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, 2014 Eric Biggers
11 * This file is free software; you can redistribute it and/or modify it under
12 * the terms of the GNU Lesser General Public License as published by the Free
13 * Software Foundation; either version 3 of the License, or (at your option) any
16 * This file is distributed in the hope that it will be useful, but WITHOUT
17 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
18 * FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
21 * You should have received a copy of the GNU Lesser General Public License
22 * along with this file; if not, see http://www.gnu.org/licenses/.
29 #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK)
30 /* On BSD, this should be included before "wimlib/list.h" so that "wimlib/list.h" can
31 * overwrite the LIST_HEAD macro. */
32 # include <sys/file.h>
40 #include "wimlib/alloca.h"
41 #include "wimlib/assert.h"
42 #include "wimlib/chunk_compressor.h"
43 #include "wimlib/endianness.h"
44 #include "wimlib/error.h"
45 #include "wimlib/file_io.h"
46 #include "wimlib/header.h"
47 #include "wimlib/inode.h"
48 #include "wimlib/integrity.h"
49 #include "wimlib/lookup_table.h"
50 #include "wimlib/metadata.h"
51 #include "wimlib/paths.h"
52 #include "wimlib/progress.h"
53 #include "wimlib/resource.h"
54 #include "wimlib/solid.h"
56 # include "wimlib/win32.h" /* win32_rename_replacement() */
58 #include "wimlib/write.h"
59 #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_SOLID 0x00000004
66 #define WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE 0x00000008
69 write_flags_to_resource_flags(int write_flags)
71 int write_resource_flags = 0;
73 if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
74 write_resource_flags |= WRITE_RESOURCE_FLAG_RECOMPRESS;
75 if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
76 write_resource_flags |= WRITE_RESOURCE_FLAG_PIPABLE;
77 if (write_flags & WIMLIB_WRITE_FLAG_SOLID)
78 write_resource_flags |= WRITE_RESOURCE_FLAG_SOLID;
79 if (write_flags & WIMLIB_WRITE_FLAG_SEND_DONE_WITH_FILE_MESSAGES)
80 write_resource_flags |= WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE;
81 return write_resource_flags;
84 struct filter_context {
89 /* Determine specified stream should be filtered out from the write.
93 * < 0 : The stream should be hard-filtered; that is, not included in the
95 * 0 : The stream should not be filtered out.
96 * > 0 : The stream should be soft-filtered; that is, it already exists in the
97 * WIM file and may not need to be written again.
100 stream_filtered(const struct wim_lookup_table_entry *lte,
101 const struct filter_context *ctx)
109 write_flags = ctx->write_flags;
112 if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE &&
113 lte->resource_location == RESOURCE_IN_WIM &&
114 lte->rspec->wim == wim)
117 if (write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS &&
118 lte->resource_location == RESOURCE_IN_WIM &&
119 lte->rspec->wim != wim)
126 stream_hard_filtered(const struct wim_lookup_table_entry *lte,
127 struct filter_context *ctx)
129 return stream_filtered(lte, ctx) < 0;
133 may_soft_filter_streams(const struct filter_context *ctx)
137 return ctx->write_flags & WIMLIB_WRITE_FLAG_OVERWRITE;
141 may_hard_filter_streams(const struct filter_context *ctx)
145 return ctx->write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS;
149 may_filter_streams(const struct filter_context *ctx)
151 return (may_soft_filter_streams(ctx) ||
152 may_hard_filter_streams(ctx));
156 /* Return true if the specified resource is compressed and the compressed data
157 * can be reused with the specified output parameters. */
159 can_raw_copy(const struct wim_lookup_table_entry *lte,
160 int write_resource_flags, int out_ctype, u32 out_chunk_size)
162 const struct wim_resource_spec *rspec;
164 if (write_resource_flags & WRITE_RESOURCE_FLAG_RECOMPRESS)
167 if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE)
170 if (lte->resource_location != RESOURCE_IN_WIM)
175 if (rspec->is_pipable != !!(write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE))
178 if (rspec->flags & WIM_RESHDR_FLAG_COMPRESSED) {
179 /* Normal compressed resource: Must use same compression type
181 return (rspec->compression_type == out_ctype &&
182 rspec->chunk_size == out_chunk_size);
185 if ((rspec->flags & WIM_RESHDR_FLAG_SOLID) &&
186 (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID))
188 /* Solid resource: Such resources may contain multiple streams,
189 * and in general only a subset of them need to be written. As
190 * a heuristic, re-use the raw data if more than two-thirds the
191 * uncompressed size is being written. */
193 /* Note: solid resources contain a header that specifies the
194 * compression type and chunk size; therefore we don't need to
195 * check if they are compatible with @out_ctype and
196 * @out_chunk_size. */
198 struct wim_lookup_table_entry *res_stream;
201 list_for_each_entry(res_stream, &rspec->stream_list, rspec_node)
202 if (res_stream->will_be_in_output_wim)
203 write_size += res_stream->size;
205 return (write_size > rspec->uncompressed_size * 2 / 3);
212 filter_resource_flags(u8 flags)
214 return (flags & ~(WIM_RESHDR_FLAG_SOLID |
215 WIM_RESHDR_FLAG_COMPRESSED |
216 WIM_RESHDR_FLAG_SPANNED |
217 WIM_RESHDR_FLAG_FREE));
221 stream_set_out_reshdr_for_reuse(struct wim_lookup_table_entry *lte)
223 const struct wim_resource_spec *rspec;
225 wimlib_assert(lte->resource_location == RESOURCE_IN_WIM);
228 if (rspec->flags & WIM_RESHDR_FLAG_SOLID) {
230 wimlib_assert(lte->flags & WIM_RESHDR_FLAG_SOLID);
232 lte->out_reshdr.offset_in_wim = lte->offset_in_res;
233 lte->out_reshdr.uncompressed_size = 0;
234 lte->out_reshdr.size_in_wim = lte->size;
236 lte->out_res_offset_in_wim = rspec->offset_in_wim;
237 lte->out_res_size_in_wim = rspec->size_in_wim;
238 lte->out_res_uncompressed_size = rspec->uncompressed_size;
240 wimlib_assert(!(lte->flags & WIM_RESHDR_FLAG_SOLID));
242 lte->out_reshdr.offset_in_wim = rspec->offset_in_wim;
243 lte->out_reshdr.uncompressed_size = rspec->uncompressed_size;
244 lte->out_reshdr.size_in_wim = rspec->size_in_wim;
246 lte->out_reshdr.flags = lte->flags;
250 /* Write the header for a stream in a pipable WIM. */
252 write_pwm_stream_header(const struct wim_lookup_table_entry *lte,
253 struct filedes *out_fd,
254 int additional_reshdr_flags)
256 struct pwm_stream_hdr stream_hdr;
260 stream_hdr.magic = cpu_to_le64(PWM_STREAM_MAGIC);
261 stream_hdr.uncompressed_size = cpu_to_le64(lte->size);
262 if (additional_reshdr_flags & PWM_RESHDR_FLAG_UNHASHED) {
263 zero_out_hash(stream_hdr.hash);
265 wimlib_assert(!lte->unhashed);
266 copy_hash(stream_hdr.hash, lte->hash);
269 reshdr_flags = filter_resource_flags(lte->flags);
270 reshdr_flags |= additional_reshdr_flags;
271 stream_hdr.flags = cpu_to_le32(reshdr_flags);
272 ret = full_write(out_fd, &stream_hdr, sizeof(stream_hdr));
274 ERROR_WITH_ERRNO("Write error");
278 struct write_streams_progress_data {
279 wimlib_progress_func_t progfunc;
281 union wimlib_progress_info progress;
282 uint64_t next_progress;
286 do_write_streams_progress(struct write_streams_progress_data *progress_data,
291 union wimlib_progress_info *progress = &progress_data->progress;
295 progress->write_streams.total_bytes -= complete_size;
296 progress->write_streams.total_streams -= complete_count;
297 if (progress_data->next_progress != ~(uint64_t)0 &&
298 progress_data->next_progress > progress->write_streams.total_bytes)
300 progress_data->next_progress = progress->write_streams.total_bytes;
303 progress->write_streams.completed_bytes += complete_size;
304 progress->write_streams.completed_streams += complete_count;
307 if (progress->write_streams.completed_bytes >= progress_data->next_progress)
309 ret = call_progress(progress_data->progfunc,
310 WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
312 progress_data->progctx);
316 if (progress_data->next_progress == progress->write_streams.total_bytes) {
317 progress_data->next_progress = ~(uint64_t)0;
319 /* Handle rate-limiting of messages */
321 /* Send new message as soon as another 1/128 of the
322 * total has been written. (Arbitrary number.) */
323 progress_data->next_progress =
324 progress->write_streams.completed_bytes +
325 progress->write_streams.total_bytes / 128;
327 /* ... Unless that would be more than 5000000 bytes, in
328 * which case send the next after the next 5000000
329 * bytes. (Another arbitrary number.) */
330 if (progress->write_streams.completed_bytes + 5000000 <
331 progress_data->next_progress)
332 progress_data->next_progress =
333 progress->write_streams.completed_bytes + 5000000;
335 /* ... But always send a message as soon as we're
336 * completely done. */
337 if (progress->write_streams.total_bytes <
338 progress_data->next_progress)
339 progress_data->next_progress =
340 progress->write_streams.total_bytes;
346 struct write_streams_ctx {
347 /* File descriptor the streams are being written to. */
348 struct filedes *out_fd;
350 /* Lookup table for the WIMStruct on whose behalf the streams are being
352 struct wim_lookup_table *lookup_table;
354 /* Compression format to use. */
357 /* Maximum uncompressed chunk size in compressed resources to use. */
360 /* Flags that affect how the streams will be written. */
361 int write_resource_flags;
363 /* Data used for issuing WRITE_STREAMS progress. */
364 struct write_streams_progress_data progress_data;
366 struct filter_context *filter_ctx;
368 /* Upper bound on the total number of bytes that need to be compressed.
