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
23static 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, len: skb_transport_offset(skb) + x->props.header_len);
33}
34
35static 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,
46 len: skb->mac_len + x->props.header_len - x->props.enc_hdr_len);
47}
48
49static void __xfrm_mode_beet_prep(struct xfrm_state *x, struct sk_buff *skb,
50 unsigned int hsize)
51{
52 struct xfrm_offload *xo = xfrm_offload(skb);
53 int phlen = 0;
54
55 if (xo->flags & XFRM_GSO_SEGMENT)
56 skb->transport_header = skb->network_header + hsize;
57
58 skb_reset_mac_len(skb);
59 if (x->sel.family != AF_INET6) {
60 phlen = IPV4_BEET_PHMAXLEN;
61 if (x->outer_mode.family == AF_INET6)
62 phlen += sizeof(struct ipv6hdr) - sizeof(struct iphdr);
63 }
64
65 pskb_pull(skb, len: skb->mac_len + hsize + (x->props.header_len - phlen));
66}
67
68/* Adjust pointers into the packet when IPsec is done at layer2 */
69static void xfrm_outer_mode_prep(struct xfrm_state *x, struct sk_buff *skb)
70{
71 switch (x->outer_mode.encap) {
72 case XFRM_MODE_IPTFS:
73 case XFRM_MODE_TUNNEL:
74 if (x->outer_mode.family == AF_INET)
75 return __xfrm_mode_tunnel_prep(x, skb,
76 hsize: sizeof(struct iphdr));
77 if (x->outer_mode.family == AF_INET6)
78 return __xfrm_mode_tunnel_prep(x, skb,
79 hsize: sizeof(struct ipv6hdr));
80 break;
81 case XFRM_MODE_TRANSPORT:
82 if (x->outer_mode.family == AF_INET)
83 return __xfrm_transport_prep(x, skb,
84 hsize: sizeof(struct iphdr));
85 if (x->outer_mode.family == AF_INET6)
86 return __xfrm_transport_prep(x, skb,
87 hsize: sizeof(struct ipv6hdr));
88 break;
89 case XFRM_MODE_BEET:
90 if (x->outer_mode.family == AF_INET)
91 return __xfrm_mode_beet_prep(x, skb,
92 hsize: sizeof(struct iphdr));
93 if (x->outer_mode.family == AF_INET6)
94 return __xfrm_mode_beet_prep(x, skb,
95 hsize: sizeof(struct ipv6hdr));
96 break;
97 case XFRM_MODE_ROUTEOPTIMIZATION:
98 case XFRM_MODE_IN_TRIGGER:
99 break;
100 }
101}
102
103static inline bool xmit_xfrm_check_overflow(struct sk_buff *skb)
104{
105 struct xfrm_offload *xo = xfrm_offload(skb);
106 __u32 seq = xo->seq.low;
107
108 seq += skb_shinfo(skb)->gso_segs;
109 if (unlikely(seq < xo->seq.low))
110 return true;
111
112 return false;
113}
114
115struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t features, bool *again)
116{
117 int err;
118 unsigned long flags;
119 struct xfrm_state *x;
120 struct softnet_data *sd;
121 struct sk_buff *skb2, *nskb, *pskb = NULL;
122 netdev_features_t esp_features = features;
123 struct xfrm_offload *xo = xfrm_offload(skb);
124 struct net_device *dev = skb->dev;
125 struct sec_path *sp;
126
127 if (!xo || (xo->flags & XFRM_XMIT))
128 return skb;
129
130 if (!(features & NETIF_F_HW_ESP))
131 esp_features = features & ~(NETIF_F_SG | NETIF_F_CSUM_MASK);
132
133 sp = skb_sec_path(skb);
134 x = sp->xvec[sp->len - 1];
135 if (xo->flags & XFRM_GRO || x->xso.dir == XFRM_DEV_OFFLOAD_IN)
136 return skb;
137
138 /* The packet was sent to HW IPsec packet offload engine,
139 * but to wrong device. Drop the packet, so it won't skip
140 * XFRM stack.
