summaryrefslogtreecommitdiffhomepage
path: root/pkg/tcpip
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/tcpip')
-rw-r--r--pkg/tcpip/BUILD9
-rw-r--r--pkg/tcpip/adapters/gonet/BUILD1
-rw-r--r--pkg/tcpip/adapters/gonet/gonet.go2
-rw-r--r--pkg/tcpip/iptables/iptables.go8
-rw-r--r--pkg/tcpip/iptables/targets.go2
-rw-r--r--pkg/tcpip/link/fdbased/BUILD1
-rw-r--r--pkg/tcpip/link/fdbased/endpoint.go2
-rw-r--r--pkg/tcpip/link/sharedmem/BUILD2
-rw-r--r--pkg/tcpip/link/sharedmem/pipe/BUILD1
-rw-r--r--pkg/tcpip/link/sharedmem/pipe/pipe_test.go3
-rw-r--r--pkg/tcpip/link/sharedmem/sharedmem.go2
-rw-r--r--pkg/tcpip/link/sharedmem/sharedmem_test.go2
-rw-r--r--pkg/tcpip/network/arp/arp.go6
-rw-r--r--pkg/tcpip/network/fragmentation/BUILD1
-rw-r--r--pkg/tcpip/network/fragmentation/fragmentation.go2
-rw-r--r--pkg/tcpip/network/fragmentation/reassembler.go2
-rw-r--r--pkg/tcpip/network/ip_test.go4
-rw-r--r--pkg/tcpip/network/ipv4/ipv4.go18
-rw-r--r--pkg/tcpip/network/ipv6/ipv6.go14
-rw-r--r--pkg/tcpip/ports/BUILD1
-rw-r--r--pkg/tcpip/ports/ports.go2
-rw-r--r--pkg/tcpip/stack/BUILD2
-rw-r--r--pkg/tcpip/stack/linkaddrcache.go2
-rw-r--r--pkg/tcpip/stack/linkaddrcache_test.go2
-rw-r--r--pkg/tcpip/stack/ndp.go411
-rw-r--r--pkg/tcpip/stack/ndp_test.go198
-rw-r--r--pkg/tcpip/stack/nic.go14
-rw-r--r--pkg/tcpip/stack/registration.go6
-rw-r--r--pkg/tcpip/stack/route.go6
-rw-r--r--pkg/tcpip/stack/stack.go70
-rw-r--r--pkg/tcpip/stack/stack_test.go60
-rw-r--r--pkg/tcpip/stack/transport_demuxer.go2
-rw-r--r--pkg/tcpip/stack/transport_demuxer_test.go92
-rw-r--r--pkg/tcpip/stack/transport_test.go14
-rw-r--r--pkg/tcpip/tcpip.go35
-rw-r--r--pkg/tcpip/timer.go161
-rw-r--r--pkg/tcpip/timer_test.go236
-rw-r--r--pkg/tcpip/transport/icmp/BUILD1
-rw-r--r--pkg/tcpip/transport/icmp/endpoint.go17
-rw-r--r--pkg/tcpip/transport/packet/BUILD1
-rw-r--r--pkg/tcpip/transport/packet/endpoint.go25
-rw-r--r--pkg/tcpip/transport/raw/BUILD1
-rw-r--r--pkg/tcpip/transport/raw/endpoint.go47
-rw-r--r--pkg/tcpip/transport/tcp/BUILD1
-rw-r--r--pkg/tcpip/transport/tcp/accept.go2
-rw-r--r--pkg/tcpip/transport/tcp/connect.go2
-rw-r--r--pkg/tcpip/transport/tcp/dual_stack_test.go8
-rw-r--r--pkg/tcpip/transport/tcp/endpoint.go182
-rw-r--r--pkg/tcpip/transport/tcp/endpoint_state.go2
-rw-r--r--pkg/tcpip/transport/tcp/forwarder.go3
-rw-r--r--pkg/tcpip/transport/tcp/protocol.go2
-rw-r--r--pkg/tcpip/transport/tcp/rcv.go6
-rw-r--r--pkg/tcpip/transport/tcp/segment_queue.go2
-rw-r--r--pkg/tcpip/transport/tcp/snd.go9
-rw-r--r--pkg/tcpip/transport/tcp/tcp_test.go200
-rw-r--r--pkg/tcpip/transport/tcp/testing/context/context.go16
-rw-r--r--pkg/tcpip/transport/udp/BUILD1
-rw-r--r--pkg/tcpip/transport/udp/endpoint.go92
-rw-r--r--pkg/tcpip/transport/udp/udp_test.go38
59 files changed, 1357 insertions, 697 deletions
diff --git a/pkg/tcpip/BUILD b/pkg/tcpip/BUILD
index 36bc3a63b..23e4b09e7 100644
--- a/pkg/tcpip/BUILD
+++ b/pkg/tcpip/BUILD
@@ -10,10 +10,12 @@ go_library(
"packet_buffer_state.go",
"tcpip.go",
"time_unsafe.go",
+ "timer.go",
],
importpath = "gvisor.dev/gvisor/pkg/tcpip",
visibility = ["//visibility:public"],
deps = [
+ "//pkg/sync",
"//pkg/tcpip/buffer",
"//pkg/waiter",
],
@@ -25,3 +27,10 @@ go_test(
srcs = ["tcpip_test.go"],
embed = [":tcpip"],
)
+
+go_test(
+ name = "tcpip_x_test",
+ size = "small",
+ srcs = ["timer_test.go"],
+ deps = [":tcpip"],
+)
diff --git a/pkg/tcpip/adapters/gonet/BUILD b/pkg/tcpip/adapters/gonet/BUILD
index 78df5a0b1..3df7d18d3 100644
--- a/pkg/tcpip/adapters/gonet/BUILD
+++ b/pkg/tcpip/adapters/gonet/BUILD
@@ -9,6 +9,7 @@ go_library(
importpath = "gvisor.dev/gvisor/pkg/tcpip/adapters/gonet",
visibility = ["//visibility:public"],
deps = [
+ "//pkg/sync",
"//pkg/tcpip",
"//pkg/tcpip/buffer",
"//pkg/tcpip/stack",
diff --git a/pkg/tcpip/adapters/gonet/gonet.go b/pkg/tcpip/adapters/gonet/gonet.go
index cd6ce930a..a2f44b496 100644
--- a/pkg/tcpip/adapters/gonet/gonet.go
+++ b/pkg/tcpip/adapters/gonet/gonet.go
@@ -20,9 +20,9 @@ import (
"errors"
"io"
"net"
- "sync"
"time"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/stack"
diff --git a/pkg/tcpip/iptables/iptables.go b/pkg/tcpip/iptables/iptables.go
index 8a72feb77..83d807a4d 100644
--- a/pkg/tcpip/iptables/iptables.go
+++ b/pkg/tcpip/iptables/iptables.go
@@ -22,13 +22,13 @@ import (
"gvisor.dev/gvisor/pkg/tcpip"
)
+// Table names.
const (
TablenameNat = "nat"
TablenameMangle = "mangle"
TablenameFilter = "filter"
)
-// TODO: Make this an iota? Faster! Do it.
// Chain names as defined by net/ipv4/netfilter/ip_tables.c.
const (
ChainNamePrerouting = "PREROUTING"
@@ -38,11 +38,15 @@ const (
ChainNamePostrouting = "POSTROUTING"
)
+// HookUnset indicates that there is no hook set for an entrypoint or
+// underflow.
const HookUnset = -1
// DefaultTables returns a default set of tables. Each chain is set to accept
// all packets.
func DefaultTables() IPTables {
+ // TODO(gvisor.dev/issue/170): We may be able to swap out some strings for
+ // iotas.
return IPTables{
Tables: map[string]Table{
TablenameNat: Table{
@@ -111,6 +115,8 @@ func DefaultTables() IPTables {
}
}
+// EmptyFilterTable returns a Table with no rules and the filter table chains
+// mapped to HookUnset.
func EmptyFilterTable() Table {
return Table{
Rules: []Rule{},
diff --git a/pkg/tcpip/iptables/targets.go b/pkg/tcpip/iptables/targets.go
index 8180dcefd..4dd281371 100644
--- a/pkg/tcpip/iptables/targets.go
+++ b/pkg/tcpip/iptables/targets.go
@@ -41,7 +41,7 @@ func (UnconditionalDropTarget) Action(packet tcpip.PacketBuffer) (Verdict, strin
// should be unreachable.
type ErrorTarget struct{}
-// Actions implements Target.Action.
+// Action implements Target.Action.
func (ErrorTarget) Action(packet tcpip.PacketBuffer) (Verdict, string) {
log.Warningf("ErrorTarget triggered.")
return Drop, ""
diff --git a/pkg/tcpip/link/fdbased/BUILD b/pkg/tcpip/link/fdbased/BUILD
index 897c94821..66cc53ed4 100644
--- a/pkg/tcpip/link/fdbased/BUILD
+++ b/pkg/tcpip/link/fdbased/BUILD
@@ -16,6 +16,7 @@ go_library(
importpath = "gvisor.dev/gvisor/pkg/tcpip/link/fdbased",
visibility = ["//visibility:public"],
deps = [
+ "//pkg/sync",
"//pkg/tcpip",
"//pkg/tcpip/buffer",
"//pkg/tcpip/header",
diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go
index fa8a703d9..b7f60178e 100644
--- a/pkg/tcpip/link/fdbased/endpoint.go
+++ b/pkg/tcpip/link/fdbased/endpoint.go
@@ -41,10 +41,10 @@ package fdbased
import (
"fmt"
- "sync"
"syscall"
"golang.org/x/sys/unix"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/header"
diff --git a/pkg/tcpip/link/sharedmem/BUILD b/pkg/tcpip/link/sharedmem/BUILD
index a4f9cdd69..09165dd4c 100644
--- a/pkg/tcpip/link/sharedmem/BUILD
+++ b/pkg/tcpip/link/sharedmem/BUILD
@@ -15,6 +15,7 @@ go_library(
visibility = ["//visibility:public"],
deps = [
"//pkg/log",
+ "//pkg/sync",
"//pkg/tcpip",
"//pkg/tcpip/buffer",
"//pkg/tcpip/header",
@@ -31,6 +32,7 @@ go_test(
],
embed = [":sharedmem"],
deps = [
+ "//pkg/sync",
"//pkg/tcpip",
"//pkg/tcpip/buffer",
"//pkg/tcpip/header",
diff --git a/pkg/tcpip/link/sharedmem/pipe/BUILD b/pkg/tcpip/link/sharedmem/pipe/BUILD
index 6b5bc542c..a0d4ad0be 100644
--- a/pkg/tcpip/link/sharedmem/pipe/BUILD
+++ b/pkg/tcpip/link/sharedmem/pipe/BUILD
@@ -21,4 +21,5 @@ go_test(
"pipe_test.go",
],
embed = [":pipe"],
+ deps = ["//pkg/sync"],
)
diff --git a/pkg/tcpip/link/sharedmem/pipe/pipe_test.go b/pkg/tcpip/link/sharedmem/pipe/pipe_test.go
index 59ef69a8b..dc239a0d0 100644
--- a/pkg/tcpip/link/sharedmem/pipe/pipe_test.go
+++ b/pkg/tcpip/link/sharedmem/pipe/pipe_test.go
@@ -18,8 +18,9 @@ import (
"math/rand"
"reflect"
"runtime"
- "sync"
"testing"
+
+ "gvisor.dev/gvisor/pkg/sync"
)
func TestSimpleReadWrite(t *testing.T) {
diff --git a/pkg/tcpip/link/sharedmem/sharedmem.go b/pkg/tcpip/link/sharedmem/sharedmem.go
index 080f9d667..655e537c4 100644
--- a/pkg/tcpip/link/sharedmem/sharedmem.go
+++ b/pkg/tcpip/link/sharedmem/sharedmem.go
@@ -23,11 +23,11 @@
package sharedmem
import (
- "sync"
"sync/atomic"
"syscall"
"gvisor.dev/gvisor/pkg/log"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/header"
diff --git a/pkg/tcpip/link/sharedmem/sharedmem_test.go b/pkg/tcpip/link/sharedmem/sharedmem_test.go
index 89603c48f..5c729a439 100644
--- a/pkg/tcpip/link/sharedmem/sharedmem_test.go
+++ b/pkg/tcpip/link/sharedmem/sharedmem_test.go
@@ -22,11 +22,11 @@ import (
"math/rand"
"os"
"strings"
- "sync"
"syscall"
"testing"
"time"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/header"
diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go
index d88119f68..1ceaebfbd 100644
--- a/pkg/tcpip/network/arp/arp.go
+++ b/pkg/tcpip/network/arp/arp.go
@@ -79,16 +79,16 @@ func (e *endpoint) MaxHeaderLength() uint16 {
func (e *endpoint) Close() {}
-func (e *endpoint) WritePacket(*stack.Route, *stack.GSO, stack.NetworkHeaderParams, stack.PacketLooping, tcpip.PacketBuffer) *tcpip.Error {
+func (e *endpoint) WritePacket(*stack.Route, *stack.GSO, stack.NetworkHeaderParams, tcpip.PacketBuffer) *tcpip.Error {
return tcpip.ErrNotSupported
}
// WritePackets implements stack.NetworkEndpoint.WritePackets.
-func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, []tcpip.PacketBuffer, stack.NetworkHeaderParams, stack.PacketLooping) (int, *tcpip.Error) {
+func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, []tcpip.PacketBuffer, stack.NetworkHeaderParams) (int, *tcpip.Error) {
return 0, tcpip.ErrNotSupported
}
-func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, loop stack.PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error {
+func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt tcpip.PacketBuffer) *tcpip.Error {
return tcpip.ErrNotSupported
}
diff --git a/pkg/tcpip/network/fragmentation/BUILD b/pkg/tcpip/network/fragmentation/BUILD
index acf1e022c..ed16076fd 100644
--- a/pkg/tcpip/network/fragmentation/BUILD
+++ b/pkg/tcpip/network/fragmentation/BUILD
@@ -28,6 +28,7 @@ go_library(
visibility = ["//visibility:public"],
deps = [
"//pkg/log",
+ "//pkg/sync",
"//pkg/tcpip",
"//pkg/tcpip/buffer",
],
diff --git a/pkg/tcpip/network/fragmentation/fragmentation.go b/pkg/tcpip/network/fragmentation/fragmentation.go
index 6da5238ec..92f2aa13a 100644
--- a/pkg/tcpip/network/fragmentation/fragmentation.go
+++ b/pkg/tcpip/network/fragmentation/fragmentation.go
@@ -19,9 +19,9 @@ package fragmentation
import (
"fmt"
"log"
- "sync"
"time"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
)
diff --git a/pkg/tcpip/network/fragmentation/reassembler.go b/pkg/tcpip/network/fragmentation/reassembler.go
index 9e002e396..0a83d81f2 100644
--- a/pkg/tcpip/network/fragmentation/reassembler.go
+++ b/pkg/tcpip/network/fragmentation/reassembler.go
@@ -18,9 +18,9 @@ import (
"container/heap"
"fmt"
"math"
- "sync"
"time"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
)
diff --git a/pkg/tcpip/network/ip_test.go b/pkg/tcpip/network/ip_test.go
index 4144a7837..f1bc33adf 100644
--- a/pkg/tcpip/network/ip_test.go
+++ b/pkg/tcpip/network/ip_test.go
@@ -239,7 +239,7 @@ func TestIPv4Send(t *testing.T) {
if err != nil {
t.Fatalf("could not find route: %v", err)
}
- if err := ep.WritePacket(&r, nil /* gso */, stack.NetworkHeaderParams{Protocol: 123, TTL: 123, TOS: stack.DefaultTOS}, stack.PacketOut, tcpip.PacketBuffer{
+ if err := ep.WritePacket(&r, nil /* gso */, stack.NetworkHeaderParams{Protocol: 123, TTL: 123, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{
Header: hdr,
Data: payload.ToVectorisedView(),
}); err != nil {
@@ -480,7 +480,7 @@ func TestIPv6Send(t *testing.T) {
if err != nil {
t.Fatalf("could not find route: %v", err)
}
- if err := ep.WritePacket(&r, nil /* gso */, stack.NetworkHeaderParams{Protocol: 123, TTL: 123, TOS: stack.DefaultTOS}, stack.PacketOut, tcpip.PacketBuffer{
+ if err := ep.WritePacket(&r, nil /* gso */, stack.NetworkHeaderParams{Protocol: 123, TTL: 123, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{
Header: hdr,
Data: payload.ToVectorisedView(),
}); err != nil {
diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go
index f856081e6..0a1453b31 100644
--- a/pkg/tcpip/network/ipv4/ipv4.go
+++ b/pkg/tcpip/network/ipv4/ipv4.go
@@ -241,11 +241,11 @@ func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadS
}
// WritePacket writes a packet to the given destination address and protocol.
-func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, loop stack.PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error {
+func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt tcpip.PacketBuffer) *tcpip.Error {
ip := e.addIPHeader(r, &pkt.Header, pkt.Data.Size(), params)
pkt.NetworkHeader = buffer.View(ip)
- if loop&stack.PacketLoop != 0 {
+ if r.Loop&stack.PacketLoop != 0 {
// The inbound path expects the network header to still be in
// the PacketBuffer's Data field.
views := make([]buffer.View, 1, 1+len(pkt.Data.Views()))
@@ -259,7 +259,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw
loopedR.Release()
}
- if loop&stack.PacketOut == 0 {
+ if r.Loop&stack.PacketOut == 0 {
return nil
}
if pkt.Header.UsedLength()+pkt.Data.Size() > int(e.linkEP.MTU()) && (gso == nil || gso.Type == stack.GSONone) {
@@ -273,11 +273,11 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw
}
// WritePackets implements stack.NetworkEndpoint.WritePackets.
