Linux Audio

Check our new training course

Loading...
v4.6
 
  1#include "audit.h"
  2#include <linux/fsnotify_backend.h>
  3#include <linux/namei.h>
  4#include <linux/mount.h>
  5#include <linux/kthread.h>
 
  6#include <linux/slab.h>
  7
  8struct audit_tree;
  9struct audit_chunk;
 10
 11struct audit_tree {
 12	atomic_t count;
 13	int goner;
 14	struct audit_chunk *root;
 15	struct list_head chunks;
 16	struct list_head rules;
 17	struct list_head list;
 18	struct list_head same_root;
 19	struct rcu_head head;
 20	char pathname[];
 21};
 22
 23struct audit_chunk {
 24	struct list_head hash;
 25	struct fsnotify_mark mark;
 26	struct list_head trees;		/* with root here */
 27	int dead;
 28	int count;
 29	atomic_long_t refs;
 30	struct rcu_head head;
 31	struct node {
 32		struct list_head list;
 33		struct audit_tree *owner;
 34		unsigned index;		/* index; upper bit indicates 'will prune' */
 35	} owners[];
 36};
 37
 38static LIST_HEAD(tree_list);
 39static LIST_HEAD(prune_list);
 40static struct task_struct *prune_thread;
 41
 42/*
 43 * One struct chunk is attached to each inode of interest.
 44 * We replace struct chunk on tagging/untagging.
 45 * Rules have pointer to struct audit_tree.
 46 * Rules have struct list_head rlist forming a list of rules over
 47 * the same tree.
 48 * References to struct chunk are collected at audit_inode{,_child}()
 49 * time and used in AUDIT_TREE rule matching.
 50 * These references are dropped at the same time we are calling
 51 * audit_free_names(), etc.
 52 *
 53 * Cyclic lists galore:
 54 * tree.chunks anchors chunk.owners[].list			hash_lock
 55 * tree.rules anchors rule.rlist				audit_filter_mutex
 56 * chunk.trees anchors tree.same_root				hash_lock
 57 * chunk.hash is a hash with middle bits of watch.inode as
 58 * a hash function.						RCU, hash_lock
 59 *
 60 * tree is refcounted; one reference for "some rules on rules_list refer to
 61 * it", one for each chunk with pointer to it.
 62 *
 63 * chunk is refcounted by embedded fsnotify_mark + .refs (non-zero refcount
 64 * of watch contributes 1 to .refs).
 65 *
 66 * node.index allows to get from node.list to containing chunk.
 67 * MSB of that sucker is stolen to mark taggings that we might have to
 68 * revert - several operations have very unpleasant cleanup logics and
 69 * that makes a difference.  Some.
 70 */
 71
 72static struct fsnotify_group *audit_tree_group;
 73
 74static struct audit_tree *alloc_tree(const char *s)
 75{
 76	struct audit_tree *tree;
 77
 78	tree = kmalloc(sizeof(struct audit_tree) + strlen(s) + 1, GFP_KERNEL);
 79	if (tree) {
 80		atomic_set(&tree->count, 1);
 81		tree->goner = 0;
 82		INIT_LIST_HEAD(&tree->chunks);
 83		INIT_LIST_HEAD(&tree->rules);
 84		INIT_LIST_HEAD(&tree->list);
 85		INIT_LIST_HEAD(&tree->same_root);
 86		tree->root = NULL;
 87		strcpy(tree->pathname, s);
 88	}
 89	return tree;
 90}
 91
 92static inline void get_tree(struct audit_tree *tree)
 93{
 94	atomic_inc(&tree->count);
 95}
 96
 97static inline void put_tree(struct audit_tree *tree)
 98{
 99	if (atomic_dec_and_test(&tree->count))
100		kfree_rcu(tree, head);
101}
102
103/* to avoid bringing the entire thing in audit.h */
104const char *audit_tree_path(struct audit_tree *tree)
105{
106	return tree->pathname;
107}
108
109static void free_chunk(struct audit_chunk *chunk)
110{
111	int i;
112
113	for (i = 0; i < chunk->count; i++) {
114		if (chunk->owners[i].owner)
115			put_tree(chunk->owners[i].owner);
116	}
117	kfree(chunk);
118}
119
120void audit_put_chunk(struct audit_chunk *chunk)
121{
122	if (atomic_long_dec_and_test(&chunk->refs))
123		free_chunk(chunk);
124}
125
126static void __put_chunk(struct rcu_head *rcu)
127{
128	struct audit_chunk *chunk = container_of(rcu, struct audit_chunk, head);
129	audit_put_chunk(chunk);
130}
131
132static void audit_tree_destroy_watch(struct fsnotify_mark *entry)
133{
134	struct audit_chunk *chunk = container_of(entry, struct audit_chunk, mark);
135	call_rcu(&chunk->head, __put_chunk);
136}
137
138static struct audit_chunk *alloc_chunk(int count)
139{
140	struct audit_chunk *chunk;
141	size_t size;
142	int i;
143
144	size = offsetof(struct audit_chunk, owners) + count * sizeof(struct node);
145	chunk = kzalloc(size, GFP_KERNEL);
146	if (!chunk)
147		return NULL;
148
149	INIT_LIST_HEAD(&chunk->hash);
150	INIT_LIST_HEAD(&chunk->trees);
151	chunk->count = count;
152	atomic_long_set(&chunk->refs, 1);
153	for (i = 0; i < count; i++) {
154		INIT_LIST_HEAD(&chunk->owners[i].list);
155		chunk->owners[i].index = i;
156	}
157	fsnotify_init_mark(&chunk->mark, audit_tree_destroy_watch);
158	chunk->mark.mask = FS_IN_IGNORED;
159	return chunk;
160}
161
162enum {HASH_SIZE = 128};
163static struct list_head chunk_hash_heads[HASH_SIZE];
164static __cacheline_aligned_in_smp DEFINE_SPINLOCK(hash_lock);
165
166static inline struct list_head *chunk_hash(const struct inode *inode)
 
167{
168	unsigned long n = (unsigned long)inode / L1_CACHE_BYTES;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
169	return chunk_hash_heads + n % HASH_SIZE;
170}
171
172/* hash_lock & entry->lock is held by caller */
173static void insert_hash(struct audit_chunk *chunk)
174{
175	struct fsnotify_mark *entry = &chunk->mark;
176	struct list_head *list;
177
178	if (!entry->inode)
179		return;
180	list = chunk_hash(entry->inode);
181	list_add_rcu(&chunk->hash, list);
182}
183
184/* called under rcu_read_lock */
185struct audit_chunk *audit_tree_lookup(const struct inode *inode)
186{
187	struct list_head *list = chunk_hash(inode);
 
188	struct audit_chunk *p;
189
190	list_for_each_entry_rcu(p, list, hash) {
191		/* mark.inode may have gone NULL, but who cares? */
192		if (p->mark.inode == inode) {
193			atomic_long_inc(&p->refs);
194			return p;
195		}
196	}
197	return NULL;
198}
199
200bool audit_tree_match(struct audit_chunk *chunk, struct audit_tree *tree)
201{
202	int n;
203	for (n = 0; n < chunk->count; n++)
204		if (chunk->owners[n].owner == tree)
205			return true;
206	return false;
207}
208
209/* tagging and untagging inodes with trees */
210
211static struct audit_chunk *find_chunk(struct node *p)
212{
213	int index = p->index & ~(1U<<31);
214	p -= index;
215	return container_of(p, struct audit_chunk, owners[0]);
216}
217
218static void untag_chunk(struct node *p)
219{
220	struct audit_chunk *chunk = find_chunk(p);
221	struct fsnotify_mark *entry = &chunk->mark;
222	struct audit_chunk *new = NULL;
223	struct audit_tree *owner;
224	int size = chunk->count - 1;
225	int i, j;
226
227	fsnotify_get_mark(entry);
228
229	spin_unlock(&hash_lock);
230
231	if (size)
232		new = alloc_chunk(size);
233
 
234	spin_lock(&entry->lock);
235	if (chunk->dead || !entry->inode) {
 
 
 
 
236		spin_unlock(&entry->lock);
 
237		if (new)
238			free_chunk(new);
239		goto out;
240	}
241
242	owner = p->owner;
243
244	if (!size) {
245		chunk->dead = 1;
246		spin_lock(&hash_lock);
247		list_del_init(&chunk->trees);
248		if (owner->root == chunk)
249			owner->root = NULL;
250		list_del_init(&p->list);
251		list_del_rcu(&chunk->hash);
252		spin_unlock(&hash_lock);
253		spin_unlock(&entry->lock);
 
