core: Apply SO_REUSEPORT to UDP sockets (#5725)

* core: Apply SO_REUSEPORT to UDP sockets

For some reason, 10 months ago when I implemented SO_REUSEPORT
for TCP, I didn't realize, or forgot, that it can be used for UDP too. It is a
much better solution than using deadline hacks to reuse a socket, at
least for TCP.

Then https://github.com/mholt/caddy-l4/issues/132 was posted,
in which we see that UDP servers never actually stopped when the
L4 app was stopped. I verified this using this command:

    $ nc -u 127.0.0.1 55353

combined with POSTing configs to the /load admin endpoint (which
alternated between an echo server and a proxy server so I could tell
which config was being used).

I refactored the code to use SO_REUSEPORT for UDP, but of course
we still need graceful reloads on all platforms, not just Unix, so I
also implemented a deadline hack similar to what we used for
TCP before. That implementation for TCP was not perfect, possibly
having a logical (not data) race condition; but for UDP so far it
seems to be working. Verified the same way I verified that SO_REUSEPORT
works.

I think this code is slightly cleaner and I'm fairly confident this code
is effective.

* Check error

* Fix return

* Fix var name

* implement Unwrap interface and clean up

* move unix packet conn to platform specific file

* implement Unwrap for unix packet conn

* Move sharedPacketConn into proper file

* Fix Windows

* move sharedPacketConn and fakeClosePacketConn to proper file

---------

Co-authored-by: Weidi Deng <weidi_deng@icloud.com>
This commit is contained in:
Matt Holt 2023-10-16 22:17:32 -06:00 committed by Matthew Holt
parent 9782ea3400
commit 648207063e
3 changed files with 181 additions and 140 deletions

View File

@ -30,7 +30,22 @@ func reuseUnixSocket(network, addr string) (any, error) {
return nil, nil
}
func listenTCPOrUnix(ctx context.Context, lnKey string, network, address string, config net.ListenConfig) (net.Listener, error) {
func listenReusable(ctx context.Context, lnKey string, network, address string, config net.ListenConfig) (any, error) {
switch network {
case "udp", "udp4", "udp6", "unixgram":
sharedPc, _, err := listenerPool.LoadOrNew(lnKey, func() (Destructor, error) {
pc, err := config.ListenPacket(ctx, network, address)
if err != nil {
return nil, err
}
return &sharedPacketConn{PacketConn: pc, key: lnKey}, nil
})
if err != nil {
return nil, err
}
return &fakeClosePacketConn{sharedPacketConn: sharedPc.(*sharedPacketConn)}, nil
default:
sharedLn, _, err := listenerPool.LoadOrNew(lnKey, func() (Destructor, error) {
ln, err := config.Listen(ctx, network, address)
if err != nil {
@ -43,6 +58,7 @@ func listenTCPOrUnix(ctx context.Context, lnKey string, network, address string,
}
return &fakeCloseListener{sharedListener: sharedLn.(*sharedListener), keepAlivePeriod: config.KeepAlive}, nil
}
}
// fakeCloseListener is a private wrapper over a listener that
// is shared. The state of fakeCloseListener is not shared.
@ -98,7 +114,7 @@ func (fcl *fakeCloseListener) Accept() (net.Conn, error) {
// so that it's clear in the code that side-effects are shared with other
// users of this listener, not just our own reference to it; we also don't
// do anything with the error because all we could do is log it, but we
// expliclty assign it to nothing so we don't forget it's there if needed
// explicitly assign it to nothing so we don't forget it's there if needed
_ = fcl.sharedListener.clearDeadline()
if netErr, ok := err.(net.Error); ok && netErr.Timeout() {
@ -172,3 +188,75 @@ func (sl *sharedListener) setDeadline() error {
func (sl *sharedListener) Destruct() error {
return sl.Listener.Close()
}
// fakeClosePacketConn is like fakeCloseListener, but for PacketConns,
// or more specifically, *net.UDPConn
type fakeClosePacketConn struct {
closed int32 // accessed atomically; belongs to this struct only
*sharedPacketConn // embedded, so we also become a net.PacketConn; its key is used in Close
}
func (fcpc *fakeClosePacketConn) ReadFrom(p []byte) (n int, addr net.Addr, err error) {
// if the listener is already "closed", return error
if atomic.LoadInt32(&fcpc.closed) == 1 {
return 0, nil, &net.OpError{
Op: "readfrom",
Net: fcpc.LocalAddr().Network(),
Addr: fcpc.LocalAddr(),
Err: errFakeClosed,
}
}
// call underlying readfrom
n, addr, err = fcpc.sharedPacketConn.ReadFrom(p)
if err != nil {
// this server was stopped, so clear the deadline and let
// any new server continue reading; but we will exit
if atomic.LoadInt32(&fcpc.closed) == 1 {
if netErr, ok := err.(net.Error); ok && netErr.Timeout() {
if err = fcpc.SetReadDeadline(time.Time{}); err != nil {
return
}
}
}
return
}
return
}
// Close won't close the underlying socket unless there is no more reference, then listenerPool will close it.
func (fcpc *fakeClosePacketConn) Close() error {
if atomic.CompareAndSwapInt32(&fcpc.closed, 0, 1) {
_ = fcpc.SetReadDeadline(time.Now()) // unblock ReadFrom() calls to kick old servers out of their loops
_, _ = listenerPool.Delete(fcpc.sharedPacketConn.key)
}
return nil
}
func (fcpc *fakeClosePacketConn) Unwrap() net.PacketConn {
return fcpc.sharedPacketConn.PacketConn
}
// sharedPacketConn is like sharedListener, but for net.PacketConns.
type sharedPacketConn struct {
net.PacketConn
key string
}
// Destruct closes the underlying socket.
func (spc *sharedPacketConn) Destruct() error {
return spc.PacketConn.Close()
}
// Unwrap returns the underlying socket
func (spc *sharedPacketConn) Unwrap() net.PacketConn {
return spc.PacketConn
}
// Interface guards (see https://github.com/caddyserver/caddy/issues/3998)
var (
_ (interface {
Unwrap() net.PacketConn
}) = (*fakeClosePacketConn)(nil)
)

