*/
/*
- * Copyright (C) 2012, 2013, 2014, 2015 Eric Biggers
+ * Copyright 2012-2023 Eric Biggers
*
* This file is free software; you can redistribute it and/or modify it under
* the terms of the GNU Lesser General Public License as published by the Free
* details.
*
* You should have received a copy of the GNU Lesser General Public License
- * along with this file; if not, see http://www.gnu.org/licenses/.
+ * along with this file; if not, see https://www.gnu.org/licenses/.
*/
/*
* - wimlib does not allow *directory* hard links, so a WIM image really does
* have a *tree* of dentries (and not an arbitrary graph of dentries).
*
- * - wimlib indexes dentries both case-insensitively and case-sensitively,
- * allowing either behavior to be used for path lookup.
+ * - wimlib supports both case-sensitive and case-insensitive path lookups.
+ * The implementation uses a single in-memory index per directory, using a
+ * collation order like that used by NTFS; see collate_dentry_names().
*
* - Multiple dentries in a directory might have the same case-insensitive
* name. But wimlib enforces that at most one dentry in a directory can have
le32 attributes;
/* A value that specifies the security descriptor for this file or
- * directory. If -1, the file or directory has no security descriptor.
- * Otherwise, it is a 0-based index into the WIM image's table of
- * security descriptors (see: `struct wim_security_data') */
- sle32 security_id;
+ * directory. If 0xFFFFFFFF, the file or directory has no security
+ * descriptor. Otherwise, it is a 0-based index into the WIM image's
+ * table of security descriptors (see: `struct wim_security_data') */
+ le32 security_id;
/* Offset, in bytes, from the start of the uncompressed metadata
* resource of this directory's child directory entries, or 0 if this
le64 last_write_time;
/*
- * Usually this is the SHA-1 message digest of the file's "contents"
- * (the unnamed data stream).
- *
- * If the file has FILE_ATTRIBUTE_REPARSE_POINT set, then this is
- * instead usually the SHA-1 message digest of the uncompressed reparse
- * point data.
- *
- * However, there are some special rules that need to be applied to
- * interpret this field correctly when extra stream entries are present.
- * See the code for details.
+ * Usually this is the SHA-1 message digest of the file's contents, or
+ * all zeroes if the file is a directory or is empty. However, special
+ * rules apply if the file has FILE_ATTRIBUTE_REPARSE_POINT set or has
+ * named data streams. See assign_stream_types_unencrypted().
*/
- u8 default_hash[SHA1_HASH_SIZE];
+ u8 main_hash[SHA1_HASH_SIZE];
/* Unknown field (maybe accidental padding) */
le32 unknown_0x54;
le32 reparse_tag;
le16 rp_reserved;
le16 rp_flags;
- } _packed_attribute reparse;
+ } __attribute__((packed)) reparse;
struct {
le64 hard_link_group_id;
- } _packed_attribute nonreparse;
+ } __attribute__((packed)) nonreparse;
};
/* Number of extra stream entries that directly follow this dentry
* field) after 8-byte alignment, then the remaining space will be a
* variable-length list of tagged metadata items. See tagged_items.c
* for more information. */
- /* u8 tagged_items[] _aligned_attribute(8); */
+ /* u8 tagged_items[] __attribute__((aligned(8))); */
-} _packed_attribute;
+} __attribute__((packed));
/* If num_extra_streams != 0, then there are that many extra stream
* entries following the dentry, starting on the next 8-byte aligned
* boundary. They are not counted in the 'length' field of the dentry.
* the null terminator. There is a null terminator character if
* @name_nbytes != 0; i.e., if this stream is named. */
utf16lechar name[];
-} _packed_attribute;
+} __attribute__((packed));
static void
do_dentry_set_name(struct wim_dentry *dentry, utf16lechar *name,
{
const struct wim_inode *inode = dentry->d_inode;
size_t len;
+ unsigned num_unnamed_streams = 0;
+ bool have_named_data_stream = false;
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);
-
- if (!(inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED)) {
- /*
- * Extra stream entries:
- *
- * - Use one extra stream entry for each named data stream
- * - Use one extra stream entry for the unnamed data stream when there is either:
- * - a reparse point stream
- * - at least one named data stream (for Windows PE bug workaround)
- * - Use one extra stream entry for the reparse point stream if there is one
- */
- bool have_named_data_stream = false;
- bool have_reparse_point_stream = false;
+ if (inode->i_extra)
+ len += ALIGN(inode->i_extra->size, 8);
+
+ /*
+ * Calculate the total length of the extra stream entries that will be
+ * written. To match DISM, some odd rules need to be followed here.
+ * See write_dentry_streams() for explanation. Keep this in sync with
+ * write_dentry_streams()!
