kernel/audit_tree.c

   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
   7 struct audit_tree;
   8 struct audit_chunk;
   9
  10 struct audit_tree {
  11         atomic_t count;
  12         int goner;
  13         struct audit_chunk *root;
  14         struct list_head chunks;
  15         struct list_head rules;
  16         struct list_head list;
  17         struct list_head same_root;
  18         struct rcu_head head;
  19         char pathname[];
  20 };
  21
  22 struct audit_chunk {
  23         struct list_head hash;
  24         struct fsnotify_mark_entry mark;
  25         struct list_head trees;         /* with root here */
  26         int dead;
  27         int count;
  28         atomic_long_t refs;
  29         struct rcu_head head;
  30         struct node {
  31                 struct list_head list;
  32                 struct audit_tree *owner;
  33                 unsigned index;         /* index; upper bit indicates 'will prune' */
  34         } owners[];
  35 };
  36
  37 static LIST_HEAD(tree_list);
  38 static LIST_HEAD(prune_list);
  39
  40 /*
  41  * One struct chunk is attached to each inode of interest.
  42  * We replace struct chunk on tagging/untagging.
  43  * Rules have pointer to struct audit_tree.
  44  * Rules have struct list_head rlist forming a list of rules over
  45  * the same tree.
  46  * References to struct chunk are collected at audit_inode{,_child}()
  47  * time and used in AUDIT_TREE rule matching.
  48  * These references are dropped at the same time we are calling
  49  * audit_free_names(), etc.
  50  *
  51  * Cyclic lists galore:
  52  * tree.chunks anchors chunk.owners[].list                      hash_lock
  53  * tree.rules anchors rule.rlist                                audit_filter_mutex
  54  * chunk.trees anchors tree.same_root                           hash_lock
  55  * chunk.hash is a hash with middle bits of watch.inode as
  56  * a hash function.                                             RCU, hash_lock
  57  *
  58  * tree is refcounted; one reference for "some rules on rules_list refer to
  59  * it", one for each chunk with pointer to it.
  60  *
  61  * chunk is refcounted by embedded inotify_watch + .refs (non-zero refcount
  62  * of watch contributes 1 to .refs).
  63  *
  64  * node.index allows to get from node.list to containing chunk.
  65  * MSB of that sucker is stolen to mark taggings that we might have to
  66  * revert - several operations have very unpleasant cleanup logics and
  67  * that makes a difference.  Some.
  68  */
  69
  70 static struct fsnotify_group *audit_tree_group;
  71
  72 static struct audit_tree *alloc_tree(const char *s)
  73 {
  74         struct audit_tree *tree;
  75
  76         tree = kmalloc(sizeof(struct audit_tree) + strlen(s) + 1, GFP_KERNEL);
  77         if (tree) {
  78                 atomic_set(&tree->count, 1);
  79                 tree->goner = 0;
  80                 INIT_LIST_HEAD(&tree->chunks);
  81                 INIT_LIST_HEAD(&tree->rules);
  82                 INIT_LIST_HEAD(&tree->list);
  83                 INIT_LIST_HEAD(&tree->same_root);
  84                 tree->root = NULL;
  85                 strcpy(tree->pathname, s);
  86         }
  87         return tree;
  88 }
  89
  90 static inline void get_tree(struct audit_tree *tree)
  91 {
  92         atomic_inc(&tree->count);
  93 }
  94
  95 static void __put_tree(struct rcu_head *rcu)
  96 {
  97         struct audit_tree *tree = container_of(rcu, struct audit_tree, head);
  98         kfree(tree);
  99 }
 100
 101 static inline void put_tree(struct audit_tree *tree)
 102 {
 103         if (atomic_dec_and_test(&tree->count))
 104                 call_rcu(&tree->head, __put_tree);
 105 }
 106
 107 /* to avoid bringing the entire thing in audit.h */
 108 const char *audit_tree_path(struct audit_tree *tree)
 109 {
 110         return tree->pathname;
 111 }
 112
 113 static void free_chunk(struct audit_chunk *chunk)
 114 {
 115         int i;
 116
 117         for (i = 0; i < chunk->count; i++) {
 118                 if (chunk->owners[i].owner)
 119                         put_tree(chunk->owners[i].owner);
 120         }
 121         kfree(chunk);
 122 }
 123
 124 void audit_put_chunk(struct audit_chunk *chunk)
 125 {
 126         if (atomic_long_dec_and_test(&chunk->refs))
 127                 free_chunk(chunk);
 128 }
 129
 130 static void __put_chunk(struct rcu_head *rcu)
 131 {
