summaryrefslogtreecommitdiffhomepage
path: root/pkg/tcpip
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/tcpip')
-rw-r--r--pkg/tcpip/link/sniffer/sniffer.go8
-rw-r--r--pkg/tcpip/stack/dhcpv6configurationfromndpra_string.go11
-rw-r--r--pkg/tcpip/stack/ndp.go9
-rw-r--r--pkg/tcpip/stack/ndp_test.go24
-rw-r--r--pkg/tcpip/transport/tcp/snd.go94
-rw-r--r--pkg/tcpip/transport/tcp/tcp_test.go17
6 files changed, 140 insertions, 23 deletions
diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go
index 0799c8f4d..da1c520ae 100644
--- a/pkg/tcpip/link/sniffer/sniffer.go
+++ b/pkg/tcpip/link/sniffer/sniffer.go
@@ -391,7 +391,7 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.P
break
}
udp := header.UDP(hdr)
- if fragmentOffset == 0 && len(udp) >= header.UDPMinimumSize {
+ if fragmentOffset == 0 {
srcPort = udp.SourcePort()
dstPort = udp.DestinationPort()
details = fmt.Sprintf("xsum: 0x%x", udp.Checksum())
@@ -405,14 +405,14 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.P
break
}
tcp := header.TCP(hdr)
- if fragmentOffset == 0 && len(tcp) >= header.TCPMinimumSize {
+ if fragmentOffset == 0 {
offset := int(tcp.DataOffset())
if offset < header.TCPMinimumSize {
details += fmt.Sprintf("invalid packet: tcp data offset too small %d", offset)
break
}
- if offset > len(tcp) && !moreFragments {
- details += fmt.Sprintf("invalid packet: tcp data offset %d larger than packet buffer length %d", offset, len(tcp))
+ if offset > vv.Size() && !moreFragments {
+ details += fmt.Sprintf("invalid packet: tcp data offset %d larger than packet buffer length %d", offset, vv.Size())
break
}
diff --git a/pkg/tcpip/stack/dhcpv6configurationfromndpra_string.go b/pkg/tcpip/stack/dhcpv6configurationfromndpra_string.go
index 8b4213eec..d199ded6a 100644
--- a/pkg/tcpip/stack/dhcpv6configurationfromndpra_string.go
+++ b/pkg/tcpip/stack/dhcpv6configurationfromndpra_string.go
@@ -12,7 +12,7 @@
// See the License for the specific language governing permissions and
// limitations under the License.
-// Code generated by "stringer -type=DHCPv6ConfigurationFromNDPRA"; DO NOT EDIT.
+// Code generated by "stringer -type DHCPv6ConfigurationFromNDPRA"; DO NOT EDIT.
package stack
@@ -22,9 +22,9 @@ func _() {
// An "invalid array index" compiler error signifies that the constant values have changed.
// Re-run the stringer command to generate them again.
var x [1]struct{}
- _ = x[DHCPv6NoConfiguration-0]
- _ = x[DHCPv6ManagedAddress-1]
- _ = x[DHCPv6OtherConfigurations-2]
+ _ = x[DHCPv6NoConfiguration-1]
+ _ = x[DHCPv6ManagedAddress-2]
+ _ = x[DHCPv6OtherConfigurations-3]
}
const _DHCPv6ConfigurationFromNDPRA_name = "DHCPv6NoConfigurationDHCPv6ManagedAddressDHCPv6OtherConfigurations"
@@ -32,8 +32,9 @@ const _DHCPv6ConfigurationFromNDPRA_name = "DHCPv6NoConfigurationDHCPv6ManagedAd
var _DHCPv6ConfigurationFromNDPRA_index = [...]uint8{0, 21, 41, 66}
func (i DHCPv6ConfigurationFromNDPRA) String() string {
+ i -= 1
if i < 0 || i >= DHCPv6ConfigurationFromNDPRA(len(_DHCPv6ConfigurationFromNDPRA_index)-1) {
- return "DHCPv6ConfigurationFromNDPRA(" + strconv.FormatInt(int64(i), 10) + ")"
+ return "DHCPv6ConfigurationFromNDPRA(" + strconv.FormatInt(int64(i+1), 10) + ")"
}
return _DHCPv6ConfigurationFromNDPRA_name[_DHCPv6ConfigurationFromNDPRA_index[i]:_DHCPv6ConfigurationFromNDPRA_index[i+1]]
}
diff --git a/pkg/tcpip/stack/ndp.go b/pkg/tcpip/stack/ndp.go
index 15343acbc..526c7d6ff 100644
--- a/pkg/tcpip/stack/ndp.go
+++ b/pkg/tcpip/stack/ndp.go
@@ -199,9 +199,11 @@ var (
type DHCPv6ConfigurationFromNDPRA int
const (
+ _ DHCPv6ConfigurationFromNDPRA = iota
+
// DHCPv6NoConfiguration indicates that no configurations are available via
// DHCPv6.
