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/resource.h"
48 # include "wimlib/win32.h" /* win32_rename_replacement() */
50 #include "wimlib/write.h"
51 #include "wimlib/xml.h"
62 /* wimlib internal flags used when writing resources. */
63 #define WRITE_RESOURCE_FLAG_RECOMPRESS 0x00000001
64 #define WRITE_RESOURCE_FLAG_PIPABLE 0x00000002
65 #define WRITE_RESOURCE_FLAG_PACK_STREAMS 0x00000004
68 write_flags_to_resource_flags(int write_flags)
70 int write_resource_flags = 0;
72 if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
73 write_resource_flags |= WRITE_RESOURCE_FLAG_RECOMPRESS;
74 if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
75 write_resource_flags |= WRITE_RESOURCE_FLAG_PIPABLE;
76 if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS)
77 write_resource_flags |= WRITE_RESOURCE_FLAG_PACK_STREAMS;
78 return write_resource_flags;
81 struct filter_context {
86 /* Determine specified stream should be filtered out from the write.
90 * < 0 : The stream should be hard-filtered; that is, not included in the
92 * 0 : The stream should not be filtered out.
93 * > 0 : The stream should be soft-filtered; that is, it already exists in the
94 * WIM file and may not need to be written again.
97 stream_filtered(const struct wim_lookup_table_entry *lte,
98 const struct filter_context *ctx)
100 int write_flags = ctx->write_flags;
101 WIMStruct *wim = ctx->wim;
106 if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE &&
107 lte->resource_location == RESOURCE_IN_WIM &&
108 lte->rspec->wim == wim)
111 if (write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS &&
112 lte->resource_location == RESOURCE_IN_WIM &&
113 lte->rspec->wim != wim)
120 stream_hard_filtered(const struct wim_lookup_table_entry *lte,
121 struct filter_context *ctx)
123 return stream_filtered(lte, ctx) < 0;
127 may_soft_filter_streams(const struct filter_context *ctx)
131 return ctx->write_flags & WIMLIB_WRITE_FLAG_OVERWRITE;
135 may_hard_filter_streams(const struct filter_context *ctx)
139 return ctx->write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS;
143 may_filter_streams(const struct filter_context *ctx)
145 return (may_soft_filter_streams(ctx) ||
146 may_hard_filter_streams(ctx));
150 /* Return true if the specified resource is compressed and the compressed data
151 * can be reused with the specified output parameters. */
153 can_raw_copy(const struct wim_lookup_table_entry *lte,
154 int write_resource_flags, int out_ctype, u32 out_chunk_size)
156 const struct wim_resource_spec *rspec;
158 if (write_resource_flags & WRITE_RESOURCE_FLAG_RECOMPRESS)
161 if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE)
164 if (lte->resource_location != RESOURCE_IN_WIM)
169 if (rspec->is_pipable != !!(write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE))
172 if (rspec->flags & WIM_RESHDR_FLAG_COMPRESSED) {
173 /* Normal compressed resource: Must use same compression type
175 return (rspec->compression_type == out_ctype &&
176 rspec->chunk_size == out_chunk_size);
179 if ((rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) &&
180 (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
182 /* Packed resource: Such resources may contain multiple streams,
183 * and in general only a subset of them need to be written. As
184 * a heuristic, re-use the raw data if more than two-thirds the
185 * uncompressed size is being written. */
187 /* Note: packed resources contain a header that specifies the
188 * compression type and chunk size; therefore we don't need to
189 * check if they are compatible with @out_ctype and
190 * @out_chunk_size. */
192 struct wim_lookup_table_entry *res_stream;
195 list_for_each_entry(res_stream, &rspec->stream_list, rspec_node)
196 if (res_stream->will_be_in_output_wim)
197 write_size += res_stream->size;
199 return (write_size > rspec->uncompressed_size * 2 / 3);
206 filter_resource_flags(u8 flags)
208 return (flags & ~(WIM_RESHDR_FLAG_PACKED_STREAMS |
209 WIM_RESHDR_FLAG_COMPRESSED |
210 WIM_RESHDR_FLAG_SPANNED |
211 WIM_RESHDR_FLAG_FREE));
215 stream_set_out_reshdr_for_reuse(struct wim_lookup_table_entry *lte)
217 const struct wim_resource_spec *rspec;
219 wimlib_assert(lte->resource_location == RESOURCE_IN_WIM);
222 if (rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
224 wimlib_assert(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS);
226 lte->out_reshdr.offset_in_wim = lte->offset_in_res;
227 lte->out_reshdr.uncompressed_size = 0;
228 lte->out_reshdr.size_in_wim = lte->size;
230 lte->out_res_offset_in_wim = rspec->offset_in_wim;
231 lte->out_res_size_in_wim = rspec->size_in_wim;
232 lte->out_res_uncompressed_size = rspec->uncompressed_size;
234 wimlib_assert(!(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS));
236 lte->out_reshdr.offset_in_wim = rspec->offset_in_wim;
237 lte->out_reshdr.uncompressed_size = rspec->uncompressed_size;
238 lte->out_reshdr.size_in_wim = rspec->size_in_wim;
240 lte->out_reshdr.flags = lte->flags;
244 /* Write the header for a stream in a pipable WIM. */
246 write_pwm_stream_header(const struct wim_lookup_table_entry *lte,
247 struct filedes *out_fd,
248 int additional_reshdr_flags)
250 struct pwm_stream_hdr stream_hdr;
254 stream_hdr.magic = cpu_to_le64(PWM_STREAM_MAGIC);
255 stream_hdr.uncompressed_size = cpu_to_le64(lte->size);
256 if (additional_reshdr_flags & PWM_RESHDR_FLAG_UNHASHED) {
257 zero_out_hash(stream_hdr.hash);
259 wimlib_assert(!lte->unhashed);
260 copy_hash(stream_hdr.hash, lte->hash);
263 reshdr_flags = filter_resource_flags(lte->flags);
264 reshdr_flags |= additional_reshdr_flags;
265 stream_hdr.flags = cpu_to_le32(reshdr_flags);
266 ret = full_write(out_fd, &stream_hdr, sizeof(stream_hdr));
268 ERROR_WITH_ERRNO("Write error");
272 struct write_streams_progress_data {
273 wimlib_progress_func_t progress_func;
274 union wimlib_progress_info progress;
275 uint64_t next_progress;
279 do_write_streams_progress(struct write_streams_progress_data *progress_data,
280 struct wim_lookup_table_entry *cur_stream,
285 union wimlib_progress_info *progress = &progress_data->progress;
288 progress->write_streams.total_bytes -= complete_size;
289 progress->write_streams.total_streams -= complete_count;
290 if (progress_data->next_progress != ~(uint64_t)0 &&
291 progress_data->next_progress > progress->write_streams.total_bytes)
293 progress_data->next_progress = progress->write_streams.total_bytes;
296 progress->write_streams.completed_bytes += complete_size;
297 progress->write_streams.completed_streams += complete_count;
300 if (progress_data->progress_func
301 && (progress->write_streams.completed_bytes >= progress_data->next_progress))
303 progress_data->progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
305 if (progress_data->next_progress == progress->write_streams.total_bytes) {
306 progress_data->next_progress = ~(uint64_t)0;
308 progress_data->next_progress =
309 min(progress->write_streams.total_bytes,
310 progress->write_streams.completed_bytes +
311 progress->write_streams.total_bytes / 100);
316 struct write_streams_ctx {
317 /* File descriptor the streams are being written to. */
318 struct filedes *out_fd;
320 /* Lookup table for the WIMStruct on whose behalf the streams are being
322 struct wim_lookup_table *lookup_table;
324 /* Compression format to use. */
327 /* Maximum uncompressed chunk size in compressed resources to use. */
330 /* Flags that affect how the streams will be written. */
331 int write_resource_flags;
333 /* Data used for issuing WRITE_STREAMS progress. */
334 struct write_streams_progress_data progress_data;
336 struct filter_context *filter_ctx;
338 /* Upper bound on the total number of bytes that need to be compressed.
340 u64 num_bytes_to_compress;
342 /* Pointer to the chunk_compressor implementation being used for
343 * compressing chunks of data, or NULL if chunks are being written
345 struct chunk_compressor *compressor;
347 /* Buffer for dividing the read data into chunks of size
348 * @out_chunk_size. */
351 /* Number of bytes in @chunk_buf that are currently filled. */
352 size_t chunk_buf_filled;
354 /* List of streams that currently have chunks being compressed. */
355 struct list_head pending_streams;
357 /* List of streams in the resource pack. Streams are moved here after
358 * @pending_streams only when writing a packed resource. */
359 struct list_head pack_streams;
361 /* Set to true if the stream currently being read was a duplicate, and
362 * therefore the corresponding stream entry needs to be freed once the
363 * read finishes. (In this case we add the duplicate entry to
364 * pending_streams rather than the entry being read.) */
365 bool stream_was_duplicate;
367 /* Current uncompressed offset in the stream being read. */
368 u64 cur_read_stream_offset;
370 /* Uncompressed size of the stream currently being read. */
371 u64 cur_read_stream_size;
373 /* Current uncompressed offset in the stream being written. */
374 u64 cur_write_stream_offset;
376 /* Uncompressed size of resource currently being written. */
377 u64 cur_write_res_size;
379 /* Array that is filled in with compressed chunk sizes as a resource is
383 /* Index of next entry in @chunk_csizes to fill in. */
386 /* Number of entries in @chunk_csizes currently allocated. */
387 size_t num_alloc_chunks;
389 /* Offset in the output file of the start of the chunks of the resource
390 * currently being written. */
391 u64 chunks_start_offset;
394 /* Reserve space for the chunk table and prepare to accumulate the chunk table
397 begin_chunk_table(struct write_streams_ctx *ctx, u64 res_expected_size)
399 u64 expected_num_chunks;
400 u64 expected_num_chunk_entries;
404 /* Calculate the number of chunks and chunk entries that should be
405 * needed for the resource. These normally will be the final values,
406 * but in PACKED_STREAMS mode some of the streams we're planning to
407 * write into the resource may be duplicates, and therefore discarded,
408 * potentially decreasing the number of chunk entries needed. */
409 expected_num_chunks = DIV_ROUND_UP(res_expected_size, ctx->out_chunk_size);
410 expected_num_chunk_entries = expected_num_chunks;
411 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
412 expected_num_chunk_entries--;
414 /* Make sure the chunk_csizes array is long enough to store the
415 * compressed size of each chunk. */
416 if (expected_num_chunks > ctx->num_alloc_chunks) {
417 u64 new_length = expected_num_chunks + 50;
419 if ((size_t)new_length != new_length) {
420 ERROR("Resource size too large (%"PRIu64" bytes!",
422 return WIMLIB_ERR_NOMEM;
425 FREE(ctx->chunk_csizes);
426 ctx->chunk_csizes = MALLOC(new_length * sizeof(ctx->chunk_csizes[0]));
427 if (ctx->chunk_csizes == NULL) {
428 ctx->num_alloc_chunks = 0;
429 return WIMLIB_ERR_NOMEM;
431 ctx->num_alloc_chunks = new_length;
434 ctx->chunk_index = 0;
436 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE)) {
437 /* Reserve space for the chunk table in the output file. In the
438 * case of packed resources this reserves the upper bound for
439 * the needed space, not necessarily the exact space which will
440 * prove to be needed. At this point, we just use @chunk_csizes
441 * for a buffer of 0's because the actual compressed chunk sizes
443 reserve_size = expected_num_chunk_entries *
444 get_chunk_entry_size(res_expected_size,
445 0 != (ctx->write_resource_flags &
446 WIM_RESHDR_FLAG_PACKED_STREAMS));
447 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)
448 reserve_size += sizeof(struct alt_chunk_table_header_disk);
449 memset(ctx->chunk_csizes, 0, reserve_size);
450 ret = full_write(ctx->out_fd, ctx->chunk_csizes, reserve_size);
458 begin_write_resource(struct write_streams_ctx *ctx, u64 res_expected_size)
462 wimlib_assert(res_expected_size != 0);
464 if (ctx->compressor != NULL) {
465 ret = begin_chunk_table(ctx, res_expected_size);
470 /* Output file descriptor is now positioned at the offset at which to
471 * write the first chunk of the resource. */
472 ctx->chunks_start_offset = ctx->out_fd->offset;
473 ctx->cur_write_stream_offset = 0;
474 ctx->cur_write_res_size = res_expected_size;
479 end_chunk_table(struct write_streams_ctx *ctx, u64 res_actual_size,
480 u64 *res_start_offset_ret, u64 *res_store_size_ret)
482 size_t actual_num_chunks;
483 size_t actual_num_chunk_entries;
484 size_t chunk_entry_size;
487 actual_num_chunks = ctx->chunk_index;
488 actual_num_chunk_entries = actual_num_chunks;
489 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
490 actual_num_chunk_entries--;
492 chunk_entry_size = get_chunk_entry_size(res_actual_size,
493 0 != (ctx->write_resource_flags &
494 WIM_RESHDR_FLAG_PACKED_STREAMS));
496 typedef le64 __attribute__((may_alias)) aliased_le64_t;
497 typedef le32 __attribute__((may_alias)) aliased_le32_t;
499 if (chunk_entry_size == 4) {
500 aliased_le32_t *entries = (aliased_le32_t*)ctx->chunk_csizes;
502 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
503 for (size_t i = 0; i < actual_num_chunk_entries; i++)
504 entries[i] = cpu_to_le32(ctx->chunk_csizes[i]);
506 u32 offset = ctx->chunk_csizes[0];
507 for (size_t i = 0; i < actual_num_chunk_entries; i++) {
508 u32 next_size = ctx->chunk_csizes[i + 1];
509 entries[i] = cpu_to_le32(offset);
514 aliased_le64_t *entries = (aliased_le64_t*)ctx->chunk_csizes;
516 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
517 for (size_t i = 0; i < actual_num_chunk_entries; i++)
518 entries[i] = cpu_to_le64(ctx->chunk_csizes[i]);
520 u64 offset = ctx->chunk_csizes[0];
521 for (size_t i = 0; i < actual_num_chunk_entries; i++) {
522 u64 next_size = ctx->chunk_csizes[i + 1];
