#include "wimlib/lookup_table.h"
#include "wimlib/metadata.h"
#include "wimlib/resource.h"
+#include "wimlib/security.h"
#include "wimlib/sha1.h"
#include "wimlib/timestamp.h"
utf16lechar stream_name[];
} _packed_attribute;
-/* WIM directory entry (on-disk format) */
+#define WIM_ADS_ENTRY_DISK_SIZE 38
+
+/* On-disk format of a WIM dentry (directory entry), located in the metadata
+ * resource for a WIM image. */
struct wim_dentry_on_disk {
+
+ /* Length of this directory entry in bytes, not including any alternate
+ * data stream entries. Should be a multiple of 8 so that the following
+ * dentry or alternate data stream entry is aligned on an 8-byte
+ * boundary. (If not, wimlib will round it up.) It must be at least as
+ * long as the fixed-length fields of the dentry (WIM_DENTRY_DISK_SIZE),
+ * plus the lengths of the file name and/or short name if present.
+ *
+ * It is also possible for this field to be 0. This situation, which is
+ * undocumented, indicates the end of a list of sibling nodes in a
+ * directory. It also means the real length is 8, because the dentry
+ * included only the length field, but that takes up 8 bytes. */
le64 length;
+
+ /* Attributes of the file or directory. This is a bitwise OR of the
+ * FILE_ATTRIBUTE_* constants and should correspond to the value
+ * retrieved by GetFileAttributes() on Windows. */
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;
+
+ /* Offset, in bytes, from the start of the uncompressed metadata
+ * resource of this directory's child directory entries, or 0 if this
+ * directory entry does not correspond to a directory or otherwise does
+ * not have any children. */
le64 subdir_offset;
+
+ /* Reserved fields */
le64 unused_1;
le64 unused_2;
+
+
+ /* Creation time, last access time, and last write time, in
+ * 100-nanosecond intervals since 12:00 a.m UTC January 1, 1601. They
+ * should correspond to the times gotten by calling GetFileTime() on
+ * Windows. */
le64 creation_time;
le64 last_access_time;
le64 last_write_time;
+
+ /* Vaguely, the SHA-1 message digest ("hash") of the file's contents.
+ * More specifically, this is for the "unnamed data stream" rather than
+ * any "alternate data streams". This hash value is used to look up the
+ * corresponding entry in the WIM's stream lookup table to actually find
+ * the file contents within the WIM.
+ *
+ * If the file has no unnamed data stream (e.g. is a directory), then
+ * this field will be all zeroes. If the unnamed data stream is empty
+ * (i.e. an "empty file"), then this field is also expected to be all
+ * zeroes. (It will be if wimlib created the WIM image, at least;
+ * otherwise it can't be ruled out that the SHA-1 message digest of 0
+ * bytes of data is given explicitly.)
+ *
+ * If the file has reparse data, then this field will instead specify
+ * the SHA-1 message digest of the reparse data. If it is somehow
+ * possible for a file to have both an unnamed data stream and reparse
+ * data, then this is not handled by wimlib.
+ *
+ * As a further special case, if this field is all zeroes but there is
+ * an alternate data stream entry with no name and a nonzero SHA-1
+ * message digest field, then that hash must be used instead of this
+ * one. (wimlib does not use this quirk on WIM images it creates.)
+ */
u8 unnamed_stream_hash[SHA1_HASH_SIZE];
+
+ /* The format of the following data is not yet completely known and they
+ * do not correspond to Microsoft's documentation.
+ *
+ * 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.
+ *
+ * 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. We have to
+ * handle this by using special code to use distinguishing features
+ * (which is possible because some information about the underlying
+ * inode is repeated in each dentry) to split up these fake hard link
+ * groups into what they actually are supposed to be.
+ */
union {
struct {
le32 rp_unknown_1;
le64 hard_link_group_id;
} _packed_attribute nonreparse;
};
+
+ /* Number of alternate data stream entries that directly follow this
+ * dentry on-disk. */
le16 num_alternate_data_streams;
+
+ /* Length of this file's UTF-16LE encoded short name (8.3 DOS-compatible
+ * name), if present, in bytes, excluding the null terminator. If this
+ * file has no short name, then this field should be 0. */
le16 short_name_nbytes;
+
+ /* Length of this file's UTF-16LE encoded "long" name, excluding the
+ * null terminator. If this file has no short name, then this field
+ * should be 0. It's expected that only the root dentry has this field
+ * set to 0. */
le16 file_name_nbytes;
- /* Follewed by variable length file name, if file_name_nbytes != 0 */
- utf16lechar file_name[];
+ /* Followed by variable length file name, in UTF16-LE, if
+ * file_name_nbytes != 0. Includes null terminator. */
+ /*utf16lechar file_name[];*/
- /* Followed by variable length short name, if short_name_nbytes != 0 */
+ /* Followed by variable length short name, in UTF16-LE, if
+ * short_name_nbytes != 0. Includes null terminator. */
/*utf16lechar short_name[];*/
} _packed_attribute;
+#define WIM_DENTRY_DISK_SIZE 102
+
/* Calculates the unaligned length, in bytes, of an on-disk WIM dentry that has
* a file name and short name that take the specified numbers of bytes. This
* excludes any alternate data stream entries that may follow the dentry. */
dentry->short_name_nbytes);
}
-/* Return %true iff the alternate data stream entry @entry has the UTF-16LE
- * stream name @name that has length @name_nbytes bytes. */
-static inline bool
-ads_entry_has_name(const struct wim_ads_entry *entry,
- const utf16lechar *name, size_t name_nbytes)
-{
- return entry->stream_name_nbytes == name_nbytes &&
- memcmp(entry->stream_name, name, name_nbytes) == 0;
-}
-
/* Duplicates a string of system-dependent encoding into a UTF-16LE string and
* returns the string and its length, in bytes, in the pointer arguments. Frees
* any existing string at the return location before overwriting it. */
/* Calculate the full path of @dentry. The full path of its parent must have
* already been calculated, or it must be the root dentry. */
-static int
+int
calculate_dentry_full_path(struct wim_dentry *dentry)
{
tchar *full_path;
return 0;
if (dentry_is_root(dentry)) {
- full_path = TSTRDUP(T("/"));
+ static const tchar _root_path[] = {WIM_PATH_SEPARATOR, T('\0')};
+ full_path = TSTRDUP(_root_path);
if (!full_path)
return WIMLIB_ERR_NOMEM;
full_path_nbytes = 1 * sizeof(tchar);
if (!full_path)
return WIMLIB_ERR_NOMEM;
memcpy(full_path, parent_full_path, parent_full_path_nbytes);
- full_path[parent_full_path_nbytes / sizeof(tchar)] = T('/');
+ full_path[parent_full_path_nbytes / sizeof(tchar)] = WIM_PATH_SEPARATOR;
#if TCHAR_IS_UTF16LE
memcpy(&full_path[parent_full_path_nbytes / sizeof(tchar) + 1],
dentry->file_name,
}
}
+/* Case-sensitive UTF-16LE dentry or stream name comparison. Used on both UNIX
+ * (always) and Windows (sometimes) */
static int
-compare_utf16le_names(const utf16lechar *name1, size_t nbytes1,
- const utf16lechar *name2, size_t nbytes2)
+compare_utf16le_names_case_sensitive(const utf16lechar *name1, size_t nbytes1,
+ const utf16lechar *name2, size_t nbytes2)
{
+ /* Return the result if the strings differ up to their minimum length.
