xref: /linux/net/xfrm/xfrm_device.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
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/gso.h>
19 #include <net/xfrm.h>
20 #include <linux/notifier.h>
21 
22 #ifdef CONFIG_XFRM_OFFLOAD
23 static void __xfrm_transport_prep(struct xfrm_state *x, struct sk_buff *skb,
24 				  unsigned int hsize)
25 {
26 	struct xfrm_offload *xo = xfrm_offload(skb);
27 
28 	skb_reset_mac_len(skb);
29 	if (xo->flags & XFRM_GSO_SEGMENT)
30 		skb->transport_header -= x->props.header_len;
31 
32 	pskb_pull(skb, skb_transport_offset(skb) + x->props.header_len);
33 }
34 
35 static void __xfrm_mode_tunnel_prep(struct xfrm_state *x, struct sk_buff *skb,
36 				    unsigned int hsize)
37 
38 {
39 	struct xfrm_offload *xo = xfrm_offload(skb);
40 
41 	if (xo->flags & XFRM_GSO_SEGMENT)
42 		skb->transport_header = skb->network_header + hsize;
43 
44 	skb_reset_mac_len(skb);
45 	pskb_pull(skb, skb->mac_len + x->props.header_len);
46 }
47 
48 static void __xfrm_mode_beet_prep(struct xfrm_state *x, struct sk_buff *skb,
49 				  unsigned int hsize)
50 {
51 	struct xfrm_offload *xo = xfrm_offload(skb);
52 	int phlen = 0;
53 
54 	if (xo->flags & XFRM_GSO_SEGMENT)
55 		skb->transport_header = skb->network_header + hsize;
56 
57 	skb_reset_mac_len(skb);
58 	if (x->sel.family != AF_INET6) {
59 		phlen = IPV4_BEET_PHMAXLEN;
60 		if (x->outer_mode.family == AF_INET6)
61 			phlen += sizeof(struct ipv6hdr) - sizeof(struct iphdr);
62 	}
63 
64 	pskb_pull(skb, skb->mac_len + hsize + (x->props.header_len - phlen));
65 }
66 
67 /* Adjust pointers into the packet when IPsec is done at layer2 */
68 static void xfrm_outer_mode_prep(struct xfrm_state *x, struct sk_buff *skb)
69 {
70 	switch (x->outer_mode.encap) {
71 	case XFRM_MODE_TUNNEL:
72 		if (x->outer_mode.family == AF_INET)
73 			return __xfrm_mode_tunnel_prep(x, skb,
74 						       sizeof(struct iphdr));
75 		if (x->outer_mode.family == AF_INET6)
76 			return __xfrm_mode_tunnel_prep(x, skb,
77 						       sizeof(struct ipv6hdr));
78 		break;
79 	case XFRM_MODE_TRANSPORT:
80 		if (x->outer_mode.family == AF_INET)
81 			return __xfrm_transport_prep(x, skb,
82 						     sizeof(struct iphdr));
83 		if (x->outer_mode.family == AF_INET6)
84 			return __xfrm_transport_prep(x, skb,
85 						     sizeof(struct ipv6hdr));
86 		break;
87 	case XFRM_MODE_BEET:
88 		if (x->outer_mode.family == AF_INET)
89 			return __xfrm_mode_beet_prep(x, skb,
90 						     sizeof(struct iphdr));
91 		if (x->outer_mode.family == AF_INET6)
92 			return __xfrm_mode_beet_prep(x, skb,
93 						     sizeof(struct ipv6hdr));
94 		break;
95 	case XFRM_MODE_ROUTEOPTIMIZATION:
96 	case XFRM_MODE_IN_TRIGGER:
97 		break;
98 	}
99 }
100 
101 static inline bool xmit_xfrm_check_overflow(struct sk_buff *skb)
102 {
103 	struct xfrm_offload *xo = xfrm_offload(skb);
104 	__u32 seq = xo->seq.low;
105 
106 	seq += skb_shinfo(skb)->gso_segs;
107 	if (unlikely(seq < xo->seq.low))
108 		return true;
109 
110 	return false;
111 }
112 
113 struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t features, bool *again)
114 {
115 	int err;
116 	unsigned long flags;
117 	struct xfrm_state *x;
118 	struct softnet_data *sd;
119 	struct sk_buff *skb2, *nskb, *pskb = NULL;
120 	netdev_features_t esp_features = features;
121 	struct xfrm_offload *xo = xfrm_offload(skb);
122 	struct net_device *dev = skb->dev;
123 	struct sec_path *sp;
124 
125 	if (!xo || (xo->flags & XFRM_XMIT))
126 		return skb;
127 
128 	if (!(features & NETIF_F_HW_ESP))
129 		esp_features = features & ~(NETIF_F_SG | NETIF_F_CSUM_MASK);
130 
131 	sp = skb_sec_path(skb);
132 	x = sp->xvec[sp->len - 1];
133 	if (xo->flags & XFRM_GRO || x->xso.dir == XFRM_DEV_OFFLOAD_IN)
134 		return skb;
135 
136 	/* The packet was sent to HW IPsec packet offload engine,
137 	 * but to wrong device. Drop the packet, so it won't skip
138 	 * XFRM stack.