370 u64 num_bytes_to_compress;
372 /* Pointer to the chunk_compressor implementation being used for
373 * compressing chunks of data, or NULL if chunks are being written
375 struct chunk_compressor *compressor;
377 /* Buffer for dividing the read data into chunks of size
378 * @out_chunk_size. */
381 /* Number of bytes in @chunk_buf that are currently filled. */
382 size_t chunk_buf_filled;
384 /* List of streams that currently have chunks being compressed. */
385 struct list_head pending_streams;
387 /* List of streams in the solid resource. Streams are moved here after
388 * @pending_streams only when writing a solid resource. */
389 struct list_head solid_streams;
391 /* Current uncompressed offset in the stream being read. */
392 u64 cur_read_stream_offset;
394 /* Uncompressed size of the stream currently being read. */
395 u64 cur_read_stream_size;
397 /* Current uncompressed offset in the stream being written. */
398 u64 cur_write_stream_offset;
400 /* Uncompressed size of resource currently being written. */
401 u64 cur_write_res_size;
403 /* Array that is filled in with compressed chunk sizes as a resource is
407 /* Index of next entry in @chunk_csizes to fill in. */
410 /* Number of entries in @chunk_csizes currently allocated. */
411 size_t num_alloc_chunks;
413 /* Offset in the output file of the start of the chunks of the resource
414 * currently being written. */
415 u64 chunks_start_offset;
418 /* Reserve space for the chunk table and prepare to accumulate the chunk table
421 begin_chunk_table(struct write_streams_ctx *ctx, u64 res_expected_size)
423 u64 expected_num_chunks;
424 u64 expected_num_chunk_entries;
428 /* Calculate the number of chunks and chunk entries that should be
429 * needed for the resource. These normally will be the final values,
430 * but in SOLID mode some of the streams we're planning to write into
431 * the resource may be duplicates, and therefore discarded, potentially
432 * decreasing the number of chunk entries needed. */
433 expected_num_chunks = DIV_ROUND_UP(res_expected_size, ctx->out_chunk_size);
434 expected_num_chunk_entries = expected_num_chunks;
435 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID))
436 expected_num_chunk_entries--;
438 /* Make sure the chunk_csizes array is long enough to store the
439 * compressed size of each chunk. */
440 if (expected_num_chunks > ctx->num_alloc_chunks) {
441 u64 new_length = expected_num_chunks + 50;
443 if ((size_t)new_length != new_length) {
444 ERROR("Resource size too large (%"PRIu64" bytes!",
446 return WIMLIB_ERR_NOMEM;
449 FREE(ctx->chunk_csizes);
450 ctx->chunk_csizes = MALLOC(new_length * sizeof(ctx->chunk_csizes[0]));
451 if (ctx->chunk_csizes == NULL) {
452 ctx->num_alloc_chunks = 0;
453 return WIMLIB_ERR_NOMEM;
455 ctx->num_alloc_chunks = new_length;
458 ctx->chunk_index = 0;
460 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE)) {
461 /* Reserve space for the chunk table in the output file. In the
462 * case of solid resources this reserves the upper bound for the
463 * needed space, not necessarily the exact space which will
464 * prove to be needed. At this point, we just use @chunk_csizes
465 * for a buffer of 0's because the actual compressed chunk sizes
467 reserve_size = expected_num_chunk_entries *
468 get_chunk_entry_size(res_expected_size,
469 0 != (ctx->write_resource_flags &
470 WRITE_RESOURCE_FLAG_SOLID));
471 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID)
472 reserve_size += sizeof(struct alt_chunk_table_header_disk);
473 memset(ctx->chunk_csizes, 0, reserve_size);
474 ret = full_write(ctx->out_fd, ctx->chunk_csizes, reserve_size);
482 begin_write_resource(struct write_streams_ctx *ctx, u64 res_expected_size)
486 wimlib_assert(res_expected_size != 0);
488 if (ctx->compressor != NULL) {
489 ret = begin_chunk_table(ctx, res_expected_size);
494 /* Output file descriptor is now positioned at the offset at which to
495 * write the first chunk of the resource. */
496 ctx->chunks_start_offset = ctx->out_fd->offset;
497 ctx->cur_write_stream_offset = 0;
498 ctx->cur_write_res_size = res_expected_size;
503 end_chunk_table(struct write_streams_ctx *ctx, u64 res_actual_size,
504 u64 *res_start_offset_ret, u64 *res_store_size_ret)
506 size_t actual_num_chunks;
507 size_t actual_num_chunk_entries;
508 size_t chunk_entry_size;
511 actual_num_chunks = ctx->chunk_index;
512 actual_num_chunk_entries = actual_num_chunks;
513 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID))
514 actual_num_chunk_entries--;
516 chunk_entry_size = get_chunk_entry_size(res_actual_size,
517 0 != (ctx->write_resource_flags &
518 WRITE_RESOURCE_FLAG_SOLID));
520 typedef le64 _may_alias_attribute aliased_le64_t;
521 typedef le32 _may_alias_attribute aliased_le32_t;
523 if (chunk_entry_size == 4) {
524 aliased_le32_t *entries = (aliased_le32_t*)ctx->chunk_csizes;
526 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
527 for (size_t i = 0; i < actual_num_chunk_entries; i++)
528 entries[i] = cpu_to_le32(ctx->chunk_csizes[i]);
530 u32 offset = ctx->chunk_csizes[0];
531 for (size_t i = 0; i < actual_num_chunk_entries; i++) {
532 u32 next_size = ctx->chunk_csizes[i + 1];
533 entries[i] = cpu_to_le32(offset);
538 aliased_le64_t *entries = (aliased_le64_t*)ctx->chunk_csizes;
540 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
541 for (size_t i = 0; i < actual_num_chunk_entries; i++)
542 entries[i] = cpu_to_le64(ctx->chunk_csizes[i]);
544 u64 offset = ctx->chunk_csizes[0];
545 for (size_t i = 0; i < actual_num_chunk_entries; i++) {
546 u64 next_size = ctx->chunk_csizes[i + 1];
547 entries[i] = cpu_to_le64(offset);
553 size_t chunk_table_size = actual_num_chunk_entries * chunk_entry_size;
554 u64 res_start_offset;
557 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
558 ret = full_write(ctx->out_fd, ctx->chunk_csizes, chunk_table_size);
561 res_end_offset = ctx->out_fd->offset;
562 res_start_offset = ctx->chunks_start_offset;
564 res_end_offset = ctx->out_fd->offset;
566 u64 chunk_table_offset;
568 chunk_table_offset = ctx->chunks_start_offset - chunk_table_size;
570 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
571 struct alt_chunk_table_header_disk hdr;
573 hdr.res_usize = cpu_to_le64(res_actual_size);
574 hdr.chunk_size = cpu_to_le32(ctx->out_chunk_size);
575 hdr.compression_format = cpu_to_le32(ctx->out_ctype);
577 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 1);
578 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 2);
579 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3);
581 ret = full_pwrite(ctx->out_fd, &hdr, sizeof(hdr),
582 chunk_table_offset - sizeof(hdr));
585 res_start_offset = chunk_table_offset - sizeof(hdr);
587 res_start_offset = chunk_table_offset;
590 ret = full_pwrite(ctx->out_fd, ctx->chunk_csizes,
591 chunk_table_size, chunk_table_offset);
596 *res_start_offset_ret = res_start_offset;
597 *res_store_size_ret = res_end_offset - res_start_offset;
602 ERROR_WITH_ERRNO("Write error");
606 /* Finish writing a WIM resource by writing or updating the chunk table (if not
607 * writing the data uncompressed) and loading its metadata into @out_reshdr. */
609 end_write_resource(struct write_streams_ctx *ctx, struct wim_reshdr *out_reshdr)
613 u64 res_uncompressed_size;
614 u64 res_offset_in_wim;
616 wimlib_assert(ctx->cur_write_stream_offset == ctx->cur_write_res_size ||
617 (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID));
618 res_uncompressed_size = ctx->cur_write_res_size;
620 if (ctx->compressor) {
621 ret = end_chunk_table(ctx, res_uncompressed_size,
622 &res_offset_in_wim, &res_size_in_wim);
626 res_offset_in_wim = ctx->chunks_start_offset;
627 res_size_in_wim = ctx->out_fd->offset - res_offset_in_wim;
629 out_reshdr->uncompressed_size = res_uncompressed_size;
630 out_reshdr->size_in_wim = res_size_in_wim;
631 out_reshdr->offset_in_wim = res_offset_in_wim;
632 DEBUG("Finished writing resource: %"PRIu64" => %"PRIu64" @ %"PRIu64"",
633 res_uncompressed_size, res_size_in_wim, res_offset_in_wim);
637 /* No more data streams of the file at @path are needed. */
639 done_with_file(const tchar *path, wimlib_progress_func_t progfunc, void *progctx)
641 union wimlib_progress_info info;
643 info.done_with_file.path_to_file = path;
645 return call_progress(progfunc, WIMLIB_PROGRESS_MSG_DONE_WITH_FILE,
650 is_file_stream(const struct wim_lookup_table_entry *lte)
652 return lte->resource_location == RESOURCE_IN_FILE_ON_DISK
654 || lte->resource_location == RESOURCE_IN_WINNT_FILE_ON_DISK
655 || lte->resource_location == RESOURCE_WIN32_ENCRYPTED
661 do_done_with_stream(struct wim_lookup_table_entry *lte,
662 wimlib_progress_func_t progfunc, void *progctx)
665 struct wim_inode *inode;
667 if (!lte->may_send_done_with_file)
670 inode = lte->file_inode;
672 wimlib_assert(inode != NULL);
673 wimlib_assert(inode->num_remaining_streams > 0);
674 if (--inode->num_remaining_streams > 0)
678 /* XXX: This logic really should be somewhere else. */
680 /* We want the path to the file, but lte->file_on_disk might actually
681 * refer to a named data stream. Temporarily strip the named data
682 * stream from the path. */
683 wchar_t *p_colon = NULL;
684 wchar_t *p_question_mark = NULL;
685 const wchar_t *p_stream_name;
687 p_stream_name = path_stream_name(lte->file_on_disk);
688 if (unlikely(p_stream_name)) {
689 p_colon = (wchar_t *)(p_stream_name - 1);
690 wimlib_assert(*p_colon == L':');
694 /* We also should use a fake Win32 path instead of a NT path */
695 if (!wcsncmp(lte->file_on_disk, L"\\??\\", 4)) {
696 p_question_mark = <e->file_on_disk[1];
697 *p_question_mark = L'\\';
701 ret = done_with_file(lte->file_on_disk, progfunc, progctx);
707 *p_question_mark = L'?';
712 /* Handle WIMLIB_WRITE_FLAG_SEND_DONE_WITH_FILE_MESSAGES mode. */
714 done_with_stream(struct wim_lookup_table_entry *lte,
715 struct write_streams_ctx *ctx)
717 if (likely(!(ctx->write_resource_flags &
718 WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE)))
720 return do_done_with_stream(lte, ctx->progress_data.progfunc,
721 ctx->progress_data.progctx);
724 /* Begin processing a stream for writing. */
726 write_stream_begin_read(struct wim_lookup_table_entry *lte, void *_ctx)
728 struct write_streams_ctx *ctx = _ctx;
731 wimlib_assert(lte->size > 0);
733 ctx->cur_read_stream_offset = 0;
734 ctx->cur_read_stream_size = lte->size;
736 /* As an optimization, we allow some streams to be "unhashed", meaning
737 * their SHA1 message digests are unknown. This is the case with
738 * streams that are added by scanning a directry tree with
739 * wimlib_add_image(), for example. Since WIM uses single-instance
740 * streams, we don't know whether such each such stream really need to
741 * written until it is actually checksummed, unless it has a unique
742 * size. In such cases we read and checksum the stream in this
743 * function, thereby advancing ahead of read_stream_list(), which will
744 * still provide the data again to write_stream_process_chunk(). This
745 * is okay because an unhashed stream cannot be in a WIM resource, which
746 * might be costly to decompress. */
747 if (ctx->lookup_table != NULL && lte->unhashed && !lte->unique_size) {
749 struct wim_lookup_table_entry *lte_new;
751 ret = hash_unhashed_stream(lte, ctx->lookup_table, <e_new);
754 if (lte_new != lte) {
755 /* Duplicate stream detected. */
757 if (lte_new->will_be_in_output_wim ||
758 stream_filtered(lte_new, ctx->filter_ctx))
760 /* The duplicate stream is already being
761 * included in the output WIM, or it would be
762 * filtered out if it had been. Skip writing
763 * this stream (and reading it again) entirely,
764 * passing its output reference count to the
765 * duplicate stream in the former case. */
766 DEBUG("Discarding duplicate stream of "
767 "length %"PRIu64, lte->size);
768 ret = do_write_streams_progress(&ctx->progress_data,
771 list_del(<e->write_streams_list);
772 list_del(<e->lookup_table_list);
773 if (lte_new->will_be_in_output_wim)
774 lte_new->out_refcnt += lte->out_refcnt;
775 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID)
776 ctx->cur_write_res_size -= lte->size;
778 ret = done_with_stream(lte, ctx);
779 free_lookup_table_entry(lte);
782 return BEGIN_STREAM_STATUS_SKIP_STREAM;
784 /* The duplicate stream can validly be written,
785 * but was not marked as such. Discard the
786 * current stream entry and use the duplicate,
787 * but actually freeing the current entry must
788 * wait until read_stream_list() has finished
789 * reading its data. */
790 DEBUG("Stream duplicate, but not already "
791 "selected for writing.");
792 list_replace(<e->write_streams_list,
793 <e_new->write_streams_list);
794 list_replace(<e->lookup_table_list,
795 <e_new->lookup_table_list);
796 lte->will_be_in_output_wim = 0;
797 lte_new->out_refcnt = lte->out_refcnt;
798 lte_new->will_be_in_output_wim = 1;
799 lte_new->may_send_done_with_file = 0;
804 list_move_tail(<e->write_streams_list, &ctx->pending_streams);
808 /* Rewrite a stream that was just written compressed as uncompressed instead.
809 * This function is optional, but if a stream did not compress to less than its
810 * original size, it might as well be written uncompressed. */
812 write_stream_uncompressed(struct wim_lookup_table_entry *lte,
813 struct filedes *out_fd)
816 u64 begin_offset = lte->out_reshdr.offset_in_wim;
817 u64 end_offset = out_fd->offset;
819 if (filedes_seek(out_fd, begin_offset) == -1)
822 ret = extract_full_stream_to_fd(lte, out_fd);
824 /* Error reading the uncompressed data. */
825 if (out_fd->offset == begin_offset &&
826 filedes_seek(out_fd, end_offset) != -1)
828 /* Nothing was actually written yet, and we successfully
829 * seeked to the end of the compressed resource, so
830 * don't issue a hard error; just keep the compressed
831 * resource instead. */
832 WARNING("Recovered compressed stream of "
833 "size %"PRIu64", continuing on.",
840 wimlib_assert(out_fd->offset - begin_offset == lte->size);
842 if (out_fd->offset < end_offset &&
843 0 != ftruncate(out_fd->fd, out_fd->offset))
845 ERROR_WITH_ERRNO("Can't truncate output file to "
846 "offset %"PRIu64, out_fd->offset);
847 return WIMLIB_ERR_WRITE;
850 lte->out_reshdr.size_in_wim = lte->size;
851 lte->out_reshdr.flags &= ~(WIM_RESHDR_FLAG_COMPRESSED |
852 WIM_RESHDR_FLAG_SOLID);
856 /* Returns true if the specified stream should be truncated from the WIM file
857 * and re-written as uncompressed. lte->out_reshdr must be filled in from the
858 * initial write of the stream. */
860 should_rewrite_stream_uncompressed(const struct write_streams_ctx *ctx,
861 const struct wim_lookup_table_entry *lte)
863 /* If the compressed data is smaller than the uncompressed data, prefer
864 * the compressed data. */
865 if (lte->out_reshdr.size_in_wim < lte->out_reshdr.uncompressed_size)
868 /* If we're not actually writing compressed data, then there's no need
870 if (!ctx->compressor)
873 /* If writing a pipable WIM, everything we write to the output is final
874 * (it might actually be a pipe!). */
875 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE)
878 /* If the stream that would need to be re-read is located in a solid
879 * resource in another WIM file, then re-reading it would be costly. So
882 * Exception: if the compressed size happens to be *exactly* the same as
883 * the uncompressed size, then the stream *must* be written uncompressed
884 * in order to remain compatible with the Windows Overlay Filesystem
885 * Filter Driver (WOF).
887 * TODO: we are currently assuming that the optimization for
888 * single-chunk resources in maybe_rewrite_stream_uncompressed()
889 * prevents this case from being triggered too often. To fully prevent
890 * excessive decompressions in degenerate cases, we really should
891 * obtain the uncompressed data by decompressing the compressed data we
892 * wrote to the output file.
894 if ((lte->flags & WIM_RESHDR_FLAG_SOLID) &&
895 (lte->out_reshdr.size_in_wim != lte->out_reshdr.uncompressed_size))
902 maybe_rewrite_stream_uncompressed(struct write_streams_ctx *ctx,
903 struct wim_lookup_table_entry *lte)
905 if (!should_rewrite_stream_uncompressed(ctx, lte))
908 /* Regular (non-solid) WIM resources with exactly one chunk and
909 * compressed size equal to uncompressed size are exactly the same as
910 * the corresponding compressed data --- since there must be 0 entries
911 * in the chunk table and the only chunk must be stored uncompressed.
