+ *res_start_offset_ret = res_start_offset;
+ *res_store_size_ret = res_end_offset - res_start_offset;
+
+ return 0;
+
+write_error:
+ ERROR_WITH_ERRNO("Write error");
+ return ret;
+}
+
+/* Finish writing a WIM resource by writing or updating the chunk table (if not
+ * writing the data uncompressed) and loading its metadata into @out_reshdr. */
+static int
+end_write_resource(struct write_blobs_ctx *ctx, struct wim_reshdr *out_reshdr)
+{
+ int ret;
+ u64 res_size_in_wim;
+ u64 res_uncompressed_size;
+ u64 res_offset_in_wim;
+
+ wimlib_assert(ctx->cur_write_blob_offset == ctx->cur_write_res_size ||
+ (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID));
+ 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;
+ return 0;
+}
+
+/* Call when no more data from the file at @path is 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);
+}
+
+static int
+do_done_with_blob(struct blob_descriptor *blob,
+ wimlib_progress_func_t progfunc, void *progctx)
+{
+ int ret;
+ struct wim_inode *inode;
+ tchar *cookie1;
+ tchar *cookie2;
+
+ if (!blob->may_send_done_with_file)
+ return 0;
+
+ inode = blob->file_inode;
+
+ wimlib_assert(inode != NULL);
+ wimlib_assert(inode->i_num_remaining_streams > 0);
+ if (--inode->i_num_remaining_streams > 0)
+ return 0;
+
+ cookie1 = progress_get_streamless_path(blob->file_on_disk);
+ cookie2 = progress_get_win32_path(blob->file_on_disk);
+
+ ret = done_with_file(blob->file_on_disk, progfunc, progctx);
+
+ progress_put_win32_path(cookie2);
+ progress_put_streamless_path(cookie1);
+
+ return ret;
+}
+
+/* Handle WIMLIB_WRITE_FLAG_SEND_DONE_WITH_FILE_MESSAGES mode. */
+static inline int
+done_with_blob(struct blob_descriptor *blob, struct write_blobs_ctx *ctx)
+{
+ if (likely(!(ctx->write_resource_flags &
+ WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE)))
+ return 0;
+ return do_done_with_blob(blob, ctx->progress_data.progfunc,
+ ctx->progress_data.progctx);
+}
+
+/* Begin processing a blob for writing. */
+static int
+write_blob_begin_read(struct blob_descriptor *blob, void *_ctx)
+{
+ struct write_blobs_ctx *ctx = _ctx;
+ int ret;
+
+ wimlib_assert(blob->size > 0);
+
+ ctx->cur_read_blob_offset = 0;
+ ctx->cur_read_blob_size = blob->size;
+
+ /* As an optimization, we allow some blobs to be "unhashed", meaning
+ * their SHA-1 message digests are unknown. This is the case with blobs
+ * that are added by scanning a directory tree with wimlib_add_image(),
+ * for example. Since WIM uses single-instance blobs, we don't know
+ * whether such each such blob really need to written until it is
+ * actually checksummed, unless it has a unique size. In such cases we
+ * read and checksum the blob in this function, thereby advancing ahead
+ * of read_blob_list(), which will still provide the data again to
+ * write_blob_process_chunk(). This is okay because an unhashed blob
+ * cannot be in a WIM resource, which might be costly to decompress. */
+ if (ctx->blob_table != NULL && blob->unhashed && !blob->unique_size) {
+
+ struct blob_descriptor *new_blob;
+
+ ret = hash_unhashed_blob(blob, ctx->blob_table, &new_blob);
+ if (ret)
+ return ret;
+ if (new_blob != blob) {
+ /* Duplicate blob detected. */
+
+ if (new_blob->will_be_in_output_wim ||
+ blob_filtered(new_blob, ctx->filter_ctx))
+ {
+ /* The duplicate blob is already being included
+ * in the output WIM, or it would be filtered
+ * out if it had been. Skip writing this blob
+ * (and reading it again) entirely, passing its
+ * output reference count to the duplicate blob
+ * in the former case. */
+ ret = do_write_blobs_progress(&ctx->progress_data,
+ blob->size, 1, true);
+ list_del(&blob->write_blobs_list);
+ list_del(&blob->blob_table_list);
+ if (new_blob->will_be_in_output_wim)
+ new_blob->out_refcnt += blob->out_refcnt;
+ if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID)
+ ctx->cur_write_res_size -= blob->size;
+ if (!ret)
+ ret = done_with_blob(blob, ctx);
+ free_blob_descriptor(blob);
+ if (ret)
+ return ret;
