Merge remote-tracking branch 'upstream/master'

This commit is contained in:
Iurii Egorov 2023-11-17 19:51:13 +03:00
commit 7af55a3e6f
27 changed files with 1110 additions and 397 deletions

View file

@ -3,23 +3,99 @@ package tun
import "encoding/binary"
// TODO: Explore SIMD and/or other assembly optimizations.
// TODO: Test native endian loads. See RFC 1071 section 2 part B.
func checksumNoFold(b []byte, initial uint64) uint64 {
ac := initial
i := 0
n := len(b)
for n >= 4 {
ac += uint64(binary.BigEndian.Uint32(b[i : i+4]))
n -= 4
i += 4
for len(b) >= 128 {
ac += uint64(binary.BigEndian.Uint32(b[:4]))
ac += uint64(binary.BigEndian.Uint32(b[4:8]))
ac += uint64(binary.BigEndian.Uint32(b[8:12]))
ac += uint64(binary.BigEndian.Uint32(b[12:16]))
ac += uint64(binary.BigEndian.Uint32(b[16:20]))
ac += uint64(binary.BigEndian.Uint32(b[20:24]))
ac += uint64(binary.BigEndian.Uint32(b[24:28]))
ac += uint64(binary.BigEndian.Uint32(b[28:32]))
ac += uint64(binary.BigEndian.Uint32(b[32:36]))
ac += uint64(binary.BigEndian.Uint32(b[36:40]))
ac += uint64(binary.BigEndian.Uint32(b[40:44]))
ac += uint64(binary.BigEndian.Uint32(b[44:48]))
ac += uint64(binary.BigEndian.Uint32(b[48:52]))
ac += uint64(binary.BigEndian.Uint32(b[52:56]))
ac += uint64(binary.BigEndian.Uint32(b[56:60]))
ac += uint64(binary.BigEndian.Uint32(b[60:64]))
ac += uint64(binary.BigEndian.Uint32(b[64:68]))
ac += uint64(binary.BigEndian.Uint32(b[68:72]))
ac += uint64(binary.BigEndian.Uint32(b[72:76]))
ac += uint64(binary.BigEndian.Uint32(b[76:80]))
ac += uint64(binary.BigEndian.Uint32(b[80:84]))
ac += uint64(binary.BigEndian.Uint32(b[84:88]))
ac += uint64(binary.BigEndian.Uint32(b[88:92]))
ac += uint64(binary.BigEndian.Uint32(b[92:96]))
ac += uint64(binary.BigEndian.Uint32(b[96:100]))
ac += uint64(binary.BigEndian.Uint32(b[100:104]))
ac += uint64(binary.BigEndian.Uint32(b[104:108]))
ac += uint64(binary.BigEndian.Uint32(b[108:112]))
ac += uint64(binary.BigEndian.Uint32(b[112:116]))
ac += uint64(binary.BigEndian.Uint32(b[116:120]))
ac += uint64(binary.BigEndian.Uint32(b[120:124]))
ac += uint64(binary.BigEndian.Uint32(b[124:128]))
b = b[128:]
}
for n >= 2 {
ac += uint64(binary.BigEndian.Uint16(b[i : i+2]))
n -= 2
i += 2
if len(b) >= 64 {
ac += uint64(binary.BigEndian.Uint32(b[:4]))
ac += uint64(binary.BigEndian.Uint32(b[4:8]))
ac += uint64(binary.BigEndian.Uint32(b[8:12]))
ac += uint64(binary.BigEndian.Uint32(b[12:16]))
ac += uint64(binary.BigEndian.Uint32(b[16:20]))
ac += uint64(binary.BigEndian.Uint32(b[20:24]))
ac += uint64(binary.BigEndian.Uint32(b[24:28]))
ac += uint64(binary.BigEndian.Uint32(b[28:32]))
ac += uint64(binary.BigEndian.Uint32(b[32:36]))
ac += uint64(binary.BigEndian.Uint32(b[36:40]))
ac += uint64(binary.BigEndian.Uint32(b[40:44]))
ac += uint64(binary.BigEndian.Uint32(b[44:48]))
ac += uint64(binary.BigEndian.Uint32(b[48:52]))
ac += uint64(binary.BigEndian.Uint32(b[52:56]))
ac += uint64(binary.BigEndian.Uint32(b[56:60]))
ac += uint64(binary.BigEndian.Uint32(b[60:64]))
b = b[64:]
}
if n == 1 {
ac += uint64(b[i]) << 8
if len(b) >= 32 {
ac += uint64(binary.BigEndian.Uint32(b[:4]))
ac += uint64(binary.BigEndian.Uint32(b[4:8]))
ac += uint64(binary.BigEndian.Uint32(b[8:12]))
ac += uint64(binary.BigEndian.Uint32(b[12:16]))
ac += uint64(binary.BigEndian.Uint32(b[16:20]))
ac += uint64(binary.BigEndian.Uint32(b[20:24]))
ac += uint64(binary.BigEndian.Uint32(b[24:28]))
ac += uint64(binary.BigEndian.Uint32(b[28:32]))
b = b[32:]
}
if len(b) >= 16 {
ac += uint64(binary.BigEndian.Uint32(b[:4]))
ac += uint64(binary.BigEndian.Uint32(b[4:8]))
ac += uint64(binary.BigEndian.Uint32(b[8:12]))
ac += uint64(binary.BigEndian.Uint32(b[12:16]))
b = b[16:]
}
if len(b) >= 8 {
ac += uint64(binary.BigEndian.Uint32(b[:4]))
ac += uint64(binary.BigEndian.Uint32(b[4:8]))
b = b[8:]
}
if len(b) >= 4 {
ac += uint64(binary.BigEndian.Uint32(b))
b = b[4:]
}
if len(b) >= 2 {
ac += uint64(binary.BigEndian.Uint16(b))
b = b[2:]
}
if len(b) == 1 {
ac += uint64(b[0]) << 8
}
return ac
}

