X-Git-Url: https://wimlib.net/git/?a=blobdiff_plain;f=src%2Fwrite.c;h=a9cb6aecaa6c6da0db11b159014e192ad29fac67;hb=a6c413a13da72aac44b0a522c2650d5e54e16a72;hp=6663564585c299e9dcec970d42d3dc4fd5c6db8d;hpb=6cca349b45e66a2b2b82aa5dcd269a4bf61c50db;p=wimlib diff --git a/src/write.c b/src/write.c index 66635645..a9cb6aec 100644 --- a/src/write.c +++ b/src/write.c @@ -33,9 +33,12 @@ #include "lzx.h" #include "xpress.h" #include + +#ifdef ENABLE_MULTITHREADED_COMPRESSION #include #include #include +#endif #ifdef WITH_NTFS_3G #include @@ -47,270 +50,36 @@ #ifdef HAVE_ALLOCA_H #include +#else +#include #endif - -/* Reopens the FILE* for a WIM read-write. */ -static int reopen_rw(WIMStruct *w) -{ - FILE *fp; - - if (fclose(w->fp) != 0) - ERROR_WITH_ERRNO("Failed to close the file `%s'", w->filename); - w->fp = NULL; - fp = fopen(w->filename, "r+b"); - if (!fp) { - ERROR_WITH_ERRNO("Failed to open `%s' for reading and writing", - w->filename); - return WIMLIB_ERR_OPEN; - } - w->fp = fp; - return 0; -} - - - -/* - * Writes a WIM file to the original file that it was read from, overwriting it. - */ -WIMLIBAPI int wimlib_overwrite(WIMStruct *w, int write_flags) +static int do_fflush(FILE *fp) { - const char *wimfile_name; - size_t wim_name_len; - int ret; - - if (!w) - return WIMLIB_ERR_INVALID_PARAM; - - write_flags &= ~WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE; - - wimfile_name = w->filename; - - DEBUG("Replacing WIM file `%s'.", wimfile_name); - - if (!wimfile_name) - return WIMLIB_ERR_NO_FILENAME; - - /* Write the WIM to a temporary file. */ - /* XXX should the temporary file be somewhere else? */ - wim_name_len = strlen(wimfile_name); - char tmpfile[wim_name_len + 10]; - memcpy(tmpfile, wimfile_name, wim_name_len); - randomize_char_array_with_alnum(tmpfile + wim_name_len, 9); - tmpfile[wim_name_len + 9] = '\0'; - - ret = wimlib_write(w, tmpfile, WIM_ALL_IMAGES, write_flags); + int ret = fflush(fp); if (ret != 0) { - ERROR("Failed to write the WIM file `%s'", tmpfile); - if (unlink(tmpfile) != 0) - WARNING("Failed to remove `%s'", tmpfile); - return ret; - } - - DEBUG("Closing original WIM file."); - /* Close the original WIM file that was opened for reading. */ - if (w->fp) { - if (fclose(w->fp) != 0) { - WARNING("Failed to close the file `%s'", wimfile_name); - } - w->fp = NULL; - } - - DEBUG("Renaming `%s' to `%s'", tmpfile, wimfile_name); - - /* Rename the new file to the old file .*/ - if (rename(tmpfile, wimfile_name) != 0) { - ERROR_WITH_ERRNO("Failed to rename `%s' to `%s'", - tmpfile, wimfile_name); - /* Remove temporary file. */ - if (unlink(tmpfile) != 0) - ERROR_WITH_ERRNO("Failed to remove `%s'", tmpfile); - return WIMLIB_ERR_RENAME; - } - - if (write_flags & WIMLIB_WRITE_FLAG_VERBOSE) - printf("Successfully renamed `%s' to `%s'\n", tmpfile, wimfile_name); - - return 0; -} - -static int check_resource_offset(struct lookup_table_entry *lte, void *arg) -{ - u64 xml_data_offset = *(u64*)arg; - if (lte->resource_entry.offset > xml_data_offset) { - ERROR("The following resource is *after* the XML data:"); - print_lookup_table_entry(lte); - return WIMLIB_ERR_RESOURCE_ORDER; + ERROR_WITH_ERRNO("Failed to flush data to output WIM file"); + return WIMLIB_ERR_WRITE; } return 0; } -WIMLIBAPI int wimlib_overwrite_xml_and_header(WIMStruct *w, int write_flags) +static int fflush_and_ftruncate(FILE *fp, off_t size) { int ret; - FILE *fp; - u8 *integrity_table = NULL; - off_t xml_end; - off_t xml_size; - size_t bytes_written; - - DEBUG("Overwriting XML and header of `%s', write_flags = %#x", - w->filename, write_flags); - - if (!w->filename) - return WIMLIB_ERR_NO_FILENAME; - - write_flags &= ~WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE; - /* Make sure that the integrity table (if present) is after the XML - * data, and that there are no stream resources, metadata resources, or - * lookup tables after the XML data. Otherwise, these data would be - * destroyed by this function. */ - if (w->hdr.integrity.offset != 0 && - w->hdr.integrity.offset < w->hdr.xml_res_entry.offset) { - ERROR("Didn't expect the integrity table to be before the XML data"); - return WIMLIB_ERR_RESOURCE_ORDER; - } - - if (w->hdr.lookup_table_res_entry.offset > - w->hdr.xml_res_entry.offset) { - ERROR("Didn't expect the lookup table to be after the XML data"); - return WIMLIB_ERR_RESOURCE_ORDER; - } - - ret = for_lookup_table_entry(w->lookup_table, check_resource_offset, - &w->hdr.xml_res_entry.offset); - if (ret != 0) - return ret; - - ret = reopen_rw(w); - if (ret != 0) - return ret; - - fp = w->fp; - - /* The old integrity table is still OK, as the SHA1 message digests in - * the integrity table include neither the header nor the XML data. - * Save it for later if it exists and an integrity table was required. - * */ - if ((write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) - && w->hdr.integrity.offset != 0) - { - DEBUG("Reading existing integrity table."); - integrity_table = MALLOC(w->hdr.integrity.size); - if (!integrity_table) - return WIMLIB_ERR_NOMEM; - - ret = read_uncompressed_resource(fp, w->hdr.integrity.offset, - w->hdr.integrity.original_size, - integrity_table); - if (ret != 0) - goto err; - DEBUG("Done reading existing integrity table."); - } - - DEBUG("Overwriting XML data."); - /* Overwrite the XML data. */ - if (fseeko(fp, w->hdr.xml_res_entry.offset, SEEK_SET) != 0) { - ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" " - "for XML data", w->hdr.xml_res_entry.offset); - ret = WIMLIB_ERR_WRITE; - goto err; - } - ret = write_xml_data(w->wim_info, WIM_ALL_IMAGES, fp, 0); - if (ret != 0) - goto err; - - DEBUG("Updating XML resource entry."); - /* Update the XML resource entry in the WIM header. */ - xml_end = ftello(fp); - if (xml_end == -1) { - ret = WIMLIB_ERR_WRITE; - goto err; - } - xml_size = xml_end - w->hdr.xml_res_entry.offset; - w->hdr.xml_res_entry.size = xml_size; - w->hdr.xml_res_entry.original_size = xml_size; - /* XML data offset is unchanged. */ - - if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) { - DEBUG("Writing integrity table."); - w->hdr.integrity.offset = xml_end; - if (integrity_table) { - /* The existing integrity table was saved. */ - bytes_written = fwrite(integrity_table, 1, - w->hdr.integrity.size, fp); - if (bytes_written != w->hdr.integrity.size) { - ERROR_WITH_ERRNO("Failed