-func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) {
- if loop&stack.PacketLoop != 0 {
+func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, params stack.NetworkHeaderParams) (int, *tcpip.Error) {
+ if r.Loop&stack.PacketLoop != 0 {
panic("multiple packets in local loop")
}
- if loop&stack.PacketOut == 0 {
+ if r.Loop&stack.PacketOut == 0 {
return len(pkts), nil
}
@@ -292,7 +292,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.Pac
// WriteHeaderIncludedPacket writes a packet already containing a network
// header through the given route.
-func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, loop stack.PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error {
+func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt tcpip.PacketBuffer) *tcpip.Error {
// The packet already has an IP header, but there are a few required
// checks.
ip := header.IPv4(pkt.Data.First())
@@ -327,10 +327,10 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, loop stack.PacketLo
ip.SetChecksum(0)
ip.SetChecksum(^ip.CalculateChecksum())
- if loop&stack.PacketLoop != 0 {
+ if r.Loop&stack.PacketLoop != 0 {
e.HandlePacket(r, pkt.Clone())
}
- if loop&stack.PacketOut == 0 {
+ if r.Loop&stack.PacketOut == 0 {
return nil
}
diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go
index 4c940e9e5..180a480fd 100644
--- a/pkg/tcpip/network/ipv6/ipv6.go
+++ b/pkg/tcpip/network/ipv6/ipv6.go
@@ -112,11 +112,11 @@ func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadS
}
// WritePacket writes a packet to the given destination address and protocol.
-func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, loop stack.PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error {
+func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt tcpip.PacketBuffer) *tcpip.Error {
ip := e.addIPHeader(r, &pkt.Header, pkt.Data.Size(), params)
pkt.NetworkHeader = buffer.View(ip)
- if loop&stack.PacketLoop != 0 {
+ if r.Loop&stack.PacketLoop != 0 {
// The inbound path expects the network header to still be in
// the PacketBuffer's Data field.
views := make([]buffer.View, 1, 1+len(pkt.Data.Views()))
@@ -130,7 +130,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw
loopedR.Release()
}
- if loop&stack.PacketOut == 0 {
+ if r.Loop&stack.PacketOut == 0 {
return nil
}
@@ -139,11 +139,11 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw
}
// WritePackets implements stack.LinkEndpoint.WritePackets.
-func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) {
- if loop&stack.PacketLoop != 0 {
+func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, params stack.NetworkHeaderParams) (int, *tcpip.Error) {
+ if r.Loop&stack.PacketLoop != 0 {
panic("not implemented")
}
- if loop&stack.PacketOut == 0 {
+ if r.Loop&stack.PacketOut == 0 {
return len(pkts), nil
}
@@ -161,7 +161,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.Pac
// WriteHeaderIncludedPacker implements stack.NetworkEndpoint. It is not yet
// supported by IPv6.
-func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, loop stack.PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error {
+func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt tcpip.PacketBuffer) *tcpip.Error {
// TODO(b/146666412): Support IPv6 header-included packets.
return tcpip.ErrNotSupported
}
diff --git a/pkg/tcpip/ports/BUILD b/pkg/tcpip/ports/BUILD
index e156b01f6..a6ef3bdcc 100644
--- a/pkg/tcpip/ports/BUILD
+++ b/pkg/tcpip/ports/BUILD
@@ -9,6 +9,7 @@ go_library(
importpath = "gvisor.dev/gvisor/pkg/tcpip/ports",
visibility = ["//visibility:public"],
deps = [
+ "//pkg/sync",
"//pkg/tcpip",
],
)
diff --git a/pkg/tcpip/ports/ports.go b/pkg/tcpip/ports/ports.go
index 6c5e19e8f..b937cb84b 100644
--- a/pkg/tcpip/ports/ports.go
+++ b/pkg/tcpip/ports/ports.go
@@ -18,9 +18,9 @@ package ports
import (
"math"
"math/rand"
- "sync"
"sync/atomic"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
)
diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD
index b8f9517d0..705e984c1 100644
--- a/pkg/tcpip/stack/BUILD
+++ b/pkg/tcpip/stack/BUILD
@@ -36,6 +36,7 @@ go_library(
"//pkg/ilist",
"//pkg/rand",
"//pkg/sleep",
+ "//pkg/sync",
"//pkg/tcpip",
"//pkg/tcpip/buffer",
"//pkg/tcpip/hash/jenkins",
@@ -83,6 +84,7 @@ go_test(
embed = [":stack"],
deps = [
"//pkg/sleep",
+ "//pkg/sync",
"//pkg/tcpip",
],
)
diff --git a/pkg/tcpip/stack/linkaddrcache.go b/pkg/tcpip/stack/linkaddrcache.go
index 267df60d1..403557fd7 100644
--- a/pkg/tcpip/stack/linkaddrcache.go
+++ b/pkg/tcpip/stack/linkaddrcache.go
@@ -16,10 +16,10 @@ package stack
import (
"fmt"
- "sync"
"time"
"gvisor.dev/gvisor/pkg/sleep"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
)
diff --git a/pkg/tcpip/stack/linkaddrcache_test.go b/pkg/tcpip/stack/linkaddrcache_test.go
index 9946b8fe8..1baa498d0 100644
--- a/pkg/tcpip/stack/linkaddrcache_test.go
+++ b/pkg/tcpip/stack/linkaddrcache_test.go
@@ -16,12 +16,12 @@ package stack
import (
"fmt"
- "sync"
"sync/atomic"
"testing"
"time"
"gvisor.dev/gvisor/pkg/sleep"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
)
diff --git a/pkg/tcpip/stack/ndp.go b/pkg/tcpip/stack/ndp.go
index 4722ec9ce..a9dd322db 100644
--- a/pkg/tcpip/stack/ndp.go
+++ b/pkg/tcpip/stack/ndp.go
@@ -115,6 +115,30 @@ var (
MinPrefixInformationValidLifetimeForUpdate = 2 * time.Hour
)
+// DHCPv6ConfigurationFromNDPRA is a configuration available via DHCPv6 that an
+// NDP Router Advertisement informed the Stack about.
+type DHCPv6ConfigurationFromNDPRA int
+
+const (
+ // DHCPv6NoConfiguration indicates that no configurations are available via
+ // DHCPv6.
+ DHCPv6NoConfiguration DHCPv6ConfigurationFromNDPRA = iota
+
+ // DHCPv6ManagedAddress indicates that addresses are available via DHCPv6.
+ //
+ // DHCPv6ManagedAddress also implies DHCPv6OtherConfigurations because DHCPv6
+ // will return all available configuration information.
+ DHCPv6ManagedAddress
+
+ // DHCPv6OtherConfigurations indicates that other configuration information is
+ // available via DHCPv6.
+ //
+ // Other configurations are configurations other than addresses. Examples of
+ // other configurations are recursive DNS server list, DNS search lists and
+ // default gateway.
+ DHCPv6OtherConfigurations
+)
+
// NDPDispatcher is the interface integrators of netstack must implement to
// receive and handle NDP related events.
type NDPDispatcher interface {
@@ -194,7 +218,20 @@ type NDPDispatcher interface {
// already known DNS servers. If called with known DNS servers, their
// valid lifetimes must be refreshed to lifetime (it may be increased,
// decreased, or completely invalidated when lifetime = 0).
+ //
+ // This function is not permitted to block indefinitely. It must not
+ // call functions on the stack itself.
OnRecursiveDNSServerOption(nicID tcpip.NICID, addrs []tcpip.Address, lifetime time.Duration)
+
+ // 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)
}
// NDPConfigurations is the NDP configurations for the netstack.
@@ -281,6 +318,9 @@ type ndpState struct {
// The addresses generated by SLAAC.
autoGenAddresses map[tcpip.Address]autoGenAddressState
+
+ // The last learned DHCPv6 configuration from an NDP RA.
+ dhcpv6Configuration DHCPv6ConfigurationFromNDPRA
}
// dadState holds the Duplicate Address Detection timer and channel to signal
@@ -299,46 +339,14 @@ type dadState struct {
// defaultRouterState holds data associated with a default router discovered by
// a Router Advertisement (RA).
type defaultRouterState struct {
- invalidationTimer *time.Timer
-
- // Used to inform the timer not to invalidate the default router (R) in
- // a race condition (T1 is a goroutine that handles an RA from R and T2
- // is the goroutine that handles R's invalidation timer firing):
- // T1: Receive a new RA from R
- // T1: Obtain the NIC's lock before processing the RA
- // T2: R's invalidation timer fires, and gets blocked on obtaining the
- // NIC's lock
- // T1: Refreshes/extends R's lifetime & releases NIC's lock
- // T2: Obtains NIC's lock & invalidates R immediately
- //
- // To resolve this, T1 will check to see if the timer already fired, and
- // inform the timer using doNotInvalidate to not invalidate R, so that
- // once T2 obtains the lock, it will see that it is set to true and do
- // nothing further.
- doNotInvalidate *bool
+ invalidationTimer tcpip.CancellableTimer
}
// onLinkPrefixState holds data associated with an on-link prefix discovered by
// a Router Advertisement's Prefix Information option (PI) when the NDP
// configurations was configured to do so.
type onLinkPrefixState struct {
- invalidationTimer *time.Timer
-
- // Used to signal the timer not to invalidate the on-link prefix (P) in
- // a race condition (T1 is a goroutine that handles a PI for P and T2
- // is the goroutine that handles P's invalidation timer firing):
- // T1: Receive a new PI for P
- // T1: Obtain the NIC's lock before processing the PI
- // T2: P's invalidation timer fires, and gets blocked on obtaining the
- // NIC's lock
- // T1: Refreshes/extends P's lifetime & releases NIC's lock
- // T2: Obtains NIC's lock & invalidates P immediately
- //
- // To resolve this, T1 will check to see if the timer already fired, and
- // inform the timer using doNotInvalidate to not invalidate P, so that
- // once T2 obtains the lock, it will see that it is set to true and do
- // nothing further.
- doNotInvalidate *bool
+ invalidationTimer tcpip.CancellableTimer
}
// autoGenAddressState holds data associated with an address generated via
@@ -348,33 +356,10 @@ type autoGenAddressState struct {
// is holding state for.
ref *referencedNetworkEndpoint
- deprecationTimer *time.Timer
+ deprecationTimer tcpip.CancellableTimer
+ invalidationTimer tcpip.CancellableTimer
- // Used to signal the timer not to deprecate the SLAAC address in a race
- // condition. Used for the same reason as doNotInvalidate, but for deprecating
- // an address.
- doNotDeprecate *bool
-
- invalidationTimer *time.Timer
-
- // Used to signal the timer not to invalidate the SLAAC address (A) in
- // a race condition (T1 is a goroutine that handles a PI for A and T2
- // is the goroutine that handles A's invalidation timer firing):
- // T1: Receive a new PI for A
- // T1: Obtain the NIC's lock before processing the PI
- // T2: A's invalidation timer fires, and gets blocked on obtaining the
- // NIC's lock
- // T1: Refreshes/extends A's lifetime & releases NIC's lock
- // T2: Obtains NIC's lock & invalidates A immediately
- //
- // To resolve this, T1 will check to see if the timer already fired, and
- // inform the timer using doNotInvalidate to not invalidate A, so that
- // once T2 obtains the lock, it will see that it is set to true and do
- // nothing further.
- doNotInvalidate *bool
-
- // Nonzero only when the address is not valid forever (invalidationTimer
- // is not nil).
+ // Nonzero only when the address is not valid forever.
validUntil time.Time
}
@@ -576,7 +561,7 @@ func (ndp *ndpState) stopDuplicateAddressDetection(addr tcpip.Address) {
// handleRA handles a Router Advertisement message that arrived on the NIC
// this ndp is for. Does nothing if the NIC is configured to not handle RAs.
//
-// The NIC that ndp belongs to and its associated stack MUST be locked.
+// The NIC that ndp belongs to MUST be locked.
func (ndp *ndpState) handleRA(ip tcpip.Address, ra header.NDPRouterAdvert) {
// Is the NIC configured to handle RAs at all?
//
@@ -588,6 +573,28 @@ func (ndp *ndpState) handleRA(ip tcpip.Address, ra header.NDPRouterAdvert) {
return
}
+ // Only worry about the DHCPv6 configuration if we have an NDPDispatcher as we
+ // only inform the dispatcher on configuration changes. We do nothing else
+ // with the information.
+ if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil {
+ var configuration DHCPv6ConfigurationFromNDPRA
+ switch {
+ case ra.ManagedAddrConfFlag():
+ configuration = DHCPv6ManagedAddress
+
+ case ra.OtherConfFlag():
+ configuration = DHCPv6OtherConfigurations
+
+ default:
+ configuration = DHCPv6NoConfiguration
+ }
+
+ if ndp.dhcpv6Configuration != configuration {
+ ndp.dhcpv6Configuration = configuration
+ ndpDisp.OnDHCPv6Configuration(ndp.nic.ID(), configuration)
+ }
+ }
+
// Is the NIC configured to discover default routers?
if ndp.configs.DiscoverDefaultRouters {
rtr, ok := ndp.defaultRouters[ip]
@@ -605,27 +612,9 @@ func (ndp *ndpState) handleRA(ip tcpip.Address, ra header.NDPRouterAdvert) {
case ok && rl != 0:
// This is an already discovered default router. Update
// the invalidation timer.
- timer := rtr.invalidationTimer
-
- // We should ALWAYS have an invalidation timer for a
- // discovered router.
- if timer == nil {
- panic("ndphandlera: RA invalidation timer should not be nil")
- }
-
- if !timer.Stop() {
- // If we reach this point, then we know the
- // timer fired after we already took the NIC
- // lock. Inform the timer not to invalidate the
- // router when it obtains the lock as we just
- // got a new RA that refreshes its lifetime to a
- // non-zero value. See
- // defaultRouterState.doNotInvalidate for more
- // details.
- *rtr.doNotInvalidate = true
- }
-
- timer.Reset(rl)
+ rtr.invalidationTimer.StopLocked()
+ rtr.invalidationTimer.Reset(rl)
+ ndp.defaultRouters[ip] = rtr
case ok && rl == 0:
// We know about the router but it is no longer to be
@@ -692,10 +681,7 @@ func (ndp *ndpState) invalidateDefaultRouter(ip tcpip.Address) {
return
}
- rtr.invalidationTimer.Stop()
- rtr.invalidationTimer = nil
- *rtr.doNotInvalidate = true
- rtr.doNotInvalidate = nil
+ rtr.invalidationTimer.StopLocked()
delete(ndp.defaultRouters, ip)
@@ -724,27 +710,15 @@ func (ndp *ndpState) rememberDefaultRouter(ip tcpip.Address, rl time.Duration) {
return
}
- // Used to signal the timer not to invalidate the default router (R) in
- // a race condition. See defaultRouterState.doNotInvalidate for more
- // details.
- var doNotInvalidate bool
-
- ndp.defaultRouters[ip] = defaultRouterState{
- invalidationTimer: time.AfterFunc(rl, func() {
- ndp.nic.stack.mu.Lock()
- defer ndp.nic.stack.mu.Unlock()
- ndp.nic.mu.Lock()
- defer ndp.nic.mu.Unlock()
-
- if doNotInvalidate {
- doNotInvalidate = false
- return
- }
-
+ state := defaultRouterState{
+ invalidationTimer: tcpip.MakeCancellableTimer(&ndp.nic.mu, func() {
ndp.invalidateDefaultRouter(ip)
}),
- doNotInvalidate: &doNotInvalidate,
}
+
+ state.invalidationTimer.Reset(rl)
+
+ ndp.defaultRouters[ip] = state
}
// rememberOnLinkPrefix remembers a newly discovered on-link prefix with IPv6
@@ -766,21 +740,17 @@ func (ndp *ndpState) rememberOnLinkPrefix(prefix tcpip.Subnet, l time.Duration)
return
}
- // Used to signal the timer not to invalidate the on-link prefix (P) in
- // a race condition. See onLinkPrefixState.doNotInvalidate for more
- // details.
- var doNotInvalidate bool
- var timer *time.Timer
+ state := onLinkPrefixState{
+ invalidationTimer: tcpip.MakeCancellableTimer(&ndp.nic.mu, func() {
+ ndp.invalidateOnLinkPrefix(prefix)
+ }),
+ }
- // Only create a timer if the lifetime is not infinite.
if l < header.NDPInfiniteLifetime {
- timer = ndp.prefixInvalidationCallback(prefix, l, &doNotInvalidate)
+ state.invalidationTimer.Reset(l)
}
- ndp.onLinkPrefixes[prefix] = onLinkPrefixState{
- invalidationTimer: timer,
- doNotInvalidate: &doNotInvalidate,
- }
+ ndp.onLinkPrefixes[prefix] = state
}
// invalidateOnLinkPrefix invalidates a discovered on-link prefix.
@@ -795,13 +765,7 @@ func (ndp *ndpState) invalidateOnLinkPrefix(prefix tcpip.Subnet) {
return
}
- if s.invalidationTimer != nil {
- s.invalidationTimer.Stop()
- s.invalidationTimer = nil
- *s.doNotInvalidate = true
- }
-
- s.doNotInvalidate = nil
+ s.invalidationTimer.StopLocked()
delete(ndp.onLinkPrefixes, prefix)
@@ -811,28 +775,6 @@ func (ndp *ndpState) invalidateOnLinkPrefix(prefix tcpip.Subnet) {
}
}
-// prefixInvalidationCallback returns a new on-link prefix invalidation timer
-// for prefix that fires after vl.
-//
-// doNotInvalidate is used to signal the timer when it fires at the same time
-// that a prefix's valid lifetime gets refreshed. See
-// onLinkPrefixState.doNotInvalidate for more details.
