*/
u8 default_hash[SHA1_HASH_SIZE];
- /* The format of the following data is not yet completely known and they
- * do not correspond to Microsoft's documentation.
+ /* Unknown field (maybe accidental padding) */
+ le32 unknown_0x54;
+
+ /*
+ * The following 8-byte union contains either information about the
+ * reparse point (for files with FILE_ATTRIBUTE_REPARSE_POINT set), or
+ * the "hard link group ID" (for other files).
*
- * If this directory entry is for a reparse point (has
- * FILE_ATTRIBUTE_REPARSE_POINT set in the 'attributes' field), then the
- * version of the following fields containing the reparse tag is valid.
- * Furthermore, the field notated as not_rpfixed, as far as I can tell,
- * is supposed to be set to 1 if reparse point fixups (a.k.a. fixing the
- * targets of absolute symbolic links) were *not* done, and otherwise 0.
+ * The reparse point information contains ReparseTag and ReparseReserved
+ * from the header of the reparse point buffer. It also contains a flag
+ * that indicates whether a reparse point fixup (for the target of an
+ * absolute symbolic link or junction) was done or not.
*
- * If this directory entry is not for a reparse point, then the version
- * of the following fields containing the hard_link_group_id is valid.
- * All MS says about this field is that "If this file is part of a hard
- * link set, all the directory entries in the set will share the same
- * value in this field.". However, more specifically I have observed
- * the following:
- * - If the file is part of a hard link set of size 1, then the
- * hard_link_group_id should be set to either 0, which is treated
- * specially as indicating "not hardlinked", or any unique value.
- * - The specific nonzero values used to identity hard link sets do
- * not matter, as long as they are unique.
- * - However, due to bugs in Microsoft's software, it is actually NOT
- * guaranteed that directory entries that share the same hard link
- * group ID are actually hard linked to each either. See
- * inode_fixup.c for the code that handles this.
+ * The "hard link group ID" is like an inode number; all dentries for
+ * the same inode share the same value. See inode_fixup.c for more
+ * information.
+ *
+ * Note that this union creates the limitation that reparse point files
+ * cannot have multiple names (hard links).
*/
union {
struct {
- le32 rp_unknown_1;
le32 reparse_tag;
- le16 rp_unknown_2;
- le16 not_rpfixed;
+ le16 rp_reserved;
+ le16 rp_flags;
} _packed_attribute reparse;
struct {
- le32 rp_unknown_1;
le64 hard_link_group_id;
} _packed_attribute nonreparse;
};
* encoded "long" name, excluding the null terminator. If zero, then
* this file has no long name. The root dentry should not have a long
* name, but all other dentries in the image should have long names. */
- le16 file_name_nbytes;
+ le16 name_nbytes;
/* Beginning of optional, variable-length fields */
- /* If file_name_nbytes != 0, the next field will be the UTF-16LE encoded
- * long file name. This will be null-terminated, so the size of this
- * field will really be file_name_nbytes + 2. */
- /*utf16lechar file_name[];*/
+ /* If name_nbytes != 0, the next field will be the UTF-16LE encoded long
+ * name. This will be null-terminated, so the size of this field will
