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;
  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         skb2 = skb;
 152
 153         do {
 154                 struct sk_buff *nskb = skb2->next;
 155
 156                 esp_features |= skb->dev->gso_partial_features;
 157                 skb_mark_not_on_list(skb2);
 158
 159                 xo = xfrm_offload(skb2);
 160                 xo->flags |= XFRM_DEV_RESUME;
 161
 162                 xfrm_outer_mode_prep(x, skb2);
 163
 164                 err = x->type_offload->xmit(x, skb2, esp_features);
 165                 if (!err) {
 166                         skb2->next = nskb;
 167                 } else if (err != -EINPROGRESS) {
 168                         XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
 169                         skb2->next = nskb;
 170                         kfree_skb_list(skb2);
 171                         return NULL;
 172                 } else {
 173                         if (skb == skb2)
 174                                 skb = nskb;
 175
 176                         if (!skb)
 177                                 return NULL;
 178
 179                         goto skip_push;
 180                 }
 181
 182                 skb_push(skb2, skb2->data - skb_mac_header(skb2));
 183
 184 skip_push:
 185                 skb2 = nskb;
 186         } while (skb2);
 187
 188         return skb;
 189 }
 190 EXPORT_SYMBOL_GPL(validate_xmit_xfrm);
 191
 192 int xfrm_dev_state_add(struct net *net, struct xfrm_state *x,
 193                        struct xfrm_user_offload *xuo)
 194 {
 195         int err;
 196         struct dst_entry *dst;
 197         struct net_device *dev;
 198         struct xfrm_state_offload *xso = &x->xso;
 199         xfrm_address_t *saddr;
 200         xfrm_address_t *daddr;
 201
 202         if (!x->type_offload)
 203                 return -EINVAL;
 204
 205         /* We don't yet support UDP encapsulation and TFC padding. */
 206         if (x->encap || x->tfcpad)
 207                 return -EINVAL;
 208
 209         dev = dev_get_by_index(net, xuo->ifindex);
 210         if (!dev) {
 211                 if (!(xuo->flags & XFRM_OFFLOAD_INBOUND)) {
 212                         saddr = &x->props.saddr;
 213                         daddr = &x->id.daddr;
 214                 } else {
 215                         saddr = &x->id.daddr;
 216                         daddr = &x->props.saddr;
 217                 }
 218
 219                 dst = __xfrm_dst_lookup(net, 0, 0, saddr, daddr,
 220                                         x->props.family,
 221                                         xfrm_smark_get(0, x));
 222                 if (IS_ERR(dst))
 223                         return 0;
 224
 225                 dev = dst->dev;
 226
 227                 dev_hold(dev);
 228                 dst_release(dst);
 229         }
 230
 231         if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_state_add) {
 232                 xso->dev = NULL;
 233                 dev_put(dev);
 234                 return 0;
 235         }
 236
 237         if (x->props.flags & XFRM_STATE_ESN &&
 238             !dev->xfrmdev_ops->xdo_dev_state_advance_esn) {
 239                 xso->dev = NULL;
 240                 dev_put(dev);
 241                 return -EINVAL;
 242         }
 243
 244         xso->dev = dev;
 245         xso->num_exthdrs = 1;
 246         xso->flags = xuo->flags;
 247
 248         err = dev->xfrmdev_ops->xdo_dev_state_add(x);
 249         if (err) {
 250                 xso->num_exthdrs = 0;
 251                 xso->flags = 0;
 252                 xso->dev = NULL;
 253                 dev_put(dev);
 254
 255                 if (err != -EOPNOTSUPP)
 256                         return err;
 257         }
 258
 259         return 0;
 260 }
 261 EXPORT_SYMBOL_GPL(xfrm_dev_state_add);
 262
 263 bool xfrm_dev_offload_ok(struct sk_buff *skb, struct xfrm_state *x)
 264 {
 265         int mtu;
 266         struct dst_entry *dst = skb_dst(skb);
 267         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
 268         struct net_device *dev = x->xso.dev;
 269
 270         if (!x->type_offload || x->encap)
 271                 return false;
 272
