4 * Support for writing WIM files; write a WIM file, overwrite a WIM file, write
5 * compressed file resources, etc.
9 * Copyright (C) 2012, 2013, 2014, 2015 Eric Biggers
11 * This file is free software; you can redistribute it and/or modify it under
12 * the terms of the GNU Lesser General Public License as published by the Free
13 * Software Foundation; either version 3 of the License, or (at your option) any
16 * This file is distributed in the hope that it will be useful, but WITHOUT
17 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
18 * FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
21 * You should have received a copy of the GNU Lesser General Public License
22 * along with this file; if not, see http://www.gnu.org/licenses/.
29 #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK)
30 /* On BSD, this should be included before "wimlib/list.h" so that "wimlib/list.h" can
31 * overwrite the LIST_HEAD macro. */
32 # include <sys/file.h>
40 #include "wimlib/alloca.h"
41 #include "wimlib/assert.h"
42 #include "wimlib/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_OVERWRITE &&
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)
146 return ctx->write_flags & WIMLIB_WRITE_FLAG_OVERWRITE;
150 may_hard_filter_blobs(const struct filter_context *ctx)
154 return ctx->write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS;
158 may_filter_blobs(const struct filter_context *ctx)
160 return (may_soft_filter_blobs(ctx) || may_hard_filter_blobs(ctx));
163 /* Return true if the specified resource is compressed and the compressed data
164 * can be reused with the specified output parameters. */
166 can_raw_copy(const struct blob_descriptor *blob,
167 int write_resource_flags, int out_ctype, u32 out_chunk_size)
169 const struct wim_resource_descriptor *rdesc;
171 if (write_resource_flags & WRITE_RESOURCE_FLAG_RECOMPRESS)
174 if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE)
177 if (blob->blob_location != BLOB_IN_WIM)
182 if (rdesc->is_pipable != !!(write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE))
185 if (rdesc->flags & WIM_RESHDR_FLAG_COMPRESSED) {
186 /* Normal compressed resource: Must use same compression type
188 return (rdesc->compression_type == out_ctype &&
189 rdesc->chunk_size == out_chunk_size);
192 if ((rdesc->flags & WIM_RESHDR_FLAG_SOLID) &&
193 (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID))
195 /* Solid resource: Such resources may contain multiple blobs,
196 * and in general only a subset of them need to be written. As
197 * a heuristic, re-use the raw data if more than two-thirds the
198 * uncompressed size is being written. */
200 /* Note: solid resources contain a header that specifies the
201 * compression type and chunk size; therefore we don't need to
202 * check if they are compatible with @out_ctype and
203 * @out_chunk_size. */
205 struct blob_descriptor *res_blob;
208 list_for_each_entry(res_blob, &rdesc->blob_list, rdesc_node)
209 if (res_blob->will_be_in_output_wim)
210 write_size += res_blob->size;
212 return (write_size > rdesc->uncompressed_size * 2 / 3);
219 reshdr_flags_for_blob(const struct blob_descriptor *blob)
221 u32 reshdr_flags = 0;
222 if (blob->is_metadata)
223 reshdr_flags |= WIM_RESHDR_FLAG_METADATA;
228 blob_set_out_reshdr_for_reuse(struct blob_descriptor *blob)
230 const struct wim_resource_descriptor *rdesc;
232 wimlib_assert(blob->blob_location == BLOB_IN_WIM);
235 if (rdesc->flags & WIM_RESHDR_FLAG_SOLID) {
236 blob->out_reshdr.offset_in_wim = blob->offset_in_res;
237 blob->out_reshdr.uncompressed_size = 0;
238 blob->out_reshdr.size_in_wim = blob->size;
240 blob->out_res_offset_in_wim = rdesc->offset_in_wim;
241 blob->out_res_size_in_wim = rdesc->size_in_wim;
242 blob->out_res_uncompressed_size = rdesc->uncompressed_size;
244 blob->out_reshdr.offset_in_wim = rdesc->offset_in_wim;
245 blob->out_reshdr.uncompressed_size = rdesc->uncompressed_size;
246 blob->out_reshdr.size_in_wim = rdesc->size_in_wim;
248 blob->out_reshdr.flags = rdesc->flags;
252 /* Write the header for a blob in a pipable WIM. */
254 write_pwm_blob_header(const struct blob_descriptor *blob,
255 struct filedes *out_fd, bool compressed)
257 struct pwm_blob_hdr blob_hdr;
261 wimlib_assert(!blob->unhashed);
263 blob_hdr.magic = cpu_to_le64(PWM_BLOB_MAGIC);
264 blob_hdr.uncompressed_size = cpu_to_le64(blob->size);
265 copy_hash(blob_hdr.hash, blob->hash);
266 reshdr_flags = reshdr_flags_for_blob(blob);
268 reshdr_flags |= WIM_RESHDR_FLAG_COMPRESSED;
269 blob_hdr.flags = cpu_to_le32(reshdr_flags);
270 ret = full_write(out_fd, &blob_hdr, sizeof(blob_hdr));
272 ERROR_WITH_ERRNO("Write error");
276 struct write_blobs_progress_data {
277 wimlib_progress_func_t progfunc;
279 union wimlib_progress_info progress;
284 do_write_blobs_progress(struct write_blobs_progress_data *progress_data,
285 u64 complete_size, u32 complete_count, bool discarded)
287 union wimlib_progress_info *progress = &progress_data->progress;
291 progress->write_streams.total_bytes -= complete_size;
292 progress->write_streams.total_streams -= complete_count;
293 if (progress_data->next_progress != ~(u64)0 &&
294 progress_data->next_progress > progress->write_streams.total_bytes)
296 progress_data->next_progress = progress->write_streams.total_bytes;
299 progress->write_streams.completed_bytes += complete_size;
300 progress->write_streams.completed_streams += complete_count;
303 if (progress->write_streams.completed_bytes >= progress_data->next_progress)
305 ret = call_progress(progress_data->progfunc,
306 WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
308 progress_data->progctx);
312 if (progress_data->next_progress == progress->write_streams.total_bytes) {
313 progress_data->next_progress = ~(u64)0;
315 /* Handle rate-limiting of messages */
317 /* Send new message as soon as another 1/128 of the
318 * total has been written. (Arbitrary number.) */
319 progress_data->next_progress =
320 progress->write_streams.completed_bytes +
321 progress->write_streams.total_bytes / 128;
323 /* ... Unless that would be more than 5000000 bytes, in
324 * which case send the next after the next 5000000
325 * bytes. (Another arbitrary number.) */
326 if (progress->write_streams.completed_bytes + 5000000 <
327 progress_data->next_progress)
328 progress_data->next_progress =
329 progress->write_streams.completed_bytes + 5000000;
331 /* ... But always send a message as soon as we're
332 * completely done. */
333 if (progress->write_streams.total_bytes <
334 progress_data->next_progress)
335 progress_data->next_progress =
336 progress->write_streams.total_bytes;
342 struct write_blobs_ctx {
343 /* File descriptor to which the blobs are being written. */
344 struct filedes *out_fd;
346 /* Blob table for the WIMStruct on whose behalf the blobs are being
348 struct blob_table *blob_table;
350 /* Compression format to use. */
353 /* Maximum uncompressed chunk size in compressed resources to use. */
356 /* Flags that affect how the blobs will be written. */
357 int write_resource_flags;
359 /* Data used for issuing WRITE_STREAMS progress. */
360 struct write_blobs_progress_data progress_data;
362 struct filter_context *filter_ctx;
364 /* Upper bound on the total number of bytes that need to be compressed.
366 u64 num_bytes_to_compress;
368 /* Pointer to the chunk_compressor implementation being used for
369 * compressing chunks of data, or NULL if chunks are being written
371 struct chunk_compressor *compressor;
373 /* A buffer of size @out_chunk_size that has been loaned out from the
374 * chunk compressor and is currently being filled with the uncompressed
375 * data of the next chunk. */
378 /* Number of bytes in @cur_chunk_buf that are currently filled. */
379 size_t cur_chunk_buf_filled;
381 /* List of blobs that currently have chunks being compressed. */
382 struct list_head blobs_being_compressed;
384 /* List of blobs in the solid resource. Blobs are moved here after
385 * @blobs_being_compressed only when writing a solid resource. */
386 struct list_head blobs_in_solid_resource;
388 /* Current uncompressed offset in the blob being read. */
389 u64 cur_read_blob_offset;
391 /* Uncompressed size of the blob currently being read. */
392 u64 cur_read_blob_size;
394 /* Current uncompressed offset in the blob being written. */
395 u64 cur_write_blob_offset;
397 /* Uncompressed size of resource currently being written. */
398 u64 cur_write_res_size;
400 /* Array that is filled in with compressed chunk sizes as a resource is
404 /* Index of next entry in @chunk_csizes to fill in. */
407 /* Number of entries in @chunk_csizes currently allocated. */
408 size_t num_alloc_chunks;
410 /* Offset in the output file of the start of the chunks of the resource
411 * currently being written. */
412 u64 chunks_start_offset;
415 /* Reserve space for the chunk table and prepare to accumulate the chunk table
418 begin_chunk_table(struct write_blobs_ctx *ctx, u64 res_expected_size)
420 u64 expected_num_chunks;
421 u64 expected_num_chunk_entries;
425 /* Calculate the number of chunks and chunk entries that should be
426 * needed for the resource. These normally will be the final values,
427 * but in SOLID mode some of the blobs we're planning to write into the
428 * resource may be duplicates, and therefore discarded, potentially
429 * decreasing the number of chunk entries needed. */
430 expected_num_chunks = DIV_ROUND_UP(res_expected_size, ctx->out_chunk_size);
431 expected_num_chunk_entries = expected_num_chunks;
432 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID))
433 expected_num_chunk_entries--;
435 /* Make sure the chunk_csizes array is long enough to store the
436 * compressed size of each chunk. */
437 if (expected_num_chunks > ctx->num_alloc_chunks) {
438 u64 new_length = expected_num_chunks + 50;
440 if ((size_t)new_length != new_length) {
441 ERROR("Resource size too large (%"PRIu64" bytes!",
443 return WIMLIB_ERR_NOMEM;
446 FREE(ctx->chunk_csizes);
447 ctx->chunk_csizes = MALLOC(new_length * sizeof(ctx->chunk_csizes[0]));
448 if (ctx->chunk_csizes == NULL) {
449 ctx->num_alloc_chunks = 0;
450 return WIMLIB_ERR_NOMEM;
452 ctx->num_alloc_chunks = new_length;
455 ctx->chunk_index = 0;
457 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE)) {
458 /* Reserve space for the chunk table in the output file. In the
459 * case of solid resources this reserves the upper bound for the
460 * needed space, not necessarily the exact space which will
461 * prove to be needed. At this point, we just use @chunk_csizes
462 * for a buffer of 0's because the actual compressed chunk sizes
464 reserve_size = expected_num_chunk_entries *
465 get_chunk_entry_size(res_expected_size,
466 0 != (ctx->write_resource_flags &
467 WRITE_RESOURCE_FLAG_SOLID));
468 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID)
469 reserve_size += sizeof(struct alt_chunk_table_header_disk);
470 memset(ctx->chunk_csizes, 0, reserve_size);
471 ret = full_write(ctx->out_fd, ctx->chunk_csizes, reserve_size);
479 begin_write_resource(struct write_blobs_ctx *ctx, u64 res_expected_size)
483 wimlib_assert(res_expected_size != 0);
485 if (ctx->compressor != NULL) {
486 ret = begin_chunk_table(ctx, res_expected_size);
491 /* Output file descriptor is now positioned at the offset at which to
492 * write the first chunk of the resource. */
493 ctx->chunks_start_offset = ctx->out_fd->offset;
494 ctx->cur_write_blob_offset = 0;
495 ctx->cur_write_res_size = res_expected_size;
500 end_chunk_table(struct write_blobs_ctx *ctx, u64 res_actual_size,
501 u64 *res_start_offset_ret, u64 *res_store_size_ret)
