* wimlib; if not, see http://www.gnu.org/licenses/.
*/
-#include "wimlib_internal.h"
-#include "dentry.h"
-#include "lookup_table.h"
-#include "buffer_io.h"
-#include "lzx.h"
-#include "xpress.h"
-#include "sha1.h"
+#ifdef HAVE_CONFIG_H
+# include "config.h"
+#endif
+
+#include "wimlib.h"
+#include "wimlib/dentry.h"
+#include "wimlib/endianness.h"
+#include "wimlib/error.h"
+#include "wimlib/file_io.h"
+#include "wimlib/lookup_table.h"
+#include "wimlib/resource.h"
+#include "wimlib/sha1.h"
#ifdef __WIN32__
-# include "win32.h"
+/* for read_win32_file_prefix(), read_win32_encrypted_file_prefix() */
+# include "wimlib/win32.h"
+#endif
+
+#ifdef WITH_NTFS_3G
+/* for read_ntfs_file_prefix() */
+# include "wimlib/ntfs_3g.h"
#endif
+#ifdef HAVE_ALLOCA_H
+# include <alloca.h>
+#endif
#include <errno.h>
+#include <fcntl.h>
#include <stdarg.h>
#include <stdlib.h>
#include <unistd.h>
-#ifdef WITH_NTFS_3G
-# include <time.h>
-# include <ntfs-3g/attrib.h>
-# include <ntfs-3g/inode.h>
-# include <ntfs-3g/dir.h>
-#endif
+/*
+ * Compressed resources
+ *
+ * A compressed resource in a WIM consists of a number of consecutive LZX or
+ * XPRESS-compressed chunks, each of which decompresses to 32768 bytes of data,
+ * 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 (less than or
+ * equal to 32768). Otherwise the details are the same.
+ */
-#if defined(__WIN32__) && !defined(INVALID_HANDLE_VALUE)
-# define INVALID_HANDLE_VALUE ((HANDLE)(-1))
-#endif
+static int 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;
+ }
+}
/*
- * Reads all or part of a compressed resource into an in-memory buffer.
+ * read_compressed_resource()-
*
- * @fp: The FILE* for the WIM file.
- * @resource_compressed_size: The compressed size of the resource.
- * @resource_uncompressed_size: The uncompressed size of the resource.
- * @resource_offset: The offset of the start of the resource from
- * the start of the stream @fp.
- * @resource_ctype: The compression type of the resource.
- * @len: The number of bytes of uncompressed data to read from
- * the resource.
- * @offset: The offset of the bytes to read within the uncompressed
- * resource.
- * @contents_len: An array into which the uncompressed data is written.
- * It must be at least @len bytes long.
+ * Read data from a compressed resource being read from a seekable WIM file.
+ * The resource may be either pipable or non-pipable.
*
- * Returns zero on success, nonzero on failure.
+ * @flags may be:
+ *
+ * 0:
+ * Just do a normal read, decompressing the data if necessary.
+ *
+ * WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS:
+ * Read the raw contents of the compressed chunks of the compressed
+ * resource. For pipable resources, this does *not* include the chunk
+ * headers. If a callback function is being used, it will be called once
+ * for each compressed chunk. For non-pipable resources, this mode
+ * excludes the chunk table. For pipable resources, this mode excludes the
+ * stream and chunk headers.
*/
static int
-read_compressed_resource(FILE *fp, u64 resource_compressed_size,
- u64 resource_uncompressed_size,
- u64 resource_offset, int resource_ctype,
- u64 len, u64 offset, void *contents_ret)
+read_compressed_resource(const struct wim_lookup_table_entry * const lte,
+ u64 size, const consume_data_callback_t cb,
+ const u32 in_chunk_size, void * const ctx_or_buf,
+ const int flags, const u64 offset)
{
+ int ret;
- DEBUG2("comp size = %"PRIu64", uncomp size = %"PRIu64", "
- "res offset = %"PRIu64"",
- resource_compressed_size,
- resource_uncompressed_size,
- resource_offset);
- DEBUG2("resource_ctype = %"TS", len = %"PRIu64", offset = %"PRIu64"",
- wimlib_get_compression_type_string(resource_ctype), len, offset);
- /* Trivial case */
- if (len == 0)
- return 0;
+ const u32 orig_chunk_size = wim_resource_chunk_size(lte);
+ const u32 orig_chunk_order = bsr32(orig_chunk_size);
+
+ wimlib_assert(is_power_of_2(orig_chunk_size));
+ wimlib_assert(cb == NULL || is_power_of_2(in_chunk_size));
+
+ /* Currently, reading raw compressed chunks is only guaranteed to work
+ * correctly when the full resource is requested. Furthermore, in such
+ * cases the requested size is specified as the compressed size, but
+ * here we change it to an uncompressed size to avoid confusing the rest
+ * of this function. */
+ if (flags & WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS) {
+ wimlib_assert(offset == 0);
+ wimlib_assert(size == lte->resource_entry.size);
+ wimlib_assert(wim_resource_chunk_size(lte) == in_chunk_size);
+ size = wim_resource_size(lte);
+ }
- int (*decompress)(const void *, unsigned, void *, unsigned);
- /* Set the appropriate decompress function. */
- if (resource_ctype == WIMLIB_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.