to write integrity " - "table"); - ret = WIMLIB_ERR_WRITE; - goto err; - } - FREE(integrity_table); - } else { - /* There was no existing integrity table, so a new one - * must be calculated. */ - ret = write_integrity_table(fp, WIM_HEADER_DISK_SIZE, - w->hdr.lookup_table_res_entry.offset + - w->hdr.lookup_table_res_entry.size, - write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS); - if (ret != 0) - return ret; - - off_t end_integrity = ftello(fp); - if (end_integrity == -1) - return WIMLIB_ERR_WRITE; - - off_t integrity_size = end_integrity - xml_end; - w->hdr.integrity.size = integrity_size; - w->hdr.integrity.original_size = integrity_size; - w->hdr.integrity.flags = 0; - } - } else { - DEBUG("Truncating file to end of XML data."); - /* No integrity table to write. The file should be truncated - * because it's possible that the old file was longer (due to it - * including an integrity table, or due to its XML data being - * longer) */ - if (fflush(fp) != 0) { - ERROR_WITH_ERRNO("Failed to flush stream for file `%s'", - w->filename); - return WIMLIB_ERR_WRITE; - } - if (ftruncate(fileno(fp), xml_end) != 0) { - ERROR_WITH_ERRNO("Failed to truncate `%s' to %"PRIu64" " - "bytes", w->filename, xml_end); - return WIMLIB_ERR_WRITE; - } - memset(&w->hdr.integrity, 0, sizeof(struct resource_entry)); - } - - DEBUG("Overwriting header."); - /* Overwrite the header. */ - if (fseeko(fp, 0, SEEK_SET) != 0) { - ERROR_WITH_ERRNO("Failed to seek to beginning of `%s'", - w->filename); - return WIMLIB_ERR_WRITE; - } - - ret = write_header(&w->hdr, fp); + ret = do_fflush(fp); if (ret != 0) return ret; - - DEBUG("Closing `%s'.", w->filename); - if (fclose(fp) != 0) { - ERROR_WITH_ERRNO("Failed to close `%s'", w->filename); + ret = ftruncate(fileno(fp), size); + if (ret != 0) { + ERROR_WITH_ERRNO("Failed to truncate output WIM file to " + "%"PRIu64" bytes", size); return WIMLIB_ERR_WRITE; } - w->fp = NULL; - DEBUG("Done."); return 0; -err: - FREE(integrity_table); - return ret; } - /* 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.) */ @@ -370,19 +139,8 @@ out: return ret; } -typedef int (*compress_func_t)(const void *, unsigned, void *, unsigned *); - -compress_func_t get_compress_func(int out_ctype) -{ - if (out_ctype == WIM_COMPRESSION_TYPE_LZX) - return lzx_compress; - else - return xpress_compress; -} - - /* - * Compresses a chunk of a WIM resource. + * Pointer to function to compresses a chunk of a WIM resource. * * @chunk: Uncompressed data of the chunk. * @chunk_size: Size of the uncompressed chunk in bytes. @@ -391,21 +149,19 @@ compress_func_t get_compress_func(int out_ctype) * @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) +typedef int (*compress_func_t)(const void *, unsigned, void *, unsigned *); + +compress_func_t get_compress_func(int out_ctype) { - compress_func_t compress = get_compress_func(ctype); - return (*compress)(chunk, chunk_size, compressed_chunk, - compressed_chunk_len_ret); + if (out_ctype == WIM_COMPRESSION_TYPE_LZX) + return lzx_compress; + else + return xpress_compress; } /* @@ -422,23 +178,17 @@ static int compress_chunk(const u8 chunk[], unsigned chunk_size, * 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, + FILE *out_fp, compress_func_t compress, 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 { + if (chunk_tab) { u8 *compressed_chunk = alloca(chunk_size); int ret; - ret = compress_chunk(chunk, chunk_size, compressed_chunk, - &out_chunk_size, out_ctype); + ret = compress(chunk, chunk_size, compressed_chunk, + &out_chunk_size); if (ret == 0) { out_chunk = compressed_chunk; } else { @@ -447,8 +197,10 @@ static int write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size, } *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset; chunk_tab->cur_offset += out_chunk_size; + } else { + out_chunk = chunk; + out_chunk_size = 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; @@ -497,8 +249,10 @@ 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 lookup_table_entry *lte - + #ifdef WITH_NTFS_3G , ntfs_inode **ni_ret #endif @@ -544,6 +298,8 @@ static int prepare_resource_for_read(struct lookup_table_entry *lte return 0; } +/* Undo prepare_resource_for_read() by closing the cached FILE * or NTFS + * attribute. */ static void end_wim_resource_read(struct lookup_table_entry *lte #ifdef WITH_NTFS_3G , ntfs_inode *ni @@ -602,6 +358,7 @@ int write_wim_resource(struct lookup_table_entry *lte, struct chunk_table *chunk_tab = NULL; bool raw; off_t file_offset; + compress_func_t compress; #ifdef WITH_NTFS_3G ntfs_inode *ni = NULL; #endif @@ -625,7 +382,8 @@ int write_wim_resource(struct lookup_table_entry *lte, /* 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); + && out_ctype != WIM_COMPRESSION_TYPE_NONE + && !(flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS)); if (raw) { flags |= WIMLIB_RESOURCE_FLAG_RAW; @@ -667,13 +425,15 @@ int write_wim_resource(struct lookup_table_entry *lte, * hash given in the lookup table entry once we've finished reading the * resource. */ SHA_CTX ctx; - if (!raw) + 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. */ - offset = 0; do { u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE); ret = read_wim_resource(lte, buf, to_read, offset, flags); @@ -682,7 +442,7 @@ int write_wim_resource(struct lookup_table_entry *lte, if (!raw) sha1_update(&ctx, buf, to_read); ret = write_wim_resource_chunk(buf, to_read, out_fp, - out_ctype, chunk_tab); + compress, chunk_tab); if (ret != 0) goto out_fclose; bytes_remaining -= to_read; @@ -748,16 +508,10 @@ int write_wim_resource(struct lookup_table_entry *lte, out_res_entry, flags); 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; + + ret = fflush_and_ftruncate(out_fp, file_offset + out_res_entry->size); + if (ret != 0) goto out_fclose; - } } else { if (out_res_entry) { out_res_entry->size = new_compressed_size; @@ -781,7 +535,7 @@ out: return ret; } - +#ifdef ENABLE_MULTITHREADED_COMPRESSION struct shared_queue { sem_t filled_slots; sem_t empty_slots; @@ -841,13 +595,6 @@ static void *shared_queue_get(struct shared_queue *q) return obj; } -static inline int shared_queue_get_filled(struct shared_queue *q) -{ - int sval; - sem_getvalue(&q->filled_slots, &sval); - return sval; -} - struct compressor_thread_params { struct shared_queue *res_to_compress_queue; struct shared_queue *compressed_res_queue; @@ -901,23 +648,67 @@ static void *compressor_thread_proc(void *arg) } DEBUG("Compressor thread terminating"); } +#endif + +static void show_stream_write_progress(u64 *cur_size, u64 *next_size, + u64 total_size, u64 one_percent, + unsigned *cur_percent, + const struct lookup_table_entry *cur_lte) +{ + if (*cur_size >= *next_size) { + printf("\r%"PRIu64" MiB of %"PRIu64" MiB " + "(uncompressed) written (%u%% done)", + *cur_size >> 20, + total_size >> 20, *cur_percent); + fflush(stdout); + *next_size += one_percent; + (*cur_percent)++; + } + *cur_size += wim_resource_size(cur_lte); +} + +static void finish_stream_write_progress(u64 total_size) +{ + printf("\r%"PRIu64" MiB of %"PRIu64" MiB " + "(uncompressed) written (100%% done)\n", + total_size >> 20, total_size >> 20); + fflush(stdout); +} static int write_stream_list_serial(struct list_head *stream_list, FILE *out_fp, int out_ctype, - int write_flags) + int write_flags, u64 total_size) { struct lookup_table_entry *lte; int ret; + u64 one_percent = total_size / 100; + u64 cur_size = 0; + u64 next_size = 0; + unsigned cur_percent = 0; + int write_resource_flags = 0; + + if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) + write_resource_flags |= WIMLIB_RESOURCE_FLAG_RECOMPRESS; + list_for_each_entry(lte, stream_list, staging_list) { + if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) { + show_stream_write_progress(&cur_size, &next_size, + total_size, one_percent, + &cur_percent, lte); + } ret = write_wim_resource(lte, out_fp, out_ctype, - <e->output_resource_entry, 0); + <e->output_resource_entry, + write_resource_flags); if (ret != 0) return ret; } + if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) + finish_stream_write_progress(total_size); return 0; } +#ifdef ENABLE_MULTITHREADED_COMPRESSION static int write_wim_chunks(struct message *msg, FILE *out_fp, struct chunk_table *chunk_tab) { @@ -930,7 +721,7 @@ static int write_wim_chunks(struct message *msg, FILE *out_fp, if (fwrite(msg->out_compressed_chunks[i], 1, chunk_csize, out_fp) != chunk_csize) { - ERROR_WITH_ERRNO("Failed to write WIM"); + ERROR_WITH_ERRNO("Failed to write WIM chunk"); return WIMLIB_ERR_WRITE; } @@ -940,25 +731,53 @@ static int write_wim_chunks(struct message *msg, FILE *out_fp, return 0; } +/* + * This function is executed by the main thread when the resources are being + * compressed in parallel. The main thread is in change of all reading of the + * uncompressed data and writing of the compressed data. The compressor threads + * *only* do compression from/to in-memory buffers. + * + * Each unit of work given to a compressor thread is up to MAX_CHUNKS_PER_MSG + * chunks of compressed data to compress, represented in a `struct message'. + * Each message is passed from the main thread to a worker thread through the + * res_to_compress_queue, and it is passed back through the + * compressed_res_queue. + */ static int main_writer_thread_proc(struct list_head *stream_list, FILE *out_fp, int out_ctype, struct shared_queue *res_to_compress_queue, struct shared_queue *compressed_res_queue, - size_t queue_size) + size_t queue_size, + int write_flags, + u64 total_size) { int ret; - struct message msgs[queue_size]; ZERO_ARRAY(msgs); + // Initially, all the messages are available to use. LIST_HEAD(available_msgs); - for (size_t i = 0; i < ARRAY_LEN(msgs); i++) list_add(&msgs[i].list, &available_msgs); - + // outstanding_resources is the list of resources that currently have + // had chunks sent off for compression. + // + // The first stream in outstanding_resources is the stream that is + // currently being written (cur_lte). + // + // The last stream in outstanding_resources is the stream that is + // currently being read and chunks fed to the compressor threads + // (next_lte). + // + // Depending on the number of threads and the sizes of the resource, + // the outstanding streams list may contain streams between cur_lte and + // next_lte that have all their chunks compressed or being compressed, + // but haven't been written yet. + // + LIST_HEAD(outstanding_resources); struct list_head *next_resource = stream_list->next; struct lookup_table_entry *next_lte = container_of(next_resource, struct lookup_table_entry, @@ -967,24 +786,27 @@ static int main_writer_thread_proc(struct list_head *stream_list, u64 next_chunk = 0; u64 next_num_chunks = wim_resource_chunks(next_lte); INIT_LIST_HEAD(&next_lte->msg_list); + list_add_tail(&next_lte->staging_list, &outstanding_resources); + + // As in write_wim_resource(), each resource we read is checksummed. SHA_CTX next_sha_ctx; sha1_init(&next_sha_ctx); - u8 next_hash[SHA1_HASH_SIZE]; - // Resources owning chunks that have been sent off for compression - LIST_HEAD(outstanding_resources); - - // Resources that are going to be written by the main thread + // Resources that don't need any chunks compressed are added to this + // list and written directly by the main thread. LIST_HEAD(my_resources); - list_add_tail(&next_lte->staging_list, &outstanding_resources); - struct lookup_table_entry *cur_lte = next_lte; struct chunk_table *cur_chunk_tab = NULL; struct lookup_table_entry *lte; struct message *msg; + u64 one_percent = total_size / 100; + u64 cur_size = 0; + u64 next_size = 0; + unsigned cur_percent = 0; + #ifdef WITH_NTFS_3G ntfs_inode *ni = NULL; #endif @@ -994,11 +816,15 @@ static int main_writer_thread_proc(struct list_head *stream_list, #else ret = prepare_resource_for_read(next_lte); #endif + if (ret != 0) + goto out; DEBUG("Initializing buffers for uncompressed " "and compressed data (%zu bytes needed)", queue_size * MAX_CHUNKS_PER_MSG * WIM_CHUNK_SIZE * 2); + // Pre-allocate all the buffers that will be needed to do the chunk + // compression. for (size_t i = 0; i < ARRAY_LEN(msgs); i++) { for (size_t j = 0; j < MAX_CHUNKS_PER_MSG; j++) { msgs[i].compressed_chunks[j] = MALLOC(WIM_CHUNK_SIZE); @@ -1014,6 +840,9 @@ static int main_writer_thread_proc(struct list_head *stream_list, } } + // This loop is executed until all resources have been written, except + // possibly a few that have been added to the @my_resources list for + // writing later. while (1) { // Send chunks to the compressor threads until either (a) there // are no more messages available since they