503 size_t actual_num_chunks;
504 size_t actual_num_chunk_entries;
505 size_t chunk_entry_size;
508 actual_num_chunks = ctx->chunk_index;
509 actual_num_chunk_entries = actual_num_chunks;
510 if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID))
511 actual_num_chunk_entries--;
513 chunk_entry_size = get_chunk_entry_size(res_actual_size,
514 0 != (ctx->write_resource_flags &
515 WRITE_RESOURCE_FLAG_SOLID));
517 typedef le64 _may_alias_attribute aliased_le64_t;
518 typedef le32 _may_alias_attribute aliased_le32_t;
520 if (chunk_entry_size == 4) {
521 aliased_le32_t *entries = (aliased_le32_t*)ctx->chunk_csizes;
523 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
524 for (size_t i = 0; i < actual_num_chunk_entries; i++)
525 entries[i] = cpu_to_le32(ctx->chunk_csizes[i]);
527 u32 offset = ctx->chunk_csizes[0];
528 for (size_t i = 0; i < actual_num_chunk_entries; i++) {
529 u32 next_size = ctx->chunk_csizes[i + 1];
530 entries[i] = cpu_to_le32(offset);
535 aliased_le64_t *entries = (aliased_le64_t*)ctx->chunk_csizes;
537 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
538 for (size_t i = 0; i < actual_num_chunk_entries; i++)
539 entries[i] = cpu_to_le64(ctx->chunk_csizes[i]);
541 u64 offset = ctx->chunk_csizes[0];
542 for (size_t i = 0; i < actual_num_chunk_entries; i++) {
543 u64 next_size = ctx->chunk_csizes[i + 1];
544 entries[i] = cpu_to_le64(offset);
550 size_t chunk_table_size = actual_num_chunk_entries * chunk_entry_size;
551 u64 res_start_offset;
554 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
555 ret = full_write(ctx->out_fd, ctx->chunk_csizes, chunk_table_size);
558 res_end_offset = ctx->out_fd->offset;
559 res_start_offset = ctx->chunks_start_offset;
561 res_end_offset = ctx->out_fd->offset;
563 u64 chunk_table_offset;
565 chunk_table_offset = ctx->chunks_start_offset - chunk_table_size;
567 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
568 struct alt_chunk_table_header_disk hdr;
570 hdr.res_usize = cpu_to_le64(res_actual_size);
571 hdr.chunk_size = cpu_to_le32(ctx->out_chunk_size);
572 hdr.compression_format = cpu_to_le32(ctx->out_ctype);
574 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 1);
575 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 2);
576 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3);
578 ret = full_pwrite(ctx->out_fd, &hdr, sizeof(hdr),
579 chunk_table_offset - sizeof(hdr));
582 res_start_offset = chunk_table_offset - sizeof(hdr);
584 res_start_offset = chunk_table_offset;
587 ret = full_pwrite(ctx->out_fd, ctx->chunk_csizes,
588 chunk_table_size, chunk_table_offset);
593 *res_start_offset_ret = res_start_offset;
594 *res_store_size_ret = res_end_offset - res_start_offset;
599 ERROR_WITH_ERRNO("Write error");
603 /* Finish writing a WIM resource by writing or updating the chunk table (if not
604 * writing the data uncompressed) and loading its metadata into @out_reshdr. */
606 end_write_resource(struct write_blobs_ctx *ctx, struct wim_reshdr *out_reshdr)
610 u64 res_uncompressed_size;
611 u64 res_offset_in_wim;
613 wimlib_assert(ctx->cur_write_blob_offset == ctx->cur_write_res_size ||
614 (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID));
615 res_uncompressed_size = ctx->cur_write_res_size;
617 if (ctx->compressor) {
618 ret = end_chunk_table(ctx, res_uncompressed_size,
619 &res_offset_in_wim, &res_size_in_wim);
623 res_offset_in_wim = ctx->chunks_start_offset;
624 res_size_in_wim = ctx->out_fd->offset - res_offset_in_wim;
626 out_reshdr->uncompressed_size = res_uncompressed_size;
627 out_reshdr->size_in_wim = res_size_in_wim;
628 out_reshdr->offset_in_wim = res_offset_in_wim;
632 /* Call when no more data from the file at @path is needed. */
634 done_with_file(const tchar *path, wimlib_progress_func_t progfunc, void *progctx)
636 union wimlib_progress_info info;
638 info.done_with_file.path_to_file = path;
640 return call_progress(progfunc, WIMLIB_PROGRESS_MSG_DONE_WITH_FILE,
645 do_done_with_blob(struct blob_descriptor *blob,
646 wimlib_progress_func_t progfunc, void *progctx)
649 struct wim_inode *inode;
651 if (!blob->may_send_done_with_file)
654 inode = blob->file_inode;
656 wimlib_assert(inode != NULL);
657 wimlib_assert(inode->num_remaining_streams > 0);
658 if (--inode->num_remaining_streams > 0)
662 /* XXX: This logic really should be somewhere else. */
664 /* We want the path to the file, but blob->file_on_disk might actually
665 * refer to a named data stream. Temporarily strip the named data
666 * stream from the path. */
667 wchar_t *p_colon = NULL;
668 wchar_t *p_question_mark = NULL;
669 const wchar_t *p_stream_name;
671 p_stream_name = path_stream_name(blob->file_on_disk);
672 if (unlikely(p_stream_name)) {
673 p_colon = (wchar_t *)(p_stream_name - 1);
674 wimlib_assert(*p_colon == L':');
678 /* We also should use a fake Win32 path instead of a NT path */
679 if (!wcsncmp(blob->file_on_disk, L"\\??\\", 4)) {
680 p_question_mark = &blob->file_on_disk[1];
681 *p_question_mark = L'\\';
685 ret = done_with_file(blob->file_on_disk, progfunc, progctx);
691 *p_question_mark = L'?';
696 /* Handle WIMLIB_WRITE_FLAG_SEND_DONE_WITH_FILE_MESSAGES mode. */
698 done_with_blob(struct blob_descriptor *blob, struct write_blobs_ctx *ctx)
700 if (likely(!(ctx->write_resource_flags &
701 WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE)))
703 return do_done_with_blob(blob, ctx->progress_data.progfunc,
704 ctx->progress_data.progctx);
707 /* Begin processing a blob for writing. */
709 write_blob_begin_read(struct blob_descriptor *blob, void *_ctx)
711 struct write_blobs_ctx *ctx = _ctx;
714 wimlib_assert(blob->size > 0);
716 ctx->cur_read_blob_offset = 0;
717 ctx->cur_read_blob_size = blob->size;
719 /* As an optimization, we allow some blobs to be "unhashed", meaning
720 * their SHA-1 message digests are unknown. This is the case with blobs
721 * that are added by scanning a directory tree with wimlib_add_image(),
722 * for example. Since WIM uses single-instance blobs, we don't know
723 * whether such each such blob really need to written until it is
724 * actually checksummed, unless it has a unique size. In such cases we
725 * read and checksum the blob in this function, thereby advancing ahead
726 * of read_blob_list(), which will still provide the data again to
727 * write_blob_process_chunk(). This is okay because an unhashed blob
728 * cannot be in a WIM resource, which might be costly to decompress. */
729 if (ctx->blob_table != NULL && blob->unhashed && !blob->unique_size) {
731 struct blob_descriptor *new_blob;
733 ret = hash_unhashed_blob(blob, ctx->blob_table, &new_blob);
736 if (new_blob != blob) {
737 /* Duplicate blob detected. */
739 if (new_blob->will_be_in_output_wim ||
740 blob_filtered(new_blob, ctx->filter_ctx))
742 /* The duplicate blob is already being included
743 * in the output WIM, or it would be filtered
744 * out if it had been. Skip writing this blob
745 * (and reading it again) entirely, passing its
746 * output reference count to the duplicate blob
747 * in the former case. */
748 ret = do_write_blobs_progress(&ctx->progress_data,
749 blob->size, 1, true);
750 list_del(&blob->write_blobs_list);
751 list_del(&blob->blob_table_list);
752 if (new_blob->will_be_in_output_wim)
753 new_blob->out_refcnt += blob->out_refcnt;
754 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID)
755 ctx->cur_write_res_size -= blob->size;
757 ret = done_with_blob(blob, ctx);
758 free_blob_descriptor(blob);
761 return BEGIN_BLOB_STATUS_SKIP_BLOB;
763 /* The duplicate blob can validly be written,
764 * but was not marked as such. Discard the
765 * current blob descriptor and use the
766 * duplicate, but actually freeing the current
767 * blob descriptor must wait until
768 * read_blob_list() has finished reading its
770 list_replace(&blob->write_blobs_list,
771 &new_blob->write_blobs_list);
772 list_replace(&blob->blob_table_list,
773 &new_blob->blob_table_list);
774 blob->will_be_in_output_wim = 0;
775 new_blob->out_refcnt = blob->out_refcnt;
776 new_blob->will_be_in_output_wim = 1;
777 new_blob->may_send_done_with_file = 0;
782 list_move_tail(&blob->write_blobs_list, &ctx->blobs_being_compressed);
786 /* Rewrite a blob that was just written compressed (as a non-solid WIM resource)
787 * as uncompressed instead. */
789 write_blob_uncompressed(struct blob_descriptor *blob, struct filedes *out_fd)
792 u64 begin_offset = blob->out_reshdr.offset_in_wim;
793 u64 end_offset = out_fd->offset;
795 if (filedes_seek(out_fd, begin_offset) == -1)
798 ret = extract_full_blob_to_fd(blob, out_fd);
800 /* Error reading the uncompressed data. */
801 if (out_fd->offset == begin_offset &&
802 filedes_seek(out_fd, end_offset) != -1)
804 /* Nothing was actually written yet, and we successfully
805 * seeked to the end of the compressed resource, so
806 * don't issue a hard error; just keep the compressed
807 * resource instead. */
808 WARNING("Recovered compressed resource of "
809 "size %"PRIu64", continuing on.", blob->size);
815 wimlib_assert(out_fd->offset - begin_offset == blob->size);
817 if (out_fd->offset < end_offset &&
818 0 != ftruncate(out_fd->fd, out_fd->offset))
820 ERROR_WITH_ERRNO("Can't truncate output file to "
821 "offset %"PRIu64, out_fd->offset);
822 return WIMLIB_ERR_WRITE;
825 blob->out_reshdr.size_in_wim = blob->size;
826 blob->out_reshdr.flags &= ~(WIM_RESHDR_FLAG_COMPRESSED |
827 WIM_RESHDR_FLAG_SOLID);
831 /* Returns true if the specified blob, which was written as a non-solid
832 * resource, should be truncated from the WIM file and re-written uncompressed.
833 * blob->out_reshdr must be filled in from the initial write of the blob. */
835 should_rewrite_blob_uncompressed(const struct write_blobs_ctx *ctx,
836 const struct blob_descriptor *blob)
838 /* If the compressed data is smaller than the uncompressed data, prefer
839 * the compressed data. */
840 if (blob->out_reshdr.size_in_wim < blob->out_reshdr.uncompressed_size)
843 /* If we're not actually writing compressed data, then there's no need
845 if (!ctx->compressor)
848 /* If writing a pipable WIM, everything we write to the output is final
849 * (it might actually be a pipe!). */
850 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE)
853 /* If the blob that would need to be re-read is located in a solid
854 * resource in another WIM file, then re-reading it would be costly. So
857 * Exception: if the compressed size happens to be *exactly* the same as
858 * the uncompressed size, then the blob *must* be written uncompressed
859 * in order to remain compatible with the Windows Overlay Filesystem
860 * Filter Driver (WOF).
862 * TODO: we are currently assuming that the optimization for
863 * single-chunk resources in maybe_rewrite_blob_uncompressed() prevents
864 * this case from being triggered too often. To fully prevent excessive
865 * decompressions in degenerate cases, we really should obtain the
866 * uncompressed data by decompressing the compressed data we wrote to
869 if (blob->blob_location == BLOB_IN_WIM &&
870 blob->size != blob->rdesc->uncompressed_size &&
871 blob->size != blob->out_reshdr.size_in_wim)
878 maybe_rewrite_blob_uncompressed(struct write_blobs_ctx *ctx,
879 struct blob_descriptor *blob)
881 if (!should_rewrite_blob_uncompressed(ctx, blob))
884 /* Regular (non-solid) WIM resources with exactly one chunk and
885 * compressed size equal to uncompressed size are exactly the same as
886 * the corresponding compressed data --- since there must be 0 entries
887 * in the chunk table and the only chunk must be stored uncompressed.
