is 'umount' asynchronous?

is 'umount' asynchronous?

[Ben Greear]

While testing NFSv4 over TCPv6 on 2.6.36-rc3 + my srcaddr= patch today,
I noticed that I was getting 'busy' errors from umount if I closed a
file in a program and then had it immediately system("umount ...");

If I retry the umount a few times, it will start returning 'already unmounted'.
It took around 50-100ms (I was sleeping 50ms between umount attempts).

Is that expected behaviour?  Seems we never needed any retries and never
got 'busy' on 2.6.34 and earlier.

Thanks,
Ben

-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com

Re: is 'umount' asynchronous?

[J. Bruce Fields]

On Tue, Sep 07, 2010 at 05:12:17PM -0700, Ben Greear wrote:
> While testing NFSv4 over TCPv6 on 2.6.36-rc3 + my srcaddr= patch today,
> I noticed that I was getting 'busy' errors from umount if I closed a
> file in a program and then had it immediately system("umount ...");
> 
> If I retry the umount a few times, it will start returning 'already unmounted'.
> It took around 50-100ms (I was sleeping 50ms between umount attempts).
> 
> Is that expected behaviour?  Seems we never needed any retries and never
> got 'busy' on 2.6.34 and earlier.

I've had this:

	# ARGH: getting a spurious EBUSY on some
	# umounts; this helps:
	sleep 1;

in my test scripts for a while, and haven't tried to investigate why
it's happening.  (Checking git logs: I added that in July 2008.)

--b.