141 */
142 if (x->xso.type == XFRM_DEV_OFFLOAD_PACKET && x->xso.dev != dev) {
143 kfree_skb(skb);
144 dev_core_stats_tx_dropped_inc(dev);
145 return NULL;
146 }
147
148 local_irq_save(flags);
149 sd = this_cpu_ptr(&softnet_data);
150 err = !skb_queue_empty(list: &sd->xfrm_backlog);
151 local_irq_restore(flags);
152
153 if (err) {
154 *again = true;
155 return skb;
156 }
157
158 if (skb_is_gso(skb) && (unlikely(x->xso.dev != dev) ||
159 unlikely(xmit_xfrm_check_overflow(skb)))) {
160 struct sk_buff *segs;
161
162 /* Packet got rerouted, fixup features and segment it. */
163 esp_features = esp_features & ~(NETIF_F_HW_ESP | NETIF_F_GSO_ESP);
164
165 segs = skb_gso_segment(skb, features: esp_features);
166 if (IS_ERR(ptr: segs)) {
167 kfree_skb(skb);
168 dev_core_stats_tx_dropped_inc(dev);
169 return NULL;
170 } else {
171 consume_skb(skb);
172 skb = segs;
173 }
174 }
175
176 if (!skb->next) {
177 esp_features |= skb->dev->gso_partial_features;
178 xfrm_outer_mode_prep(x, skb);
179
180 xo->flags |= XFRM_DEV_RESUME;
181
182 err = x->type_offload->xmit(x, skb, esp_features);
183 if (err) {
184 if (err == -EINPROGRESS)
185 return NULL;
186
187 XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
188 kfree_skb(skb);
189 return NULL;
190 }
191
192 skb_push(skb, len: skb->data - skb_mac_header(skb));
193
194 return skb;
195 }
196
197 skb_list_walk_safe(skb, skb2, nskb) {
198 esp_features |= skb->dev->gso_partial_features;
199 skb_mark_not_on_list(skb: skb2);
200
201 xo = xfrm_offload(skb: skb2);
202 xo->flags |= XFRM_DEV_RESUME;
203
204 xfrm_outer_mode_prep(x, skb: skb2);
205
206 err = x->type_offload->xmit(x, skb2, esp_features);
207 if (!err) {
208 skb2->next = nskb;
209 } else if (err != -EINPROGRESS) {
210 XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
211 skb2->next = nskb;
212 kfree_skb_list(segs: skb2);
213 return NULL;
214 } else {
215 if (skb == skb2)
216 skb = nskb;
217 else
218 pskb->next = nskb;
219
220 continue;
221 }
222
223 skb_push(skb: skb2, len: skb2->data - skb_mac_header(skb: skb2));
224 pskb = skb2;
225 }
226
227 return skb;
228}
229EXPORT_SYMBOL_GPL(validate_xmit_xfrm);
230
231int xfrm_dev_state_add(struct net *net, struct xfrm_state *x,
232 struct xfrm_user_offload *xuo,
233 struct netlink_ext_ack *extack)
234{
235 int err;
236 struct dst_entry *dst;
237 struct net_device *dev;
238 struct xfrm_dev_offload *xso = &x->xso;
239 xfrm_address_t *saddr;
240 xfrm_address_t *daddr;
241 bool is_packet_offload;
242
243 if (xuo->flags &
244 ~(XFRM_OFFLOAD_IPV6 | XFRM_OFFLOAD_INBOUND | XFRM_OFFLOAD_PACKET)) {
245 NL_SET_ERR_MSG(extack, "Unrecognized flags in offload request");
246 return -EINVAL;
247 }
248
249 if ((xuo->flags & XFRM_OFFLOAD_INBOUND && x->dir == XFRM_SA_DIR_OUT) ||
250 (!(xuo->flags & XFRM_OFFLOAD_INBOUND) && x->dir == XFRM_SA_DIR_IN)) {
251 NL_SET_ERR_MSG(extack, "Mismatched SA and offload direction");
252 return -EINVAL;
253 }
254
255 if (xuo->flags & XFRM_OFFLOAD_INBOUND && x->if_id) {
