Check in gVisor.

PiperOrigin-RevId: 194583126
Change-Id: Ica1d8821a90f74e7e745962d71801c598c652463

3 files changed, 629 insertions, 0 deletions

diff --git a/pkg/tcpip/link/fdbased/BUILD b/pkg/tcpip/link/fdbased/BUILD
new file mode 100644
index 000000000..b5ab1ea6a
--- /dev/null
+++ b/pkg/tcpip/link/fdbased/BUILD
@@ -0,0 +1,32 @@
+package(licenses = ["notice"])  # BSD
+
+load("@io_bazel_rules_go//go:def.bzl", "go_library", "go_test")
+
+go_library(
+    name = "fdbased",
+    srcs = ["endpoint.go"],
+    importpath = "gvisor.googlesource.com/gvisor/pkg/tcpip/link/fdbased",
+    visibility = [
+        "//visibility:public",
+    ],
+    deps = [
+        "//pkg/tcpip",
+        "//pkg/tcpip/buffer",
+        "//pkg/tcpip/header",
+        "//pkg/tcpip/link/rawfile",
+        "//pkg/tcpip/stack",
+    ],
+)
+
+go_test(
+    name = "fdbased_test",
+    size = "small",
+    srcs = ["endpoint_test.go"],
+    embed = [":fdbased"],
+    deps = [
+        "//pkg/tcpip",
+        "//pkg/tcpip/buffer",
+        "//pkg/tcpip/header",
+        "//pkg/tcpip/stack",
+    ],
+)
diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go
new file mode 100644
index 000000000..da74cd644
--- /dev/null
+++ b/pkg/tcpip/link/fdbased/endpoint.go
@@ -0,0 +1,261 @@
+// Copyright 2016 The Netstack Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package fdbased provides the implemention of data-link layer endpoints
+// backed by boundary-preserving file descriptors (e.g., TUN devices,
+// seqpacket/datagram sockets).
+//
+// FD based endpoints can be used in the networking stack by calling New() to
+// create a new endpoint, and then passing it as an argument to
+// Stack.CreateNIC().
+package fdbased
+
+import (
+	"syscall"
+
+	"gvisor.googlesource.com/gvisor/pkg/tcpip"
+	"gvisor.googlesource.com/gvisor/pkg/tcpip/buffer"
+	"gvisor.googlesource.com/gvisor/pkg/tcpip/header"
+	"gvisor.googlesource.com/gvisor/pkg/tcpip/link/rawfile"
+	"gvisor.googlesource.com/gvisor/pkg/tcpip/stack"
+)
+
+// BufConfig defines the shape of the vectorised view used to read packets from the NIC.
+var BufConfig = []int{128, 256, 256, 512, 1024, 2048, 4096, 8192, 16384, 32768}
+
+type endpoint struct {
+	// fd is the file descriptor used to send and receive packets.
+	fd int
+
+	// mtu (maximum transmission unit) is the maximum size of a packet.
+	mtu uint32
+
+	// hdrSize specifies the link-layer header size. If set to 0, no header
+	// is added/removed; otherwise an ethernet header is used.
+	hdrSize int
+
+	// addr is the address of the endpoint.
+	addr tcpip.LinkAddress
+
+	// caps holds the endpoint capabilities.
+	caps stack.LinkEndpointCapabilities
+
+	// closed is a function to be called when the FD's peer (if any) closes
+	// its end of the communication pipe.
+	closed func(*tcpip.Error)
+
+	vv     *buffer.VectorisedView
+	iovecs []syscall.Iovec
+	views  []buffer.View
+}
+
+// Options specify the details about the fd-based endpoint to be created.
+type Options struct {
+	FD              int
+	MTU             uint32
+	EthernetHeader  bool
+	ChecksumOffload bool
+	ClosedFunc      func(*tcpip.Error)
+	Address         tcpip.LinkAddress
+}
+
+// New creates a new fd-based endpoint.
+//
+// Makes fd non-blocking, but does not take ownership of fd, which must remain
+// open for the lifetime of the returned endpoint.
