diff options
Diffstat (limited to 'pkg/tcpip/link')
26 files changed, 581 insertions, 169 deletions
diff --git a/pkg/tcpip/link/channel/BUILD b/pkg/tcpip/link/channel/BUILD index 97a794986..7dbc05754 100644 --- a/pkg/tcpip/link/channel/BUILD +++ b/pkg/tcpip/link/channel/BUILD @@ -6,7 +6,7 @@ go_library( name = "channel", srcs = ["channel.go"], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/channel", - visibility = ["//:sandbox"], + visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/link/channel/channel.go b/pkg/tcpip/link/channel/channel.go index 18adb2085..70188551f 100644 --- a/pkg/tcpip/link/channel/channel.go +++ b/pkg/tcpip/link/channel/channel.go @@ -25,10 +25,9 @@ import ( // PacketInfo holds all the information about an outbound packet. type PacketInfo struct { - Header buffer.View - Payload buffer.View - Proto tcpip.NetworkProtocolNumber - GSO *stack.GSO + Pkt tcpip.PacketBuffer + Proto tcpip.NetworkProtocolNumber + GSO *stack.GSO } // Endpoint is link layer endpoint that stores outbound packets in a channel @@ -65,14 +64,14 @@ func (e *Endpoint) Drain() int { } } -// Inject injects an inbound packet. -func (e *Endpoint) Inject(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { - e.InjectLinkAddr(protocol, "", vv) +// InjectInbound injects an inbound packet. +func (e *Endpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { + e.InjectLinkAddr(protocol, "", pkt) } // InjectLinkAddr injects an inbound packet with a remote link address. -func (e *Endpoint) InjectLinkAddr(protocol tcpip.NetworkProtocolNumber, remote tcpip.LinkAddress, vv buffer.VectorisedView) { - e.dispatcher.DeliverNetworkPacket(e, remote, "" /* local */, protocol, vv.Clone(nil)) +func (e *Endpoint) InjectLinkAddr(protocol tcpip.NetworkProtocolNumber, remote tcpip.LinkAddress, pkt tcpip.PacketBuffer) { + e.dispatcher.DeliverNetworkPacket(e, remote, "" /* local */, protocol, pkt) } // Attach saves the stack network-layer dispatcher for use later when packets @@ -96,7 +95,7 @@ func (e *Endpoint) MTU() uint32 { func (e *Endpoint) Capabilities() stack.LinkEndpointCapabilities { caps := stack.LinkEndpointCapabilities(0) if e.GSO { - caps |= stack.CapabilityGSO + caps |= stack.CapabilityHardwareGSO } return caps } @@ -118,12 +117,55 @@ func (e *Endpoint) LinkAddress() tcpip.LinkAddress { } // WritePacket stores outbound packets into the channel. -func (e *Endpoint) WritePacket(_ *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (e *Endpoint) WritePacket(_ *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { p := PacketInfo{ - Header: hdr.View(), - Proto: protocol, - Payload: payload.ToView(), - GSO: gso, + Pkt: pkt, + Proto: protocol, + GSO: gso, + } + + select { + case e.C <- p: + default: + } + + return nil +} + +// WritePackets stores outbound packets into the channel. +func (e *Endpoint) WritePackets(_ *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + payloadView := pkts[0].Data.ToView() + n := 0 +packetLoop: + for _, pkt := range pkts { + off := pkt.DataOffset + size := pkt.DataSize + p := PacketInfo{ + Pkt: tcpip.PacketBuffer{ + Header: pkt.Header, + Data: buffer.NewViewFromBytes(payloadView[off : off+size]).ToVectorisedView(), + }, + Proto: protocol, + GSO: gso, + } + + select { + case e.C <- p: + n++ + default: + break packetLoop + } + } + + return n, nil +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *Endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { + p := PacketInfo{ + Pkt: tcpip.PacketBuffer{Data: vv}, + Proto: 0, + GSO: nil, } select { diff --git a/pkg/tcpip/link/fdbased/BUILD b/pkg/tcpip/link/fdbased/BUILD index 8fa9e3984..897c94821 100644 --- a/pkg/tcpip/link/fdbased/BUILD +++ b/pkg/tcpip/link/fdbased/BUILD @@ -14,9 +14,7 @@ go_library( "packet_dispatchers.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/fdbased", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index f80ac3435..fa8a703d9 100644 --- a/pkg/tcpip/link/fdbased/endpoint.go +++ b/pkg/tcpip/link/fdbased/endpoint.go @@ -165,6 +165,9 @@ type Options struct { // disabled. GSOMaxSize uint32 + // SoftwareGSOEnabled indicates whether software GSO is enabled or not. + SoftwareGSOEnabled bool + // PacketDispatchMode specifies the type of inbound dispatcher to be // used for this endpoint. PacketDispatchMode PacketDispatchMode @@ -242,7 +245,11 @@ func New(opts *Options) (stack.LinkEndpoint, error) { } if isSocket { if opts.GSOMaxSize != 0 { - e.caps |= stack.CapabilityGSO + if opts.SoftwareGSOEnabled { + e.caps |= stack.CapabilitySoftwareGSO + } else { + e.caps |= stack.CapabilityHardwareGSO + } e.gsoMaxSize = opts.GSOMaxSize } } @@ -379,10 +386,11 @@ const ( // WritePacket writes outbound packets to the file descriptor. If it is not // currently writable, the packet is dropped. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { if e.hdrSize > 0 { // Add ethernet header if needed. - eth := header.Ethernet(hdr.Prepend(header.EthernetMinimumSize)) + eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) + pkt.LinkHeader = buffer.View(eth) ethHdr := &header.EthernetFields{ DstAddr: r.RemoteLinkAddress, Type: protocol, @@ -397,17 +405,17 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen eth.Encode(ethHdr) } - if e.Capabilities()&stack.CapabilityGSO != 0 { + if e.Capabilities()&stack.CapabilityHardwareGSO != 0 { vnetHdr := virtioNetHdr{} vnetHdrBuf := vnetHdrToByteSlice(&vnetHdr) if gso != nil { - vnetHdr.hdrLen = uint16(hdr.UsedLength()) + vnetHdr.hdrLen = uint16(pkt.Header.UsedLength()) if gso.NeedsCsum { vnetHdr.flags = _VIRTIO_NET_HDR_F_NEEDS_CSUM vnetHdr.csumStart = header.EthernetMinimumSize + gso.L3HdrLen vnetHdr.csumOffset = gso.CsumOffset } - if gso.Type != stack.GSONone && uint16(payload.Size()) > gso.MSS { + if gso.Type != stack.GSONone && uint16(pkt.Data.Size()) > gso.MSS { switch gso.Type { case stack.GSOTCPv4: vnetHdr.gsoType = _VIRTIO_NET_HDR_GSO_TCPV4 @@ -420,18 +428,151 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen } } - return rawfile.NonBlockingWrite3(e.fds[0], vnetHdrBuf, hdr.View(), payload.ToView()) + return