This change implements the following property:
Any 2**N aligned memory region on size 2**N contains no more than one DSO.
The value N can be configured, with 16 or 18 looking like a good choice.
Additionally, DSOs are loaded at random page-aligned address inside these large
regions.
This change has dual purpose:
1. Larger values of N allow a lot more compact CFI shadow implementation.
See change I14dfea630de468eb5620e7f55f92b1397ba06217.
For example, CFI shadow for the system_server process has the following size (RSS, KB):
152 for N = 12, 32 for N = 16, 16 for N = 18.
2. Extra randomization is good for security.
This change does not result in extra RAM usage, because everything is still page-aligned.
It does result in a bit more VM fragmentation because of the gaps between shared libraries.
As it turns out, this fragmentation is barely noticeable because the kernel creates new mapping
at the highest possible address, and we do enough small mappings to almost completely fill the
gaps (ex. in the Zygote the gaps are filled with .ttf file mappings and thread stacks).
I've measured VM fragmentation as the sum of all VM gaps (unmapped regions) that are larger
than 1MB according to /proc/$PID/maps. On aosp_angler-userdebug, the numbers are (in GB):
| N = 12 | N = 18
system_server | 521.9 | 521.1
zygote64 | 522.1 | 521.3
zygote32 | 2.55 | 2.55
mediaserver | 4.00 | 4.00
Change-Id: Ia6df840dd409c82837efd1f263be420d9723c84a
Before, dynamic executables would initialize the global stack protector
twice, once for the linker, and once for the executable. This worked
because the result was the same for both initializations, because it
used getauxval(AT_RANDOM), which won't be the case once arc4random gets
used for it.
Bug: http://b/29622562
Change-Id: I7718b1ba8ee8fac7127ab2360cb1088e510fef5c
Test: ran the stack protector tests on angler (32/64bit, static/dynamic)
And clang won't let you have a function declaration where some arguments
have nullability specifiers and others don't.
Change-Id: I450b0221a3f7f068d5fe971dfbc0ba91d25710e8
Previously, arc4random would register a fork-detecting pthread_atfork
handler to not have to call getpid() after a fork. pthread_atfork uses
pthread_mutex_lock, which requires the current thread to be initialized,
preventing the use of arc4random for initializing the global stack guard,
which needs to happen before the main thread has been initialized.
Extract the arc4random fork-detection flag and use the existing
arc4random fork handler to set it.
Bug: http://b/29622562
Change-Id: I98c9329fa0e489c3f78cad52747eaaf2f5226b80
This function only exists for backwards compatibility, so leave it as it was.
Bug: http://b/26944282
Change-Id: I31973d1402660933103ee2d815649ab9569e4dfc
This patch uses __kernel_vsyscall instead of "int 0x80"
as the syscall entry point. AT_SYSINFO points to
an adapter to mask the arch specific difference and gives a
performance boost on i386 architecture.
Change-ID: Ib340c604d02c6c25714a95793737e3cfdc3fc5d7
Signed-off-by: Mingwei Shi <mingwei.shi@intel.com>
Our FORTIFY _chk functions' implementations were very repetitive and verbose
but not very helpful. We'd also screwed up and put the SSIZE_MAX checks where
they would never fire unless you actually had a buffer as large as half your
address space, which probably doesn't happen very often.
Factor out the duplication and take the opportunity to actually show details
like how big the overrun buffer was, or by how much it was overrun.
Also remove the obsolete FORTIFY event logging.
Also remove the unused __libc_fatal_no_abort.
This change doesn't improve the diagnostics from the optimized assembler
implementations.
Change-Id: I176a90701395404d50975b547a00bd2c654e1252
Add backtrace_string to convert a malloc_debug backtrace to a string.
Also move the backtrace functions to libc_malloc_debug_backtrace so that
libmemunreachable can reuse them.
Change-Id: I5ad67001c0b4d184903c762863a8588181d4873b
The major components of the rewrite:
- Completely remove the qemu shared library code. Nobody was using it
and it appears to have broken at some point.
- Adds the ability to enable/disable different options independently.
- Adds a new option that can enable the backtrace on alloc/free when
a process gets a specific signal.
- Adds a new way to enable malloc debug. If a special property is
set, and the process has an environment variable set, then debug
malloc will be enabled. This allows something that might be
a derivative of app_process to be started with an environment variable
being enabled.
