summaryrefslogtreecommitdiffhomepage
path: root/pkg
diff options
context:
space:
mode:
authorDean Deng <deandeng@google.com>2020-04-03 14:07:42 -0700
committergVisor bot <gvisor-bot@google.com>2020-04-03 14:08:54 -0700
commit5818663ebe26857845685702d99db41c7aa2cf3d (patch)
tree36337a4b8c1cd0577f29dfacd0fcfdac84c85aee /pkg
parenta94309628ebbc2e6c4997890f1b966fa7a16be20 (diff)
Add FileDescriptionImpl for Unix sockets.
This change involves several steps: - Refactor the VFS1 unix socket implementation to share methods between VFS1 and VFS2 where possible. Re-implement the rest. - Override the default PRead, Read, PWrite, Write, Ioctl, Release methods in FileDescriptionDefaultImpl. - Add functions to create and initialize a new Dentry/Inode and FileDescription for a Unix socket file. Updates #1476 PiperOrigin-RevId: 304689796
Diffstat (limited to 'pkg')
-rw-r--r--pkg/sentry/fsimpl/sockfs/BUILD1
-rw-r--r--pkg/sentry/fsimpl/sockfs/sockfs.go29
-rw-r--r--pkg/sentry/kernel/BUILD1
-rw-r--r--pkg/sentry/socket/netstack/netstack.go8
-rw-r--r--pkg/sentry/socket/unix/BUILD4
-rw-r--r--pkg/sentry/socket/unix/unix.go89
-rw-r--r--pkg/sentry/socket/unix/unix_vfs2.go348
-rw-r--r--pkg/sentry/vfs/options.go5
8 files changed, 456 insertions, 29 deletions
diff --git a/pkg/sentry/fsimpl/sockfs/BUILD b/pkg/sentry/fsimpl/sockfs/BUILD
index 790d50e65..52084ddb5 100644
--- a/pkg/sentry/fsimpl/sockfs/BUILD
+++ b/pkg/sentry/fsimpl/sockfs/BUILD
@@ -7,6 +7,7 @@ go_library(
srcs = ["sockfs.go"],
visibility = ["//pkg/sentry:internal"],
deps = [
+ "//pkg/abi/linux",
"//pkg/context",
"//pkg/sentry/fsimpl/kernfs",
"//pkg/sentry/kernel/auth",
diff --git a/pkg/sentry/fsimpl/sockfs/sockfs.go b/pkg/sentry/fsimpl/sockfs/sockfs.go
index c13511de2..3f7ad1d65 100644
--- a/pkg/sentry/fsimpl/sockfs/sockfs.go
+++ b/pkg/sentry/fsimpl/sockfs/sockfs.go
@@ -16,6 +16,7 @@
package sockfs
import (
+ "gvisor.dev/gvisor/pkg/abi/linux"
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
@@ -60,6 +61,10 @@ type filesystem struct {
}
// inode implements kernfs.Inode.
+//
+// TODO(gvisor.dev/issue/1476): Add device numbers to this inode (which are
+// not included in InodeAttrs) to store the numbers of the appropriate
+// socket device. Override InodeAttrs.Stat() accordingly.
type inode struct {
kernfs.InodeNotDirectory
kernfs.InodeNotSymlink
@@ -71,3 +76,27 @@ type inode struct {
func (i *inode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) {
return nil, syserror.ENXIO
}
+
+// InitSocket initializes a socket FileDescription, with a corresponding
+// Dentry in mnt.
+//
+// fd should be the FileDescription associated with socketImpl, i.e. its first
+// field. mnt should be the global socket mount, Kernel.socketMount.
+func InitSocket(socketImpl vfs.FileDescriptionImpl, fd *vfs.FileDescription, mnt *vfs.Mount, creds *auth.Credentials) error {
+ fsimpl := mnt.Filesystem().Impl()
+ fs := fsimpl.(*kernfs.Filesystem)
+
+ // File mode matches net/socket.c:sock_alloc.
