summaryrefslogtreecommitdiffhomepage
diff options
context:
space:
mode:
-rw-r--r--pkg/sentry/inet/inet.go4
-rw-r--r--pkg/sentry/inet/test_stack.go6
-rw-r--r--pkg/sentry/socket/hostinet/stack.go5
-rw-r--r--pkg/sentry/socket/netlink/BUILD14
-rw-r--r--pkg/sentry/socket/netlink/message.go129
-rw-r--r--pkg/sentry/socket/netlink/message_test.go312
-rw-r--r--pkg/sentry/socket/netlink/provider.go2
-rw-r--r--pkg/sentry/socket/netlink/route/BUILD2
-rw-r--r--pkg/sentry/socket/netlink/route/protocol.go238
-rw-r--r--pkg/sentry/socket/netlink/socket.go54
-rw-r--r--pkg/sentry/socket/netlink/uevent/protocol.go2
-rw-r--r--pkg/sentry/socket/netstack/stack.go55
-rw-r--r--pkg/tcpip/stack/stack.go9
-rw-r--r--test/syscalls/linux/BUILD2
-rw-r--r--test/syscalls/linux/socket_netlink_route.cc296
-rw-r--r--test/syscalls/linux/socket_netlink_util.cc45
-rw-r--r--test/syscalls/linux/socket_netlink_util.h9
17 files changed, 1022 insertions, 162 deletions
diff --git a/pkg/sentry/inet/inet.go b/pkg/sentry/inet/inet.go
index a7dfb78a7..2916a0644 100644
--- a/pkg/sentry/inet/inet.go
+++ b/pkg/sentry/inet/inet.go
@@ -28,6 +28,10 @@ type Stack interface {
// interface indexes to a slice of associated interface address properties.
InterfaceAddrs() map[int32][]InterfaceAddr
+ // AddInterfaceAddr adds an address to the network interface identified by
+ // index.
+ AddInterfaceAddr(idx int32, addr InterfaceAddr) error
+
// SupportsIPv6 returns true if the stack supports IPv6 connectivity.
SupportsIPv6() bool
diff --git a/pkg/sentry/inet/test_stack.go b/pkg/sentry/inet/test_stack.go
index dcfcbd97e..d8961fc94 100644
--- a/pkg/sentry/inet/test_stack.go
+++ b/pkg/sentry/inet/test_stack.go
@@ -47,6 +47,12 @@ func (s *TestStack) InterfaceAddrs() map[int32][]InterfaceAddr {
return s.InterfaceAddrsMap
}
+// AddInterfaceAddr implements Stack.AddInterfaceAddr.
+func (s *TestStack) AddInterfaceAddr(idx int32, addr InterfaceAddr) error {
+ s.InterfaceAddrsMap[idx] = append(s.InterfaceAddrsMap[idx], addr)
+ return nil
+}
+
// SupportsIPv6 implements Stack.SupportsIPv6.
func (s *TestStack) SupportsIPv6() bool {
return s.SupportsIPv6Flag
diff --git a/pkg/sentry/socket/hostinet/stack.go b/pkg/sentry/socket/hostinet/stack.go
index 034eca676..a48082631 100644
--- a/pkg/sentry/socket/hostinet/stack.go
+++ b/pkg/sentry/socket/hostinet/stack.go
@@ -310,6 +310,11 @@ func (s *Stack) InterfaceAddrs() map[int32][]inet.InterfaceAddr {
return addrs
}
+// AddInterfaceAddr implements inet.Stack.AddInterfaceAddr.
+func (s *Stack) AddInterfaceAddr(idx int32, addr inet.InterfaceAddr) error {
+ return syserror.EACCES
+}
+
// SupportsIPv6 implements inet.Stack.SupportsIPv6.
func (s *Stack) SupportsIPv6() bool {
return s.supportsIPv6
diff --git a/pkg/sentry/socket/netlink/BUILD b/pkg/sentry/socket/netlink/BUILD
index f8b8e467d..1911cd9b8 100644
--- a/pkg/sentry/socket/netlink/BUILD
+++ b/pkg/sentry/socket/netlink/BUILD
@@ -1,4 +1,4 @@
-load("//tools:defs.bzl", "go_library")
+load("//tools:defs.bzl", "go_library", "go_test")
package(licenses = ["notice"])
@@ -33,3 +33,15 @@ go_library(
"//pkg/waiter",
],
)
+
+go_test(
+ name = "netlink_test",
+ size = "small",
+ srcs = [
+ "message_test.go",
+ ],
+ deps = [
+ ":netlink",
+ "//pkg/abi/linux",
+ ],
+)
diff --git a/pkg/sentry/socket/netlink/message.go b/pkg/sentry/socket/netlink/message.go
index b21e0ca4b..4ea252ccb 100644
--- a/pkg/sentry/socket/netlink/message.go
+++ b/pkg/sentry/socket/netlink/message.go
@@ -30,8 +30,16 @@ func alignUp(length int, align uint) int {
return (length + int(align) - 1) &^ (int(align) - 1)
}
+// alignPad returns the length of padding required for alignment.
+//
+// Preconditions: align is a power of two.
+func alignPad(length int, align uint) int {
+ return alignUp(length, align) - length
+}
+
// Message contains a complete serialized netlink message.
type Message struct {
+ hdr linux.NetlinkMessageHeader
buf []byte
}
@@ -40,10 +48,86 @@ type Message struct {
// The header length will be updated by Finalize.
func NewMessage(hdr linux.NetlinkMessageHeader) *Message {
return &Message{
+ hdr: hdr,
buf: binary.Marshal(nil, usermem.ByteOrder, hdr),
}
}
+// ParseMessage parses the first message seen at buf, returning the rest of the
+// buffer. If message is malformed, ok of false is returned. For last message,
+// padding check is loose, if there isn't enought padding, whole buf is consumed
+// and ok is set to true.
+func ParseMessage(buf []byte) (msg *Message, rest []byte, ok bool) {
+ b := BytesView(buf)
+
+ hdrBytes, ok := b.Extract(linux.NetlinkMessageHeaderSize)
+ if !ok {
+ return
+ }
+ var hdr linux.NetlinkMessageHeader
+ binary.Unmarshal(hdrBytes, usermem.ByteOrder, &hdr)
+
+ // Msg portion.
+ totalMsgLen := int(hdr.Length)
+ _, ok = b.Extract(totalMsgLen - linux.NetlinkMessageHeaderSize)
+ if !ok {
+ return
+ }
+
+ // Padding.
+ numPad := alignPad(totalMsgLen, linux.NLMSG_ALIGNTO)
+ // Linux permits the last message not being aligned, just consume all of it.
+ // Ref: net/netlink/af_netlink.c:netlink_rcv_skb
+ if numPad > len(b) {
+ numPad = len(b)
+ }
+ _, ok = b.Extract(numPad)
+ if !ok {
+ return
+ }
+
+ return &Message{
+ hdr: hdr,
+ buf: buf[:totalMsgLen],
+ }, []byte(b), true
+}
+
+// Header returns the header of this message.
+func (m *Message) Header() linux.NetlinkMessageHeader {
+ return m.hdr
+}
+
+// GetData unmarshals the payload message header from this netlink message, and
+// returns the attributes portion.
+func (m *Message) GetData(msg interface{}) (AttrsView, bool) {
+ b := BytesView(m.buf)
+
+ _, ok := b.Extract(linux.NetlinkMessageHeaderSize)
+ if !ok {
+ return nil, false
+ }
+
+ size := int(binary.Size(msg))
+ msgBytes, ok := b.Extract(size)
+ if !ok {
+ return nil, false
+ }
+ binary.Unmarshal(msgBytes, usermem.ByteOrder, msg)
+
+ numPad := alignPad(linux.NetlinkMessageHeaderSize+size, linux.NLMSG_ALIGNTO)
+ // Linux permits the last message not being aligned, just consume all of it.
+ // Ref: net/netlink/af_netlink.c:netlink_rcv_skb
+ if numPad > len(b) {
+ numPad = len(b)
+ }
+ _, ok = b.Extract(numPad)
+ if !ok {
+ return nil, false
+ }
+
+ return AttrsView(b), true
+}
+
// Finalize returns the []byte containing the entire message, with the total
// length set in the message header. The Message must not be modified after
// calling Finalize.