523 entries[i] = cpu_to_le64(offset);
529 size_t chunk_table_size = actual_num_chunk_entries * chunk_entry_size;
530 u64 res_start_offset;
533 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
534 ret = full_write(ctx->out_fd, ctx->chunk_csizes, chunk_table_size);
537 res_end_offset = ctx->out_fd->offset;
538 res_start_offset = ctx->chunks_start_offset;
540 res_end_offset = ctx->out_fd->offset;
542 u64 chunk_table_offset;
544 chunk_table_offset = ctx->chunks_start_offset - chunk_table_size;
546 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
547 struct alt_chunk_table_header_disk hdr;
549 hdr.res_usize = cpu_to_le64(res_actual_size);
550 hdr.chunk_size = cpu_to_le32(ctx->out_chunk_size);
551 hdr.compression_format = cpu_to_le32(ctx->out_ctype);
553 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 1);
554 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 2);
555 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3);
557 ret = full_pwrite(ctx->out_fd, &hdr, sizeof(hdr),
558 chunk_table_offset - sizeof(hdr));
561 res_start_offset = chunk_table_offset - sizeof(hdr);
563 res_start_offset = chunk_table_offset;
566 ret = full_pwrite(ctx->out_fd, ctx->chunk_csizes,
567 chunk_table_size, chunk_table_offset);
572 *res_start_offset_ret = res_start_offset;
573 *res_store_size_ret = res_end_offset - res_start_offset;
578 ERROR_WITH_ERRNO("Write error");
582 /* Finish writing a WIM resource by writing or updating the chunk table (if not
583 * writing the data uncompressed) and loading its metadata into @out_reshdr. */
585 end_write_resource(struct write_streams_ctx *ctx, struct wim_reshdr *out_reshdr)
589 u64 res_uncompressed_size;
590 u64 res_offset_in_wim;
592 wimlib_assert(ctx->cur_write_stream_offset == ctx->cur_write_res_size ||
593 (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS));
594 res_uncompressed_size = ctx->cur_write_res_size;
596 if (ctx->compressor) {
597 ret = end_chunk_table(ctx, res_uncompressed_size,
598 &res_offset_in_wim, &res_size_in_wim);
602 res_offset_in_wim = ctx->chunks_start_offset;
603 res_size_in_wim = ctx->out_fd->offset - res_offset_in_wim;
605 out_reshdr->uncompressed_size = res_uncompressed_size;
606 out_reshdr->size_in_wim = res_size_in_wim;
607 out_reshdr->offset_in_wim = res_offset_in_wim;
608 DEBUG("Finished writing resource: %"PRIu64" => %"PRIu64" @ %"PRIu64"",
609 res_uncompressed_size, res_size_in_wim, res_offset_in_wim);
613 /* Begin processing a stream for writing. */
615 write_stream_begin_read(struct wim_lookup_table_entry *lte,
616 u32 flags, void *_ctx)
618 struct write_streams_ctx *ctx = _ctx;
621 wimlib_assert(lte->size > 0);
623 ctx->cur_read_stream_offset = 0;
624 ctx->cur_read_stream_size = lte->size;
626 /* As an optimization, we allow some streams to be "unhashed", meaning
627 * their SHA1 message digests are unknown. This is the case with
628 * streams that are added by scanning a directry tree with
629 * wimlib_add_image(), for example. Since WIM uses single-instance
630 * streams, we don't know whether such each such stream really need to
631 * written until it is actually checksummed, unless it has a unique
632 * size. In such cases we read and checksum the stream in this
633 * function, thereby advancing ahead of read_stream_list(), which will
634 * still provide the data again to write_stream_process_chunk(). This
635 * is okay because an unhashed stream cannot be in a WIM resource, which
636 * might be costly to decompress. */
637 ctx->stream_was_duplicate = false;
638 if (ctx->lookup_table != NULL && lte->unhashed && !lte->unique_size) {
640 wimlib_assert(!(flags & BEGIN_STREAM_FLAG_PARTIAL_RESOURCE));
642 struct wim_lookup_table_entry *lte_new;
644 ret = hash_unhashed_stream(lte, ctx->lookup_table, <e_new);
647 if (lte_new != lte) {
648 /* Duplicate stream detected. */
650 if (lte_new->will_be_in_output_wim ||
651 stream_filtered(lte_new, ctx->filter_ctx))
653 /* The duplicate stream is already being
654 * included in the output WIM, or it would be
655 * filtered out if it had been. Skip writing
656 * this stream (and reading it again) entirely,
657 * passing its output reference count to the
658 * duplicate stream in the former case. */
659 DEBUG("Discarding duplicate stream of "
660 "length %"PRIu64, lte->size);
661 do_write_streams_progress(&ctx->progress_data,
664 list_del(<e->write_streams_list);
665 list_del(<e->lookup_table_list);
666 if (lte_new->will_be_in_output_wim)
667 lte_new->out_refcnt += lte->out_refcnt;
668 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)
669 ctx->cur_write_res_size -= lte->size;
670 free_lookup_table_entry(lte);
671 return BEGIN_STREAM_STATUS_SKIP_STREAM;
673 /* The duplicate stream can validly be written,
674 * but was not marked as such. Discard the
675 * current stream entry and use the duplicate,
676 * but actually freeing the current entry must
677 * wait until read_stream_list() has finished
678 * reading its data. */
679 DEBUG("Stream duplicate, but not already "
680 "selected for writing.");
681 list_replace(<e->write_streams_list,
682 <e_new->write_streams_list);
683 list_replace(<e->lookup_table_list,
684 <e_new->lookup_table_list);
685 lte_new->out_refcnt = lte->out_refcnt;
686 lte_new->will_be_in_output_wim = 1;
687 ctx->stream_was_duplicate = true;
692 list_move_tail(<e->write_streams_list, &ctx->pending_streams);
696 /* Rewrite a stream that was just written compressed as uncompressed instead.
697 * This function is optional, but if a stream did not compress to less than its
698 * original size, it might as well be written uncompressed. */
700 write_stream_uncompressed(struct wim_lookup_table_entry *lte,
701 struct filedes *out_fd)
704 u64 begin_offset = lte->out_reshdr.offset_in_wim;
705 u64 end_offset = out_fd->offset;
707 if (filedes_seek(out_fd, begin_offset) == -1)
710 ret = extract_full_stream_to_fd(lte, out_fd);
712 /* Error reading the uncompressed data. */
713 if (out_fd->offset == begin_offset &&
714 filedes_seek(out_fd, end_offset) != -1)
716 /* Nothing was actually written yet, and we successfully
717 * seeked to the end of the compressed resource, so
718 * don't issue a hard error; just keep the compressed
719 * resource instead. */
720 WARNING("Recovered compressed stream of "
721 "size %"PRIu64", continuing on.",
728 wimlib_assert(out_fd->offset - begin_offset == lte->size);
730 if (out_fd->offset < end_offset &&
731 0 != ftruncate(out_fd->fd, out_fd->offset))
733 ERROR_WITH_ERRNO("Can't truncate output file to "
734 "offset %"PRIu64, out_fd->offset);
735 return WIMLIB_ERR_WRITE;
738 lte->out_reshdr.size_in_wim = lte->size;
739 lte->out_reshdr.flags &= ~(WIM_RESHDR_FLAG_COMPRESSED |
740 WIM_RESHDR_FLAG_PACKED_STREAMS);
744 /* Write the next chunk of (typically compressed) data to the output WIM,
745 * handling the writing of the chunk table. */
747 write_chunk(struct write_streams_ctx *ctx, const void *cchunk,
748 size_t csize, size_t usize)
752 struct wim_lookup_table_entry *lte;
753 u32 completed_stream_count;
756 lte = list_entry(ctx->pending_streams.next,
757 struct wim_lookup_table_entry, write_streams_list);
759 if (ctx->cur_write_stream_offset == 0 &&
760 !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
762 /* Starting to write a new stream in non-packed mode. */
764 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
765 int additional_reshdr_flags = 0;
766 if (ctx->compressor != NULL)
767 additional_reshdr_flags |= WIM_RESHDR_FLAG_COMPRESSED;
769 DEBUG("Writing pipable WIM stream header "
770 "(offset=%"PRIu64")", ctx->out_fd->offset);
772 ret = write_pwm_stream_header(lte, ctx->out_fd,
773 additional_reshdr_flags);
778 ret = begin_write_resource(ctx, lte->size);
783 if (ctx->compressor != NULL) {
784 /* Record the compresed chunk size. */
785 wimlib_assert(ctx->chunk_index < ctx->num_alloc_chunks);
786 ctx->chunk_csizes[ctx->chunk_index++] = csize;
788 /* If writing a pipable WIM, before the chunk data write a chunk
789 * header that provides the compressed chunk size. */
790 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
791 struct pwm_chunk_hdr chunk_hdr = {
792 .compressed_size = cpu_to_le32(csize),
794 ret = full_write(ctx->out_fd, &chunk_hdr,
801 /* Write the chunk data. */
802 ret = full_write(ctx->out_fd, cchunk, csize);
806 ctx->cur_write_stream_offset += usize;
808 completed_size = usize;
809 completed_stream_count = 0;
810 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
811 /* Wrote chunk in packed mode. It may have finished multiple
813 while (ctx->cur_write_stream_offset > lte->size) {
814 struct wim_lookup_table_entry *next;
816 ctx->cur_write_stream_offset -= lte->size;
818 wimlib_assert(!list_is_singular(&ctx->pending_streams) &&
819 !list_empty(&ctx->pending_streams));
820 next = list_entry(lte->write_streams_list.next,
821 struct wim_lookup_table_entry,
823 list_move_tail(<e->write_streams_list,
826 completed_stream_count++;
828 if (ctx->cur_write_stream_offset == lte->size) {
829 ctx->cur_write_stream_offset = 0;
830 list_move_tail(<e->write_streams_list,
832 completed_stream_count++;
835 /* Wrote chunk in non-packed mode. It may have finished a
837 if (ctx->cur_write_stream_offset == lte->size) {
839 completed_stream_count++;
841 list_del(<e->write_streams_list);
843 wimlib_assert(ctx->cur_write_stream_offset ==
844 ctx->cur_write_res_size);
846 ret = end_write_resource(ctx, <e->out_reshdr);
850 lte->out_reshdr.flags = filter_resource_flags(lte->flags);
851 if (ctx->compressor != NULL)
852 lte->out_reshdr.flags |= WIM_RESHDR_FLAG_COMPRESSED;
854 if (ctx->compressor != NULL &&
855 lte->out_reshdr.size_in_wim >= lte->out_reshdr.uncompressed_size &&
856 !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) &&
857 !(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS))
859 /* Stream did not compress to less than its original
860 * size. If we're not writing a pipable WIM (which
861 * could mean the output file descriptor is
862 * non-seekable), and the stream isn't located in a
863 * resource pack (which would make reading it again
864 * costly), truncate the file to the start of the stream
865 * and write it uncompressed instead. */
866 DEBUG("Stream of size %"PRIu64" did not compress to "
867 "less than original size; writing uncompressed.",
869 ret = write_stream_uncompressed(lte, ctx->out_fd);
873 wimlib_assert(lte->out_reshdr.uncompressed_size == lte->size);
875 ctx->cur_write_stream_offset = 0;
879 do_write_streams_progress(&ctx->progress_data, lte,
880 completed_size, completed_stream_count,
886 ERROR_WITH_ERRNO("Write error");
891 submit_chunk_for_compression(struct write_streams_ctx *ctx,
892 const void *chunk, size_t size)
894 /* While we are unable to submit the chunk for compression (due to too
895 * many chunks already outstanding), retrieve and write the next
896 * compressed chunk. */
897 while (!ctx->compressor->submit_chunk(ctx->compressor, chunk, size)) {
904 bret = ctx->compressor->get_chunk(ctx->compressor,
905 &cchunk, &csize, &usize);
909 ret = write_chunk(ctx, cchunk, csize, usize);
916 /* Process the next chunk of data to be written to a WIM resource. */
918 write_stream_process_chunk(const void *chunk, size_t size, void *_ctx)
920 struct write_streams_ctx *ctx = _ctx;
922 const u8 *chunkptr, *chunkend;
924 wimlib_assert(size != 0);
926 if (ctx->compressor == NULL) {
927 /* Write chunk uncompressed. */
928 ret = write_chunk(ctx, chunk, size, size);
931 ctx->cur_read_stream_offset += size;
935 /* Submit the chunk for compression, but take into account that the
936 * @size the chunk was provided in may not correspond to the
937 * @out_chunk_size being used for compression. */
939 chunkend = chunkptr + size;
941 const u8 *resized_chunk;
942 size_t needed_chunk_size;
944 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
945 needed_chunk_size = ctx->out_chunk_size;
947 u64 res_bytes_remaining;
949 res_bytes_remaining = ctx->cur_read_stream_size -
950 ctx->cur_read_stream_offset;
951 needed_chunk_size = min(ctx->out_chunk_size,
952 ctx->chunk_buf_filled +
953 res_bytes_remaining);
956 if (ctx->chunk_buf_filled == 0 &&
957 chunkend - chunkptr >= needed_chunk_size)
959 /* No intermediate buffering needed. */
960 resized_chunk = chunkptr;
961 chunkptr += needed_chunk_size;
962 ctx->cur_read_stream_offset += needed_chunk_size;
964 /* Intermediate buffering needed. */
965 size_t bytes_consumed;
967 bytes_consumed = min(chunkend - chunkptr,
968 needed_chunk_size - ctx->chunk_buf_filled);
970 memcpy(&ctx->chunk_buf[ctx->chunk_buf_filled],
971 chunkptr, bytes_consumed);
973 chunkptr += bytes_consumed;
974 ctx->cur_read_stream_offset += bytes_consumed;
975 ctx->chunk_buf_filled += bytes_consumed;
976 if (ctx->chunk_buf_filled == needed_chunk_size) {
977 resized_chunk = ctx->chunk_buf;
978 ctx->chunk_buf_filled = 0;
985 ret = submit_chunk_for_compression(ctx, resized_chunk,
990 } while (chunkptr != chunkend);
994 /* Finish processing a stream for writing. It may not have been completely
995 * written yet, as the chunk_compressor implementation may still have chunks
996 * buffered or being compressed. */
998 write_stream_end_read(struct wim_lookup_table_entry *lte, int status, void *_ctx)
1000 struct write_streams_ctx *ctx = _ctx;
1002 wimlib_assert(ctx->cur_read_stream_offset == ctx->cur_read_stream_size);
1003 if (ctx->stream_was_duplicate) {
1004 free_lookup_table_entry(lte);
1005 } else if (lte->unhashed && ctx->lookup_table != NULL) {
1006 list_del(<e->unhashed_list);
1007 lookup_table_insert(ctx->lookup_table, lte);
1013 /* Compute statistics about a list of streams that will be written.