+ filemode := linux.FileMode(linux.S_IFSOCK | 0600)
+ i := &inode{}
+ i.Init(creds, fs.NextIno(), filemode)
+
+ d := &kernfs.Dentry{}
+ d.Init(i)
+
+ opts := &vfs.FileDescriptionOptions{UseDentryMetadata: true}
+ if err := fd.Init(socketImpl, linux.O_RDWR, mnt, d.VFSDentry(), opts); err != nil {
+ return err
+ }
+ return nil
+}
diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD
index bb7e3cbc3..e0ff58d8c 100644
--- a/pkg/sentry/kernel/BUILD
+++ b/pkg/sentry/kernel/BUILD
@@ -169,6 +169,7 @@ go_library(
"//pkg/sentry/fs/lock",
"//pkg/sentry/fs/timerfd",
"//pkg/sentry/fsbridge",
+ "//pkg/sentry/fsimpl/kernfs",
"//pkg/sentry/fsimpl/sockfs",
"//pkg/sentry/hostcpu",
"//pkg/sentry/inet",
diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go
index 06a5b53bc..5d0085462 100644
--- a/pkg/sentry/socket/netstack/netstack.go
+++ b/pkg/sentry/socket/netstack/netstack.go
@@ -940,7 +940,7 @@ func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr us
// GetSockOpt can be used to implement the linux syscall getsockopt(2) for
// sockets backed by a commonEndpoint.
-func GetSockOpt(t *kernel.Task, s socket.Socket, ep commonEndpoint, family int, skType linux.SockType, level, name, outLen int) (interface{}, *syserr.Error) {
+func GetSockOpt(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family int, skType linux.SockType, level, name, outLen int) (interface{}, *syserr.Error) {
switch level {
case linux.SOL_SOCKET:
return getSockOptSocket(t, s, ep, family, skType, name, outLen)
@@ -966,7 +966,7 @@ func GetSockOpt(t *kernel.Task, s socket.Socket, ep commonEndpoint, family int,
}
// getSockOptSocket implements GetSockOpt when level is SOL_SOCKET.
-func getSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, family int, skType linux.SockType, name, outLen int) (interface{}, *syserr.Error) {
+func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family int, skType linux.SockType, name, outLen int) (interface{}, *syserr.Error) {
// TODO(b/124056281): Stop rejecting short optLen values in getsockopt.
switch name {
case linux.SO_ERROR:
@@ -1541,7 +1541,7 @@ func (s *SocketOperations) SetSockOpt(t *kernel.Task, level int, name int, optVa
// SetSockOpt can be used to implement the linux syscall setsockopt(2) for
// sockets backed by a commonEndpoint.
-func SetSockOpt(t *kernel.Task, s socket.Socket, ep commonEndpoint, level int, name int, optVal []byte) *syserr.Error {
+func SetSockOpt(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, level int, name int, optVal []byte) *syserr.Error {
switch level {
case linux.SOL_SOCKET:
return setSockOptSocket(t, s, ep, name, optVal)
@@ -1568,7 +1568,7 @@ func SetSockOpt(t *kernel.Task, s socket.Socket, ep commonEndpoint, level int, n
}
// setSockOptSocket implements SetSockOpt when level is SOL_SOCKET.
