2 * dentry.c - see description below
6 * Copyright (C) 2012, 2013, 2014, 2015 Eric Biggers
8 * This file is free software; you can redistribute it and/or modify it under
9 * the terms of the GNU Lesser General Public License as published by the Free
10 * Software Foundation; either version 3 of the License, or (at your option) any
13 * This file is distributed in the hope that it will be useful, but WITHOUT
14 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
15 * FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
18 * You should have received a copy of the GNU Lesser General Public License
19 * along with this file; if not, see http://www.gnu.org/licenses/.
23 * This file contains logic to deal with WIM directory entries, or "dentries":
25 * - Reading a dentry tree from a metadata resource in a WIM file
26 * - Writing a dentry tree to a metadata resource in a WIM file
27 * - Iterating through a tree of WIM dentries
28 * - Path lookup: translating a path into a WIM dentry or inode
29 * - Creating, modifying, and deleting WIM dentries
33 * - A WIM file can contain multiple images, each of which has an independent
34 * tree of dentries. "On disk", the dentry tree for an image is stored in
35 * the "metadata resource" for that image.
37 * - Multiple dentries in an image may correspond to the same inode, or "file".
38 * When this occurs, it means that the file has multiple names, or "hard
39 * links". A dentry is not a file, but rather the name of a file!
41 * - Inodes are not represented explicitly in the WIM file format. Instead,
42 * the metadata resource provides a "hard link group ID" for each dentry.
43 * wimlib handles pulling out actual inodes from this information, but this
44 * occurs in inode_fixup.c and not in this file.
46 * - wimlib does not allow *directory* hard links, so a WIM image really does
47 * have a *tree* of dentries (and not an arbitrary graph of dentries).
49 * - wimlib indexes dentries both case-insensitively and case-sensitively,
50 * allowing either behavior to be used for path lookup.
52 * - Multiple dentries in a directory might have the same case-insensitive
53 * name. But wimlib enforces that at most one dentry in a directory can have
54 * a given case-sensitive name.
63 #include "wimlib/assert.h"
64 #include "wimlib/dentry.h"
65 #include "wimlib/inode.h"
66 #include "wimlib/encoding.h"
67 #include "wimlib/endianness.h"
68 #include "wimlib/metadata.h"
69 #include "wimlib/paths.h"
71 /* On-disk format of a WIM dentry (directory entry), located in the metadata
72 * resource for a WIM image. */
73 struct wim_dentry_on_disk {
75 /* Length of this directory entry in bytes, not including any extra
76 * stream entries. Should be a multiple of 8 so that the following
77 * dentry or extra stream entry is aligned on an 8-byte boundary. (If
78 * not, wimlib will round it up.) It must be at least as long as the
79 * fixed-length fields of the dentry (WIM_DENTRY_DISK_SIZE), plus the
80 * lengths of the file name and/or short name if present, plus the size
81 * of any "extra" data.
83 * It is also possible for this field to be 0. This case indicates the
84 * end of a list of sibling entries in a directory. It also means the
85 * real length is 8, because the dentry included only the length field,
86 * but that takes up 8 bytes. */
89 /* File attributes for the file or directory. This is a bitwise OR of
90 * the FILE_ATTRIBUTE_* constants and should correspond to the value
91 * retrieved by GetFileAttributes() on Windows. */
94 /* A value that specifies the security descriptor for this file or
95 * directory. If -1, the file or directory has no security descriptor.
96 * Otherwise, it is a 0-based index into the WIM image's table of
97 * security descriptors (see: `struct wim_security_data') */
100 /* Offset, in bytes, from the start of the uncompressed metadata
101 * resource of this directory's child directory entries, or 0 if this
102 * directory entry does not correspond to a directory or otherwise does
103 * not have any children. */
106 /* Reserved fields */
110 /* Creation time, last access time, and last write time, in
111 * 100-nanosecond intervals since 12:00 a.m UTC January 1, 1601. They
112 * should correspond to the times gotten by calling GetFileTime() on
115 le64 last_access_time;
116 le64 last_write_time;
119 * Usually this is the SHA-1 message digest of the file's "contents"
120 * (the unnamed data stream).
122 * If the file has FILE_ATTRIBUTE_REPARSE_POINT set, then this is
123 * instead usually the SHA-1 message digest of the uncompressed reparse
126 * However, there are some special rules that need to be applied to
127 * interpret this field correctly when extra stream entries are present.
128 * See the code for details.
130 u8 default_hash[SHA1_HASH_SIZE];
132 /* Unknown field (maybe accidental padding) */
136 * The following 8-byte union contains either information about the
137 * reparse point (for files with FILE_ATTRIBUTE_REPARSE_POINT set), or
138 * the "hard link group ID" (for other files).
140 * The reparse point information contains ReparseTag and ReparseReserved
141 * from the header of the reparse point buffer. It also contains a flag
142 * that indicates whether a reparse point fixup (for the target of an
143 * absolute symbolic link or junction) was done or not.
145 * The "hard link group ID" is like an inode number; all dentries for
146 * the same inode share the same value. See inode_fixup.c for more
149 * Note that this union creates the limitation that reparse point files
150 * cannot have multiple names (hard links).
157 } _packed_attribute reparse;
159 le64 hard_link_group_id;
160 } _packed_attribute nonreparse;
163 /* Number of extra stream entries that directly follow this dentry
165 le16 num_extra_streams;
167 /* If nonzero, this is the length, in bytes, of this dentry's UTF-16LE
168 * encoded short name (8.3 DOS-compatible name), excluding the null
169 * terminator. If zero, then the long name of this dentry does not have
170 * a corresponding short name (but this does not exclude the possibility
171 * that another dentry for the same file has a short name). */
172 le16 short_name_nbytes;
174 /* If nonzero, this is the length, in bytes, of this dentry's UTF-16LE
175 * encoded "long" name, excluding the null terminator. If zero, then
176 * this file has no long name. The root dentry should not have a long
177 * name, but all other dentries in the image should have long names. */
178 le16 file_name_nbytes;
180 /* Beginning of optional, variable-length fields */
182 /* If file_name_nbytes != 0, the next field will be the UTF-16LE encoded
183 * long file name. This will be null-terminated, so the size of this
184 * field will really be file_name_nbytes + 2. */
185 /*utf16lechar file_name[];*/
187 /* If short_name_nbytes != 0, the next field will be the UTF-16LE
188 * encoded short name. This will be null-terminated, so the size of
189 * this field will really be short_name_nbytes + 2. */
190 /*utf16lechar short_name[];*/
192 /* If there is still space in the dentry (according to the 'length'
193 * field) after 8-byte alignment, then the remaining space will be a
194 * variable-length list of tagged metadata items. See tagged_items.c
195 * for more information. */
196 /* u8 tagged_items[] _aligned_attribute(8); */
199 /* If num_extra_streams != 0, then there are that many extra stream
200 * entries following the dentry, starting on the next 8-byte aligned
201 * boundary. They are not counted in the 'length' field of the dentry.
