resource.c: Cleanup and refactor
[wimlib] / src / resource.c
index d60f547..b38c65d 100644 (file)
@@ -27,7 +27,6 @@
 #endif
 
 #include "wimlib.h"
-#include "wimlib/dentry.h"
 #include "wimlib/endianness.h"
 #include "wimlib/error.h"
 #include "wimlib/file_io.h"
 #endif
 #include <errno.h>
 #include <fcntl.h>
-#include <stdarg.h>
 #include <stdlib.h>
 #include <unistd.h>
 
 /*
- * Reads all or part of a compressed WIM resource.
+ *                         Compressed WIM resources
  *
- * Returns zero on success, nonzero on failure.
+ * A compressed resource in a WIM consists of a number of compressed chunks,
+ * each of which decompresses to a fixed chunk size (given in the WIM header;
+ * usually 32768) except possibly the last, which always decompresses to any
+ * remaining bytes.  In addition, immediately before the chunks, a table (the
+ * "chunk table") provides the offset, in bytes relative to the end of the chunk
+ * table, of the start of each compressed chunk, except for the first chunk
+ * which is omitted as it always has an offset of 0.  Therefore, a compressed
+ * resource with N chunks will have a chunk table with N - 1 entries.
+ *
+ * Additional information:
+ *
+ * - Entries in the chunk table are 4 bytes each, except if the uncompressed
+ *   size of the resource is greater than 4 GiB, in which case the entries in
+ *   the chunk table are 8 bytes each.  In either case, the entries are unsigned
+ *   little-endian integers.
+ *
+ * - The chunk table is included in the compressed size of the resource provided
+ *   in the corresponding entry in the WIM's stream lookup table.
+ *
+ * - The compressed size of a chunk is never greater than the uncompressed size.
+ *   From the compressor's point of view, chunks that would have compressed to a
+ *   size greater than or equal to their original size are in fact stored
+ *   uncompressed.  From the decompresser's point of view, chunks with
+ *   compressed size equal to their uncompressed size are in fact uncompressed.
+ *
+ * Furthermore, wimlib supports its own "pipable" WIM format, and for this the
+ * structure of compressed resources was modified to allow piped reading and
+ * writing.  To make sequential writing possible, the chunk table is placed
+ * after the chunks rather than before the chunks, and to make sequential
+ * reading possible, each chunk is prefixed with a 4-byte header giving its
+ * compressed size as a 32-bit, unsigned, little-endian integer.  Otherwise the
+ * details are the same.
  */
+
+
+/* Decompress the specified chunk that uses the specified compression type
+ * @ctype, part of a WIM with default chunk size @wim_chunk_size.  For LZX the
+ * separate @wim_chunk_size is needed because it determines the window size used
+ * for LZX compression.  */
 static int
-read_compressed_resource(int in_fd,
-                        u64 resource_compressed_size,
-                        u64 resource_uncompressed_size,
-                        u64 resource_offset,
-                        int resource_ctype,
-                        u64 len,
-                        u64 offset,
-                        consume_data_callback_t cb,
-                        void *ctx_or_buf)
+decompress(const void *cchunk, unsigned clen,
+          void *uchunk, unsigned ulen,
+          int ctype, u32 wim_chunk_size)
+{
+       switch (ctype) {
+       case WIMLIB_COMPRESSION_TYPE_XPRESS:
+               return wimlib_xpress_decompress(cchunk,
+                                               clen,
+                                               uchunk,
+                                               ulen);
+       case WIMLIB_COMPRESSION_TYPE_LZX:
+               return wimlib_lzx_decompress2(cchunk,
+                                             clen,
+                                             uchunk,
+                                             ulen,
+                                             wim_chunk_size);
+       default:
+               wimlib_assert(0);
+               return -1;
+       }
+}
+
+/* Read data from a compressed WIM resource.  Assumes parameters were already
+ * verified by read_partial_wim_resource().  */
+static int
+read_compressed_wim_resource(const struct wim_lookup_table_entry * const lte,
+                            const u64 size, const consume_data_callback_t cb,
+                            const u32 cb_chunk_size, void * const ctx_or_buf,
+                            const int flags, const u64 offset)
 {
        int ret;
+       int errno_save;
+
+       const u32 orig_chunk_size = wim_resource_chunk_size(lte);
+       const u32 orig_chunk_order = bsr32(orig_chunk_size);
 
-       /* Trivial case */
-       if (len == 0)
+       wimlib_assert(is_power_of_2(orig_chunk_size));
+
+       /* Handle the trivial case.  */
+       if (size == 0)
                return 0;
 
-       int (*decompress)(const void *, unsigned, void *, unsigned);
-       /* Set the appropriate decompress function. */
-       if (resource_ctype == WIMLIB_COMPRESSION_TYPE_LZX)
-               decompress = wimlib_lzx_decompress;
-       else
-               decompress = wimlib_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.
+       u64 *chunk_offsets = NULL;
+       u8 *out_buf = NULL;
+       u8 *tmp_buf = NULL;
+       void *compressed_buf = NULL;
+       bool chunk_offsets_malloced = false;
+       bool out_buf_malloced = false;
+       bool tmp_buf_malloced = false;
+       bool compressed_buf_malloced = false;
+
+       /* Get the file descriptor for the WIM.  */
+       struct filedes * const in_fd = &lte->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);
+
+       /* Calculate the number of chunks the resource is divided into.  */
+       const u64 num_chunks = wim_resource_chunks(lte);
+
+       /* Calculate the 0-based index of the chunk at which the read starts.
         */
+       const u64 start_chunk = offset >> orig_chunk_order;
+
+       /* For pipe reads, we always must start from the 0th chunk.  */
+       const u64 actual_start_chunk = (is_pipe_read ? 0 : start_chunk);
+
+       /* Calculate the offset, within the start chunk, of the first byte of
+        * the read.  */
+       const u32 start_offset_in_chunk = offset & (orig_chunk_size - 1);
+
+       /* Calculate the index of the chunk that contains the last byte of the
+        * read.  */
+       const u64 end_chunk = (offset + size - 1) >> orig_chunk_order;
+
+       /* Calculate the offset, within the end chunk, of the last byte of the
+        * read.  */
+       const u32 end_offset_in_chunk = (offset + size - 1) & (orig_chunk_size - 1);
+
+       /* Calculate the number of entries in the chunk table; it's one less
+        * than the number of chunks, since the first chunk has no entry.  */
+       const u64 num_chunk_entries = num_chunks - 1;
+
+       /* Set the size of each chunk table entry based on the resource's
+        * uncompressed size.  */
+       const u64 chunk_entry_size = (wim_resource_size(lte) > (1ULL << 32)) ? 8 : 4;
+
+       /* Calculate the size, in bytes, of the full chunk table.  */
+       const u64 chunk_table_size = num_chunk_entries * chunk_entry_size;
+
+       /* Current offset to read from.  */
+       u64 cur_read_offset = lte->resource_entry.offset;
+       if (!is_pipe_read) {
+               /* Read the chunk table into memory.  */
+
+               /* Calculate the number of chunk entries are actually needed to
+                * read the requested part of the resource.  Include an entry
+                * for the first chunk even though that doesn't exist in the
+                * on-disk table, but take into account that if the last chunk
+                * required for the read is not the last chunk of the resource,
+                * an extra chunk entry is needed so that the compressed size of
+                * the last chunk of the read can be determined.  */
+               const u64 num_alloc_chunk_entries = end_chunk - start_chunk +
+                                                   1 + (end_chunk != num_chunks - 1);
+
+               /* Allocate a buffer to hold a subset of the chunk table.  It
+                * will only contain offsets for the chunks that are actually
+                * needed for this read.  For speed, allocate the buffer on the
+                * stack unless it's too large.  */
+               if ((size_t)(num_alloc_chunk_entries * sizeof(u64)) !=
+                           (num_alloc_chunk_entries * sizeof(u64)))
+                       goto oom;
+
+               if (num_alloc_chunk_entries <= STACK_MAX / sizeof(u64)) {
+                       chunk_offsets = alloca(num_alloc_chunk_entries * sizeof(u64));
+               } else {
+                       chunk_offsets = MALLOC(num_alloc_chunk_entries * sizeof(u64));
+                       if (chunk_offsets == NULL)
+                               goto oom;
+                       chunk_offsets_malloced = true;
+               }
 
