[ntp:hackers] Cool new stuff

Frank Kardel kardel at ntp.org
Sat Jul 22 22:50:00 UTC 2006


David L. Mills wrote:

> Frank,
>
> This is profoundly and absolutely ridiculous. 

Dave,
in your opinion it may be. It may not make much sense in general and 
especially not to you but
configuration mechanisms in todays operating systems allow the 
reconfiguration and even the deletion
of such interfaces (in NetBSD "lo?" is member of the cloning interface 
group - these interfaces can be created
and destroyed as one wishes). Nobody of us can control what the vendor 
will do in his scripts or what a user would
do. ntpd has to handle that and it does. The code does not even 
specifically scan for loopback interfaces - the
interfaces just happen to show up in the interface list and all the code 
knows is how to manage a global
variable "loopback_interface" that holds the loopback interface pointer 
(not the IP address but to the structure describing
the address AND the socket file descriptor) that is assigned to refclock 
peers.
I get the impression that your interpretation of "interface" is "IP 
address". My interpretation for
"interface" is "the structure that holds the socket file descriptor 
bound to a specific IP address". So to discussion
may have been on different abstraction levels.

> The loopback interface and refclock interface addresses

I was not talking about addresses - the code cares about sockets bound 
to (local) addresses. socket instances come
and go. As do (local) addresses from the interfaces currently available 
to the system.

> are not going to change. To do so would screw up links, the access 
> control list and who knows what else.

no need to tell me that. the external addresses stay stable as 
configured. the local addresses may change as the
interfaces in the system change. the acl for the local interface 
addresses are tracked with the changes.
We need to differentiate between "leaf clients" that often have changing
local IP addresses (WLAN, pppoe, "Zwangstrennung" and whatnot) and 
"servers" that have a stable address setup
(how else could they be servers if they would constantly change their 
addresses). The dynamic interface code serves
the purpose that ntpd's internal interface list stays in sync with 
reality and thus binds to all available interfaces all the time
- one of ntpd design goals - and keeps up connectivity even when network 
setup is changing. On servers it would not see
any changes - on "leaf clients" the code is likely to see frequent changes.

> In any case the excessive debug display screws up a per-minute 
> protocol watch. Lose the display

that is not an option and you know that.

> or move it up in the -d option.

I am sorry that I messed up your debug display. Is is fixed now and 
Harlan pulled it into ntp-dev. If you want
to see any significant output of network interaction wrt/ interfaces and 
addresses you need to specify debug levels of 3 and/or
4. Tell me if these levels still interfere with your debug display.

Frank


More information about the hackers mailing list