-func setSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, name int, optVal []byte) *syserr.Error {
+func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name int, optVal []byte) *syserr.Error {
switch name {
case linux.SO_SNDBUF:
if len(optVal) < sizeOfInt32 {
diff --git a/pkg/sentry/socket/unix/BUILD b/pkg/sentry/socket/unix/BUILD
index 08743deba..de2cc4bdf 100644
--- a/pkg/sentry/socket/unix/BUILD
+++ b/pkg/sentry/socket/unix/BUILD
@@ -8,23 +8,27 @@ go_library(
"device.go",
"io.go",
"unix.go",
+ "unix_vfs2.go",
],
visibility = ["//pkg/sentry:internal"],
deps = [
"//pkg/abi/linux",
"//pkg/context",
+ "//pkg/fspath",
"//pkg/refs",
"//pkg/safemem",
"//pkg/sentry/arch",
"//pkg/sentry/device",
"//pkg/sentry/fs",
"//pkg/sentry/fs/fsutil",
+ "//pkg/sentry/fsimpl/sockfs",
"//pkg/sentry/kernel",
"//pkg/sentry/kernel/time",
"//pkg/sentry/socket",
"//pkg/sentry/socket/control",
"//pkg/sentry/socket/netstack",
"//pkg/sentry/socket/unix/transport",
+ "//pkg/sentry/vfs",
"//pkg/syserr",
"//pkg/syserror",
"//pkg/tcpip",
diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go
index 4d30aa714..7c64f30fa 100644
--- a/pkg/sentry/socket/unix/unix.go
+++ b/pkg/sentry/socket/unix/unix.go
@@ -23,6 +23,7 @@ import (
"gvisor.dev/gvisor/pkg/abi/linux"
"gvisor.dev/gvisor/pkg/context"
+ "gvisor.dev/gvisor/pkg/fspath"
"gvisor.dev/gvisor/pkg/refs"
"gvisor.dev/gvisor/pkg/sentry/arch"
"gvisor.dev/gvisor/pkg/sentry/fs"
@@ -33,6 +34,7 @@ import (
"gvisor.dev/gvisor/pkg/sentry/socket/control"
"gvisor.dev/gvisor/pkg/sentry/socket/netstack"
"gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
+ "gvisor.dev/gvisor/pkg/sentry/vfs"
"gvisor.dev/gvisor/pkg/syserr"
"gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/tcpip"
@@ -52,11 +54,8 @@ type SocketOperations struct {
fsutil.FileNoSplice `state:"nosave"`
fsutil.FileNoopFlush `state:"nosave"`
fsutil.FileUseInodeUnstableAttr `state:"nosave"`
- refs.AtomicRefCount
- socket.SendReceiveTimeout
- ep transport.Endpoint
- stype linux.SockType
+ socketOpsCommon
}
// New creates a new unix socket.
@@ -75,16 +74,29 @@ func NewWithDirent(ctx context.Context, d *fs.Dirent, ep transport.Endpoint, sty
}
s := SocketOperations{
- ep: ep,
- stype: stype,
+ socketOpsCommon: socketOpsCommon{
+ ep: ep,
+ stype: stype,
+ },
}
s.EnableLeakCheck("unix.SocketOperations")
return fs.NewFile(ctx, d, flags, &s)
}
+// socketOpsCommon contains the socket operations common to VFS1 and VFS2.
+//
+// +stateify savable
+type socketOpsCommon struct {
+ refs.AtomicRefCount
+ socket.SendReceiveTimeout
+
+ ep transport.Endpoint
+ stype linux.SockType
+}
+
// DecRef implements RefCounter.DecRef.
-func (s *SocketOperations) DecRef() {
+func (s *socketOpsCommon) DecRef() {
s.DecRefWithDestructor(func() {
s.ep.Close()
})
@@ -97,7 +109,7 @@ func (s *SocketOperations) Release() {
s.DecRef()
}
-func (s *SocketOperations) isPacket() bool {
+func (s *socketOpsCommon) isPacket() bool {
switch s.stype {
case linux.SOCK_DGRAM, linux.SOCK_SEQPACKET:
return true
@@ -110,7 +122,7 @@ func (s *SocketOperations) isPacket() bool {
}
// Endpoint extracts the transport.Endpoint.
-func (s *SocketOperations) Endpoint() transport.Endpoint {
+func (s *socketOpsCommon) Endpoint() transport.Endpoint {
return s.ep
}
@@ -143,7 +155,7 @@ func extractPath(sockaddr []byte) (string, *syserr.Error) {
// GetPeerName implements the linux syscall getpeername(2) for sockets backed by
// a transport.Endpoint.
-func (s *SocketOperations) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) {
+func (s *socketOpsCommon) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) {
addr, err := s.ep.GetRemoteAddress()
if err != nil {
return nil, 0, syserr.TranslateNetstackError(err)
@@ -155,7 +167,7 @@ func (s *SocketOperations) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32,
// GetSockName implements the linux syscall getsockname(2) for sockets backed by
// a transport.Endpoint.
-func (s *SocketOperations) GetSockName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) {
+func (s *socketOpsCommon) GetSockName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) {
addr, err := s.ep.GetLocalAddress()
if err != nil {
return nil, 0, syserr.TranslateNetstackError(err)
@@ -178,7 +190,7 @@ func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr us
// Listen implements the linux syscall listen(2) for sockets backed by
// a transport.Endpoint.
