pd.if.org Git - zpackage/blob - lzma/lz/lz_decoder.c

   1 ///////////////////////////////////////////////////////////////////////////////
   2 //
   3 /// \file       lz_decoder.c
   4 /// \brief      LZ out window
   5 ///
   6 //  Authors:    Igor Pavlov
   7 //              Lasse Collin
   8 //
   9 //  This file has been put into the public domain.
  10 //  You can do whatever you want with this file.
  11 //
  12 ///////////////////////////////////////////////////////////////////////////////
  13
  14 // liblzma supports multiple LZ77-based filters. The LZ part is shared
  15 // between these filters. The LZ code takes care of dictionary handling
  16 // and passing the data between filters in the chain. The filter-specific
  17 // part decodes from the input buffer to the dictionary.
  18
  19
  20 #include "lz_decoder.h"
  21
  22
  23 typedef struct {
  24         /// Dictionary (history buffer)
  25         lzma_dict dict;
  26
  27         /// The actual LZ-based decoder e.g. LZMA
  28         lzma_lz_decoder lz;
  29
  30         /// Next filter in the chain, if any. Note that LZMA and LZMA2 are
  31         /// only allowed as the last filter, but the long-range filter in
  32         /// future can be in the middle of the chain.
  33         lzma_next_coder next;
  34
  35         /// True if the next filter in the chain has returned LZMA_STREAM_END.
  36         bool next_finished;
  37
  38         /// True if the LZ decoder (e.g. LZMA) has detected end of payload
  39         /// marker. This may become true before next_finished becomes true.
  40         bool this_finished;
  41
  42         /// Temporary buffer needed when the LZ-based filter is not the last
  43         /// filter in the chain. The output of the next filter is first
  44         /// decoded into buffer[], which is then used as input for the actual
  45         /// LZ-based decoder.
  46         struct {
  47                 size_t pos;
  48                 size_t size;
  49                 uint8_t buffer[LZMA_BUFFER_SIZE];
  50         } temp;
  51 } lzma_coder;
  52
  53
  54 static void
  55 lz_decoder_reset(lzma_coder *coder)
  56 {
  57         coder->dict.pos = 0;
  58         coder->dict.full = 0;
  59         coder->dict.buf[coder->dict.size - 1] = '\0';
  60         coder->dict.need_reset = false;
  61         return;
  62 }
  63
  64
  65 static lzma_ret
  66 decode_buffer(lzma_coder *coder,
  67                 const uint8_t *restrict in, size_t *restrict in_pos,
  68                 size_t in_size, uint8_t *restrict out,
  69                 size_t *restrict out_pos, size_t out_size)
  70 {
  71         while (true) {
  72                 // Wrap the dictionary if needed.
  73                 if (coder->dict.pos == coder->dict.size)
  74                         coder->dict.pos = 0;
  75
  76                 // Store the current dictionary position. It is needed to know
  77                 // where to start copying to the out[] buffer.
  78                 const size_t dict_start = coder->dict.pos;
  79
  80                 // Calculate how much we allow coder->lz.code() to decode.
  81                 // It must not decode past the end of the dictionary
  82                 // buffer, and we don't want it to decode more than is
  83                 // actually needed to fill the out[] buffer.
  84                 coder->dict.limit = coder->dict.pos
  85                                 + my_min(out_size - *out_pos,
  86                                         coder->dict.size - coder->dict.pos);
  87
  88                 // Call the coder->lz.code() to do the actual decoding.
  89                 const lzma_ret ret = coder->lz.code(
  90                                 coder->lz.coder, &coder->dict,
  91                                 in, in_pos, in_size);
  92
  93                 // Copy the decoded data from the dictionary to the out[]
  94                 // buffer.
  95                 const size_t copy_size = coder->dict.pos - dict_start;
  96                 assert(copy_size <= out_size - *out_pos);
  97                 memcpy(out + *out_pos, coder->dict.buf + dict_start,
  98                                 copy_size);
  99                 *out_pos += copy_size;
 100
 101                 // Reset the dictionary if so requested by coder->lz.code().
