Bug 147656

Summary: very long ipv6 time out if there is no ipv6 route to host
Product: [openSUSE] SUSE Linux 10.1 Reporter: Johannes Meixner <jsmeix>
Component: KernelAssignee: Jiri Bohac <jbohac>
Status: RESOLVED DUPLICATE QA Contact: E-mail List <qa-bugs>
Severity: Minor    
Priority: P5 - None    
Version: Beta 3   
Target Milestone: ---   
Hardware: x86-64   
OS: Linux   
Whiteboard:
Found By: Development Services Priority:
Business Priority: Blocker: ---
Marketing QA Status: --- IT Deployment: ---

Description Johannes Meixner 2006-02-02 15:17:05 UTC
In 10.1 beta 3 (using kernel-smp-2.6.16_rc1_git3-7)
there is a very long timeout (a few minutes)
if there is no ipv6 route to host and then it falls back to ipv4:
---------------------------------------------------------------------
# telnet nelson.suse.de 1234
Trying 2001:780:101:a00:280:c8ff:fe95:5a20...
telnet: connect to address 2001:780:101:a00:280:c8ff:fe95:5a20:
        Connection timed out
Trying 10.10.11.4...
Connected to nelson.suse.de.
Escape character is '^]'.
---------------------------------------------------------------------

In 10.1 beta 2 (different machine using kernel-default-2.6.16_rc1_git3-4)
it works o.k. because there is only a short delay (I guess less than
one second) if there is no ipv6 route to host:
---------------------------------------------------------------------
# telnet nelson.suse.de 1235
Trying 2001:780:101:a00:280:c8ff:fe95:5a20...
telnet: connect to address 2001:780:101:a00:280:c8ff:fe95:5a20:
        No route to host
Trying 10.10.11.4...
Connected to nelson.suse.de.
Escape character is '^]'.
---------------------------------------------------------------------

On nelson I used "netcat -l -p 1234" and "netcat -l -p 1235"
to have something listening.

The nameserver is wotan.suse.de (10.10.0.1) and it has somehow
stored a wrong ipv6 address for nelson because:
----------------------------------------------------------------------
nelson:~ # ifconfig
eth0  Link encap:Ethernet  HWaddr 00:0C:6E:DE:64:45  
      inet addr:10.10.11.4  Bcast:10.10.255.255  Mask:255.255.0.0
      inet6 addr: 2001:780:101:a00:20c:6eff:fede:6445/64 Scope:Global
      inet6 addr: fe80::20c:6eff:fede:6445/64 Scope:Link
----------------------------------------------------------------------

Even if the nameserver setting is false, it would be nice
to have again the short delay if there is no route to host.
Comment 1 Olaf Kirch 2006-02-02 16:20:09 UTC
Jiri, could you look into this please? Thanks!
Comment 2 Jiri Bohac 2006-02-02 18:51:19 UTC
Probably (99%) duplicate of #147075.

kernel-smp-2.6.16_rc1_git3-7 was built with the patch that breaks IPv6 already included and not yet removed. 

I suppose dmesg says something like:
xfrm_lookup: IPv4 route is stale (obsolete=4294967295, loops=0)
 [<c0279deb>] xfrm_lookup+0x403/0x481
 ....

Does it?
Comment 3 Johannes Meixner 2006-02-03 10:20:50 UTC
Yes.
After a reboot and one telnet test dmesg shows the line
-----------------------------------------------------------------------------
xfrm_lookup: IPv4 route is stale (obsolete=4294967295, loops=0)
-----------------------------------------------------------------------------
34 times.
This happens on caps.suse.de (10.10.11.93).
Feel free to log in but please tell me if you have to reboot it.

Comment 4 Jiri Bohac 2006-02-03 22:11:43 UTC
Ok, this is clearly a duplicate of #147075.

The problem has been fixed in recent KOTDs, so try one if this bug causes you pain.


*** This bug has been marked as a duplicate of 147075 ***