@@ -157,3 +241,48 @@ func (ms *MessageSet) AddMessage(hdr linux.NetlinkMessageHeader) *Message {
ms.Messages = append(ms.Messages, m)
return m
}
+
+// AttrsView is a view into the attributes portion of a netlink message.
+type AttrsView []byte
+
+// Empty returns whether there is no attribute left in v.
+func (v AttrsView) Empty() bool {
+ return len(v) == 0
+}
+
+// ParseFirst parses first netlink attribute at the beginning of v.
+func (v AttrsView) ParseFirst() (hdr linux.NetlinkAttrHeader, value []byte, rest AttrsView, ok bool) {
+ b := BytesView(v)
+
+ hdrBytes, ok := b.Extract(linux.NetlinkAttrHeaderSize)
+ if !ok {
+ return
+ }
+ binary.Unmarshal(hdrBytes, usermem.ByteOrder, &hdr)
+
+ value, ok = b.Extract(int(hdr.Length) - linux.NetlinkAttrHeaderSize)
+ if !ok {
+ return
+ }
+
+ _, ok = b.Extract(alignPad(int(hdr.Length), linux.NLA_ALIGNTO))
+ if !ok {
+ return
+ }
+
+ return hdr, value, AttrsView(b), ok
+}
+
+// BytesView supports extracting data from a byte slice with bounds checking.
+type BytesView []byte
+
+// Extract removes the first n bytes from v and returns it. If n is out of
+// bounds, it returns false.
+func (v *BytesView) Extract(n int) ([]byte, bool) {
+ if n < 0 || n > len(*v) {
+ return nil, false
+ }
+ extracted := (*v)[:n]
+ *v = (*v)[n:]
+ return extracted, true
+}
diff --git a/pkg/sentry/socket/netlink/message_test.go b/pkg/sentry/socket/netlink/message_test.go
new file mode 100644
index 000000000..ef13d9386
--- /dev/null
+++ b/pkg/sentry/socket/netlink/message_test.go
@@ -0,0 +1,312 @@
+// 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 message_test
+
+import (
+ "bytes"
+ "reflect"
+ "testing"
+
+ "gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/sentry/socket/netlink"
+)
+
+type dummyNetlinkMsg struct {
+ Foo uint16
+}
+
+func TestParseMessage(t *testing.T) {
+ tests := []struct {
+ desc string
+ input []byte
+
+ header linux.NetlinkMessageHeader
+ dataMsg *dummyNetlinkMsg
+ restLen int
+ ok bool
+ }{
+ {
+ desc: "valid",
+ input: []byte{
+ 0x14, 0x00, 0x00, 0x00, // Length
+ 0x01, 0x00, // Type
+ 0x02, 0x00, // Flags
+ 0x03, 0x00, 0x00, 0x00, // Seq
+ 0x04, 0x00, 0x00, 0x00, // PortID
+ 0x30, 0x31, 0x00, 0x00, // Data message with 2 bytes padding
+ },
+ header: linux.NetlinkMessageHeader{
+ Length: 20,
+ Type: 1,
+ Flags: 2,
+ Seq: 3,
+ PortID: 4,
+ },
+ dataMsg: &dummyNetlinkMsg{
+ Foo: 0x3130,
+ },
+ restLen: 0,
+ ok: true,
+ },
+ {
+ desc: "valid with next message",
+ input: []byte{
+ 0x14, 0x00, 0x00, 0x00, // Length
+ 0x01, 0x00, // Type
+ 0x02, 0x00, // Flags
+ 0x03, 0x00, 0x00, 0x00, // Seq
+ 0x04, 0x00, 0x00, 0x00, // PortID
+ 0x30, 0x31, 0x00, 0x00, // Data message with 2 bytes padding
+ 0xFF, // Next message (rest)
+ },
+ header: linux.NetlinkMessageHeader{
+ Length: 20,
+ Type: 1,
+ Flags: 2,
+ Seq: 3,
+ PortID: 4,
+ },
+ dataMsg: &dummyNetlinkMsg{
+ Foo: 0x3130,
+ },
+ restLen: 1,
+ ok: true,
+ },
+ {
+ desc: "valid for last message without padding",
+ input: []byte{
+ 0x12, 0x00, 0x00, 0x00, // Length
+ 0x01, 0x00, // Type
+ 0x02, 0x00, // Flags
+ 0x03, 0x00, 0x00, 0x00, // Seq
+ 0x04, 0x00, 0x00, 0x00, // PortID
+ 0x30, 0x31, // Data message
+ },
+ header: linux.NetlinkMessageHeader{
+ Length: 18,
+ Type: 1,
+ Flags: 2,
+ Seq: 3,
+ PortID: 4,
+ },
+ dataMsg: &dummyNetlinkMsg{
+ Foo: 0x3130,
+ },
+ restLen: 0,
+ ok: true,
+ },
+ {
+ desc: "valid for last message not to be aligned",
+ input: []byte{
+ 0x13, 0x00, 0x00, 0x00, // Length
+ 0x01, 0x00, // Type
+ 0x02, 0x00, // Flags
+ 0x03, 0x00, 0x00, 0x00, // Seq
+ 0x04, 0x00, 0x00, 0x00, // PortID
+ 0x30, 0x31, // Data message
+ 0x00, // Excessive 1 byte permitted at end
+ },
+ header: linux.NetlinkMessageHeader{
+ Length: 19,
+ Type: 1,
+ Flags: 2,
+ Seq: 3,
+ PortID: 4,
+ },
+ dataMsg: &dummyNetlinkMsg{
+ Foo: 0x3130,
+ },
+ restLen: 0,
+ ok: true,
+ },
+ {
+ desc: "header.Length too short",
+ input: []byte{
+ 0x04, 0x00, 0x00, 0x00, // Length
+ 0x01, 0x00, // Type
+ 0x02, 0x00, // Flags
+ 0x03, 0x00, 0x00, 0x00, // Seq
+ 0x04, 0x00, 0x00, 0x00, // PortID
+ 0x30, 0x31, 0x00, 0x00, // Data message with 2 bytes padding
+ },
+ ok: false,
+ },
+ {
+ desc: "header.Length too long",
+ input: []byte{
+ 0xFF, 0xFF, 0x00, 0x00, // Length
+ 0x01, 0x00, // Type
+ 0x02, 0x00, // Flags
+ 0x03, 0x00, 0x00, 0x00, // Seq
+ 0x04, 0x00, 0x00, 0x00, // PortID
+ 0x30, 0x31, 0x00, 0x00, // Data message with 2 bytes padding
+ },
+ ok: false,
+ },
+ {
+ desc: "header incomplete",
+ input: []byte{
+ 0x04, 0x00, 0x00, 0x00, // Length
+ },
+ ok: false,
+ },
+ {
+ desc: "empty message",
+ input: []byte{},
+ ok: false,
+ },
+ }
+ for _, test := range tests {
+ msg, rest, ok := netlink.ParseMessage(test.input)
+ if ok != test.ok {
+ t.Errorf("%v: got ok = %v, want = %v", test.desc, ok, test.ok)
+ continue
+ }
+ if !test.ok {
+ continue
+ }
+ if !reflect.DeepEqual(msg.Header(), test.header) {
+ t.Errorf("%v: got hdr = %+v, want = %+v", test.desc, msg.Header(), test.header)
+ }
+
+ dataMsg := &dummyNetlinkMsg{}
+ _, dataOk := msg.GetData(dataMsg)
+ if !dataOk {
+ t.Errorf("%v: GetData.ok = %v, want = true", test.desc, dataOk)
+ } else if !reflect.DeepEqual(dataMsg, test.dataMsg) {
+ t.Errorf("%v: GetData.msg = %+v, want = %+v", test.desc, dataMsg, test.dataMsg)
+ }
+
+ if got, want := rest, test.input[len(test.input)-test.restLen:]; !bytes.Equal(got, want) {
+ t.Errorf("%v: got rest = %v, want = %v", test.desc, got, want)
+ }
+ }
+}
+
+func TestAttrView(t *testing.T) {
+ tests := []struct {
+ desc string
+ input []byte
+
+ // Outputs for ParseFirst.
+ hdr linux.NetlinkAttrHeader
+ value []byte
+ restLen int
+ ok bool
+
+ // Outputs for Empty.
