diff options
Diffstat (limited to 'pkg/tcpip')
-rwxr-xr-x | pkg/tcpip/link/channel/channel.go | 12 | ||||
-rw-r--r-- | pkg/tcpip/link/fdbased/endpoint.go | 25 | ||||
-rw-r--r-- | pkg/tcpip/link/loopback/loopback.go | 2 | ||||
-rw-r--r-- | pkg/tcpip/link/sniffer/sniffer.go | 12 | ||||
-rw-r--r-- | pkg/tcpip/network/arp/arp.go | 2 | ||||
-rw-r--r-- | pkg/tcpip/network/ipv4/ipv4.go | 10 | ||||
-rw-r--r-- | pkg/tcpip/network/ipv6/ipv6.go | 12 | ||||
-rwxr-xr-x | pkg/tcpip/packet_buffer.go | 8 | ||||
-rw-r--r-- | pkg/tcpip/stack/registration.go | 8 | ||||
-rw-r--r-- | pkg/tcpip/stack/route.go | 29 | ||||
-rwxr-xr-x | pkg/tcpip/tcpip_state_autogen.go | 4 | ||||
-rw-r--r-- | pkg/tcpip/transport/tcp/connect.go | 43 |
12 files changed, 88 insertions, 79 deletions
diff --git a/pkg/tcpip/link/channel/channel.go b/pkg/tcpip/link/channel/channel.go index 9fe8e9f9d..70188551f 100755 --- a/pkg/tcpip/link/channel/channel.go +++ b/pkg/tcpip/link/channel/channel.go @@ -133,16 +133,16 @@ func (e *Endpoint) WritePacket(_ *stack.Route, gso *stack.GSO, protocol tcpip.Ne } // WritePackets stores outbound packets into the channel. -func (e *Endpoint) WritePackets(_ *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { - payloadView := payload.ToView() +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 _, hdr := range hdrs { - off := hdr.Off - size := hdr.Size + for _, pkt := range pkts { + off := pkt.DataOffset + size := pkt.DataSize p := PacketInfo{ Pkt: tcpip.PacketBuffer{ - Header: hdr.Hdr, + Header: pkt.Header, Data: buffer.NewViewFromBytes(payloadView[off : off+size]).ToVectorisedView(), }, Proto: protocol, diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index 98109c5dc..fa8a703d9 100644 --- a/pkg/tcpip/link/fdbased/endpoint.go +++ b/pkg/tcpip/link/fdbased/endpoint.go @@ -440,7 +440,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne // 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, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +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 @@ -463,9 +463,9 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.Pac iovLen++ } - n := len(hdrs) + n := len(pkts) - views := payload.Views() + views := pkts[0].Data.Views() /* * Each bondary in views can add one more iovec. * @@ -483,14 +483,20 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.Pac viewOff := 0 off := 0 nextOff := 0 - for i := range hdrs { + 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 := hdrs[i].Size - hdr := &hdrs[i].Hdr + packetSize := pkts[i].DataSize + hdr := &pkts[i].Header - off = hdrs[i].Off + off = pkts[i].DataOffset if off != nextOff { // We stop in a different point last time. size := packetSize @@ -555,6 +561,11 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.Pac 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()) diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index 563a67188..499cc608f 100644 --- a/pkg/tcpip/link/loopback/loopback.go +++ b/pkg/tcpip/link/loopback/loopback.go @@ -92,7 +92,7 @@ func (e *endpoint) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.Netw } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (e *endpoint) WritePackets(_ *stack.Route, _ *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, []tcpip.PacketBuffer, tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { panic("not implemented") } diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go index 122680e10..767f14303 100644 --- a/pkg/tcpip/link/sniffer/sniffer.go +++ b/pkg/tcpip/link/sniffer/sniffer.go @@ -233,15 +233,15 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne // 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, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { - view := payload.ToView() - for _, d := range hdrs { +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: d.Hdr, - Data: view[d.Off:][:d.Size].ToVectorisedView(), + Header: pkt.Header, + Data: view[pkt.DataOffset:][:pkt.DataSize].ToVectorisedView(), }) } - return e.lower.WritePackets(r, gso, hdrs, payload, protocol) + return e.lower.WritePackets(r, gso, pkts, protocol) } // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index 30aec9ba7..da8482509 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -84,7 +84,7 @@ func (e *endpoint) WritePacket(*stack.Route, *stack.GSO, stack.NetworkHeaderPara } // WritePackets implements stack.NetworkEndpoint.WritePackets. -func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, []stack.PacketDescriptor, buffer.VectorisedView, stack.NetworkHeaderParams, stack.PacketLooping) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, []tcpip.PacketBuffer, stack.NetworkHeaderParams, stack.PacketLooping) (int, *tcpip.Error) { return 0, tcpip.ErrNotSupported } diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index 040329a74..7059600f5 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -268,18 +268,18 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw } // WritePackets implements stack.NetworkEndpoint.WritePackets. -func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) { if loop&stack.PacketLoop != 0 { panic("multiple packets in local loop") } if loop&stack.PacketOut == 0 { - return len(hdrs), nil + return len(pkts), nil } - for i := range hdrs { - e.addIPHeader(r, &hdrs[i].Hdr, hdrs[i].Size, params) + for i := range pkts { + e.addIPHeader(r, &pkts[i].Header, pkts[i].DataSize, params) } - n, err := e.linkEP.WritePackets(r, gso, hdrs, payload, ProtocolNumber) + n, err := e.linkEP.WritePackets(r, gso, pkts, ProtocolNumber) r.Stats().IP.PacketsSent.IncrementBy(uint64(n)) return n, err } diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index 8d1578ed9..c9087ffa7 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -137,21 +137,21 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) { if loop&stack.PacketLoop != 0 { panic("not implemented") } if loop&stack.PacketOut == 0 { - return len(hdrs), nil + return len(pkts), nil } - for i := range hdrs { - hdr := &hdrs[i].Hdr - size := hdrs[i].Size + for i := range pkts { + hdr := &pkts[i].Header + size := pkts[i].DataSize e.addIPHeader(r, hdr, size, params) } - n, err := e.linkEP.WritePackets(r, gso, hdrs, payload, ProtocolNumber) + n, err := e.linkEP.WritePackets(r, gso, pkts, ProtocolNumber) r.Stats().IP.PacketsSent.IncrementBy(uint64(n)) return n, err } diff --git a/pkg/tcpip/packet_buffer.go b/pkg/tcpip/packet_buffer.go index 695f7b188..ab24372e7 100755 --- a/pkg/tcpip/packet_buffer.go +++ b/pkg/tcpip/packet_buffer.go @@ -31,6 +31,14 @@ type PacketBuffer struct { // or otherwise modified. Data buffer.VectorisedView + // DataOffset is used for GSO output. It is the offset into the Data + // field where the payload of this packet starts. + DataOffset int + + // DataSize is used for GSO output. It is the size of this packet's + // payload. + DataSize int + // Header holds the headers of outbound packets. As a packet is passed // down the stack, each layer adds to Header. Header buffer.Prependable diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go index 7fd4e4a65..61fd46d66 100644 --- a/pkg/tcpip/stack/registration.go +++ b/pkg/tcpip/stack/registration.go @@ -237,8 +237,8 @@ type NetworkEndpoint interface { WritePacket(r *Route, gso *GSO, params NetworkHeaderParams, loop PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error // WritePackets writes packets to the given destination address and - // protocol. - WritePackets(r *Route, gso *GSO, hdrs []PacketDescriptor, payload buffer.VectorisedView, params NetworkHeaderParams, loop PacketLooping) (int, *tcpip.Error) + // protocol. pkts