X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fresource.c;h=e8f1114412b5940b07fe8db69caba304772eb6ae;hp=5ae2dff2ffc7da0e3db9ee99bce3f116a7efc433;hb=e1ec9c16b55adb6947b4d1ff46dc847430db4746;hpb=cdda846b47fae4633ba1084472c982766592ac10 diff --git a/src/resource.c b/src/resource.c index 5ae2dff2..e8f11144 100644 --- a/src/resource.c +++ b/src/resource.c @@ -1,7 +1,7 @@ /* * resource.c * - * Read uncompressed and compressed metadata and file resources. + * Read uncompressed and compressed metadata and file resources from a WIM file. */ /* @@ -22,40 +22,35 @@ * wimlib; if not, see http://www.gnu.org/licenses/. */ -#include "config.h" +#include "wimlib_internal.h" +#include "dentry.h" +#include "lookup_table.h" +#include "buffer_io.h" +#include "lzx.h" +#include "xpress.h" +#include "sha1.h" -#include +#include #include +#include +#include #ifdef WITH_NTFS_3G +#include #include #include #include #endif -#include "wimlib_internal.h" -#include "lookup_table.h" -#include "io.h" -#include "lzx.h" -#include "xpress.h" -#include "sha1.h" -#include "dentry.h" -#include -#include -#ifdef HAVE_ALLOCA_H -#include -#endif - - -/* +/* * Reads all or part of a compressed resource into an in-memory buffer. * * @fp: The FILE* for the WIM file. - * @resource_compressed_size: The compressed size of the resource. + * @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. + * @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 @@ -65,9 +60,9 @@ * * Returns zero on success, nonzero on failure. */ -static int read_compressed_resource(FILE *fp, u64 resource_compressed_size, - u64 resource_uncompressed_size, - u64 resource_offset, int resource_ctype, +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, u8 contents_ret[]) { @@ -82,9 +77,9 @@ static int read_compressed_resource(FILE *fp, u64 resource_compressed_size, if (len == 0) return 0; - int (*decompress)(const void *, uint, void *, uint); + int (*decompress)(const void *, unsigned, void *, unsigned); /* Set the appropriate decompress function. */ - if (resource_ctype == WIM_COMPRESSION_TYPE_LZX) + if (resource_ctype == WIMLIB_COMPRESSION_TYPE_LZX) decompress = lzx_decompress; else decompress = xpress_decompress; @@ -98,7 +93,7 @@ static int read_compressed_resource(FILE *fp, u64 resource_compressed_size, * 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. + * follows the chunk table and therefore must have an offset of 0. */ /* Calculate how many chunks the resource conists of in its entirety. */ @@ -146,7 +141,7 @@ static int read_compressed_resource(FILE *fp, u64 resource_compressed_size, /* 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) ? + u64 chunk_entry_size = (resource_uncompressed_size >= (u64)1 << 32) ? 8 : 4; /* Size of the full chunk table in the WIM file. */ @@ -160,11 +155,11 @@ static int read_compressed_resource(FILE *fp, u64 resource_compressed_size, /* 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) ? + 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 + + 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 " @@ -191,17 +186,17 @@ static int read_compressed_resource(FILE *fp, u64 resource_compressed_size, if (chunk_entry_size == 4) { u32 *entries = (u32*)chunk_tab_buf; while (num_needed_chunk_entries--) - *chunk_tab_p++ = to_le32(*entries++); + *chunk_tab_p++ = le32_to_cpu(*entries++); } else { u64 *entries = (u64*)chunk_tab_buf; while (num_needed_chunk_entries--) - *chunk_tab_p++ = to_le64(*entries++); + *chunk_tab_p++ = le64_to_cpu(*entries++); } /* Done with the chunk table now. We must now seek to the first chunk * that is needed for the read. */ - u64 file_offset_of_first_needed_chunk = resource_offset + + 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 " @@ -231,13 +226,14 @@ static int read_compressed_resource(FILE *fp, u64 resource_compressed_size, /* Calculate the sizes of the compressed chunk and of the * uncompressed chunk. */ - uint compressed_chunk_size, uncompressed_chunk_size; + 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] - + compressed_chunk_size = chunk_offsets[i + 1 - start_chunk] - chunk_offsets[i - start_chunk]; uncompressed_chunk_size = WIM_CHUNK_SIZE; } else { @@ -245,15 +241,15 @@ static int read_compressed_resource(FILE *fp, u64 resource_compressed_size, * 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. + * 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 - + compressed_chunk_size = resource_compressed_size - chunk_table_size - chunk_offsets[i - start_chunk]; - uncompressed_chunk_size = resource_uncompressed_size % + uncompressed_chunk_size = resource_uncompressed_size % WIM_CHUNK_SIZE; /* If the remainder is 0, the last chunk actually @@ -280,7 +276,7 @@ static int read_compressed_resource(FILE *fp, u64 resource_compressed_size, end_offset = WIM_CHUNK_SIZE - 1; u64 partial_chunk_size = end_offset + 1 - start_offset; - bool is_partial_chunk = (partial_chunk_size != + bool is_partial_chunk = (partial_chunk_size != uncompressed_chunk_size); DEBUG2("start_offset = %u, end_offset = %u", start_offset, @@ -300,7 +296,7 @@ static int read_compressed_resource(FILE *fp, u64 resource_compressed_size, return WIMLIB_ERR_READ; } } - if (fread(out_p, 1, partial_chunk_size, fp) != + if (fread(out_p, 1, partial_chunk_size, fp) != partial_chunk_size) goto err; } else { @@ -308,7 +304,7 @@ static int read_compressed_resource(FILE *fp, u64 resource_compressed_size, int ret; /* Read the compressed data into compressed_buf. */ - if (fread(compressed_buf, 1, compressed_chunk_size, + if (fread(compressed_buf, 1, compressed_chunk_size, fp) != compressed_chunk_size) goto err; @@ -319,7 +315,7 @@ static int read_compressed_resource(FILE *fp, u64 resource_compressed_size, ret = decompress(compressed_buf, compressed_chunk_size, - uncompressed_buf, + uncompressed_buf, uncompressed_chunk_size); if (ret != 0) return WIMLIB_ERR_DECOMPRESSION; @@ -350,7 +346,7 @@ err: return WIMLIB_ERR_READ; } -/* +/* * Reads uncompressed data from an open file stream. */ int read_uncompressed_resource(FILE *fp, u64 offset, u64 len, @@ -375,9 +371,6 @@ int read_uncompressed_resource(FILE *fp, u64 offset, u64 len, return 0; } - - - /* 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. */ @@ -390,8 +383,21 @@ const u8 *get_resource_entry(const u8 *p, struct resource_entry *entry) p = get_u8(p, &flags); entry->size = size; entry->flags = flags; + + /* 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; + } return p; } @@ -407,6 +413,59 @@ u8 *put_resource_entry(u8 *p, const struct resource_entry *entry) return p; } +#ifdef WITH_FUSE +static FILE *wim_get_fp(WIMStruct *w) +{ + pthread_mutex_lock(&w->fp_tab_mutex); + FILE *fp; + + 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; + } + } + DEBUG("Opening extra file descriptor to `%s'", w->filename); + fp = fopen(w->filename, "rb"); + if (!fp) + ERROR_WITH_ERRNO("Failed to open `%s'", w->filename); +out: + pthread_mutex_unlock(&w->fp_tab_mutex); + return fp; +} + +static int wim_release_fp(WIMStruct *w, FILE *fp) +{ + int ret = 0; + FILE **fp_tab; + + pthread_mutex_lock(&w->fp_tab_mutex); + + for (size_t i = 0; i < w->num_allocated_fps; i++) { + if (w->fp_tab[i] == NULL) { + w->fp_tab[i] = fp; + goto out; + } + } + + 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; +} +#endif + /* * Reads some data from the resource corresponding to a WIM lookup table entry. * @@ -414,47 +473,67 @@ u8 *put_resource_entry(u8 *p, const struct resource_entry *entry) * @buf: Buffer into which to write the data. * @size: Number of bytes to read. * @offset: Offset at which to start reading the resource. - * @raw: If %true, compressed data is read literally rather than being - * decompressed first. * * Returns zero on success, nonzero on failure. */ -int read_wim_resource(const struct lookup_table_entry *lte, u8 buf[], - size_t size, u64 offset, bool raw) +int read_wim_resource(const struct wim_lookup_table_entry *lte, u8 buf[], + size_t size, u64 offset, int flags) { + int ctype; + int ret = 0; + FILE *fp; + /* We shouldn't be allowing read over-runs in any part of the library. * */ - if (raw) + if (flags & WIMLIB_RESOURCE_FLAG_RAW) wimlib_assert(offset + size <= lte->resource_entry.size); else wimlib_assert(offset + size <= lte->resource_entry.original_size); - int ctype; - int ret; - FILE *fp; 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); - wimlib_assert(lte->wim->fp); + 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 != WIM_COMPRESSION_TYPE_NONE || + wimlib_assert(ctype != WIMLIB_COMPRESSION_TYPE_NONE || (lte->resource_entry.original_size == lte->resource_entry.size)); - if (raw || ctype == WIM_COMPRESSION_TYPE_NONE) - return read_uncompressed_resource(lte->wim->fp, - lte->resource_entry.offset + offset, - size, buf); + if ((flags & WIMLIB_RESOURCE_FLAG_RAW) + || ctype == WIMLIB_COMPRESSION_TYPE_NONE) + ret = read_uncompressed_resource(fp, + lte->resource_entry.offset + offset, + size, buf); else - return read_compressed_resource(lte->wim->fp, - lte->resource_entry.size, - lte->resource_entry.original_size, - lte->resource_entry.offset, - ctype, size, offset, buf); + ret = read_compressed_resource(fp, + lte->resource_entry.size, + lte->resource_entry.original_size, + lte->resource_entry.offset, + ctype, size, offset, buf); + #ifdef WITH_FUSE + if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) { + int ret2 = wim_release_fp(lte->wim, fp); + if (ret == 0) + ret = ret2; + } + #endif break; case RESOURCE_IN_STAGING_FILE: case RESOURCE_IN_FILE_ON_DISK: @@ -471,49 +550,43 @@ int read_wim_resource(const struct lookup_table_entry *lte, u8 buf[], if (!fp) { ERROR_WITH_ERRNO("Failed