4 * Support for writing WIM files; write a WIM file, overwrite a WIM file, write
5 * compressed file resources, etc.
9 * Copyright (C) 2010 Carl Thijssen
10 * Copyright (C) 2012 Eric Biggers
12 * This file is part of wimlib, a library for working with WIM files.
14 * wimlib is free software; you can redistribute it and/or modify it under the
15 * terms of the GNU General Public License as published by the Free
16 * Software Foundation; either version 3 of the License, or (at your option)
19 * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
20 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
21 * A PARTICULAR PURPOSE. See the GNU General Public License for more
24 * You should have received a copy of the GNU General Public License
25 * along with wimlib; if not, see http://www.gnu.org/licenses/.
28 #include "wimlib_internal.h"
31 #include "lookup_table.h"
37 #ifdef ENABLE_MULTITHREADED_COMPRESSION
38 #include <semaphore.h>
45 #include <ntfs-3g/attrib.h>
46 #include <ntfs-3g/inode.h>
47 #include <ntfs-3g/dir.h>
57 static int do_fflush(FILE *fp)
61 ERROR_WITH_ERRNO("Failed to flush data to output WIM file");
62 return WIMLIB_ERR_WRITE;
67 static int fflush_and_ftruncate(FILE *fp, off_t size)
74 ret = ftruncate(fileno(fp), size);
76 ERROR_WITH_ERRNO("Failed to truncate output WIM file to "
77 "%"PRIu64" bytes", size);
78 return WIMLIB_ERR_WRITE;
83 /* Chunk table that's located at the beginning of each compressed resource in
84 * the WIM. (This is not the on-disk format; the on-disk format just has an
85 * array of offsets.) */
89 u64 original_resource_size;
90 u64 bytes_per_chunk_entry;
98 * Allocates and initializes a chunk table, and reserves space for it in the
102 begin_wim_resource_chunk_tab(const struct lookup_table_entry *lte,
105 struct chunk_table **chunk_tab_ret)
107 u64 size = wim_resource_size(lte);
108 u64 num_chunks = (size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE;
109 size_t alloc_size = sizeof(struct chunk_table) + num_chunks * sizeof(u64);
110 struct chunk_table *chunk_tab = CALLOC(1, alloc_size);
114 ERROR("Failed to allocate chunk table for %"PRIu64" byte "
116 ret = WIMLIB_ERR_NOMEM;
119 chunk_tab->file_offset = file_offset;
120 chunk_tab->num_chunks = num_chunks;
121 chunk_tab->original_resource_size = size;
122 chunk_tab->bytes_per_chunk_entry = (size >= (1ULL << 32)) ? 8 : 4;
123 chunk_tab->table_disk_size = chunk_tab->bytes_per_chunk_entry *
125 chunk_tab->cur_offset = 0;
126 chunk_tab->cur_offset_p = chunk_tab->offsets;
128 if (fwrite(chunk_tab, 1, chunk_tab->table_disk_size, out_fp) !=
129 chunk_tab->table_disk_size) {
130 ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
132 ret = WIMLIB_ERR_WRITE;
138 *chunk_tab_ret = chunk_tab;
143 * Pointer to function to compresses a chunk of a WIM resource.
145 * @chunk: Uncompressed data of the chunk.
146 * @chunk_size: Size of the uncompressed chunk in bytes.
147 * @compressed_chunk: Pointer to output buffer of size at least
148 * (@chunk_size - 1) bytes.
149 * @compressed_chunk_len_ret: Pointer to an unsigned int into which the size
150 * of the compressed chunk will be
153 * Returns zero if compressed succeeded, and nonzero if the chunk could not be
154 * compressed to any smaller than @chunk_size. This function cannot fail for
157 typedef int (*compress_func_t)(const void *, unsigned, void *, unsigned *);
159 compress_func_t get_compress_func(int out_ctype)
161 if (out_ctype == WIM_COMPRESSION_TYPE_LZX)
164 return xpress_compress;
168 * Writes a chunk of a WIM resource to an output file.
170 * @chunk: Uncompressed data of the chunk.
171 * @chunk_size: Size of the chunk (<= WIM_CHUNK_SIZE)
172 * @out_fp: FILE * to write tho chunk to.
173 * @out_ctype: Compression type to use when writing the chunk (ignored if no
174 * chunk table provided)
175 * @chunk_tab: Pointer to chunk table being created. It is updated with the
176 * offset of the chunk we write.
178 * Returns 0 on success; nonzero on failure.
180 static int write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size,
181 FILE *out_fp, compress_func_t compress,
182 struct chunk_table *chunk_tab)
185 unsigned out_chunk_size;
187 u8 *compressed_chunk = alloca(chunk_size);
190 ret = compress(chunk, chunk_size, compressed_chunk,
193 out_chunk = compressed_chunk;
196 out_chunk_size = chunk_size;
198 *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset;
199 chunk_tab->cur_offset += out_chunk_size;
202 out_chunk_size = chunk_size;
204 if (fwrite(out_chunk, 1, out_chunk_size, out_fp) != out_chunk_size) {
205 ERROR_WITH_ERRNO("Failed to write WIM resource chunk");
206 return WIMLIB_ERR_WRITE;
212 * Finishes a WIM chunk tale and writes it to the output file at the correct
215 * The final size of the full compressed resource is returned in the
216 * @compressed_size_p.
