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 inline __s32 nf_flow_timeout_delta(unsigned int timeout)
 115 {
 116         return (__s32)(timeout - (u32)jiffies);
 117 }
 118
 119 static void flow_offload_fixup_ct_timeout(struct nf_conn *ct)
 120 {
 121         const struct nf_conntrack_l4proto *l4proto;
 122         int l4num = nf_ct_protonum(ct);
 123         unsigned int timeout;
 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         if (nf_flow_timeout_delta(ct->timeout) > (__s32)timeout)
 137                 ct->timeout = nfct_time_stamp + timeout;
 138 }
 139
 140 static void flow_offload_fixup_ct_state(struct nf_conn *ct)
 141 {
 142         if (nf_ct_protonum(ct) == IPPROTO_TCP)
 143                 flow_offload_fixup_tcp(&ct->proto.tcp);
 144 }
 145
 146 static void flow_offload_fixup_ct(struct nf_conn *ct)
 147 {
 148         flow_offload_fixup_ct_state(ct);
 149         flow_offload_fixup_ct_timeout(ct);
 150 }
 151
 152 void flow_offload_free(struct flow_offload *flow)
 153 {
 154         struct flow_offload_entry *e;
 155
 156         dst_release(flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_cache);
 157         dst_release(flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_cache);
 158         e = container_of(flow, struct flow_offload_entry, flow);
 159         if (flow->flags & FLOW_OFFLOAD_DYING)
 160                 nf_ct_delete(e->ct, 0, 0);
 161         nf_ct_put(e->ct);
 162         kfree_rcu(e, rcu_head);
 163 }
 164 EXPORT_SYMBOL_GPL(flow_offload_free);
 165
 166 static u32 flow_offload_hash(const void *data, u32 len, u32 seed)
 167 {
 168         const struct flow_offload_tuple *tuple = data;
 169
 170         return jhash(tuple, offsetof(struct flow_offload_tuple, dir), seed);
 171 }
 172
 173 static u32 flow_offload_hash_obj(const void *data, u32 len, u32 seed)
 174 {
 175         const struct flow_offload_tuple_rhash *tuplehash = data;
 176
 177         return jhash(&tuplehash->tuple, offsetof(struct flow_offload_tuple, dir), seed);
 178 }
 179
 180 static int flow_offload_hash_cmp(struct rhashtable_compare_arg *arg,
 181                                         const void *ptr)
 182 {
 183         const struct flow_offload_tuple *tuple = arg->key;
 184         const struct flow_offload_tuple_rhash *x = ptr;
 185
 186         if (memcmp(&x->tuple, tuple, offsetof(struct flow_offload_tuple, dir)))
 187                 return 1;
 188
 189         return 0;
 190 }
 191
 192 static const struct rhashtable_params nf_flow_offload_rhash_params = {
 193         .head_offset            = offsetof(struct flow_offload_tuple_rhash, node),
 194         .hashfn                 = flow_offload_hash,
 195         .obj_hashfn             = flow_offload_hash_obj,
 196         .obj_cmpfn              = flow_offload_hash_cmp,
 197         .automatic_shrinking    = true,
 198 };
 199
 200 int flow_offload_add(struct nf_flowtable *flow_table, struct flow_offload *flow)
 201 {
 202         int err;
 203
 204         flow->timeout = (u32)jiffies + NF_FLOW_TIMEOUT;
 205
 206         err = rhashtable_insert_fast(&flow_table->rhashtable,
 207                                      &flow->tuplehash[0].node,
 208                                      nf_flow_offload_rhash_params);
 209         if (err < 0)
 210                 return err;
 211
 212         err = rhashtable_insert_fast(&flow_table->rhashtable,
 213                                      &flow->tuplehash[1].node,
 214                                      nf_flow_offload_rhash_params);
 215         if (err < 0) {
 216                 rhashtable_remove_fast(&flow_table->rhashtable,
 217                                        &flow->tuplehash[0].node,
 218                                        nf_flow_offload_rhash_params);
 219                 return err;
 220         }
 221
 222         return 0;
 223 }
 224 EXPORT_SYMBOL_GPL(flow_offload_add);
