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->wim->compression_type == out_ctype &&
176 rspec->wim->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,
287 struct wim_lookup_table_entry *cur_stream)
289 union wimlib_progress_info *progress = &progress_data->progress;
293 progress->write_streams.total_bytes -= size;
294 if (progress_data->next_progress != ~(uint64_t)0 &&
295 progress_data->next_progress > progress->write_streams.total_bytes)
297 progress_data->next_progress = progress->write_streams.total_bytes;
300 progress->write_streams.completed_bytes += size;
302 new_wim_part = false;
303 if (cur_stream->resource_location == RESOURCE_IN_WIM &&
304 cur_stream->rspec->wim != progress_data->prev_wim_part)
306 if (progress_data->prev_wim_part) {
308 progress->write_streams.completed_parts++;
310 progress_data->prev_wim_part = cur_stream->rspec->wim;
312 progress->write_streams.completed_streams++;
313 if (progress_data->progress_func
314 && (progress->write_streams.completed_bytes >= progress_data->next_progress
317 progress_data->progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
319 if (progress_data->next_progress == progress->write_streams.total_bytes) {
320 progress_data->next_progress = ~(uint64_t)0;
322 progress_data->next_progress =
323 min(progress->write_streams.total_bytes,
324 progress->write_streams.completed_bytes +
325 progress->write_streams.total_bytes / 100);
330 struct write_streams_ctx {
331 /* File descriptor the streams are being written to. */
332 struct filedes *out_fd;
334 /* Lookup table for the WIMStruct on whose behalf the streams are being
336 struct wim_lookup_table *lookup_table;
338 /* Compression format to use. */
341 /* Maximum uncompressed chunk size in compressed resources to use. */
344 /* Flags that affect how the streams will be written. */
345 int write_resource_flags;
347 /* Data used for issuing WRITE_STREAMS progress. */
348 struct write_streams_progress_data progress_data;
350 struct filter_context *filter_ctx;
352 /* Upper bound on the total number of bytes that need to be compressed.
354 u64 num_bytes_to_compress;
356 /* Pointer to the chunk_compressor implementation being used for
357 * compressing chunks of data, or NULL if chunks are being written
359 struct chunk_compressor *compressor;
361 /* Buffer for dividing the read data into chunks of size
362 * @out_chunk_size. */
365 /* Number of bytes in @chunk_buf that are currently filled. */
366 size_t chunk_buf_filled;
368 /* List of streams that currently have chunks being compressed. */
369 struct list_head pending_streams;
371 /* Set to true if the stream currently being read was a duplicate, and
372 * therefore the corresponding stream entry needs to be freed once the
373 * read finishes. (In this case we add the duplicate entry to
374 * pending_streams rather than the entry being read.) */
375 bool stream_was_duplicate;
377 /* Current uncompressed offset in the resource being read. */
378 u64 cur_read_res_offset;
380 /* Uncompressed size of the resource currently being read. */
381 u64 cur_read_res_size;
383 /* Current uncompressed offset in the resource being written. */
384 u64 cur_write_res_offset;
386 /* Uncompressed size of resource currently being written. */
387 u64 cur_write_res_size;
389 /* Array that is filled in with compressed chunk sizes as a resource is
393 /* Index of next entry in @chunk_csizes to fill in. */
396 /* Number of entries in @chunk_csizes currently allocated. */
397 size_t num_alloc_chunks;
399 /* Offset in the output file of the start of the chunks of the resource
400 * currently being written. */
401 u64 chunks_start_offset;
405 get_chunk_entry_size(u64 res_size, int write_resource_flags)
407 if (res_size <= UINT32_MAX ||
408 (write_resource_flags & WIM_RESHDR_FLAG_PACKED_STREAMS))
414 /* Reserve space for the chunk table and prepare to accumulate the chunk table
417 begin_chunk_table(struct write_streams_ctx *ctx, u64 res_expected_size)
419 u64 expected_num_chunks;
420 u64 expected_num_chunk_entries;
424 /* Calculate the number of chunks and chunk entries that should be
425 * needed for the resource. These normally will be the final values,
426 * but in PACKED_STREAMS mode some of the streams we're planning to
427 * write into the resource may be duplicates, and therefore discarded,
428 * potentially decreasing the number of chunk entries needed. */
429 expected_num_chunks = DIV_ROUND_UP(res_expected_size, ctx->out_chunk_size);
430 expected_num_chunk_entries = expected_num_chunks;
431 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
432 expected_num_chunk_entries--;
434 /* Make sure the chunk_csizes array is long enough to store the
435 * compressed size of each chunk. */
436 if (expected_num_chunks > ctx->num_alloc_chunks) {
437 u64 new_length = expected_num_chunks + 50;
439 if ((size_t)new_length != new_length) {
440 ERROR("Resource size too large (%"PRIu64" bytes!",
442 return WIMLIB_ERR_NOMEM;
445 FREE(ctx->chunk_csizes);
446 ctx->chunk_csizes = MALLOC(new_length * sizeof(ctx->chunk_csizes[0]));
447 if (ctx->chunk_csizes == NULL) {
448 ctx->num_alloc_chunks = 0;
449 return WIMLIB_ERR_NOMEM;
451 ctx->num_alloc_chunks = new_length;
454 ctx->chunk_index = 0;
456 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE)) {
457 /* Reserve space for the chunk table in the output file. In the
458 * case of packed resources this reserves the upper bound for
459 * the needed space, not necessarily the exact space which will
460 * prove to be needed. At this point, we just use @chunk_csizes
461 * for a buffer of 0's because the actual compressed chunk sizes
463 reserve_size = expected_num_chunk_entries *
464 get_chunk_entry_size(res_expected_size,
465 ctx->write_resource_flags);
466 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)
467 reserve_size += sizeof(struct alt_chunk_table_header_disk);
468 memset(ctx->chunk_csizes, 0, reserve_size);
469 ret = full_write(ctx->out_fd, ctx->chunk_csizes, reserve_size);
477 begin_write_resource(struct write_streams_ctx *ctx, u64 res_expected_size)
481 wimlib_assert(res_expected_size != 0);
483 if (ctx->compressor != NULL) {
484 ret = begin_chunk_table(ctx, res_expected_size);
489 /* Output file descriptor is now positioned at the offset at which to
490 * write the first chunk of the resource. */
491 ctx->chunks_start_offset = ctx->out_fd->offset;
492 ctx->cur_write_res_offset = 0;
493 ctx->cur_write_res_size = res_expected_size;
498 end_chunk_table(struct write_streams_ctx *ctx, u64 res_actual_size,
499 u64 *res_start_offset_ret, u64 *res_store_size_ret)
501 size_t actual_num_chunks;
502 size_t actual_num_chunk_entries;
503 size_t chunk_entry_size;
506 actual_num_chunks = ctx->chunk_index;
507 actual_num_chunk_entries = actual_num_chunks;
508 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
509 actual_num_chunk_entries--;
511 chunk_entry_size = get_chunk_entry_size(res_actual_size,
512 ctx->write_resource_flags);
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_res_size == ctx->cur_write_res_offset);
611 res_uncompressed_size = ctx->cur_write_res_size;
613 if (ctx->compressor) {
614 ret = end_chunk_table(ctx, res_uncompressed_size,
615 &res_offset_in_wim, &res_size_in_wim);
619 res_offset_in_wim = ctx->chunks_start_offset;
620 res_size_in_wim = ctx->out_fd->offset - res_offset_in_wim;
622 out_reshdr->uncompressed_size = res_uncompressed_size;
623 out_reshdr->size_in_wim = res_size_in_wim;
624 out_reshdr->offset_in_wim = res_offset_in_wim;
625 DEBUG("Finished writing resource: %"PRIu64" => %"PRIu64" @ %"PRIu64"",
626 res_uncompressed_size, res_size_in_wim, res_offset_in_wim);
630 /* Begin processing a stream for writing. */
632 write_stream_begin_read(struct wim_lookup_table_entry *lte,
633 bool is_partial_res, void *_ctx)
635 struct write_streams_ctx *ctx = _ctx;
638 wimlib_assert(lte->size > 0);
640 ctx->cur_read_res_offset = 0;
641 ctx->cur_read_res_size = lte->size;
643 /* As an optimization, we allow some streams to be "unhashed", meaning
644 * their SHA1 message digests are unknown. This is the case with
645 * streams that are added by scanning a directry tree with
646 * wimlib_add_image(), for example. Since WIM uses single-instance
647 * streams, we don't know whether such each such stream really need to
648 * written until it is actually checksummed, unless it has a unique
649 * size. In such cases we read and checksum the stream in this
650 * function, thereby advancing ahead of read_stream_list(), which will
651 * still provide the data again to write_stream_process_chunk(). This
652 * is okay because an unhashed stream cannot be in a WIM resource, which
653 * might be costly to decompress. */
654 ctx->stream_was_duplicate = false;
655 if (ctx->lookup_table != NULL && lte->unhashed && !lte->unique_size) {
657 wimlib_assert(!is_partial_res);
659 struct wim_lookup_table_entry *lte_new;
661 ret = hash_unhashed_stream(lte, ctx->lookup_table, <e_new);
664 if (lte_new != lte) {
665 /* Duplicate stream detected. */
667 if (lte_new->will_be_in_output_wim ||
668 stream_filtered(lte_new, ctx->filter_ctx))
670 /* The duplicate stream is already being
671 * included in the output WIM, or it would be
672 * filtered out if it had been. Skip writing
673 * this stream (and reading it again) entirely,
674 * passing its output reference count to the
675 * duplicate stream in the former case. */
676 DEBUG("Discarding duplicate stream of "
677 "length %"PRIu64, lte->size);
678 do_write_streams_progress(&ctx->progress_data,
679 lte->size, true, lte);
680 list_del(<e->write_streams_list);
681 list_del(<e->lookup_table_list);
682 if (lte_new->will_be_in_output_wim)
683 lte_new->out_refcnt += lte->out_refcnt;
684 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)
685 ctx->cur_write_res_size -= lte->size;
686 free_lookup_table_entry(lte);
687 return BEGIN_STREAM_STATUS_SKIP_STREAM;
689 /* The duplicate stream can validly be written,
690 * but was not marked as such. Discard the
691 * current stream entry and use the duplicate,
692 * but actually freeing the current entry must
693 * wait until read_stream_list() has finished
694 * reading its data. */
695 DEBUG("Stream duplicate, but not already "
696 "selected for writing.");
697 list_replace(<e->write_streams_list,
698 <e_new->write_streams_list);
699 list_replace(<e->lookup_table_list,
700 <e_new->lookup_table_list);
701 lte_new->out_refcnt = lte->out_refcnt;
702 lte_new->will_be_in_output_wim = 1;
703 ctx->stream_was_duplicate = true;
708 list_move_tail(<e->write_streams_list, &ctx->pending_streams);
712 /* Rewrite a stream that was just written compressed as uncompressed instead.
