Label some shell scripts from bridge-utils correctly. Maybe have ifdef
distro_debian around this, not sure what upstream is doing.
systemd_nspawn_t needs to manage the /etc/localtime symlink if you have a
labeled chroot.
Another dontaudit for mon_local_test_t to stop it spamming the logs.
Support a .d directory for dnsmasq config files.
Do not audit ioctl operation attempts whenever write
operations on the xserver log should not be audited.
Signed-off-by: Guido Trentalancia <guido@trentalancia.com>
Create new interface init_rlimit_inherit to allow a process started by init to inherit resource limits. systemd allows for setting of resource limits [1] but the default from SELinux is to not allow the inheritance of those limits as a service is started. This interface allows that resource limit inheritance.
The systemd .service options are LimitCPU=, LimitFSIZE=, LimitDATA=, LimitSTACK=, LimitCORE=, LimitRSS=, LimitNOFILE=, LimitAS=, LimitNPROC=, LimitMEMLOCK=, LimitLOCKS=, LimitSIGPENDING=, LimitMSGQUEUE=, LimitNICE=, LimitRTPRIO=, LimitRTTIME=
[1] https://www.freedesktop.org/software/systemd/man/systemd.exec.html
Signed-off-by: Dave Sugar <dsugar@tresys.com>
Alter interface init_startstop_service to also allow for the status permission. systemctl start <foo> and systemctl stop <foo> work correctly. But systemctl restart <foo> will fail as restart uses status to determine the action to take.
This interface is used by many other modules (like iptables, logging, apache, cron, etc... - see 'admin' interface). This allows restart to work for all these services.
Signed-off-by: Dave Sugar <dsugar@tresys.com>
We have a use case on a system where we have a systemd .service unit file that is using the SELinuxContext= [1] option to specify a context for the service being started. The same .service file (/lib/systemd/system/foo@.service) is used to start multiple instances of the same executable that are customized with a different drop-in .conf file for each. The context is customized in /lib/systemd/system/foo@.service file (based on using SELinuxContext=system_u:system_r:foo_%i_t:s0) [2]
We then create /etc/systemd/system/foo@bar.service.d/bar.conf so the final running process is in the domain foo_bar_t
We have created the following interface (in init.if) to meet our needs. The interface is very much like init_daemon_domain except for the use of spec_domtrans_pattern rather than domtrans_pattern because the automatic transition doesn't work in this case.
[1] The SELinuxContext option for systemd is explained https://www.freedesktop.org/software/systemd/man/systemd.exec.html
[2] The systemd %i (and other specifiers) along with drop-in files are explained https://www.freedesktop.org/software/systemd/man/systemd.unit.html
Signed-off-by: Dave Sugar <dsugar@tresys.com>
Those mmap() calls are non-essential, though. They're issued by
fdopen(), which can also fall back to read().
Since there are few reasons to grant read access to audit logs other
than for using ausearch, I've added the dontaudits directly to the
relevant read interfaces.
Note that not only kmod needs this permission, other libkmod consumers
like udev require it, too. Hence I'm adding the permission to the
relevant interfaces.
systemd relabels everything in /dev and /run after loading the policy.
In these directories there is a socket file for udev, /run/udev/control,
which is not allowed to be relabelled from tmpfs_t:
avc: denied { relabelfrom } for pid=1 comm="systemd"
name="control" dev="tmpfs" ino=7027
scontext=system_u:system_r:init_t tcontext=system_u:object_r:tmpfs_t
tclass=sock_file permissive=1
avc: denied { relabelto } for pid=1 comm="systemd" name="control"
dev="tmpfs" ino=7027 scontext=system_u:system_r:init_t
tcontext=system_u:object_r:udev_var_run_t tclass=sock_file
permissive=1
/run/tmpfiles.d directory (used by systemd-tmpfiles service) is also
relabelled.
This is currently denied:
avc: denied { create } for pid=1 comm="systemd" name="pts"
scontext=system_u:system_r:init_t
tcontext=system_u:object_r:devpts_t tclass=dir permissive=1
This is a minor update of the last attempt at this patch.
