From e3e8aabbcf65d37516bb97f9dc0f77df52f8cf55 Mon Sep 17 00:00:00 2001 From: Matt Holt Date: Wed, 28 Sep 2022 13:35:51 -0600 Subject: [PATCH] core: Refactor and improve listener logic (#5089) * core: Refactor, improve listener logic Deprecate: - caddy.Listen - caddy.ListenTimeout - caddy.ListenPacket Prefer caddy.NetworkAddress.Listen() instead. Change: - caddy.ListenQUIC (hopefully to remove later) - caddy.ListenerFunc signature (add context and ListenConfig) - Don't emit Alt-Svc header advertising h3 over HTTP/3 - Use quic.ListenEarly instead of quic.ListenEarlyAddr; this gives us more flexibility (e.g. possibility of HTTP/3 over UDS) but also introduces a new issue: https://github.com/lucas-clemente/quic-go/issues/3560#issuecomment-1258959608 - Unlink unix socket before and after use * Appease the linter * Keep ListenAll --- admin.go | 7 +- listen.go | 26 +- listen_unix.go | 105 +++--- listeners.go | 636 +++++++++++++++++++++++----------- modules/caddyhttp/app.go | 48 ++- modules/caddyhttp/matchers.go | 4 +- modules/caddyhttp/server.go | 40 ++- 7 files changed, 568 insertions(+), 298 deletions(-) diff --git a/admin.go b/admin.go index 36cc2f82a..a23a1ac7d 100644 --- a/admin.go +++ b/admin.go @@ -382,7 +382,7 @@ func replaceLocalAdminServer(cfg *Config) error { handler := cfg.Admin.newAdminHandler(addr, false) - ln, err := Listen(addr.Network, addr.JoinHostPort(0)) + ln, err := addr.Listen(context.TODO(), 0, net.ListenConfig{}) if err != nil { return err } @@ -403,7 +403,7 @@ func replaceLocalAdminServer(cfg *Config) error { serverMu.Lock() server := localAdminServer serverMu.Unlock() - if err := server.Serve(ln); !errors.Is(err, http.ErrServerClosed) { + if err := server.Serve(ln.(net.Listener)); !errors.Is(err, http.ErrServerClosed) { adminLogger.Error("admin server shutdown for unknown reason", zap.Error(err)) } }() @@ -549,10 +549,11 @@ func replaceRemoteAdminServer(ctx Context, cfg *Config) error { serverMu.Unlock() // start listener - ln, err := Listen(addr.Network, addr.JoinHostPort(0)) + lnAny, err := addr.Listen(ctx, 0, net.ListenConfig{}) if err != nil { return err } + ln := lnAny.(net.Listener) ln = tls.NewListener(ln, tlsConfig) go func() { diff --git a/listen.go b/listen.go index 268785a3d..1ca168744 100644 --- a/listen.go +++ b/listen.go @@ -20,7 +20,7 @@ package caddy import ( - "fmt" + "context" "net" "sync" "sync/atomic" @@ -29,21 +29,14 @@ import ( "go.uber.org/zap" ) -func ListenTimeout(network, addr string, keepAlivePeriod time.Duration) (net.Listener, error) { - // check to see if plugin provides listener - if ln, err := getListenerFromPlugin(network, addr); err != nil || ln != nil { - return ln, err - } - - lnKey := listenerKey(network, addr) +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) { sharedLn, _, err := listenerPool.LoadOrNew(lnKey, func() (Destructor, error) { - ln, err := net.Listen(network, addr) + ln, err := config.Listen(ctx, network, address) if err != nil { - // https://github.com/caddyserver/caddy/pull/4534 - if isUnixNetwork(network) && isListenBindAddressAlreadyInUseError(err) { - return nil, fmt.Errorf("%w: this can happen if Caddy was forcefully killed", err) - } return nil, err } return &sharedListener{Listener: ln, key: lnKey}, nil @@ -51,8 +44,7 @@ func ListenTimeout(network, addr string, keepAlivePeriod time.Duration) (net.Lis if err != nil { return nil, err } - - return &fakeCloseListener{sharedListener: sharedLn.(*sharedListener), keepAlivePeriod: keepAlivePeriod}, nil + return &fakeCloseListener{sharedListener: sharedLn.(*sharedListener), keepAlivePeriod: config.KeepAlive}, nil } // fakeCloseListener is a private wrapper over a listener that @@ -156,8 +148,6 @@ func (sl *sharedListener) clearDeadline() error { switch ln := sl.Listener.(type) { case *net.TCPListener: err = ln.SetDeadline(time.Time{}) - case *net.UnixListener: - err = ln.SetDeadline(time.Time{}) } sl.deadline = false } @@ -173,8 +163,6 @@ func (sl *sharedListener) setDeadline() error { switch ln := sl.Listener.(type) { case *net.TCPListener: err = ln.SetDeadline(timeInPast) - case *net.UnixListener: - err = ln.SetDeadline(timeInPast) } sl.deadline = true } diff --git a/listen_unix.go b/listen_unix.go index f7b627940..dc955d891 100644 --- a/listen_unix.go +++ b/listen_unix.go @@ -24,78 +24,88 @@ import ( "errors" "io/fs" "net" - "sync" + "sync/atomic" "syscall" - "time" "go.uber.org/zap" "golang.org/x/sys/unix" ) -// ListenTimeout is the same as Listen, but with a configurable keep-alive timeout duration. -func ListenTimeout(network, addr string, keepalivePeriod time.Duration) (net.Listener, error) { - // check to see if plugin provides listener - if ln, err := getListenerFromPlugin(network, addr); err != nil || ln != nil { - return ln, err +// reuseUnixSocket copies and reuses the unix domain socket (UDS) if we already +// have it open; if not, unlink it so we