+ * Note that we cannot use strcmp() or strncmp() here, as the strings
+ * are in UTF-16LE format. */
int result = memcmp(name1, name2, min(nbytes1, nbytes2));
if (result)
return result;
+
+ /* The strings are the same up to their minimum length, so return a
+ * result based on their lengths. */
+ if (nbytes1 < nbytes2)
+ return -1;
+ else if (nbytes1 > nbytes2)
+ return 1;
+ else
+ return 0;
+}
+
+#ifdef __WIN32__
+/* Windoze: Case-insensitive UTF-16LE dentry or stream name comparison */
+static int
+compare_utf16le_names_case_insensitive(const utf16lechar *name1, size_t nbytes1,
+ const utf16lechar *name2, size_t nbytes2)
+{
+ /* Return the result if the strings differ up to their minimum length.
+ * */
+ int result = _wcsnicmp((const wchar_t*)name1, (const wchar_t*)name2,
+ min(nbytes1 / 2, nbytes2 / 2));
+ if (result)
+ return result;
+
+ /* The strings are the same up to their minimum length, so return a
+ * result based on their lengths. */
+ if (nbytes1 < nbytes2)
+ return -1;
+ else if (nbytes1 > nbytes2)
+ return 1;
else
- return (int)nbytes1 - (int)nbytes2;
+ return 0;
}
+#endif /* __WIN32__ */
+
+#ifdef __WIN32__
+# define compare_utf16le_names compare_utf16le_names_case_insensitive
+#else
+# define compare_utf16le_names compare_utf16le_names_case_sensitive
+#endif
+
+#ifdef __WIN32__
static int
-dentry_compare_names(const struct wim_dentry *d1, const struct wim_dentry *d2)
+dentry_compare_names_case_insensitive(const struct wim_dentry *d1,
+ const struct wim_dentry *d2)
{
- return compare_utf16le_names(d1->file_name, d1->file_name_nbytes,
- d2->file_name, d2->file_name_nbytes);
+ return compare_utf16le_names_case_insensitive(d1->file_name,
+ d1->file_name_nbytes,
+ d2->file_name,
+ d2->file_name_nbytes);
}
+#endif /* __WIN32__ */
+static int
+dentry_compare_names_case_sensitive(const struct wim_dentry *d1,
+ const struct wim_dentry *d2)
+{
+ return compare_utf16le_names_case_sensitive(d1->file_name,
+ d1->file_name_nbytes,
+ d2->file_name,
+ d2->file_name_nbytes);
+}
+#ifdef __WIN32__
+# define dentry_compare_names dentry_compare_names_case_insensitive
+#else
+# define dentry_compare_names dentry_compare_names_case_sensitive
+#endif
+
+/* Return %true iff the alternate data stream entry @entry has the UTF-16LE
+ * stream name @name that has length @name_nbytes bytes. */
+static inline bool
+ads_entry_has_name(const struct wim_ads_entry *entry,
+ const utf16lechar *name, size_t name_nbytes)
+{
+ return !compare_utf16le_names(name, name_nbytes,
+ entry->stream_name,
+ entry->stream_name_nbytes);
+}
+
+/* Given a UTF-16LE filename and a directory, look up the dentry for the file.
+ * Return it if found, otherwise NULL. This is case-sensitive on UNIX and
+ * case-insensitive on Windows. */
struct wim_dentry *
get_dentry_child_with_utf16le_name(const struct wim_dentry *dentry,
const utf16lechar *name,
size_t name_nbytes)
{
- struct rb_node *node = dentry->d_inode->i_children.rb_node;
+ struct rb_node *node;
+
+#ifdef __WIN32__
+ node = dentry->d_inode->i_children_case_insensitive.rb_node;
+#else
+ node = dentry->d_inode->i_children.rb_node;
+#endif
+
struct wim_dentry *child;
while (node) {
+ #ifdef __WIN32__
+ child = rb_entry(node, struct wim_dentry, rb_node_case_insensitive);
+ #else
child = rbnode_dentry(node);
+ #endif
int result = compare_utf16le_names(name, name_nbytes,
child->file_name,
child->file_name_nbytes);
node = node->rb_left;
else if (result > 0)
node = node->rb_right;
- else
+ else {
+ #ifdef __WIN32__
+ if (!list_empty(&child->case_insensitive_conflict_list))
+ {
+ WARNING("Result of case-insensitive lookup is ambiguous "
+ "(returning \"%ls\" instead of \"%ls\")",
+ child->file_name,
+ container_of(child->case_insensitive_conflict_list.next,
+ struct wim_dentry,
+ case_insensitive_conflict_list)->file_name);
+ }
+ #endif
return child;
+ }
}
return NULL;
}
}
p = path;
while (1) {
- while (*p == cpu_to_le16('/'))
+ while (*p == cpu_to_le16(WIM_PATH_SEPARATOR))
p++;
if (*p == cpu_to_le16('\0'))
break;
pp = p;
- while (*pp != cpu_to_le16('/') && *pp != cpu_to_le16('\0'))
+ while (*pp != cpu_to_le16(WIM_PATH_SEPARATOR) &&
+ *pp != cpu_to_le16('\0'))
pp++;
cur_dentry = get_dentry_child_with_utf16le_name(parent_dentry, p,
return cur_dentry;
}
-/* Returns the dentry corresponding to the @path, or NULL if there is no such
- * dentry. */
+/*
+ * Returns the dentry in the currently selected WIM image named by @path
+ * starting from the root of the WIM image, or NULL if there is no such dentry.
+ *
+ * On Windows, the search is done case-insensitively.
