Re: Mark traffic on one machine, match on another machine?

[Jack Bates <uo4zau@nottheoilrig.com>]

Thanks for this advice Anatoly, I corrected the script to redirect all 
incoming traffic to ifb0, and now statistics on the ifb0 classes and 
qdiscs are incrementing, but they're still not incrementing in the way I 
expect. Outgoing traffic is still getting classified as expected: If I 
make a request with TOS 28, I watch the statistics of eth0.2 class 2:3 
increment, but the response causes ifb0 class 2:2 to increment, not 2:3.

I think that the TOS field is correctly matched and the mark is 
correctly saved and restored, but only outgoing traffic is getting 
classified based on the mark? Do you have any more suggestions how to 
investigate why incoming traffic isn't getting classified? or is there 
another way to classify response traffic based on the TOS/DSCP field of 
the request?

Here is the corrected script:

    1: iptables -N NEW_CONN -t mangle
    2: iptables -A NEW_CONN -t mangle -m tos --tos 12 -j MARK --set-mark 1
    3: iptables -A NEW_CONN -t mangle -m tos --tos 28 -j MARK --set-mark 2
    4: iptables -A NEW_CONN -t mangle -m mark --mark 0 -j MARK --set-mark 3
    5: iptables -A NEW_CONN -t mangle -j CONNMARK --save-mark
    6:
    7: iptables -A PREROUTING -t mangle -m conntrack --ctstate INVALID 
-j DROP
    8: iptables -A PREROUTING -t mangle -m conntrack --ctstate 
ESTABLISHED -j CONNMARK --restore-mark
    9: iptables -A PREROUTING -t mangle -m conntrack --ctstate 
NEW,RELATED -j NEW_CONN
   10:
   11: ifconfig ifb0 up
   12:
   13: insmod sch_ingress
   14: tc qdisc add dev eth0.2 ingress
   15:
   16: insmod cls_u32
   17: insmod act_mirred
   18: tc filter add dev eth0.2 parent ffff: protocol ip pref 1 handle 1 
u32 match u32 0 0 action mirred egress redirect dev ifb0
   19:
   20: # Limit up and downstream to 1mbit, to "own" the queue
   21: insmod sch_tbf
   22: tc qdisc add dev eth0.2 root handle 1 tbf rate 1mbit burst 5k 
latency 70ms
   23: tc qdisc add dev ifb0 root handle 1 tbf rate 1mbit burst 5k 
latency 70ms
   24:
   25: # High, low, and "other" priority
   26: insmod sch_prio
   27: tc qdisc add dev eth0.2 parent 1: handle 2 prio
   28: tc qdisc add dev ifb0 parent 1: handle 2 prio
   29:
   30: insmod cls_fw
   31:
   32: # Classify request traffic from proxy server based on mark
   33: tc filter add dev eth0.2 parent 2: protocol ip pref 1 handle 1 fw 
flowid :1
   34: tc filter add dev eth0.2 parent 2: protocol ip pref 1 handle 2 fw 
flowid :3
   35: tc filter add dev eth0.2 parent 2: protocol ip pref 1 handle 3 fw 
flowid :2
   36:
   37: # Classify response traffic from origin server based on restored mark
   38: tc filter add dev ifb0 parent 2: protocol ip pref 1 handle 1 fw 
flowid :1
   39: tc filter add dev ifb0 parent 2: protocol ip pref 1 handle 2 fw 
flowid :3
   40: tc filter add dev ifb0 parent 2: protocol ip pref 1 handle 3 fw 
flowid :2

On 10/12/12 12:11 PM, Anatoly Muliarski wrote:
> IMHO, the problem lies in next - incoming packets on eth0.2 have no
> mark and therefore they cannot be forwarded into ifb0 on the current
> classification scheme. You should redirect all incoming packets( or
> use non-mark redirection criteria ) into ifb0 and then classify and
> shape them on ifb0 device.
>
> To more clarify the way of a packet look at
> http://en.wikipedia.org/wiki/File:Netfilter-packet-flow.svg
>
> 2012/12/10, Jack Bates <uo4zau@nottheoilrig.com>:
>> Thank you for this clarification, and your assumptions are correct: I'm
>> trying to prioritize incoming and outgoing traffic on eth0.2, ifb0 is
>> for shaping the incoming traffic.