 225
 226 static inline bool nf_flow_has_expired(const struct flow_offload *flow)
 227 {
 228         return nf_flow_timeout_delta(flow->timeout) <= 0;
 229 }
 230
 231 static void flow_offload_del(struct nf_flowtable *flow_table,
 232                              struct flow_offload *flow)
 233 {
 234         struct flow_offload_entry *e;
 235
 236         rhashtable_remove_fast(&flow_table->rhashtable,
 237                                &flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].node,
 238                                nf_flow_offload_rhash_params);
 239         rhashtable_remove_fast(&flow_table->rhashtable,
 240                                &flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].node,
 241                                nf_flow_offload_rhash_params);
 242
 243         e = container_of(flow, struct flow_offload_entry, flow);
 244         clear_bit(IPS_OFFLOAD_BIT, &e->ct->status);
 245
 246         if (nf_flow_has_expired(flow))
 247                 flow_offload_fixup_ct(e->ct);
 248         else if (flow->flags & FLOW_OFFLOAD_TEARDOWN)
 249                 flow_offload_fixup_ct_timeout(e->ct);
 250
 251         flow_offload_free(flow);
 252 }
 253
 254 void flow_offload_teardown(struct flow_offload *flow)
 255 {
 256         struct flow_offload_entry *e;
 257
 258         flow->flags |= FLOW_OFFLOAD_TEARDOWN;
 259
 260         e = container_of(flow, struct flow_offload_entry, flow);
 261         flow_offload_fixup_ct_state(e->ct);
 262 }
 263 EXPORT_SYMBOL_GPL(flow_offload_teardown);
 264
 265 struct flow_offload_tuple_rhash *
 266 flow_offload_lookup(struct nf_flowtable *flow_table,
 267                     struct flow_offload_tuple *tuple)
 268 {
 269         struct flow_offload_tuple_rhash *tuplehash;
 270         struct flow_offload *flow;
 271         struct flow_offload_entry *e;
 272         int dir;
 273
 274         tuplehash = rhashtable_lookup(&flow_table->rhashtable, tuple,
 275                                       nf_flow_offload_rhash_params);
 276         if (!tuplehash)
 277                 return NULL;
 278
 279         dir = tuplehash->tuple.dir;
 280         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 281         if (flow->flags & (FLOW_OFFLOAD_DYING | FLOW_OFFLOAD_TEARDOWN))
 282                 return NULL;
 283
 284         e = container_of(flow, struct flow_offload_entry, flow);
 285         if (unlikely(nf_ct_is_dying(e->ct)))
 286                 return NULL;
 287
 288         return tuplehash;
 289 }
 290 EXPORT_SYMBOL_GPL(flow_offload_lookup);
 291
 292 static int
 293 nf_flow_table_iterate(struct nf_flowtable *flow_table,
 294                       void (*iter)(struct flow_offload *flow, void *data),
 295                       void *data)
 296 {
 297         struct flow_offload_tuple_rhash *tuplehash;
 298         struct rhashtable_iter hti;
 299         struct flow_offload *flow;
 300         int err = 0;
 301
 302         rhashtable_walk_enter(&flow_table->rhashtable, &hti);
 303         rhashtable_walk_start(&hti);
 304
 305         while ((tuplehash = rhashtable_walk_next(&hti))) {
 306                 if (IS_ERR(tuplehash)) {
 307                         if (PTR_ERR(tuplehash) != -EAGAIN) {
 308                                 err = PTR_ERR(tuplehash);
 309                                 break;
 310                         }
 311                         continue;
 312                 }
 313                 if (tuplehash->tuple.dir)
 314                         continue;
 315
 316                 flow = container_of(tuplehash, struct flow_offload, tuplehash[0]);
 317
 318                 iter(flow, data);
 319         }
 320         rhashtable_walk_stop(&hti);
 321         rhashtable_walk_exit(&hti);
 322
 323         return err;
 324 }
 325
 326 static void nf_flow_offload_gc_step(struct flow_offload *flow, void *data)
 327 {
