tinc - Feb 2017 - Tinc hangs if started with network disconnected

On my system, tinc is started on bootup.
If the Network cable is not connected, tincd hangs forever after these messages:

2017-02-13 11:34:01 tinc.XXX[403]: tincd 1.0.31 starting, debug level 5
2017-02-13 11:34:01 tinc.XXX[403]: /dev/net/tun is a Linux tun/tap device (tun
mode)
2017-02-13 11:34:01 tinc.XXX[403]: Executing script tinc-up
2017-02-13 11:34:02 tinc.XXX[403]: Listening on 0.0.0.0 port 655
2017-02-13 11:34:02 tinc.XXX[403]: Ready
2017-02-13 11:34:02 tinc.XXX[403]: Error looking up XXX.net port 2300: Temporary
failure in name resolution
2017-02-13 11:34:02 tinc.XXX[403]: Could not set up a meta connection to XXXVPN
2017-02-13 11:34:02 tinc.XXX[403]: Trying to re-establish outgoing connection in
5 seconds
2017-02-13 11:34:02 tinc.XXX[403]: Purging unreachable nodes

Even if I reconnect the interface, tinc does not continue.
If I restart the service manually, everything works fine, even with disconnected
network interface (it tries again and again until internet connection works).
I think, there is some kind of boot-time race condition.
It's the same problem with version 1.0.28 and 1.0.31.

Do you have an idea how to find/solve the problem?


Thanks,

Sandro

On Fri, Feb 17, 2017 at 09:55:26AM +0100, Sandro Stiller wrote:
> On my system, tinc is started on bootup.
> If the Network cable is not connected, tincd hangs forever after these
messages:
[...]
> 2017-02-13 11:34:02 tinc.XXX[403]: Error looking up XXX.net port 2300:
Temporary failure in name resolution
[...]
> Even if I reconnect the interface, tinc does not continue.
> If I restart the service manually, everything works fine, even with
disconnected network interface (it tries again and again until internet
connection works).
> I think, there is some kind of boot-time race condition.
> It's the same problem with version 1.0.28 and 1.0.31.
Does it not log anything anymore after those messages? Or does it
continue giving the "Error looking up..." message periodically?

In the first case, try sending tinc the ALRM and/or HUP signal (tincd -n
<netname> -kALRM). This should cause tinc to retry outgoing connections
immediately.

If that doesn't help, it could be that libc doesn't reload
/etc/resolv.conf after the network connection came up. Tinc has code to
force glibc to reload that file every time it wants to do a hostname
lookup. Did you compile tinc yourself? If so, can you send us a copy of
the config.h file?

In any case, which Linux distribution are you using?

-- 
Met vriendelijke groet / with kind regards,
     Guus Sliepen <guus at tinc-vpn.org>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: Digital signature
URL:
<http://www.tinc-vpn.org/pipermail/tinc/attachments/20170217/47acd0b9/attachment.sig>

Am 17.02.2017 um 10:59 schrieb Guus Sliepen:
> On Fri, Feb 17, 2017 at 09:55:26AM +0100, Sandro Stiller wrote:
> 
>> On my system, tinc is started on bootup.
>> If the Network cable is not connected, tincd hangs forever after these
messages:
> [...]
>> 2017-02-13 11:34:02 tinc.XXX[403]: Error looking up XXX.net port 2300:
Temporary failure in name resolution
> [...]
>> Even if I reconnect the interface, tinc does not continue.
>> If I restart the service manually, everything works fine, even with
disconnected network interface (it tries again and again until internet
connection works).
>> I think, there is some kind of boot-time race condition.
>> It's the same problem with version 1.0.28 and 1.0.31.
> 
> Does it not log anything anymore after those messages? Or does it
> continue giving the "Error looking up..." message periodically?
It stops working completely.
> 
> In the first case, try sending tinc the ALRM and/or HUP signal (tincd -n
> <netname> -kALRM). This should cause tinc to retry outgoing
connections
> immediately.
After that, it continues trying.

I think, I found the cause of the problem:
Is tinc using a monotonic / steady clock for timeouts or is it using the system
time?

If the system time is set during boot using the RTC (after the start of tincd),
tincd stops retrying.
I tested it by starting tincd with disconnected network, waited for 1 or 2
retries and then set the system clock to 1 hour earlier.
Tincd printed one more message and then stopped the connection attempts.
> Did you compile tinc yourself? If so, can you send us a copy of
> the config.h file?
I guess, it's not needed anymore, but here it is:
http://pastebin.com/JWbMGXvi
> In any case, which Linux distribution are you using?
It's an OpenEmbedded-based distribution  for embedded devices