target/linux/generic/hack-6.6/650-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/netfilter/Kconfig
  10 +++ b/net/netfilter/Kconfig
  11 @@ -729,7 +729,6 @@ config NF_FLOW_TABLE
  12         tristate "Netfilter flow table module"
  13         depends on NETFILTER_INGRESS
  14         depends on NF_CONNTRACK
  15 -       depends on NF_TABLES
  16         help
  17           This option adds the flow table core infrastructure.
  18
  19 @@ -1025,6 +1024,15 @@ config NETFILTER_XT_TARGET_NOTRACK
  20         depends on NETFILTER_ADVANCED
  21         select NETFILTER_XT_TARGET_CT
  22
  23 +config NETFILTER_XT_TARGET_FLOWOFFLOAD
  24 +       tristate '"FLOWOFFLOAD" target support'
  25 +       depends on NF_FLOW_TABLE
  26 +       depends on NETFILTER_INGRESS
  27 +       help
  28 +         This option adds a `FLOWOFFLOAD' target, which uses the nf_flow_offload
  29 +         module to speed up processing of packets by bypassing the usual
  30 +         netfilter chains
  31 +
  32  config NETFILTER_XT_TARGET_RATEEST
  33         tristate '"RATEEST" target support'
  34         depends on NETFILTER_ADVANCED
  35 --- a/net/netfilter/Makefile
  36 +++ b/net/netfilter/Makefile
  37 @@ -163,6 +163,7 @@ obj-$(CONFIG_NETFILTER_XT_TARGET_CLASSIF
  38  obj-$(CONFIG_NETFILTER_XT_TARGET_CONNSECMARK) += xt_CONNSECMARK.o
  39  obj-$(CONFIG_NETFILTER_XT_TARGET_CT) += xt_CT.o
  40  obj-$(CONFIG_NETFILTER_XT_TARGET_DSCP) += xt_DSCP.o
  41 +obj-$(CONFIG_NETFILTER_XT_TARGET_FLOWOFFLOAD) += xt_FLOWOFFLOAD.o
  42  obj-$(CONFIG_NETFILTER_XT_TARGET_HL) += xt_HL.o
  43  obj-$(CONFIG_NETFILTER_XT_TARGET_HMARK) += xt_HMARK.o
  44  obj-$(CONFIG_NETFILTER_XT_TARGET_LED) += xt_LED.o
  45 --- /dev/null
  46 +++ b/net/netfilter/xt_FLOWOFFLOAD.c
  47 @@ -0,0 +1,703 @@
  48 +/*
  49 + * Copyright (C) 2018-2021 Felix Fietkau <nbd@nbd.name>
  50 + *
  51 + * This program is free software; you can redistribute it and/or modify
  52 + * it under the terms of the GNU General Public License version 2 as
  53 + * published by the Free Software Foundation.
  54 + */
  55 +#include <linux/module.h>
  56 +#include <linux/init.h>
  57 +#include <linux/netfilter.h>
  58 +#include <linux/netfilter/xt_FLOWOFFLOAD.h>
  59 +#include <linux/if_vlan.h>
  60 +#include <net/ip.h>
  61 +#include <net/netfilter/nf_conntrack.h>
  62 +#include <net/netfilter/nf_conntrack_extend.h>
  63 +#include <net/netfilter/nf_conntrack_helper.h>
  64 +#include <net/netfilter/nf_flow_table.h>
  65 +
  66 +struct xt_flowoffload_hook {
  67 +       struct hlist_node list;
  68 +       struct nf_hook_ops ops;
  69 +       struct net *net;
  70 +       bool registered;
  71 +       bool used;
  72 +};
  73 +
  74 +struct xt_flowoffload_table {
  75 +       struct nf_flowtable ft;
  76 +       struct hlist_head hooks;
  77 +       struct delayed_work work;
  78 +};
  79 +
  80 +struct nf_forward_info {
  81 +       const struct net_device *indev;
  82 +       const struct net_device *outdev;
  83 +       const struct net_device *hw_outdev;
  84 +       struct id {
  85 +               __u16   id;
  86 +               __be16  proto;
  87 +       } encap[NF_FLOW_TABLE_ENCAP_MAX];
  88 +       u8 num_encaps;
  89 +       u8 ingress_vlans;
  90 +       u8 h_source[ETH_ALEN];
  91 +       u8 h_dest[ETH_ALEN];
  92 +       enum flow_offload_xmit_type xmit_type;
  93 +};
  94 +
  95 +static DEFINE_SPINLOCK(hooks_lock);
  96 +
  97 +struct xt_flowoffload_table flowtable[2];
  98 +
  99 +static unsigned int
 100 +xt_flowoffload_net_hook(void *priv, struct sk_buff *skb,
 101 +                       const struct nf_hook_state *state)
 102 +{
 103 +       struct vlan_ethhdr *veth;
 104 +       __be16 proto;
 105 +
 106 +       switch (skb->protocol) {
 107 +       case htons(ETH_P_8021Q):
 108 +               veth = (struct vlan_ethhdr *)skb_mac_header(skb);