 328         struct nf_flowtable *flow_table = data;
 329         struct flow_offload_entry *e;
 330
 331         e = container_of(flow, struct flow_offload_entry, flow);
 332         if (nf_flow_has_expired(flow) || nf_ct_is_dying(e->ct) ||
 333             (flow->flags & (FLOW_OFFLOAD_DYING | FLOW_OFFLOAD_TEARDOWN)))
 334                 flow_offload_del(flow_table, flow);
 335 }
 336
 337 static void nf_flow_offload_work_gc(struct work_struct *work)
 338 {
 339         struct nf_flowtable *flow_table;
 340
 341         flow_table = container_of(work, struct nf_flowtable, gc_work.work);
 342         nf_flow_table_iterate(flow_table, nf_flow_offload_gc_step, flow_table);
 343         queue_delayed_work(system_power_efficient_wq, &flow_table->gc_work, HZ);
 344 }
 345
 346 static int nf_flow_nat_port_tcp(struct sk_buff *skb, unsigned int thoff,
 347                                 __be16 port, __be16 new_port)
 348 {
 349         struct tcphdr *tcph;
 350
 351         if (!pskb_may_pull(skb, thoff + sizeof(*tcph)) ||
 352             skb_try_make_writable(skb, thoff + sizeof(*tcph)))
 353                 return -1;
 354
 355         tcph = (void *)(skb_network_header(skb) + thoff);
 356         inet_proto_csum_replace2(&tcph->check, skb, port, new_port, true);
 357
 358         return 0;
 359 }
 360
 361 static int nf_flow_nat_port_udp(struct sk_buff *skb, unsigned int thoff,
 362                                 __be16 port, __be16 new_port)
 363 {
 364         struct udphdr *udph;
 365
 366         if (!pskb_may_pull(skb, thoff + sizeof(*udph)) ||
 367             skb_try_make_writable(skb, thoff + sizeof(*udph)))
 368                 return -1;
 369
 370         udph = (void *)(skb_network_header(skb) + thoff);
 371         if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 372                 inet_proto_csum_replace2(&udph->check, skb, port,
 373                                          new_port, true);
 374                 if (!udph->check)
 375                         udph->check = CSUM_MANGLED_0;
 376         }
 377
 378         return 0;
 379 }
 380
 381 static int nf_flow_nat_port(struct sk_buff *skb, unsigned int thoff,
 382                             u8 protocol, __be16 port, __be16 new_port)
 383 {
 384         switch (protocol) {
 385         case IPPROTO_TCP:
 386                 if (nf_flow_nat_port_tcp(skb, thoff, port, new_port) < 0)
 387                         return NF_DROP;
 388                 break;
 389         case IPPROTO_UDP:
 390                 if (nf_flow_nat_port_udp(skb, thoff, port, new_port) < 0)
 391                         return NF_DROP;
 392                 break;
 393         }
 394
 395         return 0;
 396 }
 397
 398 int nf_flow_snat_port(const struct flow_offload *flow,
 399                       struct sk_buff *skb, unsigned int thoff,
 400                       u8 protocol, enum flow_offload_tuple_dir dir)
 401 {
 402         struct flow_ports *hdr;
 403         __be16 port, new_port;
 404
 405         if (!pskb_may_pull(skb, thoff + sizeof(*hdr)) ||
 406             skb_try_make_writable(skb, thoff + sizeof(*hdr)))
 407                 return -1;
 408
 409         hdr = (void *)(skb_network_header(skb) + thoff);
 410
 411         switch (dir) {
 412         case FLOW_OFFLOAD_DIR_ORIGINAL:
 413                 port = hdr->source;
 414                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_port;
 415                 hdr->source = new_port;
 416                 break;
 417         case FLOW_OFFLOAD_DIR_REPLY:
 418                 port = hdr->dest;
 419                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_port;
 420                 hdr->dest = new_port;
 421                 break;
 422         default:
 423                 return -1;
 424         }
 425
 426         return nf_flow_nat_port(skb, thoff, protocol, port, new_port);
 427 }
 428 EXPORT_SYMBOL_GPL(nf_flow_snat_port);
