Bug 30181 - Rework local authority configuration
Summary: Rework local authority configuration
Status: RESOLVED MOVED
Alias: None
Product: PolicyKit
Classification: Unclassified
Component: daemon (show other bugs)
Version: unspecified
Hardware: Other All
: medium normal
Assignee: David Zeuthen (not reading bugmail)
QA Contact: David Zeuthen (not reading bugmail)
URL:
Whiteboard:
Keywords:
Depends on:
Blocks:
 
Reported: 2010-09-14 06:42 UTC by David Zeuthen (not reading bugmail)
Modified: 2018-08-20 21:34 UTC (History)
0 users

See Also:
i915 platform:
i915 features:


Attachments

Description David Zeuthen (not reading bugmail) 2010-09-14 06:42:05 UTC
Consider the following .pkla files

/etc/polkit-1/localauthority/50-local.d/10-org.acme.pkla:
 [Allow group staff to do org.acme.frobnicate]
 Identity=unix-group:staff
 Action=org.acme.frobnicate
 ResultAny=yes
 ResultInactive=yes
 ResultActive=yes

and

/etc/polkit-1/localauthority/10-vendor.d/10-org.vendor.lockdown.pkla:
 [Lock org.acme.frobnicate down]
 Identity=unix-user:*
 Action=org.acme.frobnicate
 ResultAny=auth_admin
 ResultInactive=auth_admin
 ResultActive=auth_admin

where the former is provided by the 3rd party that provides the org.acme.frobnicate action (e.g. the software package) and the latter is provided by the OS vendor. 

Because of the way .pkla files work, e.g.

 1. First we run through .pkla files in order for Group identities
    that the Subject in question belongs to

 2. Then we run through .pkla files in order for the User identity
    for the Subject in question

the latter always "win" even though the former has a higher priority. This is problematic because it's a semi-reasonable thing for the OS vendor to want to do.

One solution would be to allow "Identity=*" to mean any and then declare the order to be

 0. First we run through .pkla files in order and apply any section
    where Identity=*

 1. First we run through .pkla files in order for Group identities
    that the Subject in question belongs to

 2. Then we run through .pkla files in order for the User identity
    for the Subject in question

This way OS vendors can "lock down" everything without interfering with other .pkla files.

Another option is to completely rethink how the Local Authority is configured and do something a lot simpler. This might not be too late as we haven't hit 1.0 yet.

See https://bugzilla.novell.com/show_bug.cgi?id=544579 for a real-world example of where an OS vendor runs into this problem.
Comment 1 David Zeuthen (not reading bugmail) 2010-09-14 07:14:51 UTC
Crazy idea: Maybe a better setup than .pkla files would be to do things exactly like udev does - with rules

 ACTION=="org.libvirt.manage", USER=="davidz", RESULT="auth_admin"

 # /path/to/script can print RESULT=auth_admin|yes|no|... on stdout
 ACTION=="org.libvirt.*", RUN="/path/to/script"

 # allow mounting filesystems if in a local active session
 ACTION=="org.udisks.mount", ACTIVE=="true", LOCAL=="true", RESULT="yes"

So we'd have and

 /etc/polkit-1/localauthority/rules.d/
 /var/polkit-1/localauthority/rules.d/

 (same way udev has /lib/udev/rules.d and /etc/udev/rules.d - we'd also
  include nice and useful semantics such as

    Rule files are required to have a unique name, duplicate file names are
    ignored. Files in /etc/udev/rules.d/ have precedence over files with
    the same name in /lib/udev/rules.d/. This can be used to ignore a
    default rules file if needed.

  which is from the udev(8) man page.)

and, more importantly, this allows running scripts/programs to determine if the given Subject is authorized or not. Which basically makes the PolicyKit Local Authority backend 100% scriptable. Which I think is something most admins want.
Comment 2 David Zeuthen (not reading bugmail) 2011-02-23 13:06:15 UTC
Repurposing this bug for reworking local authority configuration
Comment 3 David Zeuthen (not reading bugmail) 2012-05-17 21:25:26 UTC
I created a new branch for an experiment, see

 http://cgit.freedesktop.org/polkit/log/?h=wip/js-rule-files

that embeds a JS interpreter (running inside polkitd). The idea is that admins can drop .rules files in /etc/polkit-1/rules.d and these rules are JS scripts. This allows maximum flexibility while still allowing relatively simple files. A lot of information is available in the passed @subject value, for example pid, user-name, groups, session and seat. See below

With the rules file in [1], I get

 00:12:40.780: /etc/polkit-1/rules.d/10-example.rules:8:
action=org.freedesktop.policykit.exec
 00:12:40.780: /etc/polkit-1/rules.d/10-example.rules:9:
subject=[Subject pid=7449 seat=seat0 session=1 local=true active=true
user=davidz groups=davidz,wheel]
 00:12:40.780: /etc/polkit-1/rules.d/10-example.rules:11: now=Fri May
18 2012 00:12:40 GMT-0400 (EDT)

and if from another seat I get

 00:11:29.680: /etc/polkit-1/rules.d/10-example.rules:8:
action=org.freedesktop.policykit.exec
 00:11:29.680: /etc/polkit-1/rules.d/10-example.rules:9:
subject=[Subject pid=30980 seat= session=8 local=false active=true
user=bateman groups=bateman]
 00:11:29.680: /etc/polkit-1/rules.d/10-example.rules:11: now=Fri May
18 2012 00:11:29 GMT-0400 (EDT)

The only real missing feature is a polkit.spawn() method to run
arbitrary helpers (which is of course very expensive so should be used sparingly).

[1] : /etc/polkit-1/rules.d/10-example.rules

/* -*- mode: js; js-indent-level: 4; indent-tabs-mode: nil -*- */

polkit.addAdministratorRule(function(action, subject) {
   return ["unix-group:sys", "unix-user:root"];
});

polkit.addAuthorizationRule(function(action, subject) {
   polkit.log("action=" + action);
   polkit.log("subject=" + subject);
   var now = new Date();
   polkit.log("now=" + now);
   if (action == "org.freedesktop.policykit.exec" &&
       subject.isInGroup("staff")) {
       return "yes";
   }
   return null;
});
Comment 4 GitLab Migration User 2018-08-20 21:34:32 UTC
-- GitLab Migration Automatic Message --

This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity.

You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/polkit/polkit/issues/12.


Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.