net/netfilter/nf_flow_table_core.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 #include <linux/kernel.h>
   3 #include <linux/init.h>
   4 #include <linux/module.h>
   5 #include <linux/netfilter.h>
   6 #include <linux/rhashtable.h>
   7 #include <linux/netdevice.h>
   8 #include <net/ip.h>
   9 #include <net/ip6_route.h>
  10 #include <net/netfilter/nf_tables.h>
  11 #include <net/netfilter/nf_flow_table.h>
  12 #include <net/netfilter/nf_conntrack.h>
  13 #include <net/netfilter/nf_conntrack_core.h>
  14 #include <net/netfilter/nf_conntrack_tuple.h>
  15
  16 struct flow_offload_entry {
  17         struct flow_offload     flow;
  18         struct nf_conn          *ct;
  19         struct rcu_head         rcu_head;
  20 };
  21
  22 static DEFINE_MUTEX(flowtable_lock);
  23 static LIST_HEAD(flowtables);
  24
  25 static void
  26 flow_offload_fill_dir(struct flow_offload *flow, struct nf_conn *ct,
  27                       struct nf_flow_route *route,
  28                       enum flow_offload_tuple_dir dir)
  29 {
  30         struct flow_offload_tuple *ft = &flow->tuplehash[dir].tuple;
  31         struct nf_conntrack_tuple *ctt = &ct->tuplehash[dir].tuple;
  32         struct dst_entry *other_dst = route->tuple[!dir].dst;
  33         struct dst_entry *dst = route->tuple[dir].dst;
  34
  35         ft->dir = dir;
  36
  37         switch (ctt->src.l3num) {
  38         case NFPROTO_IPV4:
  39                 ft->src_v4 = ctt->src.u3.in;
  40                 ft->dst_v4 = ctt->dst.u3.in;
  41                 ft->mtu = ip_dst_mtu_maybe_forward(dst, true);
  42                 break;
  43         case NFPROTO_IPV6:
  44                 ft->src_v6 = ctt->src.u3.in6;
  45                 ft->dst_v6 = ctt->dst.u3.in6;
  46                 ft->mtu = ip6_dst_mtu_forward(dst);
  47                 break;
  48         }
  49
  50         ft->l3proto = ctt->src.l3num;
  51         ft->l4proto = ctt->dst.protonum;
  52         ft->src_port = ctt->src.u.tcp.port;
  53         ft->dst_port = ctt->dst.u.tcp.port;
  54
  55         ft->iifidx = other_dst->dev->ifindex;
  56         ft->dst_cache = dst;
  57 }
  58
  59 struct flow_offload *
  60 flow_offload_alloc(struct nf_conn *ct, struct nf_flow_route *route)
  61 {
  62         struct flow_offload_entry *entry;
  63         struct flow_offload *flow;
  64
  65         if (unlikely(nf_ct_is_dying(ct) ||
  66             !atomic_inc_not_zero(&ct->ct_general.use)))
  67                 return NULL;
  68
  69         entry = kzalloc(sizeof(*entry), GFP_ATOMIC);
  70         if (!entry)
  71                 goto err_ct_refcnt;
  72
  73         flow = &entry->flow;
  74
  75         if (!dst_hold_safe(route->tuple[FLOW_OFFLOAD_DIR_ORIGINAL].dst))
  76                 goto err_dst_cache_original;
  77
  78         if (!dst_hold_safe(route->tuple[FLOW_OFFLOAD_DIR_REPLY].dst))
  79                 goto err_dst_cache_reply;
  80
  81         entry->ct = ct;
  82
  83         flow_offload_fill_dir(flow, ct, route, FLOW_OFFLOAD_DIR_ORIGINAL);
  84         flow_offload_fill_dir(flow, ct, route, FLOW_OFFLOAD_DIR_REPLY);
  85
  86         if (ct->status & IPS_SRC_NAT)
  87                 flow->flags |= FLOW_OFFLOAD_SNAT;
  88         if (ct->status & IPS_DST_NAT)
  89                 flow->flags |= FLOW_OFFLOAD_DNAT;
  90
  91         return flow;
  92
  93 err_dst_cache_reply:
  94         dst_release(route->tuple[FLOW_OFFLOAD_DIR_ORIGINAL].dst);
  95 err_dst_cache_original:
  96         kfree(entry);
  97 err_ct_refcnt:
  98         nf_ct_put(ct);
  99
 100         return NULL;
 101 }
 102 EXPORT_SYMBOL_GPL(flow_offload_alloc);
 103
 104 static void flow_offload_fixup_tcp(struct ip_ct_tcp *tcp)
 105 {
 106         tcp->state = TCP_CONNTRACK_ESTABLISHED;
 107         tcp->seen[0].td_maxwin = 0;
 108         tcp->seen[1].td_maxwin = 0;
