Merge release-20200413.0-10-gf3ca5ca (automated)

6 files changed, 298 insertions, 46 deletions

diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go
index 55f9ed911..b98218753 100755
--- a/pkg/sentry/fsimpl/gofer/directory.go
+++ b/pkg/sentry/fsimpl/gofer/directory.go
@@ -15,6 +15,7 @@
 package gofer
 
 import (
+	"fmt"
 	"sync"
 	"sync/atomic"
 
@@ -22,6 +23,8 @@ import (
 	"gvisor.dev/gvisor/pkg/context"
 	"gvisor.dev/gvisor/pkg/p9"
 	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+	"gvisor.dev/gvisor/pkg/sentry/kernel/pipe"
+	"gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
 	"gvisor.dev/gvisor/pkg/sentry/vfs"
 	"gvisor.dev/gvisor/pkg/syserror"
 	"gvisor.dev/gvisor/pkg/usermem"
@@ -59,28 +62,52 @@ func (d *dentry) cacheNegativeLookupLocked(name string) {
 	d.children[name] = nil
 }
 
-// createSyntheticDirectory creates a synthetic directory with the given name
+type createSyntheticOpts struct {
+	name string
+	mode linux.FileMode
+	kuid auth.KUID
+	kgid auth.KGID
+
+	// The endpoint for a synthetic socket. endpoint should be nil if the file
+	// being created is not a socket.
+	endpoint transport.BoundEndpoint
+
+	// pipe should be nil if the file being created is not a pipe.
+	pipe *pipe.VFSPipe
+}
+
+// createSyntheticChildLocked creates a synthetic file with the given name
 // in d.
 //
 // Preconditions: d.dirMu must be locked. d.isDir(). d does not already contain
 // a child with the given name.
-func (d *dentry) createSyntheticDirectoryLocked(name string, mode linux.FileMode, kuid auth.KUID, kgid auth.KGID) {
+func (d *dentry) createSyntheticChildLocked(opts *createSyntheticOpts) {
 	d2 := &dentry{
 		refs:      1, // held by d
 		fs:        d.fs,
-		mode:      uint32(mode) | linux.S_IFDIR,
-		uid:       uint32(kuid),
-		gid:       uint32(kgid),
+		mode:      uint32(opts.mode),
+		uid:       uint32(opts.kuid),
+		gid:       uint32(opts.kgid),
 		blockSize: usermem.PageSize, // arbitrary
 		handle: handle{
 			fd: -1,
 		},
 		nlink: uint32(2),
 	}
+	switch opts.mode.FileType() {
+	case linux.S_IFDIR:
+		// Nothing else needs to be done.
+	case linux.S_IFSOCK:
+		d2.endpoint = opts.endpoint
+	case linux.S_IFIFO:
+		d2.pipe = opts.pipe
+	default:
+		panic(fmt.Sprintf("failed to create synthetic file of unrecognized type: %v", opts.mode.FileType()))
+	}
 	d2.pf.dentry = d2
 	d2.vfsd.Init(d2)
 
-	d.cacheNewChildLocked(d2, name)
+	d.cacheNewChildLocked(d2, opts.name)
 	d.syntheticChildren++
 }
 
diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go
index 98ccb42fd..4a8411371 100755
--- a/pkg/sentry/fsimpl/gofer/filesystem.go
+++ b/pkg/sentry/fsimpl/gofer/filesystem.go
@@ -22,9 +22,11 @@ import (
 	"gvisor.dev/gvisor/pkg/fspath"
 	"gvisor.dev/gvisor/pkg/p9"
 	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+	"gvisor.dev/gvisor/pkg/sentry/kernel/pipe"
 	"gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
 	"gvisor.dev/gvisor/pkg/sentry/vfs"
 	"gvisor.dev/gvisor/pkg/syserror"
+	"gvisor.dev/gvisor/pkg/usermem"
 )
 