35
tun/checksum_test.go Normal file
View file

@ -0,0 +1,35 @@
package tun
import (
"fmt"
"math/rand"
"testing"
)
func BenchmarkChecksum(b *testing.B) {
lengths := []int{
64,
128,
256,
512,
1024,
1500,
2048,
4096,
8192,
9000,
9001,
}
for _, length := range lengths {
b.Run(fmt.Sprintf("%d", length), func(b *testing.B) {
buf := make([]byte, length)
rng := rand.New(rand.NewSource(1))
rng.Read(buf)
b.ResetTimer()
for i := 0; i < b.N; i++ {
checksum(buf, 0)
}
})
}
}

View file

@ -25,7 +25,7 @@ import (
"github.com/amnezia-vpn/amnezia-wg/tun"
"golang.org/x/net/dns/dnsmessage"
"gvisor.dev/gvisor/pkg/bufferv2"
"gvisor.dev/gvisor/pkg/buffer"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/adapters/gonet"
"gvisor.dev/gvisor/pkg/tcpip/header"
@ -43,7 +43,7 @@ type netTun struct {
ep *channel.Endpoint
stack *stack.Stack
events chan tun.Event
incomingPacket chan *bufferv2.View
incomingPacket chan *buffer.View
mtu int
dnsServers []netip.Addr
hasV4, hasV6 bool
@ -61,7 +61,7 @@ func CreateNetTUN(localAddresses, dnsServers []netip.Addr, mtu int) (tun.Device,
ep: channel.New(1024, uint32(mtu), ""),
stack: stack.New(opts),
events: make(chan tun.Event, 10),
incomingPacket: make(chan *bufferv2.View),
incomingPacket: make(chan *buffer.View),
dnsServers: dnsServers,
mtu: mtu,
}
@ -84,7 +84,7 @@ func CreateNetTUN(localAddresses, dnsServers []netip.Addr, mtu int) (tun.Device,
}
protoAddr := tcpip.ProtocolAddress{
Protocol: protoNumber,
AddressWithPrefix: tcpip.Address(ip.AsSlice()).WithPrefix(),
AddressWithPrefix: tcpip.AddrFromSlice(ip.AsSlice()).WithPrefix(),
}
tcpipErr := dev.stack.AddProtocolAddress(1, protoAddr, stack.AddressProperties{})
if tcpipErr != nil {
@ -140,7 +140,7 @@ func (tun *netTun) Write(buf [][]byte, offset int) (int, error) {
continue
}
pkb := stack.NewPacketBuffer(stack.PacketBufferOptions{Payload: bufferv2.MakeWithData(packet)})
pkb := stack.NewPacketBuffer(stack.PacketBufferOptions{Payload: buffer.MakeWithData(packet)})
switch packet[0] >> 4 {
case 4:
tun.ep.InjectInbound(header.IPv4ProtocolNumber, pkb)
@ -198,7 +198,7 @@ func convertToFullAddr(endpoint netip.AddrPort) (tcpip.FullAddress, tcpip.Networ
}
return tcpip.FullAddress{
NIC: 1,
Addr: tcpip.Address(endpoint.Addr().AsSlice()),
Addr: tcpip.AddrFromSlice(endpoint.Addr().AsSlice()),
Port: endpoint.Port(),
}, protoNumber
}
@ -453,7 +453,7 @@ func (pc *PingConn) ReadFrom(p []byte) (n int, addr net.Addr, err error) {
return 0, nil, fmt.Errorf("ping read: %s", tcpipErr)
}
remoteAddr, _ := netip.AddrFromSlice([]byte(res.RemoteAddr.Addr))
remoteAddr, _ := netip.AddrFromSlice(res.RemoteAddr.Addr.AsSlice())
return res.Count, &PingAddr{remoteAddr}, nil
}

