[systemd-devel] Dependency loop due to network filesystem and quotacheck; suboptimal loop resolution.

[Lennart Poettering]

On Thu, 17.03.11 17:49, Andrey Borzenkov (arvidjaar at gmail.com) wrote:

> 
> On Thu, Mar 17, 2011 at 5:44 PM, Lennart Poettering
> <lennart at poettering.net> wrote:
> > On Thu, 17.03.11 16:50, Andrey Borzenkov (arvidjaar at gmail.com) wrote:
> >
> >>
> >> On Thu, Mar 17, 2011 at 4:43 AM, Lennart Poettering
> >> <lennart at poettering.net> wrote:
> >> > On Thu, 10.03.11 00:42, Andrey Borzenkov (arvidjaar at gmail.com) wrote:
> >> >> It is not. Suggested patch attached.
> >> >
> >> >> From: Andrey Borzenkov <arvidjaar at gmail.com>
> >> >> Subject: [PATCH] mount: do not add dependency on network filesystem to quotacheck
> >> >>
> >> >> This creates loop:
> >> >>
> >> >> fs -> quotacheck -> basic -> network -> fs
> >> >>
> >> >> It does not look like quota was enabled for them in /etc/init.d/netfs
> >> >> anyway.  If quota is required, it probably should be implemented as
> >> >> per mount point unit.
> >> >
> >> [...]
> >> > Anywaym uf you rework this patch to check for the usrquota/grpquota
> >> > options I'd merge it promptly
> >>
> >> But it does not solve the problem of usrquota being set on a _netfs
> >> filesystem. So either this has to be skipped completely or additional
> >> unit provided.
> >
> > Well, isn't it kind of a misconfiguration if people use
> > "usrquota,grpquota" on a network filesystem?
> 
> Is ext4 on iSCSI target network file system?

True.

We will probably not be able to support quota for those, since we cannot
block apps to access the fs while quotacheck is running. But I guess
even if we cannot support it really this shouldn't be an excuse for a
dep loop. Hence yepp, a patch that adds deps to quotacheck only iff a) a
device has grpquota or usrquota set AND b) is not a network fs would be
very good.

Lennart

-- 
Lennart Poettering - Red Hat, Inc.