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"
55 # include "wimlib/win32.h" /* win32_rename_replacement() */
57 #include "wimlib/write.h"
58 #include "wimlib/xml.h"
61 /* wimlib internal flags used when writing resources. */
62 #define WRITE_RESOURCE_FLAG_RECOMPRESS 0x00000001
63 #define WRITE_RESOURCE_FLAG_PIPABLE 0x00000002
64 #define WRITE_RESOURCE_FLAG_PACK_STREAMS 0x00000004
65 #define WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE 0x00000008
68 write_flags_to_resource_flags(int write_flags)
70 int write_resource_flags = 0;
72 if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
73 write_resource_flags |= WRITE_RESOURCE_FLAG_RECOMPRESS;
74 if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
75 write_resource_flags |= WRITE_RESOURCE_FLAG_PIPABLE;
76 if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS)
77 write_resource_flags |= WRITE_RESOURCE_FLAG_PACK_STREAMS;
78 if (write_flags & WIMLIB_WRITE_FLAG_SEND_DONE_WITH_FILE_MESSAGES)
79 write_resource_flags |= WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE;
80 return write_resource_flags;
83 struct filter_context {
88 /* Determine specified stream should be filtered out from the write.
92 * < 0 : The stream should be hard-filtered; that is, not included in the
94 * 0 : The stream should not be filtered out.
95 * > 0 : The stream should be soft-filtered; that is, it already exists in the
96 * WIM file and may not need to be written again.
99 stream_filtered(const struct wim_lookup_table_entry *lte,
100 const struct filter_context *ctx)
108 write_flags = ctx->write_flags;
111 if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE &&
112 lte->resource_location == RESOURCE_IN_WIM &&
113 lte->rspec->wim == wim)
116 if (write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS &&
117 lte->resource_location == RESOURCE_IN_WIM &&
118 lte->rspec->wim != wim)
125 stream_hard_filtered(const struct wim_lookup_table_entry *lte,
126 struct filter_context *ctx)
128 return stream_filtered(lte, ctx) < 0;
132 may_soft_filter_streams(const struct filter_context *ctx)
136 return ctx->write_flags & WIMLIB_WRITE_FLAG_OVERWRITE;
140 may_hard_filter_streams(const struct filter_context *ctx)
144 return ctx->write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS;
148 may_filter_streams(const struct filter_context *ctx)
150 return (may_soft_filter_streams(ctx) ||
151 may_hard_filter_streams(ctx));
155 /* Return true if the specified resource is compressed and the compressed data
156 * can be reused with the specified output parameters. */
158 can_raw_copy(const struct wim_lookup_table_entry *lte,
159 int write_resource_flags, int out_ctype, u32 out_chunk_size)
161 const struct wim_resource_spec *rspec;
163 if (write_resource_flags & WRITE_RESOURCE_FLAG_RECOMPRESS)
166 if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE)
169 if (lte->resource_location != RESOURCE_IN_WIM)
174 if (rspec->is_pipable != !!(write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE))
177 if (rspec->flags & WIM_RESHDR_FLAG_COMPRESSED) {
178 /* Normal compressed resource: Must use same compression type
180 return (rspec->compression_type == out_ctype &&
181 rspec->chunk_size == out_chunk_size);
184 if ((rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) &&
185 (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
187 /* Packed resource: Such resources may contain multiple streams,
188 * and in general only a subset of them need to be written. As
189 * a heuristic, re-use the raw data if more than two-thirds the
190 * uncompressed size is being written. */
192 /* Note: packed resources contain a header that specifies the
193 * compression type and chunk size; therefore we don't need to
194 * check if they are compatible with @out_ctype and
195 * @out_chunk_size. */
197 struct wim_lookup_table_entry *res_stream;
200 list_for_each_entry(res_stream, &rspec->stream_list, rspec_node)
201 if (res_stream->will_be_in_output_wim)
202 write_size += res_stream->size;
204 return (write_size > rspec->uncompressed_size * 2 / 3);
211 filter_resource_flags(u8 flags)
213 return (flags & ~(WIM_RESHDR_FLAG_PACKED_STREAMS |
214 WIM_RESHDR_FLAG_COMPRESSED |
215 WIM_RESHDR_FLAG_SPANNED |
216 WIM_RESHDR_FLAG_FREE));
220 stream_set_out_reshdr_for_reuse(struct wim_lookup_table_entry *lte)
222 const struct wim_resource_spec *rspec;
224 wimlib_assert(lte->resource_location == RESOURCE_IN_WIM);
227 if (rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
229 wimlib_assert(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS);
231 lte->out_reshdr.offset_in_wim = lte->offset_in_res;
232 lte->out_reshdr.uncompressed_size = 0;
233 lte->out_reshdr.size_in_wim = lte->size;
235 lte->out_res_offset_in_wim = rspec->offset_in_wim;
236 lte->out_res_size_in_wim = rspec->size_in_wim;
237 lte->out_res_uncompressed_size = rspec->uncompressed_size;
239 wimlib_assert(!(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS));
241 lte->out_reshdr.offset_in_wim = rspec->offset_in_wim;
242 lte->out_reshdr.uncompressed_size = rspec->uncompressed_size;
243 lte->out_reshdr.size_in_wim = rspec->size_in_wim;
245 lte->out_reshdr.flags = lte->flags;
249 /* Write the header for a stream in a pipable WIM. */
251 write_pwm_stream_header(const struct wim_lookup_table_entry *lte,
252 struct filedes *out_fd,
253 int additional_reshdr_flags)
255 struct pwm_stream_hdr stream_hdr;
259 stream_hdr.magic = cpu_to_le64(PWM_STREAM_MAGIC);
260 stream_hdr.uncompressed_size = cpu_to_le64(lte->size);
261 if (additional_reshdr_flags & PWM_RESHDR_FLAG_UNHASHED) {
262 zero_out_hash(stream_hdr.hash);
264 wimlib_assert(!lte->unhashed);
265 copy_hash(stream_hdr.hash, lte->hash);
268 reshdr_flags = filter_resource_flags(lte->flags);
269 reshdr_flags |= additional_reshdr_flags;
270 stream_hdr.flags = cpu_to_le32(reshdr_flags);
271 ret = full_write(out_fd, &stream_hdr, sizeof(stream_hdr));
273 ERROR_WITH_ERRNO("Write error");
277 struct write_streams_progress_data {
278 wimlib_progress_func_t progfunc;
280 union wimlib_progress_info progress;
281 uint64_t next_progress;
285 do_write_streams_progress(struct write_streams_progress_data *progress_data,
290 union wimlib_progress_info *progress = &progress_data->progress;
294 progress->write_streams.total_bytes -= complete_size;
295 progress->write_streams.total_streams -= complete_count;
296 if (progress_data->next_progress != ~(uint64_t)0 &&
297 progress_data->next_progress > progress->write_streams.total_bytes)
299 progress_data->next_progress = progress->write_streams.total_bytes;
302 progress->write_streams.completed_bytes += complete_size;
303 progress->write_streams.completed_streams += complete_count;
306 if (progress->write_streams.completed_bytes >= progress_data->next_progress)
308 ret = call_progress(progress_data->progfunc,
309 WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
311 progress_data->progctx);
315 if (progress_data->next_progress == progress->write_streams.total_bytes) {
316 progress_data->next_progress = ~(uint64_t)0;
318 /* Handle rate-limiting of messages */
320 /* Send new message as soon as another 1/128 of the
321 * total has been written. (Arbitrary number.) */
322 progress_data->next_progress =
323 progress->write_streams.completed_bytes +
324 progress->write_streams.total_bytes / 128;
326 /* ... Unless that would be more than 5000000 bytes, in
327 * which case send the next after the next 5000000
328 * bytes. (Another arbitrary number.) */
329 if (progress->write_streams.completed_bytes + 5000000 <
330 progress_data->next_progress)
331 progress_data->next_progress =
332 progress->write_streams.completed_bytes + 5000000;
334 /* ... But always send a message as soon as we're
335 * completely done. */
336 if (progress->write_streams.total_bytes <
337 progress_data->next_progress)
338 progress_data->next_progress =
339 progress->write_streams.total_bytes;
345 struct write_streams_ctx {
346 /* File descriptor the streams are being written to. */
347 struct filedes *out_fd;
349 /* Lookup table for the WIMStruct on whose behalf the streams are being
351 struct wim_lookup_table *lookup_table;
353 /* Compression format to use. */
356 /* Maximum uncompressed chunk size in compressed resources to use. */
359 /* Flags that affect how the streams will be written. */
360 int write_resource_flags;
362 /* Data used for issuing WRITE_STREAMS progress. */
363 struct write_streams_progress_data progress_data;
365 struct filter_context *filter_ctx;
367 /* Upper bound on the total number of bytes that need to be compressed.
369 u64 num_bytes_to_compress;
371 /* Pointer to the chunk_compressor implementation being used for
372 * compressing chunks of data, or NULL if chunks are being written
374 struct chunk_compressor *compressor;
376 /* Buffer for dividing the read data into chunks of size
377 * @out_chunk_size. */
380 /* Number of bytes in @chunk_buf that are currently filled. */
381 size_t chunk_buf_filled;
383 /* List of streams that currently have chunks being compressed. */
384 struct list_head pending_streams;
386 /* List of streams in the resource pack. Streams are moved here after
387 * @pending_streams only when writing a packed resource. */
388 struct list_head pack_streams;
390 /* Current uncompressed offset in the stream being read. */
391 u64 cur_read_stream_offset;
393 /* Uncompressed size of the stream currently being read. */
394 u64 cur_read_stream_size;
396 /* Current uncompressed offset in the stream being written. */
397 u64 cur_write_stream_offset;
399 /* Uncompressed size of resource currently being written. */
400 u64 cur_write_res_size;
402 /* Array that is filled in with compressed chunk sizes as a resource is
406 /* Index of next entry in @chunk_csizes to fill in. */
409 /* Number of entries in @chunk_csizes currently allocated. */
410 size_t num_alloc_chunks;
412 /* Offset in the output file of the start of the chunks of the resource
413 * currently being written. */
414 u64 chunks_start_offset;
417 /* Reserve space for the chunk table and prepare to accumulate the chunk table
420 begin_chunk_table(struct write_streams_ctx *ctx, u64 res_expected_size)
422 u64 expected_num_chunks;
423 u64 expected_num_chunk_entries;
427 /* Calculate the number of chunks and chunk entries that should be
428 * needed for the resource. These normally will be the final values,
429 * but in PACKED_STREAMS mode some of the streams we're planning to
430 * write into the resource may be duplicates, and therefore discarded,
431 * potentially decreasing the number of chunk entries needed. */
432 expected_num_chunks = DIV_ROUND_UP(res_expected_size, ctx->out_chunk_size);
433 expected_num_chunk_entries = expected_num_chunks;
434 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
435 expected_num_chunk_entries--;
437 /* Make sure the chunk_csizes array is long enough to store the
438 * compressed size of each chunk. */
439 if (expected_num_chunks > ctx->num_alloc_chunks) {
440 u64 new_length = expected_num_chunks + 50;
442 if ((size_t)new_length != new_length) {
443 ERROR("Resource size too large (%"PRIu64" bytes!",
445 return WIMLIB_ERR_NOMEM;
448 FREE(ctx->chunk_csizes);
449 ctx->chunk_csizes = MALLOC(new_length * sizeof(ctx->chunk_csizes[0]));
450 if (ctx->chunk_csizes == NULL) {
451 ctx->num_alloc_chunks = 0;
452 return WIMLIB_ERR_NOMEM;
454 ctx->num_alloc_chunks = new_length;
457 ctx->chunk_index = 0;
459 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE)) {
460 /* Reserve space for the chunk table in the output file. In the
461 * case of packed resources this reserves the upper bound for
462 * the needed space, not necessarily the exact space which will
463 * prove to be needed. At this point, we just use @chunk_csizes
464 * for a buffer of 0's because the actual compressed chunk sizes
466 reserve_size = expected_num_chunk_entries *
467 get_chunk_entry_size(res_expected_size,
468 0 != (ctx->write_resource_flags &
469 WRITE_RESOURCE_FLAG_PACK_STREAMS));
470 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)
471 reserve_size += sizeof(struct alt_chunk_table_header_disk);
472 memset(ctx->chunk_csizes, 0, reserve_size);
473 ret = full_write(ctx->out_fd, ctx->chunk_csizes, reserve_size);
481 begin_write_resource(struct write_streams_ctx *ctx, u64 res_expected_size)
485 wimlib_assert(res_expected_size != 0);
487 if (ctx->compressor != NULL) {
488 ret = begin_chunk_table(ctx, res_expected_size);
493 /* Output file descriptor is now positioned at the offset at which to
494 * write the first chunk of the resource. */
