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 static 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
  72 static struct fsnotify_group *audit_tree_group;
  73
  74 static 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
  92 static inline void get_tree(struct audit_tree *tree)
  93 {
  94         atomic_inc(&tree->count);
  95 }
  96
  97 static 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 */
 104 const char *audit_tree_path(struct audit_tree *tree)
 105 {
 106         return tree->pathname;
 107 }
 108
 109 static 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
 120 void audit_put_chunk(struct audit_chunk *chunk)
 121 {
 122         if (atomic_long_dec_and_test(&chunk->refs))
 123                 free_chunk(chunk);
 124 }
 125
 126 static 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
 132 static 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
 138 static 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
 162 enum {HASH_SIZE = 128};
 163 static struct list_head chunk_hash_heads[HASH_SIZE];
 164 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(hash_lock);
 165
 166 static 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 */
 173 static 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 */
 185 struct 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
 200 bool 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
 211 static 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
 218 static 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
 300 Fallback:
 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);
 312 out:
 313         fsnotify_put_mark(entry);
 314         spin_lock(&hash_lock);
 315 }
 316
 317 static 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 */
 356 static 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
 454 static 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
 470 static 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  */
 495 static 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
 511 static 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
 553 static void audit_schedule_prune(void);
 554
 555 /* called with audit_filter_mutex */
 556 int 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
 580 static int compare_root(struct vfsmount *mnt, void *arg)
 581 {
 582         return d_backing_inode(mnt->mnt_root) == arg;
 583 }
 584
 585 void 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);
 625 skip_it:
 626                 put_tree(tree);
 627                 mutex_lock(&audit_filter_mutex);
 628         }
 629         list_del(&cursor);
 630         mutex_unlock(&audit_filter_mutex);
 631 }
 632
 633 int 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
 647 void audit_put_tree(struct audit_tree *tree)
 648 {
 649         put_tree(tree);
 650 }
 651
 652 static 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  */
 661 static int prune_tree_thread(void *unused)
 662 {
 663         for (;;) {
 664                 if (list_empty(&prune_list)) {
 665                         set_current_state(TASK_INTERRUPTIBLE);
 666                         schedule();
 667                 }
 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
 692 static int audit_launch_prune(void)
 693 {
 694         if (prune_thread)
 695                 return 0;
 696         prune_thread = kthread_run(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         }
 703         return 0;
 704 }
 705
 706 /* called with audit_filter_mutex */
 707 int audit_add_tree_rule(struct audit_krule *rule)
 708 {
 709         struct audit_tree *seed = rule->tree, *tree;
 710         struct path path;
 711         struct vfsmount *mnt;
 712         int err;
 713
 714         rule->tree = NULL;
 715         list_for_each_entry(tree, &tree_list, list) {
 716                 if (!strcmp(seed->pathname, tree->pathname)) {
 717                         put_tree(seed);
 718                         rule->tree = tree;
 719                         list_add(&rule->rlist, &tree->rules);
 720                         return 0;
 721                 }
 722         }
 723         tree = seed;
 724         list_add(&tree->list, &tree_list);
 725         list_add(&rule->rlist, &tree->rules);
 726         /* do not set rule->tree yet */
 727         mutex_unlock(&audit_filter_mutex);
 728
 729         if (unlikely(!prune_thread)) {
 730                 err = audit_launch_prune();
 731                 if (err)
 732                         goto Err;
 733         }
 734
 735         err = kern_path(tree->pathname, 0, &path);
 736         if (err)
 737                 goto Err;
 738         mnt = collect_mounts(&path);
 739         path_put(&path);
 740         if (IS_ERR(mnt)) {
 741                 err = PTR_ERR(mnt);
 742                 goto Err;
 743         }
 744
 745         get_tree(tree);
 746         err = iterate_mounts(tag_mount, tree, mnt);
 747         drop_collected_mounts(mnt);
 748
 749         if (!err) {
 750                 struct node *node;
 751                 spin_lock(&hash_lock);
 752                 list_for_each_entry(node, &tree->chunks, list)
 753                         node->index &= ~(1U<<31);
 754                 spin_unlock(&hash_lock);
 755         } else {
 756                 trim_marked(tree);
 757                 goto Err;
 758         }
 759
 760         mutex_lock(&audit_filter_mutex);
 761         if (list_empty(&rule->rlist)) {
 762                 put_tree(tree);
 763                 return -ENOENT;
 764         }
 765         rule->tree = tree;
 766         put_tree(tree);
 767
 768         return 0;
 769 Err:
 770         mutex_lock(&audit_filter_mutex);
 771         list_del_init(&tree->list);
 772         list_del_init(&tree->rules);
 773         put_tree(tree);
 774         return err;
 775 }
 776
 777 int audit_tag_tree(char *old, char *new)
 778 {
 779         struct list_head cursor, barrier;
