1 // Copyright 2013 The Chromium Authors. All rights reserved. 2 // Use of this source code is governed by a BSD-style license that can be 3 // found in the LICENSE file. 4 5 #ifndef NET_TOOLS_BALSA_BALSA_HEADERS_H_ 6 #define NET_TOOLS_BALSA_BALSA_HEADERS_H_ 7 8 #include <algorithm> 9 #include <iosfwd> 10 #include <iterator> 11 #include <string> 12 #include <utility> 13 #include <vector> 14 15 #include "base/logging.h" 16 #include "base/port.h" 17 #include "base/strings/string_piece.h" 18 #include "net/tools/balsa/balsa_enums.h" 19 #include "net/tools/balsa/string_piece_utils.h" 20 21 namespace net { 22 23 // WARNING: 24 // Note that -no- char* returned by any function in this 25 // file is null-terminated. 26 27 // This class exists to service the specific needs of BalsaHeaders. 28 // 29 // Functional goals: 30 // 1) provide a backing-store for all of the StringPieces that BalsaHeaders 31 // returns. Every StringPiece returned from BalsaHeaders should remain 32 // valid until the BalsaHeader's object is cleared, or the header-line is 33 // erased. 34 // 2) provide a backing-store for BalsaFrame, which requires contiguous memory 35 // for its fast-path parsing functions. Note that the cost of copying is 36 // less than the cost of requiring the parser to do slow-path parsing, as 37 // it would have to check for bounds every byte, instead of every 16 bytes. 38 // 39 // This class is optimized for the case where headers are stored in one of two 40 // buffers. It doesn't make a lot of effort to densely pack memory-- in fact, 41 // it -may- be somewhat memory inefficient. This possible inefficiency allows a 42 // certain simplicity of implementation and speed which makes it worthwhile. 43 // If, in the future, better memory density is required, it should be possible 44 // to reuse the abstraction presented by this object to achieve those goals. 45 // 46 // In the most common use-case, this memory inefficiency should be relatively 47 // small. 48 // 49 // Alternate implementations of BalsaBuffer may include: 50 // - vector of strings, one per header line (similar to HTTPHeaders) 51 // - densely packed strings: 52 // - keep a sorted array/map of free-space linked lists or numbers. 53 // - use the entry that most closely first your needs. 54 // - at this point, perhaps just use a vector of strings, and let 55 // the allocator do the right thing. 56 // 57 class BalsaBuffer { 58 public: 59 static const size_t kDefaultBlocksize = 4096; 60 // We have two friends here. These exist as friends as we 61 // want to allow access to the constructors for the test 62 // class and the Balsa* classes. We put this into the 63 // header file as we want this class to be inlined into the 64 // BalsaHeaders implementation, yet be testable. 65 friend class BalsaBufferTestSpouse; 66 friend class BalsaHeaders; 67 friend class BalsaBufferTest; 68 69 // The BufferBlock is a structure used internally by the 70 // BalsaBuffer class to store the base buffer pointers to 71 // each block, as well as the important metadata for buffer 72 // sizes and bytes free. 73 struct BufferBlock { 74 public: 75 char* buffer; 76 size_t buffer_size; 77 size_t bytes_free; 78 bytes_usedBufferBlock79 size_t bytes_used() const { 80 return buffer_size - bytes_free; 81 } start_of_unused_bytesBufferBlock82 char* start_of_unused_bytes() const { 83 return buffer + bytes_used(); 84 } 85 BufferBlockBufferBlock86 BufferBlock() : buffer(NULL), buffer_size(0), bytes_free(0) {} ~BufferBlockBufferBlock87 ~BufferBlock() {} 88 BufferBlockBufferBlock89 BufferBlock(char* buf, size_t size, size_t free) : 90 buffer(buf), buffer_size(size), bytes_free(free) {} 91 // Yes we want this to be copyable (it gets stuck into vectors). 92 // For this reason, we don't use scoped ptrs, etc. here-- it 93 // is more efficient to manage this memory externally to this 94 // object. 95 }; 96 97 typedef std::vector<BufferBlock> Blocks; 98 99 ~BalsaBuffer(); 100 101 // Returns the total amount of memory used by the buffer blocks. 102 size_t GetTotalBufferBlockSize() const; 103 GetPtr(Blocks::size_type block_idx)104 const char* GetPtr(Blocks::size_type block_idx) const { 105 DCHECK_LT(block_idx, blocks_.size()) 106 << block_idx << ", " << blocks_.size(); 107 return blocks_[block_idx].buffer; 108 } 109 GetPtr(Blocks::size_type block_idx)110 char* GetPtr(Blocks::size_type block_idx) { 111 DCHECK_LT(block_idx, blocks_.size()) 112 << block_idx << ", " << blocks_.size(); 113 return blocks_[block_idx].buffer; 114 } 115 116 // This function is different from Write(), as it ensures that the data 117 // stored via subsequent calls to this function are all contiguous (and in 118 // the order in which these writes happened). This is essentially the same 119 // as a string append. 120 // 121 // You may call this function at any time between object 122 // construction/Clear(), and the calling of the 123 // NoMoreWriteToContiguousBuffer() function. 124 // 125 // You must not call this function after the NoMoreWriteToContiguousBuffer() 126 // function is called, unless a Clear() has been called since. 127 // If you do, the program will abort(). 128 // 129 // This condition is placed upon this code so that calls to Write() can 130 // append to the buffer in the first block safely, and without invaliding 131 // the StringPiece which it returns. 132 // 133 // This function's main intended user is the BalsaFrame class, which, 134 // for reasons of efficiency, requires that the buffer from which it parses 135 // the headers be contiguous. 136 // 137 void WriteToContiguousBuffer(const base::StringPiece& sp); 138 NoMoreWriteToContiguousBuffer()139 void NoMoreWriteToContiguousBuffer() { 140 can_write_to_contiguous_buffer_ = false; 141 } 142 143 // Takes a StringPiece and writes it to "permanent" storage, then returns a 144 // StringPiece which points to that data. If block_idx != NULL, it will be 145 // assigned the index of the block into which the data was stored. 