1015 * Assumes the streams are sorted such that all streams located in each distinct
1016 * WIM (specified by WIMStruct) are together. */
1018 compute_stream_list_stats(struct list_head *stream_list,
1019 struct write_streams_ctx *ctx)
1021 struct wim_lookup_table_entry *lte;
1022 u64 total_bytes = 0;
1023 u64 num_streams = 0;
1024 u64 total_parts = 0;
1025 WIMStruct *prev_wim_part = NULL;
1027 list_for_each_entry(lte, stream_list, write_streams_list) {
1029 total_bytes += lte->size;
1030 if (lte->resource_location == RESOURCE_IN_WIM) {
1031 if (prev_wim_part != lte->rspec->wim) {
1032 prev_wim_part = lte->rspec->wim;
1037 ctx->progress_data.progress.write_streams.total_bytes = total_bytes;
1038 ctx->progress_data.progress.write_streams.total_streams = num_streams;
1039 ctx->progress_data.progress.write_streams.completed_bytes = 0;
1040 ctx->progress_data.progress.write_streams.completed_streams = 0;
1041 ctx->progress_data.progress.write_streams.compression_type = ctx->out_ctype;
1042 ctx->progress_data.progress.write_streams.total_parts = total_parts;
1043 ctx->progress_data.progress.write_streams.completed_parts = 0;
1044 ctx->progress_data.next_progress = 0;
1047 /* Find streams in @stream_list that can be copied to the output WIM in raw form
1048 * rather than compressed. Delete these streams from @stream_list and move them
1049 * to @raw_copy_streams. Return the total uncompressed size of the streams that
1050 * need to be compressed. */
1052 find_raw_copy_streams(struct list_head *stream_list,
1053 int write_resource_flags,
1056 struct list_head *raw_copy_streams)
1058 struct wim_lookup_table_entry *lte, *tmp;
1059 u64 num_bytes_to_compress = 0;
1061 INIT_LIST_HEAD(raw_copy_streams);
1063 /* Initialize temporary raw_copy_ok flag. */
1064 list_for_each_entry(lte, stream_list, write_streams_list)
1065 if (lte->resource_location == RESOURCE_IN_WIM)
1066 lte->rspec->raw_copy_ok = 0;
1068 list_for_each_entry_safe(lte, tmp, stream_list, write_streams_list) {
1069 if (lte->resource_location == RESOURCE_IN_WIM &&
1070 lte->rspec->raw_copy_ok)
1072 list_move_tail(<e->write_streams_list,
1074 } else if (can_raw_copy(lte, write_resource_flags,
1075 out_ctype, out_chunk_size))
1077 lte->rspec->raw_copy_ok = 1;
1078 list_move_tail(<e->write_streams_list,
1081 num_bytes_to_compress += lte->size;
1085 return num_bytes_to_compress;
1088 /* Copy a raw compressed resource located in another WIM file to the WIM file
1091 write_raw_copy_resource(struct wim_resource_spec *in_rspec,
1092 struct filedes *out_fd)
1094 u64 cur_read_offset;
1095 u64 end_read_offset;
1096 u8 buf[BUFFER_SIZE];
1097 size_t bytes_to_read;
1099 struct filedes *in_fd;
1100 struct wim_lookup_table_entry *lte;
1101 u64 out_offset_in_wim;
1103 DEBUG("Copying raw compressed data (size_in_wim=%"PRIu64", "
1104 "uncompressed_size=%"PRIu64")",
1105 in_rspec->size_in_wim, in_rspec->uncompressed_size);
1107 /* Copy the raw data. */
1108 cur_read_offset = in_rspec->offset_in_wim;
1109 end_read_offset = cur_read_offset + in_rspec->size_in_wim;
1111 out_offset_in_wim = out_fd->offset;
1113 if (in_rspec->is_pipable) {
1114 if (cur_read_offset < sizeof(struct pwm_stream_hdr))
1115 return WIMLIB_ERR_INVALID_PIPABLE_WIM;
1116 cur_read_offset -= sizeof(struct pwm_stream_hdr);
1117 out_offset_in_wim += sizeof(struct pwm_stream_hdr);
1119 in_fd = &in_rspec->wim->in_fd;
1120 wimlib_assert(cur_read_offset != end_read_offset);
1123 bytes_to_read = min(sizeof(buf), end_read_offset - cur_read_offset);
1125 ret = full_pread(in_fd, buf, bytes_to_read, cur_read_offset);
1129 ret = full_write(out_fd, buf, bytes_to_read);
1133 cur_read_offset += bytes_to_read;
1135 } while (cur_read_offset != end_read_offset);
1137 list_for_each_entry(lte, &in_rspec->stream_list, rspec_node) {
1138 if (lte->will_be_in_output_wim) {
1139 stream_set_out_reshdr_for_reuse(lte);
1140 if (in_rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS)
1141 lte->out_res_offset_in_wim = out_offset_in_wim;
1143 lte->out_reshdr.offset_in_wim = out_offset_in_wim;
1150 /* Copy a list of raw compressed resources located in other WIM file(s) to the
1151 * WIM file being written. */
1153 write_raw_copy_resources(struct list_head *raw_copy_streams,
1154 struct filedes *out_fd,
1155 struct write_streams_progress_data *progress_data)
1157 struct wim_lookup_table_entry *lte;
1160 list_for_each_entry(lte, raw_copy_streams, write_streams_list)
1161 lte->rspec->raw_copy_ok = 1;
1163 list_for_each_entry(lte, raw_copy_streams, write_streams_list) {
1164 if (lte->rspec->raw_copy_ok) {
1165 /* Write each packed resource only one time, no matter
1166 * how many streams reference it. */
1167 ret = write_raw_copy_resource(lte->rspec, out_fd);
1170 lte->rspec->raw_copy_ok = 0;
1172 do_write_streams_progress(progress_data, lte, lte->size,
1178 /* Wait for and write all chunks pending in the compressor. */
1180 finish_remaining_chunks(struct write_streams_ctx *ctx)
1187 if (ctx->compressor == NULL)
1190 if (ctx->chunk_buf_filled != 0) {
1191 ret = submit_chunk_for_compression(ctx, ctx->chunk_buf,
1192 ctx->chunk_buf_filled);
1197 while (ctx->compressor->get_chunk(ctx->compressor, &cdata, &csize, &usize)) {
1198 ret = write_chunk(ctx, cdata, csize, usize);
1206 remove_zero_length_streams(struct list_head *stream_list)
1208 struct wim_lookup_table_entry *lte, *tmp;
1210 list_for_each_entry_safe(lte, tmp, stream_list, write_streams_list) {
1211 wimlib_assert(lte->will_be_in_output_wim);
1212 if (lte->size == 0) {
1213 list_del(<e->write_streams_list);
1214 lte->out_reshdr.offset_in_wim = 0;
1215 lte->out_reshdr.size_in_wim = 0;
1216 lte->out_reshdr.uncompressed_size = 0;
1217 lte->out_reshdr.flags = filter_resource_flags(lte->flags);
1223 * Write a list of streams to the output WIM file.
1226 * The list of streams to write, specified by a list of `struct
1227 * wim_lookup_table_entry's linked by the 'write_streams_list' member.
1230 * The file descriptor, opened for writing, to which to write the streams.
1232 * @write_resource_flags
1233 * Flags to modify how the streams are written:
1235 * WRITE_RESOURCE_FLAG_RECOMPRESS:
1236 * Force compression of all resources, even if they could otherwise
1237 * be re-used by copying the raw data, due to being located in a WIM
1238 * file with compatible compression parameters.
1240 * WRITE_RESOURCE_FLAG_PIPABLE:
1241 * Write the resources in the wimlib-specific pipable format, and
1242 * furthermore do so in such a way that no seeking backwards in
1243 * @out_fd will be performed (so it may be a pipe).
1245 * WRITE_RESOURCE_FLAG_PACK_STREAMS:
1246 * Pack all the streams into a single resource rather than writing
1247 * them in separate resources. This flag is only valid if the WIM
1248 * version number has been, or will be, set to
1249 * WIM_VERSION_PACKED_STREAMS. This flag may not be combined with
1250 * WRITE_RESOURCE_FLAG_PIPABLE.
1253 * Compression format to use to write the output streams, specified as one
1254 * of the WIMLIB_COMPRESSION_TYPE_* constants, excepting
1255 * WIMLIB_COMPRESSION_TYPE_INVALID but including
1256 * WIMLIB_COMPRESSION_TYPE_NONE.
1259 * Chunk size to use to write the streams. It must be a valid chunk size
1260 * for the specified compression format @out_ctype, unless @out_ctype is
1261 * WIMLIB_COMPRESSION_TYPE_NONE, in which case this parameter is ignored.
1264 * Number of threads to use to compress data. If 0, a default number of
1265 * threads will be chosen. The number of threads still may be decreased
1266 * from the specified value if insufficient memory is detected.
1269 * If on-the-fly deduplication of unhashed streams is desired, this
1270 * parameter must be pointer to the lookup table for the WIMStruct on whose
1271 * behalf the streams are being written. Otherwise, this parameter can be
1275 * If on-the-fly deduplication of unhashed streams is desired, this
1276 * parameter can be a pointer to a context for stream filtering used to
1277 * detect whether the duplicate stream has been hard-filtered or not. If
1278 * no streams are hard-filtered or no streams are unhashed, this parameter
1282 * If non-NULL, a progress function that will be called periodically with
1283 * WIMLIB_PROGRESS_MSG_WRITE_STREAMS messages. Note that on-the-fly
1284 * deduplication of unhashed streams may result in the total bytes provided
1285 * in the progress data to decrease from one message to the next.
1287 * This function will write the streams in @stream_list to resources in
1288 * consecutive positions in the output WIM file, or to a single packed resource
1289 * if WRITE_RESOURCE_FLAG_PACK_STREAMS was specified in @write_resource_flags.
