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 static int fflush_and_ftruncate(FILE *fp, off_t size)
76 ERROR_WITH_ERRNO("Failed to flush data to output WIM file");
77 return WIMLIB_ERR_WRITE;
79 ret = ftruncate(fileno(fp), size);
81 ERROR_WITH_ERRNO("Failed to truncate output WIM file to "
82 "%"PRIu64" bytes", size);
83 return WIMLIB_ERR_WRITE;
88 /* Chunk table that's located at the beginning of each compressed resource in
89 * the WIM. (This is not the on-disk format; the on-disk format just has an
90 * array of offsets.) */
94 u64 original_resource_size;
95 u64 bytes_per_chunk_entry;
103 * Allocates and initializes a chunk table, and reserves space for it in the
107 begin_wim_resource_chunk_tab(const struct wim_lookup_table_entry *lte,
110 struct chunk_table **chunk_tab_ret)
112 u64 size = wim_resource_size(lte);
113 u64 num_chunks = (size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE;
114 size_t alloc_size = sizeof(struct chunk_table) + num_chunks * sizeof(u64);
115 struct chunk_table *chunk_tab = CALLOC(1, alloc_size);
119 ERROR("Failed to allocate chunk table for %"PRIu64" byte "
121 ret = WIMLIB_ERR_NOMEM;
124 chunk_tab->file_offset = file_offset;
125 chunk_tab->num_chunks = num_chunks;
126 chunk_tab->original_resource_size = size;
127 chunk_tab->bytes_per_chunk_entry = (size >= (1ULL << 32)) ? 8 : 4;
128 chunk_tab->table_disk_size = chunk_tab->bytes_per_chunk_entry *
130 chunk_tab->cur_offset = 0;
131 chunk_tab->cur_offset_p = chunk_tab->offsets;
133 if (fwrite(chunk_tab, 1, chunk_tab->table_disk_size, out_fp) !=
134 chunk_tab->table_disk_size) {
135 ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
137 ret = WIMLIB_ERR_WRITE;
143 *chunk_tab_ret = chunk_tab;
148 * Pointer to function to compresses a chunk of a WIM resource.
150 * @chunk: Uncompressed data of the chunk.
151 * @chunk_size: Size of the uncompressed chunk in bytes.
152 * @compressed_chunk: Pointer to output buffer of size at least
153 * (@chunk_size - 1) bytes.
154 * @compressed_chunk_len_ret: Pointer to an unsigned int into which the size
155 * of the compressed chunk will be
158 * Returns zero if compressed succeeded, and nonzero if the chunk could not be
159 * compressed to any smaller than @chunk_size. This function cannot fail for
162 typedef int (*compress_func_t)(const void *, unsigned, void *, unsigned *);
164 compress_func_t get_compress_func(int out_ctype)
166 if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX)
169 return xpress_compress;
173 * Writes a chunk of a WIM resource to an output file.
175 * @chunk: Uncompressed data of the chunk.
176 * @chunk_size: Size of the chunk (<= WIM_CHUNK_SIZE)
177 * @out_fp: FILE * to write tho chunk to.
178 * @out_ctype: Compression type to use when writing the chunk (ignored if no
179 * chunk table provided)
180 * @chunk_tab: Pointer to chunk table being created. It is updated with the
181 * offset of the chunk we write.
183 * Returns 0 on success; nonzero on failure.
185 static int write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size,
186 FILE *out_fp, compress_func_t compress,
187 struct chunk_table *chunk_tab)
190 unsigned out_chunk_size;
192 u8 *compressed_chunk = alloca(chunk_size);
195 ret = compress(chunk, chunk_size, compressed_chunk,
198 out_chunk = compressed_chunk;
201 out_chunk_size = chunk_size;
203 *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset;
204 chunk_tab->cur_offset += out_chunk_size;
207 out_chunk_size = chunk_size;
209 if (fwrite(out_chunk, 1, out_chunk_size, out_fp) != out_chunk_size) {
210 ERROR_WITH_ERRNO("Failed to write WIM resource chunk");
211 return WIMLIB_ERR_WRITE;
217 * Finishes a WIM chunk table and writes it to the output file at the correct
220 * The final size of the full compressed resource is returned in the
221 * @compressed_size_p.
224 finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab,
225 FILE *out_fp, u64 *compressed_size_p)
227 size_t bytes_written;
228 if (fseeko(out_fp, chunk_tab->file_offset, SEEK_SET) != 0) {
229 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of output "
230 "WIM file", chunk_tab->file_offset);
231 return WIMLIB_ERR_WRITE;
234 if (chunk_tab->bytes_per_chunk_entry == 8) {
235 array_cpu_to_le64(chunk_tab->offsets, chunk_tab->num_chunks);
237 for (u64 i = 0; i < chunk_tab->num_chunks; i++)
238 ((u32*)chunk_tab->offsets)[i] =
239 cpu_to_le32(chunk_tab->offsets[i]);
241 bytes_written = fwrite((u8*)chunk_tab->offsets +
242 chunk_tab->bytes_per_chunk_entry,
243 1, chunk_tab->table_disk_size, out_fp);
244 if (bytes_written != chunk_tab->table_disk_size) {
245 ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
247 return WIMLIB_ERR_WRITE;
249 if (fseeko(out_fp, 0, SEEK_END) != 0) {
250 ERROR_WITH_ERRNO("Failed to seek to end of output WIM file");
251 return WIMLIB_ERR_WRITE;
253 *compressed_size_p = chunk_tab->cur_offset + chunk_tab->table_disk_size;
257 /* Prepare for multiple reads to a resource by caching a FILE * or NTFS
258 * attribute pointer in the lookup table entry. */
259 static int prepare_resource_for_read(struct wim_lookup_table_entry *lte
262 , ntfs_inode **ni_ret
266 switch (lte->resource_location) {
267 case RESOURCE_IN_FILE_ON_DISK:
268 if (!lte->file_on_disk_fp) {
269 lte->file_on_disk_fp = fopen(lte->file_on_disk, "rb");
270 if (!lte->file_on_disk_fp) {
271 ERROR_WITH_ERRNO("Failed to open the file "
272 "`%s'", lte->file_on_disk);
273 return WIMLIB_ERR_OPEN;
278 case RESOURCE_IN_NTFS_VOLUME:
280 struct ntfs_location *loc = lte->ntfs_loc;
283 ni = ntfs_pathname_to_inode(*loc->ntfs_vol_p, NULL, loc->path_utf8);
285 ERROR_WITH_ERRNO("Failed to open inode `%s' in NTFS "
286 "volume", loc->path_utf8);
287 return WIMLIB_ERR_NTFS_3G;
289 lte->attr = ntfs_attr_open(ni,
290 loc->is_reparse_point ? AT_REPARSE_POINT : AT_DATA,
291 (ntfschar*)loc->stream_name_utf16,
292 loc->stream_name_utf16_num_chars);
294 ERROR_WITH_ERRNO("Failed to open attribute of `%s' in "
295 "NTFS volume", loc->path_utf8);
296 ntfs_inode_close(ni);
297 return WIMLIB_ERR_NTFS_3G;
305 if (!lte->file_on_disk_fp) {
306 lte->file_on_disk_fp = win32_open_file_readonly(lte->file_on_disk);
307 if (!lte->file_on_disk_fp)
308 return WIMLIB_ERR_OPEN;
318 /* Undo prepare_resource_for_read() by closing the cached FILE * or NTFS
320 static void end_wim_resource_read(struct wim_lookup_table_entry *lte
326 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK
327 && lte->file_on_disk_fp)
329 fclose(lte->file_on_disk_fp);
330 lte->file_on_disk_fp = NULL;
333 else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME) {
335 ntfs_attr_close(lte->attr);
339 ntfs_inode_close(ni);
343 else if (lte->resource_location == RESOURCE_WIN32
344 && lte->file_on_disk_fp)
346 win32_close_file(lte->file_on_disk_fp);
347 lte->file_on_disk_fp = NULL;
353 write_uncompressed_resource_and_truncate(struct wim_lookup_table_entry *lte,
356 struct resource_entry *out_res_entry)
359 if (fseeko(out_fp, file_offset, SEEK_SET) != 0) {
360 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of "
361 "output WIM file", file_offset);
362 return WIMLIB_ERR_WRITE;
364 ret = write_wim_resource(lte, out_fp, WIMLIB_COMPRESSION_TYPE_NONE,
369 return fflush_and_ftruncate(out_fp,
370 file_offset + wim_resource_size(lte));
374 * Writes a WIM resource to a FILE * opened for writing. The resource may be
375 * written uncompressed or compressed depending on the @out_ctype parameter.