to open the file " "`%s'", lte->file_on_disk); - return WIMLIB_ERR_OPEN; + ret = WIMLIB_ERR_OPEN; + break; } } ret = read_uncompressed_resource(fp, offset, size, buf); if (fp != lte->file_on_disk_fp) fclose(fp); - return ret; break; case RESOURCE_IN_ATTACHED_BUFFER: /* The resource is directly attached uncompressed in an * in-memory buffer. */ - wimlib_assert(lte->attached_buffer); + wimlib_assert(lte->attached_buffer != NULL); memcpy(buf, lte->attached_buffer + offset, size); - return 0; break; #ifdef WITH_NTFS_3G case RESOURCE_IN_NTFS_VOLUME: - wimlib_assert(lte->ntfs_loc); - if (lte->attr) { - u64 adjusted_offset; - if (lte->ntfs_loc->is_reparse_point) - adjusted_offset = offset + 8; - else - adjusted_offset = offset; - if (ntfs_attr_pread(lte->attr, offset, size, buf) == size) { - return 0; - } else { - ERROR_WITH_ERRNO("Error reading NTFS attribute " - "at `%s'", - lte->ntfs_loc->path_utf8); - return WIMLIB_ERR_NTFS_3G; - } - } else { - wimlib_assert(0); + 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 `%s'", + lte->ntfs_loc->path_utf8); + ret = WIMLIB_ERR_NTFS_3G; } break; #endif default: - assert(0); + wimlib_assert(0); + ret = -1; + break; } + return ret; } -/* +/* * Reads all the data from the resource corresponding to a WIM lookup table * entry. * @@ -523,525 +596,99 @@ int read_wim_resource(const struct lookup_table_entry *lte, u8 buf[], * * Returns 0 on success; nonzero on failure. */ -int read_full_wim_resource(const struct lookup_table_entry *lte, u8 buf[]) -{ - return read_wim_resource(lte, buf, wim_resource_size(lte), 0, false); -} - -/* Chunk table that's located at the beginning of each compressed resource in - * the WIM. (This is not the on-disk format; the on-disk format just has an - * array of offsets.) */ -struct chunk_table { - off_t file_offset; - u64 num_chunks; - u64 original_resource_size; - u64 bytes_per_chunk_entry; - u64 table_disk_size; - u64 cur_offset; - u64 *cur_offset_p; - u64 offsets[0]; -}; - -/* - * Allocates and initializes a chunk table, and reserves space for it in the - * output file. - */ -static int -begin_wim_resource_chunk_tab(const struct lookup_table_entry *lte, - FILE *out_fp, - off_t file_offset, - struct chunk_table **chunk_tab_ret) -{ - u64 size = wim_resource_size(lte); - u64 num_chunks = (size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE; - struct chunk_table *chunk_tab = MALLOC(sizeof(struct chunk_table) + - num_chunks * sizeof(u64)); - int ret = 0; - - wimlib_assert(size != 0); - - if (!chunk_tab) { - ERROR("Failed to allocate chunk table for %"PRIu64" byte " - "resource", size); - ret = WIMLIB_ERR_NOMEM; - goto out; - } - chunk_tab->file_offset = file_offset; - chunk_tab->num_chunks = num_chunks; - chunk_tab->original_resource_size = size; - chunk_tab->bytes_per_chunk_entry = (size >= (1ULL << 32)) ? 8 : 4; - chunk_tab->table_disk_size = chunk_tab->bytes_per_chunk_entry * - (num_chunks - 1); - chunk_tab->cur_offset = 0; - chunk_tab->cur_offset_p = chunk_tab->offsets; - - if (fwrite(chunk_tab, 1, chunk_tab->table_disk_size, out_fp) != - chunk_tab->table_disk_size) { - ERROR_WITH_ERRNO("Failed to write chunk table in compressed " - "file resource"); - ret = WIMLIB_ERR_WRITE; - goto out; - } - - *chunk_tab_ret = chunk_tab; -out: - return ret; -} - -/* - * Compresses a chunk of a WIM resource. - * - * @chunk: Uncompressed data of the chunk. - * @chunk_size: Size of the uncompressed chunk in bytes. - * @compressed_chunk: Pointer to output buffer of size at least - * (@chunk_size - 1) bytes. - * @compressed_chunk_len_ret: Pointer to an unsigned int into which the size - * of the compressed chunk will be - * returned. - * @ctype: Type of compression to use. Must be WIM_COMPRESSION_TYPE_LZX - * or WIM_COMPRESSION_TYPE_XPRESS. - * - * Returns zero if compressed succeeded, and nonzero if the chunk could not be - * compressed to any smaller than @chunk_size. This function cannot fail for - * any other reasons. - */ -static int compress_chunk(const u8 chunk[], unsigned chunk_size, - u8 compressed_chunk[], - unsigned *compressed_chunk_len_ret, - int ctype) -{ - int (*compress)(const void *, unsigned, void *, unsigned *); - switch (ctype) { - case WIM_COMPRESSION_TYPE_LZX: - compress = lzx_compress; - break; - case WIM_COMPRESSION_TYPE_XPRESS: - compress = xpress_compress; - break; - default: - wimlib_assert(0); - break; - } - return (*compress)(chunk, chunk_size, compressed_chunk, - compressed_chunk_len_ret); -} - -/* - * Writes a chunk of a WIM resource to an output file. - * - * @chunk: Uncompressed data of the chunk. - * @chunk_size: Size of the chunk (<= WIM_CHUNK_SIZE) - * @out_fp: FILE * to write tho chunk to. - * @out_ctype: Compression type to use when writing the chunk (ignored if no - * chunk table provided) - * @chunk_tab: Pointer to chunk table