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. */
36 #include "wimlib_internal.h"
37 #include "buffer_io.h"
39 #include "lookup_table.h"
44 #ifdef ENABLE_MULTITHREADED_COMPRESSION
53 #include <ntfs-3g/attrib.h>
54 #include <ntfs-3g/inode.h>
55 #include <ntfs-3g/dir.h>
64 static int fflush_and_ftruncate(FILE *fp, off_t size)
70 ERROR_WITH_ERRNO("Failed to flush data to output WIM file");
71 return WIMLIB_ERR_WRITE;
73 ret = ftruncate(fileno(fp), size);
75 ERROR_WITH_ERRNO("Failed to truncate output WIM file to "
76 "%"PRIu64" bytes", size);
77 return WIMLIB_ERR_WRITE;
82 /* Chunk table that's located at the beginning of each compressed resource in
83 * the WIM. (This is not the on-disk format; the on-disk format just has an
84 * array of offsets.) */
88 u64 original_resource_size;
89 u64 bytes_per_chunk_entry;
97 * Allocates and initializes a chunk table, and reserves space for it in the
101 begin_wim_resource_chunk_tab(const struct wim_lookup_table_entry *lte,
104 struct chunk_table **chunk_tab_ret)
106 u64 size = wim_resource_size(lte);
107 u64 num_chunks = (size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE;
108 size_t alloc_size = sizeof(struct chunk_table) + num_chunks * sizeof(u64);
109 struct chunk_table *chunk_tab = CALLOC(1, alloc_size);
113 ERROR("Failed to allocate chunk table for %"PRIu64" byte "
115 ret = WIMLIB_ERR_NOMEM;
118 chunk_tab->file_offset = file_offset;
119 chunk_tab->num_chunks = num_chunks;
120 chunk_tab->original_resource_size = size;
121 chunk_tab->bytes_per_chunk_entry = (size >= (1ULL << 32)) ? 8 : 4;
122 chunk_tab->table_disk_size = chunk_tab->bytes_per_chunk_entry *
124 chunk_tab->cur_offset = 0;
125 chunk_tab->cur_offset_p = chunk_tab->offsets;
127 if (fwrite(chunk_tab, 1, chunk_tab->table_disk_size, out_fp) !=
128 chunk_tab->table_disk_size) {
129 ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
131 ret = WIMLIB_ERR_WRITE;
137 *chunk_tab_ret = chunk_tab;
142 * Pointer to function to compresses a chunk of a WIM resource.
144 * @chunk: Uncompressed data of the chunk.
145 * @chunk_size: Size of the uncompressed chunk in bytes.
146 * @compressed_chunk: Pointer to output buffer of size at least
147 * (@chunk_size - 1) bytes.
148 * @compressed_chunk_len_ret: Pointer to an unsigned int into which the size
149 * of the compressed chunk will be
152 * Returns zero if compressed succeeded, and nonzero if the chunk could not be
153 * compressed to any smaller than @chunk_size. This function cannot fail for
156 typedef int (*compress_func_t)(const void *, unsigned, void *, unsigned *);
158 compress_func_t get_compress_func(int out_ctype)
160 if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX)
163 return xpress_compress;
167 * Writes a chunk of a WIM resource to an output file.
169 * @chunk: Uncompressed data of the chunk.
170 * @chunk_size: Size of the chunk (<= WIM_CHUNK_SIZE)
171 * @out_fp: FILE * to write tho chunk to.
172 * @out_ctype: Compression type to use when writing the chunk (ignored if no
173 * chunk table provided)
174 * @chunk_tab: Pointer to chunk table being created. It is updated with the
175 * offset of the chunk we write.
177 * Returns 0 on success; nonzero on failure.
179 static int write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size,
180 FILE *out_fp, compress_func_t compress,
181 struct chunk_table *chunk_tab)
184 unsigned out_chunk_size;
186 u8 *compressed_chunk = alloca(chunk_size);
189 ret = compress(chunk, chunk_size, compressed_chunk,
192 out_chunk = compressed_chunk;
195 out_chunk_size = chunk_size;
197 *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset;
198 chunk_tab->cur_offset += out_chunk_size;
201 out_chunk_size = chunk_size;
203 if (fwrite(out_chunk, 1, out_chunk_size, out_fp) != out_chunk_size) {
204 ERROR_WITH_ERRNO("Failed to write WIM resource chunk");
205 return WIMLIB_ERR_WRITE;
211 * Finishes a WIM chunk table and writes it to the output file at the correct
214 * The final size of the full compressed resource is returned in the
215 * @compressed_size_p.
218 finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab,
219 FILE *out_fp, u64 *compressed_size_p)
221 size_t bytes_written;
222 if (fseeko(out_fp, chunk_tab->file_offset, SEEK_SET) != 0) {
223 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of output "
224 "WIM file", chunk_tab->file_offset);
225 return WIMLIB_ERR_WRITE;
228 if (chunk_tab->bytes_per_chunk_entry == 8) {
229 array_cpu_to_le64(chunk_tab->offsets, chunk_tab->num_chunks);
231 for (u64 i = 0; i < chunk_tab->num_chunks; i++)
232 ((u32*)chunk_tab->offsets)[i] =
233 cpu_to_le32(chunk_tab->offsets[i]);
235 bytes_written = fwrite((u8*)chunk_tab->offsets +
236 chunk_tab->bytes_per_chunk_entry,
237 1, chunk_tab->table_disk_size, out_fp);
238 if (bytes_written != chunk_tab->table_disk_size) {
239 ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
241 return WIMLIB_ERR_WRITE;
243 if (fseeko(out_fp, 0, SEEK_END) != 0) {
244 ERROR_WITH_ERRNO("Failed to seek to end of output WIM file");
245 return WIMLIB_ERR_WRITE;
247 *compressed_size_p = chunk_tab->cur_offset + chunk_tab->table_disk_size;
251 /* Prepare for multiple reads to a resource by caching a FILE * or NTFS
252 * attribute pointer in the lookup table entry. */
253 static int prepare_resource_for_read(struct wim_lookup_table_entry *lte
256 , ntfs_inode **ni_ret
260 switch (lte->resource_location) {
261 case RESOURCE_IN_FILE_ON_DISK:
262 if (!lte->file_on_disk_fp) {
263 lte->file_on_disk_fp = fopen(lte->file_on_disk, "rb");
264 if (!lte->file_on_disk_fp) {
265 ERROR_WITH_ERRNO("Failed to open the file "
266 "`%s'", lte->file_on_disk);
267 return WIMLIB_ERR_OPEN;
272 case RESOURCE_IN_NTFS_VOLUME:
274 struct ntfs_location *loc = lte->ntfs_loc;
277 ni = ntfs_pathname_to_inode(*loc->ntfs_vol_p, NULL, loc->path_utf8);
279 ERROR_WITH_ERRNO("Failed to open inode `%s' in NTFS "
280 "volume", loc->path_utf8);
281 return WIMLIB_ERR_NTFS_3G;
283 lte->attr = ntfs_attr_open(ni,
284 loc->is_reparse_point ? AT_REPARSE_POINT : AT_DATA,
285 (ntfschar*)loc->stream_name_utf16,
286 loc->stream_name_utf16_num_chars);
288 ERROR_WITH_ERRNO("Failed to open attribute of `%s' in "
289 "NTFS volume", loc->path_utf8);
290 ntfs_inode_close(ni);
291 return WIMLIB_ERR_NTFS_3G;
297 #if defined(__CYGWIN__) || defined(__WIN32__)