204 /* On-disk format of an extra stream entry. This represents an extra NTFS-style
205 * "stream" associated with the file, such as a named data stream. */
206 struct wim_extra_stream_entry_on_disk {
208 /* Length of this extra stream entry, in bytes. This includes all
209 * fixed-length fields, plus the name and null terminator if present,
210 * and any needed padding such that the length is a multiple of 8. */
216 /* SHA-1 message digest of this stream's uncompressed data, or all
217 * zeroes if this stream's data is of zero length. */
218 u8 hash[SHA1_HASH_SIZE];
220 /* Length of this stream's name, in bytes and excluding the null
221 * terminator; or 0 if this stream is unnamed. */
224 /* Stream name in UTF-16LE. It is @name_nbytes bytes long, excluding
225 * the null terminator. There is a null terminator character if
226 * @name_nbytes != 0; i.e., if this stream is named. */
231 do_dentry_set_name(struct wim_dentry *dentry, utf16lechar *file_name,
232 size_t file_name_nbytes)
234 FREE(dentry->file_name);
235 dentry->file_name = file_name;
236 dentry->file_name_nbytes = file_name_nbytes;
238 if (dentry_has_short_name(dentry)) {
239 FREE(dentry->short_name);
240 dentry->short_name = NULL;
241 dentry->short_name_nbytes = 0;
246 * Set the name of a WIM dentry from a UTF-16LE string.
248 * This sets the long name of the dentry. The short name will automatically be
249 * removed, since it may not be appropriate for the new long name.
251 * The @name string need not be null-terminated, since its length is specified
254 * If @name_nbytes is 0, both the long and short names of the dentry will be
257 * Only use this function on unlinked dentries, since it doesn't update the name
258 * indices. For dentries that are currently linked into the tree, use
261 * Returns 0 or WIMLIB_ERR_NOMEM.
264 dentry_set_name_utf16le(struct wim_dentry *dentry, const utf16lechar *name,
267 utf16lechar *dup = NULL;
270 dup = utf16le_dupz(name, name_nbytes);
272 return WIMLIB_ERR_NOMEM;
274 do_dentry_set_name(dentry, dup, name_nbytes);
280 * Set the name of a WIM dentry from a 'tchar' string.
282 * This sets the long name of the dentry. The short name will automatically be
283 * removed, since it may not be appropriate for the new long name.
285 * If @name is NULL or empty, both the long and short names of the dentry will
288 * Only use this function on unlinked dentries, since it doesn't update the name
289 * indices. For dentries that are currently linked into the tree, use
292 * Returns 0 or an error code resulting from a failed string conversion.
295 dentry_set_name(struct wim_dentry *dentry, const tchar *name)
297 utf16lechar *name_utf16le = NULL;
298 size_t name_utf16le_nbytes = 0;
302 ret = tstr_to_utf16le(name, tstrlen(name) * sizeof(tchar),
303 &name_utf16le, &name_utf16le_nbytes);
308 do_dentry_set_name(dentry, name_utf16le, name_utf16le_nbytes);
312 /* Calculate the minimum unaligned length, in bytes, of an on-disk WIM dentry
313 * that has names of the specified lengths. (Zero length means the
314 * corresponding name actually does not exist.) The returned value excludes
315 * tagged metadata items as well as any extra stream entries that may need to
316 * follow the dentry. */
318 dentry_min_len_with_names(u16 file_name_nbytes, u16 short_name_nbytes)
320 size_t length = sizeof(struct wim_dentry_on_disk);
321 if (file_name_nbytes)
322 length += (u32)file_name_nbytes + 2;
323 if (short_name_nbytes)
324 length += (u32)short_name_nbytes + 2;
329 /* Return the length, in bytes, required for the specified stream on-disk, when
330 * represented as an extra stream entry. */
332 stream_out_total_length(const struct wim_inode_stream *strm)
334 /* Account for the fixed length portion */
335 size_t len = sizeof(struct wim_extra_stream_entry_on_disk);
337 /* For named streams, account for the variable-length name. */
338 if (stream_is_named(strm))
339 len += utf16le_len_bytes(strm->stream_name) + 2;
341 /* Account for any necessary padding to the next 8-byte boundary. */
342 return ALIGN(len, 8);
346 * Calculate the total number of bytes that will be consumed when a dentry is
347 * written. This includes the fixed-length portion of the dentry, the name
348 * fields, any tagged metadata items, and any extra stream entries. This also
349 * includes all alignment bytes.
352 dentry_out_total_length(const struct wim_dentry *dentry)
354 const struct wim_inode *inode = dentry->d_inode;
357 len = dentry_min_len_with_names(dentry->file_name_nbytes,
358 dentry->short_name_nbytes);
361 len += ALIGN(inode->i_extra_size, 8);
363 if (!(inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED)) {
365 * Extra stream entries:
367 * - Use one extra stream entry for each named data stream
368 * - Use one extra stream entry for the unnamed data stream when there is either:
369 * - a reparse point stream
370 * - at least one named data stream (for Windows PE bug workaround)
371 * - Use one extra stream entry for the reparse point stream if there is one
373 bool have_named_data_stream = false;
374 bool have_reparse_point_stream = false;
375 for (unsigned i = 0; i < inode->i_num_streams; i++) {
376 const struct wim_inode_stream *strm = &inode->i_streams[i];
377 if (stream_is_named_data_stream(strm)) {
378 len += stream_out_total_length(strm);
379 have_named_data_stream = true;
380 } else if (strm->stream_type == STREAM_TYPE_REPARSE_POINT) {
381 wimlib_assert(inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT);
382 have_reparse_point_stream = true;
386 if (have_named_data_stream || have_reparse_point_stream) {
387 if (have_reparse_point_stream)
388 len += ALIGN(sizeof(struct wim_extra_stream_entry_on_disk), 8);
389 len += ALIGN(sizeof(struct wim_extra_stream_entry_on_disk), 8);
396 /* Internal version of for_dentry_in_tree() that omits the NULL check */
398 do_for_dentry_in_tree(struct wim_dentry *dentry,
399 int (*visitor)(struct wim_dentry *, void *), void *arg)
402 struct wim_dentry *child;
404 ret = (*visitor)(dentry, arg);
408 for_dentry_child(child, dentry) {
409 ret = do_for_dentry_in_tree(child, visitor, arg);
416 /* Internal version of for_dentry_in_tree_depth() that omits the NULL check */
418 do_for_dentry_in_tree_depth(struct wim_dentry *dentry,
419 int (*visitor)(struct wim_dentry *, void *), void *arg)
422 struct wim_dentry *child;
424 for_dentry_child_postorder(child, dentry) {
425 ret = do_for_dentry_in_tree_depth(child, visitor, arg);
429 return unlikely((*visitor)(dentry, arg));
433 * Call a function on all dentries in a tree.
435 * @arg will be passed as the second argument to each invocation of @visitor.
437 * This function does a pre-order traversal --- that is, a parent will be
438 * visited before its children. It also will visit siblings in order of
439 * case-sensitive filename. Equivalently, this function visits the entire tree
440 * in the case-sensitive lexicographic order of the full paths.
442 * It is safe to pass NULL for @root, which means that the dentry tree is empty.
443 * In this case, this function does nothing.
445 * @visitor must not modify the structure of the dentry tree during the
448 * The return value will be 0 if all calls to @visitor returned 0. Otherwise,
449 * the return value will be the first nonzero value returned by @visitor.
452 for_dentry_in_tree(struct wim_dentry *root,
453 int (*visitor)(struct wim_dentry *, void *), void *arg)
457 return do_for_dentry_in_tree(root, visitor, arg);
460 /* Like for_dentry_in_tree(), but do a depth-first traversal of the dentry tree.
461 * That is, the visitor function will be called on a dentry's children before
462 * itself. It will be safe to free a dentry when visiting it. */
464 for_dentry_in_tree_depth(struct wim_dentry *root,
465 int (*visitor)(struct wim_dentry *, void *), void *arg)
469 return do_for_dentry_in_tree_depth(root, visitor, arg);
473 * Calculate the full path to @dentry within the WIM image, if not already done.
475 * The full name will be saved in the cached value 'dentry->_full_path'.
477 * Whenever possible, use dentry_full_path() instead of calling this and
478 * accessing _full_path directly.
480 * Returns 0 or an error code resulting from a failed string conversion.
483 calculate_dentry_full_path(struct wim_dentry *dentry)
487 const struct wim_dentry *d;
489 if (dentry->_full_path)
495 ulen += d->file_name_nbytes / sizeof(utf16lechar);
497 d = d->d_parent; /* assumes d == d->d_parent for root */
498 } while (!dentry_is_root(d));
500 utf16lechar ubuf[ulen];
501 utf16lechar *p = &ubuf[ulen];
505 p -= d->file_name_nbytes / sizeof(utf16lechar);
506 memcpy(p, d->file_name, d->file_name_nbytes);
507 *--p = cpu_to_le16(WIM_PATH_SEPARATOR);
508 d = d->d_parent; /* assumes d == d->d_parent for root */
509 } while (!dentry_is_root(d));
511 wimlib_assert(p == ubuf);
513 return utf16le_to_tstr(ubuf, ulen * sizeof(utf16lechar),
514 &dentry->_full_path, &dummy);
518 * Return the full path to the @dentry within the WIM image, or NULL if the full
519 * path could not be determined due to a string conversion error.
521 * The returned memory will be cached in the dentry, so the caller is not
522 * responsible for freeing it.
525 dentry_full_path(struct wim_dentry *dentry)
527 calculate_dentry_full_path(dentry);
528 return dentry->_full_path;
532 dentry_calculate_subdir_offset(struct wim_dentry *dentry, void *_subdir_offset_p)
534 if (dentry_is_directory(dentry)) {
535 u64 *subdir_offset_p = _subdir_offset_p;
536 struct wim_dentry *child;
538 /* Set offset of directory's child dentries */
539 dentry->subdir_offset = *subdir_offset_p;
541 /* Account for child dentries */
542 for_dentry_child(child, dentry)
543 *subdir_offset_p += dentry_out_total_length(child);
545 /* Account for end-of-directory entry */
546 *subdir_offset_p += 8;
548 /* Not a directory; set subdir_offset to 0 */
549 dentry->subdir_offset = 0;
555 * Calculate the subdir offsets for a dentry tree, in preparation of writing
556 * that dentry tree to a metadata resource.
558 * The subdir offset of each dentry is the offset in the uncompressed metadata
559 * resource at which its child dentries begin, or 0 if that dentry has no
562 * The caller must initialize *subdir_offset_p to the first subdir offset that
563 * is available to use after the root dentry is written.
565 * When this function returns, *subdir_offset_p will have been advanced past the
566 * size needed for the dentry tree within the uncompressed metadata resource.
569 calculate_subdir_offsets(struct wim_dentry *root, u64 *subdir_offset_p)
571 for_dentry_in_tree(root, dentry_calculate_subdir_offset, subdir_offset_p);
574 /* Compare the UTF-16LE long filenames of two dentries case insensitively. */
576 dentry_compare_names_case_insensitive(const struct wim_dentry *d1,
577 const struct wim_dentry *d2)
579 return cmp_utf16le_strings(d1->file_name,
580 d1->file_name_nbytes / 2,
582 d2->file_name_nbytes / 2,
586 /* Compare the UTF-16LE long filenames of two dentries case sensitively. */
588 dentry_compare_names_case_sensitive(const struct wim_dentry *d1,
589 const struct wim_dentry *d2)
591 return cmp_utf16le_strings(d1->file_name,
592 d1->file_name_nbytes / 2,
594 d2->file_name_nbytes / 2,
599 _avl_dentry_compare_names_ci(const struct avl_tree_node *n1,
600 const struct avl_tree_node *n2)
602 const struct wim_dentry *d1, *d2;
604 d1 = avl_tree_entry(n1, struct wim_dentry, d_index_node_ci);
605 d2 = avl_tree_entry(n2, struct wim_dentry, d_index_node_ci);
606 return dentry_compare_names_case_insensitive(d1, d2);
610 _avl_dentry_compare_names(const struct avl_tree_node *n1,
611 const struct avl_tree_node *n2)
613 const struct wim_dentry *d1, *d2;
615 d1 = avl_tree_entry(n1, struct wim_dentry, d_index_node);
616 d2 = avl_tree_entry(n2, struct wim_dentry, d_index_node);
617 return dentry_compare_names_case_sensitive(d1, d2);
620 /* Default case sensitivity behavior for searches with
621 * WIMLIB_CASE_PLATFORM_DEFAULT specified. This can be modified by passing
622 * WIMLIB_INIT_FLAG_DEFAULT_CASE_SENSITIVE or
623 * WIMLIB_INIT_FLAG_DEFAULT_CASE_INSENSITIVE to wimlib_global_init(). */
624 bool default_ignore_case =
632 /* Case-sensitive dentry lookup. Only @file_name and @file_name_nbytes of
633 * @dummy must be valid. */
634 static struct wim_dentry *
635 dir_lookup(const struct wim_inode *dir, const struct wim_dentry *dummy)
637 struct avl_tree_node *node;
639 node = avl_tree_lookup_node(dir->i_children,
640 &dummy->d_index_node,
641 _avl_dentry_compare_names);
644 return avl_tree_entry(node, struct wim_dentry, d_index_node);
647 /* Case-insensitive dentry lookup. Only @file_name and @file_name_nbytes of
648 * @dummy must be valid. */
649 static struct wim_dentry *
650 dir_lookup_ci(const struct wim_inode *dir, const struct wim_dentry *dummy)
652 struct avl_tree_node *node;
654 node = avl_tree_lookup_node(dir->i_children_ci,
655 &dummy->d_index_node_ci,
656 _avl_dentry_compare_names_ci);
659 return avl_tree_entry(node, struct wim_dentry, d_index_node_ci);
662 /* Given a UTF-16LE filename and a directory, look up the dentry for the file.
663 * Return it if found, otherwise NULL. This has configurable case sensitivity,
664 * and @name need not be null-terminated. */
666 get_dentry_child_with_utf16le_name(const struct wim_dentry *dentry,
667 const utf16lechar *name,
669 CASE_SENSITIVITY_TYPE case_ctype)
671 const struct wim_inode *dir = dentry->d_inode;
672 bool ignore_case = will_ignore_case(case_ctype);
673 struct wim_dentry dummy;
674 struct wim_dentry *child;
676 dummy.file_name = (utf16lechar*)name;
677 dummy.file_name_nbytes = name_nbytes;
680 /* Case-sensitive lookup. */
681 return dir_lookup(dir, &dummy);
683 /* Case-insensitive lookup. */
685 child = dir_lookup_ci(dir, &dummy);
689 if (likely(list_empty(&child->d_ci_conflict_list)))
690 /* Only one dentry has this case-insensitive name; return it */
693 /* Multiple dentries have the same case-insensitive name. Choose the
694 * dentry with the same case-sensitive name, if one exists; otherwise
695 * print a warning and choose one of the possible dentries arbitrarily.
697 struct wim_dentry *alt = child;
702 if (!dentry_compare_names_case_sensitive(&dummy, alt))
704 alt = list_entry(alt->d_ci_conflict_list.next,
705 struct wim_dentry, d_ci_conflict_list);
706 } while (alt != child);
708 WARNING("Result of case-insensitive lookup is ambiguous\n"
709 " (returning \"%"TS"\" of %zu "
710 "possible files, including \"%"TS"\")",
711 dentry_full_path(child),
713 dentry_full_path(list_entry(child->d_ci_conflict_list.next,
715 d_ci_conflict_list)));
719 /* Given a 'tchar' filename and a directory, look up the dentry for the file.