888 * In this case, there's no need to rewrite anything. */
889 if (ctx->chunk_index == 1 &&
890 blob->out_reshdr.size_in_wim == blob->out_reshdr.uncompressed_size)
892 blob->out_reshdr.flags &= ~WIM_RESHDR_FLAG_COMPRESSED;
896 return write_blob_uncompressed(blob, ctx->out_fd);
899 /* Write the next chunk of (typically compressed) data to the output WIM,
900 * handling the writing of the chunk table. */
902 write_chunk(struct write_blobs_ctx *ctx, const void *cchunk,
903 size_t csize, size_t usize)
906 struct blob_descriptor *blob;
907 u32 completed_blob_count;
910 blob = list_entry(ctx->blobs_being_compressed.next,
911 struct blob_descriptor, write_blobs_list);
913 if (ctx->cur_write_blob_offset == 0 &&
914 !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID))
916 /* Starting to write a new blob in non-solid mode. */
918 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
919 ret = write_pwm_blob_header(blob, ctx->out_fd,
920 ctx->compressor != NULL);
925 ret = begin_write_resource(ctx, blob->size);
930 if (ctx->compressor != NULL) {
931 /* Record the compresed chunk size. */
932 wimlib_assert(ctx->chunk_index < ctx->num_alloc_chunks);
933 ctx->chunk_csizes[ctx->chunk_index++] = csize;
935 /* If writing a pipable WIM, before the chunk data write a chunk
936 * header that provides the compressed chunk size. */
937 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
938 struct pwm_chunk_hdr chunk_hdr = {
939 .compressed_size = cpu_to_le32(csize),
941 ret = full_write(ctx->out_fd, &chunk_hdr,
948 /* Write the chunk data. */
949 ret = full_write(ctx->out_fd, cchunk, csize);
953 ctx->cur_write_blob_offset += usize;
955 completed_size = usize;
956 completed_blob_count = 0;
957 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
958 /* Wrote chunk in solid mode. It may have finished multiple
960 struct blob_descriptor *next_blob;
962 while (blob && ctx->cur_write_blob_offset >= blob->size) {
964 ctx->cur_write_blob_offset -= blob->size;
966 if (ctx->cur_write_blob_offset)
967 next_blob = list_entry(blob->write_blobs_list.next,
968 struct blob_descriptor,
973 ret = done_with_blob(blob, ctx);
976 list_move_tail(&blob->write_blobs_list, &ctx->blobs_in_solid_resource);
977 completed_blob_count++;
982 /* Wrote chunk in non-solid mode. It may have finished a
984 if (ctx->cur_write_blob_offset == blob->size) {
986 wimlib_assert(ctx->cur_write_blob_offset ==
987 ctx->cur_write_res_size);
989 ret = end_write_resource(ctx, &blob->out_reshdr);
993 blob->out_reshdr.flags = reshdr_flags_for_blob(blob);
994 if (ctx->compressor != NULL)
995 blob->out_reshdr.flags |= WIM_RESHDR_FLAG_COMPRESSED;
997 ret = maybe_rewrite_blob_uncompressed(ctx, blob);
1001 wimlib_assert(blob->out_reshdr.uncompressed_size == blob->size);
1003 ctx->cur_write_blob_offset = 0;
1005 ret = done_with_blob(blob, ctx);
1008 list_del(&blob->write_blobs_list);
1009 completed_blob_count++;
1013 return do_write_blobs_progress(&ctx->progress_data, completed_size,
1014 completed_blob_count, false);
1017 ERROR_WITH_ERRNO("Write error");
1022 prepare_chunk_buffer(struct write_blobs_ctx *ctx)
1024 /* While we are unable to get a new chunk buffer due to too many chunks
1025 * already outstanding, retrieve and write the next compressed chunk. */
1026 while (!(ctx->cur_chunk_buf =
1027 ctx->compressor->get_chunk_buffer(ctx->compressor)))
1035 bret = ctx->compressor->get_compression_result(ctx->compressor,
1039 wimlib_assert(bret);
1041 ret = write_chunk(ctx, cchunk, csize, usize);
1048 /* Process the next chunk of data to be written to a WIM resource. */
1050 write_blob_process_chunk(const void *chunk, size_t size, void *_ctx)
1052 struct write_blobs_ctx *ctx = _ctx;
1054 const u8 *chunkptr, *chunkend;
1056 wimlib_assert(size != 0);
1058 if (ctx->compressor == NULL) {
1059 /* Write chunk uncompressed. */
1060 ret = write_chunk(ctx, chunk, size, size);
1063 ctx->cur_read_blob_offset += size;
1067 /* Submit the chunk for compression, but take into account that the
1068 * @size the chunk was provided in may not correspond to the
1069 * @out_chunk_size being used for compression. */
1071 chunkend = chunkptr + size;
1073 size_t needed_chunk_size;
1074 size_t bytes_consumed;
1076 if (!ctx->cur_chunk_buf) {
1077 ret = prepare_chunk_buffer(ctx);
1082 if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
1083 needed_chunk_size = ctx->out_chunk_size;
1085 needed_chunk_size = min(ctx->out_chunk_size,
1086 ctx->cur_chunk_buf_filled +
1087 (ctx->cur_read_blob_size -
1088 ctx->cur_read_blob_offset));
1091 bytes_consumed = min(chunkend - chunkptr,
1092 needed_chunk_size - ctx->cur_chunk_buf_filled);
1094 memcpy(&ctx->cur_chunk_buf[ctx->cur_chunk_buf_filled],
1095 chunkptr, bytes_consumed);
1097 chunkptr += bytes_consumed;
1098 ctx->cur_read_blob_offset += bytes_consumed;
1099 ctx->cur_chunk_buf_filled += bytes_consumed;
1101 if (ctx->cur_chunk_buf_filled == needed_chunk_size) {
1102 ctx->compressor->signal_chunk_filled(ctx->compressor,
1103 ctx->cur_chunk_buf_filled);
1104 ctx->cur_chunk_buf = NULL;
1105 ctx->cur_chunk_buf_filled = 0;
1107 } while (chunkptr != chunkend);
1111 /* Finish processing a blob for writing. It may not have been completely
1112 * written yet, as the chunk_compressor implementation may still have chunks
1113 * buffered or being compressed. */
1115 write_blob_end_read(struct blob_descriptor *blob, int status, void *_ctx)
1117 struct write_blobs_ctx *ctx = _ctx;
1119 wimlib_assert(ctx->cur_read_blob_offset == ctx->cur_read_blob_size || status);
1121 if (!blob->will_be_in_output_wim) {
1122 /* The blob was a duplicate. Now that its data has finished
1123 * being read, it is being discarded in favor of the duplicate
1124 * entry. It therefore is no longer needed, and we can fire the
1125 * DONE_WITH_FILE callback because the file will not be read
1128 * Note: we can't yet fire DONE_WITH_FILE for non-duplicate
1129 * blobs, since it needs to be possible to re-read the file if
1130 * it does not compress to less than its original size. */
1132 status = done_with_blob(blob, ctx);
1133 free_blob_descriptor(blob);
1134 } else if (!status && blob->unhashed && ctx->blob_table != NULL) {
1135 /* The blob was not a duplicate and was previously unhashed.
1136 * Since we passed COMPUTE_MISSING_BLOB_HASHES to
1137 * read_blob_list(), blob->hash is now computed and valid. So
1138 * turn this blob into a "hashed" blob. */
1139 list_del(&blob->unhashed_list);
1140 blob_table_insert(ctx->blob_table, blob);
1146 /* Compute statistics about a list of blobs that will be written.
1148 * Assumes the blobs are sorted such that all blobs located in each distinct WIM
1149 * (specified by WIMStruct) are together. */
1151 compute_blob_list_stats(struct list_head *blob_list,
1152 struct write_blobs_ctx *ctx)
1154 struct blob_descriptor *blob;
1155 u64 total_bytes = 0;
1157 u64 total_parts = 0;
1158 WIMStruct *prev_wim_part = NULL;
1160 list_for_each_entry(blob, blob_list, write_blobs_list) {
1162 total_bytes += blob->size;
1163 if (blob->blob_location == BLOB_IN_WIM) {
1164 if (prev_wim_part != blob->rdesc->wim) {
1165 prev_wim_part = blob->rdesc->wim;
1170 ctx->progress_data.progress.write_streams.total_bytes = total_bytes;
1171 ctx->progress_data.progress.write_streams.total_streams = num_blobs;
1172 ctx->progress_data.progress.write_streams.completed_bytes = 0;
1173 ctx->progress_data.progress.write_streams.completed_streams = 0;
1174 ctx->progress_data.progress.write_streams.compression_type = ctx->out_ctype;
1175 ctx->progress_data.progress.write_streams.total_parts = total_parts;
1176 ctx->progress_data.progress.write_streams.completed_parts = 0;
1177 ctx->progress_data.next_progress = 0;
1180 /* Find blobs in @blob_list that can be copied to the output WIM in raw form
1181 * rather than compressed. Delete these blobs from @blob_list and move them to
1182 * @raw_copy_blobs. Return the total uncompressed size of the blobs that need
1183 * to be compressed. */
1185 find_raw_copy_blobs(struct list_head *blob_list,
1186 int write_resource_flags,
1189 struct list_head *raw_copy_blobs)
1191 struct blob_descriptor *blob, *tmp;
1192 u64 num_bytes_to_compress = 0;
1194 INIT_LIST_HEAD(raw_copy_blobs);
1196 /* Initialize temporary raw_copy_ok flag. */
1197 list_for_each_entry(blob, blob_list, write_blobs_list)
1198 if (blob->blob_location == BLOB_IN_WIM)
1199 blob->rdesc->raw_copy_ok = 0;
1201 list_for_each_entry_safe(blob, tmp, blob_list, write_blobs_list) {
1202 if (blob->blob_location == BLOB_IN_WIM &&
1203 blob->rdesc->raw_copy_ok)
1205 list_move_tail(&blob->write_blobs_list,
1207 } else if (can_raw_copy(blob, write_resource_flags,
1208 out_ctype, out_chunk_size))
1210 blob->rdesc->raw_copy_ok = 1;
1211 list_move_tail(&blob->write_blobs_list,
1214 num_bytes_to_compress += blob->size;
1218 return num_bytes_to_compress;
1221 /* Copy a raw compressed resource located in another WIM file to the WIM file
1224 write_raw_copy_resource(struct wim_resource_descriptor *in_rdesc,
1225 struct filedes *out_fd)
1227 u64 cur_read_offset;
1228 u64 end_read_offset;
1229 u8 buf[BUFFER_SIZE];
1230 size_t bytes_to_read;
1232 struct filedes *in_fd;
1233 struct blob_descriptor *blob;
1234 u64 out_offset_in_wim;
1236 /* Copy the raw data. */
1237 cur_read_offset = in_rdesc->offset_in_wim;
1238 end_read_offset = cur_read_offset + in_rdesc->size_in_wim;
1240 out_offset_in_wim = out_fd->offset;
1242 if (in_rdesc->is_pipable) {
1243 if (cur_read_offset < sizeof(struct pwm_blob_hdr))
1244 return WIMLIB_ERR_INVALID_PIPABLE_WIM;
1245 cur_read_offset -= sizeof(struct pwm_blob_hdr);
1246 out_offset_in_wim += sizeof(struct pwm_blob_hdr);
1248 in_fd = &in_rdesc->wim->in_fd;
1249 wimlib_assert(cur_read_offset != end_read_offset);
1252 bytes_to_read = min(sizeof(buf), end_read_offset - cur_read_offset);
1254 ret = full_pread(in_fd, buf, bytes_to_read, cur_read_offset);
1258 ret = full_write(out_fd, buf, bytes_to_read);
1262 cur_read_offset += bytes_to_read;
1264 } while (cur_read_offset != end_read_offset);
1266 list_for_each_entry(blob, &in_rdesc->blob_list, rdesc_node) {
1267 if (blob->will_be_in_output_wim) {
1268 blob_set_out_reshdr_for_reuse(blob);
1269 if (in_rdesc->flags & WIM_RESHDR_FLAG_SOLID)
1270 blob->out_res_offset_in_wim = out_offset_in_wim;
1272 blob->out_reshdr.offset_in_wim = out_offset_in_wim;
1279 /* Copy a list of raw compressed resources located in other WIM file(s) to the
1280 * WIM file being written. */
1282 write_raw_copy_resources(struct list_head *raw_copy_blobs,
1283 struct filedes *out_fd,
1284 struct write_blobs_progress_data *progress_data)
1286 struct blob_descriptor *blob;
1289 list_for_each_entry(blob, raw_copy_blobs, write_blobs_list)
1290 blob->rdesc->raw_copy_ok = 1;
1292 list_for_each_entry(blob, raw_copy_blobs, write_blobs_list) {
1293 if (blob->rdesc->raw_copy_ok) {
1294 /* Write each solid resource only one time. */
1295 ret = write_raw_copy_resource(blob->rdesc, out_fd);
1298 blob->rdesc->raw_copy_ok = 0;
1300 ret = do_write_blobs_progress(progress_data, blob->size,
1308 /* Wait for and write all chunks pending in the compressor. */
1310 finish_remaining_chunks(struct write_blobs_ctx *ctx)
1317 if (ctx->compressor == NULL)
1320 if (ctx->cur_chunk_buf_filled != 0) {
1321 ctx->compressor->signal_chunk_filled(ctx->compressor,
1322 ctx->cur_chunk_buf_filled);
1325 while (ctx->compressor->get_compression_result(ctx->compressor, &cdata,
1328 ret = write_chunk(ctx, cdata, csize, usize);
1336 remove_empty_blobs(struct list_head *blob_list)
1338 struct blob_descriptor *blob, *tmp;
1340 list_for_each_entry_safe(blob, tmp, blob_list, write_blobs_list) {
1341 wimlib_assert(blob->will_be_in_output_wim);
1342 if (blob->size == 0) {
1343 list_del(&blob->write_blobs_list);
1344 blob->out_reshdr.offset_in_wim = 0;
1345 blob->out_reshdr.size_in_wim = 0;
1346 blob->out_reshdr.uncompressed_size = 0;
1347 blob->out_reshdr.flags = reshdr_flags_for_blob(blob);
1353 blob_is_in_file(const struct blob_descriptor *blob)
1355 return blob->blob_location == BLOB_IN_FILE_ON_DISK
1357 || blob->blob_location == BLOB_IN_WINNT_FILE_ON_DISK
1358 || blob->blob_location == BLOB_WIN32_ENCRYPTED
1364 init_done_with_file_info(struct list_head *blob_list)
1366 struct blob_descriptor *blob;
1368 list_for_each_entry(blob, blob_list, write_blobs_list) {
1369 if (blob_is_in_file(blob)) {
1370 blob->file_inode->num_remaining_streams = 0;
1371 blob->may_send_done_with_file = 1;
1373 blob->may_send_done_with_file = 0;
1377 list_for_each_entry(blob, blob_list, write_blobs_list)
1378 if (blob->may_send_done_with_file)
1379 blob->file_inode->num_remaining_streams++;
1383 * Write a list of blobs to the output WIM file.
