postgres_fdw super user checks

On Mon, Oct 17, 2016 at 3:33 AM, Jeff Janes <jeff.janes@gmail.com> wrote:

postgres_fdw has some checks to enforce that non-superusers must connect to
the foreign server with a password-based method. The reason for this is to
prevent the authentication to the foreign server from happening on the basis
of the OS user who is running the non-foreign server.

But I think these super user checks should be run against the userid of the
USER MAPPING being used for the connection, not the userid of currently
logged on user.

So, if the user mapping user is a superuser locally, this would allow
any lambda user of the local server to attempt a connection to the
remote server. It looks dangerous rather dangerous to me to authorize
that, even if the current behavior is a bit inconsistent I agree.

Your patch breaks the join pushdown logic when multiple user IDs are
involved. Per se make check.
--
Michael

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

On Mon, Oct 17, 2016 at 12:03 AM, Jeff Janes <jeff.janes@gmail.com> wrote:

postgres_fdw has some checks to enforce that non-superusers must connect to
the foreign server with a password-based method. The reason for this is to
prevent the authentication to the foreign server from happening on the basis
of the OS user who is running the non-foreign server.

But I think these super user checks should be run against the userid of the
USER MAPPING being used for the connection, not the userid of currently
logged on user.

That is, I think the last line in this script should succeed: ('jjanes' is
both a superuser, and a database):

CREATE EXTENSION IF NOT EXISTS postgres_fdw WITH SCHEMA public;
CREATE SERVER foo FOREIGN DATA WRAPPER postgres_fdw;
CREATE USER MAPPING FOR jjanes SERVER foo;
CREATE TABLE foobar1 ( x integer);
CREATE FOREIGN TABLE foobar2 ( x integer) SERVER foo OPTIONS ( table_name
'foobar1');
CREATE VIEW foobar3 AS SELECT foobar2.x FROM foobar2;
CREATE USER test;
GRANT SELECT ON TABLE foobar3 TO test;
\c jjanes test
select * from foobar3;

It connects back to itself, simply for demonstration purposes.

The attached patch implements this change in auth checking.

I agree with your analysis, that any passwordless foreign server
access with super user's user mapping should be allowed. If it's safe
to access a foreign server without password for a superuser, then it
should be safe to do so when corresponding user mapping is used even
when login user is non-superuser.

But there's one problem with the patch.

login as some useruser and run following commands.

create extension postgres_fdw;
create server foo foreign data wrapper postgres_fdw options (dbname 'postgres');
create user test;
grant USAGE ON FOREIGN server foo to test;
set role test;
create user mapping for test server foo;
create foreign table fpg_class (oid oid) server foo options
(table_name 'pg_class', schema_name 'pg_catalog');
create view fview as select * from fpg_class;
set role <some superuser>;
select * from fview limit 0;

With your patch it gives error
ERROR: password is required
DETAIL: Non-superuser cannot connect if the server does not request a password.
HINT: Target server's authentication method must be changed.

Without the patch it does not give any error.

Is that intentional?

I guess, this is because of asymmetry in check_conn_params() and
connect_pg_server(). The first one does not check any params if the
logged in user is a superuser but the later checks if only the user in
the mapping is superuser.

--
Best Wishes,
Ashutosh Bapat
EnterpriseDB Corporation
The Postgres Database Company

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

On Mon, Oct 17, 2016 at 11:48 AM, Michael Paquier
<michael.paquier@gmail.com> wrote:

On Mon, Oct 17, 2016 at 3:33 AM, Jeff Janes <jeff.janes@gmail.com> wrote:

postgres_fdw has some checks to enforce that non-superusers must connect to
the foreign server with a password-based method. The reason for this is to
prevent the authentication to the foreign server from happening on the basis
of the OS user who is running the non-foreign server.

But I think these super user checks should be run against the userid of the
USER MAPPING being used for the connection, not the userid of currently
logged on user.

So, if the user mapping user is a superuser locally, this would allow
any lambda user of the local server to attempt a connection to the
remote server. It looks dangerous rather dangerous to me to authorize
that, even if the current behavior is a bit inconsistent I agree.

