target/linux/generic/hack-5.10/650-netfilter-add-xt_FLOWOFFLOAD-target.patch

   1 From: Felix Fietkau <nbd@nbd.name>
   2 Date: Tue, 20 Feb 2018 15:56:02 +0100
   3 Subject: [PATCH] netfilter: add xt_FLOWOFFLOAD target
   4
   5 Signed-off-by: Felix Fietkau <nbd@nbd.name>
   6 ---
   7  create mode 100644 net/netfilter/xt_OFFLOAD.c
   8
   9 --- a/net/ipv4/netfilter/Kconfig
  10 +++ b/net/ipv4/netfilter/Kconfig
  11 @@ -56,8 +56,6 @@ config NF_TABLES_ARP
  12         help
  13           This option enables the ARP support for nf_tables.
  14
  15 -endif # NF_TABLES
  16 -
  17  config NF_FLOW_TABLE_IPV4
  18         tristate "Netfilter flow table IPv4 module"
  19         depends on NF_FLOW_TABLE
  20 @@ -66,6 +64,8 @@ config NF_FLOW_TABLE_IPV4
  21
  22           To compile it as a module, choose M here.
  23
  24 +endif # NF_TABLES
  25 +
  26  config NF_DUP_IPV4
  27         tristate "Netfilter IPv4 packet duplication to alternate destination"
  28         depends on !NF_CONNTRACK || NF_CONNTRACK
  29 --- a/net/ipv6/netfilter/Kconfig
  30 +++ b/net/ipv6/netfilter/Kconfig
  31 @@ -45,7 +45,6 @@ config NFT_FIB_IPV6
  32           multicast or blackhole.
  33
  34  endif # NF_TABLES_IPV6
  35 -endif # NF_TABLES
  36
  37  config NF_FLOW_TABLE_IPV6
  38         tristate "Netfilter flow table IPv6 module"
  39 @@ -55,6 +54,8 @@ config NF_FLOW_TABLE_IPV6
  40
  41           To compile it as a module, choose M here.
  42
  43 +endif # NF_TABLES
  44 +
  45  config NF_DUP_IPV6
  46         tristate "Netfilter IPv6 packet duplication to alternate destination"
  47         depends on !NF_CONNTRACK || NF_CONNTRACK
  48 --- a/net/netfilter/Kconfig
  49 +++ b/net/netfilter/Kconfig
  50 @@ -683,8 +683,6 @@ config NFT_FIB_NETDEV
  51
  52  endif # NF_TABLES_NETDEV
  53
  54 -endif # NF_TABLES
  55 -
  56  config NF_FLOW_TABLE_INET
  57         tristate "Netfilter flow table mixed IPv4/IPv6 module"
  58         depends on NF_FLOW_TABLE
  59 @@ -693,11 +691,12 @@ config NF_FLOW_TABLE_INET
  60
  61           To compile it as a module, choose M here.
  62
  63 +endif # NF_TABLES
  64 +
  65  config NF_FLOW_TABLE
  66         tristate "Netfilter flow table module"
  67         depends on NETFILTER_INGRESS
  68         depends on NF_CONNTRACK
  69 -       depends on NF_TABLES
  70         help
  71           This option adds the flow table core infrastructure.
  72
  73 @@ -977,6 +976,15 @@ config NETFILTER_XT_TARGET_NOTRACK
  74         depends on NETFILTER_ADVANCED
  75         select NETFILTER_XT_TARGET_CT
  76
  77 +config NETFILTER_XT_TARGET_FLOWOFFLOAD
  78 +       tristate '"FLOWOFFLOAD" target support'
  79 +       depends on NF_FLOW_TABLE
  80 +       depends on NETFILTER_INGRESS
  81 +       help
  82 +         This option adds a `FLOWOFFLOAD' target, which uses the nf_flow_offload
  83 +         module to speed up processing of packets by bypassing the usual
  84 +         netfilter chains
  85 +
  86  config NETFILTER_XT_TARGET_RATEEST
  87         tristate '"RATEEST" target support'
  88         depends on NETFILTER_ADVANCED
  89 --- a/net/netfilter/Makefile
  90 +++ b/net/netfilter/Makefile
  91 @@ -145,6 +145,7 @@ obj-$(CONFIG_NETFILTER_XT_TARGET_CLASSIF
  92  obj-$(CONFIG_NETFILTER_XT_TARGET_CONNSECMARK) += xt_CONNSECMARK.o
  93  obj-$(CONFIG_NETFILTER_XT_TARGET_CT) += xt_CT.o
  94  obj-$(CONFIG_NETFILTER_XT_TARGET_DSCP) += xt_DSCP.o
  95 +obj-$(CONFIG_NETFILTER_XT_TARGET_FLOWOFFLOAD) += xt_FLOWOFFLOAD.o
  96  obj-$(CONFIG_NETFILTER_XT_TARGET_HL) += xt_HL.o
  97  obj-$(CONFIG_NETFILTER_XT_TARGET_HMARK) += xt_HMARK.o
  98  obj-$(CONFIG_NETFILTER_XT_TARGET_LED) += xt_LED.o
  99 --- /dev/null
 100 +++ b/net/netfilter/xt_FLOWOFFLOAD.c
 101 @@ -0,0 +1,658 @@
 102 +/*
 103 + * Copyright (C) 2018-2021 Felix Fietkau <nbd@nbd.name>
 104 + *
 105 + * This program is free software; you can redistribute it and/or modify
 106 + * it under the terms of the GNU General Public License version 2 as
 107 + * published by the Free Software Foundation.
