xref: /openbmc/linux/net/xfrm/xfrm_device.c (revision f4356947)
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 static inline bool xmit_xfrm_check_overflow(struct sk_buff *skb)
101 {
102 	struct xfrm_offload *xo = xfrm_offload(skb);
103 	__u32 seq = xo->seq.low;
104 
105 	seq += skb_shinfo(skb)->gso_segs;
106 	if (unlikely(seq < xo->seq.low))
107 		return true;
108 
109 	return false;
110 }
111 
112 struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t features, bool *again)
113 {
114 	int err;
115 	unsigned long flags;
116 	struct xfrm_state *x;
117 	struct softnet_data *sd;
118 	struct sk_buff *skb2, *nskb, *pskb = NULL;
119 	netdev_features_t esp_features = features;
120 	struct xfrm_offload *xo = xfrm_offload(skb);
121 	struct net_device *dev = skb->dev;
122 	struct sec_path *sp;
123 
124 	if (!xo || (xo->flags & XFRM_XMIT))
125 		return skb;
126 
127 	if (!(features & NETIF_F_HW_ESP))
128 		esp_features = features & ~(NETIF_F_SG | NETIF_F_CSUM_MASK);
129 
130 	sp = skb_sec_path(skb);
131 	x = sp->xvec[sp->len - 1];
132 	if (xo->flags & XFRM_GRO || x->xso.dir == XFRM_DEV_OFFLOAD_IN)
133 		return skb;
134 
135 	/* The packet was sent to HW IPsec packet offload engine,
136 	 * but to wrong device. Drop the packet, so it won't skip
137 	 * XFRM stack.
138 	 */
139 	if (x->xso.type == XFRM_DEV_OFFLOAD_PACKET && x->xso.dev != dev) {
140 		kfree_skb(skb);
141 		dev_core_stats_tx_dropped_inc(dev);
142 		return NULL;
143 	}
144 
145 	/* This skb was already validated on the upper/virtual dev */
146 	if ((x->xso.dev != dev) && (x->xso.real_dev == dev))
147 		return skb;
148 
149 	local_irq_save(flags);
150 	sd = this_cpu_ptr(&softnet_data);
151 	err = !skb_queue_empty(&sd->xfrm_backlog);
152 	local_irq_restore(flags);
153 
154 	if (err) {
155 		*again = true;
156 		return skb;
157 	}
158 
159 	if (skb_is_gso(skb) && (unlikely(x->xso.dev != dev) ||
160 				unlikely(xmit_xfrm_check_overflow(skb)))) {
161 		struct sk_buff *segs;
162 
163 		/* Packet got rerouted, fixup features and segment it. */
164 		esp_features = esp_features & ~(NETIF_F_HW_ESP | NETIF_F_GSO_ESP);
165 
166 		segs = skb_gso_segment(skb, esp_features);
167 		if (IS_ERR(segs)) {
168 			kfree_skb(skb);
169 			dev_core_stats_tx_dropped_inc(dev);
170 			return NULL;
171 		} else {
172 			consume_skb(skb);
173 			skb = segs;
174 		}
175 	}
176 
177 	if (!skb->next) {
178 		esp_features |= skb->dev->gso_partial_features;
179 		xfrm_outer_mode_prep(x, skb);
180 
181 		xo->flags |= XFRM_DEV_RESUME;
182 
183 		err = x->type_offload->xmit(x, skb, esp_features);
184 		if (err) {
185 			if (err == -EINPROGRESS)
186 				return NULL;
187 
188 			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
189 			kfree_skb(skb);
190 			return NULL;
191 		}
192 
193 		skb_push(skb, skb->data - skb_mac_header(skb));
194 
195 		return skb;
196 	}
197 
198 	skb_list_walk_safe(skb, skb2, nskb) {
199 		esp_features |= skb->dev->gso_partial_features;
200 		skb_mark_not_on_list(skb2);
201 
202 		xo = xfrm_offload(skb2);
203 		xo->flags |= XFRM_DEV_RESUME;
204 
205 		xfrm_outer_mode_prep(x, skb2);
206 
207 		err = x->type_offload->xmit(x, skb2, esp_features);
208 		if (!err) {
209 			skb2->next = nskb;
210 		} else if (err != -EINPROGRESS) {