were all sent off, @@ -1021,8 +850,6 @@ static int main_writer_thread_proc(struct list_head *stream_list, // compressed. while (!list_empty(&available_msgs) && next_lte != NULL) { - wimlib_assert(next_chunk < next_num_chunks); - // Get a message from the available messages // list msg = container_of(available_msgs.next, @@ -1043,12 +870,18 @@ static int main_writer_thread_proc(struct list_head *stream_list, unsigned size = WIM_CHUNK_SIZE; for (unsigned i = 0; i < msg->num_chunks; i++) { + + // Read chunk @next_chunk of the stream into the + // message so that a compressor thread can + // compress it. + if (next_chunk == next_num_chunks - 1 && wim_resource_size(next_lte) % WIM_CHUNK_SIZE != 0) { size = wim_resource_size(next_lte) % WIM_CHUNK_SIZE; } + DEBUG2("Read resource (size=%u, offset=%zu)", size, next_chunk * WIM_CHUNK_SIZE); @@ -1092,13 +925,11 @@ static int main_writer_thread_proc(struct list_head *stream_list, // Advance to the next resource. // - // If the next resource needs no compression, - // just write it with this thread (not now - // though--- we could be in the middle of - // writing another resource.) Keep doing this - // until we either get to the end of the - // resources list, or we get to a resource that - // needs compression. + // If the next resource needs no compression, just write + // it with this thread (not now though--- we could be in + // the middle of writing another resource.) Keep doing + // this until we either get to the end of the resources + // list, or we get to a resource that needs compression. while (1) { if (next_resource == stream_list) { @@ -1116,8 +947,9 @@ static int main_writer_thread_proc(struct list_head *stream_list, struct lookup_table_entry, staging_list); next_resource = next_resource->next; - if ((next_lte->resource_location == RESOURCE_IN_WIM - && wimlib_get_compression_type(next_lte->wim) == out_ctype) + if ((!(write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) + && next_lte->resource_location == RESOURCE_IN_WIM + && wimlib_get_compression_type(next_lte->wim) == out_ctype) || wim_resource_size(next_lte) == 0) { list_add_tail(&next_lte->staging_list, @@ -1143,7 +975,7 @@ static int main_writer_thread_proc(struct list_head *stream_list, } // If there are no outstanding resources, there are no more - // resources to compress. + // resources that need to be written. if (list_empty(&outstanding_resources)) { DEBUG("No outstanding resources! Done"); ret = 0; @@ -1177,16 +1009,22 @@ static int main_writer_thread_proc(struct list_head *stream_list, DEBUG2("Complete msg (begin_chunk=%"PRIu64")", msg->begin_chunk); if (msg->begin_chunk == 0) { DEBUG2("Begin chunk tab"); - // This is the first set of chunks. - // Leave space for the chunk table in - // the output file. + if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) { + show_stream_write_progress(&cur_size, + &next_size, + total_size, + one_percent, + &cur_percent, + cur_lte); + } + + // This is the first set of chunks. Leave space + // for the chunk table in the output file. off_t cur_offset = ftello(out_fp); if (cur_offset == -1) { ret = WIMLIB_ERR_WRITE; goto out; } - FREE(cur_chunk_tab); - cur_chunk_tab = NULL; ret = begin_wim_resource_chunk_tab(cur_lte, out_fp, cur_offset, @@ -1194,13 +1032,20 @@ static int main_writer_thread_proc(struct list_head *stream_list, if (ret != 0) goto out; } + + // Write the compressed chunks from the message. ret = write_wim_chunks(msg, out_fp, cur_chunk_tab); if (ret != 0) goto out; list_del(&msg->list); + + // This message is available to use for different chunks + // now. list_add(&msg->list, &available_msgs); + // Was this the last chunk of the stream? If so, + // finish it. if (list_empty(&cur_lte->msg_list) && msg->begin_chunk + msg->num_chunks == cur_chunk_tab->num_chunks) { @@ -1210,7 +1055,8 @@ static int main_writer_thread_proc(struct list_head *stream_list, out_fp, &res_csize); if (ret != 0) - return ret; + goto out; + cur_lte->output_resource_entry.size = res_csize; @@ -1225,6 +1071,9 @@ static int main_writer_thread_proc(struct list_head *stream_list, cur_lte->resource_entry.flags | WIM_RESHDR_FLAG_COMPRESSED; + FREE(cur_chunk_tab); + cur_chunk_tab = NULL; + struct list_head *next = cur_lte->staging_list.next; list_del(&cur_lte->staging_list); @@ -1238,12 +1087,28 @@ static int main_writer_thread_proc(struct list_head *stream_list, staging_list); } + // Since we just finished writing a stream, + // write any streams that have been added to the + // my_resources list for direct writing by the + // main thread (e.g. resources that don't need + // to be compressed because the desired + // compression type is the same as the previous + // compression type). struct lookup_table_entry *tmp; list_for_each_entry_safe(lte, tmp, &my_resources, staging_list) { + if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) { + show_stream_write_progress(&cur_size, + &next_size, + total_size, + one_percent, + &cur_percent, + lte); + } + ret = write_wim_resource(lte, out_fp, out_ctype, @@ -1265,6 +1130,14 @@ out: #endif if (ret == 0) { list_for_each_entry(lte, &my_resources, staging_list) { + if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) { + show_stream_write_progress(&cur_size, + &next_size, + total_size, + one_percent, + &cur_percent, + lte); + } ret = write_wim_resource(lte, out_fp, out_ctype, <e->output_resource_entry, @@ -1272,6 +1145,8 @@ out: if (ret != 0) break; } + if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) + finish_stream_write_progress(total_size); } else { size_t num_available_msgs = 0; struct list_head *cur; @@ -1281,7 +1156,7 @@ out: } while (num_available_msgs < ARRAY_LEN(msgs)) { - msg = shared_queue_get(compressed_res_queue); + shared_queue_get(compressed_res_queue); num_available_msgs++; } } @@ -1300,80 +1175,99 @@ out: return ret; } + +static const char *get_data_type(int ctype) +{ + switch (ctype) { + case WIM_COMPRESSION_TYPE_NONE: + return "uncompressed"; + case WIM_COMPRESSION_TYPE_LZX: + return "LZX-compressed"; + case WIM_COMPRESSION_TYPE_XPRESS: + return "XPRESS-compressed"; + } +} + static int write_stream_list_parallel(struct list_head *stream_list, FILE *out_fp, int out_ctype, - int write_flags) + int write_flags, u64 total_size, + unsigned num_threads) { int ret; - long nthreads; struct shared_queue