net/netfilter/xt_connlimit.c

   1 /*
   2  * netfilter module to limit the number of parallel tcp
   3  * connections per IP address.
   4  *   (c) 2000 Gerd Knorr <kraxel@bytesex.org>
   5  *   Nov 2002: Martin Bene <martin.bene@icomedias.com>:
   6  *              only ignore TIME_WAIT or gone connections
   7  *   (C) CC Computer Consultants GmbH, 2007
   8  *
   9  * based on ...
  10  *
  11  * Kernel module to match connection tracking information.
  12  * GPL (C) 1999  Rusty Russell (rusty@rustcorp.com.au).
  13  */
  14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  15 #include <linux/in.h>
  16 #include <linux/in6.h>
  17 #include <linux/ip.h>
  18 #include <linux/ipv6.h>
  19 #include <linux/jhash.h>
  20 #include <linux/slab.h>
  21 #include <linux/list.h>
  22 #include <linux/rbtree.h>
  23 #include <linux/module.h>
  24 #include <linux/random.h>
  25 #include <linux/skbuff.h>
  26 #include <linux/spinlock.h>
  27 #include <linux/netfilter/nf_conntrack_tcp.h>
  28 #include <linux/netfilter/x_tables.h>
  29 #include <linux/netfilter/xt_connlimit.h>
  30 #include <net/netfilter/nf_conntrack.h>
  31 #include <net/netfilter/nf_conntrack_core.h>
  32 #include <net/netfilter/nf_conntrack_tuple.h>
  33 #include <net/netfilter/nf_conntrack_zones.h>
  34
  35 #define CONNLIMIT_SLOTS         256U
  36
  37 #ifdef CONFIG_LOCKDEP
  38 #define CONNLIMIT_LOCK_SLOTS    8U
  39 #else
  40 #define CONNLIMIT_LOCK_SLOTS    256U
  41 #endif
  42
  43 #define CONNLIMIT_GC_MAX_NODES  8
  44
  45 /* we will save the tuples of all connections we care about */
  46 struct xt_connlimit_conn {
  47         struct hlist_node               node;
  48         struct nf_conntrack_tuple       tuple;
  49         struct nf_conntrack_zone        zone;
  50         int                             cpu;
  51         u32                             jiffies32;
  52 };
  53
  54 struct xt_connlimit_rb {
  55         struct rb_node node;
  56         struct hlist_head hhead; /* connections/hosts in same subnet */
  57         union nf_inet_addr addr; /* search key */
  58 };
  59
  60 static spinlock_t xt_connlimit_locks[CONNLIMIT_LOCK_SLOTS] __cacheline_aligned_in_smp;
  61
  62 struct xt_connlimit_data {
  63         struct rb_root climit_root[CONNLIMIT_SLOTS];
  64 };
  65
  66 static u_int32_t connlimit_rnd __read_mostly;
  67 static struct kmem_cache *connlimit_rb_cachep __read_mostly;
  68 static struct kmem_cache *connlimit_conn_cachep __read_mostly;
  69
  70 static inline unsigned int connlimit_iphash(__be32 addr)
  71 {
  72         return jhash_1word((__force __u32)addr,
  73                             connlimit_rnd) % CONNLIMIT_SLOTS;
  74 }
  75
  76 static inline unsigned int
  77 connlimit_iphash6(const union nf_inet_addr *addr,
  78                   const union nf_inet_addr *mask)
  79 {
  80         union nf_inet_addr res;
  81         unsigned int i;
  82
  83         for (i = 0; i < ARRAY_SIZE(addr->ip6); ++i)
  84                 res.ip6[i] = addr->ip6[i] & mask->ip6[i];
  85
  86         return jhash2((u32 *)res.ip6, ARRAY_SIZE(res.ip6),
  87                        connlimit_rnd) % CONNLIMIT_SLOTS;
  88 }
  89
  90 static inline bool already_closed(const struct nf_conn *conn)
  91 {
  92         if (nf_ct_protonum(conn) == IPPROTO_TCP)
  93                 return conn->proto.tcp.state == TCP_CONNTRACK_TIME_WAIT ||
  94                        conn->proto.tcp.state == TCP_CONNTRACK_CLOSE;
  95         else
  96                 return 0;
  97 }
  98
  99 static int
 100 same_source_net(const union nf_inet_addr *addr,
 101                 const union nf_inet_addr *mask,
 102                 const union nf_inet_addr *u3, u_int8_t family)
 103 {
 104         if (family == NFPROTO_IPV4) {
 105                 return ntohl(addr->ip & mask->ip) -
 106                        ntohl(u3->ip & mask->ip);
 107         } else {
 108                 union nf_inet_addr lh, rh;