View file

@ -269,11 +269,11 @@ func tcpChecksumValid(pkt []byte, iphLen uint8, isV6 bool) bool {
type coalesceResult int
const (
coalesceInsufficientCap coalesceResult = 0
coalescePSHEnding coalesceResult = 1
coalesceItemInvalidCSum coalesceResult = 2
coalescePktInvalidCSum coalesceResult = 3
coalesceSuccess coalesceResult = 4
coalesceInsufficientCap coalesceResult = iota
coalescePSHEnding
coalesceItemInvalidCSum
coalescePktInvalidCSum
coalesceSuccess
)
// coalesceTCPPackets attempts to coalesce pkt with the packet described by
@ -339,42 +339,6 @@ func coalesceTCPPackets(mode canCoalesce, pkt []byte, pktBuffsIndex int, gsoSize
if gsoSize > item.gsoSize {
item.gsoSize = gsoSize
}
hdr := virtioNetHdr{
flags: unix.VIRTIO_NET_HDR_F_NEEDS_CSUM, // this turns into CHECKSUM_PARTIAL in the skb
hdrLen: uint16(headersLen),
gsoSize: uint16(item.gsoSize),
csumStart: uint16(item.iphLen),
csumOffset: 16,
}
// Recalculate the total len (IPv4) or payload len (IPv6). Recalculate the
// (IPv4) header checksum.
if isV6 {
hdr.gsoType = unix.VIRTIO_NET_HDR_GSO_TCPV6
binary.BigEndian.PutUint16(pktHead[4:], uint16(coalescedLen)-uint16(item.iphLen)) // set new payload len
} else {
hdr.gsoType = unix.VIRTIO_NET_HDR_GSO_TCPV4
pktHead[10], pktHead[11] = 0, 0 // clear checksum field
binary.BigEndian.PutUint16(pktHead[2:], uint16(coalescedLen)) // set new total length
iphCSum := ^checksum(pktHead[:item.iphLen], 0) // compute checksum
binary.BigEndian.PutUint16(pktHead[10:], iphCSum) // set checksum field
}
hdr.encode(bufs[item.bufsIndex][bufsOffset-virtioNetHdrLen:])
// Calculate the pseudo header checksum and place it at the TCP checksum
// offset. Downstream checksum offloading will combine this with computation
// of the tcp header and payload checksum.
addrLen := 4
addrOffset := ipv4SrcAddrOffset
if isV6 {
addrLen = 16
addrOffset = ipv6SrcAddrOffset
}
srcAddrAt := bufsOffset + addrOffset
srcAddr := bufs[item.bufsIndex][srcAddrAt : srcAddrAt+addrLen]
dstAddr := bufs[item.bufsIndex][srcAddrAt+addrLen : srcAddrAt+addrLen*2]
psum := pseudoHeaderChecksumNoFold(unix.IPPROTO_TCP, srcAddr, dstAddr, uint16(coalescedLen-int(item.iphLen)))
binary.BigEndian.PutUint16(pktHead[hdr.csumStart+hdr.csumOffset:], checksum([]byte{}, psum))
item.numMerged++
return coalesceSuccess
@ -390,43 +354,52 @@ const (
maxUint16 = 1<<16 - 1
)
type tcpGROResult int
const (
tcpGROResultNoop tcpGROResult = iota
tcpGROResultTableInsert
tcpGROResultCoalesced
)
// tcpGRO evaluates the TCP packet at pktI in bufs for coalescing with
// existing packets tracked in table. It will return false when pktI is not
// coalesced, otherwise true. This indicates to the caller if bufs[pktI]
// should be written to the Device.
func tcpGRO(bufs [][]byte, offset int, pktI int, table *tcpGROTable, isV6 bool) (pktCoalesced bool) {
// existing packets tracked in table. It returns a tcpGROResultNoop when no
// action was taken, tcpGROResultTableInsert when the evaluated packet was
// inserted into table, and tcpGROResultCoalesced when the evaluated packet was
