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)
  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         if (skb_is_gso(skb)) {
  57                 struct net_device *dev = skb->dev;
  58
  59                 if (unlikely(x->xso.dev != dev)) {
  60                         struct sk_buff *segs;
  61
  62                         /* Packet got rerouted, fixup features and segment it. */
  63                         esp_features = esp_features & ~(NETIF_F_HW_ESP
  64                                                         | NETIF_F_GSO_ESP);
  65
  66                         segs = skb_gso_segment(skb, esp_features);
  67                         if (IS_ERR(segs)) {
  68                                 kfree_skb(skb);
  69                                 atomic_long_inc(&dev->tx_dropped);
  70                                 return NULL;
  71                         } else {
  72                                 consume_skb(skb);
  73                                 skb = segs;
  74                         }
  75                 }
  76         }
  77
  78         if (!skb->next) {
  79                 x->outer_mode->xmit(x, skb);
  80
  81                 xo->flags |= XFRM_DEV_RESUME;
  82
  83                 err = x->type_offload->xmit(x, skb, esp_features);
  84                 if (err) {
  85                         if (err == -EINPROGRESS)
  86                                 return NULL;
  87
  88                         XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
  89                         kfree_skb(skb);
  90                         return NULL;
  91                 }
  92
  93                 skb_push(skb, skb->data - skb_mac_header(skb));
  94
  95                 return skb;
  96         }
  97
  98         skb2 = skb;
  99
 100         do {
 101                 struct sk_buff *nskb = skb2->next;
 102                 skb_mark_not_on_list(skb2);
 103
 104                 xo = xfrm_offload(skb2);
 105                 xo->flags |= XFRM_DEV_RESUME;
 106
 107                 x->outer_mode->xmit(x, skb2);
 108
 109                 err = x->type_offload->xmit(x, skb2, esp_features);
 110                 if (!err) {
 111                         skb2->next = nskb;
 112                 } else if (err != -EINPROGRESS) {
 113                         XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
 114                         skb2->next = nskb;
 115                         kfree_skb_list(skb2);
 116                         return NULL;
 117                 } else {
 118                         if (skb == skb2)
 119                                 skb = nskb;
 120
 121                         if (!skb)
 122                                 return NULL;
 123
 124                         goto skip_push;
 125                 }
 126
 127                 skb_push(skb2, skb2->data - skb_mac_header(skb2));
 128
 129 skip_push:
 130                 skb2 = nskb;
 131         } while (skb2);
 132
 133         return skb;
 134 }
 135 EXPORT_SYMBOL_GPL(validate_xmit_xfrm);
 136
 137 int xfrm_dev_state_add(struct net *net, struct xfrm_state *x,
 138                        struct xfrm_user_offload *xuo)
 139 {
 140         int err;
 141         struct dst_entry *dst;
 142         struct net_device *dev;
 143         struct xfrm_state_offload *xso = &x->xso;
 144         xfrm_address_t *saddr;
 145         xfrm_address_t *daddr;
 146
 147         if (!x->type_offload)
 148                 return -EINVAL;
 149
 150         /* We don't yet support UDP encapsulation and TFC padding. */
 151         if (x->encap || x->tfcpad)
 152                 return -EINVAL;
 153
 154         dev = dev_get_by_index(net, xuo->ifindex);
 155         if (!dev) {
 156                 if (!(xuo->flags & XFRM_OFFLOAD_INBOUND)) {
 157                         saddr = &x->props.saddr;
 158                         daddr = &x->id.daddr;
 159                 } else {
 160                         saddr = &x->id.daddr;
 161                         daddr = &x->props.saddr;
 162                 }
 163
 164                 dst = __xfrm_dst_lookup(net, 0, 0, saddr, daddr,
 165                                         x->props.family,
 166                                         xfrm_smark_get(0, x));
 167                 if (IS_ERR(dst))
 168                         return 0;
 169
 170                 dev = dst->dev;
 171
 172                 dev_hold(dev);
 173                 dst_release(dst);
 174         }
 175
 176         if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_state_add) {
 177                 xso->dev = NULL;
 178                 dev_put(dev);
 179                 return 0;
 180         }
 181
 182         if (x->props.flags & XFRM_STATE_ESN &&
 183             !dev->xfrmdev_ops->xdo_dev_state_advance_esn) {
 184                 xso->dev = NULL;
 185                 dev_put(dev);
 186                 return -EINVAL;
 187         }
 188
 189         xso->dev = dev;
 190         xso->num_exthdrs = 1;
 191         xso->flags = xuo->flags;
 192
