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
39 * dentry ------------ ------------
43 * ----------- ----------- / \
44 * dentry<---| struct | | struct |---> dentry dentry
45 * / |inode| |inode| \ /
46 * dentry ------------ ------------ dentry
50 * inode_table->array | idx 0 | idx 1 |
55 int init_inode_table(struct inode_table *table, size_t capacity)
57 table->array = CALLOC(capacity, sizeof(table->array[0]));
59 ERROR("Cannot initalize inode table: out of memory");
60 return WIMLIB_ERR_NOMEM;
62 table->num_entries = 0;
63 table->capacity = capacity;
64 INIT_HLIST_HEAD(&table->extra_inodes);
69 static size_t inode_link_count(const struct inode *inode)
71 const struct list_head *cur;
73 list_for_each(cur, &inode->dentry_list)
79 * Insert a dentry into the inode table based on its inode
82 * If there is already a dentry in the table having the same inode ID,
83 * and the inode ID is not 0, the dentry is added to the circular
84 * linked list for that inode.
86 * If the inode ID is 0, this indicates a dentry that's in a hard link
87 * inode by itself (has a link count of 1). We can't insert it into the hash
88 * table itself because we don't know what inode numbers are available to
89 * give it (this could be kept track of but would be more difficult). Instead
90 * we keep a linked list of the single dentries, and assign them inode
93 int inode_table_insert(struct dentry *dentry, void *__table)
95 struct inode_table *table = __table;
96 struct inode *d_inode = dentry->inode;
98 if (d_inode->ino == 0) {
99 /* Single inode--- Add to the list of extra inodes (we can't put
100 * it in the table itself because all the singles have a link
102 hlist_add_head(&d_inode->hlist, &table->extra_inodes);
104 wimlib_assert(d_inode->dentry_list.next == &dentry->inode_dentry_list);
105 wimlib_assert(d_inode->dentry_list.prev == &dentry->inode_dentry_list);
106 wimlib_assert(d_inode->link_count == 1);
108 /* Inode that may have multiple corresponding dentries (the code
109 * will work even if the inode actually contains only 1 dentry
114 struct hlist_node *cur;
116 /* Try adding to existing inode */
117 pos = d_inode->ino % table->capacity;
118 hlist_for_each_entry(inode, cur, &table->array[pos], hlist) {
119 if (inode->ino == d_inode->ino) {
120 list_add(&dentry->inode_dentry_list,
121 &inode->dentry_list);
127 /* Add new inode to the table */
128 hlist_add_head(&d_inode->hlist, &table->array[pos]);
130 wimlib_assert(d_inode->dentry_list.next == &dentry->inode_dentry_list);
131 wimlib_assert(d_inode->dentry_list.prev == &dentry->inode_dentry_list);
132 wimlib_assert(d_inode->link_count == 1);
134 /* XXX Make the table grow when too many entries have been
136 table->num_entries++;
141 /* Assign the inode numbers to dentries in a inode table, and return the
142 * next available inode ID. */
143 u64 assign_inode_numbers(struct hlist_head *inode_list)
146 struct hlist_node *cur;
148 hlist_for_each_entry(inode, cur, inode_list, hlist) {
149 inode->ino = cur_ino;
157 print_inode_dentries(const struct inode *inode)
159 struct dentry *dentry;
160 list_for_each_entry(dentry, &inode->dentry_list, inode_dentry_list)
161 printf("`%s'\n", dentry->full_path_utf8);
164 static void inconsistent_inode(const struct inode *inode)
166 ERROR("An inconsistent hard link group that we cannot correct has been "
168 ERROR("The dentries are located at the following paths:");
169 print_inode_dentries(inode);
