net/xfrm/xfrm_device.c

   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/xfrm.h>
  19 #include <linux/notifier.h>
  20
  21 #ifdef CONFIG_XFRM_OFFLOAD
  22 static void __xfrm_transport_prep(struct xfrm_state *x, struct sk_buff *skb,
  23                                   unsigned int hsize)
  24 {
  25         struct xfrm_offload *xo = xfrm_offload(skb);
  26
  27         skb_reset_mac_len(skb);
  28         if (xo->flags & XFRM_GSO_SEGMENT)
  29                 skb->transport_header -= x->props.header_len;
  30
  31         pskb_pull(skb, skb_transport_offset(skb) + x->props.header_len);
  32 }
  33
  34 static void __xfrm_mode_tunnel_prep(struct xfrm_state *x, struct sk_buff *skb,
  35                                     unsigned int hsize)
  36
  37 {
  38         struct xfrm_offload *xo = xfrm_offload(skb);
  39
  40         if (xo->flags & XFRM_GSO_SEGMENT)
  41                 skb->transport_header = skb->network_header + hsize;
  42
  43         skb_reset_mac_len(skb);
  44         pskb_pull(skb, skb->mac_len + x->props.header_len);
  45 }
  46
  47 static void __xfrm_mode_beet_prep(struct xfrm_state *x, struct sk_buff *skb,
  48                                   unsigned int hsize)
  49 {
  50         struct xfrm_offload *xo = xfrm_offload(skb);
  51         int phlen = 0;
  52
  53         if (xo->flags & XFRM_GSO_SEGMENT)
  54                 skb->transport_header = skb->network_header + hsize;
  55
  56         skb_reset_mac_len(skb);
  57         if (x->sel.family != AF_INET6) {
  58                 phlen = IPV4_BEET_PHMAXLEN;
  59                 if (x->outer_mode.family == AF_INET6)
  60                         phlen += sizeof(struct ipv6hdr) - sizeof(struct iphdr);
  61         }
  62
  63         pskb_pull(skb, skb->mac_len + hsize + (x->props.header_len - phlen));
  64 }
  65
  66 /* Adjust pointers into the packet when IPsec is done at layer2 */
  67 static void xfrm_outer_mode_prep(struct xfrm_state *x, struct sk_buff *skb)
  68 {
  69         switch (x->outer_mode.encap) {
  70         case XFRM_MODE_TUNNEL:
  71                 if (x->outer_mode.family == AF_INET)
  72                         return __xfrm_mode_tunnel_prep(x, skb,
  73                                                        sizeof(struct iphdr));
  74                 if (x->outer_mode.family == AF_INET6)
  75                         return __xfrm_mode_tunnel_prep(x, skb,
  76                                                        sizeof(struct ipv6hdr));
  77                 break;
  78         case XFRM_MODE_TRANSPORT:
  79                 if (x->outer_mode.family == AF_INET)
  80                         return __xfrm_transport_prep(x, skb,
  81                                                      sizeof(struct iphdr));
  82                 if (x->outer_mode.family == AF_INET6)
  83                         return __xfrm_transport_prep(x, skb,
  84                                                      sizeof(struct ipv6hdr));
  85                 break;
  86         case XFRM_MODE_BEET:
  87                 if (x->outer_mode.family == AF_INET)
  88                         return __xfrm_mode_beet_prep(x, skb,
  89                                                      sizeof(struct iphdr));
  90                 if (x->outer_mode.family == AF_INET6)
  91                         return __xfrm_mode_beet_prep(x, skb,
  92                                                      sizeof(struct ipv6hdr));
  93                 break;
  94         case XFRM_MODE_ROUTEOPTIMIZATION:
  95         case XFRM_MODE_IN_TRIGGER:
  96                 break;
  97         }
  98 }
  99
 100 static inline bool xmit_xfrm_check_overflow(struct sk_buff *skb)
 101 {
 102         struct xfrm_offload *xo = xfrm_offload(skb);
 103         __u32 seq = xo->seq.low;
 104
 105         seq += skb_shinfo(skb)->gso_segs;
