xref: /linux/net/xfrm/xfrm_device.c (revision d4b996f9ef1fe83d9ce9ad5c1ca0bd8231638ce5)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * xfrm_device.c - IPsec device offloading code.
4  *
5  * Copyright (c) 2015 secunet Security Networks AG
6  *
7  * Author:
8  * Steffen Klassert <steffen.klassert@secunet.com>
9  */
10 
11 #include <linux/errno.h>
12 #include <linux/module.h>
13 #include <linux/netdevice.h>
14 #include <linux/skbuff.h>
15 #include <linux/slab.h>
16 #include <linux/spinlock.h>
17 #include <net/dst.h>
18 #include <net/xfrm.h>
19 #include <linux/notifier.h>
20 
21 #ifdef CONFIG_XFRM_OFFLOAD
22 static void __xfrm_transport_prep(struct xfrm_state *x, struct sk_buff *skb,
23 				  unsigned int hsize)
24 {
25 	struct xfrm_offload *xo = xfrm_offload(skb);
26 
27 	skb_reset_mac_len(skb);
28 	if (xo->flags & XFRM_GSO_SEGMENT)
29 		skb->transport_header -= x->props.header_len;
30 
31 	pskb_pull(skb, skb_transport_offset(skb) + x->props.header_len);
32 }
33 
34 static void __xfrm_mode_tunnel_prep(struct xfrm_state *x, struct sk_buff *skb,
35 				    unsigned int hsize)
36 
37 {
38 	struct xfrm_offload *xo = xfrm_offload(skb);
39 
40 	if (xo->flags & XFRM_GSO_SEGMENT)
41 		skb->transport_header = skb->network_header + hsize;
42 
43 	skb_reset_mac_len(skb);
44 	pskb_pull(skb, skb->mac_len + x->props.header_len);
45 }
46 
47 static void __xfrm_mode_beet_prep(struct xfrm_state *x, struct sk_buff *skb,
48 				  unsigned int hsize)
49 {
50 	struct xfrm_offload *xo = xfrm_offload(skb);
51 	int phlen = 0;
52 
53 	if (xo->flags & XFRM_GSO_SEGMENT)
54 		skb->transport_header = skb->network_header + hsize;
55 
56 	skb_reset_mac_len(skb);
57 	if (x->sel.family != AF_INET6) {
58 		phlen = IPV4_BEET_PHMAXLEN;
59 		if (x->outer_mode.family == AF_INET6)
60 			phlen += sizeof(struct ipv6hdr) - sizeof(struct iphdr);
61 	}
62 
63 	pskb_pull(skb, skb->mac_len + hsize + (x->props.header_len - phlen));
64 }
65 
66 /* Adjust pointers into the packet when IPsec is done at layer2 */
67 static void xfrm_outer_mode_prep(struct xfrm_state *x, struct sk_buff *skb)
68 {
69 	switch (x->outer_mode.encap) {
70 	case XFRM_MODE_TUNNEL:
71 		if (x->outer_mode.family == AF_INET)
72 			return __xfrm_mode_tunnel_prep(x, skb,
73 						       sizeof(struct iphdr));
74 		if (x->outer_mode.family == AF_INET6)
75 			return __xfrm_mode_tunnel_prep(x, skb,
76 						       sizeof(struct ipv6hdr));
77 		break;
78 	case XFRM_MODE_TRANSPORT:
79 		if (x->outer_mode.family == AF_INET)
80 			return __xfrm_transport_prep(x, skb,
81 						     sizeof(struct iphdr));
82 		if (x->outer_mode.family == AF_INET6)
83 			return __xfrm_transport_prep(x, skb,
84 						     sizeof(struct ipv6hdr));
85 		break;
86 	case XFRM_MODE_BEET:
87 		if (x->outer_mode.family == AF_INET)
88 			return __xfrm_mode_beet_prep(x, skb,
89 						     sizeof(struct iphdr));
90 		if (x->outer_mode.family == AF_INET6)