256 NL_SET_ERR_MSG(extack, "XFRM if_id is not supported in RX path");
257 return -EINVAL;
258 }
259
260 is_packet_offload = xuo->flags & XFRM_OFFLOAD_PACKET;
261
262 /* We don't yet support TFC padding. */
263 if (x->tfcpad) {
264 NL_SET_ERR_MSG(extack, "TFC padding can't be offloaded");
265 return -EINVAL;
266 }
267
268 dev = dev_get_by_index(net, ifindex: xuo->ifindex);
269 if (!dev) {
270 struct xfrm_dst_lookup_params params;
271
272 if (!(xuo->flags & XFRM_OFFLOAD_INBOUND)) {
273 saddr = &x->props.saddr;
274 daddr = &x->id.daddr;
275 } else {
276 saddr = &x->id.daddr;
277 daddr = &x->props.saddr;
278 }
279
280 memset(&params, 0, sizeof(params));
281 params.net = net;
282 params.saddr = saddr;
283 params.daddr = daddr;
284 params.mark = xfrm_smark_get(mark: 0, x);
285 dst = __xfrm_dst_lookup(family: x->props.family, params: &params);
286 if (IS_ERR(ptr: dst))
287 return (is_packet_offload) ? -EINVAL : 0;
288
289 dev = dst->dev;
290
291 dev_hold(dev);
292 dst_release(dst);
293 }
294
295 if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_state_add) {
296 xso->dev = NULL;
297 dev_put(dev);
298 return (is_packet_offload) ? -EINVAL : 0;
299 }
300
301 if (!is_packet_offload && x->props.flags & XFRM_STATE_ESN &&
302 !dev->xfrmdev_ops->xdo_dev_state_advance_esn) {
303 NL_SET_ERR_MSG(extack, "Device doesn't support offload with ESN");
304 xso->dev = NULL;
305 dev_put(dev);
306 return -EINVAL;
307 }
308
309 if (!x->type_offload) {
310 NL_SET_ERR_MSG(extack, "Type doesn't support offload");
311 dev_put(dev);
312 return -EINVAL;
313 }
314
315 xso->dev = dev;
316 netdev_tracker_alloc(dev, tracker: &xso->dev_tracker, GFP_ATOMIC);
317
318 if (xuo->flags & XFRM_OFFLOAD_INBOUND)
319 xso->dir = XFRM_DEV_OFFLOAD_IN;
320 else
321 xso->dir = XFRM_DEV_OFFLOAD_OUT;
322
323 if (is_packet_offload)
324 xso->type = XFRM_DEV_OFFLOAD_PACKET;
325 else
326 xso->type = XFRM_DEV_OFFLOAD_CRYPTO;
327
328 err = dev->xfrmdev_ops->xdo_dev_state_add(dev, x, extack);
329 if (err) {
330 xso->dev = NULL;
331 xso->dir = 0;
332 netdev_put(dev, tracker: &xso->dev_tracker);
333 xso->type = XFRM_DEV_OFFLOAD_UNSPECIFIED;
334
335 xfrm_unset_type_offload(x);
336 /* User explicitly requested packet offload mode and configured
337 * policy in addition to the XFRM state. So be civil to users,
338 * and return an error instead of taking fallback path.
339 */
340 if ((err != -EOPNOTSUPP && !is_packet_offload) || is_packet_offload) {
341 NL_SET_ERR_MSG_WEAK(extack, "Device failed to offload this state");
342 return err;
343 }
344 }
345
346 return 0;
347}
348EXPORT_SYMBOL_GPL(xfrm_dev_state_add);
349
350int xfrm_dev_policy_add(struct net *net, struct xfrm_policy *xp,
351 struct xfrm_user_offload *xuo, u8 dir,
352 struct netlink_ext_ack *extack)
353{
354 struct xfrm_dev_offload *xdo = &xp->xdo;
355 struct net_device *dev;
356 int err;
357
358 if (!xuo->flags || xuo->flags & ~XFRM_OFFLOAD_PACKET) {
359 /* We support only packet offload mode and it means
360 * that user must set XFRM_OFFLOAD_PACKET bit.