 132         struct audit_chunk *chunk = container_of(rcu, struct audit_chunk, head);
 133         audit_put_chunk(chunk);
 134 }
 135
 136 static void audit_tree_destroy_watch(struct fsnotify_mark_entry *entry)
 137 {
 138         struct audit_chunk *chunk = container_of(entry, struct audit_chunk, mark);
 139         call_rcu(&chunk->head, __put_chunk);
 140 }
 141
 142 static struct audit_chunk *alloc_chunk(int count)
 143 {
 144         struct audit_chunk *chunk;
 145         size_t size;
 146         int i;
 147
 148         size = offsetof(struct audit_chunk, owners) + count * sizeof(struct node);
 149         chunk = kzalloc(size, GFP_KERNEL);
 150         if (!chunk)
 151                 return NULL;
 152
 153         INIT_LIST_HEAD(&chunk->hash);
 154         INIT_LIST_HEAD(&chunk->trees);
 155         chunk->count = count;
 156         atomic_long_set(&chunk->refs, 1);
 157         for (i = 0; i < count; i++) {
 158                 INIT_LIST_HEAD(&chunk->owners[i].list);
 159                 chunk->owners[i].index = i;
 160         }
 161         fsnotify_init_mark(&chunk->mark, audit_tree_destroy_watch);
 162         return chunk;
 163 }
 164
 165 enum {HASH_SIZE = 128};
 166 static struct list_head chunk_hash_heads[HASH_SIZE];
 167 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(hash_lock);
 168
 169 static inline struct list_head *chunk_hash(const struct inode *inode)
 170 {
 171         unsigned long n = (unsigned long)inode / L1_CACHE_BYTES;
 172         return chunk_hash_heads + n % HASH_SIZE;
 173 }
 174
 175 /* hash_lock & entry->lock is held by caller */
 176 static void insert_hash(struct audit_chunk *chunk)
 177 {
 178         struct fsnotify_mark_entry *entry = &chunk->mark;
 179         struct list_head *list;
 180
 181         if (!entry->inode)
 182                 return;
 183         list = chunk_hash(entry->inode);
 184         list_add_rcu(&chunk->hash, list);
 185 }
 186
 187 /* called under rcu_read_lock */
 188 struct audit_chunk *audit_tree_lookup(const struct inode *inode)
 189 {
 190         struct list_head *list = chunk_hash(inode);
 191         struct audit_chunk *p;
 192
 193         list_for_each_entry_rcu(p, list, hash) {
 194                 /* mark.inode may have gone NULL, but who cares? */
 195                 if (p->mark.inode == inode) {
 196                         atomic_long_inc(&p->refs);
 197                         return p;
 198                 }
 199         }
 200         return NULL;
 201 }
 202
 203 int audit_tree_match(struct audit_chunk *chunk, struct audit_tree *tree)
 204 {
 205         int n;
 206         for (n = 0; n < chunk->count; n++)
 207                 if (chunk->owners[n].owner == tree)
 208                         return 1;
 209         return 0;
 210 }
 211
 212 /* tagging and untagging inodes with trees */
 213
 214 static struct audit_chunk *find_chunk(struct node *p)
 215 {
 216         int index = p->index & ~(1U<<31);
 217         p -= index;
 218         return container_of(p, struct audit_chunk, owners[0]);
 219 }
 220
 221 static void untag_chunk(struct node *p)
 222 {
 223         struct audit_chunk *chunk = find_chunk(p);
 224         struct fsnotify_mark_entry *entry = &chunk->mark;
 225         struct audit_chunk *new;
 226         struct audit_tree *owner;
 227         int size = chunk->count - 1;
 228         int i, j;
 229
 230         fsnotify_get_mark(entry);
 231
 232         spin_unlock(&hash_lock);
 233
 234         spin_lock(&entry->lock);
 235         if (chunk->dead || !entry->inode) {
 236                 spin_unlock(&entry->lock);
 237                 goto out;
 238         }
 239
 240         owner = p->owner;
 241
 242         if (!size) {
 243                 chunk->dead = 1;
 244                 spin_lock(&hash_lock);
 245                 list_del_init(&chunk->trees);
 246                 if (owner->root == chunk)
 247                         owner->root = NULL;
 248                 list_del_init(&p->list);
 249                 list_del_rcu(&chunk->hash);
 250                 spin_unlock(&hash_lock);
 251                 spin_unlock(&entry->lock);
 252                 fsnotify_destroy_mark_by_entry(entry);
 253                 fsnotify_put_mark(entry);
 254                 goto out;
 255         }
 256
 257         new = alloc_chunk(size);
 258         if (!new)
 259                 goto Fallback;
 260         fsnotify_duplicate_mark(&new->mark, entry);
