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 static void __xfrm_transport_prep(struct xfrm_state *x, struct sk_buff *skb,
  27                                   unsigned int hsize)
  28 {
  29         struct xfrm_offload *xo = xfrm_offload(skb);
  30
  31         skb_reset_mac_len(skb);
  32         pskb_pull(skb, skb->mac_len + hsize + x->props.header_len);
  33
  34         if (xo->flags & XFRM_GSO_SEGMENT) {
  35                 skb_reset_transport_header(skb);
  36                 skb->transport_header -= x->props.header_len;
  37         }
  38 }
  39
  40 static void __xfrm_mode_tunnel_prep(struct xfrm_state *x, struct sk_buff *skb,
  41                                     unsigned int hsize)
  42
  43 {
  44         struct xfrm_offload *xo = xfrm_offload(skb);
  45
  46         if (xo->flags & XFRM_GSO_SEGMENT)
  47                 skb->transport_header = skb->network_header + hsize;
  48
  49         skb_reset_mac_len(skb);
  50         pskb_pull(skb, skb->mac_len + x->props.header_len);
  51 }
  52
  53 /* Adjust pointers into the packet when IPsec is done at layer2 */
  54 static void xfrm_outer_mode_prep(struct xfrm_state *x, struct sk_buff *skb)
  55 {
  56         switch (x->outer_mode.encap) {
  57         case XFRM_MODE_TUNNEL:
  58                 if (x->outer_mode.family == AF_INET)
  59                         return __xfrm_mode_tunnel_prep(x, skb,
  60                                                        sizeof(struct iphdr));
  61                 if (x->outer_mode.family == AF_INET6)
  62                         return __xfrm_mode_tunnel_prep(x, skb,
  63                                                        sizeof(struct ipv6hdr));
  64                 break;
  65         case XFRM_MODE_TRANSPORT:
  66                 if (x->outer_mode.family == AF_INET)
  67                         return __xfrm_transport_prep(x, skb,
  68                                                      sizeof(struct iphdr));
  69                 if (x->outer_mode.family == AF_INET6)
  70                         return __xfrm_transport_prep(x, skb,
  71                                                      sizeof(struct ipv6hdr));
  72                 break;
  73         case XFRM_MODE_ROUTEOPTIMIZATION:
  74         case XFRM_MODE_IN_TRIGGER:
  75         case XFRM_MODE_BEET:
  76                 break;
  77         }
  78 }
  79
  80 struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t features, bool *again)
  81 {
  82         int err;
  83         unsigned long flags;
  84         struct xfrm_state *x;
  85         struct sk_buff *skb2;
  86         struct softnet_data *sd;
  87         netdev_features_t esp_features = features;
  88         struct xfrm_offload *xo = xfrm_offload(skb);
  89         struct sec_path *sp;
  90
  91         if (!xo)
  92                 return skb;
  93
  94         if (!(features & NETIF_F_HW_ESP))
  95                 esp_features = features & ~(NETIF_F_SG | NETIF_F_CSUM_MASK);
  96
  97         sp = skb_sec_path(skb);
  98         x = sp->xvec[sp->len - 1];
  99         if (xo->flags & XFRM_GRO || x->xso.flags & XFRM_OFFLOAD_INBOUND)
 100                 return skb;
 101
 102         local_irq_save(flags);
 103         sd = this_cpu_ptr(&softnet_data);
 104         err = !skb_queue_empty(&sd->xfrm_backlog);
 105         local_irq_restore(flags);
 106
 107         if (err) {
 108                 *again = true;
 109                 return skb;
 110         }
 111
 112         if (skb_is_gso(skb)) {
 113                 struct net_device *dev = skb->dev;
 114
 115                 if (unlikely(x->xso.dev != dev)) {
 116                         struct sk_buff *segs;
 117
 118                         /* Packet got rerouted, fixup features and segment it. */
 119                         esp_features = esp_features & ~(NETIF_F_HW_ESP
 120                                                         | NETIF_F_GSO_ESP);
 121
 122                         segs = skb_gso_segment(skb, esp_features);
 123                         if (IS_ERR(segs)) {
 124                                 kfree_skb(skb);
 125                                 atomic_long_inc(&dev->tx_dropped);
 126                                 return NULL;
 127                         } else {
 128                                 consume_skb(skb);
 129                                 skb = segs;
 130                         }
 131                 }
 132         }
 133
 134         if (!skb->next) {
 135                 esp_features |= skb->dev->gso_partial_features;