+func New(opts *Options) tcpip.LinkEndpointID {
+	syscall.SetNonblock(opts.FD, true)
+
+	caps := stack.LinkEndpointCapabilities(0)
+	if opts.ChecksumOffload {
+		caps |= stack.CapabilityChecksumOffload
+	}
+
+	hdrSize := 0
+	if opts.EthernetHeader {
+		hdrSize = header.EthernetMinimumSize
+		caps |= stack.CapabilityResolutionRequired
+	}
+
+	e := &endpoint{
+		fd:      opts.FD,
+		mtu:     opts.MTU,
+		caps:    caps,
+		closed:  opts.ClosedFunc,
+		addr:    opts.Address,
+		hdrSize: hdrSize,
+		views:   make([]buffer.View, len(BufConfig)),
+		iovecs:  make([]syscall.Iovec, len(BufConfig)),
+	}
+	vv := buffer.NewVectorisedView(0, e.views)
+	e.vv = &vv
+	return stack.RegisterLinkEndpoint(e)
+}
+
+// Attach launches the goroutine that reads packets from the file descriptor and
+// dispatches them via the provided dispatcher.
+func (e *endpoint) Attach(dispatcher stack.NetworkDispatcher) {
+	go e.dispatchLoop(dispatcher) // S/R-FIXME
+}
+
+// MTU implements stack.LinkEndpoint.MTU. It returns the value initialized
+// during construction.
+func (e *endpoint) MTU() uint32 {
+	return e.mtu
+}
+
+// Capabilities implements stack.LinkEndpoint.Capabilities.
+func (e *endpoint) Capabilities() stack.LinkEndpointCapabilities {
+	return e.caps
+}
+
+// MaxHeaderLength returns the maximum size of the link-layer header.
+func (e *endpoint) MaxHeaderLength() uint16 {
+	return uint16(e.hdrSize)
+}
+
+// LinkAddress returns the link address of this endpoint.
+func (e *endpoint) LinkAddress() tcpip.LinkAddress {
+	return e.addr
+}
+
+// WritePacket writes outbound packets to the file descriptor. If it is not
+// currently writable, the packet is dropped.
+func (e *endpoint) WritePacket(r *stack.Route, hdr *buffer.Prependable, payload buffer.View, protocol tcpip.NetworkProtocolNumber) *tcpip.Error {
+	if e.hdrSize > 0 {
+		// Add ethernet header if needed.
+		eth := header.Ethernet(hdr.Prepend(header.EthernetMinimumSize))
+		eth.Encode(&header.EthernetFields{
+			DstAddr: r.RemoteLinkAddress,
+			SrcAddr: e.addr,
+			Type:    protocol,
+		})
+	}
+
+	if len(payload) == 0 {
+		return rawfile.NonBlockingWrite(e.fd, hdr.UsedBytes())
+
+	}
+
+	return rawfile.NonBlockingWrite2(e.fd, hdr.UsedBytes(), payload)
+}
+
+func (e *endpoint) capViews(n int, buffers []int) int {
+	c := 0
+	for i, s := range buffers {
+		c += s
+		if c >= n {
+			e.views[i].CapLength(s - (c - n))
+			return i + 1
+		}
+	}
+	return len(buffers)
+}
+
+func (e *endpoint) allocateViews(bufConfig []int) {
+	for i, v := range e.views {
+		if v != nil {
+			break
+		}
+		b := buffer.NewView(bufConfig[i])
+		e.views[i] = b
+		e.iovecs[i] = syscall.Iovec{
+			Base: &b[0],
+			Len:  uint64(len(b)),
+		}
+	}
+}
+
+// dispatch reads one packet from the file descriptor and dispatches it.
+func (e *endpoint) dispatch(d stack.NetworkDispatcher, largeV buffer.View) (bool, *tcpip.Error) {
+	e.allocateViews(BufConfig)
+
+	n, err := rawfile.BlockingReadv(e.fd, e.iovecs)
+	if err != nil {
+		return false, err
+	}
+
+	if n <= e.hdrSize {
+		return false, nil
+	}
+
+	var p tcpip.NetworkProtocolNumber
+	var addr tcpip.LinkAddress
+	if e.hdrSize > 0 {
+		eth := header.Ethernet(e.views[0])
+		p = eth.Type()
+		addr = eth.SourceAddress()
+	} else {
+		// We don't get any indication of what the packet is, so try to guess
+		// if it's an IPv4 or IPv6 packet.
+		switch header.IPVersion(e.views[0]) {
+		case header.IPv4Version:
+			p = header.IPv4ProtocolNumber
+		case header.IPv6Version:
+			p = header.IPv6ProtocolNumber
+		default:
+			return true, nil
+		}
+	}
+
+	used := e.capViews(n, BufConfig)
+	e.vv.SetViews(e.views[:used])
+	e.vv.SetSize(n)
+	e.vv.TrimFront(e.hdrSize)
+
+	d.DeliverNetworkPacket(e, addr, p, e.vv)
+
+	// Prepare e.views for another packet: release used views.
