> raven> As Jeff has said this sort of nesting of mounts is, in general, not
> raven> supported by autofs and it worked previously because the locking was
> raven> less broken in 4.1.3 than in 4.1.4 (yes its still broken in 4.1.4).
> 
> raven> Can this case be done safely?
> 
> raven> I'm not entirely sure of the answer to this but I've never liked
> raven> using locking in autofs. I introduced it only to prevent mtab
> raven> corruption during mount (correct me if you have a different
> raven> recollection Jeff). It happens during rapid mount activity such as
> raven> when there as a largish number of entries in the master map (of the
> raven> order of 50 or more).
> 
> I'm not sure why it was introduced.  When looking at the code, the only
> thing I could deduce was that we do locking in autofs in order to not issue
> a ton of mounts all at once.  Because the mount command has a timeout for
> each mount, and because each mount will take some amount of time to
> complete, there exists a possibility of getting failed mounts due to a
> timeout or lack of resources.  That was my only guess.

Yes, with lots of mounts within a short time, you can easily run out of
available reserved ports (since many of them end up in a TIME_WAIT state
for a while).  We changed mount to simply wait and retry in such cases.
I can send you the patch we're currently using, if you really want it, but
I'd implement it differently if I were to do it again (i.e. as implemented
it retries a fixed number of times; I think it would be better to retry
until the timeout specified in the timeo parameter expires so that
port starvation isn't handled differently than other problems).

> 
> I have seen mtab corruption in the field, but typically only in the
> presence of mvfs (the clearcase file system).

There aren't enough hours in the day to go into all the problems induced
by mvfs.....

_______________________________________________
autofs mailing list
[email protected]
http://linux.kernel.org/mailman/listinfo/autofs

Reply via email to