being created. It is updated with the - * offset of the chunk we write. - * - * Returns 0 on success; nonzero on failure. - */ -static int write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size, - FILE *out_fp, int out_ctype, - struct chunk_table *chunk_tab) +int read_full_wim_resource(const struct wim_lookup_table_entry *lte, u8 buf[], + int flags) { - const u8 *out_chunk; - unsigned out_chunk_size; - - wimlib_assert(chunk_size <= WIM_CHUNK_SIZE); - - if (!chunk_tab) { - out_chunk = chunk; - out_chunk_size = chunk_size; - } else { - u8 *compressed_chunk = alloca(chunk_size); - int ret; - - ret = compress_chunk(chunk, chunk_size, compressed_chunk, - &out_chunk_size, out_ctype); - if (ret == 0) { - out_chunk = compressed_chunk; - } else { - out_chunk = chunk; - out_chunk_size = chunk_size; - } - *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset; - chunk_tab->cur_offset += out_chunk_size; - } - - if (fwrite(out_chunk, 1, out_chunk_size, out_fp) != out_chunk_size) { - ERROR_WITH_ERRNO("Failed to write WIM resource chunk"); - return WIMLIB_ERR_WRITE; - } - return 0; -} - -/* - * Finishes a WIM chunk tale and writes it to the output file at the correct - * offset. - * - * The final size of the full compressed resource is returned in the - * @compressed_size_p. - */ -static int -finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab, - FILE *out_fp, u64 *compressed_size_p) -{ - size_t bytes_written; - if (fseeko(out_fp, chunk_tab->file_offset, SEEK_SET) != 0) { - ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of output " - "WIM file", chunk_tab->file_offset); - return WIMLIB_ERR_WRITE; - } - - if (chunk_tab->bytes_per_chunk_entry == 8) { - array_to_le64(chunk_tab->offsets, chunk_tab->num_chunks); - } else { - for (u64 i = 0; i < chunk_tab->num_chunks; i++) - ((u32*)chunk_tab->offsets)[i] = - to_le32(chunk_tab->offsets[i]); - } - bytes_written = fwrite((u8*)chunk_tab->offsets + - chunk_tab->bytes_per_chunk_entry, - 1, chunk_tab->table_disk_size, out_fp); - if (bytes_written != chunk_tab->table_disk_size) { - ERROR_WITH_ERRNO("Failed to write chunk table in compressed " - "file resource"); - return WIMLIB_ERR_WRITE; - } - if (fseeko(out_fp, 0, SEEK_END) != 0) { - ERROR_WITH_ERRNO("Failed to seek to end of output WIM file"); - return WIMLIB_ERR_WRITE; - } - *compressed_size_p = chunk_tab->cur_offset + chunk_tab->table_disk_size; - return 0; + return read_wim_resource(lte, buf, wim_resource_size(lte), 0, flags); } -/* - * Writes a WIM resource to a FILE * opened for writing. The resource may be - * written uncompressed or compressed depending on the @out_ctype parameter. - * - * If by chance the resource compresses to more than the original size (this may - * happen with random data or files than are pre-compressed), the resource is - * instead written uncompressed (and this is reflected in the @out_res_entry by - * removing the WIM_RESHDR_FLAG_COMPRESSED flag). - * - * @lte: The lookup table entry for the WIM resource. - * @out_fp: The FILE * to write the resource to. - * @out_ctype: The compression type of the resource to write. Note: if this is - * the same as the compression type of the WIM resource we - * need to read, we simply copy the data (i.e. we do not - * uncompress it, then compress it again). - * @out_res_entry: If non-NULL, a resource entry that is filled in with the - * offset, original size, compressed size, and compression flag - * of the output resource. +/* 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 + * resource is being extracted. * - * Returns 0 on success; nonzero on failure. - */ -static int write_wim_resource(struct lookup_table_entry *lte, - FILE *out_fp, int out_ctype, - struct resource_entry *out_res_entry) + * @extract_chunk is a function that is called to extract each chunk of the + * resource. */ +int extract_wim_resource(const struct wim_lookup_table_entry *lte, + u64 size, + extract_chunk_func_t extract_chunk, + void *extract_chunk_arg) { - u64 bytes_remaining; - u64 original_size; - u64 old_compressed_size; - u64 new_compressed_size; + u64 bytes_remaining = size; + u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)]; u64 offset = 0; int ret = 0; - struct chunk_table *chunk_tab = NULL; - bool raw; - off_t file_offset; -#ifdef WITH_NTFS_3G - ntfs_inode *ni = NULL; -#endif - - wimlib_assert(lte); - - /* Original size of the resource */ - original_size = wim_resource_size(lte); - - /* Compressed size of the resource (as it exists now) */ - old_compressed_size = wim_resource_compressed_size(lte); - - /* Current offset in output file */ - file_offset = ftello(out_fp); - if (file_offset == -1) { - ERROR_WITH_ERRNO("Failed to get offset in output " - "stream"); - return WIMLIB_ERR_WRITE; - } - - /* Are the compression types the same? If so, do a raw copy (copy - * without decompressing and recompressing the data). */ - raw = (wim_resource_compression_type(lte) == out_ctype - && out_ctype != WIM_COMPRESSION_TYPE_NONE); - if (raw) - bytes_remaining = old_compressed_size; - else - bytes_remaining = original_size; - - /* Empty resource; nothing needs to be done, so just return success. */ - if (bytes_remaining == 0) - return 0; - - /* Buffer for reading chunks for the resource */ - u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)]; - - /* If we are writing a compressed resource and not doing a raw copy, we - * need to initialize the chunk table */ - if (out_ctype != WIM_COMPRESSION_TYPE_NONE && !raw) { - ret = begin_wim_resource_chunk_tab(lte, out_fp, file_offset, - &chunk_tab); - if (ret != 0) - goto out; - } - - /* If the WIM resource is in an external file, open a FILE * to it so we - * don't have to open a temporary one in read_wim_resource() for each - * chunk. */ - if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK - && !lte->file_on_disk_fp) - { - wimlib_assert(lte->file_on_disk); - lte->file_on_disk_fp = fopen(lte->file_on_disk, "rb"); - if (!lte->file_on_disk_fp) { - ERROR_WITH_ERRNO("Failed to open the file `%s' for " - "reading", lte->file_on_disk); - ret = WIMLIB_ERR_OPEN; - goto out; - } - } -#ifdef WITH_NTFS_3G - else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME - && !lte->attr) - { - struct ntfs_location *loc = lte->ntfs_loc; - wimlib_assert(loc); - ni = ntfs_pathname_to_inode(*loc->ntfs_vol_p, NULL, loc->path_utf8); - if (!ni) { - ERROR_WITH_ERRNO("Failed to open inode `%s' in NTFS " - "volume", loc->path_utf8); - ret = WIMLIB_ERR_NTFS_3G; - goto out; - } - lte->attr = ntfs_attr_open(ni, - loc->is_reparse_point ? AT_REPARSE_POINT : AT_DATA, - (ntfschar*)loc->stream_name_utf16, - loc->stream_name_utf16_num_chars); - if (!lte->attr) { - ERROR_WITH_ERRNO("Failed to open attribute of `%s' in " - "NTFS volume", loc->path_utf8); - ret = WIMLIB_ERR_NTFS_3G; - goto out_fclose; - } - } -#endif - - /* If we aren't doing a raw copy, we will compute the SHA1 message - * digest of the resource as we read it, and verify it's the same as the - * hash given in the lookup table entry once we've finished reading the - * resource. */ + u8 hash[SHA1_HASH_SIZE]; + bool check_hash = (size == wim_resource_size(lte)); SHA_CTX ctx; - if (!raw) + + if (check_hash) sha1_init(&ctx); - /* While there are still bytes remaining in the WIM resource, read a - * chunk of the resource, update SHA1, then write that chunk using the - * desired compression type. */ - do { - u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE); - ret = read_wim_resource(lte, buf, to_read, offset, raw); + while (bytes_remaining) { + u64 to_read = min(bytes_remaining, sizeof(buf)); + ret = read_wim_resource(lte, buf, to_read, offset, 0); if (ret != 0) - goto out_fclose; - if (!raw) + return ret; + if (check_hash) sha1_update(&ctx, buf, to_read); - ret = write_wim_resource_chunk(buf, to_read, out_fp, - out_ctype, chunk_tab); - if (ret != 0) - goto out_fclose; + 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; - } while (bytes_remaining); - - /* Raw copy: The new compressed size is the same as the old compressed - * size - * - * Using WIM_COMPRESSION_TYPE_NONE: The new compressed size is the - * original size - * - * Using a different compression type: Call - * finish_wim_resource_chunk_tab() and it will provide the new - * compressed size. - */ - if (raw) { - new_compressed_size = old_compressed_size; - } else { - if (out_ctype == WIM_COMPRESSION_TYPE_NONE) - new_compressed_size = original_size; - else { - ret = finish_wim_resource_chunk_tab(chunk_tab, out_fp, - &new_compressed_size); - if (ret != 0) - goto out_fclose; - } - } - - /* Verify SHA1 message digest of the resource, unless we are doing a raw - * write (in which case we never even saw the uncompressed data). Or, - * if the hash we had before is all 0's, just re-set it to be the new - * hash. */ - if (!raw) { - u8 md[SHA1_HASH_SIZE]; - sha1_final(md, &ctx); - if (is_zero_hash(lte->hash)) { - copy_hash(lte->hash, md); - } else if (!hashes_equal(md, lte->hash)) { - ERROR("WIM resource has incorrect hash!"); - if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK) { - ERROR("We were reading it from `%s'; maybe it changed " - "while we were reading it.", - lte->file_on_disk); - } - ret = WIMLIB_ERR_INVALID_RESOURCE_HASH; - goto out_fclose; - } } - - if (!raw && new_compressed_size >= original_size && - out_ctype != WIM_COMPRESSION_TYPE_NONE) - { - /* Oops! We compressed the resource to larger than the original - * size. Write the resource uncompressed instead. */ - if (fseeko(out_fp, file_offset, SEEK_SET) != 0) { - ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" " - "of output WIM file", file_offset); - ret = WIMLIB_ERR_WRITE; - goto out_fclose; - } - ret = write_wim_resource(lte, out_fp, WIM_COMPRESSION_TYPE_NONE, - out_res_entry); - if (ret != 0) - goto out_fclose; - if (fflush(out_fp) != 0) { - ERROR_WITH_ERRNO("Failed to flush output WIM file"); - ret = WIMLIB_ERR_WRITE; - goto out_fclose; - } - if (ftruncate(fileno(out_fp), file_offset + out_res_entry->size) != 0) { - ERROR_WITH_ERRNO("Failed to truncate output WIM file"); - ret = WIMLIB_ERR_WRITE; + 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; } - goto out_fclose; - } - wimlib_assert(new_compressed_size <= original_size || raw); - if (out_res_entry) { - out_res_entry->size = new_compressed_size; - out_res_entry->original_size = original_size; - out_res_entry->offset = file_offset; - out_res_entry->flags = lte->resource_entry.flags - & ~WIM_RESHDR_FLAG_COMPRESSED; - if (out_ctype != WIM_COMPRESSION_TYPE_NONE) - out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED; } -out_fclose: - if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK - && lte->file_on_disk_fp) { - fclose(lte->file_on_disk_fp); - lte->file_on_disk_fp = NULL; - } -#ifdef WITH_NTFS_3G - else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME) { - if (lte->attr) { - ntfs_attr_close(lte->attr); - lte->attr = NULL; - } if (ni) { - ntfs_inode_close(ni); - } - } -#endif -out: - FREE(chunk_tab); - return ret; -} - -/* Like write_wim_resource(), but the resource is specified by a buffer of - * uncompressed data rather a lookup table entry; also writes the SHA1 hash of - * the buffer to @hash. */ -static int write_wim_resource_from_buffer(const u8 *buf, u64 buf_size, - FILE *out_fp, int out_ctype, - struct resource_entry *out_res_entry, - u8 hash[SHA1_HASH_SIZE]) -{ - /* Set up a temporary lookup table entry that we provide to - * write_wim_resource(). */ - struct lookup_table_entry lte; - int ret; - lte.resource_entry.flags = 0; - lte.resource_entry.original_size = buf_size; - lte.resource_entry.size = buf_size; - lte.resource_entry.offset = 0; - lte.resource_location = RESOURCE_IN_ATTACHED_BUFFER; - lte.attached_buffer = (u8*)buf; - - zero_out_hash(lte.hash); - ret = write_wim_resource(<e, out_fp, out_ctype, out_res_entry); - if (ret != 0) - return ret; - copy_hash(hash, lte.hash); return 0; } -/* - * Extracts the first @size bytes of the WIM resource specified by @lte to the - * open file descriptor @fd. - * - * Returns 0 on success; nonzero on failure. - */ -int extract_wim_resource_to_fd(const struct lookup_table_entry *lte, int fd, - u64 size) +/* 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) { - u64 bytes_remaining = size; - u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)]; - u64 offset = 0; - int ret = 0; - u8 hash[SHA1_HASH_SIZE]; - - SHA_CTX ctx; - sha1_init(&ctx); - - while (bytes_remaining) { - u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE); - ret = read_wim_resource(lte, buf, to_read, offset, false); - if (ret != 0) - break; - sha1_update(&ctx, buf, to_read); - if (full_write(fd, buf, to_read) < 0) { - ERROR_WITH_ERRNO("Error extracting WIM resource"); - return WIMLIB_ERR_WRITE; + const char *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; } - bytes_remaining -= to_read; - offset += to_read; - } - sha1_final(hash, &ctx); - if (!hashes_equal(hash, lte->hash)) { - ERROR("Invalid checksum on a WIM resource " - "(detected when extracting to external file)"); - ERROR("The following WIM resource is invalid:"); - print_lookup_table_entry(lte); - return WIMLIB_ERR_INVALID_RESOURCE_HASH; + total += ret; + p += ret; } - return 0; + return total; } -/* - * Extracts the WIM resource specified by @lte to the open file descriptor @fd. - * - * Returns 0 on success; nonzero on failure. - */ -int extract_full_wim_resource_to_fd(const struct lookup_table_entry *lte, int fd) +int extract_wim_chunk_to_fd(const u8 *buf, size_t len, u64 offset, void *arg) { - return extract_wim_resource_to_fd(lte, fd, wim_resource_size(lte)); + 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; + } } -/* +/* * 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. @@ -1049,10 +696,9 @@ int extract_full_wim_resource_to_fd(const struct lookup_table_entry *lte, int fd * The output_resource_entry, out_refcnt, and part_number fields of @lte are * updated. * - * Metadata resources are not copied (they are handled elsewhere for joining and - * splitting). + * (This function is confusing and should be refactored somehow.) */ -int copy_resource(struct lookup_table_entry *lte, void *wim) +int copy_resource(struct wim_lookup_table_entry *lte, void *wim) { WIMStruct *w = wim; int ret; @@ -1062,295 +708,11 @@ int copy_resource(struct lookup_table_entry *lte, void *wim) return 0; ret = write_wim_resource(lte, w->out_fp, - wim_resource_compression_type(lte), - <e->output_resource_entry); + wim_resource_compression_type(lte), + <e->output_resource_entry, 