target/linux/generic/hack-6.1/650-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 @@ -1023,6 +1023,15 @@ config NETFILTER_XT_TARGET_NOTRACK
  12         depends on NETFILTER_ADVANCED
  13         select NETFILTER_XT_TARGET_CT
  14
  15 +config NETFILTER_XT_TARGET_FLOWOFFLOAD
  16 +       tristate '"FLOWOFFLOAD" target support'
  17 +       depends on NF_FLOW_TABLE
  18 +       depends on NETFILTER_INGRESS
  19 +       help
  20 +         This option adds a `FLOWOFFLOAD' target, which uses the nf_flow_offload
  21 +         module to speed up processing of packets by bypassing the usual
  22 +         netfilter chains
  23 +
  24  config NETFILTER_XT_TARGET_RATEEST
  25         tristate '"RATEEST" target support'
  26         depends on NETFILTER_ADVANCED
  27 --- a/net/netfilter/Makefile
  28 +++ b/net/netfilter/Makefile
  29 @@ -154,6 +154,7 @@ obj-$(CONFIG_NETFILTER_XT_TARGET_CLASSIF
  30  obj-$(CONFIG_NETFILTER_XT_TARGET_CONNSECMARK) += xt_CONNSECMARK.o
  31  obj-$(CONFIG_NETFILTER_XT_TARGET_CT) += xt_CT.o
  32  obj-$(CONFIG_NETFILTER_XT_TARGET_DSCP) += xt_DSCP.o
  33 +obj-$(CONFIG_NETFILTER_XT_TARGET_FLOWOFFLOAD) += xt_FLOWOFFLOAD.o
  34  obj-$(CONFIG_NETFILTER_XT_TARGET_HL) += xt_HL.o
  35  obj-$(CONFIG_NETFILTER_XT_TARGET_HMARK) += xt_HMARK.o
  36  obj-$(CONFIG_NETFILTER_XT_TARGET_LED) += xt_LED.o
  37 --- /dev/null
  38 +++ b/net/netfilter/xt_FLOWOFFLOAD.c
  39 @@ -0,0 +1,698 @@
  40 +/*
  41 + * Copyright (C) 2018-2021 Felix Fietkau <nbd@nbd.name>
  42 + *
  43 + * This program is free software; you can redistribute it and/or modify
  44 + * it under the terms of the GNU General Public License version 2 as
  45 + * published by the Free Software Foundation.
  46 + */
  47 +#include <linux/module.h>
  48 +#include <linux/init.h>
  49 +#include <linux/netfilter.h>
  50 +#include <linux/netfilter/xt_FLOWOFFLOAD.h>
  51 +#include <linux/if_vlan.h>
  52 +#include <net/ip.h>
  53 +#include <net/netfilter/nf_conntrack.h>
  54 +#include <net/netfilter/nf_conntrack_extend.h>
  55 +#include <net/netfilter/nf_conntrack_helper.h>
  56 +#include <net/netfilter/nf_flow_table.h>
  57 +
  58 +struct xt_flowoffload_hook {
  59 +       struct hlist_node list;
  60 +       struct nf_hook_ops ops;
  61 +       struct net *net;
  62 +       bool registered;
  63 +       bool used;
  64 +};
  65 +
  66 +struct xt_flowoffload_table {
  67 +       struct nf_flowtable ft;
  68 +       struct hlist_head hooks;
  69 +       struct delayed_work work;
  70 +};
  71 +
  72 +struct nf_forward_info {
  73 +       const struct net_device *indev;
  74 +       const struct net_device *outdev;
  75 +       const struct net_device *hw_outdev;
  76 +       struct id {
  77 +               __u16   id;
  78 +               __be16  proto;
  79 +       } encap[NF_FLOW_TABLE_ENCAP_MAX];
  80 +       u8 num_encaps;
  81 +       u8 ingress_vlans;
  82 +       u8 h_source[ETH_ALEN];
  83 +       u8 h_dest[ETH_ALEN];
  84 +       enum flow_offload_xmit_type xmit_type;
  85 +};
  86 +
  87 +static DEFINE_SPINLOCK(hooks_lock);
  88 +
  89 +struct xt_flowoffload_table flowtable[2];
  90 +
  91 +static unsigned int
  92 +xt_flowoffload_net_hook(void *priv, struct sk_buff *skb,
  93 +                       const struct nf_hook_state *state)
  94 +{
  95 +       struct vlan_ethhdr *veth;
  96 +       __be16 proto;
  97 +
  98 +       switch (skb->protocol) {
  99 +       case htons(ETH_P_8021Q):
 100 +               veth = (struct vlan_ethhdr *)skb_mac_header(skb);
 101 +               proto = veth->h_vlan_encapsulated_proto;
 102 +               break;
 103 +       case htons(ETH_P_PPP_SES):
 104 +               proto = nf_flow_pppoe_proto(skb);
 105 +               break;
 106 +       default:
