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/inode.h"
43 #include "wimlib/integrity.h"
44 #include "wimlib/lookup_table.h"
45 #include "wimlib/metadata.h"
46 #include "wimlib/progress.h"
47 #include "wimlib/resource.h"
49 # include "wimlib/win32.h" /* win32_rename_replacement() */
51 #include "wimlib/write.h"
52 #include "wimlib/xml.h"
63 /* wimlib internal flags used when writing resources. */
64 #define WRITE_RESOURCE_FLAG_RECOMPRESS 0x00000001
65 #define WRITE_RESOURCE_FLAG_PIPABLE 0x00000002
66 #define WRITE_RESOURCE_FLAG_PACK_STREAMS 0x00000004
69 write_flags_to_resource_flags(int write_flags)
71 int write_resource_flags = 0;
73 if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
74 write_resource_flags |= WRITE_RESOURCE_FLAG_RECOMPRESS;
75 if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
76 write_resource_flags |= WRITE_RESOURCE_FLAG_PIPABLE;
77 if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS)
78 write_resource_flags |= WRITE_RESOURCE_FLAG_PACK_STREAMS;
79 return write_resource_flags;
82 struct filter_context {
87 /* Determine specified stream should be filtered out from the write.
91 * < 0 : The stream should be hard-filtered; that is, not included in the
93 * 0 : The stream should not be filtered out.
94 * > 0 : The stream should be soft-filtered; that is, it already exists in the
95 * WIM file and may not need to be written again.
98 stream_filtered(const struct wim_lookup_table_entry *lte,
99 const struct filter_context *ctx)
101 int write_flags = ctx->write_flags;
102 WIMStruct *wim = ctx->wim;
107 if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE &&
108 lte->resource_location == RESOURCE_IN_WIM &&
109 lte->rspec->wim == wim)
112 if (write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS &&
113 lte->resource_location == RESOURCE_IN_WIM &&
114 lte->rspec->wim != wim)
121 stream_hard_filtered(const struct wim_lookup_table_entry *lte,
122 struct filter_context *ctx)
124 return stream_filtered(lte, ctx) < 0;
128 may_soft_filter_streams(const struct filter_context *ctx)
132 return ctx->write_flags & WIMLIB_WRITE_FLAG_OVERWRITE;
136 may_hard_filter_streams(const struct filter_context *ctx)
140 return ctx->write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS;
144 may_filter_streams(const struct filter_context *ctx)
146 return (may_soft_filter_streams(ctx) ||
147 may_hard_filter_streams(ctx));
151 /* Return true if the specified resource is compressed and the compressed data
152 * can be reused with the specified output parameters. */
154 can_raw_copy(const struct wim_lookup_table_entry *lte,
155 int write_resource_flags, int out_ctype, u32 out_chunk_size)
157 const struct wim_resource_spec *rspec;
159 if (write_resource_flags & WRITE_RESOURCE_FLAG_RECOMPRESS)
162 if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE)
165 if (lte->resource_location != RESOURCE_IN_WIM)
170 if (rspec->is_pipable != !!(write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE))
173 if (rspec->flags & WIM_RESHDR_FLAG_COMPRESSED) {
174 /* Normal compressed resource: Must use same compression type
176 return (rspec->compression_type == out_ctype &&
177 rspec->chunk_size == out_chunk_size);
180 if ((rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) &&
181 (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
183 /* Packed resource: Such resources may contain multiple streams,
184 * and in general only a subset of them need to be written. As
185 * a heuristic, re-use the raw data if more than two-thirds the
186 * uncompressed size is being written. */
188 /* Note: packed resources contain a header that specifies the
189 * compression type and chunk size; therefore we don't need to
190 * check if they are compatible with @out_ctype and
191 * @out_chunk_size. */
193 struct wim_lookup_table_entry *res_stream;
196 list_for_each_entry(res_stream, &rspec->stream_list, rspec_node)
197 if (res_stream->will_be_in_output_wim)
198 write_size += res_stream->size;
200 return (write_size > rspec->uncompressed_size * 2 / 3);
207 filter_resource_flags(u8 flags)
209 return (flags & ~(WIM_RESHDR_FLAG_PACKED_STREAMS |
210 WIM_RESHDR_FLAG_COMPRESSED |
211 WIM_RESHDR_FLAG_SPANNED |
212 WIM_RESHDR_FLAG_FREE));
216 stream_set_out_reshdr_for_reuse(struct wim_lookup_table_entry *lte)
218 const struct wim_resource_spec *rspec;
220 wimlib_assert(lte->resource_location == RESOURCE_IN_WIM);
223 if (rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
225 wimlib_assert(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS);
227 lte->out_reshdr.offset_in_wim = lte->offset_in_res;
228 lte->out_reshdr.uncompressed_size = 0;
229 lte->out_reshdr.size_in_wim = lte->size;
231 lte->out_res_offset_in_wim = rspec->offset_in_wim;
232 lte->out_res_size_in_wim = rspec->size_in_wim;
233 lte->out_res_uncompressed_size = rspec->uncompressed_size;
235 wimlib_assert(!(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS));
237 lte->out_reshdr.offset_in_wim = rspec->offset_in_wim;
238 lte->out_reshdr.uncompressed_size = rspec->uncompressed_size;
239 lte->out_reshdr.size_in_wim = rspec->size_in_wim;
241 lte->out_reshdr.flags = lte->flags;
245 /* Write the header for a stream in a pipable WIM. */
247 write_pwm_stream_header(const struct wim_lookup_table_entry *lte,
248 struct filedes *out_fd,
249 int additional_reshdr_flags)
251 struct pwm_stream_hdr stream_hdr;
255 stream_hdr.magic = cpu_to_le64(PWM_STREAM_MAGIC);
256 stream_hdr.uncompressed_size = cpu_to_le64(lte->size);
257 if (additional_reshdr_flags & PWM_RESHDR_FLAG_UNHASHED) {
258 zero_out_hash(stream_hdr.hash);
260 wimlib_assert(!lte->unhashed);
261 copy_hash(stream_hdr.hash, lte->hash);
264 reshdr_flags = filter_resource_flags(lte->flags);
265 reshdr_flags |= additional_reshdr_flags;
266 stream_hdr.flags = cpu_to_le32(reshdr_flags);
267 ret = full_write(out_fd, &stream_hdr, sizeof(stream_hdr));
269 ERROR_WITH_ERRNO("Write error");
273 struct write_streams_progress_data {
274 wimlib_progress_func_t progfunc;
276 union wimlib_progress_info progress;
277 uint64_t next_progress;
281 do_write_streams_progress(struct write_streams_progress_data *progress_data,
282 struct wim_lookup_table_entry *cur_stream,
287 union wimlib_progress_info *progress = &progress_data->progress;
291 progress->write_streams.total_bytes -= complete_size;
292 progress->write_streams.total_streams -= complete_count;
293 if (progress_data->next_progress != ~(uint64_t)0 &&
294 progress_data->next_progress > progress->write_streams.total_bytes)
296 progress_data->next_progress = progress->write_streams.total_bytes;
299 progress->write_streams.completed_bytes += complete_size;
300 progress->write_streams.completed_streams += complete_count;
303 if (progress->write_streams.completed_bytes >= progress_data->next_progress)
305 ret = call_progress(progress_data->progfunc,
306 WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
308 progress_data->progctx);
312 if (progress_data->next_progress == progress->write_streams.total_bytes) {
313 progress_data->next_progress = ~(uint64_t)0;
315 progress_data->next_progress =
316 min(progress->write_streams.total_bytes,
317 progress->write_streams.completed_bytes +
318 progress->write_streams.total_bytes / 100);
324 struct write_streams_ctx {
325 /* File descriptor the streams are being written to. */
326 struct filedes *out_fd;
328 /* Lookup table for the WIMStruct on whose behalf the streams are being
330 struct wim_lookup_table *lookup_table;
332 /* Compression format to use. */
335 /* Maximum uncompressed chunk size in compressed resources to use. */
338 /* Flags that affect how the streams will be written. */
339 int write_resource_flags;
341 /* Data used for issuing WRITE_STREAMS progress. */
342 struct write_streams_progress_data progress_data;
344 struct filter_context *filter_ctx;
346 /* Upper bound on the total number of bytes that need to be compressed.
348 u64 num_bytes_to_compress;
350 /* Pointer to the chunk_compressor implementation being used for
351 * compressing chunks of data, or NULL if chunks are being written
353 struct chunk_compressor *compressor;
355 /* Buffer for dividing the read data into chunks of size
356 * @out_chunk_size. */
359 /* Number of bytes in @chunk_buf that are currently filled. */
360 size_t chunk_buf_filled;
362 /* List of streams that currently have chunks being compressed. */
363 struct list_head pending_streams;
365 /* List of streams in the resource pack. Streams are moved here after
366 * @pending_streams only when writing a packed resource. */
367 struct list_head pack_streams;
369 /* Set to true if the stream currently being read was a duplicate, and
370 * therefore the corresponding stream entry needs to be freed once the
371 * read finishes. (In this case we add the duplicate entry to
372 * pending_streams rather than the entry being read.) */
373 bool stream_was_duplicate;
375 /* Current uncompressed offset in the stream being read. */
376 u64 cur_read_stream_offset;
378 /* Uncompressed size of the stream currently being read. */
379 u64 cur_read_stream_size;
381 /* Current uncompressed offset in the stream being written. */
382 u64 cur_write_stream_offset;
384 /* Uncompressed size of resource currently being written. */
385 u64 cur_write_res_size;
387 /* Array that is filled in with compressed chunk sizes as a resource is
391 /* Index of next entry in @chunk_csizes to fill in. */
394 /* Number of entries in @chunk_csizes currently allocated. */
395 size_t num_alloc_chunks;
397 /* Offset in the output file of the start of the chunks of the resource
398 * currently being written. */
399 u64 chunks_start_offset;
402 /* Reserve space for the chunk table and prepare to accumulate the chunk table
405 begin_chunk_table(struct write_streams_ctx *ctx, u64 res_expected_size)
407 u64 expected_num_chunks;
408 u64 expected_num_chunk_entries;
412 /* Calculate the number of chunks and chunk entries that should be
413 * needed for the resource. These normally will be the final values,
414 * but in PACKED_STREAMS mode some of the streams we're planning to
415 * write into the resource may be duplicates, and therefore discarded,
416 * potentially decreasing the number of chunk entries needed. */
417 expected_num_chunks = DIV_ROUND_UP(res_expected_size, ctx->out_chunk_size);
418 expected_num_chunk_entries = expected_num_chunks;
419 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
420 expected_num_chunk_entries--;
422 /* Make sure the chunk_csizes array is long enough to store the
423 * compressed size of each chunk. */
424 if (expected_num_chunks > ctx->num_alloc_chunks) {
425 u64 new_length = expected_num_chunks + 50;
427 if ((size_t)new_length != new_length) {
428 ERROR("Resource size too large (%"PRIu64" bytes!",
430 return WIMLIB_ERR_NOMEM;
433 FREE(ctx->chunk_csizes);
434 ctx->chunk_csizes = MALLOC(new_length * sizeof(ctx->chunk_csizes[0]));
435 if (ctx->chunk_csizes == NULL) {
436 ctx->num_alloc_chunks = 0;
437 return WIMLIB_ERR_NOMEM;
439 ctx->num_alloc_chunks = new_length;
442 ctx->chunk_index = 0;
444 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE)) {
445 /* Reserve space for the chunk table in the output file. In the
446 * case of packed resources this reserves the upper bound for
447 * the needed space, not necessarily the exact space which will
448 * prove to be needed. At this point, we just use @chunk_csizes
449 * for a buffer of 0's because the actual compressed chunk sizes
451 reserve_size = expected_num_chunk_entries *
452 get_chunk_entry_size(res_expected_size,
453 0 != (ctx->write_resource_flags &
454 WRITE_RESOURCE_FLAG_PACK_STREAMS));
455 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)
456 reserve_size += sizeof(struct alt_chunk_table_header_disk);
457 memset(ctx->chunk_csizes, 0, reserve_size);
458 ret = full_write(ctx->out_fd, ctx->chunk_csizes, reserve_size);
466 begin_write_resource(struct write_streams_ctx *ctx, u64 res_expected_size)
470 wimlib_assert(res_expected_size != 0);
472 if (ctx->compressor != NULL) {
473 ret = begin_chunk_table(ctx, res_expected_size);
478 /* Output file descriptor is now positioned at the offset at which to
479 * write the first chunk of the resource. */
480 ctx->chunks_start_offset = ctx->out_fd->offset;
481 ctx->cur_write_stream_offset = 0;
482 ctx->cur_write_res_size = res_expected_size;
487 end_chunk_table(struct write_streams_ctx *ctx, u64 res_actual_size,
488 u64 *res_start_offset_ret, u64 *res_store_size_ret)
490 size_t actual_num_chunks;
491 size_t actual_num_chunk_entries;
492 size_t chunk_entry_size;
495 actual_num_chunks = ctx->chunk_index;
496 actual_num_chunk_entries = actual_num_chunks;
497 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
498 actual_num_chunk_entries--;
500 chunk_entry_size = get_chunk_entry_size(res_actual_size,
501 0 != (ctx->write_resource_flags &
502 WRITE_RESOURCE_FLAG_PACK_STREAMS));
504 typedef le64 __attribute__((may_alias)) aliased_le64_t;
505 typedef le32 __attribute__((may_alias)) aliased_le32_t;
507 if (chunk_entry_size == 4) {
508 aliased_le32_t *entries = (aliased_le32_t*)ctx->chunk_csizes;
510 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
511 for (size_t i = 0; i < actual_num_chunk_entries; i++)
512 entries[i] = cpu_to_le32(ctx->chunk_csizes[i]);
514 u32 offset = ctx->chunk_csizes[0];
515 for (size_t i = 0; i < actual_num_chunk_entries; i++) {
516 u32 next_size = ctx->chunk_csizes[i + 1];
517 entries[i] = cpu_to_le32(offset);
522 aliased_le64_t *entries = (aliased_le64_t*)ctx->chunk_csizes;
524 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
525 for (size_t i = 0; i < actual_num_chunk_entries; i++)
526 entries[i] = cpu_to_le64(ctx->chunk_csizes[i]);
528 u64 offset = ctx->chunk_csizes[0];