// coalesced with another packet in table.
func tcpGRO(bufs [][]byte, offset int, pktI int, table *tcpGROTable, isV6 bool) tcpGROResult {
pkt := bufs[pktI][offset:]
if len(pkt) > maxUint16 {
// A valid IPv4 or IPv6 packet will never exceed this.
return false
return tcpGROResultNoop
}
iphLen := int((pkt[0] & 0x0F) * 4)
if isV6 {
iphLen = 40
ipv6HPayloadLen := int(binary.BigEndian.Uint16(pkt[4:]))
if ipv6HPayloadLen != len(pkt)-iphLen {
return false
return tcpGROResultNoop
}
} else {
totalLen := int(binary.BigEndian.Uint16(pkt[2:]))
if totalLen != len(pkt) {
return false
return tcpGROResultNoop
}
}
if len(pkt) < iphLen {
return false
return tcpGROResultNoop
}
tcphLen := int((pkt[iphLen+12] >> 4) * 4)
if tcphLen < 20 || tcphLen > 60 {
return false
return tcpGROResultNoop
}
if len(pkt) < iphLen+tcphLen {
return false
return tcpGROResultNoop
}
if !isV6 {
if pkt[6]&ipv4FlagMoreFragments != 0 || pkt[6]<<3 != 0 || pkt[7] != 0 {
// no GRO support for fragmented segments for now
return false
return tcpGROResultNoop
}
}
tcpFlags := pkt[iphLen+tcpFlagsOffset]
@ -434,14 +407,14 @@ func tcpGRO(bufs [][]byte, offset int, pktI int, table *tcpGROTable, isV6 bool)
// not a candidate if any non-ACK flags (except PSH+ACK) are set
if tcpFlags != tcpFlagACK {
if pkt[iphLen+tcpFlagsOffset] != tcpFlagACK|tcpFlagPSH {
return false
return tcpGROResultNoop
}
pshSet = true
}
gsoSize := uint16(len(pkt) - tcphLen - iphLen)
// not a candidate if payload len is 0
if gsoSize < 1 {
return false
return tcpGROResultNoop
}
seq := binary.BigEndian.Uint32(pkt[iphLen+4:])
srcAddrOffset := ipv4SrcAddrOffset
@ -452,7 +425,7 @@ func tcpGRO(bufs [][]byte, offset int, pktI int, table *tcpGROTable, isV6 bool)
}
items, existing := table.lookupOrInsert(pkt, srcAddrOffset, srcAddrOffset+addrLen, iphLen, tcphLen, pktI)
if !existing {
return false
return tcpGROResultNoop
}
for i := len(items) - 1; i >= 0; i-- {
// In the best case of packets arriving in order iterating in reverse is
@ -470,20 +443,20 @@ func tcpGRO(bufs [][]byte, offset int, pktI int, table *tcpGROTable, isV6 bool)
switch result {
case coalesceSuccess:
table.updateAt(item, i)
return true
return tcpGROResultCoalesced
case coalesceItemInvalidCSum:
// delete the item with an invalid csum
table.deleteAt(item.key, i)
case coalescePktInvalidCSum:
// no point in inserting an item that we can't coalesce
return false
return tcpGROResultNoop
default:
}
}
}
// failed to coalesce with any other packets; store the item in the flow
table.insert(pkt, srcAddrOffset, srcAddrOffset+addrLen, iphLen, tcphLen, pktI)
return false
return tcpGROResultTableInsert
}
func isTCP4NoIPOptions(b []byte) bool {
@ -515,6 +488,64 @@ func isTCP6NoEH(b []byte) bool {
return true
}
// applyCoalesceAccounting updates bufs to account for coalescing based on the
// metadata found in table.
func applyCoalesceAccounting(bufs [][]byte, offset int, table *tcpGROTable, isV6 bool) error {
for _, items := range table.itemsByFlow {
for _, item := range items {
if item.numMerged > 0 {
hdr := virtioNetHdr{
