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