I added ro.bionic.(2nd_)?_(arch|cpu_variant) to vendor system
properties. And have init to write them to files under dev/.
This change set SELinux rules for these properties and files.
For the system properties: vendor/default.prop will set them. init will
read them.
For the files /dev/cpu_variant:.*: init will write them. bionic libc
will read them. (Basically world readable).
This is to allow libc select the right optimized routine at runtime.
Like memcpy / strcmp etc.
Test: getprop to make sure the properties are set.
Test: ls -laZ to make sure /dev/cpu_variant:.* are correctly labeled.
Change-Id: I41662493dce30eae6d41bf0985709045c44247d3
The dynamic linker calls realpath(3) on paths found in the linker config
script. Since realpath() calls lstat() on the parent paths, not allowing
getattr on /apex and its subdirectories will cause selinux denial spam
whenever something is executed from APEXes.
Silence the spam by allowing getattr on apex_mnt_dir.
Bug: 117403679
Bug: 115787633
Test: m apex.test; m; device is bootable
Change-Id: Ic659582760a3ae146e73770266bc64332b36a97c
Auditallow added in commit 72edbb3e83 ("Audit generic debugfs access for
removal", May 01 2018) has not triggered. Remove allow rule and tighten
up neverallow rule.
Test: policy compiles
Test: no collected SELinux denials.
Change-Id: I9a90463575f9eab4711b72d6f444fa9d526b80e1
It is unnecessary to use get_prop() rules for the su domain. The
su domain is always in permissive mode [1] and not subject to SELinux
enforcement. It's also possible these rules were added to avoid SELinux
denial log spam from showing up, however, there are already dontaudit
rules in place [2] to prevent this.
Delete the unnecessary rules.
[1] 96b62a60c2/private/su.te (19)
[2] 96b62a60c2/public/su.te (42)
Test: policy compiles
Change-Id: I5913f360738725bf915f0606d381029b9ba4318f
Require all SELinux domains which have permission to perform ioctls on
/dev/tun explicitly specify what ioctls they perform. Only allow the
safe defaults FIOCLEX and FIONCLEX, which are alternate, uncommon ways
to set and unset the O_CLOEXEC flag.
Remove app's ability to issue *any* ioctls on /dev/tun, period. Add
neverallow assertions (compile time assertion + CTS test) to prevent
regressions.
Limit system_server's ability to perform ioctls on /dev/tun to FIOCLEX,
FIONCLEX, TUNGETIFF, and TUNSETIFF. Testing and source code examination
shows that only TUNGETIFF and TUNSETIFF are used by system_server.
The goal of this change is to put SELinux ioctl controls in place for
/dev/tun, so we don't have to maintain the custom kernel patch at
11cee2be0c%5E%21
Delete the neverallow assertion in isolated_app.te. This is already
covered by the assertion present in app_neverallows.te.
Test: cts-tradefed run cts -m CtsHostsideNetworkTests -t com.android.cts.net.HostsideVpnTests
Test: cts-tradefed run cts -m CtsHostsideNetworkTests
Test: cts-tradefed run cts -m CtsNetTestCases
Bug: 111560739
Bug: 111560570
Change-Id: Ibe1c3a9e880db0bee438535554abdbc6d84eec45
Already has permissions to remove the scratch partition, but to allow
more refined cleansing (eg: just remove vendor override), need the
ability to mount and scrub overlay directories.
Test: manual
Bug: 117605276
Change-Id: Ibc272c0aa7ce207280023912f5f119ccf5079a7f
Create a transient SELinux domain where system_server can perform
certain JIT setup. The idea is that system_server will start in the
system_server_startup domain, setup certain JIT pages, then perform a
one-way transition into the system_server domain. From that point,
further JITing operations are disallowed.
Bug: 62356545
Test: device boots, no permission errors
Change-Id: Ic55b2cc5aba420ebcf62736622e08881a4779004
Remove blanket coredomain access to same_process_hal_file in favor of
granular access. This change takes into account audits from go/sedenials
(our internal dogfood program)
Bug: 37211678
Test: m selinux_policy
Change-Id: I5634fb65c72d13007e40c131a600585a05b8c4b5
What changed:
- Tightening neverallow forbidding vendor execution access in /system.
In it's current form the neverallow is loose because not all executables
have exec_type attribute, e.g. almost everything in /system/bin/. This
change tightens up the neverallow by instead targeting system_file_type
attribute, which must be applied to all files in /system.
