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 +1,656 @@
 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 +
 283 +static void
 284 +xt_flowoffload_hook_work(struct work_struct *work)
 285 +{
 286 +       struct xt_flowoffload_table *table;
 287 +       struct xt_flowoffload_hook *hook;
 288 +       int err;
 289 +
 290 +       table = container_of(work, struct xt_flowoffload_table, work.work);
 291 +
 292 +       spin_lock_bh(&hooks_lock);
 293 +       xt_flowoffload_register_hooks(table);
 294 +       hlist_for_each_entry(hook, &table->hooks, list)
 295 +               hook->used = false;
 296 +       spin_unlock_bh(&hooks_lock);
 297 +
 298 +       err = nf_flow_table_iterate(&table->ft, xt_flowoffload_check_hook,
 299 +                                   table);
 300 +       if (err && err != -EAGAIN)
 301 +               goto out;
 302 +
 303 +       if (!xt_flowoffload_cleanup_hooks(table))
 304 +               return;
 305 +
 306 +out:
 307 +       queue_delayed_work(system_power_efficient_wq, &table->work, HZ);
 308 +}
 309 +
 310 +static bool
 311 +xt_flowoffload_skip(struct sk_buff *skb, int family)
 312 +{
 313 +       if (skb_sec_path(skb))
 314 +               return true;
 315 +
 316 +       if (family == NFPROTO_IPV4) {
 317 +               const struct ip_options *opt = &(IPCB(skb)->opt);
 318 +
 319 +               if (unlikely(opt->optlen))
 320 +                       return true;
 321 +       }
 322 +
 323 +       return false;
 324 +}
 325 +
 326 +static bool flow_is_valid_ether_device(const struct net_device *dev)
 327 +{
 328 +       if (!dev || (dev->flags & IFF_LOOPBACK) || dev->type != ARPHRD_ETHER ||
 329 +           dev->addr_len != ETH_ALEN || !is_valid_ether_addr(dev->dev_addr))
 330 +               return false;
 331 +
 332 +       return true;
 333 +}
 334 +
 335 +static void
 336 +xt_flowoffload_route_check_path(struct nf_flow_route *route,
 337 +                               const struct nf_conn *ct,
 338 +                               enum ip_conntrack_dir dir,
 339 +                               struct net_device **out_dev)
 340 +{
 341 +       const struct dst_entry *dst = route->tuple[dir].dst;
 342 +       const void *daddr = &ct->tuplehash[!dir].tuple.src.u3;
 343 +       struct net_device_path_stack stack;
 344 +       enum net_device_path_type prev_type;
 345 +       struct net_device *dev = dst->dev;
 346 +       struct neighbour *n;
 347 +       bool last = false;
 348 +       u8 nud_state;
 349 +       int i;
 350 +
 351 +       route->tuple[!dir].in.ifindex = dev->ifindex;
 352 +       route->tuple[dir].out.ifindex = dev->ifindex;
 353 +
 354 +       if (route->tuple[dir].xmit_type == FLOW_OFFLOAD_XMIT_XFRM)
 355 +               return;
 356 +
 357 +       if ((dev->flags & IFF_LOOPBACK) ||
 358 +           dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN ||
 359 +           !is_valid_ether_addr(dev->dev_addr))
 360 +               return;
 361 +
 362 +       n = dst_neigh_lookup(dst, daddr);
 363 +       if (!n)
 364 +               return;
 365 +
 366 +       read_lock_bh(&n->lock);
 367 +       nud_state = n->nud_state;
 368 +       memcpy(route->tuple[dir].out.h_dest, n->ha, ETH_ALEN);
 369 +       read_unlock_bh(&n->lock);
 370 +       neigh_release(n);
 371 +
 372 +       if (!(nud_state & NUD_VALID))
 373 +               return;
 374 +
