X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fwrite.c;h=1891c89b20685a0caf05b3ce3bb8cb6755c51566;hp=139ecdabc73dd74da20fc45cfaaa772dd8478e59;hb=fc8276d0a3efb3df5f7512b3fa9499eb1b3449eb;hpb=9fb3aaca115429b0af2a623bf20bfceef74f047f diff --git a/src/write.c b/src/write.c index 139ecdab..1891c89b 100644 --- a/src/write.c +++ b/src/write.c @@ -33,7 +33,7 @@ #endif #ifdef __WIN32__ -# include +# include "win32.h" #endif #include "list.h" @@ -42,8 +42,6 @@ #include "dentry.h" #include "lookup_table.h" #include "xml.h" -#include "lzx.h" -#include "xpress.h" #ifdef ENABLE_MULTITHREADED_COMPRESSION # include @@ -150,29 +148,38 @@ out: } /* - * Pointer to function to compresses a chunk of a WIM resource. + * compress_func_t- Pointer to a function to compresses a chunk + * of a WIM resource. This may be either + * wimlib_xpress_compress() (xpress-compress.c) or + * wimlib_lzx_compress() (lzx-compress.c). * - * @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. + * @chunk: Uncompressed data of the chunk. + * @chunk_size: Size of the uncompressed chunk, in bytes. + * @out: Pointer to output buffer of size at least (@chunk_size - 1) bytes. + * + * Returns the size of the compressed data written to @out in bytes, or 0 if the + * data could not be compressed to (@chunk_size - 1) bytes or fewer. * - * 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. + * As a special requirement, the compression code is optimized for the WIM + * format and therefore requires (@chunk_size <= 32768). + * + * As another special requirement, the compression code will read up to 8 bytes + * off the end of the @chunk array for performance reasons. The values of these + * bytes will not affect the output of the compression, but the calling code + * must make sure that the buffer holding the uncompressed chunk is actually at + * least (@chunk_size + 8) bytes, or at least that these extra bytes are in + * mapped memory that will not cause a memory access violation if accessed. */ -typedef int (*compress_func_t)(const void *, unsigned, void *, unsigned *); +typedef unsigned (*compress_func_t)(const void *chunk, unsigned chunk_size, + void *out); compress_func_t get_compress_func(int out_ctype) { if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX) - return lzx_compress; + return wimlib_lzx_compress; else - return xpress_compress; + return wimlib_xpress_compress; } /* @@ -180,36 +187,37 @@ get_compress_func(int out_ctype) * * @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) + * @out_fp: FILE * to write the chunk to. + * @compress: Compression function to use (NULL if writing uncompressed + * data). * @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, +write_wim_resource_chunk(const void *chunk, unsigned chunk_size, FILE *out_fp, compress_func_t compress, struct chunk_table *chunk_tab) { const u8 *out_chunk; unsigned out_chunk_size; - if (chunk_tab) { + if (compress) { u8 *compressed_chunk = alloca(chunk_size); - int ret; - ret = compress(chunk, chunk_size, compressed_chunk, - &out_chunk_size); - if (ret == 0) { + out_chunk_size = compress(chunk, chunk_size, compressed_chunk); + if (out_chunk_size) { + /* Write compressed */ out_chunk = compressed_chunk; } else { + /* Write uncompressed */ out_chunk = chunk; out_chunk_size = chunk_size; } *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset; chunk_tab->cur_offset += out_chunk_size; } else { + /* Write uncompressed */ out_chunk = chunk; out_chunk_size = chunk_size; } @@ -261,107 +269,6 @@ finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab, return 0; } -/* Prepare for multiple reads to a resource by caching a FILE * or NTFS - * attribute pointer in the lookup table entry. */ -static int -prepare_resource_for_read(struct wim_lookup_table_entry *lte - - #ifdef WITH_NTFS_3G - , ntfs_inode **ni_ret - #endif - ) -{ - switch (lte->resource_location) { - case RESOURCE_IN_FILE_ON_DISK: - if (!lte->file_on_disk_fp) { - 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'", lte->file_on_disk); - return WIMLIB_ERR_OPEN; - } - } - break; -#ifdef WITH_NTFS_3G - case RESOURCE_IN_NTFS_VOLUME: - if (!lte->attr) { - struct ntfs_location *loc = lte->ntfs_loc; - ntfs_inode *ni; - wimlib_assert(loc); - ni = ntfs_pathname_to_inode(*loc->ntfs_vol_p, NULL, loc->path); - if (!ni) { - ERROR_WITH_ERRNO("Failed to open inode `%s' in NTFS " - "volume", loc->path); - return WIMLIB_ERR_NTFS_3G; - } - lte->attr = ntfs_attr_open(ni, - loc->is_reparse_point ? AT_REPARSE_POINT : AT_DATA, - loc->stream_name, - loc->stream_name_nchars); - if (!lte->attr) { - ERROR_WITH_ERRNO("Failed to open attribute of `%s' in " - "NTFS volume", loc->path); - ntfs_inode_close(ni); - return WIMLIB_ERR_NTFS_3G; - } - *ni_ret = ni; - } - break; -#endif -#ifdef __WIN32__ - case RESOURCE_WIN32: - if (lte->win32_file_on_disk_fp == INVALID_HANDLE_VALUE) { - lte->win32_file_on_disk_fp = - win32_open_file_readonly(lte->win32_file_on_disk, true); - if (lte->win32_file_on_disk_fp == INVALID_HANDLE_VALUE) { - ERROR("Win32 API: Can't open %ls", lte->win32_file_on_disk); - win32_error_last(); - return WIMLIB_ERR_OPEN; - } - } - break; -#endif - default: - break; - } - return 0; -} - -/* Undo prepare_resource_for_read() by closing the cached FILE * or NTFS - * attribute. */ -static void -end_wim_resource_read(struct wim_lookup_table_entry *lte - #ifdef WITH_NTFS_3G - , ntfs_inode *ni - #endif - ) -{ - 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 -#ifdef __WIN32__ - else if (lte->resource_location == RESOURCE_WIN32 - && lte->win32_file_on_disk_fp != INVALID_HANDLE_VALUE) - { - win32_close_file(lte->win32_file_on_disk_fp); - lte->win32_file_on_disk_fp = INVALID_HANDLE_VALUE; - } -#endif -} - static int write_uncompressed_resource_and_truncate(struct wim_lookup_table_entry *lte, FILE *out_fp, @@ -374,217 +281,157 @@ write_uncompressed_resource_and_truncate(struct wim_lookup_table_entry *lte, "output WIM file", file_offset); return WIMLIB_ERR_WRITE; } - ret = write_wim_resource(lte, out_fp, WIMLIB_COMPRESSION_TYPE_NONE, - out_res_entry, 0); - if (ret != 0) + ret = write_wim_resource(lte, out_fp, + WIMLIB_COMPRESSION_TYPE_NONE, + out_res_entry, + 0); + if (ret) return ret; return fflush_and_ftruncate(out_fp, file_offset + wim_resource_size(lte)); } -/* - * 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. - */ +struct write_resource_ctx { + compress_func_t compress; + struct chunk_table *chunk_tab; + FILE *out_fp; + SHA_CTX sha_ctx; + bool doing_sha; +}; + +static int +write_resource_cb(const void *chunk, size_t chunk_size, void *_ctx) +{ + struct write_resource_ctx *ctx = _ctx; + + if (ctx->doing_sha) + sha1_update(&ctx->sha_ctx, chunk, chunk_size); + + if (ctx->compress) { + return write_wim_resource_chunk(chunk, chunk_size, + ctx->out_fp, ctx->compress, + ctx->chunk_tab); + } else { + if (fwrite(chunk, 1, chunk_size, ctx->out_fp) != chunk_size) { + ERROR_WITH_ERRNO("Error writing to output WIM"); + return WIMLIB_ERR_WRITE; + } else { + return 0; + } + } +} + int write_wim_resource(struct wim_lookup_table_entry *lte, FILE *out_fp, int out_ctype, struct resource_entry *out_res_entry, int flags) { - u64 bytes_remaining; - u64 original_size; - u64 old_compressed_size; - u64 new_compressed_size; - u64 offset; + struct write_resource_ctx write_ctx; + u64 new_size; + off_t offset; int ret; - struct chunk_table *chunk_tab = NULL; - bool raw; - off_t file_offset; - compress_func_t compress = NULL; -#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); + if (wim_resource_size(lte) == 0) { + /* Empty resource; nothing needs to be done, so just return + * success. */ + return 0; + } - /* Current offset in output file */ - file_offset = ftello(out_fp); - if (file_offset == -1) { - ERROR_WITH_ERRNO("Failed to get offset in output " - "stream"); + offset = ftello(out_fp); + if (offset == -1) { + ERROR_WITH_ERRNO("Can't get position in output WIM"); 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 != WIMLIB_COMPRESSION_TYPE_NONE - && !(flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS)); + /* Can we simply copy the compressed data without recompressing it? */ - if (raw) { + if (!(flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS) && + lte->resource_location == RESOURCE_IN_WIM && + wimlib_get_compression_type(lte->wim) == out_ctype) + { flags |= WIMLIB_RESOURCE_FLAG_RAW; - bytes_remaining = old_compressed_size; + write_ctx.doing_sha = false; } else { - flags &= ~WIMLIB_RESOURCE_FLAG_RAW; - 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 != WIMLIB_COMPRESSION_TYPE_NONE && !raw) { - ret = begin_wim_resource_chunk_tab(lte, out_fp, file_offset, - &chunk_tab); - if (ret != 0) - goto out; + write_ctx.doing_sha = true; + sha1_init(&write_ctx.sha_ctx); } - /* 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. */ -#ifdef WITH_NTFS_3G - ret = prepare_resource_for_read(lte, &ni); -#else - ret = prepare_resource_for_read(lte); -#endif - if (ret != 0) - goto out; - - /* 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); - compress = get_compress_func(out_ctype); - } - offset = 0; - - /* 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, flags); - if (ret != 0) - goto out_fclose; - if (!raw) - sha1_update(&ctx, buf, to_read); - ret = write_wim_resource_chunk(buf, to_read, out_fp, - compress, 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 WIMLIB_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; + /* Initialize the chunk table and set the compression function if + * compressing the resource. */ + if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE || + (flags & WIMLIB_RESOURCE_FLAG_RAW)) { + write_ctx.compress = NULL; + write_ctx.chunk_tab = NULL; } else { - if (out_ctype == WIMLIB_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; - } + write_ctx.compress = get_compress_func(out_ctype); + ret = begin_wim_resource_chunk_tab(lte, out_fp, + offset, + &write_ctx.chunk_tab); + if (ret) + return ret; } - /* 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) { + /* Write the data */ + write_ctx.out_fp = out_fp; + ret = read_resource_prefix(lte, wim_resource_size(lte), + write_resource_cb, &write_ctx, 0); + + /* Verify SHA1 message digest of the resource, or set the hash for the + * first time. */ + if (write_ctx.doing_sha) { u8 md[SHA1_HASH_SIZE]; - sha1_final(md, &ctx); - if (is_zero_hash(lte->hash)) { + sha1_final(md, &write_ctx.sha_ctx); + if (lte->unhashed) { 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.", + if (lte_filename_valid(lte)) { + ERROR("We were reading it from \"%"TS"\"; maybe " + "it changed while we were reading it.", lte->file_on_disk); } ret = WIMLIB_ERR_INVALID_RESOURCE_HASH; - goto out_fclose; + goto out_free_chunk_tab; } } - if (!raw && new_compressed_size >= original_size && - out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) - { - /* Oops! We compressed the resource to larger than the original - * size. Write the resource uncompressed instead. */ - ret = write_uncompressed_resource_and_truncate(lte, - out_fp, - file_offset, - out_res_entry); - if (ret != 0) - goto out_fclose; + out_res_entry->flags = lte->resource_entry.flags; + out_res_entry->original_size = wim_resource_size(lte); + out_res_entry->offset = offset; + if (flags & WIMLIB_RESOURCE_FLAG_RAW) { + /* Doing a raw write: The new compressed size is the same as + * the compressed size in the other WIM. */ + new_size = lte->resource_entry.size; + } else if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE) { + /* Using WIMLIB_COMPRESSION_TYPE_NONE: The new compressed size + * is the original size. */ + new_size = lte->resource_entry.original_size; + out_res_entry->flags &= ~WIM_RESHDR_FLAG_COMPRESSED; } 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 != WIMLIB_COMPRESSION_TYPE_NONE) - out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED; + /* Using a different compression type: Call + * finish_wim_resource_chunk_tab() and it will provide the new + * compressed size. */ + ret = finish_wim_resource_chunk_tab(write_ctx.chunk_tab, out_fp, + &new_size); + if (ret) + goto out_free_chunk_tab; + if (new_size >= wim_resource_size(lte)) { + /* Oops! We compressed the resource to larger than the original + * size. Write the resource uncompressed instead. */ + ret = write_uncompressed_resource_and_truncate(lte, + out_fp, + offset, + out_res_entry); + goto out_free_chunk_tab; } + out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED; } + out_res_entry->size = new_size; ret = 0; -out_fclose: -#ifdef WITH_NTFS_3G - end_wim_resource_read(lte, ni); -#else - end_wim_resource_read(lte); -#endif -out: - FREE(chunk_tab); +out_free_chunk_tab: + FREE(write_ctx.chunk_tab); return ret; } @@ -688,15 +535,18 @@ compress_chunks(struct message *msg, compress_func_t compress) { for (unsigned i = 0; i < msg->num_chunks; i++) { DEBUG2("compress chunk %u of %u", i, msg->num_chunks); - int ret = compress(msg->uncompressed_chunks[i], - msg->uncompressed_chunk_sizes[i], - msg->compressed_chunks[i], - &msg->compressed_chunk_sizes[i]); - if (ret == 0) { + unsigned len = compress(msg->uncompressed_chunks[i], + msg->uncompressed_chunk_sizes[i], + msg->compressed_chunks[i]); + if (len) { + /* To be written compressed */ msg->out_compressed_chunks[i] = msg->compressed_chunks[i]; + msg->compressed_chunk_sizes[i] = len; } else { + /* To be written uncompressed */ msg->out_compressed_chunks[i] = msg->uncompressed_chunks[i]; msg->compressed_chunk_sizes[i] = msg->uncompressed_chunk_sizes[i]; + } } } @@ -725,8 +575,58 @@ compressor_thread_proc(void *arg) } #endif /* ENABLE_MULTITHREADED_COMPRESSION */ +static void +do_write_streams_progress(union wimlib_progress_info *progress, + wimlib_progress_func_t progress_func, + uint64_t size_added) +{ + progress->write_streams.completed_bytes += size_added; + progress->write_streams.completed_streams++; + if (progress_func && + progress->write_streams.completed_bytes >= progress->write_streams._private) + { + progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, + progress); + if (progress->write_streams._private == progress->write_streams.total_bytes) { + progress->write_streams._private = ~0; + } else { + progress->write_streams._private = + min(progress->write_streams.total_bytes, + progress->write_streams.completed_bytes + + progress->write_streams.total_bytes / 100); + } + } +} + +static int +sha1_chunk(const void *buf, size_t