4 * Support for writing WIM files; write a WIM file, overwrite a WIM file, write
5 * compressed file resources, etc.
9 * Copyright (C) 2012-2016 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 * override the LIST_HEAD macro. */
32 # include <sys/file.h>
40 #include "wimlib/alloca.h"
41 #include "wimlib/assert.h"
42 #include "wimlib/blob_table.h"
43 #include "wimlib/chunk_compressor.h"
44 #include "wimlib/endianness.h"
45 #include "wimlib/error.h"
46 #include "wimlib/file_io.h"
47 #include "wimlib/header.h"
48 #include "wimlib/inode.h"
49 #include "wimlib/integrity.h"
50 #include "wimlib/metadata.h"
51 #include "wimlib/paths.h"
52 #include "wimlib/progress.h"
53 #include "wimlib/resource.h"
54 #include "wimlib/solid.h"
55 #include "wimlib/win32.h" /* win32_rename_replacement() */
56 #include "wimlib/write.h"
57 #include "wimlib/xml.h"
60 /* wimlib internal flags used when writing resources. */
61 #define WRITE_RESOURCE_FLAG_RECOMPRESS 0x00000001
62 #define WRITE_RESOURCE_FLAG_PIPABLE 0x00000002
63 #define WRITE_RESOURCE_FLAG_SOLID 0x00000004
64 #define WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE 0x00000008
65 #define WRITE_RESOURCE_FLAG_SOLID_SORT 0x00000010
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;
75 if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
76 write_resource_flags |= WRITE_RESOURCE_FLAG_PIPABLE;
78 if (write_flags & WIMLIB_WRITE_FLAG_SOLID)
79 write_resource_flags |= WRITE_RESOURCE_FLAG_SOLID;
81 if (write_flags & WIMLIB_WRITE_FLAG_SEND_DONE_WITH_FILE_MESSAGES)
82 write_resource_flags |= WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE;
84 if ((write_flags & (WIMLIB_WRITE_FLAG_SOLID |
85 WIMLIB_WRITE_FLAG_NO_SOLID_SORT)) ==
86 WIMLIB_WRITE_FLAG_SOLID)
87 write_resource_flags |= WRITE_RESOURCE_FLAG_SOLID_SORT;
89 return write_resource_flags;
92 struct filter_context {
98 * Determine whether the specified blob should be filtered out from the write.
102 * < 0 : The blob should be hard-filtered; that is, not included in the output
104 * 0 : The blob should not be filtered out.
105 * > 0 : The blob should be soft-filtered; that is, it already exists in the
106 * WIM file and may not need to be written again.
109 blob_filtered(const struct blob_descriptor *blob,
110 const struct filter_context *ctx)
118 write_flags = ctx->write_flags;
121 if (write_flags & WIMLIB_WRITE_FLAG_APPEND &&
122 blob->blob_location == BLOB_IN_WIM &&
123 blob->rdesc->wim == wim)
126 if (write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS &&
127 blob->blob_location == BLOB_IN_WIM &&
128 blob->rdesc->wim != wim)
135 blob_hard_filtered(const struct blob_descriptor *blob,
136 struct filter_context *ctx)
138 return blob_filtered(blob, ctx) < 0;
142 may_soft_filter_blobs(const struct filter_context *ctx)
144 return ctx && (ctx->write_flags & WIMLIB_WRITE_FLAG_APPEND);
148 may_hard_filter_blobs(const struct filter_context *ctx)
150 return ctx && (ctx->write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS);
154 may_filter_blobs(const struct filter_context *ctx)
156 return (may_soft_filter_blobs(ctx) || may_hard_filter_blobs(ctx));
159 /* Return true if the specified blob is located in a WIM resource which can be
160 * reused in the output WIM file, without being recompressed. */
162 can_raw_copy(const struct blob_descriptor *blob, int write_resource_flags,
163 int out_ctype, u32 out_chunk_size)
165 const struct wim_resource_descriptor *rdesc;
167 /* Recompress everything if requested. */
168 if (write_resource_flags & WRITE_RESOURCE_FLAG_RECOMPRESS)
171 /* A blob not located in a WIM resource cannot be reused. */
172 if (blob->blob_location != BLOB_IN_WIM)
177 /* In the case of an in-place compaction, always reuse resources located
178 * in the WIM being compacted. */
179 if (rdesc->wim->being_compacted)
182 /* Otherwise, only reuse compressed resources. */
183 if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE ||
184 !(rdesc->flags & (WIM_RESHDR_FLAG_COMPRESSED |
185 WIM_RESHDR_FLAG_SOLID)))
188 /* When writing a pipable WIM, we can only reuse pipable resources; and
189 * when writing a non-pipable WIM, we can only reuse non-pipable
191 if (rdesc->is_pipable !=
192 !!(write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE))
195 /* When writing a solid WIM, we can only reuse solid resources; and when
196 * writing a non-solid WIM, we can only reuse non-solid resources. */
197 if (!!(rdesc->flags & WIM_RESHDR_FLAG_SOLID) !=
198 !!(write_resource_flags & WRITE_RESOURCE_FLAG_SOLID))
201 /* Note: it is theoretically possible to copy chunks of compressed data
202 * between non-solid, solid, and pipable resources. However, we don't
203 * currently implement this optimization because it would be complex and
204 * would usually go unused. */
206 if (rdesc->flags & WIM_RESHDR_FLAG_COMPRESSED) {
207 /* To re-use a non-solid resource, it must use the desired
208 * compression type and chunk size. */
209 return (rdesc->compression_type == out_ctype &&
210 rdesc->chunk_size == out_chunk_size);
212 /* Solid resource: Such resources may contain multiple blobs,
213 * and in general only a subset of them need to be written. As
214 * a heuristic, re-use the raw data if more than two-thirds the
215 * uncompressed size is being written. */
217 /* Note: solid resources contain a header that specifies the
218 * compression type and chunk size; therefore we don't need to
219 * check if they are compatible with @out_ctype and
220 * @out_chunk_size. */
222 /* Did we already decide to reuse the resource? */
223 if (rdesc->raw_copy_ok)
226 struct blob_descriptor *res_blob;
229 list_for_each_entry(res_blob, &rdesc->blob_list, rdesc_node)
230 if (res_blob->will_be_in_output_wim)
231 write_size += res_blob->size;
233 return (write_size > rdesc->uncompressed_size * 2 / 3);
238 reshdr_flags_for_blob(const struct blob_descriptor *blob)
240 u32 reshdr_flags = 0;
241 if (blob->is_metadata)
242 reshdr_flags |= WIM_RESHDR_FLAG_METADATA;
247 blob_set_out_reshdr_for_reuse(struct blob_descriptor *blob)
249 const struct wim_resource_descriptor *rdesc;
251 wimlib_assert(blob->blob_location == BLOB_IN_WIM);
254 if (rdesc->flags & WIM_RESHDR_FLAG_SOLID) {
255 blob->out_reshdr.offset_in_wim = blob->offset_in_res;
256 blob->out_reshdr.uncompressed_size = 0;
257 blob->out_reshdr.size_in_wim = blob->size;
259 blob->out_res_offset_in_wim = rdesc->offset_in_wim;
260 blob->out_res_size_in_wim = rdesc->size_in_wim;
261 blob->out_res_uncompressed_size = rdesc->uncompressed_size;
263 blob->out_reshdr.offset_in_wim = rdesc->offset_in_wim;
264 blob->out_reshdr.uncompressed_size = rdesc->uncompressed_size;
265 blob->out_reshdr.size_in_wim = rdesc->size_in_wim;
267 blob->out_reshdr.flags = rdesc->flags;
271 /* Write the header for a blob in a pipable WIM. */
273 write_pwm_blob_header(const struct blob_descriptor *blob,
274 struct filedes *out_fd, bool compressed)
276 struct pwm_blob_hdr blob_hdr;
280 wimlib_assert(!blob->unhashed);
282 blob_hdr.magic = cpu_to_le64(PWM_BLOB_MAGIC);
283 blob_hdr.uncompressed_size = cpu_to_le64(blob->size);
284 copy_hash(blob_hdr.hash, blob->hash);
285 reshdr_flags = reshdr_flags_for_blob(blob);
287 reshdr_flags |= WIM_RESHDR_FLAG_COMPRESSED;
288 blob_hdr.flags = cpu_to_le32(reshdr_flags);
289 ret = full_write(out_fd, &blob_hdr, sizeof(blob_hdr));
291 ERROR_WITH_ERRNO("Error writing blob header to WIM file");
295 struct write_blobs_progress_data {
296 wimlib_progress_func_t progfunc;
298 union wimlib_progress_info progress;
303 do_write_blobs_progress(struct write_blobs_progress_data *progress_data,
304 u64 complete_size, u64 complete_compressed_size,
305 u32 complete_count, bool discarded)
307 union wimlib_progress_info *progress = &progress_data->progress;
311 progress->write_streams.total_bytes -= complete_size;
312 progress->write_streams.total_streams -= complete_count;
313 if (progress_data->next_progress != ~(u64)0 &&
314 progress_data->next_progress > progress->write_streams.total_bytes)
316 progress_data->next_progress = progress->write_streams.total_bytes;
319 progress->write_streams.completed_bytes += complete_size;
320 progress->write_streams.completed_compressed_bytes +=
321 complete_compressed_size;
322 progress->write_streams.completed_streams += complete_count;
325 if (progress->write_streams.completed_bytes >= progress_data->next_progress) {
327 ret = call_progress(progress_data->progfunc,
328 WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
330 progress_data->progctx);
334 set_next_progress(progress->write_streams.completed_bytes,
335 progress->write_streams.total_bytes,
336 &progress_data->next_progress);
341 struct write_blobs_ctx {
342 /* File descriptor to which the blobs are being written. */
343 struct filedes *out_fd;
345 /* Blob table for the WIMStruct on whose behalf the blobs are being
347 struct blob_table *blob_table;
349 /* Compression format to use. */
352 /* Maximum uncompressed chunk size in compressed resources to use. */
355 /* Flags that affect how the blobs will be written. */
356 int write_resource_flags;
358 /* Data used for issuing WRITE_STREAMS progress. */
359 struct write_blobs_progress_data progress_data;
361 struct filter_context *filter_ctx;
363 /* Pointer to the chunk_compressor implementation being used for
364 * compressing chunks of data, or NULL if chunks are being written
366 struct chunk_compressor *compressor;
368 /* A buffer of size @out_chunk_size that has been loaned out from the
369 * chunk compressor and is currently being filled with the uncompressed
370 * data of the next chunk. */
373 /* Number of bytes in @cur_chunk_buf that are currently filled. */
374 size_t cur_chunk_buf_filled;
376 /* List of blobs that currently have chunks being compressed. */
377 struct list_head blobs_being_compressed;
379 /* List of blobs in the solid resource. Blobs are moved here after
380 * @blobs_being_compressed only when writing a solid resource. */
381 struct list_head blobs_in_solid_resource;
383 /* Current uncompressed offset in the blob being written. */
384 u64 cur_write_blob_offset;
386 /* Uncompressed size of resource currently being written. */
387 u64 cur_write_res_size;
389 /* Array that is filled in with compressed chunk sizes as a resource is
393 /* Index of next entry in @chunk_csizes to fill in. */
396 /* Number of entries in @chunk_csizes currently allocated. */
397 size_t num_alloc_chunks;
399 /* Offset in the output file of the start of the chunks of the resource
400 * currently being written. */
401 u64 chunks_start_offset;
404 /* Reserve space for the chunk table and prepare to accumulate the chunk table
407 begin_chunk_table(struct write_blobs_ctx *ctx, u64 res_expected_size)
409 u64 expected_num_chunks;
410 u64 expected_num_chunk_entries;
414 /* Calculate the number of chunks and chunk entries that should be
415 * needed for the resource. These normally will be the final values,
416 * but in SOLID mode some of the blobs we're planning to write into the
417 * resource may be duplicates, and therefore discarded, potentially
418 * decreasing the number of chunk entries needed. */
419 expected_num_chunks = DIV_ROUND_UP(res_expected_size, ctx->out_chunk_size);
420 expected_num_chunk_entries = expected_num_chunks;
421 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID))
422 expected_num_chunk_entries--;
424 /* Make sure the chunk_csizes array is long enough to store the
425 * compressed size of each chunk. */
426 if (expected_num_chunks > ctx->num_alloc_chunks) {
427 u64 new_length = expected_num_chunks + 50;
429 if ((size_t)new_length != new_length) {
430 ERROR("Resource size too large (%"PRIu64" bytes!",
432 return WIMLIB_ERR_NOMEM;
435 FREE(ctx->chunk_csizes);
436 ctx->chunk_csizes = MALLOC(new_length * sizeof(ctx->chunk_csizes[0]));
437 if (ctx->chunk_csizes == NULL) {
438 ctx->num_alloc_chunks = 0;
439 return WIMLIB_ERR_NOMEM;
441 ctx->num_alloc_chunks = new_length;
444 ctx->chunk_index = 0;
446 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE)) {
447 /* Reserve space for the chunk table in the output file. In the
448 * case of solid resources this reserves the upper bound for the
449 * needed space, not necessarily the exact space which will
