xref: /openbmc/linux/net/xfrm/xfrm_device.c (revision 31eeb6b0)
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 	netdev_tracker_alloc(dev, &xso->dev_tracker, GFP_ATOMIC);
263 	xso->real_dev = dev;
264 	xso->num_exthdrs = 1;
265 	xso->flags = xuo->flags;
266 
267 	err = dev->xfrmdev_ops->xdo_dev_state_add(x);
268 	if (err) {
269 		xso->num_exthdrs = 0;
270 		xso->flags = 0;
271 		xso->dev = NULL;
272 		xso->real_dev = NULL;
273 		dev_put_track(dev, &xso->dev_tracker);
274 
275 		if (err != -EOPNOTSUPP)
276 			return err;
277 	}
278 
279 	return 0;
280 }
281 EXPORT_SYMBOL_GPL(xfrm_dev_state_add);
282 
283 bool xfrm_dev_offload_ok(struct sk_buff *skb, struct xfrm_state *x)
284 {
285 	int mtu;
286 	struct dst_entry *dst = skb_dst(skb);
287 	struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
288 	struct net_device *dev = x->xso.dev;
289 
290 	if (!x->type_offload || x->encap)
291 		return false;
292 
293 	if ((!dev || (dev == xfrm_dst_path(dst)->dev)) &&
294 	    (!xdst->child->xfrm)) {
295 		mtu = xfrm_state_mtu(x, xdst->child_mtu_cached);
296 		if (skb->len <= mtu)
297 			goto ok;
298 
299 		if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
300 			goto ok;
301 	}
302 
303 	return false;
304 
305 ok:
306 	if (dev && dev->xfrmdev_ops && dev->xfrmdev_ops->xdo_dev_offload_ok)
307 		return x->xso.dev->xfrmdev_ops->xdo_dev_offload_ok(skb, x);
308 
309 	return true;
310 }
311 EXPORT_SYMBOL_GPL(xfrm_dev_offload_ok);
312 
313 void xfrm_dev_resume(struct sk_buff *skb)
314 {
315 	struct net_device *dev = skb->dev;
316 	int ret = NETDEV_TX_BUSY;
317 	struct netdev_queue *txq;
318 	struct softnet_data *sd;
319 	unsigned long flags;
320 
321 	rcu_read_lock();
322 	txq = netdev_core_pick_tx(dev, skb, NULL);
323 
324 	HARD_TX_LOCK(dev, txq, smp_processor_id());
325 	if (!netif_xmit_frozen_or_stopped(txq))
326 		skb = dev_hard_start_xmit(skb, dev, txq, &ret);
327 	HARD_TX_UNLOCK(dev, txq);
328 
329 	if (!dev_xmit_complete(ret)) {
330 		local_irq_save(flags);
331 		sd = this_cpu_ptr(&softnet_data);
332 		skb_queue_tail(&sd->xfrm_backlog, skb);
333 		raise_softirq_irqoff(NET_TX_SOFTIRQ);
334 		local_irq_restore(flags);
335 	}
336 	rcu_read_unlock();
337 }
338 EXPORT_SYMBOL_GPL(xfrm_dev_resume);
339 
340 void xfrm_dev_backlog(struct softnet_data *sd)
341 {
342 	struct sk_buff_head *xfrm_backlog = &sd->xfrm_backlog;
343 	struct sk_buff_head list;
344 	struct sk_buff *skb;
345 
346 	if (skb_queue_empty(xfrm_backlog))
347 		return;
348 
349 	__skb_queue_head_init(&list);
350 
351 	spin_lock(&xfrm_backlog->lock);
352 	skb_queue_splice_init(xfrm_backlog, &list);
353 	spin_unlock(&xfrm_backlog->lock);
354 
355 	while (!skb_queue_empty(&list)) {
356 		skb = __skb_dequeue(&list);
357 		xfrm_dev_resume(skb);
358 	}
359 
360 }
361 #endif
362 
363 static int xfrm_api_check(struct net_device *dev)
364 {
365 #ifdef CONFIG_XFRM_OFFLOAD
366 	if ((dev->features & NETIF_F_HW_ESP_TX_CSUM) &&
367 	    !(dev->features & NETIF_F_HW_ESP))
368 		return NOTIFY_BAD;
369 
370 	if ((dev->features & NETIF_F_HW_ESP) &&
371 	    (!(dev->xfrmdev_ops &&
372 	       dev->xfrmdev_ops->xdo_dev_state_add &&
373 	       dev->xfrmdev_ops->xdo_dev_state_delete)))
374 		return NOTIFY_BAD;
375 #else
376 	if (dev->features & (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM))
377 		return NOTIFY_BAD;
378 #endif
379 
380 	return NOTIFY_DONE;
381 }
382 
383 static int xfrm_dev_register(struct net_device *dev)
384 {
385 	return xfrm_api_check(dev);
386 }
387 
388 static int xfrm_dev_feat_change(struct net_device *dev)
389 {
390 	return xfrm_api_check(dev);
391 }
392 
393 static int xfrm_dev_down(struct net_device *dev)
394 {
395 	if (dev->features & NETIF_F_HW_ESP)
396 		xfrm_dev_state_flush(dev_net(dev), dev, true);
397 
398 	return NOTIFY_DONE;
399 }
400 
401 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
402 {
403 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
404 
405 	switch (event) {
406 	case NETDEV_REGISTER:
407 		return xfrm_dev_register(dev);
408 
409 	case NETDEV_FEAT_CHANGE:
410 		return xfrm_dev_feat_change(dev);
411 
412 	case NETDEV_DOWN:
413 	case NETDEV_UNREGISTER:
414 		return xfrm_dev_down(dev);
415 	}
416 	return NOTIFY_DONE;
417 }
418 
419 static struct notifier_block xfrm_dev_notifier = {
420 	.notifier_call	= xfrm_dev_event,
421 };
422 
423 void __init xfrm_dev_init(void)
424 {
425 	register_netdevice_notifier(&xfrm_dev_notifier);
426 }
427