Re: [PATCH v2 13/18] sysctl: move sysctl type to ctl_table_header

[Thomas Weißschuh]

On 2023-12-07 13:14:37+0100, Joel Granados wrote:
> On Wed, Dec 06, 2023 at 06:53:10AM +0100, Thomas Weißschuh wrote:
> > On 2023-12-05 14:50:01-0800, Luis Chamberlain wrote:
> > > On Tue, Dec 5, 2023 at 2:41 PM Thomas Weißschuh <linux@xxxxxxxxxxxxxx> wrote:
> > > >
> > > > On 2023-12-05 14:33:38-0800, Luis Chamberlain wrote:
> > > > > On Mon, Dec 04, 2023 at 08:52:26AM +0100, Thomas Weißschuh wrote:
> > > > > > @@ -231,7 +231,8 @@ static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
> > > > > >             return -EROFS;
> > > > > >
> > > > > >     /* Am I creating a permanently empty directory? */
> > > > > > -   if (sysctl_is_perm_empty_ctl_header(header)) {
> > > > > > +   if (header->ctl_table == sysctl_mount_point ||
> > > > > > +       sysctl_is_perm_empty_ctl_header(header)) {
> > > > > >             if (!RB_EMPTY_ROOT(&dir->root))
> > > > > >                     return -EINVAL;
> > > > > >             sysctl_set_perm_empty_ctl_header(dir_h);
> > > > >
> > > > > While you're at it.
> > > >
> > > > This hunk is completely gone in v3/the code that you merged.
> > > 
> > > It is worse in that it is not obvious:
> > > 
> > > +       if (table == sysctl_mount_point)
> > > +               sysctl_set_perm_empty_ctl_header(head);
> > > 
> > > > Which kind of unsafety do you envision here?
> > > 
> > > Making the code obvious during patch review hy this is needed /
> > > special, and if we special case this, why not remove enum, and make it
> > > specific to only that one table. The catch is that it is not
> > > immediately obvious that we actually call
> > > sysctl_set_perm_empty_ctl_header() in other places, and it begs the
> > > question if this can be cleaned up somehow.
> > 
> > Making it specific won't work because the flag needs to be transferred
> > from the leaf table to the table representing the directory.
> > 
> > What do you think of the aproach taken in the attached patch?
> > (On top of current sysctl-next, including my series)

> What would this new patch be fixing again? I could not follow ?

This patch improves upon and replaces the patch you asked to submit on
its own:  "sysctl: move sysctl type to ctl_table_header".

The current code and my original patch have to work around the fact that
the "empty" flag is first registered on a *leaf* ctl_table and from
there has to be transferred to the *directory* ctl_table somehow.
Which is confusing, at least it was for me and evidently also Luis.

The new code just directly sets the flag on the directory ctl_table and
gets rid of some now-dead code.
I should have written a proper changelog...

> Additionally, this might be another reason to set this patch aside :)

I hope we get this one in, it seems cleaner now.
If you agree I can send it as a proper standalone preparation patch.

> > 
> > Note: Current sysctl-next ist still based on v6.6.
> 
> > From 2fb9887fb2a5024c2620f2d694bc6dcc32afde67 Mon Sep 17 00:00:00 2001
> > From: =?UTF-8?q?Thomas=20Wei=C3=9Fschuh?= <linux@xxxxxxxxxxxxxx>
> > Date: Wed, 6 Dec 2023 06:17:22 +0100
> > Subject: [PATCH] sysctl: simplify handling of permanently empty directories
> > 
> > ---
> >  fs/proc/proc_sysctl.c  | 76 +++++++++++++++++++-----------------------
> >  include/linux/sysctl.h | 13 ++------
> >  2 files changed, 36 insertions(+), 53 deletions(-)

> [..]