713 * This function is optional, but if a stream did not compress to less than its
714 * original size, it might as well be written uncompressed. */
716 write_stream_uncompressed(struct wim_lookup_table_entry *lte,
717 struct filedes *out_fd)
720 u64 begin_offset = lte->out_reshdr.offset_in_wim;
721 u64 end_offset = out_fd->offset;
723 if (filedes_seek(out_fd, begin_offset) == -1)
726 ret = extract_full_stream_to_fd(lte, out_fd);
728 /* Error reading the uncompressed data. */
729 if (out_fd->offset == begin_offset &&
730 filedes_seek(out_fd, end_offset) != -1)
732 /* Nothing was actually written yet, and we successfully
733 * seeked to the end of the compressed resource, so
734 * don't issue a hard error; just keep the compressed
735 * resource instead. */
736 WARNING("Recovered compressed stream of "
737 "size %"PRIu64", continuing on.",
744 wimlib_assert(out_fd->offset - begin_offset == lte->size);
746 if (out_fd->offset < end_offset &&
747 0 != ftruncate(out_fd->fd, out_fd->offset))
749 ERROR_WITH_ERRNO("Can't truncate output file to "
750 "offset %"PRIu64, out_fd->offset);
751 return WIMLIB_ERR_WRITE;
754 lte->out_reshdr.size_in_wim = lte->size;
755 lte->out_reshdr.flags &= ~(WIM_RESHDR_FLAG_COMPRESSED |
756 WIM_RESHDR_FLAG_PACKED_STREAMS);
760 /* Write the next chunk of (typically compressed) data to the output WIM,
761 * handling the writing of the chunk table. */
763 write_chunk(struct write_streams_ctx *ctx, const void *cchunk,
764 size_t csize, size_t usize)
768 struct wim_lookup_table_entry *lte;
770 lte = list_entry(ctx->pending_streams.next,
771 struct wim_lookup_table_entry, write_streams_list);
773 if (ctx->cur_write_res_offset == 0 &&
774 !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
776 /* Starting to write a new stream in non-packed mode. */
778 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
779 int additional_reshdr_flags = 0;
780 if (ctx->compressor != NULL)
781 additional_reshdr_flags |= WIM_RESHDR_FLAG_COMPRESSED;
783 DEBUG("Writing pipable WIM stream header "
784 "(offset=%"PRIu64")", ctx->out_fd->offset);
786 ret = write_pwm_stream_header(lte, ctx->out_fd,
787 additional_reshdr_flags);
792 ret = begin_write_resource(ctx, lte->size);
797 if (ctx->compressor != NULL) {
798 /* Record the compresed chunk size. */
799 wimlib_assert(ctx->chunk_index < ctx->num_alloc_chunks);
800 ctx->chunk_csizes[ctx->chunk_index++] = csize;
802 /* If writing a pipable WIM, before the chunk data write a chunk
803 * header that provides the compressed chunk size. */
804 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
805 struct pwm_chunk_hdr chunk_hdr = {
806 .compressed_size = cpu_to_le32(csize),
808 ret = full_write(ctx->out_fd, &chunk_hdr,
815 /* Write the chunk data. */
816 ret = full_write(ctx->out_fd, cchunk, csize);
820 ctx->cur_write_res_offset += usize;
822 do_write_streams_progress(&ctx->progress_data,
825 if (ctx->cur_write_res_offset == ctx->cur_write_res_size &&
826 !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
828 wimlib_assert(ctx->cur_write_res_offset == lte->size);
830 /* Finished writing a stream in non-packed mode. */
832 ret = end_write_resource(ctx, <e->out_reshdr);
836 lte->out_reshdr.flags = filter_resource_flags(lte->flags);
837 if (ctx->compressor != NULL)
838 lte->out_reshdr.flags |= WIM_RESHDR_FLAG_COMPRESSED;
840 if (ctx->compressor != NULL &&
841 lte->out_reshdr.size_in_wim >= lte->out_reshdr.uncompressed_size &&
842 !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) &&
843 !(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS))
845 /* Stream did not compress to less than its original
846 * size. If we're not writing a pipable WIM (which
847 * could mean the output file descriptor is
848 * non-seekable), and the stream isn't located in a
849 * resource pack (which would make reading it again
850 * costly), truncate the file to the start of the stream
851 * and write it uncompressed instead. */
852 DEBUG("Stream of size %"PRIu64" did not compress to "
853 "less than original size; writing uncompressed.",
855 ret = write_stream_uncompressed(lte, ctx->out_fd);
860 wimlib_assert(lte->out_reshdr.uncompressed_size == lte->size);
862 list_del(<e->write_streams_list);
863 ctx->cur_write_res_offset = 0;
869 ERROR_WITH_ERRNO("Write error");
874 submit_chunk_for_compression(struct write_streams_ctx *ctx,
875 const void *chunk, size_t size)
877 /* While we are unable to submit the chunk for compression (due to too
878 * many chunks already outstanding), retrieve and write the next
879 * compressed chunk. */
880 while (!ctx->compressor->submit_chunk(ctx->compressor, chunk, size)) {
887 bret = ctx->compressor->get_chunk(ctx->compressor,
888 &cchunk, &csize, &usize);
892 ret = write_chunk(ctx, cchunk, csize, usize);
899 /* Process the next chunk of data to be written to a WIM resource. */
901 write_stream_process_chunk(const void *chunk, size_t size, void *_ctx)
903 struct write_streams_ctx *ctx = _ctx;
905 const u8 *chunkptr, *chunkend;
907 wimlib_assert(size != 0);
909 if (ctx->compressor == NULL) {
910 /* Write chunk uncompressed. */
911 ret = write_chunk(ctx, chunk, size, size);
914 ctx->cur_read_res_offset += size;
918 /* Submit the chunk for compression, but take into account that the
919 * @size the chunk was provided in may not correspond to the
920 * @out_chunk_size being used for compression. */
922 chunkend = chunkptr + size;
924 const u8 *resized_chunk;
925 size_t needed_chunk_size;
927 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
928 needed_chunk_size = ctx->out_chunk_size;
930 u64 res_bytes_remaining;
932 res_bytes_remaining = ctx->cur_read_res_size -
933 ctx->cur_read_res_offset;
934 needed_chunk_size = min(ctx->out_chunk_size,
935 ctx->chunk_buf_filled +
936 res_bytes_remaining);
939 if (ctx->chunk_buf_filled == 0 &&
940 chunkend - chunkptr >= needed_chunk_size)
942 /* No intermediate buffering needed. */
943 resized_chunk = chunkptr;
944 chunkptr += needed_chunk_size;
945 ctx->cur_read_res_offset += needed_chunk_size;
947 /* Intermediate buffering needed. */
948 size_t bytes_consumed;
950 bytes_consumed = min(chunkend - chunkptr,
951 needed_chunk_size - ctx->chunk_buf_filled);
953 memcpy(&ctx->chunk_buf[ctx->chunk_buf_filled],
954 chunkptr, bytes_consumed);
956 chunkptr += bytes_consumed;
957 ctx->cur_read_res_offset += bytes_consumed;
958 ctx->chunk_buf_filled += bytes_consumed;
959 if (ctx->chunk_buf_filled == needed_chunk_size) {
960 resized_chunk = ctx->chunk_buf;
961 ctx->chunk_buf_filled = 0;
968 ret = submit_chunk_for_compression(ctx, resized_chunk,
973 } while (chunkptr != chunkend);
977 /* Finish processing a stream for writing. It may not have been completely
978 * written yet, as the chunk_compressor implementation may still have chunks
979 * buffered or being compressed. */
981 write_stream_end_read(struct wim_lookup_table_entry *lte, int status, void *_ctx)
983 struct write_streams_ctx *ctx = _ctx;
985 wimlib_assert(ctx->cur_read_res_offset == ctx->cur_read_res_size);
986 if (ctx->stream_was_duplicate) {
987 free_lookup_table_entry(lte);
988 } else if (lte->unhashed && ctx->lookup_table != NULL) {
989 list_del(<e->unhashed_list);
990 lookup_table_insert(ctx->lookup_table, lte);
996 /* Compute statistics about a list of streams that will be written.
998 * Assumes the streams are sorted such that all streams located in each distinct
999 * WIM (specified by WIMStruct) are together. */
1001 compute_stream_list_stats(struct list_head *stream_list,
1002 struct write_streams_ctx *ctx)
1004 struct wim_lookup_table_entry *lte;
1005 u64 total_bytes = 0;
1006 u64 num_streams = 0;
1007 u64 total_parts = 0;
1008 WIMStruct *prev_wim_part = NULL;
1010 list_for_each_entry(lte, stream_list, write_streams_list) {
1012 total_bytes += lte->size;
1013 if (lte->resource_location == RESOURCE_IN_WIM) {
1014 if (prev_wim_part != lte->rspec->wim) {
1015 prev_wim_part = lte->rspec->wim;
1020 ctx->progress_data.progress.write_streams.total_bytes = total_bytes;
1021 ctx->progress_data.progress.write_streams.total_streams = num_streams;
1022 ctx->progress_data.progress.write_streams.completed_bytes = 0;
1023 ctx->progress_data.progress.write_streams.completed_streams = 0;
1024 ctx->progress_data.progress.write_streams.compression_type = ctx->out_ctype;
1025 ctx->progress_data.progress.write_streams.total_parts = total_parts;
1026 ctx->progress_data.progress.write_streams.completed_parts = 0;
1027 ctx->progress_data.next_progress = 0;
1028 ctx->progress_data.prev_wim_part = NULL;
1031 /* Find streams in @stream_list that can be copied to the output WIM in raw form
1032 * rather than compressed. Delete these streams from @stream_list, and move one
1033 * per resource to @raw_copy_resources. Return the total uncompressed size of
1034 * the streams that need to be compressed. */
1036 find_raw_copy_resources(struct list_head *stream_list,
1037 int write_resource_flags,
1040 struct list_head *raw_copy_resources)
1042 struct wim_lookup_table_entry *lte, *tmp;
1043 u64 num_bytes_to_compress = 0;
1045 INIT_LIST_HEAD(raw_copy_resources);
1047 /* Initialize temporary raw_copy_ok flag. */
1048 list_for_each_entry(lte, stream_list, write_streams_list)
1049 if (lte->resource_location == RESOURCE_IN_WIM)
1050 lte->rspec->raw_copy_ok = 0;
1052 list_for_each_entry_safe(lte, tmp, stream_list, write_streams_list) {
1053 if (lte->resource_location == RESOURCE_IN_WIM &&
1054 lte->rspec->raw_copy_ok)
1056 list_del(<e->write_streams_list);
1057 } else if (can_raw_copy(lte, write_resource_flags,
1058 out_ctype, out_chunk_size))
1060 lte->rspec->raw_copy_ok = 1;
1061 list_move_tail(<e->write_streams_list,
1062 raw_copy_resources);
1064 num_bytes_to_compress += lte->size;
1068 return num_bytes_to_compress;
1071 /* Copy a raw compressed resource located in another WIM file to the WIM file
1074 write_raw_copy_resource(struct wim_resource_spec *in_rspec,
1075 struct filedes *out_fd)
1077 u64 cur_read_offset;
1078 u64 end_read_offset;
1079 u8 buf[BUFFER_SIZE];
1080 size_t bytes_to_read;
1082 struct filedes *in_fd;
1083 struct wim_lookup_table_entry *lte;
1084 u64 out_offset_in_wim;
1086 DEBUG("Copying raw compressed data (size_in_wim=%"PRIu64", "
1087 "uncompressed_size=%"PRIu64")",
1088 in_rspec->size_in_wim, in_rspec->uncompressed_size);
1090 /* Copy the raw data. */
1091 cur_read_offset = in_rspec->offset_in_wim;
1092 end_read_offset = cur_read_offset + in_rspec->size_in_wim;
1094 out_offset_in_wim = out_fd->offset;
1096 if (in_rspec->is_pipable) {
1097 if (cur_read_offset < sizeof(struct pwm_stream_hdr))
1098 return WIMLIB_ERR_INVALID_PIPABLE_WIM;
1099 cur_read_offset -= sizeof(struct pwm_stream_hdr);
1100 out_offset_in_wim += sizeof(struct pwm_stream_hdr);
1102 in_fd = &in_rspec->wim->in_fd;
1103 wimlib_assert(cur_read_offset != end_read_offset);
1106 bytes_to_read = min(sizeof(buf), end_read_offset - cur_read_offset);
1108 ret = full_pread(in_fd, buf, bytes_to_read, cur_read_offset);
1112 ret = full_write(out_fd, buf, bytes_to_read);
1116 cur_read_offset += bytes_to_read;
1118 } while (cur_read_offset != end_read_offset);
1120 list_for_each_entry(lte, &in_rspec->stream_list, rspec_node) {
1121 if (lte->will_be_in_output_wim) {
1122 stream_set_out_reshdr_for_reuse(lte);
1123 if (in_rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS)
1124 lte->out_res_offset_in_wim = out_offset_in_wim;
1126 lte->out_reshdr.offset_in_wim = out_offset_in_wim;
1133 /* Copy a list of raw compressed resources located other WIM file(s) to the WIM
1134 * file being written. */
1136 write_raw_copy_resources(struct list_head *raw_copy_resources,
1137 struct filedes *out_fd,
1138 struct write_streams_progress_data *progress_data)
1140 struct wim_lookup_table_entry *lte;
1143 list_for_each_entry(lte, raw_copy_resources, write_streams_list) {
1144 ret = write_raw_copy_resource(lte->rspec, out_fd);
1147 do_write_streams_progress(progress_data, lte->size, false, lte);
1152 /* Wait for and write all chunks pending in the compressor. */
1154 finish_remaining_chunks(struct write_streams_ctx *ctx)
1161 if (ctx->compressor == NULL)
1164 if (ctx->chunk_buf_filled != 0) {
1165 ret = submit_chunk_for_compression(ctx, ctx->chunk_buf,
1166 ctx->chunk_buf_filled);
1171 while (ctx->compressor->get_chunk(ctx->compressor, &cdata, &csize, &usize)) {
1172 ret = write_chunk(ctx, cdata, csize, usize);
1180 remove_zero_length_streams(struct list_head *stream_list)
1182 struct wim_lookup_table_entry *lte, *tmp;
1184 list_for_each_entry_safe(lte, tmp, stream_list, write_streams_list) {
1185 wimlib_assert(lte->will_be_in_output_wim);
1186 if (lte->size == 0) {
1187 list_del(<e->write_streams_list);
1188 lte->out_reshdr.offset_in_wim = 0;
1189 lte->out_reshdr.size_in_wim = 0;
1190 lte->out_reshdr.uncompressed_size = 0;
1191 lte->out_reshdr.flags = filter_resource_flags(lte->flags);
1197 * Write a list of streams to the output WIM file.