-       /* Calculate how many chunks the resource consists 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++;
-
-       /* 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;
-
-       /* Allocate the chunk table.  It will only contain offsets for the
-        * chunks that are actually needed for this read. */
-       u64 *chunk_offsets;
-       bool chunk_offsets_malloced;
-       if (num_needed_chunks < 1024) {
-               chunk_offsets = alloca(num_needed_chunks * sizeof(u64));
-               chunk_offsets_malloced = false;
-       } else {
-               chunk_offsets = malloc(num_needed_chunks * sizeof(u64));
-               if (!chunk_offsets) {
-                       ERROR("Failed to allocate chunk table "
-                             "with %"PRIu64" entries", num_needed_chunks);
-                       return WIMLIB_ERR_NOMEM;
+               /* Set the implicit offset of the first chunk if it's included
+                * in the needed chunks.  */
+               if (start_chunk == 0)
+                       chunk_offsets[0] = 0;
+
+               /* Calculate the index of the first needed entry in the chunk
+                * table.  */
+               const u64 start_table_idx = (start_chunk == 0) ? 0 : start_chunk - 1;
+
+               /* Calculate the number of entries that need to be read from the
+                * chunk table.  */
+               const u64 num_needed_chunk_entries = (start_chunk == 0) ?
+                                       num_alloc_chunk_entries - 1 : num_alloc_chunk_entries;
+
+               /* Calculate the number of bytes of data that need to be read
+                * from the chunk table.  */
+               const size_t chunk_table_needed_size =
+                                       num_needed_chunk_entries * chunk_entry_size;
+
+               /* Calculate the byte offset, in the WIM file, of the first
+                * chunk table entry to read.  Take into account that if the WIM
+                * file is in the special "pipable" format, then the chunk table
+                * is at the end of the resource, not the beginning.  */
+               const u64 file_offset_of_needed_chunk_entries =
+                       lte->resource_entry.offset
+                       + (start_table_idx * chunk_entry_size)
+                       + (lte->is_pipable ? (lte->resource_entry.size - chunk_table_size) : 0);
+
+               /* Read the needed chunk table entries into the end of the
+                * chunk_offsets buffer.  */
+               void * const chunk_tab_data = (u8*)&chunk_offsets[num_alloc_chunk_entries] -
+                                             chunk_table_needed_size;
+               ret = full_pread(in_fd, chunk_tab_data, chunk_table_needed_size,
+                                file_offset_of_needed_chunk_entries);
+               if (ret)
+                       goto read_error;
+
+               /* Now fill in chunk_offsets from the entries we have read in
+                * chunk_tab_data.  Careful: chunk_offsets aliases
+                * chunk_tab_data, which breaks C's aliasing rules when we read
+                * 32-bit integers and store 64-bit integers.  But since the
+                * operations are safe as long as the compiler doesn't mess with
+                * their order, we use the gcc may_alias extension to tell the
+                * compiler that loads from the 32-bit integers may alias stores
+                * to the 64-bit integers.  */
+               {
+                       typedef le64 __attribute__((may_alias)) aliased_le64_t;
+                       typedef le32 __attribute__((may_alias)) aliased_le32_t;
+                       u64 * const chunk_offsets_p = chunk_offsets + (start_chunk == 0);
+                       u64 i;
+
+                       if (chunk_entry_size == 4) {
+                               aliased_le32_t *raw_entries = (aliased_le32_t*)chunk_tab_data;
+                               for (i = 0; i < num_needed_chunk_entries; i++)
+                                       chunk_offsets_p[i] = le32_to_cpu(raw_entries[i]);
+                       } else {
+                               aliased_le64_t *raw_entries = (aliased_le64_t*)chunk_tab_data;
+                               for (i = 0; i < num_needed_chunk_entries; i++)
+                                       chunk_offsets_p[i] = le64_to_cpu(raw_entries[i]);
+                       }
                }
-               chunk_offsets_malloced = true;
-       }
 
-       /* 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;
-
-
-       /* 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;
-
-       /* Allocate a buffer into which to read the raw chunk entries. */
-       void *chunk_tab_buf;
-       bool chunk_tab_buf_malloced = false;
-
-       /* Number of bytes we need to read from the chunk table. */
-       size_t size = num_needed_chunk_entries * chunk_entry_size;
-       if ((u64)size != num_needed_chunk_entries * chunk_entry_size) {
-               ERROR("Compressed read request too large to fit into memory!");
-               ret = WIMLIB_ERR_NOMEM;
-               goto out;
+               /* Set offset to beginning of first chunk to read.  */
+               cur_read_offset += chunk_table_size + chunk_offsets[0];
        }
 
-       if (size < 4096) {
-               chunk_tab_buf = alloca(size);
+       /* If using a callback function, allocate a temporary buffer that will
+        * be used to pass data to it.  If writing directly to a buffer instead,
+        * arrange to write data directly into it.  */
+       size_t out_buf_size;
+       u8 *out_buf_end, *out_p;
+       if (cb) {
+               out_buf_size = max(cb_chunk_size, orig_chunk_size);
+               if (out_buf_size <= STACK_MAX) {
+                       out_buf = alloca(out_buf_size);
+               } else {
+                       out_buf = MALLOC(out_buf_size);
+                       if (out_buf == NULL)
+                               goto oom;
+                       out_buf_malloced = true;
+               }
        } else {
-               chunk_tab_buf = malloc(size);
-               if (!chunk_tab_buf) {
-                       ERROR("Failed to allocate chunk table buffer of "
-                             "size %zu bytes", size);
-                       ret = WIMLIB_ERR_NOMEM;
-                       goto out;
+               out_buf_size = size;
+               out_buf = ctx_or_buf;
+       }
+       out_buf_end = out_buf + out_buf_size;
+       out_p = out_buf;
+
+       /* Unless the raw compressed data was requested, allocate a temporary
+        * buffer for reading compressed chunks, each of which can be at most
+        * @orig_chunk_size - 1 bytes.  This excludes compressed chunks that are
+        * a full @orig_chunk_size bytes, which are actually stored
+        * uncompressed.  */
+       if (!(flags & WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS)) {
+               if (orig_chunk_size - 1 <= STACK_MAX) {
+                       compressed_buf = alloca(orig_chunk_size - 1);
+               } else {
+                       compressed_buf = MALLOC(orig_chunk_size - 1);
+                       if (compressed_buf == NULL)
+                               goto oom;
+                       compressed_buf_malloced = true;
                }
-               chunk_tab_buf_malloced = true;
        }
 