172 static bool ref_inodes_consistent(const struct inode * restrict ref_inode_1,
173 const struct inode * restrict ref_inode_2)
175 wimlib_assert(ref_inode_1 != ref_inode_2);
177 if (ref_inode_1->num_ads != ref_inode_2->num_ads)
179 if (ref_inode_1->security_id != ref_inode_2->security_id
180 || ref_inode_1->attributes != ref_inode_2->attributes)
182 for (unsigned i = 0; i <= ref_inode_1->num_ads; i++) {
183 const u8 *ref_1_hash, *ref_2_hash;
184 ref_1_hash = inode_stream_hash(ref_inode_1, i);
185 ref_2_hash = inode_stream_hash(ref_inode_2, i);
186 if (!hashes_equal(ref_1_hash, ref_2_hash))
188 if (i && !ads_entries_have_same_name(ref_inode_1->ads_entries[i - 1],
189 ref_inode_2->ads_entries[i - 1]))
196 static bool inodes_consistent(const struct inode * restrict ref_inode,
197 const struct inode * restrict inode)
199 wimlib_assert(ref_inode != inode);
201 if (ref_inode->num_ads != inode->num_ads &&
204 if (ref_inode->security_id != inode->security_id
205 || ref_inode->attributes != inode->attributes)
207 for (unsigned i = 0; i <= min(ref_inode->num_ads, inode->num_ads); i++) {
208 const u8 *ref_hash, *hash;
209 ref_hash = inode_stream_hash(ref_inode, i);
210 hash = inode_stream_hash(inode, i);
211 if (!hashes_equal(ref_hash, hash) && !is_zero_hash(hash))
213 if (i && !ads_entries_have_same_name(ref_inode->ads_entries[i - 1],
214 inode->ads_entries[i - 1]))
222 print_dentry_list(const struct dentry *first_dentry)
224 const struct dentry *dentry = first_dentry;
226 printf("`%s'\n", dentry->full_path_utf8);
227 } while ((dentry = container_of(dentry->inode_dentry_list.next,
229 inode_dentry_list)) != first_dentry);
235 /* Fix up a "true" inode and check for inconsistencies */
236 static int fix_true_inode(struct inode *inode)
238 struct dentry *dentry;
239 struct dentry *ref_dentry = NULL;
240 struct inode *ref_inode;
244 bool found_short_name = false;
246 inode_for_each_dentry(dentry, inode) {
247 if (!ref_dentry || ref_dentry->inode->num_ads == 0)
249 if (dentry->short_name_len) {
250 if (found_short_name) {
251 ERROR("Multiple short names in hard link "
253 inconsistent_inode(inode);
254 return WIMLIB_ERR_INVALID_DENTRY;
256 found_short_name = true;
259 if (dentry->inode->creation_time > last_ctime)
260 last_ctime = dentry->inode->creation_time;
261 if (dentry->inode->last_write_time > last_mtime)
262 last_mtime = dentry->inode->last_write_time;
263 if (dentry->inode->last_access_time > last_atime)
264 last_atime = dentry->inode->last_access_time;
267 ref_inode = ref_dentry->inode;
268 ref_inode->link_count = 1;
270 inode_for_each_dentry(dentry, inode) {
271 if (dentry != ref_dentry) {
272 if (!inodes_consistent(ref_inode, dentry->inode)) {
273 inconsistent_inode(dentry->inode);
274 return WIMLIB_ERR_INVALID_DENTRY;
276 /* Free the unneeded `struct inode'. */
277 free_inode(dentry->inode);
278 dentry->inode = ref_inode;
279 ref_inode->link_count++;
282 ref_inode->creation_time = last_ctime;
283 ref_inode->last_write_time = last_mtime;
284 ref_inode->last_access_time = last_atime;
285 wimlib_assert(inode_link_count(ref_inode) == ref_inode->link_count);
290 * Fixes up a nominal inode.
292 * By a nominal inode we mean a group of two or more dentries that share
293 * the same hard link group ID.
295 * If dentries in the inode are found to be inconsistent, we may split the inode
296 * into several "true" inodes. @new_inodes points to a linked list of
297 * these split inodes, and if we create any, they will be added to this list.