+ */
struct wim_dentry *
get_dentry(WIMStruct *wim, const tchar *path)
{
to_parent_name(tchar *buf, size_t len)
{
ssize_t i = (ssize_t)len - 1;
- while (i >= 0 && buf[i] == T('/'))
+ while (i >= 0 && buf[i] == WIM_PATH_SEPARATOR)
i--;
- while (i >= 0 && buf[i] != T('/'))
+ while (i >= 0 && buf[i] != WIM_PATH_SEPARATOR)
i--;
- while (i >= 0 && buf[i] == T('/'))
+ while (i >= 0 && buf[i] == WIM_PATH_SEPARATOR)
i--;
buf[i + 1] = T('\0');
}
inode->i_next_stream_id = 1;
inode->i_not_rpfixed = 1;
INIT_LIST_HEAD(&inode->i_list);
- #ifdef WITH_FUSE
- if (pthread_mutex_init(&inode->i_mutex, NULL) != 0) {
- ERROR_WITH_ERRNO("Error initializing mutex");
- FREE(inode);
- return NULL;
- }
- #endif
INIT_LIST_HEAD(&inode->i_dentry);
}
return inode;
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);
+}
+
static int
init_ads_entry(struct wim_ads_entry *ads_entry, const void *name,
size_t name_nbytes, bool is_utf16le)
destroy_ads_entry(&inode->i_ads_entries[i]);
FREE(inode->i_ads_entries);
}
- #ifdef WITH_FUSE
- wimlib_assert(inode->i_num_opened_fds == 0);
- FREE(inode->i_fds);
- pthread_mutex_destroy(&inode->i_mutex);
- #endif
/* HACK: This may instead delete the inode from i_list, but the
* hlist_del() behaves the same as list_del(). */
- hlist_del(&inode->i_hlist);
- FREE(inode->i_extracted_file);
+ if (!hlist_unhashed(&inode->i_hlist))
+ hlist_del(&inode->i_hlist);
FREE(inode);
}
}
/*
* Unlinks and frees a dentry tree.
*
- * @root: The root of the tree.
- * @lookup_table: The lookup table for dentries. If non-NULL, the
- * reference counts in the lookup table for the lookup
- * table entries corresponding to the dentries will be
- * decremented.
+ * @root:
+ * The root of the tree.
+ *
+ * @lookup_table:
+ * The lookup table for dentries. If non-NULL, the reference counts in the
+ * lookup table for the lookup table entries corresponding to the dentries
+ * will be decremented.
*/
void
free_dentry_tree(struct wim_dentry *root, struct wim_lookup_table *lookup_table)
for_dentry_in_tree_depth(root, do_free_dentry, lookup_table);
}
+#ifdef __WIN32__
+
+/* Insert a dentry into the case insensitive index for a directory.
+ *
+ * This is a red-black tree, but when multiple dentries share the same
+ * case-insensitive name, only one is inserted into the tree itself; the rest
+ * are connected in a list.
+ */
+static struct wim_dentry *
+dentry_add_child_case_insensitive(struct wim_dentry *parent,
+ struct wim_dentry *child)
+{
+ struct rb_root *root;
+ struct rb_node **new;
+ struct rb_node *rb_parent;
+
+ root = &parent->d_inode->i_children_case_insensitive;
+ new = &root->rb_node;
+ rb_parent = NULL;
+ while (*new) {
+ struct wim_dentry *this = container_of(*new, struct wim_dentry,
+ rb_node_case_insensitive);
+ int result = dentry_compare_names_case_insensitive(child, this);
+
+ rb_parent = *new;
+
+ if (result < 0)
+ new = &((*new)->rb_left);
+ else if (result > 0)
+ new = &((*new)->rb_right);
+ else
+ return this;
+ }
+ rb_link_node(&child->rb_node_case_insensitive, rb_parent, new);
+ rb_insert_color(&child->rb_node_case_insensitive, root);
+ return NULL;
+}
+#endif
+
/*
* Links a dentry into the directory tree.
*
* @parent: The dentry that will be the parent of @child.
* @child: The dentry to link.
*
- * Returns non-NULL if a duplicate dentry was detected.
+ * Returns NULL if successful. If @parent already contains a dentry with the
+ * same case-sensitive name as @child, the pointer to this duplicate dentry is
+ * returned.
*/
struct wim_dentry *
dentry_add_child(struct wim_dentry * restrict parent,
struct wim_dentry * restrict child)
{
- wimlib_assert(dentry_is_directory(parent));
+ struct rb_root *root;
+ struct rb_node **new;
+ struct rb_node *rb_parent;
- struct rb_root *root = &parent->d_inode->i_children;
- struct rb_node **new = &(root->rb_node);
- struct rb_node *rb_parent = NULL;
+ wimlib_assert(dentry_is_directory(parent));
+ wimlib_assert(parent != child);
+ /* Case sensitive child dentry index */
+ root = &parent->d_inode->i_children;
+ new = &root->rb_node;
+ rb_parent = NULL;
while (*new) {
struct wim_dentry *this = rbnode_dentry(*new);
- int result = dentry_compare_names(child, this);
+ int result = dentry_compare_names_case_sensitive(child, this);
rb_parent = *new;
child->parent = parent;
rb_link_node(&child->rb_node, rb_parent, new);
rb_insert_color(&child->rb_node, root);
+
+#ifdef __WIN32__
+ {
+ struct wim_dentry *existing;
+ existing = dentry_add_child_case_insensitive(parent, child);
+ if (existing) {
+ list_add(&child->case_insensitive_conflict_list,
+ &existing->case_insensitive_conflict_list);
+ child->rb_node_case_insensitive.__rb_parent_color = 0;
+ } else {
+ INIT_LIST_HEAD(&child->case_insensitive_conflict_list);
+ }
+ }
+#endif
return NULL;
}
void
unlink_dentry(struct wim_dentry *dentry)
{
- if (!dentry_is_root(dentry))
- rb_erase(&dentry->rb_node, &dentry->parent->d_inode->i_children);
+ struct wim_dentry *parent = dentry->parent;
+
+ if (parent == dentry)
+ return;
+ rb_erase(&dentry->rb_node, &parent->d_inode->i_children);
+#ifdef __WIN32__
+ if (dentry->rb_node_case_insensitive.__rb_parent_color) {
+ /* This dentry was in the case-insensitive red-black tree. */
+ rb_erase(&dentry->rb_node_case_insensitive,
+ &parent->d_inode->i_children_case_insensitive);
+ if (!list_empty(&dentry->case_insensitive_conflict_list)) {
+ /* Make a different case-insensitively-the-same dentry
+ * be the "representative" in the red-black tree. */
+ struct list_head *next;
+ struct wim_dentry *other;
+ struct wim_dentry *existing;
+
+ next = dentry->case_insensitive_conflict_list.next;
+ other = list_entry(next, struct wim_dentry, case_insensitive_conflict_list);
+ existing = dentry_add_child_case_insensitive(parent, other);
+ wimlib_assert(existing == NULL);
+ }
+ }
+ list_del(&dentry->case_insensitive_conflict_list);
+#endif
}
/*
struct wim_lookup_table_entry *lte, *existing_lte;
sha1_buffer(buffer, size, hash);
- existing_lte = __lookup_resource(lookup_table, hash);
+ existing_lte = lookup_resource(lookup_table, hash);
if (existing_lte) {
wimlib_assert(wim_resource_size(existing_lte) == size);
lte = existing_lte;
lte = new_lookup_table_entry();
if (!lte)
return NULL;
- buffer_copy = MALLOC(size);
+ buffer_copy = memdup(buffer, size);
if (!buffer_copy) {
free_lookup_table_entry(lte);
return NULL;
}
- memcpy(buffer_copy, buffer, size);
lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER;
lte->attached_buffer = buffer_copy;
lte->resource_entry.original_size = size;
return 0;
}
+bool
+inode_has_named_stream(const struct wim_inode *inode)
+{
+ for (u16 i = 0; i < inode->i_num_ads; i++)
+ if (ads_entry_is_named_stream(&inode->i_ads_entries[i]))
+ return true;
+ return false;
+}
+
/* Set the unnamed stream of a WIM inode, given a data buffer containing the
* stream contents. */
int
inode->i_num_ads--;
}
+bool
+inode_has_unix_data(const struct wim_inode *inode)
+{
+ for (u16 i = 0; i < inode->i_num_ads; i++)
+ if (ads_entry_is_unix_data(&inode->i_ads_entries[i]))
+ return true;
+ return false;
+}
+
#ifndef __WIN32__
int
inode_get_unix_data(const struct wim_inode *inode,
}
#endif /* !__WIN32__ */
-/* Replace weird characters in filenames and alternate data stream names.
