summaryrefslogtreecommitdiffhomepage
path: root/pkg/tcpip/transport/tcp/endpoint.go
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/tcpip/transport/tcp/endpoint.go')
-rw-r--r--pkg/tcpip/transport/tcp/endpoint.go217
1 files changed, 135 insertions, 82 deletions
diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go
index dd931f88c..f9d5e0085 100644
--- a/pkg/tcpip/transport/tcp/endpoint.go
+++ b/pkg/tcpip/transport/tcp/endpoint.go
@@ -15,6 +15,7 @@
package tcp
import (
+ "encoding/binary"
"fmt"
"math"
"strings"
@@ -26,6 +27,7 @@ import (
"gvisor.dev/gvisor/pkg/sleep"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
+ "gvisor.dev/gvisor/pkg/tcpip/hash/jenkins"
"gvisor.dev/gvisor/pkg/tcpip/header"
"gvisor.dev/gvisor/pkg/tcpip/iptables"
"gvisor.dev/gvisor/pkg/tcpip/seqnum"
@@ -280,6 +282,9 @@ type endpoint struct {
// reusePort is set to true if SO_REUSEPORT is enabled.
reusePort bool
+ // bindToDevice is set to the NIC on which to bind or disabled if 0.
+ bindToDevice tcpip.NICID
+
// delay enables Nagle's algorithm.
//
// delay is a boolean (0 is false) and must be accessed atomically.
@@ -564,11 +569,11 @@ func (e *endpoint) Close() {
// in Listen() when trying to register.
if e.state == StateListen && e.isPortReserved {
if e.isRegistered {
- e.stack.UnregisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e)
+ e.stack.UnregisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.bindToDevice)
e.isRegistered = false
}
- e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort)
+ e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort, e.bindToDevice)
e.isPortReserved = false
}
@@ -625,12 +630,12 @@ func (e *endpoint) cleanupLocked() {
e.workerCleanup = false
if e.isRegistered {
- e.stack.UnregisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e)
+ e.stack.UnregisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.bindToDevice)
e.isRegistered = false
}
if e.isPortReserved {
- e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort)
+ e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort, e.bindToDevice)
e.isPortReserved = false
}
@@ -952,62 +957,9 @@ func (e *endpoint) zeroReceiveWindow(scale uint8) bool {
return ((e.rcvBufSize - e.rcvBufUsed) >> scale) == 0
}
-// SetSockOpt sets a socket option.
-func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
- switch v := opt.(type) {
- case tcpip.DelayOption:
- if v == 0 {
- atomic.StoreUint32(&e.delay, 0)
-
- // Handle delayed data.
- e.sndWaker.Assert()
- } else {
- atomic.StoreUint32(&e.delay, 1)
- }
- return nil
-
- case tcpip.CorkOption:
- if v == 0 {
- atomic.StoreUint32(&e.cork, 0)
-
- // Handle the corked data.
- e.sndWaker.Assert()
- } else {
- atomic.StoreUint32(&e.cork, 1)
- }
- return nil
-
- case tcpip.ReuseAddressOption:
- e.mu.Lock()
- e.reuseAddr = v != 0
- e.mu.Unlock()
- return nil
-
- case tcpip.ReusePortOption:
- e.mu.Lock()
- e.reusePort = v != 0
- e.mu.Unlock()
- return nil
-
- case tcpip.QuickAckOption:
- if v == 0 {
- atomic.StoreUint32(&e.slowAck, 1)
- } else {
- atomic.StoreUint32(&e.slowAck, 0)
- }
- return nil
-
- case tcpip.MaxSegOption:
- userMSS := v
- if userMSS < header.TCPMinimumMSS || userMSS > header.TCPMaximumMSS {
- return tcpip.ErrInvalidOptionValue
- }
- e.mu.Lock()
- e.userMSS = int(userMSS)
- e.mu.Unlock()
- e.notifyProtocolGoroutine(notifyMSSChanged)
- return nil
-
+// SetSockOptInt sets a socket option.