912 * In this case, there's no need to rewrite anything. */
913 if (ctx->chunk_index == 1 &&
914 lte->out_reshdr.size_in_wim == lte->out_reshdr.uncompressed_size)
916 lte->out_reshdr.flags &= ~WIM_RESHDR_FLAG_COMPRESSED;
920 return write_stream_uncompressed(lte, ctx->out_fd);
923 /* Write the next chunk of (typically compressed) data to the output WIM,
924 * handling the writing of the chunk table. */
926 write_chunk(struct write_streams_ctx *ctx, const void *cchunk,
927 size_t csize, size_t usize)
931 struct wim_lookup_table_entry *lte;
932 u32 completed_stream_count;
935 lte = list_entry(ctx->pending_streams.next,
936 struct wim_lookup_table_entry, write_streams_list);
938 if (ctx->cur_write_stream_offset == 0 &&
939 !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID))
941 /* Starting to write a new stream in non-solid mode. */
943 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
944 int additional_reshdr_flags = 0;
945 if (ctx->compressor != NULL)
946 additional_reshdr_flags |= WIM_RESHDR_FLAG_COMPRESSED;
948 DEBUG("Writing pipable WIM stream header "
949 "(offset=%"PRIu64")", ctx->out_fd->offset);
951 ret = write_pwm_stream_header(lte, ctx->out_fd,
952 additional_reshdr_flags);
957 ret = begin_write_resource(ctx, lte->size);
962 if (ctx->compressor != NULL) {
963 /* Record the compresed chunk size. */
964 wimlib_assert(ctx->chunk_index < ctx->num_alloc_chunks);
965 ctx->chunk_csizes[ctx->chunk_index++] = csize;
967 /* If writing a pipable WIM, before the chunk data write a chunk
968 * header that provides the compressed chunk size. */
969 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
970 struct pwm_chunk_hdr chunk_hdr = {
971 .compressed_size = cpu_to_le32(csize),
973 ret = full_write(ctx->out_fd, &chunk_hdr,
980 /* Write the chunk data. */
981 ret = full_write(ctx->out_fd, cchunk, csize);
985 ctx->cur_write_stream_offset += usize;
987 completed_size = usize;
988 completed_stream_count = 0;
989 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
990 /* Wrote chunk in solid mode. It may have finished multiple
992 struct wim_lookup_table_entry *next_lte;
994 while (lte && ctx->cur_write_stream_offset >= lte->size) {
996 ctx->cur_write_stream_offset -= lte->size;
998 if (ctx->cur_write_stream_offset)
999 next_lte = list_entry(lte->write_streams_list.next,
1000 struct wim_lookup_table_entry,
1001 write_streams_list);
1005 ret = done_with_stream(lte, ctx);
1008 list_move_tail(<e->write_streams_list, &ctx->solid_streams);
1009 completed_stream_count++;
1014 /* Wrote chunk in non-solid mode. It may have finished a
1016 if (ctx->cur_write_stream_offset == lte->size) {
1018 wimlib_assert(ctx->cur_write_stream_offset ==
1019 ctx->cur_write_res_size);
1021 ret = end_write_resource(ctx, <e->out_reshdr);
1025 lte->out_reshdr.flags = filter_resource_flags(lte->flags);
1026 if (ctx->compressor != NULL)
1027 lte->out_reshdr.flags |= WIM_RESHDR_FLAG_COMPRESSED;
1029 ret = maybe_rewrite_stream_uncompressed(ctx, lte);
1033 wimlib_assert(lte->out_reshdr.uncompressed_size == lte->size);
1035 ctx->cur_write_stream_offset = 0;
1037 ret = done_with_stream(lte, ctx);
1040 list_del(<e->write_streams_list);
1041 completed_stream_count++;
1045 return do_write_streams_progress(&ctx->progress_data,
1046 completed_size, completed_stream_count,
1050 ERROR_WITH_ERRNO("Write error");
1055 submit_chunk_for_compression(struct write_streams_ctx *ctx,
1056 const void *chunk, size_t size)
1058 /* While we are unable to submit the chunk for compression (due to too
1059 * many chunks already outstanding), retrieve and write the next
1060 * compressed chunk. */
1061 while (!ctx->compressor->submit_chunk(ctx->compressor, chunk, size)) {
1068 bret = ctx->compressor->get_chunk(ctx->compressor,
1069 &cchunk, &csize, &usize);
1071 wimlib_assert(bret);
1073 ret = write_chunk(ctx, cchunk, csize, usize);
1080 /* Process the next chunk of data to be written to a WIM resource. */
1082 write_stream_process_chunk(const void *chunk, size_t size, void *_ctx)
1084 struct write_streams_ctx *ctx = _ctx;
1086 const u8 *chunkptr, *chunkend;
1088 wimlib_assert(size != 0);
1090 if (ctx->compressor == NULL) {
1091 /* Write chunk uncompressed. */
1092 ret = write_chunk(ctx, chunk, size, size);
1095 ctx->cur_read_stream_offset += size;
1099 /* Submit the chunk for compression, but take into account that the
1100 * @size the chunk was provided in may not correspond to the
1101 * @out_chunk_size being used for compression. */
1103 chunkend = chunkptr + size;
1105 const u8 *resized_chunk;
1106 size_t needed_chunk_size;
1108 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
1109 needed_chunk_size = ctx->out_chunk_size;
1111 u64 res_bytes_remaining;
1113 res_bytes_remaining = ctx->cur_read_stream_size -
1114 ctx->cur_read_stream_offset;
1115 needed_chunk_size = min(ctx->out_chunk_size,
1116 ctx->chunk_buf_filled +
1117 res_bytes_remaining);
1120 if (ctx->chunk_buf_filled == 0 &&
1121 chunkend - chunkptr >= needed_chunk_size)
1123 /* No intermediate buffering needed. */
1124 resized_chunk = chunkptr;
1125 chunkptr += needed_chunk_size;
1126 ctx->cur_read_stream_offset += needed_chunk_size;
1128 /* Intermediate buffering needed. */
1129 size_t bytes_consumed;
1131 bytes_consumed = min(chunkend - chunkptr,
1132 needed_chunk_size - ctx->chunk_buf_filled);
1134 memcpy(&ctx->chunk_buf[ctx->chunk_buf_filled],
1135 chunkptr, bytes_consumed);
1137 chunkptr += bytes_consumed;
1138 ctx->cur_read_stream_offset += bytes_consumed;
1139 ctx->chunk_buf_filled += bytes_consumed;
1140 if (ctx->chunk_buf_filled == needed_chunk_size) {
1141 resized_chunk = ctx->chunk_buf;
1142 ctx->chunk_buf_filled = 0;
1149 ret = submit_chunk_for_compression(ctx, resized_chunk,
1154 } while (chunkptr != chunkend);
1158 /* Finish processing a stream for writing. It may not have been completely
1159 * written yet, as the chunk_compressor implementation may still have chunks
1160 * buffered or being compressed. */
1162 write_stream_end_read(struct wim_lookup_table_entry *lte, int status, void *_ctx)
1164 struct write_streams_ctx *ctx = _ctx;
1166 wimlib_assert(ctx->cur_read_stream_offset == ctx->cur_read_stream_size || status);
1168 if (!lte->will_be_in_output_wim) {
1169 /* The 'lte' stream was a duplicate. Now that its data has
1170 * finished being read, it is being discarded in favor of the
1171 * duplicate entry. It therefore is no longer needed, and we
1172 * can fire the DONE_WITH_FILE callback because the file will
1173 * not be read again.
1175 * Note: we can't yet fire DONE_WITH_FILE for non-duplicate
1176 * streams, since it needs to be possible to re-read the file if
1177 * it does not compress to less than its original size. */
1179 status = done_with_stream(lte, ctx);
1180 free_lookup_table_entry(lte);
1181 } else if (!status && lte->unhashed && ctx->lookup_table != NULL) {
1182 /* The 'lte' stream was not a duplicate and was previously
1183 * unhashed. Since we passed COMPUTE_MISSING_STREAM_HASHES to
1184 * read_stream_list(), lte->hash is now computed and valid. So
1185 * turn this stream into a "hashed" stream. */
1186 list_del(<e->unhashed_list);
1187 lookup_table_insert(ctx->lookup_table, lte);
1193 /* Compute statistics about a list of streams that will be written.
1195 * Assumes the streams are sorted such that all streams located in each distinct
1196 * WIM (specified by WIMStruct) are together. */
1198 compute_stream_list_stats(struct list_head *stream_list,
1199 struct write_streams_ctx *ctx)
1201 struct wim_lookup_table_entry *lte;
1202 u64 total_bytes = 0;
1203 u64 num_streams = 0;
1204 u64 total_parts = 0;
1205 WIMStruct *prev_wim_part = NULL;
1207 list_for_each_entry(lte, stream_list, write_streams_list) {
1209 total_bytes += lte->size;
1210 if (lte->resource_location == RESOURCE_IN_WIM) {
1211 if (prev_wim_part != lte->rspec->wim) {
1212 prev_wim_part = lte->rspec->wim;
1217 ctx->progress_data.progress.write_streams.total_bytes = total_bytes;
1218 ctx->progress_data.progress.write_streams.total_streams = num_streams;
1219 ctx->progress_data.progress.write_streams.completed_bytes = 0;
1220 ctx->progress_data.progress.write_streams.completed_streams = 0;
1221 ctx->progress_data.progress.write_streams.compression_type = ctx->out_ctype;
1222 ctx->progress_data.progress.write_streams.total_parts = total_parts;
1223 ctx->progress_data.progress.write_streams.completed_parts = 0;
1224 ctx->progress_data.next_progress = 0;
1227 /* Find streams in @stream_list that can be copied to the output WIM in raw form
1228 * rather than compressed. Delete these streams from @stream_list and move them
1229 * to @raw_copy_streams. Return the total uncompressed size of the streams that
1230 * need to be compressed. */
1232 find_raw_copy_streams(struct list_head *stream_list,
1233 int write_resource_flags,
1236 struct list_head *raw_copy_streams)
1238 struct wim_lookup_table_entry *lte, *tmp;
1239 u64 num_bytes_to_compress = 0;
1241 INIT_LIST_HEAD(raw_copy_streams);
1243 /* Initialize temporary raw_copy_ok flag. */
1244 list_for_each_entry(lte, stream_list, write_streams_list)
1245 if (lte->resource_location == RESOURCE_IN_WIM)
1246 lte->rspec->raw_copy_ok = 0;
1248 list_for_each_entry_safe(lte, tmp, stream_list, write_streams_list) {
1249 if (lte->resource_location == RESOURCE_IN_WIM &&
1250 lte->rspec->raw_copy_ok)
1252 list_move_tail(<e->write_streams_list,
1254 } else if (can_raw_copy(lte, write_resource_flags,
1255 out_ctype, out_chunk_size))
1257 lte->rspec->raw_copy_ok = 1;
1258 list_move_tail(<e->write_streams_list,
1261 num_bytes_to_compress += lte->size;
1265 return num_bytes_to_compress;
1268 /* Copy a raw compressed resource located in another WIM file to the WIM file
1271 write_raw_copy_resource(struct wim_resource_spec *in_rspec,
1272 struct filedes *out_fd)
1274 u64 cur_read_offset;
1275 u64 end_read_offset;
1276 u8 buf[BUFFER_SIZE];
1277 size_t bytes_to_read;
1279 struct filedes *in_fd;
1280 struct wim_lookup_table_entry *lte;
1281 u64 out_offset_in_wim;
1283 DEBUG("Copying raw compressed data (size_in_wim=%"PRIu64", "
1284 "uncompressed_size=%"PRIu64")",
1285 in_rspec->size_in_wim, in_rspec->uncompressed_size);
1287 /* Copy the raw data. */
1288 cur_read_offset = in_rspec->offset_in_wim;
1289 end_read_offset = cur_read_offset + in_rspec->size_in_wim;
1291 out_offset_in_wim = out_fd->offset;
1293 if (in_rspec->is_pipable) {
1294 if (cur_read_offset < sizeof(struct pwm_stream_hdr))
1295 return WIMLIB_ERR_INVALID_PIPABLE_WIM;
1296 cur_read_offset -= sizeof(struct pwm_stream_hdr);
1297 out_offset_in_wim += sizeof(struct pwm_stream_hdr);
1299 in_fd = &in_rspec->wim->in_fd;
1300 wimlib_assert(cur_read_offset != end_read_offset);
1303 bytes_to_read = min(sizeof(buf), end_read_offset - cur_read_offset);
1305 ret = full_pread(in_fd, buf, bytes_to_read, cur_read_offset);
1309 ret = full_write(out_fd, buf, bytes_to_read);
1313 cur_read_offset += bytes_to_read;
1315 } while (cur_read_offset != end_read_offset);
1317 list_for_each_entry(lte, &in_rspec->stream_list, rspec_node) {
1318 if (lte->will_be_in_output_wim) {
1319 stream_set_out_reshdr_for_reuse(lte);
1320 if (in_rspec->flags & WIM_RESHDR_FLAG_SOLID)
1321 lte->out_res_offset_in_wim = out_offset_in_wim;
1323 lte->out_reshdr.offset_in_wim = out_offset_in_wim;
1330 /* Copy a list of raw compressed resources located in other WIM file(s) to the
1331 * WIM file being written. */
1333 write_raw_copy_resources(struct list_head *raw_copy_streams,
1334 struct filedes *out_fd,
1335 struct write_streams_progress_data *progress_data)
1337 struct wim_lookup_table_entry *lte;
1340 list_for_each_entry(lte, raw_copy_streams, write_streams_list)
1341 lte->rspec->raw_copy_ok = 1;
1343 list_for_each_entry(lte, raw_copy_streams, write_streams_list) {
1344 if (lte->rspec->raw_copy_ok) {
1345 /* Write each solid resource only one time, no matter
1346 * how many streams reference it. */
1347 ret = write_raw_copy_resource(lte->rspec, out_fd);
1350 lte->rspec->raw_copy_ok = 0;
1352 ret = do_write_streams_progress(progress_data, lte->size,
1360 /* Wait for and write all chunks pending in the compressor. */
1362 finish_remaining_chunks(struct write_streams_ctx *ctx)
1369 if (ctx->compressor == NULL)
1372 if (ctx->chunk_buf_filled != 0) {
1373 ret = submit_chunk_for_compression(ctx, ctx->chunk_buf,
1374 ctx->chunk_buf_filled);
1379 while (ctx->compressor->get_chunk(ctx->compressor, &cdata, &csize, &usize)) {
1380 ret = write_chunk(ctx, cdata, csize, usize);
1388 remove_zero_length_streams(struct list_head *stream_list)
1390 struct wim_lookup_table_entry *lte, *tmp;
1392 list_for_each_entry_safe(lte, tmp, stream_list, write_streams_list) {
1393 wimlib_assert(lte->will_be_in_output_wim);
1394 if (lte->size == 0) {
1395 list_del(<e->write_streams_list);
1396 lte->out_reshdr.offset_in_wim = 0;
1397 lte->out_reshdr.size_in_wim = 0;
1398 lte->out_reshdr.uncompressed_size = 0;
1399 lte->out_reshdr.flags = filter_resource_flags(lte->flags);
1405 init_done_with_file_info(struct list_head *stream_list)
1407 struct wim_lookup_table_entry *lte;
1409 list_for_each_entry(lte, stream_list, write_streams_list) {
1410 if (is_file_stream(lte)) {
1411 lte->file_inode->num_remaining_streams = 0;
1412 lte->may_send_done_with_file = 1;
1414 lte->may_send_done_with_file = 0;
1418 list_for_each_entry(lte, stream_list, write_streams_list)
1419 if (lte->may_send_done_with_file)
1420 lte->file_inode->num_remaining_streams++;
1424 * Write a list of streams to the output WIM file.
