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. */
32 # include <sys/file.h>
40 #include "wimlib_internal.h"
41 #include "buffer_io.h"
43 #include "lookup_table.h"
48 #ifdef ENABLE_MULTITHREADED_COMPRESSION
57 # include <ntfs-3g/attrib.h>
58 # include <ntfs-3g/inode.h>
59 # include <ntfs-3g/dir.h>
70 #if defined(__WIN32__) && !defined(INVALID_HANDLE_VALUE)
71 # define INVALID_HANDLE_VALUE ((HANDLE)(-1))
75 fflush_and_ftruncate(FILE *fp, off_t size)
81 ERROR_WITH_ERRNO("Failed to flush data to output WIM file");
82 return WIMLIB_ERR_WRITE;
84 ret = ftruncate(fileno(fp), size);
86 ERROR_WITH_ERRNO("Failed to truncate output WIM file to "
87 "%"PRIu64" bytes", size);
88 return WIMLIB_ERR_WRITE;
93 /* Chunk table that's located at the beginning of each compressed resource in
94 * the WIM. (This is not the on-disk format; the on-disk format just has an
95 * array of offsets.) */
99 u64 original_resource_size;
100 u64 bytes_per_chunk_entry;
108 * Allocates and initializes a chunk table, and reserves space for it in the
112 begin_wim_resource_chunk_tab(const struct wim_lookup_table_entry *lte,
115 struct chunk_table **chunk_tab_ret)
117 u64 size = wim_resource_size(lte);
118 u64 num_chunks = (size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE;
119 size_t alloc_size = sizeof(struct chunk_table) + num_chunks * sizeof(u64);
120 struct chunk_table *chunk_tab = CALLOC(1, alloc_size);
124 ERROR("Failed to allocate chunk table for %"PRIu64" byte "
126 ret = WIMLIB_ERR_NOMEM;
129 chunk_tab->file_offset = file_offset;
130 chunk_tab->num_chunks = num_chunks;
131 chunk_tab->original_resource_size = size;
132 chunk_tab->bytes_per_chunk_entry = (size >= (1ULL << 32)) ? 8 : 4;
133 chunk_tab->table_disk_size = chunk_tab->bytes_per_chunk_entry *
135 chunk_tab->cur_offset = 0;
136 chunk_tab->cur_offset_p = chunk_tab->offsets;
138 if (fwrite(chunk_tab, 1, chunk_tab->table_disk_size, out_fp) !=
139 chunk_tab->table_disk_size) {
140 ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
142 ret = WIMLIB_ERR_WRITE;
148 *chunk_tab_ret = chunk_tab;
153 * Pointer to function to compresses a chunk of a WIM resource.
155 * @chunk: Uncompressed data of the chunk.
156 * @chunk_size: Size of the uncompressed chunk in bytes.
157 * @compressed_chunk: Pointer to output buffer of size at least
158 * (@chunk_size - 1) bytes.
159 * @compressed_chunk_len_ret: Pointer to an unsigned int into which the size
160 * of the compressed chunk will be
163 * Returns zero if compressed succeeded, and nonzero if the chunk could not be
164 * compressed to any smaller than @chunk_size. This function cannot fail for
167 typedef int (*compress_func_t)(const void *, unsigned, void *, unsigned *);
170 get_compress_func(int out_ctype)
172 if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX)
175 return xpress_compress;
179 * Writes a chunk of a WIM resource to an output file.
181 * @chunk: Uncompressed data of the chunk.
182 * @chunk_size: Size of the chunk (<= WIM_CHUNK_SIZE)
183 * @out_fp: FILE * to write tho chunk to.
184 * @out_ctype: Compression type to use when writing the chunk (ignored if no
185 * chunk table provided)
186 * @chunk_tab: Pointer to chunk table being created. It is updated with the
187 * offset of the chunk we write.
189 * Returns 0 on success; nonzero on failure.
192 write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size,
193 FILE *out_fp, compress_func_t compress,
194 struct chunk_table *chunk_tab)
197 unsigned out_chunk_size;
199 u8 *compressed_chunk = alloca(chunk_size);
202 ret = compress(chunk, chunk_size, compressed_chunk,
205 out_chunk = compressed_chunk;
208 out_chunk_size = chunk_size;
210 *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset;
211 chunk_tab->cur_offset += out_chunk_size;
214 out_chunk_size = chunk_size;
216 if (fwrite(out_chunk, 1, out_chunk_size, out_fp) != out_chunk_size) {
217 ERROR_WITH_ERRNO("Failed to write WIM resource chunk");
218 return WIMLIB_ERR_WRITE;
224 * Finishes a WIM chunk table and writes it to the output file at the correct
227 * The final size of the full compressed resource is returned in the
228 * @compressed_size_p.
231 finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab,
232 FILE *out_fp, u64 *compressed_size_p)
234 size_t bytes_written;
235 if (fseeko(out_fp, chunk_tab->file_offset, SEEK_SET) != 0) {
236 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of output "
237 "WIM file", chunk_tab->file_offset);
238 return WIMLIB_ERR_WRITE;
241 if (chunk_tab->bytes_per_chunk_entry == 8) {
242 array_cpu_to_le64(chunk_tab->offsets, chunk_tab->num_chunks);
244 for (u64 i = 0; i < chunk_tab->num_chunks; i++)
245 ((u32*)chunk_tab->offsets)[i] =
246 cpu_to_le32(chunk_tab->offsets[i]);
248 bytes_written = fwrite((u8*)chunk_tab->offsets +
249 chunk_tab->bytes_per_chunk_entry,
250 1, chunk_tab->table_disk_size, out_fp);
251 if (bytes_written != chunk_tab->table_disk_size) {
252 ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
254 return WIMLIB_ERR_WRITE;
256 if (fseeko(out_fp, 0, SEEK_END) != 0) {
257 ERROR_WITH_ERRNO("Failed to seek to end of output WIM file");
258 return WIMLIB_ERR_WRITE;
260 *compressed_size_p = chunk_tab->cur_offset + chunk_tab->table_disk_size;
264 /* Prepare for multiple reads to a resource by caching a FILE * or NTFS
265 * attribute pointer in the lookup table entry. */
267 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);
293 ERROR_WITH_ERRNO("Failed to open inode `%s' in NTFS "
294 "volume", loc->path);
295 return WIMLIB_ERR_NTFS_3G;
297 lte->attr = ntfs_attr_open(ni,
298 loc->is_reparse_point ? AT_REPARSE_POINT : AT_DATA,
300 loc->stream_name_nchars);
302 ERROR_WITH_ERRNO("Failed to open attribute of `%s' in "
303 "NTFS volume", loc->path);
304 ntfs_inode_close(ni);
305 return WIMLIB_ERR_NTFS_3G;
313 if (lte->win32_file_on_disk_fp == INVALID_HANDLE_VALUE) {
314 lte->win32_file_on_disk_fp =
315 win32_open_file_readonly(lte->win32_file_on_disk, true);
316 if (lte->win32_file_on_disk_fp == INVALID_HANDLE_VALUE) {
317 ERROR("Win32 API: Can't open %ls", lte->win32_file_on_disk);
319 return WIMLIB_ERR_OPEN;
330 /* Undo prepare_resource_for_read() by closing the cached FILE * or NTFS
333 end_wim_resource_read(struct wim_lookup_table_entry *lte
339 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK
340 && lte->file_on_disk_fp)
342 fclose(lte->file_on_disk_fp);
343 lte->file_on_disk_fp = NULL;
346 else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME) {
348 ntfs_attr_close(lte->attr);
352 ntfs_inode_close(ni);
356 else if (lte->resource_location == RESOURCE_WIN32
357 && lte->win32_file_on_disk_fp != INVALID_HANDLE_VALUE)
359 win32_close_file(lte->win32_file_on_disk_fp);
360 lte->win32_file_on_disk_fp = INVALID_HANDLE_VALUE;
366 write_uncompressed_resource_and_truncate(struct wim_lookup_table_entry *lte,
369 struct resource_entry *out_res_entry)
372 if (fseeko(out_fp, file_offset, SEEK_SET) != 0) {
373 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of "
374 "output WIM file", file_offset);
375 return WIMLIB_ERR_WRITE;
377 ret = write_wim_resource(lte, out_fp, WIMLIB_COMPRESSION_TYPE_NONE,
382 return fflush_and_ftruncate(out_fp,
383 file_offset + wim_resource_size(lte));
387 * Writes a WIM resource to a FILE * opened for writing. The resource may be
388 * written uncompressed or compressed depending on the @out_ctype parameter.
