4 * Support for writing WIM files; write a WIM file, overwrite a WIM file, write
5 * compressed file resources, etc.
9 * Copyright (C) 2012, 2013 Eric Biggers
11 * This file is part of wimlib, a library for working with WIM files.
13 * wimlib is free software; you can redistribute it and/or modify it under the
14 * terms of the GNU General Public License as published by the Free
15 * Software Foundation; either version 3 of the License, or (at your option)
18 * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
19 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
20 * A PARTICULAR PURPOSE. See the GNU General Public License for more
23 * You should have received a copy of the GNU General Public License
24 * along with wimlib; if not, see http://www.gnu.org/licenses/.
29 #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK)
30 /* On BSD, this should be included before "list.h" so that "list.h" can
31 * overwrite the LIST_HEAD macro. */
43 #include "wimlib_internal.h"
44 #include "buffer_io.h"
46 #include "lookup_table.h"
51 #ifdef ENABLE_MULTITHREADED_COMPRESSION
60 #include <ntfs-3g/attrib.h>
61 #include <ntfs-3g/inode.h>
62 #include <ntfs-3g/dir.h>
78 static int fflush_and_ftruncate(FILE *fp, off_t size)
84 ERROR_WITH_ERRNO("Failed to flush data to output WIM file");
85 return WIMLIB_ERR_WRITE;
87 ret = ftruncate(fileno(fp), size);
89 ERROR_WITH_ERRNO("Failed to truncate output WIM file to "
90 "%"PRIu64" bytes", size);
91 return WIMLIB_ERR_WRITE;
96 /* Chunk table that's located at the beginning of each compressed resource in
97 * the WIM. (This is not the on-disk format; the on-disk format just has an
98 * array of offsets.) */
102 u64 original_resource_size;
103 u64 bytes_per_chunk_entry;
111 * Allocates and initializes a chunk table, and reserves space for it in the
115 begin_wim_resource_chunk_tab(const struct wim_lookup_table_entry *lte,
118 struct chunk_table **chunk_tab_ret)
120 u64 size = wim_resource_size(lte);
121 u64 num_chunks = (size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE;
122 size_t alloc_size = sizeof(struct chunk_table) + num_chunks * sizeof(u64);
123 struct chunk_table *chunk_tab = CALLOC(1, alloc_size);
127 ERROR("Failed to allocate chunk table for %"PRIu64" byte "
129 ret = WIMLIB_ERR_NOMEM;
132 chunk_tab->file_offset = file_offset;
133 chunk_tab->num_chunks = num_chunks;
134 chunk_tab->original_resource_size = size;
135 chunk_tab->bytes_per_chunk_entry = (size >= (1ULL << 32)) ? 8 : 4;
136 chunk_tab->table_disk_size = chunk_tab->bytes_per_chunk_entry *
138 chunk_tab->cur_offset = 0;
139 chunk_tab->cur_offset_p = chunk_tab->offsets;
141 if (fwrite(chunk_tab, 1, chunk_tab->table_disk_size, out_fp) !=
142 chunk_tab->table_disk_size) {
143 ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
145 ret = WIMLIB_ERR_WRITE;
151 *chunk_tab_ret = chunk_tab;
156 * Pointer to function to compresses a chunk of a WIM resource.
158 * @chunk: Uncompressed data of the chunk.
159 * @chunk_size: Size of the uncompressed chunk in bytes.
160 * @compressed_chunk: Pointer to output buffer of size at least
161 * (@chunk_size - 1) bytes.
162 * @compressed_chunk_len_ret: Pointer to an unsigned int into which the size
163 * of the compressed chunk will be
166 * Returns zero if compressed succeeded, and nonzero if the chunk could not be
167 * compressed to any smaller than @chunk_size. This function cannot fail for
170 typedef int (*compress_func_t)(const void *, unsigned, void *, unsigned *);
172 compress_func_t get_compress_func(int out_ctype)
174 if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX)
177 return xpress_compress;
181 * Writes a chunk of a WIM resource to an output file.
183 * @chunk: Uncompressed data of the chunk.
184 * @chunk_size: Size of the chunk (<= WIM_CHUNK_SIZE)
185 * @out_fp: FILE * to write tho chunk to.
186 * @out_ctype: Compression type to use when writing the chunk (ignored if no
187 * chunk table provided)
188 * @chunk_tab: Pointer to chunk table being created. It is updated with the
189 * offset of the chunk we write.
191 * Returns 0 on success; nonzero on failure.
193 static int write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size,
194 FILE *out_fp, compress_func_t compress,
195 struct chunk_table *chunk_tab)
198 unsigned out_chunk_size;
200 u8 *compressed_chunk = alloca(chunk_size);
203 ret = compress(chunk, chunk_size, compressed_chunk,
206 out_chunk = compressed_chunk;
209 out_chunk_size = chunk_size;
211 *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset;
212 chunk_tab->cur_offset += out_chunk_size;
215 out_chunk_size = chunk_size;
217 if (fwrite(out_chunk, 1, out_chunk_size, out_fp) != out_chunk_size) {
218 ERROR_WITH_ERRNO("Failed to write WIM resource chunk");
219 return WIMLIB_ERR_WRITE;
225 * Finishes a WIM chunk table and writes it to the output file at the correct
228 * The final size of the full compressed resource is returned in the
229 * @compressed_size_p.