91 			return __xfrm_mode_beet_prep(x, skb,
92 						     sizeof(struct ipv6hdr));
93 		break;
94 	case XFRM_MODE_ROUTEOPTIMIZATION:
95 	case XFRM_MODE_IN_TRIGGER:
96 		break;
97 	}
98 }
99 
100 struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t features, bool *again)
101 {
102 	int err;
103 	unsigned long flags;
104 	struct xfrm_state *x;
105 	struct softnet_data *sd;
106 	struct sk_buff *skb2, *nskb, *pskb = NULL;
107 	netdev_features_t esp_features = features;
108 	struct xfrm_offload *xo = xfrm_offload(skb);
109 	struct net_device *dev = skb->dev;
110 	struct sec_path *sp;
111 
112 	if (!xo || (xo->flags & XFRM_XMIT))
113 		return skb;
114 
115 	if (!(features & NETIF_F_HW_ESP))
116 		esp_features = features & ~(NETIF_F_SG | NETIF_F_CSUM_MASK);
117 
118 	sp = skb_sec_path(skb);
119 	x = sp->xvec[sp->len - 1];
120 	if (xo->flags & XFRM_GRO || x->xso.flags & XFRM_OFFLOAD_INBOUND)
121 		return skb;
122 
123 	/* This skb was already validated on the upper/virtual dev */
124 	if ((x->xso.dev != dev) && (x->xso.real_dev == dev))
125 		return skb;
126 
127 	local_irq_save(flags);
128 	sd = this_cpu_ptr(&softnet_data);
129 	err = !skb_queue_empty(&sd->xfrm_backlog);
130 	local_irq_restore(flags);
131 
132 	if (err) {
133 		*again = true;
134 		return skb;
135 	}
136 
137 	if (skb_is_gso(skb) && unlikely(x->xso.dev != dev)) {
138 		struct sk_buff *segs;
139 
140 		/* Packet got rerouted, fixup features and segment it. */
141 		esp_features = esp_features & ~(NETIF_F_HW_ESP | NETIF_F_GSO_ESP);
142 
143 		segs = skb_gso_segment(skb, esp_features);
144 		if (IS_ERR(segs)) {
145 			kfree_skb(skb);
146 			atomic_long_inc(&dev->tx_dropped);
147 			return NULL;
148 		} else {
149 			consume_skb(skb);
150 			skb = segs;
151 		}
152 	}
153 
154 	if (!skb->next) {
155 		esp_features |= skb->dev->gso_partial_features;
156 		xfrm_outer_mode_prep(x, skb);
157 
158 		xo->flags |= XFRM_DEV_RESUME;
159 
160 		err = x->type_offload->xmit(x, skb, esp_features);
161 		if (err) {
162 			if (err == -EINPROGRESS)
163 				return NULL;
164 
165 			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
166 			kfree_skb(skb);
167 			return NULL;
168 		}
169 
170 		skb_push(skb, skb->data - skb_mac_header(skb));
171 
172 		return skb;
173 	}
174 
175 	skb_list_walk_safe(skb, skb2, nskb) {
176 		esp_features |= skb->dev->gso_partial_features;
177 		skb_mark_not_on_list(skb2);
178 
179 		xo = xfrm_offload(skb2);
180 		xo->flags |= XFRM_DEV_RESUME;
181 
182 		xfrm_outer_mode_prep(x, skb2);
183 
184 		err = x->type_offload->xmit(x, skb2, esp_features);
185 		if (!err) {
186 			skb2->next = nskb;
187 		} else if (err != -EINPROGRESS) {
188 			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
189 			skb2->next = nskb;
190 			kfree_skb_list(skb2);
191 			return NULL;
192 		} else {
193 			if (skb == skb2)
194 				skb = nskb;
195 			else