 375 +       if (dev_fill_forward_path(dev, route->tuple[dir].out.h_dest, &stack) ||
 376 +           !stack.num_paths)
 377 +               return;
 378 +
 379 +       prev_type = DEV_PATH_ETHERNET;
 380 +       for (i = 0; i <= stack.num_paths; i++) {
 381 +               const struct net_device_path *path = &stack.path[i];
 382 +               int n_encaps = route->tuple[!dir].in.num_encaps;
 383 +
 384 +               dev = (struct net_device *)path->dev;
 385 +               if (flow_is_valid_ether_device(dev)) {
 386 +                       if (route->tuple[dir].xmit_type != FLOW_OFFLOAD_XMIT_DIRECT) {
 387 +                               memcpy(route->tuple[dir].out.h_source,
 388 +                                      dev->dev_addr, ETH_ALEN);
 389 +                               route->tuple[dir].out.ifindex = dev->ifindex;
 390 +                       }
 391 +                       route->tuple[dir].xmit_type = FLOW_OFFLOAD_XMIT_DIRECT;
 392 +               }
 393 +
 394 +               switch (path->type) {
 395 +               case DEV_PATH_PPPOE:
 396 +               case DEV_PATH_VLAN:
 397 +                       if (n_encaps >= NF_FLOW_TABLE_ENCAP_MAX ||
 398 +                           i == stack.num_paths) {
 399 +                               last = true;
 400 +                               break;
 401 +                       }
 402 +
 403 +                       route->tuple[!dir].in.num_encaps++;
 404 +                       route->tuple[!dir].in.encap[n_encaps].id = path->encap.id;
 405 +                       route->tuple[!dir].in.encap[n_encaps].proto = path->encap.proto;
 406 +                       if (path->type == DEV_PATH_PPPOE)
 407 +                               memcpy(route->tuple[dir].out.h_dest,
 408 +                                      path->encap.h_dest, ETH_ALEN);
 409 +                       break;
 410 +               case DEV_PATH_BRIDGE:
 411 +                       switch (path->bridge.vlan_mode) {
 412 +                       case DEV_PATH_BR_VLAN_TAG:
 413 +                               if (n_encaps >= NF_FLOW_TABLE_ENCAP_MAX ||
 414 +                                   i == stack.num_paths) {
 415 +                                       last = true;
 416 +                                       break;
 417 +                               }
 418 +
 419 +                               route->tuple[!dir].in.num_encaps++;
 420 +                               route->tuple[!dir].in.encap[n_encaps].id =
 421 +                                       path->bridge.vlan_id;
 422 +                               route->tuple[!dir].in.encap[n_encaps].proto =
 423 +                                       path->bridge.vlan_proto;
 424 +                               break;
 425 +                       case DEV_PATH_BR_VLAN_UNTAG:
 426 +                               route->tuple[!dir].in.num_encaps--;
 427 +                               break;
 428 +                       case DEV_PATH_BR_VLAN_UNTAG_HW:
 429 +                               route->tuple[!dir].in.ingress_vlans |= BIT(n_encaps - 1);
 430 +                               break;
 431 +                       case DEV_PATH_BR_VLAN_KEEP:
 432 +                               break;
 433 +                       }
 434 +                       break;
 435 +               default:
 436 +                       last = true;
 437 +                       break;