must not be zero length. + WritePackets(r *Route, gso *GSO, pkts []tcpip.PacketBuffer, params NetworkHeaderParams, loop PacketLooping) (int, *tcpip.Error) // WriteHeaderIncludedPacket writes a packet that includes a network // header to the given destination address. @@ -373,12 +373,12 @@ type LinkEndpoint interface { WritePacket(r *Route, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error // WritePackets writes packets with the given protocol through the - // given route. + // given route. pkts must not be zero length. // // Right now, WritePackets is used only when the software segmentation // offload is enabled. If it will be used for something else, it may // require to change syscall filters. - WritePackets(r *Route, gso *GSO, hdrs []PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) + WritePackets(r *Route, gso *GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) // WriteRawPacket writes a packet directly to the link. The packet // should already have an ethernet header. diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go index 617f5a57c..34307ae07 100644 --- a/pkg/tcpip/stack/route.go +++ b/pkg/tcpip/stack/route.go @@ -17,7 +17,6 @@ package stack import ( "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" ) @@ -169,39 +168,21 @@ func (r *Route) WritePacket(gso *GSO, params NetworkHeaderParams, pkt tcpip.Pack return err } -// PacketDescriptor is a packet descriptor which contains a packet header and -// offset and size of packet data in a payload view. -type PacketDescriptor struct { - Hdr buffer.Prependable - Off int - Size int -} - -// NewPacketDescriptors allocates a set of packet descriptors. -func NewPacketDescriptors(n int, hdrSize int) []PacketDescriptor { - buf := make([]byte, n*hdrSize) - hdrs := make([]PacketDescriptor, n) - for i := range hdrs { - hdrs[i].Hdr = buffer.NewEmptyPrependableFromView(buf[i*hdrSize:][:hdrSize]) - } - return hdrs -} - // WritePackets writes the set of packets through the given route. -func (r *Route) WritePackets(gso *GSO, hdrs []PacketDescriptor, payload buffer.VectorisedView, params NetworkHeaderParams) (int, *tcpip.Error) { +func (r *Route) WritePackets(gso *GSO, pkts []tcpip.PacketBuffer, params NetworkHeaderParams) (int, *tcpip.Error) { if !r.ref.isValidForOutgoing() { return 0, tcpip.ErrInvalidEndpointState } - n, err := r.ref.ep.WritePackets(r, gso, hdrs, payload, params, r.Loop) + n, err := r.ref.ep.WritePackets(r, gso, pkts, params, r.Loop) if err != nil { - r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(len(hdrs) - n)) + r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(len(pkts) - n)) } r.ref.nic.stats.Tx.Packets.IncrementBy(uint64(n)) payloadSize := 0 for i := 0; i < n; i++ { - r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(hdrs[i].Hdr.UsedLength())) - payloadSize += hdrs[i].Size + r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(pkts[i].Header.UsedLength())) + payloadSize += pkts[i].DataSize } r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(payloadSize)) return n, err diff --git a/pkg/tcpip/tcpip_state_autogen.go b/pkg/tcpip/tcpip_state_autogen.go index 11eb1703a..89b2dab67 100755 --- a/pkg/tcpip/tcpip_state_autogen.go +++ b/pkg/tcpip/tcpip_state_autogen.go @@ -9,6 +9,8 @@ import ( func (x *PacketBuffer) save(m state.Map) { x.beforeSave() m.Save("Data", &x.Data) + m.Save("DataOffset", &x.DataOffset) + m.Save("DataSize", &x.DataSize) m.Save("Header", &x.Header) m.Save("LinkHeader", &x.LinkHeader) m.Save("NetworkHeader", &x.NetworkHeader) @@ -18,6 +20,8 @@ func (x *PacketBuffer) save(m state.Map) { func (x *PacketBuffer) afterLoad() {} func (x *PacketBuffer) load(m