361 */
362 NL_SET_ERR_MSG(extack, "Unrecognized flags in offload request");
363 return -EINVAL;
364 }
365
366 dev = dev_get_by_index(net, ifindex: xuo->ifindex);
367 if (!dev)
368 return -EINVAL;
369
370 if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_policy_add) {
371 xdo->dev = NULL;
372 dev_put(dev);
373 NL_SET_ERR_MSG(extack, "Policy offload is not supported");
374 return -EINVAL;
375 }
376
377 xdo->dev = dev;
378 netdev_tracker_alloc(dev, tracker: &xdo->dev_tracker, GFP_ATOMIC);
379 xdo->type = XFRM_DEV_OFFLOAD_PACKET;
380 switch (dir) {
381 case XFRM_POLICY_IN:
382 xdo->dir = XFRM_DEV_OFFLOAD_IN;
383 break;
384 case XFRM_POLICY_OUT:
385 xdo->dir = XFRM_DEV_OFFLOAD_OUT;
386 break;
387 case XFRM_POLICY_FWD:
388 xdo->dir = XFRM_DEV_OFFLOAD_FWD;
389 break;
390 default:
391 xdo->dev = NULL;
392 netdev_put(dev, tracker: &xdo->dev_tracker);
393 NL_SET_ERR_MSG(extack, "Unrecognized offload direction");
394 return -EINVAL;
395 }
396
397 err = dev->xfrmdev_ops->xdo_dev_policy_add(xp, extack);
398 if (err) {
399 xdo->dev = NULL;
400 xdo->type = XFRM_DEV_OFFLOAD_UNSPECIFIED;
401 xdo->dir = 0;
402 netdev_put(dev, tracker: &xdo->dev_tracker);
403 NL_SET_ERR_MSG_WEAK(extack, "Device failed to offload this policy");
404 return err;
405 }
406
407 return 0;
408}
409EXPORT_SYMBOL_GPL(xfrm_dev_policy_add);
410
411bool xfrm_dev_offload_ok(struct sk_buff *skb, struct xfrm_state *x)
412{
413 int mtu;
414 struct dst_entry *dst = skb_dst(skb);
415 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
416 struct net_device *dev = x->xso.dev;
417 bool check_tunnel_size;
418
419 if (!x->type_offload ||
420 (x->xso.type == XFRM_DEV_OFFLOAD_UNSPECIFIED && x->encap))
421 return false;
422
423 if ((!dev || dev == xfrm_dst_path(dst)->dev) &&
424 !xdst->child->xfrm) {
425 mtu = xfrm_state_mtu(x, mtu: xdst->child_mtu_cached);
426 if (skb->len <= mtu)
427 goto ok;
428
429 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
430 goto ok;
431 }
432
433 return false;
434
435ok:
436 if (!dev)
437 return true;
438
439 check_tunnel_size = x->xso.type == XFRM_DEV_OFFLOAD_PACKET &&
440 x->props.mode == XFRM_MODE_TUNNEL;
441 switch (skb_dst(skb)->ops->family) {
442 case AF_INET:
443 /* Check for IPv4 options */
444 if (ip_hdr(skb)->ihl != 5)
445 return false;
446 if (check_tunnel_size && xfrm4_tunnel_check_size(skb))
447 return false;
448 break;
449 case AF_INET6:
450 /* Check for IPv6 extensions */
451 if (ipv6_ext_hdr(nexthdr: ipv6_hdr(skb)->nexthdr))
452 return false;
453 if (check_tunnel_size && xfrm6_tunnel_check_size(skb))
454 return false;
455 break;
456 default:
457 break;
458 }
459
460 if (dev->xfrmdev_ops->xdo_dev_offload_ok)
461 return dev->xfrmdev_ops->xdo_dev_offload_ok(skb, x);
462
463 return true;
464}
465EXPORT_SYMBOL_GPL(xfrm_dev_offload_ok);
466
467void xfrm_dev_resume(struct sk_buff *skb)
