Go back to intercepting kernel32!RaiseException, and only go for
ntdll!RtlRaiseException if that fails. Fixes throw_and_catch.cc test.
Work around an issue in LLVM's win64 epilogues. We end up with an
epilogue that looks like this, and it drives the Win64 unwinder crazy
until stack overflow:
call ill_cc!__asan_handle_no_return
xor eax,eax
add rsp,40h // epilogue starts
pop rbp // CSR
ud2 // Trap here
ret // Ret?
nop word ptr [rax+rax]
sub rsp,28h // Next function
Will file a PR soon.
llvm-svn: 277874
MSVC doesn't have an exact equivalent for __builtin_frame_address, but
_AddressOfReturnAddress() + sizeof(void*) should be equivalent for all
frames build with -fno-omit-frame-pointer.
llvm-svn: 277826
These are meant to only be included on certain targets. This only disables it
for Windows ARM for now. Ideally these would be conditionally included as
appropriate.
llvm-svn: 277777
Our Report implementation calls OutputDebugString, which calls
RtlRaiseException, which can re-enter back into the ASan runtime and
cause a hang.
Don't treat this special debugger-only exception code as a noreturn
event, since the stack won't really unwind all the way.
llvm-svn: 277763
Summary:
The sanitizer allocators can works with a dynamic address space
(i.e. specified with ~0ULL).
Unfortunately, the code was broken on GetMetadata and GetChunkIdx.
The current patch is moving the Win64 memory test to a dynamic
address space. There is a migration to move every concept to a
dynamic address space on windows.
To have a better coverage, the unittest are now testing
dynamic address space on other platforms too.
Reviewers: rnk, kcc
Subscribers: kubabrecka, dberris, llvm-commits, chrisha
Differential Revision: https://reviews.llvm.org/D23170
llvm-svn: 277745
We now stash and restore the xmm registers in the trampolines so that
log handlers don't need to worry about clobbering these registers.
In response to comments in D21612.
Reviewers: rSerge, eugenis, echristo, rnk
Subscribers: mehdi_amini, llvm-commits
Differential Revision: https://reviews.llvm.org/D23051
llvm-svn: 277683
Summary:
Respect the handle_sigill common flag and handle_segv flags while we're
at it.
We still handle signals/exceptions differently on Unix and Windows. The
installation process is tricky on Windows, and difficult to push down
into sanitizer_common without concerning it with the different
static/dynamic CRT models on Windows.
Reviewers: kcc, etienneb
Subscribers: llvm-commits, kubabrecka
Differential Revision: https://reviews.llvm.org/D23098
llvm-svn: 277621
Summary:
This patch is fixing a broken unittest which make the win64 bot failing.
The bug was introduce here:
https://reviews.llvm.org/D23046
The interception code is not the same in 32-bit and in 64-bit.
The added unittest can only be patched on 32-bits.
Reviewers: rnk
Subscribers: llvm-commits, chrisha
Differential Revision: https://reviews.llvm.org/D23099
llvm-svn: 277560
In r235779, Timur bumped the buffer size up to 1<<27, or about 134
million coverage points, presumably to handle Chrome. We allocate two
arrays of uptrs with this size, and this reliably exhausts all available
address space on 32-bit Windows (2 allocations of 512MB) when ASan is
also enabled.
Let's reduce the buffer size for now to stabilize the test suite. We can
re-evaluate the approach later when we've brought the Chrome ASan
builders back to life.
Kostya said that Mike reduced the number of instrumented coverage points
that LLVM emits by half since Timur made this change, so reducing this
array size should also be safe.
With this change, the 32-bit ASan tests reliably pass for me on Windows
10.
llvm-svn: 277558
Summary:
Currently, the Scudo Hardened Allocator only gets its flags via the SCUDO_OPTIONS environment variable.
With this patch, we offer the opportunity for programs to define their own options via __scudo_default_options() which behaves like __asan_default_options() (weak symbol).
A relevant test has been added as well, and the documentation updated accordingly.
