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 @@ -959,6 +958,15 @@ 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 @@ -133,6 +133,7 @@ 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 +1,351 @@
 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 <net/ip.h>
 113 +#include <net/netfilter/nf_conntrack.h>
 114 +#include <net/netfilter/nf_flow_table.h>
 115 +
 116 +static struct nf_flowtable nf_flowtable;
 117 +static HLIST_HEAD(hooks);
 118 +static DEFINE_SPINLOCK(hooks_lock);
 119 +static struct delayed_work hook_work;
 120 +
 121 +struct xt_flowoffload_hook {
 122 +       struct hlist_node list;
 123 +       struct nf_hook_ops ops;
 124 +       struct net *net;
 125 +       bool registered;
 126 +       bool used;
 127 +};
 128 +
 129 +static unsigned int
 130 +xt_flowoffload_net_hook(void *priv, struct sk_buff *skb,
 131 +                         const struct nf_hook_state *state)
 132 +{
 133 +       switch (skb->protocol) {
 134 +       case htons(ETH_P_IP):
 135 +               return nf_flow_offload_ip_hook(priv, skb, state);
 136 +       case htons(ETH_P_IPV6):
 137 +               return nf_flow_offload_ipv6_hook(priv, skb, state);
 138 +       }
 139 +
 140 +       return NF_ACCEPT;
 141 +}
 142 +
 143 +static int
 144 +xt_flowoffload_create_hook(struct net_device *dev)
 145 +{
 146 +       struct xt_flowoffload_hook *hook;
 147 +       struct nf_hook_ops *ops;
 148 +
 149 +       hook = kzalloc(sizeof(*hook), GFP_ATOMIC);
 150 +       if (!hook)
 151 +               return -ENOMEM;
 152 +
 153 +       ops = &hook->ops;
 154 +       ops->pf = NFPROTO_NETDEV;
 155 +       ops->hooknum = NF_NETDEV_INGRESS;
 156 +       ops->priority = 10;
 157 +       ops->priv = &nf_flowtable;
 158 +       ops->hook = xt_flowoffload_net_hook;
 159 +       ops->dev = dev;
 160 +
 161 +       hlist_add_head(&hook->list, &hooks);
 162 +       mod_delayed_work(system_power_efficient_wq, &hook_work, 0);
 163 +
 164 +       return 0;
 165 +}
 166 +
 167 +static struct xt_flowoffload_hook *
 168 +flow_offload_lookup_hook(struct net_device *dev)
 169 +{
 170 +       struct xt_flowoffload_hook *hook;
 171 +
 172 +       hlist_for_each_entry(hook, &hooks, list) {
 173 +               if (hook->ops.dev == dev)
 174 +                       return hook;
 175 +       }
 176 +
 177 +       return NULL;
 178 +}
 179 +
 180 +static void
 181 +xt_flowoffload_check_device(struct net_device *dev)
 182 +{
 183 +       struct xt_flowoffload_hook *hook;
 184 +
 185 +       spin_lock_bh(&hooks_lock);
 186 +       hook = flow_offload_lookup_hook(dev);
 187 +       if (hook)
 188 +               hook->used = true;
 189 +       else
 190 +               xt_flowoffload_create_hook(dev);
 191 +       spin_unlock_bh(&hooks_lock);
 192 +}
 193 +
 194 +static void
 195 +xt_flowoffload_register_hooks(void)
 196 +{
 197 +       struct xt_flowoffload_hook *hook;
 198 +
 199 +restart:
 200 +       hlist_for_each_entry(hook, &hooks, list) {
 201 +               if (hook->registered)
 202 +                       continue;
 203 +
 204 +               hook->registered = true;
 205 +               hook->net = dev_net(hook->ops.dev);
 206 +               spin_unlock_bh(&hooks_lock);
 207 +               nf_register_net_hook(hook->net, &hook->ops);
 208 +               spin_lock_bh(&hooks_lock);
 209 +               goto restart;
 210 +       }
 211 +
 212 +}
 213 +
 214 +static void
 215 +xt_flowoffload_cleanup_hooks(void)
 216 +{
 217 +       struct xt_flowoffload_hook *hook;
 218 +
 219 +restart:
 220 +       hlist_for_each_entry(hook, &hooks, list) {
 221 +               if (hook->used || !hook->registered)
 222 +                       continue;
 223 +
 224 +               hlist_del(&hook->list);
 225 +               spin_unlock_bh(&hooks_lock);
 226 +               nf_unregister_net_hook(hook->net, &hook->ops);