1427 * The list of streams to write, specified by a list of `struct
1428 * wim_lookup_table_entry's linked by the 'write_streams_list' member.
1431 * The file descriptor, opened for writing, to which to write the streams.
1433 * @write_resource_flags
1434 * Flags to modify how the streams are written:
1436 * WRITE_RESOURCE_FLAG_RECOMPRESS:
1437 * Force compression of all resources, even if they could otherwise
1438 * be re-used by copying the raw data, due to being located in a WIM
1439 * file with compatible compression parameters.
1441 * WRITE_RESOURCE_FLAG_PIPABLE:
1442 * Write the resources in the wimlib-specific pipable format, and
1443 * furthermore do so in such a way that no seeking backwards in
1444 * @out_fd will be performed (so it may be a pipe).
1446 * WRITE_RESOURCE_FLAG_SOLID:
1447 * Combine all the streams into a single resource rather than
1448 * writing them in separate resources. This flag is only valid if
1449 * the WIM version number has been, or will be, set to
1450 * WIM_VERSION_SOLID. This flag may not be combined with
1451 * WRITE_RESOURCE_FLAG_PIPABLE.
1454 * Compression format to use to write the output streams, specified as one
1455 * of the WIMLIB_COMPRESSION_TYPE_* constants.
1456 * WIMLIB_COMPRESSION_TYPE_NONE is allowed.
1459 * Chunk size to use to write the streams. It must be a valid chunk size
1460 * for the specified compression format @out_ctype, unless @out_ctype is
1461 * WIMLIB_COMPRESSION_TYPE_NONE, in which case this parameter is ignored.
1464 * Number of threads to use to compress data. If 0, a default number of
1465 * threads will be chosen. The number of threads still may be decreased
1466 * from the specified value if insufficient memory is detected.
1469 * If on-the-fly deduplication of unhashed streams is desired, this
1470 * parameter must be pointer to the lookup table for the WIMStruct on whose
1471 * behalf the streams are being written. Otherwise, this parameter can be
1475 * If on-the-fly deduplication of unhashed streams is desired, this
1476 * parameter can be a pointer to a context for stream filtering used to
1477 * detect whether the duplicate stream has been hard-filtered or not. If
1478 * no streams are hard-filtered or no streams are unhashed, this parameter
1481 * This function will write the streams in @stream_list to resources in
1482 * consecutive positions in the output WIM file, or to a single solid resource
1483 * if WRITE_RESOURCE_FLAG_SOLID was specified in @write_resource_flags. In both
1484 * cases, the @out_reshdr of the `struct wim_lookup_table_entry' for each stream
1485 * written will be updated to specify its location, size, and flags in the
1486 * output WIM. In the solid resource case, WIM_RESHDR_FLAG_SOLID will be set in
1487 * the @flags field of each @out_reshdr, and furthermore @out_res_offset_in_wim
1488 * and @out_res_size_in_wim of each @out_reshdr will be set to the offset and
1489 * size, respectively, in the output WIM of the solid resource containing the
1490 * corresponding stream.
1492 * Each of the streams to write may be in any location supported by the
1493 * resource-handling code (specifically, read_stream_list()), such as the
1494 * contents of external file that has been logically added to the output WIM, or
1495 * a stream in another WIM file that has been imported, or even a stream in the
1496 * "same" WIM file of which a modified copy is being written. In the case that
1497 * a stream is already in a WIM file and uses compatible compression parameters,
1498 * by default this function will re-use the raw data instead of decompressing
1499 * it, then recompressing it; however, with WRITE_RESOURCE_FLAG_RECOMPRESS
1500 * specified in @write_resource_flags, this is not done.
1502 * As a further requirement, this function requires that the
1503 * @will_be_in_output_wim member be set to 1 on all streams in @stream_list as
1504 * well as any other streams not in @stream_list that will be in the output WIM
1505 * file, but set to 0 on any other streams in the output WIM's lookup table or
1506 * sharing a solid resource with a stream in @stream_list. Still furthermore,
1507 * if on-the-fly deduplication of streams is possible, then all streams in
1508 * @stream_list must also be linked by @lookup_table_list along with any other
1509 * streams that have @will_be_in_output_wim set.
1511 * This function handles on-the-fly deduplication of streams for which SHA1
1512 * message digests have not yet been calculated. Such streams may or may not
1513 * need to be written. If @lookup_table is non-NULL, then each stream in
1514 * @stream_list that has @unhashed set but not @unique_size set is checksummed
1515 * immediately before it would otherwise be read for writing in order to
1516 * determine if it is identical to another stream already being written or one
1517 * that would be filtered out of the output WIM using stream_filtered() with the
1518 * context @filter_ctx. Each such duplicate stream will be removed from
1519 * @stream_list, its reference count transfered to the pre-existing duplicate
1520 * stream, its memory freed, and will not be written. Alternatively, if a
1521 * stream in @stream_list is a duplicate with any stream in @lookup_table that
1522 * has not been marked for writing or would not be hard-filtered, it is freed
1523 * and the pre-existing duplicate is written instead, taking ownership of the
1524 * reference count and slot in the @lookup_table_list.
1526 * Returns 0 if every stream was either written successfully or did not need to
1527 * be written; otherwise returns a non-zero error code.
1530 write_stream_list(struct list_head *stream_list,
1531 struct filedes *out_fd,
1532 int write_resource_flags,
1535 unsigned num_threads,
1536 struct wim_lookup_table *lookup_table,
1537 struct filter_context *filter_ctx,
1538 wimlib_progress_func_t progfunc,
1542 struct write_streams_ctx ctx;
1543 struct list_head raw_copy_streams;
1545 wimlib_assert((write_resource_flags &
1546 (WRITE_RESOURCE_FLAG_SOLID |
1547 WRITE_RESOURCE_FLAG_PIPABLE)) !=
1548 (WRITE_RESOURCE_FLAG_SOLID |
1549 WRITE_RESOURCE_FLAG_PIPABLE));
1551 remove_zero_length_streams(stream_list);
1553 if (list_empty(stream_list)) {
1554 DEBUG("No streams to write.");
1558 /* If needed, set auxiliary information so that we can detect when the
1559 * library has finished using each external file. */
1560 if (unlikely(write_resource_flags & WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE))
1561 init_done_with_file_info(stream_list);
1563 memset(&ctx, 0, sizeof(ctx));
1565 ctx.out_fd = out_fd;
1566 ctx.lookup_table = lookup_table;
1567 ctx.out_ctype = out_ctype;
1568 ctx.out_chunk_size = out_chunk_size;
1569 ctx.write_resource_flags = write_resource_flags;
1570 ctx.filter_ctx = filter_ctx;
1573 * We normally sort the streams to write by a "sequential" order that is
1574 * optimized for reading. But when using solid compression, we instead
1575 * sort the streams by file extension and file name (when applicable;
1576 * and we don't do this for streams from solid resources) so that
1577 * similar files are grouped together, which improves the compression
1578 * ratio. This is somewhat of a hack since a stream does not
1579 * necessarily correspond one-to-one with a filename, nor is there any
1580 * guarantee that two files with similar names or extensions are
1581 * actually similar in content. A potential TODO is to sort the streams
1582 * based on some measure of similarity of their actual contents.
1585 ret = sort_stream_list_by_sequential_order(stream_list,
1586 offsetof(struct wim_lookup_table_entry,
1587 write_streams_list));
1591 compute_stream_list_stats(stream_list, &ctx);
1593 if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
1594 ret = sort_stream_list_for_solid_compression(stream_list);
1596 WARNING("Failed to sort streams for solid compression. Continuing anyways.");
1599 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) {
1600 wimlib_assert(out_chunk_size != 0);
1601 if (out_chunk_size <= STACK_MAX) {
1602 ctx.chunk_buf = alloca(out_chunk_size);
1604 ctx.chunk_buf = MALLOC(out_chunk_size);
1605 if (ctx.chunk_buf == NULL) {
1606 ret = WIMLIB_ERR_NOMEM;
1607 goto out_destroy_context;
1611 ctx.chunk_buf_filled = 0;
1613 ctx.progress_data.progfunc = progfunc;
1614 ctx.progress_data.progctx = progctx;
1616 ctx.num_bytes_to_compress = find_raw_copy_streams(stream_list,
1617 write_resource_flags,
1622 DEBUG("Writing stream list "
1623 "(offset = %"PRIu64", write_resource_flags=0x%08x, "
1624 "out_ctype=%d, out_chunk_size=%u, num_threads=%u, "
1625 "total_bytes=%"PRIu64", num_bytes_to_compress=%"PRIu64")",
1626 out_fd->offset, write_resource_flags,
1627 out_ctype, out_chunk_size, num_threads,
1628 ctx.progress_data.progress.write_streams.total_bytes,
1629 ctx.num_bytes_to_compress);
1631 if (ctx.num_bytes_to_compress == 0) {
1632 DEBUG("No compression needed; skipping to raw copy!");
1633 goto out_write_raw_copy_resources;
1636 /* Unless uncompressed output was required, allocate a chunk_compressor
1637 * to do compression. There are serial and parallel implementations of
1638 * the chunk_compressor interface. We default to parallel using the
1639 * specified number of threads, unless the upper bound on the number
1640 * bytes needing to be compressed is less than a heuristic value. */
1641 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) {
1643 #ifdef ENABLE_MULTITHREADED_COMPRESSION
1644 if (ctx.num_bytes_to_compress > max(2000000, out_chunk_size)) {
1645 ret = new_parallel_chunk_compressor(out_ctype,
1650 WARNING("Couldn't create parallel chunk compressor: %"TS".\n"
1651 " Falling back to single-threaded compression.",
1652 wimlib_get_error_string(ret));
1657 if (ctx.compressor == NULL) {
1658 ret = new_serial_chunk_compressor(out_ctype, out_chunk_size,
1661 goto out_destroy_context;
1666 ctx.progress_data.progress.write_streams.num_threads = ctx.compressor->num_threads;
1668 ctx.progress_data.progress.write_streams.num_threads = 1;
1670 DEBUG("Actually using %u threads",
1671 ctx.progress_data.progress.write_streams.num_threads);
1673 INIT_LIST_HEAD(&ctx.pending_streams);
1674 INIT_LIST_HEAD(&ctx.solid_streams);
1676 ret = call_progress(ctx.progress_data.progfunc,
1677 WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
1678 &ctx.progress_data.progress,
1679 ctx.progress_data.progctx);
1681 goto out_destroy_context;
1683 if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
1684 ret = begin_write_resource(&ctx, ctx.num_bytes_to_compress);
1686 goto out_destroy_context;
1689 /* Read the list of streams needing to be compressed, using the
1690 * specified callbacks to execute processing of the data. */
1692 struct read_stream_list_callbacks cbs = {
1693 .begin_stream = write_stream_begin_read,
1694 .begin_stream_ctx = &ctx,
1695 .consume_chunk = write_stream_process_chunk,
1696 .consume_chunk_ctx = &ctx,
1697 .end_stream = write_stream_end_read,
1698 .end_stream_ctx = &ctx,
1701 ret = read_stream_list(stream_list,
1702 offsetof(struct wim_lookup_table_entry, write_streams_list),
1704 STREAM_LIST_ALREADY_SORTED |
1705 VERIFY_STREAM_HASHES |
1706 COMPUTE_MISSING_STREAM_HASHES);
1709 goto out_destroy_context;
1711 ret = finish_remaining_chunks(&ctx);
1713 goto out_destroy_context;
1715 if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
1716 struct wim_reshdr reshdr;
1717 struct wim_lookup_table_entry *lte;
1720 ret = end_write_resource(&ctx, &reshdr);
1722 goto out_destroy_context;
1724 DEBUG("Ending solid resource: %lu %lu %lu.",
1725 reshdr.offset_in_wim,
1727 reshdr.uncompressed_size);
1730 list_for_each_entry(lte, &ctx.solid_streams, write_streams_list) {
1731 lte->out_reshdr.size_in_wim = lte->size;
1732 lte->out_reshdr.flags = filter_resource_flags(lte->flags);
1733 lte->out_reshdr.flags |= WIM_RESHDR_FLAG_SOLID;
1734 lte->out_reshdr.uncompressed_size = 0;
1735 lte->out_reshdr.offset_in_wim = offset_in_res;
1736 lte->out_res_offset_in_wim = reshdr.offset_in_wim;
1737 lte->out_res_size_in_wim = reshdr.size_in_wim;
1738 lte->out_res_uncompressed_size = reshdr.uncompressed_size;
