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         mutex_lock(&entry->group->mark_mutex);
 235         spin_lock(&entry->lock);
 236         if (chunk->dead || !entry->inode) {
 237                 spin_unlock(&entry->lock);
 238                 mutex_unlock(&entry->group->mark_mutex);
 239                 if (new)
 240                         free_chunk(new);
 241                 goto out;
 242         }
 243
 244         owner = p->owner;
 245
 246         if (!size) {
 247                 chunk->dead = 1;
 248                 spin_lock(&hash_lock);
 249                 list_del_init(&chunk->trees);
 250                 if (owner->root == chunk)
 251                         owner->root = NULL;
 252                 list_del_init(&p->list);
 253                 list_del_rcu(&chunk->hash);
 254                 spin_unlock(&hash_lock);
 255                 spin_unlock(&entry->lock);
 256                 mutex_unlock(&entry->group->mark_mutex);
 257                 fsnotify_destroy_mark(entry, audit_tree_group);
 258                 goto out;
 259         }
 260
 261         if (!new)
 262                 goto Fallback;
 263
 264         if (fsnotify_add_mark_locked(&new->mark, entry->group, entry->inode,
 265                                      NULL, 1)) {
 266                 fsnotify_put_mark(&new->mark);
 267                 goto Fallback;
 268         }
 269
 270         chunk->dead = 1;
 271         spin_lock(&hash_lock);
 272         list_replace_init(&chunk->trees, &new->trees);
 273         if (owner->root == chunk) {
 274                 list_del_init(&owner->same_root);
 275                 owner->root = NULL;
 276         }
 277
 278         for (i = j = 0; j <= size; i++, j++) {
 279                 struct audit_tree *s;
 280                 if (&chunk->owners[j] == p) {
 281                         list_del_init(&p->list);
 282                         i--;
 283                         continue;
 284                 }
 285                 s = chunk->owners[j].owner;
 286                 new->owners[i].owner = s;
 287                 new->owners[i].index = chunk->owners[j].index - j + i;
 288                 if (!s) /* result of earlier fallback */
 289                         continue;
 290                 get_tree(s);
 291                 list_replace_init(&chunk->owners[j].list, &new->owners[i].list);
 292         }
 293
 294         list_replace_rcu(&chunk->hash, &new->hash);
 295         list_for_each_entry(owner, &new->trees, same_root)
 296                 owner->root = new;
 297         spin_unlock(&hash_lock);
 298         spin_unlock(&entry->lock);
 299         mutex_unlock(&entry->group->mark_mutex);
 300         fsnotify_destroy_mark(entry, audit_tree_group);
 301         fsnotify_put_mark(&new->mark);  /* drop initial reference */
 302         goto out;
 303
 304 Fallback:
 305         // do the best we can
 306         spin_lock(&hash_lock);
 307         if (owner->root == chunk) {
 308                 list_del_init(&owner->same_root);
 309                 owner->root = NULL;
 310         }
 311         list_del_init(&p->list);
 312         p->owner = NULL;
 313         put_tree(owner);
 314         spin_unlock(&hash_lock);
 315         spin_unlock(&entry->lock);
 316         mutex_unlock(&entry->group->mark_mutex);
 317 out:
 318         fsnotify_put_mark(entry);
 319         spin_lock(&hash_lock);
 320 }
 321
 322 static int create_chunk(struct inode *inode, struct audit_tree *tree)
 323 {
 324         struct fsnotify_mark *entry;
 325         struct audit_chunk *chunk = alloc_chunk(1);
 326         if (!chunk)
 327                 return -ENOMEM;
 328
 329         entry = &chunk->mark;
 330         if (fsnotify_add_mark(entry, audit_tree_group, inode, NULL, 0)) {
 331                 fsnotify_put_mark(entry);
 332                 return -ENOSPC;
 333         }
 334
 335         spin_lock(&entry->lock);
 336         spin_lock(&hash_lock);
 337         if (tree->goner) {
 338                 spin_unlock(&hash_lock);
 339                 chunk->dead = 1;
 340                 spin_unlock(&entry->lock);
 341                 fsnotify_destroy_mark(entry, audit_tree_group);
