xref: /openbmc/linux/net/xfrm/xfrm_output.c (revision ebd09753)
1 /*
2  * xfrm_output.c - Common IPsec encapsulation code.
3  *
4  * Copyright (c) 2007 Herbert Xu <herbert@gondor.apana.org.au>
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License
8  * as published by the Free Software Foundation; either version
9  * 2 of the License, or (at your option) any later version.
10  */
11 
12 #include <linux/errno.h>
13 #include <linux/module.h>
14 #include <linux/netdevice.h>
15 #include <linux/netfilter.h>
16 #include <linux/skbuff.h>
17 #include <linux/slab.h>
18 #include <linux/spinlock.h>
19 #include <net/dst.h>
20 #include <net/xfrm.h>
21 
22 static int xfrm_output2(struct net *net, struct sock *sk, struct sk_buff *skb);
23 
24 static int xfrm_skb_check_space(struct sk_buff *skb)
25 {
26 	struct dst_entry *dst = skb_dst(skb);
27 	int nhead = dst->header_len + LL_RESERVED_SPACE(dst->dev)
28 		- skb_headroom(skb);
29 	int ntail = dst->dev->needed_tailroom - skb_tailroom(skb);
30 
31 	if (nhead <= 0) {
32 		if (ntail <= 0)
33 			return 0;
34 		nhead = 0;
35 	} else if (ntail < 0)
36 		ntail = 0;
37 
38 	return pskb_expand_head(skb, nhead, ntail, GFP_ATOMIC);
39 }
40 
41 /* Children define the path of the packet through the
42  * Linux networking.  Thus, destinations are stackable.
43  */
44 
45 static struct dst_entry *skb_dst_pop(struct sk_buff *skb)
46 {
47 	struct dst_entry *child = dst_clone(xfrm_dst_child(skb_dst(skb)));
48 
49 	skb_dst_drop(skb);
50 	return child;
51 }
52 
53 static int xfrm_output_one(struct sk_buff *skb, int err)
54 {
55 	struct dst_entry *dst = skb_dst(skb);
56 	struct xfrm_state *x = dst->xfrm;
57 	struct net *net = xs_net(x);
58 
59 	if (err <= 0)
60 		goto resume;
61 
62 	do {
63 		err = xfrm_skb_check_space(skb);
64 		if (err) {
65 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
66 			goto error_nolock;
67 		}
68 
69 		skb->mark = xfrm_smark_get(skb->mark, x);
70 
71 		err = x->outer_mode->output(x, skb);
72 		if (err) {
73 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEMODEERROR);
74 			goto error_nolock;
75 		}
76 
77 		spin_lock_bh(&x->lock);
78 
79 		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
80 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEINVALID);
81 			err = -EINVAL;
82 			goto error;
83 		}
84 
85 		err = xfrm_state_check_expire(x);
86 		if (err) {
87 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEEXPIRED);
88 			goto error;
89 		}
90 
91 		err = x->repl->overflow(x, skb);
92 		if (err) {
93 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATESEQERROR);
94 			goto error;
95 		}
96 
97 		x->curlft.bytes += skb->len;
98 		x->curlft.packets++;
99 
100 		spin_unlock_bh(&x->lock);
101 
102 		skb_dst_force(skb);
103 		if (!skb_dst(skb)) {
104 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
105 			goto error_nolock;
106 		}
107 
108 		if (xfrm_offload(skb)) {
109 			x->type_offload->encap(x, skb);
110 		} else {
111 			/* Inner headers are invalid now. */
112 			skb->encapsulation = 0;
113 
114 			err = x->type->output(x, skb);
115 			if (err == -EINPROGRESS)
116 				goto out;
117 		}
118 
119 resume:
120 		if (err) {
121 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEPROTOERROR);
122 			goto error_nolock;
123 		}
124 
125 		dst = skb_dst_pop(skb);
126 		if (!dst) {
127 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
128 			err = -EHOSTUNREACH;
129 			goto error_nolock;
130 		}
131 		skb_dst_set(skb, dst);
132 		x = dst->xfrm;
133 	} while (x && !(x->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL));
134 
135 	return 0;
136 
137 error:
138 	spin_unlock_bh(&x->lock);
139 error_nolock:
140 	kfree_skb(skb);
141 out:
142 	return err;
143 }
144 
145 int xfrm_output_resume(struct sk_buff *skb, int err)
