process: monitor for exit with kqueue on BSDs (#3441)

This adds a workaround for an xnu kernel bug that sometimes results in
SIGCHLD not being delivered. The workaround is to use kevent to listen
for EVFILT_PROC/NOTE_EXIT events instead of relying on SIGCHLD on *BSD.
 
Apple rdar: FB9529664
Refs: https://github.com/libuv/libuv/pull/3257

src/unix/internal.h

@@ -282,6 +282,7 @@ uv_handle_type uv__handle_type(int fd);
 FILE* uv__open_file(const char* path);
 int uv__getpwuid_r(uv_passwd_t* pwd);
 int uv__search_path(const char* prog, char* buf, size_t* buflen);
+void uv__wait_children(uv_loop_t* loop);
 
 /* random */
 int uv__random_devurandom(void* buf, size_t buflen);

src/unix/kqueue.c

@@ -284,6 +284,11 @@ void uv__io_poll(uv_loop_t* loop, int timeout) {
     loop->watchers[loop->nwatchers + 1] = (void*) (uintptr_t) nfds;
     for (i = 0; i < nfds; i++) {
       ev = events + i;
+      if (ev->filter == EVFILT_PROC) {
+        uv__wait_children(loop);
+        nevents++;
+        continue;
+      }
       fd = ev->ident;
       /* Skip invalidated events, see uv__platform_invalidate_fd */
       if (fd == -1)

src/unix/process.c

@@ -49,10 +49,20 @@ extern char **environ;
 # include "zos-base.h"
 #endif
 
+#if defined(__APPLE__) || defined(__DragonFly__) || defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
+#include <sys/event.h>
+#endif
 
+
+#if !(defined(__APPLE__) || defined(__DragonFly__) || defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__))
 static void uv__chld(uv_signal_t* handle, int signum) {
+  assert(signum == SIGCHLD);
+  uv__wait_children(handle->loop);
+}
+#endif
+
+void uv__wait_children(uv_loop_t* loop) {
   uv_process_t* process;
-  uv_loop_t* loop;
   int exit_status;
   int term_signal;
   int status;
@@ -61,10 +71,7 @@ static void uv__chld(uv_signal_t* handle, int signum) {
   QUEUE* q;
   QUEUE* h;
 
-  assert(signum == SIGCHLD);
-
   QUEUE_INIT(&pending);
-  loop = handle->loop;
 
   h = &loop->process_handles;
   q = QUEUE_HEAD(h);
@@ -420,7 +427,9 @@ int uv_spawn(uv_loop_t* loop,
   if (err)
     goto error;
 
+#if !(defined(__APPLE__) || defined(__DragonFly__) || defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__))
   uv_signal_start(&loop->child_watcher, uv__chld, SIGCHLD);
+#endif
 
   /* Acquire write lock to prevent opening new fds in worker threads */
   uv_rwlock_wrlock(&loop->cloexec_lock);
@@ -495,6 +504,13 @@ int uv_spawn(uv_loop_t* loop,
 
   /* Only activate this handle if exec() happened successfully */
   if (exec_errorno == 0) {
+#if defined(__APPLE__) || defined(__DragonFly__) || defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
+    struct kevent event;
+    EV_SET(&event, pid, EVFILT_PROC, EV_ADD | EV_ONESHOT, NOTE_EXIT, 0, 0);
+    if (kevent(loop->backend_fd, &event, 1, NULL, 0, NULL))
+      abort();
+#endif
+
     QUEUE_INSERT_TAIL(&loop->process_handles, &process->queue);
     uv__handle_start(process);
   }

test/test-list.h

@@ -320,6 +320,7 @@ TEST_DECLARE   (spawn_reads_child_path)
 TEST_DECLARE   (spawn_inherit_streams)
 TEST_DECLARE   (spawn_quoted_path)
 TEST_DECLARE   (spawn_tcp_server)
+TEST_DECLARE   (spawn_exercise_sigchld_issue)
 TEST_DECLARE   (fs_poll)
 TEST_DECLARE   (fs_poll_getpath)
 TEST_DECLARE   (fs_poll_close_request)
@@ -950,6 +951,7 @@ TASK_LIST_START
   TEST_ENTRY  (spawn_inherit_streams)
   TEST_ENTRY  (spawn_quoted_path)
   TEST_ENTRY  (spawn_tcp_server)
+  TEST_ENTRY  (spawn_exercise_sigchld_issue)
   TEST_ENTRY  (fs_poll)
   TEST_ENTRY  (fs_poll_getpath)
   TEST_ENTRY  (fs_poll_close_request)

test/test-spawn.c

@@ -1891,6 +1891,44 @@ TEST_IMPL(spawn_quoted_path) {
 #endif
 }
 
+TEST_IMPL(spawn_exercise_sigchld_issue) {
+  int r;
+  int i;
+  uv_process_options_t dummy_options = {0};
+  uv_process_t dummy_processes[100];
+  char* args[2];
+
+  init_process_options("spawn_helper1", exit_cb);
+
+  r = uv_spawn(uv_default_loop(), &process, &options);
+  ASSERT_EQ(r, 0);
+
+  // This test exercises a bug in the darwin kernel that causes SIGCHLD not to
+  // be delivered sometimes. Calling posix_spawn many times increases the
+  // likelihood of encountering this issue, so spin a few times to make this
+  // test more reliable.
+  dummy_options.file = args[0] = "program-that-had-better-not-exist";
+  args[1] = NULL;
+  dummy_options.args = args;
+  dummy_options.exit_cb = fail_cb;
+  dummy_options.flags = 0;
+  for (i = 0; i < 100; i++) {
+    r = uv_spawn(uv_default_loop(), &dummy_processes[i], &dummy_options);
+    if (r != UV_ENOENT)
+      ASSERT_EQ(r, UV_EACCES);
+    uv_close((uv_handle_t*) &dummy_processes[i], close_cb);
+  }
+
+  r = uv_run(uv_default_loop(), UV_RUN_DEFAULT);
+  ASSERT_EQ(r, 0);
+
+  ASSERT_EQ(exit_cb_called, 1);
+  ASSERT_EQ(close_cb_called, 101);
+
+  MAKE_VALGRIND_HAPPY();
+  return 0;
+}
+
 /* Helper for child process of spawn_inherit_streams */
 #ifndef _WIN32
 void spawn_stdin_stdout(void) {