- DHCPv6NoConfiguration DHCPv6ConfigurationFromNDPRA = iota
+ DHCPv6NoConfiguration
// DHCPv6ManagedAddress indicates that addresses are available via DHCPv6.
//
@@ -315,9 +317,6 @@ type NDPDispatcher interface {
// OnDHCPv6Configuration will be called with an updated configuration that is
// available via DHCPv6 for a specified NIC.
//
- // NDPDispatcher assumes that the initial configuration available by DHCPv6 is
- // DHCPv6NoConfiguration.
- //
// This function is not permitted to block indefinitely. It must not
// call functions on the stack itself.
OnDHCPv6Configuration(tcpip.NICID, DHCPv6ConfigurationFromNDPRA)
@@ -1808,6 +1807,8 @@ func (ndp *ndpState) cleanupState(hostOnly bool) {
if got := len(ndp.defaultRouters); got != 0 {
panic(fmt.Sprintf("ndp: still have discovered default routers after cleaning up; found = %d", got))
}
+
+ ndp.dhcpv6Configuration = 0
}
// startSolicitingRouters starts soliciting routers, as per RFC 4861 section
diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go
index 67f012840..b3d174cdd 100644
--- a/pkg/tcpip/stack/ndp_test.go
+++ b/pkg/tcpip/stack/ndp_test.go
@@ -4888,7 +4888,12 @@ func TestDHCPv6ConfigurationFromNDPDA(t *testing.T) {
}
}
- // The initial DHCPv6 configuration should be stack.DHCPv6NoConfiguration.
+ // Even if the first RA reports no DHCPv6 configurations are available, the
+ // dispatcher should get an event.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, false))
+ expectDHCPv6Event(stack.DHCPv6NoConfiguration)
+ // Receiving the same update again should not result in an event to the
+ // dispatcher.
e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, false))
expectNoDHCPv6Event()
@@ -4896,8 +4901,6 @@ func TestDHCPv6ConfigurationFromNDPDA(t *testing.T) {
// Configurations.
e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true))
expectDHCPv6Event(stack.DHCPv6OtherConfigurations)
- // Receiving the same update again should not result in an event to the
- // NDPDispatcher.
e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true))
expectNoDHCPv6Event()
@@ -4933,6 +4936,21 @@ func TestDHCPv6ConfigurationFromNDPDA(t *testing.T) {
expectDHCPv6Event(stack.DHCPv6OtherConfigurations)
e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true))
expectNoDHCPv6Event()
+
+ // Cycling the NIC should cause the last DHCPv6 configuration to be cleared.