- */
+ wimlib_assert(offset + size <= wim_resource_size(lte));
- /* 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$'s documentation includes a picture that shows the first
- * chunk starting right after the chunk entry table, labeled as offset
- * 0x10. However, in the actual file format, the offset is measured
- * from the end of the chunk entry table, so the first chunk has an
- * offset of 0. */
- if (start_chunk == 0)
- chunk_offsets[0] = 0;
+ /* Handle the trivial case. */
+ if (size == 0)
+ return 0;
- /* According to M$'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_WITH_ERRNO("Failed to seek to byte %"PRIu64" to read "
- "chunk table of compressed resource",
- file_offset_of_needed_chunk_entries);
- return WIMLIB_ERR_READ;
- }
+ 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;
+ const size_t stack_max = 32768;
- /* Number of bytes we need to read from the chunk table. */
- size_t size = num_needed_chunk_entries * chunk_entry_size;
+ /* Get the file descriptor for the WIM. */
+ struct filedes * const in_fd = <e->wim->in_fd;
- u8 chunk_tab_buf[size];
+ /* Calculate the number of chunks the resource is divided into. */
+ const u64 num_chunks = wim_resource_chunks(lte);
- if (fread(chunk_tab_buf, 1, size, fp) != size)
- goto err;
+ /* 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;
- /* Now fill in chunk_offsets from the entries we have read in
- * chunk_tab_buf. */
+ /* Calculate the 0-based index of the chunk at which the read starts.
+ */
+ const u64 start_chunk = offset >> orig_chunk_order;
+
+ /* 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 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);
+
+ /* Set the size of each chunk table entry based on the resource's
+ * uncompressed size. */
+ const u64 chunk_entry_size = (wim_resource_size(lte) > ((u64)1 << 32)) ? 8 : 4;
+
+ /* Calculate the size, in bytes, of the full chunk table. */
+ const u64 chunk_table_size = num_chunk_entries * chunk_entry_size;
+
+ /* 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 (num_alloc_chunk_entries <= stack_max) {
+ chunk_offsets = alloca(num_alloc_chunk_entries * sizeof(u64));
+ chunk_offsets_malloced = false;
+ } else {
+ chunk_offsets = MALLOC(num_alloc_chunk_entries * sizeof(u64));
+ if (!chunk_offsets) {
+ ERROR("Failed to allocate chunk table "
+ "with %"PRIu64" entries", num_alloc_chunk_entries);
+ return WIMLIB_ERR_NOMEM;
+ }
+ chunk_offsets_malloced = true;
+ }
- u64 *chunk_tab_p = chunk_offsets;
+ /* Set the implicit offset of the first chunk if it's included in the
+ * needed chunks. */
if (start_chunk == 0)
- chunk_tab_p++;
+ chunk_offsets[0] = 0;
- if (chunk_entry_size == 4) {
- u32 *entries = (u32*)chunk_tab_buf;
- while (num_needed_chunk_entries--)
- *chunk_tab_p++ = le32_to_cpu(*entries++);
- } else {
- u64 *entries = (u64*)chunk_tab_buf;
- while (num_needed_chunk_entries--)
- *chunk_tab_p++ = le64_to_cpu(*entries++);
+ /* 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;
+ if ((u64)chunk_table_needed_size !=
+ num_needed_chunk_entries * chunk_entry_size)
+ {
+ ERROR("Compressed read request too large to fit into memory!");
+ ret = WIMLIB_ERR_NOMEM;
+ goto out_free_memory;
}
- /* Done with the chunk table now. We must now seek to the first chunk
- * that is needed for the read. */
+ /* 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;
- u64 file_offset_of_first_needed_chunk = resource_offset +
- chunk_table_size + chunk_offsets[0];
- if (fseeko(fp, file_offset_of_first_needed_chunk, SEEK_SET) != 0) {
- ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" to read "
- "first chunk of compressed resource",
- file_offset_of_first_needed_chunk);
- return WIMLIB_ERR_READ;
+ /* 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]);
+ }
}
- /* Pointer to current position in the output buffer for uncompressed
- * data. */
- u8 *out_p = contents_ret;
+ /* Calculate file offset of the first chunk that needs to be read.
+ * Note: if the resource is pipable, the entries in the chunk table do
+ * *not* include the chunk headers. */
+ u64 cur_read_offset = lte->resource_entry.offset + chunk_offsets[0];
+ if (!lte->is_pipable)
+ cur_read_offset += chunk_table_size;
+ else
+ cur_read_offset += start_chunk * sizeof(struct pwm_chunk_hdr);
+
+ /* 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(in_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) {
+ ret = WIMLIB_ERR_NOMEM;
+ goto out_free_memory;
+ }
+ out_buf_malloced = true;
+ }
+ } else {
+ out_buf_size = size;
+ out_buf = ctx_or_buf;
+ }
+ out_buf_end = out_buf + out_buf_size;
+ out_p = out_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. */
- u8 compressed_buf[WIM_CHUNK_SIZE - 1];
+ /* 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) {
+ ret = WIMLIB_ERR_NOMEM;
+ goto out_free_memory;
+ }
+ compressed_buf_malloced = true;
+ }
+ }
+ /* Allocate yet another temporary buffer, this one for reading partial
+ * chunks. */
+ 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) {
+ ret = WIMLIB_ERR_NOMEM;
+ goto out_free_memory;
+ }
+ tmp_buf_malloced = true;
+ }
+ }
- /* Decompress all the chunks. */
+ /* 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 = start_chunk; i <= end_chunk; i++) {
- DEBUG2("Chunk %"PRIu64" (start %"PRIu64", end %"PRIu64").",
- i, start_chunk, end_chunk);
+ /* If the resource is pipable, skip the chunk header. */
+ if (lte->is_pipable)
+ cur_read_offset += sizeof(struct pwm_chunk_hdr);