299 if (!lte->file_on_disk_fp) {
300 lte->file_on_disk_fp = win32_open_file_readonly(lte->file_on_disk);
301 if (!lte->file_on_disk_fp)
302 return WIMLIB_ERR_OPEN;
312 /* Undo prepare_resource_for_read() by closing the cached FILE * or NTFS
314 static void end_wim_resource_read(struct wim_lookup_table_entry *lte
320 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK
321 && lte->file_on_disk_fp)
323 fclose(lte->file_on_disk_fp);
324 lte->file_on_disk_fp = NULL;
327 else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME) {
329 ntfs_attr_close(lte->attr);
333 ntfs_inode_close(ni);
336 #if defined(__CYGWIN__) || defined(__WIN32__)
337 else if (lte->resource_location == RESOURCE_WIN32
338 && lte->file_on_disk_fp)
340 win32_close_file(lte->file_on_disk_fp);
341 lte->file_on_disk_fp = NULL;
347 write_uncompressed_resource_and_truncate(struct wim_lookup_table_entry *lte,
350 struct resource_entry *out_res_entry)
353 if (fseeko(out_fp, file_offset, SEEK_SET) != 0) {
354 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of "
355 "output WIM file", file_offset);
356 return WIMLIB_ERR_WRITE;
358 ret = write_wim_resource(lte, out_fp, WIMLIB_COMPRESSION_TYPE_NONE,
363 return fflush_and_ftruncate(out_fp,
364 file_offset + wim_resource_size(lte));
368 * Writes a WIM resource to a FILE * opened for writing. The resource may be
369 * written uncompressed or compressed depending on the @out_ctype parameter.
371 * If by chance the resource compresses to more than the original size (this may
372 * happen with random data or files than are pre-compressed), the resource is
373 * instead written uncompressed (and this is reflected in the @out_res_entry by
374 * removing the WIM_RESHDR_FLAG_COMPRESSED flag).
376 * @lte: The lookup table entry for the WIM resource.
377 * @out_fp: The FILE * to write the resource to.
378 * @out_ctype: The compression type of the resource to write. Note: if this is
379 * the same as the compression type of the WIM resource we
380 * need to read, we simply copy the data (i.e. we do not
381 * uncompress it, then compress it again).
382 * @out_res_entry: If non-NULL, a resource entry that is filled in with the
383 * offset, original size, compressed size, and compression flag
384 * of the output resource.
386 * Returns 0 on success; nonzero on failure.
388 int write_wim_resource(struct wim_lookup_table_entry *lte,
389 FILE *out_fp, int out_ctype,
390 struct resource_entry *out_res_entry,
395 u64 old_compressed_size;
396 u64 new_compressed_size;
399 struct chunk_table *chunk_tab = NULL;
402 compress_func_t compress = NULL;
404 ntfs_inode *ni = NULL;
409 /* Original size of the resource */
410 original_size = wim_resource_size(lte);
412 /* Compressed size of the resource (as it exists now) */
413 old_compressed_size = wim_resource_compressed_size(lte);
415 /* Current offset in output file */
416 file_offset = ftello(out_fp);
417 if (file_offset == -1) {
418 ERROR_WITH_ERRNO("Failed to get offset in output "
420 return WIMLIB_ERR_WRITE;
423 /* Are the compression types the same? If so, do a raw copy (copy
424 * without decompressing and recompressing the data). */
425 raw = (wim_resource_compression_type(lte) == out_ctype
426 && out_ctype != WIMLIB_COMPRESSION_TYPE_NONE
427 && !(flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS));
430 flags |= WIMLIB_RESOURCE_FLAG_RAW;
431 bytes_remaining = old_compressed_size;
433 flags &= ~WIMLIB_RESOURCE_FLAG_RAW;
434 bytes_remaining = original_size;
437 /* Empty resource; nothing needs to be done, so just return success. */
438 if (bytes_remaining == 0)
441 /* Buffer for reading chunks for the resource */
442 u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
444 /* If we are writing a compressed resource and not doing a raw copy, we
445 * need to initialize the chunk table */
446 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && !raw) {
447 ret = begin_wim_resource_chunk_tab(lte, out_fp, file_offset,
453 /* If the WIM resource is in an external file, open a FILE * to it so we
454 * don't have to open a temporary one in read_wim_resource() for each
457 ret = prepare_resource_for_read(lte, &ni);
459 ret = prepare_resource_for_read(lte);
464 /* If we aren't doing a raw copy, we will compute the SHA1 message
465 * digest of the resource as we read it, and verify it's the same as the
466 * hash given in the lookup table entry once we've finished reading the
471 compress = get_compress_func(out_ctype);
475 /* While there are still bytes remaining in the WIM resource, read a
476 * chunk of the resource, update SHA1, then write that chunk using the
477 * desired compression type. */
479 u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE);
480 ret = read_wim_resource(lte, buf, to_read, offset, flags);
484 sha1_update(&ctx, buf, to_read);
485 ret = write_wim_resource_chunk(buf, to_read, out_fp,
486 compress, chunk_tab);
489 bytes_remaining -= to_read;
491 } while (bytes_remaining);
493 /* Raw copy: The new compressed size is the same as the old compressed
496 * Using WIMLIB_COMPRESSION_TYPE_NONE: The new compressed size is the
499 * Using a different compression type: Call
500 * finish_wim_resource_chunk_tab() and it will provide the new
504 new_compressed_size = old_compressed_size;
506 if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE)
507 new_compressed_size = original_size;
509 ret = finish_wim_resource_chunk_tab(chunk_tab, out_fp,
510 &new_compressed_size);
516 /* Verify SHA1 message digest of the resource, unless we are doing a raw
517 * write (in which case we never even saw the uncompressed data). Or,
518 * if the hash we had before is all 0's, just re-set it to be the new
521 u8 md[SHA1_HASH_SIZE];
522 sha1_final(md, &ctx);
523 if (is_zero_hash(lte->hash)) {
524 copy_hash(lte->hash, md);
525 } else if (!hashes_equal(md, lte->hash)) {
526 ERROR("WIM resource has incorrect hash!");
527 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK) {
528 ERROR("We were reading it from `%s'; maybe it changed "
529 "while we were reading it.",
532 ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
537 if (!raw && new_compressed_size >= original_size &&
538 out_ctype != WIMLIB_COMPRESSION_TYPE_NONE)
540 /* Oops! We compressed the resource to larger than the original
541 * size. Write the resource uncompressed instead. */
