/* Are the compression types the same? If so, do a raw copy (copy
* without decompressing and recompressing the data). */
raw = (wim_resource_compression_type(lte) == out_ctype
- && out_ctype != WIM_COMPRESSION_TYPE_NONE);
+ && out_ctype != WIM_COMPRESSION_TYPE_NONE
+ && !(flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS));
if (raw) {
flags |= WIMLIB_RESOURCE_FLAG_RAW;
}
#endif
-static void show_stream_write_progress(u64 *cur_size, u64 *next_size,
- u64 total_size, u64 one_percent,
- unsigned *cur_percent,
- const struct lookup_table_entry *cur_lte)
+void show_stream_op_progress(u64 *cur_size, u64 *next_size,
+ u64 total_size, u64 one_percent,
+ unsigned *cur_percent,
+ const struct lookup_table_entry *cur_lte,
+ const char *op)
{
if (*cur_size >= *next_size) {
printf("\r%"PRIu64" MiB of %"PRIu64" MiB "
- "(uncompressed) written (%u%% done)",
+ "(uncompressed) %s (%u%% done)",
*cur_size >> 20,
- total_size >> 20, *cur_percent);
+ total_size >> 20, op, *cur_percent);
fflush(stdout);
*next_size += one_percent;
(*cur_percent)++;
*cur_size += wim_resource_size(cur_lte);
}
-static void finish_stream_write_progress(u64 total_size)
+void finish_stream_op_progress(u64 total_size, const char *op)
{
printf("\r%"PRIu64" MiB of %"PRIu64" MiB "
- "(uncompressed) written (100%% done)\n",
- total_size >> 20, total_size >> 20);
+ "(uncompressed) %s (100%% done)\n",
+ total_size >> 20, total_size >> 20, op);
fflush(stdout);
}
u64 cur_size = 0;
u64 next_size = 0;
unsigned cur_percent = 0;
+ int write_resource_flags = 0;
+
+ if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
+ write_resource_flags |= WIMLIB_RESOURCE_FLAG_RECOMPRESS;
list_for_each_entry(lte, stream_list, staging_list) {
if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) {
- show_stream_write_progress(&cur_size, &next_size,
- total_size, one_percent,
- &cur_percent, lte);
+ show_stream_op_progress(&cur_size, &next_size,
+ total_size, one_percent,
+ &cur_percent, lte, "written");
}
ret = write_wim_resource(lte, out_fp, out_ctype,
- <e->output_resource_entry, 0);
+ <e->output_resource_entry,
+ write_resource_flags);
if (ret != 0)
return ret;
}
if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS)
- finish_stream_write_progress(total_size);
+ finish_stream_op_progress(total_size, "written");
return 0;
}
{
int ret;
-
struct message msgs[queue_size];
ZERO_ARRAY(msgs);
struct lookup_table_entry,
staging_list);
next_resource = next_resource->next;
- if ((next_lte->resource_location == RESOURCE_IN_WIM
- && wimlib_get_compression_type(next_lte->wim) == out_ctype)
+ if ((!(write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
+ && next_lte->resource_location == RESOURCE_IN_WIM
+ && wimlib_get_compression_type(next_lte->wim) == out_ctype)
|| wim_resource_size(next_lte) == 0)
{
list_add_tail(&next_lte->staging_list,
if (msg->begin_chunk == 0) {
DEBUG2("Begin chunk tab");
if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) {
- show_stream_write_progress(&cur_size,
- &next_size,
- total_size,
- one_percent,
- &cur_percent,
- cur_lte);
+ show_stream_op_progress(&cur_size,
+ &next_size,
+ total_size,
+ one_percent,
+ &cur_percent,
+ cur_lte,
+ "written");
}
// This is the first set of chunks. Leave space
staging_list)
{
if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) {
- show_stream_write_progress(&cur_size,
- &next_size,
- total_size,
- one_percent,
- &cur_percent,