+ isEmpty bool
+ }{
+ {
+ desc: "valid",
+ input: []byte{
+ 0x06, 0x00, // Length
+ 0x01, 0x00, // Type
+ 0x30, 0x31, 0x00, 0x00, // Data with 2 bytes padding
+ },
+ hdr: linux.NetlinkAttrHeader{
+ Length: 6,
+ Type: 1,
+ },
+ value: []byte{0x30, 0x31},
+ restLen: 0,
+ ok: true,
+ isEmpty: false,
+ },
+ {
+ desc: "at alignment",
+ input: []byte{
+ 0x08, 0x00, // Length
+ 0x01, 0x00, // Type
+ 0x30, 0x31, 0x32, 0x33, // Data
+ },
+ hdr: linux.NetlinkAttrHeader{
+ Length: 8,
+ Type: 1,
+ },
+ value: []byte{0x30, 0x31, 0x32, 0x33},
+ restLen: 0,
+ ok: true,
+ isEmpty: false,
+ },
+ {
+ desc: "at alignment with rest data",
+ input: []byte{
+ 0x08, 0x00, // Length
+ 0x01, 0x00, // Type
+ 0x30, 0x31, 0x32, 0x33, // Data
+ 0xFF, 0xFE, // Rest data
+ },
+ hdr: linux.NetlinkAttrHeader{
+ Length: 8,
+ Type: 1,
+ },
+ value: []byte{0x30, 0x31, 0x32, 0x33},
+ restLen: 2,
+ ok: true,
+ isEmpty: false,
+ },
+ {
+ desc: "hdr.Length too long",
+ input: []byte{
+ 0xFF, 0x00, // Length
+ 0x01, 0x00, // Type
+ 0x30, 0x31, 0x32, 0x33, // Data
+ },
+ ok: false,
+ isEmpty: false,
+ },
+ {
+ desc: "hdr.Length too short",
+ input: []byte{
+ 0x01, 0x00, // Length
+ 0x01, 0x00, // Type
+ 0x30, 0x31, 0x32, 0x33, // Data
+ },
+ ok: false,
+ isEmpty: false,
+ },
+ {
+ desc: "empty",
+ input: []byte{},
+ ok: false,
+ isEmpty: true,
+ },
+ }
+ for _, test := range tests {
+ attrs := netlink.AttrsView(test.input)
+
+ // Test ParseFirst().
+ hdr, value, rest, ok := attrs.ParseFirst()
+ if ok != test.ok {
+ t.Errorf("%v: got ok = %v, want = %v", test.desc, ok, test.ok)
+ } else if test.ok {
+ if !reflect.DeepEqual(hdr, test.hdr) {
+ t.Errorf("%v: got hdr = %+v, want = %+v", test.desc, hdr, test.hdr)
+ }
+ if !bytes.Equal(value, test.value) {
+ t.Errorf("%v: got value = %v, want = %v", test.desc, value, test.value)
+ }
+ if wantRest := test.input[len(test.input)-test.restLen:]; !bytes.Equal(rest, wantRest) {
+ t.Errorf("%v: got rest = %v, want = %v", test.desc, rest, wantRest)
+ }
+ }
+
+ // Test Empty().
+ if got, want := attrs.Empty(), test.isEmpty; got != want {
+ t.Errorf("%v: got empty = %v, want = %v", test.desc, got, want)
+ }
+ }
+}
diff --git a/pkg/sentry/socket/netlink/provider.go b/pkg/sentry/socket/netlink/provider.go
index 07f860a49..b0dc70e5c 100644
--- a/pkg/sentry/socket/netlink/provider.go
+++ b/pkg/sentry/socket/netlink/provider.go
@@ -42,7 +42,7 @@ type Protocol interface {
// If err == nil, any messages added to ms will be sent back to the
// other end of the socket. Setting ms.Multi will cause an NLMSG_DONE
// message to be sent even if ms contains no messages.
- ProcessMessage(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *MessageSet) *syserr.Error
+ ProcessMessage(ctx context.Context, msg *Message, ms *MessageSet) *syserr.Error
}
// Provider is a function that creates a new Protocol for a specific netlink
diff --git a/pkg/sentry/socket/netlink/route/BUILD b/pkg/sentry/socket/netlink/route/BUILD
index 622a1eafc..93127398d 100644
--- a/pkg/sentry/socket/netlink/route/BUILD
+++ b/pkg/sentry/socket/netlink/route/BUILD
@@ -10,13 +10,11 @@ go_library(
visibility = ["//pkg/sentry:internal"],
deps = [
"//pkg/abi/linux",
- "//pkg/binary",
"//pkg/context",
"//pkg/sentry/inet",
"//pkg/sentry/kernel",
"//pkg/sentry/kernel/auth",
"//pkg/sentry/socket/netlink",
"//pkg/syserr",
- "//pkg/usermem",
],
)
diff --git a/pkg/sentry/socket/netlink/route/protocol.go b/pkg/sentry/socket/netlink/route/protocol.go
index 2b3c7f5b3..c84d8bd7c 100644
--- a/pkg/sentry/socket/netlink/route/protocol.go
+++ b/pkg/sentry/socket/netlink/route/protocol.go
@@ -17,16 +17,15 @@ package route
import (
"bytes"
+ "syscall"
"gvisor.dev/gvisor/pkg/abi/linux"
- "gvisor.dev/gvisor/pkg/binary"
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/sentry/inet"
"gvisor.dev/gvisor/pkg/sentry/kernel"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
"gvisor.dev/gvisor/pkg/sentry/socket/netlink"
"gvisor.dev/gvisor/pkg/syserr"
- "gvisor.dev/gvisor/pkg/usermem"
)
// commandKind describes the operational class of a message type.
@@ -69,13 +68,7 @@ func (p *Protocol) CanSend() bool {
}
// dumpLinks handles RTM_GETLINK dump requests.
-func (p *Protocol) dumpLinks(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *netlink.MessageSet) *syserr.Error {
- // TODO(b/68878065): Only the dump variant of the types below are
- // supported.
- if hdr.Flags&linux.NLM_F_DUMP != linux.NLM_F_DUMP {
- return syserr.ErrNotSupported
- }
-
+func (p *Protocol) dumpLinks(ctx context.Context, msg *netlink.Message, ms *netlink.MessageSet) *syserr.Error {
// NLM_F_DUMP + RTM_GETLINK messages are supposed to include an
// ifinfomsg. However, Linux <3.9 only checked for rtgenmsg, and some
// userspace applications (including glibc) still include rtgenmsg.
@@ -99,44 +92,105 @@ func (p *Protocol) dumpLinks(ctx context.Context, hdr linux.NetlinkMessageHeader
return nil
}
- for id, i := range stack.Interfaces() {
- m := ms.AddMessage(linux.NetlinkMessageHeader{
- Type: linux.RTM_NEWLINK,
- })
+ for idx, i := range stack.Interfaces() {
+ addNewLinkMessage(ms, idx, i)
+ }
- m.Put(linux.InterfaceInfoMessage{
- Family: linux.AF_UNSPEC,
- Type: i.DeviceType,
- Index: id,
- Flags: i.Flags,
- })
+ return nil
+}
- m.PutAttrString(linux.IFLA_IFNAME, i.Name)
- m.PutAttr(linux.IFLA_MTU, i.MTU)
+// getLinks handles RTM_GETLINK requests.
+func (p *Protocol) getLink(ctx context.Context, msg *netlink.Message, ms *netlink.MessageSet) *syserr.Error {
+ stack := inet.StackFromContext(ctx)
+ if stack == nil {
+ // No network devices.
+ return nil
+ }
- mac := make([]byte, 6)
- brd := mac
- if len(i.Addr) > 0 {
- mac = i.Addr
- brd = bytes.Repeat([]byte{0xff}, len(i.Addr))
+ // Parse message.
+ var ifi linux.InterfaceInfoMessage
+ attrs, ok := msg.GetData(&ifi)
+ if !ok {
+ return syserr.ErrInvalidArgument
+ }
+
+ // Parse attributes.
+ var byName []byte
+ for !attrs.Empty() {
+ ahdr, value, rest, ok := attrs.ParseFirst()
+ if !ok {
+ return syserr.ErrInvalidArgument
}
- m.PutAttr(linux.IFLA_ADDRESS, mac)
- m.PutAttr(linux.IFLA_BROADCAST, brd)
+ attrs = rest
- // TODO(gvisor.dev/issue/578): There are many more attributes.