139 	 */
140 	if (x->xso.type == XFRM_DEV_OFFLOAD_PACKET && x->xso.dev != dev) {
141 		kfree_skb(skb);
142 		dev_core_stats_tx_dropped_inc(dev);
143 		return NULL;
144 	}
145 
146 	/* This skb was already validated on the upper/virtual dev */
147 	if ((x->xso.dev != dev) && (x->xso.real_dev == dev))
148 		return skb;
149 
150 	local_irq_save(flags);
151 	sd = this_cpu_ptr(&softnet_data);
152 	err = !skb_queue_empty(&sd->xfrm_backlog);
153 	local_irq_restore(flags);
154 
155 	if (err) {
156 		*again = true;
157 		return skb;
158 	}
159 
160 	if (skb_is_gso(skb) && (unlikely(x->xso.dev != dev) ||
161 				unlikely(xmit_xfrm_check_overflow(skb)))) {
162 		struct sk_buff *segs;
163 
164 		/* Packet got rerouted, fixup features and segment it. */
165 		esp_features = esp_features & ~(NETIF_F_HW_ESP | NETIF_F_GSO_ESP);
166 
167 		segs = skb_gso_segment(skb, esp_features);
168 		if (IS_ERR(segs)) {
169 			kfree_skb(skb);
170 			dev_core_stats_tx_dropped_inc(dev);
171 			return NULL;
172 		} else {
173 			consume_skb(skb);
174 			skb = segs;
175 		}
176 	}
177 
178 	if (!skb->next) {
179 		esp_features |= skb->dev->gso_partial_features;
180 		xfrm_outer_mode_prep(x, skb);
181 
182 		xo->flags |= XFRM_DEV_RESUME;
183 
184 		err = x->type_offload->xmit(x, skb, esp_features);
185 		if (err) {
186 			if (err == -EINPROGRESS)
187 				return NULL;
188 
189 			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
190 			kfree_skb(skb);
191 			return NULL;
192 		}
193 
194 		skb_push(skb, skb->data - skb_mac_header(skb));
195 
196 		return skb;
197 	}
198 
199 	skb_list_walk_safe(skb, skb2, nskb) {
200 		esp_features |= skb->dev->gso_partial_features;
201 		skb_mark_not_on_list(skb2);
202 
203 		xo = xfrm_offload(skb2);
204 		xo->flags |= XFRM_DEV_RESUME;
205 
206 		xfrm_outer_mode_prep(x, skb2);
207 
208 		err = x->type_offload->xmit(x, skb2, esp_features);
209 		if (!err) {
210 			skb2->next = nskb;
211 		} else if (err != -EINPROGRESS) {
212 			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
213 			skb2->next = nskb;
214 			kfree_skb_list(skb2);
215 			return NULL;
216 		} else {
217 			if (skb == skb2)
218 				skb = nskb;
219 			else
220 				pskb->next = nskb;
221 
222 			continue;
223 		}
224 
225 		skb_push(skb2, skb2->data - skb_mac_header(skb2));
226 		pskb = skb2;
227 	}
228 
229 	return skb;
230 }
231 EXPORT_SYMBOL_GPL(validate_xmit_xfrm);
232 
233 int xfrm_dev_state_add(struct net *net, struct xfrm_state *x,
234 		       struct xfrm_user_offload *xuo,
235 		       struct netlink_ext_ack *extack)
236 {
237 	int err;
238 	struct dst_entry *dst;
239 	struct net_device *dev;
240 	struct xfrm_dev_offload *xso = &x->xso;
241 	xfrm_address_t *saddr;
242 	xfrm_address_t *daddr;
243 	bool is_packet_offload;
244 
245 	if (!x->type_offload) {
