nf_flow_table_path.c

  1// SPDX-License-Identifier: GPL-2.0-only
  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/spinlock.h>
  8#include <linux/netfilter/nf_conntrack_common.h>
  9#include <linux/netfilter/nf_tables.h>
 10#include <net/ip.h>
 11#include <net/inet_dscp.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
 18static enum flow_offload_xmit_type nft_xmit_type(struct dst_entry *dst)
 19{
 20	if (dst_xfrm(dst))
 21		return FLOW_OFFLOAD_XMIT_XFRM;
 22
 23	return FLOW_OFFLOAD_XMIT_NEIGH;
 24}
 25
 26static void nft_default_forward_path(struct nf_flow_route *route,
 27				     struct dst_entry *dst_cache,
 28				     enum ip_conntrack_dir dir)
 29{
 30	route->tuple[!dir].in.ifindex	= dst_cache->dev->ifindex;
 31	route->tuple[dir].dst		= dst_cache;
 32	route->tuple[dir].xmit_type	= nft_xmit_type(dst_cache);
 33}
 34
 35static bool nft_is_valid_ether_device(const struct net_device *dev)
 36{
 37	if (!dev || (dev->flags & IFF_LOOPBACK) || dev->type != ARPHRD_ETHER ||
 38	    dev->addr_len != ETH_ALEN || !is_valid_ether_addr(dev->dev_addr))
 39		return false;
 40
 41	return true;
 42}
 43
 44static int nft_dev_fill_forward_path(const struct nf_flow_route *route,
 45				     const struct dst_entry *dst_cache,
 46				     const struct nf_conn *ct,
 47				     enum ip_conntrack_dir dir, u8 *ha,
 48				     struct net_device_path_stack *stack)
 49{
 50	const void *daddr = &ct->tuplehash[!dir].tuple.src.u3;
 51	struct net_device *dev = dst_cache->dev;
 52	struct neighbour *n;
 53	u8 nud_state;
 54
 55	if (!nft_is_valid_ether_device(dev))
 56		goto out;
 57
 58	n = dst_neigh_lookup(dst_cache, daddr);
 59	if (!n)
 60		return -1;
 61
 62	read_lock_bh(&n->lock);
 63	nud_state = n->nud_state;
 64	ether_addr_copy(ha, n->ha);
 65	read_unlock_bh(&n->lock);
 66	neigh_release(n);
 67
 68	if (!(nud_state & NUD_VALID))
 69		return -1;
 70
 71out:
 72	return dev_fill_forward_path(dev, ha, stack);
 73}
 74
 75struct nft_forward_info {
 76	const struct net_device *indev;
 77	const struct net_device *outdev;
 78	struct id {
 79		__u16	id;
 80		__be16	proto;
 81	} encap[NF_FLOW_TABLE_ENCAP_MAX];
 82	u8 num_encaps;
 83	struct flow_offload_tunnel tun;
 84	u8 num_tuns;
 85	u8 ingress_vlans;
 86	u8 h_source[ETH_ALEN];
 87	u8 h_dest[ETH_ALEN];
 88	enum flow_offload_xmit_type xmit_type;
 89};
 90
 91static void nft_dev_path_info(const struct net_device_path_stack *stack,
 92			      struct nft_forward_info *info,
 93			      unsigned char *ha, struct nf_flowtable *flowtable)
 94{
 95	const struct net_device_path *path;
 96	int i;
 97
 98	memcpy(info->h_dest, ha, ETH_ALEN);
 99
100	for (i = 0; i < stack->num_paths; i++) {
101		path = &stack->path[i];
102		switch (path->type) {
103		case DEV_PATH_ETHERNET:
104		case DEV_PATH_DSA:
105		case DEV_PATH_VLAN:
106		case DEV_PATH_PPPOE:
107		case DEV_PATH_TUN:
108			info->indev = path->dev;
109			if (is_zero_ether_addr(info->h_source))
110				memcpy(info->h_source, path->dev->dev_addr, ETH_ALEN);
111
112			if (path->type == DEV_PATH_ETHERNET)
113				break;
114			if (path->type == DEV_PATH_DSA) {
115				i = stack->num_paths;
116				break;
117			}
118
119			/* DEV_PATH_VLAN, DEV_PATH_PPPOE and DEV_PATH_TUN */
120			if (path->type == DEV_PATH_TUN) {
121				if (info->num_tuns) {
122					info->indev = NULL;
123					break;
124				}
