2 * dentry.c - see description below
6 * Copyright (C) 2012-2016 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 0xFFFFFFFF, the file or directory has no security
96 * descriptor. Otherwise, it is a 0-based index into the WIM image's
97 * table of 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. */
180 /* Beginning of optional, variable-length fields */
182 /* If name_nbytes != 0, the next field will be the UTF-16LE encoded long
183 * name. This will be null-terminated, so the size of this field will
184 * really be name_nbytes + 2. */
185 /*utf16lechar 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 *name,
234 FREE(dentry->d_name);
235 dentry->d_name = name;
236 dentry->d_name_nbytes = name_nbytes;
238 if (dentry_has_short_name(dentry)) {
239 FREE(dentry->d_short_name);
240 dentry->d_short_name = NULL;
241 dentry->d_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 name_nbytes, u16 short_name_nbytes)
320 size_t length = sizeof(struct wim_dentry_on_disk);
322 length += (u32)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->d_name_nbytes,
358 dentry->d_short_name_nbytes);
362 len += ALIGN(inode->i_extra->size, 8);
364 if (!(inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED)) {
366 * Extra stream entries:
368 * - Use one extra stream entry for each named data stream
369 * - Use one extra stream entry for the unnamed data stream when there is either:
370 * - a reparse point stream
371 * - at least one named data stream (for Windows PE bug workaround)
372 * - Use one extra stream entry for the reparse point stream if there is one
374 bool have_named_data_stream = false;
375 bool have_reparse_point_stream = false;
376 for (unsigned i = 0; i < inode->i_num_streams; i++) {
377 const struct wim_inode_stream *strm = &inode->i_streams[i];
378 if (stream_is_named_data_stream(strm)) {
379 len += stream_out_total_length(strm);
380 have_named_data_stream = true;
381 } else if (strm->stream_type == STREAM_TYPE_REPARSE_POINT) {
382 wimlib_assert(inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT);
383 have_reparse_point_stream = true;
387 if (have_named_data_stream || have_reparse_point_stream) {
388 if (have_reparse_point_stream)
389 len += ALIGN(sizeof(struct wim_extra_stream_entry_on_disk), 8);
390 len += ALIGN(sizeof(struct wim_extra_stream_entry_on_disk), 8);
397 /* Internal version of for_dentry_in_tree() that omits the NULL check */
399 do_for_dentry_in_tree(struct wim_dentry *dentry,
400 int (*visitor)(struct wim_dentry *, void *), void *arg)
403 struct wim_dentry *child;
405 ret = (*visitor)(dentry, arg);
409 for_dentry_child(child, dentry) {
410 ret = do_for_dentry_in_tree(child, visitor, arg);
417 /* Internal version of for_dentry_in_tree_depth() that omits the NULL check */
419 do_for_dentry_in_tree_depth(struct wim_dentry *dentry,
420 int (*visitor)(struct wim_dentry *, void *), void *arg)
423 struct wim_dentry *child;
425 for_dentry_child_postorder(child, dentry) {
426 ret = do_for_dentry_in_tree_depth(child, visitor, arg);
430 return unlikely((*visitor)(dentry, arg));
434 * Call a function on all dentries in a tree.
436 * @arg will be passed as the second argument to each invocation of @visitor.
438 * This function does a pre-order traversal --- that is, a parent will be
439 * visited before its children. It also will visit siblings in order of
440 * case-sensitive filename. Equivalently, this function visits the entire tree
441 * in the case-sensitive lexicographic order of the full paths.
443 * It is safe to pass NULL for @root, which means that the dentry tree is empty.
444 * In this case, this function does nothing.
446 * @visitor must not modify the structure of the dentry tree during the
449 * The return value will be 0 if all calls to @visitor returned 0. Otherwise,
450 * the return value will be the first nonzero value returned by @visitor.
453 for_dentry_in_tree(struct wim_dentry *root,
454 int (*visitor)(struct wim_dentry *, void *), void *arg)
458 return do_for_dentry_in_tree(root, visitor, arg);
461 /* Like for_dentry_in_tree(), but do a depth-first traversal of the dentry tree.
462 * That is, the visitor function will be called on a dentry's children before
463 * itself. It will be safe to free a dentry when visiting it. */
465 for_dentry_in_tree_depth(struct wim_dentry *root,
466 int (*visitor)(struct wim_dentry *, void *), void *arg)
470 return do_for_dentry_in_tree_depth(root, visitor, arg);
474 * Calculate the full path to @dentry within the WIM image, if not already done.
476 * The full name will be saved in the cached value 'dentry->d_full_path'.
478 * Whenever possible, use dentry_full_path() instead of calling this and
479 * accessing d_full_path directly.
481 * Returns 0 or an error code resulting from a failed string conversion.
484 calculate_dentry_full_path(struct wim_dentry *dentry)
487 const struct wim_dentry *d;
489 if (dentry->d_full_path)
495 ulen += d->d_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->d_name_nbytes / sizeof(utf16lechar);
506 if (d->d_name_nbytes)
507 memcpy(p, d->d_name, d->d_name_nbytes);
508 *--p = cpu_to_le16(WIM_PATH_SEPARATOR);
509 d = d->d_parent; /* assumes d == d->d_parent for root */
510 } while (!dentry_is_root(d));
512 wimlib_assert(p == ubuf);
514 return utf16le_to_tstr(ubuf, ulen * sizeof(utf16lechar),
515 &dentry->d_full_path, NULL);
519 * Return the full path to the @dentry within the WIM image, or NULL if the full
520 * path could not be determined due to a string conversion error.
522 * The returned memory will be cached in the dentry, so the caller is not
523 * responsible for freeing it.
526 dentry_full_path(struct wim_dentry *dentry)
528 calculate_dentry_full_path(dentry);
529 return dentry->d_full_path;
533 dentry_calculate_subdir_offset(struct wim_dentry *dentry, void *_subdir_offset_p)
535 if (dentry_is_directory(dentry)) {
536 u64 *subdir_offset_p = _subdir_offset_p;
537 struct wim_dentry *child;
539 /* Set offset of directory's child dentries */
540 dentry->d_subdir_offset = *subdir_offset_p;
542 /* Account for child dentries */
543 for_dentry_child(child, dentry)
544 *subdir_offset_p += dentry_out_total_length(child);
546 /* Account for end-of-directory entry */
547 *subdir_offset_p += 8;
549 /* Not a directory; set the subdir offset to 0 */
550 dentry->d_subdir_offset = 0;
556 * Calculate the subdir offsets for a dentry tree, in preparation of writing
557 * that dentry tree to a metadata resource.