+ * really be name_nbytes + 2. */
+ /*utf16lechar name[];*/
/* If short_name_nbytes != 0, the next field will be the UTF-16LE
* encoded short name. This will be null-terminated, so the size of
} _packed_attribute;
static void
-do_dentry_set_name(struct wim_dentry *dentry, utf16lechar *file_name,
- size_t file_name_nbytes)
+do_dentry_set_name(struct wim_dentry *dentry, utf16lechar *name,
+ size_t name_nbytes)
{
- FREE(dentry->file_name);
- dentry->file_name = file_name;
- dentry->file_name_nbytes = file_name_nbytes;
+ FREE(dentry->d_name);
+ dentry->d_name = name;
+ dentry->d_name_nbytes = name_nbytes;
if (dentry_has_short_name(dentry)) {
- FREE(dentry->short_name);
- dentry->short_name = NULL;
- dentry->short_name_nbytes = 0;
+ FREE(dentry->d_short_name);
+ dentry->d_short_name = NULL;
+ dentry->d_short_name_nbytes = 0;
}
}
* tagged metadata items as well as any extra stream entries that may need to
* follow the dentry. */
static size_t
-dentry_min_len_with_names(u16 file_name_nbytes, u16 short_name_nbytes)
+dentry_min_len_with_names(u16 name_nbytes, u16 short_name_nbytes)
{
size_t length = sizeof(struct wim_dentry_on_disk);
- if (file_name_nbytes)
- length += (u32)file_name_nbytes + 2;
+ if (name_nbytes)
+ length += (u32)name_nbytes + 2;
if (short_name_nbytes)
length += (u32)short_name_nbytes + 2;
return length;
const struct wim_inode *inode = dentry->d_inode;
size_t len;
- len = dentry_min_len_with_names(dentry->file_name_nbytes,
- dentry->short_name_nbytes);
+ len = dentry_min_len_with_names(dentry->d_name_nbytes,
+ dentry->d_short_name_nbytes);
len = ALIGN(len, 8);
len += ALIGN(inode->i_extra_size, 8);
/*
* Calculate the full path to @dentry within the WIM image, if not already done.
*
- * The full name will be saved in the cached value 'dentry->_full_path'.
+ * The full name will be saved in the cached value 'dentry->d_full_path'.
*
* Whenever possible, use dentry_full_path() instead of calling this and
- * accessing _full_path directly.
+ * accessing d_full_path directly.
*
* Returns 0 or an error code resulting from a failed string conversion.
*/
size_t dummy;
const struct wim_dentry *d;
- if (dentry->_full_path)
+ if (dentry->d_full_path)
return 0;
ulen = 0;
d = dentry;
do {
- ulen += d->file_name_nbytes / sizeof(utf16lechar);
+ ulen += d->d_name_nbytes / sizeof(utf16lechar);
ulen++;
d = d->d_parent; /* assumes d == d->d_parent for root */
} while (!dentry_is_root(d));
d = dentry;
do {
- p -= d->file_name_nbytes / sizeof(utf16lechar);
- memcpy(p, d->file_name, d->file_name_nbytes);
+ p -= d->d_name_nbytes / sizeof(utf16lechar);
+ memcpy(p, d->d_name, d->d_name_nbytes);
*--p = cpu_to_le16(WIM_PATH_SEPARATOR);
d = d->d_parent; /* assumes d == d->d_parent for root */
} while (!dentry_is_root(d));
wimlib_assert(p == ubuf);
return utf16le_to_tstr(ubuf, ulen * sizeof(utf16lechar),
- &dentry->_full_path, &dummy);
+ &dentry->d_full_path, &dummy);
}
/*
dentry_full_path(struct wim_dentry *dentry)
{
calculate_dentry_full_path(dentry);
- return dentry->_full_path;
+ return dentry->d_full_path;
}
static int
struct wim_dentry *child;
/* Set offset of directory's child dentries */
- dentry->subdir_offset = *subdir_offset_p;
+ dentry->d_subdir_offset = *subdir_offset_p;
/* Account for child dentries */
for_dentry_child(child, dentry)