 108 + */
 109 +#include <linux/module.h>
 110 +#include <linux/init.h>
 111 +#include <linux/netfilter.h>
 112 +#include <linux/netfilter/xt_FLOWOFFLOAD.h>
 113 +#include <net/ip.h>
 114 +#include <net/netfilter/nf_conntrack.h>
 115 +#include <net/netfilter/nf_conntrack_extend.h>
 116 +#include <net/netfilter/nf_conntrack_helper.h>
 117 +#include <net/netfilter/nf_flow_table.h>
 118 +
 119 +struct xt_flowoffload_hook {
 120 +       struct hlist_node list;
 121 +       struct nf_hook_ops ops;
 122 +       struct net *net;
 123 +       bool registered;
 124 +       bool used;
 125 +};
 126 +
 127 +struct xt_flowoffload_table {
 128 +       struct nf_flowtable ft;
 129 +       struct hlist_head hooks;
 130 +       struct delayed_work work;
 131 +};
 132 +
 133 +static DEFINE_SPINLOCK(hooks_lock);
 134 +
 135 +struct xt_flowoffload_table flowtable[2];
 136 +
 137 +static unsigned int
 138 +xt_flowoffload_net_hook(void *priv, struct sk_buff *skb,
 139 +                       const struct nf_hook_state *state)
 140 +{
 141 +       struct nf_flowtable *ft = priv;
 142 +
 143 +       if (!atomic_read(&ft->rhashtable.nelems))
 144 +               return NF_ACCEPT;
 145 +
 146 +       switch (skb->protocol) {
 147 +       case htons(ETH_P_IP):
 148 +               return nf_flow_offload_ip_hook(priv, skb, state);
 149 +       case htons(ETH_P_IPV6):
 150 +               return nf_flow_offload_ipv6_hook(priv, skb, state);
 151 +       }
 152 +
 153 +       return NF_ACCEPT;
 154 +}
 155 +
 156 +static int
 157 +xt_flowoffload_create_hook(struct xt_flowoffload_table *table,
 158 +                          struct net_device *dev)
 159 +{
 160 +       struct xt_flowoffload_hook *hook;
 161 +       struct nf_hook_ops *ops;
 162 +
 163 +       hook = kzalloc(sizeof(*hook), GFP_ATOMIC);
 164 +       if (!hook)
 165 +               return -ENOMEM;
 166 +
 167 +       ops = &hook->ops;
 168 +       ops->pf = NFPROTO_NETDEV;
 169 +       ops->hooknum = NF_NETDEV_INGRESS;
 170 +       ops->priority = 10;
 171 +       ops->priv = &table->ft;
 172 +       ops->hook = xt_flowoffload_net_hook;
 173 +       ops->dev = dev;
 174 +
 175 +       hlist_add_head(&hook->list, &table->hooks);
 176 +       mod_delayed_work(system_power_efficient_wq, &table->work, 0);
 177 +
 178 +       return 0;
 179 +}
 180 +
 181 +static struct xt_flowoffload_hook *
 182 +flow_offload_lookup_hook(struct xt_flowoffload_table *table,
 183 +                        struct net_device *dev)
 184 +{
 185 +       struct xt_flowoffload_hook *hook;
 186 +
 187 +       hlist_for_each_entry(hook, &table->hooks, list) {
 188 +               if (hook->ops.dev == dev)
 189 +                       return hook;
 190 +       }
 191 +
 192 +       return NULL;
 193 +}
 194 +
 195 +static void
 196 +xt_flowoffload_check_device(struct xt_flowoffload_table *table,
 197 +                           struct net_device *dev)
 198 +{
 199 +       struct xt_flowoffload_hook *hook;
 200 +
 201 +       if (!dev)
 202 +               return;
 203 +
 204 +       spin_lock_bh(&hooks_lock);
 205 +       hook = flow_offload_lookup_hook(table, dev);
 206 +       if (hook)
 207 +               hook->used = true;
 208 +       else
 209 +               xt_flowoffload_create_hook(table, dev);
 210 +       spin_unlock_bh(&hooks_lock);
 211 +}
 212 +
 213 +static void
 214 +xt_flowoffload_register_hooks(struct xt_flowoffload_table *table)
 215 +{
 216 +       struct xt_flowoffload_hook *hook;
 217 +
 218 +restart:
 219 +       hlist_for_each_entry(hook, &table->hooks, list) {
 220 +               if (hook->registered)
 221 +                       continue;
 222 +
 223 +               hook->registered = true;
 224 +               hook->net = dev_net(hook->ops.dev);