146 // Note that the 'permanent' storage in which it stores data may be in 147 // the first block IFF the NoMoreWriteToContiguousBuffer function has 148 // been called since the last Clear/Construction. 149 base::StringPiece Write(const base::StringPiece& sp, 150 Blocks::size_type* block_buffer_idx); 151 152 // Reserves "permanent" storage of the size indicated. Returns a pointer to 153 // the beginning of that storage, and assigns the index of the block used to 154 // block_buffer_idx. This function uses the first block IFF the 155 // NoMoreWriteToContiguousBuffer function has been called since the last 156 // Clear/Construction. 157 char* Reserve(size_t size, Blocks::size_type* block_buffer_idx); 158 159 void Clear(); 160 161 void Swap(BalsaBuffer* b); 162 163 void CopyFrom(const BalsaBuffer& b); 164 StartOfFirstBlock()165 const char* StartOfFirstBlock() const { 166 return blocks_[0].buffer; 167 } 168 EndOfFirstBlock()169 const char* EndOfFirstBlock() const { 170 return blocks_[0].buffer + blocks_[0].bytes_used(); 171 } 172 can_write_to_contiguous_buffer()173 bool can_write_to_contiguous_buffer() const { 174 return can_write_to_contiguous_buffer_; 175 } blocksize()176 size_t blocksize() const { return blocksize_; } num_blocks()177 Blocks::size_type num_blocks() const { return blocks_.size(); } buffer_size(size_t idx)178 size_t buffer_size(size_t idx) const { return blocks_[idx].buffer_size; } bytes_used(size_t idx)179 size_t bytes_used(size_t idx) const { return blocks_[idx].bytes_used(); } 180 181 protected: 182 BalsaBuffer(); 183 184 explicit BalsaBuffer(size_t blocksize); 185 186 BufferBlock AllocBlock(); 187 188 BufferBlock AllocCustomBlock(size_t blocksize); 189 190 BufferBlock CopyBlock(const BufferBlock& b); 191 192 // Cleans up the object. 193 // The block at start_idx, and all subsequent blocks 194 // will be cleared and have associated memory deleted. 195 void CleanupBlocksStartingFrom(Blocks::size_type start_idx); 196 197 // A container of BufferBlocks 198 Blocks blocks_; 199 200 // The default allocation size for a block. 201 // In general, blocksize_ bytes will be allocated for 202 // each buffer. 203 size_t blocksize_; 204 205 // If set to true, then the first block cannot be used for Write() calls as 206 // the WriteToContiguous... function will modify the base pointer for this 207 // block, and the Write() calls need to be sure that the base pointer will 208 // not be changing in order to provide the user with StringPieces which 209 // continue to be valid. 210 bool can_write_to_contiguous_buffer_; 211 }; 212 213 //////////////////////////////////////////////////////////////////////////////// 214 215 // All of the functions in the BalsaHeaders class use string pieces, by either 216 // using the StringPiece class, or giving an explicit size and char* (as these 217 // are the native representation for these string pieces). 218 // This is done for several reasons. 219 // 1) This minimizes copying/allocation/deallocation as compared to using 220 // string parameters 221 // 2) This reduces the number of strlen() calls done (as the length of any 222 // string passed in is relatively likely to be known at compile time, and for 223 // those strings passed back we obviate the need for a strlen() to determine 224 // the size of new storage allocations if a new allocation is required. 225 // 3) This class attempts to store all of its data in two linear buffers in 226 // order to enhance the speed of parsing and writing out to a buffer. As a 227 // result, many string pieces are -not- terminated by '\0', and are not 228 // c-strings. Since this is the case, we must delineate the length of the 229 // string explicitly via a length. 230 // 231 // WARNING: The side effect of using StringPiece is that if the underlying 232 // buffer changes (due to modifying the headers) the StringPieces which point 233 // to the data which was modified, may now contain "garbage", and should not 234 // be dereferenced. 235 // For example, If you fetch some component of the first-line, (request or 236 // response), and then you modify the first line, the StringPieces you 237 // originally received from the original first-line may no longer be valid). 238 // 239 // StringPieces pointing to pieces of header lines which have not been 240 // erased() or modified should be valid until the object is cleared or 241 // destroyed. 242 243 class BalsaHeaders { 244 public: 245 struct HeaderLineDescription { HeaderLineDescriptionHeaderLineDescription246 HeaderLineDescription(size_t first_character_index, 247 size_t key_end_index, 248 size_t value_begin_index, 249 size_t last_character_index, 250 size_t buffer_base_index) : 251 first_char_idx(first_character_index), 252 key_end_idx(key_end_index), 253 value_begin_idx(value_begin_index), 254 last_char_idx(last_character_index), 255 buffer_base_idx(buffer_base_index), 256 skip(false) {} 257 HeaderLineDescriptionHeaderLineDescription258 HeaderLineDescription() : 259 first_char_idx(0), 260 key_end_idx(0), 261 value_begin_idx(0), 262 last_char_idx(0), 263 buffer_base_idx(0), 264 skip(false) {} 265 266 size_t first_char_idx; 267 size_t key_end_idx; 268 size_t value_begin_idx; 269 size_t last_char_idx; 270 BalsaBuffer::Blocks::size_type buffer_base_idx; 271 bool skip; 272 }; 273 274 typedef std::vector<base::StringPiece> HeaderTokenList; 275 friend bool ParseHTTPFirstLine(const char* begin, 276 const char* end, 277 bool is_request, 278 size_t max_request_uri_length, 279 BalsaHeaders* headers, 280 BalsaFrameEnums::ErrorCode* error_code); 281 282 protected: 283 typedef std::vector<HeaderLineDescription> HeaderLines; 284 285 // Why these base classes (iterator_base, reverse_iterator_base)? Well, if 286 // we do want to export both iterator and const_iterator types (currently we 287 // only have const_iterator), then this is useful to avoid code duplication. 