target/linux/generic/hack-4.14/650-netfilter-add-xt_OFFLOAD-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_OFFLOAD 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 @@ -75,8 +75,6 @@ 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 @@ -85,6 +83,8 @@ 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 @@ -69,7 +69,6 @@ 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 @@ -79,6 +78,8 @@ 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 @@ -665,8 +665,6 @@ config NFT_FIB_NETDEV
  51
  52  endif # NF_TABLES_NETDEV
  53
  54 -endif # NF_TABLES
  55 -
  56  config NF_FLOW_TABLE_INET
  57         tristate "Netfilter flow table mixed IPv4/IPv6 module"
  58         depends on NF_FLOW_TABLE
  59 @@ -675,11 +673,12 @@ 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 @@ -968,6 +967,15 @@ 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 @@ -134,6 +134,7 @@ 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 +1,365 @@
 102 +/*
 103 + * Copyright (C) 2018 Felix Fietkau <nbd@nbd.name>
 104 + *
 105 + * This program is free software; you can redistribute it and/or modify
 106 + * it under the terms of the GNU General Public License version 2 as
 107 + * published by the Free Software Foundation.
 108 + */
 109 +#include <linux/module.h>
 110 +#include <linux/init.h>
 111 +#include <linux/netfilter.h>
 112 +#include <linux/netfilter/xt_FLOWOFFLOAD.h>
 113 +#include <net/ip.h>
 114 +#include <net/netfilter/nf_conntrack.h>
 115 +#include <net/netfilter/nf_flow_table.h>
 116 +
 117 +static struct nf_flowtable nf_flowtable;
 118 +static HLIST_HEAD(hooks);
 119 +static DEFINE_SPINLOCK(hooks_lock);
 120 +static struct delayed_work hook_work;
 121 +
 122 +struct xt_flowoffload_hook {
 123 +       struct hlist_node list;
 124 +       struct nf_hook_ops ops;
 125 +       struct net *net;
 126 +       bool registered;
 127 +       bool used;
 128 +};
 129 +
 130 +static unsigned int
 131 +xt_flowoffload_net_hook(void *priv, struct sk_buff *skb,
 132 +                         const struct nf_hook_state *state)
 133 +{
 134 +       switch (skb->protocol) {
 135 +       case htons(ETH_P_IP):
 136 +               return nf_flow_offload_ip_hook(priv, skb, state);
 137 +       case htons(ETH_P_IPV6):
 138 +               return nf_flow_offload_ipv6_hook(priv, skb, state);
 139 +       }
 140 +
 141 +       return NF_ACCEPT;
 142 +}
 143 +
 144 +static int
 145 +xt_flowoffload_create_hook(struct net_device *dev)
 146 +{
 147 +       struct xt_flowoffload_hook *hook;
 148 +       struct nf_hook_ops *ops;
 149 +
 150 +       hook = kzalloc(sizeof(*hook), GFP_ATOMIC);
 151 +       if (!hook)
 152 +               return -ENOMEM;
 153 +
 154 +       ops = &hook->ops;
 155 +       ops->pf = NFPROTO_NETDEV;
 156 +       ops->hooknum = NF_NETDEV_INGRESS;
 157 +       ops->priority = 10;
 158 +       ops->priv = &nf_flowtable;
 159 +       ops->hook = xt_flowoffload_net_hook;
 160 +       ops->dev = dev;
 161 +
 162 +       hlist_add_head(&hook->list, &hooks);
 163 +       mod_delayed_work(system_power_efficient_wq, &hook_work, 0);
 164 +
 165 +       return 0;
 166 +}
 167 +
 168 +static struct xt_flowoffload_hook *
 169 +flow_offload_lookup_hook(struct net_device *dev)
 170 +{
 171 +       struct xt_flowoffload_hook *hook;
 172 +
 173 +       hlist_for_each_entry(hook, &hooks, list) {
 174 +               if (hook->ops.dev == dev)
 175 +                       return hook;
 176 +       }
 177 +
 178 +       return NULL;
