Re: NAT stops forwarding ACKs after PMTU discovery

Re: NAT stops forwarding ACKs after PMTU discovery

[Eric Dumazet]

On Sat, 2013-08-17 at 22:55 -0700, Corey Hickey wrote:
> Hi,
> 
> If there is a better user-oriented list I should ask this question on,
> please tell me; I've asked a couple similar questions on netdev before
> and gotten some great help.
> 
> 
> I'm having a problem wherein some NATted connections stop forwarding
> packets. I've troubleshot the problem far enough to tell that it happens
> when path MTU discovery happens--the ACK to the retransmitted packet
> never gets forwarded back to my local client (and neither do the
> retransmitted ACKs).
> 
> This is my setup:
> ----------------------------------------------------------------------
>        MTU 9100                  MTU 1355  MTU 1500
> client      --> linux router          -->  vpn  --> work host
> 198.18.0.3      198.18.0.1    (eth0)                10.15.24.13
>                 192.168.61.54 (tun0)
> ----------------------------------------------------------------------
> 
> I'm running openconnect on a Linux router to connect to a cisco VPN at
> work. The Linux router does SNAT (MASQUERADE) over the virtual tun0
> interface created by openconnect.
> 
> This is the best test I have so far:
> 
> $ sudo ip route flush cache ; ssh workhost.example.com exit
> Read from socket failed: Connection reset by peer
> 
> Most of the time, this hangs for a few minutes before giving up;
> sometimes it just works fine. tcpdumps indicate that when it works, all
> packets transmitted are below the VPN's MTU; when it fails, MTU
> discovery has happened:
> 
> 1. client sends large packet, for example 1832 bytes
> 2. router sends ICMP fragmentation needed
> 3. client retransmits with a smaller packet, for example 1303 bytes
> 4. router forwards packet over VPN
> 5. work host ACKs packet
> 6. router receives ACK but does not forward it
> 7. both endpoints retransmit, but ACKs never get forwarded
> 
> The NAT table maintains an entry throughout (and continues to SNAT
> retransmits from the client), but the ACKs from the server never get
> forwarded.
> 
> fire:~# netstat-nat -n -d 10.15.24.13
> Proto NATed Address           Destination Address            State
> tcp   198.18.0.3:51076        10.15.24.13:22                 ESTABLISHED
> 
> 
> I put a LOG target on the INPUT chain, though, and apparently these
> packets are bypassing the NAT; the kernel does not think they should be
> forwarded (else they would be on the FORWARD chain).
> 
> [23335.509084] IN=tun0 OUT= MAC= SRC=10.15.24.13 DST=192.168.61.54
> LEN=64 TOS=0x00 PREC=0x00 TTL=62 ID=28270 DF PROTO=TCP SPT=22 DPT=51076
> WINDOW=134 RES=0x00 ACK FIN URGP=0
> 
> 
> The kernel on the router is 3.10-2-amd64 (from Debian testing). I had
> the same problem with 3.2.0-4-amd64 (from Debian stable), before I tried
> to see if I could upgrade the problem away.
> 
> Any help would be much appreciated. I can send full tcpdumps if needed.

This looks like the bug we had to fix recently : 

http://git.kernel.org/cgit/linux/kernel/git/davem/net.git/commit/?id=356d7d88e088687b6578ca64601b0a2c9d145296

Could you try latest net tree ?

Re: NAT stops forwarding ACKs after PMTU discovery

[Corey Hickey]

On 2013-08-18 08:24, Eric Dumazet wrote:
> This looks like the bug we had to fix recently : 
> 
> http://git.kernel.org/cgit/linux/kernel/git/davem/net.git/commit/?id=356d7d88e088687b6578ca64601b0a2c9d145296
> 
> Could you try latest net tree ?

Thanks, I will do so, hopefully this evening (busy today).

-Corey

Re: NAT stops forwarding ACKs after PMTU discovery

[Jozsef Kadlecsik]

On Sun, 18 Aug 2013, Eric Dumazet wrote:

> On Sat, 2013-08-17 at 22:55 -0700, Corey Hickey wrote:
> > Hi,
> > 
> > If there is a better user-oriented list I should ask this question on,
> > please tell me; I've asked a couple similar questions on netdev before
> > and gotten some great help.
> > 
> > 
> > I'm having a problem wherein some NATted connections stop forwarding
> > packets. I've troubleshot the problem far enough to tell that it happens
> > when path MTU discovery happens--the ACK to the retransmitted packet
> > never gets forwarded back to my local client (and neither do the
> > retransmitted ACKs).
> > 
> > This is my setup:
> > ----------------------------------------------------------------------
> >        MTU 9100                  MTU 1355  MTU 1500
> > client      --> linux router          -->  vpn  --> work host
> > 198.18.0.3      198.18.0.1    (eth0)                10.15.24.13
> >                 192.168.61.54 (tun0)
> > ----------------------------------------------------------------------
> > 
> > I'm running openconnect on a Linux router to connect to a cisco VPN at
> > work. The Linux router does SNAT (MASQUERADE) over the virtual tun0
> > interface created by openconnect.
> > 
> > This is the best test I have so far:
> > 
> > $ sudo ip route flush cache ; ssh workhost.example.com exit
> > Read from socket failed: Connection reset by peer
> > 
> > Most of the time, this hangs for a few minutes before giving up;
> > sometimes it just works fine. tcpdumps indicate that when it works, all
> > packets transmitted are below the VPN's MTU; when it fails, MTU
> > discovery has happened:
> > 
> > 1. client sends large packet, for example 1832 bytes
> > 2. router sends ICMP fragmentation needed
> > 3. client retransmits with a smaller packet, for example 1303 bytes
> > 4. router forwards packet over VPN
> > 5. work host ACKs packet
> > 6. router receives ACK but does not forward it
> > 7. both endpoints retransmit, but ACKs never get forwarded
> > 
> > The NAT table maintains an entry throughout (and continues to SNAT
> > retransmits from the client), but the ACKs from the server never get
> > forwarded.
> > 
> > fire:~# netstat-nat -n -d 10.15.24.13
> > Proto NATed Address           Destination Address            State
> > tcp   198.18.0.3:51076        10.15.24.13:22                 ESTABLISHED
> > 
> > 
> > I put a LOG target on the INPUT chain, though, and apparently these
> > packets are bypassing the NAT; the kernel does not think they should be
> > forwarded (else they would be on the FORWARD chain).
> > 
> > [23335.509084] IN=tun0 OUT= MAC= SRC=10.15.24.13 DST=192.168.61.54
> > LEN=64 TOS=0x00 PREC=0x00 TTL=62 ID=28270 DF PROTO=TCP SPT=22 DPT=51076
> > WINDOW=134 RES=0x00 ACK FIN URGP=0
> > 
> > 
> > The kernel on the router is 3.10-2-amd64 (from Debian testing). I had
> > the same problem with 3.2.0-4-amd64 (from Debian stable), before I tried
> > to see if I could upgrade the problem away.
> > 
> > Any help would be much appreciated. I can send full tcpdumps if needed.
> 
> This looks like the bug we had to fix recently : 
> 
> http://git.kernel.org/cgit/linux/kernel/git/davem/net.git/commit/?id=356d7d88e088687b6578ca64601b0a2c9d145296
> 
> Could you try latest net tree ?

It's a PMTU related problem, so I'd suggest to check first the next ones:

Aren't the ICMP error messages (i.e. fragmentation-needed) filtered out?
If not, does adding an explicit TCPMSS rule to the mangle table help?
Isn't there SACK options involved? Maybe there's a "smart" sequence 
number "anonymizer" device in the path which forgets about the SEQ numbers 
in the SACK options and thus such packets are marked as INVALID.

So tcpdumps captured at both sides could really help to tell what goes wrong.

Best regards,
Jozsef
-
E-mail  : kadlec@blackhole.kfki.hu, kadlecsik.jozsef@wigner.mta.hu
PGP key : http://www.kfki.hu/~kadlec/pgp_public_key.txt
Address : Wigner Research Centre for Physics, Hungarian Academy of Sciences
          H-1525 Budapest 114, POB. 49, Hungary

Re: NAT stops forwarding ACKs after PMTU discovery

[Eric Dumazet]

On Sun, 2013-08-18 at 23:23 +0200, Jozsef Kadlecsik wrote:

> It's a PMTU related problem, so I'd suggest to check first the next ones:
> 
> Aren't the ICMP error messages (i.e. fragmentation-needed) filtered out?
> If not, does adding an explicit TCPMSS rule to the mangle table help?

Thats totally irrelevant. tcp conntrack should not care of ICMP messages
anyway.

> Isn't there SACK options involved? Maybe there's a "smart" sequence 
> number "anonymizer" device in the path which forgets about the SEQ numbers 
> in the SACK options and thus such packets are marked as INVALID.
> 

As I understood the report, they were no SACK at all, but plain ACK
messages, and conntrack believes these ACK are outside the window
because it remembers the @end sequence of the probe, and the @maxwin of
receiver.

Lets say the probe is 2000 bytes, seq 1:2001

conntrack records end = 2001

Of course this packet is lost.

The retransmit sends a 1000 bytes packet, seq 1:1001

conntrack pass the packet.

Receiver acks it using ack 1001

conntrack drops this ack because of some confusion about the prior 2001
end, and the window (might be as small as 1000)

Code like this seems very suspect to me :

before(sack, receiver->td_end + 1)

It basically says that conntrack doesn't deal with reorders on the
network.

> So tcpdumps captured at both sides could really help to tell what goes wrong.

We probably can reproduce this bug using packetdrill.

Re: NAT stops forwarding ACKs after PMTU discovery

[Eric Dumazet]

On Sun, 2013-08-18 at 17:00 -0700, Eric Dumazet wrote:

> Code like this seems very suspect to me :
> 
> before(sack, receiver->td_end + 1)
> 

My suggestion would be to try :

diff --git a/net/netfilter/nf_conntrack_proto_tcp.c b/net/netfilter/nf_conntrack_proto_tcp.c
index 2f80107..1862902 100644
--- a/net/netfilter/nf_conntrack_proto_tcp.c
+++ b/net/netfilter/nf_conntrack_proto_tcp.c
@@ -656,12 +656,12 @@ static bool tcp_in_window(const struct nf_conn *ct,
 	pr_debug("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
 		 before(seq, sender->td_maxend + 1),
 		 (in_recv_win ? 1 : 0),
-		 before(sack, receiver->td_end + 1),
+		 before(sack, receiver->td_end + MAXACKWINDOW(sender) + 1),
 		 after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1));
 
 	if (before(seq, sender->td_maxend + 1) &&
 	    in_recv_win &&
-	    before(sack, receiver->td_end + 1) &&
+	    before(sack, receiver->td_end + MAXACKWINDOW(sender) + 1) &&
 	    after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1)) {
 		/*
 		 * Take into account window scaling (RFC 1323).