-       if (full_pread(in_fd, chunk_tab_buf, size,
-                      file_offset_of_needed_chunk_entries) != size)
-               goto read_error;
-
-       /* Now fill in chunk_offsets from the entries we have read in
-        * chunk_tab_buf. */
-
-       u64 *chunk_tab_p = chunk_offsets;
-       if (start_chunk == 0)
-               chunk_tab_p++;
-
-       if (chunk_entry_size == 4) {
-               le32 *entries = (le32*)chunk_tab_buf;
-               while (num_needed_chunk_entries--)
-                       *chunk_tab_p++ = le32_to_cpu(*entries++);
-       } else {
-               le64 *entries = (le64*)chunk_tab_buf;
-               while (num_needed_chunk_entries--)
-                       *chunk_tab_p++ = le64_to_cpu(*entries++);
+       /* Allocate yet another temporary buffer, this one for decompressing
+        * chunks for which only part of the data is needed.  */
+       if (start_offset_in_chunk != 0 ||
+           (end_offset_in_chunk != orig_chunk_size - 1 &&
+            offset + size != wim_resource_size(lte)))
+       {
+               if (orig_chunk_size <= STACK_MAX) {
+                       tmp_buf = alloca(orig_chunk_size);
+               } else {
+                       tmp_buf = MALLOC(orig_chunk_size);
+                       if (tmp_buf == NULL)
+                               goto oom;
+                       tmp_buf_malloced = true;
+               }
        }
 
-       /* Done reading the chunk table now.  Now calculate the file offset for
-        * the first byte of compressed data we need to read. */
+       /* Read, and possibly decompress, each needed chunk, either writing the
+        * data directly into the @ctx_or_buf buffer or passing it to the @cb
+        * callback function.  */
+       for (u64 i = actual_start_chunk; i <= end_chunk; i++) {
 
-       u64 cur_read_offset = resource_offset + chunk_table_size + chunk_offsets[0];
+               /* Calculate uncompressed size of next chunk.  */
+               u32 chunk_usize;
+               if ((i == num_chunks - 1) && (wim_resource_size(lte) & (orig_chunk_size - 1)))
+                       chunk_usize = (wim_resource_size(lte) & (orig_chunk_size - 1));
+               else
+                       chunk_usize = orig_chunk_size;
 
-       /* Pointer to current position in the output buffer for uncompressed
-        * data.  Alternatively, if using a callback function, we repeatedly
-        * fill a temporary buffer to feed data into the callback function.  */
-       u8 *out_p;
-       if (cb)
-               out_p = alloca(WIM_CHUNK_SIZE);
-       else
-               out_p = ctx_or_buf;
-
-       /* Buffer for compressed data.  While most compressed chunks will have a
-        * size much less than WIM_CHUNK_SIZE, WIM_CHUNK_SIZE - 1 is the maximum
-        * size in the worst-case.  This assumption is valid only if chunks that
-        * happen to compress to more than the uncompressed size (i.e. a
-        * sequence of random bytes) are always stored uncompressed. But this seems
-        * to be the case in M$'s WIM files, even though it is undocumented. */
-       void *compressed_buf = alloca(WIM_CHUNK_SIZE - 1);
-
-       /* Decompress all the chunks. */
-       for (u64 i = start_chunk; i <= end_chunk; i++) {
-
-               /* Calculate the sizes of the compressed chunk and of the
-                * uncompressed chunk. */
-               unsigned compressed_chunk_size;
-               unsigned uncompressed_chunk_size;
-               if (i != num_chunks - 1) {
-                       /* All the chunks except the last one in the resource
-                        * expand to WIM_CHUNK_SIZE uncompressed, and the amount
-                        * of compressed data for the chunk is given by the
-                        * difference of offsets in the chunk offset table. */
-                       compressed_chunk_size = chunk_offsets[i + 1 - start_chunk] -
-                                               chunk_offsets[i - start_chunk];
-                       uncompressed_chunk_size = WIM_CHUNK_SIZE;
+               /* Calculate compressed size of next chunk.  */
+               u32 chunk_csize;
+               if (is_pipe_read) {
+                       struct pwm_chunk_hdr chunk_hdr;
+
+                       ret = full_pread(in_fd, &chunk_hdr,
+                                        sizeof(chunk_hdr), cur_read_offset);
+                       if (ret)
+                               goto read_error;
+                       chunk_csize = le32_to_cpu(chunk_hdr.compressed_size);
                } else {
-                       /* The last compressed chunk consists of the remaining
-                        * bytes in the file resource, and the last uncompressed
-                        * chunk has size equal to however many bytes are left-
-                        * that is, the remainder of the uncompressed size when
-                        * divided by WIM_CHUNK_SIZE.
-                        *
-                        * Note that the resource_compressed_size includes the
-                        * chunk table, so the size of it must be subtracted. */
-                       compressed_chunk_size = resource_compressed_size -
-                                               chunk_table_size -
-                                               chunk_offsets[i - start_chunk];
-
-                       uncompressed_chunk_size = resource_uncompressed_size %
-                                                               WIM_CHUNK_SIZE;
-
-                       /* If the remainder is 0, the last chunk actually
-                        * uncompresses to a full WIM_CHUNK_SIZE bytes. */
-                       if (uncompressed_chunk_size == 0)
-                               uncompressed_chunk_size = WIM_CHUNK_SIZE;
+                       if (i == num_chunks - 1) {
+                               chunk_csize = lte->resource_entry.size -
+                                             chunk_table_size -
+                                             chunk_offsets[i - start_chunk];
+                               if (lte->is_pipable)
+                                       chunk_csize -= num_chunks * sizeof(struct pwm_chunk_hdr);
+                       } else {
+                               chunk_csize = chunk_offsets[i + 1 - start_chunk] -
+                                             chunk_offsets[i - start_chunk];
+                       }
+               }
+               if (chunk_csize == 0 || chunk_csize > orig_chunk_size) {
+                       ERROR("Invalid chunk size in compressed resource!");
+                       errno = EINVAL;
+                       ret = WIMLIB_ERR_INVALID_CHUNK_SIZE;
+                       goto out_free_memory;
                }
+               if (lte->is_pipable)
+                       cur_read_offset += sizeof(struct pwm_chunk_hdr);
 
