Fix fork locking on win32 (#1141)

* Fix fork locking on win32

- __enable_threads / set __threaded in __proc_setup as threads are required for
  win32 subprocess management
- move mmi/fds locking out of pthread_atfork.c into fork.c so it's done anytime
  __threaded is set instead of being dependent of pthreads
- explicitly yoink _pthread_onfork_prepare, _pthread_onfork_parent, and
  _pthread_onfork_child in pthread_create.c so they are linked in in-case they
  are separated from _pthread_atfork

Big Thanks to @dfyz for help with locating the issue, testing, and devising a fix!

* fix child processes not being able to open files, initialize all necessary locks on fork

libc/intrin/pthreadlock.c

@@ -20,6 +20,10 @@
 
 pthread_spinlock_t _pthread_lock_obj;
 
+void _pthread_init(void) {
+  (void)pthread_spin_init(&_pthread_lock_obj, 0);
+}
+
 void _pthread_lock(void) {
   pthread_spin_lock(&_pthread_lock_obj);
 }

libc/proc/fork.c

@@ -18,6 +18,7 @@
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "libc/assert.h"
 #include "libc/calls/calls.h"
+#include "libc/calls/state.internal.h"
 #include "libc/calls/struct/sigset.h"
 #include "libc/calls/struct/sigset.internal.h"
 #include "libc/calls/syscall-nt.internal.h"
@@ -34,12 +35,45 @@
 #include "libc/nt/thread.h"
 #include "libc/proc/proc.internal.h"
 #include "libc/runtime/internal.h"
+#include "libc/runtime/memtrack.internal.h"
 #include "libc/runtime/runtime.h"
 #include "libc/runtime/syslib.internal.h"
 #include "libc/sysv/consts/sig.h"
 #include "libc/thread/posixthread.internal.h"
 #include "libc/thread/tls.h"
 
+static void _onfork_prepare(void) {
+  if (_weaken(_pthread_onfork_prepare)) {
+    _weaken(_pthread_onfork_prepare)();
+  }
+  _pthread_lock();
+  __fds_lock();
+  __mmi_lock();
+}
+
+static void _onfork_parent(void) {
+  __mmi_unlock();
+  __fds_unlock();
+  _pthread_unlock();
+  if (_weaken(_pthread_onfork_parent)) {
+    _weaken(_pthread_onfork_parent)();
+  }
+}
+
+static void _onfork_child(void) {
+  pthread_mutexattr_t attr;
+  pthread_mutexattr_init(&attr);
+  pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_RECURSIVE);
+  extern pthread_mutex_t __mmi_lock_obj;
+  pthread_mutex_init(&__mmi_lock_obj, &attr);
+  pthread_mutex_init(&__fds_lock_obj, &attr);
+  pthread_mutexattr_destroy(&attr);
+  _pthread_init();
+  if (_weaken(_pthread_onfork_child)) {
+    _weaken(_pthread_onfork_child)();
+  }
+}
+
 int _fork(uint32_t dwCreationFlags) {
   struct Dll *e;
   int ax, dx, tid, parent;
@@ -47,8 +81,8 @@ int _fork(uint32_t dwCreationFlags) {
   BLOCK_SIGNALS;
   if (IsWindows())
     __proc_lock();
-  if (__threaded && _weaken(_pthread_onfork_prepare)) {
-    _weaken(_pthread_onfork_prepare)();
+  if (__threaded) {
+    _onfork_prepare();
   }
   if (!IsWindows()) {
     ax = sys_fork();
@@ -99,14 +133,14 @@ int _fork(uint32_t dwCreationFlags) {
     atomic_store_explicit(&pt->pt_canceled, false, memory_order_relaxed);
 
     // run user fork callbacks
-    if (__threaded && _weaken(_pthread_onfork_child)) {
-      _weaken(_pthread_onfork_child)();
+    if (__threaded) {
+      _onfork_child();
     }
     STRACE("fork() → 0 (child of %d)", parent);
   } else {
     // this is the parent process
-    if (__threaded && _weaken(_pthread_onfork_parent)) {
-      _weaken(_pthread_onfork_parent)();
+    if (__threaded) {
+      _onfork_parent();
     }
     if (IsWindows())
       __proc_unlock();

libc/proc/proc.c

@@ -233,6 +233,7 @@ static textwindows dontinstrument uint32_t __proc_worker(void *arg) {
  * Lazy initializes process tracker data structures and worker.
  */
 static textwindows void __proc_setup(void) {
+  __enable_threads();
   __proc.onbirth = CreateEvent(0, 0, 0, 0);     // auto reset
   __proc.haszombies = CreateEvent(0, 1, 0, 0);  // manual reset
   __proc.thread = CreateThread(0, 65536, __proc_worker, 0,

libc/thread/posixthread.internal.h

@@ -105,6 +105,7 @@ intptr_t _pthread_syshand(struct PosixThread *) libcesque;
 long _pthread_cancel_ack(void) libcesque;
 void _pthread_decimate(void) libcesque;
 void _pthread_free(struct PosixThread *, bool) libcesque;
+void _pthread_init(void) libcesque;
 void _pthread_lock(void) libcesque;
 void _pthread_onfork_child(void) libcesque;
 void _pthread_onfork_parent(void) libcesque;

libc/thread/pthread_atfork.c

@@ -28,7 +28,6 @@
 #include "libc/macros.internal.h"
 #include "libc/mem/mem.h"
 #include "libc/proc/proc.internal.h"
-#include "libc/runtime/memtrack.internal.h"
 #include "libc/runtime/runtime.h"
 #include "libc/str/str.h"
 #include "libc/thread/posixthread.internal.h"
@@ -47,8 +46,6 @@ static struct AtForks {
   atomic_int allocated;
 } _atforks;
 
-extern pthread_spinlock_t _pthread_lock_obj;
-
 static void _pthread_onfork(int i) {
   struct AtFork *a;
   unassert(0 <= i && i <= 2);
@@ -65,27 +62,13 @@ static void _pthread_onfork(int i) {
 
 void _pthread_onfork_prepare(void) {
   _pthread_onfork(0);
-  _pthread_lock();
-  __fds_lock();
-  __mmi_lock();
 }
 
 void _pthread_onfork_parent(void) {
-  __mmi_unlock();
-  __fds_unlock();
-  _pthread_unlock();
   _pthread_onfork(1);
 }
 
 void _pthread_onfork_child(void) {
-  pthread_mutexattr_t attr;
-  pthread_mutexattr_init(&attr);
-  pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_RECURSIVE);
-  extern pthread_mutex_t __mmi_lock_obj;
-  pthread_mutex_init(&__mmi_lock_obj, &attr);
-  pthread_mutex_init(&__fds_lock_obj, &attr);
-  pthread_mutexattr_destroy(&attr);
-  (void)pthread_spin_init(&_pthread_lock_obj, 0);
   _pthread_onfork(2);
 }
 

libc/thread/pthread_create.c

@@ -60,6 +60,9 @@ __static_yoink("nsync_mu_trylock");
 __static_yoink("nsync_mu_rlock");
 __static_yoink("nsync_mu_runlock");
 __static_yoink("_pthread_atfork");
+__static_yoink("_pthread_onfork_prepare");
+__static_yoink("_pthread_onfork_parent");
+__static_yoink("_pthread_onfork_child");
 
 #define MAP_ANON_OPENBSD  0x1000
 #define MAP_STACK_OPENBSD 0x4000