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 @@ -76,8 +76,6 @@ 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 @@ -86,6 +84,8 @@ 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 @@ -97,7 +97,6 @@ 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 @@ -107,6 +106,8 @@ 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 @@ -671,8 +671,6 @@ 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 @@ -681,11 +679,12 @@ 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 @@ -974,6 +973,15 @@ 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 +1,408 @@
 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 +#include <net/netfilter/nf_conntrack_helper.h>
 117 +
 118 +static struct nf_flowtable nf_flowtable;
 119 +static HLIST_HEAD(hooks);
 120 +static DEFINE_SPINLOCK(hooks_lock);
 121 +static struct delayed_work hook_work;
 122 +
 123 +struct xt_flowoffload_hook {
 124 +       struct hlist_node list;
 125 +       struct nf_hook_ops ops;
 126 +       struct net *net;
 127 +       bool registered;
 128 +       bool used;
 129 +};
 130 +
 131 +static unsigned int
 132 +xt_flowoffload_net_hook(void *priv, struct sk_buff *skb,
 133 +                         const struct nf_hook_state *state)
 134 +{
 135 +       switch (skb->protocol) {
 136 +       case htons(ETH_P_IP):
 137 +               return nf_flow_offload_ip_hook(priv, skb, state);
 138 +       case htons(ETH_P_IPV6):
 139 +               return nf_flow_offload_ipv6_hook(priv, skb, state);
 140 +       }
 141 +
 142 +       return NF_ACCEPT;
 143 +}
 144 +
 145 +static int
 146 +xt_flowoffload_create_hook(struct net_device *dev)
 147 +{
 148 +       struct xt_flowoffload_hook *hook;
 149 +       struct nf_hook_ops *ops;
 150 +
 151 +       hook = kzalloc(sizeof(*hook), GFP_ATOMIC);
 152 +       if (!hook)
 153 +               return -ENOMEM;
 154 +
 155 +       ops = &hook->ops;
 156 +       ops->pf = NFPROTO_NETDEV;
 157 +       ops->hooknum = NF_NETDEV_INGRESS;
 158 +       ops->priority = 10;
 159 +       ops->priv = &nf_flowtable;
 160 +       ops->hook = xt_flowoffload_net_hook;
 161 +       ops->dev = dev;
 162 +
 163 +       hlist_add_head(&hook->list, &hooks);
 164 +       mod_delayed_work(system_power_efficient_wq, &hook_work, 0);
 165 +
 166 +       return 0;
 167 +}
 168 +
 169 +static struct xt_flowoffload_hook *
 170 +flow_offload_lookup_hook(struct net_device *dev)
 171 +{
 172 +       struct xt_flowoffload_hook *hook;
 173 +
 174 +       hlist_for_each_entry(hook, &hooks, list) {
 175 +               if (hook->ops.dev == dev)
 176 +                       return hook;
 177 +       }
 178 +
 179 +       return NULL;
 180 +}
 181 +
 182 +static void
 183 +xt_flowoffload_check_device(struct net_device *dev)
 184 +{
 185 +       struct xt_flowoffload_hook *hook;
 186 +
 187 +       spin_lock_bh(&hooks_lock);
 188 +       hook = flow_offload_lookup_hook(dev);
 189 +       if (hook)
 190 +               hook->used = true;
 191 +       else
 192 +               xt_flowoffload_create_hook(dev);
 193 +       spin_unlock_bh(&hooks_lock);
 194 +}
 195 +
 196 +static void
 197 +xt_flowoffload_register_hooks(void)
 198 +{
 199 +       struct xt_flowoffload_hook *hook;
 200 +
 201 +restart:
 202 +       hlist_for_each_entry(hook, &hooks, list) {
 203 +               if (hook->registered)
 204 +                       continue;
 205 +
 206 +               hook->registered = true;
 207 +               hook->net = dev_net(hook->ops.dev);
 208 +               spin_unlock_bh(&hooks_lock);
 209 +               nf_register_net_hook(hook->net, &hook->ops);
 210 +               spin_lock_bh(&hooks_lock);
