Merge pull request #3651 from ianlewis:ip-forwarding

PiperOrigin-RevId: 332760843

4 files changed, 207 insertions, 2 deletions

diff --git a/pkg/sentry/fs/proc/BUILD b/pkg/sentry/fs/proc/BUILD
index 77c2c5c0e..b8b2281a8 100644
--- a/pkg/sentry/fs/proc/BUILD
+++ b/pkg/sentry/fs/proc/BUILD
@@ -50,6 +50,7 @@ go_library(
         "//pkg/sync",
         "//pkg/syserror",
         "//pkg/tcpip/header",
+        "//pkg/tcpip/network/ipv4",
         "//pkg/usermem",
         "//pkg/waiter",
     ],
diff --git a/pkg/sentry/fs/proc/sys_net.go b/pkg/sentry/fs/proc/sys_net.go
index 8615b60f0..e555672ad 100644
--- a/pkg/sentry/fs/proc/sys_net.go
+++ b/pkg/sentry/fs/proc/sys_net.go
@@ -26,6 +26,7 @@ import (
 	"gvisor.dev/gvisor/pkg/sentry/fs/ramfs"
 	"gvisor.dev/gvisor/pkg/sentry/inet"
 	"gvisor.dev/gvisor/pkg/sync"
+	"gvisor.dev/gvisor/pkg/tcpip/network/ipv4"
 	"gvisor.dev/gvisor/pkg/usermem"
 	"gvisor.dev/gvisor/pkg/waiter"
 )
@@ -54,7 +55,7 @@ type tcpMemInode struct {
 
 	// size stores the tcp buffer size during save, and sets the buffer
 	// size in netstack in restore. We must save/restore this here, since
-	// netstack itself is stateless.
+	// a netstack instance is created on restore.
 	size inet.TCPBufferSize
 
 	// mu protects against concurrent reads/writes to files based on this
@@ -258,6 +259,9 @@ func (f *tcpSackFile) Write(ctx context.Context, _ *fs.File, src usermem.IOSeque
 	if src.NumBytes() == 0 {
 		return 0, nil
 	}
+
+	// Only consider size of one memory page for input for performance reasons.
+	// We are only reading if it's zero or not anyway.
 	src = src.TakeFirst(usermem.PageSize - 1)
 
 	var v int32
@@ -383,11 +387,125 @@ func (p *proc) newSysNetCore(ctx context.Context, msrc *fs.MountSource, s inet.S
 	return newProcInode(ctx, d, msrc, fs.SpecialDirectory, nil)
 }
 
+// ipForwarding implements fs.InodeOperations.
+//
+// ipForwarding is used to enable/disable packet forwarding of netstack.
+//
+// +stateify savable
+type ipForwarding struct {
+	fsutil.SimpleFileInode
+
+	stack inet.Stack `state:"wait"`
+
+	// enabled stores the IPv4 forwarding state on save.
+	// We must save/restore this here, since a netstack instance
+	// is created on restore.
+	enabled *bool
+}
+
+func newIPForwardingInode(ctx context.Context, msrc *fs.MountSource, s inet.Stack) *fs.Inode {
+	ipf := &ipForwarding{
+		SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(0444), linux.PROC_SUPER_MAGIC),
+		stack:           s,
+	}
+	sattr := fs.StableAttr{
+		DeviceID:  device.ProcDevice.DeviceID(),
+		InodeID:   device.ProcDevice.NextIno(),
+		BlockSize: usermem.PageSize,
+		Type:      fs.SpecialFile,
+	}
+	return fs.NewInode(ctx, ipf, msrc, sattr)
+}
+
+// Truncate implements fs.InodeOperations.Truncate. Truncate is called when
+// O_TRUNC is specified for any kind of existing Dirent but is not called via
+// (f)truncate for proc files.
+func (*ipForwarding) Truncate(context.Context, *fs.Inode, int64) error {
+	return nil
+}
+
+// +stateify savable
+type ipForwardingFile struct {
+	fsutil.FileGenericSeek          `state:"nosave"`
+	fsutil.FileNoIoctl              `state:"nosave"`
+	fsutil.FileNoMMap               `state:"nosave"`
+	fsutil.FileNoSplice             `state:"nosave"`
+	fsutil.FileNoopFlush            `state:"nosave"`
+	fsutil.FileNoopFsync            `state:"nosave"`
+	fsutil.FileNoopRelease          `state:"nosave"`
+	fsutil.FileNotDirReaddir        `state:"nosave"`
+	fsutil.FileUseInodeUnstableAttr `state:"nosave"`
+	waiter.AlwaysReady              `state:"nosave"`
+
+	ipf *ipForwarding
+
+	stack inet.Stack `state:"wait"`
+}
+
+// GetFile implements fs.InodeOperations.GetFile.
+func (ipf *ipForwarding) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) {
+	flags.Pread = true
+	flags.Pwrite = true
+	return fs.NewFile(ctx, dirent, flags, &ipForwardingFile{
+		stack: ipf.stack,
+		ipf:   ipf,
+	}), nil
+}
+
+// Read implements fs.FileOperations.Read.
+func (f *ipForwardingFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) {
+	if offset != 0 {
+		return 0, io.EOF
+	}
+
+	if f.ipf.enabled == nil {
+		enabled := f.stack.Forwarding(ipv4.ProtocolNumber)
+		f.ipf.enabled = &enabled
+	}
+
+	val := "0\n"
+	if *f.ipf.enabled {
+		// Technically, this is not quite compatible with Linux. Linux
+		// stores these as an integer, so if you write "2" into
+		// ip_forward, you should get 2 back.
+		val = "1\n"
+	}
+	n, err := dst.CopyOut(ctx, []byte(val))
+	return int64(n), err
+}
+
+// Write implements fs.FileOperations.Write.
+//
+// Offset is ignored, multiple writes are not supported.
+func (f *ipForwardingFile) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, offset int64) (int64, error) {
+	if src.NumBytes() == 0 {
+		return 0, nil
+	}
+
+	// Only consider size of one memory page for input for performance reasons.
+	// We are only reading if it's zero or not anyway.
+	src = src.TakeFirst(usermem.PageSize - 1)
+
+	var v int32
+	n, err := usermem.CopyInt32StringInVec(ctx, src.IO, src.Addrs, &v, src.Opts)
+	if err != nil {
+		return n, err
+	}
+	if f.ipf.enabled == nil {
+		f.ipf.enabled = new(bool)
+	}
+	*f.ipf.enabled = v != 0
+	return n, f.stack.SetForwarding(ipv4.ProtocolNumber, *f.ipf.enabled)
+}
+
 func (p *proc) newSysNetIPv4Dir(ctx context.Context, msrc *fs.MountSource, s inet.Stack) *fs.Inode {
 	contents := map[string]*fs.Inode{
 		// Add tcp_sack.
 		"tcp_sack": newTCPSackInode(ctx, msrc, s),
 
