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 };
  50
  51 struct xt_connlimit_rb {
  52         struct rb_node node;
  53         struct hlist_head hhead; /* connections/hosts in same subnet */
  54         union nf_inet_addr addr; /* search key */
  55 };
  56
  57 static spinlock_t xt_connlimit_locks[CONNLIMIT_LOCK_SLOTS] __cacheline_aligned_in_smp;
  58
  59 struct xt_connlimit_data {
  60         struct rb_root climit_root[CONNLIMIT_SLOTS];
  61 };
  62
  63 static u_int32_t connlimit_rnd __read_mostly;
  64 static struct kmem_cache *connlimit_rb_cachep __read_mostly;
  65 static struct kmem_cache *connlimit_conn_cachep __read_mostly;
  66
  67 static inline unsigned int connlimit_iphash(__be32 addr)
  68 {
  69         return jhash_1word((__force __u32)addr,
  70                             connlimit_rnd) % CONNLIMIT_SLOTS;
  71 }
  72
  73 static inline unsigned int
  74 connlimit_iphash6(const union nf_inet_addr *addr)
  75 {
  76         return jhash2((u32 *)addr->ip6, ARRAY_SIZE(addr->ip6),
  77                        connlimit_rnd) % CONNLIMIT_SLOTS;
  78 }
  79
  80 static inline bool already_closed(const struct nf_conn *conn)
  81 {
  82         if (nf_ct_protonum(conn) == IPPROTO_TCP)
  83                 return conn->proto.tcp.state == TCP_CONNTRACK_TIME_WAIT ||
  84                        conn->proto.tcp.state == TCP_CONNTRACK_CLOSE;
  85         else
  86                 return 0;
  87 }
  88
  89 static int
  90 same_source(const union nf_inet_addr *addr,
  91             const union nf_inet_addr *u3, u_int8_t family)
  92 {
  93         if (family == NFPROTO_IPV4)
  94                 return ntohl(addr->ip) - ntohl(u3->ip);
  95
  96         return memcmp(addr->ip6, u3->ip6, sizeof(addr->ip6));
  97 }
  98
  99 bool nf_conncount_add(struct hlist_head *head,
 100                       const struct nf_conntrack_tuple *tuple)
 101 {
 102         struct xt_connlimit_conn *conn;
 103
 104         conn = kmem_cache_alloc(connlimit_conn_cachep, GFP_ATOMIC);
 105         if (conn == NULL)
 106                 return false;
 107         conn->tuple = *tuple;
 108         hlist_add_head(&conn->node, head);
 109         return true;
 110 }
 111 EXPORT_SYMBOL_GPL(nf_conncount_add);
 112
 113 unsigned int nf_conncount_lookup(struct net *net, struct hlist_head *head,
 114                                  const struct nf_conntrack_tuple *tuple,
 115                                  const struct nf_conntrack_zone *zone,
 116                                  bool *addit)
 117 {
 118         const struct nf_conntrack_tuple_hash *found;
 119         struct xt_connlimit_conn *conn;
 120         struct hlist_node *n;
 121         struct nf_conn *found_ct;
 122         unsigned int length = 0;
 123
 124         *addit = true;
 125
 126         /* check the saved connections */
 127         hlist_for_each_entry_safe(conn, n, head, node) {
 128                 found = nf_conntrack_find_get(net, zone, &conn->tuple);
 129                 if (found == NULL) {
 130                         hlist_del(&conn->node);
 131                         kmem_cache_free(connlimit_conn_cachep, conn);
 132                         continue;
 133                 }
 134
 135                 found_ct = nf_ct_tuplehash_to_ctrack(found);
 136
 137                 if (nf_ct_tuple_equal(&conn->tuple, tuple)) {
 138                         /*
 139                          * Just to be sure we have it only once in the list.
 140                          * We should not see tuples twice unless someone hooks
 141                          * this into a table without "-p tcp --syn".