 225 +               spin_unlock_bh(&hooks_lock);
 226 +               nf_register_net_hook(hook->net, &hook->ops);
 227 +               if (table->ft.flags & NF_FLOWTABLE_HW_OFFLOAD)
 228 +                       table->ft.type->setup(&table->ft, hook->ops.dev,
 229 +                                             FLOW_BLOCK_BIND);
 230 +               spin_lock_bh(&hooks_lock);
 231 +               goto restart;
 232 +       }
 233 +
 234 +}
 235 +
 236 +static bool
 237 +xt_flowoffload_cleanup_hooks(struct xt_flowoffload_table *table)
 238 +{
 239 +       struct xt_flowoffload_hook *hook;
 240 +       bool active = false;
 241 +
 242 +restart:
 243 +       spin_lock_bh(&hooks_lock);
 244 +       hlist_for_each_entry(hook, &table->hooks, list) {
 245 +               if (hook->used || !hook->registered) {
 246 +                       active = true;
 247 +                       continue;
 248 +               }
 249 +
 250 +               hlist_del(&hook->list);
 251 +               spin_unlock_bh(&hooks_lock);
 252 +               if (table->ft.flags & NF_FLOWTABLE_HW_OFFLOAD)
 253 +                       table->ft.type->setup(&table->ft, hook->ops.dev,
 254 +                                             FLOW_BLOCK_UNBIND);
 255 +               nf_unregister_net_hook(hook->net, &hook->ops);
 256 +               kfree(hook);
 257 +               goto restart;
 258 +       }
 259 +       spin_unlock_bh(&hooks_lock);
 260 +
 261 +       return active;
 262 +}
 263 +
 264 +static void
 265 +xt_flowoffload_check_hook(struct flow_offload *flow, void *data)
 266 +{
 267 +       struct xt_flowoffload_table *table = data;
 268 +       struct flow_offload_tuple *tuple0 = &flow->tuplehash[0].tuple;
 269 +       struct flow_offload_tuple *tuple1 = &flow->tuplehash[1].tuple;
 270 +       struct xt_flowoffload_hook *hook;
 271 +
 272 +       spin_lock_bh(&hooks_lock);
 273 +       hlist_for_each_entry(hook, &table->hooks, list) {
 274 +               if (hook->ops.dev->ifindex != tuple0->iifidx &&
 275 +                   hook->ops.dev->ifindex != tuple1->iifidx)
 276 +                       continue;
 277 +
 278 +               hook->used = true;
 279 +       }
 280 +       spin_unlock_bh(&hooks_lock);
 281 +
 282 +       cond_resched();
 283 +}
 284 +
 285 +static void
 286 +xt_flowoffload_hook_work(struct work_struct *work)
 287 +{
 288 +       struct xt_flowoffload_table *table;
 289 +       struct xt_flowoffload_hook *hook;
 290 +       int err;
 291 +
 292 +       table = container_of(work, struct xt_flowoffload_table, work.work);
 293 +
 294 +       spin_lock_bh(&hooks_lock);
 295 +       xt_flowoffload_register_hooks(table);
 296 +       hlist_for_each_entry(hook, &table->hooks, list)
 297 +               hook->used = false;
 298 +       spin_unlock_bh(&hooks_lock);
 299 +
 300 +       err = nf_flow_table_iterate(&table->ft, xt_flowoffload_check_hook,
 301 +                                   table);
 302 +       if (err && err != -EAGAIN)
 303 +               goto out;
 304 +
 305 +       if (!xt_flowoffload_cleanup_hooks(table))
 306 +               return;
 307 +
 308 +out:
 309 +       queue_delayed_work(system_power_efficient_wq, &table->work, HZ);
 310 +}
 311 +
 312 +static bool
 313 +xt_flowoffload_skip(struct sk_buff *skb, int family)
 314 +{
 315 +       if (skb_sec_path(skb))
 316 +               return true;
 317 +
 318 +       if (family == NFPROTO_IPV4) {
 319 +               const struct ip_options *opt = &(IPCB(skb)->opt);
 320 +
 321 +               if (unlikely(opt->optlen))
 322 +                       return true;
 323 +       }
 324 +
 325 +       return false;
 326 +}
 327 +
 328 +static bool flow_is_valid_ether_device(const struct net_device *dev)
 329 +{