254		fsnotify_destroy_mark(entry, audit_tree_group);
255		goto out;
256	}
257
258	if (!new)
259		goto Fallback;
260
261	fsnotify_duplicate_mark(&new->mark, entry);
262	if (fsnotify_add_mark(&new->mark, new->mark.group, new->mark.inode, NULL, 1)) {
263		fsnotify_put_mark(&new->mark);
264		goto Fallback;
265	}
266
267	chunk->dead = 1;
268	spin_lock(&hash_lock);
269	list_replace_init(&chunk->trees, &new->trees);
270	if (owner->root == chunk) {
271		list_del_init(&owner->same_root);
272		owner->root = NULL;
273	}
274
275	for (i = j = 0; j <= size; i++, j++) {
276		struct audit_tree *s;
277		if (&chunk->owners[j] == p) {
278			list_del_init(&p->list);
279			i--;
280			continue;
281		}
282		s = chunk->owners[j].owner;
283		new->owners[i].owner = s;
284		new->owners[i].index = chunk->owners[j].index - j + i;
285		if (!s) /* result of earlier fallback */
286			continue;
287		get_tree(s);
288		list_replace_init(&chunk->owners[j].list, &new->owners[i].list);
289	}
290
291	list_replace_rcu(&chunk->hash, &new->hash);
292	list_for_each_entry(owner, &new->trees, same_root)
293		owner->root = new;
294	spin_unlock(&hash_lock);
295	spin_unlock(&entry->lock);
 
296	fsnotify_destroy_mark(entry, audit_tree_group);
297	fsnotify_put_mark(&new->mark);	/* drop initial reference */
298	goto out;
299
300Fallback:
301	// do the best we can
302	spin_lock(&hash_lock);
303	if (owner->root == chunk) {
304		list_del_init(&owner->same_root);
305		owner->root = NULL;
306	}
307	list_del_init(&p->list);
308	p->owner = NULL;
309	put_tree(owner);
310	spin_unlock(&hash_lock);
311	spin_unlock(&entry->lock);
 
312out:
313	fsnotify_put_mark(entry);
314	spin_lock(&hash_lock);
315}
316
317static int create_chunk(struct inode *inode, struct audit_tree *tree)
318{
319	struct fsnotify_mark *entry;
320	struct audit_chunk *chunk = alloc_chunk(1);
321	if (!chunk)
322		return -ENOMEM;
323
324	entry = &chunk->mark;
325	if (fsnotify_add_mark(entry, audit_tree_group, inode, NULL, 0)) {
326		fsnotify_put_mark(entry);
327		return -ENOSPC;
328	}
329
330	spin_lock(&entry->lock);
331	spin_lock(&hash_lock);
332	if (tree->goner) {
333		spin_unlock(&hash_lock);
334		chunk->dead = 1;
335		spin_unlock(&entry->lock);
336		fsnotify_destroy_mark(entry, audit_tree_group);
337		fsnotify_put_mark(entry);
338		return 0;
339	}
340	chunk->owners[0].index = (1U << 31);
341	chunk->owners[0].owner = tree;
342	get_tree(tree);
343	list_add(&chunk->owners[0].list, &tree->chunks);
344	if (!tree->root) {
345		tree->root = chunk;
346		list_add(&tree->same_root, &chunk->trees);
347	}
348	insert_hash(chunk);
349	spin_unlock(&hash_lock);
350	spin_unlock(&entry->lock);
351	fsnotify_put_mark(entry);	/* drop initial reference */
352	return 0;
353}
354
355/* the first tagged inode becomes root of tree */
356static int tag_chunk(struct inode *inode, struct audit_tree *tree)
357{
358	struct fsnotify_mark *old_entry, *chunk_entry;
359	struct audit_tree *owner;
360	struct audit_chunk *chunk, *old;
361	struct node *p;
362	int n;
363
364	old_entry = fsnotify_find_inode_mark(audit_tree_group, inode);
 
365	if (!old_entry)
366		return create_chunk(inode, tree);
367
368	old = container_of(old_entry, struct audit_chunk, mark);
369
370	/* are we already there? */
371	spin_lock(&hash_lock);
372	for (n = 0; n < old->count; n++) {
373		if (old->owners[n].owner == tree) {
374			spin_unlock(&hash_lock);
375			fsnotify_put_mark(old_entry);
376			return 0;
377		}
378	}
379	spin_unlock(&hash_lock);
380
381	chunk = alloc_chunk(old->count + 1);
382	if (!chunk) {
383		fsnotify_put_mark(old_entry);
384		return -ENOMEM;
385	}
386
387	chunk_entry = &chunk->mark;
388
 
389	spin_lock(&old_entry->lock);
390	if (!old_entry->inode) {
 
 
 
 
391		/* old_entry is being shot, lets just lie */
392		spin_unlock(&old_entry->lock);
 
393		fsnotify_put_mark(old_entry);
394		free_chunk(chunk);
395		return -ENOENT;
396	}
397
398	fsnotify_duplicate_mark(chunk_entry, old_entry);
399	if (fsnotify_add_mark(chunk_entry, chunk_entry->group, chunk_entry->inode, NULL, 1)) {
400		spin_unlock(&old_entry->lock);
 
401		fsnotify_put_mark(chunk_entry);
402		fsnotify_put_mark(old_entry);
403		return -ENOSPC;
404	}
405
406	/* even though we hold old_entry->lock, this is safe since chunk_entry->lock could NEVER have been grabbed before */
407	spin_lock(&chunk_entry->lock);
408	spin_lock(&hash_lock);
409
410	/* we now hold old_entry->lock, chunk_entry->lock, and hash_lock */
411	if (tree->goner) {
412		spin_unlock(&hash_lock);
413		chunk->dead = 1;
414		spin_unlock(&chunk_entry->lock);
415		spin_unlock(&old_entry->lock);
 
416
417		fsnotify_destroy_mark(chunk_entry, audit_tree_group);
418
419		fsnotify_put_mark(chunk_entry);
420		fsnotify_put_mark(old_entry);
421		return 0;
422	}
423	list_replace_init(&old->trees, &chunk->trees);
424	for (n = 0, p = chunk->owners; n < old->count; n++, p++) {
425		struct audit_tree *s = old->owners[n].owner;
426		p->owner = s;
427		p->index = old->owners[n].index;
428		if (!s) /* result of fallback in untag */
429			continue;
430		get_tree(s);
431		list_replace_init(&old->owners[n].list, &p->list);
432	}
433	p->index = (chunk->count - 1) | (1U<<31);
434	p->owner = tree;
435	get_tree(tree);
436	list_add(&p->list, &tree->chunks);
437	list_replace_rcu(&old->hash, &chunk->hash);
438	list_for_each_entry(owner, &chunk->trees, same_root)
439		owner->root = chunk;
440	old->dead = 1;
441	if (!tree->root) {
442		tree->root = chunk;
443		list_add(&tree->same_root, &chunk->trees);
444	}
445	spin_unlock(&hash_lock);
446	spin_unlock(&chunk_entry->lock);
447	spin_unlock(&old_entry->lock);
 
