X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fresource.c;h=01d481df0f97057a074e7168270e0616e9359c2e;hp=f21fd5148355ab2c0920980a4e45306a66f9139e;hb=6cca349b45e66a2b2b82aa5dcd269a4bf61c50db;hpb=d5b841b4d3243c7c6922d9254fb4e5b9f0b58d41 diff --git a/src/resource.c b/src/resource.c index f21fd514..01d481df 100644 --- a/src/resource.c +++ b/src/resource.c @@ -27,11 +27,7 @@ #include #include -#ifdef WITH_NTFS_3G -#include -#include -#include -#endif +#include "dentry.h" #include "wimlib_internal.h" #include "lookup_table.h" @@ -39,11 +35,14 @@ #include "lzx.h" #include "xpress.h" #include "sha1.h" -#include "dentry.h" #include #include -#ifdef HAVE_ALLOCA_H -#include + +#ifdef WITH_NTFS_3G +#include +#include +#include +#include #endif @@ -420,6 +419,57 @@ u8 *put_resource_entry(u8 *p, const struct resource_entry *entry) return p; } +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; +} + /* * Reads some data from the resource corresponding to a WIM lookup table entry. * @@ -427,13 +477,11 @@ 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) + size_t size, u64 offset, int flags) { int ctype; int ret = 0; @@ -441,7 +489,7 @@ int read_wim_resource(const struct lookup_table_entry *lte, u8 buf[], /* 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); @@ -452,23 +500,38 @@ int read_wim_resource(const struct lookup_table_entry *lte, u8 buf[], * the lte->wim member. The resource may be either compressed * or uncompressed. */ wimlib_assert(lte->wim != NULL); - wimlib_assert(lte->wim->fp != NULL); + + if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) { + fp = wim_get_fp(lte->wim); + if (!fp) + return WIMLIB_ERR_OPEN; + } else { + wimlib_assert(lte->wim->fp != NULL); + fp = lte->wim->fp; + } + ctype = wim_resource_compression_type(lte); wimlib_assert(ctype != WIM_COMPRESSION_TYPE_NONE || (lte->resource_entry.original_size == lte->resource_entry.size)); - if (raw || ctype == WIM_COMPRESSION_TYPE_NONE) - ret = read_uncompressed_resource(lte->wim->fp, + if ((flags & WIMLIB_RESOURCE_FLAG_RAW) + || ctype == WIM_COMPRESSION_TYPE_NONE) + ret = read_uncompressed_resource(fp, lte->resource_entry.offset + offset, size, buf); else - ret = read_compressed_resource(lte->wim->fp, + ret = read_compressed_resource(fp, lte->resource_entry.size, lte->resource_entry.original_size, lte->resource_entry.offset, ctype, size, offset, buf); + if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) { + int ret2 = wim_release_fp(lte->wim, fp); + if (ret == 0) + ret = ret2; + } break; case RESOURCE_IN_STAGING_FILE: case RESOURCE_IN_FILE_ON_DISK: @@ -533,444 +596,10 @@ 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[]) +int read_full_wim_resource(const struct lookup_table_entry *lte, u8 buf[], + int flags) { - 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; - size_t alloc_size = sizeof(struct chunk_table) + num_chunks * sizeof(u64); - struct chunk_table *chunk_tab = CALLOC(1, alloc_size); - int ret; - - 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; - } - - ret = 0; -out: - *chunk_tab_ret = chunk_tab; - 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) -{ - 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_cpu_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] = - cpu_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; -} - -/* - * 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. - * - * 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) -{ - u64 bytes_remaining; - u64 original_size; - u64 old_compressed_size; - u64 new_compressed_size; - u64 offset; - int ret; - 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. */ - SHA_CTX ctx; - if (!raw) - 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. */ - offset = 0; - do { - u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE); - ret = read_wim_resource(lte, buf, to_read, offset, raw); - if (ret != 0) - goto out_fclose; - if (!raw) - 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; - 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; - goto out_fclose; - } - } else { - 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; - } - } - ret = 0; -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; + return read_wim_resource(lte, buf, wim_resource_size(lte), 0, flags); } /* Like write_wim_resource(), but the resource is specified by a buffer of @@ -993,7 +622,7 @@ static int write_wim_resource_from_buffer(const u8 *buf, u64 buf_size, lte.attached_buffer = (u8*)buf; zero_out_hash(lte.hash); - ret = write_wim_resource(<e, out_fp, out_ctype, out_res_entry); + ret = write_wim_resource(<e, out_fp, out_ctype, out_res_entry, 0); if (ret != 0) return ret; copy_hash(hash, lte.hash); @@ -1020,7 +649,7 @@ int extract_wim_resource_to_fd(const struct lookup_table_entry *lte, int fd, while (bytes_remaining) { u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE); - ret = read_wim_resource(lte, buf, to_read, offset, false); + ret = read_wim_resource(lte, buf, to_read, offset, 0); if (ret != 0) break; sha1_update(&ctx, buf, to_read); @@ -1074,7 +703,7 @@ int copy_resource(struct lookup_table_entry *lte, void *wim) ret = write_wim_resource(lte, w->out_fp, wim_resource_compression_type(lte), - <e->output_resource_entry); + <e->output_resource_entry, 0); if (ret != 0) return ret; lte->out_refcnt = lte->refcnt; @@ -1082,39 +711,6 @@ int copy_resource(struct lookup_table_entry *lte, void *wim) 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->d_inode->num_ads; i++) { - lte = inode_stream_lte(dentry->d_inode, 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 @@ -1178,7 +774,7 @@ int read_metadata_resource(WIMStruct *w, struct image_metadata *imd) } /* Read the metadata resource into memory. (It may be compressed.) */ - ret = read_full_wim_resource(metadata_lte, buf); + ret = read_full_wim_resource(metadata_lte, buf, 0); if (ret != 0) goto out_free_buf;