target/linux/generic/hack-5.15/650-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 @@ -707,8 +707,6 @@ config NFT_REJECT_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 @@ -717,11 +715,12 @@ 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 @@ -1010,6 +1009,15 @@ 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 @@ -143,6 +143,7 @@ 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 +1,702 @@
 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 <linux/if_vlan.h>
 114 +#include <net/ip.h>
 115 +#include <net/netfilter/nf_conntrack.h>
 116 +#include <net/netfilter/nf_conntrack_extend.h>
 117 +#include <net/netfilter/nf_conntrack_helper.h>
 118 +#include <net/netfilter/nf_flow_table.h>
 119 +
 120 +struct xt_flowoffload_hook {
 121 +       struct hlist_node list;
 122 +       struct nf_hook_ops ops;
 123 +       struct net *net;
 124 +       bool registered;
 125 +       bool used;
 126 +};
 127 +
 128 +struct xt_flowoffload_table {
 129 +       struct nf_flowtable ft;
 130 +       struct hlist_head hooks;
 131 +       struct delayed_work work;
 132 +};
 133 +
 134 +struct nf_forward_info {
 135 +       const struct net_device *indev;
 136 +       const struct net_device *outdev;
 137 +       const struct net_device *hw_outdev;
 138 +       struct id {
 139 +               __u16   id;
 140 +               __be16  proto;
 141 +       } encap[NF_FLOW_TABLE_ENCAP_MAX];
 142 +       u8 num_encaps;
 143 +       u8 ingress_vlans;
 144 +       u8 h_source[ETH_ALEN];
 145 +       u8 h_dest[ETH_ALEN];
 146 +       enum flow_offload_xmit_type xmit_type;
 147 +};
 148 +
 149 +static DEFINE_SPINLOCK(hooks_lock);
 150 +
 151 +struct xt_flowoffload_table flowtable[2];
 152 +
 153 +static unsigned int
 154 +xt_flowoffload_net_hook(void *priv, struct sk_buff *skb,
 155 +                       const struct nf_hook_state *state)
 156 +{
 157 +       struct vlan_ethhdr *veth;
 158 +       __be16 proto;
 159 +
 160 +       switch (skb->protocol) {
 161 +       case htons(ETH_P_8021Q):
 162 +               veth = (struct vlan_ethhdr *)skb_mac_header(skb);
 163 +               proto = veth->h_vlan_encapsulated_proto;
 164 +               break;
 165 +       case htons(ETH_P_PPP_SES):
 166 +               if (!nf_flow_pppoe_proto(skb, &proto))
 167 +                       return NF_ACCEPT;
 168 +               break;
 169 +       default:
 170 +               proto = skb->protocol;
 171 +               break;
 172 +       }
 173 +
 174 +       switch (proto) {
 175 +       case htons(ETH_P_IP):
 176 +               return nf_flow_offload_ip_hook(priv, skb, state);
 177 +       case htons(ETH_P_IPV6):
 178 +               return nf_flow_offload_ipv6_hook(priv, skb, state);
 179 +       }
 180 +
 181 +       return NF_ACCEPT;
 182 +}
 183 +
 184 +static int
 185 +xt_flowoffload_create_hook(struct xt_flowoffload_table *table,
 186 +                          struct net_device *dev)
 187 +{
 188 +       struct xt_flowoffload_hook *hook;
 189 +       struct nf_hook_ops *ops;
 190 +
 191 +       hook = kzalloc(sizeof(*hook), GFP_ATOMIC);
 192 +       if (!hook)
 193 +               return -ENOMEM;
 194 +
 195 +       ops = &hook->ops;
 196 +       ops->pf = NFPROTO_NETDEV;
 197 +       ops->hooknum = NF_NETDEV_INGRESS;
 198 +       ops->priority = 10;
 199 +       ops->priv = &table->ft;
 200 +       ops->hook = xt_flowoffload_net_hook;
 201 +       ops->dev = dev;
 202 +
 203 +       hlist_add_head(&hook->list, &table->hooks);
 204 +       mod_delayed_work(system_power_efficient_wq, &table->work, 0);
 205 +
 206 +       return 0;