/* Calculate the sizes of the compressed chunk and of the
- * uncompressed chunk. */
- unsigned compressed_chunk_size;
- unsigned uncompressed_chunk_size;
+ * uncompressed chunk. */
+ u32 compressed_chunk_size;
+ u32 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. */
+ /* Not the last chunk. Compressed size is given by
+ * difference of chunk table entries; uncompressed size
+ * is always the WIM chunk size. */
compressed_chunk_size = chunk_offsets[i + 1 - start_chunk] -
chunk_offsets[i - start_chunk];
- uncompressed_chunk_size = WIM_CHUNK_SIZE;
+ uncompressed_chunk_size = orig_chunk_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 -
+ /* Last chunk. Compressed size is the remaining size in
+ * the compressed resource; uncompressed size is the
+ * remaining size in the uncompressed resource. */
+ compressed_chunk_size = lte->resource_entry.size -
chunk_table_size -
chunk_offsets[i - start_chunk];
+ if (lte->is_pipable)
+ compressed_chunk_size -= num_chunks *
+ sizeof(struct pwm_chunk_hdr);
- uncompressed_chunk_size = resource_uncompressed_size %
- WIM_CHUNK_SIZE;
+ if ((wim_resource_size(lte) & (orig_chunk_size - 1)) == 0)
+ uncompressed_chunk_size = orig_chunk_size;
+ else
+ uncompressed_chunk_size = wim_resource_size(lte) &
+ (orig_chunk_size - 1);
+ }
+
+ /* Calculate how much of this chunk needs to be read. */
+
+ u32 partial_chunk_size;
+ u32 start_offset = 0;
+ u32 end_offset = orig_chunk_size - 1;
- /* 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 (flags & WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS) {
+ partial_chunk_size = compressed_chunk_size;
+ } else {
+ if (i == start_chunk)
+ start_offset = start_offset_in_chunk;
+
+ if (i == end_chunk)
+ end_offset = end_offset_in_chunk;
+
+ partial_chunk_size = end_offset + 1 - start_offset;
}
- DEBUG2("compressed_chunk_size = %u, "
- "uncompressed_chunk_size = %u",
- compressed_chunk_size, uncompressed_chunk_size);
-
-
- /* 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;
-
- u64 partial_chunk_size = end_offset + 1 - start_offset;
- bool is_partial_chunk = (partial_chunk_size !=
- uncompressed_chunk_size);
-
- DEBUG2("start_offset = %"PRIu64", end_offset = %"PRIu64"",
- start_offset, end_offset);
- DEBUG2("partial_chunk_size = %"PRIu64"", partial_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) {
- /* Probably an uncompressed chunk */
-
- if (start_offset != 0) {
- if (fseeko(fp, start_offset, SEEK_CUR) != 0) {
- ERROR_WITH_ERRNO("Uncompressed partial "
- "chunk fseek() error");
- return WIMLIB_ERR_READ;
- }
- }
- if (fread(out_p, 1, partial_chunk_size, fp) !=
- partial_chunk_size)
- goto err;
+ if (compressed_chunk_size == uncompressed_chunk_size ||
+ (flags & WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS))
+ {
+ /* Chunk stored uncompressed, or reading raw chunk data. */
+ ret = full_pread(in_fd,
+ out_p,
+ partial_chunk_size,
+ cur_read_offset + start_offset);
+ if (ret)
+ goto read_error;
} else {
- /* Compressed chunk */
- int ret;
+ /* Compressed chunk and not doing raw read. */
+ u8 *target;
- /* Read the compressed data into compressed_buf. */
- if (fread(compressed_buf, 1, compressed_chunk_size,
- fp) != compressed_chunk_size)
- goto err;
+ /* Read the compressed data into compressed_buf. */
+ ret = full_pread(in_fd,
+ compressed_buf,
+ compressed_chunk_size,
+ cur_read_offset);
+ if (ret)
+ goto read_error;
/* For partial chunks we must buffer the uncompressed
- * data because we don't need all of it. */
- if (is_partial_chunk) {
- u8 uncompressed_buf[uncompressed_chunk_size];
-
- ret = decompress(compressed_buf,
- compressed_chunk_size,
- uncompressed_buf,
- uncompressed_chunk_size);
- if (ret != 0)
- return WIMLIB_ERR_DECOMPRESSION;
- 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 != 0)
- return WIMLIB_ERR_DECOMPRESSION;
+ * data because we don't need all of it. */
+ if (partial_chunk_size == uncompressed_chunk_size)
+ target = out_p;
+ else
+ target = tmp_buf;
+
+ /* Decompress the chunk. */
+ ret = decompress(compressed_buf,
+ compressed_chunk_size,
+ target,
+ uncompressed_chunk_size,
+ wim_resource_compression_type(lte),
+ orig_chunk_size);
+ if (ret) {
+ ERROR("Failed to decompress data.");
+ ret = WIMLIB_ERR_DECOMPRESSION;
+ errno = EINVAL;
+ goto out_free_memory;
}
+ if (partial_chunk_size != uncompressed_chunk_size)
+ memcpy(out_p, tmp_buf + start_offset,
+ partial_chunk_size);
}
- /* Advance the pointer into the uncompressed output data by the
- * number of uncompressed bytes that were written. */
out_p += partial_chunk_size;
+
+ if (cb) {
+ /* Feed the data to the callback function. */
+ wimlib_assert(offset == 0);
+
+ 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(in_chunk_size, out_p - p);
+ ret = cb(p, bytes_sent, ctx_or_buf);
+ if (ret)
+ goto out_free_memory;
+ }
+ out_p = out_buf;
+ }
+ }
+ cur_read_offset += compressed_chunk_size;
}
- return 0;
+ ret = 0;
+out_free_memory:
+ if (chunk_offsets_malloced)
+ FREE(chunk_offsets);
+ if (out_buf_malloced)
+ FREE(out_buf);
+ if (compressed_buf_malloced)
+ FREE(compressed_buf);
+ if (tmp_buf_malloced)
+ FREE(tmp_buf);
+ return ret;
-err:
- if (feof(fp))
- ERROR("Unexpected EOF in compressed file resource");
- else
- ERROR_WITH_ERRNO("Error reading compressed file resource");
- return WIMLIB_ERR_READ;
+read_error:
+ ERROR_WITH_ERRNO("Error reading compressed file resource");
+ goto out_free_memory;
}
-/*
- * Reads uncompressed data from an open file stream.