377 * If by chance the resource compresses to more than the original size (this may
378 * happen with random data or files than are pre-compressed), the resource is
379 * instead written uncompressed (and this is reflected in the @out_res_entry by
380 * removing the WIM_RESHDR_FLAG_COMPRESSED flag).
382 * @lte: The lookup table entry for the WIM resource.
383 * @out_fp: The FILE * to write the resource to.
384 * @out_ctype: The compression type of the resource to write. Note: if this is
385 * the same as the compression type of the WIM resource we
386 * need to read, we simply copy the data (i.e. we do not
387 * uncompress it, then compress it again).
388 * @out_res_entry: If non-NULL, a resource entry that is filled in with the
389 * offset, original size, compressed size, and compression flag
390 * of the output resource.
392 * Returns 0 on success; nonzero on failure.
394 int write_wim_resource(struct wim_lookup_table_entry *lte,
395 FILE *out_fp, int out_ctype,
396 struct resource_entry *out_res_entry,
401 u64 old_compressed_size;
402 u64 new_compressed_size;
405 struct chunk_table *chunk_tab = NULL;
408 compress_func_t compress = NULL;
410 ntfs_inode *ni = NULL;
415 /* Original size of the resource */
416 original_size = wim_resource_size(lte);
418 /* Compressed size of the resource (as it exists now) */
419 old_compressed_size = wim_resource_compressed_size(lte);
421 /* Current offset in output file */
422 file_offset = ftello(out_fp);
423 if (file_offset == -1) {
424 ERROR_WITH_ERRNO("Failed to get offset in output "
426 return WIMLIB_ERR_WRITE;
429 /* Are the compression types the same? If so, do a raw copy (copy
430 * without decompressing and recompressing the data). */
431 raw = (wim_resource_compression_type(lte) == out_ctype
432 && out_ctype != WIMLIB_COMPRESSION_TYPE_NONE
433 && !(flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS));
436 flags |= WIMLIB_RESOURCE_FLAG_RAW;
437 bytes_remaining = old_compressed_size;
439 flags &= ~WIMLIB_RESOURCE_FLAG_RAW;
440 bytes_remaining = original_size;
443 /* Empty resource; nothing needs to be done, so just return success. */
444 if (bytes_remaining == 0)
447 /* Buffer for reading chunks for the resource */
448 u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
450 /* If we are writing a compressed resource and not doing a raw copy, we
451 * need to initialize the chunk table */
452 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && !raw) {
453 ret = begin_wim_resource_chunk_tab(lte, out_fp, file_offset,
459 /* If the WIM resource is in an external file, open a FILE * to it so we
460 * don't have to open a temporary one in read_wim_resource() for each
463 ret = prepare_resource_for_read(lte, &ni);
465 ret = prepare_resource_for_read(lte);
470 /* If we aren't doing a raw copy, we will compute the SHA1 message
471 * digest of the resource as we read it, and verify it's the same as the
472 * hash given in the lookup table entry once we've finished reading the
477 compress = get_compress_func(out_ctype);
481 /* While there are still bytes remaining in the WIM resource, read a
482 * chunk of the resource, update SHA1, then write that chunk using the
483 * desired compression type. */
485 u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE);
486 ret = read_wim_resource(lte, buf, to_read, offset, flags);
490 sha1_update(&ctx, buf, to_read);
491 ret = write_wim_resource_chunk(buf, to_read, out_fp,
492 compress, chunk_tab);
495 bytes_remaining -= to_read;
497 } while (bytes_remaining);
499 /* Raw copy: The new compressed size is the same as the old compressed
502 * Using WIMLIB_COMPRESSION_TYPE_NONE: The new compressed size is the
505 * Using a different compression type: Call
506 * finish_wim_resource_chunk_tab() and it will provide the new
510 new_compressed_size = old_compressed_size;
512 if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE)
513 new_compressed_size = original_size;
515 ret = finish_wim_resource_chunk_tab(chunk_tab, out_fp,
516 &new_compressed_size);
522 /* Verify SHA1 message digest of the resource, unless we are doing a raw
523 * write (in which case we never even saw the uncompressed data). Or,
524 * if the hash we had before is all 0's, just re-set it to be the new
527 u8 md[SHA1_HASH_SIZE];
528 sha1_final(md, &ctx);
529 if (is_zero_hash(lte->hash)) {
530 copy_hash(lte->hash, md);
531 } else if (!hashes_equal(md, lte->hash)) {
532 ERROR("WIM resource has incorrect hash!");
533 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK) {
534 ERROR("We were reading it from `%s'; maybe it changed "
535 "while we were reading it.",
538 ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
543 if (!raw && new_compressed_size >= original_size &&
544 out_ctype != WIMLIB_COMPRESSION_TYPE_NONE)
546 /* Oops! We compressed the resource to larger than the original
547 * size. Write the resource uncompressed instead. */
548 ret = write_uncompressed_resource_and_truncate(lte,
556 out_res_entry->size = new_compressed_size;
557 out_res_entry->original_size = original_size;
558 out_res_entry->offset = file_offset;
559 out_res_entry->flags = lte->resource_entry.flags
560 & ~WIM_RESHDR_FLAG_COMPRESSED;
561 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE)
562 out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED;
568 end_wim_resource_read(lte, ni);
570 end_wim_resource_read(lte);
577 #ifdef ENABLE_MULTITHREADED_COMPRESSION
579 /* Blocking shared queue (solves the producer-consumer problem) */
580 struct shared_queue {
584 unsigned filled_slots;
586 pthread_mutex_t lock;
587 pthread_cond_t msg_avail_cond;
588 pthread_cond_t space_avail_cond;
591 static int shared_queue_init(struct shared_queue *q, unsigned size)
593 wimlib_assert(size != 0);
594 q->array = CALLOC(sizeof(q->array[0]), size);
596 return WIMLIB_ERR_NOMEM;
