xref: /openbmc/linux/net/xfrm/xfrm_output.c (revision d8bcaabe)
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(skb_dst(skb)->child);
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 		if (x->props.output_mark)
70 			skb->mark = x->props.output_mark;
71 
72 		err = x->outer_mode->output(x, skb);
73 		if (err) {
74 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEMODEERROR);
75 			goto error_nolock;
76 		}
77 
78 		spin_lock_bh(&x->lock);
79 
80 		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
81 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEINVALID);
82 			err = -EINVAL;
83 			goto error;
84 		}
85 
86 		err = xfrm_state_check_expire(x);
87 		if (err) {
88 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEEXPIRED);
89 			goto error;
90 		}
91 
92 		err = x->repl->overflow(x, skb);
93 		if (err) {
94 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATESEQERROR);
95 			goto error;
96 		}
97 
98 		x->curlft.bytes += skb->len;
99 		x->curlft.packets++;
100 
101 		spin_unlock_bh(&x->lock);
102 
103 		skb_dst_force(skb);
104 
105 		if (xfrm_offload(skb)) {
106 			x->type_offload->encap(x, skb);
107 		} else {
108 			err = x->type->output(x, skb);
109 			if (err == -EINPROGRESS)
110 				goto out;
111 		}
112 
113 resume:
114 		if (err) {
115 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEPROTOERROR);
116 			goto error_nolock;
117 		}
118 
119 		dst = skb_dst_pop(skb);
120 		if (!dst) {
121 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
122 			err = -EHOSTUNREACH;
123 			goto error_nolock;
124 		}
125 		skb_dst_set(skb, dst);
126 		x = dst->xfrm;
127 	} while (x && !(x->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL));
128 
129 	return 0;
130 
131 error:
132 	spin_unlock_bh(&x->lock);
133 error_nolock:
134 	kfree_skb(skb);
135 out:
136 	return err;
137 }
138 
139 int xfrm_output_resume(struct sk_buff *skb, int err)
140 {
141 	struct net *net = xs_net(skb_dst(skb)->xfrm);
142 
143 	while (likely((err = xfrm_output_one(skb, err)) == 0)) {
144 		nf_reset(skb);
145 
146 		err = skb_dst(skb)->ops->local_out(net, skb->sk, skb);
147 		if (unlikely(err != 1))
148 			goto out;
149 
150 		if (!skb_dst(skb)->xfrm)
151 			return dst_output(net, skb->sk, skb);
152 
153 		err = nf_hook(skb_dst(skb)->ops->family,
154 			      NF_INET_POST_ROUTING, net, skb->sk, skb,
155 			      NULL, skb_dst(skb)->dev, xfrm_output2);
156 		if (unlikely(err != 1))
157 			goto out;
158 	}
159 
160 	if (err == -EINPROGRESS)
161 		err = 0;
162 
163 out:
164 	return err;
165 }
166 EXPORT_SYMBOL_GPL(xfrm_output_resume);
167 
168 static int xfrm_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
169 {
170 	return xfrm_output_resume(skb, 1);
171 }
172 
173 static int xfrm_output_gso(struct net *net, struct sock *sk, struct sk_buff *skb)
174 {
175 	struct sk_buff *segs;
176 
177 	BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_SGO_CB_OFFSET);
178 	BUILD_BUG_ON(sizeof(*IP6CB(skb)) > SKB_SGO_CB_OFFSET);
179 	segs = skb_gso_segment(skb, 0);
180 	kfree_skb(skb);
181 	if (IS_ERR(segs))
182 		return PTR_ERR(segs);
183 	if (segs == NULL)
184 		return -EINVAL;
185 
186 	do {
187 		struct sk_buff *nskb = segs->next;
188 		int err;
189 
190 		segs->next = NULL;
191 		err = xfrm_output2(net, sk, segs);
192 
193 		if (unlikely(err)) {
194 			kfree_skb_list(nskb);
195 			return err;
196 		}
197 
198 		segs = nskb;
199 	} while (segs);
200 
201 	return 0;
202 }
203 
204 int xfrm_output(struct sock *sk, struct sk_buff *skb)
205 {
206 	struct net *net = dev_net(skb_dst(skb)->dev);
207 	struct xfrm_state *x = skb_dst(skb)->xfrm;
208 	int err;
209 
210 	secpath_reset(skb);
211 	skb->encapsulation = 0;
212 
213 	if (xfrm_dev_offload_ok(skb, x)) {
214 		struct sec_path *sp;
215 
216 		sp = secpath_dup(skb->sp);
217 		if (!sp) {
218 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
219 			kfree_skb(skb);
220 			return -ENOMEM;
221 		}
222 		if (skb->sp)
223 			secpath_put(skb->sp);
224 		skb->sp = sp;
225 		skb->encapsulation = 1;
226 
227 		sp->olen++;
228 		sp->xvec[skb->sp->len++] = x;
229 		xfrm_state_hold(x);
230 
231 		if (skb_is_gso(skb)) {
232 			skb_shinfo(skb)->gso_type |= SKB_GSO_ESP;
233 
234 			return xfrm_output2(net, sk, skb);
235 		}
236 
237 		if (x->xso.dev && x->xso.dev->features & NETIF_F_HW_ESP_TX_CSUM)
238 			goto out;
239 	}
240 
241 	if (skb_is_gso(skb))
242 		return xfrm_output_gso(net, sk, skb);
243 
244 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
245 		err = skb_checksum_help(skb);
246 		if (err) {
247 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
248 			kfree_skb(skb);
249 			return err;
250 		}
251 	}
252 
253 out:
254 	return xfrm_output2(net, sk, skb);
255 }
256 EXPORT_SYMBOL_GPL(xfrm_output);
257 
258 int xfrm_inner_extract_output(struct xfrm_state *x, struct sk_buff *skb)
259 {
260 	struct xfrm_mode *inner_mode;
261 	if (x->sel.family == AF_UNSPEC)
262 		inner_mode = xfrm_ip2inner_mode(x,
263 				xfrm_af2proto(skb_dst(skb)->ops->family));
264 	else
265 		inner_mode = x->inner_mode;
266 
267 	if (inner_mode == NULL)
268 		return -EAFNOSUPPORT;
269 	return inner_mode->afinfo->extract_output(x, skb);
270 }
271 EXPORT_SYMBOL_GPL(xfrm_inner_extract_output);
272 
273 void xfrm_local_error(struct sk_buff *skb, int mtu)
274 {
275 	unsigned int proto;
276 	struct xfrm_state_afinfo *afinfo;
277 
278 	if (skb->protocol == htons(ETH_P_IP))
279 		proto = AF_INET;
280 	else if (skb->protocol == htons(ETH_P_IPV6))
281 		proto = AF_INET6;
282 	else
283 		return;
284 
285 	afinfo = xfrm_state_get_afinfo(proto);
286 	if (afinfo)
287 		afinfo->local_error(skb, mtu);
288 	rcu_read_unlock();
289 }
290 EXPORT_SYMBOL_GPL(xfrm_local_error);
291