390 * If by chance the resource compresses to more than the original size (this may
391 * happen with random data or files than are pre-compressed), the resource is
392 * instead written uncompressed (and this is reflected in the @out_res_entry by
393 * removing the WIM_RESHDR_FLAG_COMPRESSED flag).
395 * @lte: The lookup table entry for the WIM resource.
396 * @out_fp: The FILE * to write the resource to.
397 * @out_ctype: The compression type of the resource to write. Note: if this is
398 * the same as the compression type of the WIM resource we
399 * need to read, we simply copy the data (i.e. we do not
400 * uncompress it, then compress it again).
401 * @out_res_entry: If non-NULL, a resource entry that is filled in with the
402 * offset, original size, compressed size, and compression flag
403 * of the output resource.
405 * Returns 0 on success; nonzero on failure.
408 write_wim_resource(struct wim_lookup_table_entry *lte,
409 FILE *out_fp, int out_ctype,
410 struct resource_entry *out_res_entry,
415 u64 old_compressed_size;
416 u64 new_compressed_size;
419 struct chunk_table *chunk_tab = NULL;
422 compress_func_t compress = NULL;
424 ntfs_inode *ni = NULL;
429 /* Original size of the resource */
430 original_size = wim_resource_size(lte);
432 /* Compressed size of the resource (as it exists now) */
433 old_compressed_size = wim_resource_compressed_size(lte);
435 /* Current offset in output file */
436 file_offset = ftello(out_fp);
437 if (file_offset == -1) {
438 ERROR_WITH_ERRNO("Failed to get offset in output "
440 return WIMLIB_ERR_WRITE;
443 /* Are the compression types the same? If so, do a raw copy (copy
444 * without decompressing and recompressing the data). */
445 raw = (wim_resource_compression_type(lte) == out_ctype
446 && out_ctype != WIMLIB_COMPRESSION_TYPE_NONE
447 && !(flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS));
450 flags |= WIMLIB_RESOURCE_FLAG_RAW;
451 bytes_remaining = old_compressed_size;
453 flags &= ~WIMLIB_RESOURCE_FLAG_RAW;
454 bytes_remaining = original_size;
457 /* Empty resource; nothing needs to be done, so just return success. */
458 if (bytes_remaining == 0)
461 /* Buffer for reading chunks for the resource */
462 u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
464 /* If we are writing a compressed resource and not doing a raw copy, we
465 * need to initialize the chunk table */
466 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && !raw) {
467 ret = begin_wim_resource_chunk_tab(lte, out_fp, file_offset,
473 /* If the WIM resource is in an external file, open a FILE * to it so we
474 * don't have to open a temporary one in read_wim_resource() for each
477 ret = prepare_resource_for_read(lte, &ni);
479 ret = prepare_resource_for_read(lte);
484 /* If we aren't doing a raw copy, we will compute the SHA1 message
485 * digest of the resource as we read it, and verify it's the same as the
486 * hash given in the lookup table entry once we've finished reading the
491 compress = get_compress_func(out_ctype);
495 /* While there are still bytes remaining in the WIM resource, read a
496 * chunk of the resource, update SHA1, then write that chunk using the
497 * desired compression type. */
499 u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE);
500 ret = read_wim_resource(lte, buf, to_read, offset, flags);
504 sha1_update(&ctx, buf, to_read);
505 ret = write_wim_resource_chunk(buf, to_read, out_fp,
506 compress, chunk_tab);
509 bytes_remaining -= to_read;
511 } while (bytes_remaining);
513 /* Raw copy: The new compressed size is the same as the old compressed
516 * Using WIMLIB_COMPRESSION_TYPE_NONE: The new compressed size is the
519 * Using a different compression type: Call
520 * finish_wim_resource_chunk_tab() and it will provide the new
524 new_compressed_size = old_compressed_size;
526 if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE)
527 new_compressed_size = original_size;
529 ret = finish_wim_resource_chunk_tab(chunk_tab, out_fp,
530 &new_compressed_size);
536 /* Verify SHA1 message digest of the resource, unless we are doing a raw
537 * write (in which case we never even saw the uncompressed data). Or,
538 * if the hash we had before is all 0's, just re-set it to be the new
541 u8 md[SHA1_HASH_SIZE];
542 sha1_final(md, &ctx);
543 if (is_zero_hash(lte->hash)) {
544 copy_hash(lte->hash, md);
545 } else if (!hashes_equal(md, lte->hash)) {
546 ERROR("WIM resource has incorrect hash!");
547 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK) {
548 ERROR("We were reading it from `%s'; maybe it changed "
549 "while we were reading it.",
552 ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
557 if (!raw && new_compressed_size >= original_size &&
558 out_ctype != WIMLIB_COMPRESSION_TYPE_NONE)
560 /* Oops! We compressed the resource to larger than the original
561 * size. Write the resource uncompressed instead. */
562 ret = write_uncompressed_resource_and_truncate(lte,
570 out_res_entry->size = new_compressed_size;
571 out_res_entry->original_size = original_size;
572 out_res_entry->offset = file_offset;
573 out_res_entry->flags = lte->resource_entry.flags
574 & ~WIM_RESHDR_FLAG_COMPRESSED;
575 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE)
576 out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED;
582 end_wim_resource_read(lte, ni);
584 end_wim_resource_read(lte);
591 #ifdef ENABLE_MULTITHREADED_COMPRESSION
593 /* Blocking shared queue (solves the producer-consumer problem) */
594 struct shared_queue {
598 unsigned filled_slots;
600 pthread_mutex_t lock;
601 pthread_cond_t msg_avail_cond;
602 pthread_cond_t space_avail_cond;
606 shared_queue_init(struct shared_queue *q, unsigned size)
608 wimlib_assert(size != 0);
609 q->array = CALLOC(sizeof(q->array[0]), size);
611 return WIMLIB_ERR_NOMEM;
