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"
49 #ifdef ENABLE_MULTITHREADED_COMPRESSION
58 # include <ntfs-3g/attrib.h>
59 # include <ntfs-3g/inode.h>
60 # include <ntfs-3g/dir.h>
71 #if defined(__WIN32__) && !defined(INVALID_HANDLE_VALUE)
72 # define INVALID_HANDLE_VALUE ((HANDLE)(-1))
80 fflush_and_ftruncate(FILE *fp, off_t size)
86 ERROR_WITH_ERRNO("Failed to flush data to output WIM file");
87 return WIMLIB_ERR_WRITE;
89 ret = ftruncate(fileno(fp), size);
91 ERROR_WITH_ERRNO("Failed to truncate output WIM file to "
92 "%"PRIu64" bytes", size);
93 return WIMLIB_ERR_WRITE;
98 /* Chunk table that's located at the beginning of each compressed resource in
99 * the WIM. (This is not the on-disk format; the on-disk format just has an
100 * array of offsets.) */
104 u64 original_resource_size;
105 u64 bytes_per_chunk_entry;
113 * Allocates and initializes a chunk table, and reserves space for it in the
117 begin_wim_resource_chunk_tab(const struct wim_lookup_table_entry *lte,
120 struct chunk_table **chunk_tab_ret)
122 u64 size = wim_resource_size(lte);
123 u64 num_chunks = (size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE;
124 size_t alloc_size = sizeof(struct chunk_table) + num_chunks * sizeof(u64);
125 struct chunk_table *chunk_tab = CALLOC(1, alloc_size);
129 ERROR("Failed to allocate chunk table for %"PRIu64" byte "
131 ret = WIMLIB_ERR_NOMEM;
134 chunk_tab->file_offset = file_offset;
135 chunk_tab->num_chunks = num_chunks;
136 chunk_tab->original_resource_size = size;
137 chunk_tab->bytes_per_chunk_entry = (size >= (1ULL << 32)) ? 8 : 4;
138 chunk_tab->table_disk_size = chunk_tab->bytes_per_chunk_entry *
140 chunk_tab->cur_offset = 0;
141 chunk_tab->cur_offset_p = chunk_tab->offsets;
143 if (fwrite(chunk_tab, 1, chunk_tab->table_disk_size, out_fp) !=
144 chunk_tab->table_disk_size) {
145 ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
147 ret = WIMLIB_ERR_WRITE;
153 *chunk_tab_ret = chunk_tab;
158 * Pointer to function to compresses a chunk of a WIM resource.
160 * @chunk: Uncompressed data of the chunk.
161 * @chunk_size: Size of the uncompressed chunk in bytes.
162 * @compressed_chunk: Pointer to output buffer of size at least
163 * (@chunk_size - 1) bytes.
164 * @compressed_chunk_len_ret: Pointer to an unsigned int into which the size
165 * of the compressed chunk will be
168 * Returns zero if compressed succeeded, and nonzero if the chunk could not be
169 * compressed to any smaller than @chunk_size. This function cannot fail for
172 typedef int (*compress_func_t)(const void *, unsigned, void *, unsigned *);
175 get_compress_func(int out_ctype)
177 if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX)
180 return xpress_compress;
184 * Writes a chunk of a WIM resource to an output file.
186 * @chunk: Uncompressed data of the chunk.
187 * @chunk_size: Size of the chunk (<= WIM_CHUNK_SIZE)
188 * @out_fp: FILE * to write tho chunk to.
189 * @out_ctype: Compression type to use when writing the chunk (ignored if no
190 * chunk table provided)
191 * @chunk_tab: Pointer to chunk table being created. It is updated with the
192 * offset of the chunk we write.
194 * Returns 0 on success; nonzero on failure.
197 write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size,
198 FILE *out_fp, compress_func_t compress,
199 struct chunk_table *chunk_tab)
202 unsigned out_chunk_size;
204 u8 *compressed_chunk = alloca(chunk_size);
207 ret = compress(chunk, chunk_size, compressed_chunk,
210 out_chunk = compressed_chunk;
213 out_chunk_size = chunk_size;
215 *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset;
216 chunk_tab->cur_offset += out_chunk_size;
219 out_chunk_size = chunk_size;
221 if (fwrite(out_chunk, 1, out_chunk_size, out_fp) != out_chunk_size) {
222 ERROR_WITH_ERRNO("Failed to write WIM resource chunk");
223 return WIMLIB_ERR_WRITE;
229 * Finishes a WIM chunk table and writes it to the output file at the correct
232 * The final size of the full compressed resource is returned in the
233 * @compressed_size_p.
236 finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab,
237 FILE *out_fp, u64 *compressed_size_p)
239 size_t bytes_written;
240 if (fseeko(out_fp, chunk_tab->file_offset, SEEK_SET) != 0) {
241 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of output "
242 "WIM file", chunk_tab->file_offset);
243 return WIMLIB_ERR_WRITE;
246 if (chunk_tab->bytes_per_chunk_entry == 8) {
247 array_cpu_to_le64(chunk_tab->offsets, chunk_tab->num_chunks);
249 for (u64 i = 0; i < chunk_tab->num_chunks; i++)
250 ((u32*)chunk_tab->offsets)[i] =
251 cpu_to_le32(chunk_tab->offsets[i]);
253 bytes_written = fwrite((u8*)chunk_tab->offsets +
254 chunk_tab->bytes_per_chunk_entry,
255 1, chunk_tab->table_disk_size, out_fp);
256 if (bytes_written != chunk_tab->table_disk_size) {
257 ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
259 return WIMLIB_ERR_WRITE;
261 if (fseeko(out_fp, 0, SEEK_END) != 0) {
262 ERROR_WITH_ERRNO("Failed to seek to end of output WIM file");
263 return WIMLIB_ERR_WRITE;
265 *compressed_size_p = chunk_tab->cur_offset + chunk_tab->table_disk_size;
269 /* Prepare for multiple reads to a resource by caching a FILE * or NTFS
270 * attribute pointer in the lookup table entry. */
272 prepare_resource_for_read(struct wim_lookup_table_entry *lte
275 , ntfs_inode **ni_ret
279 switch (lte->resource_location) {
280 case RESOURCE_IN_FILE_ON_DISK:
281 if (!lte->file_on_disk_fp) {
282 lte->file_on_disk_fp = tfopen(lte->file_on_disk, T("rb"));
283 if (!lte->file_on_disk_fp) {
284 ERROR_WITH_ERRNO("Failed to open the file "
285 "`%"TS"'", lte->file_on_disk);
286 return WIMLIB_ERR_OPEN;
291 case RESOURCE_IN_NTFS_VOLUME:
293 struct ntfs_location *loc = lte->ntfs_loc;
296 ni = ntfs_pathname_to_inode(*loc->ntfs_vol_p, NULL, loc->path);
298 ERROR_WITH_ERRNO("Failed to open inode `%"TS"' in NTFS "
299 "volume", loc->path);
300 return WIMLIB_ERR_NTFS_3G;
302 lte->attr = ntfs_attr_open(ni,
303 loc->is_reparse_point ? AT_REPARSE_POINT : AT_DATA,
305 loc->stream_name_nchars);
307 ERROR_WITH_ERRNO("Failed to open attribute of `%"TS"' in "
308 "NTFS volume", loc->path);
309 ntfs_inode_close(ni);
310 return WIMLIB_ERR_NTFS_3G;
318 if (lte->win32_file_on_disk_fp == INVALID_HANDLE_VALUE) {
319 lte->win32_file_on_disk_fp =
320 win32_open_file_data_only(lte->file_on_disk);
321 if (lte->win32_file_on_disk_fp == INVALID_HANDLE_VALUE) {
322 ERROR("Win32 API: Can't open %"TS, lte->file_on_disk);
324 return WIMLIB_ERR_OPEN;
335 /* Undo prepare_resource_for_read() by closing the cached FILE * or NTFS
338 end_wim_resource_read(struct wim_lookup_table_entry *lte
344 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK
345 && lte->file_on_disk_fp)
347 fclose(lte->file_on_disk_fp);
348 lte->file_on_disk_fp = NULL;
351 else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME) {
353 ntfs_attr_close(lte->attr);
357 ntfs_inode_close(ni);
361 else if (lte->resource_location == RESOURCE_WIN32
362 && lte->win32_file_on_disk_fp != INVALID_HANDLE_VALUE)
364 win32_close_file(lte->win32_file_on_disk_fp);
365 lte->win32_file_on_disk_fp = INVALID_HANDLE_VALUE;
371 write_uncompressed_resource_and_truncate(struct wim_lookup_table_entry *lte,
374 struct resource_entry *out_res_entry)
377 if (fseeko(out_fp, file_offset, SEEK_SET) != 0) {
378 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of "
379 "output WIM file", file_offset);
380 return WIMLIB_ERR_WRITE;
382 ret = write_wim_resource(lte, out_fp, WIMLIB_COMPRESSION_TYPE_NONE,
387 return fflush_and_ftruncate(out_fp,
388 file_offset + wim_resource_size(lte));
392 * Writes a WIM resource to a FILE * opened for writing. The resource may be
393 * written uncompressed or compressed depending on the @out_ctype parameter.