 342                 fsnotify_put_mark(entry);
 343                 return 0;
 344         }
 345         chunk->owners[0].index = (1U << 31);
 346         chunk->owners[0].owner = tree;
 347         get_tree(tree);
 348         list_add(&chunk->owners[0].list, &tree->chunks);
 349         if (!tree->root) {
 350                 tree->root = chunk;
 351                 list_add(&tree->same_root, &chunk->trees);
 352         }
 353         insert_hash(chunk);
 354         spin_unlock(&hash_lock);
 355         spin_unlock(&entry->lock);
 356         fsnotify_put_mark(entry);       /* drop initial reference */
 357         return 0;
 358 }
 359
 360 /* the first tagged inode becomes root of tree */
 361 static int tag_chunk(struct inode *inode, struct audit_tree *tree)
 362 {
 363         struct fsnotify_mark *old_entry, *chunk_entry;
 364         struct audit_tree *owner;
 365         struct audit_chunk *chunk, *old;
 366         struct node *p;
 367         int n;
 368
 369         old_entry = fsnotify_find_inode_mark(audit_tree_group, inode);
 370         if (!old_entry)
 371                 return create_chunk(inode, tree);
 372
 373         old = container_of(old_entry, struct audit_chunk, mark);
 374
 375         /* are we already there? */
 376         spin_lock(&hash_lock);
 377         for (n = 0; n < old->count; n++) {
 378                 if (old->owners[n].owner == tree) {
 379                         spin_unlock(&hash_lock);
 380                         fsnotify_put_mark(old_entry);
 381                         return 0;
 382                 }
 383         }
 384         spin_unlock(&hash_lock);
 385
 386         chunk = alloc_chunk(old->count + 1);
 387         if (!chunk) {
 388                 fsnotify_put_mark(old_entry);
 389                 return -ENOMEM;
 390         }
 391
 392         chunk_entry = &chunk->mark;
 393
 394         mutex_lock(&old_entry->group->mark_mutex);
 395         spin_lock(&old_entry->lock);
 396         if (!old_entry->inode) {
 397                 /* old_entry is being shot, lets just lie */
 398                 spin_unlock(&old_entry->lock);
 399                 mutex_unlock(&old_entry->group->mark_mutex);
 400                 fsnotify_put_mark(old_entry);
 401                 free_chunk(chunk);
 402                 return -ENOENT;
 403         }
 404
 405         if (fsnotify_add_mark_locked(chunk_entry, old_entry->group,
 406                                      old_entry->inode, NULL, 1)) {
 407                 spin_unlock(&old_entry->lock);
 408                 mutex_unlock(&old_entry->group->mark_mutex);
 409                 fsnotify_put_mark(chunk_entry);
 410                 fsnotify_put_mark(old_entry);
 411                 return -ENOSPC;
 412         }
 413
 414         /* even though we hold old_entry->lock, this is safe since chunk_entry->lock could NEVER have been grabbed before */
 415         spin_lock(&chunk_entry->lock);
 416         spin_lock(&hash_lock);
 417
 418         /* we now hold old_entry->lock, chunk_entry->lock, and hash_lock */
 419         if (tree->goner) {
 420                 spin_unlock(&hash_lock);
 421                 chunk->dead = 1;
 422                 spin_unlock(&chunk_entry->lock);
 423                 spin_unlock(&old_entry->lock);
 424                 mutex_unlock(&old_entry->group->mark_mutex);
 425
 426                 fsnotify_destroy_mark(chunk_entry, audit_tree_group);
 427
 428                 fsnotify_put_mark(chunk_entry);
 429                 fsnotify_put_mark(old_entry);
 430                 return 0;
 431         }
 432         list_replace_init(&old->trees, &chunk->trees);
 433         for (n = 0, p = chunk->owners; n < old->count; n++, p++) {
 434                 struct audit_tree *s = old->owners[n].owner;
 435                 p->owner = s;
 436                 p->index = old->owners[n].index;
 437                 if (!s) /* result of fallback in untag */
 438                         continue;
 439                 get_tree(s);
 440                 list_replace_init(&old->owners[n].list, &p->list);
 441         }
 442         p->index = (chunk->count - 1) | (1U<<31);