 107 +               proto = skb->protocol;
 108 +               break;
 109 +       }
 110 +
 111 +       switch (proto) {
 112 +       case htons(ETH_P_IP):
 113 +               return nf_flow_offload_ip_hook(priv, skb, state);
 114 +       case htons(ETH_P_IPV6):
 115 +               return nf_flow_offload_ipv6_hook(priv, skb, state);
 116 +       }
 117 +
 118 +       return NF_ACCEPT;
 119 +}
 120 +
 121 +static int
 122 +xt_flowoffload_create_hook(struct xt_flowoffload_table *table,
 123 +                          struct net_device *dev)
 124 +{
 125 +       struct xt_flowoffload_hook *hook;
 126 +       struct nf_hook_ops *ops;
 127 +
 128 +       hook = kzalloc(sizeof(*hook), GFP_ATOMIC);
 129 +       if (!hook)
 130 +               return -ENOMEM;
 131 +
 132 +       ops = &hook->ops;
 133 +       ops->pf = NFPROTO_NETDEV;
 134 +       ops->hooknum = NF_NETDEV_INGRESS;
 135 +       ops->priority = 10;
 136 +       ops->priv = &table->ft;
 137 +       ops->hook = xt_flowoffload_net_hook;
 138 +       ops->dev = dev;
 139 +
 140 +       hlist_add_head(&hook->list, &table->hooks);
 141 +       mod_delayed_work(system_power_efficient_wq, &table->work, 0);
 142 +
 143 +       return 0;
 144 +}
 145 +
 146 +static struct xt_flowoffload_hook *
 147 +flow_offload_lookup_hook(struct xt_flowoffload_table *table,
 148 +                        struct net_device *dev)
 149 +{
 150 +       struct xt_flowoffload_hook *hook;
 151 +
 152 +       hlist_for_each_entry(hook, &table->hooks, list) {
 153 +               if (hook->ops.dev == dev)
 154 +                       return hook;
 155 +       }
 156 +
 157 +       return NULL;
 158 +}
 159 +
 160 +static void
 161 +xt_flowoffload_check_device(struct xt_flowoffload_table *table,
 162 +                           struct net_device *dev)
 163 +{
 164 +       struct xt_flowoffload_hook *hook;
 165 +
 166 +       if (!dev)
 167 +               return;
 168 +
 169 +       spin_lock_bh(&hooks_lock);
 170 +       hook = flow_offload_lookup_hook(table, dev);
 171 +       if (hook)
 172 +               hook->used = true;
 173 +       else
 174 +               xt_flowoffload_create_hook(table, dev);
 175 +       spin_unlock_bh(&hooks_lock);
 176 +}
 177 +
 178 +static void
 179 +xt_flowoffload_register_hooks(struct xt_flowoffload_table *table)
 180 +{
 181 +       struct xt_flowoffload_hook *hook;
 182 +
 183 +restart:
 184 +       hlist_for_each_entry(hook, &table->hooks, list) {
 185 +               if (hook->registered)
 186 +                       continue;
 187 +
 188 +               hook->registered = true;
 189 +               hook->net = dev_net(hook->ops.dev);
 190 +               spin_unlock_bh(&hooks_lock);
 191 +               nf_register_net_hook(hook->net, &hook->ops);
 192 +               if (table->ft.flags & NF_FLOWTABLE_HW_OFFLOAD)
 193 +                       table->ft.type->setup(&table->ft, hook->ops.dev,
 194 +                                             FLOW_BLOCK_BIND);
 195 +               spin_lock_bh(&hooks_lock);
 196 +               goto restart;
 197 +       }
 198 +
 199 +}
 200 +
 201 +static bool
 202 +xt_flowoffload_cleanup_hooks(struct xt_flowoffload_table *table)
 203 +{
 204 +       struct xt_flowoffload_hook *hook;
 205 +       bool active = false;
 206 +
 207 +restart:
 208 +       spin_lock_bh(&hooks_lock);
 209 +       hlist_for_each_entry(hook, &table->hooks, list) {
 210 +               if (hook->used || !hook->registered) {
 211 +                       active = true;
 212 +                       continue;
 213 +               }
 214 +
 215 +               hlist_del(&hook->list);
 216 +               spin_unlock_bh(&hooks_lock);
 217 +               if (table->ft.flags & NF_FLOWTABLE_HW_OFFLOAD)
