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