219 finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab,
220 FILE *out_fp, u64 *compressed_size_p)
222 size_t bytes_written;
223 if (fseeko(out_fp, chunk_tab->file_offset, SEEK_SET) != 0) {
224 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of output "
225 "WIM file", chunk_tab->file_offset);
226 return WIMLIB_ERR_WRITE;
229 if (chunk_tab->bytes_per_chunk_entry == 8) {
230 array_cpu_to_le64(chunk_tab->offsets, chunk_tab->num_chunks);
232 for (u64 i = 0; i < chunk_tab->num_chunks; i++)
233 ((u32*)chunk_tab->offsets)[i] =
234 cpu_to_le32(chunk_tab->offsets[i]);
236 bytes_written = fwrite((u8*)chunk_tab->offsets +
237 chunk_tab->bytes_per_chunk_entry,
238 1, chunk_tab->table_disk_size, out_fp);
239 if (bytes_written != chunk_tab->table_disk_size) {
240 ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
242 return WIMLIB_ERR_WRITE;
244 if (fseeko(out_fp, 0, SEEK_END) != 0) {
245 ERROR_WITH_ERRNO("Failed to seek to end of output WIM file");
246 return WIMLIB_ERR_WRITE;
248 *compressed_size_p = chunk_tab->cur_offset + chunk_tab->table_disk_size;
252 /* Prepare for multiple reads to a resource by caching a FILE * or NTFS
253 * attribute pointer in the lookup table entry. */
254 static int prepare_resource_for_read(struct lookup_table_entry *lte
257 , ntfs_inode **ni_ret
261 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK
262 && !lte->file_on_disk_fp)
264 wimlib_assert(lte->file_on_disk);
265 lte->file_on_disk_fp = fopen(lte->file_on_disk, "rb");
266 if (!lte->file_on_disk_fp) {
267 ERROR_WITH_ERRNO("Failed to open the file `%s' for "
268 "reading", lte->file_on_disk);
269 return WIMLIB_ERR_OPEN;
273 else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME
276 struct ntfs_location *loc = lte->ntfs_loc;
279 ni = ntfs_pathname_to_inode(*loc->ntfs_vol_p, NULL, loc->path_utf8);
281 ERROR_WITH_ERRNO("Failed to open inode `%s' in NTFS "
282 "volume", loc->path_utf8);
283 return WIMLIB_ERR_NTFS_3G;
285 lte->attr = ntfs_attr_open(ni,
286 loc->is_reparse_point ? AT_REPARSE_POINT : AT_DATA,
287 (ntfschar*)loc->stream_name_utf16,
288 loc->stream_name_utf16_num_chars);
290 ERROR_WITH_ERRNO("Failed to open attribute of `%s' in "
291 "NTFS volume", loc->path_utf8);
292 ntfs_inode_close(ni);
293 return WIMLIB_ERR_NTFS_3G;
301 /* Undo prepare_resource_for_read() by closing the cached FILE * or NTFS
303 static void end_wim_resource_read(struct lookup_table_entry *lte
309 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK
310 && lte->file_on_disk_fp) {
311 fclose(lte->file_on_disk_fp);
312 lte->file_on_disk_fp = NULL;
315 else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME) {
317 ntfs_attr_close(lte->attr);
321 ntfs_inode_close(ni);
327 * Writes a WIM resource to a FILE * opened for writing. The resource may be
328 * written uncompressed or compressed depending on the @out_ctype parameter.
330 * If by chance the resource compresses to more than the original size (this may
331 * happen with random data or files than are pre-compressed), the resource is
332 * instead written uncompressed (and this is reflected in the @out_res_entry by
333 * removing the WIM_RESHDR_FLAG_COMPRESSED flag).
335 * @lte: The lookup table entry for the WIM resource.
336 * @out_fp: The FILE * to write the resource to.
337 * @out_ctype: The compression type of the resource to write. Note: if this is
338 * the same as the compression type of the WIM resource we
339 * need to read, we simply copy the data (i.e. we do not
340 * uncompress it, then compress it again).
341 * @out_res_entry: If non-NULL, a resource entry that is filled in with the
342 * offset, original size, compressed size, and compression flag
343 * of the output resource.
345 * Returns 0 on success; nonzero on failure.
347 int write_wim_resource(struct lookup_table_entry *lte,
348 FILE *out_fp, int out_ctype,
349 struct resource_entry *out_res_entry,
354 u64 old_compressed_size;
355 u64 new_compressed_size;
358 struct chunk_table *chunk_tab = NULL;
361 compress_func_t compress;
363 ntfs_inode *ni = NULL;
368 /* Original size of the resource */
369 original_size = wim_resource_size(lte);
371 /* Compressed size of the resource (as it exists now) */
372 old_compressed_size = wim_resource_compressed_size(lte);
374 /* Current offset in output file */
375 file_offset = ftello(out_fp);
376 if (file_offset == -1) {
377 ERROR_WITH_ERRNO("Failed to get offset in output "
379 return WIMLIB_ERR_WRITE;
382 /* Are the compression types the same? If so, do a raw copy (copy
383 * without decompressing and recompressing the data). */
384 raw = (wim_resource_compression_type(lte) == out_ctype
385 && out_ctype != WIM_COMPRESSION_TYPE_NONE
386 && !(flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS));
389 flags |= WIMLIB_RESOURCE_FLAG_RAW;
390 bytes_remaining = old_compressed_size;
392 flags &= ~WIMLIB_RESOURCE_FLAG_RAW;
393 bytes_remaining = original_size;
396 /* Empty resource; nothing needs to be done, so just return success. */
397 if (bytes_remaining == 0)
400 /* Buffer for reading chunks for the resource */
401 u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
403 /* If we are writing a compressed resource and not doing a raw copy, we
404 * need to initialize the chunk table */
405 if (out_ctype != WIM_COMPRESSION_TYPE_NONE && !raw) {
406 ret = begin_wim_resource_chunk_tab(lte, out_fp, file_offset,
412 /* If the WIM resource is in an external file, open a FILE * to it so we
413 * don't have to open a temporary one in read_wim_resource() for each
416 ret = prepare_resource_for_read(lte, &ni);
418 ret = prepare_resource_for_read(lte);
423 /* If we aren't doing a raw copy, we will compute the SHA1 message
424 * digest of the resource as we read it, and verify it's the same as the
425 * hash given in the lookup table entry once we've finished reading the
430 compress = get_compress_func(out_ctype);
434 /* While there are still bytes remaining in the WIM resource, read a
435 * chunk of the resource, update SHA1, then write that chunk using the
436 * desired compression type. */
438 u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE);
439 ret = read_wim_resource(lte, buf, to_read, offset, flags);
443 sha1_update(&ctx, buf, to_read);
444 ret = write_wim_resource_chunk(buf, to_read, out_fp,
445 compress, chunk_tab);
448 bytes_remaining -= to_read;
450 } while (bytes_remaining);
452 /* Raw copy: The new compressed size is the same as the old compressed
455 * Using WIM_COMPRESSION_TYPE_NONE: The new compressed size is the
458 * Using a different compression type: Call
459 * finish_wim_resource_chunk_tab() and it will provide the new
463 new_compressed_size = old_compressed_size;
465 if (out_ctype == WIM_COMPRESSION_TYPE_NONE)
466 new_compressed_size = original_size;
468 ret = finish_wim_resource_chunk_tab(chunk_tab, out_fp,
469 &new_compressed_size);
475 /* Verify SHA1 message digest of the resource, unless we are doing a raw
476 * write (in which case we never even saw the uncompressed data). Or,
477 * if the hash we had before is all 0's, just re-set it to be the new