 106         if (unlikely(seq < xo->seq.low))
 107                 return true;
 108
 109         return false;
 110 }
 111
 112 struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t features, bool *again)
 113 {
 114         int err;
 115         unsigned long flags;
 116         struct xfrm_state *x;
 117         struct softnet_data *sd;
 118         struct sk_buff *skb2, *nskb, *pskb = NULL;
 119         netdev_features_t esp_features = features;
 120         struct xfrm_offload *xo = xfrm_offload(skb);
 121         struct net_device *dev = skb->dev;
 122         struct sec_path *sp;
 123
 124         if (!xo || (xo->flags & XFRM_XMIT))
 125                 return skb;
 126
 127         if (!(features & NETIF_F_HW_ESP))
 128                 esp_features = features & ~(NETIF_F_SG | NETIF_F_CSUM_MASK);
 129
 130         sp = skb_sec_path(skb);
 131         x = sp->xvec[sp->len - 1];
 132         if (xo->flags & XFRM_GRO || x->xso.flags & XFRM_OFFLOAD_INBOUND)
 133                 return skb;
 134
 135         /* This skb was already validated on the upper/virtual dev */
 136         if ((x->xso.dev != dev) && (x->xso.real_dev == dev))
 137                 return skb;
 138
 139         local_irq_save(flags);
 140         sd = this_cpu_ptr(&softnet_data);
 141         err = !skb_queue_empty(&sd->xfrm_backlog);
 142         local_irq_restore(flags);
 143
 144         if (err) {
 145                 *again = true;
 146                 return skb;
 147         }
 148
 149         if (skb_is_gso(skb) && (unlikely(x->xso.dev != dev) ||
 150                                 unlikely(xmit_xfrm_check_overflow(skb)))) {
 151                 struct sk_buff *segs;
 152
 153                 /* Packet got rerouted, fixup features and segment it. */
 154                 esp_features = esp_features & ~(NETIF_F_HW_ESP | NETIF_F_GSO_ESP);
 155
 156                 segs = skb_gso_segment(skb, esp_features);
 157                 if (IS_ERR(segs)) {
 158                         kfree_skb(skb);
 159                         atomic_long_inc(&dev->tx_dropped);
 160                         return NULL;
 161                 } else {
 162                         consume_skb(skb);
 163                         skb = segs;
 164                 }
 165         }
 166
 167         if (!skb->next) {
 168                 esp_features |= skb->dev->gso_partial_features;
 169                 xfrm_outer_mode_prep(x, skb);
 170
 171                 xo->flags |= XFRM_DEV_RESUME;
 172
 173                 err = x->type_offload->xmit(x, skb, esp_features);
 174                 if (err) {
 175                         if (err == -EINPROGRESS)
 176                                 return NULL;
 177
 178                         XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
 179                         kfree_skb(skb);
 180                         return NULL;
 181                 }
 182
 183                 skb_push(skb, skb->data - skb_mac_header(skb));
 184
 185                 return skb;
 186         }
 187
 188         skb_list_walk_safe(skb, skb2, nskb) {
 189                 esp_features |= skb->dev->gso_partial_features;
 190                 skb_mark_not_on_list(skb2);
 191
 192                 xo = xfrm_offload(skb2);
 193                 xo->flags |= XFRM_DEV_RESUME;
 194
 195                 xfrm_outer_mode_prep(x, skb2);
 196
 197                 err = x->type_offload->xmit(x, skb2, esp_features);
 198                 if (!err) {
 199                         skb2->next = nskb;
 200                 } else if (err != -EINPROGRESS) {
 201                         XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
 202                         skb2->next = nskb;
 203                         kfree_skb_list(skb2);
 204                         return NULL;
 205                 } else {
 206                         if (skb == skb2)
