0
Fork 0
mirror of https://github.com/caddyserver/caddy.git synced 2024-12-23 22:27:38 -05:00
caddy/caddyhttp/httpserver/server.go
Matt Holt d5371aff22 httpserver/all: Clean up and standardize request URL handling (#1633)
* httpserver/all: Clean up and standardize request URL handling

The HTTP server now always creates a context value on the request which
is a copy of the request's URL struct. It should not be modified by
middlewares, but it is safe to get the value out of the request and make
changes to it locally-scoped. Thus, the value in the context always
stores the original request URL information as it was received. Any
rewrites that happen will be to the request's URL field directly.

The HTTP server no longer cleans /sanitizes the request URL. It made too
many strong assumptions and ended up making a lot of middleware more
complicated, including upstream proxying (and fastcgi). To alleviate
this complexity, we no longer change the request URL. Middlewares are
responsible to access the disk safely by using http.Dir or, if not
actually opening files, they can use httpserver.SafePath().

I'm hoping this will address issues with #1624, #1584, #1582, and others.

* staticfiles: Fix test on Windows

@abiosoft: I still can't figure out exactly what this is for. 😅

* Use (potentially) changed URL for browse redirects, as before

* Use filepath.ToSlash, clean up a couple proxy test cases

* Oops, fix variable name
2017-05-01 23:11:10 -06:00

566 lines
17 KiB
Go

// Package httpserver implements an HTTP server on top of Caddy.
package httpserver
import (
"context"
"crypto/tls"
"fmt"
"io"
"log"
"net"
"net/http"
"net/url"
"os"
"path"
"path/filepath"
"runtime"
"strings"
"sync"
"time"
"github.com/lucas-clemente/quic-go/h2quic"
"github.com/mholt/caddy"
"github.com/mholt/caddy/caddyhttp/staticfiles"
"github.com/mholt/caddy/caddytls"
)
// Server is the HTTP server implementation.
type Server struct {
Server *http.Server
quicServer *h2quic.Server
listener net.Listener
listenerMu sync.Mutex
sites []*SiteConfig
connTimeout time.Duration // max time to wait for a connection before force stop
tlsGovChan chan struct{} // close to stop the TLS maintenance goroutine
vhosts *vhostTrie
}
// ensure it satisfies the interface
var _ caddy.GracefulServer = new(Server)
var defaultALPN = []string{"h2", "http/1.1"}
// makeTLSConfig extracts TLS settings from each site config to
// build a tls.Config usable in Caddy HTTP servers. The returned
// config will be nil if TLS is disabled for these sites.
func makeTLSConfig(group []*SiteConfig) (*tls.Config, error) {
var tlsConfigs []*caddytls.Config
for i := range group {
if HTTP2 && len(group[i].TLS.ALPN) == 0 {
// if no application-level protocol was configured up to now,
// default to HTTP/2, then HTTP/1.1 if necessary
group[i].TLS.ALPN = defaultALPN
}
tlsConfigs = append(tlsConfigs, group[i].TLS)
}
return caddytls.MakeTLSConfig(tlsConfigs)
}
// NewServer creates a new Server instance that will listen on addr
// and will serve the sites configured in group.
func NewServer(addr string, group []*SiteConfig) (*Server, error) {
s := &Server{
Server: makeHTTPServerWithTimeouts(addr, group),
vhosts: newVHostTrie(),
sites: group,
connTimeout: GracefulTimeout,
}
s.Server.Handler = s // this is weird, but whatever
// extract TLS settings from each site config to build
// a tls.Config, which will not be nil if TLS is enabled
tlsConfig, err := makeTLSConfig(group)
if err != nil {
return nil, err
}
s.Server.TLSConfig = tlsConfig
// Enable QUIC if desired
if QUIC {
s.quicServer = &h2quic.Server{Server: s.Server}
s.Server.Handler = s.wrapWithSvcHeaders(s.Server.Handler)
}
// if TLS is enabled, make sure we prepare the Server accordingly