len, void *ctx) +{ + sha1_update(ctx, buf, len); + return 0; +} + +static int +sha1_resource(struct wim_lookup_table_entry *lte) +{ + int ret; + SHA_CTX sha_ctx; + + sha1_init(&sha_ctx); + ret = read_resource_prefix(lte, wim_resource_size(lte), + sha1_chunk, &sha_ctx, 0); + if (ret == 0) + sha1_final(lte->hash, &sha_ctx); + return ret; +} + +enum { + STREAMS_MERGED = 0, + STREAMS_NOT_MERGED = 1, +}; + static int do_write_stream_list(struct list_head *my_resources, + struct wim_lookup_table *lookup_table, FILE *out_fp, int out_ctype, wimlib_progress_func_t progress_func, @@ -734,30 +634,70 @@ do_write_stream_list(struct list_head *my_resources, int write_resource_flags) { int ret; - struct wim_lookup_table_entry *lte, *tmp; + struct wim_lookup_table_entry *lte; + + while (!list_empty(my_resources)) { + lte = container_of(my_resources->next, + struct wim_lookup_table_entry, + write_streams_list); + list_del(<e->write_streams_list); + if (lte->unhashed && !lte->unique_size) { + struct wim_lookup_table_entry *duplicate_lte; + struct wim_lookup_table_entry **my_ptr; + + my_ptr = lte->my_ptr; + ret = sha1_resource(lte); + if (ret) + return ret; + duplicate_lte = __lookup_resource(lookup_table, lte->hash); + if (duplicate_lte) { + bool new_stream = (duplicate_lte->out_refcnt == 0); + duplicate_lte->refcnt += lte->refcnt; + duplicate_lte->out_refcnt += lte->refcnt; + *my_ptr = duplicate_lte; + free_lookup_table_entry(lte); + lte = duplicate_lte; + if (new_stream) { + DEBUG("Stream of length %"PRIu64" is duplicate " + "with one already in WIM", + wim_resource_size(lte)); + } else { + DEBUG("Discarding duplicate stream of length %"PRIu64, + wim_resource_size(lte)); + goto skip_to_progress; + } + + } else { + lookup_table_insert(lookup_table, lte); + lte->out_refcnt = lte->refcnt; + lte->unhashed = 0; + } + } + + wimlib_assert(lte->out_refcnt != 0); - list_for_each_entry_safe(lte, tmp, my_resources, staging_list) { ret = write_wim_resource(lte, out_fp, out_ctype, <e->output_resource_entry, write_resource_flags); - if (ret != 0) + if (ret) return ret; - list_del(<e->staging_list); - progress->write_streams.completed_bytes += - wim_resource_size(lte); - progress->write_streams.completed_streams++; - if (progress_func) { - progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, - progress); + if (lte->unhashed) { + lookup_table_insert(lookup_table, lte); + lte->unhashed = 0; } + skip_to_progress: + do_write_streams_progress(progress, + progress_func, + wim_resource_size(lte)); } return 0; } static int write_stream_list_serial(struct list_head *stream_list, + struct wim_lookup_table *lookup_table, FILE *out_fp, int out_ctype, int write_flags, @@ -773,7 +713,9 @@ write_stream_list_serial(struct list_head *stream_list, progress->write_streams.num_threads = 1; if (progress_func) progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress); - return do_write_stream_list(stream_list, out_fp, + return do_write_stream_list(stream_list, + lookup_table, + out_fp, out_ctype, progress_func, progress, write_resource_flags); } @@ -935,7 +877,7 @@ main_writer_thread_proc(struct list_head *stream_list, if (next_lte->resource_location == RESOURCE_IN_FILE_ON_DISK) { - ERROR("We were reading it from `%s'; " + ERROR("We were reading it from `%"TS"'; " "maybe it changed while we were " "reading it.", next_lte->file_on_disk); @@ -1146,14 +1088,8 @@ main_writer_thread_proc(struct list_head *stream_list, WIM_RESHDR_FLAG_COMPRESSED; } - progress->write_streams.completed_bytes += - wim_resource_size(cur_lte); - progress->write_streams.completed_streams++; - - if (progress_func) { - progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, - progress); - } + do_write_streams_progress(progress, progress_func, + wim_resource_size(cur_lte)); FREE(cur_chunk_tab); cur_chunk_tab = NULL; @@ -1247,6 +1183,7 @@ get_default_num_threads() static int write_stream_list_parallel(struct list_head *stream_list, + struct wim_lookup_table *lookup_table, FILE *out_fp, int out_ctype, int write_flags, @@ -1341,6 +1278,7 @@ out_destroy_res_to_compress_queue: out_serial: WARNING("Falling back to single-threaded compression"); return write_stream_list_serial(stream_list, + lookup_table, out_fp, out_ctype, write_flags, @@ -1355,18 +1293,22 @@ out_serial: * @out_ctype and up to @num_threads compressor threads. */ static int -write_stream_list(struct list_head *stream_list, FILE *out_fp, - int out_ctype, int write_flags, - unsigned num_threads, - wimlib_progress_func_t progress_func) +write_stream_list(struct list_head *stream_list, + struct wim_lookup_table *lookup_table, + FILE *out_fp, int out_ctype, int write_flags, + unsigned num_threads, wimlib_progress_func_t progress_func) { struct wim_lookup_table_entry *lte; size_t num_streams = 0; u64 total_bytes = 0; u64 total_compression_bytes = 0; union wimlib_progress_info progress; + int ret; - list_for_each_entry(lte, stream_list, staging_list) { + if (list_empty(stream_list)) + return 0; + + list_for_each_entry(lte, stream_list, write_streams_list) { num_streams++; total_bytes += wim_resource_size(lte); if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE @@ -1382,30 +1324,82 @@ write_stream_list(struct list_head *stream_list, FILE *out_fp, progress.write_streams.completed_streams = 0; progress.write_streams.num_threads = num_threads; progress.write_streams.compression_type = out_ctype; + progress.write_streams._private = 0; #ifdef ENABLE_MULTITHREADED_COMPRESSION if (total_compression_bytes >= 1000000 && num_threads != 1) - return write_stream_list_parallel(stream_list, - out_fp, - out_ctype, - write_flags, - num_threads, - progress_func, - &progress); + ret = write_stream_list_parallel(stream_list, + lookup_table, + out_fp, + out_ctype, + write_flags, + num_threads, + progress_func, + &progress); else #endif - return write_stream_list_serial(stream_list, - out_fp, - out_ctype, - write_flags, - progress_func, - &progress); + ret = write_stream_list_serial(stream_list, + lookup_table, + out_fp, + out_ctype, + write_flags, + progress_func, + &progress); + return ret; +} + +struct stream_size_table { + struct hlist_head *array; + size_t num_entries; + size_t capacity; +}; + +static int +init_stream_size_table(struct stream_size_table *tab, size_t capacity) +{ + tab->array = CALLOC(capacity, sizeof(tab->array[0])); + if (!tab->array) + return WIMLIB_ERR_NOMEM; + tab->num_entries = 0; + tab->capacity = capacity; + return 0; +} + +static void +destroy_stream_size_table(struct stream_size_table *tab) +{ + FREE(tab->array); +} + +static int +stream_size_table_insert(struct wim_lookup_table_entry *lte, void *_tab) +{ + struct stream_size_table *tab = _tab; + size_t pos; + struct wim_lookup_table_entry *hashed_lte; + struct hlist_node *tmp; + + pos = hash_u64(wim_resource_size(lte)) % tab->capacity; + lte->unique_size = 1; + hlist_for_each_entry(hashed_lte, tmp, &tab->array[pos], hash_list_2) { + if (wim_resource_size(hashed_lte) == wim_resource_size(lte)) { + lte->unique_size = 0; + hashed_lte->unique_size = 0; + break; + } + } + + hlist_add_head(<e->hash_list_2, &tab->array[pos]); + tab->num_entries++; + return 0; } + struct lte_overwrite_prepare_args { WIMStruct *wim; off_t end_offset; struct list_head *stream_list; + struct stream_size_table *stream_size_tab; }; static int @@ -1414,52 +1408,118 @@ lte_overwrite_prepare(struct wim_lookup_table_entry *lte, void *arg) struct lte_overwrite_prepare_args *args = arg; if (lte->resource_location == RESOURCE_IN_WIM && - lte->wim == args->wim && - lte->resource_entry.offset + lte->resource_entry.size > args->end_offset) + lte->wim == args->wim) { - #ifdef ENABLE_ERROR_MESSAGES - ERROR("The following resource is after the XML data:"); - print_lookup_table_entry(lte, stderr); - #endif - return WIMLIB_ERR_RESOURCE_ORDER; + /* We can't do an in place overwrite on the WIM if there are + * streams after the XML data. */ + if (lte->resource_entry.offset + + lte->resource_entry.size > args->end_offset) + { + #ifdef ENABLE_ERROR_MESSAGES + ERROR("The following resource is after the XML data:"); + print_lookup_table_entry(lte, stderr); + #endif + return WIMLIB_ERR_RESOURCE_ORDER; + } + } else { + if (!(lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA)) + list_add_tail(<e->write_streams_list, args->stream_list); } - lte->out_refcnt = lte->refcnt; - memcpy(<e->output_resource_entry, <e->resource_entry, - sizeof(struct resource_entry)); - if (!(lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA)) { - wimlib_assert(lte->resource_location != RESOURCE_NONEXISTENT); - if (lte->resource_location != RESOURCE_IN_WIM || lte->wim != args->wim) - list_add(<e->staging_list, args->stream_list); + stream_size_table_insert(lte, args->stream_size_tab); + return 0; +} + +static int +lte_set_output_res_entry(struct wim_lookup_table_entry *lte, void *_wim) +{ + if (lte->resource_location == RESOURCE_IN_WIM && + lte->wim == _wim) + { + memcpy(<e->output_resource_entry, <e->resource_entry, + sizeof(struct resource_entry)); } return 0; } +/* Given a WIM that we are going to overwrite in place with zero or more + * additional streams added, construct a list the list of new unique streams + * ('struct wim_lookup_table_entry's) that must be written, plus any unhashed + * streams that need to be added but may be identical to other hashed or + * unhashed streams. These unhashed streams are checksummed while the streams + * are being written. To aid this process, the member @unique_size is set to 1 + * on streams that have a unique size and therefore must be written. + * + * The