[ https://issues.apache.org/jira/browse/CASSANDRA-13626?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16146990#comment-16146990 ]
Sam Tunnicliffe edited comment on CASSANDRA-13626 at 8/30/17 10:01 AM: ----------------------------------------------------------------------- bq.checkpw does an even worse job of failing nicely In what way(s)? Checking with a few strings that would be caught by this validation, {{checkpw}} seems to behave as expected. This validation would help us detect that we have stored invalid hash values, so that could be useful in diagnosing unexpected auth failures & debugging their causes. That will obviously require logging when the validation fails before throwing the {{AuthenticationException}}, so we should separate it from the actual {{checkPw}} call. On the actual validation, the 22 character component is actually the salt, not the cost - the bcrypt format is {{$<id>$<cost>$<salt><digest>}}. Cost, salt and digest are all fixed length (2, 22 & 31 chars repectively), whereas id may be 1 or 2 chars, though we have only ever used a version of jbcrypt that generates the 2 char variant. So we could simplify that check to {{length == 60}}. If {{checkpw}} *is* correctly returning false when the stored hash is invalid though, we only really need to do the validation on failure, in which case we could a more thorough check than simply looking at the length, if that's worthwhile. EDIT: I know it's a pretty trivial change, but it would be nice to add some tests to go with it. was (Author: beobal): bq.checkpw does an even worse job of failing nicely In what way(s)? Checking with a few strings that would be caught by this validation, {{checkpw}} seems to behave as expected. This validation would help us detect that we have stored invalid hash values, so that could be useful in diagnosing unexpected auth failures & debugging their causes. That will obviously require logging when the validation fails before throwing the {{AuthenticationException}}, so we should separate it from the actual {{checkPw}} call. On the actual validation, the 22 character component is actually the salt, not the cost - the bcrypt format is {{$<id>$<cost>$<salt><digest>}}. Cost, salt and digest are all fixed length (2, 22 & 31 chars repectively), whereas id may be 1 or 2 chars, though we have only ever used a version of jbcrypt that generates the 2 char variant. So we could simplify that check to {{length == 60}}. If {{checkpw}} *is* correctly returning false when the stored hash is invalid though, we only really need to do the validation on failure, in which case we could a more thorough check than simply looking at the length, if that's worthwhile. > Check hashed password matches expected bcrypt hash format before checking > ------------------------------------------------------------------------- > > Key: CASSANDRA-13626 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13626 > Project: Cassandra > Issue Type: Bug > Components: Auth > Reporter: Jeff Jirsa > Assignee: Jeff Jirsa > Priority: Minor > Fix For: 3.0.x, 3.11.x, 4.x > > > We use {{Bcrypt.checkpw}} in the auth subsystem, but do a reasonably poor job > of guaranteeing that the hashed password we send to it is really a hashed > password, and {{checkpw}} does an even worse job of failing nicely. We should > at least sanity check the hash complies with the expected format prior to > validating. -- This message was sent by Atlassian JIRA (v6.4.14#64029) --------------------------------------------------------------------- To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org