529 for (size_t i = 0; i < actual_num_chunk_entries; i++) {
530 u64 next_size = ctx->chunk_csizes[i + 1];
531 entries[i] = cpu_to_le64(offset);
537 size_t chunk_table_size = actual_num_chunk_entries * chunk_entry_size;
538 u64 res_start_offset;
541 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
542 ret = full_write(ctx->out_fd, ctx->chunk_csizes, chunk_table_size);
545 res_end_offset = ctx->out_fd->offset;
546 res_start_offset = ctx->chunks_start_offset;
548 res_end_offset = ctx->out_fd->offset;
550 u64 chunk_table_offset;
552 chunk_table_offset = ctx->chunks_start_offset - chunk_table_size;
554 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
555 struct alt_chunk_table_header_disk hdr;
557 hdr.res_usize = cpu_to_le64(res_actual_size);
558 hdr.chunk_size = cpu_to_le32(ctx->out_chunk_size);
559 hdr.compression_format = cpu_to_le32(ctx->out_ctype);
561 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 1);
562 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 2);
563 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3);
565 ret = full_pwrite(ctx->out_fd, &hdr, sizeof(hdr),
566 chunk_table_offset - sizeof(hdr));
569 res_start_offset = chunk_table_offset - sizeof(hdr);
571 res_start_offset = chunk_table_offset;
574 ret = full_pwrite(ctx->out_fd, ctx->chunk_csizes,
575 chunk_table_size, chunk_table_offset);
580 *res_start_offset_ret = res_start_offset;
581 *res_store_size_ret = res_end_offset - res_start_offset;
586 ERROR_WITH_ERRNO("Write error");
590 /* Finish writing a WIM resource by writing or updating the chunk table (if not
591 * writing the data uncompressed) and loading its metadata into @out_reshdr. */
593 end_write_resource(struct write_streams_ctx *ctx, struct wim_reshdr *out_reshdr)
597 u64 res_uncompressed_size;
598 u64 res_offset_in_wim;
600 wimlib_assert(ctx->cur_write_stream_offset == ctx->cur_write_res_size ||
601 (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS));
602 res_uncompressed_size = ctx->cur_write_res_size;
604 if (ctx->compressor) {
605 ret = end_chunk_table(ctx, res_uncompressed_size,
606 &res_offset_in_wim, &res_size_in_wim);
610 res_offset_in_wim = ctx->chunks_start_offset;
611 res_size_in_wim = ctx->out_fd->offset - res_offset_in_wim;
613 out_reshdr->uncompressed_size = res_uncompressed_size;
614 out_reshdr->size_in_wim = res_size_in_wim;
615 out_reshdr->offset_in_wim = res_offset_in_wim;
616 DEBUG("Finished writing resource: %"PRIu64" => %"PRIu64" @ %"PRIu64"",
617 res_uncompressed_size, res_size_in_wim, res_offset_in_wim);
621 /* Begin processing a stream for writing. */
623 write_stream_begin_read(struct wim_lookup_table_entry *lte,
624 u32 flags, void *_ctx)
626 struct write_streams_ctx *ctx = _ctx;
629 wimlib_assert(lte->size > 0);
631 ctx->cur_read_stream_offset = 0;
632 ctx->cur_read_stream_size = lte->size;
634 /* As an optimization, we allow some streams to be "unhashed", meaning
635 * their SHA1 message digests are unknown. This is the case with
636 * streams that are added by scanning a directry tree with
637 * wimlib_add_image(), for example. Since WIM uses single-instance
638 * streams, we don't know whether such each such stream really need to
639 * written until it is actually checksummed, unless it has a unique
640 * size. In such cases we read and checksum the stream in this
641 * function, thereby advancing ahead of read_stream_list(), which will
642 * still provide the data again to write_stream_process_chunk(). This
643 * is okay because an unhashed stream cannot be in a WIM resource, which
644 * might be costly to decompress. */
645 ctx->stream_was_duplicate = false;
646 if (ctx->lookup_table != NULL && lte->unhashed && !lte->unique_size) {
648 wimlib_assert(!(flags & BEGIN_STREAM_FLAG_PARTIAL_RESOURCE));
650 struct wim_lookup_table_entry *lte_new;
652 ret = hash_unhashed_stream(lte, ctx->lookup_table, <e_new);
655 if (lte_new != lte) {
656 /* Duplicate stream detected. */
658 if (lte_new->will_be_in_output_wim ||
659 stream_filtered(lte_new, ctx->filter_ctx))
661 /* The duplicate stream is already being
662 * included in the output WIM, or it would be
663 * filtered out if it had been. Skip writing
664 * this stream (and reading it again) entirely,
665 * passing its output reference count to the
666 * duplicate stream in the former case. */
667 DEBUG("Discarding duplicate stream of "
668 "length %"PRIu64, lte->size);
669 ret = do_write_streams_progress(&ctx->progress_data,
672 list_del(<e->write_streams_list);
673 list_del(<e->lookup_table_list);
674 if (lte_new->will_be_in_output_wim)
675 lte_new->out_refcnt += lte->out_refcnt;
676 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)
677 ctx->cur_write_res_size -= lte->size;
678 free_lookup_table_entry(lte);
681 return BEGIN_STREAM_STATUS_SKIP_STREAM;
683 /* The duplicate stream can validly be written,
684 * but was not marked as such. Discard the
685 * current stream entry and use the duplicate,
686 * but actually freeing the current entry must
687 * wait until read_stream_list() has finished
688 * reading its data. */
689 DEBUG("Stream duplicate, but not already "
690 "selected for writing.");
691 list_replace(<e->write_streams_list,
692 <e_new->write_streams_list);
693 list_replace(<e->lookup_table_list,
694 <e_new->lookup_table_list);
695 lte_new->out_refcnt = lte->out_refcnt;
696 lte_new->will_be_in_output_wim = 1;
697 ctx->stream_was_duplicate = true;
702 list_move_tail(<e->write_streams_list, &ctx->pending_streams);
706 /* Rewrite a stream that was just written compressed as uncompressed instead.
707 * This function is optional, but if a stream did not compress to less than its
708 * original size, it might as well be written uncompressed. */
710 write_stream_uncompressed(struct wim_lookup_table_entry *lte,
711 struct filedes *out_fd)
714 u64 begin_offset = lte->out_reshdr.offset_in_wim;
715 u64 end_offset = out_fd->offset;
717 if (filedes_seek(out_fd, begin_offset) == -1)
720 ret = extract_full_stream_to_fd(lte, out_fd);
722 /* Error reading the uncompressed data. */
723 if (out_fd->offset == begin_offset &&
724 filedes_seek(out_fd, end_offset) != -1)
726 /* Nothing was actually written yet, and we successfully
727 * seeked to the end of the compressed resource, so
728 * don't issue a hard error; just keep the compressed
729 * resource instead. */
730 WARNING("Recovered compressed stream of "
731 "size %"PRIu64", continuing on.",
738 wimlib_assert(out_fd->offset - begin_offset == lte->size);
740 if (out_fd->offset < end_offset &&
741 0 != ftruncate(out_fd->fd, out_fd->offset))
743 ERROR_WITH_ERRNO("Can't truncate output file to "
744 "offset %"PRIu64, out_fd->offset);
745 return WIMLIB_ERR_WRITE;
748 lte->out_reshdr.size_in_wim = lte->size;
749 lte->out_reshdr.flags &= ~(WIM_RESHDR_FLAG_COMPRESSED |
750 WIM_RESHDR_FLAG_PACKED_STREAMS);
754 /* Write the next chunk of (typically compressed) data to the output WIM,
755 * handling the writing of the chunk table. */
757 write_chunk(struct write_streams_ctx *ctx, const void *cchunk,
758 size_t csize, size_t usize)
762 struct wim_lookup_table_entry *lte;
763 u32 completed_stream_count;
766 lte = list_entry(ctx->pending_streams.next,
767 struct wim_lookup_table_entry, write_streams_list);
769 if (ctx->cur_write_stream_offset == 0 &&
770 !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
772 /* Starting to write a new stream in non-packed mode. */
774 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
775 int additional_reshdr_flags = 0;
776 if (ctx->compressor != NULL)
777 additional_reshdr_flags |= WIM_RESHDR_FLAG_COMPRESSED;
779 DEBUG("Writing pipable WIM stream header "
780 "(offset=%"PRIu64")", ctx->out_fd->offset);
782 ret = write_pwm_stream_header(lte, ctx->out_fd,
783 additional_reshdr_flags);
788 ret = begin_write_resource(ctx, lte->size);
793 if (ctx->compressor != NULL) {
794 /* Record the compresed chunk size. */
795 wimlib_assert(ctx->chunk_index < ctx->num_alloc_chunks);
796 ctx->chunk_csizes[ctx->chunk_index++] = csize;
798 /* If writing a pipable WIM, before the chunk data write a chunk
799 * header that provides the compressed chunk size. */
800 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
801 struct pwm_chunk_hdr chunk_hdr = {
802 .compressed_size = cpu_to_le32(csize),
804 ret = full_write(ctx->out_fd, &chunk_hdr,
811 /* Write the chunk data. */
812 ret = full_write(ctx->out_fd, cchunk, csize);
816 ctx->cur_write_stream_offset += usize;
818 completed_size = usize;
819 completed_stream_count = 0;
820 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
821 /* Wrote chunk in packed mode. It may have finished multiple
823 while (ctx->cur_write_stream_offset > lte->size) {
824 struct wim_lookup_table_entry *next;
826 ctx->cur_write_stream_offset -= lte->size;
828 wimlib_assert(!list_is_singular(&ctx->pending_streams) &&
829 !list_empty(&ctx->pending_streams));
830 next = list_entry(lte->write_streams_list.next,
831 struct wim_lookup_table_entry,
833 list_move_tail(<e->write_streams_list,
836 completed_stream_count++;
838 if (ctx->cur_write_stream_offset == lte->size) {
839 ctx->cur_write_stream_offset = 0;
840 list_move_tail(<e->write_streams_list,
842 completed_stream_count++;
845 /* Wrote chunk in non-packed mode. It may have finished a
847 if (ctx->cur_write_stream_offset == lte->size) {
849 completed_stream_count++;
851 list_del(<e->write_streams_list);
853 wimlib_assert(ctx->cur_write_stream_offset ==
854 ctx->cur_write_res_size);
856 ret = end_write_resource(ctx, <e->out_reshdr);
860 lte->out_reshdr.flags = filter_resource_flags(lte->flags);
861 if (ctx->compressor != NULL)
862 lte->out_reshdr.flags |= WIM_RESHDR_FLAG_COMPRESSED;
864 if (ctx->compressor != NULL &&
865 lte->out_reshdr.size_in_wim >= lte->out_reshdr.uncompressed_size &&
866 !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) &&
867 !(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS))
869 /* Stream did not compress to less than its original
870 * size. If we're not writing a pipable WIM (which
871 * could mean the output file descriptor is
872 * non-seekable), and the stream isn't located in a
873 * resource pack (which would make reading it again
874 * costly), truncate the file to the start of the stream
875 * and write it uncompressed instead. */
876 DEBUG("Stream of size %"PRIu64" did not compress to "
877 "less than original size; writing uncompressed.",
879 ret = write_stream_uncompressed(lte, ctx->out_fd);
883 wimlib_assert(lte->out_reshdr.uncompressed_size == lte->size);
885 ctx->cur_write_stream_offset = 0;
889 return do_write_streams_progress(&ctx->progress_data, lte,
890 completed_size, completed_stream_count,
894 ERROR_WITH_ERRNO("Write error");
899 submit_chunk_for_compression(struct write_streams_ctx *ctx,
900 const void *chunk, size_t size)
902 /* While we are unable to submit the chunk for compression (due to too
903 * many chunks already outstanding), retrieve and write the next
904 * compressed chunk. */
905 while (!ctx->compressor->submit_chunk(ctx->compressor, chunk, size)) {
912 bret = ctx->compressor->get_chunk(ctx->compressor,
913 &cchunk, &csize, &usize);
917 ret = write_chunk(ctx, cchunk, csize, usize);
924 /* Process the next chunk of data to be written to a WIM resource. */
926 write_stream_process_chunk(const void *chunk, size_t size, void *_ctx)
928 struct write_streams_ctx *ctx = _ctx;
930 const u8 *chunkptr, *chunkend;
932 wimlib_assert(size != 0);
934 if (ctx->compressor == NULL) {
935 /* Write chunk uncompressed. */
936 ret = write_chunk(ctx, chunk, size, size);
939 ctx->cur_read_stream_offset += size;
943 /* Submit the chunk for compression, but take into account that the
944 * @size the chunk was provided in may not correspond to the
945 * @out_chunk_size being used for compression. */
947 chunkend = chunkptr + size;
949 const u8 *resized_chunk;
950 size_t needed_chunk_size;
952 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
953 needed_chunk_size = ctx->out_chunk_size;
955 u64 res_bytes_remaining;
957 res_bytes_remaining = ctx->cur_read_stream_size -
958 ctx->cur_read_stream_offset;
959 needed_chunk_size = min(ctx->out_chunk_size,
960 ctx->chunk_buf_filled +
961 res_bytes_remaining);
964 if (ctx->chunk_buf_filled == 0 &&
965 chunkend - chunkptr >= needed_chunk_size)
967 /* No intermediate buffering needed. */
968 resized_chunk = chunkptr;
969 chunkptr += needed_chunk_size;
970 ctx->cur_read_stream_offset += needed_chunk_size;
972 /* Intermediate buffering needed. */
973 size_t bytes_consumed;
975 bytes_consumed = min(chunkend - chunkptr,
976 needed_chunk_size - ctx->chunk_buf_filled);
978 memcpy(&ctx->chunk_buf[ctx->chunk_buf_filled],
979 chunkptr, bytes_consumed);
981 chunkptr += bytes_consumed;
982 ctx->cur_read_stream_offset += bytes_consumed;
983 ctx->chunk_buf_filled += bytes_consumed;
984 if (ctx->chunk_buf_filled == needed_chunk_size) {
985 resized_chunk = ctx->chunk_buf;
986 ctx->chunk_buf_filled = 0;
993 ret = submit_chunk_for_compression(ctx, resized_chunk,
998 } while (chunkptr != chunkend);
1002 /* Finish processing a stream for writing. It may not have been completely
1003 * written yet, as the chunk_compressor implementation may still have chunks
1004 * buffered or being compressed. */
1006 write_stream_end_read(struct wim_lookup_table_entry *lte, int status, void *_ctx)
1008 struct write_streams_ctx *ctx = _ctx;
1010 wimlib_assert(ctx->cur_read_stream_offset == ctx->cur_read_stream_size);
1011 if (ctx->stream_was_duplicate) {
1012 free_lookup_table_entry(lte);
1013 } else if (lte->unhashed && ctx->lookup_table != NULL) {
1014 list_del(<e->unhashed_list);
1015 lookup_table_insert(ctx->lookup_table, lte);
1021 /* Compute statistics about a list of streams that will be written.