495 ctx->chunks_start_offset = ctx->out_fd->offset;
496 ctx->cur_write_stream_offset = 0;
497 ctx->cur_write_res_size = res_expected_size;
502 end_chunk_table(struct write_streams_ctx *ctx, u64 res_actual_size,
503 u64 *res_start_offset_ret, u64 *res_store_size_ret)
505 size_t actual_num_chunks;
506 size_t actual_num_chunk_entries;
507 size_t chunk_entry_size;
510 actual_num_chunks = ctx->chunk_index;
511 actual_num_chunk_entries = actual_num_chunks;
512 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
513 actual_num_chunk_entries--;
515 chunk_entry_size = get_chunk_entry_size(res_actual_size,
516 0 != (ctx->write_resource_flags &
517 WRITE_RESOURCE_FLAG_PACK_STREAMS));
519 typedef le64 _may_alias_attribute aliased_le64_t;
520 typedef le32 _may_alias_attribute aliased_le32_t;
522 if (chunk_entry_size == 4) {
523 aliased_le32_t *entries = (aliased_le32_t*)ctx->chunk_csizes;
525 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
526 for (size_t i = 0; i < actual_num_chunk_entries; i++)
527 entries[i] = cpu_to_le32(ctx->chunk_csizes[i]);
529 u32 offset = ctx->chunk_csizes[0];
530 for (size_t i = 0; i < actual_num_chunk_entries; i++) {
531 u32 next_size = ctx->chunk_csizes[i + 1];
532 entries[i] = cpu_to_le32(offset);
537 aliased_le64_t *entries = (aliased_le64_t*)ctx->chunk_csizes;
539 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
540 for (size_t i = 0; i < actual_num_chunk_entries; i++)
541 entries[i] = cpu_to_le64(ctx->chunk_csizes[i]);
543 u64 offset = ctx->chunk_csizes[0];
544 for (size_t i = 0; i < actual_num_chunk_entries; i++) {
545 u64 next_size = ctx->chunk_csizes[i + 1];
546 entries[i] = cpu_to_le64(offset);
552 size_t chunk_table_size = actual_num_chunk_entries * chunk_entry_size;
553 u64 res_start_offset;
556 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
557 ret = full_write(ctx->out_fd, ctx->chunk_csizes, chunk_table_size);
560 res_end_offset = ctx->out_fd->offset;
561 res_start_offset = ctx->chunks_start_offset;
563 res_end_offset = ctx->out_fd->offset;
565 u64 chunk_table_offset;
567 chunk_table_offset = ctx->chunks_start_offset - chunk_table_size;
569 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
570 struct alt_chunk_table_header_disk hdr;
572 hdr.res_usize = cpu_to_le64(res_actual_size);
573 hdr.chunk_size = cpu_to_le32(ctx->out_chunk_size);
574 hdr.compression_format = cpu_to_le32(ctx->out_ctype);
576 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 1);
577 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 2);
578 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3);
580 ret = full_pwrite(ctx->out_fd, &hdr, sizeof(hdr),
581 chunk_table_offset - sizeof(hdr));
584 res_start_offset = chunk_table_offset - sizeof(hdr);
586 res_start_offset = chunk_table_offset;
589 ret = full_pwrite(ctx->out_fd, ctx->chunk_csizes,
590 chunk_table_size, chunk_table_offset);
595 *res_start_offset_ret = res_start_offset;
596 *res_store_size_ret = res_end_offset - res_start_offset;
601 ERROR_WITH_ERRNO("Write error");
605 /* Finish writing a WIM resource by writing or updating the chunk table (if not
606 * writing the data uncompressed) and loading its metadata into @out_reshdr. */
608 end_write_resource(struct write_streams_ctx *ctx, struct wim_reshdr *out_reshdr)
612 u64 res_uncompressed_size;
613 u64 res_offset_in_wim;
615 wimlib_assert(ctx->cur_write_stream_offset == ctx->cur_write_res_size ||
616 (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS));
617 res_uncompressed_size = ctx->cur_write_res_size;
619 if (ctx->compressor) {
620 ret = end_chunk_table(ctx, res_uncompressed_size,
621 &res_offset_in_wim, &res_size_in_wim);
625 res_offset_in_wim = ctx->chunks_start_offset;
626 res_size_in_wim = ctx->out_fd->offset - res_offset_in_wim;
628 out_reshdr->uncompressed_size = res_uncompressed_size;
629 out_reshdr->size_in_wim = res_size_in_wim;
630 out_reshdr->offset_in_wim = res_offset_in_wim;
631 DEBUG("Finished writing resource: %"PRIu64" => %"PRIu64" @ %"PRIu64"",
632 res_uncompressed_size, res_size_in_wim, res_offset_in_wim);
636 /* No more data streams of the file at @path are needed. */
638 done_with_file(const tchar *path, wimlib_progress_func_t progfunc, void *progctx)
640 union wimlib_progress_info info;
642 info.done_with_file.path_to_file = path;
644 return call_progress(progfunc, WIMLIB_PROGRESS_MSG_DONE_WITH_FILE,
649 is_file_stream(const struct wim_lookup_table_entry *lte)
651 return lte->resource_location == RESOURCE_IN_FILE_ON_DISK
653 || lte->resource_location == RESOURCE_IN_WINNT_FILE_ON_DISK
654 || lte->resource_location == RESOURCE_WIN32_ENCRYPTED
660 do_done_with_stream(struct wim_lookup_table_entry *lte,
661 wimlib_progress_func_t progfunc, void *progctx)
664 struct wim_inode *inode;
666 if (!lte->may_send_done_with_file)
669 inode = lte->file_inode;
671 wimlib_assert(inode != NULL);
672 wimlib_assert(inode->num_remaining_streams > 0);
673 if (--inode->num_remaining_streams > 0)
677 /* XXX: This logic really should be somewhere else. */
679 /* We want the path to the file, but lte->file_on_disk might actually
680 * refer to a named data stream. Temporarily strip the named data
681 * stream from the path. */
682 wchar_t *p_colon = NULL;
683 wchar_t *p_question_mark = NULL;
684 const wchar_t *p_stream_name;
686 p_stream_name = path_stream_name(lte->file_on_disk);
687 if (unlikely(p_stream_name)) {
688 p_colon = (wchar_t *)(p_stream_name - 1);
689 wimlib_assert(*p_colon == L':');
693 /* We also should use a fake Win32 path instead of a NT path */
694 if (!wcsncmp(lte->file_on_disk, L"\\??\\", 4)) {
695 p_question_mark = <e->file_on_disk[1];
696 *p_question_mark = L'\\';
700 ret = done_with_file(lte->file_on_disk, progfunc, progctx);
706 *p_question_mark = L'?';
711 /* Handle WIMLIB_WRITE_FLAG_SEND_DONE_WITH_FILE_MESSAGES mode. */
713 done_with_stream(struct wim_lookup_table_entry *lte,
714 struct write_streams_ctx *ctx)
716 if (likely(!(ctx->write_resource_flags &
717 WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE)))
719 return do_done_with_stream(lte, ctx->progress_data.progfunc,
720 ctx->progress_data.progctx);
723 /* Begin processing a stream for writing. */
725 write_stream_begin_read(struct wim_lookup_table_entry *lte, void *_ctx)
727 struct write_streams_ctx *ctx = _ctx;
730 wimlib_assert(lte->size > 0);
732 ctx->cur_read_stream_offset = 0;
733 ctx->cur_read_stream_size = lte->size;
735 /* As an optimization, we allow some streams to be "unhashed", meaning
736 * their SHA1 message digests are unknown. This is the case with
737 * streams that are added by scanning a directry tree with
738 * wimlib_add_image(), for example. Since WIM uses single-instance
739 * streams, we don't know whether such each such stream really need to
740 * written until it is actually checksummed, unless it has a unique
741 * size. In such cases we read and checksum the stream in this
742 * function, thereby advancing ahead of read_stream_list(), which will
743 * still provide the data again to write_stream_process_chunk(). This
744 * is okay because an unhashed stream cannot be in a WIM resource, which
745 * might be costly to decompress. */
746 if (ctx->lookup_table != NULL && lte->unhashed && !lte->unique_size) {
748 struct wim_lookup_table_entry *lte_new;
750 ret = hash_unhashed_stream(lte, ctx->lookup_table, <e_new);
753 if (lte_new != lte) {
754 /* Duplicate stream detected. */
756 if (lte_new->will_be_in_output_wim ||
757 stream_filtered(lte_new, ctx->filter_ctx))
759 /* The duplicate stream is already being
760 * included in the output WIM, or it would be
761 * filtered out if it had been. Skip writing
762 * this stream (and reading it again) entirely,
763 * passing its output reference count to the
764 * duplicate stream in the former case. */
765 DEBUG("Discarding duplicate stream of "
766 "length %"PRIu64, lte->size);
767 ret = do_write_streams_progress(&ctx->progress_data,
770 list_del(<e->write_streams_list);
771 list_del(<e->lookup_table_list);
772 if (lte_new->will_be_in_output_wim)
773 lte_new->out_refcnt += lte->out_refcnt;
774 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)
775 ctx->cur_write_res_size -= lte->size;
777 ret = done_with_stream(lte, ctx);
778 free_lookup_table_entry(lte);
781 return BEGIN_STREAM_STATUS_SKIP_STREAM;
783 /* The duplicate stream can validly be written,
784 * but was not marked as such. Discard the
785 * current stream entry and use the duplicate,
786 * but actually freeing the current entry must
787 * wait until read_stream_list() has finished
788 * reading its data. */
789 DEBUG("Stream duplicate, but not already "
790 "selected for writing.");
791 list_replace(<e->write_streams_list,
792 <e_new->write_streams_list);
793 list_replace(<e->lookup_table_list,
794 <e_new->lookup_table_list);
795 lte->will_be_in_output_wim = 0;
796 lte_new->out_refcnt = lte->out_refcnt;
797 lte_new->will_be_in_output_wim = 1;
798 lte_new->may_send_done_with_file = 0;
803 list_move_tail(<e->write_streams_list, &ctx->pending_streams);
807 /* Rewrite a stream that was just written compressed as uncompressed instead.
808 * This function is optional, but if a stream did not compress to less than its
809 * original size, it might as well be written uncompressed. */
811 write_stream_uncompressed(struct wim_lookup_table_entry *lte,
812 struct filedes *out_fd)
815 u64 begin_offset = lte->out_reshdr.offset_in_wim;
816 u64 end_offset = out_fd->offset;
818 if (filedes_seek(out_fd, begin_offset) == -1)
821 ret = extract_full_stream_to_fd(lte, out_fd);
823 /* Error reading the uncompressed data. */
824 if (out_fd->offset == begin_offset &&
825 filedes_seek(out_fd, end_offset) != -1)
827 /* Nothing was actually written yet, and we successfully
828 * seeked to the end of the compressed resource, so
829 * don't issue a hard error; just keep the compressed
830 * resource instead. */
831 WARNING("Recovered compressed stream of "
832 "size %"PRIu64", continuing on.",
839 wimlib_assert(out_fd->offset - begin_offset == lte->size);
841 if (out_fd->offset < end_offset &&
842 0 != ftruncate(out_fd->fd, out_fd->offset))
844 ERROR_WITH_ERRNO("Can't truncate output file to "
845 "offset %"PRIu64, out_fd->offset);
846 return WIMLIB_ERR_WRITE;
849 lte->out_reshdr.size_in_wim = lte->size;
850 lte->out_reshdr.flags &= ~(WIM_RESHDR_FLAG_COMPRESSED |
851 WIM_RESHDR_FLAG_PACKED_STREAMS);
855 /* Returns true if the specified stream should be truncated from the WIM file
856 * and re-written as uncompressed. lte->out_reshdr must be filled in from the
857 * initial write of the stream. */
859 should_rewrite_stream_uncompressed(const struct write_streams_ctx *ctx,
860 const struct wim_lookup_table_entry *lte)
862 /* If the compressed data is smaller than the uncompressed data, prefer
863 * the compressed data. */
864 if (lte->out_reshdr.size_in_wim < lte->out_reshdr.uncompressed_size)
867 /* If we're not actually writing compressed data, then there's no need
869 if (!ctx->compressor)
872 /* If writing a pipable WIM, everything we write to the output is final
873 * (it might actually be a pipe!). */
874 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE)
877 /* If the stream that would need to be re-read is located in a solid
878 * block in another WIM file, then re-reading it would be costly. So
881 * Exception: if the compressed size happens to be *exactly* the same as
882 * the uncompressed size, then the stream *must* be written uncompressed
883 * in order to remain compatible with the Windows Overlay Filesystem
884 * Filter Driver (WOF).
886 * TODO: we are currently assuming that the optimization for
887 * single-chunk resources in maybe_rewrite_stream_uncompressed()
888 * prevents this case from being triggered too often. To fully prevent
889 * excessive decompressions in degenerate cases, we really should
890 * obtain the uncompressed data by decompressing the compressed data we
891 * wrote to the output file.
893 if ((lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) &&
894 (lte->out_reshdr.size_in_wim != lte->out_reshdr.uncompressed_size))
901 maybe_rewrite_stream_uncompressed(struct write_streams_ctx *ctx,
902 struct wim_lookup_table_entry *lte)
904 if (!should_rewrite_stream_uncompressed(ctx, lte))
907 /* Regular (non-solid) WIM resources with exactly one chunk and
908 * compressed size equal to uncompressed size are exactly the same as
909 * the corresponding compressed data --- since there must be 0 entries
910 * in the chunk table and the only chunk must be stored uncompressed.
