1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * IPv6 specific functions of netfilter core
4 *
5 * Rusty Russell (C) 2000
6 * Patrick McHardy (C) 2006-2012
7 */
8 #include <linux/kernel.h>
9 #include <linux/init.h>
10 #include <linux/ipv6.h>
11 #include <linux/netfilter.h>
12 #include <linux/netfilter_ipv6.h>
13 #include <linux/export.h>
14 #include <net/addrconf.h>
15 #include <net/dst.h>
16 #include <net/ipv6.h>
17 #include <net/ip6_route.h>
18 #include <net/xfrm.h>
19 #include <net/netfilter/nf_queue.h>
20 #include <net/netfilter/nf_conntrack_bridge.h>
21 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
22 #include "../bridge/br_private.h"
23
ip6_route_me_harder(struct net * net,struct sock * sk_partial,struct sk_buff * skb)24 int ip6_route_me_harder(struct net *net, struct sock *sk_partial, struct sk_buff *skb)
25 {
26 const struct ipv6hdr *iph = ipv6_hdr(skb);
27 struct sock *sk = sk_to_full_sk(sk_partial);
28 struct net_device *dev = skb_dst_dev(skb);
29 struct flow_keys flkeys;
30 unsigned int hh_len;
31 struct dst_entry *dst;
32 int strict = (ipv6_addr_type(&iph->daddr) &
33 (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL));
34 struct flowi6 fl6 = {
35 .flowi6_l3mdev = l3mdev_master_ifindex(dev),
36 .flowi6_mark = skb->mark,
37 .flowi6_uid = sock_net_uid(net, sk),
38 .daddr = iph->daddr,
39 .saddr = iph->saddr,
40 .flowlabel = ip6_flowinfo(iph),
41 };
42 int err;
43
44 if (sk && sk->sk_bound_dev_if)
45 fl6.flowi6_oif = sk->sk_bound_dev_if;
46 else if (strict)
47 fl6.flowi6_oif = dev->ifindex;
48
49 fib6_rules_early_flow_dissect(net, skb, &fl6, &flkeys);
50 dst = ip6_route_output(net, sk, &fl6);
51 err = dst->error;
52 if (err) {
53 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
54 net_dbg_ratelimited("ip6_route_me_harder: No more route\n");
55 dst_release(dst);
56 return err;
57 }
58
59 /* Drop old route. */
60 skb_dst_drop(skb);
61
62 skb_dst_set(skb, dst);
63
64 #ifdef CONFIG_XFRM
65 if (!(IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED) &&
66 xfrm_decode_session(net, skb, flowi6_to_flowi(&fl6), AF_INET6) == 0) {
67 /* ignore return value from skb_dstref_steal, xfrm_lookup takes
68 * care of dropping the refcnt if needed.
69 */
70 skb_dstref_steal(skb);
71 dst = xfrm_lookup(net, dst, flowi6_to_flowi(&fl6), sk, 0);
72 if (IS_ERR(dst))
73 return PTR_ERR(dst);
74 skb_dst_set(skb, dst);
75 }
76 #endif
77
78 /* Change in oif may mean change in hh_len. */
79 hh_len = skb_dst_dev(skb)->hard_header_len;
80 if (skb_headroom(skb) < hh_len &&
81 pskb_expand_head(skb, HH_DATA_ALIGN(hh_len - skb_headroom(skb)),
82 0, GFP_ATOMIC))
83 return -ENOMEM;
84
85 return 0;
86 }
87 EXPORT_SYMBOL(ip6_route_me_harder);
88
__nf_ip6_route(struct net * net,struct dst_entry ** dst,struct flowi * fl,bool strict)89 int __nf_ip6_route(struct net *net, struct dst_entry **dst,
90 struct flowi *fl, bool strict)
91 {
92 static const struct ipv6_pinfo fake_pinfo;
93 static const struct inet_sock fake_sk = {
94 /* makes ip6_route_output set RT6_LOOKUP_F_IFACE: */
95 .sk.sk_bound_dev_if = 1,
96 .pinet6 = (struct ipv6_pinfo *) &fake_pinfo,
97 };
98 const void *sk = strict ? &fake_sk : NULL;
99 struct dst_entry *result;
100 int err;
101
102 result = ip6_route_output(net, sk, &fl->u.ip6);
103 err = result->error;
104 if (err)
105 dst_release(result);
106 else
107 *dst = result;
108 return err;
109 }
110 EXPORT_SYMBOL_GPL(__nf_ip6_route);