542 ret = write_uncompressed_resource_and_truncate(lte,
550 out_res_entry->size = new_compressed_size;
551 out_res_entry->original_size = original_size;
552 out_res_entry->offset = file_offset;
553 out_res_entry->flags = lte->resource_entry.flags
554 & ~WIM_RESHDR_FLAG_COMPRESSED;
555 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE)
556 out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED;
562 end_wim_resource_read(lte, ni);
564 end_wim_resource_read(lte);
571 #ifdef ENABLE_MULTITHREADED_COMPRESSION
573 /* Blocking shared queue (solves the producer-consumer problem) */
574 struct shared_queue {
578 unsigned filled_slots;
580 pthread_mutex_t lock;
581 pthread_cond_t msg_avail_cond;
582 pthread_cond_t space_avail_cond;
585 static int shared_queue_init(struct shared_queue *q, unsigned size)
587 wimlib_assert(size != 0);
588 q->array = CALLOC(sizeof(q->array[0]), size);
590 return WIMLIB_ERR_NOMEM;
595 pthread_mutex_init(&q->lock, NULL);
596 pthread_cond_init(&q->msg_avail_cond, NULL);
597 pthread_cond_init(&q->space_avail_cond, NULL);
601 static void shared_queue_destroy(struct shared_queue *q)
604 pthread_mutex_destroy(&q->lock);
605 pthread_cond_destroy(&q->msg_avail_cond);
606 pthread_cond_destroy(&q->space_avail_cond);
609 static void shared_queue_put(struct shared_queue *q, void *obj)
611 pthread_mutex_lock(&q->lock);
612 while (q->filled_slots == q->size)
613 pthread_cond_wait(&q->space_avail_cond, &q->lock);
615 q->back = (q->back + 1) % q->size;
616 q->array[q->back] = obj;
619 pthread_cond_broadcast(&q->msg_avail_cond);
620 pthread_mutex_unlock(&q->lock);
623 static void *shared_queue_get(struct shared_queue *q)
627 pthread_mutex_lock(&q->lock);
628 while (q->filled_slots == 0)
629 pthread_cond_wait(&q->msg_avail_cond, &q->lock);
631 obj = q->array[q->front];
632 q->array[q->front] = NULL;
633 q->front = (q->front + 1) % q->size;
636 pthread_cond_broadcast(&q->space_avail_cond);
637 pthread_mutex_unlock(&q->lock);
641 struct compressor_thread_params {
642 struct shared_queue *res_to_compress_queue;
643 struct shared_queue *compressed_res_queue;
644 compress_func_t compress;
647 #define MAX_CHUNKS_PER_MSG 2
650 struct wim_lookup_table_entry *lte;
651 u8 *uncompressed_chunks[MAX_CHUNKS_PER_MSG];
652 u8 *out_compressed_chunks[MAX_CHUNKS_PER_MSG];
653 u8 *compressed_chunks[MAX_CHUNKS_PER_MSG];
654 unsigned uncompressed_chunk_sizes[MAX_CHUNKS_PER_MSG];
655 unsigned compressed_chunk_sizes[MAX_CHUNKS_PER_MSG];
657 struct list_head list;
662 static void compress_chunks(struct message *msg, compress_func_t compress)
664 for (unsigned i = 0; i < msg->num_chunks; i++) {
665 DEBUG2("compress chunk %u of %u", i, msg->num_chunks);
666 int ret = compress(msg->uncompressed_chunks[i],
667 msg->uncompressed_chunk_sizes[i],
668 msg->compressed_chunks[i],
669 &msg->compressed_chunk_sizes[i]);
671 msg->out_compressed_chunks[i] = msg->compressed_chunks[i];
673 msg->out_compressed_chunks[i] = msg->uncompressed_chunks[i];
674 msg->compressed_chunk_sizes[i] = msg->uncompressed_chunk_sizes[i];
679 /* Compressor thread routine. This is a lot simpler than the main thread
680 * routine: just repeatedly get a group of chunks from the
681 * res_to_compress_queue, compress them, and put them in the
682 * compressed_res_queue. A NULL pointer indicates that the thread should stop.
684 static void *compressor_thread_proc(void *arg)
686 struct compressor_thread_params *params = arg;
687 struct shared_queue *res_to_compress_queue = params->res_to_compress_queue;
688 struct shared_queue *compressed_res_queue = params->compressed_res_queue;
689 compress_func_t compress = params->compress;
692 DEBUG("Compressor thread ready");
693 while ((msg = shared_queue_get(res_to_compress_queue)) != NULL) {
694 compress_chunks(msg, compress);
695 shared_queue_put(compressed_res_queue, msg);
697 DEBUG("Compressor thread terminating");
702 static int do_write_stream_list(struct list_head *my_resources,
705 wimlib_progress_func_t progress_func,
706 union wimlib_progress_info *progress,
707 int write_resource_flags)
710 struct wim_lookup_table_entry *lte, *tmp;
712 list_for_each_entry_safe(lte, tmp, my_resources, staging_list) {
713 ret = write_wim_resource(lte,
716 <e->output_resource_entry,
717 write_resource_flags);
720 list_del(<e->staging_list);
721 progress->write_streams.completed_bytes +=
722 wim_resource_size(lte);
723 progress->write_streams.completed_streams++;
725 progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
732 static int write_stream_list_serial(struct list_head *stream_list,
736 wimlib_progress_func_t progress_func,
737 union wimlib_progress_info *progress)
739 int write_resource_flags;
741 if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
742 write_resource_flags = WIMLIB_RESOURCE_FLAG_RECOMPRESS;
744 write_resource_flags = 0;
745 progress->write_streams.num_threads = 1;
747 progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress);
748 return do_write_stream_list(stream_list, out_fp,
749 out_ctype, progress_func,
750 progress, write_resource_flags);
753 #ifdef ENABLE_MULTITHREADED_COMPRESSION
754 static int write_wim_chunks(struct message *msg, FILE *out_fp,
755 struct chunk_table *chunk_tab)
757 for (unsigned i = 0; i < msg->num_chunks; i++) {
758 unsigned chunk_csize = msg->compressed_chunk_sizes[i];
760 DEBUG2("Write wim chunk %u of %u (csize = %u)",
761 i, msg->num_chunks, chunk_csize);
763 if (fwrite(msg->out_compressed_chunks[i], 1, chunk_csize, out_fp)
766 ERROR_WITH_ERRNO("Failed to write WIM chunk");
767 return WIMLIB_ERR_WRITE;
770 *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset;
771 chunk_tab->cur_offset += chunk_csize;
777 * This function is executed by the main thread when the resources are being
778 * compressed in parallel. The main thread is in change of all reading of the
779 * uncompressed data and writing of the compressed data. The compressor threads
780 * *only* do compression from/to in-memory buffers.
782 * Each unit of work given to a compressor thread is up to MAX_CHUNKS_PER_MSG
783 * chunks of compressed data to compress, represented in a `struct message'.
784 * Each message is passed from the main thread to a worker thread through the
785 * res_to_compress_queue, and it is passed back through the
786 * compressed_res_queue.