 207 +}
 208 +
 209 +static struct xt_flowoffload_hook *
 210 +flow_offload_lookup_hook(struct xt_flowoffload_table *table,
 211 +                        struct net_device *dev)
 212 +{
 213 +       struct xt_flowoffload_hook *hook;
 214 +
 215 +       hlist_for_each_entry(hook, &table->hooks, list) {
 216 +               if (hook->ops.dev == dev)
 217 +                       return hook;
 218 +       }
 219 +
 220 +       return NULL;
 221 +}
 222 +
 223 +static void
 224 +xt_flowoffload_check_device(struct xt_flowoffload_table *table,
 225 +                           struct net_device *dev)
 226 +{
 227 +       struct xt_flowoffload_hook *hook;
 228 +
 229 +       if (!dev)
 230 +               return;
 231 +
 232 +       spin_lock_bh(&hooks_lock);
 233 +       hook = flow_offload_lookup_hook(table, dev);
 234 +       if (hook)
 235 +               hook->used = true;
 236 +       else
 237 +               xt_flowoffload_create_hook(table, dev);
 238 +       spin_unlock_bh(&hooks_lock);
 239 +}
 240 +
 241 +static void
 242 +xt_flowoffload_register_hooks(struct xt_flowoffload_table *table)
 243 +{
 244 +       struct xt_flowoffload_hook *hook;
 245 +
 246 +restart:
 247 +       hlist_for_each_entry(hook, &table->hooks, list) {
 248 +               if (hook->registered)
 249 +                       continue;
 250 +
 251 +               hook->registered = true;
 252 +               hook->net = dev_net(hook->ops.dev);
 253 +               spin_unlock_bh(&hooks_lock);
 254 +               nf_register_net_hook(hook->net, &hook->ops);
 255 +               if (table->ft.flags & NF_FLOWTABLE_HW_OFFLOAD)
 256 +                       table->ft.type->setup(&table->ft, hook->ops.dev,
 257 +                                             FLOW_BLOCK_BIND);
 258 +               spin_lock_bh(&hooks_lock);
 259 +               goto restart;
 260 +       }
 261 +
 262 +}
 263 +
 264 +static bool
 265 +xt_flowoffload_cleanup_hooks(struct xt_flowoffload_table *table)
 266 +{
 267 +       struct xt_flowoffload_hook *hook;
 268 +       bool active = false;
 269 +
 270 +restart:
 271 +       spin_lock_bh(&hooks_lock);
 272 +       hlist_for_each_entry(hook, &table->hooks, list) {
 273 +               if (hook->used || !hook->registered) {
 274 +                       active = true;
 275 +                       continue;
 276 +               }
 277 +
 278 +               hlist_del(&hook->list);
 279 +               spin_unlock_bh(&hooks_lock);
 280 +               if (table->ft.flags & NF_FLOWTABLE_HW_OFFLOAD)
 281 +                       table->ft.type->setup(&table->ft, hook->ops.dev,
 282 +                                             FLOW_BLOCK_UNBIND);
 283 +               nf_unregister_net_hook(hook->net, &hook->ops);
 284 +               kfree(hook);
 285 +               goto restart;
 286 +       }
 287 +       spin_unlock_bh(&hooks_lock);
 288 +
 289 +       return active;
 290 +}
 291 +
 292 +static void
 293 +xt_flowoffload_check_hook(struct nf_flowtable *flowtable,
 294 +                         struct flow_offload *flow, void *data)
 295 +{
 296 +       struct xt_flowoffload_table *table;
 297 +       struct flow_offload_tuple *tuple0 = &flow->tuplehash[0].tuple;
 298 +       struct flow_offload_tuple *tuple1 = &flow->tuplehash[1].tuple;
 299 +       struct xt_flowoffload_hook *hook;
 300 +
 301 +       table = container_of(flowtable, struct xt_flowoffload_table, ft);
 302 +
 303 +       spin_lock_bh(&hooks_lock);
 304 +       hlist_for_each_entry(hook, &table->hooks, list) {
 305 +               if (hook->ops.dev->ifindex != tuple0->iifidx &&
