- DEBUG2("comp size = %"PRIu64", "
- "uncomp size = %"PRIu64", "
- "res offset = %"PRIu64"\n",
- resource_compressed_size,
- resource_uncompressed_size,
- resource_offset);
- DEBUG2("resource_ctype = %s, len = %"PRIu64", offset = %"PRIu64"\n",
- wimlib_get_compression_type_string(resource_ctype),
- len, offset);
- /* Trivial case */
- if (len == 0)
- return 0;
-
- int (*decompress)(const void *, uint, void *, uint);
- /* Set the appropriate decompress function. */
- if (resource_ctype == WIM_COMPRESSION_TYPE_LZX)
- decompress = lzx_decompress;
- else
- decompress = xpress_decompress;
-
- /* The structure of a compressed resource consists of a table of chunk
- * offsets followed by the chunks themselves. Each chunk consists of
- * compressed data, and there is one chunk for each WIM_CHUNK_SIZE =
- * 32768 bytes of the uncompressed file, with the last chunk having any
- * remaining bytes.
- *
- * The chunk offsets are measured relative to the end of the chunk
- * table. The first chunk is omitted from the table in the WIM file
- * because its offset is implicitly given by the fact that it directly
- * follows the chunk table and therefore must have an offset of 0.
- */
-
- /* Calculate how many chunks the resource conists of in its entirety. */
- u64 num_chunks = (resource_uncompressed_size + WIM_CHUNK_SIZE - 1) /
- WIM_CHUNK_SIZE;
- /* As mentioned, the first chunk has no entry in the chunk table. */
- u64 num_chunk_entries = num_chunks - 1;
-
-
- /* The index of the chunk that the read starts at. */
- u64 start_chunk = offset / WIM_CHUNK_SIZE;
- /* The byte offset at which the read starts, within the start chunk. */
- u64 start_chunk_offset = offset % WIM_CHUNK_SIZE;
-
- /* The index of the chunk that contains the last byte of the read. */
- u64 end_chunk = (offset + len - 1) / WIM_CHUNK_SIZE;
- /* The byte offset of the last byte of the read, within the end chunk */
- u64 end_chunk_offset = (offset + len - 1) % WIM_CHUNK_SIZE;
-
- /* Number of chunks that are actually needed to read the requested part
- * of the file. */
- u64 num_needed_chunks = end_chunk - start_chunk + 1;
-
- /* If the end chunk is not the last chunk, an extra chunk entry is
- * needed because we need to know the offset of the chunk after the last
- * chunk read to figure out the size of the last read chunk. */
- if (end_chunk != num_chunks - 1)
- num_needed_chunks++;
-
- /* Declare the chunk table. It will only contain offsets for the chunks
- * that are actually needed for this read. */
- u64 chunk_offsets[num_needed_chunks];
-
- /* Set the implicit offset of the first chunk if it is included in the
- * needed chunks.
- *
- * Note: M$'s documentation includes a picture that shows the first
- * chunk starting right after the chunk entry table, labeled as offset
- * 0x10. However, in the actual file format, the offset is measured
- * from the end of the chunk entry table, so the first chunk has an
- * offset of 0. */
- if (start_chunk == 0)
- chunk_offsets[0] = 0;
-
- /* According to M$'s documentation, if the uncompressed size of
- * the file is greater than 4 GB, the chunk entries are 8-byte
- * integers. Otherwise, they are 4-byte integers. */
- u64 chunk_entry_size = (resource_uncompressed_size >= (u64)1 << 32) ?
- 8 : 4;
-
- /* Size of the full chunk table in the WIM file. */
- u64 chunk_table_size = chunk_entry_size * num_chunk_entries;
-
- /* Read the needed chunk offsets from the table in the WIM file. */
-
- /* Index, in the WIM file, of the first needed entry in the
- * chunk table. */
- u64 start_table_idx = (start_chunk == 0) ? 0 : start_chunk - 1;
-
- /* Number of entries we need to actually read from the chunk
- * table (excludes the implicit first chunk). */
- u64 num_needed_chunk_entries = (start_chunk == 0) ?
