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 #include <linux/slab.h>
   7
   8 struct audit_tree;
   9 struct audit_chunk;
  10
  11 struct 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
  23 struct 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
  38 static LIST_HEAD(tree_list);
  39 static LIST_HEAD(prune_list);
  40
  41 /*
  42  * One struct chunk is attached to each inode of interest.
  43  * We replace struct chunk on tagging/untagging.
  44  * Rules have pointer to struct audit_tree.
  45  * Rules have struct list_head rlist forming a list of rules over
  46  * the same tree.
  47  * References to struct chunk are collected at audit_inode{,_child}()
  48  * time and used in AUDIT_TREE rule matching.
  49  * These references are dropped at the same time we are calling
  50  * audit_free_names(), etc.
  51  *
  52  * Cyclic lists galore:
  53  * tree.chunks anchors chunk.owners[].list                      hash_lock
  54  * tree.rules anchors rule.rlist                                audit_filter_mutex
  55  * chunk.trees anchors tree.same_root                           hash_lock
  56  * chunk.hash is a hash with middle bits of watch.inode as
  57  * a hash function.                                             RCU, hash_lock
  58  *
  59  * tree is refcounted; one reference for "some rules on rules_list refer to
  60  * it", one for each chunk with pointer to it.
  61  *
  62  * chunk is refcounted by embedded fsnotify_mark + .refs (non-zero refcount
  63  * of watch contributes 1 to .refs).
  64  *
  65  * node.index allows to get from node.list to containing chunk.
  66  * MSB of that sucker is stolen to mark taggings that we might have to
  67  * revert - several operations have very unpleasant cleanup logics and
  68  * that makes a difference.  Some.
  69  */
  70
  71 static struct fsnotify_group *audit_tree_group;
  72
  73 static struct audit_tree *alloc_tree(const char *s)
  74 {
  75         struct audit_tree *tree;
  76
  77         tree = kmalloc(sizeof(struct audit_tree) + strlen(s) + 1, GFP_KERNEL);
  78         if (tree) {
  79                 atomic_set(&tree->count, 1);
  80                 tree->goner = 0;
  81                 INIT_LIST_HEAD(&tree->chunks);
  82                 INIT_LIST_HEAD(&tree->rules);
  83                 INIT_LIST_HEAD(&tree->list);
  84                 INIT_LIST_HEAD(&tree->same_root);
  85                 tree->root = NULL;
  86                 strcpy(tree->pathname, s);
  87         }
  88         return tree;
  89 }
  90
  91 static inline void get_tree(struct audit_tree *tree)
  92 {
  93         atomic_inc(&tree->count);
  94 }
  95
  96 static inline void put_tree(struct audit_tree *tree)
  97 {
  98         if (atomic_dec_and_test(&tree->count))
  99                 kfree_rcu(tree, head);
 100 }
 101
 102 /* to avoid bringing the entire thing in audit.h */
 103 const char *audit_tree_path(struct audit_tree *tree)
 104 {
 105         return tree->pathname;
 106 }
 107
 108 static void free_chunk(struct audit_chunk *chunk)
 109 {
 110         int i;
 111
 112         for (i = 0; i < chunk->count; i++) {
 113                 if (chunk->owners[i].owner)
 114                         put_tree(chunk->owners[i].owner);
 115         }
 116         kfree(chunk);
 117 }
 118
 119 void audit_put_chunk(struct audit_chunk *chunk)
 120 {
 121         if (atomic_long_dec_and_test(&chunk->refs))
 122                 free_chunk(chunk);
 123 }
 124
 125 static void __put_chunk(struct rcu_head *rcu)
 126 {
 127         struct audit_chunk *chunk = container_of(rcu, struct audit_chunk, head);
 128         audit_put_chunk(chunk);
 129 }
 130
 131 static void audit_tree_destroy_watch(struct fsnotify_mark *entry)
 132 {
 133         struct audit_chunk *chunk = container_of(entry, struct audit_chunk, mark);
 134         call_rcu(&chunk->head, __put_chunk);
 135 }
 136
 137 static struct audit_chunk *alloc_chunk(int count)
 138 {
 139         struct audit_chunk *chunk;
 140         size_t size;
 141         int i;
 142
 143         size = offsetof(struct audit_chunk, owners) + count * sizeof(struct node);
