diff options
Diffstat (limited to 'pkg/tcpip/link')
-rw-r--r-- | pkg/tcpip/link/channel/channel.go | 58 | ||||
-rw-r--r-- | pkg/tcpip/link/fdbased/endpoint.go | 178 | ||||
-rw-r--r-- | pkg/tcpip/link/fdbased/endpoint_test.go | 86 | ||||
-rw-r--r-- | pkg/tcpip/link/fdbased/endpoint_unsafe.go | 9 | ||||
-rw-r--r-- | pkg/tcpip/link/fdbased/mmap.go | 3 | ||||
-rw-r--r-- | pkg/tcpip/link/fdbased/packet_dispatchers.go | 4 | ||||
-rw-r--r-- | pkg/tcpip/link/loopback/loopback.go | 8 | ||||
-rw-r--r-- | pkg/tcpip/link/muxed/injectable.go | 6 | ||||
-rw-r--r-- | pkg/tcpip/link/muxed/injectable_test.go | 4 | ||||
-rw-r--r-- | pkg/tcpip/link/sharedmem/sharedmem.go | 6 | ||||
-rw-r--r-- | pkg/tcpip/link/sharedmem/sharedmem_test.go | 26 | ||||
-rw-r--r-- | pkg/tcpip/link/sniffer/sniffer.go | 170 | ||||
-rw-r--r-- | pkg/tcpip/link/tun/device.go | 12 | ||||
-rw-r--r-- | pkg/tcpip/link/waitable/waitable.go | 8 | ||||
-rw-r--r-- | pkg/tcpip/link/waitable/waitable_test.go | 22 |
15 files changed, 273 insertions, 327 deletions
diff --git a/pkg/tcpip/link/channel/channel.go b/pkg/tcpip/link/channel/channel.go index 5944ba190..9bf67686d 100644 --- a/pkg/tcpip/link/channel/channel.go +++ b/pkg/tcpip/link/channel/channel.go @@ -28,7 +28,7 @@ import ( // PacketInfo holds all the information about an outbound packet. type PacketInfo struct { - Pkt tcpip.PacketBuffer + Pkt *stack.PacketBuffer Proto tcpip.NetworkProtocolNumber GSO *stack.GSO Route stack.Route @@ -50,13 +50,11 @@ type NotificationHandle struct { } type queue struct { + // c is the outbound packet channel. + c chan PacketInfo // mu protects fields below. - mu sync.RWMutex - // c is the outbound packet channel. Sending to c should hold mu. - c chan PacketInfo - numWrite int - numRead int - notify []*NotificationHandle + mu sync.RWMutex + notify []*NotificationHandle } func (q *queue) Close() { @@ -64,11 +62,8 @@ func (q *queue) Close() { } func (q *queue) Read() (PacketInfo, bool) { - q.mu.Lock() - defer q.mu.Unlock() select { case p := <-q.c: - q.numRead++ return p, true default: return PacketInfo{}, false @@ -76,15 +71,8 @@ func (q *queue) Read() (PacketInfo, bool) { } func (q *queue) ReadContext(ctx context.Context) (PacketInfo, bool) { - // We have to receive from channel without holding the lock, since it can - // block indefinitely. This will cause a window that numWrite - numRead - // produces a larger number, but won't go to negative. numWrite >= numRead - // still holds. select { case pkt := <-q.c: - q.mu.Lock() - defer q.mu.Unlock() - q.numRead++ return pkt, true case <-ctx.Done(): return PacketInfo{}, false @@ -93,16 +81,12 @@ func (q *queue) ReadContext(ctx context.Context) (PacketInfo, bool) { func (q *queue) Write(p PacketInfo) bool { wrote := false - - // It's important to make sure nobody can see numWrite until we increment it, - // so numWrite >= numRead holds. - q.mu.Lock() select { case q.c <- p: wrote = true - q.numWrite++ default: } + q.mu.Lock() notify := q.notify q.mu.Unlock() @@ -116,13 +100,7 @@ func (q *queue) Write(p PacketInfo) bool { } func (q *queue) Num() int { - q.mu.RLock() - defer q.mu.RUnlock() - n := q.numWrite - q.numRead - if n < 0 { - panic("numWrite < numRead") - } - return n + return len(q.c) } func (q *queue) AddNotify(notify Notification) *NotificationHandle { @@ -203,12 +181,12 @@ func (e *Endpoint) NumQueued() int { } // InjectInbound injects an inbound packet. -func (e *Endpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { +func (e *Endpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt stack.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, pkt tcpip.PacketBuffer) { +func (e *Endpoint) InjectLinkAddr(protocol tcpip.NetworkProtocolNumber, remote tcpip.LinkAddress, pkt stack.PacketBuffer) { e.dispatcher.DeliverNetworkPacket(e, remote, "" /* local */, protocol, pkt) } @@ -251,13 +229,13 @@ func (e *Endpoint) LinkAddress() tcpip.LinkAddress { } // WritePacket stores outbound packets into the channel. -func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { +func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { // Clone r then release its resource so we only get the relevant fields from // stack.Route without holding a reference to a NIC's endpoint. route := r.Clone() route.Release() p := PacketInfo{ - Pkt: pkt, + Pkt: &pkt, Proto: protocol, GSO: gso, Route: route, @@ -269,21 +247,15 @@ func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne } // WritePackets