1739 offset_in_res += lte->size;
1741 wimlib_assert(offset_in_res == reshdr.uncompressed_size);
1744 out_write_raw_copy_resources:
1745 /* Copy any compressed resources for which the raw data can be reused
1746 * without decompression. */
1747 ret = write_raw_copy_resources(&raw_copy_streams, ctx.out_fd,
1748 &ctx.progress_data);
1750 out_destroy_context:
1751 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && out_chunk_size > STACK_MAX)
1752 FREE(ctx.chunk_buf);
1753 FREE(ctx.chunk_csizes);
1755 ctx.compressor->destroy(ctx.compressor);
1756 DEBUG("Done (ret=%d)", ret);
1761 is_stream_in_solid_resource(struct wim_lookup_table_entry *lte, void *_ignore)
1763 return lte_is_partial(lte);
1767 wim_has_solid_resources(WIMStruct *wim)
1769 return for_lookup_table_entry(wim->lookup_table,
1770 is_stream_in_solid_resource, NULL);
1774 wim_write_stream_list(WIMStruct *wim,
1775 struct list_head *stream_list,
1777 unsigned num_threads,
1778 struct filter_context *filter_ctx)
1782 int write_resource_flags;
1784 write_resource_flags = write_flags_to_resource_flags(write_flags);
1786 /* wimlib v1.7.0: create a solid WIM file by default if the WIM version
1787 * has been set to WIM_VERSION_SOLID and at least one stream in the
1788 * WIM's lookup table is located in a solid resource (may be the same
1789 * WIM, or a different one in the case of export). */
1790 if (wim->hdr.wim_version == WIM_VERSION_SOLID &&
1791 wim_has_solid_resources(wim))
1793 write_resource_flags |= WRITE_RESOURCE_FLAG_SOLID;
1796 if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
1797 out_chunk_size = wim->out_solid_chunk_size;
1798 out_ctype = wim->out_solid_compression_type;
1800 out_chunk_size = wim->out_chunk_size;
1801 out_ctype = wim->out_compression_type;
1804 return write_stream_list(stream_list,
1806 write_resource_flags,
1817 write_wim_resource(struct wim_lookup_table_entry *lte,
1818 struct filedes *out_fd,
1821 int write_resource_flags)
1823 LIST_HEAD(stream_list);
1824 list_add(<e->write_streams_list, &stream_list);
1825 lte->will_be_in_output_wim = 1;
1826 return write_stream_list(&stream_list,
1828 write_resource_flags & ~WRITE_RESOURCE_FLAG_SOLID,
1839 write_wim_resource_from_buffer(const void *buf, size_t buf_size,
1840 int reshdr_flags, struct filedes *out_fd,
1843 struct wim_reshdr *out_reshdr,
1845 int write_resource_flags)
1848 struct wim_lookup_table_entry *lte;
1850 /* Set up a temporary lookup table entry to provide to
1851 * write_wim_resource(). */
1853 lte = new_lookup_table_entry();
1855 return WIMLIB_ERR_NOMEM;
1857 lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER;
1858 lte->attached_buffer = (void*)buf;
1859 lte->size = buf_size;
1860 lte->flags = reshdr_flags;
1862 if (write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
1863 sha1_buffer(buf, buf_size, lte->hash);
1869 ret = write_wim_resource(lte, out_fd, out_ctype, out_chunk_size,
1870 write_resource_flags);
1874 copy_reshdr(out_reshdr, <e->out_reshdr);
1877 copy_hash(hash, lte->hash);
1880 lte->resource_location = RESOURCE_NONEXISTENT;
1881 free_lookup_table_entry(lte);
1885 struct stream_size_table {
1886 struct hlist_head *array;
1892 init_stream_size_table(struct stream_size_table *tab, size_t capacity)
1894 tab->array = CALLOC(capacity, sizeof(tab->array[0]));
1895 if (tab->array == NULL)
1896 return WIMLIB_ERR_NOMEM;
1897 tab->num_entries = 0;
1898 tab->capacity = capacity;
1903 destroy_stream_size_table(struct stream_size_table *tab)
1909 stream_size_table_insert(struct wim_lookup_table_entry *lte, void *_tab)
1911 struct stream_size_table *tab = _tab;
1913 struct wim_lookup_table_entry *same_size_lte;
1914 struct hlist_node *tmp;
1916 pos = hash_u64(lte->size) % tab->capacity;
1917 lte->unique_size = 1;
1918 hlist_for_each_entry(same_size_lte, tmp, &tab->array[pos], hash_list_2) {
1919 if (same_size_lte->size == lte->size) {
1920 lte->unique_size = 0;
1921 same_size_lte->unique_size = 0;
1926 hlist_add_head(<e->hash_list_2, &tab->array[pos]);
1931 struct find_streams_ctx {
1934 struct list_head stream_list;
1935 struct stream_size_table stream_size_tab;
1939 reference_stream_for_write(struct wim_lookup_table_entry *lte,
1940 struct list_head *stream_list, u32 nref)
1942 if (!lte->will_be_in_output_wim) {
1943 lte->out_refcnt = 0;
1944 list_add_tail(<e->write_streams_list, stream_list);
1945 lte->will_be_in_output_wim = 1;
1947 lte->out_refcnt += nref;
1951 fully_reference_stream_for_write(struct wim_lookup_table_entry *lte,
1954 struct list_head *stream_list = _stream_list;
1955 lte->will_be_in_output_wim = 0;
1956 reference_stream_for_write(lte, stream_list, lte->refcnt);
1961 inode_find_streams_to_reference(const struct wim_inode *inode,
1962 const struct wim_lookup_table *table,
1963 struct list_head *stream_list)
1965 struct wim_lookup_table_entry *lte;
1968 wimlib_assert(inode->i_nlink > 0);
1970 for (i = 0; i <= inode->i_num_ads; i++) {
1971 lte = inode_stream_lte(inode, i, table);
1973 reference_stream_for_write(lte, stream_list,
1975 else if (!is_zero_hash(inode_stream_hash(inode, i)))
1976 return WIMLIB_ERR_RESOURCE_NOT_FOUND;
1982 do_stream_set_not_in_output_wim(struct wim_lookup_table_entry *lte, void *_ignore)
1984 lte->will_be_in_output_wim = 0;
1989 image_find_streams_to_reference(WIMStruct *wim)
1991 struct wim_image_metadata *imd;
1992 struct wim_inode *inode;
1993 struct wim_lookup_table_entry *lte;
1994 struct list_head *stream_list;
1997 imd = wim_get_current_image_metadata(wim);
1999 image_for_each_unhashed_stream(lte, imd)
2000 lte->will_be_in_output_wim = 0;
2002 stream_list = wim->private;
2003 image_for_each_inode(inode, imd) {
2004 ret = inode_find_streams_to_reference(inode,
2014 prepare_unfiltered_list_of_streams_in_output_wim(WIMStruct *wim,
2017 struct list_head *stream_list_ret)
2021 INIT_LIST_HEAD(stream_list_ret);
2023 if (streams_ok && (image == WIMLIB_ALL_IMAGES ||
2024 (image == 1 && wim->hdr.image_count == 1)))
2026 /* Fast case: Assume that all streams are being written and
2027 * that the reference counts are correct. */
2028 struct wim_lookup_table_entry *lte;
2029 struct wim_image_metadata *imd;
2032 for_lookup_table_entry(wim->lookup_table,
2033 fully_reference_stream_for_write,
2036 for (i = 0; i < wim->hdr.image_count; i++) {
2037 imd = wim->image_metadata[i];
2038 image_for_each_unhashed_stream(lte, imd)
2039 fully_reference_stream_for_write(lte, stream_list_ret);
2042 /* Slow case: Walk through the images being written and
2043 * determine the streams referenced. */
2044 for_lookup_table_entry(wim->lookup_table,
2045 do_stream_set_not_in_output_wim, NULL);
2046 wim->private = stream_list_ret;
2047 ret = for_image(wim, image, image_find_streams_to_reference);
2055 struct insert_other_if_hard_filtered_ctx {
2056 struct stream_size_table *tab;
2057 struct filter_context *filter_ctx;
2061 insert_other_if_hard_filtered(struct wim_lookup_table_entry *lte, void *_ctx)
2063 struct insert_other_if_hard_filtered_ctx *ctx = _ctx;
2065 if (!lte->will_be_in_output_wim &&
2066 stream_hard_filtered(lte, ctx->filter_ctx))
2067 stream_size_table_insert(lte, ctx->tab);
2072 determine_stream_size_uniquity(struct list_head *stream_list,
2073 struct wim_lookup_table *lt,
2074 struct filter_context *filter_ctx)
2077 struct stream_size_table tab;
2078 struct wim_lookup_table_entry *lte;
2080 ret = init_stream_size_table(&tab, 9001);
2084 if (may_hard_filter_streams(filter_ctx)) {
2085 struct insert_other_if_hard_filtered_ctx ctx = {
2087 .filter_ctx = filter_ctx,
2089 for_lookup_table_entry(lt, insert_other_if_hard_filtered, &ctx);
2092 list_for_each_entry(lte, stream_list, write_streams_list)
2093 stream_size_table_insert(lte, &tab);
2095 destroy_stream_size_table(&tab);
2100 filter_stream_list_for_write(struct list_head *stream_list,
2101 struct filter_context *filter_ctx)
2103 struct wim_lookup_table_entry *lte, *tmp;
2105 list_for_each_entry_safe(lte, tmp,
2106 stream_list, write_streams_list)
2108 int status = stream_filtered(lte, filter_ctx);
2115 /* Soft filtered. */
2117 /* Hard filtered. */
2118 lte->will_be_in_output_wim = 0;
2119 list_del(<e->lookup_table_list);
2121 list_del(<e->write_streams_list);
2127 * prepare_stream_list_for_write() -
2129 * Prepare the list of streams to write for writing a WIM containing the
2130 * specified image(s) with the specified write flags.
2133 * The WIMStruct on whose behalf the write is occurring.
2136 * Image(s) from the WIM to write; may be WIMLIB_ALL_IMAGES.
2139 * WIMLIB_WRITE_FLAG_* flags for the write operation:
2141 * STREAMS_OK: For writes of all images, assume that all streams in the
2142 * lookup table of @wim and the per-image lists of unhashed streams should
2143 * be taken as-is, and image metadata should not be searched for
2144 * references. This does not exclude filtering with OVERWRITE and
2145 * SKIP_EXTERNAL_WIMS, below.
2147 * OVERWRITE: Streams already present in @wim shall not be returned in
2150 * SKIP_EXTERNAL_WIMS: Streams already present in a WIM file, but not
2151 * @wim, shall be returned in neither @stream_list_ret nor
2152 * @lookup_table_list_ret.
2155 * List of streams, linked by write_streams_list, that need to be written
2156 * will be returned here.
2158 * Note that this function assumes that unhashed streams will be written;
2159 * it does not take into account that they may become duplicates when
2162 * @lookup_table_list_ret
2163 * List of streams, linked by lookup_table_list, that need to be included
2164 * in the WIM's lookup table will be returned here. This will be a
2165 * superset of the streams in @stream_list_ret.
2167 * This list will be a proper superset of @stream_list_ret if and only if
2168 * WIMLIB_WRITE_FLAG_OVERWRITE was specified in @write_flags and some of
2169 * the streams that would otherwise need to be written were already located
2172 * All streams in this list will have @out_refcnt set to the number of
2173 * references to the stream in the output WIM. If
2174 * WIMLIB_WRITE_FLAG_STREAMS_OK was specified in @write_flags, @out_refcnt
2175 * may be as low as 0.
2178 * A context for queries of stream filter status with stream_filtered() is
2179 * returned in this location.
2181 * In addition, @will_be_in_output_wim will be set to 1 in all stream entries
2182 * inserted into @lookup_table_list_ret and to 0 in all stream entries in the
2183 * lookup table of @wim not inserted into @lookup_table_list_ret.
2185 * Still furthermore, @unique_size will be set to 1 on all stream entries in
2186 * @stream_list_ret that have unique size among all stream entries in
2187 * @stream_list_ret and among all stream entries in the lookup table of @wim
2188 * that are ineligible for being written due to filtering.
2190 * Returns 0 on success; nonzero on read error, memory allocation error, or
2194 prepare_stream_list_for_write(WIMStruct *wim, int image,
2196 struct list_head *stream_list_ret,
2197 struct list_head *lookup_table_list_ret,
2198 struct filter_context *filter_ctx_ret)
2201 struct wim_lookup_table_entry *lte;
2203 filter_ctx_ret->write_flags = write_flags;
2204 filter_ctx_ret->wim = wim;
2206 ret = prepare_unfiltered_list_of_streams_in_output_wim(
2209 write_flags & WIMLIB_WRITE_FLAG_STREAMS_OK,
2214 INIT_LIST_HEAD(lookup_table_list_ret);
2215 list_for_each_entry(lte, stream_list_ret, write_streams_list)
2216 list_add_tail(<e->lookup_table_list, lookup_table_list_ret);
2218 ret = determine_stream_size_uniquity(stream_list_ret, wim->lookup_table,
2223 if (may_filter_streams(filter_ctx_ret))
2224 filter_stream_list_for_write(stream_list_ret, filter_ctx_ret);
2230 write_wim_streams(WIMStruct *wim, int image, int write_flags,
2231 unsigned num_threads,
2232 struct list_head *stream_list_override,
2233 struct list_head *lookup_table_list_ret)
2236 struct list_head _stream_list;
2237 struct list_head *stream_list;
2238 struct wim_lookup_table_entry *lte;
2239 struct filter_context _filter_ctx;
2240 struct filter_context *filter_ctx;
2242 if (stream_list_override == NULL) {
2243 /* Normal case: prepare stream list from image(s) being written.
