On 06/28/2013 07:22 PM, Dwight Engen wrote: > On Fri, 28 Jun 2013 17:39:13 -0400 > Brian Foster <bfoster@xxxxxxxxxx> wrote: > >> On 06/28/2013 04:28 PM, Dwight Engen wrote: >>> On Fri, 28 Jun 2013 14:50:24 -0400 >>> Brian Foster <bfoster@xxxxxxxxxx> wrote: >>> >>>> On 06/28/2013 11:11 AM, Dwight Engen wrote: ... >> >> To be honest, there aren't any real users of the eofblocks command >> from userspace that I'm aware of at the moment. I added it originally >> for a poc quota implementation I was working on for gluster, but the >> primary use case for the scanning mechanism is to allow background >> clean up of files such that post-eof speculative preallocation >> doesn't hang around for too long. > > ... and there are likely to be scenarios where waiting for the timer > would be too long? > Well one can adjust the timer via the /proc file if necessary. Our use case was along the lines of ensuring prealloc was cleared up at certain points where we wanted to track space usage (using a cluster of XFS project quotas to represent a higher level quota), iirc. It's been a while since I've looked at that code... ;) >>> Maybe this permission stuff should be a separate change since it >>> isn't really related to user namespace stuff? I just happened to be >>> in the vicinity and am happy to help :) >>> >> >> Sounds reasonable to me. :) > > If you want me to code up either option, let me know. > Feel free to. :) I agree that it's separate from the userns work. I'll put it on my todo list if you don't get around to it. Brian >> Brian _______________________________________________ xfs mailing list xfs@xxxxxxxxxxx http://oss.sgi.com/mailman/listinfo/xfs