395 * If by chance the resource compresses to more than the original size (this may
396 * happen with random data or files than are pre-compressed), the resource is
397 * instead written uncompressed (and this is reflected in the @out_res_entry by
398 * removing the WIM_RESHDR_FLAG_COMPRESSED flag).
400 * @lte: The lookup table entry for the WIM resource.
401 * @out_fp: The FILE * to write the resource to.
402 * @out_ctype: The compression type of the resource to write. Note: if this is
403 * the same as the compression type of the WIM resource we
404 * need to read, we simply copy the data (i.e. we do not
405 * uncompress it, then compress it again).
406 * @out_res_entry: If non-NULL, a resource entry that is filled in with the
407 * offset, original size, compressed size, and compression flag
408 * of the output resource.
410 * Returns 0 on success; nonzero on failure.
413 write_wim_resource(struct wim_lookup_table_entry *lte,
414 FILE *out_fp, int out_ctype,
415 struct resource_entry *out_res_entry,
420 u64 old_compressed_size;
421 u64 new_compressed_size;
424 struct chunk_table *chunk_tab = NULL;
427 compress_func_t compress = NULL;
429 ntfs_inode *ni = NULL;
434 /* Original size of the resource */
435 original_size = wim_resource_size(lte);
437 /* Compressed size of the resource (as it exists now) */
438 old_compressed_size = wim_resource_compressed_size(lte);
440 /* Current offset in output file */
441 file_offset = ftello(out_fp);
442 if (file_offset == -1) {
443 ERROR_WITH_ERRNO("Failed to get offset in output "
445 return WIMLIB_ERR_WRITE;
448 /* Are the compression types the same? If so, do a raw copy (copy
449 * without decompressing and recompressing the data). */
450 raw = (wim_resource_compression_type(lte) == out_ctype
451 && out_ctype != WIMLIB_COMPRESSION_TYPE_NONE
452 && !(flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS));
455 flags |= WIMLIB_RESOURCE_FLAG_RAW;
456 bytes_remaining = old_compressed_size;
458 flags &= ~WIMLIB_RESOURCE_FLAG_RAW;
459 bytes_remaining = original_size;
462 /* Empty resource; nothing needs to be done, so just return success. */
463 if (bytes_remaining == 0)
466 /* Buffer for reading chunks for the resource */
467 u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
469 /* If we are writing a compressed resource and not doing a raw copy, we
470 * need to initialize the chunk table */
471 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && !raw) {
472 ret = begin_wim_resource_chunk_tab(lte, out_fp, file_offset,
478 /* If the WIM resource is in an external file, open a FILE * to it so we
479 * don't have to open a temporary one in read_wim_resource() for each
482 ret = prepare_resource_for_read(lte, &ni);
484 ret = prepare_resource_for_read(lte);
489 /* If we aren't doing a raw copy, we will compute the SHA1 message
490 * digest of the resource as we read it, and verify it's the same as the
491 * hash given in the lookup table entry once we've finished reading the
496 compress = get_compress_func(out_ctype);
500 /* While there are still bytes remaining in the WIM resource, read a
501 * chunk of the resource, update SHA1, then write that chunk using the
502 * desired compression type. */
504 u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE);
505 ret = read_wim_resource(lte, buf, to_read, offset, flags);
509 sha1_update(&ctx, buf, to_read);
510 ret = write_wim_resource_chunk(buf, to_read, out_fp,
511 compress, chunk_tab);
514 bytes_remaining -= to_read;
516 } while (bytes_remaining);
518 /* Raw copy: The new compressed size is the same as the old compressed
521 * Using WIMLIB_COMPRESSION_TYPE_NONE: The new compressed size is the
524 * Using a different compression type: Call
525 * finish_wim_resource_chunk_tab() and it will provide the new
529 new_compressed_size = old_compressed_size;
531 if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE)
532 new_compressed_size = original_size;
534 ret = finish_wim_resource_chunk_tab(chunk_tab, out_fp,
535 &new_compressed_size);
541 /* Verify SHA1 message digest of the resource, unless we are doing a raw
542 * write (in which case we never even saw the uncompressed data). Or,
543 * if the hash we had before is all 0's, just re-set it to be the new
546 u8 md[SHA1_HASH_SIZE];
547 sha1_final(md, &ctx);
548 if (is_zero_hash(lte->hash)) {
549 copy_hash(lte->hash, md);
550 } else if (!hashes_equal(md, lte->hash)) {
551 ERROR("WIM resource has incorrect hash!");
552 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK) {
553 ERROR("We were reading it from `%s'; maybe it changed "
554 "while we were reading it.",
557 ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
562 if (!raw && new_compressed_size >= original_size &&
563 out_ctype != WIMLIB_COMPRESSION_TYPE_NONE)
565 /* Oops! We compressed the resource to larger than the original
566 * size. Write the resource uncompressed instead. */
567 ret = write_uncompressed_resource_and_truncate(lte,
575 out_res_entry->size = new_compressed_size;
576 out_res_entry->original_size = original_size;
577 out_res_entry->offset = file_offset;
578 out_res_entry->flags = lte->resource_entry.flags
579 & ~WIM_RESHDR_FLAG_COMPRESSED;
580 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE)
581 out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED;
587 end_wim_resource_read(lte, ni);
589 end_wim_resource_read(lte);
596 #ifdef ENABLE_MULTITHREADED_COMPRESSION
598 /* Blocking shared queue (solves the producer-consumer problem) */
599 struct shared_queue {
603 unsigned filled_slots;
605 pthread_mutex_t lock;
