Add srcPodIP field in Traceflow observations #6247
Conversation
Atish-iaf
force-pushed
the
add-srcPodIP-traceflow
branch
from
54003b1 to
f6336a8
Compare
rajnkamr
changed the title
Atish-iaf
force-pushed
the
add-srcPodIP-traceflow
branch
from
f6336a8 to
3b6a6ba
Compare
|
/test-ipv6-e2e |
1 similar comment
|
/test-ipv6-e2e |
|
/test-kind-ipv6-only-e2e |
Atish-iaf
force-pushed
the
add-srcPodIP-traceflow
branch
from
3b6a6ba to
8449791
Compare
|
/test-kind-ipv6-only-e2e |
|
/test-ipv6-e2e |
|
/test-ipv6-only-e2e |
|
/test-kind-ipv6-only-e2e |
1 similar comment
|
/test-kind-ipv6-only-e2e |
Atish-iaf
force-pushed
the
add-srcPodIP-traceflow
branch
from
8449791 to
175105d
Compare
|
/test-kind-ipv6-only-e2e |
| @@ -173,6 +173,12 @@ func (c *Controller) parsePacketIn(pktIn *ofctrl.PacketIn) (*crdv1beta1.Traceflo | |||
| ob := new(crdv1beta1.Observation) | |||
| ob.Component = crdv1beta1.ComponentSpoofGuard | |||
| ob.Action = crdv1beta1.ActionForwarded | |||
| // ctNwSrc is invalid incase of ICMP6. | |||
There was a problem hiding this comment.
I don't understand what this comment is trying to say here, it really needs to include more information.
There was a problem hiding this comment.
ctNwSrc doesn't contain original src ip incase of icmp6 packet. It should contain but don't know why it always have :: incase of icmp6. It contains original src ip in all other cases(tcp/udp/ipv4/ipv6). Maybe some limitation from OVS ?
We don't use ipSrc directly here because ipSrc and ctNwSrc are different incase of hairpin.
Incase of icmp6, even if ctNwSrc is invalid, we can use ipSrc as hairpin is not applicable in icmp6, so ipSrc contains src pod IP.
There was a problem hiding this comment.
Thanks for the information.
ctNwSrc doesn't contain original src ip incase of icmp6 packet
What kind of ICMPv6 packet specifically are you referring to, and how did you observe this condition? Was it because of an e2e test?
ICMPv6 is very complex, and some packets use multicast, and it's possible that conntrack handles different ICMPv6 packets differently.
I'm tagging @wenyingd as well.
There was a problem hiding this comment.
how did you observe this condition? Was it because of an e2e test?
Yes, all traceflow ICMPv6 testcases(using icmpv6 protocol 58) failed in dual-stack and ipv6-only e2e kind jobs.
What kind of ICMPv6 packet specifically are you referring to
Traceflow IPv6 packets with source as pod and destination an IPv6 address using ICMPv6 protocol(58). I modified the code to read related fields from traceflow packet and logged it.
For ICMP in IPv6 antctl tf -S pod2 -D 2001:4860:4860::8888 -f ipv6 following is the log :
I0502 05:44:19.714801 1 packetin.go:133] "IPv6 Packet" ipSrc="fd00:10:244:1::4"
I0502 05:44:19.714853 1 packetin.go:134] "IPv6 Packet" ipDst="2001:4860:4860::8888"
I0502 05:44:19.714884 1 packetin.go:135] "IPv6 Packet" ctNwSrc="::"
I0502 05:44:19.714959 1 packetin.go:136] "IPv6 Packet" ctNwDst="::8888"
I0502 05:44:19.715019 1 packetin.go:137] "IPv6 Packet" nextHeader(proto)=58
I0502 05:44:19.715042 1 packetin.go:457] "Packet-In" NXM_NX_CT_STATE value={"Data":33} mask={"Data":255}
I0502 05:44:19.715066 1 packetin.go:467] "Packet-In" NXM_NX_CT_ZONE=65510
I0502 05:44:19.715104 1 packetin.go:477] "Packet-In" NXM_NX_CT_MARK=3
I0502 05:44:19.715146 1 packetin.go:487] "Packet-In" NXM_NX_CT_NW_PROTO=58
For TCP in IPv6 antctl tf -S pod2 -D 2001:4860:4860::8888 -f ipv6,tcp,tcp_dst=80 following is the log :
I0502 05:44:59.722052 1 packetin.go:133] "IPv6 Packet" ipSrc="fd00:10:244:1::4"
I0502 05:44:59.722158 1 packetin.go:134] "IPv6 Packet" ipDst="2001:4860:4860::8888"
I0502 05:44:59.722247 1 packetin.go:135] "IPv6 Packet" ctNwSrc="fd00:10:244:1::4"
I0502 05:44:59.722396 1 packetin.go:136] "IPv6 Packet" ctNwDst="2001:4860:4860::8888"
I0502 05:44:59.722478 1 packetin.go:137] "IPv6 Packet" nextHeader(proto)=6
I0502 05:44:59.722529 1 packetin.go:457] "Packet-In" NXM_NX_CT_STATE value={"Data":33} mask={"Data":255}
I0502 05:44:59.722562 1 packetin.go:467] "Packet-In" NXM_NX_CT_ZONE=65510
I0502 05:44:59.722650 1 packetin.go:477] "Packet-In" NXM_NX_CT_MARK=3
I0502 05:44:59.722692 1 packetin.go:487] "Packet-In" NXM_NX_CT_NW_PROTO=6
There was a problem hiding this comment.
@Atish-iaf I was able to reproduce, thanks for the detailed information.
I do think this is an OVS-specific issue. I was able to reproduce with a "standalone" OVS bridge (i.e., without Antrea) and I reported the bug to the OVS community: openvswitch/ovs-issues#327
I suggest that you add a comment here explaining the situation, with a link to the OVS bug, and then we can proceed with this PR.
There was a problem hiding this comment.
Thanks @antoninbas
There was a problem hiding this comment.
This was indeed a bug in the OVS kernel module and has been patched in Linux: https://git.kernel.org/pub/scm/linux/kernel/git/netdev/net.git/commit/?id=7c988176b6c1. I don't know if it will be backported.
Atish-iaf
added
the
area/ops/traceflow
Atish-iaf
force-pushed
the
add-srcPodIP-traceflow
branch
from
175105d to
f85a26e
Compare
|
/test-kind-ipv6-only-e2e |
|
/test-kind-ipv6-only-e2e |
1 similar comment
|
/test-kind-ipv6-only-e2e |
Atish-iaf
force-pushed
the
add-srcPodIP-traceflow
branch
from
f85a26e to
94c63ae
Compare
Signed-off-by: Kumar Atish <kumar.atish@broadcom.com>
Atish-iaf
force-pushed
the
add-srcPodIP-traceflow
branch
from
94c63ae to
006297f
Compare
|
/test-all |
|
/test-kind-ipv6-e2e |
|
/test-kind-ipv6-only-e2e |
1 similar comment
|
/test-kind-ipv6-only-e2e |
antoninbas
added
the
action/release-note
#6099