[LSan] skip leaks from dlerror

[fmayer]

We have known false positives, and the return value is never
user-managed anyway.

[llvmbot]

@llvm/pr-subscribers-compiler-rt-sanitizer

Author: Florian Mayer (fmayer)

Changes

We have known false positives, and the return value is never
user-managed anyway.

---

Full diff: https://github.com/llvm/llvm-project/pull/142876.diff

2 Files Affected:

• (modified) compiler-rt/lib/lsan/lsan_common.cpp (+2-1)
• (modified) compiler-rt/test/hwasan/TestCases/Posix/dlerror.cpp (+2-2)

diff --git a/compiler-rt/lib/lsan/lsan_common.cpp b/compiler-rt/lib/lsan/lsan_common.cpp
index 7ab9e4ff2ac9f..b17a17e1193bc 100644
--- a/compiler-rt/lib/lsan/lsan_common.cpp
+++ b/compiler-rt/lib/lsan/lsan_common.cpp
@@ -124,7 +124,8 @@ static const char kStdSuppressions[] =
 #  endif
     // TLS leak in some glibc versions, described in
     // https://sourceware.org/bugzilla/show_bug.cgi?id=12650.
-    "leak:*tls_get_addr*\n";
+    "leak:*tls_get_addr*\n"
+    "leak:*dlerror*\n";
 
 void InitializeSuppressions() {
   CHECK_EQ(nullptr, suppression_ctx);
diff --git a/compiler-rt/test/hwasan/TestCases/Posix/dlerror.cpp b/compiler-rt/test/hwasan/TestCases/Posix/dlerror.cpp
index 15455ba5af780..b6e486b291f3a 100644
--- a/compiler-rt/test/hwasan/TestCases/Posix/dlerror.cpp
+++ b/compiler-rt/test/hwasan/TestCases/Posix/dlerror.cpp
@@ -2,7 +2,6 @@
 // This is currently not implemented, so this test is XFAIL.
 
 // RUN: %clangxx_hwasan -O0 %s -o %t && HWASAN_OPTIONS=detect_leaks=1 %run %t
-// XFAIL: *
 
 #include <assert.h>
 #include <dlfcn.h>
@@ -12,7 +11,8 @@
 #include <string.h>
 #include <unistd.h>
 
-constexpr auto kKeys = 500;
+// musl only has  128 keys
+constexpr auto kKeys = 100;
 
 int main(int argc, char **argv) {
   __hwasan_enable_allocator_tagging();

[thurstond]

There's a flag:

LSAN_FLAG(bool, use_ld_allocations, true,
          "Root set: mark as reachable all allocations made from dynamic "
          "linker. This was the old way to handle dynamic TLS, and will "
          "be removed soon. Do not use this flag.")

Obviously we shouldn't use this flag, but I wonder if something like that would be better than specifically suppressing *dlerror* (are other dl* functions going to break at some point?).

[fmayer]

There's a flag:

LSAN_FLAG(bool, use_ld_allocations, true,
          "Root set: mark as reachable all allocations made from dynamic "
          "linker. This was the old way to handle dynamic TLS, and will "
          "be removed soon. Do not use this flag.")

Obviously we shouldn't use this flag, but I wonder if something like that would be better than specifically suppressing *dlerror* (are other dl* functions going to break at some point?).

Yeah, it would be better to fix the root cause. But until then, this exclusion makes sense IMO. WDYT?

[thurstond]

Yeah, it would be better to fix the root cause. But until then, this exclusion makes sense IMO. WDYT?

SGTM.