rawfile.NonBlockingWrite3(e.fds[0], vnetHdrBuf, pkt.Header.View(), pkt.Data.ToView()) + } + + if pkt.Data.Size() == 0 { + return rawfile.NonBlockingWrite(e.fds[0], pkt.Header.View()) } - if payload.Size() == 0 { - return rawfile.NonBlockingWrite(e.fds[0], hdr.View()) + return rawfile.NonBlockingWrite3(e.fds[0], pkt.Header.View(), pkt.Data.ToView(), nil) +} + +// WritePackets writes outbound packets to the file descriptor. If it is not +// currently writable, the packet is dropped. +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + var ethHdrBuf []byte + // hdr + data + iovLen := 2 + if e.hdrSize > 0 { + // Add ethernet header if needed. + ethHdrBuf = make([]byte, header.EthernetMinimumSize) + eth := header.Ethernet(ethHdrBuf) + ethHdr := &header.EthernetFields{ + DstAddr: r.RemoteLinkAddress, + Type: protocol, + } + + // Preserve the src address if it's set in the route. + if r.LocalLinkAddress != "" { + ethHdr.SrcAddr = r.LocalLinkAddress + } else { + ethHdr.SrcAddr = e.addr + } + eth.Encode(ethHdr) + iovLen++ } - return rawfile.NonBlockingWrite3(e.fds[0], hdr.View(), payload.ToView(), nil) + n := len(pkts) + + views := pkts[0].Data.Views() + /* + * Each bondary in views can add one more iovec. + * + * payload | | | | + * ----------------------------- + * packets | | | | | | | + * ----------------------------- + * iovecs | | | | | | | | | + */ + iovec := make([]syscall.Iovec, n*iovLen+len(views)-1) + mmsgHdrs := make([]rawfile.MMsgHdr, n) + + iovecIdx := 0 + viewIdx := 0 + viewOff := 0 + off := 0 + nextOff := 0 + for i := range pkts { + // TODO(b/134618279): Different packets may have different data + // in the future. We should handle this. + if !viewsEqual(pkts[i].Data.Views(), views) { + panic("All packets in pkts should have the same Data.") + } + + prevIovecIdx := iovecIdx + mmsgHdr := &mmsgHdrs[i] + mmsgHdr.Msg.Iov = &iovec[iovecIdx] + packetSize := pkts[i].DataSize + hdr := &pkts[i].Header + + off = pkts[i].DataOffset + if off != nextOff { + // We stop in a different point last time. + size := packetSize + viewIdx = 0 + viewOff = 0 + for size > 0 { + if size >= len(views[viewIdx]) { + viewIdx++ + viewOff = 0 + size -= len(views[viewIdx]) + } else { + viewOff = size + size = 0 + } + } + } + nextOff = off + packetSize + + if ethHdrBuf != nil { + v := &iovec[iovecIdx] + v.Base = ðHdrBuf[0] + v.Len = uint64(len(ethHdrBuf)) + iovecIdx++ + } + + v := &iovec[iovecIdx] + hdrView := hdr.View() + v.Base = &hdrView[0] + v.Len = uint64(len(hdrView)) + iovecIdx++ + + for packetSize > 0 { + vec := &iovec[iovecIdx] + iovecIdx++ + + v := views[viewIdx] + vec.Base = &v[viewOff] + s := len(v) - viewOff + if s <= packetSize { + viewIdx++ + viewOff = 0 + } else { + s = packetSize + viewOff += s + } + vec.Len = uint64(s) + packetSize -= s + } + + mmsgHdr.Msg.Iovlen = uint64(iovecIdx - prevIovecIdx) + } + + packets := 0 + for packets < n { + sent, err := rawfile.NonBlockingSendMMsg(e.fds[0], mmsgHdrs) + if err != nil { + return packets, err + } + packets += sent + mmsgHdrs = mmsgHdrs[sent:] + } + return packets, nil +} + +// viewsEqual tests whether v1 and v2 refer to the same backing bytes. +func viewsEqual(vs1, vs2 []buffer.View) bool { + return len(vs1) == len(vs2) && (len(vs1) == 0 || &vs1[0] == &vs2[0]) +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { + return rawfile.NonBlockingWrite(e.fds[0], vv.ToView()) } -// WriteRawPacket writes a raw packet directly to the file descriptor. -func (e *endpoint) WriteRawPacket(dest tcpip.Address, packet []byte) *tcpip.Error { +// InjectOutobund implements stack.InjectableEndpoint.InjectOutbound. +func (e *endpoint) InjectOutbound(dest tcpip.Address, packet []byte) *tcpip.Error { return rawfile.NonBlockingWrite(e.fds[0], packet) } @@ -468,9 +609,9 @@ func (e *InjectableEndpoint) Attach(dispatcher stack.NetworkDispatcher) { e.dispatcher = dispatcher } -// Inject injects an inbound packet. -func (e *InjectableEndpoint) Inject(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { - e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, vv) +// InjectInbound injects an inbound packet. +func (e *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { + e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, pkt) } // NewInjectable creates a new fd-based InjectableEndpoint. diff --git a/pkg/tcpip/link/fdbased/endpoint_test.go b/pkg/tcpip/link/fdbased/endpoint_test.go index 04406bc9a..2066987eb 100644 --- a/pkg/tcpip/link/fdbased/endpoint_test.go +++ b/pkg/tcpip/link/fdbased/endpoint_test.go @@ -45,7 +45,7 @@ const ( type packetInfo struct { raddr tcpip.LinkAddress proto tcpip.NetworkProtocolNumber - contents buffer.View + contents tcpip.PacketBuffer } type context struct { @@ -92,8 +92,8 @@ func (c *context) cleanup() { syscall.Close(c.fds[1]) } -func (c *context) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote tcpip.LinkAddress, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { - c.ch <- packetInfo{remote, protocol, vv.ToView()} +func (c *context) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote tcpip.LinkAddress, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { + c.ch <- packetInfo{remote, protocol, pkt} } func TestNoEthernetProperties(t *testing.T) { @@ -168,7 +168,10 @@ func testWritePacket(t *testing.T, plen int, eth bool, gsoMaxSize uint32) { L3HdrLen: header.IPv4MaximumHeaderSize, } } - if err := c.ep.WritePacket(r, gso, hdr, payload.ToVectorisedView(), proto); err != nil { + if err := c.ep.WritePacket(r, gso, proto, tcpip.PacketBuffer{ + Header: hdr, + Data: payload.