 306 +                   hook->ops.dev->ifindex != tuple1->iifidx)
 307 +                       continue;
 308 +
 309 +               hook->used = true;
 310 +       }
 311 +       spin_unlock_bh(&hooks_lock);
 312 +}
 313 +
 314 +static void
 315 +xt_flowoffload_hook_work(struct work_struct *work)
 316 +{
 317 +       struct xt_flowoffload_table *table;
 318 +       struct xt_flowoffload_hook *hook;
 319 +       int err;
 320 +
 321 +       table = container_of(work, struct xt_flowoffload_table, work.work);
 322 +
 323 +       spin_lock_bh(&hooks_lock);
 324 +       xt_flowoffload_register_hooks(table);
 325 +       hlist_for_each_entry(hook, &table->hooks, list)
 326 +               hook->used = false;
 327 +       spin_unlock_bh(&hooks_lock);
 328 +
 329 +       err = nf_flow_table_iterate(&table->ft, xt_flowoffload_check_hook,
 330 +                                   NULL);
 331 +       if (err && err != -EAGAIN)
 332 +               goto out;
 333 +
 334 +       if (!xt_flowoffload_cleanup_hooks(table))
 335 +               return;
 336 +
 337 +out:
 338 +       queue_delayed_work(system_power_efficient_wq, &table->work, HZ);
 339 +}
 340 +
 341 +static bool
 342 +xt_flowoffload_skip(struct sk_buff *skb, int family)
 343 +{
 344 +       if (skb_sec_path(skb))
 345 +               return true;
 346 +
 347 +       if (family == NFPROTO_IPV4) {
 348 +               const struct ip_options *opt = &(IPCB(skb)->opt);
 349 +
 350 +               if (unlikely(opt->optlen))
 351 +                       return true;
 352 +       }
 353 +
 354 +       return false;
 355 +}
 356 +
 357 +static enum flow_offload_xmit_type nf_xmit_type(struct dst_entry *dst)
 358 +{
 359 +       if (dst_xfrm(dst))
 360 +               return FLOW_OFFLOAD_XMIT_XFRM;
 361 +
 362 +       return FLOW_OFFLOAD_XMIT_NEIGH;
 363 +}
 364 +
 365 +static void nf_default_forward_path(struct nf_flow_route *route,
 366 +                                   struct dst_entry *dst_cache,
 367 +                                   enum ip_conntrack_dir dir,
 368 +                                   struct net_device **dev)
 369 +{
 370 +       dev[!dir] = dst_cache->dev;
 371 +       route->tuple[!dir].in.ifindex   = dst_cache->dev->ifindex;
 372 +       route->tuple[dir].dst           = dst_cache;
 373 +       route->tuple[dir].xmit_type     = nf_xmit_type(dst_cache);
 374 +}
 375 +
 376 +static bool nf_is_valid_ether_device(const struct net_device *dev)
 377 +{
 378 +       if (!dev || (dev->flags & IFF_LOOPBACK) || dev->type != ARPHRD_ETHER ||
 379 +           dev->addr_len != ETH_ALEN || !is_valid_ether_addr(dev->dev_addr))
 380 +               return false;
 381 +
 382 +       return true;
 383 +}
 384 +
 385 +static void nf_dev_path_info(const struct net_device_path_stack *stack,
 386 +                            struct nf_forward_info *info,
 387 +                            unsigned char *ha)
 388 +{
 389 +       const struct net_device_path *path;
 390 +       int i;
 391 +
 392 +       memcpy(info->h_dest, ha, ETH_ALEN);
 393 +
 394 +       for (i = 0; i < stack->num_paths; i++) {
 395 +               path = &stack->path[i];
 396 +               switch (path->type) {
 397 +               case DEV_PATH_ETHERNET:
 398 +               case DEV_PATH_DSA:
 399 +               case DEV_PATH_VLAN:
 400 +               case DEV_PATH_PPPOE:
 401 +                       info->indev = path->dev;
 402 +                       if (is_zero_ether_addr(info->h_source))
 403 +                               memcpy(info->h_source, path->dev->dev_addr, ETH_ALEN);
 404 +