 330 +       if (!dev || (dev->flags & IFF_LOOPBACK) || dev->type != ARPHRD_ETHER ||
 331 +           dev->addr_len != ETH_ALEN || !is_valid_ether_addr(dev->dev_addr))
 332 +               return false;
 333 +
 334 +       return true;
 335 +}
 336 +
 337 +static void
 338 +xt_flowoffload_route_check_path(struct nf_flow_route *route,
 339 +                               const struct nf_conn *ct,
 340 +                               enum ip_conntrack_dir dir,
 341 +                               struct net_device **out_dev)
 342 +{
 343 +       const struct dst_entry *dst = route->tuple[dir].dst;
 344 +       const void *daddr = &ct->tuplehash[!dir].tuple.src.u3;
 345 +       struct net_device_path_stack stack;
 346 +       enum net_device_path_type prev_type;
 347 +       struct net_device *dev = dst->dev;
 348 +       struct neighbour *n;
 349 +       bool last = false;
 350 +       u8 nud_state;
 351 +       int i;
 352 +
 353 +       route->tuple[!dir].in.ifindex = dev->ifindex;
 354 +       route->tuple[dir].out.ifindex = dev->ifindex;
 355 +
 356 +       if (route->tuple[dir].xmit_type == FLOW_OFFLOAD_XMIT_XFRM)
 357 +               return;
 358 +
 359 +       if ((dev->flags & IFF_LOOPBACK) ||
 360 +           dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN ||
 361 +           !is_valid_ether_addr(dev->dev_addr))
 362 +               return;
 363 +
 364 +       n = dst_neigh_lookup(dst, daddr);
 365 +       if (!n)
 366 +               return;
 367 +
 368 +       read_lock_bh(&n->lock);
 369 +       nud_state = n->nud_state;
 370 +       memcpy(route->tuple[dir].out.h_dest, n->ha, ETH_ALEN);
 371 +       read_unlock_bh(&n->lock);
 372 +       neigh_release(n);
 373 +
 374 +       if (!(nud_state & NUD_VALID))
 375 +               return;
 376 +
 377 +       if (dev_fill_forward_path(dev, route->tuple[dir].out.h_dest, &stack) ||
 378 +           !stack.num_paths)
 379 +               return;
 380 +
 381 +       prev_type = DEV_PATH_ETHERNET;
 382 +       for (i = 0; i <= stack.num_paths; i++) {
 383 +               const struct net_device_path *path = &stack.path[i];
 384 +               int n_encaps = route->tuple[!dir].in.num_encaps;
 385 +
 386 +               dev = (struct net_device *)path->dev;
 387 +               if (flow_is_valid_ether_device(dev)) {
 388 +                       if (route->tuple[dir].xmit_type != FLOW_OFFLOAD_XMIT_DIRECT) {
 389 +                               memcpy(route->tuple[dir].out.h_source,
 390 +                                      dev->dev_addr, ETH_ALEN);
 391 +                               route->tuple[dir].out.ifindex = dev->ifindex;
 392 +                       }
 393 +                       route->tuple[dir].xmit_type = FLOW_OFFLOAD_XMIT_DIRECT;
 394 +               }
 395 +
 396 +               switch (path->type) {
 397 +               case DEV_PATH_PPPOE:
 398 +               case DEV_PATH_VLAN:
 399 +                       if (n_encaps >= NF_FLOW_TABLE_ENCAP_MAX ||
 400 +                           i == stack.num_paths) {
 401 +                               last = true;
 402 +                               break;
 403 +                       }
 404 +
 405 +                       route->tuple[!dir].in.num_encaps++;
 406 +                       route->tuple[!dir].in.encap[n_encaps].id = path->encap.id;
 407 +                       route->tuple[!dir].in.encap[n_encaps].proto = path->encap.proto;
 408 +                       if (path->type == DEV_PATH_PPPOE)
 409 +                               memcpy(route->tuple[dir].out.h_dest,
 410 +                                      path->encap.h_dest, ETH_ALEN);
 411 +                       break;
 412 +               case DEV_PATH_BRIDGE:
 413 +                       switch (path->bridge.vlan_mode) {