Re: NAT stops forwarding ACKs after PMTU discovery

[Corey Hickey]

On 2013-08-18 17:03, Eric Dumazet wrote:
> On Sun, 2013-08-18 at 17:00 -0700, Eric Dumazet wrote:
> 
>> Code like this seems very suspect to me :
>>
>> before(sack, receiver->td_end + 1)
>>
> 
> My suggestion would be to try :
> 
> diff --git a/net/netfilter/nf_conntrack_proto_tcp.c b/net/netfilter/nf_conntrack_proto_tcp.c

[...]

Thanks for all your suggestions--I really wasn't expecting so much on a
weekend. Here's all the data I have for tonight.


I tried the linux-next kernel, then linux-next with your patch applied.
Neither of them fix the problem, unfortunately. I have taken tcpdumps
for a working SSH and a failing SSH.

http://fatooh.org/files/tmp/linux-next-patch1.tar.bz2

[localhost]
sudo tcpdump -ni br0 -s 0 -w /tmp/local.pcap 'host 10.15.24.13 or icmp'

[router eth0]
tcpdump -ni eth0 -s 0 -w /tmp/eth0.pcap \
    'host 10.15.24.13 or (icmp and host not 69.78.33.132)'
* the exclusion here is just to remove some unrelated clutter

[router tun0]
tcpdump -ni tun0 -s 0 -w /tmp/tun0.pcap -s 0 'host 10.15.24.13 or icmp'

[remote]
tcpdump -ni eth0 -s 0 -w remote.pcap 'host 192.168.61.56'


Some notes:

1. I tested the new kernels only on the Linux router, assuming that is
where it was intended.

2. I take back what I wrote earlier about every connection that involves
PMTU discovery failed; I may have been observing this wrong. For now,
the situation is that some connections stop forwarding packets from the
remote host immediately after the retransmit, while other work fine.

3. From local.pcap, you can see that my localhost doesn't actually
transmit a large packet, yet the router's eth0 sees a large packet come
in. I think this is due to TSO, but I'm not completely sure.

4. For some reason, I cannot reproduce this when SSHing to a host at
work that is running Debian sid with 3.10-1-amd64, but I can reproduce
it when SSHing to hosts running Centos 6.4 with
2.6.32-358.6.1.el6.x86_64 (which surely has a ton of patches applied,
for whatever that's worth).

5. I have only a vague understanding of SACK; I will be reading up on
this soon. I will also look into packetdrill for reproducing the
problem, if the SSH results aren't good enough.

6. If I reduce the MTU on localhost to match the path MTU, the problem
does go away.

Thanks again for all the help,
Corey

Re: NAT stops forwarding ACKs after PMTU discovery

[Christoph Paasch]

Hello,

I would say, the problem is due to a sequence-number rewriting
middlebox, who does not correctly handle the SACK-blocks.

In remote.pcap, you see:
Packet#10: A Dup-ACK: ACK 1005503816, SACK: 1005505184-1005505648
The SACK actually covers Packet#9

In tun0.pcap, you have:
Packet#9: The one that is SACK'ed: SEQ: 3514869772
Packet#11: The Dup-ACK: ACK: 3514868404, SACK: 3570452498-3570452962

You can see that the SACK-block is not really aligned with the ACK-numbers.

Netfilter is probably dropping the Dup-ACK, because the SACK-block is
acknowledging unseen data.


There are middleboxes out there that randomize the sequence numbers, due to
an old bug in Windows, where the initial sequence number was not
sufficiently randomized. There are some of these middleboxes, who simply do
not support SACK, thus modify only the sequence numbers of the TCP-header
(https://supportforums.cisco.com/docs/DOC-12668#TCP_Sequence_Number_Randomization_and_SACK).

This results in a TCP retransmission timeout on the sender, because of
the invalid SACK-blocks, the duplicate ACKs are not accounted. This
completly kills the performance, as you can see in our presentation given at
IETF87: http://tools.ietf.org/agenda/87/slides/slides-87-tcpm-11.pdf

We have a patch that accounts DUP-ACKs with invalid SACK-blocks effectively
as duplicate acknowledgments. I can send the patches, if the
netdev-community is interested in accepting these upstream.