Changes in .fc to label /etc/rsyslog.d(/.*)? as syslog_conf_t so all rsyslog config files are labeled syslog_conf_t (not just /etc/r?syslog.conf). Update .te file to allow rsyslog to read the directory now labeled syslog_conf_t (files of this type were already readable). Final (and new) change is in logging_admin_syslog interface so files_etc_filetrans now includes the optional filename so /etc/r?syslog.conf would be labeled correctly when created in etc_t.
The overall goal of this patch is that a domain using the logging_admin_syslog is able to create/edit files in /etc/rsyslog.d and they get created as syslog_conf_t AND other files created in /etc (or other etc_t labeled directory) don't get created with the syslog_conf_t type as they are not necessarily syslog configuration files.
Dave Sugar
dsugar@tresys.com
Signed-off-by: Dave Sugar <dsugar@tresys.com>
Label RHEL specific systemd binaries /usr/lib/systemd/rhel* as initrc_exec_t.
Now in the proper location.
Signed-off-by: Dave Sugar <dsugar@tresys.com>
systemd tries to create /dev/pts directly with its context type
"devpts_t", but this is not allowed:
avc: denied { associate } for pid=1 comm="systemd" name="pts"
scontext=system_u:object_r:devpts_t
tcontext=system_u:object_r:device_t
tclass=filesystem permissive=1
The following patch creates two additional interfaces for tun_tap_device_t to grant only read or only write access (rather than both read and write access). It is possible to open a tap device for only reading or only writing and this allows policy to match that use.
Signed-off-by: Dave Sugar <dsugar@tresys.com>
The following patch allows systemd_nspawn_t to create directories under /tmp
and use them as mountpoints. Also allows systemd_nspawn_t to umount cgroup
filesystems.
Allows systemd_backlight_t to search /var/lib.
Commit 2e7553db63 ("Create / to /usr equivalence for bin, sbin, and
lib, from Russell Coker.") removed from corecommands.fc:
/usr/(.*/)?bin(/.*)? gen_context(system_u:object_r:bin_t,s0)
Consequently files like /usr/x86_64-w64-mingw32/bin/objdump get labeled
as usr_t:
$ matchpathcon /usr/x86_64-w64-mingw32/bin/objdump
/usr/x86_64-w64-mingw32/bin/objdump system_u:object_r:usr_t
Make such files labeled as bin_t again.
When auditctl logs a message to syslog, it needs to read /etc/localtime.
This is currently denied:
avc: denied { read } for pid=191 comm="auditctl" name="UTC"
dev="vda1" ino=394043 scontext=system_u:system_r:auditctl_t
tcontext=system_u:object_r:locale_t tclass=file permissive=1
avc: denied { open } for pid=191 comm="auditctl"
path="/usr/share/zoneinfo/UTC" dev="vda1" ino=394043
scontext=system_u:system_r:auditctl_t
tcontext=system_u:object_r:locale_t tclass=file permissive=1
This occurs for example at boot time when "/usr/bin/augenrules --load"
is run [1]. Here is an extract of "strace -s 256 -f /usr/bin/augenrules
--load":
[pid 635] execve("/sbin/auditctl", ["/sbin/auditctl", "-R",
"/etc/audit/audit.rules"], 0x1e77d80 /* 16 vars */) = 0
...
[pid 635] open("/etc/audit/audit.rules", O_RDONLY) = -1 ENOENT (No
such file or directory)
[pid 635] open("/etc/localtime", O_RDONLY|O_CLOEXEC) = 4
[pid 635] fstat(4, {st_mode=S_IFREG|0644, st_size=127, ...}) = 0
[pid 635] fstat(4, {st_mode=S_IFREG|0644, st_size=127, ...}) = 0
[pid 635] read(4, "TZif2\0\0\0\0\0\0\0\0\0\0\0"..., 4096) = 127
[pid 635] lseek(4, -71, SEEK_CUR) = 56
[pid 635] read(4, "TZif2\0\0\0\0\0\0\0\0\0\0\0", 4096) = 71
[pid 635] close(4) = 0
[pid 635] socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0) = 4
[pid 635] connect(4, {sa_family=AF_UNIX, sun_path="/dev/log"}, 110) = 0
[pid 635] sendto(4, "<14>Aug 12 08:59:53 auditctl: file
/etc/audit/audit.rules doesn't exist, skipping", 81, MSG_NOSIGNAL,
NULL, 0) = 81
[pid 635] exit_group(0) = ?