- */
-int
-read_uncompressed_resource(FILE *fp, u64 offset, u64 len, void *contents_ret)
+/* Skip over the chunk table at the end of pipable, compressed resource being
+ * read from a pipe. */
+static int
+skip_chunk_table(const struct wim_lookup_table_entry *lte,
+ struct filedes *in_fd)
{
- if (fseeko(fp, offset, SEEK_SET) != 0) {
- ERROR("Failed to seek to byte %"PRIu64" of input file "
- "to read uncompressed resource (len = %"PRIu64")",
- offset, len);
- return WIMLIB_ERR_READ;
+ u64 num_chunk_entries = wim_resource_chunks(lte) - 1;
+ u64 chunk_entry_size = (wim_resource_size(lte) > ((u64)1 << 32)) ? 8 : 4;
+ u64 chunk_table_size = num_chunk_entries * chunk_entry_size;
+ int ret;
+
+ if (num_chunk_entries != 0) {
+ u8 dummy;
+ ret = full_pread(in_fd, &dummy, 1,
+ in_fd->offset + chunk_table_size - 1);
+ if (ret)
+ return ret;
}
- if (fread(contents_ret, 1, len, fp) != len) {
- if (feof(fp)) {
- ERROR("Unexpected EOF in uncompressed file resource");
+ return 0;
+}
+
+/* Read and decompress data from a compressed, pipable resource being read from
+ * a pipe. */
+static int
+read_pipable_resource(const struct wim_lookup_table_entry *lte,
+ u64 size, consume_data_callback_t cb,
+ u32 in_chunk_size, void *ctx_or_buf,
+ int flags, u64 offset)
+{
+ struct filedes *in_fd;
+ int ret;
+ const u32 orig_chunk_size = wim_resource_chunk_size(lte);
+ u8 cchunk[orig_chunk_size - 1];
+
+ size_t out_buf_size;
+ u8 *out_buf, *out_buf_end, *out_p;
+ if (cb) {
+ out_buf_size = max(in_chunk_size, orig_chunk_size);
+ out_buf = alloca(out_buf_size);
+ } else {
+ out_buf_size = size;
+ out_buf = ctx_or_buf;
+ }
+ out_buf_end = out_buf + out_buf_size;
+ out_p = out_buf;
+
+ /* Get pointers to appropriate decompression function and the input file
+ * descriptor. */
+ in_fd = <e->wim->in_fd;
+
+ /* This function currently assumes the entire resource is being read at
+ * once and that the raw compressed data isn't being requested. This is
+ * based on the fact that this function currently only gets called
+ * during the operation of wimlib_extract_image_from_pipe(). */
+ wimlib_assert(!(flags & WIMLIB_READ_RESOURCE_FLAG_RAW));
+ wimlib_assert(offset == 0);
+ wimlib_assert(size == wim_resource_size(lte));
+ wimlib_assert(in_fd->offset == lte->resource_entry.offset);
+
+ u32 chunk_usize;
+ for (offset = 0; offset < size; offset += chunk_usize) {
+ struct pwm_chunk_hdr chunk_hdr;
+ u32 chunk_csize;
+
+ /* Calculate uncompressed size of next chunk. */
+ chunk_usize = min(orig_chunk_size, size - offset);
+
+ /* Read the compressed size of the next chunk from the chunk
+ * header. */
+ ret = full_read(in_fd, &chunk_hdr, sizeof(chunk_hdr));
+ if (ret)
+ goto read_error;
+
+ chunk_csize = le32_to_cpu(chunk_hdr.compressed_size);
+
+ if (chunk_csize > orig_chunk_size) {
+ errno = EINVAL;
+ ret = WIMLIB_ERR_INVALID_PIPABLE_WIM;
+ goto invalid;
+ }
+
+ /* Read chunk data. */
+ ret = full_read(in_fd, cchunk, chunk_csize);
+ if (ret)
+ goto read_error;
+
+ if (flags & WIMLIB_READ_RESOURCE_FLAG_SEEK_ONLY)
+ continue;
+
+ /* Decompress chunk if needed. Uncompressed size same
+ * as compressed size means the chunk is uncompressed.
+ */
+ if (chunk_csize == chunk_usize) {
+ memcpy(out_p, cchunk, chunk_usize);
} else {
- ERROR("Failed to read %"PRIu64" bytes from "
- "uncompressed resource at offset %"PRIu64,
- len, offset);
+ ret = (*decompress)(cchunk, chunk_csize,
+ out_p, chunk_usize,
+ wim_resource_compression_type(lte),
+ orig_chunk_size);
+ if (ret) {
+ errno = EINVAL;
+ ret = WIMLIB_ERR_DECOMPRESSION;
+ goto invalid;
+ }
+ }
+ out_p += chunk_usize;
+
+ /* Feed the uncompressed data into the callback function or copy
+ * it into the provided buffer. */
+ if (cb && (out_p == out_buf_end ||
+ offset + chunk_usize == size))
+ {
+ size_t bytes_sent;
+ const u8 *p;
+
+ for (p = out_buf; p != out_p; p += bytes_sent) {
+ bytes_sent = min(in_chunk_size, out_p - p);
+ ret = cb(p, bytes_sent, ctx_or_buf);
+ if (ret)
+ return ret;
+ }
+ out_p = out_buf;
}
- return WIMLIB_ERR_READ;
}
+
+ ret = skip_chunk_table(lte, in_fd);
+ if (ret)
+ goto read_error;
return 0;
+
+read_error:
+ ERROR_WITH_ERRNO("Error reading compressed file resource");
+ return ret;
+
+invalid:
+ ERROR("Compressed file resource is invalid");
+ return ret;
}
-/* Reads the contents of a struct resource_entry, as represented in the on-disk
- * format, from the memory pointed to by @p, and fills in the fields of @entry.
- * A pointer to the byte after the memory read at @p is returned. */
-const u8 *
-get_resource_entry(const u8 *p, struct resource_entry *entry)
+/*
+ * 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().
+ *
+ * @flags can be:
+ *
+ * 0:
+ * Just do a normal read, decompressing the data if necessary. @size and
+ * @offset are interpreted relative to the uncompressed contents of the
+ * stream.
