git.rot13.org Git - linux/blob - net/netfilter/nf_conncount.c

   1 /*
   2  * count the number of connections matching an arbitrary key.
   3  *
   4  * (C) 2017 Red Hat GmbH
   5  * Author: Florian Westphal <fw@strlen.de>
   6  *
   7  * split from xt_connlimit.c:
   8  *   (c) 2000 Gerd Knorr <kraxel@bytesex.org>
   9  *   Nov 2002: Martin Bene <martin.bene@icomedias.com>:
  10  *              only ignore TIME_WAIT or gone connections
  11  *   (C) CC Computer Consultants GmbH, 2007
  12  */
  13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  14 #include <linux/in.h>
  15 #include <linux/in6.h>
  16 #include <linux/ip.h>
  17 #include <linux/ipv6.h>
  18 #include <linux/jhash.h>
  19 #include <linux/slab.h>
  20 #include <linux/list.h>
  21 #include <linux/rbtree.h>
  22 #include <linux/module.h>
  23 #include <linux/random.h>
  24 #include <linux/skbuff.h>
  25 #include <linux/spinlock.h>
  26 #include <linux/netfilter/nf_conntrack_tcp.h>
  27 #include <linux/netfilter/x_tables.h>
  28 #include <net/netfilter/nf_conntrack.h>
  29 #include <net/netfilter/nf_conntrack_count.h>
  30 #include <net/netfilter/nf_conntrack_core.h>
  31 #include <net/netfilter/nf_conntrack_tuple.h>
  32 #include <net/netfilter/nf_conntrack_zones.h>
  33
  34 #define CONNCOUNT_SLOTS         256U
  35
  36 #define CONNCOUNT_GC_MAX_NODES  8
  37 #define MAX_KEYLEN              5
  38
  39 /* we will save the tuples of all connections we care about */
  40 struct nf_conncount_tuple {
  41         struct list_head                node;
  42         struct nf_conntrack_tuple       tuple;
  43         struct nf_conntrack_zone        zone;
  44         int                             cpu;
  45         u32                             jiffies32;
  46 };
  47
  48 struct nf_conncount_rb {
  49         struct rb_node node;
  50         struct nf_conncount_list list;
  51         u32 key[MAX_KEYLEN];
  52         struct rcu_head rcu_head;
  53 };
  54
  55 static spinlock_t nf_conncount_locks[CONNCOUNT_SLOTS] __cacheline_aligned_in_smp;
  56
  57 struct nf_conncount_data {
  58         unsigned int keylen;
  59         struct rb_root root[CONNCOUNT_SLOTS];
  60         struct net *net;
  61         struct work_struct gc_work;
  62         unsigned long pending_trees[BITS_TO_LONGS(CONNCOUNT_SLOTS)];
  63         unsigned int gc_tree;
  64 };
  65
  66 static u_int32_t conncount_rnd __read_mostly;
  67 static struct kmem_cache *conncount_rb_cachep __read_mostly;
  68 static struct kmem_cache *conncount_conn_cachep __read_mostly;
  69
  70 static inline bool already_closed(const struct nf_conn *conn)
  71 {
  72         if (nf_ct_protonum(conn) == IPPROTO_TCP)
  73                 return conn->proto.tcp.state == TCP_CONNTRACK_TIME_WAIT ||
  74                        conn->proto.tcp.state == TCP_CONNTRACK_CLOSE;
  75         else
  76                 return false;
  77 }
  78
  79 static int key_diff(const u32 *a, const u32 *b, unsigned int klen)
  80 {
  81         return memcmp(a, b, klen * sizeof(u32));
  82 }
  83
  84 static bool conn_free(struct nf_conncount_list *list,
  85                       struct nf_conncount_tuple *conn)
  86 {
  87         bool free_entry = false;
  88
  89         lockdep_assert_held(&list->list_lock);
  90
  91         list->count--;
  92         list_del(&conn->node);
  93         if (list->count == 0) {
  94                 list->dead = true;
  95                 free_entry = true;
  96         }
  97
  98         kmem_cache_free(conncount_conn_cachep, conn);
  99         return free_entry;
 100 }
 101
 102 static const struct nf_conntrack_tuple_hash *
 103 find_or_evict(struct net *net, struct nf_conncount_list *list,
 104               struct nf_conncount_tuple *conn, bool *free_entry)
 105 {
 106         const struct nf_conntrack_tuple_hash *found;
 107         unsigned long a, b;
 108         int cpu = raw_smp_processor_id();
 109         u32 age;
 110
 111         found = nf_conntrack_find_get(net, &conn->zone, &conn->tuple);
 112         if (found)
 113                 return found;
 114         b = conn->jiffies32;
 115         a = (u32)jiffies;
 116
 117         /* conn might have been added just before by another cpu and
 118          * might still be unconfirmed.  In this case, nf_conntrack_find()
 119          * returns no result.  Thus only evict if this cpu added the
 120          * stale entry or if the entry is older than two jiffies.
