Thread: Pros and cons of giving someone superuser privilege
Hi all,
I'm part of a service where we provide users with their own PostgreSQL instances. The idea is that we provide them with a website to request and manage their databases (start/stop, backups, restores, upgrades, monitoring, etc). By doing this, we avoid having to give them access to the machine where their database is running, as this would be a security concern. But in the end, the user is the sole responsible for the database.
At the moment we create an "admin" user for them and give it "createdb" and "createrole" privileges. My question is, in case we give that user the superuser privilege, what would the repercussion be concerning security (as in accessing data on the machine for example)? And what advantages would the user acquire by having that privilege (considering major admin functionality is provided via the website)?
Thanks in advance for your replies.
Cheers,
Daniel
Daniel Gomez Blanco <nanodgb@gmail.com> writes:
> I'm part of a service where we provide users with their own PostgreSQL
> instances. The idea is that we provide them with a website to request and
> manage their databases (start/stop, backups, restores, upgrades,
> monitoring, etc). By doing this, we avoid having to give them access to the
> machine where their database is running, as this would be a security
> concern. But in the end, the user is the sole responsible for the database.
> At the moment we create an "admin" user for them and give it "createdb" and
> "createrole" privileges. My question is, in case we give that user the
> superuser privilege, what would the repercussion be concerning security (as
> in accessing data on the machine for example)? And what advantages would
> the user acquire by having that privilege (considering major admin
> functionality is provided via the website)?
If you're not willing to give them a shell login on the database server,
you'd better not give them superuser, because the latter is easily
parlayed into the former on most platforms. (It's not trivial
necessarily, but we do not consider it a security bug if you can.)
However, it's also generally assumed that a person with DBA
responsibilities should be able to log in as the O/S user that runs the
database server (not root, but the postgres O/S user). So if your service
is telling people that they are "solely responsible", maybe you should
rethink that end of it.
The main functionality omission I'm noticing in your sketch is server
configuration adjustment and installation of database extensions ...
regards, tom lane
Thanks for your prompt reply! When we say they're responsible for their databases we mean in terms of the users/roles/privileges they create, and the data they're able to access. But we are responsible for OS patches, database availability, making sure that backups and restores are taken consistently, etc.
Concerning the configuration adjustment, we also provide them with a way to change the configuration (postgresql.conf and pg_hba.conf) on the web interface. Of course there are some vetoed parameters that they are not allowed to change. Basically, we want to provide all the admin functionality, for which you'd need access to the machine, via our web interface. At the moment we offer: start/stop, backups (and backups to tape), scheduling of backups, restores, configuration management, download of logs, one-click upgrades and access to monitoring. The only thing we would have to do for them is installing database extensions, but that rarely happens in our environment.
From your reply and considering why I have just said, I think that, in our case, these users would be better off without superuser access. As for their day-to-day operations they wouldn't need it as long as they can create their databases and roles, and execute the rest of operations on the web interface.
Cheers,
Daniel
On 25 April 2014 16:02, Tom Lane <tgl@sss.pgh.pa.us> wrote:
If you're not willing to give them a shell login on the database server,Daniel Gomez Blanco <nanodgb@gmail.com> writes:
> I'm part of a service where we provide users with their own PostgreSQL
> instances. The idea is that we provide them with a website to request and
> manage their databases (start/stop, backups, restores, upgrades,
> monitoring, etc). By doing this, we avoid having to give them access to the
> machine where their database is running, as this would be a security
> concern. But in the end, the user is the sole responsible for the database.
> At the moment we create an "admin" user for them and give it "createdb" and
> "createrole" privileges. My question is, in case we give that user the
> superuser privilege, what would the repercussion be concerning security (as
> in accessing data on the machine for example)? And what advantages would
> the user acquire by having that privilege (considering major admin
> functionality is provided via the website)?
you'd better not give them superuser, because the latter is easily
parlayed into the former on most platforms. (It's not trivial
necessarily, but we do not consider it a security bug if you can.)
However, it's also generally assumed that a person with DBA
responsibilities should be able to log in as the O/S user that runs the
database server (not root, but the postgres O/S user). So if your service
is telling people that they are "solely responsible", maybe you should
rethink that end of it.
The main functionality omission I'm noticing in your sketch is server
configuration adjustment and installation of database extensions ...
regards, tom lane
* Daniel Gomez Blanco (nanodgb@gmail.com) wrote:
> >From your reply and considering why I have just said, I think that, in our
> case, these users would be better off without superuser access. As for
> their day-to-day operations they wouldn't need it as long as they can
> create their databases and roles, and execute the rest of operations on the
> web interface.
Just to hit on this again- that means your users can't install any
extensions (not even those provided as part of PG through contrib).
Eventually, users will get upset by the lack of things like PostGIS.
They also can't create stored procedures using any of the untrusted
languages (of course, if they could, they could get access to the host,
so that may be a show-stopper anyway).
Thanks,
Stephen
Attachment
We provide some extensions by default, and would be willing to install more extensions if needed. The point about stored procedures is a good point though.
Thanks for your input!
Daniel
On 25 April 2014 16:52, Stephen Frost <sfrost@snowman.net> wrote:
* Daniel Gomez Blanco (nanodgb@gmail.com) wrote:Just to hit on this again- that means your users can't install any
> >From your reply and considering why I have just said, I think that, in our
> case, these users would be better off without superuser access. As for
> their day-to-day operations they wouldn't need it as long as they can
> create their databases and roles, and execute the rest of operations on the
> web interface.
extensions (not even those provided as part of PG through contrib).
Eventually, users will get upset by the lack of things like PostGIS.
They also can't create stored procedures using any of the untrusted
languages (of course, if they could, they could get access to the host,
so that may be a show-stopper anyway).
Thanks,
Stephen
On Fri, Apr 25, 2014 at 03:46:52PM +0200, Daniel Gomez Blanco wrote: > Hi all, > > I'm part of a service where we provide users with their own PostgreSQL > instances. The idea is that we provide them with a website to request and > manage their databases (start/stop, backups, restores, upgrades, monitoring, > etc). By doing this, we avoid having to give them access to the machine where > their database is running, as this would be a security concern. But in the end, > the user is the sole responsible for the database. > > At the moment we create an "admin" user for them and give it "createdb" and > "createrole" privileges. My question is, in case we give that user the > superuser privilege, what would the repercussion be concerning security (as in Have you considered that your users can _create_ superusers? I think modified Amazon Postgres blocks that, but native Postgres does not. -- Bruce Momjian <bruce@momjian.us> http://momjian.us EnterpriseDB http://enterprisedb.com + Everyone has their own god. +