1023 * Assumes the streams are sorted such that all streams located in each distinct
1024 * WIM (specified by WIMStruct) are together. */
1026 compute_stream_list_stats(struct list_head *stream_list,
1027 struct write_streams_ctx *ctx)
1029 struct wim_lookup_table_entry *lte;
1030 u64 total_bytes = 0;
1031 u64 num_streams = 0;
1032 u64 total_parts = 0;
1033 WIMStruct *prev_wim_part = NULL;
1035 list_for_each_entry(lte, stream_list, write_streams_list) {
1037 total_bytes += lte->size;
1038 if (lte->resource_location == RESOURCE_IN_WIM) {
1039 if (prev_wim_part != lte->rspec->wim) {
1040 prev_wim_part = lte->rspec->wim;
1045 ctx->progress_data.progress.write_streams.total_bytes = total_bytes;
1046 ctx->progress_data.progress.write_streams.total_streams = num_streams;
1047 ctx->progress_data.progress.write_streams.completed_bytes = 0;
1048 ctx->progress_data.progress.write_streams.completed_streams = 0;
1049 ctx->progress_data.progress.write_streams.compression_type = ctx->out_ctype;
1050 ctx->progress_data.progress.write_streams.total_parts = total_parts;
1051 ctx->progress_data.progress.write_streams.completed_parts = 0;
1052 ctx->progress_data.next_progress = 0;
1055 /* Find streams in @stream_list that can be copied to the output WIM in raw form
1056 * rather than compressed. Delete these streams from @stream_list and move them
1057 * to @raw_copy_streams. Return the total uncompressed size of the streams that
1058 * need to be compressed. */
1060 find_raw_copy_streams(struct list_head *stream_list,
1061 int write_resource_flags,
1064 struct list_head *raw_copy_streams)
1066 struct wim_lookup_table_entry *lte, *tmp;
1067 u64 num_bytes_to_compress = 0;
1069 INIT_LIST_HEAD(raw_copy_streams);
1071 /* Initialize temporary raw_copy_ok flag. */
1072 list_for_each_entry(lte, stream_list, write_streams_list)
1073 if (lte->resource_location == RESOURCE_IN_WIM)
1074 lte->rspec->raw_copy_ok = 0;
1076 list_for_each_entry_safe(lte, tmp, stream_list, write_streams_list) {
1077 if (lte->resource_location == RESOURCE_IN_WIM &&
1078 lte->rspec->raw_copy_ok)
1080 list_move_tail(<e->write_streams_list,
1082 } else if (can_raw_copy(lte, write_resource_flags,
1083 out_ctype, out_chunk_size))
1085 lte->rspec->raw_copy_ok = 1;
1086 list_move_tail(<e->write_streams_list,
1089 num_bytes_to_compress += lte->size;
1093 return num_bytes_to_compress;
1096 /* Copy a raw compressed resource located in another WIM file to the WIM file
1099 write_raw_copy_resource(struct wim_resource_spec *in_rspec,
1100 struct filedes *out_fd)
1102 u64 cur_read_offset;
1103 u64 end_read_offset;
1104 u8 buf[BUFFER_SIZE];
1105 size_t bytes_to_read;
1107 struct filedes *in_fd;
1108 struct wim_lookup_table_entry *lte;
1109 u64 out_offset_in_wim;
1111 DEBUG("Copying raw compressed data (size_in_wim=%"PRIu64", "
1112 "uncompressed_size=%"PRIu64")",
1113 in_rspec->size_in_wim, in_rspec->uncompressed_size);
1115 /* Copy the raw data. */
1116 cur_read_offset = in_rspec->offset_in_wim;
1117 end_read_offset = cur_read_offset + in_rspec->size_in_wim;
1119 out_offset_in_wim = out_fd->offset;
1121 if (in_rspec->is_pipable) {
1122 if (cur_read_offset < sizeof(struct pwm_stream_hdr))
1123 return WIMLIB_ERR_INVALID_PIPABLE_WIM;
1124 cur_read_offset -= sizeof(struct pwm_stream_hdr);
1125 out_offset_in_wim += sizeof(struct pwm_stream_hdr);
1127 in_fd = &in_rspec->wim->in_fd;
1128 wimlib_assert(cur_read_offset != end_read_offset);
1131 bytes_to_read = min(sizeof(buf), end_read_offset - cur_read_offset);
1133 ret = full_pread(in_fd, buf, bytes_to_read, cur_read_offset);
1137 ret = full_write(out_fd, buf, bytes_to_read);
1141 cur_read_offset += bytes_to_read;
1143 } while (cur_read_offset != end_read_offset);
1145 list_for_each_entry(lte, &in_rspec->stream_list, rspec_node) {
1146 if (lte->will_be_in_output_wim) {
1147 stream_set_out_reshdr_for_reuse(lte);
1148 if (in_rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS)
1149 lte->out_res_offset_in_wim = out_offset_in_wim;
1151 lte->out_reshdr.offset_in_wim = out_offset_in_wim;
1158 /* Copy a list of raw compressed resources located in other WIM file(s) to the
1159 * WIM file being written. */
1161 write_raw_copy_resources(struct list_head *raw_copy_streams,
1162 struct filedes *out_fd,
1163 struct write_streams_progress_data *progress_data)
1165 struct wim_lookup_table_entry *lte;
1168 list_for_each_entry(lte, raw_copy_streams, write_streams_list)
1169 lte->rspec->raw_copy_ok = 1;
1171 list_for_each_entry(lte, raw_copy_streams, write_streams_list) {
1172 if (lte->rspec->raw_copy_ok) {
1173 /* Write each packed resource only one time, no matter
1174 * how many streams reference it. */
1175 ret = write_raw_copy_resource(lte->rspec, out_fd);
1178 lte->rspec->raw_copy_ok = 0;
1180 ret = do_write_streams_progress(progress_data, lte, lte->size,
1188 /* Wait for and write all chunks pending in the compressor. */
1190 finish_remaining_chunks(struct write_streams_ctx *ctx)
1197 if (ctx->compressor == NULL)
1200 if (ctx->chunk_buf_filled != 0) {
1201 ret = submit_chunk_for_compression(ctx, ctx->chunk_buf,
1202 ctx->chunk_buf_filled);
1207 while (ctx->compressor->get_chunk(ctx->compressor, &cdata, &csize, &usize)) {
1208 ret = write_chunk(ctx, cdata, csize, usize);
1216 remove_zero_length_streams(struct list_head *stream_list)
1218 struct wim_lookup_table_entry *lte, *tmp;
1220 list_for_each_entry_safe(lte, tmp, stream_list, write_streams_list) {
1221 wimlib_assert(lte->will_be_in_output_wim);
1222 if (lte->size == 0) {
1223 list_del(<e->write_streams_list);
1224 lte->out_reshdr.offset_in_wim = 0;
1225 lte->out_reshdr.size_in_wim = 0;
1226 lte->out_reshdr.uncompressed_size = 0;
1227 lte->out_reshdr.flags = filter_resource_flags(lte->flags);
1233 * Write a list of streams to the output WIM file.
1236 * The list of streams to write, specified by a list of `struct
1237 * wim_lookup_table_entry's linked by the 'write_streams_list' member.
1240 * The file descriptor, opened for writing, to which to write the streams.
1242 * @write_resource_flags
1243 * Flags to modify how the streams are written:
1245 * WRITE_RESOURCE_FLAG_RECOMPRESS:
1246 * Force compression of all resources, even if they could otherwise
1247 * be re-used by copying the raw data, due to being located in a WIM
1248 * file with compatible compression parameters.
1250 * WRITE_RESOURCE_FLAG_PIPABLE:
1251 * Write the resources in the wimlib-specific pipable format, and
1252 * furthermore do so in such a way that no seeking backwards in
1253 * @out_fd will be performed (so it may be a pipe).
1255 * WRITE_RESOURCE_FLAG_PACK_STREAMS:
1256 * Pack all the streams into a single resource rather than writing
1257 * them in separate resources. This flag is only valid if the WIM
1258 * version number has been, or will be, set to
1259 * WIM_VERSION_PACKED_STREAMS. This flag may not be combined with
1260 * WRITE_RESOURCE_FLAG_PIPABLE.
1263 * Compression format to use to write the output streams, specified as one
1264 * of the WIMLIB_COMPRESSION_TYPE_* constants.
1265 * WIMLIB_COMPRESSION_TYPE_NONE is allowed.
1268 * Chunk size to use to write the streams. It must be a valid chunk size
1269 * for the specified compression format @out_ctype, unless @out_ctype is
1270 * WIMLIB_COMPRESSION_TYPE_NONE, in which case this parameter is ignored.
1273 * Number of threads to use to compress data. If 0, a default number of
1274 * threads will be chosen. The number of threads still may be decreased
1275 * from the specified value if insufficient memory is detected.
1278 * If on-the-fly deduplication of unhashed streams is desired, this
1279 * parameter must be pointer to the lookup table for the WIMStruct on whose
1280 * behalf the streams are being written. Otherwise, this parameter can be
1284 * If on-the-fly deduplication of unhashed streams is desired, this
1285 * parameter can be a pointer to a context for stream filtering used to
1286 * detect whether the duplicate stream has been hard-filtered or not. If
1287 * no streams are hard-filtered or no streams are unhashed, this parameter
1290 * This function will write the streams in @stream_list to resources in
1291 * consecutive positions in the output WIM file, or to a single packed resource
1292 * if WRITE_RESOURCE_FLAG_PACK_STREAMS was specified in @write_resource_flags.