+ *
+ * WIMLIB_READ_RESOURCE_FLAG_RAW_FULL:
+ * Only valid when the resource is compressed: Read the raw contents of
+ * the compressed resource. If the resource is non-pipable, this includes
+ * the chunk table as well as the compressed chunks. If the resource is
+ * pipable, this includes the compressed chunks--- including the chunk
+ * headers--- and the chunk table. The stream header is still *not*
+ * included.
+ *
+ * In this mode, @offset is relative to the beginning of the raw contents
+ * of the compressed resource--- that is, the chunk table if the resource
+ * is non-pipable, or the header for the first compressed chunk if the
+ * resource is pipable. @size is the number of raw bytes to read, which
+ * must not overrun the end of the resource. For example, if @offset is 0,
+ * then @size can be at most the raw size of the compressed resource
+ * (@lte->resource_entry.size).
+ *
+ * WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS:
+ * Only valid when the resource is compressed and is not being read from a
+ * pipe: Read the raw contents of the compressed chunks of the compressed
+ * resource. For pipable resources, this does *not* include the chunk
+ * headers. If a callback function is being used, it will be called once
+ * for each compressed chunk. The chunk table is excluded. Also, for
+ * pipable resources, the stream and chunk headers are excluded. In this
+ * mode, @size must be exactly the raw size of the compressed resource
+ * (@lte->resource_entry.size) and @offset must be 0.
+ *
+ * WIMLIB_READ_RESOURCE_FLAG_SEEK_ONLY:
+ * Only valid when the resource is being read from a pipe: Skip over the
+ * requested data rather than feed it to the callback function or write it
+ * into the buffer. No decompression is done.
+ * WIMLIB_READ_RESOURCE_FLAG_RAW_* may not be combined with this flag.
+ * @offset must be 0 and @size must be the uncompressed size of the
+ * resource.
+ *
+ * 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_PIPABLE_WIM (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,
+ u32 in_chunk_size,
+ void *ctx_or_buf, int flags, u64 offset)
{
- u64 size;
- u8 flags;
-
- p = get_u56(p, &size);
- p = get_u8(p, &flags);
- entry->size = size;
- entry->flags = flags;
+ struct filedes *in_fd;
+ int ret;
- /* 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... */
- p = get_u64(p, &entry->offset);
- if (entry->offset & 0xc000000000000000ULL) {
- WARNING("Truncating offset in resource entry");
- entry->offset &= 0x3fffffffffffffffULL;
- }
- p = get_u64(p, &entry->original_size);
- if (entry->original_size & 0xc000000000000000ULL) {
- WARNING("Truncating original_size in resource entry");
- entry->original_size &= 0x3fffffffffffffffULL;
+ /* Make sure the resource is actually located in a WIM file and is not
+ * somewhere else. */
+ wimlib_assert(lte->resource_location == RESOURCE_IN_WIM);
+
+ /* If a callback was specified, in_chunk_size must be a power of 2 (and
+ * not 0). */
+ wimlib_assert(cb == NULL || is_power_of_2(in_chunk_size));
+
+ /* If a callback was specified, offset must be zero. */
+ wimlib_assert(cb == NULL || offset == 0);
+
+ /* Retrieve input file descriptor for the WIM file. */
+ in_fd = <e->wim->in_fd;
+
+ /* Don't allow raw reads (either full or chunks) of uncompressed
+ * resources. */
+ wimlib_assert(!(flags & WIMLIB_READ_RESOURCE_FLAG_RAW) ||
+ resource_is_compressed(<e->resource_entry));
+
+ /* Don't allow seek-only reads unless reading from a pipe; also don't
+ * allow combining SEEK_ONLY with either RAW flag. */
+ wimlib_assert(!(flags & WIMLIB_READ_RESOURCE_FLAG_SEEK_ONLY) ||
+ (!filedes_is_seekable(in_fd) &&
+ !(flags & WIMLIB_READ_RESOURCE_FLAG_RAW)));
+
+ DEBUG("Reading WIM resource: %"PRIu64" @ +%"PRIu64" "
+ "from %"PRIu64" @ +%"PRIu64" (readflags 0x%08x, resflags 0x%02x%s)",
+ size, offset,
+ 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(<e->resource_entry))
+ {
+ /* Reading raw resource contents or reading uncompressed
+ * resource. */
+ wimlib_assert(offset + size <= lte->resource_entry.size);
+ offset += lte->resource_entry.offset;
+ if (flags & WIMLIB_READ_RESOURCE_FLAG_SEEK_ONLY) {
+ if (lte->resource_entry.size != 0) {
+ u8 dummy;
+ ret = full_pread(in_fd, &dummy, 1,
+ offset + lte->resource_entry.size - 1);
+ if (ret)
+ goto read_error;
+ }
+ } else if (cb) {
+ /* Send data to callback function */
+ u8 buf[min(in_chunk_size, size)];
+ while (size) {
+ size_t bytes_to_read = min(in_chunk_size, size);
+ ret = full_pread(in_fd, buf, bytes_to_read,
+ offset);
+ if (ret)
+ goto read_error;
+ ret = cb(buf, bytes_to_read, ctx_or_buf);
+ if (ret)
+ goto out;
+ size -= bytes_to_read;
+ offset += bytes_to_read;
+ }
+ } else {
+ /* Send data directly to a buffer */
+ ret = full_pread(in_fd, ctx_or_buf, size, offset);
+ if (ret)
+ goto read_error;
+ }