 414 +                       case DEV_PATH_BR_VLAN_TAG:
 415 +                               if (n_encaps >= NF_FLOW_TABLE_ENCAP_MAX ||
 416 +                                   i == stack.num_paths) {
 417 +                                       last = true;
 418 +                                       break;
 419 +                               }
 420 +
 421 +                               route->tuple[!dir].in.num_encaps++;
 422 +                               route->tuple[!dir].in.encap[n_encaps].id =
 423 +                                       path->bridge.vlan_id;
 424 +                               route->tuple[!dir].in.encap[n_encaps].proto =
 425 +                                       path->bridge.vlan_proto;
 426 +                               break;
 427 +                       case DEV_PATH_BR_VLAN_UNTAG:
 428 +                               route->tuple[!dir].in.num_encaps--;
 429 +                               break;
 430 +                       case DEV_PATH_BR_VLAN_UNTAG_HW:
 431 +                               route->tuple[!dir].in.ingress_vlans |= BIT(n_encaps - 1);
 432 +                               break;
 433 +                       case DEV_PATH_BR_VLAN_KEEP:
 434 +                               break;
 435 +                       }
 436 +                       break;
 437 +               default:
 438 +                       last = true;
 439 +                       break;
 440 +               }
 441 +
 442 +               if (last)
 443 +                       break;
 444 +       }
 445 +
 446 +       *out_dev = dev;
 447 +       route->tuple[dir].out.hw_ifindex = dev->ifindex;
 448 +       route->tuple[!dir].in.ifindex = dev->ifindex;
 449 +}
 450 +
 451 +static int
 452 +xt_flowoffload_route_dir(struct nf_flow_route *route, const struct nf_conn *ct,
 453 +                        enum ip_conntrack_dir dir,
 454 +                        const struct xt_action_param *par, int ifindex)
 455 +{
 456 +       struct dst_entry *dst = NULL;
 457 +       struct flowi fl;
 458 +
 459 +       memset(&fl, 0, sizeof(fl));
 460 +       switch (xt_family(par)) {
 461 +       case NFPROTO_IPV4:
 462 +               fl.u.ip4.daddr = ct->tuplehash[!dir].tuple.src.u3.ip;
 463 +               fl.u.ip4.flowi4_oif = ifindex;
 464 +               break;
 465 +       case NFPROTO_IPV6:
 466 +               fl.u.ip6.saddr = ct->tuplehash[!dir].tuple.dst.u3.in6;
 467 +               fl.u.ip6.daddr = ct->tuplehash[!dir].tuple.src.u3.in6;
 468 +               fl.u.ip6.flowi6_oif = ifindex;
 469 +               break;
 470 +       }
 471 +
 472 +       nf_route(xt_net(par), &dst, &fl, false, xt_family(par));
 473 +       if (!dst)
 474 +               return -ENOENT;
 475 +
 476 +       route->tuple[dir].dst = dst;
 477 +       if (dst_xfrm(dst))
 478 +               route->tuple[dir].xmit_type = FLOW_OFFLOAD_XMIT_XFRM;
 479 +       else
 480 +               route->tuple[dir].xmit_type = FLOW_OFFLOAD_XMIT_NEIGH;
 481 +
 482 +       return 0;
 483 +}
 484 +
 485 +static int
 486 +xt_flowoffload_route(struct sk_buff *skb, const struct nf_conn *ct,
 487 +                    const struct xt_action_param *par,
 488 +                    struct nf_flow_route *route, enum ip_conntrack_dir dir,
 489 +                    struct net_device **dev)
 490 +{
 491 +       int ret;
 492 +
 493 +       ret = xt_flowoffload_route_dir(route, ct, dir, par,
 494 +                                      dev[dir]->ifindex);
 495 +       if (ret)
 496 +               return ret;
 497 +
 498 +       ret = xt_flowoffload_route_dir(route, ct, !dir, par,
 499 +                                      dev[!dir]->ifindex);
 500 +       if (ret)
 501 +               return ret;
 502 +