+ */
+ if (inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED) {
+ num_unnamed_streams++;
+ } else {
for (unsigned i = 0; i < inode->i_num_streams; i++) {
const struct wim_inode_stream *strm = &inode->i_streams[i];
+
if (stream_is_named_data_stream(strm)) {
len += stream_out_total_length(strm);
have_named_data_stream = true;
- } else if (strm->stream_type == STREAM_TYPE_REPARSE_POINT) {
- wimlib_assert(inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT);
- have_reparse_point_stream = true;
}
}
-
- if (have_named_data_stream || have_reparse_point_stream) {
- if (have_reparse_point_stream)
- len += ALIGN(sizeof(struct wim_extra_stream_entry_on_disk), 8);
- len += ALIGN(sizeof(struct wim_extra_stream_entry_on_disk), 8);
- }
+ if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT)
+ num_unnamed_streams++;
+ if (!(inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY))
+ num_unnamed_streams++;
}
-
+ if (num_unnamed_streams > 1 || have_named_data_stream)
+ len += num_unnamed_streams *
+ ALIGN(sizeof(struct wim_extra_stream_entry_on_disk), 8);
return len;
}
* @arg will be passed as the second argument to each invocation of @visitor.
*
* This function does a pre-order traversal --- that is, a parent will be
- * visited before its children. It also will visit siblings in order of
- * case-sensitive filename. Equivalently, this function visits the entire tree
- * in the case-sensitive lexicographic order of the full paths.
+ * visited before its children. Furthermore, siblings will be visited in their
+ * collation order.
*
* It is safe to pass NULL for @root, which means that the dentry tree is empty.
* In this case, this function does nothing.
calculate_dentry_full_path(struct wim_dentry *dentry)
{
size_t ulen;
- size_t dummy;
const struct wim_dentry *d;
if (dentry->d_full_path)
d = dentry;
do {
p -= d->d_name_nbytes / sizeof(utf16lechar);
- memcpy(p, d->d_name, d->d_name_nbytes);
+ if (d->d_name_nbytes)
+ 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->d_full_path, &dummy);
+ &dentry->d_full_path, NULL);
}
/*
for_dentry_in_tree(root, dentry_calculate_subdir_offset, subdir_offset_p);
}
-/* Compare the UTF-16LE long filenames of two dentries case insensitively. */
static int
-dentry_compare_names_case_insensitive(const struct wim_dentry *d1,
- const struct wim_dentry *d2)
+dentry_compare_names(const struct wim_dentry *d1, const struct wim_dentry *d2,
+ bool ignore_case)
{
- return cmp_utf16le_strings(d1->d_name,
- d1->d_name_nbytes / 2,
- d2->d_name,
- d2->d_name_nbytes / 2,
- true);
-}
-
-/* Compare the UTF-16LE long filenames of two dentries case sensitively. */
-static int
-dentry_compare_names_case_sensitive(const struct wim_dentry *d1,
- const struct wim_dentry *d2)
-{
- return cmp_utf16le_strings(d1->d_name,
- d1->d_name_nbytes / 2,
- d2->d_name,
- d2->d_name_nbytes / 2,
- false);
-}
-
-static int
-_avl_dentry_compare_names_ci(const struct avl_tree_node *n1,
- const struct avl_tree_node *n2)
-{
- const struct wim_dentry *d1, *d2;
-
- d1 = avl_tree_entry(n1, struct wim_dentry, d_index_node_ci);
- d2 = avl_tree_entry(n2, struct wim_dentry, d_index_node_ci);
- return dentry_compare_names_case_insensitive(d1, d2);
+ return cmp_utf16le_strings(d1->d_name, d1->d_name_nbytes / 2,
+ d2->d_name, d2->d_name_nbytes / 2,
+ ignore_case);
}
+/*
+ * Collate (compare) the long filenames of two dentries. This first compares
+ * the names ignoring case, then falls back to a case-sensitive comparison if
+ * the names are the same ignoring case.
+ */
static int
-_avl_dentry_compare_names(const struct avl_tree_node *n1,
- const struct avl_tree_node *n2)
+collate_dentry_names(const struct avl_tree_node *n1,
+ const struct avl_tree_node *n2)
{
const struct wim_dentry *d1, *d2;
+ int res;
d1 = avl_tree_entry(n1, struct wim_dentry, d_index_node);
d2 = avl_tree_entry(n2, struct wim_dentry, d_index_node);
- return dentry_compare_names_case_sensitive(d1, d2);
+
+ res = dentry_compare_names(d1, d2, true);
+ if (res)
+ return res;
+ return dentry_compare_names(d1, d2, false);
}
/* Default case sensitivity behavior for searches with
* WIMLIB_INIT_FLAG_DEFAULT_CASE_SENSITIVE or
* WIMLIB_INIT_FLAG_DEFAULT_CASE_INSENSITIVE to wimlib_global_init(). */
bool default_ignore_case =
-#ifdef __WIN32__
+#ifdef _WIN32
true
#else
false
#endif
;
-/* 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)
-{
- struct avl_tree_node *node;
-
- node = avl_tree_lookup_node(dir->i_children,
- &dummy->d_index_node,
- _avl_dentry_compare_names);
- if (!node)
- return NULL;
- return avl_tree_entry(node, struct wim_dentry, d_index_node);
-}
-
-/* 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)
-{
- struct avl_tree_node *node;
-
- node = avl_tree_lookup_node(dir->i_children_ci,
- &dummy->d_index_node_ci,
- _avl_dentry_compare_names_ci);
- if (!node)
- return NULL;
- return avl_tree_entry(node, struct wim_dentry, d_index_node_ci);
-}
-
-/* Given a UTF-16LE filename and a directory, look up the dentry for the file.