-               /* Figure out how much of this chunk we actually need to read */
-               u64 start_offset;
-               if (i == start_chunk)
-                       start_offset = start_chunk_offset;
-               else
-                       start_offset = 0;
-               u64 end_offset;
-               if (i == end_chunk)
-                       end_offset = end_chunk_offset;
-               else
-                       end_offset = WIM_CHUNK_SIZE - 1;
-
-               unsigned partial_chunk_size = end_offset + 1 - start_offset;
-               bool is_partial_chunk = (partial_chunk_size != uncompressed_chunk_size);
-
-               /* This is undocumented, but chunks can be uncompressed.  This
-                * appears to always be the case when the compressed chunk size
-                * is equal to the uncompressed chunk size. */
-               if (compressed_chunk_size == uncompressed_chunk_size) {
-                       /* Uncompressed chunk */
-                       if (full_pread(in_fd,
-                                      cb ? out_p + start_offset : out_p,
-                                      partial_chunk_size,
-                                      cur_read_offset + start_offset) != partial_chunk_size)
-                       {
-                               goto read_error;
+               if (i >= start_chunk) {
+                       /* Calculate how much of this chunk needs to be read.  */
+                       u32 chunk_needed_size;
+                       u32 start_offset = 0;
+                       u32 end_offset = orig_chunk_size - 1;
+
+                       if (flags & WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS) {
+                               chunk_needed_size = chunk_csize;
+                       } else {
+                               if (i == start_chunk)
+                                       start_offset = start_offset_in_chunk;
+
+                               if (i == end_chunk)
+                                       end_offset = end_offset_in_chunk;
+
+                               chunk_needed_size = end_offset + 1 - start_offset;
                        }
-               } else {
-                       /* Compressed chunk */
 
-                       /* Read the compressed data into compressed_buf. */
-                       if (full_pread(in_fd,
-                                      compressed_buf,
-                                      compressed_chunk_size,
-                                      cur_read_offset) != compressed_chunk_size)
+                       if (chunk_csize == chunk_usize ||
+                           (flags & WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS))
                        {
-                               goto read_error;
-                       }
+                               /* Read the raw chunk data.  */
+
+                               ret = full_pread(in_fd,
+                                                out_p,
+                                                chunk_needed_size,
+                                                cur_read_offset + start_offset);
+                               if (ret)
+                                       goto read_error;
+                       } else {
+                               /* Read and decompress the chunk.  */
 
-                       /* For partial chunks and when writing directly to a
-                        * buffer, we must buffer the uncompressed data because
-                        * we don't need all of it. */
-                       if (is_partial_chunk && !cb) {
-                               u8 uncompressed_buf[uncompressed_chunk_size];
+                               u8 *target;
+
+                               ret = full_pread(in_fd,
+                                                compressed_buf,
+                                                chunk_csize,
+                                                cur_read_offset);
+                               if (ret)
+                                       goto read_error;
+
+                               if (chunk_needed_size == chunk_usize)
+                                       target = out_p;
+                               else
+                                       target = tmp_buf;
 
                                ret = decompress(compressed_buf,
-                                                compressed_chunk_size,
-                                                uncompressed_buf,
-                                                uncompressed_chunk_size);
+                                                chunk_csize,
+                                                target,
+                                                chunk_usize,
+                                                wim_resource_compression_type(lte),
+                                                orig_chunk_size);
                                if (ret) {
+                                       ERROR("Failed to decompress data!");
                                        ret = WIMLIB_ERR_DECOMPRESSION;
-                                       goto out;
+                                       errno = EINVAL;
+                                       goto out_free_memory;
                                }
-                               memcpy(out_p, uncompressed_buf + start_offset,
-                                      partial_chunk_size);
-                       } else {
-                               ret = decompress(compressed_buf,
-                                                compressed_chunk_size,
-                                                out_p,
-                                                uncompressed_chunk_size);
-                               if (ret) {
-                                       ret = WIMLIB_ERR_DECOMPRESSION;
-                                       goto out;
+                               if (chunk_needed_size != chunk_usize)
+                                       memcpy(out_p, tmp_buf + start_offset,
+                                              chunk_needed_size);
+                       }
+
+                       out_p += chunk_needed_size;
+
+                       if (cb) {
+                               /* Feed the data to the callback function.  */
+
+                               if (flags & WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS) {
+                                       ret = cb(out_buf, out_p - out_buf, ctx_or_buf);
+                                       if (ret)
+                                               goto out_free_memory;
+                                       out_p = out_buf;
+                               } else if (i == end_chunk || out_p == out_buf_end) {
+                                       size_t bytes_sent;
+                                       const u8 *p;
+
+                                       for (p = out_buf; p != out_p; p += bytes_sent) {
+                                               bytes_sent = min(cb_chunk_size, out_p - p);
+                                               ret = cb(p, bytes_sent, ctx_or_buf);
+                                               if (ret)
+                                                       goto out_free_memory;
+                                       }
+                                       out_p = out_buf;
                                }
                        }
-               }
-               if (cb) {
-                       /* Feed the data to the callback function */
-                       ret = cb(out_p + start_offset,
-                                partial_chunk_size, ctx_or_buf);
-                       if (ret)
-                               goto out;
+                       cur_read_offset += chunk_csize;
                } else {
-                       /* No callback function provided; we are writing
-                        * directly to a buffer.  Advance the pointer into this
-                        * buffer by the number of uncompressed bytes that were
-                        * written.  */
-                       out_p += partial_chunk_size;
+                       u8 dummy;
+
+                       /* Skip data only.  */
+                       cur_read_offset += chunk_csize;
+                       ret = full_pread(in_fd, &dummy, 1, cur_read_offset - 1);
+                       if (ret)
+                               goto read_error;
                }
-               cur_read_offset += compressed_chunk_size;
        }
 
+       if (is_pipe_read
+           && size == lte->resource_entry.original_size
+           && chunk_table_size)
+       {
+               u8 dummy;
+               /* Skip chunk table at end of pipable resource.  */
+
+               cur_read_offset += chunk_table_size;
+               ret = full_pread(in_fd, &dummy, 1, cur_read_offset - 1);
+               if (ret)
+                       goto read_error;
+       }
        ret = 0;
-out:
+out_free_memory:
+       errno_save = errno;
        if (chunk_offsets_malloced)
                FREE(chunk_offsets);
-       if (chunk_tab_buf_malloced)
-               FREE(chunk_tab_buf);
+       if (out_buf_malloced)
+               FREE(out_buf);
+       if (compressed_buf_malloced)
+               FREE(compressed_buf);
+       if (tmp_buf_malloced)
+               FREE(tmp_buf);
+       errno = errno_save;
        return ret;
 
+oom:
+       ERROR("Not enough memory available to read size=%"PRIu64" bytes "
+             "from compressed resource!", size);
+       errno = ENOMEM;
+       ret = WIMLIB_ERR_NOMEM;
+       goto out_free_memory;
+
 read_error:
-       ERROR_WITH_ERRNO("Error reading compressed file resource");
-       ret = WIMLIB_ERR_READ;
-       goto out;
+       ERROR_WITH_ERRNO("Error reading compressed file resource!");
+       goto out_free_memory;
 }
 
-/* Translates a WIM resource entry from the on-disk format to an in-memory
- * format. */
-void
-get_resource_entry(const struct resource_entry_disk *disk_entry,
-                  struct resource_entry *entry)
+/* Read raw data from a file descriptor at the specified offset.  */
+static int
+read_raw_file_data(struct filedes *in_fd,
+                  u64 size,
+                  consume_data_callback_t cb,
+                  u32 cb_chunk_size,
+                  void *ctx_or_buf,
+                  u64 offset)
 {
-       /* Note: disk_entry may not be 8 byte aligned--- in that case, the
-        * offset and original_size members will be unaligned.  (This should be
-        * okay since `struct resource_entry_disk' is declared as packed.) */
+       int ret;
+       u8 *tmp_buf;
+       bool tmp_buf_malloced = false;
 