1200 * The list of streams to write, specifies a list of `struct
1201 * wim_lookup_table_entry's linked by the 'write_streams_list' member.
1204 * The file descriptor, opened for writing, to which to write the streams.
1206 * @write_resource_flags
1207 * Flags to modify how the streams are written:
1209 * WRITE_RESOURCE_FLAG_RECOMPRESS:
1210 * Force compression of all resources, even if they could otherwise
1211 * be re-used by caping the raw data, due to being located in a WIM
1212 * file with compatible compression parameters.
1214 * WRITE_RESOURCE_FLAG_PIPABLE:
1215 * Write the resources in the wimlib-specific pipable format, and
1216 * furthermore do so in such a way that no seeking backwards in
1217 * @out_fd will be performed (so it may be a pipe, contrary to the
1218 * default behavior).
1220 * WRITE_RESOURCE_FLAG_PACK_STREAMS:
1221 * Pack all the streams into a single resource rather than writing
1222 * them in separate resources. This format is only valid if the
1223 * WIM version number is WIM_VERSION_PACKED_STREAMS. This flag
1224 * currently may not be combined with WRITE_RESOURCE_FLAG_PIPABLE.
1227 * Compression format to use to write the output streams, specified as one
1228 * of the WIMLIB_COMPRESSION_TYPE_* constants, excepting
1229 * WIMLIB_COMPRESSION_TYPE_INVALID but including
1230 * WIMLIB_COMPRESSION_TYPE_NONE.
1233 * Chunk size to use to write the streams. It must be a valid chunk size
1234 * for the specified compression format @out_ctype, unless @out_ctype is
1235 * WIMLIB_COMPRESSION_TYPE_NONE, in which case this parameter is ignored.
1238 * Number of threads to use to compress data. If 0, a default number of
1239 * threads will be chosen. The number of threads still may be decreased
1240 * from the specified value if insufficient memory is detected.
1243 * If on-the-fly deduplication of unhashed streams is desired, this
1244 * parameter must be pointer to the lookup table for the WIMStruct on whose
1245 * behalf the streams are being written. Otherwise, this parameter can be
1249 * If on-the-fly deduplication of unhashed streams is desired, this
1250 * parameter can be a pointer to a context for stream filtering used to
1251 * detect whether the duplicate stream has been hard-filtered or not. If
1252 * no streams are hard-filtered or no streams are unhashed, this parameter
1256 * If non-NULL, a progress function that will be called periodically with
1257 * WIMLIB_PROGRESS_MSG_WRITE_STREAMS messages. Note that on-the-fly
1258 * deduplication of unhashed streams may result in the total bytes provided
1259 * in the progress data to decrease from one message to the next.
1261 * This function will write the streams in @stream_list to resources in
1262 * consecutive positions in the output WIM file, or to a single packed resource
1263 * if WRITE_RESOURCE_FLAG_PACK_STREAMS was specified in @write_resource_flags.
1264 * In both cases, the @out_reshdr of the `struct wim_lookup_table_entry' for
1265 * each stream written will be updated to specify its location, size, and flags
1266 * in the output WIM. In the packed resource case,
1267 * WIM_RESHDR_FLAG_PACKED_STREAMS shall be set in the @flags field of the
1268 * @out_reshdr, and @out_res_offset_in_wim and @out_res_size_in_wim will also
1269 * be set to the offset and size, respectively, in the output WIM of the full
1270 * packed resource containing the corresponding stream.
1272 * Each of the streams to write may be in any location supported by the
1273 * resource-handling code (specifically, read_stream_list()), such as the
1274 * contents of external file that has been logically added to the output WIM, or
1275 * a stream in another WIM file that has been imported, or even stream in the
1276 * "same" WIM file of which a modified copy is being written. In the case that
1277 * a stream is already in a WIM file and uses compatible compression parameters,
1278 * by default this function will re-use the raw data instead of decompressing
1279 * it, then recompressing it; however, with WRITE_RESOURCE_FLAG_RECOMPRESS
1280 * specified in @write_resource_flags, this is not done.
1282 * As a further requirement, this function requires that the
1283 * @will_be_in_output_wim member be set on all streams in @stream_list as well
1284 * as any other streams not in @stream_list that will be in the output WIM file,
1285 * but not on any other streams in the output WIM's lookup table or sharing a
1286 * packed resource with a stream in @stream_list. Still furthermore, if
1287 * on-the-fly deduplication of streams is possible, then all streams in
1288 * @stream_list must also be linked by @lookup_table_list along with any other
1289 * streams that have @will_be_in_output_wim set.
1291 * This function handles on-the-fly deduplication of streams for which SHA1
1292 * message digests have not yet been calculated and it is therefore known
1293 * whether such streams are already in @stream_list or in the WIM's lookup table
1294 * at all. If @lookup_table is non-NULL, then each stream in @stream_list that
1295 * has @unhashed set but not @unique_size set is checksummed immediately before
1296 * it would otherwise be read for writing in order to determine if it is
1297 * identical to another stream already being written or one that would be
1298 * filtered out of the output WIM using stream_filtered() with the context
1299 * @filter_ctx. Each such duplicate stream will be removed from @stream_list, its
1300 * reference count transfered to the pre-existing duplicate stream, its memory
1301 * freed, and will not be written. Alternatively, if a stream in @stream_list
1302 * is a duplicate with any stream in @lookup_table that has not been marked for
1303 * writing or would not be hard-filtered, it is freed and the pre-existing
1304 * duplicate is written instead, taking ownership of the reference count and
1305 * slot in the @lookup_table_list.
1307 * Returns 0 if all streams were written successfully (or did not need to be
1308 * written); otherwise a non-zero error code.
1311 write_stream_list(struct list_head *stream_list,
1312 struct filedes *out_fd,
1313 int write_resource_flags,
1316 unsigned num_threads,
1317 struct wim_lookup_table *lookup_table,
1318 struct filter_context *filter_ctx,
1319 wimlib_progress_func_t progress_func)
1322 struct write_streams_ctx ctx;
1323 struct list_head raw_copy_resources;
1325 wimlib_assert((write_resource_flags &
1326 (WRITE_RESOURCE_FLAG_PACK_STREAMS |
1327 WRITE_RESOURCE_FLAG_PIPABLE)) !=
1328 (WRITE_RESOURCE_FLAG_PACK_STREAMS |
1329 WRITE_RESOURCE_FLAG_PIPABLE));
1331 remove_zero_length_streams(stream_list);
1333 if (list_empty(stream_list)) {
1334 DEBUG("No streams to write.");
1338 memset(&ctx, 0, sizeof(ctx));
1340 /* Pre-sorting the streams is required for compute_stream_list_stats().