788 static int main_writer_thread_proc(struct list_head *stream_list,
791 struct shared_queue *res_to_compress_queue,
792 struct shared_queue *compressed_res_queue,
795 wimlib_progress_func_t progress_func,
796 union wimlib_progress_info *progress)
799 struct chunk_table *cur_chunk_tab = NULL;
800 struct message *msgs = CALLOC(num_messages, sizeof(struct message));
801 struct wim_lookup_table_entry *next_lte = NULL;
803 // Initially, all the messages are available to use.
804 LIST_HEAD(available_msgs);
807 ret = WIMLIB_ERR_NOMEM;
811 for (size_t i = 0; i < num_messages; i++)
812 list_add(&msgs[i].list, &available_msgs);
814 // outstanding_resources is the list of resources that currently have
815 // had chunks sent off for compression.
817 // The first stream in outstanding_resources is the stream that is
818 // currently being written (cur_lte).
820 // The last stream in outstanding_resources is the stream that is
821 // currently being read and chunks fed to the compressor threads
824 // Depending on the number of threads and the sizes of the resource,
825 // the outstanding streams list may contain streams between cur_lte and
826 // next_lte that have all their chunks compressed or being compressed,
827 // but haven't been written yet.
829 LIST_HEAD(outstanding_resources);
830 struct list_head *next_resource = stream_list->next;
832 u64 next_num_chunks = 0;
834 // As in write_wim_resource(), each resource we read is checksummed.
835 SHA_CTX next_sha_ctx;
836 u8 next_hash[SHA1_HASH_SIZE];
838 // Resources that don't need any chunks compressed are added to this
839 // list and written directly by the main thread.
840 LIST_HEAD(my_resources);
842 struct wim_lookup_table_entry *cur_lte = NULL;
846 ntfs_inode *ni = NULL;
849 DEBUG("Initializing buffers for uncompressed "
850 "and compressed data (%zu bytes needed)",
851 num_messages * MAX_CHUNKS_PER_MSG * WIM_CHUNK_SIZE * 2);
853 // Pre-allocate all the buffers that will be needed to do the chunk
855 for (size_t i = 0; i < num_messages; i++) {
856 for (size_t j = 0; j < MAX_CHUNKS_PER_MSG; j++) {
857 msgs[i].compressed_chunks[j] = MALLOC(WIM_CHUNK_SIZE);
859 // The extra 8 bytes is because longest_match() in
860 // lz77.c may read a little bit off the end of the
861 // uncompressed data. It doesn't need to be
862 // initialized--- we really just need to avoid accessing
864 msgs[i].uncompressed_chunks[j] = MALLOC(WIM_CHUNK_SIZE + 8);
865 if (msgs[i].compressed_chunks[j] == NULL ||
866 msgs[i].uncompressed_chunks[j] == NULL)
868 ret = WIMLIB_ERR_NOMEM;
874 // This loop is executed until all resources have been written, except
875 // possibly a few that have been added to the @my_resources list for
878 // Send chunks to the compressor threads until either (a) there
879 // are no more messages available since they were all sent off,
880 // or (b) there are no more resources that need to be
882 while (!list_empty(&available_msgs)) {
883 if (next_chunk == next_num_chunks) {
884 // If next_chunk == next_num_chunks, there are
885 // no more chunks to write in the current
886 // stream. So, check the SHA1 message digest of
887 // the stream that was just finished (unless
888 // next_lte == NULL, which is the case the very
889 // first time this loop is entered, and also
890 // near the very end of the compression when
891 // there are no more streams.) Then, advance to
892 // the next stream (if there is one).
893 if (next_lte != NULL) {
895 end_wim_resource_read(next_lte, ni);
898 end_wim_resource_read(next_lte);
900 DEBUG2("Finalize SHA1 md (next_num_chunks=%zu)",
902 sha1_final(next_hash, &next_sha_ctx);
903 if (!hashes_equal(next_lte->hash, next_hash)) {
904 ERROR("WIM resource has incorrect hash!");
905 if (next_lte->resource_location ==
906 RESOURCE_IN_FILE_ON_DISK)
908 ERROR("We were reading it from `%s'; "
909 "maybe it changed while we were "
911 next_lte->file_on_disk);
913 ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
918 // Advance to the next resource.
920 // If the next resource needs no compression, just write
921 // it with this thread (not now though--- we could be in
922 // the middle of writing another resource.) Keep doing
923 // this until we either get to the end of the resources
924 // list, or we get to a resource that needs compression.
926 if (next_resource == stream_list) {
927 // No more resources to send for
932 next_lte = container_of(next_resource,
933 struct wim_lookup_table_entry,
935 next_resource = next_resource->next;
936 if ((!(write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
937 && wim_resource_compression_type(next_lte) == out_ctype)
938 || wim_resource_size(next_lte) == 0)
940 list_add_tail(&next_lte->staging_list,
943 list_add_tail(&next_lte->staging_list,
944 &outstanding_resources);
946 next_num_chunks = wim_resource_chunks(next_lte);
947 sha1_init(&next_sha_ctx);
948 INIT_LIST_HEAD(&next_lte->msg_list);
950 ret = prepare_resource_for_read(next_lte, &ni);
952 ret = prepare_resource_for_read(next_lte);
957 if (cur_lte == NULL) {
958 // Set cur_lte for the
967 if (next_lte == NULL) {
968 // No more resources to send for compression
972 // Get a message from the available messages
974 msg = container_of(available_msgs.next,
978 // ... and delete it from the available messages
980 list_del(&msg->list);
982 // Initialize the message with the chunks to
984 msg->num_chunks = min(next_num_chunks - next_chunk,
987 msg->complete = false;
988 msg->begin_chunk = next_chunk;
990 unsigned size = WIM_CHUNK_SIZE;
991 for (unsigned i = 0; i < msg->num_chunks; i++) {
993 // Read chunk @next_chunk of the stream into the
994 // message so that a compressor thread can
997 if (next_chunk == next_num_chunks - 1) {
998 size = MODULO_NONZERO(wim_resource_size(next_lte),
1002 DEBUG2("Read resource (size=%u, offset=%zu)",
1003 size, next_chunk * WIM_CHUNK_SIZE);
1005 msg->uncompressed_chunk_sizes[i] = size;
1007 ret = read_wim_resource(next_lte,
1008 msg->uncompressed_chunks[i],
1010 next_chunk * WIM_CHUNK_SIZE,
1014 sha1_update(&next_sha_ctx,
1015 msg->uncompressed_chunks[i], size);
1019 // Send the compression request
1020 list_add_tail(&msg->list, &next_lte->msg_list);
1021 shared_queue_put(res_to_compress_queue, msg);
1022 DEBUG2("Compression request sent");
1025 // If there are no outstanding resources, there are no more
1026 // resources that need to be written.
1027 if (list_empty(&outstanding_resources)) {
1032 // Get the next message from the queue and process it.
