> Brian Bennett <[email protected]> wrote:
> 
> Not to get too far off topic, but is there actually any evidence to back
> up that statement? I've been searching for a number of years for someone
> who can speak intelligently on the topic. As I understand it, statements
> like this are parroted simply due to key sizes. While ssh-keygen can only
> create DSA keys of 1024 bits, openssl can generate arbitrarily large DSA
> keys that can be used with OpenSSH.

From what I understand after reading openssh-unix-dev for the last few
years, the biggest issue with DSA for SSH is that the SSH RFC mandates the
use of SHA-1 hashes only for DSA keys. At the same time, the NIST spec for
DSA mandates that for keys >1024 bits you must not use SHA-1. This is why
the OpenSSH keygen utility doesn't allow DSA keys >1024 bits, because using
them has to either violate the SSH RFC, or the NIST standards for DSA.

The OpenSSH developers see replacing DSA with ECDSA as a way out of this
problem, since the spec for ECDSA support includes longer keys and better
hashes. Thus they consider support for plain DSA deprecated in favour of
moving to ECDSA.

So it's not really out of some concern for a specific weakness in DSA
itself, but rather a combination of factors:

* Their hands are tied with respect to increasing its key size without violating
  standards,
* 1024-bit keys are starting to look a bit weak, and
* They have already given people an alternative path forwards (ECDSA)

As a result, I don't really consider it a huge security risk to leave DSA
support on in our builds by default specifically for backwards-compat,
though I acknowledge that in general discouraging the generation of new
1024-bit DSA keys by users is probably wise.

Additionally, if we went ahead and shipped a SmartOS image with DSA disabled
by default, and our users have DSA keys in their authorized_keys as their
only means of remote access to a server, the change would lock them out.

In other distros and operating systems, this could be handled with a config
migration script of some kind, or a warning before upgrading. However, with
the nature of SmartOS as a live distro running from an immutable image,
neither of these mechanisms are really established and available.

Since the "failure mode" here (locking users out of their server's gz after
an upgrade) is not acceptable, and we don't have a good way around it ready
to use today, I consider that re-enabling DSA keys is worthwhile, though I
am certainly open to re-visiting this decision now and in the future if more
information or evidence becomes available.

Hope that explanation helps.

-------------------------------------------
smartos-discuss
Archives: https://www.listbox.com/member/archive/184463/=now
RSS Feed: https://www.listbox.com/member/archive/rss/184463/25769125-55cfbc00
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=25769125&id_secret=25769125-7688e9fb
Powered by Listbox: http://www.listbox.com

Reply via email to