480 u8 md[SHA1_HASH_SIZE];
481 sha1_final(md, &ctx);
482 if (is_zero_hash(lte->hash)) {
483 copy_hash(lte->hash, md);
484 } else if (!hashes_equal(md, lte->hash)) {
485 ERROR("WIM resource has incorrect hash!");
486 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK) {
487 ERROR("We were reading it from `%s'; maybe it changed "
488 "while we were reading it.",
491 ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
496 if (!raw && new_compressed_size >= original_size &&
497 out_ctype != WIM_COMPRESSION_TYPE_NONE)
499 /* Oops! We compressed the resource to larger than the original
500 * size. Write the resource uncompressed instead. */
501 if (fseeko(out_fp, file_offset, SEEK_SET) != 0) {
502 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" "
503 "of output WIM file", file_offset);
504 ret = WIMLIB_ERR_WRITE;
507 ret = write_wim_resource(lte, out_fp, WIM_COMPRESSION_TYPE_NONE,
508 out_res_entry, flags);
512 ret = fflush_and_ftruncate(out_fp, file_offset + out_res_entry->size);
517 out_res_entry->size = new_compressed_size;
518 out_res_entry->original_size = original_size;
519 out_res_entry->offset = file_offset;
520 out_res_entry->flags = lte->resource_entry.flags
521 & ~WIM_RESHDR_FLAG_COMPRESSED;
522 if (out_ctype != WIM_COMPRESSION_TYPE_NONE)
523 out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED;
529 end_wim_resource_read(lte, ni);
531 end_wim_resource_read(lte);
538 #ifdef ENABLE_MULTITHREADED_COMPRESSION
539 struct shared_queue {
542 pthread_mutex_t lock;
549 static int shared_queue_init(struct shared_queue *q, unsigned size)
551 q->array = CALLOC(sizeof(q->array[0]), size);
553 return WIMLIB_ERR_NOMEM;
555 sem_init(&q->filled_slots, 0, 0);
556 sem_init(&q->empty_slots, 0, size);
557 pthread_mutex_init(&q->lock, NULL);
564 static void shared_queue_destroy(struct shared_queue *q)
566 sem_destroy(&q->filled_slots);
567 sem_destroy(&q->empty_slots);
568 pthread_mutex_destroy(&q->lock);
572 static void shared_queue_put(struct shared_queue *q, void *obj)
574 sem_wait(&q->empty_slots);
575 pthread_mutex_lock(&q->lock);
577 q->back = (q->back + 1) % q->size;
578 q->array[q->back] = obj;
580 sem_post(&q->filled_slots);
581 pthread_mutex_unlock(&q->lock);
584 static void *shared_queue_get(struct shared_queue *q)
586 sem_wait(&q->filled_slots);
587 pthread_mutex_lock(&q->lock);
589 void *obj = q->array[q->front];
590 q->array[q->front] = NULL;
591 q->front = (q->front + 1) % q->size;
593 sem_post(&q->empty_slots);
594 pthread_mutex_unlock(&q->lock);
598 struct compressor_thread_params {
599 struct shared_queue *res_to_compress_queue;
600 struct shared_queue *compressed_res_queue;
601 compress_func_t compress;
604 #define MAX_CHUNKS_PER_MSG 2
607 struct lookup_table_entry *lte;
608 u8 *uncompressed_chunks[MAX_CHUNKS_PER_MSG];
609 u8 *out_compressed_chunks[MAX_CHUNKS_PER_MSG];
610 u8 *compressed_chunks[MAX_CHUNKS_PER_MSG];
611 unsigned uncompressed_chunk_sizes[MAX_CHUNKS_PER_MSG];
612 unsigned compressed_chunk_sizes[MAX_CHUNKS_PER_MSG];
614 struct list_head list;
619 static void compress_chunks(struct message *msg, compress_func_t compress)
621 for (unsigned i = 0; i < msg->num_chunks; i++) {
622 DEBUG2("compress chunk %u of %u", i, msg->num_chunks);
623 int ret = compress(msg->uncompressed_chunks[i],
624 msg->uncompressed_chunk_sizes[i],
625 msg->compressed_chunks[i],
626 &msg->compressed_chunk_sizes[i]);
628 msg->out_compressed_chunks[i] = msg->compressed_chunks[i];
630 msg->out_compressed_chunks[i] = msg->uncompressed_chunks[i];
631 msg->compressed_chunk_sizes[i] = msg->uncompressed_chunk_sizes[i];
636 static void *compressor_thread_proc(void *arg)
638 struct compressor_thread_params *params = arg;
639 struct shared_queue *res_to_compress_queue = params->res_to_compress_queue;
640 struct shared_queue *compressed_res_queue = params->compressed_res_queue;
641 compress_func_t compress = params->compress;
644 DEBUG("Compressor thread ready");
645 while ((msg = shared_queue_get(res_to_compress_queue)) != NULL) {
646 compress_chunks(msg, compress);
647 shared_queue_put(compressed_res_queue, msg);
649 DEBUG("Compressor thread terminating");
653 static void show_stream_write_progress(u64 *cur_size, u64 *next_size,
654 u64 total_size, u64 one_percent,
655 unsigned *cur_percent,
656 const struct lookup_table_entry *cur_lte)
658 if (*cur_size >= *next_size) {
659 printf("\r%"PRIu64" MiB of %"PRIu64" MiB "
660 "(uncompressed) written (%u%% done)",
662 total_size >> 20, *cur_percent);
664 *next_size += one_percent;
667 *cur_size += wim_resource_size(cur_lte);
670 static void finish_stream_write_progress(u64 total_size)
672 printf("\r%"PRIu64" MiB of %"PRIu64" MiB "
673 "(uncompressed) written (100%% done)\n",
674 total_size >> 20, total_size >> 20);
678 static int write_stream_list_serial(struct list_head *stream_list,
679 FILE *out_fp, int out_ctype,
680 int write_flags, u64 total_size)
682 struct lookup_table_entry *lte;
685 u64 one_percent = total_size / 100;
688 unsigned cur_percent = 0;
689 int write_resource_flags = 0;
691 if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
692 write_resource_flags |= WIMLIB_RESOURCE_FLAG_RECOMPRESS;
694 list_for_each_entry(lte, stream_list, staging_list) {
695 if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) {
696 show_stream_write_progress(&cur_size, &next_size,
697 total_size, one_percent,
700 ret = write_wim_resource(lte, out_fp, out_ctype,
701 <e->output_resource_entry,
702 write_resource_flags);
706 if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS)
707 finish_stream_write_progress(total_size);
711 #ifdef ENABLE_MULTITHREADED_COMPRESSION
712 static int write_wim_chunks(struct message *msg, FILE *out_fp,
713 struct chunk_table *chunk_tab)
715 for (unsigned i = 0; i < msg->num_chunks; i++) {
716 unsigned chunk_csize = msg->compressed_chunk_sizes[i];
718 DEBUG2("Write wim chunk %u of %u (csize = %u)",
719 i, msg->num_chunks, chunk_csize);
721 if (fwrite(msg->out_compressed_chunks[i], 1, chunk_csize, out_fp)
724 ERROR_WITH_ERRNO("Failed to write WIM chunk");
725 return WIMLIB_ERR_WRITE;
728 *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset;
729 chunk_tab->cur_offset += chunk_csize;
735 * This function is executed by the main thread when the resources are being
736 * compressed in parallel. The main thread is in change of all reading of the
737 * uncompressed data and writing of the compressed data. The compressor threads
738 * *only* do compression from/to in-memory buffers.
740 * Each unit of work given to a compressor thread is up to MAX_CHUNKS_PER_MSG
741 * chunks of compressed data to compress, represented in a `struct message'.
742 * Each message is passed from the main thread to a worker thread through the
743 * res_to_compress_queue, and it is passed back through the
744 * compressed_res_queue.
746 static int main_writer_thread_proc(struct list_head *stream_list,
749 struct shared_queue *res_to_compress_queue,
750 struct shared_queue *compressed_res_queue,
757 struct message msgs[queue_size];
760 // Initially, all the messages are available to use.
