[Cake] Cake with Deep Packet Inspection

Outback Dingo outbackdingo at gmail.com
Fri Nov 11 09:08:52 EST 2016


On Fri, Nov 11, 2016 at 9:04 PM, Noah Causin <n0manletter at gmail.com> wrote:
> Do you build your firmware from the git repository?
>
>

Yes it is built from a recent trunk, as of yesterday


>
> On 11/10/2016 11:28 PM, Outback Dingo wrote:
>>
>> On Fri, Nov 11, 2016 at 10:52 AM, Outback Dingo <outbackdingo at gmail.com>
>> wrote:
>>>
>>> On Fri, Nov 11, 2016 at 10:42 AM, Noah Causin <n0manletter at gmail.com>
>>> wrote:
>>>>
>>>> I took an extra router I had and bridged two ports on the router's
>>>> switch,
>>>> so they just pass traffic. (eth0.2 and eth0.3)
>>>>
>>>> Clients >-[Main Router] --[Extra Router]--[Cable-Modem]
>>>>
>>>> The extra router is passive.  It acts like it's part of the Ethernet
>>>> cable
>>>> between the main router and cable modem.  It does not interfere.
>>>>
>>>> The extra router needs these packages:
>>>>
>>>> kmod-ebtables, kmod-ebtables-ipv4, kmod-crypto-pcompress and the two
>>>> packages compiled from the Makefile I showed below (iptables-mod-ndpi
>>>> and
>>>> iptables-mod-ndpi)
>>>>
>>>> I edited /etc/sysctl.conf and change the last two lines to this:
>>>>
>>>> net.bridge.bridge-nf-call-ip6tables=1
>>>> net.bridge.bridge-nf-call-iptables=1
>>>>
>>>> This enables bridge firewalling, so the traffic between the two ports
>>>> can be
>>>> marked.
>>>>
>>>> I then added firewall rules to LuCIs custom firewall rules tab.
>>>>
>>>> Example Rules, modify classes as desired:
>>>>
>>>> iptables -t mangle -A FORWARD -m ndpi --steam -m mac ! --mac-source Your
>>>> Main Router's Mac Address -j DSCP --set-dscp-class cs0
>>>> iptables -t mangle -A FORWARD -m ndpi --youtube -m mac ! --mac-source
>>>> Your
>>>> Main Router's Mac Address -j DSCP --set-dscp-class cs2
>>>>
>>>> iptables -t mangle -A FORWARD -m ndpi --netflix -m mac ! --mac-source
>>>> Your
>>>> Main Router's Mac Address -j DSCP --set-dscp-class cs3
>>>> iptables -t mangle -A FORWARD -m ndpi --skype -m mac ! --mac-source Your
>>>> Main Router's Mac Address -j DSCP --set-dscp-class cs4
>>>>
>>>> Have your main router use some form of DiffServ for both upload and
>>>> download
>>>> on its WAN interface.  For upload traffic, you just need the two
>>>> packages
>>>> from the makefile to be installed on the main router and create firewall
>>>> rules like this:
>>>>
>>>> iptables -t mangle -A FORWARD -o eth2 -m ndpi --netflix -j DSCP
>>>> --set-dscp-class cs3
>>>> iptables -t mangle -A FORWARD -o eth2 -m ndpi --skype -j DSCP
>>>> --set-dscp-class cs4
>>>>
>>>> If you need help building a custom firmware image, just let me know.
>>>>
>>> Ok got it, its in-line.... right now im fighting the GFW of China
>>> using shadowsocks and chinadns so im pretty customized already,
>>> i was having issues with sqm and kernel segfaults on 4.4.30 ill build
>>> a cake and DPI image and see how that goes, maybe Ill try this
>>> after i deem sqm is stable with cake on this build. It a ZBT-WG3526
>>> router... so its ralink based
>>
>> like i stated on LEDE major issues when sqm is enabled.... even with
>> cake... is nobody else seeing these issues?
>>
>> [  569.020000]  1-...: (6 GPs behind) idle=236/0/0 softirq=50607/50613
>> fqs=1
>> [  569.020000]  (detected by 0, t=6002 jiffies, g=5799, c=5798, q=573)
