ip_local_deliver_finish: proto 108 (IPComp) isn't netns-ready

ip_local_deliver_finish: proto 108 (IPComp) isn't netns-ready

[Dilip Daya]

Hi All,

LTP-network tests produced:

	"ip_local_deliver_finish: proto 108 (IPComp) isn't netns-ready"

Question:
Is the above a bug or feature (IPComp, mode: transport) not netns
enabled?

Environment:

* v3.1-stable kernel.

* Configured two network-namespaces via "ip netns add ...".
  on a single system: netns0 (assigned as server) and
  netns1 (assigned as client).
  - assigned multiple physical NICs to netns0 and netns1.

* Entered netns1 (ip netns exec netns1 bash) to execute
  ltp-network suite of tests between netns1 and netns0.

* The following LTP tests produced messages:

  "kernel: ip_local_deliver_finish: proto 108 isn't netns-ready"

   - /usr/lib/ltp/testcases/bin/icmp4-multi-diffip06
   - /usr/lib/ltp/testcases/bin/tcp4-multi-diffip13
   - /usr/lib/ltp/testcases/bin/udp4-multi-diffip06
   - /usr/lib/ltp/testcases/bin/udp4-multi-diffnic06
   - /usr/lib/ltp/testcases/bin/udp4-multi-diffport06

* I also noticed missing ".netns_ok = 1" in:
  - http://lxr.linux.no/linux+v3.1/net/ipv4/ipcomp.c#L150


-DilipD.

Re: ip_local_deliver_finish: proto 108 (IPComp) isn't netns-ready

[Eric W. Biederman]

Dilip Daya <dilip.daya@hp.com> writes:

> Hi All,
>
> LTP-network tests produced:
>
> 	"ip_local_deliver_finish: proto 108 (IPComp) isn't netns-ready"
>
> Question:
> Is the above a bug or feature (IPComp, mode: transport) not netns
> enabled?

I would call it a missing feature.

> Environment:
>
> * v3.1-stable kernel.
>
> * Configured two network-namespaces via "ip netns add ...".
>   on a single system: netns0 (assigned as server) and
>   netns1 (assigned as client).
>   - assigned multiple physical NICs to netns0 and netns1.
>
> * Entered netns1 (ip netns exec netns1 bash) to execute
>   ltp-network suite of tests between netns1 and netns0.
>
> * The following LTP tests produced messages:
>
>   "kernel: ip_local_deliver_finish: proto 108 isn't netns-ready"
>
>    - /usr/lib/ltp/testcases/bin/icmp4-multi-diffip06
>    - /usr/lib/ltp/testcases/bin/tcp4-multi-diffip13
>    - /usr/lib/ltp/testcases/bin/udp4-multi-diffip06
>    - /usr/lib/ltp/testcases/bin/udp4-multi-diffnic06
>    - /usr/lib/ltp/testcases/bin/udp4-multi-diffport06
>
> * I also noticed missing ".netns_ok = 1" in:
>   - http://lxr.linux.no/linux+v3.1/net/ipv4/ipcomp.c#L150

The .netns_ok flag indicates that someone has audited the code and made
certain it all works with network namespaces.

Skimming the code I don't see anything that jumps out as me as
wrong.  And it looks like Alexy did the work in Jan of 2010.

   commit a92df2545402c1a08e7a158f4477a52dea0eeeed
   Author: Alexey Dobriyan <adobriyan@gmail.com>
   Date:   Mon Jan 25 10:38:34 2010 +0000
   
       netns xfrm: ipcomp support
       
       Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
       Signed-off-by: David S. Miller <davem@davemloft.net>

However it is clear that no one has actually tested ipcomp with
network namespaces.

If you have a clue what is going on with ip compression my
recommendation would be to add .netns_ok = 1.  Verify that the
everything works and send the patch.  You probably want to
verify and test ipv6 as well.  It really looks like the code
is fine and it just needs to be enabled.

Eric

Re: ip_local_deliver_finish: proto 108 (IPComp) isn't netns-ready

[Dilip Daya]

[-- Attachment #1: Type: text/plain, Size: 3957 bytes --]

Hi Eric,

> If you have a clue what is going on with ip compression my
> recommendation would be to add .netns_ok = 1.  Verify that the
> everything works and send the patch.  You probably want to
> verify and test ipv6 as well.  It really looks like the code
> is fine and it just needs to be enabled.

Thank you for the above feedback.