450 * prove to be needed. At this point, we just use @chunk_csizes
451 * for a buffer of 0's because the actual compressed chunk sizes
453 reserve_size = expected_num_chunk_entries *
454 get_chunk_entry_size(res_expected_size,
455 0 != (ctx->write_resource_flags &
456 WRITE_RESOURCE_FLAG_SOLID));
457 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID)
458 reserve_size += sizeof(struct alt_chunk_table_header_disk);
459 memset(ctx->chunk_csizes, 0, reserve_size);
460 ret = full_write(ctx->out_fd, ctx->chunk_csizes, reserve_size);
462 ERROR_WITH_ERRNO("Error reserving space for chunk "
463 "table in WIM file");
471 begin_write_resource(struct write_blobs_ctx *ctx, u64 res_expected_size)
475 wimlib_assert(res_expected_size != 0);
477 if (ctx->compressor != NULL) {
478 ret = begin_chunk_table(ctx, res_expected_size);
483 /* Output file descriptor is now positioned at the offset at which to
484 * write the first chunk of the resource. */
485 ctx->chunks_start_offset = ctx->out_fd->offset;
486 ctx->cur_write_blob_offset = 0;
487 ctx->cur_write_res_size = res_expected_size;
492 end_chunk_table(struct write_blobs_ctx *ctx, u64 res_actual_size,
493 u64 *res_start_offset_ret, u64 *res_store_size_ret)
495 size_t actual_num_chunks;
496 size_t actual_num_chunk_entries;
497 size_t chunk_entry_size;
500 actual_num_chunks = ctx->chunk_index;
501 actual_num_chunk_entries = actual_num_chunks;
502 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID))
503 actual_num_chunk_entries--;
505 chunk_entry_size = get_chunk_entry_size(res_actual_size,
506 0 != (ctx->write_resource_flags &
507 WRITE_RESOURCE_FLAG_SOLID));
509 typedef le64 _may_alias_attribute aliased_le64_t;
510 typedef le32 _may_alias_attribute aliased_le32_t;
512 if (chunk_entry_size == 4) {
513 aliased_le32_t *entries = (aliased_le32_t*)ctx->chunk_csizes;
515 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
516 for (size_t i = 0; i < actual_num_chunk_entries; i++)
517 entries[i] = cpu_to_le32(ctx->chunk_csizes[i]);
519 u32 offset = ctx->chunk_csizes[0];
520 for (size_t i = 0; i < actual_num_chunk_entries; i++) {
521 u32 next_size = ctx->chunk_csizes[i + 1];
522 entries[i] = cpu_to_le32(offset);
527 aliased_le64_t *entries = (aliased_le64_t*)ctx->chunk_csizes;
529 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
530 for (size_t i = 0; i < actual_num_chunk_entries; i++)
531 entries[i] = cpu_to_le64(ctx->chunk_csizes[i]);
533 u64 offset = ctx->chunk_csizes[0];
534 for (size_t i = 0; i < actual_num_chunk_entries; i++) {
535 u64 next_size = ctx->chunk_csizes[i + 1];
536 entries[i] = cpu_to_le64(offset);
542 size_t chunk_table_size = actual_num_chunk_entries * chunk_entry_size;
543 u64 res_start_offset;
546 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
547 ret = full_write(ctx->out_fd, ctx->chunk_csizes, chunk_table_size);
550 res_end_offset = ctx->out_fd->offset;
551 res_start_offset = ctx->chunks_start_offset;
553 res_end_offset = ctx->out_fd->offset;
555 u64 chunk_table_offset;
557 chunk_table_offset = ctx->chunks_start_offset - chunk_table_size;
559 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
560 struct alt_chunk_table_header_disk hdr;
562 hdr.res_usize = cpu_to_le64(res_actual_size);
563 hdr.chunk_size = cpu_to_le32(ctx->out_chunk_size);
564 hdr.compression_format = cpu_to_le32(ctx->out_ctype);
566 STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_XPRESS == 1);
567 STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_LZX == 2);
568 STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_LZMS == 3);
570 ret = full_pwrite(ctx->out_fd, &hdr, sizeof(hdr),
571 chunk_table_offset - sizeof(hdr));
574 res_start_offset = chunk_table_offset - sizeof(hdr);
576 res_start_offset = chunk_table_offset;
579 ret = full_pwrite(ctx->out_fd, ctx->chunk_csizes,
580 chunk_table_size, chunk_table_offset);
585 *res_start_offset_ret = res_start_offset;
586 *res_store_size_ret = res_end_offset - res_start_offset;
591 ERROR_WITH_ERRNO("Error writing chunk table to WIM file");
595 /* Finish writing a WIM resource by writing or updating the chunk table (if not
596 * writing the data uncompressed) and loading its metadata into @out_reshdr. */
598 end_write_resource(struct write_blobs_ctx *ctx, struct wim_reshdr *out_reshdr)
602 u64 res_uncompressed_size;
603 u64 res_offset_in_wim;
605 wimlib_assert(ctx->cur_write_blob_offset == ctx->cur_write_res_size ||
606 (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID));
607 res_uncompressed_size = ctx->cur_write_res_size;
609 if (ctx->compressor) {
610 ret = end_chunk_table(ctx, res_uncompressed_size,
611 &res_offset_in_wim, &res_size_in_wim);
615 res_offset_in_wim = ctx->chunks_start_offset;
616 res_size_in_wim = ctx->out_fd->offset - res_offset_in_wim;
618 out_reshdr->uncompressed_size = res_uncompressed_size;
619 out_reshdr->size_in_wim = res_size_in_wim;
620 out_reshdr->offset_in_wim = res_offset_in_wim;
624 /* Call when no more data from the file at @path is needed. */
626 done_with_file(const tchar *path, wimlib_progress_func_t progfunc, void *progctx)
628 union wimlib_progress_info info;
630 info.done_with_file.path_to_file = path;
632 return call_progress(progfunc, WIMLIB_PROGRESS_MSG_DONE_WITH_FILE,
637 do_done_with_blob(struct blob_descriptor *blob,
638 wimlib_progress_func_t progfunc, void *progctx)
641 struct wim_inode *inode;
646 if (!blob->may_send_done_with_file)
649 inode = blob->file_inode;
651 wimlib_assert(inode != NULL);
652 wimlib_assert(inode->i_num_remaining_streams > 0);
653 if (--inode->i_num_remaining_streams > 0)
656 path = blob_file_path(blob);
658 cookie1 = progress_get_streamless_path(path);
659 cookie2 = progress_get_win32_path(path);
661 ret = done_with_file(path, progfunc, progctx);
663 progress_put_win32_path(cookie2);
664 progress_put_streamless_path(cookie1);
669 /* Handle WIMLIB_WRITE_FLAG_SEND_DONE_WITH_FILE_MESSAGES mode. */
671 done_with_blob(struct blob_descriptor *blob, struct write_blobs_ctx *ctx)
673 if (likely(!(ctx->write_resource_flags &
674 WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE)))
676 return do_done_with_blob(blob, ctx->progress_data.progfunc,
677 ctx->progress_data.progctx);
680 /* Begin processing a blob for writing. */
682 write_blob_begin_read(struct blob_descriptor *blob, void *_ctx)
684 struct write_blobs_ctx *ctx = _ctx;
687 wimlib_assert(blob->size > 0);
689 /* As an optimization, we allow some blobs to be "unhashed", meaning
690 * their SHA-1 message digests are unknown. This is the case with blobs
691 * that are added by scanning a directory tree with wimlib_add_image(),
692 * for example. Since WIM uses single-instance blobs, we don't know
693 * whether such each such blob really need to written until it is
694 * actually checksummed, unless it has a unique size. In such cases we
695 * read and checksum the blob in this function, thereby advancing ahead
696 * of read_blob_list(), which will still provide the data again to
697 * write_blob_process_chunk(). This is okay because an unhashed blob
698 * cannot be in a WIM resource, which might be costly to decompress. */
699 if (ctx->blob_table != NULL && blob->unhashed && !blob->unique_size) {
701 struct blob_descriptor *new_blob;
703 ret = hash_unhashed_blob(blob, ctx->blob_table, &new_blob);
706 if (new_blob != blob) {
707 /* Duplicate blob detected. */
709 if (new_blob->will_be_in_output_wim ||
710 blob_filtered(new_blob, ctx->filter_ctx))
712 /* The duplicate blob is already being included
713 * in the output WIM, or it would be filtered
714 * out if it had been. Skip writing this blob
715 * (and reading it again) entirely, passing its
716 * output reference count to the duplicate blob
717 * in the former case. */
718 ret = do_write_blobs_progress(&ctx->progress_data,
722 list_del(&blob->write_blobs_list);
723 list_del(&blob->blob_table_list);
724 if (new_blob->will_be_in_output_wim)
725 new_blob->out_refcnt += blob->out_refcnt;
726 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID)
727 ctx->cur_write_res_size -= blob->size;
729 ret = done_with_blob(blob, ctx);
730 free_blob_descriptor(blob);
733 return BEGIN_BLOB_STATUS_SKIP_BLOB;
735 /* The duplicate blob can validly be written,
736 * but was not marked as such. Discard the
737 * current blob descriptor and use the
738 * duplicate, but actually freeing the current
739 * blob descriptor must wait until
740 * read_blob_list() has finished reading its
742 list_replace(&blob->write_blobs_list,
743 &new_blob->write_blobs_list);
744 list_replace(&blob->blob_table_list,
745 &new_blob->blob_table_list);
746 blob->will_be_in_output_wim = 0;
747 new_blob->out_refcnt = blob->out_refcnt;
748 new_blob->will_be_in_output_wim = 1;
749 new_blob->may_send_done_with_file = 0;
754 list_move_tail(&blob->write_blobs_list, &ctx->blobs_being_compressed);
758 /* Rewrite a blob that was just written compressed (as a non-solid WIM resource)
759 * as uncompressed instead. */
761 write_blob_uncompressed(struct blob_descriptor *blob, struct filedes *out_fd)
764 u64 begin_offset = blob->out_reshdr.offset_in_wim;
765 u64 end_offset = out_fd->offset;
767 if (filedes_seek(out_fd, begin_offset) == -1)
770 ret = extract_blob_to_fd(blob, out_fd, false);
772 /* Error reading the uncompressed data. */
773 if (out_fd->offset == begin_offset &&
774 filedes_seek(out_fd, end_offset) != -1)
776 /* Nothing was actually written yet, and we successfully
777 * seeked to the end of the compressed resource, so
778 * don't issue a hard error; just keep the compressed
779 * resource instead. */
780 WARNING("Recovered compressed resource of "
781 "size %"PRIu64", continuing on.", blob->size);
787 wimlib_assert(out_fd->offset - begin_offset == blob->size);
789 /* We could ftruncate() the file to 'out_fd->offset' here, but there
790 * isn't much point. Usually we will only be truncating by a few bytes
791 * and will just overwrite the data immediately. */
793 blob->out_reshdr.size_in_wim = blob->size;
794 blob->out_reshdr.flags &= ~(WIM_RESHDR_FLAG_COMPRESSED |
795 WIM_RESHDR_FLAG_SOLID);
799 /* Returns true if the specified blob, which was written as a non-solid
800 * resource, should be truncated from the WIM file and re-written uncompressed.
801 * blob->out_reshdr must be filled in from the initial write of the blob. */
803 should_rewrite_blob_uncompressed(const struct write_blobs_ctx *ctx,
804 const struct blob_descriptor *blob)
806 /* If the compressed data is smaller than the uncompressed data, prefer
807 * the compressed data. */
808 if (blob->out_reshdr.size_in_wim < blob->out_reshdr.uncompressed_size)
811 /* If we're not actually writing compressed data, then there's no need
813 if (!ctx->compressor)
816 /* If writing a pipable WIM, everything we write to the output is final
817 * (it might actually be a pipe!). */
818 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE)
821 /* If the blob that would need to be re-read is located in a solid
822 * resource in another WIM file, then re-reading it would be costly. So
825 * Exception: if the compressed size happens to be *exactly* the same as
826 * the uncompressed size, then the blob *must* be written uncompressed
827 * in order to remain compatible with the Windows Overlay Filesystem
828 * filter driver (WOF).
830 * TODO: we are currently assuming that the optimization for
831 * single-chunk resources in maybe_rewrite_blob_uncompressed() prevents
832 * this case from being triggered too often. To fully prevent excessive
833 * decompressions in degenerate cases, we really should obtain the
834 * uncompressed data by decompressing the compressed data we wrote to
837 if (blob->blob_location == BLOB_IN_WIM &&
838 blob->size != blob->rdesc->uncompressed_size &&
839 blob->size != blob->out_reshdr.size_in_wim)
846 maybe_rewrite_blob_uncompressed(struct write_blobs_ctx *ctx,
847 struct blob_descriptor *blob)
849 if (!should_rewrite_blob_uncompressed(ctx, blob))
852 /* Regular (non-solid) WIM resources with exactly one chunk and
853 * compressed size equal to uncompressed size are exactly the same as
854 * the corresponding compressed data --- since there must be 0 entries
855 * in the chunk table and the only chunk must be stored uncompressed.
