This is an archive of the discontinued LLVM Phabricator instance.

It fails for me both on x86_64 and s390x.
I think there might be a race: if we manage to write CODE lines before llvm-symbolizer exits, we just get a "Failed to use and restart external symbolizer" message.
However, if we write at a later point in time, we get a SIGPIPE.

One can induce the race as follows:

build$ PATH=$PWD/bin:$PATH gdb projects/compiler-rt/test/asan/X86_64LinuxConfig/TestCases/Linux/Output/read_binary_name_regtest.c.tmp
(gdb) b WriteToFile
Breakpoint 1 at 0x4d5970: file llvm-project/compiler-rt/lib/sanitizer_common/sanitizer_posix.cpp, line 184.
(gdb) r
Breakpoint 1, __sanitizer::WriteToFile (fd=6, buff=buff@entry=0x7ffff50ed018, buff_size=buff_size@entry=158, bytes_written=bytes_written@entry=0x7fffffffb2b8, error_p=error_p@entry=0x0)
(gdb) c
Program received signal SIGPIPE, Broken pipe.
__sanitizer::internal_write (fd=6, buf=buf@entry=0x7ffff50ed018, count=count@entry=158) at llvm-project/compiler-rt/lib/sanitizer_common/sanitizer_linux.cpp:222

Come to think of it, the most important thing actually must be that this is with -DBUILD_SHARED_LIBS=ON. I don't think static builds are affected.

OK, makes sense, LGTM.

This revision is now accepted and ready to land.Mar 24 2020, 10:48 AM

Closed by commit rGdc37287320cc: [asan] Fix read_binary_name_regtest.c test dying with SIGPIPE (authored by iii, committed by uweigand). · Explain WhyMar 26 2020, 5:55 AM

This revision was automatically updated to reflect the committed changes.

Revision Contents

Path

Size

compiler-rt/

test/

asan/

TestCases/

Linux/

read_binary_name_regtest.c

6 lines

Diff 251922

compiler-rt/test/asan/TestCases/Linux/read_binary_name_regtest.c

	// Regression test for https://crbug.com/502974, where ASan was unable to read			// Regression test for https://crbug.com/502974, where ASan was unable to read
	// the binary name because of sandbox restrictions.			// the binary name because of sandbox restrictions.
	// This test uses seccomp-BPF to restrict the readlink() system call and makes			// This test uses seccomp-BPF to restrict the readlink() system call and makes
	// sure ASan is still able to			// sure ASan is still able to
	// RUN: not ls /usr/include/linux/seccomp.h \|\| ( %clang_asan %s -o %t && ( not %run %t 2>&1 ) \| FileCheck %s )			// Disable symbolizing results, since this will invoke llvm-symbolizer, which
				// will be unable to resolve its $ORIGIN due to readlink() restriction and will
				// thus fail to start, causing the test to die with SIGPIPE when attempting to
				// talk to it.
				// RUN: not ls /usr/include/linux/seccomp.h \|\| ( %clang_asan %s -o %t && ( not env ASAN_OPTIONS=symbolize=0 %run %t 2>&1 ) \| FileCheck %s )
	// REQUIRES: shell			// REQUIRES: shell
	// UNSUPPORTED: android			// UNSUPPORTED: android

	#include <errno.h>			#include <errno.h>
	#include <stddef.h>			#include <stddef.h>
	#include <stdlib.h>			#include <stdlib.h>
	#include <stdio.h>			#include <stdio.h>
	#include <sys/prctl.h>			#include <sys/prctl.h>
	▲ Show 20 Lines • Show All 42 Lines • Show Last 20 Lines

This is an archive of the discontinued LLVM Phabricator instance.

[asan] Fix read_binary_name_regtest.c test dying with SIGPIPEClosedPublic

Details

Diff Detail

Event Timeline

Revision Contents

Diff 251922

compiler-rt/test/asan/TestCases/Linux/read_binary_name_regtest.c

[asan] Fix read_binary_name_regtest.c test dying with SIGPIPE
ClosedPublic