I also used this patch as an opportunity to rename a few variables to comply with the LLVM naming scheme, and replaced a use of Report with dieWithMessage for consistency (and to avoid a callback).
Reviewers: llvm-commits, kcc
Differential Revision: https://reviews.llvm.org/D23018
llvm-svn: 277536
Summary:
These instructions where not supported on my win7 computer.
They were happening on strstr when building chrome unittests with asan.
Reviewers: rnk
Subscribers: llvm-commits, chrisha
Differential Revision: https://reviews.llvm.org/D23081
llvm-svn: 277519
Summary:
On my install of Windows 10, RaiseException is a tail call to
kernelbase!RaiseException. Obviously, we fail to intercept that.
Instead, try hooking at the ntdll!RtlRaiseException layer. It is
unlikely that this layer will contain control flow.
Intercepting at this level requires adding a decoding for
'LEA ESP, [ESP + 0xXXXXXXXX]', which is a really obscure way to write
'SUB ESP, 0xXXXXXXXX' that avoids clobbering EFLAGS.
Reviewers: etienneb
Subscribers: llvm-commits, kubabrecka
Differential Revision: https://reviews.llvm.org/D23046
llvm-svn: 277518
Summary:
On Windows 10, this gets called after TLS has been torn down from NTDLL,
and we crash attempting to return fake_tsd. This interceptor isn't
needed after r242948 anyway, so let's remove it. The ASan runtime can
now tolerate unregistered threads calling __asan_handle_no_return.
Reviewers: vitalybuka, etienneb
Subscribers: kubabrecka, llvm-commits
Differential Revision: https://reviews.llvm.org/D23044
llvm-svn: 277478
The system implementation of OSAtomicTestAndClear returns the original bit, but the TSan interceptor has a bug which always returns zero from the function. This patch fixes this and adds a test.
Differential Revision: https://reviews.llvm.org/D23061
llvm-svn: 277461
On Darwin, there are some apps that rely on realloc(nullptr, 0) returning a valid pointer. TSan currently returns nullptr in this case, let's fix it to avoid breaking binary compatibility.
Differential Revision: https://reviews.llvm.org/D22800
llvm-svn: 277458
We were getting warnings about how 'uint32_t*' is different from
'unsigned long*' even though they are effectively the same on Windows.
llvm-svn: 277363
Summary:
Due to a QoI issuse in FreeBSD's libcxxrt-based demangler, one sanitizer
symbolizer test consistently appears to fail:
Value of: DemangleSwiftAndCXX("foo")
Actual: "float"
Expected: "foo"
This is because libcxxrt's __cxa_demangle() incorrectly demangles the "foo"
identifier to "float". It should return an error instead.
For now, XFAIL this particular test for FreeBSD, until we can fix libcxxrt
properly (which might take some time to coordinate with upstream).
Reviewers: rnk, zaks.anna, emaste
Subscribers: emaste, llvm-commits, kubabrecka
Differential Revision: https://reviews.llvm.org/D23001
llvm-svn: 277297
Summary:
On windows, an export can be redirected to an other DLL.
This patch is adding the required support to the internal
GetProcAddress implementation.
This case was encountered by instrumenting chromium (win 64-bits)
using this GN configuration:
```
is_component_build = true
is_debug = false
enable_nacl = false
is_clang = true
is_asan = true
clang_base_path = "d:\src\llvm\ninja64"
clang_use_chrome_plugins = false
clang_version = "4.0.0"
```
The operating system is win7 (x64).
Visual Studio: 2015 Professional
Reviewers: rnk
Subscribers: llvm-commits, chrisha
Differential Revision: https://reviews.llvm.org/D22880
llvm-svn: 277294
This patch adds 48-bits VMA support for tsan on aarch64. As current
mappings for aarch64, 48-bit VMA also supports PIE executable. This
limits the mapping mechanism because the PIE address bits
(usually 0aaaaXXXXXXXX) makes it harder to create a mask/xor value
to include all memory regions. I think it is possible to create a
large application VAM range by either dropping PIE support or tune
current range.