- * Return it if found, otherwise NULL. This has configurable case sensitivity,
- * and @name need not be null-terminated. */
+/*
+ * Find the dentry within the given directory that has the given UTF-16LE
+ * filename. Return it if found, otherwise return NULL. This has configurable
+ * case sensitivity, and @name need not be null-terminated.
+ */
struct wim_dentry *
-get_dentry_child_with_utf16le_name(const struct wim_dentry *dentry,
+get_dentry_child_with_utf16le_name(const struct wim_dentry *dir,
const utf16lechar *name,
size_t name_nbytes,
CASE_SENSITIVITY_TYPE case_type)
{
- const struct wim_inode *dir = dentry->d_inode;
- bool ignore_case = will_ignore_case(case_type);
- struct wim_dentry dummy;
- struct wim_dentry *child;
+ struct wim_dentry wanted;
+ struct avl_tree_node *cur = dir->d_inode->i_children;
+ struct wim_dentry *ci_match = NULL;
+
+ wanted.d_name = (utf16lechar *)name;
+ wanted.d_name_nbytes = name_nbytes;
- dummy.d_name = (utf16lechar*)name;
- dummy.d_name_nbytes = name_nbytes;
+ if (unlikely(wanted.d_name_nbytes != name_nbytes))
+ return NULL; /* overflow */
- if (!ignore_case)
- /* Case-sensitive lookup. */
- return dir_lookup(dir, &dummy);
+ /* Note: we can't use avl_tree_lookup_node() here because we need to
+ * save case-insensitive matches. */
+ while (cur) {
+ struct wim_dentry *child;
+ int res;
- /* Case-insensitive lookup. */
+ child = avl_tree_entry(cur, struct wim_dentry, d_index_node);
- child = dir_lookup_ci(dir, &dummy);
- if (!child)
+ res = dentry_compare_names(&wanted, child, true);
+ if (!res) {
+ /* case-insensitive match found */
+ ci_match = child;
+
+ res = dentry_compare_names(&wanted, child, false);
+ if (!res)
+ return child; /* case-sensitive match found */
+ }
+
+ if (res < 0)
+ cur = cur->left;
+ else
+ cur = cur->right;
+ }
+
+ /* No case-sensitive match; use a case-insensitive match if possible. */
+
+ if (!will_ignore_case(case_type))
return NULL;
- if (likely(list_empty(&child->d_ci_conflict_list)))
- /* Only one dentry has this case-insensitive name; return it */
- return child;
+ if (ci_match) {
+ size_t num_other_ci_matches = 0;
+ struct wim_dentry *other_ci_match, *d;
- /* Multiple dentries have the same case-insensitive name. Choose the
- * dentry with the same case-sensitive name, if one exists; otherwise
- * print a warning and choose one of the possible dentries arbitrarily.
- */
- struct wim_dentry *alt = child;
- size_t num_alts = 0;
+ dentry_for_each_ci_match(d, ci_match) {
+ num_other_ci_matches++;
+ other_ci_match = d;
+ }
- do {
- num_alts++;
- if (!dentry_compare_names_case_sensitive(&dummy, alt))
- return alt;
- alt = list_entry(alt->d_ci_conflict_list.next,
- struct wim_dentry, d_ci_conflict_list);
- } while (alt != child);
-
- WARNING("Result of case-insensitive lookup is ambiguous\n"
- " (returning \"%"TS"\" of %zu "
- "possible files, including \"%"TS"\")",
- dentry_full_path(child),
- num_alts,
- dentry_full_path(list_entry(child->d_ci_conflict_list.next,
- struct wim_dentry,
- d_ci_conflict_list)));
- return child;
+ if (num_other_ci_matches != 0) {
+ WARNING("Result of case-insensitive lookup is ambiguous\n"
+ " (returning \"%"TS"\" of %zu "
+ "possible files, including \"%"TS"\")",
+ dentry_full_path(ci_match), num_other_ci_matches,
+ dentry_full_path(other_ci_match));
+ }
+ }
+
+ return ci_match;
}
-/* Given a 'tchar' filename and a directory, look up the dentry for the file.