+func (e *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error {
+ switch opt {
case tcpip.ReceiveBufferSizeOption:
// Make sure the receive buffer size is within the min and max
// allowed.
@@ -1071,6 +1023,82 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
e.sndBufMu.Unlock()
return nil
+ default:
+ return nil
+ }
+}
+
+// SetSockOpt sets a socket option.
+func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
+ switch v := opt.(type) {
+ case tcpip.DelayOption:
+ if v == 0 {
+ atomic.StoreUint32(&e.delay, 0)
+
+ // Handle delayed data.
+ e.sndWaker.Assert()
+ } else {
+ atomic.StoreUint32(&e.delay, 1)
+ }
+ return nil
+
+ case tcpip.CorkOption:
+ if v == 0 {
+ atomic.StoreUint32(&e.cork, 0)
+
+ // Handle the corked data.
+ e.sndWaker.Assert()
+ } else {
+ atomic.StoreUint32(&e.cork, 1)
+ }
+ return nil
+
+ case tcpip.ReuseAddressOption:
+ e.mu.Lock()
+ e.reuseAddr = v != 0
+ e.mu.Unlock()
+ return nil
+
+ case tcpip.ReusePortOption:
+ e.mu.Lock()
+ e.reusePort = v != 0
+ e.mu.Unlock()
+ return nil
+
+ case tcpip.BindToDeviceOption:
+ e.mu.Lock()
+ defer e.mu.Unlock()
+ if v == "" {
+ e.bindToDevice = 0
+ return nil
+ }
+ for nicid, nic := range e.stack.NICInfo() {
+ if nic.Name == string(v) {
+ e.bindToDevice = nicid
+ return nil
+ }
+ }
+ return tcpip.ErrUnknownDevice
+
+ case tcpip.QuickAckOption:
+ if v == 0 {
+ atomic.StoreUint32(&e.slowAck, 1)
+ } else {
+ atomic.StoreUint32(&e.slowAck, 0)
+ }
+ return nil
+
+ case tcpip.MaxSegOption:
+ userMSS := v
+ if userMSS < header.TCPMinimumMSS || userMSS > header.TCPMaximumMSS {
+ return tcpip.ErrInvalidOptionValue
+ }
+ e.mu.Lock()
+ e.userMSS = int(userMSS)
+ e.mu.Unlock()
+ e.notifyProtocolGoroutine(notifyMSSChanged)
+ return nil
+
case tcpip.V6OnlyOption:
// We only recognize this option on v6 endpoints.
if e.netProto != header.IPv6ProtocolNumber {
@@ -1182,6 +1210,18 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) {
switch opt {
case tcpip.ReceiveQueueSizeOption:
return e.readyReceiveSize()
+ case tcpip.SendBufferSizeOption:
+ e.sndBufMu.Lock()
+ v := e.sndBufSize
+ e.sndBufMu.Unlock()
+ return v, nil
+
+ case tcpip.ReceiveBufferSizeOption:
+ e.rcvListMu.Lock()
+ v := e.rcvBufSize
+ e.rcvListMu.Unlock()
+ return v, nil
+
}
return -1, tcpip.ErrUnknownProtocolOption
}
@@ -1204,18 +1244,6 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
*o = header.TCPDefaultMSS
return nil
- case *tcpip.SendBufferSizeOption:
- e.sndBufMu.Lock()
- *o = tcpip.SendBufferSizeOption(e.sndBufSize)
- e.sndBufMu.Unlock()
- return nil
-
- case *tcpip.ReceiveBufferSizeOption:
- e.rcvListMu.Lock()
- *o = tcpip.ReceiveBufferSizeOption(e.rcvBufSize)
- e.rcvListMu.Unlock()
- return nil
-
case *tcpip.DelayOption:
*o = 0
if v := atomic.LoadUint32(&e.delay); v != 0 {
@@ -1252,6 +1280,16 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
}
return nil
+ case *tcpip.BindToDeviceOption:
+ e.mu.RLock()
+ defer e.mu.RUnlock()
+ if nic, ok := e.stack.NICInfo()[e.bindToDevice]; ok {
+ *o = tcpip.BindToDeviceOption(nic.Name)
+ return nil
+ }
+ *o = ""
+ return nil
+
case *tcpip.QuickAckOption:
*o = 1
if v := atomic.LoadUint32(&e.slowAck); v != 0 {
@@ -1458,7 +1496,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (er
if e.id.LocalPort != 0 {
// The endpoint is bound to a port, attempt to register it.