 179 +}
 180 +
 181 +static void
 182 +xt_flowoffload_check_device(struct net_device *dev)
 183 +{
 184 +       struct xt_flowoffload_hook *hook;
 185 +
 186 +       spin_lock_bh(&hooks_lock);
 187 +       hook = flow_offload_lookup_hook(dev);
 188 +       if (hook)
 189 +               hook->used = true;
 190 +       else
 191 +               xt_flowoffload_create_hook(dev);
 192 +       spin_unlock_bh(&hooks_lock);
 193 +}
 194 +
 195 +static void
 196 +xt_flowoffload_register_hooks(void)
 197 +{
 198 +       struct xt_flowoffload_hook *hook;
 199 +
 200 +restart:
 201 +       hlist_for_each_entry(hook, &hooks, list) {
 202 +               if (hook->registered)
 203 +                       continue;
 204 +
 205 +               hook->registered = true;
 206 +               hook->net = dev_net(hook->ops.dev);
 207 +               spin_unlock_bh(&hooks_lock);
 208 +               nf_register_net_hook(hook->net, &hook->ops);
 209 +               spin_lock_bh(&hooks_lock);
 210 +               goto restart;
 211 +       }
 212 +
 213 +}
 214 +
 215 +static void
 216 +xt_flowoffload_cleanup_hooks(void)
 217 +{
 218 +       struct xt_flowoffload_hook *hook;
 219 +
 220 +restart:
 221 +       hlist_for_each_entry(hook, &hooks, list) {
 222 +               if (hook->used || !hook->registered)
 223 +                       continue;
 224 +
 225 +               hlist_del(&hook->list);
 226 +               spin_unlock_bh(&hooks_lock);
 227 +               nf_unregister_net_hook(hook->net, &hook->ops);
 228 +               kfree(hook);
 229 +               spin_lock_bh(&hooks_lock);
 230 +               goto restart;
 231 +       }
 232 +
 233 +}
 234 +
 235 +static void
 236 +xt_flowoffload_check_hook(struct flow_offload *flow, void *data)
 237 +{
 238 +       struct flow_offload_tuple *tuple = &flow->tuplehash[0].tuple;
 239 +       struct xt_flowoffload_hook *hook;
 240 +       bool *found = data;
 241 +
 242 +       spin_lock_bh(&hooks_lock);
 243 +       hlist_for_each_entry(hook, &hooks, list) {
 244 +               if (hook->ops.dev->ifindex != tuple->iifidx &&
 245 +                   hook->ops.dev->ifindex != tuple->oifidx)
 246 +                       continue;
 247 +
 248 +               hook->used = true;
 249 +               *found = true;
 250 +       }
 251 +       spin_unlock_bh(&hooks_lock);
 252 +}
 253 +
 254 +static void
 255 +xt_flowoffload_hook_work(struct work_struct *work)
 256 +{
 257 +       struct xt_flowoffload_hook *hook;
 258 +       bool found = false;
 259 +       int err;
 260 +
 261 +       spin_lock_bh(&hooks_lock);
 262 +       xt_flowoffload_register_hooks();
 263 +       hlist_for_each_entry(hook, &hooks, list)
 264 +               hook->used = false;
 265 +       spin_unlock_bh(&hooks_lock);
 266 +
 267 +       err = nf_flow_table_iterate(&nf_flowtable, xt_flowoffload_check_hook,
 268 +                                   &found);
 269 +       if (err && err != -EAGAIN)
 270 +           goto out;
 271 +
 272 +       spin_lock_bh(&hooks_lock);
 273 +       xt_flowoffload_cleanup_hooks();
 274 +       spin_unlock_bh(&hooks_lock);
 275 +
 276 +out:
 277 +       if (found)
 278 +               queue_delayed_work(system_power_efficient_wq, &hook_work, HZ);
 279 +}
 280 +
 281 +static bool
 282 +xt_flowoffload_skip(struct sk_buff *skb)
 283 +{
 284 +       struct ip_options *opt = &(IPCB(skb)->opt);
 285 +
 286 +       if (unlikely(opt->optlen))
 287 +               return true;