+ ret = 0;
+ } else if (lte->is_pipable && !filedes_is_seekable(in_fd)) {
+ /* Reading compressed, pipable resource from pipe. */
+ ret = read_pipable_resource(lte, size, cb,
+ in_chunk_size,
+ ctx_or_buf, flags, offset);
+ } else {
+ /* Reading compressed, possibly pipable resource from seekable
+ * file. */
+ ret = read_compressed_resource(lte, size, cb,
+ in_chunk_size,
+ ctx_or_buf, flags, offset);
}
- return p;
+ goto out;
+
+read_error:
+ ERROR_WITH_ERRNO("Error reading data from WIM");
+out:
+ return ret;
}
-/* Copies the struct resource_entry @entry to the memory pointed to by @p in the
- * on-disk format. A pointer to the byte after the memory written at @p is
- * returned. */
-u8 *
-put_resource_entry(u8 *p, const struct resource_entry *entry)
+
+int
+read_partial_wim_resource_into_buf(const struct wim_lookup_table_entry *lte,
+ size_t size, u64 offset, void *buf)
{
- p = put_u56(p, entry->size);
- p = put_u8(p, entry->flags);
- p = put_u64(p, entry->offset);
- p = put_u64(p, entry->original_size);
- return p;
+ return read_partial_wim_resource(lte, size, NULL, 0, buf, 0, offset);
}
-#ifdef WITH_FUSE
-static FILE *
-wim_get_fp(WIMStruct *w)
+static int
+read_wim_resource_prefix(const struct wim_lookup_table_entry *lte,
+ u64 size,
+ consume_data_callback_t cb,
+ u32 in_chunk_size,
+ void *ctx_or_buf,
+ int flags)
{
- pthread_mutex_lock(&w->fp_tab_mutex);
- FILE *fp;
+ return read_partial_wim_resource(lte, size, cb, in_chunk_size,
+ ctx_or_buf, flags, 0);
+}
- wimlib_assert(w->filename != NULL);
- for (size_t i = 0; i < w->num_allocated_fps; i++) {
- if (w->fp_tab[i]) {
- fp = w->fp_tab[i];
- w->fp_tab[i] = NULL;
- goto out;
+#ifndef __WIN32__
+static int
+read_file_on_disk_prefix(const struct wim_lookup_table_entry *lte,
+ u64 size,
+ consume_data_callback_t cb,
+ u32 in_chunk_size,
+ void *ctx_or_buf,
+ int _ignored_flags)
+{
+ const tchar *filename = lte->file_on_disk;
+ int ret;
+ struct filedes fd;
+ int raw_fd;
+ u8 *out_buf;
+ bool out_buf_malloced;
+ const size_t stack_max = 32768;
+
+ DEBUG("Reading %"PRIu64" bytes from \"%"TS"\"",
+ size, lte->file_on_disk);
+
+ raw_fd = open(filename, O_RDONLY);
+ if (raw_fd < 0) {
+ ERROR_WITH_ERRNO("Can't open \"%"TS"\"", filename);
+ return WIMLIB_ERR_OPEN;
+ }
+ filedes_init(&fd, raw_fd);
+ out_buf_malloced = false;
+ if (cb) {
+ /* Send data to callback function */
+ if (in_chunk_size <= stack_max) {
+ out_buf = alloca(in_chunk_size);
+ } else {
+ out_buf = MALLOC(in_chunk_size);
+ if (out_buf == NULL) {
+ ret = WIMLIB_ERR_NOMEM;
+ goto out_close;
+ }
+ out_buf_malloced = true;
}
+
+ size_t bytes_to_read;
+ while (size) {
+ bytes_to_read = min(in_chunk_size, size);
+ ret = full_read(&fd, out_buf, bytes_to_read);
+ if (ret)
+ goto read_error;
+ ret = cb(out_buf, bytes_to_read, ctx_or_buf);
+ if (ret)
+ goto out_close;
+ size -= bytes_to_read;
+ }
+ } else {
+ /* Send data directly to a buffer */
+ ret = full_read(&fd, ctx_or_buf, size);
+ if (ret)
+ goto read_error;
}
- DEBUG("Opening extra file descriptor to `%"TS"'", w->filename);
- fp = tfopen(w->filename, T("rb"));
- if (!fp)
- ERROR_WITH_ERRNO("Failed to open `%"TS"'", w->filename);
-out:
- pthread_mutex_unlock(&w->fp_tab_mutex);
- return fp;
+ ret = 0;
+ goto out_close;
+
+read_error:
+ ERROR_WITH_ERRNO("Error reading \"%"TS"\"", filename);
+out_close:
+ filedes_close(&fd);
+ if (out_buf_malloced)
+ FREE(out_buf);
+ return ret;
}
+#endif /* !__WIN32__ */
static int
-wim_release_fp(WIMStruct *w, FILE *fp)
+read_buffer_prefix(const struct wim_lookup_table_entry *lte,
+ u64 size, consume_data_callback_t cb,
+ u32 in_chunk_size,
+ void *ctx_or_buf, int _ignored_flags)
{
- int ret = 0;
- FILE **fp_tab;
- pthread_mutex_lock(&w->fp_tab_mutex);
+ if (cb) {
+ int ret;
+ u32 chunk_size;
- for (size_t i = 0; i < w->num_allocated_fps; i++) {
- if (w->fp_tab[i] == NULL) {
- w->fp_tab[i] = fp;
- goto out;
+ for (u64 offset = 0; offset < size; offset += chunk_size) {
+ chunk_size = min(in_chunk_size, size - offset);
+ ret = cb((const u8*)lte->attached_buffer + offset,
+ chunk_size, ctx_or_buf);
+ if (ret)
+ return ret;
}
+ } else {
+ memcpy(ctx_or_buf, lte->attached_buffer, size);
}
-
- fp_tab = REALLOC(w->fp_tab, sizeof(FILE*) * (w->num_allocated_fps + 4));
- if (!fp_tab) {
- ret = WIMLIB_ERR_NOMEM;
- goto out;
- }
- w->fp_tab = fp_tab;
- memset(&w->fp_tab[w->num_allocated_fps], 0, 4 * sizeof(FILE*));
- w->fp_tab[w->num_allocated_fps] = fp;
- w->num_allocated_fps += 4;
-out:
- pthread_mutex_unlock(&w->fp_tab_mutex);
- return ret;
+ return 0;
}
-#endif /* !WITH_FUSE */
+
+typedef int (*read_resource_prefix_handler_t)(const struct wim_lookup_table_entry *lte,
+ u64 size,
+ consume_data_callback_t cb,
+ u32 in_chunk_size,
+ void *ctx_or_buf,
+ int flags);
/*
- * Reads some data from the resource corresponding to a WIM lookup table entry.