 144         chunk = kzalloc(size, GFP_KERNEL);
 145         if (!chunk)
 146                 return NULL;
 147
 148         INIT_LIST_HEAD(&chunk->hash);
 149         INIT_LIST_HEAD(&chunk->trees);
 150         chunk->count = count;
 151         atomic_long_set(&chunk->refs, 1);
 152         for (i = 0; i < count; i++) {
 153                 INIT_LIST_HEAD(&chunk->owners[i].list);
 154                 chunk->owners[i].index = i;
 155         }
 156         fsnotify_init_mark(&chunk->mark, audit_tree_destroy_watch);
 157         return chunk;
 158 }
 159
 160 enum {HASH_SIZE = 128};
 161 static struct list_head chunk_hash_heads[HASH_SIZE];
 162 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(hash_lock);
 163
 164 static inline struct list_head *chunk_hash(const struct inode *inode)
 165 {
 166         unsigned long n = (unsigned long)inode / L1_CACHE_BYTES;
 167         return chunk_hash_heads + n % HASH_SIZE;
 168 }
 169
 170 /* hash_lock & entry->lock is held by caller */
 171 static void insert_hash(struct audit_chunk *chunk)
 172 {
 173         struct fsnotify_mark *entry = &chunk->mark;
 174         struct list_head *list;
 175
 176         if (!entry->i.inode)
 177                 return;
 178         list = chunk_hash(entry->i.inode);
 179         list_add_rcu(&chunk->hash, list);
 180 }
 181
 182 /* called under rcu_read_lock */
 183 struct audit_chunk *audit_tree_lookup(const struct inode *inode)
 184 {
 185         struct list_head *list = chunk_hash(inode);
 186         struct audit_chunk *p;
 187
 188         list_for_each_entry_rcu(p, list, hash) {
 189                 /* mark.inode may have gone NULL, but who cares? */
 190                 if (p->mark.i.inode == inode) {
 191                         atomic_long_inc(&p->refs);
 192                         return p;
 193                 }
 194         }
 195         return NULL;
 196 }
 197
 198 int audit_tree_match(struct audit_chunk *chunk, struct audit_tree *tree)
 199 {
 200         int n;
 201         for (n = 0; n < chunk->count; n++)
 202                 if (chunk->owners[n].owner == tree)
 203                         return 1;
 204         return 0;
 205 }
 206
 207 /* tagging and untagging inodes with trees */
 208
 209 static struct audit_chunk *find_chunk(struct node *p)
 210 {
 211         int index = p->index & ~(1U<<31);
 212         p -= index;
 213         return container_of(p, struct audit_chunk, owners[0]);
 214 }
 215
 216 static void untag_chunk(struct node *p)
 217 {
 218         struct audit_chunk *chunk = find_chunk(p);
 219         struct fsnotify_mark *entry = &chunk->mark;
 220         struct audit_chunk *new = NULL;
 221         struct audit_tree *owner;
 222         int size = chunk->count - 1;
 223         int i, j;
 224
 225         fsnotify_get_mark(entry);
 226
 227         spin_unlock(&hash_lock);
 228
 229         if (size)
 230                 new = alloc_chunk(size);
 231
 232         spin_lock(&entry->lock);
 233         if (chunk->dead || !entry->i.inode) {
 234                 spin_unlock(&entry->lock);
 235                 if (new)
 236                         free_chunk(new);
 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(entry, audit_tree_group);
 253                 goto out;
 254         }
 255
 256         if (!new)
 257                 goto Fallback;
 258
 259         fsnotify_duplicate_mark(&new->mark, entry);
 260         if (fsnotify_add_mark(&new->mark, new->mark.group, new->mark.i.inode, NULL, 1)) {
 261                 fsnotify_put_mark(&new->mark);
 262                 goto Fallback;
 263         }
 264
 265         chunk->dead = 1;
 266         spin_lock(&hash_lock);
 267         list_replace_init(&chunk->trees, &new->trees);
 268         if (owner->root == chunk) {
 269                 list_del_init(&owner->same_root);
 270                 owner->root = NULL;
 271         }
 272
 273         for (i = j = 0; j <= size; i++, j++) {
 274                 struct audit_tree *s;
 275                 if (&chunk->owners[j] == p) {
 276                         list_del_init(&p->list);