flags: unix.VIRTIO_NET_HDR_F_NEEDS_CSUM, // this turns into CHECKSUM_PARTIAL in the skb
hdrLen: uint16(item.iphLen + item.tcphLen),
gsoSize: item.gsoSize,
csumStart: uint16(item.iphLen),
csumOffset: 16,
}
pkt := bufs[item.bufsIndex][offset:]
// Recalculate the total len (IPv4) or payload len (IPv6).
// Recalculate the (IPv4) header checksum.
if isV6 {
hdr.gsoType = unix.VIRTIO_NET_HDR_GSO_TCPV6
binary.BigEndian.PutUint16(pkt[4:], uint16(len(pkt))-uint16(item.iphLen)) // set new IPv6 header payload len
} else {
hdr.gsoType = unix.VIRTIO_NET_HDR_GSO_TCPV4
pkt[10], pkt[11] = 0, 0
binary.BigEndian.PutUint16(pkt[2:], uint16(len(pkt))) // set new total length
iphCSum := ^checksum(pkt[:item.iphLen], 0) // compute IPv4 header checksum
binary.BigEndian.PutUint16(pkt[10:], iphCSum) // set IPv4 header checksum field
}
err := hdr.encode(bufs[item.bufsIndex][offset-virtioNetHdrLen:])
if err != nil {
return err
}
// Calculate the pseudo header checksum and place it at the TCP
// checksum offset. Downstream checksum offloading will combine
// this with computation of the tcp header and payload checksum.
addrLen := 4
addrOffset := ipv4SrcAddrOffset
if isV6 {
addrLen = 16
addrOffset = ipv6SrcAddrOffset
}
srcAddrAt := offset + addrOffset
srcAddr := bufs[item.bufsIndex][srcAddrAt : srcAddrAt+addrLen]
dstAddr := bufs[item.bufsIndex][srcAddrAt+addrLen : srcAddrAt+addrLen*2]
psum := pseudoHeaderChecksumNoFold(unix.IPPROTO_TCP, srcAddr, dstAddr, uint16(len(pkt)-int(item.iphLen)))
binary.BigEndian.PutUint16(pkt[hdr.csumStart+hdr.csumOffset:], checksum([]byte{}, psum))
} else {
hdr := virtioNetHdr{}
err := hdr.encode(bufs[item.bufsIndex][offset-virtioNetHdrLen:])
if err != nil {
return err
}
}
}
}
return nil
}
// handleGRO evaluates bufs for GRO, and writes the indices of the resulting
// packets into toWrite. toWrite, tcp4Table, and tcp6Table should initially be
// empty (but non-nil), and are passed in to save allocs as the caller may reset
@ -524,23 +555,28 @@ func handleGRO(bufs [][]byte, offset int, tcp4Table, tcp6Table *tcpGROTable, toW
if offset < virtioNetHdrLen || offset > len(bufs[i])-1 {
return errors.New("invalid offset")
}
var coalesced bool
var result tcpGROResult
switch {
case isTCP4NoIPOptions(bufs[i][offset:]): // ipv4 packets w/IP options do not coalesce
coalesced = tcpGRO(bufs, offset, i, tcp4Table, false)
result = tcpGRO(bufs, offset, i, tcp4Table, false)
case isTCP6NoEH(bufs[i][offset:]): // ipv6 packets w/extension headers do not coalesce
coalesced = tcpGRO(bufs, offset, i, tcp6Table, true)
result = tcpGRO(bufs, offset, i, tcp6Table, true)
}
if !coalesced {
switch result {
case tcpGROResultNoop:
hdr := virtioNetHdr{}
err := hdr.encode(bufs[i][offset-virtioNetHdrLen:])
if err != nil {
return err
}
fallthrough
case tcpGROResultTableInsert:
*toWrite = append(*toWrite, i)
}
}
return nil
err4 := applyCoalesceAccounting(bufs, offset, tcp4Table, false)
err6 := applyCoalesceAccounting(bufs, offset, tcp6Table, true)
return errors.Join(err4, err6)
}
// tcpTSO splits packets from in into outBuffs, writing the size of each

