Bug 1275153

+++ This bug was initially created as a clone of Bug #1274345 +++

Description of problem:

Trying to make "systemd --user" + pam_selinux working correctly for confined users. If I have all policy changes then I am not able to get a correct context for "systemd --user" 

staff_u:unconfined_r:unconfined_t:s0 mgrepl 1541   1  0 09:00 ?        00:00:00 /usr/lib/systemd/systemd --user

instead of staff_t. I see

audit[1024]: USER_ROLE_CHANGE pid=1024 uid=0 auid=4294967295 ses=4294967295 subj=system_u:system_r:init_t:s0 msg='pam: default-context=unconfined_u:unconfined_r:unconfined_t:s0 selected-context=unconfined_u:unconfined_r:unconfined_t:s0 exe="/usr/lib/systemd/systemd" hostname=? addr=? terminal=? res=success'

The problem is with enabled uncofined.pp module.


Version-Release number of selected component (if applicable):

$ rpm -q libselinux
libselinux-2.4-4.fc24.x86_64

How reproducible:

1. With enabled unconfined.pp module

selinux.security_compute_user("system_u:system_r:init_t:s0","staff_u")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
OSError: [Errno 34] Numerical result out of range

which causes you continue with a context defined in "/etc/selinux/targeted/contexts/failsafe_context".


2. With disabled unconfined.pp module

selinux.security_compute_user("system_u:system_r:init_t:s0","staff_u")

returns a correct list and 

selinux.get_default_context("staff_u","system_u:system_r:init_t:s0")
[0, 'staff_u:staff_r:staff_t:s0']

--- Additional comment from Petr Lautrbach on 2015-10-22 11:11:30 EDT ---

It could be a problem in kernel which hit some kind of limit?

# echo -n system_u:system_r:init_t:s0 staff_u > /sys/fs/selinux/user
echo: write error: numerical result out of range

--- Additional comment from Petr Lautrbach on 2015-10-22 11:21:38 EDT ---

So it seems to the code in kernel security/selinux/selinuxfs.c which hits the SIMPLE_TRANSACTION_LIMIT:

security/selinux/selinuxfs.c:
 916 static ssize_t sel_write_user(struct file *file, char *buf, size_t size)
 ...
 953         for (i = 0; i < nsids; i++) {
 954                 rc = security_sid_to_context(sids[i], &newcon, &len);
 955                 if (rc) {
 956                         length = rc;
 957                         goto out;
 958                 }
 959                 if ((length + len) >= SIMPLE_TRANSACTION_LIMIT) {
 960                         kfree(newcon);
 961                         length = -ERANGE;
 962                         goto out;
 963                 }
 964                 memcpy(ptr, newcon, len);
 965                 kfree(newcon);
 966                 ptr += len;
 967                 length += len;
 968         }

I'm not sure if it's expected, or the limit could be raised or policy changed?

--- Additional comment from Paul Moore on 2015-10-22 15:41:37 EDT ---

The SIMPLE_TRANSACTION_LIMIT is due to our use of the simple_transaction_* set of APIs in the kernel, removing the current limit would require a good deal of work in the kernel.  Without knowing too much about the problem you're trying to solve, is there some workaround in userspace?  After all, the limit should be almost a full page size ...

--- Additional comment from Miroslav Grepl on 2015-10-23 02:56:24 EDT ---

The problem is we are not able to get confined users working with enabled unconfined.pp module. We have pam_selinux.so in /etc/pam.d/systemd-user to have it running in a user domain instead of init_t.

But security_compute_user() is not able to get all possible contexts with enabled unconfined.pp module because we have a lot of transition rules to staff_t. 

It is caused by

optional_policy(`
 unconfined_domain(init_t)
')

So we would need to run with confined init_t by default to avoid it. But I can not imagine it now. But we could probably review these transitions rules for unconfined domains.

--- Additional comment from Miroslav Grepl on 2015-10-23 07:33:04 EDT ---

I am playing with policy changes to make it working which I will need to discuss with upstream.

--- Additional comment from Paul Moore on 2015-10-23 11:53:38 EDT ---

Okay.  If there is a workaround in policy it will be much quicker than trying to solve this in the kernel, although we could still keep this BZ open to address the kernel limit at a later date.