FYI..my results

Environment:
- v3.0.9-stable kernel with  ".netns_ok	= 1" applied to ipcomp.c
  (see attached patch)
- Same setup as described earlier for two netns stacks on single system.
- ltp-network test results:

- /usr/lib/ltp/testcases/bin/tcp[4|6]-multi-diffip13
  IPv4: no issues
  IPv6: no issues

- /usr/lib/ltp/testcases/bin/udp4-multi-diffip06
  IPv4 -> no issues
  IPv6 -> script issue (work-in-progress)

- /usr/lib/ltp/testcases/bin/udp[4|6]-multi-diffnic06
  IPv4: no issues
  IPv6:	no issues

- /usr/lib/ltp/testcases/bin/udp[4|6]-multi-diffport06
  IPv4 -> multiple messages:
    UDP: short packet: From 10.0.9.2:1025 1008/267 to 10.0.9.1:55575
    UDP: short packet: From 10.0.0.1:54544 253/38 to 10.0.0.2:1050
    UDP: short packet: From 10.0.0.1:21601 27128/1008 to 10.0.0.2:27137

  IPv6 -> multiple messages:
    UDPv6: short packet: From [fd00:1:1:1::2]:1028 242/122 to
[fd00:1:1:1::1]:38438
    UDPv6: short packet: From [fd00:1:1:1::2]:37972 21588/29 to
[fd00:1:1:1::1]:21588
    UDPv6: short packet: From [fd00:1:1:1::2]:1026 1008/898 to
[fd00:1:1:1::1]:55703

- /usr/lib/ltp/testcases/bin/icmp4-multi-diffip06
  IPv4 -> no issues
  IPv6 -> multiple messages:
    ICMPv6 checksum failed [fd00:0001:0000:0003:0000:0000:0000:0001 >
fd00:0001:0000:0003:0000:0000:0000:0002]
				

...continuing with debug.


--DilipD.

On Wed, 2011-11-09 at 22:38 +0000, Eric W. Biederman wrote:
> Dilip Daya <dilip.daya@hp.com> writes:
> 
> > Hi All,
> >
> > LTP-network tests produced:
> >
> > 	"ip_local_deliver_finish: proto 108 (IPComp) isn't netns-ready"
> >
> > Question:
> > Is the above a bug or feature (IPComp, mode: transport) not netns
> > enabled?
> 
> I would call it a missing feature.
> 
> > Environment:
> >
> > * v3.1-stable kernel.
> >
> > * Configured two network-namespaces via "ip netns add ...".
> >   on a single system: netns0 (assigned as server) and
> >   netns1 (assigned as client).
> >   - assigned multiple physical NICs to netns0 and netns1.
> >
> > * Entered netns1 (ip netns exec netns1 bash) to execute
> >   ltp-network suite of tests between netns1 and netns0.
> >
> > * The following LTP tests produced messages:
> >
> >   "kernel: ip_local_deliver_finish: proto 108 isn't netns-ready"
> >
> >    - /usr/lib/ltp/testcases/bin/icmp4-multi-diffip06
> >    - /usr/lib/ltp/testcases/bin/tcp4-multi-diffip13
> >    - /usr/lib/ltp/testcases/bin/udp4-multi-diffip06
> >    - /usr/lib/ltp/testcases/bin/udp4-multi-diffnic06
> >    - /usr/lib/ltp/testcases/bin/udp4-multi-diffport06
> >
> > * I also noticed missing ".netns_ok = 1" in:
> >   - http://lxr.linux.no/linux+v3.1/net/ipv4/ipcomp.c#L150
> 
> The .netns_ok flag indicates that someone has audited the code and made
> certain it all works with network namespaces.
> 
> Skimming the code I don't see anything that jumps out as me as
> wrong.  And it looks like Alexy did the work in Jan of 2010.
> 
>    commit a92df2545402c1a08e7a158f4477a52dea0eeeed
>    Author: Alexey Dobriyan <adobriyan@gmail.com>
>    Date:   Mon Jan 25 10:38:34 2010 +0000
>    
>        netns xfrm: ipcomp support
>        
>        Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
>        Signed-off-by: David S. Miller <davem@davemloft.net>
> 
> However it is clear that no one has actually tested ipcomp with
> network namespaces.
> 
> If you have a clue what is going on with ip compression my
> recommendation would be to add .netns_ok = 1.  Verify that the
> everything works and send the patch.  You probably want to
> verify and test ipv6 as well.  It really looks like the code
> is fine and it just needs to be enabled.
> 
> Eric

[-- Attachment #2: ipcomp_netns.patch --]
[-- Type: text/x-patch, Size: 310 bytes --]

--- a/ipv4/ipcomp.c	2011-11-11 13:12:24.000000000 -0500
+++ b/net/ipv4/ipcomp.c	2011-11-08 09:12:37.000000000 -0500
@@ -151,6 +151,7 @@ static const struct net_protocol ipcomp4
 	.handler	=	xfrm4_rcv,
 	.err_handler	=	ipcomp4_err,
 	.no_policy	=	1,
+	.netns_ok		= 1,
 };

 static int __init ipcomp4_init(void)