4 * See dentry_tree_fix_inodes() for description.
8 * Copyright (C) 2012, 2013 Eric Biggers
10 * This file is part of wimlib, a library for working with WIM files.
12 * wimlib is free software; you can redistribute it and/or modify it under the
13 * terms of the GNU General Public License as published by the Free
14 * Software Foundation; either version 3 of the License, or (at your option)
17 * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
18 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
19 * A PARTICULAR PURPOSE. See the GNU General Public License for more
22 * You should have received a copy of the GNU General Public License
23 * along with wimlib; if not, see http://www.gnu.org/licenses/.
30 #include "wimlib/dentry.h"
31 #include "wimlib/error.h"
32 #include "wimlib/inode.h"
33 #include "wimlib/inode_table.h"
34 #include "wimlib/lookup_table.h"
36 /* Manual link count of inode (normally we can just check i_nlink) */
38 inode_link_count(const struct wim_inode *inode)
40 const struct list_head *cur;
42 list_for_each(cur, &inode->i_dentry)
48 print_inode_dentries(const struct wim_inode *inode)
50 struct wim_dentry *dentry;
51 inode_for_each_dentry(dentry, inode)
52 tfprintf(stderr, T("%"TS"\n"), dentry_full_path(dentry));
56 inconsistent_inode(const struct wim_inode *inode)
58 if (wimlib_print_errors) {
59 ERROR("An inconsistent hard link group that cannot be corrected has "
61 ERROR("The dentries are located at the following paths:");
62 print_inode_dentries(inode);
67 ads_entries_have_same_name(const struct wim_ads_entry *entry_1,
68 const struct wim_ads_entry *entry_2)
70 return entry_1->stream_name_nbytes == entry_2->stream_name_nbytes &&
71 memcmp(entry_1->stream_name, entry_2->stream_name,
72 entry_1->stream_name_nbytes) == 0;
76 ref_inodes_consistent(const struct wim_inode * restrict ref_inode_1,
77 const struct wim_inode * restrict ref_inode_2)
79 wimlib_assert(ref_inode_1 != ref_inode_2);
81 if (ref_inode_1->i_num_ads != ref_inode_2->i_num_ads)
83 if (ref_inode_1->i_security_id != ref_inode_2->i_security_id
84 || ref_inode_1->i_attributes != ref_inode_2->i_attributes)
86 for (unsigned i = 0; i <= ref_inode_1->i_num_ads; i++) {
87 const u8 *ref_1_hash, *ref_2_hash;
88 ref_1_hash = inode_stream_hash(ref_inode_1, i);
89 ref_2_hash = inode_stream_hash(ref_inode_2, i);
90 if (!hashes_equal(ref_1_hash, ref_2_hash))
92 if (i && !ads_entries_have_same_name(&ref_inode_1->i_ads_entries[i - 1],
93 &ref_inode_2->i_ads_entries[i - 1]))
100 /* Returns true iff the specified inode has any data streams with nonzero hash.
103 inode_has_data_streams(const struct wim_inode *inode)
105 for (unsigned i = 0; i <= inode->i_num_ads; i++)
106 if (!is_zero_hash(inode_stream_hash(inode, i)))
111 /* Returns true iff the specified dentry has any data streams with nonzero hash.