288 // Additionally, having this base class makes comparisons of iterators of 289 // different types (they're different types to ensure that operator= and 290 // constructors do not work in the places where they're expected to not work) 291 // work properly. There could be as many as 4 iterator types, all based on 292 // the same data as iterator_base... so it makes sense to simply have some 293 // base classes. 294 295 class iterator_base { 296 public: 297 friend class BalsaHeaders; 298 friend class reverse_iterator_base; 299 typedef std::pair<base::StringPiece, base::StringPiece> StringPiecePair; 300 typedef StringPiecePair value_type; 301 typedef value_type& reference; 302 typedef value_type* pointer; 303 304 typedef std::forward_iterator_tag iterator_category; 305 typedef ptrdiff_t difference_type; 306 307 typedef iterator_base self; 308 309 // default constructor. 310 iterator_base(); 311 312 // copy constructor. 313 iterator_base(const iterator_base& it); 314 315 reference operator*() const { 316 return Lookup(idx_); 317 } 318 319 pointer operator->() const { 320 return &(this->operator*()); 321 } 322 323 bool operator==(const self& it) const { 324 return idx_ == it.idx_; 325 } 326 327 bool operator<(const self& it) const { 328 return idx_ < it.idx_; 329 } 330 331 bool operator<=(const self& it) const { 332 return idx_ <= it.idx_; 333 } 334 335 bool operator!=(const self& it) const { 336 return !(*this == it); 337 } 338 339 bool operator>(const self& it) const { 340 return it < *this; 341 } 342 343 bool operator>=(const self& it) const { 344 return it <= *this; 345 } 346 347 // This mainly exists so that we can have interesting output for 348 // unittesting. The EXPECT_EQ, EXPECT_NE functions require that 349 // operator<< work for the classes it sees. It would be better if there 350 // was an additional traits-like system for the gUnit output... but oh 351 // well. 352 std::ostream& operator<<(std::ostream& os) const; 353 354 protected: 355 iterator_base(const BalsaHeaders* headers, HeaderLines::size_type index); 356 increment()357 void increment() { 358 const HeaderLines& header_lines = headers_->header_lines_; 359 const HeaderLines::size_type header_lines_size = header_lines.size(); 360 const HeaderLines::size_type original_idx = idx_; 361 do { 362 ++idx_; 363 } while (idx_ < header_lines_size && header_lines[idx_].skip == true); 364 // The condition below exists so that ++(end() - 1) == end(), even 365 // if there are only 'skip == true' elements between the end() iterator 366 // and the end of the vector of HeaderLineDescriptions. 367 // TODO(fenix): refactor this list so that we don't have to do 368 // linear scanning through skipped headers (and this condition is 369 // then unnecessary) 370 if (idx_ == header_lines_size) { 371 idx_ = original_idx + 1; 372 } 373 } 374 decrement()375 void decrement() { 376 const HeaderLines& header_lines = headers_->header_lines_; 377 const HeaderLines::size_type header_lines_size = header_lines.size(); 378 const HeaderLines::size_type original_idx = idx_; 379 do { 380 --idx_; 381 } while (idx_ < header_lines_size && header_lines[idx_].skip == true); 382 // The condition below exists so that --(rbegin() + 1) == rbegin(), even 383 // if there are only 'skip == true' elements between the rbegin() iterator 384 // and the beginning of the vector of HeaderLineDescriptions. 385 // TODO(fenix): refactor this list so that we don't have to do 386 // linear scanning through skipped headers (and this condition is 387 // then unnecessary) 388 if (idx_ > header_lines_size) { 389 idx_ = original_idx - 1; 390 } 391 } 392 Lookup(HeaderLines::size_type index)393 reference Lookup(HeaderLines::size_type index) const { 394 DCHECK_LT(index, headers_->header_lines_.size()); 395 const HeaderLineDescription& line = headers_->header_lines_[index]; 396 const char* stream_begin = headers_->GetPtr(line.buffer_base_idx); 397 value_ = value_type( 398 base::StringPiece(stream_begin + line.first_char_idx, 399 line.key_end_idx - line.first_char_idx), 400 base::StringPiece(stream_begin + line.value_begin_idx, 401 line.last_char_idx - line.value_begin_idx)); 402 DCHECK_GE(line.key_end_idx, line.first_char_idx); 403 DCHECK_GE(line.last_char_idx, line.value_begin_idx); 404 return value_; 405 } 406 407 const BalsaHeaders* headers_; 408 HeaderLines::size_type idx_; 409 mutable StringPiecePair value_; 410 }; 411 412 class reverse_iterator_base : public iterator_base { 413 public: 414 typedef reverse_iterator_base self; 415 typedef iterator_base::reference reference; 416 typedef iterator_base::pointer pointer; 417 using iterator_base::headers_; 418 using iterator_base::idx_; 419 reverse_iterator_base()420 reverse_iterator_base() : iterator_base() {} 421 422 // This constructor is no explicit purposely. reverse_iterator_base(const iterator_base & it)423 reverse_iterator_base(const iterator_base& it) : // NOLINT 424 iterator_base(it) { 425 } 426 427 self& operator=(const iterator_base& it) { 428 idx_ = it.idx_; 429 headers_ = it.headers_; 430 return *this; 431 } 432 433 self& operator=(const reverse_iterator_base& it) { 434 idx_ = it.idx_; 435 headers_ = it.headers_; 436 return *this; 437 } 438 439 reference operator*() const { 440 return Lookup(idx_ - 1); 441 } 442 443 pointer operator->() const { 444 return &(this->operator*()); 445 } 446 reverse_iterator_base(const reverse_iterator_base & it)447 reverse_iterator_base(const reverse_iterator_base& it) : 448 iterator_base(it) { } 449 450 protected: increment()451 void increment() { 452 --idx_; 453 iterator_base::decrement(); 454 ++idx_; 455 } 456 decrement()457 void decrement() { 458 ++idx_; 459 