856 * In this case, there's no need to rewrite anything. */
857 if (ctx->chunk_index == 1 &&
858 blob->out_reshdr.size_in_wim == blob->out_reshdr.uncompressed_size)
860 blob->out_reshdr.flags &= ~WIM_RESHDR_FLAG_COMPRESSED;
864 return write_blob_uncompressed(blob, ctx->out_fd);
867 /* Write the next chunk of (typically compressed) data to the output WIM,
868 * handling the writing of the chunk table. */
870 write_chunk(struct write_blobs_ctx *ctx, const void *cchunk,
871 size_t csize, size_t usize)
874 struct blob_descriptor *blob;
875 u32 completed_blob_count = 0;
877 blob = list_entry(ctx->blobs_being_compressed.next,
878 struct blob_descriptor, write_blobs_list);
880 if (ctx->cur_write_blob_offset == 0 &&
881 !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID))
883 /* Starting to write a new blob in non-solid mode. */
885 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
886 ret = write_pwm_blob_header(blob, ctx->out_fd,
887 ctx->compressor != NULL);
892 ret = begin_write_resource(ctx, blob->size);
897 if (ctx->compressor != NULL) {
898 /* Record the compresed chunk size. */
899 wimlib_assert(ctx->chunk_index < ctx->num_alloc_chunks);
900 ctx->chunk_csizes[ctx->chunk_index++] = csize;
902 /* If writing a pipable WIM, before the chunk data write a chunk
903 * header that provides the compressed chunk size. */
904 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
905 struct pwm_chunk_hdr chunk_hdr = {
906 .compressed_size = cpu_to_le32(csize),
908 ret = full_write(ctx->out_fd, &chunk_hdr,
915 /* Write the chunk data. */
916 ret = full_write(ctx->out_fd, cchunk, csize);
920 ctx->cur_write_blob_offset += usize;
922 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
923 /* Wrote chunk in solid mode. It may have finished multiple
925 struct blob_descriptor *next_blob;
927 while (blob && ctx->cur_write_blob_offset >= blob->size) {
929 ctx->cur_write_blob_offset -= blob->size;
931 if (ctx->cur_write_blob_offset)
932 next_blob = list_entry(blob->write_blobs_list.next,
933 struct blob_descriptor,
938 ret = done_with_blob(blob, ctx);
941 list_move_tail(&blob->write_blobs_list, &ctx->blobs_in_solid_resource);
942 completed_blob_count++;
947 /* Wrote chunk in non-solid mode. It may have finished a
949 if (ctx->cur_write_blob_offset == blob->size) {
951 wimlib_assert(ctx->cur_write_blob_offset ==
952 ctx->cur_write_res_size);
954 ret = end_write_resource(ctx, &blob->out_reshdr);
958 blob->out_reshdr.flags = reshdr_flags_for_blob(blob);
959 if (ctx->compressor != NULL)
960 blob->out_reshdr.flags |= WIM_RESHDR_FLAG_COMPRESSED;
962 ret = maybe_rewrite_blob_uncompressed(ctx, blob);
966 wimlib_assert(blob->out_reshdr.uncompressed_size == blob->size);
968 ctx->cur_write_blob_offset = 0;
970 ret = done_with_blob(blob, ctx);
973 list_del(&blob->write_blobs_list);
974 completed_blob_count++;
978 return do_write_blobs_progress(&ctx->progress_data, usize, csize,
979 completed_blob_count, false);
982 ERROR_WITH_ERRNO("Error writing chunk data to WIM file");
987 prepare_chunk_buffer(struct write_blobs_ctx *ctx)
989 /* While we are unable to get a new chunk buffer due to too many chunks
990 * already outstanding, retrieve and write the next compressed chunk. */
991 while (!(ctx->cur_chunk_buf =
992 ctx->compressor->get_chunk_buffer(ctx->compressor)))
1000 bret = ctx->compressor->get_compression_result(ctx->compressor,
1004 wimlib_assert(bret);
1006 ret = write_chunk(ctx, cchunk, csize, usize);
1013 /* Process the next chunk of data to be written to a WIM resource. */
1015 write_blob_process_chunk(const struct blob_descriptor *blob, u64 offset,
1016 const void *chunk, size_t size, void *_ctx)
1018 struct write_blobs_ctx *ctx = _ctx;
1020 const u8 *chunkptr, *chunkend;
1022 wimlib_assert(size != 0);
1024 if (ctx->compressor == NULL) {
1025 /* Write chunk uncompressed. */
1026 ret = write_chunk(ctx, chunk, size, size);
1032 /* Submit the chunk for compression, but take into account that the
1033 * @size the chunk was provided in may not correspond to the
1034 * @out_chunk_size being used for compression. */
1036 chunkend = chunkptr + size;
1038 size_t needed_chunk_size;
1039 size_t bytes_consumed;
1041 if (!ctx->cur_chunk_buf) {
1042 ret = prepare_chunk_buffer(ctx);
1047 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
1048 needed_chunk_size = ctx->out_chunk_size;
1050 needed_chunk_size = min(ctx->out_chunk_size,
1051 ctx->cur_chunk_buf_filled +
1052 (blob->size - offset));
1055 bytes_consumed = min(chunkend - chunkptr,
1056 needed_chunk_size - ctx->cur_chunk_buf_filled);
1058 memcpy(&ctx->cur_chunk_buf[ctx->cur_chunk_buf_filled],
1059 chunkptr, bytes_consumed);
1061 chunkptr += bytes_consumed;
1062 offset += bytes_consumed;
1063 ctx->cur_chunk_buf_filled += bytes_consumed;
1065 if (ctx->cur_chunk_buf_filled == needed_chunk_size) {
1066 ctx->compressor->signal_chunk_filled(ctx->compressor,
1067 ctx->cur_chunk_buf_filled);
1068 ctx->cur_chunk_buf = NULL;
1069 ctx->cur_chunk_buf_filled = 0;
1071 } while (chunkptr != chunkend);
1075 /* Finish processing a blob for writing. It may not have been completely
1076 * written yet, as the chunk_compressor implementation may still have chunks
1077 * buffered or being compressed. */
1079 write_blob_end_read(struct blob_descriptor *blob, int status, void *_ctx)
1081 struct write_blobs_ctx *ctx = _ctx;
1083 if (!blob->will_be_in_output_wim) {
1084 /* The blob was a duplicate. Now that its data has finished
1085 * being read, it is being discarded in favor of the duplicate
1086 * entry. It therefore is no longer needed, and we can fire the
1087 * DONE_WITH_FILE callback because the file will not be read
1090 * Note: we can't yet fire DONE_WITH_FILE for non-duplicate
1091 * blobs, since it needs to be possible to re-read the file if
1092 * it does not compress to less than its original size. */
1094 status = done_with_blob(blob, ctx);
1095 free_blob_descriptor(blob);
1096 } else if (!status && blob->unhashed && ctx->blob_table != NULL) {
1097 /* The blob was not a duplicate and was previously unhashed.
1098 * Since we passed COMPUTE_MISSING_BLOB_HASHES to
1099 * read_blob_list(), blob->hash is now computed and valid. So
1100 * turn this blob into a "hashed" blob. */
1101 list_del(&blob->unhashed_list);
1102 blob_table_insert(ctx->blob_table, blob);
1109 * Compute statistics about a list of blobs that will be written.
1111 * Assumes the blobs are sorted such that all blobs located in each distinct WIM
1112 * (specified by WIMStruct) are together.
1114 * For compactions, also verify that there are no overlapping resources. This
1115 * really should be checked earlier, but for now it's easiest to check here.
1118 compute_blob_list_stats(struct list_head *blob_list,
1119 struct write_blobs_ctx *ctx)
1121 struct blob_descriptor *blob;
1122 u64 total_bytes = 0;
1124 u64 total_parts = 0;
1125 WIMStruct *prev_wim_part = NULL;
1126 const struct wim_resource_descriptor *prev_rdesc = NULL;
1128 list_for_each_entry(blob, blob_list, write_blobs_list) {
1130 total_bytes += blob->size;
1131 if (blob->blob_location == BLOB_IN_WIM) {
1132 const struct wim_resource_descriptor *rdesc = blob->rdesc;
1133 WIMStruct *wim = rdesc->wim;
1135 if (prev_wim_part != wim) {
1136 prev_wim_part = wim;
1139 if (unlikely(wim->being_compacted) && rdesc != prev_rdesc) {
1140 if (prev_rdesc != NULL &&
1141 rdesc->offset_in_wim <
1142 prev_rdesc->offset_in_wim +
1143 prev_rdesc->size_in_wim)
1145 WARNING("WIM file contains overlapping "
1146 "resources! Compaction is not "
1148 return WIMLIB_ERR_RESOURCE_ORDER;
1154 ctx->progress_data.progress.write_streams.total_bytes = total_bytes;
1155 ctx->progress_data.progress.write_streams.total_streams = num_blobs;
1156 ctx->progress_data.progress.write_streams.completed_bytes = 0;
1157 ctx->progress_data.progress.write_streams.completed_streams = 0;
1158 ctx->progress_data.progress.write_streams.compression_type = ctx->out_ctype;
1159 ctx->progress_data.progress.write_streams.total_parts = total_parts;
1160 ctx->progress_data.progress.write_streams.completed_parts = 0;
1161 ctx->progress_data.next_progress = 0;
1165 /* Find blobs in @blob_list that can be copied to the output WIM in raw form
1166 * rather than compressed. Delete these blobs from @blob_list and move them to
1167 * @raw_copy_blobs. Return the total uncompressed size of the blobs that need
1168 * to be compressed. */
1170 find_raw_copy_blobs(struct list_head *blob_list, int write_resource_flags,
1171 int out_ctype, u32 out_chunk_size,
1172 struct list_head *raw_copy_blobs)
1174 struct blob_descriptor *blob, *tmp;
1175 u64 num_nonraw_bytes = 0;
1177 INIT_LIST_HEAD(raw_copy_blobs);
1179 /* Initialize temporary raw_copy_ok flag. */
1180 list_for_each_entry(blob, blob_list, write_blobs_list)
1181 if (blob->blob_location == BLOB_IN_WIM)
1182 blob->rdesc->raw_copy_ok = 0;
1184 list_for_each_entry_safe(blob, tmp, blob_list, write_blobs_list) {
1185 if (can_raw_copy(blob, write_resource_flags,
1186 out_ctype, out_chunk_size))
1188 blob->rdesc->raw_copy_ok = 1;
1189 list_move_tail(&blob->write_blobs_list, raw_copy_blobs);
1191 num_nonraw_bytes += blob->size;
1195 return num_nonraw_bytes;
1198 /* Copy a raw compressed resource located in another WIM file to the WIM file
1201 write_raw_copy_resource(struct wim_resource_descriptor *in_rdesc,
1202 struct filedes *out_fd)
1204 u64 cur_read_offset;
1205 u64 end_read_offset;
1206 u8 buf[BUFFER_SIZE];
1207 size_t bytes_to_read;
1209 struct filedes *in_fd;
1210 struct blob_descriptor *blob;
1211 u64 out_offset_in_wim;
1213 /* Copy the raw data. */
1214 cur_read_offset = in_rdesc->offset_in_wim;
1215 end_read_offset = cur_read_offset + in_rdesc->size_in_wim;
1217 out_offset_in_wim = out_fd->offset;
1219 if (in_rdesc->is_pipable) {
1220 if (cur_read_offset < sizeof(struct pwm_blob_hdr))
1221 return WIMLIB_ERR_INVALID_PIPABLE_WIM;
1222 cur_read_offset -= sizeof(struct pwm_blob_hdr);
1223 out_offset_in_wim += sizeof(struct pwm_blob_hdr);
1225 in_fd = &in_rdesc->wim->in_fd;
1226 wimlib_assert(cur_read_offset != end_read_offset);
1228 if (likely(!in_rdesc->wim->being_compacted) ||
1229 in_rdesc->offset_in_wim > out_fd->offset) {
1231 bytes_to_read = min(sizeof(buf),
1232 end_read_offset - cur_read_offset);
1234 ret = full_pread(in_fd, buf, bytes_to_read,
1237 ERROR_WITH_ERRNO("Error reading raw data "
1242 ret = full_write(out_fd, buf, bytes_to_read);
1244 ERROR_WITH_ERRNO("Error writing raw data "
1249 cur_read_offset += bytes_to_read;
1251 } while (cur_read_offset != end_read_offset);
1253 /* Optimization: the WIM file is being compacted and the
1254 * resource being written is already in the desired location.
1255 * Skip over the data instead of re-writing it. */
1257 /* Due the earlier check for overlapping resources, it should
1258 * never be the case that we already overwrote the resource. */
1259 wimlib_assert(!(in_rdesc->offset_in_wim < out_fd->offset));
1261 if (-1 == filedes_seek(out_fd, out_fd->offset + in_rdesc->size_in_wim))
1262 return WIMLIB_ERR_WRITE;
1265 list_for_each_entry(blob, &in_rdesc->blob_list, rdesc_node) {
1266 if (blob->will_be_in_output_wim) {
1267 blob_set_out_reshdr_for_reuse(blob);
1268 if (in_rdesc->flags & WIM_RESHDR_FLAG_SOLID)
1269 blob->out_res_offset_in_wim = out_offset_in_wim;
1271 blob->out_reshdr.offset_in_wim = out_offset_in_wim;
1278 /* Copy a list of raw compressed resources located in other WIM file(s) to the
1279 * WIM file being written. */
1281 write_raw_copy_resources(struct list_head *raw_copy_blobs,
1282 struct filedes *out_fd,
1283 struct write_blobs_progress_data *progress_data)
1285 struct blob_descriptor *blob;
1288 list_for_each_entry(blob, raw_copy_blobs, write_blobs_list)
1289 blob->rdesc->raw_copy_ok = 1;
1291 list_for_each_entry(blob, raw_copy_blobs, write_blobs_list) {
1292 u64 compressed_size = 0;
1294 if (blob->rdesc->raw_copy_ok) {
1295 /* Write each solid resource only one time. */
1296 ret = write_raw_copy_resource(blob->rdesc, out_fd);
1299 blob->rdesc->raw_copy_ok = 0;
1300 compressed_size = blob->rdesc->size_in_wim;
1302 ret = do_write_blobs_progress(progress_data, blob->size,
1303 compressed_size, 1, false);
1310 /* Wait for and write all chunks pending in the compressor. */
1312 finish_remaining_chunks(struct write_blobs_ctx *ctx)
1319 if (ctx->compressor == NULL)
1322 if (ctx->cur_chunk_buf_filled != 0) {
1323 ctx->compressor->signal_chunk_filled(ctx->compressor,
1324 ctx->cur_chunk_buf_filled);
1327 while (ctx->compressor->get_compression_result(ctx->compressor, &cdata,
1330 ret = write_chunk(ctx, cdata, csize, usize);
1338 validate_blob_list(struct list_head *blob_list)
1340 struct blob_descriptor *blob;
1342 list_for_each_entry(blob, blob_list, write_blobs_list) {
1343 wimlib_assert(blob->will_be_in_output_wim);
1344 wimlib_assert(blob->size != 0);
1349 init_done_with_file_info(struct list_head *blob_list)
1351 struct blob_descriptor *blob;
1353 list_for_each_entry(blob, blob_list, write_blobs_list) {
1354 if (blob_is_in_file(blob)) {
1355 blob->file_inode->i_num_remaining_streams = 0;
1356 blob->may_send_done_with_file = 1;
1358 blob->may_send_done_with_file = 0;
1362 list_for_each_entry(blob, blob_list, write_blobs_list)
1363 if (blob->may_send_done_with_file)
1364 blob->file_inode->i_num_remaining_streams++;
1368 * Write a list of blobs to the output WIM file.