114 dentry_has_data_streams(const struct wim_dentry *dentry)
116 return inode_has_data_streams(dentry->d_inode);
120 inodes_consistent(const struct wim_inode *ref_inode,
121 const struct wim_inode *inode)
123 if (ref_inode->i_security_id != inode->i_security_id) {
124 ERROR("Security ID mismatch: %d != %d",
125 ref_inode->i_security_id, inode->i_security_id);
129 if (ref_inode->i_attributes != inode->i_attributes) {
130 ERROR("Attributes mismatch: 0x%08x != 0x%08x",
131 ref_inode->i_attributes, inode->i_attributes);
135 if (inode_has_data_streams(inode)) {
136 if (ref_inode->i_num_ads != inode->i_num_ads) {
137 ERROR("Stream count mismatch: %u != %u",
138 ref_inode->i_num_ads, inode->i_num_ads);
141 for (unsigned i = 0; i <= ref_inode->i_num_ads; i++) {
142 const u8 *ref_hash, *hash;
144 ref_hash = inode_stream_hash(ref_inode, i);
145 hash = inode_stream_hash(inode, i);
146 if (!hashes_equal(ref_hash, hash) && !is_zero_hash(hash)) {
147 ERROR("Stream hash mismatch");
150 if (i && !ads_entries_have_same_name(&ref_inode->i_ads_entries[i - 1],
151 &inode->i_ads_entries[i - 1]))
153 ERROR("Stream name mismatch");
161 /* Fix up a "true" inode and check for inconsistencies */
163 fix_true_inode(struct wim_inode *inode, struct list_head *inode_list)
165 struct wim_dentry *dentry;
166 struct wim_dentry *ref_dentry = NULL;
167 struct wim_inode *ref_inode;
172 inode_for_each_dentry(dentry, inode) {
173 if (!ref_dentry || dentry->d_inode->i_num_ads > ref_dentry->d_inode->i_num_ads)
175 if (dentry->d_inode->i_creation_time > last_ctime)
176 last_ctime = dentry->d_inode->i_creation_time;
177 if (dentry->d_inode->i_last_write_time > last_mtime)
178 last_mtime = dentry->d_inode->i_last_write_time;
179 if (dentry->d_inode->i_last_access_time > last_atime)
180 last_atime = dentry->d_inode->i_last_access_time;
183 ref_inode = ref_dentry->d_inode;
184 wimlib_assert(ref_inode->i_nlink == 1);
185 list_add_tail(&ref_inode->i_list, inode_list);
187 list_del(&inode->i_dentry);
188 list_add(&ref_inode->i_dentry, &ref_dentry->d_alias);
190 inode_for_each_dentry(dentry, ref_inode) {
191 if (dentry != ref_dentry) {
192 if (!inodes_consistent(ref_inode, dentry->d_inode)) {
193 inconsistent_inode(ref_inode);
194 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
196 /* Free the unneeded `struct wim_inode'. */
197 wimlib_assert(dentry->d_inode->i_nlink == 1);
198 free_inode(dentry->d_inode);
199 dentry->d_inode = ref_inode;
200 ref_inode->i_nlink++;
203 ref_inode->i_creation_time = last_ctime;
204 ref_inode->i_last_write_time = last_mtime;
205 ref_inode->i_last_access_time = last_atime;
206 wimlib_assert(inode_link_count(ref_inode) == ref_inode->i_nlink);
211 * Fixes up a nominal inode.
213 * By a nominal inode we mean a group of two or more dentries that share the
214 * same hard link group ID.
216 * If dentries in the inode are found to be inconsistent, we may split the inode
217 * into several "true" inodes.
219 * After splitting up each nominal inode into the "true" inodes we will
220 * canonicalize the link group by getting rid of all the unnecessary `struct
221 * wim_inode's. There will be just one `struct wim_inode' for each hard link
225 fix_nominal_inode(struct wim_inode *inode, struct list_head *inode_list,
226 bool *ino_changes_needed)
228 struct wim_dentry *dentry;
229 struct hlist_node *cur, *tmp;
231 size_t num_true_inodes;
233 LIST_HEAD(dentries_with_data_streams);
234 LIST_HEAD(dentries_with_no_data_streams);
235 HLIST_HEAD(true_inodes);
237 /* Create a list of dentries in the nominal inode that have at
238 * least one data stream with a non-zero hash, and another list that
239 * contains the dentries that have a zero hash for all data streams. */
240 inode_for_each_dentry(dentry, inode) {
241 if (dentry_has_data_streams(dentry))
242 list_add(&dentry->tmp_list, &dentries_with_data_streams);
244 list_add(&dentry->tmp_list, &dentries_with_no_data_streams);
247 /* If there are no dentries with data streams, we require the nominal
248 * inode to be a true inode */
