Merge patch series "pidfs: handle multi-threaded exec and premature thread-group leader exit"

Christian Brauner <brauner@kernel.org> says:

This is another attempt at trying to make pidfd polling for
multi-threaded exec and premature thread-group leader exit consistent.

A quick recap of these two cases:

(1) During a multi-threaded exec by a subthread, i.e., non-thread-group
    leader thread, all other threads in the thread-group including the
    thread-group leader are killed and the struct pid of the
    thread-group leader will be taken over by the subthread that called
    exec. IOW, two tasks change their TIDs.

(2) A premature thread-group leader exit means that the thread-group
    leader exited before all of the other subthreads in the thread-group
    have exited.

Both cases lead to inconsistencies for pidfd polling with PIDFD_THREAD.
Any caller that holds a PIDFD_THREAD pidfd to the current thread-group
leader may or may not see an exit notification on the file descriptor
depending on when poll is performed. If the poll is performed before the
exec of the subthread has concluded an exit notification is generated
for the old thread-group leader. If the poll is performed after the exec
of the subthread has concluded no exit notification is generated for the
old thread-group leader.

The correct behavior would be to simply not generate an exit
notification on the struct pid of a subhthread exec because the struct
pid is taken over by the subthread and thus remains alive.

But this is difficult to handle because a thread-group may exit
premature as mentioned in (2). In that case an exit notification is
reliably generated but the subthreads may continue to run for an
indeterminate amount of time and thus also may exec at some point.

This tiny series tries to address this problem. If that works correctly
then no exit notifications are generated for a PIDFD_THREAD pidfd for a
thread-group leader until all subthreads have been reaped. If a
subthread should exec before no exit notification will be generated
until that task exits or it creates subthreads and repeates the cycle.

* patches from https://lore.kernel.org/r/20250320-work-pidfs-thread_group-v4-0-da678ce805bf@kernel.org:
  selftests/pidfd: third test for multi-threaded exec polling
  selftests/pidfd: second test for multi-threaded exec polling
  selftests/pidfd: first test for multi-threaded exec polling
  pidfs: improve multi-threaded exec and premature thread-group leader exit polling

Link: https://lore.kernel.org/r/20250320-work-pidfs-thread_group-v4-0-da678ce805bf@kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>

4 files changed, 225 insertions, 30 deletions

diff --git a/fs/pidfs.c b/fs/pidfs.c
index a48cc44ced6f..1b3d23e0ffdd 100644
--- a/fs/pidfs.c
+++ b/fs/pidfs.c
@@ -210,20 +210,21 @@ static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
 {
 	struct pid *pid = pidfd_pid(file);
-	bool thread = file->f_flags & PIDFD_THREAD;
 	struct task_struct *task;
 	__poll_t poll_flags = 0;
 
 	poll_wait(file, &pid->wait_pidfd, pts);
 	/*
-	 * Depending on PIDFD_THREAD, inform pollers when the thread
-	 * or the whole thread-group exits.
+	 * Don't wake waiters if the thread-group leader exited
+	 * prematurely. They either get notified when the last subthread
+	 * exits or not at all if one of the remaining subthreads execs
+	 * and assumes the struct pid of the old thread-group leader.
 	 */
 	guard(rcu)();
 	task = pid_task(pid, PIDTYPE_PID);
 	if (!task)
 		poll_flags = EPOLLIN | EPOLLRDNORM | EPOLLHUP;
-	else if (task->exit_state && (thread || thread_group_empty(task)))
+	else if (task->exit_state && !delay_group_leader(task))
 		poll_flags = EPOLLIN | EPOLLRDNORM;
 
 	return poll_flags;
diff --git a/kernel/exit.c b/kernel/exit.c
index 9916305e34d3..683766316a3d 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -743,10 +743,10 @@ static void exit_notify(struct task_struct *tsk, int group_dead)
 
