Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Arvind Sankar <nivedita@alum.mit.edu>
Date:   Wed, 1 Jan 2020 18:08:46 -0500
To:     Linus Torvalds <torvalds@linux-foundation.org>
Cc:     Arvind Sankar <nivedita@alum.mit.edu>,
        Dominik Brodowski <linux@dominikbrodowski.net>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Al Viro <viro@zeniv.linux.org.uk>,
        youling 257 <youling257@gmail.com>
Subject: Re: [PATCH] early init: open /dev/console with O_LARGEFILE
Message-ID: <20200101230845.GA454720@rani.riverdale.lan>
References: <20191231150226.GA523748@light.dominikbrodowski.net>
 <20200101003017.GA116793@rani.riverdale.lan>
 <20200101183243.GB183871@rani.riverdale.lan>
 <CAHk-=whzgLPi4szh8xOKysuS9CKaQESngc=n0omBVpwdQ822aw@mail.gmail.com>
 <20200101225049.GB438328@rani.riverdale.lan>
 <CAHk-=wgHg8z5WQBoypjyi0VPp3E44oO_Szuvk9f6v1mGMqsSZg@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <CAHk-=wgHg8z5WQBoypjyi0VPp3E44oO_Szuvk9f6v1mGMqsSZg@mail.gmail.com>
User-Agent: Mutt/1.10.1 (2018-07-13)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Wed, Jan 01, 2020 at 03:01:12PM -0800, Linus Torvalds wrote:
> On Wed, Jan 1, 2020 at 2:50 PM Arvind Sankar <nivedita@alum.mit.edu> wrote:
> >
> > Shouldn't that only affect init though? The getty's it spawns should be
> > in their own sessions.
> 
> They *should* be in their own sessions, and clearly this problem
> doesn't seem to really affect much anybody else.
> 
> But I think youling has some limited and/or odd init userspace, and I
> think it gets confused.
> 
> So my theory is that because of the file descriptor leak, that "forget
> the old controlling tty" doesn't happen, and then subsequent tty opens
> don't do the right thing.
> 
> Maybe.
> 
> But it's the only real semantic change I can see in that whole patch.
> 
>                  Linus

Ok. If you do end up going with this rather than the revert, one minor
nit with the patch -- if somehow the filp_open succeeds but one of the
f_dupfd's fails, you still need to do an fput to avoid leaking the
reference.