wimlib.net Git - wimlib/blob - src/lzx_decompress.c

   1 /*
   2  * lzx_decompress.c
   3  *
   4  * A decompressor for the LZX compression format, as used in WIM files.
   5  */
   6
   7 /*
   8  * Copyright (C) 2012, 2013, 2014, 2015 Eric Biggers
   9  *
  10  * This file is free software; you can redistribute it and/or modify it under
  11  * the terms of the GNU Lesser General Public License as published by the Free
  12  * Software Foundation; either version 3 of the License, or (at your option) any
  13  * later version.
  14  *
  15  * This file is distributed in the hope that it will be useful, but WITHOUT
  16  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
  17  * FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
  18  * details.
  19  *
  20  * You should have received a copy of the GNU Lesser General Public License
  21  * along with this file; if not, see http://www.gnu.org/licenses/.
  22  */
  23
  24 /*
  25  * LZX is an LZ77 and Huffman-code based compression format that has many
  26  * similarities to DEFLATE (the format used by zlib/gzip).  The compression
  27  * ratio is as good or better than DEFLATE.  See lzx_compress.c for a format
  28  * overview, and see https://en.wikipedia.org/wiki/LZX_(algorithm) for a
  29  * historical overview.  Here I make some pragmatic notes.
  30  *
  31  * The old specification for LZX is the document "Microsoft LZX Data Compression
  32  * Format" (1997).  It defines the LZX format as used in cabinet files.  Allowed
  33  * window sizes are 2^n where 15 <= n <= 21.  However, this document contains
  34  * several errors, so don't read too much into it...
  35  *
  36  * The new specification for LZX is the document "[MS-PATCH]: LZX DELTA
  37  * Compression and Decompression" (2014).  It defines the LZX format as used by
  38  * Microsoft's binary patcher.  It corrects several errors in the 1997 document
  39  * and extends the format in several ways --- namely, optional reference data,
  40  * up to 2^25 byte windows, and longer match lengths.
  41  *
  42  * WIM files use a more restricted form of LZX.  No LZX DELTA extensions are
  43  * present, the window is not "sliding", E8 preprocessing is done
  44  * unconditionally with a fixed file size, and the maximum window size is always
  45  * 2^15 bytes (equal to the size of each "chunk" in a compressed WIM resource).
  46  * This code is primarily intended to implement this form of LZX.  But although
  47  * not compatible with WIMGAPI, this code also supports maximum window sizes up
  48  * to 2^21 bytes.
  49  *
  50  * TODO: Add support for window sizes up to 2^25 bytes.
  51  */
  52
  53 #ifdef HAVE_CONFIG_H
  54 #  include "config.h"
  55 #endif
  56
  57 #include <string.h>
  58
  59 #include "wimlib/decompressor_ops.h"
  60 #include "wimlib/decompress_common.h"
  61 #include "wimlib/error.h"
  62 #include "wimlib/lzx_common.h"
  63 #include "wimlib/util.h"
  64
  65 /* These values are chosen for fast decompression.  */
  66 #define LZX_MAINCODE_TABLEBITS          11
  67 #define LZX_LENCODE_TABLEBITS           10
  68 #define LZX_PRECODE_TABLEBITS           6
  69 #define LZX_ALIGNEDCODE_TABLEBITS       7
  70
  71 #define LZX_READ_LENS_MAX_OVERRUN 50
  72
  73 /* Huffman decoding tables, and arrays that map symbols to codeword lengths.  */
  74 struct lzx_tables {
  75
  76         u16 maincode_decode_table[(1 << LZX_MAINCODE_TABLEBITS) +
  77                                         (LZX_MAINCODE_MAX_NUM_SYMBOLS * 2)]
  78                                         _aligned_attribute(DECODE_TABLE_ALIGNMENT);
  79         u8 maincode_lens[LZX_MAINCODE_MAX_NUM_SYMBOLS + LZX_READ_LENS_MAX_OVERRUN];
  80
  81
  82         u16 lencode_decode_table[(1 << LZX_LENCODE_TABLEBITS) +
  83                                         (LZX_LENCODE_NUM_SYMBOLS * 2)]
  84                                         _aligned_attribute(DECODE_TABLE_ALIGNMENT);
  85         u8 lencode_lens[LZX_LENCODE_NUM_SYMBOLS + LZX_READ_LENS_MAX_OVERRUN];
  86
  87
  88         u16 alignedcode_decode_table[(1 << LZX_ALIGNEDCODE_TABLEBITS) +
  89                                         (LZX_ALIGNEDCODE_NUM_SYMBOLS * 2)]
  90                                         _aligned_attribute(DECODE_TABLE_ALIGNMENT);
  91         u8 alignedcode_lens[LZX_ALIGNEDCODE_NUM_SYMBOLS];
  92 } _aligned_attribute(DECODE_TABLE_ALIGNMENT);
  93
  94 /* Least-recently used queue for match offsets.  */
  95 struct lzx_lru_queue {
  96         u32 R[LZX_NUM_RECENT_OFFSETS];
  97 };
  98
  99 static inline void
 100 lzx_lru_queue_init(struct lzx_lru_queue *queue)
 101 {
 102         for (unsigned i = 0; i < LZX_NUM_RECENT_OFFSETS; i++)
 103                 queue->R[i] = 1;
 104 }
 105
 106 /* The main LZX decompressor structure.
 107  *
 108  * Note: we keep track of most of the decompression state outside this
 109  * structure.  This structure only exists so that (1) we can store @window_order
 110  * and @num_main_syms for multiple calls to lzx_decompress(); and (2) so that we
 111  * don't have to allocate the large 'struct lzx_tables' on the stack.  */
 112 struct lzx_decompressor {
 113         unsigned window_order;
 114         unsigned num_main_syms;
 115         struct lzx_tables tables;
 116 };
 117
 118 /* Read a Huffman-encoded symbol using the precode.  */
 119 static inline u16
 120 read_huffsym_using_precode(struct input_bitstream *istream,
 121                            const u16 precode_decode_table[])
 122 {
 123         return read_huffsym(istream, precode_decode_table,
 124                             LZX_PRECODE_TABLEBITS, LZX_MAX_PRE_CODEWORD_LEN);
 125 }
 126
 127 /* Read a Huffman-encoded symbol using the main code.  */
 128 static inline u16
 129 read_huffsym_using_maincode(struct input_bitstream *istream,
 130                             const struct lzx_tables *tables)
 131 {
 132         return read_huffsym(istream, tables->maincode_decode_table,
 133                             LZX_MAINCODE_TABLEBITS, LZX_MAX_MAIN_CODEWORD_LEN);
 134 }
 135
 136 /* Read a Huffman-encoded symbol using the length code.  */
 137 static inline u16
 138 read_huffsym_using_lencode(struct input_bitstream *istream,
 139                            const struct lzx_tables *tables)
 140 {
 141         return read_huffsym(istream, tables->lencode_decode_table,
 142                             LZX_LENCODE_TABLEBITS, LZX_MAX_LEN_CODEWORD_LEN);
 143 }
 144
 145 /* Read a Huffman-encoded symbol using the aligned offset code.  */
 146 static inline u16
 147 read_huffsym_using_alignedcode(struct input_bitstream *istream,
 148                                const struct lzx_tables *tables)
 149 {
 150         return read_huffsym(istream, tables->alignedcode_decode_table,
 151                             LZX_ALIGNEDCODE_TABLEBITS, LZX_MAX_ALIGNED_CODEWORD_LEN);
 152 }
 153
 154 /*
 155  * Read the precode from the compressed input bitstream, then use it to decode
 156  * @num_lens codeword length values.
 157  *
 158  * @istream:
 159  *      The input bitstream.
 160  *
 161  * @lens:
 162  *      An array that contains the length values from the previous time the
 163  *      codeword lengths for this Huffman code were read, or all 0's if this is
 164  *      the first time.  This array must have at least (@num_lens +
 165  *      LZX_READ_LENS_MAX_OVERRUN) entries.
 166  *
 167  * @num_lens:
 168  *      Number of length values to decode.
 169  *
 170  * Returns 0 on success, or -1 if the data was invalid.
 171  */
 172 static int
 173 lzx_read_codeword_lens(struct input_bitstream *istream, u8 *lens, unsigned num_lens)
 174 {
 175         u16 precode_decode_table[(1 << LZX_PRECODE_TABLEBITS) +
 176                                         (LZX_PRECODE_NUM_SYMBOLS * 2)]
 177                                         _aligned_attribute(DECODE_TABLE_ALIGNMENT);
 178         u8 precode_lens[LZX_PRECODE_NUM_SYMBOLS];
 179         u8 *len_ptr = lens;
 180         u8 *lens_end = lens + num_lens;
 181         int ret;
 182
 183         /* Read the lengths of the precode codewords.  These are given
 184          * explicitly.  */
 185         for (int i = 0; i < LZX_PRECODE_NUM_SYMBOLS; i++) {
 186                 precode_lens[i] = bitstream_read_bits(istream,
 187                                                       LZX_PRECODE_ELEMENT_SIZE);
 188         }
 189
 190         /* Make the decoding table for the precode.  */
 191         ret = make_huffman_decode_table(precode_decode_table,
 192                                         LZX_PRECODE_NUM_SYMBOLS,
 193                                         LZX_PRECODE_TABLEBITS,
 194                                         precode_lens,
 195                                         LZX_MAX_PRE_CODEWORD_LEN);
 196         if (ret)
 197                 return ret;
 198
 199         /* Decode the codeword lengths.  */
 200         do {
 201                 unsigned presym;
 202                 u8 len;
 203
 204                 /* Read the next precode symbol.  */
 205                 presym = read_huffsym_using_precode(istream,
 206                                                     precode_decode_table);
 207                 if (presym < 17) {
 208                         /* Difference from old length  */
 209                         len = *len_ptr - presym;
 210                         if ((s8)len < 0)
 211                                 len += 17;
 212                         *len_ptr++ = len;
 213                 } else {
 214                         /* Special RLE values  */
 215
 216                         unsigned run_len;
 217
 218                         if (presym == 17) {
 219                                 /* Run of 0's  */
 220                                 run_len = 4 + bitstream_read_bits(istream, 4);
 221                                 len = 0;
 222                         } else if (presym == 18) {
 223                                 /* Longer run of 0's  */
 224                                 run_len = 20 + bitstream_read_bits(istream, 5);
 225                                 len = 0;
 226                         } else {
 227                                 /* Run of identical lengths  */
 228                                 run_len = 4 + bitstream_read_bits(istream, 1);
 229                                 presym = read_huffsym_using_precode(istream,
 230                                                                     precode_decode_table);
 231                                 if (unlikely(presym > 17))
 232                                         return -1;
 233                                 len = *len_ptr - presym;
 234                                 if ((s8)len < 0)
 235                                         len += 17;
 236                         }
 237
 238                         do {
 239                                 *len_ptr++ = len;
 240                         } while (--run_len);
 241                         /* Worst case overrun is when presym == 18,
 242                          * run_len == 20 + 31, and only 1 length was remaining.
 243                          * So LZX_READ_LENS_MAX_OVERRUN == 50.
 244                          *
 245                          * Overrun while reading the first half of maincode_lens
 246                          * can corrupt the previous values in the second half.
 247                          * This doesn't really matter because the resulting
 248                          * lengths will still be in range, and data that
 249                          * generates overruns is invalid anyway.  */
 250                 }
 251         } while (len_ptr < lens_end);
 252         return 0;
 253 }
 254
 255 /*
 256  * Read the header of an LZX block and save the block type and size in
 257  * *block_type_ret and *block_size_ret, respectively.
 258  *
 259  * If the block is compressed, also update the Huffman decode @tables with the
 260  * new Huffman codes.
 261  *
 262  * If the block is uncompressed, also update the match offset @queue with the
 263  * new match offsets.
 264  *
 265  * Return 0 on success, or -1 if the data was invalid.
 266  */
 267 static int
 268 lzx_read_block_header(struct input_bitstream *istream,
 269                       unsigned num_main_syms,
 270                       unsigned window_order,
 271                       int *block_type_ret,
 272                       u32 *block_size_ret,
 273                       struct lzx_tables *tables,
 274                       struct lzx_lru_queue *queue)
 275 {
 276         int block_type;
 277         u32 block_size;
 278         int ret;
 279
 280         bitstream_ensure_bits(istream, 4);
 281
 282         /* The first three bits tell us what kind of block it is, and should be
 283          * one of the LZX_BLOCKTYPE_* values.  */
 284         block_type = bitstream_pop_bits(istream, 3);
 285
 286         /* Read the block size.  This mirrors the behavior of
 287          * lzx_write_compressed_block() in lzx_compress.c; see that for more
 288          * details.  */
 289         if (bitstream_pop_bits(istream, 1)) {
 290                 block_size = LZX_DEFAULT_BLOCK_SIZE;
 291         } else {
 292                 u32 tmp;
 293                 block_size = 0;
 294
 295                 tmp = bitstream_read_bits(istream, 8);
 296                 block_size |= tmp;
 297                 tmp = bitstream_read_bits(istream, 8);
 298                 block_size <<= 8;
 299                 block_size |= tmp;
 300
 301                 if (window_order >= 16) {
 302                         tmp = bitstream_read_bits(istream, 8);
 303                         block_size <<= 8;
 304                         block_size |= tmp;
 305                 }
 306         }
 307
 308         switch (block_type) {
 309
 310         case LZX_BLOCKTYPE_ALIGNED:
 311
 312                 /* Read the aligned offset code and prepare its decode table.
 313                  */
 314
 315                 for (int i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++) {
 316                         tables->alignedcode_lens[i] =
 317                                 bitstream_read_bits(istream,
 318                                                     LZX_ALIGNEDCODE_ELEMENT_SIZE);
 319                 }
 320
 321                 ret = make_huffman_decode_table(tables->alignedcode_decode_table,
 322                                                 LZX_ALIGNEDCODE_NUM_SYMBOLS,
 323                                                 LZX_ALIGNEDCODE_TABLEBITS,
 324                                                 tables->alignedcode_lens,
 325                                                 LZX_MAX_ALIGNED_CODEWORD_LEN);
 326                 if (ret)
 327                         return ret;
 328
 329                 /* Fall though, since the rest of the header for aligned offset
 330                  * blocks is the same as that for verbatim blocks.  */
 331
 332         case LZX_BLOCKTYPE_VERBATIM:
 333
 334                 /* Read the main code and prepare its decode table.
