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: <20210525193735.2716374-1-keescook@chromium.org>
In-Reply-To: <20210525193735.2716374-1-keescook@chromium.org>
From:   Jann Horn <jannh@google.com>
Date:   Tue, 25 May 2021 22:49:29 +0200
Message-ID: <CAG48ez2PdgpUj3GYRLDJ9MS1uKMZ4SU77i__vhXvmbzqudzuzA@mail.gmail.com>
Subject: Re: [PATCH] proc: Check /proc/$pid/attr/ writes against file opener
To:     Kees Cook <keescook@chromium.org>
Cc:     Linus Torvalds <torvalds@linuxfoundation.org>,
        stable <stable@vger.kernel.org>,
        "Eric W. Biederman" <ebiederm@xmission.com>,
        Paul Moore <paul@paul-moore.com>,
        Casey Schaufler <casey@schaufler-ca.com>,
        Oleg Nesterov <oleg@redhat.com>,
        James Morris <jmorris@namei.org>,
        John Johansen <john.johansen@canonical.com>,
        Stephen Smalley <sds@tycho.nsa.gov>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        kernel list <linux-kernel@vger.kernel.org>,
        linux-security-module <linux-security-module@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, May 25, 2021 at 9:37 PM Kees Cook <keescook@chromium.org> wrote:
> Fix another "confused deputy" weakness[1]. Writes to /proc/$pid/attr/
> files need to check the opener credentials, since these fds do not
> transition state across execve(). Without this, it is possible to
> trick another process (which may have different credentials) to write
> to its own /proc/$pid/attr/ files, leading to unexpected and possibly
> exploitable behaviors.
>
> [1] https://www.kernel.org/doc/html/latest/security/credentials.html?highlight=confused#open-file-credentials
>
> Fixes: 1da177e4c3f41 ("Linux-2.6.12-rc2")
> Cc: stable@vger.kernel.org
> Signed-off-by: Kees Cook <keescook@chromium.org>
> ---
>  fs/proc/base.c | 4 ++++
>  1 file changed, 4 insertions(+)
>
> diff --git a/fs/proc/base.c b/fs/proc/base.c
> index 3851bfcdba56..58bbf334265b 100644
> --- a/fs/proc/base.c
> +++ b/fs/proc/base.c
> @@ -2703,6 +2703,10 @@ static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
>         void *page;
>         int rv;
>
> +       /* A task may only write when it was the opener. */
> +       if (file->f_cred != current_real_cred())
> +               return -EPERM;

With this, if a task forks, the child will then still be able to open
its parent's /proc/$pid/attr/current and trick the parent into writing
to that, right? Is that acceptable? If not, the ->open handler should
probably also check for "current->thread_pid == proc_pid(inode)", or
something like that?