 438 +               }
 439 +
 440 +               if (last)
 441 +                       break;
 442 +       }
 443 +
 444 +       *out_dev = dev;
 445 +       route->tuple[dir].out.hw_ifindex = dev->ifindex;
 446 +       route->tuple[!dir].in.ifindex = dev->ifindex;
 447 +}
 448 +
 449 +static int
 450 +xt_flowoffload_route_dir(struct nf_flow_route *route, const struct nf_conn *ct,
 451 +                        enum ip_conntrack_dir dir,
 452 +                        const struct xt_action_param *par, int ifindex)
 453 +{
 454 +       struct dst_entry *dst = NULL;
 455 +       struct flowi fl;
 456 +
 457 +       memset(&fl, 0, sizeof(fl));
 458 +       switch (xt_family(par)) {
 459 +       case NFPROTO_IPV4:
 460 +               fl.u.ip4.daddr = ct->tuplehash[!dir].tuple.src.u3.ip;
 461 +               fl.u.ip4.flowi4_oif = ifindex;
 462 +               break;
 463 +       case NFPROTO_IPV6:
 464 +               fl.u.ip6.saddr = ct->tuplehash[!dir].tuple.dst.u3.in6;
 465 +               fl.u.ip6.daddr = ct->tuplehash[!dir].tuple.src.u3.in6;
 466 +               fl.u.ip6.flowi6_oif = ifindex;
 467 +               break;
 468 +       }
 469 +
 470 +       nf_route(xt_net(par), &dst, &fl, false, xt_family(par));
 471 +       if (!dst)
 472 +               return -ENOENT;
 473 +
 474 +       route->tuple[dir].dst = dst;
 475 +       if (dst_xfrm(dst))
 476 +               route->tuple[dir].xmit_type = FLOW_OFFLOAD_XMIT_XFRM;
 477 +       else
 478 +               route->tuple[dir].xmit_type = FLOW_OFFLOAD_XMIT_NEIGH;
 479 +
 480 +       return 0;
 481 +}
 482 +
 483 +static int
 484 +xt_flowoffload_route(struct sk_buff *skb, const struct nf_conn *ct,
 485 +                    const struct xt_action_param *par,
 486 +                    struct nf_flow_route *route, enum ip_conntrack_dir dir,
 487 +                    struct net_device **dev)
 488 +{
 489 +       int ret;
 490 +
 491 +       ret = xt_flowoffload_route_dir(route, ct, dir, par,
 492 +                                      dev[dir]->ifindex);
 493 +       if (ret)
 494 +               return ret;
 495 +
 496 +       ret = xt_flowoffload_route_dir(route, ct, !dir, par,
 497 +                                      dev[!dir]->ifindex);
 498 +       if (ret)
 499 +               return ret;
 500 +
 501 +       xt_flowoffload_route_check_path(route, ct, dir, &dev[!dir]);
 502 +       xt_flowoffload_route_check_path(route, ct, !dir, &dev[dir]);
 503 +
 504 +       return 0;
 505 +}
 506 +
 507 +static unsigned int
 508 +flowoffload_tg(struct sk_buff *skb, const struct xt_action_param *par)
 509 +{
 510 +       struct xt_flowoffload_table *table;
 511 +       const struct xt_flowoffload_target_info *info = par->targinfo;
 512 +       struct tcphdr _tcph, *tcph = NULL;
 513 +       enum ip_conntrack_info ctinfo;
 514 +       enum ip_conntrack_dir dir;
 515 +       struct nf_flow_route route = {};
 516 +       struct flow_offload *flow = NULL;
 517 +       struct net_device *devs[2] = {};
 518 +       struct nf_conn *ct;
 519 +       struct net *net;
 520 +
 521 +       if (xt_flowoffload_skip(skb, xt_family(par)))
 522 +               return XT_CONTINUE;
 523 +
 524 +       ct = nf_ct_get(skb, &ctinfo);
 525 +       if (ct == NULL)
 526 +               return XT_CONTINUE;
 527 +
 528 +       switch (ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.protonum) {