1293 * In both cases, the @out_reshdr of the `struct wim_lookup_table_entry' for
1294 * each stream written will be updated to specify its location, size, and flags
1295 * in the output WIM. In the packed resource case,
1296 * WIM_RESHDR_FLAG_PACKED_STREAMS will be set in the @flags field of each
1297 * @out_reshdr, and furthermore @out_res_offset_in_wim and @out_res_size_in_wim
1298 * of each @out_reshdr will be set to the offset and size, respectively, in the
1299 * output WIM of the packed resource containing the corresponding stream.
1301 * Each of the streams to write may be in any location supported by the
1302 * resource-handling code (specifically, read_stream_list()), such as the
1303 * contents of external file that has been logically added to the output WIM, or
1304 * a stream in another WIM file that has been imported, or even a stream in the
1305 * "same" WIM file of which a modified copy is being written. In the case that
1306 * a stream is already in a WIM file and uses compatible compression parameters,
1307 * by default this function will re-use the raw data instead of decompressing
1308 * it, then recompressing it; however, with WRITE_RESOURCE_FLAG_RECOMPRESS
1309 * specified in @write_resource_flags, this is not done.
1311 * As a further requirement, this function requires that the
1312 * @will_be_in_output_wim member be set to 1 on all streams in @stream_list as
1313 * well as any other streams not in @stream_list that will be in the output WIM
1314 * file, but set to 0 on any other streams in the output WIM's lookup table or
1315 * sharing a packed resource with a stream in @stream_list. Still furthermore,
1316 * if on-the-fly deduplication of streams is possible, then all streams in
1317 * @stream_list must also be linked by @lookup_table_list along with any other
1318 * streams that have @will_be_in_output_wim set.
1320 * This function handles on-the-fly deduplication of streams for which SHA1
1321 * message digests have not yet been calculated. Such streams may or may not
1322 * need to be written. If @lookup_table is non-NULL, then each stream in
1323 * @stream_list that has @unhashed set but not @unique_size set is checksummed
1324 * immediately before it would otherwise be read for writing in order to
1325 * determine if it is identical to another stream already being written or one
1326 * that would be filtered out of the output WIM using stream_filtered() with the
1327 * context @filter_ctx. Each such duplicate stream will be removed from
1328 * @stream_list, its reference count transfered to the pre-existing duplicate
1329 * stream, its memory freed, and will not be written. Alternatively, if a
1330 * stream in @stream_list is a duplicate with any stream in @lookup_table that
1331 * has not been marked for writing or would not be hard-filtered, it is freed
1332 * and the pre-existing duplicate is written instead, taking ownership of the
1333 * reference count and slot in the @lookup_table_list.
1335 * Returns 0 if every stream was either written successfully or did not need to
1336 * be written; otherwise returns a non-zero error code.
1339 write_stream_list(struct list_head *stream_list,
1340 struct filedes *out_fd,
1341 int write_resource_flags,
1344 unsigned num_threads,
1345 struct wim_lookup_table *lookup_table,
1346 struct filter_context *filter_ctx,
1347 wimlib_progress_func_t progfunc,
1351 struct write_streams_ctx ctx;
1352 struct list_head raw_copy_streams;
1354 wimlib_assert((write_resource_flags &
1355 (WRITE_RESOURCE_FLAG_PACK_STREAMS |
1356 WRITE_RESOURCE_FLAG_PIPABLE)) !=
1357 (WRITE_RESOURCE_FLAG_PACK_STREAMS |
1358 WRITE_RESOURCE_FLAG_PIPABLE));
1360 remove_zero_length_streams(stream_list);
1362 if (list_empty(stream_list)) {
1363 DEBUG("No streams to write.");
1367 memset(&ctx, 0, sizeof(ctx));
1369 /* Pre-sorting the streams is required for compute_stream_list_stats().
1370 * Afterwards, read_stream_list() need not sort them again. */
1371 ret = sort_stream_list_by_sequential_order(stream_list,
1372 offsetof(struct wim_lookup_table_entry,
1373 write_streams_list));
1377 ctx.out_fd = out_fd;
1378 ctx.lookup_table = lookup_table;
1379 ctx.out_ctype = out_ctype;
1380 ctx.out_chunk_size = out_chunk_size;
1381 ctx.write_resource_flags = write_resource_flags;
1382 ctx.filter_ctx = filter_ctx;
1384 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) {
1385 wimlib_assert(out_chunk_size != 0);
1386 if (out_chunk_size <= STACK_MAX) {
1387 ctx.chunk_buf = alloca(out_chunk_size);
1389 ctx.chunk_buf = MALLOC(out_chunk_size);
1390 if (ctx.chunk_buf == NULL) {
1391 ret = WIMLIB_ERR_NOMEM;
1392 goto out_destroy_context;
1396 ctx.chunk_buf_filled = 0;
1398 compute_stream_list_stats(stream_list, &ctx);
1400 ctx.progress_data.progfunc = progfunc;
1401 ctx.progress_data.progctx = progctx;
1403 ctx.num_bytes_to_compress = find_raw_copy_streams(stream_list,
1404 write_resource_flags,
1409 DEBUG("Writing stream list "
1410 "(offset = %"PRIu64", write_resource_flags=0x%08x, "
1411 "out_ctype=%d, out_chunk_size=%u, num_threads=%u, "
1412 "total_bytes=%"PRIu64", num_bytes_to_compress=%"PRIu64")",
1413 out_fd->offset, write_resource_flags,
1414 out_ctype, out_chunk_size, num_threads,
1415 ctx.progress_data.progress.write_streams.total_bytes,
1416 ctx.num_bytes_to_compress);
1418 if (ctx.num_bytes_to_compress == 0) {
1419 DEBUG("No compression needed; skipping to raw copy!");
1420 goto out_write_raw_copy_resources;
1423 /* Unless uncompressed output was required, allocate a chunk_compressor
1424 * to do compression. There are serial and parallel implementations of
1425 * the chunk_compressor interface. We default to parallel using the
1426 * specified number of threads, unless the upper bound on the number
1427 * bytes needing to be compressed is less than a heuristic value. */
1428 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) {
1430 #ifdef ENABLE_MULTITHREADED_COMPRESSION
1431 if (ctx.num_bytes_to_compress > max(2000000, out_chunk_size)) {
1432 ret = new_parallel_chunk_compressor(out_ctype,
1437 DEBUG("Couldn't create parallel chunk compressor "
1438 "(status %d)", ret);
1443 if (ctx.compressor == NULL) {
1444 ret = new_serial_chunk_compressor(out_ctype, out_chunk_size,
1447 goto out_destroy_context;
1452 ctx.progress_data.progress.write_streams.num_threads = ctx.compressor->num_threads;
1454 ctx.progress_data.progress.write_streams.num_threads = 1;
1456 DEBUG("Actually using %u threads",
1457 ctx.progress_data.progress.write_streams.num_threads);
1459 INIT_LIST_HEAD(&ctx.pending_streams);
1460 INIT_LIST_HEAD(&ctx.pack_streams);
1462 ret = call_progress(ctx.progress_data.progfunc,
1463 WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
1464 &ctx.progress_data.progress,
1465 ctx.progress_data.progctx);
1467 goto out_destroy_context;
1469 if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
1470 ret = begin_write_resource(&ctx, ctx.num_bytes_to_compress);
1472 goto out_destroy_context;
1475 /* Read the list of streams needing to be compressed, using the
1476 * specified callbacks to execute processing of the data. */
1478 struct read_stream_list_callbacks cbs = {
1479 .begin_stream = write_stream_begin_read,
1480 .begin_stream_ctx = &ctx,
1481 .consume_chunk = write_stream_process_chunk,
1482 .consume_chunk_ctx = &ctx,
1483 .end_stream = write_stream_end_read,
1484 .end_stream_ctx = &ctx,
1487 ret = read_stream_list(stream_list,
1488 offsetof(struct wim_lookup_table_entry, write_streams_list),
1490 STREAM_LIST_ALREADY_SORTED |
1491 VERIFY_STREAM_HASHES |
1492 COMPUTE_MISSING_STREAM_HASHES);
1495 goto out_destroy_context;
1497 ret = finish_remaining_chunks(&ctx);
1499 goto out_destroy_context;
1501 if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
1502 struct wim_reshdr reshdr;
1503 struct wim_lookup_table_entry *lte;
1506 ret = end_write_resource(&ctx, &reshdr);
1508 goto out_destroy_context;
1510 DEBUG("Ending packed resource: %lu %lu %lu.",
1511 reshdr.offset_in_wim,
1513 reshdr.uncompressed_size);
1516 list_for_each_entry(lte, &ctx.pack_streams, write_streams_list) {
1517 lte->out_reshdr.size_in_wim = lte->size;
1518 lte->out_reshdr.flags = filter_resource_flags(lte->flags);
1519 lte->out_reshdr.flags |= WIM_RESHDR_FLAG_PACKED_STREAMS;
1520 lte->out_reshdr.uncompressed_size = 0;
1521 lte->out_reshdr.offset_in_wim = offset_in_res;
1522 lte->out_res_offset_in_wim = reshdr.offset_in_wim;
1523 lte->out_res_size_in_wim = reshdr.size_in_wim;
1524 lte->out_res_uncompressed_size = reshdr.uncompressed_size;
1525 offset_in_res += lte->size;
1527 wimlib_assert(offset_in_res == reshdr.uncompressed_size);
1530 out_write_raw_copy_resources:
1531 /* Copy any compressed resources for which the raw data can be reused
1532 * without decompression. */
1533 ret = write_raw_copy_resources(&raw_copy_streams, ctx.out_fd,
1534 &ctx.progress_data);
1536 out_destroy_context:
1537 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && out_chunk_size > STACK_MAX)
1538 FREE(ctx.chunk_buf);
1539 FREE(ctx.chunk_csizes);
1541 ctx.compressor->destroy(ctx.compressor);
1542 DEBUG("Done (ret=%d)", ret);
1547 is_stream_packed(struct wim_lookup_table_entry *lte, void *_ignore)
1549 return lte_is_partial(lte);
1553 wim_has_packed_streams(WIMStruct *wim)
1555 return for_lookup_table_entry(wim->lookup_table, is_stream_packed, NULL);
1559 wim_write_stream_list(WIMStruct *wim,
1560 struct list_head *stream_list,
1562 unsigned num_threads,
1563 struct filter_context *filter_ctx)
1567 int write_resource_flags;
1569 write_resource_flags = write_flags_to_resource_flags(write_flags);
1571 /* wimlib v1.7.0: pack streams by default if the WIM version has been
1572 * set to WIM_VERSION_PACKED_STREAMS and at least one stream in the
1573 * WIM's lookup table is located in a packed resource (may be the same
1574 * WIM, or a different one in the case of export). */
1575 if (wim->hdr.wim_version == WIM_VERSION_PACKED_STREAMS &&
1576 wim_has_packed_streams(wim))
1578 write_resource_flags |= WRITE_RESOURCE_FLAG_PACK_STREAMS;
1581 if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
1582 out_chunk_size = wim->out_pack_chunk_size;
1583 out_ctype = wim->out_pack_compression_type;
1585 out_chunk_size = wim->out_chunk_size;
1586 out_ctype = wim->out_compression_type;
1589 return write_stream_list(stream_list,
1591 write_resource_flags,
1602 write_wim_resource(struct wim_lookup_table_entry *lte,
1603 struct filedes *out_fd,
1606 int write_resource_flags)
1608 LIST_HEAD(stream_list);
1609 list_add(<e->write_streams_list, &stream_list);
1610 lte->will_be_in_output_wim = 1;
1611 return write_stream_list(&stream_list,
1613 write_resource_flags & ~WRITE_RESOURCE_FLAG_PACK_STREAMS,
1624 write_wim_resource_from_buffer(const void *buf, size_t buf_size,
1625 int reshdr_flags, struct filedes *out_fd,
1628 struct wim_reshdr *out_reshdr,
1630 int write_resource_flags)
1633 struct wim_lookup_table_entry *lte;
1635 /* Set up a temporary lookup table entry to provide to
1636 * write_wim_resource(). */
1638 lte = new_lookup_table_entry();
1640 return WIMLIB_ERR_NOMEM;
1642 lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER;
1643 lte->attached_buffer = (void*)buf;
1644 lte->size = buf_size;
1645 lte->flags = reshdr_flags;
1647 if (write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
1648 sha1_buffer(buf, buf_size, lte->hash);
1654 ret = write_wim_resource(lte, out_fd, out_ctype, out_chunk_size,
1655 write_resource_flags);
1659 copy_reshdr(out_reshdr, <e->out_reshdr);
1662 copy_hash(hash, lte->hash);
1665 lte->resource_location = RESOURCE_NONEXISTENT;
1666 free_lookup_table_entry(lte);
1670 struct stream_size_table {
1671 struct hlist_head *array;
1677 init_stream_size_table(struct stream_size_table *tab, size_t capacity)
1679 tab->array = CALLOC(capacity, sizeof(tab->array[0]));
1680 if (tab->array == NULL)
1681 return WIMLIB_ERR_NOMEM;
1682 tab->num_entries = 0;
1683 tab->capacity = capacity;
1688 destroy_stream_size_table(struct stream_size_table *tab)
1694 stream_size_table_insert(struct wim_lookup_table_entry *lte, void *_tab)
1696 struct stream_size_table *tab = _tab;
1698 struct wim_lookup_table_entry *same_size_lte;
1699 struct hlist_node *tmp;
1701 pos = hash_u64(lte->size) % tab->capacity;
1702 lte->unique_size = 1;
1703 hlist_for_each_entry(same_size_lte, tmp, &tab->array[pos], hash_list_2) {
1704 if (same_size_lte->size == lte->size) {
1705 lte->unique_size = 0;
1706 same_size_lte->unique_size = 0;
1711 hlist_add_head(<e->hash_list_2, &tab->array[pos]);
1716 struct find_streams_ctx {
1719 struct list_head stream_list;
1720 struct stream_size_table stream_size_tab;