+	for i := 0; i < used; i++ {
+		e.views[i] = nil
+	}
+
+	return true, nil
+}
+
+// dispatchLoop reads packets from the file descriptor in a loop and dispatches
+// them to the network stack.
+func (e *endpoint) dispatchLoop(d stack.NetworkDispatcher) *tcpip.Error {
+	v := buffer.NewView(header.MaxIPPacketSize)
+	for {
+		cont, err := e.dispatch(d, v)
+		if err != nil || !cont {
+			if e.closed != nil {
+				e.closed(err)
+			}
+			return err
+		}
+	}
+}
+
+// InjectableEndpoint is an injectable fd-based endpoint. The endpoint writes
+// to the FD, but does not read from it. All reads come from injected packets.
+type InjectableEndpoint struct {
+	endpoint
+
+	dispatcher stack.NetworkDispatcher
+}
+
+// Attach saves the stack network-layer dispatcher for use later when packets
+// are injected.
+func (e *InjectableEndpoint) Attach(dispatcher stack.NetworkDispatcher) {
+	e.dispatcher = dispatcher
+}
+
+// Inject injects an inbound packet.
+func (e *InjectableEndpoint) Inject(protocol tcpip.NetworkProtocolNumber, vv *buffer.VectorisedView) {
+	e.dispatcher.DeliverNetworkPacket(e, "", protocol, vv)
+}
+
+// NewInjectable creates a new fd-based InjectableEndpoint.
+func NewInjectable(fd int, mtu uint32) (tcpip.LinkEndpointID, *InjectableEndpoint) {
+	syscall.SetNonblock(fd, true)
+
+	e := &InjectableEndpoint{endpoint: endpoint{
+		fd:  fd,
+		mtu: mtu,
+	}}
+
+	return stack.RegisterLinkEndpoint(e), e
+}
diff --git a/pkg/tcpip/link/fdbased/endpoint_test.go b/pkg/tcpip/link/fdbased/endpoint_test.go
new file mode 100644
index 000000000..f7bbb28e1
--- /dev/null
+++ b/pkg/tcpip/link/fdbased/endpoint_test.go
@@ -0,0 +1,336 @@
+// Copyright 2016 The Netstack Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package fdbased
+
+import (
+	"fmt"
+	"math/rand"
+	"reflect"
+	"syscall"
+	"testing"
+	"time"
+
+	"gvisor.googlesource.com/gvisor/pkg/tcpip"
+	"gvisor.googlesource.com/gvisor/pkg/tcpip/buffer"
+	"gvisor.googlesource.com/gvisor/pkg/tcpip/header"
+	"gvisor.googlesource.com/gvisor/pkg/tcpip/stack"
+)
+
+type packetInfo struct {
+	raddr    tcpip.LinkAddress
+	proto    tcpip.NetworkProtocolNumber
+	contents buffer.View
+}
+
+type context struct {
+	t    *testing.T
+	fds  [2]int
+	ep   stack.LinkEndpoint
+	ch   chan packetInfo
+	done chan struct{}
+}
+
+func newContext(t *testing.T, opt *Options) *context {
+	fds, err := syscall.Socketpair(syscall.AF_UNIX, syscall.SOCK_SEQPACKET, 0)
+	if err != nil {
+		t.Fatalf("Socketpair failed: %v", err)
+	}
+
+	done := make(chan struct{}, 1)
+	opt.ClosedFunc = func(*tcpip.Error) {
+		done <- struct{}{}
+	}
+
+	opt.FD = fds[1]
+	ep := stack.FindLinkEndpoint(New(opt)).(*endpoint)
+
+	c := &context{
+		t:    t,
+		fds:  fds,
+		ep:   ep,
+		ch:   make(chan packetInfo, 100),
+		done: done,
+	}
+
+	ep.Attach(c)
+
+	return c
+}
+
+func (c *context) cleanup() {
+	syscall.Close(c.fds[0])
+	<-c.done
+	syscall.Close(c.fds[1])
+}
+
+func (c *context) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remoteLinkAddr tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv *buffer.VectorisedView) {
+	c.ch <- packetInfo{remoteLinkAddr, protocol, vv.ToView()}
+}
+
+func TestNoEthernetProperties(t *testing.T) {
+	const mtu = 1500
+	c := newContext(t, &Options{MTU: mtu})
+	defer c.cleanup()
+
+	if want, v := uint16(0), c.ep.MaxHeaderLength(); want != v {
+		t.Fatalf("MaxHeaderLength() = %v, want %v", v, want)
+	}
+