616 pthread_mutex_init(&q->lock, NULL);
617 pthread_cond_init(&q->msg_avail_cond, NULL);
618 pthread_cond_init(&q->space_avail_cond, NULL);
623 shared_queue_destroy(struct shared_queue *q)
626 pthread_mutex_destroy(&q->lock);
627 pthread_cond_destroy(&q->msg_avail_cond);
628 pthread_cond_destroy(&q->space_avail_cond);
632 shared_queue_put(struct shared_queue *q, void *obj)
634 pthread_mutex_lock(&q->lock);
635 while (q->filled_slots == q->size)
636 pthread_cond_wait(&q->space_avail_cond, &q->lock);
638 q->back = (q->back + 1) % q->size;
639 q->array[q->back] = obj;
642 pthread_cond_broadcast(&q->msg_avail_cond);
643 pthread_mutex_unlock(&q->lock);
647 shared_queue_get(struct shared_queue *q)
651 pthread_mutex_lock(&q->lock);
652 while (q->filled_slots == 0)
653 pthread_cond_wait(&q->msg_avail_cond, &q->lock);
655 obj = q->array[q->front];
656 q->array[q->front] = NULL;
657 q->front = (q->front + 1) % q->size;
660 pthread_cond_broadcast(&q->space_avail_cond);
661 pthread_mutex_unlock(&q->lock);
665 struct compressor_thread_params {
666 struct shared_queue *res_to_compress_queue;
667 struct shared_queue *compressed_res_queue;
668 compress_func_t compress;
671 #define MAX_CHUNKS_PER_MSG 2
674 struct wim_lookup_table_entry *lte;
675 u8 *uncompressed_chunks[MAX_CHUNKS_PER_MSG];
676 u8 *out_compressed_chunks[MAX_CHUNKS_PER_MSG];
677 u8 *compressed_chunks[MAX_CHUNKS_PER_MSG];
678 unsigned uncompressed_chunk_sizes[MAX_CHUNKS_PER_MSG];
679 unsigned compressed_chunk_sizes[MAX_CHUNKS_PER_MSG];
681 struct list_head list;
687 compress_chunks(struct message *msg, compress_func_t compress)
689 for (unsigned i = 0; i < msg->num_chunks; i++) {
690 DEBUG2("compress chunk %u of %u", i, msg->num_chunks);
691 int ret = compress(msg->uncompressed_chunks[i],
692 msg->uncompressed_chunk_sizes[i],
693 msg->compressed_chunks[i],
694 &msg->compressed_chunk_sizes[i]);
696 msg->out_compressed_chunks[i] = msg->compressed_chunks[i];
698 msg->out_compressed_chunks[i] = msg->uncompressed_chunks[i];
699 msg->compressed_chunk_sizes[i] = msg->uncompressed_chunk_sizes[i];
704 /* Compressor thread routine. This is a lot simpler than the main thread
705 * routine: just repeatedly get a group of chunks from the
706 * res_to_compress_queue, compress them, and put them in the
707 * compressed_res_queue. A NULL pointer indicates that the thread should stop.
710 compressor_thread_proc(void *arg)
712 struct compressor_thread_params *params = arg;
713 struct shared_queue *res_to_compress_queue = params->res_to_compress_queue;
714 struct shared_queue *compressed_res_queue = params->compressed_res_queue;
715 compress_func_t compress = params->compress;
718 DEBUG("Compressor thread ready");
719 while ((msg = shared_queue_get(res_to_compress_queue)) != NULL) {
720 compress_chunks(msg, compress);
721 shared_queue_put(compressed_res_queue, msg);
723 DEBUG("Compressor thread terminating");
726 #endif /* ENABLE_MULTITHREADED_COMPRESSION */
729 do_write_stream_list(struct list_head *my_resources,
732 wimlib_progress_func_t progress_func,
733 union wimlib_progress_info *progress,
734 int write_resource_flags)
737 struct wim_lookup_table_entry *lte, *tmp;
739 list_for_each_entry_safe(lte, tmp, my_resources, staging_list) {
740 ret = write_wim_resource(lte,
743 <e->output_resource_entry,
744 write_resource_flags);
747 list_del(<e->staging_list);
748 progress->write_streams.completed_bytes +=
749 wim_resource_size(lte);
750 progress->write_streams.completed_streams++;
752 progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
760 write_stream_list_serial(struct list_head *stream_list,
764 wimlib_progress_func_t progress_func,
765 union wimlib_progress_info *progress)
767 int write_resource_flags;
769 if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
770 write_resource_flags = WIMLIB_RESOURCE_FLAG_RECOMPRESS;
772 write_resource_flags = 0;
773 progress->write_streams.num_threads = 1;
775 progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress);
776 return do_write_stream_list(stream_list, out_fp,
777 out_ctype, progress_func,
778 progress, write_resource_flags);
781 #ifdef ENABLE_MULTITHREADED_COMPRESSION
783 write_wim_chunks(struct message *msg, FILE *out_fp,
784 struct chunk_table *chunk_tab)
786 for (unsigned i = 0; i < msg->num_chunks; i++) {
787 unsigned chunk_csize = msg->compressed_chunk_sizes[i];
789 DEBUG2("Write wim chunk %u of %u (csize = %u)",
790 i, msg->num_chunks, chunk_csize);
792 if (fwrite(msg->out_compressed_chunks[i], 1, chunk_csize, out_fp)
795 ERROR_WITH_ERRNO("Failed to write WIM chunk");
796 return WIMLIB_ERR_WRITE;
799 *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset;
800 chunk_tab->cur_offset += chunk_csize;
806 * This function is executed by the main thread when the resources are being
807 * compressed in parallel. The main thread is in change of all reading of the
808 * uncompressed data and writing of the compressed data. The compressor threads
809 * *only* do compression from/to in-memory buffers.
811 * Each unit of work given to a compressor thread is up to MAX_CHUNKS_PER_MSG
812 * chunks of compressed data to compress, represented in a `struct message'.
813 * Each message is passed from the main thread to a worker thread through the
814 * res_to_compress_queue, and it is passed back through the
815 * compressed_res_queue.
818 main_writer_thread_proc(struct list_head *stream_list,
821 struct shared_queue *res_to_compress_queue,
822 struct shared_queue *compressed_res_queue,
825 wimlib_progress_func_t progress_func,
826 union wimlib_progress_info *progress)
829 struct chunk_table *cur_chunk_tab = NULL;
830 struct message *msgs = CALLOC(num_messages, sizeof(struct message));
831 struct wim_lookup_table_entry *next_lte = NULL;
833 // Initially, all the messages are available to use.
