Else blocks with optional statements are not supported in CIL.
Currently, if the pp to CIL compiler comes across one of these in a pp
module, it just drops the block and outputs a warning. Fortunately,
these are very rare. In fact, this is the only place in refpolicy where
an optional else block is used, and it is not clear if it is even
needed. This patch is untested, and is more to spark discussions to see
if there are any thoughts about whether or not this piece of policy is
needed.
Signed-off-by: Steve Lawrence <slawrence@tresys.com>
With the new userspace, the only files in /var/lib/selinux are selinux
store related files, so label it and everything inside it as
semanage_store_t. semanage_var_lib_t is completely removed and now
aliases semanage_store_t for backwards compatibility. This differs from
the v2 patch in that it adds back the ability to manage
selinux_config_t, which is necessary to manage the old module store for
things like migrating from the old to new store and backwards
compatability.
Signed-off-by: Steve Lawrence <slawrence@tresys.com>
Allow sudo (1.8.9_p5 and higher) to handle /var/run/sudo/ts if it does
not exist (given the tmpfs nature of /var/run). This is done when sudo
is run in the user prefixed domain, and requires both the chown
capability as well as the proper file transition when /var/run/sudo is
created.
In order to write the grub configuration and perform the preliminary
checks, the grub-mkconfig command should run in the bootloader_t domain.
As such, update the file context definition to be bootloader_exec_t.
create_netlink_socket_perms is defined as:
{ create_socket_perms nlmsg_read nlmsg_write }
This means that it is redundant to allow create_socket_perms and
nlmsg_read/nlmsg_write.
Clean up things without allowing anything new.
iw manpage says "iw - show / manipulate wireless devices and their
configuration". Label this command ifconfig_exec_t to allow it to
manage wireless communication devices.
Debian installs iw in /sbin/iw, Fedora in /usr/sbin/iw and Arch Linux in
/usr/bin/iw (with /usr/sbin being a symlink to /usr/bin).
-This patch is needed since CIL does not allow duplicate
role declarations. The roles for system_r, staff_r, sysadm_r, and
user_r were already declared in kernel.te. Since the roles are
pulled in from require statements in the appropriate interfaces,
the duplicate role declarations could be deleted in modules for
auditadm, staff, sysadm, and userdomain.
-Move a role declaration that used an argument passed into the
userdom_base_user_template into a gen_require statement.
Since commit 0fd9dc55, logging.te contains:
term_write_all_user_ttys(syslogd_t)
As "write" is a superset of "append", this rule is no longer needed:
term_append_unallocated_ttys(syslogd_t)
While at it, add a comment which explains why
term_dontaudit_setattr_unallocated_ttys is needed.
audit.log shows that journald needs to read the kernel read buffer:
avc: denied { syslog_read } for pid=147 comm="systemd-journal" scontext=system_u:system_r:syslogd_t tcontext=system_u:system_r:kernel_t tclass=system permissive=1
Moreover journald uses RW access to /dev/kmsg, according to its code:
http://cgit.freedesktop.org/systemd/systemd/tree/src/journal/journald-kmsg.c?id=v215#n394
On ArchLinux the directory name of Network Manager in /usr/lib is
written in lowercase but not the files in /usr/bin, /var/lib, etc.
While at it, remove a useless backslash before a minus character.
Currently, all postgresql commands in are labeled as postgresql_exec_t.
This means they can only be executed by db admins. However, the "normal"
commands, such as createdb or psql, should also be executable by users.
(The users in question still need to be granted postgresql_role(), so
this is no security problem.)
If the (sadly nearly undocumented) Linux kernel feature which allows
specific user groups to send ICMP echos without CAP_NET_RAW
(configurable with the sysctl net.ipv4.ping_group_range, available since
3.0) is used, ping needs the getattr permission of the rawip_socket
class in order to work.
The kernel_delete_unlabeled_chr_files interface is called by the
(deprecated) files_delete_isid_type_chr_files interface in
kernel/files.if.