 288 +       if (skb_sec_path(skb))
 289 +               return true;
 290 +
 291 +       return false;
 292 +}
 293 +
 294 +static struct dst_entry *
 295 +xt_flowoffload_dst(const struct nf_conn *ct, enum ip_conntrack_dir dir,
 296 +                  const struct xt_action_param *par)
 297 +{
 298 +       struct dst_entry *dst = NULL;
 299 +       struct flowi fl;
 300 +
 301 +       memset(&fl, 0, sizeof(fl));
 302 +       switch (xt_family(par)) {
 303 +       case NFPROTO_IPV4:
 304 +               fl.u.ip4.daddr = ct->tuplehash[dir].tuple.src.u3.ip;
 305 +               break;
 306 +       case NFPROTO_IPV6:
 307 +               fl.u.ip6.saddr = ct->tuplehash[dir].tuple.dst.u3.in6;
 308 +               fl.u.ip6.daddr = ct->tuplehash[dir].tuple.src.u3.in6;
 309 +               break;
 310 +       }
 311 +
 312 +       nf_route(xt_net(par), &dst, &fl, false, xt_family(par));
 313 +
 314 +       return dst;
 315 +}
 316 +
 317 +static int
 318 +xt_flowoffload_route(struct sk_buff *skb, const struct nf_conn *ct,
 319 +                  const struct xt_action_param *par,
 320 +                  struct nf_flow_route *route, enum ip_conntrack_dir dir)
 321 +{
 322 +       struct dst_entry *this_dst, *other_dst;
 323 +
 324 +       this_dst = xt_flowoffload_dst(ct, dir, par);
 325 +       other_dst = xt_flowoffload_dst(ct, !dir, par);
 326 +       if (!this_dst || !other_dst)
 327 +               return -ENOENT;
 328 +
 329 +       route->tuple[dir].dst           = this_dst;
 330 +       route->tuple[dir].ifindex       = xt_in(par)->ifindex;
 331 +       route->tuple[!dir].dst          = other_dst;
 332 +       route->tuple[!dir].ifindex      = xt_out(par)->ifindex;
 333 +
 334 +       return 0;
 335 +}
 336 +
 337 +static unsigned int
 338 +flowoffload_tg(struct sk_buff *skb, const struct xt_action_param *par)
 339 +{
 340 +       const struct xt_flowoffload_target_info *info = par->targinfo;
 341 +       enum ip_conntrack_info ctinfo;
 342 +       enum ip_conntrack_dir dir;
 343 +       struct nf_flow_route route;
 344 +       struct flow_offload *flow;
 345 +       struct nf_conn *ct;
 346 +
 347 +       if (xt_flowoffload_skip(skb))
 348 +               return XT_CONTINUE;
 349 +
 350 +       ct = nf_ct_get(skb, &ctinfo);
 351 +       if (ct == NULL)
 352 +               return XT_CONTINUE;
 353 +
 354 +       switch (ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.protonum) {
 355 +       case IPPROTO_TCP:
 356 +               if (ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED)
 357 +                       return XT_CONTINUE;
 358 +               break;
 359 +       case IPPROTO_UDP:
 360 +               break;
 361 +       default:
 362 +               return XT_CONTINUE;
 363 +       }
 364 +
 365 +       if (test_bit(IPS_HELPER_BIT, &ct->status))
 366 +               return XT_CONTINUE;
 367 +
 368 +       if (ctinfo == IP_CT_NEW ||
 369 +           ctinfo == IP_CT_RELATED)
 370 +               return XT_CONTINUE;
 371 +
 372 +       if (!xt_in(par) || !xt_out(par))
 373 +               return XT_CONTINUE;
 374 +
 375 +       if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
 376 +               return XT_CONTINUE;
 377 +
 378 +       dir = CTINFO2DIR(ctinfo);
 379 +
 380 +       if (xt_flowoffload_route(skb, ct, par, &route, dir) < 0)
 381 +               goto err_flow_route;
 382 +
 383 +       flow = flow_offload_alloc(ct, &route);
 384 +       if (!flow)
 385 +               goto err_flow_alloc;
 386 +