 109 +               proto = veth->h_vlan_encapsulated_proto;
 110 +               break;
 111 +       case htons(ETH_P_PPP_SES):
 112 +               if (!nf_flow_pppoe_proto(skb, &proto))
 113 +                       return NF_ACCEPT;
 114 +               break;
 115 +       default:
 116 +               proto = skb->protocol;
 117 +               break;
 118 +       }
 119 +
 120 +       switch (proto) {
 121 +       case htons(ETH_P_IP):
 122 +               return nf_flow_offload_ip_hook(priv, skb, state);
 123 +       case htons(ETH_P_IPV6):
 124 +               return nf_flow_offload_ipv6_hook(priv, skb, state);
 125 +       }
 126 +
 127 +       return NF_ACCEPT;
 128 +}
 129 +
 130 +static int
 131 +xt_flowoffload_create_hook(struct xt_flowoffload_table *table,
 132 +                          struct net_device *dev)
 133 +{
 134 +       struct xt_flowoffload_hook *hook;
 135 +       struct nf_hook_ops *ops;
 136 +
 137 +       hook = kzalloc(sizeof(*hook), GFP_ATOMIC);
 138 +       if (!hook)
 139 +               return -ENOMEM;
 140 +
 141 +       ops = &hook->ops;
 142 +       ops->pf = NFPROTO_NETDEV;
 143 +       ops->hooknum = NF_NETDEV_INGRESS;
 144 +       ops->priority = 10;
 145 +       ops->priv = &table->ft;
 146 +       ops->hook = xt_flowoffload_net_hook;
 147 +       ops->dev = dev;
 148 +
 149 +       hlist_add_head(&hook->list, &table->hooks);
 150 +       mod_delayed_work(system_power_efficient_wq, &table->work, 0);
 151 +
 152 +       return 0;
 153 +}
 154 +
 155 +static struct xt_flowoffload_hook *
 156 +flow_offload_lookup_hook(struct xt_flowoffload_table *table,
 157 +                        struct net_device *dev)
 158 +{
 159 +       struct xt_flowoffload_hook *hook;
 160 +
 161 +       hlist_for_each_entry(hook, &table->hooks, list) {
 162 +               if (hook->ops.dev == dev)
 163 +                       return hook;
 164 +       }
 165 +
 166 +       return NULL;
 167 +}
 168 +
 169 +static void
 170 +xt_flowoffload_check_device(struct xt_flowoffload_table *table,
 171 +                           struct net_device *dev)
 172 +{
 173 +       struct xt_flowoffload_hook *hook;
 174 +
 175 +       if (!dev)
 176 +               return;
 177 +
 178 +       spin_lock_bh(&hooks_lock);
 179 +       hook = flow_offload_lookup_hook(table, dev);
 180 +       if (hook)
 181 +               hook->used = true;
 182 +       else
 183 +               xt_flowoffload_create_hook(table, dev);
 184 +       spin_unlock_bh(&hooks_lock);
 185 +}
 186 +
 187 +static void
 188 +xt_flowoffload_register_hooks(struct xt_flowoffload_table *table)
 189 +{
 190 +       struct xt_flowoffload_hook *hook;
 191 +
 192 +restart:
 193 +       hlist_for_each_entry(hook, &table->hooks, list) {
 194 +               if (hook->registered)
 195 +                       continue;
 196 +
 197 +               hook->registered = true;
 198 +               hook->net = dev_net(hook->ops.dev);
 199 +               spin_unlock_bh(&hooks_lock);
 200 +               nf_register_net_hook(hook->net, &hook->ops);
 201 +               if (table->ft.flags & NF_FLOWTABLE_HW_OFFLOAD)
 202 +                       table->ft.type->setup(&table->ft, hook->ops.dev,
 203 +                                             FLOW_BLOCK_BIND);
 204 +               spin_lock_bh(&hooks_lock);
 205 +               goto restart;
 206 +       }
 207 +
 208 +}
 209 +
 210 +static bool
 211 +xt_flowoffload_cleanup_hooks(struct xt_flowoffload_table *table)
 212 +{
 213 +       struct xt_flowoffload_hook *hook;
 214 +       bool active = false;
 215 +
 216 +restart:
 217 +       spin_lock_bh(&hooks_lock);
 218 +       hlist_for_each_entry(hook, &table->hooks, list) {
 219 +               if (hook->used || !hook->registered) {
 220 +                       active = true;