View File

@ -22,8 +22,10 @@ package caddy
import (
"context"
"errors"
"io"
"io/fs"
"net"
"os"
"sync/atomic"
"syscall"
@ -87,7 +89,7 @@ func reuseUnixSocket(network, addr string) (any, error) {
return nil, nil
}
func listenTCPOrUnix(ctx context.Context, lnKey string, network, address string, config net.ListenConfig) (net.Listener, error) {
func listenReusable(ctx context.Context, lnKey string, network, address string, config net.ListenConfig) (any, error) {
// wrap any Control function set by the user so we can also add our reusePort control without clobbering theirs
oldControl := config.Control
config.Control = func(network, address string, c syscall.RawConn) error {
@ -103,7 +105,14 @@ func listenTCPOrUnix(ctx context.Context, lnKey string, network, address string,
// we still put it in the listenerPool so we can count how many
// configs are using this socket; necessary to ensure we can know
// whether to enforce shutdown delays, for example (see #5393).
ln, err := config.Listen(ctx, network, address)
var ln io.Closer
var err error
switch network {
case "udp", "udp4", "udp6", "unixgram":
ln, err = config.ListenPacket(ctx, network, address)
default:
ln, err = config.Listen(ctx, network, address)
}
if err == nil {
listenerPool.LoadOrStore(lnKey, nil)
}
@ -117,9 +126,23 @@ func listenTCPOrUnix(ctx context.Context, lnKey string, network, address string,
unixSockets[lnKey] = ln.(*unixListener)
}
// TODO: Not 100% sure this is necessary, but we do this for net.UnixListener in listen_unix.go, so...
if unix, ok := ln.(*net.UnixConn); ok {
ln = &unixConn{unix, address, lnKey, &one}
unixSockets[lnKey] = ln.(*unixConn)
}
// lightly wrap the listener so that when it is closed,
// we can decrement the usage pool counter
return deleteListener{ln, lnKey}, err
switch specificLn := ln.(type) {
case net.Listener:
return deleteListener{specificLn, lnKey}, err
case net.PacketConn:
return deletePacketConn{specificLn, lnKey}, err
}
// other types, I guess we just return them directly
return ln, err
}
// reusePort sets SO_REUSEPORT. Ineffective for unix sockets.
@ -158,6 +181,36 @@ func (uln *unixListener) Close() error {
return uln.UnixListener.Close()
}
type unixConn struct {
*net.UnixConn
filename string
mapKey string
count *int32 // accessed atomically
}
func (uc *unixConn) Close() error {
newCount := atomic.AddInt32(uc.count, -1)
if newCount == 0 {
defer func() {
unixSocketsMu.Lock()
delete(unixSockets, uc.mapKey)
unixSocketsMu.Unlock()
_ = syscall.Unlink(uc.filename)
}()
}
return uc.UnixConn.Close()
}
func (uc *unixConn) Unwrap() net.PacketConn {
return uc.UnixConn
}
// unixSockets keeps track of the currently-active unix sockets
// so we can transfer their FDs gracefully during reloads.
var unixSockets = make(map[string]interface {
File() (*os.File, error)
})
// deleteListener is a type that simply deletes itself
// from the listenerPool when it closes. It is used
// solely for the purpose of reference counting (i.e.
@ -171,3 +224,19 @@ func (dl deleteListener) Close() error {
_, _ = listenerPool.Delete(dl.lnKey)
return dl.Listener.Close()
}
// deletePacketConn is like deleteListener, but
// for net.PacketConns.
type deletePacketConn struct {
net.PacketConn
lnKey string
}
func (dl deletePacketConn) Close() error {
_, _ = listenerPool.Delete(dl.lnKey)
return dl.PacketConn.Close()
}
func (dl deletePacketConn) Unwrap() net.PacketConn {
return dl.PacketConn
}

