diff options
Diffstat (limited to 'pkg/tcpip')
-rw-r--r-- | pkg/tcpip/buffer/prependable.go | 5 | ||||
-rw-r--r-- | pkg/tcpip/header/checksum.go | 50 | ||||
-rwxr-xr-x | pkg/tcpip/link/channel/channel.go | 29 | ||||
-rw-r--r-- | pkg/tcpip/link/fdbased/endpoint.go | 128 | ||||
-rw-r--r-- | pkg/tcpip/link/fdbased/packet_dispatchers.go | 12 | ||||
-rw-r--r-- | pkg/tcpip/link/loopback/loopback.go | 5 | ||||
-rw-r--r-- | pkg/tcpip/link/rawfile/rawfile_unsafe.go | 11 | ||||
-rw-r--r-- | pkg/tcpip/link/sniffer/sniffer.go | 23 | ||||
-rw-r--r-- | pkg/tcpip/network/arp/arp.go | 5 | ||||
-rw-r--r-- | pkg/tcpip/network/ipv4/ipv4.go | 27 | ||||
-rw-r--r-- | pkg/tcpip/network/ipv6/ipv6.go | 30 | ||||
-rw-r--r-- | pkg/tcpip/stack/registration.go | 28 | ||||
-rw-r--r-- | pkg/tcpip/stack/route.go | 48 | ||||
-rw-r--r-- | pkg/tcpip/transport/tcp/connect.go | 77 | ||||
-rw-r--r-- | pkg/tcpip/transport/tcp/endpoint.go | 18 |
15 files changed, 442 insertions, 54 deletions
diff --git a/pkg/tcpip/buffer/prependable.go b/pkg/tcpip/buffer/prependable.go index 4287464f3..48a2a2713 100644 --- a/pkg/tcpip/buffer/prependable.go +++ b/pkg/tcpip/buffer/prependable.go @@ -41,6 +41,11 @@ func NewPrependableFromView(v View) Prependable { return Prependable{buf: v, usedIdx: 0} } +// NewEmptyPrependableFromView creates a new prependable buffer from a View. +func NewEmptyPrependableFromView(v View) Prependable { + return Prependable{buf: v, usedIdx: len(v)} +} + // View returns a View of the backing buffer that contains all prepended // data so far. func (p Prependable) View() View { diff --git a/pkg/tcpip/header/checksum.go b/pkg/tcpip/header/checksum.go index 39a4d69be..9749c7f4d 100644 --- a/pkg/tcpip/header/checksum.go +++ b/pkg/tcpip/header/checksum.go @@ -23,11 +23,17 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/buffer" ) -func calculateChecksum(buf []byte, initial uint32) uint16 { +func calculateChecksum(buf []byte, odd bool, initial uint32) (uint16, bool) { v := initial + if odd { + v += uint32(buf[0]) + buf = buf[1:] + } + l := len(buf) - if l&1 != 0 { + odd = l&1 != 0 + if odd { l-- v += uint32(buf[l]) << 8 } @@ -36,7 +42,7 @@ func calculateChecksum(buf []byte, initial uint32) uint16 { v += (uint32(buf[i]) << 8) + uint32(buf[i+1]) } - return ChecksumCombine(uint16(v), uint16(v>>16)) + return ChecksumCombine(uint16(v), uint16(v>>16)), odd } // Checksum calculates the checksum (as defined in RFC 1071) of the bytes in the @@ -44,7 +50,8 @@ func calculateChecksum(buf []byte, initial uint32) uint16 { // // The initial checksum must have been computed on an even number of bytes. func Checksum(buf []byte, initial uint16) uint16 { - return calculateChecksum(buf, uint32(initial)) + s, _ := calculateChecksum(buf, false, uint32(initial)) + return s } // ChecksumVV calculates the checksum (as defined in RFC 1071) of the bytes in @@ -52,19 +59,40 @@ func Checksum(buf []byte, initial uint16) uint16 { // // The initial checksum must have been computed on an even number of bytes. func ChecksumVV(vv buffer.VectorisedView, initial uint16) uint16 { - var odd bool + return ChecksumVVWithOffset(vv, initial, 0, vv.Size()) +} + +// ChecksumVVWithOffset calculates the checksum (as defined in RFC 1071) of the +// bytes in the given VectorizedView. +// +// The initial checksum must have been computed on an even number of bytes. +func ChecksumVVWithOffset(vv buffer.VectorisedView, initial uint16, off int, size int) uint16 { + odd := false sum := initial for _, v := range vv.Views() { if len(v) == 0 { continue } - s := uint32(sum) - if odd { - s += uint32(v[0]) - v = v[1:] + + if off >= len(v) { + off -= len(v) + continue + } + v = v[off:] + + l := len(v) + if l > size { + l = size + } + v = v[:l] + + sum, odd = calculateChecksum(v, odd, uint32(sum)) + + size -= len(v) + if size == 0 { + break } - odd = len(v)&1 != 0 - sum = calculateChecksum(v, s) + off = 0 } return sum } diff --git a/pkg/tcpip/link/channel/channel.go b/pkg/tcpip/link/channel/channel.go index 95c6f0828..14f197a77 100755 --- a/pkg/tcpip/link/channel/channel.go +++ b/pkg/tcpip/link/channel/channel.go @@ -96,7 +96,7 @@ 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 } @@ -134,6 +134,33 @@ func (e *Endpoint) WritePacket(_ *stack.Route, gso *stack.GSO, hdr buffer.Prepen return nil } +// 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() + n := 0 +packetLoop: + for i := range hdrs { + hdr := &hdrs[i].Hdr + off := hdrs[i].Off + size := hdrs[i].Size + p := PacketInfo{ + Header: hdr.View(), + Proto: protocol, + Payload: buffer.NewViewFromBytes(payloadView[off : off+size]), + GSO: gso, + } + + select { + case e.C <- p: + n++ + default: + break packetLoop + } + } + + return n, nil +} + // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *Endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { p := PacketInfo{ diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index 255220d1a..ae4858529 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 } } @@ -397,7 +404,7 @@ 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 { @@ -430,6 +437,123 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen return rawfile.NonBlockingWrite3(e.fds[0], hdr.View(), payload.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, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, 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++ + } + + n := len(hdrs) + + views := payload.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 hdrs { + prevIovecIdx := iovecIdx + mmsgHdr := &mmsgHdrs[i] + mmsgHdr.Msg.Iov = &iovec[iovecIdx] + packetSize := hdrs[i].Size + hdr := &hdrs[i].Hdr + + off = hdrs[i].Off + 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 +} + // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { return rawfile.NonBlockingWrite(e.fds[0], packet.ToView()) diff --git a/pkg/tcpip/link/fdbased/packet_dispatchers.go b/pkg/tcpip/link/fdbased/packet_dispatchers.go index 9fb760891..12168a1dc 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 @@ -195,7 +195,7 @@ 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++ } @@ -226,7 +226,7 @@ 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. @@ -262,7 +262,7 @@ 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 { diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index 5f5bacf27..a3b48fa73 100644 --- a/pkg/tcpip/link/loopback/loopback.go +++ b/pkg/tcpip/link/loopback/loopback.go @@ -90,6 +90,11 @@ func (e *endpoint) WritePacket(_ *stack.Route, _ *stack.GSO, hdr buffer.Prependa return nil } +// 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) { + panic("not implemented") +} + // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { // Reject the packet if it's shorter than an ethernet header. 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/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go index de9ff98cb..39757ea2a 100644 --- a/pkg/tcpip/link/sniffer/sniffer.go +++ b/pkg/tcpip/link/sniffer/sniffer.go @@ -193,10 +193,7 @@ 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, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) { if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil { logPacket("send", protocol, hdr.View(), gso) } @@ -223,9 +220,27 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen 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, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { + e.dumpPacket(gso, hdr, payload, protocol) return e.lower.WritePacket(r, gso, hdr, payload, protocol) } +// 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 { + e.dumpPacket(gso, d.Hdr, buffer.NewVectorisedView(d.Size, []buffer.View{view[d.Off:][:d.Size]}), protocol) + } + return e.lower.WritePackets(r, gso, hdrs, payload, protocol) +} + // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil { diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index 6b1e854dc..46178459e 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -83,6 +83,11 @@ func (e *endpoint) WritePacket(*stack.Route, *stack.GSO, buffer.Prependable, buf return tcpip.ErrNotSupported } +// WritePackets implements stack.NetworkEndpoint.WritePackets. +func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, []stack.PacketDescriptor, buffer.VectorisedView, stack.NetworkHeaderParams, stack.PacketLooping) (int, *tcpip.Error) { + return 0, tcpip.ErrNotSupported +} + func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error { return tcpip.ErrNotSupported } diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index 5cd895ff0..1339f8474 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -198,10 +198,9 @@ func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, hdr buff return nil } -// WritePacket writes a packet to the given destination address and protocol. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) *tcpip.Error { +func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadSize int, params stack.NetworkHeaderParams) { ip := header.IPv4(hdr.Prepend(header.IPv4MinimumSize)) - length := uint16(hdr.UsedLength() + payload.Size()) + length := uint16(hdr.UsedLength() + payloadSize) id := uint32(0) if length > header.IPv4MaximumHeaderSize+8 { // Packets of 68 bytes or less are required by RFC 791 to not be @@ -219,6 +218,11 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen DstAddr: r.RemoteAddress, }) ip.SetChecksum(^ip.CalculateChecksum()) +} + +// WritePacket writes a packet to the given destination address and protocol. +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) *tcpip.Error { + e.addIPHeader(r, &hdr, payload.Size(), params) if loop&stack.PacketLoop != 0 { views := make([]buffer.View, 1, 1+len(payload.Views())) @@ -242,6 +246,23 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen return nil } +// 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) { + if loop&stack.PacketLoop != 0 { + panic("multiple packets in local loop") + } + if loop&stack.PacketOut == 0 { + return len(hdrs), nil + } + + for i := range hdrs { + e.addIPHeader(r, &hdrs[i].Hdr, hdrs[i].Size, params) + } + n, err := e.linkEP.WritePackets(r, gso, hdrs, payload, ProtocolNumber) + r.Stats().IP.PacketsSent.IncrementBy(uint64(n)) + return n, err +} + // WriteHeaderIncludedPacket writes a packet already containing a network // header through the given route. func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error { diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index cd1e34085..5898f8f9e 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -97,9 +97,8 @@ func (e *endpoint) GSOMaxSize() uint32 { return 0 } -// WritePacket writes a packet to the given destination address and protocol. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) *tcpip.Error { - length := uint16(hdr.UsedLength() + payload.Size()) +func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadSize int, params stack.NetworkHeaderParams) { + length := uint16(hdr.UsedLength() + payloadSize) ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) ip.Encode(&header.IPv6Fields{ PayloadLength: length, @@ -109,6 +108,11 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen SrcAddr: r.LocalAddress, DstAddr: r.RemoteAddress, }) +} + +// WritePacket writes a packet to the given destination address and protocol. +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) *tcpip.Error { + e.addIPHeader(r, &hdr, payload.Size(), params) if loop&stack.PacketLoop != 0 { views := make([]buffer.View, 1, 1+len(payload.Views())) @@ -127,6 +131,26 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen return e.linkEP.WritePacket(r, gso, hdr, payload, ProtocolNumber) } +// 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) { + if loop&stack.PacketLoop != 0 { + panic("not implemented") + } + if loop&stack.PacketOut == 0 { + return len(hdrs), nil + } + + for i := range hdrs { + hdr := &hdrs[i].Hdr + size := hdrs[i].Size + e.addIPHeader(r, hdr, size, params) + } + + n, err := e.linkEP.WritePackets(r, gso, hdrs, payload, ProtocolNumber) + r.Stats().IP.PacketsSent.IncrementBy(uint64(n)) + return n, err +} + // WriteHeaderIncludedPacker implements stack.NetworkEndpoint. It is not yet // supported by IPv6. func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error { diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go index f655d011f..0869fb084 100644 --- a/pkg/tcpip/stack/registration.go +++ b/pkg/tcpip/stack/registration.go @@ -201,6 +201,10 @@ type NetworkEndpoint interface { // protocol. WritePacket(r *Route, gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params NetworkHeaderParams, loop