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 struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t features, bool *again)
 101 {
 102         int err;
 103         unsigned long flags;
 104         struct xfrm_state *x;
 105         struct softnet_data *sd;
 106         struct sk_buff *skb2, *nskb, *pskb = NULL;
 107         netdev_features_t esp_features = features;
 108         struct xfrm_offload *xo = xfrm_offload(skb);
 109         struct net_device *dev = skb->dev;
 110         struct sec_path *sp;
 111
 112         if (!xo || (xo->flags & XFRM_XMIT))
 113                 return skb;
 114
 115         if (!(features & NETIF_F_HW_ESP))
 116                 esp_features = features & ~(NETIF_F_SG | NETIF_F_CSUM_MASK);
 117
 118         sp = skb_sec_path(skb);
 119         x = sp->xvec[sp->len - 1];
 120         if (xo->flags & XFRM_GRO || x->xso.dir == XFRM_DEV_OFFLOAD_IN)
 121                 return skb;
 122
 123         /* This skb was already validated on the upper/virtual dev */
 124         if ((x->xso.dev != dev) && (x->xso.real_dev == dev))
 125                 return skb;
 126
 127         local_irq_save(flags);
 128         sd = this_cpu_ptr(&softnet_data);
 129         err = !skb_queue_empty(&sd->xfrm_backlog);
 130         local_irq_restore(flags);
 131
 132         if (err) {
 133                 *again = true;
 134                 return skb;
 135         }
 136
 137         if (skb_is_gso(skb) && unlikely(x->xso.dev != dev)) {
 138                 struct sk_buff *segs;
 139
 140                 /* Packet got rerouted, fixup features and segment it. */
 141                 esp_features = esp_features & ~(NETIF_F_HW_ESP | NETIF_F_GSO_ESP);
 142
 143                 segs = skb_gso_segment(skb, esp_features);
 144                 if (IS_ERR(segs)) {
 145                         kfree_skb(skb);
 146                         dev_core_stats_tx_dropped_inc(dev);
 147                         return NULL;
 148                 } else {
 149                         consume_skb(skb);
 150                         skb = segs;
 151                 }
 152         }
 153
 154         if (!skb->next) {
 155                 esp_features |= skb->dev->gso_partial_features;
 156                 xfrm_outer_mode_prep(x, skb);
 157
 158                 xo->flags |= XFRM_DEV_RESUME;
 159
 160                 err = x->type_offload->xmit(x, skb, esp_features);
 161                 if (err) {
 162                         if (err == -EINPROGRESS)
 163                                 return NULL;
 164
 165                         XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
 166                         kfree_skb(skb);
 167                         return NULL;
 168                 }
 169
 170                 skb_push(skb, skb->data - skb_mac_header(skb));
 171
 172                 return skb;
 173         }
 174
 175         skb_list_walk_safe(skb, skb2, nskb) {
 176                 esp_features |= skb->dev->gso_partial_features;
 177                 skb_mark_not_on_list(skb2);
 178
 179                 xo = xfrm_offload(skb2);
 180                 xo->flags |= XFRM_DEV_RESUME;
 181
 182                 xfrm_outer_mode_prep(x, skb2);
 183
 184                 err = x->type_offload->xmit(x, skb2, esp_features);
 185                 if (!err) {
 186                         skb2->next = nskb;
 187                 } else if (err != -EINPROGRESS) {
 188                         XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
 189                         skb2->next = nskb;
 190                         kfree_skb_list(skb2);
 191                         return NULL;
 192                 } else {
 193                         if (skb == skb2)
 194                                 skb = nskb;
 195                         else
 196                                 pskb->next = nskb;
 197
 198                         continue;
 199                 }
 200
 201                 skb_push(skb2, skb2->data - skb_mac_header(skb2));
 202                 pskb = skb2;
 203         }
 204
 205         return skb;
 206 }
 207 EXPORT_SYMBOL_GPL(validate_xmit_xfrm);
 208
 209 int xfrm_dev_state_add(struct net *net, struct xfrm_state *x,
 210                        struct xfrm_user_offload *xuo,