1033 // The message will contain 1 or more data chunks that have been
1035 msg = shared_queue_get(compressed_res_queue);
1036 msg->complete = true;
1038 // Is this the next chunk in the current resource? If it's not
1039 // (i.e., an earlier chunk in a same or different resource
1040 // hasn't been compressed yet), do nothing, and keep this
1041 // message around until all earlier chunks are received.
1043 // Otherwise, write all the chunks we can.
1044 while (cur_lte != NULL &&
1045 !list_empty(&cur_lte->msg_list) &&
1046 (msg = container_of(cur_lte->msg_list.next,
1050 DEBUG2("Complete msg (begin_chunk=%"PRIu64")", msg->begin_chunk);
1051 if (msg->begin_chunk == 0) {
1052 DEBUG2("Begin chunk tab");
1054 // This is the first set of chunks. Leave space
1055 // for the chunk table in the output file.
1056 off_t cur_offset = ftello(out_fp);
1057 if (cur_offset == -1) {
1058 ret = WIMLIB_ERR_WRITE;
1061 ret = begin_wim_resource_chunk_tab(cur_lte,
1069 // Write the compressed chunks from the message.
1070 ret = write_wim_chunks(msg, out_fp, cur_chunk_tab);
1074 list_del(&msg->list);
1076 // This message is available to use for different chunks
1078 list_add(&msg->list, &available_msgs);
1080 // Was this the last chunk of the stream? If so, finish
1082 if (list_empty(&cur_lte->msg_list) &&
1083 msg->begin_chunk + msg->num_chunks == cur_chunk_tab->num_chunks)
1085 DEBUG2("Finish wim chunk tab");
1087 ret = finish_wim_resource_chunk_tab(cur_chunk_tab,
1093 if (res_csize >= wim_resource_size(cur_lte)) {
1094 /* Oops! We compressed the resource to
1095 * larger than the original size. Write
1096 * the resource uncompressed instead. */
1097 ret = write_uncompressed_resource_and_truncate(
1100 cur_chunk_tab->file_offset,
1101 &cur_lte->output_resource_entry);
1105 cur_lte->output_resource_entry.size =
1108 cur_lte->output_resource_entry.original_size =
1109 cur_lte->resource_entry.original_size;
1111 cur_lte->output_resource_entry.offset =
1112 cur_chunk_tab->file_offset;
1114 cur_lte->output_resource_entry.flags =
1115 cur_lte->resource_entry.flags |
1116 WIM_RESHDR_FLAG_COMPRESSED;
1119 progress->write_streams.completed_bytes +=
1120 wim_resource_size(cur_lte);
1121 progress->write_streams.completed_streams++;
1123 if (progress_func) {
1124 progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
1128 FREE(cur_chunk_tab);
1129 cur_chunk_tab = NULL;
1131 struct list_head *next = cur_lte->staging_list.next;
1132 list_del(&cur_lte->staging_list);
1134 if (next == &outstanding_resources)
1137 cur_lte = container_of(cur_lte->staging_list.next,
1138 struct wim_lookup_table_entry,
1141 // Since we just finished writing a stream,
1142 // write any streams that have been added to the
1143 // my_resources list for direct writing by the
1144 // main thread (e.g. resources that don't need
1145 // to be compressed because the desired
1146 // compression type is the same as the previous
1147 // compression type).
1148 ret = do_write_stream_list(&my_resources,
1161 if (ret == WIMLIB_ERR_NOMEM) {
1162 ERROR("Could not allocate enough memory for "
1163 "multi-threaded compression");
1168 end_wim_resource_read(next_lte, ni);
1170 end_wim_resource_read(next_lte);
1175 ret = do_write_stream_list(&my_resources, out_fp,
1176 out_ctype, progress_func,
1180 size_t num_available_msgs = 0;
1181 struct list_head *cur;
1183 list_for_each(cur, &available_msgs) {
1184 num_available_msgs++;
1187 while (num_available_msgs < num_messages) {
1188 shared_queue_get(compressed_res_queue);
1189 num_available_msgs++;
1195 for (size_t i = 0; i < num_messages; i++) {
1196 for (size_t j = 0; j < MAX_CHUNKS_PER_MSG; j++) {
1197 FREE(msgs[i].compressed_chunks[j]);
1198 FREE(msgs[i].uncompressed_chunks[j]);
1204 FREE(cur_chunk_tab);
1209 static int write_stream_list_parallel(struct list_head *stream_list,
1213 unsigned num_threads,
1214 wimlib_progress_func_t progress_func,
1215 union wimlib_progress_info *progress)
1218 struct shared_queue res_to_compress_queue;
1219 struct shared_queue compressed_res_queue;
1220 pthread_t *compressor_threads = NULL;
1222 if (num_threads == 0) {
1223 long nthreads = sysconf(_SC_NPROCESSORS_ONLN);
1225 WARNING("Could not determine number of processors! Assuming 1");
1228 num_threads = nthreads;
1232 progress->write_streams.num_threads = num_threads;
1233 wimlib_assert(stream_list->next != stream_list);
1235 static const double MESSAGES_PER_THREAD = 2.0;
1236 size_t queue_size = (size_t)(num_threads * MESSAGES_PER_THREAD);
1238 DEBUG("Initializing shared queues (queue_size=%zu)", queue_size);
1240 ret = shared_queue_init(&res_to_compress_queue, queue_size);
1244 ret = shared_queue_init(&compressed_res_queue, queue_size);
1246 goto out_destroy_res_to_compress_queue;
1248 struct compressor_thread_params params;
1249 params.res_to_compress_queue = &res_to_compress_queue;
1250 params.compressed_res_queue = &compressed_res_queue;
1251 params.compress = get_compress_func(out_ctype);
1253 compressor_threads = MALLOC(num_threads * sizeof(pthread_t));
1254 if (!compressor_threads) {
1255 ret = WIMLIB_ERR_NOMEM;
1256 goto out_destroy_compressed_res_queue;
1259 for (unsigned i = 0; i < num_threads; i++) {
1260 DEBUG("pthread_create thread %u", i);
1261 ret = pthread_create(&compressor_threads[i], NULL,
1262 compressor_thread_proc, ¶ms);
1265 ERROR_WITH_ERRNO("Failed to create compressor "
1273 progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress);
1275 ret = main_writer_thread_proc(stream_list,
1278 &res_to_compress_queue,
1279 &compressed_res_queue,
1285 for (unsigned i = 0; i < num_threads; i++)
1286 shared_queue_put(&res_to_compress_queue, NULL);
1288 for (unsigned i = 0; i < num_threads; i++) {
1289 if (pthread_join(compressor_threads[i], NULL)) {
1290 WARNING_WITH_ERRNO("Failed to join compressor "
1294 FREE(compressor_threads);
1295 out_destroy_compressed_res_queue:
1296 shared_queue_destroy(&compressed_res_queue);
1297 out_destroy_res_to_compress_queue:
1298 shared_queue_destroy(&res_to_compress_queue);
1299 if (ret >= 0 && ret != WIMLIB_ERR_NOMEM)
1302 WARNING("Falling back to single-threaded compression");
1303 return write_stream_list_serial(stream_list,
1314 * Write a list of streams to a WIM (@out_fp) using the compression type
1315 * @out_ctype and up to @num_threads compressor threads.