 121          */
 122         age = a - b;
 123         if (conn->cpu == cpu || age >= 2) {
 124                 *free_entry = conn_free(list, conn);
 125                 return ERR_PTR(-ENOENT);
 126         }
 127
 128         return ERR_PTR(-EAGAIN);
 129 }
 130
 131 static int __nf_conncount_add(struct net *net,
 132                               struct nf_conncount_list *list,
 133                               const struct nf_conntrack_tuple *tuple,
 134                               const struct nf_conntrack_zone *zone)
 135 {
 136         const struct nf_conntrack_tuple_hash *found;
 137         struct nf_conncount_tuple *conn, *conn_n;
 138         struct nf_conn *found_ct;
 139         unsigned int collect = 0;
 140         bool free_entry = false;
 141
 142         /* check the saved connections */
 143         list_for_each_entry_safe(conn, conn_n, &list->head, node) {
 144                 if (collect > CONNCOUNT_GC_MAX_NODES)
 145                         break;
 146
 147                 found = find_or_evict(net, list, conn, &free_entry);
 148                 if (IS_ERR(found)) {
 149                         /* Not found, but might be about to be confirmed */
 150                         if (PTR_ERR(found) == -EAGAIN) {
 151                                 if (nf_ct_tuple_equal(&conn->tuple, tuple) &&
 152                                     nf_ct_zone_id(&conn->zone, conn->zone.dir) ==
 153                                     nf_ct_zone_id(zone, zone->dir))
 154                                         return 0; /* already exists */
 155                         } else {
 156                                 collect++;
 157                         }
 158                         continue;
 159                 }
 160
 161                 found_ct = nf_ct_tuplehash_to_ctrack(found);
 162
 163                 if (nf_ct_tuple_equal(&conn->tuple, tuple) &&
 164                     nf_ct_zone_equal(found_ct, zone, zone->dir)) {
 165                         /*
 166                          * We should not see tuples twice unless someone hooks
 167                          * this into a table without "-p tcp --syn".
 168                          *
 169                          * Attempt to avoid a re-add in this case.
 170                          */
 171                         nf_ct_put(found_ct);
 172                         return 0;
 173                 } else if (already_closed(found_ct)) {
 174                         /*
 175                          * we do not care about connections which are
 176                          * closed already -> ditch it
 177                          */
 178                         nf_ct_put(found_ct);
 179                         conn_free(list, conn);
 180                         collect++;
 181                         continue;
 182                 }
 183
 184                 nf_ct_put(found_ct);
 185         }
 186
 187         if (WARN_ON_ONCE(list->count > INT_MAX))
 188                 return -EOVERFLOW;
 189
 190         conn = kmem_cache_alloc(conncount_conn_cachep, GFP_ATOMIC);
 191         if (conn == NULL)
 192                 return -ENOMEM;
 193
 194         conn->tuple = *tuple;
 195         conn->zone = *zone;
 196         conn->cpu = raw_smp_processor_id();
 197         conn->jiffies32 = (u32)jiffies;
 198         list_add_tail(&conn->node, &list->head);
 199         list->count++;
 200         return 0;
 201 }
 202
 203 int nf_conncount_add(struct net *net,
 204                      struct nf_conncount_list *list,
 205                      const struct nf_conntrack_tuple *tuple,
 206                      const struct nf_conntrack_zone *zone)
 207 {
 208         int ret;
 209
 210         /* check the saved connections */
 211         spin_lock_bh(&list->list_lock);
 212         ret = __nf_conncount_add(net, list, tuple, zone);
 213         spin_unlock_bh(&list->list_lock);
 214
 215         return ret;
 216 }
 217 EXPORT_SYMBOL_GPL(nf_conncount_add);
 218
 219 void nf_conncount_list_init(struct nf_conncount_list *list)
 220 {
 221         spin_lock_init(&list->list_lock);
 222         INIT_LIST_HEAD(&list->head);