+ if err := s.DisableNIC(nicID); err != nil {
+ t.Fatalf("s.DisableNIC(%d): %s", nicID, err)
+ }
+ if err := s.EnableNIC(nicID); err != nil {
+ t.Fatalf("s.EnableNIC(%d): %s", nicID, err)
+ }
+
+ // Receive an RA that updates the DHCPv6 configuration to Other
+ // Configurations.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true))
+ expectDHCPv6Event(stack.DHCPv6OtherConfigurations)
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true))
+ expectNoDHCPv6Event()
}
// TestRouterSolicitation tests the initial Router Solicitations that are sent
diff --git a/pkg/tcpip/transport/tcp/snd.go b/pkg/tcpip/transport/tcp/snd.go
index d8cfe3115..a3018914b 100644
--- a/pkg/tcpip/transport/tcp/snd.go
+++ b/pkg/tcpip/transport/tcp/snd.go
@@ -41,6 +41,10 @@ const (
// nDupAckThreshold is the number of duplicate ACK's required
// before fast-retransmit is entered.
nDupAckThreshold = 3
+
+ // MaxRetries is the maximum number of probe retries sender does
+ // before timing out the connection, Linux default TCP_RETR2.
+ MaxRetries = 15
)
// ccState indicates the current congestion control state for this sender.
@@ -138,6 +142,14 @@ type sender struct {
// the first segment that was retransmitted due to RTO expiration.
firstRetransmittedSegXmitTime time.Time `state:".(unixTime)"`
+ // zeroWindowProbing is set if the sender is currently probing
+ // for zero receive window.
+ zeroWindowProbing bool `state:"nosave"`
+
+ // unackZeroWindowProbes is the number of unacknowledged zero
+ // window probes.
+ unackZeroWindowProbes uint32 `state:"nosave"`
+
closed bool
writeNext *segment
writeList segmentList
@@ -479,10 +491,24 @@ func (s *sender) retransmitTimerExpired() bool {
remaining = uto - elapsed
}
- if remaining <= 0 || s.rto >= MaxRTO {
+ // Always honor the user-timeout irrespective of whether the zero
+ // window probes were acknowledged.
+ // net/ipv4/tcp_timer.c::tcp_probe_timer()
+ if remaining <= 0 || s.unackZeroWindowProbes >= MaxRetries {
return false
}
+ if s.rto >= MaxRTO {
+ // RFC 1122 section: 4.2.2.17
+ // A TCP MAY keep its offered receive window closed
+ // indefinitely. As long as the receiving TCP continues to
+ // send acknowledgments in response to the probe segments, the
+ // sending TCP MUST allow the connection to stay open.
+ if !(s.zeroWindowProbing && s.unackZeroWindowProbes == 0) {
+ return false
+ }
+ }
+
// Set new timeout. The timer will be restarted by the call to sendData
// below.
s.rto *= 2
@@ -533,6 +559,15 @@ func (s *sender) retransmitTimerExpired() bool {
// information is usable after an RTO.
s.ep.scoreboard.Reset()
s.writeNext = s.writeList.Front()
+
+ // RFC 1122 4.2.2.17: Start sending zero window probes when we still see a
+ // zero receive window after retransmission interval and we have data to
+ // send.
+ if s.zeroWindowProbing {
+ s.sendZeroWindowProbe()
+ return true
+ }
+
s.sendData()
return true
@@ -827,6 +862,34 @@ func (s *sender) handleSACKRecovery(limit int, end seqnum.Value) (dataSent bool)
return dataSent
}
+func (s *sender) sendZeroWindowProbe() {
+ ack, win := s.ep.rcv.getSendParams()
+ s.unackZeroWindowProbes++
+ // Send a zero window probe with sequence number pointing to
+ // the last acknowledged byte.
+ s.ep.sendRaw(buffer.VectorisedView{}, header.TCPFlagAck, s.sndUna-1, ack, win)
+ // Rearm the timer to continue probing.
+ s.resendTimer.enable(s.rto)
+}
+
+func (s *sender) enableZeroWindowProbing() {
+ s.zeroWindowProbing = true
+ // We piggyback the probing on the retransmit timer with the
+ // current retranmission interval, as we may start probing while
+ // segment retransmissions.