 // Sync implements vfs.FilesystemImpl.Sync.
@@ -652,7 +654,12 @@ func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts v
 				return err
 			}
 			ctx.Infof("Failed to create remote directory %q: %v; falling back to synthetic directory", name, err)
-			parent.createSyntheticDirectoryLocked(name, opts.Mode, creds.EffectiveKUID, creds.EffectiveKGID)
+			parent.createSyntheticChildLocked(&createSyntheticOpts{
+				name: name,
+				mode: linux.S_IFDIR | opts.Mode,
+				kuid: creds.EffectiveKUID,
+				kgid: creds.EffectiveKGID,
+			})
 		}
 		if fs.opts.interop != InteropModeShared {
 			parent.incLinks()
@@ -663,7 +670,12 @@ func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts v
 			// Can't create non-synthetic files in synthetic directories.
 			return syserror.EPERM
 		}
-		parent.createSyntheticDirectoryLocked(name, opts.Mode, creds.EffectiveKUID, creds.EffectiveKGID)
+		parent.createSyntheticChildLocked(&createSyntheticOpts{
+			name: name,
+			mode: linux.S_IFDIR | opts.Mode,
+			kuid: creds.EffectiveKUID,
+			kgid: creds.EffectiveKGID,
+		})
 		parent.incLinks()
 		return nil
 	})
@@ -674,6 +686,28 @@ func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v
 	return fs.doCreateAt(ctx, rp, false /* dir */, func(parent *dentry, name string) error {
 		creds := rp.Credentials()
 		_, err := parent.file.mknod(ctx, name, (p9.FileMode)(opts.Mode), opts.DevMajor, opts.DevMinor, (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID))
+		if err == syserror.EPERM {
+			switch opts.Mode.FileType() {
+			case linux.S_IFSOCK:
+				parent.createSyntheticChildLocked(&createSyntheticOpts{
+					name:     name,
+					mode:     opts.Mode,
+					kuid:     creds.EffectiveKUID,
+					kgid:     creds.EffectiveKGID,
+					endpoint: opts.Endpoint,
+				})
+				return nil
+			case linux.S_IFIFO:
+				parent.createSyntheticChildLocked(&createSyntheticOpts{
+					name: name,
+					mode: opts.Mode,
+					kuid: creds.EffectiveKUID,
+					kgid: creds.EffectiveKGID,
+					pipe: pipe.NewVFSPipe(true /* isNamed */, pipe.DefaultPipeSize, usermem.PageSize),
+				})
+				return nil
+			}
+		}
 		return err
 	}, nil)
 }
@@ -756,20 +790,21 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf
 		return nil, err
 	}
 	mnt := rp.Mount()
-	filetype := d.fileType()
-	switch {
-	case filetype == linux.S_IFREG && !d.fs.opts.regularFilesUseSpecialFileFD:
-		if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, opts.Flags&linux.O_TRUNC != 0); err != nil {
-			return nil, err
-		}
-		fd := &regularFileFD{}
-		if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{
-			AllowDirectIO: true,
-		}); err != nil {
-			return nil, err
+	switch d.fileType() {
+	case linux.S_IFREG:
+		if !d.fs.opts.regularFilesUseSpecialFileFD {
+			if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, opts.Flags&linux.O_TRUNC != 0); err != nil {
+				return nil, err
+			}
+			fd := &regularFileFD{}
+			if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{
+				AllowDirectIO: true,
+			}); err != nil {
+				return nil, err
+			}
+			return &fd.vfsfd, nil
 		}
-		return &fd.vfsfd, nil
-	case filetype == linux.S_IFDIR:
+	case linux.S_IFDIR:
 		// Can't open directories with O_CREAT.
 		if opts.Flags&linux.O_CREAT != 0 {
 			return nil, syserror.EISDIR
@@ -791,26 +826,40 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf
 			return nil, err
 		}
 		return &fd.vfsfd, nil
-	case filetype == linux.S_IFLNK:
+	case linux.S_IFLNK:
 		// Can't open symlinks without O_PATH (which is unimplemented).
 		return nil, syserror.ELOOP
-	default:
-		if opts.Flags&linux.O_DIRECT != 0 {
-			return nil, syserror.EINVAL
-		}
-		h, err := openHandle(ctx, d.file, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, opts.Flags&linux.O_TRUNC != 0)
-		if err != nil {
-			return nil, err
-		}
-		fd := &specialFileFD{
-			handle: h,
+	case linux.S_IFSOCK:
+		if d.isSynthetic() {
+			return nil, syserror.ENXIO
 		}
-		if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil {
-			h.close(ctx)
-			return nil, err
+	case linux.S_IFIFO:
+		if d.isSynthetic() {
+			return d.pipe.Open(ctx, mnt, &d.vfsd, opts.Flags)
 		}
-		return &fd.vfsfd, nil
 	}
+	return d.openSpecialFileLocked(ctx, mnt, opts)
+}
+
+func (d *dentry) openSpecialFileLocked(ctx context.Context, mnt *vfs.Mount, opts *vfs.OpenOptions) (*vfs.FileDescription, error) {
+	ats := vfs.AccessTypesForOpenFlags(opts)
+	// Treat as a special file. This is done for non-synthetic pipes as well as
+	// regular files when d.fs.opts.regularFilesUseSpecialFileFD is true.
+	if opts.Flags&linux.O_DIRECT != 0 {
+		return nil, syserror.EINVAL
+	}
+	h, err := openHandle(ctx, d.file, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, opts.Flags&linux.O_TRUNC != 0)
+	if err != nil {
+		return nil, err
+	}
+	fd := &specialFileFD{
+		handle: h,
+	}
+	if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil {
+		h.close(ctx)
+		return nil, err
+	}
+	return &fd.vfsfd, nil
 }
 