+	if want, v := uint32(mtu), c.ep.MTU(); want != v {
+		t.Fatalf("MTU() = %v, want %v", v, want)
+	}
+}
+
+func TestEthernetProperties(t *testing.T) {
+	const mtu = 1500
+	c := newContext(t, &Options{EthernetHeader: true, MTU: mtu})
+	defer c.cleanup()
+
+	if want, v := uint16(header.EthernetMinimumSize), c.ep.MaxHeaderLength(); want != v {
+		t.Fatalf("MaxHeaderLength() = %v, want %v", v, want)
+	}
+
+	if want, v := uint32(mtu), c.ep.MTU(); want != v {
+		t.Fatalf("MTU() = %v, want %v", v, want)
+	}
+}
+
+func TestAddress(t *testing.T) {
+	const mtu = 1500
+	addrs := []tcpip.LinkAddress{"", "abc", "def"}
+	for _, a := range addrs {
+		t.Run(fmt.Sprintf("Address: %q", a), func(t *testing.T) {
+			c := newContext(t, &Options{Address: a, MTU: mtu})
+			defer c.cleanup()
+
+			if want, v := a, c.ep.LinkAddress(); want != v {
+				t.Fatalf("LinkAddress() = %v, want %v", v, want)
+			}
+		})
+	}
+}
+
+func TestWritePacket(t *testing.T) {
+	const (
+		mtu   = 1500
+		laddr = tcpip.LinkAddress("\x11\x22\x33\x44\x55\x66")
+		raddr = tcpip.LinkAddress("\x77\x88\x99\xaa\xbb\xcc")
+		proto = 10
+	)
+
+	lengths := []int{0, 100, 1000}
+	eths := []bool{true, false}
+
+	for _, eth := range eths {
+		for _, plen := range lengths {
+			t.Run(fmt.Sprintf("Eth=%v,PayloadLen=%v", eth, plen), func(t *testing.T) {
+				c := newContext(t, &Options{Address: laddr, MTU: mtu, EthernetHeader: eth})
+				defer c.cleanup()
+
+				r := &stack.Route{
+					RemoteLinkAddress: raddr,
+				}
+
+				// Build header.
+				hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength()) + 100)
+				b := hdr.Prepend(100)
+				for i := range b {
+					b[i] = uint8(rand.Intn(256))
+				}
+
+				// Buiild payload and write.
+				payload := make([]byte, plen)
+				for i := range payload {
+					payload[i] = uint8(rand.Intn(256))
+				}
+				want := append(hdr.UsedBytes(), payload...)
+				if err := c.ep.WritePacket(r, &hdr, payload, proto); err != nil {
+					t.Fatalf("WritePacket failed: %v", err)
+				}
+
+				// Read from fd, then compare with what we wrote.
+				b = make([]byte, mtu)
+				n, err := syscall.Read(c.fds[0], b)
+				if err != nil {
+					t.Fatalf("Read failed: %v", err)
+				}
+				b = b[:n]
+				if eth {
+					h := header.Ethernet(b)
+					b = b[header.EthernetMinimumSize:]
+
+					if a := h.SourceAddress(); a != laddr {
+						t.Fatalf("SourceAddress() = %v, want %v", a, laddr)
+					}
+
+					if a := h.DestinationAddress(); a != raddr {
+						t.Fatalf("DestinationAddress() = %v, want %v", a, raddr)
+					}
+
+					if et := h.Type(); et != proto {
+						t.Fatalf("Type() = %v, want %v", et, proto)
+					}
+				}
+				if len(b) != len(want) {
+					t.Fatalf("Read returned %v bytes, want %v", len(b), len(want))
+				}
+				if !reflect.DeepEqual(b, want) {
+					t.Fatalf("Read returned %x, want %x", b, want)
+				}
+			})
+		}
+	}
+}
+
+func TestDeliverPacket(t *testing.T) {
+	const (
+		mtu   = 1500
+		laddr = tcpip.LinkAddress("\x11\x22\x33\x44\x55\x66")
+		raddr = tcpip.LinkAddress("\x77\x88\x99\xaa\xbb\xcc")
+		proto = 10
+	)
+
+	lengths := []int{100, 1000}
+	eths := []bool{true, false}
+
+	for _, eth := range eths {
+		for _, plen := range lengths {
+			t.Run(fmt.Sprintf("Eth=%v,PayloadLen=%v", eth, plen), func(t *testing.T) {
+				c := newContext(t, &Options{Address: laddr, MTU: mtu, EthernetHeader: eth})
+				defer c.cleanup()
+
+				// Build packet.