iterator_base::increment(); 460 --idx_; 461 } 462 reverse_iterator_base(const BalsaHeaders * headers,HeaderLines::size_type index)463 reverse_iterator_base(const BalsaHeaders* headers, 464 HeaderLines::size_type index) : 465 iterator_base(headers, index) {} 466 }; 467 468 public: 469 class const_header_lines_iterator : public iterator_base { 470 friend class BalsaHeaders; 471 public: 472 typedef const_header_lines_iterator self; const_header_lines_iterator()473 const_header_lines_iterator() : iterator_base() {} 474 const_header_lines_iterator(const const_header_lines_iterator & it)475 const_header_lines_iterator(const const_header_lines_iterator& it) : 476 iterator_base(it.headers_, it.idx_) {} 477 478 self& operator++() { 479 iterator_base::increment(); 480 return *this; 481 } 482 483 self& operator--() { 484 iterator_base::decrement(); 485 return *this; 486 } 487 protected: const_header_lines_iterator(const BalsaHeaders * headers,HeaderLines::size_type index)488 const_header_lines_iterator(const BalsaHeaders* headers, 489 HeaderLines::size_type index) : 490 iterator_base(headers, index) {} 491 }; 492 493 class const_reverse_header_lines_iterator : public reverse_iterator_base { 494 public: 495 typedef const_reverse_header_lines_iterator self; const_reverse_header_lines_iterator()496 const_reverse_header_lines_iterator() : reverse_iterator_base() {} 497 const_reverse_header_lines_iterator(const const_header_lines_iterator & it)498 const_reverse_header_lines_iterator( 499 const const_header_lines_iterator& it) : 500 reverse_iterator_base(it.headers_, it.idx_) {} 501 const_reverse_header_lines_iterator(const const_reverse_header_lines_iterator & it)502 const_reverse_header_lines_iterator( 503 const const_reverse_header_lines_iterator& it) : 504 reverse_iterator_base(it.headers_, it.idx_) {} 505 base()506 const_header_lines_iterator base() { 507 return const_header_lines_iterator(headers_, idx_); 508 } 509 510 self& operator++() { 511 reverse_iterator_base::increment(); 512 return *this; 513 } 514 515 self& operator--() { 516 reverse_iterator_base::decrement(); 517 return *this; 518 } 519 protected: const_reverse_header_lines_iterator(const BalsaHeaders * headers,HeaderLines::size_type index)520 const_reverse_header_lines_iterator(const BalsaHeaders* headers, 521 HeaderLines::size_type index) : 522 reverse_iterator_base(headers, index) {} 523 524 friend class BalsaHeaders; 525 }; 526 527 // An iterator that only stops at lines with a particular key. 528 // See also GetIteratorForKey. 529 // 530 // Check against header_lines_key_end() to determine when iteration is 531 // finished. header_lines_end() will also work. 532 class const_header_lines_key_iterator : public iterator_base { 533 friend class BalsaHeaders; 534 public: 535 typedef const_header_lines_key_iterator self; 536 const_header_lines_key_iterator(const const_header_lines_key_iterator&); 537 538 self& operator++() { 539 do { 540 iterator_base::increment(); 541 } while (!AtEnd() && 542 !StringPieceUtils::EqualIgnoreCase(key_, (**this).first)); 543 return *this; 544 } 545 546 void operator++(int ignore) { 547 ++(*this); 548 } 549 550 // Only forward-iteration makes sense, so no operator-- defined. 551 552 private: 553 const_header_lines_key_iterator(const BalsaHeaders* headers, 554 HeaderLines::size_type index, 555 const base::StringPiece& key); 556 557 // Should only be used for creating an end iterator. 558 const_header_lines_key_iterator(const BalsaHeaders* headers, 559 HeaderLines::size_type index); 560 AtEnd()561 bool AtEnd() const { 562 return *this >= headers_->header_lines_end(); 563 } 564 565 base::StringPiece key_; 566 }; 567 568 // TODO(fenix): Revisit the amount of bytes initially allocated to the second 569 // block of the balsa_buffer_. It may make sense to pre-allocate some amount 570 // (roughly the amount we'd append in new headers such as X-User-Ip, etc.) 571 BalsaHeaders(); 572 ~BalsaHeaders(); 573 header_lines_begin()574 const_header_lines_iterator header_lines_begin() { 575 return HeaderLinesBeginHelper<const_header_lines_iterator>(); 576 } 577 header_lines_begin()578 const_header_lines_iterator header_lines_begin() const { 579 return HeaderLinesBeginHelper<const_header_lines_iterator>(); 580 } 581 header_lines_end()582 const_header_lines_iterator header_lines_end() { 583 return HeaderLinesEndHelper<const_header_lines_iterator>(); 584 } 585 header_lines_end()586 const_header_lines_iterator header_lines_end() const { 587 return HeaderLinesEndHelper<const_header_lines_iterator>(); 588 } 589 header_lines_rbegin()590 const_reverse_header_lines_iterator header_lines_rbegin() { 591 return const_reverse_header_lines_iterator(header_lines_end()); 592 } 593 header_lines_rbegin()594 const_reverse_header_lines_iterator header_lines_rbegin() const { 595 return const_reverse_header_lines_iterator(header_lines_end()); 596 } 597 header_lines_rend()598 const_reverse_header_lines_iterator header_lines_rend() { 599 return const_reverse_header_lines_iterator(header_lines_begin()); 600 } 601 header_lines_rend()602 const_reverse_header_lines_iterator header_lines_rend() const { 603 return const_reverse_header_lines_iterator(header_lines_begin()); 604 } 605 header_lines_key_end()606 const_header_lines_key_iterator header_lines_key_end() const { 607 return HeaderLinesEndHelper<const_header_lines_key_iterator>(); 608 } 609 erase(const const_header_lines_iterator & it)610 void erase(const const_header_lines_iterator& it) { 611 DCHECK_EQ(it.headers_, this); 612 DCHECK_LT(it.idx_, header_lines_.size()); 613 DCHECK_GE(it.idx_, 0u); 614 header_lines_[it.idx_].skip = true; 615 } 616 617 void Clear(); 618 619 void Swap(BalsaHeaders* other); 620 621 void CopyFrom(const BalsaHeaders& other); 622 623 void HackHeader(const base::StringPiece& key, const base::StringPiece& value); 624 625 // Same as AppendToHeader, except that it will attempt to preserve 626 // header ordering. 