>>
>> But I'm still failing to filter incoming traffic based on the restored
>> mark: statistics on the ifb0 qdiscs and classes remain zero. Filtering
>> outgoing traffic is working as expected, statistics on the eth0.2
>> classes are incrementing as expected, which I think means that the
>> PREROUTING --restore-mark target is working?
>>
>> Do you have any other suggestions how to investigate why incoming
>> traffic isn't getting filtered? or is there another way to classify
>> response traffic from the origin server based on the TOS/DSCP field of
>> the request from the proxy server? Are there any more details that would
>> help explain the problem?
>>
>> Here is the clarified script:
>>
>>      1: iptables -N NEW_CONN -t mangle
>>      2: iptables -A NEW_CONN -t mangle -m tos --tos 12 -j MARK --set-mark 1
>>      3: iptables -A NEW_CONN -t mangle -m tos --tos 28 -j MARK --set-mark 2
>>      4: iptables -A NEW_CONN -t mangle -m mark --mark 0 -j MARK --set-mark 3
>>      5: iptables -A NEW_CONN -t mangle -j CONNMARK --save-mark
>>      6:
>>      7: iptables -A PREROUTING -t mangle -m conntrack --ctstate INVALID
>> -j DROP
>>      8: iptables -A PREROUTING -t mangle -m conntrack --ctstate
>> ESTABLISHED -j CONNMARK --restore-mark
>>      9: iptables -A PREROUTING -t mangle -m conntrack --ctstate
>> NEW,RELATED -j NEW_CONN
>>     10:
>>     11: ifconfig ifb0 up
>>     12:
>>     13: insmod sch_ingress
>>     14: tc qdisc add dev eth0.2 ingress
>>     15:
>>     16: # Classify response traffic from origin server based on restored
>> mark
>>     17: insmod cls_fw
>>     18: insmod act_mirred
>>     19: tc filter add dev eth0.2 parent ffff: protocol ip pref 1 handle 1
>> fw flowid 2:1 action mirred egress redirect dev ifb0
>>     20: tc filter add dev eth0.2 parent ffff: protocol ip pref 1 handle 2
>> fw flowid 2:3 action mirred egress redirect dev ifb0
>>     21: tc filter add dev eth0.2 parent ffff: protocol ip pref 1 handle 3
>> fw flowid 2:2 action mirred egress redirect dev ifb0
>>     22:
>>     23: # Limit up and downstream to 1mbit, to "own" the queue
>>     24: insmod sch_tbf
>>     25: tc qdisc add dev eth0.2 root handle 1 tbf rate 1mbit burst 5k
>> latency 70ms
>>     26: tc qdisc add dev ifb0 root handle 1 tbf rate 1mbit burst 5k
>> latency 70ms
>>     27:
>>     28: # High, low, and "other" priority
>>     29: insmod sch_prio
>>     30: tc qdisc add dev eth0.2 parent 1: handle 2 prio
>>     31: tc qdisc add dev ifb0 parent 1: handle 2 prio
>>     32:
>>     33: # Classify request traffic from proxy server based on mark
>>     34: tc filter add dev eth0.2 parent 2: protocol ip pref 1 handle 1 fw
>> flowid :1
>>     35: tc filter add dev eth0.2 parent 2: protocol ip pref 1 handle 2 fw
>> flowid :3
>>     36: tc filter add dev eth0.2 parent 2: protocol ip pref 1 handle 3 fw
>> flowid :2
>>
>> On 05/12/12 08:18 PM, Anatoly Muliarski wrote:
>>> IMHO, there is a bit of mess in your scheme.
>>> Let's clarify it :)
>>>
>>> If my assumptions are right then:
>>> 1. eth0.2 is the external shaper interface.
>>> 2. ifb0 is using for shaping incoming traffic on this interface.