 // Preconditions: d.fs.renameMu must be locked. d.dirMu must be locked.
@@ -1196,15 +1245,28 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error
 }
 
 // BoundEndpointAt implements FilesystemImpl.BoundEndpointAt.
-func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath) (transport.BoundEndpoint, error) {
+func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) {
 	var ds *[]*dentry
 	fs.renameMu.RLock()
 	defer fs.renameMuRUnlockAndCheckCaching(&ds)
-	_, err := fs.resolveLocked(ctx, rp, &ds)
+	d, err := fs.resolveLocked(ctx, rp, &ds)
 	if err != nil {
 		return nil, err
 	}
-	// TODO(gvisor.dev/issue/1476): Implement BoundEndpointAt.
+	if err := d.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil {
+		return nil, err
+	}
+	if d.isSocket() {
+		if !d.isSynthetic() {
+			d.IncRef()
+			return &endpoint{
+				dentry: d,
+				file:   d.file.file,
+				path:   opts.Addr,
+			}, nil
+		}
+		return d.endpoint, nil
+	}
 	return nil, syserror.ECONNREFUSED
 }
 
diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go
index 8b4e91d17..3235816c2 100755
--- a/pkg/sentry/fsimpl/gofer/gofer.go
+++ b/pkg/sentry/fsimpl/gofer/gofer.go
@@ -46,9 +46,11 @@ import (
 	"gvisor.dev/gvisor/pkg/p9"
 	"gvisor.dev/gvisor/pkg/sentry/fs/fsutil"
 	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+	"gvisor.dev/gvisor/pkg/sentry/kernel/pipe"
 	ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time"
 	"gvisor.dev/gvisor/pkg/sentry/memmap"
 	"gvisor.dev/gvisor/pkg/sentry/pgalloc"
+	"gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
 	"gvisor.dev/gvisor/pkg/sentry/vfs"
 	"gvisor.dev/gvisor/pkg/syserror"
 	"gvisor.dev/gvisor/pkg/unet"
@@ -585,6 +587,14 @@ type dentry struct {
 	// and target are protected by dataMu.
 	haveTarget bool
 	target     string
+
+	// If this dentry represents a synthetic socket file, endpoint is the
+	// transport endpoint bound to this file.
+	endpoint transport.BoundEndpoint
+
+	// If this dentry represents a synthetic named pipe, pipe is the pipe
+	// endpoint bound to this file.
+	pipe *pipe.VFSPipe
 }
 