 405 +                       if (path->type == DEV_PATH_ETHERNET)
 406 +                               break;
 407 +                       if (path->type == DEV_PATH_DSA) {
 408 +                               i = stack->num_paths;
 409 +                               break;
 410 +                       }
 411 +
 412 +                       /* DEV_PATH_VLAN and DEV_PATH_PPPOE */
 413 +                       if (info->num_encaps >= NF_FLOW_TABLE_ENCAP_MAX) {
 414 +                               info->indev = NULL;
 415 +                               break;
 416 +                       }
 417 +                       if (!info->outdev)
 418 +                               info->outdev = path->dev;
 419 +                       info->encap[info->num_encaps].id = path->encap.id;
 420 +                       info->encap[info->num_encaps].proto = path->encap.proto;
 421 +                       info->num_encaps++;
 422 +                       if (path->type == DEV_PATH_PPPOE)
 423 +                               memcpy(info->h_dest, path->encap.h_dest, ETH_ALEN);
 424 +                       break;
 425 +               case DEV_PATH_BRIDGE:
 426 +                       if (is_zero_ether_addr(info->h_source))
 427 +                               memcpy(info->h_source, path->dev->dev_addr, ETH_ALEN);
 428 +
 429 +                       switch (path->bridge.vlan_mode) {
 430 +                       case DEV_PATH_BR_VLAN_UNTAG_HW:
 431 +                               info->ingress_vlans |= BIT(info->num_encaps - 1);
 432 +                               break;
 433 +                       case DEV_PATH_BR_VLAN_TAG:
 434 +                               info->encap[info->num_encaps].id = path->bridge.vlan_id;
 435 +                               info->encap[info->num_encaps].proto = path->bridge.vlan_proto;
 436 +                               info->num_encaps++;
 437 +                               break;
 438 +                       case DEV_PATH_BR_VLAN_UNTAG:
 439 +                               info->num_encaps--;
 440 +                               break;
 441 +                       case DEV_PATH_BR_VLAN_KEEP:
 442 +                               break;
 443 +                       }
 444 +                       break;
 445 +               default:
 446 +                       info->indev = NULL;
 447 +                       break;
 448 +               }
 449 +       }
 450 +       if (!info->outdev)
 451 +               info->outdev = info->indev;
 452 +
 453 +       info->hw_outdev = info->indev;
 454 +
 455 +       if (nf_is_valid_ether_device(info->indev))
 456 +               info->xmit_type = FLOW_OFFLOAD_XMIT_DIRECT;
 457 +}
 458 +
 459 +static int nf_dev_fill_forward_path(const struct nf_flow_route *route,
 460 +                                    const struct dst_entry *dst_cache,
 461 +                                    const struct nf_conn *ct,
 462 +                                    enum ip_conntrack_dir dir, u8 *ha,
 463 +                                    struct net_device_path_stack *stack)
 464 +{
 465 +       const void *daddr = &ct->tuplehash[!dir].tuple.src.u3;
 466 +       struct net_device *dev = dst_cache->dev;
 467 +       struct neighbour *n;
 468 +       u8 nud_state;
 469 +
 470 +       if (!nf_is_valid_ether_device(dev))
 471 +               goto out;
 472 +
 473 +       n = dst_neigh_lookup(dst_cache, daddr);
 474 +       if (!n)
 475 +               return -1;
 476 +
 477 +       read_lock_bh(&n->lock);
 478 +       nud_state = n->nud_state;
 479 +       ether_addr_copy(ha, n->ha);