 142                          */
 143                         *addit = false;
 144                 } else if (already_closed(found_ct)) {
 145                         /*
 146                          * we do not care about connections which are
 147                          * closed already -> ditch it
 148                          */
 149                         nf_ct_put(found_ct);
 150                         hlist_del(&conn->node);
 151                         kmem_cache_free(connlimit_conn_cachep, conn);
 152                         continue;
 153                 }
 154
 155                 nf_ct_put(found_ct);
 156                 length++;
 157         }
 158
 159         return length;
 160 }
 161 EXPORT_SYMBOL_GPL(nf_conncount_lookup);
 162
 163 static void tree_nodes_free(struct rb_root *root,
 164                             struct xt_connlimit_rb *gc_nodes[],
 165                             unsigned int gc_count)
 166 {
 167         struct xt_connlimit_rb *rbconn;
 168
 169         while (gc_count) {
 170                 rbconn = gc_nodes[--gc_count];
 171                 rb_erase(&rbconn->node, root);
 172                 kmem_cache_free(connlimit_rb_cachep, rbconn);
 173         }
 174 }
 175
 176 static unsigned int
 177 count_tree(struct net *net, struct rb_root *root,
 178            const struct nf_conntrack_tuple *tuple,
 179            const union nf_inet_addr *addr,
 180            u8 family, const struct nf_conntrack_zone *zone)
 181 {
 182         struct xt_connlimit_rb *gc_nodes[CONNLIMIT_GC_MAX_NODES];
 183         struct rb_node **rbnode, *parent;
 184         struct xt_connlimit_rb *rbconn;
 185         struct xt_connlimit_conn *conn;
 186         unsigned int gc_count;
 187         bool no_gc = false;
 188
 189  restart:
 190         gc_count = 0;
 191         parent = NULL;
 192         rbnode = &(root->rb_node);
 193         while (*rbnode) {
 194                 int diff;
 195                 bool addit;
 196
 197                 rbconn = rb_entry(*rbnode, struct xt_connlimit_rb, node);
 198
 199                 parent = *rbnode;
 200                 diff = same_source(addr, &rbconn->addr, family);
 201                 if (diff < 0) {
 202                         rbnode = &((*rbnode)->rb_left);
 203                 } else if (diff > 0) {
 204                         rbnode = &((*rbnode)->rb_right);
 205                 } else {
 206                         /* same source network -> be counted! */
 207                         unsigned int count;
 208
 209                         count = nf_conncount_lookup(net, &rbconn->hhead, tuple,
 210                                                     zone, &addit);
 211
 212                         tree_nodes_free(root, gc_nodes, gc_count);
 213                         if (!addit)
 214                                 return count;
 215
 216                         if (!nf_conncount_add(&rbconn->hhead, tuple))
 217                                 return 0; /* hotdrop */
 218
 219                         return count + 1;
 220                 }
 221
 222                 if (no_gc || gc_count >= ARRAY_SIZE(gc_nodes))
 223                         continue;
 224
 225                 /* only used for GC on hhead, retval and 'addit' ignored */
 226                 nf_conncount_lookup(net, &rbconn->hhead, tuple, zone, &addit);
 227                 if (hlist_empty(&rbconn->hhead))
 228                         gc_nodes[gc_count++] = rbconn;
 229         }
 230
 231         if (gc_count) {
 232                 no_gc = true;
 233                 tree_nodes_free(root, gc_nodes, gc_count);
 234                 /* tree_node_free before new allocation permits
 235                  * allocator to re-use newly free'd object.
 236                  *
 237                  * This is a rare event; in most cases we will find
 238                  * existing node to re-use. (or gc_count is 0).