- lte);
+ show_stream_op_progress(&cur_size,
+ &next_size,
+ total_size,
+ one_percent,
+ &cur_percent,
+ lte,
+ "written");
}
ret = write_wim_resource(lte,
if (ret == 0) {
list_for_each_entry(lte, &my_resources, staging_list) {
if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) {
- show_stream_write_progress(&cur_size,
- &next_size,
- total_size,
- one_percent,
- &cur_percent,
- lte);
+ show_stream_op_progress(&cur_size,
+ &next_size,
+ total_size,
+ one_percent,
+ &cur_percent,
+ lte,
+ "written");
}
ret = write_wim_resource(lte, out_fp,
out_ctype,
break;
}
if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS)
- finish_stream_write_progress(total_size);
+ finish_stream_op_progress(total_size, "written");
} else {
size_t num_available_msgs = 0;
struct list_head *cur;
}
if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) {
- printf("Writing %s compressed data using %u threads...\n",
+ printf("Writing %s data using %u threads...\n",
get_data_type(out_ctype), num_threads);
}
num_streams++;
total_size += wim_resource_size(lte);
if (!compression_needed
- && out_ctype != WIM_COMPRESSION_TYPE_NONE
- && (lte->resource_location != RESOURCE_IN_WIM
- || wimlib_get_compression_type(lte->wim) != out_ctype)
+ &&
+ (out_ctype != WIM_COMPRESSION_TYPE_NONE
+ && (lte->resource_location != RESOURCE_IN_WIM
+ || wimlib_get_compression_type(lte->wim) != out_ctype
+ || (write_flags & WIMLIB_WRITE_FLAG_REBUILD)))
&& wim_resource_size(lte) != 0)
compression_needed = true;
}
}
/*
- * Write the lookup table, xml data, and integrity table, then overwrite the WIM
- * header.
+ * Finish writing a WIM file: write the lookup table, xml data, and integrity
+ * table (optional), then overwrite the WIM header.
+ *
+ * write_flags is a bitwise OR of the following:
+ *
+ * (public) WIMLIB_WRITE_FLAG_CHECK_INTEGRITY:
+ * Include an integrity table.
+ *
+ * (public) WIMLIB_WRITE_FLAG_SHOW_PROGRESS:
+ * Show progress information when (if) writing the integrity table.
+ *
+ * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE:
+ * Don't write the lookup table.
+ *
+ * (private) WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE:
+ * When (if) writing the integrity table, re-use entries from the
+ * existing integrity table, if possible.
+ *
+ * (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_FSYNC:
+ * fsync() the output file before closing it.
+ *
*/
int finish_write(WIMStruct *w, int image, int write_flags)
{
goto out;
if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
+ if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) {
+ struct wim_header checkpoint_hdr;
+ memcpy(&checkpoint_hdr, &hdr, sizeof(struct wim_header));
+ memset(&checkpoint_hdr.integrity, 0, sizeof(struct resource_entry));
+ if (fseeko(out, 0, SEEK_SET) != 0) {
+ ret = WIMLIB_ERR_WRITE;
+ goto out;
+ }
+ ret = write_header(&checkpoint_hdr, out);
+ if (ret != 0)
+ goto out;
+
+ if (fflush(out) != 0) {
+ ERROR_WITH_ERRNO("Can't write data to WIM");
+ ret = WIMLIB_ERR_WRITE;
+ goto out;
+ }
+
+ if (fseeko(out, 0, SEEK_END) != 0) {
+ ret = WIMLIB_ERR_WRITE;
+ goto out;
+ }
+ }
+
off_t old_lookup_table_end;
off_t new_lookup_table_end;
bool show_progress;
if (fseeko(out, 0, SEEK_SET) != 0) {
ret = WIMLIB_ERR_WRITE;
- ERROR_WITH_ERRNO("Failed to seek to beginning of WIM "
- "to overwrite header");
goto out;
}
return 0;
}
+/*
+ * Overwrite a WIM, possibly appending streams to it.