- get_malloc_leak_info() used to return one element for each pointer that
had the exact same backtrace. The new version returns information for
every one of the pointers with same backtrace. It turns out ddms already
automatically coalesces these, so the old method simply hid the fact
that there where multiple pointers with the same amount of backtrace.
- Moved all of the malloc debug specific code into the library.
Nothing related to the malloc debug data structures remains in libc.
- Removed the calls to the debug malloc cleanup routine. Instead, I
added an atexit call with the debug malloc cleanup routine. This gets
around most problems related to the timing of doing the cleanup.
The new properties and environment variables:
libc.debug.malloc.options
Set by option name (such as "backtrace"). Setting this to a bad value
will cause a usage statement to be printed to the log.
libc.debug.malloc.program
Same as before. If this is set, then only the program named will
be launched with malloc debug enabled. This is not a complete match,
but if any part of the property is in the program name, malloc debug is
enabled.
libc.debug.malloc.env_enabled
If set, then malloc debug is only enabled if the running process has the
environment variable LIBC_DEBUG_MALLOC_ENABLE set.
Bug: 19145921
Change-Id: I7b0e58cc85cc6d4118173fe1f8627a391b64c0d7
Exactly which functions get a stack protector is up to the compiler, so
let's separate the code that sets up the environment stack protection
requires and explicitly build it with -fno-stack-protector.
Bug: http://b/26276517
Change-Id: I8719e23ead1f1e81715c32c1335da868f68369b5
Correct the comment, and remove the unused functionality. getauxval(3) does
now set errno to let you know it failed to find anything, but since none of
this function's callers care anyway it seems safer to leave errno untouched
until we actually have a demonstrated need for it.
Bug: https://code.google.com/p/android/issues/detail?id=198111
Change-Id: I232a42dc5a02c8faab94c7d69bef610408276c23
The BIONIC_ROUND_UP_POWER_OF_2 macro did not have parentheses around
the whole expression. This lead to the wrong value being computed when
used as part of a mathematical expression such as this:
value = BIONIC_ROUND_UP_POWER_OF_2(value) - 1;
This only happens on 64 bit abis.
Change-Id: I6f8afbdaf16fe64a88fa0246d074b3534c9159c1
It actually means "crash immediately". Well, it's an error. And callers are
much more likely to realize their mistake if we crash immediately rather
than return EINVAL. Historically, glibc has crashed and bionic -- before
the recent changes -- returned EINVAL, so this is a behavior change.
Change-Id: I0c2373a6703b20b8a97aacc1e66368a5885e8c51
In order to run tsan unit tests, we need to support pthread spin APIs.
Bug: 18623621
Bug: 25392375
Change-Id: Icbb4a74e72e467824b3715982a01600031868e29
It removes calling to pthread_mutex_lock() at the beginning of new
thread, which helps to support thread sanitizer.
Change-Id: Ia3601c476de7976a9177b792bd74bb200cee0e13
For the __release and __release_rt functions, the previous macros
would add a dwarf cfi entry for the function with no values. This works
with libunwind since it always tries the arm unwind information first.
This change removes those entries by creating a no dwarf version of the
assembler macro.
Change-Id: Ib93e42fff5a79b8d770eab0071fdee7d2afa988d
Read /proc/stat to count online cpus is not correct for all android
kernels. Change to reading /sys/devices/system/cpu/online instead.
Bug: 24376925
Change-Id: I3785a6c7aa15a467022a9a261b457194d688fb38
I'm removing the TODO on the assumption that being compatible with glibc
is more useful than BSD. The new internal "bionic_page.h" header factors
out some duplication between libc and the linker.
Bug: http://b/22735893
Change-Id: I4aec4dcba5886fb6f6b9290a8f85660643261321
According to the comments in Posix_close(), TEMP_FAILURE_RETRY() should
not be used with close():
462bdac45c%5E%21/#F12
Kill ScopedFd by simplifying the single caller.
Change-Id: I248c40b8c2fc95f1938a6edfc245c81847fc44af
Signed-off-by: Spencer Low <CompareAndSwap@gmail.com>
Previous implementation of rwlock contains four atomic variables, which
is hard to maintain and change. So I make following changes in this CL:
1. Add pending flags in rwlock.state, so we don't need to synchronize
between different atomic variables. Using compare_and_swap operations
on rwlock.state is enough for all state change.