+ switch ahdr.Type {
+ case linux.IFLA_IFNAME:
+ if len(value) < 1 {
+ return syserr.ErrInvalidArgument
+ }
+ byName = value[:len(value)-1]
+
+ // TODO(gvisor.dev/issue/578): Support IFLA_EXT_MASK.
+ }
}
+ found := false
+ for idx, i := range stack.Interfaces() {
+ switch {
+ case ifi.Index > 0:
+ if idx != ifi.Index {
+ continue
+ }
+ case byName != nil:
+ if string(byName) != i.Name {
+ continue
+ }
+ default:
+ // Criteria not specified.
+ return syserr.ErrInvalidArgument
+ }
+
+ addNewLinkMessage(ms, idx, i)
+ found = true
+ break
+ }
+ if !found {
+ return syserr.ErrNoDevice
+ }
return nil
}
-// dumpAddrs handles RTM_GETADDR dump requests.
-func (p *Protocol) dumpAddrs(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *netlink.MessageSet) *syserr.Error {
- // TODO(b/68878065): Only the dump variant of the types below are
- // supported.
- if hdr.Flags&linux.NLM_F_DUMP != linux.NLM_F_DUMP {
- return syserr.ErrNotSupported
+// addNewLinkMessage appends RTM_NEWLINK message for the given interface into
+// the message set.
+func addNewLinkMessage(ms *netlink.MessageSet, idx int32, i inet.Interface) {
+ m := ms.AddMessage(linux.NetlinkMessageHeader{
+ Type: linux.RTM_NEWLINK,
+ })
+
+ m.Put(linux.InterfaceInfoMessage{
+ Family: linux.AF_UNSPEC,
+ Type: i.DeviceType,
+ Index: idx,
+ Flags: i.Flags,
+ })
+
+ m.PutAttrString(linux.IFLA_IFNAME, i.Name)
+ m.PutAttr(linux.IFLA_MTU, i.MTU)
+
+ mac := make([]byte, 6)
+ brd := mac
+ if len(i.Addr) > 0 {
+ mac = i.Addr
+ brd = bytes.Repeat([]byte{0xff}, len(i.Addr))
}
+ m.PutAttr(linux.IFLA_ADDRESS, mac)
+ m.PutAttr(linux.IFLA_BROADCAST, brd)
+
+ // TODO(gvisor.dev/issue/578): There are many more attributes.
+}
+// dumpAddrs handles RTM_GETADDR dump requests.
+func (p *Protocol) dumpAddrs(ctx context.Context, msg *netlink.Message, ms *netlink.MessageSet) *syserr.Error {
// RTM_GETADDR dump requests need not contain anything more than the
// netlink header and 1 byte protocol family common to all
// NETLINK_ROUTE requests.
@@ -168,6 +222,7 @@ func (p *Protocol) dumpAddrs(ctx context.Context, hdr linux.NetlinkMessageHeader
Index: uint32(id),
})
+ m.PutAttr(linux.IFA_LOCAL, []byte(a.Addr))
m.PutAttr(linux.IFA_ADDRESS, []byte(a.Addr))
// TODO(gvisor.dev/issue/578): There are many more attributes.
@@ -252,12 +307,12 @@ func fillRoute(routes []inet.Route, addr []byte) (inet.Route, *syserr.Error) {
}
// parseForDestination parses a message as format of RouteMessage-RtAttr-dst.
-func parseForDestination(data []byte) ([]byte, *syserr.Error) {
+func parseForDestination(msg *netlink.Message) ([]byte, *syserr.Error) {
var rtMsg linux.RouteMessage
- if len(data) < linux.SizeOfRouteMessage {
+ attrs, ok := msg.GetData(&rtMsg)
+ if !ok {
return nil, syserr.ErrInvalidArgument
}
- binary.Unmarshal(data[:linux.SizeOfRouteMessage], usermem.ByteOrder, &rtMsg)
// iproute2 added the RTM_F_LOOKUP_TABLE flag in version v4.4.0. See
// commit bc234301af12. Note we don't check this flag for backward
// compatibility.
@@ -265,26 +320,15 @@ func parseForDestination(data []byte) ([]byte, *syserr.Error) {
return nil, syserr.ErrNotSupported
}
- data = data[linux.SizeOfRouteMessage:]
-
- // TODO(gvisor.dev/issue/1611): Add generic attribute parsing.
- var rtAttr linux.RtAttr
- if len(data) < linux.SizeOfRtAttr {
- return nil, syserr.ErrInvalidArgument
+ // Expect first attribute is RTA_DST.
+ if hdr, value, _, ok := attrs.ParseFirst(); ok && hdr.Type == linux.RTA_DST {
+ return value, nil
}
- binary.Unmarshal(data[:linux.SizeOfRtAttr], usermem.ByteOrder, &rtAttr)
- if rtAttr.Type != linux.RTA_DST {
- return nil, syserr.ErrInvalidArgument
- }
-
- if len(data) < int(rtAttr.Len) {
- return nil, syserr.ErrInvalidArgument
- }
- return data[linux.SizeOfRtAttr:rtAttr.Len], nil
+ return nil, syserr.ErrInvalidArgument
}
// dumpRoutes handles RTM_GETROUTE requests.
-func (p *Protocol) dumpRoutes(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *netlink.MessageSet) *syserr.Error {
+func (p *Protocol) dumpRoutes(ctx context.Context, msg *netlink.Message, ms *netlink.MessageSet) *syserr.Error {
// RTM_GETROUTE dump requests need not contain anything more than the
// netlink header and 1 byte protocol family common to all
// NETLINK_ROUTE requests.
@@ -295,10 +339,11 @@ func (p *Protocol) dumpRoutes(ctx context.Context, hdr linux.NetlinkMessageHeade
return nil
}
+ hdr := msg.Header()
routeTables := stack.RouteTable()
if hdr.Flags == linux.NLM_F_REQUEST {
- dst, err := parseForDestination(data)
+ dst, err := parseForDestination(msg)
if err != nil {
return err
}
@@ -357,10 +402,55 @@ func (p *Protocol) dumpRoutes(ctx context.Context, hdr linux.NetlinkMessageHeade
return nil
}
+// newAddr handles RTM_NEWADDR requests.
+func (p *Protocol) newAddr(ctx context.Context, msg *netlink.Message, ms *netlink.MessageSet) *syserr.Error {
+ stack := inet.StackFromContext(ctx)
+ if stack == nil {
+ // No network stack.
+ return syserr.ErrProtocolNotSupported
+ }
+
+ var ifa linux.InterfaceAddrMessage
+ attrs, ok := msg.GetData(&ifa)
+ if !ok {
+ return syserr.ErrInvalidArgument
+ }
+
+ for !attrs.Empty() {
+ ahdr, value, rest, ok := attrs.ParseFirst()
+ if !ok {
+ return syserr.ErrInvalidArgument
+ }
+ attrs = rest
+
+ switch ahdr.Type {
+ case linux.IFA_LOCAL:
+ err := stack.AddInterfaceAddr(int32(ifa.Index), inet.InterfaceAddr{
+ Family: ifa.Family,
+ PrefixLen: ifa.PrefixLen,
+ Flags: ifa.Flags,
+ Addr: value,
+ })
+ if err == syscall.EEXIST {
+ flags := msg.Header().Flags
+ if flags&linux.NLM_F_EXCL != 0 {
+ return syserr.ErrExists
+ }
+ } else if err != nil {
+ return syserr.ErrInvalidArgument
+ }
+ }
+ }
+ return nil
+}
+
// ProcessMessage implements netlink.Protocol.ProcessMessage.
-func (p *Protocol) ProcessMessage(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *netlink.MessageSet) *syserr.Error {
+func (p *Protocol) ProcessMessage(ctx context.Context, msg *netlink.Message, ms *netlink.MessageSet) *syserr.Error {
+ hdr := msg.Header()
+
// All messages start with a 1 byte protocol family.
- if len(data) < 1 {
+ var family uint8
+ if _, ok := msg.GetData(&family); !ok {
// Linux ignores messages missing the protocol family. See
// net/core/rtnetlink.c:rtnetlink_rcv_msg.
return nil
@@ -374,16 +464,32 @@ func (p *Protocol) ProcessMessage(ctx context.Context, hdr linux.NetlinkMessageH
}
}
- switch hdr.Type {
- case linux.RTM_GETLINK:
- return p.dumpLinks(ctx, hdr, data, ms)
- case linux.RTM_GETADDR:
- return p.dumpAddrs(ctx, hdr, data, ms)
- case linux.RTM_GETROUTE:
- return p.dumpRoutes(ctx, hdr, data, ms)
- default:
- return syserr.ErrNotSupported
+ if hdr.Flags&linux.NLM_F_DUMP == linux.NLM_F_DUMP {
+ // TODO(b/68878065): Only the dump variant of the types below are
+ // supported.