 529 +       case IPPROTO_TCP:
 530 +               if (ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED)
 531 +                       return XT_CONTINUE;
 532 +
 533 +               tcph = skb_header_pointer(skb, par->thoff,
 534 +                                         sizeof(_tcph), &_tcph);
 535 +               if (unlikely(!tcph || tcph->fin || tcph->rst))
 536 +                       return XT_CONTINUE;
 537 +               break;
 538 +       case IPPROTO_UDP:
 539 +               break;
 540 +       default:
 541 +               return XT_CONTINUE;
 542 +       }
 543 +
 544 +       if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
 545 +           ct->status & IPS_SEQ_ADJUST)
 546 +               return XT_CONTINUE;
 547 +
 548 +       if (!nf_ct_is_confirmed(ct))
 549 +               return XT_CONTINUE;
 550 +
 551 +       devs[dir] = xt_out(par);
 552 +       devs[!dir] = xt_in(par);
 553 +
 554 +       if (!devs[dir] || !devs[!dir])
 555 +               return XT_CONTINUE;
 556 +
 557 +       if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
 558 +               return XT_CONTINUE;
 559 +
 560 +       dir = CTINFO2DIR(ctinfo);
 561 +
 562 +       if (xt_flowoffload_route(skb, ct, par, &route, dir, devs) < 0)
 563 +               goto err_flow_route;
 564 +
 565 +       flow = flow_offload_alloc(ct);
 566 +       if (!flow)
 567 +               goto err_flow_alloc;
 568 +
 569 +       if (flow_offload_route_init(flow, &route) < 0)
 570 +               goto err_flow_add;
 571 +
 572 +       if (tcph) {
 573 +               ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
 574 +               ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
 575 +       }
 576 +
 577 +       table = &flowtable[!!(info->flags & XT_FLOWOFFLOAD_HW)];
 578 +       if (flow_offload_add(&table->ft, flow) < 0)
 579 +               goto err_flow_add;
 580 +
 581 +       xt_flowoffload_check_device(table, devs[0]);
 582 +       xt_flowoffload_check_device(table, devs[1]);
 583 +
 584 +       net = read_pnet(&table->ft.net);
 585 +       if (!net)
 586 +               write_pnet(&table->ft.net, xt_net(par));
 587 +
 588 +       dst_release(route.tuple[dir].dst);
 589 +       dst_release(route.tuple[!dir].dst);
 590 +
 591 +       return XT_CONTINUE;
 592 +
 593 +err_flow_add:
 594 +       flow_offload_free(flow);
 595 +err_flow_alloc:
 596 +       dst_release(route.tuple[dir].dst);
 597 +       dst_release(route.tuple[!dir].dst);
 598 +err_flow_route:
 599 +       clear_bit(IPS_OFFLOAD_BIT, &ct->status);
 600 +
 601 +       return XT_CONTINUE;
 602 +}
 603 +
 604 +static int flowoffload_chk(const struct xt_tgchk_param *par)
 605 +{
 606 +       struct xt_flowoffload_target_info *info = par->targinfo;
 607 +
 608 +       if (info->flags & ~XT_FLOWOFFLOAD_MASK)
 609 +               return -EINVAL;
 610 +
 611 +       return 0;
 612 +}
 613 +
 614 +static struct xt_target offload_tg_reg __read_mostly = {
 615 +       .family         = NFPROTO_UNSPEC,
 616 +       .name           = "FLOWOFFLOAD",
 617 +       .revision       = 0,
 618 +       .targetsize     = sizeof(struct xt_flowoffload_target_info),
 619 +       .usersize       = sizeof(struct xt_flowoffload_target_info),
 620 +       .checkentry     = flowoffload_chk,
 621 +       .target         = flowoffload_tg,
 622 +       .me             = THIS_MODULE,