 218 +                       table->ft.type->setup(&table->ft, hook->ops.dev,
 219 +                                             FLOW_BLOCK_UNBIND);
 220 +               nf_unregister_net_hook(hook->net, &hook->ops);
 221 +               kfree(hook);
 222 +               goto restart;
 223 +       }
 224 +       spin_unlock_bh(&hooks_lock);
 225 +
 226 +       return active;
 227 +}
 228 +
 229 +static void
 230 +xt_flowoffload_check_hook(struct nf_flowtable *flowtable,
 231 +                         struct flow_offload *flow, void *data)
 232 +{
 233 +       struct xt_flowoffload_table *table;
 234 +       struct flow_offload_tuple *tuple0 = &flow->tuplehash[0].tuple;
 235 +       struct flow_offload_tuple *tuple1 = &flow->tuplehash[1].tuple;
 236 +       struct xt_flowoffload_hook *hook;
 237 +
 238 +       table = container_of(flowtable, struct xt_flowoffload_table, ft);
 239 +
 240 +       spin_lock_bh(&hooks_lock);
 241 +       hlist_for_each_entry(hook, &table->hooks, list) {
 242 +               if (hook->ops.dev->ifindex != tuple0->iifidx &&
 243 +                   hook->ops.dev->ifindex != tuple1->iifidx)
 244 +                       continue;
 245 +
 246 +               hook->used = true;
 247 +       }
 248 +       spin_unlock_bh(&hooks_lock);
 249 +}
 250 +
 251 +static void
 252 +xt_flowoffload_hook_work(struct work_struct *work)
 253 +{
 254 +       struct xt_flowoffload_table *table;
 255 +       struct xt_flowoffload_hook *hook;
 256 +       int err;
 257 +
 258 +       table = container_of(work, struct xt_flowoffload_table, work.work);
 259 +
 260 +       spin_lock_bh(&hooks_lock);
 261 +       xt_flowoffload_register_hooks(table);
 262 +       hlist_for_each_entry(hook, &table->hooks, list)
 263 +               hook->used = false;
 264 +       spin_unlock_bh(&hooks_lock);
 265 +
 266 +       err = nf_flow_table_iterate(&table->ft, xt_flowoffload_check_hook,
 267 +                                   NULL);
 268 +       if (err && err != -EAGAIN)
 269 +               goto out;
 270 +
 271 +       if (!xt_flowoffload_cleanup_hooks(table))
 272 +               return;
 273 +
 274 +out:
 275 +       queue_delayed_work(system_power_efficient_wq, &table->work, HZ);
 276 +}
 277 +
 278 +static bool
 279 +xt_flowoffload_skip(struct sk_buff *skb, int family)
 280 +{
 281 +       if (skb_sec_path(skb))
 282 +               return true;
 283 +
 284 +       if (family == NFPROTO_IPV4) {
 285 +               const struct ip_options *opt = &(IPCB(skb)->opt);
 286 +
 287 +               if (unlikely(opt->optlen))
 288 +                       return true;
 289 +       }
 290 +
 291 +       return false;
 292 +}
 293 +
 294 +static enum flow_offload_xmit_type nf_xmit_type(struct dst_entry *dst)
 295 +{
 296 +       if (dst_xfrm(dst))
 297 +               return FLOW_OFFLOAD_XMIT_XFRM;
 298 +
 299 +       return FLOW_OFFLOAD_XMIT_NEIGH;
 300 +}
 301 +
 302 +static void nf_default_forward_path(struct nf_flow_route *route,
 303 +                                   struct dst_entry *dst_cache,
 304 +                                   enum ip_conntrack_dir dir,
 305 +                                   struct net_device **dev)
 306 +{
 307 +       dev[!dir] = dst_cache->dev;
 308 +       route->tuple[!dir].in.ifindex   = dst_cache->dev->ifindex;
 309 +       route->tuple[dir].dst           = dst_cache;
 310 +       route->tuple[dir].xmit_type     = nf_xmit_type(dst_cache);
 311 +}
 312 +
 313 +static bool nf_is_valid_ether_device(const struct net_device *dev)
 314 +{
 315 +       if (!dev || (dev->flags & IFF_LOOPBACK) || dev->type != ARPHRD_ETHER ||
 316 +           dev->addr_len != ETH_ALEN || !is_valid_ether_addr(dev->dev_addr))
 317 +               return false;
 318 +
 319 +       return true;
 320 +}