601 pthread_mutex_init(&q->lock, NULL);
602 pthread_cond_init(&q->msg_avail_cond, NULL);
603 pthread_cond_init(&q->space_avail_cond, NULL);
607 static void shared_queue_destroy(struct shared_queue *q)
610 pthread_mutex_destroy(&q->lock);
611 pthread_cond_destroy(&q->msg_avail_cond);
612 pthread_cond_destroy(&q->space_avail_cond);
615 static void shared_queue_put(struct shared_queue *q, void *obj)
617 pthread_mutex_lock(&q->lock);
618 while (q->filled_slots == q->size)
619 pthread_cond_wait(&q->space_avail_cond, &q->lock);
621 q->back = (q->back + 1) % q->size;
622 q->array[q->back] = obj;
625 pthread_cond_broadcast(&q->msg_avail_cond);
626 pthread_mutex_unlock(&q->lock);
629 static void *shared_queue_get(struct shared_queue *q)
633 pthread_mutex_lock(&q->lock);
634 while (q->filled_slots == 0)
635 pthread_cond_wait(&q->msg_avail_cond, &q->lock);
637 obj = q->array[q->front];
638 q->array[q->front] = NULL;
639 q->front = (q->front + 1) % q->size;
642 pthread_cond_broadcast(&q->space_avail_cond);
643 pthread_mutex_unlock(&q->lock);
647 struct compressor_thread_params {
648 struct shared_queue *res_to_compress_queue;
649 struct shared_queue *compressed_res_queue;
650 compress_func_t compress;
653 #define MAX_CHUNKS_PER_MSG 2
656 struct wim_lookup_table_entry *lte;
657 u8 *uncompressed_chunks[MAX_CHUNKS_PER_MSG];
658 u8 *out_compressed_chunks[MAX_CHUNKS_PER_MSG];
659 u8 *compressed_chunks[MAX_CHUNKS_PER_MSG];
660 unsigned uncompressed_chunk_sizes[MAX_CHUNKS_PER_MSG];
661 unsigned compressed_chunk_sizes[MAX_CHUNKS_PER_MSG];
663 struct list_head list;
668 static void compress_chunks(struct message *msg, compress_func_t compress)
670 for (unsigned i = 0; i < msg->num_chunks; i++) {
671 DEBUG2("compress chunk %u of %u", i, msg->num_chunks);
672 int ret = compress(msg->uncompressed_chunks[i],
673 msg->uncompressed_chunk_sizes[i],
674 msg->compressed_chunks[i],
675 &msg->compressed_chunk_sizes[i]);
677 msg->out_compressed_chunks[i] = msg->compressed_chunks[i];
679 msg->out_compressed_chunks[i] = msg->uncompressed_chunks[i];
680 msg->compressed_chunk_sizes[i] = msg->uncompressed_chunk_sizes[i];
685 /* Compressor thread routine. This is a lot simpler than the main thread
686 * routine: just repeatedly get a group of chunks from the
687 * res_to_compress_queue, compress them, and put them in the
688 * compressed_res_queue. A NULL pointer indicates that the thread should stop.
690 static void *compressor_thread_proc(void *arg)
692 struct compressor_thread_params *params = arg;
693 struct shared_queue *res_to_compress_queue = params->res_to_compress_queue;
694 struct shared_queue *compressed_res_queue = params->compressed_res_queue;
695 compress_func_t compress = params->compress;
698 DEBUG("Compressor thread ready");
699 while ((msg = shared_queue_get(res_to_compress_queue)) != NULL) {
700 compress_chunks(msg, compress);
701 shared_queue_put(compressed_res_queue, msg);
703 DEBUG("Compressor thread terminating");
708 static int do_write_stream_list(struct list_head *my_resources,
711 wimlib_progress_func_t progress_func,
712 union wimlib_progress_info *progress,
713 int write_resource_flags)
716 struct wim_lookup_table_entry *lte, *tmp;
718 list_for_each_entry_safe(lte, tmp, my_resources, staging_list) {
719 ret = write_wim_resource(lte,
722 <e->output_resource_entry,
723 write_resource_flags);
726 list_del(<e->staging_list);
727 progress->write_streams.completed_bytes +=
728 wim_resource_size(lte);
729 progress->write_streams.completed_streams++;
731 progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
738 static int write_stream_list_serial(struct list_head *stream_list,
742 wimlib_progress_func_t progress_func,
743 union wimlib_progress_info *progress)
745 int write_resource_flags;
747 if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
748 write_resource_flags = WIMLIB_RESOURCE_FLAG_RECOMPRESS;
750 write_resource_flags = 0;
751 progress->write_streams.num_threads = 1;
753 progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress);
754 return do_write_stream_list(stream_list, out_fp,
755 out_ctype, progress_func,
756 progress, write_resource_flags);
759 #ifdef ENABLE_MULTITHREADED_COMPRESSION
760 static int write_wim_chunks(struct message *msg, FILE *out_fp,
761 struct chunk_table *chunk_tab)
763 for (unsigned i = 0; i < msg->num_chunks; i++) {
764 unsigned chunk_csize = msg->compressed_chunk_sizes[i];
766 DEBUG2("Write wim chunk %u of %u (csize = %u)",
767 i, msg->num_chunks, chunk_csize);
769 if (fwrite(msg->out_compressed_chunks[i], 1, chunk_csize, out_fp)
772 ERROR_WITH_ERRNO("Failed to write WIM chunk");
773 return WIMLIB_ERR_WRITE;
776 *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset;
777 chunk_tab->cur_offset += chunk_csize;
783 * This function is executed by the main thread when the resources are being
784 * compressed in parallel. The main thread is in change of all reading of the
785 * uncompressed data and writing of the compressed data. The compressor threads
786 * *only* do compression from/to in-memory buffers.
788 * Each unit of work given to a compressor thread is up to MAX_CHUNKS_PER_MSG
789 * chunks of compressed data to compress, represented in a `struct message'.
790 * Each message is passed from the main thread to a worker thread through the
791 * res_to_compress_queue, and it is passed back through the
792 * compressed_res_queue.
794 static int main_writer_thread_proc(struct list_head *stream_list,
797 struct shared_queue *res_to_compress_queue,
798 struct shared_queue *compressed_res_queue,
801 wimlib_progress_func_t progress_func,
802 union wimlib_progress_info *progress)
805 struct chunk_table *cur_chunk_tab = NULL;
806 struct message *msgs = CALLOC(num_messages, sizeof(struct message));
807 struct wim_lookup_table_entry *next_lte = NULL;
809 // Initially, all the messages are available to use.