-func (ndp *ndpState) prefixInvalidationCallback(prefix tcpip.Subnet, vl time.Duration, doNotInvalidate *bool) *time.Timer {
- return time.AfterFunc(vl, func() {
- ndp.nic.stack.mu.Lock()
- defer ndp.nic.stack.mu.Unlock()
- ndp.nic.mu.Lock()
- defer ndp.nic.mu.Unlock()
-
- if *doNotInvalidate {
- *doNotInvalidate = false
- return
- }
-
- ndp.invalidateOnLinkPrefix(prefix)
- })
-}
-
// handleOnLinkPrefixInformation handles a Prefix Information option with
// its on-link flag set, as per RFC 4861 section 6.3.4.
//
@@ -872,42 +814,17 @@ func (ndp *ndpState) handleOnLinkPrefixInformation(pi header.NDPPrefixInformatio
// This is an already discovered on-link prefix with a
// new non-zero valid lifetime.
+ //
// Update the invalidation timer.
- timer := prefixState.invalidationTimer
-
- if timer == nil && vl >= header.NDPInfiniteLifetime {
- // Had infinite valid lifetime before and
- // continues to have an invalid lifetime. Do
- // nothing further.
- return
- }
- if timer != nil && !timer.Stop() {
- // If we reach this point, then we know the timer alread fired
- // after we took the NIC lock. Inform the timer to not
- // invalidate the prefix once it obtains the lock as we just
- // got a new PI that refreshes its lifetime to a non-zero value.
- // See onLinkPrefixState.doNotInvalidate for more details.
- *prefixState.doNotInvalidate = true
- }
+ prefixState.invalidationTimer.StopLocked()
- if vl >= header.NDPInfiniteLifetime {
- // Prefix is now valid forever so we don't need
- // an invalidation timer.
- prefixState.invalidationTimer = nil
- ndp.onLinkPrefixes[prefix] = prefixState
- return
- }
-
- if timer != nil {
- // We already have a timer so just reset it to
- // expire after the new valid lifetime.
- timer.Reset(vl)
- return
+ if vl < header.NDPInfiniteLifetime {
+ // Prefix is valid for a finite lifetime, reset the timer to expire after
+ // the new valid lifetime.
+ prefixState.invalidationTimer.Reset(vl)
}
- // We do not have a timer so just create a new one.
- prefixState.invalidationTimer = ndp.prefixInvalidationCallback(prefix, vl, prefixState.doNotInvalidate)
ndp.onLinkPrefixes[prefix] = prefixState
}
@@ -917,7 +834,7 @@ func (ndp *ndpState) handleOnLinkPrefixInformation(pi header.NDPPrefixInformatio
// handleAutonomousPrefixInformation assumes that the prefix this pi is for is
// not the link-local prefix and the autonomous flag is set.
//
-// The NIC that ndp belongs to and its associated stack MUST be locked.
+// The NIC that ndp belongs to MUST be locked.
func (ndp *ndpState) handleAutonomousPrefixInformation(pi header.NDPPrefixInformation) {
vl := pi.ValidLifetime()
pl := pi.PreferredLifetime()
@@ -1026,28 +943,34 @@ func (ndp *ndpState) newAutoGenAddress(prefix tcpip.Subnet, pl, vl time.Duration
log.Fatalf("ndp: error when adding address %s: %s", protocolAddr, err)
}
- // Setup the timers to deprecate and invalidate this newly generated address.
+ state := autoGenAddressState{
+ ref: ref,
+ deprecationTimer: tcpip.MakeCancellableTimer(&ndp.nic.mu, func() {
+ addrState, ok := ndp.autoGenAddresses[addr]
+ if !ok {
+ log.Fatalf("ndp: must have an autoGenAddressess entry for the SLAAC generated IPv6 address %s", addr)
+ }
+ addrState.ref.deprecated = true
+ ndp.notifyAutoGenAddressDeprecated(addr)
+ }),
+ invalidationTimer: tcpip.MakeCancellableTimer(&ndp.nic.mu, func() {
+ ndp.invalidateAutoGenAddress(addr)
+ }),
+ }
+
+ // Setup the initial timers to deprecate and invalidate this newly generated
+ // address.
- var doNotDeprecate bool
- var pTimer *time.Timer
if !deprecated && pl < header.NDPInfiniteLifetime {
- pTimer = ndp.autoGenAddrDeprecationTimer(addr, pl, &doNotDeprecate)
+ state.deprecationTimer.Reset(pl)
}
- var doNotInvalidate bool
- var vTimer *time.Timer
if vl < header.NDPInfiniteLifetime {
- vTimer = ndp.autoGenAddrInvalidationTimer(addr, vl, &doNotInvalidate)
+ state.invalidationTimer.Reset(vl)
+ state.validUntil = time.Now().Add(vl)
}
- ndp.autoGenAddresses[addr] = autoGenAddressState{
- ref: ref,
- deprecationTimer: pTimer,
- doNotDeprecate: &doNotDeprecate,
- invalidationTimer: vTimer,
- doNotInvalidate: &doNotInvalidate,
- validUntil: time.Now().Add(vl),
- }
+ ndp.autoGenAddresses[addr] = state
}
// refreshAutoGenAddressLifetimes refreshes the lifetime of a SLAAC generated
@@ -1075,20 +998,10 @@ func (ndp *ndpState) refreshAutoGenAddressLifetimes(addr tcpip.Address, pl, vl t
// If addr was preferred for some finite lifetime before, stop the deprecation
// timer so it can be reset.
- if t := addrState.deprecationTimer; t != nil && !t.Stop() {
- *addrState.doNotDeprecate = true
- }
+ addrState.deprecationTimer.StopLocked()
- // Reset the deprecation timer.
- if pl >= header.NDPInfiniteLifetime || deprecated {
- // If addr is preferred forever or it has been deprecated already, there is
- // no need for a deprecation timer.
- addrState.deprecationTimer = nil
- } else if addrState.deprecationTimer == nil {
- // addr is now preferred for a finite lifetime.
- addrState.deprecationTimer = ndp.autoGenAddrDeprecationTimer(addr, pl, addrState.doNotDeprecate)
- } else {
- // addr continues to be preferred for a finite lifetime.
+ // Reset the deprecation timer if addr has a finite preferred lifetime.
+ if !deprecated && pl < header.NDPInfiniteLifetime {
addrState.deprecationTimer.Reset(pl)
}
@@ -1107,15 +1020,8 @@ func (ndp *ndpState) refreshAutoGenAddressLifetimes(addr tcpip.Address, pl, vl t
// Handle the infinite valid lifetime separately as we do not keep a timer in
// this case.
if vl >= header.NDPInfiniteLifetime {
- if addrState.invalidationTimer != nil {
- // Valid lifetime was finite before, but now it is valid forever.
- if !addrState.invalidationTimer.Stop() {
- *addrState.doNotInvalidate = true
- }
- addrState.invalidationTimer = nil
- addrState.validUntil = time.Time{}
- }
-
+ addrState.invalidationTimer.StopLocked()
+ addrState.validUntil = time.Time{}
return
}
@@ -1124,7 +1030,7 @@ func (ndp *ndpState) refreshAutoGenAddressLifetimes(addr tcpip.Address, pl, vl t
// If the address was originally set to be valid forever, assume the remaining
// time to be the maximum possible value.
- if addrState.invalidationTimer == nil {
+ if addrState.validUntil == (time.Time{}) {
rl = header.NDPInfiniteLifetime
} else {
rl = time.Until(addrState.validUntil)
@@ -1138,15 +1044,8 @@ func (ndp *ndpState) refreshAutoGenAddressLifetimes(addr tcpip.Address, pl, vl t
effectiveVl = MinPrefixInformationValidLifetimeForUpdate
}
- if addrState.invalidationTimer == nil {
- addrState.invalidationTimer = ndp.autoGenAddrInvalidationTimer(addr, effectiveVl, addrState.doNotInvalidate)
- } else {
- if !addrState.invalidationTimer.Stop() {
- *addrState.doNotInvalidate = true
- }
- addrState.invalidationTimer.Reset(effectiveVl)
- }
-
+ addrState.invalidationTimer.StopLocked()
+ addrState.invalidationTimer.Reset(effectiveVl)
addrState.validUntil = time.Now().Add(effectiveVl)
}
@@ -1181,27 +1080,12 @@ func (ndp *ndpState) invalidateAutoGenAddress(addr tcpip.Address) {
// The NIC that ndp belongs to MUST be locked.
func (ndp *ndpState) cleanupAutoGenAddrResourcesAndNotify(addr tcpip.Address) bool {
state, ok := ndp.autoGenAddresses[addr]
-
if !ok {
return false
}
- if state.deprecationTimer != nil {
- state.deprecationTimer.Stop()
- state.deprecationTimer = nil
- *state.doNotDeprecate = true
- }
-
- state.doNotDeprecate = nil
-
- if state.invalidationTimer != nil {
- state.invalidationTimer.Stop()
- state.invalidationTimer = nil
- *state.doNotInvalidate = true
- }
-
- state.doNotInvalidate = nil
-
+ state.deprecationTimer.StopLocked()
+ state.invalidationTimer.StopLocked()
delete(ndp.autoGenAddresses, addr)
if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil {
@@ -1214,53 +1098,6 @@ func (ndp *ndpState) cleanupAutoGenAddrResourcesAndNotify(addr tcpip.Address) bo
return true
}
-// autoGenAddrDeprecationTimer returns a new deprecation timer for an
-// auto-generated address that fires after pl.
-//
-// doNotDeprecate is used to inform the timer when it fires at the same time
-// that an auto-generated address's preferred lifetime gets refreshed. See
-// autoGenAddrState.doNotDeprecate for more details.
-func (ndp *ndpState) autoGenAddrDeprecationTimer(addr tcpip.Address, pl time.Duration, doNotDeprecate *bool) *time.Timer {
- return time.AfterFunc(pl, func() {
- ndp.nic.mu.Lock()
- defer ndp.nic.mu.Unlock()
-
- if *doNotDeprecate {
- *doNotDeprecate = false
- return
- }
-
- addrState, ok := ndp.autoGenAddresses[addr]
- if !ok {
- log.Fatalf("ndp: must have an autoGenAddressess entry for the SLAAC generated IPv6 address %s", addr)
- }
- addrState.ref.deprecated = true
- ndp.notifyAutoGenAddressDeprecated(addr)
- addrState.deprecationTimer = nil
- ndp.autoGenAddresses[addr] = addrState
- })
-}
-
-// autoGenAddrInvalidationTimer returns a new invalidation timer for an
-// auto-generated address that fires after vl.
-//
-// doNotInvalidate is used to inform the timer when it fires at the same time
-// that an auto-generated address's valid lifetime gets refreshed. See
-// autoGenAddrState.doNotInvalidate for more details.
-func (ndp *ndpState) autoGenAddrInvalidationTimer(addr tcpip.Address, vl time.Duration, doNotInvalidate *bool) *time.Timer {
- return time.AfterFunc(vl, func() {
- ndp.nic.mu.Lock()
- defer ndp.nic.mu.Unlock()
-
- if *doNotInvalidate {
- *doNotInvalidate = false
- return
- }
-
- ndp.invalidateAutoGenAddress(addr)
- })
-}
-
// cleanupHostOnlyState cleans up any state that is only useful for hosts.
//
// cleanupHostOnlyState MUST be called when ndp's NIC is transitioning from a
diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go
index 8d89859ba..f9bc18c55 100644
--- a/pkg/tcpip/stack/ndp_test.go
+++ b/pkg/tcpip/stack/ndp_test.go
@@ -162,18 +162,24 @@ type ndpRDNSSEvent struct {
rdnss ndpRDNSS
}
+type ndpDHCPv6Event struct {
+ nicID tcpip.NICID
+ configuration stack.DHCPv6ConfigurationFromNDPRA
+}
+
var _ stack.NDPDispatcher = (*ndpDispatcher)(nil)
// ndpDispatcher implements NDPDispatcher so tests can know when various NDP
// related events happen for test purposes.
type ndpDispatcher struct {
- dadC chan ndpDADEvent
- routerC chan ndpRouterEvent
- rememberRouter bool
- prefixC chan ndpPrefixEvent
- rememberPrefix bool
- autoGenAddrC chan ndpAutoGenAddrEvent
- rdnssC chan ndpRDNSSEvent
+ dadC chan ndpDADEvent
+ routerC chan ndpRouterEvent
+ rememberRouter bool
+ prefixC chan ndpPrefixEvent
+ rememberPrefix bool
+ autoGenAddrC chan ndpAutoGenAddrEvent
+ rdnssC chan ndpRDNSSEvent
+ dhcpv6ConfigurationC chan ndpDHCPv6Event
}
// Implements stack.NDPDispatcher.OnDuplicateAddressDetectionStatus.
@@ -280,6 +286,16 @@ func (n *ndpDispatcher) OnRecursiveDNSServerOption(nicID tcpip.NICID, addrs []tc
}
}
+// Implements stack.NDPDispatcher.OnDHCPv6Configuration.
+func (n *ndpDispatcher) OnDHCPv6Configuration(nicID tcpip.NICID, configuration stack.DHCPv6ConfigurationFromNDPRA) {
+ if c := n.dhcpv6ConfigurationC; c != nil {
+ c <- ndpDHCPv6Event{
+ nicID,
+ configuration,
+ }
+ }
+}
+
// TestDADResolve tests that an address successfully resolves after performing
// DAD for various values of DupAddrDetectTransmits and RetransmitTimer.
// Included in the subtests is a test to make sure that an invalid
@@ -797,21 +813,32 @@ func TestSetNDPConfigurations(t *testing.T) {
}
}
-// raBufWithOpts returns a valid NDP Router Advertisement with options.
-//
-// Note, raBufWithOpts does not populate any of the RA fields other than the
-// Router Lifetime.
-func raBufWithOpts(ip tcpip.Address, rl uint16, optSer header.NDPOptionsSerializer) tcpip.PacketBuffer {
+// raBufWithOptsAndDHCPv6 returns a valid NDP Router Advertisement with options
+// and DHCPv6 configurations specified.
+func raBufWithOptsAndDHCPv6(ip tcpip.Address, rl uint16, managedAddress, otherConfigurations bool, optSer header.NDPOptionsSerializer) tcpip.PacketBuffer {
icmpSize := header.ICMPv6HeaderSize + header.NDPRAMinimumSize + int(optSer.Length())
hdr := buffer.NewPrependable(header.IPv6MinimumSize + icmpSize)
pkt := header.ICMPv6(hdr.Prepend(icmpSize))
pkt.SetType(header.ICMPv6RouterAdvert)
pkt.SetCode(0)
- ra := header.NDPRouterAdvert(pkt.NDPPayload())
+ raPayload := pkt.NDPPayload()
+ ra := header.NDPRouterAdvert(raPayload)
+ // Populate the Router Lifetime.
+ binary.BigEndian.PutUint16(raPayload[2:], rl)
+ // Populate the Managed Address flag field.
+ if managedAddress {
+ // The Managed Addresses flag field is the 7th bit of byte #1 (0-indexing)
+ // of the RA payload.
+ raPayload[1] |= (1 << 7)
+ }
+ // Populate the Other Configurations flag field.
+ if otherConfigurations {
+ // The Other Configurations flag field is the 6th bit of byte #1
+ // (0-indexing) of the RA payload.
+ raPayload[1] |= (1 << 6)
+ }
opts := ra.Options()
opts.Serialize(optSer)
- // Populate the Router Lifetime.
- binary.BigEndian.PutUint16(pkt.NDPPayload()[2:], rl)
pkt.SetChecksum(header.ICMPv6Checksum(pkt, ip, header.IPv6AllNodesMulticastAddress, buffer.VectorisedView{}))
payloadLength := hdr.UsedLength()
iph := header.IPv6(hdr.Prepend(header.IPv6MinimumSize))
@@ -826,6 +853,23 @@ func raBufWithOpts(ip tcpip.Address, rl uint16, optSer header.NDPOptionsSerializ
return tcpip.PacketBuffer{Data: hdr.View().ToVectorisedView()}
}
+// raBufWithOpts returns a valid NDP Router Advertisement with options.
+//
+// Note, raBufWithOpts does not populate any of the RA fields other than the
+// Router Lifetime.
+func raBufWithOpts(ip tcpip.Address, rl uint16, optSer header.NDPOptionsSerializer) tcpip.PacketBuffer {
+ return raBufWithOptsAndDHCPv6(ip, rl, false, false, optSer)
+}
+
+// raBufWithDHCPv6 returns a valid NDP Router Advertisement with DHCPv6 related
+// fields set.
+//
+// Note, raBufWithDHCPv6 does not populate any of the RA fields other than the
+// DHCPv6 related ones.
+func raBufWithDHCPv6(ip tcpip.Address, managedAddresses, otherConfiguratiosns bool) tcpip.PacketBuffer {
+ return raBufWithOptsAndDHCPv6(ip, 0, managedAddresses, otherConfiguratiosns, header.NDPOptionsSerializer{})
+}
+
// raBuf returns a valid NDP Router Advertisement.
//
// Note, raBuf does not populate any of the RA fields other than the
@@ -1029,13 +1073,13 @@ func TestRouterDiscovery(t *testing.T) {
expectRouterEvent(llAddr2, true)
// Rx an RA from another router (lladdr3) with non-zero lifetime.
- l3Lifetime := time.Duration(6)
- e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr3, uint16(l3Lifetime)))
+ const l3LifetimeSeconds = 6
+ e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr3, l3LifetimeSeconds))
expectRouterEvent(llAddr3, true)
// Rx an RA from lladdr2 with lesser lifetime.
- l2Lifetime := time.Duration(2)
- e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, uint16(l2Lifetime)))
+ const l2LifetimeSeconds = 2
+ e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, l2LifetimeSeconds))
select {
case <-ndpDisp.routerC:
t.Fatal("Should not receive a router event when updating lifetimes for known routers")
@@ -1049,7 +1093,7 @@ func TestRouterDiscovery(t *testing.T) {
// Wait for the normal lifetime plus an extra bit for the
// router to get invalidated. If we don't get an invalidation
// event after this time, then something is wrong.
