On Sat, 6 Sep 2003, Buchan Milne wrote: > Well, this returns us to the original bug report, which is that netfs is > started too soon after network, and ifplugd may not have waited long > enough to guarantee the network is up before netfs starts. ifplugd only waits until the ethernet cable is connected, not for the interface to be configured, and as I reported it takes ifplugd about 3 seconds to bring the interface up, while it should only need a millisecond or so, since the interface is configured statically.
> I haven't seen this myself, but maybe our DHCP server is faster, or my > machine is slower, or it appeared after we switched to using autofs. The problem appeared when the network initscript started to use ifplugd. I actually got 2 ifplugds running and fighting because I used to have the driver module in /etc/modules. This resulted in the NFS share being mounted but the interface being down (!). Then I removed that entry from /etc/modules and got the problem I reported here. > Maybe a hack for the moment is to have a configurable pause for ifplugd? > Or maybe netfs needs to be moved to behind something else that is > guaranteed to wait long enough for the network? - The MII_NOT_SUPPORTED hack works, but this is just a workaround. Maybe ifplugd should not be used at all, or be configurable so people will only use it for external interfaces. The "network" initscript is supposed to bring up the network, and any scripts that run after it assume all relevant interfaces are configured. - Using ifplugd only for dhcp, not for statically configured interfaces is not a solution either, since in a small (corporate) network one may use dhcp but not autofs. - The only solution with ifplugd I see is fixing ifplugd to not take so long to configure the interface, and certainly not when it is run with -w .