A lambda user can use a user mapping same as a superuser if a. that
user mapping is public and/or b. it uses a view owned by super user
(RangeTblEntry::checkuser). When a is true but not b, the the user in
UserMapping is set to lambda and not superuser, so this patch is
correct here. If b is true, and lambda is able to access the view, the
superuser has granted it permissions to do so and thus intends to let
lambda use a super user user mapping. Since we trust super users to do
the right thing, I don't see why it's unsafe. Any other objects
accesses by lambda, will use a different user mapping based on the
object being accessed.

Best Wishes,
Ashutosh Bapat
EnterpriseDB Corporation
The Postgres Database Company

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

On Mon, Oct 17, 2016 at 2:18 AM, Michael Paquier
<michael.paquier@gmail.com> wrote:

On Mon, Oct 17, 2016 at 3:33 AM, Jeff Janes <jeff.janes@gmail.com> wrote:

postgres_fdw has some checks to enforce that non-superusers must connect to
the foreign server with a password-based method. The reason for this is to
prevent the authentication to the foreign server from happening on the basis
of the OS user who is running the non-foreign server.

But I think these super user checks should be run against the userid of the
USER MAPPING being used for the connection, not the userid of currently
logged on user.

So, if the user mapping user is a superuser locally, this would allow
any lambda user of the local server to attempt a connection to the
remote server. It looks dangerous rather dangerous to me to authorize
that, even if the current behavior is a bit inconsistent I agree.

I don't know what "any lambda user" means. Did you mean to write "any
random user"?

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

On Mon, Oct 17, 2016 at 10:51 PM, Robert Haas <robertmhaas@gmail.com> wrote:

On Mon, Oct 17, 2016 at 2:18 AM, Michael Paquier
<michael.paquier@gmail.com> wrote:

On Mon, Oct 17, 2016 at 3:33 AM, Jeff Janes <jeff.janes@gmail.com> wrote:

postgres_fdw has some checks to enforce that non-superusers must connect to
the foreign server with a password-based method. The reason for this is to
prevent the authentication to the foreign server from happening on the basis
of the OS user who is running the non-foreign server.

But I think these super user checks should be run against the userid of the
USER MAPPING being used for the connection, not the userid of currently
logged on user.

So, if the user mapping user is a superuser locally, this would allow
any lambda user of the local server to attempt a connection to the
remote server. It looks dangerous rather dangerous to me to authorize
that, even if the current behavior is a bit inconsistent I agree.

I don't know what "any lambda user" means. Did you mean to write "any
random user"?

Yes, in this context that would be "any non-superuser" or "any user
without superuser rights". Actually that's a French-ism. I just
translated it naturally to English to define a user that has no access
to advanced features :)
--
Michael

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

On Tue, Oct 18, 2016 at 10:38 AM, Michael Paquier <michael.paquier@gmail.com

wrote:

On Mon, Oct 17, 2016 at 10:51 PM, Robert Haas <robertmhaas@gmail.com>
wrote:

On Mon, Oct 17, 2016 at 2:18 AM, Michael Paquier
<michael.paquier@gmail.com> wrote:

On Mon, Oct 17, 2016 at 3:33 AM, Jeff Janes <jeff.janes@gmail.com>

wrote:

postgres_fdw has some checks to enforce that non-superusers must

connect to

the foreign server with a password-based method. The reason for this

is to

prevent the authentication to the foreign server from happening on the

basis

of the OS user who is running the non-foreign server.

But I think these super user checks should be run against the userid

of the

USER MAPPING being used for the connection, not the userid of currently
logged on user.

So, if the user mapping user is a superuser locally, this would allow
any lambda user of the local server to attempt a connection to the
remote server. It looks dangerous rather dangerous to me to authorize
that, even if the current behavior is a bit inconsistent I agree.

I don't know what "any lambda user" means. Did you mean to write "any
random user"?

Yes, in this context that would be "any non-superuser" or "any user
without superuser rights". Actually that's a French-ism. I just
translated it naturally to English to define a user that has no access
to advanced features :)

Thanks for the patch, but it breaking the existing functionality as per the
other
mails. Marked as "returned with feedback" in 2016-11 commitfest.