 480 +       read_unlock_bh(&n->lock);
 481 +       neigh_release(n);
 482 +
 483 +       if (!(nud_state & NUD_VALID))
 484 +               return -1;
 485 +
 486 +out:
 487 +       return dev_fill_forward_path(dev, ha, stack);
 488 +}
 489 +
 490 +static void nf_dev_forward_path(struct nf_flow_route *route,
 491 +                               const struct nf_conn *ct,
 492 +                               enum ip_conntrack_dir dir,
 493 +                               struct net_device **devs)
 494 +{
 495 +       const struct dst_entry *dst = route->tuple[dir].dst;
 496 +       struct net_device_path_stack stack;
 497 +       struct nf_forward_info info = {};
 498 +       unsigned char ha[ETH_ALEN];
 499 +       int i;
 500 +
 501 +       if (nf_dev_fill_forward_path(route, dst, ct, dir, ha, &stack) >= 0)
 502 +               nf_dev_path_info(&stack, &info, ha);
 503 +
 504 +       devs[!dir] = (struct net_device *)info.indev;
 505 +       if (!info.indev)
 506 +               return;
 507 +
 508 +       route->tuple[!dir].in.ifindex = info.indev->ifindex;
 509 +       for (i = 0; i < info.num_encaps; i++) {
 510 +               route->tuple[!dir].in.encap[i].id = info.encap[i].id;
 511 +               route->tuple[!dir].in.encap[i].proto = info.encap[i].proto;
 512 +       }
 513 +       route->tuple[!dir].in.num_encaps = info.num_encaps;
 514 +       route->tuple[!dir].in.ingress_vlans = info.ingress_vlans;
 515 +
 516 +       if (info.xmit_type == FLOW_OFFLOAD_XMIT_DIRECT) {
 517 +               memcpy(route->tuple[dir].out.h_source, info.h_source, ETH_ALEN);
 518 +               memcpy(route->tuple[dir].out.h_dest, info.h_dest, ETH_ALEN);
 519 +               route->tuple[dir].out.ifindex = info.outdev->ifindex;
 520 +               route->tuple[dir].out.hw_ifindex = info.hw_outdev->ifindex;
 521 +               route->tuple[dir].xmit_type = info.xmit_type;
 522 +       }
 523 +}
 524 +
 525 +static int
 526 +xt_flowoffload_route(struct sk_buff *skb, const struct nf_conn *ct,
 527 +                    const struct xt_action_param *par,
 528 +                    struct nf_flow_route *route, enum ip_conntrack_dir dir,
 529 +                    struct net_device **devs)
 530 +{
 531 +       struct dst_entry *this_dst = skb_dst(skb);
 532 +       struct dst_entry *other_dst = NULL;
 533 +       struct flowi fl;
 534 +
 535 +       memset(&fl, 0, sizeof(fl));
 536 +       switch (xt_family(par)) {
 537 +       case NFPROTO_IPV4:
 538 +               fl.u.ip4.daddr = ct->tuplehash[dir].tuple.src.u3.ip;
 539 +               fl.u.ip4.flowi4_oif = xt_in(par)->ifindex;
 540 +               break;
 541 +       case NFPROTO_IPV6:
 542 +               fl.u.ip6.saddr = ct->tuplehash[!dir].tuple.dst.u3.in6;
 543 +               fl.u.ip6.daddr = ct->tuplehash[dir].tuple.src.u3.in6;
 544 +               fl.u.ip6.flowi6_oif = xt_in(par)->ifindex;
 545 +               break;
 546 +       }
 547 +
 548 +       if (!dst_hold_safe(this_dst))
 549 +               return -ENOENT;
 550 +
 551 +       nf_route(xt_net(par), &other_dst, &fl, false, xt_family(par));
 552 +       if (!other_dst) {
 553 +               dst_release(this_dst);
 554 +               return -ENOENT;
 555 +       }
 556 +
 557 +       nf_default_forward_path(route, this_dst, dir, devs);
 558 +       nf_default_forward_path(route, other_dst, !dir, devs);
 559 +
 560 +       if (route->tuple[dir].xmit_type == FLOW_OFFLOAD_XMIT_NEIGH &&
 561 +           route->tuple[!dir].xmit_type == FLOW_OFFLOAD_XMIT_NEIGH) {
 562 +               nf_dev_forward_path(route, ct, dir, devs);