 211 +               goto restart;
 212 +       }
 213 +
 214 +}
 215 +
 216 +static void
 217 +xt_flowoffload_cleanup_hooks(void)
 218 +{
 219 +       struct xt_flowoffload_hook *hook;
 220 +
 221 +restart:
 222 +       hlist_for_each_entry(hook, &hooks, list) {
 223 +               if (hook->used || !hook->registered)
 224 +                       continue;
 225 +
 226 +               hlist_del(&hook->list);
 227 +               spin_unlock_bh(&hooks_lock);
 228 +               nf_unregister_net_hook(hook->net, &hook->ops);
 229 +               kfree(hook);
 230 +               spin_lock_bh(&hooks_lock);
 231 +               goto restart;
 232 +       }
 233 +
 234 +}
 235 +
 236 +static void
 237 +xt_flowoffload_check_hook(struct flow_offload *flow, void *data)
 238 +{
 239 +       struct flow_offload_tuple *tuple = &flow->tuplehash[0].tuple;
 240 +       struct xt_flowoffload_hook *hook;
 241 +       bool *found = data;
 242 +
 243 +       spin_lock_bh(&hooks_lock);
 244 +       hlist_for_each_entry(hook, &hooks, list) {
 245 +               if (hook->ops.dev->ifindex != tuple->iifidx &&
 246 +                   hook->ops.dev->ifindex != tuple->oifidx)
 247 +                       continue;
 248 +
 249 +               hook->used = true;
 250 +               *found = true;
 251 +       }
 252 +       spin_unlock_bh(&hooks_lock);
 253 +}
 254 +
 255 +static void
 256 +xt_flowoffload_hook_work(struct work_struct *work)
 257 +{
 258 +       struct xt_flowoffload_hook *hook;
 259 +       bool found = false;
 260 +       int err;
 261 +
 262 +       spin_lock_bh(&hooks_lock);
 263 +       xt_flowoffload_register_hooks();
 264 +       hlist_for_each_entry(hook, &hooks, list)
 265 +               hook->used = false;
 266 +       spin_unlock_bh(&hooks_lock);
 267 +
 268 +       err = nf_flow_table_iterate(&nf_flowtable, xt_flowoffload_check_hook,
 269 +                                   &found);
 270 +       if (err && err != -EAGAIN)
 271 +           goto out;
 272 +
 273 +       spin_lock_bh(&hooks_lock);
 274 +       xt_flowoffload_cleanup_hooks();
 275 +       spin_unlock_bh(&hooks_lock);
 276 +
 277 +out:
 278 +       if (found)
 279 +               queue_delayed_work(system_power_efficient_wq, &hook_work, HZ);
 280 +}
 281 +
 282 +static bool
 283 +xt_flowoffload_skip(struct sk_buff *skb)
 284 +{
 285 +       struct ip_options *opt = &(IPCB(skb)->opt);
 286 +
 287 +       if (unlikely(opt->optlen))
 288 +               return true;
 289 +       if (skb_sec_path(skb))
 290 +               return true;
 291 +
 292 +       return false;
 293 +}
 294 +
 295 +static struct dst_entry *
 296 +xt_flowoffload_dst(const struct nf_conn *ct, enum ip_conntrack_dir dir,
 297 +                  const struct xt_action_param *par)
 298 +{
 299 +       struct dst_entry *dst = NULL;
 300 +       struct flowi fl;
 301 +
 302 +       memset(&fl, 0, sizeof(fl));
 303 +       switch (xt_family(par)) {
 304 +       case NFPROTO_IPV4:
 305 +               fl.u.ip4.daddr = ct->tuplehash[dir].tuple.src.u3.ip;
 306 +               break;
 307 +       case NFPROTO_IPV6:
 308 +               fl.u.ip6.saddr = ct->tuplehash[dir].tuple.dst.u3.in6;
 309 +               fl.u.ip6.daddr = ct->tuplehash[dir].tuple.src.u3.in6;
 310 +               break;
 311 +       }
 312 +
 313 +       nf_route(xt_net(par), &dst, &fl, false, xt_family(par));
 314 +
 315 +       return dst;
 316 +}
 317 +
 318 +static int
 319 +xt_flowoffload_route(struct sk_buff *skb, const struct nf_conn *ct,
 320 +                  const struct xt_action_param *par,
 321 +                  struct nf_flow_route *route, enum ip_conntrack_dir dir)