559 * The subdir offset of each dentry is the offset in the uncompressed metadata
560 * resource at which its child dentries begin, or 0 if that dentry has no
563 * The caller must initialize *subdir_offset_p to the first subdir offset that
564 * is available to use after the root dentry is written.
566 * When this function returns, *subdir_offset_p will have been advanced past the
567 * size needed for the dentry tree within the uncompressed metadata resource.
570 calculate_subdir_offsets(struct wim_dentry *root, u64 *subdir_offset_p)
572 for_dentry_in_tree(root, dentry_calculate_subdir_offset, subdir_offset_p);
575 /* Compare the UTF-16LE long filenames of two dentries case insensitively. */
577 dentry_compare_names_case_insensitive(const struct wim_dentry *d1,
578 const struct wim_dentry *d2)
580 return cmp_utf16le_strings(d1->d_name,
581 d1->d_name_nbytes / 2,
583 d2->d_name_nbytes / 2,
587 /* Compare the UTF-16LE long filenames of two dentries case sensitively. */
589 dentry_compare_names_case_sensitive(const struct wim_dentry *d1,
590 const struct wim_dentry *d2)
592 return cmp_utf16le_strings(d1->d_name,
593 d1->d_name_nbytes / 2,
595 d2->d_name_nbytes / 2,
600 _avl_dentry_compare_names_ci(const struct avl_tree_node *n1,
601 const struct avl_tree_node *n2)
603 const struct wim_dentry *d1, *d2;
605 d1 = avl_tree_entry(n1, struct wim_dentry, d_index_node_ci);
606 d2 = avl_tree_entry(n2, struct wim_dentry, d_index_node_ci);
607 return dentry_compare_names_case_insensitive(d1, d2);
611 _avl_dentry_compare_names(const struct avl_tree_node *n1,
612 const struct avl_tree_node *n2)
614 const struct wim_dentry *d1, *d2;
616 d1 = avl_tree_entry(n1, struct wim_dentry, d_index_node);
617 d2 = avl_tree_entry(n2, struct wim_dentry, d_index_node);
618 return dentry_compare_names_case_sensitive(d1, d2);
621 /* Default case sensitivity behavior for searches with
622 * WIMLIB_CASE_PLATFORM_DEFAULT specified. This can be modified by passing
623 * WIMLIB_INIT_FLAG_DEFAULT_CASE_SENSITIVE or
624 * WIMLIB_INIT_FLAG_DEFAULT_CASE_INSENSITIVE to wimlib_global_init(). */
625 bool default_ignore_case =
633 /* Case-sensitive dentry lookup. Only @d_name and @d_name_nbytes of @dummy must
635 static struct wim_dentry *
636 dir_lookup(const struct wim_inode *dir, const struct wim_dentry *dummy)
638 struct avl_tree_node *node;
640 node = avl_tree_lookup_node(dir->i_children,
641 &dummy->d_index_node,
642 _avl_dentry_compare_names);
645 return avl_tree_entry(node, struct wim_dentry, d_index_node);
648 /* Case-insensitive dentry lookup. Only @d_name and @d_name_nbytes of @dummy
650 static struct wim_dentry *
651 dir_lookup_ci(const struct wim_inode *dir, const struct wim_dentry *dummy)
653 struct avl_tree_node *node;
655 node = avl_tree_lookup_node(dir->i_children_ci,
656 &dummy->d_index_node_ci,
657 _avl_dentry_compare_names_ci);
660 return avl_tree_entry(node, struct wim_dentry, d_index_node_ci);
663 /* Given a UTF-16LE filename and a directory, look up the dentry for the file.
664 * Return it if found, otherwise NULL. This has configurable case sensitivity,
665 * and @name need not be null-terminated. */
667 get_dentry_child_with_utf16le_name(const struct wim_dentry *dentry,
668 const utf16lechar *name,
670 CASE_SENSITIVITY_TYPE case_type)
672 const struct wim_inode *dir = dentry->d_inode;
673 bool ignore_case = will_ignore_case(case_type);
674 struct wim_dentry dummy;
675 struct wim_dentry *child;
677 dummy.d_name = (utf16lechar*)name;
678 dummy.d_name_nbytes = name_nbytes;
681 /* Case-sensitive lookup. */
682 return dir_lookup(dir, &dummy);
684 /* Case-insensitive lookup. */
686 child = dir_lookup_ci(dir, &dummy);
690 if (likely(list_empty(&child->d_ci_conflict_list)))
691 /* Only one dentry has this case-insensitive name; return it */
694 /* Multiple dentries have the same case-insensitive name. Choose the
695 * dentry with the same case-sensitive name, if one exists; otherwise
696 * print a warning and choose one of the possible dentries arbitrarily.
698 struct wim_dentry *alt = child;
703 if (!dentry_compare_names_case_sensitive(&dummy, alt))
705 alt = list_entry(alt->d_ci_conflict_list.next,
706 struct wim_dentry, d_ci_conflict_list);
707 } while (alt != child);
709 WARNING("Result of case-insensitive lookup is ambiguous\n"
710 " (returning \"%"TS"\" of %zu "
711 "possible files, including \"%"TS"\")",
712 dentry_full_path(child),
714 dentry_full_path(list_entry(child->d_ci_conflict_list.next,
716 d_ci_conflict_list)));
720 /* Given a 'tchar' filename and a directory, look up the dentry for the file.