196 				pskb->next = nskb;
197 
198 			continue;
199 		}
200 
201 		skb_push(skb2, skb2->data - skb_mac_header(skb2));
202 		pskb = skb2;
203 	}
204 
205 	return skb;
206 }
207 EXPORT_SYMBOL_GPL(validate_xmit_xfrm);
208 
209 int xfrm_dev_state_add(struct net *net, struct xfrm_state *x,
210 		       struct xfrm_user_offload *xuo)
211 {
212 	int err;
213 	struct dst_entry *dst;
214 	struct net_device *dev;
215 	struct xfrm_state_offload *xso = &x->xso;
216 	xfrm_address_t *saddr;
217 	xfrm_address_t *daddr;
218 
219 	if (!x->type_offload)
220 		return -EINVAL;
221 
222 	/* We don't yet support UDP encapsulation and TFC padding. */
223 	if (x->encap || x->tfcpad)
224 		return -EINVAL;
225 
226 	dev = dev_get_by_index(net, xuo->ifindex);
227 	if (!dev) {
228 		if (!(xuo->flags & XFRM_OFFLOAD_INBOUND)) {
229 			saddr = &x->props.saddr;
230 			daddr = &x->id.daddr;
231 		} else {
232 			saddr = &x->id.daddr;
233 			daddr = &x->props.saddr;
234 		}
235 
236 		dst = __xfrm_dst_lookup(net, 0, 0, saddr, daddr,
237 					x->props.family,
238 					xfrm_smark_get(0, x));
239 		if (IS_ERR(dst))
240 			return 0;
241 
242 		dev = dst->dev;
243 
244 		dev_hold(dev);
245 		dst_release(dst);
246 	}
247 
248 	if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_state_add) {
249 		xso->dev = NULL;
250 		dev_put(dev);
251 		return 0;
252 	}
253 
254 	if (x->props.flags & XFRM_STATE_ESN &&
255 	    !dev->xfrmdev_ops->xdo_dev_state_advance_esn) {
256 		xso->dev = NULL;
257 		dev_put(dev);
258 		return -EINVAL;
259 	}
260 
261 	xso->dev = dev;
262 	xso->real_dev = dev;
263 	xso->num_exthdrs = 1;
264 	xso->flags = xuo->flags;
265 
266 	err = dev->xfrmdev_ops->xdo_dev_state_add(x);
267 	if (err) {
268 		xso->num_exthdrs = 0;
269 		xso->flags = 0;
270 		xso->dev = NULL;
271 		xso->real_dev = NULL;
272 		dev_put(dev);
273 
274 		if (err != -EOPNOTSUPP)
275 			return err;
276 	}
277 
278 	return 0;
279 }
280 EXPORT_SYMBOL_GPL(xfrm_dev_state_add);
281 
282 bool xfrm_dev_offload_ok(struct sk_buff *skb, struct xfrm_state *x)
283 {
284 	int mtu;
285 	struct dst_entry *dst = skb_dst(skb);
286 	struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
287 	struct net_device *dev = x->xso.dev;
288 
289 	if (!x->type_offload || x->encap)
290 		return false;
291 
292 	if ((!dev || (dev == xfrm_dst_path(dst)->dev)) &&
293 	    (!xdst->child->xfrm)) {
294 		mtu = xfrm_state_mtu(x, xdst->child_mtu_cached);
295 		if (skb->len <= mtu)
296 			goto ok;
297 
298 		if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
299 			goto ok;
300 	}
301 
302 	return false;
303 
304 ok:
305 	if (dev && dev->xfrmdev_ops && dev->xfrmdev_ops->xdo_dev_offload_ok)
306 		return x->xso.dev->xfrmdev_ops->xdo_dev_offload_ok(skb, x);
307 
308 	return true;
309 }
310 EXPORT_SYMBOL_GPL(xfrm_dev_offload_ok);
311 
312 void xfrm_dev_resume(struct sk_buff *skb)
313 {
314 	struct net_device *dev = skb->dev;