 335                  *
 336                  * Note that the codeword lengths in the main code are encoded
 337                  * in two parts: one part for literal symbols, and one part for
 338                  * match symbols.  */
 339
 340                 ret = lzx_read_codeword_lens(istream, tables->maincode_lens,
 341                                              LZX_NUM_CHARS);
 342                 if (ret)
 343                         return ret;
 344
 345                 ret = lzx_read_codeword_lens(istream,
 346                                              tables->maincode_lens + LZX_NUM_CHARS,
 347                                              num_main_syms - LZX_NUM_CHARS);
 348                 if (ret)
 349                         return ret;
 350
 351                 ret = make_huffman_decode_table(tables->maincode_decode_table,
 352                                                 num_main_syms,
 353                                                 LZX_MAINCODE_TABLEBITS,
 354                                                 tables->maincode_lens,
 355                                                 LZX_MAX_MAIN_CODEWORD_LEN);
 356                 if (ret)
 357                         return ret;
 358
 359                 /* Read the length code and prepare its decode table.  */
 360
 361                 ret = lzx_read_codeword_lens(istream, tables->lencode_lens,
 362                                              LZX_LENCODE_NUM_SYMBOLS);
 363                 if (ret)
 364                         return ret;
 365
 366                 ret = make_huffman_decode_table(tables->lencode_decode_table,
 367                                                 LZX_LENCODE_NUM_SYMBOLS,
 368                                                 LZX_LENCODE_TABLEBITS,
 369                                                 tables->lencode_lens,
 370                                                 LZX_MAX_LEN_CODEWORD_LEN);
 371                 if (ret)
 372                         return ret;
 373
 374                 break;
 375
 376         case LZX_BLOCKTYPE_UNCOMPRESSED:
 377
 378                 /* Before reading the three LRU match offsets from the
 379                  * uncompressed block header, the stream must be aligned on a
 380                  * 16-bit boundary.  But, unexpectedly, if the stream is
 381                  * *already* aligned, the correct thing to do is to throw away
 382                  * the next 16 bits.  */
 383
 384                 bitstream_ensure_bits(istream, 1);
 385                 bitstream_align(istream);
 386                 queue->R[0] = bitstream_read_u32(istream);
 387                 queue->R[1] = bitstream_read_u32(istream);
 388                 queue->R[2] = bitstream_read_u32(istream);
 389
 390                 /* Offsets of 0 are invalid.  */
 391                 if (queue->R[0] == 0 || queue->R[1] == 0 || queue->R[2] == 0)
 392                         return -1;
 393                 break;
 394
 395         default:
 396                 /* Unrecognized block type.  */
 397                 return -1;
 398         }
 399
 400         *block_type_ret = block_type;
 401         *block_size_ret = block_size;
 402         return 0;
 403 }
 404
 405 /*
 406  * Decompress an LZX-compressed block of data.
 407  *
 408  * @block_type:
 409  *      The type of the block (LZX_BLOCKTYPE_VERBATIM or LZX_BLOCKTYPE_ALIGNED).
 410  *
 411  * @block_size:
 412  *      The size of the block, in bytes.
 413  *
 414  * @window:
 415  *      Pointer to the beginning of the decompression window.
 416  *
 417  * @window_pos:
 418  *      The position in the window at which the block starts.
 419  *
 420  * @tables:
 421  *      The Huffman decoding tables for the block.
 422  *
 423  * @queue:
 424  *      The least-recently-used queue for match offsets.
 425  *
 426  * @istream:
 427  *      The input bitstream, positioned at the start of the block data.
 428  *
 429  * Returns 0 on success, or -1 if the data was invalid.