 503 +       xt_flowoffload_route_check_path(route, ct, dir, &dev[!dir]);
 504 +       xt_flowoffload_route_check_path(route, ct, !dir, &dev[dir]);
 505 +
 506 +       return 0;
 507 +}
 508 +
 509 +static unsigned int
 510 +flowoffload_tg(struct sk_buff *skb, const struct xt_action_param *par)
 511 +{
 512 +       struct xt_flowoffload_table *table;
 513 +       const struct xt_flowoffload_target_info *info = par->targinfo;
 514 +       struct tcphdr _tcph, *tcph = NULL;
 515 +       enum ip_conntrack_info ctinfo;
 516 +       enum ip_conntrack_dir dir;
 517 +       struct nf_flow_route route = {};
 518 +       struct flow_offload *flow = NULL;
 519 +       struct net_device *devs[2] = {};
 520 +       struct nf_conn *ct;
 521 +       struct net *net;
 522 +
 523 +       if (xt_flowoffload_skip(skb, xt_family(par)))
 524 +               return XT_CONTINUE;
 525 +
 526 +       ct = nf_ct_get(skb, &ctinfo);
 527 +       if (ct == NULL)
 528 +               return XT_CONTINUE;
 529 +
 530 +       switch (ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.protonum) {
 531 +       case IPPROTO_TCP:
 532 +               if (ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED)
 533 +                       return XT_CONTINUE;
 534 +
 535 +               tcph = skb_header_pointer(skb, par->thoff,
 536 +                                         sizeof(_tcph), &_tcph);
 537 +               if (unlikely(!tcph || tcph->fin || tcph->rst))
 538 +                       return XT_CONTINUE;
 539 +               break;
 540 +       case IPPROTO_UDP:
 541 +               break;
 542 +       default:
 543 +               return XT_CONTINUE;
 544 +       }
 545 +
 546 +       if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
 547 +           ct->status & IPS_SEQ_ADJUST)
 548 +               return XT_CONTINUE;
 549 +
 550 +       if (!nf_ct_is_confirmed(ct))
 551 +               return XT_CONTINUE;
 552 +
 553 +       devs[dir] = xt_out(par);
 554 +       devs[!dir] = xt_in(par);
 555 +
 556 +       if (!devs[dir] || !devs[!dir])
 557 +               return XT_CONTINUE;
 558 +
 559 +       if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
 560 +               return XT_CONTINUE;
 561 +
 562 +       dir = CTINFO2DIR(ctinfo);
 563 +
 564 +       if (xt_flowoffload_route(skb, ct, par, &route, dir, devs) < 0)
 565 +               goto err_flow_route;
 566 +
 567 +       flow = flow_offload_alloc(ct);
 568 +       if (!flow)
 569 +               goto err_flow_alloc;
 570 +
 571 +       if (flow_offload_route_init(flow, &route) < 0)
 572 +               goto err_flow_add;
 573 +
 574 +       if (tcph) {
 575 +               ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
 576 +               ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
 577 +       }
 578 +
 579 +       table = &flowtable[!!(info->flags & XT_FLOWOFFLOAD_HW)];
 580 +       if (flow_offload_add(&table->ft, flow) < 0)
 581 +               goto err_flow_add;
 582 +
 583 +       xt_flowoffload_check_device(table, devs[0]);
 584 +       xt_flowoffload_check_device(table, devs[1]);
 585 +
 586 +       net = read_pnet(&table->ft.net);
 587 +       if (!net)
 588 +               write_pnet(&table->ft.net, xt_net(par));
 589 +
 590 +       dst_release(route.tuple[dir].dst);
 591 +       dst_release(route.tuple[!dir].dst);
 592 +
 593 +       return XT_CONTINUE;
 594 +
 595 +err_flow_add:
 596 +       flow_offload_free(flow);
 597 +err_flow_alloc:
 598 +       dst_release(route.tuple[dir].dst);
 599 +       dst_release(route.tuple[!dir].dst);
 600 +err_flow_route:
 601 +       clear_bit(IPS_OFFLOAD_BIT, &ct->status);
 602 +
 603 +       return XT_CONTINUE;
 604 +}
 605 +
 606 +static int flowoffload_chk(const struct xt_tgchk_param *par)