Regards,
Hari Babu
Fujitsu Australia

On Thu, Dec 1, 2016 at 7:11 PM, Haribabu Kommi <kommi.haribabu@gmail.com>
wrote:

On Tue, Oct 18, 2016 at 10:38 AM, Michael Paquier <
michael.paquier@gmail.com> wrote:

On Mon, Oct 17, 2016 at 10:51 PM, Robert Haas <robertmhaas@gmail.com>
wrote:

On Mon, Oct 17, 2016 at 2:18 AM, Michael Paquier
<michael.paquier@gmail.com> wrote:

On Mon, Oct 17, 2016 at 3:33 AM, Jeff Janes <jeff.janes@gmail.com>

wrote:

postgres_fdw has some checks to enforce that non-superusers must

connect to

the foreign server with a password-based method. The reason for this

is to

prevent the authentication to the foreign server from happening on

the basis

of the OS user who is running the non-foreign server.

But I think these super user checks should be run against the userid

of the

USER MAPPING being used for the connection, not the userid of

currently

logged on user.

So, if the user mapping user is a superuser locally, this would allow
any lambda user of the local server to attempt a connection to the
remote server. It looks dangerous rather dangerous to me to authorize
that, even if the current behavior is a bit inconsistent I agree.

I don't know what "any lambda user" means. Did you mean to write "any
random user"?

Yes, in this context that would be "any non-superuser" or "any user
without superuser rights". Actually that's a French-ism. I just
translated it naturally to English to define a user that has no access
to advanced features :)

Thanks for the patch, but it breaking the existing functionality as per
the other
mails. Marked as "returned with feedback" in 2016-11 commitfest.

Here is an updated patch. This version allows you use the password-less
connection if you either are the super-user directly (which is the existing
committed behavior), or if you are using the super-user's mapping because
you are querying a super-user-owned view which you have been granted access
to.

It first I thought the currently committed behavior might be a security bug
as a directly logged in superuser can use another user's user-defined
mapping but without the password restriction when querying a view made by
someone else. But consulting with the security list nearly a year ago, the
conclusion was that it is never a good idea for a superuser to blindly
query from other users' views, and that the current situation is no worse
for postgres_fdw than it is for other features, and so nothing needs to be
done about it. So that is why I've decided to allow the passwordless
solution in either situation--a superuser using someone else mapping, or
someone else using a super user's mapping.

I didn't update any comments because the existing ones seem to apply
equally well to the new code as the old code.

The regression test passes with this version because I still allow the old
behavior. I didn't add a new test to also test the new behavior, because I
don't know how to do that with the existing make check framework, and a TAP
test seems like overkill.

Cheers,

Jeff

On 07/27/2017 09:45 PM, Jeff Janes wrote:> Here is an updated patch. 
This version allows you use the password-less

connection if you either are the super-user directly (which is the
existing committed behavior), or if you are using the super-user's
mapping because you are querying a super-user-owned view which you have
been granted access to.

I have tested the patch and it passes the tests and works, and the code
looks good (I have a small nitpick below).

The feature seems useful, especially for people who already use views
for security, so the question is if this is a potential footgun. I am
leaning towards no since the superuser should be careful when grant
access to is views anyway.

It would have been nice if there was a more generic way to handle this
since 1) the security issue is not unique to postgres_fdw and 2) this
requires you to create a view. But since the patch is simple, an
improvement in itself and does not prevent any future further
improvements in this era I see no reason to let perfect be the enemy of
good.

= Nitpicking/style

I would prefer if

/* no check required if superuser */
if (superuser())
return;

if (superuser_arg(user->userid))
return;

was, for consistency with the if clause in connect_pg_server(), written as

/* no check required if superuser */
if (superuser() || superuser_arg(user->userid))
return;

Andreas

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

On Tue, Sep 12, 2017 at 1:13 AM, Andreas Karlsson <andreas@proxel.se> wrote:

On 07/27/2017 09:45 PM, Jeff Janes wrote:> Here is an updated patch. This
version allows you use the password-less

connection if you either are the super-user directly (which is the
existing committed behavior), or if you are using the super-user's mapping
because you are querying a super-user-owned view which you have been
granted access to.

I have tested the patch and it passes the tests and works, and the code
looks good (I have a small nitpick below).

The feature seems useful, especially for people who already use views for
security, so the question is if this is a potential footgun. I am leaning
towards no since the superuser should be careful when grant access to is
views anyway.