-func (s *SocketOperations) Listen(t *kernel.Task, backlog int) *syserr.Error {
+func (s *socketOpsCommon) Listen(t *kernel.Task, backlog int) *syserr.Error {
return s.ep.Listen(backlog)
}
@@ -310,6 +322,8 @@ func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error {
}
// Create the socket.
+ //
+ // TODO(gvisor.dev/issue/2324): Correctly set file permissions.
childDir, err := d.Bind(t, t.FSContext().RootDirectory(), name, bep, fs.FilePermissions{User: fs.PermMask{Read: true}})
if err != nil {
return syserr.ErrPortInUse
@@ -345,6 +359,31 @@ func extractEndpoint(t *kernel.Task, sockaddr []byte) (transport.BoundEndpoint,
return ep, nil
}
+ if kernel.VFS2Enabled {
+ p := fspath.Parse(path)
+ root := t.FSContext().RootDirectoryVFS2()
+ start := root
+ relPath := !p.Absolute
+ if relPath {
+ start = t.FSContext().WorkingDirectoryVFS2()
+ }
+ pop := vfs.PathOperation{
+ Root: root,
+ Start: start,
+ Path: p,
+ FollowFinalSymlink: true,
+ }
+ ep, e := t.Kernel().VFS().BoundEndpointAt(t, t.Credentials(), &pop)
+ root.DecRef()
+ if relPath {
+ start.DecRef()
+ }
+ if e != nil {
+ return nil, syserr.FromError(e)
+ }
+ return ep, nil
+ }
+
// Find the node in the filesystem.
root := t.FSContext().RootDirectory()
cwd := t.FSContext().WorkingDirectory()
@@ -363,12 +402,11 @@ func extractEndpoint(t *kernel.Task, sockaddr []byte) (transport.BoundEndpoint,
// No socket!
return nil, syserr.ErrConnectionRefused
}
-
return ep, nil
}
// Connect implements the linux syscall connect(2) for unix sockets.
-func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr.Error {
+func (s *socketOpsCommon) Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr.Error {
ep, err := extractEndpoint(t, sockaddr)
if err != nil {
return err
@@ -379,7 +417,7 @@ func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking boo
return s.ep.Connect(t, ep)
}
-// Writev implements fs.FileOperations.Write.
+// Write implements fs.FileOperations.Write.
func (s *SocketOperations) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, _ int64) (int64, error) {
t := kernel.TaskFromContext(ctx)
ctrl := control.New(t, s.ep, nil)
@@ -399,7 +437,7 @@ func (s *SocketOperations) Write(ctx context.Context, _ *fs.File, src usermem.IO
// SendMsg implements the linux syscall sendmsg(2) for unix sockets backed by
// a transport.Endpoint.
-func (s *SocketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to []byte, flags int, haveDeadline bool, deadline ktime.Time, controlMessages socket.ControlMessages) (int, *syserr.Error) {
+func (s *socketOpsCommon) SendMsg(t *kernel.Task, src usermem.IOSequence, to []byte, flags int, haveDeadline bool, deadline ktime.Time, controlMessages socket.ControlMessages) (int, *syserr.Error) {
w := EndpointWriter{
Ctx: t,
Endpoint: s.ep,
@@ -453,27 +491,27 @@ func (s *SocketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to []
}
// Passcred implements transport.Credentialer.Passcred.
-func (s *SocketOperations) Passcred() bool {
+func (s *socketOpsCommon) Passcred() bool {
return s.ep.Passcred()
}
// ConnectedPasscred implements transport.Credentialer.ConnectedPasscred.
-func (s *SocketOperations) ConnectedPasscred() bool {
+func (s *socketOpsCommon) ConnectedPasscred() bool {
return s.ep.ConnectedPasscred()
}
// Readiness implements waiter.Waitable.Readiness.
-func (s *SocketOperations) Readiness(mask waiter.EventMask) waiter.EventMask {
+func (s *socketOpsCommon) Readiness(mask waiter.EventMask) waiter.EventMask {
return s.ep.Readiness(mask)
}
// EventRegister implements waiter.Waitable.EventRegister.
