net: do not use RLock around Accept

It might be non-blocking, but it also might be blocking.
Cannot take the chance, as Accept might block indefinitely
and make it impossible to acquire ForkLock exclusively
(during fork+exec).

Fixes #4737.

R=golang-dev, dave, iant, mikioh.mikioh
CC=golang-dev
https://golang.org/cl/7309050

src/pkg/net/fd_unix.go

@@ -661,6 +661,9 @@ func (fd *netFD) dup() (f *os.File, err error) {
 	syscall.ForkLock.RUnlock()
 
 	// We want blocking mode for the new fd, hence the double negative.
+	// This also puts the old fd into blocking mode, meaning that
+	// I/O will block the thread instead of letting us use the epoll server.
+	// Everything will still work, just with more threads.
 	if err = syscall.SetNonblock(ns, false); err != nil {
 		return nil, &OpError{"setnonblock", fd.net, fd.laddr, err}
 	}

src/pkg/net/sock_cloexec.go

@@ -50,14 +50,14 @@ func accept(fd int) (int, syscall.Sockaddr, error) {
 	}
 
 	// See ../syscall/exec_unix.go for description of ForkLock.
-	// It is okay to hold the lock across syscall.Accept
+	// It is probably okay to hold the lock across syscall.Accept
 	// because we have put fd.sysfd into non-blocking mode.
-	syscall.ForkLock.RLock()
+	// However, a call to the File method will put it back into
+	// blocking mode. We can't take that risk, so no use of ForkLock here.
 	nfd, sa, err = syscall.Accept(fd)
 	if err == nil {
 		syscall.CloseOnExec(nfd)
 	}
-	syscall.ForkLock.RUnlock()
 	if err != nil {
 		return -1, nil, err
 	}

src/pkg/net/sys_cloexec.go

@@ -35,14 +35,14 @@ func sysSocket(f, t, p int) (int, error) {
 // descriptor as nonblocking and close-on-exec.
 func accept(fd int) (int, syscall.Sockaddr, error) {
 	// See ../syscall/exec_unix.go for description of ForkLock.
-	// It is okay to hold the lock across syscall.Accept
+	// It is probably okay to hold the lock across syscall.Accept
 	// because we have put fd.sysfd into non-blocking mode.
-	syscall.ForkLock.RLock()
+	// However, a call to the File method will put it back into
+	// blocking mode. We can't take that risk, so no use of ForkLock here.
 	nfd, sa, err := syscall.Accept(fd)
 	if err == nil {
 		syscall.CloseOnExec(nfd)
 	}
-	syscall.ForkLock.RUnlock()
 	if err != nil {
 		return -1, nil, err
 	}