 221 +                       continue;
 222 +               }
 223 +
 224 +               hlist_del(&hook->list);
 225 +               spin_unlock_bh(&hooks_lock);
 226 +               if (table->ft.flags & NF_FLOWTABLE_HW_OFFLOAD)
 227 +                       table->ft.type->setup(&table->ft, hook->ops.dev,
 228 +                                             FLOW_BLOCK_UNBIND);
 229 +               nf_unregister_net_hook(hook->net, &hook->ops);
 230 +               kfree(hook);
 231 +               goto restart;
 232 +       }
 233 +       spin_unlock_bh(&hooks_lock);
 234 +
 235 +       return active;
 236 +}
 237 +
 238 +static void
 239 +xt_flowoffload_check_hook(struct nf_flowtable *flowtable,
 240 +                         struct flow_offload *flow, void *data)
 241 +{
 242 +       struct xt_flowoffload_table *table;
 243 +       struct flow_offload_tuple *tuple0 = &flow->tuplehash[0].tuple;
 244 +       struct flow_offload_tuple *tuple1 = &flow->tuplehash[1].tuple;
 245 +       struct xt_flowoffload_hook *hook;
 246 +
 247 +       table = container_of(flowtable, struct xt_flowoffload_table, ft);
 248 +
 249 +       spin_lock_bh(&hooks_lock);
 250 +       hlist_for_each_entry(hook, &table->hooks, list) {
 251 +               if (hook->ops.dev->ifindex != tuple0->iifidx &&
 252 +                   hook->ops.dev->ifindex != tuple1->iifidx)
 253 +                       continue;
 254 +
 255 +               hook->used = true;
 256 +       }
 257 +       spin_unlock_bh(&hooks_lock);
 258 +}
 259 +
 260 +static void
 261 +xt_flowoffload_hook_work(struct work_struct *work)
 262 +{
 263 +       struct xt_flowoffload_table *table;
 264 +       struct xt_flowoffload_hook *hook;
 265 +       int err;
 266 +
 267 +       table = container_of(work, struct xt_flowoffload_table, work.work);
 268 +
 269 +       spin_lock_bh(&hooks_lock);
 270 +       xt_flowoffload_register_hooks(table);
 271 +       hlist_for_each_entry(hook, &table->hooks, list)
 272 +               hook->used = false;
 273 +       spin_unlock_bh(&hooks_lock);
 274 +
 275 +       err = nf_flow_table_iterate(&table->ft, xt_flowoffload_check_hook,
 276 +                                   NULL);
 277 +       if (err && err != -EAGAIN)
 278 +               goto out;
 279 +
 280 +       if (!xt_flowoffload_cleanup_hooks(table))
 281 +               return;
 282 +
 283 +out:
 284 +       queue_delayed_work(system_power_efficient_wq, &table->work, HZ);
 285 +}
 286 +
 287 +static bool
 288 +xt_flowoffload_skip(struct sk_buff *skb, int family)
 289 +{
 290 +       if (skb_sec_path(skb))
 291 +               return true;
 292 +
 293 +       if (family == NFPROTO_IPV4) {
 294 +               const struct ip_options *opt = &(IPCB(skb)->opt);
 295 +
 296 +               if (unlikely(opt->optlen))
 297 +                       return true;
 298 +       }
 299 +
 300 +       return false;
 301 +}
 302 +
 303 +static enum flow_offload_xmit_type nf_xmit_type(struct dst_entry *dst)
 304 +{
 305 +       if (dst_xfrm(dst))
 306 +               return FLOW_OFFLOAD_XMIT_XFRM;
 307 +
 308 +       return FLOW_OFFLOAD_XMIT_NEIGH;
 309 +}
 310 +
 311 +static void nf_default_forward_path(struct nf_flow_route *route,
 312 +                                   struct dst_entry *dst_cache,
 313 +                                   enum ip_conntrack_dir dir,
 314 +                                   struct net_device **dev)
 315 +{
 316 +       dev[!dir] = dst_cache->dev;
 317 +       route->tuple[!dir].in.ifindex   = dst_cache->dev->ifindex;
 318 +       route->tuple[dir].dst           = dst_cache;
 319 +       route->tuple[dir].xmit_type     = nf_xmit_type(dst_cache);
 320 +}
 321 +
 322 +static bool nf_is_valid_ether_device(const struct net_device *dev)
 323 +{