res_to_compress_queue; struct shared_queue compressed_res_queue; + pthread_t *compressor_threads = NULL; - nthreads = sysconf(_SC_NPROCESSORS_ONLN); - if (nthreads < 1) { - WARNING("Could not determine number of processors! Assuming 1"); - goto out_serial; + if (num_threads == 0) { + long nthreads = sysconf(_SC_NPROCESSORS_ONLN); + if (nthreads < 1) { + WARNING("Could not determine number of processors! Assuming 1"); + goto out_serial; + } else { + num_threads = nthreads; + } } wimlib_assert(stream_list->next != stream_list); - { - pthread_t compressor_threads[nthreads]; - - static const double MESSAGES_PER_THREAD = 2.0; - size_t queue_size = (size_t)(nthreads * MESSAGES_PER_THREAD); + static const double MESSAGES_PER_THREAD = 2.0; + size_t queue_size = (size_t)(num_threads * MESSAGES_PER_THREAD); - DEBUG("Initializing shared queues (queue_size=%zu)", queue_size); + DEBUG("Initializing shared queues (queue_size=%zu)", queue_size); - ret = shared_queue_init(&res_to_compress_queue, queue_size); - if (ret != 0) - goto out_serial; + ret = shared_queue_init(&res_to_compress_queue, queue_size); + if (ret != 0) + goto out_serial; - ret = shared_queue_init(&compressed_res_queue, queue_size); - if (ret != 0) - goto out_destroy_res_to_compress_queue; - - - struct compressor_thread_params params; - params.res_to_compress_queue = &res_to_compress_queue; - params.compressed_res_queue = &compressed_res_queue; - params.compress = get_compress_func(out_ctype); - - for (long i = 0; i < nthreads; i++) { - DEBUG("pthread_create thread %ld", i); - ret = pthread_create(&compressor_threads[i], NULL, - compressor_thread_proc, ¶ms); - if (ret != 0) { - ERROR_WITH_ERRNO("Failed to create compressor " - "thread %ld", i); - nthreads = i; - goto out_join; - } + ret = shared_queue_init(&compressed_res_queue, queue_size); + if (ret != 0) + goto out_destroy_res_to_compress_queue; + + struct compressor_thread_params params; + params.res_to_compress_queue = &res_to_compress_queue; + params.compressed_res_queue = &compressed_res_queue; + params.compress = get_compress_func(out_ctype); + + compressor_threads = MALLOC(num_threads * sizeof(pthread_t)); + + for (unsigned i = 0; i < num_threads; i++) { + DEBUG("pthread_create thread %u", i); + ret = pthread_create(&compressor_threads[i], NULL, + compressor_thread_proc, ¶ms); + if (ret != 0) { + ret = -1; + ERROR_WITH_ERRNO("Failed to create compressor " + "thread %u", i); + num_threads = i; + goto out_join; } + } - if (write_flags & WIMLIB_WRITE_FLAG_VERBOSE) { - printf("Writing compressed data using %ld threads...\n", - nthreads); - } + if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) { + printf("Writing %s data using %u threads...\n", + get_data_type(out_ctype), num_threads); + } - ret = main_writer_thread_proc(stream_list, - out_fp, - out_ctype, - &res_to_compress_queue, - &compressed_res_queue, - queue_size); - - out_join: - for (long i = 0; i < nthreads; i++) - shared_queue_put(&res_to_compress_queue, NULL); - - for (long i = 0; i < nthreads; i++) { - if (pthread_join(compressor_threads[i], NULL)) { - WARNING("Failed to join compressor thread %ld: %s", - i, strerror(errno)); - } + ret = main_writer_thread_proc(stream_list, + out_fp, + out_ctype, + &res_to_compress_queue, + &compressed_res_queue, + queue_size, + write_flags, + total_size); + +out_join: + for (unsigned i = 0; i < num_threads; i++) + shared_queue_put(&res_to_compress_queue, NULL); + + for (unsigned i = 0; i < num_threads; i++) { + if (pthread_join(compressor_threads[i], NULL)) { + WARNING("Failed to join compressor thread %u: %s", + i, strerror(errno)); } } + FREE(compressor_threads); shared_queue_destroy(&compressed_res_queue); out_destroy_res_to_compress_queue: shared_queue_destroy(&res_to_compress_queue); @@ -1382,11 +1276,17 @@ out_destroy_res_to_compress_queue: out_serial: WARNING("Falling back to single-threaded compression"); return write_stream_list_serial(stream_list, out_fp, - out_ctype, write_flags); + out_ctype, write_flags, total_size); } +#endif +/* + * Write a list of streams to a WIM (@out_fp) using the compression type + * @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) + int out_ctype, int write_flags, + unsigned num_threads) { struct lookup_table_entry *lte; size_t num_streams = 0; @@ -1397,13 +1297,21 @@ static int write_stream_list(struct list_head *stream_list, FILE *out_fp, num_streams++; total_size += wim_resource_size(lte); if (!compression_needed - && out_ctype != WIM_COMPRESSION_TYPE_NONE - && (lte->resource_location != RESOURCE_IN_WIM - || wimlib_get_compression_type(lte->wim) != out_ctype) + && + (out_ctype != WIM_COMPRESSION_TYPE_NONE + && (lte->resource_location != RESOURCE_IN_WIM + || wimlib_get_compression_type(lte->wim) != out_ctype + || (write_flags & WIMLIB_WRITE_FLAG_REBUILD))) && wim_resource_size(lte) != 0) compression_needed = true; } + if (num_streams == 0) { + if (write_flags & WIMLIB_WRITE_FLAG_VERBOSE) + printf("No streams to write\n"); + return 0; + } + if (write_flags & WIMLIB_WRITE_FLAG_VERBOSE) { printf("Preparing to write %zu streams " "(%"PRIu64" total bytes uncompressed)\n", @@ -1412,16 +1320,26 @@ static int write_stream_list(struct list_head *stream_list, FILE *out_fp, wimlib_get_compression_type_string(out_ctype)); } - if (compression_needed && total_size >= 0) { // XXX +#ifdef ENABLE_MULTITHREADED_COMPRESSION + if (compression_needed && total_size >= 1000000 && num_threads != 1) { return write_stream_list_parallel(stream_list, out_fp, - out_ctype, write_flags); - } else { - if (write_flags & WIMLIB_WRITE_FLAG_VERBOSE) { - puts("Using 1 thread (no compression needed)"); + out_ctype, write_flags, + total_size, num_threads); + } + else +#endif + { + if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) { + const char *reason = ""; + if (!compression_needed) + reason = " (no compression needed)"; + printf("Writing %s data using 1 thread%s\n", + get_data_type(out_ctype), reason); } return write_stream_list_serial(stream_list, out_fp, - out_ctype, write_flags); + out_ctype, write_flags, + total_size); } } @@ -1435,7 +1353,7 @@ static int dentry_find_streams_to_write(struct dentry *dentry, 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) - list_add(<e->staging_list, stream_list); + list_add_tail(<e->staging_list, stream_list); } return 0; } @@ -1446,102 +1364,121 @@ static int find_streams_to_write(WIMStruct *w) dentry_find_streams_to_write, w); } -static int write_wim_streams(WIMStruct *w, int image, int write_flags) +static int write_wim_streams(WIMStruct *w, int image, int write_flags, + unsigned num_threads) { + for_lookup_table_entry(w->lookup_table, lte_zero_out_refcnt, NULL); LIST_HEAD(stream_list); - w->private = &stream_list; for_image(w, image, find_streams_to_write); return write_stream_list(&stream_list, w->out_fp, - wimlib_get_compression_type(w), write_flags); + wimlib_get_compression_type(w), write_flags, + num_threads); } /* - * Write the lookup table, xml data, and integrity table, then overwrite the WIM - * header. + * Finish writing a WIM file: write the lookup table, xml data, and integrity + * table (optional), then overwrite the WIM header. + * + * write_flags is a bitwise OR of the following: + * + * (public) WIMLIB_WRITE_FLAG_CHECK_INTEGRITY: + * Include an integrity table. + * + * (public) WIMLIB_WRITE_FLAG_SHOW_PROGRESS: + * Show progress information when (if) writing the integrity table. + * + * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE: + * Don't write the lookup table. + * + * (private) WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE: + * When (if) writing the integrity table, re-use entries from the + * existing integrity table, if possible. + * + * (private) WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML: + * After writing the XML data but before writing the integrity + * table, write a temporary WIM header and flush the stream so that + * the WIM is less likely to become corrupted upon abrupt program + * termination. + * + * (private) WIMLIB_WRITE_FLAG_FSYNC: + * fsync() the output file before closing it. + * */ int finish_write(WIMStruct *w, int image, int write_flags) { - off_t lookup_table_offset; - off_t xml_data_offset; - off_t lookup_table_size; - off_t integrity_offset; - off_t xml_data_size; - off_t end_offset; - off_t integrity_size; int ret; struct wim_header hdr; FILE *out = w->out_fp; - if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { - /* Write the lookup table. */ - lookup_table_offset = ftello(out); - if (lookup_table_offset == -1) - return WIMLIB_ERR_WRITE; - - DEBUG("Writing lookup table (offset %"PRIu64")", - lookup_table_offset); - ret = write_lookup_table(w->lookup_table, out); - if (ret != 0) - return ret; - } - - xml_data_offset = ftello(out); - if (xml_data_offset == -1) - return WIMLIB_ERR_WRITE; - /* @hdr will be the header for the new WIM. First copy all the data * from the header in the WIMStruct; then set all the fields that may * have changed, including the resource entries, boot index, and image * count. */ memcpy(&hdr, &w->hdr, sizeof(struct wim_header)); + if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { - lookup_table_size = xml_data_offset - lookup_table_offset; - hdr.lookup_table_res_entry.offset = lookup_table_offset; - hdr.lookup_table_res_entry.size = lookup_table_size; + ret = write_lookup_table(w->lookup_table, out, &hdr.lookup_table_res_entry); + if (ret != 0) + goto out; } - hdr.lookup_table_res_entry.original_size = hdr.lookup_table_res_entry.size; - hdr.lookup_table_res_entry.flags = WIM_RESHDR_FLAG_METADATA; - DEBUG("Writing XML data (offset %"PRIu64")", xml_data_offset); ret = write_xml_data(w->wim_info, image, out, (write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE) ? - wim_info_get_total_bytes(w->wim_info) : 0); + wim_info_get_total_bytes(w->wim_info) : 0, + &hdr.xml_res_entry); if (ret != 0) - return ret; + goto out; - integrity_offset = ftello(out); - if (integrity_offset == -1) - return WIMLIB_ERR_WRITE; - xml_data_size = integrity_offset - xml_data_offset; + if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) { + if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) { + struct wim_header checkpoint_hdr; + memcpy(&checkpoint_hdr, &hdr, sizeof(struct wim_header)); + memset(&checkpoint_hdr.integrity, 0, sizeof(struct resource_entry)); + if (fseeko(out, 0, SEEK_SET) != 0) { + ret = WIMLIB_ERR_WRITE; + goto out; + } + ret = write_header(&checkpoint_hdr, out); + if (ret != 0) + goto out; - hdr.xml_res_entry.offset = xml_data_offset; - hdr.xml_res_entry.size = xml_data_size; - hdr.xml_res_entry.original_size = xml_data_size; - hdr.xml_res_entry.flags = 0; + if (fflush(out) != 0) { + ERROR_WITH_ERRNO("Can't write data to WIM"); + ret = WIMLIB_ERR_WRITE; + goto out; + } - if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) { - ret = write_integrity_table(out, WIM_HEADER_DISK_SIZE, - xml_data_offset, - write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS); + if (fseeko(out, 0, SEEK_END) != 0) { + ret = WIMLIB_ERR_WRITE; + goto out; + } + } + + off_t old_lookup_table_end; + off_t new_lookup_table_end; + bool show_progress; + if (write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE) { + old_lookup_table_end = w->hdr.lookup_table_res_entry.offset + + w->hdr.lookup_table_res_entry.size; + } else { + old_lookup_table_end = 0; + } + new_lookup_table_end = hdr.lookup_table_res_entry.offset + + hdr.lookup_table_res_entry.size; + show_progress = ((write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) != 0); + + ret = write_integrity_table(out, + &hdr.integrity, + new_lookup_table_end, + old_lookup_table_end, + show_progress); if (ret != 0) - return ret; - end_offset = ftello(out); - if (end_offset == -1) - return WIMLIB_ERR_WRITE; - integrity_size = end_offset - integrity_offset; - hdr.integrity.offset = integrity_offset; - hdr.integrity.size = integrity_size; - hdr.integrity.original_size = integrity_size; + goto out; } else { - hdr.integrity.offset = 0; - hdr.integrity.size = 0; - hdr.integrity.original_size = 0; + memset(&hdr.integrity, 0, sizeof(struct resource_entry)); } - hdr.integrity.flags = 0; - - DEBUG("Updating WIM header."); /* * In the WIM header, there is room for the resource entry for a @@ -1570,69 +1507,73 @@ int finish_write(WIMStruct *w, int image, int write_flags) hdr.boot_idx = 0; } - - if (fseeko(out, 0, SEEK_SET) != 0) - return WIMLIB_ERR_WRITE; + if (fseeko(out, 0, SEEK_SET) != 0) { + ret = WIMLIB_ERR_WRITE; + goto out; + } ret = write_header(&hdr, out); if (ret != 0) - return ret; + goto out; - DEBUG("Closing output file."); - wimlib_assert(w->out_fp != NULL); - if (fclose(w->out_fp) != 