2. Add pending_lock to protect readers/writers waiting and wake up
operations. As waiting/wakeup is not performance critical, using a
lock is easier to maintain.
3. Add writer preference option.
4. Add unit tests for rwlock.
Bug: 19109156
Change-Id: Idcaa58d695ea401d64445610b465ac5cff23ec7c
Found by the toybox id(1) which calls both getpwuid(3) and getgrgid(3) before
looking at either result. The use of a shared buffer in this code meant that
even on a single thread, the data for any of the passwd functions would be
clobbered by the data for any of the group functions (or vice versa).
This might seem like an insufficient fix, but POSIX explicitly says (for
getpwnam) that the result "might be overwritten by a subsequent call to
getpwent(), getpwnam(), or getpwuid()" and likewise for other members of
that group, plus equivalent text for the group-related functions.
Change-Id: I2272f47e91f72e043fdaf7c169fa9f6978ff4370
POSIX specifies that pthread_kill(3) and pthread_sigmask(3) are
supposed to live in signal.h rather than pthread.h.
Since signal.h now needs pthread_t and pthread_attr_t, I've moved
those defintions into include/machine/pthread_types.h to keep the
namespace clean. I also sorted some includes. The combination of these
two things seems to have exploded into a cascade of missing includes,
so this patch also cleans up all those.
Change-Id: Icfa92a39432fe83f542a797e5a113289d7e4ad0c
Make the definition of DISALLOW_COPY_AND_ASSIGN conditional. This is
so that the projects that include libnativehelper and bionic macros
do not have to be careful in which order those projects are included.
Bug: 18334516
Change-Id: Ib12a2c2b7ad2e360edcf3b26cb1be626540fadc1
Enable the -std=gnu++11 flag for libstdc++ static and
dynamic libs.
ScopeGuard uses DISABLE_ macros instead of '= delete';
Change-Id: I07e21b306f95fffd49345f7fa136cfdac61e0225
This patch fixes the problem with symbol search order
for dlsym(RTLD_DEFAULT/RTLD_NEXT, .) by loading libraries
and ld_preloads in correct order.
Bug: https://code.google.com/p/android/issues/detail?id=74255
Change-Id: If1ba5c807322409faa914e27ecb675e2c4541f0d
Attempt: 2
This patch fixes the problem with symbol search order
for dlsym(RTLD_DEFAULT/RTLD_NEXT, .) by loading libraries
and ld_preloads in correct order.
Bug: https://code.google.com/p/android/issues/detail?id=74255
Change-Id: I4cf84c70dbaabe99310230dfda12385ae5401859
On most architectures the kernel subtracts a random offset to the stack
pointer in create_elf_tables by calling arch_align_stack before writing
the auxval table and so on. On all but x86 this doesn't cause a problem
because the random offset is less than a page, but on x86 it's up to two
pages. This means that our old technique of rounding the stack pointer
doesn't work. (Our old implementation of that technique was wrong too.)
It's also incorrect to assume that the main thread's stack base and size
are constant. Likewise to assume that the main thread has a guard page.
The main thread is not like other threads.
This patch switches to reading /proc/self/maps (and checking RLIMIT_STACK)
whenever we're asked.
Bug: 17111575
Signed-off-by: Fengwei Yin <fengwei.yin@intel.com>
Change-Id: I1d4dbffe7bc7bda1d353c3a295dbf68d29f63158
Removes the leading underscores from __android_set_abort_message() and
moves its declaration into a public header file.
Bug: 17059126
Change-Id: I470c79db47ec783ea7a54b800f8b78ecbe7479ab
bionic_systrace.h contains an implementation of tracing that
can be used with systrace.py and its associated viewer. pthread_mutex
now uses this tracing to track pthread_mutex contention, which can be
enabled by using the "bionic" command line option to systrace.
Bug: 15116468
Change-Id: I30ed5b377c91ca4c36568a0e647ddf95d4e4a61a
The getentropy_linux.c is lightly modified to build on Android, but we're now
completely in sync with upstream OpenBSD's arc4random implementation.
Change-Id: If32229fc28aba908035fb38703190d41ddcabc95
Also remove __bionic_name_mem which has exactly one caller, and is only
ever expected to be used in this one place.
Change-Id: I833744f91e887639f5b2d1269f966ee9032af207
There were two bugs here:
- For 64 bit values, this did not properly round up.
- The macro rounded to the power of 2 less than value, not to the power
of 2 greater than value.
