Add TTY support on VFS2 to runsc

Updates #1623, #1487

PiperOrigin-RevId: 309777922

pkg/sentry/control/BUILD

@@ -20,9 +20,11 @@ go_library(
         "//pkg/fd",
         "//pkg/fspath",
         "//pkg/log",
+        "//pkg/sentry/fdimport",
         "//pkg/sentry/fs",
         "//pkg/sentry/fs/host",
         "//pkg/sentry/fsbridge",
+        "//pkg/sentry/fsimpl/host",
         "//pkg/sentry/kernel",
         "//pkg/sentry/kernel/auth",
         "//pkg/sentry/kernel/time",
@@ -36,6 +38,7 @@ go_library(
         "//pkg/syserror",
         "//pkg/tcpip/link/sniffer",
         "//pkg/urpc",
+        "@org_golang_x_sys//unix:go_default_library",
     ],
 )
 

pkg/sentry/control/proc.go

@@ -24,13 +24,16 @@ import (
 	"text/tabwriter"
 	"time"
 
+	"golang.org/x/sys/unix"
 	"gvisor.dev/gvisor/pkg/abi/linux"
 	"gvisor.dev/gvisor/pkg/context"
 	"gvisor.dev/gvisor/pkg/fspath"
 	"gvisor.dev/gvisor/pkg/log"
+	"gvisor.dev/gvisor/pkg/sentry/fdimport"
 	"gvisor.dev/gvisor/pkg/sentry/fs"
 	"gvisor.dev/gvisor/pkg/sentry/fs/host"
 	"gvisor.dev/gvisor/pkg/sentry/fsbridge"
+	hostvfs2 "gvisor.dev/gvisor/pkg/sentry/fsimpl/host"
 	"gvisor.dev/gvisor/pkg/sentry/kernel"
 	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
 	ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time"
@@ -84,15 +87,13 @@ type ExecArgs struct {
 	// the root group if not set explicitly.
 	KGID auth.KGID
 
-	// ExtraKGIDs is the list of additional groups to which the user
-	// belongs.
+	// ExtraKGIDs is the list of additional groups to which the user belongs.
 	ExtraKGIDs []auth.KGID
 
 	// Capabilities is the list of capabilities to give to the process.
 	Capabilities *auth.TaskCapabilities
 
-	// StdioIsPty indicates that FDs 0, 1, and 2 are connected to a host
-	// pty FD.
+	// StdioIsPty indicates that FDs 0, 1, and 2 are connected to a host pty FD.
 	StdioIsPty bool
 
 	// FilePayload determines the files to give to the new process.
@@ -117,7 +118,7 @@ func (args ExecArgs) String() string {
 
 // Exec runs a new task.
 func (proc *Proc) Exec(args *ExecArgs, waitStatus *uint32) error {
-	newTG, _, _, err := proc.execAsync(args)
+	newTG, _, _, _, err := proc.execAsync(args)
 	if err != nil {
 		return err
 	}
@@ -130,26 +131,18 @@ func (proc *Proc) Exec(args *ExecArgs, waitStatus *uint32) error {
 
 // ExecAsync runs a new task, but doesn't wait for it to finish. It is defined
 // as a function rather than a method to avoid exposing execAsync as an RPC.
-func ExecAsync(proc *Proc, args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadID, *host.TTYFileOperations, error) {
+func ExecAsync(proc *Proc, args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadID, *host.TTYFileOperations, *hostvfs2.TTYFileDescription, error) {
 	return proc.execAsync(args)
 }
 
 // execAsync runs a new task, but doesn't wait for it to finish. It returns the
 // newly created thread group and its PID. If the stdio FDs are TTYs, then a
 // TTYFileOperations that wraps the TTY is also returned.
-func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadID, *host.TTYFileOperations, error) {
+func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadID, *host.TTYFileOperations, *hostvfs2.TTYFileDescription, error) {
 	// Import file descriptors.
 	fdTable := proc.Kernel.NewFDTable()
 	defer fdTable.DecRef()
 
-	// No matter what happens, we should close all files in the FilePayload
-	// before returning. Any files that are imported will be duped.
-	defer func() {
-		for _, f := range args.FilePayload.Files {
-			f.Close()
-		}
-	}()
-
 	creds := auth.NewUserCredentials(
 		args.KUID,
 		args.KGID,
@@ -202,7 +195,7 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI
 			vfsObj := proc.Kernel.VFS()
 			file, err := ResolveExecutablePath(ctx, vfsObj, initArgs.WorkingDirectory, initArgs.Argv[0], paths)
 			if err != nil {
-				return nil, 0, nil, fmt.Errorf("error finding executable %q in PATH %v: %v", initArgs.Argv[0], paths, err)
+				return nil, 0, nil, nil, fmt.Errorf("error finding executable %q in PATH %v: %v", initArgs.Argv[0], paths, err)
 			}
 			initArgs.File = fsbridge.NewVFSFile(file)
 		} else {
@@ -214,74 +207,57 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI
 
 				// initArgs must hold a reference on MountNamespace, which will
 				// be donated to the new process in CreateProcess.
-				initArgs.MountNamespaceVFS2.IncRef()
+				initArgs.MountNamespace.IncRef()
 			}
 			f, err := initArgs.MountNamespace.ResolveExecutablePath(ctx, initArgs.WorkingDirectory, initArgs.Argv[0], paths)
 			if err != nil {
-				return nil, 0, nil, fmt.Errorf("error finding executable %q in PATH %v: %v", initArgs.Argv[0], paths, err)
+				return nil, 0, nil, nil, fmt.Errorf("error finding executable %q in PATH %v: %v", initArgs.Argv[0], paths, err)
 			}
 			initArgs.Filename = f
 		}
 	}
 