 261         if (fsnotify_add_mark(&new->mark, new->mark.group, new->mark.inode, 1)) {
 262                 free_chunk(new);
 263                 goto Fallback;
 264         }
 265
 266         chunk->dead = 1;
 267         spin_lock(&hash_lock);
 268         list_replace_init(&chunk->trees, &new->trees);
 269         if (owner->root == chunk) {
 270                 list_del_init(&owner->same_root);
 271                 owner->root = NULL;
 272         }
 273
 274         for (i = j = 0; i < size; i++, j++) {
 275                 struct audit_tree *s;
 276                 if (&chunk->owners[j] == p) {
 277                         list_del_init(&p->list);
 278                         i--;
 279                         continue;
 280                 }
 281                 s = chunk->owners[j].owner;
 282                 new->owners[i].owner = s;
 283                 new->owners[i].index = chunk->owners[j].index - j + i;
 284                 if (!s) /* result of earlier fallback */
 285                         continue;
 286                 get_tree(s);
 287                 list_replace_init(&chunk->owners[i].list, &new->owners[j].list);
 288         }
 289
 290         list_replace_rcu(&chunk->hash, &new->hash);
 291         list_for_each_entry(owner, &new->trees, same_root)
 292                 owner->root = new;
 293         spin_unlock(&hash_lock);
 294         spin_unlock(&entry->lock);
 295         fsnotify_destroy_mark_by_entry(entry);
 296         fsnotify_put_mark(entry);
 297         goto out;
 298
 299 Fallback:
 300         // do the best we can
 301         spin_lock(&hash_lock);
 302         if (owner->root == chunk) {
 303                 list_del_init(&owner->same_root);
 304                 owner->root = NULL;
 305         }
 306         list_del_init(&p->list);
 307         p->owner = NULL;
 308         put_tree(owner);
 309         spin_unlock(&hash_lock);
 310         spin_unlock(&entry->lock);
 311 out:
 312         fsnotify_put_mark(entry);
 313         spin_lock(&hash_lock);
 314 }
 315
 316 static int create_chunk(struct inode *inode, struct audit_tree *tree)
 317 {
 318         struct fsnotify_mark_entry *entry;
 319         struct audit_chunk *chunk = alloc_chunk(1);
 320         if (!chunk)
 321                 return -ENOMEM;
 322
 323         entry = &chunk->mark;
 324         if (fsnotify_add_mark(entry, audit_tree_group, inode, 0)) {
 325                 free_chunk(chunk);
 326                 return -ENOSPC;
 327         }
 328
 329         spin_lock(&entry->lock);
 330         spin_lock(&hash_lock);
 331         if (tree->goner) {
 332                 spin_unlock(&hash_lock);
 333                 chunk->dead = 1;
 334                 spin_unlock(&entry->lock);
 335                 fsnotify_destroy_mark_by_entry(entry);
 336                 fsnotify_put_mark(entry);
 337                 return 0;
 338         }
 339         chunk->owners[0].index = (1U << 31);
 340         chunk->owners[0].owner = tree;
 341         get_tree(tree);
 342         list_add(&chunk->owners[0].list, &tree->chunks);
 343         if (!tree->root) {
 344                 tree->root = chunk;
 345                 list_add(&tree->same_root, &chunk->trees);
 346         }
 347         insert_hash(chunk);
 348         spin_unlock(&hash_lock);
 349         spin_unlock(&entry->lock);
 350         return 0;
 351 }
 352
 353 /* the first tagged inode becomes root of tree */
 354 static int tag_chunk(struct inode *inode, struct audit_tree *tree)
 355 {
 356         struct fsnotify_mark_entry *old_entry, *chunk_entry;
 357         struct audit_tree *owner;
 358         struct audit_chunk *chunk, *old;
 359         struct node *p;
 360         int n;
 361
 362         spin_lock(&inode->i_lock);
 363         old_entry = fsnotify_find_mark_entry(audit_tree_group, inode);
 364         spin_unlock(&inode->i_lock);
 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                 return -ENOMEM;
 384         chunk_entry = &chunk->mark;
 385
 386         spin_lock(&old_entry->lock);
 387         if (!old_entry->inode) {
 388                 /* old_entry is being shot, lets just lie */
 389                 spin_unlock(&old_entry->lock);
 390                 fsnotify_put_mark(old_entry);
 391                 free_chunk(chunk);
 392                 return -ENOENT;
 393         }
 394
 395         fsnotify_duplicate_mark(chunk_entry, old_entry);
 396         if (fsnotify_add_mark(chunk_entry, chunk_entry->group, chunk_entry->inode, 1)) {
