[Dev] Gumd and security-manager integration

Jussi Laako jussi.laako at linux.intel.com
Mon Oct 13 09:19:35 GMT 2014


On 10.10.2014 20:03, Rafał Krypa wrote:
> Then the user removal should stop and return an error. If we are lucky, the user removal action will be retried and will work then. If not, the user would stay in half-deleted state.

We can do that, but I'd propose to at least run all the removal still, 
even despite of an error, because it would minimize amount of potential 
left-overs. We could then leave the UID in place, but disable logins for 
that user?

> I don't suggest doing rollbacks. IMHO partially removed user that cannot be used anymore is acceptable.

At least for user creation stage it would be nice to have possibility to 
rollback. But those would require having an LFS filesystem, which I 
think we should have in first place at least for IVI and such. It would 
also allow doing quick and complete factory-defaults reset.

> That's fine, it's one way to do that. Could you point me to the place where it is handled? What label are you giving to the home directory?

Imran can probably quicker point the place where it happens. It's "User" 
something. We added the code, because by default home directory was 
inheriting "System" label or something like that and it prevented user 
logins...

> Are you saying that /etc/skel handling in gumd prevents security xattrs? That would be perfect, we could simply set the proper labels there.

Intention is to copy things exactly over from there, including xattrs, 
only change naturally being UID/GID change for the items. So if it is 
possible to set correct attributes there, it would be really nice. Only 
case where I see that being not straightforward is case when you'd like 
to create some user-specific labels that no other user has. But these 
could be handled with the post-creation script.

> Does gumd daemon still need to be running for gum-utils to work? If yes, is your solution usable when image is created by mic?
> I'm only asking, it would be great to have gum-utils or libgum usable at image creation time. We wouldn't have to care about using bare useradd commands and fixing everything else it doesn't handle.

It needs to be, but now there's environment variable to switch it to use 
a regular socket for dbus communication instead of system bus daemon. So 
you could start gumd with "gumd &" and then run the gum-utils without 
need for other services like system-bus daemon. Once you are done, you 
could "killall gumd".

> But this approach is likely require some changes to gumd. At least cynara support andhandling error codes from scripts.

That shouldn't be an issue. We'll modify script handling and see into 
adding Cynara support.




More information about the Dev mailing list