246 		NL_SET_ERR_MSG(extack, "Type doesn't support offload");
247 		return -EINVAL;
248 	}
249 
250 	if (xuo->flags &
251 	    ~(XFRM_OFFLOAD_IPV6 | XFRM_OFFLOAD_INBOUND | XFRM_OFFLOAD_PACKET)) {
252 		NL_SET_ERR_MSG(extack, "Unrecognized flags in offload request");
253 		return -EINVAL;
254 	}
255 
256 	if ((xuo->flags & XFRM_OFFLOAD_INBOUND && x->dir == XFRM_SA_DIR_OUT) ||
257 	    (!(xuo->flags & XFRM_OFFLOAD_INBOUND) && x->dir == XFRM_SA_DIR_IN)) {
258 		NL_SET_ERR_MSG(extack, "Mismatched SA and offload direction");
259 		return -EINVAL;
260 	}
261 
262 	is_packet_offload = xuo->flags & XFRM_OFFLOAD_PACKET;
263 
264 	/* We don't yet support TFC padding. */
265 	if (x->tfcpad) {
266 		NL_SET_ERR_MSG(extack, "TFC padding can't be offloaded");
267 		return -EINVAL;
268 	}
269 
270 	dev = dev_get_by_index(net, xuo->ifindex);
271 	if (!dev) {
272 		struct xfrm_dst_lookup_params params;
273 
274 		if (!(xuo->flags & XFRM_OFFLOAD_INBOUND)) {
275 			saddr = &x->props.saddr;
276 			daddr = &x->id.daddr;
277 		} else {
278 			saddr = &x->id.daddr;
279 			daddr = &x->props.saddr;
280 		}
281 
282 		memset(&params, 0, sizeof(params));
283 		params.net = net;
284 		params.saddr = saddr;
285 		params.daddr = daddr;
286 		params.mark = xfrm_smark_get(0, x);
287 		dst = __xfrm_dst_lookup(x->props.family, &params);
288 		if (IS_ERR(dst))
289 			return (is_packet_offload) ? -EINVAL : 0;
290 
291 		dev = dst->dev;
292 
293 		dev_hold(dev);
294 		dst_release(dst);
295 	}
296 
297 	if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_state_add) {
298 		xso->dev = NULL;
299 		dev_put(dev);
300 		return (is_packet_offload) ? -EINVAL : 0;
301 	}
302 
303 	if (!is_packet_offload && x->props.flags & XFRM_STATE_ESN &&
304 	    !dev->xfrmdev_ops->xdo_dev_state_advance_esn) {
305 		NL_SET_ERR_MSG(extack, "Device doesn't support offload with ESN");
306 		xso->dev = NULL;
307 		dev_put(dev);
308 		return -EINVAL;
309 	}
310 
311 	xso->dev = dev;
312 	netdev_tracker_alloc(dev, &xso->dev_tracker, GFP_ATOMIC);
313 	xso->real_dev = dev;
314 
315 	if (xuo->flags & XFRM_OFFLOAD_INBOUND)
316 		xso->dir = XFRM_DEV_OFFLOAD_IN;
317 	else
318 		xso->dir = XFRM_DEV_OFFLOAD_OUT;
319 
320 	if (is_packet_offload)
321 		xso->type = XFRM_DEV_OFFLOAD_PACKET;
322 	else
323 		xso->type = XFRM_DEV_OFFLOAD_CRYPTO;
324 
325 	err = dev->xfrmdev_ops->xdo_dev_state_add(x, extack);
326 	if (err) {
327 		xso->dev = NULL;
328 		xso->dir = 0;
329 		xso->real_dev = NULL;
330 		netdev_put(dev, &xso->dev_tracker);
331 		xso->type = XFRM_DEV_OFFLOAD_UNSPECIFIED;
332 
333 		/* User explicitly requested packet offload mode and configured
334 		 * policy in addition to the XFRM state. So be civil to users,
335 		 * and return an error instead of taking fallback path.
