We already have neverallow rules for all domains about
loading policy, setting enforcing mode, and setting
checkreqprot, so we can drop redundant ones from netd and appdomain.
Add neverallow rules to domain.te for setbool and setsecparam
and exclude them from unconfined to allow fully eliminating
separate neverallow rules on the :security class from anything
other than domain.te.
Change-Id: I0122e23ccb2b243f4c5376893e0c894f01f548fc
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
This is a world-readable directory anyway and will help to
address a small number of new denials.
Change-Id: I9e53c89a19da8553cbcbef8295c02ccaaa5d564c
Signed-off-by: rpcraig <rpcraig@tycho.ncsc.mil>
Replace * or any permission set containing create with
create_socket_perms or create_stream_socket_perms.
Add net_domain() to all domains using network sockets and
delete rules already covered by domain.te or net.te.
For netlink_route_socket, only nlmsg_write needs to be separately
granted to specific domains that are permitted to modify the routing
table. Clarification: read/write permissions are just ability to
perform read/recv() or write/send() on the socket, whereas nlmsg_read/
nlmsg_write permissions control ability to observe or modify the
underlying kernel state accessed via the socket.
See security/selinux/nlmsgtab.c in the kernel for the mapping of
netlink message types to nlmsg_read or nlmsg_write.
Delete legacy rule for b/12061011.
This change does not touch any rules where only read/write were allowed
to a socket created by another domain (inherited across exec or
received across socket or binder IPC). We may wish to rewrite some or all
of those rules with the rw_socket_perms macro but that is a separate
change.
Change-Id: Ib0637ab86f6d388043eff928e5d96beb02e5450e
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
If we are going to allow all domains to search and
stat the contents of /data/security, then we should
also allow them to read the /data/security/current symlink
created by SELinuxPolicyInstallReceiver to the directory
containing the current policy update.
Change-Id: Ida352ed7ae115723964d2723f1115a87af438013
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
Add initial support for uncrypt, started via the
pre-recovery service in init.rc. On an encrypted device,
uncrypt reads an OTA zip file on /data, opens the underlying
block device, and writes the unencrypted blocks on top of the
encrypted blocks. This allows recovery, which can't normally
read encrypted partitions, to reconstruct the OTA image and apply
the update as normal.
Add an exception to the neverallow rule for sys_rawio. This is
needed to support writing to the raw block device.
Add an exception to the neverallow rule for unlabeled block devices.
The underlying block device for /data varies between devices
within the same family (for example, "flo" vs "deb"), and the existing
per-device file_context labeling isn't sufficient to cover these
differences. Until I can resolve this problem, allow access to any
block devices.
Bug: 13083922
Change-Id: I7cd4c3493c151e682866fe4645c488b464322379
Only allow to domains as required and amend the existing
neverallow on block_device:blk_file to replace the
exemption for unconfineddomain with an explicit whitelist.
The neverallow does not check other device types as specific
ones may need to be writable by device-specific domains.
Change-Id: I0f2f1f565e886ae110a719a08aa3a1e7e9f23e8c
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
Remove sys_ptrace and add a neverallow for it.
Remove sys_rawio and mknod, explicitly allow to kernel, init, and recovery,
and add a neverallow for them.
Remove sys_module. It can be added back where appropriate in device
policy if using a modular kernel. No neverallow since it is device
specific.
Change-Id: I1a7971db8d247fd53a8f9392de9e46250e91f89b
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
Only allow to specific domains as required, and add a neverallow
to prevent allowing it to other domains not explicitly whitelisted.
sdcard_type is exempted from the neverallow since more domains
require the ability to mount it, including device-specific domains.
Change-Id: Ia6476d1c877f5ead250749fb12bff863be5e9f27
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
This new type will allow us to write finer-grained
policy concerning asec containers. Some files of
these containers need to be world readable.