911 * In this case, there's no need to rewrite anything. */
912 if (ctx->chunk_index == 1 &&
913 lte->out_reshdr.size_in_wim == lte->out_reshdr.uncompressed_size)
915 lte->out_reshdr.flags &= ~WIM_RESHDR_FLAG_COMPRESSED;
919 return write_stream_uncompressed(lte, ctx->out_fd);
922 /* Write the next chunk of (typically compressed) data to the output WIM,
923 * handling the writing of the chunk table. */
925 write_chunk(struct write_streams_ctx *ctx, const void *cchunk,
926 size_t csize, size_t usize)
930 struct wim_lookup_table_entry *lte;
931 u32 completed_stream_count;
934 lte = list_entry(ctx->pending_streams.next,
935 struct wim_lookup_table_entry, write_streams_list);
937 if (ctx->cur_write_stream_offset == 0 &&
938 !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
940 /* Starting to write a new stream in non-packed mode. */
942 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
943 int additional_reshdr_flags = 0;
944 if (ctx->compressor != NULL)
945 additional_reshdr_flags |= WIM_RESHDR_FLAG_COMPRESSED;
947 DEBUG("Writing pipable WIM stream header "
948 "(offset=%"PRIu64")", ctx->out_fd->offset);
950 ret = write_pwm_stream_header(lte, ctx->out_fd,
951 additional_reshdr_flags);
956 ret = begin_write_resource(ctx, lte->size);
961 if (ctx->compressor != NULL) {
962 /* Record the compresed chunk size. */
963 wimlib_assert(ctx->chunk_index < ctx->num_alloc_chunks);
964 ctx->chunk_csizes[ctx->chunk_index++] = csize;
966 /* If writing a pipable WIM, before the chunk data write a chunk
967 * header that provides the compressed chunk size. */
968 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
969 struct pwm_chunk_hdr chunk_hdr = {
970 .compressed_size = cpu_to_le32(csize),
972 ret = full_write(ctx->out_fd, &chunk_hdr,
979 /* Write the chunk data. */
980 ret = full_write(ctx->out_fd, cchunk, csize);
984 ctx->cur_write_stream_offset += usize;
986 completed_size = usize;
987 completed_stream_count = 0;
988 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
989 /* Wrote chunk in packed mode. It may have finished multiple
991 struct wim_lookup_table_entry *next_lte;
993 while (lte && ctx->cur_write_stream_offset >= lte->size) {
995 ctx->cur_write_stream_offset -= lte->size;
997 if (ctx->cur_write_stream_offset)
998 next_lte = list_entry(lte->write_streams_list.next,
999 struct wim_lookup_table_entry,
1000 write_streams_list);
1004 ret = done_with_stream(lte, ctx);
1007 list_move_tail(<e->write_streams_list, &ctx->pack_streams);
1008 completed_stream_count++;
1013 /* Wrote chunk in non-packed mode. It may have finished a
1015 if (ctx->cur_write_stream_offset == lte->size) {
1017 wimlib_assert(ctx->cur_write_stream_offset ==
1018 ctx->cur_write_res_size);
1020 ret = end_write_resource(ctx, <e->out_reshdr);
1024 lte->out_reshdr.flags = filter_resource_flags(lte->flags);
1025 if (ctx->compressor != NULL)
1026 lte->out_reshdr.flags |= WIM_RESHDR_FLAG_COMPRESSED;
1028 ret = maybe_rewrite_stream_uncompressed(ctx, lte);
1032 wimlib_assert(lte->out_reshdr.uncompressed_size == lte->size);
1034 ctx->cur_write_stream_offset = 0;
1036 ret = done_with_stream(lte, ctx);
1039 list_del(<e->write_streams_list);
1040 completed_stream_count++;
1044 return do_write_streams_progress(&ctx->progress_data,
1045 completed_size, completed_stream_count,
1049 ERROR_WITH_ERRNO("Write error");
1054 submit_chunk_for_compression(struct write_streams_ctx *ctx,
1055 const void *chunk, size_t size)
1057 /* While we are unable to submit the chunk for compression (due to too
1058 * many chunks already outstanding), retrieve and write the next
1059 * compressed chunk. */
1060 while (!ctx->compressor->submit_chunk(ctx->compressor, chunk, size)) {
1067 bret = ctx->compressor->get_chunk(ctx->compressor,
1068 &cchunk, &csize, &usize);
1070 wimlib_assert(bret);
1072 ret = write_chunk(ctx, cchunk, csize, usize);
1079 /* Process the next chunk of data to be written to a WIM resource. */
1081 write_stream_process_chunk(const void *chunk, size_t size, void *_ctx)
1083 struct write_streams_ctx *ctx = _ctx;
1085 const u8 *chunkptr, *chunkend;
1087 wimlib_assert(size != 0);
1089 if (ctx->compressor == NULL) {
1090 /* Write chunk uncompressed. */
1091 ret = write_chunk(ctx, chunk, size, size);
1094 ctx->cur_read_stream_offset += size;
1098 /* Submit the chunk for compression, but take into account that the
1099 * @size the chunk was provided in may not correspond to the
1100 * @out_chunk_size being used for compression. */
1102 chunkend = chunkptr + size;
1104 const u8 *resized_chunk;
1105 size_t needed_chunk_size;
1107 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
1108 needed_chunk_size = ctx->out_chunk_size;
1110 u64 res_bytes_remaining;
1112 res_bytes_remaining = ctx->cur_read_stream_size -
1113 ctx->cur_read_stream_offset;
1114 needed_chunk_size = min(ctx->out_chunk_size,
1115 ctx->chunk_buf_filled +
1116 res_bytes_remaining);
1119 if (ctx->chunk_buf_filled == 0 &&
1120 chunkend - chunkptr >= needed_chunk_size)
1122 /* No intermediate buffering needed. */
1123 resized_chunk = chunkptr;
1124 chunkptr += needed_chunk_size;
1125 ctx->cur_read_stream_offset += needed_chunk_size;
1127 /* Intermediate buffering needed. */
1128 size_t bytes_consumed;
1130 bytes_consumed = min(chunkend - chunkptr,
1131 needed_chunk_size - ctx->chunk_buf_filled);
1133 memcpy(&ctx->chunk_buf[ctx->chunk_buf_filled],
1134 chunkptr, bytes_consumed);
1136 chunkptr += bytes_consumed;
1137 ctx->cur_read_stream_offset += bytes_consumed;
1138 ctx->chunk_buf_filled += bytes_consumed;
1139 if (ctx->chunk_buf_filled == needed_chunk_size) {
1140 resized_chunk = ctx->chunk_buf;
1141 ctx->chunk_buf_filled = 0;
1148 ret = submit_chunk_for_compression(ctx, resized_chunk,
1153 } while (chunkptr != chunkend);
1157 /* Finish processing a stream for writing. It may not have been completely
1158 * written yet, as the chunk_compressor implementation may still have chunks
1159 * buffered or being compressed. */
1161 write_stream_end_read(struct wim_lookup_table_entry *lte, int status, void *_ctx)
1163 struct write_streams_ctx *ctx = _ctx;
1165 wimlib_assert(ctx->cur_read_stream_offset == ctx->cur_read_stream_size || status);
1167 if (!lte->will_be_in_output_wim) {
1168 /* The 'lte' stream was a duplicate. Now that its data has
1169 * finished being read, it is being discarded in favor of the
1170 * duplicate entry. It therefore is no longer needed, and we
1171 * can fire the DONE_WITH_FILE callback because the file will
1172 * not be read again.
1174 * Note: we can't yet fire DONE_WITH_FILE for non-duplicate
1175 * streams, since it needs to be possible to re-read the file if
1176 * it does not compress to less than its original size. */
1178 status = done_with_stream(lte, ctx);
1179 free_lookup_table_entry(lte);
1180 } else if (!status && lte->unhashed && ctx->lookup_table != NULL) {
1181 /* The 'lte' stream was not a duplicate and was previously
1182 * unhashed. Since we passed COMPUTE_MISSING_STREAM_HASHES to
1183 * read_stream_list(), lte->hash is now computed and valid. So
1184 * turn this stream into a "hashed" stream. */
1185 list_del(<e->unhashed_list);
1186 lookup_table_insert(ctx->lookup_table, lte);
1192 /* Compute statistics about a list of streams that will be written.
1194 * Assumes the streams are sorted such that all streams located in each distinct
1195 * WIM (specified by WIMStruct) are together. */
1197 compute_stream_list_stats(struct list_head *stream_list,
1198 struct write_streams_ctx *ctx)
1200 struct wim_lookup_table_entry *lte;
1201 u64 total_bytes = 0;
1202 u64 num_streams = 0;
1203 u64 total_parts = 0;
1204 WIMStruct *prev_wim_part = NULL;
1206 list_for_each_entry(lte, stream_list, write_streams_list) {
1208 total_bytes += lte->size;
1209 if (lte->resource_location == RESOURCE_IN_WIM) {
1210 if (prev_wim_part != lte->rspec->wim) {
1211 prev_wim_part = lte->rspec->wim;
1216 ctx->progress_data.progress.write_streams.total_bytes = total_bytes;
1217 ctx->progress_data.progress.write_streams.total_streams = num_streams;
1218 ctx->progress_data.progress.write_streams.completed_bytes = 0;
1219 ctx->progress_data.progress.write_streams.completed_streams = 0;
1220 ctx->progress_data.progress.write_streams.compression_type = ctx->out_ctype;
1221 ctx->progress_data.progress.write_streams.total_parts = total_parts;
1222 ctx->progress_data.progress.write_streams.completed_parts = 0;
1223 ctx->progress_data.next_progress = 0;
1226 /* Find streams in @stream_list that can be copied to the output WIM in raw form
1227 * rather than compressed. Delete these streams from @stream_list and move them
1228 * to @raw_copy_streams. Return the total uncompressed size of the streams that
1229 * need to be compressed. */
1231 find_raw_copy_streams(struct list_head *stream_list,
1232 int write_resource_flags,
1235 struct list_head *raw_copy_streams)
1237 struct wim_lookup_table_entry *lte, *tmp;
1238 u64 num_bytes_to_compress = 0;
1240 INIT_LIST_HEAD(raw_copy_streams);
1242 /* Initialize temporary raw_copy_ok flag. */
1243 list_for_each_entry(lte, stream_list, write_streams_list)
1244 if (lte->resource_location == RESOURCE_IN_WIM)
1245 lte->rspec->raw_copy_ok = 0;
1247 list_for_each_entry_safe(lte, tmp, stream_list, write_streams_list) {
1248 if (lte->resource_location == RESOURCE_IN_WIM &&
1249 lte->rspec->raw_copy_ok)
1251 list_move_tail(<e->write_streams_list,
1253 } else if (can_raw_copy(lte, write_resource_flags,
1254 out_ctype, out_chunk_size))
1256 lte->rspec->raw_copy_ok = 1;
1257 list_move_tail(<e->write_streams_list,
1260 num_bytes_to_compress += lte->size;
1264 return num_bytes_to_compress;
1267 /* Copy a raw compressed resource located in another WIM file to the WIM file
1270 write_raw_copy_resource(struct wim_resource_spec *in_rspec,
1271 struct filedes *out_fd)
1273 u64 cur_read_offset;
1274 u64 end_read_offset;
1275 u8 buf[BUFFER_SIZE];
1276 size_t bytes_to_read;
1278 struct filedes *in_fd;
1279 struct wim_lookup_table_entry *lte;
1280 u64 out_offset_in_wim;
1282 DEBUG("Copying raw compressed data (size_in_wim=%"PRIu64", "
1283 "uncompressed_size=%"PRIu64")",
1284 in_rspec->size_in_wim, in_rspec->uncompressed_size);
1286 /* Copy the raw data. */
1287 cur_read_offset = in_rspec->offset_in_wim;
1288 end_read_offset = cur_read_offset + in_rspec->size_in_wim;
1290 out_offset_in_wim = out_fd->offset;
1292 if (in_rspec->is_pipable) {
1293 if (cur_read_offset < sizeof(struct pwm_stream_hdr))
1294 return WIMLIB_ERR_INVALID_PIPABLE_WIM;
1295 cur_read_offset -= sizeof(struct pwm_stream_hdr);
1296 out_offset_in_wim += sizeof(struct pwm_stream_hdr);
1298 in_fd = &in_rspec->wim->in_fd;
1299 wimlib_assert(cur_read_offset != end_read_offset);
1302 bytes_to_read = min(sizeof(buf), end_read_offset - cur_read_offset);
1304 ret = full_pread(in_fd, buf, bytes_to_read, cur_read_offset);
1308 ret = full_write(out_fd, buf, bytes_to_read);
1312 cur_read_offset += bytes_to_read;
1314 } while (cur_read_offset != end_read_offset);
1316 list_for_each_entry(lte, &in_rspec->stream_list, rspec_node) {
1317 if (lte->will_be_in_output_wim) {
1318 stream_set_out_reshdr_for_reuse(lte);
1319 if (in_rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS)
1320 lte->out_res_offset_in_wim = out_offset_in_wim;
1322 lte->out_reshdr.offset_in_wim = out_offset_in_wim;
1329 /* Copy a list of raw compressed resources located in other WIM file(s) to the
1330 * WIM file being written. */
1332 write_raw_copy_resources(struct list_head *raw_copy_streams,
1333 struct filedes *out_fd,
1334 struct write_streams_progress_data *progress_data)
1336 struct wim_lookup_table_entry *lte;
1339 list_for_each_entry(lte, raw_copy_streams, write_streams_list)
1340 lte->rspec->raw_copy_ok = 1;
1342 list_for_each_entry(lte, raw_copy_streams, write_streams_list) {
1343 if (lte->rspec->raw_copy_ok) {
1344 /* Write each packed resource only one time, no matter
1345 * how many streams reference it. */
1346 ret = write_raw_copy_resource(lte->rspec, out_fd);
1349 lte->rspec->raw_copy_ok = 0;
1351 ret = do_write_streams_progress(progress_data, lte->size,
1359 /* Wait for and write all chunks pending in the compressor. */
1361 finish_remaining_chunks(struct write_streams_ctx *ctx)
1368 if (ctx->compressor == NULL)
1371 if (ctx->chunk_buf_filled != 0) {
1372 ret = submit_chunk_for_compression(ctx, ctx->chunk_buf,
1373 ctx->chunk_buf_filled);
1378 while (ctx->compressor->get_chunk(ctx->compressor, &cdata, &csize, &usize)) {
1379 ret = write_chunk(ctx, cdata, csize, usize);
1387 remove_zero_length_streams(struct list_head *stream_list)
1389 struct wim_lookup_table_entry *lte, *tmp;
1391 list_for_each_entry_safe(lte, tmp, stream_list, write_streams_list) {
1392 wimlib_assert(lte->will_be_in_output_wim);
1393 if (lte->size == 0) {
1394 list_del(<e->write_streams_list);
1395 lte->out_reshdr.offset_in_wim = 0;
1396 lte->out_reshdr.size_in_wim = 0;
1397 lte->out_reshdr.uncompressed_size = 0;
1398 lte->out_reshdr.flags = filter_resource_flags(lte->flags);
1404 init_done_with_file_info(struct list_head *stream_list)
1406 struct wim_lookup_table_entry *lte;
1408 list_for_each_entry(lte, stream_list, write_streams_list) {
1409 if (is_file_stream(lte)) {
1410 lte->file_inode->num_remaining_streams = 0;
1411 lte->may_send_done_with_file = 1;
1413 lte->may_send_done_with_file = 0;
1417 list_for_each_entry(lte, stream_list, write_streams_list)
1418 if (lte->may_send_done_with_file)
1419 lte->file_inode->num_remaining_streams++;
1423 * Write a list of streams to the output WIM file.