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed: %v", err) } @@ -258,7 +261,10 @@ func TestPreserveSrcAddress(t *testing.T) { // WritePacket panics given a prependable with anything less than // the minimum size of the ethernet header. hdr := buffer.NewPrependable(header.EthernetMinimumSize) - if err := c.ep.WritePacket(r, nil /* gso */, hdr, buffer.VectorisedView{}, proto); err != nil { + if err := c.ep.WritePacket(r, nil /* gso */, proto, tcpip.PacketBuffer{ + Header: hdr, + Data: buffer.VectorisedView{}, + }); err != nil { t.Fatalf("WritePacket failed: %v", err) } @@ -293,11 +299,12 @@ func TestDeliverPacket(t *testing.T) { b[i] = uint8(rand.Intn(256)) } + var hdr header.Ethernet if !eth { // So that it looks like an IPv4 packet. b[0] = 0x40 } else { - hdr := make(header.Ethernet, header.EthernetMinimumSize) + hdr = make(header.Ethernet, header.EthernetMinimumSize) hdr.Encode(&header.EthernetFields{ SrcAddr: raddr, DstAddr: laddr, @@ -315,14 +322,21 @@ func TestDeliverPacket(t *testing.T) { select { case pi := <-c.ch: want := packetInfo{ - raddr: raddr, - proto: proto, - contents: b, + raddr: raddr, + proto: proto, + contents: tcpip.PacketBuffer{ + Data: buffer.View(b).ToVectorisedView(), + LinkHeader: buffer.View(hdr), + }, } if !eth { want.proto = header.IPv4ProtocolNumber want.raddr = "" } + // want.contents.Data will be a single + // view, so make pi do the same for the + // DeepEqual check. + pi.contents.Data = pi.contents.Data.ToView().ToVectorisedView() if !reflect.DeepEqual(want, pi) { t.Fatalf("Unexpected received packet: %+v, want %+v", pi, want) } diff --git a/pkg/tcpip/link/fdbased/mmap.go b/pkg/tcpip/link/fdbased/mmap.go index 8bfeb97e4..62ed1e569 100644 --- a/pkg/tcpip/link/fdbased/mmap.go +++ b/pkg/tcpip/link/fdbased/mmap.go @@ -169,9 +169,10 @@ func (d *packetMMapDispatcher) dispatch() (bool, *tcpip.Error) { var ( p tcpip.NetworkProtocolNumber remote, local tcpip.LinkAddress + eth header.Ethernet ) if d.e.hdrSize > 0 { - eth := header.Ethernet(pkt) + eth = header.Ethernet(pkt) p = eth.Type() remote = eth.SourceAddress() local = eth.DestinationAddress() @@ -189,6 +190,9 @@ func (d *packetMMapDispatcher) dispatch() (bool, *tcpip.Error) { } pkt = pkt[d.e.hdrSize:] - d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, buffer.NewVectorisedView(len(pkt), []buffer.View{buffer.View(pkt)})) + d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, tcpip.PacketBuffer{ + Data: buffer.View(pkt).ToVectorisedView(), + LinkHeader: buffer.View(eth), + }) return true, nil } diff --git a/pkg/tcpip/link/fdbased/packet_dispatchers.go b/pkg/tcpip/link/fdbased/packet_dispatchers.go index 7ca217e5b..c67d684ce 100644 --- a/pkg/tcpip/link/fdbased/packet_dispatchers.go +++ b/pkg/tcpip/link/fdbased/packet_dispatchers.go @@ -53,7 +53,7 @@ func newReadVDispatcher(fd int, e *endpoint) (linkDispatcher, error) { d := &readVDispatcher{fd: fd, e: e} d.views = make([]buffer.View, len(BufConfig)) iovLen := len(BufConfig) - if d.e.Capabilities()&stack.CapabilityGSO != 0 { + if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 { iovLen++ } d.iovecs = make([]syscall.Iovec, iovLen) @@ -63,7 +63,7 @@ func newReadVDispatcher(fd int, e *endpoint) (linkDispatcher, error) { func (d *readVDispatcher) allocateViews(bufConfig []int) { var vnetHdr [virtioNetHdrSize]byte vnetHdrOff := 0 - if d.e.Capabilities()&stack.CapabilityGSO != 0 { + if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 { // The kernel adds virtioNetHdr before each packet, but // we don't use it, so so we allocate a buffer for it, // add it in iovecs but don't add it in a view. @@ -106,7 +106,7 @@ func (d *readVDispatcher) dispatch() (bool, *tcpip.Error) { if err != nil { return false, err } - if d.e.Capabilities()&stack.CapabilityGSO != 0 { + if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 { // Skip virtioNetHdr which is added before each packet, it // isn't used and it isn't in a view. n -= virtioNetHdrSize @@ -118,9 +118,10 @@ func (d *readVDispatcher) dispatch() (bool, *tcpip.Error) { var ( p tcpip.NetworkProtocolNumber remote, local tcpip.LinkAddress + eth header.Ethernet ) if d.e.hdrSize > 0 { - eth := header.Ethernet(d.views[0]) + eth = header.Ethernet(d.views[0][:header.EthernetMinimumSize]) p = eth.Type() remote = eth.SourceAddress() local = eth.DestinationAddress() @@ -138,10 +139,13 @@ func (d *readVDispatcher) dispatch() (bool, *tcpip.Error) { } used := d.capViews(n, BufConfig) - vv := buffer.NewVectorisedView(n, d.views[:used]) - vv.TrimFront(d.e.hdrSize) + pkt := tcpip.PacketBuffer{ + Data: buffer.NewVectorisedView(n, append([]buffer.View(nil), d.views[:used]...)), + LinkHeader: buffer.View(eth), + } + pkt.Data.TrimFront(d.e.hdrSize) - d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, vv) + d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, pkt) // Prepare e.views for another packet: release used views. for i := 0; i < used; i++ { @@ -194,7 +198,7 @@ func newRecvMMsgDispatcher(fd int, e *endpoint) (linkDispatcher, error) { } d.iovecs = make([][]syscall.Iovec, MaxMsgsPerRecv) iovLen := len(BufConfig) - if d.e.Capabilities()&stack.CapabilityGSO != 0 { + if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 { // virtioNetHdr is prepended before each packet. iovLen++ } @@ -225,7 +229,7 @@ func (d *recvMMsgDispatcher) allocateViews(bufConfig []int) { for k := 0; k < len(d.views); k++ { var vnetHdr [virtioNetHdrSize]byte vnetHdrOff := 0 - if d.e.Capabilities()&stack.CapabilityGSO != 0 { + if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 { // The kernel adds virtioNetHdr before each packet, but // we don't use it, so so we allocate a buffer for it, // add it in iovecs but don't add it in a view. @@ -261,7 +265,7 @@ func (d *recvMMsgDispatcher) dispatch() (bool, *tcpip.Error) { // Process each of received packets. for k := 0; k < nMsgs; k++ { n := int(d.msgHdrs[k].Len) - if d.e.Capabilities()&stack.CapabilityGSO != 0 { + if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 { n -= virtioNetHdrSize } if n <= d.e.hdrSize { @@ -271,9 +275,10 @@ func (d *recvMMsgDispatcher) dispatch() (bool, *tcpip.Error) { var ( p tcpip.NetworkProtocolNumber remote, local tcpip.LinkAddress + eth header.Ethernet ) if d.e.hdrSize > 0 { - eth := header.Ethernet(d.views[k][0]) + eth = header.Ethernet(d.views[k][0]) p = eth.Type() remote = eth.SourceAddress() local = eth.DestinationAddress() @@ -291,9 +296,12 @@ func (d *recvMMsgDispatcher) dispatch() (bool, *tcpip.Error) { } used := d.capViews(k, int(n), BufConfig) - vv := buffer.NewVectorisedView(int(n), d.views[k][:used]) - vv.TrimFront(d.e.hdrSize) - d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, vv) + pkt := tcpip.PacketBuffer{ + Data: buffer.NewVectorisedView(int(n), append([]buffer.View(nil), d.views[k][:used]...)), + LinkHeader: buffer.View(eth), + } + pkt.Data.TrimFront(d.e.hdrSize) + d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, pkt) // Prepare e.views for another packet: release used views. for i := 0; i < used; i++ { diff --git a/pkg/tcpip/link/loopback/BUILD b/pkg/tcpip/link/loopback/BUILD index 47a54845c..f35fcdff4 100644 --- a/pkg/tcpip/link/loopback/BUILD +++ b/pkg/tcpip/link/loopback/BUILD @@ -6,10 +6,11 @@ go_library( name = "loopback", srcs = ["loopback.go"], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/loopback", - visibility = ["//:sandbox"], + visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/header", "//pkg/tcpip/stack", ], ) diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index b36629d2c..499cc608f 100644 --- a/pkg/tcpip/link/loopback/loopback.go +++ b/pkg/tcpip/link/loopback/loopback.go @@ -23,6 +23,7 @@ package loopback import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -70,21 +71,45 @@ func (*endpoint) LinkAddress() tcpip.LinkAddress { return "" } +// Wait implements stack.LinkEndpoint.Wait. +func (*endpoint) Wait() {} + // WritePacket implements stack.LinkEndpoint.WritePacket. It delivers outbound // packets to the network-layer dispatcher. -func (e *endpoint) WritePacket(_ *stack.Route, _ *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { - views := make([]buffer.View, 1, 1+len(payload.Views())) - views[0] = hdr.View() - views = append(views, payload.Views()...) - vv := buffer.NewVectorisedView(len(views[0])+payload.Size(), views) +func (e *endpoint) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { + views := make([]buffer.View, 1, 1+len(pkt.Data.Views())) + views[0] = pkt.Header.View() + views = append(views, pkt.Data.Views()...) - // Because we're immediately turning around and writing the packet back to the - // rx path, we intentionally don't preserve the remote and local link - // addresses from the stack.Route we're passed. - e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, vv) + // Because we're immediately turning around and writing the packet back + // to the rx path, we intentionally don't preserve the remote and local + // link addresses from the stack.Route we're passed. + e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, tcpip.PacketBuffer{ + Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), + }) return nil } -// Wait implements stack.LinkEndpoint.Wait. -func (*endpoint) Wait() {} +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, []tcpip.PacketBuffer, tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + panic("not implemented") +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { + // Reject the packet if it's shorter than an ethernet header. + if vv.Size() < header.EthernetMinimumSize { + return tcpip.ErrBadAddress + } + + // There should be an ethernet header at the beginning of vv. + linkHeader := header.Ethernet(vv.First()[:header.EthernetMinimumSize]) + vv.TrimFront(len(linkHeader)) + e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, linkHeader.Type(), tcpip.PacketBuffer{ + Data: vv, + LinkHeader: buffer.View(linkHeader), + }) + + return nil +} diff --git a/pkg/tcpip/link/muxed/BUILD b/pkg/tcpip/link/muxed/BUILD index 1bab380b0..1ac7948b6 100644 --- a/pkg/tcpip/link/muxed/BUILD +++ b/pkg/tcpip/link/muxed/BUILD @@ -7,9 +7,7 @@ go_library( name = "muxed", srcs = ["injectable.go"], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/muxed", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/link/muxed/injectable.go b/pkg/tcpip/link/muxed/injectable.go index 7c946101d..445b22c17 100644 --- a/pkg/tcpip/link/muxed/injectable.go +++ b/pkg/tcpip/link/muxed/injectable.go @@ -79,29 +79,47 @@ func (m *InjectableEndpoint) IsAttached() bool { return m.dispatcher != nil } -// Inject implements stack.InjectableLinkEndpoint. -func (m *InjectableEndpoint) Inject(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { - m.dispatcher.DeliverNetworkPacket(m, "" /* remote */, "" /* local */, protocol, vv) +// InjectInbound implements stack.InjectableLinkEndpoint. +func (m *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { + m.dispatcher.DeliverNetworkPacket(m, "" /* remote */, "" /* local */, protocol, pkt) +} + +// WritePackets writes outbound packets to the appropriate +// LinkInjectableEndpoint based on the RemoteAddress. HandleLocal only works if +// r.RemoteAddress has a route registered in this endpoint. +func (m *InjectableEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + endpoint, ok := m.routes[r.RemoteAddress] + if !ok { + return 0, tcpip.ErrNoRoute + } + return endpoint.WritePackets(r, gso, pkts, protocol) } // WritePacket writes outbound packets to the appropriate LinkInjectableEndpoint // based on the RemoteAddress. HandleLocal only works if r.RemoteAddress has a // route registered in this endpoint. -func (m *InjectableEndpoint) WritePacket(r *stack.Route, _ *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (m *InjectableEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { if endpoint, ok := m.routes[r.RemoteAddress]; ok { - return endpoint.WritePacket(r, nil /* gso */, hdr, payload, protocol) + return endpoint.WritePacket(r, gso, protocol, pkt) } return tcpip.ErrNoRoute } -// WriteRawPacket writes outbound packets to the appropriate +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (m *InjectableEndpoint) WriteRawPacket(buffer.VectorisedView) *tcpip.Error { + // WriteRawPacket doesn't get a route or network address, so there's + // nowhere to write this. + return tcpip.ErrNoRoute +} + +// InjectOutbound writes outbound packets to the appropriate // LinkInjectableEndpoint based on the dest address. -func (m *InjectableEndpoint) WriteRawPacket(dest tcpip.Address, packet []byte) *tcpip.Error { +func (m *InjectableEndpoint) InjectOutbound(dest tcpip.Address, packet []byte) *tcpip.Error { endpoint, ok := m.routes[dest] if !ok { return