1371 * The list of blobs to write, specified by a list of 'struct blob_descriptor' linked
1372 * by the 'write_blobs_list' member.
1375 * The file descriptor, opened for writing, to which to write the blobs.
1377 * @write_resource_flags
1378 * Flags to modify how the blobs are written:
1380 * WRITE_RESOURCE_FLAG_RECOMPRESS:
1381 * Force compression of all resources, even if they could otherwise
1382 * be re-used by copying the raw data, due to being located in a WIM
1383 * file with compatible compression parameters.
1385 * WRITE_RESOURCE_FLAG_PIPABLE:
1386 * Write the resources in the wimlib-specific pipable format, and
1387 * furthermore do so in such a way that no seeking backwards in
1388 * @out_fd will be performed (so it may be a pipe).
1390 * WRITE_RESOURCE_FLAG_SOLID:
1391 * Combine all the blobs into a single resource rather than writing
1392 * them in separate resources. This flag is only valid if the WIM
1393 * version number has been, or will be, set to WIM_VERSION_SOLID.
1394 * This flag may not be combined with WRITE_RESOURCE_FLAG_PIPABLE.
1397 * Compression format to use in the output resources, specified as one of
1398 * the WIMLIB_COMPRESSION_TYPE_* constants. WIMLIB_COMPRESSION_TYPE_NONE
1402 * Compression chunk size to use in the output resources. It must be a
1403 * valid chunk size for the specified compression format @out_ctype, unless
1404 * @out_ctype is WIMLIB_COMPRESSION_TYPE_NONE, in which case this parameter
1408 * Number of threads to use to compress data. If 0, a default number of
1409 * threads will be chosen. The number of threads still may be decreased
1410 * from the specified value if insufficient memory is detected.
1413 * If on-the-fly deduplication of unhashed blobs is desired, this parameter
1414 * must be pointer to the blob table for the WIMStruct on whose behalf the
1415 * blobs are being written. Otherwise, this parameter can be NULL.
1418 * If on-the-fly deduplication of unhashed blobs is desired, this parameter
1419 * can be a pointer to a context for blob filtering used to detect whether
1420 * the duplicate blob has been hard-filtered or not. If no blobs are
1421 * hard-filtered or no blobs are unhashed, this parameter can be NULL.
1423 * This function will write the blobs in @blob_list to resources in
1424 * consecutive positions in the output WIM file, or to a single solid resource
1425 * if WRITE_RESOURCE_FLAG_SOLID was specified in @write_resource_flags. In both
1426 * cases, the @out_reshdr of the `struct blob_descriptor' for each blob written will be
1427 * updated to specify its location, size, and flags in the output WIM. In the
1428 * solid resource case, WIM_RESHDR_FLAG_SOLID will be set in the @flags field of
1429 * each @out_reshdr, and furthermore @out_res_offset_in_wim and
1430 * @out_res_size_in_wim of each @out_reshdr will be set to the offset and size,
1431 * respectively, in the output WIM of the solid resource containing the
1432 * corresponding blob.
1434 * Each of the blobs to write may be in any location supported by the
1435 * resource-handling code (specifically, read_blob_list()), such as the contents
1436 * of external file that has been logically added to the output WIM, or a blob
1437 * in another WIM file that has been imported, or even a blob in the "same" WIM
1438 * file of which a modified copy is being written. In the case that a blob is
1439 * already in a WIM file and uses compatible compression parameters, by default
1440 * this function will re-use the raw data instead of decompressing it, then
1441 * recompressing it; however, with WRITE_RESOURCE_FLAG_RECOMPRESS
1442 * specified in @write_resource_flags, this is not done.
1444 * As a further requirement, this function requires that the
1445 * @will_be_in_output_wim member be set to 1 on all blobs in @blob_list as well
1446 * as any other blobs not in @blob_list that will be in the output WIM file, but
1447 * set to 0 on any other blobs in the output WIM's blob table or sharing a solid
1448 * resource with a blob in @blob_list. Still furthermore, if on-the-fly
1449 * deduplication of blobs is possible, then all blobs in @blob_list must also be
1450 * linked by @blob_table_list along with any other blobs that have
1451 * @will_be_in_output_wim set.
1453 * This function handles on-the-fly deduplication of blobs for which SHA-1
1454 * message digests have not yet been calculated. Such blobs may or may not need
1455 * to be written. If @blob_table is non-NULL, then each blob in @blob_list that
1456 * has @unhashed set but not @unique_size set is checksummed immediately before
1457 * it would otherwise be read for writing in order to determine if it is
1458 * identical to another blob already being written or one that would be filtered
1459 * out of the output WIM using blob_filtered() with the context @filter_ctx.
1460 * Each such duplicate blob will be removed from @blob_list, its reference count
1461 * transferred to the pre-existing duplicate blob, its memory freed, and will
1462 * not be written. Alternatively, if a blob in @blob_list is a duplicate with
1463 * any blob in @blob_table that has not been marked for writing or would not be
1464 * hard-filtered, it is freed and the pre-existing duplicate is written instead,
1465 * taking ownership of the reference count and slot in the @blob_table_list.
1467 * Returns 0 if every blob was either written successfully or did not need to be
1468 * written; otherwise returns a non-zero error code.
1471 write_blob_list(struct list_head *blob_list,
1472 struct filedes *out_fd,
1473 int write_resource_flags,
1476 unsigned num_threads,
1477 struct blob_table *blob_table,
1478 struct filter_context *filter_ctx,
1479 wimlib_progress_func_t progfunc,
1483 struct write_blobs_ctx ctx;
1484 struct list_head raw_copy_blobs;
1485 u64 num_nonraw_bytes;
1487 wimlib_assert((write_resource_flags &
1488 (WRITE_RESOURCE_FLAG_SOLID |
1489 WRITE_RESOURCE_FLAG_PIPABLE)) !=
1490 (WRITE_RESOURCE_FLAG_SOLID |
1491 WRITE_RESOURCE_FLAG_PIPABLE));
1493 validate_blob_list(blob_list);
1495 if (list_empty(blob_list))
1498 /* If needed, set auxiliary information so that we can detect when the
1499 * library has finished using each external file. */
1500 if (unlikely(write_resource_flags & WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE))
1501 init_done_with_file_info(blob_list);
1503 memset(&ctx, 0, sizeof(ctx));
1505 ctx.out_fd = out_fd;
1506 ctx.blob_table = blob_table;
1507 ctx.out_ctype = out_ctype;
1508 ctx.out_chunk_size = out_chunk_size;
1509 ctx.write_resource_flags = write_resource_flags;
1510 ctx.filter_ctx = filter_ctx;
1513 * We normally sort the blobs to write by a "sequential" order that is
1514 * optimized for reading. But when using solid compression, we instead
1515 * sort the blobs by file extension and file name (when applicable; and
1516 * we don't do this for blobs from solid resources) so that similar
1517 * files are grouped together, which improves the compression ratio.
1518 * This is somewhat of a hack since a blob does not necessarily
1519 * correspond one-to-one with a filename, nor is there any guarantee
1520 * that two files with similar names or extensions are actually similar
1521 * in content. A potential TODO is to sort the blobs based on some
1522 * measure of similarity of their actual contents.
1525 ret = sort_blob_list_by_sequential_order(blob_list,
1526 offsetof(struct blob_descriptor,
1531 ret = compute_blob_list_stats(blob_list, &ctx);
1535 if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID_SORT) {
1536 ret = sort_blob_list_for_solid_compression(blob_list);
1538 WARNING("Failed to sort blobs for solid compression. Continuing anyways.");
1541 ctx.progress_data.progfunc = progfunc;
1542 ctx.progress_data.progctx = progctx;
1544 num_nonraw_bytes = find_raw_copy_blobs(blob_list, write_resource_flags,
1545 out_ctype, out_chunk_size,
1548 /* Unless no data needs to be compressed, allocate a chunk_compressor to
1549 * do compression. There are serial and parallel implementations of the
1550 * chunk_compressor interface. We default to parallel using the
1551 * specified number of threads, unless the upper bound on the number
1552 * bytes needing to be compressed is less than a heuristic value. */
1553 if (num_nonraw_bytes != 0 && out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) {
1554 #ifdef ENABLE_MULTITHREADED_COMPRESSION
1555 if (num_nonraw_bytes > max(2000000, out_chunk_size)) {
1556 ret = new_parallel_chunk_compressor(out_ctype,
1561 WARNING("Couldn't create parallel chunk compressor: %"TS".\n"
1562 " Falling back to single-threaded compression.",
1563 wimlib_get_error_string(ret));
1568 if (ctx.compressor == NULL) {
1569 ret = new_serial_chunk_compressor(out_ctype, out_chunk_size,
1572 goto out_destroy_context;
1577 ctx.progress_data.progress.write_streams.num_threads = ctx.compressor->num_threads;
1579 ctx.progress_data.progress.write_streams.num_threads = 1;
1581 ret = call_progress(ctx.progress_data.progfunc,
1582 WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
1583 &ctx.progress_data.progress,
1584 ctx.progress_data.progctx);
1586 goto out_destroy_context;
1588 /* Copy any compressed resources for which the raw data can be reused
1589 * without decompression. */
1590 ret = write_raw_copy_resources(&raw_copy_blobs, ctx.out_fd,
1591 &ctx.progress_data);
1593 if (ret || num_nonraw_bytes == 0)
1594 goto out_destroy_context;
1596 INIT_LIST_HEAD(&ctx.blobs_being_compressed);
1598 if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
1600 INIT_LIST_HEAD(&ctx.blobs_in_solid_resource);
1602 ret = begin_write_resource(&ctx, num_nonraw_bytes);
1604 goto out_destroy_context;
1607 /* Read the list of blobs needing to be compressed, using the specified
1608 * callbacks to execute processing of the data. */
1610 struct read_blob_callbacks cbs = {
1611 .begin_blob = write_blob_begin_read,
1612 .continue_blob = write_blob_process_chunk,
1613 .end_blob = write_blob_end_read,
1617 ret = read_blob_list(blob_list,
1618 offsetof(struct blob_descriptor, write_blobs_list),
1620 BLOB_LIST_ALREADY_SORTED |
1621 VERIFY_BLOB_HASHES |
1622 COMPUTE_MISSING_BLOB_HASHES);
1625 goto out_destroy_context;
1627 ret = finish_remaining_chunks(&ctx);
1629 goto out_destroy_context;
1631 if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
1632 struct wim_reshdr reshdr;
1633 struct blob_descriptor *blob;
1636 ret = end_write_resource(&ctx, &reshdr);
1638 goto out_destroy_context;
1641 list_for_each_entry(blob, &ctx.blobs_in_solid_resource, write_blobs_list) {
1642 blob->out_reshdr.size_in_wim = blob->size;
1643 blob->out_reshdr.flags = reshdr_flags_for_blob(blob) |
1644 WIM_RESHDR_FLAG_SOLID;
1645 blob->out_reshdr.uncompressed_size = 0;
1646 blob->out_reshdr.offset_in_wim = offset_in_res;
1647 blob->out_res_offset_in_wim = reshdr.offset_in_wim;
1648 blob->out_res_size_in_wim = reshdr.size_in_wim;
1649 blob->out_res_uncompressed_size = reshdr.uncompressed_size;
1650 offset_in_res += blob->size;
1652 wimlib_assert(offset_in_res == reshdr.uncompressed_size);
1655 out_destroy_context:
1656 FREE(ctx.chunk_csizes);
1658 ctx.compressor->destroy(ctx.compressor);
1664 write_file_data_blobs(WIMStruct *wim,
1665 struct list_head *blob_list,
1667 unsigned num_threads,
1668 struct filter_context *filter_ctx)
1672 int write_resource_flags;
1674 write_resource_flags = write_flags_to_resource_flags(write_flags);
1676 if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
1677 out_chunk_size = wim->out_solid_chunk_size;
1678 out_ctype = wim->out_solid_compression_type;
1680 out_chunk_size = wim->out_chunk_size;
