4 * Code to deal with hard links in WIMs. Essentially, the WIM dentries are put
5 * into a hash table indexed by the hard link group ID field, then for each hard
6 * link group, 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 * \ / |link_group| |link_group|
39 * dentry ------------ ------------
43 * ----------- ----------- / \
44 * dentry<---| struct | | struct |---> dentry dentry
45 * / |link_group| |link_group| \ /
46 * dentry ------------ ------------ dentry
50 * link_group_table->array | idx 0 | idx 1 |
54 /* Hard link group; it's identified by its hard link group ID and points to a
55 * circularly linked list of dentries. */
59 /* Pointer to use to make a singly-linked list of link groups. */
60 struct link_group *next;
62 /* This is a pointer to the circle and not part of the circle itself.
63 * This makes it easy to iterate through other dentries hard-linked to a
64 * given dentry without having to find the "head" of the list first. */
65 struct list_head *dentry_list;
68 /* Hash table to find hard link groups, identified by their hard link group ID.
70 struct link_group_table {
71 /* Fields for the hash table */
72 struct link_group **array;
77 * Linked list of "extra" groups. These may be:
79 * - Hard link groups of size 1, which are all allowed to have 0 for
80 * their hard link group ID, meaning we cannot insert them into the
81 * hash table before calling assign_link_group_ids().
83 * - Groups we create ourselves by splitting a nominal hard link group
84 * due to inconsistencies in the dentries. These groups will share a
85 * hard link group ID with some other group until
86 * assign_link_group_ids() is called.
88 struct link_group *extra_groups;
91 /* Returns pointer to a new link group table having the specified capacity */
92 struct link_group_table *new_link_group_table(size_t capacity)
94 struct link_group_table *table;
95 struct link_group **array;
97 table = MALLOC(sizeof(struct link_group_table));
100 array = CALLOC(capacity, sizeof(array[0]));
105 table->num_entries = 0;
106 table->capacity = capacity;
107 table->array = array;
108 table->extra_groups = NULL;
111 ERROR("Failed to allocate memory for link group table with capacity %zu",
117 * Insert a dentry into the hard link group table based on its hard link group
120 * If there is already a dentry in the table having the same hard link group ID,
121 * and the hard link group ID is not 0, the dentry is added to the circular
122 * linked list for that hard link group.
124 * If the hard link group ID is 0, this indicates a dentry that's in a hard link
125 * group by itself (has a link count of 1). We can't insert it into the hash
126 * table itself because we don't know what hard link group IDs are available to
127 * give it (this could be kept track of but would be more difficult). Instead
128 * we keep a linked list of the single dentries, and assign them hard link group
131 int link_group_table_insert(struct dentry *dentry, void *__table)
133 struct link_group_table *table = __table;
135 struct link_group *group;
137 if (dentry->link_group_id == 0) {
138 /* Single group--- Add to the list of extra groups (we can't put
139 * it in the table itself because all the singles have a link
141 group = MALLOC(sizeof(struct link_group));
143 return WIMLIB_ERR_NOMEM;
144 group->link_group_id = 0;
145 group->next = table->extra_groups;
146 table->extra_groups = group;
147 INIT_LIST_HEAD(&dentry->link_group_list);
148 group->dentry_list = &dentry->link_group_list;
150 /* Hard link group that may contain multiple dentries (the code
151 * will work even if the group actually contains only 1 dentry
154 /* Try adding to existing hard link group */
155 pos = dentry->link_group_id % table->capacity;
156 group = table->array[pos];
158 if (group->link_group_id == dentry->link_group_id) {
159 list_add(&dentry->link_group_list,
166 /* Add new hard link group to the table */
168 group = MALLOC(sizeof(struct link_group));