 223         list->count = 0;
 224         list->dead = false;
 225 }
 226 EXPORT_SYMBOL_GPL(nf_conncount_list_init);
 227
 228 /* Return true if the list is empty. Must be called with BH disabled. */
 229 bool nf_conncount_gc_list(struct net *net,
 230                           struct nf_conncount_list *list)
 231 {
 232         const struct nf_conntrack_tuple_hash *found;
 233         struct nf_conncount_tuple *conn, *conn_n;
 234         struct nf_conn *found_ct;
 235         unsigned int collected = 0;
 236         bool free_entry = false;
 237         bool ret = false;
 238
 239         /* don't bother if other cpu is already doing GC */
 240         if (!spin_trylock(&list->list_lock))
 241                 return false;
 242
 243         list_for_each_entry_safe(conn, conn_n, &list->head, node) {
 244                 found = find_or_evict(net, list, conn, &free_entry);
 245                 if (IS_ERR(found)) {
 246                         if (PTR_ERR(found) == -ENOENT)  {
 247                                 if (free_entry) {
 248                                         spin_unlock(&list->list_lock);
 249                                         return true;
 250                                 }
 251                                 collected++;
 252                         }
 253                         continue;
 254                 }
 255
 256                 found_ct = nf_ct_tuplehash_to_ctrack(found);
 257                 if (already_closed(found_ct)) {
 258                         /*
 259                          * we do not care about connections which are
 260                          * closed already -> ditch it
 261                          */
 262                         nf_ct_put(found_ct);
 263                         if (conn_free(list, conn)) {
 264                                 spin_unlock(&list->list_lock);
 265                                 return true;
 266                         }
 267                         collected++;
 268                         continue;
 269                 }
 270
 271                 nf_ct_put(found_ct);
 272                 if (collected > CONNCOUNT_GC_MAX_NODES)
 273                         break;
 274         }
 275
 276         if (!list->count) {
 277                 list->dead = true;
 278                 ret = true;
 279         }
 280         spin_unlock(&list->list_lock);
 281
 282         return ret;
 283 }
 284 EXPORT_SYMBOL_GPL(nf_conncount_gc_list);
 285
 286 static void __tree_nodes_free(struct rcu_head *h)
 287 {
 288         struct nf_conncount_rb *rbconn;
 289
 290         rbconn = container_of(h, struct nf_conncount_rb, rcu_head);
 291         kmem_cache_free(conncount_rb_cachep, rbconn);
 292 }
 293
 294 static void tree_nodes_free(struct rb_root *root,
 295                             struct nf_conncount_rb *gc_nodes[],
 296                             unsigned int gc_count)
 297 {
 298         struct nf_conncount_rb *rbconn;
 299
 300         while (gc_count) {
 301                 rbconn = gc_nodes[--gc_count];
 302                 spin_lock(&rbconn->list.list_lock);
 303                 rb_erase(&rbconn->node, root);
 304                 call_rcu(&rbconn->rcu_head, __tree_nodes_free);
 305                 spin_unlock(&rbconn->list.list_lock);
 306         }
 307 }
 308
 309 static void schedule_gc_worker(struct nf_conncount_data *data, int tree)
 310 {
 311         set_bit(tree, data->pending_trees);
 312         schedule_work(&data->gc_work);
 313 }
 314
 315 static unsigned int
 316 insert_tree(struct net *net,
 317             struct nf_conncount_data *data,
 318             struct rb_root *root,
 319             unsigned int hash,
 320             const u32 *key,
 321             u8 keylen,
 322             const struct nf_conntrack_tuple *tuple,
 323             const struct nf_conntrack_zone *zone)
 324 {
 325         struct nf_conncount_rb *gc_nodes[CONNCOUNT_GC_MAX_NODES];
 326         struct rb_node **rbnode, *parent;
 327         struct nf_conncount_rb *rbconn;
 328         struct nf_conncount_tuple *conn;
 329         unsigned int count = 0, gc_count = 0;
