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