Skip to contentSkip to conversation

do I need to iptables mark in this scenario?

LARTC Ron McKown 3 participants 5 messages Apr 27, 2005 – Apr 28, 2005

lore
Share & tools
lore

Ron McKown

author
#0 lore raw

Hello,
really new to advanced linux routing, but I have managed to setup something that sort of works for my needs.

I have a linux box with this configuration:

Internet - eth0 - eth1 - many users

I'm using a NAT to allow my users Internet access.

On some users, I wish to shape their download speeds as to not overload our outside connection.

Here is what I do to shape them at 64k down: (only two users shown in this example(but I have hundreds that will be shaped)) SIXFOURd
DEV=eth1

# clean up qdiscs
tc qdisc del dev $DEV root 2> /dev/null > /dev/null
tc qdisc add dev $DEV root handle 1: cbq avpkt 1000 bandwidth 50mbit

# DELL
tc class add dev $DEV parent 1: classid 1:1 cbq rate ${SIXFOUR}kbit allot 1500 prio 5 bounded isolated
tc filter add dev $DEV parent 1: protocol ip prio 16 u32 match ip dst 192.168.0.199 flowid 1:1
tc qdisc add dev $DEV parent 1:1 sfq perturb 10

# TYR
tc class add dev $DEV parent 1: classid 1:2 cbq rate ${SIXFOUR}kbit allot 1500 prio 5 bounded isolated
tc filter add dev $DEV parent 1: protocol ip prio 16 u32 match ip dst 192.168.0.133 flowid 1:2
tc qdisc add dev $DEV parent 1:2 sfq perturb 10

each user gets their own section.
This works great for downspeed only, but I also want to shape them at 64k up as well. I can't shape them on eth0, because by the time their packets get to that interface, they've already been NAT'd to a different address.

After lots of reading on the mailing list archive, it appears the best way to handle this is to mark packets from each user, then on eth0 have HTB or CBQ (really unsure which to use) shape each marked packet accordingly. Currently, I'm matching by IP (u32?), so should I match by fwmark instead?

Any help is really appreciated or even a sample script concept would make the operation more clear!

Ron

_______________________________________________
LARTC mailing list
[email protected]
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

#1 lore raw
Quoted 2 lines · Ron McKown

Hello,
really new to advanced linux routing, but I have managed to setup

something that sort of works for my needs.

Quoted 8 lines

I have a linux box with this configuration:

Internet - eth0 - eth1 - many users

I'm using a NAT to allow my users Internet access.

On some users, I wish to shape their download speeds as to not overload

our outside connection.

Quoted 2 lines

Here is what I do to shape them at 64k down: (only two users shown in

this example(but I have hundreds that will be shaped))

Quoted 8 lines

SIXFOURd
DEV=eth1

# clean up qdiscs
tc qdisc del dev $DEV root 2> /dev/null > /dev/null
tc qdisc add dev $DEV root handle 1: cbq avpkt 1000 bandwidth 50mbit

# DELL
tc class add dev $DEV parent 1: classid 1:1 cbq rate ${SIXFOUR}kbit

allot 1500 prio 5 bounded isolated

Quoted 1 line

tc filter add dev $DEV parent 1: protocol ip prio 16 u32 match ip dst

192.168.0.199 flowid 1:1

Quoted 4 lines

tc qdisc add dev $DEV parent 1:1 sfq perturb 10

# TYR
tc class add dev $DEV parent 1: classid 1:2 cbq rate ${SIXFOUR}kbit

allot 1500 prio 5 bounded isolated

Quoted 1 line

tc filter add dev $DEV parent 1: protocol ip prio 16 u32 match ip dst

192.168.0.133 flowid 1:2

Quoted 4 lines

tc qdisc add dev $DEV parent 1:2 sfq perturb 10

each user gets their own section.
This works great for downspeed only, but I also want to shape them at

64k up as well. I can't shape them on eth0, because by the time their packets get to that interface, they've already been NAT'd to a different address.

Quoted 2 lines

After lots of reading on the mailing list archive, it appears the best

way to handle this is to mark packets from each user, then on eth0 have HTB or CBQ (really unsure which to use) shape each marked packet accordingly. Currently, I'm matching by IP (u32?), so should I match by fwmark instead?

Quoted 2 lines

Any help is really appreciated or even a sample script concept would

make the operation more clear!

Quoted 2 lines

Ron

http://lartc.org/howto/lartc.cookbook.ultimate-tc.html#AEN2241

tc qdisc add dev $DEV handle ffff: ingress
tc filter add dev $DEV parent ffff: protocol ip prio 50 u32 match ip src \
0.0.0.0/0 police rate ${DOWNLINK}kbit burst 10k drop flowid :1

This means you can limit bandwidth for incoming packets too.
I think you can shape your users' UL on eth1 with "ingress" without having to mark packets...

Regards,

Sylvain

_______________________________________________
LARTC mailing list
[email protected]
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

#2 lore raw
In reply to Ron McKown
Quoted 5 lines · Ron McKown

After lots of reading on the mailing list archive, it appears the best way to handle this is to mark packets from each user, then on eth0 have HTB or CBQ (really unsure which to use) shape each marked packet accordingly. Currently, I'm matching by IP (u32?), so should I match by fwmark instead?