 324 +       if (!dev || (dev->flags & IFF_LOOPBACK) || dev->type != ARPHRD_ETHER ||
 325 +           dev->addr_len != ETH_ALEN || !is_valid_ether_addr(dev->dev_addr))
 326 +               return false;
 327 +
 328 +       return true;
 329 +}
 330 +
 331 +static void nf_dev_path_info(const struct net_device_path_stack *stack,
 332 +                            struct nf_forward_info *info,
 333 +                            unsigned char *ha)
 334 +{
 335 +       const struct net_device_path *path;
 336 +       int i;
 337 +
 338 +       memcpy(info->h_dest, ha, ETH_ALEN);
 339 +
 340 +       for (i = 0; i < stack->num_paths; i++) {
 341 +               path = &stack->path[i];
 342 +               switch (path->type) {
 343 +               case DEV_PATH_ETHERNET:
 344 +               case DEV_PATH_DSA:
 345 +               case DEV_PATH_VLAN:
 346 +               case DEV_PATH_PPPOE:
 347 +                       info->indev = path->dev;
 348 +                       if (is_zero_ether_addr(info->h_source))
 349 +                               memcpy(info->h_source, path->dev->dev_addr, ETH_ALEN);
 350 +
 351 +                       if (path->type == DEV_PATH_ETHERNET)
 352 +                               break;
 353 +                       if (path->type == DEV_PATH_DSA) {
 354 +                               i = stack->num_paths;
 355 +                               break;
 356 +                       }
 357 +
 358 +                       /* DEV_PATH_VLAN and DEV_PATH_PPPOE */
 359 +                       if (info->num_encaps >= NF_FLOW_TABLE_ENCAP_MAX) {
 360 +                               info->indev = NULL;
 361 +                               break;
 362 +                       }
 363 +                       if (!info->outdev)
 364 +                               info->outdev = path->dev;
 365 +                       info->encap[info->num_encaps].id = path->encap.id;
 366 +                       info->encap[info->num_encaps].proto = path->encap.proto;
 367 +                       info->num_encaps++;
 368 +                       if (path->type == DEV_PATH_PPPOE)
 369 +                               memcpy(info->h_dest, path->encap.h_dest, ETH_ALEN);
 370 +                       break;
 371 +               case DEV_PATH_BRIDGE:
 372 +                       if (is_zero_ether_addr(info->h_source))
 373 +                               memcpy(info->h_source, path->dev->dev_addr, ETH_ALEN);
 374 +
 375 +                       switch (path->bridge.vlan_mode) {
 376 +                       case DEV_PATH_BR_VLAN_UNTAG_HW:
 377 +                               info->ingress_vlans |= BIT(info->num_encaps - 1);
 378 +                               break;
 379 +                       case DEV_PATH_BR_VLAN_TAG:
 380 +                               info->encap[info->num_encaps].id = path->bridge.vlan_id;
 381 +                               info->encap[info->num_encaps].proto = path->bridge.vlan_proto;
 382 +                               info->num_encaps++;
 383 +                               break;
 384 +                       case DEV_PATH_BR_VLAN_UNTAG:
 385 +                               info->num_encaps--;
 386 +                               break;
 387 +                       case DEV_PATH_BR_VLAN_KEEP:
 388 +                               break;
 389 +                       }
 390 +                       break;
 391 +               default:
 392 +                       info->indev = NULL;
 393 +                       break;
 394 +               }
 395 +       }
 396 +       if (!info->outdev)
 397 +               info->outdev = info->indev;
 398 +
 399 +       info->hw_outdev = info->indev;
 400 +
 401 +       if (nf_is_valid_ether_device(info->indev))
 402 +               info->xmit_type = FLOW_OFFLOAD_XMIT_DIRECT;
 403 +}
 404 +
 405 +static int nf_dev_fill_forward_path(const struct nf_flow_route *route,