810 LIST_HEAD(available_msgs);
813 ret = WIMLIB_ERR_NOMEM;
817 for (size_t i = 0; i < num_messages; i++)
818 list_add(&msgs[i].list, &available_msgs);
820 // outstanding_resources is the list of resources that currently have
821 // had chunks sent off for compression.
823 // The first stream in outstanding_resources is the stream that is
824 // currently being written (cur_lte).
826 // The last stream in outstanding_resources is the stream that is
827 // currently being read and chunks fed to the compressor threads
830 // Depending on the number of threads and the sizes of the resource,
831 // the outstanding streams list may contain streams between cur_lte and
832 // next_lte that have all their chunks compressed or being compressed,
833 // but haven't been written yet.
835 LIST_HEAD(outstanding_resources);
836 struct list_head *next_resource = stream_list->next;
838 u64 next_num_chunks = 0;
840 // As in write_wim_resource(), each resource we read is checksummed.
841 SHA_CTX next_sha_ctx;
842 u8 next_hash[SHA1_HASH_SIZE];
844 // Resources that don't need any chunks compressed are added to this
845 // list and written directly by the main thread.
846 LIST_HEAD(my_resources);
848 struct wim_lookup_table_entry *cur_lte = NULL;
852 ntfs_inode *ni = NULL;
855 DEBUG("Initializing buffers for uncompressed "
856 "and compressed data (%zu bytes needed)",
857 num_messages * MAX_CHUNKS_PER_MSG * WIM_CHUNK_SIZE * 2);
859 // Pre-allocate all the buffers that will be needed to do the chunk
861 for (size_t i = 0; i < num_messages; i++) {
862 for (size_t j = 0; j < MAX_CHUNKS_PER_MSG; j++) {
863 msgs[i].compressed_chunks[j] = MALLOC(WIM_CHUNK_SIZE);
865 // The extra 8 bytes is because longest_match() in
866 // lz77.c may read a little bit off the end of the
867 // uncompressed data. It doesn't need to be
868 // initialized--- we really just need to avoid accessing
870 msgs[i].uncompressed_chunks[j] = MALLOC(WIM_CHUNK_SIZE + 8);
871 if (msgs[i].compressed_chunks[j] == NULL ||
872 msgs[i].uncompressed_chunks[j] == NULL)
874 ret = WIMLIB_ERR_NOMEM;
880 // This loop is executed until all resources have been written, except
881 // possibly a few that have been added to the @my_resources list for
884 // Send chunks to the compressor threads until either (a) there
885 // are no more messages available since they were all sent off,
886 // or (b) there are no more resources that need to be
888 while (!list_empty(&available_msgs)) {
889 if (next_chunk == next_num_chunks) {
890 // If next_chunk == next_num_chunks, there are
891 // no more chunks to write in the current
892 // stream. So, check the SHA1 message digest of
893 // the stream that was just finished (unless
894 // next_lte == NULL, which is the case the very
895 // first time this loop is entered, and also
896 // near the very end of the compression when
897 // there are no more streams.) Then, advance to
898 // the next stream (if there is one).
899 if (next_lte != NULL) {
901 end_wim_resource_read(next_lte, ni);
904 end_wim_resource_read(next_lte);
906 DEBUG2("Finalize SHA1 md (next_num_chunks=%zu)",
908 sha1_final(next_hash, &next_sha_ctx);
909 if (!hashes_equal(next_lte->hash, next_hash)) {
910 ERROR("WIM resource has incorrect hash!");
911 if (next_lte->resource_location ==
912 RESOURCE_IN_FILE_ON_DISK)
914 ERROR("We were reading it from `%s'; "
915 "maybe it changed while we were "
917 next_lte->file_on_disk);
919 ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
924 // Advance to the next resource.
926 // If the next resource needs no compression, just write
927 // it with this thread (not now though--- we could be in
928 // the middle of writing another resource.) Keep doing
929 // this until we either get to the end of the resources
930 // list, or we get to a resource that needs compression.
932 if (next_resource == stream_list) {
933 // No more resources to send for
938 next_lte = container_of(next_resource,
939 struct wim_lookup_table_entry,
941 next_resource = next_resource->next;
942 if ((!(write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
943 && wim_resource_compression_type(next_lte) == out_ctype)
944 || wim_resource_size(next_lte) == 0)
946 list_add_tail(&next_lte->staging_list,
949 list_add_tail(&next_lte->staging_list,
950 &outstanding_resources);
952 next_num_chunks = wim_resource_chunks(next_lte);
953 sha1_init(&next_sha_ctx);
954 INIT_LIST_HEAD(&next_lte->msg_list);
956 ret = prepare_resource_for_read(next_lte, &ni);
958 ret = prepare_resource_for_read(next_lte);
963 if (cur_lte == NULL) {
964 // Set cur_lte for the
973 if (next_lte == NULL) {
974 // No more resources to send for compression
978 // Get a message from the available messages
980 msg = container_of(available_msgs.next,
984 // ... and delete it from the available messages
986 list_del(&msg->list);
988 // Initialize the message with the chunks to
990 msg->num_chunks = min(next_num_chunks - next_chunk,
993 msg->complete = false;
994 msg->begin_chunk = next_chunk;
996 unsigned size = WIM_CHUNK_SIZE;
997 for (unsigned i = 0; i < msg->num_chunks; i++) {
999 // Read chunk @next_chunk of the stream into the
1000 // message so that a compressor thread can
1003 if (next_chunk == next_num_chunks - 1) {
1004 size = MODULO_NONZERO(wim_resource_size(next_lte),
1008 DEBUG2("Read resource (size=%u, offset=%zu)",
1009 size, next_chunk * WIM_CHUNK_SIZE);
1011 msg->uncompressed_chunk_sizes[i] = size;
1013 ret = read_wim_resource(next_lte,
1014 msg->uncompressed_chunks[i],
1016 next_chunk * WIM_CHUNK_SIZE,
1020 sha1_update(&next_sha_ctx,
1021 msg->uncompressed_chunks[i], size);
1025 // Send the compression request
1026 list_add_tail(&msg->list, &next_lte->msg_list);
1027 shared_queue_put(res_to_compress_queue, msg);
1028 DEBUG2("Compression request sent");
1031 // If there are no outstanding resources, there are no more
1032 // resources that need to be written.
1033 if (list_empty(&outstanding_resources)) {
1038 // Get the next message from the queue and process it.
1039 // The message will contain 1 or more data chunks that have been
1041 msg = shared_queue_get(compressed_res_queue);
1042 msg->complete = true;
1044 // Is this the next chunk in the current resource? If it's not
1045 // (i.e., an earlier chunk in a same or different resource
1046 // hasn't been compressed yet), do nothing, and keep this
1047 // message around until all earlier chunks are received.
1049 // Otherwise, write all the chunks we can.