448	fsnotify_destroy_mark(old_entry, audit_tree_group);
449	fsnotify_put_mark(chunk_entry);	/* drop initial reference */
450	fsnotify_put_mark(old_entry); /* pair to fsnotify_find mark_entry */
451	return 0;
452}
453
454static void audit_tree_log_remove_rule(struct audit_krule *rule)
455{
456	struct audit_buffer *ab;
457
458	ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE);
459	if (unlikely(!ab))
460		return;
461	audit_log_format(ab, "op=");
462	audit_log_string(ab, "remove_rule");
463	audit_log_format(ab, " dir=");
464	audit_log_untrustedstring(ab, rule->tree->pathname);
465	audit_log_key(ab, rule->filterkey);
466	audit_log_format(ab, " list=%d res=1", rule->listnr);
467	audit_log_end(ab);
468}
469
470static void kill_rules(struct audit_tree *tree)
471{
472	struct audit_krule *rule, *next;
473	struct audit_entry *entry;
474
475	list_for_each_entry_safe(rule, next, &tree->rules, rlist) {
476		entry = container_of(rule, struct audit_entry, rule);
477
478		list_del_init(&rule->rlist);
479		if (rule->tree) {
480			/* not a half-baked one */
481			audit_tree_log_remove_rule(rule);
482			if (entry->rule.exe)
483				audit_remove_mark(entry->rule.exe);
484			rule->tree = NULL;
485			list_del_rcu(&entry->list);
486			list_del(&entry->rule.list);
487			call_rcu(&entry->rcu, audit_free_rule_rcu);
488		}
489	}
490}
491
492/*
493 * finish killing struct audit_tree
494 */
495static void prune_one(struct audit_tree *victim)
496{
497	spin_lock(&hash_lock);
498	while (!list_empty(&victim->chunks)) {
499		struct node *p;
500
501		p = list_entry(victim->chunks.next, struct node, list);
502
503		untag_chunk(p);
504	}
505	spin_unlock(&hash_lock);
506	put_tree(victim);
507}
508
509/* trim the uncommitted chunks from tree */
510
511static void trim_marked(struct audit_tree *tree)
512{
513	struct list_head *p, *q;
514	spin_lock(&hash_lock);
515	if (tree->goner) {
516		spin_unlock(&hash_lock);
517		return;
518	}
519	/* reorder */
520	for (p = tree->chunks.next; p != &tree->chunks; p = q) {
521		struct node *node = list_entry(p, struct node, list);
522		q = p->next;
523		if (node->index & (1U<<31)) {
524			list_del_init(p);
525			list_add(p, &tree->chunks);
526		}
527	}
528
529	while (!list_empty(&tree->chunks)) {
530		struct node *node;
531
532		node = list_entry(tree->chunks.next, struct node, list);
533
534		/* have we run out of marked? */
535		if (!(node->index & (1U<<31)))
536			break;
537
538		untag_chunk(node);
539	}
540	if (!tree->root && !tree->goner) {
541		tree->goner = 1;
542		spin_unlock(&hash_lock);
543		mutex_lock(&audit_filter_mutex);
544		kill_rules(tree);
545		list_del_init(&tree->list);
546		mutex_unlock(&audit_filter_mutex);
547		prune_one(tree);
548	} else {
549		spin_unlock(&hash_lock);
550	}
551}
552
553static void audit_schedule_prune(void);
554
555/* called with audit_filter_mutex */
556int audit_remove_tree_rule(struct audit_krule *rule)
557{
558	struct audit_tree *tree;
559	tree = rule->tree;
560	if (tree) {
561		spin_lock(&hash_lock);
562		list_del_init(&rule->rlist);
563		if (list_empty(&tree->rules) && !tree->goner) {
564			tree->root = NULL;
565			list_del_init(&tree->same_root);
566			tree->goner = 1;
567			list_move(&tree->list, &prune_list);
568			rule->tree = NULL;
569			spin_unlock(&hash_lock);
570			audit_schedule_prune();
571			return 1;
572		}
573		rule->tree = NULL;
574		spin_unlock(&hash_lock);
575		return 1;
576	}
577	return 0;
578}
579
580static int compare_root(struct vfsmount *mnt, void *arg)
581{
582	return d_backing_inode(mnt->mnt_root) == arg;
 
583}
584
585void audit_trim_trees(void)
586{
587	struct list_head cursor;
588
589	mutex_lock(&audit_filter_mutex);
590	list_add(&cursor, &tree_list);
591	while (cursor.next != &tree_list) {
592		struct audit_tree *tree;
593		struct path path;
594		struct vfsmount *root_mnt;
595		struct node *node;
596		int err;
597
598		tree = container_of(cursor.next, struct audit_tree, list);
599		get_tree(tree);
600		list_del(&cursor);
601		list_add(&cursor, &tree->list);
602		mutex_unlock(&audit_filter_mutex);
603
604		err = kern_path(tree->pathname, 0, &path);
605		if (err)
606			goto skip_it;
607
608		root_mnt = collect_mounts(&path);
609		path_put(&path);
610		if (IS_ERR(root_mnt))
611			goto skip_it;
612
613		spin_lock(&hash_lock);
614		list_for_each_entry(node, &tree->chunks, list) {
615			struct audit_chunk *chunk = find_chunk(node);
616			/* this could be NULL if the watch is dying else where... */
617			struct inode *inode = chunk->mark.inode;
618			node->index |= 1U<<31;
619			if (iterate_mounts(compare_root, inode, root_mnt))
 
 
620				node->index &= ~(1U<<31);
621		}
622		spin_unlock(&hash_lock);
623		trim_marked(tree);
624		drop_collected_mounts(root_mnt);
625skip_it:
626		put_tree(tree);
627		mutex_lock(&audit_filter_mutex);
628	}
629	list_del(&cursor);
630	mutex_unlock(&audit_filter_mutex);
631}
632
633int audit_make_tree(struct audit_krule *rule, char *pathname, u32 op)
634{
635
636	if (pathname[0] != '/' ||
637	    rule->listnr != AUDIT_FILTER_EXIT ||
638	    op != Audit_equal ||
639	    rule->inode_f || rule->watch || rule->tree)
640		return -EINVAL;
641	rule->tree = alloc_tree(pathname);
642	if (!rule->tree)
643		return -ENOMEM;
644	return 0;
645}
646
647void audit_put_tree(struct audit_tree *tree)
648{
649	put_tree(tree);
650}
651
652static int tag_mount(struct vfsmount *mnt, void *arg)
653{
654	return tag_chunk(d_backing_inode(mnt->mnt_root), arg);
655}
656
657/*
658 * That gets run when evict_chunk() ends up needing to kill audit_tree.
659 * Runs from a separate thread.
660 */
661static int prune_tree_thread(void *unused)
662{
663	for (;;) {
664		set_current_state(TASK_INTERRUPTIBLE);
665		if (list_empty(&prune_list))
666			schedule();
667		__set_current_state(TASK_RUNNING);
668
669		mutex_lock(&audit_cmd_mutex);
670		mutex_lock(&audit_filter_mutex);
671
672		while (!list_empty(&prune_list)) {
673			struct audit_tree *victim;
674
675			victim = list_entry(prune_list.next,
676					struct audit_tree, list);
677			list_del_init(&victim->list);
678
679			mutex_unlock(&audit_filter_mutex);
680
681			prune_one(victim);
682
683			mutex_lock(&audit_filter_mutex);
684		}
685
686		mutex_unlock(&audit_filter_mutex);
687		mutex_unlock(&audit_cmd_mutex);
688	}
689	return 0;
690}
691
692static int audit_launch_prune(void)
693{
694	if (prune_thread)
695		return 0;
696	prune_thread = kthread_create(prune_tree_thread, NULL,
697				"audit_prune_tree");
698	if (IS_ERR(prune_thread)) {
699		pr_err("cannot start thread audit_prune_tree");
700		prune_thread = NULL;
701		return -ENOMEM;
702	} else {
703		wake_up_process(prune_thread);
704		return 0;
705	}
 
