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/.
31 #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK)
32 /* On BSD, this should be included before "wimlib/list.h" so that "wimlib/list.h" can
33 * overwrite the LIST_HEAD macro. */
34 # include <sys/file.h>
37 #include "wimlib/chunk_compressor.h"
38 #include "wimlib/endianness.h"
39 #include "wimlib/error.h"
40 #include "wimlib/file_io.h"
41 #include "wimlib/header.h"
42 #include "wimlib/integrity.h"
43 #include "wimlib/lookup_table.h"
44 #include "wimlib/metadata.h"
45 #include "wimlib/resource.h"
47 # include "wimlib/win32.h" /* win32_rename_replacement() */
49 #include "wimlib/write.h"
50 #include "wimlib/xml.h"
61 /* wimlib internal flags used when writing resources. */
62 #define WRITE_RESOURCE_FLAG_RECOMPRESS 0x00000001
63 #define WRITE_RESOURCE_FLAG_PIPABLE 0x00000002
64 #define WRITE_RESOURCE_FLAG_PACK_STREAMS 0x00000004
67 write_flags_to_resource_flags(int write_flags)
69 int write_resource_flags = 0;
71 if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
72 write_resource_flags |= WRITE_RESOURCE_FLAG_RECOMPRESS;
73 if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
74 write_resource_flags |= WRITE_RESOURCE_FLAG_PIPABLE;
75 if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS)
76 write_resource_flags |= WRITE_RESOURCE_FLAG_PACK_STREAMS;
77 return write_resource_flags;
80 struct filter_context {
85 /* Determine specified stream should be filtered out from the write.
89 * < 0 : The stream should be hard-filtered; that is, not included in the
91 * 0 : The stream should not be filtered out.
92 * > 0 : The stream should be soft-filtered; that is, it already exists in the
93 * WIM file and may not need to be written again.
96 stream_filtered(const struct wim_lookup_table_entry *lte,
97 const struct filter_context *ctx)
99 int write_flags = ctx->write_flags;
100 WIMStruct *wim = ctx->wim;
105 if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE &&
106 lte->resource_location == RESOURCE_IN_WIM &&
107 lte->rspec->wim == wim)
110 if (write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS &&
111 lte->resource_location == RESOURCE_IN_WIM &&
112 lte->rspec->wim != wim)
119 stream_hard_filtered(const struct wim_lookup_table_entry *lte,
120 struct filter_context *ctx)
122 return stream_filtered(lte, ctx) < 0;
126 may_soft_filter_streams(const struct filter_context *ctx)
130 return ctx->write_flags & WIMLIB_WRITE_FLAG_OVERWRITE;
134 may_hard_filter_streams(const struct filter_context *ctx)
138 return ctx->write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS;
142 may_filter_streams(const struct filter_context *ctx)
144 return (may_soft_filter_streams(ctx) ||
145 may_hard_filter_streams(ctx));
149 /* Return true if the specified resource is compressed and the compressed data
150 * can be reused with the specified output parameters. */
152 can_raw_copy(const struct wim_lookup_table_entry *lte,
153 int write_resource_flags, int out_ctype, u32 out_chunk_size)
155 const struct wim_resource_spec *rspec;
157 if (write_resource_flags & WRITE_RESOURCE_FLAG_RECOMPRESS)
160 if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE)
163 if (lte->resource_location != RESOURCE_IN_WIM)
168 if (rspec->is_pipable != !!(write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE))
172 if (rspec->flags & WIM_RESHDR_FLAG_COMPRESSED) {
173 /* Normal compressed resource: Must use same compression type
175 return (rspec->compression_type == out_ctype &&
176 rspec->chunk_size == out_chunk_size);
179 /* XXX: For compatibility, we can't allow multiple packed resources per
182 if ((rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) &&
183 (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
185 /* Packed resource: Such resources may contain multiple streams,
186 * and in general only a subset of them need to be written. As
187 * a heuristic, re-use the raw data if at least half the
188 * uncompressed size is being written. */
190 /* Note: packed resources contain a header that specifies the
191 * compression type and chunk size; therefore we don't need to
192 * check if they are compatible with @out_ctype and
193 * @out_chunk_size. */
195 struct wim_lookup_table_entry *res_stream;
198 list_for_each_entry(res_stream, &rspec->stream_list, rspec_node)
199 if (res_stream->will_be_in_output_wim)
200 write_size += res_stream->size;
202 return (write_size > rspec->uncompressed_size / 2);
210 filter_resource_flags(u8 flags)
212 return (flags & ~(WIM_RESHDR_FLAG_PACKED_STREAMS |
213 WIM_RESHDR_FLAG_COMPRESSED |
214 WIM_RESHDR_FLAG_SPANNED |
215 WIM_RESHDR_FLAG_FREE));
219 stream_set_out_reshdr_for_reuse(struct wim_lookup_table_entry *lte)
221 const struct wim_resource_spec *rspec;
223 wimlib_assert(lte->resource_location == RESOURCE_IN_WIM);
226 if (rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
228 wimlib_assert(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS);
230 lte->out_reshdr.offset_in_wim = lte->offset_in_res;
231 lte->out_reshdr.uncompressed_size = 0;
232 lte->out_reshdr.size_in_wim = lte->size;
234 lte->out_res_offset_in_wim = rspec->offset_in_wim;
235 lte->out_res_size_in_wim = rspec->size_in_wim;
236 /*lte->out_res_uncompressed_size = rspec->uncompressed_size;*/
238 wimlib_assert(!(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS));
240 lte->out_reshdr.offset_in_wim = rspec->offset_in_wim;
241 lte->out_reshdr.uncompressed_size = rspec->uncompressed_size;
242 lte->out_reshdr.size_in_wim = rspec->size_in_wim;
244 lte->out_reshdr.flags = lte->flags;
248 /* Write the header for a stream in a pipable WIM. */
250 write_pwm_stream_header(const struct wim_lookup_table_entry *lte,
251 struct filedes *out_fd,
252 int additional_reshdr_flags)
254 struct pwm_stream_hdr stream_hdr;
258 stream_hdr.magic = cpu_to_le64(PWM_STREAM_MAGIC);
259 stream_hdr.uncompressed_size = cpu_to_le64(lte->size);
260 if (additional_reshdr_flags & PWM_RESHDR_FLAG_UNHASHED) {
261 zero_out_hash(stream_hdr.hash);
263 wimlib_assert(!lte->unhashed);
264 copy_hash(stream_hdr.hash, lte->hash);
267 reshdr_flags = filter_resource_flags(lte->flags);
268 reshdr_flags |= additional_reshdr_flags;
269 stream_hdr.flags = cpu_to_le32(reshdr_flags);
270 ret = full_write(out_fd, &stream_hdr, sizeof(stream_hdr));
272 ERROR_WITH_ERRNO("Write error");
276 struct write_streams_progress_data {
277 wimlib_progress_func_t progress_func;
278 union wimlib_progress_info progress;
279 uint64_t next_progress;
280 WIMStruct *prev_wim_part;
284 do_write_streams_progress(struct write_streams_progress_data *progress_data,
285 struct wim_lookup_table_entry *cur_stream,
290 union wimlib_progress_info *progress = &progress_data->progress;
294 progress->write_streams.total_bytes -= complete_size;
295 progress->write_streams.total_streams -= complete_count;
296 if (progress_data->next_progress != ~(uint64_t)0 &&
297 progress_data->next_progress > progress->write_streams.total_bytes)
299 progress_data->next_progress = progress->write_streams.total_bytes;
302 progress->write_streams.completed_bytes += complete_size;
303 progress->write_streams.completed_streams += complete_count;
306 new_wim_part = false;
307 if (cur_stream->resource_location == RESOURCE_IN_WIM &&
308 cur_stream->rspec->wim != progress_data->prev_wim_part)
310 if (progress_data->prev_wim_part) {
312 progress->write_streams.completed_parts++;
314 progress_data->prev_wim_part = cur_stream->rspec->wim;
317 if (progress_data->progress_func
318 && (progress->write_streams.completed_bytes >= progress_data->next_progress
321 progress_data->progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
323 if (progress_data->next_progress == progress->write_streams.total_bytes) {
324 progress_data->next_progress = ~(uint64_t)0;
326 progress_data->next_progress =
327 min(progress->write_streams.total_bytes,
328 progress->write_streams.completed_bytes +
329 progress->write_streams.total_bytes / 100);
334 struct write_streams_ctx {
335 /* File descriptor the streams are being written to. */
336 struct filedes *out_fd;
338 /* Lookup table for the WIMStruct on whose behalf the streams are being
340 struct wim_lookup_table *lookup_table;
342 /* Compression format to use. */
345 /* Maximum uncompressed chunk size in compressed resources to use. */
348 /* Flags that affect how the streams will be written. */
349 int write_resource_flags;
351 /* Data used for issuing WRITE_STREAMS progress. */
352 struct write_streams_progress_data progress_data;
354 struct filter_context *filter_ctx;
356 /* Upper bound on the total number of bytes that need to be compressed.
358 u64 num_bytes_to_compress;
360 /* Pointer to the chunk_compressor implementation being used for
361 * compressing chunks of data, or NULL if chunks are being written
363 struct chunk_compressor *compressor;
365 /* Buffer for dividing the read data into chunks of size
366 * @out_chunk_size. */
369 /* Number of bytes in @chunk_buf that are currently filled. */
370 size_t chunk_buf_filled;
372 /* List of streams that currently have chunks being compressed. */
373 struct list_head pending_streams;
375 /* List of streams in the resource pack. Streams are moved here after
376 * @pending_streams only when writing a packed resource. */
377 struct list_head pack_streams;
379 /* Set to true if the stream currently being read was a duplicate, and
380 * therefore the corresponding stream entry needs to be freed once the
381 * read finishes. (In this case we add the duplicate entry to
382 * pending_streams rather than the entry being read.) */
383 bool stream_was_duplicate;
385 /* Current uncompressed offset in the stream being read. */
386 u64 cur_read_stream_offset;
388 /* Uncompressed size of the stream currently being read. */
389 u64 cur_read_stream_size;
391 /* Current uncompressed offset in the stream being written. */
392 u64 cur_write_stream_offset;
394 /* Uncompressed size of resource currently being written. */
395 u64 cur_write_res_size;
397 /* Array that is filled in with compressed chunk sizes as a resource is
401 /* Index of next entry in @chunk_csizes to fill in. */
404 /* Number of entries in @chunk_csizes currently allocated. */
405 size_t num_alloc_chunks;
407 /* Offset in the output file of the start of the chunks of the resource
408 * currently being written. */
409 u64 chunks_start_offset;
412 /* Reserve space for the chunk table and prepare to accumulate the chunk table
415 begin_chunk_table(struct write_streams_ctx *ctx, u64 res_expected_size)
417 u64 expected_num_chunks;
418 u64 expected_num_chunk_entries;
422 /* Calculate the number of chunks and chunk entries that should be
423 * needed for the resource. These normally will be the final values,
424 * but in PACKED_STREAMS mode some of the streams we're planning to
425 * write into the resource may be duplicates, and therefore discarded,
426 * potentially decreasing the number of chunk entries needed. */
427 expected_num_chunks = DIV_ROUND_UP(res_expected_size, ctx->out_chunk_size);
428 expected_num_chunk_entries = expected_num_chunks;
429 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
430 expected_num_chunk_entries--;
432 /* Make sure the chunk_csizes array is long enough to store the
433 * compressed size of each chunk. */
434 if (expected_num_chunks > ctx->num_alloc_chunks) {
435 u64 new_length = expected_num_chunks + 50;
437 if ((size_t)new_length != new_length) {
438 ERROR("Resource size too large (%"PRIu64" bytes!",
440 return WIMLIB_ERR_NOMEM;
443 FREE(ctx->chunk_csizes);
444 ctx->chunk_csizes = MALLOC(new_length * sizeof(ctx->chunk_csizes[0]));
445 if (ctx->chunk_csizes == NULL) {
446 ctx->num_alloc_chunks = 0;
447 return WIMLIB_ERR_NOMEM;
449 ctx->num_alloc_chunks = new_length;
452 ctx->chunk_index = 0;
454 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE)) {