 	tsk->exit_state = EXIT_ZOMBIE;
 	/*
-	 * sub-thread or delay_group_leader(), wake up the
-	 * PIDFD_THREAD waiters.
+	 * Ignore thread-group leaders that exited before all
+	 * subthreads did.
 	 */
-	if (!thread_group_empty(tsk))
+	if (!delay_group_leader(tsk))
 		do_notify_pidfd(tsk);
 
 	if (unlikely(tsk->ptrace)) {
diff --git a/kernel/signal.c b/kernel/signal.c
index 081f19a24506..027ad9e97417 100644
--- a/kernel/signal.c
+++ b/kernel/signal.c
@@ -2180,8 +2180,7 @@ bool do_notify_parent(struct task_struct *tsk, int sig)
 	WARN_ON_ONCE(!tsk->ptrace &&
 	       (tsk->group_leader != tsk || !thread_group_empty(tsk)));
 	/*
-	 * tsk is a group leader and has no threads, wake up the
-	 * non-PIDFD_THREAD waiters.
+	 * Notify for thread-group leaders without subthreads.
 	 */
 	if (thread_group_empty(tsk))
 		do_notify_pidfd(tsk);
diff --git a/tools/testing/selftests/pidfd/pidfd_info_test.c b/tools/testing/selftests/pidfd/pidfd_info_test.c
index 09bc4ae7aed5..1758a1b0457b 100644
--- a/tools/testing/selftests/pidfd/pidfd_info_test.c
+++ b/tools/testing/selftests/pidfd/pidfd_info_test.c
@@ -236,7 +236,7 @@ static void *pidfd_info_pause_thread(void *arg)
 
 TEST_F(pidfd_info, thread_group)
 {
-	pid_t pid_leader, pid_thread;
+	pid_t pid_leader, pid_poller, pid_thread;
 	pthread_t thread;
 	int nevents, pidfd_leader, pidfd_thread, pidfd_leader_thread, ret;
 	int ipc_sockets[2];
@@ -262,6 +262,35 @@ TEST_F(pidfd_info, thread_group)
 		syscall(__NR_exit, EXIT_SUCCESS);
 	}
 
+	/*
+	 * Opening a PIDFD_THREAD aka thread-specific pidfd based on a
+	 * thread-group leader must succeed.
+	 */
+	pidfd_leader_thread = sys_pidfd_open(pid_leader, PIDFD_THREAD);
+	ASSERT_GE(pidfd_leader_thread, 0);
+
+	pid_poller = fork();
+	ASSERT_GE(pid_poller, 0);
+	if (pid_poller == 0) {
+		/*
+		 * We can't poll and wait for the old thread-group
+		 * leader to exit using a thread-specific pidfd. The
+		 * thread-group leader exited prematurely and
+		 * notification is delayed until all subthreads have
+		 * exited.
+		 */
+		fds.events = POLLIN;
+		fds.fd = pidfd_leader_thread;
+		nevents = poll(&fds, 1, 10000 /* wait 5 seconds */);
+		if (nevents != 0)
+			_exit(EXIT_FAILURE);
+		if (fds.revents & POLLIN)
+			_exit(EXIT_FAILURE);
+		if (fds.revents & POLLHUP)
+			_exit(EXIT_FAILURE);
+		_exit(EXIT_SUCCESS);
+	}
+
 	/* Retrieve the tid of the thread. */
 	EXPECT_EQ(close(ipc_sockets[1]), 0);
 	ASSERT_EQ(read_nointr(ipc_sockets[0], &pid_thread, sizeof(pid_thread)), sizeof(pid_thread));
@@ -275,12 +304,7 @@ TEST_F(pidfd_info, thread_group)
 	pidfd_thread = sys_pidfd_open(pid_thread, PIDFD_THREAD);
 	ASSERT_GE(pidfd_thread, 0);
 
-	/*
-	 * Opening a PIDFD_THREAD aka thread-specific pidfd based on a
-	 * thread-group leader must succeed.
-	 */
-	pidfd_leader_thread = sys_pidfd_open(pid_leader, PIDFD_THREAD);
-	ASSERT_GE(pidfd_leader_thread, 0);
+	ASSERT_EQ(wait_for_pid(pid_poller), 0);
 