+ return BEGIN_BLOB_STATUS_SKIP_BLOB;
+ } else {
+ /* The duplicate blob can validly be written,
+ * but was not marked as such. Discard the
+ * current blob descriptor and use the
+ * duplicate, but actually freeing the current
+ * blob descriptor must wait until
+ * read_blob_list() has finished reading its
+ * data. */
+ list_replace(&blob->write_blobs_list,
+ &new_blob->write_blobs_list);
+ list_replace(&blob->blob_table_list,
+ &new_blob->blob_table_list);
+ blob->will_be_in_output_wim = 0;
+ new_blob->out_refcnt = blob->out_refcnt;
+ new_blob->will_be_in_output_wim = 1;
+ new_blob->may_send_done_with_file = 0;
+ blob = new_blob;
+ }
+ }
+ }
+ list_move_tail(&blob->write_blobs_list, &ctx->blobs_being_compressed);
+ return 0;
+}
+
+/* Rewrite a blob that was just written compressed (as a non-solid WIM resource)
+ * as uncompressed instead. */
+static int
+write_blob_uncompressed(struct blob_descriptor *blob, struct filedes *out_fd)
+{
+ int ret;
+ u64 begin_offset = blob->out_reshdr.offset_in_wim;
+ u64 end_offset = out_fd->offset;
+
+ if (filedes_seek(out_fd, begin_offset) == -1)
+ return 0;
+
+ ret = extract_blob_to_fd(blob, 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 resource of "
+ "size %"PRIu64", continuing on.", blob->size);
+ return 0;
+ }
+ return ret;
+ }
+
+ wimlib_assert(out_fd->offset - begin_offset == blob->size);
+
+ /* We could ftruncate() the file to 'out_fd->offset' here, but there
+ * isn't much point. Usually we will only be truncating by a few bytes
+ * and will just overwrite the data immediately. */
+
+ blob->out_reshdr.size_in_wim = blob->size;
+ blob->out_reshdr.flags &= ~(WIM_RESHDR_FLAG_COMPRESSED |
+ WIM_RESHDR_FLAG_SOLID);
+ return 0;
+}
+
+/* Returns true if the specified blob, which was written as a non-solid
+ * resource, should be truncated from the WIM file and re-written uncompressed.
+ * blob->out_reshdr must be filled in from the initial write of the blob. */
+static bool
+should_rewrite_blob_uncompressed(const struct write_blobs_ctx *ctx,
+ const struct blob_descriptor *blob)
+{
+ /* If the compressed data is smaller than the uncompressed data, prefer
+ * the compressed data. */
+ if (blob->out_reshdr.size_in_wim < blob->out_reshdr.uncompressed_size)
+ return false;
+
+ /* If we're not actually writing compressed data, then there's no need
+ * for re-writing. */
+ if (!ctx->compressor)
+ return false;
+
+ /* If writing a pipable WIM, everything we write to the output is final
+ * (it might actually be a pipe!). */
+ if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE)
+ return false;
+
+ /* If the blob that would need to be re-read is located in a solid
+ * resource in another WIM file, then re-reading it would be costly. So
+ * don't do it.
+ *
+ * Exception: if the compressed size happens to be *exactly* the same as
+ * the uncompressed size, then the blob *must* be written uncompressed
+ * in order to remain compatible with the Windows Overlay Filesystem
+ * Filter Driver (WOF).
+ *
+ * TODO: we are currently assuming that the optimization for
+ * single-chunk resources in maybe_rewrite_blob_uncompressed() prevents
+ * this case from being triggered too often. To fully prevent excessive
+ * decompressions in degenerate cases, we really should obtain the
+ * uncompressed data by decompressing the compressed data we wrote to
+ * the output file.
+ */
+ if (blob->blob_location == BLOB_IN_WIM &&
+ blob->size != blob->rdesc->uncompressed_size &&
+ blob->size != blob->out_reshdr.size_in_wim)
+ return false;
+
+ return true;
+}
+
+static int
+maybe_rewrite_blob_uncompressed(struct write_blobs_ctx *ctx,
+ struct blob_descriptor *blob)
+{
+ if (!should_rewrite_blob_uncompressed(ctx, blob))
+ return 0;
+
+ /* Regular (non-solid) WIM resources with exactly one chunk and
+ * compressed size equal to uncompressed size are exactly the same as
+ * the corresponding compressed data --- since there must be 0 entries
+ * in the chunk table and the only chunk must be stored uncompressed.
