[sanitizer] Fix vfork interception on loongarch64

[xry111]

Summary

Fix a brown paper bag error made by me in D129418. I didn't set
ASAN_INTERCEPT_VFORK correctly for loongarch64, but created an all-zero
object for __interception::real_vfork. This caused anything calling
vfork() to die instantly.

Fix this issue by setting ASAN_INTERCEPT_VFORK and remove the bad
all-zero definition. Other ports have an all-zero common definition but
we don't need it at least for now.

And, enable ASAN vfork test for loongarch64 to prevent regression in the
future.

[xry111]

@tangyouling I guess this will fix some test failures you've noticed.

[xry111]

Hmm, should I modify the diff to make git clang-format happy?

[MaskRay]

If you can write down how to test this somewhere, then I can verify such changes if needed in the future.

[xry111]

In D137160#3899759, @MaskRay wrote:

If you can write down how to test this somewhere, then I can verify such changes if needed in the future.

$ cat t.c
#include <unistd.h>
int main() { vfork(); }
$ gcc t.c -fsanitize=address
$ ./a.out
AddressSanitizer:DEADLYSIGNAL
=================================================================
==101489==ERROR: AddressSanitizer: SEGV on unknown address 0x000000000000 (pc 0x000000000000 bp 0x7ffffba56030 sp 0x7ffffba56020 T0)
==101489==Hint: pc points to the zero page.
==101489==The signal is caused by a UNKNOWN memory access.
==101489==Hint: address points to the zero page.
    #0 0x0  (<unknown module>)
    #1 0x7ffff20228a4 in __libc_start_call_main (/usr/lib/libc.so.6+0x368a4)

AddressSanitizer can not provide additional info.
SUMMARY: AddressSanitizer: SEGV (<unknown module>) 
==101489==ABORTING

Using libasan from a new GCC build with this diff applied:

$ LD_LIBRARY_PATH=/home/xry111/git-repos/gcc-asan/loongarch64-unknown-linux-gnu/libsanitizer/asan/.libs ./a.out && echo 'ok'
ok

Strangely, GCC asan test does not cover vfork at all so I failed to notice this issue before. But building GCC with --with-build-config=bootstrap-asan can reveal it.

[tangyouling]

In D137160#3898970, @xry111 wrote:

@tangyouling I guess this will fix some test failures you've noticed.

I'll apply it locally.
BTW, since la.local is not yet supported, I modified it locally as follows:

-        la.local  $a0, _ZN14__interception10real_vforkE
+        pcalau12i $a0, %pc_hi20(_ZN14__interception10real_vforkE)
+        addi.d    $a0, $a0, %pc_lo12(_ZN14__interception10real_vforkE)

[SixWeining]

So this can make compiler-rt/test/asan/TestCases/Linux/vfork.cpp pass on loongarch64?

[MaskRay]

In D137160#3900958, @xry111 wrote:

In D137160#3899759, @MaskRay wrote:

If you can write down how to test this somewhere, then I can verify such changes if needed in the future.

$ cat t.c
#include <unistd.h>
int main() { vfork(); }
$ gcc t.c -fsanitize=address
$ ./a.out
AddressSanitizer:DEADLYSIGNAL
=================================================================
==101489==ERROR: AddressSanitizer: SEGV on unknown address 0x000000000000 (pc 0x000000000000 bp 0x7ffffba56030 sp 0x7ffffba56020 T0)
==101489==Hint: pc points to the zero page.
==101489==The signal is caused by a UNKNOWN memory access.
==101489==Hint: address points to the zero page.
    #0 0x0  (<unknown module>)
    #1 0x7ffff20228a4 in __libc_start_call_main (/usr/lib/libc.so.6+0x368a4)

AddressSanitizer can not provide additional info.
SUMMARY: AddressSanitizer: SEGV (<unknown module>) 
==101489==ABORTING

Using libasan from a new GCC build with this diff applied:

$ LD_LIBRARY_PATH=/home/xry111/git-repos/gcc-asan/loongarch64-unknown-linux-gnu/libsanitizer/asan/.libs ./a.out && echo 'ok'
ok

Strangely, GCC asan test does not cover vfork at all so I failed to notice this issue before. But building GCC with --with-build-config=bootstrap-asan can reveal it.