 239                  */
 240                 goto restart;
 241         }
 242
 243         /* no match, need to insert new node */
 244         rbconn = kmem_cache_alloc(connlimit_rb_cachep, GFP_ATOMIC);
 245         if (rbconn == NULL)
 246                 return 0;
 247
 248         conn = kmem_cache_alloc(connlimit_conn_cachep, GFP_ATOMIC);
 249         if (conn == NULL) {
 250                 kmem_cache_free(connlimit_rb_cachep, rbconn);
 251                 return 0;
 252         }
 253
 254         conn->tuple = *tuple;
 255         rbconn->addr = *addr;
 256
 257         INIT_HLIST_HEAD(&rbconn->hhead);
 258         hlist_add_head(&conn->node, &rbconn->hhead);
 259
 260         rb_link_node(&rbconn->node, parent, rbnode);
 261         rb_insert_color(&rbconn->node, root);
 262         return 1;
 263 }
 264
 265 static int count_them(struct net *net,
 266                       struct xt_connlimit_data *data,
 267                       const struct nf_conntrack_tuple *tuple,
 268                       const union nf_inet_addr *addr,
 269                       u_int8_t family,
 270                       const struct nf_conntrack_zone *zone)
 271 {
 272         struct rb_root *root;
 273         int count;
 274         u32 hash;
 275
 276         if (family == NFPROTO_IPV6)
 277                 hash = connlimit_iphash6(addr);
 278         else
 279                 hash = connlimit_iphash(addr->ip);
 280         root = &data->climit_root[hash];
 281
 282         spin_lock_bh(&xt_connlimit_locks[hash % CONNLIMIT_LOCK_SLOTS]);
 283
 284         count = count_tree(net, root, tuple, addr, family, zone);
 285
 286         spin_unlock_bh(&xt_connlimit_locks[hash % CONNLIMIT_LOCK_SLOTS]);
 287
 288         return count;
 289 }
 290
 291 static bool
 292 connlimit_mt(const struct sk_buff *skb, struct xt_action_param *par)
 293 {
 294         struct net *net = xt_net(par);
 295         const struct xt_connlimit_info *info = par->matchinfo;
 296         union nf_inet_addr addr;
 297         struct nf_conntrack_tuple tuple;
 298         const struct nf_conntrack_tuple *tuple_ptr = &tuple;
 299         const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
 300         enum ip_conntrack_info ctinfo;
 301         const struct nf_conn *ct;
 302         unsigned int connections;
 303
 304         ct = nf_ct_get(skb, &ctinfo);
 305         if (ct != NULL) {
 306                 tuple_ptr = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 307                 zone = nf_ct_zone(ct);
 308         } else if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb),
 309                                       xt_family(par), net, &tuple)) {
 310                 goto hotdrop;
 311         }
 312
 313         if (xt_family(par) == NFPROTO_IPV6) {
 314                 const struct ipv6hdr *iph = ipv6_hdr(skb);
 315                 unsigned int i;
 316
 317                 memcpy(&addr.ip6, (info->flags & XT_CONNLIMIT_DADDR) ?
 318                        &iph->daddr : &iph->saddr, sizeof(addr.ip6));
 319
 320                 for (i = 0; i < ARRAY_SIZE(addr.ip6); ++i)
 321                         addr.ip6[i] &= info->mask.ip6[i];
 322         } else {
 323                 const struct iphdr *iph = ip_hdr(skb);
 324                 addr.ip = (info->flags & XT_CONNLIMIT_DADDR) ?
 325                           iph->daddr : iph->saddr;
 326
 327                 addr.ip &= info->mask.ip;
 328         }
 329
 330         connections = count_them(net, info->data, tuple_ptr, &addr,
 331                                  xt_family(par), zone);
 332         if (connections == 0)
 333                 /* kmalloc failed, drop it entirely */
 334                 goto hotdrop;
 335
 336         return (connections > info->limit) ^
 337                !!(info->flags & XT_CONNLIMIT_INVERT);
 338
 339  hotdrop:
 340         par->hotdrop = true;
 341         return false;
 342 }
 343
 344 static int connlimit_mt_check(const struct xt_mtchk_param *par)
 345 {
 346         struct xt_connlimit_info *info = par->matchinfo;
 347         unsigned int i;
 348         int ret;
 349
 350         net_get_random_once(&connlimit_rnd, sizeof(connlimit_rnd));
 351
 352         ret = nf_ct_netns_get(par->net, par->family);