if s.Server.TLSConfig != nil {
// wrap the HTTP handler with a handler that does MITM detection
tlsh := &tlsHandler{next: s.Server.Handler}
s.Server.Handler = tlsh // this needs to be the "outer" handler when Serve() is called, for type assertion
// when Serve() creates the TLS listener later, that listener should
// be adding a reference the ClientHello info to a map; this callback
// will be sure to clear out that entry when the connection closes.
s.Server.ConnState = func(c net.Conn, cs http.ConnState) {
// when a connection closes or is hijacked, delete its entry
// in the map, because we are done with it.
if tlsh.listener != nil {
if cs == http.StateHijacked || cs == http.StateClosed {
tlsh.listener.helloInfosMu.Lock()
delete(tlsh.listener.helloInfos, c.RemoteAddr().String())
tlsh.listener.helloInfosMu.Unlock()
}
}
}
// As of Go 1.7, if the Server's TLSConfig is not nil, HTTP/2 is enabled only
// if TLSConfig.NextProtos includes the string "h2"
if HTTP2 && len(s.Server.TLSConfig.NextProtos) == 0 {
// some experimenting shows that this NextProtos must have at least
// one value that overlaps with the NextProtos of any other tls.Config
// that is returned from GetConfigForClient; if there is no overlap,
// the connection will fail (as of Go 1.8, Feb. 2017).
s.Server.TLSConfig.NextProtos = defaultALPN
}
}
// Compile custom middleware for every site (enables virtual hosting)
for _, site := range group {
stack := Handler(staticfiles.FileServer{Root: http.Dir(site.Root), Hide: site.HiddenFiles})
for i := len(site.middleware) - 1; i >= 0; i-- {
stack = site.middleware[i](stack)
}
site.middlewareChain = stack
s.vhosts.Insert(site.Addr.VHost(), site)
}
return s, nil
}
// makeHTTPServerWithTimeouts makes an http.Server from the group of
// configs in a way that configures timeouts (or, if not set, it uses
// the default timeouts) by combining the configuration of each
// SiteConfig in the group. (Timeouts are important for mitigating
// slowloris attacks.)
func makeHTTPServerWithTimeouts(addr string, group []*SiteConfig) *http.Server {
// find the minimum duration configured for each timeout
var min Timeouts
for _, cfg := range group {
if cfg.Timeouts.ReadTimeoutSet &&
(!min.ReadTimeoutSet || cfg.Timeouts.ReadTimeout < min.ReadTimeout) {
min.ReadTimeoutSet = true
min.ReadTimeout = cfg.Timeouts.ReadTimeout
}
if cfg.Timeouts.ReadHeaderTimeoutSet &&
(!min.ReadHeaderTimeoutSet || cfg.Timeouts.ReadHeaderTimeout < min.ReadHeaderTimeout) {
min.ReadHeaderTimeoutSet = true
min.ReadHeaderTimeout = cfg.Timeouts.ReadHeaderTimeout
}
if cfg.Timeouts.WriteTimeoutSet &&
(!min.WriteTimeoutSet || cfg.Timeouts.WriteTimeout < min.WriteTimeout) {
min.WriteTimeoutSet = true
min.WriteTimeout = cfg.Timeouts.WriteTimeout
}
if cfg.Timeouts.IdleTimeoutSet &&
(!min.IdleTimeoutSet || cfg.Timeouts.IdleTimeout < min.IdleTimeout) {
min.IdleTimeoutSet = true
min.IdleTimeout = cfg.Timeouts.IdleTimeout
}
}
// for the values that were not set, use defaults
if !min.ReadTimeoutSet {
min.ReadTimeout = defaultTimeouts.ReadTimeout
}
if !min.ReadHeaderTimeoutSet {
min.ReadHeaderTimeout = defaultTimeouts.ReadHeaderTimeout
}
if !min.WriteTimeoutSet {
min.WriteTimeout = defaultTimeouts.WriteTimeout
}
if !min.IdleTimeoutSet {
min.IdleTimeout = defaultTimeouts.IdleTimeout
}
// set the final values on the server and return it
return &http.Server{
Addr: addr,
ReadTimeout: min.ReadTimeout,
ReadHeaderTimeout: min.ReadHeaderTimeout,
WriteTimeout: min.WriteTimeout,
IdleTimeout: min.IdleTimeout,
}
}
func (s *Server) wrapWithSvcHeaders(previousHandler http.Handler) http.HandlerFunc {
return func(w http.ResponseWriter, r *http.Request) {
