+ int ret;
+ u64 res_size_in_wim;
+ u64 res_uncompressed_size;
+ u64 res_offset_in_wim;
+
+ wimlib_assert(ctx->cur_write_stream_offset == ctx->cur_write_res_size ||
+ (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS));
+ res_uncompressed_size = ctx->cur_write_res_size;
+
+ if (ctx->compressor) {
+ ret = end_chunk_table(ctx, res_uncompressed_size,
+ &res_offset_in_wim, &res_size_in_wim);
+ if (ret)
+ return ret;
+ } else {
+ res_offset_in_wim = ctx->chunks_start_offset;
+ res_size_in_wim = ctx->out_fd->offset - res_offset_in_wim;
+ }
+ out_reshdr->uncompressed_size = res_uncompressed_size;
+ out_reshdr->size_in_wim = res_size_in_wim;
+ out_reshdr->offset_in_wim = res_offset_in_wim;
+ DEBUG("Finished writing resource: %"PRIu64" => %"PRIu64" @ %"PRIu64"",
+ res_uncompressed_size, res_size_in_wim, res_offset_in_wim);
+ return 0;
+}
+
+/* No more data streams of the file at @path are needed. */
+static int
+done_with_file(const tchar *path, wimlib_progress_func_t progfunc, void *progctx)
+{
+ union wimlib_progress_info info;
+
+ info.done_with_file.path_to_file = path;
+
+ return call_progress(progfunc, WIMLIB_PROGRESS_MSG_DONE_WITH_FILE,
+ &info, progctx);
+}
+
+/* Handle WIMLIB_WRITE_FLAG_SEND_DONE_WITH_FILE_MESSAGES mode. */
+static int
+done_with_stream(struct wim_lookup_table_entry *stream,
+ wimlib_progress_func_t progfunc, void *progctx,
+ struct wim_inode *inode)
+{
+ if (stream->resource_location == RESOURCE_IN_FILE_ON_DISK
+#ifdef __WIN32__
+ || stream->resource_location == RESOURCE_IN_WINNT_FILE_ON_DISK
+ || stream->resource_location == RESOURCE_WIN32_ENCRYPTED
+#endif
+ )
+ {
+ int ret;
+
+ wimlib_assert(inode->num_unread_streams > 0);
+ if (--inode->num_unread_streams > 0)
+ return 0;
+
+ #ifdef __WIN32__
+ /* XXX: This logic really should be somewhere else. */
+
+ /* We want the path to the file, but stream->file_on_disk might
+ * actually refer to a named data stream. Temporarily strip the
+ * named data stream from the path. */
+ wchar_t *p_colon = NULL;
+ wchar_t *p_question_mark = NULL;
+ const wchar_t *p_stream_name;
+
+ p_stream_name = path_stream_name(stream->file_on_disk);
+ if (unlikely(p_stream_name)) {
+ p_colon = (wchar_t *)(p_stream_name - 1);
+ wimlib_assert(*p_colon == L':');
+ *p_colon = L'\0';
+ }
+
+ /* We also should use a fake Win32 path instead of a NT path */
+ if (!wcsncmp(stream->file_on_disk, L"\\??\\", 4)) {
+ p_question_mark = &stream->file_on_disk[1];
+ *p_question_mark = L'\\';
+ }
+ #endif
+
+ ret = done_with_file(stream->file_on_disk, progfunc, progctx);
+
+ #ifdef __WIN32__
+ if (p_colon)
+ *p_colon = L':';
+ if (p_question_mark)
+ *p_question_mark = L'?';
+ #endif
+ return ret;
+ }
+ return 0;
+}
+
+/* Begin processing a stream for writing. */
+static int
+write_stream_begin_read(struct wim_lookup_table_entry *lte, void *_ctx)
+{
+ struct write_streams_ctx *ctx = _ctx;
+ int ret;
+
+ wimlib_assert(lte->size > 0);
+
+ ctx->cur_read_stream_offset = 0;
+ ctx->cur_read_stream_size = lte->size;
+
+ if (lte->unhashed)
+ ctx->stream_inode = lte->back_inode;
+ else
+ ctx->stream_inode = NULL;
+
+ /* As an optimization, we allow some streams to be "unhashed", meaning
+ * their SHA1 message digests are unknown. This is the case with
+ * streams that are added by scanning a directry tree with
+ * wimlib_add_image(), for example. Since WIM uses single-instance
+ * streams, we don't know whether such each such stream really need to
+ * written until it is actually checksummed, unless it has a unique
+ * size. In such cases we read and checksum the stream in this
+ * function, thereby advancing ahead of read_stream_list(), which will
+ * still provide the data again to write_stream_process_chunk(). This
+ * is okay because an unhashed stream cannot be in a WIM resource, which
+ * might be costly to decompress. */
+ ctx->stream_was_duplicate = false;
+ if (ctx->lookup_table != NULL && lte->unhashed && !lte->unique_size) {
+
+ struct wim_lookup_table_entry *lte_new;
+
+ ret = hash_unhashed_stream(lte, ctx->lookup_table, <e_new);
+ if (ret)
+ return ret;
+ if (lte_new != lte) {
+ /* Duplicate stream detected. */
+
+ if (lte_new->will_be_in_output_wim ||
+ stream_filtered(lte_new, ctx->filter_ctx))
+ {
+ /* The duplicate stream is already being
+ * included in the output WIM, or it would be
+ * filtered out if it had been. Skip writing
+ * this stream (and reading it again) entirely,
+ * passing its output reference count to the
+ * duplicate stream in the former case. */
+ DEBUG("Discarding duplicate stream of "
+ "length %"PRIu64, lte->size);
+ ret = do_write_streams_progress(&ctx->progress_data,
+ lte, lte->size,
+ 1, true);
+ list_del(<e->write_streams_list);
+ list_del(<e->lookup_table_list);
+ if (lte_new->will_be_in_output_wim)
+ lte_new->out_refcnt += lte->out_refcnt;
+ if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)
+ ctx->cur_write_res_size -= lte->size;
+ if (!ret && unlikely(ctx->write_resource_flags &
+ WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE))
+ {
+ ret = done_with_stream(lte,
+ ctx->progress_data.progfunc,
+ ctx->progress_data.progctx,
+ ctx->stream_inode);
+ }
+ free_lookup_table_entry(lte);
+ if (ret)
+ return ret;
+ return BEGIN_STREAM_STATUS_SKIP_STREAM;
+ } else {
+ /* The duplicate stream can validly be written,
+ * but was not marked as such. Discard the
+ * current stream entry and use the duplicate,
+ * but actually freeing the current entry must
+ * wait until read_stream_list() has finished
+ * reading its data. */
+ DEBUG("Stream duplicate, but not already "
+ "selected for writing.");
+ list_replace(<e->write_streams_list,
+ <e_new->write_streams_list);
+ list_replace(<e->lookup_table_list,
+ <e_new->lookup_table_list);
+ lte_new->out_refcnt = lte->out_refcnt;
+ lte_new->will_be_in_output_wim = 1;
+ ctx->stream_was_duplicate = true;
+ lte = lte_new;
+ }
+ }
+ }
+ list_move_tail(<e->write_streams_list, &ctx->pending_streams);
+ return 0;
+}
+
+/* Rewrite a stream that was just written compressed as uncompressed instead.