tcpip.ErrNoRoute } - return endpoint.WriteRawPacket(dest, packet) + return endpoint.InjectOutbound(dest, packet) } // Wait implements stack.LinkEndpoint.Wait. diff --git a/pkg/tcpip/link/muxed/injectable_test.go b/pkg/tcpip/link/muxed/injectable_test.go index 3086fec00..63b249837 100644 --- a/pkg/tcpip/link/muxed/injectable_test.go +++ b/pkg/tcpip/link/muxed/injectable_test.go @@ -31,7 +31,7 @@ import ( func TestInjectableEndpointRawDispatch(t *testing.T) { endpoint, sock, dstIP := makeTestInjectableEndpoint(t) - endpoint.WriteRawPacket(dstIP, []byte{0xFA}) + endpoint.InjectOutbound(dstIP, []byte{0xFA}) buf := make([]byte, ipv4.MaxTotalSize) bytesRead, err := sock.Read(buf) @@ -50,8 +50,10 @@ func TestInjectableEndpointDispatch(t *testing.T) { hdr.Prepend(1)[0] = 0xFA packetRoute := stack.Route{RemoteAddress: dstIP} - endpoint.WritePacket(&packetRoute, nil /* gso */, hdr, - buffer.NewViewFromBytes([]byte{0xFB}).ToVectorisedView(), ipv4.ProtocolNumber) + endpoint.WritePacket(&packetRoute, nil /* gso */, ipv4.ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buffer.NewViewFromBytes([]byte{0xFB}).ToVectorisedView(), + }) buf := make([]byte, 6500) bytesRead, err := sock.Read(buf) @@ -68,8 +70,10 @@ func TestInjectableEndpointDispatchHdrOnly(t *testing.T) { hdr := buffer.NewPrependable(1) hdr.Prepend(1)[0] = 0xFA packetRoute := stack.Route{RemoteAddress: dstIP} - endpoint.WritePacket(&packetRoute, nil /* gso */, hdr, - buffer.NewView(0).ToVectorisedView(), ipv4.ProtocolNumber) + endpoint.WritePacket(&packetRoute, nil /* gso */, ipv4.ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buffer.NewView(0).ToVectorisedView(), + }) buf := make([]byte, 6500) bytesRead, err := sock.Read(buf) if err != nil { diff --git a/pkg/tcpip/link/rawfile/BUILD b/pkg/tcpip/link/rawfile/BUILD index 2e8bc772a..d8211e93d 100644 --- a/pkg/tcpip/link/rawfile/BUILD +++ b/pkg/tcpip/link/rawfile/BUILD @@ -13,8 +13,9 @@ go_library( "rawfile_unsafe.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/rawfile", - visibility = [ - "//visibility:public", + visibility = ["//visibility:public"], + deps = [ + "//pkg/tcpip", + "@org_golang_x_sys//unix:go_default_library", ], - deps = ["//pkg/tcpip"], ) diff --git a/pkg/tcpip/link/rawfile/blockingpoll_yield_unsafe.go b/pkg/tcpip/link/rawfile/blockingpoll_yield_unsafe.go index dda3b10a6..0b5a6cf49 100644 --- a/pkg/tcpip/link/rawfile/blockingpoll_yield_unsafe.go +++ b/pkg/tcpip/link/rawfile/blockingpoll_yield_unsafe.go @@ -14,7 +14,7 @@ // +build linux,amd64 linux,arm64 // +build go1.12 -// +build !go1.14 +// +build !go1.15 // Check go:linkname function signatures when updating Go version. diff --git a/pkg/tcpip/link/rawfile/rawfile_unsafe.go b/pkg/tcpip/link/rawfile/rawfile_unsafe.go index 7e286a3a6..44e25d475 100644 --- a/pkg/tcpip/link/rawfile/rawfile_unsafe.go +++ b/pkg/tcpip/link/rawfile/rawfile_unsafe.go @@ -22,6 +22,7 @@ import ( "syscall" "unsafe" + "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/tcpip" ) @@ -101,6 +102,16 @@ func NonBlockingWrite3(fd int, b1, b2, b3 []byte) *tcpip.Error { return nil } +// NonBlockingSendMMsg sends multiple messages on a socket. +func NonBlockingSendMMsg(fd int, msgHdrs []MMsgHdr) (int, *tcpip.Error) { + n, _, e := syscall.RawSyscall6(unix.SYS_SENDMMSG, uintptr(fd), uintptr(unsafe.Pointer(&msgHdrs[0])), uintptr(len(msgHdrs)), syscall.MSG_DONTWAIT, 0, 0) + if e != 0 { + return 0, TranslateErrno(e) + } + + return int(n), nil +} + // PollEvent represents the pollfd structure passed to a poll() system call. type PollEvent struct { FD int32 diff --git a/pkg/tcpip/link/sharedmem/BUILD b/pkg/tcpip/link/sharedmem/BUILD index 0a5ea3dc4..a4f9cdd69 100644 --- a/pkg/tcpip/link/sharedmem/BUILD +++ b/pkg/tcpip/link/sharedmem/BUILD @@ -12,9 +12,7 @@ go_library( "tx.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/sharedmem", - visibility = [ - "//:sandbox", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/log", "//pkg/tcpip", diff --git a/pkg/tcpip/link/sharedmem/pipe/BUILD b/pkg/tcpip/link/sharedmem/pipe/BUILD index 330ed5e94..6b5bc542c 100644 --- a/pkg/tcpip/link/sharedmem/pipe/BUILD +++ b/pkg/tcpip/link/sharedmem/pipe/BUILD @@ -12,7 +12,7 @@ go_library( "tx.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/sharedmem/pipe", - visibility = ["//:sandbox"], + visibility = ["//visibility:public"], ) go_test( diff --git a/pkg/tcpip/link/sharedmem/queue/BUILD b/pkg/tcpip/link/sharedmem/queue/BUILD index de1ce043d..8c9234d54 100644 --- a/pkg/tcpip/link/sharedmem/queue/BUILD +++ b/pkg/tcpip/link/sharedmem/queue/BUILD @@ -10,7 +10,7 @@ go_library( "tx.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/sharedmem/queue", - visibility = ["//:sandbox"], + visibility = ["//visibility:public"], deps = [ "//pkg/log", "//pkg/tcpip/link/sharedmem/pipe", diff --git a/pkg/tcpip/link/sharedmem/sharedmem.go b/pkg/tcpip/link/sharedmem/sharedmem.go index 9e71d4edf..080f9d667 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem.go +++ b/pkg/tcpip/link/sharedmem/sharedmem.go @@ -185,9 +185,10 @@ func (e *endpoint) LinkAddress() tcpip.LinkAddress { // WritePacket writes outbound packets to the file descriptor. If it is not // currently writable, the packet is dropped. -func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { // Add the ethernet header here. - eth := header.Ethernet(hdr.Prepend(header.EthernetMinimumSize)) + eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) + pkt.LinkHeader = buffer.View(eth) ethHdr := &header.EthernetFields{ DstAddr: r.RemoteLinkAddress, Type: protocol, @@ -199,10 +200,30 @@ func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, hdr buffer.Prependa } eth.Encode(ethHdr) - v := payload.ToView() + v := pkt.Data.ToView() // Transmit the packet. e.mu.Lock() - ok := e.tx.transmit(hdr.View(), v) + ok := e.tx.transmit(pkt.Header.View(), v) + e.mu.Unlock() + + if !ok { + return tcpip.ErrWouldBlock + } + + return nil +} + +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (e *endpoint) WritePackets(r *stack.Route, _ *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + panic("not implemented") +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { + v := vv.ToView() + // Transmit the packet. + e.mu.Lock() + ok := e.tx.transmit(v, buffer.View{}) e.mu.Unlock() if !ok { @@ -253,8 +274,11 @@ func (e *endpoint) dispatchLoop(d stack.NetworkDispatcher) { } // Send packet up the stack. - eth := header.Ethernet(b) - d.DeliverNetworkPacket(e, eth.SourceAddress(), eth.DestinationAddress(), eth.Type(), buffer.View(b[header.EthernetMinimumSize:]).ToVectorisedView()) + eth := header.Ethernet(b[:header.EthernetMinimumSize]) + d.DeliverNetworkPacket(e, eth.SourceAddress(), eth.DestinationAddress(), eth.Type(), tcpip.PacketBuffer{ + Data: buffer.View(b[header.EthernetMinimumSize:]).ToVectorisedView(), + LinkHeader: buffer.View(eth), + }) } // Clean state. diff --git a/pkg/tcpip/link/sharedmem/sharedmem_test.go b/pkg/tcpip/link/sharedmem/sharedmem_test.go index 0e9ba0846..89603c48f 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem_test.go +++ b/pkg/tcpip/link/sharedmem/sharedmem_test.go @@ -78,9 +78,10 @@ func (q *queueBuffers) cleanup() { } type packetInfo struct { - addr tcpip.LinkAddress - proto tcpip.NetworkProtocolNumber - vv buffer.VectorisedView + addr tcpip.LinkAddress + proto tcpip.NetworkProtocolNumber + vv buffer.VectorisedView + linkHeader buffer.View } type testContext struct { @@ -130,12 +131,12 @@ func newTestContext(t *testing.T, mtu, bufferSize uint32, addr tcpip.LinkAddress return c } -func (c *testContext) DeliverNetworkPacket(_ stack.LinkEndpoint, remoteLinkAddr, localLinkAddr tcpip.LinkAddress, proto tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { +func (c *testContext) DeliverNetworkPacket(_ stack.LinkEndpoint, remoteLinkAddr, localLinkAddr tcpip.LinkAddress, proto tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { c.mu.Lock() c.packets = append(c.packets, packetInfo{ addr: remoteLinkAddr, proto: proto, - vv: vv.Clone(nil), + vv: pkt.Data.Clone(nil), }) c.mu.Unlock() @@ -272,7 +273,10 @@ func TestSimpleSend(t *testing.T) { randomFill(buf) proto := tcpip.NetworkProtocolNumber(rand.Intn(0x10000)) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), proto); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, proto, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed: %v", err) } @@ -341,7 +345,9 @@ func TestPreserveSrcAddressInSend(t *testing.T) { hdr := buffer.NewPrependable(header.EthernetMinimumSize) proto := tcpip.NetworkProtocolNumber(rand.Intn(0x10000)) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buffer.VectorisedView{}, proto); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, proto, tcpip.PacketBuffer{ + Header: hdr, + }); err != nil { t.Fatalf("WritePacket failed: %v", err) } @@ -395,7 +401,10 @@ func TestFillTxQueue(t *testing.T) { for i := queuePipeSize / 40; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } @@ -410,7 +419,10 @@ func TestFillTxQueue(t *testing.T) { // Next attempt to write must fail. hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != want { + if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != want { t.Fatalf("WritePacket return unexpected result: got %v, want %v", err, want) } } @@ -435,7 +447,10 @@ func TestFillTxQueueAfterBadCompletion(t *testing.T) { // Send two packets so that the id slice has at least two slots. for i := 2; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } } @@ -455,7 +470,10 @@ func TestFillTxQueueAfterBadCompletion(t *testing.T) { ids := make(map[uint64]struct{}) for i := queuePipeSize / 40; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } @@ -470,7 +488,10 @@ func TestFillTxQueueAfterBadCompletion(t *testing.T) { // Next attempt to write must fail. hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != want { + if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != want { t.Fatalf("WritePacket return unexpected result: got %v, want %v", err, want) } } @@ -493,7 +514,10 @@ func TestFillTxMemory(t *testing.T) { ids := make(map[uint64]struct{}) for i := queueDataSize / bufferSize; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } @@ -509,7 +533,10 @@ func TestFillTxMemory(t *testing.T) { // Next attempt to write must fail. hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber) + err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }) if want := tcpip.ErrWouldBlock; err != want { t.Fatalf("WritePacket return unexpected result: got %v, want %v", err, want) } @@ -534,7 +561,10 @@ func TestFillTxMemoryWithMultiBuffer(t *testing.T) { // until there is only one buffer left. for i := queueDataSize/bufferSize - 1; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } @@ -547,7 +577,10 @@ func TestFillTxMemoryWithMultiBuffer(t *testing.T) { { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) uu := buffer.NewView(bufferSize).ToVectorisedView() - if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, hdr, uu, header.IPv4ProtocolNumber); err != want { + if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: uu, + }); err != want { t.Fatalf("WritePacket return unexpected result: got %v, want %v", err, want) } } @@ -555,7 +588,10 @@ func TestFillTxMemoryWithMultiBuffer(t *testing.T) { // Attempt to write the one-buffer packet again. It must succeed. { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, hdr, buf.ToVectorisedView(), header.IPv4ProtocolNumber); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + Header: hdr, + Data: buf.ToVectorisedView(), + }); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } } diff --git a/pkg/tcpip/link/sniffer/BUILD b/pkg/tcpip/link/sniffer/BUILD index 1756114e6..d6ae0368a 100644 --- a/pkg/tcpip/link/sniffer/BUILD +++ b/pkg/tcpip/link/sniffer/BUILD @@ -9,9 +9,7 @@ go_library( "sniffer.