From 51b783505b1ec164b02b48a0fd234509fba01a73 Mon Sep 17 00:00:00 2001 From: Bhasker Hariharan Date: Wed, 29 Jan 2020 15:41:51 -0800 Subject: Add support for TCP_DEFER_ACCEPT. PiperOrigin-RevId: 292233574 --- pkg/sentry/socket/netstack/netstack.go | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+) (limited to 'pkg/sentry/socket/netstack/netstack.go') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 8619cc506..049d04bf2 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -1260,6 +1260,18 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa return int32(time.Duration(v) / time.Second), nil + case linux.TCP_DEFER_ACCEPT: + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + + var v tcpip.TCPDeferAcceptOption + if err := ep.GetSockOpt(&v); err != nil { + return nil, syserr.TranslateNetstackError(err) + } + + return int32(time.Duration(v) / time.Second), nil + default: emitUnimplementedEventTCP(t, name) } @@ -1713,6 +1725,16 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) * v := usermem.ByteOrder.Uint32(optVal) return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.TCPLingerTimeoutOption(time.Second * time.Duration(v)))) + case linux.TCP_DEFER_ACCEPT: + if len(optVal) < sizeOfInt32 { + return syserr.ErrInvalidArgument + } + v := int32(usermem.ByteOrder.Uint32(optVal)) + if v < 0 { + v = 0 + } + return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.TCPDeferAcceptOption(time.Second * time.Duration(v)))) + case linux.TCP_REPAIR_OPTIONS: t.Kernel().EmitUnimplementedEvent(t) -- cgit v1.2.3 From f3d95607036b8a502c65aa7b3e8145227274dbbc Mon Sep 17 00:00:00 2001 From: Eyal Soha Date: Wed, 5 Feb 2020 17:56:00 -0800 Subject: recv() on a closed TCP socket returns ENOTCONN From RFC 793 s3.9 p58 Event Processing: If RECEIVE Call arrives in CLOSED state and the user has access to such a connection, the return should be "error: connection does not exist" Fixes #1598 PiperOrigin-RevId: 293494287 --- pkg/sentry/socket/netstack/netstack.go | 7 ++++++- pkg/tcpip/tcpip.go | 4 ++++ pkg/tcpip/transport/tcp/endpoint.go | 4 ++-- pkg/tcpip/transport/tcp/tcp_test.go | 9 ++++----- test/syscalls/linux/tcp_socket.cc | 9 +++++++++ 5 files changed, 25 insertions(+), 8 deletions(-) (limited to 'pkg/sentry/socket/netstack/netstack.go') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 049d04bf2..ed2fbcceb 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -2229,11 +2229,16 @@ func (s *SocketOperations) coalescingRead(ctx context.Context, dst usermem.IOSeq var copied int // Copy as many views as possible into the user-provided buffer. - for dst.NumBytes() != 0 { + for { + // Always do at least one fetchReadView, even if the number of bytes to + // read is 0. err = s.fetchReadView() if err != nil { break } + if dst.NumBytes() == 0 { + break + } var n int var e error diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 0fa141d58..d29d9a704 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -1124,6 +1124,10 @@ type ReadErrors struct { // InvalidEndpointState is the number of times we found the endpoint state // to be unexpected. InvalidEndpointState StatCounter + + // NotConnected is the number of times we tried to read but found that the + // endpoint was not connected. + NotConnected StatCounter } // WriteErrors collects packet write errors from an endpoint write call. diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index b5a8e15ee..e4a6b1b8b 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -1003,8 +1003,8 @@ func (e *endpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, if s == StateError { return buffer.View{}, tcpip.ControlMessages{}, he } - e.stats.ReadErrors.InvalidEndpointState.Increment() - return buffer.View{}, tcpip.ControlMessages{}, tcpip.ErrInvalidEndpointState + e.stats.ReadErrors.NotConnected.Increment() + return buffer.View{}, tcpip.ControlMessages{}, tcpip.ErrNotConnected } v, err := e.readLocked() diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 2c1505067..cc118c993 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -5405,12 +5405,11 @@ func TestEndpointBindListenAcceptState(t *testing.T) { t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) } - // Expect InvalidEndpointState errors on a read at this point. - if _, _, err := ep.Read(nil); err != tcpip.ErrInvalidEndpointState { - t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrInvalidEndpointState) + if _, _, err := ep.Read(nil); err != tcpip.ErrNotConnected { + t.Errorf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrNotConnected) } - if got := ep.Stats().(*tcp.Stats).ReadErrors.InvalidEndpointState.Value(); got != 1 { - t.Fatalf("got EP stats Stats.ReadErrors.InvalidEndpointState got %v want %v", got, 1) + if got := ep.Stats().(*tcp.Stats).ReadErrors.NotConnected.Value(); got != 1 { + t.Errorf("got EP stats Stats.ReadErrors.NotConnected got %v want %v", got, 1) } if err := ep.Listen(10); err != nil { diff --git a/test/syscalls/linux/tcp_socket.cc b/test/syscalls/linux/tcp_socket.cc index 525ccbd88..8a8b68e75 100644 --- a/test/syscalls/linux/tcp_socket.cc +++ b/test/syscalls/linux/tcp_socket.cc @@ -1339,6 +1339,15 @@ TEST_P(SimpleTcpSocketTest, SetTCPDeferAcceptGreaterThanZero) { EXPECT_EQ(get, kTCPDeferAccept); } +TEST_P(SimpleTcpSocketTest, RecvOnClosedSocket) { + auto s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + char buf[1]; + EXPECT_THAT(recv(s.get(), buf, 0, 0), SyscallFailsWithErrno(ENOTCONN)); + EXPECT_THAT(recv(s.get(), buf, sizeof(buf), 0), + SyscallFailsWithErrno(ENOTCONN)); +} + INSTANTIATE_TEST_SUITE_P(AllInetTests, SimpleTcpSocketTest, ::testing::Values(AF_INET, AF_INET6)); -- cgit v1.2.3 From 69bf39e8a47d3b4dcbbd04d2e8df476cdfab5e74 Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Thu, 13 Feb 2020 10:58:47 -0800 Subject: Internal change. PiperOrigin-RevId: 294952610 --- pkg/abi/linux/socket.go | 13 ++++ pkg/sentry/socket/control/BUILD | 1 + pkg/sentry/socket/control/control.go | 43 +++++++++++++ pkg/sentry/socket/hostinet/socket.go | 11 +++- pkg/sentry/socket/netstack/netstack.go | 37 ++++++++++-- pkg/tcpip/tcpip.go | 25 ++++++++ pkg/tcpip/transport/udp/endpoint.go | 26 ++++++++ test/syscalls/linux/socket_ip_udp_generic.cc | 44 ++++++++++++++ test/syscalls/linux/socket_ipv4_udp_unbound.cc | 84 ++++++++++++++++++++++++++ test/syscalls/linux/udp_socket_test_cases.cc | 1 - 10 files changed, 278 insertions(+), 7 deletions(-) (limited to 'pkg/sentry/socket/netstack/netstack.go') diff --git a/pkg/abi/linux/socket.go b/pkg/abi/linux/socket.go index 766ee4014..4a14ef691 100644 --- a/pkg/abi/linux/socket.go +++ b/pkg/abi/linux/socket.go @@ -411,6 +411,15 @@ type ControlMessageCredentials struct { GID uint32 } +// A ControlMessageIPPacketInfo is IP_PKTINFO socket control message. +// +// ControlMessageIPPacketInfo represents struct in_pktinfo from linux/in.h. +type ControlMessageIPPacketInfo struct { + NIC int32 + LocalAddr InetAddr + DestinationAddr InetAddr +} + // SizeOfControlMessageCredentials is the binary size of a // ControlMessageCredentials struct. var SizeOfControlMessageCredentials = int(binary.Size(ControlMessageCredentials{})) @@ -431,6 +440,10 @@ const SizeOfControlMessageTOS = 1 // SizeOfControlMessageTClass is the size of an