can have it. No-op if not a unix network. +func reuseUnixSocket(network, addr string) (any, error) { + if !isUnixNetwork(network) { + return nil, nil } socketKey := listenerKey(network, addr) - if isUnixNetwork(network) { - unixSocketsMu.Lock() - defer unixSocketsMu.Unlock() - socket, exists := unixSockets[socketKey] - if exists { - // make copy of file descriptor - socketFile, err := socket.File() // dup() deep down - if err != nil { - return nil, err - } + socket, exists := unixSockets[socketKey] + if exists { + // make copy of file descriptor + socketFile, err := socket.File() // does dup() deep down + if err != nil { + return nil, err + } - // use copy to make new listener + // use copied fd to make new Listener or PacketConn, then replace + // it in the map so that future copies always come from the most + // recent fd (as the previous ones will be closed, and we'd get + // "use of closed network connection" errors) -- note that we + // preserve the *pointer* to the counter (not just the value) so + // that all socket wrappers will refer to the same value + switch unixSocket := socket.(type) { + case *unixListener: ln, err := net.FileListener(socketFile) if err != nil { return nil, err } + atomic.AddInt32(unixSocket.count, 1) + unixSockets[socketKey] = &unixListener{ln.(*net.UnixListener), socketKey, unixSocket.count} - // the old socket fd will likely be closed soon, so replace it in the map - unixSockets[socketKey] = ln.(*net.UnixListener) - - return ln.(*net.UnixListener), nil + case *unixConn: + pc, err := net.FilePacketConn(socketFile) + if err != nil { + return nil, err + } + atomic.AddInt32(unixSocket.count, 1) + unixSockets[socketKey] = &unixConn{pc.(*net.UnixConn), addr, socketKey, unixSocket.count} } - // from what I can tell after some quick research, it's quite common for programs to - // leave their socket file behind after they close, so the typical pattern is to - // unlink it before you bind to it -- this is often crucial if the last program using - // it was killed forcefully without a chance to clean up the socket, but there is a - // race, as the comment in net.UnixListener.close() explains... oh well? - if err := syscall.Unlink(addr); err != nil && !errors.Is(err, fs.ErrNotExist) { - return nil, err - } + return unixSockets[socketKey], nil } - config := &net.ListenConfig{Control: reusePort, KeepAlive: keepalivePeriod} - - ln, err := config.Listen(context.Background(), network, addr) - if err != nil { + // from what I can tell after some quick research, it's quite common for programs to + // leave their socket file behind after they close, so the typical pattern is to + // unlink it before you bind to it -- this is often crucial if the last program using + // it was killed forcefully without a chance to clean up the socket, but there is a + // race, as the comment in net.UnixListener.close() explains... oh well, I guess? + if err := syscall.Unlink(addr); err != nil && !errors.Is(err, fs.ErrNotExist) { return nil, err } - if uln, ok := ln.(*net.UnixListener); ok { - // TODO: ideally, we should unlink the socket once we know we're done using it - // (i.e. either on exit or a new config that doesn't use this socket; in UsagePool - // terms, when the reference count reaches 0), but given that we unlink existing - // socket before we create the new one anyway (see above), we don't necessarily - // need to clean up after ourselves; still, doing so would probably be more tidy - uln.SetUnlinkOnClose(false) - unixSockets[socketKey] = uln - } + return nil, nil +} - return ln, nil +func listenTCPOrUnix(ctx context.Context, lnKey string, network, address string, config net.ListenConfig) (net.Listener, 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 { + if oldControl != nil { + if err := oldControl(network, address, c); err != nil { + return err + } + } + return reusePort(network, address, c) + } + return config.Listen(ctx, network, address) } // reusePort sets SO_REUSEPORT. Ineffective for unix sockets. func reusePort(network, address string, conn syscall.RawConn) error { + if isUnixNetwork(network) { + return nil + } return conn.Control(func(descriptor uintptr) { if err := unix.SetsockoptInt(int(descriptor), unix.SOL_SOCKET, unix.SO_REUSEPORT, 1); err != nil { Log().Error("setting SO_REUSEPORT", @@ -106,10 +116,3 @@ func reusePort(network, address string, conn syscall.RawConn) error { } }) } - -// unixSockets keeps track of the currently-active unix sockets -// so we can transfer their FDs gracefully during reloads. -var ( - unixSockets = make(map[string]*net.UnixListener) - unixSocketsMu sync.Mutex -) diff --git a/listeners.go b/listeners.go index 6a23c61a1..aad6b691c 100644 --- a/listeners.go +++ b/listeners.go @@ -19,230 +19,187 @@ import ( "crypto/tls" "errors" "fmt" + "io" "net" "net/netip" "os" "strconv" "strings" + "sync" "sync/atomic" "syscall" + "time" "github.com/lucas-clemente/quic-go" "github.com/lucas-clemente/quic-go/http3" "go.uber.org/zap" ) -// Listen