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]))
101 inodes_consistent(const struct wim_inode * restrict ref_inode,
102 const struct wim_inode * restrict inode)
104 wimlib_assert(ref_inode != inode);
106 if (ref_inode->i_num_ads != inode->i_num_ads &&
107 inode->i_num_ads != 0)
109 if (ref_inode->i_security_id != inode->i_security_id
110 || ref_inode->i_attributes != inode->i_attributes)
112 for (unsigned i = 0; i <= min(ref_inode->i_num_ads, inode->i_num_ads); i++) {
113 const u8 *ref_hash, *hash;
114 ref_hash = inode_stream_hash(ref_inode, i);
115 hash = inode_stream_hash(inode, i);
116 if (!hashes_equal(ref_hash, hash) && !is_zero_hash(hash))
118 if (i && !ads_entries_have_same_name(&ref_inode->i_ads_entries[i - 1],
119 &inode->i_ads_entries[i - 1]))
125 /* Fix up a "true" inode and check for inconsistencies */
127 fix_true_inode(struct wim_inode *inode, struct list_head *inode_list)
129 struct wim_dentry *dentry;
130 struct wim_dentry *ref_dentry = NULL;
131 struct wim_inode *ref_inode;
136 inode_for_each_dentry(dentry, inode) {
137 if (!ref_dentry || dentry->d_inode->i_num_ads > ref_dentry->d_inode->i_num_ads)
139 if (dentry->d_inode->i_creation_time > last_ctime)
140 last_ctime = dentry->d_inode->i_creation_time;
141 if (dentry->d_inode->i_last_write_time > last_mtime)
142 last_mtime = dentry->d_inode->i_last_write_time;
143 if (dentry->d_inode->i_last_access_time > last_atime)
144 last_atime = dentry->d_inode->i_last_access_time;
147 ref_inode = ref_dentry->d_inode;
148 wimlib_assert(ref_inode->i_nlink == 1);
149 list_add_tail(&ref_inode->i_list, inode_list);
151 list_del(&inode->i_dentry);
152 list_add(&ref_inode->i_dentry, &ref_dentry->d_alias);
154 inode_for_each_dentry(dentry, ref_inode) {
155 if (dentry != ref_dentry) {
156 if (!inodes_consistent(ref_inode, dentry->d_inode)) {
157 inconsistent_inode(ref_inode);
158 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
160 /* Free the unneeded `struct wim_inode'. */
161 wimlib_assert(dentry->d_inode->i_nlink == 1);
162 free_inode(dentry->d_inode);
163 dentry->d_inode = ref_inode;
164 ref_inode->i_nlink++;
167 ref_inode->i_creation_time = last_ctime;
168 ref_inode->i_last_write_time = last_mtime;
169 ref_inode->i_last_access_time = last_atime;
170 wimlib_assert(inode_link_count(ref_inode) == ref_inode->i_nlink);
175 * Fixes up a nominal inode.
177 * By a nominal inode we mean a group of two or more dentries that share the
178 * same hard link group ID.
180 * If dentries in the inode are found to be inconsistent, we may split the inode
181 * into several "true" inodes.