- * If the filename was successfully converted to UTF-16LE and the dentry was
- * found, return it; otherwise return NULL. This has configurable case
- * sensitivity. */
+/*
+ * Find the dentry within the given directory that has the given 'tstr'
+ * filename. If the filename was successfully converted to UTF-16LE and the
+ * dentry was found, return it; otherwise return NULL. This has configurable
+ * case sensitivity.
+ */
struct wim_dentry *
-get_dentry_child_with_name(const struct wim_dentry *dentry, const tchar *name,
+get_dentry_child_with_name(const struct wim_dentry *dir, const tchar *name,
CASE_SENSITIVITY_TYPE case_type)
{
int ret;
if (ret)
return NULL;
- child = get_dentry_child_with_utf16le_name(dentry,
+ child = get_dentry_child_with_utf16le_name(dir,
name_utf16le,
name_utf16le_nbytes,
case_type);
for_dentry_in_tree_depth(root, f, blob_table);
}
-/* Insert the @child dentry into the case sensitive index of the @dir directory.
- * Return NULL if successfully inserted, otherwise a pointer to the
- * already-inserted duplicate. */
-static struct wim_dentry *
-dir_index_child(struct wim_inode *dir, struct wim_dentry *child)
+/*
+ * Return the first dentry in the list of dentries which have the same
+ * case-insensitive name as the one given.
+ */
+struct wim_dentry *
+dentry_get_first_ci_match(struct wim_dentry *dentry)
{
- struct avl_tree_node *duplicate;
+ struct wim_dentry *ci_match = dentry;
- duplicate = avl_tree_insert(&dir->i_children,
- &child->d_index_node,
- _avl_dentry_compare_names);
- if (!duplicate)
- return NULL;
- return avl_tree_entry(duplicate, struct wim_dentry, d_index_node);
-}
+ for (;;) {
+ struct avl_tree_node *node;
+ struct wim_dentry *prev;
+
+ node = avl_tree_prev_in_order(&ci_match->d_index_node);
+ if (!node)
+ break;
+ prev = avl_tree_entry(node, struct wim_dentry, d_index_node);
+ if (dentry_compare_names(prev, dentry, true))
+ break;
+ ci_match = prev;
+ }
-/* Insert the @child dentry into the case insensitive index of the @dir
- * directory. Return NULL if successfully inserted, otherwise a pointer to the
- * already-inserted duplicate. */
-static struct wim_dentry *
-dir_index_child_ci(struct wim_inode *dir, struct wim_dentry *child)
-{
- struct avl_tree_node *duplicate;
+ if (ci_match == dentry)
+ return dentry_get_next_ci_match(dentry, dentry);
- duplicate = avl_tree_insert(&dir->i_children_ci,
- &child->d_index_node_ci,
- _avl_dentry_compare_names_ci);
- if (!duplicate)
- return NULL;
- return avl_tree_entry(duplicate, struct wim_dentry, d_index_node_ci);
+ return ci_match;
}
-/* Remove the specified dentry from its directory's case-sensitive index. */
-static void
-dir_unindex_child(struct wim_inode *dir, struct wim_dentry *child)
+/*
+ * Return the next dentry in the list of dentries which have the same
+ * case-insensitive name as the one given.
+ */
+struct wim_dentry *
+dentry_get_next_ci_match(struct wim_dentry *dentry, struct wim_dentry *ci_match)
{
- avl_tree_remove(&dir->i_children, &child->d_index_node);
-}
+ do {
+ struct avl_tree_node *node;
-/* Remove the specified dentry from its directory's case-insensitive index. */
-static void
-dir_unindex_child_ci(struct wim_inode *dir, struct wim_dentry *child)
-{
- avl_tree_remove(&dir->i_children_ci, &child->d_index_node_ci);
-}
+ node = avl_tree_next_in_order(&ci_match->d_index_node);
+ if (!node)
+ return NULL;
+ ci_match = avl_tree_entry(node, struct wim_dentry, d_index_node);
+ } while (ci_match == dentry);
-/* Return true iff the specified dentry is in its parent directory's
- * case-insensitive index. */
-static bool
-dentry_in_ci_index(const struct wim_dentry *dentry)
-{
- return !avl_tree_node_is_unlinked(&dentry->d_index_node_ci);
+ if (dentry_compare_names(ci_match, dentry, true))
+ return NULL;
+
+ return ci_match;
}
/*
- * Link a dentry into the tree.
+ * Link a dentry into a directory.
*
* @parent:
- * The dentry that will be the parent of @child. It must name a directory.
+ * The directory into which to link the dentry.
*
* @child:
- * The dentry to link. It must be currently unlinked.