- expectAsyncRouterInvalidationEvent(llAddr2, l2Lifetime*time.Second+defaultTimeout)
+ expectAsyncRouterInvalidationEvent(llAddr2, l2LifetimeSeconds*time.Second+defaultTimeout)
// Rx an RA from lladdr2 with huge lifetime.
e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, 1000))
@@ -1066,7 +1110,7 @@ func TestRouterDiscovery(t *testing.T) {
// Wait for the normal lifetime plus an extra bit for the
// router to get invalidated. If we don't get an invalidation
// event after this time, then something is wrong.
- expectAsyncRouterInvalidationEvent(llAddr3, l3Lifetime*time.Second+defaultTimeout)
+ expectAsyncRouterInvalidationEvent(llAddr3, l3LifetimeSeconds*time.Second+defaultTimeout)
}
// TestRouterDiscoveryMaxRouters tests that only
@@ -1701,9 +1745,8 @@ func addrForNewConnection(t *testing.T, s *stack.Stack) tcpip.Address {
t.Fatalf("s.NewEndpoint(%d, %d, _): %s", header.UDPProtocolNumber, header.IPv6ProtocolNumber, err)
}
defer ep.Close()
- v := tcpip.V6OnlyOption(1)
- if err := ep.SetSockOpt(v); err != nil {
- t.Fatalf("SetSockOpt(%+v): %s", v, err)
+ if err := ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil {
+ t.Fatalf("SetSockOpt(tcpip.V6OnlyOption, true): %s", err)
}
if err := ep.Connect(dstAddr); err != nil {
t.Fatalf("ep.Connect(%+v): %s", dstAddr, err)
@@ -1728,9 +1771,8 @@ func addrForNewConnectionWithAddr(t *testing.T, s *stack.Stack, addr tcpip.FullA
t.Fatalf("s.NewEndpoint(%d, %d, _): %s", header.UDPProtocolNumber, header.IPv6ProtocolNumber, err)
}
defer ep.Close()
- v := tcpip.V6OnlyOption(1)
- if err := ep.SetSockOpt(v); err != nil {
- t.Fatalf("SetSockOpt(%+v): %s", v, err)
+ if err := ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil {
+ t.Fatalf("SetSockOpt(tcpip.V6OnlyOption, true): %s", err)
}
if err := ep.Bind(addr); err != nil {
t.Fatalf("ep.Bind(%+v): %s", addr, err)
@@ -2066,9 +2108,8 @@ func TestAutoGenAddrTimerDeprecation(t *testing.T) {
t.Fatalf("s.NewEndpoint(%d, %d, _): %s", header.UDPProtocolNumber, header.IPv6ProtocolNumber, err)
}
defer ep.Close()
- v := tcpip.V6OnlyOption(1)
- if err := ep.SetSockOpt(v); err != nil {
- t.Fatalf("SetSockOpt(%+v): %s", v, err)
+ if err := ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil {
+ t.Fatalf("SetSockOpt(tcpip.V6OnlyOption, true): %s", err)
}
if err := ep.Connect(dstAddr); err != tcpip.ErrNoRoute {
@@ -2500,9 +2541,9 @@ func TestAutoGenAddrWithOpaqueIID(t *testing.T) {
SecretKey: secretKey,
},
})
-
- if err := s.CreateNamedNIC(nicID, nicName, e); err != nil {
- t.Fatalf("CreateNamedNIC(%d, %q, _) = %s", nicID, nicName, err)
+ opts := stack.NICOptions{Name: nicName}
+ if err := s.CreateNICWithOptions(nicID, e, opts); err != nil {
+ t.Fatalf("CreateNICWithOptions(%d, _, %+v, _) = %s", nicID, opts, err)
}
expectAutoGenAddrEvent := func(addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType) {
@@ -2954,3 +2995,94 @@ func TestCleanupHostOnlyStateOnBecomingRouter(t *testing.T) {
default:
}
}
+
+// TestDHCPv6ConfigurationFromNDPDA tests that the NDPDispatcher is properly
+// informed when new information about what configurations are available via
+// DHCPv6 is learned.
+func TestDHCPv6ConfigurationFromNDPDA(t *testing.T) {
+ const nicID = 1
+
+ ndpDisp := ndpDispatcher{
+ dhcpv6ConfigurationC: make(chan ndpDHCPv6Event, 1),
+ rememberRouter: true,
+ }
+ e := channel.New(0, 1280, linkAddr1)
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ HandleRAs: true,
+ },
+ NDPDisp: &ndpDisp,
+ })
+
+ if err := s.CreateNIC(nicID, e); err != nil {
+ t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
+ }
+
+ expectDHCPv6Event := func(configuration stack.DHCPv6ConfigurationFromNDPRA) {
+ t.Helper()
+ select {
+ case e := <-ndpDisp.dhcpv6ConfigurationC:
+ if diff := cmp.Diff(ndpDHCPv6Event{nicID: nicID, configuration: configuration}, e, cmp.AllowUnexported(e)); diff != "" {
+ t.Errorf("dhcpv6 event mismatch (-want +got):\n%s", diff)
+ }
+ default:
+ t.Fatal("expected DHCPv6 configuration event")
+ }
+ }
+
+ expectNoDHCPv6Event := func() {
+ t.Helper()
+ select {
+ case <-ndpDisp.dhcpv6ConfigurationC:
+ t.Fatal("unexpected DHCPv6 configuration event")
+ default:
+ }
+ }
+
+ // The initial DHCPv6 configuration should be stack.DHCPv6NoConfiguration.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, false))
+ expectNoDHCPv6Event()
+
+ // Receive an RA that updates the DHCPv6 configuration to Other
+ // 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()
+
+ // Receive an RA that updates the DHCPv6 configuration to Managed Address.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, true, false))
+ expectDHCPv6Event(stack.DHCPv6ManagedAddress)
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, true, false))
+ expectNoDHCPv6Event()
+
+ // Receive an RA that updates the DHCPv6 configuration to none.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, false))
+ expectDHCPv6Event(stack.DHCPv6NoConfiguration)
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, false))
+ expectNoDHCPv6Event()
+
+ // Receive an RA that updates the DHCPv6 configuration to Managed Address.
+ //
+ // Note, when the M flag is set, the O flag is redundant.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, true, true))
+ expectDHCPv6Event(stack.DHCPv6ManagedAddress)
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, true, true))
+ expectNoDHCPv6Event()
+ // Even though the DHCPv6 flags are different, the effective configuration is
+ // the same so we should not receive a new event.
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, true, false))
+ expectNoDHCPv6Event()
+ e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, true, true))
+ expectNoDHCPv6Event()
+
+ // 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()
+}
diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go
index f2d338bd1..d50693634 100644
--- a/pkg/tcpip/stack/nic.go
+++ b/pkg/tcpip/stack/nic.go
@@ -16,9 +16,9 @@ package stack
import (
"strings"
- "sync"
"sync/atomic"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/header"
@@ -27,10 +27,11 @@ import (
// NIC represents a "network interface card" to which the networking stack is
// attached.
type NIC struct {
- stack *Stack
- id tcpip.NICID
- name string
- linkEP LinkEndpoint
+ stack *Stack
+ id tcpip.NICID
+ name string
+ linkEP LinkEndpoint
+ context NICContext
mu sync.RWMutex
spoofing bool
@@ -84,7 +85,7 @@ const (
)
// newNIC returns a new NIC using the default NDP configurations from stack.
-func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint) *NIC {
+func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, ctx NICContext) *NIC {
// TODO(b/141011931): Validate a LinkEndpoint (ep) is valid. For
// example, make sure that the link address it provides is a valid
// unicast ethernet address.
@@ -98,6 +99,7 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint) *NIC {
id: id,
name: name,
linkEP: ep,
+ context: ctx,
primary: make(map[tcpip.NetworkProtocolNumber][]*referencedNetworkEndpoint),
endpoints: make(map[NetworkEndpointID]*referencedNetworkEndpoint),
mcastJoins: make(map[NetworkEndpointID]int32),
diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go
index 754323e82..ec91f60dd 100644
--- a/pkg/tcpip/stack/registration.go
+++ b/pkg/tcpip/stack/registration.go
@@ -234,15 +234,15 @@ type NetworkEndpoint interface {
// WritePacket writes a packet to the given destination address and
// protocol. It sets pkt.NetworkHeader. pkt.TransportHeader must have
// already been set.
- WritePacket(r *Route, gso *GSO, params NetworkHeaderParams, loop PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error
+ WritePacket(r *Route, gso *GSO, params NetworkHeaderParams, pkt tcpip.PacketBuffer) *tcpip.Error
// WritePackets writes packets to the given destination address and
// protocol. pkts must not be zero length.
- WritePackets(r *Route, gso *GSO, pkts []tcpip.PacketBuffer, params NetworkHeaderParams, loop PacketLooping) (int, *tcpip.Error)
+ WritePackets(r *Route, gso *GSO, pkts []tcpip.PacketBuffer, params NetworkHeaderParams) (int, *tcpip.Error)
// WriteHeaderIncludedPacket writes a packet that includes a network
// header to the given destination address.
- WriteHeaderIncludedPacket(r *Route, loop PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error
+ WriteHeaderIncludedPacket(r *Route, pkt tcpip.PacketBuffer) *tcpip.Error
// ID returns the network protocol endpoint ID.
ID() *NetworkEndpointID
diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go
index 34307ae07..517f4b941 100644
--- a/pkg/tcpip/stack/route.go
+++ b/pkg/tcpip/stack/route.go
@@ -158,7 +158,7 @@ func (r *Route) WritePacket(gso *GSO, params NetworkHeaderParams, pkt tcpip.Pack
return tcpip.ErrInvalidEndpointState
}
- err := r.ref.ep.WritePacket(r, gso, params, r.Loop, pkt)
+ err := r.ref.ep.WritePacket(r, gso, params, pkt)
if err != nil {
r.Stats().IP.OutgoingPacketErrors.Increment()
} else {
@@ -174,7 +174,7 @@ func (r *Route) WritePackets(gso *GSO, pkts []tcpip.PacketBuffer, params Network
return 0, tcpip.ErrInvalidEndpointState
}
- n, err := r.ref.ep.WritePackets(r, gso, pkts, params, r.Loop)
+ n, err := r.ref.ep.WritePackets(r, gso, pkts, params)
if err != nil {
r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(len(pkts) - n))
}
@@ -195,7 +195,7 @@ func (r *Route) WriteHeaderIncludedPacket(pkt tcpip.PacketBuffer) *tcpip.Error {
return tcpip.ErrInvalidEndpointState
}
- if err := r.ref.ep.WriteHeaderIncludedPacket(r, r.Loop, pkt); err != nil {
+ if err := r.ref.ep.WriteHeaderIncludedPacket(r, pkt); err != nil {
r.Stats().IP.OutgoingPacketErrors.Increment()
return err
}
diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go
index 807f910f6..a47ceba54 100644
--- a/pkg/tcpip/stack/stack.go
+++ b/pkg/tcpip/stack/stack.go
@@ -21,13 +21,13 @@ package stack
import (
"encoding/binary"
- "sync"
"sync/atomic"
"time"
"golang.org/x/time/rate"
"gvisor.dev/gvisor/pkg/rand"
"gvisor.dev/gvisor/pkg/sleep"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/header"
@@ -796,9 +796,30 @@ func (s *Stack) NewPacketEndpoint(cooked bool, netProto tcpip.NetworkProtocolNum
return s.rawFactory.NewPacketEndpoint(s, cooked, netProto, waiterQueue)
}
-// createNIC creates a NIC with the provided id and link-layer endpoint, and
-// optionally enable it.
-func (s *Stack) createNIC(id tcpip.NICID, name string, ep LinkEndpoint, enabled bool) *tcpip.Error {
+// NICContext is an opaque pointer used to store client-supplied NIC metadata.
+type NICContext interface{}
+
+// NICOptions specifies the configuration of a NIC as it is being created.
+// The zero value creates an enabled, unnamed NIC.
+type NICOptions struct {
+ // Name specifies the name of the NIC.
+ Name string
+
+ // Disabled specifies whether to avoid calling Attach on the passed
+ // LinkEndpoint.
+ Disabled bool
+
+ // Context specifies user-defined data that will be returned in stack.NICInfo
+ // for the NIC. Clients of this library can use it to add metadata that
+ // should be tracked alongside a NIC, to avoid having to keep a
+ // map[tcpip.NICID]metadata mirroring stack.Stack's nic map.
+ Context NICContext
+}
+
+// CreateNICWithOptions creates a NIC with the provided id, LinkEndpoint, and
+// NICOptions. See the documentation on type NICOptions for details on how
+// NICs can be configured.
+func (s *Stack) CreateNICWithOptions(id tcpip.NICID, ep LinkEndpoint, opts NICOptions) *tcpip.Error {
s.mu.Lock()
defer s.mu.Unlock()
@@ -807,38 +828,20 @@ func (s *Stack) createNIC(id tcpip.NICID, name string, ep LinkEndpoint, enabled
return tcpip.ErrDuplicateNICID
}
- n := newNIC(s, id, name, ep)
+ n := newNIC(s, id, opts.Name, ep, opts.Context)
s.nics[id] = n
- if enabled {
+ if !opts.Disabled {
return n.enable()
}
return nil
}
-// CreateNIC creates a NIC with the provided id and link-layer endpoint.
+// CreateNIC creates a NIC with the provided id and LinkEndpoint and calls
+// `LinkEndpoint.Attach` to start delivering packets to it.
func (s *Stack) CreateNIC(id tcpip.NICID, ep LinkEndpoint) *tcpip.Error {
- return s.createNIC(id, "", ep, true)
-}
-
-// CreateNamedNIC creates a NIC with the provided id and link-layer endpoint,
-// and a human-readable name.
-func (s *Stack) CreateNamedNIC(id tcpip.NICID, name string, ep LinkEndpoint) *tcpip.Error {
- return s.createNIC(id, name, ep, true)
-}
-
-// CreateDisabledNIC creates a NIC with the provided id and link-layer endpoint,
-// but leave it disable. Stack.EnableNIC must be called before the link-layer
-// endpoint starts delivering packets to it.
-func (s *Stack) CreateDisabledNIC(id tcpip.NICID, ep LinkEndpoint) *tcpip.Error {
- return s.createNIC(id, "", ep, false)
-}
-
-// CreateDisabledNamedNIC is a combination of CreateNamedNIC and
-// CreateDisabledNIC.
-func (s *Stack) CreateDisabledNamedNIC(id tcpip.NICID, name string, ep LinkEndpoint) *tcpip.Error {
- return s.createNIC(id, name, ep, false)
+ return s.CreateNICWithOptions(id, ep, NICOptions{})
}
// EnableNIC enables the given NIC so that the link-layer endpoint can start
@@ -892,6 +895,18 @@ type NICInfo struct {
MTU uint32
Stats NICStats
+
+ // Context is user-supplied data optionally supplied in CreateNICWithOptions.
+ // See type NICOptions for more details.
+ Context NICContext
+}
+
+// HasNIC returns true if the NICID is defined in the stack.
+func (s *Stack) HasNIC(id tcpip.NICID) bool {
+ s.mu.RLock()
+ _, ok := s.nics[id]
+ s.mu.RUnlock()
+ return ok
}
// NICInfo returns a map of NICIDs to their associated information.
@@ -914,6 +929,7 @@ func (s *Stack) NICInfo() map[tcpip.NICID]NICInfo {
Flags: flags,
MTU: nic.linkEP.MTU(),
Stats: nic.stats,
+ Context: nic.context,
}
}
return nics
diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go
index bf057745e..44e5229cc 100644
--- a/pkg/tcpip/stack/stack_test.go
+++ b/pkg/tcpip/stack/stack_test.go
@@ -124,7 +124,7 @@ func (f *fakeNetworkEndpoint) Capabilities() stack.LinkEndpointCapabilities {
return f.ep.Capabilities()
}
-func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, loop stack.PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error {
+func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt tcpip.PacketBuffer) *tcpip.Error {
// Increment the sent packet count in the protocol descriptor.
f.proto.sendPacketCount[int(r.RemoteAddress[0])%len(f.proto.sendPacketCount)]++
@@ -135,7 +135,7 @@ func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, params
b[1] = f.id.LocalAddress[0]
b[2] = byte(params.Protocol)
- if loop&stack.PacketLoop != 0 {
+ if r.Loop&stack.PacketLoop != 0 {
views := make([]buffer.View, 1, 1+len(pkt.Data.Views()))
views[0] = pkt.Header.View()
views = append(views, pkt.Data.Views()...)
@@ -143,7 +143,7 @@ func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, params
Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views),
})
}
- if loop&stack.PacketOut == 0 {
+ if r.Loop&stack.PacketOut == 0 {
return nil
}
@@ -151,11 +151,11 @@ func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, params
}
// WritePackets implements stack.LinkEndpoint.WritePackets.
-func (f *fakeNetworkEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) {
+func (f *fakeNetworkEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, params stack.NetworkHeaderParams) (int, *tcpip.Error) {
panic("not implemented")
}
-func (*fakeNetworkEndpoint) WriteHeaderIncludedPacket(r *stack.Route, loop stack.PacketLooping, pkt tcpip.PacketBuffer) *tcpip.Error {
+func (*fakeNetworkEndpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt tcpip.PacketBuffer) *tcpip.Error {
return tcpip.ErrNotSupported
}
@@ -2001,6 +2001,46 @@ func TestNICAutoGenAddr(t *testing.T) {
}
}
+// TestNICContextPreservation tests that you can read out via stack.NICInfo the
+// Context data you pass via NICContext.Context in stack.CreateNICWithOptions.
