[PATCH 01/14] VFS: Add additional RESOLVE_* flags [ver #18]

[Linus Torvalds]

On Mon, Mar 9, 2020 at 5:56 PM Aleksa Sarai <cyphar at cyphar.com> wrote:
>
> On 2020-03-09, David Howells <dhowells at redhat.com> wrote:
> > This is necessary for fsinfo() to use RESOLVE_* flags instead of AT_* flags
> > if the latter are to be considered deprecated for new system calls.
> >
> > Also make openat2() handle RESOLVE_NO_TRAILING_SYMLINKS.

No, please let's not do this.

We have O_NOFOLLOW, and we can't get rid of it.

So adding RESOLVE_NO_TRAILING_SYMLINKS isn't a cleanup. It's just
extra complexity for absolutely zero gain.


> After thinking about what Christian said some more, I reckon we
> shouldn't support both O_NOFOLLOW and RESOLVE_NO_TRAILING_SYMLINKS. But
> that means we'll need to cherry-pick this patch and get it into mainline
> before v5.6.

No.

It simply means that we shouldn't have RESOLVE_NO_TRAILING_SYMLINKS at all.

Adding that flag is a mistake. It causes problems like this, where
subtlenly people say "what if both flags are set".

Just don't do it.

There's no way in hell we can ever get rid of O_NOFOLLOW anyway, since
people will continue to use plain open() and openat().

So adding RESOLVE_NO_TRAILING_SYMLINKS is entirely redundant.

Don't deprecate the old flags that are going to always stay around,
don't add stupid new flags that add no value.

It's that easy.

              Linus