Skip site navigation (1) Skip section navigation (2)

[PATCH] Re: Why READ ONLY transactions?

From: Sean Chittenden <sean(at)chittenden(dot)org>
To: Christopher Browne <cbbrowne(at)libertyrms(dot)info>,pgsql-patches(at)postgresql(dot)org
Cc: Robert Treat <xzilla(at)users(dot)sourceforge(dot)net>,Gavin Sherry <swm(at)linuxworld(dot)com(dot)au>, pgsql-advocacy(at)postgresql(dot)org
Subject: [PATCH] Re: Why READ ONLY transactions?
Date: 2003-07-30 05:05:34
Message-ID: (view raw, whole thread or download thread mbox)
Lists: pgsql-advocacypgsql-hackerspgsql-patches
> >>>>    - Read only transactions, bringing a greater level of
> >>>>    security to web and enterprise applications by protecting
> >>>>    data from modification.
> >> This should be removed. Even though I added it to the press
> >> release, I've just realised it's not really a security measure
> >> against SQL injection since injected code can just specify 'SET
> >> TRANSACTION READ WRITE'. We should still mention it, but not as a
> >> security measure.
> > Aside from spec compliance, whats the bonus for having it then? Or
> > put a better way, why/when would I want to use this?
> If I am writing a "report program" that isn't supposed to do any
> updates to anything, then I would be quite happy to set things to
> READ-ONLY as it means that I won't _accidentally_ do updates.
> It's like adding a pair of suspenders to your wardrobe.  You can
> _always_, if you really try, get your pants to fall down, but this
> provides some protection.
> I would NOT call it a "security" provision, as it is fairly easily
> defeated using SET TRANSACTION.

Um, why not make it an actual full blown security feature by applying
the following patch?  This gives PostgreSQL real read only
transactions that users can't escape from.  Notes about the patch:

*) If the GUC transaction_force_read_only is set to FALSE, nothing
   changes in PostgreSQL's behavior.  The default is FALSE, letting
   users change from READ ONLY to READ WRITE at will.

*) If transaction_force_read_only is TRUE, this sandboxes the
   connection for the remainder of the connection if the session is
   set to read only.  The following bits apply:

   a) if you're a super user, you can change transaction_read_only.

   b) if you're not a super user, you can change transaction_read_only
      to true.

   c) if you're not a super user, you can always change
      transaction_read_only from false to true.  If
      transaction_force_read_only is true, you can't change
      transaction_read_only from true to false.

   d) If transaction_force_read_only is TRUE, but
      transaction_read_only is FALSE, the transaction is still READ

   e) Only super users can change transaction_force_read_only.

Basically, if you want to permanently prevent a user from ever being
able to get in a non-read only transaction, do:

\c [dbname] [db_superuser]
ALTER USER test SET default_transaction_read_only TO TRUE;
ALTER USER test SET transaction_force_read_only TO TRUE;

-- To test:
regression=# \c regression test
regression=> SHOW transaction_read_only;
(1 row)

regression=> SHOW transaction_force_read_only;
(1 row)

regression=> SET transaction_read_only TO FALSE;
ERROR:  Insufficient privileges to SET transaction_read_only TO FALSE

It's also possible to set transaction_force_read_only in
postgresql.conf making it possible to create read only databases to
non-superusers by starting postgresql with
default_transaction_read_only and transaction_force_read_only set to
TRUE.  If this patch is well received, I'll quickly bang out some
documentation for this new GUC.  From a security stand point, this is
a nifty feature.  -sc

Sean Chittenden

In response to


pgsql-hackers by date

Next:From: Peter EisentrautDate: 2003-07-30 07:54:20
Subject: Re: [HACKERS] Passing server_encoding to the client is not future-proof
Previous:From: Bruce MomjianDate: 2003-07-30 04:38:54
Subject: Re: Upgrading my BSDI box, again

pgsql-patches by date

Next:From: Fernando NasserDate: 2003-07-30 15:24:40
Subject: Re: JDBC stored procs doc patch
Previous:From: Bruce MomjianDate: 2003-07-30 01:21:33
Subject: Re: Datetime patch

pgsql-advocacy by date

Next:From: KaarelDate: 2003-07-30 07:01:46
Subject: Re: WebGUI vote on whether to drop PostgreSQL support
Previous:From: The Hermit HackerDate: 2003-07-30 03:34:23
Subject: Re: WebGUI vote on whether to drop PostgreSQL

Privacy Policy | About PostgreSQL
Copyright © 1996-2017 The PostgreSQL Global Development Group