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
5 Signed-off-by: Felix Fietkau <nbd@nbd.name>
7 create mode 100644 net/netfilter/xt_OFFLOAD.c
9 --- a/net/ipv4/netfilter/Kconfig
10 +++ b/net/ipv4/netfilter/Kconfig
11 @@ -75,8 +75,6 @@ config NF_TABLES_ARP
13 This option enables the ARP support for nf_tables.
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
22 To compile it as a module, choose M here.
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.
34 endif # NF_TABLES_IPV6
37 config NF_FLOW_TABLE_IPV6
38 tristate "Netfilter flow table IPv6 module"
39 @@ -79,6 +78,8 @@ config NF_FLOW_TABLE_IPV6
41 To compile it as a module, choose M here.
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
52 endif # NF_TABLES_NETDEV
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
61 To compile it as a module, choose M here.
66 tristate "Netfilter flow table module"
67 depends on NETFILTER_INGRESS
68 depends on NF_CONNTRACK
69 - depends on NF_TABLES
71 This option adds the flow table core infrastructure.
73 @@ -959,6 +958,15 @@ config NETFILTER_XT_TARGET_NOTRACK
74 depends on NETFILTER_ADVANCED
75 select NETFILTER_XT_TARGET_CT
77 +config NETFILTER_XT_TARGET_FLOWOFFLOAD
78 + tristate '"FLOWOFFLOAD" target support'
79 + depends on NF_FLOW_TABLE
80 + depends on NETFILTER_INGRESS
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
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
100 +++ b/net/netfilter/xt_FLOWOFFLOAD.c
103 + * Copyright (C) 2018 Felix Fietkau <nbd@nbd.name>
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.
109 +#include <linux/module.h>
110 +#include <linux/init.h>
111 +#include <linux/netfilter.h>
113 +#include <net/netfilter/nf_conntrack.h>
114 +#include <net/netfilter/nf_flow_table.h>
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;
121 +struct xt_flowoffload_hook {
122 + struct hlist_node list;
123 + struct nf_hook_ops ops;
130 +xt_flowoffload_net_hook(void *priv, struct sk_buff *skb,
131 + const struct nf_hook_state *state)
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);
144 +xt_flowoffload_create_hook(struct net_device *dev)
146 + struct xt_flowoffload_hook *hook;
147 + struct nf_hook_ops *ops;
149 + hook = kzalloc(sizeof(*hook), GFP_ATOMIC);
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;
161 + hlist_add_head(&hook->list, &hooks);
162 + mod_delayed_work(system_power_efficient_wq, &hook_work, 0);
167 +static struct xt_flowoffload_hook *
168 +flow_offload_lookup_hook(struct net_device *dev)
170 + struct xt_flowoffload_hook *hook;
172 + hlist_for_each_entry(hook, &hooks, list) {
173 + if (hook->ops.dev == dev)
181 +xt_flowoffload_check_device(struct net_device *dev)
183 + struct xt_flowoffload_hook *hook;
185 + spin_lock_bh(&hooks_lock);
186 + hook = flow_offload_lookup_hook(dev);
190 + xt_flowoffload_create_hook(dev);
191 + spin_unlock_bh(&hooks_lock);
195 +xt_flowoffload_register_hooks(void)
197 + struct xt_flowoffload_hook *hook;
200 + hlist_for_each_entry(hook, &hooks, list) {
201 + if (hook->registered)
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);
215 +xt_flowoffload_cleanup_hooks(void)
217 + struct xt_flowoffload_hook *hook;
220 + hlist_for_each_entry(hook, &hooks, list) {
221 + if (hook->used || !hook->registered)
224 + hlist_del(&hook->list);
225 + spin_unlock_bh(&hooks_lock);
226 + nf_unregister_net_hook(hook->net, &hook->ops);
228 + spin_lock_bh(&hooks_lock);
235 +xt_flowoffload_check_hook(struct flow_offload *flow, void *data)
237 + struct flow_offload_tuple *tuple = &flow->tuplehash[0].tuple;
238 + struct xt_flowoffload_hook *hook;
239 + bool *found = data;
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)
250 + spin_unlock_bh(&hooks_lock);
254 +xt_flowoffload_hook_work(struct work_struct *work)
256 + struct xt_flowoffload_hook *hook;
257 + bool found = false;
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);