1681 out_ctype = wim->out_compression_type;
1684 return write_blob_list(blob_list,
1686 write_resource_flags,
1696 /* Write the contents of the specified blob as a WIM resource. */
1698 write_wim_resource(struct blob_descriptor *blob,
1699 struct filedes *out_fd,
1702 int write_resource_flags)
1704 LIST_HEAD(blob_list);
1705 list_add(&blob->write_blobs_list, &blob_list);
1706 blob->will_be_in_output_wim = 1;
1707 return write_blob_list(&blob_list,
1709 write_resource_flags & ~WRITE_RESOURCE_FLAG_SOLID,
1719 /* Write the contents of the specified buffer as a WIM resource. */
1721 write_wim_resource_from_buffer(const void *buf,
1724 struct filedes *out_fd,
1727 struct wim_reshdr *out_reshdr,
1729 int write_resource_flags)
1732 struct blob_descriptor blob;
1734 if (unlikely(buf_size == 0)) {
1735 zero_reshdr(out_reshdr);
1737 copy_hash(hash_ret, zero_hash);
1741 blob_set_is_located_in_attached_buffer(&blob, (void *)buf, buf_size);
1742 sha1_buffer(buf, buf_size, blob.hash);
1744 blob.is_metadata = is_metadata;
1746 ret = write_wim_resource(&blob, out_fd, out_ctype, out_chunk_size,
1747 write_resource_flags);
1751 copy_reshdr(out_reshdr, &blob.out_reshdr);
1754 copy_hash(hash_ret, blob.hash);
1758 struct blob_size_table {
1759 struct hlist_head *array;
1765 init_blob_size_table(struct blob_size_table *tab, size_t capacity)
1767 tab->array = CALLOC(capacity, sizeof(tab->array[0]));
1768 if (tab->array == NULL)
1769 return WIMLIB_ERR_NOMEM;
1770 tab->num_entries = 0;
1771 tab->capacity = capacity;
1776 destroy_blob_size_table(struct blob_size_table *tab)
1782 blob_size_table_insert(struct blob_descriptor *blob, void *_tab)
1784 struct blob_size_table *tab = _tab;
1786 struct blob_descriptor *same_size_blob;
1788 pos = hash_u64(blob->size) % tab->capacity;
1789 blob->unique_size = 1;
1790 hlist_for_each_entry(same_size_blob, &tab->array[pos], hash_list_2) {
1791 if (same_size_blob->size == blob->size) {
1792 blob->unique_size = 0;
1793 same_size_blob->unique_size = 0;
1798 hlist_add_head(&blob->hash_list_2, &tab->array[pos]);
1803 struct find_blobs_ctx {
1806 struct list_head blob_list;
1807 struct blob_size_table blob_size_tab;
1811 reference_blob_for_write(struct blob_descriptor *blob,
1812 struct list_head *blob_list, u32 nref)
1814 if (!blob->will_be_in_output_wim) {
1815 blob->out_refcnt = 0;
1816 list_add_tail(&blob->write_blobs_list, blob_list);
1817 blob->will_be_in_output_wim = 1;
1819 blob->out_refcnt += nref;
1823 fully_reference_blob_for_write(struct blob_descriptor *blob, void *_blob_list)
1825 struct list_head *blob_list = _blob_list;
1826 blob->will_be_in_output_wim = 0;
1827 reference_blob_for_write(blob, blob_list, blob->refcnt);
1832 inode_find_blobs_to_reference(const struct wim_inode *inode,
1833 const struct blob_table *table,
1834 struct list_head *blob_list)
1836 wimlib_assert(inode->i_nlink > 0);
1838 for (unsigned i = 0; i < inode->i_num_streams; i++) {
1839 struct blob_descriptor *blob;
1842 blob = stream_blob(&inode->i_streams[i], table);
1844 reference_blob_for_write(blob, blob_list, inode->i_nlink);
1846 hash = stream_hash(&inode->i_streams[i]);
1847 if (!is_zero_hash(hash))
1848 return blob_not_found_error(inode, hash);
1855 do_blob_set_not_in_output_wim(struct blob_descriptor *blob, void *_ignore)
1857 blob->will_be_in_output_wim = 0;
1862 image_find_blobs_to_reference(WIMStruct *wim)
1864 struct wim_image_metadata *imd;
1865 struct wim_inode *inode;
1866 struct blob_descriptor *blob;
1867 struct list_head *blob_list;
1870 imd = wim_get_current_image_metadata(wim);
1872 image_for_each_unhashed_blob(blob, imd)
1873 blob->will_be_in_output_wim = 0;
1875 blob_list = wim->private;
1876 image_for_each_inode(inode, imd) {
1877 ret = inode_find_blobs_to_reference(inode,
1887 prepare_unfiltered_list_of_blobs_in_output_wim(WIMStruct *wim,
1890 struct list_head *blob_list_ret)
1894 INIT_LIST_HEAD(blob_list_ret);
1896 if (blobs_ok && (image == WIMLIB_ALL_IMAGES ||
1897 (image == 1 && wim->hdr.image_count == 1)))
1899 /* Fast case: Assume that all blobs are being written and that
1900 * the reference counts are correct. */
1901 struct blob_descriptor *blob;
1902 struct wim_image_metadata *imd;
1905 for_blob_in_table(wim->blob_table,
1906 fully_reference_blob_for_write,
1909 for (i = 0; i < wim->hdr.image_count; i++) {
1910 imd = wim->image_metadata[i];
1911 image_for_each_unhashed_blob(blob, imd)
1912 fully_reference_blob_for_write(blob, blob_list_ret);
1915 /* Slow case: Walk through the images being written and
1916 * determine the blobs referenced. */
1917 for_blob_in_table(wim->blob_table,
1918 do_blob_set_not_in_output_wim, NULL);
1919 wim->private = blob_list_ret;
1920 ret = for_image(wim, image, image_find_blobs_to_reference);
1928 struct insert_other_if_hard_filtered_ctx {
1929 struct blob_size_table *tab;
1930 struct filter_context *filter_ctx;
1934 insert_other_if_hard_filtered(struct blob_descriptor *blob, void *_ctx)
1936 struct insert_other_if_hard_filtered_ctx *ctx = _ctx;
1938 if (!blob->will_be_in_output_wim &&
1939 blob_hard_filtered(blob, ctx->filter_ctx))
1940 blob_size_table_insert(blob, ctx->tab);
1945 determine_blob_size_uniquity(struct list_head *blob_list,
1946 struct blob_table *lt,
1947 struct filter_context *filter_ctx)
1950 struct blob_size_table tab;
1951 struct blob_descriptor *blob;
1953 ret = init_blob_size_table(&tab, 9001);
1957 if (may_hard_filter_blobs(filter_ctx)) {
1958 struct insert_other_if_hard_filtered_ctx ctx = {
1960 .filter_ctx = filter_ctx,
1962 for_blob_in_table(lt, insert_other_if_hard_filtered, &ctx);
1965 list_for_each_entry(blob, blob_list, write_blobs_list)
1966 blob_size_table_insert(blob, &tab);
1968 destroy_blob_size_table(&tab);
1973 filter_blob_list_for_write(struct list_head *blob_list,
1974 struct filter_context *filter_ctx)
1976 struct blob_descriptor *blob, *tmp;
1978 list_for_each_entry_safe(blob, tmp, blob_list, write_blobs_list) {
1979 int status = blob_filtered(blob, filter_ctx);
1986 /* Soft filtered. */
1988 /* Hard filtered. */
1989 blob->will_be_in_output_wim = 0;
1990 list_del(&blob->blob_table_list);
1992 list_del(&blob->write_blobs_list);
1998 * prepare_blob_list_for_write() -
2000 * Prepare the list of blobs to write for writing a WIM containing the specified
2001 * image(s) with the specified write flags.
2004 * The WIMStruct on whose behalf the write is occurring.
2007 * Image(s) from the WIM to write; may be WIMLIB_ALL_IMAGES.
2010 * WIMLIB_WRITE_FLAG_* flags for the write operation:
2012 * STREAMS_OK: For writes of all images, assume that all blobs in the blob
2013 * table of @wim and the per-image lists of unhashed blobs should be taken
2014 * as-is, and image metadata should not be searched for references. This
2015 * does not exclude filtering with APPEND and SKIP_EXTERNAL_WIMS, below.
2017 * APPEND: Blobs already present in @wim shall not be returned in
2020 * SKIP_EXTERNAL_WIMS: Blobs already present in a WIM file, but not @wim,
2021 * shall be returned in neither @blob_list_ret nor @blob_table_list_ret.
2024 * List of blobs, linked by write_blobs_list, that need to be written will
2027 * Note that this function assumes that unhashed blobs will be written; it
2028 * does not take into account that they may become duplicates when actually
2031 * @blob_table_list_ret
2032 * List of blobs, linked by blob_table_list, that need to be included in
2033 * the WIM's blob table will be returned here. This will be a superset of
2034 * the blobs in @blob_list_ret.
2036 * This list will be a proper superset of @blob_list_ret if and only if
2037 * WIMLIB_WRITE_FLAG_APPEND was specified in @write_flags and some of the
2038 * blobs that would otherwise need to be written were already located in
2041 * All blobs in this list will have @out_refcnt set to the number of
2042 * references to the blob in the output WIM. If
2043 * WIMLIB_WRITE_FLAG_STREAMS_OK was specified in @write_flags, @out_refcnt
2044 * may be as low as 0.
2047 * A context for queries of blob filter status with blob_filtered() is
2048 * returned in this location.
2050 * In addition, @will_be_in_output_wim will be set to 1 in all blobs inserted
2051 * into @blob_table_list_ret and to 0 in all blobs in the blob table of @wim not
2052 * inserted into @blob_table_list_ret.
2054 * Still furthermore, @unique_size will be set to 1 on all blobs in
2055 * @blob_list_ret that have unique size among all blobs in @blob_list_ret and
2056 * among all blobs in the blob table of @wim that are ineligible for being
2057 * written due to filtering.
2059 * Returns 0 on success; nonzero on read error, memory allocation error, or
2063 prepare_blob_list_for_write(WIMStruct *wim, int image,
2065 struct list_head *blob_list_ret,
2066 struct list_head *blob_table_list_ret,
2067 struct filter_context *filter_ctx_ret)
2070 struct blob_descriptor *blob;
2072 filter_ctx_ret->write_flags = write_flags;
2073 filter_ctx_ret->wim = wim;
2075 ret = prepare_unfiltered_list_of_blobs_in_output_wim(
2078 write_flags & WIMLIB_WRITE_FLAG_STREAMS_OK,
2083 INIT_LIST_HEAD(blob_table_list_ret);
2084 list_for_each_entry(blob, blob_list_ret, write_blobs_list)
2085 list_add_tail(&blob->blob_table_list, blob_table_list_ret);
2087 ret = determine_blob_size_uniquity(blob_list_ret, wim->blob_table,
2092 if (may_filter_blobs(filter_ctx_ret))
2093 filter_blob_list_for_write(blob_list_ret, filter_ctx_ret);
2099 write_file_data(WIMStruct *wim, int image, int write_flags,
2100 unsigned num_threads,
2101 struct list_head *blob_list_override,
2102 struct list_head *blob_table_list_ret)
2105 struct list_head _blob_list;
2106 struct list_head *blob_list;
2107 struct blob_descriptor *blob;
2108 struct filter_context _filter_ctx;
2109 struct filter_context *filter_ctx;
2111 if (blob_list_override == NULL) {
2112 /* Normal case: prepare blob list from image(s) being written.
2114 blob_list = &_blob_list;
2115 filter_ctx = &_filter_ctx;
2116 ret = prepare_blob_list_for_write(wim, image, write_flags,
2118 blob_table_list_ret,
2123 /* Currently only as a result of wimlib_split() being called:
2124 * use blob list already explicitly provided. Use existing
2125 * reference counts. */
2126 blob_list = blob_list_override;
2128 INIT_LIST_HEAD(blob_table_list_ret);
2129 list_for_each_entry(blob, blob_list, write_blobs_list) {
2130 blob->out_refcnt = blob->refcnt;
2131 blob->will_be_in_output_wim = 1;
2132 blob->unique_size = 0;
2133 list_add_tail(&blob->blob_table_list, blob_table_list_ret);
2137 return write_file_data_blobs(wim,
2145 write_metadata_resources(WIMStruct *wim, int image, int write_flags)
2150 int write_resource_flags;
2152 if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)
2155 write_resource_flags = write_flags_to_resource_flags(write_flags);
2157 write_resource_flags &= ~WRITE_RESOURCE_FLAG_SOLID;
2159 ret = call_progress(wim->progfunc,
2160 WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN,
2161 NULL, wim->progctx);
2165 if (image == WIMLIB_ALL_IMAGES) {
2167 end_image = wim->hdr.image_count;
2169 start_image = image;
2173 for (int i = start_image; i <= end_image; i++) {
2174 struct wim_image_metadata *imd;
2176 imd = wim->image_metadata[i - 1];
2177 if (is_image_dirty(imd)) {
2178 /* The image was modified from the original, or was
2179 * newly added, so we have to build and write a new
2180 * metadata resource. */
2181 ret = write_metadata_resource(wim, i,
2182 write_resource_flags);
2183 } else if (is_image_unchanged_from_wim(imd, wim) &&
2184 (write_flags & (WIMLIB_WRITE_FLAG_UNSAFE_COMPACT |
2185 WIMLIB_WRITE_FLAG_APPEND)))
2187 /* The metadata resource is already in the WIM file.
