src/liblzma/simple/simple_coder.c

   1 ///////////////////////////////////////////////////////////////////////////////
   2 //
   3 /// \file       simple_coder.c
   4 /// \brief      Wrapper for simple filters
   5 ///
   6 /// Simple filters don't change the size of the data i.e. number of bytes
   7 /// in equals the number of bytes out.
   8 //
   9 //  Copyright (C) 2007 Lasse Collin
  10 //
  11 //  This library is free software; you can redistribute it and/or
  12 //  modify it under the terms of the GNU Lesser General Public
  13 //  License as published by the Free Software Foundation; either
  14 //  version 2.1 of the License, or (at your option) any later version.
  15 //
  16 //  This library is distributed in the hope that it will be useful,
  17 //  but WITHOUT ANY WARRANTY; without even the implied warranty of
  18 //  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19 //  Lesser General Public License for more details.
  20 //
  21 ///////////////////////////////////////////////////////////////////////////////
  22
  23 #include "simple_private.h"
  24
  25
  26 /// Copied or encodes/decodes more data to out[]. Checks and updates
  27 /// uncompressed_size when we are the last coder in the chain.
  28 /// If we aren't the last filter in the chain, we don't need to care about
  29 /// uncompressed size, since we don't change it; the next filter in the
  30 /// chain will check it anyway.
  31 static lzma_ret
  32 copy_or_code(lzma_coder *coder, lzma_allocator *allocator,
  33                 const uint8_t *restrict in, size_t *restrict in_pos,
  34                 size_t in_size, uint8_t *restrict out,
  35                 size_t *restrict out_pos, size_t out_size, lzma_action action)
  36 {
  37         assert(!coder->end_was_reached);
  38
  39         if (coder->next.code == NULL) {
  40                 const size_t in_avail = in_size - *in_pos;
  41
  42                 if (!coder->is_encoder) {
  43                         // Limit in_size so that we don't copy too much.
  44                         if ((lzma_vli)(in_avail) > coder->uncompressed_size)
  45                                 in_size = *in_pos + (size_t)(
  46                                                 coder->uncompressed_size);
  47                 }
  48
  49                 const size_t out_start = *out_pos;
  50                 bufcpy(in, in_pos, in_size, out, out_pos, out_size);
  51
  52                 // Check if end of stream was reached.
  53                 if (coder->is_encoder) {
  54                         if (action == LZMA_FINISH && *in_pos == in_size)
  55                                 coder->end_was_reached = true;
  56                 } else if (coder->uncompressed_size
  57                                 != LZMA_VLI_VALUE_UNKNOWN) {
  58                         coder->uncompressed_size -= *out_pos - out_start;
  59                         if (coder->uncompressed_size == 0)
  60                                 coder->end_was_reached = true;
  61                 }
  62
  63         } else {
  64                 // Call the next coder in the chain to provide us some data.
  65                 // We don't care about uncompressed_size here, because
  66                 // the next filter in the chain will do it for us (since
  67                 // we don't change the size of the data).
  68                 const lzma_ret ret = coder->next.code(
  69                                 coder->next.coder, allocator,
  70                                 in, in_pos, in_size,
  71                                 out, out_pos, out_size, action);
  72
  73                 if (ret == LZMA_STREAM_END) {
  74                         assert(!coder->is_encoder
  75                                         || action == LZMA_FINISH);
  76                         coder->end_was_reached = true;
  77
  78                 } else if (ret != LZMA_OK) {
  79                         return ret;
  80                 }
  81         }
  82
  83         return LZMA_OK;
  84 }
  85
  86
  87 static size_t
  88 call_filter(lzma_coder *coder, uint8_t *buffer, size_t size)
  89 {
  90         const size_t filtered = coder->filter(coder->simple,
  91                         coder->now_pos, coder->is_encoder,
  92                         buffer, size);
  93         coder->now_pos += filtered;
  94         return filtered;
  95 }
  96
  97
  98 static lzma_ret
  99 simple_code(lzma_coder *coder, lzma_allocator *allocator,
 100                 const uint8_t *restrict in, size_t *restrict in_pos,
 101                 size_t in_size, uint8_t *restrict out,
 102                 size_t *restrict out_pos, size_t out_size, lzma_action action)
 103 {
 104         // TODO: Add partial support for LZMA_SYNC_FLUSH. We can support it
 105         // in cases when the filter is able to filter everything. With most
 106         // simple filters it can be done at offset that is a multiple of 2,
 107         // 4, or 16. With x86 filter, it needs good luck, and thus cannot
 108         // be made to work predictably.