232 finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab,
233 FILE *out_fp, u64 *compressed_size_p)
235 size_t bytes_written;
236 if (fseeko(out_fp, chunk_tab->file_offset, SEEK_SET) != 0) {
237 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of output "
238 "WIM file", chunk_tab->file_offset);
239 return WIMLIB_ERR_WRITE;
242 if (chunk_tab->bytes_per_chunk_entry == 8) {
243 array_cpu_to_le64(chunk_tab->offsets, chunk_tab->num_chunks);
245 for (u64 i = 0; i < chunk_tab->num_chunks; i++)
246 ((u32*)chunk_tab->offsets)[i] =
247 cpu_to_le32(chunk_tab->offsets[i]);
249 bytes_written = fwrite((u8*)chunk_tab->offsets +
250 chunk_tab->bytes_per_chunk_entry,
251 1, chunk_tab->table_disk_size, out_fp);
252 if (bytes_written != chunk_tab->table_disk_size) {
253 ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
255 return WIMLIB_ERR_WRITE;
257 if (fseeko(out_fp, 0, SEEK_END) != 0) {
258 ERROR_WITH_ERRNO("Failed to seek to end of output WIM file");
259 return WIMLIB_ERR_WRITE;
261 *compressed_size_p = chunk_tab->cur_offset + chunk_tab->table_disk_size;
265 /* Prepare for multiple reads to a resource by caching a FILE * or NTFS
266 * attribute pointer in the lookup table entry. */
267 static int prepare_resource_for_read(struct wim_lookup_table_entry *lte
270 , ntfs_inode **ni_ret
274 switch (lte->resource_location) {
275 case RESOURCE_IN_FILE_ON_DISK:
276 if (!lte->file_on_disk_fp) {
277 lte->file_on_disk_fp = fopen(lte->file_on_disk, "rb");
278 if (!lte->file_on_disk_fp) {
279 ERROR_WITH_ERRNO("Failed to open the file "
280 "`%s'", lte->file_on_disk);
281 return WIMLIB_ERR_OPEN;
286 case RESOURCE_IN_NTFS_VOLUME:
288 struct ntfs_location *loc = lte->ntfs_loc;
291 ni = ntfs_pathname_to_inode(*loc->ntfs_vol_p, NULL, loc->path_utf8);
293 ERROR_WITH_ERRNO("Failed to open inode `%s' in NTFS "
294 "volume", loc->path_utf8);
295 return WIMLIB_ERR_NTFS_3G;
297 lte->attr = ntfs_attr_open(ni,
298 loc->is_reparse_point ? AT_REPARSE_POINT : AT_DATA,
299 (ntfschar*)loc->stream_name_utf16,
300 loc->stream_name_utf16_num_chars);
302 ERROR_WITH_ERRNO("Failed to open attribute of `%s' in "
303 "NTFS volume", loc->path_utf8);
304 ntfs_inode_close(ni);
305 return WIMLIB_ERR_NTFS_3G;
311 #if defined(__CYGWIN__) || defined(__WIN32__)
313 if (!lte->file_on_disk_fp) {
314 lte->file_on_disk_fp = win32_open_file_readonly(lte->file_on_disk);
315 if (!lte->file_on_disk_fp)
316 return WIMLIB_ERR_OPEN;
326 /* Undo prepare_resource_for_read() by closing the cached FILE * or NTFS
328 static void end_wim_resource_read(struct wim_lookup_table_entry *lte
334 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK
335 && lte->file_on_disk_fp)
337 fclose(lte->file_on_disk_fp);
338 lte->file_on_disk_fp = NULL;
341 else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME) {
343 ntfs_attr_close(lte->attr);
347 ntfs_inode_close(ni);
350 #if defined(__CYGWIN__) || defined(__WIN32__)
351 else if (lte->resource_location == RESOURCE_WIN32
352 && lte->file_on_disk_fp)
354 win32_close_file(lte->file_on_disk_fp);
355 lte->file_on_disk_fp = NULL;
361 write_uncompressed_resource_and_truncate(struct wim_lookup_table_entry *lte,
364 struct resource_entry *out_res_entry)
367 if (fseeko(out_fp, file_offset, SEEK_SET) != 0) {
368 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of "
369 "output WIM file", file_offset);
370 return WIMLIB_ERR_WRITE;
372 ret = write_wim_resource(lte, out_fp, WIMLIB_COMPRESSION_TYPE_NONE,
377 return fflush_and_ftruncate(out_fp,
378 file_offset + wim_resource_size(lte));
382 * Writes a WIM resource to a FILE * opened for writing. The resource may be
383 * written uncompressed or compressed depending on the @out_ctype parameter.
385 * If by chance the resource compresses to more than the original size (this may
386 * happen with random data or files than are pre-compressed), the resource is
387 * instead written uncompressed (and this is reflected in the @out_res_entry by
388 * removing the WIM_RESHDR_FLAG_COMPRESSED flag).
390 * @lte: The lookup table entry for the WIM resource.
391 * @out_fp: The FILE * to write the resource to.
392 * @out_ctype: The compression type of the resource to write. Note: if this is
393 * the same as the compression type of the WIM resource we
394 * need to read, we simply copy the data (i.e. we do not
395 * uncompress it, then compress it again).
396 * @out_res_entry: If non-NULL, a resource entry that is filled in with the
397 * offset, original size, compressed size, and compression flag
398 * of the output resource.
400 * Returns 0 on success; nonzero on failure.
402 int write_wim_resource(struct wim_lookup_table_entry *lte,
403 FILE *out_fp, int out_ctype,
404 struct resource_entry *out_res_entry,
409 u64 old_compressed_size;
410 u64 new_compressed_size;
413 struct chunk_table *chunk_tab = NULL;
416 compress_func_t compress = NULL;
418 ntfs_inode *ni = NULL;
423 /* Original size of the resource */
424 original_size = wim_resource_size(lte);
426 /* Compressed size of the resource (as it exists now) */
427 old_compressed_size = wim_resource_compressed_size(lte);
429 /* Current offset in output file */
430 file_offset = ftello(out_fp);
431 if (file_offset == -1) {
432 ERROR_WITH_ERRNO("Failed to get offset in output "
434 return WIMLIB_ERR_WRITE;
437 /* Are the compression types the same? If so, do a raw copy (copy
438 * without decompressing and recompressing the data). */
439 raw = (wim_resource_compression_type(lte) == out_ctype
440 && out_ctype != WIMLIB_COMPRESSION_TYPE_NONE
441 && !(flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS));
444 flags |= WIMLIB_RESOURCE_FLAG_RAW;
445 bytes_remaining = old_compressed_size;
447 flags &= ~WIMLIB_RESOURCE_FLAG_RAW;
448 bytes_remaining = original_size;
451 /* Empty resource; nothing needs to be done, so just return success. */
452 if (bytes_remaining == 0)
455 /* Buffer for reading chunks for the resource */
456 u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
458 /* If we are writing a compressed resource and not doing a raw copy, we
459 * need to initialize the chunk table */
460 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && !raw) {
461 ret = begin_wim_resource_chunk_tab(lte, out_fp, file_offset,
467 /* If the WIM resource is in an external file, open a FILE * to it so we
468 * don't have to open a temporary one in read_wim_resource() for each
471 ret = prepare_resource_for_read(lte, &ni);
473 ret = prepare_resource_for_read(lte);
478 /* If we aren't doing a raw copy, we will compute the SHA1 message
479 * digest of the resource as we read it, and verify it's the same as the
480 * hash given in the lookup table entry once we've finished reading the
485 compress = get_compress_func(out_ctype);
489 /* While there are still bytes remaining in the WIM resource, read a
490 * chunk of the resource, update SHA1, then write that chunk using the
491 * desired compression type. */
493 u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE);
494 ret = read_wim_resource(lte, buf, to_read, offset, flags);
498 sha1_update(&ctx, buf, to_read);
499 ret = write_wim_resource_chunk(buf, to_read, out_fp,
500 compress, chunk_tab);
503 bytes_remaining -= to_read;
505 } while (bytes_remaining);
507 /* Raw copy: The new compressed size is the same as the old compressed