 // dentryAttrMask returns a p9.AttrMask enabling all attributes used by the
@@ -791,10 +801,21 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin
 		setLocalAtime = stat.Mask&linux.STATX_ATIME != 0
 		setLocalMtime = stat.Mask&linux.STATX_MTIME != 0
 		stat.Mask &^= linux.STATX_ATIME | linux.STATX_MTIME
-		if !setLocalMtime && (stat.Mask&linux.STATX_SIZE != 0) {
-			// Truncate updates mtime.
-			setLocalMtime = true
-			stat.Mtime.Nsec = linux.UTIME_NOW
+
+		// Prepare for truncate.
+		if stat.Mask&linux.STATX_SIZE != 0 {
+			switch d.mode & linux.S_IFMT {
+			case linux.S_IFREG:
+				if !setLocalMtime {
+					// Truncate updates mtime.
+					setLocalMtime = true
+					stat.Mtime.Nsec = linux.UTIME_NOW
+				}
+			case linux.S_IFDIR:
+				return syserror.EISDIR
+			default:
+				return syserror.EINVAL
+			}
 		}
 	}
 	d.metadataMu.Lock()
@@ -1049,6 +1070,7 @@ func (d *dentry) destroyLocked() {
 		d.handle.close(ctx)
 	}
 	d.handleMu.Unlock()
+
 	if !d.file.isNil() {
 		d.file.close(ctx)
 		d.file = p9file{}
@@ -1134,7 +1156,7 @@ func (d *dentry) removexattr(ctx context.Context, creds *auth.Credentials, name
 	return d.file.removeXattr(ctx, name)
 }
 
