* Allow sanitization of libc (excluding existing global sanitizers)
and disallow sanitization of linker. The latter has not been
necessary before because HWASan is the first sanitizer to support
static binaries (with the exception of CFI, which is not used
globally).
* Static binary startup: initialize HWASan shadow very early so that
almost entire libc can be sanitized. The rest of initialization is
done in a global constructor; until that is done sanitized code can
run but can't report errors (will simply crash with SIGTRAP).
* Switch malloc_common from je_* to __sanitizer_*.
* Call hwasan functions when entering and leaving threads. We can not
intercept pthread_create when libc depends on libclang_rt.hwasan.
An alternative to this would be a callback interface like requested
here:
https://sourceware.org/glibc/wiki/ThreadPropertiesAPI
All of the above is behind a compile-time check
__has_feature(hwaddress_sanitizer). This means that HWASan actually
requires libc to be instrumented, and would not work otherwise. It's
an implementation choice that greatly reduces complexity of the tool.
Instrumented libc also guarantees that hwasan is present and
initialized in every process, which allows piecemeal sanitization
(i.e. library w/o main executable, or even individual static
libraries), unlike ASan.
Change-Id: If44c46b79b15049d1745ba46ec910ae4f355d19c
libc had some -Wimplicit-fallthrough warnings. They all seem to be
benign. We're trying to enable this flag globally, so we need to
annotate these breaks here.
Bug: 112564944
Test: Builds
Change-Id: I5afae694cc4cf26ad1a61e2c8ae91f00cda7c733
This change avoids having to run the genseccomp.py script every time a
policy file is edited, and instead generates these files at
compile-time.
Bug: None
Test: m
Test: find out/soong/ -name x86_64_global_policy.cpp # Shows files
Test: generated policies are equivalent to original policies
Change-Id: I12461fe0c5fb02c008c1b2503fbb994b8aa2f56b
Add two functions to allow objects that own a file descriptor to
enforce that only they can close their file descriptor.
Use them in FILE* and DIR*.
Bug: http://b/110100358
Test: bionic_unit_tests
Test: aosp/master boots without errors
Test: treehugger
Change-Id: Iecd6e8b26c62217271e0822dc3d2d7888b091a45
pclose(3) is now an alias for fclose(3). We could add a FORTIFY check
that you use pclose(3) if and only if you used popen(3), but there seems
little value to that when we can just do the right thing.
This patch also adds the missing locking to _fwalk --- we need to lock
both the global list of FILE*s and also each FILE* we touch. POSIX says
that "The popen() function shall ensure that any streams from previous
popen() calls that remain open in the parent process are closed in the
new child process", which we implement via _fwalk(fclose) in the child,
but we might want to just make *all* popen(3) file descriptors O_CLOEXEC
in all cases.
Ignore fewer errors in popen(3) failure cases.
Improve popen(3) test coverage.
Bug: http://b/72470344
Test: ran tests
Change-Id: Ic937594bf28ec88b375f7e5825b9c05f500af438
__libc_sysinfo is hidden, so accessing it doesn't require a relocated GOT.
It is important not to have a relocatable initializer on __libc_sysinfo,
because if it did have one, and if we initialized it before relocating the
linker, then on 32-bit x86 (which uses REL rather than RELA), the
relocation step would calculate the wrong addend and overwrite
__libc_sysinfo with garbage.
Asides:
* It'd be simpler to keep the __libc_sysinfo initializer for static
executables, but the loader pulls in libc_init_static (even though it
uses almost none of the code in that file, like __libc_init).
* The loader has called __libc_init_sysinfo three times by the time it
has relocated itself. A static executable calls it twice, while libc.so
calls it only once.
Bug: none
Test: lunch aosp_x86-userdebug ; emulator
Test: adb shell /data/nativetest/bionic-unit-tests/bionic-unit-tests
Test: adb shell /data/nativetest/bionic-unit-tests-static/bionic-unit-tests-static
Change-Id: I5944f57847db7191608f4f83dde22b49e279e6cb
Global flag --pack-dyn-relocs=android is used with clang lld.
For b/24465209, we need to override that with --pack-dyn-relocs=none.
Bug: 80093890
Bug: 24465209
Test: build with USE_CLANG_LLD=true and run dlext.compat_elf_hash_and_relocation_tables
Change-Id: Ic3837446ff0deccf4e66425b39d3b062abcacc5e
See longer explanation in b/80093890.
Clang lld does not generate expected DT_REL and DT_RELA tags
with --hash-style=both and --pack-dyn-relocs=android.
I am not sure about the extent of b/24465209, so
I would rather not to use lld for these .so files for now.
Bug: 80093890
Bug: 24465209
Test: build with USE_CLANG_LLD=true and run dlext.compat_elf_hash_and_relocation_tables
Change-Id: I94e9fe8d687daeadb0182ae26596ca11d3f8bd9b
Libraries that are direct or indirect dependencies of modules installed
to recovery partition (e.g. toybox) are marked as recovery_available:
true. This allows a recovery variant of the lib is created when it is
depended by other recovery or recovery_available modules.
Bug: 67916654
Bug: 64960723
Bug: 63673171
Bug: 29921292
Test: m -j
Change-Id: I59bf859a10a218af6591025a0fe7d1853e328405
Bug: 78133793
Bug: 78242072
Test: Builds and ran to verify best libc routines.
Change-Id: I2299b8f0f6379c9003072764978ea917b74aee4f
Merged-In: I2299b8f0f6379c9003072764978ea917b74aee4f
(cherry picked from commit 3162eb3b81)
This reverts commit 2e7145c048.
