#include <unistd.h> /* for unlink() */
#include "wimlib/assert.h"
+#include "wimlib/bitops.h"
#include "wimlib/blob_table.h"
#include "wimlib/encoding.h"
#include "wimlib/endianness.h"
#include "wimlib/resource.h"
#include "wimlib/unaligned.h"
#include "wimlib/util.h"
+#include "wimlib/win32.h"
#include "wimlib/write.h"
/* A hash table mapping SHA-1 message digests to blob descriptors */
struct blob_table {
struct hlist_head *array;
size_t num_blobs;
- size_t capacity;
+ size_t mask; /* capacity - 1; capacity is a power of 2 */
};
+static size_t
+next_power_of_2(size_t n)
+{
+ if (n <= 1)
+ return 1;
+ return (size_t)1 << (1 + flsw(n - 1));
+}
+
struct blob_table *
new_blob_table(size_t capacity)
{
struct blob_table *table;
struct hlist_head *array;
+ capacity = next_power_of_2(capacity);
+
table = MALLOC(sizeof(struct blob_table));
if (table == NULL)
goto oom;
}
table->num_blobs = 0;
- table->capacity = capacity;
+ table->mask = capacity - 1;
table->array = array;
return table;
struct blob_descriptor *
new_blob_descriptor(void)
{
- BUILD_BUG_ON(BLOB_NONEXISTENT != 0);
+ STATIC_ASSERT(BLOB_NONEXISTENT == 0);
return CALLOC(1, sizeof(struct blob_descriptor));
}
break;
case BLOB_IN_FILE_ON_DISK:
-#ifdef __WIN32__
- case BLOB_IN_WINNT_FILE_ON_DISK:
- case BLOB_WIN32_ENCRYPTED:
-#endif
#ifdef WITH_FUSE
case BLOB_IN_STAGING_FILE:
- BUILD_BUG_ON((void*)&old->file_on_disk !=
- (void*)&old->staging_file_name);
+ STATIC_ASSERT((void*)&old->file_on_disk ==
+ (void*)&old->staging_file_name);
#endif
new->file_on_disk = TSTRDUP(old->file_on_disk);
if (new->file_on_disk == NULL)
goto out_free;
break;
+#ifdef __WIN32__
+ case BLOB_IN_WINDOWS_FILE:
+ new->windows_file = clone_windows_file(old->windows_file);
+ break;
+#endif
case BLOB_IN_ATTACHED_BUFFER:
new->attached_buffer = memdup(old->attached_buffer, old->size);
if (new->attached_buffer == NULL)
return NULL;
}
-static void
+/* Release a blob descriptor from its location, if any, and set its new location
+ * to BLOB_NONEXISTENT. */
+void
blob_release_location(struct blob_descriptor *blob)
{
switch (blob->blob_location) {
- case BLOB_IN_WIM:
+ case BLOB_IN_WIM: {
+ struct wim_resource_descriptor *rdesc = blob->rdesc;
+
list_del(&blob->rdesc_node);
- if (list_empty(&blob->rdesc->blob_list))
- FREE(blob->rdesc);
+ if (list_empty(&rdesc->blob_list)) {
+ wim_decrement_refcnt(rdesc->wim);
+ FREE(rdesc);
+ }
break;
+ }
case BLOB_IN_FILE_ON_DISK:
-#ifdef __WIN32__
- case BLOB_IN_WINNT_FILE_ON_DISK:
- case BLOB_WIN32_ENCRYPTED:
-#endif
#ifdef WITH_FUSE
case BLOB_IN_STAGING_FILE:
- BUILD_BUG_ON((void*)&blob->file_on_disk !=
- (void*)&blob->staging_file_name);
+ STATIC_ASSERT((void*)&blob->file_on_disk ==
+ (void*)&blob->staging_file_name);
#endif
case BLOB_IN_ATTACHED_BUFFER:
- BUILD_BUG_ON((void*)&blob->file_on_disk !=
- (void*)&blob->attached_buffer);
+ STATIC_ASSERT((void*)&blob->file_on_disk ==
+ (void*)&blob->attached_buffer);
FREE(blob->file_on_disk);
break;
+#ifdef __WIN32__
+ case BLOB_IN_WINDOWS_FILE:
+ free_windows_file(blob->windows_file);
+ break;
+#endif
#ifdef WITH_NTFS_3G
case BLOB_IN_NTFS_VOLUME:
- if (blob->ntfs_loc)
- free_ntfs_location(blob->ntfs_loc);
+ free_ntfs_location(blob->ntfs_loc);
break;
#endif
}
+ blob->blob_location = BLOB_NONEXISTENT;
}
void
static void
blob_table_insert_raw(struct blob_table *table, struct blob_descriptor *blob)
{
- size_t i = blob->hash_short % table->capacity;
+ size_t i = blob->hash_short & table->mask;
hlist_add_head(&blob->hash_list, &table->array[i]);
}
struct hlist_node *tmp;
size_t i;
- old_capacity = table->capacity;
+ old_capacity = table->mask + 1;
new_capacity = old_capacity * 2;
new_array = CALLOC(new_capacity, sizeof(struct hlist_head));
if (new_array == NULL)
return;
old_array = table->array;
table->array = new_array;
- table->capacity = new_capacity;
+ table->mask = new_capacity - 1;
- for (i = 0; i < old_capacity; i++) {
- hlist_for_each_entry_safe(blob, tmp, &old_array[i], hash_list) {
- hlist_del(&blob->hash_list);
+ for (i = 0; i < old_capacity; i++)
+ hlist_for_each_entry_safe(blob, tmp, &old_array[i], hash_list)
blob_table_insert_raw(table, blob);
- }
- }
FREE(old_array);
}
blob_table_insert(struct blob_table *table, struct blob_descriptor *blob)
{
blob_table_insert_raw(table, blob);
- if (++table->num_blobs > table->capacity)
+ if (table->num_blobs++ > table->mask)
enlarge_blob_table(table);
}
size_t i;
struct blob_descriptor *blob;
- i = load_size_t_unaligned(hash) % table->capacity;
+ i = load_size_t_unaligned(hash) & table->mask;
hlist_for_each_entry(blob, &table->array[i], hash_list)
if (hashes_equal(hash, blob->hash))
return blob;
struct hlist_node *tmp;
int ret;
- for (size_t i = 0; i < table->capacity; i++) {
+ for (size_t i = 0; i <= table->mask; i++) {
hlist_for_each_entry_safe(blob, tmp, &table->array[i],
hash_list)
{
v = (int)blob1->blob_location - (int)blob2->blob_location;
- /* Different locations? */
+ /* Different locations? Note: "unsafe compaction mode" requires that
+ * blobs in WIMs sort before all others. For the logic here to ensure
+ * this, BLOB_IN_WIM must have the lowest value among all defined
+ * blob_locations. Statically verify that the enum values haven't
+ * changed. */
+ STATIC_ASSERT(BLOB_NONEXISTENT == 0 && BLOB_IN_WIM == 1);
if (v)
return v;
wim1 = blob1->rdesc->wim;
wim2 = blob2->rdesc->wim;
- /* Different (possibly split) WIMs? */
+ /* Different WIM files? */
if (wim1 != wim2) {
- v = memcmp(wim1->hdr.guid, wim2->hdr.guid, WIM_GUID_LEN);
+
+ /* Resources from the WIM file currently being compacted
+ * (if any) must always sort first. */
+ v = (int)wim2->being_compacted - (int)wim1->being_compacted;
+ if (v)
+ return v;
+
+ /* Different split WIMs? */
+ v = cmp_guids(wim1->hdr.guid, wim2->hdr.guid);
+ if (v)
+ return v;
+
+ /* Different part numbers in the same split WIM? */
+ v = (int)wim1->hdr.part_number - (int)wim2->hdr.part_number;
if (v)
return v;
+
+ /* Probably two WIMStructs for the same on-disk file.
