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/endpoint.go178
-rw-r--r--pkg/tcpip/link/fdbased/endpoint_test.go86
-rw-r--r--pkg/tcpip/link/fdbased/endpoint_unsafe.go9
-rw-r--r--pkg/tcpip/link/fdbased/mmap.go3
-rw-r--r--pkg/tcpip/link/fdbased/packet_dispatchers.go4
5 files changed, 153 insertions, 127 deletions
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 = &ethHdrBuf[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),
}
generated by cgit v1.2.3 (git 2.39.1) at 2024-11-08 18:53:00 +0000