1 // Copyright 2009 The Go Authors. All rights reserved. 2 // Use of this source code is governed by a BSD-style 3 // license that can be found in the LICENSE file. 4 5 // The wire protocol for HTTP's "chunked" Transfer-Encoding. 6 7 // Package internal contains HTTP internals shared by net/http and 8 // net/http/httputil. 9 package internal 10 11 import ( 12 "bufio" 13 "bytes" 14 "errors" 15 "fmt" 16 "io" 17 ) 18 19 const maxLineLength = 4096 // assumed <= bufio.defaultBufSize 20 21 var ErrLineTooLong = errors.New("header line too long") 22 23 // NewChunkedReader returns a new chunkedReader that translates the data read from r 24 // out of HTTP "chunked" format before returning it. 25 // The chunkedReader returns io.EOF when the final 0-length chunk is read. 26 // 27 // NewChunkedReader is not needed by normal applications. The http package 28 // automatically decodes chunking when reading response bodies. 29 func NewChunkedReader(r io.Reader) io.Reader { 30 br, ok := r.(*bufio.Reader) 31 if !ok { 32 br = bufio.NewReader(r) 33 } 34 return &chunkedReader{r: br} 35 } 36 37 type chunkedReader struct { 38 r *bufio.Reader 39 n uint64 // unread bytes in chunk 40 err error 41 buf [2]byte 42 checkEnd bool // whether need to check for \r\n chunk footer 43 } 44 45 func (cr *chunkedReader) beginChunk() { 46 // chunk-size CRLF 47 var line []byte 48 line, cr.err = readChunkLine(cr.r) 49 if cr.err != nil { 50 return 51 } 52 cr.n, cr.err = parseHexUint(line) 53 if cr.err != nil { 54 return 55 } 56 if cr.n == 0 { 57 cr.err = io.EOF 58 } 59 } 60 61 func (cr *chunkedReader) chunkHeaderAvailable() bool { 62 n := cr.r.Buffered() 63 if n > 0 { 64 peek, _ := cr.r.Peek(n) 65 return bytes.IndexByte(peek, '\n') >= 0 66 } 67 return false 68 } 69 70 func (cr *chunkedReader) Read(b []uint8) (n int, err error) { 71 for cr.err == nil { 72 if cr.checkEnd { 73 if n > 0 && cr.r.Buffered() < 2 { 74 // We have some data. Return early (per the io.Reader 75 // contract) instead of potentially blocking while 76 // reading more. 77 break 78 } 79 if _, cr.err = io.ReadFull(cr.r, cr.buf[:2]); cr.err == nil { 80 if string(cr.buf[:]) != "\r\n" { 81 cr.err = errors.New("malformed chunked encoding") 82 break 83 } 84 } 85 cr.checkEnd = false 86 } 87 if cr.n == 0 { 88 if n > 0 && !cr.chunkHeaderAvailable() { 89 // We've read enough. Don't potentially block 90 // reading a new chunk header. 91 break 92 } 93 cr.beginChunk() 94 continue 95 } 96 if len(b) == 0 { 97 break 98 } 99 rbuf := b 100 if uint64(len(rbuf)) > cr.n { 101 rbuf = rbuf[:cr.n] 102 } 103 var n0 int 104 n0, cr.err = cr.r.Read(rbuf) 105 n += n0 106 b = b[n0:] 107 cr.n -= uint64(n0) 108 // If we're at the end of a chunk, read the next two 109 // bytes to verify they are "\r\n". 110 if cr.n == 0 && cr.err == nil { 111 cr.checkEnd = true 112 } 113 } 114 return n, cr.err 115 } 116 117 // Read a line of bytes (up to \n) from b. 118 // Give up if the line exceeds maxLineLength. 119 // The returned bytes are owned by the bufio.Reader 120 // so they are only valid until the next bufio read. 121 func readChunkLine(b *bufio.Reader) ([]byte, error) { 122 p, err := b.ReadSlice('\n') 123 if err != nil { 124 // We always know when EOF is coming. 