2188 * For appends, we don't need to write it at all. For
2189 * compactions, we re-write existing metadata resources
2190 * along with the existing file resources, not here. */
2191 if (write_flags & WIMLIB_WRITE_FLAG_APPEND)
2192 blob_set_out_reshdr_for_reuse(imd->metadata_blob);
2195 /* The metadata resource is in a WIM file other than the
2196 * one being written to. We need to rewrite it,
2197 * possibly compressed differently; but rebuilding the
2198 * metadata itself isn't necessary. */
2199 ret = write_wim_resource(imd->metadata_blob,
2201 wim->out_compression_type,
2202 wim->out_chunk_size,
2203 write_resource_flags);
2209 return call_progress(wim->progfunc,
2210 WIMLIB_PROGRESS_MSG_WRITE_METADATA_END,
2211 NULL, wim->progctx);
2215 open_wim_writable(WIMStruct *wim, const tchar *path, int open_flags)
2217 int raw_fd = topen(path, open_flags | O_BINARY, 0644);
2219 ERROR_WITH_ERRNO("Failed to open \"%"TS"\" for writing", path);
2220 return WIMLIB_ERR_OPEN;
2222 filedes_init(&wim->out_fd, raw_fd);
2227 close_wim_writable(WIMStruct *wim, int write_flags)
2231 if (!(write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR))
2232 if (filedes_valid(&wim->out_fd))
2233 if (filedes_close(&wim->out_fd))
2234 ret = WIMLIB_ERR_WRITE;
2235 filedes_invalidate(&wim->out_fd);
2240 cmp_blobs_by_out_rdesc(const void *p1, const void *p2)
2242 const struct blob_descriptor *blob1, *blob2;
2244 blob1 = *(const struct blob_descriptor**)p1;
2245 blob2 = *(const struct blob_descriptor**)p2;
2247 if (blob1->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
2248 if (blob2->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
2249 if (blob1->out_res_offset_in_wim != blob2->out_res_offset_in_wim)
2250 return cmp_u64(blob1->out_res_offset_in_wim,
2251 blob2->out_res_offset_in_wim);
2256 if (blob2->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID)
2259 return cmp_u64(blob1->out_reshdr.offset_in_wim,
2260 blob2->out_reshdr.offset_in_wim);
2264 write_blob_table(WIMStruct *wim, int image, int write_flags,
2265 struct list_head *blob_table_list)
2269 /* Set output resource metadata for blobs already present in WIM. */
2270 if (write_flags & WIMLIB_WRITE_FLAG_APPEND) {
2271 struct blob_descriptor *blob;
2272 list_for_each_entry(blob, blob_table_list, blob_table_list) {
2273 if (blob->blob_location == BLOB_IN_WIM &&
2274 blob->rdesc->wim == wim)
2276 blob_set_out_reshdr_for_reuse(blob);
2281 ret = sort_blob_list(blob_table_list,
2282 offsetof(struct blob_descriptor, blob_table_list),
2283 cmp_blobs_by_out_rdesc);
2287 /* Add entries for metadata resources. */
2288 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)) {
2292 if (image == WIMLIB_ALL_IMAGES) {
2294 end_image = wim->hdr.image_count;
2296 start_image = image;
2300 /* Push metadata blob table entries onto the front of the list
2301 * in reverse order, so that they're written in order.
2303 for (int i = end_image; i >= start_image; i--) {
2304 struct blob_descriptor *metadata_blob;
2306 metadata_blob = wim->image_metadata[i - 1]->metadata_blob;
2307 wimlib_assert(metadata_blob->out_reshdr.flags & WIM_RESHDR_FLAG_METADATA);
2308 metadata_blob->out_refcnt = 1;
2309 list_add(&metadata_blob->blob_table_list, blob_table_list);
2313 return write_blob_table_from_blob_list(blob_table_list,
2315 wim->out_hdr.part_number,
2316 &wim->out_hdr.blob_table_reshdr,
2317 write_flags_to_resource_flags(write_flags));
2321 * Finish writing a WIM file: write the blob table, xml data, and integrity
2322 * table, then overwrite the WIM header.
2324 * The output file descriptor is closed on success, except when writing to a
2325 * user-specified file descriptor (WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR set).
2328 finish_write(WIMStruct *wim, int image, int write_flags,
2329 struct list_head *blob_table_list)
2331 int write_resource_flags;
2332 off_t old_blob_table_end = 0;
2333 struct integrity_table *old_integrity_table = NULL;
2334 off_t new_blob_table_end;
2338 write_resource_flags = write_flags_to_resource_flags(write_flags);
2340 /* In the WIM header, there is room for the resource entry for a
2341 * metadata resource labeled as the "boot metadata". This entry should
2342 * be zeroed out if there is no bootable image (boot_idx 0). Otherwise,
2343 * it should be a copy of the resource entry for the image that is
2344 * marked as bootable. */
2345 if (wim->out_hdr.boot_idx == 0) {
2346 zero_reshdr(&wim->out_hdr.boot_metadata_reshdr);
2348 copy_reshdr(&wim->out_hdr.boot_metadata_reshdr,
2349 &wim->image_metadata[
2350 wim->out_hdr.boot_idx - 1]->metadata_blob->out_reshdr);
2353 /* If appending to a WIM file containing an integrity table, we'd like
2354 * to re-use the information in the old integrity table instead of
2355 * recalculating it. But we might overwrite the old integrity table
2356 * when we expand the XML data. Read it into memory just in case. */
2357 if ((write_flags & (WIMLIB_WRITE_FLAG_APPEND |
2358 WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)) ==
2359 (WIMLIB_WRITE_FLAG_APPEND |
2360 WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)
2361 && wim_has_integrity_table(wim))
2363 old_blob_table_end = wim->hdr.blob_table_reshdr.offset_in_wim +
2364 wim->hdr.blob_table_reshdr.size_in_wim;
2365 (void)read_integrity_table(wim,
2366 old_blob_table_end - WIM_HEADER_DISK_SIZE,
2367 &old_integrity_table);
2368 /* If we couldn't read the old integrity table, we can still
2369 * re-calculate the full integrity table ourselves. Hence the
2370 * ignoring of the return value. */
2373 /* Write blob table if needed. */
2374 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_NEW_BLOBS)) {
2375 ret = write_blob_table(wim, image, write_flags,
2381 /* Write XML data. */
2382 xml_totalbytes = wim->out_fd.offset;
2383 if (write_flags & WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES)
2384 xml_totalbytes = WIM_TOTALBYTES_USE_EXISTING;
2385 ret = write_wim_xml_data(wim, image, xml_totalbytes,
2386 &wim->out_hdr.xml_data_reshdr,
2387 write_resource_flags);
2391 /* Write integrity table if needed. */
2392 if ((write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) &&
2393 wim->out_hdr.blob_table_reshdr.offset_in_wim != 0)
2395 if (write_flags & WIMLIB_WRITE_FLAG_NO_NEW_BLOBS) {
2396 /* The XML data we wrote may have overwritten part of
2397 * the old integrity table, so while calculating the new
2398 * integrity table we should temporarily update the WIM
2399 * header to remove the integrity table reference. */
2400 struct wim_header checkpoint_hdr;
2401 memcpy(&checkpoint_hdr, &wim->out_hdr, sizeof(struct wim_header));
2402 zero_reshdr(&checkpoint_hdr.integrity_table_reshdr);
2403 checkpoint_hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2404 ret = write_wim_header(&checkpoint_hdr, &wim->out_fd, 0);
2409 new_blob_table_end = wim->out_hdr.blob_table_reshdr.offset_in_wim +
2410 wim->out_hdr.blob_table_reshdr.size_in_wim;
2412 ret = write_integrity_table(wim,
2415 old_integrity_table);
2419 /* No integrity table. */
2420 zero_reshdr(&wim->out_hdr.integrity_table_reshdr);
2423 /* Now that all information in the WIM header has been determined, the
2424 * preliminary header written earlier can be overwritten, the header of
2425 * the existing WIM file can be overwritten, or the final header can be
2426 * written to the end of the pipable WIM. */
2427 wim->out_hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2428 if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
2429 ret = write_wim_header(&wim->out_hdr, &wim->out_fd, wim->out_fd.offset);
2431 ret = write_wim_header(&wim->out_hdr, &wim->out_fd, 0);
2435 ret = WIMLIB_ERR_WRITE;
2436 if (unlikely(write_flags & WIMLIB_WRITE_FLAG_UNSAFE_COMPACT)) {
2437 /* Truncate any data the compaction freed up. */
2438 if (ftruncate(wim->out_fd.fd, wim->out_fd.offset) &&
2439 errno != EINVAL) /* allow compaction on untruncatable files,
2440 e.g. block devices */
2442 ERROR_WITH_ERRNO("Failed to truncate the output WIM file");
2447 /* Possibly sync file data to disk before closing. On POSIX systems, it
2448 * is necessary to do this before using rename() to overwrite an
2449 * existing file with a new file. Otherwise, data loss would occur if
2450 * the system is abruptly terminated when the metadata for the rename
2451 * operation has been written to disk, but the new file data has not.
2453 ret = WIMLIB_ERR_WRITE;
2454 if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) {
2455 if (fsync(wim->out_fd.fd)) {
2456 ERROR_WITH_ERRNO("Error syncing data to WIM file");
2461 ret = WIMLIB_ERR_WRITE;
2462 if (close_wim_writable(wim, write_flags)) {
2463 ERROR_WITH_ERRNO("Failed to close the output WIM file");
2469 free_integrity_table(old_integrity_table);
2473 #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK)
2475 /* Set advisory lock on WIM file (if not already done so) */
2477 lock_wim_for_append(WIMStruct *wim)
2479 if (wim->locked_for_append)
2481 if (!flock(wim->in_fd.fd, LOCK_EX | LOCK_NB)) {
2482 wim->locked_for_append = 1;
2485 if (errno != EWOULDBLOCK)
2487 return WIMLIB_ERR_ALREADY_LOCKED;
2490 /* Remove advisory lock on WIM file (if present) */
2492 unlock_wim_for_append(WIMStruct *wim)
2494 if (wim->locked_for_append) {
2495 flock(wim->in_fd.fd, LOCK_UN);
2496 wim->locked_for_append = 0;
2502 * write_pipable_wim():
2504 * Perform the intermediate stages of creating a "pipable" WIM (i.e. a WIM
2505 * capable of being applied from a pipe).
2507 * Pipable WIMs are a wimlib-specific modification of the WIM format such that
2508 * images can be applied from them sequentially when the file data is sent over
2509 * a pipe. In addition, a pipable WIM can be written sequentially to a pipe.
2510 * The modifications made to the WIM format for pipable WIMs are:
2512 * - Magic characters in header are "WLPWM\0\0\0" (wimlib pipable WIM) instead
2513 * of "MSWIM\0\0\0". This lets wimlib know that the WIM is pipable and also
2514 * stops other software from trying to read the file as a normal WIM.
2516 * - The header at the beginning of the file does not contain all the normal
2517 * information; in particular it will have all 0's for the blob table and XML
2518 * data resource entries. This is because this information cannot be
2519 * determined until the blob table and XML data have been written.
2520 * Consequently, wimlib will write the full header at the very end of the
2521 * file. The header at the end, however, is only used when reading the WIM
2522 * from a seekable file (not a pipe).
2524 * - An extra copy of the XML data is placed directly after the header. This
2525 * allows image names and sizes to be determined at an appropriate time when
2526 * reading the WIM from a pipe. This copy of the XML data is ignored if the
2527 * WIM is read from a seekable file (not a pipe).
2529 * - Solid resources are not allowed. Each blob is always stored in its own
2532 * - The format of resources, or blobs, has been modified to allow them to be
2533 * used before the "blob table" has been read. Each blob is prefixed with a
2534 * `struct pwm_blob_hdr' that is basically an abbreviated form of `struct
2535 * blob_descriptor_disk' that only contains the SHA-1 message digest,
2536 * uncompressed blob size, and flags that indicate whether the blob is
2537 * compressed. The data of uncompressed blobs then follows literally, while
2538 * the data of compressed blobs follows in a modified format. Compressed
2539 * blobs do not begin with a chunk table, since the chunk table cannot be
2540 * written until all chunks have been compressed. Instead, each compressed
2541 * chunk is prefixed by a `struct pwm_chunk_hdr' that gives its size.
2542 * Furthermore, the chunk table is written at the end of the resource instead
2543 * of the start. Note: chunk offsets are given in the chunk table as if the
2544 * `struct pwm_chunk_hdr's were not present; also, the chunk table is only
2545 * used if the WIM is being read from a seekable file (not a pipe).
2547 * - Metadata blobs always come before non-metadata blobs. (This does not by
2548 * itself constitute an incompatibility with normal WIMs, since this is valid
2551 * - At least up to the end of the blobs, all components must be packed as
2552 * tightly as possible; there cannot be any "holes" in the WIM. (This does
2553 * not by itself consititute an incompatibility with normal WIMs, since this
2554 * is valid in normal WIMs.)
2556 * Note: the blob table, XML data, and header at the end are not used when
2557 * applying from a pipe. They exist to support functionality such as image
2558 * application and export when the WIM is *not* read from a pipe.
2560 * Layout of pipable WIM:
2562 * ---------+----------+--------------------+----------------+--------------+-----------+--------+
2563 * | Header | XML data | Metadata resources | File resources | Blob table | XML data | Header |
2564 * ---------+----------+--------------------+----------------+--------------+-----------+--------+
2566 * Layout of normal WIM:
2568 * +--------+-----------------------------+-------------------------+
2569 * | Header | File and metadata resources | Blob table | XML data |
2570 * +--------+-----------------------------+-------------------------+
2572 * An optional integrity table can follow the final XML data in both normal and
2573 * pipable WIMs. However, due to implementation details, wimlib currently can
2574 * only include an integrity table in a pipable WIM when writing it to a
2575 * seekable file (not a pipe).
2577 * Do note that since pipable WIMs are not supported by Microsoft's software,
2578 * wimlib does not create them unless explicitly requested (with
2579 * WIMLIB_WRITE_FLAG_PIPABLE) and as stated above they use different magic
2580 * characters to identify the file.