 109 }
 110
 111 #define NF_FLOWTABLE_TCP_PICKUP_TIMEOUT (120 * HZ)
 112 #define NF_FLOWTABLE_UDP_PICKUP_TIMEOUT (30 * HZ)
 113
 114 static void flow_offload_fixup_ct(struct nf_conn *ct)
 115 {
 116         const struct nf_conntrack_l4proto *l4proto;
 117         unsigned int timeout;
 118         int l4num;
 119
 120         l4num = nf_ct_protonum(ct);
 121         if (l4num == IPPROTO_TCP)
 122                 flow_offload_fixup_tcp(&ct->proto.tcp);
 123
 124         l4proto = nf_ct_l4proto_find(l4num);
 125         if (!l4proto)
 126                 return;
 127
 128         if (l4num == IPPROTO_TCP)
 129                 timeout = NF_FLOWTABLE_TCP_PICKUP_TIMEOUT;
 130         else if (l4num == IPPROTO_UDP)
 131                 timeout = NF_FLOWTABLE_UDP_PICKUP_TIMEOUT;
 132         else
 133                 return;
 134
 135         ct->timeout = nfct_time_stamp + timeout;
 136 }
 137
 138 void flow_offload_free(struct flow_offload *flow)
 139 {
 140         struct flow_offload_entry *e;
 141
 142         dst_release(flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_cache);
 143         dst_release(flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_cache);
 144         e = container_of(flow, struct flow_offload_entry, flow);
 145         if (flow->flags & FLOW_OFFLOAD_DYING)
 146                 nf_ct_delete(e->ct, 0, 0);
 147         nf_ct_put(e->ct);
 148         kfree_rcu(e, rcu_head);
 149 }
 150 EXPORT_SYMBOL_GPL(flow_offload_free);
 151
 152 static u32 flow_offload_hash(const void *data, u32 len, u32 seed)
 153 {
 154         const struct flow_offload_tuple *tuple = data;
 155
 156         return jhash(tuple, offsetof(struct flow_offload_tuple, dir), seed);
 157 }
 158
 159 static u32 flow_offload_hash_obj(const void *data, u32 len, u32 seed)
 160 {
 161         const struct flow_offload_tuple_rhash *tuplehash = data;
 162
 163         return jhash(&tuplehash->tuple, offsetof(struct flow_offload_tuple, dir), seed);
 164 }
 165
 166 static int flow_offload_hash_cmp(struct rhashtable_compare_arg *arg,
 167                                         const void *ptr)
 168 {
 169         const struct flow_offload_tuple *tuple = arg->key;
 170         const struct flow_offload_tuple_rhash *x = ptr;
 171
 172         if (memcmp(&x->tuple, tuple, offsetof(struct flow_offload_tuple, dir)))
 173                 return 1;
 174
 175         return 0;
 176 }
 177
 178 static const struct rhashtable_params nf_flow_offload_rhash_params = {
 179         .head_offset            = offsetof(struct flow_offload_tuple_rhash, node),
 180         .hashfn                 = flow_offload_hash,
 181         .obj_hashfn             = flow_offload_hash_obj,
 182         .obj_cmpfn              = flow_offload_hash_cmp,
 183         .automatic_shrinking    = true,
 184 };
 185
 186 int flow_offload_add(struct nf_flowtable *flow_table, struct flow_offload *flow)
 187 {
 188         int err;
 189
 190         err = rhashtable_insert_fast(&flow_table->rhashtable,
 191                                      &flow->tuplehash[0].node,
 192                                      nf_flow_offload_rhash_params);
 193         if (err < 0)
 194                 return err;
 195
 196         err = rhashtable_insert_fast(&flow_table->rhashtable,
 197                                      &flow->tuplehash[1].node,
 198                                      nf_flow_offload_rhash_params);
 199         if (err < 0) {
 200                 rhashtable_remove_fast(&flow_table->rhashtable,
 201                                        &flow->tuplehash[0].node,
 202                                        nf_flow_offload_rhash_params);
 203                 return err;
 204         }
 205
 206         flow->timeout = (u32)jiffies;
 207         return 0;
 208 }
 209 EXPORT_SYMBOL_GPL(flow_offload_add);
 210