-	// TODO(gvisor.dev/issue/1623): Use host FD when supported in VFS2.
-	var ttyFile *fs.File
-	for appFD, hostFile := range args.FilePayload.Files {
-		var appFile *fs.File
-
-		if args.StdioIsPty && appFD < 3 {
-			// Import the file as a host TTY file.
-			if ttyFile == nil {
-				var err error
-				appFile, err = host.ImportFile(ctx, int(hostFile.Fd()), true /* isTTY */)
-				if err != nil {
-					return nil, 0, nil, err
-				}
-				defer appFile.DecRef()
-
-				// Remember this in the TTY file, as we will
-				// use it for the other stdio FDs.
-				ttyFile = appFile
-			} else {
-				// Re-use the existing TTY file, as all three
-				// stdio FDs must point to the same fs.File in
-				// order to share TTY state, specifically the
-				// foreground process group id.
-				appFile = ttyFile
-			}
-		} else {
-			// Import the file as a regular host file.
-			var err error
-			appFile, err = host.ImportFile(ctx, int(hostFile.Fd()), false /* isTTY */)
+	fds := make([]int, len(args.FilePayload.Files))
+	for i, file := range args.FilePayload.Files {
+		if kernel.VFS2Enabled {
+			// Need to dup to remove ownership from os.File.
+			dup, err := unix.Dup(int(file.Fd()))
 			if err != nil {
-				return nil, 0, nil, err
+				return nil, 0, nil, nil, fmt.Errorf("duplicating payload files: %w", err)
 			}
-			defer appFile.DecRef()
+			fds[i] = dup
+		} else {
+			// VFS1 dups the file on import.
+			fds[i] = int(file.Fd())
 		}
-
-		// Add the file to the FD map.
-		if err := fdTable.NewFDAt(ctx, int32(appFD), appFile, kernel.FDFlags{}); err != nil {
-			return nil, 0, nil, err
+	}
+	ttyFile, ttyFileVFS2, err := fdimport.Import(ctx, fdTable, args.StdioIsPty, fds)
+	if err != nil {
+		if kernel.VFS2Enabled {
+			for _, fd := range fds {
+				unix.Close(fd)
+			}
 		}
+		return nil, 0, nil, nil, err
 	}
 
 	tg, tid, err := proc.Kernel.CreateProcess(initArgs)
 	if err != nil {
-		return nil, 0, nil, err
+		return nil, 0, nil, nil, err
 	}
 
-	var ttyFileOps *host.TTYFileOperations
-	if ttyFile != nil {
-		// Set the foreground process group on the TTY before starting
-		// the process.
-		ttyFileOps = ttyFile.FileOperations.(*host.TTYFileOperations)
-		ttyFileOps.InitForegroundProcessGroup(tg.ProcessGroup())
+	// Set the foreground process group on the TTY before starting the process.
+	switch {
+	case ttyFile != nil:
+		ttyFile.InitForegroundProcessGroup(tg.ProcessGroup())
+	case ttyFileVFS2 != nil:
+		ttyFileVFS2.InitForegroundProcessGroup(tg.ProcessGroup())
 	}
 
 	// Start the newly created process.
 	proc.Kernel.StartProcess(tg)
 
-	return tg, tid, ttyFileOps, nil
+	return tg, tid, ttyFile, ttyFileVFS2, nil
 }
 
 // PsArgs is the set of arguments to ps.

pkg/sentry/fdimport/BUILD

@@ -0,0 +1,19 @@
+load("//tools:defs.bzl", "go_library")
+
+package(licenses = ["notice"])
+
+go_library(
+    name = "fdimport",
+    srcs = [
+        "fdimport.go",
+    ],
+    visibility = ["//pkg/sentry:internal"],
+    deps = [
+        "//pkg/context",
+        "//pkg/sentry/fs",
+        "//pkg/sentry/fs/host",
+        "//pkg/sentry/fsimpl/host",
+        "//pkg/sentry/kernel",
+        "//pkg/sentry/vfs",
+    ],
+)

pkg/sentry/fdimport/fdimport.go

@@ -0,0 +1,129 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package fdimport
+
+import (
+	"gvisor.dev/gvisor/pkg/context"
+	"gvisor.dev/gvisor/pkg/sentry/fs"
+	"gvisor.dev/gvisor/pkg/sentry/fs/host"
+	hostvfs2 "gvisor.dev/gvisor/pkg/sentry/fsimpl/host"
+	"gvisor.dev/gvisor/pkg/sentry/kernel"
+	"gvisor.dev/gvisor/pkg/sentry/vfs"
+)
+
+// Import imports a slice of FDs into the given FDTable. If console is true,
+// sets up TTY for the first 3 FDs in the slice representing stdin, stdout,
+// stderr. Upon success, Import takes ownership of all FDs.
+func Import(ctx context.Context, fdTable *kernel.FDTable, console bool, fds []int) (*host.TTYFileOperations, *hostvfs2.TTYFileDescription, error) {
+	if kernel.VFS2Enabled {
+		ttyFile, err := importVFS2(ctx, fdTable, console, fds)
+		return nil, ttyFile, err
+	}
+	ttyFile, err := importFS(ctx, fdTable, console, fds)
+	return ttyFile, nil, err
+}
+
+func importFS(ctx context.Context, fdTable *kernel.FDTable, console bool, fds []int) (*host.TTYFileOperations, error) {
+	var ttyFile *fs.File
+	for appFD, hostFD := range fds {
+		var appFile *fs.File
+
+		if console && appFD < 3 {
+			// Import the file as a host TTY file.
+			if ttyFile == nil {
+				var err error
+				appFile, err = host.ImportFile(ctx, hostFD, true /* isTTY */)
+				if err != nil {
+					return nil, err
+				}
+				defer appFile.DecRef()
+
+				// Remember this in the TTY file, as we will
+				// use it for the other stdio FDs.
+				ttyFile = appFile
+			} else {
+				// Re-use the existing TTY file, as all three
+				// stdio FDs must point to the same fs.File in
+				// order to share TTY state, specifically the
+				// foreground process group id.
+				appFile = ttyFile
+			}
+		} else {
+			// Import the file as a regular host file.
+			var err error
+			appFile, err = host.ImportFile(ctx, hostFD, false /* isTTY */)
+			if err != nil {
+				return nil, err
+			}
+			defer appFile.DecRef()
+		}
+
+		// Add the file to the FD map.
+		if err := fdTable.NewFDAt(ctx, int32(appFD), appFile, kernel.FDFlags{}); err != nil {
+			return nil, err
+		}
+	}
+
+	if ttyFile == nil {
+		return nil, nil
+	}
+	return ttyFile.FileOperations.(*host.TTYFileOperations), nil
+}
+
+func importVFS2(ctx context.Context, fdTable *kernel.FDTable, console bool, stdioFDs []int) (*hostvfs2.TTYFileDescription, error) {
+	k := kernel.KernelFromContext(ctx)
+
+	var ttyFile *vfs.FileDescription
+	for appFD, hostFD := range stdioFDs {
+		var appFile *vfs.FileDescription
+
+		if console && appFD < 3 {
+			// Import the file as a host TTY file.
+			if ttyFile == nil {
+				var err error
+				appFile, err = hostvfs2.ImportFD(ctx, k.HostMount(), hostFD, true /* isTTY */)
+				if err != nil {
+					return nil, err
+				}
+				defer appFile.DecRef()
+
+				// Remember this in the TTY file, as we will use it for the other stdio
+				// FDs.
+				ttyFile = appFile
+			} else {
+				// Re-use the existing TTY file, as all three stdio FDs must point to
+				// the same fs.File in order to share TTY state, specifically the
+				// foreground process group id.
+				appFile = ttyFile
+			}
+		} else {
+			var err error
+			appFile, err = hostvfs2.ImportFD(ctx, k.HostMount(), hostFD, false /* isTTY */)
+			if err != nil {
+				return nil, err
+			}
+			defer appFile.DecRef()
+		}
+
+		if err := fdTable.NewFDAtVFS2(ctx, int32(appFD), appFile, kernel.FDFlags{}); err != nil {
+			return nil, err
+		}
+	}
+
+	if ttyFile == nil {
+		return nil, nil
+	}
+	return ttyFile.Impl().(*hostvfs2.TTYFileDescription), nil
+}