 321 +
 322 +static void nf_dev_path_info(const struct net_device_path_stack *stack,
 323 +                            struct nf_forward_info *info,
 324 +                            unsigned char *ha)
 325 +{
 326 +       const struct net_device_path *path;
 327 +       int i;
 328 +
 329 +       memcpy(info->h_dest, ha, ETH_ALEN);
 330 +
 331 +       for (i = 0; i < stack->num_paths; i++) {
 332 +               path = &stack->path[i];
 333 +               switch (path->type) {
 334 +               case DEV_PATH_ETHERNET:
 335 +               case DEV_PATH_DSA:
 336 +               case DEV_PATH_VLAN:
 337 +               case DEV_PATH_PPPOE:
 338 +                       info->indev = path->dev;
 339 +                       if (is_zero_ether_addr(info->h_source))
 340 +                               memcpy(info->h_source, path->dev->dev_addr, ETH_ALEN);
 341 +
 342 +                       if (path->type == DEV_PATH_ETHERNET)
 343 +                               break;
 344 +                       if (path->type == DEV_PATH_DSA) {
 345 +                               i = stack->num_paths;
 346 +                               break;
 347 +                       }
 348 +
 349 +                       /* DEV_PATH_VLAN and DEV_PATH_PPPOE */
 350 +                       if (info->num_encaps >= NF_FLOW_TABLE_ENCAP_MAX) {
 351 +                               info->indev = NULL;
 352 +                               break;
 353 +                       }
 354 +                       if (!info->outdev)
 355 +                               info->outdev = path->dev;
 356 +                       info->encap[info->num_encaps].id = path->encap.id;
 357 +                       info->encap[info->num_encaps].proto = path->encap.proto;
 358 +                       info->num_encaps++;
 359 +                       if (path->type == DEV_PATH_PPPOE)
 360 +                               memcpy(info->h_dest, path->encap.h_dest, ETH_ALEN);
 361 +                       break;
 362 +               case DEV_PATH_BRIDGE:
 363 +                       if (is_zero_ether_addr(info->h_source))
 364 +                               memcpy(info->h_source, path->dev->dev_addr, ETH_ALEN);
 365 +
 366 +                       switch (path->bridge.vlan_mode) {
 367 +                       case DEV_PATH_BR_VLAN_UNTAG_HW:
 368 +                               info->ingress_vlans |= BIT(info->num_encaps - 1);
 369 +                               break;
 370 +                       case DEV_PATH_BR_VLAN_TAG:
 371 +                               info->encap[info->num_encaps].id = path->bridge.vlan_id;
 372 +                               info->encap[info->num_encaps].proto = path->bridge.vlan_proto;
 373 +                               info->num_encaps++;
 374 +                               break;
 375 +                       case DEV_PATH_BR_VLAN_UNTAG:
 376 +                               info->num_encaps--;
 377 +                               break;
 378 +                       case DEV_PATH_BR_VLAN_KEEP:
 379 +                               break;
 380 +                       }
 381 +                       break;
 382 +               default:
 383 +                       info->indev = NULL;
 384 +                       break;
 385 +               }
 386 +       }
 387 +       if (!info->outdev)
 388 +               info->outdev = info->indev;
 389 +
 390 +       info->hw_outdev = info->indev;
 391 +
 392 +       if (nf_is_valid_ether_device(info->indev))
 393 +               info->xmit_type = FLOW_OFFLOAD_XMIT_DIRECT;
 394 +}
 395 +
 396 +static int nf_dev_fill_forward_path(const struct nf_flow_route *route,
 397 +                                    const struct dst_entry *dst_cache,