+ * This function is optional, but if a stream did not compress to less than its
+ * original size, it might as well be written uncompressed. */
+static int
+write_stream_uncompressed(struct wim_lookup_table_entry *lte,
+ struct filedes *out_fd)
+{
+ int ret;
+ u64 begin_offset = lte->out_reshdr.offset_in_wim;
+ u64 end_offset = out_fd->offset;
+
+ if (filedes_seek(out_fd, begin_offset) == -1)
+ return 0;
+
+ ret = extract_full_stream_to_fd(lte, out_fd);
+ if (ret) {
+ /* Error reading the uncompressed data. */
+ if (out_fd->offset == begin_offset &&
+ filedes_seek(out_fd, end_offset) != -1)
+ {
+ /* Nothing was actually written yet, and we successfully
+ * seeked to the end of the compressed resource, so
+ * don't issue a hard error; just keep the compressed
+ * resource instead. */
+ WARNING("Recovered compressed stream of "
+ "size %"PRIu64", continuing on.",
+ lte->size);
+ return 0;
+ }
+ return ret;
+ }
+
+ wimlib_assert(out_fd->offset - begin_offset == lte->size);
+
+ if (out_fd->offset < end_offset &&
+ 0 != ftruncate(out_fd->fd, out_fd->offset))
+ {
+ ERROR_WITH_ERRNO("Can't truncate output file to "
+ "offset %"PRIu64, out_fd->offset);
+ return WIMLIB_ERR_WRITE;
+ }
+
+ lte->out_reshdr.size_in_wim = lte->size;
+ lte->out_reshdr.flags &= ~(WIM_RESHDR_FLAG_COMPRESSED |
+ WIM_RESHDR_FLAG_PACKED_STREAMS);
+ return 0;
+}
+
+/* Write the next chunk of (typically compressed) data to the output WIM,
+ * handling the writing of the chunk table. */
+static int
+write_chunk(struct write_streams_ctx *ctx, const void *cchunk,
+ size_t csize, size_t usize)
+{
+ int ret;
+
+ struct wim_lookup_table_entry *lte;
+ u32 completed_stream_count;
+ u32 completed_size;
+
+ lte = list_entry(ctx->pending_streams.next,
+ struct wim_lookup_table_entry, write_streams_list);
+
+ if (ctx->cur_write_stream_offset == 0 &&
+ !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
+ {
+ /* Starting to write a new stream in non-packed mode. */
+
+ if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
+ int additional_reshdr_flags = 0;
+ if (ctx->compressor != NULL)
+ additional_reshdr_flags |= WIM_RESHDR_FLAG_COMPRESSED;
+
+ DEBUG("Writing pipable WIM stream header "
+ "(offset=%"PRIu64")", ctx->out_fd->offset);
+
+ ret = write_pwm_stream_header(lte, ctx->out_fd,
+ additional_reshdr_flags);
+ if (ret)
+ return ret;
+ }
+
+ ret = begin_write_resource(ctx, lte->size);
+ if (ret)
+ return ret;
+ }
+
+ if (ctx->compressor != NULL) {
+ /* Record the compresed chunk size. */
+ wimlib_assert(ctx->chunk_index < ctx->num_alloc_chunks);
+ ctx->chunk_csizes[ctx->chunk_index++] = csize;
+
+ /* If writing a pipable WIM, before the chunk data write a chunk
+ * header that provides the compressed chunk size. */
+ if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
+ struct pwm_chunk_hdr chunk_hdr = {
+ .compressed_size = cpu_to_le32(csize),
+ };
+ ret = full_write(ctx->out_fd, &chunk_hdr,
+ sizeof(chunk_hdr));
+ if (ret)
+ goto error;
+ }
+ }
+
+ /* Write the chunk data. */
+ ret = full_write(ctx->out_fd, cchunk, csize);
+ if (ret)
+ goto error;
+
+ ctx->cur_write_stream_offset += usize;
+
+ completed_size = usize;
+ completed_stream_count = 0;
+ if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
+ /* Wrote chunk in packed mode. It may have finished multiple
+ * streams. */
+ while (ctx->cur_write_stream_offset > lte->size) {
+ struct wim_lookup_table_entry *next;
+
+ ctx->cur_write_stream_offset -= lte->size;
+
+ wimlib_assert(!list_is_singular(&ctx->pending_streams) &&
+ !list_empty(&ctx->pending_streams));
+ next = list_entry(lte->write_streams_list.next,
+ struct wim_lookup_table_entry,
+ write_streams_list);
+ list_move_tail(<e->write_streams_list,
+ &ctx->pack_streams);
+ lte = next;
+ completed_stream_count++;
+ }
+ if (ctx->cur_write_stream_offset == lte->size) {
+ ctx->cur_write_stream_offset = 0;
+ list_move_tail(<e->write_streams_list,
+ &ctx->pack_streams);
+ completed_stream_count++;
+ }
+ } else {
+ /* Wrote chunk in non-packed mode. It may have finished a
+ * stream. */
+ if (ctx->cur_write_stream_offset == lte->size) {
+
+ completed_stream_count++;
+
+ list_del(<e->write_streams_list);
+
+ wimlib_assert(ctx->cur_write_stream_offset ==
+ ctx->cur_write_res_size);
+
+ ret = end_write_resource(ctx, <e->out_reshdr);
+ if (ret)
+ return ret;
+
+ lte->out_reshdr.flags = filter_resource_flags(lte->flags);
+ if (ctx->compressor != NULL)
+ lte->out_reshdr.flags |= WIM_RESHDR_FLAG_COMPRESSED;
+
+ if (ctx->compressor != NULL &&
+ lte->out_reshdr.size_in_wim >= lte->out_reshdr.uncompressed_size &&
+ !(ctx->write_resource_flags & (WRITE_RESOURCE_FLAG_PIPABLE |
+ WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE)) &&
+ !(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS))
+ {
+ /* Stream did not compress to less than its original
+ * size. If we're not writing a pipable WIM (which
+ * could mean the output file descriptor is
+ * non-seekable), and the stream isn't located in a
+ * resource pack (which would make reading it again
+ * costly), truncate the file to the start of the stream
+ * and write it uncompressed instead. */
+ DEBUG("Stream of size %"PRIu64" did not compress to "
+ "less than original size; writing uncompressed.",
+ lte->size);
+ ret = write_stream_uncompressed(lte, ctx->out_fd);
+ if (ret)
+ return ret;
+ }
+ wimlib_assert(lte->out_reshdr.uncompressed_size == lte->size);
+
+ ctx->cur_write_stream_offset = 0;
+ }
+ }
+
+ return do_write_streams_progress(&ctx->progress_data, lte,
+ completed_size, completed_stream_count,
+ false);
+
+error:
+ ERROR_WITH_ERRNO("Write error");
+ return ret;
+}
+
+static int
+submit_chunk_for_compression(struct write_streams_ctx *ctx,
+ const void *chunk, size_t size)
+{
+ /* While we are unable to submit the chunk for compression (due to too
+ * many chunks already outstanding), retrieve and write the next
+ * compressed chunk. */
+ while (!ctx->compressor->submit_chunk(ctx->compressor, chunk, size)) {
+ const void *cchunk;
+ u32 csize;
+ u32 usize;
+ bool bret;
+ int ret;
+
+ bret = ctx->compressor->get_chunk(ctx->compressor,
+ &cchunk, &csize, &usize);
+
+ wimlib_assert(bret);
+
+ ret = write_chunk(ctx, cchunk, csize, usize);
+ if (ret)
+ return ret;
+ }
+ return 0;
+}
+
+/* Process the next chunk of data to be written to a WIM resource. */
+static int
+write_stream_process_chunk(const void *chunk, size_t size, void *_ctx)
+{
+ struct write_streams_ctx *ctx = _ctx;
+ int ret;
+ const u8 *chunkptr, *chunkend;
+
+ wimlib_assert(size != 0);
+
+ if (ctx->compressor == NULL) {
+ /* Write chunk uncompressed. */
+ ret = write_chunk(ctx, chunk, size, size);
+ if (ret)
+ return ret;
+ ctx->cur_read_stream_offset += size;
+ return 0;
+ }
+
+ /* Submit the chunk for compression, but take into account that the
+ * @size the chunk was provided in may not correspond to the
+ * @out_chunk_size being used for compression. */
+ chunkptr = chunk;
+ chunkend = chunkptr + size;
+ do {
+ const u8 *resized_chunk;
+ size_t needed_chunk_size;
+
+ if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
+ needed_chunk_size = ctx->out_chunk_size;
+ } else {
+ u64 res_bytes_remaining;
+
+ res_bytes_remaining = ctx->cur_read_stream_size -
+ ctx->cur_read_stream_offset;
+ needed_chunk_size = min(ctx->out_chunk_size,
+ ctx->chunk_buf_filled +
+ res_bytes_remaining);
+ }
+
+ if (ctx->chunk_buf_filled == 0 &&
+ chunkend - chunkptr >= needed_chunk_size)
+ {
+ /* No intermediate buffering needed. */
+ resized_chunk = chunkptr;
+ chunkptr += needed_chunk_size;
+ ctx->cur_read_stream_offset += needed_chunk_size;
+ } else {
+ /* Intermediate buffering needed. */
+ size_t bytes_consumed;
+
+ bytes_consumed = min(chunkend - chunkptr,
+ needed_chunk_size - ctx->chunk_buf_filled);
+
+ memcpy(&ctx->chunk_buf[ctx->chunk_buf_filled],
+ chunkptr, bytes_consumed);
+
+ chunkptr += bytes_consumed;
+ ctx->cur_read_stream_offset += bytes_consumed;
+ ctx->chunk_buf_filled += bytes_consumed;
+ if (ctx->chunk_buf_filled == needed_chunk_size) {
+ resized_chunk = ctx->chunk_buf;
+ ctx->chunk_buf_filled = 0;
+ } else {
+ break;
+ }
+
+ }
+
+ ret = submit_chunk_for_compression(ctx, resized_chunk,
+ needed_chunk_size);
+ if (ret)
+ return ret;
+
+ } while (chunkptr != chunkend);
+ return 0;
+}
+
+/* Finish processing a stream for writing. It may not have been completely
+ * written yet, as the chunk_compressor implementation may still have chunks
+ * buffered or being compressed. */
+static int
+write_stream_end_read(struct wim_lookup_table_entry *lte, int status, void *_ctx)
+{
+ struct write_streams_ctx *ctx = _ctx;
+ if (status)
+ goto out;
+
+ wimlib_assert(ctx->cur_read_stream_offset == ctx->cur_read_stream_size);
+
+ if (unlikely(ctx->write_resource_flags &
+ WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE) &&
+ ctx->stream_inode != NULL)
+ {
+ status = done_with_stream(lte,
+ ctx->progress_data.progfunc,
+ ctx->progress_data.progctx,
+ ctx->stream_inode);
+ }
+
+ if (!ctx->stream_was_duplicate && lte->unhashed &&
+ ctx->lookup_table != NULL)
+ {
+ list_del(<e->unhashed_list);
+ lookup_table_insert(ctx->lookup_table, lte);
+ lte->unhashed = 0;
+ }
+out:
+ if (ctx->stream_was_duplicate)
+ free_lookup_table_entry(lte);
+ return status;
+}
+
+/* Compute statistics about a list of streams that will be written.