455 /* Reserve space for the chunk table in the output file. In the
456 * case of packed resources this reserves the upper bound for
457 * the needed space, not necessarily the exact space which will
458 * prove to be needed. At this point, we just use @chunk_csizes
459 * for a buffer of 0's because the actual compressed chunk sizes
461 reserve_size = expected_num_chunk_entries *
462 get_chunk_entry_size(res_expected_size,
463 0 != (ctx->write_resource_flags &
464 WIM_RESHDR_FLAG_PACKED_STREAMS));
465 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)
466 reserve_size += sizeof(struct alt_chunk_table_header_disk);
467 memset(ctx->chunk_csizes, 0, reserve_size);
468 ret = full_write(ctx->out_fd, ctx->chunk_csizes, reserve_size);
476 begin_write_resource(struct write_streams_ctx *ctx, u64 res_expected_size)
480 wimlib_assert(res_expected_size != 0);
482 if (ctx->compressor != NULL) {
483 ret = begin_chunk_table(ctx, res_expected_size);
488 /* Output file descriptor is now positioned at the offset at which to
489 * write the first chunk of the resource. */
490 ctx->chunks_start_offset = ctx->out_fd->offset;
491 ctx->cur_write_stream_offset = 0;
492 ctx->cur_write_res_size = res_expected_size;
497 end_chunk_table(struct write_streams_ctx *ctx, u64 res_actual_size,
498 u64 *res_start_offset_ret, u64 *res_store_size_ret)
500 size_t actual_num_chunks;
501 size_t actual_num_chunk_entries;
502 size_t chunk_entry_size;
505 actual_num_chunks = ctx->chunk_index;
506 actual_num_chunk_entries = actual_num_chunks;
507 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
508 actual_num_chunk_entries--;
510 chunk_entry_size = get_chunk_entry_size(res_actual_size,
511 0 != (ctx->write_resource_flags &
512 WIM_RESHDR_FLAG_PACKED_STREAMS));
514 typedef le64 __attribute__((may_alias)) aliased_le64_t;
515 typedef le32 __attribute__((may_alias)) aliased_le32_t;
517 if (chunk_entry_size == 4) {
518 aliased_le32_t *entries = (aliased_le32_t*)ctx->chunk_csizes;
520 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
521 for (size_t i = 0; i < actual_num_chunk_entries; i++)
522 entries[i] = cpu_to_le32(ctx->chunk_csizes[i]);
524 u32 offset = ctx->chunk_csizes[0];
525 for (size_t i = 0; i < actual_num_chunk_entries; i++) {
526 u32 next_size = ctx->chunk_csizes[i + 1];
527 entries[i] = cpu_to_le32(offset);
532 aliased_le64_t *entries = (aliased_le64_t*)ctx->chunk_csizes;
534 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
535 for (size_t i = 0; i < actual_num_chunk_entries; i++)
536 entries[i] = cpu_to_le64(ctx->chunk_csizes[i]);
538 u64 offset = ctx->chunk_csizes[0];
539 for (size_t i = 0; i < actual_num_chunk_entries; i++) {
540 u64 next_size = ctx->chunk_csizes[i + 1];
541 entries[i] = cpu_to_le64(offset);
547 size_t chunk_table_size = actual_num_chunk_entries * chunk_entry_size;
548 u64 res_start_offset;
551 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
552 ret = full_write(ctx->out_fd, ctx->chunk_csizes, chunk_table_size);
555 res_end_offset = ctx->out_fd->offset;
556 res_start_offset = ctx->chunks_start_offset;
558 res_end_offset = ctx->out_fd->offset;
560 u64 chunk_table_offset;
562 chunk_table_offset = ctx->chunks_start_offset - chunk_table_size;
564 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
565 struct alt_chunk_table_header_disk hdr;
567 hdr.res_usize = cpu_to_le64(res_actual_size);
568 hdr.chunk_size = cpu_to_le32(ctx->out_chunk_size);
569 hdr.compression_format = cpu_to_le32(ctx->out_ctype);
571 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 1);
572 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 2);
573 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3);
575 ret = full_pwrite(ctx->out_fd, &hdr, sizeof(hdr),
576 chunk_table_offset - sizeof(hdr));
579 res_start_offset = chunk_table_offset - sizeof(hdr);
581 res_start_offset = chunk_table_offset;
584 ret = full_pwrite(ctx->out_fd, ctx->chunk_csizes,
585 chunk_table_size, chunk_table_offset);
590 *res_start_offset_ret = res_start_offset;
591 *res_store_size_ret = res_end_offset - res_start_offset;
596 ERROR_WITH_ERRNO("Write error");
600 /* Finish writing a WIM resource by writing or updating the chunk table (if not
601 * writing the data uncompressed) and loading its metadata into @out_reshdr. */
603 end_write_resource(struct write_streams_ctx *ctx, struct wim_reshdr *out_reshdr)
607 u64 res_uncompressed_size;
608 u64 res_offset_in_wim;
610 wimlib_assert(ctx->cur_write_stream_offset == ctx->cur_write_res_size ||
611 (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS));
612 res_uncompressed_size = ctx->cur_write_res_size;
614 if (ctx->compressor) {
615 ret = end_chunk_table(ctx, res_uncompressed_size,
616 &res_offset_in_wim, &res_size_in_wim);
620 res_offset_in_wim = ctx->chunks_start_offset;
621 res_size_in_wim = ctx->out_fd->offset - res_offset_in_wim;
623 out_reshdr->uncompressed_size = res_uncompressed_size;
624 out_reshdr->size_in_wim = res_size_in_wim;
625 out_reshdr->offset_in_wim = res_offset_in_wim;
626 DEBUG("Finished writing resource: %"PRIu64" => %"PRIu64" @ %"PRIu64"",
627 res_uncompressed_size, res_size_in_wim, res_offset_in_wim);
631 /* Begin processing a stream for writing. */
633 write_stream_begin_read(struct wim_lookup_table_entry *lte,
634 bool is_partial_res, void *_ctx)
636 struct write_streams_ctx *ctx = _ctx;
639 wimlib_assert(lte->size > 0);
641 ctx->cur_read_stream_offset = 0;
642 ctx->cur_read_stream_size = lte->size;
644 /* As an optimization, we allow some streams to be "unhashed", meaning
645 * their SHA1 message digests are unknown. This is the case with
646 * streams that are added by scanning a directry tree with
647 * wimlib_add_image(), for example. Since WIM uses single-instance
648 * streams, we don't know whether such each such stream really need to
649 * written until it is actually checksummed, unless it has a unique
650 * size. In such cases we read and checksum the stream in this
651 * function, thereby advancing ahead of read_stream_list(), which will
652 * still provide the data again to write_stream_process_chunk(). This
653 * is okay because an unhashed stream cannot be in a WIM resource, which
654 * might be costly to decompress. */
655 ctx->stream_was_duplicate = false;
656 if (ctx->lookup_table != NULL && lte->unhashed && !lte->unique_size) {
658 wimlib_assert(!is_partial_res);
660 struct wim_lookup_table_entry *lte_new;
662 ret = hash_unhashed_stream(lte, ctx->lookup_table, <e_new);
665 if (lte_new != lte) {
666 /* Duplicate stream detected. */
668 if (lte_new->will_be_in_output_wim ||
669 stream_filtered(lte_new, ctx->filter_ctx))
671 /* The duplicate stream is already being
672 * included in the output WIM, or it would be
673 * filtered out if it had been. Skip writing
674 * this stream (and reading it again) entirely,
675 * passing its output reference count to the
676 * duplicate stream in the former case. */
677 DEBUG("Discarding duplicate stream of "
678 "length %"PRIu64, lte->size);
679 do_write_streams_progress(&ctx->progress_data,
682 list_del(<e->write_streams_list);
683 list_del(<e->lookup_table_list);
684 if (lte_new->will_be_in_output_wim)
685 lte_new->out_refcnt += lte->out_refcnt;
686 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)
687 ctx->cur_write_res_size -= lte->size;
688 free_lookup_table_entry(lte);
689 return BEGIN_STREAM_STATUS_SKIP_STREAM;
691 /* The duplicate stream can validly be written,
692 * but was not marked as such. Discard the
693 * current stream entry and use the duplicate,
694 * but actually freeing the current entry must
695 * wait until read_stream_list() has finished
696 * reading its data. */
697 DEBUG("Stream duplicate, but not already "
698 "selected for writing.");
699 list_replace(<e->write_streams_list,
700 <e_new->write_streams_list);
701 list_replace(<e->lookup_table_list,
702 <e_new->lookup_table_list);
703 lte_new->out_refcnt = lte->out_refcnt;
704 lte_new->will_be_in_output_wim = 1;
705 ctx->stream_was_duplicate = true;
710 list_move_tail(<e->write_streams_list, &ctx->pending_streams);
714 /* Rewrite a stream that was just written compressed as uncompressed instead.
715 * This function is optional, but if a stream did not compress to less than its
716 * original size, it might as well be written uncompressed. */
718 write_stream_uncompressed(struct wim_lookup_table_entry *lte,
719 struct filedes *out_fd)
722 u64 begin_offset = lte->out_reshdr.offset_in_wim;
723 u64 end_offset = out_fd->offset;
725 if (filedes_seek(out_fd, begin_offset) == -1)
728 ret = extract_full_stream_to_fd(lte, out_fd);
730 /* Error reading the uncompressed data. */
731 if (out_fd->offset == begin_offset &&
732 filedes_seek(out_fd, end_offset) != -1)
734 /* Nothing was actually written yet, and we successfully
735 * seeked to the end of the compressed resource, so
736 * don't issue a hard error; just keep the compressed
737 * resource instead. */
738 WARNING("Recovered compressed stream of "
739 "size %"PRIu64", continuing on.",
746 wimlib_assert(out_fd->offset - begin_offset == lte->size);
748 if (out_fd->offset < end_offset &&
749 0 != ftruncate(out_fd->fd, out_fd->offset))
751 ERROR_WITH_ERRNO("Can't truncate output file to "
752 "offset %"PRIu64, out_fd->offset);
753 return WIMLIB_ERR_WRITE;
756 lte->out_reshdr.size_in_wim = lte->size;
757 lte->out_reshdr.flags &= ~(WIM_RESHDR_FLAG_COMPRESSED |
758 WIM_RESHDR_FLAG_PACKED_STREAMS);
762 /* Write the next chunk of (typically compressed) data to the output WIM,
763 * handling the writing of the chunk table. */
765 write_chunk(struct write_streams_ctx *ctx, const void *cchunk,
766 size_t csize, size_t usize)
770 struct wim_lookup_table_entry *lte;
771 u32 completed_stream_count;
774 lte = list_entry(ctx->pending_streams.next,
775 struct wim_lookup_table_entry, write_streams_list);
777 if (ctx->cur_write_stream_offset == 0 &&
778 !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
780 /* Starting to write a new stream in non-packed mode. */
782 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
783 int additional_reshdr_flags = 0;
784 if (ctx->compressor != NULL)
785 additional_reshdr_flags |= WIM_RESHDR_FLAG_COMPRESSED;
787 DEBUG("Writing pipable WIM stream header "
788 "(offset=%"PRIu64")", ctx->out_fd->offset);
790 ret = write_pwm_stream_header(lte, ctx->out_fd,
791 additional_reshdr_flags);
796 ret = begin_write_resource(ctx, lte->size);
801 if (ctx->compressor != NULL) {
802 /* Record the compresed chunk size. */
803 wimlib_assert(ctx->chunk_index < ctx->num_alloc_chunks);
804 ctx->chunk_csizes[ctx->chunk_index++] = csize;
806 /* If writing a pipable WIM, before the chunk data write a chunk
807 * header that provides the compressed chunk size. */
808 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
809 struct pwm_chunk_hdr chunk_hdr = {
810 .compressed_size = cpu_to_le32(csize),
812 ret = full_write(ctx->out_fd, &chunk_hdr,
819 /* Write the chunk data. */
820 ret = full_write(ctx->out_fd, cchunk, csize);
824 ctx->cur_write_stream_offset += usize;
826 completed_size = usize;
827 completed_stream_count = 0;
828 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
829 /* Wrote chunk in packed mode. It may have finished multiple
831 while (ctx->cur_write_stream_offset > lte->size) {
832 struct wim_lookup_table_entry *next;
834 ctx->cur_write_stream_offset -= lte->size;
836 wimlib_assert(!list_is_singular(&ctx->pending_streams) &&
837 !list_empty(&ctx->pending_streams));
838 next = list_entry(lte->write_streams_list.next,
839 struct wim_lookup_table_entry,
841 list_move_tail(<e->write_streams_list,
844 completed_stream_count++;
846 if (ctx->cur_write_stream_offset == lte->size) {
847 ctx->cur_write_stream_offset = 0;
848 list_move_tail(<e->write_streams_list,
850 completed_stream_count++;
853 /* Wrote chunk in non-packed mode. It may have finished a
855 if (ctx->cur_write_stream_offset == lte->size) {