 277                         i--;
 278                         continue;
 279                 }
 280                 s = chunk->owners[j].owner;
 281                 new->owners[i].owner = s;
 282                 new->owners[i].index = chunk->owners[j].index - j + i;
 283                 if (!s) /* result of earlier fallback */
 284                         continue;
 285                 get_tree(s);
 286                 list_replace_init(&chunk->owners[j].list, &new->owners[i].list);
 287         }
 288
 289         list_replace_rcu(&chunk->hash, &new->hash);
 290         list_for_each_entry(owner, &new->trees, same_root)
 291                 owner->root = new;
 292         spin_unlock(&hash_lock);
 293         spin_unlock(&entry->lock);
 294         fsnotify_destroy_mark(entry, audit_tree_group);
 295         fsnotify_put_mark(&new->mark);  /* drop initial reference */
 296         goto out;
 297
 298 Fallback:
 299         // do the best we can
 300         spin_lock(&hash_lock);
 301         if (owner->root == chunk) {
 302                 list_del_init(&owner->same_root);
 303                 owner->root = NULL;
 304         }
 305         list_del_init(&p->list);
 306         p->owner = NULL;
 307         put_tree(owner);
 308         spin_unlock(&hash_lock);
 309         spin_unlock(&entry->lock);
 310 out:
 311         fsnotify_put_mark(entry);
 312         spin_lock(&hash_lock);
 313 }
 314
 315 static int create_chunk(struct inode *inode, struct audit_tree *tree)
 316 {
 317         struct fsnotify_mark *entry;
 318         struct audit_chunk *chunk = alloc_chunk(1);
 319         if (!chunk)
 320                 return -ENOMEM;
 321
 322         entry = &chunk->mark;
 323         if (fsnotify_add_mark(entry, audit_tree_group, inode, NULL, 0)) {
 324                 fsnotify_put_mark(entry);
 325                 return -ENOSPC;
 326         }
 327
 328         spin_lock(&entry->lock);
 329         spin_lock(&hash_lock);
 330         if (tree->goner) {
 331                 spin_unlock(&hash_lock);
 332                 chunk->dead = 1;
 333                 spin_unlock(&entry->lock);
 334                 fsnotify_destroy_mark(entry, audit_tree_group);
 335                 fsnotify_put_mark(entry);
 336                 return 0;
 337         }
 338         chunk->owners[0].index = (1U << 31);
 339         chunk->owners[0].owner = tree;
 340         get_tree(tree);
 341         list_add(&chunk->owners[0].list, &tree->chunks);
 342         if (!tree->root) {
 343                 tree->root = chunk;
 344                 list_add(&tree->same_root, &chunk->trees);
 345         }
 346         insert_hash(chunk);
 347         spin_unlock(&hash_lock);
 348         spin_unlock(&entry->lock);
 349         fsnotify_put_mark(entry);       /* drop initial reference */
 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 *old_entry, *chunk_entry;
 357         struct audit_tree *owner;
 358         struct audit_chunk *chunk, *old;
 359         struct node *p;
 360         int n;
 361
 362         old_entry = fsnotify_find_inode_mark(audit_tree_group, inode);
 363         if (!old_entry)
 364                 return create_chunk(inode, tree);
 365
 366         old = container_of(old_entry, struct audit_chunk, mark);
 367
 368         /* are we already there? */
 369         spin_lock(&hash_lock);
 370         for (n = 0; n < old->count; n++) {
 371                 if (old->owners[n].owner == tree) {
 372                         spin_unlock(&hash_lock);
 373                         fsnotify_put_mark(old_entry);
 374                         return 0;
 375                 }
 376         }
 377         spin_unlock(&hash_lock);
 378
 379         chunk = alloc_chunk(old->count + 1);
 380         if (!chunk) {
 381                 fsnotify_put_mark(old_entry);
 382                 return -ENOMEM;
 383         }
 384
 385         chunk_entry = &chunk->mark;
 386
 387         spin_lock(&old_entry->lock);
 388         if (!old_entry->i.inode) {
 389                 /* old_entry is being shot, lets just lie */