606 pthread_cond_t msg_avail_cond;
607 pthread_cond_t space_avail_cond;
611 shared_queue_init(struct shared_queue *q, unsigned size)
613 wimlib_assert(size != 0);
614 q->array = CALLOC(sizeof(q->array[0]), size);
616 return WIMLIB_ERR_NOMEM;
621 pthread_mutex_init(&q->lock, NULL);
622 pthread_cond_init(&q->msg_avail_cond, NULL);
623 pthread_cond_init(&q->space_avail_cond, NULL);
628 shared_queue_destroy(struct shared_queue *q)
631 pthread_mutex_destroy(&q->lock);
632 pthread_cond_destroy(&q->msg_avail_cond);
633 pthread_cond_destroy(&q->space_avail_cond);
637 shared_queue_put(struct shared_queue *q, void *obj)
639 pthread_mutex_lock(&q->lock);
640 while (q->filled_slots == q->size)
641 pthread_cond_wait(&q->space_avail_cond, &q->lock);
643 q->back = (q->back + 1) % q->size;
644 q->array[q->back] = obj;
647 pthread_cond_broadcast(&q->msg_avail_cond);
648 pthread_mutex_unlock(&q->lock);
652 shared_queue_get(struct shared_queue *q)
656 pthread_mutex_lock(&q->lock);
657 while (q->filled_slots == 0)
658 pthread_cond_wait(&q->msg_avail_cond, &q->lock);
660 obj = q->array[q->front];
661 q->array[q->front] = NULL;
662 q->front = (q->front + 1) % q->size;
665 pthread_cond_broadcast(&q->space_avail_cond);
666 pthread_mutex_unlock(&q->lock);
670 struct compressor_thread_params {
671 struct shared_queue *res_to_compress_queue;
672 struct shared_queue *compressed_res_queue;
673 compress_func_t compress;
676 #define MAX_CHUNKS_PER_MSG 2
679 struct wim_lookup_table_entry *lte;
680 u8 *uncompressed_chunks[MAX_CHUNKS_PER_MSG];
681 u8 *out_compressed_chunks[MAX_CHUNKS_PER_MSG];
682 u8 *compressed_chunks[MAX_CHUNKS_PER_MSG];
683 unsigned uncompressed_chunk_sizes[MAX_CHUNKS_PER_MSG];
684 unsigned compressed_chunk_sizes[MAX_CHUNKS_PER_MSG];
686 struct list_head list;
692 compress_chunks(struct message *msg, compress_func_t compress)
694 for (unsigned i = 0; i < msg->num_chunks; i++) {
695 DEBUG2("compress chunk %u of %u", i, msg->num_chunks);
696 int ret = compress(msg->uncompressed_chunks[i],
697 msg->uncompressed_chunk_sizes[i],
698 msg->compressed_chunks[i],
699 &msg->compressed_chunk_sizes[i]);
701 msg->out_compressed_chunks[i] = msg->compressed_chunks[i];
703 msg->out_compressed_chunks[i] = msg->uncompressed_chunks[i];
704 msg->compressed_chunk_sizes[i] = msg->uncompressed_chunk_sizes[i];
709 /* Compressor thread routine. This is a lot simpler than the main thread
710 * routine: just repeatedly get a group of chunks from the
711 * res_to_compress_queue, compress them, and put them in the
712 * compressed_res_queue. A NULL pointer indicates that the thread should stop.
715 compressor_thread_proc(void *arg)
717 struct compressor_thread_params *params = arg;
718 struct shared_queue *res_to_compress_queue = params->res_to_compress_queue;
719 struct shared_queue *compressed_res_queue = params->compressed_res_queue;
720 compress_func_t compress = params->compress;
723 DEBUG("Compressor thread ready");
724 while ((msg = shared_queue_get(res_to_compress_queue)) != NULL) {
725 compress_chunks(msg, compress);
726 shared_queue_put(compressed_res_queue, msg);
728 DEBUG("Compressor thread terminating");
731 #endif /* ENABLE_MULTITHREADED_COMPRESSION */
734 do_write_stream_list(struct list_head *my_resources,
737 wimlib_progress_func_t progress_func,
738 union wimlib_progress_info *progress,
739 int write_resource_flags)
742 struct wim_lookup_table_entry *lte, *tmp;
744 list_for_each_entry_safe(lte, tmp, my_resources, staging_list) {
745 ret = write_wim_resource(lte,
748 <e->output_resource_entry,
749 write_resource_flags);
752 list_del(<e->staging_list);
753 progress->write_streams.completed_bytes +=
754 wim_resource_size(lte);
755 progress->write_streams.completed_streams++;
757 progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
765 write_stream_list_serial(struct list_head *stream_list,
769 wimlib_progress_func_t progress_func,
770 union wimlib_progress_info *progress)
772 int write_resource_flags;
774 if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
775 write_resource_flags = WIMLIB_RESOURCE_FLAG_RECOMPRESS;
777 write_resource_flags = 0;
778 progress->write_streams.num_threads = 1;
780 progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress);
781 return do_write_stream_list(stream_list, out_fp,
782 out_ctype, progress_func,
783 progress, write_resource_flags);
786 #ifdef ENABLE_MULTITHREADED_COMPRESSION
788 write_wim_chunks(struct message *msg, FILE *out_fp,
789 struct chunk_table *chunk_tab)
791 for (unsigned i = 0; i < msg->num_chunks; i++) {
792 unsigned chunk_csize = msg->compressed_chunk_sizes[i];
794 DEBUG2("Write wim chunk %u of %u (csize = %u)",
795 i, msg->num_chunks, chunk_csize);
797 if (fwrite(msg->out_compressed_chunks[i], 1, chunk_csize, out_fp)
800 ERROR_WITH_ERRNO("Failed to write WIM chunk");
801 return WIMLIB_ERR_WRITE;
804 *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset;
805 chunk_tab->cur_offset += chunk_csize;
811 * This function is executed by the main thread when the resources are being
812 * compressed in parallel. The main thread is in change of all reading of the
813 * uncompressed data and writing of the compressed data. The compressor threads
814 * *only* do compression from/to in-memory buffers.
816 * Each unit of work given to a compressor thread is up to MAX_CHUNKS_PER_MSG
817 * chunks of compressed data to compress, represented in a `struct message'.
818 * Each message is passed from the main thread to a worker thread through the
819 * res_to_compress_queue, and it is passed back through the
820 * compressed_res_queue.