 207                                 skb = nskb;
 208                         else
 209                                 pskb->next = nskb;
 210
 211                         continue;
 212                 }
 213
 214                 skb_push(skb2, skb2->data - skb_mac_header(skb2));
 215                 pskb = skb2;
 216         }
 217
 218         return skb;
 219 }
 220 EXPORT_SYMBOL_GPL(validate_xmit_xfrm);
 221
 222 int xfrm_dev_state_add(struct net *net, struct xfrm_state *x,
 223                        struct xfrm_user_offload *xuo)
 224 {
 225         int err;
 226         struct dst_entry *dst;
 227         struct net_device *dev;
 228         struct xfrm_state_offload *xso = &x->xso;
 229         xfrm_address_t *saddr;
 230         xfrm_address_t *daddr;
 231
 232         if (!x->type_offload)
 233                 return -EINVAL;
 234
 235         /* We don't yet support UDP encapsulation and TFC padding. */
 236         if (x->encap || x->tfcpad)
 237                 return -EINVAL;
 238
 239         if (xuo->flags & ~(XFRM_OFFLOAD_IPV6 | XFRM_OFFLOAD_INBOUND))
 240                 return -EINVAL;
 241
 242         dev = dev_get_by_index(net, xuo->ifindex);
 243         if (!dev) {
 244                 if (!(xuo->flags & XFRM_OFFLOAD_INBOUND)) {
 245                         saddr = &x->props.saddr;
 246                         daddr = &x->id.daddr;
 247                 } else {
 248                         saddr = &x->id.daddr;
 249                         daddr = &x->props.saddr;
 250                 }
 251
 252                 dst = __xfrm_dst_lookup(net, 0, 0, saddr, daddr,
 253                                         x->props.family,
 254                                         xfrm_smark_get(0, x));
 255                 if (IS_ERR(dst))
 256                         return 0;
 257
 258                 dev = dst->dev;
 259
 260                 dev_hold(dev);
 261                 dst_release(dst);
 262         }
 263
 264         if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_state_add) {
 265                 xso->dev = NULL;
 266                 dev_put(dev);
 267                 return 0;
 268         }
 269
 270         if (x->props.flags & XFRM_STATE_ESN &&
 271             !dev->xfrmdev_ops->xdo_dev_state_advance_esn) {
 272                 xso->dev = NULL;
 273                 dev_put(dev);
 274                 return -EINVAL;
 275         }
 276
 277         xso->dev = dev;
 278         xso->real_dev = dev;
 279         xso->num_exthdrs = 1;
 280         /* Don't forward bit that is not implemented */
 281         xso->flags = xuo->flags & ~XFRM_OFFLOAD_IPV6;
 282
 283         err = dev->xfrmdev_ops->xdo_dev_state_add(x);
 284         if (err) {
 285                 xso->num_exthdrs = 0;
 286                 xso->flags = 0;
 287                 xso->dev = NULL;
 288                 xso->real_dev = NULL;
 289                 dev_put(dev);
 290
 291                 if (err != -EOPNOTSUPP)
 292                         return err;
 293         }
 294
 295         return 0;
 296 }
 297 EXPORT_SYMBOL_GPL(xfrm_dev_state_add);
 298
 299 bool xfrm_dev_offload_ok(struct sk_buff *skb, struct xfrm_state *x)
 300 {
 301         int mtu;
 302         struct dst_entry *dst = skb_dst(skb);
 303         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
 304         struct net_device *dev = x->xso.dev;
 305
 306         if (!x->type_offload || x->encap)
 307                 return false;
 308
 309         if ((!dev || (dev == xfrm_dst_path(dst)->dev)) &&
 310             (!xdst->child->xfrm)) {
 311                 mtu = xfrm_state_mtu(x, xdst->child_mtu_cached);
 312                 if (skb->len <= mtu)
 313                         goto ok;
 314
 315                 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
 316                         goto ok;
 317         }
 318
 319         return false;