is like net.Listen, except Caddy's listeners can overlap -// each other: multiple listeners may be created on the same socket -// at the same time. This is useful because during config changes, -// the new config is started while the old config is still running. -// When Caddy listeners are closed, the closing logic is virtualized -// so the underlying socket isn't actually closed until all uses of -// the socket have been finished. Always be sure to close listeners -// when you are done with them, just like normal listeners. -func Listen(network, addr string) (net.Listener, error) { - // a 0 timeout means Go uses its default - return ListenTimeout(network, addr, 0) +// NetworkAddress represents one or more network addresses. +// It contains the individual components for a parsed network +// address of the form accepted by ParseNetworkAddress(). +type NetworkAddress struct { + // Should be a network value accepted by Go's net package or + // by a plugin providing a listener for that network type. + Network string + + // The "main" part of the network address is the host, which + // often takes the form of a hostname, DNS name, IP address, + // or socket path. + Host string + + // For addresses that contain a port, ranges are given by + // [StartPort, EndPort]; i.e. for a single port, StartPort + // and EndPort are the same. For no port, they are 0. + StartPort uint + EndPort uint } -// getListenerFromPlugin returns a listener on the given network and address -// if a plugin has registered the network name. It may return (nil, nil) if -// no plugin can provide a listener. -func getListenerFromPlugin(network, addr string) (net.Listener, error) { - network = strings.TrimSpace(strings.ToLower(network)) +// ListenAll calls Listen() for all addresses represented by this struct, i.e. all ports in the range. +// (If the address doesn't use ports or has 1 port only, then only 1 listener will be created.) +// It returns an error if any listener failed to bind, and closes any listeners opened up to that point. +// +// TODO: Experimental API: subject to change or removal. +func (na NetworkAddress) ListenAll(ctx context.Context, config net.ListenConfig) ([]any, error) { + var listeners []any + var err error - // get listener from plugin if network type is registered - if getListener, ok := networkTypes[network]; ok { - Log().Debug("getting listener from plugin", zap.String("network", network)) - return getListener(network, addr) - } - - return nil, nil -} - -// ListenPacket returns a net.PacketConn suitable for use in a Caddy module. -// It is like Listen except for PacketConns. -// Always be sure to close the PacketConn when you are done. -func ListenPacket(network, addr string) (net.PacketConn, error) { - lnKey := listenerKey(network, addr) - - sharedPc, _, err := listenerPool.LoadOrNew(lnKey, func() (Destructor, error) { - pc, err := net.ListenPacket(network, addr) - if err != nil { - // https://github.com/caddyserver/caddy/pull/4534 - if isUnixNetwork(network) && isListenBindAddressAlreadyInUseError(err) { - return nil, fmt.Errorf("%w: this can happen if Caddy was forcefully killed", err) + // if one of the addresses has a failure, we need to close + // any that did open a socket to avoid leaking resources + defer func() { + if err == nil { + return + } + for _, ln := range listeners { + if cl, ok := ln.(io.Closer); ok { + cl.Close() } + } + }() + + // an address can contain a port range, which represents multiple addresses; + // some addresses don't use ports at all and have a port range size of 1; + // whatever the case, iterate each address represented and bind a socket + for portOffset := uint(0); portOffset < na.PortRangeSize(); portOffset++ { + select { + case <-ctx.Done(): + return nil, ctx.Err() + default: + } + + // create (or reuse) the listener ourselves + var ln any + ln, err = na.Listen(ctx, portOffset, config) + if err != nil { return nil, err } - return &sharedPacketConn{PacketConn: pc, key: lnKey}, nil - }) - if err != nil { - return nil, err + listeners = append(listeners, ln) } - return &fakeClosePacketConn{sharedPacketConn: sharedPc.