211 			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
212 			skb2->next = nskb;
213 			kfree_skb_list(skb2);
214 			return NULL;
215 		} else {
216 			if (skb == skb2)
217 				skb = nskb;
218 			else
219 				pskb->next = nskb;
220 
221 			continue;
222 		}
223 
224 		skb_push(skb2, skb2->data - skb_mac_header(skb2));
225 		pskb = skb2;
226 	}
227 
228 	return skb;
229 }
230 EXPORT_SYMBOL_GPL(validate_xmit_xfrm);
231 
232 int xfrm_dev_state_add(struct net *net, struct xfrm_state *x,
233 		       struct xfrm_user_offload *xuo,
234 		       struct netlink_ext_ack *extack)
235 {
236 	int err;
237 	struct dst_entry *dst;
238 	struct net_device *dev;
239 	struct xfrm_dev_offload *xso = &x->xso;
240 	xfrm_address_t *saddr;
241 	xfrm_address_t *daddr;
242 	bool is_packet_offload;
243 
244 	if (!x->type_offload) {
245 		NL_SET_ERR_MSG(extack, "Type doesn't support offload");
246 		return -EINVAL;
247 	}
248 
249 	/* We don't yet support UDP encapsulation and TFC padding. */
250 	if (x->encap || x->tfcpad) {
251 		NL_SET_ERR_MSG(extack, "Encapsulation and TFC padding can't be offloaded");
252 		return -EINVAL;
253 	}
254 
255 	if (xuo->flags &
256 	    ~(XFRM_OFFLOAD_IPV6 | XFRM_OFFLOAD_INBOUND | XFRM_OFFLOAD_PACKET)) {
257 		NL_SET_ERR_MSG(extack, "Unrecognized flags in offload request");
258 		return -EINVAL;
259 	}
260 
261 	is_packet_offload = xuo->flags & XFRM_OFFLOAD_PACKET;
262 	dev = dev_get_by_index(net, xuo->ifindex);
263 	if (!dev) {
264 		if (!(xuo->flags & XFRM_OFFLOAD_INBOUND)) {
265 			saddr = &x->props.saddr;
266 			daddr = &x->id.daddr;
267 		} else {
268 			saddr = &x->id.daddr;
269 			daddr = &x->props.saddr;
270 		}
271 
272 		dst = __xfrm_dst_lookup(net, 0, 0, saddr, daddr,
273 					x->props.family,
274 					xfrm_smark_get(0, x));
275 		if (IS_ERR(dst))
276 			return (is_packet_offload) ? -EINVAL : 0;
277 
278 		dev = dst->dev;
279 
280 		dev_hold(dev);
281 		dst_release(dst);
282 	}
283 
284 	if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_state_add) {
285 		xso->dev = NULL;
286 		dev_put(dev);
287 		return (is_packet_offload) ? -EINVAL : 0;
288 	}
289 
290 	if (!is_packet_offload && x->props.flags & XFRM_STATE_ESN &&
291 	    !dev->xfrmdev_ops->xdo_dev_state_advance_esn) {
292 		NL_SET_ERR_MSG(extack, "Device doesn't support offload with ESN");
293 		xso->dev = NULL;
294 		dev_put(dev);
295 		return -EINVAL;
296 	}
297 
298 	xso->dev = dev;
299 	netdev_tracker_alloc(dev, &xso->dev_tracker, GFP_ATOMIC);
300 	xso->real_dev = dev;
301 
302 	if (xuo->flags & XFRM_OFFLOAD_INBOUND)
303 		xso->dir = XFRM_DEV_OFFLOAD_IN;
304 	else
305 		xso->dir = XFRM_DEV_OFFLOAD_OUT;
306 
307 	if (is_packet_offload)
308 		xso->type = XFRM_DEV_OFFLOAD_PACKET;
309 	else
310 		xso->type = XFRM_DEV_OFFLOAD_CRYPTO;
311 
312 	err = dev->xfrmdev_ops->xdo_dev_state_add(x, extack);
313 	if (err) {
314 		xso->dev = NULL;
315 		xso->dir = 0;
316 		xso->real_dev = NULL;
317 		netdev_put(dev, &xso->dev_tracker);
318 		xso->type = XFRM_DEV_OFFLOAD_UNSPECIFIED;
319 
320 		/* User explicitly requested packet offload mode and configured
321 		 * policy in addition to the XFRM state. So be civil to users,
322 		 * and return an error instead of taking fallback path.