Cheers,
Christoph



On 19/08/13 - 01:43:18, Corey Hickey wrote:
> On 2013-08-18 17:03, Eric Dumazet wrote:
> > On Sun, 2013-08-18 at 17:00 -0700, Eric Dumazet wrote:
> > 
> >> Code like this seems very suspect to me :
> >>
> >> before(sack, receiver->td_end + 1)
> >>
> > 
> > My suggestion would be to try :
> > 
> > diff --git a/net/netfilter/nf_conntrack_proto_tcp.c b/net/netfilter/nf_conntrack_proto_tcp.c
> 
> [...]
> 
> Thanks for all your suggestions--I really wasn't expecting so much on a
> weekend. Here's all the data I have for tonight.
> 
> 
> I tried the linux-next kernel, then linux-next with your patch applied.
> Neither of them fix the problem, unfortunately. I have taken tcpdumps
> for a working SSH and a failing SSH.
> 
> http://fatooh.org/files/tmp/linux-next-patch1.tar.bz2
> 
> [localhost]
> sudo tcpdump -ni br0 -s 0 -w /tmp/local.pcap 'host 10.15.24.13 or icmp'
> 
> [router eth0]
> tcpdump -ni eth0 -s 0 -w /tmp/eth0.pcap \
>     'host 10.15.24.13 or (icmp and host not 69.78.33.132)'
> * the exclusion here is just to remove some unrelated clutter
> 
> [router tun0]
> tcpdump -ni tun0 -s 0 -w /tmp/tun0.pcap -s 0 'host 10.15.24.13 or icmp'
> 
> [remote]
> tcpdump -ni eth0 -s 0 -w remote.pcap 'host 192.168.61.56'
> 
> 
> Some notes:
> 
> 1. I tested the new kernels only on the Linux router, assuming that is
> where it was intended.
> 
> 2. I take back what I wrote earlier about every connection that involves
> PMTU discovery failed; I may have been observing this wrong. For now,
> the situation is that some connections stop forwarding packets from the
> remote host immediately after the retransmit, while other work fine.
> 
> 3. From local.pcap, you can see that my localhost doesn't actually
> transmit a large packet, yet the router's eth0 sees a large packet come
> in. I think this is due to TSO, but I'm not completely sure.
> 
> 4. For some reason, I cannot reproduce this when SSHing to a host at
> work that is running Debian sid with 3.10-1-amd64, but I can reproduce
> it when SSHing to hosts running Centos 6.4 with
> 2.6.32-358.6.1.el6.x86_64 (which surely has a ton of patches applied,
> for whatever that's worth).
> 
> 5. I have only a vague understanding of SACK; I will be reading up on
> this soon. I will also look into packetdrill for reproducing the
> problem, if the SSH results aren't good enough.
> 
> 6. If I reduce the MTU on localhost to match the path MTU, the problem
> does go away.
> 
> Thanks again for all the help,
> Corey
> --
> To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: NAT stops forwarding ACKs after PMTU discovery

[Eric Dumazet]

On Mon, 2013-08-19 at 14:33 +0200, Christoph Paasch wrote:
> Hello,
> 
> I would say, the problem is due to a sequence-number rewriting
> middlebox, who does not correctly handle the SACK-blocks.
> 
> In remote.pcap, you see:
> Packet#10: A Dup-ACK: ACK 1005503816, SACK: 1005505184-1005505648
> The SACK actually covers Packet#9
> 
> In tun0.pcap, you have:
> Packet#9: The one that is SACK'ed: SEQ: 3514869772
> Packet#11: The Dup-ACK: ACK: 3514868404, SACK: 3570452498-3570452962
> 
> You can see that the SACK-block is not really aligned with the ACK-numbers.
> 
> Netfilter is probably dropping the Dup-ACK, because the SACK-block is
> acknowledging unseen data.
> 
> 
> There are middleboxes out there that randomize the sequence numbers, due to
> an old bug in Windows, where the initial sequence number was not
> sufficiently randomized. There are some of these middleboxes, who simply do
> not support SACK, thus modify only the sequence numbers of the TCP-header
> (https://supportforums.cisco.com/docs/DOC-12668#TCP_Sequence_Number_Randomization_and_SACK).
> 
> This results in a TCP retransmission timeout on the sender, because of
> the invalid SACK-blocks, the duplicate ACKs are not accounted. This
> completly kills the performance, as you can see in our presentation given at
> IETF87: http://tools.ietf.org/agenda/87/slides/slides-87-tcpm-11.pdf
> 
> We have a patch that accounts DUP-ACKs with invalid SACK-blocks effectively
> as duplicate acknowledgments. I can send the patches, if the
> netdev-community is interested in accepting these upstream.
> 

You mean a netfilter patch, a tcp patch, or both ?

Re: NAT stops forwarding ACKs after PMTU discovery

[Christoph Paasch]