2245 stream_list = &_stream_list;
2246 filter_ctx = &_filter_ctx;
2247 ret = prepare_stream_list_for_write(wim, image, write_flags,
2249 lookup_table_list_ret,
2254 /* Currently only as a result of wimlib_split() being called:
2255 * use stream list already explicitly provided. Use existing
2256 * reference counts. */
2257 stream_list = stream_list_override;
2259 INIT_LIST_HEAD(lookup_table_list_ret);
2260 list_for_each_entry(lte, stream_list, write_streams_list) {
2261 lte->out_refcnt = lte->refcnt;
2262 lte->will_be_in_output_wim = 1;
2263 lte->unique_size = 0;
2264 list_add_tail(<e->lookup_table_list, lookup_table_list_ret);
2268 return wim_write_stream_list(wim,
2276 write_wim_metadata_resources(WIMStruct *wim, int image, int write_flags)
2281 int write_resource_flags;
2283 if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA) {
2284 DEBUG("Not writing any metadata resources.");
2288 write_resource_flags = write_flags_to_resource_flags(write_flags);
2290 write_resource_flags &= ~WRITE_RESOURCE_FLAG_SOLID;
2292 DEBUG("Writing metadata resources (offset=%"PRIu64")",
2293 wim->out_fd.offset);
2295 ret = call_progress(wim->progfunc,
2296 WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN,
2297 NULL, wim->progctx);
2301 if (image == WIMLIB_ALL_IMAGES) {
2303 end_image = wim->hdr.image_count;
2305 start_image = image;
2309 for (int i = start_image; i <= end_image; i++) {
2310 struct wim_image_metadata *imd;
2312 imd = wim->image_metadata[i - 1];
2313 /* Build a new metadata resource only if image was modified from
2314 * the original (or was newly added). Otherwise just copy the
2316 if (imd->modified) {
2317 DEBUG("Image %u was modified; building and writing new "
2318 "metadata resource", i);
2319 ret = write_metadata_resource(wim, i,
2320 write_resource_flags);
2321 } else if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) {
2322 DEBUG("Image %u was not modified; re-using existing "
2323 "metadata resource.", i);
2324 stream_set_out_reshdr_for_reuse(imd->metadata_lte);
2327 DEBUG("Image %u was not modified; copying existing "
2328 "metadata resource.", i);
2329 ret = write_wim_resource(imd->metadata_lte,
2331 wim->out_compression_type,
2332 wim->out_chunk_size,
2333 write_resource_flags);
2339 return call_progress(wim->progfunc,
2340 WIMLIB_PROGRESS_MSG_WRITE_METADATA_END,
2341 NULL, wim->progctx);
2345 open_wim_writable(WIMStruct *wim, const tchar *path, int open_flags)
2348 DEBUG("Opening \"%"TS"\" for writing.", path);
2350 raw_fd = topen(path, open_flags | O_BINARY, 0644);
2352 ERROR_WITH_ERRNO("Failed to open \"%"TS"\" for writing", path);
2353 return WIMLIB_ERR_OPEN;
2355 filedes_init(&wim->out_fd, raw_fd);
2360 close_wim_writable(WIMStruct *wim, int write_flags)
2364 if (!(write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)) {
2365 DEBUG("Closing WIM file.");
2366 if (filedes_valid(&wim->out_fd))
2367 if (filedes_close(&wim->out_fd))
2368 ret = WIMLIB_ERR_WRITE;
2370 filedes_invalidate(&wim->out_fd);
2375 cmp_streams_by_out_rspec(const void *p1, const void *p2)
2377 const struct wim_lookup_table_entry *lte1, *lte2;
2379 lte1 = *(const struct wim_lookup_table_entry**)p1;
2380 lte2 = *(const struct wim_lookup_table_entry**)p2;
2382 if (lte1->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
2383 if (lte2->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
2384 if (lte1->out_res_offset_in_wim != lte2->out_res_offset_in_wim)
2385 return cmp_u64(lte1->out_res_offset_in_wim,
2386 lte2->out_res_offset_in_wim);
2391 if (lte2->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID)
2394 return cmp_u64(lte1->out_reshdr.offset_in_wim,
2395 lte2->out_reshdr.offset_in_wim);
2399 write_wim_lookup_table(WIMStruct *wim, int image, int write_flags,
2400 struct wim_reshdr *out_reshdr,
2401 struct list_head *lookup_table_list)
2405 /* Set output resource metadata for streams already present in WIM. */
2406 if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) {
2407 struct wim_lookup_table_entry *lte;
2408 list_for_each_entry(lte, lookup_table_list, lookup_table_list)
2410 if (lte->resource_location == RESOURCE_IN_WIM &&
2411 lte->rspec->wim == wim)
2413 stream_set_out_reshdr_for_reuse(lte);
2418 ret = sort_stream_list(lookup_table_list,
2419 offsetof(struct wim_lookup_table_entry, lookup_table_list),
2420 cmp_streams_by_out_rspec);
2424 /* Add entries for metadata resources. */
2425 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)) {
2429 if (image == WIMLIB_ALL_IMAGES) {
2431 end_image = wim->hdr.image_count;
2433 start_image = image;
2437 /* Push metadata resource lookup table entries onto the front of
2438 * the list in reverse order, so that they're written in order.
2440 for (int i = end_image; i >= start_image; i--) {
2441 struct wim_lookup_table_entry *metadata_lte;
2443 metadata_lte = wim->image_metadata[i - 1]->metadata_lte;
2444 wimlib_assert(metadata_lte->out_reshdr.flags & WIM_RESHDR_FLAG_METADATA);
2445 metadata_lte->out_refcnt = 1;
2446 list_add(&metadata_lte->lookup_table_list, lookup_table_list);
2450 return write_wim_lookup_table_from_stream_list(lookup_table_list,
2452 wim->hdr.part_number,
2454 write_flags_to_resource_flags(write_flags));
2460 * Finish writing a WIM file: write the lookup table, xml data, and integrity
2461 * table, then overwrite the WIM header. By default, closes the WIM file
2462 * descriptor (@wim->out_fd) if successful.
2464 * write_flags is a bitwise OR of the following:
2466 * (public) WIMLIB_WRITE_FLAG_CHECK_INTEGRITY:
2467 * Include an integrity table.
2469 * (public) WIMLIB_WRITE_FLAG_FSYNC:
2470 * fsync() the output file before closing it.
2472 * (public) WIMLIB_WRITE_FLAG_PIPABLE:
2473 * Writing a pipable WIM, possibly to a pipe; include pipable WIM
2474 * stream headers before the lookup table and XML data, and also
2475 * write the WIM header at the end instead of seeking to the
2476 * beginning. Can't be combined with
2477 * WIMLIB_WRITE_FLAG_CHECK_INTEGRITY.
2479 * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE:
2480 * Don't write the lookup table.
2482 * (private) WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML:
2483 * After writing the XML data but before writing the integrity
2484 * table, write a temporary WIM header and flush the stream so that
2485 * the WIM is less likely to become corrupted upon abrupt program
2487 * (private) WIMLIB_WRITE_FLAG_HEADER_AT_END:
2488 * Instead of overwriting the WIM header at the beginning of the
2489 * file, simply append it to the end of the file. (Used when
2491 * (private) WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR:
2492 * Do not close the file descriptor @wim->out_fd on either success
2494 * (private) WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES:
2495 * Use the existing <TOTALBYTES> stored in the in-memory XML
2496 * information, rather than setting it to the offset of the XML
2497 * data being written.
2498 * (private) WIMLIB_WRITE_FLAG_OVERWRITE
2499 * The existing WIM file is being updated in-place. The entries
2500 * from its integrity table may be re-used.
2503 finish_write(WIMStruct *wim, int image, int write_flags,
2504 struct list_head *lookup_table_list)
2508 int write_resource_flags;
2509 off_t old_lookup_table_end = 0;
2510 off_t new_lookup_table_end;
2512 struct integrity_table *old_integrity_table = NULL;
2514 DEBUG("image=%d, write_flags=%08x", image, write_flags);
2516 write_resource_flags = write_flags_to_resource_flags(write_flags);
2518 /* In the WIM header, there is room for the resource entry for a
2519 * metadata resource labeled as the "boot metadata". This entry should
2520 * be zeroed out if there is no bootable image (boot_idx 0). Otherwise,
2521 * it should be a copy of the resource entry for the image that is
2522 * marked as bootable. This is not well documented... */
2523 if (wim->hdr.boot_idx == 0) {
2524 zero_reshdr(&wim->hdr.boot_metadata_reshdr);
2526 copy_reshdr(&wim->hdr.boot_metadata_reshdr,
2527 &wim->image_metadata[
2528 wim->hdr.boot_idx - 1]->metadata_lte->out_reshdr);
2531 /* If overwriting the WIM file containing an integrity table in-place,
2532 * we'd like to re-use the information in the old integrity table
2533 * instead of recalculating it. But we might overwrite the old
2534 * integrity table when we expand the XML data. Read it into memory
2536 if ((write_flags & (WIMLIB_WRITE_FLAG_OVERWRITE |
2537 WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)) ==
2538 (WIMLIB_WRITE_FLAG_OVERWRITE |
2539 WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)
2540 && wim_has_integrity_table(wim))
2542 old_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim +
2543 wim->hdr.lookup_table_reshdr.size_in_wim;
2544 (void)read_integrity_table(wim,
2545 old_lookup_table_end - WIM_HEADER_DISK_SIZE,
2546 &old_integrity_table);
2547 /* If we couldn't read the old integrity table, we can still
2548 * re-calculate the full integrity table ourselves. Hence the
2549 * ignoring of the return value. */
2552 /* Write lookup table. */
2553 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
2554 ret = write_wim_lookup_table(wim, image, write_flags,
2555 &wim->hdr.lookup_table_reshdr,
2558 free_integrity_table(old_integrity_table);
2563 /* Write XML data. */
2564 xml_totalbytes = wim->out_fd.offset;
2565 if (write_flags & WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES)
2566 xml_totalbytes = WIM_TOTALBYTES_USE_EXISTING;
2567 ret = write_wim_xml_data(wim, image, xml_totalbytes,
2568 &wim->hdr.xml_data_reshdr,
2569 write_resource_flags);
2571 free_integrity_table(old_integrity_table);
2575 /* Write integrity table (optional). */
2576 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
2577 if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) {
2578 struct wim_header checkpoint_hdr;
2579 memcpy(&checkpoint_hdr, &wim->hdr, sizeof(struct wim_header));
2580 zero_reshdr(&checkpoint_hdr.integrity_table_reshdr);
2581 checkpoint_hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2582 ret = write_wim_header_at_offset(&checkpoint_hdr,
2585 free_integrity_table(old_integrity_table);
2590 new_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim +
2591 wim->hdr.lookup_table_reshdr.size_in_wim;
2593 ret = write_integrity_table(wim,
2594 new_lookup_table_end,
2595 old_lookup_table_end,
2596 old_integrity_table);
2597 free_integrity_table(old_integrity_table);
2601 /* No integrity table. */
2602 zero_reshdr(&wim->hdr.integrity_table_reshdr);
2605 /* Now that all information in the WIM header has been determined, the
2606 * preliminary header written earlier can be overwritten, the header of
2607 * the existing WIM file can be overwritten, or the final header can be
2608 * written to the end of the pipable WIM. */
2609 wim->hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2611 if (write_flags & WIMLIB_WRITE_FLAG_HEADER_AT_END)
2612 hdr_offset = wim->out_fd.offset;
2613 DEBUG("Writing new header @ %"PRIu64".", hdr_offset);
2614 ret = write_wim_header_at_offset(&wim->hdr, &wim->out_fd, hdr_offset);
2618 /* Possibly sync file data to disk before closing. On POSIX systems, it
2619 * is necessary to do this before using rename() to overwrite an
2620 * existing file with a new file. Otherwise, data loss would occur if
2621 * the system is abruptly terminated when the metadata for the rename
2622 * operation has been written to disk, but the new file data has not.
2624 if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) {
2625 DEBUG("Syncing WIM file.");
2626 if (fsync(wim->out_fd.fd)) {
2627 ERROR_WITH_ERRNO("Error syncing data to WIM file");
2628 return WIMLIB_ERR_WRITE;
2632 if (close_wim_writable(wim, write_flags)) {
2633 ERROR_WITH_ERRNO("Failed to close the output WIM file");
2634 return WIMLIB_ERR_WRITE;
2640 #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK)
2642 /* Set advisory lock on WIM file (if not already done so) */
2644 lock_wim_for_append(WIMStruct *wim)
2646 if (wim->locked_for_append)
2648 if (!flock(wim->in_fd.fd, LOCK_EX | LOCK_NB)) {
2649 wim->locked_for_append = 1;
2652 if (errno != EWOULDBLOCK)
2654 return WIMLIB_ERR_ALREADY_LOCKED;
2657 /* Remove advisory lock on WIM file (if present) */
2659 unlock_wim_for_append(WIMStruct *wim)
2661 if (wim->locked_for_append) {
2662 flock(wim->in_fd.fd, LOCK_UN);
2663 wim->locked_for_append = 0;
2669 * write_pipable_wim():
2671 * Perform the intermediate stages of creating a "pipable" WIM (i.e. a WIM
2672 * capable of being applied from a pipe).
2674 * Pipable WIMs are a wimlib-specific modification of the WIM format such that
2675 * images can be applied from them sequentially when the file data is sent over
2676 * a pipe. In addition, a pipable WIM can be written sequentially to a pipe.
2677 * The modifications made to the WIM format for pipable WIMs are:
2679 * - Magic characters in header are "WLPWM\0\0\0" (wimlib pipable WIM) instead
2680 * of "MSWIM\0\0\0". This lets wimlib know that the WIM is pipable and also
2681 * stops other software from trying to read the file as a normal WIM.
2683 * - The header at the beginning of the file does not contain all the normal
2684 * information; in particular it will have all 0's for the lookup table and
2685 * XML data resource entries. This is because this information cannot be
2686 * determined until the lookup table and XML data have been written.
2687 * Consequently, wimlib will write the full header at the very end of the
2688 * file. The header at the end, however, is only used when reading the WIM
2689 * from a seekable file (not a pipe).
2691 * - An extra copy of the XML data is placed directly after the header. This
2692 * allows image names and sizes to be determined at an appropriate time when
2693 * reading the WIM from a pipe. This copy of the XML data is ignored if the
2694 * WIM is read from a seekable file (not a pipe).
2696 * - The format of resources, or streams, has been modified to allow them to be
2697 * used before the "lookup table" has been read. Each stream is prefixed with
2698 * a `struct pwm_stream_hdr' that is basically an abbreviated form of `struct
2699 * wim_lookup_table_entry_disk' that only contains the SHA1 message digest,
2700 * uncompressed stream size, and flags that indicate whether the stream is
2701 * compressed. The data of uncompressed streams then follows literally, while
2702 * the data of compressed streams follows in a modified format. Compressed
2703 * streams do not begin with a chunk table, since the chunk table cannot be
2704 * written until all chunks have been compressed. Instead, each compressed
2705 * chunk is prefixed by a `struct pwm_chunk_hdr' that gives its size.