-       /* Read the size and flags into a bitfield portably... */
-       entry->size = (((u64)disk_entry->size[0] <<  0) |
-                      ((u64)disk_entry->size[1] <<  8) |
-                      ((u64)disk_entry->size[2] << 16) |
-                      ((u64)disk_entry->size[3] << 24) |
-                      ((u64)disk_entry->size[4] << 32) |
-                      ((u64)disk_entry->size[5] << 40) |
-                      ((u64)disk_entry->size[6] << 48));
-       entry->flags = disk_entry->flags;
-       entry->offset = le64_to_cpu(disk_entry->offset);
-       entry->original_size = le64_to_cpu(disk_entry->original_size);
+       if (cb) {
+               /* Send data to callback function in chunks.  */
+               if (cb_chunk_size <= STACK_MAX) {
+                       tmp_buf = alloca(cb_chunk_size);
+               } else {
+                       tmp_buf = MALLOC(cb_chunk_size);
+                       if (tmp_buf == NULL) {
+                               ret = WIMLIB_ERR_NOMEM;
+                               goto out;
+                       }
+                       tmp_buf_malloced = true;
+               }
 
-       /* offset and original_size are truncated to 62 bits to avoid possible
-        * overflows, when converting to a signed 64-bit integer (off_t) or when
-        * adding size or original_size.  This is okay since no one would ever
-        * actually have a WIM bigger than 4611686018427387903 bytes... */
-       if (entry->offset & 0xc000000000000000ULL) {
-               WARNING("Truncating offset in resource entry");
-               entry->offset &= 0x3fffffffffffffffULL;
-       }
-       if (entry->original_size & 0xc000000000000000ULL) {
-               WARNING("Truncating original_size in resource entry");
-               entry->original_size &= 0x3fffffffffffffffULL;
+               while (size) {
+                       size_t bytes_to_read = min(cb_chunk_size, size);
+                       ret = full_pread(in_fd, tmp_buf, bytes_to_read,
+                                        offset);
+                       if (ret)
+                               goto read_error;
+                       ret = cb(tmp_buf, bytes_to_read, ctx_or_buf);
+                       if (ret)
+                               goto out;
+                       size -= bytes_to_read;
+                       offset += bytes_to_read;
+               }
+       } else {
+               /* Read data directly into buffer.  */
+               ret = full_pread(in_fd, ctx_or_buf, size, offset);
+               if (ret)
+                       goto read_error;
        }
-}
-
-/* Translates a WIM resource entry from an in-memory format into the on-disk
- * format. */
-void
-put_resource_entry(const struct resource_entry *entry,
-                  struct resource_entry_disk *disk_entry)
-{
-       /* Note: disk_entry may not be 8 byte aligned--- in that case, the
-        * offset and original_size members will be unaligned.  (This should be
-        * okay since `struct resource_entry_disk' is declared as packed.) */
-       u64 size = entry->size;
+       ret = 0;
+       goto out;
 
-       disk_entry->size[0] = size >>  0;
-       disk_entry->size[1] = size >>  8;
-       disk_entry->size[2] = size >> 16;
-       disk_entry->size[3] = size >> 24;
-       disk_entry->size[4] = size >> 32;
-       disk_entry->size[5] = size >> 40;
-       disk_entry->size[6] = size >> 48;
-       disk_entry->flags = entry->flags;
-       disk_entry->offset = cpu_to_le64(entry->offset);
-       disk_entry->original_size = cpu_to_le64(entry->original_size);
+read_error:
+       ERROR_WITH_ERRNO("Read error");
+out:
+       if (tmp_buf_malloced)
+               FREE(tmp_buf);
+       return ret;
 }
 
-static int
+/*
+ * read_partial_wim_resource()-
+ *
+ * Read a range of data from a uncompressed or compressed resource in a WIM
+ * file.  Data is written into a buffer or fed into a callback function, as
+ * documented in read_resource_prefix().
+ *
+ * By default, this function provides the uncompressed data of the resource, and
+ * @size and @offset and interpreted relative to the uncompressed contents of
+ * the resource.  The behavior can be modified by any of the following flags:
+ *
+ * WIMLIB_READ_RESOURCE_FLAG_RAW_FULL:
+ *     Read @size bytes at @offset of the raw contents of the compressed
+ *     resource.  In the case of pipable resources, this excludes the stream
+ *     header.  Exclusive with WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS.
+ *
+ * WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS:
+ *     Read the raw compressed chunks of the compressed resource.  @size must
+ *     be the full uncompressed size, @offset must be 0, and @cb_chunk_size
+ *     must be the resource chunk size.
+ *
+ * Return values:
+ *     WIMLIB_ERR_SUCCESS (0)
+ *     WIMLIB_ERR_READ                   (errno set)
+ *     WIMLIB_ERR_UNEXPECTED_END_OF_FILE (errno set to 0)
+ *     WIMLIB_ERR_NOMEM                  (errno set to ENOMEM)
+ *     WIMLIB_ERR_DECOMPRESSION          (errno set to EINVAL)
+ *     WIMLIB_ERR_INVALID_CHUNK_SIZE    (errno set to EINVAL)
+ *
+ *     or other error code returned by the @cb function.
+ */
+int
 read_partial_wim_resource(const struct wim_lookup_table_entry *lte,
-                         u64 size,
-                         consume_data_callback_t cb,
-                         void *ctx_or_buf,
-                         int flags,
-                         u64 offset)
+                         u64 size, consume_data_callback_t cb,
+                         u32 cb_chunk_size,
+                         void *ctx_or_buf, int flags, u64 offset)
 {
-       WIMStruct *wim;
-       int in_fd;
+       struct filedes *in_fd;
        int ret;
 