125 // If the caller asked for a line, there should be a line. 126 if err == io.EOF { 127 err = io.ErrUnexpectedEOF 128 } else if err == bufio.ErrBufferFull { 129 err = ErrLineTooLong 130 } 131 return nil, err 132 } 133 if len(p) >= maxLineLength { 134 return nil, ErrLineTooLong 135 } 136 p = trimTrailingWhitespace(p) 137 p, err = removeChunkExtension(p) 138 if err != nil { 139 return nil, err 140 } 141 return p, nil 142 } 143 144 func trimTrailingWhitespace(b []byte) []byte { 145 for len(b) > 0 && isASCIISpace(b[len(b)-1]) { 146 b = b[:len(b)-1] 147 } 148 return b 149 } 150 151 func isASCIISpace(b byte) bool { 152 return b == ' ' || b == '\t' || b == '\n' || b == '\r' 153 } 154 155 // removeChunkExtension removes any chunk-extension from p. 156 // For example, 157 // "0" => "0" 158 // "0;token" => "0" 159 // "0;token=val" => "0" 160 // `0;token="quoted string"` => "0" 161 func removeChunkExtension(p []byte) ([]byte, error) { 162 semi := bytes.IndexByte(p, ';') 163 if semi == -1 { 164 return p, nil 165 } 166 // TODO: care about exact syntax of chunk extensions? We're 167 // ignoring and stripping them anyway. For now just never 168 // return an error. 169 return p[:semi], nil 170 } 171 172 // NewChunkedWriter returns a new chunkedWriter that translates writes into HTTP 173 // "chunked" format before writing them to w. Closing the returned chunkedWriter 174 // sends the final 0-length chunk that marks the end of the stream. 175 // 176 // NewChunkedWriter is not needed by normal applications. The http 177 // package adds chunking automatically if handlers don't set a 178 // Content-Length header. Using newChunkedWriter inside a handler 179 // would result in double chunking or chunking with a Content-Length 180 // length, both of which are wrong. 181 func NewChunkedWriter(w io.Writer) io.WriteCloser { 182 return &chunkedWriter{w} 183 } 184 185 // Writing to chunkedWriter translates to writing in HTTP chunked Transfer 186 // Encoding wire format to the underlying Wire chunkedWriter. 187 type chunkedWriter struct { 188 Wire io.Writer 189 } 190 191 // Write the contents of data as one chunk to Wire. 192 // NOTE: Note that the corresponding chunk-writing procedure in Conn.Write has 193 // a bug since it does not check for success of io.WriteString 194 func (cw *chunkedWriter) Write(data []byte) (n int, err error) { 195 196 // Don't send 0-length data. It looks like EOF for chunked encoding. 197 if len(data) == 0 { 198 return 0, nil 199 } 200 201 if _, err = fmt.Fprintf(cw.Wire, "%x\r\n", len(data)); err != nil { 202 return 0, err 203 } 204 if n, err = cw.Wire.Write(data); err != nil { 205 return 206 } 207 if n != len(data) { 208 err = io.ErrShortWrite 209 return 210 } 211 if _, err = io.WriteString(cw.Wire, "\r\n"); err != nil { 212 return 213 } 214 if bw, ok := cw.Wire.(*FlushAfterChunkWriter); ok { 215 err = bw.Flush() 216 } 217 return 218 } 219 220 func (cw *chunkedWriter) Close() error { 221 _, err := io.WriteString(cw.Wire, "0\r\n") 222 return err 223 } 224 225 // FlushAfterChunkWriter signals from the caller of NewChunkedWriter 226 // that each chunk should be followed by a flush. It is used by the 227 // http.Transport code to keep the buffering behavior for headers and 228 // trailers, but flush out chunks aggressively in the middle for 229 // request bodies which may be generated slowly. See Issue 6574. 230 type FlushAfterChunkWriter struct { 231 *bufio.Writer 232 } 233 234 func parseHexUint(v []byte) (n uint64, err error) { 235 for i, b := range v { 236 switch { 237 case '0' <= b && b <= '9': 238 b = b - '0' 239 case 'a' <= b && b <= 'f': 240 b = b - 'a' + 10 241 case 'A' <= b && b <= 'F': 242 b = b - 'A' + 10 243 default: 244 return 0, errors.New("invalid byte in chunk length") 245 } 246 if i == 16 { 247 return 0, errors.New("http chunk length too large") 248 } 249 n <<= 4 250 n |= uint64(b) 251 } 252 return 253 } 254
