On Thu, Nov 10, 2005 at 04:06:15PM +0900, YAMAMOTO Takashi wrote:
> > It appears that if I mount /dev/wd0e on /mnt, and /dev is a tmpfs, changes
> > in /mnt are not flushed to wd0a.  Instead, dirty blocks linger until I
> > unmount /mnt.  I first noticed this because all of my tmpfs filesystems
> > showed 0 blocks available as I extracted a large .tgz file under /mnt/.
> 
> if tmpfs ENOSPC is only symptom you see,
> it's normal regarding buffer cache.
> 
> because tmpfs doesn't know the fact that some memory like buffer cache
> can be reclaimed on-demand, it behaves too conservatively if memory is
> filled up with caches.

Can I work around this with the right choice of vm settings?  Setting very
high vm.anon{min,max} and vm.file{min,max} = {0,1} didn't help.

Is uvm_wait() the right thing to do in tmpfs_reg_resize() when there
are insufficient free pages?

Dave

-- 
David Young             OJC Technologies
dyoung@ojctech.com      Urbana, IL * (217) 278-3933