Yes you could do that or if you are using recent kernel/iptables you can classify directly.

http://www.netfilter.org/patch-o-matic/pom-submitted.html

If you need to use mark just do something like

$IPTABLES -t mangle -A POSTROUTING --src 192.168.0.4 -j MARK --set-mark 34

and then match it with a filter on eth0

$TC filter add dev eth0 parent 1:0 prio 6 protocol ip handle 34 fw flowid 1:34

Andy.
_______________________________________________
LARTC mailing list
[email protected]
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

author
#3 lore raw
In reply to Ron McKown

Hi Andy,
I've followed your advice and it works brilliant. However, I did fail to mention something else which is causing a problem:

Internet - eth0 - eth1 users (192.168.x.x)
local net- eth2 (10.0.x.x)

There is another interface in this router (eth2) that should not be shaped at all (it goes to another local network).

My problem is, I can shape Internet traffic going to user on eth1 (down), and user traffic going to Internet on eth0 (up). However, if the downspeed is being shaped on the eth1, that means that customers wanting something from eth2 will also be shaped.

I thought maybe I could only mark packets with a destination to eth0, which means packets going to eth2 would be left untouched, but that doesn't appear to work, or maybe I'm making a mistake.

here's my marking rule:
iptables -t mangle -A POSTROUTING --src 192.168.0.84 -o eth0 -j MARK -- set-mark 34

and here's the cbq rules (should I be using HTB for this??) SIXFOURu
VEGA�
DEV=eth1
DEVTEST=eth0
tc class add dev $DEV parent 1: classid 1:30 cbq rate ${SIXFOUR}kbit allot 1500 prio 5 bounded isolated
tc filter add dev $DEV parent 1: prio 6 protocol ip handle 34 fw flowid 1:30
tc qdisc add dev $DEV parent 1:30 sfq perturb 10
tc class add dev $DEVTEST parent 1: classid 1:30 cbq rate ${SIXFOUR}kbit allot 1500 prio 5 bounded isolated
tc filter add dev $DEVTEST parent 1: prio 6 protocol ip handle 34 fw flowid 1:30
tc qdisc add dev $DEVTEST parent 1:30 sfq perturb 10

This current setup does not shape eth0 traffic going to eth1 (because I'm not using u32, I'm trying to shape on the mark).
On $DEV, if I replace the mark handle with a u32 ip address match, then shaping will work, but then users downloading from interface eth2 will also be shaped, which I don't want.

Am I getting close or really going down the wrong path here?

Thanks so much,
Ron

Quoted 15 lines · Andy Furniss
Quoted 5 lines · Ron McKown

After lots of reading on the mailing list archive, it appears the best way to handle this is to mark packets from each user, then on eth0 have HTB or CBQ (really unsure which to use) shape each marked packet accordingly. Currently, I'm matching by IP (u32?), so should I match by fwmark instead?

Yes you could do that or if you are using recent kernel/iptables you can classify directly.

http://www.netfilter.org/patch-o-matic/pom-submitted.html

If you need to use mark just do something like

Quoted 8 lines

$IPTABLES -t mangle -A POSTROUTING --src 192.168.0.4 -j MARK --set-mark 34

and then match it with a filter on eth0

$TC filter add dev eth0 parent 1:0 prio 6 protocol ip handle 34 fw flowid 1:34

Andy.

_______________________________________________
LARTC mailing list
[email protected]
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

#4 lore raw
In reply to Ron McKown
Quoted 15 lines · Ron McKown

Hi Andy,
I've followed your advice and it works brilliant. However, I did fail to mention something else which is causing a problem:

Internet - eth0 - eth1 users (192.168.x.x)
local net- eth2 (10.0.x.x)

There is another interface in this router (eth2) that should not be shaped at all (it goes to another local network).

My problem is, I can shape Internet traffic going to user on eth1 (down), and user traffic going to Internet on eth0 (up). However, if the downspeed is being shaped on the eth1, that means that customers wanting something from eth2 will also be shaped.

Quoted 7 lines

I thought maybe I could only mark packets with a destination to eth0, which means packets going to eth2 would be left untouched, but that doesn't appear to work, or maybe I'm making a mistake.

here's my marking rule:
iptables -t mangle -A POSTROUTING --src 192.168.0.84 -o eth0 -j MARK -- set-mark 34

I would change marking to FORWARD you can use -i and -o then so for upload from eth1 to internet.

iptables -t mangle -A FORWARD -i eth1 -o eth0 -j MARK -- set-mark 34

and download

iptables -t mangle -A FORWARD -i eth0 -o eth1 -j MARK -- set-mark 34

Quoted 2 lines

and here's the cbq rules (should I be using HTB for this??)

I always use htb because it's what I am used to - I never really played around with CBQ so can't say it's better or worse.

When I first read LARTC it said HTB was easier - so that's what I used.

Andy.
_______________________________________________
LARTC mailing list
[email protected]
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc