1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * xfrm_device.c - IPsec device offloading code.
5 * Copyright (c) 2015 secunet Security Networks AG
8 * Steffen Klassert <steffen.klassert@secunet.com>
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>
20 #include <linux/notifier.h>
22 #ifdef CONFIG_XFRM_OFFLOAD
23 static void __xfrm_transport_prep(struct xfrm_state *x, struct sk_buff *skb,
26 struct xfrm_offload *xo = xfrm_offload(skb);
28 skb_reset_mac_len(skb);
29 if (xo->flags & XFRM_GSO_SEGMENT)
30 skb->transport_header -= x->props.header_len;
32 pskb_pull(skb, skb_transport_offset(skb) + x->props.header_len);
35 static void __xfrm_mode_tunnel_prep(struct xfrm_state *x, struct sk_buff *skb,
39 struct xfrm_offload *xo = xfrm_offload(skb);
41 if (xo->flags & XFRM_GSO_SEGMENT)
42 skb->transport_header = skb->network_header + hsize;
44 skb_reset_mac_len(skb);
45 pskb_pull(skb, skb->mac_len + x->props.header_len);
48 static void __xfrm_mode_beet_prep(struct xfrm_state *x, struct sk_buff *skb,
51 struct xfrm_offload *xo = xfrm_offload(skb);
54 if (xo->flags & XFRM_GSO_SEGMENT)
55 skb->transport_header = skb->network_header + hsize;
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);
64 pskb_pull(skb, skb->mac_len + hsize + (x->props.header_len - phlen));
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)
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));
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));
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));
95 case XFRM_MODE_ROUTEOPTIMIZATION:
96 case XFRM_MODE_IN_TRIGGER:
101 static inline bool xmit_xfrm_check_overflow(struct sk_buff *skb)
103 struct xfrm_offload *xo = xfrm_offload(skb);
104 __u32 seq = xo->seq.low;
106 seq += skb_shinfo(skb)->gso_segs;
107 if (unlikely(seq < xo->seq.low))
113 struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t features, bool *again)
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;
125 if (!xo || (xo->flags & XFRM_XMIT))
128 if (!(features & NETIF_F_HW_ESP))
129 esp_features = features & ~(NETIF_F_SG | NETIF_F_CSUM_MASK);
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)
136 /* The packet was sent to HW IPsec packet offload engine,
137 * but to wrong device. Drop the packet, so it won't skip
140 if (x->xso.type == XFRM_DEV_OFFLOAD_PACKET && x->xso.dev != dev) {
142 dev_core_stats_tx_dropped_inc(dev);
146 /* This skb was already validated on the upper/virtual dev */
147 if ((x->xso.dev != dev) && (x->xso.real_dev == dev))
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);
160 if (skb_is_gso(skb) && (unlikely(x->xso.dev != dev) ||
161 unlikely(xmit_xfrm_check_overflow(skb)))) {
162 struct sk_buff *segs;
164 /* Packet got rerouted, fixup features and segment it. */
165 esp_features = esp_features & ~(NETIF_F_HW_ESP | NETIF_F_GSO_ESP);
167 segs = skb_gso_segment(skb, esp_features);
170 dev_core_stats_tx_dropped_inc(dev);
179 esp_features |= skb->dev->gso_partial_features;
180 xfrm_outer_mode_prep(x, skb);
182 xo->flags |= XFRM_DEV_RESUME;
184 err = x->type_offload->xmit(x, skb, esp_features);
186 if (err == -EINPROGRESS)
189 XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
