1
2#include <linux/kernel.h>
3#include <linux/module.h>
4#include <linux/init.h>
5#include <linux/netlink.h>
6#include <linux/netfilter.h>
7#include <linux/workqueue.h>
8#include <linux/spinlock.h>
9#include <linux/netfilter/nf_conntrack_common.h>
10#include <linux/netfilter/nf_tables.h>
11#include <net/ip.h>
12#include <net/netfilter/nf_tables.h>
13#include <net/netfilter/nf_tables_core.h>
14#include <net/netfilter/nf_conntrack_core.h>
15#include <net/netfilter/nf_conntrack_extend.h>
16#include <net/netfilter/nf_flow_table.h>
17
18struct nft_flow_offload {
19 struct nft_flowtable *flowtable;
20};
21
22static enum flow_offload_xmit_type nft_xmit_type(struct dst_entry *dst)
23{
24 if (dst_xfrm(dst))
25 return FLOW_OFFLOAD_XMIT_XFRM;
26
27 return FLOW_OFFLOAD_XMIT_NEIGH;
28}
29
30static void nft_default_forward_path(struct nf_flow_route *route,
31 struct dst_entry *dst_cache,
32 enum ip_conntrack_dir dir)
33{
34 route->tuple[!dir].in.ifindex = dst_cache->dev->ifindex;
35 route->tuple[dir].dst = dst_cache;
36 route->tuple[dir].xmit_type = nft_xmit_type(dst_cache);
37}
38
39static int nft_dev_fill_forward_path(const struct nf_flow_route *route,
40 const struct dst_entry *dst_cache,
41 const struct nf_conn *ct,
42 enum ip_conntrack_dir dir, u8 *ha,
43 struct net_device_path_stack *stack)
44{
45 const void *daddr = &ct->tuplehash[!dir].tuple.src.u3;
46 struct net_device *dev = dst_cache->dev;
47 struct neighbour *n;
48 u8 nud_state;
49
50 n = dst_neigh_lookup(dst_cache, daddr);
51 if (!n)
52 return -1;
53
54 read_lock_bh(&n->lock);
55 nud_state = n->nud_state;
56 ether_addr_copy(ha, n->ha);
57 read_unlock_bh(&n->lock);
58 neigh_release(n);
59
60 if (!(nud_state & NUD_VALID))
61 return -1;
62
63 return dev_fill_forward_path(dev, ha, stack);
64}
65
66struct nft_forward_info {
67 const struct net_device *indev;
68 const struct net_device *outdev;
69 const struct net_device *hw_outdev;
70 struct id {
71 __u16 id;
72 __be16 proto;
73 } encap[NF_FLOW_TABLE_ENCAP_MAX];
74 u8 num_encaps;
75 u8 ingress_vlans;
76 u8 h_source[ETH_ALEN];
77 u8 h_dest[ETH_ALEN];
78 enum flow_offload_xmit_type xmit_type;
79};
80
81static bool nft_is_valid_ether_device(const struct net_device *dev)
82{
83 if (!dev || (dev->flags & IFF_LOOPBACK) || dev->type != ARPHRD_ETHER ||
84 dev->addr_len != ETH_ALEN || !is_valid_ether_addr(dev->dev_addr))
85 return false;
86
87 return true;
88}
89
90static void nft_dev_path_info(const struct net_device_path_stack *stack,
91 struct nft_forward_info *info,
92 unsigned char *ha, struct nf_flowtable *flowtable)
93{
94 const struct net_device_path *path;
95 int i;
96
97 memcpy(info->h_dest, ha, ETH_ALEN);
98
99 for (i = 0; i < stack->num_paths; i++) {
100 path = &stack->path[i];
101 switch (path->type) {
102 case DEV_PATH_ETHERNET:
103 case DEV_PATH_DSA:
104 case DEV_PATH_VLAN:
105 case DEV_PATH_PPPOE:
106 info->indev = path->dev;
107 if (is_zero_ether_addr(info->h_source))
108 memcpy(info->h_source, path->dev->dev_addr, ETH_ALEN);
109
110 if (path->type == DEV_PATH_ETHERNET)
111 break;
112 if (path->type == DEV_PATH_DSA) {
113 i = stack->num_paths;
114 break;
115 }
116
117
118 if (info->num_encaps >= NF_FLOW_TABLE_ENCAP_MAX) {
119 info->indev = NULL;
120 break;
121 }
122 info->outdev = path->dev;
123 info->encap[info->num_encaps].id = path->encap.id;
124 info->encap[info->num_encaps].proto = path->encap.proto;
125 info->num_encaps++;