 406 +                                    const struct dst_entry *dst_cache,
 407 +                                    const struct nf_conn *ct,
 408 +                                    enum ip_conntrack_dir dir, u8 *ha,
 409 +                                    struct net_device_path_stack *stack)
 410 +{
 411 +       const void *daddr = &ct->tuplehash[!dir].tuple.src.u3;
 412 +       struct net_device *dev = dst_cache->dev;
 413 +       struct neighbour *n;
 414 +       u8 nud_state;
 415 +
 416 +       if (!nf_is_valid_ether_device(dev))
 417 +               goto out;
 418 +
 419 +       n = dst_neigh_lookup(dst_cache, daddr);
 420 +       if (!n)
 421 +               return -1;
 422 +
 423 +       read_lock_bh(&n->lock);
 424 +       nud_state = n->nud_state;
 425 +       ether_addr_copy(ha, n->ha);
 426 +       read_unlock_bh(&n->lock);
 427 +       neigh_release(n);
 428 +
 429 +       if (!(nud_state & NUD_VALID))
 430 +               return -1;
 431 +
 432 +out:
 433 +       return dev_fill_forward_path(dev, ha, stack);
 434 +}
 435 +
 436 +static void nf_dev_forward_path(struct nf_flow_route *route,
 437 +                               const struct nf_conn *ct,
 438 +                               enum ip_conntrack_dir dir,
 439 +                               struct net_device **devs)
 440 +{
 441 +       const struct dst_entry *dst = route->tuple[dir].dst;
 442 +       struct net_device_path_stack stack;
 443 +       struct nf_forward_info info = {};
 444 +       unsigned char ha[ETH_ALEN];
 445 +       int i;
 446 +
 447 +       if (nf_dev_fill_forward_path(route, dst, ct, dir, ha, &stack) >= 0)
 448 +               nf_dev_path_info(&stack, &info, ha);
 449 +
 450 +       devs[!dir] = (struct net_device *)info.indev;
 451 +       if (!info.indev)
 452 +               return;
 453 +
 454 +       route->tuple[!dir].in.ifindex = info.indev->ifindex;
 455 +       for (i = 0; i < info.num_encaps; i++) {
 456 +               route->tuple[!dir].in.encap[i].id = info.encap[i].id;
 457 +               route->tuple[!dir].in.encap[i].proto = info.encap[i].proto;
 458 +       }
 459 +       route->tuple[!dir].in.num_encaps = info.num_encaps;
 460 +       route->tuple[!dir].in.ingress_vlans = info.ingress_vlans;
 461 +
 462 +       if (info.xmit_type == FLOW_OFFLOAD_XMIT_DIRECT) {
 463 +               memcpy(route->tuple[dir].out.h_source, info.h_source, ETH_ALEN);
 464 +               memcpy(route->tuple[dir].out.h_dest, info.h_dest, ETH_ALEN);
 465 +               route->tuple[dir].out.ifindex = info.outdev->ifindex;
 466 +               route->tuple[dir].out.hw_ifindex = info.hw_outdev->ifindex;
 467 +               route->tuple[dir].xmit_type = info.xmit_type;
 468 +       }
 469 +}
 470 +
 471 +static int
 472 +xt_flowoffload_route(struct sk_buff *skb, const struct nf_conn *ct,
 473 +                    const struct xt_action_param *par,
 474 +                    struct nf_flow_route *route, enum ip_conntrack_dir dir,
 475 +                    struct net_device **devs)
 476 +{
 477 +       struct dst_entry *this_dst = skb_dst(skb);
 478 +       struct dst_entry *other_dst = NULL;
 479 +       struct flowi fl;
 480 +
 481 +       memset(&fl, 0, sizeof(fl));
 482 +       switch (xt_family(par)) {
 483 +       case NFPROTO_IPV4:
 484 +               fl.u.ip4.daddr = ct->tuplehash[dir].tuple.src.u3.ip;
 485 +               fl.u.ip4.flowi4_oif = xt_in(par)->ifindex;
 486 +               break;
 487 +       case NFPROTO_IPV6:
 488 +               fl.u.ip6.saddr = ct->tuplehash[!dir].tuple.dst.u3.in6;
 489 +               fl.u.ip6.daddr = ct->tuplehash[dir].tuple.src.u3.in6;
 490 +               fl.u.ip6.flowi6_oif = xt_in(par)->ifindex;
 491 +               break;
 492 +       }
 493 +
 494 +       if (!dst_hold_safe(this_dst))
