* changes:
fix: boot fails on ASAN builds with VNDK restriction
templatize ld.config.txt
Make default namespace for system processes isolated
vndk: add vendor/lib(64)/vndk for vendor namespace
Implement system namespace for vendor process
Use the legacy linker namespace configs when target is sanitized
Permit /system/lib/vndk-sp/hw/ for 'sphal' namespace
Add system/${LIB}/vndk to default namespace for vendor.
Add [test] section for /data/[nativetest|benchmarktest]
/data/nativetest[64] processes are running with the new ns config
Fix the bug that caused boot failure on ASAN builds with VNDK
restriciton. The major cause is because incorrect (old) ld.config.txt
was used when the build is sanitized, which prevented the dynamic linker
to find some VNDK libs that only exist in /system/lib/vndk; the old
ld.config.txt does not have the directory in its search paths. So, this
CL fixes the problem by having the same ld.config.txt for both sanitized
and non-sanitizied builds.
Furthermore, ld.config.txt is modified so that dependency to
libclang_rt* libs are redirected to those in /system/lib directory. This
ensures that the sanitizer runtime libs are not dual loaded but are
provided for both platform and vendors.
Bug: 65217017
Test: SANITIZE_TARGET=integer_overflow SANITIZE_TARGET_DIAG=integer_overflow m
on 2017 pixel devices. The build is successful and the device boots to
the UI.
Merged-In: I0e21e20d9aca340b984968e07d4ce542ae10fd31
Change-Id: I0e21e20d9aca340b984968e07d4ce542ae10fd31
(cherry picked from commit faefa6bd36)
Lists of libraries in between the linker namespaces are no longer
hard-coded in ld.config.txt, but instead come from Soong.
Bug: 37139976
Test: build 2017 pixel device with BOARD_VNDK_VERSION=current m -j
Test: the device is bootable, basic functionalities (camera, camcorder,
wifi, bt, gps, etc.) work.
Merged-In: I8170e6c3f6ee04b16359791d64cc46bd2714a073
Change-Id: I8170e6c3f6ee04b16359791d64cc46bd2714a073
(cherry picked from commit 367984602a)
The default namespace for system process now becomes isolated, which
means it can load only the libs that are in search.paths and under
permitted.paths.
/system/framework, /system/app, /system/priv-app, /vendor/app, /data,
etc are added to the permitted paths since libart should be able to
dlopen odex files under the locations.
Following directories become unavailable:
/system/lib/vndk
/system/lib/vndk-sp
Bug: 37013858
Bug: 64888291
Bug: 64950640
Test: 2017 pixel devices builds and boots
Test: android.jni.cts.JniStaticTest#test_linker_namespaces passes
Merged-In: I2bbe9cc19940c3633c2fb901f9bf8ab813e38c13
Change-Id: I2bbe9cc19940c3633c2fb901f9bf8ab813e38c13
(cherry picked from commit 55f05d79f9)
When the vndk is extended by vendor, the libs in vendor/lib(64)/vndk
will override system/lib(64)/vndk libs. Vendor binaries will search
vndk libs in vendor/lib(64)/vndk prior to system/lib(64)/vndk.
Also, remove a workaround for libsdm-color.so
Bug: 65032513
Test: build and boot check for libsdm-color.so will be loaded on boot
Test: currently no libs exist on vendor/lib(64)/vndk. No test for it
Merged-In: I99ed37eb1a9e92c83533e2d92c9c06db11f01e97
Change-Id: I99ed37eb1a9e92c83533e2d92c9c06db11f01e97
(cherry picked from commit 3094de9684)
For vendor process default namespace searches as following order:
1. /vendor/lib/(hw|egl), /vendor/lib: Vendor libs + VNDK-vnd-ext
2. /system/lib/vndk-$(ver): VNDK libs
3. /vendor/lib/vndk-sp-$(ver): VNDK-SP-vnd-ext
4. /system/lib/vndk-sp-$(ver): VNDK-SP
and searches system namespace (/system/lib) only for LL-NDK libs.
This configuarion is used only with BOARD_VNDK_VERSION is defined.
Bug: 37192038
Test: build with 'BOARD_VNDK_VERSION=current' and boot
Merged-In: If9778b9266a084846ba8fe73e6bff25828359d33
Change-Id: If9778b9266a084846ba8fe73e6bff25828359d33
(cherry picked from commit 24c29f1be4)
The new linker namespace config causes problem when the target is
sanitized: vendor libs which are loaded in the sphal namespace can't
link against to libclang_rt* libs which are in /system/lib directory
because the directory is not allowed for sphal namsapce.
Long-term solution would be installing libclang_rt* to both /system/lib
and /vendor/lib so that vendor libs can link against to the one in
/vendor/lib.
Until the work is done, let's just disable the new linker namespace
config when the target is to be sanitized.
Bug: 63535130
Test: make SANITIZE_TARGET=integer_overflow
SANITIZE_TARGET_DIAG=integer_overflow builds and boots to the UI
Merged-In: I6afb69885aaa3d25e554b9ca699a572248bfc50a
Change-Id: I6afb69885aaa3d25e554b9ca699a572248bfc50a
(cherry picked from commit fff6f75fc1)
android.hidl.memory@1.0-impl.so is a SP-HAL located in
/system/lib/vndk-sp/hw. This can't be moved to /vendor/lib since it is a
framework HAL.
Bug: 62930720
Test: 2017 pixel builds and boots with BOARD_VNDK_VERSION on
Merged-In: I9c456983ef68120c5e8c629efc6dd66a26220ecb
Change-Id: I9c456983ef68120c5e8c629efc6dd66a26220ecb
(cherry picked from commit 7d4cf3fd33)
With BOARD_VNDK_VERSION=current, vndk libs will be installed in
system/${LIB}/vndk. To make them available for vendor, it must be
added to default namespace.
