On Sat, May 09, 2020 at 02:47:02PM -0500, Eric Sandeen wrote:
> vfs/ext3/4 quota allows the administrator to push out the grace time
> for soft quota with the setquota -T command:
>
> setquota -T [ -u | -g ] [ -F quotaformat ] name block-grace inode-grace -a | filesystem...
>
> -T, --edit-times
> Alter times for individual user/group when softlimit is enforced.
> Times block-grace and inode-grace are specified in seconds or can
> be string 'unset'.
>
> Essentially, if you do "setquota -T -u username 1d 1d" and "username" is
> over their soft quotas and into their grace time, it will extend the
> grace time expiry to 1d from now.
>
> xfs can't do this, today. The patch below is a first cut at allowing us
> to do this, and userspace updates are needed as well (I have those in a
> patch stack.)
>
> I'm not looking so much for patch review right now, though, what I'm
> wondering is if this is a change we can make from the ABI perspective?
>
> Because today, if you try to pass in a UID other than 0 (i.e. the
> default grace period) it just gets ignored by the kernel, not rejected.
>
> So there's no real way to know that the grace period adjustment failed
> on an older kernel. We could consider that a bug and fix it, or
> consider it a change in behavior that we can't just make without
> at least some form of versioning. Thoughts?
>
> Anyway, the patch below moves the disk quota grace period adjustment out
> from "if id == 0" and allows the change for any ID; it only sets the
> default grace value in the "id == 0" case.
>
> Signed-off-by: Eric Sandeen <sandeen@xxxxxxxxxx>
> ---
>
> diff --git a/fs/xfs/xfs_qm_syscalls.c b/fs/xfs/xfs_qm_syscalls.c
> index f48561b7e947..e58ee98f938c 100644
> --- a/fs/xfs/xfs_qm_syscalls.c
> +++ b/fs/xfs/xfs_qm_syscalls.c
> @@ -555,32 +555,41 @@ xfs_qm_scall_setqlim(
> ddq->d_rtbwarns = cpu_to_be16(newlim->d_rt_spc_warns);
>
> if (id == 0) {
> - /*
> - * Timelimits for the super user set the relative time
> - * the other users can be over quota for this file system.
> - * If it is zero a default is used. Ditto for the default
> - * soft and hard limit values (already done, above), and
> - * for warnings.
> - */
> - if (newlim->d_fieldmask & QC_SPC_TIMER) {
> - defq->btimelimit = newlim->d_spc_timer;
> - ddq->d_btimer = cpu_to_be32(newlim->d_spc_timer);
> - }
> - if (newlim->d_fieldmask & QC_INO_TIMER) {
> - defq->itimelimit = newlim->d_ino_timer;
> - ddq->d_itimer = cpu_to_be32(newlim->d_ino_timer);
> - }
> - if (newlim->d_fieldmask & QC_RT_SPC_TIMER) {
> - defq->rtbtimelimit = newlim->d_rt_spc_timer;
> - ddq->d_rtbtimer = cpu_to_be32(newlim->d_rt_spc_timer);
> - }
> if (newlim->d_fieldmask & QC_SPC_WARNS)
> defq->bwarnlimit = newlim->d_spc_warns;
> if (newlim->d_fieldmask & QC_INO_WARNS)
> defq->iwarnlimit = newlim->d_ino_warns;
> if (newlim->d_fieldmask & QC_RT_SPC_WARNS)
> defq->rtbwarnlimit = newlim->d_rt_spc_warns;
> - } else {
> + }
> +
> + /*
> + * Timelimits for the super user set the relative time the other users
> + * can be over quota for this file system. If it is zero a default is
> + * used. Ditto for the default soft and hard limit values (already
> + * done, above), and for warnings.
> + *
> + * For other IDs, userspace can bump out the grace period if over
> + * the soft limit.
> + */
> + if (newlim->d_fieldmask & QC_SPC_TIMER) {
> + if (!id)
> + defq->btimelimit = newlim->d_spc_timer;
> + ddq->d_btimer = cpu_to_be32(newlim->d_spc_timer);
> + }
> + if (newlim->d_fieldmask & QC_INO_TIMER) {
> + printk("setting inode timer to %d\n", newlim->d_ino_timer);
Stray printk here.
> + if (!id)
> + defq->itimelimit = newlim->d_ino_timer;
> + ddq->d_itimer = cpu_to_be32(newlim->d_ino_timer);
> + }
> + if (newlim->d_fieldmask & QC_RT_SPC_TIMER) {
> + if (!id)
> + defq->rtbtimelimit = newlim->d_rt_spc_timer;
> + ddq->d_rtbtimer = cpu_to_be32(newlim->d_rt_spc_timer);
> + }
Otherwise I guess this looks reasonable. It might help to patchbomb all
the kernel/userspace/fstests changes together so I don't have to go
scurrying around to find the pieces(?)
--D
> +
> + if (id != 0) {
> /*
> * If the user is now over quota, start the timelimit.
> * The user will not be 'warned'.
>
>