823 main_writer_thread_proc(struct list_head *stream_list,
826 struct shared_queue *res_to_compress_queue,
827 struct shared_queue *compressed_res_queue,
830 wimlib_progress_func_t progress_func,
831 union wimlib_progress_info *progress)
834 struct chunk_table *cur_chunk_tab = NULL;
835 struct message *msgs = CALLOC(num_messages, sizeof(struct message));
836 struct wim_lookup_table_entry *next_lte = NULL;
838 // Initially, all the messages are available to use.
839 LIST_HEAD(available_msgs);
842 ret = WIMLIB_ERR_NOMEM;
846 for (size_t i = 0; i < num_messages; i++)
847 list_add(&msgs[i].list, &available_msgs);
849 // outstanding_resources is the list of resources that currently have
850 // had chunks sent off for compression.
852 // The first stream in outstanding_resources is the stream that is
853 // currently being written (cur_lte).
855 // The last stream in outstanding_resources is the stream that is
856 // currently being read and chunks fed to the compressor threads
859 // Depending on the number of threads and the sizes of the resource,
860 // the outstanding streams list may contain streams between cur_lte and
861 // next_lte that have all their chunks compressed or being compressed,
862 // but haven't been written yet.
864 LIST_HEAD(outstanding_resources);
865 struct list_head *next_resource = stream_list->next;
867 u64 next_num_chunks = 0;
869 // As in write_wim_resource(), each resource we read is checksummed.
870 SHA_CTX next_sha_ctx;
871 u8 next_hash[SHA1_HASH_SIZE];
873 // Resources that don't need any chunks compressed are added to this
874 // list and written directly by the main thread.
875 LIST_HEAD(my_resources);
877 struct wim_lookup_table_entry *cur_lte = NULL;
881 ntfs_inode *ni = NULL;
884 DEBUG("Initializing buffers for uncompressed "
885 "and compressed data (%zu bytes needed)",
886 num_messages * MAX_CHUNKS_PER_MSG * WIM_CHUNK_SIZE * 2);
888 // Pre-allocate all the buffers that will be needed to do the chunk
890 for (size_t i = 0; i < num_messages; i++) {
891 for (size_t j = 0; j < MAX_CHUNKS_PER_MSG; j++) {
892 msgs[i].compressed_chunks[j] = MALLOC(WIM_CHUNK_SIZE);
894 // The extra 8 bytes is because longest_match() in
895 // lz77.c may read a little bit off the end of the
896 // uncompressed data. It doesn't need to be
897 // initialized--- we really just need to avoid accessing
899 msgs[i].uncompressed_chunks[j] = MALLOC(WIM_CHUNK_SIZE + 8);
900 if (msgs[i].compressed_chunks[j] == NULL ||
901 msgs[i].uncompressed_chunks[j] == NULL)
903 ret = WIMLIB_ERR_NOMEM;
909 // This loop is executed until all resources have been written, except
910 // possibly a few that have been added to the @my_resources list for
913 // Send chunks to the compressor threads until either (a) there
914 // are no more messages available since they were all sent off,
915 // or (b) there are no more resources that need to be
917 while (!list_empty(&available_msgs)) {
918 if (next_chunk == next_num_chunks) {
919 // If next_chunk == next_num_chunks, there are
920 // no more chunks to write in the current
921 // stream. So, check the SHA1 message digest of
922 // the stream that was just finished (unless
923 // next_lte == NULL, which is the case the very
924 // first time this loop is entered, and also
925 // near the very end of the compression when
926 // there are no more streams.) Then, advance to
927 // the next stream (if there is one).
928 if (next_lte != NULL) {
930 end_wim_resource_read(next_lte, ni);
933 end_wim_resource_read(next_lte);
935 DEBUG2("Finalize SHA1 md (next_num_chunks=%zu)",
937 sha1_final(next_hash, &next_sha_ctx);
938 if (!hashes_equal(next_lte->hash, next_hash)) {
939 ERROR("WIM resource has incorrect hash!");
940 if (next_lte->resource_location ==
941 RESOURCE_IN_FILE_ON_DISK)
943 ERROR("We were reading it from `%"TS"'; "
944 "maybe it changed while we were "
946 next_lte->file_on_disk);
948 ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
953 // Advance to the next resource.
955 // If the next resource needs no compression, just write
956 // it with this thread (not now though--- we could be in
957 // the middle of writing another resource.) Keep doing
958 // this until we either get to the end of the resources
959 // list, or we get to a resource that needs compression.
961 if (next_resource == stream_list) {
962 // No more resources to send for
967 next_lte = container_of(next_resource,
968 struct wim_lookup_table_entry,
970 next_resource = next_resource->next;
971 if ((!(write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
972 && wim_resource_compression_type(next_lte) == out_ctype)
973 || wim_resource_size(next_lte) == 0)
975 list_add_tail(&next_lte->staging_list,
978 list_add_tail(&next_lte->staging_list,
979 &outstanding_resources);
981 next_num_chunks = wim_resource_chunks(next_lte);
982 sha1_init(&next_sha_ctx);
983 INIT_LIST_HEAD(&next_lte->msg_list);
985 ret = prepare_resource_for_read(next_lte, &ni);
987 ret = prepare_resource_for_read(next_lte);
992 if (cur_lte == NULL) {
993 // Set cur_lte for the
1002 if (next_lte == NULL) {
1003 // No more resources to send for compression
1007 // Get a message from the available messages
1009 msg = container_of(available_msgs.next,
1013 // ... and delete it from the available messages
1015 list_del(&msg->list);
1017 // Initialize the message with the chunks to
1019 msg->num_chunks = min(next_num_chunks - next_chunk,
1020 MAX_CHUNKS_PER_MSG);
1021 msg->lte = next_lte;
1022 msg->complete = false;
1023 msg->begin_chunk = next_chunk;
1025 unsigned size = WIM_CHUNK_SIZE;
1026 for (unsigned i = 0; i < msg->num_chunks; i++) {
1028 // Read chunk @next_chunk of the stream into the
1029 // message so that a compressor thread can
1032 if (next_chunk == next_num_chunks - 1) {
1033 size = MODULO_NONZERO(wim_resource_size(next_lte),
1037 DEBUG2("Read resource (size=%u, offset=%zu)",
1038 size, next_chunk * WIM_CHUNK_SIZE);
1040 msg->uncompressed_chunk_sizes[i] = size;
1042 ret = read_wim_resource(next_lte,
1043 msg->uncompressed_chunks[i],
1045 next_chunk * WIM_CHUNK_SIZE,
1049 sha1_update(&next_sha_ctx,
1050 msg->uncompressed_chunks[i], size);
1054 // Send the compression request
1055 list_add_tail(&msg->list, &next_lte->msg_list);
1056 shared_queue_put(res_to_compress_queue, msg);
1057 DEBUG2("Compression request sent");
1060 // If there are no outstanding resources, there are no more
1061 // resources that need to be written.
1062 if (list_empty(&outstanding_resources)) {
1067 // Get the next message from the queue and process it.
1068 // The message will contain 1 or more data chunks that have been
1070 msg = shared_queue_get(compressed_res_queue);
1071 msg->complete = true;
1073 // Is this the next chunk in the current resource? If it's not
1074 // (i.e., an earlier chunk in a same or different resource
1075 // hasn't been compressed yet), do nothing, and keep this
1076 // message around until all earlier chunks are received.