s.quicServer.SetQuicHeaders(w.Header())
previousHandler.ServeHTTP(w, r)
}
}
// Listen creates an active listener for s that can be
// used to serve requests.
func (s *Server) Listen() (net.Listener, error) {
if s.Server == nil {
return nil, fmt.Errorf("Server field is nil")
}
ln, err := net.Listen("tcp", s.Server.Addr)
if err != nil {
var succeeded bool
if runtime.GOOS == "windows" {
// Windows has been known to keep sockets open even after closing the listeners.
// Tests reveal this error case easily because they call Start() then Stop()
// in succession. TODO: Better way to handle this? And why limit this to Windows?
for i := 0; i < 20; i++ {
time.Sleep(100 * time.Millisecond)
ln, err = net.Listen("tcp", s.Server.Addr)
if err == nil {
succeeded = true
break
}
}
}
if !succeeded {
return nil, err
}
}
if tcpLn, ok := ln.(*net.TCPListener); ok {
ln = tcpKeepAliveListener{TCPListener: tcpLn}
}
cln := ln.(caddy.Listener)
for _, site := range s.sites {
for _, m := range site.listenerMiddleware {
cln = m(cln)
}
}
// Very important to return a concrete caddy.Listener
// implementation for graceful restarts.
return cln.(caddy.Listener), nil
}
// ListenPacket creates udp connection for QUIC if it is enabled,
func (s *Server) ListenPacket() (net.PacketConn, error) {
if QUIC {
udpAddr, err := net.ResolveUDPAddr("udp", s.Server.Addr)
if err != nil {
return nil, err
}
return net.ListenUDP("udp", udpAddr)
}
return nil, nil
}
// Serve serves requests on ln. It blocks until ln is closed.
func (s *Server) Serve(ln net.Listener) error {
s.listenerMu.Lock()
s.listener = ln
s.listenerMu.Unlock()
if s.Server.TLSConfig != nil {
// Create TLS listener - note that we do not replace s.listener
// with this TLS listener; tls.listener is unexported and does
// not implement the File() method we need for graceful restarts
// on POSIX systems.
// TODO: Is this ^ still relevant anymore? Maybe we can now that it's a net.Listener...
ln = newTLSListener(ln, s.Server.TLSConfig)
if handler, ok := s.Server.Handler.(*tlsHandler); ok {
handler.listener = ln.(*tlsHelloListener)
}
// Rotate TLS session ticket keys
s.tlsGovChan = caddytls.RotateSessionTicketKeys(s.Server.TLSConfig)
}
err := s.Server.Serve(ln)
if QUIC {
s.quicServer.Close()
}
return err
}
// ServePacket serves QUIC requests on pc until it is closed.
func (s *Server) ServePacket(pc net.PacketConn) error {
if QUIC {
err := s.quicServer.Serve(pc.(*net.UDPConn))
return fmt.Errorf("serving QUIC connections: %v", err)
}
return nil
}
// ServeHTTP is the entry point of all HTTP requests.
func (s *Server) ServeHTTP(w http.ResponseWriter, r *http.Request) {
defer func() {
// We absolutely need to be sure we stay alive up here,
// even though, in theory, the errors middleware does this.
if rec := recover(); rec != nil {
log.Printf("[PANIC] %v", rec)
DefaultErrorFunc(w, r, http.StatusInternalServerError)
}
}()
// copy the original, unchanged URL into the context
// so it can be referenced by middlewares
urlCopy := *r.URL
if r.URL.User != nil {
userInfo := new(url.Userinfo)
*userInfo = *r.URL.User
urlCopy.User = userInfo
}
c := context.WithValue(r.Context(), OriginalURLCtxKey, urlCopy)
r = r.WithContext(c)
w.Header().Set("Server", "Caddy")
status, _ := s.serveHTTP(w, r)
// Fallback error response in case error handling wasn't chained in
if status >= 400 {
DefaultErrorFunc(w, r, status)
}
}
func (s *Server) serveHTTP(w http.ResponseWriter, r *http.Request) (int, error) {
// strip out the port because it's not used in virtual
// hosting; the port is irrelevant because each listener
// is on a different port.
hostname, _, err := net.SplitHostPort(r.Host)
if err != nil {
hostname = r.Host
}