 563 +               nf_dev_forward_path(route, ct, !dir, devs);
 564 +       }
 565 +
 566 +       return 0;
 567 +}
 568 +
 569 +static unsigned int
 570 +flowoffload_tg(struct sk_buff *skb, const struct xt_action_param *par)
 571 +{
 572 +       struct xt_flowoffload_table *table;
 573 +       const struct xt_flowoffload_target_info *info = par->targinfo;
 574 +       struct tcphdr _tcph, *tcph = NULL;
 575 +       enum ip_conntrack_info ctinfo;
 576 +       enum ip_conntrack_dir dir;
 577 +       struct nf_flow_route route = {};
 578 +       struct flow_offload *flow = NULL;
 579 +       struct net_device *devs[2] = {};
 580 +       struct nf_conn *ct;
 581 +       struct net *net;
 582 +
 583 +       if (xt_flowoffload_skip(skb, xt_family(par)))
 584 +               return XT_CONTINUE;
 585 +
 586 +       ct = nf_ct_get(skb, &ctinfo);
 587 +       if (ct == NULL)
 588 +               return XT_CONTINUE;
 589 +
 590 +       switch (ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.protonum) {
 591 +       case IPPROTO_TCP:
 592 +               if (ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED)
 593 +                       return XT_CONTINUE;
 594 +
 595 +               tcph = skb_header_pointer(skb, par->thoff,
 596 +                                         sizeof(_tcph), &_tcph);
 597 +               if (unlikely(!tcph || tcph->fin || tcph->rst))
 598 +                       return XT_CONTINUE;
 599 +               break;
 600 +       case IPPROTO_UDP:
 601 +               break;
 602 +       default:
 603 +               return XT_CONTINUE;
 604 +       }
 605 +
 606 +       if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
 607 +           ct->status & (IPS_SEQ_ADJUST | IPS_NAT_CLASH))
 608 +               return XT_CONTINUE;
 609 +
 610 +       if (!nf_ct_is_confirmed(ct))
 611 +               return XT_CONTINUE;
 612 +
 613 +       dir = CTINFO2DIR(ctinfo);
 614 +
 615 +       devs[dir] = xt_out(par);
 616 +       devs[!dir] = xt_in(par);
 617 +
 618 +       if (!devs[dir] || !devs[!dir])
 619 +               return XT_CONTINUE;
 620 +
 621 +       if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
 622 +               return XT_CONTINUE;
 623 +
 624 +       if (xt_flowoffload_route(skb, ct, par, &route, dir, devs) < 0)
 625 +               goto err_flow_route;
 626 +
 627 +       flow = flow_offload_alloc(ct);
 628 +       if (!flow)
 629 +               goto err_flow_alloc;
 630 +
 631 +       flow_offload_route_init(flow, &route);
 632 +
 633 +       if (tcph) {
 634 +               ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
 635 +               ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
 636 +       }
 637 +
 638 +       table = &flowtable[!!(info->flags & XT_FLOWOFFLOAD_HW)];
 639 +
 640 +       net = read_pnet(&table->ft.net);
 641 +       if (!net)
 642 +               write_pnet(&table->ft.net, xt_net(par));
 643 +
 644 +       if (flow_offload_add(&table->ft, flow) < 0)
 645 +               goto err_flow_add;
 646 +
 647 +       xt_flowoffload_check_device(table, devs[0]);
 648 +       xt_flowoffload_check_device(table, devs[1]);
 649 +
 650 +       return XT_CONTINUE;
 651 +
 652 +err_flow_add:
 653 +       flow_offload_free(flow);
 654 +err_flow_alloc:
 655 +       dst_release(route.tuple[dir].dst);
 656 +       dst_release(route.tuple[!dir].dst);
 657 +err_flow_route:
 658 +       clear_bit(IPS_OFFLOAD_BIT, &ct->status);
 659 +
 660 +       return XT_CONTINUE;
 661 +}
 662 +
 663 +static int flowoffload_chk(const struct xt_tgchk_param *par)
 664 +{
