[PATCH v2] selftests/pidfd: wait for POLLHUP before reading exit info

[fujo]

The CLONE_AUTOREAP tests poll for POLLIN and then immediately call
PIDFD_GET_INFO with PIDFD_INFO_EXIT.  PIDFD_INFO_EXIT is only published
from release_task() -> pidfs_exit(), which runs after the pidfd already
reported POLLIN for the exited task.  A poller woken by
do_notify_pidfd() can therefore read a mask without PIDFD_INFO_EXIT;
under KASAN the autoreap_multithreaded case fails this way.

POLLHUP is returned once the task linkage has been removed from
struct pid, which happens after pidfs_exit() has published the exit
information.  Wait for POLLHUP and check it in revents.

With this change all tests pass ten consecutive runs on an unpatched
kernel.

Signed-off-by: fujo <hooyan2024@xxxxxxxxx>
---
 .../selftests/pidfd/pidfd_autoreap_test.c     | 36 +++++++++----------
 1 file changed, 18 insertions(+), 18 deletions(-)

diff --git a/tools/testing/selftests/pidfd/pidfd_autoreap_test.c b/tools/testing/selftests/pidfd/pidfd_autoreap_test.c
index 1c586482f..b4f3699ef 100644
--- a/tools/testing/selftests/pidfd/pidfd_autoreap_test.c
+++ b/tools/testing/selftests/pidfd/pidfd_autoreap_test.c
@@ -149,7 +149,7 @@ TEST(autoreap_rejects_thread)
 
 /*
  * Basic test: create an autoreap child, let it exit, verify:
- * - pidfd becomes readable (poll returns POLLIN)
+ * - pidfd becomes readable (poll returns POLLHUP)
  * - PIDFD_GET_INFO returns the correct exit code
  * - waitpid() returns -1/ECHILD (no zombie)
  */
@@ -172,10 +172,10 @@ TEST(autoreap_basic)
 
 	/* Wait for the child to exit via pidfd poll. */
 	pfd.fd = pidfd;
-	pfd.events = POLLIN;
+	pfd.events = POLLHUP;
 	ret = poll(&pfd, 1, 5000);
 	ASSERT_EQ(ret, 1);
-	ASSERT_TRUE(pfd.revents & POLLIN);
+	ASSERT_TRUE(pfd.revents & POLLHUP);
 
 	/* Verify exit info via PIDFD_GET_INFO. */
 	ret = ioctl(pidfd, PIDFD_GET_INFO, &info);
@@ -225,10 +225,10 @@ TEST(autoreap_signaled)
 
 	/* Wait for exit via pidfd. */
 	pfd.fd = pidfd;
-	pfd.events = POLLIN;
+	pfd.events = POLLHUP;
 	ret = poll(&pfd, 1, 5000);
 	ASSERT_EQ(ret, 1);
-	ASSERT_TRUE(pfd.revents & POLLIN);
+	ASSERT_TRUE(pfd.revents & POLLHUP);
 
 	/* Verify signal info. */
 	ret = ioctl(pidfd, PIDFD_GET_INFO, &info);
@@ -334,10 +334,10 @@ TEST(autoreap_reparent)
 
 	/* Wait for it to exit via pidfd poll. */
 	pfd.fd = pidfd;
-	pfd.events = POLLIN;
+	pfd.events = POLLHUP;
 	ret = poll(&pfd, 1, 5000);
 	ASSERT_EQ(ret, 1);
-	ASSERT_TRUE(pfd.revents & POLLIN);
+	ASSERT_TRUE(pfd.revents & POLLHUP);
 
 	/*
 	 * The grandchild should have been autoreaped even though
@@ -421,10 +421,10 @@ TEST(autoreap_multithreaded)
 
 	/* Wait for the process to fully exit via pidfd poll. */
 	pfd.fd = pidfd;
-	pfd.events = POLLIN;
+	pfd.events = POLLHUP;
 	ret = poll(&pfd, 1, 5000);
 	ASSERT_EQ(ret, 1);
-	ASSERT_TRUE(pfd.revents & POLLIN);
+	ASSERT_TRUE(pfd.revents & POLLHUP);
 
 	/* Verify exit info. */
 	ret = ioctl(pidfd, PIDFD_GET_INFO, &info);
@@ -512,7 +512,7 @@ TEST(autoreap_no_inherit)
 
 	/* Wait for the autoreap child to exit. */
 	pfd.fd = pidfd;
-	pfd.events = POLLIN;
+	pfd.events = POLLHUP;
 	ret = poll(&pfd, 1, 5000);
 	ASSERT_EQ(ret, 1);
 
@@ -571,7 +571,7 @@ TEST(nnp_sets_no_new_privs)
 
 	/* Wait for child to exit. */
 	pfd.fd = pidfd;
-	pfd.events = POLLIN;
+	pfd.events = POLLHUP;
 	ret = poll(&pfd, 1, 5000);
 	ASSERT_EQ(ret, 1);
 
@@ -631,7 +631,7 @@ TEST(autoreap_no_new_privs_unset)
 	ASSERT_GE(pidfd, 0);
 
 	pfd.fd = pidfd;
-	pfd.events = POLLIN;
+	pfd.events = POLLHUP;
 	ret = poll(&pfd, 1, 5000);
 	ASSERT_EQ(ret, 1);
 
@@ -695,10 +695,10 @@ TEST(autokill_basic)
 
 	/* Wait for the child to die via the pidfd_open'd fd. */
 	pfd.fd = pollfd_fd;
-	pfd.events = POLLIN;
+	pfd.events = POLLHUP;
 	ret = poll(&pfd, 1, 5000);
 	ASSERT_EQ(ret, 1);
-	ASSERT_TRUE(pfd.revents & POLLIN);
+	ASSERT_TRUE(pfd.revents & POLLHUP);
 
 	/* Child should be autoreaped — no zombie. */
 	usleep(100000);
@@ -799,7 +799,7 @@ TEST(autokill_pidfd_open_no_effect)
 
 	/* Verify child is still alive by polling the clone3 pidfd. */
 	pfd.fd = pidfd;
-	pfd.events = POLLIN;
+	pfd.events = POLLHUP;
 	ret = poll(&pfd, 1, 0);
 	ASSERT_EQ(ret, 0) {
 		TH_LOG("Child died after closing pidfd_open fd — should still be alive");
@@ -813,10 +813,10 @@ TEST(autokill_pidfd_open_no_effect)
 	close(pidfd);
 
 	pfd.fd = open_fd;
-	pfd.events = POLLIN;
+	pfd.events = POLLHUP;
 	ret = poll(&pfd, 1, 5000);
 	ASSERT_EQ(ret, 1);
-	ASSERT_TRUE(pfd.revents & POLLIN);
+	ASSERT_TRUE(pfd.revents & POLLHUP);
 
 	/* Child should be autoreaped — no zombie. */
 	usleep(100000);
@@ -884,7 +884,7 @@ TEST(autokill_without_nnp_with_cap)
 
 	/* Wait for child to exit. */
 	pfd.fd = pidfd;
-	pfd.events = POLLIN;
+	pfd.events = POLLHUP;
 	ret = poll(&pfd, 1, 5000);
 	ASSERT_EQ(ret, 1);
 
-- 
2.53.0