IPv6 stateful testing

• Question regarding configuration for IPv6 stateful testing. Running version 2.27 using Mellanox Connect-X4 100Gps NICs
• Ran stateful tests against a firewall with IPv4 very successfully. The trex_cfg.yaml file contains:

- port_limit: 2

  version: 2

  interfaces: ['03:00.0', '03:00.1']

  port_info:

      - ip: 131.247.125.10

        default_gw: 131.247.125.11

      - ip: 131.247.126.10

        default_gw: 131.247.126.11

  platform:

      master_thread_id: 0

      latency_thread_id: 1

      dual_if:

        - socket: 0

          threads: [2,4,6,8,10,12,14,16,18]

 And  using the stock cap2/sfr_agg_tcp14_udp11_http200msec_new_high_new_nir_profile_ipg_mix.yaml file with only the client and server addresses modified to:

 The cli used for the tests was:

# ./t-rex-64 -f cap2/sfr_agg_tcp14_udp11_http200msec_new_high_new_nir_profile_ipg_mix.yaml -m 55 -d 600 -e -c 9

For the tests involved IPv6, modified the cap2/sfr_agg_tcp14_udp11_http200msec_new_high_new_nir_profile_ipg_mix.yaml with these additional lines:

  src_ipv6 : [0x2607,0xfe50,0x0000,0xf802,0x0000,0x0000]

  dst_ipv6 : [0x2607,0xfe50,0x0000,0xf803,0x0000,0x0000]

 And the cli used was:

# ./t-rex-64 -f cap2/sfr_agg_tcp14_udp11_http200msec_new_high_new_nir_profile_ipg_mix.yaml -m 55 -d 600 -e -c 9 --ipv6

 set up the firewall DUT with static routes and NDP entries for the IPv6 prefixes so that traffic would be properly forwarded (which it is based on captures I've done of the traffic).  Under v4 the '-e' flag made sure that the traffic sourced from NIC0 was from the client IP range and the traffic sourced from NIC1 was from the server range (so the firewall was happy with outside and inside addresses it saw). 

But, for the v6 testing, the source v6 addresses are keeping the 16.0.0.0/12 IPv4 encoded addresses coming from NIC0 as you can see below, but showing a mix of v6 prefixes coming from that NIC instead of just the 2607:fe50:0:f802::/64 prefix. Have configured as the source (this is a capture of packets transmitted by NIC0): 

62 0.205888089 2607:fe50:0:f802::1003:557a -> 2607:fe50:0:f803::3003:557a TCP 78 1025 > 443 [ACK] Seq=1 Ack=1 Win=32768 Len=0

63 0.205898872 2607:fe50:0:f803::1003:55b5 -> 2607:fe50:0:f802::3003:55b5 TCP 1538 [TCP segment of a reassembled PDU]

<-------Logs - Removed------------->

69 0.205926516 2607:fe50:0:f803::1008:55b1 -> 2607:fe50:0:f802::3008:55b1 TCP 1538 [TCP Previous segment not captured] [TCP segment of a reassem

Seems like the -e flag isn't using the src_ipv6 prefix for NIC0 for source addresses and flipping that to the dst_ipv6 prefix as the source addresses for NIC1 like it does for IPv4 testing. 

Did you test without -e CLI. Are your profile include UDP one packet. -e should not be use anymore given our Stateless support.

 Can you test it with out cap2/simple_http.yaml

 753 0.038209464 2607:fe50:0:f802::1006:32d3 -> 2607:fe50:0:f803::3006:32d3 UDP 76 Source port: 1032  Destination port: 10000 [ILLEGAL CHECKSUM (0)]

754 0.038220503 2607:fe50:0:f803::3006:232e -> 2607:fe50:0:f802::1006:232e TCP 98 1494 > 1039 [PSH, ACK] Seq=1 Ack=1 Win=32768 Len=20

 Also seeing a mix of client (:1006:) and server (:3006:) IPv4 encoded hosts but that doesn't really matter since it's the host portion.  The /64 prefix is the bigger issue since the firewall being tested sees the server-addressed prefix as a spoofed packet when coming from the client side of the firewall.

 Couldn't find a cap2/simple_http.yaml, but did find http_simple.yaml and tried that one (with the addition of these lines):

  src_ipv6 : [0x2607,0xfe50,0x0000,0xf802,0x0000,0x0000]

  dst_ipv6 : [0x2607,0xfe50,0x0000,0xf803,0x0000,0x0000]

 Using this CLI I'm still seeing mixed source prefixes:

# ./t-rex-64 -f cap2/http_simple.yaml -m 55 -d 600 --ipv6  -c 9

 763 2.826488587 2607:fe50:0:f803::3000:aaa9 -> 2607:fe50:0:f802::1000:ab TCP 1538 [TCP segment of a reassembled PDU]

764 2.826489107 2607:fe50:0:f802::1000:1f -> 2607:fe50:0:f803::3000:1c74 TCP 74 1024 > 80 [ACK] Seq=250 Ack=17521 Win=32768 Len=0 

 So, still seeing the mixed prefixes as well as the mixed host addresses without the -e option. 

As mentioned using a yaml file with UDP flows.  So  modified cap2/imix_fast_1g_100k_flows.yaml to include the v6 addresses and ran it with ./t-rex-64 -f cap2/imix_fast_1g_100k_flows.yaml -m 55 -d 600 --ipv6  -c 9.  That seemed to work just fine:

 49 6.381512150 2607:fe50:0:f802::100b:d04d -> 2607:fe50:0:f803::300b:d04d UDP 1534 Source port: 1024  Destination port: 2001 [ILLEGAL CHECKSUM (0)]

Both the prefix and the IPv4 encoded host portions were correct.  So is there a way to make this work for the TCP flows

# ./t-rex-64 -v

Version : v2.27  

DPDK version : DPDK 17.02.0-rc3  

User    : hhaim  

Date    : Jun 18 2017 , 10:54:07

Uuid    : 37c1c20c-53fb-11e7-8930-0006f62b3e88   

Git SHA : 7d0fb503142103a36968206bf9727e55144bef9c   

This bug was fixed in v2.24 see here

 https://trex-tgn.cisco.com/youtrack/issueMobile/trex-408 

Looking into the fix,

https://github.com/cisco-system-traffic-generator/trex-core/commit/0c40dbd93fe9d0aaed3af9b2a1fcd84c283c4aaf

Could you do  test

Try cap2/http_simple_ipv6.yaml and change default ipv6 src/dst in the file to your need.

 This file is used in our unit-test so it must work

 Just for information our forum

https://groups.google.com/forum/m/#!forum/trex-tgn

 Is more active with much more registered users, so you will probably get answer faster there.

 We are using the devNet mainly for jive blogs