761 LIST_HEAD(available_msgs);
762 for (size_t i = 0; i < ARRAY_LEN(msgs); i++)
763 list_add(&msgs[i].list, &available_msgs);
765 // outstanding_resources is the list of resources that currently have
766 // had chunks sent off for compression.
768 // The first stream in outstanding_resources is the stream that is
769 // currently being written (cur_lte).
771 // The last stream in outstanding_resources is the stream that is
772 // currently being read and chunks fed to the compressor threads
775 // Depending on the number of threads and the sizes of the resource,
776 // the outstanding streams list may contain streams between cur_lte and
777 // next_lte that have all their chunks compressed or being compressed,
778 // but haven't been written yet.
780 LIST_HEAD(outstanding_resources);
781 struct list_head *next_resource = stream_list->next;
782 struct lookup_table_entry *next_lte = container_of(next_resource,
783 struct lookup_table_entry,
785 next_resource = next_resource->next;
787 u64 next_num_chunks = wim_resource_chunks(next_lte);
788 INIT_LIST_HEAD(&next_lte->msg_list);
789 list_add_tail(&next_lte->staging_list, &outstanding_resources);
791 // As in write_wim_resource(), each resource we read is checksummed.
792 SHA_CTX next_sha_ctx;
793 sha1_init(&next_sha_ctx);
794 u8 next_hash[SHA1_HASH_SIZE];
796 // Resources that don't need any chunks compressed are added to this
797 // list and written directly by the main thread.
798 LIST_HEAD(my_resources);
800 struct lookup_table_entry *cur_lte = next_lte;
801 struct chunk_table *cur_chunk_tab = NULL;
802 struct lookup_table_entry *lte;
805 u64 one_percent = total_size / 100;
808 unsigned cur_percent = 0;
811 ntfs_inode *ni = NULL;
815 ret = prepare_resource_for_read(next_lte, &ni);
817 ret = prepare_resource_for_read(next_lte);
822 DEBUG("Initializing buffers for uncompressed "
823 "and compressed data (%zu bytes needed)",
824 queue_size * MAX_CHUNKS_PER_MSG * WIM_CHUNK_SIZE * 2);
826 // Pre-allocate all the buffers that will be needed to do the chunk
828 for (size_t i = 0; i < ARRAY_LEN(msgs); i++) {
829 for (size_t j = 0; j < MAX_CHUNKS_PER_MSG; j++) {
830 msgs[i].compressed_chunks[j] = MALLOC(WIM_CHUNK_SIZE);
831 msgs[i].uncompressed_chunks[j] = MALLOC(WIM_CHUNK_SIZE);
832 if (msgs[i].compressed_chunks[j] == NULL ||
833 msgs[i].uncompressed_chunks[j] == NULL)
835 ERROR("Could not allocate enough memory for "
836 "multi-threaded compression");
837 ret = WIMLIB_ERR_NOMEM;
843 // This loop is executed until all resources have been written, except
844 // possibly a few that have been added to the @my_resources list for
847 // Send chunks to the compressor threads until either (a) there
848 // are no more messages available since they were all sent off,
849 // or (b) there are no more resources that need to be
851 while (!list_empty(&available_msgs) && next_lte != NULL) {
853 // Get a message from the available messages
855 msg = container_of(available_msgs.next,
859 // ... and delete it from the available messages
861 list_del(&msg->list);
863 // Initialize the message with the chunks to
865 msg->num_chunks = min(next_num_chunks - next_chunk,
868 msg->complete = false;
869 msg->begin_chunk = next_chunk;
871 unsigned size = WIM_CHUNK_SIZE;
872 for (unsigned i = 0; i < msg->num_chunks; i++) {
874 // Read chunk @next_chunk of the stream into the
875 // message so that a compressor thread can
878 if (next_chunk == next_num_chunks - 1 &&
879 wim_resource_size(next_lte) % WIM_CHUNK_SIZE != 0)
881 size = wim_resource_size(next_lte) % WIM_CHUNK_SIZE;
885 DEBUG2("Read resource (size=%u, offset=%zu)",
886 size, next_chunk * WIM_CHUNK_SIZE);
888 msg->uncompressed_chunk_sizes[i] = size;
890 ret = read_wim_resource(next_lte,
891 msg->uncompressed_chunks[i],
893 next_chunk * WIM_CHUNK_SIZE,
897 sha1_update(&next_sha_ctx,
898 msg->uncompressed_chunks[i], size);
902 // Send the compression request
903 list_add_tail(&msg->list, &next_lte->msg_list);
904 shared_queue_put(res_to_compress_queue, msg);
905 DEBUG2("Compression request sent");
907 if (next_chunk != next_num_chunks)
908 // More chunks to send for this resource
911 // Done sending compression requests for a resource!
912 // Check the SHA1 message digest.
913 DEBUG2("Finalize SHA1 md (next_num_chunks=%zu)", next_num_chunks);
914 sha1_final(next_hash, &next_sha_ctx);
915 if (!hashes_equal(next_lte->hash, next_hash)) {
916 ERROR("WIM resource has incorrect hash!");
917 if (next_lte->resource_location == RESOURCE_IN_FILE_ON_DISK) {
918 ERROR("We were reading it from `%s'; maybe it changed "
919 "while we were reading it.",
920 next_lte->file_on_disk);
922 ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
926 // Advance to the next resource.
928 // If the next resource needs no compression, just write
929 // it with this thread (not now though--- we could be in
930 // the middle of writing another resource.) Keep doing
931 // this until we either get to the end of the resources
932 // list, or we get to a resource that needs compression.
935 if (next_resource == stream_list) {
940 end_wim_resource_read(next_lte, ni);
943 end_wim_resource_read(next_lte);
946 next_lte = container_of(next_resource,
947 struct lookup_table_entry,
949 next_resource = next_resource->next;
950 if ((!(write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
951 && next_lte->resource_location == RESOURCE_IN_WIM
952 && wimlib_get_compression_type(next_lte->wim) == out_ctype)
953 || wim_resource_size(next_lte) == 0)
955 list_add_tail(&next_lte->staging_list,
958 list_add_tail(&next_lte->staging_list,
959 &outstanding_resources);
961 next_num_chunks = wim_resource_chunks(next_lte);
962 sha1_init(&next_sha_ctx);
963 INIT_LIST_HEAD(&next_lte->msg_list);
965 ret = prepare_resource_for_read(next_lte, &ni);
967 ret = prepare_resource_for_read(next_lte);
971 DEBUG2("Updated next_lte");
977 // If there are no outstanding resources, there are no more
978 // resources that need to be written.
979 if (list_empty(&outstanding_resources)) {
980 DEBUG("No outstanding resources! Done");
985 // Get the next message from the queue and process it.