Signed-off-by: Sven Vermeulen <sven.vermeulen@siphos.be>
The kernel_delete_unlabeled_blk_files interface is called by the
(deprecated) files_delete_isid_type_blk_files in kernel/files.if.
Signed-off-by: Sven Vermeulen <sven.vermeulen@siphos.be>
The kernel_delete_unlabeled_sockets interface is called by the
(deprecated) files_delete_isid_type_sock_files interface in
kernel/files.if.
Signed-off-by: Sven Vermeulen <sven.vermeulen@siphos.be>
The kernel_delete_unlabeled_pipes interface is called by the
(deprecated) files_delete_isid_type_fifo_files interface in
kernel/files.if.
Signed-off-by: Sven Vermeulen <sven.vermeulen@siphos.be>
The kernel_delete_unlabeled_symlinks interface is called by the
files_delete_isid_type_symlinks interface (in kernel/files.if). This
interface is deprecated (and calls kernel_delete_unlabeled_symlinks).
Signed-off-by: Sven Vermeulen <sven.vermeulen@siphos.be>
These are seldomly-used tools from the util-linux package.
Please check if they are located in /sbin instead of /usr/sbin on other
distributions.
Signed-off-by: Luis Ressel <aranea@aixah.de>
With init_daemon_pid_file supporting class parameters, all calls to
init_daemon_run_dir can now be transformed into init_daemon_pid_file
calls.
Update the init_daemon_run_dir interface so it gives a warning when
used, and use the init_daemon_pid_file interface underlyingly.
Signed-off-by: Sven Vermeulen <sven.vermeulen@siphos.be>
For some daemons, it is the init script that is responsible for creating
the PID file of the daemon. As we do not want to update the init SELinux
policy module for each of these situations, we need to introduce an
interface that can be called by the SELinux policy module of the caller
(the daemon domain).
The initial suggestion was to transform the init_daemon_run_dir
interface, which offers a similar approach for directories in /run, into
a class-agnostic interface. Several names have been suggested, such as
init_script_spec_run_content or init_script_generic_run_filetrans_spec,
but in the end init_daemon_pid_file was used.
The interface requires the class(es) on which the file transition should
occur, like so:
init_daemon_pid_file(xdm_var_run_t, dir, "xdm")
init_daemon_pid_file(postgresql_var_run_t, file, "postgresql.pid")
Signed-off-by: Sven Vermeulen <sven.vermeulen@siphos.be>
When an unconfined_t root user runs dmesg, the kernel complains with
this message in its logs (when SELinux is in enforcing mode):
dmesg (16289): Attempt to access syslog with CAP_SYS_ADMIN but no
CAP_SYSLOG (deprecated).
audit.log contains following AVC:
avc: denied { syslog } for pid=16289 comm="dmesg" capability=34
scontext=unconfined_u:unconfined_r:unconfined_t
tcontext=unconfined_u:unconfined_r:unconfined_t tclass=capability2
The dropbox application has a feature called "LAN Sync" which works on
TCP & UDP port 17500. Marking this port as dropbox_port_t (instead of
the currently default unreserved_port_t) allows for more fine-grained
access control to this resource.
Signed-off-by: Sven Vermeulen <sven.vermeulen@siphos.be>
Several DMs offer the possibility to shutdown the system. I personally
don't think a bool is neccessary for this permission, but I wouldn't
oppose one either.
Currently, the /usr/share/cvs/contrib/rcs2log is only labeled as bin_t
for redhat distributions. Moving this to the general one as it is also
in use on other distributions
Signed-off-by: Sven Vermeulen <sven.vermeulen@siphos.be>
The LightDM application stores its xauth file in a subdirectory
(/var/run/lightdm/root) which is labeled as xdm_var_run_t. As a result,
X11 (xserver_t) needs search rights to this location.
With this setup, X is run as follows:
/usr/bin/X :0 -auth /var/run/lightdm/root/:0
Changes since v1:
- Use read_files_pattern instead of separate allow rules
Signed-off-by: Jason Zaman <jason@perfinion.com>
Signed-off-by: Sven Vermeulen <sven.vermeulen@siphos.be>
Since commit 2d0c9cec mls_file_read_up and mls_file_write_down
interfaces are deprecated even though they are still present.
