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         32
  36 #define CONNLIMIT_LOCK_SLOTS    32
  37 #define CONNLIMIT_GC_MAX_NODES  8
  38
  39 /* we will save the tuples of all connections we care about */
  40 struct xt_connlimit_conn {
  41         struct hlist_node               node;
  42         struct nf_conntrack_tuple       tuple;
  43         union nf_inet_addr              addr;
  44 };
  45
  46 struct xt_connlimit_rb {
  47         struct rb_node node;
  48         struct hlist_head hhead; /* connections/hosts in same subnet */
  49         union nf_inet_addr addr; /* search key */
  50 };
  51
  52 struct xt_connlimit_data {
  53         struct rb_root climit_root4[CONNLIMIT_SLOTS];
  54         struct rb_root climit_root6[CONNLIMIT_SLOTS];
  55         spinlock_t              locks[CONNLIMIT_LOCK_SLOTS];
  56 };
  57
  58 static u_int32_t connlimit_rnd __read_mostly;
  59 static struct kmem_cache *connlimit_rb_cachep __read_mostly;
  60 static struct kmem_cache *connlimit_conn_cachep __read_mostly;
  61
  62 static inline unsigned int connlimit_iphash(__be32 addr)
  63 {
  64         return jhash_1word((__force __u32)addr,
  65                             connlimit_rnd) % CONNLIMIT_SLOTS;
  66 }
  67
  68 static inline unsigned int
  69 connlimit_iphash6(const union nf_inet_addr *addr,
  70                   const union nf_inet_addr *mask)
  71 {
  72         union nf_inet_addr res;
  73         unsigned int i;
  74
  75         for (i = 0; i < ARRAY_SIZE(addr->ip6); ++i)
  76                 res.ip6[i] = addr->ip6[i] & mask->ip6[i];
  77
  78         return jhash2((u32 *)res.ip6, ARRAY_SIZE(res.ip6),
  79                        connlimit_rnd) % CONNLIMIT_SLOTS;
  80 }
  81
  82 static inline bool already_closed(const struct nf_conn *conn)
  83 {
  84         if (nf_ct_protonum(conn) == IPPROTO_TCP)
  85                 return conn->proto.tcp.state == TCP_CONNTRACK_TIME_WAIT ||
  86                        conn->proto.tcp.state == TCP_CONNTRACK_CLOSE;
  87         else
  88                 return 0;
  89 }
  90
  91 static int
  92 same_source_net(const union nf_inet_addr *addr,
  93                 const union nf_inet_addr *mask,
  94                 const union nf_inet_addr *u3, u_int8_t family)
  95 {
  96         if (family == NFPROTO_IPV4) {
  97                 return ntohl(addr->ip & mask->ip) -
  98                        ntohl(u3->ip & mask->ip);
  99         } else {
 100                 union nf_inet_addr lh, rh;
 101                 unsigned int i;
 102
 103                 for (i = 0; i < ARRAY_SIZE(addr->ip6); ++i) {
 104                         lh.ip6[i] = addr->ip6[i] & mask->ip6[i];
 105                         rh.ip6[i] = u3->ip6[i] & mask->ip6[i];
 106                 }
 107
 108                 return memcmp(&lh.ip6, &rh.ip6, sizeof(lh.ip6));
 109         }
 110 }
 111
 112 static bool add_hlist(struct hlist_head *head,
 113                       const struct nf_conntrack_tuple *tuple,
 114                       const union nf_inet_addr *addr)
 115 {
 116         struct xt_connlimit_conn *conn;
 117
 118         conn = kmem_cache_alloc(connlimit_conn_cachep, GFP_ATOMIC);
 119         if (conn == NULL)
 120                 return false;
 121         conn->tuple = *tuple;
 122         conn->addr = *addr;
 123         hlist_add_head(&conn->node, head);
 124         return true;
 125 }
 126
 127 static unsigned int check_hlist(struct net *net,
 128                                 struct hlist_head *head,
 129                                 const struct nf_conntrack_tuple *tuple,
 130                                 bool *addit)
 131 {
 132         const struct nf_conntrack_tuple_hash *found;
 133         struct xt_connlimit_conn *conn;
 134         struct hlist_node *n;
 135         struct nf_conn *found_ct;
 136         unsigned int length = 0;
 137
 138         *addit = true;
 139         rcu_read_lock();
 140
 141         /* check the saved connections */
 142         hlist_for_each_entry_safe(conn, n, head, node) {
 143                 found    = nf_conntrack_find_get(net, NF_CT_DEFAULT_ZONE,
 144                                                  &conn->tuple);
 145                 if (found == NULL) {
 146                         hlist_del(&conn->node);
 147                         kmem_cache_free(connlimit_conn_cachep, conn);
 148                         continue;
 149                 }
 150
 151                 found_ct = nf_ct_tuplehash_to_ctrack(found);
 152
 153                 if (nf_ct_tuple_equal(&conn->tuple, tuple)) {
 154                         /*
 155                          * Just to be sure we have it only once in the list.