170 return WIMLIB_ERR_NOMEM;
171 group->link_group_id = dentry->link_group_id;
172 group->next = table->array[pos];
173 INIT_LIST_HEAD(&dentry->link_group_list);
174 group->dentry_list = &dentry->link_group_list;
175 table->array[pos] = group;
177 /* XXX Make the table grow when too many entries have been
179 table->num_entries++;
184 static void free_link_group_list(struct link_group *group)
186 struct link_group *next_group;
188 next_group = group->next;
194 /* Frees a link group table. */
195 void free_link_group_table(struct link_group_table *table)
199 for (size_t i = 0; i < table->capacity; i++)
200 free_link_group_list(table->array[i]);
201 free_link_group_list(table->extra_groups);
207 assign_link_group_ids_to_list(struct link_group *group, u64 id,
208 struct link_group **extra_groups)
210 struct dentry *dentry;
211 struct list_head *cur_head;
212 struct link_group *prev_group = NULL;
213 struct link_group *cur_group = group;
215 cur_head = cur_group->dentry_list;
217 dentry = container_of(cur_head,
220 dentry->link_group_id = id;
221 cur_head = cur_head->next;
222 } while (cur_head != cur_group->dentry_list);
223 cur_group->link_group_id = id;
225 prev_group = cur_group;
226 cur_group = cur_group->next;
228 if (group && extra_groups) {
229 prev_group->next = *extra_groups;
230 *extra_groups = group;
235 /* Insert the link groups in the `extra_groups' list into the hash table */
236 static void insert_extra_groups(struct link_group_table *table)
238 struct link_group *group, *next_group;
241 group = table->extra_groups;
243 next_group = group->next;
244 pos = group->link_group_id % table->capacity;
245 group->next = table->array[pos];
246 table->array[pos] = group;
249 table->extra_groups = NULL;
252 /* Assign the link group IDs to dentries in a link group table, and return the
253 * next available link group ID. */
254 u64 assign_link_group_ids(struct link_group_table *table)
256 DEBUG("Assigning link groups");
257 struct link_group *extra_groups = table->extra_groups;
259 /* Assign consecutive link group IDs to each link group in the hash
262 for (size_t i = 0; i < table->capacity; i++) {
263 id = assign_link_group_ids_to_list(table->array[i], id,
264 &table->extra_groups);
265 table->array[i] = NULL;
268 /* Assign link group IDs to the "extra" link groups and insert them into
270 id = assign_link_group_ids_to_list(extra_groups, id, NULL);
271 insert_extra_groups(table);
277 static void inconsistent_link_group(const struct dentry *first_dentry)
279 const struct dentry *dentry = first_dentry;
281 ERROR("An inconsistent hard link group that we cannot correct has been "
283 ERROR("The dentries are located at the following paths:");
285 ERROR("`%s'", dentry->full_path_utf8);
286 } while ((dentry = container_of(dentry->link_group_list.next,
288 link_group_list)) != first_dentry);
291 static bool ref_dentries_consistent(const struct dentry * restrict ref_dentry_1,
292 const struct dentry * restrict ref_dentry_2)
294 wimlib_assert(ref_dentry_1 != ref_dentry_2);
296 if (ref_dentry_1->num_ads != ref_dentry_2->num_ads)
298 if (ref_dentry_1->security_id != ref_dentry_2->security_id
299 || ref_dentry_1->attributes != ref_dentry_2->attributes)
301 for (unsigned i = 0; i <= ref_dentry_1->num_ads; i++) {
302 const u8 *ref_1_hash, *ref_2_hash;
303 ref_1_hash = dentry_stream_hash(ref_dentry_1, i);
304 ref_2_hash = dentry_stream_hash(ref_dentry_2, i);
305 if (!hashes_equal(ref_1_hash, ref_2_hash))
307 if (i && !ads_entries_have_same_name(&ref_dentry_1->ads_entries[i - 1],
308 &ref_dentry_2->ads_entries[i - 1]))
315 static bool dentries_consistent(const struct dentry * restrict ref_dentry,
316 const struct dentry * restrict dentry)
318 wimlib_assert(ref_dentry != dentry);
320 if (ref_dentry->num_ads != dentry->num_ads && dentry->num_ads != 0)
322 if (ref_dentry->security_id != dentry->security_id
323 || ref_dentry->attributes != dentry->attributes)