+ * The dentry to link into the directory. It must be currently unlinked.
*
- * Returns NULL if successful. If @parent already contains a dentry with the
- * same case-sensitive name as @child, returns a pointer to this duplicate
- * dentry.
+ * Returns NULL if successful; or, if @parent already contains a dentry with the
+ * same case-sensitive name as @child, then a pointer to this duplicate dentry
+ * is returned.
*/
struct wim_dentry *
dentry_add_child(struct wim_dentry *parent, struct wim_dentry *child)
{
- struct wim_dentry *duplicate;
- struct wim_inode *dir;
+ struct wim_inode *dir = parent->d_inode;
+ struct avl_tree_node *duplicate;
wimlib_assert(parent != child);
-
- dir = parent->d_inode;
-
wimlib_assert(inode_is_directory(dir));
- duplicate = dir_index_child(dir, child);
+ duplicate = avl_tree_insert(&dir->i_children, &child->d_index_node,
+ collate_dentry_names);
if (duplicate)
- return duplicate;
+ return avl_tree_entry(duplicate, struct wim_dentry, d_index_node);
- duplicate = dir_index_child_ci(dir, child);
- if (duplicate) {
- list_add(&child->d_ci_conflict_list, &duplicate->d_ci_conflict_list);
- avl_tree_node_set_unlinked(&child->d_index_node_ci);
- } else {
- INIT_LIST_HEAD(&child->d_ci_conflict_list);
- }
child->d_parent = parent;
return NULL;
}
-/* Unlink a dentry from the tree. */
+/* Unlink a dentry from its parent directory. */
void
unlink_dentry(struct wim_dentry *dentry)
{
- struct wim_inode *dir;
-
/* Do nothing if the dentry is root or it's already unlinked. Not
* actually necessary based on the current callers, but we do the check
* here to be safe. */
if (unlikely(dentry->d_parent == dentry))
return;
- dir = dentry->d_parent->d_inode;
-
- dir_unindex_child(dir, dentry);
-
- if (dentry_in_ci_index(dentry)) {
-
- dir_unindex_child_ci(dir, dentry);
-
- if (!list_empty(&dentry->d_ci_conflict_list)) {
- /* Make a different case-insensitively-the-same dentry
- * be the "representative" in the search index. */
- struct list_head *next;
- struct wim_dentry *other;
- struct wim_dentry *existing;
-
- next = dentry->d_ci_conflict_list.next;
- other = list_entry(next, struct wim_dentry, d_ci_conflict_list);
- existing = dir_index_child_ci(dir, other);
- wimlib_assert(existing == NULL);
- }
- }
- list_del(&dentry->d_ci_conflict_list);
+ avl_tree_remove(&dentry->d_parent->d_inode->i_children,
+ &dentry->d_index_node);
/* Not actually necessary, but to be safe don't retain the now-obsolete
* parent pointer. */
p++;
if (unlikely(p < end)) {
- inode->i_extra = memdup(p, end - p);
+ inode->i_extra = MALLOC(sizeof(struct wim_inode_extra) +
+ end - p);
if (!inode->i_extra)
return WIMLIB_ERR_NOMEM;
- inode->i_extra_size = end - p;
+ inode->i_extra->size = end - p;
+ memcpy(inode->i_extra->data, p, end - p);
}
return 0;
}
/*
* Set the type of each stream for an unencrypted file.
*
- * 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.
+ * To specify the streams of each file, the WIM provides a main_hash and an
+ * optional list of "extra stream entries". Each extra stream entry is a
+ * (name, hash) pair where the name is optional. Hashes can be the special
+ * value of zero_hash, which means the stream is empty (zero-length).
+ *
+ * While extra stream entries with names always refer to "named data streams",
+ * the main hash and any extra unnamed hashes can be hard to interpret. This is
+ * because the WIM file format unfortunately doesn't make it very clear which is
+ * the unnamed data stream (i.e. standard file contents) and which is the
+ * reparse stream. The way this ambiguity is resolved (based on what MS
+ * software seems to do) is by (1) a file can have at most one unnamed data
+ * stream and at most one reparse stream, (2) a reparse stream is present if and
+ * only if the file has FILE_ATTRIBUTE_REPARSE_POINT, and (3) the reparse
+ * stream, if present, is stored before the unnamed data stream if present
+ * (considering main_hash to come before any extra hashes). Note: directories
+ * need not have an unnamed data stream stored, even with a zero hash, as
+ * "unnamed data stream" isn't meaningful for a directory in the first place.
