feeding randomness to a TLS server

[Abhijit Menon-Sen]

Hi.

I have a program linked to cryptlib that runs in a chroot jail and acts
as a TLS server (for various clients, e.g. IMAP, POP, HTTP). Right now,
I'm calling cryptAddRandom(NULL, CRYPT_RANDOM_SLOWPOLL) once at startup
before I chroot, and never call it again.

Do I need to call it again? When would be a good time? Every time I
accept a connection, or perhaps when a session finishes?

Suggestions welcome.

-- ams

_______________________________________________
Cryptlib mailing list
Cryptlib@... via Mail: cryptlib-request@...
Archive: ftp://ftp.franken.de/pub/crypt/cryptlib/archives/
http://news.gmane.org/gmane.comp.encryption.cryptlib
Posts from non-subscribed addresses are blocked to prevent spam, please
subscribe in order to post messages.

[Peter Gutmann]

Abhijit Menon-Sen <ams@...> writes:

>I have a program linked to cryptlib that runs in a chroot jail and acts as a
>TLS server (for various clients, e.g. IMAP, POP, HTTP). Right now, I'm
>calling cryptAddRandom(NULL, CRYPT_RANDOM_SLOWPOLL) once at startup before I
>chroot, and never call it again.
>
>Do I need to call it again? When would be a good time? Every time I accept a
>connection, or perhaps when a session finishes?

It won't hurt to call it again, but it's not absolutely necessary, cryptlib
does continuous re-seeding as it runs.

Peter.

_______________________________________________
Cryptlib mailing list
Cryptlib@... via Mail: cryptlib-request@...
Archive: ftp://ftp.franken.de/pub/crypt/cryptlib/archives/
http://news.gmane.org/gmane.comp.encryption.cryptlib
Posts from non-subscribed addresses are blocked to prevent spam, please
subscribe in order to post messages.

[Abhijit Menon-Sen]

At 2007-08-22 03:02:50 +1200, pgut001@... wrote:
>
> It won't hurt to call it again, but it's not absolutely necessary,
> cryptlib does continuous re-seeding as it runs.

Thanks.

We're running the cryptlib process inside a chroot jail, and /dev/random
isn't accessible without some gymnastics we'd rather not do (e.g. mount
a special device filesystem under the jail on FreeBSD). So we'd like to
open /dev/urandom before we chroot, and feed cryptlib data from it.

Would you consider a "cryptAddRandomSource()" function that could either
record an fd for use in getDevRandomData(), or register a callback to a
function that could be called from slowPoll() to fetch bytes? (Or any
other functionally equivalent solution, really.)

-- ams

_______________________________________________
Cryptlib mailing list
Cryptlib@... via Mail: cryptlib-request@...
Archive: ftp://ftp.franken.de/pub/crypt/cryptlib/archives/
http://news.gmane.org/gmane.comp.encryption.cryptlib
Posts from non-subscribed addresses are blocked to prevent spam, please
subscribe in order to post messages.

[Peter Gutmann]

Abhijit Menon-Sen <ams@...> writes:

>Would you consider a "cryptAddRandomSource()" function that could either
>record an fd for use in getDevRandomData(), or register a callback to a
>function that could be called from slowPoll() to fetch bytes? (Or any other
>functionally equivalent solution, really.)

There isn't any way of doing this in way that's even remotely portable across
OSes, if you need something like this you'd have to add support for it
yourself.

(The easiest fix, without having to modify cryptlib at all, would probably be
to provide a simulated EGD inside the jail).

The /dev/random stuff will be fixed in the next release, since it's a PRNG
read there'd have to be something truly weird happening if noBytes !=
DEVRANDOM_BYTES, but you're right, it should report the noBytes value.

Peter.


_______________________________________________
Cryptlib mailing list
Cryptlib@... via Mail: cryptlib-request@...
Archive: ftp://ftp.franken.de/pub/crypt/cryptlib/archives/
http://news.gmane.org/gmane.comp.encryption.cryptlib
Posts from non-subscribed addresses are blocked to prevent spam, please
subscribe in order to post messages.

[Abhijit Menon-Sen]

At 2007-08-31 19:03:47 +1200, pgut001@... wrote:
>
> (The easiest fix, without having to modify cryptlib at all, would
> probably be to provide a simulated EGD inside the jail).

That's what we did, eventually, since cryptlib has access to no other
entropy sources inside the jail. We open /dev/urandom at startup, keep
the fd around while we (fork and) chroot, and then feed cryptlib data
from that fd on /var/run/egd-pool.

It works all right (though it's *awfully* slow unless we increase the
quality assigned to EGD data in random/unix.c).

-- ams

_______________________________________________
Cryptlib mailing list
Cryptlib@... via Mail: cryptlib-request@...
Archive: ftp://ftp.franken.de/pub/crypt/cryptlib/archives/
http://news.gmane.org/gmane.comp.encryption.cryptlib
Posts from non-subscribed addresses are blocked to prevent spam, please
subscribe in order to post messages.