+ * read_resource_prefix()-
+ *
+ * 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.
*
- * @lte: The WIM lookup table entry for the resource.
- * @buf: Buffer into which to write the data.
- * @size: Number of bytes to read.
- * @offset: Offset at which to start reading the resource.
+ * 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).
*
- * Returns zero on success, nonzero on failure.
+ * 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.
*/
int
-read_wim_resource(const struct wim_lookup_table_entry *lte, void *buf,
- size_t size, u64 offset, int flags)
+read_resource_prefix(const struct wim_lookup_table_entry *lte,
+ u64 size, consume_data_callback_t cb, u32 in_chunk_size,
+ void *ctx_or_buf, int flags)
{
- int ctype;
- int ret = 0;
- FILE *fp;
-
- /* We shouldn't be allowing read over-runs in any part of the library.
- * */
- if (flags & WIMLIB_RESOURCE_FLAG_RAW)
- wimlib_assert(offset + size <= lte->resource_entry.size);
- else
- wimlib_assert(offset + size <= lte->resource_entry.original_size);
-
- switch (lte->resource_location) {
- case RESOURCE_IN_WIM:
- /* The resource is in a WIM file, and its WIMStruct is given by
- * the lte->wim member. The resource may be either compressed
- * or uncompressed. */
- wimlib_assert(lte->wim != NULL);
-
- #ifdef WITH_FUSE
- if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) {
- fp = wim_get_fp(lte->wim);
- if (!fp)
- return WIMLIB_ERR_OPEN;
- } else
- #endif
- {
- wimlib_assert(!(flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED));
- wimlib_assert(lte->wim->fp != NULL);
- fp = lte->wim->fp;
- }
-
- ctype = wim_resource_compression_type(lte);
-
- wimlib_assert(ctype != WIMLIB_COMPRESSION_TYPE_NONE ||
- (lte->resource_entry.original_size ==
- lte->resource_entry.size));
-
- if ((flags & WIMLIB_RESOURCE_FLAG_RAW)
- || ctype == WIMLIB_COMPRESSION_TYPE_NONE)
- ret = read_uncompressed_resource(fp,
- lte->resource_entry.offset + offset,
- size, buf);
- else
- ret = read_compressed_resource(fp,
- lte->resource_entry.size,
- lte->resource_entry.original_size,
- lte->resource_entry.offset,
- ctype, size, offset, buf);
+ static const read_resource_prefix_handler_t handlers[] = {
+ [RESOURCE_IN_WIM] = read_wim_resource_prefix,
+ #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,
#ifdef WITH_FUSE
- if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) {
- int ret2 = wim_release_fp(lte->wim, fp);
- if (ret == 0)
- ret = ret2;
- }
+ [RESOURCE_IN_STAGING_FILE] = read_file_on_disk_prefix,
#endif
- break;
- case RESOURCE_IN_STAGING_FILE:
- case RESOURCE_IN_FILE_ON_DISK:
- /* The resource is in some file on the external filesystem and
- * needs to be read uncompressed */
- wimlib_assert(lte->file_on_disk != NULL);
- BUILD_BUG_ON(<e->file_on_disk != <e->staging_file_name);
- /* Use existing file pointer if available; otherwise open one
- * temporarily */
- if (lte->file_on_disk_fp) {
- fp = lte->file_on_disk_fp;
- } else {
- fp = tfopen(lte->file_on_disk, T("rb"));
- if (!fp) {
- ERROR_WITH_ERRNO("Failed to open the file "
- "`%"TS"'", lte->file_on_disk);
- ret = WIMLIB_ERR_OPEN;
- break;
- }
- }
- ret = read_uncompressed_resource(fp, offset, size, buf);
- if (fp != lte->file_on_disk_fp)
- fclose(fp);
- break;
-#ifdef __WIN32__
- case RESOURCE_WIN32:
- wimlib_assert(lte->win32_file_on_disk_fp != INVALID_HANDLE_VALUE);
- ret = win32_read_file(lte->file_on_disk,
- lte->win32_file_on_disk_fp, offset,
- size, buf);
- break;
-#endif
- case RESOURCE_IN_ATTACHED_BUFFER:
- /* The resource is directly attached uncompressed in an
- * in-memory buffer. */
- wimlib_assert(lte->attached_buffer != NULL);
- memcpy(buf, lte->attached_buffer + offset, size);
- break;
-#ifdef WITH_NTFS_3G
- case RESOURCE_IN_NTFS_VOLUME:
- wimlib_assert(lte->ntfs_loc != NULL);
- wimlib_assert(lte->attr != NULL);
- if (lte->ntfs_loc->is_reparse_point)
- offset += 8;
- if (ntfs_attr_pread(lte->attr, offset, size, buf) != size) {
- ERROR_WITH_ERRNO("Error reading NTFS attribute "
- "at `%"TS"'",
- lte->ntfs_loc->path);
- ret = WIMLIB_ERR_NTFS_3G;
- }
- break;
-#endif
- default:
- wimlib_assert(0);
- ret = -1;
- break;
+ #ifdef WITH_NTFS_3G
+ [RESOURCE_IN_NTFS_VOLUME] = read_ntfs_file_prefix,
+ #endif
+ #ifdef __WIN32__
+ [RESOURCE_WIN32_ENCRYPTED] = read_win32_encrypted_file_prefix,
+ #endif
+ };
+ wimlib_assert(lte->resource_location < ARRAY_LEN(handlers)
+ && handlers[lte->resource_location] != NULL);
+ wimlib_assert(cb == NULL || in_chunk_size > 0);
+ return handlers[lte->resource_location](lte, size, cb, in_chunk_size, ctx_or_buf, flags);
+}
+
+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, 0, buf, 0);
+}
+
+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;
}
- return ret;
+
+ buf = MALLOC(lte->resource_entry.original_size);
+ if (!buf)
+ return WIMLIB_ERR_NOMEM;
+
+ ret = read_full_resource_into_buf(lte, buf);
+ if (ret) {
+ FREE(buf);
+ return ret;
+ }
+
+ *buf_ret = buf;
+ return 0;
}
-/*
- * Reads all the data from the resource corresponding to a WIM lookup table
- * entry.
