Re: Maintaining connection

[Dan Williams <dcbw redhat com>]

On Thu, 2015-03-12 at 11:54 -0400, Stuart Gathman wrote:
> On 03/11/2015 06:23 PM, Dan Williams wrote:
> > On Wed, 2015-03-11 at 15:57 -0500, Alex Ferm wrote:
> > > Hello, I'm trying to write a python script that resets NetworkManager
> > > when the state is not "NM_STATE_CONNECTED_GLOBAL". Does NetworkManager
> > > time out and retry automatically during the "NM_STATE_CONNECTING" state?
> > > Also, how is the "NM_STATE_CONNECTED_GLOBAL" determined(ie: is there a
> > > periodic ping or something?)?
> > What's the reason to reset NM when it reports something isn't connected?
> > Just to ensure always-on connectivity as hard as possible?  Also, what
> > do you mean by "reset", what specific actions are you running to do so?
> In general, it would be useful if NM was able to detect that networking 
> was hosed and reset things.  How would it know, however? I generally 
> ping the gateway manually to check the connection.  Are gateways always 
> pingable?  Just checking for a connection is not sufficient - e.g. when 
> wpa_supplicant hangs (radio is operational, but incoming and outgoing 
> packets have expired key and are discarded).  Microsoft used to ping one 

If the packets have an expired key, then that implies that key rotation
has not happened.  That's done on a timer that's negotiated during the
EAP or WPA-PSK setup.  If you had wpa_supplicant debug logs you could
probably figure out what's going on in the supplicant.

> of their servers.  But that could be a problem anywhere in between the 
> PC and their server.

NM does that with it's connectivity checking functionality, but of
course since it's "phone home" functionality more or less, that must be
enabled by the user/admin.  NM does advertise the result via dbus/nmcli,
but does not yet do anything with it internally.  It's also not
per-interface yet, so you couldn't use the result to determine whether a
specific interface was hosted or not.

Dan