I think the argument is not with the change of authentication, which
I agree is more secure, but with the default configuration for the
default domain. It used to be that "admin" requires a password
(adminadmin), but now it doesn't, which breaks existing tests. I
think users will complain about this too.
I can make a test to run if I remove "--user admin --passwordfile <f>",
like Shing Wai suggested, or make <f> an empty file. Strangely, if
<f> contains a single line
AS_ADMIN_PASSWORD=""
it still fails.
On 09/14/09 21:20, Bill Shannon wrote:
> Sorry guys, I'm just getting back to the thread.
>
> While this change is working as intended, the thing that surprises me
> is how many places have *incorrect* authentication information for the
> domain. If you pass in *no* authentication information (which is what
> I was expecting people were doing for anonymous login), then it pretty
> much works as before. But if you pass in *incorrect* authentication
> information (wrong username or wrong password), it now fails. This is
> to avoid the case where you login to a domain using what you know are
> the correct username and password, and the login succeeds, but in fact
> a configuration error in the domain set it up for anonymous/unauthenticated
> access. Before this change you would have a false sense of security and
> the configuration error would go undetected. After the change the error
> is detected.
>
> If you have a password file with an incorrect password for the domain,
> you need to fix that, most likely by simply removing the password from
> the password file.
>
> If you've logged into the domain using "asadmin login", or created a
> domain using the --savelogin option, you might have a saved password
> that's now incorrect. You can either remove the ~/.asadminpass file
> or use "asadmin login" to specify the correct password.
>
> Hopefully the problems caused by this change are just transitory until
> all the configuration errors are corrected. If it proves to be more of
> a problem than that, we can consider restoring the previous behavior, at
> the cost of reintroducing the "false sense of security" problem.
>
>
> Kedar Mhaswade wrote on 9/14/09 6:35 PM:
>> Kin-man Chung wrote:
>>> In the interest of backward compatibility, asadmin should just ignore
>>> the --passwordfile options if the domain was created with no password.
>>> Otherwise a lot of scripts and ant tasks will no longer work. This is
>>> a serious problem for devtests, which need to run in various versions
>>> of the appserver.
>> Probably you are right and we might have to do something if this
>> remains confusing but I don't think there is any v2-compatibility issue
>> here. After talking to Bill, I agree with him that the change he's made
>> is correct and makes the domain more secure. (A long-winding explanation
>> ...)
>>
>> The only "change" is that in v2, the setup.xml
>> used to create the domain with "admin"/"adminadmin" and in v3 the
>> .zip bundles create a preconfigured domain with "admin"/no-password.
>>
>> v3-final is going to be incompatible with earlier releases of v3
>> (Prelude/Preview) because v3 (Prelude/Preview) ignored all
>> user names/passwords with default domain (that is bundled in
>> the web.zip/glassfish.zip bundles). I am not sure if we must
>> preserve v3-final's compatibility with Preview/Prelude.
>>
>> One idea is to make the password for "admin" user "adminadmin"
>> instead of no-password, which will fix most of the tests, but
>> this idea is is rather weird too (conflicts with --no-password
>> option on create-domain).
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe_at_glassfish.dev.java.net
> For additional commands, e-mail: dev-help_at_glassfish.dev.java.net
>