>>>
>>> To set marks for the traffic you should use something like these:
>>>
>>> iptables -t manlge -N NEW_CONN
>>> iptables -t mangle -A NEW_CONN -m tos --tos 12 -j MARK --set-mark 1
>>> iptables -t mangle -A NEW_CONN -m tos --tos 28 -j MARK --set-mark 2
>>> # to mark all the unmatched traffic and treat it separately in your
>>> shaper
>>> iptables -t mangle -A NEW_CONN -m mark --mark 0  -j MARK --set-mark 3
>>> iptables -t mangle -A NEW_CONN  -j CONNMARK --save-mark
>>>
>>> iptables -t mangle -A PREROUTING -m conntrack --cstate INVALID -j DROP
>>> iptables -t mangle -A PREROUTING -m conntrack --cstate ESTABLISHED -j
>>> CONNMARK --restore-mark
>>> iptables -t mangle -A PREROUTING -m conntrack --cstate NEW,RELATED -g
>>> NEW_CONN
>>>
>>>
>>> 2012/12/5, Jack Bates <uo4zau@nottheoilrig.com>:
>>>> Thanks for this help, I think the marks are getting saved/restored
>>>> successfully, and I can successfully classify request traffic from the
>>>> proxy server based on the mark (statistics on the eth0.2 classes are
>>>> incrementing) but response traffic from the origin server isn't getting
>>>> classified yet (statistics on the ifb0 classes remain zero).
>>>>
>>>> Here is my complete work in progress script. How can I investigate why
>>>> response traffic isn't getting classified? Are there any more details
>>>> that would help explain why not?
>>>>
>>>> How can I classify traffic based on the restored mark? or is there
>>>> another way to classify response traffic from the origin server based on
>>>> the TOS/DSCP field of the request?
>>>>
>>>>       1: # Save/restore connection mark based on TOS field
>>>>       2: iptables -A PREROUTING -t mangle -j CONNMARK --restore-mark
>>>>       3: iptables -A PREROUTING -t mangle -m tos --tos 12 -j MARK
>>>> --set-mark
>>>> 1
>>>>       4: iptables -A PREROUTING -t mangle -m tos --tos 28 -j MARK
>>>> --set-mark
>>>> 2
>>>>       5: iptables -A PREROUTING -t mangle -j CONNMARK --save-mark
>>>>       6:
>>>>       7: ifconfig ifb0 up
>>>>       8:
>>>>       9: insmod sch_ingress
>>>>      10: tc qdisc add dev eth0.2 ingress
>>>>      11:
>>>>      12: # Classify response traffic from origin server based on restored
>>>> mark
>>>>      13: insmod cls_fw
>>>>      14: insmod act_mirred
>>>>      15: tc filter add dev eth0.2 parent ffff: protocol ip pref 1 handle
>>>> 1
>>>> fw flowid 2:1 action mirred egress redirect dev ifb0
>>>>      16: tc filter add dev eth0.2 parent ffff: protocol ip pref 1 handle
>>>> 2
>>>> fw flowid 2:3 action mirred egress redirect dev ifb0
>>>>      17:
>>>>      18: # Send unmarked traffic to class 2:2
>>>>      19: insmod cls_u32
>>>>      20: tc filter add dev eth0.2 parent ffff: protocol ip pref 1 u32
>>>> match u32 0 0 flowid 2:2 action mirred egress redirect dev ifb0
>>>>      21:
>>>>      22: # Limit up and downstream to 1mbit, to "own" the queue
>>>>      23: insmod sch_tbf
>>>>      24: tc qdisc add dev eth0.2 root handle 1 tbf rate 1mbit burst 5k
>>>> latency 70ms
>>>>      25: tc qdisc add dev ifb0 root handle 1 tbf rate 1mbit burst 5k
>>>> latency 70ms
>>>>      26:
>>>>      27: # High, low, and "other" priority
>>>>      28: insmod sch_prio
>>>>      29: tc qdisc add dev eth0.2 parent 1: handle 2 prio
>>>>      30: tc qdisc add dev ifb0 parent 1: handle 2 prio
>>>>      31:
>>>>      32: # Classify request traffic from proxy server based on mark
>>>>      33: tc filter add dev eth0.2 parent 2: protocol ip pref 1 handle 1
>>>> fw
>>>> flowid :1
>>>>      34: tc filter add dev eth0.2 parent 2: protocol ip pref 1 handle 2
>>>> fw
>>>> flowid :3
>>>>
>>>> On 05/12/12 01:12 AM, Eliezer Croitoru wrote:
>>>>> Thanks Anatoly,
>>>>>
>>>>> My idea was that based on a mark he will jump the packets to another
>>>>> table which HE will mark TOS/DSCP.