 397                 spin_unlock(&old_entry->lock);
 398                 free_chunk(chunk);
 399                 fsnotify_put_mark(old_entry);
 400                 return -ENOSPC;
 401         }
 402
 403         /* even though we hold old_entry->lock, this is safe since chunk_entry->lock could NEVER have been grabbed before */
 404         spin_lock(&chunk_entry->lock);
 405         spin_lock(&hash_lock);
 406
 407         /* we now hold old_entry->lock, chunk_entry->lock, and hash_lock */
 408         if (tree->goner) {
 409                 spin_unlock(&hash_lock);
 410                 chunk->dead = 1;
 411                 spin_unlock(&chunk_entry->lock);
 412                 spin_unlock(&old_entry->lock);
 413
 414                 fsnotify_destroy_mark_by_entry(chunk_entry);
 415
 416                 fsnotify_put_mark(chunk_entry);
 417                 fsnotify_put_mark(old_entry);
 418                 return 0;
 419         }
 420         list_replace_init(&old->trees, &chunk->trees);
 421         for (n = 0, p = chunk->owners; n < old->count; n++, p++) {
 422                 struct audit_tree *s = old->owners[n].owner;
 423                 p->owner = s;
 424                 p->index = old->owners[n].index;
 425                 if (!s) /* result of fallback in untag */
 426                         continue;
 427                 get_tree(s);
 428                 list_replace_init(&old->owners[n].list, &p->list);
 429         }
 430         p->index = (chunk->count - 1) | (1U<<31);
 431         p->owner = tree;
 432         get_tree(tree);
 433         list_add(&p->list, &tree->chunks);
 434         list_replace_rcu(&old->hash, &chunk->hash);
 435         list_for_each_entry(owner, &chunk->trees, same_root)
 436                 owner->root = chunk;
 437         old->dead = 1;
 438         if (!tree->root) {
 439                 tree->root = chunk;
 440                 list_add(&tree->same_root, &chunk->trees);
 441         }
 442         spin_unlock(&hash_lock);
 443         spin_unlock(&chunk_entry->lock);
 444         spin_unlock(&old_entry->lock);
 445         fsnotify_destroy_mark_by_entry(old_entry);
 446         fsnotify_put_mark(old_entry);
 447         return 0;
 448 }
 449
 450 static void kill_rules(struct audit_tree *tree)
 451 {
 452         struct audit_krule *rule, *next;
 453         struct audit_entry *entry;
 454         struct audit_buffer *ab;
 455
 456         list_for_each_entry_safe(rule, next, &tree->rules, rlist) {
 457                 entry = container_of(rule, struct audit_entry, rule);
 458
 459                 list_del_init(&rule->rlist);
 460                 if (rule->tree) {
 461                         /* not a half-baked one */
 462                         ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE);
 463                         audit_log_format(ab, "op=");
 464                         audit_log_string(ab, "remove rule");
 465                         audit_log_format(ab, " dir=");
 466                         audit_log_untrustedstring(ab, rule->tree->pathname);
 467                         audit_log_key(ab, rule->filterkey);
 468                         audit_log_format(ab, " list=%d res=1", rule->listnr);
 469                         audit_log_end(ab);
 470                         rule->tree = NULL;
 471                         list_del_rcu(&entry->list);
 472                         list_del(&entry->rule.list);
 473                         call_rcu(&entry->rcu, audit_free_rule_rcu);
 474                 }
 475         }
 476 }
 477
 478 /*
 479  * finish killing struct audit_tree
 480  */
 481 static void prune_one(struct audit_tree *victim)
 482 {
 483         spin_lock(&hash_lock);
 484         while (!list_empty(&victim->chunks)) {
 485                 struct node *p;
 486
 487                 p = list_entry(victim->chunks.next, struct node, list);
 488
 489                 untag_chunk(p);
 490         }
 491         spin_unlock(&hash_lock);
 492         put_tree(victim);
 493 }
 494
 495 /* trim the uncommitted chunks from tree */
 496
 497 static void trim_marked(struct audit_tree *tree)
 498 {
 499         struct list_head *p, *q;
 500         spin_lock(&hash_lock);
 501         if (tree->goner) {
 502                 spin_unlock(&hash_lock);
 503                 return;
 504         }
 505         /* reorder */
 506         for (p = tree->chunks.next; p != &tree->chunks; p = q) {
