Windows: -fsanitize=address doesn't work with exceptions

[mikey-b]

Hi all,

On Windows 10 64bit, LDC2 1.27.0-beta1. I am having issues with throwing exceptions in constructors. Address sanatiser is able to catch the issue as an access violation. Here is example code to trigger this issue:

import std.stdio;

class example: Exception {
	this(string msg, string file = __FILE__, size_t line = __LINE__) {
		super(msg, file, line);
	}
}

class test {
	this() {		
		throw new example("this is a test");
	}
}

void main() {
	try {
		auto t = new test();
	} catch (example e) {
		writeln(e.msg);
	}
}

Compiled with:
ldc2.exe -g -fsanitize='address' .\test.d -of='test.exe'

Access Violation is on "writeln(e.msg)" line.

Kind regards,
Mike Brown

[JohanEngelen]

Please paste the full ASan output here.

[mikey-b]

Hi Johan,

Sure, here it is (It is identical on all runs)

Kind regards,
Mike

.\test.exe
=================================================================
==14804==ERROR: AddressSanitizer: access-violation on unknown address 0x000000000020 (pc 0x7ff6fdf51506 bp 0x00e7804ffa80 sp 0x00e7804ff9a0 T0)
==14804==The signal is caused by a READ memory access.
==14804==Hint: address points to the zero page.
==14804==*** WARNING: Failed to initialize DbgHelp! ***
==14804==*** Most likely this means that the app is already ***
==14804==*** using DbgHelp, possibly with incompatible flags. ***
==14804==*** Due to technical reasons, symbolization might crash ***
==14804==*** or produce wrong results. ***
#0 0x7ff6fdf51505 in _Dmain C:\Users\Mike Brown\Documents\Projects\nibia-lang\test.d:19
#1 0x7ff6fdfc4ec4 in _D2rt6dmain212_d_run_main2UAAamPUQgZiZ6runAllMFZv (C:\Users\Mike Brown\Documents\Projects\nibia-lang\test.exe+0x140074ec4)
#2 0x7ff6fdfc4b96 in _d_run_main2 (C:\Users\Mike Brown\Documents\Projects\nibia-lang\test.exe+0x140074b96)
#3 0x7ff6fdfc4e52 in _d_wrun_main (C:\Users\Mike Brown\Documents\Projects\nibia-lang\test.exe+0x140074e52)
#4 0x7ff6fdf51993 in wmain C:\Program Files\LDC 1.27\import\core\internal\entrypoint.d:32
#5 0x7ff6fdfeac33 in invoke_main d:\agent_work\4\s\src\vctools\crt\vcstartup\src\startup\exe_common.inl:90
#6 0x7ff6fdfeac33 in __scrt_common_main_seh d:\agent_work\4\s\src\vctools\crt\vcstartup\src\startup\exe_common.inl:288
#7 0x7ffcd2f47033 (C:\WINDOWS\System32\KERNEL32.DLL+0x180017033)
#8 0x7ffcd4a42650 (C:\WINDOWS\SYSTEM32\ntdll.dll+0x180052650)

AddressSanitizer can not provide additional info.
SUMMARY: AddressSanitizer: access-violation C:\Users\Mike Brown\Documents\Projects\nibia-lang\test.d:19 in _Dmain
==14804==ABORTING

[mikey-b]

Hi again Johan,

I apologise, it isnt the same on each run. Here is another ASAN report:

.\test.exe

==9532==ERROR: AddressSanitizer: unknown-crash on address 0x0279dcbe5e88 at pc 0x7ff6fdf6b4d9 bp 0x00d05432e640 sp 0x00d05432e680
READ of size 4096 at 0x0279dcbe5e88 thread T0
#0 0x7ff6fdf6b4d8 in __asan_wrap_memmove D:\a\1\s\compiler-rt\lib\sanitizer_common\sanitizer_common_interceptors.inc:793
#1 0x7ff6fdfeea82 in _fwrite_nolock minkernel\crts\ucrt\src\appcrt\stdio\fwrite.cpp:94
#2 0x7ff6fdfee97e in <lambda_5856287d7ecd2be6c9197bb4007c3f6e>::operator() minkernel\crts\ucrt\src\appcrt\stdio\fwrite.cpp:37
#3 0x7ff6fdfee921 in __crt_seh_guarded_call::operator()<<lambda_2831f20263db5b546e098b45503eb778>,<lambda_5856287d7ecd2be6c9197bb4007c3f6e> &,<lambda_0838d7e100fbcbd261b69cfea6abb102>