 	/*
 	 * Note that pidfd_leader is a thread-group pidfd, so polling on it
@@ -389,7 +413,7 @@ static void *pidfd_info_thread_exec(void *arg)
 
 TEST_F(pidfd_info, thread_group_exec)
 {
-	pid_t pid_leader, pid_thread;
+	pid_t pid_leader, pid_poller, pid_thread;
 	pthread_t thread;
 	int nevents, pidfd_leader, pidfd_leader_thread, pidfd_thread, ret;
 	int ipc_sockets[2];
@@ -415,6 +439,37 @@ TEST_F(pidfd_info, thread_group_exec)
 		syscall(__NR_exit, EXIT_SUCCESS);
 	}
 
+	/* Open a thread-specific pidfd for the thread-group leader. */
+	pidfd_leader_thread = sys_pidfd_open(pid_leader, PIDFD_THREAD);
+	ASSERT_GE(pidfd_leader_thread, 0);
+
+	pid_poller = fork();
+	ASSERT_GE(pid_poller, 0);
+	if (pid_poller == 0) {
+		/*
+		 * We can't poll and wait for the old thread-group
+		 * leader to exit using a thread-specific pidfd. The
+		 * thread-group leader exited prematurely and
+		 * notification is delayed until all subthreads have
+		 * exited.
+		 *
+		 * When the thread has execed it will taken over the old
+		 * thread-group leaders struct pid. Calling poll after
+		 * the thread execed will thus block again because a new
+		 * thread-group has started.
+		 */
+		fds.events = POLLIN;
+		fds.fd = pidfd_leader_thread;
+		nevents = poll(&fds, 1, 10000 /* wait 5 seconds */);
+		if (nevents != 0)
+			_exit(EXIT_FAILURE);
+		if (fds.revents & POLLIN)
+			_exit(EXIT_FAILURE);
+		if (fds.revents & POLLHUP)
+			_exit(EXIT_FAILURE);
+		_exit(EXIT_SUCCESS);
+	}
+
 	/* Retrieve the tid of the thread. */
 	EXPECT_EQ(close(ipc_sockets[1]), 0);
 	ASSERT_EQ(read_nointr(ipc_sockets[0], &pid_thread, sizeof(pid_thread)), sizeof(pid_thread));
@@ -423,32 +478,158 @@ TEST_F(pidfd_info, thread_group_exec)
 	pidfd_thread = sys_pidfd_open(pid_thread, PIDFD_THREAD);
 	ASSERT_GE(pidfd_thread, 0);
 
-	/* Open a thread-specific pidfd for the thread-group leader. */
-	pidfd_leader_thread = sys_pidfd_open(pid_leader, PIDFD_THREAD);
-	ASSERT_GE(pidfd_leader_thread, 0);
+	/* Now that we've opened a thread-specific pidfd the thread can exec. */
+	ASSERT_EQ(write_nointr(ipc_sockets[0], &pid_thread, sizeof(pid_thread)), sizeof(pid_thread));
+	EXPECT_EQ(close(ipc_sockets[0]), 0);
+
+	ASSERT_EQ(wait_for_pid(pid_poller), 0);
+
+	/* Wait until the kernel has SIGKILLed the thread. */
+	fds.events = POLLHUP;
+	fds.fd = pidfd_thread;
+	nevents = poll(&fds, 1, -1);
+	ASSERT_EQ(nevents, 1);
+	/* The thread has been reaped. */
+	ASSERT_TRUE(!!(fds.revents & POLLHUP));
+
+	/* Retrieve thread-specific exit info from pidfd. */
+	ASSERT_EQ(ioctl(pidfd_thread, PIDFD_GET_INFO, &info), 0);
+	ASSERT_FALSE(!!(info.mask & PIDFD_INFO_CREDS));
+	ASSERT_TRUE(!!(info.mask & PIDFD_INFO_EXIT));
+	/*
+	 * While the kernel will have SIGKILLed the whole thread-group
+	 * during exec it will cause the individual threads to exit
+	 * cleanly.
+	 */
+	ASSERT_TRUE(WIFEXITED(info.exit_code));
+	ASSERT_EQ(WEXITSTATUS(info.exit_code), 0);
+
+	/*
+	 * The thread-group leader is still alive, the thread has taken
+	 * over its struct pid and thus its pid number.
+	 */
+	info.mask = PIDFD_INFO_CGROUPID | PIDFD_INFO_EXIT;
+	ASSERT_EQ(ioctl(pidfd_leader, PIDFD_GET_INFO, &info), 0);
+	ASSERT_TRUE(!!(info.mask & PIDFD_INFO_CREDS));
+	ASSERT_FALSE(!!(info.mask & PIDFD_INFO_EXIT));
+	ASSERT_EQ(info.pid, pid_leader);
+
+	/* Take down the thread-group leader. */
+	EXPECT_EQ(sys_pidfd_send_signal(pidfd_leader, SIGKILL, NULL, 0), 0);
 