 429
 430 int nf_flow_dnat_port(const struct flow_offload *flow,
 431                       struct sk_buff *skb, unsigned int thoff,
 432                       u8 protocol, enum flow_offload_tuple_dir dir)
 433 {
 434         struct flow_ports *hdr;
 435         __be16 port, new_port;
 436
 437         if (!pskb_may_pull(skb, thoff + sizeof(*hdr)) ||
 438             skb_try_make_writable(skb, thoff + sizeof(*hdr)))
 439                 return -1;
 440
 441         hdr = (void *)(skb_network_header(skb) + thoff);
 442
 443         switch (dir) {
 444         case FLOW_OFFLOAD_DIR_ORIGINAL:
 445                 port = hdr->dest;
 446                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_port;
 447                 hdr->dest = new_port;
 448                 break;
 449         case FLOW_OFFLOAD_DIR_REPLY:
 450                 port = hdr->source;
 451                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_port;
 452                 hdr->source = new_port;
 453                 break;
 454         default:
 455                 return -1;
 456         }
 457
 458         return nf_flow_nat_port(skb, thoff, protocol, port, new_port);
 459 }
 460 EXPORT_SYMBOL_GPL(nf_flow_dnat_port);
 461
 462 int nf_flow_table_init(struct nf_flowtable *flowtable)
 463 {
 464         int err;
 465
 466         INIT_DEFERRABLE_WORK(&flowtable->gc_work, nf_flow_offload_work_gc);
 467
 468         err = rhashtable_init(&flowtable->rhashtable,
 469                               &nf_flow_offload_rhash_params);
 470         if (err < 0)
 471                 return err;
 472
 473         queue_delayed_work(system_power_efficient_wq,
 474                            &flowtable->gc_work, HZ);
 475
 476         mutex_lock(&flowtable_lock);
 477         list_add(&flowtable->list, &flowtables);
 478         mutex_unlock(&flowtable_lock);
 479
 480         return 0;
 481 }
 482 EXPORT_SYMBOL_GPL(nf_flow_table_init);
 483
 484 static void nf_flow_table_do_cleanup(struct flow_offload *flow, void *data)
 485 {
 486         struct net_device *dev = data;
 487         struct flow_offload_entry *e;
 488
 489         e = container_of(flow, struct flow_offload_entry, flow);
 490
 491         if (!dev) {
 492                 flow_offload_teardown(flow);
 493                 return;
 494         }
 495         if (net_eq(nf_ct_net(e->ct), dev_net(dev)) &&
 496             (flow->tuplehash[0].tuple.iifidx == dev->ifindex ||
 497              flow->tuplehash[1].tuple.iifidx == dev->ifindex))
 498                 flow_offload_dead(flow);
 499 }
 500
 501 static void nf_flow_table_iterate_cleanup(struct nf_flowtable *flowtable,
 502                                           struct net_device *dev)
 503 {
 504         nf_flow_table_iterate(flowtable, nf_flow_table_do_cleanup, dev);
 505         flush_delayed_work(&flowtable->gc_work);
 506 }
 507
 508 void nf_flow_table_cleanup(struct net_device *dev)
 509 {
 510         struct nf_flowtable *flowtable;
 511
 512         mutex_lock(&flowtable_lock);
 513         list_for_each_entry(flowtable, &flowtables, list)
 514                 nf_flow_table_iterate_cleanup(flowtable, dev);
 515         mutex_unlock(&flowtable_lock);
 516 }
 517 EXPORT_SYMBOL_GPL(nf_flow_table_cleanup);
 518
 519 void nf_flow_table_free(struct nf_flowtable *flow_table)
 520 {
 521         mutex_lock(&flowtable_lock);
 522         list_del(&flow_table->list);
 523         mutex_unlock(&flowtable_lock);
 524         cancel_delayed_work_sync(&flow_table->gc_work);
 525         nf_flow_table_iterate(flow_table, nf_flow_table_do_cleanup, NULL);
 526         nf_flow_table_iterate(flow_table, nf_flow_offload_gc_step, flow_table);
 527         rhashtable_destroy(&flow_table->rhashtable);
 528 }
 529 EXPORT_SYMBOL_GPL(nf_flow_table_free);
 530
 531 MODULE_LICENSE("GPL");
 532 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");