336 		 */
337 		if ((err != -EOPNOTSUPP && !is_packet_offload) || is_packet_offload) {
338 			NL_SET_ERR_MSG_WEAK(extack, "Device failed to offload this state");
339 			return err;
340 		}
341 	}
342 
343 	return 0;
344 }
345 EXPORT_SYMBOL_GPL(xfrm_dev_state_add);
346 
347 int xfrm_dev_policy_add(struct net *net, struct xfrm_policy *xp,
348 			struct xfrm_user_offload *xuo, u8 dir,
349 			struct netlink_ext_ack *extack)
350 {
351 	struct xfrm_dev_offload *xdo = &xp->xdo;
352 	struct net_device *dev;
353 	int err;
354 
355 	if (!xuo->flags || xuo->flags & ~XFRM_OFFLOAD_PACKET) {
356 		/* We support only packet offload mode and it means
357 		 * that user must set XFRM_OFFLOAD_PACKET bit.
358 		 */
359 		NL_SET_ERR_MSG(extack, "Unrecognized flags in offload request");
360 		return -EINVAL;
361 	}
362 
363 	dev = dev_get_by_index(net, xuo->ifindex);
364 	if (!dev)
365 		return -EINVAL;
366 
367 	if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_policy_add) {
368 		xdo->dev = NULL;
369 		dev_put(dev);
370 		NL_SET_ERR_MSG(extack, "Policy offload is not supported");
371 		return -EINVAL;
372 	}
373 
374 	xdo->dev = dev;
375 	netdev_tracker_alloc(dev, &xdo->dev_tracker, GFP_ATOMIC);
376 	xdo->real_dev = dev;
377 	xdo->type = XFRM_DEV_OFFLOAD_PACKET;
378 	switch (dir) {
379 	case XFRM_POLICY_IN:
380 		xdo->dir = XFRM_DEV_OFFLOAD_IN;
381 		break;
382 	case XFRM_POLICY_OUT:
383 		xdo->dir = XFRM_DEV_OFFLOAD_OUT;
384 		break;
385 	case XFRM_POLICY_FWD:
386 		xdo->dir = XFRM_DEV_OFFLOAD_FWD;
387 		break;
388 	default:
389 		xdo->dev = NULL;
390 		netdev_put(dev, &xdo->dev_tracker);
391 		NL_SET_ERR_MSG(extack, "Unrecognized offload direction");
392 		return -EINVAL;
393 	}
394 
395 	err = dev->xfrmdev_ops->xdo_dev_policy_add(xp, extack);
396 	if (err) {
397 		xdo->dev = NULL;
398 		xdo->real_dev = NULL;
399 		xdo->type = XFRM_DEV_OFFLOAD_UNSPECIFIED;
400 		xdo->dir = 0;
401 		netdev_put(dev, &xdo->dev_tracker);
402 		NL_SET_ERR_MSG_WEAK(extack, "Device failed to offload this policy");
403 		return err;
404 	}
405 
406 	return 0;
407 }
408 EXPORT_SYMBOL_GPL(xfrm_dev_policy_add);
409 
410 bool xfrm_dev_offload_ok(struct sk_buff *skb, struct xfrm_state *x)
411 {
412 	int mtu;
413 	struct dst_entry *dst = skb_dst(skb);
414 	struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
415 	struct net_device *dev = x->xso.dev;
416 
417 	if (!x->type_offload ||
418 	    (x->xso.type == XFRM_DEV_OFFLOAD_UNSPECIFIED && x->encap))
419 		return false;
420 
421 	if (x->xso.type == XFRM_DEV_OFFLOAD_PACKET ||
422 	    ((!dev || (dev == xfrm_dst_path(dst)->dev)) &&
423 	     !xdst->child->xfrm)) {
424 		mtu = xfrm_state_mtu(x, xdst->child_mtu_cached);
425 		if (skb->len <= mtu)
426 			goto ok;
427 
428 		if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
429 			goto ok;
430 	}
431 
432 	return false;
433 
434 ok:
435 	if (dev && dev->xfrmdev_ops && dev->xfrmdev_ops->xdo_dev_offload_ok)