pkg/sentry/fsimpl/host/host.go

@@ -422,7 +422,7 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount) (*vfs.F
 	// TODO(gvisor.dev/issue/1672): Whitelist specific file types here, so that
 	// we don't allow importing arbitrary file types without proper support.
 	if i.isTTY {
-		fd := &ttyFD{
+		fd := &TTYFileDescription{
 			fileDescription: fileDescription{inode: i},
 			termios:         linux.DefaultSlaveTermios,
 		}

pkg/sentry/fsimpl/host/tty.go

@@ -26,15 +26,15 @@ import (
 	"gvisor.dev/gvisor/pkg/usermem"
 )
 
-// ttyFD implements vfs.FileDescriptionImpl for a host file descriptor
-// that wraps a TTY FD.
-type ttyFD struct {
+// TTYFileDescription implements vfs.FileDescriptionImpl for a host file
+// descriptor that wraps a TTY FD.
+type TTYFileDescription struct {
 	fileDescription
 
 	// mu protects the fields below.
 	mu sync.Mutex `state:"nosave"`
 
-	// session is the session attached to this ttyFD.
+	// session is the session attached to this TTYFileDescription.
 	session *kernel.Session
 
 	// fgProcessGroup is the foreground process group that is currently
@@ -48,7 +48,7 @@ type ttyFD struct {
 // InitForegroundProcessGroup sets the foreground process group and session for
 // the TTY. This should only be called once, after the foreground process group
 // has been created, but before it has started running.
-func (t *ttyFD) InitForegroundProcessGroup(pg *kernel.ProcessGroup) {
+func (t *TTYFileDescription) InitForegroundProcessGroup(pg *kernel.ProcessGroup) {
 	t.mu.Lock()
 	defer t.mu.Unlock()
 	if t.fgProcessGroup != nil {
@@ -59,14 +59,14 @@ func (t *ttyFD) InitForegroundProcessGroup(pg *kernel.ProcessGroup) {
 }
 
 // ForegroundProcessGroup returns the foreground process for the TTY.
-func (t *ttyFD) ForegroundProcessGroup() *kernel.ProcessGroup {
+func (t *TTYFileDescription) ForegroundProcessGroup() *kernel.ProcessGroup {
 	t.mu.Lock()
 	defer t.mu.Unlock()
 	return t.fgProcessGroup
 }
 
 // Release implements fs.FileOperations.Release.
-func (t *ttyFD) Release() {
+func (t *TTYFileDescription) Release() {
 	t.mu.Lock()
 	t.fgProcessGroup = nil
 	t.mu.Unlock()
@@ -78,7 +78,7 @@ func (t *ttyFD) Release() {
 //
 // Reading from a TTY is only allowed for foreground process groups. Background
 // process groups will either get EIO or a SIGTTIN.
-func (t *ttyFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) {
+func (t *TTYFileDescription) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) {
 	t.mu.Lock()
 	defer t.mu.Unlock()
 
@@ -96,7 +96,7 @@ func (t *ttyFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64,
 //
 // Reading from a TTY is only allowed for foreground process groups. Background
 // process groups will either get EIO or a SIGTTIN.
-func (t *ttyFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) {
+func (t *TTYFileDescription) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) {
 	t.mu.Lock()
 	defer t.mu.Unlock()
 
@@ -111,7 +111,7 @@ func (t *ttyFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadO
 }
 
 // PWrite implements vfs.FileDescriptionImpl.
-func (t *ttyFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) {
+func (t *TTYFileDescription) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) {
 	t.mu.Lock()
 	defer t.mu.Unlock()
 
@@ -126,7 +126,7 @@ func (t *ttyFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64
 }
 
 // Write implements vfs.FileDescriptionImpl.
-func (t *ttyFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) {
+func (t *TTYFileDescription) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) {
 	t.mu.Lock()
 	defer t.mu.Unlock()
 
@@ -141,7 +141,7 @@ func (t *ttyFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.Writ
 }
 
 // Ioctl implements vfs.FileDescriptionImpl.
-func (t *ttyFD) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) {
+func (t *TTYFileDescription) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) {
 	// Ignore arg[0]. This is the real FD:
 	fd := t.inode.hostFD
 	ioctl := args[1].Uint64()
@@ -321,7 +321,7 @@ func (t *ttyFD) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArgum
 // is a bit convoluted, but documented inline.
 //
 // Preconditions: t.mu must be held.
-func (t *ttyFD) checkChange(ctx context.Context, sig linux.Signal) error {
+func (t *TTYFileDescription) checkChange(ctx context.Context, sig linux.Signal) error {
 	task := kernel.TaskFromContext(ctx)
 	if task == nil {
 		// No task? Linux does not have an analog for this case, but

pkg/sentry/vfs/context.go

@@ -49,3 +49,27 @@ func RootFromContext(ctx context.Context) VirtualDentry {
 	}
 	return VirtualDentry{}
 }
+
+type rootContext struct {
+	context.Context
+	root VirtualDentry
+}
+
+// WithRoot returns a copy of ctx with the given root.
+func WithRoot(ctx context.Context, root VirtualDentry) context.Context {
+	return &rootContext{
+		Context: ctx,
+		root:    root,
+	}
+}
+
+// Value implements Context.Value.
+func (rc rootContext) Value(key interface{}) interface{} {
+	switch key {
+	case CtxRoot:
+		rc.root.IncRef()
+		return rc.root
+	default:
+		return rc.Context.Value(key)
+	}
+}

pkg/sentry/vfs/epoll.go

@@ -192,6 +192,7 @@ func (ep *EpollInstance) AddInterest(file *FileDescription, num int32, event lin
 		mask:     mask,
 		userData: event.Data,
 	}
+	epi.waiter.Callback = epi
 	ep.interest[key] = epi
 	wmask := waiter.EventMaskFromLinux(mask)
 	file.EventRegister(&epi.waiter, wmask)