325 for (unsigned i = 0; i <= min(ref_dentry->num_ads, dentry->num_ads); i++) {
326 const u8 *ref_hash, *hash;
327 ref_hash = dentry_stream_hash(ref_dentry, i);
328 hash = dentry_stream_hash(dentry, i);
329 if (!hashes_equal(ref_hash, hash) && !is_zero_hash(hash))
331 if (i && !ads_entries_have_same_name(&ref_dentry->ads_entries[i - 1],
332 &dentry->ads_entries[i - 1]))
340 print_dentry_list(const struct dentry *first_dentry)
342 const struct dentry *dentry = first_dentry;
344 printf("`%s'\n", dentry->full_path_utf8);
345 } while ((dentry = container_of(dentry->link_group_list.next,
347 link_group_list)) != first_dentry);
351 /* Fix up a "true" link group and check for inconsistencies */
353 fix_true_link_group(struct dentry *first_dentry)
355 struct dentry *dentry;
356 struct dentry *ref_dentry = NULL;
360 bool found_short_name = false;
362 dentry = first_dentry;
364 if (!ref_dentry || ref_dentry->num_ads == 0)
366 if (dentry->short_name_len) {
367 if (found_short_name) {
368 ERROR("Multiple short names in hard link "
370 inconsistent_link_group(first_dentry);
371 return WIMLIB_ERR_INVALID_DENTRY;
373 found_short_name = true;
376 if (dentry->creation_time > last_ctime)
377 last_ctime = dentry->creation_time;
378 if (dentry->last_write_time > last_mtime)
379 last_mtime = dentry->last_write_time;
380 if (dentry->last_access_time > last_atime)
381 last_atime = dentry->last_access_time;
382 } while ((dentry = container_of(dentry->link_group_list.next,
384 link_group_list)) != first_dentry);
387 ref_dentry->ads_entries_status = ADS_ENTRIES_OWNER;
388 dentry = first_dentry;
390 if (dentry != ref_dentry) {
391 if (!dentries_consistent(ref_dentry, dentry)) {
392 inconsistent_link_group(first_dentry);
393 return WIMLIB_ERR_INVALID_DENTRY;
395 copy_hash(dentry->hash, ref_dentry->hash);
396 dentry_free_ads_entries(dentry);
397 dentry->num_ads = ref_dentry->num_ads;
398 dentry->ads_entries = ref_dentry->ads_entries;
399 dentry->ads_entries_status = ADS_ENTRIES_USER;
401 dentry->creation_time = last_ctime;
402 dentry->last_write_time = last_mtime;
403 dentry->last_access_time = last_atime;
404 } while ((dentry = container_of(dentry->link_group_list.next,
406 link_group_list)) != first_dentry);
411 * Fixes up a nominal link group.
413 * By a nominal link group we mean a group of two or more dentries that share
414 * the same hard link group ID.
416 * If dentries in the group are found to be inconsistent, we may split the group
417 * into several "true" hard link groups. @new_groups points to a linked list of
418 * these split groups, and if we create any, they will be added to this list.
420 * After splitting up each nominal link group into the "true" link groups we
421 * will canonicalize the link groups. To do this, we:
423 * - Assign all the dentries in the link group the most recent timestamp
424 * among all the corresponding timestamps in the link group, for each of
425 * the three categories of time stamps.
427 * - Make sure the dentry->hash field is valid in all the dentries, if
428 * possible (this field may be all zeroes, and in the context of a hard
429 * link group this must be interpreted as implicitly refering to the same
430 * stream as another dentry in the hard link group that does NOT have all
431 * zeroes for this field).
433 * - Make sure dentry->num_ads is the same in all the dentries in the link
434 * group. In some cases, it's possible for it to be set to 0 when it
435 * actually must be interpreted as being the same as the number of
436 * alternate data streams in another dentry in the hard link group that has
437 * a nonzero number of alternate data streams.
439 * - Make sure only the dentry->ads_entries array is only allocated for one
440 * dentry in the hard link group. This dentry will have
441 * dentry->ads_entries_status set to ADS_ENTRIES_OWNER, while the others
442 * will have dentry->ads_entries_status set to ADS_ENTRIES_USER.