 387 +       if (flow_offload_add(&nf_flowtable, flow) < 0)
 388 +               goto err_flow_add;
 389 +
 390 +       xt_flowoffload_check_device(xt_in(par));
 391 +       xt_flowoffload_check_device(xt_out(par));
 392 +
 393 +       if (info->flags & XT_FLOWOFFLOAD_HW)
 394 +               nf_flow_offload_hw_add(xt_net(par), flow, ct);
 395 +
 396 +       return XT_CONTINUE;
 397 +
 398 +err_flow_add:
 399 +       flow_offload_free(flow);
 400 +err_flow_alloc:
 401 +       dst_release(route.tuple[!dir].dst);
 402 +err_flow_route:
 403 +       clear_bit(IPS_OFFLOAD_BIT, &ct->status);
 404 +       return XT_CONTINUE;
 405 +}
 406 +
 407 +
 408 +static int flowoffload_chk(const struct xt_tgchk_param *par)
 409 +{
 410 +       struct xt_flowoffload_target_info *info = par->targinfo;
 411 +
 412 +       if (info->flags & ~XT_FLOWOFFLOAD_MASK)
 413 +               return -EINVAL;
 414 +
 415 +       return 0;
 416 +}
 417 +
 418 +static struct xt_target offload_tg_reg __read_mostly = {
 419 +       .family         = NFPROTO_UNSPEC,
 420 +       .name           = "FLOWOFFLOAD",
 421 +       .revision       = 0,
 422 +       .targetsize     = sizeof(struct xt_flowoffload_target_info),
 423 +       .usersize       = sizeof(struct xt_flowoffload_target_info),
 424 +       .checkentry     = flowoffload_chk,
 425 +       .target         = flowoffload_tg,
 426 +       .me             = THIS_MODULE,
 427 +};
 428 +
 429 +static int xt_flowoffload_table_init(struct nf_flowtable *table)
 430 +{
 431 +       table->flags = NF_FLOWTABLE_F_HW;
 432 +       nf_flow_table_init(table);
 433 +       return 0;
 434 +}
 435 +
 436 +static void xt_flowoffload_table_cleanup(struct nf_flowtable *table)
 437 +{
 438 +       nf_flow_table_free(table);
 439 +}
 440 +
 441 +static int __init xt_flowoffload_tg_init(void)
 442 +{
 443 +       int ret;
 444 +
 445 +       INIT_DELAYED_WORK(&hook_work, xt_flowoffload_hook_work);
 446 +
 447 +       ret = xt_flowoffload_table_init(&nf_flowtable);
 448 +       if (ret)
 449 +               return ret;
 450 +
 451 +       ret = xt_register_target(&offload_tg_reg);
 452 +       if (ret)
 453 +               xt_flowoffload_table_cleanup(&nf_flowtable);
 454 +
 455 +       return ret;
 456 +}
 457 +
 458 +static void __exit xt_flowoffload_tg_exit(void)
 459 +{
 460 +       xt_unregister_target(&offload_tg_reg);
 461 +       xt_flowoffload_table_cleanup(&nf_flowtable);
 462 +}
 463 +
 464 +MODULE_LICENSE("GPL");
 465 +module_init(xt_flowoffload_tg_init);
 466 +module_exit(xt_flowoffload_tg_exit);
 467 --- a/net/netfilter/nf_flow_table_core.c
 468 +++ b/net/netfilter/nf_flow_table_core.c
 469 @@ -6,7 +6,6 @@
 470  #include <linux/netdevice.h>
 471  #include <net/ip.h>
 472  #include <net/ip6_route.h>
 473 -#include <net/netfilter/nf_tables.h>
 474  #include <net/netfilter/nf_flow_table.h>
 475  #include <net/netfilter/nf_conntrack.h>
 476  #include <net/netfilter/nf_conntrack_core.h>
 477 --- /dev/null
 478 +++ b/include/uapi/linux/netfilter/xt_FLOWOFFLOAD.h
 479 @@ -0,0 +1,17 @@
 480 +/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
 481 +#ifndef _XT_FLOWOFFLOAD_H
 482 +#define _XT_FLOWOFFLOAD_H
 483 +
 484 +#include <linux/types.h>
 485 +
 486 +enum {
 487 +       XT_FLOWOFFLOAD_HW       = 1 << 0,
 488 +
 489 +       XT_FLOWOFFLOAD_MASK     = XT_FLOWOFFLOAD_HW
 490 +};
 491 +
 492 +struct xt_flowoffload_target_info {
 493 +       __u32 flags;
 494 +};
 495 +
 496 +#endif /* _XT_FLOWOFFLOAD_H */