/* Account for end-of-directory entry */
*subdir_offset_p += 8;
} else {
- /* Not a directory; set subdir_offset to 0 */
- dentry->subdir_offset = 0;
+ /* Not a directory; set the subdir offset to 0 */
+ dentry->d_subdir_offset = 0;
}
return 0;
}
dentry_compare_names_case_insensitive(const struct wim_dentry *d1,
const struct wim_dentry *d2)
{
- return cmp_utf16le_strings(d1->file_name,
- d1->file_name_nbytes / 2,
- d2->file_name,
- d2->file_name_nbytes / 2,
+ return cmp_utf16le_strings(d1->d_name,
+ d1->d_name_nbytes / 2,
+ d2->d_name,
+ d2->d_name_nbytes / 2,
true);
}
dentry_compare_names_case_sensitive(const struct wim_dentry *d1,
const struct wim_dentry *d2)
{
- return cmp_utf16le_strings(d1->file_name,
- d1->file_name_nbytes / 2,
- d2->file_name,
- d2->file_name_nbytes / 2,
+ return cmp_utf16le_strings(d1->d_name,
+ d1->d_name_nbytes / 2,
+ d2->d_name,
+ d2->d_name_nbytes / 2,
false);
}
#endif
;
-/* Case-sensitive dentry lookup. Only @file_name and @file_name_nbytes of
- * @dummy must be valid. */
+/* Case-sensitive dentry lookup. Only @d_name and @d_name_nbytes of @dummy must
+ * be valid. */
static struct wim_dentry *
dir_lookup(const struct wim_inode *dir, const struct wim_dentry *dummy)
{
return avl_tree_entry(node, struct wim_dentry, d_index_node);
}
-/* Case-insensitive dentry lookup. Only @file_name and @file_name_nbytes of
- * @dummy must be valid. */
+/* Case-insensitive dentry lookup. Only @d_name and @d_name_nbytes of @dummy
+ * must be valid. */
static struct wim_dentry *
dir_lookup_ci(const struct wim_inode *dir, const struct wim_dentry *dummy)
{
get_dentry_child_with_utf16le_name(const struct wim_dentry *dentry,
const utf16lechar *name,
size_t name_nbytes,
- CASE_SENSITIVITY_TYPE case_ctype)
+ CASE_SENSITIVITY_TYPE case_type)
{
const struct wim_inode *dir = dentry->d_inode;
- bool ignore_case = will_ignore_case(case_ctype);
+ bool ignore_case = will_ignore_case(case_type);
struct wim_dentry dummy;
struct wim_dentry *child;
- dummy.file_name = (utf16lechar*)name;
- dummy.file_name_nbytes = name_nbytes;
+ dummy.d_name = (utf16lechar*)name;
+ dummy.d_name_nbytes = name_nbytes;
if (!ignore_case)
/* Case-sensitive lookup. */
return 0;
}
-static int
-dentry_clear_inode_visited(struct wim_dentry *dentry, void *_ignore)
-{
- dentry->d_inode->i_visited = 0;
- return 0;
-}
-
-void
-dentry_tree_clear_inode_visited(struct wim_dentry *root)
-{
- for_dentry_in_tree(root, dentry_clear_inode_visited, NULL);
-}
-
/*
* Free a WIM dentry.
*
{
if (dentry) {
d_disassociate(dentry);
- FREE(dentry->file_name);
- FREE(dentry->short_name);
- FREE(dentry->_full_path);
+ FREE(dentry->d_name);
+ FREE(dentry->d_short_name);
+ FREE(dentry->d_full_path);
FREE(dentry);
}
}
* There will be an unnamed data stream, a reparse point stream, or both an
* unnamed data stream and a reparse point stream. In addition, there may be
* named data streams.
+ *
+ * NOTE: if the file has a reparse point stream or at least one named data
+ * stream, then WIMGAPI puts *all* streams in the extra stream entries and
+ * leaves the default stream hash zeroed. wimlib now does the same. However,
+ * for input we still support the default hash field being used, since wimlib
+ * used to use it and MS software is somewhat accepting of it as well.