445 fix_nominal_link_group(struct link_group *group,
446 struct link_group **new_groups)
448 struct dentry *tmp, *dentry, *ref_dentry;
450 size_t num_true_link_groups;
451 struct list_head *head;
453 LIST_HEAD(dentries_with_data_streams);
454 LIST_HEAD(dentries_with_no_data_streams);
455 LIST_HEAD(true_link_groups);
457 /* Create a list of dentries in the nominal hard link group that have at
458 * least one data stream with a non-zero hash, and another list that
459 * contains the dentries that have a zero hash for all data streams. */
460 dentry = container_of(group->dentry_list, struct dentry,
463 for (unsigned i = 0; i <= dentry->num_ads; i++) {
465 hash = dentry_stream_hash(dentry, i);
466 if (!is_zero_hash(hash)) {
467 list_add(&dentry->tmp_list,
468 &dentries_with_data_streams);
472 list_add(&dentry->tmp_list,
473 &dentries_with_no_data_streams);
475 dentry = container_of(dentry->link_group_list.next,
478 } while (&dentry->link_group_list != group->dentry_list);
480 /* If there are no dentries with data streams, we require the nominal
481 * link group to be a true link group */
482 if (list_empty(&dentries_with_data_streams)) {
485 size_t size = dentry_link_group_size(dentry);
487 DEBUG("Found link group of size %zu without "
488 "any data streams:", size);
489 print_dentry_list(dentry);
490 DEBUG("We are going to interpret it as true "
491 "link group, provided that the dentries "
496 return fix_true_link_group(container_of(group->dentry_list,
501 /* One or more dentries had data streams specified. We check each of
502 * these dentries for consistency with the others to form a set of true
504 num_true_link_groups = 0;
505 list_for_each_entry_safe(dentry, tmp, &dentries_with_data_streams,
508 list_del(&dentry->tmp_list);
510 /* Look for a true link group that is consistent with
511 * this dentry and add this dentry to it. Or, if none
512 * of the true link groups are consistent with this
513 * dentry, make a new one. */
514 list_for_each_entry(ref_dentry, &true_link_groups, tmp_list) {
515 if (ref_dentries_consistent(ref_dentry, dentry)) {
516 list_add(&dentry->link_group_list,
517 &ref_dentry->link_group_list);
521 num_true_link_groups++;
522 list_add(&dentry->tmp_list, &true_link_groups);
523 INIT_LIST_HEAD(&dentry->link_group_list);
528 wimlib_assert(num_true_link_groups != 0);
530 /* If there were dentries with no data streams, we require there to only
531 * be one true link group so that we know which link group to assign the
532 * streamless dentries to. */
533 if (!list_empty(&dentries_with_no_data_streams)) {
534 if (num_true_link_groups != 1) {
535 ERROR("Hard link group ambiguity detected!");
536 ERROR("We split up hard link group 0x%"PRIx64" due to "
537 "inconsistencies,", group->link_group_id);
538 ERROR("but dentries with no stream information remained. "
539 "We don't know which true hard link");
540 ERROR("group to assign them to.");
541 return WIMLIB_ERR_INVALID_DENTRY;
543 /* Assign the streamless dentries to the one and only true link
545 ref_dentry = container_of(true_link_groups.next,
548 list_for_each_entry(dentry, &dentries_with_no_data_streams, tmp_list)
549 list_add(&dentry->link_group_list, &ref_dentry->link_group_list);
551 if (num_true_link_groups != 1) {
554 printf("Split nominal link group 0x%"PRIx64" into %zu "
556 group->link_group_id, num_true_link_groups);
557 puts("------------------------------------------------------------------------------");
559 list_for_each_entry(dentry, &true_link_groups, tmp_list) {
560 printf("[Split link group %zu]\n", i++);
561 print_dentry_list(dentry);
564 puts("------------------------------------------------------------------------------");
569 list_for_each_entry(dentry, &true_link_groups, tmp_list) {
570 ret = fix_true_link_group(dentry);
575 /* Make new `struct link_group's for the new link groups */
576 for (head = true_link_groups.next->next;
577 head != &true_link_groups;
580 dentry = container_of(head, struct dentry, tmp_list);
581 group = MALLOC(sizeof(*group));
583 ERROR("Out of memory");
584 return WIMLIB_ERR_NOMEM;
586 group->link_group_id = dentry->link_group_id;
587 group->dentry_list = &dentry->link_group_list;
588 group->next = *new_groups;
595 * Goes through each link group and shares the ads_entries (Alternate Data
596 * Stream entries) field of each dentry among members of a hard link group.
598 * In the process, the dentries in each link group are checked for consistency.
599 * If they contain data features that indicate they cannot really be in the same
600 * hard link group, this should be an error, but in reality this case needs to
601 * be handled, so we split the dentries into different hard link groups.
603 * One of the dentries in each hard link group group is arbitrarily assigned the
604 * role of "owner" of the memory pointed to by the @ads_entries field,
605 * (ADS_ENTRIES_OWNER), while the others are "users" (ADS_ENTRIES_USER) who are
606 * not allowed to free the memory.
608 int fix_link_groups(struct link_group_table *table)
610 for (u64 i = 0; i < table->capacity; i++) {
611 struct link_group *group = table->array[i];
614 ret = fix_nominal_link_group(group, &table->extra_groups);