2 #include <linux/inotify.h>
3 #include <linux/namei.h>
4 #include <linux/mount.h>
12 struct audit_chunk *root;
13 struct list_head chunks;
14 struct list_head rules;
15 struct list_head list;
16 struct list_head same_root;
22 struct list_head hash;
23 struct inotify_watch watch;
24 struct list_head trees; /* with root here */
29 struct list_head list;
30 struct audit_tree *owner;
31 unsigned index; /* index; upper bit indicates 'will prune' */
35 static LIST_HEAD(tree_list);
36 static LIST_HEAD(prune_list);
39 * One struct chunk is attached to each inode of interest.
40 * We replace struct chunk on tagging/untagging.
41 * Rules have pointer to struct audit_tree.
42 * Rules have struct list_head rlist forming a list of rules over
44 * References to struct chunk are collected at audit_inode{,_child}()
45 * time and used in AUDIT_TREE rule matching.
46 * These references are dropped at the same time we are calling
47 * audit_free_names(), etc.
49 * Cyclic lists galore:
50 * tree.chunks anchors chunk.owners[].list hash_lock
51 * tree.rules anchors rule.rlist audit_filter_mutex
52 * chunk.trees anchors tree.same_root hash_lock
53 * chunk.hash is a hash with middle bits of watch.inode as
54 * a hash function. RCU, hash_lock
56 * tree is refcounted; one reference for "some rules on rules_list refer to
57 * it", one for each chunk with pointer to it.
59 * chunk is refcounted by embedded inotify_watch.
61 * node.index allows to get from node.list to containing chunk.
62 * MSB of that sucker is stolen to mark taggings that we might have to
63 * revert - several operations have very unpleasant cleanup logics and
64 * that makes a difference. Some.
67 static struct inotify_handle *rtree_ih;
69 static struct audit_tree *alloc_tree(const char *s)
71 struct audit_tree *tree;
73 tree = kmalloc(sizeof(struct audit_tree) + strlen(s) + 1, GFP_KERNEL);
75 atomic_set(&tree->count, 1);
77 INIT_LIST_HEAD(&tree->chunks);
78 INIT_LIST_HEAD(&tree->rules);
79 INIT_LIST_HEAD(&tree->list);
80 INIT_LIST_HEAD(&tree->same_root);
82 strcpy(tree->pathname, s);
87 static inline void get_tree(struct audit_tree *tree)
89 atomic_inc(&tree->count);
92 static void __put_tree(struct rcu_head *rcu)
94 struct audit_tree *tree = container_of(rcu, struct audit_tree, head);
98 static inline void put_tree(struct audit_tree *tree)
100 if (atomic_dec_and_test(&tree->count))
101 call_rcu(&tree->head, __put_tree);
104 /* to avoid bringing the entire thing in audit.h */
105 const char *audit_tree_path(struct audit_tree *tree)
107 return tree->pathname;
110 static struct audit_chunk *alloc_chunk(int count)
112 struct audit_chunk *chunk;
116 size = offsetof(struct audit_chunk, owners) + count * sizeof(struct node);
117 chunk = kzalloc(size, GFP_KERNEL);
121 INIT_LIST_HEAD(&chunk->hash);
122 INIT_LIST_HEAD(&chunk->trees);
123 chunk->count = count;
124 for (i = 0; i < count; i++) {
125 INIT_LIST_HEAD(&chunk->owners[i].list);
126 chunk->owners[i].index = i;
128 inotify_init_watch(&chunk->watch);
132 static void __free_chunk(struct rcu_head *rcu)
134 struct audit_chunk *chunk = container_of(rcu, struct audit_chunk, head);
137 for (i = 0; i < chunk->count; i++) {
138 if (chunk->owners[i].owner)
139 put_tree(chunk->owners[i].owner);
144 static inline void free_chunk(struct audit_chunk *chunk)
146 call_rcu(&chunk->head, __free_chunk);
149 void audit_put_chunk(struct audit_chunk *chunk)
151 put_inotify_watch(&chunk->watch);
154 enum {HASH_SIZE = 128};
155 static struct list_head chunk_hash_heads[HASH_SIZE];
156 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(hash_lock);