857 completed_stream_count++;
859 list_del(<e->write_streams_list);
861 wimlib_assert(ctx->cur_write_stream_offset ==
862 ctx->cur_write_res_size);
864 ret = end_write_resource(ctx, <e->out_reshdr);
868 lte->out_reshdr.flags = filter_resource_flags(lte->flags);
869 if (ctx->compressor != NULL)
870 lte->out_reshdr.flags |= WIM_RESHDR_FLAG_COMPRESSED;
872 if (ctx->compressor != NULL &&
873 lte->out_reshdr.size_in_wim >= lte->out_reshdr.uncompressed_size &&
874 !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) &&
875 !(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS))
877 /* Stream did not compress to less than its original
878 * size. If we're not writing a pipable WIM (which
879 * could mean the output file descriptor is
880 * non-seekable), and the stream isn't located in a
881 * resource pack (which would make reading it again
882 * costly), truncate the file to the start of the stream
883 * and write it uncompressed instead. */
884 DEBUG("Stream of size %"PRIu64" did not compress to "
885 "less than original size; writing uncompressed.",
887 ret = write_stream_uncompressed(lte, ctx->out_fd);
891 wimlib_assert(lte->out_reshdr.uncompressed_size == lte->size);
893 ctx->cur_write_stream_offset = 0;
897 do_write_streams_progress(&ctx->progress_data, lte,
898 completed_size, completed_stream_count,
904 ERROR_WITH_ERRNO("Write error");
909 submit_chunk_for_compression(struct write_streams_ctx *ctx,
910 const void *chunk, size_t size)
912 /* While we are unable to submit the chunk for compression (due to too
913 * many chunks already outstanding), retrieve and write the next
914 * compressed chunk. */
915 while (!ctx->compressor->submit_chunk(ctx->compressor, chunk, size)) {
922 bret = ctx->compressor->get_chunk(ctx->compressor,
923 &cchunk, &csize, &usize);
927 ret = write_chunk(ctx, cchunk, csize, usize);
934 /* Process the next chunk of data to be written to a WIM resource. */
936 write_stream_process_chunk(const void *chunk, size_t size, void *_ctx)
938 struct write_streams_ctx *ctx = _ctx;
940 const u8 *chunkptr, *chunkend;
942 wimlib_assert(size != 0);
944 if (ctx->compressor == NULL) {
945 /* Write chunk uncompressed. */
946 ret = write_chunk(ctx, chunk, size, size);
949 ctx->cur_read_stream_offset += size;
953 /* Submit the chunk for compression, but take into account that the
954 * @size the chunk was provided in may not correspond to the
955 * @out_chunk_size being used for compression. */
957 chunkend = chunkptr + size;
959 const u8 *resized_chunk;
960 size_t needed_chunk_size;
962 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
963 needed_chunk_size = ctx->out_chunk_size;
965 u64 res_bytes_remaining;
967 res_bytes_remaining = ctx->cur_read_stream_size -
968 ctx->cur_read_stream_offset;
969 needed_chunk_size = min(ctx->out_chunk_size,
970 ctx->chunk_buf_filled +
971 res_bytes_remaining);
974 if (ctx->chunk_buf_filled == 0 &&
975 chunkend - chunkptr >= needed_chunk_size)
977 /* No intermediate buffering needed. */
978 resized_chunk = chunkptr;
979 chunkptr += needed_chunk_size;
980 ctx->cur_read_stream_offset += needed_chunk_size;
982 /* Intermediate buffering needed. */
983 size_t bytes_consumed;
985 bytes_consumed = min(chunkend - chunkptr,
986 needed_chunk_size - ctx->chunk_buf_filled);
988 memcpy(&ctx->chunk_buf[ctx->chunk_buf_filled],
989 chunkptr, bytes_consumed);
991 chunkptr += bytes_consumed;
992 ctx->cur_read_stream_offset += bytes_consumed;
993 ctx->chunk_buf_filled += bytes_consumed;
994 if (ctx->chunk_buf_filled == needed_chunk_size) {
995 resized_chunk = ctx->chunk_buf;
996 ctx->chunk_buf_filled = 0;
1003 ret = submit_chunk_for_compression(ctx, resized_chunk,
1008 } while (chunkptr != chunkend);
1012 /* Finish processing a stream for writing. It may not have been completely
1013 * written yet, as the chunk_compressor implementation may still have chunks
1014 * buffered or being compressed. */
1016 write_stream_end_read(struct wim_lookup_table_entry *lte, int status, void *_ctx)
1018 struct write_streams_ctx *ctx = _ctx;
1020 wimlib_assert(ctx->cur_read_stream_offset == ctx->cur_read_stream_size);
1021 if (ctx->stream_was_duplicate) {
1022 free_lookup_table_entry(lte);
1023 } else if (lte->unhashed && ctx->lookup_table != NULL) {
1024 list_del(<e->unhashed_list);
1025 lookup_table_insert(ctx->lookup_table, lte);
1031 /* Compute statistics about a list of streams that will be written.
1033 * Assumes the streams are sorted such that all streams located in each distinct
1034 * WIM (specified by WIMStruct) are together. */
1036 compute_stream_list_stats(struct list_head *stream_list,
1037 struct write_streams_ctx *ctx)
1039 struct wim_lookup_table_entry *lte;
1040 u64 total_bytes = 0;
1041 u64 num_streams = 0;
1042 u64 total_parts = 0;
1043 WIMStruct *prev_wim_part = NULL;
1045 list_for_each_entry(lte, stream_list, write_streams_list) {
1047 total_bytes += lte->size;
1048 if (lte->resource_location == RESOURCE_IN_WIM) {
1049 if (prev_wim_part != lte->rspec->wim) {
1050 prev_wim_part = lte->rspec->wim;
1055 ctx->progress_data.progress.write_streams.total_bytes = total_bytes;
1056 ctx->progress_data.progress.write_streams.total_streams = num_streams;
1057 ctx->progress_data.progress.write_streams.completed_bytes = 0;
1058 ctx->progress_data.progress.write_streams.completed_streams = 0;
1059 ctx->progress_data.progress.write_streams.compression_type = ctx->out_ctype;
1060 ctx->progress_data.progress.write_streams.total_parts = total_parts;
1061 ctx->progress_data.progress.write_streams.completed_parts = 0;
1062 ctx->progress_data.next_progress = 0;
1063 ctx->progress_data.prev_wim_part = NULL;
1066 /* Find streams in @stream_list that can be copied to the output WIM in raw form
1067 * rather than compressed. Delete these streams from @stream_list, and move one
1068 * per resource to @raw_copy_resources. Return the total uncompressed size of
1069 * the streams that need to be compressed. */
1071 find_raw_copy_resources(struct list_head *stream_list,
1072 int write_resource_flags,
1075 struct list_head *raw_copy_resources)
1077 struct wim_lookup_table_entry *lte, *tmp;
1078 u64 num_bytes_to_compress = 0;
1080 INIT_LIST_HEAD(raw_copy_resources);
1082 /* Initialize temporary raw_copy_ok flag. */
1083 list_for_each_entry(lte, stream_list, write_streams_list)
1084 if (lte->resource_location == RESOURCE_IN_WIM)
1085 lte->rspec->raw_copy_ok = 0;
1087 list_for_each_entry_safe(lte, tmp, stream_list, write_streams_list) {
1088 if (lte->resource_location == RESOURCE_IN_WIM &&
1089 lte->rspec->raw_copy_ok)
1091 list_del(<e->write_streams_list);
1092 } else if (can_raw_copy(lte, write_resource_flags,
1093 out_ctype, out_chunk_size))
1095 lte->rspec->raw_copy_ok = 1;
1096 list_move_tail(<e->write_streams_list,
1097 raw_copy_resources);
1099 num_bytes_to_compress += lte->size;
1103 return num_bytes_to_compress;
1106 /* Copy a raw compressed resource located in another WIM file to the WIM file
1109 write_raw_copy_resource(struct wim_resource_spec *in_rspec,
1110 struct filedes *out_fd)
1112 u64 cur_read_offset;
1113 u64 end_read_offset;
1114 u8 buf[BUFFER_SIZE];
1115 size_t bytes_to_read;
1117 struct filedes *in_fd;
1118 struct wim_lookup_table_entry *lte;
1119 u64 out_offset_in_wim;
1121 DEBUG("Copying raw compressed data (size_in_wim=%"PRIu64", "
1122 "uncompressed_size=%"PRIu64")",
1123 in_rspec->size_in_wim, in_rspec->uncompressed_size);
1125 /* Copy the raw data. */
1126 cur_read_offset = in_rspec->offset_in_wim;
1127 end_read_offset = cur_read_offset + in_rspec->size_in_wim;
1129 out_offset_in_wim = out_fd->offset;
1131 if (in_rspec->is_pipable) {
1132 if (cur_read_offset < sizeof(struct pwm_stream_hdr))
1133 return WIMLIB_ERR_INVALID_PIPABLE_WIM;
1134 cur_read_offset -= sizeof(struct pwm_stream_hdr);
1135 out_offset_in_wim += sizeof(struct pwm_stream_hdr);
1137 in_fd = &in_rspec->wim->in_fd;
1138 wimlib_assert(cur_read_offset != end_read_offset);
1141 bytes_to_read = min(sizeof(buf), end_read_offset - cur_read_offset);
1143 ret = full_pread(in_fd, buf, bytes_to_read, cur_read_offset);
1147 ret = full_write(out_fd, buf, bytes_to_read);
1151 cur_read_offset += bytes_to_read;
1153 } while (cur_read_offset != end_read_offset);
1155 list_for_each_entry(lte, &in_rspec->stream_list, rspec_node) {
1156 if (lte->will_be_in_output_wim) {
1157 stream_set_out_reshdr_for_reuse(lte);
1158 if (in_rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS)
1159 lte->out_res_offset_in_wim = out_offset_in_wim;
1161 lte->out_reshdr.offset_in_wim = out_offset_in_wim;
1168 /* Copy a list of raw compressed resources located other WIM file(s) to the WIM
1169 * file being written. */
1171 write_raw_copy_resources(struct list_head *raw_copy_resources,
1172 struct filedes *out_fd,
1173 struct write_streams_progress_data *progress_data)
1175 struct wim_lookup_table_entry *lte;
1178 list_for_each_entry(lte, raw_copy_resources, write_streams_list) {
1179 ret = write_raw_copy_resource(lte->rspec, out_fd);
1182 do_write_streams_progress(progress_data, lte, lte->size,
1188 /* Wait for and write all chunks pending in the compressor. */
1190 finish_remaining_chunks(struct write_streams_ctx *ctx)
1197 if (ctx->compressor == NULL)
1200 if (ctx->chunk_buf_filled != 0) {
1201 ret = submit_chunk_for_compression(ctx, ctx->chunk_buf,
1202 ctx->chunk_buf_filled);
1207 while (ctx->compressor->get_chunk(ctx->compressor, &cdata, &csize, &usize)) {
1208 ret = write_chunk(ctx, cdata, csize, usize);
1216 remove_zero_length_streams(struct list_head *stream_list)
1218 struct wim_lookup_table_entry *lte, *tmp;
1220 list_for_each_entry_safe(lte, tmp, stream_list, write_streams_list) {
1221 wimlib_assert(lte->will_be_in_output_wim);
1222 if (lte->size == 0) {
1223 list_del(<e->write_streams_list);
1224 lte->out_reshdr.offset_in_wim = 0;
1225 lte->out_reshdr.size_in_wim = 0;
1226 lte->out_reshdr.uncompressed_size = 0;
1227 lte->out_reshdr.flags = filter_resource_flags(lte->flags);
1233 * Write a list of streams to the output WIM file.
1236 * The list of streams to write, specified by a list of `struct
1237 * wim_lookup_table_entry's linked by the 'write_streams_list' member.
1240 * The file descriptor, opened for writing, to which to write the streams.
1242 * @write_resource_flags
1243 * Flags to modify how the streams are written:
1245 * WRITE_RESOURCE_FLAG_RECOMPRESS:
1246 * Force compression of all resources, even if they could otherwise
1247 * be re-used by copying the raw data, due to being located in a WIM
1248 * file with compatible compression parameters.
1250 * WRITE_RESOURCE_FLAG_PIPABLE:
1251 * Write the resources in the wimlib-specific pipable format, and
1252 * furthermore do so in such a way that no seeking backwards in
1253 * @out_fd will be performed (so it may be a pipe).
1255 * WRITE_RESOURCE_FLAG_PACK_STREAMS:
1256 * Pack all the streams into a single resource rather than writing
1257 * them in separate resources. This flag is only valid if the WIM
1258 * version number has been, or will be, set to
1259 * WIM_VERSION_PACKED_STREAMS. This flag may not be combined with
1260 * WRITE_RESOURCE_FLAG_PIPABLE.
1263 * Compression format to use to write the output streams, specified as one
1264 * of the WIMLIB_COMPRESSION_TYPE_* constants, excepting
1265 * WIMLIB_COMPRESSION_TYPE_INVALID but including
1266 * WIMLIB_COMPRESSION_TYPE_NONE.
1269 * Chunk size to use to write the streams. It must be a valid chunk size
1270 * for the specified compression format @out_ctype, unless @out_ctype is
1271 * WIMLIB_COMPRESSION_TYPE_NONE, in which case this parameter is ignored.
1274 * Number of threads to use to compress data. If 0, a default number of
1275 * threads will be chosen. The number of threads still may be decreased
1276 * from the specified value if insufficient memory is detected.