out_refcnt member of each 'struct wim_lookup_table_entry' is set to + * indicate the number of times the stream is referenced in only the streams + * that are being written; this may still be adjusted later when unhashed + * streams are being resolved. + */ static int -wim_find_new_streams(WIMStruct *wim, off_t end_offset, - struct list_head *stream_list) +prepare_streams_for_overwrite(WIMStruct *wim, off_t end_offset, + struct list_head *stream_list) { + int ret; + struct stream_size_table stream_size_tab; struct lte_overwrite_prepare_args args = { .wim = wim, .end_offset = end_offset, .stream_list = stream_list, + .stream_size_tab = &stream_size_tab, }; - return for_lookup_table_entry(wim->lookup_table, - lte_overwrite_prepare, &args); + ret = init_stream_size_table(&stream_size_tab, 9001); + if (ret) + return ret; + + INIT_LIST_HEAD(stream_list); + for (int i = 0; i < wim->hdr.image_count; i++) { + struct wim_image_metadata *imd; + struct wim_lookup_table_entry *lte; + + imd = wim->image_metadata[i]; + image_for_each_unhashed_stream(lte, imd) { + ret = lte_overwrite_prepare(lte, &args); + if (ret) + goto out_destroy_stream_size_table; + } + } + ret = for_lookup_table_entry(wim->lookup_table, + lte_overwrite_prepare, &args); + if (ret) + goto out_destroy_stream_size_table; + + for (int i = 0; i < wim->hdr.image_count; i++) + lte_set_output_res_entry(wim->image_metadata[i]->metadata_lte, + wim); + ret = for_lookup_table_entry(wim->lookup_table, + lte_set_output_res_entry, wim); +out_destroy_stream_size_table: + destroy_stream_size_table(&stream_size_tab); + return ret; } + +struct find_streams_ctx { + struct list_head stream_list; + struct stream_size_table stream_size_tab; +}; + static int inode_find_streams_to_write(struct wim_inode *inode, struct wim_lookup_table *table, - struct list_head *stream_list) + struct list_head *stream_list, + struct stream_size_table *tab) { struct wim_lookup_table_entry *lte; for (unsigned i = 0; i <= inode->i_num_ads; i++) { lte = inode_stream_lte(inode, i, table); if (lte) { - if (lte->out_refcnt == 0) - list_add_tail(<e->staging_list, stream_list); + if (lte->out_refcnt == 0) { + if (lte->unhashed) + stream_size_table_insert(lte, tab); + list_add_tail(<e->write_streams_list, stream_list); + } lte->out_refcnt += inode->i_nlink; } } @@ -1469,31 +1529,85 @@ inode_find_streams_to_write(struct wim_inode *inode, static int image_find_streams_to_write(WIMStruct *w) { + struct wim_image_metadata *imd; + struct find_streams_ctx *ctx; struct wim_inode *inode; - struct hlist_node *cur; - struct hlist_head *inode_list; + struct wim_lookup_table_entry *lte; + + ctx = w->private; + imd = wim_get_current_image_metadata(w); - inode_list = &wim_get_current_image_metadata(w)->inode_list; - hlist_for_each_entry(inode, cur, inode_list, i_hlist) { + image_for_each_unhashed_stream(lte, imd) { + lte->out_refcnt = 0; + wimlib_assert(lte->unhashed); + wimlib_assert(lte->my_ptr != NULL); + } + + /* Go through this image's inodes to find any streams that have not been + * found yet. */ + image_for_each_inode(inode, imd) { inode_find_streams_to_write(inode, w->lookup_table, - (struct list_head*)w->private); + &ctx->stream_list, + &ctx->stream_size_tab); } return 0; } +/* Given a WIM that from which one or all of the images is being written, build + * the list of unique streams ('struct wim_lookup_table_entry's) that must be + * written, plus any unhashed streams that need to be written but may be + * identical to other hashed or unhashed streams being written. These unhashed + * streams are checksummed while the streams are being written. To aid this + * process, the member @unique_size is set to 1 on streams that have a unique + * size and therefore must be written. + * + * The out_refcnt member of each 'struct wim_lookup_table_entry' is set to + * indicate the number of times the stream is referenced in only the streams + * that are being written; this may still be adjusted later when unhashed + * streams are being resolved. + */ static int -write_wim_streams(WIMStruct *w, int image, int write_flags, - unsigned num_threads, - wimlib_progress_func_t progress_func) +prepare_stream_list(WIMStruct *wim, int image, struct list_head *stream_list) { + int ret; + struct find_streams_ctx ctx; - for_lookup_table_entry(w->lookup_table, lte_zero_out_refcnt, NULL); - LIST_HEAD(stream_list); - w->private = &stream_list; - for_image(w, image, image_find_streams_to_write); - return write_stream_list(&stream_list, w->out_fp, - wimlib_get_compression_type(w), write_flags, - num_threads, progress_func); + for_lookup_table_entry(wim->lookup_table, lte_zero_out_refcnt, NULL); + ret = init_stream_size_table(&ctx.stream_size_tab, 9001); + if (ret) + return