158 static inline struct list_head *chunk_hash(const struct inode *inode)
160 unsigned long n = (unsigned long)inode / L1_CACHE_BYTES;
161 return chunk_hash_heads + n % HASH_SIZE;
164 /* hash_lock is held by caller */
165 static void insert_hash(struct audit_chunk *chunk)
167 struct list_head *list = chunk_hash(chunk->watch.inode);
168 list_add_rcu(&chunk->hash, list);
171 /* called under rcu_read_lock */
172 struct audit_chunk *audit_tree_lookup(const struct inode *inode)
174 struct list_head *list = chunk_hash(inode);
175 struct list_head *pos;
177 list_for_each_rcu(pos, list) {
178 struct audit_chunk *p = container_of(pos, struct audit_chunk, hash);
179 if (p->watch.inode == inode) {
180 get_inotify_watch(&p->watch);
187 int audit_tree_match(struct audit_chunk *chunk, struct audit_tree *tree)
190 for (n = 0; n < chunk->count; n++)
191 if (chunk->owners[n].owner == tree)
196 /* tagging and untagging inodes with trees */
198 static void untag_chunk(struct audit_chunk *chunk, struct node *p)
200 struct audit_chunk *new;
201 struct audit_tree *owner;
202 int size = chunk->count - 1;
205 mutex_lock(&chunk->watch.inode->inotify_mutex);
207 mutex_unlock(&chunk->watch.inode->inotify_mutex);
215 spin_lock(&hash_lock);
216 list_del_init(&chunk->trees);
217 if (owner->root == chunk)
219 list_del_init(&p->list);
220 list_del_rcu(&chunk->hash);
221 spin_unlock(&hash_lock);
222 inotify_evict_watch(&chunk->watch);
223 mutex_unlock(&chunk->watch.inode->inotify_mutex);
224 put_inotify_watch(&chunk->watch);
228 new = alloc_chunk(size);
231 if (inotify_clone_watch(&chunk->watch, &new->watch) < 0) {
237 spin_lock(&hash_lock);
238 list_replace_init(&chunk->trees, &new->trees);
239 if (owner->root == chunk) {
240 list_del_init(&owner->same_root);
244 for (i = j = 0; i < size; i++, j++) {
245 struct audit_tree *s;
246 if (&chunk->owners[j] == p) {
247 list_del_init(&p->list);
251 s = chunk->owners[j].owner;
252 new->owners[i].owner = s;
253 new->owners[i].index = chunk->owners[j].index - j + i;
254 if (!s) /* result of earlier fallback */
257 list_replace_init(&chunk->owners[i].list, &new->owners[j].list);
260 list_replace_rcu(&chunk->hash, &new->hash);
261 list_for_each_entry(owner, &new->trees, same_root)
263 spin_unlock(&hash_lock);
264 inotify_evict_watch(&chunk->watch);
265 mutex_unlock(&chunk->watch.inode->inotify_mutex);
266 put_inotify_watch(&chunk->watch);
270 // do the best we can
271 spin_lock(&hash_lock);
272 if (owner->root == chunk) {
273 list_del_init(&owner->same_root);
276 list_del_init(&p->list);
279 spin_unlock(&hash_lock);
280 mutex_unlock(&chunk->watch.inode->inotify_mutex);
283 static int create_chunk(struct inode *inode, struct audit_tree *tree)
285 struct audit_chunk *chunk = alloc_chunk(1);
289 if (inotify_add_watch(rtree_ih, &chunk->watch, inode, IN_IGNORED | IN_DELETE_SELF) < 0) {
294 mutex_lock(&inode->inotify_mutex);
295 spin_lock(&hash_lock);
297 spin_unlock(&hash_lock);
299 inotify_evict_watch(&chunk->watch);
300 mutex_unlock(&inode->inotify_mutex);
301 put_inotify_watch(&chunk->watch);
304 chunk->owners[0].index = (1U << 31);
305 chunk->owners[0].owner = tree;
307 list_add(&chunk->owners[0].list, &tree->chunks);
310 list_add(&tree->same_root, &chunk->trees);
313 spin_unlock(&hash_lock);
314 mutex_unlock(&inode->inotify_mutex);
318 /* the first tagged inode becomes root of tree */
319 static int tag_chunk(struct inode *inode, struct audit_tree *tree)
321 struct inotify_watch *watch;
322 struct audit_tree *owner;
323 struct audit_chunk *chunk, *old;
327 if (inotify_find_watch(rtree_ih, inode, &watch) < 0)
328 return create_chunk(inode, tree);