510 * Using WIMLIB_COMPRESSION_TYPE_NONE: The new compressed size is the
513 * Using a different compression type: Call
514 * finish_wim_resource_chunk_tab() and it will provide the new
518 new_compressed_size = old_compressed_size;
520 if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE)
521 new_compressed_size = original_size;
523 ret = finish_wim_resource_chunk_tab(chunk_tab, out_fp,
524 &new_compressed_size);
530 /* Verify SHA1 message digest of the resource, unless we are doing a raw
531 * write (in which case we never even saw the uncompressed data). Or,
532 * if the hash we had before is all 0's, just re-set it to be the new
535 u8 md[SHA1_HASH_SIZE];
536 sha1_final(md, &ctx);
537 if (is_zero_hash(lte->hash)) {
538 copy_hash(lte->hash, md);
539 } else if (!hashes_equal(md, lte->hash)) {
540 ERROR("WIM resource has incorrect hash!");
541 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK) {
542 ERROR("We were reading it from `%s'; maybe it changed "
543 "while we were reading it.",
546 ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
551 if (!raw && new_compressed_size >= original_size &&
552 out_ctype != WIMLIB_COMPRESSION_TYPE_NONE)
554 /* Oops! We compressed the resource to larger than the original
555 * size. Write the resource uncompressed instead. */
556 ret = write_uncompressed_resource_and_truncate(lte,
564 out_res_entry->size = new_compressed_size;
565 out_res_entry->original_size = original_size;
566 out_res_entry->offset = file_offset;
567 out_res_entry->flags = lte->resource_entry.flags
568 & ~WIM_RESHDR_FLAG_COMPRESSED;
569 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE)
570 out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED;
576 end_wim_resource_read(lte, ni);
578 end_wim_resource_read(lte);
585 #ifdef ENABLE_MULTITHREADED_COMPRESSION
587 /* Blocking shared queue (solves the producer-consumer problem) */
588 struct shared_queue {
592 unsigned filled_slots;
594 pthread_mutex_t lock;
595 pthread_cond_t msg_avail_cond;
596 pthread_cond_t space_avail_cond;
599 static int shared_queue_init(struct shared_queue *q, unsigned size)
601 wimlib_assert(size != 0);
602 q->array = CALLOC(sizeof(q->array[0]), size);
604 return WIMLIB_ERR_NOMEM;
609 pthread_mutex_init(&q->lock, NULL);
610 pthread_cond_init(&q->msg_avail_cond, NULL);
611 pthread_cond_init(&q->space_avail_cond, NULL);
615 static void shared_queue_destroy(struct shared_queue *q)
618 pthread_mutex_destroy(&q->lock);
619 pthread_cond_destroy(&q->msg_avail_cond);
620 pthread_cond_destroy(&q->space_avail_cond);
623 static void shared_queue_put(struct shared_queue *q, void *obj)
625 pthread_mutex_lock(&q->lock);
626 while (q->filled_slots == q->size)
627 pthread_cond_wait(&q->space_avail_cond, &q->lock);
629 q->back = (q->back + 1) % q->size;
630 q->array[q->back] = obj;
633 pthread_cond_broadcast(&q->msg_avail_cond);
634 pthread_mutex_unlock(&q->lock);
637 static void *shared_queue_get(struct shared_queue *q)
641 pthread_mutex_lock(&q->lock);
642 while (q->filled_slots == 0)
643 pthread_cond_wait(&q->msg_avail_cond, &q->lock);
645 obj = q->array[q->front];
646 q->array[q->front] = NULL;
647 q->front = (q->front + 1) % q->size;
650 pthread_cond_broadcast(&q->space_avail_cond);
651 pthread_mutex_unlock(&q->lock);
655 struct compressor_thread_params {
656 struct shared_queue *res_to_compress_queue;
657 struct shared_queue *compressed_res_queue;
658 compress_func_t compress;
661 #define MAX_CHUNKS_PER_MSG 2
664 struct wim_lookup_table_entry *lte;
665 u8 *uncompressed_chunks[MAX_CHUNKS_PER_MSG];
666 u8 *out_compressed_chunks[MAX_CHUNKS_PER_MSG];
667 u8 *compressed_chunks[MAX_CHUNKS_PER_MSG];
668 unsigned uncompressed_chunk_sizes[MAX_CHUNKS_PER_MSG];
669 unsigned compressed_chunk_sizes[MAX_CHUNKS_PER_MSG];
671 struct list_head list;
676 static void compress_chunks(struct message *msg, compress_func_t compress)
678 for (unsigned i = 0; i < msg->num_chunks; i++) {
679 DEBUG2("compress chunk %u of %u", i, msg->num_chunks);
680 int ret = compress(msg->uncompressed_chunks[i],
681 msg->uncompressed_chunk_sizes[i],
682 msg->compressed_chunks[i],
683 &msg->compressed_chunk_sizes[i]);
685 msg->out_compressed_chunks[i] = msg->compressed_chunks[i];
687 msg->out_compressed_chunks[i] = msg->uncompressed_chunks[i];
688 msg->compressed_chunk_sizes[i] = msg->uncompressed_chunk_sizes[i];
693 /* Compressor thread routine. This is a lot simpler than the main thread
694 * routine: just repeatedly get a group of chunks from the
695 * res_to_compress_queue, compress them, and put them in the
696 * compressed_res_queue. A NULL pointer indicates that the thread should stop.
698 static void *compressor_thread_proc(void *arg)
700 struct compressor_thread_params *params = arg;
701 struct shared_queue *res_to_compress_queue = params->res_to_compress_queue;
702 struct shared_queue *compressed_res_queue = params->compressed_res_queue;
703 compress_func_t compress = params->compress;
706 DEBUG("Compressor thread ready");
707 while ((msg = shared_queue_get(res_to_compress_queue)) != NULL) {
708 compress_chunks(msg, compress);
709 shared_queue_put(compressed_res_queue, msg);
711 DEBUG("Compressor thread terminating");
716 static int do_write_stream_list(struct list_head *my_resources,
719 wimlib_progress_func_t progress_func,
720 union wimlib_progress_info *progress,
721 int write_resource_flags)
724 struct wim_lookup_table_entry *lte, *tmp;
726 list_for_each_entry_safe(lte, tmp, my_resources, staging_list) {
727 ret = write_wim_resource(lte,
730 <e->output_resource_entry,
731 write_resource_flags);
734 list_del(<e->staging_list);
735 progress->write_streams.completed_bytes +=
736 wim_resource_size(lte);
737 progress->write_streams.completed_streams++;
739 progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
746 static int write_stream_list_serial(struct list_head *stream_list,
750 wimlib_progress_func_t progress_func,
751 union wimlib_progress_info *progress)
753 int write_resource_flags;
755 if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
756 write_resource_flags = WIMLIB_RESOURCE_FLAG_RECOMPRESS;
758 write_resource_flags = 0;
759 progress->write_streams.num_threads = 1;
761 progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress);
762 return do_write_stream_list(stream_list, out_fp,
763 out_ctype, progress_func,
764 progress, write_resource_flags);
767 #ifdef ENABLE_MULTITHREADED_COMPRESSION
768 static int write_wim_chunks(struct message *msg, FILE *out_fp,
769 struct chunk_table *chunk_tab)
771 for (unsigned i = 0; i < msg->num_chunks; i++) {
772 unsigned chunk_csize = msg->compressed_chunk_sizes[i];
774 DEBUG2("Write wim chunk %u of %u (csize = %u)",
775 i, msg->num_chunks, chunk_csize);
777 if (fwrite(msg->out_compressed_chunks[i], 1, chunk_csize, out_fp)
780 ERROR_WITH_ERRNO("Failed to write WIM chunk");
781 return WIMLIB_ERR_WRITE;
784 *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset;
785 chunk_tab->cur_offset += chunk_csize;
791 * This function is executed by the main thread when the resources are being
792 * compressed in parallel. The main thread is in change of all reading of the
793 * uncompressed data and writing of the compressed data. The compressor threads
794 * *only* do compression from/to in-memory buffers.