Change-Id: If8cb41536a9d2f5c1bc213676f1e67a7903a36b0
It's okay for a program to choose to drag in stdio, but it's unfortunate
if even the minimal "int main() { return 42; }" drags in stdio...
This brings the minimal static binary on ARM down from 78KiB to 46KiB.
Given that we don't have a separate -lpthread it's not obvious to me that
we can shave this down any further. I'm not sure whether this is a worthwhile
change for that reason. (And the fact that dynamic binaries, the usual case,
are unaffected either way.)
Change-Id: I02f91dcff37d14354314a30b72fed2563f431c88
This is actually revision 1.33, which is no longer the latest, but it's
as close to head as we can currently reasonably get. I've also switched
to the OpenBSD getentropy_linux.c implementation of getentropy, lightly
modified to try to report an error on failure.
Bug: 14499627
Change-Id: Ia7c561184b1f366c9bf66f248aa60f0d53535fcb
This allows an easier way to share config parameters between unit tests
and the bionic code.
It also fixes a problem where the 32 bit bionic tests based on glibc, or
the cts list executable did not have the pvalloc,valloc tests.
Change-Id: Ib47942cb8a278252faa7498a6ef23e9578db544f
mbrtoc32 and c32rtomb get their implementations from mbrtowc and wcrtomb. The
wc functions now simply call the c32 functions.
Bug: 14646575
Change-Id: I49d4b95fed0f9d790260c996c4d0f8bfd1686324
The problem with the original patch was that using syscall(3) means that
errno can be set, but pthread_create(3) was abusing the TLS errno slot as
a pthread_mutex_t for the thread startup handshake.
There was also a mistake in the check for syscall failures --- it should
have checked against -1 instead of 0 (not just because that's the default
idiom, but also here because futex(2) can legitimately return values > 0).
This patch stops abusing the TLS errno slot and adds a pthread_mutex_t to
pthread_internal_t instead. (Note that for LP64 sizeof(pthread_mutex_t) >
sizeof(uintptr_t), so we could potentially clobber other TLS slots too.)
I've also rewritten the LP32 compatibility stubs to directly reuse the
code from the .h file.
This reverts commit 75c55ff84e.
Bug: 15195455
Change-Id: I6ffb13e5cf6a35d8f59f692d94192aae9ab4593d
This reverts commit ced906c849.
Causes issues on art / dalvik due to a broken return value
check and other undiagnosed issues.
bug: 15195455
Change-Id: I5d6bbb389ecefb0e33a5237421a9d56d32a9317c
Also hide part of the system properties compatibility code, since
we needed to touch that to keep it building.
I'll remove __futex_syscall4 and futex in a later patch.
Bug: 11156955
Change-Id: Ibbf42414c5bb07fb9f1c4a169922844778e4eeae
To use jemalloc, add MALLOC_IMPL = jemalloc in a board config file
and you get the new version automatically.
Update the pthread_create_key tests since jemalloc uses a few keys.
Add a new test to verify memalign works as expected.
Bug: 981363
Change-Id: I16eb152b291a95bd2499e90492fc6b4bd7053836
If libnetd_client can't be found, operate as before and use the default netId
potentially overriden by a more specific netId passed in to
android_get*fornet().
(cherry picked from commit 559c7842cc)
Change-Id: I42ef3293172651870fb46d2de22464c4f03e8e0b
+ Name the dispatch header correctly (NetdClientDispatch.h).
+ Hide the global dispatch variable (__netdClientDispatch).
+ Explain why it's okay to read the variable without locking.
+ Use quotes instead of angle-brackets for non-system includes.
+ Add necessary declarations for C compiles (and not just C++).
Change-Id: Id0932165e71d81da5fce77a684f40c2263f58e61
The library exists outside bionic. It is dynamically loaded, to replace selected
standard socket syscalls with versions that talk to netd.
Change connect() to use the library if available.
(cherry picked from commit 3a6b627a14df8111b03e452f2df4b5f4938e0e49)
Change-Id: Ib6198e19dbc306521a26fcecfdf6e8424d163fc9
This more general interface lets liblog give us any fatal log message,
regardless of source. This means we can remove the special case for
LOG_ALWAYS_FATAL with a simpler scheme that automatically works for
the VM too.
Change-Id: Ia6dbf7c3dbabf223081bd5159294835d954bb067