On 19/08/13 - 06:24:17, Eric Dumazet wrote:
> On Mon, 2013-08-19 at 14:33 +0200, Christoph Paasch wrote:
> > Hello,
> > 
> > I would say, the problem is due to a sequence-number rewriting
> > middlebox, who does not correctly handle the SACK-blocks.
> > 
> > In remote.pcap, you see:
> > Packet#10: A Dup-ACK: ACK 1005503816, SACK: 1005505184-1005505648
> > The SACK actually covers Packet#9
> > 
> > In tun0.pcap, you have:
> > Packet#9: The one that is SACK'ed: SEQ: 3514869772
> > Packet#11: The Dup-ACK: ACK: 3514868404, SACK: 3570452498-3570452962
> > 
> > You can see that the SACK-block is not really aligned with the ACK-numbers.
> > 
> > Netfilter is probably dropping the Dup-ACK, because the SACK-block is
> > acknowledging unseen data.
> > 
> > 
> > There are middleboxes out there that randomize the sequence numbers, due to
> > an old bug in Windows, where the initial sequence number was not
> > sufficiently randomized. There are some of these middleboxes, who simply do
> > not support SACK, thus modify only the sequence numbers of the TCP-header
> > (https://supportforums.cisco.com/docs/DOC-12668#TCP_Sequence_Number_Randomization_and_SACK).
> > 
> > This results in a TCP retransmission timeout on the sender, because of
> > the invalid SACK-blocks, the duplicate ACKs are not accounted. This
> > completly kills the performance, as you can see in our presentation given at
> > IETF87: http://tools.ietf.org/agenda/87/slides/slides-87-tcpm-11.pdf
> > 
> > We have a patch that accounts DUP-ACKs with invalid SACK-blocks effectively
> > as duplicate acknowledgments. I can send the patches, if the
> > netdev-community is interested in accepting these upstream.
> > 
> 
> You mean a netfilter patch, a tcp patch, or both ?

It's a TCP-patch, that interprets duplicate-acks with invalid SACK-blocks as
duplicate acks in tcp_sock->sacked_out.


Christoph

Re: NAT stops forwarding ACKs after PMTU discovery

[Eric Dumazet]

On Mon, 2013-08-19 at 15:49 +0200, Christoph Paasch wrote:

> 
> It's a TCP-patch, that interprets duplicate-acks with invalid SACK-blocks as
> duplicate acks in tcp_sock->sacked_out.

Yeah, but here, this is conntrack who is blocking the thing.

TCP receiver has no chance to 'fix' it.

See conntrack is one of those buggy middle box as well.

So if you want to properly handle this mess, you'll also have to fix
conntrack.

Re: NAT stops forwarding ACKs after PMTU discovery

[Phil Oester]

On Mon, Aug 19, 2013 at 06:58:05AM -0700, Eric Dumazet wrote:
> Yeah, but here, this is conntrack who is blocking the thing.
> 
> TCP receiver has no chance to 'fix' it.
> 
> See conntrack is one of those buggy middle box as well.
> 
> So if you want to properly handle this mess, you'll also have to fix
> conntrack.

Better to fix the box which is randomizing the acks and ignoring
the SACKs.  Usually these are older Cisco PIX/ASA devices which just
need a code upgrade.  

Phil

Re: NAT stops forwarding ACKs after PMTU discovery

[Christoph Paasch]

On 19/08/13 - 06:58:05, Eric Dumazet wrote:
> On Mon, 2013-08-19 at 15:49 +0200, Christoph Paasch wrote:
> > It's a TCP-patch, that interprets duplicate-acks with invalid SACK-blocks as
> > duplicate acks in tcp_sock->sacked_out.
> 
> Yeah, but here, this is conntrack who is blocking the thing.

Yes, of course. I know that here conntrack is blocking the dup-ACKs.

Sorry for having added TCP-stack specific things in a conntrack-thread.
I just wanted to highlight that the TCP-stack has also problems with
sequence-number rewriting middleboxes.

> TCP receiver has no chance to 'fix' it.
> 
> See conntrack is one of those buggy middle box as well.
> 
> So if you want to properly handle this mess, you'll also have to fix
> conntrack.

Re: NAT stops forwarding ACKs after PMTU discovery

[Eric Dumazet]

On Mon, 2013-08-19 at 07:35 -0700, Phil Oester wrote:
> On Mon, Aug 19, 2013 at 06:58:05AM -0700, Eric Dumazet wrote:
> > Yeah, but here, this is conntrack who is blocking the thing.
> > 
> > TCP receiver has no chance to 'fix' it.
> > 
> > See conntrack is one of those buggy middle box as well.
> > 
> > So if you want to properly handle this mess, you'll also have to fix
> > conntrack.
> 
> Better to fix the box which is randomizing the acks and ignoring
> the SACKs.  Usually these are older Cisco PIX/ASA devices which just
> need a code upgrade.  


Having a patch to relax things could be evaluated, if not adding new
problems ( http://en.wikipedia.org/wiki/Robustness_principle ) 

Re: NAT stops forwarding ACKs after PMTU discovery

[Christoph Paasch]

On 19/08/13 - 07:35:09, Phil Oester wrote:
> On Mon, Aug 19, 2013 at 06:58:05AM -0700, Eric Dumazet wrote:
> > Yeah, but here, this is conntrack who is blocking the thing.
> > 
> > TCP receiver has no chance to 'fix' it.
> > 
> > See conntrack is one of those buggy middle box as well.
> > 
> > So if you want to properly handle this mess, you'll also have to fix
> > conntrack.
> 
> Better to fix the box which is randomizing the acks and ignoring
> the SACKs.  Usually these are older Cisco PIX/ASA devices which just
> need a code upgrade.  

I agree, the problem are these horrid middleboxes...

Unfortunately, they will hardly go away in the near futur. Rather the
opposite is the case.


If you have a public server running, I would be interested in the count of
invalid SACK-blocks received (netstat -s | grep TCPSACKDiscard). This is an
indication for such kind of middlebox between your server and the client,
implying that these connections cannot benefit from TCP-FastRetransmission
and each packet-loss will require an RTO to recover.