ret; + for_lookup_table_entry(wim->lookup_table, stream_size_table_insert, + &ctx.stream_size_tab); + INIT_LIST_HEAD(&ctx.stream_list); + wim->private = &ctx; + for_image(wim, image, image_find_streams_to_write); + destroy_stream_size_table(&ctx.stream_size_tab); + + INIT_LIST_HEAD(stream_list); + list_splice(&ctx.stream_list, stream_list); + return 0; +} + +/* Writes the streams for the specified @image in @wim to @wim->out_fp. + */ +static int +write_wim_streams(WIMStruct *wim, int image, int write_flags, + unsigned num_threads, + wimlib_progress_func_t progress_func) +{ + int ret; + struct list_head stream_list; + + ret = prepare_stream_list(wim, image, &stream_list); + if (ret) + return ret; + return write_stream_list(&stream_list, + wim->lookup_table, + wim->out_fp, + wimlib_get_compression_type(wim), + write_flags, + num_threads, + progress_func); } /* @@ -1540,7 +1654,7 @@ finish_write(WIMStruct *w, int image, int write_flags, memcpy(&hdr, &w->hdr, sizeof(struct wim_header)); if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { - ret = write_lookup_table(w->lookup_table, out, &hdr.lookup_table_res_entry); + ret = write_lookup_table(w, image, &hdr.lookup_table_res_entry); if (ret != 0) goto out; } @@ -1626,7 +1740,7 @@ finish_write(WIMStruct *w, int image, int write_flags, } else { memcpy(&hdr.boot_metadata_res_entry, &w->image_metadata[ - hdr.boot_idx - 1].metadata_lte->output_resource_entry, + hdr.boot_idx - 1]->metadata_lte->output_resource_entry, sizeof(struct resource_entry)); } @@ -1638,7 +1752,7 @@ finish_write(WIMStruct *w, int image, int write_flags, } ret = write_header(&hdr, out); - if (ret != 0) + if (ret) goto out; if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) { @@ -1668,12 +1782,12 @@ lock_wim(WIMStruct *w, FILE *fp) ret = flock(fileno(fp), LOCK_EX | LOCK_NB); if (ret != 0) { if (errno == EWOULDBLOCK) { - ERROR("`%s' is already being modified or has been " + ERROR("`%"TS"' is already being modified or has been " "mounted read-write\n" " by another process!", w->filename); ret = WIMLIB_ERR_ALREADY_LOCKED; } else { - WARNING_WITH_ERRNO("Failed to lock `%s'", + WARNING_WITH_ERRNO("Failed to lock `%"TS"'", w->filename); ret = 0; } @@ -1686,24 +1800,24 @@ lock_wim(WIMStruct *w, FILE *fp) #endif static int -open_wim_writable(WIMStruct *w, const mbchar *path, - bool trunc, bool readable) +open_wim_writable(WIMStruct *w, const tchar *path, + bool trunc, bool also_readable) { - const char *mode; + const tchar *mode; if (trunc) - if (readable) - mode = "w+b"; + if (also_readable) + mode = T("w+b"); else - mode = "wb"; + mode = T("wb"); else - mode = "r+b"; + mode = T("r+b"); wimlib_assert(w->out_fp == NULL); - w->out_fp = fopen(path, mode); + w->out_fp = tfopen(path, mode); if (w->out_fp) { return 0; } else { - ERROR_WITH_ERRNO("Failed to open `%s' for writing", path); + ERROR_WITH_ERRNO("Failed to open `%"TS"' for writing", path); return WIMLIB_ERR_OPEN; } } @@ -1722,7 +1836,7 @@ close_wim_writable(WIMStruct *w) /* Open file stream and write dummy header for WIM. */ int -begin_write(WIMStruct *w, const mbchar *path, int write_flags) +begin_write(WIMStruct *w, const tchar *path, int write_flags) { int ret; ret = open_wim_writable(w, path, true, @@ -1735,7 +1849,7 @@ begin_write(WIMStruct *w, const mbchar *path, int write_flags) /* Writes a stand-alone WIM to a file. */ WIMLIBAPI int -wimlib_write(WIMStruct *w, const mbchar *path, +wimlib_write(WIMStruct *w, const tchar *path, int image, int write_flags, unsigned num_threads, wimlib_progress_func_t progress_func) { @@ -1756,19 +1870,19 @@ wimlib_write(WIMStruct *w, const mbchar *path, } ret = begin_write(w, path, write_flags); - if (ret != 0) + if (ret) goto out; ret = write_wim_streams(w, image, write_flags, num_threads, progress_func); - if (ret != 0) + if (ret) goto out; if (progress_func) progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN, NULL); ret = for_image(w, image, write_metadata_resource); - if (ret != 0) + if (ret) goto out; if (progress_func) @@ -1777,7 +1891,7 @@ wimlib_write(WIMStruct *w, const mbchar *path, ret = finish_write(w, image, write_flags, progress_func); out: close_wim_writable(w); - DEBUG("wimlib_write(path=%s) = %d", path, ret); + DEBUG("wimlib_write(path=%"TS") = %d", path, ret); return ret; } @@ -1785,7 +1899,7 @@ static bool any_images_modified(WIMStruct *w) { for (int i = 0; i < w->hdr.image_count; i++) - if (w->image_metadata[i].modified) + if (w->image_metadata[i]->modified) return true; return false; } @@ -1855,9 +1969,8 @@ overwrite_wim_inplace(WIMStruct *w, int write_flags, int ret; struct list_head stream_list; off_t old_wim_end; - bool found_modified_image; - DEBUG("Overwriting `%s' in-place", w->filename); + DEBUG("Overwriting `%"TS"' in-place", w->filename); /* Make sure that the integrity table (if present) is after the