+ *
+ * Assumes the streams are sorted such that all streams located in each distinct
+ * WIM (specified by WIMStruct) are together. */
+static void
+compute_stream_list_stats(struct list_head *stream_list,
+ struct write_streams_ctx *ctx)
+{
+ struct wim_lookup_table_entry *lte;
+ u64 total_bytes = 0;
+ u64 num_streams = 0;
+ u64 total_parts = 0;
+ WIMStruct *prev_wim_part = NULL;
+
+ list_for_each_entry(lte, stream_list, write_streams_list) {
+ num_streams++;
+ total_bytes += lte->size;
+ if (lte->resource_location == RESOURCE_IN_WIM) {
+ if (prev_wim_part != lte->rspec->wim) {
+ prev_wim_part = lte->rspec->wim;
+ total_parts++;
+ }
+ }
+ }
+ ctx->progress_data.progress.write_streams.total_bytes = total_bytes;
+ ctx->progress_data.progress.write_streams.total_streams = num_streams;
+ ctx->progress_data.progress.write_streams.completed_bytes = 0;
+ ctx->progress_data.progress.write_streams.completed_streams = 0;
+ ctx->progress_data.progress.write_streams.compression_type = ctx->out_ctype;
+ ctx->progress_data.progress.write_streams.total_parts = total_parts;
+ ctx->progress_data.progress.write_streams.completed_parts = 0;
+ ctx->progress_data.next_progress = 0;
+}
+
+/* Find streams in @stream_list that can be copied to the output WIM in raw form
+ * rather than compressed. Delete these streams from @stream_list and move them
+ * to @raw_copy_streams. Return the total uncompressed size of the streams that
+ * need to be compressed. */
+static u64
+find_raw_copy_streams(struct list_head *stream_list,
+ int write_resource_flags,
+ int out_ctype,
+ u32 out_chunk_size,
+ struct list_head *raw_copy_streams)
+{
+ struct wim_lookup_table_entry *lte, *tmp;
+ u64 num_bytes_to_compress = 0;
+
+ INIT_LIST_HEAD(raw_copy_streams);
+
+ /* Initialize temporary raw_copy_ok flag. */
+ list_for_each_entry(lte, stream_list, write_streams_list)
+ if (lte->resource_location == RESOURCE_IN_WIM)
+ lte->rspec->raw_copy_ok = 0;
+
+ list_for_each_entry_safe(lte, tmp, stream_list, write_streams_list) {
+ if (lte->resource_location == RESOURCE_IN_WIM &&
+ lte->rspec->raw_copy_ok)
+ {
+ list_move_tail(<e->write_streams_list,
+ raw_copy_streams);
+ } else if (can_raw_copy(lte, write_resource_flags,
+ out_ctype, out_chunk_size))
+ {
+ lte->rspec->raw_copy_ok = 1;
+ list_move_tail(<e->write_streams_list,
+ raw_copy_streams);
+ } else {
+ num_bytes_to_compress += lte->size;
+ }
+ }
+
+ return num_bytes_to_compress;
+}
+
+/* Copy a raw compressed resource located in another WIM file to the WIM file
+ * being written. */
+static int
+write_raw_copy_resource(struct wim_resource_spec *in_rspec,
+ struct filedes *out_fd)
+{
+ u64 cur_read_offset;
+ u64 end_read_offset;
+ u8 buf[BUFFER_SIZE];
+ size_t bytes_to_read;
+ int ret;
+ struct filedes *in_fd;
+ struct wim_lookup_table_entry *lte;
+ u64 out_offset_in_wim;
+
+ DEBUG("Copying raw compressed data (size_in_wim=%"PRIu64", "
+ "uncompressed_size=%"PRIu64")",
+ in_rspec->size_in_wim, in_rspec->uncompressed_size);
+
+ /* Copy the raw data. */
+ cur_read_offset = in_rspec->offset_in_wim;
+ end_read_offset = cur_read_offset + in_rspec->size_in_wim;
+
+ out_offset_in_wim = out_fd->offset;
+
+ if (in_rspec->is_pipable) {
+ if (cur_read_offset < sizeof(struct pwm_stream_hdr))
+ return WIMLIB_ERR_INVALID_PIPABLE_WIM;
+ cur_read_offset -= sizeof(struct pwm_stream_hdr);
+ out_offset_in_wim += sizeof(struct pwm_stream_hdr);
+ }
+ in_fd = &in_rspec->wim->in_fd;
+ wimlib_assert(cur_read_offset != end_read_offset);
+ do {
+
+ bytes_to_read = min(sizeof(buf), end_read_offset - cur_read_offset);
+
+ ret = full_pread(in_fd, buf, bytes_to_read, cur_read_offset);
+ if (ret)
+ return ret;
+
+ ret = full_write(out_fd, buf, bytes_to_read);
+ if (ret)
+ return ret;
+
+ cur_read_offset += bytes_to_read;
+
+ } while (cur_read_offset != end_read_offset);
+
+ list_for_each_entry(lte, &in_rspec->stream_list, rspec_node) {
+ if (lte->will_be_in_output_wim) {
+ stream_set_out_reshdr_for_reuse(lte);
+ if (in_rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS)
+ lte->out_res_offset_in_wim = out_offset_in_wim;
+ else
+ lte->out_reshdr.offset_in_wim = out_offset_in_wim;
+
+ }
+ }
+ return 0;
+}
+
+/* Copy a list of raw compressed resources located in other WIM file(s) to the
+ * WIM file being written. */
+static int
+write_raw_copy_resources(struct list_head *raw_copy_streams,
+ struct filedes *out_fd,
+ struct write_streams_progress_data *progress_data)
+{
+ struct wim_lookup_table_entry *lte;
+ int ret;
+
+ list_for_each_entry(lte, raw_copy_streams, write_streams_list)
+ lte->rspec->raw_copy_ok = 1;
+
+ list_for_each_entry(lte, raw_copy_streams, write_streams_list) {
+ if (lte->rspec->raw_copy_ok) {
+ /* Write each packed resource only one time, no matter
+ * how many streams reference it. */
+ ret = write_raw_copy_resource(lte->rspec, out_fd);
+ if (ret)
+ return ret;
+ lte->rspec->raw_copy_ok = 0;
+ }
+ ret = do_write_streams_progress(progress_data, lte, lte->size,
+ 1, false);
+ if (ret)
+ return ret;
+ }
+ return 0;
+}
+
+/* Wait for and write all chunks pending in the compressor. */
+static int
+finish_remaining_chunks(struct write_streams_ctx *ctx)
+{
+ const void *cdata;
+ u32 csize;
+ u32 usize;
+ int ret;
+
+ if (ctx->compressor == NULL)
+ return 0;
+
+ if (ctx->chunk_buf_filled != 0) {
+ ret = submit_chunk_for_compression(ctx, ctx->chunk_buf,
+ ctx->chunk_buf_filled);
+ if (ret)
+ return ret;
+ }
+
+ while (ctx->compressor->get_chunk(ctx->compressor, &cdata, &csize, &usize)) {
+ ret = write_chunk(ctx, cdata, csize, usize);
+ if (ret)
+ return ret;
+ }
+ return 0;
+}
+
+static void
+remove_zero_length_streams(struct list_head *stream_list)
+{
+ struct wim_lookup_table_entry *lte, *tmp;
+
+ list_for_each_entry_safe(lte, tmp, stream_list, write_streams_list) {
+ wimlib_assert(lte->will_be_in_output_wim);
+ if (lte->size == 0) {
+ list_del(<e->write_streams_list);
+ lte->out_reshdr.offset_in_wim = 0;
+ lte->out_reshdr.size_in_wim = 0;
+ lte->out_reshdr.uncompressed_size = 0;
+ lte->out_reshdr.flags = filter_resource_flags(lte->flags);
+ }
+ }
+}
+
+/*
+ * Write a list of streams to the output WIM file.