1078 // Otherwise, write all the chunks we can.
1079 while (cur_lte != NULL &&
1080 !list_empty(&cur_lte->msg_list) &&
1081 (msg = container_of(cur_lte->msg_list.next,
1085 DEBUG2("Complete msg (begin_chunk=%"PRIu64")", msg->begin_chunk);
1086 if (msg->begin_chunk == 0) {
1087 DEBUG2("Begin chunk tab");
1089 // This is the first set of chunks. Leave space
1090 // for the chunk table in the output file.
1091 off_t cur_offset = ftello(out_fp);
1092 if (cur_offset == -1) {
1093 ret = WIMLIB_ERR_WRITE;
1096 ret = begin_wim_resource_chunk_tab(cur_lte,
1104 // Write the compressed chunks from the message.
1105 ret = write_wim_chunks(msg, out_fp, cur_chunk_tab);
1109 list_del(&msg->list);
1111 // This message is available to use for different chunks
1113 list_add(&msg->list, &available_msgs);
1115 // Was this the last chunk of the stream? If so, finish
1117 if (list_empty(&cur_lte->msg_list) &&
1118 msg->begin_chunk + msg->num_chunks == cur_chunk_tab->num_chunks)
1120 DEBUG2("Finish wim chunk tab");
1122 ret = finish_wim_resource_chunk_tab(cur_chunk_tab,
1128 if (res_csize >= wim_resource_size(cur_lte)) {
1129 /* Oops! We compressed the resource to
1130 * larger than the original size. Write
1131 * the resource uncompressed instead. */
1132 ret = write_uncompressed_resource_and_truncate(
1135 cur_chunk_tab->file_offset,
1136 &cur_lte->output_resource_entry);
1140 cur_lte->output_resource_entry.size =
1143 cur_lte->output_resource_entry.original_size =
1144 cur_lte->resource_entry.original_size;
1146 cur_lte->output_resource_entry.offset =
1147 cur_chunk_tab->file_offset;
1149 cur_lte->output_resource_entry.flags =
1150 cur_lte->resource_entry.flags |
1151 WIM_RESHDR_FLAG_COMPRESSED;
1154 progress->write_streams.completed_bytes +=
1155 wim_resource_size(cur_lte);
1156 progress->write_streams.completed_streams++;
1158 if (progress_func) {
1159 progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
1163 FREE(cur_chunk_tab);
1164 cur_chunk_tab = NULL;
1166 struct list_head *next = cur_lte->staging_list.next;
1167 list_del(&cur_lte->staging_list);
1169 if (next == &outstanding_resources)
1172 cur_lte = container_of(cur_lte->staging_list.next,
1173 struct wim_lookup_table_entry,
1176 // Since we just finished writing a stream,
1177 // write any streams that have been added to the
1178 // my_resources list for direct writing by the
1179 // main thread (e.g. resources that don't need
1180 // to be compressed because the desired
1181 // compression type is the same as the previous
1182 // compression type).
1183 ret = do_write_stream_list(&my_resources,
1196 if (ret == WIMLIB_ERR_NOMEM) {
1197 ERROR("Could not allocate enough memory for "
1198 "multi-threaded compression");
1203 end_wim_resource_read(next_lte, ni);
1205 end_wim_resource_read(next_lte);
1210 ret = do_write_stream_list(&my_resources, out_fp,
1211 out_ctype, progress_func,
1215 size_t num_available_msgs = 0;
1216 struct list_head *cur;
1218 list_for_each(cur, &available_msgs) {
1219 num_available_msgs++;
1222 while (num_available_msgs < num_messages) {
1223 shared_queue_get(compressed_res_queue);
1224 num_available_msgs++;
1230 for (size_t i = 0; i < num_messages; i++) {
1231 for (size_t j = 0; j < MAX_CHUNKS_PER_MSG; j++) {
1232 FREE(msgs[i].compressed_chunks[j]);
1233 FREE(msgs[i].uncompressed_chunks[j]);
1239 FREE(cur_chunk_tab);
1244 get_default_num_threads()
1247 return win32_get_number_of_processors();
1249 return sysconf(_SC_NPROCESSORS_ONLN);
1254 write_stream_list_parallel(struct list_head *stream_list,
1258 unsigned num_threads,
1259 wimlib_progress_func_t progress_func,
1260 union wimlib_progress_info *progress)
1263 struct shared_queue res_to_compress_queue;
1264 struct shared_queue compressed_res_queue;
1265 pthread_t *compressor_threads = NULL;
1267 if (num_threads == 0) {
1268 long nthreads = get_default_num_threads();
1269 if (nthreads < 1 || nthreads > UINT_MAX) {
1270 WARNING("Could not determine number of processors! Assuming 1");
1273 num_threads = nthreads;
1277 progress->write_streams.num_threads = num_threads;
1278 wimlib_assert(stream_list->next != stream_list);
1280 static const double MESSAGES_PER_THREAD = 2.0;
1281 size_t queue_size = (size_t)(num_threads * MESSAGES_PER_THREAD);
1283 DEBUG("Initializing shared queues (queue_size=%zu)", queue_size);
1285 ret = shared_queue_init(&res_to_compress_queue, queue_size);
1289 ret = shared_queue_init(&compressed_res_queue, queue_size);
1291 goto out_destroy_res_to_compress_queue;
1293 struct compressor_thread_params params;
1294 params.res_to_compress_queue = &res_to_compress_queue;
1295 params.compressed_res_queue = &compressed_res_queue;
1296 params.compress = get_compress_func(out_ctype);
1298 compressor_threads = MALLOC(num_threads * sizeof(pthread_t));
1299 if (!compressor_threads) {
1300 ret = WIMLIB_ERR_NOMEM;
1301 goto out_destroy_compressed_res_queue;
1304 for (unsigned i = 0; i < num_threads; i++) {
1305 DEBUG("pthread_create thread %u", i);
1306 ret = pthread_create(&compressor_threads[i], NULL,
1307 compressor_thread_proc, ¶ms);
1310 ERROR_WITH_ERRNO("Failed to create compressor "
1318 progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress);
1320 ret = main_writer_thread_proc(stream_list,
1323 &res_to_compress_queue,
1324 &compressed_res_queue,
1330 for (unsigned i = 0; i < num_threads; i++)
1331 shared_queue_put(&res_to_compress_queue, NULL);
1333 for (unsigned i = 0; i < num_threads; i++) {
1334 if (pthread_join(compressor_threads[i], NULL)) {
1335 WARNING_WITH_ERRNO("Failed to join compressor "
1339 FREE(compressor_threads);
1340 out_destroy_compressed_res_queue:
1341 shared_queue_destroy(&compressed_res_queue);
1342 out_destroy_res_to_compress_queue:
1343 shared_queue_destroy(&res_to_compress_queue);
1344 if (ret >= 0 && ret != WIMLIB_ERR_NOMEM)
1347 WARNING("Falling back to single-threaded compression");
1348 return write_stream_list_serial(stream_list,
1359 * Write a list of streams to a WIM (@out_fp) using the compression type
1360 * @out_ctype and up to @num_threads compressor threads.