>>>>>
>>>>> Eliezer
>>>>>
>>>>> On 12/5/2012 4:39 AM, Anatoly Muliarski wrote:
>>>>>> Hi Jack,
>>>>>>
>>>>>> --restore-mark should be used for existing connections and to mark new
>>>>>> ones you may use something like that:
>>>>>>
>>>>>> iptables -t mangle -A PREROUTING -i eth0.2 -m tos --tos 1 -m conntrack
>>>>>> --cstate NEW,RELATED -j MARK --set-mark 1
>>>>>>
>>>>>> The main idea consists in marking packets on the physical input
>>>>>> interface and shaping them on ifb0( where they arrive already marked
>>>>>> ).
>>>>>> iptables' packet marks exist only in memory of one computer, TOS/DSCP
>>>>>> may be used for transmitting a map of them to another one.
>>>>>> BTW, use --restore-mark on the output interface of your shaper too.
>>>>>>
>>>>>> 2012/12/3, Jack Bates <uo4zau@nottheoilrig.com>:
>>>>>>> Thanks a lot for your help, how can I evaluate --restore-mark before
>>>>>>> I
>>>>>>> classify and shape response traffic from the origin server?
>>>>>>>
>>>>>>> I think you mean something like:
>>>>>>>
>>>>>>>       # Copy ctmart to nfmark (e.g. 1, 2)
>>>>>>>       iptables -A PREROUTING -t mangle -i eth0.2 -j CONNMARK
>>>>>>> --restore-mark
>>>>>>>
>>>>>>>       # Classify by nfmark (e.g. 1, 2), send unmarked traffic to class
>>>>>>> 2:2
>>>>>>>       tc filter add dev eth0.2 parent ffff: protocol ip handle 1 fw
>>>>>>> flowid
>>>>>>> 2:1 action mirred egress redirect dev ifb0
>>>>>>>       tc filter add dev eth0.2 parent ffff: protocol ip handle 2 fw
>>>>>>> flowid
>>>>>>> 2:3 action mirred egress redirect dev ifb0
>>>>>>>       tc filter add dev eth0.2 parent ffff: protocol ip u32 match u32 0
>>>>>>> 0
>>>>>>> flowid 2:2 action mirred egress redirect dev ifb0
>>>>>>>
>>>>>>> Just how can I get --restore-mark to evaluate before tc filter?
>>>>>>>
>>>>>>> Another way I can imagine is with the CLASSIFY target:
>>>>>>>
>>>>>>>       # Send unmarked traffic to class 2:2
>>>>>>>       iptables -A PREROUTING -t mangle -i eth0.2 -m connmark --mark 1
>>>>>>> -j
>>>>>>> CLASSIFY 2:1
>>>>>>>       iptables -A PREROUTING -t mangle -i eth0.2 -m connmark --mark 2
>>>>>>> -j
>>>>>>> CLASSIFY 2:3
>>>>>>>       iptables -A PREROUTING -t mangle -i eth0.2 -j CLASSIFY 2:2
>>>>>>>
>>>>>>> But I have the same challenge, how can I evaluate the CLASSIFY target
>>>>>>> before I shape traffic?
>>>>>>>
>>>>>>> Or is there another way to classify and shape response traffic from
>>>>>>> the
>>>>>>> origin server based on the TOS/DSCP field of the request?
>>>>>>>
>>>>>>> On 03/12/12 03:52 AM, Eliezer Croitoru wrote:
>>>>>>>> You use iptables mark + restore mark based on connection tracking.
>>>>>>>> you can mark the TOS on the outgoing postrouting table.
>>>>>>>> you can take a look at the iptabes man pages:
>>>>>>>> http://ipset.netfilter.org/iptables.man.html
>>>>>>>> which has --restore-mark  exaple.
>>>>>>>>
>>>>>>>> Eliezer
>>>>>>>>
>>>>>>>> On 12/3/2012 10:43 AM, Jack Bates wrote:
>>>>>>>>> I can imagine a couple ways of classifying traffic from our proxy
>>>>>>>>> server
>>>>>>>>> based on the TOS/DSCP field, and also how to set the connection
>>>>>>>>> mark
>>>>>>>>> based on this field. But how do I classify and shape response
>>>>>>>>> traffic
>>>>>>>>> from the origin server based on the connection mark?
>>>>>>> --
>>>>>>> To unsubscribe from this list: send the line "unsubscribe netfilter"
>>>>>>> in
>>>>>>> the body of a message to majordomo@vger.kernel.org
>>>>>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>>>>>>
>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>>
>>
>
>