+ * In this case, there's no need to rewrite anything. */
+ if (ctx->chunk_index == 1 &&
+ blob->out_reshdr.size_in_wim == blob->out_reshdr.uncompressed_size)
+ {
+ blob->out_reshdr.flags &= ~WIM_RESHDR_FLAG_COMPRESSED;
+ return 0;
+ }
+
+ return write_blob_uncompressed(blob, ctx->out_fd);
+}
+
+/* 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_blobs_ctx *ctx, const void *cchunk,
+ size_t csize, size_t usize)
+{
+ int ret;
+ struct blob_descriptor *blob;
+ u32 completed_blob_count;
+ u32 completed_size;
+
+ blob = list_entry(ctx->blobs_being_compressed.next,
+ struct blob_descriptor, write_blobs_list);
+
+ if (ctx->cur_write_blob_offset == 0 &&
+ !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID))
+ {
+ /* Starting to write a new blob in non-solid mode. */
+
+ if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
+ ret = write_pwm_blob_header(blob, ctx->out_fd,
+ ctx->compressor != NULL);
+ if (ret)
+ return ret;
+ }
+
+ ret = begin_write_resource(ctx, blob->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 write_error;
+ }
+ }
+
+ /* Write the chunk data. */
+ ret = full_write(ctx->out_fd, cchunk, csize);
+ if (ret)
+ goto write_error;
+
+ ctx->cur_write_blob_offset += usize;
+
+ completed_size = usize;
+ completed_blob_count = 0;
+ if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
+ /* Wrote chunk in solid mode. It may have finished multiple
+ * blobs. */
+ struct blob_descriptor *next_blob;
+
+ while (blob && ctx->cur_write_blob_offset >= blob->size) {
+
+ ctx->cur_write_blob_offset -= blob->size;
+
+ if (ctx->cur_write_blob_offset)
+ next_blob = list_entry(blob->write_blobs_list.next,
+ struct blob_descriptor,
+ write_blobs_list);
+ else
+ next_blob = NULL;
+
+ ret = done_with_blob(blob, ctx);
+ if (ret)
+ return ret;
+ list_move_tail(&blob->write_blobs_list, &ctx->blobs_in_solid_resource);
+ completed_blob_count++;
+
+ blob = next_blob;
+ }
+ } else {
+ /* Wrote chunk in non-solid mode. It may have finished a
+ * blob. */
+ if (ctx->cur_write_blob_offset == blob->size) {
+
+ wimlib_assert(ctx->cur_write_blob_offset ==
+ ctx->cur_write_res_size);
+
+ ret = end_write_resource(ctx, &blob->out_reshdr);
+ if (ret)
+ return ret;
+
+ blob->out_reshdr.flags = reshdr_flags_for_blob(blob);
+ if (ctx->compressor != NULL)
+ blob->out_reshdr.flags |= WIM_RESHDR_FLAG_COMPRESSED;
+
+ ret = maybe_rewrite_blob_uncompressed(ctx, blob);
+ if (ret)
+ return ret;
+
+ wimlib_assert(blob->out_reshdr.uncompressed_size == blob->size);
+
+ ctx->cur_write_blob_offset = 0;
+
+ ret = done_with_blob(blob, ctx);
+ if (ret)
+ return ret;
+ list_del(&blob->write_blobs_list);
+ completed_blob_count++;
+ }
+ }
+
+ return do_write_blobs_progress(&ctx->progress_data, completed_size,
+ completed_blob_count, false);
+
+write_error:
+ ERROR_WITH_ERRNO("Write error");
+ return ret;
+}
+
+static int
+prepare_chunk_buffer(struct write_blobs_ctx *ctx)
+{
+ /* While we are unable to get a new chunk buffer due to too many chunks
+ * already outstanding, retrieve and write the next compressed chunk. */
+ while (!(ctx->cur_chunk_buf =
+ ctx->compressor->get_chunk_buffer(ctx->compressor)))
+ {
+ const void *cchunk;
+ u32 csize;
+ u32 usize;
+ bool bret;
+ int ret;
+
+ bret = ctx->compressor->get_compression_result(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_blob_process_chunk(const void *chunk, size_t size, void *_ctx)
+{
+ struct write_blobs_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_blob_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 {
+ size_t needed_chunk_size;
+ size_t bytes_consumed;
+
+ if (!ctx->cur_chunk_buf) {
+ ret = prepare_chunk_buffer(ctx);
+ if (ret)
+ return ret;
+ }
+
+ if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
+ needed_chunk_size = ctx->out_chunk_size;
+ } else {
+ needed_chunk_size = min(ctx->out_chunk_size,
+ ctx->cur_chunk_buf_filled +
+ (ctx->cur_read_blob_size -
+ ctx->cur_read_blob_offset));
+ }
+
+ bytes_consumed = min(chunkend - chunkptr,
+ needed_chunk_size - ctx->cur_chunk_buf_filled);
+
+ memcpy(&ctx->cur_chunk_buf[ctx->cur_chunk_buf_filled],
+ chunkptr, bytes_consumed);
+
+ chunkptr += bytes_consumed;
+ ctx->cur_read_blob_offset += bytes_consumed;
+ ctx->cur_chunk_buf_filled += bytes_consumed;
+
+ if (ctx->cur_chunk_buf_filled == needed_chunk_size) {
+ ctx->compressor->signal_chunk_filled(ctx->compressor,
+ ctx->cur_chunk_buf_filled);
+ ctx->cur_chunk_buf = NULL;
+ ctx->cur_chunk_buf_filled = 0;
+ }
+ } while (chunkptr != chunkend);
+ return 0;
+}
+
+/* Finish processing a blob 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_blob_end_read(struct blob_descriptor *blob, int status, void *_ctx)
+{
+ struct write_blobs_ctx *ctx = _ctx;
+
+ wimlib_assert(ctx->cur_read_blob_offset == ctx->cur_read_blob_size || status);
+
+ if (!blob->will_be_in_output_wim) {
+ /* The blob was a duplicate. Now that its data has finished
+ * being read, it is being discarded in favor of the duplicate
+ * entry. It therefore is no longer needed, and we can fire the
+ * DONE_WITH_FILE callback because the file will not be read
+ * again.