+				b := make([]byte, plen)
+				all := b
+				for i := range b {
+					b[i] = uint8(rand.Intn(256))
+				}
+
+				if !eth {
+					// So that it looks like an IPv4 packet.
+					b[0] = 0x40
+				} else {
+					hdr := make(header.Ethernet, header.EthernetMinimumSize)
+					hdr.Encode(&header.EthernetFields{
+						SrcAddr: raddr,
+						DstAddr: laddr,
+						Type:    proto,
+					})
+					all = append(hdr, b...)
+				}
+
+				// Write packet via the file descriptor.
+				if _, err := syscall.Write(c.fds[0], all); err != nil {
+					t.Fatalf("Write failed: %v", err)
+				}
+
+				// Receive packet through the endpoint.
+				select {
+				case pi := <-c.ch:
+					want := packetInfo{
+						raddr:    raddr,
+						proto:    proto,
+						contents: b,
+					}
+					if !eth {
+						want.proto = header.IPv4ProtocolNumber
+						want.raddr = ""
+					}
+					if !reflect.DeepEqual(want, pi) {
+						t.Fatalf("Unexpected received packet: %+v, want %+v", pi, want)
+					}
+				case <-time.After(10 * time.Second):
+					t.Fatalf("Timed out waiting for packet")
+				}
+			})
+		}
+	}
+}
+
+func TestBufConfigMaxLength(t *testing.T) {
+	got := 0
+	for _, i := range BufConfig {
+		got += i
+	}
+	want := header.MaxIPPacketSize // maximum TCP packet size
+	if got < want {
+		t.Errorf("total buffer size is invalid: got %d, want >= %d", got, want)
+	}
+}
+
+func TestBufConfigFirst(t *testing.T) {
+	// The stack assumes that the TCP/IP header is enterily contained in the first view.
+	// Therefore, the first view needs to be large enough to contain the maximum TCP/IP
+	// header, which is 120 bytes (60 bytes for IP + 60 bytes for TCP).
+	want := 120
+	got := BufConfig[0]
+	if got < want {
+		t.Errorf("first view has an invalid size: got %d, want >= %d", got, want)
+	}
+}
+
+func build(bufConfig []int) *endpoint {
+	e := &endpoint{
+		views:  make([]buffer.View, len(bufConfig)),
+		iovecs: make([]syscall.Iovec, len(bufConfig)),
+	}
+	e.allocateViews(bufConfig)
+	return e
+}
+
+var capLengthTestCases = []struct {
+	comment     string
+	config      []int
+	n           int
+	wantUsed    int
+	wantLengths []int
+}{
+	{
+		comment:     "Single slice",
+		config:      []int{2},
+		n:           1,
+		wantUsed:    1,
+		wantLengths: []int{1},
+	},
+	{
+		comment:     "Multiple slices",
+		config:      []int{1, 2},
+		n:           2,
+		wantUsed:    2,
+		wantLengths: []int{1, 1},
+	},
+	{
+		comment:     "Entire buffer",
+		config:      []int{1, 2},
+		n:           3,
+		wantUsed:    2,
+		wantLengths: []int{1, 2},
+	},
+	{
+		comment:     "Entire buffer but not on the last slice",
+		config:      []int{1, 2, 3},
+		n:           3,
+		wantUsed:    2,
+		wantLengths: []int{1, 2, 3},
+	},
+}
+
+func TestCapLength(t *testing.T) {
+	for _, c := range capLengthTestCases {
+		e := build(c.config)
+		used := e.capViews(c.n, c.config)
+		if used != c.wantUsed {
+			t.Errorf("Test \"%s\" failed when calling capViews(%d, %v). Got %d. Want %d", c.comment, c.n, c.config, used, c.wantUsed)
+		}
+		lengths := make([]int, len(e.views))
+		for i, v := range e.views {
+			lengths[i] = len(v)
+		}
+		if !reflect.DeepEqual(lengths, c.wantLengths) {
+			t.Errorf("Test \"%s\" failed when calling capViews(%d, %v). Got %v. Want %v", c.comment, c.n, c.config, lengths, c.wantLengths)
+		}
+
+	}
+}