syscall: simplify closing of extra fds in plan9 StartProcess

Reviving earlier work by @ality in https://golang.org/cl/57890043
to make the closing of extra file descriptors in syscall.StartProcess
less race-prone. Instead of making a list of open fds in the parent
before forking, the child can read through the list of open fds and
close the ones not explicitly requested.  Also eliminate the
complication of keeping open any extra fds which were inherited by
the parent when it started.

This CL will be followed by one to eliminate the ForkLock in plan9,
which is now redundant.

Fixes #5605

Change-Id: I6b4b942001baa54248b656c52dced3b62021c486
Reviewed-on: https://go-review.googlesource.com/22610
Run-TryBot: David du Colombier <0intro@gmail.com>
Reviewed-by: David du Colombier <0intro@gmail.com>

src/syscall/exec_plan9.go

@@ -61,6 +61,41 @@ import (
 
 var ForkLock sync.RWMutex
 
+// gstringb reads a non-empty string from b, prefixed with a 16-bit length in little-endian order.
+// It returns the string as a byte slice, or nil if b is too short to contain the length or
+// the full string.
+//go:nosplit
+func gstringb(b []byte) []byte {
+	if len(b) < 2 {
+		return nil
+	}
+	n, b := gbit16(b)
+	if int(n) > len(b) {
+		return nil
+	}
+	return b[:n]
+}
+
+// Offset of the name field in a 9P directory entry - see UnmarshalDir() in dir_plan9.go
+const nameOffset = 39
+
+// gdirname returns the first filename from a buffer of directory entries,
+// and a slice containing the remaining directory entries.
+// If the buffer doesn't start with a valid directory entry, the returned name is nil.
+//go:nosplit
+func gdirname(buf []byte) (name []byte, rest []byte) {
+	if len(buf) < 2 {
+		return
+	}
+	size, buf := gbit16(buf)
+	if size < STATFIXLEN || int(size) > len(buf) {
+		return
+	}
+	name = gstringb(buf[nameOffset:size])
+	rest = buf[size:]
+	return
+}
+
 // StringSlicePtr converts a slice of strings to a slice of pointers
 // to NUL-terminated byte arrays. If any string contains a NUL byte
 // this function panics instead of returning an error.
@@ -104,20 +139,13 @@ func readdirnames(dirfd int) (names []string, err error) {
 		if n == 0 {
 			break
 		}
-		for i := 0; i < n; {
-			m, _ := gbit16(buf[i:])
-			m += 2
-
-			if m < STATFIXLEN {
+		for b := buf[:n]; len(b) > 0; {
+			var s []byte
+			s, b = gdirname(b)
+			if s == nil {
 				return nil, ErrBadStat
 			}
-
-			s, _, ok := gstring(buf[i+41:])
-			if !ok {
-				return nil, ErrBadStat
-			}
-			names = append(names, s)
-			i += int(m)
+			names = append(names, string(s))
 		}
 	}
 	return
@@ -152,16 +180,8 @@ func readdupdevice() (fds []int, err error) {
 	return
 }
 
-var startupFds []int
-
-// Plan 9 does not allow clearing the OCEXEC flag
-// from the underlying channel backing an open file descriptor,
-// therefore we store a list of already opened file descriptors
-// inside startupFds and skip them when manually closing descriptors
-// not meant to be passed to a child exec.
-func init() {
-	startupFds, _ = readdupdevice()
-}
+// name of the directory containing names and control files for all open file descriptors
+var dupdev, _ = BytePtrFromString("#d")
 
 // forkAndExecInChild forks the process, calling dup onto 0..len(fd)
 // and finally invoking exec(argv0, argvv, envv) in the child.
@@ -174,7 +194,7 @@ func init() {
 // The calls to RawSyscall are okay because they are assembly
 // functions that do not grow the stack.
 //go:norace
-func forkAndExecInChild(argv0 *byte, argv []*byte, envv []envItem, dir *byte, attr *ProcAttr, fdsToClose []int, pipe int, rflag int) (pid int, err error) {
+func forkAndExecInChild(argv0 *byte, argv []*byte, envv []envItem, dir *byte, attr *ProcAttr, pipe int, rflag int) (pid int, err error) {
 	// Declare all variables at top in case any
 	// declarations require heap allocation (e.g., errbuf).
 	var (
@@ -184,6 +204,8 @@ func forkAndExecInChild(argv0 *byte, argv []*byte, envv []envItem, dir *byte, at
 		clearenv int
 		envfd    int
 		errbuf   [ERRMAX]byte
+		statbuf  [STATMAX]byte
+		dupdevfd int
 	)
 
 	// Guard against side effects of shuffling fds below.
@@ -218,14 +240,39 @@ func forkAndExecInChild(argv0 *byte, argv []*byte, envv []envItem, dir *byte, at
 	// Fork succeeded, now in child.
 