1317 static int write_stream_list(struct list_head *stream_list, FILE *out_fp,
1318 int out_ctype, int write_flags,
1319 unsigned num_threads,
1320 wimlib_progress_func_t progress_func)
1322 struct wim_lookup_table_entry *lte;
1323 size_t num_streams = 0;
1324 u64 total_bytes = 0;
1325 u64 total_compression_bytes = 0;
1326 union wimlib_progress_info progress;
1328 list_for_each_entry(lte, stream_list, staging_list) {
1330 total_bytes += wim_resource_size(lte);
1331 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE
1332 && (wim_resource_compression_type(lte) != out_ctype ||
1333 (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)))
1335 total_compression_bytes += wim_resource_size(lte);
1338 progress.write_streams.total_bytes = total_bytes;
1339 progress.write_streams.total_streams = num_streams;
1340 progress.write_streams.completed_bytes = 0;
1341 progress.write_streams.completed_streams = 0;
1342 progress.write_streams.num_threads = num_threads;
1343 progress.write_streams.compression_type = out_ctype;
1345 #ifdef ENABLE_MULTITHREADED_COMPRESSION
1346 if (total_compression_bytes >= 1000000 && num_threads != 1)
1347 return write_stream_list_parallel(stream_list,
1356 return write_stream_list_serial(stream_list,
1364 struct lte_overwrite_prepare_args {
1367 struct list_head *stream_list;
1370 static int lte_overwrite_prepare(struct wim_lookup_table_entry *lte, void *arg)
1372 struct lte_overwrite_prepare_args *args = arg;
1374 if (lte->resource_location == RESOURCE_IN_WIM &&
1375 lte->wim == args->wim &&
1376 lte->resource_entry.offset + lte->resource_entry.size > args->end_offset)
1378 #ifdef ENABLE_ERROR_MESSAGES
1379 ERROR("The following resource is after the XML data:");
1380 print_lookup_table_entry(lte, stderr);
1382 return WIMLIB_ERR_RESOURCE_ORDER;
1385 lte->out_refcnt = lte->refcnt;
1386 memcpy(<e->output_resource_entry, <e->resource_entry,
1387 sizeof(struct resource_entry));
1388 if (!(lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA)) {
1389 wimlib_assert(lte->resource_location != RESOURCE_NONEXISTENT);
1390 if (lte->resource_location != RESOURCE_IN_WIM || lte->wim != args->wim)
1391 list_add(<e->staging_list, args->stream_list);
1396 static int wim_find_new_streams(WIMStruct *wim, off_t end_offset,
1397 struct list_head *stream_list)
1399 struct lte_overwrite_prepare_args args = {
1401 .end_offset = end_offset,
1402 .stream_list = stream_list,
1405 return for_lookup_table_entry(wim->lookup_table,
1406 lte_overwrite_prepare, &args);
1409 static int inode_find_streams_to_write(struct wim_inode *inode,
1410 struct wim_lookup_table *table,
1411 struct list_head *stream_list)
1413 struct wim_lookup_table_entry *lte;
1414 for (unsigned i = 0; i <= inode->i_num_ads; i++) {
1415 lte = inode_stream_lte(inode, i, table);
1417 if (lte->out_refcnt == 0)
1418 list_add_tail(<e->staging_list, stream_list);
1419 lte->out_refcnt += inode->i_nlink;
1425 static int image_find_streams_to_write(WIMStruct *w)
1427 struct wim_inode *inode;
1428 struct hlist_node *cur;
1429 struct hlist_head *inode_list;
1431 inode_list = &wim_get_current_image_metadata(w)->inode_list;
1432 hlist_for_each_entry(inode, cur, inode_list, i_hlist) {
1433 inode_find_streams_to_write(inode, w->lookup_table,
1434 (struct list_head*)w->private);
1439 static int write_wim_streams(WIMStruct *w, int image, int write_flags,
1440 unsigned num_threads,
1441 wimlib_progress_func_t progress_func)
1444 for_lookup_table_entry(w->lookup_table, lte_zero_out_refcnt, NULL);
1445 LIST_HEAD(stream_list);
1446 w->private = &stream_list;
1447 for_image(w, image, image_find_streams_to_write);
1448 return write_stream_list(&stream_list, w->out_fp,
1449 wimlib_get_compression_type(w), write_flags,
1450 num_threads, progress_func);
1454 * Finish writing a WIM file: write the lookup table, xml data, and integrity
1455 * table (optional), then overwrite the WIM header.
1457 * write_flags is a bitwise OR of the following:
1459 * (public) WIMLIB_WRITE_FLAG_CHECK_INTEGRITY:
1460 * Include an integrity table.
1462 * (public) WIMLIB_WRITE_FLAG_SHOW_PROGRESS:
1463 * Show progress information when (if) writing the integrity table.
1465 * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE:
1466 * Don't write the lookup table.
1468 * (private) WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE:
1469 * When (if) writing the integrity table, re-use entries from the
1470 * existing integrity table, if possible.
1472 * (private) WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML:
1473 * After writing the XML data but before writing the integrity
1474 * table, write a temporary WIM header and flush the stream so that
1475 * the WIM is less likely to become corrupted upon abrupt program
1478 * (private) WIMLIB_WRITE_FLAG_FSYNC:
1479 * fsync() the output file before closing it.
1482 int finish_write(WIMStruct *w, int image, int write_flags,
1483 wimlib_progress_func_t progress_func)
1486 struct wim_header hdr;
1487 FILE *out = w->out_fp;
1489 /* @hdr will be the header for the new WIM. First copy all the data
1490 * from the header in the WIMStruct; then set all the fields that may
1491 * have changed, including the resource entries, boot index, and image
1493 memcpy(&hdr, &w->hdr, sizeof(struct wim_header));
1495 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
1496 ret = write_lookup_table(w->lookup_table, out, &hdr.lookup_table_res_entry);
1501 ret = write_xml_data(w->wim_info, image, out,
1502 (write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE) ?