 398 +                                    const struct nf_conn *ct,
 399 +                                    enum ip_conntrack_dir dir, u8 *ha,
 400 +                                    struct net_device_path_stack *stack)
 401 +{
 402 +       const void *daddr = &ct->tuplehash[!dir].tuple.src.u3;
 403 +       struct net_device *dev = dst_cache->dev;
 404 +       struct neighbour *n;
 405 +       u8 nud_state;
 406 +
 407 +       if (!nf_is_valid_ether_device(dev))
 408 +               goto out;
 409 +
 410 +       n = dst_neigh_lookup(dst_cache, daddr);
 411 +       if (!n)
 412 +               return -1;
 413 +
 414 +       read_lock_bh(&n->lock);
 415 +       nud_state = n->nud_state;
 416 +       ether_addr_copy(ha, n->ha);
 417 +       read_unlock_bh(&n->lock);
 418 +       neigh_release(n);
 419 +
 420 +       if (!(nud_state & NUD_VALID))
 421 +               return -1;
 422 +
 423 +out:
 424 +       return dev_fill_forward_path(dev, ha, stack);
 425 +}
 426 +
 427 +static void nf_dev_forward_path(struct nf_flow_route *route,
 428 +                               const struct nf_conn *ct,
 429 +                               enum ip_conntrack_dir dir,
 430 +                               struct net_device **devs)
 431 +{
 432 +       const struct dst_entry *dst = route->tuple[dir].dst;
 433 +       struct net_device_path_stack stack;
 434 +       struct nf_forward_info info = {};
 435 +       unsigned char ha[ETH_ALEN];
 436 +       int i;
 437 +
 438 +       if (nf_dev_fill_forward_path(route, dst, ct, dir, ha, &stack) >= 0)
 439 +               nf_dev_path_info(&stack, &info, ha);
 440 +
 441 +       devs[!dir] = (struct net_device *)info.indev;
 442 +       if (!info.indev)
 443 +               return;
 444 +
 445 +       route->tuple[!dir].in.ifindex = info.indev->ifindex;
 446 +       for (i = 0; i < info.num_encaps; i++) {
 447 +               route->tuple[!dir].in.encap[i].id = info.encap[i].id;
 448 +               route->tuple[!dir].in.encap[i].proto = info.encap[i].proto;
 449 +       }
 450 +       route->tuple[!dir].in.num_encaps = info.num_encaps;
 451 +       route->tuple[!dir].in.ingress_vlans = info.ingress_vlans;
 452 +
 453 +       if (info.xmit_type == FLOW_OFFLOAD_XMIT_DIRECT) {
 454 +               memcpy(route->tuple[dir].out.h_source, info.h_source, ETH_ALEN);
 455 +               memcpy(route->tuple[dir].out.h_dest, info.h_dest, ETH_ALEN);
 456 +               route->tuple[dir].out.ifindex = info.outdev->ifindex;
 457 +               route->tuple[dir].out.hw_ifindex = info.hw_outdev->ifindex;
 458 +               route->tuple[dir].xmit_type = info.xmit_type;
 459 +       }
 460 +}
 461 +
 462 +static int
 463 +xt_flowoffload_route(struct sk_buff *skb, const struct nf_conn *ct,
 464 +                    const struct xt_action_param *par,
 465 +                    struct nf_flow_route *route, enum ip_conntrack_dir dir,
 466 +                    struct net_device **devs)
 467 +{
 468 +       struct dst_entry *this_dst = skb_dst(skb);
 469 +       struct dst_entry *other_dst = NULL;
 470 +       struct flowi fl;
 471 +
 472 +       memset(&fl, 0, sizeof(fl));
 473 +       switch (xt_family(par)) {
 474 +       case NFPROTO_IPV4:
 475 +               fl.u.ip4.daddr = ct->tuplehash[dir].tuple.src.u3.ip;
 476 +               fl.u.ip4.flowi4_oif = xt_in(par)->ifindex;
 477 +               break;
 478 +       case NFPROTO_IPV6:
 479 +               fl.u.ip6.saddr = ct->tuplehash[!dir].tuple.dst.u3.in6;
 480 +               fl.u.ip6.daddr = ct->tuplehash[dir].tuple.src.u3.in6;
 481 +               fl.u.ip6.flowi6_oif = xt_in(par)->ifindex;
 482 +               break;
 483 +       }
 484 +
 485 +       nf_route(xt_net(par), &other_dst, &fl, false, xt_family(par));
 486 +       if (!other_dst)