 109                 unsigned int i;
 110
 111                 for (i = 0; i < ARRAY_SIZE(addr->ip6); ++i) {
 112                         lh.ip6[i] = addr->ip6[i] & mask->ip6[i];
 113                         rh.ip6[i] = u3->ip6[i] & mask->ip6[i];
 114                 }
 115
 116                 return memcmp(&lh.ip6, &rh.ip6, sizeof(lh.ip6));
 117         }
 118 }
 119
 120 bool nf_conncount_add(struct hlist_head *head,
 121                       const struct nf_conntrack_tuple *tuple,
 122                       const struct nf_conntrack_zone *zone)
 123 {
 124         struct xt_connlimit_conn *conn;
 125
 126         conn = kmem_cache_alloc(connlimit_conn_cachep, GFP_ATOMIC);
 127         if (conn == NULL)
 128                 return false;
 129         conn->tuple = *tuple;
 130         conn->zone = *zone;
 131         conn->cpu = raw_smp_processor_id();
 132         conn->jiffies32 = (u32)jiffies;
 133         hlist_add_head(&conn->node, head);
 134         return true;
 135 }
 136 EXPORT_SYMBOL_GPL(nf_conncount_add);
 137
 138 static const struct nf_conntrack_tuple_hash *
 139 find_or_evict(struct net *net, struct xt_connlimit_conn *conn)
 140 {
 141         const struct nf_conntrack_tuple_hash *found;
 142         unsigned long a, b;
 143         int cpu = raw_smp_processor_id();
 144         u32 age;
 145
 146         found = nf_conntrack_find_get(net, &conn->zone, &conn->tuple);
 147         if (found)
 148                 return found;
 149         b = conn->jiffies32;
 150         a = (u32)jiffies;
 151
 152         /* conn might have been added just before by another cpu and
 153          * might still be unconfirmed.  In this case, nf_conntrack_find()
 154          * returns no result.  Thus only evict if this cpu added the
 155          * stale entry or if the entry is older than two jiffies.
 156          */
 157         age = a - b;
 158         if (conn->cpu == cpu || age >= 2) {
 159                 hlist_del(&conn->node);
 160                 kmem_cache_free(connlimit_conn_cachep, conn);
 161                 return ERR_PTR(-ENOENT);
 162         }
 163
 164         return ERR_PTR(-EAGAIN);
 165 }
 166
 167 unsigned int nf_conncount_lookup(struct net *net, struct hlist_head *head,
 168                                  const struct nf_conntrack_tuple *tuple,
 169                                  const struct nf_conntrack_zone *zone,
 170                                  bool *addit)
 171 {
 172         const struct nf_conntrack_tuple_hash *found;
 173         struct xt_connlimit_conn *conn;
 174         struct nf_conn *found_ct;
 175         struct hlist_node *n;
 176         unsigned int length = 0;
 177
 178         *addit = true;
 179
 180         /* check the saved connections */
 181         hlist_for_each_entry_safe(conn, n, head, node) {
 182                 found = find_or_evict(net, conn);
 183                 if (IS_ERR(found)) {
 184                         /* Not found, but might be about to be confirmed */
 185                         if (PTR_ERR(found) == -EAGAIN) {
 186                                 length++;
 187                                 if (!tuple)
 188                                         continue;
 189
 190                                 if (nf_ct_tuple_equal(&conn->tuple, tuple) &&
 191                                     nf_ct_zone_id(&conn->zone, conn->zone.dir) ==
 192                                     nf_ct_zone_id(zone, zone->dir))
 193                                         *addit = false;
 194                         }
 195                         continue;
 196                 }
 197
 198                 found_ct = nf_ct_tuplehash_to_ctrack(found);
 199
 200                 if (nf_ct_tuple_equal(&conn->tuple, tuple) &&
 201                     nf_ct_zone_equal(found_ct, zone, zone->dir)) {
 202                         /*
 203                          * Just to be sure we have it only once in the list.