1386 * The list of blobs to write, specified by a list of 'struct blob_descriptor' linked
1387 * by the 'write_blobs_list' member.
1390 * The file descriptor, opened for writing, to which to write the blobs.
1392 * @write_resource_flags
1393 * Flags to modify how the blobs are written:
1395 * WRITE_RESOURCE_FLAG_RECOMPRESS:
1396 * Force compression of all resources, even if they could otherwise
1397 * be re-used by copying the raw data, due to being located in a WIM
1398 * file with compatible compression parameters.
1400 * WRITE_RESOURCE_FLAG_PIPABLE:
1401 * Write the resources in the wimlib-specific pipable format, and
1402 * furthermore do so in such a way that no seeking backwards in
1403 * @out_fd will be performed (so it may be a pipe).
1405 * WRITE_RESOURCE_FLAG_SOLID:
1406 * Combine all the blobs into a single resource rather than writing
1407 * them in separate resources. This flag is only valid if the WIM
1408 * version number has been, or will be, set to WIM_VERSION_SOLID.
1409 * This flag may not be combined with WRITE_RESOURCE_FLAG_PIPABLE.
1412 * Compression format to use in the output resources, specified as one of
1413 * the WIMLIB_COMPRESSION_TYPE_* constants. WIMLIB_COMPRESSION_TYPE_NONE
1417 * Compression chunk size to use in the output resources. It must be a
1418 * valid chunk size for the specified compression format @out_ctype, unless
1419 * @out_ctype is WIMLIB_COMPRESSION_TYPE_NONE, in which case this parameter
1423 * Number of threads to use to compress data. If 0, a default number of
1424 * threads will be chosen. The number of threads still may be decreased
1425 * from the specified value if insufficient memory is detected.
1428 * If on-the-fly deduplication of unhashed blobs is desired, this parameter
1429 * must be pointer to the blob table for the WIMStruct on whose behalf the
1430 * blobs are being written. Otherwise, this parameter can be NULL.
1433 * If on-the-fly deduplication of unhashed blobs is desired, this parameter
1434 * can be a pointer to a context for blob filtering used to detect whether
1435 * the duplicate blob has been hard-filtered or not. If no blobs are
1436 * hard-filtered or no blobs are unhashed, this parameter can be NULL.
1438 * This function will write the blobs in @blob_list to resources in
1439 * consecutive positions in the output WIM file, or to a single solid resource
1440 * if WRITE_RESOURCE_FLAG_SOLID was specified in @write_resource_flags. In both
1441 * cases, the @out_reshdr of the `struct blob_descriptor' for each blob written will be
1442 * updated to specify its location, size, and flags in the output WIM. In the
1443 * solid resource case, WIM_RESHDR_FLAG_SOLID will be set in the @flags field of
1444 * each @out_reshdr, and furthermore @out_res_offset_in_wim and
1445 * @out_res_size_in_wim of each @out_reshdr will be set to the offset and size,
1446 * respectively, in the output WIM of the solid resource containing the
1447 * corresponding blob.
1449 * Each of the blobs to write may be in any location supported by the
1450 * resource-handling code (specifically, read_blob_list()), such as the contents
1451 * of external file that has been logically added to the output WIM, or a blob
1452 * in another WIM file that has been imported, or even a blob in the "same" WIM
1453 * file of which a modified copy is being written. In the case that a blob is
1454 * already in a WIM file and uses compatible compression parameters, by default
1455 * this function will re-use the raw data instead of decompressing it, then
1456 * recompressing it; however, with WRITE_RESOURCE_FLAG_RECOMPRESS
1457 * specified in @write_resource_flags, this is not done.
1459 * As a further requirement, this function requires that the
1460 * @will_be_in_output_wim member be set to 1 on all blobs in @blob_list as well
1461 * as any other blobs not in @blob_list that will be in the output WIM file, but
1462 * set to 0 on any other blobs in the output WIM's blob table or sharing a solid
1463 * resource with a blob in @blob_list. Still furthermore, if on-the-fly
1464 * deduplication of blobs is possible, then all blobs in @blob_list must also be
1465 * linked by @blob_table_list along with any other blobs that have
1466 * @will_be_in_output_wim set.
1468 * This function handles on-the-fly deduplication of blobs for which SHA-1
1469 * message digests have not yet been calculated. Such blobs may or may not need
1470 * to be written. If @blob_table is non-NULL, then each blob in @blob_list that
1471 * has @unhashed set but not @unique_size set is checksummed immediately before
1472 * it would otherwise be read for writing in order to determine if it is
1473 * identical to another blob already being written or one that would be filtered
1474 * out of the output WIM using blob_filtered() with the context @filter_ctx.
1475 * Each such duplicate blob will be removed from @blob_list, its reference count
1476 * transfered to the pre-existing duplicate blob, its memory freed, and will not
1477 * be written. Alternatively, if a blob in @blob_list is a duplicate with any
1478 * blob in @blob_table that has not been marked for writing or would not be
1479 * hard-filtered, it is freed and the pre-existing duplicate is written instead,
1480 * taking ownership of the reference count and slot in the @blob_table_list.
1482 * Returns 0 if every blob was either written successfully or did not need to be
1483 * written; otherwise returns a non-zero error code.
1486 write_blob_list(struct list_head *blob_list,
1487 struct filedes *out_fd,
1488 int write_resource_flags,
1491 unsigned num_threads,
1492 struct blob_table *blob_table,
1493 struct filter_context *filter_ctx,
1494 wimlib_progress_func_t progfunc,
1498 struct write_blobs_ctx ctx;
1499 struct list_head raw_copy_blobs;
1501 wimlib_assert((write_resource_flags &
1502 (WRITE_RESOURCE_FLAG_SOLID |
1503 WRITE_RESOURCE_FLAG_PIPABLE)) !=
1504 (WRITE_RESOURCE_FLAG_SOLID |
1505 WRITE_RESOURCE_FLAG_PIPABLE));
1507 remove_empty_blobs(blob_list);
1509 if (list_empty(blob_list))
1512 /* If needed, set auxiliary information so that we can detect when the
1513 * library has finished using each external file. */
1514 if (unlikely(write_resource_flags & WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE))
1515 init_done_with_file_info(blob_list);
1517 memset(&ctx, 0, sizeof(ctx));
1519 ctx.out_fd = out_fd;
1520 ctx.blob_table = blob_table;
1521 ctx.out_ctype = out_ctype;
1522 ctx.out_chunk_size = out_chunk_size;
1523 ctx.write_resource_flags = write_resource_flags;
1524 ctx.filter_ctx = filter_ctx;
1527 * We normally sort the blobs to write by a "sequential" order that is
1528 * optimized for reading. But when using solid compression, we instead
1529 * sort the blobs by file extension and file name (when applicable; and
1530 * we don't do this for blobs from solid resources) so that similar
1531 * files are grouped together, which improves the compression ratio.
1532 * This is somewhat of a hack since a blob does not necessarily
1533 * correspond one-to-one with a filename, nor is there any guarantee
1534 * that two files with similar names or extensions are actually similar
1535 * in content. A potential TODO is to sort the blobs based on some
1536 * measure of similarity of their actual contents.