627 // Note that this will always append to an existing header, if available, 628 // without moving the header around, or collapsing multiple header lines 629 // with the same key together. For this reason, it only 'attempts' to 630 // preserve header ordering. 631 // TODO(fenix): remove this function and rename all occurances 632 // of it in the code to AppendToHeader when the condition above 633 // has been satisified. 634 void HackAppendToHeader(const base::StringPiece& key, 635 const base::StringPiece& value); 636 637 // Replaces header entries with key 'key' if they exist, or appends 638 // a new header if none exist. See 'AppendHeader' below for additional 639 // comments about ContentLength and TransferEncoding headers. Note that this 640 // will allocate new storage every time that it is called. 641 // TODO(fenix): modify this function to reuse existing storage 642 // if it is available. 643 void ReplaceOrAppendHeader(const base::StringPiece& key, 644 const base::StringPiece& value); 645 646 // Append a new header entry to the header object. Clients who wish to append 647 // Content-Length header should use SetContentLength() method instead of 648 // adding the content length header using AppendHeader (manually adding the 649 // content length header will not update the content_length_ and 650 // content_length_status_ values). 651 // Similarly, clients who wish to add or remove the transfer encoding header 652 // in order to apply or remove chunked encoding should use SetChunkEncoding() 653 // instead. 654 void AppendHeader(const base::StringPiece& key, 655 const base::StringPiece& value); 656 657 // Appends ',value' to an existing header named 'key'. If no header with the 658 // correct key exists, it will call AppendHeader(key, value). Calling this 659 // function on a key which exists several times in the headers will produce 660 // unpredictable results. 661 void AppendToHeader(const base::StringPiece& key, 662 const base::StringPiece& value); 663 664 // Prepends 'value,' to an existing header named 'key'. If no header with the 665 // correct key exists, it will call AppendHeader(key, value). Calling this 666 // function on a key which exists several times in the headers will produce 667 // unpredictable results. 668 void PrependToHeader(const base::StringPiece& key, 669 const base::StringPiece& value); 670 671 const base::StringPiece GetHeader(const base::StringPiece& key) const; 672 673 // Iterates over all currently valid header lines, appending their 674 // values into the vector 'out', in top-to-bottom order. 675 // Header-lines which have been erased are not currently valid, and 676 // will not have their values appended. Empty values will be 677 // represented as empty string. If 'key' doesn't exist in the headers at 678 // all, out will not be changed. We do not clear the vector out 679 // before adding new entries. If there are header lines with matching 680 // key but empty value then they are also added to the vector out. 681 // (Basically empty values are not treated in any special manner). 682 // 683 // Example: 684 // Input header: 685 // "GET / HTTP/1.0\r\n" 686 // "key1: v1\r\n" 687 // "key1: \r\n" 688 // "key1:\r\n" 689 // "key1: v1\r\n" 690 // "key1:v2\r\n" 691 // 692 // vector out is initially: ["foo"] 693 // vector out after GetAllOfHeader("key1", &out) is: 694 // ["foo", "v1", "", "", "v2", "v1", "v2"] 695 696 void GetAllOfHeader(const base::StringPiece& key, 697 std::vector<base::StringPiece>* out) const; 698 699 // Joins all values for key into a comma-separated string in out. 700 // More efficient than calling JoinStrings on result of GetAllOfHeader if 701 // you don't need the intermediate vector<StringPiece>. 702 void GetAllOfHeaderAsString(const base::StringPiece& key, 703 std::string* out) const; 704 705 // Returns true if RFC 2616 Section 14 indicates that header can 706 // have multiple values. 707 static bool IsMultivaluedHeader(const base::StringPiece& header); 708 709 // Determine if a given header is present. HasHeader(const base::StringPiece & key)710 inline bool HasHeader(const base::StringPiece& key) const { 711 return (GetConstHeaderLinesIterator(key, header_lines_.begin()) != 712 header_lines_.end()); 713 } 714 715 // Returns true iff any header 'key' exists with non-empty value. 716 bool HasNonEmptyHeader(const base::StringPiece& key) const; 717 718 const_header_lines_iterator GetHeaderPosition( 719 const base::StringPiece& key) const; 720 721 // Returns a forward-only iterator that only stops at lines matching key. 722 // String backing 'key' must remain valid for lifetime of iterator. 723 // 724 // Check returned iterator against header_lines_key_end() to determine when 725 // iteration is finished. 726 const_header_lines_key_iterator GetIteratorForKey( 727 const base::StringPiece& key) const; 728 729 void RemoveAllOfHeader(const base::StringPiece& key); 730 731 // Removes all headers starting with 'key' [case insensitive] 732 void RemoveAllHeadersWithPrefix(const base::StringPiece& key); 733 734 // Returns the lower bound of memory used by this header object, including 735 // all internal buffers and data structure. Some of the memory used cannot be 736 // directly measure. For example, memory used for bookkeeping by standard 737 // containers. 738 size_t GetMemoryUsedLowerBound() const; 739 740 // Returns the upper bound on the required buffer space to fully write out 741 // the header object (this include the first line, all header lines, and the 742 // final CRLF that marks the ending of the header). 