1 // Copyright 2014 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 // Package hpack implements HPACK, a compression format for 6 // efficiently representing HTTP header fields in the context of HTTP/2. 7 // 8 // See http://tools.ietf.org/html/draft-ietf-httpbis-header-compression-09 9 package hpack 10 11 import ( 12 "bytes" 13 "errors" 14 "fmt" 15 ) 16 17 // A DecodingError is something the spec defines as a decoding error. 18 type DecodingError struct { 19 Err error 20 } 21 22 func (de DecodingError) Error() string { 23 return fmt.Sprintf("decoding error: %v", de.Err) 24 } 25 26 // An InvalidIndexError is returned when an encoder references a table 27 // entry before the static table or after the end of the dynamic table. 28 type InvalidIndexError int 29 30 func (e InvalidIndexError) Error() string { 31 return fmt.Sprintf("invalid indexed representation index %d", int(e)) 32 } 33 34 // A HeaderField is a name-value pair. Both the name and value are 35 // treated as opaque sequences of octets. 36 type HeaderField struct { 37 Name, Value string 38 39 // Sensitive means that this header field should never be 40 // indexed. 41 Sensitive bool 42 } 43 44 // IsPseudo reports whether the header field is an http2 pseudo header. 45 // That is, it reports whether it starts with a colon. 46 // It is not otherwise guaranteed to be a valid pseudo header field, 47 // though. 48 func (hf HeaderField) IsPseudo() bool { 49 return len(hf.Name) != 0 && hf.Name[0] == ':' 50 } 51 52 func (hf HeaderField) String() string { 53 var suffix string 54 if hf.Sensitive { 55 suffix = " (sensitive)" 56 } 57 return fmt.Sprintf("header field %q = %q%s", hf.Name, hf.Value, suffix) 58 } 59 60 // Size returns the size of an entry per RFC 7541 section 4.1. 61 func (hf HeaderField) Size() uint32 { 62 // http://http2.github.io/http2-spec/compression.html#rfc.section.4.1 63 // "The size of the dynamic table is the sum of the size of 64 // its entries. The size of an entry is the sum of its name's 65 // length in octets (as defined in Section 5.2), its value's 66 // length in octets (see Section 5.2), plus 32. The size of 67 // an entry is calculated using the length of the name and 68 // value without any Huffman encoding applied." 69 70 // This can overflow if somebody makes a large HeaderField 71 // Name and/or Value by hand, but we don't care, because that 72 // won't happen on the wire because the encoding doesn't allow 73 // it. 74 return uint32(len(hf.Name) + len(hf.Value) + 32) 75 } 76 77 // A Decoder is the decoding context for incremental processing of 78 // header blocks. 79 type Decoder struct { 80 dynTab dynamicTable 81 emit func(f HeaderField) 82 83 emitEnabled bool // whether calls to emit are enabled 84 maxStrLen int // 0 means unlimited 85 86 // buf is the unparsed buffer. It's only written to 87 // saveBuf if it was truncated in the middle of a header 88 // block. Because it's usually not owned, we can only 89 // process it under Write. 90 buf []byte // not owned; only valid during Write 91 92 // saveBuf is previous data passed to Write which we weren't able 93 // to fully parse before. Unlike buf, we own this data. 94 saveBuf bytes.Buffer 95 } 96 97 // NewDecoder returns a new decoder with the provided maximum dynamic 98 // table size. The emitFunc will be called for each valid field 99 // parsed, in the same goroutine as calls to Write, before Write returns. 100 func NewDecoder(maxDynamicTableSize uint32, emitFunc func(f HeaderField)) *Decoder { 101 d := &Decoder{ 102 emit: emitFunc, 103 emitEnabled: true, 104 } 105 d.dynTab.table.init() 106 d.dynTab.allowedMaxSize = maxDynamicTableSize 107 d.dynTab.setMaxSize(maxDynamicTableSize) 108 return d 109 } 110 111 // ErrStringLength is returned by Decoder.Write when the max string length 112 // (as configured by Decoder.SetMaxStringLength) would be violated. 