266 + err = nf_flow_table_iterate(&nf_flowtable, xt_flowoffload_check_hook,
268 + if (err && err != -EAGAIN)
271 + spin_lock_bh(&hooks_lock);
272 + xt_flowoffload_cleanup_hooks();
273 + spin_unlock_bh(&hooks_lock);
277 + queue_delayed_work(system_power_efficient_wq, &hook_work, HZ);
281 +xt_flowoffload_skip(struct sk_buff *skb)
283 + struct ip_options *opt = &(IPCB(skb)->opt);
285 + if (unlikely(opt->optlen))
287 + if (skb_sec_path(skb))
294 +xt_flowoffload_route(struct sk_buff *skb, const struct nf_conn *ct,
295 + const struct xt_action_param *par,
296 + struct nf_flow_route *route, enum ip_conntrack_dir dir)
298 + struct dst_entry *this_dst = skb_dst(skb);
299 + struct dst_entry *other_dst = NULL;
302 + memset(&fl, 0, sizeof(fl));
303 + switch (xt_family(par)) {
305 + fl.u.ip4.daddr = ct->tuplehash[!dir].tuple.dst.u3.ip;
308 + fl.u.ip6.daddr = ct->tuplehash[!dir].tuple.dst.u3.in6;
312 + nf_route(xt_net(par), &other_dst, &fl, false, xt_family(par));
316 + route->tuple[dir].dst = this_dst;
317 + route->tuple[dir].ifindex = xt_in(par)->ifindex;
318 + route->tuple[!dir].dst = other_dst;
319 + route->tuple[!dir].ifindex = xt_out(par)->ifindex;
325 +flowoffload_tg(struct sk_buff *skb, const struct xt_action_param *par)
327 + enum ip_conntrack_info ctinfo;
328 + enum ip_conntrack_dir dir;
329 + struct nf_flow_route route;
330 + struct flow_offload *flow;
331 + struct nf_conn *ct;
333 + if (xt_flowoffload_skip(skb))
334 + return XT_CONTINUE;
336 + ct = nf_ct_get(skb, &ctinfo);
338 + return XT_CONTINUE;
340 + switch (ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.protonum) {
342 + if (ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED)
343 + return XT_CONTINUE;
348 + return XT_CONTINUE;
351 + if (test_bit(IPS_HELPER_BIT, &ct->status))
352 + return XT_CONTINUE;
354 + if (ctinfo == IP_CT_NEW ||
355 + ctinfo == IP_CT_RELATED)
356 + return XT_CONTINUE;
358 + if (!xt_in(par) || !xt_out(par))
359 + return XT_CONTINUE;
361 + if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
362 + return XT_CONTINUE;
364 + dir = CTINFO2DIR(ctinfo);
366 + if (xt_flowoffload_route(skb, ct, par, &route, dir) < 0)
367 + goto err_flow_route;
369 + flow = flow_offload_alloc(ct, &route);
371 + goto err_flow_alloc;
373 + if (flow_offload_add(&nf_flowtable, flow) < 0)
376 + xt_flowoffload_check_device(xt_in(par));
377 + xt_flowoffload_check_device(xt_out(par));
379 + return XT_CONTINUE;
382 + flow_offload_free(flow);
384 + dst_release(route.tuple[!dir].dst);
386 + clear_bit(IPS_OFFLOAD_BIT, &ct->status);
387 + return XT_CONTINUE;
391 +static int flowoffload_chk(const struct xt_tgchk_param *par)
396 +static struct xt_target offload_tg_reg __read_mostly = {
397 + .family = NFPROTO_UNSPEC,
398 + .name = "FLOWOFFLOAD",
400 + .checkentry = flowoffload_chk,
401 + .target = flowoffload_tg,
405 +static int xt_flowoffload_table_init(struct nf_flowtable *table)
407 + nf_flow_table_init(table);
411 +static void xt_flowoffload_table_cleanup(struct nf_flowtable *table)
413 + nf_flow_table_free(table);
416 +static int __init xt_flowoffload_tg_init(void)
420 + INIT_DELAYED_WORK(&hook_work, xt_flowoffload_hook_work);
422 + ret = xt_flowoffload_table_init(&nf_flowtable);
426 + ret = xt_register_target(&offload_tg_reg);
428 + xt_flowoffload_table_cleanup(&nf_flowtable);
433 +static void __exit xt_flowoffload_tg_exit(void)
435 + xt_unregister_target(&offload_tg_reg);
436 + xt_flowoffload_table_cleanup(&nf_flowtable);
439 +MODULE_LICENSE("GPL");
440 +module_init(xt_flowoffload_tg_init);
441 +module_exit(xt_flowoffload_tg_exit);
442 --- a/net/netfilter/nf_flow_table_core.c
443 +++ b/net/netfilter/nf_flow_table_core.c
445 #include <linux/netdevice.h>
447 #include <net/ip6_route.h>
448 -#include <net/netfilter/nf_tables.h>
449 #include <net/netfilter/nf_flow_table.h>
450 #include <net/netfilter/nf_conntrack.h>
451 #include <net/netfilter/nf_conntrack_core.h>