stores outbound packets into the channel. -func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { // Clone r then release its resource so we only get the relevant fields from // stack.Route without holding a reference to a NIC's endpoint. route := r.Clone() route.Release() - payloadView := pkts[0].Data.ToView() n := 0 - for _, pkt := range pkts { - off := pkt.DataOffset - size := pkt.DataSize + for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() { p := PacketInfo{ - Pkt: tcpip.PacketBuffer{ - Header: pkt.Header, - Data: buffer.NewViewFromBytes(payloadView[off : off+size]).ToVectorisedView(), - }, + Pkt: pkt, Proto: protocol, GSO: gso, Route: route, @@ -301,7 +273,7 @@ func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.Pac // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *Endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { p := PacketInfo{ - Pkt: tcpip.PacketBuffer{Data: vv}, + Pkt: &stack.PacketBuffer{Data: vv}, Proto: 0, GSO: nil, } diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index b7f60178e..b857ce9d0 100644 --- a/pkg/tcpip/link/fdbased/endpoint.go +++ b/pkg/tcpip/link/fdbased/endpoint.go @@ -91,7 +91,7 @@ func (p PacketDispatchMode) String() string { case PacketMMap: return "PacketMMap" default: - return fmt.Sprintf("unknown packet dispatch mode %v", p) + return fmt.Sprintf("unknown packet dispatch mode '%d'", p) } } @@ -386,7 +386,7 @@ 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, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { if e.hdrSize > 0 { // Add ethernet header if needed. eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) @@ -405,9 +405,9 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne eth.Encode(ethHdr) } + fd := e.fds[pkt.Hash%uint32(len(e.fds))] if e.Capabilities()&stack.CapabilityHardwareGSO != 0 { vnetHdr := virtioNetHdr{} - vnetHdrBuf := vnetHdrToByteSlice(&vnetHdr) if gso != nil { vnetHdr.hdrLen = uint16(pkt.Header.UsedLength()) if gso.NeedsCsum { @@ -428,130 +428,120 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne } } - return rawfile.NonBlockingWrite3(e.fds[0], vnetHdrBuf, pkt.Header.View(), pkt.Data.ToView()) + vnetHdrBuf := vnetHdrToByteSlice(&vnetHdr) + return rawfile.NonBlockingWrite3(fd, vnetHdrBuf, pkt.Header.View(), pkt.Data.ToView()) } if pkt.Data.Size() == 0 { - return rawfile.NonBlockingWrite(e.fds[0], pkt.Header.View()) + return rawfile.NonBlockingWrite(fd, pkt.Header.View()) } - return rawfile.NonBlockingWrite3(e.fds[0], pkt.Header.View(), pkt.Data.ToView(), nil) + return rawfile.NonBlockingWrite3(fd, 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++ - } +// +// NOTE: This API uses sendmmsg to batch packets. As a result the underlying FD +// picked to write the packet out has to be the same for all packets in the +// list. In other words all packets in the batch should belong to the same +// flow. +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + n := pkts.Len() - 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) + i := 0 + for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() { + var ethHdrBuf []byte + iovLen := 0 + 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, + } - 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.") + // 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++ } - 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 + var vnetHdrBuf []byte + vnetHdr := virtioNetHdr{} + if e.Capabilities()&stack.CapabilityHardwareGSO != 0 { + if gso != nil { + 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(pkt.Data.Size()) > gso.MSS { + switch gso.Type { + case stack.GSOTCPv4: + vnetHdr.gsoType = _VIRTIO_NET_HDR_GSO_TCPV4 + case stack.GSOTCPv6: + vnetHdr.gsoType = _VIRTIO_NET_HDR_GSO_TCPV6 + default: + panic(fmt.Sprintf("Unknown gso type: %v", gso.Type)) + } + vnetHdr.gsoSize = gso.MSS } } + vnetHdrBuf = vnetHdrToByteSlice(&vnetHdr) + iovLen++ } - nextOff = off + packetSize + iovecs := make([]syscall.Iovec, iovLen+1+len(pkt.Data.Views())) + mmsgHdr := &mmsgHdrs[i] + mmsgHdr.Msg.Iov = &iovecs[0] + iovecIdx := 0 + if vnetHdrBuf != nil { + v := &iovecs[iovecIdx] + v.Base = &vnetHdrBuf[0] + v.Len = uint64(len(vnetHdrBuf)) + iovecIdx++ + } if ethHdrBuf != nil { - v := &iovec[iovecIdx] + v := &iovecs[iovecIdx] v.Base = ðHdrBuf[0] v.Len = uint64(len(ethHdrBuf)) iovecIdx++ } - - v := &iovec[iovecIdx] + pktSize := uint64(0) + // Encode L3 Header + v := &iovecs[iovecIdx] + hdr := &pkt.Header hdrView := hdr.View() v.Base = &hdrView[0] v.Len = uint64(len(hdrView)) + pktSize += v.Len iovecIdx++ - for packetSize > 0 { - vec := &iovec[iovecIdx] + // Now encode the Transport Payload. + pktViews := pkt.Data.Views() + for i := range pktViews { + vec := &iovecs[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 + vec.Base = &pktViews[i][0] + vec.Len = uint64(len(pktViews[i])) + pktSize += vec.Len } - - mmsgHdr.Msg.Iovlen = uint64(iovecIdx - prevIovecIdx) + mmsgHdr.Msg.Iovlen = uint64(iovecIdx) + i++ } packets := 0 for packets < n { - sent, err := rawfile.NonBlockingSendMMsg(e.fds[0], mmsgHdrs) + fd := e.fds[pkts.Front().Hash%uint32(len(e.fds))] + sent, err := rawfile.NonBlockingSendMMsg(fd, mmsgHdrs) if err != nil { return packets, err } @@ -610,7 +600,7 @@ func (e *InjectableEndpoint) Attach(dispatcher stack.NetworkDispatcher) { } // InjectInbound injects an inbound packet. -func (e *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { +func (e *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, pkt) } diff --git a/pkg/tcpip/link/fdbased/endpoint_test.go b/pkg/tcpip/link/fdbased/endpoint_test.go index 2066987eb..3bfb15a8e 100644 --- a/pkg/tcpip/link/fdbased/endpoint_test.go +++ b/pkg/tcpip/link/fdbased/endpoint_test.go @@ -45,40 +45,46 @@ const ( type packetInfo struct { raddr tcpip.LinkAddress proto tcpip.NetworkProtocolNumber - contents tcpip.PacketBuffer + contents stack.PacketBuffer } type context struct { - t *testing.T - fds [2]int - ep stack.LinkEndpoint - ch chan packetInfo - done chan struct{} + t *testing.T + readFDs []int + writeFDs []int + ep stack.LinkEndpoint + ch chan packetInfo + done chan struct{} } func newContext(t *testing.T, opt *Options) *context { - fds, err := syscall.Socketpair(syscall.AF_UNIX, syscall.SOCK_SEQPACKET, 0) + firstFDPair, err := syscall.Socketpair(syscall.AF_UNIX, syscall.SOCK_SEQPACKET, 0) + if err != nil { + t.Fatalf("Socketpair failed: %v", err) + } + secondFDPair, err := syscall.Socketpair(syscall.AF_UNIX, syscall.SOCK_SEQPACKET, 0) if err != nil { t.Fatalf("Socketpair failed: %v", err) } - done := make(chan struct{}, 1) + done := make(chan struct{}, 2) opt.ClosedFunc = func(*tcpip.Error) { done <- struct{}{} } - opt.FDs = []int{fds[1]} + opt.FDs = []int{firstFDPair[1], secondFDPair[1]} ep, err := New(opt) if err != nil { t.Fatalf("Failed to create FD endpoint: %v", err) } c := &context{ - t: t, - fds: fds, - ep: ep, - ch: make(chan packetInfo, 100), - done: done, + t: t, + readFDs: []int{firstFDPair[0], secondFDPair[0]}, + writeFDs: opt.FDs, + ep: ep, + ch: make(chan packetInfo, 100), + done: done, } ep.Attach(c) @@ -87,12 +93,17 @@ func newContext(t *testing.T, opt *Options) *context { } func (c *context) cleanup() { - syscall.Close(c.fds[0]) + for _, fd := range c.readFDs { + syscall.Close(fd) + } + <-c.done <-c.done - syscall.Close(c.fds[1]) + for _, fd := range c.writeFDs { + syscall.Close(fd) + } } -func (c *context) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote tcpip.LinkAddress, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { +func (c *context) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote tcpip.LinkAddress, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { c.ch <- packetInfo{remote, protocol, pkt} } @@ -136,7 +147,7 @@ func TestAddress(t *testing.T) { } } -func testWritePacket(t *testing.T, plen int, eth bool, gsoMaxSize uint32) { +func testWritePacket(t *testing.T, plen int, eth bool, gsoMaxSize uint32, hash uint32) { c := newContext(t, &Options{Address: laddr, MTU: mtu, EthernetHeader: eth, GSOMaxSize: gsoMaxSize}) defer c.cleanup() @@ -168,16 +179,18 @@ func testWritePacket(t *testing.T, plen int, eth bool, gsoMaxSize uint32) { L3HdrLen: header.IPv4MaximumHeaderSize, } } - if err := c.ep.WritePacket(r, gso, proto, tcpip.PacketBuffer{ + if err := c.ep.WritePacket(r, gso, proto, stack.PacketBuffer{ Header: hdr, Data: payload.ToVectorisedView(), + Hash: hash, }); err != nil { t.Fatalf("WritePacket failed: %v", err) } - // Read from fd, then compare with what we wrote. + // Read from the corresponding FD, then compare with what we wrote. b = make([]byte, mtu) - n, err := syscall.Read(c.fds[0], b) + fd := c.readFDs[hash%uint32(len(c.readFDs))] + n, err := syscall.Read(fd, b) if err != nil { t.Fatalf("Read failed: %v", err) } @@ -238,7 +251,7 @@ func TestWritePacket(t *testing.T) { t.Run( fmt.Sprintf("Eth=%v,PayloadLen=%v,GSOMaxSize=%v", eth, plen, gso), func(t *testing.T) { - testWritePacket(t, plen, eth, gso) + testWritePacket(t, plen, eth, gso, 0) }, ) } @@ -246,6 +259,27 @@ func TestWritePacket(t *testing.T) { } } +func TestHashedWritePacket(t *testing.T) { + lengths := []int{0, 100, 1000} + eths := []bool{true, false} + gsos := []uint32{0, 32768} + hashes := []uint32{0, 1} + for _, eth := range eths { + for _, plen := range lengths { + for _, gso := range gsos { + for _, hash := range hashes { + t.Run( + fmt.Sprintf("Eth=%v,PayloadLen=%v,GSOMaxSize=%v,Hash=%d", eth, plen, gso, hash), + func(t *testing.T) { + testWritePacket(t, plen, eth, gso, hash) + }, + ) + } + } + } + } +} + func TestPreserveSrcAddress(t *testing.T) { baddr := tcpip.LinkAddress("\xcc\xbb\xaa\x77\x88\x99") @@ -261,7 +295,7 @@ 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 */, proto, tcpip.PacketBuffer{ + if err := c.ep.WritePacket(r, nil /* gso */, proto, stack.PacketBuffer{ Header: hdr, Data: buffer.VectorisedView{}, }); err != nil { @@ -270,7 +304,7 @@ func TestPreserveSrcAddress(t *testing.T) { // Read from the FD, then compare with what we wrote. b := make([]byte, mtu) - n, err := syscall.Read(c.fds[0], b) + n, err := syscall.Read(c.readFDs[0], b) if err != nil { t.Fatalf("Read failed: %v", err) } @@ -314,7 +348,7 @@ func TestDeliverPacket(t *testing.T) { } // Write packet via the file descriptor. - if _, err := syscall.Write(c.fds[0], all); err != nil { + if _, err := syscall.Write(c.readFDs[0], all); err != nil { t.Fatalf("Write failed: %v", err) } @@ -324,7 +358,7 @@ func TestDeliverPacket(t *testing.T) { want := packetInfo{ raddr: raddr, proto: proto, - contents: tcpip.PacketBuffer{ + contents: stack.PacketBuffer{ Data: buffer.View(b).ToVectorisedView(), LinkHeader: buffer.View(hdr), }, diff --git a/pkg/tcpip/link/fdbased/endpoint_unsafe.go b/pkg/tcpip/link/fdbased/endpoint_unsafe.go index 97a477b61..d81858353 100644 --- a/pkg/tcpip/link/fdbased/endpoint_unsafe.go +++ b/pkg/tcpip/link/fdbased/endpoint_unsafe.go @@ -24,9 +24,10 @@ import ( const virtioNetHdrSize = int(unsafe.Sizeof(virtioNetHdr{})) func vnetHdrToByteSlice(hdr *virtioNetHdr) (slice []byte) { - sh := (*reflect.SliceHeader)(unsafe.Pointer(&slice)) - sh.Data = uintptr(unsafe.Pointer(hdr)) - sh.Len = virtioNetHdrSize - sh.Cap = virtioNetHdrSize + *(*reflect.SliceHeader)(unsafe.Pointer(&slice)) = reflect.SliceHeader{ + Data: uintptr((unsafe.Pointer(hdr))), + Len: virtioNetHdrSize, + Cap: virtioNetHdrSize, + } return } diff --git a/pkg/tcpip/link/fdbased/mmap.go b/pkg/tcpip/link/fdbased/mmap.go index 62ed1e569..fe2bf3b0b 100644 --- a/pkg/tcpip/link/fdbased/mmap.go +++ b/pkg/tcpip/link/fdbased/mmap.go @@ -25,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/link/rawfile" + "gvisor.dev/gvisor/pkg/tcpip/stack" ) const ( @@ -190,7 +191,7 @@ func (d *packetMMapDispatcher) dispatch() (bool, *tcpip.Error) { } pkt = pkt[d.e.hdrSize:] - d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, tcpip.PacketBuffer{ + d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, stack.PacketBuffer{ Data: buffer.View(pkt).ToVectorisedView(), LinkHeader: buffer.View(eth), }) diff --git a/pkg/tcpip/link/fdbased/packet_dispatchers.go b/pkg/tcpip/link/fdbased/packet_dispatchers.go index c67d684ce..cb4cbea69 100644 --- a/pkg/tcpip/link/fdbased/packet_dispatchers.go +++ b/pkg/tcpip/link/fdbased/packet_dispatchers.go @@ -139,7 +139,7 @@ func (d *readVDispatcher) dispatch() (bool, *tcpip.Error) { } used := d.capViews(n, BufConfig) - pkt := tcpip.PacketBuffer{ + pkt := stack.PacketBuffer{ Data: buffer.NewVectorisedView(n, append([]buffer.View(nil), d.views[:used]...)), LinkHeader: buffer.View(eth), } @@ -296,7 +296,7 @@ func (d *recvMMsgDispatcher) dispatch() (bool, *tcpip.Error) { } used := d.capViews(k, int(n), BufConfig) - pkt := tcpip.PacketBuffer{ + pkt := stack.PacketBuffer{ Data: buffer.NewVectorisedView(int(n), append([]buffer.View(nil), d.views[k][:used]...)), LinkHeader: buffer.View(eth), } diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index 499cc608f..1e2255bfa 100644 --- a/pkg/tcpip/link/loopback/loopback.go +++ b/pkg/tcpip/link/loopback/loopback.go @@ -76,7 +76,7 @@ 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, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { +func (e *endpoint) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { views := make([]buffer.View, 1, 1+len(pkt.Data.Views())) views[0] = pkt.Header.View() views = append(views, pkt.Data.Views()...) @@ -84,7 +84,7 @@ func (e *endpoint) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.Netw // 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{ + e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, stack.PacketBuffer{ Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), }) @@ -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, []tcpip.PacketBuffer, tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, stack.PacketBufferList, tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { panic("not implemented") } @@ -106,7 +106,7 @@ func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { // 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{ + e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, linkHeader.Type(), stack.PacketBuffer{ Data: vv, LinkHeader: buffer.View(linkHeader), }) diff --git a/pkg/tcpip/link/muxed/injectable.go b/pkg/tcpip/link/muxed/injectable.go index 445b22c17..a5478ce17 100644 --- a/pkg/tcpip/link/muxed/injectable.go +++ b/pkg/tcpip/link/muxed/injectable.go @@ -80,14 +80,14 @@ func (m *InjectableEndpoint) IsAttached() bool { } // InjectInbound implements stack.InjectableLinkEndpoint. -func (m *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { +func (m *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt stack.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) { +func (m *InjectableEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { endpoint, ok := m.routes[r.RemoteAddress] if !ok { return 0, tcpip.ErrNoRoute @@ -98,7 +98,7 @@ func (m *InjectableEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts [ // 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, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { +func (m *InjectableEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { if endpoint, ok := m.routes[r.RemoteAddress]; ok { return endpoint.WritePacket(r, gso, protocol, pkt) } diff --git a/pkg/tcpip/link/muxed/injectable_test.go b/pkg/tcpip/link/muxed/injectable_test.go index 63b249837..87c734c1f 100644 --- a/pkg/tcpip/link/muxed/injectable_test.go +++ b/pkg/tcpip/link/muxed/injectable_test.go @@ -50,7 +50,7 @@ func TestInjectableEndpointDispatch(t *testing.T) { hdr.Prepend(1)[0] = 0xFA packetRoute := stack.Route{RemoteAddress: dstIP} - endpoint.WritePacket(&packetRoute, nil /* gso */, ipv4.ProtocolNumber, tcpip.PacketBuffer{ + endpoint.WritePacket(&packetRoute, nil /* gso */, ipv4.ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: buffer.NewViewFromBytes([]byte{0xFB}).ToVectorisedView(), }) @@ -70,7 +70,7 @@ func TestInjectableEndpointDispatchHdrOnly(t *testing.T) { hdr := buffer.NewPrependable(1) hdr.Prepend(1)[0] = 0xFA packetRoute := stack.Route{RemoteAddress: dstIP} - endpoint.WritePacket(&packetRoute, nil /* gso */, ipv4.ProtocolNumber, tcpip.PacketBuffer{ + endpoint.WritePacket(&packetRoute, nil /* gso */, ipv4.ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: buffer.NewView(0).ToVectorisedView(), }) diff --git a/pkg/tcpip/link/sharedmem/sharedmem.go b/pkg/tcpip/link/sharedmem/sharedmem.go index 655e537c4..0796d717e 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem.go +++ b/pkg/tcpip/link/sharedmem/sharedmem.go @@ -185,7 +185,7 @@ 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, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { +func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { // Add the ethernet header here. eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) pkt.LinkHeader = buffer.View(eth) @@ -214,7 +214,7 @@ func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.Netw } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (e *endpoint) WritePackets(r *stack.Route, _ *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(r *stack.Route, _ *stack.GSO, pkts stack.PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { panic("not implemented") } @@ -275,7 +275,7 @@ func (e *endpoint) dispatchLoop(d stack.NetworkDispatcher) { // Send packet up the stack. eth := header.Ethernet(b[:header.EthernetMinimumSize]) - d.DeliverNetworkPacket(e, eth.SourceAddress(), eth.DestinationAddress(), eth.Type(), tcpip.PacketBuffer{ + d.DeliverNetworkPacket(e, eth.SourceAddress(), eth.DestinationAddress(), eth.Type(), stack.PacketBuffer{ Data: buffer.View(b[header.EthernetMinimumSize:]).ToVectorisedView(), LinkHeader: buffer.View(eth), }) diff --git a/pkg/tcpip/link/sharedmem/sharedmem_test.go b/pkg/tcpip/link/sharedmem/sharedmem_test.go index 5c729a439..27ea3f531 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem_test.go +++ b/pkg/tcpip/link/sharedmem/sharedmem_test.go @@ -131,7 +131,7 @@ 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, pkt tcpip.PacketBuffer) { +func (c *testContext) DeliverNetworkPacket(_ stack.LinkEndpoint, remoteLinkAddr, localLinkAddr tcpip.LinkAddress, proto tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { c.mu.Lock() c.packets = append(c.packets, packetInfo{ addr: remoteLinkAddr, @@ -273,7 +273,7 @@ func TestSimpleSend(t *testing.T) { randomFill(buf) proto := tcpip.NetworkProtocolNumber(rand.Intn(0x10000)) - if err := c.ep.WritePacket(&r, nil /* gso */, proto, tcpip.