 102                 if (coder->dict.need_reset) {
 103                         lz_decoder_reset(coder);
 104
 105                         // Since we reset dictionary, we don't check if
 106                         // dictionary became full.
 107                         if (ret != LZMA_OK || *out_pos == out_size)
 108                                 return ret;
 109                 } else {
 110                         // Return if everything got decoded or an error
 111                         // occurred, or if there's no more data to decode.
 112                         //
 113                         // Note that detecting if there's something to decode
 114                         // is done by looking if dictionary become full
 115                         // instead of looking if *in_pos == in_size. This
 116                         // is because it is possible that all the input was
 117                         // consumed already but some data is pending to be
 118                         // written to the dictionary.
 119                         if (ret != LZMA_OK || *out_pos == out_size
 120                                         || coder->dict.pos < coder->dict.size)
 121                                 return ret;
 122                 }
 123         }
 124 }
 125
 126
 127 static lzma_ret
 128 lz_decode(void *coder_ptr,
 129                 const lzma_allocator *allocator lzma_attribute((__unused__)),
 130                 const uint8_t *restrict in, size_t *restrict in_pos,
 131                 size_t in_size, uint8_t *restrict out,
 132                 size_t *restrict out_pos, size_t out_size,
 133                 lzma_action action)
 134 {
 135         lzma_coder *coder = coder_ptr;
 136
 137         if (coder->next.code == NULL)
 138                 return decode_buffer(coder, in, in_pos, in_size,
 139                                 out, out_pos, out_size);
 140
 141         // We aren't the last coder in the chain, we need to decode
 142         // our input to a temporary buffer.
 143         while (*out_pos < out_size) {
 144                 // Fill the temporary buffer if it is empty.
 145                 if (!coder->next_finished
 146                                 && coder->temp.pos == coder->temp.size) {
 147                         coder->temp.pos = 0;
 148                         coder->temp.size = 0;
 149
 150                         const lzma_ret ret = coder->next.code(
 151                                         coder->next.coder,
 152                                         allocator, in, in_pos, in_size,
 153                                         coder->temp.buffer, &coder->temp.size,
 154                                         LZMA_BUFFER_SIZE, action);
 155
 156                         if (ret == LZMA_STREAM_END)
 157                                 coder->next_finished = true;
 158                         else if (ret != LZMA_OK || coder->temp.size == 0)
 159                                 return ret;
 160                 }
 161
 162                 if (coder->this_finished) {
 163                         if (coder->temp.size != 0)
 164                                 return LZMA_DATA_ERROR;
 165
 166                         if (coder->next_finished)
 167                                 return LZMA_STREAM_END;
 168
 169                         return LZMA_OK;
 170                 }
 171
 172                 const lzma_ret ret = decode_buffer(coder, coder->temp.buffer,
 173                                 &coder->temp.pos, coder->temp.size,
 174                                 out, out_pos, out_size);
 175
 176                 if (ret == LZMA_STREAM_END)
 177                         coder->this_finished = true;
 178                 else if (ret != LZMA_OK)
 179                         return ret;
 180                 else if (coder->next_finished && *out_pos < out_size)
 181                         return LZMA_DATA_ERROR;
 182         }
 183
 184         return LZMA_OK;
 185 }
 186
 187
 188 static void
 189 lz_decoder_end(void *coder_ptr, const lzma_allocator *allocator)
 190 {
 191         lzma_coder *coder = coder_ptr;
 192
 193         lzma_next_end(&coder->next, allocator);
 194         lzma_free(coder->dict.buf, allocator);
 195
 196         if (coder->lz.end != NULL)
 197                 coder->lz.end(coder->lz.coder, allocator);
 198         else
 199                 lzma_free(coder->lz.coder, allocator);
 200
 201         lzma_free(coder, allocator);
 202         return;
 203 }
 204
 205
 206 extern lzma_ret
 207 lzma_lz_decoder_init(lzma_next_coder *next, const lzma_allocator *allocator,
 208                 const lzma_filter_info *filters,
 209                 lzma_ret (*lz_init)(lzma_lz_decoder *lz,
 210                         const lzma_allocator *allocator, const void *options,
 211                         lzma_lz_options *lz_options))
 212 {
 213         // Allocate the base structure if it isn't already allocated.