1539 ret = sort_blob_list_by_sequential_order(blob_list,
1540 offsetof(struct blob_descriptor,
1545 compute_blob_list_stats(blob_list, &ctx);
1547 if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID_SORT) {
1548 ret = sort_blob_list_for_solid_compression(blob_list);
1550 WARNING("Failed to sort blobs for solid compression. Continuing anyways.");
1553 ctx.progress_data.progfunc = progfunc;
1554 ctx.progress_data.progctx = progctx;
1556 ctx.num_bytes_to_compress = find_raw_copy_blobs(blob_list,
1557 write_resource_flags,
1562 if (ctx.num_bytes_to_compress == 0)
1563 goto out_write_raw_copy_resources;
1565 /* Unless uncompressed output was required, allocate a chunk_compressor
1566 * to do compression. There are serial and parallel implementations of
1567 * the chunk_compressor interface. We default to parallel using the
1568 * specified number of threads, unless the upper bound on the number
1569 * bytes needing to be compressed is less than a heuristic value. */
1570 if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) {
1572 #ifdef ENABLE_MULTITHREADED_COMPRESSION
1573 if (ctx.num_bytes_to_compress > max(2000000, out_chunk_size)) {
1574 ret = new_parallel_chunk_compressor(out_ctype,
1579 WARNING("Couldn't create parallel chunk compressor: %"TS".\n"
1580 " Falling back to single-threaded compression.",
1581 wimlib_get_error_string(ret));
1586 if (ctx.compressor == NULL) {
1587 ret = new_serial_chunk_compressor(out_ctype, out_chunk_size,
1590 goto out_destroy_context;
1595 ctx.progress_data.progress.write_streams.num_threads = ctx.compressor->num_threads;
1597 ctx.progress_data.progress.write_streams.num_threads = 1;
1599 INIT_LIST_HEAD(&ctx.blobs_being_compressed);
1600 INIT_LIST_HEAD(&ctx.blobs_in_solid_resource);
1602 ret = call_progress(ctx.progress_data.progfunc,
1603 WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
1604 &ctx.progress_data.progress,
1605 ctx.progress_data.progctx);
1607 goto out_destroy_context;
1609 if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
1610 ret = begin_write_resource(&ctx, ctx.num_bytes_to_compress);
1612 goto out_destroy_context;
1615 /* Read the list of blobs needing to be compressed, using the specified
1616 * callbacks to execute processing of the data. */
1618 struct read_blob_list_callbacks cbs = {
1619 .begin_blob = write_blob_begin_read,
1620 .begin_blob_ctx = &ctx,
1621 .consume_chunk = write_blob_process_chunk,
1622 .consume_chunk_ctx = &ctx,
1623 .end_blob = write_blob_end_read,
1624 .end_blob_ctx = &ctx,
1627 ret = read_blob_list(blob_list,
1628 offsetof(struct blob_descriptor, write_blobs_list),
1630 BLOB_LIST_ALREADY_SORTED |
1631 VERIFY_BLOB_HASHES |
1632 COMPUTE_MISSING_BLOB_HASHES);
1635 goto out_destroy_context;
1637 ret = finish_remaining_chunks(&ctx);
1639 goto out_destroy_context;
1641 if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
1642 struct wim_reshdr reshdr;
1643 struct blob_descriptor *blob;
1646 ret = end_write_resource(&ctx, &reshdr);
1648 goto out_destroy_context;
1651 list_for_each_entry(blob, &ctx.blobs_in_solid_resource, write_blobs_list) {
1652 blob->out_reshdr.size_in_wim = blob->size;
1653 blob->out_reshdr.flags = reshdr_flags_for_blob(blob) |
1654 WIM_RESHDR_FLAG_SOLID;
1655 blob->out_reshdr.uncompressed_size = 0;
1656 blob->out_reshdr.offset_in_wim = offset_in_res;
1657 blob->out_res_offset_in_wim = reshdr.offset_in_wim;
1658 blob->out_res_size_in_wim = reshdr.size_in_wim;
1659 blob->out_res_uncompressed_size = reshdr.uncompressed_size;
1660 offset_in_res += blob->size;
1662 wimlib_assert(offset_in_res == reshdr.uncompressed_size);
1665 out_write_raw_copy_resources:
1666 /* Copy any compressed resources for which the raw data can be reused
1667 * without decompression. */
1668 ret = write_raw_copy_resources(&raw_copy_blobs, ctx.out_fd,
1669 &ctx.progress_data);
1671 out_destroy_context:
1672 FREE(ctx.chunk_csizes);
1674 ctx.compressor->destroy(ctx.compressor);
1680 write_file_data_blobs(WIMStruct *wim,
1681 struct list_head *blob_list,
1683 unsigned num_threads,
1684 struct filter_context *filter_ctx)
1688 int write_resource_flags;
1690 write_resource_flags = write_flags_to_resource_flags(write_flags);
1692 if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
1693 out_chunk_size = wim->out_solid_chunk_size;
1694 out_ctype = wim->out_solid_compression_type;
1696 out_chunk_size = wim->out_chunk_size;
1697 out_ctype = wim->out_compression_type;
1700 return write_blob_list(blob_list,
1702 write_resource_flags,
1712 /* Write the contents of the specified blob as a WIM resource. */
1714 write_wim_resource(struct blob_descriptor *blob,
1715 struct filedes *out_fd,
1718 int write_resource_flags)
1720 LIST_HEAD(blob_list);
1721 list_add(&blob->write_blobs_list, &blob_list);
1722 blob->will_be_in_output_wim = 1;
1723 return write_blob_list(&blob_list,
1725 write_resource_flags & ~WRITE_RESOURCE_FLAG_SOLID,
1735 /* Write the contents of the specified buffer as a WIM resource. */
1737 write_wim_resource_from_buffer(const void *buf,
1740 struct filedes *out_fd,
1743 struct wim_reshdr *out_reshdr,
1745 int write_resource_flags)
1748 struct blob_descriptor blob;
1750 blob_set_is_located_in_attached_buffer(&blob, (void *)buf, buf_size);
1751 sha1_buffer(buf, buf_size, blob.hash);
1753 blob.is_metadata = is_metadata;
1755 ret = write_wim_resource(&blob, out_fd, out_ctype, out_chunk_size,
1756 write_resource_flags);
1760 copy_reshdr(out_reshdr, &blob.out_reshdr);
1763 copy_hash(hash_ret, blob.hash);
1767 struct blob_size_table {
1768 struct hlist_head *array;
1774 init_blob_size_table(struct blob_size_table *tab, size_t capacity)
1776 tab->array = CALLOC(capacity, sizeof(tab->array[0]));
1777 if (tab->array == NULL)
1778 return WIMLIB_ERR_NOMEM;
1779 tab->num_entries = 0;
1780 tab->capacity = capacity;
1785 destroy_blob_size_table(struct blob_size_table *tab)
1791 blob_size_table_insert(struct blob_descriptor *blob, void *_tab)
1793 struct blob_size_table *tab = _tab;
1795 struct blob_descriptor *same_size_blob;
1797 pos = hash_u64(blob->size) % tab->capacity;
1798 blob->unique_size = 1;
1799 hlist_for_each_entry(same_size_blob, &tab->array[pos], hash_list_2) {
1800 if (same_size_blob->size == blob->size) {
1801 blob->unique_size = 0;
1802 same_size_blob->unique_size = 0;
1807 hlist_add_head(&blob->hash_list_2, &tab->array[pos]);
1812 struct find_blobs_ctx {
1815 struct list_head blob_list;
1816 struct blob_size_table blob_size_tab;
1820 reference_blob_for_write(struct blob_descriptor *blob,
1821 struct list_head *blob_list, u32 nref)
1823 if (!blob->will_be_in_output_wim) {
1824 blob->out_refcnt = 0;
1825 list_add_tail(&blob->write_blobs_list, blob_list);
1826 blob->will_be_in_output_wim = 1;
1828 blob->out_refcnt += nref;
1832 fully_reference_blob_for_write(struct blob_descriptor *blob, void *_blob_list)
1834 struct list_head *blob_list = _blob_list;
1835 blob->will_be_in_output_wim = 0;
1836 reference_blob_for_write(blob, blob_list, blob->refcnt);
1841 inode_find_blobs_to_reference(const struct wim_inode *inode,
1842 const struct blob_table *table,
1843 struct list_head *blob_list)
1845 wimlib_assert(inode->i_nlink > 0);
1847 for (unsigned i = 0; i < inode->i_num_streams; i++) {
1848 struct blob_descriptor *blob;
1851 blob = stream_blob(&inode->i_streams[i], table);
1853 reference_blob_for_write(blob, blob_list, inode->i_nlink);
1855 hash = stream_hash(&inode->i_streams[i]);
1856 if (!is_zero_hash(hash))
1857 return blob_not_found_error(inode, hash);
1864 do_blob_set_not_in_output_wim(struct blob_descriptor *blob, void *_ignore)
1866 blob->will_be_in_output_wim = 0;
1871 image_find_blobs_to_reference(WIMStruct *wim)
1873 struct wim_image_metadata *imd;
1874 struct wim_inode *inode;
1875 struct blob_descriptor *blob;
1876 struct list_head *blob_list;
1879 imd = wim_get_current_image_metadata(wim);
1881 image_for_each_unhashed_blob(blob, imd)
1882 blob->will_be_in_output_wim = 0;
1884 blob_list = wim->private;
1885 image_for_each_inode(inode, imd) {
1886 ret = inode_find_blobs_to_reference(inode,
1896 prepare_unfiltered_list_of_blobs_in_output_wim(WIMStruct *wim,
1899 struct list_head *blob_list_ret)
1903 INIT_LIST_HEAD(blob_list_ret);
1905 if (blobs_ok && (image == WIMLIB_ALL_IMAGES ||
1906 (image == 1 && wim->hdr.image_count == 1)))
1908 /* Fast case: Assume that all blobs are being written and that
1909 * the reference counts are correct. */
1910 struct blob_descriptor *blob;
1911 struct wim_image_metadata *imd;
1914 for_blob_in_table(wim->blob_table,
1915 fully_reference_blob_for_write,
1918 for (i = 0; i < wim->hdr.image_count; i++) {
1919 imd = wim->image_metadata[i];
1920 image_for_each_unhashed_blob(blob, imd)
1921 fully_reference_blob_for_write(blob, blob_list_ret);
1924 /* Slow case: Walk through the images being written and
1925 * determine the blobs referenced. */
1926 for_blob_in_table(wim->blob_table,
1927 do_blob_set_not_in_output_wim, NULL);
1928 wim->private = blob_list_ret;
1929 ret = for_image(wim, image, image_find_blobs_to_reference);
1937 struct insert_other_if_hard_filtered_ctx {
1938 struct blob_size_table *tab;
1939 struct filter_context *filter_ctx;
1943 insert_other_if_hard_filtered(struct blob_descriptor *blob, void *_ctx)
1945 struct insert_other_if_hard_filtered_ctx *ctx = _ctx;
1947 if (!blob->will_be_in_output_wim &&
1948 blob_hard_filtered(blob, ctx->filter_ctx))
1949 blob_size_table_insert(blob, ctx->tab);
1954 determine_blob_size_uniquity(struct list_head *blob_list,
1955 struct blob_table *lt,
1956 struct filter_context *filter_ctx)
1959 struct blob_size_table tab;
1960 struct blob_descriptor *blob;
1962 ret = init_blob_size_table(&tab, 9001);
1966 if (may_hard_filter_blobs(filter_ctx)) {
1967 struct insert_other_if_hard_filtered_ctx ctx = {
1969 .filter_ctx = filter_ctx,
1971 for_blob_in_table(lt, insert_other_if_hard_filtered, &ctx);
1974 list_for_each_entry(blob, blob_list, write_blobs_list)
1975 blob_size_table_insert(blob, &tab);
1977 destroy_blob_size_table(&tab);
1982 filter_blob_list_for_write(struct list_head *blob_list,
1983 struct filter_context *filter_ctx)
1985 struct blob_descriptor *blob, *tmp;
1987 list_for_each_entry_safe(blob, tmp, blob_list, write_blobs_list) {
1988 int status = blob_filtered(blob, filter_ctx);
1995 /* Soft filtered. */
1997 /* Hard filtered. */
1998 blob->will_be_in_output_wim = 0;
1999 list_del(&blob->blob_table_list);
2001 list_del(&blob->write_blobs_list);
2007 * prepare_blob_list_for_write() -
2009 * Prepare the list of blobs to write for writing a WIM containing the specified
2010 * image(s) with the specified write flags.
2013 * The WIMStruct on whose behalf the write is occurring.
2016 * Image(s) from the WIM to write; may be WIMLIB_ALL_IMAGES.
2019 * WIMLIB_WRITE_FLAG_* flags for the write operation:
2021 * STREAMS_OK: For writes of all images, assume that all blobs in the blob
2022 * table of @wim and the per-image lists of unhashed blobs should be taken
2023 * as-is, and image metadata should not be searched for references. This
2024 * does not exclude filtering with OVERWRITE and SKIP_EXTERNAL_WIMS, below.
2026 * OVERWRITE: Blobs already present in @wim shall not be returned in
2029 * SKIP_EXTERNAL_WIMS: Blobs already present in a WIM file, but not @wim,
2030 * shall be returned in neither @blob_list_ret nor @blob_table_list_ret.
2033 * List of blobs, linked by write_blobs_list, that need to be written will
2036 * Note that this function assumes that unhashed blobs will be written; it
2037 * does not take into account that they may become duplicates when actually
2040 * @blob_table_list_ret
2041 * List of blobs, linked by blob_table_list, that need to be included in
2042 * the WIM's blob table will be returned here. This will be a superset of
2043 * the blobs in @blob_list_ret.
2045 * This list will be a proper superset of @blob_list_ret if and only if
2046 * WIMLIB_WRITE_FLAG_OVERWRITE was specified in @write_flags and some of
2047 * the blobs that would otherwise need to be written were already located
2050 * All blobs in this list will have @out_refcnt set to the number of
2051 * references to the blob in the output WIM. If
2052 * WIMLIB_WRITE_FLAG_STREAMS_OK was specified in @write_flags, @out_refcnt
2053 * may be as low as 0.
2056 * A context for queries of blob filter status with blob_filtered() is
2057 * returned in this location.
2059 * In addition, @will_be_in_output_wim will be set to 1 in all blobs inserted
2060 * into @blob_table_list_ret and to 0 in all blobs in the blob table of @wim not
2061 * inserted into @blob_table_list_ret.
2063 * Still furthermore, @unique_size will be set to 1 on all blobs in
2064 * @blob_list_ret that have unique size among all blobs in @blob_list_ret and
2065 * among all blobs in the blob table of @wim that are ineligible for being
2066 * written due to filtering.
2068 * Returns 0 on success; nonzero on read error, memory allocation error, or
2072 prepare_blob_list_for_write(WIMStruct *wim, int image,
2074 struct list_head *blob_list_ret,
2075 struct list_head *blob_table_list_ret,
2076 struct filter_context *filter_ctx_ret)
2079 struct blob_descriptor *blob;
2081 filter_ctx_ret->write_flags = write_flags;
2082 filter_ctx_ret->wim = wim;
2084 ret = prepare_unfiltered_list_of_blobs_in_output_wim(
2087 write_flags & WIMLIB_WRITE_FLAG_STREAMS_OK,
2092 INIT_LIST_HEAD(blob_table_list_ret);
2093 list_for_each_entry(blob, blob_list_ret, write_blobs_list)
2094 list_add_tail(&blob->blob_table_list, blob_table_list_ret);
2096 ret = determine_blob_size_uniquity(blob_list_ret, wim->blob_table,
2101 if (may_filter_blobs(filter_ctx_ret))
2102 filter_blob_list_for_write(blob_list_ret, filter_ctx_ret);
2108 write_file_data(WIMStruct *wim, int image, int write_flags,
2109 unsigned num_threads,
2110 struct list_head *blob_list_override,
2111 struct list_head *blob_table_list_ret)
2114 struct list_head _blob_list;
2115 struct list_head *blob_list;
2116 struct blob_descriptor *blob;
2117 struct filter_context _filter_ctx;
2118 struct filter_context *filter_ctx;
2120 if (blob_list_override == NULL) {
2121 /* Normal case: prepare blob list from image(s) being written.