XML * data, and that there are no stream resources, metadata resources, or @@ -1888,18 +2001,17 @@ overwrite_wim_inplace(WIMStruct *w, int write_flags, write_flags |= WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE | WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML; } - INIT_LIST_HEAD(&stream_list); - ret = wim_find_new_streams(w, old_wim_end, &stream_list); - if (ret != 0) + ret = prepare_streams_for_overwrite(w, old_wim_end, &stream_list); + if (ret) return ret; ret = open_wim_writable(w, w->filename, false, (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0); - if (ret != 0) + if (ret) return ret; ret = lock_wim(w, w->out_fp); - if (ret != 0) { + if (ret) { fclose(w->out_fp); w->out_fp = NULL; return ret; @@ -1913,27 +2025,22 @@ overwrite_wim_inplace(WIMStruct *w, int write_flags, return WIMLIB_ERR_WRITE; } - if (!list_empty(&stream_list)) { - DEBUG("Writing newly added streams (offset = %"PRIu64")", - old_wim_end); - ret = write_stream_list(&stream_list, w->out_fp, - wimlib_get_compression_type(w), - write_flags, num_threads, - progress_func); - if (ret != 0) - goto out_ftruncate; - } else { - DEBUG("No new streams were added"); - } + DEBUG("Writing newly added streams (offset = %"PRIu64")", + old_wim_end); + ret = write_stream_list(&stream_list, + w->lookup_table, + w->out_fp, + wimlib_get_compression_type(w), + write_flags, num_threads, + progress_func); + if (ret) + goto out_ftruncate; - found_modified_image = false; for (int i = 0; i < w->hdr.image_count; i++) { - if (!found_modified_image) - found_modified_image = w->image_metadata[i].modified; - if (found_modified_image) { + if (w->image_metadata[i]->modified) { select_wim_image(w, i + 1); ret = write_metadata_resource(w); - if (ret != 0) + if (ret) goto out_ftruncate; } } @@ -1943,11 +2050,11 @@ overwrite_wim_inplace(WIMStruct *w, int write_flags, out_ftruncate: close_wim_writable(w); if (ret != 0 && !(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { - WARNING("Truncating `%s' to its original size (%"PRIu64" bytes)", + WARNING("Truncating `%"TS"' to its original size (%"PRIu64" bytes)", w->filename, old_wim_end); /* Return value of truncate() is ignored because this is already * an error path. */ - (void)truncate(w->filename, old_wim_end); + (void)ttruncate(w->filename, old_wim_end); } w->wim_locked = 0; return ret; @@ -1961,29 +2068,40 @@ overwrite_wim_via_tmpfile(WIMStruct *w, int write_flags, size_t wim_name_len; int ret; - DEBUG("Overwriting `%s' via a temporary file", w->filename); + DEBUG("Overwriting `%"TS"' via a temporary file", w->filename); /* Write the WIM to a temporary file in the same directory as the * original WIM. */ - wim_name_len = strlen(w->filename); - mbchar tmpfile[wim_name_len + 10]; - memcpy(tmpfile, w->filename, wim_name_len); + wim_name_len = tstrlen(w->filename); + tchar tmpfile[wim_name_len + 10]; + tmemcpy(tmpfile, w->filename, wim_name_len); randomize_char_array_with_alnum(tmpfile + wim_name_len, 9); - tmpfile[wim_name_len + 9] = '\0'; + tmpfile[wim_name_len + 9] = T('\0'); ret = wimlib_write(w, tmpfile, WIMLIB_ALL_IMAGES, write_flags | WIMLIB_WRITE_FLAG_FSYNC, num_threads, progress_func); if (ret != 0) { - ERROR("Failed to write the WIM file `%s'", tmpfile); + ERROR("Failed to write the WIM file `%"TS"'", tmpfile); goto err; } - DEBUG("Renaming `%s' to `%s'", tmpfile, w->filename); + DEBUG("Renaming `%"TS"' to `%"TS"'", tmpfile, w->filename); + +#ifdef __WIN32__ + /* Windows won't let you delete open files unless FILE_SHARE_DELETE was + * specified to CreateFile(). The WIM was opened with fopen(), which + * didn't provided this flag to CreateFile, so the handle must be closed + * before executing the rename(). */ + if (w->fp != NULL) { + fclose(w->fp); + w->fp = NULL; + } +#endif /* Rename the new file to the old file .*/ - if (rename(tmpfile, w->filename) != 0) { - ERROR_WITH_ERRNO("Failed to rename `%s' to `%s'", + if (trename(tmpfile, w->filename) != 0) { + ERROR_WITH_ERRNO("Failed to rename `%"TS"' to `%"TS"'", tmpfile, w->filename); ret = WIMLIB_ERR_RENAME; goto err; @@ -2003,10 +2121,10 @@ overwrite_wim_via_tmpfile(WIMStruct *w, int write_flags, } /* Re-open the WIM read-only. */ - w->fp = fopen(w->filename, "rb"); + w->fp = tfopen(w->filename, T("rb")); if (w->fp == NULL) { ret = WIMLIB_ERR_REOPEN; - WARNING_WITH_ERRNO("Failed to re-open `%s' read-only", + WARNING_WITH_ERRNO("Failed to re-open `%"TS"' read-only", w->filename); FREE(w->filename); w->filename = NULL; @@ -2014,8 +2132,8 @@ overwrite_wim_via_tmpfile(WIMStruct *w, int write_flags, return ret; err: /* Remove temporary file. */ - if (unlink(tmpfile) != 0) - WARNING_WITH_ERRNO("Failed to remove `%s'", tmpfile); + if (tunlink(tmpfile) != 0) + WARNING_WITH_ERRNO("Failed to remove `%"TS"'", tmpfile); return ret; }