+ *
+ * Note: we can't yet fire DONE_WITH_FILE for non-duplicate
+ * blobs, since it needs to be possible to re-read the file if
+ * it does not compress to less than its original size. */
+ if (!status)
+ status = done_with_blob(blob, ctx);
+ free_blob_descriptor(blob);
+ } else if (!status && blob->unhashed && ctx->blob_table != NULL) {
+ /* The blob was not a duplicate and was previously unhashed.
+ * Since we passed COMPUTE_MISSING_BLOB_HASHES to
+ * read_blob_list(), blob->hash is now computed and valid. So
+ * turn this blob into a "hashed" blob. */
+ list_del(&blob->unhashed_list);
+ blob_table_insert(ctx->blob_table, blob);
+ blob->unhashed = 0;
+ }
+ return status;
+}
+
+/* Compute statistics about a list of blobs that will be written.
+ *
+ * Assumes the blobs are sorted such that all blobs located in each distinct WIM
+ * (specified by WIMStruct) are together. */
+static void
+compute_blob_list_stats(struct list_head *blob_list,
+ struct write_blobs_ctx *ctx)
+{
+ struct blob_descriptor *blob;
+ u64 total_bytes = 0;
+ u64 num_blobs = 0;
+ u64 total_parts = 0;
+ WIMStruct *prev_wim_part = NULL;
+
+ list_for_each_entry(blob, blob_list, write_blobs_list) {
+ num_blobs++;
+ total_bytes += blob->size;
+ if (blob->blob_location == BLOB_IN_WIM) {
+ if (prev_wim_part != blob->rdesc->wim) {
+ prev_wim_part = blob->rdesc->wim;
+ total_parts++;
+ }
+ }
+ }
+ ctx->progress_data.progress.write_streams.total_bytes = total_bytes;
+ ctx->progress_data.progress.write_streams.total_streams = num_blobs;
+ 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 blobs in @blob_list that can be copied to the output WIM in raw form
+ * rather than compressed. Delete these blobs from @blob_list and move them to
+ * @raw_copy_blobs. Return the total uncompressed size of the blobs that need
+ * to be compressed. */
+static u64
+find_raw_copy_blobs(struct list_head *blob_list,
+ int write_resource_flags,
+ int out_ctype,
+ u32 out_chunk_size,
+ struct list_head *raw_copy_blobs)
+{
+ struct blob_descriptor *blob, *tmp;
+ u64 num_bytes_to_compress = 0;
+
+ INIT_LIST_HEAD(raw_copy_blobs);
+
+ /* Initialize temporary raw_copy_ok flag. */
+ list_for_each_entry(blob, blob_list, write_blobs_list)
+ if (blob->blob_location == BLOB_IN_WIM)
+ blob->rdesc->raw_copy_ok = 0;
+
+ list_for_each_entry_safe(blob, tmp, blob_list, write_blobs_list) {
+ if (blob->blob_location == BLOB_IN_WIM &&
+ blob->rdesc->raw_copy_ok)
+ {
+ list_move_tail(&blob->write_blobs_list,
+ raw_copy_blobs);
+ } else if (can_raw_copy(blob, write_resource_flags,
+ out_ctype, out_chunk_size))
+ {
+ blob->rdesc->raw_copy_ok = 1;
+ list_move_tail(&blob->write_blobs_list,
+ raw_copy_blobs);
+ } else {
+ num_bytes_to_compress += blob->size;
+ }
+ }
+
+ return num_bytes_to_compress;