 665 +       struct xt_flowoffload_target_info *info = par->targinfo;
 666 +
 667 +       if (info->flags & ~XT_FLOWOFFLOAD_MASK)
 668 +               return -EINVAL;
 669 +
 670 +       return 0;
 671 +}
 672 +
 673 +static struct xt_target offload_tg_reg __read_mostly = {
 674 +       .family         = NFPROTO_UNSPEC,
 675 +       .name           = "FLOWOFFLOAD",
 676 +       .revision       = 0,
 677 +       .targetsize     = sizeof(struct xt_flowoffload_target_info),
 678 +       .usersize       = sizeof(struct xt_flowoffload_target_info),
 679 +       .checkentry     = flowoffload_chk,
 680 +       .target         = flowoffload_tg,
 681 +       .me             = THIS_MODULE,
 682 +};
 683 +
 684 +static int flow_offload_netdev_event(struct notifier_block *this,
 685 +                                    unsigned long event, void *ptr)
 686 +{
 687 +       struct xt_flowoffload_hook *hook0, *hook1;
 688 +       struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 689 +
 690 +       if (event != NETDEV_UNREGISTER)
 691 +               return NOTIFY_DONE;
 692 +
 693 +       spin_lock_bh(&hooks_lock);
 694 +       hook0 = flow_offload_lookup_hook(&flowtable[0], dev);
 695 +       if (hook0)
 696 +               hlist_del(&hook0->list);
 697 +
 698 +       hook1 = flow_offload_lookup_hook(&flowtable[1], dev);
 699 +       if (hook1)
 700 +               hlist_del(&hook1->list);
 701 +       spin_unlock_bh(&hooks_lock);
 702 +
 703 +       if (hook0) {
 704 +               nf_unregister_net_hook(hook0->net, &hook0->ops);
 705 +               kfree(hook0);
 706 +       }
 707 +
 708 +       if (hook1) {
 709 +               nf_unregister_net_hook(hook1->net, &hook1->ops);
 710 +               kfree(hook1);
 711 +       }
 712 +
 713 +       nf_flow_table_cleanup(dev);
 714 +
 715 +       return NOTIFY_DONE;
 716 +}
 717 +
 718 +static struct notifier_block flow_offload_netdev_notifier = {
 719 +       .notifier_call  = flow_offload_netdev_event,
 720 +};
 721 +
 722 +static int nf_flow_rule_route_inet(struct net *net,
 723 +                                  const struct flow_offload *flow,
 724 +                                  enum flow_offload_tuple_dir dir,
 725 +                                  struct nf_flow_rule *flow_rule)
 726 +{
 727 +       const struct flow_offload_tuple *flow_tuple = &flow->tuplehash[dir].tuple;
 728 +       int err;
 729 +
 730 +       switch (flow_tuple->l3proto) {
 731 +       case NFPROTO_IPV4:
 732 +               err = nf_flow_rule_route_ipv4(net, flow, dir, flow_rule);
 733 +               break;
 734 +       case NFPROTO_IPV6:
 735 +               err = nf_flow_rule_route_ipv6(net, flow, dir, flow_rule);
 736 +               break;
 737 +       default:
 738 +               err = -1;
 739 +               break;
 740 +       }
 741 +
 742 +       return err;
 743 +}
 744 +
 745 +static struct nf_flowtable_type flowtable_inet = {
 746 +       .family         = NFPROTO_INET,
 747 +       .init           = nf_flow_table_init,
 748 +       .setup          = nf_flow_table_offload_setup,
 749 +       .action         = nf_flow_rule_route_inet,
 750 +       .free           = nf_flow_table_free,
 751 +       .hook           = xt_flowoffload_net_hook,
 752 +       .owner          = THIS_MODULE,
 753 +};
 754 +
 755 +static int init_flowtable(struct xt_flowoffload_table *tbl)
 756 +{
 757 +       INIT_DELAYED_WORK(&tbl->work, xt_flowoffload_hook_work);
 758 +       tbl->ft.type = &flowtable_inet;
 759 +       tbl->ft.flags = NF_FLOWTABLE_COUNTER;
 760 +
 761 +       return nf_flow_table_init(&tbl->ft);
 762 +}