>> [  569.020000] Task dump for CPU 1:
>> [  569.020000] swapper/1       R running      0     0      1 0x00100000
>> [  569.020000] Stack : 00000000 00000001 00000015 00000000 00000000
>> 00000001 8045a2a4 80410000
>> [  569.020000]    8041275c 00000001 00000001 80412540 80412724
>> 80410000 00000000 800132e8
>> [  569.020000]    1100fc03 00000001 8fc60000 8fc61ec0 80410000
>> 8005c870 1100fc03 00000001
>> [  569.020000]    00000000 80410000 8045a2a4 8005c868 80410000
>> 8001ade0 1100fc03 00000000
>> [  569.020000]    00000004 804124a0 000000a0 8001ade8 fe9ff9db
>> aefffbff fff77fe7 dfffffef
>> [  569.020000]    ...
>> [  569.020000] Call Trace:
>> [  569.020000] [<8000bba8>] __schedule+0x574/0x758
>> [  569.020000] [<800132e8>] r4k_wait_irqoff+0x0/0x20
>> [  569.020000]
>> [  569.020000] rcu_sched kthread starved for 6001 jiffies! g5799 c5798
>> f0x0 s3 ->state=0x1
>> [  629.130000] INFO: rcu_sched detected stalls on CPUs/tasks:
>> [  629.130000]  1-...: (8 GPs behind) idle=756/0/0 softirq=50607/50613
>> fqs=1
>> [  629.130000]  (detected by 0, t=6002 jiffies, g=5801, c=5800, q=653)
>> [  629.130000] Task dump for CPU 1:
>> [  629.130000] swapper/1       R running      0     0      1 0x00100000
>> [  629.130000] Stack : 00000000 bfd47c67 00000094 ffffffff 00000081
>> 00000011 8045a2a4 80410000
>> [  629.130000]    8041275c 00000001 00000001 80412540 80412724
>> 80410000 000010d9 800132e8
>> [  629.130000]    1100fc03 00000001 8fc60000 8fc61ec0 80410000
>> 8005c870 1100fc03 00000001
>> [  629.130000]    00000000 80410000 8045a2a4 8005c868 80410000
>> 8001ade0 1100fc03 00000000
>> [  629.130000]    00000004 804124a0 000000a0 8001ade8 fe9ff9db
>> aefffbff fff77fe7 dfffffef
>> [  629.130000]    ...
>> [  629.130000] Call Trace:
>> [  629.130000] [<8000bba8>] __schedule+0x574/0x758
>> [  629.130000] [<800132e8>] r4k_wait_irqoff+0x0/0x20
>> [  629.130000]
>> [  629.130000] rcu_sched kthread starved for 6001 jiffies! g5801 c5800
>> f0x0 s3 ->state=0x1
>> [  692.940000] INFO: rcu_sched detected stalls on CPUs/tasks:
>> [  692.940000]  1-...: (0 ticks this GP) idle=4c2/0/0
>> softirq=50613/50613 fqs=0
>> [  692.940000]  2-...: (25 GPs behind) idle=55a/0/0 softirq=28459/28460
>> fqs=0
>> [  692.940000]  3-...: (25 GPs behind) idle=5c2/0/0 softirq=30711/30982
>> fqs=0
>> [  692.940000]  (detected by 0, t=6002 jiffies, g=5824, c=5823, q=3190)
>> [  692.940000] Task dump for CPU 1:
>> [  692.940000] swapper/1       R running      0     0      1 0x00100000
>> [  692.940000] Stack : 00000000 3646fb94 000000af ffffffff 0000008f
>> d2f1a9fc 8045a2a4 80410000
>> [  692.940000]    8041275c 00000001 00000001 80412540 80412724
>> 80410000 000010d9 800132e8
>> [  692.940000]    00000000 00000001 8fc60000 8fc61ec0 80410000
>> 8005c870 1100fc03 00000001
>> [  692.940000]    00000000 80410000 8045a2a4 8005c868 80410000
>> 8001ade0 1100fc03 00000000
>> [  692.940000]    00000004 804124a0 000000a0 8001ade8 fe9ff9db
>> aefffbff fff77fe7 dfffffef
>> [  692.940000]    ...
>> [  692.940000] Call Trace:
>> [  692.940000] [<8000bba8>] __schedule+0x574/0x758
>> [  692.940000] [<800132e8>] r4k_wait_irqoff+0x0/0x20
>> [  692.940000]
>> [  692.940000] Task dump for CPU 2:
>> [  692.940000] swapper/2       R running      0     0      1 0x00100000
>> [  692.940000] Stack : 00000000 00000001 00000010 00000000 00000000