// look up the virtualhost; if no match, serve error
vhost, pathPrefix := s.vhosts.Match(hostname + r.URL.Path)
if vhost == nil {
// check for ACME challenge even if vhost is nil;
// could be a new host coming online soon
if caddytls.HTTPChallengeHandler(w, r, "localhost", caddytls.DefaultHTTPAlternatePort) {
return 0, nil
}
// otherwise, log the error and write a message to the client
remoteHost, _, err := net.SplitHostPort(r.RemoteAddr)
if err != nil {
remoteHost = r.RemoteAddr
}
WriteTextResponse(w, http.StatusNotFound, "No such site at "+s.Server.Addr)
log.Printf("[INFO] %s - No such site at %s (Remote: %s, Referer: %s)",
hostname, s.Server.Addr, remoteHost, r.Header.Get("Referer"))
return 0, nil
}
// we still check for ACME challenge if the vhost exists,
// because we must apply its HTTP challenge config settings
if s.proxyHTTPChallenge(vhost, w, r) {
return 0, nil
}
// trim the path portion of the site address from the beginning of
// the URL path, so a request to example.com/foo/blog on the site
// defined as example.com/foo appears as /blog instead of /foo/blog.
if pathPrefix != "/" {
r.URL.Path = strings.TrimPrefix(r.URL.Path, pathPrefix)
if !strings.HasPrefix(r.URL.Path, "/") {
r.URL.Path = "/" + r.URL.Path
}
}
// Apply the path-based request body size limit
// The error returned by MaxBytesReader is meant to be handled
// by whichever middleware/plugin that receives it when calling
// .Read() or a similar method on the request body
// TODO: Make this middleware instead?
if r.Body != nil {
for _, pathlimit := range vhost.MaxRequestBodySizes {
if Path(r.URL.Path).Matches(pathlimit.Path) {
r.Body = MaxBytesReader(w, r.Body, pathlimit.Limit)
break
}
}
}
return vhost.middlewareChain.ServeHTTP(w, r)
}
// proxyHTTPChallenge solves the ACME HTTP challenge if r is the HTTP
// request for the challenge. If it is, and if the request has been
// fulfilled (response written), true is returned; false otherwise.
// If you don't have a vhost, just call the challenge handler directly.
func (s *Server) proxyHTTPChallenge(vhost *SiteConfig, w http.ResponseWriter, r *http.Request) bool {
if vhost.Addr.Port != caddytls.HTTPChallengePort {
return false
}
if vhost.TLS != nil && vhost.TLS.Manual {
return false
}
altPort := caddytls.DefaultHTTPAlternatePort
if vhost.TLS != nil && vhost.TLS.AltHTTPPort != "" {
altPort = vhost.TLS.AltHTTPPort
}
return caddytls.HTTPChallengeHandler(w, r, vhost.ListenHost, altPort)
}
// Address returns the address s was assigned to listen on.
func (s *Server) Address() string {
return s.Server.Addr
}
// Stop stops s gracefully (or forcefully after timeout) and
// closes its listener.
func (s *Server) Stop() error {
ctx, cancel := context.WithTimeout(context.Background(), s.connTimeout)
defer cancel()
err := s.Server.Shutdown(ctx)
if err != nil {
return err
}
// signal any TLS governor goroutines to exit
if s.tlsGovChan != nil {
close(s.tlsGovChan)
}
return nil
}
// OnStartupComplete lists the sites served by this server
// and any relevant information, assuming caddy.Quiet == false.
func (s *Server) OnStartupComplete() {
if caddy.Quiet {
return
}
for _, site := range s.sites {
output := site.Addr.String()
if caddy.IsLoopback(s.Address()) && !caddy.IsLoopback(site.Addr.Host) {
output += " (only accessible on this machine)"
}
fmt.Println(output)
log.Println(output)
}
}
// defaultTimeouts stores the default timeout values to use
// if left unset by user configuration. NOTE: Default timeouts
// are disabled (see issue #1464).
var defaultTimeouts Timeouts
// tcpKeepAliveListener sets TCP keep-alive timeouts on accepted
// connections. It's used by ListenAndServe and ListenAndServeTLS so
// dead TCP connections (e.g. closing laptop mid-download) eventually
