package gziphandler import ( "bufio" "compress/gzip" "fmt" "io" "net" "net/http" "strconv" "strings" "sync" ) const ( vary = "Vary" acceptEncoding = "Accept-Encoding" contentEncoding = "Content-Encoding" contentType = "Content-Type" contentLength = "Content-Length" ) type codings map[string]float64 const ( // DefaultQValue is the default qvalue to assign to an encoding if no explicit qvalue is set. // This is actually kind of ambiguous in RFC 2616, so hopefully it's correct. // The examples seem to indicate that it is. DefaultQValue = 1.0 // DefaultMinSize defines the minimum size to reach to enable compression. // It's 512 bytes. DefaultMinSize = 512 ) // gzipWriterPools stores a sync.Pool for each compression level for reuse of // gzip.Writers. Use poolIndex to covert a compression level to an index into // gzipWriterPools. var gzipWriterPools [gzip.BestCompression - gzip.BestSpeed + 2]*sync.Pool func init() { for i := gzip.BestSpeed; i <= gzip.BestCompression; i++ { addLevelPool(i) } addLevelPool(gzip.DefaultCompression) } // poolIndex maps a compression level to its index into gzipWriterPools. It // assumes that level is a valid gzip compression level. func poolIndex(level int) int { // gzip.DefaultCompression == -1, so we need to treat it special. if level == gzip.DefaultCompression { return gzip.BestCompression - gzip.BestSpeed + 1 } return level - gzip.BestSpeed } func addLevelPool(level int) { gzipWriterPools[poolIndex(level)] = &sync.Pool{ New: func() interface{} { // NewWriterLevel only returns error on a bad level, we are guaranteeing // that this will be a valid level so it is okay to ignore the returned // error. w, _ := gzip.NewWriterLevel(nil, level) return w }, } } // GzipResponseWriter provides an http.ResponseWriter interface, which gzips // bytes before writing them to the underlying response. This doesn't close the // writers, so don't forget to do that. // It can be configured to skip response smaller than minSize. type GzipResponseWriter struct { http.ResponseWriter index int // Index for gzipWriterPools. gw *gzip.Writer code int // Saves the WriteHeader value. minSize int // Specifed the minimum response size to gzip. If the response length is bigger than this value, it is compressed. buf []byte // Holds the first part of the write before reaching the minSize or the end of the write. contentTypes []string // Only compress if the response is one of these content-types. All are accepted if empty. } // Write appends data to the gzip writer. func (w *GzipResponseWriter) Write(b []byte) (int, error) { // If content type is not set. if _, ok := w.Header()[contentType]; !ok { // It infer it from the uncompressed body. w.Header().Set(contentType, http.DetectContentType(b)) } // GZIP responseWriter is initialized. Use the GZIP responseWriter. if w.gw != nil { n, err := w.gw.Write(b) return n, err } // Save the write into a buffer for later use in GZIP responseWriter (if content is long enough) or at close with regular responseWriter. // On the first write, w.buf changes from nil to a valid slice w.buf = append(w.buf, b...) // If the global writes are bigger than the minSize and we're about to write // a response containing a content type we want to handle, enable // compression. if len(w.buf) >= w.minSize && handleContentType(w.contentTypes, w) { err := w.startGzip() if err != nil { return 0, err } } return len(b), nil } // startGzip initialize any GZIP specific informations. func (w *GzipResponseWriter) startGzip() error { // Set the GZIP header. w.Header().Set(contentEncoding, "gzip") // if the Content-Length is already set, then calls to Write on gzip // will fail to set the Content-Length header since its already set // See: https://github.com/golang/go/issues/14975. w.Header().Del(contentLength) // Write the header to gzip response. if w.code != 0 { w.ResponseWriter.WriteHeader(w.code) } // Initialize the GZIP response. w.init() // Flush the buffer into the gzip reponse. n, err := w.gw.Write(w.buf) // This should never happen (per io.Writer docs), but if the write didn't // accept the entire buffer but returned no specific error, we have no clue // what's going on, so abort just to be safe. if err == nil && n < len(w.buf) { return io.ErrShortWrite } w.buf = nil return err } // WriteHeader just saves the response code until close or GZIP effective writes. func (w *GzipResponseWriter) WriteHeader(code int) { w.code = code } // init graps a new gzip writer from the gzipWriterPool and writes the correct // content encoding header. func (w *GzipResponseWriter) init() { // Bytes written during ServeHTTP are redirected to this gzip writer // before being written to the underlying response. gzw := gzipWriterPools[w.index].Get().(*gzip.Writer) gzw.Reset(w.ResponseWriter) w.gw = gzw } // Close will close the gzip.Writer and will put it back in the gzipWriterPool. func (w *GzipResponseWriter) Close() error { if w.gw == nil { // Gzip not trigged yet, write out regular response. if w.code != 0 { w.ResponseWriter.WriteHeader(w.code) } if w.buf != nil { _, writeErr := w.ResponseWriter.Write(w.buf) // Returns the error if any at write. if writeErr != nil { return fmt.Errorf("gziphandler: write to regular responseWriter at close gets error: %q", writeErr.Error()) } } return nil } err := w.gw.Close() gzipWriterPools[w.index].Put(w.gw) w.gw = nil return err } // Flush flushes the underlying *gzip.Writer and then the underlying // http.ResponseWriter if it is an http.Flusher. This makes GzipResponseWriter // an http.Flusher. func (w *GzipResponseWriter) Flush() { if w.gw != nil { w.gw.Flush() } if fw, ok := w.ResponseWriter.