Change-Id: Iefee74214d664acd262edecbb4f981d633ff96ce
Signed-off-by: rpcraig <rpcraig@tycho.ncsc.mil>
- Add write_logd, read_logd and control_logd macros added along
with contexts for user space logd.
- Specify above on domain wide, or service-by-service basis
- Add logd rules.
- deprecate access_logcat as unused.
- 'allow <domain> zygote:unix_dgram_socket write;' rule added to
deal with fd inheritance. ToDo: investigate means to allow
references to close, and reopen in context of application
or call setsockcreatecon() to label them in child context.
Change-Id: I35dbb9d5122c5ed9b8c8f128abf24a871d6b26d8
Rather then allowing open,read,write to raw block devices, one
should relabel it to something more specific.
vold should be re-worked so we can drop it from this assert.
Change-Id: Ie891a9eaf0814ea3878d32b18b4e9f4d7dac4faf
Linux defines two capabilities for Mandatory Access Control (MAC)
security modules, CAP_MAC_OVERRIDE (override MAC access restrictions)
and CAP_MAC_ADMIN (allow MAC configuration or state changes).
SELinux predates these capabilities and did not originally use them,
but later made use of CAP_MAC_ADMIN as a way to control the ability
to set security context values unknown to the currently loaded
SELinux policy on files. That facility is used in Linux for e.g.
livecd creation where a file security context that is being set
on a generated filesystem is not known to the build host policy.
Internally, files with such labels are treated as having the unlabeled
security context for permission checking purposes until/unless the
context is later defined through a policy reload.
CAP_MAC_OVERRIDE is never checked by SELinux, so it never needs
to be allowed. CAP_MAC_ADMIN is only checked if setting an
unknown security context value; the only legitimate use I can see
in Android is the recovery console, where a context may need to be set
on /system that is not defined in the recovery policy.
Remove these capabilities from unconfined domains, allow
mac_admin for the recovery domain, and add neverallow rules.
Change-Id: Ief673e12bc3caf695f3fb67cabe63e68f5f58150
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
When adbd runs as root, it transitions into the
su domain. Add the various rules to support this.
This is needed to run the adbd and shell domains in
enforcing on userdebug / eng devices without breaking
developer workflows.
Change-Id: Ib33c0dd2dd6172035230514ac84fcaed2ecf44d6
powervr_device is obsoleted by the more general gpu_device.
akm_device and accelerometer_device are obsoleted by the more
general sensors_device.
We could also drop the file_contexts entries altogether and
take them to device-specific policy (in this case, they all
came from crespo, so that is obsolete for master).
Change-Id: I63cef43b0d66bc99b80b64655416cc050f443e7d
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
We do not want to permit connecting to arbitrary unconfined services
left running in the init domain. I do not know how this was originally
triggered and thus cannot test that it is fixed. Possible causes:
- another service was left running in init domain, e.g. dumpstate,
- there was a socket entry for the service in the init.rc file
and the service was launched via logwrapper and therefore init did
not know how to label the socket.
The former should be fixed. The latter can be solved either by
removing use of logwrapper or by specifying the socket context
explicitly in the init.rc file now.
Change-Id: I09ececaaaea2ccafb7637ca08707566c1155a298
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
Just use notdevfile_class_set to pick up all non-device file classes.
Change-Id: Ib3604537ccfc25da67823f0f2b5d70b84edfaadf
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
Otherwise all domains can create/write files that are executable
by all other domains. If I understand correctly, this should
only be necessary for app domains executing content from legacy
unlabeled userdata partitions on existing devices and zygote
and system_server mappings of dalvikcache files, so only allow
it for those domains.
If required for others, add it to the individual
domain .te file, not for all domains.
Change-Id: I6f5715eb1ecf2911e70772b9ab4e531feea18819
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
Now that we set /sys/fs/selinux/checkreqprot via init.rc,
restrict the ability to set it to only the kernel domain.