1341 * Afterwards, read_stream_list() need not sort them again. */
1342 ret = sort_stream_list_by_sequential_order(stream_list,
1343 offsetof(struct wim_lookup_table_entry,
1344 write_streams_list));
1348 ctx.out_fd = out_fd;
1349 ctx.lookup_table = lookup_table;
1350 ctx.out_ctype = out_ctype;
1351 ctx.out_chunk_size = out_chunk_size;
1352 ctx.write_resource_flags = write_resource_flags;
1353 ctx.filter_ctx = filter_ctx;
1355 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) {
1356 wimlib_assert(out_chunk_size != 0);
1357 if (out_chunk_size <= STACK_MAX) {
1358 ctx.chunk_buf = alloca(out_chunk_size);
1360 ctx.chunk_buf = MALLOC(out_chunk_size);
1361 if (ctx.chunk_buf == NULL) {
1362 ret = WIMLIB_ERR_NOMEM;
1363 goto out_destroy_context;
1367 ctx.chunk_buf_filled = 0;
1369 compute_stream_list_stats(stream_list, &ctx);
1371 ctx.progress_data.progress_func = progress_func;
1373 ctx.num_bytes_to_compress = find_raw_copy_resources(stream_list,
1374 write_resource_flags,
1377 &raw_copy_resources);
1379 DEBUG("Writing stream list "
1380 "(offset = %"PRIu64", write_resource_flags=0x%08x, "
1381 "out_ctype=%d, out_chunk_size=%u, num_threads=%u, "
1382 "total_bytes=%"PRIu64", num_bytes_to_compress=%"PRIu64")",
1383 out_fd->offset, write_resource_flags,
1384 out_ctype, out_chunk_size, num_threads,
1385 ctx.progress_data.progress.write_streams.total_bytes,
1386 ctx.num_bytes_to_compress);
1388 if (ctx.num_bytes_to_compress == 0) {
1389 DEBUG("No compression needed; skipping to raw copy!");
1390 goto out_write_raw_copy_resources;
1393 /* Unless uncompressed output was required, allocate a chunk_compressor
1394 * to do compression. There are serial and parallel implementations of
1395 * the chunk_compressor interface. We default to parallel using the
1396 * specified number of threads, unless the upper bound on the number
1397 * bytes needing to be compressed is less 2000000 (heuristic value). */
1398 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) {
1400 #ifdef ENABLE_MULTITHREADED_COMPRESSION
1401 if (ctx.num_bytes_to_compress >= 2000000) {
1402 ret = new_parallel_chunk_compressor(out_ctype,
1407 DEBUG("Couldn't create parallel chunk compressor "
1408 "(status %d)", ret);
1413 if (ctx.compressor == NULL) {
1414 ret = new_serial_chunk_compressor(out_ctype, out_chunk_size,
1417 goto out_destroy_context;
1422 ctx.progress_data.progress.write_streams.num_threads = ctx.compressor->num_threads;
1424 ctx.progress_data.progress.write_streams.num_threads = 1;
1426 DEBUG("Actually using %u threads",
1427 ctx.progress_data.progress.write_streams.num_threads);
1429 INIT_LIST_HEAD(&ctx.pending_streams);
1431 if (ctx.progress_data.progress_func) {
1432 (*ctx.progress_data.progress_func)(WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
1433 &ctx.progress_data.progress);
1436 if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
1437 ret = begin_write_resource(&ctx, ctx.num_bytes_to_compress);
1439 goto out_destroy_context;
1442 /* Read the list of streams needing to be compressed, using the
1443 * specified callbacks to execute processing of the data. */
1445 struct read_stream_list_callbacks cbs = {
1446 .begin_stream = write_stream_begin_read,
1447 .begin_stream_ctx = &ctx,
1448 .consume_chunk = write_stream_process_chunk,
1449 .consume_chunk_ctx = &ctx,
1450 .end_stream = write_stream_end_read,
1451 .end_stream_ctx = &ctx,
1454 ret = read_stream_list(stream_list,
1455 offsetof(struct wim_lookup_table_entry, write_streams_list),
1457 STREAM_LIST_ALREADY_SORTED |
1458 VERIFY_STREAM_HASHES |
1459 COMPUTE_MISSING_STREAM_HASHES);
1462 goto out_destroy_context;
1464 ret = finish_remaining_chunks(&ctx);
1466 goto out_destroy_context;
1468 if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
1469 struct wim_reshdr reshdr;
1470 struct wim_lookup_table_entry *lte;
1473 ret = end_write_resource(&ctx, &reshdr);
1475 goto out_destroy_context;
1477 DEBUG("Ending packed resource: %lu %lu %lu.",
1478 reshdr.offset_in_wim,
1480 reshdr.uncompressed_size);
1483 list_for_each_entry(lte, &ctx.pending_streams, write_streams_list) {
1484 lte->out_reshdr.size_in_wim = lte->size;
1485 lte->out_reshdr.flags = filter_resource_flags(lte->flags);
1486 lte->out_reshdr.flags |= WIM_RESHDR_FLAG_PACKED_STREAMS;
1487 lte->out_reshdr.uncompressed_size = 0;
1488 lte->out_reshdr.offset_in_wim = offset_in_res;
1489 lte->out_res_offset_in_wim = reshdr.offset_in_wim;
1490 lte->out_res_size_in_wim = reshdr.size_in_wim;
1491 /*lte->out_res_uncompressed_size = reshdr.uncompressed_size;*/
1492 offset_in_res += lte->size;
1494 wimlib_assert(offset_in_res == reshdr.uncompressed_size);
1497 out_write_raw_copy_resources:
1498 /* Copy any compressed resources for which the raw data can be reused
1499 * without decompression. */
1500 ret = write_raw_copy_resources(&raw_copy_resources, ctx.out_fd,
1501 &ctx.progress_data);
1503 out_destroy_context:
1504 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && out_chunk_size > STACK_MAX)
1505 FREE(ctx.chunk_buf);
1506 FREE(ctx.chunk_csizes);
1508 ctx.compressor->destroy(ctx.compressor);
1509 DEBUG("Done (ret=%d)", ret);
1514 wim_write_stream_list(WIMStruct *wim,
1515 struct list_head *stream_list,
1517 unsigned num_threads,
1518 struct filter_context *filter_ctx,
1519 wimlib_progress_func_t progress_func)
1523 int write_resource_flags;
1525 write_resource_flags = write_flags_to_resource_flags(write_flags);
1527 if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
1528 out_chunk_size = wim->out_pack_chunk_size;
1529 out_ctype = wim->out_pack_compression_type;
1531 out_chunk_size = wim->out_chunk_size;
1532 out_ctype = wim->out_compression_type;
1535 return write_stream_list(stream_list,
1537 write_resource_flags,
1547 write_wim_resource(struct wim_lookup_table_entry *lte,
1548 struct filedes *out_fd,
1551 int write_resource_flags)
1553 LIST_HEAD(stream_list);
1554 list_add(<e->write_streams_list, &stream_list);
1555 lte->will_be_in_output_wim = 1;
1556 return write_stream_list(&stream_list,
1558 write_resource_flags & ~WRITE_RESOURCE_FLAG_PACK_STREAMS,
1568 write_wim_resource_from_buffer(const void *buf, size_t buf_size,
1569 int reshdr_flags, struct filedes *out_fd,
1572 struct wim_reshdr *out_reshdr,
1574 int write_resource_flags)
1577 struct wim_lookup_table_entry *lte;
1579 /* Set up a temporary lookup table entry to provide to
1580 * write_wim_resource(). */
1582 lte = new_lookup_table_entry();
1584 return WIMLIB_ERR_NOMEM;
1586 lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER;
1587 lte->attached_buffer = (void*)buf;
1588 lte->size = buf_size;
1589 lte->flags = reshdr_flags;
1591 if (write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
1592 sha1_buffer(buf, buf_size, lte->hash);
1598 ret = write_wim_resource(lte, out_fd, out_ctype, out_chunk_size,
1599 write_resource_flags);
1603 copy_reshdr(out_reshdr, <e->out_reshdr);
1606 copy_hash(hash, lte->hash);
1609 lte->resource_location = RESOURCE_NONEXISTENT;
1610 free_lookup_table_entry(lte);
1614 struct stream_size_table {
1615 struct hlist_head *array;
1621 init_stream_size_table(struct stream_size_table *tab, size_t capacity)
1623 tab->array = CALLOC(capacity, sizeof(tab->array[0]));
1624 if (tab->array == NULL)
1625 return WIMLIB_ERR_NOMEM;
1626 tab->num_entries = 0;
1627 tab->capacity = capacity;
1632 destroy_stream_size_table(struct stream_size_table *tab)
1638 stream_size_table_insert(struct wim_lookup_table_entry *lte, void *_tab)
1640 struct stream_size_table *tab = _tab;
1642 struct wim_lookup_table_entry *same_size_lte;
1643 struct hlist_node *tmp;
1645 pos = hash_u64(lte->size) % tab->capacity;
1646 lte->unique_size = 1;
1647 hlist_for_each_entry(same_size_lte, tmp, &tab->array[pos], hash_list_2) {
1648 if (same_size_lte->size == lte->size) {
1649 lte->unique_size = 0;
1650 same_size_lte->unique_size = 0;
1655 hlist_add_head(<e->hash_list_2, &tab->array[pos]);
1660 struct find_streams_ctx {
1663 struct list_head stream_list;
1664 struct stream_size_table stream_size_tab;
1668 reference_stream_for_write(struct wim_lookup_table_entry *lte,
1669 struct list_head *stream_list, u32 nref)
1671 if (!lte->will_be_in_output_wim) {
1672 lte->out_refcnt = 0;
1673 list_add_tail(<e->write_streams_list, stream_list);
1674 lte->will_be_in_output_wim = 1;
1676 lte->out_refcnt += nref;
1680 fully_reference_stream_for_write(struct wim_lookup_table_entry *lte,
1683 struct list_head *stream_list = _stream_list;
1684 lte->will_be_in_output_wim = 0;
1685 reference_stream_for_write(lte, stream_list, lte->refcnt);
1690 inode_find_streams_to_reference(const struct wim_inode *inode,
1691 const struct wim_lookup_table *table,
1692 struct list_head *stream_list)
1694 struct wim_lookup_table_entry *lte;
1697 wimlib_assert(inode->i_nlink > 0);
1699 for (i = 0; i <= inode->i_num_ads; i++) {
1700 lte = inode_stream_lte(inode, i, table);
1702 reference_stream_for_write(lte, stream_list,
1704 else if (!is_zero_hash(inode_stream_hash(inode, i)))
1705 return WIMLIB_ERR_RESOURCE_NOT_FOUND;
1711 do_stream_set_not_in_output_wim(struct wim_lookup_table_entry *lte, void *_ignore)
1713 lte->will_be_in_output_wim = 0;
1718 image_find_streams_to_reference(WIMStruct *wim)
1720 struct wim_image_metadata *imd;
1721 struct wim_inode *inode;
1722 struct wim_lookup_table_entry *lte;
1723 struct list_head *stream_list;
1726 imd = wim_get_current_image_metadata(wim);
1728 image_for_each_unhashed_stream(lte, imd)
1729 lte->will_be_in_output_wim = 0;
1731 stream_list = wim->private;
1732 image_for_each_inode(inode, imd) {
1733 ret = inode_find_streams_to_reference(inode,
1743 prepare_unfiltered_list_of_streams_in_output_wim(WIMStruct *wim,
1746 struct list_head *stream_list_ret)
1750 INIT_LIST_HEAD(stream_list_ret);
1752 if (streams_ok && (image == WIMLIB_ALL_IMAGES ||
1753 (image == 1 && wim->hdr.image_count == 1)))
1755 /* Fast case: Assume that all streams are being written and
1756 * that the reference counts are correct. */
1757 struct wim_lookup_table_entry *lte;
1758 struct wim_image_metadata *imd;
1761 for_lookup_table_entry(wim->lookup_table,
1762 fully_reference_stream_for_write,
1765 for (i = 0; i < wim->hdr.image_count; i++) {
1766 imd = wim->image_metadata[i];
1767 image_for_each_unhashed_stream(lte, imd)
1768 fully_reference_stream_for_write(lte, stream_list_ret);
1771 /* Slow case: Walk through the images being written and
1772 * determine the streams referenced. */
1773 for_lookup_table_entry(wim->lookup_table,
1774 do_stream_set_not_in_output_wim, NULL);
1775 wim->private = stream_list_ret;
1776 ret = for_image(wim, image, image_find_streams_to_reference);
1784 struct insert_other_if_hard_filtered_ctx {
1785 struct stream_size_table *tab;
1786 struct filter_context *filter_ctx;
1790 insert_other_if_hard_filtered(struct wim_lookup_table_entry *lte, void *_ctx)
1792 struct insert_other_if_hard_filtered_ctx *ctx = _ctx;
1794 if (!lte->will_be_in_output_wim &&
1795 stream_hard_filtered(lte, ctx->filter_ctx))
1796 stream_size_table_insert(lte, ctx->tab);
1801 determine_stream_size_uniquity(struct list_head *stream_list,
1802 struct wim_lookup_table *lt,
1803 struct filter_context *filter_ctx)
1806 struct stream_size_table tab;
1807 struct wim_lookup_table_entry *lte;
1809 ret = init_stream_size_table(&tab, lt->capacity);
1813 if (may_hard_filter_streams(filter_ctx)) {
1814 struct insert_other_if_hard_filtered_ctx ctx = {
1816 .filter_ctx = filter_ctx,
1818 for_lookup_table_entry(lt, insert_other_if_hard_filtered, &ctx);
1821 list_for_each_entry(lte, stream_list, write_streams_list)
1822 stream_size_table_insert(lte, &tab);
1824 destroy_stream_size_table(&tab);
1829 filter_stream_list_for_write(struct list_head *stream_list,
1830 struct filter_context *filter_ctx)
1832 struct wim_lookup_table_entry *lte, *tmp;
1834 list_for_each_entry_safe(lte, tmp,
1835 stream_list, write_streams_list)
1837 int status = stream_filtered(lte, filter_ctx);
1844 /* Soft filtered. */
1846 /* Hard filtered. */
1847 lte->will_be_in_output_wim = 0;
1848 list_del(<e->lookup_table_list);
1850 list_del(<e->write_streams_list);
1856 * prepare_stream_list_for_write() -
1858 * Prepare the list of streams to write for writing a WIM containing the
1859 * specified image(s) with the specified write flags.