IPV6_TCLASS control message. const SizeOfControlMessageTClass = 4 +// SizeOfControlMessageIPPacketInfo is the size of an IP_PKTINFO +// control message. +const SizeOfControlMessageIPPacketInfo = 12 + // SCM_MAX_FD is the maximum number of FDs accepted in a single sendmsg call. // From net/scm.h. const SCM_MAX_FD = 253 diff --git a/pkg/sentry/socket/control/BUILD b/pkg/sentry/socket/control/BUILD index 79e16d6e8..4d42d29cb 100644 --- a/pkg/sentry/socket/control/BUILD +++ b/pkg/sentry/socket/control/BUILD @@ -19,6 +19,7 @@ go_library( "//pkg/sentry/socket", "//pkg/sentry/socket/unix/transport", "//pkg/syserror", + "//pkg/tcpip", "//pkg/usermem", ], ) diff --git a/pkg/sentry/socket/control/control.go b/pkg/sentry/socket/control/control.go index 6145a7fc3..4667373d2 100644 --- a/pkg/sentry/socket/control/control.go +++ b/pkg/sentry/socket/control/control.go @@ -26,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/usermem" ) @@ -338,6 +339,22 @@ func PackTClass(t *kernel.Task, tClass int32, buf []byte) []byte { ) } +// PackIPPacketInfo packs an IP_PKTINFO socket control message. +func PackIPPacketInfo(t *kernel.Task, packetInfo tcpip.IPPacketInfo, buf []byte) []byte { + var p linux.ControlMessageIPPacketInfo + p.NIC = int32(packetInfo.NIC) + copy(p.LocalAddr[:], []byte(packetInfo.LocalAddr)) + copy(p.DestinationAddr[:], []byte(packetInfo.DestinationAddr)) + + return putCmsgStruct( + buf, + linux.SOL_IP, + linux.IP_PKTINFO, + t.Arch().Width(), + p, + ) +} + // PackControlMessages packs control messages into the given buffer. // // We skip control messages specific to Unix domain sockets. @@ -362,6 +379,10 @@ func PackControlMessages(t *kernel.Task, cmsgs socket.ControlMessages, buf []byt buf = PackTClass(t, cmsgs.IP.TClass, buf) } + if cmsgs.IP.HasIPPacketInfo { + buf = PackIPPacketInfo(t, cmsgs.IP.PacketInfo, buf) + } + return buf } @@ -394,6 +415,16 @@ func CmsgsSpace(t *kernel.Task, cmsgs socket.ControlMessages) int { return space } +// NewIPPacketInfo returns the IPPacketInfo struct. +func NewIPPacketInfo(packetInfo linux.ControlMessageIPPacketInfo) tcpip.IPPacketInfo { + var p tcpip.IPPacketInfo + p.NIC = tcpip.NICID(packetInfo.NIC) + copy([]byte(p.LocalAddr), packetInfo.LocalAddr[:]) + copy([]byte(p.DestinationAddr), packetInfo.DestinationAddr[:]) + + return p +} + // Parse parses a raw socket control message into portable objects. func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (socket.ControlMessages, error) { var ( @@ -468,6 +499,18 @@ func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (socket.Con binary.Unmarshal(buf[i:i+linux.SizeOfControlMessageTOS], usermem.ByteOrder, &cmsgs.IP.TOS) i += binary.AlignUp(length, width) + case linux.IP_PKTINFO: + if length < linux.SizeOfControlMessageIPPacketInfo { + return socket.ControlMessages{}, syserror.EINVAL + } + + cmsgs.IP.HasIPPacketInfo = true + var packetInfo linux.ControlMessageIPPacketInfo + binary.Unmarshal(buf[i:i+linux.SizeOfControlMessageIPPacketInfo], usermem.ByteOrder, &packetInfo) + + cmsgs.IP.PacketInfo = NewIPPacketInfo(packetInfo) + i += binary.AlignUp(length, width) + default: return socket.ControlMessages{}, syserror.EINVAL } diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go index de76388ac..22f78d2e2 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -289,7 +289,7 @@ func (s *socketOperations) GetSockOpt(t *kernel.Task, level int, name int, outPt switch level { case linux.SOL_IP: switch name { - case linux.IP_TOS, linux.IP_RECVTOS: + case linux.IP_TOS, linux.IP_RECVTOS, linux.IP_PKTINFO: optlen = sizeofInt32 } case linux.SOL_IPV6: @@ -336,6 +336,8 @@ func (s *socketOperations) SetSockOpt(t *kernel.Task, level int, name int, opt [ switch name { case linux.IP_TOS, linux.IP_RECVTOS: optlen = sizeofInt32 + case linux.IP_PKTINFO: + optlen = linux.SizeOfControlMessageIPPacketInfo } case linux.SOL_IPV6: switch name { @@ -473,7 +475,14 @@ func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags case syscall.IP_TOS: controlMessages.IP.HasTOS = true binary.Unmarshal(unixCmsg.Data[:linux.SizeOfControlMessageTOS], usermem.ByteOrder, &controlMessages.IP.TOS) + + case syscall.IP_PKTINFO: + controlMessages.IP.HasIPPacketInfo = true + var packetInfo linux.ControlMessageIPPacketInfo + binary.Unmarshal(unixCmsg.Data[:linux.SizeOfControlMessageIPPacketInfo], usermem.ByteOrder, &packetInfo) + controlMessages.IP.PacketInfo = control.NewIPPacketInfo(packetInfo) } + case syscall.SOL_IPV6: switch unixCmsg.Header.Type { case syscall.IPV6_TCLASS: diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index ed2fbcceb..9757fbfba 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -1414,6 +1414,21 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in } return o, nil + case linux.IP_PKTINFO: + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + + v, err := ep.GetSockOptBool(tcpip.ReceiveIPPacketInfoOption) + if err != nil { + return nil, syserr.TranslateNetstackError(err) + } + var o int32 + if v { + o = 1 + } + return o, nil + default: emitUnimplementedEventIP(t, name) } @@ -1762,6 +1777,7 @@ func setSockOptIPv6(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) linux.IPV6_IPSEC_POLICY, linux.IPV6_JOIN_ANYCAST, linux.IPV6_LEAVE_ANYCAST, + // TODO(b/148887420): Add support for IPV6_PKTINFO. linux.IPV6_PKTINFO, linux.IPV6_ROUTER_ALERT, linux.IPV6_XFRM_POLICY, @@ -1949,6 +1965,16 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s } return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReceiveTOSOption, v != 0)) + case linux.IP_PKTINFO: + if len(optVal) == 0 { + return nil + } + v, err := parseIntOrChar(optVal) + if err != nil { + return err + } + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReceiveIPPacketInfoOption, v != 0)) + case linux.IP_ADD_SOURCE_MEMBERSHIP, linux.IP_BIND_ADDRESS_NO_PORT, linux.IP_BLOCK_SOURCE, @@ -1964,7 +1990,6 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s linux.IP_NODEFRAG, linux.IP_OPTIONS, linux.IP_PASSSEC, - linux.IP_PKTINFO, linux.IP_RECVERR, linux.IP_RECVFRAGSIZE, linux.IP_RECVOPTS, @@ -2395,10 +2420,12 @@ func (s *SocketOperations) nonBlockingRead(ctx context.Context, dst usermem.IOSe func (s *SocketOperations) controlMessages() socket.ControlMessages { return socket.ControlMessages{ IP: tcpip.ControlMessages{ - HasTimestamp: s.readCM.HasTimestamp && s.sockOptTimestamp, - Timestamp: s.readCM.Timestamp, - HasTOS: s.readCM.HasTOS, - TOS: s.readCM.TOS, + HasTimestamp: s.readCM.HasTimestamp && s.sockOptTimestamp, + Timestamp: s.readCM.Timestamp, + HasTOS: s.readCM.HasTOS, + TOS: s.readCM.TOS, + HasIPPacketInfo: s.readCM.HasIPPacketInfo, + PacketInfo: s.readCM.PacketInfo, }, } } diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 0e944712f..9ca39ce40 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -328,6 +328,12 @@ type ControlMessages struct { // Tclass is the IPv6 traffic class of the associated packet. TClass int32 + + // HasIPPacketInfo indicates whether PacketInfo is set. + HasIPPacketInfo bool + + // PacketInfo holds interface and address data on an incoming packet. + PacketInfo IPPacketInfo } // Endpoint is the interface implemented by transport protocols (e.g., tcp, udp) @@ -503,6 +509,11 @@ const ( // V6OnlyOption is used by {G,S}etSockOptBool to specify whether an IPv6 // socket is to be restricted to sending and receiving IPv6 packets only. V6OnlyOption + + // ReceiveIPPacketInfoOption is used by {G,S}etSockOptBool to specify + // if more inforamtion is provided with incoming packets such + // as interface index and address. + ReceiveIPPacketInfoOption ) // SockOptInt represents socket options which values have the int type. @@ -685,6 +696,20 @@ type IPv4TOSOption uint8 // for all subsequent outgoing IPv6 packets from the endpoint. type IPv6TrafficClassOption uint8 +// IPPacketInfo is the message struture for IP_PKTINFO. +// +// +stateify savable +type IPPacketInfo struct { + // NIC is the ID of the NIC to be used. + NIC NICID + + // LocalAddr is the local address. + LocalAddr Address + + // DestinationAddr is the destination address. + DestinationAddr Address +} + // Route is a row in the routing table. It specifies through which NIC (and // gateway) sets of packets should be routed. A row is considered viable if the // masked target address matches the destination address in the row. diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index c9cbed8f4..3fe91cac2 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -29,6 +29,7 @@ import ( type udpPacket struct { udpPacketEntry senderAddress tcpip.FullAddress + packetInfo tcpip.IPPacketInfo data buffer.VectorisedView `state:".(buffer.VectorisedView)"` timestamp int64 tos uint8 @@ -118,6 +119,9 @@ type endpoint struct { // as ancillary data to ControlMessages on Read. receiveTOS bool + // receiveIPPacketInfo determines if the packet info is returned by Read. + receiveIPPacketInfo bool + // shutdownFlags represent the current shutdown state of the endpoint. shutdownFlags tcpip.ShutdownFlags @@ -254,11 +258,17 @@ func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMess } e.mu.RLock() receiveTOS := e.receiveTOS + receiveIPPacketInfo := e.receiveIPPacketInfo e.mu.RUnlock() if receiveTOS { cm.HasTOS = true cm.TOS = p.tos } + + if receiveIPPacketInfo { + cm.HasIPPacketInfo = true + cm.PacketInfo = p.packetInfo + } return p.data.ToView(), cm, nil } @@ -495,6 +505,13 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { } e.v6only = v + return nil + + case tcpip.ReceiveIPPacketInfoOption: + e.mu.Lock() + e.receiveIPPacketInfo = v + e.mu.Unlock() + return nil } return nil @@ -703,6 +720,12 @@ func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { e.mu.RUnlock() return v, nil + + case tcpip.ReceiveIPPacketInfoOption: + e.mu.RLock() + v := e.receiveIPPacketInfo + e.mu.RUnlock() + return v, nil } return false, tcpip.ErrUnknownProtocolOption @@ -1247,6 +1270,9 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk switch r.NetProto { case header.IPv4ProtocolNumber: packet.tos, _ = header.IPv4(pkt.NetworkHeader).TOS() + packet.packetInfo.LocalAddr = r.LocalAddress + packet.packetInfo.DestinationAddr = r.RemoteAddress + packet.packetInfo.NIC = r.NICID() } packet.timestamp = e.stack.NowNanoseconds() diff --git a/test/syscalls/linux/socket_ip_udp_generic.cc b/test/syscalls/linux/socket_ip_udp_generic.cc index 53290bed7..db5663ecd 100644 --- a/test/syscalls/linux/socket_ip_udp_generic.cc +++ b/test/syscalls/linux/socket_ip_udp_generic.cc @@ -357,5 +357,49 @@ TEST_P(UDPSocketPairTest, SetReuseAddrReusePort) { EXPECT_EQ(get, kSockOptOn); } +// Test getsockopt for a socket which is not set with IP_PKTINFO option. +TEST_P(UDPSocketPairTest, IPPKTINFODefault) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + int get = -1; + socklen_t get_len = sizeof(get); + + ASSERT_THAT( + getsockopt(sockets->first_fd(), SOL_IP, IP_PKTINFO, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kSockOptOff); +} + +// Test setsockopt and getsockopt for a socket with IP_PKTINFO option. +TEST_P(UDPSocketPairTest, SetAndGetIPPKTINFO) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + int level = SOL_IP; + int type = IP_PKTINFO; + + // Check getsockopt before IP_PKTINFO is set. + int get = -1; + socklen_t get_len = sizeof(get); + + ASSERT_THAT(setsockopt(sockets->first_fd(), level, type, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceedsWithValue(0)); + + ASSERT_THAT(getsockopt(sockets->first_fd(), level, type, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get, kSockOptOn); + EXPECT_EQ(get_len, sizeof(get)); + + ASSERT_THAT(setsockopt(sockets->first_fd(), level, type, &kSockOptOff, + sizeof(kSockOptOff)), + SyscallSucceedsWithValue(0)); + + ASSERT_THAT(getsockopt(sockets->first_fd(), level, type, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get, kSockOptOff); + EXPECT_EQ(get_len, sizeof(get)); +} + } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_ipv4_udp_unbound.cc b/test/syscalls/linux/socket_ipv4_udp_unbound.cc index 990ccf23c..bc4b07a62 100644 --- a/test/syscalls/linux/socket_ipv4_udp_unbound.cc +++ b/test/syscalls/linux/socket_ipv4_udp_unbound.cc @@ -15,6 +15,7 @@ #include "test/syscalls/linux/socket_ipv4_udp_unbound.h" #include +#include #include #include #include @@ -2128,5 +2129,88 @@ TEST_P(IPv4UDPUnboundSocketTest, ReuseAddrReusePortDistribution) { SyscallSucceedsWithValue(kMessageSize)); } +// Test that socket will receive packet info control message. +TEST_P(IPv4UDPUnboundSocketTest, SetAndReceiveIPPKTINFO) { + // TODO(gvisor.dev/issue/1202): ioctl() is not supported by hostinet. + SKIP_IF((IsRunningWithHostinet())); + + auto sender = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + auto receiver = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + auto sender_addr = V4Loopback(); + int level = SOL_IP; + int type = IP_PKTINFO; + + ASSERT_THAT( + bind(receiver->get(), reinterpret_cast(&sender_addr.addr), + sender_addr.addr_len), + SyscallSucceeds()); + socklen_t sender_addr_len = sender_addr.addr_len; + ASSERT_THAT(getsockname(receiver->get(), + reinterpret_cast(&sender_addr.addr), + &sender_addr_len), + SyscallSucceeds()); + EXPECT_EQ(sender_addr_len, sender_addr.addr_len); + + auto receiver_addr = V4Loopback(); + reinterpret_cast(&receiver_addr.addr)->sin_port = + reinterpret_cast(&sender_addr.addr)->sin_port; + ASSERT_THAT( + connect(sender->get(), reinterpret_cast(&receiver_addr.addr), + receiver_addr.addr_len), + SyscallSucceeds()); + + // Allow socket to receive control message. + ASSERT_THAT( + setsockopt(receiver->get(), level, type, &kSockOptOn, sizeof(kSockOptOn)), + SyscallSucceeds()); + + // Prepare message to send. + constexpr size_t kDataLength = 1024; + msghdr sent_msg = {}; + iovec sent_iov = {}; + char sent_data[kDataLength]; + sent_iov.iov_base = sent_data; + sent_iov.iov_len = kDataLength; + sent_msg.msg_iov = &sent_iov; + sent_msg.msg_iovlen = 1; + sent_msg.msg_flags = 0; + + ASSERT_THAT(RetryEINTR(sendmsg)(sender->get(), &sent_msg, 0), + SyscallSucceedsWithValue(kDataLength)); + + msghdr received_msg = {}; + iovec received_iov = {}; + char received_data[kDataLength]; + char received_cmsg_buf[CMSG_SPACE(sizeof(in_pktinfo))] = {}; + size_t cmsg_data_len = sizeof(in_pktinfo); + received_iov.iov_base = received_data; + received_iov.iov_len = kDataLength; + received_msg.msg_iov = &received_iov; + received_msg.msg_iovlen = 1; + received_msg.msg_controllen = CMSG_LEN(cmsg_data_len); + received_msg.msg_control = received_cmsg_buf; + + ASSERT_THAT(RetryEINTR(recvmsg)(receiver->get(), &received_msg, 0), + SyscallSucceedsWithValue(kDataLength)); + + cmsghdr* cmsg = CMSG_FIRSTHDR(&received_msg); + ASSERT_NE(cmsg, nullptr); + EXPECT_EQ(cmsg->cmsg_len, CMSG_LEN(cmsg_data_len)); + EXPECT_EQ(cmsg->cmsg_level, level); + EXPECT_EQ(cmsg->cmsg_type, type); + + // Get loopback index. + ifreq ifr = {}; + absl::SNPrintF(ifr.ifr_name, IFNAMSIZ, "lo"); + ASSERT_THAT(ioctl(sender->get(), SIOCGIFINDEX, &ifr), SyscallSucceeds()); + ASSERT_NE(ifr.ifr_ifindex, 0); + + // Check the data + in_pktinfo received_pktinfo = {}; + memcpy(&received_pktinfo, CMSG_DATA(cmsg), sizeof(in_pktinfo)); + EXPECT_EQ(received_pktinfo.ipi_ifindex, ifr.ifr_ifindex); + EXPECT_EQ(received_pktinfo.ipi_spec_dst.s_addr, htonl(INADDR_LOOPBACK)); + EXPECT_EQ(received_pktinfo.ipi_addr.s_addr, htonl(INADDR_LOOPBACK)); +} } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/udp_socket_test_cases.cc b/test/syscalls/linux/udp_socket_test_cases.cc index a2f6ef8cc..9f8de6b48 100644 --- a/test/syscalls/linux/udp_socket_test_cases.cc +++ b/test/syscalls/linux/udp_socket_test_cases.cc @@ -1495,6 +1495,5 @@ TEST_P(UdpSocketTest, SendAndReceiveTOS) { memcpy(&received_tos, CMSG_DATA(cmsg), sizeof(received_tos)); EXPECT_EQ(received_tos, sent_tos); } - } // namespace testing } // namespace gvisor -- cgit v1.2.3 From 56fd9504aab44a738d3df164cbee8e572b309f28 Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Tue, 18 Feb 2020 15:44:22 -0800 Subject: Enable IPV6_RECVTCLASS socket option for datagram sockets Added the ability to get/set the IP_RECVTCLASS socket option on UDP endpoints. If enabled, traffic class from the incoming Network Header passed as ancillary data in the ControlMessages. Adding Get/SetSockOptBool to decrease the overhead of getting/setting simple options. (This was absorbed in a CL that will be landing before this one). Test: * Added unit test to udp_test.go that tests getting/setting as well as verifying that we receive expected TOS from incoming packet. * Added a syscall test for verifying getting/setting * Removed test skip for existing syscall test to enable end to end test. PiperOrigin-RevId: 295840218 --- pkg/sentry/socket/control/control.go | 2 +- pkg/sentry/socket/netstack/netstack.go | 27 +++++- pkg/tcpip/checker/checker.go | 14 +++ pkg/tcpip/tcpip.go | 15 ++- pkg/tcpip/transport/udp/endpoint.go | 38 +++++++- pkg/tcpip/transport/udp/udp_test.go | 120 ++++++++++++++---------- test/syscalls/linux/ip_socket_test_util.h | 16 ++-- test/syscalls/linux/socket_ip_udp_generic.cc | 133 +++++++++++++++++++-------- test/syscalls/linux/udp_socket_test_cases.cc | 4 - 9 files changed, 260 insertions(+), 109 deletions(-) (limited to 'pkg/sentry/socket/netstack/netstack.go') diff --git a/pkg/sentry/socket/control/control.go b/pkg/sentry/socket/control/control.go index 4667373d2..8834a1e1a 100644 --- a/pkg/sentry/socket/control/control.go +++ b/pkg/sentry/socket/control/control.go @@ -329,7 +329,7 @@ func PackTOS(t *kernel.Task, tos uint8, buf []byte) []byte { } // PackTClass packs an IPV6_TCLASS socket control message. -func PackTClass(t *kernel.Task, tClass int32, buf []byte) []byte { +func PackTClass(t *kernel.Task, tClass uint32, buf []byte) []byte { return putCmsgStruct( buf, linux.SOL_IPV6, diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 9757fbfba..e187276c5 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -1318,6 +1318,22 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf } return ib, nil + case linux.IPV6_RECVTCLASS: + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + + v, err := ep.GetSockOptBool(tcpip.ReceiveTClassOption) + if err != nil { + return nil, syserr.TranslateNetstackError(err) + } + + var o int32 + if v { + o = 1 + } + return o, nil + default: emitUnimplementedEventIPv6(t, name) } @@ -1803,6 +1819,14 @@ func setSockOptIPv6(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) } return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.IPv6TrafficClassOption(v))) + case linux.IPV6_RECVTCLASS: + v, err := parseIntOrChar(optVal) + if err != nil { + return err + } + + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReceiveTClassOption, v != 0)) + default: emitUnimplementedEventIPv6(t, name) } @@ -2086,7 +2110,6 @@ func emitUnimplementedEventIPv6(t *kernel.Task, name int) { linux.IPV6_RECVPATHMTU, linux.IPV6_RECVPKTINFO, linux.IPV6_RECVRTHDR, - linux.IPV6_RECVTCLASS, linux.IPV6_RTHDR, linux.IPV6_RTHDRDSTOPTS, linux.IPV6_TCLASS, @@ -2424,6 +2447,8 @@ func (s *SocketOperations) controlMessages() socket.ControlMessages { Timestamp: s.readCM.Timestamp, HasTOS: s.readCM.HasTOS, TOS: s.readCM.TOS, + HasTClass: s.readCM.HasTClass, + TClass: s.readCM.TClass, HasIPPacketInfo: s.readCM.HasIPPacketInfo, PacketInfo: s.readCM.PacketInfo, }, diff --git a/pkg/tcpip/checker/checker.go b/pkg/tcpip/checker/checker.go index 4d6ae0871..c6c160dfc 100644 --- a/pkg/tcpip/checker/checker.go +++ b/pkg/tcpip/checker/checker.go @@ -161,6 +161,20 @@ func FragmentFlags(flags uint8) NetworkChecker { } } +// ReceiveTClass creates a checker that checks the TCLASS field in +// ControlMessages. +func ReceiveTClass(want uint32) ControlMessagesChecker { + return func(t *testing.T, cm tcpip.ControlMessages) { + t.Helper() + if !cm.HasTClass { + t.Fatalf("got cm.HasTClass = %t, want cm.TClass = %d", cm.HasTClass, want) + } + if got := cm.TClass; got != want { + t.Fatalf("got cm.TClass = %d, want %d", got, want) + } + } +} + // ReceiveTOS creates a checker that checks the TOS field in ControlMessages. func ReceiveTOS(want uint8) ControlMessagesChecker { return func(t *testing.T, cm tcpip.ControlMessages) { diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 9ca39ce40..ce5527391 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -323,11 +323,11 @@ type ControlMessages struct { // TOS is the IPv4 type of service of the associated packet. TOS uint8 - // HasTClass indicates whether Tclass is valid/set. + // HasTClass indicates whether TClass is valid/set. HasTClass bool - // Tclass is the IPv6 traffic class of the associated packet. - TClass int32 + // TClass is the IPv6 traffic class of the associated packet. + TClass uint32 // HasIPPacketInfo indicates whether PacketInfo is set. HasIPPacketInfo bool @@ -502,9 +502,13 @@ type WriteOptions struct { type SockOptBool int const ( + // ReceiveTClassOption is used by SetSockOpt/GetSockOpt to specify if the + // IPV6_TCLASS ancillary message is passed with incoming packets. + ReceiveTClassOption SockOptBool = iota + // ReceiveTOSOption is used by SetSockOpt/GetSockOpt to specify if the TOS // ancillary message is passed with incoming packets. - ReceiveTOSOption SockOptBool = iota + ReceiveTOSOption // V6OnlyOption is used by {G,S}etSockOptBool to specify whether an IPv6 // socket is to be restricted to sending and receiving IPv6 packets only. @@ -514,6 +518,9 @@ const ( // if more inforamtion is provided with incoming packets such // as interface index and address. ReceiveIPPacketInfoOption + + // TODO(b/146901447): convert existing bool socket options to be handled via + // Get/SetSockOptBool ) // SockOptInt represents socket options which values have the int type. diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 3fe91cac2..eff7f3600 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -32,7 +32,8 @@ type udpPacket struct { packetInfo tcpip.IPPacketInfo data buffer.VectorisedView `state:".