Close extraneous file descriptors in containers

Without this patch, containers inherit the open file descriptors of the daemon, so my "exec 42>&2" allows us to "echo >&42 some nasty error with some bad advice" directly into the daemon log. :)

Also, "hack/dind" was already doing this due to issues caused by the inheritance, so I'm removing that hack too since this patch obsoletes it by generalizing it for all containers.

Docker-DCO-1.1-Signed-off-by: Andrew Page <admwiggin@gmail.com> (github: tianon)

libcontainer/nsinit/init.go

@@ -130,12 +130,16 @@ func setupNetwork(container *libcontainer.Container, context libcontainer.Contex
 	return nil
 }
 
-// finalizeNamespace drops the caps and sets the correct user
-// and working dir before execing the command inside the namespace
+// finalizeNamespace drops the caps, sets the correct user
+// and working dir, and closes any leaky file descriptors
+// before execing the command inside the namespace
 func finalizeNamespace(container *libcontainer.Container) error {
 	if err := capabilities.DropCapabilities(container); err != nil {
 		return fmt.Errorf("drop capabilities %s", err)
 	}
+	if err := system.CloseFdsFrom(3); err != nil {
+		return fmt.Errorf("close open file descriptors %s", err)
+	}
 	if err := setupUser(container); err != nil {
 		return fmt.Errorf("setup user %s", err)
 	}

system/fds_linux.go

@@ -0,0 +1,38 @@
+package system
+
+import (
+	"io/ioutil"
+	"strconv"
+	"syscall"
+)
+
+// Works similarly to OpenBSD's "closefrom(2)":
+//   The closefrom() call deletes all descriptors numbered fd and higher from
+//   the per-process file descriptor table.  It is effectively the same as
+//   calling close(2) on each descriptor.
+// http://www.openbsd.org/cgi-bin/man.cgi?query=closefrom&sektion=2
+//
+// See also http://stackoverflow.com/a/918469/433558
+func CloseFdsFrom(minFd int) error {
+	fdList, err := ioutil.ReadDir("/proc/self/fd")
+	if err != nil {
+		return err
+	}
+	for _, fi := range fdList {
+		fd, err := strconv.Atoi(fi.Name())
+		if err != nil {
+			// ignore non-numeric file names
+			continue
+		}
+
+		if fd < minFd {
+			// ignore descriptors lower than our specified minimum
+			continue
+		}
+
+		// intentionally ignore errors from syscall.Close
+		syscall.Close(fd)
+		// the cases where this might fail are basically file descriptors that have already been closed (including and especially the one that was created when ioutil.ReadDir did the "opendir" syscall)
+	}
+	return nil
+}

system/fds_unsupported.go

@@ -0,0 +1,12 @@
+// +build !linux
+
+package system
+
+import (
+	"fmt"
+	"runtime"
+)
+
+func CloseFdsFrom(minFd int) error {
+	return fmt.Errorf("CloseFdsFrom is unsupported on this platform (%s/%s)", runtime.GOOS, runtime.GOARCH)
+}