Cheers,
Christoph

Re: NAT stops forwarding ACKs after PMTU discovery

[Eric Dumazet]

On Mon, 2013-08-19 at 17:33 +0200, Christoph Paasch wrote:

> 
> Unfortunately, they will hardly go away in the near futur. Rather the
> opposite is the case.
> 
> 
> If you have a public server running, I would be interested in the count of
> invalid SACK-blocks received (netstat -s | grep TCPSACKDiscard). This is an
> indication for such kind of middlebox between your server and the client,
> implying that these connections cannot benefit from TCP-FastRetransmission
> and each packet-loss will require an RTO to recover.
> 

If the (random) sequence offset is small rather than completely out of
window, it's going to be hard to detect all problems.

Show us your patch ;)

Re: NAT stops forwarding ACKs after PMTU discovery

[Christoph Paasch]

On 19/08/13 - 09:00:31, Eric Dumazet wrote:
> On Mon, 2013-08-19 at 17:33 +0200, Christoph Paasch wrote:
> > Unfortunately, they will hardly go away in the near futur. Rather the
> > opposite is the case.
> > 
> > 
> > If you have a public server running, I would be interested in the count of
> > invalid SACK-blocks received (netstat -s | grep TCPSACKDiscard). This is an
> > indication for such kind of middlebox between your server and the client,
> > implying that these connections cannot benefit from TCP-FastRetransmission
> > and each packet-loss will require an RTO to recover.
> > 
> 
> If the (random) sequence offset is small rather than completely out of
> window, it's going to be hard to detect all problems.

Yes, it is not possible to make it 100% perfect. But considering the size of
the seq-space, the probability is rather low that the SACK-block falls
in-window.

> Show us your patch ;)

Will send it soon. Have to rebase on net-next,... :)
(it's several months ago that I did this)


Christoph

Re: NAT stops forwarding ACKs after PMTU discovery

[Phil Oester]

On Mon, Aug 19, 2013 at 07:15:19PM +0200, Christoph Paasch wrote:
> On 19/08/13 - 09:00:31, Eric Dumazet wrote:
> > On Mon, 2013-08-19 at 17:33 +0200, Christoph Paasch wrote:
> > > Unfortunately, they will hardly go away in the near futur. Rather the
> > > opposite is the case.
> > > 
> > > 
> > > If you have a public server running, I would be interested in the count of
> > > invalid SACK-blocks received (netstat -s | grep TCPSACKDiscard). This is an
> > > indication for such kind of middlebox between your server and the client,
> > > implying that these connections cannot benefit from TCP-FastRetransmission
> > > and each packet-loss will require an RTO to recover.
> > > 
> > 
> > If the (random) sequence offset is small rather than completely out of
> > window, it's going to be hard to detect all problems.

It is not small unfortunately.  The bug is that with ISN randomization enabled
the PIX leaves the SACK sequence numbers untouched.  For reference, the cisco
bug is CSCse14419.

> Yes, it is not possible to make it 100% perfect. But considering the size of
> the seq-space, the probability is rather low that the SACK-block falls
> in-window.

s/rather low/nil/

> > Show us your patch ;)

Or just disable SACK on the box behind the problematic PIX.

Phil

Re: NAT stops forwarding ACKs after PMTU discovery

[Eric Dumazet]

On Mon, 2013-08-19 at 19:15 +0200, Christoph Paasch wrote:

> 
> Will send it soon. Have to rebase on net-next,... :)
> (it's several months ago that I did this)

No hurry.

This is a very unlikely problem anyway.

Trace taken on a random Google server :

TcpExtTCPSACKDiscard 44666
TcpExtTCPSACKReneging 173593
TcpExtTCPSACKReorder 1331306
TcpExtTCPSackFailures 791292
TcpExtTCPSackMerged 261702691
TcpExtTCPSackRecovery 57267494
TcpExtTCPSackRecoveryFail 2008511
TcpExtTCPSackShiftFallback 841017416
TcpExtTCPSackShifted 619337144
...
TcpOutSegs 42215528084
TcpRetransSegs 441868477

Re: NAT stops forwarding ACKs after PMTU discovery

[Corey Hickey]

On 2013-08-19 05:33, Christoph Paasch wrote:
> Hello,
> 
> I would say, the problem is due to a sequence-number rewriting
> middlebox, who does not correctly handle the SACK-blocks.
> 
> In remote.pcap, you see:
> Packet#10: A Dup-ACK: ACK 1005503816, SACK: 1005505184-1005505648
> The SACK actually covers Packet#9
> 
> In tun0.pcap, you have:
> Packet#9: The one that is SACK'ed: SEQ: 3514869772
> Packet#11: The Dup-ACK: ACK: 3514868404, SACK: 3570452498-3570452962
> 
> You can see that the SACK-block is not really aligned with the ACK-numbers.
> 
> Netfilter is probably dropping the Dup-ACK, because the SACK-block is
> acknowledging unseen data.
> 
> 
> There are middleboxes out there that randomize the sequence numbers, due to
> an old bug in Windows, where the initial sequence number was not
> sufficiently randomized. There are some of these middleboxes, who simply do
> not support SACK, thus modify only the sequence numbers of the TCP-header
> (https://supportforums.cisco.com/docs/DOC-12668#TCP_Sequence_Number_Randomization_and_SACK).
> 
> This results in a TCP retransmission timeout on the sender, because of
> the invalid SACK-blocks, the duplicate ACKs are not accounted. This
> completly kills the performance, as you can see in our presentation given at
> IETF87: http://tools.ietf.org/agenda/87/slides/slides-87-tcpm-11.pdf

This makes good sense. I normally look at these in wireshark with
relative sequence numbers turned on, and I see in bad/tun0.pcap that the
SACK values are very high, but are normal in bad/remote.pcap.