 211 static inline bool nf_flow_has_expired(const struct flow_offload *flow)
 212 {
 213         return (__s32)(flow->timeout - (u32)jiffies) <= 0;
 214 }
 215
 216 static void flow_offload_del(struct nf_flowtable *flow_table,
 217                              struct flow_offload *flow)
 218 {
 219         struct flow_offload_entry *e;
 220
 221         rhashtable_remove_fast(&flow_table->rhashtable,
 222                                &flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].node,
 223                                nf_flow_offload_rhash_params);
 224         rhashtable_remove_fast(&flow_table->rhashtable,
 225                                &flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].node,
 226                                nf_flow_offload_rhash_params);
 227
 228         e = container_of(flow, struct flow_offload_entry, flow);
 229         clear_bit(IPS_OFFLOAD_BIT, &e->ct->status);
 230
 231         if (nf_flow_has_expired(flow))
 232                 flow_offload_fixup_ct(e->ct);
 233
 234         flow_offload_free(flow);
 235 }
 236
 237 void flow_offload_teardown(struct flow_offload *flow)
 238 {
 239         struct flow_offload_entry *e;
 240
 241         flow->flags |= FLOW_OFFLOAD_TEARDOWN;
 242
 243         e = container_of(flow, struct flow_offload_entry, flow);
 244         flow_offload_fixup_ct(e->ct);
 245 }
 246 EXPORT_SYMBOL_GPL(flow_offload_teardown);
 247
 248 struct flow_offload_tuple_rhash *
 249 flow_offload_lookup(struct nf_flowtable *flow_table,
 250                     struct flow_offload_tuple *tuple)
 251 {
 252         struct flow_offload_tuple_rhash *tuplehash;
 253         struct flow_offload *flow;
 254         struct flow_offload_entry *e;
 255         int dir;
 256
 257         tuplehash = rhashtable_lookup(&flow_table->rhashtable, tuple,
 258                                       nf_flow_offload_rhash_params);
 259         if (!tuplehash)
 260                 return NULL;
 261
 262         dir = tuplehash->tuple.dir;
 263         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 264         if (flow->flags & (FLOW_OFFLOAD_DYING | FLOW_OFFLOAD_TEARDOWN))
 265                 return NULL;
 266
 267         e = container_of(flow, struct flow_offload_entry, flow);
 268         if (unlikely(nf_ct_is_dying(e->ct)))
 269                 return NULL;
 270
 271         return tuplehash;
 272 }
 273 EXPORT_SYMBOL_GPL(flow_offload_lookup);
 274
 275 static int
 276 nf_flow_table_iterate(struct nf_flowtable *flow_table,
 277                       void (*iter)(struct flow_offload *flow, void *data),
 278                       void *data)
 279 {
 280         struct flow_offload_tuple_rhash *tuplehash;
 281         struct rhashtable_iter hti;
 282         struct flow_offload *flow;
 283         int err = 0;
 284
 285         rhashtable_walk_enter(&flow_table->rhashtable, &hti);
 286         rhashtable_walk_start(&hti);
 287
 288         while ((tuplehash = rhashtable_walk_next(&hti))) {
 289                 if (IS_ERR(tuplehash)) {
 290                         if (PTR_ERR(tuplehash) != -EAGAIN) {
 291                                 err = PTR_ERR(tuplehash);
 292                                 break;
 293                         }
 294                         continue;
 295                 }
 296                 if (tuplehash->tuple.dir)
 297                         continue;
 298
 299                 flow = container_of(tuplehash, struct flow_offload, tuplehash[0]);
 300
 301                 iter(flow, data);
 302         }
 303         rhashtable_walk_stop(&hti);
 304         rhashtable_walk_exit(&hti);
 305
 306         return err;
 307 }
 308
 309 static void nf_flow_offload_gc_step(struct flow_offload *flow, void *data)
 310 {
 311         struct nf_flowtable *flow_table = data;
 312         struct flow_offload_entry *e;
 313
