[slow unmount] revert work-around for bug 543617 please

Andy Whitcroft apw at canonical.com
Wed May 5 12:10:40 UTC 2010


On Tue, May 04, 2010 at 11:57:19AM -0700, Kees Cook wrote:
> Hello,
> 
> I think that 5e1941884c700b7b97bcad52c2d8212ac56a7ebc is causing more
> problems than it tries to solve[1].  i.e. if any filesystem is under heavy
> load, you can't umount any other filesystems until the sync finishes.
> 
> Additionally, trivial stuff (even with an idle system) take an unexpectedly
> long time (e.g. umount of a bind mount takes 1 second instead of nearly 0).
> 
> Try this in Karmic and Lucid:
> 
> sudo -s
> mkdir -p /mnt/test
> mount -o bind /tmp /mnt/test
> time umount /mnt/test
> 
> Karmic: real    0m0.010s
> Lucid: real    0m1.022s
> 
> -Kees
> 
> [1] https://bugs.launchpad.net/ubuntu/+source/linux/+bug/543617

I believe Surbhi is now testing a potential upstream fix.  If that
testing is good, then we would clearly revert and replace it.  I am
therefore inclined to wait on that testing, rather than jump the SRU
process twice.

Otherwise it does appear that this has a not so healthy side.

-apw




More information about the kernel-team mailing list