 330         bool do_gc = true;
 331
 332         spin_lock_bh(&nf_conncount_locks[hash]);
 333 restart:
 334         parent = NULL;
 335         rbnode = &(root->rb_node);
 336         while (*rbnode) {
 337                 int diff;
 338                 rbconn = rb_entry(*rbnode, struct nf_conncount_rb, node);
 339
 340                 parent = *rbnode;
 341                 diff = key_diff(key, rbconn->key, keylen);
 342                 if (diff < 0) {
 343                         rbnode = &((*rbnode)->rb_left);
 344                 } else if (diff > 0) {
 345                         rbnode = &((*rbnode)->rb_right);
 346                 } else {
 347                         int ret;
 348
 349                         ret = nf_conncount_add(net, &rbconn->list, tuple, zone);
 350                         if (ret)
 351                                 count = 0; /* hotdrop */
 352                         else
 353                                 count = rbconn->list.count;
 354                         tree_nodes_free(root, gc_nodes, gc_count);
 355                         goto out_unlock;
 356                 }
 357
 358                 if (gc_count >= ARRAY_SIZE(gc_nodes))
 359                         continue;
 360
 361                 if (do_gc && nf_conncount_gc_list(net, &rbconn->list))
 362                         gc_nodes[gc_count++] = rbconn;
 363         }
 364
 365         if (gc_count) {
 366                 tree_nodes_free(root, gc_nodes, gc_count);
 367                 schedule_gc_worker(data, hash);
 368                 gc_count = 0;
 369                 do_gc = false;
 370                 goto restart;
 371         }
 372
 373         /* expected case: match, insert new node */
 374         rbconn = kmem_cache_alloc(conncount_rb_cachep, GFP_ATOMIC);
 375         if (rbconn == NULL)
 376                 goto out_unlock;
 377
 378         conn = kmem_cache_alloc(conncount_conn_cachep, GFP_ATOMIC);
 379         if (conn == NULL) {
 380                 kmem_cache_free(conncount_rb_cachep, rbconn);
 381                 goto out_unlock;
 382         }
 383
 384         conn->tuple = *tuple;
 385         conn->zone = *zone;
 386         memcpy(rbconn->key, key, sizeof(u32) * keylen);
 387
 388         nf_conncount_list_init(&rbconn->list);
 389         list_add(&conn->node, &rbconn->list.head);
 390         count = 1;
 391         rbconn->list.count = count;
 392
 393         rb_link_node_rcu(&rbconn->node, parent, rbnode);
 394         rb_insert_color(&rbconn->node, root);
 395 out_unlock:
 396         spin_unlock_bh(&nf_conncount_locks[hash]);
 397         return count;
 398 }
 399
 400 static unsigned int
 401 count_tree(struct net *net,
 402            struct nf_conncount_data *data,
 403            const u32 *key,
 404            const struct nf_conntrack_tuple *tuple,
 405            const struct nf_conntrack_zone *zone)
 406 {
 407         struct rb_root *root;
 408         struct rb_node *parent;
 409         struct nf_conncount_rb *rbconn;
 410         unsigned int hash;
 411         u8 keylen = data->keylen;
 412
 413         hash = jhash2(key, data->keylen, conncount_rnd) % CONNCOUNT_SLOTS;
 414         root = &data->root[hash];
 415
 416         parent = rcu_dereference_raw(root->rb_node);
 417         while (parent) {
 418                 int diff;
 419
 420                 rbconn = rb_entry(parent, struct nf_conncount_rb, node);
 421
 422                 diff = key_diff(key, rbconn->key, keylen);
 423                 if (diff < 0) {
 424                         parent = rcu_dereference_raw(parent->rb_left);
 425                 } else if (diff > 0) {
 426                         parent = rcu_dereference_raw(parent->rb_right);
 427                 } else {
 428                         int ret;
 429
 430                         if (!tuple) {
 431                                 nf_conncount_gc_list(net, &rbconn->list);
 432                                 return rbconn->list.count;
 433                         }
 434
 435                         spin_lock_bh(&rbconn->list.list_lock);
 436                         /* Node might be about to be free'd.