720 * If the filename was successfully converted to UTF-16LE and the dentry was
721 * found, return it; otherwise return NULL. This has configurable case
724 get_dentry_child_with_name(const struct wim_dentry *dentry, const tchar *name,
725 CASE_SENSITIVITY_TYPE case_type)
728 const utf16lechar *name_utf16le;
729 size_t name_utf16le_nbytes;
730 struct wim_dentry *child;
732 ret = tstr_get_utf16le_and_len(name, &name_utf16le,
733 &name_utf16le_nbytes);
737 child = get_dentry_child_with_utf16le_name(dentry,
741 tstr_put_utf16le(name_utf16le);
745 /* This is the UTF-16LE version of get_dentry(), currently private to this file
746 * because no one needs it besides get_dentry(). */
747 static struct wim_dentry *
748 get_dentry_utf16le(WIMStruct *wim, const utf16lechar *path,
749 CASE_SENSITIVITY_TYPE case_type)
751 struct wim_dentry *cur_dentry;
752 const utf16lechar *name_start, *name_end;
754 /* Start with the root directory of the image. Note: this will be NULL
755 * if an image has been added directly with wimlib_add_empty_image() but
756 * no files have been added yet; in that case we fail with ENOENT. */
757 cur_dentry = wim_get_current_root_dentry(wim);
761 if (cur_dentry == NULL) {
766 if (*name_start && !dentry_is_directory(cur_dentry)) {
771 while (*name_start == cpu_to_le16(WIM_PATH_SEPARATOR))
777 name_end = name_start;
780 } while (*name_end != cpu_to_le16(WIM_PATH_SEPARATOR) && *name_end);
782 cur_dentry = get_dentry_child_with_utf16le_name(cur_dentry,
784 (u8*)name_end - (u8*)name_start,
786 name_start = name_end;
791 * WIM path lookup: translate a path in the currently selected WIM image to the
792 * corresponding dentry, if it exists.
795 * The WIMStruct for the WIM. The search takes place in the currently
799 * The path to look up, given relative to the root of the WIM image.
800 * Characters with value WIM_PATH_SEPARATOR are taken to be path
801 * separators. Leading path separators are ignored, whereas one or more
802 * trailing path separators cause the path to only match a directory.
805 * The case-sensitivity behavior of this function, as one of the following
808 * - WIMLIB_CASE_SENSITIVE: Perform the search case sensitively. This means
809 * that names must match exactly.
811 * - WIMLIB_CASE_INSENSITIVE: Perform the search case insensitively. This
812 * means that names are considered to match if they are equal when
813 * transformed to upper case. If a path component matches multiple names
814 * case-insensitively, the name that matches the path component
815 * case-sensitively is chosen, if existent; otherwise one
816 * case-insensitively matching name is chosen arbitrarily.
818 * - WIMLIB_CASE_PLATFORM_DEFAULT: Perform either case-sensitive or
819 * case-insensitive search, depending on the value of the global variable
820 * default_ignore_case.
822 * In any case, no Unicode normalization is done before comparing strings.
824 * Returns a pointer to the dentry that is the result of the lookup, or NULL if
825 * no such dentry exists. If NULL is returned, errno is set to one of the
828 * ENOTDIR if one of the path components used as a directory existed but
829 * was not, in fact, a directory.
835 * - This function does not consider a reparse point to be a directory, even
836 * if it has FILE_ATTRIBUTE_DIRECTORY set.
838 * - This function does not dereference symbolic links or junction points
839 * when performing the search.
841 * - Since this function ignores leading slashes, the empty path is valid and
842 * names the root directory of the WIM image.
844 * - An image added with wimlib_add_empty_image() does not have a root
845 * directory yet, and this function will fail with ENOENT for any path on
849 get_dentry(WIMStruct *wim, const tchar *path, CASE_SENSITIVITY_TYPE case_type)
852 const utf16lechar *path_utf16le;
853 struct wim_dentry *dentry;
855 ret = tstr_get_utf16le(path, &path_utf16le);
858 dentry = get_dentry_utf16le(wim, path_utf16le, case_type);
859 tstr_put_utf16le(path_utf16le);
863 /* Modify @path, which is a null-terminated string @len 'tchars' in length,
864 * in-place to produce the path to its parent directory. */
866 to_parent_name(tchar *path, size_t len)
868 ssize_t i = (ssize_t)len - 1;
869 while (i >= 0 && path[i] == WIM_PATH_SEPARATOR)
871 while (i >= 0 && path[i] != WIM_PATH_SEPARATOR)
873 while (i >= 0 && path[i] == WIM_PATH_SEPARATOR)
875 path[i + 1] = T('\0');
878 /* Similar to get_dentry(), but returns the dentry named by @path with the last
879 * component stripped off.
881 * Note: The returned dentry is NOT guaranteed to be a directory. */
883 get_parent_dentry(WIMStruct *wim, const tchar *path,
884 CASE_SENSITIVITY_TYPE case_type)
886 size_t path_len = tstrlen(path);
887 tchar buf[path_len + 1];
889 tmemcpy(buf, path, path_len + 1);
890 to_parent_name(buf, path_len);
891 return get_dentry(wim, buf, case_type);
895 * Create an unlinked dentry.
897 * @name specifies the long name to give the new dentry. If NULL or empty, the
898 * new dentry will be given no long name.
900 * The new dentry will have no short name and no associated inode.
902 * On success, returns 0 and a pointer to the new, allocated dentry is stored in
903 * *dentry_ret. On failure, returns WIMLIB_ERR_NOMEM or an error code resulting
904 * from a failed string conversion.
907 new_dentry(const tchar *name, struct wim_dentry **dentry_ret)
909 struct wim_dentry *dentry;
912 dentry = CALLOC(1, sizeof(struct wim_dentry));
914 return WIMLIB_ERR_NOMEM;
917 ret = dentry_set_name(dentry, name);
923 dentry->d_parent = dentry;
924 *dentry_ret = dentry;
928 /* Like new_dentry(), but also allocate an inode and associate it with the
929 * dentry. If set_timestamps=true, the timestamps for the inode will be set to
930 * the current time; otherwise, they will be left 0. */
932 new_dentry_with_new_inode(const tchar *name, bool set_timestamps,
933 struct wim_dentry **dentry_ret)
935 struct wim_dentry *dentry;
936 struct wim_inode *inode;
939 ret = new_dentry(name, &dentry);
943 inode = new_inode(dentry, set_timestamps);
946 return WIMLIB_ERR_NOMEM;
949 *dentry_ret = dentry;
953 /* Like new_dentry(), but also associate the new dentry with the specified inode
954 * and acquire a reference to each of the inode's blobs. */
956 new_dentry_with_existing_inode(const tchar *name, struct wim_inode *inode,
957 struct wim_dentry **dentry_ret)
959 int ret = new_dentry(name, dentry_ret);
962 d_associate(*dentry_ret, inode);
963 inode_ref_blobs(inode);
967 /* Create an unnamed dentry with a new inode for a directory with the default
970 new_filler_directory(struct wim_dentry **dentry_ret)
973 struct wim_dentry *dentry;
975 ret = new_dentry_with_new_inode(NULL, true, &dentry);
978 /* Leave the inode number as 0; this is allowed for non
979 * hard-linked files. */
980 dentry->d_inode->i_attributes = FILE_ATTRIBUTE_DIRECTORY;
981 *dentry_ret = dentry;
986 dentry_clear_inode_visited(struct wim_dentry *dentry, void *_ignore)
988 dentry->d_inode->i_visited = 0;
993 dentry_tree_clear_inode_visited(struct wim_dentry *root)
995 for_dentry_in_tree(root, dentry_clear_inode_visited, NULL);
1001 * In addition to freeing the dentry itself, this disassociates the dentry from
1002 * its inode. If the inode is no longer in use, it will be freed as well.