721 * If the filename was successfully converted to UTF-16LE and the dentry was
722 * found, return it; otherwise return NULL. This has configurable case
725 get_dentry_child_with_name(const struct wim_dentry *dentry, const tchar *name,
726 CASE_SENSITIVITY_TYPE case_type)
729 const utf16lechar *name_utf16le;
730 size_t name_utf16le_nbytes;
731 struct wim_dentry *child;
733 ret = tstr_get_utf16le_and_len(name, &name_utf16le,
734 &name_utf16le_nbytes);
738 child = get_dentry_child_with_utf16le_name(dentry,
742 tstr_put_utf16le(name_utf16le);
746 /* This is the UTF-16LE version of get_dentry(), currently private to this file
747 * because no one needs it besides get_dentry(). */
748 static struct wim_dentry *
749 get_dentry_utf16le(WIMStruct *wim, const utf16lechar *path,
750 CASE_SENSITIVITY_TYPE case_type)
752 struct wim_dentry *cur_dentry;
753 const utf16lechar *name_start, *name_end;
755 /* Start with the root directory of the image. Note: this will be NULL
756 * if an image has been added directly with wimlib_add_empty_image() but
757 * no files have been added yet; in that case we fail with ENOENT. */
758 cur_dentry = wim_get_current_root_dentry(wim);
762 if (cur_dentry == NULL) {
767 if (*name_start && !dentry_is_directory(cur_dentry)) {
772 while (*name_start == cpu_to_le16(WIM_PATH_SEPARATOR))
778 name_end = name_start;
781 } while (*name_end != cpu_to_le16(WIM_PATH_SEPARATOR) && *name_end);
783 cur_dentry = get_dentry_child_with_utf16le_name(cur_dentry,
785 (u8*)name_end - (u8*)name_start,
787 name_start = name_end;
792 * WIM path lookup: translate a path in the currently selected WIM image to the
793 * corresponding dentry, if it exists.
796 * The WIMStruct for the WIM. The search takes place in the currently
800 * The path to look up, given relative to the root of the WIM image.
801 * Characters with value WIM_PATH_SEPARATOR are taken to be path
802 * separators. Leading path separators are ignored, whereas one or more
803 * trailing path separators cause the path to only match a directory.
806 * The case-sensitivity behavior of this function, as one of the following
809 * - WIMLIB_CASE_SENSITIVE: Perform the search case sensitively. This means
810 * that names must match exactly.
812 * - WIMLIB_CASE_INSENSITIVE: Perform the search case insensitively. This
813 * means that names are considered to match if they are equal when
814 * transformed to upper case. If a path component matches multiple names
815 * case-insensitively, the name that matches the path component
816 * case-sensitively is chosen, if existent; otherwise one
817 * case-insensitively matching name is chosen arbitrarily.
819 * - WIMLIB_CASE_PLATFORM_DEFAULT: Perform either case-sensitive or
820 * case-insensitive search, depending on the value of the global variable
821 * default_ignore_case.
823 * In any case, no Unicode normalization is done before comparing strings.
825 * Returns a pointer to the dentry that is the result of the lookup, or NULL if
826 * no such dentry exists. If NULL is returned, errno is set to one of the
829 * ENOTDIR if one of the path components used as a directory existed but
830 * was not, in fact, a directory.
836 * - This function does not consider a reparse point to be a directory, even
837 * if it has FILE_ATTRIBUTE_DIRECTORY set.
839 * - This function does not dereference symbolic links or junction points
840 * when performing the search.
842 * - Since this function ignores leading slashes, the empty path is valid and
843 * names the root directory of the WIM image.
845 * - An image added with wimlib_add_empty_image() does not have a root
846 * directory yet, and this function will fail with ENOENT for any path on
850 get_dentry(WIMStruct *wim, const tchar *path, CASE_SENSITIVITY_TYPE case_type)
853 const utf16lechar *path_utf16le;
854 struct wim_dentry *dentry;
856 ret = tstr_get_utf16le(path, &path_utf16le);
859 dentry = get_dentry_utf16le(wim, path_utf16le, case_type);
860 tstr_put_utf16le(path_utf16le);
864 /* Modify @path, which is a null-terminated string @len 'tchars' in length,
865 * in-place to produce the path to its parent directory. */
867 to_parent_name(tchar *path, size_t len)
869 ssize_t i = (ssize_t)len - 1;
870 while (i >= 0 && path[i] == WIM_PATH_SEPARATOR)
872 while (i >= 0 && path[i] != WIM_PATH_SEPARATOR)
874 while (i >= 0 && path[i] == WIM_PATH_SEPARATOR)
876 path[i + 1] = T('\0');
879 /* Similar to get_dentry(), but returns the dentry named by @path with the last
880 * component stripped off.
882 * Note: The returned dentry is NOT guaranteed to be a directory. */
884 get_parent_dentry(WIMStruct *wim, const tchar *path,
885 CASE_SENSITIVITY_TYPE case_type)
887 size_t path_len = tstrlen(path);
888 tchar buf[path_len + 1];
890 tmemcpy(buf, path, path_len + 1);
891 to_parent_name(buf, path_len);
892 return get_dentry(wim, buf, case_type);
896 * Create an unlinked dentry.
898 * @name specifies the long name to give the new dentry. If NULL or empty, the
899 * new dentry will be given no long name.
901 * The new dentry will have no short name and no associated inode.
903 * On success, returns 0 and a pointer to the new, allocated dentry is stored in
904 * *dentry_ret. On failure, returns WIMLIB_ERR_NOMEM or an error code resulting
905 * from a failed string conversion.
908 new_dentry(const tchar *name, struct wim_dentry **dentry_ret)
910 struct wim_dentry *dentry;
913 dentry = CALLOC(1, sizeof(struct wim_dentry));
915 return WIMLIB_ERR_NOMEM;
918 ret = dentry_set_name(dentry, name);
924 dentry->d_parent = dentry;
925 *dentry_ret = dentry;
929 /* Like new_dentry(), but also allocate an inode and associate it with the
930 * dentry. If set_timestamps=true, the timestamps for the inode will be set to
931 * the current time; otherwise, they will be left 0. */
933 new_dentry_with_new_inode(const tchar *name, bool set_timestamps,
934 struct wim_dentry **dentry_ret)
936 struct wim_dentry *dentry;
937 struct wim_inode *inode;
940 ret = new_dentry(name, &dentry);
944 inode = new_inode(dentry, set_timestamps);
947 return WIMLIB_ERR_NOMEM;
950 *dentry_ret = dentry;
954 /* Like new_dentry(), but also associate the new dentry with the specified inode
955 * and acquire a reference to each of the inode's blobs. */
957 new_dentry_with_existing_inode(const tchar *name, struct wim_inode *inode,
958 struct wim_dentry **dentry_ret)
960 int ret = new_dentry(name, dentry_ret);
963 d_associate(*dentry_ret, inode);
964 inode_ref_blobs(inode);
968 /* Create an unnamed dentry with a new inode for a directory with the default
971 new_filler_directory(struct wim_dentry **dentry_ret)
974 struct wim_dentry *dentry;
976 ret = new_dentry_with_new_inode(NULL, true, &dentry);
979 /* Leave the inode number as 0; this is allowed for non
980 * hard-linked files. */
981 dentry->d_inode->i_attributes = FILE_ATTRIBUTE_DIRECTORY;
982 *dentry_ret = dentry;
989 * In addition to freeing the dentry itself, this disassociates the dentry from
990 * its inode. If the inode is no longer in use, it will be freed as well.