 390                 spin_unlock(&old_entry->lock);
 391                 fsnotify_put_mark(old_entry);
 392                 free_chunk(chunk);
 393                 return -ENOENT;
 394         }
 395
 396         fsnotify_duplicate_mark(chunk_entry, old_entry);
 397         if (fsnotify_add_mark(chunk_entry, chunk_entry->group, chunk_entry->i.inode, NULL, 1)) {
 398                 spin_unlock(&old_entry->lock);
 399                 fsnotify_put_mark(chunk_entry);
 400                 fsnotify_put_mark(old_entry);
 401                 return -ENOSPC;
 402         }
 403
 404         /* even though we hold old_entry->lock, this is safe since chunk_entry->lock could NEVER have been grabbed before */
 405         spin_lock(&chunk_entry->lock);
 406         spin_lock(&hash_lock);
 407
 408         /* we now hold old_entry->lock, chunk_entry->lock, and hash_lock */
 409         if (tree->goner) {
 410                 spin_unlock(&hash_lock);
 411                 chunk->dead = 1;
 412                 spin_unlock(&chunk_entry->lock);
 413                 spin_unlock(&old_entry->lock);
 414
 415                 fsnotify_destroy_mark(chunk_entry, audit_tree_group);
 416
 417                 fsnotify_put_mark(chunk_entry);
 418                 fsnotify_put_mark(old_entry);
 419                 return 0;
 420         }
 421         list_replace_init(&old->trees, &chunk->trees);
 422         for (n = 0, p = chunk->owners; n < old->count; n++, p++) {
 423                 struct audit_tree *s = old->owners[n].owner;
 424                 p->owner = s;
 425                 p->index = old->owners[n].index;
 426                 if (!s) /* result of fallback in untag */
 427                         continue;
 428                 get_tree(s);
 429                 list_replace_init(&old->owners[n].list, &p->list);
 430         }
 431         p->index = (chunk->count - 1) | (1U<<31);
 432         p->owner = tree;
 433         get_tree(tree);
 434         list_add(&p->list, &tree->chunks);
 435         list_replace_rcu(&old->hash, &chunk->hash);
 436         list_for_each_entry(owner, &chunk->trees, same_root)
 437                 owner->root = chunk;
 438         old->dead = 1;
 439         if (!tree->root) {
 440                 tree->root = chunk;
 441                 list_add(&tree->same_root, &chunk->trees);
 442         }
 443         spin_unlock(&hash_lock);
 444         spin_unlock(&chunk_entry->lock);
 445         spin_unlock(&old_entry->lock);
 446         fsnotify_destroy_mark(old_entry, audit_tree_group);
 447         fsnotify_put_mark(chunk_entry); /* drop initial reference */
 448         fsnotify_put_mark(old_entry); /* pair to fsnotify_find mark_entry */
 449         return 0;
 450 }
 451
 452 static void audit_log_remove_rule(struct audit_krule *rule)
 453 {
 454         struct audit_buffer *ab;
 455
 456         ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE);
 457         if (unlikely(!ab))
 458                 return;
 459         audit_log_format(ab, "op=");
 460         audit_log_string(ab, "remove rule");
 461         audit_log_format(ab, " dir=");
 462         audit_log_untrustedstring(ab, rule->tree->pathname);
 463         audit_log_key(ab, rule->filterkey);
 464         audit_log_format(ab, " list=%d res=1", rule->listnr);
 465         audit_log_end(ab);
 466 }
 467
 468 static void kill_rules(struct audit_tree *tree)
 469 {
 470         struct audit_krule *rule, *next;
 471         struct audit_entry *entry;
 472
 473         list_for_each_entry_safe(rule, next, &tree->rules, rlist) {
 474                 entry = container_of(rule, struct audit_entry, rule);
 475
 476                 list_del_init(&rule->rlist);
 477                 if (rule->tree) {
 478                         /* not a half-baked one */
 479                         audit_log_remove_rule(rule);
 480                         rule->tree = NULL;
 481                         list_del_rcu(&entry->list);
 482                         list_del(&entry->rule.list);
 483                         call_rcu(&entry->rcu, audit_free_rule_rcu);
 484                 }