1724 reference_stream_for_write(struct wim_lookup_table_entry *lte,
1725 struct list_head *stream_list, u32 nref)
1727 if (!lte->will_be_in_output_wim) {
1728 lte->out_refcnt = 0;
1729 list_add_tail(<e->write_streams_list, stream_list);
1730 lte->will_be_in_output_wim = 1;
1732 lte->out_refcnt += nref;
1736 fully_reference_stream_for_write(struct wim_lookup_table_entry *lte,
1739 struct list_head *stream_list = _stream_list;
1740 lte->will_be_in_output_wim = 0;
1741 reference_stream_for_write(lte, stream_list, lte->refcnt);
1746 inode_find_streams_to_reference(const struct wim_inode *inode,
1747 const struct wim_lookup_table *table,
1748 struct list_head *stream_list)
1750 struct wim_lookup_table_entry *lte;
1753 wimlib_assert(inode->i_nlink > 0);
1755 for (i = 0; i <= inode->i_num_ads; i++) {
1756 lte = inode_stream_lte(inode, i, table);
1758 reference_stream_for_write(lte, stream_list,
1760 else if (!is_zero_hash(inode_stream_hash(inode, i)))
1761 return WIMLIB_ERR_RESOURCE_NOT_FOUND;
1767 do_stream_set_not_in_output_wim(struct wim_lookup_table_entry *lte, void *_ignore)
1769 lte->will_be_in_output_wim = 0;
1774 image_find_streams_to_reference(WIMStruct *wim)
1776 struct wim_image_metadata *imd;
1777 struct wim_inode *inode;
1778 struct wim_lookup_table_entry *lte;
1779 struct list_head *stream_list;
1782 imd = wim_get_current_image_metadata(wim);
1784 image_for_each_unhashed_stream(lte, imd)
1785 lte->will_be_in_output_wim = 0;
1787 stream_list = wim->private;
1788 image_for_each_inode(inode, imd) {
1789 ret = inode_find_streams_to_reference(inode,
1799 prepare_unfiltered_list_of_streams_in_output_wim(WIMStruct *wim,
1802 struct list_head *stream_list_ret)
1806 INIT_LIST_HEAD(stream_list_ret);
1808 if (streams_ok && (image == WIMLIB_ALL_IMAGES ||
1809 (image == 1 && wim->hdr.image_count == 1)))
1811 /* Fast case: Assume that all streams are being written and
1812 * that the reference counts are correct. */
1813 struct wim_lookup_table_entry *lte;
1814 struct wim_image_metadata *imd;
1817 for_lookup_table_entry(wim->lookup_table,
1818 fully_reference_stream_for_write,
1821 for (i = 0; i < wim->hdr.image_count; i++) {
1822 imd = wim->image_metadata[i];
1823 image_for_each_unhashed_stream(lte, imd)
1824 fully_reference_stream_for_write(lte, stream_list_ret);
1827 /* Slow case: Walk through the images being written and
1828 * determine the streams referenced. */
1829 for_lookup_table_entry(wim->lookup_table,
1830 do_stream_set_not_in_output_wim, NULL);
1831 wim->private = stream_list_ret;
1832 ret = for_image(wim, image, image_find_streams_to_reference);
1840 struct insert_other_if_hard_filtered_ctx {
1841 struct stream_size_table *tab;
1842 struct filter_context *filter_ctx;
1846 insert_other_if_hard_filtered(struct wim_lookup_table_entry *lte, void *_ctx)
1848 struct insert_other_if_hard_filtered_ctx *ctx = _ctx;
1850 if (!lte->will_be_in_output_wim &&
1851 stream_hard_filtered(lte, ctx->filter_ctx))
1852 stream_size_table_insert(lte, ctx->tab);
1857 determine_stream_size_uniquity(struct list_head *stream_list,
1858 struct wim_lookup_table *lt,
1859 struct filter_context *filter_ctx)
1862 struct stream_size_table tab;
1863 struct wim_lookup_table_entry *lte;
1865 ret = init_stream_size_table(&tab, 9001);
1869 if (may_hard_filter_streams(filter_ctx)) {
1870 struct insert_other_if_hard_filtered_ctx ctx = {
1872 .filter_ctx = filter_ctx,
1874 for_lookup_table_entry(lt, insert_other_if_hard_filtered, &ctx);
1877 list_for_each_entry(lte, stream_list, write_streams_list)
1878 stream_size_table_insert(lte, &tab);
1880 destroy_stream_size_table(&tab);
1885 filter_stream_list_for_write(struct list_head *stream_list,
1886 struct filter_context *filter_ctx)
1888 struct wim_lookup_table_entry *lte, *tmp;
1890 list_for_each_entry_safe(lte, tmp,
1891 stream_list, write_streams_list)
1893 int status = stream_filtered(lte, filter_ctx);
1900 /* Soft filtered. */
1902 /* Hard filtered. */
1903 lte->will_be_in_output_wim = 0;
1904 list_del(<e->lookup_table_list);
1906 list_del(<e->write_streams_list);
1912 * prepare_stream_list_for_write() -
1914 * Prepare the list of streams to write for writing a WIM containing the
1915 * specified image(s) with the specified write flags.
1918 * The WIMStruct on whose behalf the write is occurring.
1921 * Image(s) from the WIM to write; may be WIMLIB_ALL_IMAGES.
1924 * WIMLIB_WRITE_FLAG_* flags for the write operation:
1926 * STREAMS_OK: For writes of all images, assume that all streams in the
1927 * lookup table of @wim and the per-image lists of unhashed streams should
1928 * be taken as-is, and image metadata should not be searched for
1929 * references. This does not exclude filtering with OVERWRITE and
1930 * SKIP_EXTERNAL_WIMS, below.
1932 * OVERWRITE: Streams already present in @wim shall not be returned in
1935 * SKIP_EXTERNAL_WIMS: Streams already present in a WIM file, but not
1936 * @wim, shall be returned in neither @stream_list_ret nor
1937 * @lookup_table_list_ret.
1940 * List of streams, linked by write_streams_list, that need to be written
1941 * will be returned here.
1943 * Note that this function assumes that unhashed streams will be written;
1944 * it does not take into account that they may become duplicates when
1947 * @lookup_table_list_ret
1948 * List of streams, linked by lookup_table_list, that need to be included
1949 * in the WIM's lookup table will be returned here. This will be a
1950 * superset of the streams in @stream_list_ret.
1952 * This list will be a proper superset of @stream_list_ret if and only if
1953 * WIMLIB_WRITE_FLAG_OVERWRITE was specified in @write_flags and some of
1954 * the streams that would otherwise need to be written were already located
1957 * All streams in this list will have @out_refcnt set to the number of
1958 * references to the stream in the output WIM. If
1959 * WIMLIB_WRITE_FLAG_STREAMS_OK was specified in @write_flags, @out_refcnt
1960 * may be as low as 0.
1963 * A context for queries of stream filter status with stream_filtered() is
1964 * returned in this location.
1966 * In addition, @will_be_in_output_wim will be set to 1 in all stream entries
1967 * inserted into @lookup_table_list_ret and to 0 in all stream entries in the
1968 * lookup table of @wim not inserted into @lookup_table_list_ret.
1970 * Still furthermore, @unique_size will be set to 1 on all stream entries in
1971 * @stream_list_ret that have unique size among all stream entries in
1972 * @stream_list_ret and among all stream entries in the lookup table of @wim
1973 * that are ineligible for being written due to filtering.
1975 * Returns 0 on success; nonzero on read error, memory allocation error, or
1979 prepare_stream_list_for_write(WIMStruct *wim, int image,
1981 struct list_head *stream_list_ret,
1982 struct list_head *lookup_table_list_ret,
1983 struct filter_context *filter_ctx_ret)
1986 struct wim_lookup_table_entry *lte;
1988 filter_ctx_ret->write_flags = write_flags;
1989 filter_ctx_ret->wim = wim;
1991 ret = prepare_unfiltered_list_of_streams_in_output_wim(
1994 write_flags & WIMLIB_WRITE_FLAG_STREAMS_OK,
1999 INIT_LIST_HEAD(lookup_table_list_ret);
2000 list_for_each_entry(lte, stream_list_ret, write_streams_list)
2001 list_add_tail(<e->lookup_table_list, lookup_table_list_ret);
2003 ret = determine_stream_size_uniquity(stream_list_ret, wim->lookup_table,
2008 if (may_filter_streams(filter_ctx_ret))
2009 filter_stream_list_for_write(stream_list_ret, filter_ctx_ret);
2015 write_wim_streams(WIMStruct *wim, int image, int write_flags,
2016 unsigned num_threads,
2017 struct list_head *stream_list_override,
2018 struct list_head *lookup_table_list_ret)
2021 struct list_head _stream_list;
2022 struct list_head *stream_list;
2023 struct wim_lookup_table_entry *lte;
2024 struct filter_context _filter_ctx;
2025 struct filter_context *filter_ctx;
2027 if (stream_list_override == NULL) {
2028 /* Normal case: prepare stream list from image(s) being written.
2030 stream_list = &_stream_list;
2031 filter_ctx = &_filter_ctx;
2032 ret = prepare_stream_list_for_write(wim, image, write_flags,
2034 lookup_table_list_ret,
2039 /* Currently only as a result of wimlib_split() being called:
2040 * use stream list already explicitly provided. Use existing
2041 * reference counts. */
2042 stream_list = stream_list_override;
2044 INIT_LIST_HEAD(lookup_table_list_ret);
2045 list_for_each_entry(lte, stream_list, write_streams_list) {
2046 lte->out_refcnt = lte->refcnt;
2047 lte->will_be_in_output_wim = 1;
2048 lte->unique_size = 0;
2049 list_add_tail(<e->lookup_table_list, lookup_table_list_ret);
2053 return wim_write_stream_list(wim,
2061 write_wim_metadata_resources(WIMStruct *wim, int image, int write_flags)
2066 int write_resource_flags;
2068 if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA) {
2069 DEBUG("Not writing any metadata resources.");
2073 write_resource_flags = write_flags_to_resource_flags(write_flags);
2075 write_resource_flags &= ~WRITE_RESOURCE_FLAG_PACK_STREAMS;
2077 DEBUG("Writing metadata resources (offset=%"PRIu64")",
2078 wim->out_fd.offset);
2080 ret = call_progress(wim->progfunc,
2081 WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN,
2082 NULL, wim->progctx);
2086 if (image == WIMLIB_ALL_IMAGES) {
2088 end_image = wim->hdr.image_count;
2090 start_image = image;
2094 for (int i = start_image; i <= end_image; i++) {
2095 struct wim_image_metadata *imd;
2097 imd = wim->image_metadata[i - 1];
2098 /* Build a new metadata resource only if image was modified from
2099 * the original (or was newly added). Otherwise just copy the
2101 if (imd->modified) {
2102 DEBUG("Image %u was modified; building and writing new "
2103 "metadata resource", i);
2104 ret = write_metadata_resource(wim, i,
2105 write_resource_flags);
2106 } else if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) {
2107 DEBUG("Image %u was not modified; re-using existing "
2108 "metadata resource.", i);
2109 stream_set_out_reshdr_for_reuse(imd->metadata_lte);
2112 DEBUG("Image %u was not modified; copying existing "
2113 "metadata resource.", i);
2114 ret = write_wim_resource(imd->metadata_lte,
2116 wim->out_compression_type,
2117 wim->out_chunk_size,
2118 write_resource_flags);
2124 return call_progress(wim->progfunc,
2125 WIMLIB_PROGRESS_MSG_WRITE_METADATA_END,
2126 NULL, wim->progctx);
2130 open_wim_writable(WIMStruct *wim, const tchar *path, int open_flags)
2133 DEBUG("Opening \"%"TS"\" for writing.", path);
2135 raw_fd = topen(path, open_flags | O_BINARY, 0644);
2137 ERROR_WITH_ERRNO("Failed to open \"%"TS"\" for writing", path);
2138 return WIMLIB_ERR_OPEN;
2140 filedes_init(&wim->out_fd, raw_fd);
2145 close_wim_writable(WIMStruct *wim, int write_flags)
2149 if (!(write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)) {
2150 DEBUG("Closing WIM file.");
2151 if (filedes_valid(&wim->out_fd))
2152 if (filedes_close(&wim->out_fd))
2153 ret = WIMLIB_ERR_WRITE;
2155 filedes_invalidate(&wim->out_fd);
2160 cmp_streams_by_out_rspec(const void *p1, const void *p2)
2162 const struct wim_lookup_table_entry *lte1, *lte2;
2164 lte1 = *(const struct wim_lookup_table_entry**)p1;
2165 lte2 = *(const struct wim_lookup_table_entry**)p2;
2167 if (lte1->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
2168 if (lte2->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
2169 if (lte1->out_res_offset_in_wim != lte2->out_res_offset_in_wim)
2170 return cmp_u64(lte1->out_res_offset_in_wim,
2171 lte2->out_res_offset_in_wim);
2176 if (lte2->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS)
2179 return cmp_u64(lte1->out_reshdr.offset_in_wim,
2180 lte2->out_reshdr.offset_in_wim);
2184 write_wim_lookup_table(WIMStruct *wim, int image, int write_flags,
2185 struct wim_reshdr *out_reshdr,
2186 struct list_head *lookup_table_list)
2190 /* Set output resource metadata for streams already present in WIM. */
2191 if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) {
2192 struct wim_lookup_table_entry *lte;
2193 list_for_each_entry(lte, lookup_table_list, lookup_table_list)
2195 if (lte->resource_location == RESOURCE_IN_WIM &&
2196 lte->rspec->wim == wim)
2198 stream_set_out_reshdr_for_reuse(lte);
2203 ret = sort_stream_list(lookup_table_list,
2204 offsetof(struct wim_lookup_table_entry, lookup_table_list),
2205 cmp_streams_by_out_rspec);
2209 /* Add entries for metadata resources. */
2210 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)) {
2214 if (image == WIMLIB_ALL_IMAGES) {
2216 end_image = wim->hdr.image_count;
2218 start_image = image;
2222 /* Push metadata resource lookup table entries onto the front of
2223 * the list in reverse order, so that they're written in order.