Thanks but how to apply a patch to https://gcc.gnu.org/git/?p=gcc.git 's libasan is a step I missed...

[xry111]

Thanks but how to apply a patch to https://gcc.gnu.org/git/?p=gcc.git 's libasan is a step I missed...

cd /path/to/git-repos/gcc
patch -d libsanitizer -Np3 < /path/to/D137160.diff

[xry111]

In D137160#3901074, @SixWeining wrote:

So this can make compiler-rt/test/asan/TestCases/Linux/vfork.cpp pass on loongarch64?

The test case compiled and ran OK with patched GCC.

[SixWeining]

In D137160#3901128, @xry111 wrote:

In D137160#3901074, @SixWeining wrote:

So this can make compiler-rt/test/asan/TestCases/Linux/vfork.cpp pass on loongarch64?

The test case compiled and ran OK with patched GCC.

Good! So do we need to enable this test on loongarch64? Currently it is:

// REQUIRES: aarch64-target-arch || x86_64-target-arch || i386-target-arch || arm-target-arch || riscv64-target-arch

[xry111]

Enable ASAN vfork test for loongarch64.

[SixWeining]

LGTM. Thanks.

[tangyouling]

LGTM.

compiler-rt/lib/asan/asan_interceptors.h

 #if SANITIZER_GLIBC
 # define ASAN_INTERCEPT___STRDUP 1
 #else
 # define ASAN_INTERCEPT___STRDUP 0
 #endif
 
 #if SANITIZER_LINUX &&                                                \
     (defined(__arm__) || defined(__aarch64__) || defined(__i386__) || \
-     defined(__x86_64__) || SANITIZER_RISCV64)
+     defined(__x86_64__) || SANITIZER_RISCV64 || SANITIZER_LOONGARCH64)
 # define ASAN_INTERCEPT_VFORK 1
 #else
 # define ASAN_INTERCEPT_VFORK 0
 #endif
 
 #if SANITIZER_NETBSD
 # define ASAN_INTERCEPT_PTHREAD_ATFORK 1
 #else

compiler-rt/lib/sanitizer_common/sanitizer_common_interceptors_vfork_loongarch64.inc.S

 #if defined(__loongarch_lp64) && defined(__linux__)
 
 #include "sanitizer_common/sanitizer_asm.h"
 
 ASM_HIDDEN(COMMON_INTERCEPTOR_SPILL_AREA)
 ASM_HIDDEN(_ZN14__interception10real_vforkE)
 
-.bss
-.type _ZN14__interception10real_vforkE, @object
-.size _ZN14__interception10real_vforkE, 8
-_ZN14__interception10real_vforkE:
-        .zero     8
-
 .text
 .globl ASM_WRAPPER_NAME(vfork)
 ASM_TYPE_FUNCTION(ASM_WRAPPER_NAME(vfork))
 ASM_WRAPPER_NAME(vfork):
         // Save ra in the off-stack spill area.
         // allocate space on stack
         addi.d    $sp, $sp, -16
         // store $ra value

compiler-rt/test/asan/TestCases/Linux/vfork.cpp

 // https://github.com/google/sanitizers/issues/925
 // RUN: %clang_asan -O0 %s -o %t && %run %t 2>&1
 
-// REQUIRES: aarch64-target-arch || x86_64-target-arch || i386-target-arch || arm-target-arch || riscv64-target-arch
+// REQUIRES: aarch64-target-arch || x86_64-target-arch || i386-target-arch || arm-target-arch || riscv64-target-arch || loongarch64-target-arch
 
 #include <assert.h>
 #include <sys/types.h>
 #include <sys/wait.h>
 #include <unistd.h>
 #include <stdio.h>
 #include <sanitizer/asan_interface.h>