743 size_t GetSizeForWriteBuffer() const; 744 745 // The following WriteHeader* methods are template member functions that 746 // place one requirement on the Buffer class: it must implement a Write 747 // method that takes a pointer and a length. The buffer passed in is not 748 // required to be stretchable. For non-stretchable buffers, the user must 749 // call GetSizeForWriteBuffer() to find out the upper bound on the output 750 // buffer space required to make sure that the entire header is serialized. 751 // BalsaHeaders will not check that there is adequate space in the buffer 752 // object during the write. 753 754 // Writes the entire header and the final CRLF that marks the end of the HTTP 755 // header section to the buffer. After this method returns, no more header 756 // data should be written to the buffer. 757 template <typename Buffer> WriteHeaderAndEndingToBuffer(Buffer * buffer)758 void WriteHeaderAndEndingToBuffer(Buffer* buffer) const { 759 WriteToBuffer(buffer); 760 WriteHeaderEndingToBuffer(buffer); 761 } 762 763 // Writes the final CRLF to the buffer to terminate the HTTP header section. 764 // After this method returns, no more header data should be written to the 765 // buffer. 766 template <typename Buffer> WriteHeaderEndingToBuffer(Buffer * buffer)767 static void WriteHeaderEndingToBuffer(Buffer* buffer) { 768 buffer->Write("\r\n", 2); 769 } 770 771 // Writes the entire header to the buffer without the CRLF that terminates 772 // the HTTP header. This lets users append additional header lines using 773 // WriteHeaderLineToBuffer and then terminate the header with 774 // WriteHeaderEndingToBuffer as the header is serialized to the 775 // buffer, without having to first copy the header. 776 template <typename Buffer> WriteToBuffer(Buffer * buffer)777 void WriteToBuffer(Buffer* buffer) const { 778 // write the first line. 779 const size_t firstline_len = whitespace_4_idx_ - non_whitespace_1_idx_; 780 const char* stream_begin = GetPtr(firstline_buffer_base_idx_); 781 buffer->Write(stream_begin + non_whitespace_1_idx_, firstline_len); 782 buffer->Write("\r\n", 2); 783 const HeaderLines::size_type end = header_lines_.size(); 784 for (HeaderLines::size_type i = 0; i < end; ++i) { 785 const HeaderLineDescription& line = header_lines_[i]; 786 if (line.skip) { 787 continue; 788 } 789 const char* line_ptr = GetPtr(line.buffer_base_idx); 790 WriteHeaderLineToBuffer( 791 buffer, 792 base::StringPiece(line_ptr + line.first_char_idx, 793 line.key_end_idx - line.first_char_idx), 794 base::StringPiece(line_ptr + line.value_begin_idx, 795 line.last_char_idx - line.value_begin_idx)); 796 } 797 } 798 799 // Takes a header line in the form of a key/value pair and append it to the 800 // buffer. This function should be called after WriteToBuffer to 801 // append additional header lines to the header without copying the header. 802 // When the user is done with appending to the buffer, 803 // WriteHeaderEndingToBuffer must be used to terminate the HTTP 804 // header in the buffer. This method is a no-op if key is empty. 805 template <typename Buffer> WriteHeaderLineToBuffer(Buffer * buffer,const base::StringPiece & key,const base::StringPiece & value)806 static void WriteHeaderLineToBuffer(Buffer* buffer, 807 const base::StringPiece& key, 808 const base::StringPiece& value) { 809 // if the key is empty, we don't want to write the rest because it 810 // will not be a well-formed header line. 811 if (!key.empty()) { 812 buffer->Write(key.data(), key.size()); 813 buffer->Write(": ", 2); 814 buffer->Write(value.data(), value.size()); 815 buffer->Write("\r\n", 2); 816 } 817 } 818 819 // Dump the textural representation of the header object to a string, which 820 // is suitable for writing out to logs. All CRLF will be printed out as \n. 821 // This function can be called on a header object in any state. Raw header 822 // data will be printed out if the header object is not completely parsed, 823 // e.g., when there was an error in the middle of parsing. 824 // The header content is appended to the string; the original content is not 825 // cleared. 826 void DumpToString(std::string* str) const; 827 first_line()828 const base::StringPiece first_line() const { 829 DCHECK_GE(whitespace_4_idx_, non_whitespace_1_idx_); 830 return base::StringPiece(BeginningOfFirstLine() + non_whitespace_1_idx_, 831 whitespace_4_idx_ - non_whitespace_1_idx_); 832 } 833 834 // Returns the parsed value of the response code if it has been parsed. 835 // Guaranteed to return 0 when unparsed (though it is a much better idea to 836 // verify that the BalsaFrame had no errors while parsing). 837 // This may return response codes which are outside the normal bounds of 838 // HTTP response codes-- it is up to the user of this class to ensure that 839 // the response code is one which is interpretable. parsed_response_code()840 size_t parsed_response_code() const { return parsed_response_code_; } 841 request_method()842 const base::StringPiece request_method() const { 843 DCHECK_GE(whitespace_2_idx_, non_whitespace_1_idx_); 844 return base::StringPiece(BeginningOfFirstLine() + non_whitespace_1_idx_, 845 whitespace_2_idx_ - non_whitespace_1_idx_); 846 } 847 response_version()848 const base::StringPiece response_version() const { 849 // Note: There is no difference between request_method() and 850 // response_version(). They both could be called 851 // GetFirstTokenFromFirstline()... but that wouldn't be anywhere near as 852 // descriptive. 853 return request_method(); 854 } 855 request_uri()856 const base::StringPiece request_uri() const { 857 DCHECK_GE(whitespace_3_idx_, non_whitespace_2_idx_); 858 return base::StringPiece(BeginningOfFirstLine() + non_whitespace_2_idx_, 859 whitespace_3_idx_ - non_whitespace_2_idx_); 860 } 861 response_code()862 const base::StringPiece response_code() const { 863 // Note: There is no difference between request_uri() and response_code(). 