On Thu, 17 Feb 2000, Jeremy Fitzhardinge wrote:
> On 18-Feb-00 Steven N. Hirsch wrote:
> > I'm running 2.2.14 w/ your autofs kernel patches and latest utilities.
>
> OK, so that's 4.0.0pre6?
Correct.
> > Twice in the last week, I've had an automounted NFS export "disappear" and
> > refuse to remount without restarting autofs.
>
> Is the symptom that the /net/cy directory still exists, but there's
> nothing mounted on it? Is there anything mounted underneath?
/net/cy is empty except for 'usr/src/*' (the nested mount), i.e.
everything exported from the 'cy:/' volume vanishes.
> If you do a force expire (kill -USR1 automount), does it delete the
> mountpoint and allow a remount?
I haven't tried this.
> > I lost all incoming messages from a run of 'fetchmail' (my incoming mail
> > spool is nfs-mounted and they went into the bit-bucket).
>
> (Fetchmail throws things away if something breaks? Bad fetchmail!)
There was furious outgoing net activity while this was happening. I'm
really not sure what in blazes it did with the data!
> > As root, I umounted 'cy:/usr/src' and tried to cd into /net/cy several
> > times to trigger a remount of 'cy:/'. No luck. It would mount
> > 'cy:/usr/src', i.e. the only thing under /net/cy would be 'usr/src/*'.
> > The root volume was just not visible.
>
> So did it mount /net/cy/usr/src, but not /net/cy itself? That's very
> strange.
Yes, that's exactly the case. I tried several times after a forcible
'umount' of /net/cy/usr/src.
> If for some reason the first mount of /net/cy failed then it
> will go on to mount /net/cy/usr/src, creating a mount-point if
> necessary. If it times that out, then it should clean things up
> properly. Rather than unmounting things manually, you should send
> automount a USR1 to get it to clean things up (or at least send a USR1
> after doing the umount).
I'll try this when it happens again.
> > I tried to shut it down and restart, but something was busy:
> >
> > Feb 17 19:42:23 pii automount[532]: shutting down, path = /misc
> > Feb 17 19:42:23 pii automount[11490]: expired /net/cy
> > Feb 17 19:42:23 pii automount[536]: shutdown failed: filesystem still busy
>
> Was there something still using the /net directory (as cwd perhaps)?
Probably that's the case, as my home directory is /net/home/hirsch.
> > Only after forcibly trying to shut down, did the root volume reappear.
> > This has happened about 3x over the past week.
>
> It looks to me like the first mount failed, but the second succeeded; I've
> seen similar things here when DNS is in a slightly strange state, where it
> fails one query and succeeds for the same one immediately after.
Yes, but that doesn't explain why I was able to clear /net/cy/* using a
umount, then have a second attempt also fail to mount the cy:/ export.
Unless... Are you saying that a SIGUSR1 is required in _addition_ (or in
lieu of) a successful 'umount'?
> To get more info about what its trying to do in detail, you can
> compile automount with DEBUG set (see near the top of
> daemon/automount.c), or look at the NFS server logs to see what mount
> attempts it actually saw.
Good thought. I'll stroll through them.
>
> I guess a fix I can apply is to never allow partial mounts; if one
> mount fails, back out any that succeeded and fail the entire mount.
I'm in favor of this solution. Principle of least surprise and all that..
Steve