1290 * In both cases, the @out_reshdr of the `struct wim_lookup_table_entry' for
1291 * each stream written will be updated to specify its location, size, and flags
1292 * in the output WIM. In the packed resource case,
1293 * WIM_RESHDR_FLAG_PACKED_STREAMS will be set in the @flags field of each
1294 * @out_reshdr, and furthermore @out_res_offset_in_wim and @out_res_size_in_wim
1295 * of each @out_reshdr will be set to the offset and size, respectively, in the
1296 * output WIM of the packed resource containing the corresponding stream.
1298 * Each of the streams to write may be in any location supported by the
1299 * resource-handling code (specifically, read_stream_list()), such as the
1300 * contents of external file that has been logically added to the output WIM, or
1301 * a stream in another WIM file that has been imported, or even a stream in the
1302 * "same" WIM file of which a modified copy is being written. In the case that
1303 * a stream is already in a WIM file and uses compatible compression parameters,
1304 * by default this function will re-use the raw data instead of decompressing
1305 * it, then recompressing it; however, with WRITE_RESOURCE_FLAG_RECOMPRESS
1306 * specified in @write_resource_flags, this is not done.
1308 * As a further requirement, this function requires that the
1309 * @will_be_in_output_wim member be set to 1 on all streams in @stream_list as
1310 * well as any other streams not in @stream_list that will be in the output WIM
1311 * file, but set to 0 on any other streams in the output WIM's lookup table or
1312 * sharing a packed resource with a stream in @stream_list. Still furthermore,
1313 * if on-the-fly deduplication of streams is possible, then all streams in
1314 * @stream_list must also be linked by @lookup_table_list along with any other
1315 * streams that have @will_be_in_output_wim set.
1317 * This function handles on-the-fly deduplication of streams for which SHA1
1318 * message digests have not yet been calculated. Such streams may or may not
1319 * need to be written. If @lookup_table is non-NULL, then each stream in
1320 * @stream_list that has @unhashed set but not @unique_size set is checksummed
1321 * immediately before it would otherwise be read for writing in order to
1322 * determine if it is identical to another stream already being written or one
1323 * that would be filtered out of the output WIM using stream_filtered() with the
1324 * context @filter_ctx. Each such duplicate stream will be removed from
1325 * @stream_list, its reference count transfered to the pre-existing duplicate
1326 * stream, its memory freed, and will not be written. Alternatively, if a
1327 * stream in @stream_list is a duplicate with any stream in @lookup_table that
1328 * has not been marked for writing or would not be hard-filtered, it is freed
1329 * and the pre-existing duplicate is written instead, taking ownership of the
1330 * reference count and slot in the @lookup_table_list.
1332 * Returns 0 if every stream was either written successfully or did not need to
1333 * be written; otherwise returns a non-zero error code.
1336 write_stream_list(struct list_head *stream_list,
1337 struct filedes *out_fd,
1338 int write_resource_flags,
1341 unsigned num_threads,
1342 struct wim_lookup_table *lookup_table,
1343 struct filter_context *filter_ctx,
1344 wimlib_progress_func_t progress_func)
1347 struct write_streams_ctx ctx;
1348 struct list_head raw_copy_streams;
1350 wimlib_assert((write_resource_flags &
1351 (WRITE_RESOURCE_FLAG_PACK_STREAMS |
1352 WRITE_RESOURCE_FLAG_PIPABLE)) !=
1353 (WRITE_RESOURCE_FLAG_PACK_STREAMS |
1354 WRITE_RESOURCE_FLAG_PIPABLE));
1356 remove_zero_length_streams(stream_list);
1358 if (list_empty(stream_list)) {
1359 DEBUG("No streams to write.");
1363 memset(&ctx, 0, sizeof(ctx));
1365 /* Pre-sorting the streams is required for compute_stream_list_stats().
1366 * Afterwards, read_stream_list() need not sort them again. */
1367 ret = sort_stream_list_by_sequential_order(stream_list,
1368 offsetof(struct wim_lookup_table_entry,
1369 write_streams_list));
1373 ctx.out_fd = out_fd;
1374 ctx.lookup_table = lookup_table;
1375 ctx.out_ctype = out_ctype;
1376 ctx.out_chunk_size = out_chunk_size;
1377 ctx.write_resource_flags = write_resource_flags;
1378 ctx.filter_ctx = filter_ctx;
1380 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) {
1381 wimlib_assert(out_chunk_size != 0);
1382 if (out_chunk_size <= STACK_MAX) {
1383 ctx.chunk_buf = alloca(out_chunk_size);
1385 ctx.chunk_buf = MALLOC(out_chunk_size);
1386 if (ctx.chunk_buf == NULL) {
1387 ret = WIMLIB_ERR_NOMEM;
1388 goto out_destroy_context;
1392 ctx.chunk_buf_filled = 0;
1394 compute_stream_list_stats(stream_list, &ctx);
1396 ctx.progress_data.progress_func = progress_func;
1398 ctx.num_bytes_to_compress = find_raw_copy_streams(stream_list,
1399 write_resource_flags,
1404 DEBUG("Writing stream list "
1405 "(offset = %"PRIu64", write_resource_flags=0x%08x, "
1406 "out_ctype=%d, out_chunk_size=%u, num_threads=%u, "
1407 "total_bytes=%"PRIu64", num_bytes_to_compress=%"PRIu64")",
1408 out_fd->offset, write_resource_flags,
1409 out_ctype, out_chunk_size, num_threads,
1410 ctx.progress_data.progress.write_streams.total_bytes,
1411 ctx.num_bytes_to_compress);
1413 if (ctx.num_bytes_to_compress == 0) {
1414 DEBUG("No compression needed; skipping to raw copy!");
1415 goto out_write_raw_copy_resources;
1418 /* Unless uncompressed output was required, allocate a chunk_compressor
1419 * to do compression. There are serial and parallel implementations of
1420 * the chunk_compressor interface. We default to parallel using the
1421 * specified number of threads, unless the upper bound on the number
1422 * bytes needing to be compressed is less than a heuristic value. */
1423 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) {
1425 #ifdef ENABLE_MULTITHREADED_COMPRESSION
1426 if (ctx.num_bytes_to_compress > max(2000000, out_chunk_size)) {
1427 ret = new_parallel_chunk_compressor(out_ctype,
1432 DEBUG("Couldn't create parallel chunk compressor "
1433 "(status %d)", ret);
1438 if (ctx.compressor == NULL) {
1439 ret = new_serial_chunk_compressor(out_ctype, out_chunk_size,
1442 goto out_destroy_context;
1447 ctx.progress_data.progress.write_streams.num_threads = ctx.compressor->num_threads;
1449 ctx.progress_data.progress.write_streams.num_threads = 1;
1451 DEBUG("Actually using %u threads",
1452 ctx.progress_data.progress.write_streams.num_threads);
1454 INIT_LIST_HEAD(&ctx.pending_streams);
1455 INIT_LIST_HEAD(&ctx.pack_streams);
1457 if (ctx.progress_data.progress_func) {
1458 (*ctx.progress_data.progress_func)(WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
1459 &ctx.progress_data.progress);
1462 if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
1463 ret = begin_write_resource(&ctx, ctx.num_bytes_to_compress);
1465 goto out_destroy_context;
1468 /* Read the list of streams needing to be compressed, using the
1469 * specified callbacks to execute processing of the data. */
1471 struct read_stream_list_callbacks cbs = {
1472 .begin_stream = write_stream_begin_read,
1473 .begin_stream_ctx = &ctx,
1474 .consume_chunk = write_stream_process_chunk,
1475 .consume_chunk_ctx = &ctx,
1476 .end_stream = write_stream_end_read,
1477 .end_stream_ctx = &ctx,
1480 ret = read_stream_list(stream_list,
1481 offsetof(struct wim_lookup_table_entry, write_streams_list),
1483 STREAM_LIST_ALREADY_SORTED |
1484 VERIFY_STREAM_HASHES |
1485 COMPUTE_MISSING_STREAM_HASHES);
1488 goto out_destroy_context;
1490 ret = finish_remaining_chunks(&ctx);
1492 goto out_destroy_context;
1494 if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
1495 struct wim_reshdr reshdr;
1496 struct wim_lookup_table_entry *lte;
1499 ret = end_write_resource(&ctx, &reshdr);
1501 goto out_destroy_context;
1503 DEBUG("Ending packed resource: %lu %lu %lu.",
1504 reshdr.offset_in_wim,
1506 reshdr.uncompressed_size);
1509 list_for_each_entry(lte, &ctx.pack_streams, write_streams_list) {
1510 lte->out_reshdr.size_in_wim = lte->size;
1511 lte->out_reshdr.flags = filter_resource_flags(lte->flags);
1512 lte->out_reshdr.flags |= WIM_RESHDR_FLAG_PACKED_STREAMS;
1513 lte->out_reshdr.uncompressed_size = 0;
1514 lte->out_reshdr.offset_in_wim = offset_in_res;
1515 lte->out_res_offset_in_wim = reshdr.offset_in_wim;
1516 lte->out_res_size_in_wim = reshdr.size_in_wim;
1517 lte->out_res_uncompressed_size = reshdr.uncompressed_size;
1518 offset_in_res += lte->size;
1520 wimlib_assert(offset_in_res == reshdr.uncompressed_size);
1523 out_write_raw_copy_resources:
1524 /* Copy any compressed resources for which the raw data can be reused
1525 * without decompression. */
1526 ret = write_raw_copy_resources(&raw_copy_streams, ctx.out_fd,
1527 &ctx.progress_data);
1529 out_destroy_context:
1530 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && out_chunk_size > STACK_MAX)
1531 FREE(ctx.chunk_buf);
1532 FREE(ctx.chunk_csizes);
1534 ctx.compressor->destroy(ctx.compressor);
1535 DEBUG("Done (ret=%d)", ret);
1540 wim_write_stream_list(WIMStruct *wim,
1541 struct list_head *stream_list,
1543 unsigned num_threads,
1544 struct filter_context *filter_ctx,
1545 wimlib_progress_func_t progress_func)
1549 int write_resource_flags;
1551 write_resource_flags = write_flags_to_resource_flags(write_flags);
1553 /* wimlib v1.6.3: pack streams by default if the WIM version number is
1554 * that usually used in solid archives. */
1555 if (wim->hdr.wim_version == WIM_VERSION_PACKED_STREAMS)
1556 write_resource_flags |= WRITE_RESOURCE_FLAG_PACK_STREAMS;
1558 if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
1559 out_chunk_size = wim->out_pack_chunk_size;
1560 out_ctype = wim->out_pack_compression_type;
1562 out_chunk_size = wim->out_chunk_size;
1563 out_ctype = wim->out_compression_type;
1566 return write_stream_list(stream_list,
1568 write_resource_flags,
1578 write_wim_resource(struct wim_lookup_table_entry *lte,
1579 struct filedes *out_fd,
1582 int write_resource_flags)
1584 LIST_HEAD(stream_list);
1585 list_add(<e->write_streams_list, &stream_list);
1586 lte->will_be_in_output_wim = 1;
1587 return write_stream_list(&stream_list,
1589 write_resource_flags & ~WRITE_RESOURCE_FLAG_PACK_STREAMS,
1599 write_wim_resource_from_buffer(const void *buf, size_t buf_size,
1600 int reshdr_flags, struct filedes *out_fd,
1603 struct wim_reshdr *out_reshdr,
1605 int write_resource_flags)
1608 struct wim_lookup_table_entry *lte;
1610 /* Set up a temporary lookup table entry to provide to
1611 * write_wim_resource(). */
1613 lte = new_lookup_table_entry();
1615 return WIMLIB_ERR_NOMEM;
1617 lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER;
1618 lte->attached_buffer = (void*)buf;
1619 lte->size = buf_size;
1620 lte->flags = reshdr_flags;
1622 if (write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
1623 sha1_buffer(buf, buf_size, lte->hash);
1629 ret = write_wim_resource(lte, out_fd, out_ctype, out_chunk_size,
1630 write_resource_flags);
1634 copy_reshdr(out_reshdr, <e->out_reshdr);
1637 copy_hash(hash, lte->hash);
1640 lte->resource_location = RESOURCE_NONEXISTENT;
1641 free_lookup_table_entry(lte);
1645 struct stream_size_table {
1646 struct hlist_head *array;
1652 init_stream_size_table(struct stream_size_table *tab, size_t capacity)
1654 tab->array = CALLOC(capacity, sizeof(tab->array[0]));
1655 if (tab->array == NULL)
1656 return WIMLIB_ERR_NOMEM;
1657 tab->num_entries = 0;
1658 tab->capacity = capacity;
1663 destroy_stream_size_table(struct stream_size_table *tab)
1669 stream_size_table_insert(struct wim_lookup_table_entry *lte, void *_tab)
1671 struct stream_size_table *tab = _tab;
1673 struct wim_lookup_table_entry *same_size_lte;
1674 struct hlist_node *tmp;
1676 pos = hash_u64(lte->size) % tab->capacity;
1677 lte->unique_size = 1;
1678 hlist_for_each_entry(same_size_lte, tmp, &tab->array[pos], hash_list_2) {
1679 if (same_size_lte->size == lte->size) {
1680 lte->unique_size = 0;
1681 same_size_lte->unique_size = 0;
1686 hlist_add_head(<e->hash_list_2, &tab->array[pos]);
1691 struct find_streams_ctx {
1694 struct list_head stream_list;
1695 struct stream_size_table stream_size_tab;
1699 reference_stream_for_write(struct wim_lookup_table_entry *lte,
1700 struct list_head *stream_list, u32 nref)
1702 if (!lte->will_be_in_output_wim) {
1703 lte->out_refcnt = 0;
1704 list_add_tail(<e->write_streams_list, stream_list);