 109         if (action == LZMA_SYNC_FLUSH)
 110                 return LZMA_HEADER_ERROR;
 111
 112         // Flush already filtered data from coder->buffer[] to out[].
 113         if (coder->pos < coder->filtered) {
 114                 bufcpy(coder->buffer, &coder->pos, coder->filtered,
 115                                 out, out_pos, out_size);
 116
 117                 // If we couldn't flush all the filtered data, return to
 118                 // application immediatelly.
 119                 if (coder->pos < coder->filtered)
 120                         return LZMA_OK;
 121
 122                 if (coder->end_was_reached) {
 123                         assert(coder->filtered == coder->size);
 124                         return LZMA_STREAM_END;
 125                 }
 126         }
 127
 128         // If we get here, there is no filtered data left in the buffer.
 129         coder->filtered = 0;
 130
 131         assert(!coder->end_was_reached);
 132
 133         // If there is more output space left than there is unfiltered data
 134         // in coder->buffer[], flush coder->buffer[] to out[], and copy/code
 135         // more data to out[] hopefully filling it completely. Then filter
 136         // the data in out[]. This step is where most of the data gets
 137         // filtered if the buffer sizes used by the application are reasonable.
 138         const size_t out_avail = out_size - *out_pos;
 139         const size_t buf_avail = coder->size - coder->pos;
 140         if (out_avail > buf_avail) {
 141                 // Store the old position so that we know from which byte
 142                 // to start filtering.
 143                 const size_t out_start = *out_pos;
 144
 145                 // Flush data from coder->buffer[] to out[], but don't reset
 146                 // coder->pos and coder->size yet. This way the coder can be
 147                 // restarted if the next filter in the chain returns e.g.
 148                 // LZMA_MEM_ERROR.
 149                 memcpy(out + *out_pos, coder->buffer + coder->pos, buf_avail);
 150                 *out_pos += buf_avail;
 151
 152                 // Copy/Encode/Decode more data to out[].
 153                 {
 154                         const lzma_ret ret = copy_or_code(coder, allocator,
 155                                         in, in_pos, in_size,
 156                                         out, out_pos, out_size, action);
 157                         assert(ret != LZMA_STREAM_END);
 158                         if (ret != LZMA_OK)
 159                                 return ret;
 160                 }
 161
 162                 // Filter out[].
 163                 const size_t size = *out_pos - out_start;
 164                 const size_t filtered = call_filter(
 165                                 coder, out + out_start, size);
 166
 167                 const size_t unfiltered = size - filtered;
 168                 assert(unfiltered <= coder->allocated / 2);
 169
 170                 // Now we can update coder->pos and coder->size, because
 171                 // the next coder in the chain (if any) was successful.
 172                 coder->pos = 0;
 173                 coder->size = unfiltered;
 174
 175                 if (coder->end_was_reached) {
 176                         // The last byte has been copied to out[] already.
 177                         // They are left as is.
 178                         coder->size = 0;
 179
 180                 } else if (unfiltered > 0) {
 181                         // There is unfiltered data left in out[]. Copy it to
 182                         // coder->buffer[] and rewind *out_pos appropriately.
 183                         *out_pos -= unfiltered;
 184                         memcpy(coder->buffer, out + *out_pos, unfiltered);
 185                 }
 186         } else if (coder->pos > 0) {
 187                 memmove(coder->buffer, coder->buffer + coder->pos, buf_avail);
 188                 coder->size -= coder->pos;
 189                 coder->pos = 0;
 190         }
 191
 192         assert(coder->pos == 0);
 193
 194         // If coder->buffer[] isn't empty, try to fill it by copying/decoding
 195         // more data. Then filter coder->buffer[] and copy the successfully
 196         // filtered data to out[]. It is probable, that some filtered and
 197         // unfiltered data will be left to coder->buffer[].