+ * Just sort by pointer. */
+ return wim1 < wim2 ? -1 : 1;
}
- /* Different part numbers in the same WIM? */
- v = (int)wim1->hdr.part_number - (int)wim2->hdr.part_number;
- if (v)
- return v;
+ /* Same WIM file */
+ /* Sort by increasing resource offset */
if (blob1->rdesc->offset_in_wim != blob2->rdesc->offset_in_wim)
return cmp_u64(blob1->rdesc->offset_in_wim,
blob2->rdesc->offset_in_wim);
+ /* The blobs are in the same solid resource. Sort by increasing
+ * offset in the resource. */
return cmp_u64(blob1->offset_in_res, blob2->offset_in_res);
case BLOB_IN_FILE_ON_DISK:
#ifdef WITH_FUSE
case BLOB_IN_STAGING_FILE:
-#endif
-#ifdef __WIN32__
- case BLOB_IN_WINNT_FILE_ON_DISK:
- case BLOB_WIN32_ENCRYPTED:
- /* Windows: compare by starting LCN (logical cluster number) */
- v = cmp_u64(blob1->sort_key, blob2->sort_key);
- if (v)
- return v;
#endif
/* Compare files by path: just a heuristic that will place files
* in the same directory next to each other. */
return tstrcmp(blob1->file_on_disk, blob2->file_on_disk);
+#ifdef __WIN32__
+ case BLOB_IN_WINDOWS_FILE:
+ return cmp_windows_files(blob1->windows_file, blob2->windows_file);
+#endif
#ifdef WITH_NTFS_3G
case BLOB_IN_NTFS_VOLUME:
return cmp_ntfs_locations(blob1->ntfs_loc, blob2->ntfs_loc);
rdesc = rdescs[i];
- wim_res_hdr_to_desc(&reshdr, wim, rdesc);
+ wim_reshdr_to_desc(&reshdr, wim, rdesc);
/* For solid resources, the uncompressed size, compression type,
* and chunk size are stored in the resource itself, not in the
/* Compression format numbers must be the same as in
* WIMGAPI to be compatible here. */
- BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_NONE != 0);
- BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 1);
- BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 2);
- BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3);
+ STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_NONE == 0);
+ STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_XPRESS == 1);
+ STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_LZX == 2);
+ STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_LZMS == 3);
rdesc->compression_type = le32_to_cpu(hdr.compression_format);
-
rdesc->chunk_size = le32_to_cpu(hdr.chunk_size);
-
- DEBUG("Solid resource %zu/%zu: %"PRIu64" => %"PRIu64" "
- "(%"TS"/%"PRIu32") @ +%"PRIu64"",
- i + 1, num_rdescs,
- rdesc->uncompressed_size,
- rdesc->size_in_wim,
- wimlib_get_compression_type_string(rdesc->compression_type),
- rdesc->chunk_size,
- rdesc->offset_in_wim);
}
return 0;
}
if (ret)
goto out_free_rdescs;
+ wim->refcnt += num_rdescs;
+
*rdescs_ret = rdescs;
*num_rdescs_ret = num_rdescs;
return 0;
free_solid_rdescs(struct wim_resource_descriptor **rdescs, size_t num_rdescs)
{
if (rdescs) {
- for (size_t i = 0; i < num_rdescs; i++)
- if (list_empty(&rdescs[i]->blob_list))
+ for (size_t i = 0; i < num_rdescs; i++) {
+ if (list_empty(&rdescs[i]->blob_list)) {
+ rdescs[i]->wim->refcnt--;
FREE(rdescs[i]);
+ }
+ }
FREE(rdescs);
}
}
struct blob_table *table = NULL;
struct blob_descriptor *cur_blob = NULL;
size_t num_duplicate_blobs = 0;
+ size_t num_empty_blobs = 0;
size_t num_wrong_part_blobs = 0;
u32 image_index = 0;
struct wim_resource_descriptor **cur_solid_rdescs = NULL;
size_t cur_num_solid_rdescs = 0;
- DEBUG("Reading blob table.");
-
/* Calculate the number of entries in the blob table. */
num_entries = wim->hdr.blob_table_reshdr.uncompressed_size /
sizeof(struct blob_descriptor_disk);
/* Allocate a hash table to map SHA-1 message digests into blob
* descriptors. This is the in-memory "blob table". */
- table = new_blob_table(num_entries * 2 + 1);
+ table = new_blob_table(num_entries);
if (!table)
goto oom;
/* Get the resource header */
get_wim_reshdr(&disk_entry->reshdr, &reshdr);
- DEBUG("reshdr: size_in_wim=%"PRIu64", "
- "uncompressed_size=%"PRIu64", "
- "offset_in_wim=%"PRIu64", "
- "flags=0x%02x",
- reshdr.size_in_wim, reshdr.uncompressed_size,
- reshdr.offset_in_wim, reshdr.flags);
-
/* Ignore SOLID flag if it isn't supposed to be used in this WIM
* version. */
if (wim->hdr.wim_version == WIM_VERSION_DEFAULT)
goto out;
}
- /* How to handle an uncompressed resource with its
- * uncompressed size different from its compressed size?
