rspec->wim->chunk_size == out_chunk_size);
}
+ /* XXX: For compatibility, we can't allow multiple packed resources per
+ * WIM. */
+#if 0
if ((rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) &&
(write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
{
return (write_size > rspec->uncompressed_size / 2);
}
+#endif
+
return false;
}
return ret;
}
-#if 0
-static int
-seek_and_truncate(struct filedes *out_fd, off_t offset)
-{
- if (filedes_seek(out_fd, offset) == -1 ||
- ftruncate(out_fd->fd, offset))
- {
- ERROR_WITH_ERRNO("Failed to truncate output WIM file");
- return WIMLIB_ERR_WRITE;
- }
- return 0;
-}
-#endif
-
struct write_streams_progress_data {
wimlib_progress_func_t progress_func;
union wimlib_progress_info progress;
static void
do_write_streams_progress(struct write_streams_progress_data *progress_data,
- u64 size,
- bool discarded,
- struct wim_lookup_table_entry *cur_stream)
+ struct wim_lookup_table_entry *cur_stream,
+ u64 complete_size,
+ u32 complete_count,
+ bool discarded)
{
union wimlib_progress_info *progress = &progress_data->progress;
bool new_wim_part;
if (discarded) {
- progress->write_streams.total_bytes -= size;
+ progress->write_streams.total_bytes -= complete_size;
+ progress->write_streams.total_streams -= complete_count;
if (progress_data->next_progress != ~(uint64_t)0 &&
progress_data->next_progress > progress->write_streams.total_bytes)
{
progress_data->next_progress = progress->write_streams.total_bytes;
}
} else {
- progress->write_streams.completed_bytes += size;
+ progress->write_streams.completed_bytes += complete_size;
+ progress->write_streams.completed_streams += complete_count;
}
+
new_wim_part = false;
if (cur_stream->resource_location == RESOURCE_IN_WIM &&
cur_stream->rspec->wim != progress_data->prev_wim_part)
}
progress_data->prev_wim_part = cur_stream->rspec->wim;
}
- progress->write_streams.completed_streams++;
+
if (progress_data->progress_func
&& (progress->write_streams.completed_bytes >= progress_data->next_progress
|| new_wim_part))
/* List of streams that currently have chunks being compressed. */
struct list_head pending_streams;
+ /* List of streams in the resource pack. Streams are moved here after
+ * @pending_streams only when writing a packed resource. */
+ struct list_head pack_streams;
+
/* Set to true if the stream currently being read was a duplicate, and
* therefore the corresponding stream entry needs to be freed once the
* read finishes. (In this case we add the duplicate entry to
* pending_streams rather than the entry being read.) */
bool stream_was_duplicate;
- /* Current uncompressed offset in the resource being read. */
- u64 cur_read_res_offset;
+ /* Current uncompressed offset in the stream being read. */
+ u64 cur_read_stream_offset;
- /* Uncompressed size of the resource currently being read. */
- u64 cur_read_res_size;
+ /* Uncompressed size of the stream currently being read. */
+ u64 cur_read_stream_size;
- /* Current uncompressed offset in the resource being written. */
- u64 cur_write_res_offset;
+ /* Current uncompressed offset in the stream being written. */
+ u64 cur_write_stream_offset;
/* Uncompressed size of resource currently being written. */
u64 cur_write_res_size;
u64 chunks_start_offset;
};
-static u64
-get_chunk_entry_size(u64 res_size, int write_resource_flags)
-{
- if (res_size <= UINT32_MAX ||
- (write_resource_flags & WIM_RESHDR_FLAG_PACKED_STREAMS))
- return 4;
- else
- return 8;
-}
-
/* Reserve space for the chunk table and prepare to accumulate the chunk table
* in memory. */
static int
* are unknown. */
reserve_size = expected_num_chunk_entries *
get_chunk_entry_size(res_expected_size,
- ctx->write_resource_flags);
+ 0 != (ctx->write_resource_flags &
+ WIM_RESHDR_FLAG_PACKED_STREAMS));
if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)
reserve_size += sizeof(struct alt_chunk_table_header_disk);
memset(ctx->chunk_csizes, 0, reserve_size);
/* Output file descriptor is now positioned at the offset at which to
* write the first chunk of the resource. */
ctx->chunks_start_offset = ctx->out_fd->offset;
- ctx->cur_write_res_offset = 0;
+ ctx->cur_write_stream_offset = 0;
ctx->cur_write_res_size = res_expected_size;
return 0;
}
actual_num_chunk_entries--;
chunk_entry_size = get_chunk_entry_size(res_actual_size,
- ctx->write_resource_flags);
+ 0 != (ctx->write_resource_flags &
+ WIM_RESHDR_FLAG_PACKED_STREAMS));