468{
469 struct net_device *dev = skb->dev;
470 int ret = NETDEV_TX_BUSY;
471 struct netdev_queue *txq;
472 struct softnet_data *sd;
473 unsigned long flags;
474
475 rcu_read_lock();
476 txq = netdev_core_pick_tx(dev, skb, NULL);
477
478 HARD_TX_LOCK(dev, txq, smp_processor_id());
479 if (!netif_xmit_frozen_or_stopped(dev_queue: txq))
480 skb = dev_hard_start_xmit(skb, dev, txq, ret: &ret);
481 HARD_TX_UNLOCK(dev, txq);
482
483 if (!dev_xmit_complete(rc: ret)) {
484 local_irq_save(flags);
485 sd = this_cpu_ptr(&softnet_data);
486 skb_queue_tail(list: &sd->xfrm_backlog, newsk: skb);
487 raise_softirq_irqoff(nr: NET_TX_SOFTIRQ);
488 local_irq_restore(flags);
489 }
490 rcu_read_unlock();
491}
492EXPORT_SYMBOL_GPL(xfrm_dev_resume);
493
494void xfrm_dev_backlog(struct softnet_data *sd)
495{
496 struct sk_buff_head *xfrm_backlog = &sd->xfrm_backlog;
497 struct sk_buff_head list;
498 struct sk_buff *skb;
499
500 if (skb_queue_empty(list: xfrm_backlog))
501 return;
502
503 __skb_queue_head_init(list: &list);
504
505 spin_lock(lock: &xfrm_backlog->lock);
506 skb_queue_splice_init(list: xfrm_backlog, head: &list);
507 spin_unlock(lock: &xfrm_backlog->lock);
508
509 while (!skb_queue_empty(list: &list)) {
510 skb = __skb_dequeue(list: &list);
511 xfrm_dev_resume(skb);
512 }
513
514}
515#endif
516
517static int xfrm_api_check(struct net_device *dev)
518{
519#ifdef CONFIG_XFRM_OFFLOAD
520 if ((dev->features & NETIF_F_HW_ESP_TX_CSUM) &&
521 !(dev->features & NETIF_F_HW_ESP))
522 return NOTIFY_BAD;
523
524 if ((dev->features & NETIF_F_HW_ESP) &&
525 (!(dev->xfrmdev_ops &&
526 dev->xfrmdev_ops->xdo_dev_state_add &&
527 dev->xfrmdev_ops->xdo_dev_state_delete)))
528 return NOTIFY_BAD;
529#else
530 if (dev->features & (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM))
531 return NOTIFY_BAD;
532#endif
533
534 return NOTIFY_DONE;
535}
536
537static int xfrm_dev_down(struct net_device *dev)
538{
539 if (dev->features & NETIF_F_HW_ESP) {
540 xfrm_dev_state_flush(net: dev_net(dev), dev, task_valid: true);
541 xfrm_dev_policy_flush(net: dev_net(dev), dev, task_valid: true);
542 }
543
544 return NOTIFY_DONE;
545}
546
547static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
548{
549 struct net_device *dev = netdev_notifier_info_to_dev(info: ptr);
550
551 switch (event) {
552 case NETDEV_REGISTER:
553 return xfrm_api_check(dev);
554
555 case NETDEV_FEAT_CHANGE:
556 return xfrm_api_check(dev);
557
558 case NETDEV_DOWN:
559 case NETDEV_UNREGISTER:
560 return xfrm_dev_down(dev);
561 }
562 return NOTIFY_DONE;
563}
564
565static struct notifier_block xfrm_dev_notifier = {
566 .notifier_call = xfrm_dev_event,
567};
568
569void __init xfrm_dev_init(void)
570{
571 register_netdevice_notifier(nb: &xfrm_dev_notifier);
572}
573

source code of linux/net/xfrm/xfrm_device.c