net/xfrm/xfrm_device.c

   1 /*
   2  * xfrm_device.c - IPsec device offloading code.
   3  *
   4  * Copyright (c) 2015 secunet Security Networks AG
   5  *
   6  * Author:
   7  * Steffen Klassert <steffen.klassert@secunet.com>
   8  *
   9  * This program is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU General Public License
  11  * as published by the Free Software Foundation; either version
  12  * 2 of the License, or (at your option) any later version.
  13  */
  14
  15 #include <linux/errno.h>
  16 #include <linux/module.h>
  17 #include <linux/netdevice.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/slab.h>
  20 #include <linux/spinlock.h>
  21 #include <net/dst.h>
  22 #include <net/xfrm.h>
  23 #include <linux/notifier.h>
  24
  25 #ifdef CONFIG_XFRM_OFFLOAD
  26 struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t features, bool *again)
  27 {
  28         int err;
  29         unsigned long flags;
  30         struct xfrm_state *x;
  31         struct sk_buff *skb2;
  32         struct softnet_data *sd;
  33         netdev_features_t esp_features = features;
  34         struct xfrm_offload *xo = xfrm_offload(skb);
  35
  36         if (!xo || (xo->flags & XFRM_XMIT))
  37                 return skb;
  38
  39         if (!(features & NETIF_F_HW_ESP))
  40                 esp_features = features & ~(NETIF_F_SG | NETIF_F_CSUM_MASK);
  41
  42         x = skb->sp->xvec[skb->sp->len - 1];
  43         if (xo->flags & XFRM_GRO || x->xso.flags & XFRM_OFFLOAD_INBOUND)
  44                 return skb;
  45
  46         local_irq_save(flags);
  47         sd = this_cpu_ptr(&softnet_data);
  48         err = !skb_queue_empty(&sd->xfrm_backlog);
  49         local_irq_restore(flags);
  50
  51         if (err) {
  52                 *again = true;
  53                 return skb;
  54         }
  55
  56         xo->flags |= XFRM_XMIT;
  57
  58         if (skb_is_gso(skb)) {
  59                 struct net_device *dev = skb->dev;
  60
  61                 if (unlikely(x->xso.dev != dev)) {
  62                         struct sk_buff *segs;
  63
  64                         /* Packet got rerouted, fixup features and segment it. */
  65                         esp_features = esp_features & ~(NETIF_F_HW_ESP
  66                                                         | NETIF_F_GSO_ESP);
  67
  68                         segs = skb_gso_segment(skb, esp_features);
  69                         if (IS_ERR(segs)) {
  70                                 kfree_skb(skb);
  71                                 atomic_long_inc(&dev->tx_dropped);
  72                                 return NULL;
  73                         } else {
  74                                 consume_skb(skb);
  75                                 skb = segs;
  76                         }
  77                 }
  78         }
  79
  80         if (!skb->next) {
  81                 x->outer_mode->xmit(x, skb);
  82
  83                 xo->flags |= XFRM_DEV_RESUME;
  84
  85                 err = x->type_offload->xmit(x, skb, esp_features);
  86                 if (err) {
  87                         if (err == -EINPROGRESS)
  88                                 return NULL;
  89
  90                         XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
  91                         kfree_skb(skb);
  92                         return NULL;
  93                 }
  94
  95                 skb_push(skb, skb->data - skb_mac_header(skb));
  96
  97                 return skb;
  98         }
  99
 100         skb2 = skb;
 101
 102         do {
 103                 struct sk_buff *nskb = skb2->next;
 104                 skb2->next = NULL;
 105
 106                 xo = xfrm_offload(skb2);
 107                 xo->flags |= XFRM_DEV_RESUME;
 108
 109                 x->outer_mode->xmit(x, skb2);
 110
 111                 err = x->type_offload->xmit(x, skb2, esp_features);
 112                 if (!err) {
 113                         skb2->next = nskb;