1279 * If on-the-fly deduplication of unhashed streams is desired, this
1280 * parameter must be pointer to the lookup table for the WIMStruct on whose
1281 * behalf the streams are being written. Otherwise, this parameter can be
1285 * If on-the-fly deduplication of unhashed streams is desired, this
1286 * parameter can be a pointer to a context for stream filtering used to
1287 * detect whether the duplicate stream has been hard-filtered or not. If
1288 * no streams are hard-filtered or no streams are unhashed, this parameter
1292 * If non-NULL, a progress function that will be called periodically with
1293 * WIMLIB_PROGRESS_MSG_WRITE_STREAMS messages. Note that on-the-fly
1294 * deduplication of unhashed streams may result in the total bytes provided
1295 * in the progress data to decrease from one message to the next.
1297 * This function will write the streams in @stream_list to resources in
1298 * consecutive positions in the output WIM file, or to a single packed resource
1299 * if WRITE_RESOURCE_FLAG_PACK_STREAMS was specified in @write_resource_flags.
1300 * In both cases, the @out_reshdr of the `struct wim_lookup_table_entry' for
1301 * each stream written will be updated to specify its location, size, and flags
1302 * in the output WIM. In the packed resource case,
1303 * WIM_RESHDR_FLAG_PACKED_STREAMS will be set in the @flags field of each
1304 * @out_reshdr, and furthermore @out_res_offset_in_wim and @out_res_size_in_wim
1305 * of each @out_reshdr will be set to the offset and size, respectively, in the
1306 * output WIM of the packed resource containing the corresponding stream.
1308 * Each of the streams to write may be in any location supported by the
1309 * resource-handling code (specifically, read_stream_list()), such as the
1310 * contents of external file that has been logically added to the output WIM, or
1311 * a stream in another WIM file that has been imported, or even a stream in the
1312 * "same" WIM file of which a modified copy is being written. In the case that
1313 * a stream is already in a WIM file and uses compatible compression parameters,
1314 * by default this function will re-use the raw data instead of decompressing
1315 * it, then recompressing it; however, with WRITE_RESOURCE_FLAG_RECOMPRESS
1316 * specified in @write_resource_flags, this is not done.
1318 * As a further requirement, this function requires that the
1319 * @will_be_in_output_wim member be set to 1 on all streams in @stream_list as
1320 * well as any other streams not in @stream_list that will be in the output WIM
1321 * file, but set to 0 on any other streams in the output WIM's lookup table or
1322 * sharing a packed resource with a stream in @stream_list. Still furthermore,
1323 * if on-the-fly deduplication of streams is possible, then all streams in
1324 * @stream_list must also be linked by @lookup_table_list along with any other
1325 * streams that have @will_be_in_output_wim set.
1327 * This function handles on-the-fly deduplication of streams for which SHA1
1328 * message digests have not yet been calculated. Such streams may or may not
1329 * need to be written. If @lookup_table is non-NULL, then each stream in
1330 * @stream_list that has @unhashed set but not @unique_size set is checksummed
1331 * immediately before it would otherwise be read for writing in order to
1332 * determine if it is identical to another stream already being written or one
1333 * that would be filtered out of the output WIM using stream_filtered() with the
1334 * context @filter_ctx. Each such duplicate stream will be removed from
1335 * @stream_list, its reference count transfered to the pre-existing duplicate
1336 * stream, its memory freed, and will not be written. Alternatively, if a
1337 * stream in @stream_list is a duplicate with any stream in @lookup_table that
1338 * has not been marked for writing or would not be hard-filtered, it is freed
1339 * and the pre-existing duplicate is written instead, taking ownership of the
1340 * reference count and slot in the @lookup_table_list.
1342 * Returns 0 if every stream was either written successfully or did not need to
1343 * be written; otherwise returns a non-zero error code.
1346 write_stream_list(struct list_head *stream_list,
1347 struct filedes *out_fd,
1348 int write_resource_flags,
1351 unsigned num_threads,
1352 struct wim_lookup_table *lookup_table,
1353 struct filter_context *filter_ctx,
1354 wimlib_progress_func_t progress_func)
1357 struct write_streams_ctx ctx;
1358 struct list_head raw_copy_resources;
1360 wimlib_assert((write_resource_flags &
1361 (WRITE_RESOURCE_FLAG_PACK_STREAMS |
1362 WRITE_RESOURCE_FLAG_PIPABLE)) !=
1363 (WRITE_RESOURCE_FLAG_PACK_STREAMS |
1364 WRITE_RESOURCE_FLAG_PIPABLE));
1366 remove_zero_length_streams(stream_list);
1368 if (list_empty(stream_list)) {
1369 DEBUG("No streams to write.");
1373 memset(&ctx, 0, sizeof(ctx));
1375 /* Pre-sorting the streams is required for compute_stream_list_stats().
1376 * Afterwards, read_stream_list() need not sort them again. */
1377 ret = sort_stream_list_by_sequential_order(stream_list,
1378 offsetof(struct wim_lookup_table_entry,
1379 write_streams_list));
1383 ctx.out_fd = out_fd;
1384 ctx.lookup_table = lookup_table;
1385 ctx.out_ctype = out_ctype;
1386 ctx.out_chunk_size = out_chunk_size;
1387 ctx.write_resource_flags = write_resource_flags;
1388 ctx.filter_ctx = filter_ctx;
1390 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) {
1391 wimlib_assert(out_chunk_size != 0);
1392 if (out_chunk_size <= STACK_MAX) {
1393 ctx.chunk_buf = alloca(out_chunk_size);
1395 ctx.chunk_buf = MALLOC(out_chunk_size);
1396 if (ctx.chunk_buf == NULL) {
1397 ret = WIMLIB_ERR_NOMEM;
1398 goto out_destroy_context;
1402 ctx.chunk_buf_filled = 0;
1404 compute_stream_list_stats(stream_list, &ctx);
1406 ctx.progress_data.progress_func = progress_func;
1408 ctx.num_bytes_to_compress = find_raw_copy_resources(stream_list,
1409 write_resource_flags,
1412 &raw_copy_resources);
1414 DEBUG("Writing stream list "
1415 "(offset = %"PRIu64", write_resource_flags=0x%08x, "
1416 "out_ctype=%d, out_chunk_size=%u, num_threads=%u, "
1417 "total_bytes=%"PRIu64", num_bytes_to_compress=%"PRIu64")",
1418 out_fd->offset, write_resource_flags,
1419 out_ctype, out_chunk_size, num_threads,
1420 ctx.progress_data.progress.write_streams.total_bytes,
1421 ctx.num_bytes_to_compress);
1423 if (ctx.num_bytes_to_compress == 0) {
1424 DEBUG("No compression needed; skipping to raw copy!");
1425 goto out_write_raw_copy_resources;
1428 /* Unless uncompressed output was required, allocate a chunk_compressor
1429 * to do compression. There are serial and parallel implementations of
1430 * the chunk_compressor interface. We default to parallel using the
1431 * specified number of threads, unless the upper bound on the number
1432 * bytes needing to be compressed is less 2000000 (heuristic value). */
1433 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) {
1435 #ifdef ENABLE_MULTITHREADED_COMPRESSION
1436 if (ctx.num_bytes_to_compress >= 2000000) {
1437 ret = new_parallel_chunk_compressor(out_ctype,
1442 DEBUG("Couldn't create parallel chunk compressor "
1443 "(status %d)", ret);
1448 if (ctx.compressor == NULL) {
1449 ret = new_serial_chunk_compressor(out_ctype, out_chunk_size,
1452 goto out_destroy_context;
1457 ctx.progress_data.progress.write_streams.num_threads = ctx.compressor->num_threads;
1459 ctx.progress_data.progress.write_streams.num_threads = 1;
1461 DEBUG("Actually using %u threads",
1462 ctx.progress_data.progress.write_streams.num_threads);
1464 INIT_LIST_HEAD(&ctx.pending_streams);
1465 INIT_LIST_HEAD(&ctx.pack_streams);
1467 if (ctx.progress_data.progress_func) {
1468 (*ctx.progress_data.progress_func)(WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
1469 &ctx.progress_data.progress);
1472 if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
1473 ret = begin_write_resource(&ctx, ctx.num_bytes_to_compress);
1475 goto out_destroy_context;
1478 /* Read the list of streams needing to be compressed, using the
1479 * specified callbacks to execute processing of the data. */
1481 struct read_stream_list_callbacks cbs = {
1482 .begin_stream = write_stream_begin_read,
1483 .begin_stream_ctx = &ctx,
1484 .consume_chunk = write_stream_process_chunk,
1485 .consume_chunk_ctx = &ctx,
1486 .end_stream = write_stream_end_read,
1487 .end_stream_ctx = &ctx,
1490 ret = read_stream_list(stream_list,
1491 offsetof(struct wim_lookup_table_entry, write_streams_list),
1493 STREAM_LIST_ALREADY_SORTED |
1494 VERIFY_STREAM_HASHES |
1495 COMPUTE_MISSING_STREAM_HASHES);
1498 goto out_destroy_context;
1500 ret = finish_remaining_chunks(&ctx);
1502 goto out_destroy_context;
1504 if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
1505 struct wim_reshdr reshdr;
1506 struct wim_lookup_table_entry *lte;
1509 ret = end_write_resource(&ctx, &reshdr);
1511 goto out_destroy_context;
1513 DEBUG("Ending packed resource: %lu %lu %lu.",
1514 reshdr.offset_in_wim,
1516 reshdr.uncompressed_size);
1519 list_for_each_entry(lte, &ctx.pack_streams, write_streams_list) {
1520 lte->out_reshdr.size_in_wim = lte->size;
1521 lte->out_reshdr.flags = filter_resource_flags(lte->flags);
1522 lte->out_reshdr.flags |= WIM_RESHDR_FLAG_PACKED_STREAMS;
1523 lte->out_reshdr.uncompressed_size = 0;
1524 lte->out_reshdr.offset_in_wim = offset_in_res;
1525 lte->out_res_offset_in_wim = reshdr.offset_in_wim;
1526 lte->out_res_size_in_wim = reshdr.size_in_wim;
1527 /*lte->out_res_uncompressed_size = reshdr.uncompressed_size;*/
1528 offset_in_res += lte->size;
1530 wimlib_assert(offset_in_res == reshdr.uncompressed_size);
1533 out_write_raw_copy_resources:
1534 /* Copy any compressed resources for which the raw data can be reused
1535 * without decompression. */
1536 ret = write_raw_copy_resources(&raw_copy_resources, ctx.out_fd,
1537 &ctx.progress_data);
1539 out_destroy_context:
1540 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && out_chunk_size > STACK_MAX)
1541 FREE(ctx.chunk_buf);
1542 FREE(ctx.chunk_csizes);
1544 ctx.compressor->destroy(ctx.compressor);
1545 DEBUG("Done (ret=%d)", ret);
1550 wim_write_stream_list(WIMStruct *wim,
1551 struct list_head *stream_list,
1553 unsigned num_threads,
1554 struct filter_context *filter_ctx,
1555 wimlib_progress_func_t progress_func)
1559 int write_resource_flags;
1561 write_resource_flags = write_flags_to_resource_flags(write_flags);
1563 if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
1564 out_chunk_size = wim->out_pack_chunk_size;
1565 out_ctype = wim->out_pack_compression_type;
1567 out_chunk_size = wim->out_chunk_size;
1568 out_ctype = wim->out_compression_type;
1571 return write_stream_list(stream_list,
1573 write_resource_flags,
1583 write_wim_resource(struct wim_lookup_table_entry *lte,
1584 struct filedes *out_fd,
1587 int write_resource_flags)
1589 LIST_HEAD(stream_list);
1590 list_add(<e->write_streams_list, &stream_list);
1591 lte->will_be_in_output_wim = 1;
1592 return write_stream_list(&stream_list,
1594 write_resource_flags & ~WRITE_RESOURCE_FLAG_PACK_STREAMS,
1604 write_wim_resource_from_buffer(const void *buf, size_t buf_size,
1605 int reshdr_flags, struct filedes *out_fd,
1608 struct wim_reshdr *out_reshdr,
1610 int write_resource_flags)
1613 struct wim_lookup_table_entry *lte;
1615 /* Set up a temporary lookup table entry to provide to
1616 * write_wim_resource(). */
1618 lte = new_lookup_table_entry();
1620 return WIMLIB_ERR_NOMEM;
1622 lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER;
1623 lte->attached_buffer = (void*)buf;
1624 lte->size = buf_size;
1625 lte->flags = reshdr_flags;
1627 if (write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
1628 sha1_buffer(buf, buf_size, lte->hash);
1634 ret = write_wim_resource(lte, out_fd, out_ctype, out_chunk_size,
1635 write_resource_flags);
1639 copy_reshdr(out_reshdr, <e->out_reshdr);
1642 copy_hash(hash, lte->hash);
1645 lte->resource_location = RESOURCE_NONEXISTENT;
1646 free_lookup_table_entry(lte);
1650 struct stream_size_table {
1651 struct hlist_head *array;
1657 init_stream_size_table(struct stream_size_table *tab, size_t capacity)
1659 tab->array = CALLOC(capacity, sizeof(tab->array[0]));
1660 if (tab->array == NULL)
1661 return WIMLIB_ERR_NOMEM;
1662 tab->num_entries = 0;
1663 tab->capacity = capacity;
1668 destroy_stream_size_table(struct stream_size_table *tab)
1674 stream_size_table_insert(struct wim_lookup_table_entry *lte, void *_tab)
1676 struct stream_size_table *tab = _tab;
1678 struct wim_lookup_table_entry *same_size_lte;
1679 struct hlist_node *tmp;
1681 pos = hash_u64(lte->size) % tab->capacity;
1682 lte->unique_size = 1;
1683 hlist_for_each_entry(same_size_lte, tmp, &tab->array[pos], hash_list_2) {
1684 if (same_size_lte->size == lte->size) {
1685 lte->unique_size = 0;
1686 same_size_lte->unique_size = 0;