- *
- * In particular we do not want the path separator to appear in any names, as
- * that would make it possible for a "malicious" WIM to extract itself to any
- * location it wanted to. */
-static void
-replace_forbidden_characters(utf16lechar *name)
-{
- utf16lechar *p;
-
- for (p = name; *p; p++) {
- #ifdef __WIN32__
- if (wcschr(L"<>:\"/\\|?*", (wchar_t)*p))
- #else
- if (*p == cpu_to_le16('/'))
- #endif
- {
- if (name) {
- WARNING("File, directory, or stream name \"%"WS"\"\n"
- " contains forbidden characters; "
- "substituting replacement characters.",
- name);
- name = NULL;
- }
- #ifdef __WIN32__
- *p = cpu_to_le16(0xfffd);
- #else
- *p = cpu_to_le16('?');
- #endif
- }
- }
-}
-
/*
* Reads the alternate data stream entries of a WIM dentry.
*
- * @p: Pointer to buffer that starts with the first alternate stream entry.
+ * @p:
+ * Pointer to buffer that starts with the first alternate stream entry.
*
- * @inode: Inode to load the alternate data streams into.
- * @inode->i_num_ads must have been set to the number of
- * alternate data streams that are expected.
+ * @inode:
+ * Inode to load the alternate data streams into. @inode->i_num_ads must
+ * have been set to the number of alternate data streams that are expected.
*
- * @remaining_size: Number of bytes of data remaining in the buffer pointed
- * to by @p.
+ * @remaining_size:
+ * Number of bytes of data remaining in the buffer pointed to by @p.
*
+ * On success, inode->i_ads_entries is set to an array of `struct
+ * wim_ads_entry's of length inode->i_num_ads. On failure, @inode is not
+ * modified.
*
- * Return 0 on success or nonzero on failure. On success, inode->i_ads_entries
- * is set to an array of `struct wim_ads_entry's of length inode->i_num_ads. On
- * failure, @inode is not modified.
+ * Return values:
+ * WIMLIB_ERR_SUCCESS (0)
+ * WIMLIB_ERR_INVALID_METADATA_RESOURCE
+ * WIMLIB_ERR_NOMEM
*/
static int
read_ads_entries(const u8 * restrict p, struct wim_inode * restrict inode,
struct wim_ads_entry *ads_entries;
int ret;
+ BUILD_BUG_ON(sizeof(struct wim_ads_entry_on_disk) != WIM_ADS_ENTRY_DISK_SIZE);
+
+ /* Allocate an array for our in-memory representation of the alternate
+ * data stream entries. */
num_ads = inode->i_num_ads;
ads_entries = CALLOC(num_ads, sizeof(inode->i_ads_entries[0]));
if (!ads_entries)
goto out_of_memory;
+ /* Read the entries into our newly allocated buffer. */
for (u16 i = 0; i < num_ads; i++) {
u64 length;
struct wim_ads_entry *cur_entry;
length = le64_to_cpu(disk_entry->length);
/* Make sure the length field is neither so small it doesn't
- * include all the fixed-length data, or so large it overflows
+ * include all the fixed-length data nor so large it overflows
* the metadata resource buffer. */
if (length < sizeof(struct wim_ads_entry_on_disk) ||
length > nbytes_remaining)
/* Read the rest of the fixed-length data. */
cur_entry->reserved = le64_to_cpu(disk_entry->reserved);
-
copy_hash(cur_entry->hash, disk_entry->hash);
- cur_entry->stream_name_nbytes = le16_to_cpu(cur_entry->stream_name_nbytes);
+ cur_entry->stream_name_nbytes = le16_to_cpu(disk_entry->stream_name_nbytes);
/* If stream_name_nbytes != 0, this is a named stream.
* Otherwise this is an unnamed stream, or in some cases (bugs
* the fact that the real unnamed stream entry has a nonzero
* hash field. */
if (cur_entry->stream_name_nbytes) {
- u64 length_no_padding;
-
/* The name is encoded in UTF16-LE, which uses 2-byte
* coding units, so the length of the name had better be
* an even number of bytes... */
memcpy(cur_entry->stream_name,
disk_entry->stream_name,
cur_entry->stream_name_nbytes);
- cur_entry->stream_name[cur_entry->stream_name_nbytes / 2] = 0;
- replace_forbidden_characters(cur_entry->stream_name);
+ cur_entry->stream_name[cur_entry->stream_name_nbytes / 2] = cpu_to_le16(0);
}
/* It's expected that the size of every ADS entry is a multiple
nbytes_remaining -= length;
}
inode->i_ads_entries = ads_entries;
-#ifdef WITH_FUSE
inode->i_next_stream_id = inode->i_num_ads + 1;
-#endif
ret = 0;
goto out;
out_of_memory:
goto out_free_ads_entries;
out_invalid:
ERROR("An alternate data stream entry is invalid");
- ret = WIMLIB_ERR_INVALID_DENTRY;
+ ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE;
out_free_ads_entries:
if (ads_entries) {
for (u16 i = 0; i < num_ads; i++)
* @metadata_resource_len:
* Length of the metadata resource buffer, in bytes.
*
- * @offset: Offset of the dentry within the metadata resource.
+ * @offset: Offset of the dentry within the metadata resource.
*
* @dentry: A `struct wim_dentry' that will be filled in by this function.
*
* buffers. On success, the dentry->length field must be examined. If zero,
* this was a special "end of directory" dentry and not a real dentry. If
* nonzero, this was a real dentry.