 437                          * We need to defer to insert_tree() in this case.
 438                          */
 439                         if (rbconn->list.count == 0) {
 440                                 spin_unlock_bh(&rbconn->list.list_lock);
 441                                 break;
 442                         }
 443
 444                         /* same source network -> be counted! */
 445                         ret = __nf_conncount_add(net, &rbconn->list, tuple, zone);
 446                         spin_unlock_bh(&rbconn->list.list_lock);
 447                         if (ret)
 448                                 return 0; /* hotdrop */
 449                         else
 450                                 return rbconn->list.count;
 451                 }
 452         }
 453
 454         if (!tuple)
 455                 return 0;
 456
 457         return insert_tree(net, data, root, hash, key, keylen, tuple, zone);
 458 }
 459
 460 static void tree_gc_worker(struct work_struct *work)
 461 {
 462         struct nf_conncount_data *data = container_of(work, struct nf_conncount_data, gc_work);
 463         struct nf_conncount_rb *gc_nodes[CONNCOUNT_GC_MAX_NODES], *rbconn;
 464         struct rb_root *root;
 465         struct rb_node *node;
 466         unsigned int tree, next_tree, gc_count = 0;
 467
 468         tree = data->gc_tree % CONNCOUNT_SLOTS;
 469         root = &data->root[tree];
 470
 471         local_bh_disable();
 472         rcu_read_lock();
 473         for (node = rb_first(root); node != NULL; node = rb_next(node)) {
 474                 rbconn = rb_entry(node, struct nf_conncount_rb, node);
 475                 if (nf_conncount_gc_list(data->net, &rbconn->list))
 476                         gc_count++;
 477         }
 478         rcu_read_unlock();
 479         local_bh_enable();
 480
 481         cond_resched();
 482
 483         spin_lock_bh(&nf_conncount_locks[tree]);
 484         if (gc_count < ARRAY_SIZE(gc_nodes))
 485                 goto next; /* do not bother */
 486
 487         gc_count = 0;
 488         node = rb_first(root);
 489         while (node != NULL) {
 490                 rbconn = rb_entry(node, struct nf_conncount_rb, node);
 491                 node = rb_next(node);
 492
 493                 if (rbconn->list.count > 0)
 494                         continue;
 495
 496                 gc_nodes[gc_count++] = rbconn;
 497                 if (gc_count >= ARRAY_SIZE(gc_nodes)) {
 498                         tree_nodes_free(root, gc_nodes, gc_count);
 499                         gc_count = 0;
 500                 }
 501         }
 502
 503         tree_nodes_free(root, gc_nodes, gc_count);
 504 next:
 505         clear_bit(tree, data->pending_trees);
 506
 507         next_tree = (tree + 1) % CONNCOUNT_SLOTS;
 508         next_tree = find_next_bit(data->pending_trees, next_tree, CONNCOUNT_SLOTS);
 509
 510         if (next_tree < CONNCOUNT_SLOTS) {
 511                 data->gc_tree = next_tree;
 512                 schedule_work(work);
 513         }
 514
 515         spin_unlock_bh(&nf_conncount_locks[tree]);
 516 }
 517
 518 /* Count and return number of conntrack entries in 'net' with particular 'key'.
 519  * If 'tuple' is not null, insert it into the accounting data structure.
 520  * Call with RCU read lock.