 198         if (coder->size > 0) {
 199                 {
 200                         const lzma_ret ret = copy_or_code(coder, allocator,
 201                                         in, in_pos, in_size,
 202                                         coder->buffer, &coder->size,
 203                                         coder->allocated, action);
 204                         assert(ret != LZMA_STREAM_END);
 205                         if (ret != LZMA_OK)
 206                                 return ret;
 207                 }
 208
 209                 coder->filtered = call_filter(
 210                                 coder, coder->buffer, coder->size);
 211
 212                 // Everything is considered to be filtered if coder->buffer[]
 213                 // contains the last bytes of the data.
 214                 if (coder->end_was_reached)
 215                         coder->filtered = coder->size;
 216
 217                 // Flush as much as possible.
 218                 bufcpy(coder->buffer, &coder->pos, coder->filtered,
 219                                 out, out_pos, out_size);
 220         }
 221
 222         // Check if we got everything done.
 223         if (coder->end_was_reached && coder->pos == coder->size)
 224                 return LZMA_STREAM_END;
 225
 226         return LZMA_OK;
 227 }
 228
 229
 230 static void
 231 simple_coder_end(lzma_coder *coder, lzma_allocator *allocator)
 232 {
 233         lzma_next_coder_end(&coder->next, allocator);
 234         lzma_free(coder->simple, allocator);
 235         lzma_free(coder, allocator);
 236         return;
 237 }
 238
 239
 240 extern lzma_ret
 241 lzma_simple_coder_init(lzma_next_coder *next, lzma_allocator *allocator,
 242                 const lzma_filter_info *filters,
 243                 size_t (*filter)(lzma_simple *simple, uint32_t now_pos,
 244                         bool is_encoder, uint8_t *buffer, size_t size),
 245                 size_t simple_size, size_t unfiltered_max, bool is_encoder)
 246 {
 247         // Allocate memory for the lzma_coder structure if needed.
 248         if (next->coder == NULL) {
 249                 // Here we allocate space also for the temporary buffer. We
 250                 // need twice the size of unfiltered_max, because then it
 251                 // is always possible to filter at least unfiltered_max bytes
 252                 // more data in coder->buffer[] if it can be filled completely.
 253                 next->coder = lzma_alloc(sizeof(lzma_coder)
 254                                 + 2 * unfiltered_max, allocator);
 255                 if (next->coder == NULL)
 256                         return LZMA_MEM_ERROR;
 257
 258                 next->code = &simple_code;
 259                 next->end = &simple_coder_end;
 260
 261                 next->coder->next = LZMA_NEXT_CODER_INIT;
 262                 next->coder->filter = filter;
 263                 next->coder->allocated = 2 * unfiltered_max;
 264
 265                 // Allocate memory for filter-specific data structure.
 266                 if (simple_size > 0) {
 267                         next->coder->simple = lzma_alloc(
 268                                         simple_size, allocator);
 269                         if (next->coder->simple == NULL)
 270                                 return LZMA_MEM_ERROR;
 271                 } else {
 272                         next->coder->simple = NULL;
 273                 }
 274         }
 275
 276         if (filters[0].options != NULL) {
 277                 const lzma_options_simple *simple = filters[0].options;
 278                 next->coder->now_pos = simple->start_offset;
 279         } else {
 280                 next->coder->now_pos = 0;
 281         }
 282
 283         // Reset variables.
 284         next->coder->is_encoder = is_encoder;
 285         next->coder->end_was_reached = false;
 286         next->coder->uncompressed_size = filters[0].uncompressed_size;
 287         next->coder->pos = 0;
 288         next->coder->filtered = 0;
 289         next->coder->size = 0;
 290
 291         return lzma_next_filter_init(
 292                         &next->coder->next, allocator, filters + 1);
 293 }