113 var ErrStringLength = errors.New("hpack: string too long") 114 115 // SetMaxStringLength sets the maximum size of a HeaderField name or 116 // value string. If a string exceeds this length (even after any 117 // decompression), Write will return ErrStringLength. 118 // A value of 0 means unlimited and is the default from NewDecoder. 119 func (d *Decoder) SetMaxStringLength(n int) { 120 d.maxStrLen = n 121 } 122 123 // SetEmitFunc changes the callback used when new header fields 124 // are decoded. 125 // It must be non-nil. It does not affect EmitEnabled. 126 func (d *Decoder) SetEmitFunc(emitFunc func(f HeaderField)) { 127 d.emit = emitFunc 128 } 129 130 // SetEmitEnabled controls whether the emitFunc provided to NewDecoder 131 // should be called. The default is true. 132 // 133 // This facility exists to let servers enforce MAX_HEADER_LIST_SIZE 134 // while still decoding and keeping in-sync with decoder state, but 135 // without doing unnecessary decompression or generating unnecessary 136 // garbage for header fields past the limit. 137 func (d *Decoder) SetEmitEnabled(v bool) { d.emitEnabled = v } 138 139 // EmitEnabled reports whether calls to the emitFunc provided to NewDecoder 140 // are currently enabled. The default is true. 141 func (d *Decoder) EmitEnabled() bool { return d.emitEnabled } 142 143 // TODO: add method *Decoder.Reset(maxSize, emitFunc) to let callers re-use Decoders and their 144 // underlying buffers for garbage reasons. 145 146 func (d *Decoder) SetMaxDynamicTableSize(v uint32) { 147 d.dynTab.setMaxSize(v) 148 } 149 150 // SetAllowedMaxDynamicTableSize sets the upper bound that the encoded 151 // stream (via dynamic table size updates) may set the maximum size 152 // to. 153 func (d *Decoder) SetAllowedMaxDynamicTableSize(v uint32) { 154 d.dynTab.allowedMaxSize = v 155 } 156 157 type dynamicTable struct { 158 // http://http2.github.io/http2-spec/compression.html#rfc.section.2.3.2 159 table headerFieldTable 160 size uint32 // in bytes 161 maxSize uint32 // current maxSize 162 allowedMaxSize uint32 // maxSize may go up to this, inclusive 163 } 164 165 func (dt *dynamicTable) setMaxSize(v uint32) { 166 dt.maxSize = v 167 dt.evict() 168 } 169 170 func (dt *dynamicTable) add(f HeaderField) { 171 dt.table.addEntry(f) 172 dt.size += f.Size() 173 dt.evict() 174 } 175 176 // If we're too big, evict old stuff. 177 func (dt *dynamicTable) evict() { 178 var n int 179 for dt.size > dt.maxSize && n < dt.table.len() { 180 dt.size -= dt.table.ents[n].Size() 181 n++ 182 } 183 dt.table.evictOldest(n) 184 } 185 186 func (d *Decoder) maxTableIndex() int { 187 // This should never overflow. RFC 7540 Section 6.5.2 limits the size of 188 // the dynamic table to 2^32 bytes, where each entry will occupy more than 189 // one byte. Further, the staticTable has a fixed, small length. 190 return d.dynTab.table.len() + staticTable.len() 191 } 192 193 func (d *Decoder) at(i uint64) (hf HeaderField, ok bool) { 194 // See Section 2.3.3. 195 if i == 0 { 196 return 197 } 198 if i <= uint64(staticTable.len()) { 199 return staticTable.ents[i-1], true 200 } 201 if i > uint64(d.maxTableIndex()) { 202 return 203 } 204 // In the dynamic table, newer entries have lower indices. 205 // However, dt.ents[0] is the oldest entry. Hence, dt.ents is 206 // the reversed dynamic table. 207 dt := d.dynTab.table 208 return dt.ents[dt.len()-(int(i)-staticTable.len())], true 209 } 210 211 // Decode decodes an entire block. 212 // 213 // TODO: remove this method and make it incremental later? This is 214 // easier for debugging now. 215 func (d *Decoder) DecodeFull(p []byte) ([]HeaderField, error) { 216 var hf []HeaderField 217 saveFunc := d.emit 218 defer func() { d.emit = saveFunc }() 219 d.emit = func(f HeaderField) { hf = append(hf, f) } 220 if _, err := d.Write(p); err != nil { 221 return nil, err 222 } 223 if err := d.Close(); err != nil { 224 return nil, err 225 } 226 return hf, nil 227 } 228 229 func (d *Decoder) Close() error { 230 if d.saveBuf.Len() > 0 { 231 d.saveBuf.Reset() 232 return DecodingError{errors.New("truncated headers")} 233 } 234 return nil 235 } 236 237 func (d *Decoder) Write(p []byte) (n int, err error) { 238 if len(p) == 0 { 239 // Prevent state machine CPU attacks (making us redo 240 // work up to the point of finding out we don't have 241 // enough data) 242 return 243 } 244 // Only copy the data if we have to. Optimistically assume 245 // that p will contain a complete header block. 