+func TestNICContextPreservation(t *testing.T) {
+ var ctx *int
+ tests := []struct {
+ name string
+ opts stack.NICOptions
+ want stack.NICContext
+ }{
+ {
+ "context_set",
+ stack.NICOptions{Context: ctx},
+ ctx,
+ },
+ {
+ "context_not_set",
+ stack.NICOptions{},
+ nil,
+ },
+ }
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ s := stack.New(stack.Options{})
+ id := tcpip.NICID(1)
+ ep := channel.New(0, 0, tcpip.LinkAddress("\x00\x00\x00\x00\x00\x00"))
+ if err := s.CreateNICWithOptions(id, ep, test.opts); err != nil {
+ t.Fatalf("got stack.CreateNICWithOptions(%d, %+v, %+v) = %s, want nil", id, ep, test.opts, err)
+ }
+ nicinfos := s.NICInfo()
+ nicinfo, ok := nicinfos[id]
+ if !ok {
+ t.Fatalf("got nicinfos[%d] = _, %t, want _, true; nicinfos = %+v", id, ok, nicinfos)
+ }
+ if got, want := nicinfo.Context == test.want, true; got != want {
+ t.Fatal("got nicinfo.Context == ctx = %t, want %t; nicinfo.Context = %p, ctx = %p", got, want, nicinfo.Context, test.want)
+ }
+ })
+ }
+}
+
// TestNICAutoGenAddrWithOpaque tests the auto-generation of IPv6 link-local
// addresses with opaque interface identifiers. Link Local addresses should
// always be generated with opaque IIDs if configured to use them, even if the
@@ -2097,8 +2137,9 @@ func TestNICAutoGenAddrWithOpaque(t *testing.T) {
e := channel.New(10, 1280, test.linkAddr)
s := stack.New(opts)
- if err := s.CreateNamedNIC(nicID, test.nicName, e); err != nil {
- t.Fatalf("CreateNamedNIC(%d, %q, _) = %s", nicID, test.nicName, err)
+ nicOpts := stack.NICOptions{Name: test.nicName}
+ if err := s.CreateNICWithOptions(nicID, e, nicOpts); err != nil {
+ t.Fatalf("CreateNICWithOptions(%d, _, %+v) = %s", nicID, opts, err)
}
addr, err := s.GetMainNICAddress(nicID, header.IPv6ProtocolNumber)
@@ -2156,8 +2197,9 @@ func TestNoLinkLocalAutoGenForLoopbackNIC(t *testing.T) {
e := loopback.New()
s := stack.New(opts)
- if err := s.CreateNamedNIC(nicID, nicName, e); err != nil {
- t.Fatalf("CreateNamedNIC(%d, %q, _) = %s", nicID, nicName, err)
+ nicOpts := stack.NICOptions{Name: nicName}
+ if err := s.CreateNICWithOptions(nicID, e, nicOpts); err != nil {
+ t.Fatalf("CreateNICWithOptions(%d, _, %+v) = %s", nicID, nicOpts, err)
}
addr, err := s.GetMainNICAddress(nicID, header.IPv6ProtocolNumber)
diff --git a/pkg/tcpip/stack/transport_demuxer.go b/pkg/tcpip/stack/transport_demuxer.go
index 67c21be42..f384a91de 100644
--- a/pkg/tcpip/stack/transport_demuxer.go
+++ b/pkg/tcpip/stack/transport_demuxer.go
@@ -18,8 +18,8 @@ import (
"fmt"
"math/rand"
"sort"
- "sync"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/hash/jenkins"
"gvisor.dev/gvisor/pkg/tcpip/header"
diff --git a/pkg/tcpip/stack/transport_demuxer_test.go b/pkg/tcpip/stack/transport_demuxer_test.go
index 3b28b06d0..5e9237de9 100644
--- a/pkg/tcpip/stack/transport_demuxer_test.go
+++ b/pkg/tcpip/stack/transport_demuxer_test.go
@@ -41,7 +41,7 @@ const (
type testContext struct {
t *testing.T
- linkEPs map[string]*channel.Endpoint
+ linkEps map[tcpip.NICID]*channel.Endpoint
s *stack.Stack
ep tcpip.Endpoint
@@ -61,35 +61,29 @@ func (c *testContext) createV6Endpoint(v6only bool) {
c.t.Fatalf("NewEndpoint failed: %v", err)
}
- var v tcpip.V6OnlyOption
- if v6only {
- v = 1
- }
- if err := c.ep.SetSockOpt(v); err != nil {
+ if err := c.ep.SetSockOptBool(tcpip.V6OnlyOption, v6only); err != nil {
c.t.Fatalf("SetSockOpt failed: %v", err)
}
}
-// newDualTestContextMultiNic creates the testing context and also linkEpNames
-// named NICs.
-func newDualTestContextMultiNic(t *testing.T, mtu uint32, linkEpNames []string) *testContext {
+// newDualTestContextMultiNIC creates the testing context and also linkEpIDs NICs.
+func newDualTestContextMultiNIC(t *testing.T, mtu uint32, linkEpIDs []tcpip.NICID) *testContext {
s := stack.New(stack.Options{
NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()},
TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}})
- linkEPs := make(map[string]*channel.Endpoint)
- for i, linkEpName := range linkEpNames {
- channelEP := channel.New(256, mtu, "")
- nicID := tcpip.NICID(i + 1)
- if err := s.CreateNamedNIC(nicID, linkEpName, channelEP); err != nil {
+ linkEps := make(map[tcpip.NICID]*channel.Endpoint)
+ for _, linkEpID := range linkEpIDs {
+ channelEp := channel.New(256, mtu, "")
+ if err := s.CreateNIC(linkEpID, channelEp); err != nil {
t.Fatalf("CreateNIC failed: %v", err)
}
- linkEPs[linkEpName] = channelEP
+ linkEps[linkEpID] = channelEp
- if err := s.AddAddress(nicID, ipv4.ProtocolNumber, stackAddr); err != nil {
+ if err := s.AddAddress(linkEpID, ipv4.ProtocolNumber, stackAddr); err != nil {
t.Fatalf("AddAddress IPv4 failed: %v", err)
}
- if err := s.AddAddress(nicID, ipv6.ProtocolNumber, stackV6Addr); err != nil {
+ if err := s.AddAddress(linkEpID, ipv6.ProtocolNumber, stackV6Addr); err != nil {
t.Fatalf("AddAddress IPv6 failed: %v", err)
}
}
@@ -108,7 +102,7 @@ func newDualTestContextMultiNic(t *testing.T, mtu uint32, linkEpNames []string)
return &testContext{
t: t,
s: s,
- linkEPs: linkEPs,
+ linkEps: linkEps,
}
}
@@ -125,7 +119,7 @@ func newPayload() []byte {
return b
}
-func (c *testContext) sendV6Packet(payload []byte, h *headers, linkEpName string) {
+func (c *testContext) sendV6Packet(payload []byte, h *headers, linkEpID tcpip.NICID) {
// Allocate a buffer for data and headers.
buf := buffer.NewView(header.UDPMinimumSize + header.IPv6MinimumSize + len(payload))
copy(buf[len(buf)-len(payload):], payload)
@@ -156,7 +150,7 @@ func (c *testContext) sendV6Packet(payload []byte, h *headers, linkEpName string
u.SetChecksum(^u.CalculateChecksum(xsum))
// Inject packet.
- c.linkEPs[linkEpName].InjectInbound(ipv6.ProtocolNumber, tcpip.PacketBuffer{
+ c.linkEps[linkEpID].InjectInbound(ipv6.ProtocolNumber, tcpip.PacketBuffer{
Data: buf.ToVectorisedView(),
})
}
@@ -186,7 +180,7 @@ func TestTransportDemuxerRegister(t *testing.T) {
func TestDistribution(t *testing.T) {
type endpointSockopts struct {
reuse int
- bindToDevice string
+ bindToDevice tcpip.NICID
}
for _, test := range []struct {
name string
@@ -194,71 +188,71 @@ func TestDistribution(t *testing.T) {
endpoints []endpointSockopts
// wantedDistribution is the wanted ratio of packets received on each
// endpoint for each NIC on which packets are injected.
- wantedDistributions map[string][]float64
+ wantedDistributions map[tcpip.NICID][]float64
}{
{
"BindPortReuse",
// 5 endpoints that all have reuse set.
[]endpointSockopts{
- endpointSockopts{1, ""},
- endpointSockopts{1, ""},
- endpointSockopts{1, ""},
- endpointSockopts{1, ""},
- endpointSockopts{1, ""},
+ {1, 0},
+ {1, 0},
+ {1, 0},
+ {1, 0},
+ {1, 0},
},
- map[string][]float64{
+ map[tcpip.NICID][]float64{
// Injected packets on dev0 get distributed evenly.
- "dev0": []float64{0.2, 0.2, 0.2, 0.2, 0.2},
+ 1: {0.2, 0.2, 0.2, 0.2, 0.2},
},
},
{
"BindToDevice",
// 3 endpoints with various bindings.
[]endpointSockopts{
- endpointSockopts{0, "dev0"},
- endpointSockopts{0, "dev1"},
- endpointSockopts{0, "dev2"},
+ {0, 1},
+ {0, 2},
+ {0, 3},
},
- map[string][]float64{
+ map[tcpip.NICID][]float64{
// Injected packets on dev0 go only to the endpoint bound to dev0.
- "dev0": []float64{1, 0, 0},
+ 1: {1, 0, 0},
// Injected packets on dev1 go only to the endpoint bound to dev1.
- "dev1": []float64{0, 1, 0},
+ 2: {0, 1, 0},
// Injected packets on dev2 go only to the endpoint bound to dev2.
- "dev2": []float64{0, 0, 1},
+ 3: {0, 0, 1},
},
},
{
"ReuseAndBindToDevice",
// 6 endpoints with various bindings.
[]endpointSockopts{
- endpointSockopts{1, "dev0"},
- endpointSockopts{1, "dev0"},
- endpointSockopts{1, "dev1"},
- endpointSockopts{1, "dev1"},
- endpointSockopts{1, "dev1"},
- endpointSockopts{1, ""},
+ {1, 1},
+ {1, 1},
+ {1, 2},
+ {1, 2},
+ {1, 2},
+ {1, 0},
},
- map[string][]float64{
+ map[tcpip.NICID][]float64{
// Injected packets on dev0 get distributed among endpoints bound to
// dev0.
- "dev0": []float64{0.5, 0.5, 0, 0, 0, 0},
+ 1: {0.5, 0.5, 0, 0, 0, 0},
// Injected packets on dev1 get distributed among endpoints bound to
// dev1 or unbound.
- "dev1": []float64{0, 0, 1. / 3, 1. / 3, 1. / 3, 0},
+ 2: {0, 0, 1. / 3, 1. / 3, 1. / 3, 0},
// Injected packets on dev999 go only to the unbound.
- "dev999": []float64{0, 0, 0, 0, 0, 1},
+ 1000: {0, 0, 0, 0, 0, 1},
},
},
} {
t.Run(test.name, func(t *testing.T) {
for device, wantedDistribution := range test.wantedDistributions {
- t.Run(device, func(t *testing.T) {
- var devices []string
+ t.Run(string(device), func(t *testing.T) {
+ var devices []tcpip.NICID
for d := range test.wantedDistributions {
devices = append(devices, d)
}
- c := newDualTestContextMultiNic(t, defaultMTU, devices)
+ c := newDualTestContextMultiNIC(t, defaultMTU, devices)
defer c.cleanup()
c.createV6Endpoint(false)
diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go
index 748ce4ea5..f50604a8a 100644
--- a/pkg/tcpip/stack/transport_test.go
+++ b/pkg/tcpip/stack/transport_test.go
@@ -102,13 +102,23 @@ func (*fakeTransportEndpoint) SetSockOpt(interface{}) *tcpip.Error {
return tcpip.ErrInvalidEndpointState
}
+// SetSockOptBool sets a socket option. Currently not supported.
+func (*fakeTransportEndpoint) SetSockOptBool(tcpip.SockOptBool, bool) *tcpip.Error {
+ return tcpip.ErrInvalidEndpointState
+}
+
// SetSockOptInt sets a socket option. Currently not supported.
-func (*fakeTransportEndpoint) SetSockOptInt(tcpip.SockOpt, int) *tcpip.Error {
+func (*fakeTransportEndpoint) SetSockOptInt(tcpip.SockOptInt, int) *tcpip.Error {
return tcpip.ErrInvalidEndpointState
}
+// GetSockOptBool implements tcpip.Endpoint.GetSockOptBool.
+func (*fakeTransportEndpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) {
+ return false, tcpip.ErrUnknownProtocolOption
+}
+
// GetSockOptInt implements tcpip.Endpoint.GetSockOptInt.
-func (*fakeTransportEndpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) {
+func (*fakeTransportEndpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) {
return -1, tcpip.ErrUnknownProtocolOption
}
diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go
index d02950c7a..a0a3d1e53 100644
--- a/pkg/tcpip/tcpip.go
+++ b/pkg/tcpip/tcpip.go
@@ -35,10 +35,10 @@ import (
"reflect"
"strconv"
"strings"
- "sync"
"sync/atomic"
"time"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/waiter"
)
@@ -422,17 +422,25 @@ type Endpoint interface {
// SetSockOpt sets a socket option. opt should be one of the *Option types.
SetSockOpt(opt interface{}) *Error
+ // SetSockOptBool sets a socket option, for simple cases where a value
+ // has the bool type.
+ SetSockOptBool(opt SockOptBool, v bool) *Error
+
// SetSockOptInt sets a socket option, for simple cases where a value
// has the int type.
- SetSockOptInt(opt SockOpt, v int) *Error
+ SetSockOptInt(opt SockOptInt, v int) *Error
// GetSockOpt gets a socket option. opt should be a pointer to one of the
// *Option types.
GetSockOpt(opt interface{}) *Error
+ // GetSockOptBool gets a socket option for simple cases where a return
+ // value has the bool type.
+ GetSockOptBool(SockOptBool) (bool, *Error)
+
// GetSockOptInt gets a socket option for simple cases where a return
// value has the int type.
- GetSockOptInt(SockOpt) (int, *Error)
+ GetSockOptInt(SockOptInt) (int, *Error)
// State returns a socket's lifecycle state. The returned value is
// protocol-specific and is primarily used for diagnostics.
@@ -484,13 +492,22 @@ type WriteOptions struct {
Atomic bool
}
-// SockOpt represents socket options which values have the int type.
-type SockOpt int
+// SockOptBool represents socket options which values have the bool type.
+type SockOptBool int
+
+const (
+ // V6OnlyOption is used by {G,S}etSockOptBool to specify whether an IPv6
+ // socket is to be restricted to sending and receiving IPv6 packets only.
+ V6OnlyOption SockOptBool = iota
+)
+
+// SockOptInt represents socket options which values have the int type.
+type SockOptInt int
const (
// ReceiveQueueSizeOption is used in GetSockOptInt to specify that the
// number of unread bytes in the input buffer should be returned.
- ReceiveQueueSizeOption SockOpt = iota
+ ReceiveQueueSizeOption SockOptInt = iota
// SendBufferSizeOption is used by SetSockOptInt/GetSockOptInt to
// specify the send buffer size option.
@@ -517,10 +534,6 @@ const (
// the endpoint should be cleared and returned.
type ErrorOption struct{}
-// V6OnlyOption is used by SetSockOpt/GetSockOpt to specify whether an IPv6
-// socket is to be restricted to sending and receiving IPv6 packets only.
-type V6OnlyOption int
-
// CorkOption is used by SetSockOpt/GetSockOpt to specify if data should be
// held until segments are full by the TCP transport protocol.
type CorkOption int
@@ -535,7 +548,7 @@ type ReusePortOption int
// BindToDeviceOption is used by SetSockOpt/GetSockOpt to specify that sockets
// should bind only on a specific NIC.
-type BindToDeviceOption string
+type BindToDeviceOption NICID
// QuickAckOption is stubbed out in SetSockOpt/GetSockOpt.
type QuickAckOption int
diff --git a/pkg/tcpip/timer.go b/pkg/tcpip/timer.go
new file mode 100644
index 000000000..f5f01f32f
--- /dev/null
+++ b/pkg/tcpip/timer.go
@@ -0,0 +1,161 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package tcpip
+
+import (
+ "sync"
+ "time"
+)
+
+// cancellableTimerInstance is a specific instance of CancellableTimer.
+//
+// Different instances are created each time CancellableTimer is Reset so each
+// timer has its own earlyReturn signal. This is to address a bug when a
+// CancellableTimer is stopped and reset in quick succession resulting in a
+// timer instance's earlyReturn signal being affected or seen by another timer
+// instance.
+//
+// Consider the following sceneario where timer instances share a common
+// earlyReturn signal (T1 creates, stops and resets a Cancellable timer under a
+// lock L; T2, T3, T4 and T5 are goroutines that handle the first (A), second
+// (B), third (C), and fourth (D) instance of the timer firing, respectively):
+// T1: Obtain L
+// T1: Create a new CancellableTimer w/ lock L (create instance A)
+// T2: instance A fires, blocked trying to obtain L.
+// T1: Attempt to stop instance A (set earlyReturn = true)
+// T1: Reset timer (create instance B)
+// T3: instance B fires, blocked trying to obtain L.
+// T1: Attempt to stop instance B (set earlyReturn = true)
+// T1: Reset timer (create instance C)
+// T4: instance C fires, blocked trying to obtain L.
+// T1: Attempt to stop instance C (set earlyReturn = true)
+// T1: Reset timer (create instance D)
+// T5: instance D fires, blocked trying to obtain L.
