Cryptographic weakness on Debian systems
Posted May 13, 2008 20:30 UTC (Tue) by
lambda (subscriber, #40735)
In reply to:
Cryptographic weakness on Debian systems by bronson
Parent article:
Cryptographic weakness on Debian systems
There were two lines that the Debian maintainer commented out; one that was pulling in
some entropy from uninitialized buffers, and one that was the actual interface to seed the
random number generator. It was perfectly reasonable to remove the line that was pulling in
entropy from an uninitialized buffer; it was in the ssleay_rand_bytes function,
which is supposed to use the provided buffer to output random bytes, and for some reason it
happened to be using that uninitialized data to mix a little bit of extra entropy into the pool
(you're not going to get very good entropy from that, but it's not particularly harmful, other than
making tools like Valgrind complain).
But in the other case, in ssleay_rand_add, the buffer is an input buffer, and it is
the very function used to seed the random number generator with actual entropy. Commenting
that line out was completely and utterly wrong, and if someone was providing that function with
an uninitialized buffer, it's the call site that should have been fixed, not that function. The
amazing thing is that this passed whatever review processes Debian has in place, was sent to the
openssl-dev mailing list, and still no one noticed. The openssl people are claiming that openssl-
dev is the wrong place to send it, but even still, some of the people who replied on that thread
have @openssl.org addresses, so it's a fairly reasonable assumption to make that the openssl
developers did read that thread and had no problem with the patch.
(
Log in to post comments)