706}
707
708/* called with audit_filter_mutex */
709int audit_add_tree_rule(struct audit_krule *rule)
710{
711	struct audit_tree *seed = rule->tree, *tree;
712	struct path path;
713	struct vfsmount *mnt;
714	int err;
715
716	rule->tree = NULL;
717	list_for_each_entry(tree, &tree_list, list) {
718		if (!strcmp(seed->pathname, tree->pathname)) {
719			put_tree(seed);
720			rule->tree = tree;
721			list_add(&rule->rlist, &tree->rules);
722			return 0;
723		}
724	}
725	tree = seed;
726	list_add(&tree->list, &tree_list);
727	list_add(&rule->rlist, &tree->rules);
728	/* do not set rule->tree yet */
729	mutex_unlock(&audit_filter_mutex);
730
731	if (unlikely(!prune_thread)) {
732		err = audit_launch_prune();
733		if (err)
734			goto Err;
735	}
736
737	err = kern_path(tree->pathname, 0, &path);
738	if (err)
739		goto Err;
740	mnt = collect_mounts(&path);
741	path_put(&path);
742	if (IS_ERR(mnt)) {
743		err = PTR_ERR(mnt);
744		goto Err;
745	}
746
747	get_tree(tree);
748	err = iterate_mounts(tag_mount, tree, mnt);
749	drop_collected_mounts(mnt);
750
751	if (!err) {
752		struct node *node;
753		spin_lock(&hash_lock);
754		list_for_each_entry(node, &tree->chunks, list)
755			node->index &= ~(1U<<31);
756		spin_unlock(&hash_lock);
757	} else {
758		trim_marked(tree);
759		goto Err;
760	}
761
762	mutex_lock(&audit_filter_mutex);
763	if (list_empty(&rule->rlist)) {
764		put_tree(tree);
765		return -ENOENT;
766	}
767	rule->tree = tree;
768	put_tree(tree);
769
770	return 0;
771Err:
772	mutex_lock(&audit_filter_mutex);
773	list_del_init(&tree->list);
774	list_del_init(&tree->rules);
775	put_tree(tree);
776	return err;
777}
778
779int audit_tag_tree(char *old, char *new)
780{
781	struct list_head cursor, barrier;
782	int failed = 0;
783	struct path path1, path2;
784	struct vfsmount *tagged;
785	int err;
786
787	err = kern_path(new, 0, &path2);
788	if (err)
789		return err;
790	tagged = collect_mounts(&path2);
791	path_put(&path2);
792	if (IS_ERR(tagged))
793		return PTR_ERR(tagged);
794
795	err = kern_path(old, 0, &path1);
796	if (err) {
797		drop_collected_mounts(tagged);
798		return err;
799	}
800
801	mutex_lock(&audit_filter_mutex);
802	list_add(&barrier, &tree_list);
803	list_add(&cursor, &barrier);
804
805	while (cursor.next != &tree_list) {
806		struct audit_tree *tree;
807		int good_one = 0;
808
809		tree = container_of(cursor.next, struct audit_tree, list);
810		get_tree(tree);
811		list_del(&cursor);
812		list_add(&cursor, &tree->list);
813		mutex_unlock(&audit_filter_mutex);
814
815		err = kern_path(tree->pathname, 0, &path2);
816		if (!err) {
817			good_one = path_is_under(&path1, &path2);
818			path_put(&path2);
819		}
820
821		if (!good_one) {
822			put_tree(tree);
823			mutex_lock(&audit_filter_mutex);
824			continue;
825		}
826
827		failed = iterate_mounts(tag_mount, tree, tagged);
828		if (failed) {
829			put_tree(tree);
830			mutex_lock(&audit_filter_mutex);
831			break;
832		}
833
834		mutex_lock(&audit_filter_mutex);
835		spin_lock(&hash_lock);
836		if (!tree->goner) {
837			list_del(&tree->list);
838			list_add(&tree->list, &tree_list);
839		}
840		spin_unlock(&hash_lock);
841		put_tree(tree);
842	}
843
844	while (barrier.prev != &tree_list) {
845		struct audit_tree *tree;
846
847		tree = container_of(barrier.prev, struct audit_tree, list);
848		get_tree(tree);
849		list_del(&tree->list);
850		list_add(&tree->list, &barrier);
851		mutex_unlock(&audit_filter_mutex);
852
853		if (!failed) {
854			struct node *node;
855			spin_lock(&hash_lock);
856			list_for_each_entry(node, &tree->chunks, list)
857				node->index &= ~(1U<<31);
858			spin_unlock(&hash_lock);
859		} else {
860			trim_marked(tree);
861		}
862
863		put_tree(tree);
864		mutex_lock(&audit_filter_mutex);
865	}
866	list_del(&barrier);
867	list_del(&cursor);
868	mutex_unlock(&audit_filter_mutex);
869	path_put(&path1);
870	drop_collected_mounts(tagged);
871	return failed;
872}
873
874
875static void audit_schedule_prune(void)
876{
877	wake_up_process(prune_thread);
878}
879
880/*
881 * ... and that one is done if evict_chunk() decides to delay until the end
882 * of syscall.  Runs synchronously.
883 */
884void audit_kill_trees(struct list_head *list)
885{
886	mutex_lock(&audit_cmd_mutex);
887	mutex_lock(&audit_filter_mutex);
888
889	while (!list_empty(list)) {
890		struct audit_tree *victim;
891
892		victim = list_entry(list->next, struct audit_tree, list);
893		kill_rules(victim);
894		list_del_init(&victim->list);
895
896		mutex_unlock(&audit_filter_mutex);
897
898		prune_one(victim);
899
900		mutex_lock(&audit_filter_mutex);
901	}
902
903	mutex_unlock(&audit_filter_mutex);
904	mutex_unlock(&audit_cmd_mutex);
905}
906
907/*
908 *  Here comes the stuff asynchronous to auditctl operations
909 */
910
911static void evict_chunk(struct audit_chunk *chunk)
912{
913	struct audit_tree *owner;
914	struct list_head *postponed = audit_killed_trees();
915	int need_prune = 0;
916	int n;
917
918	if (chunk->dead)
919		return;
920
921	chunk->dead = 1;
922	mutex_lock(&audit_filter_mutex);
923	spin_lock(&hash_lock);
924	while (!list_empty(&chunk->trees)) {
925		owner = list_entry(chunk->trees.next,
926				   struct audit_tree, same_root);
927		owner->goner = 1;
928		owner->root = NULL;
929		list_del_init(&owner->same_root);
930		spin_unlock(&hash_lock);
931		if (!postponed) {
932			kill_rules(owner);
933			list_move(&owner->list, &prune_list);
934			need_prune = 1;
935		} else {
936			list_move(&owner->list, postponed);
937		}
938		spin_lock(&hash_lock);
939	}
940	list_del_rcu(&chunk->hash);
941	for (n = 0; n < chunk->count; n++)
942		list_del_init(&chunk->owners[n].list);
943	spin_unlock(&hash_lock);
944	mutex_unlock(&audit_filter_mutex);
945	if (need_prune)
946		audit_schedule_prune();
947}
948
949static int audit_tree_handle_event(struct fsnotify_group *group,
950				   struct inode *to_tell,
951				   struct fsnotify_mark *inode_mark,
952				   struct fsnotify_mark *vfsmount_mark,
953				   u32 mask, void *data, int data_type,
954				   const unsigned char *file_name, u32 cookie)
 
955{
956	return 0;
957}
958
959static void audit_tree_freeing_mark(struct fsnotify_mark *entry, struct fsnotify_group *group)
960{
961	struct audit_chunk *chunk = container_of(entry, struct audit_chunk, mark);
962
963	evict_chunk(chunk);
964
965	/*
966	 * We are guaranteed to have at least one reference to the mark from
967	 * either the inode or the caller of fsnotify_destroy_mark().
968	 */
969	BUG_ON(atomic_read(&entry->refcnt) < 1);
970}
971
972static const struct fsnotify_ops audit_tree_ops = {
973	.handle_event = audit_tree_handle_event,
974	.freeing_mark = audit_tree_freeing_mark,
 