 507                 struct node *node = list_entry(p, struct node, list);
 508                 q = p->next;
 509                 if (node->index & (1U<<31)) {
 510                         list_del_init(p);
 511                         list_add(p, &tree->chunks);
 512                 }
 513         }
 514
 515         while (!list_empty(&tree->chunks)) {
 516                 struct node *node;
 517
 518                 node = list_entry(tree->chunks.next, struct node, list);
 519
 520                 /* have we run out of marked? */
 521                 if (!(node->index & (1U<<31)))
 522                         break;
 523
 524                 untag_chunk(node);
 525         }
 526         if (!tree->root && !tree->goner) {
 527                 tree->goner = 1;
 528                 spin_unlock(&hash_lock);
 529                 mutex_lock(&audit_filter_mutex);
 530                 kill_rules(tree);
 531                 list_del_init(&tree->list);
 532                 mutex_unlock(&audit_filter_mutex);
 533                 prune_one(tree);
 534         } else {
 535                 spin_unlock(&hash_lock);
 536         }
 537 }
 538
 539 static void audit_schedule_prune(void);
 540
 541 /* called with audit_filter_mutex */
 542 int audit_remove_tree_rule(struct audit_krule *rule)
 543 {
 544         struct audit_tree *tree;
 545         tree = rule->tree;
 546         if (tree) {
 547                 spin_lock(&hash_lock);
 548                 list_del_init(&rule->rlist);
 549                 if (list_empty(&tree->rules) && !tree->goner) {
 550                         tree->root = NULL;
 551                         list_del_init(&tree->same_root);
 552                         tree->goner = 1;
 553                         list_move(&tree->list, &prune_list);
 554                         rule->tree = NULL;
 555                         spin_unlock(&hash_lock);
 556                         audit_schedule_prune();
 557                         return 1;
 558                 }
 559                 rule->tree = NULL;
 560                 spin_unlock(&hash_lock);
 561                 return 1;
 562         }
 563         return 0;
 564 }
 565
 566 void audit_trim_trees(void)
 567 {
 568         struct list_head cursor;
 569
 570         mutex_lock(&audit_filter_mutex);
 571         list_add(&cursor, &tree_list);
 572         while (cursor.next != &tree_list) {
 573                 struct audit_tree *tree;
 574                 struct path path;
 575                 struct vfsmount *root_mnt;
 576                 struct node *node;
 577                 struct list_head list;
 578                 int err;
 579
 580                 tree = container_of(cursor.next, struct audit_tree, list);
 581                 get_tree(tree);
 582                 list_del(&cursor);
 583                 list_add(&cursor, &tree->list);
 584                 mutex_unlock(&audit_filter_mutex);
 585
 586                 err = kern_path(tree->pathname, 0, &path);
 587                 if (err)
 588                         goto skip_it;
 589
 590                 root_mnt = collect_mounts(&path);
 591                 path_put(&path);
 592                 if (!root_mnt)
 593                         goto skip_it;
 594
 595                 list_add_tail(&list, &root_mnt->mnt_list);
 596                 spin_lock(&hash_lock);
 597                 list_for_each_entry(node, &tree->chunks, list) {
 598                         struct audit_chunk *chunk = find_chunk(node);
 599                         /* this could be NULL if the watch is dieing else where... */
 600                         struct inode *inode = chunk->mark.inode;
 601                         struct vfsmount *mnt;
 602                         node->index |= 1U<<31;
 603                         list_for_each_entry(mnt, &list, mnt_list) {
 604                                 if (mnt->mnt_root->d_inode == inode) {
 605                                         node->index &= ~(1U<<31);
 606                                         break;
 607                                 }
 608                         }
 609                 }
 610                 spin_unlock(&hash_lock);
 611                 trim_marked(tree);
 612                 put_tree(tree);
 613                 list_del_init(&list);
 614                 drop_collected_mounts(root_mnt);
 615 skip_it:
 616                 mutex_lock(&audit_filter_mutex);
 617         }
 618         list_del(&cursor);
 619         mutex_unlock(&audit_filter_mutex);