 495 +               return -ENOENT;
 496 +
 497 +       nf_route(xt_net(par), &other_dst, &fl, false, xt_family(par));
 498 +       if (!other_dst) {
 499 +               dst_release(this_dst);
 500 +               return -ENOENT;
 501 +       }
 502 +
 503 +       nf_default_forward_path(route, this_dst, dir, devs);
 504 +       nf_default_forward_path(route, other_dst, !dir, devs);
 505 +
 506 +       if (route->tuple[dir].xmit_type == FLOW_OFFLOAD_XMIT_NEIGH &&
 507 +           route->tuple[!dir].xmit_type == FLOW_OFFLOAD_XMIT_NEIGH) {
 508 +               nf_dev_forward_path(route, ct, dir, devs);
 509 +               nf_dev_forward_path(route, ct, !dir, devs);
 510 +       }
 511 +
 512 +       return 0;
 513 +}
 514 +
 515 +static unsigned int
 516 +flowoffload_tg(struct sk_buff *skb, const struct xt_action_param *par)
 517 +{
 518 +       struct xt_flowoffload_table *table;
 519 +       const struct xt_flowoffload_target_info *info = par->targinfo;
 520 +       struct tcphdr _tcph, *tcph = NULL;
 521 +       enum ip_conntrack_info ctinfo;
 522 +       enum ip_conntrack_dir dir;
 523 +       struct nf_flow_route route = {};
 524 +       struct flow_offload *flow = NULL;
 525 +       struct net_device *devs[2] = {};
 526 +       struct nf_conn *ct;
 527 +       struct net *net;
 528 +
 529 +       if (xt_flowoffload_skip(skb, xt_family(par)))
 530 +               return XT_CONTINUE;
 531 +
 532 +       ct = nf_ct_get(skb, &ctinfo);
 533 +       if (ct == NULL)
 534 +               return XT_CONTINUE;
 535 +
 536 +       switch (ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.protonum) {
 537 +       case IPPROTO_TCP:
 538 +               if (ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED)
 539 +                       return XT_CONTINUE;
 540 +
 541 +               tcph = skb_header_pointer(skb, par->thoff,
 542 +                                         sizeof(_tcph), &_tcph);
 543 +               if (unlikely(!tcph || tcph->fin || tcph->rst))
 544 +                       return XT_CONTINUE;
 545 +               break;
 546 +       case IPPROTO_UDP:
 547 +               break;
 548 +       default:
 549 +               return XT_CONTINUE;
 550 +       }
 551 +
 552 +       if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
 553 +           ct->status & (IPS_SEQ_ADJUST | IPS_NAT_CLASH))
 554 +               return XT_CONTINUE;
 555 +
 556 +       if (!nf_ct_is_confirmed(ct))
 557 +               return XT_CONTINUE;
 558 +
 559 +       dir = CTINFO2DIR(ctinfo);
 560 +
 561 +       devs[dir] = xt_out(par);
 562 +       devs[!dir] = xt_in(par);
 563 +
 564 +       if (!devs[dir] || !devs[!dir])
 565 +               return XT_CONTINUE;
 566 +
 567 +       if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
 568 +               return XT_CONTINUE;
 569 +
 570 +       if (xt_flowoffload_route(skb, ct, par, &route, dir, devs) < 0)
 571 +               goto err_flow_route;
 572 +
 573 +       flow = flow_offload_alloc(ct);
 574 +       if (!flow)
 575 +               goto err_flow_alloc;
 576 +
 577 +       flow_offload_route_init(flow, &route);
 578 +
 579 +       if (tcph) {
 580 +               ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
 581 +               ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
 582 +       }
 583 +
 584 +       table = &flowtable[!!(info->flags & XT_FLOWOFFLOAD_HW)];
 585 +
 586 +       net = read_pnet(&table->ft.net);
 587 +       if (!net)
 588 +               write_pnet(&table->ft.net, xt_net(par));
 589 +
 590 +       __set_bit(NF_FLOW_HW_BIDIRECTIONAL, &flow->flags);
 591 +       if (flow_offload_add(&table->ft, flow) < 0)
 592 +               goto err_flow_add;
 593 +
 594 +       xt_flowoffload_check_device(table, devs[0]);
 595 +       xt_flowoffload_check_device(table, devs[1]);