 443         p->owner = tree;
 444         get_tree(tree);
 445         list_add(&p->list, &tree->chunks);
 446         list_replace_rcu(&old->hash, &chunk->hash);
 447         list_for_each_entry(owner, &chunk->trees, same_root)
 448                 owner->root = chunk;
 449         old->dead = 1;
 450         if (!tree->root) {
 451                 tree->root = chunk;
 452                 list_add(&tree->same_root, &chunk->trees);
 453         }
 454         spin_unlock(&hash_lock);
 455         spin_unlock(&chunk_entry->lock);
 456         spin_unlock(&old_entry->lock);
 457         mutex_unlock(&old_entry->group->mark_mutex);
 458         fsnotify_destroy_mark(old_entry, audit_tree_group);
 459         fsnotify_put_mark(chunk_entry); /* drop initial reference */
 460         fsnotify_put_mark(old_entry); /* pair to fsnotify_find mark_entry */
 461         return 0;
 462 }
 463
 464 static void audit_tree_log_remove_rule(struct audit_krule *rule)
 465 {
 466         struct audit_buffer *ab;
 467
 468         ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE);
 469         if (unlikely(!ab))
 470                 return;
 471         audit_log_format(ab, "op=remove_rule");
 472         audit_log_format(ab, " dir=");
 473         audit_log_untrustedstring(ab, rule->tree->pathname);
 474         audit_log_key(ab, rule->filterkey);
 475         audit_log_format(ab, " list=%d res=1", rule->listnr);
 476         audit_log_end(ab);
 477 }
 478
 479 static void kill_rules(struct audit_tree *tree)
 480 {
 481         struct audit_krule *rule, *next;
 482         struct audit_entry *entry;
 483
 484         list_for_each_entry_safe(rule, next, &tree->rules, rlist) {
 485                 entry = container_of(rule, struct audit_entry, rule);
 486
 487                 list_del_init(&rule->rlist);
 488                 if (rule->tree) {
 489                         /* not a half-baked one */
 490                         audit_tree_log_remove_rule(rule);
 491                         if (entry->rule.exe)
 492                                 audit_remove_mark(entry->rule.exe);
 493                         rule->tree = NULL;
 494                         list_del_rcu(&entry->list);
 495                         list_del(&entry->rule.list);
 496                         call_rcu(&entry->rcu, audit_free_rule_rcu);
 497                 }
 498         }
 499 }
 500
 501 /*
 502  * finish killing struct audit_tree
 503  */
 504 static void prune_one(struct audit_tree *victim)
 505 {
 506         spin_lock(&hash_lock);
 507         while (!list_empty(&victim->chunks)) {
 508                 struct node *p;
 509
 510                 p = list_entry(victim->chunks.next, struct node, list);
 511
 512                 untag_chunk(p);
 513         }
 514         spin_unlock(&hash_lock);
 515         put_tree(victim);
 516 }
 517
 518 /* trim the uncommitted chunks from tree */
 519
 520 static void trim_marked(struct audit_tree *tree)
 521 {
 522         struct list_head *p, *q;
 523         spin_lock(&hash_lock);
 524         if (tree->goner) {
 525                 spin_unlock(&hash_lock);
 526                 return;
 527         }
 528         /* reorder */
 529         for (p = tree->chunks.next; p != &tree->chunks; p = q) {
 530                 struct node *node = list_entry(p, struct node, list);
 531                 q = p->next;
 532                 if (node->index & (1U<<31)) {
 533                         list_del_init(p);
 534                         list_add(p, &tree->chunks);
 535                 }
 536         }
 537
 538         while (!list_empty(&tree->chunks)) {
 539                 struct node *node;
 540
 541                 node = list_entry(tree->chunks.next, struct node, list);
 542
 543                 /* have we run out of marked? */
 544                 if (!(node->index & (1U<<31)))
 545                         break;
 546
 547                 untag_chunk(node);
 548         }
 549         if (!tree->root && !tree->goner) {
 550                 tree->goner = 1;
 551                 spin_unlock(&hash_lock);
 552                 mutex_lock(&audit_filter_mutex);