246 if d.saveBuf.Len() == 0 { 247 d.buf = p 248 } else { 249 d.saveBuf.Write(p) 250 d.buf = d.saveBuf.Bytes() 251 d.saveBuf.Reset() 252 } 253 254 for len(d.buf) > 0 { 255 err = d.parseHeaderFieldRepr() 256 if err == errNeedMore { 257 // Extra paranoia, making sure saveBuf won't 258 // get too large. All the varint and string 259 // reading code earlier should already catch 260 // overlong things and return ErrStringLength, 261 // but keep this as a last resort. 262 const varIntOverhead = 8 // conservative 263 if d.maxStrLen != 0 && int64(len(d.buf)) > 2*(int64(d.maxStrLen)+varIntOverhead) { 264 return 0, ErrStringLength 265 } 266 d.saveBuf.Write(d.buf) 267 return len(p), nil 268 } 269 if err != nil { 270 break 271 } 272 } 273 return len(p), err 274 } 275 276 // errNeedMore is an internal sentinel error value that means the 277 // buffer is truncated and we need to read more data before we can 278 // continue parsing. 279 var errNeedMore = errors.New("need more data") 280 281 type indexType int 282 283 const ( 284 indexedTrue indexType = iota 285 indexedFalse 286 indexedNever 287 ) 288 289 func (v indexType) indexed() bool { return v == indexedTrue } 290 func (v indexType) sensitive() bool { return v == indexedNever } 291 292 // returns errNeedMore if there isn't enough data available. 293 // any other error is fatal. 294 // consumes d.buf iff it returns nil. 295 // precondition: must be called with len(d.buf) > 0 296 func (d *Decoder) parseHeaderFieldRepr() error { 297 b := d.buf[0] 298 switch { 299 case b&128 != 0: 300 // Indexed representation. 301 // High bit set? 302 // http://http2.github.io/http2-spec/compression.html#rfc.section.6.1 303 return d.parseFieldIndexed() 304 case b&192 == 64: 305 // 6.2.1 Literal Header Field with Incremental Indexing 306 // 0b10xxxxxx: top two bits are 10 307 // http://http2.github.io/http2-spec/compression.html#rfc.section.6.2.1 308 return d.parseFieldLiteral(6, indexedTrue) 309 case b&240 == 0: 310 // 6.2.2 Literal Header Field without Indexing 311 // 0b0000xxxx: top four bits are 0000 312 // http://http2.github.io/http2-spec/compression.html#rfc.section.6.2.2 313 return d.parseFieldLiteral(4, indexedFalse) 314 case b&240 == 16: 315 // 6.2.3 Literal Header Field never Indexed 316 // 0b0001xxxx: top four bits are 0001 317 // http://http2.github.io/http2-spec/compression.html#rfc.section.6.2.3 318 return d.parseFieldLiteral(4, indexedNever) 319 case b&224 == 32: 320 // 6.3 Dynamic Table Size Update 321 // Top three bits are '001'. 322 // http://http2.github.io/http2-spec/compression.html#rfc.section.6.3 323 return d.parseDynamicTableSizeUpdate() 324 } 325 326 return DecodingError{errors.New("invalid encoding")} 327 } 328 329 // (same invariants and behavior as parseHeaderFieldRepr) 330 func (d *Decoder) parseFieldIndexed() error { 331 buf := d.buf 332 idx, buf, err := readVarInt(7, buf) 333 if err != nil { 334 return err 335 } 336 hf, ok := d.at(idx) 337 if !ok { 338 return DecodingError{InvalidIndexError(idx)} 339 } 340 d.buf = buf 341 return d.callEmit(HeaderField{Name: hf.Name, Value: hf.Value}) 342 } 343 344 // (same invariants and behavior as parseHeaderFieldRepr) 345 func (d *Decoder) parseFieldLiteral(n uint8, it indexType) error { 346 buf := d.buf 347 nameIdx, buf, err := readVarInt(n, buf) 348 if err != nil { 349 return err 350 } 351 352 var hf HeaderField 353 wantStr := d.emitEnabled || it.indexed() 354 if nameIdx > 0 { 355 ihf, ok := d.at(nameIdx) 