summaryrefslogtreecommitdiffhomepage
path: root/pkg/tcpip/link/fdbased
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/tcpip/link/fdbased')
-rw-r--r--pkg/tcpip/link/fdbased/BUILD1
-rw-r--r--pkg/tcpip/link/fdbased/endpoint.go111
-rw-r--r--pkg/tcpip/link/fdbased/endpoint_test.go91
-rw-r--r--pkg/tcpip/link/fdbased/mmap.go2
-rw-r--r--pkg/tcpip/link/fdbased/packet_dispatchers.go12
5 files changed, 136 insertions, 81 deletions
diff --git a/pkg/tcpip/link/fdbased/BUILD b/pkg/tcpip/link/fdbased/BUILD
index aa6db9aea..507b44abc 100644
--- a/pkg/tcpip/link/fdbased/BUILD
+++ b/pkg/tcpip/link/fdbased/BUILD
@@ -15,6 +15,7 @@ go_library(
visibility = ["//visibility:public"],
deps = [
"//pkg/binary",
+ "//pkg/iovec",
"//pkg/sync",
"//pkg/tcpip",
"//pkg/tcpip/buffer",
diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go
index affa1bbdf..c18bb91fb 100644
--- a/pkg/tcpip/link/fdbased/endpoint.go
+++ b/pkg/tcpip/link/fdbased/endpoint.go
@@ -45,6 +45,7 @@ import (
"golang.org/x/sys/unix"
"gvisor.dev/gvisor/pkg/binary"
+ "gvisor.dev/gvisor/pkg/iovec"
"gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
@@ -385,26 +386,35 @@ const (
_VIRTIO_NET_HDR_GSO_TCPV6 = 4
)
-// 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 stack.PacketBuffer) *tcpip.Error {
+// AddHeader implements stack.LinkEndpoint.AddHeader.
+func (e *endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) {
if e.hdrSize > 0 {
// Add ethernet header if needed.
eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize))
pkt.LinkHeader = buffer.View(eth)
ethHdr := &header.EthernetFields{
- DstAddr: r.RemoteLinkAddress,
+ DstAddr: remote,
Type: protocol,
}
// Preserve the src address if it's set in the route.
- if r.LocalLinkAddress != "" {
- ethHdr.SrcAddr = r.LocalLinkAddress
+ if local != "" {
+ ethHdr.SrcAddr = local
} else {
ethHdr.SrcAddr = e.addr
}
eth.Encode(ethHdr)
}
+}
+
+// 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 *stack.PacketBuffer) *tcpip.Error {
+ if e.hdrSize > 0 {
+ e.AddHeader(r.LocalLinkAddress, r.RemoteLinkAddress, protocol, pkt)
+ }
+
+ var builder iovec.Builder
fd := e.fds[pkt.Hash%uint32(len(e.fds))]
if e.Capabilities()&stack.CapabilityHardwareGSO != 0 {
@@ -430,47 +440,28 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne
}
vnetHdrBuf := binary.Marshal(make([]byte, 0, virtioNetHdrSize), binary.LittleEndian, vnetHdr)
- return rawfile.NonBlockingWrite3(fd, vnetHdrBuf, pkt.Header.View(), pkt.Data.ToView())
+ builder.Add(vnetHdrBuf)
}
- if pkt.Data.Size() == 0 {
- return rawfile.NonBlockingWrite(fd, pkt.Header.View())
- }
- if pkt.Header.UsedLength() == 0 {
- return rawfile.NonBlockingWrite(fd, pkt.Data.ToView())
+ builder.Add(pkt.Header.View())
+ for _, v := range pkt.Data.Views() {
+ builder.Add(v)
}
- return rawfile.NonBlockingWrite3(fd, pkt.Header.View(), pkt.Data.ToView(), nil)
+ return rawfile.NonBlockingWriteIovec(fd, builder.Build())
}
func (e *endpoint) sendBatch(batchFD int, batch []*stack.PacketBuffer) (int, *tcpip.Error) {
// Send a batch of packets through batchFD.
mmsgHdrs := make([]rawfile.MMsgHdr, 0, len(batch))
for _, pkt := range batch {
- 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: pkt.EgressRoute.RemoteLinkAddress,
- Type: pkt.NetworkProtocolNumber,
- }
-
- // Preserve the src address if it's set in the route.