+       /* Verify parameters.  */
        wimlib_assert(lte->resource_location == RESOURCE_IN_WIM);
-
-       wim = lte->wim;
-       in_fd = wim->in_fd;
-
-       if (lte->resource_entry.flags & WIM_RESHDR_FLAG_COMPRESSED &&
-           !(flags & WIMLIB_RESOURCE_FLAG_RAW))
-       {
-               ret = read_compressed_resource(in_fd,
-                                              lte->resource_entry.size,
-                                              lte->resource_entry.original_size,
-                                              lte->resource_entry.offset,
-                                              wimlib_get_compression_type(wim),
-                                              size,
-                                              offset,
-                                              cb,
-                                              ctx_or_buf);
+       in_fd = &lte->wim->in_fd;
+       if (cb)
+               wimlib_assert(is_power_of_2(cb_chunk_size));
+       if (flags & WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS) {
+               /* Raw chunks mode is subject to the restrictions noted.  */
+               wimlib_assert(!(flags & WIMLIB_READ_RESOURCE_FLAG_RAW_FULL));
+               wimlib_assert(cb_chunk_size == wim_resource_chunk_size(lte));
+               wimlib_assert(size == lte->resource_entry.original_size);
+               wimlib_assert(offset == 0);
+       } else if (flags & WIMLIB_READ_RESOURCE_FLAG_RAW_FULL) {
+               /* Raw full mode:  read must not overrun end of store size.  */
+               wimlib_assert(offset + size >= size &&
+                             offset + size <= lte->resource_entry.size);
        } else {
-               offset += lte->resource_entry.offset;
-               if (cb) {
-                       /* Send data to callback function */
-                       u8 buf[min(WIM_CHUNK_SIZE, size)];
-                       while (size) {
-                               size_t bytes_to_read = min(WIM_CHUNK_SIZE, size);
-                               size_t bytes_read = full_pread(in_fd, buf,
-                                                              bytes_to_read, offset);
-                               if (bytes_read != bytes_to_read)
-                                       goto read_error;
-                               ret = cb(buf, bytes_read, ctx_or_buf);
-                               if (ret)
-                                       goto out;
-                               size -= bytes_read;
-                               offset += bytes_read;
-                       }
-               } else {
-                       /* Send data directly to a buffer */
-                       if (full_pread(in_fd, ctx_or_buf, size, offset) != size)
-                               goto read_error;
-               }
-               ret = 0;
+               /* Normal mode:  read must not overrun end of original size.  */
+               wimlib_assert(offset + size >= size &&
+                             offset + size <= lte->resource_entry.original_size);
        }
-       goto out;
-read_error:
-       ERROR_WITH_ERRNO("Error reading data from WIM");
-       ret = WIMLIB_ERR_READ;
-out:
-       if (ret) {
-               if (errno == 0)
-                       errno = EIO;
+
+       DEBUG("Reading WIM resource: %"PRIu64" @ +%"PRIu64" "
+             "from %"PRIu64"(%"PRIu64") @ +%"PRIu64" "
+             "(readflags 0x%08x, resflags 0x%02x%s)",
+             size, offset,
+             lte->resource_entry.size,
+             lte->resource_entry.original_size,
+             lte->resource_entry.offset,
+             flags, lte->resource_entry.flags,
+             (lte->is_pipable ? ", pipable" : ""));
+
+       if ((flags & WIMLIB_READ_RESOURCE_FLAG_RAW_FULL) ||
+           !resource_is_compressed(&lte->resource_entry)) {
+               return read_raw_file_data(in_fd,
+                                         size,
+                                         cb,
+                                         cb_chunk_size,
+                                         ctx_or_buf,
+                                         offset + lte->resource_entry.offset);
+       } else {
+               return read_compressed_wim_resource(lte, size, cb,
+                                                   cb_chunk_size,
+                                                   ctx_or_buf, flags, offset);
        }
-       return ret;
 }
 
-
 int
 read_partial_wim_resource_into_buf(const struct wim_lookup_table_entry *lte,
                                   size_t size, u64 offset, void *buf)
 {
-       return read_partial_wim_resource(lte, size, NULL, buf, 0, offset);
+       return read_partial_wim_resource(lte, size, NULL, 0, buf, 0, offset);
 }
 
 static int
 read_wim_resource_prefix(const struct wim_lookup_table_entry *lte,
                         u64 size,
                         consume_data_callback_t cb,
+                        u32 cb_chunk_size,
                         void *ctx_or_buf,
                         int flags)
 {
-       return read_partial_wim_resource(lte, size, cb, ctx_or_buf, flags, 0);
+       return read_partial_wim_resource(lte, size, cb, cb_chunk_size,
+                                        ctx_or_buf, flags, 0);
 }
 
-
 #ifndef __WIN32__
+/* This function handles reading resource data that is located in an external
+ * file,  such as a file that has been added to the WIM image through execution
+ * of a wimlib_add_command.
+ *
+ * This assumes the file can be accessed using the standard POSIX open(),
+ * read(), and close().  On Windows this will not necessarily be the case (since
+ * the file may need FILE_FLAG_BACKUP_SEMANTICS to be opened, or the file may be
+ * encrypted), so Windows uses its own code for its equivalent case.
+ */
 static int
 read_file_on_disk_prefix(const struct wim_lookup_table_entry *lte,
                         u64 size,
                         consume_data_callback_t cb,
+                        u32 cb_chunk_size,
                         void *ctx_or_buf,
                         int _ignored_flags)
 {
-       const tchar *filename = lte->file_on_disk;
        int ret;
-       int fd;
-       size_t bytes_read;
+       int raw_fd;
+       struct filedes fd;
+
+       wimlib_assert(size <= wim_resource_size(lte));
+       DEBUG("Reading %"PRIu64" bytes from \"%"TS"\"", size, lte->file_on_disk);
 
-       fd = open(filename, O_RDONLY);
-       if (fd < 0) {
-               ERROR_WITH_ERRNO("Can't open \"%"TS"\"", filename);
+       raw_fd = open(lte->file_on_disk, O_BINARY | O_RDONLY);
+       if (raw_fd < 0) {
+               ERROR_WITH_ERRNO("Can't open \"%"TS"\"", lte->file_on_disk);
                return WIMLIB_ERR_OPEN;
        }
-       if (cb) {
-               /* Send data to callback function */
-               u8 buf[min(WIM_CHUNK_SIZE, size)];
-               size_t bytes_to_read;
-               while (size) {
-                       bytes_to_read = min(WIM_CHUNK_SIZE, size);
-                       bytes_read = full_read(fd, buf, bytes_to_read);
-                       if (bytes_read != bytes_to_read)
-                               goto read_error;
-                       ret = cb(buf, bytes_read, ctx_or_buf);
-                       if (ret)
-                               goto out_close;
-                       size -= bytes_read;
-               }
-       } else {
-               /* Send data directly to a buffer */
-               bytes_read = full_read(fd, ctx_or_buf, size);
-               if (bytes_read != size)
-                       goto read_error;
-       }
-       ret = 0;
-       goto out_close;
-read_error:
-       ERROR_WITH_ERRNO("Error reading \"%"TS"\"", filename);
-       ret = WIMLIB_ERR_READ;
-out_close:
-       close(fd);
+       filedes_init(&fd, raw_fd);
+       ret = read_raw_file_data(&fd, size, cb, cb_chunk_size, ctx_or_buf, 0);
+       filedes_close(&fd);
        return ret;
 }
 #endif /* !__WIN32__ */
 
+/* This function handles the trivial case of reading resource data that is, in
+ * fact, already located in an in-memory buffer.  */
 static int
 read_buffer_prefix(const struct wim_lookup_table_entry *lte,
                   u64 size, consume_data_callback_t cb,
+                  u32 cb_chunk_size,
                   void *ctx_or_buf, int _ignored_flags)
 {
-       const void *inbuf = lte->attached_buffer;
-       int ret;
+       wimlib_assert(size <= wim_resource_size(lte));
 
        if (cb) {
-               while (size) {
-                       size_t chunk_size = min(WIM_CHUNK_SIZE, size);
-                       ret = cb(inbuf, chunk_size, ctx_or_buf);
+               /* Feed the data into the callback function in
+                * appropriately-sized chunks.  */
+               int ret;
+               u32 chunk_size;
+
+               for (u64 offset = 0; offset < size; offset += chunk_size) {
+                       chunk_size = min(cb_chunk_size, size - offset);
+                       ret = cb((const u8*)lte->attached_buffer + offset,
+                                chunk_size, ctx_or_buf);
                        if (ret)
                                return ret;
-                       size -= chunk_size;
-                       inbuf += chunk_size;
                }
        } else {
-               memcpy(ctx_or_buf, inbuf, size);
+               /* Copy the data directly into the specified buffer.  */
+               memcpy(ctx_or_buf, lte->attached_buffer, size);
        }
        return 0;
 }
@@ -595,34 +746,49 @@ read_buffer_prefix(const struct wim_lookup_table_entry *lte,
 typedef int (*read_resource_prefix_handler_t)(const struct wim_lookup_table_entry *lte,
                                              u64 size,
                                              consume_data_callback_t cb,
+                                             u32 cb_chunk_size,
                                              void *ctx_or_buf,
                                              int flags);
 