249 if (list_empty(&dentries_with_data_streams)) {
251 unsigned nominal_group_size = inode_link_count(inode);
252 if (nominal_group_size > 1) {
253 DEBUG("Found link group of size %u without "
254 "any data streams:", nominal_group_size);
255 print_inode_dentries(inode);
256 DEBUG("We are going to interpret it as true "
257 "link group, provided that the dentries "
261 return fix_true_inode(inode, inode_list);
264 /* One or more dentries had data streams specified. We check each of
265 * these dentries for consistency with the others to form a set of true
268 list_for_each_entry(dentry, &dentries_with_data_streams, tmp_list) {
269 /* Look for a true inode that is consistent with this dentry and
270 * add this dentry to it. Or, if none of the true inodes are
271 * consistent with this dentry, add a new one (if that happens,
272 * we have split the hard link group). */
273 hlist_for_each_entry(inode, cur, &true_inodes, i_hlist) {
274 if (ref_inodes_consistent(inode, dentry->d_inode)) {
275 inode_add_dentry(dentry, inode);
280 INIT_LIST_HEAD(&dentry->d_inode->i_dentry);
281 inode_add_dentry(dentry, dentry->d_inode);
282 hlist_add_head(&dentry->d_inode->i_hlist, &true_inodes);
287 wimlib_assert(num_true_inodes != 0);
289 /* If there were dentries with no data streams, we require there to only
290 * be one true inode so that we know which inode to assign the
291 * streamless dentries to. */
292 if (!list_empty(&dentries_with_no_data_streams)) {
293 if (num_true_inodes != 1) {
294 ERROR("Hard link ambiguity detected!");
295 ERROR("We split up inode 0x%"PRIx64" due to "
296 "inconsistencies,", inode->i_ino);
297 ERROR("but dentries with no stream information remained. "
298 "We don't know which inode");
299 ERROR("to assign them to.");
300 ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE;
301 goto out_cleanup_true_inode_list;
303 inode = container_of(true_inodes.first, struct wim_inode, i_hlist);
304 /* Assign the streamless dentries to the one and only true
306 list_for_each_entry(dentry, &dentries_with_no_data_streams, tmp_list)
307 inode_add_dentry(dentry, inode);
309 if (num_true_inodes != 1) {
311 inode = container_of(true_inodes.first, struct wim_inode, i_hlist);
313 tprintf(T("Split nominal inode 0x%"PRIx64" into %zu "
314 "inodes:\n"), inode->i_ino, num_true_inodes);
315 tputs(T("----------------------------------------------------"
316 "--------------------------"));
318 hlist_for_each_entry(inode, cur, &true_inodes, i_hlist) {
319 tprintf(T("[Split inode %zu]\n"), i++);
320 print_inode_dentries(inode);
323 tputs(T("----------------------------------------------------"
324 "--------------------------"));
326 *ino_changes_needed = true;
329 hlist_for_each_entry_safe(inode, cur, tmp, &true_inodes, i_hlist) {
330 hlist_del_init(&inode->i_hlist);
331 ret = fix_true_inode(inode, inode_list);
333 goto out_cleanup_true_inode_list;
337 out_cleanup_true_inode_list:
338 hlist_for_each_entry_safe(inode, cur, tmp, &true_inodes, i_hlist)
339 hlist_del_init(&inode->i_hlist);
345 fix_inodes(struct wim_inode_table *table, struct list_head *inode_list,
346 bool *ino_changes_needed)
348 struct wim_inode *inode;
349 struct hlist_node *cur, *tmp;
351 INIT_LIST_HEAD(inode_list);
352 for (u64 i = 0; i < table->capacity; i++) {
353 hlist_for_each_entry_safe(inode, cur, tmp, &table->array[i], i_hlist) {
354 hlist_del_init(&inode->i_hlist);
355 ret = fix_nominal_inode(inode, inode_list, ino_changes_needed);
360 list_splice_tail(&table->extra_inodes, inode_list);
364 /* Insert a dentry into the inode table based on the inode number of the
365 * attached inode (which came from the hard link group ID field of the on-disk
368 inode_table_insert(struct wim_dentry *dentry, void *_table)
370 struct wim_inode_table *table = _table;
371 struct wim_inode *d_inode = dentry->d_inode;
373 if (d_inode->i_ino == 0) {
374 /* A dentry with a hard link group ID of 0 indicates that it's
375 * in a hard link group by itself. Add it to the list of extra