 620 }
 621
 622 static int is_under(struct vfsmount *mnt, struct dentry *dentry,
 623                     struct path *path)
 624 {
 625         if (mnt != path->mnt) {
 626                 for (;;) {
 627                         if (mnt->mnt_parent == mnt)
 628                                 return 0;
 629                         if (mnt->mnt_parent == path->mnt)
 630                                         break;
 631                         mnt = mnt->mnt_parent;
 632                 }
 633                 dentry = mnt->mnt_mountpoint;
 634         }
 635         return is_subdir(dentry, path->dentry);
 636 }
 637
 638 int audit_make_tree(struct audit_krule *rule, char *pathname, u32 op)
 639 {
 640
 641         if (pathname[0] != '/' ||
 642             rule->listnr != AUDIT_FILTER_EXIT ||
 643             op != Audit_equal ||
 644             rule->inode_f || rule->watch || rule->tree)
 645                 return -EINVAL;
 646         rule->tree = alloc_tree(pathname);
 647         if (!rule->tree)
 648                 return -ENOMEM;
 649         return 0;
 650 }
 651
 652 void audit_put_tree(struct audit_tree *tree)
 653 {
 654         put_tree(tree);
 655 }
 656
 657 /* called with audit_filter_mutex */
 658 int audit_add_tree_rule(struct audit_krule *rule)
 659 {
 660         struct audit_tree *seed = rule->tree, *tree;
 661         struct path path;
 662         struct vfsmount *mnt, *p;
 663         struct list_head list;
 664         int err;
 665
 666         list_for_each_entry(tree, &tree_list, list) {
 667                 if (!strcmp(seed->pathname, tree->pathname)) {
 668                         put_tree(seed);
 669                         rule->tree = tree;
 670                         list_add(&rule->rlist, &tree->rules);
 671                         return 0;
 672                 }
 673         }
 674         tree = seed;
 675         list_add(&tree->list, &tree_list);
 676         list_add(&rule->rlist, &tree->rules);
 677         /* do not set rule->tree yet */
 678         mutex_unlock(&audit_filter_mutex);
 679
 680         err = kern_path(tree->pathname, 0, &path);
 681         if (err)
 682                 goto Err;
 683         mnt = collect_mounts(&path);
 684         path_put(&path);
 685         if (!mnt) {
 686                 err = -ENOMEM;
 687                 goto Err;
 688         }
 689         list_add_tail(&list, &mnt->mnt_list);
 690
 691         get_tree(tree);
 692         list_for_each_entry(p, &list, mnt_list) {
 693                 err = tag_chunk(p->mnt_root->d_inode, tree);
 694                 if (err)
 695                         break;
 696         }
 697
 698         list_del(&list);
 699         drop_collected_mounts(mnt);
 700
 701         if (!err) {
 702                 struct node *node;
 703                 spin_lock(&hash_lock);
 704                 list_for_each_entry(node, &tree->chunks, list)
 705                         node->index &= ~(1U<<31);
 706                 spin_unlock(&hash_lock);
 707         } else {
 708                 trim_marked(tree);
 709                 goto Err;
 710         }
 711
 712         mutex_lock(&audit_filter_mutex);
 713         if (list_empty(&rule->rlist)) {
 714                 put_tree(tree);
 715                 return -ENOENT;
 716         }
 717         rule->tree = tree;
 718         put_tree(tree);
 719
 720         return 0;
 721 Err:
 722         mutex_lock(&audit_filter_mutex);
 723         list_del_init(&tree->list);
 724         list_del_init(&tree->rules);
 725         put_tree(tree);
 726         return err;
 727 }
 728
 729 int audit_tag_tree(char *old, char *new)
 730 {
 731         struct list_head cursor, barrier;
 732         int failed = 0;
 733         struct path path;
 734         struct vfsmount *tagged;
 735         struct list_head list;
 736         struct vfsmount *mnt;
 737         struct dentry *dentry;
 738         int err;
 739
 740         err = kern_path(new, 0, &path);
 741         if (err)
 742                 return err;
 743         tagged = collect_mounts(&path);
 744         path_put(&path);
 745         if (!tagged)
 746                 return -ENOMEM;
 747
 748         err = kern_path(old, 0, &path);
 749         if (err) {
 750                 drop_collected_mounts(tagged);
 751                 return err;
 752         }
 753         mnt = mntget(path.mnt);
 754         dentry = dget(path.dentry);
 755         path_put(&path);
 756