323 		 *
324 		 * This WARN_ON() can be seen as a documentation for driver
325 		 * authors to do not return -EOPNOTSUPP in packet offload mode.
326 		 */
327 		WARN_ON(err == -EOPNOTSUPP && is_packet_offload);
328 		if (err != -EOPNOTSUPP || is_packet_offload) {
329 			NL_SET_ERR_MSG_WEAK(extack, "Device failed to offload this state");
330 			return err;
331 		}
332 	}
333 
334 	return 0;
335 }
336 EXPORT_SYMBOL_GPL(xfrm_dev_state_add);
337 
338 int xfrm_dev_policy_add(struct net *net, struct xfrm_policy *xp,
339 			struct xfrm_user_offload *xuo, u8 dir,
340 			struct netlink_ext_ack *extack)
341 {
342 	struct xfrm_dev_offload *xdo = &xp->xdo;
343 	struct net_device *dev;
344 	int err;
345 
346 	if (!xuo->flags || xuo->flags & ~XFRM_OFFLOAD_PACKET) {
347 		/* We support only packet offload mode and it means
348 		 * that user must set XFRM_OFFLOAD_PACKET bit.
349 		 */
350 		NL_SET_ERR_MSG(extack, "Unrecognized flags in offload request");
351 		return -EINVAL;
352 	}
353 
354 	dev = dev_get_by_index(net, xuo->ifindex);
355 	if (!dev)
356 		return -EINVAL;
357 
358 	if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_policy_add) {
359 		xdo->dev = NULL;
360 		dev_put(dev);
361 		NL_SET_ERR_MSG(extack, "Policy offload is not supported");
362 		return -EINVAL;
363 	}
364 
365 	xdo->dev = dev;
366 	netdev_tracker_alloc(dev, &xdo->dev_tracker, GFP_ATOMIC);
367 	xdo->real_dev = dev;
368 	xdo->type = XFRM_DEV_OFFLOAD_PACKET;
369 	switch (dir) {
370 	case XFRM_POLICY_IN:
371 		xdo->dir = XFRM_DEV_OFFLOAD_IN;
372 		break;
373 	case XFRM_POLICY_OUT:
374 		xdo->dir = XFRM_DEV_OFFLOAD_OUT;
375 		break;
376 	case XFRM_POLICY_FWD:
377 		xdo->dir = XFRM_DEV_OFFLOAD_FWD;
378 		break;
379 	default:
380 		xdo->dev = NULL;
381 		netdev_put(dev, &xdo->dev_tracker);
382 		NL_SET_ERR_MSG(extack, "Unrecognized offload direction");
383 		return -EINVAL;
384 	}
385 
386 	err = dev->xfrmdev_ops->xdo_dev_policy_add(xp, extack);
387 	if (err) {
388 		xdo->dev = NULL;
389 		xdo->real_dev = NULL;
390 		xdo->type = XFRM_DEV_OFFLOAD_UNSPECIFIED;
391 		xdo->dir = 0;
392 		netdev_put(dev, &xdo->dev_tracker);
393 		NL_SET_ERR_MSG_WEAK(extack, "Device failed to offload this policy");
394 		return err;
395 	}
396 
397 	return 0;
398 }
399 EXPORT_SYMBOL_GPL(xfrm_dev_policy_add);
400 
401 bool xfrm_dev_offload_ok(struct sk_buff *skb, struct xfrm_state *x)
402 {
403 	int mtu;
404 	struct dst_entry *dst = skb_dst(skb);
405 	struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
406 	struct net_device *dev = x->xso.dev;
407 
408 	if (!x->type_offload || x->encap)
409 		return false;
410 
411 	if (x->xso.type == XFRM_DEV_OFFLOAD_PACKET ||
412 	    ((!dev || (dev == xfrm_dst_path(dst)->dev)) &&
413 	     !xdst->child->xfrm)) {
414 		mtu = xfrm_state_mtu(x, xdst->child_mtu_cached);
415 		if (skb->len <= mtu)
416 			goto ok;
417 
418 		if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
419 			goto ok;
420 	}
421 
422 	return false;
423 
424 ok:
425 	if (dev && dev->xfrmdev_ops && dev->xfrmdev_ops->xdo_dev_offload_ok)
426 		return x->xso.dev->xfrmdev_ops->xdo_dev_offload_ok(skb, x);
427 
428 	return true;
429 }
430 EXPORT_SYMBOL_GPL(xfrm_dev_offload_ok);
431 
432 void xfrm_dev_resume(struct sk_buff *skb)
433 {
434 	struct net_device *dev = skb->dev;
435 	int ret = NETDEV_TX_BUSY;
436 	struct netdev_queue *txq;
437 	struct softnet_data *sd;
438 	unsigned long flags;
439 
440 	rcu_read_lock();
441 	txq = netdev_core_pick_tx(dev, skb, NULL);
442 
443 	HARD_TX_LOCK(dev, txq, smp_processor_id());
444 	if (!netif_xmit_frozen_or_stopped(txq))
445 		skb = dev_hard_start_xmit(skb, dev, txq, &ret);
446 	HARD_TX_UNLOCK(dev, txq);
447 
448 	if (!dev_xmit_complete(ret)) {
449 		local_irq_save(flags);
450 		sd = this_cpu_ptr(&softnet_data);
451 		skb_queue_tail(&sd->xfrm_backlog, skb);
452 		raise_softirq_irqoff(NET_TX_SOFTIRQ);
453 		local_irq_restore(flags);
454 	}
455 	rcu_read_unlock();
456 }
457 EXPORT_SYMBOL_GPL(xfrm_dev_resume);
458 
459 void xfrm_dev_backlog(struct softnet_data *sd)
460 {
461 	struct sk_buff_head *xfrm_backlog = &sd->xfrm_backlog;
462 	struct sk_buff_head list;
463 	struct sk_buff *skb;
464 
465 	if (skb_queue_empty(xfrm_backlog))
466 		return;
467 
468 	__skb_queue_head_init(&list);
469 
470 	spin_lock(&xfrm_backlog->lock);
471 	skb_queue_splice_init(xfrm_backlog, &list);
472 	spin_unlock(&xfrm_backlog->lock);
473 
474 	while (!skb_queue_empty(&list)) {
475 		skb = __skb_dequeue(&list);
476 		xfrm_dev_resume(skb);
477 	}
478 
479 }
480 #endif
481 
482 static int xfrm_api_check(struct net_device *dev)
483 {
484 #ifdef CONFIG_XFRM_OFFLOAD
485 	if ((dev->features & NETIF_F_HW_ESP_TX_CSUM) &&
486 	    !(dev->features & NETIF_F_HW_ESP))
487 		return NOTIFY_BAD;
488 
489 	if ((dev->features & NETIF_F_HW_ESP) &&
490 	    (!(dev->xfrmdev_ops &&
491 	       dev->xfrmdev_ops->xdo_dev_state_add &&
492 	       dev->xfrmdev_ops->xdo_dev_state_delete)))
493 		return NOTIFY_BAD;
494 #else
495 	if (dev->features & (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM))
496 		return NOTIFY_BAD;
497 #endif
498 
499 	return NOTIFY_DONE;
500 }
501 
502 static int xfrm_dev_down(struct net_device *dev)
503 {
504 	if (dev->features & NETIF_F_HW_ESP) {
505 		xfrm_dev_state_flush(dev_net(dev), dev, true);
506 		xfrm_dev_policy_flush(dev_net(dev), dev, true);
507 	}
508 
509 	return NOTIFY_DONE;
510 }
511 
512 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
513 {
514 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
515 
516 	switch (event) {
517 	case NETDEV_REGISTER:
518 		return xfrm_api_check(dev);
519 
520 	case NETDEV_FEAT_CHANGE:
521 		return xfrm_api_check(dev);
522 
523 	case NETDEV_DOWN:
524 	case NETDEV_UNREGISTER:
525 		return xfrm_dev_down(dev);
526 	}
527 	return NOTIFY_DONE;
528 }
529 
530 static struct notifier_block xfrm_dev_notifier = {
531 	.notifier_call	= xfrm_dev_event,
532 };
533 
534 void __init xfrm_dev_init(void)
535 {
536 	register_netdevice_notifier(&xfrm_dev_notifier);
537 }
538