 353         if (ret < 0) {
 354                 pr_info("cannot load conntrack support for "
 355                         "address family %u\n", par->family);
 356                 return ret;
 357         }
 358
 359         /* init private data */
 360         info->data = kmalloc(sizeof(struct xt_connlimit_data), GFP_KERNEL);
 361         if (info->data == NULL) {
 362                 nf_ct_netns_put(par->net, par->family);
 363                 return -ENOMEM;
 364         }
 365
 366         for (i = 0; i < ARRAY_SIZE(info->data->climit_root); ++i)
 367                 info->data->climit_root[i] = RB_ROOT;
 368
 369         return 0;
 370 }
 371
 372 void nf_conncount_cache_free(struct hlist_head *hhead)
 373 {
 374         struct xt_connlimit_conn *conn;
 375         struct hlist_node *n;
 376
 377         hlist_for_each_entry_safe(conn, n, hhead, node)
 378                 kmem_cache_free(connlimit_conn_cachep, conn);
 379 }
 380 EXPORT_SYMBOL_GPL(nf_conncount_cache_free);
 381
 382 static void destroy_tree(struct rb_root *r)
 383 {
 384         struct xt_connlimit_rb *rbconn;
 385         struct rb_node *node;
 386
 387         while ((node = rb_first(r)) != NULL) {
 388                 rbconn = rb_entry(node, struct xt_connlimit_rb, node);
 389
 390                 rb_erase(node, r);
 391
 392                 nf_conncount_cache_free(&rbconn->hhead);
 393
 394                 kmem_cache_free(connlimit_rb_cachep, rbconn);
 395         }
 396 }
 397
 398 static void connlimit_mt_destroy(const struct xt_mtdtor_param *par)
 399 {
 400         const struct xt_connlimit_info *info = par->matchinfo;
 401         unsigned int i;
 402
 403         nf_ct_netns_put(par->net, par->family);
 404
 405         for (i = 0; i < ARRAY_SIZE(info->data->climit_root); ++i)
 406                 destroy_tree(&info->data->climit_root[i]);
 407
 408         kfree(info->data);
 409 }
 410
 411 static struct xt_match connlimit_mt_reg __read_mostly = {
 412         .name       = "connlimit",
 413         .revision   = 1,
 414         .family     = NFPROTO_UNSPEC,
 415         .checkentry = connlimit_mt_check,
 416         .match      = connlimit_mt,
 417         .matchsize  = sizeof(struct xt_connlimit_info),
 418         .usersize   = offsetof(struct xt_connlimit_info, data),
 419         .destroy    = connlimit_mt_destroy,
 420         .me         = THIS_MODULE,
 421 };
 422
 423 static int __init connlimit_mt_init(void)
 424 {
 425         int ret, i;
 426
 427         BUILD_BUG_ON(CONNLIMIT_LOCK_SLOTS > CONNLIMIT_SLOTS);
 428         BUILD_BUG_ON((CONNLIMIT_SLOTS % CONNLIMIT_LOCK_SLOTS) != 0);
 429
 430         for (i = 0; i < CONNLIMIT_LOCK_SLOTS; ++i)
 431                 spin_lock_init(&xt_connlimit_locks[i]);
 432
 433         connlimit_conn_cachep = kmem_cache_create("xt_connlimit_conn",
 434                                            sizeof(struct xt_connlimit_conn),
 435                                            0, 0, NULL);
 436         if (!connlimit_conn_cachep)
 437                 return -ENOMEM;
 438
 439         connlimit_rb_cachep = kmem_cache_create("xt_connlimit_rb",
 440                                            sizeof(struct xt_connlimit_rb),
 441                                            0, 0, NULL);
 442         if (!connlimit_rb_cachep) {
 443                 kmem_cache_destroy(connlimit_conn_cachep);
 444                 return -ENOMEM;
 445         }
 446         ret = xt_register_match(&connlimit_mt_reg);
 447         if (ret != 0) {
 448                 kmem_cache_destroy(connlimit_conn_cachep);
 449                 kmem_cache_destroy(connlimit_rb_cachep);
 450         }
 451         return ret;
 452 }
 453
 454 static void __exit connlimit_mt_exit(void)
 455 {
 456         xt_unregister_match(&connlimit_mt_reg);
 457         kmem_cache_destroy(connlimit_conn_cachep);
 458         kmem_cache_destroy(connlimit_rb_cachep);
 459 }
 460
 461 module_init(connlimit_mt_init);
 462 module_exit(connlimit_mt_exit);
 463 MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>");
 464 MODULE_DESCRIPTION("Xtables: Number of connections matching");
 465 MODULE_LICENSE("GPL");
 466 MODULE_ALIAS("ipt_connlimit");
 467 MODULE_ALIAS("ip6t_connlimit");