1426 * The list of streams to write, specified by a list of `struct
1427 * wim_lookup_table_entry's linked by the 'write_streams_list' member.
1430 * The file descriptor, opened for writing, to which to write the streams.
1432 * @write_resource_flags
1433 * Flags to modify how the streams are written:
1435 * WRITE_RESOURCE_FLAG_RECOMPRESS:
1436 * Force compression of all resources, even if they could otherwise
1437 * be re-used by copying the raw data, due to being located in a WIM
1438 * file with compatible compression parameters.
1440 * WRITE_RESOURCE_FLAG_PIPABLE:
1441 * Write the resources in the wimlib-specific pipable format, and
1442 * furthermore do so in such a way that no seeking backwards in
1443 * @out_fd will be performed (so it may be a pipe).
1445 * WRITE_RESOURCE_FLAG_PACK_STREAMS:
1446 * Pack all the streams into a single resource rather than writing
1447 * them in separate resources. This flag is only valid if the WIM
1448 * version number has been, or will be, set to
1449 * WIM_VERSION_PACKED_STREAMS. This flag may not be combined with
1450 * WRITE_RESOURCE_FLAG_PIPABLE.
1453 * Compression format to use to write the output streams, specified as one
1454 * of the WIMLIB_COMPRESSION_TYPE_* constants.
1455 * WIMLIB_COMPRESSION_TYPE_NONE is allowed.
1458 * Chunk size to use to write the streams. It must be a valid chunk size
1459 * for the specified compression format @out_ctype, unless @out_ctype is
1460 * WIMLIB_COMPRESSION_TYPE_NONE, in which case this parameter is ignored.
1463 * Number of threads to use to compress data. If 0, a default number of
1464 * threads will be chosen. The number of threads still may be decreased
1465 * from the specified value if insufficient memory is detected.
1468 * If on-the-fly deduplication of unhashed streams is desired, this
1469 * parameter must be pointer to the lookup table for the WIMStruct on whose
1470 * behalf the streams are being written. Otherwise, this parameter can be
1474 * If on-the-fly deduplication of unhashed streams is desired, this
1475 * parameter can be a pointer to a context for stream filtering used to
1476 * detect whether the duplicate stream has been hard-filtered or not. If
1477 * no streams are hard-filtered or no streams are unhashed, this parameter
1480 * This function will write the streams in @stream_list to resources in
1481 * consecutive positions in the output WIM file, or to a single packed resource
1482 * if WRITE_RESOURCE_FLAG_PACK_STREAMS was specified in @write_resource_flags.
1483 * In both cases, the @out_reshdr of the `struct wim_lookup_table_entry' for
1484 * each stream written will be updated to specify its location, size, and flags
1485 * in the output WIM. In the packed resource case,
1486 * WIM_RESHDR_FLAG_PACKED_STREAMS will be set in the @flags field of each
1487 * @out_reshdr, and furthermore @out_res_offset_in_wim and @out_res_size_in_wim
1488 * of each @out_reshdr will be set to the offset and size, respectively, in the
1489 * output WIM of the packed resource containing the corresponding stream.
1491 * Each of the streams to write may be in any location supported by the
1492 * resource-handling code (specifically, read_stream_list()), such as the
1493 * contents of external file that has been logically added to the output WIM, or
1494 * a stream in another WIM file that has been imported, or even a stream in the
1495 * "same" WIM file of which a modified copy is being written. In the case that
1496 * a stream is already in a WIM file and uses compatible compression parameters,
1497 * by default this function will re-use the raw data instead of decompressing
1498 * it, then recompressing it; however, with WRITE_RESOURCE_FLAG_RECOMPRESS
1499 * specified in @write_resource_flags, this is not done.
1501 * As a further requirement, this function requires that the
1502 * @will_be_in_output_wim member be set to 1 on all streams in @stream_list as
1503 * well as any other streams not in @stream_list that will be in the output WIM
1504 * file, but set to 0 on any other streams in the output WIM's lookup table or
1505 * sharing a packed resource with a stream in @stream_list. Still furthermore,
1506 * if on-the-fly deduplication of streams is possible, then all streams in
1507 * @stream_list must also be linked by @lookup_table_list along with any other
1508 * streams that have @will_be_in_output_wim set.
1510 * This function handles on-the-fly deduplication of streams for which SHA1
1511 * message digests have not yet been calculated. Such streams may or may not
1512 * need to be written. If @lookup_table is non-NULL, then each stream in
1513 * @stream_list that has @unhashed set but not @unique_size set is checksummed
1514 * immediately before it would otherwise be read for writing in order to
1515 * determine if it is identical to another stream already being written or one
1516 * that would be filtered out of the output WIM using stream_filtered() with the
1517 * context @filter_ctx. Each such duplicate stream will be removed from
1518 * @stream_list, its reference count transfered to the pre-existing duplicate
1519 * stream, its memory freed, and will not be written. Alternatively, if a
1520 * stream in @stream_list is a duplicate with any stream in @lookup_table that
1521 * has not been marked for writing or would not be hard-filtered, it is freed
1522 * and the pre-existing duplicate is written instead, taking ownership of the
1523 * reference count and slot in the @lookup_table_list.
1525 * Returns 0 if every stream was either written successfully or did not need to
1526 * be written; otherwise returns a non-zero error code.
1529 write_stream_list(struct list_head *stream_list,
1530 struct filedes *out_fd,
1531 int write_resource_flags,
1534 unsigned num_threads,
1535 struct wim_lookup_table *lookup_table,
1536 struct filter_context *filter_ctx,
1537 wimlib_progress_func_t progfunc,
1541 struct write_streams_ctx ctx;
1542 struct list_head raw_copy_streams;
1544 wimlib_assert((write_resource_flags &
1545 (WRITE_RESOURCE_FLAG_PACK_STREAMS |
1546 WRITE_RESOURCE_FLAG_PIPABLE)) !=
1547 (WRITE_RESOURCE_FLAG_PACK_STREAMS |
1548 WRITE_RESOURCE_FLAG_PIPABLE));
1550 remove_zero_length_streams(stream_list);
1552 if (list_empty(stream_list)) {
1553 DEBUG("No streams to write.");
1557 /* If needed, set auxiliary information so that we can detect when the
1558 * library has finished using each external file. */
1559 if (unlikely(write_resource_flags & WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE))
1560 init_done_with_file_info(stream_list);
1562 memset(&ctx, 0, sizeof(ctx));
1564 /* Pre-sorting the streams is required for compute_stream_list_stats().
1565 * Afterwards, read_stream_list() need not sort them again. */
1566 ret = sort_stream_list_by_sequential_order(stream_list,
1567 offsetof(struct wim_lookup_table_entry,
1568 write_streams_list));
1572 ctx.out_fd = out_fd;
1573 ctx.lookup_table = lookup_table;
1574 ctx.out_ctype = out_ctype;
1575 ctx.out_chunk_size = out_chunk_size;
1576 ctx.write_resource_flags = write_resource_flags;
1577 ctx.filter_ctx = filter_ctx;
1579 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) {
1580 wimlib_assert(out_chunk_size != 0);
1581 if (out_chunk_size <= STACK_MAX) {
1582 ctx.chunk_buf = alloca(out_chunk_size);
1584 ctx.chunk_buf = MALLOC(out_chunk_size);
1585 if (ctx.chunk_buf == NULL) {
1586 ret = WIMLIB_ERR_NOMEM;
1587 goto out_destroy_context;
1591 ctx.chunk_buf_filled = 0;
1593 compute_stream_list_stats(stream_list, &ctx);
1595 ctx.progress_data.progfunc = progfunc;
1596 ctx.progress_data.progctx = progctx;
1598 ctx.num_bytes_to_compress = find_raw_copy_streams(stream_list,
1599 write_resource_flags,
1604 DEBUG("Writing stream list "
1605 "(offset = %"PRIu64", write_resource_flags=0x%08x, "
1606 "out_ctype=%d, out_chunk_size=%u, num_threads=%u, "
1607 "total_bytes=%"PRIu64", num_bytes_to_compress=%"PRIu64")",
1608 out_fd->offset, write_resource_flags,
1609 out_ctype, out_chunk_size, num_threads,
1610 ctx.progress_data.progress.write_streams.total_bytes,
1611 ctx.num_bytes_to_compress);
1613 if (ctx.num_bytes_to_compress == 0) {
1614 DEBUG("No compression needed; skipping to raw copy!");
1615 goto out_write_raw_copy_resources;
1618 /* Unless uncompressed output was required, allocate a chunk_compressor
1619 * to do compression. There are serial and parallel implementations of
1620 * the chunk_compressor interface. We default to parallel using the
1621 * specified number of threads, unless the upper bound on the number
1622 * bytes needing to be compressed is less than a heuristic value. */
1623 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) {
1625 #ifdef ENABLE_MULTITHREADED_COMPRESSION
1626 if (ctx.num_bytes_to_compress > max(2000000, out_chunk_size)) {
1627 ret = new_parallel_chunk_compressor(out_ctype,
1632 WARNING("Couldn't create parallel chunk compressor: %"TS".\n"
1633 " Falling back to single-threaded compression.",
1634 wimlib_get_error_string(ret));
1639 if (ctx.compressor == NULL) {
1640 ret = new_serial_chunk_compressor(out_ctype, out_chunk_size,
1643 goto out_destroy_context;
1648 ctx.progress_data.progress.write_streams.num_threads = ctx.compressor->num_threads;
1650 ctx.progress_data.progress.write_streams.num_threads = 1;
1652 DEBUG("Actually using %u threads",
1653 ctx.progress_data.progress.write_streams.num_threads);
1655 INIT_LIST_HEAD(&ctx.pending_streams);
1656 INIT_LIST_HEAD(&ctx.pack_streams);
1658 ret = call_progress(ctx.progress_data.progfunc,
1659 WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
1660 &ctx.progress_data.progress,
1661 ctx.progress_data.progctx);
1663 goto out_destroy_context;
1665 if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
1666 ret = begin_write_resource(&ctx, ctx.num_bytes_to_compress);
1668 goto out_destroy_context;
1671 /* Read the list of streams needing to be compressed, using the
1672 * specified callbacks to execute processing of the data. */
1674 struct read_stream_list_callbacks cbs = {
1675 .begin_stream = write_stream_begin_read,
1676 .begin_stream_ctx = &ctx,
1677 .consume_chunk = write_stream_process_chunk,
1678 .consume_chunk_ctx = &ctx,
1679 .end_stream = write_stream_end_read,
1680 .end_stream_ctx = &ctx,
1683 ret = read_stream_list(stream_list,
1684 offsetof(struct wim_lookup_table_entry, write_streams_list),
1686 STREAM_LIST_ALREADY_SORTED |
1687 VERIFY_STREAM_HASHES |
1688 COMPUTE_MISSING_STREAM_HASHES);
1691 goto out_destroy_context;
1693 ret = finish_remaining_chunks(&ctx);
1695 goto out_destroy_context;
1697 if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
1698 struct wim_reshdr reshdr;
1699 struct wim_lookup_table_entry *lte;
1702 ret = end_write_resource(&ctx, &reshdr);
1704 goto out_destroy_context;
1706 DEBUG("Ending packed resource: %lu %lu %lu.",
1707 reshdr.offset_in_wim,
1709 reshdr.uncompressed_size);
1712 list_for_each_entry(lte, &ctx.pack_streams, write_streams_list) {
1713 lte->out_reshdr.size_in_wim = lte->size;
1714 lte->out_reshdr.flags = filter_resource_flags(lte->flags);
1715 lte->out_reshdr.flags |= WIM_RESHDR_FLAG_PACKED_STREAMS;
1716 lte->out_reshdr.uncompressed_size = 0;
1717 lte->out_reshdr.offset_in_wim = offset_in_res;
1718 lte->out_res_offset_in_wim = reshdr.offset_in_wim;
1719 lte->out_res_size_in_wim = reshdr.size_in_wim;
1720 lte->out_res_uncompressed_size = reshdr.uncompressed_size;
1721 offset_in_res += lte->size;
1723 wimlib_assert(offset_in_res == reshdr.uncompressed_size);
1726 out_write_raw_copy_resources:
1727 /* Copy any compressed resources for which the raw data can be reused
1728 * without decompression. */
1729 ret = write_raw_copy_resources(&raw_copy_streams, ctx.out_fd,
1730 &ctx.progress_data);