1503 wim_info_get_total_bytes(w->wim_info) : 0,
1504 &hdr.xml_res_entry);
1508 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
1509 if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) {
1510 struct wim_header checkpoint_hdr;
1511 memcpy(&checkpoint_hdr, &hdr, sizeof(struct wim_header));
1512 memset(&checkpoint_hdr.integrity, 0, sizeof(struct resource_entry));
1513 if (fseeko(out, 0, SEEK_SET) != 0) {
1514 ERROR_WITH_ERRNO("Failed to seek to beginning "
1515 "of WIM being written");
1516 ret = WIMLIB_ERR_WRITE;
1519 ret = write_header(&checkpoint_hdr, out);
1523 if (fflush(out) != 0) {
1524 ERROR_WITH_ERRNO("Can't write data to WIM");
1525 ret = WIMLIB_ERR_WRITE;
1529 if (fseeko(out, 0, SEEK_END) != 0) {
1530 ERROR_WITH_ERRNO("Failed to seek to end "
1531 "of WIM being written");
1532 ret = WIMLIB_ERR_WRITE;
1537 off_t old_lookup_table_end;
1538 off_t new_lookup_table_end;
1539 if (write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE) {
1540 old_lookup_table_end = w->hdr.lookup_table_res_entry.offset +
1541 w->hdr.lookup_table_res_entry.size;
1543 old_lookup_table_end = 0;
1545 new_lookup_table_end = hdr.lookup_table_res_entry.offset +
1546 hdr.lookup_table_res_entry.size;
1548 ret = write_integrity_table(out,
1550 new_lookup_table_end,
1551 old_lookup_table_end,
1556 memset(&hdr.integrity, 0, sizeof(struct resource_entry));
1560 * In the WIM header, there is room for the resource entry for a
1561 * metadata resource labeled as the "boot metadata". This entry should
1562 * be zeroed out if there is no bootable image (boot_idx 0). Otherwise,
1563 * it should be a copy of the resource entry for the image that is
1564 * marked as bootable. This is not well documented...
1567 /* Set image count and boot index correctly for single image writes */
1568 if (image != WIMLIB_ALL_IMAGES) {
1569 hdr.image_count = 1;
1570 if (hdr.boot_idx == image)
1576 if (hdr.boot_idx == 0) {
1577 memset(&hdr.boot_metadata_res_entry, 0,
1578 sizeof(struct resource_entry));
1580 memcpy(&hdr.boot_metadata_res_entry,
1582 hdr.boot_idx - 1].metadata_lte->output_resource_entry,
1583 sizeof(struct resource_entry));
1586 if (fseeko(out, 0, SEEK_SET) != 0) {
1587 ERROR_WITH_ERRNO("Failed to seek to beginning of WIM "
1589 ret = WIMLIB_ERR_WRITE;
1593 ret = write_header(&hdr, out);
1597 if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) {
1598 if (fflush(out) != 0
1599 || fsync(fileno(out)) != 0)
1601 ERROR_WITH_ERRNO("Error flushing data to WIM file");
1602 ret = WIMLIB_ERR_WRITE;
1606 if (fclose(out) != 0) {
1607 ERROR_WITH_ERRNO("Failed to close the WIM file");
1609 ret = WIMLIB_ERR_WRITE;
1615 #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK)
1616 int lock_wim(WIMStruct *w, FILE *fp)
1619 if (fp && !w->wim_locked) {
1620 ret = flock(fileno(fp), LOCK_EX | LOCK_NB);
1622 if (errno == EWOULDBLOCK) {
1623 ERROR("`%s' is already being modified or has been "
1624 "mounted read-write\n"
1625 " by another process!", w->filename);
1626 ret = WIMLIB_ERR_ALREADY_LOCKED;
1628 WARNING_WITH_ERRNO("Failed to lock `%s'",
1640 static int open_wim_writable(WIMStruct *w, const char *path,
1641 bool trunc, bool readable)
1652 wimlib_assert(w->out_fp == NULL);
1653 w->out_fp = fopen(path, mode);
1657 ERROR_WITH_ERRNO("Failed to open `%s' for writing", path);
1658 return WIMLIB_ERR_OPEN;
1663 void close_wim_writable(WIMStruct *w)
1666 if (fclose(w->out_fp) != 0) {
1667 WARNING_WITH_ERRNO("Failed to close output WIM");
1673 /* Open file stream and write dummy header for WIM. */
1674 int begin_write(WIMStruct *w, const char *path, int write_flags)
1677 ret = open_wim_writable(w, path, true,
1678 (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0);
1681 /* Write dummy header. It will be overwritten later. */
1682 return write_header(&w->hdr, w->out_fp);
1685 /* Writes a stand-alone WIM to a file. */
1686 WIMLIBAPI int wimlib_write(WIMStruct *w, const char *path,
1687 int image, int write_flags, unsigned num_threads,
1688 wimlib_progress_func_t progress_func)
1693 return WIMLIB_ERR_INVALID_PARAM;
1695 write_flags &= WIMLIB_WRITE_MASK_PUBLIC;
1697 if (image != WIMLIB_ALL_IMAGES &&
1698 (image < 1 || image > w->hdr.image_count))
1699 return WIMLIB_ERR_INVALID_IMAGE;
1701 if (w->hdr.total_parts != 1) {
1702 ERROR("Cannot call wimlib_write() on part of a split WIM");
1703 return WIMLIB_ERR_SPLIT_UNSUPPORTED;
1706 ret = begin_write(w, path, write_flags);
1710 ret = write_wim_streams(w, image, write_flags, num_threads,
1716 progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN, NULL);
1718 ret = for_image(w, image, write_metadata_resource);
1723 progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_END, NULL);
1725 ret = finish_write(w, image, write_flags, progress_func);
1727 close_wim_writable(w);
1728 DEBUG("wimlib_write(path=%s) = %d", path, ret);
1732 static bool any_images_modified(WIMStruct *w)
1734 for (int i = 0; i < w->hdr.image_count; i++)
1735 if (w->image_metadata[i].modified)
1741 * Overwrite a WIM, possibly appending streams to it.
1743 * A WIM looks like (or is supposed to look like) the following:
1745 * Header (212 bytes)
1746 * Streams and metadata resources (variable size)
1747 * Lookup table (variable size)
1748 * XML data (variable size)
1749 * Integrity table (optional) (variable size)
1751 * If we are not adding any streams or metadata resources, the lookup table is
1752 * unchanged--- so we only need to overwrite the XML data, integrity table, and
1753 * header. This operation is potentially unsafe if the program is abruptly
1754 * terminated while the XML data or integrity table are being overwritten, but
1755 * before the new header has been written. To partially alleviate this problem,
1756 * a special flag (WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) is passed to
1757 * finish_write() to cause a temporary WIM header to be written after the XML
1758 * data has been written. This may prevent the WIM from becoming corrupted if
1759 * the program is terminated while the integrity table is being calculated (but
1760 * no guarantees, due to write re-ordering...).
1762 * If we are adding new streams or images (metadata resources), the lookup table
1763 * needs to be changed, and those streams need to be written. In this case, we
1764 * try to perform a safe update of the WIM file by writing the streams *after*
1765 * the end of the previous WIM, then writing the new lookup table, XML data, and
1766 * (optionally) integrity table following the new streams. This will produce a
1767 * layout like the following:
1769 * Header (212 bytes)
1770 * (OLD) Streams and metadata resources (variable size)
1771 * (OLD) Lookup table (variable size)
1772 * (OLD) XML data (variable size)
1773 * (OLD) Integrity table (optional) (variable size)
1774 * (NEW) Streams and metadata resources (variable size)
1775 * (NEW) Lookup table (variable size)
1776 * (NEW) XML data (variable size)
1777 * (NEW) Integrity table (optional) (variable size)
1779 * At all points, the WIM is valid as nothing points to the new data yet. Then,
1780 * the header is overwritten to point to the new lookup table, XML data, and
1781 * integrity table, to produce the following layout:
1783 * Header (212 bytes)
1784 * Streams and metadata resources (variable size)
1785 * Nothing (variable size)
1786 * More Streams and metadata resources (variable size)
1787 * Lookup table (variable size)
1788 * XML data (variable size)
1789 * Integrity table (optional) (variable size)
1791 * This method allows an image to be appended to a large WIM very quickly, and
1792 * is is crash-safe except in the case of write re-ordering, but the
1793 * disadvantage is that a small hole is left in the WIM where the old lookup
1794 * table, xml data, and integrity table were. (These usually only take up a
1795 * small amount of space compared to the streams, however.)