2706 * Furthermore, the chunk table is written at the end of the resource instead
2707 * of the start. Note: chunk offsets are given in the chunk table as if the
2708 * `struct pwm_chunk_hdr's were not present; also, the chunk table is only
2709 * used if the WIM is being read from a seekable file (not a pipe).
2711 * - Metadata resources always come before other file resources (streams).
2712 * (This does not by itself constitute an incompatibility with normal WIMs,
2713 * since this is valid in normal WIMs.)
2715 * - At least up to the end of the file resources, all components must be packed
2716 * as tightly as possible; there cannot be any "holes" in the WIM. (This does
2717 * not by itself consititute an incompatibility with normal WIMs, since this
2718 * is valid in normal WIMs.)
2720 * Note: the lookup table, XML data, and header at the end are not used when
2721 * applying from a pipe. They exist to support functionality such as image
2722 * application and export when the WIM is *not* read from a pipe.
2724 * Layout of pipable WIM:
2726 * ---------+----------+--------------------+----------------+--------------+-----------+--------+
2727 * | Header | XML data | Metadata resources | File resources | Lookup table | XML data | Header |
2728 * ---------+----------+--------------------+----------------+--------------+-----------+--------+
2730 * Layout of normal WIM:
2732 * +--------+-----------------------------+-------------------------+
2733 * | Header | File and metadata resources | Lookup table | XML data |
2734 * +--------+-----------------------------+-------------------------+
2736 * An optional integrity table can follow the final XML data in both normal and
2737 * pipable WIMs. However, due to implementation details, wimlib currently can
2738 * only include an integrity table in a pipable WIM when writing it to a
2739 * seekable file (not a pipe).
2741 * Do note that since pipable WIMs are not supported by Microsoft's software,
2742 * wimlib does not create them unless explicitly requested (with
2743 * WIMLIB_WRITE_FLAG_PIPABLE) and as stated above they use different magic
2744 * characters to identify the file.
2747 write_pipable_wim(WIMStruct *wim, int image, int write_flags,
2748 unsigned num_threads,
2749 struct list_head *stream_list_override,
2750 struct list_head *lookup_table_list_ret)
2753 struct wim_reshdr xml_reshdr;
2755 WARNING("Creating a pipable WIM, which will "
2757 " with Microsoft's software (wimgapi/imagex/Dism).");
2759 /* At this point, the header at the beginning of the file has already
2762 /* For efficiency, when wimlib adds an image to the WIM with
2763 * wimlib_add_image(), the SHA1 message digests of files is not
2764 * calculated; instead, they are calculated while the files are being
2765 * written. However, this does not work when writing a pipable WIM,
2766 * since when writing a stream to a pipable WIM, its SHA1 message digest
2767 * needs to be known before the stream data is written. Therefore,
2768 * before getting much farther, we need to pre-calculate the SHA1
2769 * message digests of all streams that will be written. */
2770 ret = wim_checksum_unhashed_streams(wim);
2774 /* Write extra copy of the XML data. */
2775 ret = write_wim_xml_data(wim, image, WIM_TOTALBYTES_OMIT,
2777 WRITE_RESOURCE_FLAG_PIPABLE);
2781 /* Write metadata resources for the image(s) being included in the
2783 ret = write_wim_metadata_resources(wim, image, write_flags);
2787 /* Write streams needed for the image(s) being included in the output
2788 * WIM, or streams needed for the split WIM part. */
2789 return write_wim_streams(wim, image, write_flags, num_threads,
2790 stream_list_override, lookup_table_list_ret);
2792 /* The lookup table, XML data, and header at end are handled by
2793 * finish_write(). */
2796 /* Write a standalone WIM or split WIM (SWM) part to a new file or to a file
2799 write_wim_part(WIMStruct *wim,
2800 const void *path_or_fd,
2803 unsigned num_threads,
2804 unsigned part_number,
2805 unsigned total_parts,
2806 struct list_head *stream_list_override,
2810 struct wim_header hdr_save;
2811 struct list_head lookup_table_list;
2813 if (total_parts == 1)
2814 DEBUG("Writing standalone WIM.");
2816 DEBUG("Writing split WIM part %u/%u", part_number, total_parts);
2817 if (image == WIMLIB_ALL_IMAGES)
2818 DEBUG("Including all images.");
2820 DEBUG("Including image %d only.", image);
2821 if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)
2822 DEBUG("File descriptor: %d", *(const int*)path_or_fd);
2824 DEBUG("Path: \"%"TS"\"", (const tchar*)path_or_fd);
2825 DEBUG("Write flags: 0x%08x", write_flags);
2827 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)
2828 DEBUG("\tCHECK_INTEGRITY");
2830 if (write_flags & WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)
2831 DEBUG("\tNO_CHECK_INTEGRITY");
2833 if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
2836 if (write_flags & WIMLIB_WRITE_FLAG_NOT_PIPABLE)
2837 DEBUG("\tNOT_PIPABLE");
2839 if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
2840 DEBUG("\tRECOMPRESS");
2842 if (write_flags & WIMLIB_WRITE_FLAG_FSYNC)
2845 if (write_flags & WIMLIB_WRITE_FLAG_REBUILD)
2848 if (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE)
2849 DEBUG("\tSOFT_DELETE");
2851 if (write_flags & WIMLIB_WRITE_FLAG_IGNORE_READONLY_FLAG)
2852 DEBUG("\tIGNORE_READONLY_FLAG");
2854 if (write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS)
2855 DEBUG("\tSKIP_EXTERNAL_WIMS");
2857 if (write_flags & WIMLIB_WRITE_FLAG_STREAMS_OK)
2858 DEBUG("\tSTREAMS_OK");
2860 if (write_flags & WIMLIB_WRITE_FLAG_RETAIN_GUID)
2861 DEBUG("\tRETAIN_GUID");
2863 if (write_flags & WIMLIB_WRITE_FLAG_SOLID)
2866 if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)
2867 DEBUG("\tFILE_DESCRIPTOR");
2869 if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)
2870 DEBUG("\tNO_METADATA");
2872 if (write_flags & WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES)
2873 DEBUG("\tUSE_EXISTING_TOTALBYTES");
2875 if (num_threads == 0)
2876 DEBUG("Number of threads: autodetect");
2878 DEBUG("Number of threads: %u", num_threads);
2879 DEBUG("Progress function: %s", (wim->progfunc ? "yes" : "no"));
2880 DEBUG("Stream list: %s", (stream_list_override ? "specified" : "autodetect"));
2881 DEBUG("GUID: %s", (write_flags &
2882 WIMLIB_WRITE_FLAG_RETAIN_GUID) ? "retain"
2883 : guid ? "explicit" : "generate new");
2885 /* Internally, this is always called with a valid part number and total
2887 wimlib_assert(total_parts >= 1);
2888 wimlib_assert(part_number >= 1 && part_number <= total_parts);
2890 /* A valid image (or all images) must be specified. */
2891 if (image != WIMLIB_ALL_IMAGES &&
2892 (image < 1 || image > wim->hdr.image_count))
2893 return WIMLIB_ERR_INVALID_IMAGE;
2895 /* If we need to write metadata resources, make sure the ::WIMStruct has
2896 * the needed information attached (e.g. is not a resource-only WIM,
2897 * such as a non-first part of a split WIM). */
2898 if (!wim_has_metadata(wim) &&
2899 !(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA))
2900 return WIMLIB_ERR_METADATA_NOT_FOUND;
2902 /* Check for contradictory flags. */
2903 if ((write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2904 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))
2905 == (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2906 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))
2907 return WIMLIB_ERR_INVALID_PARAM;
2909 if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2910 WIMLIB_WRITE_FLAG_NOT_PIPABLE))
2911 == (WIMLIB_WRITE_FLAG_PIPABLE |
2912 WIMLIB_WRITE_FLAG_NOT_PIPABLE))
2913 return WIMLIB_ERR_INVALID_PARAM;
2915 /* Save previous header, then start initializing the new one. */
2916 memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header));
2918 /* Set default integrity, pipable, and solid flags. */
2919 if (!(write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2920 WIMLIB_WRITE_FLAG_NOT_PIPABLE)))
2921 if (wim_is_pipable(wim)) {
2922 DEBUG("WIM is pipable; default to PIPABLE.");
2923 write_flags |= WIMLIB_WRITE_FLAG_PIPABLE;
2926 if (!(write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2927 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)))
2928 if (wim_has_integrity_table(wim)) {
2929 DEBUG("Integrity table present; default to CHECK_INTEGRITY.");
2930 write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY;
2933 if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2934 WIMLIB_WRITE_FLAG_SOLID))
2935 == (WIMLIB_WRITE_FLAG_PIPABLE |
2936 WIMLIB_WRITE_FLAG_SOLID))
2938 ERROR("Cannot specify both PIPABLE and SOLID!");
2939 return WIMLIB_ERR_INVALID_PARAM;
2942 /* Set appropriate magic number. */
2943 if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
2944 wim->hdr.magic = PWM_MAGIC;
2946 wim->hdr.magic = WIM_MAGIC;
2948 /* Set appropriate version number. */
2949 if ((write_flags & WIMLIB_WRITE_FLAG_SOLID) ||
2950 wim->out_compression_type == WIMLIB_COMPRESSION_TYPE_LZMS)
2951 wim->hdr.wim_version = WIM_VERSION_SOLID;
2953 wim->hdr.wim_version = WIM_VERSION_DEFAULT;
2955 /* Clear header flags that will be set automatically. */
2956 wim->hdr.flags &= ~(WIM_HDR_FLAG_METADATA_ONLY |
2957 WIM_HDR_FLAG_RESOURCE_ONLY |
2958 WIM_HDR_FLAG_SPANNED |
2959 WIM_HDR_FLAG_WRITE_IN_PROGRESS);
2961 /* Set SPANNED header flag if writing part of a split WIM. */
2962 if (total_parts != 1)
2963 wim->hdr.flags |= WIM_HDR_FLAG_SPANNED;
2965 /* Set part number and total parts of split WIM. This will be 1 and 1
2966 * if the WIM is standalone. */
2967 wim->hdr.part_number = part_number;
2968 wim->hdr.total_parts = total_parts;
2970 /* Set compression type if different. */
2971 if (wim->compression_type != wim->out_compression_type) {
2972 ret = set_wim_hdr_cflags(wim->out_compression_type, &wim->hdr);
2973 wimlib_assert(ret == 0);
2976 /* Set chunk size if different. */
2977 wim->hdr.chunk_size = wim->out_chunk_size;
2980 if (!(write_flags & WIMLIB_WRITE_FLAG_RETAIN_GUID)) {
2982 memcpy(wim->hdr.guid, guid, WIMLIB_GUID_LEN);
2984 randomize_byte_array(wim->hdr.guid, WIMLIB_GUID_LEN);
2987 /* Clear references to resources that have not been written yet. */
2988 zero_reshdr(&wim->hdr.lookup_table_reshdr);
2989 zero_reshdr(&wim->hdr.xml_data_reshdr);
2990 zero_reshdr(&wim->hdr.boot_metadata_reshdr);
2991 zero_reshdr(&wim->hdr.integrity_table_reshdr);
2993 /* Set image count and boot index correctly for single image writes. */
2994 if (image != WIMLIB_ALL_IMAGES) {
2995 wim->hdr.image_count = 1;
2996 if (wim->hdr.boot_idx == image)
2997 wim->hdr.boot_idx = 1;
2999 wim->hdr.boot_idx = 0;
3002 /* Split WIMs can't be bootable. */
3003 if (total_parts != 1)
3004 wim->hdr.boot_idx = 0;
3006 /* Initialize output file descriptor. */
3007 if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR) {
3008 /* File descriptor was explicitly provided. Return error if
3009 * file descriptor is not seekable, unless writing a pipable WIM
3011 wim->out_fd.fd = *(const int*)path_or_fd;
3012 wim->out_fd.offset = 0;
3013 if (!filedes_is_seekable(&wim->out_fd)) {
3014 ret = WIMLIB_ERR_INVALID_PARAM;
3015 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
3016 goto out_restore_hdr;
3017 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
3018 ERROR("Can't include integrity check when "
3019 "writing pipable WIM to pipe!");
3020 goto out_restore_hdr;
3025 /* Filename of WIM to write was provided; open file descriptor
3027 ret = open_wim_writable(wim, (const tchar*)path_or_fd,
3028 O_TRUNC | O_CREAT | O_RDWR);
3030 goto out_restore_hdr;
3033 /* Write initial header. This is merely a "dummy" header since it
3034 * doesn't have all the information yet, so it will be overwritten later
3035 * (unless writing a pipable WIM). */
3036 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
3037 wim->hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
3038 ret = write_wim_header(&wim->hdr, &wim->out_fd);
3039 wim->hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS;
3041 goto out_restore_hdr;
3043 /* Write metadata resources and streams. */
3044 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE)) {
3045 /* Default case: create a normal (non-pipable) WIM. */
3046 ret = write_wim_streams(wim, image, write_flags, num_threads,
3047 stream_list_override,
3048 &lookup_table_list);
3050 goto out_restore_hdr;
3052 ret = write_wim_metadata_resources(wim, image, write_flags);
3054 goto out_restore_hdr;
3056 /* Non-default case: create pipable WIM. */
3057 ret = write_pipable_wim(wim, image, write_flags, num_threads,
3058 stream_list_override,
3059 &lookup_table_list);
3061 goto out_restore_hdr;
3062 write_flags |= WIMLIB_WRITE_FLAG_HEADER_AT_END;
3066 /* Write lookup table, XML data, and (optional) integrity table. */
3067 ret = finish_write(wim, image, write_flags, &lookup_table_list);
3069 memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header));
3070 (void)close_wim_writable(wim, write_flags);
3071 DEBUG("ret=%d", ret);
3075 /* Write a standalone WIM to a file or file descriptor. */
3077 write_standalone_wim(WIMStruct *wim, const void *path_or_fd,
3078 int image, int write_flags, unsigned num_threads)
3080 return write_wim_part(wim, path_or_fd, image, write_flags,
3081 num_threads, 1, 1, NULL, NULL);
3084 /* API function documented in wimlib.h */
3086 wimlib_write(WIMStruct *wim, const tchar *path,
3087 int image, int write_flags, unsigned num_threads)
3089 if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC)
3090 return WIMLIB_ERR_INVALID_PARAM;
3092 if (path == NULL || path[0] == T('\0'))
3093 return WIMLIB_ERR_INVALID_PARAM;
3095 return write_standalone_wim(wim, path, image, write_flags, num_threads);
3098 /* API function documented in wimlib.h */
3100 wimlib_write_to_fd(WIMStruct *wim, int fd,
3101 int image, int write_flags, unsigned num_threads)
3103 if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC)
3104 return WIMLIB_ERR_INVALID_PARAM;
3107 return WIMLIB_ERR_INVALID_PARAM;
3109 write_flags |= WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR;
3111 return write_standalone_wim(wim, &fd, image, write_flags, num_threads);
3115 any_images_modified(WIMStruct *wim)
3117 for (int i = 0; i < wim->hdr.image_count; i++)
3118 if (wim->image_metadata[i]->modified)
3124 check_resource_offset(struct wim_lookup_table_entry *lte, void *_wim)
3126 const WIMStruct *wim = _wim;
3127 off_t end_offset = *(const off_t*)wim->private;
3129 if (lte->resource_location == RESOURCE_IN_WIM && lte->rspec->wim == wim &&
3130 lte->rspec->offset_in_wim + lte->rspec->size_in_wim > end_offset)
3131 return WIMLIB_ERR_RESOURCE_ORDER;
3135 /* Make sure no file or metadata resources are located after the XML data (or
3136 * integrity table if present)--- otherwise we can't safely overwrite the WIM in
3137 * place and we return WIMLIB_ERR_RESOURCE_ORDER. */
3139 check_resource_offsets(WIMStruct *wim, off_t end_offset)
3144 wim->private = &end_offset;
3145 ret = for_lookup_table_entry(wim->lookup_table, check_resource_offset, wim);
3149 for (i = 0; i < wim->hdr.image_count; i++) {
3150 ret = check_resource_offset(wim->image_metadata[i]->metadata_lte, wim);
3158 * Overwrite a WIM, possibly appending streams to it.