 553                 kill_rules(tree);
 554                 list_del_init(&tree->list);
 555                 mutex_unlock(&audit_filter_mutex);
 556                 prune_one(tree);
 557         } else {
 558                 spin_unlock(&hash_lock);
 559         }
 560 }
 561
 562 static void audit_schedule_prune(void);
 563
 564 /* called with audit_filter_mutex */
 565 int audit_remove_tree_rule(struct audit_krule *rule)
 566 {
 567         struct audit_tree *tree;
 568         tree = rule->tree;
 569         if (tree) {
 570                 spin_lock(&hash_lock);
 571                 list_del_init(&rule->rlist);
 572                 if (list_empty(&tree->rules) && !tree->goner) {
 573                         tree->root = NULL;
 574                         list_del_init(&tree->same_root);
 575                         tree->goner = 1;
 576                         list_move(&tree->list, &prune_list);
 577                         rule->tree = NULL;
 578                         spin_unlock(&hash_lock);
 579                         audit_schedule_prune();
 580                         return 1;
 581                 }
 582                 rule->tree = NULL;
 583                 spin_unlock(&hash_lock);
 584                 return 1;
 585         }
 586         return 0;
 587 }
 588
 589 static int compare_root(struct vfsmount *mnt, void *arg)
 590 {
 591         return d_backing_inode(mnt->mnt_root) == arg;
 592 }
 593
 594 void audit_trim_trees(void)
 595 {
 596         struct list_head cursor;
 597
 598         mutex_lock(&audit_filter_mutex);
 599         list_add(&cursor, &tree_list);
 600         while (cursor.next != &tree_list) {
 601                 struct audit_tree *tree;
 602                 struct path path;
 603                 struct vfsmount *root_mnt;
 604                 struct node *node;
 605                 int err;
 606
 607                 tree = container_of(cursor.next, struct audit_tree, list);
 608                 get_tree(tree);
 609                 list_del(&cursor);
 610                 list_add(&cursor, &tree->list);
 611                 mutex_unlock(&audit_filter_mutex);
 612
 613                 err = kern_path(tree->pathname, 0, &path);
 614                 if (err)
 615                         goto skip_it;
 616
 617                 root_mnt = collect_mounts(&path);
 618                 path_put(&path);
 619                 if (IS_ERR(root_mnt))
 620                         goto skip_it;
 621
 622                 spin_lock(&hash_lock);
 623                 list_for_each_entry(node, &tree->chunks, list) {
 624                         struct audit_chunk *chunk = find_chunk(node);
 625                         /* this could be NULL if the watch is dying else where... */
 626                         struct inode *inode = chunk->mark.inode;
 627                         node->index |= 1U<<31;
 628                         if (iterate_mounts(compare_root, inode, root_mnt))
 629                                 node->index &= ~(1U<<31);
 630                 }
 631                 spin_unlock(&hash_lock);
 632                 trim_marked(tree);
 633                 drop_collected_mounts(root_mnt);
 634 skip_it:
 635                 put_tree(tree);
 636                 mutex_lock(&audit_filter_mutex);
 637         }
 638         list_del(&cursor);
 639         mutex_unlock(&audit_filter_mutex);
 640 }
 641
 642 int audit_make_tree(struct audit_krule *rule, char *pathname, u32 op)
 643 {
 644
 645         if (pathname[0] != '/' ||
 646             rule->listnr != AUDIT_FILTER_EXIT ||
 647             op != Audit_equal ||
 648             rule->inode_f || rule->watch || rule->tree)
 649                 return -EINVAL;
 650         rule->tree = alloc_tree(pathname);
 651         if (!rule->tree)
 652                 return -ENOMEM;
 653         return 0;
 654 }
 655
 656 void audit_put_tree(struct audit_tree *tree)
 657 {
 658         put_tree(tree);
 659 }
 660
 661 static int tag_mount(struct vfsmount *mnt, void *arg)
 662 {
 663         return tag_chunk(d_backing_inode(mnt->mnt_root), arg);
 664 }
 665
 666 /*
 667  * That gets run when evict_chunk() ends up needing to kill audit_tree.