+ *
+ * @stream_list
+ * The list of streams to write, specified by a list of `struct
+ * wim_lookup_table_entry's linked by the 'write_streams_list' member.
+ *
+ * @out_fd
+ * The file descriptor, opened for writing, to which to write the streams.
+ *
+ * @write_resource_flags
+ * Flags to modify how the streams are written:
+ *
+ * WRITE_RESOURCE_FLAG_RECOMPRESS:
+ * Force compression of all resources, even if they could otherwise
+ * be re-used by copying the raw data, due to being located in a WIM
+ * file with compatible compression parameters.
+ *
+ * WRITE_RESOURCE_FLAG_PIPABLE:
+ * Write the resources in the wimlib-specific pipable format, and
+ * furthermore do so in such a way that no seeking backwards in
+ * @out_fd will be performed (so it may be a pipe).
+ *
+ * WRITE_RESOURCE_FLAG_PACK_STREAMS:
+ * Pack all the streams into a single resource rather than writing
+ * them in separate resources. This flag is only valid if the WIM
+ * version number has been, or will be, set to
+ * WIM_VERSION_PACKED_STREAMS. This flag may not be combined with
+ * WRITE_RESOURCE_FLAG_PIPABLE.
+ *
+ * @out_ctype
+ * Compression format to use to write the output streams, specified as one
+ * of the WIMLIB_COMPRESSION_TYPE_* constants.
+ * WIMLIB_COMPRESSION_TYPE_NONE is allowed.
+ *
+ * @out_chunk_size
+ * Chunk size to use to write the streams. It must be a valid chunk size
+ * for the specified compression format @out_ctype, unless @out_ctype is
+ * WIMLIB_COMPRESSION_TYPE_NONE, in which case this parameter is ignored.
+ *
+ * @num_threads
+ * Number of threads to use to compress data. If 0, a default number of
+ * threads will be chosen. The number of threads still may be decreased
+ * from the specified value if insufficient memory is detected.
+ *
+ * @lookup_table
+ * If on-the-fly deduplication of unhashed streams is desired, this
+ * parameter must be pointer to the lookup table for the WIMStruct on whose
+ * behalf the streams are being written. Otherwise, this parameter can be
+ * NULL.
+ *
+ * @filter_ctx
+ * If on-the-fly deduplication of unhashed streams is desired, this
+ * parameter can be a pointer to a context for stream filtering used to
+ * detect whether the duplicate stream has been hard-filtered or not. If
+ * no streams are hard-filtered or no streams are unhashed, this parameter
+ * can be NULL.
+ *
+ * This function will write the streams in @stream_list to resources in
+ * consecutive positions in the output WIM file, or to a single packed resource
+ * if WRITE_RESOURCE_FLAG_PACK_STREAMS was specified in @write_resource_flags.
+ * In both cases, the @out_reshdr of the `struct wim_lookup_table_entry' for
+ * each stream written will be updated to specify its location, size, and flags
+ * in the output WIM. In the packed resource case,
+ * WIM_RESHDR_FLAG_PACKED_STREAMS will be set in the @flags field of each
+ * @out_reshdr, and furthermore @out_res_offset_in_wim and @out_res_size_in_wim
+ * of each @out_reshdr will be set to the offset and size, respectively, in the
+ * output WIM of the packed resource containing the corresponding stream.
+ *
+ * Each of the streams to write may be in any location supported by the
+ * resource-handling code (specifically, read_stream_list()), such as the
+ * contents of external file that has been logically added to the output WIM, or
+ * a stream in another WIM file that has been imported, or even a stream in the
+ * "same" WIM file of which a modified copy is being written. In the case that
+ * a stream is already in a WIM file and uses compatible compression parameters,
+ * by default this function will re-use the raw data instead of decompressing
+ * it, then recompressing it; however, with WRITE_RESOURCE_FLAG_RECOMPRESS
+ * specified in @write_resource_flags, this is not done.
+ *
+ * As a further requirement, this function requires that the
+ * @will_be_in_output_wim member be set to 1 on all streams in @stream_list as
+ * well as any other streams not in @stream_list that will be in the output WIM
+ * file, but set to 0 on any other streams in the output WIM's lookup table or
+ * sharing a packed resource with a stream in @stream_list. Still furthermore,
+ * if on-the-fly deduplication of streams is possible, then all streams in
+ * @stream_list must also be linked by @lookup_table_list along with any other
+ * streams that have @will_be_in_output_wim set.
+ *
+ * This function handles on-the-fly deduplication of streams for which SHA1
+ * message digests have not yet been calculated. Such streams may or may not
+ * need to be written. If @lookup_table is non-NULL, then each stream in
+ * @stream_list that has @unhashed set but not @unique_size set is checksummed
+ * immediately before it would otherwise be read for writing in order to
+ * determine if it is identical to another stream already being written or one
+ * that would be filtered out of the output WIM using stream_filtered() with the
+ * context @filter_ctx. Each such duplicate stream will be removed from
+ * @stream_list, its reference count transfered to the pre-existing duplicate
+ * stream, its memory freed, and will not be written. Alternatively, if a
+ * stream in @stream_list is a duplicate with any stream in @lookup_table that
+ * has not been marked for writing or would not be hard-filtered, it is freed
+ * and the pre-existing duplicate is written instead, taking ownership of the
+ * reference count and slot in the @lookup_table_list.
+ *
+ * Returns 0 if every stream was either written successfully or did not need to
+ * be written; otherwise returns a non-zero error code.
+ */
+static int
+write_stream_list(struct list_head *stream_list,
+ struct filedes *out_fd,
+ int write_resource_flags,
+ int out_ctype,
+ u32 out_chunk_size,
+ unsigned num_threads,
+ struct wim_lookup_table *lookup_table,
+ struct filter_context *filter_ctx,
+ wimlib_progress_func_t progfunc,
+ void *progctx)
+{
+ int ret;
+ struct write_streams_ctx ctx;
+ struct list_head raw_copy_streams;
+
+ wimlib_assert((write_resource_flags &
+ (WRITE_RESOURCE_FLAG_PACK_STREAMS |
+ WRITE_RESOURCE_FLAG_PIPABLE)) !=
+ (WRITE_RESOURCE_FLAG_PACK_STREAMS |
+ WRITE_RESOURCE_FLAG_PIPABLE));
+
+ remove_zero_length_streams(stream_list);
+
+ if (list_empty(stream_list)) {
+ DEBUG("No streams to write.");
+ return 0;
+ }
+
+ memset(&ctx, 0, sizeof(ctx));
+
+ /* Pre-sorting the streams is required for compute_stream_list_stats().