 227 +               kfree(hook);
 228 +               spin_lock_bh(&hooks_lock);
 229 +               goto restart;
 230 +       }
 231 +
 232 +}
 233 +
 234 +static void
 235 +xt_flowoffload_check_hook(struct flow_offload *flow, void *data)
 236 +{
 237 +       struct flow_offload_tuple *tuple = &flow->tuplehash[0].tuple;
 238 +       struct xt_flowoffload_hook *hook;
 239 +       bool *found = data;
 240 +
 241 +       spin_lock_bh(&hooks_lock);
 242 +       hlist_for_each_entry(hook, &hooks, list) {
 243 +               if (hook->ops.dev->ifindex != tuple->iifidx &&
 244 +                   hook->ops.dev->ifindex != tuple->oifidx)
 245 +                       continue;
 246 +
 247 +               hook->used = true;
 248 +               *found = true;
 249 +       }
 250 +       spin_unlock_bh(&hooks_lock);
 251 +}
 252 +
 253 +static void
 254 +xt_flowoffload_hook_work(struct work_struct *work)
 255 +{
 256 +       struct xt_flowoffload_hook *hook;
 257 +       bool found = false;
 258 +       int err;
 259 +
 260 +       spin_lock_bh(&hooks_lock);
 261 +       xt_flowoffload_register_hooks();
 262 +       hlist_for_each_entry(hook, &hooks, list)
 263 +               hook->used = false;
 264 +       spin_unlock_bh(&hooks_lock);
 265 +
 266 +       err = nf_flow_table_iterate(&nf_flowtable, xt_flowoffload_check_hook,
 267 +                                   &found);
 268 +       if (err && err != -EAGAIN)
 269 +           goto out;
 270 +
 271 +       spin_lock_bh(&hooks_lock);
 272 +       xt_flowoffload_cleanup_hooks();
 273 +       spin_unlock_bh(&hooks_lock);
 274 +
 275 +out:
 276 +       if (found)
 277 +               queue_delayed_work(system_power_efficient_wq, &hook_work, HZ);
 278 +}
 279 +
 280 +static bool
 281 +xt_flowoffload_skip(struct sk_buff *skb)
 282 +{
 283 +       struct ip_options *opt = &(IPCB(skb)->opt);
 284 +
 285 +       if (unlikely(opt->optlen))
 286 +               return true;
 287 +       if (skb_sec_path(skb))
 288 +               return true;
 289 +
 290 +       return false;
 291 +}
 292 +
 293 +static struct dst_entry *
 294 +xt_flowoffload_dst(const struct nf_conn *ct, enum ip_conntrack_dir dir,
 295 +                  const struct xt_action_param *par)
 296 +{
 297 +       struct dst_entry *dst;
 298 +       struct flowi fl;
 299 +
 300 +       memset(&fl, 0, sizeof(fl));
 301 +       switch (xt_family(par)) {
 302 +       case NFPROTO_IPV4:
 303 +               fl.u.ip4.daddr = ct->tuplehash[dir].tuple.src.u3.ip;
 304 +               break;
 305 +       case NFPROTO_IPV6:
 306 +               fl.u.ip6.daddr = ct->tuplehash[dir].tuple.src.u3.in6;
 307 +               break;
 308 +       }
 309 +
 310 +       nf_route(xt_net(par), &dst, &fl, false, xt_family(par));
 311 +
 312 +       return dst;
 313 +}
 314 +
 315 +static int
 316 +xt_flowoffload_route(struct sk_buff *skb, const struct nf_conn *ct,
 317 +                  const struct xt_action_param *par,
 318 +                  struct nf_flow_route *route, enum ip_conntrack_dir dir)
 319 +{
 320 +       struct dst_entry *this_dst, *other_dst;
 321 +
 322 +       this_dst = xt_flowoffload_dst(ct, dir, par);
 323 +       other_dst = xt_flowoffload_dst(ct, !dir, par);
 324 +       if (!this_dst || !other_dst)
 325 +               return -ENOENT;
 326 +
 327 +       route->tuple[dir].dst           = this_dst;
 328 +       route->tuple[dir].ifindex       = xt_in(par)->ifindex;
 329 +       route->tuple[!dir].dst          = other_dst;
 330 +       route->tuple[!dir].ifindex      = xt_out(par)->ifindex;
 331 +
 332 +       return 0;
 333 +}
 334 +
 335 +static unsigned int
 336 +flowoffload_tg(struct sk_buff *skb, const struct xt_action_param *par)
 337 +{
 338 +       enum ip_conntrack_info ctinfo;
 339 +       enum ip_conntrack_dir dir;
 340 +       struct nf_flow_route route;
 341 +       struct flow_offload *flow;