993 free_dentry(struct wim_dentry *dentry)
996 d_disassociate(dentry);
997 FREE(dentry->d_name);
998 FREE(dentry->d_short_name);
999 FREE(dentry->d_full_path);
1005 do_free_dentry(struct wim_dentry *dentry, void *_ignore)
1007 free_dentry(dentry);
1012 do_free_dentry_and_unref_blobs(struct wim_dentry *dentry, void *blob_table)
1014 inode_unref_blobs(dentry->d_inode, blob_table);
1015 free_dentry(dentry);
1020 * Free all dentries in a tree.
1023 * The root of the dentry tree to free. If NULL, this function has no
1027 * A pointer to the blob table for the WIM, or NULL if not specified. If
1028 * specified, this function will decrement the reference counts of the
1029 * blobs referenced by the dentries.
1031 * This function also releases references to the corresponding inodes.
1033 * This function does *not* unlink @root from its parent directory, if it has
1034 * one. If @root has a parent, the caller must unlink @root before calling this
1038 free_dentry_tree(struct wim_dentry *root, struct blob_table *blob_table)
1040 int (*f)(struct wim_dentry *, void *);
1043 f = do_free_dentry_and_unref_blobs;
1047 for_dentry_in_tree_depth(root, f, blob_table);
1050 /* Insert the @child dentry into the case sensitive index of the @dir directory.
1051 * Return NULL if successfully inserted, otherwise a pointer to the
1052 * already-inserted duplicate. */
1053 static struct wim_dentry *
1054 dir_index_child(struct wim_inode *dir, struct wim_dentry *child)
1056 struct avl_tree_node *duplicate;
1058 duplicate = avl_tree_insert(&dir->i_children,
1059 &child->d_index_node,
1060 _avl_dentry_compare_names);
1063 return avl_tree_entry(duplicate, struct wim_dentry, d_index_node);
1066 /* Insert the @child dentry into the case insensitive index of the @dir
1067 * directory. Return NULL if successfully inserted, otherwise a pointer to the
1068 * already-inserted duplicate. */
1069 static struct wim_dentry *
1070 dir_index_child_ci(struct wim_inode *dir, struct wim_dentry *child)
1072 struct avl_tree_node *duplicate;
1074 duplicate = avl_tree_insert(&dir->i_children_ci,
1075 &child->d_index_node_ci,
1076 _avl_dentry_compare_names_ci);
1079 return avl_tree_entry(duplicate, struct wim_dentry, d_index_node_ci);
1082 /* Remove the specified dentry from its directory's case-sensitive index. */
1084 dir_unindex_child(struct wim_inode *dir, struct wim_dentry *child)
1086 avl_tree_remove(&dir->i_children, &child->d_index_node);
1089 /* Remove the specified dentry from its directory's case-insensitive index. */
1091 dir_unindex_child_ci(struct wim_inode *dir, struct wim_dentry *child)
1093 avl_tree_remove(&dir->i_children_ci, &child->d_index_node_ci);
1096 /* Return true iff the specified dentry is in its parent directory's
1097 * case-insensitive index. */
1099 dentry_in_ci_index(const struct wim_dentry *dentry)
1101 return !avl_tree_node_is_unlinked(&dentry->d_index_node_ci);
1105 * Link a dentry into the tree.
1108 * The dentry that will be the parent of @child. It must name a directory.
1111 * The dentry to link. It must be currently unlinked.
1113 * Returns NULL if successful. If @parent already contains a dentry with the
1114 * same case-sensitive name as @child, returns a pointer to this duplicate
1118 dentry_add_child(struct wim_dentry *parent, struct wim_dentry *child)
1120 struct wim_dentry *duplicate;
1121 struct wim_inode *dir;
1123 wimlib_assert(parent != child);
1125 dir = parent->d_inode;
1127 wimlib_assert(inode_is_directory(dir));
1129 duplicate = dir_index_child(dir, child);
1133 duplicate = dir_index_child_ci(dir, child);
1135 list_add(&child->d_ci_conflict_list, &duplicate->d_ci_conflict_list);
1136 avl_tree_node_set_unlinked(&child->d_index_node_ci);
1138 INIT_LIST_HEAD(&child->d_ci_conflict_list);
1140 child->d_parent = parent;
1144 /* Unlink a dentry from the tree. */
1146 unlink_dentry(struct wim_dentry *dentry)
1148 struct wim_inode *dir;
1150 /* Do nothing if the dentry is root or it's already unlinked. Not
1151 * actually necessary based on the current callers, but we do the check
1152 * here to be safe. */
1153 if (unlikely(dentry->d_parent == dentry))
1156 dir = dentry->d_parent->d_inode;
1158 dir_unindex_child(dir, dentry);
1160 if (dentry_in_ci_index(dentry)) {
1162 dir_unindex_child_ci(dir, dentry);
1164 if (!list_empty(&dentry->d_ci_conflict_list)) {
1165 /* Make a different case-insensitively-the-same dentry
1166 * be the "representative" in the search index. */
1167 struct list_head *next;
1168 struct wim_dentry *other;
1169 struct wim_dentry *existing;
1171 next = dentry->d_ci_conflict_list.next;
1172 other = list_entry(next, struct wim_dentry, d_ci_conflict_list);
1173 existing = dir_index_child_ci(dir, other);
1174 wimlib_assert(existing == NULL);
1177 list_del(&dentry->d_ci_conflict_list);
1179 /* Not actually necessary, but to be safe don't retain the now-obsolete
1180 * parent pointer. */
1181 dentry->d_parent = dentry;
1185 read_extra_data(const u8 *p, const u8 *end, struct wim_inode *inode)
1187 while (((uintptr_t)p & 7) && p < end)
1190 if (unlikely(p < end)) {
1191 inode->i_extra = MALLOC(sizeof(struct wim_inode_extra) +
1193 if (!inode->i_extra)
1194 return WIMLIB_ERR_NOMEM;
1195 inode->i_extra->size = end - p;
1196 memcpy(inode->i_extra->data, p, end - p);
1202 * Set the type of each stream for an encrypted file.