runsc/boot/BUILD

@@ -12,7 +12,6 @@ go_library(
         "controller.go",
         "debug.go",
         "events.go",
-        "fds.go",
         "fs.go",
         "limits.go",
         "loader.go",
@@ -43,6 +42,7 @@ go_library(
         "//pkg/sentry/arch:registers_go_proto",
         "//pkg/sentry/control",
         "//pkg/sentry/devices/memdev",
+        "//pkg/sentry/fdimport",
         "//pkg/sentry/fs",
         "//pkg/sentry/fs/dev",
         "//pkg/sentry/fs/gofer",
@@ -53,6 +53,7 @@ go_library(
         "//pkg/sentry/fs/tmpfs",
         "//pkg/sentry/fs/tty",
         "//pkg/sentry/fs/user",
+        "//pkg/sentry/fsimpl/devpts",
         "//pkg/sentry/fsimpl/devtmpfs",
         "//pkg/sentry/fsimpl/gofer",
         "//pkg/sentry/fsimpl/host",

runsc/boot/fds.go

@@ -1,108 +0,0 @@
-// Copyright 2018 The gVisor Authors.
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-package boot
-
-import (
-	"fmt"
-
-	"gvisor.dev/gvisor/pkg/context"
-	"gvisor.dev/gvisor/pkg/sentry/fs"
-	"gvisor.dev/gvisor/pkg/sentry/fs/host"
-	vfshost "gvisor.dev/gvisor/pkg/sentry/fsimpl/host"
-	"gvisor.dev/gvisor/pkg/sentry/kernel"
-)
-
-// createFDTable creates an FD table that contains stdin, stdout, and stderr.
-// If console is true, then ioctl calls will be passed through to the host FD.
-// Upon success, createFDMap dups then closes stdioFDs.
-func createFDTable(ctx context.Context, console bool, stdioFDs []int) (*kernel.FDTable, error) {
-	if len(stdioFDs) != 3 {
-		return nil, fmt.Errorf("stdioFDs should contain exactly 3 FDs (stdin, stdout, and stderr), but %d FDs received", len(stdioFDs))
-	}
-
-	if kernel.VFS2Enabled {
-		return createFDTableVFS2(ctx, console, stdioFDs)
-	}
-
-	k := kernel.KernelFromContext(ctx)
-	fdTable := k.NewFDTable()
-	defer fdTable.DecRef()
-
-	var ttyFile *fs.File
-	for appFD, hostFD := range stdioFDs {
-		var appFile *fs.File
-
-		if console && appFD < 3 {
-			// Import the file as a host TTY file.
-			if ttyFile == nil {
-				var err error
-				appFile, err = host.ImportFile(ctx, hostFD, true /* isTTY */)
-				if err != nil {
-					return nil, err
-				}
-				defer appFile.DecRef()
-
-				// Remember this in the TTY file, as we will
-				// use it for the other stdio FDs.
-				ttyFile = appFile
-			} else {
-				// Re-use the existing TTY file, as all three
-				// stdio FDs must point to the same fs.File in
-				// order to share TTY state, specifically the
-				// foreground process group id.
-				appFile = ttyFile
-			}
-		} else {
-			// Import the file as a regular host file.
-			var err error
-			appFile, err = host.ImportFile(ctx, hostFD, false /* isTTY */)
-			if err != nil {
-				return nil, err
-			}
-			defer appFile.DecRef()
-		}
-
-		// Add the file to the FD map.
-		if err := fdTable.NewFDAt(ctx, int32(appFD), appFile, kernel.FDFlags{}); err != nil {
-			return nil, err
-		}
-	}
-
-	fdTable.IncRef()
-	return fdTable, nil
-}
-
-func createFDTableVFS2(ctx context.Context, console bool, stdioFDs []int) (*kernel.FDTable, error) {
-	k := kernel.KernelFromContext(ctx)
-	fdTable := k.NewFDTable()
-	defer fdTable.DecRef()
-
-	for appFD, hostFD := range stdioFDs {
-		// TODO(gvisor.dev/issue/1482): Add TTY support.
-		appFile, err := vfshost.ImportFD(ctx, k.HostMount(), hostFD, false)
-		if err != nil {
-			return nil, err
-		}
-
-		if err := fdTable.NewFDAtVFS2(ctx, int32(appFD), appFile, kernel.FDFlags{}); err != nil {
-			appFile.DecRef()
-			return nil, err
-		}
-		appFile.DecRef()
-	}
-
-	fdTable.IncRef()
-	return fdTable, nil
-}

runsc/boot/loader.go

@@ -27,16 +27,18 @@ import (
 	specs "github.com/opencontainers/runtime-spec/specs-go"
 	"golang.org/x/sys/unix"
 	"gvisor.dev/gvisor/pkg/abi/linux"
+	"gvisor.dev/gvisor/pkg/context"
 	"gvisor.dev/gvisor/pkg/cpuid"
 	"gvisor.dev/gvisor/pkg/log"
 	"gvisor.dev/gvisor/pkg/memutil"
 	"gvisor.dev/gvisor/pkg/rand"
 	"gvisor.dev/gvisor/pkg/sentry/arch"
 	"gvisor.dev/gvisor/pkg/sentry/control"
+	"gvisor.dev/gvisor/pkg/sentry/fdimport"
 	"gvisor.dev/gvisor/pkg/sentry/fs"
 	"gvisor.dev/gvisor/pkg/sentry/fs/host"
 	"gvisor.dev/gvisor/pkg/sentry/fs/user"
-	vfs2host "gvisor.dev/gvisor/pkg/sentry/fsimpl/host"
+	hostvfs2 "gvisor.dev/gvisor/pkg/sentry/fsimpl/host"
 	"gvisor.dev/gvisor/pkg/sentry/inet"
 	"gvisor.dev/gvisor/pkg/sentry/kernel"
 	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
@@ -143,6 +145,9 @@ type execProcess struct {
 	// tty will be nil if the process is not attached to a terminal.
 	tty *host.TTYFileOperations
 
+	// tty will be nil if the process is not attached to a terminal.
+	ttyVFS2 *hostvfs2.TTYFileDescription
+
 	// pidnsPath is the pid namespace path in spec
 	pidnsPath string
 }
@@ -333,7 +338,7 @@ func New(args Args) (*Loader, error) {
 
 	if kernel.VFS2Enabled {
 		// Set up host mount that will be used for imported fds.
-		hostFilesystem := vfs2host.NewFilesystem(k.VFS())
+		hostFilesystem := hostvfs2.NewFilesystem(k.VFS())
 		defer hostFilesystem.DecRef()
 		hostMount, err := k.VFS().NewDisconnectedMount(hostFilesystem, nil, &vfs.MountOptions{})
 		if err != nil {
@@ -528,6 +533,8 @@ func (l *Loader) run() error {
 