typedef le64 __attribute__((may_alias)) aliased_le64_t;
typedef le32 __attribute__((may_alias)) aliased_le32_t;
u64 res_uncompressed_size;
u64 res_offset_in_wim;
- wimlib_assert(ctx->cur_write_res_size == ctx->cur_write_res_offset);
+ 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) {
wimlib_assert(lte->size > 0);
- ctx->cur_read_res_offset = 0;
- ctx->cur_read_res_size = lte->size;
+ ctx->cur_read_stream_offset = 0;
+ ctx->cur_read_stream_size = lte->size;
/* As an optimization, we allow some streams to be "unhashed", meaning
* their SHA1 message digests are unknown. This is the case with
DEBUG("Discarding duplicate stream of "
"length %"PRIu64, lte->size);
do_write_streams_progress(&ctx->progress_data,
- lte->size, true, lte);
+ 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)
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
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_res_offset == 0 &&
+ 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 (ret)
goto error;
- ctx->cur_write_res_offset += usize;
+ ctx->cur_write_stream_offset += usize;
- do_write_streams_progress(&ctx->progress_data,
- usize, false, lte);
+ 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;
- if (ctx->cur_write_res_offset == ctx->cur_write_res_size &&
- !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
- {
- struct wim_lookup_table_entry *lte;
+ ctx->cur_write_stream_offset -= lte->size;
- lte = list_entry(ctx->pending_streams.next,
- struct wim_lookup_table_entry, write_streams_list);
- wimlib_assert(ctx->cur_write_res_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) {
- /* Finished writing a stream in non-packed mode. */
+ completed_stream_count++;
- ret = end_write_resource(ctx, <e->out_reshdr);
- if (ret)
- return ret;
+ list_del(<e->write_streams_list);
- wimlib_assert(lte->out_reshdr.uncompressed_size == lte->size);
+ wimlib_assert(ctx->cur_write_stream_offset ==
+ ctx->cur_write_res_size);
- lte->out_reshdr.flags = filter_resource_flags(lte->flags);
- if (ctx->compressor != NULL)
- lte->out_reshdr.flags |= WIM_RESHDR_FLAG_COMPRESSED;
+ ret = end_write_resource(ctx, <e->out_reshdr);
+ if (ret)
+ return ret;
- list_del(<e->write_streams_list);
- ctx->cur_write_res_offset = 0;
+ 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) &&
+ !(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;
+ }
}
+ do_write_streams_progress(&ctx->progress_data, lte,
+ completed_size, completed_stream_count,
+ false);
+
return 0;
error:
ret = write_chunk(ctx, chunk, size, size);
if (ret)
return ret;
- ctx->cur_read_res_offset += size;
+ ctx->cur_read_stream_offset += size;
return 0;
}
} else {
u64 res_bytes_remaining;
- res_bytes_remaining = ctx->cur_read_res_size -
- ctx->cur_read_res_offset;
+ 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);
/* No intermediate buffering needed. */
resized_chunk = chunkptr;
chunkptr += needed_chunk_size;
- ctx->cur_read_res_offset += needed_chunk_size;
+ ctx->cur_read_stream_offset += needed_chunk_size;
} else {
/* Intermediate buffering needed. */
size_t bytes_consumed;
memcpy(&ctx->chunk_buf[ctx->chunk_buf_filled],
chunkptr, bytes_consumed);
- resized_chunk = ctx->chunk_buf;
-
chunkptr += bytes_consumed;
- ctx->cur_read_res_offset += 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;
{
struct write_streams_ctx *ctx = _ctx;
if (status == 0)
- wimlib_assert(ctx->cur_read_res_offset == ctx->cur_read_res_size);
+ wimlib_assert(ctx->cur_read_stream_offset == ctx->cur_read_stream_size);
if (ctx->stream_was_duplicate) {
free_lookup_table_entry(lte);
} else if (lte->unhashed && ctx->lookup_table != NULL) {
ret = write_raw_copy_resource(lte->rspec, out_fd);
if (ret)
return ret;
- do_write_streams_progress(progress_data, lte->size, false, lte);
+ do_write_streams_progress(progress_data, lte, lte->size,
+ 1, false);
}
return 0;
}
ctx.write_resource_flags = write_resource_flags;
ctx.filter_ctx = filter_ctx;
- if (out_chunk_size != 0) {
+ 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 {
* bytes needing to be compressed is less 2000000 (heuristic value). */
if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) {
- if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZMS &&
- ctx.lookup_table != NULL) {
- WARNING("LZMS compression not implemented; data will "
- "actually be written uncompressed.");
- }
-
+ #ifdef ENABLE_MULTITHREADED_COMPRESSION
if (ctx.num_bytes_to_compress >= 2000000) {
