Bug 840396

What does:

cat /proc/mounts

say? That should have all the options. Also the version of gfs2-utils that you are using should not have a mount.gfs2 helper, so if there is an issue, then it is in util-linux-ng rather than gfs2-utils

cat /proc/mounts says pretty much the same, just with different formatting:

/dev/mapper/mpathcp1 /mnt gfs2 rw,noatime 0 0

And yes, no mount.gfs2 anymore => reassigning.

Looking at the options, I'd have expected statfs_quantum to show up, at least, which is a bit odd. The mount command gets its info from mtab, which sometimes has different options to /proc/mounts. However, if the statfs_quantum is set to 0, then that should show up under /proc/mounts

So... when you say that you mounted gfs2 using the line in fstab, how did you do that? Via the gfs2 script, or just via a mount command on the command line. If the latter, what was the exact command line that you used?

I was mounting by "mount /mnt" => the /etc/fstab line should be used.

Btw. also on regular basis I'm mounting by "mount /mountpoint" in /etc/rc.d/rc.local, because the initscript in F16 seems to be broken (or it is a consequence of doing a direct yum update from F15, I don't know yet and therefore have not reported yet) and tries to mount the gfs volume before cman and gfs2-cluster are running (or at least one of them), the results being a kern_stop for that volume on all nodes:(

(In reply to comment #4)
> I was mounting by "mount /mnt" => the /etc/fstab line should be used.
> 
> Btw. also on regular basis I'm mounting by "mount /mountpoint" in
> /etc/rc.d/rc.local, because the initscript in F16 seems to be broken (or it
> is a consequence of doing a direct yum update from F15, I don't know yet and
> therefore have not reported yet) and tries to mount the gfs volume before
> cman and gfs2-cluster are running (or at least one of them), the results
> being a kern_stop for that volume on all nodes:(

This is probably due to the sysvinit->systemd transition: while gfs2's init scripts were updated to systemd units, the cman etc initscripts weren't, possibly causing a broken startup order. This should be fixed in Fedora 17 as cman has gone away, and also gfs2's mount order is handled by systemd itself. However, it is a separate issue so it would be best to open a separate bz for it if it's still a problem.

I've just tried this in f17. It seems to work so long as statfs_quantum is not set to 0. That is a bit odd, since the code clearly says:

      val = sdp->sd_tune.gt_statfs_quantum;
        if (val != 30)
                seq_printf(s, ",statfs_quantum=%d", val);

So it should print when set to anything except the default of 30. But... looking elsewhere I also see this:

    gt->gt_quota_quantum = args.ar_quota_quantum;
        if (args.ar_statfs_quantum) {
                gt->gt_statfs_slow = 0;
                gt->gt_statfs_quantum = args.ar_statfs_quantum;
        }
        else {
                gt->gt_statfs_slow = 1;
                gt->gt_statfs_quantum = 30;
        }

So it looks like we reset the quantum to be 30 secs when we set statfs_slow on. So we have a mismatch between our internal variables and the external representation in this case. So I don't think there is any issue relating to the mount command at all. It looks to me as if this is working mostly as expected, bar the 0 value for statfs_quantum.

Does this work if you specify a non-zero statfs_quantum too?

Steve, indeed, different values than zero are shown in mount :)
Nevertheless I'd expect all the mount options (e.g. noquota) to be shown, otherwise its a bit confusing.

Yes, just trying to track down where the actual issue lies first, and it sounds like it is a kernel issue, rather than userland.

There are some (backward compatibility) reasons why some of the options don't get shown, or only get shown in certain ways. GFS has the quota=[on/off/account] option so that noquota is a synonym for quota=off which is also the default and thus doesn't get shown. So we could show quota=off when noquota is requested, but not noquota itself - at least not without adding significant extra complication to the argument parsing code.

The issue with statfs_quantum=0 just looks like a bug to me, so we'll certainly fix that one.

Created attachment 605674 [details]
Proposed patch

As posted upstream

Now in Linus' tree, so fixed upstream.