834 LIST_HEAD(available_msgs);
837 ret = WIMLIB_ERR_NOMEM;
841 for (size_t i = 0; i < num_messages; i++)
842 list_add(&msgs[i].list, &available_msgs);
844 // outstanding_resources is the list of resources that currently have
845 // had chunks sent off for compression.
847 // The first stream in outstanding_resources is the stream that is
848 // currently being written (cur_lte).
850 // The last stream in outstanding_resources is the stream that is
851 // currently being read and chunks fed to the compressor threads
854 // Depending on the number of threads and the sizes of the resource,
855 // the outstanding streams list may contain streams between cur_lte and
856 // next_lte that have all their chunks compressed or being compressed,
857 // but haven't been written yet.
859 LIST_HEAD(outstanding_resources);
860 struct list_head *next_resource = stream_list->next;
862 u64 next_num_chunks = 0;
864 // As in write_wim_resource(), each resource we read is checksummed.
865 SHA_CTX next_sha_ctx;
866 u8 next_hash[SHA1_HASH_SIZE];
868 // Resources that don't need any chunks compressed are added to this
869 // list and written directly by the main thread.
870 LIST_HEAD(my_resources);
872 struct wim_lookup_table_entry *cur_lte = NULL;
876 ntfs_inode *ni = NULL;
879 DEBUG("Initializing buffers for uncompressed "
880 "and compressed data (%zu bytes needed)",
881 num_messages * MAX_CHUNKS_PER_MSG * WIM_CHUNK_SIZE * 2);
883 // Pre-allocate all the buffers that will be needed to do the chunk
885 for (size_t i = 0; i < num_messages; i++) {
886 for (size_t j = 0; j < MAX_CHUNKS_PER_MSG; j++) {
887 msgs[i].compressed_chunks[j] = MALLOC(WIM_CHUNK_SIZE);
889 // The extra 8 bytes is because longest_match() in
890 // lz77.c may read a little bit off the end of the
891 // uncompressed data. It doesn't need to be
892 // initialized--- we really just need to avoid accessing
894 msgs[i].uncompressed_chunks[j] = MALLOC(WIM_CHUNK_SIZE + 8);
895 if (msgs[i].compressed_chunks[j] == NULL ||
896 msgs[i].uncompressed_chunks[j] == NULL)
898 ret = WIMLIB_ERR_NOMEM;
904 // This loop is executed until all resources have been written, except
905 // possibly a few that have been added to the @my_resources list for
908 // Send chunks to the compressor threads until either (a) there
909 // are no more messages available since they were all sent off,
910 // or (b) there are no more resources that need to be
912 while (!list_empty(&available_msgs)) {
913 if (next_chunk == next_num_chunks) {
914 // If next_chunk == next_num_chunks, there are
915 // no more chunks to write in the current
916 // stream. So, check the SHA1 message digest of
917 // the stream that was just finished (unless
918 // next_lte == NULL, which is the case the very
919 // first time this loop is entered, and also
920 // near the very end of the compression when
921 // there are no more streams.) Then, advance to
922 // the next stream (if there is one).
923 if (next_lte != NULL) {
925 end_wim_resource_read(next_lte, ni);
928 end_wim_resource_read(next_lte);
930 DEBUG2("Finalize SHA1 md (next_num_chunks=%zu)",
932 sha1_final(next_hash, &next_sha_ctx);
933 if (!hashes_equal(next_lte->hash, next_hash)) {
934 ERROR("WIM resource has incorrect hash!");
935 if (next_lte->resource_location ==
936 RESOURCE_IN_FILE_ON_DISK)
938 ERROR("We were reading it from `%s'; "
939 "maybe it changed while we were "
941 next_lte->file_on_disk);
943 ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
948 // Advance to the next resource.
950 // If the next resource needs no compression, just write
951 // it with this thread (not now though--- we could be in
952 // the middle of writing another resource.) Keep doing
953 // this until we either get to the end of the resources
954 // list, or we get to a resource that needs compression.
956 if (next_resource == stream_list) {
957 // No more resources to send for
962 next_lte = container_of(next_resource,
963 struct wim_lookup_table_entry,
965 next_resource = next_resource->next;
966 if ((!(write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
967 && wim_resource_compression_type(next_lte) == out_ctype)
968 || wim_resource_size(next_lte) == 0)
970 list_add_tail(&next_lte->staging_list,
973 list_add_tail(&next_lte->staging_list,
974 &outstanding_resources);
976 next_num_chunks = wim_resource_chunks(next_lte);
977 sha1_init(&next_sha_ctx);
978 INIT_LIST_HEAD(&next_lte->msg_list);
980 ret = prepare_resource_for_read(next_lte, &ni);
982 ret = prepare_resource_for_read(next_lte);
987 if (cur_lte == NULL) {
988 // Set cur_lte for the
997 if (next_lte == NULL) {
998 // No more resources to send for compression
1002 // Get a message from the available messages
1004 msg = container_of(available_msgs.next,
1008 // ... and delete it from the available messages
1010 list_del(&msg->list);
1012 // Initialize the message with the chunks to
1014 msg->num_chunks = min(next_num_chunks - next_chunk,
1015 MAX_CHUNKS_PER_MSG);
1016 msg->lte = next_lte;
1017 msg->complete = false;
1018 msg->begin_chunk = next_chunk;
1020 unsigned size = WIM_CHUNK_SIZE;
1021 for (unsigned i = 0; i < msg->num_chunks; i++) {
1023 // Read chunk @next_chunk of the stream into the
1024 // message so that a compressor thread can
1027 if (next_chunk == next_num_chunks - 1) {
1028 size = MODULO_NONZERO(wim_resource_size(next_lte),
1032 DEBUG2("Read resource (size=%u, offset=%zu)",
1033 size, next_chunk * WIM_CHUNK_SIZE);
1035 msg->uncompressed_chunk_sizes[i] = size;
1037 ret = read_wim_resource(next_lte,
1038 msg->uncompressed_chunks[i],
1040 next_chunk * WIM_CHUNK_SIZE,
1044 sha1_update(&next_sha_ctx,
1045 msg->uncompressed_chunks[i], size);
1049 // Send the compression request
1050 list_add_tail(&msg->list, &next_lte->msg_list);
1051 shared_queue_put(res_to_compress_queue, msg);
1052 DEBUG2("Compression request sent");
1055 // If there are no outstanding resources, there are no more
1056 // resources that need to be written.
1057 if (list_empty(&outstanding_resources)) {
1062 // Get the next message from the queue and process it.
1063 // The message will contain 1 or more data chunks that have been
1065 msg = shared_queue_get(compressed_res_queue);
1066 msg->complete = true;
1068 // Is this the next chunk in the current resource? If it's not
1069 // (i.e., an earlier chunk in a same or different resource
1070 // hasn't been compressed yet), do nothing, and keep this
1071 // message around until all earlier chunks are received.
