Re: [PATCH 4/6] make struct mountpoint bear the dentry reference to mountpoint, not struct mount

[Al Viro]

On Sun, Jul 07, 2019 at 02:17:38PM -0700, Linus Torvalds wrote:
> On Fri, Jul 5, 2019 at 5:22 PM Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:
> >
> > +static HLIST_HEAD(unmounted);  /* protected by namespace_sem */
> > +static LIST_HEAD(ex_mountpoints);
> 
> What protects the ex_mountpoints list?
> 
> It looks like it's the mount_lock, but why isn't that documented?
> 
> It sure isn't namespace_sem from the comment above.

It is namespace_sem.  Of all put_mountpoint() callers only the one
from mntput_no_expire() (disposing of stuck MNT_LOCKed children)
is not under namespace_sem; all such are told to use ex_mountpoints
for disposal.  See
+               if (!list)
+                       list = &ex_mountpoints;
+               dput_to_list(dentry, list);
in there.  Only one call site gets non-default disposal list -
                list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts,  mnt_child) {
-                       umount_mnt(p);
+                       umount_mnt(p, &list);
                }
in mntput_no_expire() passes a local list to umount_mnt() (which passes it
to put_mountpoint()).

And namespace_unlock() empties ex_mountpoints before dropping namespace_sem -
the contents gets moved to a local list, which is fed to shrink_dentry_list()
as soon as we drop namespace_sem.

Protection of the disposal list is up to the callers of put_mountpoint();
for ex_mountpoints it's namespace_sem, for the one in mntput_no_expire()
we don't need any exclusion whatsoever - no other thread can access it.

IOW, the comment re namespace_sem applies to ex_mountpoints as well.