 156                          * We should not see tuples twice unless someone hooks
 157                          * this into a table without "-p tcp --syn".
 158                          */
 159                         *addit = false;
 160                 } else if (already_closed(found_ct)) {
 161                         /*
 162                          * we do not care about connections which are
 163                          * closed already -> ditch it
 164                          */
 165                         nf_ct_put(found_ct);
 166                         hlist_del(&conn->node);
 167                         kmem_cache_free(connlimit_conn_cachep, conn);
 168                         continue;
 169                 }
 170
 171                 nf_ct_put(found_ct);
 172                 length++;
 173         }
 174
 175         rcu_read_unlock();
 176
 177         return length;
 178 }
 179
 180 static void tree_nodes_free(struct rb_root *root,
 181                             struct xt_connlimit_rb *gc_nodes[],
 182                             unsigned int gc_count)
 183 {
 184         struct xt_connlimit_rb *rbconn;
 185
 186         while (gc_count) {
 187                 rbconn = gc_nodes[--gc_count];
 188                 rb_erase(&rbconn->node, root);
 189                 kmem_cache_free(connlimit_rb_cachep, rbconn);
 190         }
 191 }
 192
 193 static unsigned int
 194 count_tree(struct net *net, struct rb_root *root,
 195            const struct nf_conntrack_tuple *tuple,
 196            const union nf_inet_addr *addr, const union nf_inet_addr *mask,
 197            u8 family)
 198 {
 199         struct xt_connlimit_rb *gc_nodes[CONNLIMIT_GC_MAX_NODES];
 200         struct rb_node **rbnode, *parent;
 201         struct xt_connlimit_rb *rbconn;
 202         struct xt_connlimit_conn *conn;
 203         unsigned int gc_count;
 204         bool no_gc = false;
 205
 206  restart:
 207         gc_count = 0;
 208         parent = NULL;
 209         rbnode = &(root->rb_node);
 210         while (*rbnode) {
 211                 int diff;
 212                 bool addit;
 213
 214                 rbconn = container_of(*rbnode, struct xt_connlimit_rb, node);
 215
 216                 parent = *rbnode;
 217                 diff = same_source_net(addr, mask, &rbconn->addr, family);
 218                 if (diff < 0) {
 219                         rbnode = &((*rbnode)->rb_left);
 220                 } else if (diff > 0) {
 221                         rbnode = &((*rbnode)->rb_right);
 222                 } else {
 223                         /* same source network -> be counted! */
 224                         unsigned int count;
 225                         count = check_hlist(net, &rbconn->hhead, tuple, &addit);
 226
 227                         tree_nodes_free(root, gc_nodes, gc_count);
 228                         if (!addit)
 229                                 return count;
 230
 231                         if (!add_hlist(&rbconn->hhead, tuple, addr))
 232                                 return 0; /* hotdrop */
 233
 234                         return count + 1;
 235                 }
 236
 237                 if (no_gc || gc_count >= ARRAY_SIZE(gc_nodes))
 238                         continue;
 239
 240                 /* only used for GC on hhead, retval and 'addit' ignored */
 241                 check_hlist(net, &rbconn->hhead, tuple, &addit);
 242                 if (hlist_empty(&rbconn->hhead))
 243                         gc_nodes[gc_count++] = rbconn;
 244         }
 245
 246         if (gc_count) {
 247                 no_gc = true;
 248                 tree_nodes_free(root, gc_nodes, gc_count);
 249                 /* tree_node_free before new allocation permits
 250                  * allocator to re-use newly free'd object.
 251                  *
 252                  * This is a rare event; in most cases we will find
 253                  * existing node to re-use. (or gc_count is 0).
