Right now, the rtld prefork hook locks the rtld bind lock in the read

mode.  This allows the binder to be functional in the child after the
fork (assuming no lazy loading of a filter is needed), but other rtld
services which require write lock on rtld_bind_lock cause deadlock, if
called by child.

Change the _rtld_atfork() to lock the bind lock in write mode, making
the rtld fully functional after the fork.

Pre-resolve the symbols which are called by the libthr' fork()
interposer, since dynamic resolution causes deadlock due to the
rtld_bind_lock already owned in the write mode.

Reported and tested by:	pho
Sponsored by:	The FreeBSD Foundation
MFC after:	2 weeks

lib/libthr/thread/thr_fork.c

@@ -57,6 +57,7 @@
  *
  */
 
+#include <sys/syscall.h>
 #include "namespace.h"
 #include <errno.h>
 #include <link.h>
@@ -174,8 +175,15 @@ _fork(void)
 		was_threaded = 0;
 	}
 
-	/* Fork a new process: */
-	if ((ret = __sys_fork()) == 0) {
+	/*
+	 * Fork a new process.
+	 * There is no easy way to pre-resolve the __sys_fork symbol
+	 * without performing the fork.  Use the syscall(2)
+	 * indirection, the syscall symbol is resolved in
+	 * _thr_rtld_init() with side-effect free call.
+	 */
+	ret = syscall(SYS_fork);
+	if (ret == 0) {
 		/* Child process */
 		errsave = errno;
 		curthread->cancel_pending = 0;
@@ -250,6 +258,5 @@ _fork(void)
 	}
 	errno = errsave;
 
-	/* Return the process ID: */
 	return (ret);
 }

lib/libthr/thread/thr_rtld.c

@@ -32,10 +32,12 @@
   */
 #include <sys/cdefs.h>
 #include <sys/mman.h>
+#include <sys/syscall.h>
 #include <link.h>
 #include <stdlib.h>
 #include <string.h>
 
+#include "libc_private.h"
 #include "rtld_lock.h"
 #include "thr_private.h"
 
@@ -207,7 +209,24 @@ _thr_rtld_init(void)
 	li.thread_set_flag = _thr_rtld_set_flag;
 	li.thread_clr_flag = _thr_rtld_clr_flag;
 	li.at_fork = NULL;
-	
+
+	/*
+	 * Preresolve the symbols needed for the fork interposer.  We
+	 * call _rtld_atfork_pre() and _rtld_atfork_post() with NULL
+	 * argument to indicate that no actual locking inside the
+	 * functions should happen.  Neither rtld compat locks nor
+	 * libthr rtld locks cannot work there:
+	 * - compat locks do not handle the case of two locks taken
+	 *   in write mode (the signal mask for the thread is corrupted);
+	 * - libthr locks would work, but locked rtld_bind_lock prevents
+	 *   symbol resolution for _rtld_atfork_post.
+	 */
+	_rtld_atfork_pre(NULL);
+	_rtld_atfork_post(NULL);
+	_malloc_prefork();
+	_malloc_postfork();
+	syscall(SYS_getpid);
+
 	/* mask signals, also force to resolve __sys_sigprocmask PLT */
 	_thr_signal_block(curthread);
 	_rtld_thread_init(&li);

libexec/rtld-elf/rtld_lock.c

@@ -365,8 +365,19 @@ _rtld_atfork_pre(int *locks)
 {
 	RtldLockState ls[2];
 
+	if (locks == NULL)
+		return;
+
+	/*
+	 * Warning: this does not work with the rtld compat locks
+	 * above, since the thread signal mask is corrupted (set to
+	 * all signals blocked) if two locks are taken in write mode.
+	 * The caller of the _rtld_atfork_pre() must provide the
+	 * working implementation of the locks, and libthr locks are
+	 * fine.
+	 */
 	wlock_acquire(rtld_phdr_lock, &ls[0]);
-	rlock_acquire(rtld_bind_lock, &ls[1]);
+	wlock_acquire(rtld_bind_lock, &ls[1]);
 
 	/* XXXKIB: I am really sorry for this. */
 	locks[0] = ls[1].lockstate;
@@ -378,6 +389,9 @@ _rtld_atfork_post(int *locks)
 {
 	RtldLockState ls[2];
 
+	if (locks == NULL)
+		return;
+
 	bzero(ls, sizeof(ls));
 	ls[0].lockstate = locks[2];
 	ls[1].lockstate = locks[0];