-func (s *SocketOperations) EventRegister(e *waiter.Entry, mask waiter.EventMask) {
+func (s *socketOpsCommon) EventRegister(e *waiter.Entry, mask waiter.EventMask) {
s.ep.EventRegister(e, mask)
}
// EventUnregister implements waiter.Waitable.EventUnregister.
-func (s *SocketOperations) EventUnregister(e *waiter.Entry) {
+func (s *socketOpsCommon) EventUnregister(e *waiter.Entry) {
s.ep.EventUnregister(e)
}
@@ -485,7 +523,7 @@ func (s *SocketOperations) SetSockOpt(t *kernel.Task, level int, name int, optVa
// Shutdown implements the linux syscall shutdown(2) for sockets backed by
// a transport.Endpoint.
-func (s *SocketOperations) Shutdown(t *kernel.Task, how int) *syserr.Error {
+func (s *socketOpsCommon) Shutdown(t *kernel.Task, how int) *syserr.Error {
f, err := netstack.ConvertShutdown(how)
if err != nil {
return err
@@ -511,7 +549,7 @@ func (s *SocketOperations) Read(ctx context.Context, _ *fs.File, dst usermem.IOS
// RecvMsg implements the linux syscall recvmsg(2) for sockets backed by
// a transport.Endpoint.
-func (s *SocketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, haveDeadline bool, deadline ktime.Time, senderRequested bool, controlDataLen uint64) (n int, msgFlags int, senderAddr linux.SockAddr, senderAddrLen uint32, controlMessages socket.ControlMessages, err *syserr.Error) {
+func (s *socketOpsCommon) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, haveDeadline bool, deadline ktime.Time, senderRequested bool, controlDataLen uint64) (n int, msgFlags int, senderAddr linux.SockAddr, senderAddrLen uint32, controlMessages socket.ControlMessages, err *syserr.Error) {
trunc := flags&linux.MSG_TRUNC != 0
peek := flags&linux.MSG_PEEK != 0
dontWait := flags&linux.MSG_DONTWAIT != 0
@@ -648,12 +686,12 @@ func (s *SocketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags
}
// State implements socket.Socket.State.
-func (s *SocketOperations) State() uint32 {
+func (s *socketOpsCommon) State() uint32 {
return s.ep.State()
}
// Type implements socket.Socket.Type.
-func (s *SocketOperations) Type() (family int, skType linux.SockType, protocol int) {
+func (s *socketOpsCommon) Type() (family int, skType linux.SockType, protocol int) {
// Unix domain sockets always have a protocol of 0.
return linux.AF_UNIX, s.stype, 0
}
@@ -706,4 +744,5 @@ func (*provider) Pair(t *kernel.Task, stype linux.SockType, protocol int) (*fs.F
func init() {
socket.RegisterProvider(linux.AF_UNIX, &provider{})
+ socket.RegisterProviderVFS2(linux.AF_UNIX, &providerVFS2{})
}
diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go
new file mode 100644
index 000000000..ca1388e2c
--- /dev/null
+++ b/pkg/sentry/socket/unix/unix_vfs2.go
@@ -0,0 +1,348 @@
+// 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 unix
+
+import (
+ "gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/context"
+ "gvisor.dev/gvisor/pkg/fspath"
+ "gvisor.dev/gvisor/pkg/sentry/arch"
+ "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs"
+ "gvisor.dev/gvisor/pkg/sentry/kernel"
+ "gvisor.dev/gvisor/pkg/sentry/socket/control"
+ "gvisor.dev/gvisor/pkg/sentry/socket/netstack"
+ "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
+ "gvisor.dev/gvisor/pkg/sentry/vfs"
+ "gvisor.dev/gvisor/pkg/syserr"
+ "gvisor.dev/gvisor/pkg/syserror"
+ "gvisor.dev/gvisor/pkg/tcpip"
+ "gvisor.dev/gvisor/pkg/usermem"
+ "gvisor.dev/gvisor/pkg/waiter"
+)
+
+// SocketVFS2 implements socket.SocketVFS2 (and by extension,
+// vfs.FileDescriptionImpl) for Unix sockets.