1204 * All data streams of the encrypted file should have been packed into a single
1205 * stream in the format provided by ReadEncryptedFileRaw() on Windows. We
1206 * assign this stream type STREAM_TYPE_EFSRPC_RAW_DATA.
1208 * Encrypted files can't have a reparse point stream. In the on-disk NTFS
1209 * format they can, but as far as I know the reparse point stream of an
1210 * encrypted file can't be stored in the WIM format in a way that's compatible
1211 * with WIMGAPI, nor is there even any way for it to be read or written on
1212 * Windows when the process does not have access to the file encryption key.
1215 assign_stream_types_encrypted(struct wim_inode *inode)
1217 for (unsigned i = 0; i < inode->i_num_streams; i++) {
1218 struct wim_inode_stream *strm = &inode->i_streams[i];
1219 if (!stream_is_named(strm) && !is_zero_hash(strm->_stream_hash))
1221 strm->stream_type = STREAM_TYPE_EFSRPC_RAW_DATA;
1228 * Set the type of each stream for an unencrypted file.
1230 * There will be an unnamed data stream, a reparse point stream, or both an
1231 * unnamed data stream and a reparse point stream. In addition, there may be
1232 * named data streams.
1234 * NOTE: if the file has a reparse point stream or at least one named data
1235 * stream, then WIMGAPI puts *all* streams in the extra stream entries and
1236 * leaves the default stream hash zeroed. wimlib now does the same. However,
1237 * for input we still support the default hash field being used, since wimlib
1238 * used to use it and MS software is somewhat accepting of it as well.
1241 assign_stream_types_unencrypted(struct wim_inode *inode)
1243 bool found_reparse_point_stream = false;
1244 bool found_unnamed_data_stream = false;
1245 struct wim_inode_stream *unnamed_stream_with_zero_hash = NULL;
1247 for (unsigned i = 0; i < inode->i_num_streams; i++) {
1248 struct wim_inode_stream *strm = &inode->i_streams[i];
1250 if (stream_is_named(strm)) {
1251 /* Named data stream */
1252 strm->stream_type = STREAM_TYPE_DATA;
1253 } else if (i != 0 || !is_zero_hash(strm->_stream_hash)) {
1254 /* Unnamed stream in the extra stream entries, OR the
1255 * default stream in the dentry provided that it has a
1257 if ((inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) &&
1258 !found_reparse_point_stream) {
1259 found_reparse_point_stream = true;
1260 strm->stream_type = STREAM_TYPE_REPARSE_POINT;
1261 } else if (!found_unnamed_data_stream) {
1262 found_unnamed_data_stream = true;
1263 strm->stream_type = STREAM_TYPE_DATA;
1265 } else if (!unnamed_stream_with_zero_hash) {
1266 unnamed_stream_with_zero_hash = strm;
1270 if (unnamed_stream_with_zero_hash) {
1271 int type = STREAM_TYPE_UNKNOWN;
1272 if ((inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) &&
1273 !found_reparse_point_stream) {
1274 type = STREAM_TYPE_REPARSE_POINT;
1275 } else if (!found_unnamed_data_stream) {
1276 type = STREAM_TYPE_DATA;
1278 unnamed_stream_with_zero_hash->stream_type = type;
1283 * Read and interpret the collection of streams for the specified inode.
1286 setup_inode_streams(const u8 *p, const u8 *end, struct wim_inode *inode,
1287 unsigned num_extra_streams, const u8 *default_hash,
1290 const u8 *orig_p = p;
1292 inode->i_num_streams = 1 + num_extra_streams;
1294 if (unlikely(inode->i_num_streams > ARRAY_LEN(inode->i_embedded_streams))) {
1295 inode->i_streams = CALLOC(inode->i_num_streams,
1296 sizeof(inode->i_streams[0]));
1297 if (!inode->i_streams)
1298 return WIMLIB_ERR_NOMEM;
1301 /* Use the default hash field for the first stream */
1302 inode->i_streams[0].stream_name = (utf16lechar *)NO_STREAM_NAME;
1303 copy_hash(inode->i_streams[0]._stream_hash, default_hash);
1304 inode->i_streams[0].stream_type = STREAM_TYPE_UNKNOWN;
1305 inode->i_streams[0].stream_id = 0;
1307 /* Read the extra stream entries */
1308 for (unsigned i = 1; i < inode->i_num_streams; i++) {
1309 struct wim_inode_stream *strm;
1310 const struct wim_extra_stream_entry_on_disk *disk_strm;
1314 strm = &inode->i_streams[i];
1316 strm->stream_id = i;
1318 /* Do we have at least the size of the fixed-length data we know
1320 if ((end - p) < sizeof(struct wim_extra_stream_entry_on_disk))
1321 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1323 disk_strm = (const struct wim_extra_stream_entry_on_disk *)p;
1325 /* Read the length field */
1326 length = ALIGN(le64_to_cpu(disk_strm->length), 8);
1328 /* Make sure the length field is neither so small it doesn't
1329 * include all the fixed-length data nor so large it overflows
1330 * the metadata resource buffer. */
1331 if (length < sizeof(struct wim_extra_stream_entry_on_disk) ||
1333 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1335 /* Read the rest of the fixed-length data. */
1337 copy_hash(strm->_stream_hash, disk_strm->hash);
1338 name_nbytes = le16_to_cpu(disk_strm->name_nbytes);
1340 /* If stream_name_nbytes != 0, the stream is named. */
1341 if (name_nbytes != 0) {
1342 /* The name is encoded in UTF16-LE, which uses 2-byte
1343 * coding units, so the length of the name had better be
1344 * an even number of bytes. */
1345 if (name_nbytes & 1)
1346 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1348 /* Add the length of the stream name to get the length
1349 * we actually need to read. Make sure this isn't more
1350 * than the specified length of the entry. */
1351 if (sizeof(struct wim_extra_stream_entry_on_disk) +
1352 name_nbytes > length)
1353 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1355 strm->stream_name = utf16le_dupz(disk_strm->name,
1357 if (!strm->stream_name)
1358 return WIMLIB_ERR_NOMEM;
1360 strm->stream_name = (utf16lechar *)NO_STREAM_NAME;
1363 strm->stream_type = STREAM_TYPE_UNKNOWN;
1368 inode->i_next_stream_id = inode->i_num_streams;
1370 /* Now, assign a type to each stream. Unfortunately this requires
1371 * various hacks because stream types aren't explicitly provided in the
1372 * WIM on-disk format. */
1374 if (unlikely(inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED))
1375 assign_stream_types_encrypted(inode);
1377 assign_stream_types_unencrypted(inode);
1379 *offset_p += p - orig_p;
1383 /* Read a dentry, including all extra stream entries that follow it, from an
1384 * uncompressed metadata resource buffer. */
1386 read_dentry(const u8 * restrict buf, size_t buf_len,
1387 u64 *offset_p, struct wim_dentry **dentry_ret)
1389 u64 offset = *offset_p;
1392 const struct wim_dentry_on_disk *disk_dentry;
1393 struct wim_dentry *dentry;
1394 struct wim_inode *inode;
1395 u16 short_name_nbytes;
1397 u64 calculated_size;
1400 STATIC_ASSERT(sizeof(struct wim_dentry_on_disk) == WIM_DENTRY_DISK_SIZE);
1402 /* Before reading the whole dentry, we need to read just the length.