2225 for (int i = end_image; i >= start_image; i--) {
2226 struct wim_lookup_table_entry *metadata_lte;
2228 metadata_lte = wim->image_metadata[i - 1]->metadata_lte;
2229 wimlib_assert(metadata_lte->out_reshdr.flags & WIM_RESHDR_FLAG_METADATA);
2230 metadata_lte->out_refcnt = 1;
2231 list_add(&metadata_lte->lookup_table_list, lookup_table_list);
2235 return write_wim_lookup_table_from_stream_list(lookup_table_list,
2237 wim->hdr.part_number,
2239 write_flags_to_resource_flags(write_flags));
2245 * Finish writing a WIM file: write the lookup table, xml data, and integrity
2246 * table, then overwrite the WIM header. By default, closes the WIM file
2247 * descriptor (@wim->out_fd) if successful.
2249 * write_flags is a bitwise OR of the following:
2251 * (public) WIMLIB_WRITE_FLAG_CHECK_INTEGRITY:
2252 * Include an integrity table.
2254 * (public) WIMLIB_WRITE_FLAG_FSYNC:
2255 * fsync() the output file before closing it.
2257 * (public) WIMLIB_WRITE_FLAG_PIPABLE:
2258 * Writing a pipable WIM, possibly to a pipe; include pipable WIM
2259 * stream headers before the lookup table and XML data, and also
2260 * write the WIM header at the end instead of seeking to the
2261 * beginning. Can't be combined with
2262 * WIMLIB_WRITE_FLAG_CHECK_INTEGRITY.
2264 * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE:
2265 * Don't write the lookup table.
2267 * (private) WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE:
2268 * When (if) writing the integrity table, re-use entries from the
2269 * existing integrity table, if possible.
2271 * (private) WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML:
2272 * After writing the XML data but before writing the integrity
2273 * table, write a temporary WIM header and flush the stream so that
2274 * the WIM is less likely to become corrupted upon abrupt program
2276 * (private) WIMLIB_WRITE_FLAG_HEADER_AT_END:
2277 * Instead of overwriting the WIM header at the beginning of the
2278 * file, simply append it to the end of the file. (Used when
2280 * (private) WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR:
2281 * Do not close the file descriptor @wim->out_fd on either success
2283 * (private) WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES:
2284 * Use the existing <TOTALBYTES> stored in the in-memory XML
2285 * information, rather than setting it to the offset of the XML
2286 * data being written.
2289 finish_write(WIMStruct *wim, int image, int write_flags,
2290 struct list_head *lookup_table_list)
2294 int write_resource_flags;
2295 off_t old_lookup_table_end;
2296 off_t new_lookup_table_end;
2299 DEBUG("image=%d, write_flags=%08x", image, write_flags);
2301 write_resource_flags = write_flags_to_resource_flags(write_flags);
2303 /* In the WIM header, there is room for the resource entry for a
2304 * metadata resource labeled as the "boot metadata". This entry should
2305 * be zeroed out if there is no bootable image (boot_idx 0). Otherwise,
2306 * it should be a copy of the resource entry for the image that is
2307 * marked as bootable. This is not well documented... */
2308 if (wim->hdr.boot_idx == 0) {
2309 zero_reshdr(&wim->hdr.boot_metadata_reshdr);
2311 copy_reshdr(&wim->hdr.boot_metadata_reshdr,
2312 &wim->image_metadata[
2313 wim->hdr.boot_idx - 1]->metadata_lte->out_reshdr);
2316 /* Write lookup table. (Save old position first.) */
2317 old_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim +
2318 wim->hdr.lookup_table_reshdr.size_in_wim;
2319 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
2320 ret = write_wim_lookup_table(wim, image, write_flags,
2321 &wim->hdr.lookup_table_reshdr,
2327 /* Write XML data. */
2328 xml_totalbytes = wim->out_fd.offset;
2329 if (write_flags & WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES)
2330 xml_totalbytes = WIM_TOTALBYTES_USE_EXISTING;
2331 ret = write_wim_xml_data(wim, image, xml_totalbytes,
2332 &wim->hdr.xml_data_reshdr,
2333 write_resource_flags);
2337 /* Write integrity table (optional). */
2338 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
2339 if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) {
2340 struct wim_header checkpoint_hdr;
2341 memcpy(&checkpoint_hdr, &wim->hdr, sizeof(struct wim_header));
2342 zero_reshdr(&checkpoint_hdr.integrity_table_reshdr);
2343 checkpoint_hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2344 ret = write_wim_header_at_offset(&checkpoint_hdr,
2350 if (!(write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE))
2351 old_lookup_table_end = 0;
2353 if (wim->hdr.integrity_table_reshdr.offset_in_wim <
2354 wim->hdr.xml_data_reshdr.offset_in_wim +
2355 wim->hdr.xml_data_reshdr.size_in_wim)
2357 /* Old integrity table was partially overwritten by the
2359 old_lookup_table_end = 0;
2362 new_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim +
2363 wim->hdr.lookup_table_reshdr.size_in_wim;
2365 ret = write_integrity_table(wim,
2366 new_lookup_table_end,
2367 old_lookup_table_end);
2371 /* No integrity table. */
2372 zero_reshdr(&wim->hdr.integrity_table_reshdr);
2375 /* Now that all information in the WIM header has been determined, the
2376 * preliminary header written earlier can be overwritten, the header of
2377 * the existing WIM file can be overwritten, or the final header can be
2378 * written to the end of the pipable WIM. */
2379 wim->hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2381 if (write_flags & WIMLIB_WRITE_FLAG_HEADER_AT_END)
2382 hdr_offset = wim->out_fd.offset;
2383 DEBUG("Writing new header @ %"PRIu64".", hdr_offset);
2384 ret = write_wim_header_at_offset(&wim->hdr, &wim->out_fd, hdr_offset);
2388 /* Possibly sync file data to disk before closing. On POSIX systems, it
2389 * is necessary to do this before using rename() to overwrite an
2390 * existing file with a new file. Otherwise, data loss would occur if
2391 * the system is abruptly terminated when the metadata for the rename
2392 * operation has been written to disk, but the new file data has not.
2394 if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) {
2395 DEBUG("Syncing WIM file.");
2396 if (fsync(wim->out_fd.fd)) {
2397 ERROR_WITH_ERRNO("Error syncing data to WIM file");
2398 return WIMLIB_ERR_WRITE;
2402 if (close_wim_writable(wim, write_flags)) {
2403 ERROR_WITH_ERRNO("Failed to close the output WIM file");
2404 return WIMLIB_ERR_WRITE;
2410 #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK)
2412 /* Set advisory lock on WIM file (if not already done so) */
2414 lock_wim_for_append(WIMStruct *wim)
2416 if (wim->locked_for_append)
2418 if (!flock(wim->in_fd.fd, LOCK_EX | LOCK_NB)) {
2419 wim->locked_for_append = 1;
2422 if (errno != EWOULDBLOCK)
2424 return WIMLIB_ERR_ALREADY_LOCKED;
2427 /* Remove advisory lock on WIM file (if present) */
2429 unlock_wim_for_append(WIMStruct *wim)
2431 if (wim->locked_for_append) {
2432 flock(wim->in_fd.fd, LOCK_UN);
2433 wim->locked_for_append = 0;
2439 * write_pipable_wim():
2441 * Perform the intermediate stages of creating a "pipable" WIM (i.e. a WIM
2442 * capable of being applied from a pipe).
2444 * Pipable WIMs are a wimlib-specific modification of the WIM format such that
2445 * images can be applied from them sequentially when the file data is sent over
2446 * a pipe. In addition, a pipable WIM can be written sequentially to a pipe.
2447 * The modifications made to the WIM format for pipable WIMs are:
2449 * - Magic characters in header are "WLPWM\0\0\0" (wimlib pipable WIM) instead
2450 * of "MSWIM\0\0\0". This lets wimlib know that the WIM is pipable and also
2451 * stops other software from trying to read the file as a normal WIM.
2453 * - The header at the beginning of the file does not contain all the normal
2454 * information; in particular it will have all 0's for the lookup table and
2455 * XML data resource entries. This is because this information cannot be
2456 * determined until the lookup table and XML data have been written.
2457 * Consequently, wimlib will write the full header at the very end of the
2458 * file. The header at the end, however, is only used when reading the WIM
2459 * from a seekable file (not a pipe).
2461 * - An extra copy of the XML data is placed directly after the header. This
2462 * allows image names and sizes to be determined at an appropriate time when
2463 * reading the WIM from a pipe. This copy of the XML data is ignored if the
2464 * WIM is read from a seekable file (not a pipe).
2466 * - The format of resources, or streams, has been modified to allow them to be
2467 * used before the "lookup table" has been read. Each stream is prefixed with
2468 * a `struct pwm_stream_hdr' that is basically an abbreviated form of `struct
2469 * wim_lookup_table_entry_disk' that only contains the SHA1 message digest,
2470 * uncompressed stream size, and flags that indicate whether the stream is
2471 * compressed. The data of uncompressed streams then follows literally, while
2472 * the data of compressed streams follows in a modified format. Compressed
2473 * streams do not begin with a chunk table, since the chunk table cannot be
2474 * written until all chunks have been compressed. Instead, each compressed
2475 * chunk is prefixed by a `struct pwm_chunk_hdr' that gives its size.
2476 * Furthermore, the chunk table is written at the end of the resource instead
2477 * of the start. Note: chunk offsets are given in the chunk table as if the
2478 * `struct pwm_chunk_hdr's were not present; also, the chunk table is only
2479 * used if the WIM is being read from a seekable file (not a pipe).
2481 * - Metadata resources always come before other file resources (streams).
2482 * (This does not by itself constitute an incompatibility with normal WIMs,
2483 * since this is valid in normal WIMs.)
2485 * - At least up to the end of the file resources, all components must be packed
2486 * as tightly as possible; there cannot be any "holes" in the WIM. (This does
2487 * not by itself consititute an incompatibility with normal WIMs, since this
2488 * is valid in normal WIMs.)
2490 * Note: the lookup table, XML data, and header at the end are not used when
2491 * applying from a pipe. They exist to support functionality such as image
2492 * application and export when the WIM is *not* read from a pipe.
2494 * Layout of pipable WIM:
2496 * ---------+----------+--------------------+----------------+--------------+-----------+--------+
2497 * | Header | XML data | Metadata resources | File resources | Lookup table | XML data | Header |
2498 * ---------+----------+--------------------+----------------+--------------+-----------+--------+
2500 * Layout of normal WIM:
2502 * +--------+-----------------------------+-------------------------+
2503 * | Header | File and metadata resources | Lookup table | XML data |
2504 * +--------+-----------------------------+-------------------------+
2506 * An optional integrity table can follow the final XML data in both normal and
2507 * pipable WIMs. However, due to implementation details, wimlib currently can
2508 * only include an integrity table in a pipable WIM when writing it to a
2509 * seekable file (not a pipe).
2511 * Do note that since pipable WIMs are not supported by Microsoft's software,
2512 * wimlib does not create them unless explicitly requested (with
2513 * WIMLIB_WRITE_FLAG_PIPABLE) and as stated above they use different magic
2514 * characters to identify the file.