 114                 } else if (err != -EINPROGRESS) {
 115                         XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
 116                         skb2->next = nskb;
 117                         kfree_skb_list(skb2);
 118                         return NULL;
 119                 } else {
 120                         if (skb == skb2)
 121                                 skb = nskb;
 122
 123                         if (!skb)
 124                                 return NULL;
 125
 126                         goto skip_push;
 127                 }
 128
 129                 skb_push(skb2, skb2->data - skb_mac_header(skb2));
 130
 131 skip_push:
 132                 skb2 = nskb;
 133         } while (skb2);
 134
 135         return skb;
 136 }
 137 EXPORT_SYMBOL_GPL(validate_xmit_xfrm);
 138
 139 int xfrm_dev_state_add(struct net *net, struct xfrm_state *x,
 140                        struct xfrm_user_offload *xuo)
 141 {
 142         int err;
 143         struct dst_entry *dst;
 144         struct net_device *dev;
 145         struct xfrm_state_offload *xso = &x->xso;
 146         xfrm_address_t *saddr;
 147         xfrm_address_t *daddr;
 148
 149         if (!x->type_offload)
 150                 return -EINVAL;
 151
 152         /* We don't yet support UDP encapsulation and TFC padding. */
 153         if (x->encap || x->tfcpad)
 154                 return -EINVAL;
 155
 156         if (xuo->flags & ~(XFRM_OFFLOAD_IPV6 | XFRM_OFFLOAD_INBOUND))
 157                 return -EINVAL;
 158
 159         dev = dev_get_by_index(net, xuo->ifindex);
 160         if (!dev) {
 161                 if (!(xuo->flags & XFRM_OFFLOAD_INBOUND)) {
 162                         saddr = &x->props.saddr;
 163                         daddr = &x->id.daddr;
 164                 } else {
 165                         saddr = &x->id.daddr;
 166                         daddr = &x->props.saddr;
 167                 }
 168
 169                 dst = __xfrm_dst_lookup(net, 0, 0, saddr, daddr,
 170                                         x->props.family,
 171                                         xfrm_smark_get(0, x));
 172                 if (IS_ERR(dst))
 173                         return 0;
 174
 175                 dev = dst->dev;
 176
 177                 dev_hold(dev);
 178                 dst_release(dst);
 179         }
 180
 181         if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_state_add) {
 182                 xso->dev = NULL;
 183                 dev_put(dev);
 184                 return 0;
 185         }
 186
 187         if (x->props.flags & XFRM_STATE_ESN &&
 188             !dev->xfrmdev_ops->xdo_dev_state_advance_esn) {
 189                 xso->dev = NULL;
 190                 dev_put(dev);
 191                 return -EINVAL;
 192         }
 193
 194         xso->dev = dev;
 195         xso->num_exthdrs = 1;
 196         /* Don't forward bit that is not implemented */
 197         xso->flags = xuo->flags & ~XFRM_OFFLOAD_IPV6;
 198
 199         err = dev->xfrmdev_ops->xdo_dev_state_add(x);
 200         if (err) {
 201                 xso->dev = NULL;
 202                 dev_put(dev);
 203                 return err;
 204         }
 205
 206         return 0;
 207 }
 208 EXPORT_SYMBOL_GPL(xfrm_dev_state_add);
 209
 210 bool xfrm_dev_offload_ok(struct sk_buff *skb, struct xfrm_state *x)
 211 {
 212         int mtu;
 213         struct dst_entry *dst = skb_dst(skb);
 214         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
 215         struct net_device *dev = x->xso.dev;
 216
 217         if (!x->type_offload || x->encap)
 218                 return false;
 219
 220         if ((!dev || (dev == xfrm_dst_path(dst)->dev)) &&
 221             (!xdst->child->xfrm && x->type->get_mtu)) {
 222                 mtu = x->type->get_mtu(x, xdst->child_mtu_cached);
 223
 224                 if (skb->len <= mtu)
 225                         goto ok;
 226