356 if !ok { 357 return DecodingError{InvalidIndexError(nameIdx)} 358 } 359 hf.Name = ihf.Name 360 } else { 361 hf.Name, buf, err = d.readString(buf, wantStr) 362 if err != nil { 363 return err 364 } 365 } 366 hf.Value, buf, err = d.readString(buf, wantStr) 367 if err != nil { 368 return err 369 } 370 d.buf = buf 371 if it.indexed() { 372 d.dynTab.add(hf) 373 } 374 hf.Sensitive = it.sensitive() 375 return d.callEmit(hf) 376 } 377 378 func (d *Decoder) callEmit(hf HeaderField) error { 379 if d.maxStrLen != 0 { 380 if len(hf.Name) > d.maxStrLen || len(hf.Value) > d.maxStrLen { 381 return ErrStringLength 382 } 383 } 384 if d.emitEnabled { 385 d.emit(hf) 386 } 387 return nil 388 } 389 390 // (same invariants and behavior as parseHeaderFieldRepr) 391 func (d *Decoder) parseDynamicTableSizeUpdate() error { 392 // RFC 7541, sec 4.2: This dynamic table size update MUST occur at the 393 // beginning of the first header block following the change to the dynamic table size. 394 if d.dynTab.size > 0 { 395 return DecodingError{errors.New("dynamic table size update MUST occur at the beginning of a header block")} 396 } 397 398 buf := d.buf 399 size, buf, err := readVarInt(5, buf) 400 if err != nil { 401 return err 402 } 403 if size > uint64(d.dynTab.allowedMaxSize) { 404 return DecodingError{errors.New("dynamic table size update too large")} 405 } 406 d.dynTab.setMaxSize(uint32(size)) 407 d.buf = buf 408 return nil 409 } 410 411 var errVarintOverflow = DecodingError{errors.New("varint integer overflow")} 412 413 // readVarInt reads an unsigned variable length integer off the 414 // beginning of p. n is the parameter as described in 415 // http://http2.github.io/http2-spec/compression.html#rfc.section.5.1. 416 // 417 // n must always be between 1 and 8. 418 // 419 // The returned remain buffer is either a smaller suffix of p, or err != nil. 420 // The error is errNeedMore if p doesn't contain a complete integer. 421 func readVarInt(n byte, p []byte) (i uint64, remain []byte, err error) { 422 if n < 1 || n > 8 { 423 panic("bad n") 424 } 425 if len(p) == 0 { 426 return 0, p, errNeedMore 427 } 428 i = uint64(p[0]) 429 if n < 8 { 430 i &= (1 << uint64(n)) - 1 431 } 432 if i < (1<<uint64(n))-1 { 433 return i, p[1:], nil 434 } 435 436 origP := p 437 p = p[1:] 438 var m uint64 439 for len(p) > 0 { 440 b := p[0] 441 p = p[1:] 442 i += uint64(b&127) << m 443 if b&128 == 0 { 444 return i, p, nil 445 } 446 m += 7 447 if m >= 63 { // TODO: proper overflow check. making this up. 448 return 0, origP, errVarintOverflow 449 } 450 } 451 return 0, origP, errNeedMore 452 } 453 454 // readString decodes an hpack string from p. 455 // 456 // wantStr is whether s will be used. If false, decompression and 457 // []byte->string garbage are skipped if s will be ignored 458 // anyway. This does mean that huffman decoding errors for non-indexed 459 // strings past the MAX_HEADER_LIST_SIZE are ignored, but the server 460 // is returning an error anyway, and because they're not indexed, the error 461 // won't affect the decoding state. 462 func (d *Decoder) readString(p []byte, wantStr bool) (s string, remain []byte, err error) { 463 if len(p) == 0 { 464 return "", p, errNeedMore 465 } 466 isHuff := p[0]&128 != 0 467 strLen, p, err := readVarInt(7, p) 468 if err != nil { 469 return "", p, err 470 } 471 if d.maxStrLen != 0 && strLen > uint64(d.maxStrLen) { 472 return "", nil, ErrStringLength 473 } 474 if uint64(len(p)) < strLen { 475 return "", p, errNeedMore 476 } 477 if !isHuff { 478 if wantStr { 479 s = string(p[:strLen]) 480 } 481 return s, p[strLen:], nil 482 } 483 484 if wantStr { 485 buf := bufPool.Get().(*bytes.Buffer) 486 buf.Reset() // don't trust others 487 defer bufPool.Put(buf) 488 if err := huffmanDecode(buf, d.maxStrLen, p[:strLen]); err != nil { 489 buf.Reset() 490 return "", nil, err 491 } 492 s = buf.String() 493 buf.Reset() // be nice to GC 494 } 495 return s, p[strLen:], nil 496 } 497