 607 +{
 608 +       struct xt_flowoffload_target_info *info = par->targinfo;
 609 +
 610 +       if (info->flags & ~XT_FLOWOFFLOAD_MASK)
 611 +               return -EINVAL;
 612 +
 613 +       return 0;
 614 +}
 615 +
 616 +static struct xt_target offload_tg_reg __read_mostly = {
 617 +       .family         = NFPROTO_UNSPEC,
 618 +       .name           = "FLOWOFFLOAD",
 619 +       .revision       = 0,
 620 +       .targetsize     = sizeof(struct xt_flowoffload_target_info),
 621 +       .usersize       = sizeof(struct xt_flowoffload_target_info),
 622 +       .checkentry     = flowoffload_chk,
 623 +       .target         = flowoffload_tg,
 624 +       .me             = THIS_MODULE,
 625 +};
 626 +
 627 +static int flow_offload_netdev_event(struct notifier_block *this,
 628 +                                    unsigned long event, void *ptr)
 629 +{
 630 +       struct xt_flowoffload_hook *hook0, *hook1;
 631 +       struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 632 +
 633 +       if (event != NETDEV_UNREGISTER)
 634 +               return NOTIFY_DONE;
 635 +
 636 +       spin_lock_bh(&hooks_lock);
 637 +       hook0 = flow_offload_lookup_hook(&flowtable[0], dev);
 638 +       if (hook0)
 639 +               hlist_del(&hook0->list);
 640 +
 641 +       hook1 = flow_offload_lookup_hook(&flowtable[1], dev);
 642 +       if (hook1)
 643 +               hlist_del(&hook1->list);
 644 +       spin_unlock_bh(&hooks_lock);
 645 +
 646 +       if (hook0) {
 647 +               nf_unregister_net_hook(hook0->net, &hook0->ops);
 648 +               kfree(hook0);
 649 +       }
 650 +
 651 +       if (hook1) {
 652 +               nf_unregister_net_hook(hook1->net, &hook1->ops);
 653 +               kfree(hook1);
 654 +       }
 655 +
 656 +       nf_flow_table_cleanup(dev);
 657 +
 658 +       return NOTIFY_DONE;
 659 +}
 660 +
 661 +static struct notifier_block flow_offload_netdev_notifier = {
 662 +       .notifier_call  = flow_offload_netdev_event,
 663 +};
 664 +
 665 +static unsigned int
 666 +nf_flow_offload_inet_hook(void *priv, struct sk_buff *skb,
 667 +                         const struct nf_hook_state *state)
 668 +{
 669 +       switch (skb->protocol) {
 670 +       case htons(ETH_P_IP):
 671 +               return nf_flow_offload_ip_hook(priv, skb, state);
 672 +       case htons(ETH_P_IPV6):
 673 +               return nf_flow_offload_ipv6_hook(priv, skb, state);
 674 +       }
 675 +
 676 +       return NF_ACCEPT;
 677 +}
 678 +
 679 +static int nf_flow_rule_route_inet(struct net *net,
 680 +                                  const struct flow_offload *flow,
 681 +                                  enum flow_offload_tuple_dir dir,
 682 +                                  struct nf_flow_rule *flow_rule)
 683 +{
 684 +       const struct flow_offload_tuple *flow_tuple = &flow->tuplehash[dir].tuple;
 685 +       int err;
 686 +
 687 +       switch (flow_tuple->l3proto) {
 688 +       case NFPROTO_IPV4:
 689 +               err = nf_flow_rule_route_ipv4(net, flow, dir, flow_rule);
 690 +               break;
 691 +       case NFPROTO_IPV6:
 692 +               err = nf_flow_rule_route_ipv6(net, flow, dir, flow_rule);
 693 +               break;
 694 +       default:
 695 +               err = -1;
 696 +               break;
 697 +       }
 698 +
 699 +       return err;
 700 +}
 701 +
 702 +static struct nf_flowtable_type flowtable_inet = {
 703 +       .family         = NFPROTO_INET,
 704 +       .init           = nf_flow_table_init,
 705 +       .setup          = nf_flow_table_offload_setup,
 706 +       .action         = nf_flow_rule_route_inet,
 707 +       .free           = nf_flow_table_free,
 708 +       .hook           = nf_flow_offload_inet_hook,