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/sniffer", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/log", "//pkg/tcpip", diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go index e401dce44..3392b7edd 100644 --- a/pkg/tcpip/link/sniffer/sniffer.go +++ b/pkg/tcpip/link/sniffer/sniffer.go @@ -49,6 +49,13 @@ var LogPackets uint32 = 1 // LogPacketsToFile must be accessed atomically. var LogPacketsToFile uint32 = 1 +var transportProtocolMinSizes map[tcpip.TransportProtocolNumber]int = map[tcpip.TransportProtocolNumber]int{ + header.ICMPv4ProtocolNumber: header.IPv4MinimumSize, + header.ICMPv6ProtocolNumber: header.IPv6MinimumSize, + header.UDPProtocolNumber: header.UDPMinimumSize, + header.TCPProtocolNumber: header.TCPMinimumSize, +} + type endpoint struct { dispatcher stack.NetworkDispatcher lower stack.LinkEndpoint @@ -116,19 +123,19 @@ func NewWithFile(lower stack.LinkEndpoint, file *os.File, snapLen uint32) (stack // DeliverNetworkPacket implements the stack.NetworkDispatcher interface. It is // called by the link-layer endpoint being wrapped when a packet arrives, and // logs the packet before forwarding to the actual dispatcher. -func (e *endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { +func (e *endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil { - logPacket("recv", protocol, vv.First(), nil) + logPacket("recv", protocol, pkt.Data.First(), nil) } if e.file != nil && atomic.LoadUint32(&LogPacketsToFile) == 1 { - vs := vv.Views() - length := vv.Size() + vs := pkt.Data.Views() + length := pkt.Data.Size() if length > int(e.maxPCAPLen) { length = int(e.maxPCAPLen) } buf := bytes.NewBuffer(make([]byte, 0, pcapPacketHeaderLen+length)) - if err := binary.Write(buf, binary.BigEndian, newPCAPPacketHeader(uint32(length), uint32(vv.Size()))); err != nil { + if err := binary.Write(buf, binary.BigEndian, newPCAPPacketHeader(uint32(length), uint32(pkt.Data.Size()))); err != nil { panic(err) } for _, v := range vs { @@ -147,7 +154,7 @@ func (e *endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local panic(err) } } - e.dispatcher.DeliverNetworkPacket(e, remote, local, protocol, vv) + e.dispatcher.DeliverNetworkPacket(e, remote, local, protocol, pkt) } // Attach implements the stack.LinkEndpoint interface. It saves the dispatcher @@ -193,22 +200,19 @@ func (e *endpoint) GSOMaxSize() uint32 { return 0 } -// WritePacket implements the stack.LinkEndpoint interface. It is called by -// higher-level protocols to write packets; it just logs the packet and forwards -// the request to the lower endpoint. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (e *endpoint) dumpPacket(gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil { - logPacket("send", protocol, hdr.View(), gso) + logPacket("send", protocol, pkt.Header.View(), gso) } if e.file != nil && atomic.LoadUint32(&LogPacketsToFile) == 1 { - hdrBuf := hdr.View() - length := len(hdrBuf) + payload.Size() + hdrBuf := pkt.Header.View() + length := len(hdrBuf) + pkt.Data.Size() if length > int(e.maxPCAPLen) { length = int(e.maxPCAPLen) } buf := bytes.NewBuffer(make([]byte, 0, pcapPacketHeaderLen+length)) - if err := binary.Write(buf, binary.BigEndian, newPCAPPacketHeader(uint32(length), uint32(len(hdrBuf)+payload.Size()))); err != nil { + if err := binary.Write(buf, binary.BigEndian, newPCAPPacketHeader(uint32(length), uint32(len(hdrBuf)+pkt.Data.Size()))); err != nil { panic(err) } if len(hdrBuf) > length { @@ -218,26 +222,75 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen panic(err) } length -= len(hdrBuf) - if length > 0 { - for _, v := range payload.Views() { - if len(v) > length { - v = v[:length] - } - n, err := buf.Write(v) - if err != nil { - panic(err) - } - length -= n - if length == 0 { - break - } - } + logVectorisedView(pkt.Data, length, buf) + if _, err := e.file.Write(buf.Bytes()); err != nil { + panic(err) } + } +} + +// WritePacket implements the stack.LinkEndpoint interface. It is called by +// higher-level protocols to write packets; it just logs the packet and +// forwards the request to the lower endpoint. +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { + e.dumpPacket(gso, protocol, pkt) + return e.lower.WritePacket(r, gso, protocol, pkt) +} + +// WritePackets implements the stack.LinkEndpoint interface. It is called by +// higher-level protocols to write packets; it just logs the packet and +// forwards the request to the lower endpoint. +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + view := pkts[0].Data.ToView() + for _, pkt := range pkts { + e.dumpPacket(gso, protocol, tcpip.PacketBuffer{ + Header: pkt.Header, + Data: view[pkt.DataOffset:][:pkt.DataSize].ToVectorisedView(), + }) + } + return e.lower.WritePackets(r, gso, pkts, protocol) +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { + if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil { + logPacket("send", 0, buffer.View("[raw packet, no header available]"), nil /* gso */) + } + if e.file != nil && atomic.LoadUint32(&LogPacketsToFile) == 1 { + length := vv.Size() + if length > int(e.maxPCAPLen) { + length = int(e.maxPCAPLen) + } + + buf := bytes.NewBuffer(make([]byte, 0, pcapPacketHeaderLen+length)) + if err := binary.Write(buf, binary.BigEndian, newPCAPPacketHeader(uint32(length), uint32(vv.Size()))); err != nil { + panic(err) + } + logVectorisedView(vv, length, buf) if _, err := e.file.Write(buf.Bytes()); err != nil { panic(err) } } - return e.lower.WritePacket(r, gso, hdr, payload, protocol) + return e.lower.WriteRawPacket(vv) +} + +func logVectorisedView(vv buffer.VectorisedView, length int, buf *bytes.Buffer) { + if length <= 0 { + return + } + for _, v := range vv.Views() { + if len(v) > length { + v = v[:length] + } + n, err := buf.Write(v) + if err != nil { + panic(err) + } + length -= n + if length == 0 { + return + } + } } // Wait implements stack.LinkEndpoint.Wait. @@ -287,6 +340,13 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie return } + // We aren't guaranteed to have a transport header - it's possible for + // writes via raw endpoints to contain only network headers. + if minSize, ok := transportProtocolMinSizes[tcpip.TransportProtocolNumber(transProto)]; ok && len(b) < minSize { + log.Infof("%s %v -> %v transport protocol: %d, but no transport header found (possible raw packet)", prefix, src, dst, transProto) + return + } + // Figure out the transport layer info. transName := "unknown" srcPort := uint16(0) diff --git a/pkg/tcpip/link/tun/BUILD b/pkg/tcpip/link/tun/BUILD index 92dce8fac..a71a493fc 100644 --- a/pkg/tcpip/link/tun/BUILD +++ b/pkg/tcpip/link/tun/BUILD @@ -6,7 +6,5 @@ go_library( name = "tun", srcs = ["tun_unsafe.go"], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/tun", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], ) diff --git a/pkg/tcpip/link/waitable/BUILD b/pkg/tcpip/link/waitable/BUILD index 0746dc8ec..134837943 100644 --- a/pkg/tcpip/link/waitable/BUILD +++ b/pkg/tcpip/link/waitable/BUILD @@ -9,9 +9,7 @@ go_library( "waitable.