1691 hlist_add_head(<e->hash_list_2, &tab->array[pos]);
1696 struct find_streams_ctx {
1699 struct list_head stream_list;
1700 struct stream_size_table stream_size_tab;
1704 reference_stream_for_write(struct wim_lookup_table_entry *lte,
1705 struct list_head *stream_list, u32 nref)
1707 if (!lte->will_be_in_output_wim) {
1708 lte->out_refcnt = 0;
1709 list_add_tail(<e->write_streams_list, stream_list);
1710 lte->will_be_in_output_wim = 1;
1712 lte->out_refcnt += nref;
1716 fully_reference_stream_for_write(struct wim_lookup_table_entry *lte,
1719 struct list_head *stream_list = _stream_list;
1720 lte->will_be_in_output_wim = 0;
1721 reference_stream_for_write(lte, stream_list, lte->refcnt);
1726 inode_find_streams_to_reference(const struct wim_inode *inode,
1727 const struct wim_lookup_table *table,
1728 struct list_head *stream_list)
1730 struct wim_lookup_table_entry *lte;
1733 wimlib_assert(inode->i_nlink > 0);
1735 for (i = 0; i <= inode->i_num_ads; i++) {
1736 lte = inode_stream_lte(inode, i, table);
1738 reference_stream_for_write(lte, stream_list,
1740 else if (!is_zero_hash(inode_stream_hash(inode, i)))
1741 return WIMLIB_ERR_RESOURCE_NOT_FOUND;
1747 do_stream_set_not_in_output_wim(struct wim_lookup_table_entry *lte, void *_ignore)
1749 lte->will_be_in_output_wim = 0;
1754 image_find_streams_to_reference(WIMStruct *wim)
1756 struct wim_image_metadata *imd;
1757 struct wim_inode *inode;
1758 struct wim_lookup_table_entry *lte;
1759 struct list_head *stream_list;
1762 imd = wim_get_current_image_metadata(wim);
1764 image_for_each_unhashed_stream(lte, imd)
1765 lte->will_be_in_output_wim = 0;
1767 stream_list = wim->private;
1768 image_for_each_inode(inode, imd) {
1769 ret = inode_find_streams_to_reference(inode,
1779 prepare_unfiltered_list_of_streams_in_output_wim(WIMStruct *wim,
1782 struct list_head *stream_list_ret)
1786 INIT_LIST_HEAD(stream_list_ret);
1788 if (streams_ok && (image == WIMLIB_ALL_IMAGES ||
1789 (image == 1 && wim->hdr.image_count == 1)))
1791 /* Fast case: Assume that all streams are being written and
1792 * that the reference counts are correct. */
1793 struct wim_lookup_table_entry *lte;
1794 struct wim_image_metadata *imd;
1797 for_lookup_table_entry(wim->lookup_table,
1798 fully_reference_stream_for_write,
1801 for (i = 0; i < wim->hdr.image_count; i++) {
1802 imd = wim->image_metadata[i];
1803 image_for_each_unhashed_stream(lte, imd)
1804 fully_reference_stream_for_write(lte, stream_list_ret);
1807 /* Slow case: Walk through the images being written and
1808 * determine the streams referenced. */
1809 for_lookup_table_entry(wim->lookup_table,
1810 do_stream_set_not_in_output_wim, NULL);
1811 wim->private = stream_list_ret;
1812 ret = for_image(wim, image, image_find_streams_to_reference);
1820 struct insert_other_if_hard_filtered_ctx {
1821 struct stream_size_table *tab;
1822 struct filter_context *filter_ctx;
1826 insert_other_if_hard_filtered(struct wim_lookup_table_entry *lte, void *_ctx)
1828 struct insert_other_if_hard_filtered_ctx *ctx = _ctx;
1830 if (!lte->will_be_in_output_wim &&
1831 stream_hard_filtered(lte, ctx->filter_ctx))
1832 stream_size_table_insert(lte, ctx->tab);
1837 determine_stream_size_uniquity(struct list_head *stream_list,
1838 struct wim_lookup_table *lt,
1839 struct filter_context *filter_ctx)
1842 struct stream_size_table tab;
1843 struct wim_lookup_table_entry *lte;
1845 ret = init_stream_size_table(&tab, lt->capacity);
1849 if (may_hard_filter_streams(filter_ctx)) {
1850 struct insert_other_if_hard_filtered_ctx ctx = {
1852 .filter_ctx = filter_ctx,
1854 for_lookup_table_entry(lt, insert_other_if_hard_filtered, &ctx);
1857 list_for_each_entry(lte, stream_list, write_streams_list)
1858 stream_size_table_insert(lte, &tab);
1860 destroy_stream_size_table(&tab);
1865 filter_stream_list_for_write(struct list_head *stream_list,
1866 struct filter_context *filter_ctx)
1868 struct wim_lookup_table_entry *lte, *tmp;
1870 list_for_each_entry_safe(lte, tmp,
1871 stream_list, write_streams_list)
1873 int status = stream_filtered(lte, filter_ctx);
1880 /* Soft filtered. */
1882 /* Hard filtered. */
1883 lte->will_be_in_output_wim = 0;
1884 list_del(<e->lookup_table_list);
1886 list_del(<e->write_streams_list);
1892 * prepare_stream_list_for_write() -
1894 * Prepare the list of streams to write for writing a WIM containing the
1895 * specified image(s) with the specified write flags.
1898 * The WIMStruct on whose behalf the write is occurring.
1901 * Image(s) from the WIM to write; may be WIMLIB_ALL_IMAGES.
1904 * WIMLIB_WRITE_FLAG_* flags for the write operation:
1906 * STREAMS_OK: For writes of all images, assume that all streams in the
1907 * lookup table of @wim and the per-image lists of unhashed streams should
1908 * be taken as-is, and image metadata should not be searched for
1909 * references. This does not exclude filtering with OVERWRITE and
1910 * SKIP_EXTERNAL_WIMS, below.
1912 * OVERWRITE: Streams already present in @wim shall not be returned in
1915 * SKIP_EXTERNAL_WIMS: Streams already present in a WIM file, but not
1916 * @wim, shall be be returned in neither @stream_list_ret nor
1917 * @lookup_table_list_ret.
1920 * List of streams, linked by write_streams_list, that need to be written
1921 * will be returned here.
1923 * Note that this function assumes that unhashed streams will be written;
1924 * it does not take into account that they may become duplicates when
1927 * @lookup_table_list_ret
1928 * List of streams, linked by lookup_table_list, that need to be included
1929 * in the WIM's lookup table will be returned here. This will be a
1930 * superset of the streams in @stream_list_ret.
1932 * This list will be a proper superset of @stream_list_ret if and only if
1933 * WIMLIB_WRITE_FLAG_OVERWRITE was specified in @write_flags and some of
1934 * the streams that would otherwise need to be written were already located
1937 * All streams in this list will have @out_refcnt set to the number of
1938 * references to the stream in the output WIM. If
1939 * WIMLIB_WRITE_FLAG_STREAMS_OK was specified in @write_flags, @out_refcnt
1940 * may be as low as 0.
1943 * A context for queries of stream filter status with stream_filtered() is
1944 * returned in this location.
1946 * In addition, @will_be_in_output_wim will be set to 1 in all stream entries
1947 * inserted into @lookup_table_list_ret and to 0 in all stream entries in the
1948 * lookup table of @wim not inserted into @lookup_table_list_ret.
1950 * Still furthermore, @unique_size will be set to 1 on all stream entries in
1951 * @stream_list_ret that have unique size among all stream entries in
1952 * @stream_list_ret and among all stream entries in the lookup table of @wim
1953 * that are ineligible for being written due to filtering.
1955 * Returns 0 on success; nonzero on read error, memory allocation error, or
1959 prepare_stream_list_for_write(WIMStruct *wim, int image,
1961 struct list_head *stream_list_ret,
1962 struct list_head *lookup_table_list_ret,
1963 struct filter_context *filter_ctx_ret)
1966 struct wim_lookup_table_entry *lte;
1968 filter_ctx_ret->write_flags = write_flags;
1969 filter_ctx_ret->wim = wim;
1971 ret = prepare_unfiltered_list_of_streams_in_output_wim(
1974 write_flags & WIMLIB_WRITE_FLAG_STREAMS_OK,
1979 INIT_LIST_HEAD(lookup_table_list_ret);
1980 list_for_each_entry(lte, stream_list_ret, write_streams_list)
1981 list_add_tail(<e->lookup_table_list, lookup_table_list_ret);
1983 ret = determine_stream_size_uniquity(stream_list_ret, wim->lookup_table,
1988 if (may_filter_streams(filter_ctx_ret))
1989 filter_stream_list_for_write(stream_list_ret, filter_ctx_ret);
1995 write_wim_streams(WIMStruct *wim, int image, int write_flags,
1996 unsigned num_threads,
1997 wimlib_progress_func_t progress_func,
1998 struct list_head *stream_list_override,
1999 struct list_head *lookup_table_list_ret)
2002 struct list_head _stream_list;
2003 struct list_head *stream_list;
2004 struct wim_lookup_table_entry *lte;
2005 struct filter_context _filter_ctx;
2006 struct filter_context *filter_ctx;
2008 if (stream_list_override == NULL) {
2009 /* Normal case: prepare stream list from image(s) being written.
2011 stream_list = &_stream_list;
2012 filter_ctx = &_filter_ctx;
2013 ret = prepare_stream_list_for_write(wim, image, write_flags,
2015 lookup_table_list_ret,
2020 /* Currently only as a result of wimlib_split() being called:
2021 * use stream list already explicitly provided. Use existing
2022 * reference counts. */
2023 stream_list = stream_list_override;
2025 INIT_LIST_HEAD(lookup_table_list_ret);
2026 list_for_each_entry(lte, stream_list, write_streams_list) {
2027 lte->out_refcnt = lte->refcnt;
2028 lte->will_be_in_output_wim = 1;
2029 lte->unique_size = 0;
2030 list_add_tail(<e->lookup_table_list, lookup_table_list_ret);
2034 return wim_write_stream_list(wim,
2043 write_wim_metadata_resources(WIMStruct *wim, int image, int write_flags,
2044 wimlib_progress_func_t progress_func)
2049 int write_resource_flags;
2051 if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA) {
2052 DEBUG("Not writing any metadata resources.");
2056 write_resource_flags = write_flags_to_resource_flags(write_flags);
2058 write_resource_flags &= ~WRITE_RESOURCE_FLAG_PACK_STREAMS;
2060 DEBUG("Writing metadata resources (offset=%"PRIu64")",
2061 wim->out_fd.offset);
2064 progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN, NULL);
2066 if (image == WIMLIB_ALL_IMAGES) {
2068 end_image = wim->hdr.image_count;
2070 start_image = image;
2074 for (int i = start_image; i <= end_image; i++) {
2075 struct wim_image_metadata *imd;
2077 imd = wim->image_metadata[i - 1];
2078 /* Build a new metadata resource only if image was modified from
2079 * the original (or was newly added). Otherwise just copy the
2081 if (imd->modified) {
2082 DEBUG("Image %u was modified; building and writing new "
2083 "metadata resource", i);
2084 ret = write_metadata_resource(wim, i,
2085 write_resource_flags);
2086 } else if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) {
2087 DEBUG("Image %u was not modified; re-using existing "
2088 "metadata resource.", i);
2089 stream_set_out_reshdr_for_reuse(imd->metadata_lte);
2092 DEBUG("Image %u was not modified; copying existing "
2093 "metadata resource.", i);
2094 ret = write_wim_resource(imd->metadata_lte,
2096 wim->out_compression_type,
2097 wim->out_chunk_size,
2098 write_resource_flags);
2104 progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_END, NULL);
2109 open_wim_writable(WIMStruct *wim, const tchar *path, int open_flags)
2112 DEBUG("Opening \"%"TS"\" for writing.", path);
2114 raw_fd = topen(path, open_flags | O_BINARY, 0644);
2116 ERROR_WITH_ERRNO("Failed to open \"%"TS"\" for writing", path);
2117 return WIMLIB_ERR_OPEN;
2119 filedes_init(&wim->out_fd, raw_fd);
2124 close_wim_writable(WIMStruct *wim, int write_flags)
2128 if (!(write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)) {
2129 DEBUG("Closing WIM file.");
2130 if (filedes_valid(&wim->out_fd))
2131 if (filedes_close(&wim->out_fd))
2132 ret = WIMLIB_ERR_WRITE;
2134 filedes_invalidate(&wim->out_fd);
2139 cmp_streams_by_out_rspec(const void *p1, const void *p2)
2141 const struct wim_lookup_table_entry *lte1, *lte2;
2143 lte1 = *(const struct wim_lookup_table_entry**)p1;
2144 lte2 = *(const struct wim_lookup_table_entry**)p2;
2146 if (lte1->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
2147 if (lte2->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
2148 if (lte1->out_res_offset_in_wim != lte2->out_res_offset_in_wim)
2149 return cmp_u64(lte1->out_res_offset_in_wim,
2150 lte2->out_res_offset_in_wim);
2155 if (lte2->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS)
2158 return cmp_u64(lte1->out_reshdr.offset_in_wim,
2159 lte2->out_reshdr.offset_in_wim);
2163 write_wim_lookup_table(WIMStruct *wim, int image, int write_flags,
2164 struct wim_reshdr *out_reshdr,
2165 struct list_head *lookup_table_list)
2169 /* Set output resource metadata for streams already present in WIM. */
2170 if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) {
2171 struct wim_lookup_table_entry *lte;
2172 list_for_each_entry(lte, lookup_table_list, lookup_table_list)
2174 if (lte->resource_location == RESOURCE_IN_WIM &&
2175 lte->rspec->wim == wim)
2177 stream_set_out_reshdr_for_reuse(lte);
2182 ret = sort_stream_list(lookup_table_list,
2183 offsetof(struct wim_lookup_table_entry, lookup_table_list),
2184 cmp_streams_by_out_rspec);
2188 /* Add entries for metadata resources. */
2189 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)) {
2193 if (image == WIMLIB_ALL_IMAGES) {
2195 end_image = wim->hdr.image_count;
2197 start_image = image;
2201 /* Push metadata resource lookup table entries onto the front of
2202 * the list in reverse order, so that they're written in order.