1403 * This is because a dentry of length 8 (that is, just the length field)
1404 * terminates the list of sibling directory entries. */
1406 /* Check for buffer overrun. */
1407 if (unlikely(offset + sizeof(u64) > buf_len ||
1408 offset + sizeof(u64) < offset))
1409 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1411 /* Get pointer to the dentry data. */
1413 disk_dentry = (const struct wim_dentry_on_disk*)p;
1415 /* Get dentry length. */
1416 length = ALIGN(le64_to_cpu(disk_dentry->length), 8);
1418 /* Check for end-of-directory. */
1424 /* Validate dentry length. */
1425 if (unlikely(length < sizeof(struct wim_dentry_on_disk)))
1426 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1428 /* Check for buffer overrun. */
1429 if (unlikely(offset + length > buf_len ||
1430 offset + length < offset))
1431 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1433 /* Allocate new dentry structure, along with a preliminary inode. */
1434 ret = new_dentry_with_new_inode(NULL, false, &dentry);
1438 inode = dentry->d_inode;
1440 /* Read more fields: some into the dentry, and some into the inode. */
1441 inode->i_attributes = le32_to_cpu(disk_dentry->attributes);
1442 inode->i_security_id = le32_to_cpu(disk_dentry->security_id);
1443 dentry->d_subdir_offset = le64_to_cpu(disk_dentry->subdir_offset);
1444 inode->i_creation_time = le64_to_cpu(disk_dentry->creation_time);
1445 inode->i_last_access_time = le64_to_cpu(disk_dentry->last_access_time);
1446 inode->i_last_write_time = le64_to_cpu(disk_dentry->last_write_time);
1447 inode->i_unknown_0x54 = le32_to_cpu(disk_dentry->unknown_0x54);
1449 if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1450 inode->i_reparse_tag = le32_to_cpu(disk_dentry->reparse.reparse_tag);
1451 inode->i_rp_reserved = le16_to_cpu(disk_dentry->reparse.rp_reserved);
1452 inode->i_rp_flags = le16_to_cpu(disk_dentry->reparse.rp_flags);
1453 /* Leave inode->i_ino at 0. Note: this means that WIM cannot
1454 * represent multiple hard links to a reparse point file. */
1456 inode->i_ino = le64_to_cpu(disk_dentry->nonreparse.hard_link_group_id);
1459 /* Now onto reading the names. There are two of them: the (long) file
1460 * name, and the short name. */
1462 short_name_nbytes = le16_to_cpu(disk_dentry->short_name_nbytes);
1463 name_nbytes = le16_to_cpu(disk_dentry->name_nbytes);
1465 if (unlikely((short_name_nbytes & 1) | (name_nbytes & 1))) {
1466 ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1467 goto err_free_dentry;
1470 /* We now know the length of the file name and short name. Make sure
1471 * the length of the dentry is large enough to actually hold them. */
1472 calculated_size = dentry_min_len_with_names(name_nbytes,
1475 if (unlikely(length < calculated_size)) {
1476 ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1477 goto err_free_dentry;
1480 /* Advance p to point past the base dentry, to the first name. */
1481 p += sizeof(struct wim_dentry_on_disk);
1483 /* Read the filename if present. Note: if the filename is empty, there
1484 * is no null terminator following it. */
1486 dentry->d_name = utf16le_dupz(p, name_nbytes);
1487 if (unlikely(!dentry->d_name)) {
1488 ret = WIMLIB_ERR_NOMEM;
1489 goto err_free_dentry;
1491 dentry->d_name_nbytes = name_nbytes;
1492 p += (u32)name_nbytes + 2;
1495 /* Read the short filename if present. Note: if there is no short
1496 * filename, there is no null terminator following it. */
1497 if (short_name_nbytes) {
1498 dentry->d_short_name = utf16le_dupz(p, short_name_nbytes);
1499 if (unlikely(!dentry->d_short_name)) {
1500 ret = WIMLIB_ERR_NOMEM;
1501 goto err_free_dentry;
1503 dentry->d_short_name_nbytes = short_name_nbytes;
1504 p += (u32)short_name_nbytes + 2;
1507 /* Read extra data at end of dentry (but before extra stream entries).
1508 * This may contain tagged metadata items. */
1509 ret = read_extra_data(p, &buf[offset + length], inode);
1511 goto err_free_dentry;
1515 /* Set up the inode's collection of streams. */
1516 ret = setup_inode_streams(&buf[offset],
1519 le16_to_cpu(disk_dentry->num_extra_streams),
1520 disk_dentry->default_hash,
1523 goto err_free_dentry;
1525 *offset_p = offset; /* Sets offset of next dentry in directory */
1526 *dentry_ret = dentry;
1530 free_dentry(dentry);
1535 dentry_is_dot_or_dotdot(const struct wim_dentry *dentry)
1537 if (dentry->d_name_nbytes <= 4) {
1538 if (dentry->d_name_nbytes == 4) {
1539 if (dentry->d_name[0] == cpu_to_le16('.') &&
1540 dentry->d_name[1] == cpu_to_le16('.'))