0) { + if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) { + if (fflush(out) != 0 + || fsync(fileno(out)) != 0) + { + ERROR_WITH_ERRNO("Error flushing data to WIM file"); + ret = WIMLIB_ERR_WRITE; + } + } +out: + if (fclose(out) != 0) { ERROR_WITH_ERRNO("Failed to close the WIM file"); - ret = WIMLIB_ERR_WRITE; + if (ret == 0) + ret = WIMLIB_ERR_WRITE; } w->out_fp = NULL; return ret; } +static void close_wim_writable(WIMStruct *w) +{ + if (w->out_fp) { + if (fclose(w->out_fp) != 0) { + WARNING("Failed to close output WIM: %s", + strerror(errno)); + } + w->out_fp = NULL; + } +} + /* Open file stream and write dummy header for WIM. */ int begin_write(WIMStruct *w, const char *path, int write_flags) { - const char *mode; - DEBUG("Opening `%s' for new WIM", path); - - /* checking the integrity requires going back over the file to read it. - * XXX - * (It also would be possible to keep a running sha1sum as the file is - * written-- this would be faster, but a bit more complicated) */ + int ret; + bool need_readable = false; + bool trunc = true; if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) - mode = "w+b"; - else - mode = "wb"; - - if (w->out_fp) - fclose(w->out_fp); - - w->out_fp = fopen(path, mode); - if (!w->out_fp) { - ERROR_WITH_ERRNO("Failed to open the file `%s' for writing", - path); - return WIMLIB_ERR_OPEN; - } + need_readable = true; + ret = open_wim_writable(w, path, trunc, need_readable); + if (ret != 0) + return ret; /* Write dummy header. It will be overwritten later. */ return write_header(&w->hdr, w->out_fp); } /* Writes a stand-alone WIM to a file. */ WIMLIBAPI int wimlib_write(WIMStruct *w, const char *path, - int image, int write_flags) + int image, int write_flags, unsigned num_threads) { int ret; if (!w || !path) return WIMLIB_ERR_INVALID_PARAM; - write_flags &= ~WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE; - if (image != WIM_ALL_IMAGES && (image < 1 || image > w->hdr.image_count)) return WIMLIB_ERR_INVALID_IMAGE; - if (w->hdr.total_parts != 1) { ERROR("Cannot call wimlib_write() on part of a split WIM"); return WIMLIB_ERR_SPLIT_UNSUPPORTED; @@ -1645,29 +1586,315 @@ WIMLIBAPI int wimlib_write(WIMStruct *w, const char *path, ret = begin_write(w, path, write_flags); if (ret != 0) - return ret; + goto out; - for_lookup_table_entry(w->lookup_table, lte_zero_out_refcnt, NULL); + ret = write_wim_streams(w, image, write_flags, num_threads); + if (ret != 0) + goto out; - ret = write_wim_streams(w, image, write_flags); + if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) + printf("Writing image metadata...\n"); - if (ret != 0) { - /*ERROR("Failed to write WIM file resources to `%s'", path);*/ - return ret; + ret = for_image(w, image, write_metadata_resource); + if (ret != 0) + goto out; + + ret = finish_write(w, image, write_flags); + if (ret == 0 && (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS)) + printf("Successfully wrote `%s'\n", path); +out: + close_wim_writable(w); + return ret; +} + +static int lte_overwrite_prepare(struct lookup_table_entry *lte, + void *ignore) +{ + memcpy(<e->output_resource_entry, <e->resource_entry, + sizeof(struct resource_entry)); + lte->out_refcnt = 0; + return 0; +} + +static int check_resource_offset(struct lookup_table_entry *lte, void *arg) +{ + off_t end_offset = *(u64*)arg; + + wimlib_assert(lte->out_refcnt <= lte->refcnt); + if (lte->out_refcnt < lte->refcnt) { + if (lte->resource_entry.offset + lte->resource_entry.size > end_offset) { + ERROR("The following resource is after the XML data:"); + print_lookup_table_entry(lte); + return WIMLIB_ERR_RESOURCE_ORDER; + } } + return 0; +} - ret = for_image(w, image, write_metadata_resource); +static int find_new_streams(struct lookup_table_entry *lte, void *arg) +{ + if (lte->out_refcnt == lte->refcnt) + list_add(<e->staging_list, (struct list_head*)arg); + else + lte->out_refcnt = lte->refcnt; + return 0; +} - if (ret != 0) { - /*ERROR("Failed to write WIM image metadata to `%s'", path);*/ +/* + * Overwrite a WIM, possibly appending streams to it. + * + * A WIM looks like (or is supposed to look like) the following: + * + * Header (212 bytes) + * Streams and metadata resources (variable size) + * Lookup table (variable size) + * XML data (variable size) + * Integrity table (optional) (variable size) + * + * If we are not adding any streams or metadata resources, the lookup table is + * unchanged--- so we only need to overwrite the XML data, integrity table, and + * header. This operation is potentially unsafe if the program is abruptly + * terminated while the XML data or integrity table are being overwritten, but + * before the new header has been written. To partially alleviate this problem, + * a special flag (WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) is passed to + * finish_write() to cause a temporary WIM header to be written after the XML + * data has been written. This may prevent the WIM from becoming corrupted if + * the program is terminated while the integrity table is being calculated (but + * no guarantees, due to write re-ordering...). + * + * If we are adding new streams or images (metadata resources), the lookup table + * needs to be changed, and those streams need to be written. In this case, we + * try to perform a safe update of the WIM file by writing the streams *after* + * the end of the previous WIM, then writing the new lookup table, XML data, and + * (optionally) integrity table following the new streams. This will produce a + * layout like the following: + * + * Header (212 bytes) + * (OLD) Streams and metadata resources (variable size) + * (OLD) Lookup table (variable size) + * (OLD) XML data (variable size) + * (OLD) Integrity table (optional) (variable size) + * (NEW) Streams and metadata resources (variable size) + * (NEW) Lookup table (variable size) + * (NEW) XML data (variable size) + * (NEW) Integrity table (optional) (variable size) + * + * At all points, the WIM is valid as nothing points to the new data yet. Then, + * the header is overwritten to point to the new lookup table, XML data, and + * integrity table, to produce the following layout: + * + * Header (212 bytes) + * Streams and metadata resources (variable size) + * Nothing (variable size) + * More Streams and metadata resources (variable size) + * Lookup