 	/*
-	 * We can poll and wait for the old thread-group leader to exit
-	 * using a thread-specific pidfd.
-	 *
-	 * This only works until the thread has execed. When the thread
-	 * has execed it will have taken over the old thread-group
-	 * leaders struct pid. Calling poll after the thread execed will
-	 * thus block again because a new thread-group has started (Yes,
-	 * it's fscked.).
+	 * Afte the exec we're dealing with an empty thread-group so now
+	 * we must see an exit notification on the thread-specific pidfd
+	 * for the thread-group leader as there's no subthread that can
+	 * revive the struct pid.
 	 */
 	fds.events = POLLIN;
 	fds.fd = pidfd_leader_thread;
 	nevents = poll(&fds, 1, -1);
 	ASSERT_EQ(nevents, 1);
-	/* The thread-group leader has exited. */
 	ASSERT_TRUE(!!(fds.revents & POLLIN));
-	/* The thread-group leader hasn't been reaped. */
 	ASSERT_FALSE(!!(fds.revents & POLLHUP));
 
+	EXPECT_EQ(sys_waitid(P_PIDFD, pidfd_leader, NULL, WEXITED), 0);
+
+	/* Retrieve exit information for the thread-group leader. */
+	info.mask = PIDFD_INFO_CGROUPID | PIDFD_INFO_EXIT;
+	ASSERT_EQ(ioctl(pidfd_leader, PIDFD_GET_INFO, &info), 0);
+	ASSERT_FALSE(!!(info.mask & PIDFD_INFO_CREDS));
+	ASSERT_TRUE(!!(info.mask & PIDFD_INFO_EXIT));
+
+	EXPECT_EQ(close(pidfd_leader), 0);
+	EXPECT_EQ(close(pidfd_thread), 0);
+}
+
+static void *pidfd_info_thread_exec_sane(void *arg)
+{
+	pid_t pid_thread = gettid();
+	int ipc_socket = *(int *)arg;
+
+	/* Inform the grand-parent what the tid of this thread is. */
+	if (write_nointr(ipc_socket, &pid_thread, sizeof(pid_thread)) != sizeof(pid_thread))
+		return NULL;
+
+	if (read_nointr(ipc_socket, &pid_thread, sizeof(pid_thread)) != sizeof(pid_thread))
+		return NULL;
+
+	close(ipc_socket);
+
+	sys_execveat(AT_FDCWD, "pidfd_exec_helper", NULL, NULL, 0);
+	return NULL;
+}
+
+TEST_F(pidfd_info, thread_group_exec_thread)
+{
+	pid_t pid_leader, pid_poller, pid_thread;
+	pthread_t thread;
+	int nevents, pidfd_leader, pidfd_leader_thread, pidfd_thread, ret;
+	int ipc_sockets[2];
+	struct pollfd fds = {};
+	struct pidfd_info info = {
+		.mask = PIDFD_INFO_CGROUPID | PIDFD_INFO_EXIT,
+	};
+
+	ret = socketpair(AF_LOCAL, SOCK_STREAM | SOCK_CLOEXEC, 0, ipc_sockets);
+	EXPECT_EQ(ret, 0);
+
+	pid_leader = create_child(&pidfd_leader, 0);
+	EXPECT_GE(pid_leader, 0);
+
+	if (pid_leader == 0) {
+		close(ipc_sockets[0]);
+
+		/* The thread will outlive the thread-group leader. */
+		if (pthread_create(&thread, NULL, pidfd_info_thread_exec_sane, &ipc_sockets[1]))