1705 lte->will_be_in_output_wim = 1;
1707 lte->out_refcnt += nref;
1711 fully_reference_stream_for_write(struct wim_lookup_table_entry *lte,
1714 struct list_head *stream_list = _stream_list;
1715 lte->will_be_in_output_wim = 0;
1716 reference_stream_for_write(lte, stream_list, lte->refcnt);
1721 inode_find_streams_to_reference(const struct wim_inode *inode,
1722 const struct wim_lookup_table *table,
1723 struct list_head *stream_list)
1725 struct wim_lookup_table_entry *lte;
1728 wimlib_assert(inode->i_nlink > 0);
1730 for (i = 0; i <= inode->i_num_ads; i++) {
1731 lte = inode_stream_lte(inode, i, table);
1733 reference_stream_for_write(lte, stream_list,
1735 else if (!is_zero_hash(inode_stream_hash(inode, i)))
1736 return WIMLIB_ERR_RESOURCE_NOT_FOUND;
1742 do_stream_set_not_in_output_wim(struct wim_lookup_table_entry *lte, void *_ignore)
1744 lte->will_be_in_output_wim = 0;
1749 image_find_streams_to_reference(WIMStruct *wim)
1751 struct wim_image_metadata *imd;
1752 struct wim_inode *inode;
1753 struct wim_lookup_table_entry *lte;
1754 struct list_head *stream_list;
1757 imd = wim_get_current_image_metadata(wim);
1759 image_for_each_unhashed_stream(lte, imd)
1760 lte->will_be_in_output_wim = 0;
1762 stream_list = wim->private;
1763 image_for_each_inode(inode, imd) {
1764 ret = inode_find_streams_to_reference(inode,
1774 prepare_unfiltered_list_of_streams_in_output_wim(WIMStruct *wim,
1777 struct list_head *stream_list_ret)
1781 INIT_LIST_HEAD(stream_list_ret);
1783 if (streams_ok && (image == WIMLIB_ALL_IMAGES ||
1784 (image == 1 && wim->hdr.image_count == 1)))
1786 /* Fast case: Assume that all streams are being written and
1787 * that the reference counts are correct. */
1788 struct wim_lookup_table_entry *lte;
1789 struct wim_image_metadata *imd;
1792 for_lookup_table_entry(wim->lookup_table,
1793 fully_reference_stream_for_write,
1796 for (i = 0; i < wim->hdr.image_count; i++) {
1797 imd = wim->image_metadata[i];
1798 image_for_each_unhashed_stream(lte, imd)
1799 fully_reference_stream_for_write(lte, stream_list_ret);
1802 /* Slow case: Walk through the images being written and
1803 * determine the streams referenced. */
1804 for_lookup_table_entry(wim->lookup_table,
1805 do_stream_set_not_in_output_wim, NULL);
1806 wim->private = stream_list_ret;
1807 ret = for_image(wim, image, image_find_streams_to_reference);
1815 struct insert_other_if_hard_filtered_ctx {
1816 struct stream_size_table *tab;
1817 struct filter_context *filter_ctx;
1821 insert_other_if_hard_filtered(struct wim_lookup_table_entry *lte, void *_ctx)
1823 struct insert_other_if_hard_filtered_ctx *ctx = _ctx;
1825 if (!lte->will_be_in_output_wim &&
1826 stream_hard_filtered(lte, ctx->filter_ctx))
1827 stream_size_table_insert(lte, ctx->tab);
1832 determine_stream_size_uniquity(struct list_head *stream_list,
1833 struct wim_lookup_table *lt,
1834 struct filter_context *filter_ctx)
1837 struct stream_size_table tab;
1838 struct wim_lookup_table_entry *lte;
1840 ret = init_stream_size_table(&tab, 9001);
1844 if (may_hard_filter_streams(filter_ctx)) {
1845 struct insert_other_if_hard_filtered_ctx ctx = {
1847 .filter_ctx = filter_ctx,
1849 for_lookup_table_entry(lt, insert_other_if_hard_filtered, &ctx);
1852 list_for_each_entry(lte, stream_list, write_streams_list)
1853 stream_size_table_insert(lte, &tab);
1855 destroy_stream_size_table(&tab);
1860 filter_stream_list_for_write(struct list_head *stream_list,
1861 struct filter_context *filter_ctx)
1863 struct wim_lookup_table_entry *lte, *tmp;
1865 list_for_each_entry_safe(lte, tmp,
1866 stream_list, write_streams_list)
1868 int status = stream_filtered(lte, filter_ctx);
1875 /* Soft filtered. */
1877 /* Hard filtered. */
1878 lte->will_be_in_output_wim = 0;
1879 list_del(<e->lookup_table_list);
1881 list_del(<e->write_streams_list);
1887 * prepare_stream_list_for_write() -
1889 * Prepare the list of streams to write for writing a WIM containing the
1890 * specified image(s) with the specified write flags.
1893 * The WIMStruct on whose behalf the write is occurring.
1896 * Image(s) from the WIM to write; may be WIMLIB_ALL_IMAGES.
1899 * WIMLIB_WRITE_FLAG_* flags for the write operation:
1901 * STREAMS_OK: For writes of all images, assume that all streams in the
1902 * lookup table of @wim and the per-image lists of unhashed streams should
1903 * be taken as-is, and image metadata should not be searched for
1904 * references. This does not exclude filtering with OVERWRITE and
1905 * SKIP_EXTERNAL_WIMS, below.
1907 * OVERWRITE: Streams already present in @wim shall not be returned in
1910 * SKIP_EXTERNAL_WIMS: Streams already present in a WIM file, but not
1911 * @wim, shall be be returned in neither @stream_list_ret nor
1912 * @lookup_table_list_ret.
1915 * List of streams, linked by write_streams_list, that need to be written
1916 * will be returned here.
1918 * Note that this function assumes that unhashed streams will be written;
1919 * it does not take into account that they may become duplicates when
1922 * @lookup_table_list_ret
1923 * List of streams, linked by lookup_table_list, that need to be included
1924 * in the WIM's lookup table will be returned here. This will be a
1925 * superset of the streams in @stream_list_ret.
1927 * This list will be a proper superset of @stream_list_ret if and only if
1928 * WIMLIB_WRITE_FLAG_OVERWRITE was specified in @write_flags and some of
1929 * the streams that would otherwise need to be written were already located
1932 * All streams in this list will have @out_refcnt set to the number of
1933 * references to the stream in the output WIM. If
1934 * WIMLIB_WRITE_FLAG_STREAMS_OK was specified in @write_flags, @out_refcnt
1935 * may be as low as 0.
1938 * A context for queries of stream filter status with stream_filtered() is
1939 * returned in this location.
1941 * In addition, @will_be_in_output_wim will be set to 1 in all stream entries
1942 * inserted into @lookup_table_list_ret and to 0 in all stream entries in the
1943 * lookup table of @wim not inserted into @lookup_table_list_ret.
1945 * Still furthermore, @unique_size will be set to 1 on all stream entries in
1946 * @stream_list_ret that have unique size among all stream entries in
1947 * @stream_list_ret and among all stream entries in the lookup table of @wim
1948 * that are ineligible for being written due to filtering.
1950 * Returns 0 on success; nonzero on read error, memory allocation error, or
1954 prepare_stream_list_for_write(WIMStruct *wim, int image,
1956 struct list_head *stream_list_ret,
1957 struct list_head *lookup_table_list_ret,
1958 struct filter_context *filter_ctx_ret)
1961 struct wim_lookup_table_entry *lte;
1963 filter_ctx_ret->write_flags = write_flags;
1964 filter_ctx_ret->wim = wim;
1966 ret = prepare_unfiltered_list_of_streams_in_output_wim(
1969 write_flags & WIMLIB_WRITE_FLAG_STREAMS_OK,
1974 INIT_LIST_HEAD(lookup_table_list_ret);
1975 list_for_each_entry(lte, stream_list_ret, write_streams_list)
1976 list_add_tail(<e->lookup_table_list, lookup_table_list_ret);
1978 ret = determine_stream_size_uniquity(stream_list_ret, wim->lookup_table,
1983 if (may_filter_streams(filter_ctx_ret))
1984 filter_stream_list_for_write(stream_list_ret, filter_ctx_ret);
1990 write_wim_streams(WIMStruct *wim, int image, int write_flags,
1991 unsigned num_threads,
1992 wimlib_progress_func_t progress_func,
1993 struct list_head *stream_list_override,
1994 struct list_head *lookup_table_list_ret)
1997 struct list_head _stream_list;
1998 struct list_head *stream_list;
1999 struct wim_lookup_table_entry *lte;
2000 struct filter_context _filter_ctx;
2001 struct filter_context *filter_ctx;
2003 if (stream_list_override == NULL) {
2004 /* Normal case: prepare stream list from image(s) being written.
2006 stream_list = &_stream_list;
2007 filter_ctx = &_filter_ctx;
2008 ret = prepare_stream_list_for_write(wim, image, write_flags,
2010 lookup_table_list_ret,
2015 /* Currently only as a result of wimlib_split() being called:
2016 * use stream list already explicitly provided. Use existing
2017 * reference counts. */
2018 stream_list = stream_list_override;
2020 INIT_LIST_HEAD(lookup_table_list_ret);
2021 list_for_each_entry(lte, stream_list, write_streams_list) {
2022 lte->out_refcnt = lte->refcnt;
2023 lte->will_be_in_output_wim = 1;
2024 lte->unique_size = 0;
2025 list_add_tail(<e->lookup_table_list, lookup_table_list_ret);
2029 return wim_write_stream_list(wim,
2038 write_wim_metadata_resources(WIMStruct *wim, int image, int write_flags,
2039 wimlib_progress_func_t progress_func)
2044 int write_resource_flags;
2046 if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA) {
2047 DEBUG("Not writing any metadata resources.");
2051 write_resource_flags = write_flags_to_resource_flags(write_flags);
2053 write_resource_flags &= ~WRITE_RESOURCE_FLAG_PACK_STREAMS;
2055 DEBUG("Writing metadata resources (offset=%"PRIu64")",
2056 wim->out_fd.offset);
2059 progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN, NULL);
2061 if (image == WIMLIB_ALL_IMAGES) {
2063 end_image = wim->hdr.image_count;
2065 start_image = image;
2069 for (int i = start_image; i <= end_image; i++) {
2070 struct wim_image_metadata *imd;
2072 imd = wim->image_metadata[i - 1];
2073 /* Build a new metadata resource only if image was modified from
2074 * the original (or was newly added). Otherwise just copy the
2076 if (imd->modified) {
2077 DEBUG("Image %u was modified; building and writing new "
2078 "metadata resource", i);
2079 ret = write_metadata_resource(wim, i,
2080 write_resource_flags);
2081 } else if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) {
2082 DEBUG("Image %u was not modified; re-using existing "
2083 "metadata resource.", i);
2084 stream_set_out_reshdr_for_reuse(imd->metadata_lte);
2087 DEBUG("Image %u was not modified; copying existing "
2088 "metadata resource.", i);
2089 ret = write_wim_resource(imd->metadata_lte,
2091 wim->out_compression_type,
2092 wim->out_chunk_size,
2093 write_resource_flags);
2099 progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_END, NULL);
2104 open_wim_writable(WIMStruct *wim, const tchar *path, int open_flags)
2107 DEBUG("Opening \"%"TS"\" for writing.", path);
2109 raw_fd = topen(path, open_flags | O_BINARY, 0644);
2111 ERROR_WITH_ERRNO("Failed to open \"%"TS"\" for writing", path);
2112 return WIMLIB_ERR_OPEN;
2114 filedes_init(&wim->out_fd, raw_fd);
2119 close_wim_writable(WIMStruct *wim, int write_flags)
2123 if (!(write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)) {
2124 DEBUG("Closing WIM file.");
2125 if (filedes_valid(&wim->out_fd))
2126 if (filedes_close(&wim->out_fd))
2127 ret = WIMLIB_ERR_WRITE;
2129 filedes_invalidate(&wim->out_fd);
2134 cmp_streams_by_out_rspec(const void *p1, const void *p2)
2136 const struct wim_lookup_table_entry *lte1, *lte2;
2138 lte1 = *(const struct wim_lookup_table_entry**)p1;
2139 lte2 = *(const struct wim_lookup_table_entry**)p2;
2141 if (lte1->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
2142 if (lte2->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
2143 if (lte1->out_res_offset_in_wim != lte2->out_res_offset_in_wim)
2144 return cmp_u64(lte1->out_res_offset_in_wim,
2145 lte2->out_res_offset_in_wim);
2150 if (lte2->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS)
2153 return cmp_u64(lte1->out_reshdr.offset_in_wim,
2154 lte2->out_reshdr.offset_in_wim);
2158 write_wim_lookup_table(WIMStruct *wim, int image, int write_flags,
2159 struct wim_reshdr *out_reshdr,
2160 struct list_head *lookup_table_list)
2164 /* Set output resource metadata for streams already present in WIM. */
2165 if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) {
2166 struct wim_lookup_table_entry *lte;
2167 list_for_each_entry(lte, lookup_table_list, lookup_table_list)
2169 if (lte->resource_location == RESOURCE_IN_WIM &&
2170 lte->rspec->wim == wim)
2172 stream_set_out_reshdr_for_reuse(lte);
2177 ret = sort_stream_list(lookup_table_list,
2178 offsetof(struct wim_lookup_table_entry, lookup_table_list),
2179 cmp_streams_by_out_rspec);
2183 /* Add entries for metadata resources. */
2184 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)) {
2188 if (image == WIMLIB_ALL_IMAGES) {
2190 end_image = wim->hdr.image_count;
2192 start_image = image;
2196 /* Push metadata resource lookup table entries onto the front of
2197 * the list in reverse order, so that they're written in order.