125				info->tun.src_v6 = path->tun.src_v6;
126				info->tun.dst_v6 = path->tun.dst_v6;
127				info->tun.l3_proto = path->tun.l3_proto;
128				info->num_tuns++;
129			} else {
130				if (info->num_encaps >= NF_FLOW_TABLE_ENCAP_MAX) {
131					info->indev = NULL;
132					break;
133				}
134				info->encap[info->num_encaps].id =
135					path->encap.id;
136				info->encap[info->num_encaps].proto =
137					path->encap.proto;
138				info->num_encaps++;
139			}
140			if (path->type == DEV_PATH_PPPOE)
141				memcpy(info->h_dest, path->encap.h_dest, ETH_ALEN);
142			break;
143		case DEV_PATH_BRIDGE:
144			if (is_zero_ether_addr(info->h_source))
145				memcpy(info->h_source, path->dev->dev_addr, ETH_ALEN);
146
147			switch (path->bridge.vlan_mode) {
148			case DEV_PATH_BR_VLAN_UNTAG_HW:
149				info->ingress_vlans |= BIT(info->num_encaps - 1);
150				break;
151			case DEV_PATH_BR_VLAN_TAG:
152				if (info->num_encaps >= NF_FLOW_TABLE_ENCAP_MAX) {
153					info->indev = NULL;
154					break;
155				}
156				info->encap[info->num_encaps].id = path->bridge.vlan_id;
157				info->encap[info->num_encaps].proto = path->bridge.vlan_proto;
158				info->num_encaps++;
159				break;
160			case DEV_PATH_BR_VLAN_UNTAG:
161				if (WARN_ON_ONCE(info->num_encaps-- == 0)) {
162					info->indev = NULL;
163					break;
164				}
165				break;
166			case DEV_PATH_BR_VLAN_KEEP:
167				break;
168			}
169			info->xmit_type = FLOW_OFFLOAD_XMIT_DIRECT;
170			break;
171		default:
172			info->indev = NULL;
173			break;
174		}
175	}
176	info->outdev = info->indev;
177
178	if (nf_flowtable_hw_offload(flowtable) &&
179	    nft_is_valid_ether_device(info->indev))
180		info->xmit_type = FLOW_OFFLOAD_XMIT_DIRECT;
181}
182
183static bool nft_flowtable_find_dev(const struct net_device *dev,
184				   struct nft_flowtable *ft)
185{
186	struct nft_hook *hook;
187	bool found = false;
188
189	list_for_each_entry_rcu(hook, &ft->hook_list, list) {
190		if (!nft_hook_find_ops_rcu(hook, dev))
191			continue;
192
193		found = true;
194		break;
195	}
196
197	return found;
198}
199
200static int nft_flow_tunnel_update_route(const struct nft_pktinfo *pkt,
201					struct flow_offload_tunnel *tun,
202					struct nf_flow_route *route,
203					enum ip_conntrack_dir dir)
204{
205	struct dst_entry *cur_dst = route->tuple[dir].dst;
206	struct dst_entry *tun_dst = NULL;
207	struct flowi fl = {};
208
209	switch (nft_pf(pkt)) {
210	case NFPROTO_IPV4:
211		fl.u.ip4.daddr = tun->dst_v4.s_addr;
212		fl.u.ip4.saddr = tun->src_v4.s_addr;
213		fl.u.ip4.flowi4_iif = nft_in(pkt)->ifindex;
214		fl.u.ip4.flowi4_dscp = ip4h_dscp(ip_hdr(pkt->skb));
215		fl.u.ip4.flowi4_mark = pkt->skb->mark;
216		fl.u.ip4.flowi4_flags = FLOWI_FLAG_ANYSRC;
217		break;
218	case NFPROTO_IPV6:
219		fl.u.ip6.daddr = tun->dst_v6;
220		fl.u.ip6.saddr = tun->src_v6;
221		fl.u.ip6.flowi6_iif = nft_in(pkt)->ifindex;
222		fl.u.ip6.flowlabel = ip6_flowinfo(ipv6_hdr(pkt->skb));
223		fl.u.ip6.flowi6_mark = pkt->skb->mark;
224		fl.u.ip6.flowi6_flags = FLOWI_FLAG_ANYSRC;
225		break;
226	}
227
228	nf_route(nft_net(pkt), &tun_dst, &fl, false, nft_pf(pkt));
229	if (!tun_dst)
230		return -ENOENT;
231
232	route->tuple[dir].dst = tun_dst;
233	dst_release(cur_dst);
234
235	return 0;
236}
237
238static void nft_dev_forward_path(const struct nft_pktinfo *pkt,
239				 struct nf_flow_route *route,