vccrt\vcruntime\inc\internal_shared.h:204
#4 0x7ff6fdfeebca in __acrt_lock_stream_and_call minkernel\crts\ucrt\inc\corecrt_internal_stdio.h:252
#5 0x7ff6fdfeebca in fwrite minkernel\crts\ucrt\src\appcrt\stdio\fwrite.cpp:33
#6 0x7ff6fdf52851 in _D3std5stdio__T13trustedFwriteTaZQsFNbNiNePOS4core4stdcQBx6_iobufxAaZm C:\Program Files\LDC 1.27\import\std\stdio.d:4799
#7 0x7ff6fdf51bd1 in _D3std5stdio4File17LockingTextWriter__T3putTAyaZQjMFNfMQlZv C:\Program Files\LDC 1.27\import\std\stdio.d:3260
#8 0x7ff6fdf516ef in _D3std5stdio__T7writelnTAyaZQnFNfQjZv C:\Users\Mike Brown\Documents\Projects\nibia-lang\test.d:4328
#9 0x7ff6fdf51515 in _Dmain C:\Users\Mike Brown\Documents\Projects\nibia-lang\test.d:19
#10 0x7ff6fdfc4ec4 in _D2rt6dmain212_d_run_main2UAAamPUQgZiZ6runAllMFZv (C:\Users\Mike Brown\Documents\Projects\nibia-lang\test.exe+0x140074ec4)
#11 0x7ff6fdfc4b96 in _d_run_main2 (C:\Users\Mike Brown\Documents\Projects\nibia-lang\test.exe+0x140074b96)
#12 0x7ff6fdfc4e52 in _d_wrun_main (C:\Users\Mike Brown\Documents\Projects\nibia-lang\test.exe+0x140074e52)
#13 0x7ff6fdf51993 in wmain C:\Program Files\LDC 1.27\import\core\internal\entrypoint.d:32
#14 0x7ff6fdfeac33 in invoke_main d:\agent_work\4\s\src\vctools\crt\vcstartup\src\startup\exe_common.inl:90
#15 0x7ff6fdfeac33 in __scrt_common_main_seh d:\agent_work\4\s\src\vctools\crt\vcstartup\src\startup\exe_common.inl:288
#16 0x7ffcd2f47033 (C:\WINDOWS\System32\KERNEL32.DLL+0x180017033)
#17 0x7ffcd4a42650 (C:\WINDOWS\SYSTEM32\ntdll.dll+0x180052650)

Address 0x0279dcbe5e88 is located in the low shadow area.
SUMMARY: AddressSanitizer: unknown-crash D:\a\1\s\compiler-rt\lib\sanitizer_common\sanitizer_common_interceptors.inc:793 in __asan_wrap_memmove
==9532==ABORTING

[JohanEngelen]

I'm quite sure this is a problem with ASan on Windows and exceptions. Can you simplify the testcase so that it does not use a class, or phobos stdio. I think simply throwing an exception and catching it will trigger the problem.

[JohanEngelen]

google/sanitizers#749

[mikey-b]

Hi Johan,

I can confirm that

class example: Exception {
	this(string msg, string file = __FILE__, size_t line = __LINE__) {
		super(msg, file, line);
	}
}

void main() {
	try {
		throw new example("test");
	} catch (example e) {
		auto a = (e.msg == "test");
	}
}

also causes the issue. But access to e.msg is required. Is there any fix or workaround available?

Kind regards,
Mike Brown

[JohanEngelen]

I don't know the workaround, but when googling a bit, it looks like it is a known issue and some people report that with changing settings in VS it does work...

[kinke]

I am having issues with throwing exceptions in constructors.

The testcase obviously works just fine without ASan, just for clarification.

[mikey-b]

Hi Martin, I'm not 100% sure. I was using throw in the constructor to return an instance of something with a different type. To work around constructors only returning the specific type. I think I was having issues storing the returned thrown object, hence why I started fuzzing and then asan testing. I've given up on that, and avoiding exceptions now. I now have a factory function that returns a variant. So.. I'm not sure Kind regards Mike

…

On Sat, 19 Jun 2021, 19:09 Martin Kinkelin, ***@***.***> wrote: I am having issues with throwing exceptions in constructors. The testcase obviously works just fine without ASan, just for clarification. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#3760 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAOAIW2SEANM2F4LWWH2JODTTTMMRANCNFSM46SODWCQ> .

[kinke]

Wow, sounds extremely weird! :D - A factory returning a variant seems much more reasonable. If you think the original code should really have worked (edit: or reported a better error), please file that as a separate issue with according code.