+// T1: Release L
+//
+// Now that T1 has released L, any of the 4 timer instances can take L and check
+// earlyReturn. If the timers simply check earlyReturn and then do nothing
+// further, then instance D will never early return even though it was not
+// requested to stop. If the timers reset earlyReturn before early returning,
+// then all but one of the timers will do work when only one was expected to.
+// If CancellableTimer resets earlyReturn when resetting, then all the timers
+// will fire (again, when only one was expected to).
+//
+// To address the above concerns the simplest solution was to give each timer
+// its own earlyReturn signal.
+type cancellableTimerInstance struct {
+ timer *time.Timer
+
+ // Used to inform the timer to early return when it gets stopped while the
+ // lock the timer tries to obtain when fired is held (T1 is a goroutine that
+ // tries to cancel the timer and T2 is the goroutine that handles the timer
+ // firing):
+ // T1: Obtain the lock, then call StopLocked()
+ // T2: timer fires, and gets blocked on obtaining the lock
+ // T1: Releases lock
+ // T2: Obtains lock does unintended work
+ //
+ // To resolve this, T1 will check to see if the timer already fired, and
+ // inform the timer using earlyReturn to return early so that once T2 obtains
+ // the lock, it will see that it is set to true and do nothing further.
+ earlyReturn *bool
+}
+
+// stop stops the timer instance t from firing if it hasn't fired already. If it
+// has fired and is blocked at obtaining the lock, earlyReturn will be set to
+// true so that it will early return when it obtains the lock.
+func (t *cancellableTimerInstance) stop() {
+ if t.timer != nil {
+ t.timer.Stop()
+ *t.earlyReturn = true
+ }
+}
+
+// CancellableTimer is a timer that does some work and can be safely cancelled
+// when it fires at the same time some "related work" is being done.
+//
+// The term "related work" is defined as some work that needs to be done while
+// holding some lock that the timer must also hold while doing some work.
+type CancellableTimer struct {
+ // The active instance of a cancellable timer.
+ instance cancellableTimerInstance
+
+ // locker is the lock taken by the timer immediately after it fires and must
+ // be held when attempting to stop the timer.
+ //
+ // Must never change after being assigned.
+ locker sync.Locker
+
+ // fn is the function that will be called when a timer fires and has not been
+ // signaled to early return.
+ //
+ // fn MUST NOT attempt to lock locker.
+ //
+ // Must never change after being assigned.
+ fn func()
+}
+
+// StopLocked prevents the Timer from firing if it has not fired already.
+//
+// If the timer is blocked on obtaining the t.locker lock when StopLocked is
+// called, it will early return instead of calling t.fn.
+//
+// Note, t will be modified.
+//
+// t.locker MUST be locked.
+func (t *CancellableTimer) StopLocked() {
+ t.instance.stop()
+
+ // Nothing to do with the stopped instance anymore.
+ t.instance = cancellableTimerInstance{}
+}
+
+// Reset changes the timer to expire after duration d.
+//
+// Note, t will be modified.
+//
+// Reset should only be called on stopped or expired timers. To be safe, callers
+// should always call StopLocked before calling Reset.
+func (t *CancellableTimer) Reset(d time.Duration) {
+ // Create a new instance.
+ earlyReturn := false
+ t.instance = cancellableTimerInstance{
+ timer: time.AfterFunc(d, func() {
+ t.locker.Lock()
+ defer t.locker.Unlock()
+
+ if earlyReturn {
+ // If we reach this point, it means that the timer fired while another
+ // goroutine called StopLocked while it had the lock. Simply return
+ // here and do nothing further.
+ earlyReturn = false
+ return
+ }
+
+ t.fn()
+ }),
+ earlyReturn: &earlyReturn,
+ }
+}
+
+// MakeCancellableTimer returns an unscheduled CancellableTimer with the given
+// locker and fn.
+//
+// fn MUST NOT attempt to lock locker.
+//
+// Callers must call Reset to schedule the timer to fire.
+func MakeCancellableTimer(locker sync.Locker, fn func()) CancellableTimer {
+ return CancellableTimer{locker: locker, fn: fn}
+}
diff --git a/pkg/tcpip/timer_test.go b/pkg/tcpip/timer_test.go
new file mode 100644
index 000000000..2d20f7ef3
--- /dev/null
+++ b/pkg/tcpip/timer_test.go
@@ -0,0 +1,236 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package tcpip_test
+
+import (
+ "sync"
+ "testing"
+ "time"
+
+ "gvisor.dev/gvisor/pkg/tcpip"
+)
+
+const (
+ shortDuration = 1 * time.Nanosecond
+ middleDuration = 100 * time.Millisecond
+ longDuration = 1 * time.Second
+)
+
+func TestCancellableTimerFire(t *testing.T) {
+ t.Parallel()
+
+ ch := make(chan struct{})
+ var lock sync.Mutex
+
+ timer := tcpip.MakeCancellableTimer(&lock, func() {
+ ch <- struct{}{}
+ })
+ timer.Reset(shortDuration)
+
+ // Wait for timer to fire.
+ select {
+ case <-ch:
+ case <-time.After(middleDuration):
+ t.Fatal("timed out waiting for timer to fire")
+ }
+
+ // The timer should have fired only once.
+ select {
+ case <-ch:
+ t.Fatal("no other timers should have fired")
+ case <-time.After(middleDuration):
+ }
+}
+
+func TestCancellableTimerResetFromLongDuration(t *testing.T) {
+ t.Parallel()
+
+ ch := make(chan struct{})
+ var lock sync.Mutex
+
+ timer := tcpip.MakeCancellableTimer(&lock, func() { ch <- struct{}{} })
+ timer.Reset(middleDuration)
+
+ lock.Lock()
+ timer.StopLocked()
+ lock.Unlock()
+
+ timer.Reset(shortDuration)
+
+ // Wait for timer to fire.
+ select {
+ case <-ch:
+ case <-time.After(middleDuration):
+ t.Fatal("timed out waiting for timer to fire")
+ }
+
+ // The timer should have fired only once.
+ select {
+ case <-ch:
+ t.Fatal("no other timers should have fired")
+ case <-time.After(middleDuration):
+ }
+}
+
+func TestCancellableTimerResetFromShortDuration(t *testing.T) {
+ t.Parallel()
+
+ ch := make(chan struct{})
+ var lock sync.Mutex
+
+ lock.Lock()
+ timer := tcpip.MakeCancellableTimer(&lock, func() { ch <- struct{}{} })
+ timer.Reset(shortDuration)
+ timer.StopLocked()
+ lock.Unlock()
+
+ // Wait for timer to fire if it wasn't correctly stopped.
+ select {
+ case <-ch:
+ t.Fatal("timer fired after being stopped")
+ case <-time.After(middleDuration):
+ }
+
+ timer.Reset(shortDuration)
+
+ // Wait for timer to fire.
+ select {
+ case <-ch:
+ case <-time.After(middleDuration):
+ t.Fatal("timed out waiting for timer to fire")
+ }
+
+ // The timer should have fired only once.
+ select {
+ case <-ch:
+ t.Fatal("no other timers should have fired")
+ case <-time.After(middleDuration):
+ }
+}
+
+func TestCancellableTimerImmediatelyStop(t *testing.T) {
+ t.Parallel()
+
+ ch := make(chan struct{})
+ var lock sync.Mutex
+
+ for i := 0; i < 1000; i++ {
+ lock.Lock()
+ timer := tcpip.MakeCancellableTimer(&lock, func() { ch <- struct{}{} })
+ timer.Reset(shortDuration)
+ timer.StopLocked()
+ lock.Unlock()
+ }
+
+ // Wait for timer to fire if it wasn't correctly stopped.
+ select {
+ case <-ch:
+ t.Fatal("timer fired after being stopped")
+ case <-time.After(middleDuration):
+ }
+}
+
+func TestCancellableTimerStoppedResetWithoutLock(t *testing.T) {
+ t.Parallel()
+
+ ch := make(chan struct{})
+ var lock sync.Mutex
+
+ lock.Lock()
+ timer := tcpip.MakeCancellableTimer(&lock, func() { ch <- struct{}{} })
+ timer.Reset(shortDuration)
+ timer.StopLocked()
+ lock.Unlock()
+
+ for i := 0; i < 10; i++ {
+ timer.Reset(middleDuration)
+
+ lock.Lock()
+ // Sleep until the timer fires and gets blocked trying to take the lock.
+ time.Sleep(middleDuration * 2)
+ timer.StopLocked()
+ lock.Unlock()
+ }
+
+ // Wait for double the duration so timers that weren't correctly stopped can
+ // fire.
+ select {
+ case <-ch:
+ t.Fatal("timer fired after being stopped")
+ case <-time.After(middleDuration * 2):
+ }
+}
+
+func TestManyCancellableTimerResetAfterBlockedOnLock(t *testing.T) {
+ t.Parallel()
+
+ ch := make(chan struct{})
+ var lock sync.Mutex
+
+ lock.Lock()
+ timer := tcpip.MakeCancellableTimer(&lock, func() { ch <- struct{}{} })
+ timer.Reset(shortDuration)
+ for i := 0; i < 10; i++ {
+ // Sleep until the timer fires and gets blocked trying to take the lock.
+ time.Sleep(middleDuration)
+ timer.StopLocked()
+ timer.Reset(shortDuration)
+ }
+ lock.Unlock()
+
+ // Wait for double the duration for the last timer to fire.
+ select {
+ case <-ch:
+ case <-time.After(middleDuration):
+ t.Fatal("timed out waiting for timer to fire")
+ }
+
+ // The timer should have fired only once.
+ select {
+ case <-ch:
+ t.Fatal("no other timers should have fired")
+ case <-time.After(middleDuration):
+ }
+}
+
+func TestManyCancellableTimerResetUnderLock(t *testing.T) {
+ t.Parallel()
+
+ ch := make(chan struct{})
+ var lock sync.Mutex
+
+ lock.Lock()
+ timer := tcpip.MakeCancellableTimer(&lock, func() { ch <- struct{}{} })
+ timer.Reset(shortDuration)
+ for i := 0; i < 10; i++ {
+ timer.StopLocked()
+ timer.Reset(shortDuration)
+ }
+ lock.Unlock()
+
+ // Wait for double the duration for the last timer to fire.
+ select {
+ case <-ch:
+ case <-time.After(middleDuration):
+ t.Fatal("timed out waiting for timer to fire")
+ }
+
+ // The timer should have fired only once.
+ select {
+ case <-ch:
+ t.Fatal("no other timers should have fired")
+ case <-time.After(middleDuration):
+ }
+}
diff --git a/pkg/tcpip/transport/icmp/BUILD b/pkg/tcpip/transport/icmp/BUILD
index d8c5b5058..3aa23d529 100644
--- a/pkg/tcpip/transport/icmp/BUILD
+++ b/pkg/tcpip/transport/icmp/BUILD
@@ -28,6 +28,7 @@ go_library(
visibility = ["//visibility:public"],
deps = [
"//pkg/sleep",
+ "//pkg/sync",
"//pkg/tcpip",
"//pkg/tcpip/buffer",
"//pkg/tcpip/header",
diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go
index 9c40931b5..330786f4c 100644
--- a/pkg/tcpip/transport/icmp/endpoint.go
+++ b/pkg/tcpip/transport/icmp/endpoint.go
@@ -15,8 +15,7 @@
package icmp
import (
- "sync"
-
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/header"
@@ -350,13 +349,23 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
return nil
}
+// SetSockOptBool sets a socket option. Currently not supported.
+func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error {
+ return nil
+}
+
// SetSockOptInt sets a socket option. Currently not supported.
-func (e *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error {
+func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error {
return nil
}
+// GetSockOptBool implements tcpip.Endpoint.GetSockOptBool.
+func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) {
+ return false, tcpip.ErrUnknownProtocolOption
+}
+
// GetSockOptInt implements tcpip.Endpoint.GetSockOptInt.
-func (e *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) {
+func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) {
switch opt {
case tcpip.ReceiveQueueSizeOption:
v := 0
diff --git a/pkg/tcpip/transport/packet/BUILD b/pkg/tcpip/transport/packet/BUILD
index 44b58ff6b..4858d150c 100644
--- a/pkg/tcpip/transport/packet/BUILD
+++ b/pkg/tcpip/transport/packet/BUILD
@@ -28,6 +28,7 @@ go_library(
deps = [
"//pkg/log",
"//pkg/sleep",
+ "//pkg/sync",
"//pkg/tcpip",
"//pkg/tcpip/buffer",
"//pkg/tcpip/header",
diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go
index 0010b5e5f..fc5bc69fa 100644
--- a/pkg/tcpip/transport/packet/endpoint.go
+++ b/pkg/tcpip/transport/packet/endpoint.go
@@ -25,8 +25,7 @@
package packet
import (
- "sync"
-
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/header"
@@ -247,17 +246,17 @@ func (ep *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask {
// used with SetSockOpt, and this function always returns
// tcpip.ErrNotSupported.
func (ep *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
- return tcpip.ErrNotSupported
+ return tcpip.ErrUnknownProtocolOption
}
-// SetSockOptInt implements tcpip.Endpoint.SetSockOptInt.
-func (ep *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error {
+// SetSockOptBool implements tcpip.Endpoint.SetSockOptBool.
+func (ep *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error {
return tcpip.ErrUnknownProtocolOption
}
-// GetSockOptInt implements tcpip.Endpoint.GetSockOptInt.
-func (ep *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) {
- return 0, tcpip.ErrNotSupported
+// SetSockOptInt implements tcpip.Endpoint.SetSockOptInt.
+func (ep *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error {
+ return tcpip.ErrUnknownProtocolOption
}
// GetSockOpt implements tcpip.Endpoint.GetSockOpt.
@@ -265,6 +264,16 @@ func (ep *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
return tcpip.ErrNotSupported
}
+// GetSockOptBool implements tcpip.Endpoint.GetSockOptBool.
+func (ep *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) {
+ return false, tcpip.ErrNotSupported
+}
+
+// GetSockOptInt implements tcpip.Endpoint.GetSockOptInt.
+func (ep *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) {
+ return 0, tcpip.ErrNotSupported
+}
+
// HandlePacket implements stack.PacketEndpoint.HandlePacket.
func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) {
ep.rcvMu.Lock()
diff --git a/pkg/tcpip/transport/raw/BUILD b/pkg/tcpip/transport/raw/BUILD
index 00991ac8e..2f2131ff7 100644
--- a/pkg/tcpip/transport/raw/BUILD
+++ b/pkg/tcpip/transport/raw/BUILD
@@ -29,6 +29,7 @@ go_library(
deps = [
"//pkg/log",
"//pkg/sleep",
+ "//pkg/sync",
"//pkg/tcpip",
"//pkg/tcpip/buffer",
"//pkg/tcpip/header",
diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go
index 5aafe2615..ee9c4c58b 100644
--- a/pkg/tcpip/transport/raw/endpoint.go
+++ b/pkg/tcpip/transport/raw/endpoint.go
@@ -26,8 +26,7 @@
package raw
import (
- "sync"
-
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/header"
@@ -509,13 +508,38 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
return tcpip.ErrUnknownProtocolOption
}
+// SetSockOptBool implements tcpip.Endpoint.SetSockOptBool.
+func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error {
+ return tcpip.ErrUnknownProtocolOption
+}
+
// SetSockOptInt implements tcpip.Endpoint.SetSockOptInt.
-func (ep *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error {
+func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error {
return tcpip.ErrUnknownProtocolOption
}
+// GetSockOpt implements tcpip.Endpoint.GetSockOpt.
+func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
+ switch o := opt.(type) {
+ case tcpip.ErrorOption:
+ return nil
+
+ case *tcpip.KeepaliveEnabledOption:
+ *o = 0
+ return nil
+
+ default:
+ return tcpip.ErrUnknownProtocolOption
+ }
+}
+
+// GetSockOptBool implements tcpip.Endpoint.GetSockOptBool.
+func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) {
+ return false, tcpip.ErrUnknownProtocolOption
+}
+
// GetSockOptInt implements tcpip.Endpoint.GetSockOptInt.
-func (e *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) {
+func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) {
switch opt {
case tcpip.ReceiveQueueSizeOption:
v := 0
@@ -544,21 +568,6 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) {
return -1, tcpip.ErrUnknownProtocolOption
}
-// GetSockOpt implements tcpip.Endpoint.GetSockOpt.
-func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
- switch o := opt.(type) {
- case tcpip.ErrorOption:
- return nil
-
- case *tcpip.KeepaliveEnabledOption:
- *o = 0
- return nil
-
- default:
- return tcpip.ErrUnknownProtocolOption
- }
-}
-
// HandlePacket implements stack.RawTransportEndpoint.HandlePacket.
func (e *endpoint) HandlePacket(route *stack.Route, pkt tcpip.PacketBuffer) {
e.rcvMu.Lock()
diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD
index 3b353d56c..353bd06f4 100644
--- a/pkg/tcpip/transport/tcp/BUILD
+++ b/pkg/tcpip/transport/tcp/BUILD
@@ -48,6 +48,7 @@ go_library(
"//pkg/log",
"//pkg/rand",
"//pkg/sleep",
+ "//pkg/sync",
"//pkg/tcpip",
"//pkg/tcpip/buffer",
"//pkg/tcpip/hash/jenkins",
diff --git a/pkg/tcpip/transport/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go
index 5422ae80c..1ea996936 100644
--- a/pkg/tcpip/transport/tcp/accept.go
+++ b/pkg/tcpip/transport/tcp/accept.go
@@ -19,11 +19,11 @@ import (
"encoding/binary"
"hash"
"io"
- "sync"
"time"
"gvisor.dev/gvisor/pkg/rand"
"gvisor.dev/gvisor/pkg/sleep"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/header"
diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go
index cdd69f360..613ec1775 100644
--- a/pkg/tcpip/transport/tcp/connect.go
+++ b/pkg/tcpip/transport/tcp/connect.go
@@ -16,11 +16,11 @@ package tcp
import (
"encoding/binary"
- "sync"
"time"
"gvisor.dev/gvisor/pkg/rand"
"gvisor.dev/gvisor/pkg/sleep"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/hash/jenkins"
diff --git a/pkg/tcpip/transport/tcp/dual_stack_test.go b/pkg/tcpip/transport/tcp/dual_stack_test.go
index dfaa4a559..4f361b226 100644
--- a/pkg/tcpip/transport/tcp/dual_stack_test.go
+++ b/pkg/tcpip/transport/tcp/dual_stack_test.go
@@ -391,9 +391,8 @@ func testV4Accept(t *testing.T, c *context.Context) {
// Make sure we get the same error when calling the original ep and the
// new one. This validates that v4-mapped endpoints are still able to
// query the V6Only flag, whereas pure v4 endpoints are not.