+ if s.firstRetransmittedSegXmitTime.IsZero() {
+ s.firstRetransmittedSegXmitTime = time.Now()
+ }
+ s.resendTimer.enable(s.rto)
+}
+
+func (s *sender) disableZeroWindowProbing() {
+ s.zeroWindowProbing = false
+ s.unackZeroWindowProbes = 0
+ s.firstRetransmittedSegXmitTime = time.Time{}
+ s.resendTimer.disable()
+}
+
// sendData sends new data segments. It is called when data becomes available or
// when the send window opens up.
func (s *sender) sendData() {
@@ -875,6 +938,13 @@ func (s *sender) sendData() {
s.ep.disableKeepaliveTimer()
}
+ // If the sender has advertized zero receive window and we have
+ // data to be sent out, start zero window probing to query the
+ // the remote for it's receive window size.
+ if s.writeNext != nil && s.sndWnd == 0 {
+ s.enableZeroWindowProbing()
+ }
+
// Enable the timer if we have pending data and it's not enabled yet.
if !s.resendTimer.enabled() && s.sndUna != s.sndNxt {
s.resendTimer.enable(s.rto)
@@ -1122,8 +1192,26 @@ func (s *sender) handleRcvdSegment(seg *segment) {
// Stash away the current window size.
s.sndWnd = seg.window
- // Ignore ack if it doesn't acknowledge any new data.
ack := seg.ackNumber
+
+ // Disable zero window probing if remote advertizes a non-zero receive
+ // window. This can be with an ACK to the zero window probe (where the
+ // acknumber refers to the already acknowledged byte) OR to any previously
+ // unacknowledged segment.
+ if s.zeroWindowProbing && seg.window > 0 &&
+ (ack == s.sndUna || (ack-1).InRange(s.sndUna, s.sndNxt)) {
+ s.disableZeroWindowProbing()
+ }
+
+ // On receiving the ACK for the zero window probe, account for it and
+ // skip trying to send any segment as we are still probing for
+ // receive window to become non-zero.
+ if s.zeroWindowProbing && s.unackZeroWindowProbes > 0 && ack == s.sndUna {
+ s.unackZeroWindowProbes--
+ return
+ }
+
+ // Ignore ack if it doesn't acknowledge any new data.
if (ack - 1).InRange(s.sndUna, s.sndNxt) {
s.dupAckCount = 0
@@ -1143,7 +1231,7 @@ func (s *sender) handleRcvdSegment(seg *segment) {
}
// When an ack is received we must rearm the timer.
- // RFC 6298 5.2
+ // RFC 6298 5.3
s.resendTimer.enable(s.rto)
// Remove all acknowledged data from the write list.
diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go
index 33e2b9a09..49e4ba214 100644
--- a/pkg/tcpip/transport/tcp/tcp_test.go
+++ b/pkg/tcpip/transport/tcp/tcp_test.go
@@ -1900,7 +1900,7 @@ func TestZeroWindowSend(t *testing.T) {
c := context.New(t, defaultMTU)
defer c.Cleanup()
- c.CreateConnected(789, 0, -1 /* epRcvBuf */)
+ c.CreateConnected(789 /* iss */, 0 /* rcvWnd */, -1 /* epRcvBuf */)
data := []byte{1, 2, 3}
view := buffer.NewView(len(data))
@@ -1911,8 +1911,17 @@ func TestZeroWindowSend(t *testing.T) {
t.Fatalf("Write failed: %v", err)
}
- // Since the window is currently zero, check that no packet is received.
- c.CheckNoPacket("Packet received when window is zero")
+ // Check if we got a zero-window probe.
+ b := c.GetPacket()
+ checker.IPv4(t, b,
+ checker.PayloadLen(header.TCPMinimumSize),
+ checker.TCP(
+ checker.DstPort(context.TestPort),
+ checker.SeqNum(uint32(c.IRS)),
+ checker.AckNum(790),
+ checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)),
+ ),
+ )
// Open up the window. Data should be received now.
c.SendPacket(nil, &context.Headers{
@@ -1925,7 +1934,7 @@ func TestZeroWindowSend(t *testing.T) {
})
// Check that data is received.
- b := c.GetPacket()
+ b = c.GetPacket()
checker.IPv4(t, b,
checker.PayloadLen(len(data)+header.TCPMinimumSize),
checker.TCP(
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-11 10:36:08 +0000