 487 +               return -ENOENT;
 488 +
 489 +       nf_default_forward_path(route, this_dst, dir, devs);
 490 +       nf_default_forward_path(route, other_dst, !dir, devs);
 491 +
 492 +       if (route->tuple[dir].xmit_type == FLOW_OFFLOAD_XMIT_NEIGH &&
 493 +           route->tuple[!dir].xmit_type == FLOW_OFFLOAD_XMIT_NEIGH) {
 494 +               nf_dev_forward_path(route, ct, dir, devs);
 495 +               nf_dev_forward_path(route, ct, !dir, devs);
 496 +       }
 497 +
 498 +       return 0;
 499 +}
 500 +
 501 +static unsigned int
 502 +flowoffload_tg(struct sk_buff *skb, const struct xt_action_param *par)
 503 +{
 504 +       struct xt_flowoffload_table *table;
 505 +       const struct xt_flowoffload_target_info *info = par->targinfo;
 506 +       struct tcphdr _tcph, *tcph = NULL;
 507 +       enum ip_conntrack_info ctinfo;
 508 +       enum ip_conntrack_dir dir;
 509 +       struct nf_flow_route route = {};
 510 +       struct flow_offload *flow = NULL;
 511 +       struct net_device *devs[2] = {};
 512 +       struct nf_conn *ct;
 513 +       struct net *net;
 514 +
 515 +       if (xt_flowoffload_skip(skb, xt_family(par)))
 516 +               return XT_CONTINUE;
 517 +
 518 +       ct = nf_ct_get(skb, &ctinfo);
 519 +       if (ct == NULL)
 520 +               return XT_CONTINUE;
 521 +
 522 +       switch (ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.protonum) {
 523 +       case IPPROTO_TCP:
 524 +               if (ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED)
 525 +                       return XT_CONTINUE;
 526 +
 527 +               tcph = skb_header_pointer(skb, par->thoff,
 528 +                                         sizeof(_tcph), &_tcph);
 529 +               if (unlikely(!tcph || tcph->fin || tcph->rst))
 530 +                       return XT_CONTINUE;
 531 +               break;
 532 +       case IPPROTO_UDP:
 533 +               break;
 534 +       default:
 535 +               return XT_CONTINUE;
 536 +       }
 537 +
 538 +       if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
 539 +           ct->status & (IPS_SEQ_ADJUST | IPS_NAT_CLASH))
 540 +               return XT_CONTINUE;
 541 +
 542 +       if (!nf_ct_is_confirmed(ct))
 543 +               return XT_CONTINUE;
 544 +
 545 +       devs[dir] = xt_out(par);
 546 +       devs[!dir] = xt_in(par);
 547 +
 548 +       if (!devs[dir] || !devs[!dir])
 549 +               return XT_CONTINUE;
 550 +
 551 +       if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
 552 +               return XT_CONTINUE;
 553 +
 554 +       dir = CTINFO2DIR(ctinfo);
 555 +
 556 +       if (xt_flowoffload_route(skb, ct, par, &route, dir, devs) < 0)
 557 +               goto err_flow_route;
 558 +
 559 +       flow = flow_offload_alloc(ct);
 560 +       if (!flow)
 561 +               goto err_flow_alloc;
 562 +
 563 +       if (flow_offload_route_init(flow, &route) < 0)
 564 +               goto err_flow_add;
 565 +
 566 +       if (tcph) {
 567 +               ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
 568 +               ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
 569 +       }
 570 +
 571 +       table = &flowtable[!!(info->flags & XT_FLOWOFFLOAD_HW)];
 572 +
 573 +       net = read_pnet(&table->ft.net);
 574 +       if (!net)
 575 +               write_pnet(&table->ft.net, xt_net(par));
 576 +
 577 +       if (flow_offload_add(&table->ft, flow) < 0)
 578 +               goto err_flow_add;
 579 +
 580 +       xt_flowoffload_check_device(table, devs[0]);
 581 +       xt_flowoffload_check_device(table, devs[1]);
 582 +
 583 +       dst_release(route.tuple[!dir].dst);
 584 +
 585 +       return XT_CONTINUE;
 586 +
 587 +err_flow_add:
 588 +       flow_offload_free(flow);
 589 +err_flow_alloc:
 590 +       dst_release(route.tuple[!dir].dst);
 591 +err_flow_route:
 592 +       clear_bit(IPS_OFFLOAD_BIT, &ct->status);
 593 +
 594 +       return XT_CONTINUE;
 595 +}
 596 +
 597 +static int flowoffload_chk(const struct xt_tgchk_param *par)
 598 +{
 599 +       struct xt_flowoffload_target_info *info = par->targinfo;
 600 +
 601 +       if (info->flags & ~XT_FLOWOFFLOAD_MASK)
 602 +               return -EINVAL;
 603 +
 604 +       return 0;
 605 +}
 606 +
 607 +static struct xt_target offload_tg_reg __read_mostly = {
 608 +       .family         = NFPROTO_UNSPEC,
 609 +       .name           = "FLOWOFFLOAD",
 610 +       .revision       = 0,
 611 +       .targetsize     = sizeof(struct xt_flowoffload_target_info),
 612 +       .usersize       = sizeof(struct xt_flowoffload_target_info),
 613 +       .checkentry     = flowoffload_chk,
 614 +       .target         = flowoffload_tg,
 615 +       .me             = THIS_MODULE,
 616 +};
 617 +
 618 +static int flow_offload_netdev_event(struct notifier_block *this,
 619 +                                    unsigned long event, void *ptr)
 620 +{
 621 +       struct xt_flowoffload_hook *hook0, *hook1;
 622 +       struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 623 +
 624 +       if (event != NETDEV_UNREGISTER)
 625 +               return NOTIFY_DONE;
 626 +
 627 +       spin_lock_bh(&hooks_lock);
 628 +       hook0 = flow_offload_lookup_hook(&flowtable[0], dev);
 629 +       if (hook0)
 630 +               hlist_del(&hook0->list);
 631 +
 632 +       hook1 = flow_offload_lookup_hook(&flowtable[1], dev);
 633 +       if (hook1)
 634 +               hlist_del(&hook1->list);
 635 +       spin_unlock_bh(&hooks_lock);
 636 +
 637 +       if (hook0) {
 638 +               nf_unregister_net_hook(hook0->net, &hook0->ops);
 639 +               kfree(hook0);
 640 +       }
 641 +
 642 +       if (hook1) {
 643 +               nf_unregister_net_hook(hook1->net, &hook1->ops);
 644 +               kfree(hook1);
 645 +       }
 646 +
 647 +       nf_flow_table_cleanup(dev);
 648 +
 649 +       return NOTIFY_DONE;
 650 +}
 651 +
 652 +static struct notifier_block flow_offload_netdev_notifier = {
 653 +       .notifier_call  = flow_offload_netdev_event,
 654 +};
 655 +
 656 +static int nf_flow_rule_route_inet(struct net *net,
 657 +                                  const struct flow_offload *flow,
 658 +                                  enum flow_offload_tuple_dir dir,
 659 +                                  struct nf_flow_rule *flow_rule)
 660 +{
 661 +       const struct flow_offload_tuple *flow_tuple = &flow->tuplehash[dir].tuple;
 662 +       int err;
 663 +
 664 +       switch (flow_tuple->l3proto) {
 665 +       case NFPROTO_IPV4:
 666 +               err = nf_flow_rule_route_ipv4(net, flow, dir, flow_rule);
 667 +               break;
 668 +       case NFPROTO_IPV6:
 669 +               err = nf_flow_rule_route_ipv6(net, flow, dir, flow_rule);
 670 +               break;
 671 +       default:
 672 +               err = -1;
 673 +               break;
 674 +       }
 675 +
 676 +       return err;
 677 +}
 678 +
 679 +static struct nf_flowtable_type flowtable_inet = {
 680 +       .family         = NFPROTO_INET,
 681 +       .init           = nf_flow_table_init,
 682 +       .setup          = nf_flow_table_offload_setup,
 683 +       .action         = nf_flow_rule_route_inet,
 684 +       .free           = nf_flow_table_free,
 685 +       .hook           = xt_flowoffload_net_hook,
 686 +       .owner          = THIS_MODULE,
 687 +};
 688 +
 689 +static int init_flowtable(struct xt_flowoffload_table *tbl)
 690 +{
 691 +       INIT_DELAYED_WORK(&tbl->work, xt_flowoffload_hook_work);