*/
static void
assign_stream_types_unencrypted(struct wim_inode *inode)
if (stream_is_named(strm)) {
/* Named data stream */
strm->stream_type = STREAM_TYPE_DATA;
- } else if (!is_zero_hash(strm->_stream_hash)) {
+ } else if (i != 0 || !is_zero_hash(strm->_stream_hash)) {
+ /* Unnamed stream in the extra stream entries, OR the
+ * default stream in the dentry provided that it has a
+ * nonzero hash. */
if ((inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) &&
!found_reparse_point_stream) {
found_reparse_point_stream = true;
found_unnamed_data_stream = true;
strm->stream_type = STREAM_TYPE_DATA;
}
- } else {
- /* If no stream name is specified and the hash is zero,
- * then remember this stream for later so that we can
- * assign it to the unnamed data stream if we don't find
- * a better candidate. */
+ } else if (!unnamed_stream_with_zero_hash) {
unnamed_stream_with_zero_hash = strm;
}
}
- if (!found_unnamed_data_stream && unnamed_stream_with_zero_hash != NULL)
- unnamed_stream_with_zero_hash->stream_type = STREAM_TYPE_DATA;
+ if (unnamed_stream_with_zero_hash) {
+ int type = STREAM_TYPE_UNKNOWN;
+ if ((inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) &&
+ !found_reparse_point_stream) {
+ type = STREAM_TYPE_REPARSE_POINT;
+ } else if (!found_unnamed_data_stream) {
+ type = STREAM_TYPE_DATA;
+ }
+ unnamed_stream_with_zero_hash->stream_type = type;
+ }
}
/*
struct wim_dentry *dentry;
struct wim_inode *inode;
u16 short_name_nbytes;
- u16 file_name_nbytes;
+ u16 name_nbytes;
u64 calculated_size;
int ret;
- BUILD_BUG_ON(sizeof(struct wim_dentry_on_disk) != WIM_DENTRY_DISK_SIZE);
+ STATIC_ASSERT(sizeof(struct wim_dentry_on_disk) == WIM_DENTRY_DISK_SIZE);
/* Before reading the whole dentry, we need to read just the length.
* This is because a dentry of length 8 (that is, just the length field)
/* Read more fields: some into the dentry, and some into the inode. */
inode->i_attributes = le32_to_cpu(disk_dentry->attributes);
inode->i_security_id = le32_to_cpu(disk_dentry->security_id);
- dentry->subdir_offset = le64_to_cpu(disk_dentry->subdir_offset);
+ dentry->d_subdir_offset = le64_to_cpu(disk_dentry->subdir_offset);
inode->i_creation_time = le64_to_cpu(disk_dentry->creation_time);
inode->i_last_access_time = le64_to_cpu(disk_dentry->last_access_time);
inode->i_last_write_time = le64_to_cpu(disk_dentry->last_write_time);
+ inode->i_unknown_0x54 = le32_to_cpu(disk_dentry->unknown_0x54);
- /* I don't know what's going on here. It seems like M$ screwed up the
- * reparse points, then put the fields in the same place and didn't
- * document it. So we have some fields we read for reparse points, and
- * some fields in the same place for non-reparse-points. */
if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
- inode->i_rp_unknown_1 = le32_to_cpu(disk_dentry->reparse.rp_unknown_1);
inode->i_reparse_tag = le32_to_cpu(disk_dentry->reparse.reparse_tag);
- inode->i_rp_unknown_2 = le16_to_cpu(disk_dentry->reparse.rp_unknown_2);
- inode->i_not_rpfixed = le16_to_cpu(disk_dentry->reparse.not_rpfixed);
+ inode->i_rp_reserved = le16_to_cpu(disk_dentry->reparse.rp_reserved);
+ inode->i_rp_flags = le16_to_cpu(disk_dentry->reparse.rp_flags);
/* Leave inode->i_ino at 0. Note: this means that WIM cannot
* represent multiple hard links to a reparse point file. */
} else {
- inode->i_rp_unknown_1 = le32_to_cpu(disk_dentry->nonreparse.rp_unknown_1);