1073 // Otherwise, write all the chunks we can.
1074 while (cur_lte != NULL &&
1075 !list_empty(&cur_lte->msg_list) &&
1076 (msg = container_of(cur_lte->msg_list.next,
1080 DEBUG2("Complete msg (begin_chunk=%"PRIu64")", msg->begin_chunk);
1081 if (msg->begin_chunk == 0) {
1082 DEBUG2("Begin chunk tab");
1084 // This is the first set of chunks. Leave space
1085 // for the chunk table in the output file.
1086 off_t cur_offset = ftello(out_fp);
1087 if (cur_offset == -1) {
1088 ret = WIMLIB_ERR_WRITE;
1091 ret = begin_wim_resource_chunk_tab(cur_lte,
1099 // Write the compressed chunks from the message.
1100 ret = write_wim_chunks(msg, out_fp, cur_chunk_tab);
1104 list_del(&msg->list);
1106 // This message is available to use for different chunks
1108 list_add(&msg->list, &available_msgs);
1110 // Was this the last chunk of the stream? If so, finish
1112 if (list_empty(&cur_lte->msg_list) &&
1113 msg->begin_chunk + msg->num_chunks == cur_chunk_tab->num_chunks)
1115 DEBUG2("Finish wim chunk tab");
1117 ret = finish_wim_resource_chunk_tab(cur_chunk_tab,
1123 if (res_csize >= wim_resource_size(cur_lte)) {
1124 /* Oops! We compressed the resource to
1125 * larger than the original size. Write
1126 * the resource uncompressed instead. */
1127 ret = write_uncompressed_resource_and_truncate(
1130 cur_chunk_tab->file_offset,
1131 &cur_lte->output_resource_entry);
1135 cur_lte->output_resource_entry.size =
1138 cur_lte->output_resource_entry.original_size =
1139 cur_lte->resource_entry.original_size;
1141 cur_lte->output_resource_entry.offset =
1142 cur_chunk_tab->file_offset;
1144 cur_lte->output_resource_entry.flags =
1145 cur_lte->resource_entry.flags |
1146 WIM_RESHDR_FLAG_COMPRESSED;
1149 progress->write_streams.completed_bytes +=
1150 wim_resource_size(cur_lte);
1151 progress->write_streams.completed_streams++;
1153 if (progress_func) {
1154 progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
1158 FREE(cur_chunk_tab);
1159 cur_chunk_tab = NULL;
1161 struct list_head *next = cur_lte->staging_list.next;
1162 list_del(&cur_lte->staging_list);
1164 if (next == &outstanding_resources)
1167 cur_lte = container_of(cur_lte->staging_list.next,
1168 struct wim_lookup_table_entry,
1171 // Since we just finished writing a stream,
1172 // write any streams that have been added to the
1173 // my_resources list for direct writing by the
1174 // main thread (e.g. resources that don't need
1175 // to be compressed because the desired
1176 // compression type is the same as the previous
1177 // compression type).
1178 ret = do_write_stream_list(&my_resources,
1191 if (ret == WIMLIB_ERR_NOMEM) {
1192 ERROR("Could not allocate enough memory for "
1193 "multi-threaded compression");
1198 end_wim_resource_read(next_lte, ni);
1200 end_wim_resource_read(next_lte);
1205 ret = do_write_stream_list(&my_resources, out_fp,
1206 out_ctype, progress_func,
1210 size_t num_available_msgs = 0;
1211 struct list_head *cur;
1213 list_for_each(cur, &available_msgs) {
1214 num_available_msgs++;
1217 while (num_available_msgs < num_messages) {
1218 shared_queue_get(compressed_res_queue);
1219 num_available_msgs++;
1225 for (size_t i = 0; i < num_messages; i++) {
1226 for (size_t j = 0; j < MAX_CHUNKS_PER_MSG; j++) {
1227 FREE(msgs[i].compressed_chunks[j]);
1228 FREE(msgs[i].uncompressed_chunks[j]);
1234 FREE(cur_chunk_tab);
1239 get_default_num_threads()
1242 return win32_get_number_of_processors();
1244 return sysconf(_SC_NPROCESSORS_ONLN);
1249 write_stream_list_parallel(struct list_head *stream_list,
1253 unsigned num_threads,
1254 wimlib_progress_func_t progress_func,
1255 union wimlib_progress_info *progress)
1258 struct shared_queue res_to_compress_queue;
1259 struct shared_queue compressed_res_queue;
1260 pthread_t *compressor_threads = NULL;
1262 if (num_threads == 0) {
1263 long nthreads = get_default_num_threads();
1264 if (nthreads < 1 || nthreads > UINT_MAX) {
1265 WARNING("Could not determine number of processors! Assuming 1");
1268 num_threads = nthreads;
1272 progress->write_streams.num_threads = num_threads;
1273 wimlib_assert(stream_list->next != stream_list);
1275 static const double MESSAGES_PER_THREAD = 2.0;
1276 size_t queue_size = (size_t)(num_threads * MESSAGES_PER_THREAD);
1278 DEBUG("Initializing shared queues (queue_size=%zu)", queue_size);
1280 ret = shared_queue_init(&res_to_compress_queue, queue_size);
1284 ret = shared_queue_init(&compressed_res_queue, queue_size);
1286 goto out_destroy_res_to_compress_queue;
1288 struct compressor_thread_params params;
1289 params.res_to_compress_queue = &res_to_compress_queue;
1290 params.compressed_res_queue = &compressed_res_queue;
1291 params.compress = get_compress_func(out_ctype);
1293 compressor_threads = MALLOC(num_threads * sizeof(pthread_t));
1294 if (!compressor_threads) {
1295 ret = WIMLIB_ERR_NOMEM;
1296 goto out_destroy_compressed_res_queue;
1299 for (unsigned i = 0; i < num_threads; i++) {
1300 DEBUG("pthread_create thread %u", i);
1301 ret = pthread_create(&compressor_threads[i], NULL,
1302 compressor_thread_proc, ¶ms);
1305 ERROR_WITH_ERRNO("Failed to create compressor "
1313 progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress);
1315 ret = main_writer_thread_proc(stream_list,
1318 &res_to_compress_queue,
1319 &compressed_res_queue,
1325 for (unsigned i = 0; i < num_threads; i++)
1326 shared_queue_put(&res_to_compress_queue, NULL);
1328 for (unsigned i = 0; i < num_threads; i++) {
1329 if (pthread_join(compressor_threads[i], NULL)) {
1330 WARNING_WITH_ERRNO("Failed to join compressor "
1334 FREE(compressor_threads);
1335 out_destroy_compressed_res_queue:
1336 shared_queue_destroy(&compressed_res_queue);
1337 out_destroy_res_to_compress_queue:
1338 shared_queue_destroy(&res_to_compress_queue);
1339 if (ret >= 0 && ret != WIMLIB_ERR_NOMEM)
1342 WARNING("Falling back to single-threaded compression");
1343 return write_stream_list_serial(stream_list,
1354 * Write a list of streams to a WIM (@out_fp) using the compression type
1355 * @out_ctype and up to @num_threads compressor threads.