+		// Add ip_forward.
+		"ip_forward": newIPForwardingInode(ctx, msrc, s),
+
 		// The following files are simple stubs until they are
 		// implemented in netstack, most of these files are
 		// configuration related. We use the value closest to the
diff --git a/pkg/sentry/fs/proc/sys_net_state.go b/pkg/sentry/fs/proc/sys_net_state.go
index 6eba709c6..4cb4741af 100644
--- a/pkg/sentry/fs/proc/sys_net_state.go
+++ b/pkg/sentry/fs/proc/sys_net_state.go
@@ -14,7 +14,11 @@
 
 package proc
 
-import "fmt"
+import (
+	"fmt"
+
+	"gvisor.dev/gvisor/pkg/tcpip/network/ipv4"
+)
 
 // beforeSave is invoked by stateify.
 func (t *tcpMemInode) beforeSave() {
@@ -40,3 +44,12 @@ func (s *tcpSack) afterLoad() {
 		}
 	}
 }
+
+// afterLoad is invoked by stateify.
+func (ipf *ipForwarding) afterLoad() {
+	if ipf.enabled != nil {
+		if err := ipf.stack.SetForwarding(ipv4.ProtocolNumber, *ipf.enabled); err != nil {
+			panic(fmt.Sprintf("failed to set IPv4 forwarding [%v]: %v", *ipf.enabled, err))
+		}
+	}
+}
diff --git a/pkg/sentry/fs/proc/sys_net_test.go b/pkg/sentry/fs/proc/sys_net_test.go
index 355e83d47..6ef5738e7 100644
--- a/pkg/sentry/fs/proc/sys_net_test.go
+++ b/pkg/sentry/fs/proc/sys_net_test.go
@@ -123,3 +123,76 @@ func TestConfigureRecvBufferSize(t *testing.T) {
 		}
 	}
 }
+
+// TestIPForwarding tests the implementation of
+// /proc/sys/net/ipv4/ip_forwarding
+func TestIPForwarding(t *testing.T) {
+	ctx := context.Background()
+	s := inet.NewTestStack()
+
+	var cases = []struct {
+		comment string
+		initial bool
+		str     string
+		final   bool
+	}{
+		{
+			comment: `Forwarding is disabled; write 1 and enable forwarding`,
+			initial: false,
+			str:     "1",
+			final:   true,
+		},
+		{
+			comment: `Forwarding is disabled; write 0 and disable forwarding`,
+			initial: false,
+			str:     "0",
+			final:   false,
+		},
+		{
+			comment: `Forwarding is enabled; write 1 and enable forwarding`,
+			initial: true,
+			str:     "1",
+			final:   true,
+		},
+		{
+			comment: `Forwarding is enabled; write 0 and disable forwarding`,
+			initial: true,
+			str:     "0",
+			final:   false,
+		},
+		{
+			comment: `Forwarding is disabled; write 2404 and enable forwarding`,
+			initial: false,
+			str:     "2404",
+			final:   true,
+		},
+		{
+			comment: `Forwarding is enabled; write 2404 and enable forwarding`,
+			initial: true,
+			str:     "2404",
+			final:   true,
+		},
+	}
+	for _, c := range cases {
+		t.Run(c.comment, func(t *testing.T) {
+			s.IPForwarding = c.initial
+			ipf := &ipForwarding{stack: s}
+			file := &ipForwardingFile{
+				stack: s,
+				ipf:   ipf,
+			}
+
+			// Write the values.
+			src := usermem.BytesIOSequence([]byte(c.str))
+			if n, err := file.Write(ctx, nil, src, 0); n != int64(len(c.str)) || err != nil {
+				t.Errorf("file.Write(ctx, nil, %q, 0) = (%d, %v); want (%d, nil)", c.str, n, err, len(c.str))
+			}
+
+			// Read the values from the stack and check them.
+			if got, want := s.IPForwarding, c.final; got != want {
+				t.Errorf("s.IPForwarding incorrect; got: %v, want: %v", got, want)
+			}
+
+		})
+	}
+}