(buffer.VectorisedView)"` timestamp int64 - tos uint8 + // tos stores either the receiveTOS or receiveTClass value. + tos uint8 } // EndpointState represents the state of a UDP endpoint. @@ -119,6 +120,10 @@ type endpoint struct { // as ancillary data to ControlMessages on Read. receiveTOS bool + // receiveTClass determines if the incoming IPv6 TClass header field is + // passed as ancillary data to ControlMessages on Read. + receiveTClass bool + // receiveIPPacketInfo determines if the packet info is returned by Read. receiveIPPacketInfo bool @@ -258,13 +263,18 @@ func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMess } e.mu.RLock() receiveTOS := e.receiveTOS + receiveTClass := e.receiveTClass receiveIPPacketInfo := e.receiveIPPacketInfo e.mu.RUnlock() if receiveTOS { cm.HasTOS = true cm.TOS = p.tos } - + if receiveTClass { + cm.HasTClass = true + // Although TClass is an 8-bit value it's read in the CMsg as a uint32. + cm.TClass = uint32(p.tos) + } if receiveIPPacketInfo { cm.HasIPPacketInfo = true cm.PacketInfo = p.packetInfo @@ -490,6 +500,17 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { e.mu.Unlock() return nil + case tcpip.ReceiveTClassOption: + // We only support this option on v6 endpoints. + if e.NetProto != header.IPv6ProtocolNumber { + return tcpip.ErrNotSupported + } + + e.mu.Lock() + e.receiveTClass = v + e.mu.Unlock() + return nil + case tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. if e.NetProto != header.IPv6ProtocolNumber { @@ -709,6 +730,17 @@ func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { e.mu.RUnlock() return v, nil + case tcpip.ReceiveTClassOption: + // We only support this option on v6 endpoints. + if e.NetProto != header.IPv6ProtocolNumber { + return false, tcpip.ErrNotSupported + } + + e.mu.RLock() + v := e.receiveTClass + e.mu.RUnlock() + return v, nil + case tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. if e.NetProto != header.IPv6ProtocolNumber { @@ -1273,6 +1305,8 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk packet.packetInfo.LocalAddr = r.LocalAddress packet.packetInfo.DestinationAddr = r.RemoteAddress packet.packetInfo.NIC = r.NICID() + case header.IPv6ProtocolNumber: + packet.tos, _ = header.IPv6(pkt.NetworkHeader).TOS() } packet.timestamp = e.stack.NowNanoseconds() diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index f0ff3fe71..34b7c2360 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -409,6 +409,7 @@ func (c *testContext) injectV6Packet(payload []byte, h *header4Tuple, valid bool // Initialize the IP header. ip := header.IPv6(buf) ip.Encode(&header.IPv6Fields{ + TrafficClass: testTOS, PayloadLength: uint16(header.UDPMinimumSize + len(payload)), NextHeader: uint8(udp.ProtocolNumber), HopLimit: 65, @@ -1336,7 +1337,7 @@ func TestSetTTL(t *testing.T) { } } -func TestTOSV4(t *testing.T) { +func TestSetTOS(t *testing.T) { for _, flow := range []testFlow{unicastV4, multicastV4, broadcast} { t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { c := newDualTestContext(t, defaultMTU) @@ -1347,23 +1348,23 @@ func TestTOSV4(t *testing.T) { const tos = testTOS var v tcpip.IPv4TOSOption if err := c.ep.GetSockOpt(&v); err != nil { - c.t.Errorf("GetSockopt failed: %s", err) + c.t.Errorf("GetSockopt(%T) failed: %s", v, err) } // Test for expected default value. if v != 0 { - c.t.Errorf("got GetSockOpt(...) = %#v, want = %#v", v, 0) + c.t.Errorf("got GetSockOpt(%T) = 0x%x, want = 0x%x", v, v, 0) } if err := c.ep.SetSockOpt(tcpip.IPv4TOSOption(tos)); err != nil { - c.t.Errorf("SetSockOpt(%#v) failed: %s", tcpip.IPv4TOSOption(tos), err) + c.t.Errorf("SetSockOpt(%T, 0x%x) failed: %s", v, tcpip.IPv4TOSOption(tos), err) } if err := c.ep.GetSockOpt(&v); err != nil { - c.t.Errorf("GetSockopt failed: %s", err) + c.t.Errorf("GetSockopt(%T) failed: %s", v, err) } if want := tcpip.IPv4TOSOption(tos); v != want { - c.t.Errorf("got GetSockOpt(...) = %#v, want = %#v", v, want) + c.t.Errorf("got GetSockOpt(%T) = 0x%x, want = 0x%x", v, v, want) } testWrite(c, flow, checker.TOS(tos, 0)) @@ -1371,7 +1372,7 @@ func TestTOSV4(t *testing.T) { } } -func TestTOSV6(t *testing.T) { +func TestSetTClass(t *testing.T) { for _, flow := range []testFlow{unicastV4in6, unicastV6, unicastV6Only, multicastV4in6, multicastV6, broadcastIn6} { t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { c := newDualTestContext(t, defaultMTU) @@ -1379,71 +1380,92 @@ func TestTOSV6(t *testing.T) { c.createEndpointForFlow(flow) - const tos = testTOS + const tClass = testTOS var v tcpip.IPv6TrafficClassOption if err := c.ep.GetSockOpt(&v); err != nil { - c.t.Errorf("GetSockopt failed: %s", err) + c.t.Errorf("GetSockopt(%T) failed: %s", v, err) } // Test for expected default value. if v != 0 { - c.t.Errorf("got GetSockOpt(...) = %#v, want = %#v", v, 0) + c.t.Errorf("got GetSockOpt(%T) = 0x%x, want = 0x%x", v, v, 0) } - if err := c.ep.SetSockOpt(tcpip.IPv6TrafficClassOption(tos)); err != nil { - c.t.Errorf("SetSockOpt failed: %s", err) + if err := c.ep.SetSockOpt(tcpip.IPv6TrafficClassOption(tClass)); err != nil { + c.t.Errorf("SetSockOpt(%T, 0x%x) failed: %s", v, tcpip.IPv6TrafficClassOption(tClass), err) } if err := c.ep.GetSockOpt(&v); err != nil { - c.t.Errorf("GetSockopt failed: %s", err) + c.t.Errorf("GetSockopt(%T) failed: %s", v, err) } - if want := tcpip.IPv6TrafficClassOption(tos); v != want { - c.t.Errorf("got GetSockOpt(...) = %#v, want = %#v", v, want) + if want := tcpip.IPv6TrafficClassOption(tClass); v != want { + c.t.Errorf("got GetSockOpt(%T) = 0x%x, want = 0x%x", v, v, want) } - testWrite(c, flow, checker.TOS(tos, 0)) + // The header getter for TClass is called TOS, so use that checker. + testWrite(c, flow, checker.TOS(tClass, 0)) }) } } -func TestReceiveTOSV4(t *testing.T) { - for _, flow := range []testFlow{unicastV4, broadcast} { - t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() +func TestReceiveTosTClass(t *testing.T) { + testCases := []struct { + name string + getReceiveOption tcpip.SockOptBool + tests []testFlow + }{ + {"ReceiveTosOption", tcpip.ReceiveTOSOption, []testFlow{unicastV4, broadcast}}, + {"ReceiveTClassOption", tcpip.ReceiveTClassOption, []testFlow{unicastV4in6, unicastV6, unicastV6Only, broadcastIn6}}, + } + for _, testCase := range testCases { + for _, flow := range testCase.tests { + t.Run(fmt.Sprintf("%s:flow:%s", testCase.name, flow), func(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() - c.createEndpointForFlow(flow) + c.createEndpointForFlow(flow) + option := testCase.getReceiveOption + name := testCase.name - // Verify that setting and reading the option works. - v, err := c.ep.GetSockOptBool(tcpip.ReceiveTOSOption) - if err != nil { - c.t.Fatal("GetSockOptBool(tcpip.ReceiveTOSOption) failed:", err) - } - // Test for expected default value. - if v != false { - c.t.Errorf("got GetSockOptBool(tcpip.ReceiveTOSOption) = %t, want = %t", v, false) - } + // Verify that setting and reading the option works. + v, err := c.ep.GetSockOptBool(option) + if err != nil { + c.t.Errorf("GetSockoptBool(%s) failed: %s", name, err) + } + // Test for expected default value. + if v != false { + c.t.Errorf("got GetSockOptBool(%s) = %t, want = %t", name, v, false) + } - want := true - if err := c.ep.SetSockOptBool(tcpip.ReceiveTOSOption, want); err != nil { - c.t.Fatalf("SetSockOptBool(tcpip.ReceiveTOSOption, %t) failed: %s", want, err) - } + want := true + if err := c.ep.SetSockOptBool(option, want); err != nil { + c.t.Fatalf("SetSockOptBool(%s, %t) failed: %s", name, want, err) + } - got, err := c.ep.GetSockOptBool(tcpip.ReceiveTOSOption) - if err != nil { - c.t.Fatal("GetSockOptBool(tcpip.ReceiveTOSOption) failed:", err) - } - if got != want { - c.t.Fatalf("got GetSockOptBool(tcpip.ReceiveTOSOption) = %t, want = %t", got, want) - } + got, err := c.ep.GetSockOptBool(option) + if err != nil { + c.t.Errorf("GetSockoptBool(%s) failed: %s", name, err) + } - // Verify that the correct received TOS is handed through as - // ancillary data to the ControlMessages struct. - if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - c.t.Fatal("Bind failed:", err) - } - testRead(c, flow, checker.ReceiveTOS(testTOS)) - }) + if got != want { + c.t.Errorf("got GetSockOptBool(%s) = %t, want = %t", name, got, want) + } + + // Verify that the correct received TOS or TClass is handed through as + // ancillary data to the ControlMessages struct. + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + c.t.Fatalf("Bind failed: %s", err) + } + switch option { + case tcpip.ReceiveTClassOption: + testRead(c, flow, checker.ReceiveTClass(testTOS)) + case tcpip.ReceiveTOSOption: + testRead(c, flow, checker.ReceiveTOS(testTOS)) + default: + t.Fatalf("unknown test variant: %s", name) + } + }) + } } } diff --git a/test/syscalls/linux/ip_socket_test_util.h b/test/syscalls/linux/ip_socket_test_util.h index 083ebbcf0..39fd6709d 100644 --- a/test/syscalls/linux/ip_socket_test_util.h +++ b/test/syscalls/linux/ip_socket_test_util.h @@ -84,20 +84,20 @@ SocketPairKind DualStackUDPBidirectionalBindSocketPair(int type); // SocketPairs created with AF_INET and the given type. SocketPairKind IPv4UDPUnboundSocketPair(int type); -// IPv4UDPUnboundSocketPair returns a SocketKind that represents -// a SimpleSocket created with AF_INET, SOCK_DGRAM, and the given type. +// IPv4UDPUnboundSocket returns a SocketKind that represents a SimpleSocket +// created with AF_INET, SOCK_DGRAM, and the given type. SocketKind IPv4UDPUnboundSocket(int type); -// IPv6UDPUnboundSocketPair returns a SocketKind that represents -// a SimpleSocket created with AF_INET6, SOCK_DGRAM, and the given type. +// IPv6UDPUnboundSocket returns a SocketKind that represents a SimpleSocket +// created with AF_INET6, SOCK_DGRAM, and the given type. SocketKind IPv6UDPUnboundSocket(int type); -// IPv4TCPUnboundSocketPair returns a SocketKind that represents -// a SimpleSocket created with AF_INET, SOCK_STREAM and the given type. +// IPv4TCPUnboundSocket returns a SocketKind that represents a SimpleSocket +// created with AF_INET, SOCK_STREAM and the given type. SocketKind IPv4TCPUnboundSocket(int type); -// IPv6TCPUnboundSocketPair returns a SocketKind that represents -// a SimpleSocket created with AF_INET6, SOCK_STREAM and the given type. +// IPv6TCPUnboundSocket returns a SocketKind that represents a SimpleSocket +// created with AF_INET6, SOCK_STREAM and the given type. SocketKind IPv6TCPUnboundSocket(int type); // IfAddrHelper is a helper class that determines the local interfaces present diff --git a/test/syscalls/linux/socket_ip_udp_generic.cc b/test/syscalls/linux/socket_ip_udp_generic.cc index db5663ecd..1c533fdf2 100644 --- a/test/syscalls/linux/socket_ip_udp_generic.cc +++ b/test/syscalls/linux/socket_ip_udp_generic.cc @@ -14,6 +14,7 @@ #include "test/syscalls/linux/socket_ip_udp_generic.h" +#include #include #include #include @@ -209,46 +210,6 @@ TEST_P(UDPSocketPairTest, SetMulticastLoopChar) { EXPECT_EQ(get, kSockOptOn); } -// Ensure that Receiving TOS is off by default. -TEST_P(UDPSocketPairTest, RecvTosDefault) { - auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); - - int get = -1; - socklen_t get_len = sizeof(get); - ASSERT_THAT( - getsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, &get, &get_len), - SyscallSucceedsWithValue(0)); - EXPECT_EQ(get_len, sizeof(get)); - EXPECT_EQ(get, kSockOptOff); -} - -// Test that setting and getting IP_RECVTOS works as expected. -TEST_P(UDPSocketPairTest, SetRecvTos) { - auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); - - ASSERT_THAT(setsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, - &kSockOptOff, sizeof(kSockOptOff)), - SyscallSucceeds()); - - int get = -1; - socklen_t get_len = sizeof(get); - ASSERT_THAT( - getsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, &get, &get_len), - SyscallSucceedsWithValue(0)); - EXPECT_EQ(get_len, sizeof(get)); - EXPECT_EQ(get, kSockOptOff); - - ASSERT_THAT(setsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, - &kSockOptOn, sizeof(kSockOptOn)), - SyscallSucceeds()); - - ASSERT_THAT( - getsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, &get, &get_len), - SyscallSucceedsWithValue(0)); - EXPECT_EQ(get_len, sizeof(get)); - EXPECT_EQ(get, kSockOptOn); -} - TEST_P(UDPSocketPairTest, ReuseAddrDefault) { auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); @@ -401,5 +362,97 @@ TEST_P(UDPSocketPairTest, SetAndGetIPPKTINFO) { EXPECT_EQ(get_len, sizeof(get)); } +// Holds TOS or TClass information for IPv4 or IPv6 respectively. +struct RecvTosOption { + int level; + int option; +}; + +RecvTosOption GetRecvTosOption(int domain) { + TEST_CHECK(domain == AF_INET || domain == AF_INET6); + RecvTosOption opt; + switch (domain) { + case AF_INET: + opt.level = IPPROTO_IP; + opt.option = IP_RECVTOS; + break; + case AF_INET6: + opt.level = IPPROTO_IPV6; + opt.option = IPV6_RECVTCLASS; + break; + } + return opt; +} + +// Ensure that Receiving TOS or TCLASS is off by default. +TEST_P(UDPSocketPairTest, RecvTosDefault) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + RecvTosOption t = GetRecvTosOption(GetParam().domain); + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT( + getsockopt(sockets->first_fd(), t.level, t.option, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kSockOptOff); +} + +// Test that setting and getting IP_RECVTOS or IPV6_RECVTCLASS works as +// expected. +TEST_P(UDPSocketPairTest, SetRecvTos) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + RecvTosOption t = GetRecvTosOption(GetParam().domain); + + ASSERT_THAT(setsockopt(sockets->first_fd(), t.level, t.option, &kSockOptOff, + sizeof(kSockOptOff)), + SyscallSucceeds()); + + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT( + getsockopt(sockets->first_fd(), t.level, t.option, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kSockOptOff); + + ASSERT_THAT(setsockopt(sockets->first_fd(), t.level, t.option, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceeds()); + + ASSERT_THAT( + getsockopt(sockets->first_fd(), t.level, t.option, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kSockOptOn); +} + +// Test that any socket (including IPv6 only) accepts the IPv4 TOS option: this +// mirrors behavior in