240				 const struct nf_conn *ct,
241				 enum ip_conntrack_dir dir,
242				 struct nft_flowtable *ft)
243{
244	const struct dst_entry *dst = route->tuple[dir].dst;
245	struct net_device_path_stack stack;
246	struct nft_forward_info info = {};
247	unsigned char ha[ETH_ALEN];
248	int i;
249
250	if (nft_dev_fill_forward_path(route, dst, ct, dir, ha, &stack) >= 0)
251		nft_dev_path_info(&stack, &info, ha, &ft->data);
252
253	if (info.outdev)
254		route->tuple[dir].out.ifindex = info.outdev->ifindex;
255
256	if (!info.indev || !nft_flowtable_find_dev(info.indev, ft))
257		return;
258
259	route->tuple[!dir].in.ifindex = info.indev->ifindex;
260	for (i = 0; i < info.num_encaps; i++) {
261		route->tuple[!dir].in.encap[i].id = info.encap[i].id;
262		route->tuple[!dir].in.encap[i].proto = info.encap[i].proto;
263	}
264
265	if (info.num_tuns &&
266	    !nft_flow_tunnel_update_route(pkt, &info.tun, route, dir)) {
267		route->tuple[!dir].in.tun.src_v6 = info.tun.dst_v6;
268		route->tuple[!dir].in.tun.dst_v6 = info.tun.src_v6;
269		route->tuple[!dir].in.tun.l3_proto = info.tun.l3_proto;
270		route->tuple[!dir].in.num_tuns = info.num_tuns;
271	}
272
273	route->tuple[!dir].in.num_encaps = info.num_encaps;
274	route->tuple[!dir].in.ingress_vlans = info.ingress_vlans;
275
276	if (info.xmit_type == FLOW_OFFLOAD_XMIT_DIRECT) {
277		memcpy(route->tuple[dir].out.h_source, info.h_source, ETH_ALEN);
278		memcpy(route->tuple[dir].out.h_dest, info.h_dest, ETH_ALEN);
279		route->tuple[dir].xmit_type = info.xmit_type;
280	}
281}
282
283int nft_flow_route(const struct nft_pktinfo *pkt, const struct nf_conn *ct,
284		   struct nf_flow_route *route, enum ip_conntrack_dir dir,
285		   struct nft_flowtable *ft)
286{
287	struct dst_entry *this_dst = skb_dst(pkt->skb);
288	struct dst_entry *other_dst = NULL;
289	struct flowi fl;
290
291	memset(&fl, 0, sizeof(fl));
292	switch (nft_pf(pkt)) {
293	case NFPROTO_IPV4:
294		fl.u.ip4.daddr = ct->tuplehash[dir].tuple.src.u3.ip;
295		fl.u.ip4.saddr = ct->tuplehash[!dir].tuple.src.u3.ip;
296		fl.u.ip4.flowi4_oif = nft_in(pkt)->ifindex;
297		fl.u.ip4.flowi4_iif = this_dst->dev->ifindex;
298		fl.u.ip4.flowi4_dscp = ip4h_dscp(ip_hdr(pkt->skb));
299		fl.u.ip4.flowi4_mark = pkt->skb->mark;
300		fl.u.ip4.flowi4_flags = FLOWI_FLAG_ANYSRC;
301		break;
302	case NFPROTO_IPV6:
303		fl.u.ip6.daddr = ct->tuplehash[dir].tuple.src.u3.in6;
304		fl.u.ip6.saddr = ct->tuplehash[!dir].tuple.src.u3.in6;
305		fl.u.ip6.flowi6_oif = nft_in(pkt)->ifindex;
306		fl.u.ip6.flowi6_iif = this_dst->dev->ifindex;
307		fl.u.ip6.flowlabel = ip6_flowinfo(ipv6_hdr(pkt->skb));
308		fl.u.ip6.flowi6_mark = pkt->skb->mark;
309		fl.u.ip6.flowi6_flags = FLOWI_FLAG_ANYSRC;
310		break;
311	}
312
313	if (!dst_hold_safe(this_dst))
314		return -ENOENT;
315
316	nf_route(nft_net(pkt), &other_dst, &fl, false, nft_pf(pkt));
317	if (!other_dst) {
318		dst_release(this_dst);
319		return -ENOENT;
320	}
321
322	nft_default_forward_path(route, this_dst, dir);
323	nft_default_forward_path(route, other_dst, !dir);
324
325	if (route->tuple[dir].xmit_type	== FLOW_OFFLOAD_XMIT_NEIGH)
326		nft_dev_forward_path(pkt, route, ct, dir, ft);
327	if (route->tuple[!dir].xmit_type == FLOW_OFFLOAD_XMIT_NEIGH)
328		nft_dev_forward_path(pkt, route, ct, !dir, ft);
329
330	return 0;
331}
332EXPORT_SYMBOL_GPL(nft_flow_route);