1732 out_destroy_context:
1733 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && out_chunk_size > STACK_MAX)
1734 FREE(ctx.chunk_buf);
1735 FREE(ctx.chunk_csizes);
1737 ctx.compressor->destroy(ctx.compressor);
1738 DEBUG("Done (ret=%d)", ret);
1743 is_stream_packed(struct wim_lookup_table_entry *lte, void *_ignore)
1745 return lte_is_partial(lte);
1749 wim_has_packed_streams(WIMStruct *wim)
1751 return for_lookup_table_entry(wim->lookup_table, is_stream_packed, NULL);
1755 wim_write_stream_list(WIMStruct *wim,
1756 struct list_head *stream_list,
1758 unsigned num_threads,
1759 struct filter_context *filter_ctx)
1763 int write_resource_flags;
1765 write_resource_flags = write_flags_to_resource_flags(write_flags);
1767 /* wimlib v1.7.0: pack streams by default if the WIM version has been
1768 * set to WIM_VERSION_PACKED_STREAMS and at least one stream in the
1769 * WIM's lookup table is located in a packed resource (may be the same
1770 * WIM, or a different one in the case of export). */
1771 if (wim->hdr.wim_version == WIM_VERSION_PACKED_STREAMS &&
1772 wim_has_packed_streams(wim))
1774 write_resource_flags |= WRITE_RESOURCE_FLAG_PACK_STREAMS;
1777 if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
1778 out_chunk_size = wim->out_pack_chunk_size;
1779 out_ctype = wim->out_pack_compression_type;
1781 out_chunk_size = wim->out_chunk_size;
1782 out_ctype = wim->out_compression_type;
1785 return write_stream_list(stream_list,
1787 write_resource_flags,
1798 write_wim_resource(struct wim_lookup_table_entry *lte,
1799 struct filedes *out_fd,
1802 int write_resource_flags)
1804 LIST_HEAD(stream_list);
1805 list_add(<e->write_streams_list, &stream_list);
1806 lte->will_be_in_output_wim = 1;
1807 return write_stream_list(&stream_list,
1809 write_resource_flags & ~WRITE_RESOURCE_FLAG_PACK_STREAMS,
1820 write_wim_resource_from_buffer(const void *buf, size_t buf_size,
1821 int reshdr_flags, struct filedes *out_fd,
1824 struct wim_reshdr *out_reshdr,
1826 int write_resource_flags)
1829 struct wim_lookup_table_entry *lte;
1831 /* Set up a temporary lookup table entry to provide to
1832 * write_wim_resource(). */
1834 lte = new_lookup_table_entry();
1836 return WIMLIB_ERR_NOMEM;
1838 lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER;
1839 lte->attached_buffer = (void*)buf;
1840 lte->size = buf_size;
1841 lte->flags = reshdr_flags;
1843 if (write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
1844 sha1_buffer(buf, buf_size, lte->hash);
1850 ret = write_wim_resource(lte, out_fd, out_ctype, out_chunk_size,
1851 write_resource_flags);
1855 copy_reshdr(out_reshdr, <e->out_reshdr);
1858 copy_hash(hash, lte->hash);
1861 lte->resource_location = RESOURCE_NONEXISTENT;
1862 free_lookup_table_entry(lte);
1866 struct stream_size_table {
1867 struct hlist_head *array;
1873 init_stream_size_table(struct stream_size_table *tab, size_t capacity)
1875 tab->array = CALLOC(capacity, sizeof(tab->array[0]));
1876 if (tab->array == NULL)
1877 return WIMLIB_ERR_NOMEM;
1878 tab->num_entries = 0;
1879 tab->capacity = capacity;
1884 destroy_stream_size_table(struct stream_size_table *tab)
1890 stream_size_table_insert(struct wim_lookup_table_entry *lte, void *_tab)
1892 struct stream_size_table *tab = _tab;
1894 struct wim_lookup_table_entry *same_size_lte;
1895 struct hlist_node *tmp;
1897 pos = hash_u64(lte->size) % tab->capacity;
1898 lte->unique_size = 1;
1899 hlist_for_each_entry(same_size_lte, tmp, &tab->array[pos], hash_list_2) {
1900 if (same_size_lte->size == lte->size) {
1901 lte->unique_size = 0;
1902 same_size_lte->unique_size = 0;
1907 hlist_add_head(<e->hash_list_2, &tab->array[pos]);
1912 struct find_streams_ctx {
1915 struct list_head stream_list;
1916 struct stream_size_table stream_size_tab;
1920 reference_stream_for_write(struct wim_lookup_table_entry *lte,
1921 struct list_head *stream_list, u32 nref)
1923 if (!lte->will_be_in_output_wim) {
1924 lte->out_refcnt = 0;
1925 list_add_tail(<e->write_streams_list, stream_list);
1926 lte->will_be_in_output_wim = 1;
1928 lte->out_refcnt += nref;
1932 fully_reference_stream_for_write(struct wim_lookup_table_entry *lte,
1935 struct list_head *stream_list = _stream_list;
1936 lte->will_be_in_output_wim = 0;
1937 reference_stream_for_write(lte, stream_list, lte->refcnt);
1942 inode_find_streams_to_reference(const struct wim_inode *inode,
1943 const struct wim_lookup_table *table,
1944 struct list_head *stream_list)
1946 struct wim_lookup_table_entry *lte;
1949 wimlib_assert(inode->i_nlink > 0);
1951 for (i = 0; i <= inode->i_num_ads; i++) {
1952 lte = inode_stream_lte(inode, i, table);
1954 reference_stream_for_write(lte, stream_list,
1956 else if (!is_zero_hash(inode_stream_hash(inode, i)))
1957 return WIMLIB_ERR_RESOURCE_NOT_FOUND;
1963 do_stream_set_not_in_output_wim(struct wim_lookup_table_entry *lte, void *_ignore)
1965 lte->will_be_in_output_wim = 0;
1970 image_find_streams_to_reference(WIMStruct *wim)
1972 struct wim_image_metadata *imd;
1973 struct wim_inode *inode;
1974 struct wim_lookup_table_entry *lte;
1975 struct list_head *stream_list;
1978 imd = wim_get_current_image_metadata(wim);
1980 image_for_each_unhashed_stream(lte, imd)
1981 lte->will_be_in_output_wim = 0;
1983 stream_list = wim->private;
1984 image_for_each_inode(inode, imd) {
1985 ret = inode_find_streams_to_reference(inode,
1995 prepare_unfiltered_list_of_streams_in_output_wim(WIMStruct *wim,
1998 struct list_head *stream_list_ret)
2002 INIT_LIST_HEAD(stream_list_ret);
2004 if (streams_ok && (image == WIMLIB_ALL_IMAGES ||
2005 (image == 1 && wim->hdr.image_count == 1)))
2007 /* Fast case: Assume that all streams are being written and
2008 * that the reference counts are correct. */
2009 struct wim_lookup_table_entry *lte;
2010 struct wim_image_metadata *imd;
2013 for_lookup_table_entry(wim->lookup_table,
2014 fully_reference_stream_for_write,
2017 for (i = 0; i < wim->hdr.image_count; i++) {
2018 imd = wim->image_metadata[i];
2019 image_for_each_unhashed_stream(lte, imd)
2020 fully_reference_stream_for_write(lte, stream_list_ret);
2023 /* Slow case: Walk through the images being written and
2024 * determine the streams referenced. */
2025 for_lookup_table_entry(wim->lookup_table,
2026 do_stream_set_not_in_output_wim, NULL);
2027 wim->private = stream_list_ret;
2028 ret = for_image(wim, image, image_find_streams_to_reference);
2036 struct insert_other_if_hard_filtered_ctx {
2037 struct stream_size_table *tab;
2038 struct filter_context *filter_ctx;
2042 insert_other_if_hard_filtered(struct wim_lookup_table_entry *lte, void *_ctx)
2044 struct insert_other_if_hard_filtered_ctx *ctx = _ctx;
2046 if (!lte->will_be_in_output_wim &&
2047 stream_hard_filtered(lte, ctx->filter_ctx))
2048 stream_size_table_insert(lte, ctx->tab);
2053 determine_stream_size_uniquity(struct list_head *stream_list,
2054 struct wim_lookup_table *lt,
2055 struct filter_context *filter_ctx)
2058 struct stream_size_table tab;
2059 struct wim_lookup_table_entry *lte;
2061 ret = init_stream_size_table(&tab, 9001);
2065 if (may_hard_filter_streams(filter_ctx)) {
2066 struct insert_other_if_hard_filtered_ctx ctx = {
2068 .filter_ctx = filter_ctx,
2070 for_lookup_table_entry(lt, insert_other_if_hard_filtered, &ctx);
2073 list_for_each_entry(lte, stream_list, write_streams_list)
2074 stream_size_table_insert(lte, &tab);
2076 destroy_stream_size_table(&tab);
2081 filter_stream_list_for_write(struct list_head *stream_list,
2082 struct filter_context *filter_ctx)
2084 struct wim_lookup_table_entry *lte, *tmp;
2086 list_for_each_entry_safe(lte, tmp,
2087 stream_list, write_streams_list)
2089 int status = stream_filtered(lte, filter_ctx);
2096 /* Soft filtered. */
2098 /* Hard filtered. */
2099 lte->will_be_in_output_wim = 0;
2100 list_del(<e->lookup_table_list);
2102 list_del(<e->write_streams_list);
2108 * prepare_stream_list_for_write() -
2110 * Prepare the list of streams to write for writing a WIM containing the
2111 * specified image(s) with the specified write flags.
2114 * The WIMStruct on whose behalf the write is occurring.
2117 * Image(s) from the WIM to write; may be WIMLIB_ALL_IMAGES.
2120 * WIMLIB_WRITE_FLAG_* flags for the write operation:
2122 * STREAMS_OK: For writes of all images, assume that all streams in the
2123 * lookup table of @wim and the per-image lists of unhashed streams should
2124 * be taken as-is, and image metadata should not be searched for
2125 * references. This does not exclude filtering with OVERWRITE and
2126 * SKIP_EXTERNAL_WIMS, below.
2128 * OVERWRITE: Streams already present in @wim shall not be returned in
2131 * SKIP_EXTERNAL_WIMS: Streams already present in a WIM file, but not
2132 * @wim, shall be returned in neither @stream_list_ret nor
2133 * @lookup_table_list_ret.
2136 * List of streams, linked by write_streams_list, that need to be written
2137 * will be returned here.
2139 * Note that this function assumes that unhashed streams will be written;
2140 * it does not take into account that they may become duplicates when
2143 * @lookup_table_list_ret
2144 * List of streams, linked by lookup_table_list, that need to be included
2145 * in the WIM's lookup table will be returned here. This will be a
2146 * superset of the streams in @stream_list_ret.
2148 * This list will be a proper superset of @stream_list_ret if and only if
2149 * WIMLIB_WRITE_FLAG_OVERWRITE was specified in @write_flags and some of
2150 * the streams that would otherwise need to be written were already located
2153 * All streams in this list will have @out_refcnt set to the number of
2154 * references to the stream in the output WIM. If
2155 * WIMLIB_WRITE_FLAG_STREAMS_OK was specified in @write_flags, @out_refcnt
2156 * may be as low as 0.
2159 * A context for queries of stream filter status with stream_filtered() is
2160 * returned in this location.
2162 * In addition, @will_be_in_output_wim will be set to 1 in all stream entries
2163 * inserted into @lookup_table_list_ret and to 0 in all stream entries in the
2164 * lookup table of @wim not inserted into @lookup_table_list_ret.
2166 * Still furthermore, @unique_size will be set to 1 on all stream entries in
2167 * @stream_list_ret that have unique size among all stream entries in
2168 * @stream_list_ret and among all stream entries in the lookup table of @wim
2169 * that are ineligible for being written due to filtering.
2171 * Returns 0 on success; nonzero on read error, memory allocation error, or
2175 prepare_stream_list_for_write(WIMStruct *wim, int image,
2177 struct list_head *stream_list_ret,
2178 struct list_head *lookup_table_list_ret,
2179 struct filter_context *filter_ctx_ret)
2182 struct wim_lookup_table_entry *lte;
2184 filter_ctx_ret->write_flags = write_flags;
2185 filter_ctx_ret->wim = wim;
2187 ret = prepare_unfiltered_list_of_streams_in_output_wim(
2190 write_flags & WIMLIB_WRITE_FLAG_STREAMS_OK,
2195 INIT_LIST_HEAD(lookup_table_list_ret);
2196 list_for_each_entry(lte, stream_list_ret, write_streams_list)
2197 list_add_tail(<e->lookup_table_list, lookup_table_list_ret);
2199 ret = determine_stream_size_uniquity(stream_list_ret, wim->lookup_table,
2204 if (may_filter_streams(filter_ctx_ret))
2205 filter_stream_list_for_write(stream_list_ret, filter_ctx_ret);
2211 write_wim_streams(WIMStruct *wim, int image, int write_flags,
2212 unsigned num_threads,
2213 struct list_head *stream_list_override,
2214 struct list_head *lookup_table_list_ret)
2217 struct list_head _stream_list;
2218 struct list_head *stream_list;
2219 struct wim_lookup_table_entry *lte;
2220 struct filter_context _filter_ctx;
2221 struct filter_context *filter_ctx;
2223 if (stream_list_override == NULL) {
2224 /* Normal case: prepare stream list from image(s) being written.