 668  * Runs from a separate thread.
 669  */
 670 static int prune_tree_thread(void *unused)
 671 {
 672         for (;;) {
 673                 if (list_empty(&prune_list)) {
 674                         set_current_state(TASK_INTERRUPTIBLE);
 675                         schedule();
 676                 }
 677
 678                 mutex_lock(&audit_cmd_mutex);
 679                 mutex_lock(&audit_filter_mutex);
 680
 681                 while (!list_empty(&prune_list)) {
 682                         struct audit_tree *victim;
 683
 684                         victim = list_entry(prune_list.next,
 685                                         struct audit_tree, list);
 686                         list_del_init(&victim->list);
 687
 688                         mutex_unlock(&audit_filter_mutex);
 689
 690                         prune_one(victim);
 691
 692                         mutex_lock(&audit_filter_mutex);
 693                 }
 694
 695                 mutex_unlock(&audit_filter_mutex);
 696                 mutex_unlock(&audit_cmd_mutex);
 697         }
 698         return 0;
 699 }
 700
 701 static int audit_launch_prune(void)
 702 {
 703         if (prune_thread)
 704                 return 0;
 705         prune_thread = kthread_run(prune_tree_thread, NULL,
 706                                 "audit_prune_tree");
 707         if (IS_ERR(prune_thread)) {
 708                 pr_err("cannot start thread audit_prune_tree");
 709                 prune_thread = NULL;
 710                 return -ENOMEM;
 711         }
 712         return 0;
 713 }
 714
 715 /* called with audit_filter_mutex */
 716 int audit_add_tree_rule(struct audit_krule *rule)
 717 {
 718         struct audit_tree *seed = rule->tree, *tree;
 719         struct path path;
 720         struct vfsmount *mnt;
 721         int err;
 722
 723         rule->tree = NULL;
 724         list_for_each_entry(tree, &tree_list, list) {
 725                 if (!strcmp(seed->pathname, tree->pathname)) {
 726                         put_tree(seed);
 727                         rule->tree = tree;
 728                         list_add(&rule->rlist, &tree->rules);
 729                         return 0;
 730                 }
 731         }
 732         tree = seed;
 733         list_add(&tree->list, &tree_list);
 734         list_add(&rule->rlist, &tree->rules);
 735         /* do not set rule->tree yet */
 736         mutex_unlock(&audit_filter_mutex);
 737
 738         if (unlikely(!prune_thread)) {
 739                 err = audit_launch_prune();
 740                 if (err)
 741                         goto Err;
 742         }
 743
 744         err = kern_path(tree->pathname, 0, &path);
 745         if (err)
 746                 goto Err;
 747         mnt = collect_mounts(&path);
 748         path_put(&path);
 749         if (IS_ERR(mnt)) {
 750                 err = PTR_ERR(mnt);
 751                 goto Err;
 752         }
 753
 754         get_tree(tree);
 755         err = iterate_mounts(tag_mount, tree, mnt);
 756         drop_collected_mounts(mnt);
 757
 758         if (!err) {
 759                 struct node *node;
 760                 spin_lock(&hash_lock);
 761                 list_for_each_entry(node, &tree->chunks, list)
 762                         node->index &= ~(1U<<31);
 763                 spin_unlock(&hash_lock);
 764         } else {
 765                 trim_marked(tree);
 766                 goto Err;
 767         }
 768
 769         mutex_lock(&audit_filter_mutex);
 770         if (list_empty(&rule->rlist)) {
 771                 put_tree(tree);
 772                 return -ENOENT;
 773         }
 774         rule->tree = tree;
 775         put_tree(tree);
 776
 777         return 0;
 778 Err:
 779         mutex_lock(&audit_filter_mutex);
 780         list_del_init(&tree->list);