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, proto, stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != nil { @@ -345,7 +345,7 @@ func TestPreserveSrcAddressInSend(t *testing.T) { hdr := buffer.NewPrependable(header.EthernetMinimumSize) proto := tcpip.NetworkProtocolNumber(rand.Intn(0x10000)) - if err := c.ep.WritePacket(&r, nil /* gso */, proto, tcpip.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, proto, stack.PacketBuffer{ Header: hdr, }); err != nil { t.Fatalf("WritePacket failed: %v", err) @@ -401,7 +401,7 @@ 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 */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != nil { @@ -419,7 +419,7 @@ 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 */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != want { @@ -447,7 +447,7 @@ 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 */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != nil { @@ -470,7 +470,7 @@ 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 */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != nil { @@ -488,7 +488,7 @@ 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 */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != want { @@ -514,7 +514,7 @@ 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 */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != nil { @@ -533,7 +533,7 @@ 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 */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }) @@ -561,7 +561,7 @@ 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 */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != nil { @@ -577,7 +577,7 @@ 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 */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: uu, }); err != want { @@ -588,7 +588,7 @@ 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 */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != nil { diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go index 3392b7edd..be2537a82 100644 --- a/pkg/tcpip/link/sniffer/sniffer.go +++ b/pkg/tcpip/link/sniffer/sniffer.go @@ -21,11 +21,9 @@ package sniffer import ( - "bytes" "encoding/binary" "fmt" "io" - "os" "sync/atomic" "time" @@ -42,12 +40,12 @@ import ( // LogPackets must be accessed atomically. var LogPackets uint32 = 1 -// LogPacketsToFile is a flag used to enable or disable logging packets to a -// pcap file. Valid values are 0 or 1. A file must have been specified when the +// LogPacketsToPCAP is a flag used to enable or disable logging packets to a +// pcap writer. Valid values are 0 or 1. A writer must have been specified when the // sniffer was created for this flag to have effect. // -// LogPacketsToFile must be accessed atomically. -var LogPacketsToFile uint32 = 1 +// LogPacketsToPCAP must be accessed atomically. +var LogPacketsToPCAP uint32 = 1 var transportProtocolMinSizes map[tcpip.TransportProtocolNumber]int = map[tcpip.TransportProtocolNumber]int{ header.ICMPv4ProtocolNumber: header.IPv4MinimumSize, @@ -59,7 +57,7 @@ var transportProtocolMinSizes map[tcpip.TransportProtocolNumber]int = map[tcpip. type endpoint struct { dispatcher stack.NetworkDispatcher lower stack.LinkEndpoint - file *os.File + writer io.Writer maxPCAPLen uint32 } @@ -99,23 +97,22 @@ func writePCAPHeader(w io.Writer, maxLen uint32) error { }) } -// NewWithFile creates a new sniffer link-layer endpoint. It wraps around -// another endpoint and logs packets and they traverse the endpoint. +// NewWithWriter creates a new sniffer link-layer endpoint. It wraps around +// another endpoint and logs packets as they traverse the endpoint. // -// Packets can be logged to file in the pcap format. A sniffer created -// with this function will not emit packets using the standard log -// package. +// Packets are logged to