Replace mls_file_read_up with mls_file_read_all_levels and
mls_file_write_down with mls_file_write_all_levels.
When an unconfined user uses truecrypt to mount an encrypted file, dmsetup is
called to setup a new device. This program works with udev to configure the
new device and uses SysV semaphores to synchronize states. As udev runs
dmsetup in lvm_t domain, the first dmsetup process needs to create lvm_t
semaphores (not unconfined_t) and hence needs to run in lvm_t domain.
More details are available in the archives on the ML:
http://oss.tresys.com/pipermail/refpolicy/2014-May/007111.html
VBoxCreateUSBNode.sh creates character special files in /dev/vboxusb each time
a new USB device appears. This script is called by udev.
audit.log on a system in permissive mode before this patch contains:
type=AVC msg=audit(1396889711.890:175): avc: denied { execute } for pid=26284 comm="systemd-udevd" name="VBoxCreateUSBNode.sh" dev="sda5" ino=5899405 scontext=system_u:system_r:udev_t tcontext=unconfined_u:object_r:usr_t tclass=file
type=AVC msg=audit(1396889711.890:175): avc: denied { execute_no_trans } for pid=26284 comm="systemd-udevd" path="/usr/share/virtualbox/VBoxCreateUSBNode.sh" dev="sda5" ino=5899405 scontext=system_u:system_r:udev_t tcontext=unconfined_u:object_r:usr_t tclass=file
On ArchLinux, glibc package installs /usr/bin/getconf as a hard link to a file
in /usr/lib/getconf/. For example on a x86_64 machine:
$ ls -i -l /usr/bin/getconf /usr/lib/getconf/XBS5_LP64_OFF64
5900355 -rwxr-xr-x. 4 root root 22880 Feb 28 04:53 /usr/bin/getconf
5900355 -rwxr-xr-x. 4 root root 22880 Feb 28 04:53 /usr/lib/getconf/XBS5_LP64_OFF64
Such configuration produces an instability when labeling the files with
"restorecon -Rv /":
restorecon reset /usr/bin/getconf context unconfined_u:object_r:lib_t:s0->unconfined_u:object_r:bin_t:s0
restorecon reset /usr/lib/getconf/XBS5_LP64_OFF64 context unconfined_u:object_r:bin_t:s0->unconfined_u:object_r:lib_t:s0
As /usr/lib/getconf directory only contains executable programs, this issue is
fixed by labeling this directory and its content "bin_t".
/sys/fs/cgroup is a tmpfs which contains cgroup mounts and symlinks such as
cpu and cpuacct. Running restorecon makes this warning happen:
restorecon: Warning no default label for /sys/fs/cgroup/cpu
Declare a file context for every symlink in the cgroup tmpfs montpoint to
no longer have such warning.
Even if there is not FHS provision for this, systemd is using
/dev/hugepages to mount the hugetlbfs fs by default.
The needed file contexts are already present
Second part of the support of security_t under /sys/fs/selinux - when
asked not to audit getting attributes on the selinux file system, have
this propagate to the sysfs parts as well.
Signed-off-by: Sven Vermeulen <sven.vermeulen@siphos.be>
Because it is no longer a top-level file system, we need to enhance some
of the interfaces with the appropriate rights towards sysfs_t.
First set to allow getattr rights on the file system, which now also
means getattr on the sysfs_t file system as well as search privileges in
sysfs_t.
Signed-off-by: Sven Vermeulen <sven.vermeulen@siphos.be>
On Debian, this directory can contain locally trusted certificates that
will be then be symlinked to /etc/ssl/certs by
update-ca-certificates(8), the files should be labelled as cert_t.
The postgresql manpages are installed under a private directory, some of
them are symlinked to the usual location.
Properly labeling them ensure that mandb can read them.