 757         list_add_tail(&list, &tagged->mnt_list);
 758
 759         mutex_lock(&audit_filter_mutex);
 760         list_add(&barrier, &tree_list);
 761         list_add(&cursor, &barrier);
 762
 763         while (cursor.next != &tree_list) {
 764                 struct audit_tree *tree;
 765                 struct vfsmount *p;
 766
 767                 tree = container_of(cursor.next, struct audit_tree, list);
 768                 get_tree(tree);
 769                 list_del(&cursor);
 770                 list_add(&cursor, &tree->list);
 771                 mutex_unlock(&audit_filter_mutex);
 772
 773                 err = kern_path(tree->pathname, 0, &path);
 774                 if (err) {
 775                         put_tree(tree);
 776                         mutex_lock(&audit_filter_mutex);
 777                         continue;
 778                 }
 779
 780                 spin_lock(&vfsmount_lock);
 781                 if (!is_under(mnt, dentry, &path)) {
 782                         spin_unlock(&vfsmount_lock);
 783                         path_put(&path);
 784                         put_tree(tree);
 785                         mutex_lock(&audit_filter_mutex);
 786                         continue;
 787                 }
 788                 spin_unlock(&vfsmount_lock);
 789                 path_put(&path);
 790
 791                 list_for_each_entry(p, &list, mnt_list) {
 792                         failed = tag_chunk(p->mnt_root->d_inode, tree);
 793                         if (failed)
 794                                 break;
 795                 }
 796
 797                 if (failed) {
 798                         put_tree(tree);
 799                         mutex_lock(&audit_filter_mutex);
 800                         break;
 801                 }
 802
 803                 mutex_lock(&audit_filter_mutex);
 804                 spin_lock(&hash_lock);
 805                 if (!tree->goner) {
 806                         list_del(&tree->list);
 807                         list_add(&tree->list, &tree_list);
 808                 }
 809                 spin_unlock(&hash_lock);
 810                 put_tree(tree);
 811         }
 812
 813         while (barrier.prev != &tree_list) {
 814                 struct audit_tree *tree;
 815
 816                 tree = container_of(barrier.prev, struct audit_tree, list);
 817                 get_tree(tree);
 818                 list_del(&tree->list);
 819                 list_add(&tree->list, &barrier);
 820                 mutex_unlock(&audit_filter_mutex);
 821
 822                 if (!failed) {
 823                         struct node *node;
 824                         spin_lock(&hash_lock);
 825                         list_for_each_entry(node, &tree->chunks, list)
 826                                 node->index &= ~(1U<<31);
 827                         spin_unlock(&hash_lock);
 828                 } else {
 829                         trim_marked(tree);
 830                 }
 831
 832                 put_tree(tree);
 833                 mutex_lock(&audit_filter_mutex);
 834         }
 835         list_del(&barrier);
 836         list_del(&cursor);
 837         list_del(&list);
 838         mutex_unlock(&audit_filter_mutex);
 839         dput(dentry);
 840         mntput(mnt);
 841         drop_collected_mounts(tagged);
 842         return failed;
 843 }
 844
 845 /*
 846  * That gets run when evict_chunk() ends up needing to kill audit_tree.
 847  * Runs from a separate thread.
 848  */
 849 static int prune_tree_thread(void *unused)
 850 {
 851         mutex_lock(&audit_cmd_mutex);
 852         mutex_lock(&audit_filter_mutex);
 853
 854         while (!list_empty(&prune_list)) {
 855                 struct audit_tree *victim;
 856
 857                 victim = list_entry(prune_list.next, struct audit_tree, list);
 858                 list_del_init(&victim->list);
 859
 860                 mutex_unlock(&audit_filter_mutex);
 861
 862                 prune_one(victim);
 863
 864                 mutex_lock(&audit_filter_mutex);
 865         }
 866
 867         mutex_unlock(&audit_filter_mutex);
 868         mutex_unlock(&audit_cmd_mutex);
 869         return 0;
 870 }
 871
 872 static void audit_schedule_prune(void)
 873 {
 874         kthread_run(prune_tree_thread, NULL, "audit_prune_tree");
 875 }
 876
 877 /*
 878  * ... and that one is done if evict_chunk() decides to delay until the end
 879  * of syscall.  Runs synchronously.