+ switch hdr.Type {
+ case linux.RTM_GETLINK:
+ return p.dumpLinks(ctx, msg, ms)
+ case linux.RTM_GETADDR:
+ return p.dumpAddrs(ctx, msg, ms)
+ case linux.RTM_GETROUTE:
+ return p.dumpRoutes(ctx, msg, ms)
+ default:
+ return syserr.ErrNotSupported
+ }
+ } else if hdr.Flags&linux.NLM_F_REQUEST == linux.NLM_F_REQUEST {
+ switch hdr.Type {
+ case linux.RTM_GETLINK:
+ return p.getLink(ctx, msg, ms)
+ case linux.RTM_GETROUTE:
+ return p.dumpRoutes(ctx, msg, ms)
+ case linux.RTM_NEWADDR:
+ return p.newAddr(ctx, msg, ms)
+ default:
+ return syserr.ErrNotSupported
+ }
}
+ return syserr.ErrNotSupported
}
// init registers the NETLINK_ROUTE provider.
diff --git a/pkg/sentry/socket/netlink/socket.go b/pkg/sentry/socket/netlink/socket.go
index c4b95debb..2ca02567d 100644
--- a/pkg/sentry/socket/netlink/socket.go
+++ b/pkg/sentry/socket/netlink/socket.go
@@ -644,47 +644,38 @@ func (s *Socket) sendResponse(ctx context.Context, ms *MessageSet) *syserr.Error
return nil
}
-func (s *Socket) dumpErrorMesage(ctx context.Context, hdr linux.NetlinkMessageHeader, ms *MessageSet, err *syserr.Error) *syserr.Error {
+func dumpErrorMesage(hdr linux.NetlinkMessageHeader, ms *MessageSet, err *syserr.Error) {
m := ms.AddMessage(linux.NetlinkMessageHeader{
Type: linux.NLMSG_ERROR,
})
-
m.Put(linux.NetlinkErrorMessage{
Error: int32(-err.ToLinux().Number()),
Header: hdr,
})
- return nil
+}
+func dumpAckMesage(hdr linux.NetlinkMessageHeader, ms *MessageSet) {
+ m := ms.AddMessage(linux.NetlinkMessageHeader{
+ Type: linux.NLMSG_ERROR,
+ })
+ m.Put(linux.NetlinkErrorMessage{
+ Error: 0,
+ Header: hdr,
+ })
}
// processMessages handles each message in buf, passing it to the protocol
// handler for final handling.
func (s *Socket) processMessages(ctx context.Context, buf []byte) *syserr.Error {
for len(buf) > 0 {
- if len(buf) < linux.NetlinkMessageHeaderSize {
+ msg, rest, ok := ParseMessage(buf)
+ if !ok {
// Linux ignores messages that are too short. See
// net/netlink/af_netlink.c:netlink_rcv_skb.
break
}
-
- var hdr linux.NetlinkMessageHeader
- binary.Unmarshal(buf[:linux.NetlinkMessageHeaderSize], usermem.ByteOrder, &hdr)
-
- if hdr.Length < linux.NetlinkMessageHeaderSize || uint64(hdr.Length) > uint64(len(buf)) {
- // Linux ignores malformed messages. See
- // net/netlink/af_netlink.c:netlink_rcv_skb.
- break
- }
-
- // Data from this message.
- data := buf[linux.NetlinkMessageHeaderSize:hdr.Length]
-
- // Advance to the next message.
- next := alignUp(int(hdr.Length), linux.NLMSG_ALIGNTO)
- if next >= len(buf)-1 {
- next = len(buf) - 1
- }
- buf = buf[next:]
+ buf = rest
+ hdr := msg.Header()
// Ignore control messages.
if hdr.Type < linux.NLMSG_MIN_TYPE {
@@ -692,19 +683,10 @@ func (s *Socket) processMessages(ctx context.Context, buf []byte) *syserr.Error
}
ms := NewMessageSet(s.portID, hdr.Seq)
- var err *syserr.Error
- // TODO(b/68877377): ACKs not supported yet.
- if hdr.Flags&linux.NLM_F_ACK == linux.NLM_F_ACK {
- err = syserr.ErrNotSupported
- } else {
-
- err = s.protocol.ProcessMessage(ctx, hdr, data, ms)
- }
- if err != nil {
- ms = NewMessageSet(s.portID, hdr.Seq)
- if err := s.dumpErrorMesage(ctx, hdr, ms, err); err != nil {
- return err
- }
+ if err := s.protocol.ProcessMessage(ctx, msg, ms); err != nil {
+ dumpErrorMesage(hdr, ms, err)
+ } else if hdr.Flags&linux.NLM_F_ACK == linux.NLM_F_ACK {
+ dumpAckMesage(hdr, ms)
}
if err := s.sendResponse(ctx, ms); err != nil {
diff --git a/pkg/sentry/socket/netlink/uevent/protocol.go b/pkg/sentry/socket/netlink/uevent/protocol.go
index 1ee4296bc..029ba21b5 100644
--- a/pkg/sentry/socket/netlink/uevent/protocol.go
+++ b/pkg/sentry/socket/netlink/uevent/protocol.go
@@ -49,7 +49,7 @@ func (p *Protocol) CanSend() bool {
}
// ProcessMessage implements netlink.Protocol.ProcessMessage.
-func (p *Protocol) ProcessMessage(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *netlink.MessageSet) *syserr.Error {
+func (p *Protocol) ProcessMessage(ctx context.Context, msg *netlink.Message, ms *netlink.MessageSet) *syserr.Error {
// Silently ignore all messages.
return nil
}
diff --git a/pkg/sentry/socket/netstack/stack.go b/pkg/sentry/socket/netstack/stack.go
index 31ea66eca..0692482e9 100644
--- a/pkg/sentry/socket/netstack/stack.go
+++ b/pkg/sentry/socket/netstack/stack.go
@@ -20,6 +20,8 @@ import (
"gvisor.dev/gvisor/pkg/sentry/inet"
"gvisor.dev/gvisor/pkg/sentry/socket/netfilter"
"gvisor.dev/gvisor/pkg/syserr"
+ "gvisor.dev/gvisor/pkg/syserror"
+ "gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/header"
"gvisor.dev/gvisor/pkg/tcpip/iptables"
"gvisor.dev/gvisor/pkg/tcpip/network/ipv4"
@@ -88,6 +90,59 @@ func (s *Stack) InterfaceAddrs() map[int32][]inet.InterfaceAddr {
return nicAddrs
}
+// AddInterfaceAddr implements inet.Stack.AddInterfaceAddr.
+func (s *Stack) AddInterfaceAddr(idx int32, addr inet.InterfaceAddr) error {
+ var (
+ protocol tcpip.NetworkProtocolNumber
+ address tcpip.Address
+ )
+ switch addr.Family {
+ case linux.AF_INET:
+ if len(addr.Addr) < header.IPv4AddressSize {
+ return syserror.EINVAL
+ }
+ if addr.PrefixLen > header.IPv4AddressSize*8 {
+ return syserror.EINVAL
+ }
+ protocol = ipv4.ProtocolNumber
+ address = tcpip.Address(addr.Addr[:header.IPv4AddressSize])
+
+ case linux.AF_INET6:
+ if len(addr.Addr) < header.IPv6AddressSize {
+ return syserror.EINVAL
+ }
+ if addr.PrefixLen > header.IPv6AddressSize*8 {
+ return syserror.EINVAL
+ }
+ protocol = ipv6.ProtocolNumber
+ address = tcpip.Address(addr.Addr[:header.IPv6AddressSize])
+
+ default:
+ return syserror.ENOTSUP
+ }
+
+ protocolAddress := tcpip.ProtocolAddress{
+ Protocol: protocol,
+ AddressWithPrefix: tcpip.AddressWithPrefix{
+ Address: address,
+ PrefixLen: int(addr.PrefixLen),
+ },
+ }
+
+ // Attach address to interface.