- if pkt.EgressRoute.LocalLinkAddress != "" {
- ethHdr.SrcAddr = pkt.EgressRoute.LocalLinkAddress
- } else {
- ethHdr.SrcAddr = e.addr
- }
- eth.Encode(ethHdr)
- iovLen++
+ e.AddHeader(pkt.EgressRoute.LocalLinkAddress, pkt.EgressRoute.RemoteLinkAddress, pkt.NetworkProtocolNumber, pkt)
}
- vnetHdr := virtioNetHdr{}
var vnetHdrBuf []byte
if e.Capabilities()&stack.CapabilityHardwareGSO != 0 {
+ vnetHdr := virtioNetHdr{}
if pkt.GSOOptions != nil {
vnetHdr.hdrLen = uint16(pkt.Header.UsedLength())
if pkt.GSOOptions.NeedsCsum {
@@ -491,45 +482,19 @@ func (e *endpoint) sendBatch(batchFD int, batch []*stack.PacketBuffer) (int, *tc
}
}
vnetHdrBuf = binary.Marshal(make([]byte, 0, virtioNetHdrSize), binary.LittleEndian, vnetHdr)
- iovLen++
}
- iovecs := make([]syscall.Iovec, iovLen+1+len(pkt.Data.Views()))
+ var builder iovec.Builder
+ builder.Add(vnetHdrBuf)
+ builder.Add(pkt.Header.View())
+ for _, v := range pkt.Data.Views() {
+ builder.Add(v)
+ }
+ iovecs := builder.Build()
+
var mmsgHdr rawfile.MMsgHdr
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 := &iovecs[iovecIdx]
- v.Base = &ethHdrBuf[0]
- v.Len = uint64(len(ethHdrBuf))
- 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++
-
- // Now encode the Transport Payload.
- pktViews := pkt.Data.Views()
- for i := range pktViews {
- vec := &iovecs[iovecIdx]
- iovecIdx++
- vec.Base = &pktViews[i][0]
- vec.Len = uint64(len(pktViews[i]))
- pktSize += vec.Len
- }
- mmsgHdr.Msg.Iovlen = uint64(iovecIdx)
+ mmsgHdr.Msg.Iovlen = uint64(len(iovecs))
mmsgHdrs = append(mmsgHdrs, mmsgHdr)
}
@@ -626,6 +591,14 @@ func (e *endpoint) GSOMaxSize() uint32 {
return e.gsoMaxSize
}
+// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType.
+func (e *endpoint) ARPHardwareType() header.ARPHardwareType {
+ if e.hdrSize > 0 {
+ return header.ARPHardwareEther
+ }
+ return header.ARPHardwareNone
+}
+
// InjectableEndpoint is an injectable fd-based endpoint. The endpoint writes
// to the FD, but does not read from it. All reads come from injected packets.
type InjectableEndpoint struct {
@@ -641,8 +614,8 @@ func (e *InjectableEndpoint) Attach(dispatcher stack.NetworkDispatcher) {
}
// InjectInbound injects an inbound packet.