 430  */
 431 static int
 432 lzx_decompress_block(int block_type, u32 block_size,
 433                      u8 *window, u32 window_pos,
 434                      const struct lzx_tables *tables,
 435                      struct lzx_lru_queue *queue,
 436                      struct input_bitstream *istream)
 437 {
 438         u8 *window_ptr = &window[window_pos];
 439         u8 *window_end = window_ptr + block_size;
 440         unsigned mainsym;
 441         u32 match_len;
 442         unsigned offset_slot;
 443         u32 match_offset;
 444         unsigned num_extra_bits;
 445         unsigned ones_if_aligned = 0U - (block_type == LZX_BLOCKTYPE_ALIGNED);
 446
 447         while (window_ptr != window_end) {
 448
 449                 mainsym = read_huffsym_using_maincode(istream, tables);
 450                 if (mainsym < LZX_NUM_CHARS) {
 451                         /* Literal  */
 452                         *window_ptr++ = mainsym;
 453                         continue;
 454                 }
 455
 456                 /* Match  */
 457
 458                 /* Decode the length header and offset slot.  */
 459                 mainsym -= LZX_NUM_CHARS;
 460                 match_len = mainsym % LZX_NUM_LEN_HEADERS;
 461                 offset_slot = mainsym / LZX_NUM_LEN_HEADERS;
 462
 463                 /* If needed, read a length symbol to decode the full length. */
 464                 if (match_len == LZX_NUM_PRIMARY_LENS)
 465                         match_len += read_huffsym_using_lencode(istream, tables);
 466                 match_len += LZX_MIN_MATCH_LEN;
 467
 468                 if (offset_slot < LZX_NUM_RECENT_OFFSETS) {
 469                         /* Repeat offset  */
 470
 471                         /* Note: This isn't a real LRU queue, since using the R2
 472                          * offset doesn't bump the R1 offset down to R2.  This
 473                          * quirk allows all 3 recent offsets to be handled by
 474                          * the same code.  (For R0, the swap is a no-op.)  */
 475                         match_offset = queue->R[offset_slot];
 476                         queue->R[offset_slot] = queue->R[0];
 477                         queue->R[0] = match_offset;
 478                 } else {
 479                         /* Explicit offset  */
 480
 481                         /* Look up the number of extra bits that need to be read
 482                          * to decode offsets with this offset slot.  */
 483                         num_extra_bits = lzx_extra_offset_bits[offset_slot];
 484
 485                         /* Start with the offset slot base value.  */
 486                         match_offset = lzx_offset_slot_base[offset_slot];
 487
 488                         /* In aligned offset blocks, the low-order 3 bits of
 489                          * each offset are encoded using the aligned offset
 490                          * code.  Otherwise, all the extra bits are literal.  */
 491
 492                         if ((num_extra_bits & ones_if_aligned) >= LZX_NUM_ALIGNED_OFFSET_BITS) {
 493                                 match_offset +=
 494                                         bitstream_read_bits(istream,
 495                                                             num_extra_bits -
 496                                                                 LZX_NUM_ALIGNED_OFFSET_BITS)
 497                                                         << LZX_NUM_ALIGNED_OFFSET_BITS;
 498                                 match_offset += read_huffsym_using_alignedcode(istream, tables);
 499                         } else {
 500                                 match_offset += bitstream_read_bits(istream, num_extra_bits);
 501                         }
 502
 503                         /* Adjust the offset.  */
 504                         match_offset -= LZX_OFFSET_ADJUSTMENT;
 505
 506                         /* Update the match offset LRU queue.  */
 507                         BUILD_BUG_ON(LZX_NUM_RECENT_OFFSETS != 3);
 508                         queue->R[2] = queue->R[1];
 509                         queue->R[1] = queue->R[0];
 510                         queue->R[0] = match_offset;
 511                 }
 512
 513                 /* Validate the match, then copy it to the current position.  */
 514
 515                 if (unlikely(match_len > window_end - window_ptr))
 516                         return -1;
 517
 518                 if (unlikely(match_offset > window_ptr - window))
 519                         return -1;
 520
 521                 lz_copy(window_ptr, match_len, match_offset, window_end,
 522                         LZX_MIN_MATCH_LEN);
 523
 524                 window_ptr += match_len;
 525         }
 526         return 0;
 527 }
 528
 529 static int
 530 lzx_decompress(const void *compressed_data, size_t compressed_size,
 531                void *uncompressed_data, size_t uncompressed_size,
 532                void *_dec)
 533 {
 534         struct lzx_decompressor *dec = _dec;
 535         struct input_bitstream istream;
 536         struct lzx_lru_queue queue;
 537         u32 window_pos;
 538         int block_type;
 539         u32 block_size;
 540         bool may_have_e8_byte;
 541         int ret;
 542
 543         init_input_bitstream(&istream, compressed_data, compressed_size);
 544
 545         /* Initialize the recent offsets queue.  */
 546         lzx_lru_queue_init(&queue);
 547
 548         /* Codeword lengths begin as all 0's for delta encoding purposes.  */
 549         memset(dec->tables.maincode_lens, 0, dec->num_main_syms);
 550         memset(dec->tables.lencode_lens, 0, LZX_LENCODE_NUM_SYMBOLS);
 551
 552         /* Set this to true if there may be 0xe8 bytes in the uncompressed data.