 692 +       tbl->ft.type = &flowtable_inet;
 693 +       tbl->ft.flags = NF_FLOWTABLE_COUNTER;
 694 +
 695 +       return nf_flow_table_init(&tbl->ft);
 696 +}
 697 +
 698 +static int __init xt_flowoffload_tg_init(void)
 699 +{
 700 +       int ret;
 701 +
 702 +       register_netdevice_notifier(&flow_offload_netdev_notifier);
 703 +
 704 +       ret = init_flowtable(&flowtable[0]);
 705 +       if (ret)
 706 +               return ret;
 707 +
 708 +       ret = init_flowtable(&flowtable[1]);
 709 +       if (ret)
 710 +               goto cleanup;
 711 +
 712 +       flowtable[1].ft.flags |= NF_FLOWTABLE_HW_OFFLOAD;
 713 +
 714 +       ret = xt_register_target(&offload_tg_reg);
 715 +       if (ret)
 716 +               goto cleanup2;
 717 +
 718 +       return 0;
 719 +
 720 +cleanup2:
 721 +       nf_flow_table_free(&flowtable[1].ft);
 722 +cleanup:
 723 +       nf_flow_table_free(&flowtable[0].ft);
 724 +       return ret;
 725 +}
 726 +
 727 +static void __exit xt_flowoffload_tg_exit(void)
 728 +{
 729 +       xt_unregister_target(&offload_tg_reg);
 730 +       unregister_netdevice_notifier(&flow_offload_netdev_notifier);
 731 +       nf_flow_table_free(&flowtable[0].ft);
 732 +       nf_flow_table_free(&flowtable[1].ft);
 733 +}
 734 +
 735 +MODULE_LICENSE("GPL");
 736 +module_init(xt_flowoffload_tg_init);
 737 +module_exit(xt_flowoffload_tg_exit);
 738 --- a/net/netfilter/nf_flow_table_core.c
 739 +++ b/net/netfilter/nf_flow_table_core.c
 740 @@ -7,7 +7,6 @@
 741  #include <linux/netdevice.h>
 742  #include <net/ip.h>
 743  #include <net/ip6_route.h>
 744 -#include <net/netfilter/nf_tables.h>
 745  #include <net/netfilter/nf_flow_table.h>
 746  #include <net/netfilter/nf_conntrack.h>
 747  #include <net/netfilter/nf_conntrack_core.h>
 748 @@ -381,8 +380,7 @@ flow_offload_lookup(struct nf_flowtable
 749  }
 750  EXPORT_SYMBOL_GPL(flow_offload_lookup);
 751
 752 -static int
 753 -nf_flow_table_iterate(struct nf_flowtable *flow_table,
 754 +int nf_flow_table_iterate(struct nf_flowtable *flow_table,
 755                       void (*iter)(struct nf_flowtable *flowtable,
 756                                    struct flow_offload *flow, void *data),
 757                       void *data)
 758 @@ -436,6 +434,7 @@ static void nf_flow_offload_gc_step(stru
 759                 nf_flow_offload_stats(flow_table, flow);
 760         }
 761  }
 762 +EXPORT_SYMBOL_GPL(nf_flow_table_iterate);
 763
 764  void nf_flow_table_gc_run(struct nf_flowtable *flow_table)
 765  {
 766 --- /dev/null
 767 +++ b/include/uapi/linux/netfilter/xt_FLOWOFFLOAD.h
 768 @@ -0,0 +1,17 @@
 769 +/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
 770 +#ifndef _XT_FLOWOFFLOAD_H
 771 +#define _XT_FLOWOFFLOAD_H
 772 +
 773 +#include <linux/types.h>
 774 +
 775 +enum {
 776 +       XT_FLOWOFFLOAD_HW       = 1 << 0,
 777 +
 778 +       XT_FLOWOFFLOAD_MASK     = XT_FLOWOFFLOAD_HW
 779 +};
 780 +
 781 +struct xt_flowoffload_target_info {
 782 +       __u32 flags;
 783 +};
 784 +
 785 +#endif /* _XT_FLOWOFFLOAD_H */
 786 --- a/include/net/netfilter/nf_flow_table.h
 787 +++ b/include/net/netfilter/nf_flow_table.h
 788 @@ -280,6 +280,11 @@ void nf_flow_table_free(struct nf_flowta
 789
 790  void flow_offload_teardown(struct flow_offload *flow);
 791
 792 +int nf_flow_table_iterate(struct nf_flowtable *flow_table,
 793 +                      void (*iter)(struct nf_flowtable *flowtable,
 794 +                                   struct flow_offload *flow, void *data),
 795 +                      void *data);
 796 +
 797  void nf_flow_snat_port(const struct flow_offload *flow,
 798                        struct sk_buff *skb, unsigned int thoff,
 799                        u8 protocol, enum flow_offload_tuple_dir dir);