2199 for (int i = end_image; i >= start_image; i--) {
2200 struct wim_lookup_table_entry *metadata_lte;
2202 metadata_lte = wim->image_metadata[i - 1]->metadata_lte;
2203 wimlib_assert(metadata_lte->out_reshdr.flags & WIM_RESHDR_FLAG_METADATA);
2204 metadata_lte->out_refcnt = 1;
2205 list_add(&metadata_lte->lookup_table_list, lookup_table_list);
2209 return write_wim_lookup_table_from_stream_list(lookup_table_list,
2211 wim->hdr.part_number,
2213 write_flags_to_resource_flags(write_flags));
2219 * Finish writing a WIM file: write the lookup table, xml data, and integrity
2220 * table, then overwrite the WIM header. By default, closes the WIM file
2221 * descriptor (@wim->out_fd) if successful.
2223 * write_flags is a bitwise OR of the following:
2225 * (public) WIMLIB_WRITE_FLAG_CHECK_INTEGRITY:
2226 * Include an integrity table.
2228 * (public) WIMLIB_WRITE_FLAG_FSYNC:
2229 * fsync() the output file before closing it.
2231 * (public) WIMLIB_WRITE_FLAG_PIPABLE:
2232 * Writing a pipable WIM, possibly to a pipe; include pipable WIM
2233 * stream headers before the lookup table and XML data, and also
2234 * write the WIM header at the end instead of seeking to the
2235 * beginning. Can't be combined with
2236 * WIMLIB_WRITE_FLAG_CHECK_INTEGRITY.
2238 * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE:
2239 * Don't write the lookup table.
2241 * (private) WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE:
2242 * When (if) writing the integrity table, re-use entries from the
2243 * existing integrity table, if possible.
2245 * (private) WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML:
2246 * After writing the XML data but before writing the integrity
2247 * table, write a temporary WIM header and flush the stream so that
2248 * the WIM is less likely to become corrupted upon abrupt program
2250 * (private) WIMLIB_WRITE_FLAG_HEADER_AT_END:
2251 * Instead of overwriting the WIM header at the beginning of the
2252 * file, simply append it to the end of the file. (Used when
2254 * (private) WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR:
2255 * Do not close the file descriptor @wim->out_fd on either success
2257 * (private) WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES:
2258 * Use the existing <TOTALBYTES> stored in the in-memory XML
2259 * information, rather than setting it to the offset of the XML
2260 * data being written.
2263 finish_write(WIMStruct *wim, int image, int write_flags,
2264 wimlib_progress_func_t progress_func,
2265 struct list_head *lookup_table_list)
2269 int write_resource_flags;
2270 off_t old_lookup_table_end;
2271 off_t new_lookup_table_end;
2274 DEBUG("image=%d, write_flags=%08x", image, write_flags);
2276 write_resource_flags = write_flags_to_resource_flags(write_flags);
2278 /* In the WIM header, there is room for the resource entry for a
2279 * metadata resource labeled as the "boot metadata". This entry should
2280 * be zeroed out if there is no bootable image (boot_idx 0). Otherwise,
2281 * it should be a copy of the resource entry for the image that is
2282 * marked as bootable. This is not well documented... */
2283 if (wim->hdr.boot_idx == 0) {
2284 zero_reshdr(&wim->hdr.boot_metadata_reshdr);
2286 copy_reshdr(&wim->hdr.boot_metadata_reshdr,
2287 &wim->image_metadata[
2288 wim->hdr.boot_idx - 1]->metadata_lte->out_reshdr);
2291 /* Write lookup table. (Save old position first.) */
2292 old_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim +
2293 wim->hdr.lookup_table_reshdr.size_in_wim;
2294 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
2295 ret = write_wim_lookup_table(wim, image, write_flags,
2296 &wim->hdr.lookup_table_reshdr,
2302 /* Write XML data. */
2303 xml_totalbytes = wim->out_fd.offset;
2304 if (write_flags & WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES)
2305 xml_totalbytes = WIM_TOTALBYTES_USE_EXISTING;
2306 ret = write_wim_xml_data(wim, image, xml_totalbytes,
2307 &wim->hdr.xml_data_reshdr,
2308 write_resource_flags);
2312 /* Write integrity table (optional). */
2313 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
2314 if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) {
2315 struct wim_header checkpoint_hdr;
2316 memcpy(&checkpoint_hdr, &wim->hdr, sizeof(struct wim_header));
2317 zero_reshdr(&checkpoint_hdr.integrity_table_reshdr);
2318 checkpoint_hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2319 ret = write_wim_header_at_offset(&checkpoint_hdr,
2325 if (!(write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE))
2326 old_lookup_table_end = 0;
2328 new_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim +
2329 wim->hdr.lookup_table_reshdr.size_in_wim;
2331 ret = write_integrity_table(wim,
2332 new_lookup_table_end,
2333 old_lookup_table_end,
2338 /* No integrity table. */
2339 zero_reshdr(&wim->hdr.integrity_table_reshdr);
2342 /* Now that all information in the WIM header has been determined, the
2343 * preliminary header written earlier can be overwritten, the header of
2344 * the existing WIM file can be overwritten, or the final header can be
2345 * written to the end of the pipable WIM. */
2346 wim->hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2348 if (write_flags & WIMLIB_WRITE_FLAG_HEADER_AT_END)
2349 hdr_offset = wim->out_fd.offset;
2350 DEBUG("Writing new header @ %"PRIu64".", hdr_offset);
2351 ret = write_wim_header_at_offset(&wim->hdr, &wim->out_fd, hdr_offset);
2355 /* Possibly sync file data to disk before closing. On POSIX systems, it
2356 * is necessary to do this before using rename() to overwrite an
2357 * existing file with a new file. Otherwise, data loss would occur if
2358 * the system is abruptly terminated when the metadata for the rename
2359 * operation has been written to disk, but the new file data has not.
2361 if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) {
2362 DEBUG("Syncing WIM file.");
2363 if (fsync(wim->out_fd.fd)) {
2364 ERROR_WITH_ERRNO("Error syncing data to WIM file");
2365 return WIMLIB_ERR_WRITE;
2369 if (close_wim_writable(wim, write_flags)) {
2370 ERROR_WITH_ERRNO("Failed to close the output WIM file");
2371 return WIMLIB_ERR_WRITE;
2377 #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK)
2379 lock_wim(WIMStruct *wim, int fd)
2382 if (fd != -1 && !wim->wim_locked) {
2383 ret = flock(fd, LOCK_EX | LOCK_NB);
2385 if (errno == EWOULDBLOCK) {
2386 ERROR("`%"TS"' is already being modified or has been "
2387 "mounted read-write\n"
2388 " by another process!", wim->filename);
2389 ret = WIMLIB_ERR_ALREADY_LOCKED;
2391 WARNING_WITH_ERRNO("Failed to lock `%"TS"'",
2396 wim->wim_locked = 1;
2404 * write_pipable_wim():
2406 * Perform the intermediate stages of creating a "pipable" WIM (i.e. a WIM
2407 * capable of being applied from a pipe).
2409 * Pipable WIMs are a wimlib-specific modification of the WIM format such that
2410 * images can be applied from them sequentially when the file data is sent over
2411 * a pipe. In addition, a pipable WIM can be written sequentially to a pipe.
2412 * The modifications made to the WIM format for pipable WIMs are:
2414 * - Magic characters in header are "WLPWM\0\0\0" (wimlib pipable WIM) instead
2415 * of "MSWIM\0\0\0". This lets wimlib know that the WIM is pipable and also
2416 * stops other software from trying to read the file as a normal WIM.
2418 * - The header at the beginning of the file does not contain all the normal
2419 * information; in particular it will have all 0's for the lookup table and
2420 * XML data resource entries. This is because this information cannot be
2421 * determined until the lookup table and XML data have been written.
2422 * Consequently, wimlib will write the full header at the very end of the
2423 * file. The header at the end, however, is only used when reading the WIM
2424 * from a seekable file (not a pipe).
2426 * - An extra copy of the XML data is placed directly after the header. This
2427 * allows image names and sizes to be determined at an appropriate time when
2428 * reading the WIM from a pipe. This copy of the XML data is ignored if the
2429 * WIM is read from a seekable file (not a pipe).
2431 * - The format of resources, or streams, has been modified to allow them to be
2432 * used before the "lookup table" has been read. Each stream is prefixed with
2433 * a `struct pwm_stream_hdr' that is basically an abbreviated form of `struct
2434 * wim_lookup_table_entry_disk' that only contains the SHA1 message digest,
2435 * uncompressed stream size, and flags that indicate whether the stream is
2436 * compressed. The data of uncompressed streams then follows literally, while
2437 * the data of compressed streams follows in a modified format. Compressed
2438 * streams do not begin with a chunk table, since the chunk table cannot be
2439 * written until all chunks have been compressed. Instead, each compressed
2440 * chunk is prefixed by a `struct pwm_chunk_hdr' that gives its size.
2441 * Furthermore, the chunk table is written at the end of the resource instead
2442 * of the start. Note: chunk offsets are given in the chunk table as if the
2443 * `struct pwm_chunk_hdr's were not present; also, the chunk table is only
2444 * used if the WIM is being read from a seekable file (not a pipe).
2446 * - Metadata resources always come before other file resources (streams).
2447 * (This does not by itself constitute an incompatibility with normal WIMs,
2448 * since this is valid in normal WIMs.)
2450 * - At least up to the end of the file resources, all components must be packed
2451 * as tightly as possible; there cannot be any "holes" in the WIM. (This does
2452 * not by itself consititute an incompatibility with normal WIMs, since this
2453 * is valid in normal WIMs.)
2455 * Note: the lookup table, XML data, and header at the end are not used when
2456 * applying from a pipe. They exist to support functionality such as image
2457 * application and export when the WIM is *not* read from a pipe.
2459 * Layout of pipable WIM:
2461 * ---------+----------+--------------------+----------------+--------------+-----------+--------+
2462 * | Header | XML data | Metadata resources | File resources | Lookup table | XML data | Header |
2463 * ---------+----------+--------------------+----------------+--------------+-----------+--------+
2465 * Layout of normal WIM:
2467 * +--------+-----------------------------+-------------------------+
2468 * | Header | File and metadata resources | Lookup table | XML data |
2469 * +--------+-----------------------------+-------------------------+
2471 * An optional integrity table can follow the final XML data in both normal and
2472 * pipable WIMs. However, due to implementation details, wimlib currently can
2473 * only include an integrity table in a pipable WIM when writing it to a
2474 * seekable file (not a pipe).
2476 * Do note that since pipable WIMs are not supported by Microsoft's software,
2477 * wimlib does not create them unless explicitly requested (with
2478 * WIMLIB_WRITE_FLAG_PIPABLE) and as stated above they use different magic
2479 * characters to identify the file.