It also changes slight the way addresses are packed in SyncVar structure:
previously it assumes x86_64 as the maximum VMA range. Since ID is 14 bits
wide, shifting 48 bits should be ok.
Tested on x86_64, ppc64le and aarch64 (39 and 48 bits VMA).
llvm-svn: 277137
This addresses some comments from D21612, which contains the following changes:
- Update __xray_patch() and __xray_unpatch() API documentation to not imply asynchrony.
- Introduce a scope cleanup mechanism to make sure we can roll-back changes to the XRayPatching global atomic.
- Introduce a few more comments for potential extension points for other platforms (for the implementation details of patching and un-patching).
Reviewers: eugenis, rnk, kcc, echristo, majnemer
Subscribers: llvm-commits, mehdi_amini
Differential Revision: https://reviews.llvm.org/D22911
llvm-svn: 277124
Summary:
This patch is re-introducing the code to fix the
dynamic hooking on windows and to fix a compiler
warning on Apple.
Related patches:
* https://reviews.llvm.org/D22641
* https://reviews.llvm.org/D22610
* https://reviews.llvm.org/rL276311
* https://reviews.llvm.org/rL276490
Both architecture are using different techniques to
hook on library functions (memchr, strcpy,...).
On Apple, the function is not dynamically hooked and
the symbol always points to a valid function
(i.e. can't be null). The REAL macro returns the
symbol.
On windows, the function is dynamically patch and the
REAL(...) function may or may not be null. It depend
on whether or not the function was hooked correctly.
Also, on windows memcpy and memmove are the same.
```
#if !defined(__APPLE__)
[...]
# define REAL(x) __interception::PTR_TO_REAL(x)
# define ASSIGN_REAL(dst, src) REAL(dst) = REAL(src)
[...]
#else // __APPLE__
[...]
# define REAL(x) x
# define ASSIGN_REAL(x, y)
[...]
#endif // __APPLE__
Reviewers: rnk
Subscribers: kcc, hans, kubabrecka, llvm-commits, bruno, chrisha
Differential Revision: https://reviews.llvm.org/D22758
llvm-svn: 276885
When we delay signals we can deliver them when the signal
is blocked. This can be surprising to the program.
Intercept signal blocking functions merely to process
pending signals. As the result, at worst we will delay
a signal till return from the signal blocking function.
llvm-svn: 276876
The kernel on Nexus 5X returns error_code in ucontext which has
correct FSR_WRITE flag, but empty (zero) abort type field. Removing
the checks means that we will report all SEGVs as READ on very old
kernels, but will properly distinguish READ vs WRITE on moderately
old ones.
llvm-svn: 276803
This test attempts to allocate 100 512MB aligned pages of memory. This
is implemented in the usual way by allocating size + alignment bytes and
aligning the result. As a result, this test allocates 51.2GB of memory.
Windows allocates swap for all memory allocated, and our bots do not
have this much swap available.
Avoid the failure by using a more reasonable alignment, like 16MB, as we
do on 32-bit.
llvm-svn: 276779
This reverts commit r276333.
As I commented in the review (https://reviews.llvm.org/D22415), this change isn't needed because CMAKE_C_FLAGS is implicitly added by CMake to the command line for all C source files.
With this patch enabled CMAKE_C_FLAGS is duplicated on all C sources, and applied to ASM sources, which is not ideal.
I sent an email about this to llvm-commits on the commit thread. I suspect the problem the patch author was actually seeing is that CMAKE_C_FLAGS isn't applied to ASM files, and the builtins library has quite a few of those. The correct solution there is to specify CMAKE_ASM_FLAGS with whatever flags need to be passed to the compiler when compiling ASM files.
If there are other problems with flag propagation, please let me know.
llvm-svn: 276683
sanitizer_common_interceptors.inc:667:12: warning: address of function 'memchr' will always evaluate to 'true' [-Wpointer-bool-conversion]
if (REAL(memchr)) {
~~ ^~~~~~
llvm-svn: 276539
Kostya Serebryany