Обсуждение: Fix for running from admin account on win32

Attached is a two part patch for the admin functionality on win32.

The first part is a simple bugfix to backend/port/win32/security.c. The
function that checks if the user has admin privileges didn't check it
the SID in the token was enabled or not. All actual access checks done
by the OS does check this, so we should too :-) This is required for the
second part of the patch to work, but also in some scenarios with
third-party tools that modify the token.


The second part enables pg_ctl to give up admin privleges when starting
the server. This works for both standalone and service, but only when
running on Windows 2000 or newer. The APIs simply didn't exist in NT4.
pg_ctl still works in NT4, but is unable to give up privileges. Since we
still do the privilege check in the postmaster, this is not a problem.
This has to be implemented in pg_ctl, because if it's done in-process
it's possi ble to get the admin privs back.

It also implements a job object wrapper around all processes created.
This only works when running as a service, because the job object is
destroyed when pg_ctl exits (it's automatically destroyed when the last
handle is closed). However, when running as a service it increases
security further by preventing new processes from being started with a
different user, access to clipboard, windows restarting and desktop
access. It also limits further any chance of accessing admin privileges,
more than we have today.

Finally, the job object provides an excellent point for monitoring the
server. It will contain aggregate statistics of how many processes are
running (or have been running), how much CPU is being used (has been
used), memory activity etc. As a whole for postmaster+all children, not
one a piece. This functionality is all provided by default by the
windows performance monitor when you use job objects.


It turned out the mingw headers *and* libraries were incomplete wrt
these functions, so I had to do it with runtime loading of DLLs. Since I
had to do this anyway, it was trivial to do this for all the NT4
functions, and just have it work there. So the discussion I started
yesterday about NT4 compatibility doesn't really apply to this case -
but it's still a good discussionto have I think.


//Magnus


D:\msys\1.0\local\pgsql\bin>postmaster -D ..\data
Execution of PostgreSQL by a user with administrative permissions is not
permitted.
The server must be started under an unprivileged user ID to prevent
possible system security compromises.  See the documentation for
more information on how to properly start the server.

D:\msys\1.0\local\pgsql\bin>pg_ctl -D ..\data start
postmaster starting

D:\msys\1.0\local\pgsql\bin>LOG:  database system was shut down at
2006-01-14 17
:42:14 W. Europe Standard Time
LOG:  checkpoint record is at 0/39FD88
LOG:  redo record is at 0/39FD88; undo record is at 0/0; shutdown TRUE
LOG:  next transaction ID: 582; next OID: 16389
LOG:  next MultiXactId: 1; next MultiXactOffset: 0
LOG:  database system is ready
LOG:  transaction ID wrap limit is 2147484146, limited by database
"postgres"

Naturally, just a minute after sending the patch, I realised how it can
be done better ;)

If the job object code is moved to the postmaster, it'll work when not
running as a service as well. And there's no way to break out of the job
object -  it's just the other part.

Yes, I'll set up a new patch :-) Meanwhlie, if there are any other
comments on this one, I'm still interested in those.

//Magnus


> -----Original Message-----
> From: pgsql-patches-owner@postgresql.org
> [mailto:pgsql-patches-owner@postgresql.org] On Behalf Of
> Magnus Hagander
> Sent: Saturday, January 14, 2006 6:09 PM
> To: pgsql-patches@postgresql.org
> Subject: [PATCHES] Fix for running from admin account on win32
>
> Attached is a two part patch for the admin functionality on win32.
>
> The first part is a simple bugfix to
> backend/port/win32/security.c. The function that checks if
> the user has admin privileges didn't check it the SID in the
> token was enabled or not. All actual access checks done by
> the OS does check this, so we should too :-) This is required
> for the second part of the patch to work, but also in some
> scenarios with third-party tools that modify the token.
>
>
> The second part enables pg_ctl to give up admin privleges
> when starting the server. This works for both standalone and
> service, but only when running on Windows 2000 or newer. The
> APIs simply didn't exist in NT4.
> pg_ctl still works in NT4, but is unable to give up
> privileges. Since we still do the privilege check in the
> postmaster, this is not a problem.
> This has to be implemented in pg_ctl, because if it's done
> in-process it's possi ble to get the admin privs back.
>
> It also implements a job object wrapper around all processes created.
> This only works when running as a service, because the job
> object is destroyed when pg_ctl exits (it's automatically
> destroyed when the last handle is closed). However, when
> running as a service it increases security further by
> preventing new processes from being started with a different
> user, access to clipboard, windows restarting and desktop
> access. It also limits further any chance of accessing admin
> privileges, more than we have today.
>
> Finally, the job object provides an excellent point for
> monitoring the server. It will contain aggregate statistics
> of how many processes are running (or have been running), how
> much CPU is being used (has been used), memory activity etc.
> As a whole for postmaster+all children, not one a piece. This
> functionality is all provided by default by the windows
> performance monitor when you use job objects.
>
>
> It turned out the mingw headers *and* libraries were
> incomplete wrt these functions, so I had to do it with
> runtime loading of DLLs. Since I had to do this anyway, it
> was trivial to do this for all the NT4 functions, and just
> have it work there. So the discussion I started yesterday
> about NT4 compatibility doesn't really apply to this case -
> but it's still a good discussionto have I think.
>
>
> //Magnus
>
>
> D:\msys\1.0\local\pgsql\bin>postmaster -D ..\data Execution
> of PostgreSQL by a user with administrative permissions is
> not permitted.
> The server must be started under an unprivileged user ID to
> prevent possible system security compromises.  See the
> documentation for more information on how to properly start
> the server.
>
> D:\msys\1.0\local\pgsql\bin>pg_ctl -D ..\data start
> postmaster starting
>
> D:\msys\1.0\local\pgsql\bin>LOG:  database system was shut down at
> 2006-01-14 17
> :42:14 W. Europe Standard Time
> LOG:  checkpoint record is at 0/39FD88
> LOG:  redo record is at 0/39FD88; undo record is at 0/0; shutdown TRUE
> LOG:  next transaction ID: 582; next OID: 16389
> LOG:  next MultiXactId: 1; next MultiXactOffset: 0
> LOG:  database system is ready
> LOG:  transaction ID wrap limit is 2147484146, limited by
> database "postgres"
>

"Magnus Hagander" <mha@sollentuna.net> writes:
> If the job object code is moved to the postmaster, it'll work when not
> running as a service as well.

I'd just as soon keep all that Windows-specific cruft in pg_ctl.
So I think the way you've got it set up is fine.

            regards, tom lane

"Magnus Hagander" <mha@sollentuna.net> writes:
> Attached is a two part patch for the admin functionality on win32.

> The first part is a simple bugfix to backend/port/win32/security.c. The
> function that checks if the user has admin privileges didn't check it
> the SID in the token was enabled or not.

Applied back to 8.0.

> The second part enables pg_ctl to give up admin privleges when starting
> the server.

Applied to HEAD only.

            regards, tom lane