+ if err := s.Stack.AddProtocolAddressWithOptions(tcpip.NICID(idx), protocolAddress, stack.CanBePrimaryEndpoint); err != nil {
+ return syserr.TranslateNetstackError(err).ToError()
+ }
+
+ // Add route for local network.
+ s.Stack.AddRoute(tcpip.Route{
+ Destination: protocolAddress.AddressWithPrefix.Subnet(),
+ Gateway: "", // No gateway for local network.
+ NIC: tcpip.NICID(idx),
+ })
+ return nil
+}
+
// TCPReceiveBufferSize implements inet.Stack.TCPReceiveBufferSize.
func (s *Stack) TCPReceiveBufferSize() (inet.TCPBufferSize, error) {
var rs tcp.ReceiveBufferSizeOption
diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go
index 7057b110e..b793f1d74 100644
--- a/pkg/tcpip/stack/stack.go
+++ b/pkg/tcpip/stack/stack.go
@@ -795,6 +795,8 @@ func (s *Stack) Forwarding() bool {
// SetRouteTable assigns the route table to be used by this stack. It
// specifies which NIC to use for given destination address ranges.
+//
+// This method takes ownership of the table.
func (s *Stack) SetRouteTable(table []tcpip.Route) {
s.mu.Lock()
defer s.mu.Unlock()
@@ -809,6 +811,13 @@ func (s *Stack) GetRouteTable() []tcpip.Route {
return append([]tcpip.Route(nil), s.routeTable...)
}
+// AddRoute appends a route to the route table.
+func (s *Stack) AddRoute(route tcpip.Route) {
+ s.mu.Lock()
+ defer s.mu.Unlock()
+ s.routeTable = append(s.routeTable, route)
+}
+
// NewEndpoint creates a new transport layer endpoint of the given protocol.
func (s *Stack) NewEndpoint(transport tcpip.TransportProtocolNumber, network tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) {
t, ok := s.transportProtocols[transport]
diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD
index 273b014d6..f2e3c7072 100644
--- a/test/syscalls/linux/BUILD
+++ b/test/syscalls/linux/BUILD
@@ -2769,9 +2769,11 @@ cc_binary(
deps = [
":socket_netlink_util",
":socket_test_util",
+ "//test/util:capability_util",
"//test/util:cleanup",
"//test/util:file_descriptor",
"@com_google_absl//absl/strings:str_format",
+ "@com_google_absl//absl/types:optional",
gtest,
"//test/util:test_main",
"//test/util:test_util",
diff --git a/test/syscalls/linux/socket_netlink_route.cc b/test/syscalls/linux/socket_netlink_route.cc
index 1e28e658d..e5aed1eec 100644
--- a/test/syscalls/linux/socket_netlink_route.cc
+++ b/test/syscalls/linux/socket_netlink_route.cc
@@ -14,6 +14,7 @@
#include <arpa/inet.h>
#include <ifaddrs.h>
+#include <linux/if.h>
#include <linux/netlink.h>
#include <linux/rtnetlink.h>
#include <sys/socket.h>
@@ -25,8 +26,10 @@
#include "gtest/gtest.h"
#include "absl/strings/str_format.h"
+#include "absl/types/optional.h"
#include "test/syscalls/linux/socket_netlink_util.h"
#include "test/syscalls/linux/socket_test_util.h"
+#include "test/util/capability_util.h"
#include "test/util/cleanup.h"
#include "test/util/file_descriptor.h"
#include "test/util/test_util.h"
@@ -38,6 +41,8 @@ namespace testing {
namespace {
+constexpr uint32_t kSeq = 12345;
+
using ::testing::AnyOf;
using ::testing::Eq;
@@ -113,58 +118,224 @@ void CheckGetLinkResponse(const struct nlmsghdr* hdr, int seq, int port) {
// TODO(mpratt): Check ifinfomsg contents and following attrs.
}
+PosixError DumpLinks(
+ const FileDescriptor& fd, uint32_t seq,
+ const std::function<void(const struct nlmsghdr* hdr)>& fn) {
+ struct request {
+ struct nlmsghdr hdr;
+ struct ifinfomsg ifm;
+ };
+
+ struct request req = {};
+ req.hdr.nlmsg_len = sizeof(req);
+ req.hdr.nlmsg_type = RTM_GETLINK;
+ req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
+ req.hdr.nlmsg_seq = seq;
+ req.ifm.ifi_family = AF_UNSPEC;
+
+ return NetlinkRequestResponse(fd, &req, sizeof(req), fn, false);
+}
+
TEST(NetlinkRouteTest, GetLinkDump) {
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
uint32_t port = ASSERT_NO_ERRNO_AND_VALUE(NetlinkPortID(fd.get()));
+ // Loopback is common among all tests, check that it's found.
+ bool loopbackFound = false;
+ ASSERT_NO_ERRNO(DumpLinks(fd, kSeq, [&](const struct nlmsghdr* hdr) {
+ CheckGetLinkResponse(hdr, kSeq, port);
+ if (hdr->nlmsg_type != RTM_NEWLINK) {
+ return;
+ }
+ ASSERT_GE(hdr->nlmsg_len, NLMSG_SPACE(sizeof(struct ifinfomsg)));
+ const struct ifinfomsg* msg =
+ reinterpret_cast<const struct ifinfomsg*>(NLMSG_DATA(hdr));
+ std::cout << "Found interface idx=" << msg->ifi_index
+ << ", type=" << std::hex << msg->ifi_type;
+ if (msg->ifi_type == ARPHRD_LOOPBACK) {
+ loopbackFound = true;
+ EXPECT_NE(msg->ifi_flags & IFF_LOOPBACK, 0);
+ }
+ }));
+ EXPECT_TRUE(loopbackFound);
+}
+
+struct Link {
+ int index;
+ std::string name;
+};
+
+PosixErrorOr<absl::optional<Link>> FindLoopbackLink() {
+ ASSIGN_OR_RETURN_ERRNO(FileDescriptor fd, NetlinkBoundSocket(NETLINK_ROUTE));
+
+ absl::optional<Link> link;
+ RETURN_IF_ERRNO(DumpLinks(fd, kSeq, [&](const struct nlmsghdr* hdr) {
+ if (hdr->nlmsg_type != RTM_NEWLINK ||
+ hdr->nlmsg_len < NLMSG_SPACE(sizeof(struct ifinfomsg))) {
+ return;
+ }
+ const struct ifinfomsg* msg =
+ reinterpret_cast<const struct ifinfomsg*>(NLMSG_DATA(hdr));
+ if (msg->ifi_type == ARPHRD_LOOPBACK) {
+ const auto* rta = FindRtAttr(hdr, msg, IFLA_IFNAME);
+ if (rta == nullptr) {
+ // Ignore links that do not have a name.
+ return;
+ }
+
+ link = Link();
+ link->index = msg->ifi_index;
+ link->name = std::string(reinterpret_cast<const char*>(RTA_DATA(rta)));
+ }
+ }));
+ return link;
+}
+
+// CheckLinkMsg checks a netlink message against an expected link.
+void CheckLinkMsg(const struct nlmsghdr* hdr, const Link& link) {
+ ASSERT_THAT(hdr->nlmsg_type, Eq(RTM_NEWLINK));
+ ASSERT_GE(hdr->nlmsg_len, NLMSG_SPACE(sizeof(struct ifinfomsg)));
+ const struct ifinfomsg* msg =
+ reinterpret_cast<const struct ifinfomsg*>(NLMSG_DATA(hdr));
+ EXPECT_EQ(msg->ifi_index, link.index);
+
+ const struct rtattr* rta = FindRtAttr(hdr, msg, IFLA_IFNAME);
+ EXPECT_NE(nullptr, rta) << "IFLA_IFNAME not found in message.";
+ if (rta != nullptr) {
+ std::string name(reinterpret_cast<const char*>(RTA_DATA(rta)));
+ EXPECT_EQ(name, link.name);
+ }
+}
+
+TEST(NetlinkRouteTest, GetLinkByIndex) {
+ absl::optional<Link> loopback_link =
+ ASSERT_NO_ERRNO_AND_VALUE(FindLoopbackLink());
+ ASSERT_TRUE(loopback_link.has_value());
+
+ FileDescriptor fd =
+ ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
+
struct request {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
};
- constexpr uint32_t kSeq = 12345;
-
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETLINK;
- req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
+ req.hdr.nlmsg_flags = NLM_F_REQUEST;
req.hdr.nlmsg_seq = kSeq;
req.ifm.ifi_family = AF_UNSPEC;
+ req.ifm.ifi_index = loopback_link->index;
- // Loopback is common among all tests, check that it's found.