2123 blob_list = &_blob_list;
2124 filter_ctx = &_filter_ctx;
2125 ret = prepare_blob_list_for_write(wim, image, write_flags,
2127 blob_table_list_ret,
2132 /* Currently only as a result of wimlib_split() being called:
2133 * use blob list already explicitly provided. Use existing
2134 * reference counts. */
2135 blob_list = blob_list_override;
2137 INIT_LIST_HEAD(blob_table_list_ret);
2138 list_for_each_entry(blob, blob_list, write_blobs_list) {
2139 blob->out_refcnt = blob->refcnt;
2140 blob->will_be_in_output_wim = 1;
2141 blob->unique_size = 0;
2142 list_add_tail(&blob->blob_table_list, blob_table_list_ret);
2146 return write_file_data_blobs(wim,
2154 write_metadata_resources(WIMStruct *wim, int image, int write_flags)
2159 int write_resource_flags;
2161 if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)
2164 write_resource_flags = write_flags_to_resource_flags(write_flags);
2166 write_resource_flags &= ~WRITE_RESOURCE_FLAG_SOLID;
2168 ret = call_progress(wim->progfunc,
2169 WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN,
2170 NULL, wim->progctx);
2174 if (image == WIMLIB_ALL_IMAGES) {
2176 end_image = wim->hdr.image_count;
2178 start_image = image;
2182 for (int i = start_image; i <= end_image; i++) {
2183 struct wim_image_metadata *imd;
2185 imd = wim->image_metadata[i - 1];
2186 /* Build a new metadata resource only if image was modified from
2187 * the original (or was newly added). Otherwise just copy the
2189 if (imd->modified) {
2190 ret = write_metadata_resource(wim, i,
2191 write_resource_flags);
2192 } else if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) {
2193 blob_set_out_reshdr_for_reuse(imd->metadata_blob);
2196 ret = write_wim_resource(imd->metadata_blob,
2198 wim->out_compression_type,
2199 wim->out_chunk_size,
2200 write_resource_flags);
2206 return call_progress(wim->progfunc,
2207 WIMLIB_PROGRESS_MSG_WRITE_METADATA_END,
2208 NULL, wim->progctx);
2212 open_wim_writable(WIMStruct *wim, const tchar *path, int open_flags)
2214 int raw_fd = topen(path, open_flags | O_BINARY, 0644);
2216 ERROR_WITH_ERRNO("Failed to open \"%"TS"\" for writing", path);
2217 return WIMLIB_ERR_OPEN;
2219 filedes_init(&wim->out_fd, raw_fd);
2224 close_wim_writable(WIMStruct *wim, int write_flags)
2228 if (!(write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR))
2229 if (filedes_valid(&wim->out_fd))
2230 if (filedes_close(&wim->out_fd))
2231 ret = WIMLIB_ERR_WRITE;
2232 filedes_invalidate(&wim->out_fd);
2237 cmp_blobs_by_out_rdesc(const void *p1, const void *p2)
2239 const struct blob_descriptor *blob1, *blob2;
2241 blob1 = *(const struct blob_descriptor**)p1;
2242 blob2 = *(const struct blob_descriptor**)p2;
2244 if (blob1->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
2245 if (blob2->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
2246 if (blob1->out_res_offset_in_wim != blob2->out_res_offset_in_wim)
2247 return cmp_u64(blob1->out_res_offset_in_wim,
2248 blob2->out_res_offset_in_wim);
2253 if (blob2->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID)
2256 return cmp_u64(blob1->out_reshdr.offset_in_wim,
2257 blob2->out_reshdr.offset_in_wim);
2261 write_blob_table(WIMStruct *wim, int image, int write_flags,
2262 struct list_head *blob_table_list)
2266 /* Set output resource metadata for blobs already present in WIM. */
2267 if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) {
2268 struct blob_descriptor *blob;
2269 list_for_each_entry(blob, blob_table_list, blob_table_list) {
2270 if (blob->blob_location == BLOB_IN_WIM &&
2271 blob->rdesc->wim == wim)
2273 blob_set_out_reshdr_for_reuse(blob);
2278 ret = sort_blob_list(blob_table_list,
2279 offsetof(struct blob_descriptor, blob_table_list),
2280 cmp_blobs_by_out_rdesc);
2284 /* Add entries for metadata resources. */
2285 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)) {
2289 if (image == WIMLIB_ALL_IMAGES) {
2291 end_image = wim->hdr.image_count;
2293 start_image = image;
2297 /* Push metadata blob table entries onto the front of the list
2298 * in reverse order, so that they're written in order.
2300 for (int i = end_image; i >= start_image; i--) {
2301 struct blob_descriptor *metadata_blob;
2303 metadata_blob = wim->image_metadata[i - 1]->metadata_blob;
2304 wimlib_assert(metadata_blob->out_reshdr.flags & WIM_RESHDR_FLAG_METADATA);
2305 metadata_blob->out_refcnt = 1;
2306 list_add(&metadata_blob->blob_table_list, blob_table_list);
2310 return write_blob_table_from_blob_list(blob_table_list,
2312 wim->out_hdr.part_number,
2313 &wim->out_hdr.blob_table_reshdr,
2314 write_flags_to_resource_flags(write_flags));
2318 * Finish writing a WIM file: write the blob table, xml data, and integrity
2319 * table, then overwrite the WIM header.
2321 * The output file descriptor is closed on success, except when writing to a
2322 * user-specified file descriptor (WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR set).
2325 finish_write(WIMStruct *wim, int image, int write_flags,
2326 struct list_head *blob_table_list)
2328 int write_resource_flags;
2329 off_t old_blob_table_end = 0;
2330 struct integrity_table *old_integrity_table = NULL;
2331 off_t new_blob_table_end;
2335 write_resource_flags = write_flags_to_resource_flags(write_flags);
2337 /* In the WIM header, there is room for the resource entry for a
2338 * metadata resource labeled as the "boot metadata". This entry should
2339 * be zeroed out if there is no bootable image (boot_idx 0). Otherwise,
2340 * it should be a copy of the resource entry for the image that is
2341 * marked as bootable. */
2342 if (wim->out_hdr.boot_idx == 0) {
2343 zero_reshdr(&wim->out_hdr.boot_metadata_reshdr);
2345 copy_reshdr(&wim->out_hdr.boot_metadata_reshdr,
2346 &wim->image_metadata[
2347 wim->out_hdr.boot_idx - 1]->metadata_blob->out_reshdr);
2350 /* If overwriting the WIM file containing an integrity table in-place,
2351 * we'd like to re-use the information in the old integrity table
2352 * instead of recalculating it. But we might overwrite the old
2353 * integrity table when we expand the XML data. Read it into memory
2355 if ((write_flags & (WIMLIB_WRITE_FLAG_OVERWRITE |
2356 WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)) ==
2357 (WIMLIB_WRITE_FLAG_OVERWRITE |
2358 WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)
2359 && wim_has_integrity_table(wim))
2361 old_blob_table_end = wim->hdr.blob_table_reshdr.offset_in_wim +
2362 wim->hdr.blob_table_reshdr.size_in_wim;
2363 (void)read_integrity_table(wim,
2364 old_blob_table_end - WIM_HEADER_DISK_SIZE,
2365 &old_integrity_table);
2366 /* If we couldn't read the old integrity table, we can still
2367 * re-calculate the full integrity table ourselves. Hence the
2368 * ignoring of the return value. */
2371 /* Write blob table if needed. */
2372 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_NEW_BLOBS)) {
2373 ret = write_blob_table(wim, image, write_flags,
2376 free_integrity_table(old_integrity_table);
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);
2389 free_integrity_table(old_integrity_table);
2393 /* Write integrity table if needed. */
2394 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
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);
2406 free_integrity_table(old_integrity_table);
2411 new_blob_table_end = wim->out_hdr.blob_table_reshdr.offset_in_wim +
2412 wim->out_hdr.blob_table_reshdr.size_in_wim;
2414 ret = write_integrity_table(wim,
2417 old_integrity_table);
2418 free_integrity_table(old_integrity_table);
2422 /* No integrity table. */
2423 zero_reshdr(&wim->out_hdr.integrity_table_reshdr);
2426 /* Now that all information in the WIM header has been determined, the
2427 * preliminary header written earlier can be overwritten, the header of
2428 * the existing WIM file can be overwritten, or the final header can be
2429 * written to the end of the pipable WIM. */
2430 wim->out_hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2431 if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
2432 ret = write_wim_header(&wim->out_hdr, &wim->out_fd, wim->out_fd.offset);
2434 ret = write_wim_header(&wim->out_hdr, &wim->out_fd, 0);
2438 /* Possibly sync file data to disk before closing. On POSIX systems, it
2439 * is necessary to do this before using rename() to overwrite an
2440 * existing file with a new file. Otherwise, data loss would occur if
2441 * the system is abruptly terminated when the metadata for the rename
2442 * operation has been written to disk, but the new file data has not.
2444 if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) {
2445 if (fsync(wim->out_fd.fd)) {
2446 ERROR_WITH_ERRNO("Error syncing data to WIM file");
2447 return WIMLIB_ERR_WRITE;
2451 if (close_wim_writable(wim, write_flags)) {
2452 ERROR_WITH_ERRNO("Failed to close the output WIM file");
2453 return WIMLIB_ERR_WRITE;
2459 #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK)
2461 /* Set advisory lock on WIM file (if not already done so) */
2463 lock_wim_for_append(WIMStruct *wim)
2465 if (wim->locked_for_append)
2467 if (!flock(wim->in_fd.fd, LOCK_EX | LOCK_NB)) {
2468 wim->locked_for_append = 1;
2471 if (errno != EWOULDBLOCK)
2473 return WIMLIB_ERR_ALREADY_LOCKED;
2476 /* Remove advisory lock on WIM file (if present) */
2478 unlock_wim_for_append(WIMStruct *wim)
2480 if (wim->locked_for_append) {
2481 flock(wim->in_fd.fd, LOCK_UN);
2482 wim->locked_for_append = 0;
2488 * write_pipable_wim():
2490 * Perform the intermediate stages of creating a "pipable" WIM (i.e. a WIM
2491 * capable of being applied from a pipe).
2493 * Pipable WIMs are a wimlib-specific modification of the WIM format such that
2494 * images can be applied from them sequentially when the file data is sent over
2495 * a pipe. In addition, a pipable WIM can be written sequentially to a pipe.
2496 * The modifications made to the WIM format for pipable WIMs are:
2498 * - Magic characters in header are "WLPWM\0\0\0" (wimlib pipable WIM) instead
2499 * of "MSWIM\0\0\0". This lets wimlib know that the WIM is pipable and also
2500 * stops other software from trying to read the file as a normal WIM.
2502 * - The header at the beginning of the file does not contain all the normal
2503 * information; in particular it will have all 0's for the blob table and XML
2504 * data resource entries. This is because this information cannot be
2505 * determined until the blob table and XML data have been written.
2506 * Consequently, wimlib will write the full header at the very end of the
2507 * file. The header at the end, however, is only used when reading the WIM
2508 * from a seekable file (not a pipe).
2510 * - An extra copy of the XML data is placed directly after the header. This
2511 * allows image names and sizes to be determined at an appropriate time when
2512 * reading the WIM from a pipe. This copy of the XML data is ignored if the
2513 * WIM is read from a seekable file (not a pipe).
2515 * - Solid resources are not allowed. Each blob is always stored in its own
2518 * - The format of resources, or blobs, has been modified to allow them to be
2519 * used before the "blob table" has been read. Each blob is prefixed with a
2520 * `struct pwm_blob_hdr' that is basically an abbreviated form of `struct
2521 * blob_descriptor_disk' that only contains the SHA-1 message digest,
2522 * uncompressed blob size, and flags that indicate whether the blob is
2523 * compressed. The data of uncompressed blobs then follows literally, while
2524 * the data of compressed blobs follows in a modified format. Compressed
2525 * blobs do not begin with a chunk table, since the chunk table cannot be
2526 * written until all chunks have been compressed. Instead, each compressed
2527 * chunk is prefixed by a `struct pwm_chunk_hdr' that gives its size.
2528 * Furthermore, the chunk table is written at the end of the resource instead
2529 * of the start. Note: chunk offsets are given in the chunk table as if the
2530 * `struct pwm_chunk_hdr's were not present; also, the chunk table is only
2531 * used if the WIM is being read from a seekable file (not a pipe).