PacketLooping) *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) + // WriteHeaderIncludedPacket writes a packet that includes a network // header to the given destination address. WriteHeaderIncludedPacket(r *Route, payload buffer.VectorisedView, loop PacketLooping) *tcpip.Error @@ -283,7 +287,11 @@ const ( CapabilitySaveRestore CapabilityDisconnectOk CapabilityLoopback - CapabilityGSO + CapabilityHardwareGSO + + // CapabilitySoftwareGSO indicates the link endpoint supports of sending + // multiple packets using a single call (LinkEndpoint.WritePackets). + CapabilitySoftwareGSO ) // LinkEndpoint is the interface implemented by data link layer protocols (e.g., @@ -318,6 +326,14 @@ type LinkEndpoint interface { // r.LocalLinkAddress if it is provided. WritePacket(r *Route, gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error + // WritePackets writes packets with the given protocol through the + // given route. + // + // 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) + // WriteRawPacket writes a packet directly to the link. The packet // should already have an ethernet header. WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error @@ -421,8 +437,14 @@ type GSOType int // Types of gso segments. const ( GSONone GSOType = iota + + // Hardware GSO types: GSOTCPv4 GSOTCPv6 + + // GSOSW is used for software GSO segments which have to be sent by + // endpoint.WritePackets. + GSOSW ) // GSO contains generic segmentation offload properties. @@ -450,3 +472,7 @@ type GSOEndpoint interface { // GSOMaxSize returns the maximum GSO packet size. GSOMaxSize() uint32 } + +// SoftwareGSOMaxSize is a maximum allowed size of a software GSO segment. +// This isn't a hard limit, because it is never set into packet headers. +const SoftwareGSOMaxSize = (1 << 16) diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go index e72373964..1a0a51b57 100644 --- a/pkg/tcpip/stack/route.go +++ b/pkg/tcpip/stack/route.go @@ -47,8 +47,8 @@ type Route struct { // starts. ref *referencedNetworkEndpoint - // loop controls where WritePacket should send packets. - loop PacketLooping + // Loop controls where WritePacket should send packets. + Loop PacketLooping } // makeRoute initializes a new route. It takes ownership of the provided @@ -69,7 +69,7 @@ func makeRoute(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr tcpip LocalLinkAddress: localLinkAddr, RemoteAddress: remoteAddr, ref: ref, - loop: loop, + Loop: loop, } } @@ -159,7 +159,7 @@ func (r *Route) WritePacket(gso *GSO, hdr buffer.Prependable, payload buffer.Vec return tcpip.ErrInvalidEndpointState } - err := r.ref.ep.WritePacket(r, gso, hdr, payload, params, r.loop) + err := r.ref.ep.WritePacket(r, gso, hdr, payload, params, r.Loop) if err != nil { r.Stats().IP.OutgoingPacketErrors.Increment() } else { @@ -169,6 +169,44 @@ func (r *Route) WritePacket(gso *GSO, hdr buffer.Prependable, payload buffer.Vec 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) { + if !r.ref.isValidForOutgoing() { + return 0, tcpip.ErrInvalidEndpointState + } + + n, err := r.ref.ep.WritePackets(r, gso, hdrs, payload, params, r.Loop) + if err != nil { + r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(len(hdrs) - 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(payloadSize)) + return n, err +} + // WriteHeaderIncludedPacket writes a packet already containing a network // header through the given route. func (r *Route) WriteHeaderIncludedPacket(payload buffer.VectorisedView) *tcpip.Error { @@ -176,7 +214,7 @@ func (r *Route) WriteHeaderIncludedPacket(payload buffer.VectorisedView) *tcpip. return tcpip.ErrInvalidEndpointState } - if err := r.ref.ep.WriteHeaderIncludedPacket(r, payload, r.loop); err != nil { + if err := r.ref.ep.WriteHeaderIncludedPacket(r, payload, r.Loop); err != nil { r.Stats().IP.OutgoingPacketErrors.Increment() return err } diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 5ea036bea..082135374 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -607,17 +607,11 @@ func (e *endpoint) sendTCP(r *stack.Route, id stack.TransportEndpointID, data bu return nil } -// sendTCP sends a TCP segment with the provided options via the provided -// network endpoint and under