 204                          * We should not see tuples twice unless someone hooks
 205                          * this into a table without "-p tcp --syn".
 206                          */
 207                         *addit = false;
 208                 } else if (already_closed(found_ct)) {
 209                         /*
 210                          * we do not care about connections which are
 211                          * closed already -> ditch it
 212                          */
 213                         nf_ct_put(found_ct);
 214                         hlist_del(&conn->node);
 215                         kmem_cache_free(connlimit_conn_cachep, conn);
 216                         continue;
 217                 }
 218
 219                 nf_ct_put(found_ct);
 220                 length++;
 221         }
 222
 223         return length;
 224 }
 225 EXPORT_SYMBOL_GPL(nf_conncount_lookup);
 226
 227 static void tree_nodes_free(struct rb_root *root,
 228                             struct xt_connlimit_rb *gc_nodes[],
 229                             unsigned int gc_count)
 230 {
 231         struct xt_connlimit_rb *rbconn;
 232
 233         while (gc_count) {
 234                 rbconn = gc_nodes[--gc_count];
 235                 rb_erase(&rbconn->node, root);
 236                 kmem_cache_free(connlimit_rb_cachep, rbconn);
 237         }
 238 }
 239
 240 static unsigned int
 241 count_tree(struct net *net, struct rb_root *root,
 242            const struct nf_conntrack_tuple *tuple,
 243            const union nf_inet_addr *addr, const union nf_inet_addr *mask,
 244            u8 family, const struct nf_conntrack_zone *zone)
 245 {
 246         struct xt_connlimit_rb *gc_nodes[CONNLIMIT_GC_MAX_NODES];
 247         struct rb_node **rbnode, *parent;
 248         struct xt_connlimit_rb *rbconn;
 249         struct xt_connlimit_conn *conn;
 250         unsigned int gc_count;
 251         bool no_gc = false;
 252
 253  restart:
 254         gc_count = 0;
 255         parent = NULL;
 256         rbnode = &(root->rb_node);
 257         while (*rbnode) {
 258                 int diff;
 259                 bool addit;
 260
 261                 rbconn = rb_entry(*rbnode, struct xt_connlimit_rb, node);
 262
 263                 parent = *rbnode;
 264                 diff = same_source_net(addr, mask, &rbconn->addr, family);
 265                 if (diff < 0) {
 266                         rbnode = &((*rbnode)->rb_left);
 267                 } else if (diff > 0) {
 268                         rbnode = &((*rbnode)->rb_right);
 269                 } else {
 270                         /* same source network -> be counted! */
 271                         unsigned int count;
 272
 273                         count = nf_conncount_lookup(net, &rbconn->hhead, tuple,
 274                                                     zone, &addit);
 275
 276                         tree_nodes_free(root, gc_nodes, gc_count);
 277                         if (!addit)
 278                                 return count;
 279
 280                         if (!nf_conncount_add(&rbconn->hhead, tuple, zone))
 281                                 return 0; /* hotdrop */
 282
 283                         return count + 1;
 284                 }
 285
 286                 if (no_gc || gc_count >= ARRAY_SIZE(gc_nodes))
 287                         continue;
 288
 289                 /* only used for GC on hhead, retval and 'addit' ignored */
 290                 nf_conncount_lookup(net, &rbconn->hhead, tuple, zone, &addit);
 291                 if (hlist_empty(&rbconn->hhead))
 292                         gc_nodes[gc_count++] = rbconn;
 293         }
 294
 295         if (gc_count) {
 296                 no_gc = true;
 297                 tree_nodes_free(root, gc_nodes, gc_count);
 298                 /* tree_node_free before new allocation permits
 299                  * allocator to re-use newly free'd object.
 300                  *
 301                  * This is a rare event; in most cases we will find
 302                  * existing node to re-use. (or gc_count is 0).