 314         e = container_of(flow, struct flow_offload_entry, flow);
 315         if (nf_flow_has_expired(flow) || nf_ct_is_dying(e->ct) ||
 316             (flow->flags & (FLOW_OFFLOAD_DYING | FLOW_OFFLOAD_TEARDOWN)))
 317                 flow_offload_del(flow_table, flow);
 318 }
 319
 320 static void nf_flow_offload_work_gc(struct work_struct *work)
 321 {
 322         struct nf_flowtable *flow_table;
 323
 324         flow_table = container_of(work, struct nf_flowtable, gc_work.work);
 325         nf_flow_table_iterate(flow_table, nf_flow_offload_gc_step, flow_table);
 326         queue_delayed_work(system_power_efficient_wq, &flow_table->gc_work, HZ);
 327 }
 328
 329 static int nf_flow_nat_port_tcp(struct sk_buff *skb, unsigned int thoff,
 330                                 __be16 port, __be16 new_port)
 331 {
 332         struct tcphdr *tcph;
 333
 334         if (!pskb_may_pull(skb, thoff + sizeof(*tcph)) ||
 335             skb_try_make_writable(skb, thoff + sizeof(*tcph)))
 336                 return -1;
 337
 338         tcph = (void *)(skb_network_header(skb) + thoff);
 339         inet_proto_csum_replace2(&tcph->check, skb, port, new_port, true);
 340
 341         return 0;
 342 }
 343
 344 static int nf_flow_nat_port_udp(struct sk_buff *skb, unsigned int thoff,
 345                                 __be16 port, __be16 new_port)
 346 {
 347         struct udphdr *udph;
 348
 349         if (!pskb_may_pull(skb, thoff + sizeof(*udph)) ||
 350             skb_try_make_writable(skb, thoff + sizeof(*udph)))
 351                 return -1;
 352
 353         udph = (void *)(skb_network_header(skb) + thoff);
 354         if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 355                 inet_proto_csum_replace2(&udph->check, skb, port,
 356                                          new_port, true);
 357                 if (!udph->check)
 358                         udph->check = CSUM_MANGLED_0;
 359         }
 360
 361         return 0;
 362 }
 363
 364 static int nf_flow_nat_port(struct sk_buff *skb, unsigned int thoff,
 365                             u8 protocol, __be16 port, __be16 new_port)
 366 {
 367         switch (protocol) {
 368         case IPPROTO_TCP:
 369                 if (nf_flow_nat_port_tcp(skb, thoff, port, new_port) < 0)
 370                         return NF_DROP;
 371                 break;
 372         case IPPROTO_UDP:
 373                 if (nf_flow_nat_port_udp(skb, thoff, port, new_port) < 0)
 374                         return NF_DROP;
 375                 break;
 376         }
 377
 378         return 0;
 379 }
 380
 381 int nf_flow_snat_port(const struct flow_offload *flow,
 382                       struct sk_buff *skb, unsigned int thoff,
 383                       u8 protocol, enum flow_offload_tuple_dir dir)
 384 {
 385         struct flow_ports *hdr;
 386         __be16 port, new_port;
 387
 388         if (!pskb_may_pull(skb, thoff + sizeof(*hdr)) ||
 389             skb_try_make_writable(skb, thoff + sizeof(*hdr)))
 390                 return -1;
 391
 392         hdr = (void *)(skb_network_header(skb) + thoff);
 393
 394         switch (dir) {
 395         case FLOW_OFFLOAD_DIR_ORIGINAL:
 396                 port = hdr->source;
 397                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_port;
 398                 hdr->source = new_port;
 399                 break;
 400         case FLOW_OFFLOAD_DIR_REPLY:
 401                 port = hdr->dest;
 402                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_port;
 403                 hdr->dest = new_port;
 404                 break;
 405         default:
 406                 return -1;
 407         }
 408
 409         return nf_flow_nat_port(skb, thoff, protocol, port, new_port);
 410 }
 411 EXPORT_SYMBOL_GPL(nf_flow_snat_port);
 412
 413 int nf_flow_dnat_port(const struct flow_offload *flow,