126 if (path->type == DEV_PATH_PPPOE)
127 memcpy(info->h_dest, path->encap.h_dest, ETH_ALEN);
128 break;
129 case DEV_PATH_BRIDGE:
130 if (is_zero_ether_addr(info->h_source))
131 memcpy(info->h_source, path->dev->dev_addr, ETH_ALEN);
132
133 switch (path->bridge.vlan_mode) {
134 case DEV_PATH_BR_VLAN_UNTAG_HW:
135 info->ingress_vlans |= BIT(info->num_encaps - 1);
136 break;
137 case DEV_PATH_BR_VLAN_TAG:
138 info->encap[info->num_encaps].id = path->bridge.vlan_id;
139 info->encap[info->num_encaps].proto = path->bridge.vlan_proto;
140 info->num_encaps++;
141 break;
142 case DEV_PATH_BR_VLAN_UNTAG:
143 info->num_encaps--;
144 break;
145 case DEV_PATH_BR_VLAN_KEEP:
146 break;
147 }
148 info->xmit_type = FLOW_OFFLOAD_XMIT_DIRECT;
149 break;
150 default:
151 info->indev = NULL;
152 break;
153 }
154 }
155 if (!info->outdev)
156 info->outdev = info->indev;
157
158 info->hw_outdev = info->indev;
159
160 if (nf_flowtable_hw_offload(flowtable) &&
161 nft_is_valid_ether_device(info->indev))
162 info->xmit_type = FLOW_OFFLOAD_XMIT_DIRECT;
163}
164
165static bool nft_flowtable_find_dev(const struct net_device *dev,
166 struct nft_flowtable *ft)
167{
168 struct nft_hook *hook;
169 bool found = false;
170
171 list_for_each_entry_rcu(hook, &ft->hook_list, list) {
172 if (hook->ops.dev != dev)
173 continue;
174
175 found = true;
176 break;
177 }
178
179 return found;
180}
181
182static void nft_dev_forward_path(struct nf_flow_route *route,
183 const struct nf_conn *ct,
184 enum ip_conntrack_dir dir,
185 struct nft_flowtable *ft)
186{
187 const struct dst_entry *dst = route->tuple[dir].dst;
188 struct net_device_path_stack stack;
189 struct nft_forward_info info = {};
190 unsigned char ha[ETH_ALEN];
191 int i;
192
193 if (nft_dev_fill_forward_path(route, dst, ct, dir, ha, &stack) >= 0)
194 nft_dev_path_info(&stack, &info, ha, &ft->data);
195
196 if (!info.indev || !nft_flowtable_find_dev(info.indev, ft))
197 return;
198
199 route->tuple[!dir].in.ifindex = info.indev->ifindex;
200 for (i = 0; i < info.num_encaps; i++) {
201 route->tuple[!dir].in.encap[i].id = info.encap[i].id;
202 route->tuple[!dir].in.encap[i].proto = info.encap[i].proto;
203 }
204 route->tuple[!dir].in.num_encaps = info.num_encaps;
205 route->tuple[!dir].in.ingress_vlans = info.ingress_vlans;
206
207 if (info.xmit_type == FLOW_OFFLOAD_XMIT_DIRECT) {
208 memcpy(route->tuple[dir].out.h_source, info.h_source, ETH_ALEN);
209 memcpy(route->tuple[dir].out.h_dest, info.h_dest, ETH_ALEN);
210 route->tuple[dir].out.ifindex = info.outdev->ifindex;
211 route->tuple[dir].out.hw_ifindex = info.hw_outdev->ifindex;
212 route->tuple[dir].xmit_type = info.xmit_type;
213 }
214}
215
216static int nft_flow_route(const struct nft_pktinfo *pkt,
217 const struct nf_conn *ct,
218 struct nf_flow_route *route,
219 enum ip_conntrack_dir dir,
220 struct nft_flowtable *ft)
221{
222 struct dst_entry *this_dst = skb_dst(pkt->skb);
223 struct dst_entry *other_dst = NULL;
224 struct flowi fl;
225
226 memset(&fl, 0, sizeof(fl));
227 switch (nft_pf(pkt)) {
228 case NFPROTO_IPV4:
229 fl.u.ip4.daddr = ct->tuplehash[dir].tuple.src.u3.ip;
230 fl.u.ip4.flowi4_oif = nft_in(pkt)->ifindex;
231 break;
232 case NFPROTO_IPV6:
233 fl.u.ip6.daddr = ct->tuplehash[dir].tuple.src.u3.in6;
234 fl.u.ip6.flowi6_oif = nft_in(pkt)->ifindex;
235 break;
236 }
237
238 nf_route(nft_net(pkt), &other_dst, &fl, false, nft_pf(pkt));