1050 while (cur_lte != NULL &&
1051 !list_empty(&cur_lte->msg_list) &&
1052 (msg = container_of(cur_lte->msg_list.next,
1056 DEBUG2("Complete msg (begin_chunk=%"PRIu64")", msg->begin_chunk);
1057 if (msg->begin_chunk == 0) {
1058 DEBUG2("Begin chunk tab");
1060 // This is the first set of chunks. Leave space
1061 // for the chunk table in the output file.
1062 off_t cur_offset = ftello(out_fp);
1063 if (cur_offset == -1) {
1064 ret = WIMLIB_ERR_WRITE;
1067 ret = begin_wim_resource_chunk_tab(cur_lte,
1075 // Write the compressed chunks from the message.
1076 ret = write_wim_chunks(msg, out_fp, cur_chunk_tab);
1080 list_del(&msg->list);
1082 // This message is available to use for different chunks
1084 list_add(&msg->list, &available_msgs);
1086 // Was this the last chunk of the stream? If so, finish
1088 if (list_empty(&cur_lte->msg_list) &&
1089 msg->begin_chunk + msg->num_chunks == cur_chunk_tab->num_chunks)
1091 DEBUG2("Finish wim chunk tab");
1093 ret = finish_wim_resource_chunk_tab(cur_chunk_tab,
1099 if (res_csize >= wim_resource_size(cur_lte)) {
1100 /* Oops! We compressed the resource to
1101 * larger than the original size. Write
1102 * the resource uncompressed instead. */
1103 ret = write_uncompressed_resource_and_truncate(
1106 cur_chunk_tab->file_offset,
1107 &cur_lte->output_resource_entry);
1111 cur_lte->output_resource_entry.size =
1114 cur_lte->output_resource_entry.original_size =
1115 cur_lte->resource_entry.original_size;
1117 cur_lte->output_resource_entry.offset =
1118 cur_chunk_tab->file_offset;
1120 cur_lte->output_resource_entry.flags =
1121 cur_lte->resource_entry.flags |
1122 WIM_RESHDR_FLAG_COMPRESSED;
1125 progress->write_streams.completed_bytes +=
1126 wim_resource_size(cur_lte);
1127 progress->write_streams.completed_streams++;
1129 if (progress_func) {
1130 progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
1134 FREE(cur_chunk_tab);
1135 cur_chunk_tab = NULL;
1137 struct list_head *next = cur_lte->staging_list.next;
1138 list_del(&cur_lte->staging_list);
1140 if (next == &outstanding_resources)
1143 cur_lte = container_of(cur_lte->staging_list.next,
1144 struct wim_lookup_table_entry,
1147 // Since we just finished writing a stream,
1148 // write any streams that have been added to the
1149 // my_resources list for direct writing by the
1150 // main thread (e.g. resources that don't need
1151 // to be compressed because the desired
1152 // compression type is the same as the previous
1153 // compression type).
1154 ret = do_write_stream_list(&my_resources,
1167 if (ret == WIMLIB_ERR_NOMEM) {
1168 ERROR("Could not allocate enough memory for "
1169 "multi-threaded compression");
1174 end_wim_resource_read(next_lte, ni);
1176 end_wim_resource_read(next_lte);
1181 ret = do_write_stream_list(&my_resources, out_fp,
1182 out_ctype, progress_func,
1186 size_t num_available_msgs = 0;
1187 struct list_head *cur;
1189 list_for_each(cur, &available_msgs) {
1190 num_available_msgs++;
1193 while (num_available_msgs < num_messages) {
1194 shared_queue_get(compressed_res_queue);
1195 num_available_msgs++;
1201 for (size_t i = 0; i < num_messages; i++) {
1202 for (size_t j = 0; j < MAX_CHUNKS_PER_MSG; j++) {
1203 FREE(msgs[i].compressed_chunks[j]);
1204 FREE(msgs[i].uncompressed_chunks[j]);
1210 FREE(cur_chunk_tab);
1214 static long get_default_num_threads()
1217 return win32_get_number_of_processors();
1219 return sysconf(_SC_NPROCESSORS_ONLN);
1223 static int write_stream_list_parallel(struct list_head *stream_list,
1227 unsigned num_threads,
1228 wimlib_progress_func_t progress_func,
1229 union wimlib_progress_info *progress)
1232 struct shared_queue res_to_compress_queue;
1233 struct shared_queue compressed_res_queue;
1234 pthread_t *compressor_threads = NULL;
1236 if (num_threads == 0) {
1237 long nthreads = get_default_num_threads();
1238 if (nthreads < 1 || nthreads > UINT_MAX) {
1239 WARNING("Could not determine number of processors! Assuming 1");
1242 num_threads = nthreads;
1246 progress->write_streams.num_threads = num_threads;
1247 wimlib_assert(stream_list->next != stream_list);
1249 static const double MESSAGES_PER_THREAD = 2.0;
1250 size_t queue_size = (size_t)(num_threads * MESSAGES_PER_THREAD);
1252 DEBUG("Initializing shared queues (queue_size=%zu)", queue_size);
1254 ret = shared_queue_init(&res_to_compress_queue, queue_size);
1258 ret = shared_queue_init(&compressed_res_queue, queue_size);
1260 goto out_destroy_res_to_compress_queue;
1262 struct compressor_thread_params params;
1263 params.res_to_compress_queue = &res_to_compress_queue;
1264 params.compressed_res_queue = &compressed_res_queue;
1265 params.compress = get_compress_func(out_ctype);
1267 compressor_threads = MALLOC(num_threads * sizeof(pthread_t));
1268 if (!compressor_threads) {
1269 ret = WIMLIB_ERR_NOMEM;
1270 goto out_destroy_compressed_res_queue;
1273 for (unsigned i = 0; i < num_threads; i++) {
1274 DEBUG("pthread_create thread %u", i);
1275 ret = pthread_create(&compressor_threads[i], NULL,
1276 compressor_thread_proc, ¶ms);
1279 ERROR_WITH_ERRNO("Failed to create compressor "
1287 progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress);
1289 ret = main_writer_thread_proc(stream_list,
1292 &res_to_compress_queue,
1293 &compressed_res_queue,
1299 for (unsigned i = 0; i < num_threads; i++)
1300 shared_queue_put(&res_to_compress_queue, NULL);
1302 for (unsigned i = 0; i < num_threads; i++) {
1303 if (pthread_join(compressor_threads[i], NULL)) {
1304 WARNING_WITH_ERRNO("Failed to join compressor "
1308 FREE(compressor_threads);
1309 out_destroy_compressed_res_queue:
1310 shared_queue_destroy(&compressed_res_queue);
1311 out_destroy_res_to_compress_queue:
1312 shared_queue_destroy(&res_to_compress_queue);
1313 if (ret >= 0 && ret != WIMLIB_ERR_NOMEM)
1316 WARNING("Falling back to single-threaded compression");
1317 return write_stream_list_serial(stream_list,
1328 * Write a list of streams to a WIM (@out_fp) using the compression type
1329 * @out_ctype and up to @num_threads compressor threads.