- *
- * Based on a simple test, WIMGAPI seems to handle this
- * as follows:
- *
- * if (size_in_wim > uncompressed_size) {
- * Ignore uncompressed_size; use size_in_wim
- * instead.
- * } else {
- * Honor uncompressed_size, but treat the part of
- * the file data above size_in_wim as all zeros.
- * }
- *
- * So we will do the same. */
- if (unlikely(!(reshdr.flags &
- WIM_RESHDR_FLAG_COMPRESSED) &&
- (reshdr.size_in_wim >
- reshdr.uncompressed_size)))
+ if (unlikely(!(reshdr.flags & WIM_RESHDR_FLAG_COMPRESSED) &&
+ (reshdr.size_in_wim != reshdr.uncompressed_size)))
{
- reshdr.uncompressed_size = reshdr.size_in_wim;
+ ERROR("Uncompressed resource has "
+ "size_in_wim != uncompressed_size");
+ ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
+ goto out;
}
/* Set up a resource descriptor for this blob. */
if (!rdesc)
goto oom;
- wim_res_hdr_to_desc(&reshdr, wim, rdesc);
-
- blob_set_is_located_in_nonsolid_wim_resource(cur_blob, rdesc);
+ wim_reshdr_to_desc_and_blob(&reshdr, wim, rdesc, cur_blob);
+ wim->refcnt++;
}
/* cur_blob is now a blob bound to a resource. */
/* Ignore entries with all zeroes in the hash field. */
- if (is_zero_hash(cur_blob->hash))
+ if (unlikely(is_zero_hash(cur_blob->hash)))
goto free_cur_blob_and_continue;
+ /* Verify that the blob has nonzero size. */
+ if (unlikely(cur_blob->size == 0)) {
+ num_empty_blobs++;
+ goto free_cur_blob_and_continue;
+ }
+
/* Verify that the part number matches that of the underlying
* WIM file. */
- if (part_number != wim->hdr.part_number) {
+ if (unlikely(part_number != wim->hdr.part_number)) {
num_wrong_part_blobs++;
goto free_cur_blob_and_continue;
}
* this overrides the actual locations of the metadata
* resources themselves in the WIM file as well as any
* information written in the XML data. */
- DEBUG("Found metadata resource for image %"PRIu32" at "
- "offset %"PRIu64".",
- image_index + 1,
- reshdr.offset_in_wim);
-
wim->image_metadata[image_index++]->metadata_blob = cur_blob;
} else {
/* Blob table entry for a non-metadata blob. */
if (num_duplicate_blobs > 0)
WARNING("Ignoring %zu duplicate blobs", num_duplicate_blobs);
+ if (num_empty_blobs > 0)
+ WARNING("Ignoring %zu empty blobs", num_empty_blobs);
+
if (num_wrong_part_blobs > 0) {
WARNING("Ignoring %zu blobs with wrong part number",
num_wrong_part_blobs);
}
- DEBUG("Done reading blob table.");
wim->blob_table = table;
ret = 0;
goto out_free_buf;
}
}
- DEBUG("Writing WIM blob table (size=%zu, offset=%"PRIu64")",
- table_size, out_fd->offset);
-
table_buf = MALLOC(table_size);
if (table_buf == NULL) {
ERROR("Failed to allocate %zu bytes for temporary blob table",
NULL,
write_resource_flags);
FREE(table_buf);
- DEBUG("ret=%d", ret);
return ret;
}