796 * Each unit of work given to a compressor thread is up to MAX_CHUNKS_PER_MSG
797 * chunks of compressed data to compress, represented in a `struct message'.
798 * Each message is passed from the main thread to a worker thread through the
799 * res_to_compress_queue, and it is passed back through the
800 * compressed_res_queue.
802 static int main_writer_thread_proc(struct list_head *stream_list,
805 struct shared_queue *res_to_compress_queue,
806 struct shared_queue *compressed_res_queue,
809 wimlib_progress_func_t progress_func,
810 union wimlib_progress_info *progress)
813 struct chunk_table *cur_chunk_tab = NULL;
814 struct message *msgs = CALLOC(num_messages, sizeof(struct message));
815 struct wim_lookup_table_entry *next_lte = NULL;
817 // Initially, all the messages are available to use.
818 LIST_HEAD(available_msgs);
821 ret = WIMLIB_ERR_NOMEM;
825 for (size_t i = 0; i < num_messages; i++)
826 list_add(&msgs[i].list, &available_msgs);
828 // outstanding_resources is the list of resources that currently have
829 // had chunks sent off for compression.
831 // The first stream in outstanding_resources is the stream that is
832 // currently being written (cur_lte).
834 // The last stream in outstanding_resources is the stream that is
835 // currently being read and chunks fed to the compressor threads
838 // Depending on the number of threads and the sizes of the resource,
839 // the outstanding streams list may contain streams between cur_lte and
840 // next_lte that have all their chunks compressed or being compressed,
841 // but haven't been written yet.
843 LIST_HEAD(outstanding_resources);
844 struct list_head *next_resource = stream_list->next;
846 u64 next_num_chunks = 0;
848 // As in write_wim_resource(), each resource we read is checksummed.
849 SHA_CTX next_sha_ctx;
850 u8 next_hash[SHA1_HASH_SIZE];
852 // Resources that don't need any chunks compressed are added to this
853 // list and written directly by the main thread.
854 LIST_HEAD(my_resources);
856 struct wim_lookup_table_entry *cur_lte = NULL;
860 ntfs_inode *ni = NULL;
863 DEBUG("Initializing buffers for uncompressed "
864 "and compressed data (%zu bytes needed)",
865 num_messages * MAX_CHUNKS_PER_MSG * WIM_CHUNK_SIZE * 2);
867 // Pre-allocate all the buffers that will be needed to do the chunk
869 for (size_t i = 0; i < num_messages; i++) {
870 for (size_t j = 0; j < MAX_CHUNKS_PER_MSG; j++) {
871 msgs[i].compressed_chunks[j] = MALLOC(WIM_CHUNK_SIZE);
873 // The extra 8 bytes is because longest_match() in
874 // lz77.c may read a little bit off the end of the
875 // uncompressed data. It doesn't need to be
876 // initialized--- we really just need to avoid accessing
878 msgs[i].uncompressed_chunks[j] = MALLOC(WIM_CHUNK_SIZE + 8);
879 if (msgs[i].compressed_chunks[j] == NULL ||
880 msgs[i].uncompressed_chunks[j] == NULL)
882 ret = WIMLIB_ERR_NOMEM;
888 // This loop is executed until all resources have been written, except
889 // possibly a few that have been added to the @my_resources list for
892 // Send chunks to the compressor threads until either (a) there
893 // are no more messages available since they were all sent off,
894 // or (b) there are no more resources that need to be
896 while (!list_empty(&available_msgs)) {
897 if (next_chunk == next_num_chunks) {
898 // If next_chunk == next_num_chunks, there are
899 // no more chunks to write in the current
900 // stream. So, check the SHA1 message digest of
901 // the stream that was just finished (unless
902 // next_lte == NULL, which is the case the very
903 // first time this loop is entered, and also
904 // near the very end of the compression when
905 // there are no more streams.) Then, advance to
906 // the next stream (if there is one).
907 if (next_lte != NULL) {
909 end_wim_resource_read(next_lte, ni);
912 end_wim_resource_read(next_lte);
914 DEBUG2("Finalize SHA1 md (next_num_chunks=%zu)",
916 sha1_final(next_hash, &next_sha_ctx);
917 if (!hashes_equal(next_lte->hash, next_hash)) {
918 ERROR("WIM resource has incorrect hash!");
919 if (next_lte->resource_location ==
920 RESOURCE_IN_FILE_ON_DISK)
922 ERROR("We were reading it from `%s'; "
923 "maybe it changed while we were "
925 next_lte->file_on_disk);
927 ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
932 // Advance to the next resource.
934 // If the next resource needs no compression, just write
935 // it with this thread (not now though--- we could be in
936 // the middle of writing another resource.) Keep doing
937 // this until we either get to the end of the resources
938 // list, or we get to a resource that needs compression.
940 if (next_resource == stream_list) {
941 // No more resources to send for
946 next_lte = container_of(next_resource,
947 struct wim_lookup_table_entry,
949 next_resource = next_resource->next;
950 if ((!(write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
951 && wim_resource_compression_type(next_lte) == out_ctype)
952 || wim_resource_size(next_lte) == 0)
954 list_add_tail(&next_lte->staging_list,
957 list_add_tail(&next_lte->staging_list,
958 &outstanding_resources);
960 next_num_chunks = wim_resource_chunks(next_lte);
961 sha1_init(&next_sha_ctx);
962 INIT_LIST_HEAD(&next_lte->msg_list);
964 ret = prepare_resource_for_read(next_lte, &ni);
966 ret = prepare_resource_for_read(next_lte);
971 if (cur_lte == NULL) {
972 // Set cur_lte for the
981 if (next_lte == NULL) {
982 // No more resources to send for compression
986 // Get a message from the available messages
988 msg = container_of(available_msgs.next,
992 // ... and delete it from the available messages
994 list_del(&msg->list);
996 // Initialize the message with the chunks to
998 msg->num_chunks = min(next_num_chunks - next_chunk,
1000 msg->lte = next_lte;
1001 msg->complete = false;
1002 msg->begin_chunk = next_chunk;
1004 unsigned size = WIM_CHUNK_SIZE;
1005 for (unsigned i = 0; i < msg->num_chunks; i++) {
1007 // Read chunk @next_chunk of the stream into the
1008 // message so that a compressor thread can
1011 if (next_chunk == next_num_chunks - 1) {
1012 size = MODULO_NONZERO(wim_resource_size(next_lte),
1016 DEBUG2("Read resource (size=%u, offset=%zu)",
1017 size, next_chunk * WIM_CHUNK_SIZE);
1019 msg->uncompressed_chunk_sizes[i] = size;
1021 ret = read_wim_resource(next_lte,
1022 msg->uncompressed_chunks[i],
1024 next_chunk * WIM_CHUNK_SIZE,
1028 sha1_update(&next_sha_ctx,
1029 msg->uncompressed_chunks[i], size);
1033 // Send the compression request
1034 list_add_tail(&msg->list, &next_lte->msg_list);
1035 shared_queue_put(res_to_compress_queue, msg);
1036 DEBUG2("Compression request sent");
1039 // If there are no outstanding resources, there are no more
1040 // resources that need to be written.
