[ntp:hackers] [ntp-dev] pls test new interface code

Frank Kardel kardel at ntp.org
Sun Jul 23 06:49:35 UTC 2006


Danny Mayer wrote:

>Frank Kardel wrote:
>  
>
>>One of the new major features in ntp-dev aside from autogen
>>is the dynamic interface code. This code solves the long standing
>>problem (bug #51) that ntpd would loose contact with the world
>>when the IP address changed (e.g.  ppp, "Zwangstrennung", WLAN).
>>
>>The way it works is that ntpd repeatedly scans the list of network
>>interfaces and reconfigures when changes are found.  On systems
>>that provide the routing socket interface (*BSD and derived)  this
>>is almost instantaneous. On all other systems (Windows, Linux and the rest)
>>the interface list is scanned at regular intervals (5 minutes for now).
>>    
>>
>
>Windows can also receive notifications of address changes, but that
>would need to be coded of course.
>  
>
Good.

>>Currently known issues:
>>   - Windows needs to accept the fact that sockets may be
>>     closed during the lifetime of the daemon (bug 671)
>>    
>>
>
>I will need to add code to allow the completion code to be notified.
>
>  
>
Just hook in into ntp_io,c:init_async_notifications() for the
Windows case. When Windows detects any interface/routing change
just call timer_interfacetimeout(current_time) while the timeout
handler is not running - that'll be it.

>>   - size (due to compiler padding) differences between some
>>     64 bit archs and the rest (bug 672)
>>
>>    
>>
>
>I'm not sure I understand the problem here. What does it matter what the
>size of the structure is? Is sizeof(variable) not sufficient? 
>
It is the over-the-wire format. The size of the transmitted data block 
is checked.
given different sizes the cross platform functionality would not be there.

>Also,
>exactly where is the problem?
>  
>
Alpha/Mips 64 bit compilers prefer to pad structure sizes to a 64 bit 
boundary.
It should be fixed now.

>Danny
>  
>
Frank


More information about the hackers mailing list