376 * inodes rather than inserting it into the hash lists. */
377 list_add_tail(&d_inode->i_list, &table->extra_inodes);
380 struct wim_inode *inode;
381 struct hlist_node *cur;
383 /* Try adding this dentry to an existing inode */
384 pos = d_inode->i_ino % table->capacity;
385 hlist_for_each_entry(inode, cur, &table->array[pos], i_hlist) {
386 if (inode->i_ino == d_inode->i_ino) {
387 if (unlikely((inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY) ||
388 (d_inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY)))
390 ERROR("Unsupported directory hard link "
391 "\"%"TS"\" <=> \"%"TS"\"",
392 dentry_full_path(dentry),
393 dentry_full_path(inode_first_dentry(inode)));
394 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
396 inode_add_dentry(dentry, inode);
401 /* No inode in the table has the same number as this one, so add
402 * it to the table. */
403 hlist_add_head(&d_inode->i_hlist, &table->array[pos]);
405 /* XXX Make the table grow when too many entries have been
407 table->num_entries++;
414 * dentry_tree_fix_inodes():
416 * This function takes as input a tree of WIM dentries that initially has a
417 * different inode associated with each dentry. Sets of dentries that should
418 * share the same inode (a.k.a. hard link groups) are built using the i_ino
419 * field of each inode, then the link count and alias list for one inode in each
420 * set is set correctly and the unnecessary struct wim_inode's freed. The
421 * effect is to correctly associate exactly one struct wim_inode with each
422 * original inode, regardless of how many dentries are aliases for that inode.
424 * The special inode number of 0 indicates that the dentry is in a hard link
425 * group by itself, and therefore has a 'struct wim_inode' with i_nlink=1 to
428 * This function also checks the dentries in each hard link group for
429 * consistency. In some WIMs, such as install.wim for some versions of Windows
430 * 7, dentries can share the same hard link group ID but not actually be hard
431 * linked to each other (based on conflicting information, such as file
432 * contents). This should be an error, but this case needs be handled. So,
433 * each "nominal" inode (the inode based on the inode numbers provided in the
434 * WIM) is examined for consistency and may be split into multiple "true" inodes
435 * that are maximally sized consistent sets of dentries.
437 * On success, the list of "true" inodes, linked by the i_hlist field,
438 * is returned in the hlist @inode_list.
441 * WIMLIB_ERR_SUCCESS (0)
442 * WIMLIB_ERR_INVALID_METADATA_RESOURCE
446 dentry_tree_fix_inodes(struct wim_dentry *root, struct list_head *inode_list)
448 struct wim_inode_table inode_tab;
450 bool ino_changes_needed;
451 struct wim_inode *inode;
453 DEBUG("Inserting dentries into inode table");
454 ret = init_inode_table(&inode_tab, 9001);
458 ret = for_dentry_in_tree(root, inode_table_insert, &inode_tab);
460 goto out_destroy_inode_table;
462 DEBUG("Cleaning up the hard link groups");
463 ino_changes_needed = false;
464 ret = fix_inodes(&inode_tab, inode_list, &ino_changes_needed);
466 goto out_destroy_inode_table;
468 if (ino_changes_needed) {
471 WARNING("The WIM image contains invalid hard links. Fixing.");
473 list_for_each_entry(inode, inode_list, i_list) {
474 if (inode->i_nlink > 1)
475 inode->i_ino = cur_ino++;
480 /* On success, all the inodes have been moved to the image inode list,
481 * so there's no need to delete from from the hash lists in the inode
482 * table before freeing the hash buckets array directly. */
484 goto out_destroy_inode_table_raw;
485 out_destroy_inode_table:
486 for (size_t i = 0; i < inode_tab.capacity; i++) {
487 struct hlist_node *cur, *tmp;
488 hlist_for_each_entry_safe(inode, cur, tmp, &inode_tab.array[i], i_hlist)
489 hlist_del_init(&inode->i_hlist);
492 struct wim_inode *tmp;
493 list_for_each_entry_safe(inode, tmp, &inode_tab.extra_inodes, i_list)
494 list_del_init(&inode->i_list);
496 out_destroy_inode_table_raw:
497 destroy_inode_table(&inode_tab);