 485         }
 486 }
 487
 488 /*
 489  * finish killing struct audit_tree
 490  */
 491 static void prune_one(struct audit_tree *victim)
 492 {
 493         spin_lock(&hash_lock);
 494         while (!list_empty(&victim->chunks)) {
 495                 struct node *p;
 496
 497                 p = list_entry(victim->chunks.next, struct node, list);
 498
 499                 untag_chunk(p);
 500         }
 501         spin_unlock(&hash_lock);
 502         put_tree(victim);
 503 }
 504
 505 /* trim the uncommitted chunks from tree */
 506
 507 static void trim_marked(struct audit_tree *tree)
 508 {
 509         struct list_head *p, *q;
 510         spin_lock(&hash_lock);
 511         if (tree->goner) {
 512                 spin_unlock(&hash_lock);
 513                 return;
 514         }
 515         /* reorder */
 516         for (p = tree->chunks.next; p != &tree->chunks; p = q) {
 517                 struct node *node = list_entry(p, struct node, list);
 518                 q = p->next;
 519                 if (node->index & (1U<<31)) {
 520                         list_del_init(p);
 521                         list_add(p, &tree->chunks);
 522                 }
 523         }
 524
 525         while (!list_empty(&tree->chunks)) {
 526                 struct node *node;
 527
 528                 node = list_entry(tree->chunks.next, struct node, list);
 529
 530                 /* have we run out of marked? */
 531                 if (!(node->index & (1U<<31)))
 532                         break;
 533
 534                 untag_chunk(node);
 535         }
 536         if (!tree->root && !tree->goner) {
 537                 tree->goner = 1;
 538                 spin_unlock(&hash_lock);
 539                 mutex_lock(&audit_filter_mutex);
 540                 kill_rules(tree);
 541                 list_del_init(&tree->list);
 542                 mutex_unlock(&audit_filter_mutex);
 543                 prune_one(tree);
 544         } else {
 545                 spin_unlock(&hash_lock);
 546         }
 547 }
 548
 549 static void audit_schedule_prune(void);
 550
 551 /* called with audit_filter_mutex */
 552 int audit_remove_tree_rule(struct audit_krule *rule)
 553 {
 554         struct audit_tree *tree;
 555         tree = rule->tree;
 556         if (tree) {
 557                 spin_lock(&hash_lock);
 558                 list_del_init(&rule->rlist);
 559                 if (list_empty(&tree->rules) && !tree->goner) {
 560                         tree->root = NULL;
 561                         list_del_init(&tree->same_root);
 562                         tree->goner = 1;
 563                         list_move(&tree->list, &prune_list);
 564                         rule->tree = NULL;
 565                         spin_unlock(&hash_lock);
 566                         audit_schedule_prune();
 567                         return 1;
 568                 }
 569                 rule->tree = NULL;
 570                 spin_unlock(&hash_lock);
 571                 return 1;
 572         }
 573         return 0;
 574 }
 575
 576 static int compare_root(struct vfsmount *mnt, void *arg)
 577 {
 578         return mnt->mnt_root->d_inode == arg;
 579 }
 580
 581 void audit_trim_trees(void)
 582 {
 583         struct list_head cursor;
 584
 585         mutex_lock(&audit_filter_mutex);
 586         list_add(&cursor, &tree_list);
 587         while (cursor.next != &tree_list) {
 588                 struct audit_tree *tree;
 589                 struct path path;
 590                 struct vfsmount *root_mnt;
 591                 struct node *node;
 592                 int err;
 593
 594                 tree = container_of(cursor.next, struct audit_tree, list);
 595                 get_tree(tree);
 596                 list_del(&cursor);
 597                 list_add(&cursor, &tree->list);
 598                 mutex_unlock(&audit_filter_mutex);
 599
 600                 err = kern_path(tree->pathname, 0, &path);
 601                 if (err)
 602                         goto skip_it;
 603
 604                 root_mnt = collect_mounts(&path);