0); if (ret != 0) return ret; lte->out_refcnt = lte->refcnt; lte->part_number = w->hdr.part_number; return 0; } - -/* - * Writes a dentry's resources, including the main file resource as well as all - * alternate data streams, to the output file. - * - * @dentry: The dentry for the file. - * @wim_p: A pointer to the WIMStruct containing @dentry. - * - * @return zero on success, nonzero on failure. - */ -int write_dentry_resources(struct dentry *dentry, void *wim_p) -{ - WIMStruct *w = wim_p; - int ret = 0; - struct lookup_table_entry *lte; - int ctype = wimlib_get_compression_type(w); - - if (w->write_flags & WIMLIB_WRITE_FLAG_VERBOSE) { - wimlib_assert(dentry->full_path_utf8); - printf("Writing streams for `%s'\n", dentry->full_path_utf8); - } - - for (unsigned i = 0; i <= dentry->num_ads; i++) { - lte = dentry_stream_lte(dentry, i, w->lookup_table); - if (lte && ++lte->out_refcnt == 1) { - ret = write_wim_resource(lte, w->out_fp, ctype, - <e->output_resource_entry); - if (ret != 0) - break; - } - } - return ret; -} - -/* - * Reads the metadata metadata resource from the WIM file. The metadata - * resource consists of the security data, followed by the directory entry for - * the root directory, followed by all the other directory entries in the - * filesystem. The subdir_offset field of each directory entry gives the start - * of its child entries from the beginning of the metadata resource. An - * end-of-directory is signaled by a directory entry of length '0', really of - * length 8, because that's how long the 'length' field is. - * - * @fp: The FILE* for the input WIM file. - * @wim_ctype: The compression type of the WIM file. - * @imd: Pointer to the image metadata structure. Its `metadata_lte' - * member specifies the lookup table entry for the metadata - * resource. The rest of the image metadata entry will be filled - * in by this function. - * - * @return: Zero on success, nonzero on failure. - */ -int read_metadata_resource(WIMStruct *w, struct image_metadata *imd) -{ - u8 *buf; - u32 dentry_offset; - int ret; - struct dentry *dentry; - struct link_group_table *lgt; - const struct lookup_table_entry *metadata_lte; - u64 metadata_len; - u64 metadata_offset; - - metadata_lte = imd->metadata_lte; - metadata_len = wim_resource_size(metadata_lte); - metadata_offset = metadata_lte->resource_entry.offset; - - DEBUG("Reading metadata resource: length = %"PRIu64", " - "offset = %"PRIu64"", metadata_len, metadata_offset); - - /* There is no way the metadata resource could possibly be less than (8 - * + WIM_DENTRY_DISK_SIZE) bytes, where the 8 is for security data (with - * no security descriptors) and WIM_DENTRY_DISK_SIZE is for the root - * dentry. */ - if (metadata_len < 8 + WIM_DENTRY_DISK_SIZE) { - ERROR("Expected at least %u bytes for the metadata resource", - 8 + WIM_DENTRY_DISK_SIZE); - return WIMLIB_ERR_INVALID_RESOURCE_SIZE; - } - - /* Allocate memory for the uncompressed metadata resource. */ - buf = MALLOC(metadata_len); - - if (!buf) { - ERROR("Failed to allocate %"PRIu64" bytes for uncompressed " - "metadata resource", metadata_len); - return WIMLIB_ERR_NOMEM; - } - - /* Read the metadata resource into memory. (It may be compressed.) */ - ret = read_full_wim_resource(metadata_lte, buf); - if (ret != 0) - goto out_free_buf; - - DEBUG("Finished reading metadata resource into memory."); - - /* The root directory entry starts after security data, aligned on an - * 8-byte boundary within the metadata resource. - * - * The security data starts with a 4-byte integer giving its total - * length, so if we round that up to an 8-byte boundary that gives us - * the offset of the root dentry. - * - * Here we read the security data into a wim_security_data structure, - * and if successful, go ahead and calculate the offset in the metadata - * resource of the root dentry. */ - - ret = read_security_data(buf, metadata_len, &imd->security_data); - if (ret != 0) - goto out_free_buf; - - get_u32(buf, &dentry_offset); - if (dentry_offset == 0) - dentry_offset = 8; - dentry_offset = (dentry_offset + 7) & ~7; - - /* Allocate memory for the root dentry and read it into memory */ - dentry = MALLOC(sizeof(struct dentry)); - if (!dentry) { - ERROR("Failed to allocate %zu bytes for root dentry", - sizeof(struct dentry)); - ret = WIMLIB_ERR_NOMEM; - goto out_free_security_data; - } - - ret = read_dentry(buf, metadata_len, dentry_offset, dentry); - - /* This is the root dentry, so set its pointers correctly. */ - dentry->parent = dentry; - dentry->next = dentry; - dentry->prev = dentry; - if (ret != 0) - goto out_free_dentry_tree; - - /* Now read the entire directory entry tree into memory. */ - DEBUG("Reading dentry tree"); - ret = read_dentry_tree(buf, metadata_len, dentry); - if (ret != 0) - goto out_free_dentry_tree; - - /* Calculate the full paths in the dentry tree. */ - DEBUG("Calculating dentry full paths"); - ret = for_dentry_in_tree(dentry, calculate_dentry_full_path, NULL); - if (ret != 0) - goto out_free_dentry_tree; - - /* Build hash table that maps hard link group IDs to dentry sets */ - DEBUG("Building link group table"); - lgt = new_link_group_table(9001); - if (!lgt) - goto out_free_dentry_tree; - ret = for_dentry_in_tree(dentry, link_group_table_insert, lgt); - if (ret != 0) - goto out_free_lgt; - - DEBUG("Fixing inconsistencies in the link groups"); - ret = fix_link_groups(lgt); - if (ret != 0) - goto out_free_lgt; - - DEBUG("Running miscellaneous verifications on the dentry tree"); - ret = for_dentry_in_tree(dentry, verify_dentry, w); - if (ret != 0) - goto out_free_lgt; - - DEBUG("Done reading image metadata"); - - imd->lgt = lgt; - imd->root_dentry = dentry; - goto out_free_buf; -out_free_lgt: - free_link_group_table(lgt); -out_free_dentry_tree: - free_dentry_tree(dentry, NULL); -out_free_security_data: - free_security_data(imd->security_data); - imd->security_data = NULL; -out_free_buf: - FREE(buf); - return ret; -} - -/* Write the metadata resource for the current WIM image. */ -int write_metadata_resource(WIMStruct *w) -{ - u8 *buf; - u8 *p; - int ret; - u64 subdir_offset; - struct dentry *root; - struct lookup_table_entry *lte; - u64 metadata_original_size; - const struct wim_security_data *sd; - const unsigned random_tail_len = 20; - - DEBUG("Writing metadata resource for image %d", w->current_image); - - root = wim_root_dentry(w); - sd = wim_security_data(w); - - /* We do not allow the security data pointer to be NULL, although it may - * point to an empty security data with no entries. */ - wimlib_assert(sd); - - /* Offset of first child of the root dentry. It's equal to: - * - The total length of the security data, rounded to the next 8-byte - * boundary, - * - plus the total length of the root dentry, - * - plus 8 bytes for an end-of-directory entry following the root - * dentry (shouldn't really be needed, but just in case...) - */ - subdir_offset = ((sd->total_length + 7) & ~7) + - dentry_correct_total_length(root) + 8; - - /* Calculate the subdirectory offsets for the entire dentry tree. */ - calculate_subdir_offsets(root, &subdir_offset); - - /* Total length of the metadata resource (uncompressed) */ - metadata_original_size = subdir_offset + random_tail_len; - - /* Allocate a buffer to contain the uncompressed metadata resource */ - buf = MALLOC(metadata_original_size); - if (!buf) { - ERROR("Failed to allocate %"PRIu64" bytes for " - "metadata resource", metadata_original_size); - return WIMLIB_ERR_NOMEM; - } - - /* Write the security data into the resource buffer */ - p = write_security_data(sd, buf); - - /* Write the dentry tree into the resource buffer */ - DEBUG("Writing dentry tree."); - p = write_dentry_tree(root, p); - - /* - * Append 20 random bytes to the metadata resource so that we don't have - * identical metadata resources if we happen to append exactly the same - * image twice without any changes in timestamps. If this were to - * happen, it would cause confusion about the number and order of images - * in the WIM. - */ - randomize_byte_array(p, random_tail_len); - - /* We MUST have exactly filled the buffer; otherwise we calculated its - * size incorrectly or wrote the data incorrectly. */ - wimlib_assert(p - buf + random_tail_len == metadata_original_size); - - /* Get the lookup table entry for the metadata resource so we can update - * it. */ - lte = wim_metadata_lookup_table_entry(w); - - /* Write the metadata resource to the output WIM using the proper - * compression type. The lookup table entry for the metadata resource - * is updated. */ - ret = write_wim_resource_from_buffer(buf, metadata_original_size, - w->out_fp, - wimlib_get_compression_type(w), - <e->output_resource_entry, - lte->hash); - if (ret != 0) - goto out; - - /* It's very likely the SHA1 message digest of the metadata resource, so - * re-insert the lookup table entry into the lookup table. */ - lookup_table_unlink(w->lookup_table, lte); - lookup_table_insert(w->lookup_table, lte); - - /* We do not allow a metadata resource to be referenced multiple times, - * and the 20 random bytes appended to it should make it extremely - * likely for each metadata resource to be unique, even if the exact - * same image is captured. */ - wimlib_assert(lte->out_refcnt == 0); - lte->out_refcnt = 1; - - /* Make sure that the resource entry is written marked with the metadata - * flag. */ - lte->output_resource_entry.flags |= WIM_RESHDR_FLAG_METADATA; -out: - /* All the data has been written to the new WIM; no need for the buffer - * anymore */ - FREE(buf); - return ret; -}