+ * Afterwards, read_stream_list() need not sort them again. */
+ ret = sort_stream_list_by_sequential_order(stream_list,
+ offsetof(struct wim_lookup_table_entry,
+ write_streams_list));
+ if (ret)
+ return ret;
+
+ ctx.out_fd = out_fd;
+ ctx.lookup_table = lookup_table;
+ ctx.out_ctype = out_ctype;
+ ctx.out_chunk_size = out_chunk_size;
+ ctx.write_resource_flags = write_resource_flags;
+ ctx.filter_ctx = filter_ctx;
+
+ if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) {
+ wimlib_assert(out_chunk_size != 0);
+ if (out_chunk_size <= STACK_MAX) {
+ ctx.chunk_buf = alloca(out_chunk_size);
+ } else {
+ ctx.chunk_buf = MALLOC(out_chunk_size);
+ if (ctx.chunk_buf == NULL) {
+ ret = WIMLIB_ERR_NOMEM;
+ goto out_destroy_context;
+ }
+ }
+ }
+ ctx.chunk_buf_filled = 0;
+
+ compute_stream_list_stats(stream_list, &ctx);
+
+ ctx.progress_data.progfunc = progfunc;
+ ctx.progress_data.progctx = progctx;
+
+ ctx.num_bytes_to_compress = find_raw_copy_streams(stream_list,
+ write_resource_flags,
+ out_ctype,
+ out_chunk_size,
+ &raw_copy_streams);
+
+ DEBUG("Writing stream list "
+ "(offset = %"PRIu64", write_resource_flags=0x%08x, "
+ "out_ctype=%d, out_chunk_size=%u, num_threads=%u, "
+ "total_bytes=%"PRIu64", num_bytes_to_compress=%"PRIu64")",
+ out_fd->offset, write_resource_flags,
+ out_ctype, out_chunk_size, num_threads,
+ ctx.progress_data.progress.write_streams.total_bytes,
+ ctx.num_bytes_to_compress);
+
+ if (ctx.num_bytes_to_compress == 0) {
+ DEBUG("No compression needed; skipping to raw copy!");
+ goto out_write_raw_copy_resources;
+ }
+
+ /* Unless uncompressed output was required, allocate a chunk_compressor
+ * to do compression. There are serial and parallel implementations of
+ * the chunk_compressor interface. We default to parallel using the
+ * specified number of threads, unless the upper bound on the number
+ * bytes needing to be compressed is less than a heuristic value. */
+ if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) {
+
+ #ifdef ENABLE_MULTITHREADED_COMPRESSION
+ if (ctx.num_bytes_to_compress > max(2000000, out_chunk_size)) {
+ ret = new_parallel_chunk_compressor(out_ctype,
+ out_chunk_size,
+ num_threads, 0,
+ &ctx.compressor);
+ if (ret) {
+ DEBUG("Couldn't create parallel chunk compressor "
+ "(status %d)", ret);
+ }
+ }
+ #endif
+
+ if (ctx.compressor == NULL) {
+ ret = new_serial_chunk_compressor(out_ctype, out_chunk_size,
+ &ctx.compressor);
+ if (ret)
+ goto out_destroy_context;
+ }
+ }
+
+ if (ctx.compressor)
+ ctx.progress_data.progress.write_streams.num_threads = ctx.compressor->num_threads;
+ else
+ ctx.progress_data.progress.write_streams.num_threads = 1;
+
+ DEBUG("Actually using %u threads",
+ ctx.progress_data.progress.write_streams.num_threads);
+
+ INIT_LIST_HEAD(&ctx.pending_streams);
+ INIT_LIST_HEAD(&ctx.pack_streams);
+
+ ret = call_progress(ctx.progress_data.progfunc,
+ WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
+ &ctx.progress_data.progress,
+ ctx.progress_data.progctx);
+ if (ret)
+ goto out_destroy_context;
+
+ if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
+ ret = begin_write_resource(&ctx, ctx.num_bytes_to_compress);
+ if (ret)
+ goto out_destroy_context;
+ }
+
+ /* Read the list of streams needing to be compressed, using the
+ * specified callbacks to execute processing of the data. */
+
+ struct read_stream_list_callbacks cbs = {
+ .begin_stream = write_stream_begin_read,
+ .begin_stream_ctx = &ctx,
+ .consume_chunk = write_stream_process_chunk,
+ .consume_chunk_ctx = &ctx,
+ .end_stream = write_stream_end_read,
+ .end_stream_ctx = &ctx,
+ };
+
+ ret = read_stream_list(stream_list,
+ offsetof(struct wim_lookup_table_entry, write_streams_list),
+ &cbs,
+ STREAM_LIST_ALREADY_SORTED |
+ VERIFY_STREAM_HASHES |
+ COMPUTE_MISSING_STREAM_HASHES);
+
+ if (ret)
+ goto out_destroy_context;
+
+ ret = finish_remaining_chunks(&ctx);
+ if (ret)
+ goto out_destroy_context;
+
+ if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
+ struct wim_reshdr reshdr;
+ struct wim_lookup_table_entry *lte;
+ u64 offset_in_res;
+
+ ret = end_write_resource(&ctx, &reshdr);
+ if (ret)
+ goto out_destroy_context;
+
+ DEBUG("Ending packed resource: %lu %lu %lu.",
+ reshdr.offset_in_wim,
+ reshdr.size_in_wim,
+ reshdr.uncompressed_size);
+
+ offset_in_res = 0;
+ list_for_each_entry(lte, &ctx.pack_streams, write_streams_list) {
+ lte->out_reshdr.size_in_wim = lte->size;
+ lte->out_reshdr.flags = filter_resource_flags(lte->flags);
+ lte->out_reshdr.flags |= WIM_RESHDR_FLAG_PACKED_STREAMS;
+ lte->out_reshdr.uncompressed_size = 0;
+ lte->out_reshdr.offset_in_wim = offset_in_res;
+ lte->out_res_offset_in_wim = reshdr.offset_in_wim;
+ lte->out_res_size_in_wim = reshdr.size_in_wim;
+ lte->out_res_uncompressed_size = reshdr.uncompressed_size;
+ offset_in_res += lte->size;
+ }
+ wimlib_assert(offset_in_res == reshdr.uncompressed_size);
+ }
+
+out_write_raw_copy_resources:
+ /* Copy any compressed resources for which the raw data can be reused
+ * without decompression. */
+ ret = write_raw_copy_resources(&raw_copy_streams, ctx.out_fd,
+ &ctx.progress_data);
+
+out_destroy_context:
+ if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && out_chunk_size > STACK_MAX)
+ FREE(ctx.chunk_buf);
+ FREE(ctx.chunk_csizes);
+ if (ctx.compressor)
+ ctx.compressor->destroy(ctx.compressor);
+ DEBUG("Done (ret=%d)", ret);
+ return ret;
+}
+
+static int
+is_stream_packed(struct wim_lookup_table_entry *lte, void *_ignore)
+{
+ return lte_is_partial(lte);
+}
+
+static bool
+wim_has_packed_streams(WIMStruct *wim)
+{
+ return for_lookup_table_entry(wim->lookup_table, is_stream_packed, NULL);
+}
+
+static int
+wim_write_stream_list(WIMStruct *wim,
+ struct list_head *stream_list,
+ int write_flags,
+ unsigned num_threads,
+ struct filter_context *filter_ctx)
+{
+ int out_ctype;
+ u32 out_chunk_size;
+ int write_resource_flags;
+
+ write_resource_flags = write_flags_to_resource_flags(write_flags);
+
+ /* wimlib v1.7.0: pack streams by default if the WIM version has been
+ * set to WIM_VERSION_PACKED_STREAMS and at least one stream in the
+ * WIM's lookup table is located in a packed resource (may be the same
+ * WIM, or a different one in the case of export). */
+ if (wim->hdr.wim_version == WIM_VERSION_PACKED_STREAMS &&
+ wim_has_packed_streams(wim))
+ {
+ write_resource_flags |= WRITE_RESOURCE_FLAG_PACK_STREAMS;
+ }
+
+ if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
+ out_chunk_size = wim->out_pack_chunk_size;
+ out_ctype = wim->out_pack_compression_type;
+ } else {
+ out_chunk_size = wim->out_chunk_size;
+ out_ctype = wim->out_compression_type;
+ }
+
+ return write_stream_list(stream_list,
+ &wim->out_fd,
+ write_resource_flags,
+ out_ctype,
+ out_chunk_size,
+ num_threads,
+ wim->lookup_table,
+ filter_ctx,
+ wim->progfunc,
+ wim->progctx);
+}
+
+static int
+write_wim_resource(struct wim_lookup_table_entry *lte,
+ struct filedes *out_fd,
+ int out_ctype,
+ u32 out_chunk_size,
+ int write_resource_flags)
+{
+ LIST_HEAD(stream_list);
+ list_add(<e->write_streams_list, &stream_list);
+ lte->will_be_in_output_wim = 1;
+ return write_stream_list(&stream_list,
+ out_fd,
