On 27.2.2020 14.20, Dominick Grift wrote:
Russell Coker <russell@xxxxxxxxxxxx> writes:
allow systemd_logind_t init_var_run_t:chr_file write;
audit2allow shows me that the above is attempted on Debian/Unstable. What's
this inaccessible directory about anyway?
systemd-userruntimedir (245) now also creates it in /run/user/%{USERID}
The relevant code has this comment:
/* Set up inaccessible nodes now so they're available if we decide to
use them with user namespaces. */
probably used for InaccessiblePath= directive but I am not sure.
Yes, these are bind mounted over the path which is wanted inaccessible.
Perhaps this could be improved by giving them a dedicated label and then
some new TE rules could prevent anything other than PID1 from managing
them. Now if a service has CAP_SYS_ADMIN and is not blocked by seccomp
filters from using mount and umount system calls, it could dismantle the
bind mount.
-Topi
# ls -lZ /run/systemd/inaccessible
total 0
b---------. 1 root root system_u:object_r:init_var_run_t:s0 0, 0 Feb 27 13:36
blk
c---------. 1 root root system_u:object_r:init_var_run_t:s0 0, 0 Feb 27 13:36
chr
d---------. 2 root root system_u:object_r:init_var_run_t:s0 40 Feb 27 13:36
dir
p---------. 1 root root system_u:object_r:init_var_run_t:s0 0 Feb 27 13:36
fifo
----------. 1 root root system_u:object_r:init_var_run_t:s0 0 Feb 27 13:36
reg
s---------. 1 root root system_u:object_r:init_var_run_t:s0 0 Feb 27 13:36
sock