2583 write_pipable_wim(WIMStruct *wim, int image, int write_flags,
2584 unsigned num_threads,
2585 struct list_head *blob_list_override,
2586 struct list_head *blob_table_list_ret)
2589 struct wim_reshdr xml_reshdr;
2591 WARNING("Creating a pipable WIM, which will "
2593 " with Microsoft's software (WIMGAPI/ImageX/DISM).");
2595 /* At this point, the header at the beginning of the file has already
2599 * For efficiency, wimlib normally delays calculating each newly added
2600 * stream's hash until while that stream being written, or just before
2601 * it is written. However, when writing a pipable WIM (potentially to a
2602 * pipe), we first have to write the metadata resources, which contain
2603 * all the hashes. Moreover each blob is prefixed with its hash (struct
2604 * pwm_blob_hdr). Thus, we have to calculate all the hashes before
2607 ret = wim_checksum_unhashed_blobs(wim);
2611 /* Write extra copy of the XML data. */
2612 ret = write_wim_xml_data(wim, image, WIM_TOTALBYTES_OMIT,
2613 &xml_reshdr, WRITE_RESOURCE_FLAG_PIPABLE);
2617 /* Write metadata resources for the image(s) being included in the
2619 ret = write_metadata_resources(wim, image, write_flags);
2623 /* Write file data needed for the image(s) being included in the output
2624 * WIM, or file data needed for the split WIM part. */
2625 return write_file_data(wim, image, write_flags,
2626 num_threads, blob_list_override,
2627 blob_table_list_ret);
2629 /* The blob table, XML data, and header at end are handled by
2630 * finish_write(). */
2634 should_default_to_solid_compression(WIMStruct *wim, int write_flags)
2636 return wim->out_hdr.wim_version == WIM_VERSION_SOLID &&
2637 !(write_flags & (WIMLIB_WRITE_FLAG_SOLID |
2638 WIMLIB_WRITE_FLAG_PIPABLE)) &&
2639 wim_has_solid_resources(wim);
2642 /* Update the images' filecount/bytecount stats (in the XML info) to take into
2643 * account any recent modifications. */
2645 update_image_stats(WIMStruct *wim)
2647 if (!wim_has_metadata(wim))
2649 for (int i = 0; i < wim->hdr.image_count; i++) {
2650 struct wim_image_metadata *imd = wim->image_metadata[i];
2651 if (imd->stats_outdated) {
2652 int ret = xml_update_image_info(wim, i + 1);
2655 imd->stats_outdated = false;
2661 /* Write a standalone WIM or split WIM (SWM) part to a new file or to a file
2664 write_wim_part(WIMStruct *wim,
2665 const void *path_or_fd,
2668 unsigned num_threads,
2669 unsigned part_number,
2670 unsigned total_parts,
2671 struct list_head *blob_list_override,
2675 struct list_head blob_table_list;
2677 /* Internally, this is always called with a valid part number and total
2679 wimlib_assert(total_parts >= 1);
2680 wimlib_assert(part_number >= 1 && part_number <= total_parts);
2682 /* A valid image (or all images) must be specified. */
2683 if (image != WIMLIB_ALL_IMAGES &&
2684 (image < 1 || image > wim->hdr.image_count))
2685 return WIMLIB_ERR_INVALID_IMAGE;
2687 /* If we need to write metadata resources, make sure the ::WIMStruct has
2688 * the needed information attached (e.g. is not a resource-only WIM,
2689 * such as a non-first part of a split WIM). */
2690 if (!wim_has_metadata(wim) &&
2691 !(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA))
2692 return WIMLIB_ERR_METADATA_NOT_FOUND;
2694 /* Check for contradictory flags. */
2695 if ((write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2696 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))
2697 == (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2698 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))
2699 return WIMLIB_ERR_INVALID_PARAM;
2701 if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2702 WIMLIB_WRITE_FLAG_NOT_PIPABLE))
2703 == (WIMLIB_WRITE_FLAG_PIPABLE |
2704 WIMLIB_WRITE_FLAG_NOT_PIPABLE))
2705 return WIMLIB_ERR_INVALID_PARAM;
2707 /* Only wimlib_overwrite() accepts UNSAFE_COMPACT. */
2708 if (write_flags & WIMLIB_WRITE_FLAG_UNSAFE_COMPACT)
2709 return WIMLIB_ERR_INVALID_PARAM;
2711 /* Include an integrity table by default if no preference was given and
2712 * the WIM already had an integrity table. */
2713 if (!(write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2714 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))) {
2715 if (wim_has_integrity_table(wim))
2716 write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY;
2719 /* Write a pipable WIM by default if no preference was given and the WIM
2720 * was already pipable. */
2721 if (!(write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2722 WIMLIB_WRITE_FLAG_NOT_PIPABLE))) {
2723 if (wim_is_pipable(wim))
2724 write_flags |= WIMLIB_WRITE_FLAG_PIPABLE;
2727 if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2728 WIMLIB_WRITE_FLAG_SOLID))
2729 == (WIMLIB_WRITE_FLAG_PIPABLE |
2730 WIMLIB_WRITE_FLAG_SOLID))
2732 ERROR("Solid compression is unsupported in pipable WIMs");
2733 return WIMLIB_ERR_INVALID_PARAM;
2736 /* Start initializing the new file header. */
2737 memset(&wim->out_hdr, 0, sizeof(wim->out_hdr));
2739 /* Set the magic number. */
2740 if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
2741 wim->out_hdr.magic = PWM_MAGIC;
2743 wim->out_hdr.magic = WIM_MAGIC;
2745 /* Set the version number. */
2746 if ((write_flags & WIMLIB_WRITE_FLAG_SOLID) ||
2747 wim->out_compression_type == WIMLIB_COMPRESSION_TYPE_LZMS)
2748 wim->out_hdr.wim_version = WIM_VERSION_SOLID;
2750 wim->out_hdr.wim_version = WIM_VERSION_DEFAULT;
2752 /* Default to solid compression if it is valid in the chosen WIM file
2753 * format and the WIMStruct references any solid resources. This is
2754 * useful when exporting an image from a solid WIM. */
2755 if (should_default_to_solid_compression(wim, write_flags))
2756 write_flags |= WIMLIB_WRITE_FLAG_SOLID;
2758 /* Set the header flags. */
2759 wim->out_hdr.flags = (wim->hdr.flags & (WIM_HDR_FLAG_RP_FIX |
2760 WIM_HDR_FLAG_READONLY));
2761 if (total_parts != 1)
2762 wim->out_hdr.flags |= WIM_HDR_FLAG_SPANNED;
2763 if (wim->out_compression_type != WIMLIB_COMPRESSION_TYPE_NONE) {
2764 wim->out_hdr.flags |= WIM_HDR_FLAG_COMPRESSION;
2765 switch (wim->out_compression_type) {
2766 case WIMLIB_COMPRESSION_TYPE_XPRESS:
2767 wim->out_hdr.flags |= WIM_HDR_FLAG_COMPRESS_XPRESS;
2769 case WIMLIB_COMPRESSION_TYPE_LZX:
2770 wim->out_hdr.flags |= WIM_HDR_FLAG_COMPRESS_LZX;
2772 case WIMLIB_COMPRESSION_TYPE_LZMS:
2773 wim->out_hdr.flags |= WIM_HDR_FLAG_COMPRESS_LZMS;
2778 /* Set the chunk size. */
2779 wim->out_hdr.chunk_size = wim->out_chunk_size;
2782 if (write_flags & WIMLIB_WRITE_FLAG_RETAIN_GUID)
2783 guid = wim->hdr.guid;
2785 copy_guid(wim->out_hdr.guid, guid);
2787 generate_guid(wim->out_hdr.guid);
2789 /* Set the part number and total parts. */
2790 wim->out_hdr.part_number = part_number;
2791 wim->out_hdr.total_parts = total_parts;
2793 /* Set the image count. */
2794 if (image == WIMLIB_ALL_IMAGES)
2795 wim->out_hdr.image_count = wim->hdr.image_count;
2797 wim->out_hdr.image_count = 1;
2799 /* Set the boot index. */
2800 wim->out_hdr.boot_idx = 0;
2801 if (total_parts == 1) {
2802 if (image == WIMLIB_ALL_IMAGES)
2803 wim->out_hdr.boot_idx = wim->hdr.boot_idx;
2804 else if (image == wim->hdr.boot_idx)
2805 wim->out_hdr.boot_idx = 1;
2808 /* Update image stats if needed. */
2809 ret = update_image_stats(wim);
2813 /* Set up the output file descriptor. */
2814 if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR) {
2815 /* File descriptor was explicitly provided. */
2816 filedes_init(&wim->out_fd, *(const int *)path_or_fd);
2817 if (!filedes_is_seekable(&wim->out_fd)) {
2818 /* The file descriptor is a pipe. */
2819 ret = WIMLIB_ERR_INVALID_PARAM;
2820 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
2822 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
2823 ERROR("Can't include integrity check when "
2824 "writing pipable WIM to pipe!");
2829 /* Filename of WIM to write was provided; open file descriptor
2831 ret = open_wim_writable(wim, (const tchar*)path_or_fd,
2832 O_TRUNC | O_CREAT | O_RDWR);
2837 /* Write initial header. This is merely a "dummy" header since it
2838 * doesn't have resource entries filled in yet, so it will be
2839 * overwritten later (unless writing a pipable WIM). */
2840 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
2841 wim->out_hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2842 ret = write_wim_header(&wim->out_hdr, &wim->out_fd, wim->out_fd.offset);
2843 wim->out_hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2847 /* Write file data and metadata resources. */
2848 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE)) {
2849 /* Default case: create a normal (non-pipable) WIM. */
2850 ret = write_file_data(wim, image, write_flags,
2857 ret = write_metadata_resources(wim, image, write_flags);
2861 /* Non-default case: create pipable WIM. */
2862 ret = write_pipable_wim(wim, image, write_flags, num_threads,
2869 /* Write blob table, XML data, and (optional) integrity table. */
2870 ret = finish_write(wim, image, write_flags, &blob_table_list);
2872 (void)close_wim_writable(wim, write_flags);
2876 /* Write a standalone WIM to a file or file descriptor. */
2878 write_standalone_wim(WIMStruct *wim, const void *path_or_fd,
2879 int image, int write_flags, unsigned num_threads)
2881 return write_wim_part(wim, path_or_fd, image, write_flags,
2882 num_threads, 1, 1, NULL, NULL);
2885 /* API function documented in wimlib.h */
2887 wimlib_write(WIMStruct *wim, const tchar *path,
2888 int image, int write_flags, unsigned num_threads)
2890 if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC)
2891 return WIMLIB_ERR_INVALID_PARAM;
2893 if (path == NULL || path[0] == T('\0'))
2894 return WIMLIB_ERR_INVALID_PARAM;
2896 return write_standalone_wim(wim, path, image, write_flags, num_threads);
2899 /* API function documented in wimlib.h */
2901 wimlib_write_to_fd(WIMStruct *wim, int fd,
2902 int image, int write_flags, unsigned num_threads)
2904 if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC)
2905 return WIMLIB_ERR_INVALID_PARAM;
2908 return WIMLIB_ERR_INVALID_PARAM;
2910 write_flags |= WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR;
2912 return write_standalone_wim(wim, &fd, image, write_flags, num_threads);
2915 /* Have there been any changes to images in the specified WIM, including updates
2916 * as well as deletions and additions of entire images, but excluding changes to
2917 * the XML document? */
2919 any_images_changed(WIMStruct *wim)
2921 if (wim->image_deletion_occurred)
2923 for (int i = 0; i < wim->hdr.image_count; i++)
2924 if (!is_image_unchanged_from_wim(wim->image_metadata[i], wim))
2930 check_resource_offset(struct blob_descriptor *blob, void *_wim)
2932 const WIMStruct *wim = _wim;
2933 off_t end_offset = *(const off_t*)wim->private;
2935 if (blob->blob_location == BLOB_IN_WIM &&
2936 blob->rdesc->wim == wim &&
2937 blob->rdesc->offset_in_wim + blob->rdesc->size_in_wim > end_offset)
2938 return WIMLIB_ERR_RESOURCE_ORDER;
2942 /* Make sure no file or metadata resources are located after the XML data (or
2943 * integrity table if present)--- otherwise we can't safely append to the WIM
2944 * file and we return WIMLIB_ERR_RESOURCE_ORDER. */
2946 check_resource_offsets(WIMStruct *wim, off_t end_offset)
2951 wim->private = &end_offset;
2952 ret = for_blob_in_table(wim->blob_table, check_resource_offset, wim);
2956 for (i = 0; i < wim->hdr.image_count; i++) {
2957 ret = check_resource_offset(wim->image_metadata[i]->metadata_blob, wim);
2965 free_blob_if_invalidated(struct blob_descriptor *blob, void *_wim)
2967 const WIMStruct *wim = _wim;
2969 if (!blob->will_be_in_output_wim &&
2970 blob->blob_location == BLOB_IN_WIM && blob->rdesc->wim == wim)
2972 blob_table_unlink(wim->blob_table, blob);
2973 free_blob_descriptor(blob);
2979 * Overwrite a WIM, possibly appending new resources to it.