239 if (!other_dst)
240 return -ENOENT;
241
242 nft_default_forward_path(route, this_dst, dir);
243 nft_default_forward_path(route, other_dst, !dir);
244
245 if (route->tuple[dir].xmit_type == FLOW_OFFLOAD_XMIT_NEIGH &&
246 route->tuple[!dir].xmit_type == FLOW_OFFLOAD_XMIT_NEIGH) {
247 nft_dev_forward_path(route, ct, dir, ft);
248 nft_dev_forward_path(route, ct, !dir, ft);
249 }
250
251 return 0;
252}
253
254static bool nft_flow_offload_skip(struct sk_buff *skb, int family)
255{
256 if (skb_sec_path(skb))
257 return true;
258
259 if (family == NFPROTO_IPV4) {
260 const struct ip_options *opt;
261
262 opt = &(IPCB(skb)->opt);
263
264 if (unlikely(opt->optlen))
265 return true;
266 }
267
268 return false;
269}
270
271static void nft_flow_offload_eval(const struct nft_expr *expr,
272 struct nft_regs *regs,
273 const struct nft_pktinfo *pkt)
274{
275 struct nft_flow_offload *priv = nft_expr_priv(expr);
276 struct nf_flowtable *flowtable = &priv->flowtable->data;
277 struct tcphdr _tcph, *tcph = NULL;
278 struct nf_flow_route route = {};
279 enum ip_conntrack_info ctinfo;
280 struct flow_offload *flow;
281 enum ip_conntrack_dir dir;
282 struct nf_conn *ct;
283 int ret;
284
285 if (nft_flow_offload_skip(pkt->skb, nft_pf(pkt)))
286 goto out;
287
288 ct = nf_ct_get(pkt->skb, &ctinfo);
289 if (!ct)
290 goto out;
291
292 switch (ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.protonum) {
293 case IPPROTO_TCP:
294 tcph = skb_header_pointer(pkt->skb, nft_thoff(pkt),
295 sizeof(_tcph), &_tcph);
296 if (unlikely(!tcph || tcph->fin || tcph->rst))
297 goto out;
298 break;
299 case IPPROTO_UDP:
300 break;
301 default:
302 goto out;
303 }
304
305 if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
306 ct->status & (IPS_SEQ_ADJUST | IPS_NAT_CLASH))
307 goto out;
308
309 if (!nf_ct_is_confirmed(ct))
310 goto out;
311
312 if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
313 goto out;
314
315 dir = CTINFO2DIR(ctinfo);
316 if (nft_flow_route(pkt, ct, &route, dir, priv->flowtable) < 0)
317 goto err_flow_route;
318
319 flow = flow_offload_alloc(ct);
320 if (!flow)
321 goto err_flow_alloc;
322
323 if (flow_offload_route_init(flow, &route) < 0)
324 goto err_flow_add;
325
326 if (tcph) {
327 ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
328 ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
329 }
330
331 ret = flow_offload_add(flowtable, flow);
332 if (ret < 0)
333 goto err_flow_add;
334
335 dst_release(route.tuple[!dir].dst);
336 return;
337
338err_flow_add:
339 flow_offload_free(flow);
340err_flow_alloc:
341 dst_release(route.tuple[!dir].dst);
342err_flow_route:
343 clear_bit(IPS_OFFLOAD_BIT, &ct->status);
344out:
345 regs->verdict.code = NFT_BREAK;
346}
347
348static int nft_flow_offload_validate(const struct nft_ctx *ctx,
349 const struct nft_expr *expr,
350 const struct nft_data **data)
351{
352 unsigned int hook_mask = (1 << NF_INET_FORWARD);
353
354 return nft_chain_validate_hooks(ctx->chain, hook_mask);
355}
356
357static const struct nla_policy nft_flow_offload_policy[NFTA_FLOW_MAX + 1] = {
358 [NFTA_FLOW_TABLE_NAME] = { .type = NLA_STRING,
359 .len = NFT_NAME_MAXLEN - 1 },
360};
361
362static int nft_flow_offload_init(const struct nft_ctx *ctx,
363 const struct nft_expr *expr,
364 const struct nlattr * const tb[])
365{
366 struct nft_flow_offload *priv = nft_expr_priv(expr);
367 u8 genmask = nft_genmask_next(ctx->net);
