Re: [PATCH] do_open(): Fix O_DIRECTORY | O_CREAT behavior

From: Christian Brauner <brauner@kernel.org>
To: Pedro Falcato <pedro.falcato@gmail.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>,
	Alexander Viro <viro@zeniv.linux.org.uk>,
	linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org,
	Aleksa Sarai <cyphar@cyphar.com>
Subject: Re: [PATCH] do_open(): Fix O_DIRECTORY | O_CREAT behavior
Date: Tue, 28 Mar 2023 10:12:27 +0200	[thread overview]
Message-ID: <20230328081227.keyadx3gdymr7fzf@wittgenstein> (raw)
In-Reply-To: <CAKbZUD1N-jsrro_9ix12vNmjL0iUqqvicCv7MHyj19O5LJs1aQ@mail.gmail.com>

On Mon, Mar 27, 2023 at 09:13:18PM +0100, Pedro Falcato wrote:
> On Tue, Mar 21, 2023 at 4:17 PM Christian Brauner <brauner@kernel.org> wrote:
> > It would be very nice if we had tests for the new behavior. So if @Pedro
> > would be up for it that would be highly appreciated. If not I'll put it
> > on my ToDo...
> 
> Where do you want them? selftests? I have a relatively self-contained
> ""testsuite"" of namei stuff that could fit in there well, after some
> cleanup.

I think I would prefer to have them as part of xfstests:
https://git.kernel.org/pub/scm/fs/xfs/xfstests-dev.git

as that's where nearly all of the fs testing is taking place. It's never
great when developers have to run 3 separate testsuites to get
meaningful coverage. So having it central to xfstests would be my
preference.

A while ago I added a testsuite that tests generic core VFS behavior
it's located under src/vfs:
https://git.kernel.org/pub/scm/fs/xfs/xfstests-dev.git/tree/src/vfs

and covers a lot of different things. So I would ask you to consider
adding a new testsuite into that file:

https://git.kernel.org/pub/scm/fs/xfs/xfstests-dev.git/tree/src/vfs/vfstest.c

I think the structure should be somewhat understandable. Then create a
new test in xfstests using the "new" helper in the generic sectionA

> ./new generic
Next test id is 728
Append a name to the ID? Test name will be 728-$name. y,[n]:
Creating test file '728'
Add to group(s) [auto] (separate by space, ? for list): auto quick
Creating skeletal script for you to edit ...

then call the vfstest binary from the generated test case:

echo "Silence is golden"

$here/src/vfs/vfstest --test-THAT-NEW-SWITCH-NAME-YOU-ADDED --device "$TEST_DEV" \
        --mount "$TEST_DIR" --fstype "$FSTYP"

status=$?
exit

(You can also submit this to LTP or tell them about this change and
they'll likely add tests in addition to xfstests.)

> 
> > The expectation often is that this particular combination would create
> > and open a directory. This suggests users who tried to use that
> > combination would stumble upon the counterintuitive behavior no matter
> > if pre-v5.7 or post v5.7 and quickly realize neither semantics give them
> > what they want. For some examples see the code examples in [1] to [3]
> > and the discussion in [4].
> 
> Ok so, silly question: Could it not be desirable to have these
> semantics (open a dir or mkdir, atomically)?
> It does seem to be why POSIX left this edge case implementation
> defined, and if folks are asking for it, could it be the right move?
> 
> And yes, I do understand (from reading the room) that no one here is
> too excited about this possibility.

Forgive me for being lazy and instead of repeating everything I'll just
leave a link to the other part of the thread
https://lore.kernel.org/lkml/20230328075735.d3rs27jjvarmn6dw@wittgenstein