1041 if (list_empty(&outstanding_resources)) {
1046 // Get the next message from the queue and process it.
1047 // The message will contain 1 or more data chunks that have been
1049 msg = shared_queue_get(compressed_res_queue);
1050 msg->complete = true;
1052 // Is this the next chunk in the current resource? If it's not
1053 // (i.e., an earlier chunk in a same or different resource
1054 // hasn't been compressed yet), do nothing, and keep this
1055 // message around until all earlier chunks are received.
1057 // Otherwise, write all the chunks we can.
1058 while (cur_lte != NULL &&
1059 !list_empty(&cur_lte->msg_list) &&
1060 (msg = container_of(cur_lte->msg_list.next,
1064 DEBUG2("Complete msg (begin_chunk=%"PRIu64")", msg->begin_chunk);
1065 if (msg->begin_chunk == 0) {
1066 DEBUG2("Begin chunk tab");
1068 // This is the first set of chunks. Leave space
1069 // for the chunk table in the output file.
1070 off_t cur_offset = ftello(out_fp);
1071 if (cur_offset == -1) {
1072 ret = WIMLIB_ERR_WRITE;
1075 ret = begin_wim_resource_chunk_tab(cur_lte,
1083 // Write the compressed chunks from the message.
1084 ret = write_wim_chunks(msg, out_fp, cur_chunk_tab);
1088 list_del(&msg->list);
1090 // This message is available to use for different chunks
1092 list_add(&msg->list, &available_msgs);
1094 // Was this the last chunk of the stream? If so, finish
1096 if (list_empty(&cur_lte->msg_list) &&
1097 msg->begin_chunk + msg->num_chunks == cur_chunk_tab->num_chunks)
1099 DEBUG2("Finish wim chunk tab");
1101 ret = finish_wim_resource_chunk_tab(cur_chunk_tab,
1107 if (res_csize >= wim_resource_size(cur_lte)) {
1108 /* Oops! We compressed the resource to
1109 * larger than the original size. Write
1110 * the resource uncompressed instead. */
1111 ret = write_uncompressed_resource_and_truncate(
1114 cur_chunk_tab->file_offset,
1115 &cur_lte->output_resource_entry);
1119 cur_lte->output_resource_entry.size =
1122 cur_lte->output_resource_entry.original_size =
1123 cur_lte->resource_entry.original_size;
1125 cur_lte->output_resource_entry.offset =
1126 cur_chunk_tab->file_offset;
1128 cur_lte->output_resource_entry.flags =
1129 cur_lte->resource_entry.flags |
1130 WIM_RESHDR_FLAG_COMPRESSED;
1133 progress->write_streams.completed_bytes +=
1134 wim_resource_size(cur_lte);
1135 progress->write_streams.completed_streams++;
1137 if (progress_func) {
1138 progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
1142 FREE(cur_chunk_tab);
1143 cur_chunk_tab = NULL;
1145 struct list_head *next = cur_lte->staging_list.next;
1146 list_del(&cur_lte->staging_list);
1148 if (next == &outstanding_resources)
1151 cur_lte = container_of(cur_lte->staging_list.next,
1152 struct wim_lookup_table_entry,
1155 // Since we just finished writing a stream,
1156 // write any streams that have been added to the
1157 // my_resources list for direct writing by the
1158 // main thread (e.g. resources that don't need
1159 // to be compressed because the desired
1160 // compression type is the same as the previous
1161 // compression type).
1162 ret = do_write_stream_list(&my_resources,
1175 if (ret == WIMLIB_ERR_NOMEM) {
1176 ERROR("Could not allocate enough memory for "
1177 "multi-threaded compression");
1182 end_wim_resource_read(next_lte, ni);
1184 end_wim_resource_read(next_lte);
1189 ret = do_write_stream_list(&my_resources, out_fp,
1190 out_ctype, progress_func,
1194 size_t num_available_msgs = 0;
1195 struct list_head *cur;
1197 list_for_each(cur, &available_msgs) {
1198 num_available_msgs++;
1201 while (num_available_msgs < num_messages) {
1202 shared_queue_get(compressed_res_queue);
1203 num_available_msgs++;
1209 for (size_t i = 0; i < num_messages; i++) {
1210 for (size_t j = 0; j < MAX_CHUNKS_PER_MSG; j++) {
1211 FREE(msgs[i].compressed_chunks[j]);
1212 FREE(msgs[i].uncompressed_chunks[j]);
1218 FREE(cur_chunk_tab);
1222 static long get_default_num_threads()
1225 SYSTEM_INFO sysinfo;
1226 GetSystemInfo(&sysinfo);
1227 return sysinfo.dwNumberOfProcessors;
1229 return sysconf(_SC_NPROCESSORS_ONLN);
1233 static int write_stream_list_parallel(struct list_head *stream_list,
1237 unsigned num_threads,
1238 wimlib_progress_func_t progress_func,
1239 union wimlib_progress_info *progress)
1242 struct shared_queue res_to_compress_queue;
1243 struct shared_queue compressed_res_queue;
1244 pthread_t *compressor_threads = NULL;
1246 if (num_threads == 0) {
1247 long nthreads = get_default_num_threads();
1248 if (nthreads < 1 || nthreads > UINT_MAX) {
1249 WARNING("Could not determine number of processors! Assuming 1");
1252 num_threads = nthreads;
1256 progress->write_streams.num_threads = num_threads;
1257 wimlib_assert(stream_list->next != stream_list);
1259 static const double MESSAGES_PER_THREAD = 2.0;
1260 size_t queue_size = (size_t)(num_threads * MESSAGES_PER_THREAD);
1262 DEBUG("Initializing shared queues (queue_size=%zu)", queue_size);
1264 ret = shared_queue_init(&res_to_compress_queue, queue_size);
1268 ret = shared_queue_init(&compressed_res_queue, queue_size);
1270 goto out_destroy_res_to_compress_queue;
1272 struct compressor_thread_params params;
1273 params.res_to_compress_queue = &res_to_compress_queue;
1274 params.compressed_res_queue = &compressed_res_queue;
1275 params.compress = get_compress_func(out_ctype);
1277 compressor_threads = MALLOC(num_threads * sizeof(pthread_t));
1278 if (!compressor_threads) {
1279 ret = WIMLIB_ERR_NOMEM;
1280 goto out_destroy_compressed_res_queue;
1283 for (unsigned i = 0; i < num_threads; i++) {
1284 DEBUG("pthread_create thread %u", i);
1285 ret = pthread_create(&compressor_threads[i], NULL,
1286 compressor_thread_proc, ¶ms);
1289 ERROR_WITH_ERRNO("Failed to create compressor "
1297 progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress);
1299 ret = main_writer_thread_proc(stream_list,
1302 &res_to_compress_queue,
1303 &compressed_res_queue,
1309 for (unsigned i = 0; i < num_threads; i++)
1310 shared_queue_put(&res_to_compress_queue, NULL);
1312 for (unsigned i = 0; i < num_threads; i++) {
1313 if (pthread_join(compressor_threads[i], NULL)) {
1314 WARNING_WITH_ERRNO("Failed to join compressor "
1318 FREE(compressor_threads);
1319 out_destroy_compressed_res_queue:
1320 shared_queue_destroy(&compressed_res_queue);
1321 out_destroy_res_to_compress_queue:
1322 shared_queue_destroy(&res_to_compress_queue);
1323 if (ret >= 0 && ret != WIMLIB_ERR_NOMEM)
1326 WARNING("Falling back to single-threaded compression");
1327 return write_stream_list_serial(stream_list,
1338 * Write a list of streams to a WIM (@out_fp) using the compression type
1339 * @out_ctype and up to @num_threads compressor threads.