1005 free_dentry(struct wim_dentry *dentry)
1008 d_disassociate(dentry);
1009 FREE(dentry->file_name);
1010 FREE(dentry->short_name);
1011 FREE(dentry->_full_path);
1017 do_free_dentry(struct wim_dentry *dentry, void *_ignore)
1019 free_dentry(dentry);
1024 do_free_dentry_and_unref_blobs(struct wim_dentry *dentry, void *blob_table)
1026 inode_unref_blobs(dentry->d_inode, blob_table);
1027 free_dentry(dentry);
1032 * Free all dentries in a tree.
1035 * The root of the dentry tree to free. If NULL, this function has no
1039 * A pointer to the blob table for the WIM, or NULL if not specified. If
1040 * specified, this function will decrement the reference counts of the
1041 * blobs referenced by the dentries.
1043 * This function also releases references to the corresponding inodes.
1045 * This function does *not* unlink @root from its parent directory, if it has
1046 * one. If @root has a parent, the caller must unlink @root before calling this
1050 free_dentry_tree(struct wim_dentry *root, struct blob_table *blob_table)
1052 int (*f)(struct wim_dentry *, void *);
1055 f = do_free_dentry_and_unref_blobs;
1059 for_dentry_in_tree_depth(root, f, blob_table);
1062 /* Insert the @child dentry into the case sensitive index of the @dir directory.
1063 * Return NULL if successfully inserted, otherwise a pointer to the
1064 * already-inserted duplicate. */
1065 static struct wim_dentry *
1066 dir_index_child(struct wim_inode *dir, struct wim_dentry *child)
1068 struct avl_tree_node *duplicate;
1070 duplicate = avl_tree_insert(&dir->i_children,
1071 &child->d_index_node,
1072 _avl_dentry_compare_names);
1075 return avl_tree_entry(duplicate, struct wim_dentry, d_index_node);
1078 /* Insert the @child dentry into the case insensitive index of the @dir
1079 * directory. Return NULL if successfully inserted, otherwise a pointer to the
1080 * already-inserted duplicate. */
1081 static struct wim_dentry *
1082 dir_index_child_ci(struct wim_inode *dir, struct wim_dentry *child)
1084 struct avl_tree_node *duplicate;
1086 duplicate = avl_tree_insert(&dir->i_children_ci,
1087 &child->d_index_node_ci,
1088 _avl_dentry_compare_names_ci);
1091 return avl_tree_entry(duplicate, struct wim_dentry, d_index_node_ci);
1094 /* Remove the specified dentry from its directory's case-sensitive index. */
1096 dir_unindex_child(struct wim_inode *dir, struct wim_dentry *child)
1098 avl_tree_remove(&dir->i_children, &child->d_index_node);
1101 /* Remove the specified dentry from its directory's case-insensitive index. */
1103 dir_unindex_child_ci(struct wim_inode *dir, struct wim_dentry *child)
1105 avl_tree_remove(&dir->i_children_ci, &child->d_index_node_ci);
1108 /* Return true iff the specified dentry is in its parent directory's
1109 * case-insensitive index. */
1111 dentry_in_ci_index(const struct wim_dentry *dentry)
1113 return !avl_tree_node_is_unlinked(&dentry->d_index_node_ci);
1117 * Link a dentry into the tree.
1120 * The dentry that will be the parent of @child. It must name a directory.
1123 * The dentry to link. It must be currently unlinked.
1125 * Returns NULL if successful. If @parent already contains a dentry with the
1126 * same case-sensitive name as @child, returns a pointer to this duplicate
1130 dentry_add_child(struct wim_dentry *parent, struct wim_dentry *child)
1132 struct wim_dentry *duplicate;
1133 struct wim_inode *dir;
1135 wimlib_assert(parent != child);
1137 dir = parent->d_inode;
1139 wimlib_assert(inode_is_directory(dir));
1141 duplicate = dir_index_child(dir, child);
1145 duplicate = dir_index_child_ci(dir, child);
1147 list_add(&child->d_ci_conflict_list, &duplicate->d_ci_conflict_list);
1148 avl_tree_node_set_unlinked(&child->d_index_node_ci);
1150 INIT_LIST_HEAD(&child->d_ci_conflict_list);
1152 child->d_parent = parent;
1156 /* Unlink a dentry from the tree. */
1158 unlink_dentry(struct wim_dentry *dentry)
1160 struct wim_inode *dir;
1162 /* Do nothing if the dentry is root or it's already unlinked. Not
1163 * actually necessary based on the current callers, but we do the check
1164 * here to be safe. */
1165 if (unlikely(dentry->d_parent == dentry))
1168 dir = dentry->d_parent->d_inode;
1170 dir_unindex_child(dir, dentry);
1172 if (dentry_in_ci_index(dentry)) {
1174 dir_unindex_child_ci(dir, dentry);
1176 if (!list_empty(&dentry->d_ci_conflict_list)) {
1177 /* Make a different case-insensitively-the-same dentry
1178 * be the "representative" in the search index. */
1179 struct list_head *next;
1180 struct wim_dentry *other;
1181 struct wim_dentry *existing;
1183 next = dentry->d_ci_conflict_list.next;
1184 other = list_entry(next, struct wim_dentry, d_ci_conflict_list);
1185 existing = dir_index_child_ci(dir, other);
1186 wimlib_assert(existing == NULL);
1189 list_del(&dentry->d_ci_conflict_list);
1191 /* Not actually necessary, but to be safe don't retain the now-obsolete
1192 * parent pointer. */
1193 dentry->d_parent = dentry;
1197 read_extra_data(const u8 *p, const u8 *end, struct wim_inode *inode)
1199 while (((uintptr_t)p & 7) && p < end)
1202 if (unlikely(p < end)) {
1203 inode->i_extra = memdup(p, end - p);
1204 if (!inode->i_extra)
1205 return WIMLIB_ERR_NOMEM;
1206 inode->i_extra_size = end - p;
1212 * Set the type of each stream for an encrypted file.
1214 * All data streams of the encrypted file should have been packed into a single
1215 * stream in the format provided by ReadEncryptedFileRaw() on Windows. We
1216 * assign this stream type STREAM_TYPE_EFSRPC_RAW_DATA.
1218 * Encrypted files can't have a reparse point stream. In the on-disk NTFS
1219 * format they can, but as far as I know the reparse point stream of an
1220 * encrypted file can't be stored in the WIM format in a way that's compatible
1221 * with WIMGAPI, nor is there even any way for it to be read or written on
1222 * Windows when the process does not have access to the file encryption key.
1225 assign_stream_types_encrypted(struct wim_inode *inode)
1227 for (unsigned i = 0; i < inode->i_num_streams; i++) {
1228 struct wim_inode_stream *strm = &inode->i_streams[i];
1229 if (!stream_is_named(strm) && !is_zero_hash(strm->_stream_hash))
1231 strm->stream_type = STREAM_TYPE_EFSRPC_RAW_DATA;
1238 * Set the type of each stream for an unencrypted file.
