linux-kernel - Re: 3.14 regression: huge latency in read/select on tun

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-ID: <1396531176.12330.6.camel@edumazet-glaptop2.roam.corp.google.com>
Date:	Thu, 03 Apr 2014 06:19:36 -0700
From:	Eric Dumazet <eric.dumazet@...il.com>
To:	Ortwin Glück <odi@....ch>
Cc:	linux-kernel@...r.kernel.org, netdev@...r.kernel.org
Subject: Re: 3.14 regression: huge latency in read/select on tun

On Thu, 2014-04-03 at 09:52 +0200, Ortwin Glück wrote:
> On 02.04.2014 18:42, Eric Dumazet wrote:
> > You could post a tcpdump maybe ?
> 
> This is a dump from on the tun interface (traffic inside the tunnel).
> It's an scp upload of a 2MB file. I aborted after a while when not much
> progress was seen anymore, but you get the idea.
> 
> What I see:
> - Frame 28: long delay because I had to enter the key password
> - Frame 31: the packets seems to be out of order. It has a sequence
> number that is far ahead of previously sent sequence no's. Some
> duplicates and retransmissions follow. That pattern repeats randomly all
> over the trace.
> - from Frame 73 on: ping pong of ACKs and data, while the data is
> delayed by ~200ms caused by the select(). The ping pong also shows that
> there are virtually no packets in flight (see also Window size).

It seems TSO support is broken.

Your trace have no 'big' packets but holes like :

00:41:35.832695 IP 10.243.88.168.35846 > 10.185.85.69.22: Flags [.], ack 6093, win 310, options [nop,nop,TS val 78054 ecr 3191538976], length 0
00:41:35.855990 IP 10.243.88.168.35846 > 10.185.85.69.22: Flags [P.], seq 6696:6776, ack 6093, win 310, options [nop,nop,TS val 78060 ecr 3191538976], length 80
00:41:36.047465 IP 10.185.85.69.22 > 10.243.88.168.35846: Flags [P.], seq 6093:6141, ack 6776, win 218, options [nop,nop,TS val 3191539191 ecr 78060], length 48

missing 6776:14144 sequence (6 segments)

00:41:36.048766 IP 10.243.88.168.35846 > 10.185.85.69.22: Flags [.], seq 14144:15372, ack 6141, win 310, options [nop,nop,TS val 78108 ecr 3191539191], length 1228
00:41:36.057995 IP 10.185.85.69.22 > 10.243.88.168.35846: Flags [.], ack 6776, win 218, options [nop,nop,TS val 3191539202 ecr 78060,nop,nop,sack 1 {14144:15372}], length 0

You are 'saved' because retransmits do not use TSO :

00:41:36.058050 IP 10.243.88.168.35846 > 10.185.85.69.22: Flags [.], seq 6776:8004, ack 6141, win 310, options [nop,nop,TS val 78110 ecr 3191539202], length 1228
00:41:36.058070 IP 10.243.88.168.35846 > 10.185.85.69.22: Flags [.], seq 8004:9232, ack 6141, win 310, options [nop,nop,TS val 78110 ecr 3191539202], length 1228

Also receiver seems to be unable to open TCP window, but thats a secondary problem.



--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/