2.3.14: posting to shared mailbox results in 550 Permission denied

[Simon Matter]

> On Fri, May 29, 2009 at 04:51:58PM +0200, Simon Matter wrote:
>> In auth_unix.c/mymemberof() I have added the following line at the
>> beginning of the function:
>>
>> syslog(LOG_ERR, "mymemberof() userid: %s, identifier: %s",
>> auth_state->userid, identifier);
>
> Yikes.  Seriously?  auth_state->userid huh?
>
>> Now I see the following difference:
>>
>> 2.3.13:
>> May 29 16:47:42 client140 lmtp[16435]: mymemberof() userid: (null),
>> identifier: anyone
>> May 29 16:47:42 client140 lmtp[16435]: mymemberof() userid: (null),
>> identifier: anonymous
>>
>> 2.3.14:
>> May 29 16:44:42 client140 lmtp[16145]: mymemberof() userid: ,
>> identifier:
>> anyone
>> May 29 16:44:42 client140 lmtp[16145]: mymemberof() userid: ,
>> identifier:
>> anonymous
>>
>> I think what you said should be NULL can be seen here in 2.3.13 but not
>> in
>> 2.3.14. What do you think?
>
> I think it's surprising that you don't get a segfault dereferencing a
> NULL pointer for auth_state in the first place, in either of them.
> But it's even more odd that you're seeing differences with userid.
> It being "" (a pointer to a zero byte) is pretty bad.
>
> I'd better go back re-read the code.
>
> By the way, on my testbed running current 2.3.14 based code:
>
> RCPT TO:<+shared.foo>
> 250 2.1.5 ok
>
> # cyr slott02 dbtool show shared.foo
> shared.foo      0 default anyone        lrs     anonymous       p
>
> This is 32 bit.
>
> When I debug it, I get a NULL pointer for auth_state at verify_user.

Maybe it's a stupid question but doesn't the NULL mean anonymous?
When you say 'current 2.3.14' do you mean current CVS/GIT? I think current
CVS is fine as well as 2.3.13. CVS seems fixed by the path Gabor pointed
out. At least I have added the patch from GIT to my RPM and will test it
on different systems today.

Simon