 	// If we are restoring, we do not want to create a process.
 	// l.restore is set by the container manager when a restore call is made.
+	var ttyFile *host.TTYFileOperations
+	var ttyFileVFS2 *hostvfs2.TTYFileDescription
 	if !l.restore {
 		if l.conf.ProfileEnable {
 			pprof.Initialize()
@@ -542,13 +549,14 @@ func (l *Loader) run() error {
 		// Create the FD map, which will set stdin, stdout, and stderr.  If console
 		// is true, then ioctl calls will be passed through to the host fd.
 		ctx := l.rootProcArgs.NewContext(l.k)
-		fdTable, err := createFDTable(ctx, l.console, l.stdioFDs)
+		var err error
+
+		// CreateProcess takes a reference on FDMap if successful. We won't need
+		// ours either way.
+		l.rootProcArgs.FDTable, ttyFile, ttyFileVFS2, err = createFDTable(ctx, l.console, l.stdioFDs)
 		if err != nil {
 			return fmt.Errorf("importing fds: %v", err)
 		}
-		// CreateProcess takes a reference on FDMap if successful. We won't need
-		// ours either way.
-		l.rootProcArgs.FDTable = fdTable
 
 		// Setup the root container file system.
 		l.startGoferMonitor(l.sandboxID, l.goferFDs)
@@ -591,14 +599,16 @@ func (l *Loader) run() error {
 		ep.pidnsPath = ns.Path
 	}
 	if l.console {
-		ttyFile, _ := l.rootProcArgs.FDTable.Get(0)
-		defer ttyFile.DecRef()
-		ep.tty = ttyFile.FileOperations.(*host.TTYFileOperations)
-
-		// Set the foreground process group on the TTY to the global
-		// init process group, since that is what we are about to
-		// start running.
-		ep.tty.InitForegroundProcessGroup(ep.tg.ProcessGroup())
+		// Set the foreground process group on the TTY to the global init process
+		// group, since that is what we are about to start running.
+		switch {
+		case ttyFileVFS2 != nil:
+			ep.ttyVFS2 = ttyFileVFS2
+			ttyFileVFS2.InitForegroundProcessGroup(ep.tg.ProcessGroup())
+		case ttyFile != nil:
+			ep.tty = ttyFile
+			ttyFile.InitForegroundProcessGroup(ep.tg.ProcessGroup())
+		}
 	}
 
 	// Handle signals by forwarding them to the root container process
@@ -719,7 +729,7 @@ func (l *Loader) startContainer(spec *specs.Spec, conf *Config, cid string, file
 
 	// Create the FD map, which will set stdin, stdout, and stderr.
 	ctx := procArgs.NewContext(l.k)
-	fdTable, err := createFDTable(ctx, false, stdioFDs)
+	fdTable, _, _, err := createFDTable(ctx, false, stdioFDs)
 	if err != nil {
 		return fmt.Errorf("importing fds: %v", err)
 	}
@@ -804,14 +814,14 @@ func (l *Loader) destroyContainer(cid string) error {
 	l.mu.Lock()
 	defer l.mu.Unlock()
 
-	_, _, started, err := l.threadGroupFromIDLocked(execID{cid: cid})
+	tg, err := l.tryThreadGroupFromIDLocked(execID{cid: cid})
 	if err != nil {
 		// Container doesn't exist.
 		return err
 	}
 
-	// The container exists, has it been started?
-	if started {
+	// The container exists, but has it been started?
+	if tg != nil {
 		if err := l.signalAllProcesses(cid, int32(linux.SIGKILL)); err != nil {
 			return fmt.Errorf("sending SIGKILL to all container processes: %v", err)
 		}
@@ -853,48 +863,65 @@ func (l *Loader) executeAsync(args *control.ExecArgs) (kernel.ThreadID, error) {
 	l.mu.Lock()
 	defer l.mu.Unlock()
 
-	tg, _, started, err := l.threadGroupFromIDLocked(execID{cid: args.ContainerID})
+	tg, err := l.tryThreadGroupFromIDLocked(execID{cid: args.ContainerID})
 	if err != nil {
 		return 0, err
 	}
-	if !started {
+	if tg == nil {
 		return 0, fmt.Errorf("container %q not started", args.ContainerID)
 	}
 
-	// TODO(gvisor.dev/issue/1623): Add VFS2 support
-
 	// Get the container MountNamespace from the Task.
-	tg.Leader().WithMuLocked(func(t *kernel.Task) {
+	if kernel.VFS2Enabled {
 		// task.MountNamespace() does not take a ref, so we must do so ourselves.
-		args.MountNamespace = t.MountNamespace()
-		args.MountNamespace.IncRef()
-	})
-	if args.MountNamespace != nil {
-		defer args.MountNamespace.DecRef()
+		args.MountNamespaceVFS2 = tg.Leader().MountNamespaceVFS2()
+		args.MountNamespaceVFS2.IncRef()
+	} else {
+		tg.Leader().WithMuLocked(func(t *kernel.Task) {
+			// task.MountNamespace() does not take a ref, so we must do so ourselves.
+			args.MountNamespace = t.MountNamespace()
+			args.MountNamespace.IncRef()
+		})
 	}
 
 	// Add the HOME environment variable if it is not already set.
-	root := args.MountNamespace.Root()
-	defer root.DecRef()
-	ctx := fs.WithRoot(l.k.SupervisorContext(), root)
-	envv, err := user.MaybeAddExecUserHome(ctx, args.MountNamespace, args.KUID, args.Envv)
-	if err != nil {
-		return 0, err
+	if kernel.VFS2Enabled {
+		defer args.MountNamespaceVFS2.DecRef()
+
+		root := args.MountNamespaceVFS2.Root()
+		defer root.DecRef()
+		ctx := vfs.WithRoot(l.k.SupervisorContext(), root)
+		envv, err := user.MaybeAddExecUserHomeVFS2(ctx, args.MountNamespaceVFS2, args.KUID, args.Envv)
+		if err != nil {
+			return 0, err
+		}
+		args.Envv = envv
+	} else {
+		defer args.MountNamespace.DecRef()
+
+		root := args.MountNamespace.Root()
+		defer root.DecRef()
+		ctx := fs.WithRoot(l.k.SupervisorContext(), root)
+		envv, err := user.MaybeAddExecUserHome(ctx, args.MountNamespace, args.KUID, args.Envv)
+		if err != nil {
+			return 0, err
+		}
+		args.Envv = envv
 	}
-	args.Envv = envv
 