- Adding a general neverallow forbidding all access to files in /system
(bar exceptions)
TODOs:
- Remove loopholes once Treble violations are fixed across all internal
build targets.
Bug: 111243627
Test: m selinux_policy; build-only change
Change-Id: I150195756c0c3258904c3da0812bbd942ea2f229
Start enforcing the use of ioctl restrictions on all Android block
devices. Domains which perform ioctls on block devices must be explicit
about what ioctls they issue. The only ioctls allowed by default are
BLKGETSIZE64, BLKSSZGET, FIOCLEX, and FIONCLEX.
Test: device boots and no problems.
Change-Id: I1195756b20cf2b50bede1eb04a48145a97a35867
Allow a process to determine if a fifo_file (aka pipe, created from the
pipe() or pipe2() syscall) is a tty.
Addresses the following denials:
type=1400 audit(0.0:1307): avc: denied { ioctl } for comm="ls" path="pipe:[213117]" dev="pipefs" ino=213117 ioctlcmd=5401 scontext=u:r:hal_dumpstate_impl:s0 tcontext=u:r:hal_dumpstate_impl:s0 tclass=fifo_file permissive=0
type=1400 audit(0.0:22): avc: denied { ioctl } for comm="sh" path="pipe:[54971]" dev="pipefs" ino=54971 ioctlcmd=5401 scontext=u:r:untrusted_app_27:s0:c512,c768 tcontext=u:r:untrusted_app_27:s0:c512,c768 tclass=fifo_file permissive=0 app=com.zhihu.android
type=1400 audit(0.0:237): avc: denied { ioctl } for comm="sh" path="pipe:[56997]" dev="pipefs" ino=56997 ioctlcmd=5401 scontext=u:r:untrusted_app_25:s0:c512,c768 tcontext=u:r:untrusted_app_25:s0:c512,c768 tclass=fifo_file permissive=0 app=fm.qingting.qtradio
Test: policy compiles and device builds.
Change-Id: Ic4c6441d0ec33de8cda3f13ff529e98374897364
What changed:
- Tightening neverallow forbidding vendor execution access in /system.
In it's current form the neverallow is loose because not all executables
have exec_type attribute, e.g. almost everything in /system/bin/. This
change tightens up the neverallow by instead targeting system_file_type
attribute, which must be applied to all files in /system.
- Adding a general neverallow forbidding all access to files in /system
(bar exceptions)
TODOs:
- Remove loopholes once Treble violations are fixed across all internal
build targets.
Bug: 111243627
Test: m selinux_policy; build-only change
Change-Id: Ic8d71c8d139cad687ad7d7c9db7111240475f175
Add a neverallow rule requiring fine-grain ioctl filtering for most file
and socket object classes. Only chr_file and blk_file are excluded. The
goal is to ensure that any file descriptor which supports ioctl commands
uses a whitelist.
Further refine the list of file / socket objects which require ioctl
filtering. The previous ioctl filtering did not cover the following:
1) ioctls on /proc/PID files
2) ioctls on directories in /dev
3) PDX unix domain sockets
Add FIONCLEX to the list of globally safe ioctls. FIOCLEX and FIONCLEX
are alternate, uncommon ways to set the O_CLOEXEC flag on a file
descriptor, which is a harmless operation.
Test: device boots and no problems.
Change-Id: I6ba31fbe2f21935243a344d33d67238d72a8e618
Add a service in mediaswcodec to load updated codecs,
and restrict it to userdebug/eng. Reuse existing
mediaextractor_update_service since the codec update
service is identical, this avoids adding a new one
for now as we may not need the service anymore
after switching to APEX.
Bug: 111407413
Bug: 117290290
Change-Id: Ia75256f47433bd13ed819c70c1fb34ecd5d507b4
Policy w.r.t to apps:
- cgroup access from untrusted apps and priv app is neverallow'ed.
- other apps (e.g. vendor apps) need to explicitly declare appropriate
access rules to cgroups.
Policy w.r.t native domains:
- libcutils exports API to /dev/{cpuset, stune}/*. This API is used
abundantly in native vendor code. So we are not going to limit non-app
access to cgroup.
Bug: 110043362
Bug: 117666318
Test: m selinux_policy, boot device
Change-Id: I83aee21ca3e8941725c70706769ea9dbdc76b9c5
same_process_hal_file is exempted from many Treble neverallows. We want
to know which processes access this type to eventually constrain access
to it.