1358 write_stream_list(struct list_head *stream_list, FILE *out_fp,
1359 int out_ctype, int write_flags,
1360 unsigned num_threads,
1361 wimlib_progress_func_t progress_func)
1363 struct wim_lookup_table_entry *lte;
1364 size_t num_streams = 0;
1365 u64 total_bytes = 0;
1366 u64 total_compression_bytes = 0;
1367 union wimlib_progress_info progress;
1369 list_for_each_entry(lte, stream_list, staging_list) {
1371 total_bytes += wim_resource_size(lte);
1372 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE
1373 && (wim_resource_compression_type(lte) != out_ctype ||
1374 (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)))
1376 total_compression_bytes += wim_resource_size(lte);
1379 progress.write_streams.total_bytes = total_bytes;
1380 progress.write_streams.total_streams = num_streams;
1381 progress.write_streams.completed_bytes = 0;
1382 progress.write_streams.completed_streams = 0;
1383 progress.write_streams.num_threads = num_threads;
1384 progress.write_streams.compression_type = out_ctype;
1386 #ifdef ENABLE_MULTITHREADED_COMPRESSION
1387 if (total_compression_bytes >= 1000000 && num_threads != 1)
1388 return write_stream_list_parallel(stream_list,
1397 return write_stream_list_serial(stream_list,
1405 struct lte_overwrite_prepare_args {
1408 struct list_head *stream_list;
1412 lte_overwrite_prepare(struct wim_lookup_table_entry *lte, void *arg)
1414 struct lte_overwrite_prepare_args *args = arg;
1416 if (lte->resource_location == RESOURCE_IN_WIM &&
1417 lte->wim == args->wim &&
1418 lte->resource_entry.offset + lte->resource_entry.size > args->end_offset)
1420 #ifdef ENABLE_ERROR_MESSAGES
1421 ERROR("The following resource is after the XML data:");
1422 print_lookup_table_entry(lte, stderr);
1424 return WIMLIB_ERR_RESOURCE_ORDER;
1427 lte->out_refcnt = lte->refcnt;
1428 memcpy(<e->output_resource_entry, <e->resource_entry,
1429 sizeof(struct resource_entry));
1430 if (!(lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA)) {
1431 wimlib_assert(lte->resource_location != RESOURCE_NONEXISTENT);
1432 if (lte->resource_location != RESOURCE_IN_WIM || lte->wim != args->wim)
1433 list_add(<e->staging_list, args->stream_list);
1439 wim_find_new_streams(WIMStruct *wim, off_t end_offset,
1440 struct list_head *stream_list)
1442 struct lte_overwrite_prepare_args args = {
1444 .end_offset = end_offset,
1445 .stream_list = stream_list,
1448 return for_lookup_table_entry(wim->lookup_table,
1449 lte_overwrite_prepare, &args);
1453 inode_find_streams_to_write(struct wim_inode *inode,
1454 struct wim_lookup_table *table,
1455 struct list_head *stream_list)
1457 struct wim_lookup_table_entry *lte;
1458 for (unsigned i = 0; i <= inode->i_num_ads; i++) {
1459 lte = inode_stream_lte(inode, i, table);
1461 if (lte->out_refcnt == 0)
1462 list_add_tail(<e->staging_list, stream_list);
1463 lte->out_refcnt += inode->i_nlink;
1470 image_find_streams_to_write(WIMStruct *w)
1472 struct wim_inode *inode;
1473 struct hlist_node *cur;
1474 struct hlist_head *inode_list;
1476 inode_list = &wim_get_current_image_metadata(w)->inode_list;
1477 hlist_for_each_entry(inode, cur, inode_list, i_hlist) {
1478 inode_find_streams_to_write(inode, w->lookup_table,
1479 (struct list_head*)w->private);
1485 write_wim_streams(WIMStruct *w, int image, int write_flags,
1486 unsigned num_threads,
1487 wimlib_progress_func_t progress_func)
1490 for_lookup_table_entry(w->lookup_table, lte_zero_out_refcnt, NULL);
1491 LIST_HEAD(stream_list);
1492 w->private = &stream_list;
1493 for_image(w, image, image_find_streams_to_write);
1494 return write_stream_list(&stream_list, w->out_fp,
1495 wimlib_get_compression_type(w), write_flags,
1496 num_threads, progress_func);
1500 * Finish writing a WIM file: write the lookup table, xml data, and integrity
1501 * table (optional), then overwrite the WIM header.
1503 * write_flags is a bitwise OR of the following:
1505 * (public) WIMLIB_WRITE_FLAG_CHECK_INTEGRITY:
1506 * Include an integrity table.
1508 * (public) WIMLIB_WRITE_FLAG_SHOW_PROGRESS:
1509 * Show progress information when (if) writing the integrity table.
1511 * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE:
1512 * Don't write the lookup table.
1514 * (private) WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE:
1515 * When (if) writing the integrity table, re-use entries from the
1516 * existing integrity table, if possible.
1518 * (private) WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML:
1519 * After writing the XML data but before writing the integrity
1520 * table, write a temporary WIM header and flush the stream so that
1521 * the WIM is less likely to become corrupted upon abrupt program
1524 * (private) WIMLIB_WRITE_FLAG_FSYNC:
1525 * fsync() the output file before closing it.
1529 finish_write(WIMStruct *w, int image, int write_flags,
1530 wimlib_progress_func_t progress_func)
1533 struct wim_header hdr;
1534 FILE *out = w->out_fp;
1536 /* @hdr will be the header for the new WIM. First copy all the data
1537 * from the header in the WIMStruct; then set all the fields that may
1538 * have changed, including the resource entries, boot index, and image
1540 memcpy(&hdr, &w->hdr, sizeof(struct wim_header));
1542 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
1543 ret = write_lookup_table(w->lookup_table, out, &hdr.lookup_table_res_entry);
1548 ret = write_xml_data(w->wim_info, image, out,
1549 (write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE) ?