975};
976
977static int __init audit_tree_init(void)
978{
979	int i;
980
981	audit_tree_group = fsnotify_alloc_group(&audit_tree_ops);
982	if (IS_ERR(audit_tree_group))
983		audit_panic("cannot initialize fsnotify group for rectree watches");
984
985	for (i = 0; i < HASH_SIZE; i++)
986		INIT_LIST_HEAD(&chunk_hash_heads[i]);
987
988	return 0;
989}
990__initcall(audit_tree_init);
v4.17
   1// SPDX-License-Identifier: GPL-2.0
   2#include "audit.h"
   3#include <linux/fsnotify_backend.h>
   4#include <linux/namei.h>
   5#include <linux/mount.h>
   6#include <linux/kthread.h>
   7#include <linux/refcount.h>
   8#include <linux/slab.h>
   9
  10struct audit_tree;
  11struct audit_chunk;
  12
  13struct audit_tree {
  14	refcount_t count;
  15	int goner;
  16	struct audit_chunk *root;
  17	struct list_head chunks;
  18	struct list_head rules;
  19	struct list_head list;
  20	struct list_head same_root;
  21	struct rcu_head head;
  22	char pathname[];
  23};
  24
  25struct audit_chunk {
  26	struct list_head hash;
  27	struct fsnotify_mark mark;
  28	struct list_head trees;		/* with root here */
  29	int dead;
  30	int count;
  31	atomic_long_t refs;
  32	struct rcu_head head;
  33	struct node {
  34		struct list_head list;
  35		struct audit_tree *owner;
  36		unsigned index;		/* index; upper bit indicates 'will prune' */
  37	} owners[];
  38};
  39
  40static LIST_HEAD(tree_list);
  41static LIST_HEAD(prune_list);
  42static struct task_struct *prune_thread;
  43
  44/*
  45 * One struct chunk is attached to each inode of interest.
  46 * We replace struct chunk on tagging/untagging.
  47 * Rules have pointer to struct audit_tree.
  48 * Rules have struct list_head rlist forming a list of rules over
  49 * the same tree.
  50 * References to struct chunk are collected at audit_inode{,_child}()
  51 * time and used in AUDIT_TREE rule matching.
  52 * These references are dropped at the same time we are calling
  53 * audit_free_names(), etc.
  54 *
  55 * Cyclic lists galore:
  56 * tree.chunks anchors chunk.owners[].list			hash_lock
  57 * tree.rules anchors rule.rlist				audit_filter_mutex
  58 * chunk.trees anchors tree.same_root				hash_lock
  59 * chunk.hash is a hash with middle bits of watch.inode as
  60 * a hash function.						RCU, hash_lock
  61 *
  62 * tree is refcounted; one reference for "some rules on rules_list refer to
  63 * it", one for each chunk with pointer to it.
  64 *
  65 * chunk is refcounted by embedded fsnotify_mark + .refs (non-zero refcount
  66 * of watch contributes 1 to .refs).
  67 *
  68 * node.index allows to get from node.list to containing chunk.
  69 * MSB of that sucker is stolen to mark taggings that we might have to
  70 * revert - several operations have very unpleasant cleanup logics and
  71 * that makes a difference.  Some.
  72 */
  73
  74static struct fsnotify_group *audit_tree_group;
  75
  76static struct audit_tree *alloc_tree(const char *s)
  77{
  78	struct audit_tree *tree;
  79
  80	tree = kmalloc(sizeof(struct audit_tree) + strlen(s) + 1, GFP_KERNEL);
  81	if (tree) {
  82		refcount_set(&tree->count, 1);
  83		tree->goner = 0;
  84		INIT_LIST_HEAD(&tree->chunks);
  85		INIT_LIST_HEAD(&tree->rules);
  86		INIT_LIST_HEAD(&tree->list);
  87		INIT_LIST_HEAD(&tree->same_root);
  88		tree->root = NULL;
  89		strcpy(tree->pathname, s);
  90	}
  91	return tree;
  92}
  93
  94static inline void get_tree(struct audit_tree *tree)
  95{
  96	refcount_inc(&tree->count);
  97}
  98
  99static inline void put_tree(struct audit_tree *tree)
 100{
 101	if (refcount_dec_and_test(&tree->count))
 102		kfree_rcu(tree, head);
 103}
 104
 105/* to avoid bringing the entire thing in audit.h */
 106const char *audit_tree_path(struct audit_tree *tree)
 107{
 108	return tree->pathname;
 109}
 110
 111static void free_chunk(struct audit_chunk *chunk)
 112{
 113	int i;
 114
 115	for (i = 0; i < chunk->count; i++) {
 116		if (chunk->owners[i].owner)
 117			put_tree(chunk->owners[i].owner);
 118	}
 119	kfree(chunk);
 120}
 121
 122void audit_put_chunk(struct audit_chunk *chunk)
 123{
 124	if (atomic_long_dec_and_test(&chunk->refs))
 125		free_chunk(chunk);
 126}
 127
 128static void __put_chunk(struct rcu_head *rcu)
 129{
 130	struct audit_chunk *chunk = container_of(rcu, struct audit_chunk, head);
 131	audit_put_chunk(chunk);
 132}
 133
 134static void audit_tree_destroy_watch(struct fsnotify_mark *entry)
 135{
 136	struct audit_chunk *chunk = container_of(entry, struct audit_chunk, mark);
 137	call_rcu(&chunk->head, __put_chunk);
 138}
 139
 140static struct audit_chunk *alloc_chunk(int count)
 141{
 142	struct audit_chunk *chunk;
 143	size_t size;
 144	int i;
 145
 146	size = offsetof(struct audit_chunk, owners) + count * sizeof(struct node);
 147	chunk = kzalloc(size, GFP_KERNEL);
 148	if (!chunk)
 149		return NULL;
 150
 151	INIT_LIST_HEAD(&chunk->hash);
 152	INIT_LIST_HEAD(&chunk->trees);
 153	chunk->count = count;
 154	atomic_long_set(&chunk->refs, 1);
 155	for (i = 0; i < count; i++) {
 156		INIT_LIST_HEAD(&chunk->owners[i].list);
 157		chunk->owners[i].index = i;
 158	}
 159	fsnotify_init_mark(&chunk->mark, audit_tree_group);
 160	chunk->mark.mask = FS_IN_IGNORED;
 161	return chunk;
 162}
 163
 164enum {HASH_SIZE = 128};
 165static struct list_head chunk_hash_heads[HASH_SIZE];
 166static __cacheline_aligned_in_smp DEFINE_SPINLOCK(hash_lock);
 167
 168/* Function to return search key in our hash from inode. */
 169static unsigned long inode_to_key(const struct inode *inode)
 170{
 171	return (unsigned long)inode;
 172}
 173
 174/*
 175 * Function to return search key in our hash from chunk. Key 0 is special and
 176 * should never be present in the hash.
 177 */
 178static unsigned long chunk_to_key(struct audit_chunk *chunk)
 179{
 180	/*
 181	 * We have a reference to the mark so it should be attached to a
 182	 * connector.
 183	 */
 184	if (WARN_ON_ONCE(!chunk->mark.connector))
 185		return 0;
 186	return (unsigned long)chunk->mark.connector->inode;
 187}
 188
 189static inline struct list_head *chunk_hash(unsigned long key)
 190{
 191	unsigned long n = key / L1_CACHE_BYTES;
 192	return chunk_hash_heads + n % HASH_SIZE;
 193}
 194
 195/* hash_lock & entry->lock is held by caller */
 196static void insert_hash(struct audit_chunk *chunk)
 197{
 198	unsigned long key = chunk_to_key(chunk);
 199	struct list_head *list;
 200
 201	if (!(chunk->mark.flags & FSNOTIFY_MARK_FLAG_ATTACHED))
 202		return;
 203	list = chunk_hash(key);
 204	list_add_rcu(&chunk->hash, list);
 205}
 206
 207/* called under rcu_read_lock */
 208struct audit_chunk *audit_tree_lookup(const struct inode *inode)
 209{
 210	unsigned long key = inode_to_key(inode);
 211	struct list_head *list = chunk_hash(key);
 212	struct audit_chunk *p;
 213
 214	list_for_each_entry_rcu(p, list, hash) {
 215		if (chunk_to_key(p) == key) {
 