1542 } else if (dentry->d_name_nbytes == 2) {
1543 if (dentry->d_name[0] == cpu_to_le16('.'))
1551 dentry_contains_embedded_null(const struct wim_dentry *dentry)
1553 for (unsigned i = 0; i < dentry->d_name_nbytes / 2; i++)
1554 if (dentry->d_name[i] == cpu_to_le16('\0'))
1560 should_ignore_dentry(struct wim_dentry *dir, const struct wim_dentry *dentry)
1562 /* All dentries except the root must be named. */
1563 if (!dentry_has_long_name(dentry)) {
1564 WARNING("Ignoring unnamed file in directory \"%"TS"\"",
1565 dentry_full_path(dir));
1569 /* Don't allow files named "." or "..". Such filenames could be used in
1570 * path traversal attacks. */
1571 if (dentry_is_dot_or_dotdot(dentry)) {
1572 WARNING("Ignoring file named \".\" or \"..\" in directory "
1573 "\"%"TS"\"", dentry_full_path(dir));
1577 /* Don't allow filenames containing embedded null characters. Although
1578 * the null character is already considered an unsupported character for
1579 * extraction by all targets, it is probably a good idea to just forbid
1580 * such names entirely. */
1581 if (dentry_contains_embedded_null(dentry)) {
1582 WARNING("Ignoring filename with embedded null character in "
1583 "directory \"%"TS"\"", dentry_full_path(dir));
1591 read_dentry_tree_recursive(const u8 * restrict buf, size_t buf_len,
1592 struct wim_dentry * restrict dir, unsigned depth)
1594 u64 cur_offset = dir->d_subdir_offset;
1596 /* Disallow extremely deep or cyclic directory structures */
1597 if (unlikely(depth >= 16384)) {
1598 ERROR("Directory structure too deep!");
1599 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1603 struct wim_dentry *child;
1604 struct wim_dentry *duplicate;
1607 /* Read next child of @dir. */
1608 ret = read_dentry(buf, buf_len, &cur_offset, &child);
1612 /* Check for end of directory. */
1616 /* Ignore dentries with bad names. */
1617 if (unlikely(should_ignore_dentry(dir, child))) {
1622 /* Link the child into the directory. */
1623 duplicate = dentry_add_child(dir, child);
1624 if (unlikely(duplicate)) {
1625 /* We already found a dentry with this same
1626 * case-sensitive long name. Only keep the first one.
1628 WARNING("Ignoring duplicate file \"%"TS"\" "
1629 "(the WIM image already contains a file "
1630 "at that path with the exact same name)",
1631 dentry_full_path(duplicate));
1636 /* If this child is a directory that itself has children, call
1637 * this procedure recursively. */
1638 if (child->d_subdir_offset != 0) {
1639 if (likely(dentry_is_directory(child))) {
1640 ret = read_dentry_tree_recursive(buf,
1647 WARNING("Ignoring children of "
1648 "non-directory file \"%"TS"\"",
1649 dentry_full_path(child));
1656 * Read a tree of dentries from a WIM metadata resource.
1659 * Buffer containing an uncompressed WIM metadata resource.
1662 * Length of the uncompressed metadata resource, in bytes.
1665 * Offset in the metadata resource of the root of the dentry tree.
1668 * On success, either NULL or a pointer to the root dentry is written to
1669 * this location. The former case only occurs in the unexpected case that
1670 * the tree began with an end-of-directory entry.
1673 * WIMLIB_ERR_SUCCESS (0)
1674 * WIMLIB_ERR_INVALID_METADATA_RESOURCE
1678 read_dentry_tree(const u8 *buf, size_t buf_len,
1679 u64 root_offset, struct wim_dentry **root_ret)
1682 struct wim_dentry *root;
1684 ret = read_dentry(buf, buf_len, &root_offset, &root);
1688 if (likely(root != NULL)) {
1689 if (unlikely(dentry_has_long_name(root) ||
1690 dentry_has_short_name(root)))
1692 WARNING("The root directory has a nonempty name; "
1694 dentry_set_name(root, NULL);
1697 if (unlikely(!dentry_is_directory(root))) {
1698 ERROR("The root of the WIM image is not a directory!");
1699 ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1700 goto err_free_dentry_tree;
1703 if (likely(root->d_subdir_offset != 0)) {
1704 ret = read_dentry_tree_recursive(buf, buf_len, root, 0);
1706 goto err_free_dentry_tree;
1709 WARNING("The metadata resource has no directory entries; "
1710 "treating as an empty image.");
1715 err_free_dentry_tree:
1716 free_dentry_tree(root, NULL);
1721 write_extra_stream_entry(u8 * restrict p, const utf16lechar * restrict name,
1722 const u8 * restrict hash)
1724 struct wim_extra_stream_entry_on_disk *disk_strm =
1725 (struct wim_extra_stream_entry_on_disk *)p;
1729 if (name == NO_STREAM_NAME)
1732 name_nbytes = utf16le_len_bytes(name);
1734 disk_strm->reserved = 0;
1735 copy_hash(disk_strm->hash, hash);
1736 disk_strm->name_nbytes = cpu_to_le16(name_nbytes);
1737 p += sizeof(struct wim_extra_stream_entry_on_disk);
1738 if (name_nbytes != 0)
1739 p = mempcpy(p, name, name_nbytes + 2);
1740 /* Align to 8-byte boundary */
1741 while ((uintptr_t)p & 7)
1743 disk_strm->length = cpu_to_le64(p - orig_p);
1748 * Write a WIM dentry to an output buffer.
1750 * This includes any extra stream entries that may follow the dentry itself.
1753 * The dentry to write.
1756 * The memory location to which to write the data.
1758 * Returns a pointer to the byte following the last written.