1862 * The WIMStruct on whose behalf the write is occurring.
1865 * Image(s) from the WIM to write; may be WIMLIB_ALL_IMAGES.
1868 * WIMLIB_WRITE_FLAG_* flags for the write operation:
1870 * STREAMS_OK: For writes of all images, assume that all streams in the
1871 * lookup table of @wim and the per-image lists of unhashed streams should
1872 * be taken as-is, and image metadata should not be searched for
1873 * references. This does not exclude filtering with OVERWRITE and
1874 * SKIP_EXTERNAL_WIMS, below.
1876 * OVERWRITE: Streams already present in @wim shall not be returned in
1879 * SKIP_EXTERNAL_WIMS: Streams already present in a WIM file, but not
1880 * @wim, shall be be returned in neither @stream_list_ret nor
1881 * @lookup_table_list_ret.
1884 * List of streams, linked by write_streams_list, that need to be written
1885 * will be returned here.
1887 * Note that this function assumes that unhashed streams will be written;
1888 * it does not take into account that they may become duplicates when
1891 * @lookup_table_list_ret
1892 * List of streams, linked by lookup_table_list, that need to be included
1893 * in the WIM's lookup table will be returned here. This will be a
1894 * superset of the streams in @stream_list_ret.
1896 * This list will be a proper superset of @stream_list_ret if and only if
1897 * WIMLIB_WRITE_FLAG_OVERWRITE was specified in @write_flags and some of
1898 * the streams that would otherwise need to be written were already located
1901 * All streams in this list will have @out_refcnt set to the number of
1902 * references to the stream in the output WIM. If
1903 * WIMLIB_WRITE_FLAG_STREAMS_OK was specified in @write_flags, @out_refcnt
1904 * may be as low as 0.
1907 * A context for queries of stream filter status with stream_filtered() is
1908 * returned in this location.
1910 * In addition, @will_be_in_output_wim will be set to 1 in all stream entries
1911 * inserted into @lookup_table_list_ret and to 0 in all stream entries in the
1912 * lookup table of @wim not inserted into @lookup_table_list_ret.
1914 * Still furthermore, @unique_size will be set to 1 on all stream entries in
1915 * @stream_list_ret that have unique size among all stream entries in
1916 * @stream_list_ret and among all stream entries in the lookup table of @wim
1917 * that are ineligible for being written due to filtering.
1919 * Returns 0 on success; nonzero on read error, memory allocation error, or
1923 prepare_stream_list_for_write(WIMStruct *wim, int image,
1925 struct list_head *stream_list_ret,
1926 struct list_head *lookup_table_list_ret,
1927 struct filter_context *filter_ctx_ret)
1930 struct wim_lookup_table_entry *lte;
1932 filter_ctx_ret->write_flags = write_flags;
1933 filter_ctx_ret->wim = wim;
1935 ret = prepare_unfiltered_list_of_streams_in_output_wim(
1938 write_flags & WIMLIB_WRITE_FLAG_STREAMS_OK,
1943 INIT_LIST_HEAD(lookup_table_list_ret);
1944 list_for_each_entry(lte, stream_list_ret, write_streams_list)
1945 list_add_tail(<e->lookup_table_list, lookup_table_list_ret);
1947 ret = determine_stream_size_uniquity(stream_list_ret, wim->lookup_table,
1952 if (may_filter_streams(filter_ctx_ret))
1953 filter_stream_list_for_write(stream_list_ret, filter_ctx_ret);
1959 write_wim_streams(WIMStruct *wim, int image, int write_flags,
1960 unsigned num_threads,
1961 wimlib_progress_func_t progress_func,
1962 struct list_head *stream_list_override,
1963 struct list_head *lookup_table_list_ret)
1966 struct list_head _stream_list;
1967 struct list_head *stream_list;
1968 struct wim_lookup_table_entry *lte;
1969 struct filter_context _filter_ctx;
1970 struct filter_context *filter_ctx;
1972 if (stream_list_override == NULL) {
1973 /* Normal case: prepare stream list from image(s) being written.
1975 stream_list = &_stream_list;
1976 filter_ctx = &_filter_ctx;
1977 ret = prepare_stream_list_for_write(wim, image, write_flags,
1979 lookup_table_list_ret,
1984 /* Currently only as a result of wimlib_split() being called:
1985 * use stream list already explicitly provided. Use existing
1986 * reference counts. */
1987 stream_list = stream_list_override;
1989 INIT_LIST_HEAD(lookup_table_list_ret);
1990 list_for_each_entry(lte, stream_list, write_streams_list) {
1991 lte->out_refcnt = lte->refcnt;
1992 lte->will_be_in_output_wim = 1;
1993 lte->unique_size = 0;
1994 list_add_tail(<e->lookup_table_list, lookup_table_list_ret);
1998 return wim_write_stream_list(wim,
2007 write_wim_metadata_resources(WIMStruct *wim, int image, int write_flags,
2008 wimlib_progress_func_t progress_func)
2013 int write_resource_flags;
2015 if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA) {
2016 DEBUG("Not writing any metadata resources.");
2020 write_resource_flags = write_flags_to_resource_flags(write_flags);
2022 write_resource_flags &= ~WRITE_RESOURCE_FLAG_PACK_STREAMS;
2024 DEBUG("Writing metadata resources (offset=%"PRIu64")",
2025 wim->out_fd.offset);
2028 progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN, NULL);
2030 if (image == WIMLIB_ALL_IMAGES) {
2032 end_image = wim->hdr.image_count;
2034 start_image = image;
2038 for (int i = start_image; i <= end_image; i++) {
2039 struct wim_image_metadata *imd;
2041 imd = wim->image_metadata[i - 1];
2042 /* Build a new metadata resource only if image was modified from
2043 * the original (or was newly added). Otherwise just copy the
2045 if (imd->modified) {
2046 DEBUG("Image %u was modified; building and writing new "
2047 "metadata resource", i);
2048 ret = write_metadata_resource(wim, i,
2049 write_resource_flags);
2050 } else if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) {
2051 DEBUG("Image %u was not modified; re-using existing "
2052 "metadata resource.", i);
2053 stream_set_out_reshdr_for_reuse(imd->metadata_lte);
2056 DEBUG("Image %u was not modified; copying existing "
2057 "metadata resource.", i);
2058 ret = write_wim_resource(imd->metadata_lte,
2060 wim->out_compression_type,
2061 wim->out_chunk_size,
2062 write_resource_flags);
2068 progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_END, NULL);
2073 open_wim_writable(WIMStruct *wim, const tchar *path, int open_flags)
2076 DEBUG("Opening \"%"TS"\" for writing.", path);
2078 raw_fd = topen(path, open_flags | O_BINARY, 0644);
2080 ERROR_WITH_ERRNO("Failed to open \"%"TS"\" for writing", path);
2081 return WIMLIB_ERR_OPEN;
2083 filedes_init(&wim->out_fd, raw_fd);
2088 close_wim_writable(WIMStruct *wim, int write_flags)
2092 if (!(write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)) {
2093 DEBUG("Closing WIM file.");
2094 if (filedes_valid(&wim->out_fd))
2095 if (filedes_close(&wim->out_fd))
2096 ret = WIMLIB_ERR_WRITE;
2098 filedes_invalidate(&wim->out_fd);
2103 cmp_streams_by_out_rspec(const void *p1, const void *p2)
2105 const struct wim_lookup_table_entry *lte1, *lte2;
2107 lte1 = *(const struct wim_lookup_table_entry**)p1;
2108 lte2 = *(const struct wim_lookup_table_entry**)p2;
2110 if (lte1->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
2111 if (lte2->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
2112 if (lte1->out_res_offset_in_wim != lte2->out_res_offset_in_wim)
2113 return cmp_u64(lte1->out_res_offset_in_wim,
2114 lte2->out_res_offset_in_wim);
2119 if (lte2->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS)
2122 return cmp_u64(lte1->out_reshdr.offset_in_wim,
2123 lte2->out_reshdr.offset_in_wim);
2127 write_wim_lookup_table(WIMStruct *wim, int image, int write_flags,
2128 struct wim_reshdr *out_reshdr,
2129 struct list_head *lookup_table_list)
2133 /* Set output resource metadata for streams already present in WIM. */
2134 if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) {
2135 struct wim_lookup_table_entry *lte;
2136 list_for_each_entry(lte, lookup_table_list, lookup_table_list)
2138 if (lte->resource_location == RESOURCE_IN_WIM &&
2139 lte->rspec->wim == wim)
2141 stream_set_out_reshdr_for_reuse(lte);
2146 ret = sort_stream_list(lookup_table_list,
2147 offsetof(struct wim_lookup_table_entry, lookup_table_list),
2148 cmp_streams_by_out_rspec);
2152 /* Add entries for metadata resources. */
2153 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)) {
2157 if (image == WIMLIB_ALL_IMAGES) {
2159 end_image = wim->hdr.image_count;
2161 start_image = image;
2165 /* Push metadata resource lookup table entries onto the front of
2166 * the list in reverse order, so that they're written in order.
2168 for (int i = end_image; i >= start_image; i--) {
2169 struct wim_lookup_table_entry *metadata_lte;
2171 metadata_lte = wim->image_metadata[i - 1]->metadata_lte;
2172 wimlib_assert(metadata_lte->out_reshdr.flags & WIM_RESHDR_FLAG_METADATA);
2173 metadata_lte->out_refcnt = 1;
2174 list_add(&metadata_lte->lookup_table_list, lookup_table_list);
2178 return write_wim_lookup_table_from_stream_list(lookup_table_list,
2180 wim->hdr.part_number,
2182 write_flags_to_resource_flags(write_flags));
2188 * Finish writing a WIM file: write the lookup table, xml data, and integrity
2189 * table, then overwrite the WIM header. By default, closes the WIM file
2190 * descriptor (@wim->out_fd) if successful.
2192 * write_flags is a bitwise OR of the following:
2194 * (public) WIMLIB_WRITE_FLAG_CHECK_INTEGRITY:
2195 * Include an integrity table.
2197 * (public) WIMLIB_WRITE_FLAG_FSYNC:
2198 * fsync() the output file before closing it.
2200 * (public) WIMLIB_WRITE_FLAG_PIPABLE:
2201 * Writing a pipable WIM, possibly to a pipe; include pipable WIM
2202 * stream headers before the lookup table and XML data, and also
2203 * write the WIM header at the end instead of seeking to the
2204 * beginning. Can't be combined with
2205 * WIMLIB_WRITE_FLAG_CHECK_INTEGRITY.
2207 * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE:
2208 * Don't write the lookup table.
2210 * (private) WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE:
2211 * When (if) writing the integrity table, re-use entries from the
2212 * existing integrity table, if possible.
2214 * (private) WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML:
2215 * After writing the XML data but before writing the integrity
2216 * table, write a temporary WIM header and flush the stream so that
2217 * the WIM is less likely to become corrupted upon abrupt program
2219 * (private) WIMLIB_WRITE_FLAG_HEADER_AT_END:
2220 * Instead of overwriting the WIM header at the beginning of the
2221 * file, simply append it to the end of the file. (Used when
2223 * (private) WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR:
2224 * Do not close the file descriptor @wim->out_fd on either success
2226 * (private) WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES:
2227 * Use the existing <TOTALBYTES> stored in the in-memory XML
2228 * information, rather than setting it to the offset of the XML
2229 * data being written.