1331 static int write_stream_list(struct list_head *stream_list, FILE *out_fp,
1332 int out_ctype, int write_flags,
1333 unsigned num_threads,
1334 wimlib_progress_func_t progress_func)
1336 struct wim_lookup_table_entry *lte;
1337 size_t num_streams = 0;
1338 u64 total_bytes = 0;
1339 u64 total_compression_bytes = 0;
1340 union wimlib_progress_info progress;
1342 list_for_each_entry(lte, stream_list, staging_list) {
1344 total_bytes += wim_resource_size(lte);
1345 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE
1346 && (wim_resource_compression_type(lte) != out_ctype ||
1347 (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)))
1349 total_compression_bytes += wim_resource_size(lte);
1352 progress.write_streams.total_bytes = total_bytes;
1353 progress.write_streams.total_streams = num_streams;
1354 progress.write_streams.completed_bytes = 0;
1355 progress.write_streams.completed_streams = 0;
1356 progress.write_streams.num_threads = num_threads;
1357 progress.write_streams.compression_type = out_ctype;
1359 #ifdef ENABLE_MULTITHREADED_COMPRESSION
1360 if (total_compression_bytes >= 1000000 && num_threads != 1)
1361 return write_stream_list_parallel(stream_list,
1370 return write_stream_list_serial(stream_list,
1378 struct lte_overwrite_prepare_args {
1381 struct list_head *stream_list;
1384 static int lte_overwrite_prepare(struct wim_lookup_table_entry *lte, void *arg)
1386 struct lte_overwrite_prepare_args *args = arg;
1388 if (lte->resource_location == RESOURCE_IN_WIM &&
1389 lte->wim == args->wim &&
1390 lte->resource_entry.offset + lte->resource_entry.size > args->end_offset)
1392 #ifdef ENABLE_ERROR_MESSAGES
1393 ERROR("The following resource is after the XML data:");
1394 print_lookup_table_entry(lte, stderr);
1396 return WIMLIB_ERR_RESOURCE_ORDER;
1399 lte->out_refcnt = lte->refcnt;
1400 memcpy(<e->output_resource_entry, <e->resource_entry,
1401 sizeof(struct resource_entry));
1402 if (!(lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA)) {
1403 wimlib_assert(lte->resource_location != RESOURCE_NONEXISTENT);
1404 if (lte->resource_location != RESOURCE_IN_WIM || lte->wim != args->wim)
1405 list_add(<e->staging_list, args->stream_list);
1410 static int wim_find_new_streams(WIMStruct *wim, off_t end_offset,
1411 struct list_head *stream_list)
1413 struct lte_overwrite_prepare_args args = {
1415 .end_offset = end_offset,
1416 .stream_list = stream_list,
1419 return for_lookup_table_entry(wim->lookup_table,
1420 lte_overwrite_prepare, &args);
1423 static int inode_find_streams_to_write(struct wim_inode *inode,
1424 struct wim_lookup_table *table,
1425 struct list_head *stream_list)
1427 struct wim_lookup_table_entry *lte;
1428 for (unsigned i = 0; i <= inode->i_num_ads; i++) {
1429 lte = inode_stream_lte(inode, i, table);
1431 if (lte->out_refcnt == 0)
1432 list_add_tail(<e->staging_list, stream_list);
1433 lte->out_refcnt += inode->i_nlink;
1439 static int image_find_streams_to_write(WIMStruct *w)
1441 struct wim_inode *inode;
1442 struct hlist_node *cur;
1443 struct hlist_head *inode_list;
1445 inode_list = &wim_get_current_image_metadata(w)->inode_list;
1446 hlist_for_each_entry(inode, cur, inode_list, i_hlist) {
1447 inode_find_streams_to_write(inode, w->lookup_table,
1448 (struct list_head*)w->private);
1453 static int write_wim_streams(WIMStruct *w, int image, int write_flags,
1454 unsigned num_threads,
1455 wimlib_progress_func_t progress_func)
1458 for_lookup_table_entry(w->lookup_table, lte_zero_out_refcnt, NULL);
1459 LIST_HEAD(stream_list);
1460 w->private = &stream_list;
1461 for_image(w, image, image_find_streams_to_write);
1462 return write_stream_list(&stream_list, w->out_fp,
1463 wimlib_get_compression_type(w), write_flags,
1464 num_threads, progress_func);
1468 * Finish writing a WIM file: write the lookup table, xml data, and integrity
1469 * table (optional), then overwrite the WIM header.
1471 * write_flags is a bitwise OR of the following:
1473 * (public) WIMLIB_WRITE_FLAG_CHECK_INTEGRITY:
1474 * Include an integrity table.
1476 * (public) WIMLIB_WRITE_FLAG_SHOW_PROGRESS:
1477 * Show progress information when (if) writing the integrity table.
1479 * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE:
1480 * Don't write the lookup table.
1482 * (private) WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE:
1483 * When (if) writing the integrity table, re-use entries from the
1484 * existing integrity table, if possible.
1486 * (private) WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML:
1487 * After writing the XML data but before writing the integrity
1488 * table, write a temporary WIM header and flush the stream so that
1489 * the WIM is less likely to become corrupted upon abrupt program
1492 * (private) WIMLIB_WRITE_FLAG_FSYNC:
1493 * fsync() the output file before closing it.
1496 int finish_write(WIMStruct *w, int image, int write_flags,
1497 wimlib_progress_func_t progress_func)
1500 struct wim_header hdr;
1501 FILE *out = w->out_fp;
1503 /* @hdr will be the header for the new WIM. First copy all the data
1504 * from the header in the WIMStruct; then set all the fields that may
1505 * have changed, including the resource entries, boot index, and image
1507 memcpy(&hdr, &w->hdr, sizeof(struct wim_header));
1509 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
1510 ret = write_lookup_table(w->lookup_table, out, &hdr.lookup_table_res_entry);
1515 ret = write_xml_data(w->wim_info, image, out,
1516 (write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE) ?