986 // The message will contain 1 or more data chunks that have been
988 DEBUG2("Waiting for message");
989 msg = shared_queue_get(compressed_res_queue);
990 msg->complete = true;
992 DEBUG2("Received msg (begin_chunk=%"PRIu64")", msg->begin_chunk);
994 list_for_each_entry(msg, &cur_lte->msg_list, list) {
995 DEBUG2("complete=%d", msg->complete);
998 // Is this the next chunk in the current resource? If it's not
999 // (i.e., an earlier chunk in a same or different resource
1000 // hasn't been compressed yet), do nothing, and keep this
1001 // message around until all earlier chunks are received.
1003 // Otherwise, write all the chunks we can.
1004 while (!list_empty(&cur_lte->msg_list)
1005 && (msg = container_of(cur_lte->msg_list.next,
1009 DEBUG2("Complete msg (begin_chunk=%"PRIu64")", msg->begin_chunk);
1010 if (msg->begin_chunk == 0) {
1011 DEBUG2("Begin chunk tab");
1012 if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) {
1013 show_stream_write_progress(&cur_size,
1021 // This is the first set of chunks. Leave space
1022 // for the chunk table in the output file.
1023 off_t cur_offset = ftello(out_fp);
1024 if (cur_offset == -1) {
1025 ret = WIMLIB_ERR_WRITE;
1028 ret = begin_wim_resource_chunk_tab(cur_lte,
1036 // Write the compressed chunks from the message.
1037 ret = write_wim_chunks(msg, out_fp, cur_chunk_tab);
1041 list_del(&msg->list);
1043 // This message is available to use for different chunks
1045 list_add(&msg->list, &available_msgs);
1047 // Was this the last chunk of the stream? If so,
1049 if (list_empty(&cur_lte->msg_list) &&
1050 msg->begin_chunk + msg->num_chunks == cur_chunk_tab->num_chunks)
1052 DEBUG2("Finish wim chunk tab");
1054 ret = finish_wim_resource_chunk_tab(cur_chunk_tab,
1061 cur_lte->output_resource_entry.size =
1064 cur_lte->output_resource_entry.original_size =
1065 cur_lte->resource_entry.original_size;
1067 cur_lte->output_resource_entry.offset =
1068 cur_chunk_tab->file_offset;
1070 cur_lte->output_resource_entry.flags =
1071 cur_lte->resource_entry.flags |
1072 WIM_RESHDR_FLAG_COMPRESSED;
1074 FREE(cur_chunk_tab);
1075 cur_chunk_tab = NULL;
1077 struct list_head *next = cur_lte->staging_list.next;
1078 list_del(&cur_lte->staging_list);
1080 if (next == &outstanding_resources) {
1081 DEBUG("No more outstanding resources");
1085 cur_lte = container_of(cur_lte->staging_list.next,
1086 struct lookup_table_entry,
1090 // Since we just finished writing a stream,
1091 // write any streams that have been added to the
1092 // my_resources list for direct writing by the
1093 // main thread (e.g. resources that don't need
1094 // to be compressed because the desired
1095 // compression type is the same as the previous
1096 // compression type).
1097 struct lookup_table_entry *tmp;
1098 list_for_each_entry_safe(lte,
1103 if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) {
1104 show_stream_write_progress(&cur_size,
1112 ret = write_wim_resource(lte,
1115 <e->output_resource_entry,
1117 list_del(<e->staging_list);
1127 end_wim_resource_read(cur_lte, ni);
1129 end_wim_resource_read(cur_lte);
1132 list_for_each_entry(lte, &my_resources, staging_list) {
1133 if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) {
1134 show_stream_write_progress(&cur_size,
1141 ret = write_wim_resource(lte, out_fp,
1143 <e->output_resource_entry,
1148 if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS)
1149 finish_stream_write_progress(total_size);
1151 size_t num_available_msgs = 0;
1152 struct list_head *cur;
1154 list_for_each(cur, &available_msgs) {
1155 num_available_msgs++;
1158 while (num_available_msgs < ARRAY_LEN(msgs)) {
1159 shared_queue_get(compressed_res_queue);
1160 num_available_msgs++;
1164 DEBUG("Freeing messages");
1166 for (size_t i = 0; i < ARRAY_LEN(msgs); i++) {
1167 for (size_t j = 0; j < MAX_CHUNKS_PER_MSG; j++) {
1168 FREE(msgs[i].compressed_chunks[j]);
1169 FREE(msgs[i].uncompressed_chunks[j]);
1173 if (cur_chunk_tab != NULL)
1174 FREE(cur_chunk_tab);
1179 static const char *get_data_type(int ctype)
1182 case WIM_COMPRESSION_TYPE_NONE:
1183 return "uncompressed";
1184 case WIM_COMPRESSION_TYPE_LZX:
1185 return "LZX-compressed";
1186 case WIM_COMPRESSION_TYPE_XPRESS:
1187 return "XPRESS-compressed";
1191 static int write_stream_list_parallel(struct list_head *stream_list,
1192 FILE *out_fp, int out_ctype,
1193 int write_flags, u64 total_size,
1194 unsigned num_threads)
1197 struct shared_queue res_to_compress_queue;
1198 struct shared_queue compressed_res_queue;
1199 pthread_t *compressor_threads = NULL;
1201 if (num_threads == 0) {
1202 long nthreads = sysconf(_SC_NPROCESSORS_ONLN);
1204 WARNING("Could not determine number of processors! Assuming 1");
1207 num_threads = nthreads;
1211 wimlib_assert(stream_list->next != stream_list);
1213 static const double MESSAGES_PER_THREAD = 2.0;
1214 size_t queue_size = (size_t)(num_threads * MESSAGES_PER_THREAD);
1216 DEBUG("Initializing shared queues (queue_size=%zu)", queue_size);
1218 ret = shared_queue_init(&res_to_compress_queue, queue_size);
1222 ret = shared_queue_init(&compressed_res_queue, queue_size);
1224 goto out_destroy_res_to_compress_queue;
1226 struct compressor_thread_params params;
1227 params.res_to_compress_queue = &res_to_compress_queue;
1228 params.compressed_res_queue = &compressed_res_queue;
1229 params.compress = get_compress_func(out_ctype);
1231 compressor_threads = MALLOC(num_threads * sizeof(pthread_t));
1233 for (unsigned i = 0; i < num_threads; i++) {
1234 DEBUG("pthread_create thread %u", i);
1235 ret = pthread_create(&compressor_threads[i], NULL,
1236 compressor_thread_proc, ¶ms);
1239 ERROR_WITH_ERRNO("Failed to create compressor "
1246 if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) {
1247 printf("Writing %s data using %u threads...\n",
1248 get_data_type(out_ctype), num_threads);
1251 ret = main_writer_thread_proc(stream_list,
1254 &res_to_compress_queue,
1255 &compressed_res_queue,
1261 for (unsigned i = 0; i < num_threads; i++)
1262 shared_queue_put(&res_to_compress_queue, NULL);
1264 for (unsigned i = 0; i < num_threads; i++) {
1265 if (pthread_join(compressor_threads[i], NULL)) {
1266 WARNING("Failed to join compressor thread %u: %s",
1267 i, strerror(errno));
1270 FREE(compressor_threads);
1271 shared_queue_destroy(&compressed_res_queue);
1272 out_destroy_res_to_compress_queue:
1273 shared_queue_destroy(&res_to_compress_queue);
1274 if (ret >= 0 && ret != WIMLIB_ERR_NOMEM)
1277 WARNING("Falling back to single-threaded compression");
1278 return write_stream_list_serial(stream_list, out_fp,
1279 out_ctype, write_flags, total_size);
1284 * Write a list of streams to a WIM (@out_fp) using the compression type
1285 * @out_ctype and up to @num_threads compressor threads.