It would have been nice if there was a more generic way to handle this
since 1) the security issue is not unique to postgres_fdw and 2) this
requires you to create a view. But since the patch is simple, an
improvement in itself and does not prevent any future further improvements
in this era I see no reason to let perfect be the enemy of good.

Thanks for the review.

I think that foreign tables ought to behave as views do, where they run as
the owner rather than the invoker. No one has talked me out of it, but no
one has supported me on it either. But I think it is too late to change
that now. Wrapping it in a view is not hard, but it sure clutters up a
schema. I don't think this can be made too generic, because each database
has a quite different security model, so the solution will be much
different.

Attached is a new patch which fixes the style issue you mentioned.

Cheers,

Jeff

On Thu, Sep 14, 2017 at 2:33 PM, Jeff Janes <jeff.janes@gmail.com> wrote:

I think that foreign tables ought to behave as views do, where they run as
the owner rather than the invoker. No one has talked me out of it, but no
one has supported me on it either. But I think it is too late to change
that now.

That's an interesting point. I think that you can imagine use cases
for either method. Obviously, if what you want to do is drill a hole
through the Internet to another server and then expose it to some of
your fellow users, having the FDW run with the owner's permissions
(and credentials) is exactly right. But there's another use case too,
which is where you have something that looks like a multi-user
sharding cluster. You want each person's own credentials to carry
over to everything they do remotely.

I feel like the USER MAPPING stuff is a pretty clunky and annoying way
of trying to make this work, no matter which of those use cases you
happen to have. But I'm not exactly sure what would be better,
either, and like you say, it's a bit late to be breaking compatibility
at this point.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

On 09/14/2017 08:33 PM, Jeff Janes wrote:> Attached is a new patch which
fixes the style issue you mentioned.

Thanks, the patch looks good no,w and as far as I can tell there was no
need to update the comments or the documentation so I am setting this as
ready for committer.

Andreas

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

On Thu, Sep 14, 2017 at 1:08 PM, Robert Haas <robertmhaas@gmail.com> wrote:

On Thu, Sep 14, 2017 at 2:33 PM, Jeff Janes <jeff.janes@gmail.com> wrote:

I think that foreign tables ought to behave as views do, where they run

as

the owner rather than the invoker. No one has talked me out of it, but

no

one has supported me on it either. But I think it is too late to change
that now.

That's an interesting point. I think that you can imagine use cases
for either method. Obviously, if what you want to do is drill a hole
through the Internet to another server and then expose it to some of
your fellow users, having the FDW run with the owner's permissions
(and credentials) is exactly right. But there's another use case too,
which is where you have something that looks like a multi-user
sharding cluster. You want each person's own credentials to carry
over to everything they do remotely.

OK. And if you want the first one, you can wrap it in a view currently,
but if it were changed I don't know what you would do if you want the 2nd
one (other than having every user create their own set of foreign tables).
So I guess the current situation is more flexible.

It does seem like it would then be a good idea to have a user mapping
option of "pass_the_hash" which would look up md5 hash from the pg_authid
(if the local username is the same as the remote user name) and use that to
connect to the foreign server, as an alternative option to recording the
password in plain text in the mapping itself. But that would require some
changes to libpq, not just postgres_fdw.

And that wouldn't work for SCRAM. I guess that SCRAM does have some
feature to allow this kind of delegation, but I don't know enough about it
to know how hard it would be to implement in postgres_fdw or how useful it
would be to have.

I feel like the USER MAPPING stuff is a pretty clunky and annoying way
of trying to make this work, no matter which of those use cases you
happen to have. But I'm not exactly sure what would be better,
either, and like you say, it's a bit late to be breaking compatibility
at this point.

Yeah, I have not been finding it enjoyable. How much flexibility does the
SQL/MED spec even give us (I don't have access to the spec)? From what I
could tell, it requires USER MAPPING to exist but doesn't give any details,
and doesn't say there can't be something else one could optionally use
instead.

Cheers,

Jeff

On Wed, Oct 4, 2017 at 6:13 PM, Jeff Janes <jeff.janes@gmail.com> wrote:

OK. And if you want the first one, you can wrap it in a view currently, but
if it were changed I don't know what you would do if you want the 2nd one
(other than having every user create their own set of foreign tables). So I
guess the current situation is more flexible.