Change-Id: I975061fd0e69c158db9bdb23e6ba77948e3fead1
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
/proc/sys/net could use its own type to help distinguish
among some of the proc access rules. Fix dhcp and netd
because of this.
Change-Id: I6e16cba660f07bc25f437bf43e1eba851a88d538
Signed-off-by: rpcraig <rpcraig@tycho.ncsc.mil>
init can't handle binder calls. It's always incorrect
to allow init:binder call, and represents a binder call
to a service without an SELinux domain. Adding this
allow rule was a mistake; the dumpstate SELinux domain didn't
exist at the time this rule was written, and dumpstate was
running under init's domain.
Add a neverallow rule to prevent the reintroduction of
this bug.
Change-Id: I78d35e675fd142d880f15329471778c18972bf50
execmem permission controls the ability to make an anonymous
mapping executable or to make a private file mapping writable
and executable. Remove this permission from domain (i.e.
all domains) by default, and add it explicitly to app domains.
It is already allowed in other specific .te files as required.
There may be additional cases in device-specific policy where
it is required for proprietary binaries.
Change-Id: I902ac6f8cf2e93d46b3a976bc4dabefa3905fce6
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
Confine the domain for an adb shell in -user builds only.
The shell domain in non-user builds is left permissive.
init_shell (shell spawned by init, e.g. console service)
remains unconfined by this change.
Introduce a shelldomain attribute for rules common to all shell
domains, assign it to the shell types, and add shelldomain.te for
its rules.
Change-Id: I01ee2c7ef80b61a9db151abe182ef9af7623c461
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
And allow any SELinux domain to read these timezone
related files.
Addresses the following denial:
<5>[ 4.746399] type=1400 audit(3430294.470:7): avc: denied { open } for pid=197 comm="time_daemon" name="tzdata" dev="mmcblk0p28" ino=618992 scontext=u:r:time:s0 tcontext=u:object_r:system_data_file:s0 tclass=file
Change-Id: Iff32465e62729d7aad8c79607848d89ce0aede86
Remove init, ueventd, watchdogd, healthd and adbd from the set of
domains traceable by debuggerd. bionic/linker/debugger.cpp sets up
handlers for all dynamically linked programs in Android but this
should not apply for statically linked programs.
Exclude ptrace access from unconfineddomain.
Prohibit ptrace access to init via neverallow.
Change-Id: I70d742233fbe40cb4d1772a4e6cd9f8f767f2c3a
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
As per the discussion in:
https://android-review.googlesource.com/#/c/71184/
init sets the enforcing mode in its code prior to switching to
the init domain via a setcon command in the init.rc file. Hence,
the setenforce permission is checked while still running in the
kernel domain. Further, as init has no reason to ever set the
enforcing mode again, we do not need to allow setenforce to the
init domain and this prevents reverting to permissive
mode via an errant write by init later. We could technically
dontaudit the kernel setenforce access instead since the first
call to setenforce happens while still permissive (and thus we
never need to allow it in policy) but we allow it to more accurately
represent what is possible.
Change-Id: I70b5e6d8c99e0566145b9c8df863cc8a34019284
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
The build is broken. Reverting temporarily to fix breakage.
libsepol.check_assertion_helper: neverallow on line 4758 violated by allow init kernel:security { setenforce };
Error while expanding policy
make: *** [out/target/product/mako/obj/ETC/sepolicy_intermediates/sepolicy] Error 1
make: *** Waiting for unfinished jobs....
This reverts commit bf12e22514.
Change-Id: I78a05756d8ce3c7d06e1d9d27e6135f4b352bb85
As per the discussion in:
https://android-review.googlesource.com/#/c/71184/
init sets the enforcing mode in its code prior to switching to
the init domain via a setcon command in the init.rc file. Hence,
the setenforce permission is checked while still running in the
kernel domain. Further, as init has no reason to ever set the
enforcing mode again, we do not need to allow setenforce to the
init domain and this prevents reverting to permissive
mode via an errant write by init later. We could technically
dontaudit the kernel setenforce access instead since the first
call to setenforce happens while still permissive (and thus we
never need to allow it in policy) but we allow it to more accurately
represent what is possible.