 623 +};
 624 +
 625 +static int flow_offload_netdev_event(struct notifier_block *this,
 626 +                                    unsigned long event, void *ptr)
 627 +{
 628 +       struct xt_flowoffload_hook *hook0, *hook1;
 629 +       struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 630 +
 631 +       if (event != NETDEV_UNREGISTER)
 632 +               return NOTIFY_DONE;
 633 +
 634 +       spin_lock_bh(&hooks_lock);
 635 +       hook0 = flow_offload_lookup_hook(&flowtable[0], dev);
 636 +       if (hook0)
 637 +               hlist_del(&hook0->list);
 638 +
 639 +       hook1 = flow_offload_lookup_hook(&flowtable[1], dev);
 640 +       if (hook1)
 641 +               hlist_del(&hook1->list);
 642 +       spin_unlock_bh(&hooks_lock);
 643 +
 644 +       if (hook0) {
 645 +               nf_unregister_net_hook(hook0->net, &hook0->ops);
 646 +               kfree(hook0);
 647 +       }
 648 +
 649 +       if (hook1) {
 650 +               nf_unregister_net_hook(hook1->net, &hook1->ops);
 651 +               kfree(hook1);
 652 +       }
 653 +
 654 +       nf_flow_table_cleanup(dev);
 655 +
 656 +       return NOTIFY_DONE;
 657 +}
 658 +
 659 +static struct notifier_block flow_offload_netdev_notifier = {
 660 +       .notifier_call  = flow_offload_netdev_event,
 661 +};
 662 +
 663 +static unsigned int
 664 +nf_flow_offload_inet_hook(void *priv, struct sk_buff *skb,
 665 +                         const struct nf_hook_state *state)
 666 +{
 667 +       switch (skb->protocol) {
 668 +       case htons(ETH_P_IP):
 669 +               return nf_flow_offload_ip_hook(priv, skb, state);
 670 +       case htons(ETH_P_IPV6):
 671 +               return nf_flow_offload_ipv6_hook(priv, skb, state);
 672 +       }
 673 +
 674 +       return NF_ACCEPT;
 675 +}
 676 +
 677 +static int nf_flow_rule_route_inet(struct net *net,
 678 +                                  const struct flow_offload *flow,
 679 +                                  enum flow_offload_tuple_dir dir,
 680 +                                  struct nf_flow_rule *flow_rule)
 681 +{
 682 +       const struct flow_offload_tuple *flow_tuple = &flow->tuplehash[dir].tuple;
 683 +       int err;
 684 +
 685 +       switch (flow_tuple->l3proto) {
 686 +       case NFPROTO_IPV4:
 687 +               err = nf_flow_rule_route_ipv4(net, flow, dir, flow_rule);
 688 +               break;
 689 +       case NFPROTO_IPV6:
 690 +               err = nf_flow_rule_route_ipv6(net, flow, dir, flow_rule);
 691 +               break;
 692 +       default:
 693 +               err = -1;
 694 +               break;
 695 +       }
 696 +
 697 +       return err;
 698 +}
 699 +
 700 +static struct nf_flowtable_type flowtable_inet = {
 701 +       .family         = NFPROTO_INET,
 702 +       .init           = nf_flow_table_init,
 703 +       .setup          = nf_flow_table_offload_setup,
 704 +       .action         = nf_flow_rule_route_inet,
 705 +       .free           = nf_flow_table_free,
 706 +       .hook           = nf_flow_offload_inet_hook,
 707 +       .owner          = THIS_MODULE,
 708 +};
 709 +
 710 +static int init_flowtable(struct xt_flowoffload_table *tbl)
 711 +{
 712 +       INIT_DELAYED_WORK(&tbl->work, xt_flowoffload_hook_work);
 713 +       tbl->ft.type = &flowtable_inet;
 714 +
 715 +       return nf_flow_table_init(&tbl->ft);
 716 +}
 717 +