+ *
+ * A WIM looks like (or is supposed to look like) the following:
+ *
+ * Header (212 bytes)
+ * Streams and metadata resources (variable size)
+ * Lookup table (variable size)
+ * XML data (variable size)
+ * Integrity table (optional) (variable size)
+ *
+ * If we are not adding any streams or metadata resources, the lookup table is
+ * unchanged--- so we only need to overwrite the XML data, integrity table, and
+ * header. This operation is potentially unsafe if the program is abruptly
+ * terminated while the XML data or integrity table are being overwritten, but
+ * before the new header has been written. To partially alleviate this problem,
+ * a special flag (WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) is passed to
+ * finish_write() to cause a temporary WIM header to be written after the XML
+ * data has been written. This may prevent the WIM from becoming corrupted if
+ * the program is terminated while the integrity table is being calculated (but
+ * no guarantees, due to write re-ordering...).
+ *
+ * If we are adding new streams or images (metadata resources), the lookup table
+ * needs to be changed, and those streams need to be written. In this case, we
+ * try to perform a safe update of the WIM file by writing the streams *after*
+ * the end of the previous WIM, then writing the new lookup table, XML data, and
+ * (optionally) integrity table following the new streams. This will produce a
+ * layout like the following:
+ *
+ * Header (212 bytes)
+ * (OLD) Streams and metadata resources (variable size)
+ * (OLD) Lookup table (variable size)
+ * (OLD) XML data (variable size)
+ * (OLD) Integrity table (optional) (variable size)
+ * (NEW) Streams and metadata resources (variable size)
+ * (NEW) Lookup table (variable size)
+ * (NEW) XML data (variable size)
+ * (NEW) Integrity table (optional) (variable size)
+ *
+ * At all points, the WIM is valid as nothing points to the new data yet. Then,
+ * the header is overwritten to point to the new lookup table, XML data, and
+ * integrity table, to produce the following layout:
+ *
+ * Header (212 bytes)
+ * Streams and metadata resources (variable size)
+ * Nothing (variable size)
+ * More Streams and metadata resources (variable size)
+ * Lookup table (variable size)
+ * XML data (variable size)
+ * Integrity table (optional) (variable size)
+ *
+ * This method allows an image to be appended to a large WIM very quickly, and
+ * is is crash-safe except in the case of write re-ordering, but the
+ * disadvantage is that a small hole is left in the WIM where the old lookup
+ * table, xml data, and integrity table were. (These usually only take up a
+ * small amount of space compared to the streams, however.
+ */
static int overwrite_wim_inplace(WIMStruct *w, int write_flags,
unsigned num_threads,
int modified_image_idx)
if (ret != 0)
return ret;
- DEBUG("old_wim_end = %"PRIu64, old_wim_end);
+ if (modified_image_idx == w->hdr.image_count && !w->deletion_occurred) {
+ /* If no images have been modified and no images have been
+ * deleted, a new lookup table does not need to be written. */
+ wimlib_assert(list_empty(&stream_list));
+ old_wim_end = w->hdr.lookup_table_res_entry.offset +
+ w->hdr.lookup_table_res_entry.size;
+ write_flags |= WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE |
+ WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML;
+ }
INIT_LIST_HEAD(&stream_list);
for_lookup_table_entry(w->lookup_table, find_new_streams,
&stream_list);
- {
- u64 num_new_streams = 0;
- struct list_head *cur;
- list_for_each(cur, &stream_list)
- num_new_streams++;
- DEBUG("%"PRIu64" new streams to write", num_new_streams);
- }
-
- {
- bool trunc = false;
- bool need_readable = false;
- if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)
- need_readable = true;
- ret = open_wim_writable(w, w->filename, trunc, need_readable);
- if (ret != 0)
- return ret;
- }
+ ret = open_wim_writable(w, w->filename, false,
+ (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0);
+ if (ret != 0)
+ return ret;
if (fseeko(w->out_fp, old_wim_end, SEEK_SET) != 0) {
ERROR_WITH_ERRNO("Can't seek to end of WIM");
return WIMLIB_ERR_SPLIT_UNSUPPORTED;
}
- if (!w->deletion_occurred && !(write_flags & WIMLIB_WRITE_FLAG_REBUILD)) {
+ if ((!w->deletion_occurred || (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE))
+ && !(write_flags & WIMLIB_WRITE_FLAG_REBUILD))
+ {
int i, modified_image_idx;
for (i = 0; i < w->hdr.image_count && !w->image_metadata[i].modified; i++)
;
git://wimlib.net / wimlib / blobdiff