1517 wim_info_get_total_bytes(w->wim_info) : 0,
1518 &hdr.xml_res_entry);
1522 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
1523 if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) {
1524 struct wim_header checkpoint_hdr;
1525 memcpy(&checkpoint_hdr, &hdr, sizeof(struct wim_header));
1526 memset(&checkpoint_hdr.integrity, 0, sizeof(struct resource_entry));
1527 if (fseeko(out, 0, SEEK_SET) != 0) {
1528 ERROR_WITH_ERRNO("Failed to seek to beginning "
1529 "of WIM being written");
1530 ret = WIMLIB_ERR_WRITE;
1533 ret = write_header(&checkpoint_hdr, out);
1537 if (fflush(out) != 0) {
1538 ERROR_WITH_ERRNO("Can't write data to WIM");
1539 ret = WIMLIB_ERR_WRITE;
1543 if (fseeko(out, 0, SEEK_END) != 0) {
1544 ERROR_WITH_ERRNO("Failed to seek to end "
1545 "of WIM being written");
1546 ret = WIMLIB_ERR_WRITE;
1551 off_t old_lookup_table_end;
1552 off_t new_lookup_table_end;
1553 if (write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE) {
1554 old_lookup_table_end = w->hdr.lookup_table_res_entry.offset +
1555 w->hdr.lookup_table_res_entry.size;
1557 old_lookup_table_end = 0;
1559 new_lookup_table_end = hdr.lookup_table_res_entry.offset +
1560 hdr.lookup_table_res_entry.size;
1562 ret = write_integrity_table(out,
1564 new_lookup_table_end,
1565 old_lookup_table_end,
1570 memset(&hdr.integrity, 0, sizeof(struct resource_entry));
1574 * In the WIM header, there is room for the resource entry for a
1575 * metadata resource labeled as the "boot metadata". This entry should
1576 * be zeroed out if there is no bootable image (boot_idx 0). Otherwise,
1577 * it should be a copy of the resource entry for the image that is
1578 * marked as bootable. This is not well documented...
1581 /* Set image count and boot index correctly for single image writes */
1582 if (image != WIMLIB_ALL_IMAGES) {
1583 hdr.image_count = 1;
1584 if (hdr.boot_idx == image)
1590 if (hdr.boot_idx == 0) {
1591 memset(&hdr.boot_metadata_res_entry, 0,
1592 sizeof(struct resource_entry));
1594 memcpy(&hdr.boot_metadata_res_entry,
1596 hdr.boot_idx - 1].metadata_lte->output_resource_entry,
1597 sizeof(struct resource_entry));
1600 if (fseeko(out, 0, SEEK_SET) != 0) {
1601 ERROR_WITH_ERRNO("Failed to seek to beginning of WIM "
1603 ret = WIMLIB_ERR_WRITE;
1607 ret = write_header(&hdr, out);
1611 if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) {
1612 if (fflush(out) != 0
1613 || fsync(fileno(out)) != 0)
1615 ERROR_WITH_ERRNO("Error flushing data to WIM file");
1616 ret = WIMLIB_ERR_WRITE;
1620 if (fclose(out) != 0) {
1621 ERROR_WITH_ERRNO("Failed to close the WIM file");
1623 ret = WIMLIB_ERR_WRITE;
1629 #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK)
1630 int lock_wim(WIMStruct *w, FILE *fp)
1633 if (fp && !w->wim_locked) {
1634 ret = flock(fileno(fp), LOCK_EX | LOCK_NB);
1636 if (errno == EWOULDBLOCK) {
1637 ERROR("`%s' is already being modified or has been "
1638 "mounted read-write\n"
1639 " by another process!", w->filename);
1640 ret = WIMLIB_ERR_ALREADY_LOCKED;
1642 WARNING_WITH_ERRNO("Failed to lock `%s'",
1654 static int open_wim_writable(WIMStruct *w, const char *path,
1655 bool trunc, bool readable)
1666 wimlib_assert(w->out_fp == NULL);
1667 w->out_fp = fopen(path, mode);
1671 ERROR_WITH_ERRNO("Failed to open `%s' for writing", path);
1672 return WIMLIB_ERR_OPEN;
1677 void close_wim_writable(WIMStruct *w)
1680 if (fclose(w->out_fp) != 0) {
1681 WARNING_WITH_ERRNO("Failed to close output WIM");
1687 /* Open file stream and write dummy header for WIM. */
1688 int begin_write(WIMStruct *w, const char *path, int write_flags)
1691 ret = open_wim_writable(w, path, true,
1692 (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0);
1695 /* Write dummy header. It will be overwritten later. */
1696 return write_header(&w->hdr, w->out_fp);
1699 /* Writes a stand-alone WIM to a file. */
1700 WIMLIBAPI int wimlib_write(WIMStruct *w, const char *path,
1701 int image, int write_flags, unsigned num_threads,
1702 wimlib_progress_func_t progress_func)
1707 return WIMLIB_ERR_INVALID_PARAM;
1709 write_flags &= WIMLIB_WRITE_MASK_PUBLIC;
1711 if (image != WIMLIB_ALL_IMAGES &&
1712 (image < 1 || image > w->hdr.image_count))
1713 return WIMLIB_ERR_INVALID_IMAGE;
1715 if (w->hdr.total_parts != 1) {
1716 ERROR("Cannot call wimlib_write() on part of a split WIM");
1717 return WIMLIB_ERR_SPLIT_UNSUPPORTED;
1720 ret = begin_write(w, path, write_flags);
1724 ret = write_wim_streams(w, image, write_flags, num_threads,
1730 progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN, NULL);
1732 ret = for_image(w, image, write_metadata_resource);
1737 progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_END, NULL);
1739 ret = finish_write(w, image, write_flags, progress_func);
1741 close_wim_writable(w);
1742 DEBUG("wimlib_write(path=%s) = %d", path, ret);
1746 static bool any_images_modified(WIMStruct *w)
1748 for (int i = 0; i < w->hdr.image_count; i++)
1749 if (w->image_metadata[i].modified)
1755 * Overwrite a WIM, possibly appending streams to it.
1757 * A WIM looks like (or is supposed to look like) the following:
1759 * Header (212 bytes)
1760 * Streams and metadata resources (variable size)
1761 * Lookup table (variable size)
1762 * XML data (variable size)
1763 * Integrity table (optional) (variable size)
1765 * If we are not adding any streams or metadata resources, the lookup table is
1766 * unchanged--- so we only need to overwrite the XML data, integrity table, and
1767 * header. This operation is potentially unsafe if the program is abruptly
1768 * terminated while the XML data or integrity table are being overwritten, but
1769 * before the new header has been written. To partially alleviate this problem,
1770 * a special flag (WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) is passed to
1771 * finish_write() to cause a temporary WIM header to be written after the XML
1772 * data has been written. This may prevent the WIM from becoming corrupted if
1773 * the program is terminated while the integrity table is being calculated (but
1774 * no guarantees, due to write re-ordering...).
1776 * If we are adding new streams or images (metadata resources), the lookup table
1777 * needs to be changed, and those streams need to be written. In this case, we
1778 * try to perform a safe update of the WIM file by writing the streams *after*
1779 * the end of the previous WIM, then writing the new lookup table, XML data, and
1780 * (optionally) integrity table following the new streams. This will produce a
1781 * layout like the following:
1783 * Header (212 bytes)
1784 * (OLD) Streams and metadata resources (variable size)
1785 * (OLD) Lookup table (variable size)
1786 * (OLD) XML data (variable size)
1787 * (OLD) Integrity table (optional) (variable size)
1788 * (NEW) Streams and metadata resources (variable size)
1789 * (NEW) Lookup table (variable size)
1790 * (NEW) XML data (variable size)
1791 * (NEW) Integrity table (optional) (variable size)
1793 * At all points, the WIM is valid as nothing points to the new data yet. Then,
1794 * the header is overwritten to point to the new lookup table, XML data, and
1795 * integrity table, to produce the following layout:
1797 * Header (212 bytes)
1798 * Streams and metadata resources (variable size)
1799 * Nothing (variable size)
1800 * More Streams and metadata resources (variable size)
1801 * Lookup table (variable size)
1802 * XML data (variable size)
1803 * Integrity table (optional) (variable size)
1805 * This method allows an image to be appended to a large WIM very quickly, and
1806 * is is crash-safe except in the case of write re-ordering, but the
1807 * disadvantage is that a small hole is left in the WIM where the old lookup
1808 * table, xml data, and integrity table were. (These usually only take up a
1809 * small amount of space compared to the streams, however.)