2517 write_pipable_wim(WIMStruct *wim, int image, int write_flags,
2518 unsigned num_threads,
2519 struct list_head *stream_list_override,
2520 struct list_head *lookup_table_list_ret)
2523 struct wim_reshdr xml_reshdr;
2525 WARNING("Creating a pipable WIM, which will "
2527 " with Microsoft's software (wimgapi/imagex/Dism).");
2529 /* At this point, the header at the beginning of the file has already
2532 /* For efficiency, when wimlib adds an image to the WIM with
2533 * wimlib_add_image(), the SHA1 message digests of files is not
2534 * calculated; instead, they are calculated while the files are being
2535 * written. However, this does not work when writing a pipable WIM,
2536 * since when writing a stream to a pipable WIM, its SHA1 message digest
2537 * needs to be known before the stream data is written. Therefore,
2538 * before getting much farther, we need to pre-calculate the SHA1
2539 * message digests of all streams that will be written. */
2540 ret = wim_checksum_unhashed_streams(wim);
2544 /* Write extra copy of the XML data. */
2545 ret = write_wim_xml_data(wim, image, WIM_TOTALBYTES_OMIT,
2547 WRITE_RESOURCE_FLAG_PIPABLE);
2551 /* Write metadata resources for the image(s) being included in the
2553 ret = write_wim_metadata_resources(wim, image, write_flags);
2557 /* Write streams needed for the image(s) being included in the output
2558 * WIM, or streams needed for the split WIM part. */
2559 return write_wim_streams(wim, image, write_flags, num_threads,
2560 stream_list_override, lookup_table_list_ret);
2562 /* The lookup table, XML data, and header at end are handled by
2563 * finish_write(). */
2566 /* Write a standalone WIM or split WIM (SWM) part to a new file or to a file
2569 write_wim_part(WIMStruct *wim,
2570 const void *path_or_fd,
2573 unsigned num_threads,
2574 unsigned part_number,
2575 unsigned total_parts,
2576 struct list_head *stream_list_override,
2580 struct wim_header hdr_save;
2581 struct list_head lookup_table_list;
2583 if (total_parts == 1)
2584 DEBUG("Writing standalone WIM.");
2586 DEBUG("Writing split WIM part %u/%u", part_number, total_parts);
2587 if (image == WIMLIB_ALL_IMAGES)
2588 DEBUG("Including all images.");
2590 DEBUG("Including image %d only.", image);
2591 if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)
2592 DEBUG("File descriptor: %d", *(const int*)path_or_fd);
2594 DEBUG("Path: \"%"TS"\"", (const tchar*)path_or_fd);
2595 DEBUG("Write flags: 0x%08x", write_flags);
2597 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)
2598 DEBUG("\tCHECK_INTEGRITY");
2600 if (write_flags & WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)
2601 DEBUG("\tNO_CHECK_INTEGRITY");
2603 if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
2606 if (write_flags & WIMLIB_WRITE_FLAG_NOT_PIPABLE)
2607 DEBUG("\tNOT_PIPABLE");
2609 if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
2610 DEBUG("\tRECOMPRESS");
2612 if (write_flags & WIMLIB_WRITE_FLAG_FSYNC)
2615 if (write_flags & WIMLIB_WRITE_FLAG_REBUILD)
2618 if (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE)
2619 DEBUG("\tSOFT_DELETE");
2621 if (write_flags & WIMLIB_WRITE_FLAG_IGNORE_READONLY_FLAG)
2622 DEBUG("\tIGNORE_READONLY_FLAG");
2624 if (write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS)
2625 DEBUG("\tSKIP_EXTERNAL_WIMS");
2627 if (write_flags & WIMLIB_WRITE_FLAG_STREAMS_OK)
2628 DEBUG("\tSTREAMS_OK");
2630 if (write_flags & WIMLIB_WRITE_FLAG_RETAIN_GUID)
2631 DEBUG("\tRETAIN_GUID");
2633 if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS)
2634 DEBUG("\tPACK_STREAMS");
2636 if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)
2637 DEBUG("\tFILE_DESCRIPTOR");
2639 if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)
2640 DEBUG("\tNO_METADATA");
2642 if (write_flags & WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES)
2643 DEBUG("\tUSE_EXISTING_TOTALBYTES");
2645 if (num_threads == 0)
2646 DEBUG("Number of threads: autodetect");
2648 DEBUG("Number of threads: %u", num_threads);
2649 DEBUG("Progress function: %s", (wim->progfunc ? "yes" : "no"));
2650 DEBUG("Stream list: %s", (stream_list_override ? "specified" : "autodetect"));
2651 DEBUG("GUID: %s", (write_flags &
2652 WIMLIB_WRITE_FLAG_RETAIN_GUID) ? "retain"
2653 : guid ? "explicit" : "generate new");
2655 /* Internally, this is always called with a valid part number and total
2657 wimlib_assert(total_parts >= 1);
2658 wimlib_assert(part_number >= 1 && part_number <= total_parts);
2660 /* A valid image (or all images) must be specified. */
2661 if (image != WIMLIB_ALL_IMAGES &&
2662 (image < 1 || image > wim->hdr.image_count))
2663 return WIMLIB_ERR_INVALID_IMAGE;
2665 /* If we need to write metadata resources, make sure the ::WIMStruct has
2666 * the needed information attached (e.g. is not a resource-only WIM,
2667 * such as a non-first part of a split WIM). */
2668 if (!wim_has_metadata(wim) &&
2669 !(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA))
2670 return WIMLIB_ERR_METADATA_NOT_FOUND;
2672 /* Check for contradictory flags. */
2673 if ((write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2674 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))
2675 == (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2676 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))
2677 return WIMLIB_ERR_INVALID_PARAM;
2679 if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2680 WIMLIB_WRITE_FLAG_NOT_PIPABLE))
2681 == (WIMLIB_WRITE_FLAG_PIPABLE |
2682 WIMLIB_WRITE_FLAG_NOT_PIPABLE))
2683 return WIMLIB_ERR_INVALID_PARAM;
2685 /* Save previous header, then start initializing the new one. */
2686 memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header));
2688 /* Set default integrity, pipable, and packed stream flags. */
2689 if (!(write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2690 WIMLIB_WRITE_FLAG_NOT_PIPABLE)))
2691 if (wim_is_pipable(wim)) {
2692 DEBUG("WIM is pipable; default to PIPABLE.");
2693 write_flags |= WIMLIB_WRITE_FLAG_PIPABLE;
2696 if (!(write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2697 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)))
2698 if (wim_has_integrity_table(wim)) {
2699 DEBUG("Integrity table present; default to CHECK_INTEGRITY.");
2700 write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY;
2703 if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2704 WIMLIB_WRITE_FLAG_PACK_STREAMS))
2705 == (WIMLIB_WRITE_FLAG_PIPABLE |
2706 WIMLIB_WRITE_FLAG_PACK_STREAMS))
2708 ERROR("Cannot specify both PIPABLE and PACK_STREAMS!");
2709 return WIMLIB_ERR_INVALID_PARAM;
2712 /* Set appropriate magic number. */
2713 if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
2714 wim->hdr.magic = PWM_MAGIC;
2716 wim->hdr.magic = WIM_MAGIC;
2718 /* Set appropriate version number. */
2719 if ((write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS) ||
2720 wim->out_compression_type == WIMLIB_COMPRESSION_TYPE_LZMS)
2721 wim->hdr.wim_version = WIM_VERSION_PACKED_STREAMS;
2723 wim->hdr.wim_version = WIM_VERSION_DEFAULT;
2725 /* Clear header flags that will be set automatically. */
2726 wim->hdr.flags &= ~(WIM_HDR_FLAG_METADATA_ONLY |
2727 WIM_HDR_FLAG_RESOURCE_ONLY |
2728 WIM_HDR_FLAG_SPANNED |
2729 WIM_HDR_FLAG_WRITE_IN_PROGRESS);
2731 /* Set SPANNED header flag if writing part of a split WIM. */
2732 if (total_parts != 1)
2733 wim->hdr.flags |= WIM_HDR_FLAG_SPANNED;
2735 /* Set part number and total parts of split WIM. This will be 1 and 1
2736 * if the WIM is standalone. */
2737 wim->hdr.part_number = part_number;
2738 wim->hdr.total_parts = total_parts;
2740 /* Set compression type if different. */
2741 if (wim->compression_type != wim->out_compression_type) {
2742 ret = set_wim_hdr_cflags(wim->out_compression_type, &wim->hdr);
2743 wimlib_assert(ret == 0);
2746 /* Set chunk size if different. */
2747 wim->hdr.chunk_size = wim->out_chunk_size;
2750 if (!(write_flags & WIMLIB_WRITE_FLAG_RETAIN_GUID)) {
2752 memcpy(wim->hdr.guid, guid, WIMLIB_GUID_LEN);
2754 randomize_byte_array(wim->hdr.guid, WIMLIB_GUID_LEN);
2757 /* Clear references to resources that have not been written yet. */
2758 zero_reshdr(&wim->hdr.lookup_table_reshdr);
2759 zero_reshdr(&wim->hdr.xml_data_reshdr);
2760 zero_reshdr(&wim->hdr.boot_metadata_reshdr);
2761 zero_reshdr(&wim->hdr.integrity_table_reshdr);
2763 /* Set image count and boot index correctly for single image writes. */
2764 if (image != WIMLIB_ALL_IMAGES) {
2765 wim->hdr.image_count = 1;
2766 if (wim->hdr.boot_idx == image)
2767 wim->hdr.boot_idx = 1;
2769 wim->hdr.boot_idx = 0;
2772 /* Split WIMs can't be bootable. */
2773 if (total_parts != 1)
2774 wim->hdr.boot_idx = 0;
2776 /* Initialize output file descriptor. */
2777 if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR) {
2778 /* File descriptor was explicitly provided. Return error if
2779 * file descriptor is not seekable, unless writing a pipable WIM
2781 wim->out_fd.fd = *(const int*)path_or_fd;
2782 wim->out_fd.offset = 0;
2783 if (!filedes_is_seekable(&wim->out_fd)) {
2784 ret = WIMLIB_ERR_INVALID_PARAM;
2785 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
2786 goto out_restore_hdr;
2787 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
2788 ERROR("Can't include integrity check when "
2789 "writing pipable WIM to pipe!");
2790 goto out_restore_hdr;
2795 /* Filename of WIM to write was provided; open file descriptor
2797 ret = open_wim_writable(wim, (const tchar*)path_or_fd,
2798 O_TRUNC | O_CREAT | O_RDWR);
2800 goto out_restore_hdr;
2803 /* Write initial header. This is merely a "dummy" header since it
2804 * doesn't have all the information yet, so it will be overwritten later
2805 * (unless writing a pipable WIM). */
2806 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
2807 wim->hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2808 ret = write_wim_header(&wim->hdr, &wim->out_fd);
2809 wim->hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2811 goto out_restore_hdr;
2813 /* Write metadata resources and streams. */
2814 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE)) {
2815 /* Default case: create a normal (non-pipable) WIM. */
2816 ret = write_wim_streams(wim, image, write_flags, num_threads,
2817 stream_list_override,
2818 &lookup_table_list);
2820 goto out_restore_hdr;
2822 ret = write_wim_metadata_resources(wim, image, write_flags);
2824 goto out_restore_hdr;
2826 /* Non-default case: create pipable WIM. */
2827 ret = write_pipable_wim(wim, image, write_flags, num_threads,
2828 stream_list_override,
2829 &lookup_table_list);
2831 goto out_restore_hdr;
2832 write_flags |= WIMLIB_WRITE_FLAG_HEADER_AT_END;
2836 /* Write lookup table, XML data, and (optional) integrity table. */
2837 ret = finish_write(wim, image, write_flags, &lookup_table_list);
2839 memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header));
2840 (void)close_wim_writable(wim, write_flags);
2841 DEBUG("ret=%d", ret);
2845 /* Write a standalone WIM to a file or file descriptor. */
2847 write_standalone_wim(WIMStruct *wim, const void *path_or_fd,
2848 int image, int write_flags, unsigned num_threads)
2850 return write_wim_part(wim, path_or_fd, image, write_flags,
2851 num_threads, 1, 1, NULL, NULL);
2854 /* API function documented in wimlib.h */
2856 wimlib_write(WIMStruct *wim, const tchar *path,
2857 int image, int write_flags, unsigned num_threads)
2859 if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC)
2860 return WIMLIB_ERR_INVALID_PARAM;
2862 if (path == NULL || path[0] == T('\0'))
2863 return WIMLIB_ERR_INVALID_PARAM;
2865 return write_standalone_wim(wim, path, image, write_flags, num_threads);
2868 /* API function documented in wimlib.h */
2870 wimlib_write_to_fd(WIMStruct *wim, int fd,
2871 int image, int write_flags, unsigned num_threads)
2873 if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC)
2874 return WIMLIB_ERR_INVALID_PARAM;
2877 return WIMLIB_ERR_INVALID_PARAM;
2879 write_flags |= WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR;
2881 return write_standalone_wim(wim, &fd, image, write_flags, num_threads);
2885 any_images_modified(WIMStruct *wim)
2887 for (int i = 0; i < wim->hdr.image_count; i++)
2888 if (wim->image_metadata[i]->modified)
2894 check_resource_offset(struct wim_lookup_table_entry *lte, void *_wim)
2896 const WIMStruct *wim = _wim;
2897 off_t end_offset = *(const off_t*)wim->private;
2899 if (lte->resource_location == RESOURCE_IN_WIM && lte->rspec->wim == wim &&
2900 lte->rspec->offset_in_wim + lte->rspec->size_in_wim > end_offset)
2901 return WIMLIB_ERR_RESOURCE_ORDER;
2905 /* Make sure no file or metadata resources are located after the XML data (or
2906 * integrity table if present)--- otherwise we can't safely overwrite the WIM in
2907 * place and we return WIMLIB_ERR_RESOURCE_ORDER. */
2909 check_resource_offsets(WIMStruct *wim, off_t end_offset)
2914 wim->private = &end_offset;
2915 ret = for_lookup_table_entry(wim->lookup_table, check_resource_offset, wim);
2919 for (i = 0; i < wim->hdr.image_count; i++) {
2920 ret = check_resource_offset(wim->image_metadata[i]->metadata_lte, wim);
2928 * Overwrite a WIM, possibly appending streams to it.