+			syscall(__NR_exit, EXIT_FAILURE);
+
+		/*
+		 * Pause the thread-group leader. It will be killed once
+		 * the subthread execs.
+		 */
+		pause();
+		syscall(__NR_exit, EXIT_SUCCESS);
+	}
+
+	/* Retrieve the tid of the thread. */
+	EXPECT_EQ(close(ipc_sockets[1]), 0);
+	ASSERT_EQ(read_nointr(ipc_sockets[0], &pid_thread, sizeof(pid_thread)), sizeof(pid_thread));
+
+	/* Opening a thread as a PIDFD_THREAD must succeed. */
+	pidfd_thread = sys_pidfd_open(pid_thread, PIDFD_THREAD);
+	ASSERT_GE(pidfd_thread, 0);
+
+	/* Open a thread-specific pidfd for the thread-group leader. */
+	pidfd_leader_thread = sys_pidfd_open(pid_leader, PIDFD_THREAD);
+	ASSERT_GE(pidfd_leader_thread, 0);
+
+	pid_poller = fork();
+	ASSERT_GE(pid_poller, 0);
+	if (pid_poller == 0) {
+		/*
+		 * The subthread will now exec. The struct pid of the old
+		 * thread-group leader will be assumed by the subthread which
+		 * becomes the new thread-group leader. So no exit notification
+		 * must be generated. Wait for 5 seconds and call it a success
+		 * if no notification has been received.
+		 */
+		fds.events = POLLIN;
+		fds.fd = pidfd_leader_thread;
+		nevents = poll(&fds, 1, 10000 /* wait 5 seconds */);
+		if (nevents != 0)
+			_exit(EXIT_FAILURE);
+		if (fds.revents & POLLIN)
+			_exit(EXIT_FAILURE);
+		if (fds.revents & POLLHUP)
+			_exit(EXIT_FAILURE);
+		_exit(EXIT_SUCCESS);
+	}
+
 	/* Now that we've opened a thread-specific pidfd the thread can exec. */
 	ASSERT_EQ(write_nointr(ipc_sockets[0], &pid_thread, sizeof(pid_thread)), sizeof(pid_thread));
 	EXPECT_EQ(close(ipc_sockets[0]), 0);
+	ASSERT_EQ(wait_for_pid(pid_poller), 0);
 
 	/* Wait until the kernel has SIGKILLed the thread. */
 	fds.events = POLLHUP;
@@ -482,6 +663,20 @@ TEST_F(pidfd_info, thread_group_exec)
 
 	/* Take down the thread-group leader. */
 	EXPECT_EQ(sys_pidfd_send_signal(pidfd_leader, SIGKILL, NULL, 0), 0);
+
+	/*
+	 * Afte the exec we're dealing with an empty thread-group so now
+	 * we must see an exit notification on the thread-specific pidfd
+	 * for the thread-group leader as there's no subthread that can
+	 * revive the struct pid.
+	 */
+	fds.events = POLLIN;
+	fds.fd = pidfd_leader_thread;
+	nevents = poll(&fds, 1, -1);
+	ASSERT_EQ(nevents, 1);
+	ASSERT_TRUE(!!(fds.revents & POLLIN));
+	ASSERT_FALSE(!!(fds.revents & POLLHUP));
+
 	EXPECT_EQ(sys_waitid(P_PIDFD, pidfd_leader, NULL, WEXITED), 0);
 
 	/* Retrieve exit information for the thread-group leader. */