On Fri, Aug 22, 2014, at 07:42 AM, Tom Eastep wrote:
> ... if you populate the mangle table in the tcstart script, then
> iptables-restore will replace your rules.
Noted. And avoided.
> for what you are doing, 'start' is preferred.
Noted. Moved.
> > atm, moving the ENTIRE script to 'started' results in the mangle
> > chain being preserved ... although the classification of traffic is
> > NOT working as intended (yet).
Found the problem -- a MARK collision in mangle.
My QoS script's creating class numbering, starting with a "low latency" class
at classid == 1:2
/lib.private
qos_control() {
...
CLASS_LOW_LATENCY=2 # 1:2 Low latency (VoIP)
...
${TC_BIN} class add dev ${intfc} parent 1:1 classid
1:${CLASS_LOW_LATENCY} hfsc \
sc m1 ${uplink}kbit d 200ms m2 $((10*$uplink/10))kbit
\
ul rate ${uplink}kbit
...
}
Checking the mangle table, this conflicts directly with my MultiISP config's
'prov2' "MARK set 0x2",
/provider
...
prov2 2 2 main EXTIF2 10.0.0.1 track,fallback
INTIF
/mangle
...
MARK(2):P INTIF:xx.xx.xx.xx - tcp ...
...
Changing the qos_script's class numbering to avoid the collision,
/lib.private
qos_control() {
...
- CLASS_LOW_LATENCY=2 # 1:2 Low latency (VoIP)
+ CLASS_LOW_LATENCY=10 # 1:10 Low latency (VoIP)
...
${TC_BIN} class add dev ${intfc} parent 1:1 classid
1:${CLASS_LOW_LATENCY} hfsc \
sc m1 ${uplink}kbit d 200ms m2 $((10*$uplink/10))kbit
\
ul rate ${uplink}kbit
...
}
Now classification is occurring as expected for VOIP traffic
tc -s -d class show dev eth0
...
class hfsc 1:10 parent 1:1 leaf 10: sc m1 5400Kbit d 200.0ms m2
5400Kbit ul m1 0bit d 0us m2 5400Kbit
Sent 232651 bytes 780 pkt (dropped 0, overlimits 0 requeues 0)
backlog 0b 0p requeues 0
period 780 work 232651 bytes rtwork 232651 bytes level 0
...
VOIP still *sounds* lousy, but that's a different issue.
------------------------------------------------------------------------------
Slashdot TV.
Video for Nerds. Stuff that matters.
http://tv.slashdot.org/
_______________________________________________
Shorewall-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/shorewall-users