View File

@ -28,7 +28,6 @@ import (
"strings"
"sync"
"sync/atomic"
"syscall"
"time"
"github.com/quic-go/quic-go"
@ -149,11 +148,13 @@ func (na NetworkAddress) Listen(ctx context.Context, portOffset uint, config net
}
func (na NetworkAddress) listen(ctx context.Context, portOffset uint, config net.ListenConfig) (any, error) {
var ln any
var err error
var address string
var unixFileMode fs.FileMode
var isAbtractUnixSocket bool
var (
ln any
err error
address string
unixFileMode fs.FileMode
isAbtractUnixSocket bool
)
// split unix socket addr early so lnKey
// is independent of permissions bits
@ -181,27 +182,10 @@ func (na NetworkAddress) listen(ctx context.Context, portOffset uint, config net
lnKey := listenerKey(na.Network, address)
switch na.Network {
case "tcp", "tcp4", "tcp6", "unix", "unixpacket":
ln, err = listenTCPOrUnix(ctx, lnKey, na.Network, address, config)
case "unixgram":
ln, err = config.ListenPacket(ctx, na.Network, address)
case "udp", "udp4", "udp6":
sharedPc, _, err := listenerPool.LoadOrNew(lnKey, func() (Destructor, error) {
pc, err := config.ListenPacket(ctx, na.Network, address)
if err != nil {
return nil, err
}
return &sharedPacketConn{PacketConn: pc, key: lnKey}, nil
})
if err != nil {
return nil, err
}
spc := sharedPc.(*sharedPacketConn)
ln = &fakeClosePacketConn{spc: spc, UDPConn: spc.PacketConn.(*net.UDPConn)}
}
if strings.HasPrefix(na.Network, "ip") {
ln, err = config.ListenPacket(ctx, na.Network, address)
} else {
ln, err = listenReusable(ctx, lnKey, na.Network, address, config)
}
if err != nil {
return nil, err
@ -210,13 +194,6 @@ func (na NetworkAddress) listen(ctx context.Context, portOffset uint, config net
return nil, fmt.Errorf("unsupported network type: %s", na.Network)
}
// TODO: Not 100% sure this is necessary, but we do this for net.UnixListener in listen_unix.go, so...
if unix, ok := ln.(*net.UnixConn); ok {
one := int32(1)
ln = &unixConn{unix, address, lnKey, &one}
unixSockets[lnKey] = unix
}
if IsUnixNetwork(na.Network) {
if !isAbtractUnixSocket {
if err := os.Chmod(address, unixFileMode); err != nil {
@ -555,20 +532,8 @@ func ListenQUIC(ln net.PacketConn, tlsConf *tls.Config, activeRequests *int64) (
// and the request counter will reflect current http server
ctx, cancel := sql.sqs.addState(tlsConf, activeRequests)
// TODO: to serve QUIC over a unix socket, currently we need to hold onto
// the underlying net.PacketConn (which we wrap as unixConn to keep count
// of closes) because closing the quic.EarlyListener doesn't actually close
// the underlying PacketConn, but we need to for unix sockets since we dup
// the file descriptor and thus need to close the original; track issue:
// https://github.com/quic-go/quic-go/issues/3560#issuecomment-1258959608
var unix *unixConn
if uc, ok := ln.(*unixConn); ok {
unix = uc
}
return &fakeCloseQuicListener{
sharedQuicListener: sql,
uc: unix,
context: ctx,
contextCancel: cancel,
}, nil
@ -677,17 +642,6 @@ func (sql *sharedQuicListener) Destruct() error {
return sql.packetConn.Close()
}
// sharedPacketConn is like sharedListener, but for net.PacketConns.
type sharedPacketConn struct {
net.PacketConn
key string
}
// Destruct closes the underlying socket.
func (spc *sharedPacketConn) Destruct() error {
return spc.PacketConn.Close()
}
// fakeClosedErr returns an error value that is not temporary
// nor a timeout, suitable for making the caller think the