linux. +TEST_P(UDPSocketPairTest, TOSRecvMismatch) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + RecvTosOption t = GetRecvTosOption(AF_INET); + int get = -1; + socklen_t get_len = sizeof(get); + + ASSERT_THAT( + getsockopt(sockets->first_fd(), t.level, t.option, &get, &get_len), + SyscallSucceedsWithValue(0)); +} + +// Test that an IPv4 socket does not support the IPv6 TClass option. +TEST_P(UDPSocketPairTest, TClassRecvMismatch) { + // This should only test AF_INET sockets for the mismatch behavior. + SKIP_IF(GetParam().domain != AF_INET); + + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + int get = -1; + socklen_t get_len = sizeof(get); + + ASSERT_THAT(getsockopt(sockets->first_fd(), IPPROTO_IPV6, IPV6_RECVTCLASS, + &get, &get_len), + SyscallFailsWithErrno(EOPNOTSUPP)); +} + } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/udp_socket_test_cases.cc b/test/syscalls/linux/udp_socket_test_cases.cc index 9f8de6b48..57b1a357c 100644 --- a/test/syscalls/linux/udp_socket_test_cases.cc +++ b/test/syscalls/linux/udp_socket_test_cases.cc @@ -1349,9 +1349,6 @@ TEST_P(UdpSocketTest, TimestampIoctlPersistence) { // outgoing packets, and that a receiving socket with IP_RECVTOS or // IPV6_RECVTCLASS will create the corresponding control message. TEST_P(UdpSocketTest, SetAndReceiveTOS) { - // TODO(b/144868438): IPV6_RECVTCLASS not supported for netstack. - SKIP_IF((GetParam() != AddressFamily::kIpv4) && IsRunningOnGvisor() && - !IsRunningWithHostinet()); ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); ASSERT_THAT(connect(t_, addr_[0], addrlen_), SyscallSucceeds()); @@ -1422,7 +1419,6 @@ TEST_P(UdpSocketTest, SetAndReceiveTOS) { // TOS byte on outgoing packets, and that a receiving socket with IP_RECVTOS or // IPV6_RECVTCLASS will create the corresponding control message. TEST_P(UdpSocketTest, SendAndReceiveTOS) { - // TODO(b/144868438): IPV6_RECVTCLASS not supported for netstack. // TODO(b/146661005): Setting TOS via cmsg not supported for netstack. SKIP_IF(IsRunningOnGvisor() && !IsRunningWithHostinet()); ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); -- cgit v1.2.3 From abf7ebcd38e8c2750f4542f29115140bb2b44a9b Mon Sep 17 00:00:00 2001 From: Nayana Bidari Date: Thu, 27 Feb 2020 10:59:32 -0800 Subject: Internal change. PiperOrigin-RevId: 297638665 --- pkg/sentry/socket/netstack/netstack.go | 40 +++++++++-- pkg/tcpip/transport/packet/endpoint.go | 21 +++++- test/syscalls/linux/packet_socket.cc | 124 ++++++++++++++++++++++++++++++--- 3 files changed, 167 insertions(+), 18 deletions(-) (limited to 'pkg/sentry/socket/netstack/netstack.go') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index e187276c5..48c268bfa 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -712,14 +712,40 @@ func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking boo // Bind implements the linux syscall bind(2) for sockets backed by // tcpip.Endpoint. func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { - addr, family, err := AddressAndFamily(sockaddr) - if err != nil { - return err - } - if err := s.checkFamily(family, true /* exact */); err != nil { - return err + family := usermem.ByteOrder.Uint16(sockaddr) + var addr tcpip.FullAddress + + // Bind for AF_PACKET requires only family, protocol and ifindex. + // In function AddressAndFamily, we check the address length which is + // not needed for AF_PACKET bind. + if family == linux.AF_PACKET { + var a linux.SockAddrLink + if len(sockaddr) < sockAddrLinkSize { + return syserr.ErrInvalidArgument + } + binary.Unmarshal(sockaddr[:sockAddrLinkSize], usermem.ByteOrder, &a) + + if a.Protocol != uint16(s.protocol) { + return syserr.ErrInvalidArgument + } + + addr = tcpip.FullAddress{ + NIC: tcpip.NICID(a.InterfaceIndex), + Addr: tcpip.Address(a.HardwareAddr[:header.EthernetAddressSize]), + } + } else { + var err *syserr.Error + addr, family, err = AddressAndFamily(sockaddr) + if err != nil { + return err + } + + if err = s.checkFamily(family, true /* exact */); err != nil { + return err + } + + addr = s.mapFamily(addr, family) } - addr = s.mapFamily(addr, family) // Issue the bind request to the endpoint. return syserr.TranslateNetstackError(s.Endpoint.Bind(addr)) diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go index 5722815e9..09a1cd436 100644 --- a/pkg/tcpip/transport/packet/endpoint.go +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -76,6 +76,7 @@ type endpoint struct { sndBufSize int closed bool stats tcpip.TransportEndpointStats `state:"nosave"` + bound bool } // NewEndpoint returns a new packet endpoint. @@ -125,6 +126,7 @@ func (ep *endpoint) Close() { } ep.closed = true + ep.bound = false ep.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut) } @@ -216,7 +218,24 @@ func (ep *endpoint) Bind(addr tcpip.FullAddress) *tcpip.Error { // sll_family (should be AF_PACKET), sll_protocol, and sll_ifindex." // - packet(7). - return tcpip.ErrNotSupported + ep.mu.Lock() + defer ep.mu.Unlock() + + if ep.bound { + return tcpip.ErrAlreadyBound + } + + // Unregister endpoint with all the nics. + ep.stack.UnregisterPacketEndpoint(0, ep.netProto, ep) + + // Bind endpoint to receive packets from specific interface. + if err := ep.stack.RegisterPacketEndpoint(addr.NIC, ep.netProto, ep); err != nil { + return err + } + + ep.bound = true + + return nil } // GetLocalAddress implements tcpip.Endpoint.GetLocalAddress. diff --git a/test/syscalls/linux/packet_socket.cc b/test/syscalls/linux/packet_socket.cc index 92ae55eec..bc22de788 100644 --- a/test/syscalls/linux/packet_socket.cc +++ b/test/syscalls/linux/packet_socket.cc @@ -13,6 +13,7 @@ // limitations under the License. #include +#include #include #include #include @@ -163,16 +164,11 @@ int CookedPacketTest::GetLoopbackIndex() { return ifr.ifr_ifindex; } -// Receive via a packet socket. -TEST_P(CookedPacketTest, Receive) { - // Let's use a simple IP payload: a UDP datagram. - FileDescriptor udp_sock = - ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, 0)); - SendUDPMessage(udp_sock.get()); - +// Receive and verify the message via packet socket on interface. +void ReceiveMessage(int sock, int ifindex) { // Wait for the socket to become readable. struct pollfd pfd = {}; - pfd.fd = socket_; + pfd.fd = sock; pfd.events = POLLIN; EXPECT_THAT(RetryEINTR(poll)(&pfd, 1, 2000), SyscallSucceedsWithValue(1)); @@ -182,9 +178,10 @@ TEST_P(CookedPacketTest, Receive) { char buf[64]; struct sockaddr_ll src = {}; socklen_t src_len = sizeof(src); - ASSERT_THAT(recvfrom(socket_, buf, sizeof(buf), 0, + ASSERT_THAT(recvfrom(sock, buf, sizeof(buf), 0, reinterpret_cast(&src), &src_len), SyscallSucceedsWithValue(packet_size)); + // sockaddr_ll ends with an 8 byte physical address field, but ethernet // addresses only use 6 bytes. Linux used to return sizeof(sockaddr_ll)-2 // here, but since commit b2cf86e1563e33a14a1c69b3e508d15dc12f804c returns @@ -194,7 +191,7 @@ TEST_P(CookedPacketTest, Receive) { // TODO(b/129292371): Verify protocol once we return it. // Verify the source address. EXPECT_EQ(src.sll_family, AF_PACKET); - EXPECT_EQ(src.sll_ifindex, GetLoopbackIndex()); + EXPECT_EQ(src.sll_ifindex, ifindex); EXPECT_EQ(src.sll_halen, ETH_ALEN); // This came from the loopback device, so the address is all 