 273         if ((!dev || (dev == xfrm_dst_path(dst)->dev)) &&
 274             (!xdst->child->xfrm && x->type->get_mtu)) {
 275                 mtu = x->type->get_mtu(x, xdst->child_mtu_cached);
 276
 277                 if (skb->len <= mtu)
 278                         goto ok;
 279
 280                 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
 281                         goto ok;
 282         }
 283
 284         return false;
 285
 286 ok:
 287         if (dev && dev->xfrmdev_ops && dev->xfrmdev_ops->xdo_dev_offload_ok)
 288                 return x->xso.dev->xfrmdev_ops->xdo_dev_offload_ok(skb, x);
 289
 290         return true;
 291 }
 292 EXPORT_SYMBOL_GPL(xfrm_dev_offload_ok);
 293
 294 void xfrm_dev_resume(struct sk_buff *skb)
 295 {
 296         struct net_device *dev = skb->dev;
 297         int ret = NETDEV_TX_BUSY;
 298         struct netdev_queue *txq;
 299         struct softnet_data *sd;
 300         unsigned long flags;
 301
 302         rcu_read_lock();
 303         txq = netdev_core_pick_tx(dev, skb, NULL);
 304
 305         HARD_TX_LOCK(dev, txq, smp_processor_id());
 306         if (!netif_xmit_frozen_or_stopped(txq))
 307                 skb = dev_hard_start_xmit(skb, dev, txq, &ret);
 308         HARD_TX_UNLOCK(dev, txq);
 309
 310         if (!dev_xmit_complete(ret)) {
 311                 local_irq_save(flags);
 312                 sd = this_cpu_ptr(&softnet_data);
 313                 skb_queue_tail(&sd->xfrm_backlog, skb);
 314                 raise_softirq_irqoff(NET_TX_SOFTIRQ);
 315                 local_irq_restore(flags);
 316         }
 317         rcu_read_unlock();
 318 }
 319 EXPORT_SYMBOL_GPL(xfrm_dev_resume);
 320
 321 void xfrm_dev_backlog(struct softnet_data *sd)
 322 {
 323         struct sk_buff_head *xfrm_backlog = &sd->xfrm_backlog;
 324         struct sk_buff_head list;
 325         struct sk_buff *skb;
 326
 327         if (skb_queue_empty(xfrm_backlog))
 328                 return;
 329
 330         __skb_queue_head_init(&list);
 331
 332         spin_lock(&xfrm_backlog->lock);
 333         skb_queue_splice_init(xfrm_backlog, &list);
 334         spin_unlock(&xfrm_backlog->lock);
 335
 336         while (!skb_queue_empty(&list)) {
 337                 skb = __skb_dequeue(&list);
 338                 xfrm_dev_resume(skb);
 339         }
 340
 341 }
 342 #endif
 343
 344 static int xfrm_api_check(struct net_device *dev)
 345 {
 346 #ifdef CONFIG_XFRM_OFFLOAD
 347         if ((dev->features & NETIF_F_HW_ESP_TX_CSUM) &&
 348             !(dev->features & NETIF_F_HW_ESP))
 349                 return NOTIFY_BAD;
 350
 351         if ((dev->features & NETIF_F_HW_ESP) &&
 352             (!(dev->xfrmdev_ops &&
 353                dev->xfrmdev_ops->xdo_dev_state_add &&
 354                dev->xfrmdev_ops->xdo_dev_state_delete)))
 355                 return NOTIFY_BAD;
 356 #else
 357         if (dev->features & (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM))
 358                 return NOTIFY_BAD;
 359 #endif
 360
 361         return NOTIFY_DONE;
 362 }
 363
 364 static int xfrm_dev_register(struct net_device *dev)
 365 {
 366         return xfrm_api_check(dev);
 367 }
 368
 369 static int xfrm_dev_feat_change(struct net_device *dev)
 370 {
 371         return xfrm_api_check(dev);
 372 }
 373
 374 static int xfrm_dev_down(struct net_device *dev)
 375 {
 376         if (dev->features & NETIF_F_HW_ESP)
 377                 xfrm_dev_state_flush(dev_net(dev), dev, true);
 378
 379         return NOTIFY_DONE;
 380 }
 381
 382 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
 383 {
 384         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 385
 386         switch (event) {
 387         case NETDEV_REGISTER:
 388                 return xfrm_dev_register(dev);
 389
 390         case NETDEV_FEAT_CHANGE:
 391                 return xfrm_dev_feat_change(dev);
 392
 393         case NETDEV_DOWN:
 394                 return xfrm_dev_down(dev);
 395         }
 396         return NOTIFY_DONE;
 397 }
 398
 399 static struct notifier_block xfrm_dev_notifier = {
 400         .notifier_call  = xfrm_dev_event,
 401 };
 402
 403 void __init xfrm_dev_init(void)
 404 {
 405         register_netdevice_notifier(&xfrm_dev_notifier);
 406 }