436 		return x->xso.dev->xfrmdev_ops->xdo_dev_offload_ok(skb, x);
437 
438 	return true;
439 }
440 EXPORT_SYMBOL_GPL(xfrm_dev_offload_ok);
441 
442 void xfrm_dev_resume(struct sk_buff *skb)
443 {
444 	struct net_device *dev = skb->dev;
445 	int ret = NETDEV_TX_BUSY;
446 	struct netdev_queue *txq;
447 	struct softnet_data *sd;
448 	unsigned long flags;
449 
450 	rcu_read_lock();
451 	txq = netdev_core_pick_tx(dev, skb, NULL);
452 
453 	HARD_TX_LOCK(dev, txq, smp_processor_id());
454 	if (!netif_xmit_frozen_or_stopped(txq))
455 		skb = dev_hard_start_xmit(skb, dev, txq, &ret);
456 	HARD_TX_UNLOCK(dev, txq);
457 
458 	if (!dev_xmit_complete(ret)) {
459 		local_irq_save(flags);
460 		sd = this_cpu_ptr(&softnet_data);
461 		skb_queue_tail(&sd->xfrm_backlog, skb);
462 		raise_softirq_irqoff(NET_TX_SOFTIRQ);
463 		local_irq_restore(flags);
464 	}
465 	rcu_read_unlock();
466 }
467 EXPORT_SYMBOL_GPL(xfrm_dev_resume);
468 
469 void xfrm_dev_backlog(struct softnet_data *sd)
470 {
471 	struct sk_buff_head *xfrm_backlog = &sd->xfrm_backlog;
472 	struct sk_buff_head list;
473 	struct sk_buff *skb;
474 
475 	if (skb_queue_empty(xfrm_backlog))
476 		return;
477 
478 	__skb_queue_head_init(&list);
479 
480 	spin_lock(&xfrm_backlog->lock);
481 	skb_queue_splice_init(xfrm_backlog, &list);
482 	spin_unlock(&xfrm_backlog->lock);
483 
484 	while (!skb_queue_empty(&list)) {
485 		skb = __skb_dequeue(&list);
486 		xfrm_dev_resume(skb);
487 	}
488 
489 }
490 #endif
491 
492 static int xfrm_api_check(struct net_device *dev)
493 {
494 #ifdef CONFIG_XFRM_OFFLOAD
495 	if ((dev->features & NETIF_F_HW_ESP_TX_CSUM) &&
496 	    !(dev->features & NETIF_F_HW_ESP))
497 		return NOTIFY_BAD;
498 
499 	if ((dev->features & NETIF_F_HW_ESP) &&
500 	    (!(dev->xfrmdev_ops &&
501 	       dev->xfrmdev_ops->xdo_dev_state_add &&
502 	       dev->xfrmdev_ops->xdo_dev_state_delete)))
503 		return NOTIFY_BAD;
504 #else
505 	if (dev->features & (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM))
506 		return NOTIFY_BAD;
507 #endif
508 
509 	return NOTIFY_DONE;
510 }
511 
512 static int xfrm_dev_down(struct net_device *dev)
513 {
514 	if (dev->features & NETIF_F_HW_ESP) {
515 		xfrm_dev_state_flush(dev_net(dev), dev, true);
516 		xfrm_dev_policy_flush(dev_net(dev), dev, true);
517 	}
518 
519 	return NOTIFY_DONE;
520 }
521 
522 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
523 {
524 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
525 
526 	switch (event) {
527 	case NETDEV_REGISTER:
528 		return xfrm_api_check(dev);
529 
530 	case NETDEV_FEAT_CHANGE:
531 		return xfrm_api_check(dev);
532 
533 	case NETDEV_DOWN:
534 	case NETDEV_UNREGISTER:
535 		return xfrm_dev_down(dev);
536 	}
537 	return NOTIFY_DONE;
538 }
539 
540 static struct notifier_block xfrm_dev_notifier = {
541 	.notifier_call	= xfrm_dev_event,
542 };
543 
544 void __init xfrm_dev_init(void)
545 {
546 	register_netdevice_notifier(&xfrm_dev_notifier);
547 }
548