1363 write_stream_list(struct list_head *stream_list, FILE *out_fp,
1364 int out_ctype, int write_flags,
1365 unsigned num_threads,
1366 wimlib_progress_func_t progress_func)
1368 struct wim_lookup_table_entry *lte;
1369 size_t num_streams = 0;
1370 u64 total_bytes = 0;
1371 u64 total_compression_bytes = 0;
1372 union wimlib_progress_info progress;
1374 list_for_each_entry(lte, stream_list, staging_list) {
1376 total_bytes += wim_resource_size(lte);
1377 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE
1378 && (wim_resource_compression_type(lte) != out_ctype ||
1379 (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)))
1381 total_compression_bytes += wim_resource_size(lte);
1384 progress.write_streams.total_bytes = total_bytes;
1385 progress.write_streams.total_streams = num_streams;
1386 progress.write_streams.completed_bytes = 0;
1387 progress.write_streams.completed_streams = 0;
1388 progress.write_streams.num_threads = num_threads;
1389 progress.write_streams.compression_type = out_ctype;
1391 #ifdef ENABLE_MULTITHREADED_COMPRESSION
1392 if (total_compression_bytes >= 1000000 && num_threads != 1)
1393 return write_stream_list_parallel(stream_list,
1402 return write_stream_list_serial(stream_list,
1410 struct lte_overwrite_prepare_args {
1413 struct list_head *stream_list;
1417 lte_overwrite_prepare(struct wim_lookup_table_entry *lte, void *arg)
1419 struct lte_overwrite_prepare_args *args = arg;
1421 if (lte->resource_location == RESOURCE_IN_WIM &&
1422 lte->wim == args->wim &&
1423 lte->resource_entry.offset + lte->resource_entry.size > args->end_offset)
1425 #ifdef ENABLE_ERROR_MESSAGES
1426 ERROR("The following resource is after the XML data:");
1427 print_lookup_table_entry(lte, stderr);
1429 return WIMLIB_ERR_RESOURCE_ORDER;
1432 lte->out_refcnt = lte->refcnt;
1433 memcpy(<e->output_resource_entry, <e->resource_entry,
1434 sizeof(struct resource_entry));
1435 if (!(lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA)) {
1436 wimlib_assert(lte->resource_location != RESOURCE_NONEXISTENT);
1437 if (lte->resource_location != RESOURCE_IN_WIM || lte->wim != args->wim)
1438 list_add(<e->staging_list, args->stream_list);
1444 wim_find_new_streams(WIMStruct *wim, off_t end_offset,
1445 struct list_head *stream_list)
1447 struct lte_overwrite_prepare_args args = {
1449 .end_offset = end_offset,
1450 .stream_list = stream_list,
1453 return for_lookup_table_entry(wim->lookup_table,
1454 lte_overwrite_prepare, &args);
1458 inode_find_streams_to_write(struct wim_inode *inode,
1459 struct wim_lookup_table *table,
1460 struct list_head *stream_list)
1462 struct wim_lookup_table_entry *lte;
1463 for (unsigned i = 0; i <= inode->i_num_ads; i++) {
1464 lte = inode_stream_lte(inode, i, table);
1466 if (lte->out_refcnt == 0)
1467 list_add_tail(<e->staging_list, stream_list);
1468 lte->out_refcnt += inode->i_nlink;
1475 image_find_streams_to_write(WIMStruct *w)
1477 struct wim_inode *inode;
1478 struct hlist_node *cur;
1479 struct hlist_head *inode_list;
1481 inode_list = &wim_get_current_image_metadata(w)->inode_list;
1482 hlist_for_each_entry(inode, cur, inode_list, i_hlist) {
1483 inode_find_streams_to_write(inode, w->lookup_table,
1484 (struct list_head*)w->private);
1490 write_wim_streams(WIMStruct *w, int image, int write_flags,
1491 unsigned num_threads,
1492 wimlib_progress_func_t progress_func)
1495 for_lookup_table_entry(w->lookup_table, lte_zero_out_refcnt, NULL);
1496 LIST_HEAD(stream_list);
1497 w->private = &stream_list;
1498 for_image(w, image, image_find_streams_to_write);
1499 return write_stream_list(&stream_list, w->out_fp,
1500 wimlib_get_compression_type(w), write_flags,
1501 num_threads, progress_func);
1505 * Finish writing a WIM file: write the lookup table, xml data, and integrity
1506 * table (optional), then overwrite the WIM header.
1508 * write_flags is a bitwise OR of the following:
1510 * (public) WIMLIB_WRITE_FLAG_CHECK_INTEGRITY:
1511 * Include an integrity table.
1513 * (public) WIMLIB_WRITE_FLAG_SHOW_PROGRESS:
1514 * Show progress information when (if) writing the integrity table.
1516 * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE:
1517 * Don't write the lookup table.
1519 * (private) WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE:
1520 * When (if) writing the integrity table, re-use entries from the
1521 * existing integrity table, if possible.
1523 * (private) WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML:
1524 * After writing the XML data but before writing the integrity
1525 * table, write a temporary WIM header and flush the stream so that
1526 * the WIM is less likely to become corrupted upon abrupt program
1529 * (private) WIMLIB_WRITE_FLAG_FSYNC:
1530 * fsync() the output file before closing it.
1534 finish_write(WIMStruct *w, int image, int write_flags,
1535 wimlib_progress_func_t progress_func)
1538 struct wim_header hdr;
1539 FILE *out = w->out_fp;
1541 /* @hdr will be the header for the new WIM. First copy all the data
1542 * from the header in the WIMStruct; then set all the fields that may
1543 * have changed, including the resource entries, boot index, and image
1545 memcpy(&hdr, &w->hdr, sizeof(struct wim_header));
1547 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
1548 ret = write_lookup_table(w->lookup_table, out, &hdr.lookup_table_res_entry);
1553 ret = write_xml_data(w->wim_info, image, out,
1554 (write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE) ?
1555 wim_info_get_total_bytes(w->wim_info) : 0,
1556 &hdr.xml_res_entry);
1560 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
1561 if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) {
1562 struct wim_header checkpoint_hdr;
1563 memcpy(&checkpoint_hdr, &hdr, sizeof(struct wim_header));
1564 memset(&checkpoint_hdr.integrity, 0, sizeof(struct resource_entry));
1565 if (fseeko(out, 0, SEEK_SET) != 0) {
1566 ERROR_WITH_ERRNO("Failed to seek to beginning "
1567 "of WIM being written");
1568 ret = WIMLIB_ERR_WRITE;
1571 ret = write_header(&checkpoint_hdr, out);
1575 if (fflush(out) != 0) {
1576 ERROR_WITH_ERRNO("Can't write data to WIM");
1577 ret = WIMLIB_ERR_WRITE;
1581 if (fseeko(out, 0, SEEK_END) != 0) {
1582 ERROR_WITH_ERRNO("Failed to seek to end "
1583 "of WIM being written");
1584 ret = WIMLIB_ERR_WRITE;
1589 off_t old_lookup_table_end;
1590 off_t new_lookup_table_end;
1591 if (write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE) {
1592 old_lookup_table_end = w->hdr.lookup_table_res_entry.offset +
1593 w->hdr.lookup_table_res_entry.size;
1595 old_lookup_table_end = 0;
1597 new_lookup_table_end = hdr.lookup_table_res_entry.offset +
1598 hdr.lookup_table_res_entry.size;
1600 ret = write_integrity_table(out,
1602 new_lookup_table_end,
1603 old_lookup_table_end,
1608 memset(&hdr.integrity, 0, sizeof(struct resource_entry));
1612 * In the WIM header, there is room for the resource entry for a
1613 * metadata resource labeled as the "boot metadata". This entry should
1614 * be zeroed out if there is no bootable image (boot_idx 0). Otherwise,
1615 * it should be a copy of the resource entry for the image that is
1616 * marked as bootable. This is not well documented...