 880  */
 881 void audit_kill_trees(struct list_head *list)
 882 {
 883         mutex_lock(&audit_cmd_mutex);
 884         mutex_lock(&audit_filter_mutex);
 885
 886         while (!list_empty(list)) {
 887                 struct audit_tree *victim;
 888
 889                 victim = list_entry(list->next, struct audit_tree, list);
 890                 kill_rules(victim);
 891                 list_del_init(&victim->list);
 892
 893                 mutex_unlock(&audit_filter_mutex);
 894
 895                 prune_one(victim);
 896
 897                 mutex_lock(&audit_filter_mutex);
 898         }
 899
 900         mutex_unlock(&audit_filter_mutex);
 901         mutex_unlock(&audit_cmd_mutex);
 902 }
 903
 904 /*
 905  *  Here comes the stuff asynchronous to auditctl operations
 906  */
 907
 908 /* inode->inotify_mutex is locked */
 909 static void evict_chunk(struct audit_chunk *chunk)
 910 {
 911         struct audit_tree *owner;
 912         struct list_head *postponed = audit_killed_trees();
 913         int need_prune = 0;
 914         int n;
 915
 916         if (chunk->dead)
 917                 return;
 918
 919         chunk->dead = 1;
 920         mutex_lock(&audit_filter_mutex);
 921         spin_lock(&hash_lock);
 922         while (!list_empty(&chunk->trees)) {
 923                 owner = list_entry(chunk->trees.next,
 924                                    struct audit_tree, same_root);
 925                 owner->goner = 1;
 926                 owner->root = NULL;
 927                 list_del_init(&owner->same_root);
 928                 spin_unlock(&hash_lock);
 929                 if (!postponed) {
 930                         kill_rules(owner);
 931                         list_move(&owner->list, &prune_list);
 932                         need_prune = 1;
 933                 } else {
 934                         list_move(&owner->list, postponed);
 935                 }
 936                 spin_lock(&hash_lock);
 937         }
 938         list_del_rcu(&chunk->hash);
 939         for (n = 0; n < chunk->count; n++)
 940                 list_del_init(&chunk->owners[n].list);
 941         spin_unlock(&hash_lock);
 942         if (need_prune)
 943                 audit_schedule_prune();
 944         mutex_unlock(&audit_filter_mutex);
 945 }
 946
 947 static int audit_tree_handle_event(struct fsnotify_group *group, struct fsnotify_event *event)
 948 {
 949         BUG();
 950         return -EOPNOTSUPP;
 951 }
 952
 953 static void audit_tree_freeing_mark(struct fsnotify_mark_entry *entry, struct fsnotify_group *group)
 954 {
 955         struct audit_chunk *chunk = container_of(entry, struct audit_chunk, mark);
 956
 957         evict_chunk(chunk);
 958         fsnotify_put_mark(entry);
 959 }
 960
 961 static bool audit_tree_send_event(struct fsnotify_group *group, struct inode *inode,
 962                                   __u32 mask, void *data, int data_type)
 963 {
 964         return 0;
 965 }
 966
 967 static const struct fsnotify_ops audit_tree_ops = {
 968         .handle_event = audit_tree_handle_event,
 969         .should_send_event = audit_tree_send_event,
 970         .free_group_priv = NULL,
 971         .free_event_priv = NULL,
 972         .freeing_mark = audit_tree_freeing_mark,
 973 };
 974
 975 static int __init audit_tree_init(void)
 976 {
 977         int i;
 978
 979         audit_tree_group = fsnotify_obtain_group(0, &audit_tree_ops);
 980         if (IS_ERR(audit_tree_group))
 981                 audit_panic("cannot initialize inotify handle for rectree watches");
 982
 983         for (i = 0; i < HASH_SIZE; i++)
 984                 INIT_LIST_HEAD(&chunk_hash_heads[i]);
 985
 986         return 0;
 987 }
 988 __initcall(audit_tree_init);