183 * After splitting up each nominal inode into the "true" inodes we will
184 * canonicalize the link group by getting rid of all the unnecessary `struct
185 * wim_inode's. There will be just one `struct wim_inode' for each hard link
189 fix_nominal_inode(struct wim_inode *inode, struct list_head *inode_list,
190 bool *ino_changes_needed)
192 struct wim_dentry *dentry;
193 struct hlist_node *cur, *tmp;
195 size_t num_true_inodes;
197 LIST_HEAD(dentries_with_data_streams);
198 LIST_HEAD(dentries_with_no_data_streams);
199 HLIST_HEAD(true_inodes);
201 /* Create a list of dentries in the nominal inode that have at
202 * least one data stream with a non-zero hash, and another list that
203 * contains the dentries that have a zero hash for all data streams. */
204 inode_for_each_dentry(dentry, inode) {
205 for (unsigned i = 0; i <= dentry->d_inode->i_num_ads; i++) {
207 hash = inode_stream_hash(dentry->d_inode, i);
208 if (!is_zero_hash(hash)) {
209 list_add(&dentry->tmp_list,
210 &dentries_with_data_streams);
214 list_add(&dentry->tmp_list,
215 &dentries_with_no_data_streams);
220 /* If there are no dentries with data streams, we require the nominal
221 * inode to be a true inode */
222 if (list_empty(&dentries_with_data_streams)) {
224 unsigned nominal_group_size = inode_link_count(inode);
225 if (nominal_group_size > 1) {
226 DEBUG("Found link group of size %u without "
227 "any data streams:", nominal_group_size);
228 print_inode_dentries(inode);
229 DEBUG("We are going to interpret it as true "
230 "link group, provided that the dentries "
234 return fix_true_inode(inode, inode_list);
237 /* One or more dentries had data streams specified. We check each of
238 * these dentries for consistency with the others to form a set of true
241 list_for_each_entry(dentry, &dentries_with_data_streams, tmp_list) {
242 /* Look for a true inode that is consistent with this dentry and
243 * add this dentry to it. Or, if none of the true inodes are
244 * consistent with this dentry, add a new one (if that happens,
245 * we have split the hard link group). */
246 hlist_for_each_entry(inode, cur, &true_inodes, i_hlist) {
247 if (ref_inodes_consistent(inode, dentry->d_inode)) {
248 inode_add_dentry(dentry, inode);
253 INIT_LIST_HEAD(&dentry->d_inode->i_dentry);
254 inode_add_dentry(dentry, dentry->d_inode);
255 hlist_add_head(&dentry->d_inode->i_hlist, &true_inodes);
260 wimlib_assert(num_true_inodes != 0);
262 /* If there were dentries with no data streams, we require there to only
263 * be one true inode so that we know which inode to assign the
264 * streamless dentries to. */
265 if (!list_empty(&dentries_with_no_data_streams)) {
266 if (num_true_inodes != 1) {
267 ERROR("Hard link ambiguity detected!");
268 ERROR("We split up inode 0x%"PRIx64" due to "
269 "inconsistencies,", inode->i_ino);
270 ERROR("but dentries with no stream information remained. "
271 "We don't know which inode");
272 ERROR("to assign them to.");
273 ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE;
274 goto out_cleanup_true_inode_list;
276 inode = container_of(true_inodes.first, struct wim_inode, i_hlist);
277 /* Assign the streamless dentries to the one and only true
279 list_for_each_entry(dentry, &dentries_with_no_data_streams, tmp_list)
280 inode_add_dentry(dentry, inode);
282 if (num_true_inodes != 1) {
284 inode = container_of(true_inodes.first, struct wim_inode, i_hlist);
286 tprintf(T("Split nominal inode 0x%"PRIx64" into %zu "
287 "inodes:\n"), inode->i_ino, num_true_inodes);
288 tputs(T("----------------------------------------------------"
289 "--------------------------"));
291 hlist_for_each_entry(inode, cur, &true_inodes, i_hlist) {
292 tprintf(T("[Split inode %zu]\n"), i++);
293 print_inode_dentries(inode);
296 tputs(T("----------------------------------------------------"
297 "--------------------------"));
299 *ino_changes_needed = true;
302 hlist_for_each_entry_safe(inode, cur, tmp, &true_inodes, i_hlist) {
303 hlist_del_init(&inode->i_hlist);
304 ret = fix_true_inode(inode, inode_list);
306 goto out_cleanup_true_inode_list;
310 out_cleanup_true_inode_list:
311 hlist_for_each_entry_safe(inode, cur, tmp, &true_inodes, i_hlist)
312 hlist_del_init(&inode->i_hlist);
318 fix_inodes(struct wim_inode_table *table, struct list_head *inode_list,
319 bool *ino_changes_needed)
321 struct wim_inode *inode;
322 struct hlist_node *cur, *tmp;
324 INIT_LIST_HEAD(inode_list);
325 for (u64 i = 0; i < table->capacity; i++) {
326 hlist_for_each_entry_safe(inode, cur, tmp, &table->array[i], i_hlist) {
327 hlist_del_init(&inode->i_hlist);