 227                 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
 228                         goto ok;
 229         }
 230
 231         return false;
 232
 233 ok:
 234         if (dev && dev->xfrmdev_ops && dev->xfrmdev_ops->xdo_dev_offload_ok)
 235                 return x->xso.dev->xfrmdev_ops->xdo_dev_offload_ok(skb, x);
 236
 237         return true;
 238 }
 239 EXPORT_SYMBOL_GPL(xfrm_dev_offload_ok);
 240
 241 void xfrm_dev_resume(struct sk_buff *skb)
 242 {
 243         struct net_device *dev = skb->dev;
 244         int ret = NETDEV_TX_BUSY;
 245         struct netdev_queue *txq;
 246         struct softnet_data *sd;
 247         unsigned long flags;
 248
 249         rcu_read_lock();
 250         txq = netdev_pick_tx(dev, skb, NULL);
 251
 252         HARD_TX_LOCK(dev, txq, smp_processor_id());
 253         if (!netif_xmit_frozen_or_stopped(txq))
 254                 skb = dev_hard_start_xmit(skb, dev, txq, &ret);
 255         HARD_TX_UNLOCK(dev, txq);
 256
 257         if (!dev_xmit_complete(ret)) {
 258                 local_irq_save(flags);
 259                 sd = this_cpu_ptr(&softnet_data);
 260                 skb_queue_tail(&sd->xfrm_backlog, skb);
 261                 raise_softirq_irqoff(NET_TX_SOFTIRQ);
 262                 local_irq_restore(flags);
 263         }
 264         rcu_read_unlock();
 265 }
 266 EXPORT_SYMBOL_GPL(xfrm_dev_resume);
 267
 268 void xfrm_dev_backlog(struct softnet_data *sd)
 269 {
 270         struct sk_buff_head *xfrm_backlog = &sd->xfrm_backlog;
 271         struct sk_buff_head list;
 272         struct sk_buff *skb;
 273
 274         if (skb_queue_empty(xfrm_backlog))
 275                 return;
 276
 277         __skb_queue_head_init(&list);
 278
 279         spin_lock(&xfrm_backlog->lock);
 280         skb_queue_splice_init(xfrm_backlog, &list);
 281         spin_unlock(&xfrm_backlog->lock);
 282
 283         while (!skb_queue_empty(&list)) {
 284                 skb = __skb_dequeue(&list);
 285                 xfrm_dev_resume(skb);
 286         }
 287
 288 }
 289 #endif
 290
 291 static int xfrm_api_check(struct net_device *dev)
 292 {
 293 #ifdef CONFIG_XFRM_OFFLOAD
 294         if ((dev->features & NETIF_F_HW_ESP_TX_CSUM) &&
 295             !(dev->features & NETIF_F_HW_ESP))
 296                 return NOTIFY_BAD;
 297
 298         if ((dev->features & NETIF_F_HW_ESP) &&
 299             (!(dev->xfrmdev_ops &&
 300                dev->xfrmdev_ops->xdo_dev_state_add &&
 301                dev->xfrmdev_ops->xdo_dev_state_delete)))
 302                 return NOTIFY_BAD;
 303 #else
 304         if (dev->features & (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM))
 305                 return NOTIFY_BAD;
 306 #endif
 307
 308         return NOTIFY_DONE;
 309 }
 310
 311 static int xfrm_dev_register(struct net_device *dev)
 312 {
 313         return xfrm_api_check(dev);
 314 }
 315
 316 static int xfrm_dev_feat_change(struct net_device *dev)
 317 {
 318         return xfrm_api_check(dev);
 319 }
 320
 321 static int xfrm_dev_down(struct net_device *dev)
 322 {
 323         if (dev->features & NETIF_F_HW_ESP)
 324                 xfrm_dev_state_flush(dev_net(dev), dev, true);
 325
 326         return NOTIFY_DONE;
 327 }
 328
 329 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
 330 {
 331         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 332
 333         switch (event) {
 334         case NETDEV_REGISTER:
 335                 return xfrm_dev_register(dev);
 336
 337         case NETDEV_FEAT_CHANGE:
 338                 return xfrm_dev_feat_change(dev);
 339
 340         case NETDEV_DOWN:
 341         case NETDEV_UNREGISTER:
 342                 return xfrm_dev_down(dev);
 343         }
 344         return NOTIFY_DONE;
 345 }
 346
 347 static struct notifier_block xfrm_dev_notifier = {
 348         .notifier_call  = xfrm_dev_event,
 349 };
 350
 351 void __init xfrm_dev_init(void)
 352 {
 353         register_netdevice_notifier(&xfrm_dev_notifier);
 354 }