the provided identity. -func sendTCP(r *stack.Route, id stack.TransportEndpointID, data buffer.VectorisedView, ttl, tos uint8, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) *tcpip.Error { +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) { optLen := len(opts) - // Allocate a buffer for the TCP header. - hdr := buffer.NewPrependable(header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen) - - if rcvWnd > 0xffff { - rcvWnd = 0xffff - } - + hdr := &d.Hdr + packetSize := d.Size + off := d.Off // Initialize the header. tcp := header.TCP(hdr.Prepend(header.TCPMinimumSize + optLen)) tcp.Encode(&header.TCPFields{ @@ -631,7 +625,7 @@ func sendTCP(r *stack.Route, id stack.TransportEndpointID, data buffer.Vectorise }) copy(tcp[header.TCPMinimumSize:], opts) - length := uint16(hdr.UsedLength() + data.Size()) + length := uint16(hdr.UsedLength() + packetSize) xsum := r.PseudoHeaderChecksum(ProtocolNumber, length) // Only calculate the checksum if offloading isn't supported. if gso != nil && gso.NeedsCsum { @@ -641,14 +635,71 @@ func sendTCP(r *stack.Route, id stack.TransportEndpointID, data buffer.Vectorise // header and data and get the right sum of the TCP packet. tcp.SetChecksum(xsum) } else if r.Capabilities()&stack.CapabilityTXChecksumOffload == 0 { - xsum = header.ChecksumVV(data, xsum) + xsum = header.ChecksumVVWithOffset(data, xsum, off, packetSize) tcp.SetChecksum(^tcp.CalculateChecksum(xsum)) } +} + +func sendTCPBatch(r *stack.Route, id stack.TransportEndpointID, data buffer.VectorisedView, ttl, tos uint8, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) *tcpip.Error { + optLen := len(opts) + if rcvWnd > 0xffff { + rcvWnd = 0xffff + } + + mss := int(gso.MSS) + n := (data.Size() + mss - 1) / mss + + hdrs := stack.NewPacketDescriptors(n, header.TCPMinimumSize+int(r.MaxHeaderLength())+optLen) + + size := data.Size() + off := 0 + for i := 0; i < n; i++ { + packetSize := mss + if packetSize > size { + packetSize = size + } + size -= packetSize + hdrs[i].Off = off + hdrs[i].Size = packetSize + buildTCPHdr(r, id, &hdrs[i], data, 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}) + if err != nil { + r.Stats().TCP.SegmentSendErrors.IncrementBy(uint64(n - sent)) + } + r.Stats().TCP.SegmentsSent.IncrementBy(uint64(sent)) + return err +} + +// sendTCP sends a TCP segment with the provided options via the provided +// network endpoint and under the provided identity. +func sendTCP(r *stack.Route, id stack.TransportEndpointID, data buffer.VectorisedView, ttl, tos uint8, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) *tcpip.Error { + optLen := len(opts) + if rcvWnd > 0xffff { + rcvWnd = 0xffff + } + + if r.Loop&stack.PacketLoop == 0 && gso != nil && gso.Type == stack.GSOSW && int(gso.MSS) < data.Size() { + 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(), + } + buildTCPHdr(r, id, d, data, flags, seq, ack, rcvWnd, opts, gso) + if ttl == 0 { ttl = r.DefaultTTL() } - if err := r.WritePacket(gso, hdr, data, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}); err != nil { + if err := r.WritePacket(gso, d.Hdr, data, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}); err != nil { r.Stats().TCP.SegmentSendErrors.Increment() return err } diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index a1b784b49..8b9cb4c33 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -2327,11 +2327,7 @@ func (e *endpoint) completeState() stack.TCPEndpointState { return s } -func (e *endpoint) initGSO() { - if e.route.Capabilities()&stack.CapabilityGSO == 0 { - return - } - +func (e *endpoint) initHardwareGSO() { gso := &stack.GSO{} switch e.route.NetProto { case header.IPv4ProtocolNumber: @@ -2349,6 +2345,18 @@ func (e *endpoint) initGSO() { e.gso = gso } +func (e *endpoint) initGSO() { + if e.route.Capabilities()&stack.CapabilityHardwareGSO != 0 { + e.initHardwareGSO() + } else if e.route.Capabilities()&stack.CapabilitySoftwareGSO != 0 { + e.gso = &stack.GSO{ + MaxSize: e.route.GSOMaxSize(), + Type: stack.GSOSW, + NeedsCsum: false, + } + } +} + // State implements tcpip.Endpoint.State. It exports the endpoint's protocol // state for diagnostics. func (e *endpoint) State() uint32 { |