2533 * - Metadata blobs always come before non-metadata blobs. (This does not by
2534 * itself constitute an incompatibility with normal WIMs, since this is valid
2537 * - At least up to the end of the blobs, all components must be packed as
2538 * tightly as possible; there cannot be any "holes" in the WIM. (This does
2539 * not by itself consititute an incompatibility with normal WIMs, since this
2540 * is valid in normal WIMs.)
2542 * Note: the blob table, XML data, and header at the end are not used when
2543 * applying from a pipe. They exist to support functionality such as image
2544 * application and export when the WIM is *not* read from a pipe.
2546 * Layout of pipable WIM:
2548 * ---------+----------+--------------------+----------------+--------------+-----------+--------+
2549 * | Header | XML data | Metadata resources | File resources | Blob table | XML data | Header |
2550 * ---------+----------+--------------------+----------------+--------------+-----------+--------+
2552 * Layout of normal WIM:
2554 * +--------+-----------------------------+-------------------------+
2555 * | Header | File and metadata resources | Blob table | XML data |
2556 * +--------+-----------------------------+-------------------------+
2558 * An optional integrity table can follow the final XML data in both normal and
2559 * pipable WIMs. However, due to implementation details, wimlib currently can
2560 * only include an integrity table in a pipable WIM when writing it to a
2561 * seekable file (not a pipe).
2563 * Do note that since pipable WIMs are not supported by Microsoft's software,
2564 * wimlib does not create them unless explicitly requested (with
2565 * WIMLIB_WRITE_FLAG_PIPABLE) and as stated above they use different magic
2566 * characters to identify the file.
2569 write_pipable_wim(WIMStruct *wim, int image, int write_flags,
2570 unsigned num_threads,
2571 struct list_head *blob_list_override,
2572 struct list_head *blob_table_list_ret)
2575 struct wim_reshdr xml_reshdr;
2577 WARNING("Creating a pipable WIM, which will "
2579 " with Microsoft's software (WIMGAPI/ImageX/DISM).");
2581 /* At this point, the header at the beginning of the file has already
2584 /* For efficiency, when wimlib adds an image to the WIM with
2585 * wimlib_add_image(), the SHA-1 message digests of files are not
2586 * calculated; instead, they are calculated while the files are being
2587 * written. However, this does not work when writing a pipable WIM,
2588 * since when writing a blob to a pipable WIM, its SHA-1 message digest
2589 * needs to be known before the blob data is written. Therefore, before
2590 * getting much farther, we need to pre-calculate the SHA-1 message
2591 * digests of all blobs that will be written. */
2592 ret = wim_checksum_unhashed_blobs(wim);
2596 /* Write extra copy of the XML data. */
2597 ret = write_wim_xml_data(wim, image, WIM_TOTALBYTES_OMIT,
2598 &xml_reshdr, WRITE_RESOURCE_FLAG_PIPABLE);
2602 /* Write metadata resources for the image(s) being included in the
2604 ret = write_metadata_resources(wim, image, write_flags);
2608 /* Write file data needed for the image(s) being included in the output
2609 * WIM, or file data needed for the split WIM part. */
2610 return write_file_data(wim, image, write_flags,
2611 num_threads, blob_list_override,
2612 blob_table_list_ret);
2614 /* The blob table, XML data, and header at end are handled by
2615 * finish_write(). */
2619 should_default_to_solid_compression(WIMStruct *wim, int write_flags)
2621 return wim->out_hdr.wim_version == WIM_VERSION_SOLID &&
2622 !(write_flags & (WIMLIB_WRITE_FLAG_SOLID |
2623 WIMLIB_WRITE_FLAG_PIPABLE)) &&
2624 wim_has_solid_resources(wim);
2627 /* Write a standalone WIM or split WIM (SWM) part to a new file or to a file
2630 write_wim_part(WIMStruct *wim,
2631 const void *path_or_fd,
2634 unsigned num_threads,
2635 unsigned part_number,
2636 unsigned total_parts,
2637 struct list_head *blob_list_override,
2641 struct list_head blob_table_list;
2643 /* Internally, this is always called with a valid part number and total
2645 wimlib_assert(total_parts >= 1);
2646 wimlib_assert(part_number >= 1 && part_number <= total_parts);
2648 /* A valid image (or all images) must be specified. */
2649 if (image != WIMLIB_ALL_IMAGES &&
2650 (image < 1 || image > wim->hdr.image_count))
2651 return WIMLIB_ERR_INVALID_IMAGE;
2653 /* If we need to write metadata resources, make sure the ::WIMStruct has
2654 * the needed information attached (e.g. is not a resource-only WIM,
2655 * such as a non-first part of a split WIM). */
2656 if (!wim_has_metadata(wim) &&
2657 !(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA))
2658 return WIMLIB_ERR_METADATA_NOT_FOUND;
2660 /* Check for contradictory flags. */
2661 if ((write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2662 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))
2663 == (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2664 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))
2665 return WIMLIB_ERR_INVALID_PARAM;
2667 if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2668 WIMLIB_WRITE_FLAG_NOT_PIPABLE))
2669 == (WIMLIB_WRITE_FLAG_PIPABLE |
2670 WIMLIB_WRITE_FLAG_NOT_PIPABLE))
2671 return WIMLIB_ERR_INVALID_PARAM;
2673 /* Include an integrity table by default if no preference was given and
2674 * the WIM already had an integrity table. */
2675 if (!(write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2676 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))) {
2677 if (wim_has_integrity_table(wim))
2678 write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY;
2681 /* Write a pipable WIM by default if no preference was given and the WIM
2682 * was already pipable. */
2683 if (!(write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2684 WIMLIB_WRITE_FLAG_NOT_PIPABLE))) {
2685 if (wim_is_pipable(wim))
2686 write_flags |= WIMLIB_WRITE_FLAG_PIPABLE;
2689 if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
2690 WIMLIB_WRITE_FLAG_SOLID))
2691 == (WIMLIB_WRITE_FLAG_PIPABLE |
2692 WIMLIB_WRITE_FLAG_SOLID))
2694 ERROR("Solid compression is unsupported in pipable WIMs");
2695 return WIMLIB_ERR_INVALID_PARAM;
2698 /* Start initializing the new file header. */
2699 memset(&wim->out_hdr, 0, sizeof(wim->out_hdr));
2701 /* Set the magic number. */
2702 if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
2703 wim->out_hdr.magic = PWM_MAGIC;
2705 wim->out_hdr.magic = WIM_MAGIC;
2707 /* Set the version number. */
2708 if ((write_flags & WIMLIB_WRITE_FLAG_SOLID) ||
2709 wim->out_compression_type == WIMLIB_COMPRESSION_TYPE_LZMS)
2710 wim->out_hdr.wim_version = WIM_VERSION_SOLID;
2712 wim->out_hdr.wim_version = WIM_VERSION_DEFAULT;
2714 /* Default to solid compression if it is valid in the chosen WIM file
2715 * format and the WIMStruct references any solid resources. This is
2716 * useful when exporting an image from a solid WIM. */
2717 if (should_default_to_solid_compression(wim, write_flags))
2718 write_flags |= WIMLIB_WRITE_FLAG_SOLID;
2720 /* Set the header flags. */
2721 wim->out_hdr.flags = (wim->hdr.flags & (WIM_HDR_FLAG_RP_FIX |
2722 WIM_HDR_FLAG_READONLY));
2723 if (total_parts != 1)
2724 wim->out_hdr.flags |= WIM_HDR_FLAG_SPANNED;
2725 if (wim->out_compression_type != WIMLIB_COMPRESSION_TYPE_NONE) {
2726 wim->out_hdr.flags |= WIM_HDR_FLAG_COMPRESSION;
2727 switch (wim->out_compression_type) {
2728 case WIMLIB_COMPRESSION_TYPE_XPRESS:
2729 wim->out_hdr.flags |= WIM_HDR_FLAG_COMPRESS_XPRESS;
2731 case WIMLIB_COMPRESSION_TYPE_LZX:
2732 wim->out_hdr.flags |= WIM_HDR_FLAG_COMPRESS_LZX;
2734 case WIMLIB_COMPRESSION_TYPE_LZMS:
2735 wim->out_hdr.flags |= WIM_HDR_FLAG_COMPRESS_LZMS;
2740 /* Set the chunk size. */
2741 wim->out_hdr.chunk_size = wim->out_chunk_size;
2744 if (write_flags & WIMLIB_WRITE_FLAG_RETAIN_GUID)
2745 guid = wim->hdr.guid;
2747 memcpy(wim->out_hdr.guid, guid, WIMLIB_GUID_LEN);
2749 randomize_byte_array(wim->out_hdr.guid, WIMLIB_GUID_LEN);
2751 /* Set the part number and total parts. */
2752 wim->out_hdr.part_number = part_number;
2753 wim->out_hdr.total_parts = total_parts;
2755 /* Set the image count. */
2756 if (image == WIMLIB_ALL_IMAGES)
2757 wim->out_hdr.image_count = wim->hdr.image_count;
2759 wim->out_hdr.image_count = 1;
2761 /* Set the boot index. */
2762 wim->out_hdr.boot_idx = 0;
2763 if (total_parts == 1) {
2764 if (image == WIMLIB_ALL_IMAGES)
2765 wim->out_hdr.boot_idx = wim->hdr.boot_idx;
2766 else if (image == wim->hdr.boot_idx)
2767 wim->out_hdr.boot_idx = 1;
2770 /* Set up the output file descriptor. */
2771 if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR) {
2772 /* File descriptor was explicitly provided. */
2773 filedes_init(&wim->out_fd, *(const int *)path_or_fd);
2774 if (!filedes_is_seekable(&wim->out_fd)) {
2775 /* The file descriptor is a pipe. */
2776 ret = WIMLIB_ERR_INVALID_PARAM;
2777 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
2779 if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
2780 ERROR("Can't include integrity check when "
2781 "writing pipable WIM to pipe!");
2786 /* Filename of WIM to write was provided; open file descriptor
2788 ret = open_wim_writable(wim, (const tchar*)path_or_fd,
2789 O_TRUNC | O_CREAT | O_RDWR);
2794 /* Write initial header. This is merely a "dummy" header since it
2795 * doesn't have resource entries filled in yet, so it will be
2796 * overwritten later (unless writing a pipable WIM). */
2797 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
2798 wim->out_hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2799 ret = write_wim_header(&wim->out_hdr, &wim->out_fd, wim->out_fd.offset);
2800 wim->out_hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS;
2804 /* Write file data and metadata resources. */
2805 if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE)) {
2806 /* Default case: create a normal (non-pipable) WIM. */
2807 ret = write_file_data(wim, image, write_flags,
2814 ret = write_metadata_resources(wim, image, write_flags);
2818 /* Non-default case: create pipable WIM. */
2819 ret = write_pipable_wim(wim, image, write_flags, num_threads,
2826 /* Write blob table, XML data, and (optional) integrity table. */
2827 ret = finish_write(wim, image, write_flags, &blob_table_list);
2829 (void)close_wim_writable(wim, write_flags);
2833 /* Write a standalone WIM to a file or file descriptor. */
2835 write_standalone_wim(WIMStruct *wim, const void *path_or_fd,
2836 int image, int write_flags, unsigned num_threads)
2838 return write_wim_part(wim, path_or_fd, image, write_flags,
2839 num_threads, 1, 1, NULL, NULL);
2842 /* API function documented in wimlib.h */
2844 wimlib_write(WIMStruct *wim, const tchar *path,
2845 int image, int write_flags, unsigned num_threads)
2847 if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC)
2848 return WIMLIB_ERR_INVALID_PARAM;
2850 if (path == NULL || path[0] == T('\0'))
2851 return WIMLIB_ERR_INVALID_PARAM;
2853 return write_standalone_wim(wim, path, image, write_flags, num_threads);
2856 /* API function documented in wimlib.h */
2858 wimlib_write_to_fd(WIMStruct *wim, int fd,
2859 int image, int write_flags, unsigned num_threads)
2861 if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC)
2862 return WIMLIB_ERR_INVALID_PARAM;
2865 return WIMLIB_ERR_INVALID_PARAM;
2867 write_flags |= WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR;
2869 return write_standalone_wim(wim, &fd, image, write_flags, num_threads);
2873 any_images_modified(WIMStruct *wim)
2875 for (int i = 0; i < wim->hdr.image_count; i++)
2876 if (wim->image_metadata[i]->modified)
2882 check_resource_offset(struct blob_descriptor *blob, void *_wim)
2884 const WIMStruct *wim = _wim;
2885 off_t end_offset = *(const off_t*)wim->private;
2887 if (blob->blob_location == BLOB_IN_WIM &&
2888 blob->rdesc->wim == wim &&
2889 blob->rdesc->offset_in_wim + blob->rdesc->size_in_wim > end_offset)
2890 return WIMLIB_ERR_RESOURCE_ORDER;
2894 /* Make sure no file or metadata resources are located after the XML data (or
2895 * integrity table if present)--- otherwise we can't safely overwrite the WIM in
2896 * place and we return WIMLIB_ERR_RESOURCE_ORDER. */
2898 check_resource_offsets(WIMStruct *wim, off_t end_offset)
2903 wim->private = &end_offset;
2904 ret = for_blob_in_table(wim->blob_table, check_resource_offset, wim);
2908 for (i = 0; i < wim->hdr.image_count; i++) {
2909 ret = check_resource_offset(wim->image_metadata[i]->metadata_blob, wim);
2917 * Overwrite a WIM, possibly appending new resources to it.