194 skb_push(skb, skb->data - skb_mac_header(skb));
199 skb_list_walk_safe(skb, skb2, nskb) {
200 esp_features |= skb->dev->gso_partial_features;
201 skb_mark_not_on_list(skb2);
203 xo = xfrm_offload(skb2);
204 xo->flags |= XFRM_DEV_RESUME;
206 xfrm_outer_mode_prep(x, skb2);
208 err = x->type_offload->xmit(x, skb2, esp_features);
211 } else if (err != -EINPROGRESS) {
212 XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
214 kfree_skb_list(skb2);
225 skb_push(skb2, skb2->data - skb_mac_header(skb2));
231 EXPORT_SYMBOL_GPL(validate_xmit_xfrm);
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)
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;
245 if (!x->type_offload) {
246 NL_SET_ERR_MSG(extack, "Type doesn't support offload");
250 /* We don't yet support UDP encapsulation and TFC padding. */
251 if (x->encap || x->tfcpad) {
252 NL_SET_ERR_MSG(extack, "Encapsulation and TFC padding can't be offloaded");
257 ~(XFRM_OFFLOAD_IPV6 | XFRM_OFFLOAD_INBOUND | XFRM_OFFLOAD_PACKET)) {
258 NL_SET_ERR_MSG(extack, "Unrecognized flags in offload request");
262 is_packet_offload = xuo->flags & XFRM_OFFLOAD_PACKET;
263 dev = dev_get_by_index(net, xuo->ifindex);
265 if (!(xuo->flags & XFRM_OFFLOAD_INBOUND)) {
266 saddr = &x->props.saddr;
267 daddr = &x->id.daddr;
269 saddr = &x->id.daddr;
270 daddr = &x->props.saddr;
273 dst = __xfrm_dst_lookup(net, 0, 0, saddr, daddr,
275 xfrm_smark_get(0, x));
277 return (is_packet_offload) ? -EINVAL : 0;
285 if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_state_add) {
288 return (is_packet_offload) ? -EINVAL : 0;
291 if (!is_packet_offload && x->props.flags & XFRM_STATE_ESN &&
292 !dev->xfrmdev_ops->xdo_dev_state_advance_esn) {
293 NL_SET_ERR_MSG(extack, "Device doesn't support offload with ESN");
300 netdev_tracker_alloc(dev, &xso->dev_tracker, GFP_ATOMIC);
303 if (xuo->flags & XFRM_OFFLOAD_INBOUND)
304 xso->dir = XFRM_DEV_OFFLOAD_IN;
306 xso->dir = XFRM_DEV_OFFLOAD_OUT;
308 if (is_packet_offload)
309 xso->type = XFRM_DEV_OFFLOAD_PACKET;
311 xso->type = XFRM_DEV_OFFLOAD_CRYPTO;
313 err = dev->xfrmdev_ops->xdo_dev_state_add(x, extack);
317 xso->real_dev = NULL;
318 netdev_put(dev, &xso->dev_tracker);
319 xso->type = XFRM_DEV_OFFLOAD_UNSPECIFIED;
321 /* User explicitly requested packet offload mode and configured
322 * policy in addition to the XFRM state. So be civil to users,
323 * and return an error instead of taking fallback path.
325 * This WARN_ON() can be seen as a documentation for driver
326 * authors to do not return -EOPNOTSUPP in packet offload mode.
328 WARN_ON(err == -EOPNOTSUPP && is_packet_offload);
329 if (err != -EOPNOTSUPP || is_packet_offload) {
330 NL_SET_ERR_MSG_WEAK(extack, "Device failed to offload this state");
337 EXPORT_SYMBOL_GPL(xfrm_dev_state_add);
339 int xfrm_dev_policy_add(struct net *net, struct xfrm_policy *xp,
340 struct xfrm_user_offload *xuo, u8 dir,
341 struct netlink_ext_ack *extack)
343 struct xfrm_dev_offload *xdo = &xp->xdo;
344 struct net_device *dev;
347 if (!xuo->flags || xuo->flags & ~XFRM_OFFLOAD_PACKET) {
348 /* We support only packet offload mode and it means
349 * that user must set XFRM_OFFLOAD_PACKET bit.