1341 static int write_stream_list(struct list_head *stream_list, FILE *out_fp,
1342 int out_ctype, int write_flags,
1343 unsigned num_threads,
1344 wimlib_progress_func_t progress_func)
1346 struct wim_lookup_table_entry *lte;
1347 size_t num_streams = 0;
1348 u64 total_bytes = 0;
1349 u64 total_compression_bytes = 0;
1350 union wimlib_progress_info progress;
1352 list_for_each_entry(lte, stream_list, staging_list) {
1354 total_bytes += wim_resource_size(lte);
1355 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE
1356 && (wim_resource_compression_type(lte) != out_ctype ||
1357 (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)))
1359 total_compression_bytes += wim_resource_size(lte);
1362 progress.write_streams.total_bytes = total_bytes;
1363 progress.write_streams.total_streams = num_streams;
1364 progress.write_streams.completed_bytes = 0;
1365 progress.write_streams.completed_streams = 0;
1366 progress.write_streams.num_threads = num_threads;
1367 progress.write_streams.compression_type = out_ctype;
1369 #ifdef ENABLE_MULTITHREADED_COMPRESSION
1370 if (total_compression_bytes >= 1000000 && num_threads != 1)
1371 return write_stream_list_parallel(stream_list,
1380 return write_stream_list_serial(stream_list,
1388 struct lte_overwrite_prepare_args {
1391 struct list_head *stream_list;
1394 static int lte_overwrite_prepare(struct wim_lookup_table_entry *lte, void *arg)
1396 struct lte_overwrite_prepare_args *args = arg;
1398 if (lte->resource_location == RESOURCE_IN_WIM &&
1399 lte->wim == args->wim &&
1400 lte->resource_entry.offset + lte->resource_entry.size > args->end_offset)
1402 #ifdef ENABLE_ERROR_MESSAGES
1403 ERROR("The following resource is after the XML data:");
1404 print_lookup_table_entry(lte, stderr);
1406 return WIMLIB_ERR_RESOURCE_ORDER;
1409 lte->out_refcnt = lte->refcnt;
1410 memcpy(<e->output_resource_entry, <e->resource_entry,
1411 sizeof(struct resource_entry));
1412 if (!(lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA)) {
1413 wimlib_assert(lte->resource_location != RESOURCE_NONEXISTENT);
1414 if (lte->resource_location != RESOURCE_IN_WIM || lte->wim != args->wim)
1415 list_add(<e->staging_list, args->stream_list);
1420 static int wim_find_new_streams(WIMStruct *wim, off_t end_offset,
1421 struct list_head *stream_list)
1423 struct lte_overwrite_prepare_args args = {
1425 .end_offset = end_offset,
1426 .stream_list = stream_list,
1429 return for_lookup_table_entry(wim->lookup_table,
1430 lte_overwrite_prepare, &args);
1433 static int inode_find_streams_to_write(struct wim_inode *inode,
1434 struct wim_lookup_table *table,
1435 struct list_head *stream_list)
1437 struct wim_lookup_table_entry *lte;
1438 for (unsigned i = 0; i <= inode->i_num_ads; i++) {
1439 lte = inode_stream_lte(inode, i, table);
1441 if (lte->out_refcnt == 0)
1442 list_add_tail(<e->staging_list, stream_list);
1443 lte->out_refcnt += inode->i_nlink;
1449 static int image_find_streams_to_write(WIMStruct *w)
1451 struct wim_inode *inode;
1452 struct hlist_node *cur;
1453 struct hlist_head *inode_list;
1455 inode_list = &wim_get_current_image_metadata(w)->inode_list;
1456 hlist_for_each_entry(inode, cur, inode_list, i_hlist) {
1457 inode_find_streams_to_write(inode, w->lookup_table,
1458 (struct list_head*)w->private);
1463 static int write_wim_streams(WIMStruct *w, int image, int write_flags,
1464 unsigned num_threads,
1465 wimlib_progress_func_t progress_func)
1468 for_lookup_table_entry(w->lookup_table, lte_zero_out_refcnt, NULL);
1469 LIST_HEAD(stream_list);
1470 w->private = &stream_list;
1471 for_image(w, image, image_find_streams_to_write);
1472 return write_stream_list(&stream_list, w->out_fp,
1473 wimlib_get_compression_type(w), write_flags,
1474 num_threads, progress_func);
1478 * Finish writing a WIM file: write the lookup table, xml data, and integrity
1479 * table (optional), then overwrite the WIM header.
1481 * write_flags is a bitwise OR of the following:
1483 * (public) WIMLIB_WRITE_FLAG_CHECK_INTEGRITY:
1484 * Include an integrity table.
1486 * (public) WIMLIB_WRITE_FLAG_SHOW_PROGRESS:
1487 * Show progress information when (if) writing the integrity table.
1489 * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE:
1490 * Don't write the lookup table.
1492 * (private) WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE:
1493 * When (if) writing the integrity table, re-use entries from the
1494 * existing integrity table, if possible.
1496 * (private) WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML:
1497 * After writing the XML data but before writing the integrity
1498 * table, write a temporary WIM header and flush the stream so that
1499 * the WIM is less likely to become corrupted upon abrupt program
1502 * (private) WIMLIB_WRITE_FLAG_FSYNC:
1503 * fsync() the output file before closing it.
1506 int finish_write(WIMStruct *w, int image, int write_flags,
1507 wimlib_progress_func_t progress_func)
1510 struct wim_header hdr;
1511 FILE *out = w->out_fp;
1513 /* @hdr will be the header for the new WIM. First copy all the data
1514 * from the header in the WIMStruct; then set all the fields that may
1515 * have changed, including the resource entries, boot index, and image
1517 memcpy(&hdr, &w->hdr, sizeof(struct wim_header));
1519 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
1520 ret = write_lookup_table(w->lookup_table, out, &hdr.lookup_table_res_entry);
1525 ret = write_xml_data(w->wim_info, image, out,
1526 (write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE) ?