2919 * A WIM looks like (or is supposed to look like) the following:
2921 * Header (212 bytes)
2922 * Resources for metadata and files (variable size)
2923 * Blob table (variable size)
2924 * XML data (variable size)
2925 * Integrity table (optional) (variable size)
2927 * If we are not adding any new files or metadata, then the blob table is
2928 * unchanged--- so we only need to overwrite the XML data, integrity table, and
2929 * header. This operation is potentially unsafe if the program is abruptly
2930 * terminated while the XML data or integrity table are being overwritten, but
2931 * before the new header has been written. To partially alleviate this problem,
2932 * we write a temporary header after the XML data has been written. This may
2933 * prevent the WIM from becoming corrupted if the program is terminated while
2934 * the integrity table is being calculated (but no guarantees, due to write
2937 * If we are adding new blobs, including new file data as well as any metadata
2938 * for any new images, then the blob table needs to be changed, and those blobs
2939 * need to be written. In this case, we try to perform a safe update of the WIM
2940 * file by writing the blobs *after* the end of the previous WIM, then writing
2941 * the new blob table, XML data, and (optionally) integrity table following the
2942 * new blobs. This will produce a layout like the following:
2944 * Header (212 bytes)
2945 * (OLD) Resources for metadata and files (variable size)
2946 * (OLD) Blob table (variable size)
2947 * (OLD) XML data (variable size)
2948 * (OLD) Integrity table (optional) (variable size)
2949 * (NEW) Resources for metadata and files (variable size)
2950 * (NEW) Blob table (variable size)
2951 * (NEW) XML data (variable size)
2952 * (NEW) Integrity table (optional) (variable size)
2954 * At all points, the WIM is valid as nothing points to the new data yet. Then,
2955 * the header is overwritten to point to the new blob table, XML data, and
2956 * integrity table, to produce the following layout:
2958 * Header (212 bytes)
2959 * Resources for metadata and files (variable size)
2960 * Nothing (variable size)
2961 * Resources for metadata and files (variable size)
2962 * Blob table (variable size)
2963 * XML data (variable size)
2964 * Integrity table (optional) (variable size)
2966 * This method allows an image to be appended to a large WIM very quickly, and
2967 * is crash-safe except in the case of write re-ordering, but the disadvantage
2968 * is that a small hole is left in the WIM where the old blob table, xml data,
2969 * and integrity table were. (These usually only take up a small amount of
2970 * space compared to the blobs, however.)
2973 overwrite_wim_inplace(WIMStruct *wim, int write_flags, unsigned num_threads)
2977 u64 old_blob_table_end, old_xml_begin, old_xml_end;
2978 struct list_head blob_list;
2979 struct list_head blob_table_list;
2980 struct filter_context filter_ctx;
2982 /* Include an integrity table by default if no preference was given and
2983 * the WIM already had an integrity table. */
2984 if (!(write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
2985 WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)))
2986 if (wim_has_integrity_table(wim))
2987 write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY;
2989 /* Start preparing the updated file header. */
2990 memcpy(&wim->out_hdr, &wim->hdr, sizeof(wim->out_hdr));
2992 /* If using solid compression, the version number must be set to
2993 * WIM_VERSION_SOLID. */
2994 if (write_flags & WIMLIB_WRITE_FLAG_SOLID)
2995 wim->out_hdr.wim_version = WIM_VERSION_SOLID;
2997 /* Default to solid compression if it is valid in the chosen WIM file
2998 * format and the WIMStruct references any solid resources. This is
2999 * useful when updating a solid WIM. */
3000 if (should_default_to_solid_compression(wim, write_flags))
3001 write_flags |= WIMLIB_WRITE_FLAG_SOLID;
3003 /* Set additional flags for overwrite. */
3004 write_flags |= WIMLIB_WRITE_FLAG_OVERWRITE |
3005 WIMLIB_WRITE_FLAG_STREAMS_OK;
3007 /* Make sure there is no data after the XML data, except possibily an
3008 * integrity table. If this were the case, then this data would be
3010 old_xml_begin = wim->hdr.xml_data_reshdr.offset_in_wim;
3011 old_xml_end = old_xml_begin + wim->hdr.xml_data_reshdr.size_in_wim;
3012 old_blob_table_end = wim->hdr.blob_table_reshdr.offset_in_wim +
3013 wim->hdr.blob_table_reshdr.size_in_wim;
3014 if (wim_has_integrity_table(wim) &&
3015 wim->hdr.integrity_table_reshdr.offset_in_wim < old_xml_end) {
3016 WARNING("Didn't expect the integrity table to be before the XML data");
3017 ret = WIMLIB_ERR_RESOURCE_ORDER;
3021 if (old_blob_table_end > old_xml_begin) {
3022 WARNING("Didn't expect the blob table to be after the XML data");
3023 ret = WIMLIB_ERR_RESOURCE_ORDER;
3027 /* Set @old_wim_end, which indicates the point beyond which we don't
3028 * allow any file and metadata resources to appear without returning
3029 * WIMLIB_ERR_RESOURCE_ORDER (due to the fact that we would otherwise
3030 * overwrite these resources). */
3031 if (!wim->image_deletion_occurred && !any_images_modified(wim)) {
3032 /* If no images have been modified and no images have been
3033 * deleted, a new blob table does not need to be written. We
3034 * shall write the new XML data and optional integrity table
3035 * immediately after the blob table. Note that this may
3036 * overwrite an existing integrity table. */
3037 old_wim_end = old_blob_table_end;
3038 write_flags |= WIMLIB_WRITE_FLAG_NO_NEW_BLOBS;
3039 } else if (wim_has_integrity_table(wim)) {
3040 /* Old WIM has an integrity table; begin writing new blobs after
3042 old_wim_end = wim->hdr.integrity_table_reshdr.offset_in_wim +
3043 wim->hdr.integrity_table_reshdr.size_in_wim;
3045 /* No existing integrity table; begin writing new blobs after
3046 * the old XML data. */
3047 old_wim_end = old_xml_end;
3050 ret = check_resource_offsets(wim, old_wim_end);
3054 ret = prepare_blob_list_for_write(wim, WIMLIB_ALL_IMAGES, write_flags,
3055 &blob_list, &blob_table_list,
3060 if (write_flags & WIMLIB_WRITE_FLAG_NO_NEW_BLOBS)
3061 wimlib_assert(list_empty(&blob_list));
3063 ret = open_wim_writable(wim, wim->filename, O_RDWR);
3067 ret = lock_wim_for_append(wim);
3071 /* Set WIM_HDR_FLAG_WRITE_IN_PROGRESS flag in header. */
3072 wim->hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
3073 ret = write_wim_header_flags(wim->hdr.flags, &wim->out_fd);
3074 wim->hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS;
3076 ERROR_WITH_ERRNO("Error updating WIM header flags");
3077 goto out_unlock_wim;
3080 if (filedes_seek(&wim->out_fd, old_wim_end) == -1) {
3081 ERROR_WITH_ERRNO("Can't seek to end of WIM");
3082 ret = WIMLIB_ERR_WRITE;
3083 goto out_restore_hdr;
3086 ret = write_file_data_blobs(wim, &blob_list, write_flags,
3087 num_threads, &filter_ctx);
3091 ret = write_metadata_resources(wim, WIMLIB_ALL_IMAGES, write_flags);
3095 ret = finish_write(wim, WIMLIB_ALL_IMAGES, write_flags,
3100 unlock_wim_for_append(wim);
3104 if (!(write_flags & WIMLIB_WRITE_FLAG_NO_NEW_BLOBS)) {
3105 WARNING("Truncating \"%"TS"\" to its original size "
3106 "(%"PRIu64" bytes)", wim->filename, old_wim_end);
3107 /* Return value of ftruncate() is ignored because this is
3108 * already an error path. */
3109 (void)ftruncate(wim->out_fd.fd, old_wim_end);
3112 (void)write_wim_header_flags(wim->hdr.flags, &wim->out_fd);
3114 unlock_wim_for_append(wim);
3116 (void)close_wim_writable(wim, write_flags);
3122 overwrite_wim_via_tmpfile(WIMStruct *wim, int write_flags, unsigned num_threads)
3124 size_t wim_name_len;
3127 /* Write the WIM to a temporary file in the same directory as the
3129 wim_name_len = tstrlen(wim->filename);
3130 tchar tmpfile[wim_name_len + 10];
3131 tmemcpy(tmpfile, wim->filename, wim_name_len);
3132 randomize_char_array_with_alnum(tmpfile + wim_name_len, 9);
3133 tmpfile[wim_name_len + 9] = T('\0');
3135 ret = wimlib_write(wim, tmpfile, WIMLIB_ALL_IMAGES,
3137 WIMLIB_WRITE_FLAG_FSYNC |
3138 WIMLIB_WRITE_FLAG_RETAIN_GUID,
3145 if (filedes_valid(&wim->in_fd)) {
3146 filedes_close(&wim->in_fd);
3147 filedes_invalidate(&wim->in_fd);
3150 /* Rename the new WIM file to the original WIM file. Note: on Windows
3151 * this actually calls win32_rename_replacement(), not _wrename(), so
3152 * that removing the existing destination file can be handled. */
3153 ret = trename(tmpfile, wim->filename);
3155 ERROR_WITH_ERRNO("Failed to rename `%"TS"' to `%"TS"'",
3156 tmpfile, wim->filename);
3163 return WIMLIB_ERR_RENAME;
3166 union wimlib_progress_info progress;
3167 progress.rename.from = tmpfile;
3168 progress.rename.to = wim->filename;
3169 return call_progress(wim->progfunc, WIMLIB_PROGRESS_MSG_RENAME,
3170 &progress, wim->progctx);
3173 /* Determine if the specified WIM file may be updated by appending in-place
3174 * rather than writing and replacing it with an entirely new file. */
3176 can_overwrite_wim_inplace(const WIMStruct *wim, int write_flags)
3178 /* REBUILD flag forces full rebuild. */
3179 if (write_flags & WIMLIB_WRITE_FLAG_REBUILD)
3182 /* Image deletions cause full rebuild by default. */
3183 if (wim->image_deletion_occurred &&
3184 !(write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE))
3187 /* Pipable WIMs cannot be updated in place, nor can a non-pipable WIM be
3188 * turned into a pipable WIM in-place. */
3189 if (wim_is_pipable(wim) || (write_flags & WIMLIB_WRITE_FLAG_PIPABLE))
3192 /* The default compression type and compression chunk size selected for
3193 * the output WIM must be the same as those currently used for the WIM.
3195 if (wim->compression_type != wim->out_compression_type)
3197 if (wim->chunk_size != wim->out_chunk_size)
3203 /* API function documented in wimlib.h */
3205 wimlib_overwrite(WIMStruct *wim, int write_flags, unsigned num_threads)
3210 if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC)
3211 return WIMLIB_ERR_INVALID_PARAM;
3214 return WIMLIB_ERR_NO_FILENAME;
3216 orig_hdr_flags = wim->hdr.flags;
3217 if (write_flags & WIMLIB_WRITE_FLAG_IGNORE_READONLY_FLAG)
3218 wim->hdr.flags &= ~WIM_HDR_FLAG_READONLY;
3219 ret = can_modify_wim(wim);
3220 wim->hdr.flags = orig_hdr_flags;
3224 if (can_overwrite_wim_inplace(wim, write_flags)) {
3225 ret = overwrite_wim_inplace(wim, write_flags, num_threads);
3226 if (ret != WIMLIB_ERR_RESOURCE_ORDER)
3228 WARNING("Falling back to re-building entire WIM");
3230 return overwrite_wim_via_tmpfile(wim, write_flags, num_threads);