-// Preconditions: !d.file.isNil(). d.isRegularFile() || d.isDirectory().
+// Preconditions: !d.isSynthetic(). d.isRegularFile() || d.isDirectory().
 func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool) error {
 	// O_TRUNC unconditionally requires us to obtain a new handle (opened with
 	// O_TRUNC).
diff --git a/pkg/sentry/fsimpl/gofer/socket.go b/pkg/sentry/fsimpl/gofer/socket.go
new file mode 100755
index 000000000..73835df91
--- /dev/null
+++ b/pkg/sentry/fsimpl/gofer/socket.go
@@ -0,0 +1,141 @@
+// 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 gofer
+
+import (
+	"syscall"
+
+	"gvisor.dev/gvisor/pkg/abi/linux"
+	"gvisor.dev/gvisor/pkg/context"
+	"gvisor.dev/gvisor/pkg/log"
+	"gvisor.dev/gvisor/pkg/p9"
+	"gvisor.dev/gvisor/pkg/sentry/fsimpl/host"
+	"gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
+	"gvisor.dev/gvisor/pkg/syserr"
+	"gvisor.dev/gvisor/pkg/waiter"
+)
+
+func (d *dentry) isSocket() bool {
+	return d.fileType() == linux.S_IFSOCK
+}
+
+// endpoint is a Gofer-backed transport.BoundEndpoint.
+//
+// An endpoint's lifetime is the time between when filesystem.BoundEndpointAt()
+// is called and either BoundEndpoint.BidirectionalConnect or
+// BoundEndpoint.UnidirectionalConnect is called.
+type endpoint struct {
+	// dentry is the filesystem dentry which produced this endpoint.
+	dentry *dentry
+
+	// file is the p9 file that contains a single unopened fid.
+	file p9.File
+
+	// path is the sentry path where this endpoint is bound.
+	path string
+}
+
+func sockTypeToP9(t linux.SockType) (p9.ConnectFlags, bool) {
+	switch t {
+	case linux.SOCK_STREAM:
+		return p9.StreamSocket, true
+	case linux.SOCK_SEQPACKET:
+		return p9.SeqpacketSocket, true
+	case linux.SOCK_DGRAM:
+		return p9.DgramSocket, true
+	}
+	return 0, false
+}
+
+// BidirectionalConnect implements ConnectableEndpoint.BidirectionalConnect.
+func (e *endpoint) BidirectionalConnect(ctx context.Context, ce transport.ConnectingEndpoint, returnConnect func(transport.Receiver, transport.ConnectedEndpoint)) *syserr.Error {
+	cf, ok := sockTypeToP9(ce.Type())
+	if !ok {
+		return syserr.ErrConnectionRefused
+	}
+
+	// No lock ordering required as only the ConnectingEndpoint has a mutex.
+	ce.Lock()
+
+	// Check connecting state.
+	if ce.Connected() {
+		ce.Unlock()
+		return syserr.ErrAlreadyConnected
+	}
+	if ce.Listening() {
+		ce.Unlock()
+		return syserr.ErrInvalidEndpointState
+	}
+
+	c, err := e.newConnectedEndpoint(ctx, cf, ce.WaiterQueue())
+	if err != nil {
+		ce.Unlock()
+		return err
+	}
+
+	returnConnect(c, c)
+	ce.Unlock()
+	c.Init()
+
+	return nil
+}
+
+// UnidirectionalConnect implements
+// transport.BoundEndpoint.UnidirectionalConnect.
+func (e *endpoint) UnidirectionalConnect(ctx context.Context) (transport.ConnectedEndpoint, *syserr.Error) {
+	c, err := e.newConnectedEndpoint(ctx, p9.DgramSocket, &waiter.Queue{})
+	if err != nil {
+		return nil, err
+	}
+	c.Init()
+
+	// We don't need the receiver.
+	c.CloseRecv()
+	c.Release()
+
+	return c, nil
+}
+
+func (e *endpoint) newConnectedEndpoint(ctx context.Context, flags p9.ConnectFlags, queue *waiter.Queue) (*host.SCMConnectedEndpoint, *syserr.Error) {
+	hostFile, err := e.file.Connect(flags)
+	if err != nil {
+		return nil, syserr.ErrConnectionRefused
+	}
+	// Dup the fd so that the new endpoint can manage its lifetime.
+	hostFD, err := syscall.Dup(hostFile.FD())
+	if err != nil {
+		log.Warningf("Could not dup host socket fd %d: %v", hostFile.FD(), err)
+		return nil, syserr.FromError(err)
+	}
+	// After duplicating, we no longer need hostFile.
+	hostFile.Close()
+
+	c, serr := host.NewSCMEndpoint(ctx, hostFD, queue, e.path)
+	if serr != nil {
+		log.Warningf("Gofer returned invalid host socket for BidirectionalConnect; file %+v flags %+v: %v", e.file, flags, serr)
+		return nil, serr
+	}
+	return c, nil
+}
+
+// Release implements transport.BoundEndpoint.Release.
+func (e *endpoint) Release() {
+	e.dentry.DecRef()
+}
+
+// Passcred implements transport.BoundEndpoint.Passcred.
+func (e *endpoint) Passcred() bool {
+	return false
+}
diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go
index 9e8d80414..40c1b2104 100755
--- a/pkg/sentry/fsimpl/kernfs/filesystem.go
+++ b/pkg/sentry/fsimpl/kernfs/filesystem.go
@@ -766,7 +766,7 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error
 }
 
 // BoundEndpointAt implements FilesystemImpl.BoundEndpointAt.
-func (fs *Filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath) (transport.BoundEndpoint, error) {
+func (fs *Filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) {
 	fs.mu.RLock()
 	_, _, err := fs.walkExistingLocked(ctx, rp)
 	fs.mu.RUnlock()
diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go
index 5b62f9ebb..388b98bef 100755
--- a/pkg/sentry/fsimpl/tmpfs/filesystem.go
+++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go
@@ -697,7 +697,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error
 }
 
 // BoundEndpointAt implements FilesystemImpl.BoundEndpointAt.
-func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath) (transport.BoundEndpoint, error) {
+func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) {
 	fs.mu.RLock()
 	defer fs.mu.RUnlock()
 	d, err := resolveLocked(rp)