315 	int ret = NETDEV_TX_BUSY;
316 	struct netdev_queue *txq;
317 	struct softnet_data *sd;
318 	unsigned long flags;
319 
320 	rcu_read_lock();
321 	txq = netdev_core_pick_tx(dev, skb, NULL);
322 
323 	HARD_TX_LOCK(dev, txq, smp_processor_id());
324 	if (!netif_xmit_frozen_or_stopped(txq))
325 		skb = dev_hard_start_xmit(skb, dev, txq, &ret);
326 	HARD_TX_UNLOCK(dev, txq);
327 
328 	if (!dev_xmit_complete(ret)) {
329 		local_irq_save(flags);
330 		sd = this_cpu_ptr(&softnet_data);
331 		skb_queue_tail(&sd->xfrm_backlog, skb);
332 		raise_softirq_irqoff(NET_TX_SOFTIRQ);
333 		local_irq_restore(flags);
334 	}
335 	rcu_read_unlock();
336 }
337 EXPORT_SYMBOL_GPL(xfrm_dev_resume);
338 
339 void xfrm_dev_backlog(struct softnet_data *sd)
340 {
341 	struct sk_buff_head *xfrm_backlog = &sd->xfrm_backlog;
342 	struct sk_buff_head list;
343 	struct sk_buff *skb;
344 
345 	if (skb_queue_empty(xfrm_backlog))
346 		return;
347 
348 	__skb_queue_head_init(&list);
349 
350 	spin_lock(&xfrm_backlog->lock);
351 	skb_queue_splice_init(xfrm_backlog, &list);
352 	spin_unlock(&xfrm_backlog->lock);
353 
354 	while (!skb_queue_empty(&list)) {
355 		skb = __skb_dequeue(&list);
356 		xfrm_dev_resume(skb);
357 	}
358 
359 }
360 #endif
361 
362 static int xfrm_api_check(struct net_device *dev)
363 {
364 #ifdef CONFIG_XFRM_OFFLOAD
365 	if ((dev->features & NETIF_F_HW_ESP_TX_CSUM) &&
366 	    !(dev->features & NETIF_F_HW_ESP))
367 		return NOTIFY_BAD;
368 
369 	if ((dev->features & NETIF_F_HW_ESP) &&
370 	    (!(dev->xfrmdev_ops &&
371 	       dev->xfrmdev_ops->xdo_dev_state_add &&
372 	       dev->xfrmdev_ops->xdo_dev_state_delete)))
373 		return NOTIFY_BAD;
374 #else
375 	if (dev->features & (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM))
376 		return NOTIFY_BAD;
377 #endif
378 
379 	return NOTIFY_DONE;
380 }
381 
382 static int xfrm_dev_register(struct net_device *dev)
383 {
384 	return xfrm_api_check(dev);
385 }
386 
387 static int xfrm_dev_feat_change(struct net_device *dev)
388 {
389 	return xfrm_api_check(dev);
390 }
391 
392 static int xfrm_dev_down(struct net_device *dev)
393 {
394 	if (dev->features & NETIF_F_HW_ESP)
395 		xfrm_dev_state_flush(dev_net(dev), dev, true);
396 
397 	return NOTIFY_DONE;
398 }
399 
400 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
401 {
402 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
403 
404 	switch (event) {
405 	case NETDEV_REGISTER:
406 		return xfrm_dev_register(dev);
407 
408 	case NETDEV_FEAT_CHANGE:
409 		return xfrm_dev_feat_change(dev);
410 
411 	case NETDEV_DOWN:
412 	case NETDEV_UNREGISTER:
413 		return xfrm_dev_down(dev);
414 	}
415 	return NOTIFY_DONE;
416 }
417 
418 static struct notifier_block xfrm_dev_notifier = {
419 	.notifier_call	= xfrm_dev_event,
420 };
421 
422 void __init xfrm_dev_init(void)
423 {
424 	register_netdevice_notifier(&xfrm_dev_notifier);
425 }
426