4 * Code to deal with hard links in WIMs. Essentially, the WIM dentries are put
5 * into a hash table indexed by the inode ID field, then for each hard
6 * inode, a linked list is made to connect the dentries.
10 * Copyright (C) 2012 Eric Biggers
12 * This file is part of wimlib, a library for working with WIM files.
14 * wimlib is free software; you can redistribute it and/or modify it under the
15 * terms of the GNU General Public License as published by the Free
16 * Software Foundation; either version 3 of the License, or (at your option)
19 * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
20 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
21 * A PARTICULAR PURPOSE. See the GNU General Public License for more
24 * You should have received a copy of the GNU General Public License
25 * along with wimlib; if not, see http://www.gnu.org/licenses/.
28 #include "wimlib_internal.h"
31 #include "lookup_table.h"
36 * / \ ----------- -----------
37 * | dentry<---| struct | | struct |---> dentry
38 * \ / | inode | | inode |
39 * dentry ------------ ------------
43 * ----------- ----------- / \
44 * dentry<---| struct | | struct |---> dentry dentry
45 * / | inode | | inode | \ /
46 * dentry ------------ ------------ dentry
50 * inode_table->array | idx 0 | idx 1 |
54 /* Hash table to find inodes, identified by their inode ID.
57 /* Fields for the hash table */
58 struct hlist_head *array;
63 * Linked list of "extra" inodes. These may be:
65 * - inodes with link count 1, which are all allowed to have 0 for their
66 * inode number, meaning we cannot insert them into the hash table
67 * before calling assign_inode_numbers().
69 * - Groups we create ourselves by splitting a nominal inode due to
70 * inconsistencies in the dentries. These inodes will share a inode
71 * ID with some other inode until assign_inode_numbers() is called.
73 struct hlist_head extra_inodes;
76 static inline void destroy_inode_table(struct inode_table *table)
81 static int init_inode_table(struct inode_table *table, size_t capacity)
83 table->array = CALLOC(capacity, sizeof(table->array[0]));
85 ERROR("Cannot initalize inode table: out of memory");
86 return WIMLIB_ERR_NOMEM;
88 table->num_entries = 0;
89 table->capacity = capacity;
90 INIT_HLIST_HEAD(&table->extra_inodes);
94 static inline size_t inode_link_count(const struct inode *inode)
96 const struct list_head *cur;
98 list_for_each(cur, &inode->dentry_list)
104 * Insert a dentry into the inode table based on its inode
107 * If there is already a dentry in the table having the same inode ID,
108 * and the inode ID is not 0, the dentry is added to the circular
109 * linked list for that inode.
111 * If the inode ID is 0, this indicates a dentry that's in a hard link
112 * inode by itself (has a link count of 1). We can't insert it into the hash
113 * table itself because we don't know what inode numbers are available to
114 * give it (this could be kept track of but would be more difficult). Instead
115 * we keep a linked list of the single dentries, and assign them inode
118 static int inode_table_insert(struct dentry *dentry, void *__table)
120 struct inode_table *table = __table;
121 struct inode *d_inode = dentry->d_inode;
123 if (d_inode->ino == 0) {
124 /* Single inode--- Add to the list of extra inodes (we can't put
125 * it in the table itself because all the singles have a link
127 hlist_add_head(&d_inode->hlist, &table->extra_inodes);
129 wimlib_assert(d_inode->dentry_list.next == &dentry->inode_dentry_list);
130 wimlib_assert(d_inode->dentry_list.prev == &dentry->inode_dentry_list);
131 wimlib_assert(d_inode->link_count == 1);
133 /* Inode that may have multiple corresponding dentries (the code
134 * will work even if the inode actually contains only 1 dentry
139 struct hlist_node *cur;
141 /* Try adding to existing inode */
142 pos = d_inode->ino % table->capacity;
143 hlist_for_each_entry(inode, cur, &table->array[pos], hlist) {
144 if (inode->ino == d_inode->ino) {
145 inode_add_dentry(dentry, inode);
151 /* Add new inode to the table */
152 hlist_add_head(&d_inode->hlist, &table->array[pos]);