1527 wim_info_get_total_bytes(w->wim_info) : 0,
1528 &hdr.xml_res_entry);
1532 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
1533 if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) {
1534 struct wim_header checkpoint_hdr;
1535 memcpy(&checkpoint_hdr, &hdr, sizeof(struct wim_header));
1536 memset(&checkpoint_hdr.integrity, 0, sizeof(struct resource_entry));
1537 if (fseeko(out, 0, SEEK_SET) != 0) {
1538 ERROR_WITH_ERRNO("Failed to seek to beginning "
1539 "of WIM being written");
1540 ret = WIMLIB_ERR_WRITE;
1543 ret = write_header(&checkpoint_hdr, out);
1547 if (fflush(out) != 0) {
1548 ERROR_WITH_ERRNO("Can't write data to WIM");
1549 ret = WIMLIB_ERR_WRITE;
1553 if (fseeko(out, 0, SEEK_END) != 0) {
1554 ERROR_WITH_ERRNO("Failed to seek to end "
1555 "of WIM being written");
1556 ret = WIMLIB_ERR_WRITE;
1561 off_t old_lookup_table_end;
1562 off_t new_lookup_table_end;
1563 if (write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE) {
1564 old_lookup_table_end = w->hdr.lookup_table_res_entry.offset +
1565 w->hdr.lookup_table_res_entry.size;
1567 old_lookup_table_end = 0;
1569 new_lookup_table_end = hdr.lookup_table_res_entry.offset +
1570 hdr.lookup_table_res_entry.size;
1572 ret = write_integrity_table(out,
1574 new_lookup_table_end,
1575 old_lookup_table_end,
1580 memset(&hdr.integrity, 0, sizeof(struct resource_entry));
1584 * In the WIM header, there is room for the resource entry for a
1585 * metadata resource labeled as the "boot metadata". This entry should
1586 * be zeroed out if there is no bootable image (boot_idx 0). Otherwise,
1587 * it should be a copy of the resource entry for the image that is
1588 * marked as bootable. This is not well documented...
1591 /* Set image count and boot index correctly for single image writes */
1592 if (image != WIMLIB_ALL_IMAGES) {
1593 hdr.image_count = 1;
1594 if (hdr.boot_idx == image)
1600 if (hdr.boot_idx == 0) {
1601 memset(&hdr.boot_metadata_res_entry, 0,
1602 sizeof(struct resource_entry));
1604 memcpy(&hdr.boot_metadata_res_entry,
1606 hdr.boot_idx - 1].metadata_lte->output_resource_entry,
1607 sizeof(struct resource_entry));
1610 if (fseeko(out, 0, SEEK_SET) != 0) {
1611 ERROR_WITH_ERRNO("Failed to seek to beginning of WIM "
1613 ret = WIMLIB_ERR_WRITE;
1617 ret = write_header(&hdr, out);
1621 if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) {
1622 if (fflush(out) != 0
1623 || fsync(fileno(out)) != 0)
1625 ERROR_WITH_ERRNO("Error flushing data to WIM file");
1626 ret = WIMLIB_ERR_WRITE;
1630 if (fclose(out) != 0) {
1631 ERROR_WITH_ERRNO("Failed to close the WIM file");
1633 ret = WIMLIB_ERR_WRITE;
1639 #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK)
1640 int lock_wim(WIMStruct *w, FILE *fp)
1643 if (fp && !w->wim_locked) {
1644 ret = flock(fileno(fp), LOCK_EX | LOCK_NB);
1646 if (errno == EWOULDBLOCK) {
1647 ERROR("`%s' is already being modified or has been "
1648 "mounted read-write\n"
1649 " by another process!", w->filename);
1650 ret = WIMLIB_ERR_ALREADY_LOCKED;
1652 WARNING_WITH_ERRNO("Failed to lock `%s'",
1664 static int open_wim_writable(WIMStruct *w, const char *path,
1665 bool trunc, bool readable)
1676 wimlib_assert(w->out_fp == NULL);
1677 w->out_fp = fopen(path, mode);
1681 ERROR_WITH_ERRNO("Failed to open `%s' for writing", path);
1682 return WIMLIB_ERR_OPEN;
1687 void close_wim_writable(WIMStruct *w)
1690 if (fclose(w->out_fp) != 0) {
1691 WARNING_WITH_ERRNO("Failed to close output WIM");
1697 /* Open file stream and write dummy header for WIM. */
1698 int begin_write(WIMStruct *w, const char *path, int write_flags)
1701 ret = open_wim_writable(w, path, true,
1702 (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0);
1705 /* Write dummy header. It will be overwritten later. */
1706 return write_header(&w->hdr, w->out_fp);
1709 /* Writes a stand-alone WIM to a file. */
1710 WIMLIBAPI int wimlib_write(WIMStruct *w, const char *path,
1711 int image, int write_flags, unsigned num_threads,
1712 wimlib_progress_func_t progress_func)
1717 return WIMLIB_ERR_INVALID_PARAM;
1719 write_flags &= WIMLIB_WRITE_MASK_PUBLIC;
1721 if (image != WIMLIB_ALL_IMAGES &&
1722 (image < 1 || image > w->hdr.image_count))
1723 return WIMLIB_ERR_INVALID_IMAGE;
1725 if (w->hdr.total_parts != 1) {
1726 ERROR("Cannot call wimlib_write() on part of a split WIM");
1727 return WIMLIB_ERR_SPLIT_UNSUPPORTED;
1730 ret = begin_write(w, path, write_flags);
1734 ret = write_wim_streams(w, image, write_flags, num_threads,
1740 progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN, NULL);
1742 ret = for_image(w, image, write_metadata_resource);
1747 progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_END, NULL);
1749 ret = finish_write(w, image, write_flags, progress_func);
1751 close_wim_writable(w);
1752 DEBUG("wimlib_write(path=%s) = %d", path, ret);
1756 static bool any_images_modified(WIMStruct *w)
1758 for (int i = 0; i < w->hdr.image_count; i++)
1759 if (w->image_metadata[i].modified)
1765 * Overwrite a WIM, possibly appending streams to it.
1767 * A WIM looks like (or is supposed to look like) the following:
1769 * Header (212 bytes)
1770 * Streams and metadata resources (variable size)
1771 * Lookup table (variable size)
1772 * XML data (variable size)
1773 * Integrity table (optional) (variable size)
1775 * If we are not adding any streams or metadata resources, the lookup table is
1776 * unchanged--- so we only need to overwrite the XML data, integrity table, and
1777 * header. This operation is potentially unsafe if the program is abruptly
1778 * terminated while the XML data or integrity table are being overwritten, but
1779 * before the new header has been written. To partially alleviate this problem,
1780 * a special flag (WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) is passed to
1781 * finish_write() to cause a temporary WIM header to be written after the XML
1782 * data has been written. This may prevent the WIM from becoming corrupted if
1783 * the program is terminated while the integrity table is being calculated (but
1784 * no guarantees, due to write re-ordering...).
1786 * If we are adding new streams or images (metadata resources), the lookup table
1787 * needs to be changed, and those streams need to be written. In this case, we
1788 * try to perform a safe update of the WIM file by writing the streams *after*
1789 * the end of the previous WIM, then writing the new lookup table, XML data, and
1790 * (optionally) integrity table following the new streams. This will produce a
1791 * layout like the following:
1793 * Header (212 bytes)
1794 * (OLD) Streams and metadata resources (variable size)
1795 * (OLD) Lookup table (variable size)
1796 * (OLD) XML data (variable size)
1797 * (OLD) Integrity table (optional) (variable size)
1798 * (NEW) Streams and metadata resources (variable size)
1799 * (NEW) Lookup table (variable size)
1800 * (NEW) XML data (variable size)
1801 * (NEW) Integrity table (optional) (variable size)
1803 * At all points, the WIM is valid as nothing points to the new data yet. Then,
1804 * the header is overwritten to point to the new lookup table, XML data, and
1805 * integrity table, to produce the following layout:
1807 * Header (212 bytes)
1808 * Streams and metadata resources (variable size)
1809 * Nothing (variable size)
1810 * More Streams and metadata resources (variable size)
1811 * Lookup table (variable size)
1812 * XML data (variable size)
1813 * Integrity table (optional) (variable size)
1815 * This method allows an image to be appended to a large WIM very quickly, and
1816 * is is crash-safe except in the case of write re-ordering, but the
1817 * disadvantage is that a small hole is left in the WIM where the old lookup
1818 * table, xml data, and integrity table were. (These usually only take up a
1819 * small amount of space compared to the streams, however.)