1240 * There will be an unnamed data stream, a reparse point stream, or both an
1241 * unnamed data stream and a reparse point stream. In addition, there may be
1242 * named data streams.
1245 assign_stream_types_unencrypted(struct wim_inode *inode)
1247 bool found_reparse_point_stream = false;
1248 bool found_unnamed_data_stream = false;
1249 struct wim_inode_stream *unnamed_stream_with_zero_hash = NULL;
1251 for (unsigned i = 0; i < inode->i_num_streams; i++) {
1252 struct wim_inode_stream *strm = &inode->i_streams[i];
1254 if (stream_is_named(strm)) {
1255 /* Named data stream */
1256 strm->stream_type = STREAM_TYPE_DATA;
1257 } else if (!is_zero_hash(strm->_stream_hash)) {
1258 if ((inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) &&
1259 !found_reparse_point_stream) {
1260 found_reparse_point_stream = true;
1261 strm->stream_type = STREAM_TYPE_REPARSE_POINT;
1262 } else if (!found_unnamed_data_stream) {
1263 found_unnamed_data_stream = true;
1264 strm->stream_type = STREAM_TYPE_DATA;
1267 /* If no stream name is specified and the hash is zero,
1268 * then remember this stream for later so that we can
1269 * assign it to the unnamed data stream if we don't find
1270 * a better candidate. */
1271 unnamed_stream_with_zero_hash = strm;
1275 if (!found_unnamed_data_stream && unnamed_stream_with_zero_hash != NULL)
1276 unnamed_stream_with_zero_hash->stream_type = STREAM_TYPE_DATA;
1280 * Read and interpret the collection of streams for the specified inode.
1283 setup_inode_streams(const u8 *p, const u8 *end, struct wim_inode *inode,
1284 unsigned num_extra_streams, const u8 *default_hash,
1287 const u8 *orig_p = p;
1289 inode->i_num_streams = 1 + num_extra_streams;
1291 if (unlikely(inode->i_num_streams > ARRAY_LEN(inode->i_embedded_streams))) {
1292 inode->i_streams = CALLOC(inode->i_num_streams,
1293 sizeof(inode->i_streams[0]));
1294 if (!inode->i_streams)
1295 return WIMLIB_ERR_NOMEM;
1298 /* Use the default hash field for the first stream */
1299 inode->i_streams[0].stream_name = (utf16lechar *)NO_STREAM_NAME;
1300 copy_hash(inode->i_streams[0]._stream_hash, default_hash);
1301 inode->i_streams[0].stream_type = STREAM_TYPE_UNKNOWN;
1302 inode->i_streams[0].stream_id = 0;
1304 /* Read the extra stream entries */
1305 for (unsigned i = 1; i < inode->i_num_streams; i++) {
1306 struct wim_inode_stream *strm;
1307 const struct wim_extra_stream_entry_on_disk *disk_strm;
1311 strm = &inode->i_streams[i];
1313 strm->stream_id = i;
1315 /* Do we have at least the size of the fixed-length data we know
1317 if ((end - p) < sizeof(struct wim_extra_stream_entry_on_disk))
1318 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1320 disk_strm = (const struct wim_extra_stream_entry_on_disk *)p;
1322 /* Read the length field */
1323 length = ALIGN(le64_to_cpu(disk_strm->length), 8);
1325 /* Make sure the length field is neither so small it doesn't
1326 * include all the fixed-length data nor so large it overflows
1327 * the metadata resource buffer. */
1328 if (length < sizeof(struct wim_extra_stream_entry_on_disk) ||
1330 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1332 /* Read the rest of the fixed-length data. */
1334 copy_hash(strm->_stream_hash, disk_strm->hash);
1335 name_nbytes = le16_to_cpu(disk_strm->name_nbytes);
1337 /* If stream_name_nbytes != 0, the stream is named. */
1338 if (name_nbytes != 0) {
1339 /* The name is encoded in UTF16-LE, which uses 2-byte
1340 * coding units, so the length of the name had better be
1341 * an even number of bytes. */
1342 if (name_nbytes & 1)
1343 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1345 /* Add the length of the stream name to get the length
1346 * we actually need to read. Make sure this isn't more
1347 * than the specified length of the entry. */
1348 if (sizeof(struct wim_extra_stream_entry_on_disk) +
1349 name_nbytes > length)
1350 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1352 strm->stream_name = utf16le_dupz(disk_strm->name,
1354 if (!strm->stream_name)
1355 return WIMLIB_ERR_NOMEM;
1357 strm->stream_name = (utf16lechar *)NO_STREAM_NAME;
1360 strm->stream_type = STREAM_TYPE_UNKNOWN;
1365 inode->i_next_stream_id = inode->i_num_streams;
1367 /* Now, assign a type to each stream. Unfortunately this requires
1368 * various hacks because stream types aren't explicitly provided in the
1369 * WIM on-disk format. */
1371 if (unlikely(inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED))
1372 assign_stream_types_encrypted(inode);
1374 assign_stream_types_unencrypted(inode);
1376 *offset_p += p - orig_p;
1380 /* Read a dentry, including all extra stream entries that follow it, from an
1381 * uncompressed metadata resource buffer. */
1383 read_dentry(const u8 * restrict buf, size_t buf_len,
1384 u64 *offset_p, struct wim_dentry **dentry_ret)
1386 u64 offset = *offset_p;
1389 const struct wim_dentry_on_disk *disk_dentry;
1390 struct wim_dentry *dentry;
1391 struct wim_inode *inode;
1392 u16 short_name_nbytes;
1393 u16 file_name_nbytes;
1394 u64 calculated_size;
1397 BUILD_BUG_ON(sizeof(struct wim_dentry_on_disk) != WIM_DENTRY_DISK_SIZE);
1399 /* Before reading the whole dentry, we need to read just the length.