(*sharedPacketConn)}, nil + return listeners, nil } -// ListenQUIC returns a quic.EarlyListener suitable for use in a Caddy module. -// Note that the context passed to Accept is currently ignored, so using -// a context other than context.Background is meaningless. -// This API is EXPERIMENTAL and may change. -func ListenQUIC(addr string, tlsConf *tls.Config, activeRequests *int64) (quic.EarlyListener, error) { - lnKey := listenerKey("udp", addr) +// Listen is similar to net.Listen, with a few differences: +// +// Listen announces on the network address using the port calculated by adding +// portOffset to the start port. (For network types that do not use ports, the +// portOffset is ignored.) +// +// The provided ListenConfig is used to create the listener. Its Control function, +// if set, may be wrapped by an internally-used Control function. The provided +// context may be used to cancel long operations early. The context is not used +// to close the listener after it has been created. +// +// Caddy's listeners can overlap each other: multiple listeners may be created on +// the same socket at the same time. This is useful because during config changes, +// the new config is started while the old config is still running. How this is +// accomplished varies by platform and network type. For example, on Unix, SO_REUSEPORT +// is set except on Unix sockets, for which the file descriptor is duplicated and +// reused; on Windows, the close logic is virtualized using timeouts. Like normal +// listeners, be sure to Close() them when you are done. +// +// This method returns any type, as the implementations of listeners for various +// network types are not interchangeable. The type of listener returned is switched +// on the network type. Stream-based networks ("tcp", "unix", "unixpacket", etc.) +// return a net.Listener; datagram-based networks ("udp", "unixgram", etc.) return +// a net.PacketConn; and so forth. The actual concrete types are not guaranteed to +// be standard, exported types (wrapping is necessary to provide graceful reloads). +// +// Unix sockets will be unlinked before being created, to ensure we can bind to +// it even if the previous program using it exited uncleanly; it will also be +// unlinked upon a graceful exit (or when a new config does not use that socket). +// +// TODO: Experimental API: subject to change or removal. +func (na NetworkAddress) Listen(ctx context.Context, portOffset uint, config net.ListenConfig) (any, error) { + if na.IsUnixNetwork() { + unixSocketsMu.Lock() + defer unixSocketsMu.Unlock() + } - sharedEl, _, err := listenerPool.LoadOrNew(lnKey, func() (Destructor, error) { - el, err := quic.ListenAddrEarly(addr, http3.ConfigureTLSConfig(tlsConf), &quic.Config{ - RequireAddressValidation: func(clientAddr net.Addr) bool { - var highLoad bool - if activeRequests != nil { - highLoad = atomic.LoadInt64(activeRequests) > 1000 // TODO: make tunable? - } - return highLoad - }, + // check to see if plugin provides listener + if ln, err := getListenerFromPlugin(ctx, na.Network, na.JoinHostPort(portOffset), config); ln != nil || err != nil { + return ln, err + } + + // create (or reuse) the listener ourselves + return na.listen(ctx, portOffset, config) +} + +func (na NetworkAddress) listen(ctx context.Context, portOffset uint, config net.ListenConfig) (any, error) { + var ln any + var err error + + address := na.JoinHostPort(portOffset) + + // if this is a unix socket, see if we already have it open + if socket, err := reuseUnixSocket(na.Network, address); socket != nil || err != nil { + return socket, err + } + + 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 } - return &sharedQuicListener{EarlyListener: el, key: lnKey}, nil - }) + ln = &fakeClosePacketConn{sharedPacketConn: sharedPc.(*sharedPacketConn)} + } + if strings.HasPrefix(na.Network, "ip") { + ln, err = config.ListenPacket(ctx, na.Network, address) + } if err != nil { return nil, err } - - ctx, cancel := context.WithCancel(context.Background()) - return &fakeCloseQuicListener{ - sharedQuicListener: sharedEl.(*sharedQuicListener), - context: ctx, - contextCancel: cancel, - }, nil -} - -// ListenerUsage returns the current usage count of the given listener address. -func ListenerUsage(network, addr string) int { - count, _ := listenerPool.References(listenerKey(network, addr)) - return count -} - -func listenerKey(network, addr string) string { - return network + "/" + addr -} - -type fakeCloseQuicListener struct { - closed int32 // accessed atomically; belongs to this struct only - *sharedQuicListener // embedded, so we also become a quic.EarlyListener - context context.Context - contextCancel context.CancelFunc -} - -// Currently Accept ignores the passed context, however a situation where -// someone would need a hotswappable QUIC-only (not http3, since it uses context.Background here) -// server on which Accept would be called with non-empty contexts -// (mind that the default net listeners' Accept doesn't take a context argument) -// sounds way too rare for us to sacrifice efficiency here. -func (fcql *fakeCloseQuicListener) Accept(_ context.Context) (quic.EarlyConnection, error) { - conn, err := fcql.sharedQuicListener.Accept(fcql.context) - if err == nil { - return conn, nil + if ln == nil { + return nil, fmt.Errorf("unsupported network type: %s", na.Network) } - // if the listener is "closed", return a fake closed error instead - if atomic.LoadInt32(&fcql.closed) == 1 && errors.Is(err, context.Canceled) { - return nil, fakeClosedErr(fcql) + // if new listener is a unix socket, make sure we can reuse it later + // (we do our own "unlink on close" -- not required, but more tidy) + one := int32(1) + switch unix := ln.