 414                       struct sk_buff *skb, unsigned int thoff,
 415                       u8 protocol, enum flow_offload_tuple_dir dir)
 416 {
 417         struct flow_ports *hdr;
 418         __be16 port, new_port;
 419
 420         if (!pskb_may_pull(skb, thoff + sizeof(*hdr)) ||
 421             skb_try_make_writable(skb, thoff + sizeof(*hdr)))
 422                 return -1;
 423
 424         hdr = (void *)(skb_network_header(skb) + thoff);
 425
 426         switch (dir) {
 427         case FLOW_OFFLOAD_DIR_ORIGINAL:
 428                 port = hdr->dest;
 429                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_port;
 430                 hdr->dest = new_port;
 431                 break;
 432         case FLOW_OFFLOAD_DIR_REPLY:
 433                 port = hdr->source;
 434                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_port;
 435                 hdr->source = new_port;
 436                 break;
 437         default:
 438                 return -1;
 439         }
 440
 441         return nf_flow_nat_port(skb, thoff, protocol, port, new_port);
 442 }
 443 EXPORT_SYMBOL_GPL(nf_flow_dnat_port);
 444
 445 int nf_flow_table_init(struct nf_flowtable *flowtable)
 446 {
 447         int err;
 448
 449         INIT_DEFERRABLE_WORK(&flowtable->gc_work, nf_flow_offload_work_gc);
 450
 451         err = rhashtable_init(&flowtable->rhashtable,
 452                               &nf_flow_offload_rhash_params);
 453         if (err < 0)
 454                 return err;
 455
 456         queue_delayed_work(system_power_efficient_wq,
 457                            &flowtable->gc_work, HZ);
 458
 459         mutex_lock(&flowtable_lock);
 460         list_add(&flowtable->list, &flowtables);
 461         mutex_unlock(&flowtable_lock);
 462
 463         return 0;
 464 }
 465 EXPORT_SYMBOL_GPL(nf_flow_table_init);
 466
 467 static void nf_flow_table_do_cleanup(struct flow_offload *flow, void *data)
 468 {
 469         struct net_device *dev = data;
 470         struct flow_offload_entry *e;
 471
 472         e = container_of(flow, struct flow_offload_entry, flow);
 473
 474         if (!dev) {
 475                 flow_offload_teardown(flow);
 476                 return;
 477         }
 478         if (net_eq(nf_ct_net(e->ct), dev_net(dev)) &&
 479             (flow->tuplehash[0].tuple.iifidx == dev->ifindex ||
 480              flow->tuplehash[1].tuple.iifidx == dev->ifindex))
 481                 flow_offload_dead(flow);
 482 }
 483
 484 static void nf_flow_table_iterate_cleanup(struct nf_flowtable *flowtable,
 485                                           struct net_device *dev)
 486 {
 487         nf_flow_table_iterate(flowtable, nf_flow_table_do_cleanup, dev);
 488         flush_delayed_work(&flowtable->gc_work);
 489 }
 490
 491 void nf_flow_table_cleanup(struct net_device *dev)
 492 {
 493         struct nf_flowtable *flowtable;
 494
 495         mutex_lock(&flowtable_lock);
 496         list_for_each_entry(flowtable, &flowtables, list)
 497                 nf_flow_table_iterate_cleanup(flowtable, dev);
 498         mutex_unlock(&flowtable_lock);
 499 }
 500 EXPORT_SYMBOL_GPL(nf_flow_table_cleanup);
 501
 502 void nf_flow_table_free(struct nf_flowtable *flow_table)
 503 {
 504         mutex_lock(&flowtable_lock);
 505         list_del(&flow_table->list);
 506         mutex_unlock(&flowtable_lock);
 507         cancel_delayed_work_sync(&flow_table->gc_work);
 508         nf_flow_table_iterate(flow_table, nf_flow_table_do_cleanup, NULL);
 509         nf_flow_table_iterate(flow_table, nf_flow_offload_gc_step, flow_table);
 510         rhashtable_destroy(&flow_table->rhashtable);
 511 }
 512 EXPORT_SYMBOL_GPL(nf_flow_table_free);
 513
 514 MODULE_LICENSE("GPL");
 515 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");