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