 /*
- * Read the first @size bytes from a generic "resource", which may be located in
- * the WIM (compressed or uncompressed), in an external file, or directly in an
- * in-memory buffer.
+ * read_resource_prefix()-
+ *
+ * Reads the first @size bytes from a generic "resource", which may be located
+ * in any one of several locations, such as in a WIM file (compressed or
+ * uncompressed), in an external file, or directly in an in-memory buffer.
+ *
+ * This function feeds the data either to a callback function (@cb != NULL,
+ * passing it @ctx_or_buf), or write it directly into a buffer (@cb == NULL,
+ * @ctx_or_buf specifies the buffer, which must have room for at least @size
+ * bytes).
  *
- * Feed the data either to a callback function (cb != NULL, passing it
- * ctx_or_buf), or write it directly into a buffer (cb == NULL, ctx_or_buf
- * specifies the buffer, which must have room for @size bytes).
+ * When (@cb != NULL), @cb_chunk_size specifies the maximum size of data chunks
+ * to feed the callback function.  @cb_chunk_size must be positive, and if the
+ * resource is in a WIM file, must be a power of 2.  All chunks, except possibly
+ * the last one, will be this size.  If (@cb == NULL), @cb_chunk_size is
+ * ignored.
  *
- * When using a callback function, it is called with chunks up to 32768 bytes in
- * size until the resource is exhausted.
+ * If the resource is located in a WIM file, @flags can be set as documented in
+ * read_partial_wim_resource().  Otherwise @flags are ignored.
  *
- * If the resource is located in a WIM file, @flags can be:
- *   * WIMLIB_RESOURCE_FLAG_RAW if the raw compressed data is to be supplied
- *     instead of the uncompressed data.
- * Otherwise, the @flags are ignored.
+ * Returns 0 on success; nonzero on error.  A nonzero value will be returned if
+ * the resource data cannot be successfully read (for a number of different
+ * reasons, depending on the resource location), or if a callback function was
+ * specified and it returned nonzero.
  */
 int
 read_resource_prefix(const struct wim_lookup_table_entry *lte,
-                    u64 size, consume_data_callback_t cb, void *ctx_or_buf,
-                    int flags)
+                    u64 size, consume_data_callback_t cb, u32 cb_chunk_size,
+                    void *ctx_or_buf, int flags)
 {
+       /* This function merely verifies several preconditions, then passes
+        * control to an appropriate function for understanding each possible
+        * resource location.  */
        static const read_resource_prefix_handler_t handlers[] = {
                [RESOURCE_IN_WIM]             = read_wim_resource_prefix,
-       #ifndef __WIN32__
+       #ifdef __WIN32__
+               [RESOURCE_IN_FILE_ON_DISK]    = read_win32_file_prefix,
+       #else
                [RESOURCE_IN_FILE_ON_DISK]    = read_file_on_disk_prefix,
        #endif
                [RESOURCE_IN_ATTACHED_BUFFER] = read_buffer_prefix,
@@ -633,20 +799,79 @@ read_resource_prefix(const struct wim_lookup_table_entry *lte,
                [RESOURCE_IN_NTFS_VOLUME]     = read_ntfs_file_prefix,
        #endif
        #ifdef __WIN32__
-               [RESOURCE_WIN32]              = read_win32_file_prefix,
                [RESOURCE_WIN32_ENCRYPTED]    = read_win32_encrypted_file_prefix,
        #endif
        };
        wimlib_assert(lte->resource_location < ARRAY_LEN(handlers)
                      && handlers[lte->resource_location] != NULL);
-       return handlers[lte->resource_location](lte, size, cb, ctx_or_buf, flags);
+       wimlib_assert(cb == NULL || cb_chunk_size > 0);
+       return handlers[lte->resource_location](lte, size, cb, cb_chunk_size,
+                                               ctx_or_buf, flags);
 }
 
+/* Read the full uncompressed data of the specified resource into the specified
+ * buffer, which must have space for at least lte->resource_entry.original_size
+ * bytes.  */
 int
 read_full_resource_into_buf(const struct wim_lookup_table_entry *lte,
                            void *buf)
 {
-       return read_resource_prefix(lte, wim_resource_size(lte), NULL, buf, 0);
+       return read_resource_prefix(lte, wim_resource_size(lte),
+                                   NULL, 0, buf, 0);
+}
+
+/* Read the full uncompressed data of the specified resource.  A buffer
+ * sufficient to hold the data is allocated and returned in @buf_ret.  */
+int
+read_full_resource_into_alloc_buf(const struct wim_lookup_table_entry *lte,
+                                 void **buf_ret)
+{
+       int ret;
+       void *buf;
+
+       if ((size_t)lte->resource_entry.original_size !=
+           lte->resource_entry.original_size)
+       {
+               ERROR("Can't read %"PRIu64" byte resource into "
+                     "memory", lte->resource_entry.original_size);
+               return WIMLIB_ERR_NOMEM;
+       }
+
+       buf = MALLOC(lte->resource_entry.original_size);
+       if (buf == NULL)
+               return WIMLIB_ERR_NOMEM;
+
+       ret = read_full_resource_into_buf(lte, buf);
+       if (ret) {
+               FREE(buf);
+               return ret;
+       }
+
+       *buf_ret = buf;
+       return 0;
+}
+
+/* Retrieve the full uncompressed data of the specified WIM resource, provided
+ * as a raw `struct resource_entry'.  */
+int
+res_entry_to_data(const struct resource_entry *res_entry,
+                 WIMStruct *wim, void **buf_ret)
+{
+       int ret;
+       struct wim_lookup_table_entry *lte;
+
+       lte = new_lookup_table_entry();
+       if (lte == NULL)
+               return WIMLIB_ERR_NOMEM;
+
+       copy_resource_entry(&lte->resource_entry, res_entry);
+       lte->unhashed = 1;
+       lte->part_number = wim->hdr.part_number;
+       lte_init_wim(lte, wim);
+
+       ret = read_full_resource_into_alloc_buf(lte, buf_ret);
+       free_lookup_table_entry(lte);
+       return ret;
 }
 
 struct extract_ctx {
@@ -665,12 +890,12 @@ extract_chunk_sha1_wrapper(const void *chunk, size_t chunk_size,
        return ctx->extract_chunk(chunk, chunk_size, ctx->extract_chunk_arg);
 }
 
