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