 	// Start the process.
 	proc := control.Proc{Kernel: l.k}
 	args.PIDNamespace = tg.PIDNamespace()
-	newTG, tgid, ttyFile, err := control.ExecAsync(&proc, args)
+	newTG, tgid, ttyFile, ttyFileVFS2, err := control.ExecAsync(&proc, args)
 	if err != nil {
 		return 0, err
 	}
 
 	eid := execID{cid: args.ContainerID, pid: tgid}
 	l.processes[eid] = &execProcess{
-		tg:  newTG,
-		tty: ttyFile,
+		tg:      newTG,
+		tty:     ttyFile,
+		ttyVFS2: ttyFileVFS2,
 	}
 	log.Debugf("updated processes: %v", l.processes)
 
@@ -905,7 +932,7 @@ func (l *Loader) executeAsync(args *control.ExecArgs) (kernel.ThreadID, error) {
 func (l *Loader) waitContainer(cid string, waitStatus *uint32) error {
 	// Don't defer unlock, as doing so would make it impossible for
 	// multiple clients to wait on the same container.
-	tg, _, err := l.threadGroupFromID(execID{cid: cid})
+	tg, err := l.threadGroupFromID(execID{cid: cid})
 	if err != nil {
 		return fmt.Errorf("can't wait for container %q: %v", cid, err)
 	}
@@ -924,7 +951,7 @@ func (l *Loader) waitPID(tgid kernel.ThreadID, cid string, waitStatus *uint32) e
 
 	// Try to find a process that was exec'd
 	eid := execID{cid: cid, pid: tgid}
-	execTG, _, err := l.threadGroupFromID(eid)
+	execTG, err := l.threadGroupFromID(eid)
 	if err == nil {
 		ws := l.wait(execTG)
 		*waitStatus = ws
@@ -938,7 +965,7 @@ func (l *Loader) waitPID(tgid kernel.ThreadID, cid string, waitStatus *uint32) e
 
 	// The caller may be waiting on a process not started directly via exec.
 	// In this case, find the process in the container's PID namespace.
-	initTG, _, err := l.threadGroupFromID(execID{cid: cid})
+	initTG, err := l.threadGroupFromID(execID{cid: cid})
 	if err != nil {
 		return fmt.Errorf("waiting for PID %d: %v", tgid, err)
 	}
@@ -1089,8 +1116,7 @@ func (l *Loader) signal(cid string, pid, signo int32, mode SignalDeliveryMode) e
 			return fmt.Errorf("PID (%d) cannot be set when signaling all processes", pid)
 		}
 		// Check that the container has actually started before signaling it.
-		_, _, err := l.threadGroupFromID(execID{cid: cid})
-		if err != nil {
+		if _, err := l.threadGroupFromID(execID{cid: cid}); err != nil {
 			return err
 		}
 		if err := l.signalAllProcesses(cid, signo); err != nil {
@@ -1104,7 +1130,7 @@ func (l *Loader) signal(cid string, pid, signo int32, mode SignalDeliveryMode) e
 }
 
 func (l *Loader) signalProcess(cid string, tgid kernel.ThreadID, signo int32) error {
-	execTG, _, err := l.threadGroupFromID(execID{cid: cid, pid: tgid})
+	execTG, err := l.threadGroupFromID(execID{cid: cid, pid: tgid})
 	if err == nil {
 		// Send signal directly to the identified process.
 		return l.k.SendExternalSignalThreadGroup(execTG, &arch.SignalInfo{Signo: signo})
@@ -1113,7 +1139,7 @@ func (l *Loader) signalProcess(cid string, tgid kernel.ThreadID, signo int32) er
 	// The caller may be signaling a process not started directly via exec.
 	// In this case, find the process in the container's PID namespace and
 	// signal it.
-	initTG, _, err := l.threadGroupFromID(execID{cid: cid})
+	initTG, err := l.threadGroupFromID(execID{cid: cid})
 	if err != nil {
 		return fmt.Errorf("no thread group found: %v", err)
 	}
@@ -1127,17 +1153,35 @@ func (l *Loader) signalProcess(cid string, tgid kernel.ThreadID, signo int32) er
 	return l.k.SendExternalSignalThreadGroup(tg, &arch.SignalInfo{Signo: signo})
 }
 
+// signalForegrondProcessGroup looks up foreground process group from the TTY
+// for the given "tgid" inside container "cid", and send the signal to it.
 func (l *Loader) signalForegrondProcessGroup(cid string, tgid kernel.ThreadID, signo int32) error {
-	// Lookup foreground process group from the TTY for the given process,
-	// and send the signal to it.
-	tg, tty, err := l.threadGroupFromID(execID{cid: cid, pid: tgid})
+	l.mu.Lock()
+	tg, err := l.tryThreadGroupFromIDLocked(execID{cid: cid, pid: tgid})
+	if err != nil {
+		l.mu.Unlock()
+		return fmt.Errorf("no thread group found: %v", err)
+	}
+	if tg == nil {
+		l.mu.Unlock()
+		return fmt.Errorf("container %q not started", cid)
+	}
+
+	tty, ttyVFS2, err := l.ttyFromIDLocked(execID{cid: cid, pid: tgid})
+	l.mu.Unlock()
 	if err != nil {
 		return fmt.Errorf("no thread group found: %v", err)
 	}
-	if tty == nil {
+
+	var pg *kernel.ProcessGroup
+	switch {
+	case ttyVFS2 != nil:
+		pg = ttyVFS2.ForegroundProcessGroup()
+	case tty != nil:
+		pg = tty.ForegroundProcessGroup()
+	default:
 		return fmt.Errorf("no TTY attached")
 	}
-	pg := tty.ForegroundProcessGroup()
 	if pg == nil {
 		// No foreground process group has been set. Signal the
 		// original thread group.
@@ -1168,33 +1212,57 @@ func (l *Loader) signalAllProcesses(cid string, signo int32) error {
 	return l.k.SendContainerSignal(cid, &arch.SignalInfo{Signo: signo})
 }
 
-// threadGroupFromID same as threadGroupFromIDLocked except that it acquires
-// mutex before calling it.
-func (l *Loader) threadGroupFromID(key execID) (*kernel.ThreadGroup, *host.TTYFileOperations, error) {
+// threadGroupFromID is similar to tryThreadGroupFromIDLocked except that it
+// acquires mutex before calling it and fails in case container hasn't started
+// yet.
+func (l *Loader) threadGroupFromID(key execID) (*kernel.ThreadGroup, error) {
 	l.mu.Lock()
 	defer l.mu.Unlock()
-	tg, tty, ok, err := l.threadGroupFromIDLocked(key)
+	tg, err := l.tryThreadGroupFromIDLocked(key)
 	if err != nil {
-		return nil, nil, err
+		return nil, err
 	}
-	if !ok {
-		return nil, nil, fmt.Errorf("container %q not started", key.cid)
+	if tg == nil {
+		return nil, fmt.Errorf("container %q not started", key.cid)
 	}
-	return tg, tty, nil
+	return tg, nil
 }
 