(type) { + case *net.UnixListener: + unix.SetUnlinkOnClose(false) + ln = &unixListener{unix, lnKey, &one} + unixSockets[lnKey] = ln.(*unixListener) + case *net.UnixConn: + ln = &unixConn{unix, address, lnKey, &one} + unixSockets[lnKey] = ln.(*unixConn) } - return nil, err -} -func (fcql *fakeCloseQuicListener) Close() error { - if atomic.CompareAndSwapInt32(&fcql.closed, 0, 1) { - fcql.contextCancel() - _, _ = listenerPool.Delete(fcql.sharedQuicListener.key) - } - return nil -} - -// fakeClosedErr returns an error value that is not temporary -// nor a timeout, suitable for making the caller think the -// listener is actually closed -func fakeClosedErr(l interface{ Addr() net.Addr }) error { - return &net.OpError{ - Op: "accept", - Net: l.Addr().Network(), - Addr: l.Addr(), - Err: errFakeClosed, - } -} - -// ErrFakeClosed is the underlying error value returned by -// fakeCloseListener.Accept() after Close() has been called, -// indicating that it is pretending to be closed so that the -// server using it can terminate, while the underlying -// socket is actually left open. -var errFakeClosed = fmt.Errorf("listener 'closed' 😉") - -// fakeClosePacketConn is like fakeCloseListener, but for PacketConns. -type fakeClosePacketConn struct { - closed int32 // accessed atomically; belongs to this struct only - *sharedPacketConn // embedded, so we also become a net.PacketConn -} - -func (fcpc *fakeClosePacketConn) Close() error { - if atomic.CompareAndSwapInt32(&fcpc.closed, 0, 1) { - _, _ = listenerPool.Delete(fcpc.sharedPacketConn.key) - } - return nil -} - -// Supports QUIC implementation: https://github.com/caddyserver/caddy/issues/3998 -func (fcpc fakeClosePacketConn) SetReadBuffer(bytes int) error { - if conn, ok := fcpc.PacketConn.(interface{ SetReadBuffer(int) error }); ok { - return conn.SetReadBuffer(bytes) - } - return fmt.Errorf("SetReadBuffer() not implemented for %T", fcpc.PacketConn) -} - -// Supports QUIC implementation: https://github.com/caddyserver/caddy/issues/3998 -func (fcpc fakeClosePacketConn) SyscallConn() (syscall.RawConn, error) { - if conn, ok := fcpc.PacketConn.(interface { - SyscallConn() (syscall.RawConn, error) - }); ok { - return conn.SyscallConn() - } - return nil, fmt.Errorf("SyscallConn() not implemented for %T", fcpc.PacketConn) -} - -// sharedQuicListener is like sharedListener, but for quic.EarlyListeners. -type sharedQuicListener struct { - quic.EarlyListener - key string -} - -// Destruct closes the underlying QUIC listener. -func (sql *sharedQuicListener) Destruct() error { - return sql.EarlyListener.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() -} - -// NetworkAddress contains the individual components -// for a parsed network address of the form accepted -// by ParseNetworkAddress(). Network should be a -// network value accepted by Go's net package. Port -// ranges are given by [StartPort, EndPort]. -type NetworkAddress struct { - Network string - Host string - StartPort uint - EndPort uint + return ln, nil } // IsUnixNetwork returns true if na.Network is @@ -260,17 +217,27 @@ func (na NetworkAddress) JoinHostPort(offset uint) string { return net.JoinHostPort(na.Host, strconv.Itoa(int(na.StartPort+offset))) } +// Expand returns one NetworkAddress for each port in the port range. +// +// This is EXPERIMENTAL and subject to change or removal. func (na NetworkAddress) Expand() []NetworkAddress { size := na.PortRangeSize() addrs := make([]NetworkAddress, size) for portOffset := uint(0); portOffset < size; portOffset++ { - na2 := na - na2.StartPort, na2.EndPort = na.StartPort+portOffset, na.StartPort+portOffset - addrs[portOffset] = na2 + addrs[portOffset] = na.At(portOffset) } return addrs } +// At returns a NetworkAddress with a port range of just 1 +// at the given port offset; i.e. a NetworkAddress that +// represents precisely 1 address only. +func (na NetworkAddress) At(portOffset uint) NetworkAddress { + na2 := na + na2.StartPort, na2.EndPort = na.StartPort+portOffset, na.StartPort+portOffset + return na2 +} + // PortRangeSize returns how many ports are in // pa's port range. Port ranges are inclusive, // so the size is the difference of start and @@ -326,20 +293,6 @@ func isUnixNetwork(netw string) bool { return netw == "unix" || netw == "unixgram" || netw == "unixpacket" } -func isListenBindAddressAlreadyInUseError(err error) bool { - switch networkOperationError := err.(type) { - case *net.OpError: - switch syscallError := networkOperationError.Err.