Re: [PATCH 5/9] makecfg: handle metadir quota options

["Darrick J. Wong" <djwong@kernel.org>]

On Fri, Sep 11, 2026 at 08:48:36AM -0700, Christoph Hellwig wrote:
> On Tue, Sep 08, 2026 at 11:01:33PM -0700, Darrick J. Wong wrote:
> >  
> > +/* These should correspond to XFS_[UGP]UOTA_{ACCT,ENFD} */
> 
> CAn we reuse those?  If not at least add the missing Q in QUOTA here :)

Oh!  Yes, we can just reuse them since xfs_log_format.h is in
/usr/include/xfs/.  Well that simplifies things a lot.

> > diff --git a/configure.ac b/configure.ac
> > index 40ccd88aa2a950..8a2505e37c70ab 100644
> > --- a/configure.ac
> > +++ b/configure.ac
> > @@ -195,6 +195,7 @@ if test "$have_listmount" = "yes"; then
> >  fi
> >  AC_HAVE_STATMOUNT_SUPPORTED_MASK
> >  AC_HAVE_FANOTIFY_MOUNTINFO
> > +AC_HAVE_QUOTACTL_FD
> 
> Shouldn't the AC_HAVE_QUOTACTL_FD go into a separate patch, including
> a generic helper?  I assume most other uqotactl users would prefer
> that as well when available.

Yes.  I could hoist xfsquotactl from quota/linux.c into libfrog, and
then adapt it to use quotactl_fd if available:

int
xfsquotactl(
	int			mnt_fd,
	const char		*device,
	enum xfs_quota_cmd	xcommand,
	uint			xtype,
	uint			id,
	void			*addr)
{
	const int		op = QCMD(xcommand_to_qcommand(xcommand),
					  xtype_to_qtype(xtype));
	int			ret = -1;

	errno = ENOSYS;
#ifdef HAVE_QUOTACTL_FD
	if (mnt_fd >= 0)
		ret = syscall(SYS_quotactl_fd, mnt_fd, op, id, addr);
#endif
	if (ret != -1 || errno != ENOSYS)
		return ret;

	return quotactl(op, device, id, addr);
}

This will require some amount of refactoring in xfs_quota, since it
doesn't currently try to open() the filesystem mountpoint.  OTOH now it
will no longer fail if a sysadmin does an lvrename while it's running.

However, doing all that got intense, so I'm splitting all that out into
a separate hoist/cleanup/adapt series.  For now, the get_qflags function
in spaceman will look like this:

	/* GETQSTAT returns qflags for all quota types, not just user */
	ret = quotactl(QCMD(Q_XGETQSTAT, USRQUOTA), f->fs_path.fs_name, 0,
			(void *)&qstat);

until the new series ports it to xfsquotactl.

> Otherwise this looks good from a quick look.

--D