When src is at the end page, the sse2 strlcpy SSE2 optimized version
can issue a movdqu instruction that can cross the page boundary. If
the next page is not allocated to that process, it leads to
segmentation fault. This is a rare but has be caught multiple times
during robustness testing.
We isolated a way to reproduce that issue outside of an Android device
and we have been able to resolve this particular case. However, we
ran some additional compliance and robustness tests and found several
other similar page crossing issues with this implementation.
In conclusion, this optimization needs to be re-written from scratch
because its design is at cause. In the meantime, it is better to
remove it.
Change-Id: If90450de430ba9b7cd9282a422783beabd701f3d
Signed-off-by: Jeremy Compostella <jeremy.compostella@intel.com>
We saw crashes from pthread_exit+debuggerd on LP32
(https://issuetracker.google.com/72291624), and it seems like the
equivalent problem should exist with system(3). I fixed posix_spawn(3)
as part of that bug, so the easiest fix is probably to reuse that.
Bug: http://b/72470344
Test: ran tests
Change-Id: I05f838706f2b4a14ac3ee21292833e6c8579b0d4
Otherwise clang inlines it into pthread_sigmask(3), which breaks libsigchain.
Bug: http://b/73344857
Test: ran tests, plus the app this broke
Change-Id: Ie4a1dc8f9c6ba58d1a2fa69aeff961c70b74767d
Caused problems for some Google source switching to unified headers
because the legacy headers just used #define instead.
Bug: http://b/73210428
Test: builds
Change-Id: If8a36f284031e57af419903610029a8c34a23d91
Add the capability to read /etc/passwd and /etc/group for getpw* and
getgr* functions.
Bug: 27999086
Test: pwd, grp, grp_pwd_file unit tests
Test: Read in custom users/groups from /etc/{passwd,group}
Change-Id: Idc1f054af8a7ca34743a90493495f0ccc775a0d8
The error handling was a mess, resulting in an infinite loop
(and a test timeout) if you actually took the /dev/urandom
fallback. I'm also unconvinced that the getrandom(2) path was correct
because of the various reasons why we might get a short count back
(http://man7.org/linux/man-pages/man2/getrandom.2.html). In this version,
the two paths look very similar except for the getrandom(2) failure case,
where it's now much clearer that we just give up so that we can boot
(http://b/33059407).
Bug: http://b/33059407
Bug: http://b/67015565
Test: ran tests on hikey and a ICS Galaxy Nexus
Change-Id: Ie715d59b46f7a70bae66943c316168838787b40d
This doesn't address `struct sigaction` and `sigaction`. That will
come later.
Bug: http://b/72493232
Test: ran tests
Change-Id: I4134346757ce3a4dac6feae413361cec16223386
Makes no difference to the benchmarks, but does make the code a bit
more readable.
Bug: http://b/68672236
Test: ran tests, benchmarks
Change-Id: I63fa5f78d077c86e4f4f194f2c76ab5510c29109
This is a squash revert of a4a4854 and 1b0f2b4
Revert "Remove a test for backwards compatibility we no longer support."
This reverts commit a4a485454a.
# This is the commit message #2:
Revert "Remove obsolete workaround."
This reverts commit 1b0f2b49d5.
Bug: 72143978
Bug: 24465209
Test: Tested failing case on sailfish, reverted back all CL's since
Test: 3471433 for b/24465209 and apps open
This is also slightly faster for the no VDSO case (56ns vs 66ns).
Bug: N/A
Test: ran tests, benchmarks
Change-Id: I2b0edd06ee6942eb57c32678279278a53ca5ee9b
Introduce new flag to mark soinfo as TLS_NODELETE when
there are thread_local dtors associated with dso_handle
belonging to it.
Test: bionic-unit-tests --gtest_filter=dl*
Test: bionic-unit-tests-glibc --gtest_filter=dl*
Bug: https://github.com/android-ndk/ndk/issues/360
Change-Id: I724ef89fc899788f95c47e6372c38b3313f18fed
Toolbox is moving to Android.bp, and its genrule needs access to these
kernel headers.
Test: mmma system/core/toolbox
Change-Id: I213a36e1f4bb5d4e3e3e105ed7ab01ec945014a2
There are no meaningful changes here, just a minimal conversion to two
C++ templates to make further changes easier.
Bug: N/A
Test: ran tests, benchmarks
Change-Id: I958fbf17a85f19dd8f17bfb4bbb9314d220daa3b
Reinitializing system properties can result in crashes later in the
program, and is generally not recommended or even supported. This
change moves the actual logic for system properties into a class that
can be tested in isolation, without reinitializing the actual system
property area used in libc.
Bug: 62197783
Test: boot devices, ensure properties work
Test: system property unit tests and benchmarks
Change-Id: I9ae6e1b56c62f51a4d3fdb5b62b8926cef545649
This adds support for reading a serialized
/dev/__properties__/property_info file, which contains a
serialized trie that maps property names to the SELinux context to
which they belong.
Performance wise on walleye, this change reduces the start up cost in
libc from ~3000us to ~430us. On a benchmark that calls
__system_property_find() for each property set on the system, it
reduces the time per iteration from ~650us to ~292us.
Bug: 36001741
Test: Boot bullhead, walleye, run unit tests
Test: Benchmark initialization and lookup performance
Change-Id: I0887a3a7da88eb51b6d1bd494fa5bce593423599