+ *
+ * With those rules in mind, one would expect that the first unnamed stream
+ * would use main_hash, and the second (if present) would use an extra stream
+ * entry. However, there is another quirk that we must be compatible with:
+ * sometimes main_hash isn't used and only extra stream entries are used. To
+ * handle this, we ignore main_hash if it is zero and there is at least one
+ * unnamed extra stream entry. This works correctly as long as a zero main_hash
+ * and an unnamed extra stream entry is never used to represent an empty reparse
+ * stream and an unnamed data stream. (It's not, as the reparse stream always
+ * goes in the extra stream entries in this case. See write_dentry_streams().)
*/
static void
assign_stream_types_unencrypted(struct wim_inode *inode)
{
- bool found_reparse_point_stream = false;
+ bool found_reparse_stream = false;
bool found_unnamed_data_stream = false;
- struct wim_inode_stream *unnamed_stream_with_zero_hash = NULL;
for (unsigned i = 0; i < inode->i_num_streams; i++) {
struct wim_inode_stream *strm = &inode->i_streams[i];
if (stream_is_named(strm)) {
- /* Named data stream */
+ /* Named extra stream entry */
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 extra stream entry or a nonzero main_hash */
if ((inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) &&
- !found_reparse_point_stream) {
- found_reparse_point_stream = true;
+ !found_reparse_stream) {
+ found_reparse_stream = true;
strm->stream_type = STREAM_TYPE_REPARSE_POINT;
} else if (!found_unnamed_data_stream) {
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. */
- unnamed_stream_with_zero_hash = strm;
- }
+ } /* Else, too many unnamed streams were found. */
+
+ } /* Else, it's a zero main_hash. */
}
- if (!found_unnamed_data_stream && unnamed_stream_with_zero_hash != NULL)
- unnamed_stream_with_zero_hash->stream_type = STREAM_TYPE_DATA;
+ /* If needed, use the zero main_hash. */
+ if (!found_reparse_stream && !found_unnamed_data_stream) {
+ inode->i_streams[0].stream_type =
+ (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) ?
+ STREAM_TYPE_REPARSE_POINT : STREAM_TYPE_DATA;
+ }
}
/*
*/
static int
setup_inode_streams(const u8 *p, const u8 *end, struct wim_inode *inode,
- unsigned num_extra_streams, const u8 *default_hash,
+ unsigned num_extra_streams, const u8 *main_hash,
u64 *offset_p)
{
const u8 *orig_p = p;
return WIMLIB_ERR_NOMEM;
}
- /* Use the default hash field for the first stream */
+ /* Use main_hash for the first stream. */
inode->i_streams[0].stream_name = (utf16lechar *)NO_STREAM_NAME;
- copy_hash(inode->i_streams[0]._stream_hash, default_hash);
+ copy_hash(inode->i_streams[0]._stream_hash, main_hash);
inode->i_streams[0].stream_type = STREAM_TYPE_UNKNOWN;
inode->i_streams[0].stream_id = 0;
- /* Read the extra stream entries */
+ /* Read the extra stream entries. */
for (unsigned i = 1; i < inode->i_num_streams; i++) {
struct wim_inode_stream *strm;
const struct wim_extra_stream_entry_on_disk *disk_strm;
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)
&buf[buf_len],
inode,
le16_to_cpu(disk_dentry->num_extra_streams),
- disk_dentry->default_hash,
+ disk_dentry->main_hash,
&offset);
if (ret)
goto err_free_dentry;
return ret;
}
-/* Is the dentry named "." or ".." ? */
static bool
dentry_is_dot_or_dotdot(const struct wim_dentry *dentry)
{
return false;
}
+static bool
+dentry_contains_embedded_null(const struct wim_dentry *dentry)
+{
+ for (unsigned i = 0; i < dentry->d_name_nbytes / 2; i++)
+ if (dentry->d_name[i] == cpu_to_le16('\0'))
+ return true;
+ return false;
+}
+
+static bool
+should_ignore_dentry(struct wim_dentry *dir, const struct wim_dentry *dentry)
+{
+ /* All dentries except the root must be named. */
+ if (!dentry_has_long_name(dentry)) {
+ WARNING("Ignoring unnamed file in directory \"%"TS"\"",
+ dentry_full_path(dir));
+ return true;
+ }
+
+ /* Don't allow files named "." or "..". Such filenames could be used in
+ * path traversal attacks. */
+ if (dentry_is_dot_or_dotdot(dentry)) {
+ WARNING("Ignoring file named \".\" or \"..\" in directory "
+ "\"%"TS"\"", dentry_full_path(dir));
+ return true;
+ }
+
+ /* Don't allow filenames containing embedded null characters. Although
+ * the null character is already considered an unsupported character for
+ * extraction by all targets, it is probably a good idea to just forbid
+ * such names entirely. */
+ if (dentry_contains_embedded_null(dentry)) {
+ WARNING("Ignoring filename with embedded null character in "
+ "directory \"%"TS"\"", dentry_full_path(dir));
+ return true;
+ }
+
+ return false;
+}
+
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->d_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->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;
- }
+ /* Disallow extremely deep or cyclic directory structures */
+ if (unlikely(depth >= 16384)) {
+ ERROR("Directory structure too deep!");
+ return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