- *
- * @lte: The WIM lookup table entry for the resource.
- * @buf: Buffer into which to write the data. It must be at least
- * wim_resource_size(lte) bytes long.
- *
- * Returns 0 on success; nonzero on failure.
- */
int
-read_full_wim_resource(const struct wim_lookup_table_entry *lte,
- void *buf, int flags)
+res_entry_to_data(const struct resource_entry *res_entry,
+ WIMStruct *wim, void **buf_ret)
{
- return read_wim_resource(lte, buf, wim_resource_size(lte), 0, flags);
+ int ret;
+ struct wim_lookup_table_entry *lte;
+
+ lte = new_lookup_table_entry();
+ if (!lte)
+ return WIMLIB_ERR_NOMEM;
+
+ copy_resource_entry(<e->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 {
+ SHA_CTX sha_ctx;
+ consume_data_callback_t extract_chunk;
+ void *extract_chunk_arg;
+};
+
+static int
+extract_chunk_sha1_wrapper(const void *chunk, size_t chunk_size,
+ void *_ctx)
+{
+ struct extract_ctx *ctx = _ctx;
+
+ sha1_update(&ctx->sha_ctx, chunk, 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
int
extract_wim_resource(const struct wim_lookup_table_entry *lte,
u64 size,
- extract_chunk_func_t extract_chunk,
+ consume_data_callback_t extract_chunk,
void *extract_chunk_arg)
{
- u64 bytes_remaining = size;
- u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
- u64 offset = 0;
- int ret = 0;
- u8 hash[SHA1_HASH_SIZE];
- bool check_hash = (size == wim_resource_size(lte));
- SHA_CTX ctx;
-
- if (check_hash)
- sha1_init(&ctx);
-
- while (bytes_remaining) {
- u64 to_read = min(bytes_remaining, sizeof(buf));
- ret = read_wim_resource(lte, buf, to_read, offset, 0);
- if (ret != 0)
- return ret;
- if (check_hash)
- sha1_update(&ctx, buf, to_read);
- ret = extract_chunk(buf, to_read, offset, extract_chunk_arg);
- if (ret != 0) {
- ERROR_WITH_ERRNO("Error extracting WIM resource");
- return ret;
- }
- bytes_remaining -= to_read;
- offset += to_read;
- }
- if (check_hash) {
- sha1_final(hash, &ctx);
- if (!hashes_equal(hash, lte->hash)) {
- #ifdef ENABLE_ERROR_MESSAGES
- ERROR("Invalid checksum on the following WIM resource:");
- print_lookup_table_entry(lte, stderr);
- #endif
- return WIMLIB_ERR_INVALID_RESOURCE_HASH;
+ int ret;
+ if (size == wim_resource_size(lte)) {
+ /* Do SHA1 */
+ struct extract_ctx ctx;
+ ctx.extract_chunk = extract_chunk;
+ ctx.extract_chunk_arg = extract_chunk_arg;
+ 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)) {
+ 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 0;
+ return ret;
}
-/* Write @n bytes from @buf to the file descriptor @fd, retrying on internupt
- * and on short writes.
- *
- * Returns short count and set errno on failure. */
-static ssize_t
-full_write(int fd, const void *buf, size_t n)
+static int
+extract_wim_chunk_to_fd(const void *buf, size_t len, void *_fd_p)
{
- const void *p = buf;
- ssize_t ret;
- ssize_t total = 0;
-
- while (total != n) {
- ret = write(fd, p, n);
- if (ret < 0) {
- if (errno == EINTR)
- continue;
- else
- break;
- }
- total += ret;
- p += ret;
- }
- return total;
+ struct filedes *fd = _fd_p;
+ int ret = full_write(fd, buf, len);
+ if (ret)
+ ERROR_WITH_ERRNO("Error writing to file descriptor");
+ return ret;
}
int
-extract_wim_chunk_to_fd(const void *buf, size_t len, u64 offset, void *arg)
+extract_wim_resource_to_fd(const struct wim_lookup_table_entry *lte,
+ struct filedes *fd, u64 size)
{
- int fd = *(int*)arg;
- ssize_t ret = full_write(fd, buf, len);
- if (ret < len) {
- ERROR_WITH_ERRNO("Error writing to file descriptor");
- return WIMLIB_ERR_WRITE;
- } else {
- return 0;
- }
+ return extract_wim_resource(lte, size, extract_wim_chunk_to_fd, fd);
}
-/*
- * 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.)
- */
+
+static int
+sha1_chunk(const void *buf, size_t len, void *ctx)
+{
+ sha1_update(ctx, buf, len);
+ return 0;
+}
+
+/* Calculate the SHA1 message digest of a stream. */
int
-copy_resource(struct wim_lookup_table_entry *lte, void *wim)
+sha1_resource(struct wim_lookup_table_entry *lte)
{
- WIMStruct *w = wim;
int ret;
+ SHA_CTX sha_ctx;
+
+ sha1_init(&sha_ctx);
+ ret = read_resource_prefix(lte, wim_resource_size(lte),
+ sha1_chunk, wim_resource_chunk_size(lte),
+ &sha_ctx, 0);
+ if (ret == 0)
+ sha1_final(lte->hash, &sha_ctx);
+ return ret;
+}
- ret = write_wim_resource(lte, w->out_fp,
- wim_resource_compression_type(lte),
- <e->output_resource_entry, 0);
- if (ret == 0) {
- lte->out_refcnt = lte->refcnt;
- lte->part_number = w->hdr.part_number;
+/* 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)
+{
+ /* 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.) */
+
+ /* 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 should be
+ * 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);
}
git://wimlib.net / wimlib / blobdiff