Change-Id: I617876c479666a03167b8fce270c82a8d45c7cc6
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
Limit the ability to write to the files that configure kernel
usermodehelpers and security-sensitive proc settings to the init domain.
Permissive domains can also continue to set these values.
The current list is not exhaustive, just an initial set.
Not all of these files will exist on all kernels/devices.
Controlling access to certain kernel usermodehelpers, e.g. cgroup
release_agent, will require kernel changes to support and cannot be
addressed here.
Expected output on e.g. flo after the change:
ls -Z /sys/kernel/uevent_helper /proc/sys/fs/suid_dumpable /proc/sys/kernel/core_pattern /proc/sys/kernel/dmesg_restrict /proc/sys/kernel/hotplug /proc/sys/kernel/kptr_restrict /proc/sys/kernel/poweroff_cmd /proc/sys/kernel/randomize_va_space /proc/sys/kernel/usermodehelper
-rw-r--r-- root root u:object_r:usermodehelper:s0 uevent_helper
-rw-r--r-- root root u:object_r:proc_security:s0 suid_dumpable
-rw-r--r-- root root u:object_r:usermodehelper:s0 core_pattern
-rw-r--r-- root root u:object_r:proc_security:s0 dmesg_restrict
-rw-r--r-- root root u:object_r:usermodehelper:s0 hotplug
-rw-r--r-- root root u:object_r:proc_security:s0 kptr_restrict
-rw-r--r-- root root u:object_r:usermodehelper:s0 poweroff_cmd
-rw-r--r-- root root u:object_r:proc_security:s0 randomize_va_space
-rw------- root root u:object_r:usermodehelper:s0 bset
-rw------- root root u:object_r:usermodehelper:s0 inheritable
Change-Id: I3f24b4bb90f0916ead863be6afd66d15ac5e8de0
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
This label was originally used for Motorola
Xoom devices. nvmap is the tegra gpu memory
manager and the various nvhost drivers are
for tegra graphics related functionality,
i.e. display serial interface, image signal
processor, or media processing stuff.
Only grouper and tilapia presently need this
policy.
Change-Id: I2a7000f69abf3185724d88d428e8237e0ca436ec
Also make su and shell permissive in non-user builds to allow
use of setenforce without violating the neverallow rule.
Change-Id: Ie76ee04e90d5a76dfaa5f56e9e3eb7e283328a3f
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
Every device has a CPU. This is not device specific.
Allow every domain to read these files/directories.
For unknown reasons, these files are accessed by A LOT
of processes.
Allow ueventd to write to these files. This addresses
the following denials seen on mako:
<5>[ 4.935602] type=1400 audit(1383167737.512:4): avc: denied { read } for pid=140 comm="ueventd" name="cpu0" dev="sysfs" ino=3163 scontext=u:r:ueventd:s0 tcontext=u:object_r:sysfs_devices_system_cpu:s0 tclass=dir
<5>[ 4.935785] type=1400 audit(1383167737.512:5): avc: denied { open } for pid=140 comm="ueventd" name="cpu0" dev="sysfs" ino=3163 scontext=u:r:ueventd:s0 tcontext=u:object_r:sysfs_devices_system_cpu:s0 tclass=dir
<5>[ 4.935937] type=1400 audit(1383167737.512:6): avc: denied { search } for pid=140 comm="ueventd" name="cpu0" dev="sysfs" ino=3163 scontext=u:r:ueventd:s0 tcontext=u:object_r:sysfs_devices_system_cpu:s0 tclass=dir
<5>[ 4.936120] type=1400 audit(1383167737.512:7): avc: denied { write } for pid=140 comm="ueventd" name="uevent" dev="sysfs" ino=3164 scontext=u:r:ueventd:s0 tcontext=u:object_r:sysfs_devices_system_cpu:s0 tclass=file
<5>[ 4.936303] type=1400 audit(1383167737.512:8): avc: denied { open } for pid=140 comm="ueventd" name="uevent" dev="sysfs" ino=3164 scontext=u:r:ueventd:s0 tcontext=u:object_r:sysfs_devices_system_cpu:s0 tclass=file
Change-Id: I4766dc571762d8fae06aa8c26828c070b80f5936
* Keep ueventd in permissive
* Drop unconfined macro to collect logs
* Restore allow rules to current NSA maintained policy
Change-Id: Ic4ee8e24ccd8887fed151ae1e4f197512849f57b
/dev/hw_random is accessed only by init and by EntropyMixer (which
runs inside system_server). Other domains are denied access because
apps/services should be obtaining randomness from the Linux RNG.