3160 * A WIM looks like (or is supposed to look like) the following:
3162 * Header (212 bytes)
3163 * Streams and metadata resources (variable size)
3164 * Lookup table (variable size)
3165 * XML data (variable size)
3166 * Integrity table (optional) (variable size)
3168 * If we are not adding any streams or metadata resources, the lookup table is
3169 * unchanged--- so we only need to overwrite the XML data, integrity table, and
3170 * header. This operation is potentially unsafe if the program is abruptly
3171 * terminated while the XML data or integrity table are being overwritten, but
3172 * before the new header has been written. To partially alleviate this problem,
3173 * a special flag (WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) is passed to
3174 * finish_write() to cause a temporary WIM header to be written after the XML
3175 * data has been written. This may prevent the WIM from becoming corrupted if
3176 * the program is terminated while the integrity table is being calculated (but
3177 * no guarantees, due to write re-ordering...).
3179 * If we are adding new streams or images (metadata resources), the lookup table
3180 * needs to be changed, and those streams need to be written. In this case, we
3181 * try to perform a safe update of the WIM file by writing the streams *after*
3182 * the end of the previous WIM, then writing the new lookup table, XML data, and
3183 * (optionally) integrity table following the new streams. This will produce a
3184 * layout like the following:
3186 * Header (212 bytes)
3187 * (OLD) Streams and metadata resources (variable size)
3188 * (OLD) Lookup table (variable size)
3189 * (OLD) XML data (variable size)
3190 * (OLD) Integrity table (optional) (variable size)
3191 * (NEW) Streams and metadata resources (variable size)
3192 * (NEW) Lookup table (variable size)
3193 * (NEW) XML data (variable size)
3194 * (NEW) Integrity table (optional) (variable size)
3196 * At all points, the WIM is valid as nothing points to the new data yet. Then,
3197 * the header is overwritten to point to the new lookup table, XML data, and
3198 * integrity table, to produce the following layout:
3200 * Header (212 bytes)
3201 * Streams and metadata resources (variable size)
3202 * Nothing (variable size)
3203 * More Streams and metadata resources (variable size)
3204 * Lookup table (variable size)
3205 * XML data (variable size)
3206 * Integrity table (optional) (variable size)
3208 * This method allows an image to be appended to a large WIM very quickly, and
3209 * is crash-safe except in the case of write re-ordering, but the
3210 * disadvantage is that a small hole is left in the WIM where the old lookup
3211 * table, xml data, and integrity table were. (These usually only take up a
3212 * small amount of space compared to the streams, however.)
3215 overwrite_wim_inplace(WIMStruct *wim, int write_flags, unsigned num_threads)
3219 u64 old_lookup_table_end, old_xml_begin, old_xml_end;
3220 struct wim_header hdr_save;
3221 struct list_head stream_list;
3222 struct list_head lookup_table_list;
3223 struct filter_context filter_ctx;
3225 DEBUG("Overwriting `%"TS"' in-place", wim->filename);
3227 /* Save original header so it can be restored in case of error */
3228 memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header));
3230 /* Set default integrity flag. */
3231 if (!(write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
3232 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)))
3233 if (wim_has_integrity_table(wim))
3234 write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY;
3236 /* Set WIM version if writing solid resources. */
3237 if (write_flags & WIMLIB_WRITE_FLAG_SOLID)
3238 wim->hdr.wim_version = WIM_VERSION_SOLID;
3240 /* Set additional flags for overwrite. */
3241 write_flags |= WIMLIB_WRITE_FLAG_OVERWRITE |
3242 WIMLIB_WRITE_FLAG_STREAMS_OK;
3244 /* Make sure that the integrity table (if present) is after the XML
3245 * data, and that there are no stream resources, metadata resources, or
3246 * lookup tables after the XML data. Otherwise, these data would be
3248 old_xml_begin = wim->hdr.xml_data_reshdr.offset_in_wim;
3249 old_xml_end = old_xml_begin + wim->hdr.xml_data_reshdr.size_in_wim;
3250 old_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim +
3251 wim->hdr.lookup_table_reshdr.size_in_wim;
3252 if (wim->hdr.integrity_table_reshdr.offset_in_wim != 0 &&
3253 wim->hdr.integrity_table_reshdr.offset_in_wim < old_xml_end) {
3254 WARNING("Didn't expect the integrity table to be before the XML data");
3255 ret = WIMLIB_ERR_RESOURCE_ORDER;
3256 goto out_restore_memory_hdr;
3259 if (old_lookup_table_end > old_xml_begin) {
3260 WARNING("Didn't expect the lookup table to be after the XML data");
3261 ret = WIMLIB_ERR_RESOURCE_ORDER;
3262 goto out_restore_memory_hdr;
3265 /* Set @old_wim_end, which indicates the point beyond which we don't
3266 * allow any file and metadata resources to appear without returning
3267 * WIMLIB_ERR_RESOURCE_ORDER (due to the fact that we would otherwise
3268 * overwrite these resources). */
3269 if (!wim->image_deletion_occurred && !any_images_modified(wim)) {
3270 /* If no images have been modified and no images have been
3271 * deleted, a new lookup table does not need to be written. We
3272 * shall write the new XML data and optional integrity table
3273 * immediately after the lookup table. Note that this may
3274 * overwrite an existing integrity table. */
3275 DEBUG("Skipping writing lookup table "
3276 "(no images modified or deleted)");
3277 old_wim_end = old_lookup_table_end;
3278 write_flags |= WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE |
3279 WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML;
3280 } else if (wim->hdr.integrity_table_reshdr.offset_in_wim != 0) {
3281 /* Old WIM has an integrity table; begin writing new streams
3283 old_wim_end = wim->hdr.integrity_table_reshdr.offset_in_wim +
3284 wim->hdr.integrity_table_reshdr.size_in_wim;
3286 /* No existing integrity table; begin writing new streams after
3287 * the old XML data. */
3288 old_wim_end = old_xml_end;
3291 ret = check_resource_offsets(wim, old_wim_end);
3293 goto out_restore_memory_hdr;
3295 ret = prepare_stream_list_for_write(wim, WIMLIB_ALL_IMAGES, write_flags,
3296 &stream_list, &lookup_table_list,
3299 goto out_restore_memory_hdr;
3301 ret = open_wim_writable(wim, wim->filename, O_RDWR);
3303 goto out_restore_memory_hdr;
3305 ret = lock_wim_for_append(wim);
3309 /* Set WIM_HDR_FLAG_WRITE_IN_PROGRESS flag in header. */
3310 wim->hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
3311 ret = write_wim_header_flags(wim->hdr.flags, &wim->out_fd);
3313 ERROR_WITH_ERRNO("Error updating WIM header flags");
3314 goto out_unlock_wim;
3317 if (filedes_seek(&wim->out_fd, old_wim_end) == -1) {
3318 ERROR_WITH_ERRNO("Can't seek to end of WIM");
3319 ret = WIMLIB_ERR_WRITE;
3320 goto out_restore_physical_hdr;
3323 ret = wim_write_stream_list(wim,
3331 ret = write_wim_metadata_resources(wim, WIMLIB_ALL_IMAGES, write_flags);
3335 ret = finish_write(wim, WIMLIB_ALL_IMAGES, write_flags,
3336 &lookup_table_list);
3340 unlock_wim_for_append(wim);
3344 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
3345 WARNING("Truncating `%"TS"' to its original size (%"PRIu64" bytes)",
3346 wim->filename, old_wim_end);
3347 /* Return value of ftruncate() is ignored because this is
3348 * already an error path. */
3349 (void)ftruncate(wim->out_fd.fd, old_wim_end);
3351 out_restore_physical_hdr:
3352 (void)write_wim_header_flags(hdr_save.flags, &wim->out_fd);
3354 unlock_wim_for_append(wim);
3356 (void)close_wim_writable(wim, write_flags);
3357 out_restore_memory_hdr:
3358 memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header));
3363 overwrite_wim_via_tmpfile(WIMStruct *wim, int write_flags, unsigned num_threads)
3365 size_t wim_name_len;
3368 DEBUG("Overwriting `%"TS"' via a temporary file", wim->filename);
3370 /* Write the WIM to a temporary file in the same directory as the
3372 wim_name_len = tstrlen(wim->filename);
3373 tchar tmpfile[wim_name_len + 10];
3374 tmemcpy(tmpfile, wim->filename, wim_name_len);
3375 randomize_char_array_with_alnum(tmpfile + wim_name_len, 9);
3376 tmpfile[wim_name_len + 9] = T('\0');
3378 ret = wimlib_write(wim, tmpfile, WIMLIB_ALL_IMAGES,
3380 WIMLIB_WRITE_FLAG_FSYNC |
3381 WIMLIB_WRITE_FLAG_RETAIN_GUID,
3388 if (filedes_valid(&wim->in_fd)) {
3389 filedes_close(&wim->in_fd);
3390 filedes_invalidate(&wim->in_fd);
3393 /* Rename the new WIM file to the original WIM file. Note: on Windows
3394 * this actually calls win32_rename_replacement(), not _wrename(), so
3395 * that removing the existing destination file can be handled. */
3396 DEBUG("Renaming `%"TS"' to `%"TS"'", tmpfile, wim->filename);
3397 ret = trename(tmpfile, wim->filename);
3399 ERROR_WITH_ERRNO("Failed to rename `%"TS"' to `%"TS"'",
3400 tmpfile, wim->filename);
3407 return WIMLIB_ERR_RENAME;
3410 union wimlib_progress_info progress;
3411 progress.rename.from = tmpfile;
3412 progress.rename.to = wim->filename;
3413 return call_progress(wim->progfunc, WIMLIB_PROGRESS_MSG_RENAME,
3414 &progress, wim->progctx);
3417 /* Determine if the specified WIM file may be updated by appending in-place
3418 * rather than writing and replacing it with an entirely new file. */
3420 can_overwrite_wim_inplace(const WIMStruct *wim, int write_flags)
3422 /* REBUILD flag forces full rebuild. */
3423 if (write_flags & WIMLIB_WRITE_FLAG_REBUILD)
3426 /* Image deletions cause full rebuild by default. */
3427 if (wim->image_deletion_occurred &&
3428 !(write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE))
3431 /* Pipable WIMs cannot be updated in place, nor can a non-pipable WIM be
3432 * turned into a pipable WIM in-place. */
3433 if (wim_is_pipable(wim) || (write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
3436 /* The default compression type and compression chunk size selected for
3437 * the output WIM must be the same as those currently used for the WIM.
3439 if (wim->compression_type != wim->out_compression_type)
3441 if (wim->chunk_size != wim->out_chunk_size)
3447 /* API function documented in wimlib.h */
3449 wimlib_overwrite(WIMStruct *wim, int write_flags, unsigned num_threads)
3454 if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC)
3455 return WIMLIB_ERR_INVALID_PARAM;
3458 return WIMLIB_ERR_NO_FILENAME;
3460 orig_hdr_flags = wim->hdr.flags;
3461 if (write_flags & WIMLIB_WRITE_FLAG_IGNORE_READONLY_FLAG)
3462 wim->hdr.flags &= ~WIM_HDR_FLAG_READONLY;
3463 ret = can_modify_wim(wim);
3464 wim->hdr.flags = orig_hdr_flags;
3468 if (can_overwrite_wim_inplace(wim, write_flags)) {
3469 ret = overwrite_wim_inplace(wim, write_flags, num_threads);
3470 if (ret != WIMLIB_ERR_RESOURCE_ORDER)
3472 WARNING("Falling back to re-building entire WIM");
3474 return overwrite_wim_via_tmpfile(wim, write_flags, num_threads);