+type SocketVFS2 struct {
+ vfsfd vfs.FileDescription
+ vfs.FileDescriptionDefaultImpl
+ vfs.DentryMetadataFileDescriptionImpl
+
+ socketOpsCommon
+}
+
+// NewVFS2File creates and returns a new vfs.FileDescription for a unix socket.
+func NewVFS2File(t *kernel.Task, ep transport.Endpoint, stype linux.SockType) (*vfs.FileDescription, *syserr.Error) {
+ sock := NewFDImpl(ep, stype)
+ vfsfd := &sock.vfsfd
+ if err := sockfs.InitSocket(sock, vfsfd, t.Kernel().SocketMount(), t.Credentials()); err != nil {
+ return nil, syserr.FromError(err)
+ }
+ return vfsfd, nil
+}
+
+// NewFDImpl creates and returns a new SocketVFS2.
+func NewFDImpl(ep transport.Endpoint, stype linux.SockType) *SocketVFS2 {
+ // You can create AF_UNIX, SOCK_RAW sockets. They're the same as
+ // SOCK_DGRAM and don't require CAP_NET_RAW.
+ if stype == linux.SOCK_RAW {
+ stype = linux.SOCK_DGRAM
+ }
+
+ return &SocketVFS2{
+ socketOpsCommon: socketOpsCommon{
+ ep: ep,
+ stype: stype,
+ },
+ }
+}
+
+// GetSockOpt implements the linux syscall getsockopt(2) for sockets backed by
+// a transport.Endpoint.
+func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) {
+ return netstack.GetSockOpt(t, s, s.ep, linux.AF_UNIX, s.ep.Type(), level, name, outLen)
+}
+
+// blockingAccept implements a blocking version of accept(2), that is, if no
+// connections are ready to be accept, it will block until one becomes ready.
+func (s *SocketVFS2) blockingAccept(t *kernel.Task) (transport.Endpoint, *syserr.Error) {
+ // Register for notifications.
+ e, ch := waiter.NewChannelEntry(nil)
+ s.socketOpsCommon.EventRegister(&e, waiter.EventIn)
+ defer s.socketOpsCommon.EventUnregister(&e)
+
+ // Try to accept the connection; if it fails, then wait until we get a
+ // notification.
+ for {
+ if ep, err := s.ep.Accept(); err != syserr.ErrWouldBlock {
+ return ep, err
+ }
+
+ if err := t.Block(ch); err != nil {
+ return nil, syserr.FromError(err)
+ }
+ }
+}
+
+// Accept implements the linux syscall accept(2) for sockets backed by
+// a transport.Endpoint.
+func (s *SocketVFS2) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (int32, linux.SockAddr, uint32, *syserr.Error) {
+ // Issue the accept request to get the new endpoint.
+ ep, err := s.ep.Accept()
+ if err != nil {
+ if err != syserr.ErrWouldBlock || !blocking {
+ return 0, nil, 0, err
+ }
+
+ var err *syserr.Error
+ ep, err = s.blockingAccept(t)
+ if err != nil {
+ return 0, nil, 0, err
+ }
+ }
+
+ // We expect this to be a FileDescription here.
+ ns, err := NewVFS2File(t, ep, s.stype)
+ if err != nil {
+ return 0, nil, 0, err
+ }
+ defer ns.DecRef()
+
+ if flags&linux.SOCK_NONBLOCK != 0 {
+ ns.SetStatusFlags(t, t.Credentials(), linux.SOCK_NONBLOCK)
+ }
+
+ var addr linux.SockAddr
+ var addrLen uint32
+ if peerRequested {
+ // Get address of the peer.
+ var err *syserr.Error
+ addr, addrLen, err = ns.Impl().(*SocketVFS2).GetPeerName(t)
+ if err != nil {
+ return 0, nil, 0, err
+ }
+ }
+
+ fd, e := t.NewFDFromVFS2(0, ns, kernel.FDFlags{
+ CloseOnExec: flags&linux.SOCK_CLOEXEC != 0,
+ })
+ if e != nil {
+ return 0, nil, 0, syserr.FromError(e)
+ }
+
+ // TODO: add vfs2 sockets to global table.
+ return fd, addr, addrLen, nil
+}
+
+// Bind implements the linux syscall bind(2) for unix sockets.