864 // They both could be called GetSecondtTokenFromFirstline(), but, as noted 865 // in an earlier comment, that wouldn't be as descriptive. 866 return request_uri(); 867 } 868 request_version()869 const base::StringPiece request_version() const { 870 DCHECK_GE(whitespace_4_idx_, non_whitespace_3_idx_); 871 return base::StringPiece(BeginningOfFirstLine() + non_whitespace_3_idx_, 872 whitespace_4_idx_ - non_whitespace_3_idx_); 873 } 874 response_reason_phrase()875 const base::StringPiece response_reason_phrase() const { 876 // Note: There is no difference between request_version() and 877 // response_reason_phrase(). They both could be called 878 // GetThirdTokenFromFirstline(), but, as noted in an earlier comment, that 879 // wouldn't be as descriptive. 880 return request_version(); 881 } 882 883 // Note that SetFirstLine will not update the internal indices for the 884 // various bits of the first-line (and may set them all to zero). 885 // If you'd like to use the accessors for the various bits of the firstline, 886 // then you should use the Set* functions, or SetFirstlineFromStringPieces, 887 // below, instead. 888 // 889 void SetFirstlineFromStringPieces(const base::StringPiece& firstline_a, 890 const base::StringPiece& firstline_b, 891 const base::StringPiece& firstline_c); 892 SetRequestFirstlineFromStringPieces(const base::StringPiece & method,const base::StringPiece & uri,const base::StringPiece & version)893 void SetRequestFirstlineFromStringPieces(const base::StringPiece& method, 894 const base::StringPiece& uri, 895 const base::StringPiece& version) { 896 SetFirstlineFromStringPieces(method, uri, version); 897 } 898 SetResponseFirstlineFromStringPieces(const base::StringPiece & version,const base::StringPiece & code,const base::StringPiece & reason_phrase)899 void SetResponseFirstlineFromStringPieces( 900 const base::StringPiece& version, 901 const base::StringPiece& code, 902 const base::StringPiece& reason_phrase) { 903 SetFirstlineFromStringPieces(version, code, reason_phrase); 904 } 905 906 // These functions are exactly the same, except that their names are 907 // different. This is done so that the code using this class is more 908 // expressive. 909 void SetRequestMethod(const base::StringPiece& method); 910 void SetResponseVersion(const base::StringPiece& version); 911 912 void SetRequestUri(const base::StringPiece& uri); 913 void SetResponseCode(const base::StringPiece& code); set_parsed_response_code(size_t parsed_response_code)914 void set_parsed_response_code(size_t parsed_response_code) { 915 parsed_response_code_ = parsed_response_code; 916 } 917 void SetParsedResponseCodeAndUpdateFirstline(size_t parsed_response_code); 918 919 // These functions are exactly the same, except that their names are 920 // different. This is done so that the code using this class is more 921 // expressive. 922 void SetRequestVersion(const base::StringPiece& version); 923 void SetResponseReasonPhrase(const base::StringPiece& reason_phrase); 924 925 // The biggest problem with SetFirstLine is that we don't want to use a 926 // separate buffer for it. The second biggest problem with it is that the 927 // first biggest problem requires that we store offsets into a buffer instead 928 // of pointers into a buffer. Cuteness aside, SetFirstLine doesn't parse 929 // the individual fields of the firstline, and so accessors to those fields 930 // will not work properly after calling SetFirstLine. If you want those 931 // accessors to work, use the Set* functions above this one. 932 // SetFirstLine is stuff useful, however, if all you care about is correct 933 // serialization with the rest of the header object. 934 void SetFirstLine(const base::StringPiece& line); 935 936 // Simple accessors to some of the internal state transfer_encoding_is_chunked()937 bool transfer_encoding_is_chunked() const { 938 return transfer_encoding_is_chunked_; 939 } 940 ResponseCodeImpliesNoBody(size_t code)941 static bool ResponseCodeImpliesNoBody(size_t code) { 942 // From HTTP spec section 6.1.1 all 1xx responses must not have a body, 943 // as well as 204 No Content and 304 Not Modified. 944 return ((code >= 100) && (code <= 199)) || (code == 204) || (code == 304); 945 } 946 947 // Note: never check this for requests. Nothing bad will happen if you do, 948 // but spec does not allow requests framed by connection close. 949 // TODO(vitaliyl): refactor. is_framed_by_connection_close()950 bool is_framed_by_connection_close() const { 951 // We declare that response is framed by connection close if it has no 952 // content-length, no transfer encoding, and is allowed to have a body by 953 // the HTTP spec. 954 // parsed_response_code_ is 0 for requests, so ResponseCodeImpliesNoBody 955 // will return false. 956 return (content_length_status_ == BalsaHeadersEnums::NO_CONTENT_LENGTH) && 957 !transfer_encoding_is_chunked_ && 958 !ResponseCodeImpliesNoBody(parsed_response_code_); 959 } 960 content_length()961 size_t content_length() const { return content_length_; } content_length_status()962 BalsaHeadersEnums::ContentLengthStatus content_length_status() const { 963 return content_length_status_; 964 } 965 966 // SetContentLength and SetChunkEncoding modifies the header object to use 967 // content-length and transfer-encoding headers in a consistent manner. They 968 // set all internal flags and status so client can get a consistent view from 969 // various accessors. 