 718 +static int __init xt_flowoffload_tg_init(void)
 719 +{
 720 +       int ret;
 721 +
 722 +       register_netdevice_notifier(&flow_offload_netdev_notifier);
 723 +
 724 +       ret = init_flowtable(&flowtable[0]);
 725 +       if (ret)
 726 +               return ret;
 727 +
 728 +       ret = init_flowtable(&flowtable[1]);
 729 +       if (ret)
 730 +               goto cleanup;
 731 +
 732 +       flowtable[1].ft.flags = NF_FLOWTABLE_HW_OFFLOAD;
 733 +
 734 +       ret = xt_register_target(&offload_tg_reg);
 735 +       if (ret)
 736 +               goto cleanup2;
 737 +
 738 +       return 0;
 739 +
 740 +cleanup2:
 741 +       nf_flow_table_free(&flowtable[1].ft);
 742 +cleanup:
 743 +       nf_flow_table_free(&flowtable[0].ft);
 744 +       return ret;
 745 +}
 746 +
 747 +static void __exit xt_flowoffload_tg_exit(void)
 748 +{
 749 +       xt_unregister_target(&offload_tg_reg);
 750 +       unregister_netdevice_notifier(&flow_offload_netdev_notifier);
 751 +       nf_flow_table_free(&flowtable[0].ft);
 752 +       nf_flow_table_free(&flowtable[1].ft);
 753 +}
 754 +
 755 +MODULE_LICENSE("GPL");
 756 +module_init(xt_flowoffload_tg_init);
 757 +module_exit(xt_flowoffload_tg_exit);
 758 --- a/net/netfilter/nf_flow_table_core.c
 759 +++ b/net/netfilter/nf_flow_table_core.c
 760 @@ -7,7 +7,6 @@
 761  #include <linux/netdevice.h>
 762  #include <net/ip.h>
 763  #include <net/ip6_route.h>
 764 -#include <net/netfilter/nf_tables.h>
 765  #include <net/netfilter/nf_flow_table.h>
 766  #include <net/netfilter/nf_conntrack.h>
 767  #include <net/netfilter/nf_conntrack_core.h>
 768 @@ -395,8 +394,7 @@ flow_offload_lookup(struct nf_flowtable
 769  }
 770  EXPORT_SYMBOL_GPL(flow_offload_lookup);
 771
 772 -static int
 773 -nf_flow_table_iterate(struct nf_flowtable *flow_table,
 774 +int nf_flow_table_iterate(struct nf_flowtable *flow_table,
 775                       void (*iter)(struct flow_offload *flow, void *data),
 776                       void *data)
 777  {
 778 @@ -428,6 +426,7 @@ nf_flow_table_iterate(struct nf_flowtabl
 779
 780         return err;
 781  }
 782 +EXPORT_SYMBOL_GPL(nf_flow_table_iterate);
 783
 784  static bool flow_offload_stale_dst(struct flow_offload_tuple *tuple)
 785  {
 786 --- /dev/null
 787 +++ b/include/uapi/linux/netfilter/xt_FLOWOFFLOAD.h
 788 @@ -0,0 +1,17 @@
 789 +/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
 790 +#ifndef _XT_FLOWOFFLOAD_H
 791 +#define _XT_FLOWOFFLOAD_H
 792 +
 793 +#include <linux/types.h>
 794 +
 795 +enum {
 796 +       XT_FLOWOFFLOAD_HW       = 1 << 0,
 797 +
 798 +       XT_FLOWOFFLOAD_MASK     = XT_FLOWOFFLOAD_HW
 799 +};
 800 +
 801 +struct xt_flowoffload_target_info {
 802 +       __u32 flags;
 803 +};
 804 +
 805 +#endif /* _XT_FLOWOFFLOAD_H */
 806 --- a/include/net/netfilter/nf_flow_table.h
 807 +++ b/include/net/netfilter/nf_flow_table.h
 808 @@ -270,6 +270,10 @@ void nf_flow_table_free(struct nf_flowta
 809
 810  void flow_offload_teardown(struct flow_offload *flow);
 811
 812 +int nf_flow_table_iterate(struct nf_flowtable *flow_table,
 813 +                         void (*iter)(struct flow_offload *flow, void *data),
 814 +                         void *data);
 815 +
 816  void nf_flow_snat_port(const struct flow_offload *flow,
 817                        struct sk_buff *skb, unsigned int thoff,
 818                        u8 protocol, enum flow_offload_tuple_dir dir);