- bool loopbackFound = false;
+ bool found = false;
ASSERT_NO_ERRNO(NetlinkRequestResponse(
fd, &req, sizeof(req),
[&](const struct nlmsghdr* hdr) {
- CheckGetLinkResponse(hdr, kSeq, port);
- if (hdr->nlmsg_type != RTM_NEWLINK) {
- return;
- }
- ASSERT_GE(hdr->nlmsg_len, NLMSG_SPACE(sizeof(struct ifinfomsg)));
- const struct ifinfomsg* msg =
- reinterpret_cast<const struct ifinfomsg*>(NLMSG_DATA(hdr));
- std::cout << "Found interface idx=" << msg->ifi_index
- << ", type=" << std::hex << msg->ifi_type;
- if (msg->ifi_type == ARPHRD_LOOPBACK) {
- loopbackFound = true;
- EXPECT_NE(msg->ifi_flags & IFF_LOOPBACK, 0);
- }
+ CheckLinkMsg(hdr, *loopback_link);
+ found = true;
},
false));
- EXPECT_TRUE(loopbackFound);
+ EXPECT_TRUE(found) << "Netlink response does not contain any links.";
}
-TEST(NetlinkRouteTest, MsgHdrMsgUnsuppType) {
+TEST(NetlinkRouteTest, GetLinkByName) {
+ absl::optional<Link> loopback_link =
+ ASSERT_NO_ERRNO_AND_VALUE(FindLoopbackLink());
+ ASSERT_TRUE(loopback_link.has_value());
+
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
struct request {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
+ struct rtattr rtattr;
+ char ifname[IFNAMSIZ];
+ char pad[NLMSG_ALIGNTO + RTA_ALIGNTO];
};
- constexpr uint32_t kSeq = 12345;
+ struct request req = {};
+ req.hdr.nlmsg_type = RTM_GETLINK;
+ req.hdr.nlmsg_flags = NLM_F_REQUEST;
+ req.hdr.nlmsg_seq = kSeq;
+ req.ifm.ifi_family = AF_UNSPEC;
+ req.rtattr.rta_type = IFLA_IFNAME;
+ req.rtattr.rta_len = RTA_LENGTH(loopback_link->name.size() + 1);
+ strncpy(req.ifname, loopback_link->name.c_str(), sizeof(req.ifname));
+ req.hdr.nlmsg_len =
+ NLMSG_LENGTH(sizeof(req.ifm)) + NLMSG_ALIGN(req.rtattr.rta_len);
+
+ bool found = false;
+ ASSERT_NO_ERRNO(NetlinkRequestResponse(
+ fd, &req, sizeof(req),
+ [&](const struct nlmsghdr* hdr) {
+ CheckLinkMsg(hdr, *loopback_link);
+ found = true;
+ },
+ false));
+ EXPECT_TRUE(found) << "Netlink response does not contain any links.";
+}
+
+TEST(NetlinkRouteTest, GetLinkByIndexNotFound) {
+ FileDescriptor fd =
+ ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
+
+ struct request {
+ struct nlmsghdr hdr;
+ struct ifinfomsg ifm;
+ };
+
+ struct request req = {};
+ req.hdr.nlmsg_len = sizeof(req);
+ req.hdr.nlmsg_type = RTM_GETLINK;
+ req.hdr.nlmsg_flags = NLM_F_REQUEST;
+ req.hdr.nlmsg_seq = kSeq;
+ req.ifm.ifi_family = AF_UNSPEC;
+ req.ifm.ifi_index = 1234590;
+
+ EXPECT_THAT(NetlinkRequestAckOrError(fd, kSeq, &req, sizeof(req)),
+ PosixErrorIs(ENODEV, ::testing::_));
+}
+
+TEST(NetlinkRouteTest, GetLinkByNameNotFound) {
+ const std::string name = "nodevice?!";
+
+ FileDescriptor fd =
+ ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
+
+ struct request {
+ struct nlmsghdr hdr;
+ struct ifinfomsg ifm;
+ struct rtattr rtattr;
+ char ifname[IFNAMSIZ];
+ char pad[NLMSG_ALIGNTO + RTA_ALIGNTO];
+ };
+
+ struct request req = {};
+ req.hdr.nlmsg_type = RTM_GETLINK;
+ req.hdr.nlmsg_flags = NLM_F_REQUEST;
+ req.hdr.nlmsg_seq = kSeq;
+ req.ifm.ifi_family = AF_UNSPEC;
+ req.rtattr.rta_type = IFLA_IFNAME;
+ req.rtattr.rta_len = RTA_LENGTH(name.size() + 1);
+ strncpy(req.ifname, name.c_str(), sizeof(req.ifname));
+ req.hdr.nlmsg_len =
+ NLMSG_LENGTH(sizeof(req.ifm)) + NLMSG_ALIGN(req.rtattr.rta_len);
+
+ EXPECT_THAT(NetlinkRequestAckOrError(fd, kSeq, &req, sizeof(req)),
+ PosixErrorIs(ENODEV, ::testing::_));
+}
+
+TEST(NetlinkRouteTest, MsgHdrMsgUnsuppType) {
+ FileDescriptor fd =
+ ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
+
+ struct request {
+ struct nlmsghdr hdr;
+ struct ifinfomsg ifm;
+ };
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
@@ -175,18 +346,8 @@ TEST(NetlinkRouteTest, MsgHdrMsgUnsuppType) {
req.hdr.nlmsg_seq = kSeq;
req.ifm.ifi_family = AF_UNSPEC;
- ASSERT_NO_ERRNO(NetlinkRequestResponse(
- fd, &req, sizeof(req),
- [&](const struct nlmsghdr* hdr) {
- EXPECT_THAT(hdr->nlmsg_type, Eq(NLMSG_ERROR));
- EXPECT_EQ(hdr->nlmsg_seq, kSeq);
- EXPECT_GE(hdr->nlmsg_len, sizeof(*hdr) + sizeof(struct nlmsgerr));
-
- const struct nlmsgerr* msg =
- reinterpret_cast<const struct nlmsgerr*>(NLMSG_DATA(hdr));
- EXPECT_EQ(msg->error, -EOPNOTSUPP);
- },
- true));
+ EXPECT_THAT(NetlinkRequestAckOrError(fd, kSeq, &req, sizeof(req)),
+ PosixErrorIs(EOPNOTSUPP, ::testing::_));
}
TEST(NetlinkRouteTest, MsgHdrMsgTrunc) {
@@ -198,8 +359,6 @@ TEST(NetlinkRouteTest, MsgHdrMsgTrunc) {
struct ifinfomsg ifm;
};
- constexpr uint32_t kSeq = 12345;
-
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETLINK;
@@ -238,8 +397,6 @@ TEST(NetlinkRouteTest, MsgTruncMsgHdrMsgTrunc) {
struct ifinfomsg ifm;
};
- constexpr uint32_t kSeq = 12345;
-
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETLINK;
@@ -282,8 +439,6 @@ TEST(NetlinkRouteTest, ControlMessageIgnored) {
struct ifinfomsg ifm;
};
- constexpr uint32_t kSeq = 12345;
-
struct request req = {};
// This control message is ignored. We still receive a response for the
@@ -317,8 +472,6 @@ TEST(NetlinkRouteTest, GetAddrDump) {
struct rtgenmsg rgm;
};
- constexpr uint32_t kSeq = 12345;
-
struct request req;
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETADDR;
@@ -367,6 +520,57 @@ TEST(NetlinkRouteTest, LookupAll) {
ASSERT_GT(count, 0);
}
+TEST(NetlinkRouteTest, AddAddr) {
+ SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
+
+ absl::optional<Link> loopback_link =
+ ASSERT_NO_ERRNO_AND_VALUE(FindLoopbackLink());
+ ASSERT_TRUE(loopback_link.has_value());
+
+ FileDescriptor fd =
+ ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
+
+ struct request {
+ struct nlmsghdr hdr;
+ struct ifaddrmsg ifa;
+ struct rtattr rtattr;
+ struct in_addr addr;
+ char pad[NLMSG_ALIGNTO + RTA_ALIGNTO];
+ };
+
+ struct request req = {};
+ req.hdr.nlmsg_type = RTM_NEWADDR;
+ req.hdr.nlmsg_seq = kSeq;
+ req.ifa.ifa_family = AF_INET;
+ req.ifa.ifa_prefixlen = 24;
+ req.ifa.ifa_flags = 0;
+ req.ifa.ifa_scope = 0;
+ req.ifa.ifa_index = loopback_link->index;
+ req.rtattr.rta_type = IFA_LOCAL;
+ req.rtattr.rta_len = RTA_LENGTH(sizeof(req.addr));
+ inet_pton(AF_INET, "10.0.0.1", &req.addr);
+ req.hdr.nlmsg_len =
+ NLMSG_LENGTH(sizeof(req.ifa)) + NLMSG_ALIGN(req.rtattr.rta_len);
+
+ // Create should succeed, as no such address in kernel.