1287 static int write_stream_list(struct list_head *stream_list, FILE *out_fp,
1288 int out_ctype, int write_flags,
1289 unsigned num_threads)
1291 struct lookup_table_entry *lte;
1292 size_t num_streams = 0;
1294 bool compression_needed = false;
1296 list_for_each_entry(lte, stream_list, staging_list) {
1298 total_size += wim_resource_size(lte);
1299 if (!compression_needed
1301 (out_ctype != WIM_COMPRESSION_TYPE_NONE
1302 && (lte->resource_location != RESOURCE_IN_WIM
1303 || wimlib_get_compression_type(lte->wim) != out_ctype
1304 || (write_flags & WIMLIB_WRITE_FLAG_REBUILD)))
1305 && wim_resource_size(lte) != 0)
1306 compression_needed = true;
1309 if (num_streams == 0) {
1310 if (write_flags & WIMLIB_WRITE_FLAG_VERBOSE)
1311 printf("No streams to write\n");
1315 if (write_flags & WIMLIB_WRITE_FLAG_VERBOSE) {
1316 printf("Preparing to write %zu streams "
1317 "(%"PRIu64" total bytes uncompressed)\n",
1318 num_streams, total_size);
1319 printf("Using compression type %s\n",
1320 wimlib_get_compression_type_string(out_ctype));
1323 #ifdef ENABLE_MULTITHREADED_COMPRESSION
1324 if (compression_needed && total_size >= 1000000 && num_threads != 1) {
1325 return write_stream_list_parallel(stream_list, out_fp,
1326 out_ctype, write_flags,
1327 total_size, num_threads);
1332 if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) {
1333 const char *reason = "";
1334 if (!compression_needed)
1335 reason = " (no compression needed)";
1336 printf("Writing %s data using 1 thread%s\n",
1337 get_data_type(out_ctype), reason);
1340 return write_stream_list_serial(stream_list, out_fp,
1341 out_ctype, write_flags,
1347 static int dentry_find_streams_to_write(struct dentry *dentry,
1351 struct list_head *stream_list = w->private;
1352 struct lookup_table_entry *lte;
1353 for (unsigned i = 0; i <= dentry->d_inode->num_ads; i++) {
1354 lte = inode_stream_lte(dentry->d_inode, i, w->lookup_table);
1355 if (lte && ++lte->out_refcnt == 1)
1356 list_add_tail(<e->staging_list, stream_list);
1361 static int find_streams_to_write(WIMStruct *w)
1363 return for_dentry_in_tree(wim_root_dentry(w),
1364 dentry_find_streams_to_write, w);
1367 static int write_wim_streams(WIMStruct *w, int image, int write_flags,
1368 unsigned num_threads)
1371 for_lookup_table_entry(w->lookup_table, lte_zero_out_refcnt, NULL);
1372 LIST_HEAD(stream_list);
1373 w->private = &stream_list;
1374 for_image(w, image, find_streams_to_write);
1375 return write_stream_list(&stream_list, w->out_fp,
1376 wimlib_get_compression_type(w), write_flags,
1381 * Finish writing a WIM file: write the lookup table, xml data, and integrity
1382 * table (optional), then overwrite the WIM header.
1384 * write_flags is a bitwise OR of the following:
1386 * (public) WIMLIB_WRITE_FLAG_CHECK_INTEGRITY:
1387 * Include an integrity table.
1389 * (public) WIMLIB_WRITE_FLAG_SHOW_PROGRESS:
1390 * Show progress information when (if) writing the integrity table.
1392 * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE:
1393 * Don't write the lookup table.
1395 * (private) WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE:
1396 * When (if) writing the integrity table, re-use entries from the
1397 * existing integrity table, if possible.
1399 * (private) WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML:
1400 * After writing the XML data but before writing the integrity
1401 * table, write a temporary WIM header and flush the stream so that
1402 * the WIM is less likely to become corrupted upon abrupt program
1405 * (private) WIMLIB_WRITE_FLAG_FSYNC:
1406 * fsync() the output file before closing it.
1409 int finish_write(WIMStruct *w, int image, int write_flags)
1412 struct wim_header hdr;
1413 FILE *out = w->out_fp;
1415 /* @hdr will be the header for the new WIM. First copy all the data
1416 * from the header in the WIMStruct; then set all the fields that may
1417 * have changed, including the resource entries, boot index, and image
1419 memcpy(&hdr, &w->hdr, sizeof(struct wim_header));
1421 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
1422 ret = write_lookup_table(w->lookup_table, out, &hdr.lookup_table_res_entry);
1427 ret = write_xml_data(w->wim_info, image, out,
1428 (write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE) ?
1429 wim_info_get_total_bytes(w->wim_info) : 0,
1430 &hdr.xml_res_entry);
1434 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
1435 if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) {
1436 struct wim_header checkpoint_hdr;
1437 memcpy(&checkpoint_hdr, &hdr, sizeof(struct wim_header));
1438 memset(&checkpoint_hdr.integrity, 0, sizeof(struct resource_entry));
1439 if (fseeko(out, 0, SEEK_SET) != 0) {
1440 ret = WIMLIB_ERR_WRITE;
1443 ret = write_header(&checkpoint_hdr, out);
1447 if (fflush(out) != 0) {
1448 ERROR_WITH_ERRNO("Can't write data to WIM");
1449 ret = WIMLIB_ERR_WRITE;
1453 if (fseeko(out, 0, SEEK_END) != 0) {
1454 ret = WIMLIB_ERR_WRITE;
1459 off_t old_lookup_table_end;
1460 off_t new_lookup_table_end;
1462 if (write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE) {
1463 old_lookup_table_end = w->hdr.lookup_table_res_entry.offset +
1464 w->hdr.lookup_table_res_entry.size;
1466 old_lookup_table_end = 0;
1468 new_lookup_table_end = hdr.lookup_table_res_entry.offset +
1469 hdr.lookup_table_res_entry.size;
1470 show_progress = ((write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) != 0);
1472 ret = write_integrity_table(out,
1474 new_lookup_table_end,
1475 old_lookup_table_end,
1480 memset(&hdr.integrity, 0, sizeof(struct resource_entry));
1484 * In the WIM header, there is room for the resource entry for a
1485 * metadata resource labeled as the "boot metadata". This entry should
1486 * be zeroed out if there is no bootable image (boot_idx 0). Otherwise,
1487 * it should be a copy of the resource entry for the image that is
1488 * marked as bootable. This is not well documented...