 254                  */
 255                 goto restart;
 256         }
 257
 258         /* no match, need to insert new node */
 259         rbconn = kmem_cache_alloc(connlimit_rb_cachep, GFP_ATOMIC);
 260         if (rbconn == NULL)
 261                 return 0;
 262
 263         conn = kmem_cache_alloc(connlimit_conn_cachep, GFP_ATOMIC);
 264         if (conn == NULL) {
 265                 kmem_cache_free(connlimit_rb_cachep, rbconn);
 266                 return 0;
 267         }
 268
 269         conn->tuple = *tuple;
 270         conn->addr = *addr;
 271         rbconn->addr = *addr;
 272
 273         INIT_HLIST_HEAD(&rbconn->hhead);
 274         hlist_add_head(&conn->node, &rbconn->hhead);
 275
 276         rb_link_node(&rbconn->node, parent, rbnode);
 277         rb_insert_color(&rbconn->node, root);
 278         return 1;
 279 }
 280
 281 static int count_them(struct net *net,
 282                       struct xt_connlimit_data *data,
 283                       const struct nf_conntrack_tuple *tuple,
 284                       const union nf_inet_addr *addr,
 285                       const union nf_inet_addr *mask,
 286                       u_int8_t family)
 287 {
 288         struct rb_root *root;
 289         int count;
 290         u32 hash;
 291
 292         if (family == NFPROTO_IPV6) {
 293                 hash = connlimit_iphash6(addr, mask);
 294                 root = &data->climit_root6[hash];
 295         } else {
 296                 hash = connlimit_iphash(addr->ip & mask->ip);
 297                 root = &data->climit_root4[hash];
 298         }
 299
 300         spin_lock_bh(&data->locks[hash % CONNLIMIT_LOCK_SLOTS]);
 301
 302         count = count_tree(net, root, tuple, addr, mask, family);
 303
 304         spin_unlock_bh(&data->locks[hash % CONNLIMIT_LOCK_SLOTS]);
 305
 306         return count;
 307 }
 308
 309 static bool
 310 connlimit_mt(const struct sk_buff *skb, struct xt_action_param *par)
 311 {
 312         struct net *net = dev_net(par->in ? par->in : par->out);
 313         const struct xt_connlimit_info *info = par->matchinfo;
 314         union nf_inet_addr addr;
 315         struct nf_conntrack_tuple tuple;
 316         const struct nf_conntrack_tuple *tuple_ptr = &tuple;
 317         enum ip_conntrack_info ctinfo;
 318         const struct nf_conn *ct;
 319         unsigned int connections;
 320
 321         ct = nf_ct_get(skb, &ctinfo);
 322         if (ct != NULL)
 323                 tuple_ptr = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 324         else if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb),
 325                                     par->family, &tuple))
 326                 goto hotdrop;
 327
 328         if (par->family == NFPROTO_IPV6) {
 329                 const struct ipv6hdr *iph = ipv6_hdr(skb);
 330                 memcpy(&addr.ip6, (info->flags & XT_CONNLIMIT_DADDR) ?
 331                        &iph->daddr : &iph->saddr, sizeof(addr.ip6));
 332         } else {
 333                 const struct iphdr *iph = ip_hdr(skb);
 334                 addr.ip = (info->flags & XT_CONNLIMIT_DADDR) ?
 335                           iph->daddr : iph->saddr;
 336         }
 337
 338         connections = count_them(net, info->data, tuple_ptr, &addr,
 339                                  &info->mask, par->family);
 340         if (connections == 0)
 341                 /* kmalloc failed, drop it entirely */
 342                 goto hotdrop;
 343
 344         return (connections > info->limit) ^
 345                !!(info->flags & XT_CONNLIMIT_INVERT);
 346
 347  hotdrop:
 348         par->hotdrop = true;
 349         return false;
 350 }
 351
 352 static int connlimit_mt_check(const struct xt_mtchk_param *par)
 353 {
 354         struct xt_connlimit_info *info = par->matchinfo;
 355         unsigned int i;
 356         int ret;
 357
 358         if (unlikely(!connlimit_rnd)) {
 359                 u_int32_t rand;
 360
 361                 do {
 362                         get_random_bytes(&rand, sizeof(rand));
 363                 } while (!rand);
 364                 cmpxchg(&connlimit_rnd, 0, rand);
 365         }
 366         ret = nf_ct_l3proto_try_module_get(par->family);
 367         if (ret < 0) {