 763 +
 764 +static int __init xt_flowoffload_tg_init(void)
 765 +{
 766 +       int ret;
 767 +
 768 +       register_netdevice_notifier(&flow_offload_netdev_notifier);
 769 +
 770 +       ret = init_flowtable(&flowtable[0]);
 771 +       if (ret)
 772 +               return ret;
 773 +
 774 +       ret = init_flowtable(&flowtable[1]);
 775 +       if (ret)
 776 +               goto cleanup;
 777 +
 778 +       flowtable[1].ft.flags |= NF_FLOWTABLE_HW_OFFLOAD;
 779 +
 780 +       ret = xt_register_target(&offload_tg_reg);
 781 +       if (ret)
 782 +               goto cleanup2;
 783 +
 784 +       return 0;
 785 +
 786 +cleanup2:
 787 +       nf_flow_table_free(&flowtable[1].ft);
 788 +cleanup:
 789 +       nf_flow_table_free(&flowtable[0].ft);
 790 +       return ret;
 791 +}
 792 +
 793 +static void __exit xt_flowoffload_tg_exit(void)
 794 +{
 795 +       xt_unregister_target(&offload_tg_reg);
 796 +       unregister_netdevice_notifier(&flow_offload_netdev_notifier);
 797 +       nf_flow_table_free(&flowtable[0].ft);
 798 +       nf_flow_table_free(&flowtable[1].ft);
 799 +}
 800 +
 801 +MODULE_LICENSE("GPL");
 802 +module_init(xt_flowoffload_tg_init);
 803 +module_exit(xt_flowoffload_tg_exit);
 804 --- a/net/netfilter/nf_flow_table_core.c
 805 +++ b/net/netfilter/nf_flow_table_core.c
 806 @@ -7,7 +7,6 @@
 807  #include <linux/netdevice.h>
 808  #include <net/ip.h>
 809  #include <net/ip6_route.h>
 810 -#include <net/netfilter/nf_tables.h>
 811  #include <net/netfilter/nf_flow_table.h>
 812  #include <net/netfilter/nf_conntrack.h>
 813  #include <net/netfilter/nf_conntrack_core.h>
 814 @@ -373,8 +372,7 @@ flow_offload_lookup(struct nf_flowtable
 815  }
 816  EXPORT_SYMBOL_GPL(flow_offload_lookup);
 817
 818 -static int
 819 -nf_flow_table_iterate(struct nf_flowtable *flow_table,
 820 +int nf_flow_table_iterate(struct nf_flowtable *flow_table,
 821                       void (*iter)(struct nf_flowtable *flowtable,
 822                                    struct flow_offload *flow, void *data),
 823                       void *data)
 824 @@ -428,6 +426,7 @@ static void nf_flow_offload_gc_step(stru
 825                 nf_flow_offload_stats(flow_table, flow);
 826         }
 827  }
 828 +EXPORT_SYMBOL_GPL(nf_flow_table_iterate);
 829
 830  void nf_flow_table_gc_run(struct nf_flowtable *flow_table)
 831  {
 832 --- /dev/null
 833 +++ b/include/uapi/linux/netfilter/xt_FLOWOFFLOAD.h
 834 @@ -0,0 +1,17 @@
 835 +/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
 836 +#ifndef _XT_FLOWOFFLOAD_H
 837 +#define _XT_FLOWOFFLOAD_H
 838 +
 839 +#include <linux/types.h>
 840 +
 841 +enum {
 842 +       XT_FLOWOFFLOAD_HW       = 1 << 0,
 843 +
 844 +       XT_FLOWOFFLOAD_MASK     = XT_FLOWOFFLOAD_HW
 845 +};
 846 +
 847 +struct xt_flowoffload_target_info {
 848 +       __u32 flags;
 849 +};
 850 +
 851 +#endif /* _XT_FLOWOFFLOAD_H */
 852 --- a/include/net/netfilter/nf_flow_table.h
 853 +++ b/include/net/netfilter/nf_flow_table.h
 854 @@ -276,6 +276,11 @@ void nf_flow_table_free(struct nf_flowta
 855
 856  void flow_offload_teardown(struct flow_offload *flow);
 857
 858 +int nf_flow_table_iterate(struct nf_flowtable *flow_table,
 859 +                      void (*iter)(struct nf_flowtable *flowtable,
 860 +                                   struct flow_offload *flow, void *data),
 861 +                      void *data);
 862 +
 863  void nf_flow_snat_port(const struct flow_offload *flow,
 864                        struct sk_buff *skb, unsigned int thoff,
 865                        u8 protocol, enum flow_offload_tuple_dir dir);