Oracle 8i compromise questions

[Jack Donovan <jake.donovan () gmail com>]

Hello all,

A client of mine reported a compromise of an outdated Oracle 8i
(8.174) database server running on Windows 2000, which they wanted me
to try and figure out the root cause of.  The compromise was very
generic, and allowed the attackers  OS access which they used to
install some standard backdoor stuff and grab the password hashes for
the system which they broke overnight offline (presumably via
RainbowCrack).  We know this because the usernames and passwords were
attempted against other systems belonging to the client, albeit
unsuccessfully.  My knowledge of Oracle is limited, but I assumed it
would be easy to spot the initial event, since the attacker's
themselves were not very careful.

After a quick analysis of the system and network data, the earliest
action I can find appears to be a PWD<database_name>.ORA file that was
overwritten, which is the ORACLE Remote Password File allowing folks
remote access as SYSDBA.  Shortly after the timestamp on the ORA file,
there was a successful login entry in the application log to the
Oracle as as the sys user, as follows:

Application,Oracle.<dbname>,INFORMATION,<host>,<date/time>,34,None,Audit
trail: ACTION : 'connect sys' OSPRIV : DBA CLIENT USER:
<attacker_login_name_presumably> CLIENT TERMINAL: <attacker_host> 
STATUS: SUCCEEDED ( 0 )  .

> From what I understand, any user found in the PWD<db>.ORA file will
show up in the logs as 'connect sys', so this makes sense.  There was
a similar login attempt that failed roughly 10 minutes before this,
though Oracle didn't keep track of the attacker host in the failed
attempt.  Within roughly 1 minute of the successfull Oracle login, the
attacker's toolkit was uploaded (based on the timestamps I've got).

The questions I have are:

1) What is the most common attack method used for an Oracle database
that would result in this file being overwritten?  Is it likely this
was the attack vector, or a backup access method with some other
method being used to initially break the system?

2) If the overwritten authentication and subsequent login to the
database was the method of entry, how did the attackers get local
filesystem access?  I know how to do this in MSSQL etc., and I know
there's the PL/SQL stuff in Oracle, but it seems nontrivial to me how
one would go about using that for this purpose.

My assumption is these are somewhat basic questions, but my searching
around Google hasn't turned up anything concrete yet.  I found a
number of listener vulnerabilities for doing things like arbitrarily
assigning the log file and overwriting, but most of them seem to be
circa 2000, and I'm fairly certain the system had been patched since
then (though the exact patch status of the system was unknown).

It's possible there is some buffer overflow at work, but the service
itself didn't die, and there was no error in the system , security, or
application logs that would imply such an attack.  My initial thought
when the compromise was reported was brute-force password attack,
because the password used was terrible.  But the overwritten PWD file
seems counter to that theory.

The attacker's did not appear terribly subtle (nice bright shiny FTP
banners), so the simplest solution here is probably the most likely. 
Any thoughts on what that solution might be, or pointers to where I
should be looking, would be much appreciated.

Thanks,
J