>> 00010001 8045a2a4 80410000
>> [  692.940000]    8041275c 00000001 00000000 804125e0 80412724
>> 80410000 00000000 800132e8
>> [  692.940000]    1100fc03 00000002 8fc62000 8fc63ec0 80410000
>> 8005c870 1100fc03 00000002
>> [  692.940000]    00000000 80410000 8045a2a4 8005c868 80410000
>> 8001ade0 1100fc03 00000000
>> [  692.940000]    00000004 804124a0 000000a0 8001ade8 fbffbfdf
>> 7fff7b3b 7ffaeff8 67fbffff
>> [  692.940000]    ...
>> [  692.940000] Call Trace:
>> [  692.940000] [<8000bba8>] __schedule+0x574/0x758
>> [  692.940000] [<800132e8>] r4k_wait_irqoff+0x0/0x20
>> [  692.940000]
>> [  692.940000] Task dump for CPU 3:
>> [  692.940000] swapper/3       R running      0     0      1 0x00100000
>> [  692.940000] Stack : 00000000 a69c5765 000000a3 ffffffff 0000008f
>> 00000020 8045a2a4 80410000
>> [  692.940000]    8041275c 00000001 00000001 80412680 80412724
>> 80410000 000010d9 800132e8
>> [  692.940000]    1100fc03 00000003 8fc64000 8fc65ec0 80410000
>> 8005c870 1100fc03 00000003
>> [  692.940000]    00000000 80410000 8045a2a4 8005c868 80410000
>> 8001ade0 1100fc03 00000000
>> [  692.940000]    00000004 804124a0 000000a0 8001ade8 fefffbdb
>> aefffbfb fff77fe7 dfffffef
>> [  692.940000]    ...
>> [  692.940000] Call Trace:
>> [  692.940000] [<8000bba8>] __schedule+0x574/0x758
>> [  692.940000] [<800132e8>] r4k_wait_irqoff+0x0/0x20
>> [  692.940000]
>> [  692.940000] rcu_sched kthread starved for 6002 jiffies! g5824 c5823
>> f0x0 s3 ->state=0x1
>> root at lede:~# packet_write_wait: Connection to 192.168.20.1 port 22: Broken
>> pipe
>>
>>
>>>> On 11/10/2016 9:06 PM, Outback Dingo wrote:
>>>>
>>>> On Fri, Nov 11, 2016 at 9:55 AM, Noah Causin <n0manletter at gmail.com>
>>>> wrote:
>>>>
>>>> I finally got my project working.
>>>>
>>>> I integrated the NDPI Deep Packet Inspection engine into my LEDE build,
>>>> so I
>>>> could prioritize applications using Cake.
>>>>
>>>> http://www.ntop.org/products/deep-packet-inspection/ndpi/
>>>>
>>>> NDPI integrates into IPTables, which allows me to DSCP mark packets.
>>>> Cake
>>>> reads the DiffServ markings and puts the traffic into appropriate
>>>> classes.
>>>>
>>>> I found a Makefile which successfully compiles:
>>>>
>>>> https://github.com/981213/lede_src/blob/0d344bc2958838dcbc547a8f0a3d8842e6f6d2f8/package/my_package/ndpi-netfilter/Makefile
>>>>
>>>> The system works very well.  Steam traffic is deprioritized to allow
>>>> applications like YouTube, Netflix, and Skype to receive higher amounts
>>>> of
>>>> the available bandwidth.
>>>>
>>>> What I do for ingress is bridge two ports on an extra router, enable
>>>> bridge
>>>> firewalling, and create IPTables rules to mark downstream packets.  The
>>>> router I use is a D-Link DGL-5500, which is comparable to an Archer C7.
>>>>
>>>> IPv6 support is not available in this netfilter module, but the IPv4
>>>> support
>>>> is great.
>>>>
>>>> Noah Causin
>>>>
>>>> very nice.... however, can you describe a bit better how you
>>>> configured for ingress ? a second router?  configuration file ? your
>>>> high level description seems a bit confusing to me
>>>>
>>>>
>>>> _______________________________________________
>>>> Cake mailing list
>>>> Cake at lists.bufferbloat.net
>>>> https://lists.bufferbloat.net/listinfo/cake
>>>>
>>>>
>


More information about the Cake mailing list