- num_needed_chunks - 1 : num_needed_chunks;
-
- /* Skip over unneeded chunk table entries. */
- u64 file_offset_of_needed_chunk_entries = resource_offset +
- start_table_idx * chunk_entry_size;
- if (fseeko(fp, file_offset_of_needed_chunk_entries, SEEK_SET) != 0) {
- ERROR("Failed to seek to byte %"PRIu64" "
- "to read chunk table of compressed "
- "resource: %m\n",
- file_offset_of_needed_chunk_entries);
- return WIMLIB_ERR_READ;
+struct data_range {
+ u64 offset;
+ u64 size;
+};
+
+/* Alternate chunk table format for resources with WIM_RESHDR_FLAG_CONCAT set.
+ */
+struct alt_chunk_table_header_disk {
+ /* Uncompressed size of the resource. */
+ le64 res_usize;
+
+ /* Number of bytes each compressed chunk decompresses into, except
+ * possibly the last which decompresses into the remainder. */
+ le32 chunk_size;
+
+ /* ??? */
+ le32 unknown;
+
+ /* This header is directly followed by a table of compressed sizes of
+ * the chunks. */
+} _packed_attribute;
+
+/* Read data from a compressed WIM resource. */
+static int
+read_compressed_wim_resource(const struct wim_resource_spec * const rspec,
+ const struct data_range * const ranges,
+ const size_t num_ranges,
+ const consume_data_callback_t cb,
+ void * const cb_ctx,
+ const bool raw_chunks_mode)
+{
+ int ret;
+ int errno_save;
+
+ u64 *chunk_offsets = NULL;
+ u8 *ubuf = NULL;
+ void *cbuf = NULL;
+ bool chunk_offsets_malloced = false;
+ bool ubuf_malloced = false;
+ bool cbuf_malloced = false;
+
+ /* Sanity checks */
+ wimlib_assert(rspec != NULL);
+ wimlib_assert(rspec->ctype != WIMLIB_COMPRESSION_TYPE_NONE);
+ wimlib_assert(is_power_of_2(rspec->cchunk_size));
+ wimlib_assert(cb != NULL);
+ wimlib_assert(num_ranges != 0);
+ for (size_t i = 0; i < num_ranges; i++) {
+ wimlib_assert(ranges[i].size != 0);
+ wimlib_assert(ranges[i].offset + ranges[i].size >= ranges[i].size);
+ wimlib_assert(ranges[i].offset + ranges[i].size <= rspec->uncompressed_size);
+ }
+ for (size_t i = 0; i < num_ranges - 1; i++)
+ wimlib_assert(ranges[i].offset + ranges[i].size <= ranges[i + 1].offset);
+
+ /* Get the offsets of the first and last bytes of the read. */
+ const u64 first_offset = ranges[0].offset;
+ const u64 last_offset = ranges[num_ranges - 1].offset + ranges[num_ranges - 1].size - 1;
+
+ /* Get the file descriptor for the WIM. */
+ struct filedes * const in_fd = &rspec->wim->in_fd;
+
+ /* Determine if we're reading a pipable resource from a pipe or not. */
+ const bool is_pipe_read = !filedes_is_seekable(in_fd);
+
+ /* Determine if the chunk table is in an altenate format. */
+ const bool alt_chunk_table = (rspec->flags & WIM_RESHDR_FLAG_CONCAT) && !is_pipe_read;
+
+ /* Get the maximum size of uncompressed chunks in this resource, which
+ * we require be a power of 2. */
+ u32 chunk_size;
+ u64 cur_read_offset = rspec->offset_in_wim;
+ if (alt_chunk_table) {
+ /* Alternate chunk table format. */
+ struct alt_chunk_table_header_disk hdr;
+
+ ret = full_pread(in_fd, &hdr, sizeof(hdr), cur_read_offset);
+ if (ret)
+ goto read_error;
+ cur_read_offset += sizeof(hdr);
+
+ chunk_size = le32_to_cpu(hdr.chunk_size);
+
+ if (!is_power_of_2(chunk_size)) {
+ ERROR("Invalid compressed resource: "
+ "expected power-of-2 chunk size (got %u)", chunk_size);
+ ret = WIMLIB_ERR_INVALID_CHUNK_SIZE;
+ goto out_free_memory;
+ }
+ } else {
+ chunk_size = rspec->cchunk_size;