 342 +       struct nf_conn *ct;
 343 +
 344 +       if (xt_flowoffload_skip(skb))
 345 +               return XT_CONTINUE;
 346 +
 347 +       ct = nf_ct_get(skb, &ctinfo);
 348 +       if (ct == NULL)
 349 +               return XT_CONTINUE;
 350 +
 351 +       switch (ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.protonum) {
 352 +       case IPPROTO_TCP:
 353 +               if (ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED)
 354 +                       return XT_CONTINUE;
 355 +               break;
 356 +       case IPPROTO_UDP:
 357 +               break;
 358 +       default:
 359 +               return XT_CONTINUE;
 360 +       }
 361 +
 362 +       if (test_bit(IPS_HELPER_BIT, &ct->status))
 363 +               return XT_CONTINUE;
 364 +
 365 +       if (ctinfo == IP_CT_NEW ||
 366 +           ctinfo == IP_CT_RELATED)
 367 +               return XT_CONTINUE;
 368 +
 369 +       if (!xt_in(par) || !xt_out(par))
 370 +               return XT_CONTINUE;
 371 +
 372 +       if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
 373 +               return XT_CONTINUE;
 374 +
 375 +       dir = CTINFO2DIR(ctinfo);
 376 +
 377 +       if (xt_flowoffload_route(skb, ct, par, &route, dir) < 0)
 378 +               goto err_flow_route;
 379 +
 380 +       flow = flow_offload_alloc(ct, &route);
 381 +       if (!flow)
 382 +               goto err_flow_alloc;
 383 +
 384 +       if (flow_offload_add(&nf_flowtable, flow) < 0)
 385 +               goto err_flow_add;
 386 +
 387 +       xt_flowoffload_check_device(xt_in(par));
 388 +       xt_flowoffload_check_device(xt_out(par));
 389 +
 390 +       return XT_CONTINUE;
 391 +
 392 +err_flow_add:
 393 +       flow_offload_free(flow);
 394 +err_flow_alloc:
 395 +       dst_release(route.tuple[!dir].dst);
 396 +err_flow_route:
 397 +       clear_bit(IPS_OFFLOAD_BIT, &ct->status);
 398 +       return XT_CONTINUE;
 399 +}
 400 +
 401 +
 402 +static int flowoffload_chk(const struct xt_tgchk_param *par)
 403 +{
 404 +       return 0;
 405 +}
 406 +
 407 +static struct xt_target offload_tg_reg __read_mostly = {
 408 +       .family         = NFPROTO_UNSPEC,
 409 +       .name           = "FLOWOFFLOAD",
 410 +       .revision       = 0,
 411 +       .checkentry     = flowoffload_chk,
 412 +       .target         = flowoffload_tg,
 413 +       .me             = THIS_MODULE,
 414 +};
 415 +
 416 +static int xt_flowoffload_table_init(struct nf_flowtable *table)
 417 +{
 418 +       nf_flow_table_init(table);
 419 +       return 0;
 420 +}
 421 +
 422 +static void xt_flowoffload_table_cleanup(struct nf_flowtable *table)
 423 +{
 424 +       nf_flow_table_free(table);
 425 +}
 426 +
 427 +static int __init xt_flowoffload_tg_init(void)
 428 +{
 429 +       int ret;
 430 +
 431 +       INIT_DELAYED_WORK(&hook_work, xt_flowoffload_hook_work);
 432 +
 433 +       ret = xt_flowoffload_table_init(&nf_flowtable);
 434 +       if (ret)
 435 +               return ret;
 436 +
 437 +       ret = xt_register_target(&offload_tg_reg);
 438 +       if (ret)
 439 +               xt_flowoffload_table_cleanup(&nf_flowtable);
 440 +
 441 +       return ret;
 442 +}
 443 +
 444 +static void __exit xt_flowoffload_tg_exit(void)
 445 +{
 446 +       xt_unregister_target(&offload_tg_reg);
 447 +       xt_flowoffload_table_cleanup(&nf_flowtable);
 448 +}
 449 +
 450 +MODULE_LICENSE("GPL");
 451 +module_init(xt_flowoffload_tg_init);
 452 +module_exit(xt_flowoffload_tg_exit);
 453 --- a/net/netfilter/nf_flow_table_core.c
 454 +++ b/net/netfilter/nf_flow_table_core.c
 455 @@ -6,7 +6,6 @@
 456  #include <linux/netdevice.h>
 457  #include <net/ip.h>
 458  #include <net/ip6_route.h>
 459 -#include <net/netfilter/nf_tables.h>
 460  #include <net/netfilter/nf_flow_table.h>
 461  #include <net/netfilter/nf_conntrack.h>
 462  #include <net/netfilter/nf_conntrack_core.h>