+ *
+ * Return values:
+ * WIMLIB_ERR_SUCCESS (0)
+ * WIMLIB_ERR_INVALID_METADATA_RESOURCE
+ * WIMLIB_ERR_NOMEM
*/
int
read_dentry(const u8 * restrict metadata_resource, u64 metadata_resource_len,
u16 file_name_nbytes;
int ret;
struct wim_inode *inode;
- const struct wim_dentry_on_disk *disk_dentry;
const u8 *p = &metadata_resource[offset];
+ const struct wim_dentry_on_disk *disk_dentry =
+ (const struct wim_dentry_on_disk*)p;
+
+ BUILD_BUG_ON(sizeof(struct wim_dentry_on_disk) != WIM_DENTRY_DISK_SIZE);
if ((uintptr_t)p & 7)
WARNING("WIM dentry is not 8-byte aligned");
dentry_common_init(dentry);
- /*Make sure the dentry really fits into the metadata resource.*/
- if (offset + sizeof(u64) < offset ||
- offset + sizeof(u64) > metadata_resource_len)
+ /* 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)
+ * terminates the list of sibling directory entries. */
+ if (offset + sizeof(u64) > metadata_resource_len ||
+ offset + sizeof(u64) < offset)
{
ERROR("Directory entry starting at %"PRIu64" ends past the "
"end of the metadata resource (size %"PRIu64")",
offset, metadata_resource_len);
- return WIMLIB_ERR_INVALID_DENTRY;
+ return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
}
-
- disk_dentry = (const struct wim_dentry_on_disk*)p;
-
- /* 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)
- * terminates the list of sibling directory entries. */
dentry->length = le64_to_cpu(disk_dentry->length);
/* A zero length field (really a length of 8, since that's how big the
* directory entry is...) indicates that this is the end of directory
* dentry. We do not read it into memory as an actual dentry, so just
- * return successfully in that case. */
+ * return successfully in this case. */
if (dentry->length == 8)
dentry->length = 0;
if (dentry->length == 0)
return 0;
/* Now that we have the actual length provided in the on-disk structure,
- * make sure it doesn't overflow the metadata buffer. */
- if (offset + dentry->length >= metadata_resource_len
- || offset + dentry->length < offset)
+ * again make sure it doesn't overflow the metadata resource buffer. */
+ if (offset + dentry->length > metadata_resource_len ||
+ offset + dentry->length < offset)
{
ERROR("Directory entry at offset %"PRIu64" and with size "
"%"PRIu64" ends past the end of the metadata resource "
"(size %"PRIu64")",
offset, dentry->length, metadata_resource_len);
- return WIMLIB_ERR_INVALID_DENTRY;
+ return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
}
/* Make sure the dentry length is at least as large as the number of
if (dentry->length < sizeof(struct wim_dentry_on_disk)) {
ERROR("Directory entry has invalid length of %"PRIu64" bytes",
dentry->length);
- return WIMLIB_ERR_INVALID_DENTRY;
+ return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
}
/* Allocate a `struct wim_inode' for this `struct wim_dentry'. */
if ((short_name_nbytes & 1) | (file_name_nbytes & 1))
{
ERROR("Dentry name is not valid UTF-16LE (odd number of bytes)!");
- ret = WIMLIB_ERR_INVALID_DENTRY;
+ ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE;
goto out_free_inode;
}
ERROR("Unexpected end of directory entry! (Expected "
"at least %"PRIu64" bytes, got %"PRIu64" bytes.)",
calculated_size, dentry->length);
- ret = WIMLIB_ERR_INVALID_DENTRY;
+ ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE;
goto out_free_inode;
}
memcpy(file_name, p, file_name_nbytes);
p += file_name_nbytes + 2;
file_name[file_name_nbytes / 2] = cpu_to_le16(0);
- replace_forbidden_characters(file_name);
} else {
file_name = NULL;
}
memcpy(short_name, p, short_name_nbytes);
p += short_name_nbytes + 2;
short_name[short_name_nbytes / 2] = cpu_to_le16(0);
- replace_forbidden_characters(short_name);
} else {
short_name = NULL;
}
* be included in the dentry->length field for some reason.
*/
if (inode->i_num_ads != 0) {
- ret = WIMLIB_ERR_INVALID_DENTRY;
+ ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE;
if (offset + dentry->length > metadata_resource_len ||
(ret = read_ads_entries(&metadata_resource[offset + dentry->length],
inode,
goto out_free_short_name;
}
}
-out_success:
/* We've read all the data for this dentry. Set the names and their
* lengths, and we've done. */
dentry->d_inode = inode;
return ret;
}
+static const tchar *
+dentry_get_file_type_string(const struct wim_dentry *dentry)
+{
+ const struct wim_inode *inode = dentry->d_inode;
+ if (inode_is_directory(inode))
+ return T("directory");
+ else if (inode_is_symlink(inode))
+ return T("symbolic link");
+ else
+ return T("file");
+}
+
/* Reads the children of a dentry, and all their children, ..., etc. from the
* metadata resource and into the dentry tree.
*
- * @metadata_resource: An array that contains the uncompressed metadata
- * resource for the WIM file.
+ * @metadata_resource:
+ * An array that contains the uncompressed metadata resource for the WIM
+ * file.
*
- * @metadata_resource_len: The length of the uncompressed metadata resource, in
- * bytes.
+ * @metadata_resource_len:
+ * The length of the uncompressed metadata resource, in bytes.
*
- * @dentry: A pointer to a `struct wim_dentry' that is the root of the directory
- * tree and has already been read from the metadata resource. It
- * does not need to be the real root because this procedure is
- * called recursively.
+ * @dentry:
+ * A pointer to a `struct wim_dentry' that is the root of the directory
+ * tree and has already been read from the metadata resource. It does not
+ * need to be the real root because this procedure is called recursively.
*
- * Returns zero on success; nonzero on failure.