writer in the pcap format. A sniffer created with this +// function will not emit packets using the standard log package. // // snapLen is the maximum amount of a packet to be saved. Packets with a length -// less than or equal too snapLen will be saved in their entirety. Longer +// less than or equal to snapLen will be saved in their entirety. Longer // packets will be truncated to snapLen. -func NewWithFile(lower stack.LinkEndpoint, file *os.File, snapLen uint32) (stack.LinkEndpoint, error) { - if err := writePCAPHeader(file, snapLen); err != nil { +func NewWithWriter(lower stack.LinkEndpoint, writer io.Writer, snapLen uint32) (stack.LinkEndpoint, error) { + if err := writePCAPHeader(writer, snapLen); err != nil { return nil, err } return &endpoint{ lower: lower, - file: file, + writer: writer, maxPCAPLen: snapLen, }, nil } @@ -123,37 +120,8 @@ 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, pkt tcpip.PacketBuffer) { - if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil { - logPacket("recv", protocol, pkt.Data.First(), nil) - } - if e.file != nil && atomic.LoadUint32(&LogPacketsToFile) == 1 { - 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(pkt.Data.Size()))); err != nil { - panic(err) - } - for _, v := range vs { - if length == 0 { - break - } - if len(v) > length { - v = v[:length] - } - if _, err := buf.Write([]byte(v)); err != nil { - panic(err) - } - length -= len(v) - } - if _, err := e.file.Write(buf.Bytes()); err != nil { - panic(err) - } - } +func (e *endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { + e.dumpPacket("recv", nil, protocol, &pkt) e.dispatcher.DeliverNetworkPacket(e, remote, local, protocol, pkt) } @@ -200,31 +168,43 @@ func (e *endpoint) GSOMaxSize() uint32 { return 0 } -func (e *endpoint) dumpPacket(gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { - if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil { - logPacket("send", protocol, pkt.Header.View(), gso) +func (e *endpoint) dumpPacket(prefix string, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + writer := e.writer + if writer == nil && atomic.LoadUint32(&LogPackets) == 1 { + first := pkt.Header.View() + if len(first) == 0 { + first = pkt.Data.First() + } + logPacket(prefix, protocol, first, gso) } - if e.file != nil && atomic.LoadUint32(&LogPacketsToFile) == 1 { - hdrBuf := pkt.Header.View() - length := len(hdrBuf) + pkt.Data.Size() - if length > int(e.maxPCAPLen) { - length = int(e.maxPCAPLen) + if writer != nil && atomic.LoadUint32(&LogPacketsToPCAP) == 1 { + totalLength := pkt.Header.UsedLength() + pkt.Data.Size() + length := totalLength + if max := int(e.maxPCAPLen); length > max { + length = max } - - buf := bytes.NewBuffer(make([]byte, 0, pcapPacketHeaderLen+length)) - if err := binary.Write(buf, binary.BigEndian, newPCAPPacketHeader(uint32(length), uint32(len(hdrBuf)+pkt.Data.Size()))); err != nil { + if err := binary.Write(writer, binary.BigEndian, newPCAPPacketHeader(uint32(length), uint32(totalLength))); err != nil { panic(err) } - if len(hdrBuf) > length { - hdrBuf = hdrBuf[:length] - } - if _, err := buf.Write(hdrBuf); err != nil { - panic(err) + write := func(b []byte) { + if len(b) > length { + b = b[:length] + } + for len(b) != 0 { + n, err := writer.Write(b) + if err != nil { + panic(err) + } + b = b[n:] + length -= n + } } - length -= len(hdrBuf) - logVectorisedView(pkt.Data, length, buf) - if _, err := e.file.Write(buf.Bytes()); err != nil { - panic(err) + write(pkt.Header.View()) + for _, view := range pkt.Data.Views() { + if length == 0 { + break + } + write(view) } } } @@ -232,69 +212,31 @@ func (e *endpoint) dumpPacket(gso *stack.GSO, protocol tcpip.NetworkProtocolNumb // 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) +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { + e.dumpPacket("send", 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(), - }) +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() { + e.dumpPacket("send", gso, protocol, pkt) } 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) - } - } + e.dumpPacket("send", nil, 0, &stack.PacketBuffer{ + Data: vv, + }) 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. -func (*endpoint) Wait() {} +func (e *endpoint) Wait() { e.lower.Wait() } func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.View, gso *stack.GSO) { // Figure out the network layer info. diff --git a/pkg/tcpip/link/tun/device.go b/pkg/tcpip/link/tun/device.go