 216			atomic_long_inc(&p->refs);
 217			return p;
 218		}
 219	}
 220	return NULL;
 221}
 222
 223bool audit_tree_match(struct audit_chunk *chunk, struct audit_tree *tree)
 224{
 225	int n;
 226	for (n = 0; n < chunk->count; n++)
 227		if (chunk->owners[n].owner == tree)
 228			return true;
 229	return false;
 230}
 231
 232/* tagging and untagging inodes with trees */
 233
 234static struct audit_chunk *find_chunk(struct node *p)
 235{
 236	int index = p->index & ~(1U<<31);
 237	p -= index;
 238	return container_of(p, struct audit_chunk, owners[0]);
 239}
 240
 241static void untag_chunk(struct node *p)
 242{
 243	struct audit_chunk *chunk = find_chunk(p);
 244	struct fsnotify_mark *entry = &chunk->mark;
 245	struct audit_chunk *new = NULL;
 246	struct audit_tree *owner;
 247	int size = chunk->count - 1;
 248	int i, j;
 249
 250	fsnotify_get_mark(entry);
 251
 252	spin_unlock(&hash_lock);
 253
 254	if (size)
 255		new = alloc_chunk(size);
 256
 257	mutex_lock(&entry->group->mark_mutex);
 258	spin_lock(&entry->lock);
 259	/*
 260	 * mark_mutex protects mark from getting detached and thus also from
 261	 * mark->connector->inode getting NULL.
 262	 */
 263	if (chunk->dead || !(entry->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) {
 264		spin_unlock(&entry->lock);
 265		mutex_unlock(&entry->group->mark_mutex);
 266		if (new)
 267			fsnotify_put_mark(&new->mark);
 268		goto out;
 269	}
 270
 271	owner = p->owner;
 272
 273	if (!size) {
 274		chunk->dead = 1;
 275		spin_lock(&hash_lock);
 276		list_del_init(&chunk->trees);
 277		if (owner->root == chunk)
 278			owner->root = NULL;
 279		list_del_init(&p->list);
 280		list_del_rcu(&chunk->hash);
 281		spin_unlock(&hash_lock);
 282		spin_unlock(&entry->lock);
 283		mutex_unlock(&entry->group->mark_mutex);
 284		fsnotify_destroy_mark(entry, audit_tree_group);
 285		goto out;
 286	}
 287
 288	if (!new)
 289		goto Fallback;
 290
 291	if (fsnotify_add_mark_locked(&new->mark, entry->connector->inode,
 292				     NULL, 1)) {
 293		fsnotify_put_mark(&new->mark);
 294		goto Fallback;
 295	}
 296
 297	chunk->dead = 1;
 298	spin_lock(&hash_lock);
 299	list_replace_init(&chunk->trees, &new->trees);
 300	if (owner->root == chunk) {
 301		list_del_init(&owner->same_root);
 302		owner->root = NULL;
 303	}
 304
 305	for (i = j = 0; j <= size; i++, j++) {
 306		struct audit_tree *s;
 307		if (&chunk->owners[j] == p) {
 308			list_del_init(&p->list);
 309			i--;
 310			continue;
 311		}
 312		s = chunk->owners[j].owner;
 313		new->owners[i].owner = s;
 314		new->owners[i].index = chunk->owners[j].index - j + i;
 315		if (!s) /* result of earlier fallback */
 316			continue;
 317		get_tree(s);
 318		list_replace_init(&chunk->owners[j].list, &new->owners[i].list);
 319	}
 320
 321	list_replace_rcu(&chunk->hash, &new->hash);
 322	list_for_each_entry(owner, &new->trees, same_root)
 323		owner->root = new;
 324	spin_unlock(&hash_lock);
 325	spin_unlock(&entry->lock);
 326	mutex_unlock(&entry->group->mark_mutex);
 327	fsnotify_destroy_mark(entry, audit_tree_group);
 328	fsnotify_put_mark(&new->mark);	/* drop initial reference */
 329	goto out;
 330
 331Fallback:
 332	// do the best we can
 333	spin_lock(&hash_lock);
 334	if (owner->root == chunk) {
 335		list_del_init(&owner->same_root);
 336		owner->root = NULL;
 337	}
 338	list_del_init(&p->list);
 339	p->owner = NULL;
 340	put_tree(owner);
 341	spin_unlock(&hash_lock);
 342	spin_unlock(&entry->lock);
 343	mutex_unlock(&entry->group->mark_mutex);
 344out:
 345	fsnotify_put_mark(entry);
 346	spin_lock(&hash_lock);
 347}
 348
 349static int create_chunk(struct inode *inode, struct audit_tree *tree)
 350{
 351	struct fsnotify_mark *entry;
 352	struct audit_chunk *chunk = alloc_chunk(1);
 353	if (!chunk)
 354		return -ENOMEM;
 355
 356	entry = &chunk->mark;
 357	if (fsnotify_add_mark(entry, inode, NULL, 0)) {
 358		fsnotify_put_mark(entry);
 359		return -ENOSPC;
 360	}
 361
 362	spin_lock(&entry->lock);
 363	spin_lock(&hash_lock);
 364	if (tree->goner) {
 365		spin_unlock(&hash_lock);
 366		chunk->dead = 1;
 367		spin_unlock(&entry->lock);
 368		fsnotify_destroy_mark(entry, audit_tree_group);
 369		fsnotify_put_mark(entry);
 370		return 0;
 371	}
 372	chunk->owners[0].index = (1U << 31);
 373	chunk->owners[0].owner = tree;
 374	get_tree(tree);
 375	list_add(&chunk->owners[0].list, &tree->chunks);
 376	if (!tree->root) {
 377		tree->root = chunk;
 378		list_add(&tree->same_root, &chunk->trees);
 379	}
 380	insert_hash(chunk);
 381	spin_unlock(&hash_lock);
 382	spin_unlock(&entry->lock);
 383	fsnotify_put_mark(entry);	/* drop initial reference */
 384	return 0;
 385}
 386
 387/* the first tagged inode becomes root of tree */
 388static int tag_chunk(struct inode *inode, struct audit_tree *tree)
 389{
 390	struct fsnotify_mark *old_entry, *chunk_entry;
 391	struct audit_tree *owner;
 392	struct audit_chunk *chunk, *old;
 393	struct node *p;
 394	int n;
 395
 396	old_entry = fsnotify_find_mark(&inode->i_fsnotify_marks,
 397				       audit_tree_group);
 398	if (!old_entry)
 399		return create_chunk(inode, tree);
 400
 401	old = container_of(old_entry, struct audit_chunk, mark);
 402
 403	/* are we already there? */
 404	spin_lock(&hash_lock);
 405	for (n = 0; n < old->count; n++) {
 406		if (old->owners[n].owner == tree) {
 407			spin_unlock(&hash_lock);
 408			fsnotify_put_mark(old_entry);
 409			return 0;
 410		}
 411	}
 412	spin_unlock(&hash_lock);
 413
 414	chunk = alloc_chunk(old->count + 1);
 415	if (!chunk) {
 416		fsnotify_put_mark(old_entry);
 417		return -ENOMEM;
 418	}
 419
 420	chunk_entry = &chunk->mark;
 421
 422	mutex_lock(&old_entry->group->mark_mutex);
 423	spin_lock(&old_entry->lock);
 424	/*
 425	 * mark_mutex protects mark from getting detached and thus also from
 426	 * mark->connector->inode getting NULL.
 427	 */
 428	if (!(old_entry->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) {
 429		/* old_entry is being shot, lets just lie */
 430		spin_unlock(&old_entry->lock);
 431		mutex_unlock(&old_entry->group->mark_mutex);
 432		fsnotify_put_mark(old_entry);
 433		fsnotify_put_mark(&chunk->mark);
 434		return -ENOENT;
 435	}
 436
 437	if (fsnotify_add_mark_locked(chunk_entry,
 438			     old_entry->connector->inode, NULL, 1)) {
 439		spin_unlock(&old_entry->lock);
 440		mutex_unlock(&old_entry->group->mark_mutex);
 441		fsnotify_put_mark(chunk_entry);
 442		fsnotify_put_mark(old_entry);
 443		return -ENOSPC;
 444	}
 445
 446	/* even though we hold old_entry->lock, this is safe since chunk_entry->lock could NEVER have been grabbed before */
 447	spin_lock(&chunk_entry->lock);
 448	spin_lock(&hash_lock);
 449
 450	/* we now hold old_entry->lock, chunk_entry->lock, and hash_lock */
 451	if (tree->goner) {
 452		spin_unlock(&hash_lock);
 453		chunk->dead = 1;
 454		spin_unlock(&chunk_entry->lock);
 455		spin_unlock(&old_entry->lock);
 456		mutex_unlock(&old_entry->group->mark_mutex);
 457
 458		fsnotify_destroy_mark(chunk_entry, audit_tree_group);
 459
 460		fsnotify_put_mark(chunk_entry);
 461		fsnotify_put_mark(old_entry);
 462		return 0;
 463	}
 464	list_replace_init(&old->trees, &chunk->trees);
 465	for (n = 0, p = chunk->owners; n < old->count; n++, p++) {
 466		struct audit_tree *s = old->owners[n].owner;
 467		p->owner = s;
 468		p->index = old->owners[n].index;
 469		if (!s) /* result of fallback in untag */
 470			continue;
 471		get_tree(s);
 472		list_replace_init(&old->owners[n].list, &p->list);
 473	}
 474	p->index = (chunk->count - 1) | (1U<<31);
 475	p->owner = tree;
 476	get_tree(tree);
 477	list_add(&p->list, &tree->chunks);
 478	list_replace_rcu(&old->hash, &chunk->hash);
 479	list_for_each_entry(owner, &chunk->trees, same_root)
 480		owner->root = chunk;
 481	old->dead = 1;
 482	if (!tree->root) {
 483		tree->root = chunk;
 484		list_add(&tree->same_root, &chunk->trees);
 485	}
 486	spin_unlock(&hash_lock);
 487	spin_unlock(&chunk_entry->lock);
 488	spin_unlock(&old_entry->lock);
 489	mutex_unlock(&old_entry->group->mark_mutex);
 490	fsnotify_destroy_mark(old_entry, audit_tree_group);
 491	fsnotify_put_mark(chunk_entry);	/* drop initial reference */
 492	fsnotify_put_mark(old_entry); /* pair to fsnotify_find mark_entry */
 493	return 0;
 494}
 495
 496static void audit_tree_log_remove_rule(struct audit_krule *rule)
 497{
 498	struct audit_buffer *ab;
 499
 500	ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE);
 501	if (unlikely(!ab))
 502		return;
 503	audit_log_format(ab, "op=remove_rule");
 