 322 +{
 323 +       struct dst_entry *this_dst, *other_dst;
 324 +
 325 +       this_dst = xt_flowoffload_dst(ct, dir, par);
 326 +       other_dst = xt_flowoffload_dst(ct, !dir, par);
 327 +       if (!this_dst || !other_dst)
 328 +               return -ENOENT;
 329 +
 330 +       if (dst_xfrm(this_dst) || dst_xfrm(other_dst))
 331 +               return -EINVAL;
 332 +
 333 +       route->tuple[dir].dst           = this_dst;
 334 +       route->tuple[dir].ifindex       = xt_in(par)->ifindex;
 335 +       route->tuple[!dir].dst          = other_dst;
 336 +       route->tuple[!dir].ifindex      = xt_out(par)->ifindex;
 337 +
 338 +       return 0;
 339 +}
 340 +
 341 +static unsigned int
 342 +flowoffload_tg(struct sk_buff *skb, const struct xt_action_param *par)
 343 +{
 344 +       const struct xt_flowoffload_target_info *info = par->targinfo;
 345 +       const struct nf_conn_help *help;
 346 +       enum ip_conntrack_info ctinfo;
 347 +       enum ip_conntrack_dir dir;
 348 +       struct nf_flow_route route;
 349 +       struct flow_offload *flow;
 350 +       struct nf_conn *ct;
 351 +       struct net *net;
 352 +
 353 +       if (xt_flowoffload_skip(skb))
 354 +               return XT_CONTINUE;
 355 +
 356 +       ct = nf_ct_get(skb, &ctinfo);
 357 +       if (ct == NULL)
 358 +               return XT_CONTINUE;
 359 +
 360 +       switch (ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.protonum) {
 361 +       case IPPROTO_TCP:
 362 +               if (ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED)
 363 +                       return XT_CONTINUE;
 364 +               break;
 365 +       case IPPROTO_UDP:
 366 +               break;
 367 +       default:
 368 +               return XT_CONTINUE;
 369 +       }
 370 +
 371 +       help = nfct_help(ct);
 372 +       if (help)
 373 +               return XT_CONTINUE;
 374 +
 375 +       if (ctinfo == IP_CT_NEW ||
 376 +           ctinfo == IP_CT_RELATED)
 377 +               return XT_CONTINUE;
 378 +
 379 +       if (!xt_in(par) || !xt_out(par))
 380 +               return XT_CONTINUE;
 381 +
 382 +       if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
 383 +               return XT_CONTINUE;
 384 +
 385 +       dir = CTINFO2DIR(ctinfo);
 386 +
 387 +       if (xt_flowoffload_route(skb, ct, par, &route, dir) < 0)
 388 +               goto err_flow_route;
 389 +
 390 +       flow = flow_offload_alloc(ct, &route);
 391 +       if (!flow)
 392 +               goto err_flow_alloc;
 393 +
 394 +       if (flow_offload_add(&nf_flowtable, flow) < 0)
 395 +               goto err_flow_add;
 396 +
 397 +       xt_flowoffload_check_device(xt_in(par));
 398 +       xt_flowoffload_check_device(xt_out(par));
 399 +
 400 +       net = read_pnet(&nf_flowtable.ft_net);
 401 +       if (!net)
 402 +               write_pnet(&nf_flowtable.ft_net, xt_net(par));
 403 +
 404 +       if (info->flags & XT_FLOWOFFLOAD_HW)
 405 +               nf_flow_offload_hw_add(xt_net(par), flow, ct);
 406 +
 407 +       return XT_CONTINUE;
 408 +
 409 +err_flow_add:
 410 +       flow_offload_free(flow);
 411 +err_flow_alloc:
 412 +       dst_release(route.tuple[!dir].dst);
 413 +err_flow_route:
 414 +       clear_bit(IPS_OFFLOAD_BIT, &ct->status);
 415 +       return XT_CONTINUE;
 416 +}
 417 +
 418 +
 419 +static int flowoffload_chk(const struct xt_tgchk_param *par)
 420 +{
 421 +       struct xt_flowoffload_target_info *info = par->targinfo;
 422 +
 423 +       if (info->flags & ~XT_FLOWOFFLOAD_MASK)
 424 +               return -EINVAL;
 425 +
 426 +       return 0;
 427 +}
 428 +
 429 +static struct xt_target offload_tg_reg __read_mostly = {
