On 04/24/2012 01:37 PM, Guido Günther wrote: I agree that we need all four limits (is the difference between hard and soft limits the amount of space reserved only for privileged users, so that regular users can't starve root out from ENOSPC cleanup that requires just a bit more filesystem usage?). But <hardlimit> seems ambiguous compared to <inodes_hardlimit>. Should we use size_hardlimit/inode_hardlimit instead (and likewise for softlimit)? And we want to let the user provide scaling on input (reusing some of the same code as <memory> handling). I guess inodes aren't really bytes, though, so a unit='KiB' looks a bit weird in conjunction with an inode limit. Are there any kernel constraints on use of limits? For example, can softlimit appear in isolation, or must it appear with hardlimit; and if it _can_ appear in isolation, does that mean the hardlimit matches the softlimit or that the hardlimit is unlimited? Should an explicit limit of 0 be treated as unlimited? Your patches came through unthreaded, which makes it harder to follow. It helps when all of the child patches are in-reply-to the 0/n cover letter. -- Eric Blake eblake(a)redhat.com +1-919-301-3266 Libvirt virtualization library http://libvirt.org