 504	audit_log_format(ab, " dir=");
 505	audit_log_untrustedstring(ab, rule->tree->pathname);
 506	audit_log_key(ab, rule->filterkey);
 507	audit_log_format(ab, " list=%d res=1", rule->listnr);
 508	audit_log_end(ab);
 509}
 510
 511static void kill_rules(struct audit_tree *tree)
 512{
 513	struct audit_krule *rule, *next;
 514	struct audit_entry *entry;
 515
 516	list_for_each_entry_safe(rule, next, &tree->rules, rlist) {
 517		entry = container_of(rule, struct audit_entry, rule);
 518
 519		list_del_init(&rule->rlist);
 520		if (rule->tree) {
 521			/* not a half-baked one */
 522			audit_tree_log_remove_rule(rule);
 523			if (entry->rule.exe)
 524				audit_remove_mark(entry->rule.exe);
 525			rule->tree = NULL;
 526			list_del_rcu(&entry->list);
 527			list_del(&entry->rule.list);
 528			call_rcu(&entry->rcu, audit_free_rule_rcu);
 529		}
 530	}
 531}
 532
 533/*
 534 * finish killing struct audit_tree
 535 */
 536static void prune_one(struct audit_tree *victim)
 537{
 538	spin_lock(&hash_lock);
 539	while (!list_empty(&victim->chunks)) {
 540		struct node *p;
 541
 542		p = list_entry(victim->chunks.next, struct node, list);
 543
 544		untag_chunk(p);
 545	}
 546	spin_unlock(&hash_lock);
 547	put_tree(victim);
 548}
 549
 550/* trim the uncommitted chunks from tree */
 551
 552static void trim_marked(struct audit_tree *tree)
 553{
 554	struct list_head *p, *q;
 555	spin_lock(&hash_lock);
 556	if (tree->goner) {
 557		spin_unlock(&hash_lock);
 558		return;
 559	}
 560	/* reorder */
 561	for (p = tree->chunks.next; p != &tree->chunks; p = q) {
 562		struct node *node = list_entry(p, struct node, list);
 563		q = p->next;
 564		if (node->index & (1U<<31)) {
 565			list_del_init(p);
 566			list_add(p, &tree->chunks);
 567		}
 568	}
 569
 570	while (!list_empty(&tree->chunks)) {
 571		struct node *node;
 572
 573		node = list_entry(tree->chunks.next, struct node, list);
 574
 575		/* have we run out of marked? */
 576		if (!(node->index & (1U<<31)))
 577			break;
 578
 579		untag_chunk(node);
 580	}
 581	if (!tree->root && !tree->goner) {
 582		tree->goner = 1;
 583		spin_unlock(&hash_lock);
 584		mutex_lock(&audit_filter_mutex);
 585		kill_rules(tree);
 586		list_del_init(&tree->list);
 587		mutex_unlock(&audit_filter_mutex);
 588		prune_one(tree);
 589	} else {
 590		spin_unlock(&hash_lock);
 591	}
 592}
 593
 594static void audit_schedule_prune(void);
 595
 596/* called with audit_filter_mutex */
 597int audit_remove_tree_rule(struct audit_krule *rule)
 598{
 599	struct audit_tree *tree;
 600	tree = rule->tree;
 601	if (tree) {
 602		spin_lock(&hash_lock);
 603		list_del_init(&rule->rlist);
 604		if (list_empty(&tree->rules) && !tree->goner) {
 605			tree->root = NULL;
 606			list_del_init(&tree->same_root);
 607			tree->goner = 1;
 608			list_move(&tree->list, &prune_list);
 609			rule->tree = NULL;
 610			spin_unlock(&hash_lock);
 611			audit_schedule_prune();
 612			return 1;
 613		}
 614		rule->tree = NULL;
 615		spin_unlock(&hash_lock);
 616		return 1;
 617	}
 618	return 0;
 619}
 620
 621static int compare_root(struct vfsmount *mnt, void *arg)
 622{
 623	return inode_to_key(d_backing_inode(mnt->mnt_root)) ==
 624	       (unsigned long)arg;
 625}
 626
 627void audit_trim_trees(void)
 628{
 629	struct list_head cursor;
 630
 631	mutex_lock(&audit_filter_mutex);
 632	list_add(&cursor, &tree_list);
 633	while (cursor.next != &tree_list) {
 634		struct audit_tree *tree;
 635		struct path path;
 636		struct vfsmount *root_mnt;
 637		struct node *node;
 638		int err;
 639
 640		tree = container_of(cursor.next, struct audit_tree, list);
 641		get_tree(tree);
 642		list_del(&cursor);
 643		list_add(&cursor, &tree->list);
 644		mutex_unlock(&audit_filter_mutex);
 645
 646		err = kern_path(tree->pathname, 0, &path);
 647		if (err)
 648			goto skip_it;
 649
 650		root_mnt = collect_mounts(&path);
 651		path_put(&path);
 652		if (IS_ERR(root_mnt))
 653			goto skip_it;
 654
 655		spin_lock(&hash_lock);
 656		list_for_each_entry(node, &tree->chunks, list) {
 657			struct audit_chunk *chunk = find_chunk(node);
 658			/* this could be NULL if the watch is dying else where... */
 
 659			node->index |= 1U<<31;
 660			if (iterate_mounts(compare_root,
 661					   (void *)chunk_to_key(chunk),
 662					   root_mnt))
 663				node->index &= ~(1U<<31);
 664		}
 665		spin_unlock(&hash_lock);
 666		trim_marked(tree);
 667		drop_collected_mounts(root_mnt);
 668skip_it:
 669		put_tree(tree);
 670		mutex_lock(&audit_filter_mutex);
 671	}
 672	list_del(&cursor);
 673	mutex_unlock(&audit_filter_mutex);
 674}
 675
 676int audit_make_tree(struct audit_krule *rule, char *pathname, u32 op)
 677{
 678
 679	if (pathname[0] != '/' ||
 680	    rule->listnr != AUDIT_FILTER_EXIT ||
 681	    op != Audit_equal ||
 682	    rule->inode_f || rule->watch || rule->tree)
 683		return -EINVAL;
 684	rule->tree = alloc_tree(pathname);
 685	if (!rule->tree)
 686		return -ENOMEM;
 687	return 0;
 688}
 689
 690void audit_put_tree(struct audit_tree *tree)
 691{
 692	put_tree(tree);
 693}
 694
 695static int tag_mount(struct vfsmount *mnt, void *arg)
 696{
 697	return tag_chunk(d_backing_inode(mnt->mnt_root), arg);
 698}
 699
 700/*
 701 * That gets run when evict_chunk() ends up needing to kill audit_tree.
 702 * Runs from a separate thread.
 703 */
 704static int prune_tree_thread(void *unused)
 705{
 706	for (;;) {
 707		if (list_empty(&prune_list)) {
 708			set_current_state(TASK_INTERRUPTIBLE);
 709			schedule();
 710		}
 711
 712		audit_ctl_lock();
 713		mutex_lock(&audit_filter_mutex);
 714
 715		while (!list_empty(&prune_list)) {
 716			struct audit_tree *victim;
 717
 718			victim = list_entry(prune_list.next,
 719					struct audit_tree, list);
 720			list_del_init(&victim->list);
 721
 722			mutex_unlock(&audit_filter_mutex);
 723
 724			prune_one(victim);
 725
 726			mutex_lock(&audit_filter_mutex);
 727		}
 728
 729		mutex_unlock(&audit_filter_mutex);
 730		audit_ctl_unlock();
 731	}
 732	return 0;
 733}
 734
 735static int audit_launch_prune(void)
 736{
 737	if (prune_thread)
 738		return 0;
 739	prune_thread = kthread_run(prune_tree_thread, NULL,
 740				"audit_prune_tree");
 741	if (IS_ERR(prune_thread)) {
 742		pr_err("cannot start thread audit_prune_tree");
 743		prune_thread = NULL;
 744		return -ENOMEM;
 
 
 