1490 if (hdr.boot_idx == 0 || !w->image_metadata
1491 || (image != WIM_ALL_IMAGES && image != hdr.boot_idx)) {
1492 memset(&hdr.boot_metadata_res_entry, 0,
1493 sizeof(struct resource_entry));
1495 memcpy(&hdr.boot_metadata_res_entry,
1497 hdr.boot_idx - 1].metadata_lte->output_resource_entry,
1498 sizeof(struct resource_entry));
1501 /* Set image count and boot index correctly for single image writes */
1502 if (image != WIM_ALL_IMAGES) {
1503 hdr.image_count = 1;
1504 if (hdr.boot_idx == image)
1510 if (fseeko(out, 0, SEEK_SET) != 0) {
1511 ret = WIMLIB_ERR_WRITE;
1515 ret = write_header(&hdr, out);
1519 if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) {
1520 if (fflush(out) != 0
1521 || fsync(fileno(out)) != 0)
1523 ERROR_WITH_ERRNO("Error flushing data to WIM file");
1524 ret = WIMLIB_ERR_WRITE;
1528 if (fclose(out) != 0) {
1529 ERROR_WITH_ERRNO("Failed to close the WIM file");
1531 ret = WIMLIB_ERR_WRITE;
1537 static void close_wim_writable(WIMStruct *w)
1540 if (fclose(w->out_fp) != 0) {
1541 WARNING("Failed to close output WIM: %s",
1548 /* Open file stream and write dummy header for WIM. */
1549 int begin_write(WIMStruct *w, const char *path, int write_flags)
1552 bool need_readable = false;
1554 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)
1555 need_readable = true;
1557 ret = open_wim_writable(w, path, trunc, need_readable);
1560 /* Write dummy header. It will be overwritten later. */
1561 return write_header(&w->hdr, w->out_fp);
1564 /* Writes a stand-alone WIM to a file. */
1565 WIMLIBAPI int wimlib_write(WIMStruct *w, const char *path,
1566 int image, int write_flags, unsigned num_threads)
1571 return WIMLIB_ERR_INVALID_PARAM;
1573 write_flags &= WIMLIB_WRITE_MASK_PUBLIC;
1575 if (image != WIM_ALL_IMAGES &&
1576 (image < 1 || image > w->hdr.image_count))
1577 return WIMLIB_ERR_INVALID_IMAGE;
1579 if (w->hdr.total_parts != 1) {
1580 ERROR("Cannot call wimlib_write() on part of a split WIM");
1581 return WIMLIB_ERR_SPLIT_UNSUPPORTED;
1584 if (image == WIM_ALL_IMAGES)
1585 DEBUG("Writing all images to `%s'.", path);
1587 DEBUG("Writing image %d to `%s'.", image, path);
1589 ret = begin_write(w, path, write_flags);
1593 ret = write_wim_streams(w, image, write_flags, num_threads);
1597 if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS)
1598 printf("Writing image metadata...\n");
1600 ret = for_image(w, image, write_metadata_resource);
1604 ret = finish_write(w, image, write_flags);
1605 if (ret == 0 && (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS))
1606 printf("Successfully wrote `%s'\n", path);
1608 close_wim_writable(w);
1612 static int lte_overwrite_prepare(struct lookup_table_entry *lte,
1615 memcpy(<e->output_resource_entry, <e->resource_entry,
1616 sizeof(struct resource_entry));
1617 lte->out_refcnt = 0;
1621 static int check_resource_offset(struct lookup_table_entry *lte, void *arg)
1623 off_t end_offset = *(u64*)arg;
1625 wimlib_assert(lte->out_refcnt <= lte->refcnt);
1626 if (lte->out_refcnt < lte->refcnt) {
1627 if (lte->resource_entry.offset + lte->resource_entry.size > end_offset) {
1628 ERROR("The following resource is after the XML data:");
1629 print_lookup_table_entry(lte);
1630 return WIMLIB_ERR_RESOURCE_ORDER;
1636 static int find_new_streams(struct lookup_table_entry *lte, void *arg)
1638 if (lte->out_refcnt == lte->refcnt)
1639 list_add(<e->staging_list, (struct list_head*)arg);
1641 lte->out_refcnt = lte->refcnt;
1646 * Overwrite a WIM, possibly appending streams to it.
1648 * A WIM looks like (or is supposed to look like) the following:
1650 * Header (212 bytes)
1651 * Streams and metadata resources (variable size)
1652 * Lookup table (variable size)
1653 * XML data (variable size)
1654 * Integrity table (optional) (variable size)
1656 * If we are not adding any streams or metadata resources, the lookup table is
1657 * unchanged--- so we only need to overwrite the XML data, integrity table, and
1658 * header. This operation is potentially unsafe if the program is abruptly
1659 * terminated while the XML data or integrity table are being overwritten, but
1660 * before the new header has been written. To partially alleviate this problem,
1661 * a special flag (WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) is passed to
1662 * finish_write() to cause a temporary WIM header to be written after the XML
1663 * data has been written. This may prevent the WIM from becoming corrupted if
1664 * the program is terminated while the integrity table is being calculated (but
1665 * no guarantees, due to write re-ordering...).
1667 * If we are adding new streams or images (metadata resources), the lookup table
1668 * needs to be changed, and those streams need to be written. In this case, we
1669 * try to perform a safe update of the WIM file by writing the streams *after*
1670 * the end of the previous WIM, then writing the new lookup table, XML data, and
1671 * (optionally) integrity table following the new streams. This will produce a
1672 * layout like the following:
1674 * Header (212 bytes)
1675 * (OLD) Streams and metadata resources (variable size)
1676 * (OLD) Lookup table (variable size)
1677 * (OLD) XML data (variable size)
1678 * (OLD) Integrity table (optional) (variable size)
1679 * (NEW) Streams and metadata resources (variable size)
1680 * (NEW) Lookup table (variable size)
1681 * (NEW) XML data (variable size)
1682 * (NEW) Integrity table (optional) (variable size)
1684 * At all points, the WIM is valid as nothing points to the new data yet. Then,
1685 * the header is overwritten to point to the new lookup table, XML data, and
1686 * integrity table, to produce the following layout:
1688 * Header (212 bytes)
1689 * Streams and metadata resources (variable size)
1690 * Nothing (variable size)
1691 * More Streams and metadata resources (variable size)
1692 * Lookup table (variable size)
1693 * XML data (variable size)
1694 * Integrity table (optional) (variable size)
1696 * This method allows an image to be appended to a large WIM very quickly, and
1697 * is is crash-safe except in the case of write re-ordering, but the
1698 * disadvantage is that a small hole is left in the WIM where the old lookup
1699 * table, xml data, and integrity table were. (These usually only take up a
1700 * small amount of space compared to the streams, however.