More precisely, auditctl uses vsyslog() from glibc, which uses
localtime_r() to fetch the time is in local timezone.
[1] On a systemd system, this command is automatically run by
auditd.service unit, cf.
https://github.com/linux-audit/audit-userspace/blob/v2.7.7/init.d/auditd.service#L21
Remove complement (~) and wildcard (*) in allow rules so that there are no
unintentional additions when new permissions are declared.
This patch does not add or remove permissions from any rules.
When a program uses init_module() to load a module, the kernel checks
for system:load_module permission in the process type [1].
For example when systemd loads ip_tables modules (since
1d3087978a),
the following AVC denial gets reported:
avc: denied { module_load } for pid=1 comm="systemd"
scontext=system_u:system_r:init_t tcontext=system_u:system_r:init_t
tclass=system permissive=1
[1] The relevant kernel code is selinux_kernel_module_from_file() in
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/security/selinux/hooks.c?h=v4.11#n3836
/* init_module */
if (file == NULL)
return avc_has_perm(sid, sid, SECCLASS_SYSTEM,
SYSTEM__MODULE_LOAD, NULL);
In this code, both source and target SIDs are current_sid().
Since Xen 4.7, /dev/xen/privcmd is used instead of /proc/xen/privcmd.
Add the device into the policy so `xenstored` can work.
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Not auditing this turns out to be the wrong choice for
several reasons.
For normal application functioning the user domain
should be able to create netlink_kobject_uvent_socket
sockets.
Signed-off-by: Guido Trentalancia <guido@trentalancia.com>
This won't match subdirectories of /usr/lib, but that shouldn't be a
problem, since we have "allow domain lib_t ..." anyway.
We can't match on "/usr/(.*/)?man(/.*)?", since that'd result in a few
false positives; in particular, the files
/usr/share/xmlto/format/docbook/man
/usr/share/bash-completion/completions/man
Kernel commit 6941857e82ae ("selinux: add a map permission check
for mmap") added a map permission check on mmap so that we can
distinguish memory mapped access (since it has different implications
for revocation). The purpose of a separate map permission check on
mmap(2) is to permit policy to prohibit memory mapping of specific files
for which we need to ensure that every access is revalidated, particularly
useful for scenarios where we expect the file to be relabeled at runtime
in order to reflect state changes (e.g. cross-domain solution, assured
pipeline without data copying). The kernel commit is anticipated to
be included in Linux 4.13.
This refpolicy change defines map permission for refpolicy. It mirrors
the definition in the kernel classmap by adding it to the common
definitions for files and sockets. This will break compatibility for
kernels that predate the dynamic class/perm mapping support (< 2.6.33,
< RHEL 6); on such kernels, one would instead need to add map permission
to the end of each file and socket access vector.
This change only allows map permission as needed, e.g. only in the
mmap_file_perms and exec_file_perms object permission sets
(since map is always required there) and only in specific interfaces
or modules where denials were observed in limited testing.
It is important to note that effective use of this permission requires
complete removal of unconfined, as otherwise unconfined domains will be
able to map all file types and therefore bypass the intended protection.
If we wanted to exclude map permission to all file types by default from
unconfined, we would need to add it to the list of permissions excluded from
files_unconfined_type in kernel/files.te.
Policies that depend on this permission not being allowed to specific file
types should also make use of neverallow rules to ensure that this is not
undermined by any allow rule, and ensure that they are performing neverallow
checking at policy build time (e.g. make validate) or runtime (e.g.
semanage.conf expand-check=1).
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
Every Infiniband network will have a default pkey, so that is labeled.
The rest of the pkey configuration is network specific. The policy allows
access to the default and unlabeled pkeys for sysadm and staff users.
kernel_t is allowed access to all pkeys, which it needs to process and
route management datagrams.
Endports are all unlabeled by default, sysadm users are allowed to
manage the subnet on unlabeled endports. kernel_t is allowed to manage
the subnet on all ibendports, which is required for configuring the HCA.
This patch requires selinux series: "SELinux user space support for
Infiniband RDMA", due to the new ipkeycon labeling mechanism.