2232 finish_write(WIMStruct *wim, int image, int write_flags,
2233 wimlib_progress_func_t progress_func,
2234 struct list_head *lookup_table_list)
2238 int write_resource_flags;
2239 off_t old_lookup_table_end;
2240 off_t new_lookup_table_end;
2243 DEBUG("image=%d, write_flags=%08x", image, write_flags);
2245 write_resource_flags = write_flags_to_resource_flags(write_flags);
2247 /* In the WIM header, there is room for the resource entry for a
2248 * metadata resource labeled as the "boot metadata". This entry should
2249 * be zeroed out if there is no bootable image (boot_idx 0). Otherwise,
2250 * it should be a copy of the resource entry for the image that is
2251 * marked as bootable. This is not well documented... */
2252 if (wim->hdr.boot_idx == 0) {
2253 zero_reshdr(&wim->hdr.boot_metadata_reshdr);
2255 copy_reshdr(&wim->hdr.boot_metadata_reshdr,
2256 &wim->image_metadata[
2257 wim->hdr.boot_idx - 1]->metadata_lte->out_reshdr);
2260 /* Write lookup table. (Save old position first.) */
2261 old_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim +
2262 wim->hdr.lookup_table_reshdr.size_in_wim;
2263 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
2264 ret = write_wim_lookup_table(wim, image, write_flags,
2265 &wim->hdr.lookup_table_reshdr,
2271 /* Write XML data. */
2272 xml_totalbytes = wim->out_fd.offset;
2273 if (write_flags & WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES)
2274 xml_totalbytes = WIM_TOTALBYTES_USE_EXISTING;
2275 ret = write_wim_xml_data(wim, image, xml_totalbytes,
2276 &wim->hdr.xml_data_reshdr,
2277 write_resource_flags);
2281 /* Write integrity table (optional). */
2282 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
2283 if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) {
2284 struct wim_header checkpoint_hdr;
2285 memcpy(&checkpoint_hdr, &wim->hdr, sizeof(struct wim_header));
2286 zero_reshdr(&checkpoint_hdr.integrity_table_reshdr);
2287 checkpoint_hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2288 ret = write_wim_header_at_offset(&checkpoint_hdr,
2294 if (!(write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE))
2295 old_lookup_table_end = 0;
2297 new_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim +
2298 wim->hdr.lookup_table_reshdr.size_in_wim;
2300 ret = write_integrity_table(wim,
2301 new_lookup_table_end,
2302 old_lookup_table_end,
2307 /* No integrity table. */
2308 zero_reshdr(&wim->hdr.integrity_table_reshdr);
2311 /* Now that all information in the WIM header has been determined, the
2312 * preliminary header written earlier can be overwritten, the header of
2313 * the existing WIM file can be overwritten, or the final header can be
2314 * written to the end of the pipable WIM. */
2315 wim->hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2317 if (write_flags & WIMLIB_WRITE_FLAG_HEADER_AT_END)
2318 hdr_offset = wim->out_fd.offset;
2319 DEBUG("Writing new header @ %"PRIu64".", hdr_offset);
2320 ret = write_wim_header_at_offset(&wim->hdr, &wim->out_fd, hdr_offset);
2324 /* Possibly sync file data to disk before closing. On POSIX systems, it
2325 * is necessary to do this before using rename() to overwrite an
2326 * existing file with a new file. Otherwise, data loss would occur if
2327 * the system is abruptly terminated when the metadata for the rename
2328 * operation has been written to disk, but the new file data has not.
2330 if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) {
2331 DEBUG("Syncing WIM file.");
2332 if (fsync(wim->out_fd.fd)) {
2333 ERROR_WITH_ERRNO("Error syncing data to WIM file");
2334 return WIMLIB_ERR_WRITE;
2338 if (close_wim_writable(wim, write_flags)) {
2339 ERROR_WITH_ERRNO("Failed to close the output WIM file");
2340 return WIMLIB_ERR_WRITE;
2346 #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK)
2348 lock_wim(WIMStruct *wim, int fd)
2351 if (fd != -1 && !wim->wim_locked) {
2352 ret = flock(fd, LOCK_EX | LOCK_NB);
2354 if (errno == EWOULDBLOCK) {
2355 ERROR("`%"TS"' is already being modified or has been "
2356 "mounted read-write\n"
2357 " by another process!", wim->filename);
2358 ret = WIMLIB_ERR_ALREADY_LOCKED;
2360 WARNING_WITH_ERRNO("Failed to lock `%"TS"'",
2365 wim->wim_locked = 1;
2373 * write_pipable_wim():
2375 * Perform the intermediate stages of creating a "pipable" WIM (i.e. a WIM
2376 * capable of being applied from a pipe).
2378 * Pipable WIMs are a wimlib-specific modification of the WIM format such that
2379 * images can be applied from them sequentially when the file data is sent over
2380 * a pipe. In addition, a pipable WIM can be written sequentially to a pipe.
2381 * The modifications made to the WIM format for pipable WIMs are:
2383 * - Magic characters in header are "WLPWM\0\0\0" (wimlib pipable WIM) instead
2384 * of "MSWIM\0\0\0". This lets wimlib know that the WIM is pipable and also
2385 * stops other software from trying to read the file as a normal WIM.
2387 * - The header at the beginning of the file does not contain all the normal
2388 * information; in particular it will have all 0's for the lookup table and
2389 * XML data resource entries. This is because this information cannot be
2390 * determined until the lookup table and XML data have been written.
2391 * Consequently, wimlib will write the full header at the very end of the
2392 * file. The header at the end, however, is only used when reading the WIM
2393 * from a seekable file (not a pipe).
2395 * - An extra copy of the XML data is placed directly after the header. This
2396 * allows image names and sizes to be determined at an appropriate time when
2397 * reading the WIM from a pipe. This copy of the XML data is ignored if the
2398 * WIM is read from a seekable file (not a pipe).
2400 * - The format of resources, or streams, has been modified to allow them to be
2401 * used before the "lookup table" has been read. Each stream is prefixed with
2402 * a `struct pwm_stream_hdr' that is basically an abbreviated form of `struct
2403 * wim_lookup_table_entry_disk' that only contains the SHA1 message digest,
2404 * uncompressed stream size, and flags that indicate whether the stream is
2405 * compressed. The data of uncompressed streams then follows literally, while
2406 * the data of compressed streams follows in a modified format. Compressed
2407 * streams do not begin with a chunk table, since the chunk table cannot be
2408 * written until all chunks have been compressed. Instead, each compressed
2409 * chunk is prefixed by a `struct pwm_chunk_hdr' that gives its size.
2410 * Furthermore, the chunk table is written at the end of the resource instead
2411 * of the start. Note: chunk offsets are given in the chunk table as if the
2412 * `struct pwm_chunk_hdr's were not present; also, the chunk table is only
2413 * used if the WIM is being read from a seekable file (not a pipe).
2415 * - Metadata resources always come before other file resources (streams).
2416 * (This does not by itself constitute an incompatibility with normal WIMs,
2417 * since this is valid in normal WIMs.)
2419 * - At least up to the end of the file resources, all components must be packed
2420 * as tightly as possible; there cannot be any "holes" in the WIM. (This does
2421 * not by itself consititute an incompatibility with normal WIMs, since this
2422 * is valid in normal WIMs.)
2424 * Note: the lookup table, XML data, and header at the end are not used when
2425 * applying from a pipe. They exist to support functionality such as image
2426 * application and export when the WIM is *not* read from a pipe.
2428 * Layout of pipable WIM:
2430 * ---------+----------+--------------------+----------------+--------------+-----------+--------+
2431 * | Header | XML data | Metadata resources | File resources | Lookup table | XML data | Header |
2432 * ---------+----------+--------------------+----------------+--------------+-----------+--------+
2434 * Layout of normal WIM:
2436 * +--------+-----------------------------+-------------------------+
2437 * | Header | File and metadata resources | Lookup table | XML data |
2438 * +--------+-----------------------------+-------------------------+
2440 * An optional integrity table can follow the final XML data in both normal and
2441 * pipable WIMs. However, due to implementation details, wimlib currently can
2442 * only include an integrity table in a pipable WIM when writing it to a
2443 * seekable file (not a pipe).
2445 * Do note that since pipable WIMs are not supported by Microsoft's software,
2446 * wimlib does not create them unless explicitly requested (with
2447 * WIMLIB_WRITE_FLAG_PIPABLE) and as stated above they use different magic
2448 * characters to identify the file.
2451 write_pipable_wim(WIMStruct *wim, int image, int write_flags,
2452 unsigned num_threads, wimlib_progress_func_t progress_func,
2453 struct list_head *stream_list_override,
2454 struct list_head *lookup_table_list_ret)
2457 struct wim_reshdr xml_reshdr;
2459 WARNING("Creating a pipable WIM, which will "
2461 " with Microsoft's software (wimgapi/imagex/Dism).");
2463 /* At this point, the header at the beginning of the file has already
2466 /* For efficiency, when wimlib adds an image to the WIM with
2467 * wimlib_add_image(), the SHA1 message digests of files is not
2468 * calculated; instead, they are calculated while the files are being
2469 * written. However, this does not work when writing a pipable WIM,
2470 * since when writing a stream to a pipable WIM, its SHA1 message digest
2471 * needs to be known before the stream data is written. Therefore,
2472 * before getting much farther, we need to pre-calculate the SHA1
2473 * message digests of all streams that will be written. */
2474 ret = wim_checksum_unhashed_streams(wim);
2478 /* Write extra copy of the XML data. */
2479 ret = write_wim_xml_data(wim, image, WIM_TOTALBYTES_OMIT,
2481 WRITE_RESOURCE_FLAG_PIPABLE);
2485 /* Write metadata resources for the image(s) being included in the
2487 ret = write_wim_metadata_resources(wim, image, write_flags,
2492 /* Write streams needed for the image(s) being included in the output
2493 * WIM, or streams needed for the split WIM part. */
2494 return write_wim_streams(wim, image, write_flags, num_threads,
2495 progress_func, stream_list_override,
2496 lookup_table_list_ret);
2498 /* The lookup table, XML data, and header at end are handled by
2499 * finish_write(). */
2502 /* Write a standalone WIM or split WIM (SWM) part to a new file or to a file
2505 write_wim_part(WIMStruct *wim,
2506 const void *path_or_fd,
2509 unsigned num_threads,
2510 wimlib_progress_func_t progress_func,
2511 unsigned part_number,
2512 unsigned total_parts,
2513 struct list_head *stream_list_override,
2517 struct wim_header hdr_save;
2518 struct list_head lookup_table_list;
2520 if (total_parts == 1)
2521 DEBUG("Writing standalone WIM.");
2523 DEBUG("Writing split WIM part %u/%u", part_number, total_parts);
2524 if (image == WIMLIB_ALL_IMAGES)
2525 DEBUG("Including all images.");
2527 DEBUG("Including image %d only.", image);
2528 if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)
2529 DEBUG("File descriptor: %d", *(const int*)path_or_fd);
2531 DEBUG("Path: \"%"TS"\"", (const tchar*)path_or_fd);
2532 DEBUG("Write flags: 0x%08x", write_flags);
2534 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)
2535 DEBUG("\tCHECK_INTEGRITY");
2537 if (write_flags & WIMLIB_WRITE_FLAG_REBUILD)
2540 if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
2541 DEBUG("\tRECOMPRESS");
2543 if (write_flags & WIMLIB_WRITE_FLAG_FSYNC)
2546 if (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE)
2549 if (write_flags & WIMLIB_WRITE_FLAG_IGNORE_READONLY_FLAG)
2550 DEBUG("\tIGNORE_READONLY_FLAG");
2552 if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
2555 if (write_flags & WIMLIB_WRITE_FLAG_NOT_PIPABLE)
2556 DEBUG("\tNOT_PIPABLE");
2558 if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS)
2559 DEBUG("\tPACK_STREAMS");
2561 if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)
2562 DEBUG("\tFILE_DESCRIPTOR");
2564 if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)
2565 DEBUG("\tNO_METADATA");
2567 if (write_flags & WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES)
2568 DEBUG("\tUSE_EXISTING_TOTALBYTES");
2570 if (num_threads == 0)