1797 static int overwrite_wim_inplace(WIMStruct *w, int write_flags,
1798 unsigned num_threads,
1799 wimlib_progress_func_t progress_func)
1802 struct list_head stream_list;
1804 bool found_modified_image;
1806 DEBUG("Overwriting `%s' in-place", w->filename);
1808 /* Make sure that the integrity table (if present) is after the XML
1809 * data, and that there are no stream resources, metadata resources, or
1810 * lookup tables after the XML data. Otherwise, these data would be
1812 if (w->hdr.integrity.offset != 0 &&
1813 w->hdr.integrity.offset < w->hdr.xml_res_entry.offset) {
1814 ERROR("Didn't expect the integrity table to be before the XML data");
1815 return WIMLIB_ERR_RESOURCE_ORDER;
1818 if (w->hdr.lookup_table_res_entry.offset > w->hdr.xml_res_entry.offset) {
1819 ERROR("Didn't expect the lookup table to be after the XML data");
1820 return WIMLIB_ERR_RESOURCE_ORDER;
1824 if (w->hdr.integrity.offset)
1825 old_wim_end = w->hdr.integrity.offset + w->hdr.integrity.size;
1827 old_wim_end = w->hdr.xml_res_entry.offset + w->hdr.xml_res_entry.size;
1829 if (!w->deletion_occurred && !any_images_modified(w)) {
1830 /* If no images have been modified and no images have been
1831 * deleted, a new lookup table does not need to be written. */
1832 old_wim_end = w->hdr.lookup_table_res_entry.offset +
1833 w->hdr.lookup_table_res_entry.size;
1834 write_flags |= WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE |
1835 WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML;
1837 INIT_LIST_HEAD(&stream_list);
1838 ret = wim_find_new_streams(w, old_wim_end, &stream_list);
1842 ret = open_wim_writable(w, w->filename, false,
1843 (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0);
1847 ret = lock_wim(w, w->out_fp);
1854 if (fseeko(w->out_fp, old_wim_end, SEEK_SET) != 0) {
1855 ERROR_WITH_ERRNO("Can't seek to end of WIM");
1859 return WIMLIB_ERR_WRITE;
1862 if (!list_empty(&stream_list)) {
1863 DEBUG("Writing newly added streams (offset = %"PRIu64")",
1865 ret = write_stream_list(&stream_list, w->out_fp,
1866 wimlib_get_compression_type(w),
1867 write_flags, num_threads,
1872 DEBUG("No new streams were added");
1875 found_modified_image = false;
1876 for (int i = 0; i < w->hdr.image_count; i++) {
1877 if (!found_modified_image)
1878 found_modified_image = w->image_metadata[i].modified;
1879 if (found_modified_image) {
1880 select_wim_image(w, i + 1);
1881 ret = write_metadata_resource(w);
1886 write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE;
1887 ret = finish_write(w, WIMLIB_ALL_IMAGES, write_flags,
1890 close_wim_writable(w);
1891 if (ret != 0 && !(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
1892 WARNING("Truncating `%s' to its original size (%"PRIu64" bytes)",
1893 w->filename, old_wim_end);
1894 /* Return value of truncate() is ignored because this is already
1896 (void)truncate(w->filename, old_wim_end);
1902 static int overwrite_wim_via_tmpfile(WIMStruct *w, int write_flags,
1903 unsigned num_threads,
1904 wimlib_progress_func_t progress_func)
1906 size_t wim_name_len;
1909 DEBUG("Overwriting `%s' via a temporary file", w->filename);
1911 /* Write the WIM to a temporary file in the same directory as the
1913 wim_name_len = strlen(w->filename);
1914 char tmpfile[wim_name_len + 10];
1915 memcpy(tmpfile, w->filename, wim_name_len);
1916 randomize_char_array_with_alnum(tmpfile + wim_name_len, 9);
1917 tmpfile[wim_name_len + 9] = '\0';
1919 ret = wimlib_write(w, tmpfile, WIMLIB_ALL_IMAGES,
1920 write_flags | WIMLIB_WRITE_FLAG_FSYNC,
1921 num_threads, progress_func);
1923 ERROR("Failed to write the WIM file `%s'", tmpfile);
1927 DEBUG("Renaming `%s' to `%s'", tmpfile, w->filename);
1929 /* Rename the new file to the old file .*/
1930 if (rename(tmpfile, w->filename) != 0) {
1931 ERROR_WITH_ERRNO("Failed to rename `%s' to `%s'",
1932 tmpfile, w->filename);
1933 ret = WIMLIB_ERR_RENAME;
1937 if (progress_func) {
1938 union wimlib_progress_info progress;
1939 progress.rename.from = tmpfile;
1940 progress.rename.to = w->filename;
1941 progress_func(WIMLIB_PROGRESS_MSG_RENAME, &progress);
1944 /* Close the original WIM file that was opened for reading. */
1945 if (w->fp != NULL) {
1950 /* Re-open the WIM read-only. */
1951 w->fp = fopen(w->filename, "rb");
1952 if (w->fp == NULL) {
1953 ret = WIMLIB_ERR_REOPEN;
1954 WARNING_WITH_ERRNO("Failed to re-open `%s' read-only",
1961 /* Remove temporary file. */
1962 if (unlink(tmpfile) != 0)
1963 WARNING_WITH_ERRNO("Failed to remove `%s'", tmpfile);
1968 * Writes a WIM file to the original file that it was read from, overwriting it.
1970 WIMLIBAPI int wimlib_overwrite(WIMStruct *w, int write_flags,
1971 unsigned num_threads,
1972 wimlib_progress_func_t progress_func)
1974 write_flags &= WIMLIB_WRITE_MASK_PUBLIC;
1977 return WIMLIB_ERR_NO_FILENAME;
1979 if (w->hdr.total_parts != 1) {
1980 ERROR("Cannot modify a split WIM");
1981 return WIMLIB_ERR_SPLIT_UNSUPPORTED;
1984 if ((!w->deletion_occurred || (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE))
1985 && !(write_flags & WIMLIB_WRITE_FLAG_REBUILD))
1988 ret = overwrite_wim_inplace(w, write_flags, num_threads,
1990 if (ret == WIMLIB_ERR_RESOURCE_ORDER)
1991 WARNING("Falling back to re-building entire WIM");
1995 return overwrite_wim_via_tmpfile(w, write_flags, num_threads,