I see the same thing in good/tun0.pcap, though, so I don't fully
understand what is making it work sometimes and not others.

I will show this thread and the Cisco docs to the network engineer at
work, and maybe we can get the SEQ randomization turned off (or at least
test it).

> We have a patch that accounts DUP-ACKs with invalid SACK-blocks effectively
> as duplicate acknowledgments. I can send the patches, if the
> netdev-community is interested in accepting these upstream.

I'll keep my eye out and test them if they come up.


Thanks,
Corey

Re: NAT stops forwarding ACKs after PMTU discovery

[Jozsef Kadlecsik]

On Mon, 19 Aug 2013, Eric Dumazet wrote:

> On Mon, 2013-08-19 at 15:49 +0200, Christoph Paasch wrote:
>
> > It's a TCP-patch, that interprets duplicate-acks with invalid SACK-blocks as
> > duplicate acks in tcp_sock->sacked_out.
> 
> Yeah, but here, this is conntrack who is blocking the thing.
> 
> TCP receiver has no chance to 'fix' it.
> 
> See conntrack is one of those buggy middle box as well.
> 
> So if you want to properly handle this mess, you'll also have to fix
> conntrack.

I beg you pardon: why conntrack should be relaxed, when it is expected
to do more strict TCP checkings (RFC5961, Section 5.).

Also, it's clearly a broken middle box. Don't shoot the messenger.

Best regards,
Jozsef
-
E-mail  : kadlec@blackhole.kfki.hu, kadlecsik.jozsef@wigner.mta.hu
PGP key : http://www.kfki.hu/~kadlec/pgp_public_key.txt
Address : Wigner Research Centre for Physics, Hungarian Academy of Sciences
          H-1525 Budapest 114, POB. 49, Hungary

Re: NAT stops forwarding ACKs after PMTU discovery

[Christoph Paasch]

On 19/08/13 - 22:13:59, Jozsef Kadlecsik wrote:
> On Mon, 19 Aug 2013, Eric Dumazet wrote:
> 
> > On Mon, 2013-08-19 at 15:49 +0200, Christoph Paasch wrote:
> >
> > > It's a TCP-patch, that interprets duplicate-acks with invalid SACK-blocks as
> > > duplicate acks in tcp_sock->sacked_out.
> > 
> > Yeah, but here, this is conntrack who is blocking the thing.
> > 
> > TCP receiver has no chance to 'fix' it.
> > 
> > See conntrack is one of those buggy middle box as well.
> > 
> > So if you want to properly handle this mess, you'll also have to fix
> > conntrack.
> 
> I beg you pardon: why conntrack should be relaxed, when it is expected
> to do more strict TCP checkings (RFC5961, Section 5.).

There is no mention of SACK in this RFC.
The duplicate ACKs with invalid SACK-blocks are valid with respect to
RFC5961, Section 5.

Actually, no RFC says that dup-ACKs with invalid SACK-blocks should
be discarded.


Cheers,
Christoph

Re: NAT stops forwarding ACKs after PMTU discovery

[Eric Dumazet]

On Mon, 2013-08-19 at 22:13 +0200, Jozsef Kadlecsik wrote:
> On Mon, 19 Aug 2013, Eric Dumazet wrote:
> 
> > On Mon, 2013-08-19 at 15:49 +0200, Christoph Paasch wrote:
> >
> > > It's a TCP-patch, that interprets duplicate-acks with invalid SACK-blocks as
> > > duplicate acks in tcp_sock->sacked_out.
> > 
> > Yeah, but here, this is conntrack who is blocking the thing.
> > 
> > TCP receiver has no chance to 'fix' it.
> > 
> > See conntrack is one of those buggy middle box as well.
> > 
> > So if you want to properly handle this mess, you'll also have to fix
> > conntrack.
> 
> I beg you pardon: why conntrack should be relaxed, when it is expected
> to do more strict TCP checkings (RFC5961, Section 5.).
> 
> Also, it's clearly a broken middle box. Don't shoot the messenger.

Frames are dropped by conntrack, before TCP receiver can even have a
choice.

So Christoph patch would be of no use for Corey.

I do not think I shot anyone, only stated the truth.

We have workarounds in our stack to 'fix' bugs from others, there
is no shame in this.

