On Sun, Dec 23, 2018 at 05:09:21PM +0100, Dominick Grift wrote: > On Sun, Dec 23, 2018 at 11:46:48AM +0100, Dominick Grift wrote: > > On Sun, Dec 23, 2018 at 11:20:00AM +0100, Dominick Grift wrote: > > > On Sat, Dec 22, 2018 at 02:58:41AM +0000, Sugar, David wrote: > > > > > > > > On 12/21/18 5:34 AM, Dominick Grift wrote: > > > > > On Fri, Dec 21, 2018 at 01:41:25AM +0000, David Sugar wrote: > > > > >> These are changes needed when pam_fallock created files in /run/faillock > > > > >> (which is labeled faillog_t). sudo and xdm (and probably other domains) > > > > >> will create files in this directory for successful and failed logins > > > > >> attempts. > > > > > The pam stuff has become a bit broken in my view. > > > > > > > > > > We use to use auth_use_pam() for these kinds of things but the interface was forgotten and not updated properly. > > > > > > > > > > So for example sudo does not even call auth_use_pam() and a lot of stuff was added directly to the login_pgm domain that should have been added to auth_use_pam() instead. > > > > > > > > > > My opinion is that this belongs in auth_use_pam() > > > > > > > > Dominick, > > > > > > > > I see those interfaces.? It looks like xdm_t already uses > > > > auth_login_pgm_domain(xdm_t).? It also isn't really clear to me what the > > > > difference is between auth_login_pgm_domain() and auth_use_pam().? I > > > > will make updates moving my change into auth_use_pam() and also update > > > > sudo_role_template() to use (I think) auth_login_pgm_domain (). > > > > > > sudo is not an auth_login_pgm_domain() i believe > > > > > > the auth_use_pam() is a subset of auth_login_pgm_domain() > > > > > > so login_pgm domains are pam clients plus extras needed to log in users > > > > > > a auth_use_pam() (pam client) has a pam stack but it might not actually do logins > > > > > > sudo uses pam but its not a real login program, so afaik sudo should call auth_use_pam() > > > xdm is a login_pgm, so is sshd etc > > > > > > systemd is also a pam client, but not a login program > > > > And yes systemd needs to be able to create these /run/faillock/USER files as well, but if you test this on RHEL then you wont see it because > > RHEL doesnt use /etc/pam.d/systemd-user (i suppose) > > > > so: > > > > 1. auth_use_pam() == "pam clients" (programs that have a file in /etc/pam.d), they use pam for authentication of some sort > > 2, auth_login_pgm_domain() == superset (special pam clients that need permissions to do actual logins) > > Another interesting detail is that pam_faillock clients need cap_dac_override to be able to write records to /run/faillock/USER files > I wonder whether that is a bug > > for example sshd (root) creates /run/faillock/joe with joe.root and 0600 but then sshd (root) needs cap_dac_override to write records to that file > Probably should have created the files with 0660 ... to avoid the need for cap_dac_override... I filed a bugzilla for this, just to be sure: https://bugzilla.redhat.com/show_bug.cgi?id=1661822 > > > > > > > > > > > > > > I will resubmit this patch, > > > > > > > > --- snip --- > > > > > > -- > > > Key fingerprint = 5F4D 3CDB D3F8 3652 FBD8 02D5 3B6C 5F1D 2C7B 6B02 > > > https://sks-keyservers.net/pks/lookup?op=get&search=0x3B6C5F1D2C7B6B02 > > > Dominick Grift > > > > > > > > -- > > Key fingerprint = 5F4D 3CDB D3F8 3652 FBD8 02D5 3B6C 5F1D 2C7B 6B02 > > https://sks-keyservers.net/pks/lookup?op=get&search=0x3B6C5F1D2C7B6B02 > > Dominick Grift > > > > -- > Key fingerprint = 5F4D 3CDB D3F8 3652 FBD8 02D5 3B6C 5F1D 2C7B 6B02 > https://sks-keyservers.net/pks/lookup?op=get&search=0x3B6C5F1D2C7B6B02 > Dominick Grift -- Key fingerprint = 5F4D 3CDB D3F8 3652 FBD8 02D5 3B6C 5F1D 2C7B 6B02 https://sks-keyservers.net/pks/lookup?op=get&search=0x3B6C5F1D2C7B6B02 Dominick Grift