So where does that leave this patch? I don't think it makes this
patch a bad idea, although I kind of lean towads the view that the
patch should just be checking superuser_arg(), not superuser() ||
superuser_arg().

It does seem like it would then be a good idea to have a user mapping option
of "pass_the_hash" which would look up md5 hash from the pg_authid (if the
local username is the same as the remote user name) and use that to connect
to the foreign server, as an alternative option to recording the password in
plain text in the mapping itself. But that would require some changes to
libpq, not just postgres_fdw.

And that wouldn't work for SCRAM. I guess that SCRAM does have some feature
to allow this kind of delegation, but I don't know enough about it to know
how hard it would be to implement in postgres_fdw or how useful it would be
to have.

We really need some kind of method for delegating authentication. I
don't know how it should work.

Generally, password authentication is a silly choice for automated
logins because then you've got to store the password someplace from
which it can be digitally stolen; stealing a password from someone's
brain is a different kind of problem. It's not even a good method for
this situation, yet it's the only one we allow. I think that bites,
but I don't really know what to do about it.

Yeah, I have not been finding it enjoyable. How much flexibility does the
SQL/MED spec even give us (I don't have access to the spec)?

Sorry, I don't know.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

On Thu, Oct 5, 2017 at 6:44 AM, Robert Haas <robertmhaas@gmail.com> wrote:

On Wed, Oct 4, 2017 at 6:13 PM, Jeff Janes <jeff.janes@gmail.com> wrote:

OK. And if you want the first one, you can wrap it in a view currently,

but

if it were changed I don't know what you would do if you want the 2nd one
(other than having every user create their own set of foreign tables).

So I

guess the current situation is more flexible.

So where does that leave this patch?

Sorry, I thought we were just having a digression. I didn't think that
part was about this patch specifically, but what more could be done later.

I don't think it makes this
patch a bad idea, although I kind of lean towads the view that the
patch should just be checking superuser_arg(), not superuser() ||
superuser_arg().

I don't see a reason to block a directly-logged-in superuser from using a
mapping. I asked in the closed list whether the current (released)
behavior was a security bug, and the answer was no. And I don't know why
else to block superusers from doing something other than as a security
bug. Also it would create a backwards compatibility hazard to revoke the
ability now.

Cheers,

Jeff

On Thu, Sep 14, 2017 at 04:08:08PM -0400, Robert Haas wrote:

On Thu, Sep 14, 2017 at 2:33 PM, Jeff Janes <jeff.janes@gmail.com> wrote:

I think that foreign tables ought to behave as views do, where they run as
the owner rather than the invoker. No one has talked me out of it, but no
one has supported me on it either. But I think it is too late to change
that now.

That's an interesting point. I think that you can imagine use cases
for either method. Obviously, if what you want to do is drill a hole
through the Internet to another server and then expose it to some of
your fellow users, having the FDW run with the owner's permissions
(and credentials) is exactly right. But there's another use case too,
which is where you have something that looks like a multi-user
sharding cluster. You want each person's own credentials to carry
over to everything they do remotely.

Hmmm, I don't think that's really right.

What I'd like instead is for the FDW client to tell the FDW server the
session_user/current_user on behalf of which it's acting, and let the
FDW server decide how to proceed. This could be done by doing a SET
SESSION fdw.client.session_user... and so on.

We use Kerberos principal names as PG user/role names, _with_ @REALM
included, so a user foo@BAR is likely to make sense to the FDW server.

Of course, if you're not using Kerberos then the local and remote user
namespaces might be completely distinct, but by letting the FDW server
know a) the FDW client's username (via authentication) and b) the true
username on the client side (via SET/set_config()), the server might
have enough information to decide whether it trusts (a) to impersonate
(b) and how to map (b) to a local user.

Nico
--

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

On 4 October 2017 at 18:13, Jeff Janes <jeff.janes@gmail.com> wrote:

On Thu, Sep 14, 2017 at 1:08 PM, Robert Haas <robertmhaas@gmail.com> wrote:

On Thu, Sep 14, 2017 at 2:33 PM, Jeff Janes <jeff.janes@gmail.com> wrote:

I think that foreign tables ought to behave as views do, where they run
as
the owner rather than the invoker. No one has talked me out of it, but
no
one has supported me on it either. But I think it is too late to change
that now.