2226 stream_list = &_stream_list;
2227 filter_ctx = &_filter_ctx;
2228 ret = prepare_stream_list_for_write(wim, image, write_flags,
2230 lookup_table_list_ret,
2235 /* Currently only as a result of wimlib_split() being called:
2236 * use stream list already explicitly provided. Use existing
2237 * reference counts. */
2238 stream_list = stream_list_override;
2240 INIT_LIST_HEAD(lookup_table_list_ret);
2241 list_for_each_entry(lte, stream_list, write_streams_list) {
2242 lte->out_refcnt = lte->refcnt;
2243 lte->will_be_in_output_wim = 1;
2244 lte->unique_size = 0;
2245 list_add_tail(<e->lookup_table_list, lookup_table_list_ret);
2249 return wim_write_stream_list(wim,
2257 write_wim_metadata_resources(WIMStruct *wim, int image, int write_flags)
2262 int write_resource_flags;
2264 if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA) {
2265 DEBUG("Not writing any metadata resources.");
2269 write_resource_flags = write_flags_to_resource_flags(write_flags);
2271 write_resource_flags &= ~WRITE_RESOURCE_FLAG_PACK_STREAMS;
2273 DEBUG("Writing metadata resources (offset=%"PRIu64")",
2274 wim->out_fd.offset);
2276 ret = call_progress(wim->progfunc,
2277 WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN,
2278 NULL, wim->progctx);
2282 if (image == WIMLIB_ALL_IMAGES) {
2284 end_image = wim->hdr.image_count;
2286 start_image = image;
2290 for (int i = start_image; i <= end_image; i++) {
2291 struct wim_image_metadata *imd;
2293 imd = wim->image_metadata[i - 1];
2294 /* Build a new metadata resource only if image was modified from
2295 * the original (or was newly added). Otherwise just copy the
2297 if (imd->modified) {
2298 DEBUG("Image %u was modified; building and writing new "
2299 "metadata resource", i);
2300 ret = write_metadata_resource(wim, i,
2301 write_resource_flags);
2302 } else if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) {
2303 DEBUG("Image %u was not modified; re-using existing "
2304 "metadata resource.", i);
2305 stream_set_out_reshdr_for_reuse(imd->metadata_lte);
2308 DEBUG("Image %u was not modified; copying existing "
2309 "metadata resource.", i);
2310 ret = write_wim_resource(imd->metadata_lte,
2312 wim->out_compression_type,
2313 wim->out_chunk_size,
2314 write_resource_flags);
2320 return call_progress(wim->progfunc,
2321 WIMLIB_PROGRESS_MSG_WRITE_METADATA_END,
2322 NULL, wim->progctx);
2326 open_wim_writable(WIMStruct *wim, const tchar *path, int open_flags)
2329 DEBUG("Opening \"%"TS"\" for writing.", path);
2331 raw_fd = topen(path, open_flags | O_BINARY, 0644);
2333 ERROR_WITH_ERRNO("Failed to open \"%"TS"\" for writing", path);
2334 return WIMLIB_ERR_OPEN;
2336 filedes_init(&wim->out_fd, raw_fd);
2341 close_wim_writable(WIMStruct *wim, int write_flags)
2345 if (!(write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)) {
2346 DEBUG("Closing WIM file.");
2347 if (filedes_valid(&wim->out_fd))
2348 if (filedes_close(&wim->out_fd))
2349 ret = WIMLIB_ERR_WRITE;
2351 filedes_invalidate(&wim->out_fd);
2356 cmp_streams_by_out_rspec(const void *p1, const void *p2)
2358 const struct wim_lookup_table_entry *lte1, *lte2;
2360 lte1 = *(const struct wim_lookup_table_entry**)p1;
2361 lte2 = *(const struct wim_lookup_table_entry**)p2;
2363 if (lte1->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
2364 if (lte2->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
2365 if (lte1->out_res_offset_in_wim != lte2->out_res_offset_in_wim)
2366 return cmp_u64(lte1->out_res_offset_in_wim,
2367 lte2->out_res_offset_in_wim);
2372 if (lte2->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS)
2375 return cmp_u64(lte1->out_reshdr.offset_in_wim,
2376 lte2->out_reshdr.offset_in_wim);
2380 write_wim_lookup_table(WIMStruct *wim, int image, int write_flags,
2381 struct wim_reshdr *out_reshdr,
2382 struct list_head *lookup_table_list)
2386 /* Set output resource metadata for streams already present in WIM. */
2387 if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) {
2388 struct wim_lookup_table_entry *lte;
2389 list_for_each_entry(lte, lookup_table_list, lookup_table_list)
2391 if (lte->resource_location == RESOURCE_IN_WIM &&
2392 lte->rspec->wim == wim)
2394 stream_set_out_reshdr_for_reuse(lte);
2399 ret = sort_stream_list(lookup_table_list,
2400 offsetof(struct wim_lookup_table_entry, lookup_table_list),
2401 cmp_streams_by_out_rspec);
2405 /* Add entries for metadata resources. */
2406 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)) {
2410 if (image == WIMLIB_ALL_IMAGES) {
2412 end_image = wim->hdr.image_count;
2414 start_image = image;
2418 /* Push metadata resource lookup table entries onto the front of
2419 * the list in reverse order, so that they're written in order.
2421 for (int i = end_image; i >= start_image; i--) {
2422 struct wim_lookup_table_entry *metadata_lte;
2424 metadata_lte = wim->image_metadata[i - 1]->metadata_lte;
2425 wimlib_assert(metadata_lte->out_reshdr.flags & WIM_RESHDR_FLAG_METADATA);
2426 metadata_lte->out_refcnt = 1;
2427 list_add(&metadata_lte->lookup_table_list, lookup_table_list);
2431 return write_wim_lookup_table_from_stream_list(lookup_table_list,
2433 wim->hdr.part_number,
2435 write_flags_to_resource_flags(write_flags));
2441 * Finish writing a WIM file: write the lookup table, xml data, and integrity
2442 * table, then overwrite the WIM header. By default, closes the WIM file
2443 * descriptor (@wim->out_fd) if successful.
2445 * write_flags is a bitwise OR of the following:
2447 * (public) WIMLIB_WRITE_FLAG_CHECK_INTEGRITY:
2448 * Include an integrity table.
2450 * (public) WIMLIB_WRITE_FLAG_FSYNC:
2451 * fsync() the output file before closing it.
2453 * (public) WIMLIB_WRITE_FLAG_PIPABLE:
2454 * Writing a pipable WIM, possibly to a pipe; include pipable WIM
2455 * stream headers before the lookup table and XML data, and also
2456 * write the WIM header at the end instead of seeking to the
2457 * beginning. Can't be combined with
2458 * WIMLIB_WRITE_FLAG_CHECK_INTEGRITY.
2460 * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE:
2461 * Don't write the lookup table.
2463 * (private) WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML:
2464 * After writing the XML data but before writing the integrity
2465 * table, write a temporary WIM header and flush the stream so that
2466 * the WIM is less likely to become corrupted upon abrupt program
2468 * (private) WIMLIB_WRITE_FLAG_HEADER_AT_END:
2469 * Instead of overwriting the WIM header at the beginning of the
2470 * file, simply append it to the end of the file. (Used when
2472 * (private) WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR:
2473 * Do not close the file descriptor @wim->out_fd on either success
2475 * (private) WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES:
2476 * Use the existing <TOTALBYTES> stored in the in-memory XML
2477 * information, rather than setting it to the offset of the XML
2478 * data being written.
2479 * (private) WIMLIB_WRITE_FLAG_OVERWRITE
2480 * The existing WIM file is being updated in-place. The entries
2481 * from its integrity table may be re-used.
2484 finish_write(WIMStruct *wim, int image, int write_flags,
2485 struct list_head *lookup_table_list)
2489 int write_resource_flags;
2490 off_t old_lookup_table_end = 0;
2491 off_t new_lookup_table_end;
2493 struct integrity_table *old_integrity_table = NULL;
2495 DEBUG("image=%d, write_flags=%08x", image, write_flags);
2497 write_resource_flags = write_flags_to_resource_flags(write_flags);
2499 /* In the WIM header, there is room for the resource entry for a
2500 * metadata resource labeled as the "boot metadata". This entry should
2501 * be zeroed out if there is no bootable image (boot_idx 0). Otherwise,
2502 * it should be a copy of the resource entry for the image that is
2503 * marked as bootable. This is not well documented... */
2504 if (wim->hdr.boot_idx == 0) {
2505 zero_reshdr(&wim->hdr.boot_metadata_reshdr);
2507 copy_reshdr(&wim->hdr.boot_metadata_reshdr,
2508 &wim->image_metadata[
2509 wim->hdr.boot_idx - 1]->metadata_lte->out_reshdr);
2512 /* If overwriting the WIM file containing an integrity table in-place,
2513 * we'd like to re-use the information in the old integrity table
2514 * instead of recalculating it. But we might overwrite the old
2515 * integrity table when we expand the XML data. Read it into memory
2517 if ((write_flags & (WIMLIB_WRITE_FLAG_OVERWRITE |
2518 WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)) ==
2519 (WIMLIB_WRITE_FLAG_OVERWRITE |
2520 WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)
2521 && wim_has_integrity_table(wim))
2523 old_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim +
2524 wim->hdr.lookup_table_reshdr.size_in_wim;
2525 (void)read_integrity_table(wim,
2526 old_lookup_table_end - WIM_HEADER_DISK_SIZE,
2527 &old_integrity_table);
2528 /* If we couldn't read the old integrity table, we can still
2529 * re-calculate the full integrity table ourselves. Hence the
2530 * ignoring of the return value. */
2533 /* Write lookup table. */
2534 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
2535 ret = write_wim_lookup_table(wim, image, write_flags,
2536 &wim->hdr.lookup_table_reshdr,
2539 free_integrity_table(old_integrity_table);
2544 /* Write XML data. */
2545 xml_totalbytes = wim->out_fd.offset;
2546 if (write_flags & WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES)
2547 xml_totalbytes = WIM_TOTALBYTES_USE_EXISTING;
2548 ret = write_wim_xml_data(wim, image, xml_totalbytes,
2549 &wim->hdr.xml_data_reshdr,
2550 write_resource_flags);
2552 free_integrity_table(old_integrity_table);
2556 /* Write integrity table (optional). */
2557 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
2558 if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) {
2559 struct wim_header checkpoint_hdr;
2560 memcpy(&checkpoint_hdr, &wim->hdr, sizeof(struct wim_header));
2561 zero_reshdr(&checkpoint_hdr.integrity_table_reshdr);
2562 checkpoint_hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2563 ret = write_wim_header_at_offset(&checkpoint_hdr,
2566 free_integrity_table(old_integrity_table);
2571 new_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim +
2572 wim->hdr.lookup_table_reshdr.size_in_wim;
2574 ret = write_integrity_table(wim,
2575 new_lookup_table_end,
2576 old_lookup_table_end,
2577 old_integrity_table);
2578 free_integrity_table(old_integrity_table);
2582 /* No integrity table. */
2583 zero_reshdr(&wim->hdr.integrity_table_reshdr);
2586 /* Now that all information in the WIM header has been determined, the
2587 * preliminary header written earlier can be overwritten, the header of
2588 * the existing WIM file can be overwritten, or the final header can be
2589 * written to the end of the pipable WIM. */
2590 wim->hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2592 if (write_flags & WIMLIB_WRITE_FLAG_HEADER_AT_END)
2593 hdr_offset = wim->out_fd.offset;
2594 DEBUG("Writing new header @ %"PRIu64".", hdr_offset);
2595 ret = write_wim_header_at_offset(&wim->hdr, &wim->out_fd, hdr_offset);
2599 /* Possibly sync file data to disk before closing. On POSIX systems, it
2600 * is necessary to do this before using rename() to overwrite an
2601 * existing file with a new file. Otherwise, data loss would occur if
2602 * the system is abruptly terminated when the metadata for the rename
2603 * operation has been written to disk, but the new file data has not.
2605 if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) {
2606 DEBUG("Syncing WIM file.");
2607 if (fsync(wim->out_fd.fd)) {
2608 ERROR_WITH_ERRNO("Error syncing data to WIM file");
2609 return WIMLIB_ERR_WRITE;
2613 if (close_wim_writable(wim, write_flags)) {
2614 ERROR_WITH_ERRNO("Failed to close the output WIM file");
2615 return WIMLIB_ERR_WRITE;
2621 #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK)
2623 /* Set advisory lock on WIM file (if not already done so) */
2625 lock_wim_for_append(WIMStruct *wim)
2627 if (wim->locked_for_append)
2629 if (!flock(wim->in_fd.fd, LOCK_EX | LOCK_NB)) {
2630 wim->locked_for_append = 1;
2633 if (errno != EWOULDBLOCK)
2635 return WIMLIB_ERR_ALREADY_LOCKED;
2638 /* Remove advisory lock on WIM file (if present) */
2640 unlock_wim_for_append(WIMStruct *wim)
2642 if (wim->locked_for_append) {
2643 flock(wim->in_fd.fd, LOCK_UN);
2644 wim->locked_for_append = 0;
2650 * write_pipable_wim():
2652 * Perform the intermediate stages of creating a "pipable" WIM (i.e. a WIM
2653 * capable of being applied from a pipe).
2655 * Pipable WIMs are a wimlib-specific modification of the WIM format such that
2656 * images can be applied from them sequentially when the file data is sent over
2657 * a pipe. In addition, a pipable WIM can be written sequentially to a pipe.
2658 * The modifications made to the WIM format for pipable WIMs are:
2660 * - Magic characters in header are "WLPWM\0\0\0" (wimlib pipable WIM) instead
2661 * of "MSWIM\0\0\0". This lets wimlib know that the WIM is pipable and also
2662 * stops other software from trying to read the file as a normal WIM.
2664 * - The header at the beginning of the file does not contain all the normal
2665 * information; in particular it will have all 0's for the lookup table and
2666 * XML data resource entries. This is because this information cannot be
2667 * determined until the lookup table and XML data have been written.
2668 * Consequently, wimlib will write the full header at the very end of the
2669 * file. The header at the end, however, is only used when reading the WIM
2670 * from a seekable file (not a pipe).
2672 * - An extra copy of the XML data is placed directly after the header. This
2673 * allows image names and sizes to be determined at an appropriate time when
2674 * reading the WIM from a pipe. This copy of the XML data is ignored if the
2675 * WIM is read from a seekable file (not a pipe).
2677 * - The format of resources, or streams, has been modified to allow them to be
2678 * used before the "lookup table" has been read. Each stream is prefixed with
2679 * a `struct pwm_stream_hdr' that is basically an abbreviated form of `struct
2680 * wim_lookup_table_entry_disk' that only contains the SHA1 message digest,
2681 * uncompressed stream size, and flags that indicate whether the stream is
2682 * compressed. The data of uncompressed streams then follows literally, while
2683 * the data of compressed streams follows in a modified format. Compressed
2684 * streams do not begin with a chunk table, since the chunk table cannot be
2685 * written until all chunks have been compressed. Instead, each compressed
2686 * chunk is prefixed by a `struct pwm_chunk_hdr' that gives its size.
