netdev - Re: net_sched 00/07: classful multiqueue dummy scheduler

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

Message-ID: <4AA5175F.6030600@trash.net>
Date:	Mon, 07 Sep 2009 16:23:27 +0200
From:	Patrick McHardy <kaber@...sh.net>
To:	Eric Dumazet <eric.dumazet@...il.com>
CC:	David Miller <davem@...emloft.net>, netdev@...r.kernel.org
Subject: Re: net_sched 00/07: classful multiqueue dummy scheduler

Patrick McHardy wrote:
> Eric Dumazet wrote:
>> Had very litle time to test this, but got problems very fast, if rate estimator configured.
> 
> I didn't test that, but I'll look into it.
> 
>> qdisc mq 1: root
>>  Sent 528702 bytes 3491 pkt (dropped 0, overlimits 0 requeues 0)
>>  rate 177925Kbit 49pps backlog 0b 0p requeues 0
>> qdisc pfifo 8001: parent 1:1 limit 1000p
>>  Sent 528702 bytes 3491 pkt (dropped 0, overlimits 0 requeues 0)
>>  rate 25400bit 21pps backlog 0b 0p requeues 0
>>
>> <<<crash>>>
> 
> Did you capture the crash?
> 
>> (On another term I had a "ping -i 0.1 192.168.20.120" that gave :
>>
>> 2009/08/07 14:53:42.498 64 bytes from 192.168.20.120: icmp_seq=1982 ttl=64 time=0.126 ms
>> 2009/08/07 14:53:42.598 64 bytes from 192.168.20.120: icmp_seq=1983 ttl=64 time=0.118 ms
>> 2009/08/07 14:53:42.698 64 bytes from 192.168.20.120: icmp_seq=1984 ttl=64 time=0.114 ms
>> 2009/08/07 14:53:42.798 64 bytes from 192.168.20.120: icmp_seq=1985 ttl=64 time=0.123 ms
>> 2009/08/07 14:53:42.898 64 bytes from 192.168.20.120: icmp_seq=1986 ttl=64 time=0.126 ms
>> 2009/08/07 14:53:42.998 64 bytes from 192.168.20.120: icmp_seq=1987 ttl=64 time=0.119 ms
>> 2009/08/07 14:53:43.098 64 bytes from 192.168.20.120: icmp_seq=1988 ttl=64 time=0.122 ms
>> 2009/08/07 14:53:43.198 64 bytes from 192.168.20.120: icmp_seq=1989 ttl=64 time=0.119 ms
>> 2009/08/07 14:53:43.298 64 bytes from 192.168.20.120: icmp_seq=1990 ttl=64 time=0.117 ms
>> 2009/08/07 14:53:43.398 64 bytes from 192.168.20.120: icmp_seq=1991 ttl=64 time=0.117 ms
>> ping: sendmsg: No buffer space available
> 
> Was this also with rate estimators? No buffer space available
> indicates that some class/qdisc isn't dequeued or the packets
> are leaking, so the output of tc -s -d qdisc show ... might be
> helpful.

I figured out the bug, which is likely responsible for both
problems. When grafting a mq class and creating a rate estimator,
the new qdisc is not attached to the device queue yet and also
doesn't have TC_H_ROOT as parent, so qdisc_create() selects
qdisc_root_sleeping_lock() for the estimator, which belongs to
the qdisc that is getting replaced.

This is a patch I used for testing, but I'll come up with
something more elegant (I hope) as a final fix :)



View attachment "x" of type "text/plain" (1498 bytes)