0s. for (int i = 0; i < src.sll_halen; i++) { @@ -222,6 +219,18 @@ TEST_P(CookedPacketTest, Receive) { EXPECT_EQ(strncmp(payload, kMessage, sizeof(kMessage)), 0); } +// Receive via a packet socket. +TEST_P(CookedPacketTest, Receive) { + // Let's use a simple IP payload: a UDP datagram. + FileDescriptor udp_sock = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, 0)); + SendUDPMessage(udp_sock.get()); + + // Receive and verify the data. + int loopback_index = GetLoopbackIndex(); + ReceiveMessage(socket_, loopback_index); +} + // Send via a packet socket. TEST_P(CookedPacketTest, Send) { // TODO(b/129292371): Remove once we support packet socket writing. @@ -313,6 +322,101 @@ TEST_P(CookedPacketTest, Send) { EXPECT_EQ(src.sin_addr.s_addr, htonl(INADDR_LOOPBACK)); } +// Bind and receive via packet socket. +TEST_P(CookedPacketTest, BindReceive) { + struct sockaddr_ll bind_addr = {}; + bind_addr.sll_family = AF_PACKET; + bind_addr.sll_protocol = htons(GetParam()); + bind_addr.sll_ifindex = GetLoopbackIndex(); + + ASSERT_THAT(bind(socket_, reinterpret_cast(&bind_addr), + sizeof(bind_addr)), + SyscallSucceeds()); + + // Let's use a simple IP payload: a UDP datagram. + FileDescriptor udp_sock = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, 0)); + SendUDPMessage(udp_sock.get()); + + // Receive and verify the data. + ReceiveMessage(socket_, bind_addr.sll_ifindex); +} + +// Double Bind socket. +TEST_P(CookedPacketTest, DoubleBind) { + struct sockaddr_ll bind_addr = {}; + bind_addr.sll_family = AF_PACKET; + bind_addr.sll_protocol = htons(GetParam()); + bind_addr.sll_ifindex = GetLoopbackIndex(); + + ASSERT_THAT(bind(socket_, reinterpret_cast(&bind_addr), + sizeof(bind_addr)), + SyscallSucceeds()); + + // Binding socket again should fail. + ASSERT_THAT( + bind(socket_, reinterpret_cast(&bind_addr), + sizeof(bind_addr)), + // Linux 4.09 returns EINVAL here, but some time before 4.19 it switched + // to EADDRINUSE. + AnyOf(SyscallFailsWithErrno(EADDRINUSE), SyscallFailsWithErrno(EINVAL))); +} + +// Bind and verify we do not receive data on interface which is not bound +TEST_P(CookedPacketTest, BindDrop) { + // Let's use a simple IP payload: a UDP datagram. + FileDescriptor udp_sock = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, 0)); + + struct ifaddrs* if_addr_list = nullptr; + auto cleanup = Cleanup([&if_addr_list]() { freeifaddrs(if_addr_list); }); + + ASSERT_THAT(getifaddrs(&if_addr_list), SyscallSucceeds()); + + // Get interface other than loopback. + struct ifreq ifr = {}; + for (struct ifaddrs* i = if_addr_list; i; i = i->ifa_next) { + if (strcmp(i->ifa_name, "lo") != 0) { + strncpy(ifr.ifr_name, i->ifa_name, sizeof(ifr.ifr_name)); + break; + } + } + + // Skip if no interface is available other than loopback. + if (strlen(ifr.ifr_name) == 0) { + GTEST_SKIP(); + } + + // Get interface index. + EXPECT_THAT(ioctl(socket_, SIOCGIFINDEX, &ifr), SyscallSucceeds()); + EXPECT_NE(ifr.ifr_ifindex, 0); + + // Bind to packet socket requires only family, protocol and ifindex. + struct sockaddr_ll bind_addr = {}; + bind_addr.sll_family = AF_PACKET; + bind_addr.sll_protocol = htons(GetParam()); + bind_addr.sll_ifindex = ifr.ifr_ifindex; + + ASSERT_THAT(bind(socket_, reinterpret_cast(&bind_addr), + sizeof(bind_addr)), + SyscallSucceeds()); + + // Send to loopback interface. + struct sockaddr_in dest = {}; + dest.sin_addr.s_addr = htonl(INADDR_LOOPBACK); + dest.sin_family = AF_INET; + dest.sin_port = kPort; + EXPECT_THAT(sendto(udp_sock.get(), kMessage, sizeof(kMessage), 0, + reinterpret_cast(&dest), sizeof(dest)), + SyscallSucceedsWithValue(sizeof(kMessage))); + + // Wait and make sure the socket never receives any data. + struct pollfd pfd = {}; + pfd.fd = socket_; + pfd.events = POLLIN; + EXPECT_THAT(RetryEINTR(poll)(&pfd, 1, 1000), SyscallSucceedsWithValue(0)); +} + INSTANTIATE_TEST_SUITE_P(AllInetTests, CookedPacketTest, ::testing::Values(ETH_P_IP, ETH_P_ALL)); -- cgit v1.2.3 From 42fb7d349137bd8847e7c3df6493fde3bc8e6e89 Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Mon, 2 Mar 2020 10:32:20 -0800 Subject: socket: take readMu to access readView DATA RACE in netstack.(*SocketOperations).fetchReadView Write at 0x00c001dca138 by goroutine 1001: gvisor.dev/gvisor/pkg/sentry/socket/netstack.(*SocketOperations).fetchReadView() pkg/sentry/socket/netstack/netstack.go:418 +0x85 gvisor.dev/gvisor/pkg/sentry/socket/netstack.(*SocketOperations).coalescingRead() pkg/sentry/socket/netstack/netstack.go:2309 +0x67 gvisor.dev/gvisor/pkg/sentry/socket/netstack.(*SocketOperations).nonBlockingRead() pkg/sentry/socket/netstack/netstack.go:2378 +0x183d Previous read at 0x00c001dca138 by goroutine 1111: gvisor.dev/gvisor/pkg/sentry/socket/netstack.(*SocketOperations).Ioctl() pkg/sentry/socket/netstack/netstack.go:2666 +0x533 gvisor.dev/gvisor/pkg/sentry/syscalls/linux.Ioctl() Reported-by: syzbot+d4c3885fcc346f08deb6@syzkaller.appspotmail.com PiperOrigin-RevId: 298387377 --- pkg/sentry/socket/netstack/netstack.go | 2 ++ 1 file changed, 2 insertions(+) (limited to 'pkg/sentry/socket/netstack/netstack.go') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 48c268bfa..1eeb37446 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -2663,7 +2663,9 @@ func (s *SocketOperations) Ioctl(ctx context.Context, _ *fs.File, io usermem.IO, } // Add bytes removed from the endpoint but not yet sent to the caller. + s.readMu.Lock() v += len(s.readView) + s.readMu.Unlock() if v > math.MaxInt32 { v = math.MaxInt32 -- cgit v1.2.3 From 43abb24657e737dee1108ff0d512b2e1b6d8a3f6 Mon Sep 17 00:00:00 2001 From: Nayana Bidari Date: Mon, 2 Mar 2020 16:30:51 -0800 Subject: Fix panic caused by invalid address for Bind in packet sockets. PiperOrigin-RevId: 298476533 --- pkg/sentry/socket/netstack/netstack.go | 4 ++++ test/syscalls/linux/packet_socket.cc | 13 +++++++++++++ 2 files changed, 17 insertions(+) (limited to 'pkg/sentry/socket/netstack/netstack.go') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 1eeb37446..13a9a60b4 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -712,6 +712,10 @@ func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking boo // Bind implements the linux syscall bind(2) for sockets backed by // tcpip.Endpoint. func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { + if len(sockaddr) < 2 { + return syserr.ErrInvalidArgument + } + family := usermem.ByteOrder.Uint16(sockaddr) var addr tcpip.FullAddress diff --git a/test/syscalls/linux/packet_socket.cc b/test/syscalls/linux/packet_socket.cc index bc22de788..248762ca9 100644 --- a/test/syscalls/linux/packet_socket.cc +++ b/test/syscalls/linux/packet_socket.cc @@ -417,6 +417,19 @@ TEST_P(CookedPacketTest, BindDrop) { EXPECT_THAT(RetryEINTR(poll)(&pfd, 1, 1000), SyscallSucceedsWithValue(0)); } +// Bind with invalid address. +TEST_P(CookedPacketTest, BindFail) { + // Null address. + ASSERT_THAT(bind(socket_, nullptr, sizeof(struct sockaddr)), + SyscallFailsWithErrno(EFAULT)); + + // Address of size 1. + uint8_t addr = 0; + ASSERT_THAT( + bind(socket_, reinterpret_cast(&addr), sizeof(addr)), + SyscallFailsWithErrno(EINVAL)); +} + INSTANTIATE_TEST_SUITE_P(AllInetTests, CookedPacketTest, ::testing::Values(ETH_P_IP, ETH_P_ALL)); -- cgit v1.2.3