 605                 path_put(&path);
 606                 if (IS_ERR(root_mnt))
 607                         goto skip_it;
 608
 609                 spin_lock(&hash_lock);
 610                 list_for_each_entry(node, &tree->chunks, list) {
 611                         struct audit_chunk *chunk = find_chunk(node);
 612                         /* this could be NULL if the watch is dying else where... */
 613                         struct inode *inode = chunk->mark.i.inode;
 614                         node->index |= 1U<<31;
 615                         if (iterate_mounts(compare_root, inode, root_mnt))
 616                                 node->index &= ~(1U<<31);
 617                 }
 618                 spin_unlock(&hash_lock);
 619                 trim_marked(tree);
 620                 drop_collected_mounts(root_mnt);
 621 skip_it:
 622                 put_tree(tree);
 623                 mutex_lock(&audit_filter_mutex);
 624         }
 625         list_del(&cursor);
 626         mutex_unlock(&audit_filter_mutex);
 627 }
 628
 629 int audit_make_tree(struct audit_krule *rule, char *pathname, u32 op)
 630 {
 631
 632         if (pathname[0] != '/' ||
 633             rule->listnr != AUDIT_FILTER_EXIT ||
 634             op != Audit_equal ||
 635             rule->inode_f || rule->watch || rule->tree)
 636                 return -EINVAL;
 637         rule->tree = alloc_tree(pathname);
 638         if (!rule->tree)
 639                 return -ENOMEM;
 640         return 0;
 641 }
 642
 643 void audit_put_tree(struct audit_tree *tree)
 644 {
 645         put_tree(tree);
 646 }
 647
 648 static int tag_mount(struct vfsmount *mnt, void *arg)
 649 {
 650         return tag_chunk(mnt->mnt_root->d_inode, arg);
 651 }
 652
 653 /* called with audit_filter_mutex */
 654 int audit_add_tree_rule(struct audit_krule *rule)
 655 {
 656         struct audit_tree *seed = rule->tree, *tree;
 657         struct path path;
 658         struct vfsmount *mnt;
 659         int err;
 660
 661         rule->tree = NULL;
 662         list_for_each_entry(tree, &tree_list, list) {
 663                 if (!strcmp(seed->pathname, tree->pathname)) {
 664                         put_tree(seed);
 665                         rule->tree = tree;
 666                         list_add(&rule->rlist, &tree->rules);
 667                         return 0;
 668                 }
 669         }
 670         tree = seed;
 671         list_add(&tree->list, &tree_list);
 672         list_add(&rule->rlist, &tree->rules);
 673         /* do not set rule->tree yet */
 674         mutex_unlock(&audit_filter_mutex);
 675
 676         err = kern_path(tree->pathname, 0, &path);
 677         if (err)
 678                 goto Err;
 679         mnt = collect_mounts(&path);
 680         path_put(&path);
 681         if (IS_ERR(mnt)) {
 682                 err = PTR_ERR(mnt);
 683                 goto Err;
 684         }
 685
 686         get_tree(tree);
 687         err = iterate_mounts(tag_mount, tree, mnt);
 688         drop_collected_mounts(mnt);
 689
 690         if (!err) {
 691                 struct node *node;
 692                 spin_lock(&hash_lock);
 693                 list_for_each_entry(node, &tree->chunks, list)
 694                         node->index &= ~(1U<<31);
 695                 spin_unlock(&hash_lock);
 696         } else {
 697                 trim_marked(tree);
 698                 goto Err;
 699         }
 700
 701         mutex_lock(&audit_filter_mutex);
 702         if (list_empty(&rule->rlist)) {
 703                 put_tree(tree);
 704                 return -ENOENT;
 705         }
 706         rule->tree = tree;
 707         put_tree(tree);
 708
 709         return 0;
 710 Err:
 711         mutex_lock(&audit_filter_mutex);
 712         list_del_init(&tree->list);
 713         list_del_init(&tree->rules);
 714         put_tree(tree);
 715         return err;
 716 }
 717
 718 int audit_tag_tree(char *old, char *new)
 719 {
 720         struct list_head cursor, barrier;
 721         int failed = 0;
 722         struct path path1, path2;