 781         list_del_init(&tree->rules);
 782         put_tree(tree);
 783         return err;
 784 }
 785
 786 int audit_tag_tree(char *old, char *new)
 787 {
 788         struct list_head cursor, barrier;
 789         int failed = 0;
 790         struct path path1, path2;
 791         struct vfsmount *tagged;
 792         int err;
 793
 794         err = kern_path(new, 0, &path2);
 795         if (err)
 796                 return err;
 797         tagged = collect_mounts(&path2);
 798         path_put(&path2);
 799         if (IS_ERR(tagged))
 800                 return PTR_ERR(tagged);
 801
 802         err = kern_path(old, 0, &path1);
 803         if (err) {
 804                 drop_collected_mounts(tagged);
 805                 return err;
 806         }
 807
 808         mutex_lock(&audit_filter_mutex);
 809         list_add(&barrier, &tree_list);
 810         list_add(&cursor, &barrier);
 811
 812         while (cursor.next != &tree_list) {
 813                 struct audit_tree *tree;
 814                 int good_one = 0;
 815
 816                 tree = container_of(cursor.next, struct audit_tree, list);
 817                 get_tree(tree);
 818                 list_del(&cursor);
 819                 list_add(&cursor, &tree->list);
 820                 mutex_unlock(&audit_filter_mutex);
 821
 822                 err = kern_path(tree->pathname, 0, &path2);
 823                 if (!err) {
 824                         good_one = path_is_under(&path1, &path2);
 825                         path_put(&path2);
 826                 }
 827
 828                 if (!good_one) {
 829                         put_tree(tree);
 830                         mutex_lock(&audit_filter_mutex);
 831                         continue;
 832                 }
 833
 834                 failed = iterate_mounts(tag_mount, tree, tagged);
 835                 if (failed) {
 836                         put_tree(tree);
 837                         mutex_lock(&audit_filter_mutex);
 838                         break;
 839                 }
 840
 841                 mutex_lock(&audit_filter_mutex);
 842                 spin_lock(&hash_lock);
 843                 if (!tree->goner) {
 844                         list_del(&tree->list);
 845                         list_add(&tree->list, &tree_list);
 846                 }
 847                 spin_unlock(&hash_lock);
 848                 put_tree(tree);
 849         }
 850
 851         while (barrier.prev != &tree_list) {
 852                 struct audit_tree *tree;
 853
 854                 tree = container_of(barrier.prev, struct audit_tree, list);
 855                 get_tree(tree);
 856                 list_del(&tree->list);
 857                 list_add(&tree->list, &barrier);
 858                 mutex_unlock(&audit_filter_mutex);
 859
 860                 if (!failed) {
 861                         struct node *node;
 862                         spin_lock(&hash_lock);
 863                         list_for_each_entry(node, &tree->chunks, list)
 864                                 node->index &= ~(1U<<31);
 865                         spin_unlock(&hash_lock);
 866                 } else {
 867                         trim_marked(tree);
 868                 }
 869
 870                 put_tree(tree);
 871                 mutex_lock(&audit_filter_mutex);
 872         }
 873         list_del(&barrier);
 874         list_del(&cursor);
 875         mutex_unlock(&audit_filter_mutex);
 876         path_put(&path1);
 877         drop_collected_mounts(tagged);
 878         return failed;
 879 }
 880
 881
 882 static void audit_schedule_prune(void)
 883 {
 884         wake_up_process(prune_thread);
 885 }
 886
 887 /*
 888  * ... and that one is done if evict_chunk() decides to delay until the end
 889  * of syscall.  Runs synchronously.