inode->i_ino = le64_to_cpu(disk_dentry->nonreparse.hard_link_group_id);
}
* name, and the short name. */
short_name_nbytes = le16_to_cpu(disk_dentry->short_name_nbytes);
- file_name_nbytes = le16_to_cpu(disk_dentry->file_name_nbytes);
+ name_nbytes = le16_to_cpu(disk_dentry->name_nbytes);
- if (unlikely((short_name_nbytes & 1) | (file_name_nbytes & 1))) {
+ if (unlikely((short_name_nbytes & 1) | (name_nbytes & 1))) {
ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE;
goto err_free_dentry;
}
/* We now know the length of the file name and short name. Make sure
* the length of the dentry is large enough to actually hold them. */
- calculated_size = dentry_min_len_with_names(file_name_nbytes,
+ calculated_size = dentry_min_len_with_names(name_nbytes,
short_name_nbytes);
if (unlikely(length < calculated_size)) {
/* Read the filename if present. Note: if the filename is empty, there
* is no null terminator following it. */
- if (file_name_nbytes) {
- dentry->file_name = utf16le_dupz(p, file_name_nbytes);
- if (dentry->file_name == NULL) {
+ if (name_nbytes) {
+ dentry->d_name = utf16le_dupz(p, name_nbytes);
+ if (unlikely(!dentry->d_name)) {
ret = WIMLIB_ERR_NOMEM;
goto err_free_dentry;
}
- dentry->file_name_nbytes = file_name_nbytes;
- p += (u32)file_name_nbytes + 2;
+ dentry->d_name_nbytes = name_nbytes;
+ p += (u32)name_nbytes + 2;
}
/* Read the short filename if present. Note: if there is no short
* filename, there is no null terminator following it. */
if (short_name_nbytes) {
- dentry->short_name = utf16le_dupz(p, short_name_nbytes);
- if (dentry->short_name == NULL) {
+ dentry->d_short_name = utf16le_dupz(p, short_name_nbytes);
+ if (unlikely(!dentry->d_short_name)) {
ret = WIMLIB_ERR_NOMEM;
goto err_free_dentry;
}
- dentry->short_name_nbytes = short_name_nbytes;
+ dentry->d_short_name_nbytes = short_name_nbytes;
p += (u32)short_name_nbytes + 2;
}
static bool
dentry_is_dot_or_dotdot(const struct wim_dentry *dentry)
{
- if (dentry->file_name_nbytes <= 4) {
- if (dentry->file_name_nbytes == 4) {
- if (dentry->file_name[0] == cpu_to_le16('.') &&
- dentry->file_name[1] == cpu_to_le16('.'))
+ if (dentry->d_name_nbytes <= 4) {
+ if (dentry->d_name_nbytes == 4) {
+ if (dentry->d_name[0] == cpu_to_le16('.') &&
+ dentry->d_name[1] == cpu_to_le16('.'))
return true;
- } else if (dentry->file_name_nbytes == 2) {
- if (dentry->file_name[0] == cpu_to_le16('.'))
+ } else if (dentry->d_name_nbytes == 2) {
+ if (dentry->d_name[0] == cpu_to_le16('.'))
return true;
}
}
static int
read_dentry_tree_recursive(const u8 * restrict buf, size_t buf_len,
- struct wim_dentry * restrict dir)
+ struct wim_dentry * restrict dir, unsigned depth)
{
- u64 cur_offset = dir->subdir_offset;
-
- /* Check for cyclic directory structure, which would cause infinite
- * recursion if not handled. */
- for (struct wim_dentry *d = dir->d_parent;
- !dentry_is_root(d); d = d->d_parent)
- {
- if (unlikely(d->subdir_offset == cur_offset)) {
- ERROR("Cyclic directory structure detected: children "
- "of \"%"TS"\" coincide with children of \"%"TS"\"",
- dentry_full_path(dir), dentry_full_path(d));
- return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
- }
+ u64 cur_offset = dir->d_subdir_offset;
+
+ /* Disallow extremely deep or cyclic directory structures */
+ if (unlikely(depth >= 16384)) {
+ ERROR("Directory structure too deep!");