 596 +
 597 +       return XT_CONTINUE;
 598 +
 599 +err_flow_add:
 600 +       flow_offload_free(flow);
 601 +err_flow_alloc:
 602 +       dst_release(route.tuple[dir].dst);
 603 +       dst_release(route.tuple[!dir].dst);
 604 +err_flow_route:
 605 +       clear_bit(IPS_OFFLOAD_BIT, &ct->status);
 606 +
 607 +       return XT_CONTINUE;
 608 +}
 609 +
 610 +static int flowoffload_chk(const struct xt_tgchk_param *par)
 611 +{
 612 +       struct xt_flowoffload_target_info *info = par->targinfo;
 613 +
 614 +       if (info->flags & ~XT_FLOWOFFLOAD_MASK)
 615 +               return -EINVAL;
 616 +
 617 +       return 0;
 618 +}
 619 +
 620 +static struct xt_target offload_tg_reg __read_mostly = {
 621 +       .family         = NFPROTO_UNSPEC,
 622 +       .name           = "FLOWOFFLOAD",
 623 +       .revision       = 0,
 624 +       .targetsize     = sizeof(struct xt_flowoffload_target_info),
 625 +       .usersize       = sizeof(struct xt_flowoffload_target_info),
 626 +       .checkentry     = flowoffload_chk,
 627 +       .target         = flowoffload_tg,
 628 +       .me             = THIS_MODULE,
 629 +};
 630 +
 631 +static int flow_offload_netdev_event(struct notifier_block *this,
 632 +                                    unsigned long event, void *ptr)
 633 +{
 634 +       struct xt_flowoffload_hook *hook0, *hook1;
 635 +       struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 636 +
 637 +       if (event != NETDEV_UNREGISTER)
 638 +               return NOTIFY_DONE;
 639 +
 640 +       spin_lock_bh(&hooks_lock);
 641 +       hook0 = flow_offload_lookup_hook(&flowtable[0], dev);
 642 +       if (hook0)
 643 +               hlist_del(&hook0->list);
 644 +
 645 +       hook1 = flow_offload_lookup_hook(&flowtable[1], dev);
 646 +       if (hook1)
 647 +               hlist_del(&hook1->list);
 648 +       spin_unlock_bh(&hooks_lock);
 649 +
 650 +       if (hook0) {
 651 +               nf_unregister_net_hook(hook0->net, &hook0->ops);
 652 +               kfree(hook0);
 653 +       }
 654 +
 655 +       if (hook1) {
 656 +               nf_unregister_net_hook(hook1->net, &hook1->ops);
 657 +               kfree(hook1);
 658 +       }
 659 +
 660 +       nf_flow_table_cleanup(dev);
 661 +
 662 +       return NOTIFY_DONE;
 663 +}
 664 +
 665 +static struct notifier_block flow_offload_netdev_notifier = {
 666 +       .notifier_call  = flow_offload_netdev_event,
 667 +};
 668 +
 669 +static int nf_flow_rule_route_inet(struct net *net,
 670 +                                  struct flow_offload *flow,
 671 +                                  enum flow_offload_tuple_dir dir,
 672 +                                  struct nf_flow_rule *flow_rule)
 673 +{
 674 +       const struct flow_offload_tuple *flow_tuple = &flow->tuplehash[dir].tuple;
 675 +       int err;
 676 +
 677 +       switch (flow_tuple->l3proto) {
 678 +       case NFPROTO_IPV4:
 679 +               err = nf_flow_rule_route_ipv4(net, flow, dir, flow_rule);
 680 +               break;
 681 +       case NFPROTO_IPV6:
 682 +               err = nf_flow_rule_route_ipv6(net, flow, dir, flow_rule);
 683 +               break;
 684 +       default:
 685 +               err = -1;
 686 +               break;
 687 +       }
 688 +
 689 +       return err;
 690 +}
 691 +
 692 +static struct nf_flowtable_type flowtable_inet = {
 693 +       .family         = NFPROTO_INET,
 694 +       .init           = nf_flow_table_init,
 695 +       .setup          = nf_flow_table_offload_setup,
 696 +       .action         = nf_flow_rule_route_inet,
 697 +       .free           = nf_flow_table_free,
 698 +       .hook           = xt_flowoffload_net_hook,
 699 +       .owner          = THIS_MODULE,
 700 +};
 701 +
 702 +static int init_flowtable(struct xt_flowoffload_table *tbl)
 703 +{