 320
 321 ok:
 322         if (dev && dev->xfrmdev_ops && dev->xfrmdev_ops->xdo_dev_offload_ok)
 323                 return x->xso.dev->xfrmdev_ops->xdo_dev_offload_ok(skb, x);
 324
 325         return true;
 326 }
 327 EXPORT_SYMBOL_GPL(xfrm_dev_offload_ok);
 328
 329 void xfrm_dev_resume(struct sk_buff *skb)
 330 {
 331         struct net_device *dev = skb->dev;
 332         int ret = NETDEV_TX_BUSY;
 333         struct netdev_queue *txq;
 334         struct softnet_data *sd;
 335         unsigned long flags;
 336
 337         rcu_read_lock();
 338         txq = netdev_core_pick_tx(dev, skb, NULL);
 339
 340         HARD_TX_LOCK(dev, txq, smp_processor_id());
 341         if (!netif_xmit_frozen_or_stopped(txq))
 342                 skb = dev_hard_start_xmit(skb, dev, txq, &ret);
 343         HARD_TX_UNLOCK(dev, txq);
 344
 345         if (!dev_xmit_complete(ret)) {
 346                 local_irq_save(flags);
 347                 sd = this_cpu_ptr(&softnet_data);
 348                 skb_queue_tail(&sd->xfrm_backlog, skb);
 349                 raise_softirq_irqoff(NET_TX_SOFTIRQ);
 350                 local_irq_restore(flags);
 351         }
 352         rcu_read_unlock();
 353 }
 354 EXPORT_SYMBOL_GPL(xfrm_dev_resume);
 355
 356 void xfrm_dev_backlog(struct softnet_data *sd)
 357 {
 358         struct sk_buff_head *xfrm_backlog = &sd->xfrm_backlog;
 359         struct sk_buff_head list;
 360         struct sk_buff *skb;
 361
 362         if (skb_queue_empty(xfrm_backlog))
 363                 return;
 364
 365         __skb_queue_head_init(&list);
 366
 367         spin_lock(&xfrm_backlog->lock);
 368         skb_queue_splice_init(xfrm_backlog, &list);
 369         spin_unlock(&xfrm_backlog->lock);
 370
 371         while (!skb_queue_empty(&list)) {
 372                 skb = __skb_dequeue(&list);
 373                 xfrm_dev_resume(skb);
 374         }
 375
 376 }
 377 #endif
 378
 379 static int xfrm_api_check(struct net_device *dev)
 380 {
 381 #ifdef CONFIG_XFRM_OFFLOAD
 382         if ((dev->features & NETIF_F_HW_ESP_TX_CSUM) &&
 383             !(dev->features & NETIF_F_HW_ESP))
 384                 return NOTIFY_BAD;
 385
 386         if ((dev->features & NETIF_F_HW_ESP) &&
 387             (!(dev->xfrmdev_ops &&
 388                dev->xfrmdev_ops->xdo_dev_state_add &&
 389                dev->xfrmdev_ops->xdo_dev_state_delete)))
 390                 return NOTIFY_BAD;
 391 #else
 392         if (dev->features & (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM))
 393                 return NOTIFY_BAD;
 394 #endif
 395
 396         return NOTIFY_DONE;
 397 }
 398
 399 static int xfrm_dev_register(struct net_device *dev)
 400 {
 401         return xfrm_api_check(dev);
 402 }
 403
 404 static int xfrm_dev_feat_change(struct net_device *dev)
 405 {
 406         return xfrm_api_check(dev);
 407 }
 408
 409 static int xfrm_dev_down(struct net_device *dev)
 410 {
 411         if (dev->features & NETIF_F_HW_ESP)
 412                 xfrm_dev_state_flush(dev_net(dev), dev, true);
 413
 414         return NOTIFY_DONE;
 415 }
 416
 417 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
 418 {
 419         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 420
 421         switch (event) {
 422         case NETDEV_REGISTER:
 423                 return xfrm_dev_register(dev);
 424
 425         case NETDEV_FEAT_CHANGE:
 426                 return xfrm_dev_feat_change(dev);
 427
 428         case NETDEV_DOWN:
 429         case NETDEV_UNREGISTER:
 430                 return xfrm_dev_down(dev);
 431         }
 432         return NOTIFY_DONE;
 433 }
 434
 435 static struct notifier_block xfrm_dev_notifier = {
 436         .notifier_call  = xfrm_dev_event,
 437 };
 438
 439 void __init xfrm_dev_init(void)
 440 {
 441         register_netdevice_notifier(&xfrm_dev_notifier);
 442 }