 723         struct vfsmount *tagged;
 724         int err;
 725
 726         err = kern_path(new, 0, &path2);
 727         if (err)
 728                 return err;
 729         tagged = collect_mounts(&path2);
 730         path_put(&path2);
 731         if (IS_ERR(tagged))
 732                 return PTR_ERR(tagged);
 733
 734         err = kern_path(old, 0, &path1);
 735         if (err) {
 736                 drop_collected_mounts(tagged);
 737                 return err;
 738         }
 739
 740         mutex_lock(&audit_filter_mutex);
 741         list_add(&barrier, &tree_list);
 742         list_add(&cursor, &barrier);
 743
 744         while (cursor.next != &tree_list) {
 745                 struct audit_tree *tree;
 746                 int good_one = 0;
 747
 748                 tree = container_of(cursor.next, struct audit_tree, list);
 749                 get_tree(tree);
 750                 list_del(&cursor);
 751                 list_add(&cursor, &tree->list);
 752                 mutex_unlock(&audit_filter_mutex);
 753
 754                 err = kern_path(tree->pathname, 0, &path2);
 755                 if (!err) {
 756                         good_one = path_is_under(&path1, &path2);
 757                         path_put(&path2);
 758                 }
 759
 760                 if (!good_one) {
 761                         put_tree(tree);
 762                         mutex_lock(&audit_filter_mutex);
 763                         continue;
 764                 }
 765
 766                 failed = iterate_mounts(tag_mount, tree, tagged);
 767                 if (failed) {
 768                         put_tree(tree);
 769                         mutex_lock(&audit_filter_mutex);
 770                         break;
 771                 }
 772
 773                 mutex_lock(&audit_filter_mutex);
 774                 spin_lock(&hash_lock);
 775                 if (!tree->goner) {
 776                         list_del(&tree->list);
 777                         list_add(&tree->list, &tree_list);
 778                 }
 779                 spin_unlock(&hash_lock);
 780                 put_tree(tree);
 781         }
 782
 783         while (barrier.prev != &tree_list) {
 784                 struct audit_tree *tree;
 785
 786                 tree = container_of(barrier.prev, struct audit_tree, list);
 787                 get_tree(tree);
 788                 list_del(&tree->list);
 789                 list_add(&tree->list, &barrier);
 790                 mutex_unlock(&audit_filter_mutex);
 791
 792                 if (!failed) {
 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                 }
 801
 802                 put_tree(tree);
 803                 mutex_lock(&audit_filter_mutex);
 804         }
 805         list_del(&barrier);
 806         list_del(&cursor);
 807         mutex_unlock(&audit_filter_mutex);
 808         path_put(&path1);
 809         drop_collected_mounts(tagged);
 810         return failed;
 811 }
 812
 813 /*
 814  * That gets run when evict_chunk() ends up needing to kill audit_tree.
 815  * Runs from a separate thread.
 816  */
 817 static int prune_tree_thread(void *unused)
 818 {
 819         mutex_lock(&audit_cmd_mutex);
 820         mutex_lock(&audit_filter_mutex);
 821
 822         while (!list_empty(&prune_list)) {
 823                 struct audit_tree *victim;
 824
 825                 victim = list_entry(prune_list.next, struct audit_tree, list);
 826                 list_del_init(&victim->list);
 827
 828                 mutex_unlock(&audit_filter_mutex);
 829
 830                 prune_one(victim);
 831
 832                 mutex_lock(&audit_filter_mutex);
 833         }
 834
 835         mutex_unlock(&audit_filter_mutex);
 836         mutex_unlock(&audit_cmd_mutex);
 837         return 0;
 838 }
 839
 840 static void audit_schedule_prune(void)
 841 {
 842         kthread_run(prune_tree_thread, NULL, "audit_prune_tree");
 843 }
 844
 845 /*
 846  * ... and that one is done if evict_chunk() decides to delay until the end
 847  * of syscall.  Runs synchronously.