2687 * Furthermore, the chunk table is written at the end of the resource instead
2688 * of the start. Note: chunk offsets are given in the chunk table as if the
2689 * `struct pwm_chunk_hdr's were not present; also, the chunk table is only
2690 * used if the WIM is being read from a seekable file (not a pipe).
2692 * - Metadata resources always come before other file resources (streams).
2693 * (This does not by itself constitute an incompatibility with normal WIMs,
2694 * since this is valid in normal WIMs.)
2696 * - At least up to the end of the file resources, all components must be packed
2697 * as tightly as possible; there cannot be any "holes" in the WIM. (This does
2698 * not by itself consititute an incompatibility with normal WIMs, since this
2699 * is valid in normal WIMs.)
2701 * Note: the lookup table, XML data, and header at the end are not used when
2702 * applying from a pipe. They exist to support functionality such as image
2703 * application and export when the WIM is *not* read from a pipe.
2705 * Layout of pipable WIM:
2707 * ---------+----------+--------------------+----------------+--------------+-----------+--------+
2708 * | Header | XML data | Metadata resources | File resources | Lookup table | XML data | Header |
2709 * ---------+----------+--------------------+----------------+--------------+-----------+--------+
2711 * Layout of normal WIM:
2713 * +--------+-----------------------------+-------------------------+
2714 * | Header | File and metadata resources | Lookup table | XML data |
2715 * +--------+-----------------------------+-------------------------+
2717 * An optional integrity table can follow the final XML data in both normal and
2718 * pipable WIMs. However, due to implementation details, wimlib currently can
2719 * only include an integrity table in a pipable WIM when writing it to a
2720 * seekable file (not a pipe).
2722 * Do note that since pipable WIMs are not supported by Microsoft's software,
2723 * wimlib does not create them unless explicitly requested (with
2724 * WIMLIB_WRITE_FLAG_PIPABLE) and as stated above they use different magic
2725 * characters to identify the file.
2728 write_pipable_wim(WIMStruct *wim, int image, int write_flags,
2729 unsigned num_threads,
2730 struct list_head *stream_list_override,
2731 struct list_head *lookup_table_list_ret)
2734 struct wim_reshdr xml_reshdr;
2736 WARNING("Creating a pipable WIM, which will "
2738 " with Microsoft's software (wimgapi/imagex/Dism).");
2740 /* At this point, the header at the beginning of the file has already
2743 /* For efficiency, when wimlib adds an image to the WIM with
2744 * wimlib_add_image(), the SHA1 message digests of files is not
2745 * calculated; instead, they are calculated while the files are being
2746 * written. However, this does not work when writing a pipable WIM,
2747 * since when writing a stream to a pipable WIM, its SHA1 message digest
2748 * needs to be known before the stream data is written. Therefore,
2749 * before getting much farther, we need to pre-calculate the SHA1
2750 * message digests of all streams that will be written. */
2751 ret = wim_checksum_unhashed_streams(wim);
2755 /* Write extra copy of the XML data. */
2756 ret = write_wim_xml_data(wim, image, WIM_TOTALBYTES_OMIT,
2758 WRITE_RESOURCE_FLAG_PIPABLE);
2762 /* Write metadata resources for the image(s) being included in the
2764 ret = write_wim_metadata_resources(wim, image, write_flags);
2768 /* Write streams needed for the image(s) being included in the output
2769 * WIM, or streams needed for the split WIM part. */
2770 return write_wim_streams(wim, image, write_flags, num_threads,
2771 stream_list_override, lookup_table_list_ret);
2773 /* The lookup table, XML data, and header at end are handled by
2774 * finish_write(). */
2777 /* Write a standalone WIM or split WIM (SWM) part to a new file or to a file
2780 write_wim_part(WIMStruct *wim,
2781 const void *path_or_fd,
2784 unsigned num_threads,
2785 unsigned part_number,
2786 unsigned total_parts,
2787 struct list_head *stream_list_override,
2791 struct wim_header hdr_save;
2792 struct list_head lookup_table_list;
2794 if (total_parts == 1)
2795 DEBUG("Writing standalone WIM.");
2797 DEBUG("Writing split WIM part %u/%u", part_number, total_parts);
2798 if (image == WIMLIB_ALL_IMAGES)
2799 DEBUG("Including all images.");
2801 DEBUG("Including image %d only.", image);
2802 if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)
2803 DEBUG("File descriptor: %d", *(const int*)path_or_fd);
2805 DEBUG("Path: \"%"TS"\"", (const tchar*)path_or_fd);
2806 DEBUG("Write flags: 0x%08x", write_flags);
2808 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)
2809 DEBUG("\tCHECK_INTEGRITY");
2811 if (write_flags & WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)
2812 DEBUG("\tNO_CHECK_INTEGRITY");
2814 if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
2817 if (write_flags & WIMLIB_WRITE_FLAG_NOT_PIPABLE)
2818 DEBUG("\tNOT_PIPABLE");
2820 if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
2821 DEBUG("\tRECOMPRESS");
2823 if (write_flags & WIMLIB_WRITE_FLAG_FSYNC)
2826 if (write_flags & WIMLIB_WRITE_FLAG_REBUILD)
2829 if (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE)
2830 DEBUG("\tSOFT_DELETE");
2832 if (write_flags & WIMLIB_WRITE_FLAG_IGNORE_READONLY_FLAG)
2833 DEBUG("\tIGNORE_READONLY_FLAG");
2835 if (write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS)
2836 DEBUG("\tSKIP_EXTERNAL_WIMS");
2838 if (write_flags & WIMLIB_WRITE_FLAG_STREAMS_OK)
2839 DEBUG("\tSTREAMS_OK");
2841 if (write_flags & WIMLIB_WRITE_FLAG_RETAIN_GUID)
2842 DEBUG("\tRETAIN_GUID");
2844 if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS)
2845 DEBUG("\tPACK_STREAMS");
2847 if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)
2848 DEBUG("\tFILE_DESCRIPTOR");
2850 if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)
2851 DEBUG("\tNO_METADATA");
2853 if (write_flags & WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES)
2854 DEBUG("\tUSE_EXISTING_TOTALBYTES");
2856 if (num_threads == 0)
2857 DEBUG("Number of threads: autodetect");
2859 DEBUG("Number of threads: %u", num_threads);
2860 DEBUG("Progress function: %s", (wim->progfunc ? "yes" : "no"));
2861 DEBUG("Stream list: %s", (stream_list_override ? "specified" : "autodetect"));
2862 DEBUG("GUID: %s", (write_flags &
2863 WIMLIB_WRITE_FLAG_RETAIN_GUID) ? "retain"
2864 : guid ? "explicit" : "generate new");
2866 /* Internally, this is always called with a valid part number and total
2868 wimlib_assert(total_parts >= 1);
2869 wimlib_assert(part_number >= 1 && part_number <= total_parts);
2871 /* A valid image (or all images) must be specified. */
2872 if (image != WIMLIB_ALL_IMAGES &&
2873 (image < 1 || image > wim->hdr.image_count))
2874 return WIMLIB_ERR_INVALID_IMAGE;
2876 /* If we need to write metadata resources, make sure the ::WIMStruct has
2877 * the needed information attached (e.g. is not a resource-only WIM,
2878 * such as a non-first part of a split WIM). */
2879 if (!wim_has_metadata(wim) &&
2880 !(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA))
2881 return WIMLIB_ERR_METADATA_NOT_FOUND;
2883 /* Check for contradictory flags. */
2884 if ((write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2885 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))
2886 == (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2887 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))
2888 return WIMLIB_ERR_INVALID_PARAM;
2890 if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2891 WIMLIB_WRITE_FLAG_NOT_PIPABLE))
2892 == (WIMLIB_WRITE_FLAG_PIPABLE |
2893 WIMLIB_WRITE_FLAG_NOT_PIPABLE))
2894 return WIMLIB_ERR_INVALID_PARAM;
2896 /* Save previous header, then start initializing the new one. */
2897 memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header));
2899 /* Set default integrity, pipable, and packed stream flags. */
2900 if (!(write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2901 WIMLIB_WRITE_FLAG_NOT_PIPABLE)))
2902 if (wim_is_pipable(wim)) {
2903 DEBUG("WIM is pipable; default to PIPABLE.");
2904 write_flags |= WIMLIB_WRITE_FLAG_PIPABLE;
2907 if (!(write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2908 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)))
2909 if (wim_has_integrity_table(wim)) {
2910 DEBUG("Integrity table present; default to CHECK_INTEGRITY.");
2911 write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY;
2914 if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2915 WIMLIB_WRITE_FLAG_PACK_STREAMS))
2916 == (WIMLIB_WRITE_FLAG_PIPABLE |
2917 WIMLIB_WRITE_FLAG_PACK_STREAMS))
2919 ERROR("Cannot specify both PIPABLE and PACK_STREAMS!");
2920 return WIMLIB_ERR_INVALID_PARAM;
2923 /* Set appropriate magic number. */
2924 if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
2925 wim->hdr.magic = PWM_MAGIC;
2927 wim->hdr.magic = WIM_MAGIC;
2929 /* Set appropriate version number. */
2930 if ((write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS) ||
2931 wim->out_compression_type == WIMLIB_COMPRESSION_TYPE_LZMS)
2932 wim->hdr.wim_version = WIM_VERSION_PACKED_STREAMS;
2934 wim->hdr.wim_version = WIM_VERSION_DEFAULT;
2936 /* Clear header flags that will be set automatically. */
2937 wim->hdr.flags &= ~(WIM_HDR_FLAG_METADATA_ONLY |
2938 WIM_HDR_FLAG_RESOURCE_ONLY |
2939 WIM_HDR_FLAG_SPANNED |
2940 WIM_HDR_FLAG_WRITE_IN_PROGRESS);
2942 /* Set SPANNED header flag if writing part of a split WIM. */
2943 if (total_parts != 1)
2944 wim->hdr.flags |= WIM_HDR_FLAG_SPANNED;
2946 /* Set part number and total parts of split WIM. This will be 1 and 1
2947 * if the WIM is standalone. */
2948 wim->hdr.part_number = part_number;
2949 wim->hdr.total_parts = total_parts;
2951 /* Set compression type if different. */
2952 if (wim->compression_type != wim->out_compression_type) {
2953 ret = set_wim_hdr_cflags(wim->out_compression_type, &wim->hdr);
2954 wimlib_assert(ret == 0);
2957 /* Set chunk size if different. */
2958 wim->hdr.chunk_size = wim->out_chunk_size;
2961 if (!(write_flags & WIMLIB_WRITE_FLAG_RETAIN_GUID)) {
2963 memcpy(wim->hdr.guid, guid, WIMLIB_GUID_LEN);
2965 randomize_byte_array(wim->hdr.guid, WIMLIB_GUID_LEN);
2968 /* Clear references to resources that have not been written yet. */
2969 zero_reshdr(&wim->hdr.lookup_table_reshdr);
2970 zero_reshdr(&wim->hdr.xml_data_reshdr);
2971 zero_reshdr(&wim->hdr.boot_metadata_reshdr);
2972 zero_reshdr(&wim->hdr.integrity_table_reshdr);
2974 /* Set image count and boot index correctly for single image writes. */
2975 if (image != WIMLIB_ALL_IMAGES) {
2976 wim->hdr.image_count = 1;
2977 if (wim->hdr.boot_idx == image)
2978 wim->hdr.boot_idx = 1;
2980 wim->hdr.boot_idx = 0;
2983 /* Split WIMs can't be bootable. */
2984 if (total_parts != 1)
2985 wim->hdr.boot_idx = 0;
2987 /* Initialize output file descriptor. */
2988 if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR) {
2989 /* File descriptor was explicitly provided. Return error if
2990 * file descriptor is not seekable, unless writing a pipable WIM
2992 wim->out_fd.fd = *(const int*)path_or_fd;
2993 wim->out_fd.offset = 0;
2994 if (!filedes_is_seekable(&wim->out_fd)) {
2995 ret = WIMLIB_ERR_INVALID_PARAM;
2996 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
2997 goto out_restore_hdr;
2998 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
2999 ERROR("Can't include integrity check when "
3000 "writing pipable WIM to pipe!");
3001 goto out_restore_hdr;
3006 /* Filename of WIM to write was provided; open file descriptor
3008 ret = open_wim_writable(wim, (const tchar*)path_or_fd,
3009 O_TRUNC | O_CREAT | O_RDWR);
3011 goto out_restore_hdr;
3014 /* Write initial header. This is merely a "dummy" header since it
3015 * doesn't have all the information yet, so it will be overwritten later
3016 * (unless writing a pipable WIM). */
3017 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
3018 wim->hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
3019 ret = write_wim_header(&wim->hdr, &wim->out_fd);
3020 wim->hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS;
3022 goto out_restore_hdr;
3024 /* Write metadata resources and streams. */
3025 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE)) {
3026 /* Default case: create a normal (non-pipable) WIM. */
3027 ret = write_wim_streams(wim, image, write_flags, num_threads,
3028 stream_list_override,
3029 &lookup_table_list);
3031 goto out_restore_hdr;
3033 ret = write_wim_metadata_resources(wim, image, write_flags);
3035 goto out_restore_hdr;
3037 /* Non-default case: create pipable WIM. */
3038 ret = write_pipable_wim(wim, image, write_flags, num_threads,
3039 stream_list_override,
3040 &lookup_table_list);
3042 goto out_restore_hdr;
3043 write_flags |= WIMLIB_WRITE_FLAG_HEADER_AT_END;
3047 /* Write lookup table, XML data, and (optional) integrity table. */
3048 ret = finish_write(wim, image, write_flags, &lookup_table_list);
3050 memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header));
3051 (void)close_wim_writable(wim, write_flags);
3052 DEBUG("ret=%d", ret);
3056 /* Write a standalone WIM to a file or file descriptor. */
3058 write_standalone_wim(WIMStruct *wim, const void *path_or_fd,
3059 int image, int write_flags, unsigned num_threads)
3061 return write_wim_part(wim, path_or_fd, image, write_flags,
3062 num_threads, 1, 1, NULL, NULL);
3065 /* API function documented in wimlib.h */
3067 wimlib_write(WIMStruct *wim, const tchar *path,
3068 int image, int write_flags, unsigned num_threads)
3070 if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC)
3071 return WIMLIB_ERR_INVALID_PARAM;
3073 if (path == NULL || path[0] == T('\0'))
3074 return WIMLIB_ERR_INVALID_PARAM;
3076 return write_standalone_wim(wim, path, image, write_flags, num_threads);
3079 /* API function documented in wimlib.h */
3081 wimlib_write_to_fd(WIMStruct *wim, int fd,
3082 int image, int write_flags, unsigned num_threads)
3084 if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC)
3085 return WIMLIB_ERR_INVALID_PARAM;
3088 return WIMLIB_ERR_INVALID_PARAM;
3090 write_flags |= WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR;
3092 return write_standalone_wim(wim, &fd, image, write_flags, num_threads);
3096 any_images_modified(WIMStruct *wim)
3098 for (int i = 0; i < wim->hdr.image_count; i++)
3099 if (wim->image_metadata[i]->modified)
3105 check_resource_offset(struct wim_lookup_table_entry *lte, void *_wim)
3107 const WIMStruct *wim = _wim;
3108 off_t end_offset = *(const off_t*)wim->private;
3110 if (lte->resource_location == RESOURCE_IN_WIM && lte->rspec->wim == wim &&
3111 lte->rspec->offset_in_wim + lte->rspec->size_in_wim > end_offset)
3112 return WIMLIB_ERR_RESOURCE_ORDER;
3116 /* Make sure no file or metadata resources are located after the XML data (or
3117 * integrity table if present)--- otherwise we can't safely overwrite the WIM in
3118 * place and we return WIMLIB_ERR_RESOURCE_ORDER. */
3120 check_resource_offsets(WIMStruct *wim, off_t end_offset)
3125 wim->private = &end_offset;
3126 ret = for_lookup_table_entry(wim->lookup_table, check_resource_offset, wim);
3130 for (i = 0; i < wim->hdr.image_count; i++) {
3131 ret = check_resource_offset(wim->image_metadata[i]->metadata_lte, wim);
3139 * Overwrite a WIM, possibly appending streams to it.