+ return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
}
for (;;) {
/* If this child is a directory that itself has children, call
* this procedure recursively. */
- if (child->subdir_offset != 0) {
+ if (child->d_subdir_offset != 0) {
if (likely(dentry_is_directory(child))) {
ret = read_dentry_tree_recursive(buf,
buf_len,
- child);
+ child,
+ depth + 1);
if (ret)
return ret;
} else {
int ret;
struct wim_dentry *root;
- DEBUG("Reading dentry tree (root_offset=%"PRIu64")", root_offset);
-
ret = read_dentry(buf, buf_len, &root_offset, &root);
if (ret)
return ret;
goto err_free_dentry_tree;
}
- if (likely(root->subdir_offset != 0)) {
- ret = read_dentry_tree_recursive(buf, buf_len, root);
+ if (likely(root->d_subdir_offset != 0)) {
+ ret = read_dentry_tree_recursive(buf, buf_len, root, 0);
if (ret)
goto err_free_dentry_tree;
}
disk_dentry->attributes = cpu_to_le32(inode->i_attributes);
disk_dentry->security_id = cpu_to_le32(inode->i_security_id);
- disk_dentry->subdir_offset = cpu_to_le64(dentry->subdir_offset);
+ disk_dentry->subdir_offset = cpu_to_le64(dentry->d_subdir_offset);
disk_dentry->unused_1 = cpu_to_le64(0);
disk_dentry->unused_2 = cpu_to_le64(0);
disk_dentry->creation_time = cpu_to_le64(inode->i_creation_time);
disk_dentry->last_access_time = cpu_to_le64(inode->i_last_access_time);
disk_dentry->last_write_time = cpu_to_le64(inode->i_last_write_time);
+ disk_dentry->unknown_0x54 = cpu_to_le32(inode->i_unknown_0x54);
if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
- disk_dentry->reparse.rp_unknown_1 = cpu_to_le32(inode->i_rp_unknown_1);
disk_dentry->reparse.reparse_tag = cpu_to_le32(inode->i_reparse_tag);
- disk_dentry->reparse.rp_unknown_2 = cpu_to_le16(inode->i_rp_unknown_2);
- disk_dentry->reparse.not_rpfixed = cpu_to_le16(inode->i_not_rpfixed);
+ disk_dentry->reparse.rp_reserved = cpu_to_le16(inode->i_rp_reserved);
+ disk_dentry->reparse.rp_flags = cpu_to_le16(inode->i_rp_flags);
} else {
- disk_dentry->nonreparse.rp_unknown_1 = cpu_to_le32(inode->i_rp_unknown_1);
disk_dentry->nonreparse.hard_link_group_id =
cpu_to_le64((inode->i_nlink == 1) ? 0 : inode->i_ino);
}
- disk_dentry->short_name_nbytes = cpu_to_le16(dentry->short_name_nbytes);
- disk_dentry->file_name_nbytes = cpu_to_le16(dentry->file_name_nbytes);
+ disk_dentry->short_name_nbytes = cpu_to_le16(dentry->d_short_name_nbytes);
+ disk_dentry->name_nbytes = cpu_to_le16(dentry->d_name_nbytes);
p += sizeof(struct wim_dentry_on_disk);
wimlib_assert(dentry_is_root(dentry) != dentry_has_long_name(dentry));
if (dentry_has_long_name(dentry))
- p = mempcpy(p, dentry->file_name, (u32)dentry->file_name_nbytes + 2);
+ p = mempcpy(p, dentry->d_name, (u32)dentry->d_name_nbytes + 2);
if (dentry_has_short_name(dentry))
- p = mempcpy(p, dentry->short_name, (u32)dentry->short_name_nbytes + 2);
+ p = mempcpy(p, dentry->d_short_name, (u32)dentry->d_short_name_nbytes + 2);
/* Align to 8-byte boundary */
while ((uintptr_t)p & 7)
static int
write_dir_dentries(struct wim_dentry *dir, void *_pp)
{
- if (dir->subdir_offset != 0) {
+ if (dir->d_subdir_offset != 0) {
u8 **pp = _pp;
u8 *p = *pp;
struct wim_dentry *child;
u8 *
write_dentry_tree(struct wim_dentry *root, u8 *p)
{
- DEBUG("Writing dentry tree.");
-
- wimlib_assert(root != NULL);
-
/* write root dentry and end-of-directory entry following it */
p = write_dentry(root, p);
*(u64*)p = 0;