(type) { - case *os.SyscallError: - if syscallError.Syscall == "bind" { - return true - } - } - } - - return false -} - // ParseNetworkAddress parses addr into its individual // components. The input string is expected to be of // the form "network/host:port-range" where any part is @@ -439,6 +392,209 @@ func JoinNetworkAddress(network, host, port string) string { return a } +// DEPRECATED: Use NetworkAddress.Listen instead. This function will likely be changed or removed in the future. +func Listen(network, addr string) (net.Listener, error) { + // a 0 timeout means Go uses its default + return ListenTimeout(network, addr, 0) +} + +// DEPRECATED: Use NetworkAddress.Listen instead. This function will likely be changed or removed in the future. +func ListenTimeout(network, addr string, keepalivePeriod time.Duration) (net.Listener, error) { + netAddr, err := ParseNetworkAddress(JoinNetworkAddress(network, addr, "")) + if err != nil { + return nil, err + } + + ln, err := netAddr.Listen(context.TODO(), 0, net.ListenConfig{KeepAlive: keepalivePeriod}) + if err != nil { + return nil, err + } + + return ln.(net.Listener), nil +} + +// DEPRECATED: Use NetworkAddress.Listen instead. This function will likely be changed or removed in the future. +func ListenPacket(network, addr string) (net.PacketConn, error) { + netAddr, err := ParseNetworkAddress(JoinNetworkAddress(network, addr, "")) + if err != nil { + return nil, err + } + + ln, err := netAddr.Listen(context.TODO(), 0, net.ListenConfig{}) + if err != nil { + return nil, err + } + + return ln.(net.PacketConn), nil +} + +// ListenQUIC returns a quic.EarlyListener suitable for use in a Caddy module. +// The network will be transformed into a QUIC-compatible type (if unix, then +// unixgram will be used; otherwise, udp will be used). +// +// NOTE: This API is EXPERIMENTAL and may be changed or removed. +// +// TODO: See if we can find a more elegant solution closer to the new NetworkAddress.Listen API. +func ListenQUIC(ln net.PacketConn, tlsConf *tls.Config, activeRequests *int64) (quic.EarlyListener, error) { + lnKey := listenerKey(ln.LocalAddr().Network(), ln.LocalAddr().String()) + + sharedEarlyListener, _, err := listenerPool.LoadOrNew(lnKey, func() (Destructor, error) { + earlyLn, err := quic.ListenEarly(ln, http3.ConfigureTLSConfig(tlsConf), &quic.Config{ + RequireAddressValidation: func(clientAddr net.Addr) bool { + var highLoad bool + if activeRequests != nil { + highLoad = atomic.LoadInt64(activeRequests) > 1000 // TODO: make tunable? + } + return highLoad + }, + }) + if err != nil { + return nil, err + } + + return &sharedQuicListener{EarlyListener: earlyLn, key: lnKey}, nil + }) + if err != nil { + return nil, err + } + + // 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/lucas-clemente/quic-go/issues/3560#issuecomment-1258959608 + var unix *unixConn + if uc, ok := ln.(*unixConn); ok { + unix = uc + } + + ctx, cancel := context.WithCancel(context.Background()) + return &fakeCloseQuicListener{ + sharedQuicListener: sharedEarlyListener.(*sharedQuicListener), + uc: unix, + context: ctx, + contextCancel: cancel, + }, nil +} + +// ListenerUsage returns the current usage count of the given listener address. +func ListenerUsage(network, addr string) int { + count, _ := listenerPool.References(listenerKey(network, addr)) + return count +} + +// sharedQuicListener is like sharedListener, but for quic.EarlyListeners. +type sharedQuicListener struct { + quic.EarlyListener + key string +} + +// Destruct closes the underlying QUIC listener. +func (sql *sharedQuicListener) Destruct() error { + return sql.EarlyListener.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 +func fakeClosedErr(l interface{ Addr() net.Addr }) error { + return &net.OpError{ + Op: "accept", + Net: l.Addr().Network(), + Addr: l.Addr(), + Err: errFakeClosed, + } +} + +// errFakeClosed is the underlying error value returned by +// fakeCloseListener.Accept() after Close() has been called, +// indicating that it is pretending to be closed so that the +// server using it can terminate, while the underlying +// socket is actually left open. +var errFakeClosed = fmt.Errorf("listener 'closed' 😉") + +// fakeClosePacketConn is like fakeCloseListener, but for PacketConns. +type fakeClosePacketConn struct { + closed int32 // accessed atomically; belongs to this struct only + *sharedPacketConn // embedded, so we also become a net.PacketConn +} + +func (fcpc *fakeClosePacketConn) Close() error { + if atomic.CompareAndSwapInt32(&fcpc.closed, 0, 1) { + _, _ = listenerPool.Delete(fcpc.sharedPacketConn.key) + } + return nil +} + +// Supports QUIC implementation: https://github.com/caddyserver/caddy/issues/3998 +func (fcpc fakeClosePacketConn) SetReadBuffer(bytes int) error { + if conn, ok := fcpc.PacketConn.(interface{ SetReadBuffer(int) error }); ok { + return conn.SetReadBuffer(bytes) + } + return fmt.Errorf("SetReadBuffer() not implemented for %T", fcpc.PacketConn) +} + +// Supports QUIC implementation: https://github.com/caddyserver/caddy/issues/3998 +func (fcpc fakeClosePacketConn) SyscallConn() (syscall.RawConn, error) { + if conn, ok := fcpc.PacketConn.