 890  */
 891 void audit_kill_trees(struct list_head *list)
 892 {
 893         mutex_lock(&audit_cmd_mutex);
 894         mutex_lock(&audit_filter_mutex);
 895
 896         while (!list_empty(list)) {
 897                 struct audit_tree *victim;
 898
 899                 victim = list_entry(list->next, struct audit_tree, list);
 900                 kill_rules(victim);
 901                 list_del_init(&victim->list);
 902
 903                 mutex_unlock(&audit_filter_mutex);
 904
 905                 prune_one(victim);
 906
 907                 mutex_lock(&audit_filter_mutex);
 908         }
 909
 910         mutex_unlock(&audit_filter_mutex);
 911         mutex_unlock(&audit_cmd_mutex);
 912 }
 913
 914 /*
 915  *  Here comes the stuff asynchronous to auditctl operations
 916  */
 917
 918 static void evict_chunk(struct audit_chunk *chunk)
 919 {
 920         struct audit_tree *owner;
 921         struct list_head *postponed = audit_killed_trees();
 922         int need_prune = 0;
 923         int n;
 924
 925         if (chunk->dead)
 926                 return;
 927
 928         chunk->dead = 1;
 929         mutex_lock(&audit_filter_mutex);
 930         spin_lock(&hash_lock);
 931         while (!list_empty(&chunk->trees)) {
 932                 owner = list_entry(chunk->trees.next,
 933                                    struct audit_tree, same_root);
 934                 owner->goner = 1;
 935                 owner->root = NULL;
 936                 list_del_init(&owner->same_root);
 937                 spin_unlock(&hash_lock);
 938                 if (!postponed) {
 939                         kill_rules(owner);
 940                         list_move(&owner->list, &prune_list);
 941                         need_prune = 1;
 942                 } else {
 943                         list_move(&owner->list, postponed);
 944                 }
 945                 spin_lock(&hash_lock);
 946         }
 947         list_del_rcu(&chunk->hash);
 948         for (n = 0; n < chunk->count; n++)
 949                 list_del_init(&chunk->owners[n].list);
 950         spin_unlock(&hash_lock);
 951         mutex_unlock(&audit_filter_mutex);
 952         if (need_prune)
 953                 audit_schedule_prune();
 954 }
 955
 956 static int audit_tree_handle_event(struct fsnotify_group *group,
 957                                    struct inode *to_tell,
 958                                    struct fsnotify_mark *inode_mark,
 959                                    struct fsnotify_mark *vfsmount_mark,
 960                                    u32 mask, const void *data, int data_type,
 961                                    const unsigned char *file_name, u32 cookie)
 962 {
 963         return 0;
 964 }
 965
 966 static void audit_tree_freeing_mark(struct fsnotify_mark *entry, struct fsnotify_group *group)
 967 {
 968         struct audit_chunk *chunk = container_of(entry, struct audit_chunk, mark);
 969
 970         evict_chunk(chunk);
 971
 972         /*
 973          * We are guaranteed to have at least one reference to the mark from
 974          * either the inode or the caller of fsnotify_destroy_mark().
 975          */
 976         BUG_ON(atomic_read(&entry->refcnt) < 1);
 977 }
 978
 979 static const struct fsnotify_ops audit_tree_ops = {
 980         .handle_event = audit_tree_handle_event,
 981         .freeing_mark = audit_tree_freeing_mark,
 982 };
 983
 984 static int __init audit_tree_init(void)
 985 {
 986         int i;
 987
 988         audit_tree_group = fsnotify_alloc_group(&audit_tree_ops);
 989         if (IS_ERR(audit_tree_group))
 990                 audit_panic("cannot initialize fsnotify group for rectree watches");
 991
 992         for (i = 0; i < HASH_SIZE; i++)
 993                 INIT_LIST_HEAD(&chunk_hash_heads[i]);
 994
 995         return 0;
 996 }
 997 __initcall(audit_tree_init);