-// threadGroupFromIDLocked returns the thread group and TTY for the given
-// execution ID. TTY may be nil if the process is not attached to a terminal.
-// Also returns a boolean indicating whether the container has already started.
-// Returns error if execution ID is invalid or if the container cannot be
-// found (maybe it has been deleted). Caller must hold 'mu'.
-func (l *Loader) threadGroupFromIDLocked(key execID) (*kernel.ThreadGroup, *host.TTYFileOperations, bool, error) {
+// tryThreadGroupFromIDLocked returns the thread group for the given execution
+// ID. It may return nil in case the container has not started yet. Returns
+// error if execution ID is invalid or if the container cannot be found (maybe
+// it has been deleted). Caller must hold 'mu'.
+func (l *Loader) tryThreadGroupFromIDLocked(key execID) (*kernel.ThreadGroup, error) {
 	ep := l.processes[key]
 	if ep == nil {
-		return nil, nil, false, fmt.Errorf("container %q not found", key.cid)
+		return nil, fmt.Errorf("container %q not found", key.cid)
 	}
-	if ep.tg == nil {
-		return nil, nil, false, nil
+	return ep.tg, nil
+}
+
+// ttyFromIDLocked returns the TTY files for the given execution ID. It may
+// return nil in case the container has not started yet. Returns error if
+// execution ID is invalid or if the container cannot be found (maybe it has
+// been deleted). Caller must hold 'mu'.
+func (l *Loader) ttyFromIDLocked(key execID) (*host.TTYFileOperations, *hostvfs2.TTYFileDescription, error) {
+	ep := l.processes[key]
+	if ep == nil {
+		return nil, nil, fmt.Errorf("container %q not found", key.cid)
 	}
-	return ep.tg, ep.tty, true, nil
+	return ep.tty, ep.ttyVFS2, nil
+}
+
+func createFDTable(ctx context.Context, console bool, stdioFDs []int) (*kernel.FDTable, *host.TTYFileOperations, *hostvfs2.TTYFileDescription, error) {
+	if len(stdioFDs) != 3 {
+		return nil, nil, nil, fmt.Errorf("stdioFDs should contain exactly 3 FDs (stdin, stdout, and stderr), but %d FDs received", len(stdioFDs))
+	}
+
+	k := kernel.KernelFromContext(ctx)
+	fdTable := k.NewFDTable()
+	ttyFile, ttyFileVFS2, err := fdimport.Import(ctx, fdTable, console, stdioFDs)
+	if err != nil {
+		fdTable.DecRef()
+		return nil, nil, nil, err
+	}
+	return fdTable, ttyFile, ttyFileVFS2, nil
 }

runsc/boot/vfs.go

@@ -26,6 +26,7 @@ import (
 	"gvisor.dev/gvisor/pkg/fspath"
 	"gvisor.dev/gvisor/pkg/sentry/devices/memdev"
 	"gvisor.dev/gvisor/pkg/sentry/fs"
+	devpts2 "gvisor.dev/gvisor/pkg/sentry/fsimpl/devpts"
 	devtmpfsimpl "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs"
 	goferimpl "gvisor.dev/gvisor/pkg/sentry/fsimpl/gofer"
 	procimpl "gvisor.dev/gvisor/pkg/sentry/fsimpl/proc"
@@ -41,37 +42,28 @@ import (
 )
 
 func registerFilesystems(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials) error {
-
-	vfsObj.MustRegisterFilesystemType(rootFsName, &goferimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
+	vfsObj.MustRegisterFilesystemType(devpts2.Name, &devpts2.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
+		AllowUserList: true,
+		// TODO(b/29356795): Users may mount this once the terminals are in a
+		//  usable state.
+		AllowUserMount: false,
+	})
+	vfsObj.MustRegisterFilesystemType(devtmpfsimpl.Name, &devtmpfsimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
+		AllowUserMount: true,
+		AllowUserList:  true,
+	})
+	vfsObj.MustRegisterFilesystemType(goferimpl.Name, &goferimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
 		AllowUserList: true,
 	})
-
-	vfsObj.MustRegisterFilesystemType(bind, &goferimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
-		AllowUserList: true,
-	})
-
-	vfsObj.MustRegisterFilesystemType(devpts, &devtmpfsimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
+	vfsObj.MustRegisterFilesystemType(procimpl.Name, &procimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
 		AllowUserMount: true,
 		AllowUserList:  true,
 	})
-
-	vfsObj.MustRegisterFilesystemType(devtmpfs, &devtmpfsimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
+	vfsObj.MustRegisterFilesystemType(sysimpl.Name, &sysimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
 		AllowUserMount: true,
 		AllowUserList:  true,
 	})
-	vfsObj.MustRegisterFilesystemType(proc, &procimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
-		AllowUserMount: true,
-		AllowUserList:  true,
-	})
-	vfsObj.MustRegisterFilesystemType(sysfs, &sysimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
-		AllowUserMount: true,
-		AllowUserList:  true,
-	})
-	vfsObj.MustRegisterFilesystemType(tmpfs, &tmpfsimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
-		AllowUserMount: true,
-		AllowUserList:  true,
-	})
-	vfsObj.MustRegisterFilesystemType(nonefs, &sysimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
+	vfsObj.MustRegisterFilesystemType(tmpfsimpl.Name, &tmpfsimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
 		AllowUserMount: true,
 		AllowUserList:  true,
 	})
@@ -108,7 +100,6 @@ func setupContainerVFS2(ctx context.Context, conf *Config, mntr *containerMounte
 }
 
 func setExecutablePathVFS2(ctx context.Context, procArgs *kernel.CreateProcessArgs) error {
-
 	exe := procArgs.Argv[0]
 