+func (s *SocketVFS2) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error {
+ p, e := extractPath(sockaddr)
+ if e != nil {
+ return e
+ }
+
+ bep, ok := s.ep.(transport.BoundEndpoint)
+ if !ok {
+ // This socket can't be bound.
+ return syserr.ErrInvalidArgument
+ }
+
+ return s.ep.Bind(tcpip.FullAddress{Addr: tcpip.Address(p)}, func() *syserr.Error {
+ // Is it abstract?
+ if p[0] == 0 {
+ if t.IsNetworkNamespaced() {
+ return syserr.ErrInvalidEndpointState
+ }
+ if err := t.AbstractSockets().Bind(p[1:], bep, s); err != nil {
+ // syserr.ErrPortInUse corresponds to EADDRINUSE.
+ return syserr.ErrPortInUse
+ }
+ } else {
+ path := fspath.Parse(p)
+ root := t.FSContext().RootDirectoryVFS2()
+ defer root.DecRef()
+ start := root
+ relPath := !path.Absolute
+ if relPath {
+ start = t.FSContext().WorkingDirectoryVFS2()
+ defer start.DecRef()
+ }
+ pop := vfs.PathOperation{
+ Root: root,
+ Start: start,
+ Path: path,
+ }
+ err := t.Kernel().VFS().MknodAt(t, t.Credentials(), &pop, &vfs.MknodOptions{
+ // TODO(gvisor.dev/issue/2324): The file permissions should be taken
+ // from s and t.FSContext().Umask() (see net/unix/af_unix.c:unix_bind),
+ // but VFS1 just always uses 0400. Resolve this inconsistency.
+ Mode: linux.S_IFSOCK | 0400,
+ Endpoint: bep,
+ })
+ if err == syserror.EEXIST {
+ return syserr.ErrAddressInUse
+ }
+ return syserr.FromError(err)
+ }
+
+ return nil
+ })
+}
+
+// Ioctl implements vfs.FileDescriptionImpl.
+func (s *SocketVFS2) Ioctl(ctx context.Context, uio usermem.IO, args arch.SyscallArguments) (uintptr, error) {
+ return netstack.Ioctl(ctx, s.ep, uio, args)
+}
+
+// PRead implements vfs.FileDescriptionImpl.
+func (s *SocketVFS2) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) {
+ return 0, syserror.ESPIPE
+}
+
+// Read implements vfs.FileDescriptionImpl.
+func (s *SocketVFS2) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) {
+ // All flags other than RWF_NOWAIT should be ignored.
+ // TODO(gvisor.dev/issue/1476): Support RWF_NOWAIT.
+ if opts.Flags != 0 {
+ return 0, syserror.EOPNOTSUPP
+ }
+
+ if dst.NumBytes() == 0 {
+ return 0, nil
+ }
+ return dst.CopyOutFrom(ctx, &EndpointReader{
+ Ctx: ctx,
+ Endpoint: s.ep,
+ NumRights: 0,
+ Peek: false,
+ From: nil,
+ })
+}
+
+// PWrite implements vfs.FileDescriptionImpl.
+func (s *SocketVFS2) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) {
+ return 0, syserror.ESPIPE
+}
+
+// Write implements vfs.FileDescriptionImpl.
+func (s *SocketVFS2) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) {
+ // All flags other than RWF_NOWAIT should be ignored.
+ // TODO(gvisor.dev/issue/1476): Support RWF_NOWAIT.
+ if opts.Flags != 0 {
+ return 0, syserror.EOPNOTSUPP
+ }
+
+ t := kernel.TaskFromContext(ctx)
+ ctrl := control.New(t, s.ep, nil)
+
+ if src.NumBytes() == 0 {
+ nInt, err := s.ep.SendMsg(ctx, [][]byte{}, ctrl, nil)
+ return int64(nInt), err.ToError()
+ }
+
+ return src.CopyInTo(ctx, &EndpointWriter{
+ Ctx: ctx,
+ Endpoint: s.ep,
+ Control: ctrl,
+ To: nil,
+ })
+}
+
+// Release implements vfs.FileDescriptionImpl.
+func (s *SocketVFS2) Release() {
+ // Release only decrements a reference on s because s may be referenced in
+ // the abstract socket namespace.