1761 write_dentry(const struct wim_dentry * restrict dentry, u8 * restrict p)
1763 const struct wim_inode *inode;
1764 struct wim_dentry_on_disk *disk_dentry;
1767 wimlib_assert(((uintptr_t)p & 7) == 0); /* 8 byte aligned */
1770 inode = dentry->d_inode;
1771 disk_dentry = (struct wim_dentry_on_disk*)p;
1773 disk_dentry->attributes = cpu_to_le32(inode->i_attributes);
1774 disk_dentry->security_id = cpu_to_le32(inode->i_security_id);
1775 disk_dentry->subdir_offset = cpu_to_le64(dentry->d_subdir_offset);
1777 disk_dentry->unused_1 = cpu_to_le64(0);
1778 disk_dentry->unused_2 = cpu_to_le64(0);
1780 disk_dentry->creation_time = cpu_to_le64(inode->i_creation_time);
1781 disk_dentry->last_access_time = cpu_to_le64(inode->i_last_access_time);
1782 disk_dentry->last_write_time = cpu_to_le64(inode->i_last_write_time);
1783 disk_dentry->unknown_0x54 = cpu_to_le32(inode->i_unknown_0x54);
1784 if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1785 disk_dentry->reparse.reparse_tag = cpu_to_le32(inode->i_reparse_tag);
1786 disk_dentry->reparse.rp_reserved = cpu_to_le16(inode->i_rp_reserved);
1787 disk_dentry->reparse.rp_flags = cpu_to_le16(inode->i_rp_flags);
1789 disk_dentry->nonreparse.hard_link_group_id =
1790 cpu_to_le64((inode->i_nlink == 1) ? 0 : inode->i_ino);
1793 disk_dentry->short_name_nbytes = cpu_to_le16(dentry->d_short_name_nbytes);
1794 disk_dentry->name_nbytes = cpu_to_le16(dentry->d_name_nbytes);
1795 p += sizeof(struct wim_dentry_on_disk);
1797 wimlib_assert(dentry_is_root(dentry) != dentry_has_long_name(dentry));
1799 if (dentry_has_long_name(dentry))
1800 p = mempcpy(p, dentry->d_name, (u32)dentry->d_name_nbytes + 2);
1802 if (dentry_has_short_name(dentry))
1803 p = mempcpy(p, dentry->d_short_name, (u32)dentry->d_short_name_nbytes + 2);
1805 /* Align to 8-byte boundary */
1806 while ((uintptr_t)p & 7)
1809 if (inode->i_extra) {
1810 /* Extra tagged items --- not usually present. */
1811 p = mempcpy(p, inode->i_extra->data, inode->i_extra->size);
1813 /* Align to 8-byte boundary */
1814 while ((uintptr_t)p & 7)
1818 disk_dentry->length = cpu_to_le64(p - orig_p);
1822 if (unlikely(inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED)) {
1823 const struct wim_inode_stream *efs_strm;
1826 efs_strm = inode_get_unnamed_stream(inode, STREAM_TYPE_EFSRPC_RAW_DATA);
1827 efs_hash = efs_strm ? stream_hash(efs_strm) : zero_hash;
1828 copy_hash(disk_dentry->default_hash, efs_hash);
1829 disk_dentry->num_extra_streams = cpu_to_le16(0);
1832 * Extra stream entries:
1834 * - Use one extra stream entry for each named data stream
1835 * - Use one extra stream entry for the unnamed data stream when there is either:
1836 * - a reparse point stream
1837 * - at least one named data stream (for Windows PE bug workaround)
1838 * - Use one extra stream entry for the reparse point stream if there is one
1840 bool have_named_data_stream = false;
1841 bool have_reparse_point_stream = false;
1842 const u8 *unnamed_data_stream_hash = zero_hash;
1843 const u8 *reparse_point_hash;
1844 for (unsigned i = 0; i < inode->i_num_streams; i++) {
1845 const struct wim_inode_stream *strm = &inode->i_streams[i];
1846 if (strm->stream_type == STREAM_TYPE_DATA) {
1847 if (stream_is_named(strm))
1848 have_named_data_stream = true;
1850 unnamed_data_stream_hash = stream_hash(strm);
1851 } else if (strm->stream_type == STREAM_TYPE_REPARSE_POINT) {
1852 have_reparse_point_stream = true;
1853 reparse_point_hash = stream_hash(strm);
1857 if (unlikely(have_reparse_point_stream || have_named_data_stream)) {
1859 unsigned num_extra_streams = 0;
1861 copy_hash(disk_dentry->default_hash, zero_hash);
1863 if (have_reparse_point_stream) {
1864 p = write_extra_stream_entry(p, NO_STREAM_NAME,
1865 reparse_point_hash);
1866 num_extra_streams++;
1869 p = write_extra_stream_entry(p, NO_STREAM_NAME,
1870 unnamed_data_stream_hash);
1871 num_extra_streams++;
1873 for (unsigned i = 0; i < inode->i_num_streams; i++) {
1874 const struct wim_inode_stream *strm = &inode->i_streams[i];
1875 if (stream_is_named_data_stream(strm)) {
1876 p = write_extra_stream_entry(p, strm->stream_name,
1878 num_extra_streams++;
1881 wimlib_assert(num_extra_streams <= 0xFFFF);
1883 disk_dentry->num_extra_streams = cpu_to_le16(num_extra_streams);
1885 copy_hash(disk_dentry->default_hash, unnamed_data_stream_hash);
1886 disk_dentry->num_extra_streams = cpu_to_le16(0);
1894 write_dir_dentries(struct wim_dentry *dir, void *_pp)
1896 if (dir->d_subdir_offset != 0) {
1899 struct wim_dentry *child;
1901 /* write child dentries */
1902 for_dentry_child(child, dir)
1903 p = write_dentry(child, p);
1905 /* write end of directory entry */
1914 * Write a directory tree to the metadata resource.
1917 * The root of a dentry tree on which calculate_subdir_offsets() has been
1918 * called. This cannot be NULL; if the dentry tree is empty, the caller is
1919 * expected to first generate a dummy root directory.
1922 * Pointer to a buffer with enough space for the dentry tree. This size
1923 * must have been obtained by calculate_subdir_offsets().
1925 * Returns a pointer to the byte following the last written.
1928 write_dentry_tree(struct wim_dentry *root, u8 *p)
1930 /* write root dentry and end-of-directory entry following it */
1931 p = write_dentry(root, p);
1935 /* write the rest of the dentry tree */
1936 for_dentry_in_tree(root, write_dir_dentries, &p);