 136                 xfrm_outer_mode_prep(x, skb);
 137
 138                 xo->flags |= XFRM_DEV_RESUME;
 139
 140                 err = x->type_offload->xmit(x, skb, esp_features);
 141                 if (err) {
 142                         if (err == -EINPROGRESS)
 143                                 return NULL;
 144
 145                         XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
 146                         kfree_skb(skb);
 147                         return NULL;
 148                 }
 149
 150                 skb_push(skb, skb->data - skb_mac_header(skb));
 151
 152                 return skb;
 153         }
 154
 155         skb2 = skb;
 156
 157         do {
 158                 struct sk_buff *nskb = skb2->next;
 159
 160                 esp_features |= skb->dev->gso_partial_features;
 161                 skb_mark_not_on_list(skb2);
 162
 163                 xo = xfrm_offload(skb2);
 164                 xo->flags |= XFRM_DEV_RESUME;
 165
 166                 xfrm_outer_mode_prep(x, skb2);
 167
 168                 err = x->type_offload->xmit(x, skb2, esp_features);
 169                 if (!err) {
 170                         skb2->next = nskb;
 171                 } else if (err != -EINPROGRESS) {
 172                         XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
 173                         skb2->next = nskb;
 174                         kfree_skb_list(skb2);
 175                         return NULL;
 176                 } else {
 177                         if (skb == skb2)
 178                                 skb = nskb;
 179
 180                         if (!skb)
 181                                 return NULL;
 182
 183                         goto skip_push;
 184                 }
 185
 186                 skb_push(skb2, skb2->data - skb_mac_header(skb2));
 187
 188 skip_push:
 189                 skb2 = nskb;
 190         } while (skb2);
 191
 192         return skb;
 193 }
 194 EXPORT_SYMBOL_GPL(validate_xmit_xfrm);
 195
 196 int xfrm_dev_state_add(struct net *net, struct xfrm_state *x,
 197                        struct xfrm_user_offload *xuo)
 198 {
 199         int err;
 200         struct dst_entry *dst;
 201         struct net_device *dev;
 202         struct xfrm_state_offload *xso = &x->xso;
 203         xfrm_address_t *saddr;
 204         xfrm_address_t *daddr;
 205
 206         if (!x->type_offload)
 207                 return -EINVAL;
 208
 209         /* We don't yet support UDP encapsulation and TFC padding. */
 210         if (x->encap || x->tfcpad)
 211                 return -EINVAL;
 212
 213         dev = dev_get_by_index(net, xuo->ifindex);
 214         if (!dev) {
 215                 if (!(xuo->flags & XFRM_OFFLOAD_INBOUND)) {
 216                         saddr = &x->props.saddr;
 217                         daddr = &x->id.daddr;
 218                 } else {
 219                         saddr = &x->id.daddr;
 220                         daddr = &x->props.saddr;
 221                 }
 222
 223                 dst = __xfrm_dst_lookup(net, 0, 0, saddr, daddr,
 224                                         x->props.family,
 225                                         xfrm_smark_get(0, x));
 226                 if (IS_ERR(dst))
 227                         return 0;
 228
 229                 dev = dst->dev;
 230
 231                 dev_hold(dev);
 232                 dst_release(dst);
 233         }
 234
 235         if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_state_add) {
 236                 xso->dev = NULL;
 237                 dev_put(dev);
 238                 return 0;
 239         }
 240
 241         if (x->props.flags & XFRM_STATE_ESN &&
 242             !dev->xfrmdev_ops->xdo_dev_state_advance_esn) {
 243                 xso->dev = NULL;
 244                 dev_put(dev);
 245                 return -EINVAL;
 246         }
 247
 248         xso->dev = dev;
 249         xso->num_exthdrs = 1;
 250         xso->flags = xuo->flags;
 251
 252         err = dev->xfrmdev_ops->xdo_dev_state_add(x);
 253         if (err) {
 254                 xso->num_exthdrs = 0;
 255                 xso->flags = 0;
 256                 xso->dev = NULL;
 257                 dev_put(dev);
 258
 259                 if (err != -EOPNOTSUPP)
 260                         return err;
 261         }
 262
 263         return 0;
 264 }
 265 EXPORT_SYMBOL_GPL(xfrm_dev_state_add);
 266
 267 bool xfrm_dev_offload_ok(struct sk_buff *skb, struct xfrm_state *x)
 268 {
 269         int mtu;