154 wimlib_assert(d_inode->dentry_list.next == &dentry->inode_dentry_list);
155 wimlib_assert(d_inode->dentry_list.prev == &dentry->inode_dentry_list);
156 wimlib_assert(d_inode->link_count == 1);
158 /* XXX Make the table grow when too many entries have been
160 table->num_entries++;
165 /* Assign the inode numbers to dentries in a inode table, and return the
166 * next available inode ID. */
167 u64 assign_inode_numbers(struct hlist_head *inode_list)
169 DEBUG("Assigning inode numbers");
171 struct hlist_node *cur;
173 hlist_for_each_entry(inode, cur, inode_list, hlist) {
174 inode->ino = cur_ino;
181 static void print_inode_dentries(const struct inode *inode)
183 struct dentry *dentry;
184 inode_for_each_dentry(dentry, inode)
185 printf("`%s'\n", dentry->full_path_utf8);
188 static void inconsistent_inode(const struct inode *inode)
190 ERROR("An inconsistent hard link group that cannot be corrected has "
192 ERROR("The dentries are located at the following paths:");
193 #ifdef ENABLE_ERROR_MESSAGES
194 print_inode_dentries(inode);
198 static bool ref_inodes_consistent(const struct inode * restrict ref_inode_1,
199 const struct inode * restrict ref_inode_2)
201 wimlib_assert(ref_inode_1 != ref_inode_2);
203 if (ref_inode_1->num_ads != ref_inode_2->num_ads)
205 if (ref_inode_1->security_id != ref_inode_2->security_id
206 || ref_inode_1->attributes != ref_inode_2->attributes)
208 for (unsigned i = 0; i <= ref_inode_1->num_ads; i++) {
209 const u8 *ref_1_hash, *ref_2_hash;
210 ref_1_hash = inode_stream_hash(ref_inode_1, i);
211 ref_2_hash = inode_stream_hash(ref_inode_2, i);
212 if (!hashes_equal(ref_1_hash, ref_2_hash))
214 if (i && !ads_entries_have_same_name(&ref_inode_1->ads_entries[i - 1],
215 &ref_inode_2->ads_entries[i - 1]))
222 static bool inodes_consistent(const struct inode * restrict ref_inode,
223 const struct inode * restrict inode)
225 wimlib_assert(ref_inode != inode);
227 if (ref_inode->num_ads != inode->num_ads &&
230 if (ref_inode->security_id != inode->security_id
231 || ref_inode->attributes != inode->attributes)
233 for (unsigned i = 0; i <= min(ref_inode->num_ads, inode->num_ads); i++) {
234 const u8 *ref_hash, *hash;
235 ref_hash = inode_stream_hash(ref_inode, i);
236 hash = inode_stream_hash(inode, i);
237 if (!hashes_equal(ref_hash, hash) && !is_zero_hash(hash))
239 if (i && !ads_entries_have_same_name(&ref_inode->ads_entries[i - 1],
240 &inode->ads_entries[i - 1]))
246 /* Fix up a "true" inode and check for inconsistencies */
247 static int fix_true_inode(struct inode *inode, struct hlist_head *inode_list)
249 struct dentry *dentry;
250 struct dentry *ref_dentry = NULL;
251 struct inode *ref_inode;
256 inode_for_each_dentry(dentry, inode) {
257 if (!ref_dentry || dentry->d_inode->num_ads > ref_dentry->d_inode->num_ads)
259 if (dentry->d_inode->creation_time > last_ctime)
260 last_ctime = dentry->d_inode->creation_time;
261 if (dentry->d_inode->last_write_time > last_mtime)
262 last_mtime = dentry->d_inode->last_write_time;
263 if (dentry->d_inode->last_access_time > last_atime)
264 last_atime = dentry->d_inode->last_access_time;
267 ref_inode = ref_dentry->d_inode;
268 ref_inode->link_count = 1;
269 hlist_add_head(&ref_inode->hlist, inode_list);
271 list_del(&inode->dentry_list);
272 list_add(&ref_inode->dentry_list, &ref_dentry->inode_dentry_list);
274 inode_for_each_dentry(dentry, ref_inode) {
275 if (dentry != ref_dentry) {
276 if (!inodes_consistent(ref_inode, dentry->d_inode)) {
277 inconsistent_inode(ref_inode);
278 return WIMLIB_ERR_INVALID_DENTRY;
280 /* Free the unneeded `struct inode'. */
281 free_inode(dentry->d_inode);
282 dentry->d_inode = ref_inode;
283 ref_inode->link_count++;
286 ref_inode->creation_time = last_ctime;
287 ref_inode->last_write_time = last_mtime;
288 ref_inode->last_access_time = last_atime;
289 wimlib_assert(inode_link_count(ref_inode) == ref_inode->link_count);
294 * Fixes up a nominal inode.
296 * By a nominal inode we mean a group of two or more dentries that share
297 * the same hard link group ID.
299 * If dentries in the inode are found to be inconsistent, we may split the inode
300 * into several "true" inodes.