(interface { + SyscallConn() (syscall.RawConn, error) + }); ok { + return conn.SyscallConn() + } + return nil, fmt.Errorf("SyscallConn() not implemented for %T", fcpc.PacketConn) +} + +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 +} + +// Currently Accept ignores the passed context, however a situation where +// someone would need a hotswappable QUIC-only (not http3, since it uses context.Background here) +// server on which Accept would be called with non-empty contexts +// (mind that the default net listeners' Accept doesn't take a context argument) +// sounds way too rare for us to sacrifice efficiency here. +func (fcql *fakeCloseQuicListener) Accept(_ context.Context) (quic.EarlyConnection, error) { + conn, err := fcql.sharedQuicListener.Accept(fcql.context) + if err == nil { + return conn, nil + } + + // if the listener is "closed", return a fake closed error instead + if atomic.LoadInt32(&fcql.closed) == 1 && errors.Is(err, context.Canceled) { + return nil, fakeClosedErr(fcql) + } + return nil, err +} + +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 +} + // RegisterNetwork registers a network type with Caddy so that if a listener is // created for that network type, getListener will be invoked to get the listener. // This should be called during init() and will panic if the network type is standard @@ -460,11 +616,77 @@ func RegisterNetwork(network string, getListener ListenerFunc) { networkTypes[network] = getListener } +type unixListener struct { + *net.UnixListener + mapKey string + count *int32 // accessed atomically +} + +func (uln *unixListener) Close() error { + newCount := atomic.AddInt32(uln.count, -1) + if newCount == 0 { + defer func() { + addr := uln.Addr().String() + unixSocketsMu.Lock() + delete(unixSockets, uln.mapKey) + unixSocketsMu.Unlock() + _ = syscall.Unlink(addr) + }() + } + 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() +} + +// 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 +) + +// getListenerFromPlugin returns a listener on the given network and address +// if a plugin has registered the network name. It may return (nil, nil) if +// no plugin can provide a listener. +func getListenerFromPlugin(ctx context.Context, network, addr string, config net.ListenConfig) (any, error) { + // get listener from plugin if network type is registered + if getListener, ok := networkTypes[network]; ok { + Log().Debug("getting listener from plugin", zap.String("network", network)) + return getListener(ctx, network, addr, config) + } + + return nil, nil +} + +func listenerKey(network, addr string) string { + return network + "/" + addr +} + // ListenerFunc is a function that can return a listener given a network and address. // The listeners must be capable of overlapping: with Caddy, new configs are loaded // before old ones are unloaded, so listeners may overlap briefly if the configs // both need the same listener. EXPERIMENTAL and subject to change. -type ListenerFunc func(network, addr string) (net.Listener, error) +type ListenerFunc func(ctx context.Context, network, addr string, cfg net.ListenConfig) (any, error) var networkTypes = map[string]ListenerFunc{} diff --git a/modules/caddyhttp/app.go b/modules/caddyhttp/app.go index c9a554314..33d96d867 100644 --- a/modules/caddyhttp/app.go +++ b/modules/caddyhttp/app.go @@ -18,6 +18,7 @@ import ( "context" "crypto/tls" "fmt" + "net" "net/http" "strconv" "sync" @@ -387,10 +388,11 @@ func (app *App) Start() error { for portOffset := uint(0); portOffset < listenAddr.PortRangeSize(); portOffset++ { // create the listener for this socket hostport := listenAddr.JoinHostPort(portOffset) - ln, err := caddy.ListenTimeout(listenAddr.Network, hostport, time.Duration(srv.KeepAliveInterval)) + lnAny, err := listenAddr.Listen(app.ctx, portOffset, net.ListenConfig{KeepAlive: time.Duration(srv.KeepAliveInterval)}) if err != nil { - return fmt.Errorf("%s: listening on %s: %v", listenAddr.Network, hostport, err) + return fmt.Errorf("listening on %s: %v", listenAddr.At(portOffset), err) } + ln := lnAny.(net.Listener) // wrap listener before TLS (up to the TLS placeholder wrapper) var lnWrapperIdx int @@ -409,10 +411,27 @@ func (app *App) Start() error { ln = tls.NewListener(ln, tlsCfg) // enable HTTP/3 if configured - if srv.protocol("h3") && !listenAddr.IsUnixNetwork() { - app.logger.Info("enabling HTTP/3 listener", zap.String("addr", hostport)) - if err := srv.serveHTTP3(hostport, tlsCfg); err != nil { - return err + if srv.protocol("h3") { + // Can't serve HTTP/3 on the same socket as HTTP/1 and 2 because it uses + // a different transport mechanism... which is fine, but the OS doesn't + // differentiate between a SOCK_STREAM file and a SOCK_DGRAM file; they + // are still one file on the system. So even though "unixpacket" and + // "unixgram" are different network types just as "tcp" and "udp" are, + // the OS will not let us use the same file as both STREAM and DGRAM. + if len(srv.Protocols) > 1 && listenAddr.IsUnixNetwork() { + app.logger.Warn("HTTP/3 disabled because Unix can't multiplex STREAM and DGRAM on same socket", + zap.String("file", hostport)) + for i := range srv.Protocols { + if srv.Protocols[i] == "h3" { + srv.Protocols = append(srv.Protocols[:i], srv.Protocols[i+1:]...) + break + } + } + } else { + app.logger.Info("enabling HTTP/3 listener", zap.String("addr", hostport)) + if err := srv.serveHTTP3(listenAddr.At(portOffset), tlsCfg); err != nil { + return err + } } } } @@ -424,11 +443,10 @@ func (app *App) Start() error { // if binding to port 0, the OS chooses a port for us; // but the user won't know the port unless we print it - if listenAddr.StartPort == 0 && listenAddr.EndPort == 0 { + if !listenAddr.IsUnixNetwork() && listenAddr.StartPort == 0 && listenAddr.EndPort == 0 { app.logger.Info("port 0 listener", zap.String("input_address", lnAddr), - zap.String("actual_address", ln.Addr().String()), - ) + zap.String("actual_address", ln.Addr().String())) } app.logger.Debug("starting server loop", @@ -533,6 +551,18 @@ func (app *App) Stop() error { if server.h3server == nil { return } + + // TODO: we have to manually close our listeners because quic-go won't + // close listeners it didn't create along with the server itself... + // see https://github.com/lucas-clemente/quic-go/issues/3560 + for _, el := range server.h3listeners { + if err := el.Close(); err != nil { + app.logger.Error("HTTP/3 listener close", + zap.Error(err), + zap.String("address", el.LocalAddr().String())) + } + } + // TODO: CloseGracefully, once implemented upstream (see https://github.com/lucas-clemente/quic-go/issues/2103) if err := server.h3server.Close(); err != nil { app.logger.Error("HTTP/3 server shutdown", diff --git a/modules/caddyhttp/matchers.go b/modules/caddyhttp/matchers.go index bca94be8e..e39ba3fc0 100644 --- a/modules/caddyhttp/matchers.go +++ b/modules/caddyhttp/matchers.go @@ -156,7 +156,9 @@ type ( MatchHeaderRE map[string]*MatchRegexp // MatchProtocol matches requests by protocol. Recognized values are - // "http", "https", and "grpc". + // "http", "https", and "grpc" for broad protocol matches, or specific + // HTTP versions can be specified like so: "http/1", "http/1.1", + // "http/2", "http/3", or minimum versions: "http/2+", etc. MatchProtocol string // MatchRemoteIP matches requests by client IP (or CIDR range). diff --git a/modules/caddyhttp/server.go b/modules/caddyhttp/server.go index 5daa5b2bf..4d47d266e 100644 --- a/modules/caddyhttp/server.go +++ b/modules/caddyhttp/server.go @@ -170,9 +170,10 @@ type Server struct { errorLogger *zap.Logger ctx caddy.Context - server *http.Server - h3server *http3.Server - addresses []caddy.NetworkAddress + server *http.Server + h3server *http3.Server + h3listeners []net.PacketConn // TODO: we have to hold these because quic-go won't close listeners it didn't create + addresses []caddy.NetworkAddress shutdownAt time.Time shutdownAtMu *sync.RWMutex @@ -193,9 +194,11 @@ func (s *Server) ServeHTTP(w http.ResponseWriter, r *http.Request) { atomic.AddInt64(&s.activeRequests, 1) defer atomic.AddInt64(&s.activeRequests, -1) - err := s.h3server.SetQuicHeaders(w.Header()) - if err != nil { - s.logger.Error("setting HTTP/3 Alt-Svc header", zap.Error(err)) + if r.ProtoMajor < 3 { + err := s.h3server.SetQuicHeaders(w.Header()) + if err != nil { + s.logger.Error("setting HTTP/3 Alt-Svc header", zap.Error(err)) + } } } @@ -493,8 +496,27 @@ func (s *Server) findLastRouteWithHostMatcher() int { // serveHTTP3 creates a QUIC listener, configures an HTTP/3 server if // not already done, and then uses that server to serve HTTP/3 over // the listener, with Server s as the handler. -func (s *Server) serveHTTP3(hostport string, tlsCfg *tls.Config) error { - h3ln, err := caddy.ListenQUIC(hostport, tlsCfg, &s.activeRequests) +func (s *Server) serveHTTP3(addr caddy.NetworkAddress, tlsCfg *tls.Config) error { + switch addr.Network { + case "unix": + addr.Network = "unixgram" + case "tcp": + addr.Network = "udp" + case "tcp4": + addr.Network = "udp4" + case "tcp6": + addr.Network = "udp6" + default: + return fmt.Errorf("unsure what network to use for HTTP/3 given network type: %s", addr.Network) + } + + lnAny, err := addr.Listen(s.ctx, 0, net.ListenConfig{}) + if err != nil { + return err + } + ln := lnAny.(net.PacketConn) + + h3ln, err := caddy.ListenQUIC(ln, tlsCfg, &s.activeRequests) if err != nil { return fmt.Errorf("starting HTTP/3 QUIC listener: %v", err) } @@ -512,6 +534,8 @@ func (s *Server) serveHTTP3(hostport string, tlsCfg *tls.Config) error { } } + s.h3listeners = append(s.h3listeners, lnAny.(net.PacketConn)) + //nolint:errcheck go s.h3server.ServeListener(h3ln)