970 void SetContentLength(size_t length); 971 void SetChunkEncoding(bool chunk_encode); 972 973 protected: 974 friend class BalsaFrame; 975 friend class SpdyFrame; 976 friend class HTTPMessage; 977 friend class BalsaHeadersTokenUtils; 978 BeginningOfFirstLine()979 const char* BeginningOfFirstLine() const { 980 return GetPtr(firstline_buffer_base_idx_); 981 } 982 GetPtr(BalsaBuffer::Blocks::size_type block_idx)983 char* GetPtr(BalsaBuffer::Blocks::size_type block_idx) { 984 return balsa_buffer_.GetPtr(block_idx); 985 } 986 GetPtr(BalsaBuffer::Blocks::size_type block_idx)987 const char* GetPtr(BalsaBuffer::Blocks::size_type block_idx) const { 988 return balsa_buffer_.GetPtr(block_idx); 989 } 990 WriteFromFramer(const char * ptr,size_t size)991 void WriteFromFramer(const char* ptr, size_t size) { 992 balsa_buffer_.WriteToContiguousBuffer(base::StringPiece(ptr, size)); 993 } 994 DoneWritingFromFramer()995 void DoneWritingFromFramer() { 996 balsa_buffer_.NoMoreWriteToContiguousBuffer(); 997 } 998 OriginalHeaderStreamBegin()999 const char* OriginalHeaderStreamBegin() const { 1000 return balsa_buffer_.StartOfFirstBlock(); 1001 } 1002 OriginalHeaderStreamEnd()1003 const char* OriginalHeaderStreamEnd() const { 1004 return balsa_buffer_.EndOfFirstBlock(); 1005 } 1006 GetReadableBytesFromHeaderStream()1007 size_t GetReadableBytesFromHeaderStream() const { 1008 return OriginalHeaderStreamEnd() - OriginalHeaderStreamBegin(); 1009 } 1010 GetReadablePtrFromHeaderStream(const char ** p,size_t * s)1011 void GetReadablePtrFromHeaderStream(const char** p, size_t* s) { 1012 *p = OriginalHeaderStreamBegin(); 1013 *s = GetReadableBytesFromHeaderStream(); 1014 } 1015 1016 base::StringPiece GetValueFromHeaderLineDescription( 1017 const HeaderLineDescription& line) const; 1018 1019 void AddAndMakeDescription(const base::StringPiece& key, 1020 const base::StringPiece& value, 1021 HeaderLineDescription* d); 1022 1023 void AppendOrPrependAndMakeDescription(const base::StringPiece& key, 1024 const base::StringPiece& value, 1025 bool append, 1026 HeaderLineDescription* d); 1027 1028 // Removes all header lines with the given key starting at start. 1029 void RemoveAllOfHeaderStartingAt(const base::StringPiece& key, 1030 HeaderLines::iterator start); 1031 1032 // If the 'key' does not exist in the headers, calls 1033 // AppendHeader(key, value). Otherwise if append is true, appends ',value' 1034 // to the first existing header with key 'key'. If append is false, prepends 1035 // 'value,' to the first existing header with key 'key'. 1036 void AppendOrPrependToHeader(const base::StringPiece& key, 1037 const base::StringPiece& value, 1038 bool append); 1039 1040 HeaderLines::const_iterator GetConstHeaderLinesIterator( 1041 const base::StringPiece& key, 1042 HeaderLines::const_iterator start) const; 1043 1044 HeaderLines::iterator GetHeaderLinesIteratorNoSkip( 1045 const base::StringPiece& key, 1046 HeaderLines::iterator start); 1047 1048 HeaderLines::iterator GetHeaderLinesIterator( 1049 const base::StringPiece& key, 1050 HeaderLines::iterator start); 1051 1052 template <typename IteratorType> HeaderLinesBeginHelper()1053 const IteratorType HeaderLinesBeginHelper() const { 1054 if (header_lines_.empty()) { 1055 return IteratorType(this, 0); 1056 } 1057 const HeaderLines::size_type header_lines_size = header_lines_.size(); 1058 for (HeaderLines::size_type i = 0; i < header_lines_size; ++i) { 1059 if (header_lines_[i].skip == false) { 1060 return IteratorType(this, i); 1061 } 1062 } 1063 return IteratorType(this, 0); 1064 } 1065 1066 template <typename IteratorType> HeaderLinesEndHelper()1067 const IteratorType HeaderLinesEndHelper() const { 1068 if (header_lines_.empty()) { 1069 return IteratorType(this, 0); 1070 } 1071 const HeaderLines::size_type header_lines_size = header_lines_.size(); 1072 HeaderLines::size_type i = header_lines_size; 1073 do { 1074 --i; 1075 if (header_lines_[i].skip == false) { 1076 return IteratorType(this, i + 1); 1077 } 1078 } while (i != 0); 1079 return IteratorType(this, 0); 1080 } 1081 1082 // At the moment, this function will always return the original headers. 1083 // In the future, it may not do so after erasing header lines, modifying 1084 // header lines, or modifying the first line. 1085 // For this reason, it is strongly suggested that use of this function is 1086 // only acceptable for the purpose of debugging parse errors seen by the 1087 // BalsaFrame class. OriginalHeadersForDebugging()1088 base::StringPiece OriginalHeadersForDebugging() const { 1089 return base::StringPiece(OriginalHeaderStreamBegin(), 1090 OriginalHeaderStreamEnd() - OriginalHeaderStreamBegin()); 1091 } 1092 1093 BalsaBuffer balsa_buffer_; 1094 1095 size_t content_length_; 1096 BalsaHeadersEnums::ContentLengthStatus content_length_status_; 1097 size_t parsed_response_code_; 1098 // HTTP firstlines all have the following structure: 1099 // LWS NONWS LWS NONWS LWS NONWS NOTCRLF CRLF 1100 // [\t \r\n]+ [^\t ]+ [\t ]+ [^\t ]+ [\t ]+ [^\t ]+ [^\r\n]+ "\r\n" 1101 // ws1 nws1 ws2 nws2 ws3 nws3 ws4 1102 // | [-------) [-------) [----------------) 1103 // REQ: method request_uri version 1104 // RESP: version statuscode reason 1105 // 1106 // The first NONWS->LWS component we'll call firstline_a. 1107 // The second firstline_b, and the third firstline_c. 1108 // 1109 // firstline_a goes from nws1 to (but not including) ws2 1110 // firstline_b goes from nws2 to (but not including) ws3 1111 // firstline_c goes from nws3 to (but not including) ws4 1112 // 1113 // In the code: 1114 // ws1 == whitespace_1_idx_ 1115 // nws1 == non_whitespace_1_idx_ 1116 // ws2 == whitespace_2_idx_ 1117 // nws2 == non_whitespace_2_idx_ 1118 // ws3 == whitespace_3_idx_ 1119 // nws3 == non_whitespace_3_idx_ 1120 // ws4 == whitespace_4_idx_ 1121 BalsaBuffer::Blocks::size_type firstline_buffer_base_idx_; 1122 size_t whitespace_1_idx_; 1123 size_t non_whitespace_1_idx_; 1124 size_t whitespace_2_idx_; 1125 size_t non_whitespace_2_idx_; 1126 size_t whitespace_3_idx_; 1127 size_t non_whitespace_3_idx_; 1128 size_t whitespace_4_idx_; 1129 size_t end_of_firstline_idx_; 1130 1131 bool transfer_encoding_is_chunked_; 1132 1133 HeaderLines header_lines_; 1134 }; 1135 1136 } // namespace net 1137 1138 #endif // NET_TOOLS_BALSA_BALSA_HEADERS_H_ 1139