+ req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_CREATE | NLM_F_ACK;
+ EXPECT_NO_ERRNO(
+ NetlinkRequestAckOrError(fd, req.hdr.nlmsg_seq, &req, req.hdr.nlmsg_len));
+
+ // Replace an existing address should succeed.
+ req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_REPLACE | NLM_F_ACK;
+ req.hdr.nlmsg_seq++;
+ EXPECT_NO_ERRNO(
+ NetlinkRequestAckOrError(fd, req.hdr.nlmsg_seq, &req, req.hdr.nlmsg_len));
+
+ // Create exclusive should fail, as we created the address above.
+ req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_CREATE | NLM_F_EXCL | NLM_F_ACK;
+ req.hdr.nlmsg_seq++;
+ EXPECT_THAT(
+ NetlinkRequestAckOrError(fd, req.hdr.nlmsg_seq, &req, req.hdr.nlmsg_len),
+ PosixErrorIs(EEXIST, ::testing::_));
+}
+
// GetRouteDump tests a RTM_GETROUTE + NLM_F_DUMP request.
TEST(NetlinkRouteTest, GetRouteDump) {
FileDescriptor fd =
@@ -378,8 +582,6 @@ TEST(NetlinkRouteTest, GetRouteDump) {
struct rtmsg rtm;
};
- constexpr uint32_t kSeq = 12345;
-
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETROUTE;
@@ -538,8 +740,6 @@ TEST(NetlinkRouteTest, RecvmsgTrunc) {
struct rtgenmsg rgm;
};
- constexpr uint32_t kSeq = 12345;
-
struct request req;
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETADDR;
@@ -615,8 +815,6 @@ TEST(NetlinkRouteTest, RecvmsgTruncPeek) {
struct rtgenmsg rgm;
};
- constexpr uint32_t kSeq = 12345;
-
struct request req;
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETADDR;
@@ -695,8 +893,6 @@ TEST(NetlinkRouteTest, NoPasscredNoCreds) {
struct rtgenmsg rgm;
};
- constexpr uint32_t kSeq = 12345;
-
struct request req;
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETADDR;
@@ -743,8 +939,6 @@ TEST(NetlinkRouteTest, PasscredCreds) {
struct rtgenmsg rgm;
};
- constexpr uint32_t kSeq = 12345;
-
struct request req;
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETADDR;
diff --git a/test/syscalls/linux/socket_netlink_util.cc b/test/syscalls/linux/socket_netlink_util.cc
index cd2212a1a..952eecfe8 100644
--- a/test/syscalls/linux/socket_netlink_util.cc
+++ b/test/syscalls/linux/socket_netlink_util.cc
@@ -16,6 +16,7 @@
#include <linux/if_arp.h>
#include <linux/netlink.h>
+#include <linux/rtnetlink.h>
#include <sys/socket.h>
#include <vector>
@@ -71,9 +72,10 @@ PosixError NetlinkRequestResponse(
iov.iov_base = buf.data();
iov.iov_len = buf.size();
- // Response is a series of NLM_F_MULTI messages, ending with a NLMSG_DONE
- // message.
+ // If NLM_F_MULTI is set, response is a series of messages that ends with a
+ // NLMSG_DONE message.
int type = -1;
+ int flags = 0;
do {
int len;
RETURN_ERROR_IF_SYSCALL_FAIL(len = RetryEINTR(recvmsg)(fd.get(), &msg, 0));
@@ -89,6 +91,7 @@ PosixError NetlinkRequestResponse(
for (struct nlmsghdr* hdr = reinterpret_cast<struct nlmsghdr*>(buf.data());
NLMSG_OK(hdr, len); hdr = NLMSG_NEXT(hdr, len)) {
fn(hdr);
+ flags = hdr->nlmsg_flags;
type = hdr->nlmsg_type;
// Done should include an integer payload for dump_done_errno.
// See net/netlink/af_netlink.c:netlink_dump
@@ -98,11 +101,11 @@ PosixError NetlinkRequestResponse(
EXPECT_GE(hdr->nlmsg_len, NLMSG_LENGTH(sizeof(int)));
}
}
- } while (type != NLMSG_DONE && type != NLMSG_ERROR);
+ } while ((flags & NLM_F_MULTI) && type != NLMSG_DONE && type != NLMSG_ERROR);
if (expect_nlmsgerr) {
EXPECT_EQ(type, NLMSG_ERROR);
- } else {
+ } else if (flags & NLM_F_MULTI) {
EXPECT_EQ(type, NLMSG_DONE);
}
return NoError();
@@ -146,5 +149,39 @@ PosixError NetlinkRequestResponseSingle(
return NoError();
}
+PosixError NetlinkRequestAckOrError(const FileDescriptor& fd, uint32_t seq,
+ void* request, size_t len) {
+ // Dummy negative number for no error message received.
+ // We won't get a negative error number so there will be no confusion.
+ int err = -42;
+ RETURN_IF_ERRNO(NetlinkRequestResponse(
+ fd, request, len,
+ [&](const struct nlmsghdr* hdr) {
+ EXPECT_EQ(NLMSG_ERROR, hdr->nlmsg_type);
+ EXPECT_EQ(hdr->nlmsg_seq, seq);
+ EXPECT_GE(hdr->nlmsg_len, sizeof(*hdr) + sizeof(struct nlmsgerr));
+
+ const struct nlmsgerr* msg =
+ reinterpret_cast<const struct nlmsgerr*>(NLMSG_DATA(hdr));
+ err = -msg->error;
+ },
+ true));
+ return PosixError(err);
+}
+
+const struct rtattr* FindRtAttr(const struct nlmsghdr* hdr,
+ const struct ifinfomsg* msg, int16_t attr) {
+ const int ifi_space = NLMSG_SPACE(sizeof(*msg));
+ int attrlen = hdr->nlmsg_len - ifi_space;
+ const struct rtattr* rta = reinterpret_cast<const struct rtattr*>(
+ reinterpret_cast<const uint8_t*>(hdr) + NLMSG_ALIGN(ifi_space));
+ for (; RTA_OK(rta, attrlen); rta = RTA_NEXT(rta, attrlen)) {
+ if (rta->rta_type == attr) {
+ return rta;
+ }
+ }
+ return nullptr;
+}
+
} // namespace testing
} // namespace gvisor
diff --git a/test/syscalls/linux/socket_netlink_util.h b/test/syscalls/linux/socket_netlink_util.h
index 3678c0599..e13ead406 100644
--- a/test/syscalls/linux/socket_netlink_util.h
+++ b/test/syscalls/linux/socket_netlink_util.h
@@ -19,6 +19,7 @@
// socket.h has to be included before if_arp.h.
#include <linux/if_arp.h>
#include <linux/netlink.h>
+#include <linux/rtnetlink.h>
#include "test/util/file_descriptor.h"
#include "test/util/posix_error.h"
@@ -47,6 +48,14 @@ PosixError NetlinkRequestResponseSingle(
const FileDescriptor& fd, void* request, size_t len,
const std::function<void(const struct nlmsghdr* hdr)>& fn);
+// Send the passed request then expect and return an ack or error.
+PosixError NetlinkRequestAckOrError(const FileDescriptor& fd, uint32_t seq,
+ void* request, size_t len);
+
+// Find rtnetlink attribute in message.
+const struct rtattr* FindRtAttr(const struct nlmsghdr* hdr,
+ const struct ifinfomsg* msg, int16_t attr);
+
} // namespace testing
} // namespace gvisor