 521  */
 522 unsigned int nf_conncount_count(struct net *net,
 523                                 struct nf_conncount_data *data,
 524                                 const u32 *key,
 525                                 const struct nf_conntrack_tuple *tuple,
 526                                 const struct nf_conntrack_zone *zone)
 527 {
 528         return count_tree(net, data, key, tuple, zone);
 529 }
 530 EXPORT_SYMBOL_GPL(nf_conncount_count);
 531
 532 struct nf_conncount_data *nf_conncount_init(struct net *net, unsigned int family,
 533                                             unsigned int keylen)
 534 {
 535         struct nf_conncount_data *data;
 536         int ret, i;
 537
 538         if (keylen % sizeof(u32) ||
 539             keylen / sizeof(u32) > MAX_KEYLEN ||
 540             keylen == 0)
 541                 return ERR_PTR(-EINVAL);
 542
 543         net_get_random_once(&conncount_rnd, sizeof(conncount_rnd));
 544
 545         data = kmalloc(sizeof(*data), GFP_KERNEL);
 546         if (!data)
 547                 return ERR_PTR(-ENOMEM);
 548
 549         ret = nf_ct_netns_get(net, family);
 550         if (ret < 0) {
 551                 kfree(data);
 552                 return ERR_PTR(ret);
 553         }
 554
 555         for (i = 0; i < ARRAY_SIZE(data->root); ++i)
 556                 data->root[i] = RB_ROOT;
 557
 558         data->keylen = keylen / sizeof(u32);
 559         data->net = net;
 560         INIT_WORK(&data->gc_work, tree_gc_worker);
 561
 562         return data;
 563 }
 564 EXPORT_SYMBOL_GPL(nf_conncount_init);
 565
 566 void nf_conncount_cache_free(struct nf_conncount_list *list)
 567 {
 568         struct nf_conncount_tuple *conn, *conn_n;
 569
 570         list_for_each_entry_safe(conn, conn_n, &list->head, node)
 571                 kmem_cache_free(conncount_conn_cachep, conn);
 572 }
 573 EXPORT_SYMBOL_GPL(nf_conncount_cache_free);
 574
 575 static void destroy_tree(struct rb_root *r)
 576 {
 577         struct nf_conncount_rb *rbconn;
 578         struct rb_node *node;
 579
 580         while ((node = rb_first(r)) != NULL) {
 581                 rbconn = rb_entry(node, struct nf_conncount_rb, node);
 582
 583                 rb_erase(node, r);
 584
 585                 nf_conncount_cache_free(&rbconn->list);
 586
 587                 kmem_cache_free(conncount_rb_cachep, rbconn);
 588         }
 589 }
 590
 591 void nf_conncount_destroy(struct net *net, unsigned int family,
 592                           struct nf_conncount_data *data)
 593 {
 594         unsigned int i;
 595
 596         cancel_work_sync(&data->gc_work);
 597         nf_ct_netns_put(net, family);
 598
 599         for (i = 0; i < ARRAY_SIZE(data->root); ++i)
 600                 destroy_tree(&data->root[i]);
 601
 602         kfree(data);
 603 }
 604 EXPORT_SYMBOL_GPL(nf_conncount_destroy);
 605
 606 static int __init nf_conncount_modinit(void)
 607 {
 608         int i;
 609
 610         for (i = 0; i < CONNCOUNT_SLOTS; ++i)
 611                 spin_lock_init(&nf_conncount_locks[i]);
 612
 613         conncount_conn_cachep = kmem_cache_create("nf_conncount_tuple",
 614                                            sizeof(struct nf_conncount_tuple),
 615                                            0, 0, NULL);
 616         if (!conncount_conn_cachep)
 617                 return -ENOMEM;
 618
 619         conncount_rb_cachep = kmem_cache_create("nf_conncount_rb",
 620                                            sizeof(struct nf_conncount_rb),
 621                                            0, 0, NULL);
 622         if (!conncount_rb_cachep) {
 623                 kmem_cache_destroy(conncount_conn_cachep);
 624                 return -ENOMEM;
 625         }
 626
 627         return 0;
 628 }
 629
 630 static void __exit nf_conncount_modexit(void)
 631 {
 632         kmem_cache_destroy(conncount_conn_cachep);
 633         kmem_cache_destroy(conncount_rb_cachep);
 634 }
 635
 636 module_init(nf_conncount_modinit);
 637 module_exit(nf_conncount_modexit);
 638 MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>");
 639 MODULE_AUTHOR("Florian Westphal <fw@strlen.de>");
 640 MODULE_DESCRIPTION("netfilter: count number of connections matching a key");
 641 MODULE_LICENSE("GPL");