-func (e *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) {
- e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, pkt)
+func (e *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) {
+ e.dispatcher.DeliverNetworkPacket("" /* remote */, "" /* local */, protocol, pkt)
}
// NewInjectable creates a new fd-based InjectableEndpoint.
diff --git a/pkg/tcpip/link/fdbased/endpoint_test.go b/pkg/tcpip/link/fdbased/endpoint_test.go
index 3bfb15a8e..7b995b85a 100644
--- a/pkg/tcpip/link/fdbased/endpoint_test.go
+++ b/pkg/tcpip/link/fdbased/endpoint_test.go
@@ -45,7 +45,7 @@ const (
type packetInfo struct {
raddr tcpip.LinkAddress
proto tcpip.NetworkProtocolNumber
- contents stack.PacketBuffer
+ contents *stack.PacketBuffer
}
type context struct {
@@ -103,10 +103,14 @@ func (c *context) cleanup() {
}
}
-func (c *context) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote tcpip.LinkAddress, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) {
+func (c *context) DeliverNetworkPacket(remote tcpip.LinkAddress, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) {
c.ch <- packetInfo{remote, protocol, pkt}
}
+func (c *context) DeliverOutboundPacket(remote tcpip.LinkAddress, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) {
+ panic("unimplemented")
+}
+
func TestNoEthernetProperties(t *testing.T) {
c := newContext(t, &Options{MTU: mtu})
defer c.cleanup()
@@ -179,7 +183,7 @@ func testWritePacket(t *testing.T, plen int, eth bool, gsoMaxSize uint32, hash u
L3HdrLen: header.IPv4MaximumHeaderSize,
}
}
- if err := c.ep.WritePacket(r, gso, proto, stack.PacketBuffer{
+ if err := c.ep.WritePacket(r, gso, proto, &stack.PacketBuffer{
Header: hdr,
Data: payload.ToVectorisedView(),
Hash: hash,
@@ -295,7 +299,7 @@ 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, stack.PacketBuffer{
+ if err := c.ep.WritePacket(r, nil /* gso */, proto, &stack.PacketBuffer{
Header: hdr,
Data: buffer.VectorisedView{},
}); err != nil {
@@ -358,7 +362,7 @@ func TestDeliverPacket(t *testing.T) {
want := packetInfo{
raddr: raddr,
proto: proto,
- contents: stack.PacketBuffer{
+ contents: &stack.PacketBuffer{
Data: buffer.View(b).ToVectorisedView(),
LinkHeader: buffer.View(hdr),
},
@@ -500,3 +504,80 @@ func TestRecvMMsgDispatcherCapLength(t *testing.T) {
}
}
+
+// fakeNetworkDispatcher delivers packets to pkts.
+type fakeNetworkDispatcher struct {
+ pkts []*stack.PacketBuffer
+}
+
+func (d *fakeNetworkDispatcher) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) {
+ d.pkts = append(d.pkts, pkt)
+}
+
+func (d *fakeNetworkDispatcher) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) {
+ panic("unimplemented")
+}
+
+func TestDispatchPacketFormat(t *testing.T) {
+ for _, test := range []struct {
+ name string
+ newDispatcher func(fd int, e *endpoint) (linkDispatcher, error)
+ }{
+ {
+ name: "readVDispatcher",
+ newDispatcher: newReadVDispatcher,
+ },
+ {
+ name: "recvMMsgDispatcher",
+ newDispatcher: newRecvMMsgDispatcher,
+ },
+ } {
+ t.Run(test.name, func(t *testing.T) {
+ // Create a socket pair to send/recv.
+ fds, err := syscall.Socketpair(syscall.AF_UNIX, syscall.SOCK_DGRAM, 0)
+ if err != nil {
+ t.Fatal(err)
+ }
+ defer syscall.Close(fds[0])
+ defer syscall.Close(fds[1])
+
+ data := []byte{
+ // Ethernet header.
+ 1, 2, 3, 4, 5, 60,
+ 1, 2, 3, 4, 5, 61,
+ 8, 0,
+ // Mock network header.
+ 40, 41, 42, 43,
+ }
+ err = syscall.Sendmsg(fds[1], data, nil, nil, 0)
+ if err != nil {
+ t.Fatal(err)
+ }
+
+ // Create and run dispatcher once.
+ sink := &fakeNetworkDispatcher{}
+ d, err := test.newDispatcher(fds[0], &endpoint{
+ hdrSize: header.EthernetMinimumSize,
+ dispatcher: sink,
+ })
+ if err != nil {
+ t.Fatal(err)
+ }
+ if ok, err := d.dispatch(); !ok || err != nil {
+ t.Fatalf("d.dispatch() = %v, %v", ok, err)
+ }
+
+ // Verify packet.