 704 +       INIT_DELAYED_WORK(&tbl->work, xt_flowoffload_hook_work);
 705 +       tbl->ft.type = &flowtable_inet;
 706 +       tbl->ft.flags = NF_FLOWTABLE_COUNTER;
 707 +
 708 +       return nf_flow_table_init(&tbl->ft);
 709 +}
 710 +
 711 +static int __init xt_flowoffload_tg_init(void)
 712 +{
 713 +       int ret;
 714 +
 715 +       register_netdevice_notifier(&flow_offload_netdev_notifier);
 716 +
 717 +       ret = init_flowtable(&flowtable[0]);
 718 +       if (ret)
 719 +               return ret;
 720 +
 721 +       ret = init_flowtable(&flowtable[1]);
 722 +       if (ret)
 723 +               goto cleanup;
 724 +
 725 +       flowtable[1].ft.flags |= NF_FLOWTABLE_HW_OFFLOAD;
 726 +
 727 +       ret = xt_register_target(&offload_tg_reg);
 728 +       if (ret)
 729 +               goto cleanup2;
 730 +
 731 +       return 0;
 732 +
 733 +cleanup2:
 734 +       nf_flow_table_free(&flowtable[1].ft);
 735 +cleanup:
 736 +       nf_flow_table_free(&flowtable[0].ft);
 737 +       return ret;
 738 +}
 739 +
 740 +static void __exit xt_flowoffload_tg_exit(void)
 741 +{
 742 +       xt_unregister_target(&offload_tg_reg);
 743 +       unregister_netdevice_notifier(&flow_offload_netdev_notifier);
 744 +       nf_flow_table_free(&flowtable[0].ft);
 745 +       nf_flow_table_free(&flowtable[1].ft);
 746 +}
 747 +
 748 +MODULE_LICENSE("GPL");
 749 +module_init(xt_flowoffload_tg_init);
 750 +module_exit(xt_flowoffload_tg_exit);
 751 --- a/net/netfilter/nf_flow_table_core.c
 752 +++ b/net/netfilter/nf_flow_table_core.c
 753 @@ -7,7 +7,6 @@
 754  #include <linux/netdevice.h>
 755  #include <net/ip.h>
 756  #include <net/ip6_route.h>
 757 -#include <net/netfilter/nf_tables.h>
 758  #include <net/netfilter/nf_flow_table.h>
 759  #include <net/netfilter/nf_conntrack.h>
 760  #include <net/netfilter/nf_conntrack_core.h>
 761 @@ -377,8 +376,7 @@ flow_offload_lookup(struct nf_flowtable
 762  }
 763  EXPORT_SYMBOL_GPL(flow_offload_lookup);
 764
 765 -static int
 766 -nf_flow_table_iterate(struct nf_flowtable *flow_table,
 767 +int nf_flow_table_iterate(struct nf_flowtable *flow_table,
 768                       void (*iter)(struct nf_flowtable *flowtable,
 769                                    struct flow_offload *flow, void *data),
 770                       void *data)
 771 @@ -439,6 +437,7 @@ static void nf_flow_offload_gc_step(stru
 772                 nf_flow_offload_stats(flow_table, flow);
 773         }
 774  }
 775 +EXPORT_SYMBOL_GPL(nf_flow_table_iterate);
 776
 777  void nf_flow_table_gc_run(struct nf_flowtable *flow_table)
 778  {
 779 --- /dev/null
 780 +++ b/include/uapi/linux/netfilter/xt_FLOWOFFLOAD.h
 781 @@ -0,0 +1,17 @@
 782 +/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
 783 +#ifndef _XT_FLOWOFFLOAD_H
 784 +#define _XT_FLOWOFFLOAD_H
 785 +
 786 +#include <linux/types.h>
 787 +
 788 +enum {
 789 +       XT_FLOWOFFLOAD_HW       = 1 << 0,
 790 +
 791 +       XT_FLOWOFFLOAD_MASK     = XT_FLOWOFFLOAD_HW
 792 +};
 793 +
 794 +struct xt_flowoffload_target_info {
 795 +       __u32 flags;
 796 +};
 797 +
 798 +#endif /* _XT_FLOWOFFLOAD_H */
 799 --- a/include/net/netfilter/nf_flow_table.h
 800 +++ b/include/net/netfilter/nf_flow_table.h
 801 @@ -293,6 +293,11 @@ void nf_flow_table_free(struct nf_flowta
 802
 803  void flow_offload_teardown(struct flow_offload *flow);
 804
 805 +int nf_flow_table_iterate(struct nf_flowtable *flow_table,
 806 +                      void (*iter)(struct nf_flowtable *flowtable,
 807 +                                   struct flow_offload *flow, void *data),
 808 +                      void *data);
 809 +
 810  void nf_flow_snat_port(const struct flow_offload *flow,
 811                        struct sk_buff *skb, unsigned int thoff,
 812                        u8 protocol, enum flow_offload_tuple_dir dir);