+ * Return values:
+ * WIMLIB_ERR_SUCCESS (0)
+ * WIMLIB_ERR_INVALID_METADATA_RESOURCE
+ * WIMLIB_ERR_NOMEM
*/
int
-read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len,
- struct wim_dentry *dentry)
+read_dentry_tree(const u8 * restrict metadata_resource,
+ u64 metadata_resource_len,
+ struct wim_dentry * restrict dentry)
{
u64 cur_offset = dentry->subdir_offset;
struct wim_dentry *child;
+ struct wim_dentry *duplicate;
+ struct wim_dentry *parent;
struct wim_dentry cur_child;
int ret;
if (cur_offset == 0)
return 0;
+ /* Check for cyclic directory structure */
+ for (parent = dentry->parent; !dentry_is_root(parent); parent = parent->parent)
+ {
+ if (unlikely(parent->subdir_offset == cur_offset)) {
+ ERROR("Cyclic directory structure directed: children "
+ "of \"%"TS"\" coincide with children of \"%"TS"\"",
+ dentry_full_path(dentry),
+ dentry_full_path(parent));
+ return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
+ }
+ }
+
/* Find and read all the children of @dentry. */
- while (1) {
+ for (;;) {
/* Read next child of @dentry into @cur_child. */
ret = read_dentry(metadata_resource, metadata_resource_len,
cur_offset, &cur_child);
- if (ret != 0)
+ if (ret)
break;
/* Check for end of directory. */
/* Not end of directory. Allocate this child permanently and
* link it to the parent and previous child. */
- child = MALLOC(sizeof(struct wim_dentry));
+ child = memdup(&cur_child, sizeof(struct wim_dentry));
if (!child) {
- ERROR("Failed to allocate %zu bytes for new dentry",
- sizeof(struct wim_dentry));
+ ERROR("Failed to allocate new dentry!");
ret = WIMLIB_ERR_NOMEM;
break;
}
- memcpy(child, &cur_child, sizeof(struct wim_dentry));
- dentry_add_child(dentry, child);
- inode_add_dentry(child, child->d_inode);
-
- /* If there are children of this child, call this procedure
- * recursively. */
- if (child->subdir_offset != 0) {
- ret = read_dentry_tree(metadata_resource,
- metadata_resource_len, child);
- if (ret != 0)
- break;
- }
/* Advance to the offset of the next child. Note: We need to
* advance by the TOTAL length of the dentry, not by the length
- * child->length, which although it does take into account the
- * padding, it DOES NOT take into account alternate stream
+ * cur_child.length, which although it does take into account
+ * the padding, it DOES NOT take into account alternate stream
* entries. */
cur_offset += dentry_total_length(child);
+
+ if (unlikely(!dentry_has_long_name(child))) {
+ WARNING("Ignoring unnamed dentry in "
+ "directory \"%"TS"\"",
+ dentry_full_path(dentry));
+ free_dentry(child);
+ continue;
+ }
+
+ duplicate = dentry_add_child(dentry, child);
+ if (unlikely(duplicate)) {
+ const tchar *child_type, *duplicate_type;
+ child_type = dentry_get_file_type_string(child);
+ duplicate_type = dentry_get_file_type_string(duplicate);
+ WARNING("Ignoring duplicate %"TS" \"%"TS"\" "
+ "(the WIM image already contains a %"TS" "
+ "at that path with the exact same name)",
+ child_type, dentry_full_path(duplicate),
+ duplicate_type);
+ free_dentry(child);
+ continue;
+ }
+
+ inode_add_dentry(child, child->d_inode);
+ /* If there are children of this child, call this
+ * procedure recursively. */
+ if (child->subdir_offset != 0) {
+ if (likely(dentry_is_directory(child))) {
+ ret = read_dentry_tree(metadata_resource,
+ metadata_resource_len,
+ child);
+ if (ret)
+ break;
+ } else {
+ WARNING("Ignoring children of non-directory \"%"TS"\"",
+ dentry_full_path(child));
+ }
+ }
}
return ret;
}
* @p: The memory location to write the data to.
*
* Returns the pointer to the byte after the last byte we wrote as part of the
- * dentry, including any alternate data streams entry.
+ * dentry, including any alternate data stream entries.
*/
static u8 *
write_dentry(const struct wim_dentry * restrict dentry, u8 * restrict p)
wimlib_assert(((uintptr_t)p & 7) == 0); /* 8 byte aligned */
orig_p = p;
- inode = dentry->d_inode;
+ inode = dentry->d_inode;
disk_dentry = (struct wim_dentry_on_disk*)p;
disk_dentry->attributes = cpu_to_le32(inode->i_attributes);
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);
-
- if (inode->i_resolved)
- hash = inode->i_lte->hash;
- else
- hash = inode->i_hash;
- copy_hash(disk_dentry->unnamed_stream_hash, inode_stream_hash(inode, 0));
+ hash = inode_stream_hash(inode, 0);
+ copy_hash(disk_dentry->unnamed_stream_hash, hash);
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->file_name_nbytes = cpu_to_le16(dentry->file_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, dentry->file_name_nbytes + 2);
/* We calculate the correct length of the dentry ourselves because the
* dentry->length field may been set to an unexpected value from when we
* read the dentry in (for example, there may have been unknown data
- * appended to the end of the dentry...) */
+ * appended to the end of the dentry...). Furthermore, the dentry may
+ * have been renamed, thus changing its needed length. */
disk_dentry->length = cpu_to_le64(p - orig_p);
- /* Write the alternate data streams entries, if there are any. */
- for (u16 i = 0; i < inode->i_num_ads; i++)
- {
+ /* Write the alternate data streams entries, if any. */
+ for (u16 i = 0; i < inode->i_num_ads; i++) {
const struct wim_ads_entry *ads_entry =
- &inode->i_ads_entries[i];
+ &inode->i_ads_entries[i];
struct wim_ads_entry_on_disk *disk_ads_entry =
(struct wim_ads_entry_on_disk*)p;
+ orig_p = p;
disk_ads_entry->reserved = cpu_to_le64(ads_entry->reserved);
- orig_p = p;
hash = inode_stream_hash(inode, i + 1);
copy_hash(disk_ads_entry->hash, hash);
* Returns pointer to the byte after the last byte we wrote.