}
for (;;) {
if (child == NULL)
return 0;
- /* All dentries except the root should be named. */
- if (unlikely(!dentry_has_long_name(child))) {
- WARNING("Ignoring unnamed dentry in "
- "directory \"%"TS"\"", dentry_full_path(dir));
- free_dentry(child);
- continue;
- }
-
- /* Don't allow files named "." or "..". */
- if (unlikely(dentry_is_dot_or_dotdot(child))) {
- WARNING("Ignoring file named \".\" or \"..\"; "
- "potentially malicious archive!!!");
+ /* Ignore dentries with bad names. */
+ if (unlikely(should_ignore_dentry(dir, child))) {
free_dentry(child);
continue;
}
if (likely(dentry_is_directory(child))) {
ret = read_dentry_tree_recursive(buf,
buf_len,
- child);
+ child,
+ depth + 1);
if (ret)
return ret;
} else {
}
if (likely(root->d_subdir_offset != 0)) {
- ret = read_dentry_tree_recursive(buf, buf_len, root);
+ ret = read_dentry_tree_recursive(buf, buf_len, root, 0);
if (ret)
goto err_free_dentry_tree;
}
return p;
}
+/*
+ * Write the stream references for a WIM dentry. To be compatible with DISM, we
+ * follow the below rules:
+ *
+ * 1. If the file has FILE_ATTRIBUTE_ENCRYPTED, then only the EFSRPC_RAW_DATA
+ * stream is stored. Otherwise, the streams that are stored are:
+ * - Reparse stream if the file has FILE_ATTRIBUTE_REPARSE_POINT
+ * - Unnamed data stream if the file doesn't have FILE_ATTRIBUTE_DIRECTORY
+ * - Named data streams
+ *
+ * 2. If only one stream is being stored and it is the EFSRPC_RAW_DATA, unnamed
+ * data, or reparse stream, then its hash goes in main_hash, and no extra
+ * stream entries are stored. Otherwise, *all* streams go in the extra
+ * stream entries, and main_hash is left zeroed!
+ *
+ * 3. If both the reparse stream and unnamed data stream are being stored, then
+ * the reparse stream comes first.
+ *
+ * 4. The unnamed stream(s) come before the named stream(s). (Actually, DISM
+ * puts the named streams between the first and second unnamed streams, but
+ * this is incompatible with itself... Tested with DISM 10.0.20348.681.)
+ *
+ * wimlib v1.14.1 and earlier behaved slightly differently for directories.
+ * First, wimlib always put the hash of the reparse stream in an extra stream
+ * entry, never in main_hash. This difference vs. DISM went unnoticed for a
+ * long time, but eventually it was found that it broke the Windows 8 setup
+ * wizard. Second, when a directory had any extra streams, wimlib created an
+ * extra stream entry to represent the (empty) unnamed data stream. However,
+ * DISM now rejects that (though I think it used to accept it). There isn't
+ * really any such thing as "unnamed data stream" for a directory.
+ *
+ * Keep this in sync with dentry_out_total_length()!
+ */
+static u8 *
+write_dentry_streams(const struct wim_inode *inode,
+ struct wim_dentry_on_disk *disk_dentry, u8 *p)
+{
+ const u8 *unnamed_data_stream_hash = zero_hash;
+ const u8 *reparse_stream_hash = zero_hash;
+ const u8 *efsrpc_stream_hash = zero_hash;
+ const u8 *unnamed_stream_hashes[2] = { zero_hash };
+ unsigned num_unnamed_streams = 0;
+ unsigned num_named_streams = 0;
+
+ for (unsigned i = 0; i < inode->i_num_streams; i++) {
+ const struct wim_inode_stream *strm = &inode->i_streams[i];
+
+ switch (strm->stream_type) {
+ case STREAM_TYPE_DATA:
+ if (stream_is_named(strm))
+ num_named_streams++;
+ else
+ unnamed_data_stream_hash = stream_hash(strm);
+ break;
+ case STREAM_TYPE_REPARSE_POINT:
+ reparse_stream_hash = stream_hash(strm);
+ break;
+ case STREAM_TYPE_EFSRPC_RAW_DATA:
+ efsrpc_stream_hash = stream_hash(strm);
+ break;
+ }
+ }
+
+ if (inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED) {
+ unnamed_stream_hashes[num_unnamed_streams++] = efsrpc_stream_hash;
+ num_named_streams = 0;
+ } else {
+ if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT)
+ unnamed_stream_hashes[num_unnamed_streams++] = reparse_stream_hash;
+ if (!(inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY))
+ unnamed_stream_hashes[num_unnamed_streams++] = unnamed_data_stream_hash;
+ }
+
+ if (num_unnamed_streams <= 1 && num_named_streams == 0) {
+ /* No extra stream entries are needed. */
+ copy_hash(disk_dentry->main_hash, unnamed_stream_hashes[0]);
+ disk_dentry->num_extra_streams = 0;
+ return p;
+ }
+
+ /* Else, all streams go in extra stream entries. */
+ copy_hash(disk_dentry->main_hash, zero_hash);
+ wimlib_assert(num_unnamed_streams + num_named_streams <= 0xFFFF);
+ disk_dentry->num_extra_streams = cpu_to_le16(num_unnamed_streams +
+ num_named_streams);
+ for (unsigned i = 0; i < num_unnamed_streams; i++)
+ p = write_extra_stream_entry(p, NO_STREAM_NAME,
+ unnamed_stream_hashes[i]);
+ for (unsigned i = 0; i < inode->i_num_streams; i++) {
+ const struct wim_inode_stream *strm = &inode->i_streams[i];
+
+ if (stream_is_named_data_stream(strm)) {
+ p = write_extra_stream_entry(p, strm->stream_name,
+ stream_hash(strm));
+ }
+ }
+ return p;
+}
+
/*
* Write a WIM dentry to an output buffer.