Change-Id: Ifde851004301ffd41b2189151a64a0c5989c630f
Some file types used as domain entrypoints were missing the
exec_type attribute. Add it and add a neverallow rule to
keep it that way.
Change-Id: I7563f3e03940a27ae40ed4d6bb74181c26148849
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
This is a follow-up CL to the extraction of "system_app" domain
from the "system" domain which left the "system" domain encompassing
just the system_server.
Since this change cannot be made atomically across different
repositories, it temporarily adds a typealias "server" pointing to
"system_server". Once all other repositories have been switched to
"system_server", this alias will be removed.
Change-Id: I90a6850603dcf60049963462c5572d36de62bc00
Remove sys_nice capability from domains; this does not appear to be necessary
and should not be possible in particular for app domains. If we encounter
specific instances where it should be granted, we can add it back on a
per-domain basis. Allow it explicitly for the system_server. Unconfined
domains get it via unconfined_domain() and the rules in unconfined.te.
Change-Id: I9669db80a04a90a22241b2fbc5236a28dcde8c6e
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
3.4 goldfish kernel supports sysfs labeling so we no longer need this.
Change-Id: I77514a8f3102ac8be957c57d95e7de7d5901f69d
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
Apps attempting to write to /dev/random or /dev/urandom currently
succeed, but a policy violation is logged. These two Linux RNG
devices are meant to be written to by arbitrary apps. Thus, there's
no reason to deny this capability.
Bug: 10679705
Change-Id: Ife401f1dd2182889471eef7e90fcc92e96f9c4d6
For some reason, the debuggerd socket isn't getting properly
labeled. Work around this bug for now by allowing all domains
to connect to all unix stream sockets.
Bug: 9858255
Change-Id: If994e51b0201ea8cae46341efc76dc71a4e577c8
This is my first attempt at creating an enforcing SELinux domain for
apps, untrusted_apps, and isolated_apps. Much of these rules are based on the
contents of app.te as of commit 11153ef349
with extensive modifications, some of which are included below.
* Allow communication with netd/dnsproxyd, to allow netd to handle
dns requests
* Allow binder communications with the DNS server
* Allow binder communications with surfaceflinger
* Allow an app to bind to tcp/udp ports
* Allow all domains to read files from the root partition, assuming
the DAC allows access.
In addition, I added a bunch of "neverallow" rules, to assert that
certain capabilities are never added.
This change has a high probability of breaking someone, somewhere.
If it does, then I'm happy to fix the breakage, rollback this change,
or put untrusted_app into permissive mode.
Change-Id: I83f220135d20ab4f70fbd7be9401b5b1def1fe35
Remove "self:process ptrace" from all SELinux enforced domains.
In general, a process should never need to ptrace itself.
We can add this back to more narrowly scoped domains as needed.
Add a bunch of neverallow assertions to netd.te, to verify that netd
never gets unexpected capabilities.
Change-Id: Ie862dc95bec84068536bb64705667e36210c5f4e