+ if got, want := len(sink.pkts), 1; got != want {
+ t.Fatalf("len(sink.pkts) = %d, want %d", got, want)
+ }
+ pkt := sink.pkts[0]
+ if got, want := len(pkt.LinkHeader), header.EthernetMinimumSize; got != want {
+ t.Errorf("len(pkt.LinkHeader) = %d, want %d", got, want)
+ }
+ if got, want := pkt.Data.Size(), 4; got != want {
+ t.Errorf("pkt.Data.Size() = %d, want %d", got, want)
+ }
+ })
+ }
+}
diff --git a/pkg/tcpip/link/fdbased/mmap.go b/pkg/tcpip/link/fdbased/mmap.go
index fe2bf3b0b..2dfd29aa9 100644
--- a/pkg/tcpip/link/fdbased/mmap.go
+++ b/pkg/tcpip/link/fdbased/mmap.go
@@ -191,7 +191,7 @@ func (d *packetMMapDispatcher) dispatch() (bool, *tcpip.Error) {
}
pkt = pkt[d.e.hdrSize:]
- d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, stack.PacketBuffer{
+ d.e.dispatcher.DeliverNetworkPacket(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 cb4cbea69..d8f2504b3 100644
--- a/pkg/tcpip/link/fdbased/packet_dispatchers.go
+++ b/pkg/tcpip/link/fdbased/packet_dispatchers.go
@@ -139,13 +139,13 @@ func (d *readVDispatcher) dispatch() (bool, *tcpip.Error) {
}
used := d.capViews(n, BufConfig)
- pkt := stack.PacketBuffer{
+ pkt := &stack.PacketBuffer{
Data: buffer.NewVectorisedView(n, append([]buffer.View(nil), d.views[:used]...)),
LinkHeader: buffer.View(eth),
}
pkt.Data.TrimFront(d.e.hdrSize)
- d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, pkt)
+ d.e.dispatcher.DeliverNetworkPacket(remote, local, p, pkt)
// Prepare e.views for another packet: release used views.
for i := 0; i < used; i++ {
@@ -169,7 +169,7 @@ type recvMMsgDispatcher struct {
// iovecs is an array of array of iovec records where each iovec base
// pointer and length are initialzed to the corresponding view above,
- // except when GSO is neabled then the first iovec in each array of
+ // except when GSO is enabled then the first iovec in each array of
// iovecs points to a buffer for the vnet header which is stripped
// before the views are passed up the stack for further processing.
iovecs [][]syscall.Iovec
@@ -278,7 +278,7 @@ func (d *recvMMsgDispatcher) dispatch() (bool, *tcpip.Error) {
eth header.Ethernet
)
if d.e.hdrSize > 0 {
- eth = header.Ethernet(d.views[k][0])
+ eth = header.Ethernet(d.views[k][0][:header.EthernetMinimumSize])
p = eth.Type()
remote = eth.SourceAddress()
local = eth.DestinationAddress()
@@ -296,12 +296,12 @@ func (d *recvMMsgDispatcher) dispatch() (bool, *tcpip.Error) {
}
used := d.capViews(k, int(n), BufConfig)
- pkt := stack.PacketBuffer{
+ pkt := &stack.PacketBuffer{
Data: buffer.NewVectorisedView(int(n), append([]buffer.View(nil), d.views[k][:used]...)),
LinkHeader: buffer.View(eth),
}
pkt.Data.TrimFront(d.e.hdrSize)
- d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, pkt)
+ d.e.dispatcher.DeliverNetworkPacket(remote, local, p, pkt)
// Prepare e.views for another packet: release used views.
for i := 0; i < used; i++ {
generated by cgit v1.2.3 (git 2.39.1) at 2025-04-13 15:34:10 +0000