 	// Close fds we don't need.
-	for i = 0; i < len(fdsToClose); i++ {
-		if fdsToClose[i] != pipe {
-			RawSyscall(SYS_CLOSE, uintptr(fdsToClose[i]), 0, 0)
+	r1, _, _ = RawSyscall(SYS_OPEN, uintptr(unsafe.Pointer(dupdev)), uintptr(O_RDONLY), 0)
+	dupdevfd = int(r1)
+	if dupdevfd == -1 {
+		goto childerror
+	}
+dirloop:
+	for {
+		r1, _, _ = RawSyscall6(SYS_PREAD, uintptr(dupdevfd), uintptr(unsafe.Pointer(&statbuf[0])), uintptr(len(statbuf)), ^uintptr(0), ^uintptr(0), 0)
+		n := int(r1)
+		switch n {
+		case -1:
+			goto childerror
+		case 0:
+			break dirloop
+		}
+		for b := statbuf[:n]; len(b) > 0; {
+			var s []byte
+			s, b = gdirname(b)
+			if s == nil {
+				copy(errbuf[:], ErrBadStat.Error())
+				goto childerror1
+			}
+			if s[len(s)-1] == 'l' {
+				// control file for descriptor <N> is named <N>ctl
+				continue
+			}
+			closeFdExcept(int(atoi(s)), pipe, dupdevfd, fd)
 		}
 	}
+	RawSyscall(SYS_CLOSE, uintptr(dupdevfd), 0, 0)
 
+	// Write new environment variables.
 	if envv != nil {
-		// Write new environment variables.
 		for i = 0; i < len(envv); i++ {
 			r1, _, _ = RawSyscall(SYS_CREATE, uintptr(unsafe.Pointer(envv[i].name)), uintptr(O_WRONLY), uintptr(0666))
 
@@ -313,6 +360,7 @@ func forkAndExecInChild(argv0 *byte, argv []*byte, envv []envItem, dir *byte, at
 childerror:
 	// send error string on pipe
 	RawSyscall(SYS_ERRSTR, uintptr(unsafe.Pointer(&errbuf[0])), uintptr(len(errbuf)), 0)
+childerror1:
 	errbuf[len(errbuf)-1] = 0
 	i = 0
 	for i < len(errbuf) && errbuf[i] != 0 {
@@ -332,6 +380,20 @@ childerror:
 	panic("unreached")
 }
 
+// close the numbered file descriptor, unless it is fd1, fd2, or a member of fds.
+//go:nosplit
+func closeFdExcept(n int, fd1 int, fd2 int, fds []int) {
+	if n == fd1 || n == fd2 {
+		return
+	}
+	for _, fd := range fds {
+		if n == fd {
+			return
+		}
+	}
+	RawSyscall(SYS_CLOSE, uintptr(n), 0, 0)
+}
+
 func cexecPipe(p []int) error {
 	e := Pipe(p)
 	if e != nil {
@@ -433,49 +495,15 @@ func forkExec(argv0 string, argv []string, attr *ProcAttr) (pid int, err error)
 	// Acquire the fork lock to prevent other threads from creating new fds before we fork.
 	ForkLock.Lock()
 
-	// get a list of open fds, excluding stdin,stdout and stderr that need to be closed in the child.
-	// no new fds can be created while we hold the ForkLock for writing.
-	openFds, e := readdupdevice()
-	if e != nil {
-		ForkLock.Unlock()
-		return 0, e
-	}
-
-	fdsToClose := make([]int, 0, len(openFds))
-	for _, fd := range openFds {
-		doClose := true
-
-		// exclude files opened at startup.
-		for _, sfd := range startupFds {
-			if fd == sfd {
-				doClose = false
-				break
-			}
-		}
-
-		// exclude files explicitly requested by the caller.
-		for _, rfd := range attr.Files {
-			if fd == int(rfd) {
-				doClose = false
-				break
-			}
-		}
-
-		if doClose {
-			fdsToClose = append(fdsToClose, fd)
-		}
-	}
-
 	// Allocate child status pipe close on exec.
-	e = cexecPipe(p[:])
+	e := cexecPipe(p[:])
 
 	if e != nil {
 		return 0, e
 	}
-	fdsToClose = append(fdsToClose, p[0])
 
 	// Kick off child.
-	pid, err = forkAndExecInChild(argv0p, argvp, envvParsed, dir, attr, fdsToClose, p[1], sys.Rfork)
+	pid, err = forkAndExecInChild(argv0p, argvp, envvParsed, dir, attr, p[1], sys.Rfork)
 
 	if err != nil {
 		if p[0] >= 0 {
@@ -493,8 +521,10 @@ func forkExec(argv0 string, argv []string, attr *ProcAttr) (pid int, err error)
 	Close(p[0])
 
 	if err != nil || n != 0 {
-		if n != 0 {
+		if n > 0 {
 			err = NewError(string(errbuf[:n]))
+		} else if err == nil {
+			err = NewError("failed to read exec status")
 		}
 
 		// Child failed; wait for it to exit, to make sure

src/syscall/syscall_plan9.go

@@ -56,6 +56,7 @@ func Syscall6(trap, a1, a2, a3, a4, a5, a6 uintptr) (r1, r2 uintptr, err ErrorSt
 func RawSyscall(trap, a1, a2, a3 uintptr) (r1, r2, err uintptr)
 func RawSyscall6(trap, a1, a2, a3, a4, a5, a6 uintptr) (r1, r2, err uintptr)
 
+//go:nosplit
 func atoi(b []byte) (n uint) {
 	n = 0
 	for i := 0; i < len(b); i++ {