1811 static int overwrite_wim_inplace(WIMStruct *w, int write_flags,
1812 unsigned num_threads,
1813 wimlib_progress_func_t progress_func)
1816 struct list_head stream_list;
1818 bool found_modified_image;
1820 DEBUG("Overwriting `%s' in-place", w->filename);
1822 /* Make sure that the integrity table (if present) is after the XML
1823 * data, and that there are no stream resources, metadata resources, or
1824 * lookup tables after the XML data. Otherwise, these data would be
1826 if (w->hdr.integrity.offset != 0 &&
1827 w->hdr.integrity.offset < w->hdr.xml_res_entry.offset) {
1828 ERROR("Didn't expect the integrity table to be before the XML data");
1829 return WIMLIB_ERR_RESOURCE_ORDER;
1832 if (w->hdr.lookup_table_res_entry.offset > w->hdr.xml_res_entry.offset) {
1833 ERROR("Didn't expect the lookup table to be after the XML data");
1834 return WIMLIB_ERR_RESOURCE_ORDER;
1838 if (w->hdr.integrity.offset)
1839 old_wim_end = w->hdr.integrity.offset + w->hdr.integrity.size;
1841 old_wim_end = w->hdr.xml_res_entry.offset + w->hdr.xml_res_entry.size;
1843 if (!w->deletion_occurred && !any_images_modified(w)) {
1844 /* If no images have been modified and no images have been
1845 * deleted, a new lookup table does not need to be written. */
1846 old_wim_end = w->hdr.lookup_table_res_entry.offset +
1847 w->hdr.lookup_table_res_entry.size;
1848 write_flags |= WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE |
1849 WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML;
1851 INIT_LIST_HEAD(&stream_list);
1852 ret = wim_find_new_streams(w, old_wim_end, &stream_list);
1856 ret = open_wim_writable(w, w->filename, false,
1857 (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0);
1861 ret = lock_wim(w, w->out_fp);
1868 if (fseeko(w->out_fp, old_wim_end, SEEK_SET) != 0) {
1869 ERROR_WITH_ERRNO("Can't seek to end of WIM");
1873 return WIMLIB_ERR_WRITE;
1876 if (!list_empty(&stream_list)) {
1877 DEBUG("Writing newly added streams (offset = %"PRIu64")",
1879 ret = write_stream_list(&stream_list, w->out_fp,
1880 wimlib_get_compression_type(w),
1881 write_flags, num_threads,
1886 DEBUG("No new streams were added");
1889 found_modified_image = false;
1890 for (int i = 0; i < w->hdr.image_count; i++) {
1891 if (!found_modified_image)
1892 found_modified_image = w->image_metadata[i].modified;
1893 if (found_modified_image) {
1894 select_wim_image(w, i + 1);
1895 ret = write_metadata_resource(w);
1900 write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE;
1901 ret = finish_write(w, WIMLIB_ALL_IMAGES, write_flags,
1904 close_wim_writable(w);
1905 if (ret != 0 && !(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
1906 WARNING("Truncating `%s' to its original size (%"PRIu64" bytes)",
1907 w->filename, old_wim_end);
1908 /* Return value of truncate() is ignored because this is already
1910 (void)truncate(w->filename, old_wim_end);
1916 static int overwrite_wim_via_tmpfile(WIMStruct *w, int write_flags,
1917 unsigned num_threads,
1918 wimlib_progress_func_t progress_func)
1920 size_t wim_name_len;
1923 DEBUG("Overwriting `%s' via a temporary file", w->filename);
1925 /* Write the WIM to a temporary file in the same directory as the
1927 wim_name_len = strlen(w->filename);
1928 char tmpfile[wim_name_len + 10];
1929 memcpy(tmpfile, w->filename, wim_name_len);
1930 randomize_char_array_with_alnum(tmpfile + wim_name_len, 9);
1931 tmpfile[wim_name_len + 9] = '\0';
1933 ret = wimlib_write(w, tmpfile, WIMLIB_ALL_IMAGES,
1934 write_flags | WIMLIB_WRITE_FLAG_FSYNC,
1935 num_threads, progress_func);
1937 ERROR("Failed to write the WIM file `%s'", tmpfile);
1941 DEBUG("Renaming `%s' to `%s'", tmpfile, w->filename);
1943 /* Rename the new file to the old file .*/
1944 if (rename(tmpfile, w->filename) != 0) {
1945 ERROR_WITH_ERRNO("Failed to rename `%s' to `%s'",
1946 tmpfile, w->filename);
1947 ret = WIMLIB_ERR_RENAME;
1951 if (progress_func) {
1952 union wimlib_progress_info progress;
1953 progress.rename.from = tmpfile;
1954 progress.rename.to = w->filename;
1955 progress_func(WIMLIB_PROGRESS_MSG_RENAME, &progress);
1958 /* Close the original WIM file that was opened for reading. */
1959 if (w->fp != NULL) {
1964 /* Re-open the WIM read-only. */
1965 w->fp = fopen(w->filename, "rb");
1966 if (w->fp == NULL) {
1967 ret = WIMLIB_ERR_REOPEN;
1968 WARNING_WITH_ERRNO("Failed to re-open `%s' read-only",
1975 /* Remove temporary file. */
1976 if (unlink(tmpfile) != 0)
1977 WARNING_WITH_ERRNO("Failed to remove `%s'", tmpfile);
1982 * Writes a WIM file to the original file that it was read from, overwriting it.
1984 WIMLIBAPI int wimlib_overwrite(WIMStruct *w, int write_flags,
1985 unsigned num_threads,
1986 wimlib_progress_func_t progress_func)
1988 write_flags &= WIMLIB_WRITE_MASK_PUBLIC;
1991 return WIMLIB_ERR_NO_FILENAME;
1993 if (w->hdr.total_parts != 1) {
1994 ERROR("Cannot modify a split WIM");
1995 return WIMLIB_ERR_SPLIT_UNSUPPORTED;
1998 if ((!w->deletion_occurred || (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE))
1999 && !(write_flags & WIMLIB_WRITE_FLAG_REBUILD))
2002 ret = overwrite_wim_inplace(w, write_flags, num_threads,
2004 if (ret == WIMLIB_ERR_RESOURCE_ORDER)
2005 WARNING("Falling back to re-building entire WIM");
2009 return overwrite_wim_via_tmpfile(w, write_flags, num_threads,