2981 * A WIM looks like (or is supposed to look like) the following:
2983 * Header (212 bytes)
2984 * Resources for metadata and files (variable size)
2985 * Blob table (variable size)
2986 * XML data (variable size)
2987 * Integrity table (optional) (variable size)
2989 * If we are not adding any new files or metadata, then the blob table is
2990 * unchanged--- so we only need to overwrite the XML data, integrity table, and
2991 * header. This operation is potentially unsafe if the program is abruptly
2992 * terminated while the XML data or integrity table are being overwritten, but
2993 * before the new header has been written. To partially alleviate this problem,
2994 * we write a temporary header after the XML data has been written. This may
2995 * prevent the WIM from becoming corrupted if the program is terminated while
2996 * the integrity table is being calculated (but no guarantees, due to write
2999 * If we are adding new blobs, including new file data as well as any metadata
3000 * for any new images, then the blob table needs to be changed, and those blobs
3001 * need to be written. In this case, we try to perform a safe update of the WIM
3002 * file by writing the blobs *after* the end of the previous WIM, then writing
3003 * the new blob table, XML data, and (optionally) integrity table following the
3004 * new blobs. This will produce a layout like the following:
3006 * Header (212 bytes)
3007 * (OLD) Resources for metadata and files (variable size)
3008 * (OLD) Blob table (variable size)
3009 * (OLD) XML data (variable size)
3010 * (OLD) Integrity table (optional) (variable size)
3011 * (NEW) Resources for metadata and files (variable size)
3012 * (NEW) Blob table (variable size)
3013 * (NEW) XML data (variable size)
3014 * (NEW) Integrity table (optional) (variable size)
3016 * At all points, the WIM is valid as nothing points to the new data yet. Then,
3017 * the header is overwritten to point to the new blob table, XML data, and
3018 * integrity table, to produce the following layout:
3020 * Header (212 bytes)
3021 * Resources for metadata and files (variable size)
3022 * Nothing (variable size)
3023 * Resources for metadata and files (variable size)
3024 * Blob table (variable size)
3025 * XML data (variable size)
3026 * Integrity table (optional) (variable size)
3028 * This function allows an image to be appended to a large WIM very quickly, and
3029 * is crash-safe except in the case of write re-ordering, but the disadvantage
3030 * is that a small hole is left in the WIM where the old blob table, xml data,
3031 * and integrity table were. (These usually only take up a small amount of
3032 * space compared to the blobs, however.)
3034 * Finally, this function also supports "compaction" overwrites as an
3035 * alternative to the normal "append" overwrites described above. In a
3036 * compaction, data is written starting immediately from the end of the header.
3037 * All existing resources are written first, in order by file offset. New
3038 * resources are written afterwards, and at the end any extra data is truncated
3039 * from the file. The advantage of this approach is that is that the WIM file
3040 * ends up fully optimized, without any holes remaining. The main disadavantage
3041 * is that this operation is fundamentally unsafe and cannot be interrupted
3042 * without data corruption. Consequently, compactions are only ever done when
3043 * explicitly requested by the library user with the flag
3044 * WIMLIB_WRITE_FLAG_UNSAFE_COMPACT. (Another disadvantage is that a compaction
3045 * can be much slower than an append.)
3048 overwrite_wim_inplace(WIMStruct *wim, int write_flags, unsigned num_threads)
3052 struct list_head blob_list;
3053 struct list_head blob_table_list;
3054 struct filter_context filter_ctx;
3056 /* Include an integrity table by default if no preference was given and
3057 * the WIM already had an integrity table. */
3058 if (!(write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
3059 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)))
3060 if (wim_has_integrity_table(wim))
3061 write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY;
3063 /* Start preparing the updated file header. */
3064 memcpy(&wim->out_hdr, &wim->hdr, sizeof(wim->out_hdr));
3066 /* If using solid compression, the version number must be set to
3067 * WIM_VERSION_SOLID. */
3068 if (write_flags & WIMLIB_WRITE_FLAG_SOLID)
3069 wim->out_hdr.wim_version = WIM_VERSION_SOLID;
3071 /* Default to solid compression if it is valid in the chosen WIM file
3072 * format and the WIMStruct references any solid resources. This is
3073 * useful when updating a solid WIM. */
3074 if (should_default_to_solid_compression(wim, write_flags))
3075 write_flags |= WIMLIB_WRITE_FLAG_SOLID;
3077 if (unlikely(write_flags & WIMLIB_WRITE_FLAG_UNSAFE_COMPACT)) {
3079 /* In-place compaction */
3081 WARNING("The WIM file \"%"TS"\" is being compacted in place.\n"
3082 " Do *not* interrupt the operation, or else "
3083 "the WIM file will be\n"
3084 " corrupted!", wim->filename);
3085 wim->being_compacted = 1;
3086 old_wim_end = WIM_HEADER_DISK_SIZE;
3088 ret = prepare_blob_list_for_write(wim, WIMLIB_ALL_IMAGES,
3089 write_flags, &blob_list,
3090 &blob_table_list, &filter_ctx);
3094 /* Prevent new files from being deduplicated with existing blobs
3095 * in the WIM that we haven't decided to write. Such blobs will
3096 * be overwritten during the compaction. */
3097 for_blob_in_table(wim->blob_table, free_blob_if_invalidated, wim);
3099 if (wim_has_metadata(wim)) {
3100 /* Add existing metadata resources to be compacted along
3101 * with the file resources. */
3102 for (int i = 0; i < wim->hdr.image_count; i++) {
3103 struct wim_image_metadata *imd = wim->image_metadata[i];
3104 if (is_image_unchanged_from_wim(imd, wim)) {
3105 fully_reference_blob_for_write(imd->metadata_blob,
3111 u64 old_blob_table_end, old_xml_begin, old_xml_end;
3113 /* Set additional flags for append. */
3114 write_flags |= WIMLIB_WRITE_FLAG_APPEND |
3115 WIMLIB_WRITE_FLAG_STREAMS_OK;
3117 /* Make sure there is no data after the XML data, except
3118 * possibily an integrity table. If this were the case, then
3119 * this data would be overwritten. */
3120 old_xml_begin = wim->hdr.xml_data_reshdr.offset_in_wim;
3121 old_xml_end = old_xml_begin + wim->hdr.xml_data_reshdr.size_in_wim;
3122 if (wim->hdr.blob_table_reshdr.offset_in_wim == 0)
3123 old_blob_table_end = WIM_HEADER_DISK_SIZE;
3125 old_blob_table_end = wim->hdr.blob_table_reshdr.offset_in_wim +
3126 wim->hdr.blob_table_reshdr.size_in_wim;
3127 if (wim_has_integrity_table(wim) &&
3128 wim->hdr.integrity_table_reshdr.offset_in_wim < old_xml_end) {
3129 WARNING("Didn't expect the integrity table to be "
3130 "before the XML data");
3131 ret = WIMLIB_ERR_RESOURCE_ORDER;
3135 if (old_blob_table_end > old_xml_begin) {
3136 WARNING("Didn't expect the blob table to be after "
3138 ret = WIMLIB_ERR_RESOURCE_ORDER;
3141 /* Set @old_wim_end, which indicates the point beyond which we
3142 * don't allow any file and metadata resources to appear without
3143 * returning WIMLIB_ERR_RESOURCE_ORDER (due to the fact that we
3144 * would otherwise overwrite these resources). */
3145 if (!any_images_changed(wim)) {
3146 /* If no images have been modified, added, or deleted,
3147 * then a new blob table does not need to be written.
3148 * We shall write the new XML data and optional
3149 * integrity table immediately after the blob table.
3150 * Note that this may overwrite an existing integrity
3152 old_wim_end = old_blob_table_end;
3153 write_flags |= WIMLIB_WRITE_FLAG_NO_NEW_BLOBS;
3154 } else if (wim_has_integrity_table(wim)) {
3155 /* Old WIM has an integrity table; begin writing new
3156 * blobs after it. */
3157 old_wim_end = wim->hdr.integrity_table_reshdr.offset_in_wim +
3158 wim->hdr.integrity_table_reshdr.size_in_wim;
3160 /* No existing integrity table; begin writing new blobs
3161 * after the old XML data. */
3162 old_wim_end = old_xml_end;
3165 ret = check_resource_offsets(wim, old_wim_end);
3169 ret = prepare_blob_list_for_write(wim, WIMLIB_ALL_IMAGES,
3170 write_flags, &blob_list,
3171 &blob_table_list, &filter_ctx);
3175 if (write_flags & WIMLIB_WRITE_FLAG_NO_NEW_BLOBS)
3176 wimlib_assert(list_empty(&blob_list));
3179 /* Update image stats if needed. */
3180 ret = update_image_stats(wim);
3184 ret = open_wim_writable(wim, wim->filename, O_RDWR);
3188 ret = lock_wim_for_append(wim);
3192 /* Set WIM_HDR_FLAG_WRITE_IN_PROGRESS flag in header. */
3193 wim->hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
3194 ret = write_wim_header_flags(wim->hdr.flags, &wim->out_fd);
3195 wim->hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS;
3197 ERROR_WITH_ERRNO("Error updating WIM header flags");
3198 goto out_unlock_wim;
3201 if (filedes_seek(&wim->out_fd, old_wim_end) == -1) {
3202 ERROR_WITH_ERRNO("Can't seek to end of WIM");
3203 ret = WIMLIB_ERR_WRITE;
3204 goto out_restore_hdr;
3207 ret = write_file_data_blobs(wim, &blob_list, write_flags,
3208 num_threads, &filter_ctx);
3212 ret = write_metadata_resources(wim, WIMLIB_ALL_IMAGES, write_flags);
3216 ret = finish_write(wim, WIMLIB_ALL_IMAGES, write_flags,
3221 unlock_wim_for_append(wim);
3225 if (!(write_flags & (WIMLIB_WRITE_FLAG_NO_NEW_BLOBS |
3226 WIMLIB_WRITE_FLAG_UNSAFE_COMPACT))) {
3227 WARNING("Truncating \"%"TS"\" to its original size "
3228 "(%"PRIu64" bytes)", wim->filename, old_wim_end);
3229 if (ftruncate(wim->out_fd.fd, old_wim_end))
3230 WARNING_WITH_ERRNO("Failed to truncate WIM file!");
3233 (void)write_wim_header_flags(wim->hdr.flags, &wim->out_fd);
3235 unlock_wim_for_append(wim);
3237 (void)close_wim_writable(wim, write_flags);
3239 wim->being_compacted = 0;
3244 overwrite_wim_via_tmpfile(WIMStruct *wim, int write_flags, unsigned num_threads)
3246 size_t wim_name_len;
3249 /* Write the WIM to a temporary file in the same directory as the
3251 wim_name_len = tstrlen(wim->filename);
3252 tchar tmpfile[wim_name_len + 10];
3253 tmemcpy(tmpfile, wim->filename, wim_name_len);
3254 get_random_alnum_chars(tmpfile + wim_name_len, 9);
3255 tmpfile[wim_name_len + 9] = T('\0');
3257 ret = wimlib_write(wim, tmpfile, WIMLIB_ALL_IMAGES,
3259 WIMLIB_WRITE_FLAG_FSYNC |
3260 WIMLIB_WRITE_FLAG_RETAIN_GUID,
3267 if (filedes_valid(&wim->in_fd)) {
3268 filedes_close(&wim->in_fd);
3269 filedes_invalidate(&wim->in_fd);
3272 /* Rename the new WIM file to the original WIM file. Note: on Windows
3273 * this actually calls win32_rename_replacement(), not _wrename(), so
3274 * that removing the existing destination file can be handled. */
3275 ret = trename(tmpfile, wim->filename);
3277 ERROR_WITH_ERRNO("Failed to rename `%"TS"' to `%"TS"'",
3278 tmpfile, wim->filename);
3285 return WIMLIB_ERR_RENAME;
3288 union wimlib_progress_info progress;
3289 progress.rename.from = tmpfile;
3290 progress.rename.to = wim->filename;
3291 return call_progress(wim->progfunc, WIMLIB_PROGRESS_MSG_RENAME,
3292 &progress, wim->progctx);
3295 /* Determine if the specified WIM file may be updated in-place rather than by
3296 * writing and replacing it with an entirely new file. */
3298 can_overwrite_wim_inplace(const WIMStruct *wim, int write_flags)
3300 /* REBUILD flag forces full rebuild. */
3301 if (write_flags & WIMLIB_WRITE_FLAG_REBUILD)
3304 /* Image deletions cause full rebuild by default. */
3305 if (wim->image_deletion_occurred &&
3306 !(write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE))
3309 /* Pipable WIMs cannot be updated in place, nor can a non-pipable WIM be
3310 * turned into a pipable WIM in-place. */
3311 if (wim_is_pipable(wim) || (write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
3314 /* The default compression type and compression chunk size selected for
3315 * the output WIM must be the same as those currently used for the WIM.
3317 if (wim->compression_type != wim->out_compression_type)
3319 if (wim->chunk_size != wim->out_chunk_size)
3325 /* API function documented in wimlib.h */
3327 wimlib_overwrite(WIMStruct *wim, int write_flags, unsigned num_threads)
3332 if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC)
3333 return WIMLIB_ERR_INVALID_PARAM;
3336 return WIMLIB_ERR_NO_FILENAME;
3338 if (unlikely(write_flags & WIMLIB_WRITE_FLAG_UNSAFE_COMPACT)) {
3340 * In UNSAFE_COMPACT mode:
3341 * - RECOMPRESS is forbidden
3342 * - REBUILD is ignored
3343 * - SOFT_DELETE and NO_SOLID_SORT are implied
3345 if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
3346 return WIMLIB_ERR_COMPACTION_NOT_POSSIBLE;
3347 write_flags &= ~WIMLIB_WRITE_FLAG_REBUILD;
3348 write_flags |= WIMLIB_WRITE_FLAG_SOFT_DELETE;
3349 write_flags |= WIMLIB_WRITE_FLAG_NO_SOLID_SORT;
3352 orig_hdr_flags = wim->hdr.flags;
3353 if (write_flags & WIMLIB_WRITE_FLAG_IGNORE_READONLY_FLAG)
3354 wim->hdr.flags &= ~WIM_HDR_FLAG_READONLY;
3355 ret = can_modify_wim(wim);
3356 wim->hdr.flags = orig_hdr_flags;
3360 if (can_overwrite_wim_inplace(wim, write_flags)) {
3361 ret = overwrite_wim_inplace(wim, write_flags, num_threads);
3362 if (ret != WIMLIB_ERR_RESOURCE_ORDER)
3364 WARNING("Falling back to re-building entire WIM");
3366 if (write_flags & WIMLIB_WRITE_FLAG_UNSAFE_COMPACT)
3367 return WIMLIB_ERR_COMPACTION_NOT_POSSIBLE;
3368 return overwrite_wim_via_tmpfile(wim, write_flags, num_threads);