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=740591
Associate the new xattrfs attribute to fs_t and the pseudo filesystems
that we know support xattr
This patch adds the attribute to the following (pseudo) filesystems
- device_t
- devpts_t
- fs_t
- hugetlbfs
- sysfs_t
- tmpfs_t
In glibc, the get_nprocs method reads /sys/devices/system/cpu/online, so
we need to grant most domains read access to this file. As we don't want
them to have read access on sysfs_t by default, create a new type
(cpu_online_t) and assign it to the file, and grant domains read access
to the file.
This does require systems to relabel the file upon every boot, something
distributions do in their bootup scripts, as /sys devices don't keep
their context.
Signed-off-by: Sven Vermeulen <sven.vermeulen@siphos.be>
When sudo is invoked (sudo -i) the audit log gets quite a lot of denials
related to the getattr permission against tty_device_t:chr_file for the
*_sudo_t domain. However, no additional logging (that would hint at a
need) by sudo, nor any functional issues come up.
Hence the dontaudit call.
Signed-off-by: Sven Vermeulen <sven.vermeulen@siphos.be>
Unprivileged users don't need to read kallsyms and /boot/System.map.
This allow rule was introduced in the initial revision of userdomain.if in
2005, with commit b16c6b8c32:
# cjp: why?
bootloader_read_kernel_symbol_table($1_t)
FATsort is an utility to sort directory entries on FAT partitions, see
http://fatsort.sourceforge.net/ . It requires direct access to the
block devices.
If loopback devices are not pregenerated (kernel option
CONFIG_BLK_DEV_LOOP_MIN_COUNT=0), mount needs to write to
/dev/loop-control do create them dynamically when needed.
Also allow sshd_t domain to chroot(2) in this directory as explained in
the README.privsep file in the openssh tarball.
Thanks to Russell Coker for this patch
dpkg is now using rpm_execcon()/setexecfilecon()-like function to
transition to the dpkg_script_t domain. This function will fail in
enforcing mode if the transition is not allowed.
Make it consistent with sysadm_r:sysadm_t.
If you build targeted policy then consider direct_initrc=y
If you build with direct_initrc=n then both unconfined_r:unconfined_t,
as well as sysadm_r:sysadm_t rely on run_init for running services on
behalf of the system.
Signed-off-by: Dominick Grift <dominick.grift@gmail.com>
Callers on init_run_daemon() role and domain transition on all
init_script_file_type to system_r and initrc_t respectively.
The old behavior of role and domain transitioning on init daemon entry
files was causing problems with programs that can be run both by system
and session.
Signed-off-by: Dominick Grift <dominick.grift@gmail.com>
Udev is writing persistent rules in /etc/udev/rules.d to ensure the
network interfaces and storage devices have a persistent name.
This patch has been taken from the Fedora policy
It would not be sufficient in the current shape anyways because
unconfined_r is not associated with xserver_t
Signed-off-by: Dominick Grift <dominick.grift@gmail.com>
The syslog-ng logger has (build-optional) support for capabilities. If
capabilities support is enabled, running it without setcap/getcap
permissions gives the following upon start:
* Starting syslog-ng ...
syslog-ng: Error setting capabilities, capability management disabled;
error='Permission denied' [ ok ]
Granting only setcap (initial AVC seen) does not fully help either:
* Starting syslog-ng ...
Error managing capability set, cap_set_proc returned an error;
With setcap and getcap enabled, syslog-ng starts and functions fine.
See also https://bugs.gentoo.org/show_bug.cgi?id=488718
Reported-by: Vincent Brillault <gentoo@lerya.net>
Signed-off-by: Sven Vermeulen <sven.vermeulen@siphos.be>
With new userspace, trying to build a SELinux policy (and load it)
fails:
~# semodule -B
libsemanage.semanage_install_active: Unable to create sybolic link from
/etc/selinux/mcs/modules/active/policy.kern to
/etc/selinux/mcs/policy/policy.28 error code 0. (Permission denied).
AVC shows a denial for the semodule command, running as semanage_t,
trying to create a lnk_file in semanage_module_t.