Signed-off-by: Daniel Jurgens <danielj@mellanox.com>
This is the base part of the policy needed to support libmtp (an
Initiator implementation of the Media Transfer Protocol).
Signed-off-by: Guido Trentalancia <guido at trentalancia.net>
Linux kernel commit da69a5306ab9 ("selinux: support distinctions among all
network address families") triggers a build error if a new address family
is added without defining a corresponding SELinux security class. As a
result, the smc_socket class was added to the kernel to resolve a build
failure as part of merge commit 3051bf36c25d that introduced AF_SMC circa
Linux 4.11. Define this security class and its access vector, note that it
is enabled as part of the extended_socket_class policy capability, and add
it to the socket_class_set macro.
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
This permission was added to the kernel in commit 791ec491c372
("prlimit,security,selinux: add a security hook for prlimit")
circa Linux 4.12 in order to control the ability to get the resource
limits of another process. It is only checked when acting on another
process, so getrlimit permission is not required for use of getrlimit(2).
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
Setting this service option to 'full' or 'strict' will also remount the
/etc directory. Allow this in the policy.
This fixes the systemd-networkd service, but will also positively affect
any other service using the above hardening option.
Since app-admin/setools-4.1.0, some python internals try to create and
execute a file in /tmp during semanage initalization, causing semanage
to crash. Here's the backtrace (with the path
"/usr/lib64/python3.4/site-packages" replaced by "py" for brevity):
Traceback (most recent call last):
File "/usr/lib/python-exec/python3.4/semanage", line 28, in <module>
import seobject
File "py/seobject.py", line 34, in <module>
import sepolicy
File "py/sepolicy/__init__.py", line 8, in <module>
import setools
File "py/setools/__init__.py", line 77, in <module>
from .infoflow import InfoFlowAnalysis
File "py/setools/infoflow.py", line 22, in <module>
import networkx as nx
File "py/networkx/__init__.py", line 93, in <module>
import networkx.linalg
File "py/networkx/linalg/__init__.py", line 9, in <module>
from networkx.linalg.algebraicconnectivity import *
File "py/networkx/linalg/algebraicconnectivity.py", line 18, in <module>
from numpy import (array, asmatrix, asarray, dot, matrix, ndarray, ones,
File "py/numpy/__init__.py", line 180, in <module>
from . import add_newdocs
File "py/numpy/add_newdocs.py", line 13, in <module>
from numpy.lib import add_newdoc
File "py/numpy/lib/__init__.py", line 8, in <module>
from .type_check import *
File "py/numpy/lib/type_check.py", line 11, in <module>
import numpy.core.numeric as _nx
File "py/numpy/core/__init__.py", line 22, in <module>
from . import _internal # for freeze programs
File "py/numpy/core/_internal.py", line 14, in <module>
import ctypes
File "/usr/lib64/python3.4/ctypes/__init__.py", line 541, in <module>
_reset_cache()
File "/usr/lib64/python3.4/ctypes/__init__.py", line 280, in _reset_cache
CFUNCTYPE(c_int)(lambda: None)
MemoryError
Improve the initrc domain within the init module with some permissions
needed for a smoother boot.
Let the iptables init scripts read the iptables configuration.
Signed-off-by: Guido Trentalancia <guido@trentalancia.net>
Improve the locallogin module by curbing on dac_override permissions
in the sulogin domain (read/search permissions only).
Thanks to Dominick Grift for suggesting this.
Other modules are likely affected by the same issue.
Signed-off-by: Guido Trentalancia <guido@trentalancia.net>
This patch allows setfiles to use file handles inherited from apt (for dpkg
postinst scripts), adds those rsync permissions that were rejected previously
due to not using interfaces, allows fsadm_t to stat /run/mount/utab, and
allows system_cronjob_t some access it requires (including net_admin for
when it runs utilities that set buffers).
This patch is slightly more involved than just running sed. It also adds
typealias rules and doesn't change the FC entries.
The /dev/apm_bios device doesn't exist on modern systems. I have left that
policy in for the moment on the principle of making one change per patch. But
I might send another patch to remove that as it won't exist with modern
kernels.
I have used optional sections for dbus and xserver as requested and also
fixed a minor issue of a rule not being in the correct section.
Please merge this.