 709 +       .owner          = THIS_MODULE,
 710 +};
 711 +
 712 +static int init_flowtable(struct xt_flowoffload_table *tbl)
 713 +{
 714 +       INIT_DELAYED_WORK(&tbl->work, xt_flowoffload_hook_work);
 715 +       tbl->ft.type = &flowtable_inet;
 716 +
 717 +       return nf_flow_table_init(&tbl->ft);
 718 +}
 719 +
 720 +static int __init xt_flowoffload_tg_init(void)
 721 +{
 722 +       int ret;
 723 +
 724 +       register_netdevice_notifier(&flow_offload_netdev_notifier);
 725 +
 726 +       ret = init_flowtable(&flowtable[0]);
 727 +       if (ret)
 728 +               return ret;
 729 +
 730 +       ret = init_flowtable(&flowtable[1]);
 731 +       if (ret)
 732 +               goto cleanup;
 733 +
 734 +       flowtable[1].ft.flags = NF_FLOWTABLE_HW_OFFLOAD;
 735 +
 736 +       ret = xt_register_target(&offload_tg_reg);
 737 +       if (ret)
 738 +               goto cleanup2;
 739 +
 740 +       return 0;
 741 +
 742 +cleanup2:
 743 +       nf_flow_table_free(&flowtable[1].ft);
 744 +cleanup:
 745 +       nf_flow_table_free(&flowtable[0].ft);
 746 +       return ret;
 747 +}
 748 +
 749 +static void __exit xt_flowoffload_tg_exit(void)
 750 +{
 751 +       xt_unregister_target(&offload_tg_reg);
 752 +       unregister_netdevice_notifier(&flow_offload_netdev_notifier);
 753 +       nf_flow_table_free(&flowtable[0].ft);
 754 +       nf_flow_table_free(&flowtable[1].ft);
 755 +}
 756 +
 757 +MODULE_LICENSE("GPL");
 758 +module_init(xt_flowoffload_tg_init);
 759 +module_exit(xt_flowoffload_tg_exit);
 760 --- a/net/netfilter/nf_flow_table_core.c
 761 +++ b/net/netfilter/nf_flow_table_core.c
 762 @@ -7,7 +7,6 @@
 763  #include <linux/netdevice.h>
 764  #include <net/ip.h>
 765  #include <net/ip6_route.h>
 766 -#include <net/netfilter/nf_tables.h>
 767  #include <net/netfilter/nf_flow_table.h>
 768  #include <net/netfilter/nf_conntrack.h>
 769  #include <net/netfilter/nf_conntrack_core.h>
 770 @@ -356,8 +355,7 @@ flow_offload_lookup(struct nf_flowtable
 771  }
 772  EXPORT_SYMBOL_GPL(flow_offload_lookup);
 773
 774 -static int
 775 -nf_flow_table_iterate(struct nf_flowtable *flow_table,
 776 +int nf_flow_table_iterate(struct nf_flowtable *flow_table,
 777                       void (*iter)(struct flow_offload *flow, void *data),
 778                       void *data)
 779  {
 780 @@ -389,6 +387,7 @@ nf_flow_table_iterate(struct nf_flowtabl
 781
 782         return err;
 783  }
 784 +EXPORT_SYMBOL_GPL(nf_flow_table_iterate);
 785
 786  static void nf_flow_offload_gc_step(struct flow_offload *flow, void *data)
 787  {
 788 --- /dev/null
 789 +++ b/include/uapi/linux/netfilter/xt_FLOWOFFLOAD.h
 790 @@ -0,0 +1,17 @@
 791 +/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
 792 +#ifndef _XT_FLOWOFFLOAD_H
 793 +#define _XT_FLOWOFFLOAD_H
 794 +
 795 +#include <linux/types.h>
 796 +
 797 +enum {
 798 +       XT_FLOWOFFLOAD_HW       = 1 << 0,
 799 +
 800 +       XT_FLOWOFFLOAD_MASK     = XT_FLOWOFFLOAD_HW
 801 +};
 802 +
 803 +struct xt_flowoffload_target_info {
 804 +       __u32 flags;
 805 +};
 806 +
 807 +#endif /* _XT_FLOWOFFLOAD_H */
 808 --- a/include/net/netfilter/nf_flow_table.h
 809 +++ b/include/net/netfilter/nf_flow_table.h
 810 @@ -266,6 +266,10 @@ void nf_flow_table_free(struct nf_flowta
 811
 812  void flow_offload_teardown(struct flow_offload *flow);
 813
 814 +int nf_flow_table_iterate(struct nf_flowtable *flow_table,
 815 +                      void (*iter)(struct flow_offload *flow, void *data),
 816 +                      void *data);
 817 +
 818  int nf_flow_snat_port(const struct flow_offload *flow,
 819                       struct sk_buff *skb, unsigned int thoff,
 820                       u8 protocol, enum flow_offload_tuple_dir dir);