*
while ((uintptr_t)p & 7)
*p++ = 0;
- if (inode->i_extra_size) {
+ if (inode->i_extra) {
/* Extra tagged items --- not usually present. */
- p = mempcpy(p, inode->i_extra, inode->i_extra_size);
+ p = mempcpy(p, inode->i_extra->data, inode->i_extra->size);
/* Align to 8-byte boundary */
while ((uintptr_t)p & 7)
disk_dentry->length = cpu_to_le64(p - orig_p);
- /* Streams */
-
- if (unlikely(inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED)) {
- const struct wim_inode_stream *efs_strm;
- const u8 *efs_hash;
-
- efs_strm = inode_get_unnamed_stream(inode, STREAM_TYPE_EFSRPC_RAW_DATA);
- efs_hash = efs_strm ? stream_hash(efs_strm) : zero_hash;
- copy_hash(disk_dentry->default_hash, efs_hash);
- disk_dentry->num_extra_streams = cpu_to_le16(0);
- } else {
- /*
- * Extra stream entries:
- *
- * - Use one extra stream entry for each named data stream
- * - Use one extra stream entry for the unnamed data stream when there is either:
- * - a reparse point stream
- * - at least one named data stream (for Windows PE bug workaround)
- * - Use one extra stream entry for the reparse point stream if there is one
- */
- bool have_named_data_stream = false;
- bool have_reparse_point_stream = false;
- const u8 *unnamed_data_stream_hash = zero_hash;
- const u8 *reparse_point_hash;
- for (unsigned i = 0; i < inode->i_num_streams; i++) {
- const struct wim_inode_stream *strm = &inode->i_streams[i];
- if (strm->stream_type == STREAM_TYPE_DATA) {
- if (stream_is_named(strm))
- have_named_data_stream = true;
- else
- unnamed_data_stream_hash = stream_hash(strm);
- } else if (strm->stream_type == STREAM_TYPE_REPARSE_POINT) {
- have_reparse_point_stream = true;
- reparse_point_hash = stream_hash(strm);
- }
- }
-
- if (unlikely(have_reparse_point_stream || have_named_data_stream)) {
-
- unsigned num_extra_streams = 0;
-
- copy_hash(disk_dentry->default_hash, zero_hash);
-
- if (have_reparse_point_stream) {
- p = write_extra_stream_entry(p, NO_STREAM_NAME,
- reparse_point_hash);
- num_extra_streams++;
- }
-
- p = write_extra_stream_entry(p, NO_STREAM_NAME,
- unnamed_data_stream_hash);
- num_extra_streams++;
-
- for (unsigned i = 0; i < inode->i_num_streams; i++) {
- const struct wim_inode_stream *strm = &inode->i_streams[i];
- if (stream_is_named_data_stream(strm)) {
- p = write_extra_stream_entry(p, strm->stream_name,
- stream_hash(strm));
- num_extra_streams++;
- }
- }
- wimlib_assert(num_extra_streams <= 0xFFFF);
-
- disk_dentry->num_extra_streams = cpu_to_le16(num_extra_streams);
- } else {
- copy_hash(disk_dentry->default_hash, unnamed_data_stream_hash);
- disk_dentry->num_extra_streams = cpu_to_le16(0);
- }
- }
-
- return p;
+ /*
+ * Set disk_dentry->main_hash and disk_dentry->num_extra_streams,
+ * and write any extra stream entries that are needed.
+ */
+ return write_dentry_streams(inode, disk_dentry, p);
}
static int
git://wimlib.net / wimlib / blobdiff