+ s.DecRef()
+}
+
+// Readiness implements waiter.Waitable.Readiness.
+func (s *SocketVFS2) Readiness(mask waiter.EventMask) waiter.EventMask {
+ return s.socketOpsCommon.Readiness(mask)
+}
+
+// EventRegister implements waiter.Waitable.EventRegister.
+func (s *SocketVFS2) EventRegister(e *waiter.Entry, mask waiter.EventMask) {
+ s.socketOpsCommon.EventRegister(e, mask)
+}
+
+// EventUnregister implements waiter.Waitable.EventUnregister.
+func (s *SocketVFS2) EventUnregister(e *waiter.Entry) {
+ s.socketOpsCommon.EventUnregister(e)
+}
+
+// SetSockOpt implements the linux syscall setsockopt(2) for sockets backed by
+// a transport.Endpoint.
+func (s *SocketVFS2) SetSockOpt(t *kernel.Task, level int, name int, optVal []byte) *syserr.Error {
+ return netstack.SetSockOpt(t, s, s.ep, level, name, optVal)
+}
+
+// providerVFS2 is a unix domain socket provider for VFS2.
+type providerVFS2 struct{}
+
+func (*providerVFS2) Socket(t *kernel.Task, stype linux.SockType, protocol int) (*vfs.FileDescription, *syserr.Error) {
+ // Check arguments.
+ if protocol != 0 && protocol != linux.AF_UNIX /* PF_UNIX */ {
+ return nil, syserr.ErrProtocolNotSupported
+ }
+
+ // Create the endpoint and socket.
+ var ep transport.Endpoint
+ switch stype {
+ case linux.SOCK_DGRAM, linux.SOCK_RAW:
+ ep = transport.NewConnectionless(t)
+ case linux.SOCK_SEQPACKET, linux.SOCK_STREAM:
+ ep = transport.NewConnectioned(t, stype, t.Kernel())
+ default:
+ return nil, syserr.ErrInvalidArgument
+ }
+
+ f, err := NewVFS2File(t, ep, stype)
+ if err != nil {
+ ep.Close()
+ return nil, err
+ }
+ return f, nil
+}
+
+// Pair creates a new pair of AF_UNIX connected sockets.
+func (*providerVFS2) Pair(t *kernel.Task, stype linux.SockType, protocol int) (*vfs.FileDescription, *vfs.FileDescription, *syserr.Error) {
+ // Check arguments.
+ if protocol != 0 && protocol != linux.AF_UNIX /* PF_UNIX */ {
+ return nil, nil, syserr.ErrProtocolNotSupported
+ }
+
+ switch stype {
+ case linux.SOCK_STREAM, linux.SOCK_DGRAM, linux.SOCK_SEQPACKET, linux.SOCK_RAW:
+ // Ok
+ default:
+ return nil, nil, syserr.ErrInvalidArgument
+ }
+
+ // Create the endpoints and sockets.
+ ep1, ep2 := transport.NewPair(t, stype, t.Kernel())
+ s1, err := NewVFS2File(t, ep1, stype)
+ if err != nil {
+ ep1.Close()
+ ep2.Close()
+ return nil, nil, err
+ }
+ s2, err := NewVFS2File(t, ep2, stype)
+ if err != nil {
+ s1.DecRef()
+ ep2.Close()
+ return nil, nil, err
+ }
+
+ return s1, s2, nil
+}
diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go
index 3e90dc4ed..2f04bf882 100644
--- a/pkg/sentry/vfs/options.go
+++ b/pkg/sentry/vfs/options.go
@@ -16,6 +16,7 @@ package vfs
import (
"gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
)
// GetDentryOptions contains options to VirtualFilesystem.GetDentryAt() and
@@ -44,6 +45,10 @@ type MknodOptions struct {
// DevMinor are the major and minor device numbers for the created device.
DevMajor uint32
DevMinor uint32
+
+ // Endpoint is the endpoint to bind to the created file, if a socket file is
+ // being created for bind(2) on a Unix domain socket.
+ Endpoint transport.BoundEndpoint
}
// MountFlags contains flags as specified for mount(2), e.g. MS_NOEXEC.