2482 write_pipable_wim(WIMStruct *wim, int image, int write_flags,
2483 unsigned num_threads, wimlib_progress_func_t progress_func,
2484 struct list_head *stream_list_override,
2485 struct list_head *lookup_table_list_ret)
2488 struct wim_reshdr xml_reshdr;
2490 WARNING("Creating a pipable WIM, which will "
2492 " with Microsoft's software (wimgapi/imagex/Dism).");
2494 /* At this point, the header at the beginning of the file has already
2497 /* For efficiency, when wimlib adds an image to the WIM with
2498 * wimlib_add_image(), the SHA1 message digests of files is not
2499 * calculated; instead, they are calculated while the files are being
2500 * written. However, this does not work when writing a pipable WIM,
2501 * since when writing a stream to a pipable WIM, its SHA1 message digest
2502 * needs to be known before the stream data is written. Therefore,
2503 * before getting much farther, we need to pre-calculate the SHA1
2504 * message digests of all streams that will be written. */
2505 ret = wim_checksum_unhashed_streams(wim);
2509 /* Write extra copy of the XML data. */
2510 ret = write_wim_xml_data(wim, image, WIM_TOTALBYTES_OMIT,
2512 WRITE_RESOURCE_FLAG_PIPABLE);
2516 /* Write metadata resources for the image(s) being included in the
2518 ret = write_wim_metadata_resources(wim, image, write_flags,
2523 /* Write streams needed for the image(s) being included in the output
2524 * WIM, or streams needed for the split WIM part. */
2525 return write_wim_streams(wim, image, write_flags, num_threads,
2526 progress_func, stream_list_override,
2527 lookup_table_list_ret);
2529 /* The lookup table, XML data, and header at end are handled by
2530 * finish_write(). */
2533 /* Write a standalone WIM or split WIM (SWM) part to a new file or to a file
2536 write_wim_part(WIMStruct *wim,
2537 const void *path_or_fd,
2540 unsigned num_threads,
2541 wimlib_progress_func_t progress_func,
2542 unsigned part_number,
2543 unsigned total_parts,
2544 struct list_head *stream_list_override,
2548 struct wim_header hdr_save;
2549 struct list_head lookup_table_list;
2551 if (total_parts == 1)
2552 DEBUG("Writing standalone WIM.");
2554 DEBUG("Writing split WIM part %u/%u", part_number, total_parts);
2555 if (image == WIMLIB_ALL_IMAGES)
2556 DEBUG("Including all images.");
2558 DEBUG("Including image %d only.", image);
2559 if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)
2560 DEBUG("File descriptor: %d", *(const int*)path_or_fd);
2562 DEBUG("Path: \"%"TS"\"", (const tchar*)path_or_fd);
2563 DEBUG("Write flags: 0x%08x", write_flags);
2565 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)
2566 DEBUG("\tCHECK_INTEGRITY");
2568 if (write_flags & WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)
2569 DEBUG("\tNO_CHECK_INTEGRITY");
2571 if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
2574 if (write_flags & WIMLIB_WRITE_FLAG_NOT_PIPABLE)
2575 DEBUG("\tNOT_PIPABLE");
2577 if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
2578 DEBUG("\tRECOMPRESS");
2580 if (write_flags & WIMLIB_WRITE_FLAG_FSYNC)
2583 if (write_flags & WIMLIB_WRITE_FLAG_REBUILD)
2586 if (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE)
2587 DEBUG("\tSOFT_DELETE");
2589 if (write_flags & WIMLIB_WRITE_FLAG_IGNORE_READONLY_FLAG)
2590 DEBUG("\tIGNORE_READONLY_FLAG");
2592 if (write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS)
2593 DEBUG("\tSKIP_EXTERNAL_WIMS");
2595 if (write_flags & WIMLIB_WRITE_FLAG_STREAMS_OK)
2596 DEBUG("\tSTREAMS_OK");
2598 if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS)
2599 DEBUG("\tPACK_STREAMS");
2601 if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)
2602 DEBUG("\tFILE_DESCRIPTOR");
2604 if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)
2605 DEBUG("\tNO_METADATA");
2607 if (write_flags & WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES)
2608 DEBUG("\tUSE_EXISTING_TOTALBYTES");
2610 if (num_threads == 0)
2611 DEBUG("Number of threads: autodetect");
2613 DEBUG("Number of threads: %u", num_threads);
2614 DEBUG("Progress function: %s", (progress_func ? "yes" : "no"));
2615 DEBUG("Stream list: %s", (stream_list_override ? "specified" : "autodetect"));
2616 DEBUG("GUID: %s", ((guid || wim->guid_set_explicitly) ?
2617 "specified" : "generate new"));
2619 /* Internally, this is always called with a valid part number and total
2621 wimlib_assert(total_parts >= 1);
2622 wimlib_assert(part_number >= 1 && part_number <= total_parts);
2624 /* A valid image (or all images) must be specified. */
2625 if (image != WIMLIB_ALL_IMAGES &&
2626 (image < 1 || image > wim->hdr.image_count))
2627 return WIMLIB_ERR_INVALID_IMAGE;
2629 /* If we need to write metadata resources, make sure the ::WIMStruct has
2630 * the needed information attached (e.g. is not a resource-only WIM,
2631 * such as a non-first part of a split WIM). */
2632 if (!wim_has_metadata(wim) &&
2633 !(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA))
2634 return WIMLIB_ERR_METADATA_NOT_FOUND;
2636 /* Check for contradictory flags. */
2637 if ((write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2638 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))
2639 == (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2640 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))
2641 return WIMLIB_ERR_INVALID_PARAM;
2643 if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2644 WIMLIB_WRITE_FLAG_NOT_PIPABLE))
2645 == (WIMLIB_WRITE_FLAG_PIPABLE |
2646 WIMLIB_WRITE_FLAG_NOT_PIPABLE))
2647 return WIMLIB_ERR_INVALID_PARAM;
2649 /* Save previous header, then start initializing the new one. */
2650 memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header));
2652 /* Set default integrity, pipable, and packed stream flags. */
2653 if (!(write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2654 WIMLIB_WRITE_FLAG_NOT_PIPABLE)))
2655 if (wim_is_pipable(wim)) {
2656 DEBUG("WIM is pipable; default to PIPABLE.");
2657 write_flags |= WIMLIB_WRITE_FLAG_PIPABLE;
2660 if (!(write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2661 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)))
2662 if (wim_has_integrity_table(wim)) {
2663 DEBUG("Integrity table present; default to CHECK_INTEGRITY.");
2664 write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY;
2667 if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2668 WIMLIB_WRITE_FLAG_PACK_STREAMS))
2669 == (WIMLIB_WRITE_FLAG_PIPABLE |
2670 WIMLIB_WRITE_FLAG_PACK_STREAMS))
2672 ERROR("Cannot specify both PIPABLE and PACK_STREAMS!");
2673 return WIMLIB_ERR_INVALID_PARAM;
2676 /* Set appropriate magic number. */
2677 if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
2678 wim->hdr.magic = PWM_MAGIC;
2680 wim->hdr.magic = WIM_MAGIC;
2682 /* Set appropriate version number. */
2683 if ((write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS) ||
2684 wim->out_compression_type == WIMLIB_COMPRESSION_TYPE_LZMS)
2685 wim->hdr.wim_version = WIM_VERSION_PACKED_STREAMS;
2687 wim->hdr.wim_version = WIM_VERSION_DEFAULT;
2689 /* Clear header flags that will be set automatically. */
2690 wim->hdr.flags &= ~(WIM_HDR_FLAG_METADATA_ONLY |
2691 WIM_HDR_FLAG_RESOURCE_ONLY |
2692 WIM_HDR_FLAG_SPANNED |
2693 WIM_HDR_FLAG_WRITE_IN_PROGRESS);
2695 /* Set SPANNED header flag if writing part of a split WIM. */
2696 if (total_parts != 1)
2697 wim->hdr.flags |= WIM_HDR_FLAG_SPANNED;
2699 /* Set part number and total parts of split WIM. This will be 1 and 1
2700 * if the WIM is standalone. */
2701 wim->hdr.part_number = part_number;
2702 wim->hdr.total_parts = total_parts;
2704 /* Set compression type if different. */
2705 if (wim->compression_type != wim->out_compression_type) {
2706 ret = set_wim_hdr_cflags(wim->out_compression_type, &wim->hdr);
2707 wimlib_assert(ret == 0);
2710 /* Set chunk size if different. */
2711 wim->hdr.chunk_size = wim->out_chunk_size;
2713 /* Use GUID if specified; otherwise generate a new one. */
2715 memcpy(wim->hdr.guid, guid, WIMLIB_GUID_LEN);
2716 else if (!wim->guid_set_explicitly)
2717 randomize_byte_array(wim->hdr.guid, WIMLIB_GUID_LEN);
2719 /* Clear references to resources that have not been written yet. */
2720 zero_reshdr(&wim->hdr.lookup_table_reshdr);
2721 zero_reshdr(&wim->hdr.xml_data_reshdr);
2722 zero_reshdr(&wim->hdr.boot_metadata_reshdr);
2723 zero_reshdr(&wim->hdr.integrity_table_reshdr);
2725 /* Set image count and boot index correctly for single image writes. */
2726 if (image != WIMLIB_ALL_IMAGES) {
2727 wim->hdr.image_count = 1;
2728 if (wim->hdr.boot_idx == image)
2729 wim->hdr.boot_idx = 1;
2731 wim->hdr.boot_idx = 0;
2734 /* Split WIMs can't be bootable. */
2735 if (total_parts != 1)
2736 wim->hdr.boot_idx = 0;
2738 /* Initialize output file descriptor. */
2739 if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR) {
2740 /* File descriptor was explicitly provided. Return error if
2741 * file descriptor is not seekable, unless writing a pipable WIM
2743 wim->out_fd.fd = *(const int*)path_or_fd;
2744 wim->out_fd.offset = 0;
2745 if (!filedes_is_seekable(&wim->out_fd)) {
2746 ret = WIMLIB_ERR_INVALID_PARAM;
2747 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
2748 goto out_restore_hdr;
2749 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
2750 ERROR("Can't include integrity check when "
2751 "writing pipable WIM to pipe!");
2752 goto out_restore_hdr;
2757 /* Filename of WIM to write was provided; open file descriptor
2759 ret = open_wim_writable(wim, (const tchar*)path_or_fd,
2760 O_TRUNC | O_CREAT | O_RDWR);
2762 goto out_restore_hdr;
2765 /* Write initial header. This is merely a "dummy" header since it
2766 * doesn't have all the information yet, so it will be overwritten later
2767 * (unless writing a pipable WIM). */
2768 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
2769 wim->hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2770 ret = write_wim_header(&wim->hdr, &wim->out_fd);
2771 wim->hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2773 goto out_restore_hdr;
2775 /* Write metadata resources and streams. */
2776 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE)) {
2777 /* Default case: create a normal (non-pipable) WIM. */
2778 ret = write_wim_streams(wim, image, write_flags, num_threads,
2779 progress_func, stream_list_override,
2780 &lookup_table_list);
2782 goto out_restore_hdr;
2784 ret = write_wim_metadata_resources(wim, image, write_flags,
2787 goto out_restore_hdr;
2789 /* Non-default case: create pipable WIM. */
2790 ret = write_pipable_wim(wim, image, write_flags, num_threads,
2791 progress_func, stream_list_override,
2792 &lookup_table_list);
2794 goto out_restore_hdr;
2795 write_flags |= WIMLIB_WRITE_FLAG_HEADER_AT_END;
2799 /* Write lookup table, XML data, and (optional) integrity table. */
2800 ret = finish_write(wim, image, write_flags, progress_func,
2801 &lookup_table_list);
2803 memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header));
2804 (void)close_wim_writable(wim, write_flags);
2805 DEBUG("ret=%d", ret);
2809 /* Write a standalone WIM to a file or file descriptor. */
2811 write_standalone_wim(WIMStruct *wim, const void *path_or_fd,
2812 int image, int write_flags, unsigned num_threads,
2813 wimlib_progress_func_t progress_func)
2815 return write_wim_part(wim, path_or_fd, image, write_flags,
2816 num_threads, progress_func, 1, 1, NULL, NULL);
2819 /* API function documented in wimlib.h */
2821 wimlib_write(WIMStruct *wim, const tchar *path,
2822 int image, int write_flags, unsigned num_threads,
2823 wimlib_progress_func_t progress_func)
2825 if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC)
2826 return WIMLIB_ERR_INVALID_PARAM;
2828 if (path == NULL || path[0] == T('\0'))
2829 return WIMLIB_ERR_INVALID_PARAM;
2831 return write_standalone_wim(wim, path, image, write_flags,
2832 num_threads, progress_func);
2835 /* API function documented in wimlib.h */
2837 wimlib_write_to_fd(WIMStruct *wim, int fd,
2838 int image, int write_flags, unsigned num_threads,
2839 wimlib_progress_func_t progress_func)
2841 if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC)
2842 return WIMLIB_ERR_INVALID_PARAM;
2845 return WIMLIB_ERR_INVALID_PARAM;
2847 write_flags |= WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR;
2849 return write_standalone_wim(wim, &fd, image, write_flags,
2850 num_threads, progress_func);
2854 any_images_modified(WIMStruct *wim)
2856 for (int i = 0; i < wim->hdr.image_count; i++)
2857 if (wim->image_metadata[i]->modified)
2863 check_resource_offset(struct wim_lookup_table_entry *lte, void *_wim)
2865 const WIMStruct *wim = _wim;
2866 off_t end_offset = *(const off_t*)wim->private;
2868 if (lte->resource_location == RESOURCE_IN_WIM && lte->rspec->wim == wim &&
2869 lte->rspec->offset_in_wim + lte->rspec->size_in_wim > end_offset)
2870 return WIMLIB_ERR_RESOURCE_ORDER;
2874 /* Make sure no file or metadata resources are located after the XML data (or
2875 * integrity table if present)--- otherwise we can't safely overwrite the WIM in
2876 * place and we return WIMLIB_ERR_RESOURCE_ORDER. */
2878 check_resource_offsets(WIMStruct *wim, off_t end_offset)
2883 wim->private = &end_offset;
2884 ret = for_lookup_table_entry(wim->lookup_table, check_resource_offset, wim);
2888 for (i = 0; i < wim->hdr.image_count; i++) {
2889 ret = check_resource_offset(wim->image_metadata[i]->metadata_lte, wim);
2897 * Overwrite a WIM, possibly appending streams to it.