-/* Extracts the first @size bytes of a WIM resource to somewhere.  In the
- * process, the SHA1 message digest of the resource is checked if the full
+/* Extracts the first @size bytes of a resource to somewhere.  In the process,
+ * the SHA1 message digest of the uncompressed resource is checked if the full
  * resource is being extracted.
  *
- * @extract_chunk is a function that is called to extract each chunk of the
- * resource. */
+ * @extract_chunk is a function that will be called to extract each chunk of the
+ * resource.  */
 int
 extract_wim_resource(const struct wim_lookup_table_entry *lte,
                     u64 size,
@@ -686,24 +911,26 @@ extract_wim_resource(const struct wim_lookup_table_entry *lte,
                sha1_init(&ctx.sha_ctx);
                ret = read_resource_prefix(lte, size,
                                           extract_chunk_sha1_wrapper,
+                                          wim_resource_chunk_size(lte),
                                           &ctx, 0);
                if (ret == 0) {
                        u8 hash[SHA1_HASH_SIZE];
                        sha1_final(hash, &ctx.sha_ctx);
                        if (!hashes_equal(hash, lte->hash)) {
-                       #ifdef ENABLE_ERROR_MESSAGES
-                               ERROR("Invalid SHA1 message digest "
-                                     "on the following WIM resource:");
-                               print_lookup_table_entry(lte, stderr);
-                               if (lte->resource_location == RESOURCE_IN_WIM)
-                                       ERROR("The WIM file appears to be corrupt!");
-                       #endif
+                               if (wimlib_print_errors) {
+                                       ERROR("Invalid SHA1 message digest "
+                                             "on the following WIM resource:");
+                                       print_lookup_table_entry(lte, stderr);
+                                       if (lte->resource_location == RESOURCE_IN_WIM)
+                                               ERROR("The WIM file appears to be corrupt!");
+                               }
                                ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
                        }
                }
        } else {
                /* Don't do SHA1 */
                ret = read_resource_prefix(lte, size, extract_chunk,
+                                          wim_resource_chunk_size(lte),
                                           extract_chunk_arg, 0);
        }
        return ret;
@@ -712,21 +939,21 @@ extract_wim_resource(const struct wim_lookup_table_entry *lte,
 static int
 extract_wim_chunk_to_fd(const void *buf, size_t len, void *_fd_p)
 {
-       int fd = *(int*)_fd_p;
-       ssize_t ret = full_write(fd, buf, len);
-       if (ret < len) {
+       struct filedes *fd = _fd_p;
+       int ret = full_write(fd, buf, len);
+       if (ret)
                ERROR_WITH_ERRNO("Error writing to file descriptor");
-               return WIMLIB_ERR_WRITE;
-       } else {
-               return 0;
-       }
+       return ret;
 }
 
+/* Extract the first @size bytes of the specified resource to the specified file
+ * descriptor.  If @size is the full size of the resource, its SHA1 message
+ * digest is also checked.  */
 int
 extract_wim_resource_to_fd(const struct wim_lookup_table_entry *lte,
-                          int fd, u64 size)
+                          struct filedes *fd, u64 size)
 {
-       return extract_wim_resource(lte, size, extract_wim_chunk_to_fd, &fd);
+       return extract_wim_resource(lte, size, extract_wim_chunk_to_fd, fd);
 }
 
 
@@ -737,7 +964,7 @@ sha1_chunk(const void *buf, size_t len, void *ctx)
        return 0;
 }
 
-/* Calculate the SHA1 message digest of a stream. */
+/* Calculate the SHA1 message digest of a resource, storing it in @lte->hash.  */
 int
 sha1_resource(struct wim_lookup_table_entry *lte)
 {
@@ -746,34 +973,69 @@ sha1_resource(struct wim_lookup_table_entry *lte)
 
        sha1_init(&sha_ctx);
        ret = read_resource_prefix(lte, wim_resource_size(lte),
-                                  sha1_chunk, &sha_ctx, 0);
+                                  sha1_chunk, wim_resource_chunk_size(lte),
+                                  &sha_ctx, 0);
        if (ret == 0)
                sha1_final(lte->hash, &sha_ctx);
+
        return ret;
 }
 
-/*
- * Copies the file resource specified by the lookup table entry @lte from the
- * input WIM to the output WIM that has its FILE * given by
- * ((WIMStruct*)wim)->out_fp.
- *
- * The output_resource_entry, out_refcnt, and part_number fields of @lte are
- * updated.
- *
- * (This function is confusing and should be refactored somehow.)
- */
-int
-copy_resource(struct wim_lookup_table_entry *lte, void *wim)
+/* Translates a WIM resource entry from the on-disk format into an in-memory
+ * format.  */
+void
+get_resource_entry(const struct resource_entry_disk *disk_entry,
+                  struct resource_entry *entry)
 {
-       WIMStruct *w = wim;
-       int ret;
+       /* Note: disk_entry may not be 8 byte aligned--- in that case, the
+        * offset and original_size members will be unaligned.  (This is okay
+        * since `struct resource_entry_disk' is declared as packed.)  */
 
-       ret = write_wim_resource(lte, w->out_fd,
-                                wim_resource_compression_type(lte),
-                                &lte->output_resource_entry, 0);
-       if (ret == 0) {
-               lte->out_refcnt = lte->refcnt;
-               lte->part_number = w->hdr.part_number;
+       /* Read the size and flags into a bitfield portably... */
+       entry->size = (((u64)disk_entry->size[0] <<  0) |
+                      ((u64)disk_entry->size[1] <<  8) |
+                      ((u64)disk_entry->size[2] << 16) |
+                      ((u64)disk_entry->size[3] << 24) |
+                      ((u64)disk_entry->size[4] << 32) |
+                      ((u64)disk_entry->size[5] << 40) |
+                      ((u64)disk_entry->size[6] << 48));
+       entry->flags = disk_entry->flags;
+       entry->offset = le64_to_cpu(disk_entry->offset);
+       entry->original_size = le64_to_cpu(disk_entry->original_size);
+
+       /* offset and original_size are truncated to 62 bits to avoid possible
+        * overflows, when converting to a signed 64-bit integer (off_t) or when
+        * adding size or original_size.  This is okay since no one would ever
+        * actually have a WIM bigger than 4611686018427387903 bytes... */
+       if (entry->offset & 0xc000000000000000ULL) {
+               WARNING("Truncating offset in resource entry");
+               entry->offset &= 0x3fffffffffffffffULL;
        }
-       return ret;
+       if (entry->original_size & 0xc000000000000000ULL) {
+               WARNING("Truncating original_size in resource entry");
+               entry->original_size &= 0x3fffffffffffffffULL;
+       }
+}
+
+/* Translates a WIM resource entry from an in-memory format into the on-disk
+ * format. */
+void
+put_resource_entry(const struct resource_entry *entry,
+                  struct resource_entry_disk *disk_entry)
+{
+       /* Note: disk_entry may not be 8 byte aligned--- in that case, the
+        * offset and original_size members will be unaligned.  (This is okay
+        * since `struct resource_entry_disk' is declared as packed.)  */
+       u64 size = entry->size;
+
+       disk_entry->size[0] = size >>  0;
+       disk_entry->size[1] = size >>  8;
+       disk_entry->size[2] = size >> 16;
+       disk_entry->size[3] = size >> 24;
+       disk_entry->size[4] = size >> 32;
+       disk_entry->size[5] = size >> 40;
+       disk_entry->size[6] = size >> 48;
+       disk_entry->flags = entry->flags;
+       disk_entry->offset = cpu_to_le64(entry->offset);
+       disk_entry->original_size = cpu_to_le64(entry->original_size);
 }