146 {
147 	struct net *net = xs_net(skb_dst(skb)->xfrm);
148 
149 	while (likely((err = xfrm_output_one(skb, err)) == 0)) {
150 		nf_reset(skb);
151 
152 		err = skb_dst(skb)->ops->local_out(net, skb->sk, skb);
153 		if (unlikely(err != 1))
154 			goto out;
155 
156 		if (!skb_dst(skb)->xfrm)
157 			return dst_output(net, skb->sk, skb);
158 
159 		err = nf_hook(skb_dst(skb)->ops->family,
160 			      NF_INET_POST_ROUTING, net, skb->sk, skb,
161 			      NULL, skb_dst(skb)->dev, xfrm_output2);
162 		if (unlikely(err != 1))
163 			goto out;
164 	}
165 
166 	if (err == -EINPROGRESS)
167 		err = 0;
168 
169 out:
170 	return err;
171 }
172 EXPORT_SYMBOL_GPL(xfrm_output_resume);
173 
174 static int xfrm_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
175 {
176 	return xfrm_output_resume(skb, 1);
177 }
178 
179 static int xfrm_output_gso(struct net *net, struct sock *sk, struct sk_buff *skb)
180 {
181 	struct sk_buff *segs;
182 
183 	BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_SGO_CB_OFFSET);
184 	BUILD_BUG_ON(sizeof(*IP6CB(skb)) > SKB_SGO_CB_OFFSET);
185 	segs = skb_gso_segment(skb, 0);
186 	kfree_skb(skb);
187 	if (IS_ERR(segs))
188 		return PTR_ERR(segs);
189 	if (segs == NULL)
190 		return -EINVAL;
191 
192 	do {
193 		struct sk_buff *nskb = segs->next;
194 		int err;
195 
196 		skb_mark_not_on_list(segs);
197 		err = xfrm_output2(net, sk, segs);
198 
199 		if (unlikely(err)) {
200 			kfree_skb_list(nskb);
201 			return err;
202 		}
203 
204 		segs = nskb;
205 	} while (segs);
206 
207 	return 0;
208 }
209 
210 int xfrm_output(struct sock *sk, struct sk_buff *skb)
211 {
212 	struct net *net = dev_net(skb_dst(skb)->dev);
213 	struct xfrm_state *x = skb_dst(skb)->xfrm;
214 	int err;
215 
216 	secpath_reset(skb);
217 
218 	if (xfrm_dev_offload_ok(skb, x)) {
219 		struct sec_path *sp;
220 
221 		sp = secpath_dup(skb->sp);
222 		if (!sp) {
223 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
224 			kfree_skb(skb);
225 			return -ENOMEM;
226 		}
227 		if (skb->sp)
228 			secpath_put(skb->sp);
229 		skb->sp = sp;
230 		skb->encapsulation = 1;
231 
232 		sp->olen++;
233 		sp->xvec[skb->sp->len++] = x;
234 		xfrm_state_hold(x);
235 
236 		if (skb_is_gso(skb)) {
237 			skb_shinfo(skb)->gso_type |= SKB_GSO_ESP;
238 
239 			return xfrm_output2(net, sk, skb);
240 		}
241 
242 		if (x->xso.dev && x->xso.dev->features & NETIF_F_HW_ESP_TX_CSUM)
243 			goto out;
244 	}
245 
246 	if (skb_is_gso(skb))
247 		return xfrm_output_gso(net, sk, skb);
248 
249 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
250 		err = skb_checksum_help(skb);
251 		if (err) {
252 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
253 			kfree_skb(skb);
254 			return err;
255 		}
256 	}
257 
258 out:
259 	return xfrm_output2(net, sk, skb);
260 }
261 EXPORT_SYMBOL_GPL(xfrm_output);
262 
263 int xfrm_inner_extract_output(struct xfrm_state *x, struct sk_buff *skb)
264 {
265 	struct xfrm_mode *inner_mode;
266 	if (x->sel.family == AF_UNSPEC)
267 		inner_mode = xfrm_ip2inner_mode(x,
268 				xfrm_af2proto(skb_dst(skb)->ops->family));
269 	else
270 		inner_mode = x->inner_mode;
271 
272 	if (inner_mode == NULL)
273 		return -EAFNOSUPPORT;
274 	return inner_mode->afinfo->extract_output(x, skb);
275 }
276 EXPORT_SYMBOL_GPL(xfrm_inner_extract_output);
277 
278 void xfrm_local_error(struct sk_buff *skb, int mtu)
279 {
280 	unsigned int proto;
281 	struct xfrm_state_afinfo *afinfo;
282 
283 	if (skb->protocol == htons(ETH_P_IP))
284 		proto = AF_INET;
285 	else if (skb->protocol == htons(ETH_P_IPV6))
286 		proto = AF_INET6;
287 	else
288 		return;
289 
290 	afinfo = xfrm_state_get_afinfo(proto);
291 	if (afinfo) {
292 		afinfo->local_error(skb, mtu);
293 		rcu_read_unlock();
294 	}
295 }
296 EXPORT_SYMBOL_GPL(xfrm_local_error);
297