 553          */
 554         may_have_e8_byte = false;
 555
 556         /* The compressed data will consist of one or more blocks.  The
 557          * following loop decompresses one block, and it runs until there all
 558          * the compressed data has been decompressed, so there are no more
 559          * blocks.  */
 560
 561         for (window_pos = 0;
 562              window_pos < uncompressed_size;
 563              window_pos += block_size)
 564         {
 565                 ret = lzx_read_block_header(&istream, dec->num_main_syms,
 566                                             dec->window_order, &block_type,
 567                                             &block_size, &dec->tables, &queue);
 568                 if (ret)
 569                         return ret;
 570
 571                 if (block_size > uncompressed_size - window_pos)
 572                         return -1;
 573
 574                 if (block_type != LZX_BLOCKTYPE_UNCOMPRESSED) {
 575
 576                         /* Compressed block.  */
 577
 578                         ret = lzx_decompress_block(block_type,
 579                                                    block_size,
 580                                                    uncompressed_data,
 581                                                    window_pos,
 582                                                    &dec->tables,
 583                                                    &queue,
 584                                                    &istream);
 585                         if (ret)
 586                                 return ret;
 587
 588                         /* If the first 0xe8 byte was in this block, it must
 589                          * have been encoded as a literal using mainsym 0xe8. */
 590                         if (dec->tables.maincode_lens[0xe8] != 0)
 591                                 may_have_e8_byte = true;
 592                 } else {
 593
 594                         /* Uncompressed block.  */
 595                         const u8 *p;
 596
 597                         p = bitstream_read_bytes(&istream, block_size);
 598                         if (!p)
 599                                 return -1;
 600
 601                         memcpy(&((u8*)uncompressed_data)[window_pos], p, block_size);
 602
 603                         /* Re-align the bitstream if an odd number of bytes was
 604                          * read.  */
 605                         if (block_size & 1)
 606                                 bitstream_read_byte(&istream);
 607
 608                         may_have_e8_byte = true;
 609                 }
 610         }
 611
 612         /* Postprocess the data unless it cannot possibly contain 0xe8 bytes  */
 613         if (may_have_e8_byte)
 614                 lzx_undo_e8_preprocessing(uncompressed_data, uncompressed_size);
 615
 616         return 0;
 617 }
 618
 619 static void
 620 lzx_free_decompressor(void *_dec)
 621 {
 622         struct lzx_decompressor *dec = _dec;
 623
 624         ALIGNED_FREE(dec);
 625 }
 626
 627 static int
 628 lzx_create_decompressor(size_t max_block_size, void **dec_ret)
 629 {
 630         struct lzx_decompressor *dec;
 631         unsigned window_order;
 632
 633         window_order = lzx_get_window_order(max_block_size);
 634         if (window_order == 0)
 635                 return WIMLIB_ERR_INVALID_PARAM;
 636
 637         /* The aligned allocation is needed to ensure that the lzx_tables are
 638          * aligned properly.  */
 639         dec = ALIGNED_MALLOC(sizeof(struct lzx_decompressor),
 640                              DECODE_TABLE_ALIGNMENT);
 641         if (!dec)
 642                 return WIMLIB_ERR_NOMEM;
 643
 644         dec->window_order = window_order;
 645         dec->num_main_syms = lzx_get_num_main_syms(window_order);
 646
 647         *dec_ret = dec;
 648         return 0;
 649 }
 650
 651 const struct decompressor_ops lzx_decompressor_ops = {
 652         .create_decompressor = lzx_create_decompressor,
 653         .decompress          = lzx_decompress,
 654         .free_decompressor   = lzx_free_decompressor,
 655 };