2571 DEBUG("Number of threads: autodetect");
2573 DEBUG("Number of threads: %u", num_threads);
2574 DEBUG("Progress function: %s", (progress_func ? "yes" : "no"));
2575 DEBUG("Stream list: %s", (stream_list_override ? "specified" : "autodetect"));
2576 DEBUG("GUID: %s", ((guid || wim->guid_set_explicitly) ?
2577 "specified" : "generate new"));
2579 /* Internally, this is always called with a valid part number and total
2581 wimlib_assert(total_parts >= 1);
2582 wimlib_assert(part_number >= 1 && part_number <= total_parts);
2584 /* A valid image (or all images) must be specified. */
2585 if (image != WIMLIB_ALL_IMAGES &&
2586 (image < 1 || image > wim->hdr.image_count))
2587 return WIMLIB_ERR_INVALID_IMAGE;
2589 /* If we need to write metadata resources, make sure the ::WIMStruct has
2590 * the needed information attached (e.g. is not a resource-only WIM,
2591 * such as a non-first part of a split WIM). */
2592 if (!wim_has_metadata(wim) &&
2593 !(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA))
2594 return WIMLIB_ERR_METADATA_NOT_FOUND;
2596 /* Check for contradictory flags. */
2597 if ((write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2598 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))
2599 == (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2600 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))
2601 return WIMLIB_ERR_INVALID_PARAM;
2603 if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2604 WIMLIB_WRITE_FLAG_NOT_PIPABLE))
2605 == (WIMLIB_WRITE_FLAG_PIPABLE |
2606 WIMLIB_WRITE_FLAG_NOT_PIPABLE))
2607 return WIMLIB_ERR_INVALID_PARAM;
2609 /* Save previous header, then start initializing the new one. */
2610 memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header));
2612 /* Set default integrity, pipable, and packed stream flags. */
2613 if (!(write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2614 WIMLIB_WRITE_FLAG_NOT_PIPABLE)))
2615 if (wim_is_pipable(wim)) {
2616 DEBUG("WIM is pipable; default to PIPABLE.");
2617 write_flags |= WIMLIB_WRITE_FLAG_PIPABLE;
2620 if (!(write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2621 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)))
2622 if (wim_has_integrity_table(wim)) {
2623 DEBUG("Integrity table present; default to CHECK_INTEGRITY.");
2624 write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY;
2627 if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2628 WIMLIB_WRITE_FLAG_PACK_STREAMS))
2629 == (WIMLIB_WRITE_FLAG_PIPABLE |
2630 WIMLIB_WRITE_FLAG_PACK_STREAMS))
2632 ERROR("Cannot specify both PIPABLE and PACK_STREAMS!");
2633 return WIMLIB_ERR_INVALID_PARAM;
2636 /* Set appropriate magic number. */
2637 if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
2638 wim->hdr.magic = PWM_MAGIC;
2640 wim->hdr.magic = WIM_MAGIC;
2642 /* Set appropriate version number. */
2643 if ((write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS) ||
2644 wim->out_compression_type == WIMLIB_COMPRESSION_TYPE_LZMS)
2645 wim->hdr.wim_version = WIM_VERSION_PACKED_STREAMS;
2647 wim->hdr.wim_version = WIM_VERSION_DEFAULT;
2649 /* Clear header flags that will be set automatically. */
2650 wim->hdr.flags &= ~(WIM_HDR_FLAG_METADATA_ONLY |
2651 WIM_HDR_FLAG_RESOURCE_ONLY |
2652 WIM_HDR_FLAG_SPANNED |
2653 WIM_HDR_FLAG_WRITE_IN_PROGRESS);
2655 /* Set SPANNED header flag if writing part of a split WIM. */
2656 if (total_parts != 1)
2657 wim->hdr.flags |= WIM_HDR_FLAG_SPANNED;
2659 /* Set part number and total parts of split WIM. This will be 1 and 1
2660 * if the WIM is standalone. */
2661 wim->hdr.part_number = part_number;
2662 wim->hdr.total_parts = total_parts;
2664 /* Set compression type if different. */
2665 if (wim->compression_type != wim->out_compression_type) {
2666 ret = set_wim_hdr_cflags(wim->out_compression_type, &wim->hdr);
2667 wimlib_assert(ret == 0);
2670 /* Set chunk size if different. */
2671 wim->hdr.chunk_size = wim->out_chunk_size;
2673 /* Use GUID if specified; otherwise generate a new one. */
2675 memcpy(wim->hdr.guid, guid, WIMLIB_GUID_LEN);
2676 else if (!wim->guid_set_explicitly)
2677 randomize_byte_array(wim->hdr.guid, WIMLIB_GUID_LEN);
2679 /* Clear references to resources that have not been written yet. */
2680 zero_reshdr(&wim->hdr.lookup_table_reshdr);
2681 zero_reshdr(&wim->hdr.xml_data_reshdr);
2682 zero_reshdr(&wim->hdr.boot_metadata_reshdr);
2683 zero_reshdr(&wim->hdr.integrity_table_reshdr);
2685 /* Set image count and boot index correctly for single image writes. */
2686 if (image != WIMLIB_ALL_IMAGES) {
2687 wim->hdr.image_count = 1;
2688 if (wim->hdr.boot_idx == image)
2689 wim->hdr.boot_idx = 1;
2691 wim->hdr.boot_idx = 0;
2694 /* Split WIMs can't be bootable. */
2695 if (total_parts != 1)
2696 wim->hdr.boot_idx = 0;
2698 /* Initialize output file descriptor. */
2699 if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR) {
2700 /* File descriptor was explicitly provided. Return error if
2701 * file descriptor is not seekable, unless writing a pipable WIM
2703 wim->out_fd.fd = *(const int*)path_or_fd;
2704 wim->out_fd.offset = 0;
2705 if (!filedes_is_seekable(&wim->out_fd)) {
2706 ret = WIMLIB_ERR_INVALID_PARAM;
2707 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
2708 goto out_restore_hdr;
2709 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
2710 ERROR("Can't include integrity check when "
2711 "writing pipable WIM to pipe!");
2712 goto out_restore_hdr;
2717 /* Filename of WIM to write was provided; open file descriptor
2719 ret = open_wim_writable(wim, (const tchar*)path_or_fd,
2720 O_TRUNC | O_CREAT | O_RDWR);
2722 goto out_restore_hdr;
2725 /* Write initial header. This is merely a "dummy" header since it
2726 * doesn't have all the information yet, so it will be overwritten later
2727 * (unless writing a pipable WIM). */
2728 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
2729 wim->hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2730 ret = write_wim_header(&wim->hdr, &wim->out_fd);
2731 wim->hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2733 goto out_restore_hdr;
2735 /* Write metadata resources and streams. */
2736 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE)) {
2737 /* Default case: create a normal (non-pipable) WIM. */
2738 ret = write_wim_streams(wim, image, write_flags, num_threads,
2739 progress_func, stream_list_override,
2740 &lookup_table_list);
2742 goto out_restore_hdr;
2744 ret = write_wim_metadata_resources(wim, image, write_flags,
2747 goto out_restore_hdr;
2749 /* Non-default case: create pipable WIM. */
2750 ret = write_pipable_wim(wim, image, write_flags, num_threads,
2751 progress_func, stream_list_override,
2752 &lookup_table_list);
2754 goto out_restore_hdr;
2755 write_flags |= WIMLIB_WRITE_FLAG_HEADER_AT_END;
2759 /* Write lookup table, XML data, and (optional) integrity table. */
2760 ret = finish_write(wim, image, write_flags, progress_func,
2761 &lookup_table_list);
2763 memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header));
2764 (void)close_wim_writable(wim, write_flags);
2765 DEBUG("ret=%d", ret);
2769 /* Write a standalone WIM to a file or file descriptor. */
2771 write_standalone_wim(WIMStruct *wim, const void *path_or_fd,
2772 int image, int write_flags, unsigned num_threads,
2773 wimlib_progress_func_t progress_func)
2775 return write_wim_part(wim, path_or_fd, image, write_flags,
2776 num_threads, progress_func, 1, 1, NULL, NULL);
2779 /* API function documented in wimlib.h */
2781 wimlib_write(WIMStruct *wim, const tchar *path,
2782 int image, int write_flags, unsigned num_threads,
2783 wimlib_progress_func_t progress_func)
2786 return WIMLIB_ERR_INVALID_PARAM;
2788 write_flags &= WIMLIB_WRITE_MASK_PUBLIC;
2790 return write_standalone_wim(wim, path, image, write_flags,
2791 num_threads, progress_func);
2794 /* API function documented in wimlib.h */
2796 wimlib_write_to_fd(WIMStruct *wim, int fd,
2797 int image, int write_flags, unsigned num_threads,
2798 wimlib_progress_func_t progress_func)
2801 return WIMLIB_ERR_INVALID_PARAM;
2803 write_flags &= WIMLIB_WRITE_MASK_PUBLIC;
2804 write_flags |= WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR;
2806 return write_standalone_wim(wim, &fd, image, write_flags,
2807 num_threads, progress_func);
2811 any_images_modified(WIMStruct *wim)
2813 for (int i = 0; i < wim->hdr.image_count; i++)
2814 if (wim->image_metadata[i]->modified)
2820 check_resource_offset(struct wim_lookup_table_entry *lte, void *_wim)
2822 const WIMStruct *wim = _wim;
2823 off_t end_offset = *(const off_t*)wim->private;
2825 if (lte->resource_location == RESOURCE_IN_WIM && lte->rspec->wim == wim &&
2826 lte->rspec->offset_in_wim + lte->rspec->size_in_wim > end_offset)
2827 return WIMLIB_ERR_RESOURCE_ORDER;
2831 /* Make sure no file or metadata resources are located after the XML data (or
2832 * integrity table if present)--- otherwise we can't safely overwrite the WIM in
2833 * place and we return WIMLIB_ERR_RESOURCE_ORDER. */
2835 check_resource_offsets(WIMStruct *wim, off_t end_offset)
2840 wim->private = &end_offset;
2841 ret = for_lookup_table_entry(wim->lookup_table, check_resource_offset, wim);
2845 for (i = 0; i < wim->hdr.image_count; i++) {
2846 ret = check_resource_offset(wim->image_metadata[i]->metadata_lte, wim);
2854 * Overwrite a WIM, possibly appending streams to it.