Bug: 37211678
Test: m selinux_policy
Change-Id: I61c0df21250eb1b1ae2d9c5fa9c801a828539813
By convention, auditallow statements are typically put into
userdebug_or_eng blocks, to ensure we don't accidentally ship
unnecessary audit rules. Let's do the same here.
Test: policy compiles.
Change-Id: Ib3eac94284eea3c1ae2f3dacddcb2eaeca95230e
Set up a new service for sw media codec services.
Bug: 111407413
Test: cts-tradefed run cts-dev --module CtsMediaTestCases --compatibility:module-arg CtsMediaTestCases:include-annotation:android.platform.test.annotations.RequiresDevice
Change-Id: Ia1c6a9ef3f0c1d84b2be8756eb1853ffa0597f8e
What changed:
- Removed cgroup access from untrusted and priv apps.
- Settings app writes to /dev/stune/foreground/tasks, so system_app domain
retains access to cgroup.
- libcutils exports API to /dev/{cpuset, stune}/*. This API seems to be used
abundantly in native code. So added a blanket allow rule for (coredomain - apps)
to access cgroups.
- For now, only audit cgroup access from vendor domains. Ultimately, we want to
either constrain vendor access to individual domains or, even better, remove
vendor access and have platform manage cgroups exclusively.
Changes from original aosp/692189 which was reverted:
- There seem to be spurious denials from vendor-specific apps. So added
back access from { appdomain -all_untrusted_apps -priv_app } to cgroup.
Audit this access with intent to write explicit per-domain rules for it.
Bug: 110043362
Test: adb shell setprop ro.config.per_app_memcg true, device correctly populates
/dev/memcg on a per app basis on a device that supports that.
Test: aosp_sailfish, wahoo boot without cgroup denials
This reverts commit cacea25ed0.
Change-Id: I05ab404f348a864e8409d811346c8a0bf49bc47a
ebc3a1a34c enabled ioctl filtering on
normal files and directories. However, no per-ioctl permissions were
enforced for symbolic links, named pipes ("mkfifo"), or
named sockets.
Start enforcing fine-grain ioctl restrictions for symbolic links, named
pipes, and named sockets.
Motivation: Prevent FS_IOC_ENABLE_VERIFY and FS_IOC_MEASURE_VERITY from
being usable on nonsensical filesystem objects and provide a layer of
defense for kernel bugs.
Test: Device boots and no obvious problem.
Change-Id: Id81b496ab64f37a0918f3dfd8fa9aaa3227009cc
Remove kernel attack surface associated with ioctls on plain files. In
particular, we want to ensure that the ioctls FS_IOC_ENABLE_VERITY and
FS_IOC_MEASURE_VERITY are not exposed outside a whitelisted set of
entities. However, it's straight forward enough to turn on ioctl
whitelisting for everything, so we choose to do so.
Test: policy compiles and device boots
Test: device boots with data wipe
Test: device boots without data wipe
Change-Id: I545ae76dddaa2193890eeb1d404db79d1ffa13c2
This reverts commit 9899568f6c.
Reason for revert: Reports of high numbers of SELinux denials
showing up on the SELinux dashboard.
Bug: 110043362
Change-Id: Id8fc260c47ffd269ac2f15ff7dab668c959e3ab0
What changed:
- Removed cgroup access from untrusted and priv apps.
- Settings app writes to /dev/stune/foreground/tasks, so system_app domain
retains access to cgroup.
- libcutils exports API to /dev/{cpuset, stune}/*. This API seems to be used
abundantly in native code. So added a blanket allow rule for (coredomain - apps)
to access cgroups.
- For now, only audit cgroup access from vendor domains. Ultimately, we want to
either constrain vendor access to individual domains or, even better, remove
vendor access and have platform manage cgroups exclusively.
Bug: 110043362
Test: adb shell setprop ro.config.per_app_memcg true, device correctly populates
/dev/memcg on a per app basis on a device that supports that.
Test: aosp_sailfish, wahoo boot without cgroup denials
Change-Id: I9e441b26792f1edb1663c660bcff422ec7a6332b
apexd is a new daemon for managing APEX packages installed
on the device. It hosts a single binder service, "apexservice".
Bug: 112455435
Test: builds, binder service can be registered,
apexes can be accessed, verified and mounted
Change-Id: I634ad100f10b2edcd9a9c0df0d33896fa5d4ed97
system_file_type is a new attribute used to identify files which exist
on the /system partition. It's useful for allow rules in init, which are
based off of a blacklist of writable files. Additionally, it's useful
for constructing neverallow rules to prevent regressions.