2204 for (int i = end_image; i >= start_image; i--) {
2205 struct wim_lookup_table_entry *metadata_lte;
2207 metadata_lte = wim->image_metadata[i - 1]->metadata_lte;
2208 wimlib_assert(metadata_lte->out_reshdr.flags & WIM_RESHDR_FLAG_METADATA);
2209 metadata_lte->out_refcnt = 1;
2210 list_add(&metadata_lte->lookup_table_list, lookup_table_list);
2214 return write_wim_lookup_table_from_stream_list(lookup_table_list,
2216 wim->hdr.part_number,
2218 write_flags_to_resource_flags(write_flags));
2224 * Finish writing a WIM file: write the lookup table, xml data, and integrity
2225 * table, then overwrite the WIM header. By default, closes the WIM file
2226 * descriptor (@wim->out_fd) if successful.
2228 * write_flags is a bitwise OR of the following:
2230 * (public) WIMLIB_WRITE_FLAG_CHECK_INTEGRITY:
2231 * Include an integrity table.
2233 * (public) WIMLIB_WRITE_FLAG_FSYNC:
2234 * fsync() the output file before closing it.
2236 * (public) WIMLIB_WRITE_FLAG_PIPABLE:
2237 * Writing a pipable WIM, possibly to a pipe; include pipable WIM
2238 * stream headers before the lookup table and XML data, and also
2239 * write the WIM header at the end instead of seeking to the
2240 * beginning. Can't be combined with
2241 * WIMLIB_WRITE_FLAG_CHECK_INTEGRITY.
2243 * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE:
2244 * Don't write the lookup table.
2246 * (private) WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE:
2247 * When (if) writing the integrity table, re-use entries from the
2248 * existing integrity table, if possible.
2250 * (private) WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML:
2251 * After writing the XML data but before writing the integrity
2252 * table, write a temporary WIM header and flush the stream so that
2253 * the WIM is less likely to become corrupted upon abrupt program
2255 * (private) WIMLIB_WRITE_FLAG_HEADER_AT_END:
2256 * Instead of overwriting the WIM header at the beginning of the
2257 * file, simply append it to the end of the file. (Used when
2259 * (private) WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR:
2260 * Do not close the file descriptor @wim->out_fd on either success
2262 * (private) WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES:
2263 * Use the existing <TOTALBYTES> stored in the in-memory XML
2264 * information, rather than setting it to the offset of the XML
2265 * data being written.
2268 finish_write(WIMStruct *wim, int image, int write_flags,
2269 wimlib_progress_func_t progress_func,
2270 struct list_head *lookup_table_list)
2274 int write_resource_flags;
2275 off_t old_lookup_table_end;
2276 off_t new_lookup_table_end;
2279 DEBUG("image=%d, write_flags=%08x", image, write_flags);
2281 write_resource_flags = write_flags_to_resource_flags(write_flags);
2283 /* In the WIM header, there is room for the resource entry for a
2284 * metadata resource labeled as the "boot metadata". This entry should
2285 * be zeroed out if there is no bootable image (boot_idx 0). Otherwise,
2286 * it should be a copy of the resource entry for the image that is
2287 * marked as bootable. This is not well documented... */
2288 if (wim->hdr.boot_idx == 0) {
2289 zero_reshdr(&wim->hdr.boot_metadata_reshdr);
2291 copy_reshdr(&wim->hdr.boot_metadata_reshdr,
2292 &wim->image_metadata[
2293 wim->hdr.boot_idx - 1]->metadata_lte->out_reshdr);
2296 /* Write lookup table. (Save old position first.) */
2297 old_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim +
2298 wim->hdr.lookup_table_reshdr.size_in_wim;
2299 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
2300 ret = write_wim_lookup_table(wim, image, write_flags,
2301 &wim->hdr.lookup_table_reshdr,
2307 /* Write XML data. */
2308 xml_totalbytes = wim->out_fd.offset;
2309 if (write_flags & WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES)
2310 xml_totalbytes = WIM_TOTALBYTES_USE_EXISTING;
2311 ret = write_wim_xml_data(wim, image, xml_totalbytes,
2312 &wim->hdr.xml_data_reshdr,
2313 write_resource_flags);
2317 /* Write integrity table (optional). */
2318 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
2319 if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) {
2320 struct wim_header checkpoint_hdr;
2321 memcpy(&checkpoint_hdr, &wim->hdr, sizeof(struct wim_header));
2322 zero_reshdr(&checkpoint_hdr.integrity_table_reshdr);
2323 checkpoint_hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2324 ret = write_wim_header_at_offset(&checkpoint_hdr,
2330 if (!(write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE))
2331 old_lookup_table_end = 0;
2333 new_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim +
2334 wim->hdr.lookup_table_reshdr.size_in_wim;
2336 ret = write_integrity_table(wim,
2337 new_lookup_table_end,
2338 old_lookup_table_end,
2343 /* No integrity table. */
2344 zero_reshdr(&wim->hdr.integrity_table_reshdr);
2347 /* Now that all information in the WIM header has been determined, the
2348 * preliminary header written earlier can be overwritten, the header of
2349 * the existing WIM file can be overwritten, or the final header can be
2350 * written to the end of the pipable WIM. */
2351 wim->hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2353 if (write_flags & WIMLIB_WRITE_FLAG_HEADER_AT_END)
2354 hdr_offset = wim->out_fd.offset;
2355 DEBUG("Writing new header @ %"PRIu64".", hdr_offset);
2356 ret = write_wim_header_at_offset(&wim->hdr, &wim->out_fd, hdr_offset);
2360 /* Possibly sync file data to disk before closing. On POSIX systems, it
2361 * is necessary to do this before using rename() to overwrite an
2362 * existing file with a new file. Otherwise, data loss would occur if
2363 * the system is abruptly terminated when the metadata for the rename
2364 * operation has been written to disk, but the new file data has not.
2366 if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) {
2367 DEBUG("Syncing WIM file.");
2368 if (fsync(wim->out_fd.fd)) {
2369 ERROR_WITH_ERRNO("Error syncing data to WIM file");
2370 return WIMLIB_ERR_WRITE;
2374 if (close_wim_writable(wim, write_flags)) {
2375 ERROR_WITH_ERRNO("Failed to close the output WIM file");
2376 return WIMLIB_ERR_WRITE;
2382 #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK)
2384 lock_wim(WIMStruct *wim, int fd)
2387 if (fd != -1 && !wim->wim_locked) {
2388 ret = flock(fd, LOCK_EX | LOCK_NB);
2390 if (errno == EWOULDBLOCK) {
2391 ERROR("`%"TS"' is already being modified or has been "
2392 "mounted read-write\n"
2393 " by another process!", wim->filename);
2394 ret = WIMLIB_ERR_ALREADY_LOCKED;
2396 WARNING_WITH_ERRNO("Failed to lock `%"TS"'",
2401 wim->wim_locked = 1;
2409 * write_pipable_wim():
2411 * Perform the intermediate stages of creating a "pipable" WIM (i.e. a WIM
2412 * capable of being applied from a pipe).
2414 * Pipable WIMs are a wimlib-specific modification of the WIM format such that
2415 * images can be applied from them sequentially when the file data is sent over
2416 * a pipe. In addition, a pipable WIM can be written sequentially to a pipe.
2417 * The modifications made to the WIM format for pipable WIMs are:
2419 * - Magic characters in header are "WLPWM\0\0\0" (wimlib pipable WIM) instead
2420 * of "MSWIM\0\0\0". This lets wimlib know that the WIM is pipable and also
2421 * stops other software from trying to read the file as a normal WIM.
2423 * - The header at the beginning of the file does not contain all the normal
2424 * information; in particular it will have all 0's for the lookup table and
2425 * XML data resource entries. This is because this information cannot be
2426 * determined until the lookup table and XML data have been written.
2427 * Consequently, wimlib will write the full header at the very end of the
2428 * file. The header at the end, however, is only used when reading the WIM
2429 * from a seekable file (not a pipe).
2431 * - An extra copy of the XML data is placed directly after the header. This
2432 * allows image names and sizes to be determined at an appropriate time when
2433 * reading the WIM from a pipe. This copy of the XML data is ignored if the
2434 * WIM is read from a seekable file (not a pipe).
2436 * - The format of resources, or streams, has been modified to allow them to be
2437 * used before the "lookup table" has been read. Each stream is prefixed with
2438 * a `struct pwm_stream_hdr' that is basically an abbreviated form of `struct
2439 * wim_lookup_table_entry_disk' that only contains the SHA1 message digest,
2440 * uncompressed stream size, and flags that indicate whether the stream is
2441 * compressed. The data of uncompressed streams then follows literally, while
2442 * the data of compressed streams follows in a modified format. Compressed
2443 * streams do not begin with a chunk table, since the chunk table cannot be
2444 * written until all chunks have been compressed. Instead, each compressed
2445 * chunk is prefixed by a `struct pwm_chunk_hdr' that gives its size.
2446 * Furthermore, the chunk table is written at the end of the resource instead
2447 * of the start. Note: chunk offsets are given in the chunk table as if the
2448 * `struct pwm_chunk_hdr's were not present; also, the chunk table is only
2449 * used if the WIM is being read from a seekable file (not a pipe).
2451 * - Metadata resources always come before other file resources (streams).
2452 * (This does not by itself constitute an incompatibility with normal WIMs,
2453 * since this is valid in normal WIMs.)
2455 * - At least up to the end of the file resources, all components must be packed
2456 * as tightly as possible; there cannot be any "holes" in the WIM. (This does
2457 * not by itself consititute an incompatibility with normal WIMs, since this
2458 * is valid in normal WIMs.)
2460 * Note: the lookup table, XML data, and header at the end are not used when
2461 * applying from a pipe. They exist to support functionality such as image
2462 * application and export when the WIM is *not* read from a pipe.
2464 * Layout of pipable WIM:
2466 * ---------+----------+--------------------+----------------+--------------+-----------+--------+
2467 * | Header | XML data | Metadata resources | File resources | Lookup table | XML data | Header |
2468 * ---------+----------+--------------------+----------------+--------------+-----------+--------+
2470 * Layout of normal WIM:
2472 * +--------+-----------------------------+-------------------------+
2473 * | Header | File and metadata resources | Lookup table | XML data |
2474 * +--------+-----------------------------+-------------------------+
2476 * An optional integrity table can follow the final XML data in both normal and
2477 * pipable WIMs. However, due to implementation details, wimlib currently can
2478 * only include an integrity table in a pipable WIM when writing it to a
2479 * seekable file (not a pipe).
2481 * Do note that since pipable WIMs are not supported by Microsoft's software,
2482 * wimlib does not create them unless explicitly requested (with
2483 * WIMLIB_WRITE_FLAG_PIPABLE) and as stated above they use different magic
2484 * characters to identify the file.
