linux-kernel - Re: drivers/char/random.c: more ruminations

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20140610031017.9893.qmail@ns.horizon.com>
Date:	9 Jun 2014 23:10:17 -0400
From:	"George Spelvin" <linux@...izon.com>
To:	linux@...izon.com, tytso@....edu
Cc:	hpa@...ux.intel.com, linux-kernel@...r.kernel.org,
	mingo@...nel.org, price@....edu
Subject: Re: drivers/char/random.c: more ruminations

>> I have an idea for a patch to change _xfer_secondary_pool
>> to use extract_buf rather than extract_entropy; is all that
>> FIPS stuff needed for purely internal transfers?

> That's not the part of extract_entropy() which is critical.  What's
> critical is the control over only transfering entropy when there is at
> least a certain minimum amount of entropy.  This provides the
> Yarrow-like avalanche property which is required to provide recovery
> after the internal state of the entropy pools have been exposed.

I do understand the importance of catastrophic reseeding, but
I don't see that implemented in extract_entropy().

Extract_entropy() itself contains a call to xfer_secondary_pool, but
that is a no-op in the case I'm considering (when it's called from
_xfer_secondary_pool) because in that case, r is the the input_pool,
which doesn't have an r->pull pool.

The number of bytes transferred is adjusted by account(), but
it's only adjusted downward.  (If it were adjusted up, that would be
a buffer overrun.)

Normal reads seem to ask for a reseed equal in size to read itself,
which is "ctastrophic" enough.

The only *minimum* reseed transfer size I can see is enforced in
_xfer_secondary_pool and push_to_pool, which use random_read_wakeup_bits
(default 64) as a minimum reseed.  (Prehaps increase this?)
standards, IMHO.)

Now, having said all that, I'll try to avoid embarrassing myself
in public again and ask... what am I missing?

On other matters...

If I play around with getting the entropy locking right, do you
have a strong preference for the single-lock or two-lock design?
I can't decide which to start with.

The latter is an extensive core code change, but I can easily convince
myself there are no deadlock issues.  That's the one I described
with separate adder and extractor locks and code paths, two "amount
of entropy ever added" variables, ane one "amont of entropy ever removed".

The single-lock design is hopefully less code, but (to me, at least) less
obviously deadlock-free.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/