*/
u8 *
-write_dentry_tree(const struct wim_dentry *root, u8 *p)
+write_dentry_tree(const struct wim_dentry * restrict root, u8 * restrict p)
{
DEBUG("Writing dentry tree.");
wimlib_assert(dentry_is_root(root));
/* Recursively write the rest of the dentry tree. */
return write_dentry_tree_recursive(root, p);
}
+
+
+static int
+init_wimlib_dentry(struct wimlib_dir_entry *wdentry,
+ struct wim_dentry *dentry,
+ const WIMStruct *wim)
+{
+ int ret;
+ size_t dummy;
+ const struct wim_inode *inode = dentry->d_inode;
+ struct wim_lookup_table_entry *lte;
+
+#if TCHAR_IS_UTF16LE
+ wdentry->filename = dentry->file_name;
+ wdentry->dos_name = dentry->short_name;
+#else
+ if (dentry_has_long_name(dentry)) {
+ ret = utf16le_to_tstr(dentry->file_name,
+ dentry->file_name_nbytes,
+ (tchar**)&wdentry->filename,
+ &dummy);
+ if (ret)
+ return ret;
+ }
+ if (dentry_has_short_name(dentry)) {
+ ret = utf16le_to_tstr(dentry->short_name,
+ dentry->short_name_nbytes,
+ (tchar**)&wdentry->dos_name,
+ &dummy);
+ if (ret)
+ return ret;
+ }
+#endif
+ ret = calculate_dentry_full_path(dentry);
+ if (ret)
+ return ret;
+ wdentry->full_path = dentry->_full_path;
+
+ for (struct wim_dentry *d = dentry; !dentry_is_root(d); d = d->parent)
+ wdentry->depth++;
+
+ if (inode->i_security_id >= 0) {
+ const struct wim_security_data *sd = wim_const_security_data(wim);
+ wdentry->security_descriptor = sd->descriptors[inode->i_security_id];
+ wdentry->security_descriptor_size = sd->sizes[inode->i_security_id];
+ }
+ wdentry->reparse_tag = inode->i_reparse_tag;
+ wdentry->num_links = inode->i_nlink;
+ wdentry->attributes = inode->i_attributes;
+ wdentry->hard_link_group_id = inode->i_ino;
+ wdentry->creation_time = wim_timestamp_to_timespec(inode->i_creation_time);
+ wdentry->last_write_time = wim_timestamp_to_timespec(inode->i_last_write_time);
+ wdentry->last_access_time = wim_timestamp_to_timespec(inode->i_last_access_time);
+
+ lte = inode_unnamed_lte(inode, wim->lookup_table);
+ if (lte)
+ lte_to_wimlib_resource_entry(lte, &wdentry->streams[0].resource);
+
+ for (unsigned i = 0; i < inode->i_num_ads; i++) {
+ if (inode->i_ads_entries[i].stream_name == NULL)
+ continue;
+ lte = inode_stream_lte(inode, i + 1, wim->lookup_table);
+ wdentry->num_named_streams++;
+ if (lte) {
+ lte_to_wimlib_resource_entry(lte, &wdentry->streams[
+ wdentry->num_named_streams].resource);
+ }
+ #if TCHAR_IS_UTF16LE
+ wdentry->streams[wdentry->num_named_streams].stream_name =
+ inode->i_ads_entries[i].stream_name;
+ #else
+ size_t dummy;
+
+ ret = utf16le_to_tstr(inode->i_ads_entries[i].stream_name,
+ inode->i_ads_entries[i].stream_name_nbytes,
+ (tchar**)&wdentry->streams[
+ wdentry->num_named_streams].stream_name,
+ &dummy);
+ if (ret)
+ return ret;
+ #endif
+ }
+ return 0;
+}
+
+static void
+free_wimlib_dentry(struct wimlib_dir_entry *wdentry)
+{
+#if !TCHAR_IS_UTF16LE
+ FREE((tchar*)wdentry->filename);
+ FREE((tchar*)wdentry->dos_name);
+ for (unsigned i = 1; i <= wdentry->num_named_streams; i++)
+ FREE((tchar*)wdentry->streams[i].stream_name);
+#endif
+ FREE(wdentry);
+}
+
+struct iterate_dir_tree_ctx {
+ WIMStruct *wim;
+ int flags;
+ wimlib_iterate_dir_tree_callback_t cb;
+ void *user_ctx;
+};
+
+static int
+do_iterate_dir_tree(WIMStruct *wim,
+ struct wim_dentry *dentry, int flags,
+ wimlib_iterate_dir_tree_callback_t cb,
+ void *user_ctx);
+
+static int
+call_do_iterate_dir_tree(struct wim_dentry *dentry, void *_ctx)
+{
+ struct iterate_dir_tree_ctx *ctx = _ctx;
+ return do_iterate_dir_tree(ctx->wim, dentry, ctx->flags,
+ ctx->cb, ctx->user_ctx);
+}
+
+static int
+do_iterate_dir_tree(WIMStruct *wim,
+ struct wim_dentry *dentry, int flags,
+ wimlib_iterate_dir_tree_callback_t cb,
+ void *user_ctx)
+{
+ struct wimlib_dir_entry *wdentry;
+ int ret = WIMLIB_ERR_NOMEM;
+
+
+ wdentry = CALLOC(1, sizeof(struct wimlib_dir_entry) +
+ (1 + dentry->d_inode->i_num_ads) *
+ sizeof(struct wimlib_stream_entry));
+ if (!wdentry)
+ goto out;
+
+ ret = init_wimlib_dentry(wdentry, dentry, wim);
+ if (ret)
+ goto out_free_wimlib_dentry;
+
+ if (!(flags & WIMLIB_ITERATE_DIR_TREE_FLAG_CHILDREN)) {
+ ret = (*cb)(wdentry, user_ctx);
+ if (ret)
+ goto out_free_wimlib_dentry;
+ }
+
+ if (flags & (WIMLIB_ITERATE_DIR_TREE_FLAG_RECURSIVE |
+ WIMLIB_ITERATE_DIR_TREE_FLAG_CHILDREN))
+ {
+ struct iterate_dir_tree_ctx ctx = {
+ .wim = wim,
+ .flags = flags &= ~WIMLIB_ITERATE_DIR_TREE_FLAG_CHILDREN,
+ .cb = cb,
+ .user_ctx = user_ctx,
+ };
+ ret = for_dentry_child(dentry, call_do_iterate_dir_tree, &ctx);
+ }
+out_free_wimlib_dentry:
+ free_wimlib_dentry(wdentry);
+out:
+ return ret;
+}
+
+struct image_iterate_dir_tree_ctx {
+ const tchar *path;
+ int flags;
+ wimlib_iterate_dir_tree_callback_t cb;
+ void *user_ctx;
+};
+
+
+static int
+image_do_iterate_dir_tree(WIMStruct *wim)
+{
+ struct image_iterate_dir_tree_ctx *ctx = wim->private;
+ struct wim_dentry *dentry;
+
+ dentry = get_dentry(wim, ctx->path);
+ if (!dentry)
+ return WIMLIB_ERR_PATH_DOES_NOT_EXIST;
+ return do_iterate_dir_tree(wim, dentry, ctx->flags, ctx->cb, ctx->user_ctx);
+}
+
+/* API function documented in wimlib.h */
+WIMLIBAPI int
+wimlib_iterate_dir_tree(WIMStruct *wim, int image, const tchar *path,
+ int flags,
+ wimlib_iterate_dir_tree_callback_t cb, void *user_ctx)
+{
+ struct image_iterate_dir_tree_ctx ctx = {
+ .path = path,
+ .flags = flags,
+ .cb = cb,
+ .user_ctx = user_ctx,
+ };
+ wim->private = &ctx;
+ return for_image(wim, image, image_do_iterate_dir_tree);
+}
+
+/* Returns %true iff the metadata of @inode and @template_inode are reasonably
+ * consistent with them being the same, unmodified file. */
+static bool
+inode_metadata_consistent(const struct wim_inode *inode,
+ const struct wim_inode *template_inode,
+ const struct wim_lookup_table *template_lookup_table)
+{
+ /* Must have exact same creation time and last write time. */
+ if (inode->i_creation_time != template_inode->i_creation_time ||
+ inode->i_last_write_time != template_inode->i_last_write_time)
+ return false;
+
+ /* Last access time may have stayed the same or increased, but certainly
+ * shouldn't have decreased. */
+ if (inode->i_last_access_time < template_inode->i_last_access_time)
+ return false;
+
+ /* Must have same number of alternate data stream entries. */
+ if (inode->i_num_ads != template_inode->i_num_ads)
+ return false;
+
+ /* If the stream entries for the inode are for some reason not resolved,
+ * then the hashes are already available and the point of this function
+ * is defeated. */
+ if (!inode->i_resolved)
+ return false;
+
+ /* Iterate through each stream and do some more checks. */
+ for (unsigned i = 0; i <= inode->i_num_ads; i++) {
+ const struct wim_lookup_table_entry *lte, *template_lte;
+
+ lte = inode_stream_lte_resolved(inode, i);
+ template_lte = inode_stream_lte(template_inode, i,
+ template_lookup_table);
+
+ /* Compare stream sizes. */
+ if (lte && template_lte) {
+ if (wim_resource_size(lte) != wim_resource_size(template_lte))
+ return false;
+
+ /* If hash happens to be available, compare with template. */
+ if (!lte->unhashed && !template_lte->unhashed &&
+ !hashes_equal(lte->hash, template_lte->hash))
+ return false;
+
+ } else if (lte && wim_resource_size(lte)) {
+ return false;
+ } else if (template_lte && wim_resource_size(template_lte)) {
+ return false;
+ }
+ }
+
+ /* All right, barring a full checksum and given that the inodes share a
+ * path and the user isn't trying to trick us, these inodes most likely
+ * refer to the same file. */
+ return true;
+}
+
+/**
+ * Given an inode @inode that has been determined to be "the same" as another
+ * inode @template_inode in either the same WIM or another WIM, retrieve some
+ * useful stream information (e.g. checksums) from @template_inode.
