| 1 |
On 04/03/2010 10:11 AM, Michał Górny wrote: |
| 2 |
> Hello, |
| 3 |
> |
| 4 |
> I am using umask 027 on my Gentoo boxes, and setgid bit set on a few |
| 5 |
> directories crucial to userpriv-enabled merges. This way, I do not have |
| 6 |
> to worry about running e.g. layman through 'sg' or similar tools, as |
| 7 |
> all newly-created files inherit portage group ownership, and |
| 8 |
> newly-created directories inherit the setgid bit. |
| 9 |
> |
| 10 |
> I would like to be able to use similar solution for compiled kernel |
| 11 |
> sources, i.e. through setting the setgid bit on /usr/src. But in fact |
| 12 |
> it is impossible as portage forces setting it's own permissions on all |
| 13 |
> installed files, thus newly-installed kernel sources do not inherit the |
| 14 |
> parent group ownership nor the setgid bit. |
| 15 |
> |
| 16 |
> Now the question is: should such behaviour be considered really correct |
| 17 |
> and necessary? In my opinion, if user sets setuid/setgid on a parent |
| 18 |
> directory, shklee knows what shklee is doing and emerge should not |
| 19 |
> override this system-specific ownership inheritance. |
| 20 |
> |
| 21 |
|
| 22 |
Your issue seems somewhat related to this bug: |
| 23 |
|
| 24 |
http://bugs.gentoo.org/show_bug.cgi?id=141619 |
| 25 |
|
| 26 |
My first inclination is to use configuration file for stuff like |
| 27 |
this, since it's not really possible to distinguish ad hoc |
| 28 |
permission modifications done by the user from incorrect permissions |
| 29 |
that are due to other reasons such as faulty ebuilds. It would |
| 30 |
probably also be a good idea to record file permissions in |
| 31 |
/var/db/pkg/*/*/CONTENTS, so that we'd have some way know when |
| 32 |
permissions differ from those initially set by the ebuild, and a way |
| 33 |
to detect collisions in directory permissions between 2 different |
| 34 |
ebuilds that install files in the same directory. |
| 35 |
-- |
| 36 |
Thanks, |
| 37 |
Zac |
Archives