1702 static int overwrite_wim_inplace(WIMStruct *w, int write_flags,
1703 unsigned num_threads,
1704 int modified_image_idx)
1707 struct list_head stream_list;
1710 DEBUG("Overwriting `%s' in-place", w->filename);
1712 /* Make sure that the integrity table (if present) is after the XML
1713 * data, and that there are no stream resources, metadata resources, or
1714 * lookup tables after the XML data. Otherwise, these data would be
1716 if (w->hdr.integrity.offset != 0 &&
1717 w->hdr.integrity.offset < w->hdr.xml_res_entry.offset) {
1718 ERROR("Didn't expect the integrity table to be before the XML data");
1719 return WIMLIB_ERR_RESOURCE_ORDER;
1722 if (w->hdr.lookup_table_res_entry.offset > w->hdr.xml_res_entry.offset) {
1723 ERROR("Didn't expect the lookup table to be after the XML data");
1724 return WIMLIB_ERR_RESOURCE_ORDER;
1727 DEBUG("Identifying newly added streams");
1728 for_lookup_table_entry(w->lookup_table, lte_overwrite_prepare, NULL);
1729 INIT_LIST_HEAD(&stream_list);
1730 for (int i = modified_image_idx; i < w->hdr.image_count; i++) {
1731 DEBUG("Identifiying streams in image %d", i + 1);
1732 wimlib_assert(w->image_metadata[i].modified);
1733 wimlib_assert(!w->image_metadata[i].has_been_mounted_rw);
1734 wimlib_assert(w->image_metadata[i].root_dentry != NULL);
1735 wimlib_assert(w->image_metadata[i].metadata_lte != NULL);
1736 w->private = &stream_list;
1737 for_dentry_in_tree(w->image_metadata[i].root_dentry,
1738 dentry_find_streams_to_write, w);
1741 if (w->hdr.integrity.offset)
1742 old_wim_end = w->hdr.integrity.offset + w->hdr.integrity.size;
1744 old_wim_end = w->hdr.xml_res_entry.offset + w->hdr.xml_res_entry.size;
1746 ret = for_lookup_table_entry(w->lookup_table, check_resource_offset,
1751 if (modified_image_idx == w->hdr.image_count && !w->deletion_occurred) {
1752 /* If no images have been modified and no images have been
1753 * deleted, a new lookup table does not need to be written. */
1754 wimlib_assert(list_empty(&stream_list));
1755 old_wim_end = w->hdr.lookup_table_res_entry.offset +
1756 w->hdr.lookup_table_res_entry.size;
1757 write_flags |= WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE |
1758 WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML;
1761 INIT_LIST_HEAD(&stream_list);
1762 for_lookup_table_entry(w->lookup_table, find_new_streams,
1765 ret = open_wim_writable(w, w->filename, false,
1766 (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0);
1770 if (fseeko(w->out_fp, old_wim_end, SEEK_SET) != 0) {
1771 ERROR_WITH_ERRNO("Can't seek to end of WIM");
1772 return WIMLIB_ERR_WRITE;
1775 if (!list_empty(&stream_list)) {
1776 DEBUG("Writing newly added streams (offset = %"PRIu64")",
1778 ret = write_stream_list(&stream_list, w->out_fp,
1779 wimlib_get_compression_type(w),
1780 write_flags, num_threads);
1784 DEBUG("No new streams were added");
1787 for (int i = modified_image_idx; i < w->hdr.image_count; i++) {
1788 select_wim_image(w, i + 1);
1789 ret = write_metadata_resource(w);
1793 write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE;
1794 ret = finish_write(w, WIM_ALL_IMAGES, write_flags);
1796 close_wim_writable(w);
1798 WARNING("Truncating `%s' to its original size (%"PRIu64" bytes)",
1799 w->filename, old_wim_end);
1800 truncate(w->filename, old_wim_end);
1805 static int overwrite_wim_via_tmpfile(WIMStruct *w, int write_flags,
1806 unsigned num_threads)
1808 size_t wim_name_len;
1811 DEBUG("Overwrining `%s' via a temporary file", w->filename);
1813 /* Write the WIM to a temporary file in the same directory as the
1815 wim_name_len = strlen(w->filename);
1816 char tmpfile[wim_name_len + 10];
1817 memcpy(tmpfile, w->filename, wim_name_len);
1818 randomize_char_array_with_alnum(tmpfile + wim_name_len, 9);
1819 tmpfile[wim_name_len + 9] = '\0';
1821 ret = wimlib_write(w, tmpfile, WIM_ALL_IMAGES,
1822 write_flags | WIMLIB_WRITE_FLAG_FSYNC,
1825 ERROR("Failed to write the WIM file `%s'", tmpfile);
1829 /* Close the original WIM file that was opened for reading. */
1830 if (w->fp != NULL) {
1835 DEBUG("Renaming `%s' to `%s'", tmpfile, w->filename);
1837 /* Rename the new file to the old file .*/
1838 if (rename(tmpfile, w->filename) != 0) {
1839 ERROR_WITH_ERRNO("Failed to rename `%s' to `%s'",
1840 tmpfile, w->filename);
1841 ret = WIMLIB_ERR_RENAME;
1845 if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS)
1846 printf("Successfully renamed `%s' to `%s'\n", tmpfile, w->filename);
1848 /* Re-open the WIM read-only. */
1849 w->fp = fopen(w->filename, "rb");
1850 if (w->fp == NULL) {
1851 ret = WIMLIB_ERR_REOPEN;
1852 WARNING("Failed to re-open `%s' read-only: %s",
1853 w->filename, strerror(errno));
1857 /* Remove temporary file. */
1858 if (unlink(tmpfile) != 0)
1859 WARNING("Failed to remove `%s': %s", tmpfile, strerror(errno));
1864 * Writes a WIM file to the original file that it was read from, overwriting it.
1866 WIMLIBAPI int wimlib_overwrite(WIMStruct *w, int write_flags,
1867 unsigned num_threads)
1870 return WIMLIB_ERR_INVALID_PARAM;
1872 write_flags &= WIMLIB_WRITE_MASK_PUBLIC;
1875 return WIMLIB_ERR_NO_FILENAME;
1877 if (w->hdr.total_parts != 1) {
1878 ERROR("Cannot modify a split WIM");
1879 return WIMLIB_ERR_SPLIT_UNSUPPORTED;
1882 if ((!w->deletion_occurred || (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE))
1883 && !(write_flags & WIMLIB_WRITE_FLAG_REBUILD))
1885 int i, modified_image_idx;
1886 for (i = 0; i < w->hdr.image_count && !w->image_metadata[i].modified; i++)
1888 modified_image_idx = i;
1889 for (; i < w->hdr.image_count && w->image_metadata[i].modified &&
1890 !w->image_metadata[i].has_been_mounted_rw; i++)
1892 if (i == w->hdr.image_count) {
1893 return overwrite_wim_inplace(w, write_flags, num_threads,
1894 modified_image_idx);
1897 return overwrite_wim_via_tmpfile(w, write_flags, num_threads);
1901 WIMLIBAPI int wimlib_overwrite_xml_and_header(WIMStruct *wim, int write_flags)
1903 return wimlib_overwrite(wim, write_flags, 1);