351 NL_SET_ERR_MSG(extack, "Unrecognized flags in offload request");
355 dev = dev_get_by_index(net, xuo->ifindex);
359 if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_policy_add) {
362 NL_SET_ERR_MSG(extack, "Policy offload is not supported");
367 netdev_tracker_alloc(dev, &xdo->dev_tracker, GFP_ATOMIC);
369 xdo->type = XFRM_DEV_OFFLOAD_PACKET;
372 xdo->dir = XFRM_DEV_OFFLOAD_IN;
374 case XFRM_POLICY_OUT:
375 xdo->dir = XFRM_DEV_OFFLOAD_OUT;
377 case XFRM_POLICY_FWD:
378 xdo->dir = XFRM_DEV_OFFLOAD_FWD;
382 netdev_put(dev, &xdo->dev_tracker);
383 NL_SET_ERR_MSG(extack, "Unrecognized offload direction");
387 err = dev->xfrmdev_ops->xdo_dev_policy_add(xp, extack);
390 xdo->real_dev = NULL;
391 xdo->type = XFRM_DEV_OFFLOAD_UNSPECIFIED;
393 netdev_put(dev, &xdo->dev_tracker);
394 NL_SET_ERR_MSG_WEAK(extack, "Device failed to offload this policy");
400 EXPORT_SYMBOL_GPL(xfrm_dev_policy_add);
402 bool xfrm_dev_offload_ok(struct sk_buff *skb, struct xfrm_state *x)
405 struct dst_entry *dst = skb_dst(skb);
406 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
407 struct net_device *dev = x->xso.dev;
409 if (!x->type_offload || x->encap)
412 if (x->xso.type == XFRM_DEV_OFFLOAD_PACKET ||
413 ((!dev || (dev == xfrm_dst_path(dst)->dev)) &&
414 !xdst->child->xfrm)) {
415 mtu = xfrm_state_mtu(x, xdst->child_mtu_cached);
419 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
426 if (dev && dev->xfrmdev_ops && dev->xfrmdev_ops->xdo_dev_offload_ok)
427 return x->xso.dev->xfrmdev_ops->xdo_dev_offload_ok(skb, x);
431 EXPORT_SYMBOL_GPL(xfrm_dev_offload_ok);
433 void xfrm_dev_resume(struct sk_buff *skb)
435 struct net_device *dev = skb->dev;
436 int ret = NETDEV_TX_BUSY;
437 struct netdev_queue *txq;
438 struct softnet_data *sd;
442 txq = netdev_core_pick_tx(dev, skb, NULL);
444 HARD_TX_LOCK(dev, txq, smp_processor_id());
445 if (!netif_xmit_frozen_or_stopped(txq))
446 skb = dev_hard_start_xmit(skb, dev, txq, &ret);
447 HARD_TX_UNLOCK(dev, txq);
449 if (!dev_xmit_complete(ret)) {
450 local_irq_save(flags);
451 sd = this_cpu_ptr(&softnet_data);
452 skb_queue_tail(&sd->xfrm_backlog, skb);
453 raise_softirq_irqoff(NET_TX_SOFTIRQ);
454 local_irq_restore(flags);
458 EXPORT_SYMBOL_GPL(xfrm_dev_resume);
460 void xfrm_dev_backlog(struct softnet_data *sd)
462 struct sk_buff_head *xfrm_backlog = &sd->xfrm_backlog;
463 struct sk_buff_head list;
466 if (skb_queue_empty(xfrm_backlog))
469 __skb_queue_head_init(&list);
471 spin_lock(&xfrm_backlog->lock);
472 skb_queue_splice_init(xfrm_backlog, &list);
473 spin_unlock(&xfrm_backlog->lock);
475 while (!skb_queue_empty(&list)) {
476 skb = __skb_dequeue(&list);
477 xfrm_dev_resume(skb);
483 static int xfrm_api_check(struct net_device *dev)
485 #ifdef CONFIG_XFRM_OFFLOAD
486 if ((dev->features & NETIF_F_HW_ESP_TX_CSUM) &&
487 !(dev->features & NETIF_F_HW_ESP))
490 if ((dev->features & NETIF_F_HW_ESP) &&
491 (!(dev->xfrmdev_ops &&
492 dev->xfrmdev_ops->xdo_dev_state_add &&
493 dev->xfrmdev_ops->xdo_dev_state_delete)))
496 if (dev->features & (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM))
503 static int xfrm_dev_down(struct net_device *dev)
505 if (dev->features & NETIF_F_HW_ESP) {
506 xfrm_dev_state_flush(dev_net(dev), dev, true);
507 xfrm_dev_policy_flush(dev_net(dev), dev, true);
513 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
515 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
518 case NETDEV_REGISTER:
519 return xfrm_api_check(dev);
521 case NETDEV_FEAT_CHANGE:
522 return xfrm_api_check(dev);
525 case NETDEV_UNREGISTER:
526 return xfrm_dev_down(dev);
531 static struct notifier_block xfrm_dev_notifier = {
532 .notifier_call = xfrm_dev_event,
535 void __init xfrm_dev_init(void)
537 register_netdevice_notifier(&xfrm_dev_notifier);