This patch fixes the policy for sulogin. It is very important
because without this patch, sulogin cannot work properly and
it should be considered that it is used as an emergency shell
when there are serious consistency errors in the system, so it
constitutes the only way to recover the system in such
circumstances.
Nowadays, sulogin never uses PAM (at least not the official one
from util-linux), so obsolete, confusing and buggy policy has
been removed.
Extensive testing carried out while creating this patch indicates
that there aren't other permissions needed to successfully run
sulogin.
This second version should apply cleanly to the latest git tree.
Signed-off-by: Guido Trentalancia <guido@trentalancia.net>
This patch fixes the xserver module so that the hidden .ICEauthority
file is created with the proper context (file transition).
It also optimizes a similar interface used for xauth home files.
Signed-off-by: Guido Trentalancia <guido@trentalancia.net>
This patch adds three new file contexts for script executables
belonging to new Gnome applications (weather application and
sound recorder).
Signed-off-by: Guido Trentalancia <guido@trentalancia.net>
Added mono_run for unconfined and also xserver_role and allow it to dbus
chat with xdm.
Allow sysadm_t to read kmsg.
Allow user domains to dbus chat with kerneloops for the kerneloops desktop
gui. Also allow them to chat with devicekit disk and power daemons.
Allow gconfd_t to read /var/lib/gconf/defaults and /proc/filesystems
Allow devicekit_power_t to chat to xdm via dbus and log via syslog.
Allow mount_t to do more with it's runtime files and stat more filesystem
types.
Allow xauth to send sigchld to xdm.
Allow semanage to search policy_src_t dirs and read /dev/urandom.
Patches for modutils, at least one of which is needed to generate an initramfs
on Debian.
Patch to allow lvm to talk to fifos from dpkg_script_t for postinst scripts
etc.
Patch for brctl to allow it to create sysfs files.
This patch adds a lot of policy that is needed to setup an initramfs and grub
on Debian nowadays.
Also changed a comment about ia64 to correctly mention EFI.
Put in libx32 subs entries that refer to directories with fc entries.
Allow dpkg_t to transition to dpkg_script_t when it executes bin_t for
dpkg-reconfigure.
Some dontaudit rules for mta processes spawned by mon for notification.
Lots of tiny changes that are obvious.
This patch adds an interface to manage systemd_passwd_var_run_t symlinks that
I'll add another patch to use shortly.
It has a number of changes needed by systemd_logind_t to set permissions for
local logins.
It has some more permissions that systemd_machined_t needs, I don't think it's
everything that systemd_machined_t needs but it's a start.
It has some changes for udev_t for systemd-udevd.
This patch lets mandb_t search init_var_run_t dirs which it needs when running
with systems. Also allows it to fs_getattr_xattr_fs() because it seemed
pointless to put that in a separate patch.
Allow init_t to do several things that it requires when init is systemd.
Allow various operations on var_log_t to access var_log_t symlinks too.
Let auditd setattr it's directory.
On systems such as Arch Linux, all programs which are usually located in
/bin, /sbin, /usr/bin and /usr/sbin are present in /usr/bin and the
other locations are symbolic links to this directory. With such a
configuration, the file contexts which define types for files in
/bin, /sbin and /usr/sbin need to be duplicated to provide definitions
for /usr/bin/...
As the "/bin vs. /usr/bin" part of the needed definitions has already
been done with the "usr merge" patches, the next step consists in
duplicating file contexts for /usr/sbin. This is what this patch does
for all modules which are not in contrib.
This is the second iteration of an idea I have previously posted on
http://oss.tresys.com/pipermail/refpolicy/2017-March/009176.html
mount.fc defines file contexts for /usr/bin/mount.*, /usr/bin/umount.*
and /usr/sbin/mount\.zfs. These patterns are not consistent for two
reasons:
- some distributions use /sbin/mount... for other file systems that zfs.
For example Debian uses /sbin/mount.ntfs-3g
(https://packages.debian.org/jessie/amd64/ntfs-3g/filelist)
- mount_exec_t type should only be applied to mount, umount, mount.$FS
and umount.udisks2, not mountpoint.
Replace the file patterns with ones that do not match mountpoint and
match every mount and umount programs in /usr/bin and /usr/sbin.