 780         int failed = 0;
 781         struct path path1, path2;
 782         struct vfsmount *tagged;
 783         int err;
 784
 785         err = kern_path(new, 0, &path2);
 786         if (err)
 787                 return err;
 788         tagged = collect_mounts(&path2);
 789         path_put(&path2);
 790         if (IS_ERR(tagged))
 791                 return PTR_ERR(tagged);
 792
 793         err = kern_path(old, 0, &path1);
 794         if (err) {
 795                 drop_collected_mounts(tagged);
 796                 return err;
 797         }
 798
 799         mutex_lock(&audit_filter_mutex);
 800         list_add(&barrier, &tree_list);
 801         list_add(&cursor, &barrier);
 802
 803         while (cursor.next != &tree_list) {
 804                 struct audit_tree *tree;
 805                 int good_one = 0;
 806
 807                 tree = container_of(cursor.next, struct audit_tree, list);
 808                 get_tree(tree);
 809                 list_del(&cursor);
 810                 list_add(&cursor, &tree->list);
 811                 mutex_unlock(&audit_filter_mutex);
 812
 813                 err = kern_path(tree->pathname, 0, &path2);
 814                 if (!err) {
 815                         good_one = path_is_under(&path1, &path2);
 816                         path_put(&path2);
 817                 }
 818
 819                 if (!good_one) {
 820                         put_tree(tree);
 821                         mutex_lock(&audit_filter_mutex);
 822                         continue;
 823                 }
 824
 825                 failed = iterate_mounts(tag_mount, tree, tagged);
 826                 if (failed) {
 827                         put_tree(tree);
 828                         mutex_lock(&audit_filter_mutex);
 829                         break;
 830                 }
 831
 832                 mutex_lock(&audit_filter_mutex);
 833                 spin_lock(&hash_lock);
 834                 if (!tree->goner) {
 835                         list_del(&tree->list);
 836                         list_add(&tree->list, &tree_list);
 837                 }
 838                 spin_unlock(&hash_lock);
 839                 put_tree(tree);
 840         }
 841
 842         while (barrier.prev != &tree_list) {
 843                 struct audit_tree *tree;
 844
 845                 tree = container_of(barrier.prev, struct audit_tree, list);
 846                 get_tree(tree);
 847                 list_del(&tree->list);
 848                 list_add(&tree->list, &barrier);
 849                 mutex_unlock(&audit_filter_mutex);
 850
 851                 if (!failed) {
 852                         struct node *node;
 853                         spin_lock(&hash_lock);
 854                         list_for_each_entry(node, &tree->chunks, list)
 855                                 node->index &= ~(1U<<31);
 856                         spin_unlock(&hash_lock);
 857                 } else {
 858                         trim_marked(tree);
 859                 }
 860
 861                 put_tree(tree);
 862                 mutex_lock(&audit_filter_mutex);
 863         }
 864         list_del(&barrier);
 865         list_del(&cursor);
 866         mutex_unlock(&audit_filter_mutex);
 867         path_put(&path1);
 868         drop_collected_mounts(tagged);
 869         return failed;
 870 }
 871
 872
 873 static void audit_schedule_prune(void)
 874 {
 875         wake_up_process(prune_thread);
 876 }
 877
 878 /*
 879  * ... and that one is done if evict_chunk() decides to delay until the end
 880  * of syscall.  Runs synchronously.
 881  */
 882 void audit_kill_trees(struct list_head *list)
 883 {
 884         mutex_lock(&audit_cmd_mutex);
 885         mutex_lock(&audit_filter_mutex);
 886
 887         while (!list_empty(list)) {
 888                 struct audit_tree *victim;
 889
 890                 victim = list_entry(list->next, struct audit_tree, list);
 891                 kill_rules(victim);
 892                 list_del_init(&victim->list);
 893
 894                 mutex_unlock(&audit_filter_mutex);
 895
 896                 prune_one(victim);
 897
 898                 mutex_lock(&audit_filter_mutex);
 899         }
 900
 901         mutex_unlock(&audit_filter_mutex);
 902         mutex_unlock(&audit_cmd_mutex);
 903 }
 904
 905 /*
 906  *  Here comes the stuff asynchronous to auditctl operations
 907  */
 908
 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         mutex_unlock(&audit_filter_mutex);
 943         if (need_prune)
 944                 audit_schedule_prune();
 945 }
 946
 947 static int audit_tree_handle_event(struct fsnotify_group *group,
 948                                    struct inode *to_tell,
 949                                    struct fsnotify_mark *inode_mark,
 950                                    struct fsnotify_mark *vfsmount_mark,
 951                                    u32 mask, void *data, int data_type,
 952                                    const unsigned char *file_name, u32 cookie)
 953 {
 954         return 0;
 955 }
 956
 957 static void audit_tree_freeing_mark(struct fsnotify_mark *entry, struct fsnotify_group *group)
 958 {
 959         struct audit_chunk *chunk = container_of(entry, struct audit_chunk, mark);
 960
 961         evict_chunk(chunk);
 962
 963         /*
 964          * We are guaranteed to have at least one reference to the mark from
 965          * either the inode or the caller of fsnotify_destroy_mark().
 966          */
 967         BUG_ON(atomic_read(&entry->refcnt) < 1);
 968 }
 969
 970 static const struct fsnotify_ops audit_tree_ops = {
 971         .handle_event = audit_tree_handle_event,
 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_alloc_group(&audit_tree_ops);
 980         if (IS_ERR(audit_tree_group))
 981                 audit_panic("cannot initialize fsnotify group 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);