index 6ff47a742..617446ea2 100644 --- a/pkg/tcpip/link/tun/device.go +++ b/pkg/tcpip/link/tun/device.go @@ -98,7 +98,12 @@ func (d *Device) SetIff(s *stack.Stack, name string, flags uint16) error { prefix = "tap" } - endpoint, err := attachOrCreateNIC(s, name, prefix) + linkCaps := stack.CapabilityNone + if isTap { + linkCaps |= stack.CapabilityResolutionRequired + } + + endpoint, err := attachOrCreateNIC(s, name, prefix, linkCaps) if err != nil { return syserror.EINVAL } @@ -109,7 +114,7 @@ func (d *Device) SetIff(s *stack.Stack, name string, flags uint16) error { return nil } -func attachOrCreateNIC(s *stack.Stack, name, prefix string) (*tunEndpoint, error) { +func attachOrCreateNIC(s *stack.Stack, name, prefix string, linkCaps stack.LinkEndpointCapabilities) (*tunEndpoint, error) { for { // 1. Try to attach to an existing NIC. if name != "" { @@ -135,6 +140,7 @@ func attachOrCreateNIC(s *stack.Stack, name, prefix string) (*tunEndpoint, error nicID: id, name: name, } + endpoint.Endpoint.LinkEPCapabilities = linkCaps if endpoint.name == "" { endpoint.name = fmt.Sprintf("%s%d", prefix, id) } @@ -207,7 +213,7 @@ func (d *Device) Write(data []byte) (int64, error) { remote = tcpip.LinkAddress(zeroMAC[:]) } - pkt := tcpip.PacketBuffer{ + pkt := stack.PacketBuffer{ Data: buffer.View(data).ToVectorisedView(), } if ethHdr != nil { diff --git a/pkg/tcpip/link/waitable/waitable.go b/pkg/tcpip/link/waitable/waitable.go index a8de38979..2b3741276 100644 --- a/pkg/tcpip/link/waitable/waitable.go +++ b/pkg/tcpip/link/waitable/waitable.go @@ -50,7 +50,7 @@ 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, pkt tcpip.PacketBuffer) { +func (e *Endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { if !e.dispatchGate.Enter() { return } @@ -99,7 +99,7 @@ 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, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { +func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { if !e.writeGate.Enter() { return nil } @@ -112,9 +112,9 @@ func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne // 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) { +func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { if !e.writeGate.Enter() { - return len(pkts), nil + return pkts.Len(), nil } n, err := e.lower.WritePackets(r, gso, pkts, protocol) diff --git a/pkg/tcpip/link/waitable/waitable_test.go b/pkg/tcpip/link/waitable/waitable_test.go index 31b11a27a..54eb5322b 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, pkt tcpip.PacketBuffer) { +func (e *countedEndpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { e.dispatchCount++ } @@ -65,15 +65,15 @@ func (e *countedEndpoint) LinkAddress() tcpip.LinkAddress { return e.linkAddr } -func (e *countedEndpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { +func (e *countedEndpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.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) WritePackets(r *stack.Route, _ *stack.GSO, pkts stack.PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + e.writeCount += pkts.Len() + return pkts.Len(), nil } func (e *countedEndpoint) WriteRawPacket(buffer.VectorisedView) *tcpip.Error { @@ -89,21 +89,21 @@ func TestWaitWrite(t *testing.T) { wep := New(ep) // Write and check that it goes through. - wep.WritePacket(nil, nil /* gso */, 0, tcpip.PacketBuffer{}) + wep.WritePacket(nil, nil /* gso */, 0, stack.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 */, 0, tcpip.PacketBuffer{}) + wep.WritePacket(nil, nil /* gso */, 0, stack.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 */, 0, tcpip.PacketBuffer{}) + wep.WritePacket(nil, nil /* gso */, 0, stack.PacketBuffer{}) if want := 2; ep.writeCount != want { t.Fatalf("Unexpected writeCount: got=%v, want=%v", ep.writeCount, want) } @@ -120,21 +120,21 @@ func TestWaitDispatch(t *testing.T) { } // Dispatch and check that it goes through. - ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, tcpip.PacketBuffer{}) + ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, stack.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, tcpip.PacketBuffer{}) + ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, stack.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, tcpip.PacketBuffer{}) + ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, stack.PacketBuffer{}) if want := 2; ep.dispatchCount != want { t.Fatalf("Unexpected dispatchCount: got=%v, want=%v", ep.dispatchCount, want) } |