 368                 pr_info("cannot load conntrack support for "
 369                         "address family %u\n", par->family);
 370                 return ret;
 371         }
 372
 373         /* init private data */
 374         info->data = kmalloc(sizeof(struct xt_connlimit_data), GFP_KERNEL);
 375         if (info->data == NULL) {
 376                 nf_ct_l3proto_module_put(par->family);
 377                 return -ENOMEM;
 378         }
 379
 380         for (i = 0; i < ARRAY_SIZE(info->data->locks); ++i)
 381                 spin_lock_init(&info->data->locks[i]);
 382
 383         for (i = 0; i < ARRAY_SIZE(info->data->climit_root4); ++i)
 384                 info->data->climit_root4[i] = RB_ROOT;
 385         for (i = 0; i < ARRAY_SIZE(info->data->climit_root6); ++i)
 386                 info->data->climit_root6[i] = RB_ROOT;
 387
 388         return 0;
 389 }
 390
 391 static void destroy_tree(struct rb_root *r)
 392 {
 393         struct xt_connlimit_conn *conn;
 394         struct xt_connlimit_rb *rbconn;
 395         struct hlist_node *n;
 396         struct rb_node *node;
 397
 398         while ((node = rb_first(r)) != NULL) {
 399                 rbconn = container_of(node, struct xt_connlimit_rb, node);
 400
 401                 rb_erase(node, r);
 402
 403                 hlist_for_each_entry_safe(conn, n, &rbconn->hhead, node)
 404                         kmem_cache_free(connlimit_conn_cachep, conn);
 405
 406                 kmem_cache_free(connlimit_rb_cachep, rbconn);
 407         }
 408 }
 409
 410 static void connlimit_mt_destroy(const struct xt_mtdtor_param *par)
 411 {
 412         const struct xt_connlimit_info *info = par->matchinfo;
 413         unsigned int i;
 414
 415         nf_ct_l3proto_module_put(par->family);
 416
 417         for (i = 0; i < ARRAY_SIZE(info->data->climit_root4); ++i)
 418                 destroy_tree(&info->data->climit_root4[i]);
 419         for (i = 0; i < ARRAY_SIZE(info->data->climit_root6); ++i)
 420                 destroy_tree(&info->data->climit_root6[i]);
 421
 422         kfree(info->data);
 423 }
 424
 425 static struct xt_match connlimit_mt_reg __read_mostly = {
 426         .name       = "connlimit",
 427         .revision   = 1,
 428         .family     = NFPROTO_UNSPEC,
 429         .checkentry = connlimit_mt_check,
 430         .match      = connlimit_mt,
 431         .matchsize  = sizeof(struct xt_connlimit_info),
 432         .destroy    = connlimit_mt_destroy,
 433         .me         = THIS_MODULE,
 434 };
 435
 436 static int __init connlimit_mt_init(void)
 437 {
 438         int ret;
 439
 440         BUILD_BUG_ON(CONNLIMIT_LOCK_SLOTS > CONNLIMIT_SLOTS);
 441         BUILD_BUG_ON((CONNLIMIT_SLOTS % CONNLIMIT_LOCK_SLOTS) != 0);
 442
 443         connlimit_conn_cachep = kmem_cache_create("xt_connlimit_conn",
 444                                            sizeof(struct xt_connlimit_conn),
 445                                            0, 0, NULL);
 446         if (!connlimit_conn_cachep)
 447                 return -ENOMEM;
 448
 449         connlimit_rb_cachep = kmem_cache_create("xt_connlimit_rb",
 450                                            sizeof(struct xt_connlimit_rb),
 451                                            0, 0, NULL);
 452         if (!connlimit_rb_cachep) {
 453                 kmem_cache_destroy(connlimit_conn_cachep);
 454                 return -ENOMEM;
 455         }
 456         ret = xt_register_match(&connlimit_mt_reg);
 457         if (ret != 0) {
 458                 kmem_cache_destroy(connlimit_conn_cachep);
 459                 kmem_cache_destroy(connlimit_rb_cachep);
 460         }
 461         return ret;
 462 }
 463
 464 static void __exit connlimit_mt_exit(void)
 465 {
 466         xt_unregister_match(&connlimit_mt_reg);
 467         kmem_cache_destroy(connlimit_conn_cachep);
 468         kmem_cache_destroy(connlimit_rb_cachep);
 469 }
 470
 471 module_init(connlimit_mt_init);
 472 module_exit(connlimit_mt_exit);
 473 MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>");
 474 MODULE_DESCRIPTION("Xtables: Number of connections matching");
 475 MODULE_LICENSE("GPL");
 476 MODULE_ALIAS("ipt_connlimit");
 477 MODULE_ALIAS("ip6t_connlimit");