 193         err = dev->xfrmdev_ops->xdo_dev_state_add(x);
 194         if (err) {
 195                 xso->num_exthdrs = 0;
 196                 xso->flags = 0;
 197                 xso->dev = NULL;
 198                 dev_put(dev);
 199
 200                 if (err != -EOPNOTSUPP)
 201                         return err;
 202         }
 203
 204         return 0;
 205 }
 206 EXPORT_SYMBOL_GPL(xfrm_dev_state_add);
 207
 208 bool xfrm_dev_offload_ok(struct sk_buff *skb, struct xfrm_state *x)
 209 {
 210         int mtu;
 211         struct dst_entry *dst = skb_dst(skb);
 212         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
 213         struct net_device *dev = x->xso.dev;
 214
 215         if (!x->type_offload || x->encap)
 216                 return false;
 217
 218         if ((!dev || (dev == xfrm_dst_path(dst)->dev)) &&
 219             (!xdst->child->xfrm && x->type->get_mtu)) {
 220                 mtu = x->type->get_mtu(x, xdst->child_mtu_cached);
 221
 222                 if (skb->len <= mtu)
 223                         goto ok;
 224
 225                 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
 226                         goto ok;
 227         }
 228
 229         return false;
 230
 231 ok:
 232         if (dev && dev->xfrmdev_ops && dev->xfrmdev_ops->xdo_dev_offload_ok)
 233                 return x->xso.dev->xfrmdev_ops->xdo_dev_offload_ok(skb, x);
 234
 235         return true;
 236 }
 237 EXPORT_SYMBOL_GPL(xfrm_dev_offload_ok);
 238
 239 void xfrm_dev_resume(struct sk_buff *skb)
 240 {
 241         struct net_device *dev = skb->dev;
 242         int ret = NETDEV_TX_BUSY;
 243         struct netdev_queue *txq;
 244         struct softnet_data *sd;
 245         unsigned long flags;
 246
 247         rcu_read_lock();
 248         txq = netdev_pick_tx(dev, skb, NULL);
 249
 250         HARD_TX_LOCK(dev, txq, smp_processor_id());
 251         if (!netif_xmit_frozen_or_stopped(txq))
 252                 skb = dev_hard_start_xmit(skb, dev, txq, &ret);
 253         HARD_TX_UNLOCK(dev, txq);
 254
 255         if (!dev_xmit_complete(ret)) {
 256                 local_irq_save(flags);
 257                 sd = this_cpu_ptr(&softnet_data);
 258                 skb_queue_tail(&sd->xfrm_backlog, skb);
 259                 raise_softirq_irqoff(NET_TX_SOFTIRQ);
 260                 local_irq_restore(flags);
 261         }
 262         rcu_read_unlock();
 263 }
 264 EXPORT_SYMBOL_GPL(xfrm_dev_resume);
 265
 266 void xfrm_dev_backlog(struct softnet_data *sd)
 267 {
 268         struct sk_buff_head *xfrm_backlog = &sd->xfrm_backlog;
 269         struct sk_buff_head list;
 270         struct sk_buff *skb;
 271
 272         if (skb_queue_empty(xfrm_backlog))
 273                 return;
 274
 275         __skb_queue_head_init(&list);
 276
 277         spin_lock(&xfrm_backlog->lock);
 278         skb_queue_splice_init(xfrm_backlog, &list);
 279         spin_unlock(&xfrm_backlog->lock);
 280
 281         while (!skb_queue_empty(&list)) {
 282                 skb = __skb_dequeue(&list);
 283                 xfrm_dev_resume(skb);
 284         }
 285
 286 }
 287 #endif
 288
 289 static int xfrm_api_check(struct net_device *dev)
 290 {
 291 #ifdef CONFIG_XFRM_OFFLOAD
 292         if ((dev->features & NETIF_F_HW_ESP_TX_CSUM) &&
 293             !(dev->features & NETIF_F_HW_ESP))
 294                 return NOTIFY_BAD;
 295
 296         if ((dev->features & NETIF_F_HW_ESP) &&
 297             (!(dev->xfrmdev_ops &&
 298                dev->xfrmdev_ops->xdo_dev_state_add &&
 299                dev->xfrmdev_ops->xdo_dev_state_delete)))
 300                 return NOTIFY_BAD;
 301 #else
 302         if (dev->features & (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM))
 303                 return NOTIFY_BAD;
 304 #endif
 305
 306         return NOTIFY_DONE;
 307 }
 308
 309 static int xfrm_dev_register(struct net_device *dev)
 310 {
 311         return xfrm_api_check(dev);
 312 }
 313
 314 static int xfrm_dev_feat_change(struct net_device *dev)
 315 {
 316         return xfrm_api_check(dev);
 317 }
 318
 319 static int xfrm_dev_down(struct net_device *dev)
 320 {
 321         if (dev->features & NETIF_F_HW_ESP)
 322                 xfrm_dev_state_flush(dev_net(dev), dev, true);
 323
 324         return NOTIFY_DONE;
 325 }
 326
 327 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
 328 {
 329         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 330
 331         switch (event) {
 332         case NETDEV_REGISTER:
 333                 return xfrm_dev_register(dev);
 334
 335         case NETDEV_FEAT_CHANGE:
 336                 return xfrm_dev_feat_change(dev);
 337
 338         case NETDEV_DOWN:
 339                 return xfrm_dev_down(dev);
 340         }
 341         return NOTIFY_DONE;
 342 }
 343
 344 static struct notifier_block xfrm_dev_notifier = {
 345         .notifier_call  = xfrm_dev_event,
 346 };
 347
 348 void __init xfrm_dev_init(void)
 349 {
 350         register_netdevice_notifier(&xfrm_dev_notifier);
 351 }