ret = new_parallel_chunk_compressor(out_ctype,
out_chunk_size,
"(status %d)", ret);
}
}
+ #endif
if (ctx.compressor == NULL) {
ret = new_serial_chunk_compressor(out_ctype, out_chunk_size,
ctx.progress_data.progress.write_streams.num_threads);
INIT_LIST_HEAD(&ctx.pending_streams);
+ INIT_LIST_HEAD(&ctx.pack_streams);
if (ctx.progress_data.progress_func) {
(*ctx.progress_data.progress_func)(WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
reshdr.uncompressed_size);
offset_in_res = 0;
- list_for_each_entry(lte, &ctx.pending_streams, write_streams_list) {
+ 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;
&ctx.progress_data);
out_destroy_context:
- if (out_chunk_size > STACK_MAX)
+ if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && out_chunk_size > STACK_MAX)
FREE(ctx.chunk_buf);
FREE(ctx.chunk_csizes);
if (ctx.compressor)
return ret;
}
+static int
+wim_write_stream_list(WIMStruct *wim,
+ struct list_head *stream_list,
+ int write_flags,
+ unsigned num_threads,
+ struct filter_context *filter_ctx,
+ wimlib_progress_func_t progress_func)
+{
+ int out_ctype;
+ u32 out_chunk_size;
+ int write_resource_flags;
+
+ write_resource_flags = write_flags_to_resource_flags(write_flags);
+
+ 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,
+ progress_func);
+}
+
static int
write_wim_resource(struct wim_lookup_table_entry *lte,
struct filedes *out_fd,
}
}
- return write_stream_list(stream_list,
- &wim->out_fd,
- write_flags_to_resource_flags(write_flags),
- wim->out_compression_type,
- wim->out_chunk_size,
- num_threads,
- wim->lookup_table,
- filter_ctx,
- progress_func);
+ return wim_write_stream_list(wim,
+ stream_list,
+ write_flags,
+ num_threads,
+ filter_ctx,
+ progress_func);
}
static int
if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS)
DEBUG("\tPACK_STREAMS");
- if (write_flags & WIMLIB_WRITE_FLAG_NO_PACK_STREAMS)
- DEBUG("\tNO_PACK_STREAMS");
-
if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)
DEBUG("\tFILE_DESCRIPTOR");
WIMLIB_WRITE_FLAG_NOT_PIPABLE))
return WIMLIB_ERR_INVALID_PARAM;
- if ((write_flags & (WIMLIB_WRITE_FLAG_PACK_STREAMS |
- WIMLIB_WRITE_FLAG_NO_PACK_STREAMS))
- == (WIMLIB_WRITE_FLAG_PACK_STREAMS |
- WIMLIB_WRITE_FLAG_NO_PACK_STREAMS))
- return WIMLIB_ERR_INVALID_PARAM;
-
/* Save previous header, then start initializing the new one. */
memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header));
write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY;
}
- if (!(write_flags & (WIMLIB_WRITE_FLAG_PACK_STREAMS |
- WIMLIB_WRITE_FLAG_NO_PACK_STREAMS)))
- if (wim->hdr.wim_version == WIM_VERSION_PACKED_STREAMS) {
- DEBUG("WIM version 3584; default to PACK_STREAMS.");
- write_flags |= WIMLIB_WRITE_FLAG_PACK_STREAMS;
- }
-
if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
WIMLIB_WRITE_FLAG_PACK_STREAMS))
== (WIMLIB_WRITE_FLAG_PIPABLE |
wim->hdr.magic = WIM_MAGIC;
/* Set appropriate version number. */
- if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS)
+ if ((write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS) ||
+ wim->out_compression_type == WIMLIB_COMPRESSION_TYPE_LZMS)
wim->hdr.wim_version = WIM_VERSION_PACKED_STREAMS;
else
wim->hdr.wim_version = WIM_VERSION_DEFAULT;
if (wim_has_integrity_table(wim))
write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY;
- /* Set default packed flag. */
- if (!(write_flags & (WIMLIB_WRITE_FLAG_PACK_STREAMS |
- WIMLIB_WRITE_FLAG_NO_PACK_STREAMS)))
- {
- #if 0
- if (wim->hdr.wim_version == WIM_VERSION_PACKED_STREAMS)
- write_flags |= WIMLIB_WRITE_FLAG_PACK_STREAMS;
- #endif
- /* wimlib allows multiple packs in a single WIM, but they don't
- * seem to be compatible with WIMGAPI. Write new streams
- * unpacked. */
- } else if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS) {
+ /* Set WIM version if adding packed streams. */
+ if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS)
wim->hdr.wim_version = WIM_VERSION_PACKED_STREAMS;
- }
/* Set additional flags for overwrite. */
write_flags |= WIMLIB_WRITE_FLAG_OVERWRITE |
goto out_restore_physical_hdr;
}
- ret = write_stream_list(&stream_list,
- &wim->out_fd,
- write_flags_to_resource_flags(write_flags),
- wim->compression_type,
- wim->chunk_size,
- num_threads,
- wim->lookup_table,
- &filter_ctx,
- progress_func);
+ ret = wim_write_stream_list(wim,
+ &stream_list,
+ write_flags,
+ num_threads,
+ &filter_ctx,
+ progress_func);
if (ret)
goto out_truncate;
git://wimlib.net / wimlib / blobdiff