// go away.
//
// Borrowed from the Go standard library.
type tcpKeepAliveListener struct {
*net.TCPListener
}
// Accept accepts the connection with a keep-alive enabled.
func (ln tcpKeepAliveListener) Accept() (c net.Conn, err error) {
tc, err := ln.AcceptTCP()
if err != nil {
return
}
tc.SetKeepAlive(true)
tc.SetKeepAlivePeriod(3 * time.Minute)
return tc, nil
}
// File implements caddy.Listener; it returns the underlying file of the listener.
func (ln tcpKeepAliveListener) File() (*os.File, error) {
return ln.TCPListener.File()
}
// MaxBytesExceeded is the error type returned by MaxBytesReader
// when the request body exceeds the limit imposed
type MaxBytesExceeded struct{}
func (err MaxBytesExceeded) Error() string {
return "http: request body too large"
}
// MaxBytesReader and its associated methods are borrowed from the
// Go Standard library (comments intact). The only difference is that
// it returns a MaxBytesExceeded error instead of a generic error message
// when the request body has exceeded the requested limit
func MaxBytesReader(w http.ResponseWriter, r io.ReadCloser, n int64) io.ReadCloser {
return &maxBytesReader{w: w, r: r, n: n}
}
type maxBytesReader struct {
w http.ResponseWriter
r io.ReadCloser // underlying reader
n int64 // max bytes remaining
err error // sticky error
}
func (l *maxBytesReader) Read(p []byte) (n int, err error) {
if l.err != nil {
return 0, l.err
}
if len(p) == 0 {
return 0, nil
}
// If they asked for a 32KB byte read but only 5 bytes are
// remaining, no need to read 32KB. 6 bytes will answer the
// question of the whether we hit the limit or go past it.
if int64(len(p)) > l.n+1 {
p = p[:l.n+1]
}
n, err = l.r.Read(p)
if int64(n) <= l.n {
l.n -= int64(n)
l.err = err
return n, err
}
n = int(l.n)
l.n = 0
// The server code and client code both use
// maxBytesReader. This "requestTooLarge" check is
// only used by the server code. To prevent binaries
// which only using the HTTP Client code (such as
// cmd/go) from also linking in the HTTP server, don't
// use a static type assertion to the server
// "*response" type. Check this interface instead:
type requestTooLarger interface {
requestTooLarge()
}
if res, ok := l.w.(requestTooLarger); ok {
res.requestTooLarge()
}
l.err = MaxBytesExceeded{}
return n, l.err
}
func (l *maxBytesReader) Close() error {
return l.r.Close()
}
// DefaultErrorFunc responds to an HTTP request with a simple description
// of the specified HTTP status code.
func DefaultErrorFunc(w http.ResponseWriter, r *http.Request, status int) {
WriteTextResponse(w, status, fmt.Sprintf("%d %s\n", status, http.StatusText(status)))
}
// WriteTextResponse writes body with code status to w. The body will
// be interpreted as plain text.
func WriteTextResponse(w http.ResponseWriter, status int, body string) {
w.Header().Set("Content-Type", "text/plain; charset=utf-8")
w.Header().Set("X-Content-Type-Options", "nosniff")
w.WriteHeader(status)
w.Write([]byte(body))
}
// SafePath joins siteRoot and reqPath and converts it to a path that can
// be used to access a path on the local disk. It ensures the path does
// not traverse outside of the site root.
//
// If opening a file, use http.Dir instead.
func SafePath(siteRoot, reqPath string) string {
reqPath = filepath.ToSlash(reqPath)
reqPath = strings.Replace(reqPath, "\x00", "", -1) // NOTE: Go 1.9 checks for null bytes in the syscall package
if siteRoot == "" {
siteRoot = "."
}
return filepath.Join(siteRoot, filepath.FromSlash(path.Clean("/"+reqPath)))
}
// OriginalURLCtxKey is the key for accessing the original, incoming URL on an HTTP request.
const OriginalURLCtxKey = caddy.CtxKey("original_url")