330 old = container_of(watch, struct audit_chunk, watch);
332 /* are we already there? */
333 spin_lock(&hash_lock);
334 for (n = 0; n < old->count; n++) {
335 if (old->owners[n].owner == tree) {
336 spin_unlock(&hash_lock);
337 put_inotify_watch(watch);
341 spin_unlock(&hash_lock);
343 chunk = alloc_chunk(old->count + 1);
347 mutex_lock(&inode->inotify_mutex);
348 if (inotify_clone_watch(&old->watch, &chunk->watch) < 0) {
349 mutex_unlock(&inode->inotify_mutex);
353 spin_lock(&hash_lock);
355 spin_unlock(&hash_lock);
357 inotify_evict_watch(&chunk->watch);
358 mutex_unlock(&inode->inotify_mutex);
359 put_inotify_watch(&chunk->watch);
362 list_replace_init(&old->trees, &chunk->trees);
363 for (n = 0, p = chunk->owners; n < old->count; n++, p++) {
364 struct audit_tree *s = old->owners[n].owner;
366 p->index = old->owners[n].index;
367 if (!s) /* result of fallback in untag */
370 list_replace_init(&old->owners[n].list, &p->list);
372 p->index = (chunk->count - 1) | (1U<<31);
375 list_add(&p->list, &tree->chunks);
376 list_replace_rcu(&old->hash, &chunk->hash);
377 list_for_each_entry(owner, &chunk->trees, same_root)
382 list_add(&tree->same_root, &chunk->trees);
384 spin_unlock(&hash_lock);
385 inotify_evict_watch(&old->watch);
386 mutex_unlock(&inode->inotify_mutex);
387 put_inotify_watch(&old->watch);
391 static struct audit_chunk *find_chunk(struct node *p)
393 int index = p->index & ~(1U<<31);
395 return container_of(p, struct audit_chunk, owners[0]);
398 static void kill_rules(struct audit_tree *tree)
400 struct audit_krule *rule, *next;
401 struct audit_entry *entry;
402 struct audit_buffer *ab;
404 list_for_each_entry_safe(rule, next, &tree->rules, rlist) {
405 entry = container_of(rule, struct audit_entry, rule);
407 list_del_init(&rule->rlist);
409 /* not a half-baked one */
410 ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE);
411 audit_log_format(ab, "op=remove rule dir=");
412 audit_log_untrustedstring(ab, rule->tree->pathname);
413 if (rule->filterkey) {
414 audit_log_format(ab, " key=");
415 audit_log_untrustedstring(ab, rule->filterkey);
417 audit_log_format(ab, " key=(null)");
418 audit_log_format(ab, " list=%d res=1", rule->listnr);
421 list_del_rcu(&entry->list);
422 call_rcu(&entry->rcu, audit_free_rule_rcu);
428 * finish killing struct audit_tree
430 static void prune_one(struct audit_tree *victim)
432 spin_lock(&hash_lock);
433 while (!list_empty(&victim->chunks)) {
435 struct audit_chunk *chunk;
437 p = list_entry(victim->chunks.next, struct node, list);
438 chunk = find_chunk(p);
439 get_inotify_watch(&chunk->watch);
440 spin_unlock(&hash_lock);
442 untag_chunk(chunk, p);
444 put_inotify_watch(&chunk->watch);
445 spin_lock(&hash_lock);
447 spin_unlock(&hash_lock);
451 /* trim the uncommitted chunks from tree */
453 static void trim_marked(struct audit_tree *tree)
455 struct list_head *p, *q;
456 spin_lock(&hash_lock);
458 spin_unlock(&hash_lock);
462 for (p = tree->chunks.next; p != &tree->chunks; p = q) {
463 struct node *node = list_entry(p, struct node, list);
465 if (node->index & (1U<<31)) {
467 list_add(p, &tree->chunks);
471 while (!list_empty(&tree->chunks)) {
473 struct audit_chunk *chunk;
475 node = list_entry(tree->chunks.next, struct node, list);
477 /* have we run out of marked? */
478 if (!(node->index & (1U<<31)))
481 chunk = find_chunk(node);
482 get_inotify_watch(&chunk->watch);
483 spin_unlock(&hash_lock);
485 untag_chunk(chunk, node);
487 put_inotify_watch(&chunk->watch);
488 spin_lock(&hash_lock);
490 if (!tree->root && !tree->goner) {
492 spin_unlock(&hash_lock);
493 mutex_lock(&audit_filter_mutex);