1550 wim_info_get_total_bytes(w->wim_info) : 0,
1551 &hdr.xml_res_entry);
1555 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
1556 if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) {
1557 struct wim_header checkpoint_hdr;
1558 memcpy(&checkpoint_hdr, &hdr, sizeof(struct wim_header));
1559 memset(&checkpoint_hdr.integrity, 0, sizeof(struct resource_entry));
1560 if (fseeko(out, 0, SEEK_SET) != 0) {
1561 ERROR_WITH_ERRNO("Failed to seek to beginning "
1562 "of WIM being written");
1563 ret = WIMLIB_ERR_WRITE;
1566 ret = write_header(&checkpoint_hdr, out);
1570 if (fflush(out) != 0) {
1571 ERROR_WITH_ERRNO("Can't write data to WIM");
1572 ret = WIMLIB_ERR_WRITE;
1576 if (fseeko(out, 0, SEEK_END) != 0) {
1577 ERROR_WITH_ERRNO("Failed to seek to end "
1578 "of WIM being written");
1579 ret = WIMLIB_ERR_WRITE;
1584 off_t old_lookup_table_end;
1585 off_t new_lookup_table_end;
1586 if (write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE) {
1587 old_lookup_table_end = w->hdr.lookup_table_res_entry.offset +
1588 w->hdr.lookup_table_res_entry.size;
1590 old_lookup_table_end = 0;
1592 new_lookup_table_end = hdr.lookup_table_res_entry.offset +
1593 hdr.lookup_table_res_entry.size;
1595 ret = write_integrity_table(out,
1597 new_lookup_table_end,
1598 old_lookup_table_end,
1603 memset(&hdr.integrity, 0, sizeof(struct resource_entry));
1607 * In the WIM header, there is room for the resource entry for a
1608 * metadata resource labeled as the "boot metadata". This entry should
1609 * be zeroed out if there is no bootable image (boot_idx 0). Otherwise,
1610 * it should be a copy of the resource entry for the image that is
1611 * marked as bootable. This is not well documented...
1614 /* Set image count and boot index correctly for single image writes */
1615 if (image != WIMLIB_ALL_IMAGES) {
1616 hdr.image_count = 1;
1617 if (hdr.boot_idx == image)
1623 if (hdr.boot_idx == 0) {
1624 memset(&hdr.boot_metadata_res_entry, 0,
1625 sizeof(struct resource_entry));
1627 memcpy(&hdr.boot_metadata_res_entry,
1629 hdr.boot_idx - 1].metadata_lte->output_resource_entry,
1630 sizeof(struct resource_entry));
1633 if (fseeko(out, 0, SEEK_SET) != 0) {
1634 ERROR_WITH_ERRNO("Failed to seek to beginning of WIM "
1636 ret = WIMLIB_ERR_WRITE;
1640 ret = write_header(&hdr, out);
1644 if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) {
1645 if (fflush(out) != 0
1646 || fsync(fileno(out)) != 0)
1648 ERROR_WITH_ERRNO("Error flushing data to WIM file");
1649 ret = WIMLIB_ERR_WRITE;
1653 if (fclose(out) != 0) {
1654 ERROR_WITH_ERRNO("Failed to close the WIM file");
1656 ret = WIMLIB_ERR_WRITE;
1662 #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK)
1664 lock_wim(WIMStruct *w, FILE *fp)
1667 if (fp && !w->wim_locked) {
1668 ret = flock(fileno(fp), LOCK_EX | LOCK_NB);
1670 if (errno == EWOULDBLOCK) {
1671 ERROR("`%s' is already being modified or has been "
1672 "mounted read-write\n"
1673 " by another process!", w->filename);
1674 ret = WIMLIB_ERR_ALREADY_LOCKED;
1676 WARNING_WITH_ERRNO("Failed to lock `%s'",
1689 open_wim_writable(WIMStruct *w, const mbchar *path,
1690 bool trunc, bool readable)
1701 wimlib_assert(w->out_fp == NULL);
1702 w->out_fp = fopen(path, mode);
1706 ERROR_WITH_ERRNO("Failed to open `%s' for writing", path);
1707 return WIMLIB_ERR_OPEN;
1713 close_wim_writable(WIMStruct *w)
1716 if (fclose(w->out_fp) != 0) {
1717 WARNING_WITH_ERRNO("Failed to close output WIM");
1723 /* Open file stream and write dummy header for WIM. */
1725 begin_write(WIMStruct *w, const mbchar *path, int write_flags)
1728 ret = open_wim_writable(w, path, true,
1729 (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0);
1732 /* Write dummy header. It will be overwritten later. */
1733 return write_header(&w->hdr, w->out_fp);
1736 /* Writes a stand-alone WIM to a file. */
1738 wimlib_write(WIMStruct *w, const mbchar *path,
1739 int image, int write_flags, unsigned num_threads,
1740 wimlib_progress_func_t progress_func)
1745 return WIMLIB_ERR_INVALID_PARAM;
1747 write_flags &= WIMLIB_WRITE_MASK_PUBLIC;
1749 if (image != WIMLIB_ALL_IMAGES &&
1750 (image < 1 || image > w->hdr.image_count))
1751 return WIMLIB_ERR_INVALID_IMAGE;
1753 if (w->hdr.total_parts != 1) {
1754 ERROR("Cannot call wimlib_write() on part of a split WIM");
1755 return WIMLIB_ERR_SPLIT_UNSUPPORTED;
1758 ret = begin_write(w, path, write_flags);
1762 ret = write_wim_streams(w, image, write_flags, num_threads,
1768 progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN, NULL);
1770 ret = for_image(w, image, write_metadata_resource);
1775 progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_END, NULL);
1777 ret = finish_write(w, image, write_flags, progress_func);
1779 close_wim_writable(w);
1780 DEBUG("wimlib_write(path=%s) = %d", path, ret);
1785 any_images_modified(WIMStruct *w)
1787 for (int i = 0; i < w->hdr.image_count; i++)
1788 if (w->image_metadata[i].modified)
1794 * Overwrite a WIM, possibly appending streams to it.
1796 * A WIM looks like (or is supposed to look like) the following:
1798 * Header (212 bytes)
1799 * Streams and metadata resources (variable size)
1800 * Lookup table (variable size)
1801 * XML data (variable size)
1802 * Integrity table (optional) (variable size)
1804 * If we are not adding any streams or metadata resources, the lookup table is
1805 * unchanged--- so we only need to overwrite the XML data, integrity table, and
1806 * header. This operation is potentially unsafe if the program is abruptly
1807 * terminated while the XML data or integrity table are being overwritten, but
1808 * before the new header has been written. To partially alleviate this problem,
1809 * a special flag (WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) is passed to
1810 * finish_write() to cause a temporary WIM header to be written after the XML
1811 * data has been written. This may prevent the WIM from becoming corrupted if
1812 * the program is terminated while the integrity table is being calculated (but
1813 * no guarantees, due to write re-ordering...).
1815 * If we are adding new streams or images (metadata resources), the lookup table
1816 * needs to be changed, and those streams need to be written. In this case, we
1817 * try to perform a safe update of the WIM file by writing the streams *after*
1818 * the end of the previous WIM, then writing the new lookup table, XML data, and
1819 * (optionally) integrity table following the new streams. This will produce a
1820 * layout like the following:
1822 * Header (212 bytes)
1823 * (OLD) Streams and metadata resources (variable size)
1824 * (OLD) Lookup table (variable size)
1825 * (OLD) XML data (variable size)
1826 * (OLD) Integrity table (optional) (variable size)
1827 * (NEW) Streams and metadata resources (variable size)
1828 * (NEW) Lookup table (variable size)
1829 * (NEW) XML data (variable size)
1830 * (NEW) Integrity table (optional) (variable size)
1832 * At all points, the WIM is valid as nothing points to the new data yet. Then,
1833 * the header is overwritten to point to the new lookup table, XML data, and
1834 * integrity table, to produce the following layout:
1836 * Header (212 bytes)
1837 * Streams and metadata resources (variable size)
1838 * Nothing (variable size)
1839 * More Streams and metadata resources (variable size)
1840 * Lookup table (variable size)
1841 * XML data (variable size)
1842 * Integrity table (optional) (variable size)
1844 * This method allows an image to be appended to a large WIM very quickly, and
1845 * is is crash-safe except in the case of write re-ordering, but the
1846 * disadvantage is that a small hole is left in the WIM where the old lookup
1847 * table, xml data, and integrity table were. (These usually only take up a
1848 * small amount of space compared to the streams, however.)
