[
https://issues.apache.org/jira/browse/ACCUMULO-4044?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14981088#comment-14981088
]
Christopher Tubbs commented on ACCUMULO-4044:
---------------------------------------------
When UnixCrypt (truncated 8x7-bit chars + 2 byte salt with DES) was replaced
with MD5 (and now SHA-256 and SHA-512), a thin wrapper format was established
for compatibility. The old DES password format was replaced with
{{$<id>$<salt>$<hash>}} where {{<id>}} was a numerical identifier of the
algorithm to use (1=MD5, 6=SHA512). This format was established to be
backwards-compatible and future-proof, and is itself (in a way) a "thin
wrapper" around the hash+salted bytes already.
Ideally, I'd like to not create a "thin wrapper" around this already existing
wrapper format... it's unnecessarily complicated and adds code better suited to
the library we're using. The most we should have to do is pull out the bytes
using UTF-8 and send them to the library. If our format cannot start with
{{$}}, then we're free to replace our format directly and distinguish between
our old format and the new, while gaining the future-proofing benefits of this
standardized format. The only difference is that anything not in this format
will be assumed to be SHA-256 with 8-byte salt instead of the UnixCrypt DES
format. If our format can start with {{$}} (or is likely to... I'm okay with
forcing some users to reset their password if it's highly improbable), then we
do need to add a wrapper on top of the wrapper.
As for doing it for all the data in ZK, maybe that's a good idea, too. I'll
withhold judgment until that use case is articulated in context.
> Stronger/standardized password hashing
> --------------------------------------
>
> Key: ACCUMULO-4044
> URL: https://issues.apache.org/jira/browse/ACCUMULO-4044
> Project: Accumulo
> Issue Type: Improvement
> Reporter: Christopher Tubbs
> Fix For: 1.8.0
>
>
> Currently, Accumulo stores hashed passwords using SHA-256 and an 8-byte salt,
> in a custom output format.
> Instead, we should switch to using commons-codec's Crypt class to create
> crypt(3) style hashes, the default of which is to use SHA-512 with a 16-byte
> salt. The format is stored in a standard way, with an identifier to determine
> the hashing method which was used.
> We'd have to make sure that we can tell the difference between the new format
> and the old format, so we know how to properly verify user credentials. This
> would be easy if we stored the new form in a different zookeeper node, but we
> could also use a delimiter (not a fan of the delimiter, personally, because
> I'd prefer the standard format, unmodified). We might be able to
> automatically migrate to the new format upon authentication, so we can
> eventually drop the old format entirely++.
> ++ When we do eventually drop the old format, users will need to reset their
> passwords, or have an admin user do it for them. This shouldn't be a big
> issue if we wait a sufficient number of releases to drop the old format.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