2487 write_pipable_wim(WIMStruct *wim, int image, int write_flags,
2488 unsigned num_threads, wimlib_progress_func_t progress_func,
2489 struct list_head *stream_list_override,
2490 struct list_head *lookup_table_list_ret)
2493 struct wim_reshdr xml_reshdr;
2495 WARNING("Creating a pipable WIM, which will "
2497 " with Microsoft's software (wimgapi/imagex/Dism).");
2499 /* At this point, the header at the beginning of the file has already
2502 /* For efficiency, when wimlib adds an image to the WIM with
2503 * wimlib_add_image(), the SHA1 message digests of files is not
2504 * calculated; instead, they are calculated while the files are being
2505 * written. However, this does not work when writing a pipable WIM,
2506 * since when writing a stream to a pipable WIM, its SHA1 message digest
2507 * needs to be known before the stream data is written. Therefore,
2508 * before getting much farther, we need to pre-calculate the SHA1
2509 * message digests of all streams that will be written. */
2510 ret = wim_checksum_unhashed_streams(wim);
2514 /* Write extra copy of the XML data. */
2515 ret = write_wim_xml_data(wim, image, WIM_TOTALBYTES_OMIT,
2517 WRITE_RESOURCE_FLAG_PIPABLE);
2521 /* Write metadata resources for the image(s) being included in the
2523 ret = write_wim_metadata_resources(wim, image, write_flags,
2528 /* Write streams needed for the image(s) being included in the output
2529 * WIM, or streams needed for the split WIM part. */
2530 return write_wim_streams(wim, image, write_flags, num_threads,
2531 progress_func, stream_list_override,
2532 lookup_table_list_ret);
2534 /* The lookup table, XML data, and header at end are handled by
2535 * finish_write(). */
2538 /* Write a standalone WIM or split WIM (SWM) part to a new file or to a file
2541 write_wim_part(WIMStruct *wim,
2542 const void *path_or_fd,
2545 unsigned num_threads,
2546 wimlib_progress_func_t progress_func,
2547 unsigned part_number,
2548 unsigned total_parts,
2549 struct list_head *stream_list_override,
2553 struct wim_header hdr_save;
2554 struct list_head lookup_table_list;
2556 if (total_parts == 1)
2557 DEBUG("Writing standalone WIM.");
2559 DEBUG("Writing split WIM part %u/%u", part_number, total_parts);
2560 if (image == WIMLIB_ALL_IMAGES)
2561 DEBUG("Including all images.");
2563 DEBUG("Including image %d only.", image);
2564 if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)
2565 DEBUG("File descriptor: %d", *(const int*)path_or_fd);
2567 DEBUG("Path: \"%"TS"\"", (const tchar*)path_or_fd);
2568 DEBUG("Write flags: 0x%08x", write_flags);
2570 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)
2571 DEBUG("\tCHECK_INTEGRITY");
2573 if (write_flags & WIMLIB_WRITE_FLAG_REBUILD)
2576 if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
2577 DEBUG("\tRECOMPRESS");
2579 if (write_flags & WIMLIB_WRITE_FLAG_FSYNC)
2582 if (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE)
2585 if (write_flags & WIMLIB_WRITE_FLAG_IGNORE_READONLY_FLAG)
2586 DEBUG("\tIGNORE_READONLY_FLAG");
2588 if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
2591 if (write_flags & WIMLIB_WRITE_FLAG_NOT_PIPABLE)
2592 DEBUG("\tNOT_PIPABLE");
2594 if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS)
2595 DEBUG("\tPACK_STREAMS");
2597 if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)
2598 DEBUG("\tFILE_DESCRIPTOR");
2600 if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)
2601 DEBUG("\tNO_METADATA");
2603 if (write_flags & WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES)
2604 DEBUG("\tUSE_EXISTING_TOTALBYTES");
2606 if (num_threads == 0)
2607 DEBUG("Number of threads: autodetect");
2609 DEBUG("Number of threads: %u", num_threads);
2610 DEBUG("Progress function: %s", (progress_func ? "yes" : "no"));
2611 DEBUG("Stream list: %s", (stream_list_override ? "specified" : "autodetect"));
2612 DEBUG("GUID: %s", ((guid || wim->guid_set_explicitly) ?
2613 "specified" : "generate new"));
2615 /* Internally, this is always called with a valid part number and total
2617 wimlib_assert(total_parts >= 1);
2618 wimlib_assert(part_number >= 1 && part_number <= total_parts);
2620 /* A valid image (or all images) must be specified. */
2621 if (image != WIMLIB_ALL_IMAGES &&
2622 (image < 1 || image > wim->hdr.image_count))
2623 return WIMLIB_ERR_INVALID_IMAGE;
2625 /* If we need to write metadata resources, make sure the ::WIMStruct has
2626 * the needed information attached (e.g. is not a resource-only WIM,
2627 * such as a non-first part of a split WIM). */
2628 if (!wim_has_metadata(wim) &&
2629 !(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA))
2630 return WIMLIB_ERR_METADATA_NOT_FOUND;
2632 /* Check for contradictory flags. */
2633 if ((write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2634 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))
2635 == (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2636 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))
2637 return WIMLIB_ERR_INVALID_PARAM;
2639 if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2640 WIMLIB_WRITE_FLAG_NOT_PIPABLE))
2641 == (WIMLIB_WRITE_FLAG_PIPABLE |
2642 WIMLIB_WRITE_FLAG_NOT_PIPABLE))
2643 return WIMLIB_ERR_INVALID_PARAM;
2645 /* Save previous header, then start initializing the new one. */
2646 memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header));
2648 /* Set default integrity, pipable, and packed stream flags. */
2649 if (!(write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2650 WIMLIB_WRITE_FLAG_NOT_PIPABLE)))
2651 if (wim_is_pipable(wim)) {
2652 DEBUG("WIM is pipable; default to PIPABLE.");
2653 write_flags |= WIMLIB_WRITE_FLAG_PIPABLE;
2656 if (!(write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2657 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)))
2658 if (wim_has_integrity_table(wim)) {
2659 DEBUG("Integrity table present; default to CHECK_INTEGRITY.");
2660 write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY;
2663 if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2664 WIMLIB_WRITE_FLAG_PACK_STREAMS))
2665 == (WIMLIB_WRITE_FLAG_PIPABLE |
2666 WIMLIB_WRITE_FLAG_PACK_STREAMS))
2668 ERROR("Cannot specify both PIPABLE and PACK_STREAMS!");
2669 return WIMLIB_ERR_INVALID_PARAM;
2672 /* Set appropriate magic number. */
2673 if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
2674 wim->hdr.magic = PWM_MAGIC;
2676 wim->hdr.magic = WIM_MAGIC;
2678 /* Set appropriate version number. */
2679 if ((write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS) ||
2680 wim->out_compression_type == WIMLIB_COMPRESSION_TYPE_LZMS)
2681 wim->hdr.wim_version = WIM_VERSION_PACKED_STREAMS;
2683 wim->hdr.wim_version = WIM_VERSION_DEFAULT;
2685 /* Clear header flags that will be set automatically. */
2686 wim->hdr.flags &= ~(WIM_HDR_FLAG_METADATA_ONLY |
2687 WIM_HDR_FLAG_RESOURCE_ONLY |
2688 WIM_HDR_FLAG_SPANNED |
2689 WIM_HDR_FLAG_WRITE_IN_PROGRESS);
2691 /* Set SPANNED header flag if writing part of a split WIM. */
2692 if (total_parts != 1)
2693 wim->hdr.flags |= WIM_HDR_FLAG_SPANNED;
2695 /* Set part number and total parts of split WIM. This will be 1 and 1
2696 * if the WIM is standalone. */
2697 wim->hdr.part_number = part_number;
2698 wim->hdr.total_parts = total_parts;
2700 /* Set compression type if different. */
2701 if (wim->compression_type != wim->out_compression_type) {
2702 ret = set_wim_hdr_cflags(wim->out_compression_type, &wim->hdr);
2703 wimlib_assert(ret == 0);
2706 /* Set chunk size if different. */
2707 wim->hdr.chunk_size = wim->out_chunk_size;
2709 /* Use GUID if specified; otherwise generate a new one. */
2711 memcpy(wim->hdr.guid, guid, WIMLIB_GUID_LEN);
2712 else if (!wim->guid_set_explicitly)
2713 randomize_byte_array(wim->hdr.guid, WIMLIB_GUID_LEN);
2715 /* Clear references to resources that have not been written yet. */
2716 zero_reshdr(&wim->hdr.lookup_table_reshdr);
2717 zero_reshdr(&wim->hdr.xml_data_reshdr);
2718 zero_reshdr(&wim->hdr.boot_metadata_reshdr);
2719 zero_reshdr(&wim->hdr.integrity_table_reshdr);
2721 /* Set image count and boot index correctly for single image writes. */
2722 if (image != WIMLIB_ALL_IMAGES) {
2723 wim->hdr.image_count = 1;
2724 if (wim->hdr.boot_idx == image)
2725 wim->hdr.boot_idx = 1;
2727 wim->hdr.boot_idx = 0;
2730 /* Split WIMs can't be bootable. */
2731 if (total_parts != 1)
2732 wim->hdr.boot_idx = 0;
2734 /* Initialize output file descriptor. */
2735 if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR) {
2736 /* File descriptor was explicitly provided. Return error if
2737 * file descriptor is not seekable, unless writing a pipable WIM
2739 wim->out_fd.fd = *(const int*)path_or_fd;
2740 wim->out_fd.offset = 0;
2741 if (!filedes_is_seekable(&wim->out_fd)) {
2742 ret = WIMLIB_ERR_INVALID_PARAM;
2743 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
2744 goto out_restore_hdr;
2745 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
2746 ERROR("Can't include integrity check when "
2747 "writing pipable WIM to pipe!");
2748 goto out_restore_hdr;
2753 /* Filename of WIM to write was provided; open file descriptor
2755 ret = open_wim_writable(wim, (const tchar*)path_or_fd,
2756 O_TRUNC | O_CREAT | O_RDWR);
2758 goto out_restore_hdr;
2761 /* Write initial header. This is merely a "dummy" header since it
2762 * doesn't have all the information yet, so it will be overwritten later
2763 * (unless writing a pipable WIM). */
2764 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
2765 wim->hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2766 ret = write_wim_header(&wim->hdr, &wim->out_fd);
2767 wim->hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2769 goto out_restore_hdr;
2771 /* Write metadata resources and streams. */
2772 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE)) {
2773 /* Default case: create a normal (non-pipable) WIM. */
2774 ret = write_wim_streams(wim, image, write_flags, num_threads,
2775 progress_func, stream_list_override,
2776 &lookup_table_list);
2778 goto out_restore_hdr;
2780 ret = write_wim_metadata_resources(wim, image, write_flags,
2783 goto out_restore_hdr;
2785 /* Non-default case: create pipable WIM. */
2786 ret = write_pipable_wim(wim, image, write_flags, num_threads,
2787 progress_func, stream_list_override,
2788 &lookup_table_list);
2790 goto out_restore_hdr;
2791 write_flags |= WIMLIB_WRITE_FLAG_HEADER_AT_END;
2795 /* Write lookup table, XML data, and (optional) integrity table. */
2796 ret = finish_write(wim, image, write_flags, progress_func,
2797 &lookup_table_list);
2799 memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header));
2800 (void)close_wim_writable(wim, write_flags);
2801 DEBUG("ret=%d", ret);
2805 /* Write a standalone WIM to a file or file descriptor. */
2807 write_standalone_wim(WIMStruct *wim, const void *path_or_fd,
2808 int image, int write_flags, unsigned num_threads,
2809 wimlib_progress_func_t progress_func)
2811 return write_wim_part(wim, path_or_fd, image, write_flags,
2812 num_threads, progress_func, 1, 1, NULL, NULL);
2815 /* API function documented in wimlib.h */
2817 wimlib_write(WIMStruct *wim, const tchar *path,
2818 int image, int write_flags, unsigned num_threads,
2819 wimlib_progress_func_t progress_func)
2822 return WIMLIB_ERR_INVALID_PARAM;
2824 write_flags &= WIMLIB_WRITE_MASK_PUBLIC;
2826 return write_standalone_wim(wim, path, image, write_flags,
2827 num_threads, progress_func);
2830 /* API function documented in wimlib.h */
2832 wimlib_write_to_fd(WIMStruct *wim, int fd,
2833 int image, int write_flags, unsigned num_threads,
2834 wimlib_progress_func_t progress_func)
2837 return WIMLIB_ERR_INVALID_PARAM;
2839 write_flags &= WIMLIB_WRITE_MASK_PUBLIC;
2840 write_flags |= WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR;
2842 return write_standalone_wim(wim, &fd, image, write_flags,
2843 num_threads, progress_func);
2847 any_images_modified(WIMStruct *wim)
2849 for (int i = 0; i < wim->hdr.image_count; i++)
2850 if (wim->image_metadata[i]->modified)
2856 check_resource_offset(struct wim_lookup_table_entry *lte, void *_wim)
2858 const WIMStruct *wim = _wim;
2859 off_t end_offset = *(const off_t*)wim->private;
2861 if (lte->resource_location == RESOURCE_IN_WIM && lte->rspec->wim == wim &&
2862 lte->rspec->offset_in_wim + lte->rspec->size_in_wim > end_offset)
2863 return WIMLIB_ERR_RESOURCE_ORDER;
2867 /* Make sure no file or metadata resources are located after the XML data (or
2868 * integrity table if present)--- otherwise we can't safely overwrite the WIM in
2869 * place and we return WIMLIB_ERR_RESOURCE_ORDER. */
2871 check_resource_offsets(WIMStruct *wim, off_t end_offset)
2876 wim->private = &end_offset;
2877 ret = for_lookup_table_entry(wim->lookup_table, check_resource_offset, wim);
2881 for (i = 0; i < wim->hdr.image_count; i++) {
2882 ret = check_resource_offset(wim->image_metadata[i]->metadata_lte, wim);
2890 * Overwrite a WIM, possibly appending streams to it.