// listener is actually closed
@ -707,39 +661,9 @@ func fakeClosedErr(l interface{ Addr() net.Addr }) error {
// socket is actually left open.
var errFakeClosed = fmt.Errorf("listener 'closed' 😉")
// fakeClosePacketConn is like fakeCloseListener, but for PacketConns,
// or more specifically, *net.UDPConn
type fakeClosePacketConn struct {
closed int32 // accessed atomically; belongs to this struct only
spc *sharedPacketConn // its key is used in Close
*net.UDPConn // embedded, so we also become a net.PacketConn and enable several other optimizations done by quic-go
}
// interface guard for extra optimizations
// needed by QUIC implementation: https://github.com/caddyserver/caddy/issues/3998, https://github.com/caddyserver/caddy/issues/5605
var _ quic.OOBCapablePacketConn = (*fakeClosePacketConn)(nil)
// https://pkg.go.dev/golang.org/x/net/ipv4#NewPacketConn is used by quic-go and requires a net.PacketConn type assertable to a net.Conn,
// but doesn't actually use these methods, the only methods needed are `ReadMsgUDP` and `SyscallConn`.
var _ net.Conn = (*fakeClosePacketConn)(nil)
// Unwrap returns the underlying net.UDPConn for quic-go optimization
func (fcpc *fakeClosePacketConn) Unwrap() any {
return fcpc.UDPConn
}
// Close won't close the underlying socket unless there is no more reference, then listenerPool will close it.
func (fcpc *fakeClosePacketConn) Close() error {
if atomic.CompareAndSwapInt32(&fcpc.closed, 0, 1) {
_, _ = listenerPool.Delete(fcpc.spc.key)
}
return nil
}
type fakeCloseQuicListener struct {
closed int32 // accessed atomically; belongs to this struct only
*sharedQuicListener // embedded, so we also become a quic.EarlyListener
uc *unixConn // underlying unix socket, if UDS
context context.Context
contextCancel context.CancelFunc
}
@ -766,11 +690,6 @@ func (fcql *fakeCloseQuicListener) Close() error {
if atomic.CompareAndSwapInt32(&fcql.closed, 0, 1) {
fcql.contextCancel()
_, _ = listenerPool.Delete(fcql.sharedQuicListener.key)
if fcql.uc != nil {
// unix sockets need to be closed ourselves because we dup() the file
// descriptor when we reuse them, so this avoids a resource leak
fcql.uc.Close()
}
}
return nil
}
@ -796,34 +715,7 @@ func RegisterNetwork(network string, getListener ListenerFunc) {
networkTypes[network] = getListener
}
type unixConn struct {
*net.UnixConn
filename string
mapKey string
count *int32 // accessed atomically
}
func (uc *unixConn) Close() error {
newCount := atomic.AddInt32(uc.count, -1)
if newCount == 0 {
defer func() {
unixSocketsMu.Lock()
delete(unixSockets, uc.mapKey)
unixSocketsMu.Unlock()
_ = syscall.Unlink(uc.filename)
}()
}
return uc.UnixConn.Close()
}
// unixSockets keeps track of the currently-active unix sockets
// so we can transfer their FDs gracefully during reloads.
var (
unixSockets = make(map[string]interface {
File() (*os.File, error)
})
unixSocketsMu sync.Mutex
)
var unixSocketsMu sync.Mutex
// getListenerFromPlugin returns a listener on the given network and address
// if a plugin has registered the network name. It may return (nil, nil) if
@ -867,11 +759,3 @@ type ListenerWrapper interface {
var listenerPool = NewUsagePool()
const maxPortSpan = 65535
// Interface guards (see https://github.com/caddyserver/caddy/issues/3998)
var (
_ (interface{ SetReadBuffer(int) error }) = (*fakeClosePacketConn)(nil)
_ (interface {
SyscallConn() (syscall.RawConn, error)
}) = (*fakeClosePacketConn)(nil)
)