1619 /* Set image count and boot index correctly for single image writes */
1620 if (image != WIMLIB_ALL_IMAGES) {
1621 hdr.image_count = 1;
1622 if (hdr.boot_idx == image)
1628 if (hdr.boot_idx == 0) {
1629 memset(&hdr.boot_metadata_res_entry, 0,
1630 sizeof(struct resource_entry));
1632 memcpy(&hdr.boot_metadata_res_entry,
1634 hdr.boot_idx - 1].metadata_lte->output_resource_entry,
1635 sizeof(struct resource_entry));
1638 if (fseeko(out, 0, SEEK_SET) != 0) {
1639 ERROR_WITH_ERRNO("Failed to seek to beginning of WIM "
1641 ret = WIMLIB_ERR_WRITE;
1645 ret = write_header(&hdr, out);
1649 if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) {
1650 if (fflush(out) != 0
1651 || fsync(fileno(out)) != 0)
1653 ERROR_WITH_ERRNO("Error flushing data to WIM file");
1654 ret = WIMLIB_ERR_WRITE;
1658 if (fclose(out) != 0) {
1659 ERROR_WITH_ERRNO("Failed to close the WIM file");
1661 ret = WIMLIB_ERR_WRITE;
1667 #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK)
1669 lock_wim(WIMStruct *w, FILE *fp)
1672 if (fp && !w->wim_locked) {
1673 ret = flock(fileno(fp), LOCK_EX | LOCK_NB);
1675 if (errno == EWOULDBLOCK) {
1676 ERROR("`%"TS"' is already being modified or has been "
1677 "mounted read-write\n"
1678 " by another process!", w->filename);
1679 ret = WIMLIB_ERR_ALREADY_LOCKED;
1681 WARNING_WITH_ERRNO("Failed to lock `%"TS"'",
1694 open_wim_writable(WIMStruct *w, const tchar *path,
1695 bool trunc, bool readable)
1706 wimlib_assert(w->out_fp == NULL);
1707 w->out_fp = tfopen(path, mode);
1711 ERROR_WITH_ERRNO("Failed to open `%"TS"' for writing", path);
1712 return WIMLIB_ERR_OPEN;
1718 close_wim_writable(WIMStruct *w)
1721 if (fclose(w->out_fp) != 0) {
1722 WARNING_WITH_ERRNO("Failed to close output WIM");
1728 /* Open file stream and write dummy header for WIM. */
1730 begin_write(WIMStruct *w, const tchar *path, int write_flags)
1733 ret = open_wim_writable(w, path, true,
1734 (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0);
1737 /* Write dummy header. It will be overwritten later. */
1738 return write_header(&w->hdr, w->out_fp);
1741 /* Writes a stand-alone WIM to a file. */
1743 wimlib_write(WIMStruct *w, const tchar *path,
1744 int image, int write_flags, unsigned num_threads,
1745 wimlib_progress_func_t progress_func)
1750 return WIMLIB_ERR_INVALID_PARAM;
1752 write_flags &= WIMLIB_WRITE_MASK_PUBLIC;
1754 if (image != WIMLIB_ALL_IMAGES &&
1755 (image < 1 || image > w->hdr.image_count))
1756 return WIMLIB_ERR_INVALID_IMAGE;
1758 if (w->hdr.total_parts != 1) {
1759 ERROR("Cannot call wimlib_write() on part of a split WIM");
1760 return WIMLIB_ERR_SPLIT_UNSUPPORTED;
1763 ret = begin_write(w, path, write_flags);
1767 ret = write_wim_streams(w, image, write_flags, num_threads,
1773 progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN, NULL);
1775 ret = for_image(w, image, write_metadata_resource);
1780 progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_END, NULL);
1782 ret = finish_write(w, image, write_flags, progress_func);
1784 close_wim_writable(w);
1785 DEBUG("wimlib_write(path=%"TS") = %d", path, ret);
1790 any_images_modified(WIMStruct *w)
1792 for (int i = 0; i < w->hdr.image_count; i++)
1793 if (w->image_metadata[i].modified)
1799 * Overwrite a WIM, possibly appending streams to it.
1801 * A WIM looks like (or is supposed to look like) the following:
1803 * Header (212 bytes)
1804 * Streams and metadata resources (variable size)
1805 * Lookup table (variable size)
1806 * XML data (variable size)
1807 * Integrity table (optional) (variable size)
1809 * If we are not adding any streams or metadata resources, the lookup table is
1810 * unchanged--- so we only need to overwrite the XML data, integrity table, and
1811 * header. This operation is potentially unsafe if the program is abruptly
1812 * terminated while the XML data or integrity table are being overwritten, but
1813 * before the new header has been written. To partially alleviate this problem,
1814 * a special flag (WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) is passed to
1815 * finish_write() to cause a temporary WIM header to be written after the XML
1816 * data has been written. This may prevent the WIM from becoming corrupted if
1817 * the program is terminated while the integrity table is being calculated (but
1818 * no guarantees, due to write re-ordering...).
1820 * If we are adding new streams or images (metadata resources), the lookup table
1821 * needs to be changed, and those streams need to be written. In this case, we
1822 * try to perform a safe update of the WIM file by writing the streams *after*
1823 * the end of the previous WIM, then writing the new lookup table, XML data, and
1824 * (optionally) integrity table following the new streams. This will produce a
1825 * layout like the following:
1827 * Header (212 bytes)
1828 * (OLD) Streams and metadata resources (variable size)
1829 * (OLD) Lookup table (variable size)
1830 * (OLD) XML data (variable size)
1831 * (OLD) Integrity table (optional) (variable size)
1832 * (NEW) Streams and metadata resources (variable size)
1833 * (NEW) Lookup table (variable size)
1834 * (NEW) XML data (variable size)
1835 * (NEW) Integrity table (optional) (variable size)
1837 * At all points, the WIM is valid as nothing points to the new data yet. Then,
1838 * the header is overwritten to point to the new lookup table, XML data, and
1839 * integrity table, to produce the following layout:
1841 * Header (212 bytes)
1842 * Streams and metadata resources (variable size)
1843 * Nothing (variable size)
1844 * More Streams and metadata resources (variable size)
1845 * Lookup table (variable size)
1846 * XML data (variable size)
1847 * Integrity table (optional) (variable size)
1849 * This method allows an image to be appended to a large WIM very quickly, and
1850 * is is crash-safe except in the case of write re-ordering, but the
1851 * disadvantage is that a small hole is left in the WIM where the old lookup
1852 * table, xml data, and integrity table were. (These usually only take up a
1853 * small amount of space compared to the streams, however.)