495 list_del_init(&tree->list);
496 mutex_unlock(&audit_filter_mutex);
499 spin_unlock(&hash_lock);
503 /* called with audit_filter_mutex */
504 int audit_remove_tree_rule(struct audit_krule *rule)
506 struct audit_tree *tree;
509 spin_lock(&hash_lock);
510 list_del_init(&rule->rlist);
511 if (list_empty(&tree->rules) && !tree->goner) {
513 list_del_init(&tree->same_root);
515 list_move(&tree->list, &prune_list);
517 spin_unlock(&hash_lock);
518 audit_schedule_prune();
522 spin_unlock(&hash_lock);
528 void audit_trim_trees(void)
530 struct list_head cursor;
532 mutex_lock(&audit_filter_mutex);
533 list_add(&cursor, &tree_list);
534 while (cursor.next != &tree_list) {
535 struct audit_tree *tree;
537 struct vfsmount *root_mnt;
539 struct list_head list;
542 tree = container_of(cursor.next, struct audit_tree, list);
545 list_add(&cursor, &tree->list);
546 mutex_unlock(&audit_filter_mutex);
548 err = path_lookup(tree->pathname, 0, &nd);
552 root_mnt = collect_mounts(nd.path.mnt, nd.path.dentry);
557 list_add_tail(&list, &root_mnt->mnt_list);
558 spin_lock(&hash_lock);
559 list_for_each_entry(node, &tree->chunks, list) {
560 struct audit_chunk *chunk = find_chunk(node);
561 struct inode *inode = chunk->watch.inode;
562 struct vfsmount *mnt;
563 node->index |= 1U<<31;
564 list_for_each_entry(mnt, &list, mnt_list) {
565 if (mnt->mnt_root->d_inode == inode) {
566 node->index &= ~(1U<<31);
571 spin_unlock(&hash_lock);
574 list_del_init(&list);
575 drop_collected_mounts(root_mnt);
577 mutex_lock(&audit_filter_mutex);
580 mutex_unlock(&audit_filter_mutex);
583 static int is_under(struct vfsmount *mnt, struct dentry *dentry,
584 struct nameidata *nd)
586 if (mnt != nd->path.mnt) {
588 if (mnt->mnt_parent == mnt)
590 if (mnt->mnt_parent == nd->path.mnt)
592 mnt = mnt->mnt_parent;
594 dentry = mnt->mnt_mountpoint;
596 return is_subdir(dentry, nd->path.dentry);
599 int audit_make_tree(struct audit_krule *rule, char *pathname, u32 op)
602 if (pathname[0] != '/' ||
603 rule->listnr != AUDIT_FILTER_EXIT ||
605 rule->inode_f || rule->watch || rule->tree)
607 rule->tree = alloc_tree(pathname);
613 void audit_put_tree(struct audit_tree *tree)
618 /* called with audit_filter_mutex */
619 int audit_add_tree_rule(struct audit_krule *rule)
621 struct audit_tree *seed = rule->tree, *tree;
623 struct vfsmount *mnt, *p;
624 struct list_head list;
627 list_for_each_entry(tree, &tree_list, list) {
628 if (!strcmp(seed->pathname, tree->pathname)) {
631 list_add(&rule->rlist, &tree->rules);
636 list_add(&tree->list, &tree_list);
637 list_add(&rule->rlist, &tree->rules);
638 /* do not set rule->tree yet */
639 mutex_unlock(&audit_filter_mutex);
641 err = path_lookup(tree->pathname, 0, &nd);
644 mnt = collect_mounts(nd.path.mnt, nd.path.dentry);
650 list_add_tail(&list, &mnt->mnt_list);
653 list_for_each_entry(p, &list, mnt_list) {
654 err = tag_chunk(p->mnt_root->d_inode, tree);
660 drop_collected_mounts(mnt);
664 spin_lock(&hash_lock);
665 list_for_each_entry(node, &tree->chunks, list)
666 node->index &= ~(1U<<31);
667 spin_unlock(&hash_lock);
673 mutex_lock(&audit_filter_mutex);
674 if (list_empty(&rule->rlist)) {
683 mutex_lock(&audit_filter_mutex);
684 list_del_init(&tree->list);
685 list_del_init(&tree->rules);
690 int audit_tag_tree(char *old, char *new)
692 struct list_head cursor, barrier;
695 struct vfsmount *tagged;
696 struct list_head list;
697 struct vfsmount *mnt;
698 struct dentry *dentry;
701 err = path_lookup(new, 0, &nd);
704 tagged = collect_mounts(nd.path.mnt, nd.path.dentry);
709 err = path_lookup(old, 0, &nd);
711 drop_collected_mounts(tagged);
