xref: /openbmc/linux/net/xfrm/xfrm_output.c (revision c67e8ec0)
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 			err = -EHOSTUNREACH;
106 			goto error_nolock;
107 		}
108 
109 		if (xfrm_offload(skb)) {
110 			x->type_offload->encap(x, skb);
111 		} else {
112 			/* Inner headers are invalid now. */
113 			skb->encapsulation = 0;
114 
115 			err = x->type->output(x, skb);
116 			if (err == -EINPROGRESS)
117 				goto out;
118 		}
119 
120 resume:
121 		if (err) {
122 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEPROTOERROR);
123 			goto error_nolock;
124 		}
125 
126 		dst = skb_dst_pop(skb);
127 		if (!dst) {
128 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
129 			err = -EHOSTUNREACH;
130 			goto error_nolock;
131 		}
132 		skb_dst_set(skb, dst);
133 		x = dst->xfrm;
134 	} while (x && !(x->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL));
135 
136 	return 0;
137 
138 error:
139 	spin_unlock_bh(&x->lock);
140 error_nolock:
141 	kfree_skb(skb);
142 out:
143 	return err;
144 }
145 
146 int xfrm_output_resume(struct sk_buff *skb, int err)
147 {
148 	struct net *net = xs_net(skb_dst(skb)->xfrm);
149 
150 	while (likely((err = xfrm_output_one(skb, err)) == 0)) {
151 		nf_reset(skb);
152 
153 		err = skb_dst(skb)->ops->local_out(net, skb->sk, skb);
154 		if (unlikely(err != 1))
155 			goto out;
156 
157 		if (!skb_dst(skb)->xfrm)
158 			return dst_output(net, skb->sk, skb);
159 
160 		err = nf_hook(skb_dst(skb)->ops->family,
161 			      NF_INET_POST_ROUTING, net, skb->sk, skb,
162 			      NULL, skb_dst(skb)->dev, xfrm_output2);
163 		if (unlikely(err != 1))
164 			goto out;
165 	}
166 
167 	if (err == -EINPROGRESS)
168 		err = 0;
169 
170 out:
171 	return err;
172 }
173 EXPORT_SYMBOL_GPL(xfrm_output_resume);
174 
175 static int xfrm_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
176 {
177 	return xfrm_output_resume(skb, 1);
178 }
179 
180 static int xfrm_output_gso(struct net *net, struct sock *sk, struct sk_buff *skb)
181 {
182 	struct sk_buff *segs;
183 
184 	BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_SGO_CB_OFFSET);
185 	BUILD_BUG_ON(sizeof(*IP6CB(skb)) > SKB_SGO_CB_OFFSET);
186 	segs = skb_gso_segment(skb, 0);
187 	kfree_skb(skb);
188 	if (IS_ERR(segs))
189 		return PTR_ERR(segs);
190 	if (segs == NULL)
191 		return -EINVAL;
192 
193 	do {
194 		struct sk_buff *nskb = segs->next;
195 		int err;
196 
197 		skb_mark_not_on_list(segs);
198 		err = xfrm_output2(net, sk, segs);
199 
200 		if (unlikely(err)) {
201 			kfree_skb_list(nskb);
202 			return err;
203 		}
204 
205 		segs = nskb;
206 	} while (segs);
207 
208 	return 0;
209 }
210 
211 int xfrm_output(struct sock *sk, struct sk_buff *skb)
212 {
213 	struct net *net = dev_net(skb_dst(skb)->dev);
214 	struct xfrm_state *x = skb_dst(skb)->xfrm;
215 	int err;
216 
217 	secpath_reset(skb);
218 
219 	if (xfrm_dev_offload_ok(skb, x)) {
220 		struct sec_path *sp;
221 
222 		sp = secpath_set(skb);
223 		if (!sp) {
224 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
225 			kfree_skb(skb);
226 			return -ENOMEM;
227 		}
228 		skb->encapsulation = 1;
229 
230 		sp->olen++;
231 		sp->xvec[sp->len++] = x;
232 		xfrm_state_hold(x);
233 
234 		if (skb_is_gso(skb)) {
235 			skb_shinfo(skb)->gso_type |= SKB_GSO_ESP;
236 
237 			return xfrm_output2(net, sk, skb);
238 		}
239 
240 		if (x->xso.dev && x->xso.dev->features & NETIF_F_HW_ESP_TX_CSUM)
241 			goto out;
242 	}
243 
244 	if (skb_is_gso(skb))
245 		return xfrm_output_gso(net, sk, skb);
246 
247 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
248 		err = skb_checksum_help(skb);
249 		if (err) {
250 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
251 			kfree_skb(skb);
252 			return err;
253 		}
254 	}
255 
256 out:
257 	return xfrm_output2(net, sk, skb);
258 }
259 EXPORT_SYMBOL_GPL(xfrm_output);
260 
261 int xfrm_inner_extract_output(struct xfrm_state *x, struct sk_buff *skb)
262 {
263 	struct xfrm_mode *inner_mode;
264 	if (x->sel.family == AF_UNSPEC)
265 		inner_mode = xfrm_ip2inner_mode(x,
266 				xfrm_af2proto(skb_dst(skb)->ops->family));
267 	else
268 		inner_mode = x->inner_mode;
269 
270 	if (inner_mode == NULL)
271 		return -EAFNOSUPPORT;
272 	return inner_mode->afinfo->extract_output(x, skb);
273 }
274 EXPORT_SYMBOL_GPL(xfrm_inner_extract_output);
275 
276 void xfrm_local_error(struct sk_buff *skb, int mtu)
277 {
278 	unsigned int proto;
279 	struct xfrm_state_afinfo *afinfo;
280 
281 	if (skb->protocol == htons(ETH_P_IP))
282 		proto = AF_INET;
283 	else if (skb->protocol == htons(ETH_P_IPV6))
284 		proto = AF_INET6;
285 	else
286 		return;
287 
288 	afinfo = xfrm_state_get_afinfo(proto);
289 	if (afinfo) {
290 		afinfo->local_error(skb, mtu);
291 		rcu_read_unlock();
292 	}
293 }
294 EXPORT_SYMBOL_GPL(xfrm_local_error);
295