Re: Serious problem within authentication subsystem in 7.0

Hi,

Ok, I know the mailing list web page states: YOU MUST TRY ELSEWHERE FIRST! and
this should, technically speaking, be reported as a bug however given that the
security implications are potentially severe I thought here would be best in
the first instance.

I have only briefly looked into this problem as I have just now discovered it.
Essentially, in our environment, we require password authentication as a
defacto. However it appears that once a user has authenticated with the
backend it is possible for that user to trivially assume root dba privileges or
privileges of any other dba user.

To demonstrate the problem:

Consider two systems:

pgsqlserver 192.168.1.1 - backend system
pgsqlclient 192.168.1.2 - client system

Our pg_hba.conf (on pgsqlserver) now looks something similar to:

local all password
host all 127.0.0.1 255.255.255.255 password
host all 192.168.1.2 255.255.255.255 password

Now making connections from pgsqlclient (192.168.1.2) would require password
authentication. To show that this works, entering an incorrect passwd...

pgsqlclient:/home/matt 11:33am > psql -h pgsqlserver -U matt matt
Password:
psql: Password authentication failed for user 'matt'

Now a correct password:

pgsqlclient:/home/matt 11:36am > psql -h pgsqlserver -U matt matt
Password:
Welcome to psql, the PostgreSQL interactive terminal.

Type: \copyright for distribution terms
\h for help with SQL commands
\? for help on internal slash commands
\g or terminate with semicolon to execute query
\q to quit

matt=>

Ok, so at the top level password authentication works, now to the problem...

Once authenticated it is possible to trivially assume another users identity
without further authentication e.g.

matt=> \c template1 postgres
You are now connected to database template1 as user postgres.
template1=#

Or, assume any other users identity:

matt=> \c www www
You are now connected to database www as user www.
www=>

Ouch.

I have not tested to see if this is specific to the password authentication
method or a general problem relating to any of the supported methods and I have
limited time to investigate this at the moment.

Is there anyone who specifically maintains the authentication subsystem that I
could communicate with directly? I would be interested to offer whatever
assistance I can.

Regards,
Matt.

Matt Sullivan <matt(at)sullivan(dot)gen(dot)nz> writes:
> Essentially, in our environment, we require password authentication as
> a defacto. However it appears that once a user has authenticated with
> the backend it is possible for that user to trivially assume root dba
> privileges or privileges of any other dba user.

It appears that psql will auto-supply the previously entered password,
so if you were using the same password for all your accounts then this
might happen. Otherwise it's pretty hard to believe. That new
connection is to a new backend; there's no way for it to know that you
were previously connected.

Offhand I think it would be a good idea for psql to insist on a new
password if the \connect command gives a new user name...

regards, tom lane

On Mon, 22 May 2000, Tom Lane wrote:

> Matt Sullivan <matt(at)sullivan(dot)gen(dot)nz> writes:
> > Essentially, in our environment, we require password authentication as
> > a defacto. However it appears that once a user has authenticated with
> > the backend it is possible for that user to trivially assume root dba
> > privileges or privileges of any other dba user.
>
> It appears that psql will auto-supply the previously entered password,
> so if you were using the same password for all your accounts then this
> might happen. Otherwise it's pretty hard to believe. That new
> connection is to a new backend; there's no way for it to know that you
> were previously connected.
>
> Offhand I think it would be a good idea for psql to insist on a new
> password if the \connect command gives a new user name...

Ok, phew...

matt=> \c wwwdata wwwdata
Password authentication failed for user 'wwwdata'
Previous connection kept
matt=>

This would infer though that the passwd data is cached within each instance of
psql which could present it's own set of security risks.

I would think that it should probably be *forgotton* after authentication is
established and required on any new \connect. This might present some issues
with pg_dump etc. I guess though.

Matt.

Matt Sullivan <matt(at)sullivan(dot)gen(dot)nz> writes:
> This would infer though that the passwd data is cached within each
> instance of psql which could present it's own set of security risks.

Yeah. There's been discussion about that, and the consensus seems to
be that the advantages outweigh the very small risks. (One of the
disadvantages of forgetting the password is that PQreset can't work...)

> I would think that it should probably be *forgotton* after
> authentication is established and required on any new \connect. This
> might present some issues with pg_dump etc. I guess though.

pg_dump is actually pretty nearly useless in a password-auth
installation; to run the restore script, you'd have to manually enter
a password each time it hits a \connect command :-(.

The best idea I've heard for fixing this is to invent a quasi-suid
mechanism: the pg_dump script would be started as postgres (enter
password for same, once) and it would NOT do any \connect commands.
Instead it would issue some kind of "SET effective_user = 'name'"
command, which would determine the ownership assigned to subsequently-
created objects, but the backend would still remember that the user
was "really" postgres. Presumably this SET command would only be
allowed to superusers, so the backend must remember that the user is
really postgres, or it'll reject SET effective_user commands after
the first one.

The devil is in the details, of course, and the details here would be
to figure out which operations should pay attention to effective_user
and which to the true userid. But it seems doable.

(Hey Peter, wanna put this on your todo list for that privilege-system
work?)

regards, tom lane