1 files changed, 45 insertions, 25 deletions

diff --git a/src/syscall/exec_linux.go b/src/syscall/exec_linux.go
index e631cd6470..db52360717 100644
--- a/src/syscall/exec_linux.go
+++ b/src/syscall/exec_linux.go
@@ -63,15 +63,46 @@ func runtime_AfterForkInChild()
 // functions that do not grow the stack.
 //go:norace
 func forkAndExecInChild(argv0 *byte, argv, envv []*byte, chroot, dir *byte, attr *ProcAttr, sys *SysProcAttr, pipe int) (pid int, err Errno) {
+	// Set up and fork. This returns immediately in the parent or
+	// if there's an error.
+	r1, err1, p, locked := forkAndExecInChild1(argv0, argv, envv, chroot, dir, attr, sys, pipe)
+	if locked {
+		runtime_AfterFork()
+	}
+	if err1 != 0 {
+		return 0, err1
+	}
+
+	// parent; return PID
+	pid = int(r1)
+
+	if sys.UidMappings != nil || sys.GidMappings != nil {
+		Close(p[0])
+		err := writeUidGidMappings(pid, sys)
+		var err2 Errno
+		if err != nil {
+			err2 = err.(Errno)
+		}
+		RawSyscall(SYS_WRITE, uintptr(p[1]), uintptr(unsafe.Pointer(&err2)), unsafe.Sizeof(err2))
+		Close(p[1])
+	}
+
+	return pid, 0
+}
+
+//go:norace
+func forkAndExecInChild1(argv0 *byte, argv, envv []*byte, chroot, dir *byte, attr *ProcAttr, sys *SysProcAttr, pipe int) (r1 uintptr, err1 Errno, p [2]int, locked bool) {
+	// vfork requires that the child not touch any of the parent's
+	// active stack frames. Hence, the child does all post-fork
+	// processing in this stack frame and never returns, while the
+	// parent returns immediately from this frame and does all
+	// post-fork processing in the outer frame.
 	// Declare all variables at top in case any
 	// declarations require heap allocation (e.g., err1).
 	var (
-		r1     uintptr
-		err1   Errno
 		err2   Errno
 		nextfd int
 		i      int
-		p      [2]int
 	)
 
 	// Record parent PID so child can test if it has died.
@@ -94,13 +125,15 @@ func forkAndExecInChild(argv0 *byte, argv, envv []*byte, chroot, dir *byte, attr
 	// synchronizing writing of User ID/Group ID mappings.
 	if sys.UidMappings != nil || sys.GidMappings != nil {
 		if err := forkExecPipe(p[:]); err != nil {
-			return 0, err.(Errno)
+			err1 = err.(Errno)
+			return
 		}
 	}
 
 	// About to call fork.
 	// No more allocation or calls of non-assembly functions.
 	runtime_BeforeFork()
+	locked = true
 	switch {
 	case runtime.GOARCH == "amd64" && sys.Cloneflags&CLONE_NEWUSER == 0:
 		r1, err1 = rawVforkSyscall(SYS_CLONE, uintptr(SIGCHLD|CLONE_VFORK|CLONE_VM)|sys.Cloneflags)
@@ -109,27 +142,14 @@ func forkAndExecInChild(argv0 *byte, argv, envv []*byte, chroot, dir *byte, attr
 	default:
 		r1, _, err1 = RawSyscall6(SYS_CLONE, uintptr(SIGCHLD)|sys.Cloneflags, 0, 0, 0, 0, 0)
 	}
-	if err1 != 0 {
-		runtime_AfterFork()
-		return 0, err1
-	}
-
-	if r1 != 0 {
-		// parent; return PID
-		runtime_AfterFork()
-		pid = int(r1)
-
-		if sys.UidMappings != nil || sys.GidMappings != nil {
-			Close(p[0])
-			err := writeUidGidMappings(pid, sys)
-			if err != nil {
-				err2 = err.(Errno)
-			}
-			RawSyscall(SYS_WRITE, uintptr(p[1]), uintptr(unsafe.Pointer(&err2)), unsafe.Sizeof(err2))
-			Close(p[1])
-		}
-
-		return pid, 0
+	if err1 != 0 || r1 != 0 {
+		// If we're in the parent, we must return immediately
+		// so we're not in the same stack frame as the child.
+		// This can at most use the return PC, which the child
+		// will not modify, and the results of
+		// rawVforkSyscall, which must have been written after
+		// the child was replaced.
+		return
 	}
 
 	// Fork succeeded, now in child.