1821 static int overwrite_wim_inplace(WIMStruct *w, int write_flags,
1822 unsigned num_threads,
1823 wimlib_progress_func_t progress_func)
1826 struct list_head stream_list;
1828 bool found_modified_image;
1830 DEBUG("Overwriting `%s' in-place", w->filename);
1832 /* Make sure that the integrity table (if present) is after the XML
1833 * data, and that there are no stream resources, metadata resources, or
1834 * lookup tables after the XML data. Otherwise, these data would be
1836 if (w->hdr.integrity.offset != 0 &&
1837 w->hdr.integrity.offset < w->hdr.xml_res_entry.offset) {
1838 ERROR("Didn't expect the integrity table to be before the XML data");
1839 return WIMLIB_ERR_RESOURCE_ORDER;
1842 if (w->hdr.lookup_table_res_entry.offset > w->hdr.xml_res_entry.offset) {
1843 ERROR("Didn't expect the lookup table to be after the XML data");
1844 return WIMLIB_ERR_RESOURCE_ORDER;
1848 if (w->hdr.integrity.offset)
1849 old_wim_end = w->hdr.integrity.offset + w->hdr.integrity.size;
1851 old_wim_end = w->hdr.xml_res_entry.offset + w->hdr.xml_res_entry.size;
1853 if (!w->deletion_occurred && !any_images_modified(w)) {
1854 /* If no images have been modified and no images have been
1855 * deleted, a new lookup table does not need to be written. */
1856 old_wim_end = w->hdr.lookup_table_res_entry.offset +
1857 w->hdr.lookup_table_res_entry.size;
1858 write_flags |= WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE |
1859 WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML;
1861 INIT_LIST_HEAD(&stream_list);
1862 ret = wim_find_new_streams(w, old_wim_end, &stream_list);
1866 ret = open_wim_writable(w, w->filename, false,
1867 (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0);
1871 ret = lock_wim(w, w->out_fp);
1878 if (fseeko(w->out_fp, old_wim_end, SEEK_SET) != 0) {
1879 ERROR_WITH_ERRNO("Can't seek to end of WIM");
1883 return WIMLIB_ERR_WRITE;
1886 if (!list_empty(&stream_list)) {
1887 DEBUG("Writing newly added streams (offset = %"PRIu64")",
1889 ret = write_stream_list(&stream_list, w->out_fp,
1890 wimlib_get_compression_type(w),
1891 write_flags, num_threads,
1896 DEBUG("No new streams were added");
1899 found_modified_image = false;
1900 for (int i = 0; i < w->hdr.image_count; i++) {
1901 if (!found_modified_image)
1902 found_modified_image = w->image_metadata[i].modified;
1903 if (found_modified_image) {
1904 select_wim_image(w, i + 1);
1905 ret = write_metadata_resource(w);
1910 write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE;
1911 ret = finish_write(w, WIMLIB_ALL_IMAGES, write_flags,
1914 close_wim_writable(w);
1915 if (ret != 0 && !(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
1916 WARNING("Truncating `%s' to its original size (%"PRIu64" bytes)",
1917 w->filename, old_wim_end);
1918 /* Return value of truncate() is ignored because this is already
1920 (void)truncate(w->filename, old_wim_end);
1926 static int overwrite_wim_via_tmpfile(WIMStruct *w, int write_flags,
1927 unsigned num_threads,
1928 wimlib_progress_func_t progress_func)
1930 size_t wim_name_len;
1933 DEBUG("Overwriting `%s' via a temporary file", w->filename);
1935 /* Write the WIM to a temporary file in the same directory as the
1937 wim_name_len = strlen(w->filename);
1938 char tmpfile[wim_name_len + 10];
1939 memcpy(tmpfile, w->filename, wim_name_len);
1940 randomize_char_array_with_alnum(tmpfile + wim_name_len, 9);
1941 tmpfile[wim_name_len + 9] = '\0';
1943 ret = wimlib_write(w, tmpfile, WIMLIB_ALL_IMAGES,
1944 write_flags | WIMLIB_WRITE_FLAG_FSYNC,
1945 num_threads, progress_func);
1947 ERROR("Failed to write the WIM file `%s'", tmpfile);
1951 DEBUG("Renaming `%s' to `%s'", tmpfile, w->filename);
1953 /* Rename the new file to the old file .*/
1954 if (rename(tmpfile, w->filename) != 0) {
1955 ERROR_WITH_ERRNO("Failed to rename `%s' to `%s'",
1956 tmpfile, w->filename);
1957 ret = WIMLIB_ERR_RENAME;
1961 if (progress_func) {
1962 union wimlib_progress_info progress;
1963 progress.rename.from = tmpfile;
1964 progress.rename.to = w->filename;
1965 progress_func(WIMLIB_PROGRESS_MSG_RENAME, &progress);
1968 /* Close the original WIM file that was opened for reading. */
1969 if (w->fp != NULL) {
1974 /* Re-open the WIM read-only. */
1975 w->fp = fopen(w->filename, "rb");
1976 if (w->fp == NULL) {
1977 ret = WIMLIB_ERR_REOPEN;
1978 WARNING_WITH_ERRNO("Failed to re-open `%s' read-only",
1985 /* Remove temporary file. */
1986 if (unlink(tmpfile) != 0)
1987 WARNING_WITH_ERRNO("Failed to remove `%s'", tmpfile);
1992 * Writes a WIM file to the original file that it was read from, overwriting it.
1994 WIMLIBAPI int wimlib_overwrite(WIMStruct *w, int write_flags,
1995 unsigned num_threads,
1996 wimlib_progress_func_t progress_func)
1998 write_flags &= WIMLIB_WRITE_MASK_PUBLIC;
2001 return WIMLIB_ERR_NO_FILENAME;
2003 if (w->hdr.total_parts != 1) {
2004 ERROR("Cannot modify a split WIM");
2005 return WIMLIB_ERR_SPLIT_UNSUPPORTED;
2008 if ((!w->deletion_occurred || (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE))
2009 && !(write_flags & WIMLIB_WRITE_FLAG_REBUILD))
2012 ret = overwrite_wim_inplace(w, write_flags, num_threads,
2014 if (ret == WIMLIB_ERR_RESOURCE_ORDER)
2015 WARNING("Falling back to re-building entire WIM");
2019 return overwrite_wim_via_tmpfile(w, write_flags, num_threads,