368 struct nft_flowtable *flowtable;
369
370 if (!tb[NFTA_FLOW_TABLE_NAME])
371 return -EINVAL;
372
373 flowtable = nft_flowtable_lookup(ctx->table, tb[NFTA_FLOW_TABLE_NAME],
374 genmask);
375 if (IS_ERR(flowtable))
376 return PTR_ERR(flowtable);
377
378 priv->flowtable = flowtable;
379 flowtable->use++;
380
381 return nf_ct_netns_get(ctx->net, ctx->family);
382}
383
384static void nft_flow_offload_deactivate(const struct nft_ctx *ctx,
385 const struct nft_expr *expr,
386 enum nft_trans_phase phase)
387{
388 struct nft_flow_offload *priv = nft_expr_priv(expr);
389
390 nf_tables_deactivate_flowtable(ctx, priv->flowtable, phase);
391}
392
393static void nft_flow_offload_activate(const struct nft_ctx *ctx,
394 const struct nft_expr *expr)
395{
396 struct nft_flow_offload *priv = nft_expr_priv(expr);
397
398 priv->flowtable->use++;
399}
400
401static void nft_flow_offload_destroy(const struct nft_ctx *ctx,
402 const struct nft_expr *expr)
403{
404 nf_ct_netns_put(ctx->net, ctx->family);
405}
406
407static int nft_flow_offload_dump(struct sk_buff *skb, const struct nft_expr *expr)
408{
409 struct nft_flow_offload *priv = nft_expr_priv(expr);
410
411 if (nla_put_string(skb, NFTA_FLOW_TABLE_NAME, priv->flowtable->name))
412 goto nla_put_failure;
413
414 return 0;
415
416nla_put_failure:
417 return -1;
418}
419
420static struct nft_expr_type nft_flow_offload_type;
421static const struct nft_expr_ops nft_flow_offload_ops = {
422 .type = &nft_flow_offload_type,
423 .size = NFT_EXPR_SIZE(sizeof(struct nft_flow_offload)),
424 .eval = nft_flow_offload_eval,
425 .init = nft_flow_offload_init,
426 .activate = nft_flow_offload_activate,
427 .deactivate = nft_flow_offload_deactivate,
428 .destroy = nft_flow_offload_destroy,
429 .validate = nft_flow_offload_validate,
430 .dump = nft_flow_offload_dump,
431};
432
433static struct nft_expr_type nft_flow_offload_type __read_mostly = {
434 .name = "flow_offload",
435 .ops = &nft_flow_offload_ops,
436 .policy = nft_flow_offload_policy,
437 .maxattr = NFTA_FLOW_MAX,
438 .owner = THIS_MODULE,
439};
440
441static int flow_offload_netdev_event(struct notifier_block *this,
442 unsigned long event, void *ptr)
443{
444 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
445
446 if (event != NETDEV_DOWN)
447 return NOTIFY_DONE;
448
449 nf_flow_table_cleanup(dev);
450
451 return NOTIFY_DONE;
452}
453
454static struct notifier_block flow_offload_netdev_notifier = {
455 .notifier_call = flow_offload_netdev_event,
456};
457
458static int __init nft_flow_offload_module_init(void)
459{
460 int err;
461
462 err = register_netdevice_notifier(&flow_offload_netdev_notifier);
463 if (err)
464 goto err;
465
466 err = nft_register_expr(&nft_flow_offload_type);
467 if (err < 0)
468 goto register_expr;
469
470 return 0;
471
472register_expr:
473 unregister_netdevice_notifier(&flow_offload_netdev_notifier);
474err:
475 return err;
476}
477
478static void __exit nft_flow_offload_module_exit(void)
479{
480 nft_unregister_expr(&nft_flow_offload_type);
481 unregister_netdevice_notifier(&flow_offload_netdev_notifier);
482}
483
484module_init(nft_flow_offload_module_init);
485module_exit(nft_flow_offload_module_exit);
486
487MODULE_LICENSE("GPL");
488MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
489MODULE_ALIAS_NFT_EXPR("flow_offload");
490MODULE_DESCRIPTION("nftables hardware flow offload module");
491