2856 * A WIM looks like (or is supposed to look like) the following:
2858 * Header (212 bytes)
2859 * Streams and metadata resources (variable size)
2860 * Lookup table (variable size)
2861 * XML data (variable size)
2862 * Integrity table (optional) (variable size)
2864 * If we are not adding any streams or metadata resources, the lookup table is
2865 * unchanged--- so we only need to overwrite the XML data, integrity table, and
2866 * header. This operation is potentially unsafe if the program is abruptly
2867 * terminated while the XML data or integrity table are being overwritten, but
2868 * before the new header has been written. To partially alleviate this problem,
2869 * a special flag (WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) is passed to
2870 * finish_write() to cause a temporary WIM header to be written after the XML
2871 * data has been written. This may prevent the WIM from becoming corrupted if
2872 * the program is terminated while the integrity table is being calculated (but
2873 * no guarantees, due to write re-ordering...).
2875 * If we are adding new streams or images (metadata resources), the lookup table
2876 * needs to be changed, and those streams need to be written. In this case, we
2877 * try to perform a safe update of the WIM file by writing the streams *after*
2878 * the end of the previous WIM, then writing the new lookup table, XML data, and
2879 * (optionally) integrity table following the new streams. This will produce a
2880 * layout like the following:
2882 * Header (212 bytes)
2883 * (OLD) Streams and metadata resources (variable size)
2884 * (OLD) Lookup table (variable size)
2885 * (OLD) XML data (variable size)
2886 * (OLD) Integrity table (optional) (variable size)
2887 * (NEW) Streams and metadata resources (variable size)
2888 * (NEW) Lookup table (variable size)
2889 * (NEW) XML data (variable size)
2890 * (NEW) Integrity table (optional) (variable size)
2892 * At all points, the WIM is valid as nothing points to the new data yet. Then,
2893 * the header is overwritten to point to the new lookup table, XML data, and
2894 * integrity table, to produce the following layout:
2896 * Header (212 bytes)
2897 * Streams and metadata resources (variable size)
2898 * Nothing (variable size)
2899 * More Streams and metadata resources (variable size)
2900 * Lookup table (variable size)
2901 * XML data (variable size)
2902 * Integrity table (optional) (variable size)
2904 * This method allows an image to be appended to a large WIM very quickly, and
2905 * is is crash-safe except in the case of write re-ordering, but the
2906 * disadvantage is that a small hole is left in the WIM where the old lookup
2907 * table, xml data, and integrity table were. (These usually only take up a
2908 * small amount of space compared to the streams, however.)
2911 overwrite_wim_inplace(WIMStruct *wim, int write_flags,
2912 unsigned num_threads,
2913 wimlib_progress_func_t progress_func)
2917 u64 old_lookup_table_end, old_xml_begin, old_xml_end;
2918 struct wim_header hdr_save;
2919 struct list_head stream_list;
2920 struct list_head lookup_table_list;
2921 struct filter_context filter_ctx;
2923 DEBUG("Overwriting `%"TS"' in-place", wim->filename);
2925 /* Save original header so it can be restored in case of error */
2926 memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header));
2928 /* Set default integrity flag. */
2929 if (!(write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2930 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)))
2931 if (wim_has_integrity_table(wim))
2932 write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY;
2934 /* Set WIM version if adding packed streams. */
2935 if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS)
2936 wim->hdr.wim_version = WIM_VERSION_PACKED_STREAMS;
2938 /* Set additional flags for overwrite. */
2939 write_flags |= WIMLIB_WRITE_FLAG_OVERWRITE |
2940 WIMLIB_WRITE_FLAG_STREAMS_OK;
2942 /* Make sure that the integrity table (if present) is after the XML
2943 * data, and that there are no stream resources, metadata resources, or
2944 * lookup tables after the XML data. Otherwise, these data would be
2946 old_xml_begin = wim->hdr.xml_data_reshdr.offset_in_wim;
2947 old_xml_end = old_xml_begin + wim->hdr.xml_data_reshdr.size_in_wim;
2948 old_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim +
2949 wim->hdr.lookup_table_reshdr.size_in_wim;
2950 if (wim->hdr.integrity_table_reshdr.offset_in_wim != 0 &&
2951 wim->hdr.integrity_table_reshdr.offset_in_wim < old_xml_end) {
2952 WARNING("Didn't expect the integrity table to be before the XML data");
2953 ret = WIMLIB_ERR_RESOURCE_ORDER;
2954 goto out_restore_memory_hdr;
2957 if (old_lookup_table_end > old_xml_begin) {
2958 WARNING("Didn't expect the lookup table to be after the XML data");
2959 ret = WIMLIB_ERR_RESOURCE_ORDER;
2960 goto out_restore_memory_hdr;
2963 /* Set @old_wim_end, which indicates the point beyond which we don't
2964 * allow any file and metadata resources to appear without returning
2965 * WIMLIB_ERR_RESOURCE_ORDER (due to the fact that we would otherwise
2966 * overwrite these resources). */
2967 if (!wim->deletion_occurred && !any_images_modified(wim)) {
2968 /* If no images have been modified and no images have been
2969 * deleted, a new lookup table does not need to be written. We
2970 * shall write the new XML data and optional integrity table
2971 * immediately after the lookup table. Note that this may
2972 * overwrite an existing integrity table. */
2973 DEBUG("Skipping writing lookup table "
2974 "(no images modified or deleted)");
2975 old_wim_end = old_lookup_table_end;
2976 write_flags |= WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE |
2977 WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML;
2978 } else if (wim->hdr.integrity_table_reshdr.offset_in_wim != 0) {
2979 /* Old WIM has an integrity table; begin writing new streams
2981 old_wim_end = wim->hdr.integrity_table_reshdr.offset_in_wim +
2982 wim->hdr.integrity_table_reshdr.size_in_wim;
2984 /* No existing integrity table; begin writing new streams after
2985 * the old XML data. */
2986 old_wim_end = old_xml_end;
2989 ret = check_resource_offsets(wim, old_wim_end);
2991 goto out_restore_memory_hdr;
2993 ret = prepare_stream_list_for_write(wim, WIMLIB_ALL_IMAGES, write_flags,
2994 &stream_list, &lookup_table_list,
2997 goto out_restore_memory_hdr;
2999 ret = open_wim_writable(wim, wim->filename, O_RDWR);
3001 goto out_restore_memory_hdr;
3003 ret = lock_wim(wim, wim->out_fd.fd);
3007 /* Set WIM_HDR_FLAG_WRITE_IN_PROGRESS flag in header. */
3008 wim->hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
3009 ret = write_wim_header_flags(wim->hdr.flags, &wim->out_fd);
3011 ERROR_WITH_ERRNO("Error updating WIM header flags");
3012 goto out_unlock_wim;
3015 if (filedes_seek(&wim->out_fd, old_wim_end) == -1) {
3016 ERROR_WITH_ERRNO("Can't seek to end of WIM");
3017 ret = WIMLIB_ERR_WRITE;
3018 goto out_restore_physical_hdr;
3021 ret = wim_write_stream_list(wim,
3030 ret = write_wim_metadata_resources(wim, WIMLIB_ALL_IMAGES,
3031 write_flags, progress_func);
3035 write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE;
3036 ret = finish_write(wim, WIMLIB_ALL_IMAGES, write_flags,
3037 progress_func, &lookup_table_list);
3041 wim->wim_locked = 0;
3045 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
3046 WARNING("Truncating `%"TS"' to its original size (%"PRIu64" bytes)",
3047 wim->filename, old_wim_end);
3048 /* Return value of ftruncate() is ignored because this is
3049 * already an error path. */
3050 (void)ftruncate(wim->out_fd.fd, old_wim_end);
3052 out_restore_physical_hdr:
3053 (void)write_wim_header_flags(hdr_save.flags, &wim->out_fd);
3055 wim->wim_locked = 0;
3057 (void)close_wim_writable(wim, write_flags);
3058 out_restore_memory_hdr:
3059 memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header));
3064 overwrite_wim_via_tmpfile(WIMStruct *wim, int write_flags,
3065 unsigned num_threads,
3066 wimlib_progress_func_t progress_func)
3068 size_t wim_name_len;
3071 DEBUG("Overwriting `%"TS"' via a temporary file", wim->filename);
3073 /* Write the WIM to a temporary file in the same directory as the
3075 wim_name_len = tstrlen(wim->filename);
3076 tchar tmpfile[wim_name_len + 10];
3077 tmemcpy(tmpfile, wim->filename, wim_name_len);
3078 randomize_char_array_with_alnum(tmpfile + wim_name_len, 9);
3079 tmpfile[wim_name_len + 9] = T('\0');
3081 ret = wimlib_write(wim, tmpfile, WIMLIB_ALL_IMAGES,
3082 write_flags | WIMLIB_WRITE_FLAG_FSYNC,
3083 num_threads, progress_func);
3091 /* Rename the new WIM file to the original WIM file. Note: on Windows
3092 * this actually calls win32_rename_replacement(), not _wrename(), so
3093 * that removing the existing destination file can be handled. */
3094 DEBUG("Renaming `%"TS"' to `%"TS"'", tmpfile, wim->filename);
3095 ret = trename(tmpfile, wim->filename);
3097 ERROR_WITH_ERRNO("Failed to rename `%"TS"' to `%"TS"'",
3098 tmpfile, wim->filename);
3105 return WIMLIB_ERR_RENAME;
3108 if (progress_func) {
3109 union wimlib_progress_info progress;
3110 progress.rename.from = tmpfile;
3111 progress.rename.to = wim->filename;
3112 progress_func(WIMLIB_PROGRESS_MSG_RENAME, &progress);
3117 /* Determine if the specified WIM file may be updated by appending in-place
3118 * rather than writing and replacing it with an entirely new file. */
3120 can_overwrite_wim_inplace(const WIMStruct *wim, int write_flags)
3122 /* REBUILD flag forces full rebuild. */
3123 if (write_flags & WIMLIB_WRITE_FLAG_REBUILD)
3126 /* Deletions cause full rebuild by default. */
3127 if (wim->deletion_occurred && !(write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE))
3130 /* Pipable WIMs cannot be updated in place, nor can a non-pipable WIM be
3131 * turned into a pipable WIM in-place. */
3132 if (wim_is_pipable(wim) || (write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
3135 /* wimlib allows multiple packs in a single WIM, but they don't seem to
3136 * be compatible with WIMGAPI, so force all streams to be repacked if
3137 * the WIM already may have contained a pack and PACK_STREAMS was
3139 if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS &&
3140 wim->hdr.wim_version == WIM_VERSION_PACKED_STREAMS)
3143 /* The default compression type and compression chunk size selected for
3144 * the output WIM must be the same as those currently used for the WIM.
3146 if (wim->compression_type != wim->out_compression_type)
3148 if (wim->chunk_size != wim->out_chunk_size)
3154 /* API function documented in wimlib.h */
3156 wimlib_overwrite(WIMStruct *wim, int write_flags,
3157 unsigned num_threads,
3158 wimlib_progress_func_t progress_func)
3163 write_flags &= WIMLIB_WRITE_MASK_PUBLIC;
3165 if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)
3166 return WIMLIB_ERR_INVALID_PARAM;
3169 return WIMLIB_ERR_NO_FILENAME;
3171 orig_hdr_flags = wim->hdr.flags;
3172 if (write_flags & WIMLIB_WRITE_FLAG_IGNORE_READONLY_FLAG)
3173 wim->hdr.flags &= ~WIM_HDR_FLAG_READONLY;
3174 ret = can_modify_wim(wim);
3175 wim->hdr.flags = orig_hdr_flags;
3179 if (can_overwrite_wim_inplace(wim, write_flags)) {
3180 ret = overwrite_wim_inplace(wim, write_flags, num_threads,
3182 if (ret != WIMLIB_ERR_RESOURCE_ORDER)
3184 WARNING("Falling back to re-building entire WIM");
3186 return overwrite_wim_via_tmpfile(wim, write_flags, num_threads,