328 ret = fix_nominal_inode(inode, inode_list, ino_changes_needed);
333 list_splice_tail(&table->extra_inodes, inode_list);
337 /* Insert a dentry into the inode table based on the inode number of the
338 * attached inode (which came from the hard link group ID field of the on-disk
341 inode_table_insert(struct wim_dentry *dentry, void *_table)
343 struct wim_inode_table *table = _table;
344 struct wim_inode *d_inode = dentry->d_inode;
346 if (d_inode->i_ino == 0) {
347 /* A dentry with a hard link group ID of 0 indicates that it's
348 * in a hard link group by itself. Add it to the list of extra
349 * inodes rather than inserting it into the hash lists. */
350 list_add_tail(&d_inode->i_list, &table->extra_inodes);
353 struct wim_inode *inode;
354 struct hlist_node *cur;
356 /* Try adding this dentry to an existing inode */
357 pos = d_inode->i_ino % table->capacity;
358 hlist_for_each_entry(inode, cur, &table->array[pos], i_hlist) {
359 if (inode->i_ino == d_inode->i_ino) {
360 if (unlikely((inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY) ||
361 (d_inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY)))
363 ERROR("Unsupported directory hard link "
364 "\"%"TS"\" <=> \"%"TS"\"",
365 dentry_full_path(dentry),
366 dentry_full_path(inode_first_dentry(inode)));
367 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
369 inode_add_dentry(dentry, inode);
374 /* No inode in the table has the same number as this one, so add
375 * it to the table. */
376 hlist_add_head(&d_inode->i_hlist, &table->array[pos]);
378 /* XXX Make the table grow when too many entries have been
380 table->num_entries++;
387 * dentry_tree_fix_inodes():
389 * This function takes as input a tree of WIM dentries that initially has a
390 * different inode associated with each dentry. Sets of dentries that should
391 * share the same inode (a.k.a. hard link groups) are built using the i_ino
392 * field of each inode, then the link count and alias list for one inode in each
393 * set is set correctly and the unnecessary struct wim_inode's freed. The
394 * effect is to correctly associate exactly one struct wim_inode with each
395 * original inode, regardless of how many dentries are aliases for that inode.
397 * The special inode number of 0 indicates that the dentry is in a hard link
398 * group by itself, and therefore has a 'struct wim_inode' with i_nlink=1 to
401 * This function also checks the dentries in each hard link group for
402 * consistency. In some WIMs, such as install.wim for some versions of Windows
403 * 7, dentries can share the same hard link group ID but not actually be hard
404 * linked to each other (based on conflicting information, such as file
405 * contents). This should be an error, but this case needs be handled. So,
406 * each "nominal" inode (the inode based on the inode numbers provided in the
407 * WIM) is examined for consistency and may be split into multiple "true" inodes
408 * that are maximally sized consistent sets of dentries.
410 * On success, the list of "true" inodes, linked by the i_hlist field,
411 * is returned in the hlist @inode_list.
414 * WIMLIB_ERR_SUCCESS (0)
415 * WIMLIB_ERR_INVALID_METADATA_RESOURCE
419 dentry_tree_fix_inodes(struct wim_dentry *root, struct list_head *inode_list)
421 struct wim_inode_table inode_tab;
423 bool ino_changes_needed;
424 struct wim_inode *inode;
426 DEBUG("Inserting dentries into inode table");
427 ret = init_inode_table(&inode_tab, 9001);
431 ret = for_dentry_in_tree(root, inode_table_insert, &inode_tab);
433 goto out_destroy_inode_table;
435 DEBUG("Cleaning up the hard link groups");
436 ino_changes_needed = false;
437 ret = fix_inodes(&inode_tab, inode_list, &ino_changes_needed);
439 goto out_destroy_inode_table;
441 if (ino_changes_needed) {
444 WARNING("Re-assigning inode numbers due to inode inconsistencies");
445 list_for_each_entry(inode, inode_list, i_list) {
446 if (inode->i_nlink > 1)
447 inode->i_ino = cur_ino++;
452 /* On success, all the inodes have been moved to the image inode list,
453 * so there's no need to delete from from the hash lists in the inode
454 * table before freeing the hash buckets array directly. */
456 goto out_destroy_inode_table_raw;
457 out_destroy_inode_table:
458 for (size_t i = 0; i < inode_tab.capacity; i++) {
459 struct hlist_node *cur, *tmp;
460 hlist_for_each_entry_safe(inode, cur, tmp, &inode_tab.array[i], i_hlist)
461 hlist_del_init(&inode->i_hlist);
464 struct wim_inode *tmp;
465 list_for_each_entry_safe(inode, tmp, &inode_tab.extra_inodes, i_list)
466 list_del_init(&inode->i_list);
468 out_destroy_inode_table_raw:
469 destroy_inode_table(&inode_tab);