(http.Flusher); ok { fw.Flush() } } // Hijack implements http.Hijacker. If the underlying ResponseWriter is a // Hijacker, its Hijack method is returned. Otherwise an error is returned. func (w *GzipResponseWriter) Hijack() (net.Conn, *bufio.ReadWriter, error) { if hj, ok := w.ResponseWriter.(http.Hijacker); ok { return hj.Hijack() } return nil, nil, fmt.Errorf("http.Hijacker interface is not supported") } // verify Hijacker interface implementation var _ http.Hijacker = &GzipResponseWriter{} // MustNewGzipLevelHandler behaves just like NewGzipLevelHandler except that in // an error case it panics rather than returning an error. func MustNewGzipLevelHandler(level int) func(http.Handler) http.Handler { wrap, err := NewGzipLevelHandler(level) if err != nil { panic(err) } return wrap } // NewGzipLevelHandler returns a wrapper function (often known as middleware) // which can be used to wrap an HTTP handler to transparently gzip the response // body if the client supports it (via the Accept-Encoding header). Responses will // be encoded at the given gzip compression level. An error will be returned only // if an invalid gzip compression level is given, so if one can ensure the level // is valid, the returned error can be safely ignored. func NewGzipLevelHandler(level int) (func(http.Handler) http.Handler, error) { return NewGzipLevelAndMinSize(level, DefaultMinSize) } // NewGzipLevelAndMinSize behave as NewGzipLevelHandler except it let the caller // specify the minimum size before compression. func NewGzipLevelAndMinSize(level, minSize int) (func(http.Handler) http.Handler, error) { return GzipHandlerWithOpts(CompressionLevel(level), MinSize(minSize)) } func GzipHandlerWithOpts(opts ...option) (func(http.Handler) http.Handler, error) { c := &config{ level: gzip.DefaultCompression, minSize: DefaultMinSize, } for _, o := range opts { o(c) } if err := c.validate(); err != nil { return nil, err } return func(h http.Handler) http.Handler { index := poolIndex(c.level) return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) { w.Header().Add(vary, acceptEncoding) if acceptsGzip(r) { gw := &GzipResponseWriter{ ResponseWriter: w, index: index, minSize: c.minSize, contentTypes: c.contentTypes, } defer gw.Close() h.ServeHTTP(gw, r) } else { h.ServeHTTP(w, r) } }) }, nil } // Used for functional configuration. type config struct { minSize int level int contentTypes []string } func (c *config) validate() error { if c.level != gzip.DefaultCompression && (c.level < gzip.BestSpeed || c.level > gzip.BestCompression) { return fmt.Errorf("invalid compression level requested: %d", c.level) } if c.minSize < 0 { return fmt.Errorf("minimum size must be more than zero") } return nil } type option func(c *config) func MinSize(size int) option { return func(c *config) { c.minSize = size } } func CompressionLevel(level int) option { return func(c *config) { c.level = level } } func ContentTypes(types []string) option { return func(c *config) { c.contentTypes = []string{} for _, v := range types { c.contentTypes = append(c.contentTypes, strings.ToLower(v)) } } } // GzipHandler wraps an HTTP handler, to transparently gzip the response body if // the client supports it (via the Accept-Encoding header). This will compress at // the default compression level. func GzipHandler(h http.Handler) http.Handler { wrapper, _ := NewGzipLevelHandler(gzip.DefaultCompression) return wrapper(h) } // acceptsGzip returns true if the given HTTP request indicates that it will // accept a gzipped response. func acceptsGzip(r *http.Request) bool { acceptedEncodings, _ := parseEncodings(r.Header.Get(acceptEncoding)) return acceptedEncodings["gzip"] > 0.0 } // returns true if we've been configured to compress the specific content type. func handleContentType(contentTypes []string, w http.ResponseWriter) bool { // If contentTypes is empty we handle all content types. if len(contentTypes) == 0 { return true } ct := strings.ToLower(w.Header().Get(contentType)) for _, c := range contentTypes { if c == ct { return true } } return false } // parseEncodings attempts to parse a list of codings, per RFC 2616, as might // appear in an Accept-Encoding header. It returns a map of content-codings to // quality values, and an error containing the errors encountered. It's probably // safe to ignore those, because silently ignoring errors is how the internet // works. // // See: http://tools.ietf.org/html/rfc2616#section-14.3. func parseEncodings(s string) (codings, error) { c := make(codings) var e []string for _, ss := range strings.Split(s, ",") { coding, qvalue, err := parseCoding(ss) if err != nil { e = append(e, err.Error()) } else { c[coding] = qvalue } } // TODO (adammck): Use a proper multi-error struct, so the individual errors // can be extracted if anyone cares. if len(e) > 0 { return c, fmt.Errorf("errors while parsing encodings: %s", strings.Join(e, ", ")) } return c, nil } // parseCoding parses a single conding (content-coding with an optional qvalue), // as might appear in an Accept-Encoding header. It attempts to forgive minor // formatting errors. func parseCoding(s string) (coding string, qvalue float64, err error) { for n, part := range strings.Split(s, ";") { part = strings.TrimSpace(part) qvalue = DefaultQValue if n == 0 { coding = strings.ToLower(part) } else if strings.HasPrefix(part, "q=") { qvalue, err = strconv.ParseFloat(strings.TrimPrefix(part, "q="), 64) if qvalue < 0.0 { qvalue = 0.0 } else if qvalue > 1.0 { qvalue = 1.0 } } } if coding == "" { err = fmt.Errorf("empty content-coding") } return }