 303                  */
 304                 goto restart;
 305         }
 306
 307         /* no match, need to insert new node */
 308         rbconn = kmem_cache_alloc(connlimit_rb_cachep, GFP_ATOMIC);
 309         if (rbconn == NULL)
 310                 return 0;
 311
 312         conn = kmem_cache_alloc(connlimit_conn_cachep, GFP_ATOMIC);
 313         if (conn == NULL) {
 314                 kmem_cache_free(connlimit_rb_cachep, rbconn);
 315                 return 0;
 316         }
 317
 318         conn->tuple = *tuple;
 319         conn->zone = *zone;
 320         rbconn->addr = *addr;
 321
 322         INIT_HLIST_HEAD(&rbconn->hhead);
 323         hlist_add_head(&conn->node, &rbconn->hhead);
 324
 325         rb_link_node(&rbconn->node, parent, rbnode);
 326         rb_insert_color(&rbconn->node, root);
 327         return 1;
 328 }
 329
 330 static int count_them(struct net *net,
 331                       struct xt_connlimit_data *data,
 332                       const struct nf_conntrack_tuple *tuple,
 333                       const union nf_inet_addr *addr,
 334                       const union nf_inet_addr *mask,
 335                       u_int8_t family,
 336                       const struct nf_conntrack_zone *zone)
 337 {
 338         struct rb_root *root;
 339         int count;
 340         u32 hash;
 341
 342         if (family == NFPROTO_IPV6)
 343                 hash = connlimit_iphash6(addr, mask);
 344         else
 345                 hash = connlimit_iphash(addr->ip & mask->ip);
 346         root = &data->climit_root[hash];
 347
 348         spin_lock_bh(&xt_connlimit_locks[hash % CONNLIMIT_LOCK_SLOTS]);
 349
 350         count = count_tree(net, root, tuple, addr, mask, family, zone);
 351
 352         spin_unlock_bh(&xt_connlimit_locks[hash % CONNLIMIT_LOCK_SLOTS]);
 353
 354         return count;
 355 }
 356
 357 static bool
 358 connlimit_mt(const struct sk_buff *skb, struct xt_action_param *par)
 359 {
 360         struct net *net = xt_net(par);
 361         const struct xt_connlimit_info *info = par->matchinfo;
 362         union nf_inet_addr addr;
 363         struct nf_conntrack_tuple tuple;
 364         const struct nf_conntrack_tuple *tuple_ptr = &tuple;
 365         const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
 366         enum ip_conntrack_info ctinfo;
 367         const struct nf_conn *ct;
 368         unsigned int connections;
 369
 370         ct = nf_ct_get(skb, &ctinfo);
 371         if (ct != NULL) {
 372                 tuple_ptr = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 373                 zone = nf_ct_zone(ct);
 374         } else if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb),
 375                                       xt_family(par), net, &tuple)) {
 376                 goto hotdrop;
 377         }
 378
 379         if (xt_family(par) == NFPROTO_IPV6) {
 380                 const struct ipv6hdr *iph = ipv6_hdr(skb);
 381                 memcpy(&addr.ip6, (info->flags & XT_CONNLIMIT_DADDR) ?
 382                        &iph->daddr : &iph->saddr, sizeof(addr.ip6));
 383         } else {
 384                 const struct iphdr *iph = ip_hdr(skb);
 385                 addr.ip = (info->flags & XT_CONNLIMIT_DADDR) ?
 386                           iph->daddr : iph->saddr;
 387         }
 388
 389         connections = count_them(net, info->data, tuple_ptr, &addr,
 390                                  &info->mask, xt_family(par), zone);
 391         if (connections == 0)
 392                 /* kmalloc failed, drop it entirely */
 393                 goto hotdrop;
 394
 395         return (connections > info->limit) ^
 396                !!(info->flags & XT_CONNLIMIT_INVERT);
 397
 398  hotdrop:
 399         par->hotdrop = true;
 400         return false;
 401 }
 402
 403 static int connlimit_mt_check(const struct xt_mtchk_param *par)
 404 {
 405         struct xt_connlimit_info *info = par->matchinfo;
 406         unsigned int i;
 407         int ret;
 408
 409         net_get_random_once(&connlimit_rnd, sizeof(connlimit_rnd));
 410
 411         ret = nf_ct_netns_get(par->net, par->family);
 412         if (ret < 0) {