299 * After splitting up each nominal inode into the "true" inodes we
300 * will canonicalize the link group by getting rid of all the superfluous
301 * `struct inodes'. There will be just one `struct inode' for each hard link
305 fix_nominal_inode(struct inode *inode, struct hlist_head *inode_list)
307 struct dentry *dentry, *ref_dentry;
308 struct hlist_node *cur, *tmp;
310 size_t num_true_inodes;
312 wimlib_assert(inode->link_count == inode_link_count(inode));
314 LIST_HEAD(dentries_with_data_streams);
315 LIST_HEAD(dentries_with_no_data_streams);
316 HLIST_HEAD(true_inodes);
318 /* Create a list of dentries in the nominal inode that have at
319 * least one data stream with a non-zero hash, and another list that
320 * contains the dentries that have a zero hash for all data streams. */
321 inode_for_each_dentry(dentry, inode) {
322 for (unsigned i = 0; i <= dentry->inode->num_ads; i++) {
324 hash = inode_stream_hash(dentry->inode, i);
325 if (!is_zero_hash(hash)) {
326 list_add(&dentry->tmp_list,
327 &dentries_with_data_streams);
331 list_add(&dentry->tmp_list,
332 &dentries_with_no_data_streams);
337 /* If there are no dentries with data streams, we require the nominal
338 * inode to be a true inode */
339 if (list_empty(&dentries_with_data_streams)) {
340 DEBUG("No data streams");
343 if (inode->link_count > 1) {
344 DEBUG("Found link group of size %zu without "
345 "any data streams:", inode->link_count);
346 print_inode_dentries(inode);
347 DEBUG("We are going to interpret it as true "
348 "link group, provided that the dentries "
353 hlist_add_head(&inode->hlist, inode_list);
354 return fix_true_inode(inode);
357 /* One or more dentries had data streams specified. We check each of
358 * these dentries for consistency with the others to form a set of true
361 list_for_each_entry(dentry, &dentries_with_data_streams, tmp_list) {
362 /* Look for a true inode that is consistent with
363 * this dentry and add this dentry to it. Or, if none
364 * of the true inodes are consistent with this
365 * dentry, make a new one. */
366 hlist_for_each_entry(inode, cur, &true_inodes, hlist) {
367 if (ref_inodes_consistent(inode, dentry->inode)) {
368 list_add(&dentry->inode_dentry_list,
369 &inode->dentry_list);
374 INIT_LIST_HEAD(&dentry->inode->dentry_list);
375 list_add(&dentry->inode_dentry_list, &dentry->inode->dentry_list);
376 hlist_add_head(&dentry->inode->hlist, &true_inodes);
381 wimlib_assert(num_true_inodes != 0);
383 /* If there were dentries with no data streams, we require there to only
384 * be one true inode so that we know which inode to assign the
385 * streamless dentries to. */
386 if (!list_empty(&dentries_with_no_data_streams)) {
387 if (num_true_inodes != 1) {
388 ERROR("Hard inode ambiguity detected!");
389 ERROR("We split up inode 0x%"PRIx64" due to "
390 "inconsistencies,", inode->ino);
391 ERROR("but dentries with no stream information remained. "
392 "We don't know which true hard link");
393 ERROR("inode to assign them to.");
394 return WIMLIB_ERR_INVALID_DENTRY;
396 inode = container_of(true_inodes.first,
399 /* Assign the streamless dentries to the one and only true
401 list_for_each_entry(dentry, &dentries_with_no_data_streams, tmp_list)
402 list_add(&dentry->inode_dentry_list, &inode->dentry_list);
404 if (num_true_inodes != 1) {
407 inode = container_of(true_inodes.first,
411 printf("Split nominal inode 0x%"PRIx64" into %zu "
413 inode->ino, num_true_inodes);
414 puts("------------------------------------------------------------------------------");
416 hlist_for_each_entry(inode, cur, &true_inodes, hlist) {
417 printf("[Split inode %zu]\n", i++);
418 print_inode_dentries(inode);
421 puts("------------------------------------------------------------------------------");
426 hlist_for_each_entry_safe(inode, cur, tmp, &true_inodes, hlist) {
427 hlist_add_head(&inode->hlist, inode_list);
428 ret = fix_true_inode(inode);
436 * Goes through each inode and shares the inodes among members of a hard
439 * In the process, the dentries in each inode are checked for consistency.
440 * If they contain data features that indicate they cannot really be in the same
441 * inode, this should be an error, but in reality this case needs to
442 * be handled, so we split the dentries into different inodes.
444 int fix_inodes(struct inode_table *table, struct hlist_head *inode_list)
447 struct hlist_node *cur, *tmp;
449 INIT_HLIST_HEAD(inode_list);
450 for (u64 i = 0; i < table->capacity; i++) {
451 hlist_for_each_entry_safe(inode, cur, tmp, &table->array[i], hlist) {
452 ret = fix_nominal_inode(inode, inode_list);
457 hlist_for_each_safe(cur, tmp, &table->extra_inodes)
458 hlist_add_head(cur, inode_list);