2930 * A WIM looks like (or is supposed to look like) the following:
2932 * Header (212 bytes)
2933 * Streams and metadata resources (variable size)
2934 * Lookup table (variable size)
2935 * XML data (variable size)
2936 * Integrity table (optional) (variable size)
2938 * If we are not adding any streams or metadata resources, the lookup table is
2939 * unchanged--- so we only need to overwrite the XML data, integrity table, and
2940 * header. This operation is potentially unsafe if the program is abruptly
2941 * terminated while the XML data or integrity table are being overwritten, but
2942 * before the new header has been written. To partially alleviate this problem,
2943 * a special flag (WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) is passed to
2944 * finish_write() to cause a temporary WIM header to be written after the XML
2945 * data has been written. This may prevent the WIM from becoming corrupted if
2946 * the program is terminated while the integrity table is being calculated (but
2947 * no guarantees, due to write re-ordering...).
2949 * If we are adding new streams or images (metadata resources), the lookup table
2950 * needs to be changed, and those streams need to be written. In this case, we
2951 * try to perform a safe update of the WIM file by writing the streams *after*
2952 * the end of the previous WIM, then writing the new lookup table, XML data, and
2953 * (optionally) integrity table following the new streams. This will produce a
2954 * layout like the following:
2956 * Header (212 bytes)
2957 * (OLD) Streams and metadata resources (variable size)
2958 * (OLD) Lookup table (variable size)
2959 * (OLD) XML data (variable size)
2960 * (OLD) Integrity table (optional) (variable size)
2961 * (NEW) Streams and metadata resources (variable size)
2962 * (NEW) Lookup table (variable size)
2963 * (NEW) XML data (variable size)
2964 * (NEW) Integrity table (optional) (variable size)
2966 * At all points, the WIM is valid as nothing points to the new data yet. Then,
2967 * the header is overwritten to point to the new lookup table, XML data, and
2968 * integrity table, to produce the following layout:
2970 * Header (212 bytes)
2971 * Streams and metadata resources (variable size)
2972 * Nothing (variable size)
2973 * More Streams and metadata resources (variable size)
2974 * Lookup table (variable size)
2975 * XML data (variable size)
2976 * Integrity table (optional) (variable size)
2978 * This method allows an image to be appended to a large WIM very quickly, and
2979 * is crash-safe except in the case of write re-ordering, but the
2980 * disadvantage is that a small hole is left in the WIM where the old lookup
2981 * table, xml data, and integrity table were. (These usually only take up a
2982 * small amount of space compared to the streams, however.)
2985 overwrite_wim_inplace(WIMStruct *wim, int write_flags, unsigned num_threads)
2989 u64 old_lookup_table_end, old_xml_begin, old_xml_end;
2990 struct wim_header hdr_save;
2991 struct list_head stream_list;
2992 struct list_head lookup_table_list;
2993 struct filter_context filter_ctx;
2995 DEBUG("Overwriting `%"TS"' in-place", wim->filename);
2997 /* Save original header so it can be restored in case of error */
2998 memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header));
3000 /* Set default integrity flag. */
3001 if (!(write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
3002 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)))
3003 if (wim_has_integrity_table(wim))
3004 write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY;
3006 /* Set WIM version if adding packed streams. */
3007 if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS)
3008 wim->hdr.wim_version = WIM_VERSION_PACKED_STREAMS;
3010 /* Set additional flags for overwrite. */
3011 write_flags |= WIMLIB_WRITE_FLAG_OVERWRITE |
3012 WIMLIB_WRITE_FLAG_STREAMS_OK;
3014 /* Make sure that the integrity table (if present) is after the XML
3015 * data, and that there are no stream resources, metadata resources, or
3016 * lookup tables after the XML data. Otherwise, these data would be
3018 old_xml_begin = wim->hdr.xml_data_reshdr.offset_in_wim;
3019 old_xml_end = old_xml_begin + wim->hdr.xml_data_reshdr.size_in_wim;
3020 old_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim +
3021 wim->hdr.lookup_table_reshdr.size_in_wim;
3022 if (wim->hdr.integrity_table_reshdr.offset_in_wim != 0 &&
3023 wim->hdr.integrity_table_reshdr.offset_in_wim < old_xml_end) {
3024 WARNING("Didn't expect the integrity table to be before the XML data");
3025 ret = WIMLIB_ERR_RESOURCE_ORDER;
3026 goto out_restore_memory_hdr;
3029 if (old_lookup_table_end > old_xml_begin) {
3030 WARNING("Didn't expect the lookup table to be after the XML data");
3031 ret = WIMLIB_ERR_RESOURCE_ORDER;
3032 goto out_restore_memory_hdr;
3035 /* Set @old_wim_end, which indicates the point beyond which we don't
3036 * allow any file and metadata resources to appear without returning
3037 * WIMLIB_ERR_RESOURCE_ORDER (due to the fact that we would otherwise
3038 * overwrite these resources). */
3039 if (!wim->deletion_occurred && !any_images_modified(wim)) {
3040 /* If no images have been modified and no images have been
3041 * deleted, a new lookup table does not need to be written. We
3042 * shall write the new XML data and optional integrity table
3043 * immediately after the lookup table. Note that this may
3044 * overwrite an existing integrity table. */
3045 DEBUG("Skipping writing lookup table "
3046 "(no images modified or deleted)");
3047 old_wim_end = old_lookup_table_end;
3048 write_flags |= WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE |
3049 WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML;
3050 } else if (wim->hdr.integrity_table_reshdr.offset_in_wim != 0) {
3051 /* Old WIM has an integrity table; begin writing new streams
3053 old_wim_end = wim->hdr.integrity_table_reshdr.offset_in_wim +
3054 wim->hdr.integrity_table_reshdr.size_in_wim;
3056 /* No existing integrity table; begin writing new streams after
3057 * the old XML data. */
3058 old_wim_end = old_xml_end;
3061 ret = check_resource_offsets(wim, old_wim_end);
3063 goto out_restore_memory_hdr;
3065 ret = prepare_stream_list_for_write(wim, WIMLIB_ALL_IMAGES, write_flags,
3066 &stream_list, &lookup_table_list,
3069 goto out_restore_memory_hdr;
3071 ret = open_wim_writable(wim, wim->filename, O_RDWR);
3073 goto out_restore_memory_hdr;
3075 ret = lock_wim_for_append(wim);
3079 /* Set WIM_HDR_FLAG_WRITE_IN_PROGRESS flag in header. */
3080 wim->hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
3081 ret = write_wim_header_flags(wim->hdr.flags, &wim->out_fd);
3083 ERROR_WITH_ERRNO("Error updating WIM header flags");
3084 goto out_unlock_wim;
3087 if (filedes_seek(&wim->out_fd, old_wim_end) == -1) {
3088 ERROR_WITH_ERRNO("Can't seek to end of WIM");
3089 ret = WIMLIB_ERR_WRITE;
3090 goto out_restore_physical_hdr;
3093 ret = wim_write_stream_list(wim,
3101 ret = write_wim_metadata_resources(wim, WIMLIB_ALL_IMAGES, write_flags);
3105 write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE;
3106 ret = finish_write(wim, WIMLIB_ALL_IMAGES, write_flags,
3107 &lookup_table_list);
3111 unlock_wim_for_append(wim);
3115 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
3116 WARNING("Truncating `%"TS"' to its original size (%"PRIu64" bytes)",
3117 wim->filename, old_wim_end);
3118 /* Return value of ftruncate() is ignored because this is
3119 * already an error path. */
3120 (void)ftruncate(wim->out_fd.fd, old_wim_end);
3122 out_restore_physical_hdr:
3123 (void)write_wim_header_flags(hdr_save.flags, &wim->out_fd);
3125 unlock_wim_for_append(wim);
3127 (void)close_wim_writable(wim, write_flags);
3128 out_restore_memory_hdr:
3129 memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header));
3134 overwrite_wim_via_tmpfile(WIMStruct *wim, int write_flags, unsigned num_threads)
3136 size_t wim_name_len;
3139 DEBUG("Overwriting `%"TS"' via a temporary file", wim->filename);
3141 /* Write the WIM to a temporary file in the same directory as the
3143 wim_name_len = tstrlen(wim->filename);
3144 tchar tmpfile[wim_name_len + 10];
3145 tmemcpy(tmpfile, wim->filename, wim_name_len);
3146 randomize_char_array_with_alnum(tmpfile + wim_name_len, 9);
3147 tmpfile[wim_name_len + 9] = T('\0');
3149 ret = wimlib_write(wim, tmpfile, WIMLIB_ALL_IMAGES,
3151 WIMLIB_WRITE_FLAG_FSYNC |
3152 WIMLIB_WRITE_FLAG_RETAIN_GUID,
3159 if (filedes_valid(&wim->in_fd)) {
3160 filedes_close(&wim->in_fd);
3161 filedes_invalidate(&wim->in_fd);
3164 /* Rename the new WIM file to the original WIM file. Note: on Windows
3165 * this actually calls win32_rename_replacement(), not _wrename(), so
3166 * that removing the existing destination file can be handled. */
3167 DEBUG("Renaming `%"TS"' to `%"TS"'", tmpfile, wim->filename);
3168 ret = trename(tmpfile, wim->filename);
3170 ERROR_WITH_ERRNO("Failed to rename `%"TS"' to `%"TS"'",
3171 tmpfile, wim->filename);
3178 return WIMLIB_ERR_RENAME;
3181 union wimlib_progress_info progress;
3182 progress.rename.from = tmpfile;
3183 progress.rename.to = wim->filename;
3184 return call_progress(wim->progfunc, WIMLIB_PROGRESS_MSG_RENAME,
3185 &progress, wim->progctx);
3188 /* Determine if the specified WIM file may be updated by appending in-place
3189 * rather than writing and replacing it with an entirely new file. */
3191 can_overwrite_wim_inplace(const WIMStruct *wim, int write_flags)
3193 /* REBUILD flag forces full rebuild. */
3194 if (write_flags & WIMLIB_WRITE_FLAG_REBUILD)
3197 /* Deletions cause full rebuild by default. */
3198 if (wim->deletion_occurred && !(write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE))
3201 /* Pipable WIMs cannot be updated in place, nor can a non-pipable WIM be
3202 * turned into a pipable WIM in-place. */
3203 if (wim_is_pipable(wim) || (write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
3206 /* The default compression type and compression chunk size selected for
3207 * the output WIM must be the same as those currently used for the WIM.
3209 if (wim->compression_type != wim->out_compression_type)
3211 if (wim->chunk_size != wim->out_chunk_size)
3217 /* API function documented in wimlib.h */
3219 wimlib_overwrite(WIMStruct *wim, int write_flags, unsigned num_threads)
3224 if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC)
3225 return WIMLIB_ERR_INVALID_PARAM;
3228 return WIMLIB_ERR_NO_FILENAME;
3230 orig_hdr_flags = wim->hdr.flags;
3231 if (write_flags & WIMLIB_WRITE_FLAG_IGNORE_READONLY_FLAG)
3232 wim->hdr.flags &= ~WIM_HDR_FLAG_READONLY;
3233 ret = can_modify_wim(wim);
3234 wim->hdr.flags = orig_hdr_flags;
3238 if (can_overwrite_wim_inplace(wim, write_flags)) {
3239 ret = overwrite_wim_inplace(wim, write_flags, num_threads);
3240 if (ret != WIMLIB_ERR_RESOURCE_ORDER)
3242 WARNING("Falling back to re-building entire WIM");
3244 return overwrite_wim_via_tmpfile(wim, write_flags, num_threads);