 	// Absolute paths can be used directly.
@@ -120,11 +111,9 @@ func setExecutablePathVFS2(ctx context.Context, procArgs *kernel.CreateProcessAr
 	// Paths with '/' in them should be joined to the working directory, or
 	// to the root if working directory is not set.
 	if strings.IndexByte(exe, '/') > 0 {
-
 		if !path.IsAbs(procArgs.WorkingDirectory) {
 			return fmt.Errorf("working directory %q must be absolute", procArgs.WorkingDirectory)
 		}
-
 		procArgs.Filename = path.Join(procArgs.WorkingDirectory, exe)
 		return nil
 	}
@@ -144,21 +133,17 @@ func setExecutablePathVFS2(ctx context.Context, procArgs *kernel.CreateProcessAr
 	creds := procArgs.Credentials
 
 	for _, p := range paths {
-
 		binPath := path.Join(p, exe)
-
 		pop := &vfs.PathOperation{
 			Root:               root,
 			Start:              root,
 			Path:               fspath.Parse(binPath),
 			FollowFinalSymlink: true,
 		}
-
 		opts := &vfs.OpenOptions{
 			FileExec: true,
 			Flags:    linux.O_RDONLY,
 		}
-
 		dentry, err := root.Mount().Filesystem().VirtualFilesystem().OpenAt(ctx, creds, pop, opts)
 		if err == syserror.ENOENT || err == syserror.EACCES {
 			// Didn't find it here.
@@ -209,7 +194,6 @@ func (c *containerMounter) setupVFS2(ctx context.Context, conf *Config, procArgs
 }
 
 func (c *containerMounter) createMountNamespaceVFS2(ctx context.Context, conf *Config, creds *auth.Credentials) (*vfs.MountNamespace, error) {
-
 	fd := c.fds.remove()
 	opts := strings.Join(p9MountOptionsVFS2(fd, conf.FileAccess), ",")
 
@@ -222,7 +206,6 @@ func (c *containerMounter) createMountNamespaceVFS2(ctx context.Context, conf *C
 }
 
 func (c *containerMounter) mountSubmountsVFS2(ctx context.Context, conf *Config, mns *vfs.MountNamespace, creds *auth.Credentials) error {
-
 	c.prepareMountsVFS2()
 
 	for _, submount := range c.mounts {
@@ -256,7 +239,6 @@ func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *Config,
 	if err != nil {
 		return fmt.Errorf("mountOptions failed: %w", err)
 	}
-
 	if fsName == "" {
 		// Filesystem is not supported (e.g. cgroup), just skip it.
 		return nil
@@ -277,7 +259,7 @@ func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *Config,
 	// All writes go to upper, be paranoid and make lower readonly.
 	opts.ReadOnly = useOverlay
 
-	if err := c.k.VFS().MountAt(ctx, creds, "", target, submount.Type, opts); err != nil {
+	if err := c.k.VFS().MountAt(ctx, creds, "", target, fsName, opts); err != nil {
 		return fmt.Errorf("failed to mount %q (type: %s): %w, opts: %v", submount.Destination, submount.Type, err, opts)
 	}
 	log.Infof("Mounted %q to %q type: %s, internal-options: %q", submount.Source, submount.Destination, submount.Type, opts)

runsc/container/container_test.go

@@ -2189,63 +2189,70 @@ func TestTTYField(t *testing.T) {
 	}
 
 	for _, test := range testCases {
-		t.Run(test.name, func(t *testing.T) {
-			conf := testutil.TestConfig(t)
+		for _, vfs2 := range []bool{false, true} {
+			name := test.name
+			if vfs2 {
+				name += "-vfs2"
+			}
+			t.Run(name, func(t *testing.T) {
+				conf := testutil.TestConfig(t)
+				conf.VFS2 = vfs2
 
-			// We will run /bin/sleep, possibly with an open TTY.
-			cmd := []string{"/bin/sleep", "10000"}
-			if test.useTTY {
-				// Run inside the "pty-runner".
-				cmd = append([]string{testApp, "pty-runner"}, cmd...)
-			}
+				// We will run /bin/sleep, possibly with an open TTY.
+				cmd := []string{"/bin/sleep", "10000"}
+				if test.useTTY {
+					// Run inside the "pty-runner".
+					cmd = append([]string{testApp, "pty-runner"}, cmd...)
+				}
 
-			spec := testutil.NewSpecWithArgs(cmd...)
-			_, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf)
-			if err != nil {
-				t.Fatalf("error setting up container: %v", err)
-			}
-			defer cleanup()
-
-			// Create and start the container.
-			args := Args{
-				ID:        testutil.RandomContainerID(),
-				Spec:      spec,
-				BundleDir: bundleDir,
-			}
-			c, err := New(conf, args)
-			if err != nil {
-				t.Fatalf("error creating container: %v", err)
-			}
-			defer c.Destroy()
-			if err := c.Start(conf); err != nil {
-				t.Fatalf("error starting container: %v", err)
-			}
-
-			// Wait for sleep to be running, and check the TTY
-			// field.
-			var gotTTYField string
-			cb := func() error {
-				ps, err := c.Processes()
+				spec := testutil.NewSpecWithArgs(cmd...)
+				_, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf)
 				if err != nil {
-					err = fmt.Errorf("error getting process data from container: %v", err)
-					return &backoff.PermanentError{Err: err}
+					t.Fatalf("error setting up container: %v", err)
 				}
-				for _, p := range ps {
-					if strings.Contains(p.Cmd, "sleep") {
-						gotTTYField = p.TTY
-						return nil
-					}
-				}
-				return fmt.Errorf("sleep not running")
-			}
-			if err := testutil.Poll(cb, 30*time.Second); err != nil {
-				t.Fatalf("error waiting for sleep process: %v", err)
-			}
+				defer cleanup()
 
-			if gotTTYField != test.wantTTYField {
-				t.Errorf("tty field got %q, want %q", gotTTYField, test.wantTTYField)
-			}
-		})
+				// Create and start the container.
+				args := Args{
+					ID:        testutil.RandomContainerID(),
+					Spec:      spec,
+					BundleDir: bundleDir,
+				}
+				c, err := New(conf, args)
+				if err != nil {
+					t.Fatalf("error creating container: %v", err)
+				}
+				defer c.Destroy()
+				if err := c.Start(conf); err != nil {
+					t.Fatalf("error starting container: %v", err)
+				}
+
+				// Wait for sleep to be running, and check the TTY
+				// field.
+				var gotTTYField string
+				cb := func() error {
+					ps, err := c.Processes()
+					if err != nil {
+						err = fmt.Errorf("error getting process data from container: %v", err)
+						return &backoff.PermanentError{Err: err}
+					}
+					for _, p := range ps {
+						if strings.Contains(p.Cmd, "sleep") {
+							gotTTYField = p.TTY
+							return nil
+						}
+					}
+					return fmt.Errorf("sleep not running")
+				}
+				if err := testutil.Poll(cb, 30*time.Second); err != nil {
+					t.Fatalf("error waiting for sleep process: %v", err)
+				}
+
+				if gotTTYField != test.wantTTYField {
+					t.Errorf("tty field got %q, want %q", gotTTYField, test.wantTTYField)
+				}
+			})
+		}
 	}
 }