That's an interesting point. I think that you can imagine use cases
for either method. Obviously, if what you want to do is drill a hole
through the Internet to another server and then expose it to some of
your fellow users, having the FDW run with the owner's permissions
(and credentials) is exactly right. But there's another use case too,
which is where you have something that looks like a multi-user
sharding cluster. You want each person's own credentials to carry
over to everything they do remotely.

OK. And if you want the first one, you can wrap it in a view currently, but
if it were changed I don't know what you would do if you want the 2nd one
(other than having every user create their own set of foreign tables). So I
guess the current situation is more flexible.

Sounds like it would be a useful option on a Foreign Server to allow
it to run queries as either the invoker or the owner. We have that
choice for functions, so we already have the concept and syntax
available. We could have another default at FDW level that specifies
what the default is for that type of FDW, and if that is not
specified, we keep it like it currently is.

--
Simon Riggs http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

On Thu, Oct 5, 2017 at 1:02 PM, Jeff Janes <jeff.janes@gmail.com> wrote:

I don't see a reason to block a directly-logged-in superuser from using a
mapping. I asked in the closed list whether the current (released)
behavior was a security bug, and the answer was no. And I don't know why
else to block superusers from doing something other than as a security bug.
Also it would create a backwards compatibility hazard to revoke the ability
now.

Well, my thought was that we ought to be consistent about whose
authorization matters. If we're using the view owner's credentials in
general, then we also (defensibly, anyway) ought to use the view
owner's superuser-ness to decide whether to enforce this restriction.
Using either the view owner's superuser-ness or the session user's
superuser-ness kind of puts you halfway in the middle. The view
owner's rights are what matters mostly, but your own rights also
matter a little bit around the edges. That's a little strange.

I don't have violently strong opinions about this - does anyone else
have a view?

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

Robert, all,

* Robert Haas (robertmhaas@gmail.com) wrote:

On Thu, Oct 5, 2017 at 1:02 PM, Jeff Janes <jeff.janes@gmail.com> wrote:

I don't see a reason to block a directly-logged-in superuser from using a
mapping. I asked in the closed list whether the current (released)
behavior was a security bug, and the answer was no. And I don't know why
else to block superusers from doing something other than as a security bug.
Also it would create a backwards compatibility hazard to revoke the ability
now.

Well, my thought was that we ought to be consistent about whose
authorization matters. If we're using the view owner's credentials in
general, then we also (defensibly, anyway) ought to use the view
owner's superuser-ness to decide whether to enforce this restriction.
Using either the view owner's superuser-ness or the session user's
superuser-ness kind of puts you halfway in the middle. The view
owner's rights are what matters mostly, but your own rights also
matter a little bit around the edges. That's a little strange.

I don't have violently strong opinions about this - does anyone else
have a view?

I haven't been following this closely, but I tend to agree with you- if
we're using the view owner's privileges then that's what everything
should be based on, not whether the caller is a superuser or not.

Consider a security-definer function. Clearly, such a function should
always run as the owner of the function, even if the caller is a
superuser. Running as the caller instead of the owner of the function
when the caller is a superuser because that would allow the function to
access more clearly isn't a good idea, imv.

Yes, that means that sometimes when superusers run things they get
permission denied errors. That's always been the case, and is correct.

Thanks!

Stephen

On Thu, Oct 5, 2017 at 1:16 PM, Nico Williams <nico@cryptonector.com> wrote:

That's an interesting point. I think that you can imagine use cases
for either method. Obviously, if what you want to do is drill a hole
through the Internet to another server and then expose it to some of
your fellow users, having the FDW run with the owner's permissions
(and credentials) is exactly right. But there's another use case too,
which is where you have something that looks like a multi-user
sharding cluster. You want each person's own credentials to carry
over to everything they do remotely.

Hmmm, I don't think that's really right.

What I'd like instead is for the FDW client to tell the FDW server the
session_user/current_user on behalf of which it's acting, and let the
FDW server decide how to proceed. This could be done by doing a SET
SESSION fdw.client.session_user... and so on.

Isn't that the same thing as the second use case I mentioned?

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company