Additionally, add commented out tests which enforce that all files on
the /system partition have the system_file_type attribute. These tests
will be uncommented in a future change after all the device-specific
policies are cleaned up.
Test: Device boots and no obvious problems.
Change-Id: Id9bae6625f042594c8eba74ca712abb09702c1e5
/system/usr/share/zoneinfo is currently labeled zoneinfo_data_file,
a label shared with /data/misc/zoneinfo. However, each of these
directory locations has different security characteristics. In
particular, the files in /system/usr/share/zoneinfo must never be
writable, whereas /data/misc/zoneinfo may be written to by system_server.
Reusing the same label hides these different security characteristics.
Create a separate label for /system/usr/share/zoneinfo.
Test: Device boots and no obvious problems.
Change-Id: I8cf16ff038b06b38f77388e546d9b7a6865f7879
Assert that only apps and installd may open private app files.
Remove "open" permission for mediaserver/vold and remove their
neverallow exemption.
Test: verify no related audit messages in the logs.
Test: build
Fixes: 80300620
Fixes: 80418809
Bug: 80190017
Change-Id: If0c1862a273af1fedd8898f334c9b0aa6b9be728
This change limits global access to /system files down to:
/system/bin/linker*
/system/lib[64]/*
/system/etc/ld.config*
/system/etc/seccomp_policy/*
/system/etc/security/cacerts/*
/system/usr/share/zoneinfo/*
Bug: 111243627
Test: boot device, browse internet without denials to system_* types.
Test: VtsHalDrmV1_{1, 0}TargetTest without denials
Change-Id: I69894b29733979c2bc944ac80229e84de5d519f4
kernel commit 2a4c22426955d4fc04069811997b7390c0fb858e (fs: switch order
of CAP_DAC_OVERRIDE and CAP_DAC_READ_SEARCH checks) swapped the order of
dac_override and dac_read_search checks. Domains that have dac_override
will now generate spurious denials for dac_read_search unless they also
have that permission. Since dac_override is a strict superset of
dac_read_search, grant dac_read_search to all domains that already have
dac_override to get rid of the denials.
Bug: 114280985
Bug: crbug.com/877588
Test: Booted on a device running 4.14.
Change-Id: I5c1c136b775cceeb7f170e139e8d4279e73267a4
In cases when a device upgrades to system-as-root from O to P, it needs a mount
point for an already existing partition that is accessed by both system and
vendor.
Devices launching with P must not have /mnt/vendor accessible to system.
Bug: 78598545
Test: m selinx_policy
Change-Id: Ia7bcde44e2b8657a7ad9e0d9bae7a7259f40936f
Add additional compile time constraints on the ability to ptrace various
sensitive domains.
llkd: remove some domains which llkd should never ptrace, even on
debuggable builds, such as kernel threads and init.
crash_dump neverallows: Remove the ptrace neverallow checks because
it duplicates other neverallow assertions spread throughout the policy.
Test: policy compiles and device boots
Change-Id: Ia4240d1ce7143b983bb048e046bb4729d0af5a6e
Metadata needs to be erased as part of fastboot flashall -w.
Test: fastboot erase metadata
Bug: 113648914
Change-Id: I38a0debd9face16cad9d9a13a48549f3f58652fa
But in a very restricted form:
1) Nobody can initiate calls into init
2) Nobody can transfer binder objects into init, except servicemanager
Bug: 112684055
Test: device boots
Change-Id: Icfb218f2871e234284c74e096eccd7a2e786cf94
Not needed for modern Android versions. These rules are really, really
old.
Test: "adb bugreport" continues to work
Test: Generating a bugreport via key combo continues to work.
Change-Id: Ibc1157fb36abd7fc701db3819474f25210a3cb5f
llkd needs the ptrace capabilities and dac override to monitor for
live lock conditions on the stack dumps.
Test: compile
Bug: 33808187
Change-Id: Ibc1e4cc10395fa9685c4ef0ca214daf212a5e126
commit 9b2e0cbeea added a new
self:global_capability_class_set macro that covers both self:capability
and self:cap_userns. Apply the new macro to various self:capability
references that have cropped up since then.
Bug: 112307595
Test: policy diff shows new rules are all cap_userns
Change-Id: I3eb38ef07532a8e693fd549dfdbc4a6df5329609