1856 overwrite_wim_inplace(WIMStruct *w, int write_flags,
1857 unsigned num_threads,
1858 wimlib_progress_func_t progress_func)
1861 struct list_head stream_list;
1863 bool found_modified_image;
1865 DEBUG("Overwriting `%"TS"' in-place", w->filename);
1867 /* Make sure that the integrity table (if present) is after the XML
1868 * data, and that there are no stream resources, metadata resources, or
1869 * lookup tables after the XML data. Otherwise, these data would be
1871 if (w->hdr.integrity.offset != 0 &&
1872 w->hdr.integrity.offset < w->hdr.xml_res_entry.offset) {
1873 ERROR("Didn't expect the integrity table to be before the XML data");
1874 return WIMLIB_ERR_RESOURCE_ORDER;
1877 if (w->hdr.lookup_table_res_entry.offset > w->hdr.xml_res_entry.offset) {
1878 ERROR("Didn't expect the lookup table to be after the XML data");
1879 return WIMLIB_ERR_RESOURCE_ORDER;
1883 if (w->hdr.integrity.offset)
1884 old_wim_end = w->hdr.integrity.offset + w->hdr.integrity.size;
1886 old_wim_end = w->hdr.xml_res_entry.offset + w->hdr.xml_res_entry.size;
1888 if (!w->deletion_occurred && !any_images_modified(w)) {
1889 /* If no images have been modified and no images have been
1890 * deleted, a new lookup table does not need to be written. */
1891 old_wim_end = w->hdr.lookup_table_res_entry.offset +
1892 w->hdr.lookup_table_res_entry.size;
1893 write_flags |= WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE |
1894 WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML;
1896 INIT_LIST_HEAD(&stream_list);
1897 ret = wim_find_new_streams(w, old_wim_end, &stream_list);
1901 ret = open_wim_writable(w, w->filename, false,
1902 (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0);
1906 ret = lock_wim(w, w->out_fp);
1913 if (fseeko(w->out_fp, old_wim_end, SEEK_SET) != 0) {
1914 ERROR_WITH_ERRNO("Can't seek to end of WIM");
1918 return WIMLIB_ERR_WRITE;
1921 if (!list_empty(&stream_list)) {
1922 DEBUG("Writing newly added streams (offset = %"PRIu64")",
1924 ret = write_stream_list(&stream_list, w->out_fp,
1925 wimlib_get_compression_type(w),
1926 write_flags, num_threads,
1931 DEBUG("No new streams were added");
1934 found_modified_image = false;
1935 for (int i = 0; i < w->hdr.image_count; i++) {
1936 if (!found_modified_image)
1937 found_modified_image = w->image_metadata[i].modified;
1938 if (found_modified_image) {
1939 select_wim_image(w, i + 1);
1940 ret = write_metadata_resource(w);
1945 write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE;
1946 ret = finish_write(w, WIMLIB_ALL_IMAGES, write_flags,
1949 close_wim_writable(w);
1950 if (ret != 0 && !(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
1951 WARNING("Truncating `%"TS"' to its original size (%"PRIu64" bytes)",
1952 w->filename, old_wim_end);
1953 /* Return value of truncate() is ignored because this is already
1955 (void)ttruncate(w->filename, old_wim_end);
1962 overwrite_wim_via_tmpfile(WIMStruct *w, int write_flags,
1963 unsigned num_threads,
1964 wimlib_progress_func_t progress_func)
1966 size_t wim_name_len;
1969 DEBUG("Overwriting `%"TS"' via a temporary file", w->filename);
1971 /* Write the WIM to a temporary file in the same directory as the
1973 wim_name_len = tstrlen(w->filename);
1974 tchar tmpfile[wim_name_len + 10];
1975 tmemcpy(tmpfile, w->filename, wim_name_len);
1976 randomize_char_array_with_alnum(tmpfile + wim_name_len, 9);
1977 tmpfile[wim_name_len + 9] = T('\0');
1979 ret = wimlib_write(w, tmpfile, WIMLIB_ALL_IMAGES,
1980 write_flags | WIMLIB_WRITE_FLAG_FSYNC,
1981 num_threads, progress_func);
1983 ERROR("Failed to write the WIM file `%"TS"'", tmpfile);
1987 DEBUG("Renaming `%"TS"' to `%"TS"'", tmpfile, w->filename);
1990 /* Windows won't let you delete open files unless FILE_SHARE_DELETE was
1991 * specified to CreateFile(). The WIM was opened with fopen(), which
1992 * didn't provided this flag to CreateFile, so the handle must be closed
1993 * before executing the rename(). */
1994 if (w->fp != NULL) {
2000 /* Rename the new file to the old file .*/
2001 if (trename(tmpfile, w->filename) != 0) {
2002 ERROR_WITH_ERRNO("Failed to rename `%"TS"' to `%"TS"'",
2003 tmpfile, w->filename);
2004 ret = WIMLIB_ERR_RENAME;
2008 if (progress_func) {
2009 union wimlib_progress_info progress;
2010 progress.rename.from = tmpfile;
2011 progress.rename.to = w->filename;
2012 progress_func(WIMLIB_PROGRESS_MSG_RENAME, &progress);
2015 /* Close the original WIM file that was opened for reading. */
2016 if (w->fp != NULL) {
2021 /* Re-open the WIM read-only. */
2022 w->fp = tfopen(w->filename, T("rb"));
2023 if (w->fp == NULL) {
2024 ret = WIMLIB_ERR_REOPEN;
2025 WARNING_WITH_ERRNO("Failed to re-open `%"TS"' read-only",
2032 /* Remove temporary file. */
2033 if (tunlink(tmpfile) != 0)
2034 WARNING_WITH_ERRNO("Failed to remove `%"TS"'", tmpfile);
2039 * Writes a WIM file to the original file that it was read from, overwriting it.
2042 wimlib_overwrite(WIMStruct *w, int write_flags,
2043 unsigned num_threads,
2044 wimlib_progress_func_t progress_func)
2046 write_flags &= WIMLIB_WRITE_MASK_PUBLIC;
2049 return WIMLIB_ERR_NO_FILENAME;
2051 if (w->hdr.total_parts != 1) {
2052 ERROR("Cannot modify a split WIM");
2053 return WIMLIB_ERR_SPLIT_UNSUPPORTED;
2056 if ((!w->deletion_occurred || (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE))
2057 && !(write_flags & WIMLIB_WRITE_FLAG_REBUILD))
2060 ret = overwrite_wim_inplace(w, write_flags, num_threads,
2062 if (ret == WIMLIB_ERR_RESOURCE_ORDER)
2063 WARNING("Falling back to re-building entire WIM");
2067 return overwrite_wim_via_tmpfile(w, write_flags, num_threads,