714 mnt = mntget(nd.path.mnt);
715 dentry = dget(nd.path.dentry);
718 if (dentry == tagged->mnt_root && dentry == mnt->mnt_root)
719 follow_up(&mnt, &dentry);
721 list_add_tail(&list, &tagged->mnt_list);
723 mutex_lock(&audit_filter_mutex);
724 list_add(&barrier, &tree_list);
725 list_add(&cursor, &barrier);
727 while (cursor.next != &tree_list) {
728 struct audit_tree *tree;
731 tree = container_of(cursor.next, struct audit_tree, list);
734 list_add(&cursor, &tree->list);
735 mutex_unlock(&audit_filter_mutex);
737 err = path_lookup(tree->pathname, 0, &nd);
740 mutex_lock(&audit_filter_mutex);
744 spin_lock(&vfsmount_lock);
745 if (!is_under(mnt, dentry, &nd)) {
746 spin_unlock(&vfsmount_lock);
749 mutex_lock(&audit_filter_mutex);
752 spin_unlock(&vfsmount_lock);
755 list_for_each_entry(p, &list, mnt_list) {
756 failed = tag_chunk(p->mnt_root->d_inode, tree);
763 mutex_lock(&audit_filter_mutex);
767 mutex_lock(&audit_filter_mutex);
768 spin_lock(&hash_lock);
770 list_del(&tree->list);
771 list_add(&tree->list, &tree_list);
773 spin_unlock(&hash_lock);
777 while (barrier.prev != &tree_list) {
778 struct audit_tree *tree;
780 tree = container_of(barrier.prev, struct audit_tree, list);
782 list_del(&tree->list);
783 list_add(&tree->list, &barrier);
784 mutex_unlock(&audit_filter_mutex);
788 spin_lock(&hash_lock);
789 list_for_each_entry(node, &tree->chunks, list)
790 node->index &= ~(1U<<31);
791 spin_unlock(&hash_lock);
797 mutex_lock(&audit_filter_mutex);
802 mutex_unlock(&audit_filter_mutex);
805 drop_collected_mounts(tagged);
810 * That gets run when evict_chunk() ends up needing to kill audit_tree.
811 * Runs from a separate thread, with audit_cmd_mutex held.
813 void audit_prune_trees(void)
815 mutex_lock(&audit_filter_mutex);
817 while (!list_empty(&prune_list)) {
818 struct audit_tree *victim;
820 victim = list_entry(prune_list.next, struct audit_tree, list);
821 list_del_init(&victim->list);
823 mutex_unlock(&audit_filter_mutex);
827 mutex_lock(&audit_filter_mutex);
830 mutex_unlock(&audit_filter_mutex);
834 * Here comes the stuff asynchronous to auditctl operations
837 /* inode->inotify_mutex is locked */
838 static void evict_chunk(struct audit_chunk *chunk)
840 struct audit_tree *owner;
847 mutex_lock(&audit_filter_mutex);
848 spin_lock(&hash_lock);
849 while (!list_empty(&chunk->trees)) {
850 owner = list_entry(chunk->trees.next,
851 struct audit_tree, same_root);
854 list_del_init(&owner->same_root);
855 spin_unlock(&hash_lock);
857 list_move(&owner->list, &prune_list);
858 audit_schedule_prune();
859 spin_lock(&hash_lock);
861 list_del_rcu(&chunk->hash);
862 for (n = 0; n < chunk->count; n++)
863 list_del_init(&chunk->owners[n].list);
864 spin_unlock(&hash_lock);
865 mutex_unlock(&audit_filter_mutex);
868 static void handle_event(struct inotify_watch *watch, u32 wd, u32 mask,
869 u32 cookie, const char *dname, struct inode *inode)
871 struct audit_chunk *chunk = container_of(watch, struct audit_chunk, watch);
873 if (mask & IN_IGNORED) {
875 put_inotify_watch(watch);
879 static void destroy_watch(struct inotify_watch *watch)
881 struct audit_chunk *chunk = container_of(watch, struct audit_chunk, watch);
885 static const struct inotify_operations rtree_inotify_ops = {
886 .handle_event = handle_event,
887 .destroy_watch = destroy_watch,
890 static int __init audit_tree_init(void)
894 rtree_ih = inotify_init(&rtree_inotify_ops);
895 if (IS_ERR(rtree_ih))
896 audit_panic("cannot initialize inotify handle for rectree watches");
898 for (i = 0; i < HASH_SIZE; i++)
899 INIT_LIST_HEAD(&chunk_hash_heads[i]);
903 __initcall(audit_tree_init);