Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <6535286b-2532-dc86-3c6e-1b1e9bce358f@kernel.dk>
 <CAHk-=wjrayP=rOB+v+2eTP8micykkM76t=6vp-hyy+vWYkL8=A@mail.gmail.com> <4bcf3012-a4ad-ac2d-e70b-17f17441eea9@kernel.dk>
In-Reply-To: <4bcf3012-a4ad-ac2d-e70b-17f17441eea9@kernel.dk>
From:   Linus Torvalds <torvalds@linux-foundation.org>
Date:   Fri, 20 Nov 2020 16:23:41 -0800
Message-ID: <CAHk-=wimYoUtY4ygMNknkKZHqgYBZbkU4Koo5cE6ar8XjHkzGg@mail.gmail.com>
Subject: Re: [GIT PULL] io_uring fixes for 5.10-rc
To:     Jens Axboe <axboe@kernel.dk>, Al Viro <viro@zeniv.linux.org.uk>
Cc:     io-uring <io-uring@vger.kernel.org>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Fri, Nov 20, 2020 at 1:36 PM Jens Axboe <axboe@kernel.dk> wrote:
>
> I don't disagree with you on that. I've been a bit gun shy on touching
> the VFS side of things, but this one isn't too bad. I hacked up a patch
> that allows io_uring to do LOOKUP_RCU and a quick test seems to indicate
> it's fine. On top of that, we just propagate the error if we do fail and
> get rid of that odd retry loop.

Ok, this looks better to me (but is obviously not 5.10 material).

That said, I think I'd prefer to keep 'struct nameidata' internal to
just fs/namei.c, and maybe we can just expert that

        struct nameidata nd;

        set_nameidata(&nd, req->open.dfd, req->open.filename);
        file = path_openat(&nd, &op, op.lookup_flags | LOOKUP_RCU);
        restore_nameidata();
        return filp == ERR_PTR(-ECHILD) ? -EAGAIN : filp;

as a helper from namei.c instead? Call it "do_filp_open_rcu()" or something?

That "force_nonblock" test seems a bit off, though. Why is that RCU
case only done when "!force_nonblock"? It would seem that if
force_nonblock is set, you want to do this too?

Al? You can see the background on lkml, but basically io_uring wants
to punt file open to a kernel thread, except if it can just be done
directly without blocking (which is pretty much that RCU lookup case).

And the thing that triggered this is that /proc/self/ can only be done
directly - not in a kernel thread. So the RCU case actually ends up
being interesting in that it would handl those things.

            Linus