+ write_resource_flags & ~WRITE_RESOURCE_FLAG_PACK_STREAMS,
+ out_ctype,
+ out_chunk_size,
+ 1,
+ NULL,
+ NULL,
+ NULL,
+ NULL);
+}
+
+int
+write_wim_resource_from_buffer(const void *buf, size_t buf_size,
+ int reshdr_flags, struct filedes *out_fd,
+ int out_ctype,
+ u32 out_chunk_size,
+ struct wim_reshdr *out_reshdr,
+ u8 *hash,
+ int write_resource_flags)
+{
+ int ret;
+ struct wim_lookup_table_entry *lte;
+
+ /* Set up a temporary lookup table entry to provide to
+ * write_wim_resource(). */
+
+ lte = new_lookup_table_entry();
+ if (lte == NULL)
+ return WIMLIB_ERR_NOMEM;
+
+ lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER;
+ lte->attached_buffer = (void*)buf;
+ lte->size = buf_size;
+ lte->flags = reshdr_flags;
+
+ if (write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
+ sha1_buffer(buf, buf_size, lte->hash);
+ lte->unhashed = 0;
+ } else {
+ lte->unhashed = 1;
+ }
+
+ ret = write_wim_resource(lte, out_fd, out_ctype, out_chunk_size,
+ write_resource_flags);
+ if (ret)
+ goto out_free_lte;
+
+ copy_reshdr(out_reshdr, <e->out_reshdr);
+
+ if (hash)
+ copy_hash(hash, lte->hash);
+ ret = 0;
+out_free_lte:
+ lte->resource_location = RESOURCE_NONEXISTENT;
+ free_lookup_table_entry(lte);
+ return ret;
+}
+
+struct stream_size_table {
+ struct hlist_head *array;
+ size_t num_entries;
+ size_t capacity;
+};
+
+static int
+init_stream_size_table(struct stream_size_table *tab, size_t capacity)
+{
+ tab->array = CALLOC(capacity, sizeof(tab->array[0]));
+ if (tab->array == NULL)
+ return WIMLIB_ERR_NOMEM;
+ tab->num_entries = 0;
+ tab->capacity = capacity;
+ return 0;
+}
+
+static void
+destroy_stream_size_table(struct stream_size_table *tab)
+{
+ FREE(tab->array);
+}
+
+static int
+stream_size_table_insert(struct wim_lookup_table_entry *lte, void *_tab)
+{
+ struct stream_size_table *tab = _tab;
+ size_t pos;
+ struct wim_lookup_table_entry *same_size_lte;
+ struct hlist_node *tmp;
+
+ pos = hash_u64(lte->size) % tab->capacity;
+ lte->unique_size = 1;
+ hlist_for_each_entry(same_size_lte, tmp, &tab->array[pos], hash_list_2) {
+ if (same_size_lte->size == lte->size) {
+ lte->unique_size = 0;
+ same_size_lte->unique_size = 0;
+ break;
+ }
+ }
+
+ hlist_add_head(<e->hash_list_2, &tab->array[pos]);
+ tab->num_entries++;
+ return 0;
+}
+
+struct find_streams_ctx {
+ WIMStruct *wim;
+ int write_flags;
+ struct list_head stream_list;
+ struct stream_size_table stream_size_tab;
+};
+
+static void
+reference_stream_for_write(struct wim_lookup_table_entry *lte,
+ struct list_head *stream_list, u32 nref)
+{
+ if (!lte->will_be_in_output_wim) {
+ lte->out_refcnt = 0;
+ list_add_tail(<e->write_streams_list, stream_list);
+ lte->will_be_in_output_wim = 1;
+ }
+ lte->out_refcnt += nref;
+}
+
+static int
+fully_reference_stream_for_write(struct wim_lookup_table_entry *lte,
+ void *_stream_list)
+{
+ struct list_head *stream_list = _stream_list;
+ lte->will_be_in_output_wim = 0;
+ reference_stream_for_write(lte, stream_list, lte->refcnt);
+ return 0;
+}
+
+static int
+inode_find_streams_to_reference(const struct wim_inode *inode,
+ const struct wim_lookup_table *table,
+ struct list_head *stream_list)
+{
+ struct wim_lookup_table_entry *lte;
+ unsigned i;
+
+ wimlib_assert(inode->i_nlink > 0);
+
+ for (i = 0; i <= inode->i_num_ads; i++) {
+ lte = inode_stream_lte(inode, i, table);
+ if (lte)
+ reference_stream_for_write(lte, stream_list,
+ inode->i_nlink);
+ else if (!is_zero_hash(inode_stream_hash(inode, i)))
+ return WIMLIB_ERR_RESOURCE_NOT_FOUND;
+ }
+ return 0;
+}
+
+static int
+do_stream_set_not_in_output_wim(struct wim_lookup_table_entry *lte, void *_ignore)
+{
+ lte->will_be_in_output_wim = 0;
+ return 0;
+}
+
+static int
+image_find_streams_to_reference(WIMStruct *wim)
+{
+ struct wim_image_metadata *imd;
+ struct wim_inode *inode;
+ struct wim_lookup_table_entry *lte;
+ struct list_head *stream_list;
+ int ret;
+
+ imd = wim_get_current_image_metadata(wim);
+
+ image_for_each_unhashed_stream(lte, imd)
+ lte->will_be_in_output_wim = 0;
+
+ stream_list = wim->private;
+ image_for_each_inode(inode, imd) {
+ ret = inode_find_streams_to_reference(inode,
+ wim->lookup_table,
+ stream_list);
+ if (ret)
+ return ret;
+ }
+ return 0;
+}
+
+static int
+prepare_unfiltered_list_of_streams_in_output_wim(WIMStruct *wim,
+ int image,
+ int streams_ok,
+ struct list_head *stream_list_ret)
+{
+ int ret;
+
+ INIT_LIST_HEAD(stream_list_ret);
+
+ if (streams_ok && (image == WIMLIB_ALL_IMAGES ||
+ (image == 1 && wim->hdr.image_count == 1)))
+ {
+ /* Fast case: Assume that all streams are being written and
+ * that the reference counts are correct. */
+ struct wim_lookup_table_entry *lte;
+ struct wim_image_metadata *imd;
+ unsigned i;
+
+ for_lookup_table_entry(wim->lookup_table,
+ fully_reference_stream_for_write,
+ stream_list_ret);
+
+ for (i = 0; i < wim->hdr.image_count; i++) {
+ imd = wim->image_metadata[i];
+ image_for_each_unhashed_stream(lte, imd)
+ fully_reference_stream_for_write(lte, stream_list_ret);
+ }
+ } else {
+ /* Slow case: Walk through the images being written and
+ * determine the streams referenced. */
+ for_lookup_table_entry(wim->lookup_table,
+ do_stream_set_not_in_output_wim, NULL);
+ wim->private = stream_list_ret;
+ ret = for_image(wim, image, image_find_streams_to_reference);
+ if (ret)
+ return ret;
+ }
+
+ return 0;
+}
+
+struct insert_other_if_hard_filtered_ctx {
+ struct stream_size_table *tab;
+ struct filter_context *filter_ctx;
+};
+
+static int
+insert_other_if_hard_filtered(struct wim_lookup_table_entry *lte, void *_ctx)
+{
+ struct insert_other_if_hard_filtered_ctx *ctx = _ctx;
+
+ if (!lte->will_be_in_output_wim &&
+ stream_hard_filtered(lte, ctx->filter_ctx))
+ stream_size_table_insert(lte, ctx->tab);
+ return 0;
+}
+
+static int
+determine_stream_size_uniquity(struct list_head *stream_list,
+ struct wim_lookup_table *lt,
+ struct filter_context *filter_ctx)
+{
+ int ret;
+ struct stream_size_table tab;
+ struct wim_lookup_table_entry *lte;
+
+ ret = init_stream_size_table(&tab, 9001);
+ if (ret)
+ return ret;
+
+ if (may_hard_filter_streams(filter_ctx)) {
+ struct insert_other_if_hard_filtered_ctx ctx = {
+ .tab = &tab,
+ .filter_ctx = filter_ctx,
+ };
+ for_lookup_table_entry(lt, insert_other_if_hard_filtered, &ctx);
+ }
+
+ list_for_each_entry(lte, stream_list, write_streams_list)
+ stream_size_table_insert(lte, &tab);
+
+ destroy_stream_size_table(&tab);
+ return 0;
+}
+
+static void
+filter_stream_list_for_write(struct list_head *stream_list,
+ struct filter_context *filter_ctx)
+{
+ struct wim_lookup_table_entry *lte, *tmp;
+
+ list_for_each_entry_safe(lte, tmp,
+ stream_list, write_streams_list)
+ {
+ int status = stream_filtered(lte, filter_ctx);
+
+ if (status == 0) {
+ /* Not filtered. */
+ continue;
+ } else {
+ if (status > 0) {
+ /* Soft filtered. */
+ } else {
+ /* Hard filtered. */
+ lte->will_be_in_output_wim = 0;
+ list_del(<e->lookup_table_list);
+ }
+ list_del(<e->write_streams_list);
+ }
+ }
+}
+
+/*
+ * prepare_stream_list_for_write() -
+ *
+ * Prepare the list of streams to write for writing a WIM containing the
+ * specified image(s) with the specified write flags.