Glad to see you are interested in RFC 5961 support, as conntrack is
known to break the ACK challenges in response to RST messages (section
3)

Re: NAT stops forwarding ACKs after PMTU discovery

[Jozsef Kadlecsik]

On Mon, 19 Aug 2013, Eric Dumazet wrote:

> On Mon, 2013-08-19 at 22:13 +0200, Jozsef Kadlecsik wrote:
> > On Mon, 19 Aug 2013, Eric Dumazet wrote:
> > 
> > > On Mon, 2013-08-19 at 15:49 +0200, Christoph Paasch wrote:
> > >
> > > > It's a TCP-patch, that interprets duplicate-acks with invalid 
> > > > SACK-blocks as duplicate acks in tcp_sock->sacked_out.
> > > 
> > > Yeah, but here, this is conntrack who is blocking the thing.
> > > 
> > > TCP receiver has no chance to 'fix' it.
> > > 
> > > See conntrack is one of those buggy middle box as well.
> > > 
> > > So if you want to properly handle this mess, you'll also have to fix
> > > conntrack.
> > 
> > I beg you pardon: why conntrack should be relaxed, when it is expected
> > to do more strict TCP checkings (RFC5961, Section 5.).
> > 
> > Also, it's clearly a broken middle box. Don't shoot the messenger.
> 
> Frames are dropped by conntrack, before TCP receiver can even have a
> choice.
> 
> So Christoph patch would be of no use for Corey.

Yes, exactly.
 
> I do not think I shot anyone, only stated the truth.

There's a middlebox in the path wich breaks SACK completely and conntrack 
drops (technically marks as INVALID) the packets with bogus SACK options.

It can be fixed by fixing the middlebox, or disabling SACK by the 
TCPOPTSTRIP target, or by relaxing conntrack. For the latter, the next 
untested patch may be sufficient:

diff --git a/net/netfilter/nf_conntrack_proto_tcp.c b/net/netfilter/nf_conntrack_proto_tcp.c
index 7dcc376..8b5d783 100644
--- a/net/netfilter/nf_conntrack_proto_tcp.c
+++ b/net/netfilter/nf_conntrack_proto_tcp.c
@@ -649,6 +649,11 @@ static bool tcp_in_window(const struct nf_conn *ct,
 		 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 		 receiver->td_scale);
 
+	/* Fall back to ACK when SACK is bogus */
+	if (!(before(sack, receiver->td_end + 1) &&
+	      after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1)))
+		sack = ack;
+	      
 	pr_debug("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
 		 before(seq, sender->td_maxend + 1),
 		 after(end, sender->td_end - receiver->td_maxwin - 1),

However it is good to cover the issue thus?
 
> We have workarounds in our stack to 'fix' bugs from others, there
> is no shame in this.
> 
> Glad to see you are interested in RFC 5961 support, as conntrack is
> known to break the ACK challenges in response to RST messages (section
> 3)

*Any* netfilter configuration where the non-allowed TCP packets are 
dropped and not rejected breaks the ACK challenges. That is why I consider 
only section 5 useful from RFC 5961.

Best regards,
Jozsef
-
E-mail  : kadlec@blackhole.kfki.hu, kadlecsik.jozsef@wigner.mta.hu
PGP key : http://www.kfki.hu/~kadlec/pgp_public_key.txt
Address : Wigner Research Centre for Physics, Hungarian Academy of Sciences
          H-1525 Budapest 114, POB. 49, Hungary

Re: NAT stops forwarding ACKs after PMTU discovery

[Corey Hickey]

On 2013-08-19 15:07, Jozsef Kadlecsik wrote:
> It can be fixed by fixing the middlebox, or disabling SACK by the 
> TCPOPTSTRIP target, or by relaxing conntrack. For the latter, the next 
> untested patch may be sufficient:
> 
> diff --git a/net/netfilter/nf_conntrack_proto_tcp.c b/net/netfilter/nf_conntrack_proto_tcp.c
> index 7dcc376..8b5d783 100644
> --- a/net/netfilter/nf_conntrack_proto_tcp.c
> +++ b/net/netfilter/nf_conntrack_proto_tcp.c
> @@ -649,6 +649,11 @@ static bool tcp_in_window(const struct nf_conn *ct,
>  		 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
>  		 receiver->td_scale);
>  
> +	/* Fall back to ACK when SACK is bogus */
> +	if (!(before(sack, receiver->td_end + 1) &&
> +	      after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1)))
> +		sack = ack;
> +	      
>  	pr_debug("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
>  		 before(seq, sender->td_maxend + 1),
>  		 after(end, sender->td_end - receiver->td_maxwin - 1),
> 
> However it is good to cover the issue thus?

This didn't quite apply to my kernel tree, but I let patch apply it with
fuzz=2 and got:

diff --git a/net/netfilter/nf_conntrack_proto_tcp.c b/net/netfilter/nf_conntrack_proto_tcp.c
index 2f80107..94b326b 100644
--- a/net/netfilter/nf_conntrack_proto_tcp.c
+++ b/net/netfilter/nf_conntrack_proto_tcp.c
@@ -653,6 +653,11 @@ static bool tcp_in_window(const struct nf_conn *ct,
        in_recv_win = !receiver->td_maxwin ||
                      after(end, sender->td_end - receiver->td_maxwin - 1);
 
+       /* Fall back to ACK when SACK is bogus */
+       if (!(before(sack, receiver->td_end + 1) &&
+             after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1)))
+               sack = ack;
+
        pr_debug("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
                 before(seq, sender->td_maxend + 1),
                 (in_recv_win ? 1 : 0),



I can confirm, that does indeed work! Thank you. I will continue watching
this thread for other things to test, if need be.

-Corey