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: <20210104232123.31378-1-stephen.s.brennan@oracle.com>
 <20210105055935.GT3579531@ZenIV.linux.org.uk> <20210105165005.GV3579531@ZenIV.linux.org.uk>
 <20210105195937.GX3579531@ZenIV.linux.org.uk>
In-Reply-To: <20210105195937.GX3579531@ZenIV.linux.org.uk>
From:   Linus Torvalds <torvalds@linux-foundation.org>
Date:   Tue, 5 Jan 2021 12:38:31 -0800
Message-ID: <CAHk-=wiP9EAP=JHGKG5LUCusVjVzTQoPVyweJkrX5dP=T_NxXw@mail.gmail.com>
Subject: Re: [PATCH v4] proc: Allow pid_revalidate() during LOOKUP_RCU
To:     Al Viro <viro@zeniv.linux.org.uk>
Cc:     Alexey Dobriyan <adobriyan@gmail.com>,
        James Morris <jmorris@namei.org>,
        "Serge E. Hallyn" <serge@hallyn.com>,
        LSM List <linux-security-module@vger.kernel.org>,
        Paul Moore <paul@paul-moore.com>,
        Stephen Smalley <stephen.smalley.work@gmail.com>,
        Eric Paris <eparis@parisplace.org>,
        SElinux list <selinux@vger.kernel.org>,
        Casey Schaufler <casey@schaufler-ca.com>,
        Eric Biederman <ebiederm@xmission.com>,
        linux-fsdevel <linux-fsdevel@vger.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Matthew Wilcox <willy@infradead.org>,
        Stephen Brennan <stephen.s.brennan@oracle.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, Jan 5, 2021 at 12:00 PM Al Viro <viro@zeniv.linux.org.uk> wrote:
>
> We are not guaranteed the locking environment that would prevent
> dentry getting renamed right under us.  And it's possible for
> old long name to be freed after rename, leading to UAF here.

This whole thing isn't important enough to get the dentry lock. It's
more of a hint than anything else.

Why isn't the fix to just use READ_ONCE() of the name pointer, and do
it under RCU?

That's what dentry_name() does for the much more complex case of
actually even following parent data for a depth up to 4, much less
just a single name.

So instead of

                       spin_lock(&dentry->d_lock);
                       audit_log_untrustedstring(ab, dentry->d_name.name);
                       spin_unlock(&dentry->d_lock);

why not

                       rcu_read_lock();
                       audit_log_untrustedstring(ab,
READ_ONCE(dentry->d_name.name));
                       rcu_read_unlock();

which looks a lot more in line with the other dentry path functions.

Maybe even have this as part of fs/d_path.c and try to get rid of
magic internal dentry name knowledge from the audit code?

                  Linus