rktio: always set signal mask to empty after fork

Saving and restoring the signal-mask state does not work right, since
rktio itself may block SIGCHLD in some cases, and it doesn't seem
useful/right to preserve the mask after fork.

pkgs/racket-test-core/tests/racket/subprocess.rktl

@@ -595,7 +595,10 @@
                         exe
                         (path->complete-path "unix_check.c" (or (current-load-relative-directory)
                                                                 (current-directory)))))
-      (test #t 'subprocess-state (system* exe)))
+      (test #t 'subprocess-state (let ([o (open-output-bytes)])
+                                   (or (parameterize ([current-output-port o])
+                                         (system* exe))
+                                       (get-output-bytes o)))))
     (delete-directory/files dir)))
 
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;

pkgs/racket-test-core/tests/racket/unix_check.c

@@ -1,5 +1,6 @@
 #include <stdio.h>
 #include <signal.h>
+#include <stdio.h>
 
 /* Make sure a child process is started with normal properties, such
    as no blocked signals */
@@ -11,10 +12,12 @@ int main()
   int i;
 
   /* SIGPROF tends to be near the end of the range of signal IDs */
-  for (i = 1; i < SIGPROF; i++) {
+  for (i = 1; i <= SIGPROF; i++) {
     if (sigaction(i, NULL, &sa) == 0) {
-      if (sa.sa_handler != SIG_DFL)
+      if (sa.sa_handler != SIG_DFL) {
+        printf("handler %d: %p\n", i, (void *)sa.sa_handler);
         return 1;
+      }
     } else {
       /* sigaction sometimes isn't allowed at all on SIGKILL,
          for example, so ignore that failure */
@@ -24,8 +27,22 @@ int main()
   sigemptyset(&set);
   sigprocmask(SIG_BLOCK, &set, &old_set);
 
-  return (sigismember(&old_set, SIGCHLD)
-          || sigismember(&old_set, SIGINT)
-          || sigismember(&old_set, SIGHUP)
-          || sigismember(&old_set, SIGQUIT));
+  if (sigismember(&old_set, SIGCHLD)) {
+    printf("masked: SIGCHLD\n");
+    return 1;
+  }
+  if (sigismember(&old_set, SIGINT)) {
+    printf("masked: SIGINT\n");
+    return 1;
+  }
+  if (sigismember(&old_set, SIGHUP)) {
+    printf("masked: SIGHUP\n");
+    return 1;
+  }
+  if (sigismember(&old_set, SIGQUIT)) {
+    printf("masked: SIGQUIT\n");
+    return 1;
+  }
+
+  return 0;
 }

racket/src/rktio/rktio.h

@@ -1004,7 +1004,9 @@ RKTIO_EXTERN void rktio_install_os_signal_handler(rktio_t *rktio);
    Ctl-C on Windows) to signal the handle of `rktio` and also records
    the signal for reporting via `rktio_poll_os_signal`. Only one
    `rktio` can be registered this way at a time. This function must
-   not be called in two threads at the same time. */
+   not be called in two threads at the same time; more generally, it
+   can only be called when `rktio_will_modify_os_signal_handler`
+   can be called for SIGINT, etc. */
 
 RKTIO_EXTERN_NOERR int rktio_poll_os_signal(rktio_t *rktio);
 /* Returns one of the following, not counting the last one: */
@@ -1021,11 +1023,11 @@ RKTIO_EXTERN void rktio_will_modify_os_signal_handler(int sig_id);
    about to be modified within the process but outside of rktio, where
    `sig_id` is a signal identifier --- such as SIGINT or SIGTERM. This
    notification allows rktio to record the current signal disposition
-   so that it can be restored after forking a new Unix process.
-   Signal registrations should happen only before multiple threads use
-   rktio, and registration of the signal can happen before any
-   `rktio_init` call. On the first `rktio_will_modify_os_signal_handler`
-   call, the signal mask is also recorded to be restored in a fork. */
+   so that it can be restored after forking a new Unix process. Signal
+   registrations should happen only before multiple threads use rktio,
+   and registration of the signal can happen before any `rktio_init`
+   call. After a signal is registered, trying to re-register it after
+   threads start is harmless. */
 
 /*************************************************/
 /* Time and date                                 */

racket/src/rktio/rktio_process.c

@@ -982,6 +982,10 @@ int rktio_process_init(rktio_t *rktio)
      it's a per-thread setting on Linux, and we want SIGCHLD blocked everywhere. */
   block_sigchld();
 
+  /* We'll set a handler later (possibly after other threads start),
+     so register SIGCHLD now: */
+  rktio_will_modify_os_signal_handler(SIGCHLD);
+
   centralized_start_child_signal_handler();
 #endif
 

racket/src/rktio/rktio_signal.c

@@ -127,18 +127,10 @@ typedef struct signal_handler_saved_disposition {
 } signal_handler_saved_disposition;
 
 static signal_handler_saved_disposition *saved_dispositions;
-#ifdef RKTIO_SYSTEM_UNIX
-static sigset_t initial_procmask;
-#endif
 
 void rktio_will_modify_os_signal_handler(int sig_id) {
   signal_handler_saved_disposition *saved;
 
-#ifdef RKTIO_SYSTEM_UNIX
-  if (saved_dispositions == NULL)
-    sigprocmask(SIG_SETMASK, NULL, &initial_procmask);
-#endif
-
   for (saved = saved_dispositions; saved; saved = saved->next)
     if (saved->sig_id == sig_id)
       return;
@@ -156,13 +148,13 @@ void rktio_will_modify_os_signal_handler(int sig_id) {
 #ifdef RKTIO_SYSTEM_UNIX
 /* called in a child thread after `fork */
 void rktio_restore_modified_signal_handlers() {
-  if (saved_dispositions) {
   signal_handler_saved_disposition *saved;
+  sigset_t set;
 
   for (saved = saved_dispositions; saved; saved = saved->next)
     sigaction(saved->sig_id, &saved->sa, NULL);
 
-    sigprocmask(SIG_SETMASK, &initial_procmask, NULL);
-  }
+  sigemptyset(&set);
+  sigprocmask(SIG_SETMASK, &set, NULL);
 }
 #endif