On 08/20/2013 04:05 PM, Dan Pou wrote: > On Fri, Aug 09, 2013 at 08:51:17AM -0400, Stephen Smalley wrote: >> On 08/08/2013 03:58 PM, Dan Pou wrote: >>> Thanks for the suggestion about strace, that is pointing to the problem. >>> I need to check the policy rules I have been adding to see how I got >>> here: >>> >>> write(3, "user_u:sysadm_r:sysadm_t:s0\0", 28) = -1 EINVAL (Invalid >>> argument) >>> >>> This is the second write. When I test the same code with sshd_t to >>> <username> transition, I get one write, with a successful return of >>> default_context_with_level. >>> >>> Thanks, this gives me something to go off of. >> >> Could be that security_compute_user() (which writes to >> /sys/fs/selinux/user) is getting an error or no results and thus the >> code is trying to fall back to the failsafe context (as specified in >> /etc/selinux/$SELINUXTYPE/contexts/failsafe_context), which isn't legal >> for user_u. >> The failsafe is to permit root / admin logins under such a situation. >> >> I'd look to see what the result of writing to /sys/fs/selinux/user was >> and what was written to it in the strace output. You can also directly >> call security_compute_user via the compute_user tool under >> libselinux/utils if you obtain and build the sources yourself. That >> utility doesn't appear to get included in the libselinux-utils rpm though. >> >> > > I addeded the system_r:my_daemon_t:s0 user_r:user_t:s0 role transition > to /etc/selinux/mls/contexts/default_contexts. > This got me to actually writing user_u:user_r:user_t:s0 for setexeccon, > but I am still failing. It looks like it is failing in the > selinux_trans_to_raw_context. I was thinking this was an issue with > declaring the transition. > What steps do I need to setup a role_transition and/or type_transistion? > > I tried adding the following to no avail: > type_transition my_daemon_t non_security_file_type:process user_t; > Do I need more type_transitions, or addition role_transition > declarations (aside from /etc/selinux/mls/contexts/default_context)? > > Other info: > The daemon (without SELinux) just sets the euid/egid to match the user > requesting the job. So I inserted calls to > getseuserbyname > get_default_context_with_rolelevel or get_default_context_with_level > then setexeccon > the returned context before execve'ing. If you are getting to the point of setexeccon() with the correct security context, then you are past the point of any type transition or role transition rules. The transition rules are just to define defaults in the absence of an explicit setexeccon(); it is the allow rules that govern what is permitted. You do need both role allow and TE allow rules in your policy, of course. Did you add the interface calls that I listed earlier for your daemon domain? However, lack of permission there should show up as an AVC denial; you might re-test with dontaudit rules stripped to confirm (semodule -DB). If the error truly lies during the trans_to_raw_context, then maybe you have a problem in your label translation daemon? Are you running mcstransd or something else? What happens if you simply stop the label daemon (it isn't required for operation; it just provides translation of MLS labels to and from more human-readable form and deals with complex label encodings ala the old MITRE label encodings format). -- This message was distributed to subscribers of the selinux mailing list. If you no longer wish to subscribe, send mail to majordomo@xxxxxxxxxxxxx with the words "unsubscribe selinux" without quotes as the message.