+ *
+ * @wim
+ * The WIMStruct on whose behalf the write is occurring.
+ *
+ * @image
+ * Image(s) from the WIM to write; may be WIMLIB_ALL_IMAGES.
+ *
+ * @write_flags
+ * WIMLIB_WRITE_FLAG_* flags for the write operation:
+ *
+ * STREAMS_OK: For writes of all images, assume that all streams in the
+ * lookup table of @wim and the per-image lists of unhashed streams should
+ * be taken as-is, and image metadata should not be searched for
+ * references. This does not exclude filtering with OVERWRITE and
+ * SKIP_EXTERNAL_WIMS, below.
+ *
+ * OVERWRITE: Streams already present in @wim shall not be returned in
+ * @stream_list_ret.
+ *
+ * SKIP_EXTERNAL_WIMS: Streams already present in a WIM file, but not
+ * @wim, shall be returned in neither @stream_list_ret nor
+ * @lookup_table_list_ret.
+ *
+ * @stream_list_ret
+ * List of streams, linked by write_streams_list, that need to be written
+ * will be returned here.
+ *
+ * Note that this function assumes that unhashed streams will be written;
+ * it does not take into account that they may become duplicates when
+ * actually hashed.
+ *
+ * @lookup_table_list_ret
+ * List of streams, linked by lookup_table_list, that need to be included
+ * in the WIM's lookup table will be returned here. This will be a
+ * superset of the streams in @stream_list_ret.
+ *
+ * This list will be a proper superset of @stream_list_ret if and only if
+ * WIMLIB_WRITE_FLAG_OVERWRITE was specified in @write_flags and some of
+ * the streams that would otherwise need to be written were already located
+ * in the WIM file.
+ *
+ * All streams in this list will have @out_refcnt set to the number of
+ * references to the stream in the output WIM. If
+ * WIMLIB_WRITE_FLAG_STREAMS_OK was specified in @write_flags, @out_refcnt
+ * may be as low as 0.
+ *
+ * @filter_ctx_ret
+ * A context for queries of stream filter status with stream_filtered() is
+ * returned in this location.
+ *
+ * In addition, @will_be_in_output_wim will be set to 1 in all stream entries
+ * inserted into @lookup_table_list_ret and to 0 in all stream entries in the
+ * lookup table of @wim not inserted into @lookup_table_list_ret.
+ *
+ * Still furthermore, @unique_size will be set to 1 on all stream entries in
+ * @stream_list_ret that have unique size among all stream entries in
+ * @stream_list_ret and among all stream entries in the lookup table of @wim
+ * that are ineligible for being written due to filtering.
+ *
+ * Returns 0 on success; nonzero on read error, memory allocation error, or
+ * otherwise.
+ */
+static int
+prepare_stream_list_for_write(WIMStruct *wim, int image,
+ int write_flags,
+ struct list_head *stream_list_ret,
+ struct list_head *lookup_table_list_ret,
+ struct filter_context *filter_ctx_ret)
+{
+ int ret;
+ struct wim_lookup_table_entry *lte;
+
+ filter_ctx_ret->write_flags = write_flags;
+ filter_ctx_ret->wim = wim;
+
+ ret = prepare_unfiltered_list_of_streams_in_output_wim(
+ wim,
+ image,
+ write_flags & WIMLIB_WRITE_FLAG_STREAMS_OK,
+ stream_list_ret);
+ if (ret)
+ return ret;
+
+ INIT_LIST_HEAD(lookup_table_list_ret);
+ list_for_each_entry(lte, stream_list_ret, write_streams_list)
+ list_add_tail(<e->lookup_table_list, lookup_table_list_ret);
+
+ ret = determine_stream_size_uniquity(stream_list_ret, wim->lookup_table,
+ filter_ctx_ret);
+ if (ret)
+ return ret;
+
+ if (may_filter_streams(filter_ctx_ret))
+ filter_stream_list_for_write(stream_list_ret, filter_ctx_ret);
+
+ return 0;
+}
+
+static int
+write_wim_streams(WIMStruct *wim, int image, int write_flags,
+ unsigned num_threads,
+ struct list_head *stream_list_override,
+ struct list_head *lookup_table_list_ret)
+{
+ int ret;
+ struct list_head _stream_list;
+ struct list_head *stream_list;
+ struct wim_lookup_table_entry *lte;
+ struct filter_context _filter_ctx;
+ struct filter_context *filter_ctx;
+
+ if (stream_list_override == NULL) {
+ /* Normal case: prepare stream list from image(s) being written.
+ */
+ stream_list = &_stream_list;
+ filter_ctx = &_filter_ctx;
+ ret = prepare_stream_list_for_write(wim, image, write_flags,
+ stream_list,
+ lookup_table_list_ret,
+ filter_ctx);
+ if (ret)
+ return ret;
+ } else {
+ /* Currently only as a result of wimlib_split() being called:
+ * use stream list already explicitly provided. Use existing
+ * reference counts. */
+ stream_list = stream_list_override;
+ filter_ctx = NULL;
+ INIT_LIST_HEAD(lookup_table_list_ret);
+ list_for_each_entry(lte, stream_list, write_streams_list) {
+ lte->out_refcnt = lte->refcnt;
+ lte->will_be_in_output_wim = 1;
+ lte->unique_size = 0;
+ list_add_tail(<e->lookup_table_list, lookup_table_list_ret);
+ }
+ }
+
+ /* If needed, set auxiliary information so that we can detect when
+ * wimlib has finished using each external file. */
+ if (unlikely(write_flags & WIMLIB_WRITE_FLAG_SEND_DONE_WITH_FILE_MESSAGES)) {
+ list_for_each_entry(lte, stream_list, write_streams_list)
+ if (lte->unhashed)
+ lte->back_inode->num_unread_streams = 0;
+ list_for_each_entry(lte, stream_list, write_streams_list)
+ if (lte->unhashed)
+ lte->back_inode->num_unread_streams++;
+ }
+
+ return wim_write_stream_list(wim,
+ stream_list,
+ write_flags,
+ num_threads,
+ filter_ctx);
+}
+
+static int
+write_wim_metadata_resources(WIMStruct *wim, int image, int write_flags)
+{
+ int ret;
+ int start_image;
+ int end_image;
+ int write_resource_flags;
+
+ if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA) {
+ DEBUG("Not writing any metadata resources.");
+ return 0;
+ }
+
+ write_resource_flags = write_flags_to_resource_flags(write_flags);
+
+ write_resource_flags &= ~WRITE_RESOURCE_FLAG_PACK_STREAMS;
+
+ DEBUG("Writing metadata resources (offset=%"PRIu64")",
+ wim->out_fd.offset);
+
+ ret = call_progress(wim->progfunc,
+ WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN,
+ NULL, wim->progctx);
+ if (ret)
+ return ret;
+
+ if (image == WIMLIB_ALL_IMAGES) {
+ start_image = 1;
+ end_image = wim->hdr.image_count;
+ } else {
+ start_image = image;
+ end_image = image;
+ }
+
+ for (int i = start_image; i <= end_image; i++) {
+ struct wim_image_metadata *imd;
+
+ imd = wim->image_metadata[i - 1];
+ /* Build a new metadata resource only if image was modified from
+ * the original (or was newly added). Otherwise just copy the
+ * existing one. */
+ if (imd->modified) {
+ DEBUG("Image %u was modified; building and writing new "
+ "metadata resource", i);
+ ret = write_metadata_resource(wim, i,
+ write_resource_flags);
+ } else if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) {
+ DEBUG("Image %u was not modified; re-using existing "
+ "metadata resource.", i);
+ stream_set_out_reshdr_for_reuse(imd->metadata_lte);
+ ret = 0;
+ } else {
+ DEBUG("Image %u was not modified; copying existing "
+ "metadata resource.", i);
+ ret = write_wim_resource(imd->metadata_lte,
+ &wim->out_fd,
+ wim->out_compression_type,
+ wim->out_chunk_size,
+ write_resource_flags);
+ }
+ if (ret)
+ return ret;
+ }
+
+ return call_progress(wim->progfunc,
+ WIMLIB_PROGRESS_MSG_WRITE_METADATA_END,
+ NULL, wim->progctx);
+}
+
+static int
+open_wim_writable(WIMStruct *wim, const tchar *path, int open_flags)
+{
+ int raw_fd;
+ DEBUG("Opening \"%"TS"\" for writing.", path);
+
+ raw_fd = topen(path, open_flags | O_BINARY, 0644);
+ if (raw_fd < 0) {
