From: Theodore Ts'o <tytso@mit.edu>
Subject: Re: [PATCH, RFC] random: introduce getrandom(2) system call
Date: Sun, 20 Jul 2014 13:24:57 -0400
Message-ID: <20140720172457.GG5017@thunk.org>
References: <20140720162622.29664.qmail@ns.horizon.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Cc: linux-crypto@vger.kernel.org, linux-kernel@vger.kernel.org
To: George Spelvin <linux@horizon.com>
Content-Disposition: inline
In-Reply-To: <20140720162622.29664.qmail@ns.horizon.com>
Sender: linux-crypto-owner@vger.kernel.org

On Sun, Jul 20, 2014 at 12:26:22PM -0400, George Spelvin wrote:
> One basic question... why limit this to /dev/random?
> 
> If we're trying to avoid fd exhaustion attacks, wouldn't an "atomically
> read a file into a buffer" system call (that could be used on
> /dev/urandom, or /etc/hostname, or /proc/foo, or...) be more useful?
> 
> E.g.
> 
> ssize_t readat(int dirfd, char const *path, struct stat *st,
> 	char *buf, size_t len, int flags);
> 
> It's basically equivalent to openat(), optional fstat() (if st is non-NULL),
> read(), close(), but it doesn't allocate an fd number.
> 
> Is it necessary to have a system call just for entropy?
> 
> If you want a "urandom that blocks until seeded", you can always create
> another device node for the purpose.

I'd really rather not go down this path.  Your readat(2) proposal is
interesting, but it adds a whole lot of complications.  For example,
just simply booting a new kernel doesn't guarantee that a new device
node for "blocks until seeded" will exist.  So that means a lot of
applications will just either continue to use /dev/urandom, or have to
put in fallback code to first try the new device name, and then fall
back to /dev/urandom.  (And of course, they have to deal with what to
do if /dev/urandom doesn't exist --- which presumably would be
raise(SIGKILL), but we're now talking about a number of lines of codes
that application writers would have to get right.)

Readat(2) would also have to get linked into auditing, and LSM, and
honestly, it's a lot more work that I'm not all that interested in
doing and trying to get right.

> If you do stick with a random-specific call, specifying the entropy
> in bits (with some specified convention for the last fractional byte)
> is anothet interesting idea.  Perhaps too prone to bugs, though.
> (People thinking it's bytes and producing low-entropy keys.)

Definitely not worth the complexity.

							- Ted