2892 * A WIM looks like (or is supposed to look like) the following:
2894 * Header (212 bytes)
2895 * Streams and metadata resources (variable size)
2896 * Lookup table (variable size)
2897 * XML data (variable size)
2898 * Integrity table (optional) (variable size)
2900 * If we are not adding any streams or metadata resources, the lookup table is
2901 * unchanged--- so we only need to overwrite the XML data, integrity table, and
2902 * header. This operation is potentially unsafe if the program is abruptly
2903 * terminated while the XML data or integrity table are being overwritten, but
2904 * before the new header has been written. To partially alleviate this problem,
2905 * a special flag (WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) is passed to
2906 * finish_write() to cause a temporary WIM header to be written after the XML
2907 * data has been written. This may prevent the WIM from becoming corrupted if
2908 * the program is terminated while the integrity table is being calculated (but
2909 * no guarantees, due to write re-ordering...).
2911 * If we are adding new streams or images (metadata resources), the lookup table
2912 * needs to be changed, and those streams need to be written. In this case, we
2913 * try to perform a safe update of the WIM file by writing the streams *after*
2914 * the end of the previous WIM, then writing the new lookup table, XML data, and
2915 * (optionally) integrity table following the new streams. This will produce a
2916 * layout like the following:
2918 * Header (212 bytes)
2919 * (OLD) Streams and metadata resources (variable size)
2920 * (OLD) Lookup table (variable size)
2921 * (OLD) XML data (variable size)
2922 * (OLD) Integrity table (optional) (variable size)
2923 * (NEW) Streams and metadata resources (variable size)
2924 * (NEW) Lookup table (variable size)
2925 * (NEW) XML data (variable size)
2926 * (NEW) Integrity table (optional) (variable size)
2928 * At all points, the WIM is valid as nothing points to the new data yet. Then,
2929 * the header is overwritten to point to the new lookup table, XML data, and
2930 * integrity table, to produce the following layout:
2932 * Header (212 bytes)
2933 * Streams and metadata resources (variable size)
2934 * Nothing (variable size)
2935 * More Streams and metadata resources (variable size)
2936 * Lookup table (variable size)
2937 * XML data (variable size)
2938 * Integrity table (optional) (variable size)
2940 * This method allows an image to be appended to a large WIM very quickly, and
2941 * is is crash-safe except in the case of write re-ordering, but the
2942 * disadvantage is that a small hole is left in the WIM where the old lookup
2943 * table, xml data, and integrity table were. (These usually only take up a
2944 * small amount of space compared to the streams, however.)
2947 overwrite_wim_inplace(WIMStruct *wim, int write_flags,
2948 unsigned num_threads,
2949 wimlib_progress_func_t progress_func)
2953 u64 old_lookup_table_end, old_xml_begin, old_xml_end;
2954 struct wim_header hdr_save;
2955 struct list_head stream_list;
2956 struct list_head lookup_table_list;
2957 struct filter_context filter_ctx;
2959 DEBUG("Overwriting `%"TS"' in-place", wim->filename);
2961 /* Save original header so it can be restored in case of error */
2962 memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header));
2964 /* Set default integrity flag. */
2965 if (!(write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2966 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)))
2967 if (wim_has_integrity_table(wim))
2968 write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY;
2970 /* Set WIM version if adding packed streams. */
2971 if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS)
2972 wim->hdr.wim_version = WIM_VERSION_PACKED_STREAMS;
2974 /* Set additional flags for overwrite. */
2975 write_flags |= WIMLIB_WRITE_FLAG_OVERWRITE |
2976 WIMLIB_WRITE_FLAG_STREAMS_OK;
2978 /* Make sure that the integrity table (if present) is after the XML
2979 * data, and that there are no stream resources, metadata resources, or
2980 * lookup tables after the XML data. Otherwise, these data would be
2982 old_xml_begin = wim->hdr.xml_data_reshdr.offset_in_wim;
2983 old_xml_end = old_xml_begin + wim->hdr.xml_data_reshdr.size_in_wim;
2984 old_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim +
2985 wim->hdr.lookup_table_reshdr.size_in_wim;
2986 if (wim->hdr.integrity_table_reshdr.offset_in_wim != 0 &&
2987 wim->hdr.integrity_table_reshdr.offset_in_wim < old_xml_end) {
2988 WARNING("Didn't expect the integrity table to be before the XML data");
2989 ret = WIMLIB_ERR_RESOURCE_ORDER;
2990 goto out_restore_memory_hdr;
2993 if (old_lookup_table_end > old_xml_begin) {
2994 WARNING("Didn't expect the lookup table to be after the XML data");
2995 ret = WIMLIB_ERR_RESOURCE_ORDER;
2996 goto out_restore_memory_hdr;
2999 /* Set @old_wim_end, which indicates the point beyond which we don't
3000 * allow any file and metadata resources to appear without returning
3001 * WIMLIB_ERR_RESOURCE_ORDER (due to the fact that we would otherwise
3002 * overwrite these resources). */
3003 if (!wim->deletion_occurred && !any_images_modified(wim)) {
3004 /* If no images have been modified and no images have been
3005 * deleted, a new lookup table does not need to be written. We
3006 * shall write the new XML data and optional integrity table
3007 * immediately after the lookup table. Note that this may
3008 * overwrite an existing integrity table. */
3009 DEBUG("Skipping writing lookup table "
3010 "(no images modified or deleted)");
3011 old_wim_end = old_lookup_table_end;
3012 write_flags |= WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE |
3013 WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML;
3014 } else if (wim->hdr.integrity_table_reshdr.offset_in_wim != 0) {
3015 /* Old WIM has an integrity table; begin writing new streams
3017 old_wim_end = wim->hdr.integrity_table_reshdr.offset_in_wim +
3018 wim->hdr.integrity_table_reshdr.size_in_wim;
3020 /* No existing integrity table; begin writing new streams after
3021 * the old XML data. */
3022 old_wim_end = old_xml_end;
3025 ret = check_resource_offsets(wim, old_wim_end);
3027 goto out_restore_memory_hdr;
3029 ret = prepare_stream_list_for_write(wim, WIMLIB_ALL_IMAGES, write_flags,
3030 &stream_list, &lookup_table_list,
3033 goto out_restore_memory_hdr;
3035 ret = open_wim_writable(wim, wim->filename, O_RDWR);
3037 goto out_restore_memory_hdr;
3039 ret = lock_wim(wim, wim->out_fd.fd);
3043 /* Set WIM_HDR_FLAG_WRITE_IN_PROGRESS flag in header. */
3044 wim->hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
3045 ret = write_wim_header_flags(wim->hdr.flags, &wim->out_fd);
3047 ERROR_WITH_ERRNO("Error updating WIM header flags");
3048 goto out_unlock_wim;
3051 if (filedes_seek(&wim->out_fd, old_wim_end) == -1) {
3052 ERROR_WITH_ERRNO("Can't seek to end of WIM");
3053 ret = WIMLIB_ERR_WRITE;
3054 goto out_restore_physical_hdr;
3057 ret = wim_write_stream_list(wim,
3066 ret = write_wim_metadata_resources(wim, WIMLIB_ALL_IMAGES,
3067 write_flags, progress_func);
3071 write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE;
3072 ret = finish_write(wim, WIMLIB_ALL_IMAGES, write_flags,
3073 progress_func, &lookup_table_list);
3077 wim->wim_locked = 0;
3081 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
3082 WARNING("Truncating `%"TS"' to its original size (%"PRIu64" bytes)",
3083 wim->filename, old_wim_end);
3084 /* Return value of ftruncate() is ignored because this is
3085 * already an error path. */
3086 (void)ftruncate(wim->out_fd.fd, old_wim_end);
3088 out_restore_physical_hdr:
3089 (void)write_wim_header_flags(hdr_save.flags, &wim->out_fd);
3091 wim->wim_locked = 0;
3093 (void)close_wim_writable(wim, write_flags);
3094 out_restore_memory_hdr:
3095 memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header));
3100 overwrite_wim_via_tmpfile(WIMStruct *wim, int write_flags,
3101 unsigned num_threads,
3102 wimlib_progress_func_t progress_func)
3104 size_t wim_name_len;
3107 DEBUG("Overwriting `%"TS"' via a temporary file", wim->filename);
3109 /* Write the WIM to a temporary file in the same directory as the
3111 wim_name_len = tstrlen(wim->filename);
3112 tchar tmpfile[wim_name_len + 10];
3113 tmemcpy(tmpfile, wim->filename, wim_name_len);
3114 randomize_char_array_with_alnum(tmpfile + wim_name_len, 9);
3115 tmpfile[wim_name_len + 9] = T('\0');
3117 ret = wimlib_write(wim, tmpfile, WIMLIB_ALL_IMAGES,
3118 write_flags | WIMLIB_WRITE_FLAG_FSYNC,
3119 num_threads, progress_func);
3127 /* Rename the new WIM file to the original WIM file. Note: on Windows
3128 * this actually calls win32_rename_replacement(), not _wrename(), so
3129 * that removing the existing destination file can be handled. */
3130 DEBUG("Renaming `%"TS"' to `%"TS"'", tmpfile, wim->filename);
3131 ret = trename(tmpfile, wim->filename);
3133 ERROR_WITH_ERRNO("Failed to rename `%"TS"' to `%"TS"'",
3134 tmpfile, wim->filename);
3141 return WIMLIB_ERR_RENAME;
3144 if (progress_func) {
3145 union wimlib_progress_info progress;
3146 progress.rename.from = tmpfile;
3147 progress.rename.to = wim->filename;
3148 progress_func(WIMLIB_PROGRESS_MSG_RENAME, &progress);
3153 /* Determine if the specified WIM file may be updated by appending in-place
3154 * rather than writing and replacing it with an entirely new file. */
3156 can_overwrite_wim_inplace(const WIMStruct *wim, int write_flags)
3158 /* REBUILD flag forces full rebuild. */
3159 if (write_flags & WIMLIB_WRITE_FLAG_REBUILD)
3162 /* Deletions cause full rebuild by default. */
3163 if (wim->deletion_occurred && !(write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE))
3166 /* Pipable WIMs cannot be updated in place, nor can a non-pipable WIM be
3167 * turned into a pipable WIM in-place. */
3168 if (wim_is_pipable(wim) || (write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
3171 /* wimlib allows multiple packs in a single WIM, but they don't seem to
3172 * be compatible with WIMGAPI, so force all streams to be repacked if
3173 * the WIM already may have contained a pack and PACK_STREAMS was
3175 if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS &&
3176 wim->hdr.wim_version == WIM_VERSION_PACKED_STREAMS)
3179 /* The default compression type and compression chunk size selected for
3180 * the output WIM must be the same as those currently used for the WIM.
3182 if (wim->compression_type != wim->out_compression_type)
3184 if (wim->chunk_size != wim->out_chunk_size)
3190 /* API function documented in wimlib.h */
3192 wimlib_overwrite(WIMStruct *wim, int write_flags,
3193 unsigned num_threads,
3194 wimlib_progress_func_t progress_func)
3199 write_flags &= WIMLIB_WRITE_MASK_PUBLIC;
3201 if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)
3202 return WIMLIB_ERR_INVALID_PARAM;
3205 return WIMLIB_ERR_NO_FILENAME;
3207 orig_hdr_flags = wim->hdr.flags;
3208 if (write_flags & WIMLIB_WRITE_FLAG_IGNORE_READONLY_FLAG)
3209 wim->hdr.flags &= ~WIM_HDR_FLAG_READONLY;
3210 ret = can_modify_wim(wim);
3211 wim->hdr.flags = orig_hdr_flags;
3215 if (can_overwrite_wim_inplace(wim, write_flags)) {
3216 ret = overwrite_wim_inplace(wim, write_flags, num_threads,
3218 if (ret != WIMLIB_ERR_RESOURCE_ORDER)
3220 WARNING("Falling back to re-building entire WIM");
3222 return overwrite_wim_via_tmpfile(wim, write_flags, num_threads,