table (variable size) + * XML data (variable size) + * Integrity table (optional) (variable size) + * + * This method allows an image to be appended to a large WIM very quickly, and + * is is crash-safe except in the case of write re-ordering, but the + * disadvantage is that a small hole is left in the WIM where the old lookup + * table, xml data, and integrity table were. (These usually only take up a + * small amount of space compared to the streams, however. + */ +static int overwrite_wim_inplace(WIMStruct *w, int write_flags, + unsigned num_threads, + int modified_image_idx) +{ + int ret; + struct list_head stream_list; + off_t old_wim_end; + + DEBUG("Overwriting `%s' 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 + * lookup tables after the XML data. Otherwise, these data would be + * overwritten. */ + if (w->hdr.integrity.offset != 0 && + w->hdr.integrity.offset < w->hdr.xml_res_entry.offset) { + ERROR("Didn't expect the integrity table to be before the XML data"); + return WIMLIB_ERR_RESOURCE_ORDER; + } + + if (w->hdr.lookup_table_res_entry.offset > w->hdr.xml_res_entry.offset) { + ERROR("Didn't expect the lookup table to be after the XML data"); + return WIMLIB_ERR_RESOURCE_ORDER; + } + + DEBUG("Identifying newly added streams"); + for_lookup_table_entry(w->lookup_table, lte_overwrite_prepare, NULL); + INIT_LIST_HEAD(&stream_list); + for (int i = modified_image_idx; i < w->hdr.image_count; i++) { + DEBUG("Identifiying streams in image %d", i + 1); + wimlib_assert(w->image_metadata[i].modified); + wimlib_assert(!w->image_metadata[i].has_been_mounted_rw); + wimlib_assert(w->image_metadata[i].root_dentry != NULL); + wimlib_assert(w->image_metadata[i].metadata_lte != NULL); + w->private = &stream_list; + for_dentry_in_tree(w->image_metadata[i].root_dentry, + dentry_find_streams_to_write, w); + } + + if (w->hdr.integrity.offset) + old_wim_end = w->hdr.integrity.offset + w->hdr.integrity.size; + else + old_wim_end = w->hdr.xml_res_entry.offset + w->hdr.xml_res_entry.size; + + ret = for_lookup_table_entry(w->lookup_table, check_resource_offset, + &old_wim_end); + if (ret != 0) return ret; + + if (modified_image_idx == w->hdr.image_count) { + /* If no images are modified, a new lookup table does not need + * to be written. */ + wimlib_assert(list_empty(&stream_list)); + old_wim_end = w->hdr.lookup_table_res_entry.offset + + w->hdr.lookup_table_res_entry.size; + write_flags |= WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE | + WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML; } - ret = finish_write(w, image, write_flags); + INIT_LIST_HEAD(&stream_list); + for_lookup_table_entry(w->lookup_table, find_new_streams, + &stream_list); + + ret = open_wim_writable(w, w->filename, false, + (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0); if (ret != 0) return ret; - if (write_flags & WIMLIB_WRITE_FLAG_VERBOSE) - printf("Successfully wrote `%s'\n", path); - return 0; + if (fseeko(w->out_fp, old_wim_end, SEEK_SET) != 0) { + ERROR_WITH_ERRNO("Can't seek to end of WIM"); + 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); + if (ret != 0) + goto out_ftruncate; + } else { + DEBUG("No new streams were added"); + } + + for (int i = modified_image_idx; i < w->hdr.image_count; i++) { + select_wim_image(w, i + 1); + ret = write_metadata_resource(w); + if (ret != 0) + goto out_ftruncate; + } + write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE; + ret = finish_write(w, WIM_ALL_IMAGES, write_flags); +out_ftruncate: + close_wim_writable(w); + if (ret != 0) { + WARNING("Truncating `%s' to its original size (%"PRIu64" bytes)", + w->filename, old_wim_end); + truncate(w->filename, old_wim_end); + } + return ret; +} + +static int overwrite_wim_via_tmpfile(WIMStruct *w, int write_flags, + unsigned num_threads) +{ + size_t wim_name_len; + int ret; + + DEBUG("Overwrining `%s' 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); + char tmpfile[wim_name_len + 10]; + memcpy(tmpfile, w->filename, wim_name_len); + randomize_char_array_with_alnum(tmpfile + wim_name_len, 9); + tmpfile[wim_name_len + 9] = '\0'; + + ret = wimlib_write(w, tmpfile, WIM_ALL_IMAGES, + write_flags | WIMLIB_WRITE_FLAG_FSYNC, + num_threads); + if (ret != 0) { + ERROR("Failed to write the WIM file `%s'", tmpfile); + goto err; + } + + /* Close the original WIM file that was opened for reading. */ + if (w->fp != NULL) { + fclose(w->fp); + w->fp = NULL; + } + + DEBUG("Renaming `%s' to `%s'", tmpfile, w->filename); + + /* Rename the new file to the old file .*/ + if (rename(tmpfile, w->filename) != 0) { + ERROR_WITH_ERRNO("Failed to rename `%s' to `%s'", + tmpfile, w->filename); + ret = WIMLIB_ERR_RENAME; + goto err; + } + + if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) + printf("Successfully renamed `%s' to `%s'\n", tmpfile, w->filename); + + /* Re-open the WIM read-only. */ + w->fp = fopen(w->filename, "rb"); + if (w->fp == NULL) { + ret = WIMLIB_ERR_REOPEN; + WARNING("Failed to re-open `%s' read-only: %s", + w->filename, strerror(errno)); + } + return ret; +err: + /* Remove temporary file. */ + if (unlink(tmpfile) != 0) + WARNING("Failed to remove `%s': %s", tmpfile, strerror(errno)); + return ret; +} + +/* + * Writes a WIM file to the original file that it was read from, overwriting it. + */ +WIMLIBAPI int wimlib_overwrite(WIMStruct *w, int write_flags, + unsigned num_threads) +{ + if (!w) + return WIMLIB_ERR_INVALID_PARAM; + + write_flags &= WIMLIB_WRITE_MASK_PUBLIC; + + if (!w->filename) + return WIMLIB_ERR_NO_FILENAME; + + if (w->hdr.total_parts != 1) { + ERROR("Cannot modify a split WIM"); + return WIMLIB_ERR_SPLIT_UNSUPPORTED; + } + + if (!w->deletion_occurred && !(write_flags & WIMLIB_WRITE_FLAG_REBUILD)) { + int i, modified_image_idx; + for (i = 0; i < w->hdr.image_count && !w->image_metadata[i].modified; i++) + ; + modified_image_idx = i; + for (; i < w->hdr.image_count && w->image_metadata[i].modified && + !w->image_metadata[i].has_been_mounted_rw; i++) + ; + if (i == w->hdr.image_count) { + return overwrite_wim_inplace(w, write_flags, num_threads, + modified_image_idx); + } + } + return overwrite_wim_via_tmpfile(w, write_flags, num_threads); +} + +/* Deprecated */ +WIMLIBAPI int wimlib_overwrite_xml_and_header(WIMStruct *wim, int write_flags) +{ + return wimlib_overwrite(wim, write_flags, 1); }