 848  */
 849 void audit_kill_trees(struct list_head *list)
 850 {
 851         mutex_lock(&audit_cmd_mutex);
 852         mutex_lock(&audit_filter_mutex);
 853
 854         while (!list_empty(list)) {
 855                 struct audit_tree *victim;
 856
 857                 victim = list_entry(list->next, struct audit_tree, list);
 858                 kill_rules(victim);
 859                 list_del_init(&victim->list);
 860
 861                 mutex_unlock(&audit_filter_mutex);
 862
 863                 prune_one(victim);
 864
 865                 mutex_lock(&audit_filter_mutex);
 866         }
 867
 868         mutex_unlock(&audit_filter_mutex);
 869         mutex_unlock(&audit_cmd_mutex);
 870 }
 871
 872 /*
 873  *  Here comes the stuff asynchronous to auditctl operations
 874  */
 875
 876 static void evict_chunk(struct audit_chunk *chunk)
 877 {
 878         struct audit_tree *owner;
 879         struct list_head *postponed = audit_killed_trees();
 880         int need_prune = 0;
 881         int n;
 882
 883         if (chunk->dead)
 884                 return;
 885
 886         chunk->dead = 1;
 887         mutex_lock(&audit_filter_mutex);
 888         spin_lock(&hash_lock);
 889         while (!list_empty(&chunk->trees)) {
 890                 owner = list_entry(chunk->trees.next,
 891                                    struct audit_tree, same_root);
 892                 owner->goner = 1;
 893                 owner->root = NULL;
 894                 list_del_init(&owner->same_root);
 895                 spin_unlock(&hash_lock);
 896                 if (!postponed) {
 897                         kill_rules(owner);
 898                         list_move(&owner->list, &prune_list);
 899                         need_prune = 1;
 900                 } else {
 901                         list_move(&owner->list, postponed);
 902                 }
 903                 spin_lock(&hash_lock);
 904         }
 905         list_del_rcu(&chunk->hash);
 906         for (n = 0; n < chunk->count; n++)
 907                 list_del_init(&chunk->owners[n].list);
 908         spin_unlock(&hash_lock);
 909         if (need_prune)
 910                 audit_schedule_prune();
 911         mutex_unlock(&audit_filter_mutex);
 912 }
 913
 914 static int audit_tree_handle_event(struct fsnotify_group *group,
 915                                    struct fsnotify_mark *inode_mark,
 916                                    struct fsnotify_mark *vfsmonut_mark,
 917                                    struct fsnotify_event *event)
 918 {
 919         BUG();
 920         return -EOPNOTSUPP;
 921 }
 922
 923 static void audit_tree_freeing_mark(struct fsnotify_mark *entry, struct fsnotify_group *group)
 924 {
 925         struct audit_chunk *chunk = container_of(entry, struct audit_chunk, mark);
 926
 927         evict_chunk(chunk);
 928
 929         /*
 930          * We are guaranteed to have at least one reference to the mark from
 931          * either the inode or the caller of fsnotify_destroy_mark().
 932          */
 933         BUG_ON(atomic_read(&entry->refcnt) < 1);
 934 }
 935
 936 static bool audit_tree_send_event(struct fsnotify_group *group, struct inode *inode,
 937                                   struct fsnotify_mark *inode_mark,
 938                                   struct fsnotify_mark *vfsmount_mark,
 939                                   __u32 mask, void *data, int data_type)
 940 {
 941         return false;
 942 }
 943
 944 static const struct fsnotify_ops audit_tree_ops = {
 945         .handle_event = audit_tree_handle_event,
 946         .should_send_event = audit_tree_send_event,
 947         .free_group_priv = NULL,
 948         .free_event_priv = NULL,
 949         .freeing_mark = audit_tree_freeing_mark,
 950 };
 951
 952 static int __init audit_tree_init(void)
 953 {
 954         int i;
 955
 956         audit_tree_group = fsnotify_alloc_group(&audit_tree_ops);
 957         if (IS_ERR(audit_tree_group))
 958                 audit_panic("cannot initialize fsnotify group for rectree watches");
 959
 960         for (i = 0; i < HASH_SIZE; i++)
 961                 INIT_LIST_HEAD(&chunk_hash_heads[i]);
 962
 963         return 0;
 964 }
 965 __initcall(audit_tree_init);