+ *
+ * This assumes that the streams for @inode have been resolved (to point
+ * directly to the appropriate `struct wim_lookup_table_entry's) but do not
+ * necessarily have checksum information filled in.
+ */
+static int
+inode_copy_checksums(struct wim_inode *inode,
+ struct wim_inode *template_inode,
+ WIMStruct *wim,
+ WIMStruct *template_wim)
+{
+ for (unsigned i = 0; i <= inode->i_num_ads; i++) {
+ struct wim_lookup_table_entry *lte, *template_lte;
+ struct wim_lookup_table_entry *replace_lte;
+
+ lte = inode_stream_lte_resolved(inode, i);
+ template_lte = inode_stream_lte(template_inode, i,
+ template_wim->lookup_table);
+
+ /* Only take action if both entries exist, the entry for @inode
+ * has no checksum calculated, but the entry for @template_inode
+ * does. */
+ if (!lte || !template_lte ||
+ !lte->unhashed || template_lte->unhashed)
+ continue;
+
+ wimlib_assert(lte->refcnt == inode->i_nlink);
+
+ /* If the WIM of the template image is the same as the WIM of
+ * the new image, then @template_lte can be used directly.
+ *
+ * Otherwise, look for a stream with the same hash in the WIM of
+ * the new image. If found, use it; otherwise re-use the entry
+ * being discarded, filling in the hash. */
+
+ if (wim == template_wim)
+ replace_lte = template_lte;
+ else
+ replace_lte = lookup_resource(wim->lookup_table,
+ template_lte->hash);
+
+ list_del(<e->unhashed_list);
+ if (replace_lte) {
+ free_lookup_table_entry(lte);
+ } else {
+ copy_hash(lte->hash, template_lte->hash);
+ lte->unhashed = 0;
+ lookup_table_insert(wim->lookup_table, lte);
+ lte->refcnt = 0;
+ replace_lte = lte;
+ }
+
+ if (i == 0)
+ inode->i_lte = replace_lte;
+ else
+ inode->i_ads_entries[i - 1].lte = replace_lte;
+
+ replace_lte->refcnt += inode->i_nlink;
+ }
+ return 0;
+}
+
+struct reference_template_args {
+ WIMStruct *wim;
+ WIMStruct *template_wim;
+};
+
+static int
+dentry_reference_template(struct wim_dentry *dentry, void *_args)
+{
+ int ret;
+ struct wim_dentry *template_dentry;
+ struct wim_inode *inode, *template_inode;
+ struct reference_template_args *args = _args;
+ WIMStruct *wim = args->wim;
+ WIMStruct *template_wim = args->template_wim;
+
+ if (dentry->d_inode->i_visited)
+ return 0;
+
+ ret = calculate_dentry_full_path(dentry);
+ if (ret)
+ return ret;
+
+ template_dentry = get_dentry(template_wim, dentry->_full_path);
+ if (!template_dentry) {
+ DEBUG("\"%"TS"\": newly added file", dentry->_full_path);
+ return 0;
+ }
+
+ inode = dentry->d_inode;
+ template_inode = template_dentry->d_inode;
+
+ if (inode_metadata_consistent(inode, template_inode,
+ template_wim->lookup_table)) {
+ /*DEBUG("\"%"TS"\": No change detected", dentry->_full_path);*/
+ ret = inode_copy_checksums(inode, template_inode,
+ wim, template_wim);
+ inode->i_visited = 1;
+ } else {
+ DEBUG("\"%"TS"\": change detected!", dentry->_full_path);
+ ret = 0;
+ }
+ return ret;
+}
+
+/* API function documented in wimlib.h */
+WIMLIBAPI int
+wimlib_reference_template_image(WIMStruct *wim, int new_image,
+ WIMStruct *template_wim, int template_image,
+ int flags, wimlib_progress_func_t progress_func)
+{
+ int ret;
+ struct wim_image_metadata *new_imd;
+
+ if (wim == NULL || template_wim == NULL)
+ return WIMLIB_ERR_INVALID_PARAM;
+
+ if (wim == template_wim && new_image == template_image)
+ return WIMLIB_ERR_INVALID_PARAM;
+
+ if (new_image < 1 || new_image > wim->hdr.image_count)
+ return WIMLIB_ERR_INVALID_IMAGE;
+
+ if (!wim_has_metadata(wim))
+ return WIMLIB_ERR_METADATA_NOT_FOUND;
+
+ new_imd = wim->image_metadata[new_image - 1];
+ if (!new_imd->modified)
+ return WIMLIB_ERR_INVALID_PARAM;
+
+ ret = select_wim_image(template_wim, template_image);
+ if (ret)
+ return ret;
+
+ struct reference_template_args args = {
+ .wim = wim,
+ .template_wim = template_wim,
+ };
+
+ ret = for_dentry_in_tree(new_imd->root_dentry,
+ dentry_reference_template, &args);
+ dentry_tree_clear_inode_visited(new_imd->root_dentry);
+ return ret;
+}