1851 overwrite_wim_inplace(WIMStruct *w, int write_flags,
1852 unsigned num_threads,
1853 wimlib_progress_func_t progress_func)
1856 struct list_head stream_list;
1858 bool found_modified_image;
1860 DEBUG("Overwriting `%s' in-place", w->filename);
1862 /* Make sure that the integrity table (if present) is after the XML
1863 * data, and that there are no stream resources, metadata resources, or
1864 * lookup tables after the XML data. Otherwise, these data would be
1866 if (w->hdr.integrity.offset != 0 &&
1867 w->hdr.integrity.offset < w->hdr.xml_res_entry.offset) {
1868 ERROR("Didn't expect the integrity table to be before the XML data");
1869 return WIMLIB_ERR_RESOURCE_ORDER;
1872 if (w->hdr.lookup_table_res_entry.offset > w->hdr.xml_res_entry.offset) {
1873 ERROR("Didn't expect the lookup table to be after the XML data");
1874 return WIMLIB_ERR_RESOURCE_ORDER;
1878 if (w->hdr.integrity.offset)
1879 old_wim_end = w->hdr.integrity.offset + w->hdr.integrity.size;
1881 old_wim_end = w->hdr.xml_res_entry.offset + w->hdr.xml_res_entry.size;
1883 if (!w->deletion_occurred && !any_images_modified(w)) {
1884 /* If no images have been modified and no images have been
1885 * deleted, a new lookup table does not need to be written. */
1886 old_wim_end = w->hdr.lookup_table_res_entry.offset +
1887 w->hdr.lookup_table_res_entry.size;
1888 write_flags |= WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE |
1889 WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML;
1891 INIT_LIST_HEAD(&stream_list);
1892 ret = wim_find_new_streams(w, old_wim_end, &stream_list);
1896 ret = open_wim_writable(w, w->filename, false,
1897 (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0);
1901 ret = lock_wim(w, w->out_fp);
1908 if (fseeko(w->out_fp, old_wim_end, SEEK_SET) != 0) {
1909 ERROR_WITH_ERRNO("Can't seek to end of WIM");
1913 return WIMLIB_ERR_WRITE;
1916 if (!list_empty(&stream_list)) {
1917 DEBUG("Writing newly added streams (offset = %"PRIu64")",
1919 ret = write_stream_list(&stream_list, w->out_fp,
1920 wimlib_get_compression_type(w),
1921 write_flags, num_threads,
1926 DEBUG("No new streams were added");
1929 found_modified_image = false;
1930 for (int i = 0; i < w->hdr.image_count; i++) {
1931 if (!found_modified_image)
1932 found_modified_image = w->image_metadata[i].modified;
1933 if (found_modified_image) {
1934 select_wim_image(w, i + 1);
1935 ret = write_metadata_resource(w);
1940 write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE;
1941 ret = finish_write(w, WIMLIB_ALL_IMAGES, write_flags,
1944 close_wim_writable(w);
1945 if (ret != 0 && !(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
1946 WARNING("Truncating `%s' to its original size (%"PRIu64" bytes)",
1947 w->filename, old_wim_end);
1948 /* Return value of truncate() is ignored because this is already
1950 (void)truncate(w->filename, old_wim_end);
1957 overwrite_wim_via_tmpfile(WIMStruct *w, int write_flags,
1958 unsigned num_threads,
1959 wimlib_progress_func_t progress_func)
1961 size_t wim_name_len;
1964 DEBUG("Overwriting `%s' via a temporary file", w->filename);
1966 /* Write the WIM to a temporary file in the same directory as the
1968 wim_name_len = strlen(w->filename);
1969 mbchar tmpfile[wim_name_len + 10];
1970 memcpy(tmpfile, w->filename, wim_name_len);
1971 randomize_char_array_with_alnum(tmpfile + wim_name_len, 9);
1972 tmpfile[wim_name_len + 9] = '\0';
1974 ret = wimlib_write(w, tmpfile, WIMLIB_ALL_IMAGES,
1975 write_flags | WIMLIB_WRITE_FLAG_FSYNC,
1976 num_threads, progress_func);
1978 ERROR("Failed to write the WIM file `%s'", tmpfile);
1982 DEBUG("Renaming `%s' to `%s'", tmpfile, w->filename);
1984 /* Rename the new file to the old file .*/
1985 if (rename(tmpfile, w->filename) != 0) {
1986 ERROR_WITH_ERRNO("Failed to rename `%s' to `%s'",
1987 tmpfile, w->filename);
1988 ret = WIMLIB_ERR_RENAME;
1992 if (progress_func) {
1993 union wimlib_progress_info progress;
1994 progress.rename.from = tmpfile;
1995 progress.rename.to = w->filename;
1996 progress_func(WIMLIB_PROGRESS_MSG_RENAME, &progress);
1999 /* Close the original WIM file that was opened for reading. */
2000 if (w->fp != NULL) {
2005 /* Re-open the WIM read-only. */
2006 w->fp = fopen(w->filename, "rb");
2007 if (w->fp == NULL) {
2008 ret = WIMLIB_ERR_REOPEN;
2009 WARNING_WITH_ERRNO("Failed to re-open `%s' read-only",
2016 /* Remove temporary file. */
2017 if (unlink(tmpfile) != 0)
2018 WARNING_WITH_ERRNO("Failed to remove `%s'", tmpfile);
2023 * Writes a WIM file to the original file that it was read from, overwriting it.
2026 wimlib_overwrite(WIMStruct *w, int write_flags,
2027 unsigned num_threads,
2028 wimlib_progress_func_t progress_func)
2030 write_flags &= WIMLIB_WRITE_MASK_PUBLIC;
2033 return WIMLIB_ERR_NO_FILENAME;
2035 if (w->hdr.total_parts != 1) {
2036 ERROR("Cannot modify a split WIM");
2037 return WIMLIB_ERR_SPLIT_UNSUPPORTED;
2040 if ((!w->deletion_occurred || (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE))
2041 && !(write_flags & WIMLIB_WRITE_FLAG_REBUILD))
2044 ret = overwrite_wim_inplace(w, write_flags, num_threads,
2046 if (ret == WIMLIB_ERR_RESOURCE_ORDER)
2047 WARNING("Falling back to re-building entire WIM");
2051 return overwrite_wim_via_tmpfile(w, write_flags, num_threads,