302 * After splitting up each nominal inode into the "true" inodes we will
303 * canonicalize the link group by getting rid of all the unnecessary `struct
304 * inodes'. There will be just one `struct inode' for each hard link group
307 static int fix_nominal_inode(struct inode *inode,
308 struct hlist_head *inode_list)
310 struct dentry *dentry;
311 struct hlist_node *cur, *tmp;
313 size_t num_true_inodes;
315 wimlib_assert(inode->link_count == inode_link_count(inode));
317 LIST_HEAD(dentries_with_data_streams);
318 LIST_HEAD(dentries_with_no_data_streams);
319 HLIST_HEAD(true_inodes);
321 /* Create a list of dentries in the nominal inode that have at
322 * least one data stream with a non-zero hash, and another list that
323 * contains the dentries that have a zero hash for all data streams. */
324 inode_for_each_dentry(dentry, inode) {
325 for (unsigned i = 0; i <= dentry->d_inode->num_ads; i++) {
327 hash = inode_stream_hash(dentry->d_inode, i);
328 if (!is_zero_hash(hash)) {
329 list_add(&dentry->tmp_list,
330 &dentries_with_data_streams);
334 list_add(&dentry->tmp_list,
335 &dentries_with_no_data_streams);
340 /* If there are no dentries with data streams, we require the nominal
341 * inode to be a true inode */
342 if (list_empty(&dentries_with_data_streams)) {
344 if (inode->link_count > 1) {
345 DEBUG("Found link group of size %u without "
346 "any data streams:", inode->link_count);
347 print_inode_dentries(inode);
348 DEBUG("We are going to interpret it as true "
349 "link group, provided that the dentries "
353 return fix_true_inode(inode, inode_list);
356 /* One or more dentries had data streams specified. We check each of
357 * these dentries for consistency with the others to form a set of true
360 list_for_each_entry(dentry, &dentries_with_data_streams, tmp_list) {
361 /* Look for a true inode that is consistent with this dentry and
362 * add this dentry to it. Or, if none of the true inodes are
363 * consistent with this dentry, add a new one (if that happens,
364 * we have split the hard link group). */
365 hlist_for_each_entry(inode, cur, &true_inodes, hlist) {
366 if (ref_inodes_consistent(inode, dentry->d_inode)) {
367 inode_add_dentry(dentry, inode);
372 INIT_LIST_HEAD(&dentry->d_inode->dentry_list);
373 inode_add_dentry(dentry, dentry->d_inode);
374 hlist_add_head(&dentry->d_inode->hlist, &true_inodes);
379 wimlib_assert(num_true_inodes != 0);
381 /* If there were dentries with no data streams, we require there to only
382 * be one true inode so that we know which inode to assign the
383 * streamless dentries to. */
384 if (!list_empty(&dentries_with_no_data_streams)) {
385 if (num_true_inodes != 1) {
386 ERROR("Hard inode ambiguity detected!");
387 ERROR("We split up inode 0x%"PRIx64" due to "
388 "inconsistencies,", inode->ino);
389 ERROR("but dentries with no stream information remained. "
390 "We don't know which inode");
391 ERROR("to assign them to.");
392 return WIMLIB_ERR_INVALID_DENTRY;
394 inode = container_of(true_inodes.first, struct inode, hlist);
395 /* Assign the streamless dentries to the one and only true
397 list_for_each_entry(dentry, &dentries_with_no_data_streams, tmp_list)
398 inode_add_dentry(dentry, inode);
401 if (num_true_inodes != 1) {
402 inode = container_of(true_inodes.first, struct inode, hlist);
404 printf("Split nominal inode 0x%"PRIx64" into %zu "
406 inode->ino, num_true_inodes);
407 puts("------------------------------------------------------------------------------");
409 hlist_for_each_entry(inode, cur, &true_inodes, hlist) {
410 printf("[Split inode %zu]\n", i++);
411 print_inode_dentries(inode);
414 puts("------------------------------------------------------------------------------");
418 hlist_for_each_entry_safe(inode, cur, tmp, &true_inodes, hlist) {
419 ret = fix_true_inode(inode, inode_list);
427 * Goes through each hard link group (dentries sharing the same hard link group
428 * ID field) that's been inserted into the inode table and shares the `struct
429 * inode's among members of each hard link group.
431 * In the process, the dentries belonging to each inode are checked for
432 * consistency. If they contain data features that indicate they cannot really
433 * correspond to the same inode, this should be an error, but in reality this
434 * case needs to be handled, so we split the dentries into different inodes.
436 * After this function returns, the inodes are no longer in the inode table, and
437 * the inode table should be destroyed. A list of the inodes, including all
438 * split inodes as well as the inodes that were good before, is returned in the
441 static int fix_inodes(struct inode_table *table, struct hlist_head *inode_list)
444 struct hlist_node *cur, *tmp;
446 INIT_HLIST_HEAD(inode_list);
447 for (u64 i = 0; i < table->capacity; i++) {
448 hlist_for_each_entry_safe(inode, cur, tmp, &table->array[i], hlist) {
449 ret = fix_nominal_inode(inode, inode_list);
454 hlist_for_each_safe(cur, tmp, &table->extra_inodes)
455 hlist_add_head(cur, inode_list);
459 int dentry_tree_fix_inodes(struct dentry *root, struct hlist_head *inode_list)
461 struct inode_table inode_tab;
464 DEBUG("Inserting dentries into inode table");
465 ret = init_inode_table(&inode_tab, 9001);
469 for_dentry_in_tree(root, inode_table_insert, &inode_tab);
471 DEBUG("Cleaning up the hard link groups");
472 ret = fix_inodes(&inode_tab, inode_list);
473 destroy_inode_table(&inode_tab);