- var v tcpip.V6OnlyOption
- expected := c.EP.GetSockOpt(&v)
- if err := nep.GetSockOpt(&v); err != expected {
+ _, expected := c.EP.GetSockOptBool(tcpip.V6OnlyOption)
+ if _, err := nep.GetSockOptBool(tcpip.V6OnlyOption); err != expected {
t.Fatalf("GetSockOpt returned unexpected value: got %v, want %v", err, expected)
}
@@ -531,8 +530,7 @@ func TestV6AcceptOnV6(t *testing.T) {
// Make sure we can still query the v6 only status of the new endpoint,
// that is, that it is in fact a v6 socket.
- var v tcpip.V6OnlyOption
- if err := nep.GetSockOpt(&v); err != nil {
+ if _, err := nep.GetSockOptBool(tcpip.V6OnlyOption); err != nil {
t.Fatalf("GetSockOpt failed failed: %v", err)
}
diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go
index fe629aa40..cca511fb9 100644
--- a/pkg/tcpip/transport/tcp/endpoint.go
+++ b/pkg/tcpip/transport/tcp/endpoint.go
@@ -19,12 +19,12 @@ import (
"fmt"
"math"
"strings"
- "sync"
"sync/atomic"
"time"
"gvisor.dev/gvisor/pkg/rand"
"gvisor.dev/gvisor/pkg/sleep"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/hash/jenkins"
@@ -885,8 +885,14 @@ func (e *endpoint) ModerateRecvBuf(copied int) {
// reject valid data that might already be in flight as the
// acceptable window will shrink.
if rcvWnd > e.rcvBufSize {
+ availBefore := e.receiveBufferAvailableLocked()
e.rcvBufSize = rcvWnd
- e.notifyProtocolGoroutine(notifyReceiveWindowChanged)
+ availAfter := e.receiveBufferAvailableLocked()
+ mask := uint32(notifyReceiveWindowChanged)
+ if crossed, above := e.windowCrossedACKThreshold(availAfter - availBefore); crossed && above {
+ mask |= notifyNonZeroReceiveWindow
+ }
+ e.notifyProtocolGoroutine(mask)
}
// We only update prevCopied when we grow the buffer because in cases
@@ -955,11 +961,12 @@ func (e *endpoint) readLocked() (buffer.View, *tcpip.Error) {
}
e.rcvBufUsed -= len(v)
- // If the window was zero before this read and if the read freed up
- // enough buffer space for the scaled window to be non-zero then notify
- // the protocol goroutine to send a window update.
- if e.zeroWindow && !e.zeroReceiveWindow(e.rcv.rcvWndScale) {
- e.zeroWindow = false
+
+ // If the window was small before this read and if the read freed up
+ // enough buffer space, to either fit an aMSS or half a receive buffer
+ // (whichever smaller), then notify the protocol goroutine to send a
+ // window update.
+ if crossed, above := e.windowCrossedACKThreshold(len(v)); crossed && above {
e.notifyProtocolGoroutine(notifyNonZeroReceiveWindow)
}
@@ -1133,20 +1140,65 @@ func (e *endpoint) Peek(vec [][]byte) (int64, tcpip.ControlMessages, *tcpip.Erro
return num, tcpip.ControlMessages{}, nil
}
-// zeroReceiveWindow checks if the receive window to be announced now would be
-// zero, based on the amount of available buffer and the receive window scaling.
+// windowCrossedACKThreshold checks if the receive window to be announced now
+// would be under aMSS or under half receive buffer, whichever smaller. This is
+// useful as a receive side silly window syndrome prevention mechanism. If
+// window grows to reasonable value, we should send ACK to the sender to inform
+// the rx space is now large. We also want ensure a series of small read()'s
+// won't trigger a flood of spurious tiny ACK's.
//
-// It must be called with rcvListMu held.
-func (e *endpoint) zeroReceiveWindow(scale uint8) bool {
- if e.rcvBufUsed >= e.rcvBufSize {
- return true
+// For large receive buffers, the threshold is aMSS - once reader reads more
+// than aMSS we'll send ACK. For tiny receive buffers, the threshold is half of
+// receive buffer size. This is chosen arbitrairly.
+// crossed will be true if the window size crossed the ACK threshold.
+// above will be true if the new window is >= ACK threshold and false
+// otherwise.
+func (e *endpoint) windowCrossedACKThreshold(deltaBefore int) (crossed bool, above bool) {
+ newAvail := e.receiveBufferAvailableLocked()
+ oldAvail := newAvail - deltaBefore
+ if oldAvail < 0 {
+ oldAvail = 0
+ }
+
+ threshold := int(e.amss)
+ if threshold > e.rcvBufSize/2 {
+ threshold = e.rcvBufSize / 2
+ }
+
+ switch {
+ case oldAvail < threshold && newAvail >= threshold:
+ return true, true
+ case oldAvail >= threshold && newAvail < threshold:
+ return true, false
+ }
+ return false, false
+}
+
+// SetSockOptBool sets a socket option.
+func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error {
+ switch opt {
+ case tcpip.V6OnlyOption:
+ // We only recognize this option on v6 endpoints.
+ if e.NetProto != header.IPv6ProtocolNumber {
+ return tcpip.ErrInvalidEndpointState
+ }
+
+ e.mu.Lock()
+ defer e.mu.Unlock()
+
+ // We only allow this to be set when we're in the initial state.
+ if e.state != StateInitial {
+ return tcpip.ErrInvalidEndpointState
+ }
+
+ e.v6only = v
}
- return ((e.rcvBufSize - e.rcvBufUsed) >> scale) == 0
+ return nil
}
// SetSockOptInt sets a socket option.
-func (e *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error {
+func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error {
switch opt {
case tcpip.ReceiveBufferSizeOption:
// Make sure the receive buffer size is within the min and max
@@ -1181,10 +1233,16 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error {
size = math.MaxInt32 / 2
}
+ availBefore := e.receiveBufferAvailableLocked()
e.rcvBufSize = size
+ availAfter := e.receiveBufferAvailableLocked()
+
e.rcvAutoParams.disabled = true
- if e.zeroWindow && !e.zeroReceiveWindow(scale) {
- e.zeroWindow = false
+
+ // Immediately send an ACK to uncork the sender silly window
+ // syndrome prevetion, when our available space grows above aMSS
+ // or half receive buffer, whichever smaller.
+ if crossed, above := e.windowCrossedACKThreshold(availAfter - availBefore); crossed && above {
mask |= notifyNonZeroReceiveWindow
}
e.rcvListMu.Unlock()
@@ -1256,19 +1314,14 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
return nil
case tcpip.BindToDeviceOption:
- e.mu.Lock()
- defer e.mu.Unlock()
- if v == "" {
- e.bindToDevice = 0
- return nil
+ id := tcpip.NICID(v)
+ if id != 0 && !e.stack.HasNIC(id) {
+ return tcpip.ErrUnknownDevice
}
- for nicID, nic := range e.stack.NICInfo() {
- if nic.Name == string(v) {
- e.bindToDevice = nicID
- return nil
- }
- }
- return tcpip.ErrUnknownDevice
+ e.mu.Lock()
+ e.bindToDevice = id
+ e.mu.Unlock()
+ return nil
case tcpip.QuickAckOption:
if v == 0 {
@@ -1289,23 +1342,6 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
e.notifyProtocolGoroutine(notifyMSSChanged)
return nil
- case tcpip.V6OnlyOption:
- // We only recognize this option on v6 endpoints.
- if e.NetProto != header.IPv6ProtocolNumber {
- return tcpip.ErrInvalidEndpointState
- }
-
- e.mu.Lock()
- defer e.mu.Unlock()
-
- // We only allow this to be set when we're in the initial state.
- if e.state != StateInitial {
- return tcpip.ErrInvalidEndpointState
- }
-
- e.v6only = v != 0
- return nil
-
case tcpip.TTLOption:
e.mu.Lock()
e.ttl = uint8(v)
@@ -1446,8 +1482,27 @@ func (e *endpoint) readyReceiveSize() (int, *tcpip.Error) {
return e.rcvBufUsed, nil
}
+// GetSockOptBool implements tcpip.Endpoint.GetSockOptBool.
+func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) {
+ switch opt {
+ case tcpip.V6OnlyOption:
+ // We only recognize this option on v6 endpoints.
+ if e.NetProto != header.IPv6ProtocolNumber {
+ return false, tcpip.ErrUnknownProtocolOption
+ }
+
+ e.mu.Lock()
+ v := e.v6only
+ e.mu.Unlock()
+
+ return v, nil
+ }
+
+ return false, tcpip.ErrUnknownProtocolOption
+}
+
// GetSockOptInt implements tcpip.Endpoint.GetSockOptInt.
-func (e *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) {
+func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) {
switch opt {
case tcpip.ReceiveQueueSizeOption:
return e.readyReceiveSize()
@@ -1525,12 +1580,8 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
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 = ""
+ *o = tcpip.BindToDeviceOption(e.bindToDevice)
+ e.mu.RUnlock()
return nil
case *tcpip.QuickAckOption:
@@ -1540,22 +1591,6 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
}
return nil
- case *tcpip.V6OnlyOption:
- // We only recognize this option on v6 endpoints.
- if e.NetProto != header.IPv6ProtocolNumber {
- return tcpip.ErrUnknownProtocolOption
- }
-
- e.mu.Lock()
- v := e.v6only
- e.mu.Unlock()
-
- *o = 0
- if v {
- *o = 1
- }
- return nil
-
case *tcpip.TTLOption:
e.mu.Lock()
*o = tcpip.TTLOption(e.ttl)
@@ -2225,13 +2260,10 @@ func (e *endpoint) readyToRead(s *segment) {
if s != nil {
s.incRef()
e.rcvBufUsed += s.data.Size()
- // Check if the receive window is now closed. If so make sure
- // we set the zero window before we deliver the segment to ensure
- // that a subsequent read of the segment will correctly trigger
- // a non-zero notification.
- if avail := e.receiveBufferAvailableLocked(); avail>>e.rcv.rcvWndScale == 0 {
+ // Increase counter if the receive window falls down below MSS
+ // or half receive buffer size, whichever smaller.
+ if crossed, above := e.windowCrossedACKThreshold(-s.data.Size()); crossed && !above {
e.stats.ReceiveErrors.ZeroRcvWindowState.Increment()
- e.zeroWindow = true
}
e.rcvList.PushBack(s)
} else {
diff --git a/pkg/tcpip/transport/tcp/endpoint_state.go b/pkg/tcpip/transport/tcp/endpoint_state.go
index 7aa4c3f0e..4b8d867bc 100644
--- a/pkg/tcpip/transport/tcp/endpoint_state.go
+++ b/pkg/tcpip/transport/tcp/endpoint_state.go
@@ -16,9 +16,9 @@ package tcp
import (
"fmt"
- "sync"
"time"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/header"
"gvisor.dev/gvisor/pkg/tcpip/stack"
diff --git a/pkg/tcpip/transport/tcp/forwarder.go b/pkg/tcpip/transport/tcp/forwarder.go
index 4983bca81..7eb613be5 100644
--- a/pkg/tcpip/transport/tcp/forwarder.go
+++ b/pkg/tcpip/transport/tcp/forwarder.go
@@ -15,8 +15,7 @@
package tcp
import (
- "sync"
-
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/header"
"gvisor.dev/gvisor/pkg/tcpip/seqnum"
diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go
index bc718064c..9a8f64aa6 100644
--- a/pkg/tcpip/transport/tcp/protocol.go
+++ b/pkg/tcpip/transport/tcp/protocol.go
@@ -22,9 +22,9 @@ package tcp
import (
"strings"
- "sync"
"time"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/header"
diff --git a/pkg/tcpip/transport/tcp/rcv.go b/pkg/tcpip/transport/tcp/rcv.go
index 0a5534959..05c8488f8 100644
--- a/pkg/tcpip/transport/tcp/rcv.go
+++ b/pkg/tcpip/transport/tcp/rcv.go
@@ -98,12 +98,6 @@ func (r *receiver) getSendParams() (rcvNxt seqnum.Value, rcvWnd seqnum.Size) {
// in such cases we may need to send an ack to indicate to our peer that it can
// resume sending data.
func (r *receiver) nonZeroWindow() {
- if (r.rcvAcc-r.rcvNxt)>>r.rcvWndScale != 0 {
- // We never got around to announcing a zero window size, so we
- // don't need to immediately announce a nonzero one.
- return
- }
-
// Immediately send an ack.
r.ep.snd.sendAck()
}
diff --git a/pkg/tcpip/transport/tcp/segment_queue.go b/pkg/tcpip/transport/tcp/segment_queue.go
index e0759225e..bd20a7ee9 100644
--- a/pkg/tcpip/transport/tcp/segment_queue.go
+++ b/pkg/tcpip/transport/tcp/segment_queue.go
@@ -15,7 +15,7 @@
package tcp
import (
- "sync"
+ "gvisor.dev/gvisor/pkg/sync"
)
// segmentQueue is a bounded, thread-safe queue of TCP segments.
diff --git a/pkg/tcpip/transport/tcp/snd.go b/pkg/tcpip/transport/tcp/snd.go
index 8a947dc66..fdff7ed81 100644
--- a/pkg/tcpip/transport/tcp/snd.go
+++ b/pkg/tcpip/transport/tcp/snd.go
@@ -16,11 +16,11 @@ package tcp
import (
"math"
- "sync"
"sync/atomic"
"time"
"gvisor.dev/gvisor/pkg/sleep"
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/header"
@@ -442,6 +442,13 @@ func (s *sender) retransmitTimerExpired() bool {
return true
}
+ // TODO(b/147297758): Band-aid fix, retransmitTimer can fire in some edge cases
+ // when writeList is empty. Remove this once we have a proper fix for this
+ // issue.
+ if s.writeList.Front() == nil {
+ return true
+ }
+
s.ep.stack.Stats().TCP.Timeouts.Increment()
s.ep.stats.SendErrors.Timeouts.Increment()
diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go
index e8fe4dab5..6edfa8dce 100644
--- a/pkg/tcpip/transport/tcp/tcp_test.go
+++ b/pkg/tcpip/transport/tcp/tcp_test.go
@@ -1083,12 +1083,12 @@ func TestTrafficClassV6(t *testing.T) {
func TestConnectBindToDevice(t *testing.T) {
for _, test := range []struct {
name string
- device string
+ device tcpip.NICID
want tcp.EndpointState
}{
- {"RightDevice", "nic1", tcp.StateEstablished},
- {"WrongDevice", "nic2", tcp.StateSynSent},
- {"AnyDevice", "", tcp.StateEstablished},
+ {"RightDevice", 1, tcp.StateEstablished},
+ {"WrongDevice", 2, tcp.StateSynSent},
+ {"AnyDevice", 0, tcp.StateEstablished},
} {
t.Run(test.name, func(t *testing.T) {
c := context.New(t, defaultMTU)
@@ -2091,10 +2091,14 @@ func TestZeroScaledWindowReceive(t *testing.T) {
)
}
- // Read some data. An ack should be sent in response to that.
- v, _, err := c.EP.Read(nil)
- if err != nil {
- t.Fatalf("Read failed: %v", err)
+ // Read at least 1MSS of data. An ack should be sent in response to that.
+ sz := 0
+ for sz < defaultMTU {
+ v, _, err := c.EP.Read(nil)
+ if err != nil {
+ t.Fatalf("Read failed: %v", err)
+ }
+ sz += len(v)
}
checker.IPv4(t, c.GetPacket(),
@@ -2103,7 +2107,7 @@ func TestZeroScaledWindowReceive(t *testing.T) {
checker.DstPort(context.TestPort),
checker.SeqNum(uint32(c.IRS)+1),
checker.AckNum(uint32(790+sent)),
- checker.Window(uint16(len(v)>>ws)),
+ checker.Window(uint16(sz>>ws)),
checker.TCPFlags(header.TCPFlagAck),
),
)
@@ -3794,46 +3798,41 @@ func TestBindToDeviceOption(t *testing.T) {
}
defer ep.Close()
- if err := s.CreateNamedNIC(321, "my_device", loopback.New()); err != nil {
- t.Errorf("CreateNamedNIC failed: %v", err)
- }
-
- // Make an nameless NIC.
- if err := s.CreateNIC(54321, loopback.New()); err != nil {
+ if err := s.CreateNIC(321, loopback.New()); err != nil {
t.Errorf("CreateNIC failed: %v", err)
}
- // strPtr is used instead of taking the address of string literals, which is
+ // nicIDPtr is used instead of taking the address of NICID literals, which is
// a compiler error.