View file

@ -35,8 +35,8 @@ func tcp4PacketMutateIPFields(srcIPPort, dstIPPort netip.AddrPort, flags header.
srcAs4 := srcIPPort.Addr().As4()
dstAs4 := dstIPPort.Addr().As4()
ipFields := &header.IPv4Fields{
SrcAddr: tcpip.Address(srcAs4[:]),
DstAddr: tcpip.Address(dstAs4[:]),
SrcAddr: tcpip.AddrFromSlice(srcAs4[:]),
DstAddr: tcpip.AddrFromSlice(dstAs4[:]),
Protocol: unix.IPPROTO_TCP,
TTL: 64,
TotalLength: uint16(totalLen),
@ -72,8 +72,8 @@ func tcp6PacketMutateIPFields(srcIPPort, dstIPPort netip.AddrPort, flags header.
srcAs16 := srcIPPort.Addr().As16()
dstAs16 := dstIPPort.Addr().As16()
ipFields := &header.IPv6Fields{
SrcAddr: tcpip.Address(srcAs16[:]),
DstAddr: tcpip.Address(dstAs16[:]),
SrcAddr: tcpip.AddrFromSlice(srcAs16[:]),
DstAddr: tcpip.AddrFromSlice(dstAs16[:]),
TransportProtocol: unix.IPPROTO_TCP,
HopLimit: 64,
PayloadLength: uint16(segmentSize + 20),

View file

@ -127,6 +127,9 @@ func (tun *NativeTun) MTU() (int, error) {
// TODO: This is a temporary hack. We really need to be monitoring the interface in real time and adapting to MTU changes.
func (tun *NativeTun) ForceMTU(mtu int) {
if tun.close.Load() {
return
}
update := tun.forcedMTU != mtu
tun.forcedMTU = mtu
if update {