[PATCH] The directives such as user, rlimit_core should also be effective on master process

[Igor Sysoev]

On 05 Aug 2014, at 21:10, Quanah Gibson-Mount <quanah at zimbra.com> wrote:

> --On Tuesday, August 05, 2014 12:58 PM +0400 Maxim Dounin <mdounin at mdounin.ru> wrote:
> 
>> Hello!
>> 
>> On Tue, Aug 05, 2014 at 02:26:59AM -0500, Kunal Pariani wrote:
>> 
>>> # HG changeset patch
>>> # User Kunal Pariani <kpariani at zimbra.com>
>>> # Date 1407194790 25200
>>> #      Mon Aug 04 16:26:30 2014 -0700
>>> # Node ID f25ab24517bb5e45b1b7fa1a1502b907f2cff213
>>> # Parent  f8764e20fcd7f87d98fe97f82b2a8d0a77ed9097
>>> The directives such as user, rlimit_core should also be effective on
>>> master process
>> 
>> No, thanks.  The master process should remain root to be able to
>> read priveleged configuration files during configuration reload,
>> open listening sockets on priveleged ports and so on.
> 
> The drop to the user happens *after* the files & ports have been opened already.  This is how many *nix server processes work, as an additional security measure.  We've been using nginx with this patch since nginx 0.5.37, so it's quite heavily tested.
> 
> See also: <http://www.dwheeler.com/secure-class/Secure-Programs-HOWTO/minimize-privileges.html>
> and
> <http://www.ibm.com/developerworks/linux/library/l-sppriv/index.html>

nginx master process does not process incoming connections. It just keep listen ports open.
If master process will drop root privileges it will be unable to open new listen ports,
files, etc. after graceful reconfiguration on the HUP signal.


-- 
Igor Sysoev
http://nginx.com