+ ERROR_WITH_ERRNO("Failed to open \"%"TS"\" for writing", path);
+ return WIMLIB_ERR_OPEN;
+ }
+ filedes_init(&wim->out_fd, raw_fd);
+ return 0;
+}
+
+static int
+close_wim_writable(WIMStruct *wim, int write_flags)
+{
+ int ret = 0;
+
+ if (!(write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)) {
+ DEBUG("Closing WIM file.");
+ if (filedes_valid(&wim->out_fd))
+ if (filedes_close(&wim->out_fd))
+ ret = WIMLIB_ERR_WRITE;
+ }
+ filedes_invalidate(&wim->out_fd);
+ return ret;
+}
+
+static int
+cmp_streams_by_out_rspec(const void *p1, const void *p2)
+{
+ const struct wim_lookup_table_entry *lte1, *lte2;
+
+ lte1 = *(const struct wim_lookup_table_entry**)p1;
+ lte2 = *(const struct wim_lookup_table_entry**)p2;
+
+ if (lte1->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
+ if (lte2->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
+ if (lte1->out_res_offset_in_wim != lte2->out_res_offset_in_wim)
+ return cmp_u64(lte1->out_res_offset_in_wim,
+ lte2->out_res_offset_in_wim);
+ } else {
+ return 1;
+ }
+ } else {
+ if (lte2->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS)
+ return -1;
+ }
+ return cmp_u64(lte1->out_reshdr.offset_in_wim,
+ lte2->out_reshdr.offset_in_wim);
+}
+
+static int
+write_wim_lookup_table(WIMStruct *wim, int image, int write_flags,
+ struct wim_reshdr *out_reshdr,
+ struct list_head *lookup_table_list)
+{
+ int ret;
+
+ /* Set output resource metadata for streams already present in WIM. */
+ if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) {
+ struct wim_lookup_table_entry *lte;
+ list_for_each_entry(lte, lookup_table_list, lookup_table_list)
+ {
+ if (lte->resource_location == RESOURCE_IN_WIM &&
+ lte->rspec->wim == wim)
+ {
+ stream_set_out_reshdr_for_reuse(lte);
+ }
+ }
+ }
+
+ ret = sort_stream_list(lookup_table_list,
+ offsetof(struct wim_lookup_table_entry, lookup_table_list),
+ cmp_streams_by_out_rspec);
+ if (ret)
+ return ret;
+
+ /* Add entries for metadata resources. */
+ if (!(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)) {
+ int start_image;
+ int end_image;
+
+ if (image == WIMLIB_ALL_IMAGES) {
+ start_image = 1;
+ end_image = wim->hdr.image_count;
+ } else {
+ start_image = image;
+ end_image = image;
+ }
+
+ /* Push metadata resource lookup table entries onto the front of
+ * the list in reverse order, so that they're written in order.
+ */
+ for (int i = end_image; i >= start_image; i--) {
+ struct wim_lookup_table_entry *metadata_lte;
+
+ metadata_lte = wim->image_metadata[i - 1]->metadata_lte;
+ wimlib_assert(metadata_lte->out_reshdr.flags & WIM_RESHDR_FLAG_METADATA);
+ metadata_lte->out_refcnt = 1;
+ list_add(&metadata_lte->lookup_table_list, lookup_table_list);
+ }
+ }
+
+ return write_wim_lookup_table_from_stream_list(lookup_table_list,
+ &wim->out_fd,
+ wim->hdr.part_number,
+ out_reshdr,
+ write_flags_to_resource_flags(write_flags));
+}
+
+/*
+ * finish_write():
+ *
+ * Finish writing a WIM file: write the lookup table, xml data, and integrity
+ * table, then overwrite the WIM header. By default, closes the WIM file
+ * descriptor (@wim->out_fd) if successful.
+ *
+ * write_flags is a bitwise OR of the following:
+ *
+ * (public) WIMLIB_WRITE_FLAG_CHECK_INTEGRITY:
+ * Include an integrity table.
+ *
+ * (public) WIMLIB_WRITE_FLAG_FSYNC:
+ * fsync() the output file before closing it.
+ *
+ * (public) WIMLIB_WRITE_FLAG_PIPABLE:
+ * Writing a pipable WIM, possibly to a pipe; include pipable WIM
+ * stream headers before the lookup table and XML data, and also
+ * write the WIM header at the end instead of seeking to the
+ * beginning. Can't be combined with
+ * WIMLIB_WRITE_FLAG_CHECK_INTEGRITY.
+ *
+ * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE:
+ * Don't write the lookup table.
+ *
+ * (private) WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML:
+ * After writing the XML data but before writing the integrity
+ * table, write a temporary WIM header and flush the stream so that
+ * the WIM is less likely to become corrupted upon abrupt program
+ * termination.
+ * (private) WIMLIB_WRITE_FLAG_HEADER_AT_END:
+ * Instead of overwriting the WIM header at the beginning of the
+ * file, simply append it to the end of the file. (Used when
+ * writing to pipe.)
+ * (private) WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR:
+ * Do not close the file descriptor @wim->out_fd on either success
+ * on failure.
+ * (private) WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES:
+ * Use the existing <TOTALBYTES> stored in the in-memory XML
+ * information, rather than setting it to the offset of the XML
+ * data being written.
+ * (private) WIMLIB_WRITE_FLAG_OVERWRITE
+ * The existing WIM file is being updated in-place. The entries
+ * from its integrity table may be re-used.
+ */
+static int
+finish_write(WIMStruct *wim, int image, int write_flags,
+ struct list_head *lookup_table_list)
+{
+ int ret;
+ off_t hdr_offset;
+ int write_resource_flags;
+ off_t old_lookup_table_end = 0;
+ off_t new_lookup_table_end;
+ u64 xml_totalbytes;
+ struct integrity_table *old_integrity_table = NULL;
+
+ DEBUG("image=%d, write_flags=%08x", image, write_flags);
+
+ write_resource_flags = write_flags_to_resource_flags(write_flags);
+
+ /* In the WIM header, there is room for the resource entry for a
+ * metadata resource labeled as the "boot metadata". This entry should
+ * be zeroed out if there is no bootable image (boot_idx 0). Otherwise,
+ * it should be a copy of the resource entry for the image that is
+ * marked as bootable. This is not well documented... */
+ if (wim->hdr.boot_idx == 0) {
+ zero_reshdr(&wim->hdr.boot_metadata_reshdr);
+ } else {
+ copy_reshdr(&wim->hdr.boot_metadata_reshdr,
+ &wim->image_metadata[
+ wim->hdr.boot_idx - 1]->metadata_lte->out_reshdr);
+ }
+
+ /* If overwriting the WIM file containing an integrity table in-place,
+ * we'd like to re-use the information in the old integrity table
+ * instead of recalculating it. But we might overwrite the old
+ * integrity table when we expand the XML data. Read it into memory
+ * just in case. */
+ if ((write_flags & (WIMLIB_WRITE_FLAG_OVERWRITE |
+ WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)) ==
+ (WIMLIB_WRITE_FLAG_OVERWRITE |
+ WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)
+ && wim_has_integrity_table(wim))
+ {
+ old_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim +
+ wim->hdr.lookup_table_reshdr.size_in_wim;
+ (void)read_integrity_table(wim,
+ old_lookup_table_end - WIM_HEADER_DISK_SIZE,
+ &old_integrity_table);
+ /* If we couldn't read the old integrity table, we can still
+ * re-calculate the full integrity table ourselves. Hence the
+ * ignoring of the return value. */
+ }
+
+ /* Write lookup table. */
+ if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
+ ret = write_wim_lookup_table(wim, image, write_flags,
+ &wim->hdr.lookup_table_reshdr,
+ lookup_table_list);
+ if (ret) {
+ free_integrity_table(old_integrity_table);
+ return ret;
+ }
+ }
+
+ /* Write XML data. */
+ xml_totalbytes = wim->out_fd.offset;
+ if (write_flags & WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES)
+ xml_totalbytes = WIM_TOTALBYTES_USE_EXISTING;
+ ret = write_wim_xml_data(wim, image, xml_totalbytes,
+ &wim->hdr.xml_data_reshdr,
+ write_resource_flags);
+ if (ret) {
+ free_integrity_table(old_integrity_table);
+ return ret;
+ }