netdev - Re: [REGRESSION] Userland interface breaks due to hard HFSC

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: <065a0dac-499f-7375-ddb4-1800e8ef61d1@mojatatu.com>
Date: Fri, 6 Oct 2023 19:12:46 -0300
From: Pedro Tammela <pctammela@...atatu.com>
To: Christian Theune <ct@...ingcircus.io>, stable@...r.kernel.org,
 netdev@...r.kernel.org
Cc: regressions@...ts.linux.dev, davem@...emloft.net, edumazet@...gle.com,
 kuba@...nel.org, pabeni@...hat.com
Subject: Re: [REGRESSION] Userland interface breaks due to hard HFSC_FSC
 requirement

On 06/10/2023 05:37, Christian Theune wrote:
> Hi,
> 
> (prefix, I was not aware of the regression reporting process and incorrectly reported this informally with the developers mentioned in the change)
> 
> I upgraded from 6.1.38 to 6.1.55 this morning and it broke my traffic shaping script, leaving me with a non-functional uplink on a remote router.
> 
> The script errors out like this:
> 
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2053]: + ext=ispA
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2053]: + ext_ingress=ifb0
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2053]: + modprobe ifb
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2053]: + modprobe act_mirred
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2053]: + tc qdisc del dev ispA root
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2061]: Error: Cannot delete qdisc with handle of zero.
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2053]: + true
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2053]: + tc qdisc del dev ispA ingress
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2064]: Error: Cannot find specified qdisc on specified device.
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2053]: + true
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2053]: + tc qdisc del dev ifb0 root
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2066]: Error: Cannot delete qdisc with handle of zero.
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2053]: + true
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2053]: + tc qdisc del dev ifb0 ingress
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2067]: Error: Cannot find specified qdisc on specified device.
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2053]: + true
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2053]: + tc qdisc add dev ispA handle ffff: ingress
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2053]: + ifconfig ifb0 up
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2053]: + tc filter add dev ispA parent ffff: protocol all u32 match u32 0 0 action mirred egress redirect dev ifb0
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2053]: + tc qdisc add dev ifb0 root handle 1: hfsc default 1
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2053]: + tc class add dev ifb0 parent 1: classid 1:999 hfsc rt m2 2.5gbit
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2053]: + tc class add dev ifb0 parent 1:999 classid 1:1 hfsc sc rate 50mbit
> Oct 06 05:49:22 wendy00 isp-setup-shaping-start[2077]: Error: Invalid parent - parent class must have FSC.
> 
> The error message is also a bit weird (but that’s likely due to iproute2 being weird) as the CLI interface for `tc` and the error message do not map well. (I think I would have to choose `hfsc sc` on the parent to enable the FSC option which isn’t mentioned anywhere in the hfsc manpage).
> 
> The breaking change was introduced in 6.1.53[1] and a multitude of other currently supported kernels:
> 

Hi,

Your script is actually incorrect.
`man 7 tc-hfsc` goes in depth into why, but I just wanna highlight this 
section:
SEPARATE LS / RT SCs
        Another difference from the original HFSC paper is that RT and 
LS SCs can be specified separately. Moreover, leaf classes are
        allowed to have only either RT SC or LS SC. For interior 
classes, only LS SCs make sense: any RT SC will be ignored.

The last part ("For interior classes...") was what the referenced patch 
fixed. We were mistakenly allowing RTs into "interior classes" which the 
implementation never accounted for and this was a source of crashes. I'm 
surprised you were lucky enough to never crash the kernel ;)
-=
I believe the script could be updated to the following and still achieve 
the same results:
tc class add dev ifb0 parent 1: classid 1:999 hfsc ls m2 2.5gbit
tc class add dev ifb0 parent 1:999 classid 1:1 hfsc rt rate 50mbit