Set a new TLS when using CLONE_VM.

sandbox/linux/services/credentials_unittest.cc

Index: sandbox/linux/services/credentials_unittest.cc
diff --git a/sandbox/linux/services/credentials_unittest.cc b/sandbox/linux/services/credentials_unittest.cc
index 6b93c86c3ecbc2104282b50da80e32f3d9ad06d3..3ce4fc3fc61dd224fab55b27662cba6d7b2839df 100644
--- a/sandbox/linux/services/credentials_unittest.cc
+++ b/sandbox/linux/services/credentials_unittest.cc
@@ -6,6 +6,7 @@
 
 #include <errno.h>
 #include <fcntl.h>
+#include <pthread.h>
 #include <stdio.h>
 #include <sys/capability.h>
 #include <sys/stat.h>
@@ -237,6 +238,18 @@ SANDBOX_TEST(Credentials, SetCapabilitiesMatchesLibCap2) {
   CHECK_EQ(0, cap_compare(expected_cap.get(), actual_cap.get()));
 }
 
+// Disabled on ASAN because of crbug.com/451603.
+SANDBOX_TEST(Credentials, DISABLE_ON_ASAN(DropFileSystemAccessPreservesTLS)) {
+  // Probably missing kernel support.
+  if (!Credentials::MoveToNewUserNS()) return;
+  CHECK(Credentials::DropFileSystemAccess(ProcUtil::OpenProc().get()));
+
+  // In glibc, pthread_getattr_np makes an assertion about the cached PID/TID in
+  // TLS.
+  pthread_attr_t attr;
+  EXPECT_EQ(0, pthread_getattr_np(pthread_self(), &attr));
+}
+

[jln (very slow on Chromium), 2015/07/21 23:39:17]

Could you add another test that creates a thread that looks at getpid() and
sys_gettid() and checks that they're ok?

[rickyz (no longer on Chrome), 2015/07/22 04:00:32]

Sorry, I wasn't quite sure what kind of test you were thinking about. If I
create another thread, it'll have its own TLS, and the cached tid should be
correct there. Even if the cached pid is incorrectly set to -1, getpid() will
make a getpid syscall and update the cached value, so we wouldn't be able to
tell that anything was amiss.

[jln (very slow on Chromium), 2015/07/22 17:30:06]

Yeah, sorry I meant another "thread" with the vfork() we're doing. But that
would require quite a bit of work to expose that stuff to testing and we don't
really want to run too much code in that "thread" anyways.

Is there anything we could test with raise() ? I vaguely remember raise() using
some of the cached values in the TCB/PCB.

 }  // namespace.
 
 }  // namespace sandbox.