3141 * A WIM looks like (or is supposed to look like) the following:
3143 * Header (212 bytes)
3144 * Streams and metadata resources (variable size)
3145 * Lookup table (variable size)
3146 * XML data (variable size)
3147 * Integrity table (optional) (variable size)
3149 * If we are not adding any streams or metadata resources, the lookup table is
3150 * unchanged--- so we only need to overwrite the XML data, integrity table, and
3151 * header. This operation is potentially unsafe if the program is abruptly
3152 * terminated while the XML data or integrity table are being overwritten, but
3153 * before the new header has been written. To partially alleviate this problem,
3154 * a special flag (WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) is passed to
3155 * finish_write() to cause a temporary WIM header to be written after the XML
3156 * data has been written. This may prevent the WIM from becoming corrupted if
3157 * the program is terminated while the integrity table is being calculated (but
3158 * no guarantees, due to write re-ordering...).
3160 * If we are adding new streams or images (metadata resources), the lookup table
3161 * needs to be changed, and those streams need to be written. In this case, we
3162 * try to perform a safe update of the WIM file by writing the streams *after*
3163 * the end of the previous WIM, then writing the new lookup table, XML data, and
3164 * (optionally) integrity table following the new streams. This will produce a
3165 * layout like the following:
3167 * Header (212 bytes)
3168 * (OLD) Streams and metadata resources (variable size)
3169 * (OLD) Lookup table (variable size)
3170 * (OLD) XML data (variable size)
3171 * (OLD) Integrity table (optional) (variable size)
3172 * (NEW) Streams and metadata resources (variable size)
3173 * (NEW) Lookup table (variable size)
3174 * (NEW) XML data (variable size)
3175 * (NEW) Integrity table (optional) (variable size)
3177 * At all points, the WIM is valid as nothing points to the new data yet. Then,
3178 * the header is overwritten to point to the new lookup table, XML data, and
3179 * integrity table, to produce the following layout:
3181 * Header (212 bytes)
3182 * Streams and metadata resources (variable size)
3183 * Nothing (variable size)
3184 * More Streams and metadata resources (variable size)
3185 * Lookup table (variable size)
3186 * XML data (variable size)
3187 * Integrity table (optional) (variable size)
3189 * This method allows an image to be appended to a large WIM very quickly, and
3190 * is crash-safe except in the case of write re-ordering, but the
3191 * disadvantage is that a small hole is left in the WIM where the old lookup
3192 * table, xml data, and integrity table were. (These usually only take up a
3193 * small amount of space compared to the streams, however.)
3196 overwrite_wim_inplace(WIMStruct *wim, int write_flags, unsigned num_threads)
3200 u64 old_lookup_table_end, old_xml_begin, old_xml_end;
3201 struct wim_header hdr_save;
3202 struct list_head stream_list;
3203 struct list_head lookup_table_list;
3204 struct filter_context filter_ctx;
3206 DEBUG("Overwriting `%"TS"' in-place", wim->filename);
3208 /* Save original header so it can be restored in case of error */
3209 memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header));
3211 /* Set default integrity flag. */
3212 if (!(write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
3213 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)))
3214 if (wim_has_integrity_table(wim))
3215 write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY;
3217 /* Set WIM version if adding packed streams. */
3218 if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS)
3219 wim->hdr.wim_version = WIM_VERSION_PACKED_STREAMS;
3221 /* Set additional flags for overwrite. */
3222 write_flags |= WIMLIB_WRITE_FLAG_OVERWRITE |
3223 WIMLIB_WRITE_FLAG_STREAMS_OK;
3225 /* Make sure that the integrity table (if present) is after the XML
3226 * data, and that there are no stream resources, metadata resources, or
3227 * lookup tables after the XML data. Otherwise, these data would be
3229 old_xml_begin = wim->hdr.xml_data_reshdr.offset_in_wim;
3230 old_xml_end = old_xml_begin + wim->hdr.xml_data_reshdr.size_in_wim;
3231 old_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim +
3232 wim->hdr.lookup_table_reshdr.size_in_wim;
3233 if (wim->hdr.integrity_table_reshdr.offset_in_wim != 0 &&
3234 wim->hdr.integrity_table_reshdr.offset_in_wim < old_xml_end) {
3235 WARNING("Didn't expect the integrity table to be before the XML data");
3236 ret = WIMLIB_ERR_RESOURCE_ORDER;
3237 goto out_restore_memory_hdr;
3240 if (old_lookup_table_end > old_xml_begin) {
3241 WARNING("Didn't expect the lookup table to be after the XML data");
3242 ret = WIMLIB_ERR_RESOURCE_ORDER;
3243 goto out_restore_memory_hdr;
3246 /* Set @old_wim_end, which indicates the point beyond which we don't
3247 * allow any file and metadata resources to appear without returning
3248 * WIMLIB_ERR_RESOURCE_ORDER (due to the fact that we would otherwise
3249 * overwrite these resources). */
3250 if (!wim->image_deletion_occurred && !any_images_modified(wim)) {
3251 /* If no images have been modified and no images have been
3252 * deleted, a new lookup table does not need to be written. We
3253 * shall write the new XML data and optional integrity table
3254 * immediately after the lookup table. Note that this may
3255 * overwrite an existing integrity table. */
3256 DEBUG("Skipping writing lookup table "
3257 "(no images modified or deleted)");
3258 old_wim_end = old_lookup_table_end;
3259 write_flags |= WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE |
3260 WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML;
3261 } else if (wim->hdr.integrity_table_reshdr.offset_in_wim != 0) {
3262 /* Old WIM has an integrity table; begin writing new streams
3264 old_wim_end = wim->hdr.integrity_table_reshdr.offset_in_wim +
3265 wim->hdr.integrity_table_reshdr.size_in_wim;
3267 /* No existing integrity table; begin writing new streams after
3268 * the old XML data. */
3269 old_wim_end = old_xml_end;
3272 ret = check_resource_offsets(wim, old_wim_end);
3274 goto out_restore_memory_hdr;
3276 ret = prepare_stream_list_for_write(wim, WIMLIB_ALL_IMAGES, write_flags,
3277 &stream_list, &lookup_table_list,
3280 goto out_restore_memory_hdr;
3282 ret = open_wim_writable(wim, wim->filename, O_RDWR);
3284 goto out_restore_memory_hdr;
3286 ret = lock_wim_for_append(wim);
3290 /* Set WIM_HDR_FLAG_WRITE_IN_PROGRESS flag in header. */
3291 wim->hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
3292 ret = write_wim_header_flags(wim->hdr.flags, &wim->out_fd);
3294 ERROR_WITH_ERRNO("Error updating WIM header flags");
3295 goto out_unlock_wim;
3298 if (filedes_seek(&wim->out_fd, old_wim_end) == -1) {
3299 ERROR_WITH_ERRNO("Can't seek to end of WIM");
3300 ret = WIMLIB_ERR_WRITE;
3301 goto out_restore_physical_hdr;
3304 ret = wim_write_stream_list(wim,
3312 ret = write_wim_metadata_resources(wim, WIMLIB_ALL_IMAGES, write_flags);
3316 ret = finish_write(wim, WIMLIB_ALL_IMAGES, write_flags,
3317 &lookup_table_list);
3321 unlock_wim_for_append(wim);
3325 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
3326 WARNING("Truncating `%"TS"' to its original size (%"PRIu64" bytes)",
3327 wim->filename, old_wim_end);
3328 /* Return value of ftruncate() is ignored because this is
3329 * already an error path. */
3330 (void)ftruncate(wim->out_fd.fd, old_wim_end);
3332 out_restore_physical_hdr:
3333 (void)write_wim_header_flags(hdr_save.flags, &wim->out_fd);
3335 unlock_wim_for_append(wim);
3337 (void)close_wim_writable(wim, write_flags);
3338 out_restore_memory_hdr:
3339 memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header));
3344 overwrite_wim_via_tmpfile(WIMStruct *wim, int write_flags, unsigned num_threads)
3346 size_t wim_name_len;
3349 DEBUG("Overwriting `%"TS"' via a temporary file", wim->filename);
3351 /* Write the WIM to a temporary file in the same directory as the
3353 wim_name_len = tstrlen(wim->filename);
3354 tchar tmpfile[wim_name_len + 10];
3355 tmemcpy(tmpfile, wim->filename, wim_name_len);
3356 randomize_char_array_with_alnum(tmpfile + wim_name_len, 9);
3357 tmpfile[wim_name_len + 9] = T('\0');
3359 ret = wimlib_write(wim, tmpfile, WIMLIB_ALL_IMAGES,
3361 WIMLIB_WRITE_FLAG_FSYNC |
3362 WIMLIB_WRITE_FLAG_RETAIN_GUID,
3369 if (filedes_valid(&wim->in_fd)) {
3370 filedes_close(&wim->in_fd);
3371 filedes_invalidate(&wim->in_fd);
3374 /* Rename the new WIM file to the original WIM file. Note: on Windows
3375 * this actually calls win32_rename_replacement(), not _wrename(), so
3376 * that removing the existing destination file can be handled. */
3377 DEBUG("Renaming `%"TS"' to `%"TS"'", tmpfile, wim->filename);
3378 ret = trename(tmpfile, wim->filename);
3380 ERROR_WITH_ERRNO("Failed to rename `%"TS"' to `%"TS"'",
3381 tmpfile, wim->filename);
3388 return WIMLIB_ERR_RENAME;
3391 union wimlib_progress_info progress;
3392 progress.rename.from = tmpfile;
3393 progress.rename.to = wim->filename;
3394 return call_progress(wim->progfunc, WIMLIB_PROGRESS_MSG_RENAME,
3395 &progress, wim->progctx);
3398 /* Determine if the specified WIM file may be updated by appending in-place
3399 * rather than writing and replacing it with an entirely new file. */
3401 can_overwrite_wim_inplace(const WIMStruct *wim, int write_flags)
3403 /* REBUILD flag forces full rebuild. */
3404 if (write_flags & WIMLIB_WRITE_FLAG_REBUILD)
3407 /* Image deletions cause full rebuild by default. */
3408 if (wim->image_deletion_occurred &&
3409 !(write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE))
3412 /* Pipable WIMs cannot be updated in place, nor can a non-pipable WIM be
3413 * turned into a pipable WIM in-place. */
3414 if (wim_is_pipable(wim) || (write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
3417 /* The default compression type and compression chunk size selected for
3418 * the output WIM must be the same as those currently used for the WIM.
3420 if (wim->compression_type != wim->out_compression_type)
3422 if (wim->chunk_size != wim->out_chunk_size)
3428 /* API function documented in wimlib.h */
3430 wimlib_overwrite(WIMStruct *wim, int write_flags, unsigned num_threads)
3435 if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC)
3436 return WIMLIB_ERR_INVALID_PARAM;
3439 return WIMLIB_ERR_NO_FILENAME;
3441 orig_hdr_flags = wim->hdr.flags;
3442 if (write_flags & WIMLIB_WRITE_FLAG_IGNORE_READONLY_FLAG)
3443 wim->hdr.flags &= ~WIM_HDR_FLAG_READONLY;
3444 ret = can_modify_wim(wim);
3445 wim->hdr.flags = orig_hdr_flags;
3449 if (can_overwrite_wim_inplace(wim, write_flags)) {
3450 ret = overwrite_wim_inplace(wim, write_flags, num_threads);
3451 if (ret != WIMLIB_ERR_RESOURCE_ORDER)
3453 WARNING("Falling back to re-building entire WIM");
3455 return overwrite_wim_via_tmpfile(wim, write_flags, num_threads);