- err := e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, e.id, e, e.reusePort)
+ err := e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, e.id, e, e.reusePort, e.bindToDevice)
if err != nil {
return err
}
@@ -1468,17 +1506,32 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (er
// address/port for both local and remote (otherwise this
// endpoint would be trying to connect to itself).
sameAddr := e.id.LocalAddress == e.id.RemoteAddress
- if _, err := e.stack.PickEphemeralPort(func(p uint16) (bool, *tcpip.Error) {
+
+ // Calculate a port offset based on the destination IP/port and
+ // src IP to ensure that for a given tuple (srcIP, destIP,
+ // destPort) the offset used as a starting point is the same to
+ // ensure that we can cycle through the port space effectively.
+ h := jenkins.Sum32(e.stack.PortSeed())
+ h.Write([]byte(e.id.LocalAddress))
+ h.Write([]byte(e.id.RemoteAddress))
+ portBuf := make([]byte, 2)
+ binary.LittleEndian.PutUint16(portBuf, e.id.RemotePort)
+ h.Write(portBuf)
+ portOffset := h.Sum32()
+
+ if _, err := e.stack.PickEphemeralPortStable(portOffset, func(p uint16) (bool, *tcpip.Error) {
if sameAddr && p == e.id.RemotePort {
return false, nil
}
- if !e.stack.IsPortAvailable(netProtos, ProtocolNumber, e.id.LocalAddress, p, false) {
+ // reusePort is false below because connect cannot reuse a port even if
+ // reusePort was set.
+ if !e.stack.IsPortAvailable(netProtos, ProtocolNumber, e.id.LocalAddress, p, false /* reusePort */, e.bindToDevice) {
return false, nil
}
id := e.id
id.LocalPort = p
- switch e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, id, e, e.reusePort) {
+ switch e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, id, e, e.reusePort, e.bindToDevice) {
case nil:
e.id = id
return true, nil
@@ -1496,7 +1549,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (er
// before Connect: in such a case we don't want to hold on to
// reservations anymore.
if e.isPortReserved {
- e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, origID.LocalAddress, origID.LocalPort)
+ e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, origID.LocalAddress, origID.LocalPort, e.bindToDevice)
e.isPortReserved = false
}
@@ -1640,7 +1693,7 @@ func (e *endpoint) Listen(backlog int) (err *tcpip.Error) {
}
// Register the endpoint.
- if err := e.stack.RegisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.reusePort); err != nil {
+ if err := e.stack.RegisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.reusePort, e.bindToDevice); err != nil {
return err
}
@@ -1721,7 +1774,7 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) (err *tcpip.Error) {
}
}
- port, err := e.stack.ReservePort(netProtos, ProtocolNumber, addr.Addr, addr.Port, e.reusePort)
+ port, err := e.stack.ReservePort(netProtos, ProtocolNumber, addr.Addr, addr.Port, e.reusePort, e.bindToDevice)
if err != nil {
return err
}
@@ -1731,16 +1784,16 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) (err *tcpip.Error) {
e.id.LocalPort = port
// Any failures beyond this point must remove the port registration.
- defer func() {
+ defer func(bindToDevice tcpip.NICID) {
if err != nil {
- e.stack.ReleasePort(netProtos, ProtocolNumber, addr.Addr, port)
+ e.stack.ReleasePort(netProtos, ProtocolNumber, addr.Addr, port, bindToDevice)
e.isPortReserved = false
e.effectiveNetProtos = nil
e.id.LocalPort = 0
e.id.LocalAddress = ""
e.boundNICID = 0
}
- }()
+ }(e.bindToDevice)
// If an address is specified, we must ensure that it's one of our
// local addresses.