state.Map) { m.Load("Data", &x.Data) + m.Load("DataOffset", &x.DataOffset) + m.Load("DataSize", &x.DataSize) m.Load("Header", &x.Header) m.Load("LinkHeader", &x.LinkHeader) m.Load("NetworkHeader", &x.NetworkHeader) diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 75b7c0828..00c0c9a92 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -631,11 +631,11 @@ func (e *endpoint) sendTCP(r *stack.Route, id stack.TransportEndpointID, data bu return nil } -func buildTCPHdr(r *stack.Route, id stack.TransportEndpointID, d *stack.PacketDescriptor, data buffer.VectorisedView, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) { +func buildTCPHdr(r *stack.Route, id stack.TransportEndpointID, pkt *tcpip.PacketBuffer, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) { optLen := len(opts) - hdr := &d.Hdr - packetSize := d.Size - off := d.Off + hdr := &pkt.Header + packetSize := pkt.DataSize + off := pkt.DataOffset // Initialize the header. tcp := header.TCP(hdr.Prepend(header.TCPMinimumSize + optLen)) tcp.Encode(&header.TCPFields{ @@ -659,7 +659,7 @@ func buildTCPHdr(r *stack.Route, id stack.TransportEndpointID, d *stack.PacketDe // header and data and get the right sum of the TCP packet. tcp.SetChecksum(xsum) } else if r.Capabilities()&stack.CapabilityTXChecksumOffload == 0 { - xsum = header.ChecksumVVWithOffset(data, xsum, off, packetSize) + xsum = header.ChecksumVVWithOffset(pkt.Data, xsum, off, packetSize) tcp.SetChecksum(^tcp.CalculateChecksum(xsum)) } @@ -674,7 +674,13 @@ func sendTCPBatch(r *stack.Route, id stack.TransportEndpointID, data buffer.Vect mss := int(gso.MSS) n := (data.Size() + mss - 1) / mss - hdrs := stack.NewPacketDescriptors(n, header.TCPMinimumSize+int(r.MaxHeaderLength())+optLen) + // Allocate one big slice for all the headers. + hdrSize := header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen + buf := make([]byte, n*hdrSize) + pkts := make([]tcpip.PacketBuffer, n) + for i := range pkts { + pkts[i].Header = buffer.NewEmptyPrependableFromView(buf[i*hdrSize:][:hdrSize]) + } size := data.Size() off := 0 @@ -684,16 +690,17 @@ func sendTCPBatch(r *stack.Route, id stack.TransportEndpointID, data buffer.Vect packetSize = size } size -= packetSize - hdrs[i].Off = off - hdrs[i].Size = packetSize - buildTCPHdr(r, id, &hdrs[i], data, flags, seq, ack, rcvWnd, opts, gso) + pkts[i].DataOffset = off + pkts[i].DataSize = packetSize + pkts[i].Data = data + buildTCPHdr(r, id, &pkts[i], flags, seq, ack, rcvWnd, opts, gso) off += packetSize seq = seq.Add(seqnum.Size(packetSize)) } if ttl == 0 { ttl = r.DefaultTTL() } - sent, err := r.WritePackets(gso, hdrs, data, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}) + sent, err := r.WritePackets(gso, pkts, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}) if err != nil { r.Stats().TCP.SegmentSendErrors.IncrementBy(uint64(n - sent)) } @@ -713,20 +720,18 @@ func sendTCP(r *stack.Route, id stack.TransportEndpointID, data buffer.Vectorise return sendTCPBatch(r, id, data, ttl, tos, flags, seq, ack, rcvWnd, opts, gso) } - d := &stack.PacketDescriptor{ - Hdr: buffer.NewPrependable(header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen), - Off: 0, - Size: data.Size(), + pkt := tcpip.PacketBuffer{ + Header: buffer.NewPrependable(header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen), + DataOffset: 0, + DataSize: data.Size(), + Data: data, } - buildTCPHdr(r, id, d, data, flags, seq, ack, rcvWnd, opts, gso) + buildTCPHdr(r, id, &pkt, flags, seq, ack, rcvWnd, opts, gso) if ttl == 0 { ttl = r.DefaultTTL() } - if err := r.WritePacket(gso, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}, tcpip.PacketBuffer{ - Header: d.Hdr, - Data: data, - }); err != nil { + if err := r.WritePacket(gso, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}, pkt); err != nil { r.Stats().TCP.SegmentSendErrors.Increment() return err } |