 413                 pr_info("cannot load conntrack support for "
 414                         "address family %u\n", par->family);
 415                 return ret;
 416         }
 417
 418         /* init private data */
 419         info->data = kmalloc(sizeof(struct xt_connlimit_data), GFP_KERNEL);
 420         if (info->data == NULL) {
 421                 nf_ct_netns_put(par->net, par->family);
 422                 return -ENOMEM;
 423         }
 424
 425         for (i = 0; i < ARRAY_SIZE(info->data->climit_root); ++i)
 426                 info->data->climit_root[i] = RB_ROOT;
 427
 428         return 0;
 429 }
 430
 431 void nf_conncount_cache_free(struct hlist_head *hhead)
 432 {
 433         struct xt_connlimit_conn *conn;
 434         struct hlist_node *n;
 435
 436         hlist_for_each_entry_safe(conn, n, hhead, node)
 437                 kmem_cache_free(connlimit_conn_cachep, conn);
 438 }
 439 EXPORT_SYMBOL_GPL(nf_conncount_cache_free);
 440
 441 static void destroy_tree(struct rb_root *r)
 442 {
 443         struct xt_connlimit_rb *rbconn;
 444         struct rb_node *node;
 445
 446         while ((node = rb_first(r)) != NULL) {
 447                 rbconn = rb_entry(node, struct xt_connlimit_rb, node);
 448
 449                 rb_erase(node, r);
 450
 451                 nf_conncount_cache_free(&rbconn->hhead);
 452
 453                 kmem_cache_free(connlimit_rb_cachep, rbconn);
 454         }
 455 }
 456
 457 static void connlimit_mt_destroy(const struct xt_mtdtor_param *par)
 458 {
 459         const struct xt_connlimit_info *info = par->matchinfo;
 460         unsigned int i;
 461
 462         nf_ct_netns_put(par->net, par->family);
 463
 464         for (i = 0; i < ARRAY_SIZE(info->data->climit_root); ++i)
 465                 destroy_tree(&info->data->climit_root[i]);
 466
 467         kfree(info->data);
 468 }
 469
 470 static struct xt_match connlimit_mt_reg __read_mostly = {
 471         .name       = "connlimit",
 472         .revision   = 1,
 473         .family     = NFPROTO_UNSPEC,
 474         .checkentry = connlimit_mt_check,
 475         .match      = connlimit_mt,
 476         .matchsize  = sizeof(struct xt_connlimit_info),
 477         .usersize   = offsetof(struct xt_connlimit_info, data),
 478         .destroy    = connlimit_mt_destroy,
 479         .me         = THIS_MODULE,
 480 };
 481
 482 static int __init connlimit_mt_init(void)
 483 {
 484         int ret, i;
 485
 486         BUILD_BUG_ON(CONNLIMIT_LOCK_SLOTS > CONNLIMIT_SLOTS);
 487         BUILD_BUG_ON((CONNLIMIT_SLOTS % CONNLIMIT_LOCK_SLOTS) != 0);
 488
 489         for (i = 0; i < CONNLIMIT_LOCK_SLOTS; ++i)
 490                 spin_lock_init(&xt_connlimit_locks[i]);
 491
 492         connlimit_conn_cachep = kmem_cache_create("xt_connlimit_conn",
 493                                            sizeof(struct xt_connlimit_conn),
 494                                            0, 0, NULL);
 495         if (!connlimit_conn_cachep)
 496                 return -ENOMEM;
 497
 498         connlimit_rb_cachep = kmem_cache_create("xt_connlimit_rb",
 499                                            sizeof(struct xt_connlimit_rb),
 500                                            0, 0, NULL);
 501         if (!connlimit_rb_cachep) {
 502                 kmem_cache_destroy(connlimit_conn_cachep);
 503                 return -ENOMEM;
 504         }
 505         ret = xt_register_match(&connlimit_mt_reg);
 506         if (ret != 0) {
 507                 kmem_cache_destroy(connlimit_conn_cachep);
 508                 kmem_cache_destroy(connlimit_rb_cachep);
 509         }
 510         return ret;
 511 }
 512
 513 static void __exit connlimit_mt_exit(void)
 514 {
 515         xt_unregister_match(&connlimit_mt_reg);
 516         kmem_cache_destroy(connlimit_conn_cachep);
 517         kmem_cache_destroy(connlimit_rb_cachep);
 518 }
 519
 520 module_init(connlimit_mt_init);
 521 module_exit(connlimit_mt_exit);
 522 MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>");
 523 MODULE_DESCRIPTION("Xtables: Number of connections matching");
 524 MODULE_LICENSE("GPL");
 525 MODULE_ALIAS("ipt_connlimit");
 526 MODULE_ALIAS("ip6t_connlimit");