 270         struct dst_entry *dst = skb_dst(skb);
 271         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
 272         struct net_device *dev = x->xso.dev;
 273
 274         if (!x->type_offload || x->encap)
 275                 return false;
 276
 277         if ((!dev || (dev == xfrm_dst_path(dst)->dev)) &&
 278             (!xdst->child->xfrm && x->type->get_mtu)) {
 279                 mtu = x->type->get_mtu(x, xdst->child_mtu_cached);
 280
 281                 if (skb->len <= mtu)
 282                         goto ok;
 283
 284                 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
 285                         goto ok;
 286         }
 287
 288         return false;
 289
 290 ok:
 291         if (dev && dev->xfrmdev_ops && dev->xfrmdev_ops->xdo_dev_offload_ok)
 292                 return x->xso.dev->xfrmdev_ops->xdo_dev_offload_ok(skb, x);
 293
 294         return true;
 295 }
 296 EXPORT_SYMBOL_GPL(xfrm_dev_offload_ok);
 297
 298 void xfrm_dev_resume(struct sk_buff *skb)
 299 {
 300         struct net_device *dev = skb->dev;
 301         int ret = NETDEV_TX_BUSY;
 302         struct netdev_queue *txq;
 303         struct softnet_data *sd;
 304         unsigned long flags;
 305
 306         rcu_read_lock();
 307         txq = netdev_core_pick_tx(dev, skb, NULL);
 308
 309         HARD_TX_LOCK(dev, txq, smp_processor_id());
 310         if (!netif_xmit_frozen_or_stopped(txq))
 311                 skb = dev_hard_start_xmit(skb, dev, txq, &ret);
 312         HARD_TX_UNLOCK(dev, txq);
 313
 314         if (!dev_xmit_complete(ret)) {
 315                 local_irq_save(flags);
 316                 sd = this_cpu_ptr(&softnet_data);
 317                 skb_queue_tail(&sd->xfrm_backlog, skb);
 318                 raise_softirq_irqoff(NET_TX_SOFTIRQ);
 319                 local_irq_restore(flags);
 320         }
 321         rcu_read_unlock();
 322 }
 323 EXPORT_SYMBOL_GPL(xfrm_dev_resume);
 324
 325 void xfrm_dev_backlog(struct softnet_data *sd)
 326 {
 327         struct sk_buff_head *xfrm_backlog = &sd->xfrm_backlog;
 328         struct sk_buff_head list;
 329         struct sk_buff *skb;
 330
 331         if (skb_queue_empty(xfrm_backlog))
 332                 return;
 333
 334         __skb_queue_head_init(&list);
 335
 336         spin_lock(&xfrm_backlog->lock);
 337         skb_queue_splice_init(xfrm_backlog, &list);
 338         spin_unlock(&xfrm_backlog->lock);
 339
 340         while (!skb_queue_empty(&list)) {
 341                 skb = __skb_dequeue(&list);
 342                 xfrm_dev_resume(skb);
 343         }
 344
 345 }
 346 #endif
 347
 348 static int xfrm_api_check(struct net_device *dev)
 349 {
 350 #ifdef CONFIG_XFRM_OFFLOAD
 351         if ((dev->features & NETIF_F_HW_ESP_TX_CSUM) &&
 352             !(dev->features & NETIF_F_HW_ESP))
 353                 return NOTIFY_BAD;
 354
 355         if ((dev->features & NETIF_F_HW_ESP) &&
 356             (!(dev->xfrmdev_ops &&
 357                dev->xfrmdev_ops->xdo_dev_state_add &&
 358                dev->xfrmdev_ops->xdo_dev_state_delete)))
 359                 return NOTIFY_BAD;
 360 #else
 361         if (dev->features & (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM))
 362                 return NOTIFY_BAD;
 363 #endif
 364
 365         return NOTIFY_DONE;
 366 }
 367
 368 static int xfrm_dev_register(struct net_device *dev)
 369 {
 370         return xfrm_api_check(dev);
 371 }
 372
 373 static int xfrm_dev_feat_change(struct net_device *dev)
 374 {
 375         return xfrm_api_check(dev);
 376 }
 377
 378 static int xfrm_dev_down(struct net_device *dev)
 379 {
 380         if (dev->features & NETIF_F_HW_ESP)
 381                 xfrm_dev_state_flush(dev_net(dev), dev, true);
 382
 383         return NOTIFY_DONE;
 384 }
 385
 386 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
 387 {
 388         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 389
 390         switch (event) {
 391         case NETDEV_REGISTER:
 392                 return xfrm_dev_register(dev);
 393
 394         case NETDEV_FEAT_CHANGE:
 395                 return xfrm_dev_feat_change(dev);
 396
 397         case NETDEV_DOWN:
 398                 return xfrm_dev_down(dev);
 399         }
 400         return NOTIFY_DONE;
 401 }
 402
 403 static struct notifier_block xfrm_dev_notifier = {
 404         .notifier_call  = xfrm_dev_event,
 405 };
 406
 407 void __init xfrm_dev_init(void)
 408 {
 409         register_netdevice_notifier(&xfrm_dev_notifier);
 410 }