2899 * A WIM looks like (or is supposed to look like) the following:
2901 * Header (212 bytes)
2902 * Streams and metadata resources (variable size)
2903 * Lookup table (variable size)
2904 * XML data (variable size)
2905 * Integrity table (optional) (variable size)
2907 * If we are not adding any streams or metadata resources, the lookup table is
2908 * unchanged--- so we only need to overwrite the XML data, integrity table, and
2909 * header. This operation is potentially unsafe if the program is abruptly
2910 * terminated while the XML data or integrity table are being overwritten, but
2911 * before the new header has been written. To partially alleviate this problem,
2912 * a special flag (WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) is passed to
2913 * finish_write() to cause a temporary WIM header to be written after the XML
2914 * data has been written. This may prevent the WIM from becoming corrupted if
2915 * the program is terminated while the integrity table is being calculated (but
2916 * no guarantees, due to write re-ordering...).
2918 * If we are adding new streams or images (metadata resources), the lookup table
2919 * needs to be changed, and those streams need to be written. In this case, we
2920 * try to perform a safe update of the WIM file by writing the streams *after*
2921 * the end of the previous WIM, then writing the new lookup table, XML data, and
2922 * (optionally) integrity table following the new streams. This will produce a
2923 * layout like the following:
2925 * Header (212 bytes)
2926 * (OLD) Streams and metadata resources (variable size)
2927 * (OLD) Lookup table (variable size)
2928 * (OLD) XML data (variable size)
2929 * (OLD) Integrity table (optional) (variable size)
2930 * (NEW) Streams and metadata resources (variable size)
2931 * (NEW) Lookup table (variable size)
2932 * (NEW) XML data (variable size)
2933 * (NEW) Integrity table (optional) (variable size)
2935 * At all points, the WIM is valid as nothing points to the new data yet. Then,
2936 * the header is overwritten to point to the new lookup table, XML data, and
2937 * integrity table, to produce the following layout:
2939 * Header (212 bytes)
2940 * Streams and metadata resources (variable size)
2941 * Nothing (variable size)
2942 * More Streams and metadata resources (variable size)
2943 * Lookup table (variable size)
2944 * XML data (variable size)
2945 * Integrity table (optional) (variable size)
2947 * This method allows an image to be appended to a large WIM very quickly, and
2948 * is is crash-safe except in the case of write re-ordering, but the
2949 * disadvantage is that a small hole is left in the WIM where the old lookup
2950 * table, xml data, and integrity table were. (These usually only take up a
2951 * small amount of space compared to the streams, however.)
2954 overwrite_wim_inplace(WIMStruct *wim, int write_flags,
2955 unsigned num_threads,
2956 wimlib_progress_func_t progress_func)
2960 u64 old_lookup_table_end, old_xml_begin, old_xml_end;
2961 struct wim_header hdr_save;
2962 struct list_head stream_list;
2963 struct list_head lookup_table_list;
2964 struct filter_context filter_ctx;
2966 DEBUG("Overwriting `%"TS"' in-place", wim->filename);
2968 /* Save original header so it can be restored in case of error */
2969 memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header));
2971 /* Set default integrity flag. */
2972 if (!(write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2973 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)))
2974 if (wim_has_integrity_table(wim))
2975 write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY;
2977 /* Set WIM version if adding packed streams. */
2978 if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS)
2979 wim->hdr.wim_version = WIM_VERSION_PACKED_STREAMS;
2981 /* Set additional flags for overwrite. */
2982 write_flags |= WIMLIB_WRITE_FLAG_OVERWRITE |
2983 WIMLIB_WRITE_FLAG_STREAMS_OK;
2985 /* Make sure that the integrity table (if present) is after the XML
2986 * data, and that there are no stream resources, metadata resources, or
2987 * lookup tables after the XML data. Otherwise, these data would be
2989 old_xml_begin = wim->hdr.xml_data_reshdr.offset_in_wim;
2990 old_xml_end = old_xml_begin + wim->hdr.xml_data_reshdr.size_in_wim;
2991 old_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim +
2992 wim->hdr.lookup_table_reshdr.size_in_wim;
2993 if (wim->hdr.integrity_table_reshdr.offset_in_wim != 0 &&
2994 wim->hdr.integrity_table_reshdr.offset_in_wim < old_xml_end) {
2995 WARNING("Didn't expect the integrity table to be before the XML data");
2996 ret = WIMLIB_ERR_RESOURCE_ORDER;
2997 goto out_restore_memory_hdr;
3000 if (old_lookup_table_end > old_xml_begin) {
3001 WARNING("Didn't expect the lookup table to be after the XML data");
3002 ret = WIMLIB_ERR_RESOURCE_ORDER;
3003 goto out_restore_memory_hdr;
3006 /* Set @old_wim_end, which indicates the point beyond which we don't
3007 * allow any file and metadata resources to appear without returning
3008 * WIMLIB_ERR_RESOURCE_ORDER (due to the fact that we would otherwise
3009 * overwrite these resources). */
3010 if (!wim->deletion_occurred && !any_images_modified(wim)) {
3011 /* If no images have been modified and no images have been
3012 * deleted, a new lookup table does not need to be written. We
3013 * shall write the new XML data and optional integrity table
3014 * immediately after the lookup table. Note that this may
3015 * overwrite an existing integrity table. */
3016 DEBUG("Skipping writing lookup table "
3017 "(no images modified or deleted)");
3018 old_wim_end = old_lookup_table_end;
3019 write_flags |= WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE |
3020 WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML;
3021 } else if (wim->hdr.integrity_table_reshdr.offset_in_wim != 0) {
3022 /* Old WIM has an integrity table; begin writing new streams
3024 old_wim_end = wim->hdr.integrity_table_reshdr.offset_in_wim +
3025 wim->hdr.integrity_table_reshdr.size_in_wim;
3027 /* No existing integrity table; begin writing new streams after
3028 * the old XML data. */
3029 old_wim_end = old_xml_end;
3032 ret = check_resource_offsets(wim, old_wim_end);
3034 goto out_restore_memory_hdr;
3036 ret = prepare_stream_list_for_write(wim, WIMLIB_ALL_IMAGES, write_flags,
3037 &stream_list, &lookup_table_list,
3040 goto out_restore_memory_hdr;
3042 ret = open_wim_writable(wim, wim->filename, O_RDWR);
3044 goto out_restore_memory_hdr;
3046 ret = lock_wim(wim, wim->out_fd.fd);
3050 /* Set WIM_HDR_FLAG_WRITE_IN_PROGRESS flag in header. */
3051 wim->hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
3052 ret = write_wim_header_flags(wim->hdr.flags, &wim->out_fd);
3054 ERROR_WITH_ERRNO("Error updating WIM header flags");
3055 goto out_unlock_wim;
3058 if (filedes_seek(&wim->out_fd, old_wim_end) == -1) {
3059 ERROR_WITH_ERRNO("Can't seek to end of WIM");
3060 ret = WIMLIB_ERR_WRITE;
3061 goto out_restore_physical_hdr;
3064 ret = wim_write_stream_list(wim,
3073 ret = write_wim_metadata_resources(wim, WIMLIB_ALL_IMAGES,
3074 write_flags, progress_func);
3078 write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE;
3079 ret = finish_write(wim, WIMLIB_ALL_IMAGES, write_flags,
3080 progress_func, &lookup_table_list);
3084 wim->wim_locked = 0;
3088 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
3089 WARNING("Truncating `%"TS"' to its original size (%"PRIu64" bytes)",
3090 wim->filename, old_wim_end);
3091 /* Return value of ftruncate() is ignored because this is
3092 * already an error path. */
3093 (void)ftruncate(wim->out_fd.fd, old_wim_end);
3095 out_restore_physical_hdr:
3096 (void)write_wim_header_flags(hdr_save.flags, &wim->out_fd);
3098 wim->wim_locked = 0;
3100 (void)close_wim_writable(wim, write_flags);
3101 out_restore_memory_hdr:
3102 memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header));
3107 overwrite_wim_via_tmpfile(WIMStruct *wim, int write_flags,
3108 unsigned num_threads,
3109 wimlib_progress_func_t progress_func)
3111 size_t wim_name_len;
3114 DEBUG("Overwriting `%"TS"' via a temporary file", wim->filename);
3116 /* Write the WIM to a temporary file in the same directory as the
3118 wim_name_len = tstrlen(wim->filename);
3119 tchar tmpfile[wim_name_len + 10];
3120 tmemcpy(tmpfile, wim->filename, wim_name_len);
3121 randomize_char_array_with_alnum(tmpfile + wim_name_len, 9);
3122 tmpfile[wim_name_len + 9] = T('\0');
3124 ret = wimlib_write(wim, tmpfile, WIMLIB_ALL_IMAGES,
3125 write_flags | WIMLIB_WRITE_FLAG_FSYNC,
3126 num_threads, progress_func);
3132 if (filedes_valid(&wim->in_fd)) {
3133 filedes_close(&wim->in_fd);
3134 filedes_invalidate(&wim->in_fd);
3137 /* Rename the new WIM file to the original WIM file. Note: on Windows
3138 * this actually calls win32_rename_replacement(), not _wrename(), so
3139 * that removing the existing destination file can be handled. */
3140 DEBUG("Renaming `%"TS"' to `%"TS"'", tmpfile, wim->filename);
3141 ret = trename(tmpfile, wim->filename);
3143 ERROR_WITH_ERRNO("Failed to rename `%"TS"' to `%"TS"'",
3144 tmpfile, wim->filename);
3151 return WIMLIB_ERR_RENAME;
3154 if (progress_func) {
3155 union wimlib_progress_info progress;
3156 progress.rename.from = tmpfile;
3157 progress.rename.to = wim->filename;
3158 progress_func(WIMLIB_PROGRESS_MSG_RENAME, &progress);
3163 /* Determine if the specified WIM file may be updated by appending in-place
3164 * rather than writing and replacing it with an entirely new file. */
3166 can_overwrite_wim_inplace(const WIMStruct *wim, int write_flags)
3168 /* REBUILD flag forces full rebuild. */
3169 if (write_flags & WIMLIB_WRITE_FLAG_REBUILD)
3172 /* Deletions cause full rebuild by default. */
3173 if (wim->deletion_occurred && !(write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE))
3176 /* Pipable WIMs cannot be updated in place, nor can a non-pipable WIM be
3177 * turned into a pipable WIM in-place. */
3178 if (wim_is_pipable(wim) || (write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
3181 /* The default compression type and compression chunk size selected for
3182 * the output WIM must be the same as those currently used for the WIM.
3184 if (wim->compression_type != wim->out_compression_type)
3186 if (wim->chunk_size != wim->out_chunk_size)
3192 /* API function documented in wimlib.h */
3194 wimlib_overwrite(WIMStruct *wim, int write_flags,
3195 unsigned num_threads,
3196 wimlib_progress_func_t progress_func)
3201 if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC)
3202 return WIMLIB_ERR_INVALID_PARAM;
3205 return WIMLIB_ERR_NO_FILENAME;
3207 orig_hdr_flags = wim->hdr.flags;
3208 if (write_flags & WIMLIB_WRITE_FLAG_IGNORE_READONLY_FLAG)
3209 wim->hdr.flags &= ~WIM_HDR_FLAG_READONLY;
3210 ret = can_modify_wim(wim);
3211 wim->hdr.flags = orig_hdr_flags;
3215 if (can_overwrite_wim_inplace(wim, write_flags)) {
3216 ret = overwrite_wim_inplace(wim, write_flags, num_threads,
3218 if (ret != WIMLIB_ERR_RESOURCE_ORDER)
3220 WARNING("Falling back to re-building entire WIM");
3222 return overwrite_wim_via_tmpfile(wim, write_flags, num_threads,