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/link/waitable", - visibility = [ - "//visibility:public", - ], + visibility = ["//visibility:public"], deps = [ "//pkg/gate", "//pkg/tcpip", diff --git a/pkg/tcpip/link/waitable/waitable.go b/pkg/tcpip/link/waitable/waitable.go index 5a1791cb5..a8de38979 100644 --- a/pkg/tcpip/link/waitable/waitable.go +++ b/pkg/tcpip/link/waitable/waitable.go @@ -50,12 +50,12 @@ func New(lower stack.LinkEndpoint) *Endpoint { // It is called by the link-layer endpoint being wrapped when a packet arrives, // and only forwards to the actual dispatcher if Wait or WaitDispatch haven't // been called. -func (e *Endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { +func (e *Endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { if !e.dispatchGate.Enter() { return } - e.dispatcher.DeliverNetworkPacket(e, remote, local, protocol, vv) + e.dispatcher.DeliverNetworkPacket(e, remote, local, protocol, pkt) e.dispatchGate.Leave() } @@ -99,12 +99,36 @@ func (e *Endpoint) LinkAddress() tcpip.LinkAddress { // WritePacket implements stack.LinkEndpoint.WritePacket. It is called by // higher-level protocols to write packets. It only forwards packets to the // lower endpoint if Wait or WaitWrite haven't been called. -func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { if !e.writeGate.Enter() { return nil } - err := e.lower.WritePacket(r, gso, hdr, payload, protocol) + err := e.lower.WritePacket(r, gso, protocol, pkt) + e.writeGate.Leave() + return err +} + +// WritePackets implements stack.LinkEndpoint.WritePackets. It is called by +// higher-level protocols to write packets. It only forwards packets to the +// lower endpoint if Wait or WaitWrite haven't been called. +func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + if !e.writeGate.Enter() { + return len(pkts), nil + } + + n, err := e.lower.WritePackets(r, gso, pkts, protocol) + e.writeGate.Leave() + return n, err +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *Endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { + if !e.writeGate.Enter() { + return nil + } + + err := e.lower.WriteRawPacket(vv) e.writeGate.Leave() return err } diff --git a/pkg/tcpip/link/waitable/waitable_test.go b/pkg/tcpip/link/waitable/waitable_test.go index ae23c96b7..31b11a27a 100644 --- a/pkg/tcpip/link/waitable/waitable_test.go +++ b/pkg/tcpip/link/waitable/waitable_test.go @@ -35,7 +35,7 @@ type countedEndpoint struct { dispatcher stack.NetworkDispatcher } -func (e *countedEndpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { +func (e *countedEndpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { e.dispatchCount++ } @@ -65,7 +65,18 @@ func (e *countedEndpoint) LinkAddress() tcpip.LinkAddress { return e.linkAddr } -func (e *countedEndpoint) WritePacket(r *stack.Route, _ *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (e *countedEndpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { + e.writeCount++ + return nil +} + +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (e *countedEndpoint) WritePackets(r *stack.Route, _ *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + e.writeCount += len(pkts) + return len(pkts), nil +} + +func (e *countedEndpoint) WriteRawPacket(buffer.VectorisedView) *tcpip.Error { e.writeCount++ return nil } @@ -78,21 +89,21 @@ func TestWaitWrite(t *testing.T) { wep := New(ep) // Write and check that it goes through. - wep.WritePacket(nil, nil /* gso */, buffer.Prependable{}, buffer.VectorisedView{}, 0) + wep.WritePacket(nil, nil /* gso */, 0, tcpip.PacketBuffer{}) if want := 1; ep.writeCount != want { t.Fatalf("Unexpected writeCount: got=%v, want=%v", ep.writeCount, want) } // Wait on dispatches, then try to write. It must go through. wep.WaitDispatch() - wep.WritePacket(nil, nil /* gso */, buffer.Prependable{}, buffer.VectorisedView{}, 0) + wep.WritePacket(nil, nil /* gso */, 0, tcpip.PacketBuffer{}) if want := 2; ep.writeCount != want { t.Fatalf("Unexpected writeCount: got=%v, want=%v", ep.writeCount, want) } // Wait on writes, then try to write. It must not go through. wep.WaitWrite() - wep.WritePacket(nil, nil /* gso */, buffer.Prependable{}, buffer.VectorisedView{}, 0) + wep.WritePacket(nil, nil /* gso */, 0, tcpip.PacketBuffer{}) if want := 2; ep.writeCount != want { t.Fatalf("Unexpected writeCount: got=%v, want=%v", ep.writeCount, want) } @@ -109,21 +120,21 @@ func TestWaitDispatch(t *testing.T) { } // Dispatch and check that it goes through. - ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, buffer.VectorisedView{}) + ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, tcpip.PacketBuffer{}) if want := 1; ep.dispatchCount != want { t.Fatalf("Unexpected dispatchCount: got=%v, want=%v", ep.dispatchCount, want) } // Wait on writes, then try to dispatch. It must go through. wep.WaitWrite() - ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, buffer.VectorisedView{}) + ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, tcpip.PacketBuffer{}) if want := 2; ep.dispatchCount != want { t.Fatalf("Unexpected dispatchCount: got=%v, want=%v", ep.dispatchCount, want) } // Wait on dispatches, then try to dispatch. It must not go through. wep.WaitDispatch() - ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, buffer.VectorisedView{}) + ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, tcpip.PacketBuffer{}) if want := 2; ep.dispatchCount != want { t.Fatalf("Unexpected dispatchCount: got=%v, want=%v", ep.dispatchCount, want) } |