- strPtr := func(s string) *string {
+ nicIDPtr := func(s tcpip.NICID) *tcpip.NICID {
return &s
}
testActions := []struct {
name string
- setBindToDevice *string
+ setBindToDevice *tcpip.NICID
setBindToDeviceError *tcpip.Error
getBindToDevice tcpip.BindToDeviceOption
}{
- {"GetDefaultValue", nil, nil, ""},
- {"BindToNonExistent", strPtr("non_existent_device"), tcpip.ErrUnknownDevice, ""},
- {"BindToExistent", strPtr("my_device"), nil, "my_device"},
- {"UnbindToDevice", strPtr(""), nil, ""},
+ {"GetDefaultValue", nil, nil, 0},
+ {"BindToNonExistent", nicIDPtr(999), tcpip.ErrUnknownDevice, 0},
+ {"BindToExistent", nicIDPtr(321), nil, 321},
+ {"UnbindToDevice", nicIDPtr(0), nil, 0},
}
for _, testAction := range testActions {
t.Run(testAction.name, func(t *testing.T) {
if testAction.setBindToDevice != nil {
bindToDevice := tcpip.BindToDeviceOption(*testAction.setBindToDevice)
- if got, want := ep.SetSockOpt(bindToDevice), testAction.setBindToDeviceError; got != want {
- t.Errorf("SetSockOpt(%v) got %v, want %v", bindToDevice, got, want)
+ if gotErr, wantErr := ep.SetSockOpt(bindToDevice), testAction.setBindToDeviceError; gotErr != wantErr {
+ t.Errorf("SetSockOpt(%v) got %v, want %v", bindToDevice, gotErr, wantErr)
}
}
- bindToDevice := tcpip.BindToDeviceOption("to be modified by GetSockOpt")
- if ep.GetSockOpt(&bindToDevice) != nil {
- t.Errorf("GetSockOpt got %v, want %v", ep.GetSockOpt(&bindToDevice), nil)
+ bindToDevice := tcpip.BindToDeviceOption(88888)
+ if err := ep.GetSockOpt(&bindToDevice); err != nil {
+ t.Errorf("GetSockOpt got %v, want %v", err, nil)
}
if got, want := bindToDevice, testAction.getBindToDevice; got != want {
- t.Errorf("bindToDevice got %q, want %q", got, want)
+ t.Errorf("bindToDevice got %d, want %d", got, want)
}
})
}
@@ -4027,12 +4026,12 @@ func TestConnectAvoidsBoundPorts(t *testing.T) {
switch network {
case "ipv4":
case "ipv6":
- if err := ep.SetSockOpt(tcpip.V6OnlyOption(1)); err != nil {
- t.Fatalf("SetSockOpt(V6OnlyOption(1)) failed: %v", err)
+ if err := ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil {
+ t.Fatalf("SetSockOpt(V6OnlyOption(true)) failed: %v", err)
}
case "dual":
- if err := ep.SetSockOpt(tcpip.V6OnlyOption(0)); err != nil {
- t.Fatalf("SetSockOpt(V6OnlyOption(0)) failed: %v", err)
+ if err := ep.SetSockOptBool(tcpip.V6OnlyOption, false); err != nil {
+ t.Fatalf("SetSockOpt(V6OnlyOption(false)) failed: %v", err)
}
default:
t.Fatalf("unknown network: '%s'", network)
@@ -6561,3 +6560,140 @@ func TestKeepaliveWithUserTimeout(t *testing.T) {
t.Errorf("got c.Stack().Stats().TCP.EstablishedTimedout = %v, want = %v", got, want)
}
}
+
+func TestIncreaseWindowOnReceive(t *testing.T) {
+ // This test ensures that the endpoint sends an ack,
+ // after recv() when the window grows to more than 1 MSS.
+ c := context.New(t, defaultMTU)
+ defer c.Cleanup()
+
+ const rcvBuf = 65535 * 10
+ c.CreateConnected(789, 30000, rcvBuf)
+
+ // Write chunks of ~30000 bytes. It's important that two
+ // payloads make it equal or longer than MSS.
+ remain := rcvBuf
+ sent := 0
+ data := make([]byte, defaultMTU/2)
+ lastWnd := uint16(0)
+
+ for remain > len(data) {
+ c.SendPacket(data, &context.Headers{
+ SrcPort: context.TestPort,
+ DstPort: c.Port,
+ Flags: header.TCPFlagAck,
+ SeqNum: seqnum.Value(790 + sent),
+ AckNum: c.IRS.Add(1),
+ RcvWnd: 30000,
+ })
+ sent += len(data)
+ remain -= len(data)
+
+ lastWnd = uint16(remain)
+ if remain > 0xffff {
+ lastWnd = 0xffff
+ }
+ checker.IPv4(t, c.GetPacket(),
+ checker.PayloadLen(header.TCPMinimumSize),
+ checker.TCP(
+ checker.DstPort(context.TestPort),
+ checker.SeqNum(uint32(c.IRS)+1),
+ checker.AckNum(uint32(790+sent)),
+ checker.Window(lastWnd),
+ checker.TCPFlags(header.TCPFlagAck),
+ ),
+ )
+ }
+
+ if lastWnd == 0xffff || lastWnd == 0 {
+ t.Fatalf("expected small, non-zero window: %d", lastWnd)
+ }
+
+ // We now have < 1 MSS in the buffer space. Read the data! An
+ // ack should be sent in response to that. The window was not
+ // zero, but it grew to larger than MSS.
+ if _, _, err := c.EP.Read(nil); err != nil {
+ t.Fatalf("Read failed: %v", err)
+ }
+
+ if _, _, err := c.EP.Read(nil); err != nil {
+ t.Fatalf("Read failed: %v", err)
+ }
+
+ // After reading two packets, we surely crossed MSS. See the ack:
+ checker.IPv4(t, c.GetPacket(),
+ checker.PayloadLen(header.TCPMinimumSize),
+ checker.TCP(
+ checker.DstPort(context.TestPort),
+ checker.SeqNum(uint32(c.IRS)+1),
+ checker.AckNum(uint32(790+sent)),
+ checker.Window(uint16(0xffff)),
+ checker.TCPFlags(header.TCPFlagAck),
+ ),
+ )
+}
+
+func TestIncreaseWindowOnBufferResize(t *testing.T) {
+ // This test ensures that the endpoint sends an ack,
+ // after available recv buffer grows to more than 1 MSS.
+ c := context.New(t, defaultMTU)
+ defer c.Cleanup()
+
+ const rcvBuf = 65535 * 10
+ c.CreateConnected(789, 30000, rcvBuf)
+
+ // Write chunks of ~30000 bytes. It's important that two
+ // payloads make it equal or longer than MSS.
+ remain := rcvBuf
+ sent := 0
+ data := make([]byte, defaultMTU/2)
+ lastWnd := uint16(0)
+
+ for remain > len(data) {
+ c.SendPacket(data, &context.Headers{
+ SrcPort: context.TestPort,
+ DstPort: c.Port,
+ Flags: header.TCPFlagAck,
+ SeqNum: seqnum.Value(790 + sent),
+ AckNum: c.IRS.Add(1),
+ RcvWnd: 30000,
+ })
+ sent += len(data)
+ remain -= len(data)
+
+ lastWnd = uint16(remain)
+ if remain > 0xffff {
+ lastWnd = 0xffff
+ }
+ checker.IPv4(t, c.GetPacket(),
+ checker.PayloadLen(header.TCPMinimumSize),
+ checker.TCP(
+ checker.DstPort(context.TestPort),
+ checker.SeqNum(uint32(c.IRS)+1),
+ checker.AckNum(uint32(790+sent)),
+ checker.Window(lastWnd),
+ checker.TCPFlags(header.TCPFlagAck),
+ ),
+ )
+ }
+
+ if lastWnd == 0xffff || lastWnd == 0 {
+ t.Fatalf("expected small, non-zero window: %d", lastWnd)
+ }
+
+ // Increasing the buffer from should generate an ACK,
+ // since window grew from small value to larger equal MSS
+ c.EP.SetSockOptInt(tcpip.ReceiveBufferSizeOption, rcvBuf*2)
+
+ // After reading two packets, we surely crossed MSS. See the ack:
+ checker.IPv4(t, c.GetPacket(),
+ checker.PayloadLen(header.TCPMinimumSize),
+ checker.TCP(
+ checker.DstPort(context.TestPort),
+ checker.SeqNum(uint32(c.IRS)+1),
+ checker.AckNum(uint32(790+sent)),
+ checker.Window(uint16(0xffff)),
+ checker.TCPFlags(header.TCPFlagAck),
+ ),
+ )
+}
diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go
index b0a376eba..822907998 100644
--- a/pkg/tcpip/transport/tcp/testing/context/context.go
+++ b/pkg/tcpip/transport/tcp/testing/context/context.go
@@ -158,15 +158,17 @@ func New(t *testing.T, mtu uint32) *Context {
if testing.Verbose() {
wep = sniffer.New(ep)
}
- if err := s.CreateNamedNIC(1, "nic1", wep); err != nil {
- t.Fatalf("CreateNIC failed: %v", err)
+ opts := stack.NICOptions{Name: "nic1"}
+ if err := s.CreateNICWithOptions(1, wep, opts); err != nil {
+ t.Fatalf("CreateNICWithOptions(_, _, %+v) failed: %v", opts, err)
}
wep2 := stack.LinkEndpoint(channel.New(1000, mtu, ""))
if testing.Verbose() {
wep2 = sniffer.New(channel.New(1000, mtu, ""))
}
- if err := s.CreateNamedNIC(2, "nic2", wep2); err != nil {
- t.Fatalf("CreateNIC failed: %v", err)
+ opts2 := stack.NICOptions{Name: "nic2"}
+ if err := s.CreateNICWithOptions(2, wep2, opts2); err != nil {
+ t.Fatalf("CreateNICWithOptions(_, _, %+v) failed: %v", opts2, err)
}
if err := s.AddAddress(1, ipv4.ProtocolNumber, StackAddr); err != nil {
@@ -473,11 +475,7 @@ func (c *Context) CreateV6Endpoint(v6only bool) {
c.t.Fatalf("NewEndpoint failed: %v", err)
}
- var v tcpip.V6OnlyOption
- if v6only {
- v = 1
- }
- if err := c.EP.SetSockOpt(v); err != nil {
+ if err := c.EP.SetSockOptBool(tcpip.V6OnlyOption, v6only); err != nil {
c.t.Fatalf("SetSockOpt failed failed: %v", err)
}
}
diff --git a/pkg/tcpip/transport/udp/BUILD b/pkg/tcpip/transport/udp/BUILD
index 97e4d5825..57ff123e3 100644
--- a/pkg/tcpip/transport/udp/BUILD
+++ b/pkg/tcpip/transport/udp/BUILD
@@ -30,6 +30,7 @@ go_library(
visibility = ["//visibility:public"],
deps = [
"//pkg/sleep",
+ "//pkg/sync",
"//pkg/tcpip",
"//pkg/tcpip/buffer",
"//pkg/tcpip/header",
diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go
index 1ac4705af..a4ff29a7d 100644
--- a/pkg/tcpip/transport/udp/endpoint.go
+++ b/pkg/tcpip/transport/udp/endpoint.go
@@ -15,8 +15,7 @@
package udp
import (
- "sync"
-
+ "gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/header"
@@ -456,14 +455,9 @@ func (e *endpoint) Peek([][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) {
return 0, tcpip.ControlMessages{}, nil
}
-// SetSockOptInt implements tcpip.Endpoint.SetSockOptInt.
-func (e *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error {
- return nil
-}
-
-// SetSockOpt implements tcpip.Endpoint.SetSockOpt.
-func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
- switch v := opt.(type) {
+// SetSockOptBool implements tcpip.Endpoint.SetSockOptBool.
+func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error {
+ switch opt {
case tcpip.V6OnlyOption:
// We only recognize this option on v6 endpoints.
if e.NetProto != header.IPv6ProtocolNumber {
@@ -478,8 +472,20 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
return tcpip.ErrInvalidEndpointState
}
- e.v6only = v != 0
+ e.v6only = v
+ }
+
+ return nil
+}
+
+// SetSockOptInt implements tcpip.Endpoint.SetSockOptInt.
+func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error {
+ return nil
+}
+// SetSockOpt implements tcpip.Endpoint.SetSockOpt.
+func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
+ switch v := opt.(type) {
case tcpip.TTLOption:
e.mu.Lock()
e.ttl = uint8(v)
@@ -624,19 +630,14 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
e.mu.Unlock()
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
- }
+ id := tcpip.NICID(v)
+ if id != 0 && !e.stack.HasNIC(id) {
+ return tcpip.ErrUnknownDevice
}
- return tcpip.ErrUnknownDevice
+ e.mu.Lock()
+ e.bindToDevice = id
+ e.mu.Unlock()
+ return nil
case tcpip.BroadcastOption:
e.mu.Lock()
@@ -660,8 +661,27 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
return nil
}
+// GetSockOptBool implements tcpip.Endpoint.GetSockOptBool.
+func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) {
+ switch opt {
+ case tcpip.V6OnlyOption:
+ // We only recognize this option on v6 endpoints.
+ if e.NetProto != header.IPv6ProtocolNumber {
+ return false, tcpip.ErrUnknownProtocolOption
+ }
+
+ e.mu.Lock()
+ v := e.v6only
+ e.mu.Unlock()
+
+ return v, nil
+ }
+
+ return false, tcpip.ErrUnknownProtocolOption
+}
+
// GetSockOptInt implements tcpip.Endpoint.GetSockOptInt.
-func (e *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) {
+func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) {
switch opt {
case tcpip.ReceiveQueueSizeOption:
v := 0
@@ -695,22 +715,6 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
case tcpip.ErrorOption:
return nil
- case *tcpip.V6OnlyOption:
- // We only recognize this option on v6 endpoints.
- if e.NetProto != header.IPv6ProtocolNumber {
- return tcpip.ErrUnknownProtocolOption
- }
-
- e.mu.Lock()
- v := e.v6only
- e.mu.Unlock()
-
- *o = 0
- if v {
- *o = 1
- }
- return nil
-
case *tcpip.TTLOption:
e.mu.Lock()
*o = tcpip.TTLOption(e.ttl)
@@ -757,12 +761,8 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
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 = tcpip.BindToDeviceOption("")
+ *o = tcpip.BindToDeviceOption(e.bindToDevice)
+ e.mu.RUnlock()
return nil
case *tcpip.KeepaliveEnabledOption:
diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go
index 7051a7a9c..0a82bc4fa 100644
--- a/pkg/tcpip/transport/udp/udp_test.go
+++ b/pkg/tcpip/transport/udp/udp_test.go
@@ -335,7 +335,7 @@ func (c *testContext) createEndpointForFlow(flow testFlow) {
c.createEndpoint(flow.sockProto())
if flow.isV6Only() {
- if err := c.ep.SetSockOpt(tcpip.V6OnlyOption(1)); err != nil {
+ if err := c.ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil {
c.t.Fatalf("SetSockOpt failed: %v", err)
}
} else if flow.isBroadcast() {
@@ -508,46 +508,42 @@ func TestBindToDeviceOption(t *testing.T) {
}
defer ep.Close()
- if err := s.CreateNamedNIC(321, "my_device", loopback.New()); err != nil {
- t.Errorf("CreateNamedNIC failed: %v", err)
+ opts := stack.NICOptions{Name: "my_device"}
+ if err := s.CreateNICWithOptions(321, loopback.New(), opts); err != nil {
+ t.Errorf("CreateNICWithOptions(_, _, %+v) failed: %v", opts, err)
}
- // Make an nameless NIC.
- if err := s.CreateNIC(54321, loopback.New()); err != nil {
- t.Errorf("CreateNIC failed: %v", err)
- }
-
- // strPtr is used instead of taking the address of string literals, which is
+ // nicIDPtr is used instead of taking the address of NICID literals, which is
// a compiler error.
- strPtr := func(s string) *string {
+ nicIDPtr := func(s tcpip.NICID) *tcpip.NICID {
return &s
}
testActions := []struct {
name string
- setBindToDevice *string
+ setBindToDevice *tcpip.NICID
setBindToDeviceError *tcpip.Error
getBindToDevice tcpip.BindToDeviceOption
}{
- {"GetDefaultValue", nil, nil, ""},
- {"BindToNonExistent", strPtr("non_existent_device"), tcpip.ErrUnknownDevice, ""},
- {"BindToExistent", strPtr("my_device"), nil, "my_device"},
- {"UnbindToDevice", strPtr(""), nil, ""},
+ {"GetDefaultValue", nil, nil, 0},
+ {"BindToNonExistent", nicIDPtr(999), tcpip.ErrUnknownDevice, 0},
+ {"BindToExistent", nicIDPtr(321), nil, 321},
+ {"UnbindToDevice", nicIDPtr(0), nil, 0},
}
for _, testAction := range testActions {
t.Run(testAction.name, func(t *testing.T) {
if testAction.setBindToDevice != nil {
bindToDevice := tcpip.BindToDeviceOption(*testAction.setBindToDevice)
- if got, want := ep.SetSockOpt(bindToDevice), testAction.setBindToDeviceError; got != want {
- t.Errorf("SetSockOpt(%v) got %v, want %v", bindToDevice, got, want)
+ if gotErr, wantErr := ep.SetSockOpt(bindToDevice), testAction.setBindToDeviceError; gotErr != wantErr {
+ t.Errorf("SetSockOpt(%v) got %v, want %v", bindToDevice, gotErr, wantErr)
}
}
- bindToDevice := tcpip.BindToDeviceOption("to be modified by GetSockOpt")
- if ep.GetSockOpt(&bindToDevice) != nil {
- t.Errorf("GetSockOpt got %v, want %v", ep.GetSockOpt(&bindToDevice), nil)
+ bindToDevice := tcpip.BindToDeviceOption(88888)
+ if err := ep.GetSockOpt(&bindToDevice); err != nil {
+ t.Errorf("GetSockOpt got %v, want %v", err, nil)
}
if got, want := bindToDevice, testAction.getBindToDevice; got != want {
- t.Errorf("bindToDevice got %q, want %q", got, want)
+ t.Errorf("bindToDevice got %d, want %d", got, want)
}
})
}