111
br_ip6_fragment(struct net * net,struct sock * sk,struct sk_buff * skb,struct nf_bridge_frag_data * data,int (* output)(struct net *,struct sock * sk,const struct nf_bridge_frag_data * data,struct sk_buff *))112 int br_ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
113 struct nf_bridge_frag_data *data,
114 int (*output)(struct net *, struct sock *sk,
115 const struct nf_bridge_frag_data *data,
116 struct sk_buff *))
117 {
118 int frag_max_size = BR_INPUT_SKB_CB(skb)->frag_max_size;
119 u8 tstamp_type = skb->tstamp_type;
120 ktime_t tstamp = skb->tstamp;
121 struct ip6_frag_state state;
122 u8 *prevhdr, nexthdr = 0;
123 unsigned int mtu, hlen, nexthdr_offset;
124 int hroom, err = 0;
125 __be32 frag_id;
126
127 err = ip6_find_1stfragopt(skb, &prevhdr);
128 if (err < 0)
129 goto blackhole;
130 hlen = err;
131 nexthdr = *prevhdr;
132 nexthdr_offset = prevhdr - skb_network_header(skb);
133
134 mtu = skb->dev->mtu;
135 if (frag_max_size > mtu ||
136 frag_max_size < IPV6_MIN_MTU)
137 goto blackhole;
138
139 mtu = frag_max_size;
140 if (mtu < hlen + sizeof(struct frag_hdr) + 8)
141 goto blackhole;
142 mtu -= hlen + sizeof(struct frag_hdr);
143
144 frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
145 &ipv6_hdr(skb)->saddr);
146
147 if (skb->ip_summed == CHECKSUM_PARTIAL &&
148 (err = skb_checksum_help(skb)))
149 goto blackhole;
150
151 prevhdr = skb_network_header(skb) + nexthdr_offset;
152 hroom = LL_RESERVED_SPACE(skb->dev);
153 if (skb_has_frag_list(skb)) {
154 unsigned int first_len = skb_pagelen(skb);
155 struct ip6_fraglist_iter iter;
156 struct sk_buff *frag2;
157
158 if (first_len - hlen > mtu)
159 goto blackhole;
160
161 if (skb_cloned(skb) ||
162 skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
163 goto slow_path;
164
165 skb_walk_frags(skb, frag2) {
166 if (frag2->len > mtu)
167 goto blackhole;
168
169 /* Partially cloned skb? */
170 if (skb_shared(frag2) ||
171 skb_headroom(frag2) < (hlen + hroom + sizeof(struct frag_hdr)))
172 goto slow_path;
173 }
174
175 err = ip6_fraglist_init(skb, hlen, prevhdr, nexthdr, frag_id,
176 &iter);
177 if (err < 0)
178 goto blackhole;
179
180 for (;;) {
181 /* Prepare header of the next frame,
182 * before previous one went down.
183 */
184 if (iter.frag)
185 ip6_fraglist_prepare(skb, &iter);
186
187 skb_set_delivery_time(skb, tstamp, tstamp_type);
188 err = output(net, sk, data, skb);
189 if (err || !iter.frag)
190 break;
191
192 skb = ip6_fraglist_next(&iter);
193 }
194
195 kfree(iter.tmp_hdr);
196 if (!err)
197 return 0;
198
199 kfree_skb_list(iter.frag);
200 return err;
201 }
202 slow_path:
203 /* This is a linearized skbuff, the original geometry is lost for us.
204 * This may also be a clone skbuff, we could preserve the geometry for
205 * the copies but probably not worth the effort.
206 */
207 ip6_frag_init(skb, hlen, mtu, skb->dev->needed_tailroom,
208 LL_RESERVED_SPACE(skb->dev), prevhdr, nexthdr, frag_id,
209 &state);
210
211 while (state.left > 0) {
212 struct sk_buff *skb2;
213
214 skb2 = ip6_frag_next(skb, &state);
215 if (IS_ERR(skb2)) {
216 err = PTR_ERR(skb2);
217 goto blackhole;
218 }
219
220 skb_set_delivery_time(skb2, tstamp, tstamp_type);
221 err = output(net, sk, data, skb2);
222 if (err)
223 goto blackhole;
224 }
225 consume_skb(skb);
226 return err;
227
228 blackhole:
229 kfree_skb(skb);
230 return 0;
231 }
232 EXPORT_SYMBOL_GPL(br_ip6_fragment);
233