This patch doesn't do everything that is needed to have systemd-nspawn work.
But it does everything that is needed and which I have written in a clear and
uncontroversial way. I think it's best to get this upstream now and then
either have a separate discussion about the more difficult issues, or wait
until I devise a way of solving those problems that's not too hacky.
Who knows, maybe someone else will devise a brilliant solution to the remaining
issues after this is accepted upstream.
Also there's a tiny patch for systemd_machined_t that is required by
systemd_nspawn_t.
Description: systemd-nspawn
Author: Russell Coker <russell@coker.com.au>
Last-Update: 2017-03-29
I believe that I have addressed all the issues Chris raised, so here's a newer
version of the patch which applies to today's git version.
Description: systemd-resolved, sessions, and tmpfiles patches
Author: Russell Coker <russell@coker.com.au>
Last-Update: 2017-03-26
Here's the latest version of my patch to remove all /var/run when it's not
needed. I have removed the subst thing from the patch, but kept a
distro_debian bit that relies on it. So with this patch the policy won't
install if you build it with distro_debian unless you have my subst patch.
Chris, if your automated tests require that it build and install with
distro_debian then skip the patch for sysnetwork.fc.
From Russell Coker
The following patch adds dontaudit rules for where the net_admin capability
is requested due to SO_SNDBUFFORCE. This forces the caller to use SO_SNDBUF
which gives the same result but possibly a smaller buffer.
From Russell Coker
When using setfiles to validate file contexts of Debian modular policy
(with DISTRO=debian and MONOLITHIC=n), it fails with:
tmp/all_mods.fc: line 527 is missing fields
tmp/all_mods.fc: line 527 is missing fields
tmp/all_mods.fc: Invalid argument
Here is the content of tmp/all_mods.fc around line 527:
# this is a static /dev dir "backup mount"
# if you want to disable udev, youll have to boot permissive and relabel!
/dev/\.static -d system_u:object_r:device_t
/dev/\.static/dev -d system_u:object_r:device_t
/dev/\.static/dev/(.*)? <<none>>
'
The quote of "you'll" has been eaten by m4 and there is a spurious quote
on the last line, which is reported by setfiles. Fix this by removing
the quote in the comment.
Here is an example of a failed build on Travis-CI:
https://travis-ci.org/fishilico/selinux-refpolicy-patched/jobs/205951446
* reorder file contexts
* rename getty_var_run_t into getty_runtime_t and getty_etc_t into getty_conf_t
* remove unneeded and confusing init permissions
* delete access on generic pids
* remove invalid etc filetrans: getty has no create permissions for its config files
* restrict logfile access
These are needed by several patches I'm about to send.
Description: some new interfaces for init/systemd
Author: Russell Coker <russell@coker.com.au>
Last-Update: 2017-02-24
This patch defines rw_inherited_file_perms. It's needed by a few patches
I'm going to send soon so I need to get it in before they go in.
Also it's generally a good thing to have. We should reconsider some of the
other policy for whether it should use this.
When with systemd a program does not ship a systemd unit file but only a init script, systemd creates a pseudo service on the fly.
To be able to act on this service, add the target attribute init_script_file_type to the init_ACTION_all_units interfaces.
Useful for monit.
* no negative permission matching for newrole_t:process
* do not label /usr/lib/selinux as policy_src_t, otherwise semodule can not run /usr/lib/selinux/hll/pp
* reorder label for /run/restorecond.pid
* fix systemd related denials
userdomains should not alter labels of kernel pseudo filesystems, but allowing setfiles/restorecon(d) to check the contexts helps spotting incorrect labels
When you merged the mon patch you removed the ability for mon_t to execute
lib_t files.
The following patch re-enables the ability to execute alert scripts.
The following patch defines new macros rw_inherited_fifo_file_perms and
rw_inherited_term_perms for the obvious reason.
I've had this in Debian for a while and some Debian policy relies on it.
I think it's appropriate to include this before including any policy that
relies on it because it's an obvious foundation for writing good policy.
We could have inherited perms macros for other object types, but terminals
and fifos are the main ones that get inherited. The next best candidate
for such a macro is a sock_file, and that's largely due to systemd setting
programs stdout/stderr to unix domain sockets.