 430 +       .family         = NFPROTO_UNSPEC,
 431 +       .name           = "FLOWOFFLOAD",
 432 +       .revision       = 0,
 433 +       .targetsize     = sizeof(struct xt_flowoffload_target_info),
 434 +       .usersize       = sizeof(struct xt_flowoffload_target_info),
 435 +       .checkentry     = flowoffload_chk,
 436 +       .target         = flowoffload_tg,
 437 +       .me             = THIS_MODULE,
 438 +};
 439 +
 440 +static int xt_flowoffload_table_init(struct nf_flowtable *table)
 441 +{
 442 +       table->flags = NF_FLOWTABLE_F_HW;
 443 +       nf_flow_table_init(table);
 444 +       return 0;
 445 +}
 446 +
 447 +static void xt_flowoffload_table_cleanup(struct nf_flowtable *table)
 448 +{
 449 +       nf_flow_table_free(table);
 450 +}
 451 +
 452 +static int flow_offload_netdev_event(struct notifier_block *this,
 453 +                                    unsigned long event, void *ptr)
 454 +{
 455 +       struct xt_flowoffload_hook *hook = NULL;
 456 +       struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 457 +
 458 +       if (event != NETDEV_UNREGISTER)
 459 +               return NOTIFY_DONE;
 460 +
 461 +       spin_lock_bh(&hooks_lock);
 462 +       hook = flow_offload_lookup_hook(dev);
 463 +       if (hook) {
 464 +               hlist_del(&hook->list);
 465 +       }
 466 +       spin_unlock_bh(&hooks_lock);
 467 +       if (hook) {
 468 +               nf_unregister_net_hook(hook->net, &hook->ops);
 469 +               kfree(hook);
 470 +       }
 471 +
 472 +       nf_flow_table_cleanup(dev_net(dev), dev);
 473 +
 474 +       return NOTIFY_DONE;
 475 +}
 476 +
 477 +static struct notifier_block flow_offload_netdev_notifier = {
 478 +       .notifier_call  = flow_offload_netdev_event,
 479 +};
 480 +
 481 +static int __init xt_flowoffload_tg_init(void)
 482 +{
 483 +       int ret;
 484 +
 485 +       register_netdevice_notifier(&flow_offload_netdev_notifier);
 486 +
 487 +       INIT_DELAYED_WORK(&hook_work, xt_flowoffload_hook_work);
 488 +
 489 +       ret = xt_flowoffload_table_init(&nf_flowtable);
 490 +       if (ret)
 491 +               return ret;
 492 +
 493 +       ret = xt_register_target(&offload_tg_reg);
 494 +       if (ret)
 495 +               xt_flowoffload_table_cleanup(&nf_flowtable);
 496 +
 497 +       return ret;
 498 +}
 499 +
 500 +static void __exit xt_flowoffload_tg_exit(void)
 501 +{
 502 +       xt_unregister_target(&offload_tg_reg);
 503 +       xt_flowoffload_table_cleanup(&nf_flowtable);
 504 +       unregister_netdevice_notifier(&flow_offload_netdev_notifier);
 505 +}
 506 +
 507 +MODULE_LICENSE("GPL");
 508 +module_init(xt_flowoffload_tg_init);
 509 +module_exit(xt_flowoffload_tg_exit);
 510 --- a/net/netfilter/nf_flow_table_core.c
 511 +++ b/net/netfilter/nf_flow_table_core.c
 512 @@ -6,7 +6,6 @@
 513  #include <linux/netdevice.h>
 514  #include <net/ip.h>
 515  #include <net/ip6_route.h>
 516 -#include <net/netfilter/nf_tables.h>
 517  #include <net/netfilter/nf_flow_table.h>
 518  #include <net/netfilter/nf_conntrack.h>
 519  #include <net/netfilter/nf_conntrack_core.h>
 520 --- /dev/null
 521 +++ b/include/uapi/linux/netfilter/xt_FLOWOFFLOAD.h
 522 @@ -0,0 +1,17 @@
 523 +/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
 524 +#ifndef _XT_FLOWOFFLOAD_H
 525 +#define _XT_FLOWOFFLOAD_H
 526 +
 527 +#include <linux/types.h>
 528 +
 529 +enum {
 530 +       XT_FLOWOFFLOAD_HW       = 1 << 0,
 531 +
 532 +       XT_FLOWOFFLOAD_MASK     = XT_FLOWOFFLOAD_HW
 533 +};
 534 +
 535 +struct xt_flowoffload_target_info {
 536 +       __u32 flags;
 537 +};
 538 +
 539 +#endif /* _XT_FLOWOFFLOAD_H */