Bug: 63866913
Test: build and boot with BOARD_VNDK_VERSION=current
Merged-In: I9e467a6125fc89513754b56a2420975559144f98
Change-Id: I9e467a6125fc89513754b56a2420975559144f98
(cherry picked from commit 90a2487776)
We have both system and vendor tests in the same directory
/data/nativetest. Since we can't distinguish system and vendor tests at
runtime, we choose to run all of them with the legacy namespace
configuration where /system/lib and /vendor/lib are both accessible
without any restriction. Furthermore, /system/lib/vndk-sp is added to
the list of accessible directories for tests since some libs (such as
libion.so) exist only in the directory but not in /system/lib.
This will change in the future when we install system and vendor tests
into different directories (vendor tests goes into ./vendor subdir).
Then, we will run the tests with [system] or [vendor] configurations
depending on their paths.
Bug: 63597267
Test: run bionic-unit-tests
Test: run linker-unit-tests
Merged-In: I810003b2da0b30335200c130f484ea7b041f9f80
Change-Id: I810003b2da0b30335200c130f484ea7b041f9f80
(cherry picked from commit c034a43d44)
Native tests under /data/nativetest[64] directory is now running with
the same linker namespace configuration as /system/bin/* processes.
This allows us to stop mimicing the linker namespace configuration of the
/system/bin/* processes using LD_LIBRARY_PATH.
Bug: 63597267
Test: run bionic-unit-tests
Test: run linker-unit-tests
Merged-In: If8e2ed0b8016e4e07bf6829735b8e02f952042d0
Change-Id: If8e2ed0b8016e4e07bf6829735b8e02f952042d0
(cherry picked from commit 9d1e50b809)
This probably doesn't have any actual effect, because there should only
be one smartsocket listener so this will happen to work, but this looks
sketchy.
Test: mma
Test: adb server nodaemon & (sleep 1; adb kill-server)
Change-Id: Id00ab1fc1fc029949c675ae6e3b1a77f3e9e4c69
Add a function to iterate over Regs with a human-friendly name and the
value of the register, to make it easy to string-format registers.
Test: mma
Change-Id: I8f33830fa79b245b5030b7bc9bc3c1c7ced80c75
- Move the load bias stored out of ElfInterface into Elf. For the compressed
sections, the load bias was not the same as the data from the uncompressed
section.
- Move the initialization of the compressed section into Init. It was too easy
to forget to call the init of the compressed section.
- Do not automatically add in load bias to the pc before calling ElfInterface
code. Do all of the pc manipulations in the Elf object.
- Change the interface GetFunctionName code to pass in the load_bias instead
of modifying the pc inside the code.
- Modify the Step function to pass in the elf offset, not add it to the pc.
It is necessary to have two different relative values when executing the
Step: a pc that is relative to the beginning of the elf for the reading data
the actual instructions when trying to determine if this is in a signal
frame, and a pc that is relative to the map for finding the appropriate
unwind information.
- Add a feature to Unwinder so that an unwind can be stopped if it ends up
in map that has a specified suffix. This is so that the ART unwinding
code doesn't require skipping the compressed section. Instead, stop at
if trying to unwind through a known suffix code that means the code is
in java code. This is important because the compressed section data is
not only used by the jave compiled code, so that will continue to work.
- Fix tests for restructuring, add new tests for new functionality.
Test: Ran art test 137-cfi using new unwinder as default.
Test: Ran new unit tests.
Change-Id: I42e658c64c5e14f698ba34944a3043afac967884
This associates every service with a list of HIDL services
it provides. If these are disabled, hwservicemanager will
request for the service to startup.
Bug: 64678982
Test: manual with the light service
Change-Id: Ibf8a6f1cd38312c91c798b74574fa792f23c2df4
Children of init that use any of the SELinux wrapper functions,
including make_dir(), mkdir_recursive(), and plenty others, need to
first initialize the sehandle with SelabelInitialize().
I wish there were a better solution, but early init doesn't actually
want this handle initialized, so that is a valid use case. Ueventd
needs to initialize this before fork()'ing, so lazy initialization is
not universally acceptable either. Likely we won't have other
children that fork() then exec() init again, so this should be okay.
Bug: 62875318
Test: init unit tests
Test: sailfish creates directories with correct SELabel after wipe
Change-Id: I6de937604a060e18945427418f15b90e0b9d5c37
subcontext_test had been failing due to setexeccon() failing to
transition to vendor_init context. This is a good thing as nothing
other than init should be able to transition into this context.
I don't want to add code to skip the setexeccon() call only for the
tests, so I instead call setexeccon() with the return value of
getcon(). This works however only for root, so these tests are
skipped for non-root.
Test: init unit tests
Change-Id: I8a415599e0ec5506511202f7f5018c0e5265837d
Also move the Backtrace::CreateNew function into the same place as ::Create
in preparation for the switch to the new unwinder as the base.
Test: Verify that names are demangled properly.
Change-Id: I6274cd43ea52210523e1e1ce23af5b22f62f4573
wait_for_prop sets a flag that prevents the action queue from
continuing while otherwise allowing init's main loop to continue
executing. This cannot be done from a subcontext, so it's moved to
normal init.
All property functions need work in any case, particularly once
property_service is moved out of init.
Bug: 62875318
Test: boot sailfish and see that the previous failure related to this
is fixed
Change-Id: Ib9e0d0bdbd0ff22ab0e5c3fe6db620700af266c6
Jiyong Park