1400 * This is because a dentry of length 8 (that is, just the length field)
1401 * terminates the list of sibling directory entries. */
1403 /* Check for buffer overrun. */
1404 if (unlikely(offset + sizeof(u64) > buf_len ||
1405 offset + sizeof(u64) < offset))
1406 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1408 /* Get pointer to the dentry data. */
1410 disk_dentry = (const struct wim_dentry_on_disk*)p;
1412 /* Get dentry length. */
1413 length = ALIGN(le64_to_cpu(disk_dentry->length), 8);
1415 /* Check for end-of-directory. */
1421 /* Validate dentry length. */
1422 if (unlikely(length < sizeof(struct wim_dentry_on_disk)))
1423 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1425 /* Check for buffer overrun. */
1426 if (unlikely(offset + length > buf_len ||
1427 offset + length < offset))
1428 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1430 /* Allocate new dentry structure, along with a preliminary inode. */
1431 ret = new_dentry_with_new_inode(NULL, false, &dentry);
1435 inode = dentry->d_inode;
1437 /* Read more fields: some into the dentry, and some into the inode. */
1438 inode->i_attributes = le32_to_cpu(disk_dentry->attributes);
1439 inode->i_security_id = le32_to_cpu(disk_dentry->security_id);
1440 dentry->subdir_offset = le64_to_cpu(disk_dentry->subdir_offset);
1441 inode->i_creation_time = le64_to_cpu(disk_dentry->creation_time);
1442 inode->i_last_access_time = le64_to_cpu(disk_dentry->last_access_time);
1443 inode->i_last_write_time = le64_to_cpu(disk_dentry->last_write_time);
1444 inode->i_unknown_0x54 = le32_to_cpu(disk_dentry->unknown_0x54);
1446 if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1447 inode->i_reparse_tag = le32_to_cpu(disk_dentry->reparse.reparse_tag);
1448 inode->i_rp_reserved = le16_to_cpu(disk_dentry->reparse.rp_reserved);
1449 inode->i_rp_flags = le16_to_cpu(disk_dentry->reparse.rp_flags);
1450 /* Leave inode->i_ino at 0. Note: this means that WIM cannot
1451 * represent multiple hard links to a reparse point file. */
1453 inode->i_ino = le64_to_cpu(disk_dentry->nonreparse.hard_link_group_id);
1456 /* Now onto reading the names. There are two of them: the (long) file
1457 * name, and the short name. */
1459 short_name_nbytes = le16_to_cpu(disk_dentry->short_name_nbytes);
1460 file_name_nbytes = le16_to_cpu(disk_dentry->file_name_nbytes);
1462 if (unlikely((short_name_nbytes & 1) | (file_name_nbytes & 1))) {
1463 ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1464 goto err_free_dentry;
1467 /* We now know the length of the file name and short name. Make sure
1468 * the length of the dentry is large enough to actually hold them. */
1469 calculated_size = dentry_min_len_with_names(file_name_nbytes,
1472 if (unlikely(length < calculated_size)) {
1473 ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1474 goto err_free_dentry;
1477 /* Advance p to point past the base dentry, to the first name. */
1478 p += sizeof(struct wim_dentry_on_disk);
1480 /* Read the filename if present. Note: if the filename is empty, there
1481 * is no null terminator following it. */
1482 if (file_name_nbytes) {
1483 dentry->file_name = utf16le_dupz(p, file_name_nbytes);
1484 if (dentry->file_name == NULL) {
1485 ret = WIMLIB_ERR_NOMEM;
1486 goto err_free_dentry;
1488 dentry->file_name_nbytes = file_name_nbytes;
1489 p += (u32)file_name_nbytes + 2;
1492 /* Read the short filename if present. Note: if there is no short
1493 * filename, there is no null terminator following it. */
1494 if (short_name_nbytes) {
1495 dentry->short_name = utf16le_dupz(p, short_name_nbytes);
1496 if (dentry->short_name == NULL) {
1497 ret = WIMLIB_ERR_NOMEM;
1498 goto err_free_dentry;
1500 dentry->short_name_nbytes = short_name_nbytes;
1501 p += (u32)short_name_nbytes + 2;
1504 /* Read extra data at end of dentry (but before extra stream entries).
1505 * This may contain tagged metadata items. */
1506 ret = read_extra_data(p, &buf[offset + length], inode);
1508 goto err_free_dentry;
1512 /* Set up the inode's collection of streams. */
1513 ret = setup_inode_streams(&buf[offset],
1516 le16_to_cpu(disk_dentry->num_extra_streams),
1517 disk_dentry->default_hash,
1520 goto err_free_dentry;
1522 *offset_p = offset; /* Sets offset of next dentry in directory */
1523 *dentry_ret = dentry;
1527 free_dentry(dentry);
1531 /* Is the dentry named "." or ".." ? */
1533 dentry_is_dot_or_dotdot(const struct wim_dentry *dentry)
1535 if (dentry->file_name_nbytes <= 4) {
1536 if (dentry->file_name_nbytes == 4) {
1537 if (dentry->file_name[0] == cpu_to_le16('.') &&
1538 dentry->file_name[1] == cpu_to_le16('.'))
1540 } else if (dentry->file_name_nbytes == 2) {
1541 if (dentry->file_name[0] == cpu_to_le16('.'))
1549 read_dentry_tree_recursive(const u8 * restrict buf, size_t buf_len,
1550 struct wim_dentry * restrict dir)
1552 u64 cur_offset = dir->subdir_offset;
1554 /* Check for cyclic directory structure, which would cause infinite
1555 * recursion if not handled. */
1556 for (struct wim_dentry *d = dir->d_parent;
1557 !dentry_is_root(d); d = d->d_parent)
1559 if (unlikely(d->subdir_offset == cur_offset)) {
1560 ERROR("Cyclic directory structure detected: children "
1561 "of \"%"TS"\" coincide with children of \"%"TS"\"",
1562 dentry_full_path(dir), dentry_full_path(d));
1563 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1568 struct wim_dentry *child;
1569 struct wim_dentry *duplicate;
1572 /* Read next child of @dir. */
1573 ret = read_dentry(buf, buf_len, &cur_offset, &child);
1577 /* Check for end of directory. */
1581 /* All dentries except the root should be named. */
1582 if (unlikely(!dentry_has_long_name(child))) {
1583 WARNING("Ignoring unnamed dentry in "
1584 "directory \"%"TS"\"", dentry_full_path(dir));
1589 /* Don't allow files named "." or "..". */
1590 if (unlikely(dentry_is_dot_or_dotdot(child))) {
1591 WARNING("Ignoring file named \".\" or \"..\"; "
1592 "potentially malicious archive!!!");
1597 /* Link the child into the directory. */
1598 duplicate = dentry_add_child(dir, child);
1599 if (unlikely(duplicate)) {
1600 /* We already found a dentry with this same
1601 * case-sensitive long name. Only keep the first one.
1603 WARNING("Ignoring duplicate file \"%"TS"\" "
1604 "(the WIM image already contains a file "
1605 "at that path with the exact same name)",
1606 dentry_full_path(duplicate));
1611 /* If this child is a directory that itself has children, call
1612 * this procedure recursively. */
1613 if (child->subdir_offset != 0) {
1614 if (likely(dentry_is_directory(child))) {
1615 ret = read_dentry_tree_recursive(buf,
1621 WARNING("Ignoring children of "
1622 "non-directory file \"%"TS"\"",
1623 dentry_full_path(child));
1630 * Read a tree of dentries from a WIM metadata resource.
1633 * Buffer containing an uncompressed WIM metadata resource.
1636 * Length of the uncompressed metadata resource, in bytes.
1639 * Offset in the metadata resource of the root of the dentry tree.
1642 * On success, either NULL or a pointer to the root dentry is written to
1643 * this location. The former case only occurs in the unexpected case that
1644 * the tree began with an end-of-directory entry.
1647 * WIMLIB_ERR_SUCCESS (0)
1648 * WIMLIB_ERR_INVALID_METADATA_RESOURCE
1652 read_dentry_tree(const u8 *buf, size_t buf_len,
1653 u64 root_offset, struct wim_dentry **root_ret)
1656 struct wim_dentry *root;
1658 DEBUG("Reading dentry tree (root_offset=%"PRIu64")", root_offset);
1660 ret = read_dentry(buf, buf_len, &root_offset, &root);
1664 if (likely(root != NULL)) {
1665 if (unlikely(dentry_has_long_name(root) ||
1666 dentry_has_short_name(root)))
1668 WARNING("The root directory has a nonempty name; "
1670 dentry_set_name(root, NULL);
1673 if (unlikely(!dentry_is_directory(root))) {
1674 ERROR("The root of the WIM image is not a directory!");
1675 ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1676 goto err_free_dentry_tree;
1679 if (likely(root->subdir_offset != 0)) {
1680 ret = read_dentry_tree_recursive(buf, buf_len, root);
1682 goto err_free_dentry_tree;
1685 WARNING("The metadata resource has no directory entries; "
1686 "treating as an empty image.");
1691 err_free_dentry_tree:
1692 free_dentry_tree(root, NULL);
1697 write_extra_stream_entry(u8 * restrict p, const utf16lechar * restrict name,
1698 const u8 * restrict hash)
1700 struct wim_extra_stream_entry_on_disk *disk_strm =
1701 (struct wim_extra_stream_entry_on_disk *)p;
1705 if (name == NO_STREAM_NAME)
1708 name_nbytes = utf16le_len_bytes(name);
1710 disk_strm->reserved = 0;
1711 copy_hash(disk_strm->hash, hash);
1712 disk_strm->name_nbytes = cpu_to_le16(name_nbytes);
1713 p += sizeof(struct wim_extra_stream_entry_on_disk);
1714 if (name_nbytes != 0)
1715 p = mempcpy(p, name, name_nbytes + 2);
1716 /* Align to 8-byte boundary */
1717 while ((uintptr_t)p & 7)
1719 disk_strm->length = cpu_to_le64(p - orig_p);
1724 * Write a WIM dentry to an output buffer.