 745	}
 746	return 0;
 747}
 748
 749/* called with audit_filter_mutex */
 750int audit_add_tree_rule(struct audit_krule *rule)
 751{
 752	struct audit_tree *seed = rule->tree, *tree;
 753	struct path path;
 754	struct vfsmount *mnt;
 755	int err;
 756
 757	rule->tree = NULL;
 758	list_for_each_entry(tree, &tree_list, list) {
 759		if (!strcmp(seed->pathname, tree->pathname)) {
 760			put_tree(seed);
 761			rule->tree = tree;
 762			list_add(&rule->rlist, &tree->rules);
 763			return 0;
 764		}
 765	}
 766	tree = seed;
 767	list_add(&tree->list, &tree_list);
 768	list_add(&rule->rlist, &tree->rules);
 769	/* do not set rule->tree yet */
 770	mutex_unlock(&audit_filter_mutex);
 771
 772	if (unlikely(!prune_thread)) {
 773		err = audit_launch_prune();
 774		if (err)
 775			goto Err;
 776	}
 777
 778	err = kern_path(tree->pathname, 0, &path);
 779	if (err)
 780		goto Err;
 781	mnt = collect_mounts(&path);
 782	path_put(&path);
 783	if (IS_ERR(mnt)) {
 784		err = PTR_ERR(mnt);
 785		goto Err;
 786	}
 787
 788	get_tree(tree);
 789	err = iterate_mounts(tag_mount, tree, mnt);
 790	drop_collected_mounts(mnt);
 791
 792	if (!err) {
 793		struct node *node;
 794		spin_lock(&hash_lock);
 795		list_for_each_entry(node, &tree->chunks, list)
 796			node->index &= ~(1U<<31);
 797		spin_unlock(&hash_lock);
 798	} else {
 799		trim_marked(tree);
 800		goto Err;
 801	}
 802
 803	mutex_lock(&audit_filter_mutex);
 804	if (list_empty(&rule->rlist)) {
 805		put_tree(tree);
 806		return -ENOENT;
 807	}
 808	rule->tree = tree;
 809	put_tree(tree);
 810
 811	return 0;
 812Err:
 813	mutex_lock(&audit_filter_mutex);
 814	list_del_init(&tree->list);
 815	list_del_init(&tree->rules);
 816	put_tree(tree);
 817	return err;
 818}
 819
 820int audit_tag_tree(char *old, char *new)
 821{
 822	struct list_head cursor, barrier;
 823	int failed = 0;
 824	struct path path1, path2;
 825	struct vfsmount *tagged;
 826	int err;
 827
 828	err = kern_path(new, 0, &path2);
 829	if (err)
 830		return err;
 831	tagged = collect_mounts(&path2);
 832	path_put(&path2);
 833	if (IS_ERR(tagged))
 834		return PTR_ERR(tagged);
 835
 836	err = kern_path(old, 0, &path1);
 837	if (err) {
 838		drop_collected_mounts(tagged);
 839		return err;
 840	}
 841
 842	mutex_lock(&audit_filter_mutex);
 843	list_add(&barrier, &tree_list);
 844	list_add(&cursor, &barrier);
 845
 846	while (cursor.next != &tree_list) {
 847		struct audit_tree *tree;
 848		int good_one = 0;
 849
 850		tree = container_of(cursor.next, struct audit_tree, list);
 851		get_tree(tree);
 852		list_del(&cursor);
 853		list_add(&cursor, &tree->list);
 854		mutex_unlock(&audit_filter_mutex);
 855
 856		err = kern_path(tree->pathname, 0, &path2);
 857		if (!err) {
 858			good_one = path_is_under(&path1, &path2);
 859			path_put(&path2);
 860		}
 861
 862		if (!good_one) {
 863			put_tree(tree);
 864			mutex_lock(&audit_filter_mutex);
 865			continue;
 866		}
 867
 868		failed = iterate_mounts(tag_mount, tree, tagged);
 869		if (failed) {
 870			put_tree(tree);
 871			mutex_lock(&audit_filter_mutex);
 872			break;
 873		}
 874
 875		mutex_lock(&audit_filter_mutex);
 876		spin_lock(&hash_lock);
 877		if (!tree->goner) {
 878			list_del(&tree->list);
 879			list_add(&tree->list, &tree_list);
 880		}
 881		spin_unlock(&hash_lock);
 882		put_tree(tree);
 883	}
 884
 885	while (barrier.prev != &tree_list) {
 886		struct audit_tree *tree;
 887
 888		tree = container_of(barrier.prev, struct audit_tree, list);
 889		get_tree(tree);
 890		list_del(&tree->list);
 891		list_add(&tree->list, &barrier);
 892		mutex_unlock(&audit_filter_mutex);
 893
 894		if (!failed) {
 895			struct node *node;
 896			spin_lock(&hash_lock);
 897			list_for_each_entry(node, &tree->chunks, list)
 898				node->index &= ~(1U<<31);
 899			spin_unlock(&hash_lock);
 900		} else {
 901			trim_marked(tree);
 902		}
 903
 904		put_tree(tree);
 905		mutex_lock(&audit_filter_mutex);
 906	}
 907	list_del(&barrier);
 908	list_del(&cursor);
 909	mutex_unlock(&audit_filter_mutex);
 910	path_put(&path1);
 911	drop_collected_mounts(tagged);
 912	return failed;
 913}
 914
 915
 916static void audit_schedule_prune(void)
 917{
 918	wake_up_process(prune_thread);
 919}
 920
 921/*
 922 * ... and that one is done if evict_chunk() decides to delay until the end
 923 * of syscall.  Runs synchronously.
 924 */
 925void audit_kill_trees(struct list_head *list)
 926{
 927	audit_ctl_lock();
 928	mutex_lock(&audit_filter_mutex);
 929
 930	while (!list_empty(list)) {
 931		struct audit_tree *victim;
 932
 933		victim = list_entry(list->next, struct audit_tree, list);
 934		kill_rules(victim);
 935		list_del_init(&victim->list);
 936
 937		mutex_unlock(&audit_filter_mutex);
 938
 939		prune_one(victim);
 940
 941		mutex_lock(&audit_filter_mutex);
 942	}
 943
 944	mutex_unlock(&audit_filter_mutex);
 945	audit_ctl_unlock();
 946}
 947
 948/*
 949 *  Here comes the stuff asynchronous to auditctl operations
 950 */
 951
 952static void evict_chunk(struct audit_chunk *chunk)
 953{
 954	struct audit_tree *owner;
 955	struct list_head *postponed = audit_killed_trees();
 956	int need_prune = 0;
 957	int n;
 958
 959	if (chunk->dead)
 960		return;
 961
 962	chunk->dead = 1;
 963	mutex_lock(&audit_filter_mutex);
 964	spin_lock(&hash_lock);
 965	while (!list_empty(&chunk->trees)) {
 966		owner = list_entry(chunk->trees.next,
 967				   struct audit_tree, same_root);
 968		owner->goner = 1;
 969		owner->root = NULL;
 970		list_del_init(&owner->same_root);
 971		spin_unlock(&hash_lock);
 972		if (!postponed) {
 973			kill_rules(owner);
 974			list_move(&owner->list, &prune_list);
 975			need_prune = 1;
 976		} else {
 977			list_move(&owner->list, postponed);
 978		}
 979		spin_lock(&hash_lock);
 980	}
 981	list_del_rcu(&chunk->hash);
 982	for (n = 0; n < chunk->count; n++)
 983		list_del_init(&chunk->owners[n].list);
 984	spin_unlock(&hash_lock);
 985	mutex_unlock(&audit_filter_mutex);
 986	if (need_prune)
 987		audit_schedule_prune();
 988}
 989
 990static int audit_tree_handle_event(struct fsnotify_group *group,
 991				   struct inode *to_tell,
 992				   struct fsnotify_mark *inode_mark,
 993				   struct fsnotify_mark *vfsmount_mark,
 994				   u32 mask, const void *data, int data_type,
 995				   const unsigned char *file_name, u32 cookie,
 996				   struct fsnotify_iter_info *iter_info)
 997{
 998	return 0;
 999}
1000
1001static void audit_tree_freeing_mark(struct fsnotify_mark *entry, struct fsnotify_group *group)
1002{
1003	struct audit_chunk *chunk = container_of(entry, struct audit_chunk, mark);
1004
1005	evict_chunk(chunk);
1006
1007	/*
1008	 * We are guaranteed to have at least one reference to the mark from
1009	 * either the inode or the caller of fsnotify_destroy_mark().
1010	 */
1011	BUG_ON(refcount_read(&entry->refcnt) < 1);
1012}
1013
1014static const struct fsnotify_ops audit_tree_ops = {
1015	.handle_event = audit_tree_handle_event,
1016	.freeing_mark = audit_tree_freeing_mark,
1017	.free_mark = audit_tree_destroy_watch,
1018};
1019
1020static int __init audit_tree_init(void)
1021{
1022	int i;
1023
1024	audit_tree_group = fsnotify_alloc_group(&audit_tree_ops);
1025	if (IS_ERR(audit_tree_group))
1026		audit_panic("cannot initialize fsnotify group for rectree watches");
1027
1028	for (i = 0; i < HASH_SIZE; i++)
1029		INIT_LIST_HEAD(&chunk_hash_heads[i]);
1030
1031	return 0;
1032}
1033__initcall(audit_tree_init);