wgengine/magicsock: set UDP socket buffer sizes to 7MB

- At high data rates more buffer space is required in order to avoid
  packet loss during any cause of delay.
- On slower machines more buffer space is required in order to avoid
  packet loss while decryption & tun writing is underway.
- On higher latency network paths more buffer space is required in order
  to overcome BDP.
- On Linux set with SO_*BUFFORCE to bypass net.core.{r,w}mem_max.
- 7MB is the current default maximum on macOS 12.6
- Windows test is omitted, as Windows does not support getsockopt for
  these options.

Signed-off-by: James Tucker <james@tailscale.com>

wgengine/magicsock/magicsock.go

@@ -69,6 +69,11 @@
 	// _linux variant.
 	discoMagic1 = 0x5453f09f
 	discoMagic2 = 0x92ac
+
+	// UDP socket read/write buffer size (7MB). The value of 7MB is chosen as it
+	// is the max supported by a default configuration of macOS. Some platforms
+	// will silently clamp the value.
+	socketBufferSize = 7 << 20
 )
 
 // useDerpRoute reports whether magicsock should enable the DERP
@@ -2893,6 +2898,7 @@ func (c *Conn) bindSocket(ruc *RebindingUDPConn, network string, curPortFate cur
 			c.logf("magicsock: unable to bind %v port %d: %v", network, port, err)
 			continue
 		}
+		trySetSocketBuffer(pconn, c.logf)
 		// Success.
 		ruc.setConnLocked(pconn)
 		if network == "udp4" {
@@ -3948,6 +3954,20 @@ func (de *endpoint) handlePongConnLocked(m *disco.Pong, di *discoInfo, src netip
 	return
 }
 
+// portableTrySetSocketBuffer sets SO_SNDBUF and SO_RECVBUF on pconn to socketBufferSize,
+// logging an error if it occurs.
+func portableTrySetSocketBuffer(pconn nettype.PacketConn, logf logger.Logf) {
+	if c, ok := pconn.(*net.UDPConn); ok {
+		// Attempt to increase the buffer size, and allow failures.
+		if err := c.SetReadBuffer(socketBufferSize); err != nil {
+			logf("magicsock: failed to set UDP read buffer size to %d: %v", socketBufferSize, err)
+		}
+		if err := c.SetWriteBuffer(socketBufferSize); err != nil {
+			logf("magicsock: failed to set UDP write buffer size to %d: %v", socketBufferSize, err)
+		}
+	}
+}
+
 // addrLatency is an IPPort with an associated latency.
 type addrLatency struct {
 	netip.AddrPort

wgengine/magicsock/magicsock_default.go

@@ -10,8 +10,15 @@
 import (
 	"errors"
 	"io"
+
+	"tailscale.com/types/logger"
+	"tailscale.com/types/nettype"
 )
 
 func (c *Conn) listenRawDisco(family string) (io.Closer, error) {
 	return nil, errors.New("raw disco listening not supported on this OS")
 }
+
+func trySetSocketBuffer(pconn nettype.PacketConn, logf logger.Logf) {
+	portableTrySetSocketBuffer(pconn, logf)
+}

wgengine/magicsock/magicsock_linux.go

@@ -12,6 +12,7 @@
 	"io"
 	"net"
 	"net/netip"
+	"syscall"
 	"time"
 	"unsafe"
 
@@ -20,6 +21,8 @@
 	"tailscale.com/envknob"
 	"tailscale.com/net/netns"
 	"tailscale.com/types/key"
+	"tailscale.com/types/logger"
+	"tailscale.com/types/nettype"
 )
 
 const (
@@ -288,3 +291,30 @@ func setBPF(conn net.PacketConn, filter []bpf.RawInstruction) error {
 	}
 	return nil
 }
+
+// trySetSocketBuffer attempts to set SO_SNDBUFFORCE and SO_RECVBUFFORCE which
+// can overcome the limit of net.core.{r,w}mem_max, but require CAP_NET_ADMIN.
+// It falls back to the portable implementation if that fails, which may be
+// silently capped to net.core.{r,w}mem_max.
+func trySetSocketBuffer(pconn nettype.PacketConn, logf logger.Logf) {
+	if c, ok := pconn.(*net.UDPConn); ok {
+		var errRcv, errSnd error
+		rc, err := c.SyscallConn()
+		if err == nil {
+			rc.Control(func(fd uintptr) {
+				errRcv = syscall.SetsockoptInt(int(fd), syscall.SOL_SOCKET, syscall.SO_RCVBUFFORCE, socketBufferSize)
+				if errRcv != nil {
+					logf("magicsock: failed to force-set UDP read buffer size to %d: %v", socketBufferSize, errRcv)
+				}
+				errSnd = syscall.SetsockoptInt(int(fd), syscall.SOL_SOCKET, syscall.SO_SNDBUFFORCE, socketBufferSize)
+				if errSnd != nil {
+					logf("magicsock: failed to force-set UDP write buffer size to %d: %v", socketBufferSize, errSnd)
+				}
+			})
+		}
+
+		if err != nil || errRcv != nil || errSnd != nil {
+			portableTrySetSocketBuffer(pconn, logf)
+		}
+	}
+}

wgengine/magicsock/magicsock_unix_test.go

@@ -0,0 +1,62 @@
+// Copyright (c) 2022 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build unix
+// +build unix
+
+package magicsock
+
+import (
+	"net"
+	"syscall"
+	"testing"
+
+	"tailscale.com/types/nettype"
+)
+
+func TestTrySetSocketBuffer(t *testing.T) {
+	c, err := net.ListenPacket("udp", ":0")
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer c.Close()
+
+	rc, err := c.(*net.UDPConn).SyscallConn()
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	getBufs := func() (int, int) {
+		var rcv, snd int
+		rc.Control(func(fd uintptr) {
+			rcv, err = syscall.GetsockoptInt(int(fd), syscall.SOL_SOCKET, syscall.SO_RCVBUF)
+			if err != nil {
+				t.Errorf("getsockopt(SO_RCVBUF): %v", err)
+			}
+			snd, err = syscall.GetsockoptInt(int(fd), syscall.SOL_SOCKET, syscall.SO_SNDBUF)
+			if err != nil {
+				t.Errorf("getsockopt(SO_SNDBUF): %v", err)
+			}
+		})
+		return rcv, snd
+	}
+
+	curRcv, curSnd := getBufs()
+
+	trySetSocketBuffer(c.(nettype.PacketConn), t.Logf)
+
+	newRcv, newSnd := getBufs()
+
+	if curRcv > newRcv {
+		t.Errorf("SO_RCVBUF decreased: %v -> %v", curRcv, newRcv)
+	}
+	if curSnd > newSnd {
+		t.Errorf("SO_SNDBUF decreased: %v -> %v", curSnd, newSnd)
+	}
+
+	// On many systems we may not increase the value, particularly running as a
+	// regular user, so log the information for manual verification.
+	t.Logf("SO_RCVBUF: %v -> %v", curRcv, newRcv)
+	t.Logf("SO_SNDBUF: %v -> %v", curRcv, newRcv)
+}