1726 * This includes any extra stream entries that may follow the dentry itself.
1729 * The dentry to write.
1732 * The memory location to which to write the data.
1734 * Returns a pointer to the byte following the last written.
1737 write_dentry(const struct wim_dentry * restrict dentry, u8 * restrict p)
1739 const struct wim_inode *inode;
1740 struct wim_dentry_on_disk *disk_dentry;
1743 wimlib_assert(((uintptr_t)p & 7) == 0); /* 8 byte aligned */
1746 inode = dentry->d_inode;
1747 disk_dentry = (struct wim_dentry_on_disk*)p;
1749 disk_dentry->attributes = cpu_to_le32(inode->i_attributes);
1750 disk_dentry->security_id = cpu_to_le32(inode->i_security_id);
1751 disk_dentry->subdir_offset = cpu_to_le64(dentry->subdir_offset);
1753 disk_dentry->unused_1 = cpu_to_le64(0);
1754 disk_dentry->unused_2 = cpu_to_le64(0);
1756 disk_dentry->creation_time = cpu_to_le64(inode->i_creation_time);
1757 disk_dentry->last_access_time = cpu_to_le64(inode->i_last_access_time);
1758 disk_dentry->last_write_time = cpu_to_le64(inode->i_last_write_time);
1759 disk_dentry->unknown_0x54 = cpu_to_le32(inode->i_unknown_0x54);
1760 if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1761 disk_dentry->reparse.reparse_tag = cpu_to_le32(inode->i_reparse_tag);
1762 disk_dentry->reparse.rp_reserved = cpu_to_le16(inode->i_rp_reserved);
1763 disk_dentry->reparse.rp_flags = cpu_to_le16(inode->i_rp_flags);
1765 disk_dentry->nonreparse.hard_link_group_id =
1766 cpu_to_le64((inode->i_nlink == 1) ? 0 : inode->i_ino);
1769 disk_dentry->short_name_nbytes = cpu_to_le16(dentry->short_name_nbytes);
1770 disk_dentry->file_name_nbytes = cpu_to_le16(dentry->file_name_nbytes);
1771 p += sizeof(struct wim_dentry_on_disk);
1773 wimlib_assert(dentry_is_root(dentry) != dentry_has_long_name(dentry));
1775 if (dentry_has_long_name(dentry))
1776 p = mempcpy(p, dentry->file_name, (u32)dentry->file_name_nbytes + 2);
1778 if (dentry_has_short_name(dentry))
1779 p = mempcpy(p, dentry->short_name, (u32)dentry->short_name_nbytes + 2);
1781 /* Align to 8-byte boundary */
1782 while ((uintptr_t)p & 7)
1785 if (inode->i_extra_size) {
1786 /* Extra tagged items --- not usually present. */
1787 p = mempcpy(p, inode->i_extra, inode->i_extra_size);
1789 /* Align to 8-byte boundary */
1790 while ((uintptr_t)p & 7)
1794 disk_dentry->length = cpu_to_le64(p - orig_p);
1798 if (unlikely(inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED)) {
1799 const struct wim_inode_stream *efs_strm;
1802 efs_strm = inode_get_unnamed_stream(inode, STREAM_TYPE_EFSRPC_RAW_DATA);
1803 efs_hash = efs_strm ? stream_hash(efs_strm) : zero_hash;
1804 copy_hash(disk_dentry->default_hash, efs_hash);
1805 disk_dentry->num_extra_streams = cpu_to_le16(0);
1808 * Extra stream entries:
1810 * - Use one extra stream entry for each named data stream
1811 * - Use one extra stream entry for the unnamed data stream when there is either:
1812 * - a reparse point stream
1813 * - at least one named data stream (for Windows PE bug workaround)
1814 * - Use one extra stream entry for the reparse point stream if there is one
1816 bool have_named_data_stream = false;
1817 bool have_reparse_point_stream = false;
1818 const u8 *unnamed_data_stream_hash = zero_hash;
1819 const u8 *reparse_point_hash;
1820 for (unsigned i = 0; i < inode->i_num_streams; i++) {
1821 const struct wim_inode_stream *strm = &inode->i_streams[i];
1822 if (strm->stream_type == STREAM_TYPE_DATA) {
1823 if (stream_is_named(strm))
1824 have_named_data_stream = true;
1826 unnamed_data_stream_hash = stream_hash(strm);
1827 } else if (strm->stream_type == STREAM_TYPE_REPARSE_POINT) {
1828 have_reparse_point_stream = true;
1829 reparse_point_hash = stream_hash(strm);
1833 if (unlikely(have_reparse_point_stream || have_named_data_stream)) {
1835 unsigned num_extra_streams = 0;
1837 copy_hash(disk_dentry->default_hash, zero_hash);
1839 if (have_reparse_point_stream) {
1840 p = write_extra_stream_entry(p, NO_STREAM_NAME,
1841 reparse_point_hash);
1842 num_extra_streams++;
1845 p = write_extra_stream_entry(p, NO_STREAM_NAME,
1846 unnamed_data_stream_hash);
1847 num_extra_streams++;
1849 for (unsigned i = 0; i < inode->i_num_streams; i++) {
1850 const struct wim_inode_stream *strm = &inode->i_streams[i];
1851 if (stream_is_named_data_stream(strm)) {
1852 p = write_extra_stream_entry(p, strm->stream_name,
1854 num_extra_streams++;
1857 wimlib_assert(num_extra_streams <= 0xFFFF);
1859 disk_dentry->num_extra_streams = cpu_to_le16(num_extra_streams);
1861 copy_hash(disk_dentry->default_hash, unnamed_data_stream_hash);
1862 disk_dentry->num_extra_streams = cpu_to_le16(0);
1870 write_dir_dentries(struct wim_dentry *dir, void *_pp)
1872 if (dir->subdir_offset != 0) {
1875 struct wim_dentry *child;
1877 /* write child dentries */
1878 for_dentry_child(child, dir)
1879 p = write_dentry(child, p);
1881 /* write end of directory entry */
1890 * Write a directory tree to the metadata resource.
1893 * The root of a dentry tree on which calculate_subdir_offsets() has been
1894 * called. This cannot be NULL; if the dentry tree is empty, the caller is
1895 * expected to first generate a dummy root directory.
1898 * Pointer to a buffer with enough space for the dentry tree. This size
1899 * must have been obtained by calculate_subdir_offsets().
1901 * Returns a pointer to the byte following the last written.
1904 write_dentry_tree(struct wim_dentry *root, u8 *p)
1906 DEBUG("Writing dentry tree.");
1908 wimlib_assert(root != NULL);
1910 /* write root dentry and end-of-directory entry following it */
1911 p = write_dentry(root, p);
1915 /* write the rest of the dentry tree */
1916 for_dentry_in_tree(root, write_dir_dentries, &p);