 214         lzma_coder *coder = next->coder;
 215         if (coder == NULL) {
 216                 coder = lzma_alloc(sizeof(lzma_coder), allocator);
 217                 if (coder == NULL)
 218                         return LZMA_MEM_ERROR;
 219
 220                 next->coder = coder;
 221                 next->code = &lz_decode;
 222                 next->end = &lz_decoder_end;
 223
 224                 coder->dict.buf = NULL;
 225                 coder->dict.size = 0;
 226                 coder->lz = LZMA_LZ_DECODER_INIT;
 227                 coder->next = LZMA_NEXT_CODER_INIT;
 228         }
 229
 230         // Allocate and initialize the LZ-based decoder. It will also give
 231         // us the dictionary size.
 232         lzma_lz_options lz_options;
 233         return_if_error(lz_init(&coder->lz, allocator,
 234                         filters[0].options, &lz_options));
 235
 236         // If the dictionary size is very small, increase it to 4096 bytes.
 237         // This is to prevent constant wrapping of the dictionary, which
 238         // would slow things down. The downside is that since we don't check
 239         // separately for the real dictionary size, we may happily accept
 240         // corrupt files.
 241         if (lz_options.dict_size < 4096)
 242                 lz_options.dict_size = 4096;
 243
 244         // Make dictionary size a multipe of 16. Some LZ-based decoders like
 245         // LZMA use the lowest bits lzma_dict.pos to know the alignment of the
 246         // data. Aligned buffer is also good when memcpying from the
 247         // dictionary to the output buffer, since applications are
 248         // recommended to give aligned buffers to liblzma.
 249         //
 250         // Avoid integer overflow.
 251         if (lz_options.dict_size > SIZE_MAX - 15)
 252                 return LZMA_MEM_ERROR;
 253
 254         lz_options.dict_size = (lz_options.dict_size + 15) & ~((size_t)(15));
 255
 256         // Allocate and initialize the dictionary.
 257         if (coder->dict.size != lz_options.dict_size) {
 258                 lzma_free(coder->dict.buf, allocator);
 259                 coder->dict.buf
 260                                 = lzma_alloc(lz_options.dict_size, allocator);
 261                 if (coder->dict.buf == NULL)
 262                         return LZMA_MEM_ERROR;
 263
 264                 coder->dict.size = lz_options.dict_size;
 265         }
 266
 267         lz_decoder_reset(next->coder);
 268
 269         // Use the preset dictionary if it was given to us.
 270         if (lz_options.preset_dict != NULL
 271                         && lz_options.preset_dict_size > 0) {
 272                 // If the preset dictionary is bigger than the actual
 273                 // dictionary, copy only the tail.
 274                 const size_t copy_size = my_min(lz_options.preset_dict_size,
 275                                 lz_options.dict_size);
 276                 const size_t offset = lz_options.preset_dict_size - copy_size;
 277                 memcpy(coder->dict.buf, lz_options.preset_dict + offset,
 278                                 copy_size);
 279                 coder->dict.pos = copy_size;
 280                 coder->dict.full = copy_size;
 281         }
 282
 283         // Miscellaneous initializations
 284         coder->next_finished = false;
 285         coder->this_finished = false;
 286         coder->temp.pos = 0;
 287         coder->temp.size = 0;
 288
 289         // Initialize the next filter in the chain, if any.
 290         return lzma_next_filter_init(&coder->next, allocator, filters + 1);
 291 }
 292
 293
 294 extern uint64_t
 295 lzma_lz_decoder_memusage(size_t dictionary_size)
 296 {
 297         return sizeof(lzma_coder) + (uint64_t)(dictionary_size);
 298 }
 299
 300
 301 extern void
 302 lzma_lz_decoder_uncompressed(void *coder_ptr, lzma_vli uncompressed_size)
 303 {
 304         lzma_coder *coder = coder_ptr;
 305         coder->lz.set_uncompressed(coder->lz.coder, uncompressed_size);
 306 }