 211                        struct netlink_ext_ack *extack)
 212 {
 213         int err;
 214         struct dst_entry *dst;
 215         struct net_device *dev;
 216         struct xfrm_dev_offload *xso = &x->xso;
 217         xfrm_address_t *saddr;
 218         xfrm_address_t *daddr;
 219
 220         if (!x->type_offload) {
 221                 NL_SET_ERR_MSG(extack, "Type doesn't support offload");
 222                 return -EINVAL;
 223         }
 224
 225         /* We don't yet support UDP encapsulation and TFC padding. */
 226         if (x->encap || x->tfcpad) {
 227                 NL_SET_ERR_MSG(extack, "Encapsulation and TFC padding can't be offloaded");
 228                 return -EINVAL;
 229         }
 230
 231         if (xuo->flags & ~(XFRM_OFFLOAD_IPV6 | XFRM_OFFLOAD_INBOUND)) {
 232                 NL_SET_ERR_MSG(extack, "Unrecognized flags in offload request");
 233                 return -EINVAL;
 234         }
 235
 236         dev = dev_get_by_index(net, xuo->ifindex);
 237         if (!dev) {
 238                 if (!(xuo->flags & XFRM_OFFLOAD_INBOUND)) {
 239                         saddr = &x->props.saddr;
 240                         daddr = &x->id.daddr;
 241                 } else {
 242                         saddr = &x->id.daddr;
 243                         daddr = &x->props.saddr;
 244                 }
 245
 246                 dst = __xfrm_dst_lookup(net, 0, 0, saddr, daddr,
 247                                         x->props.family,
 248                                         xfrm_smark_get(0, x));
 249                 if (IS_ERR(dst))
 250                         return 0;
 251
 252                 dev = dst->dev;
 253
 254                 dev_hold(dev);
 255                 dst_release(dst);
 256         }
 257
 258         if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_state_add) {
 259                 xso->dev = NULL;
 260                 dev_put(dev);
 261                 return 0;
 262         }
 263
 264         if (x->props.flags & XFRM_STATE_ESN &&
 265             !dev->xfrmdev_ops->xdo_dev_state_advance_esn) {
 266                 NL_SET_ERR_MSG(extack, "Device doesn't support offload with ESN");
 267                 xso->dev = NULL;
 268                 dev_put(dev);
 269                 return -EINVAL;
 270         }
 271
 272         xso->dev = dev;
 273         netdev_tracker_alloc(dev, &xso->dev_tracker, GFP_ATOMIC);
 274         xso->real_dev = dev;
 275
 276         if (xuo->flags & XFRM_OFFLOAD_INBOUND)
 277                 xso->dir = XFRM_DEV_OFFLOAD_IN;
 278         else
 279                 xso->dir = XFRM_DEV_OFFLOAD_OUT;
 280
 281         err = dev->xfrmdev_ops->xdo_dev_state_add(x);
 282         if (err) {
 283                 xso->dev = NULL;
 284                 xso->dir = 0;
 285                 xso->real_dev = NULL;
 286                 netdev_put(dev, &xso->dev_tracker);
 287
 288                 if (err != -EOPNOTSUPP) {
 289                         NL_SET_ERR_MSG(extack, "Device failed to offload this state");
 290                         return err;
 291                 }
 292         }
 293
 294         return 0;
 295 }
 296 EXPORT_SYMBOL_GPL(xfrm_dev_state_add);
 297
 298 bool xfrm_dev_offload_ok(struct sk_buff *skb, struct xfrm_state *x)
 299 {
 300         int mtu;
 301         struct dst_entry *dst = skb_dst(skb);
 302         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
 303         struct net_device *dev = x->xso.dev;
 304
 305         if (!x->type_offload || x->encap)
 306                 return false;
 307
 308         if ((!dev || (dev == xfrm_dst_path(dst)->dev)) &&
 309             (!xdst->child->xfrm)) {
 310                 mtu = xfrm_state_mtu(x, xdst->child_mtu_cached);
 311                 if (skb->len <= mtu)
 312                         goto ok;
 313
 314                 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
 315                         goto ok;
 316         }
 317
 318         return false;
 319
 320 ok:
 321         if (dev && dev->xfrmdev_ops && dev->xfrmdev_ops->xdo_dev_offload_ok)
 322                 return x->xso.dev->xfrmdev_ops->xdo_dev_offload_ok(skb, x);
 323
 324         return true;
 325 }
 326 EXPORT_SYMBOL_GPL(xfrm_dev_offload_ok);
 327
 328 void xfrm_dev_resume(struct sk_buff *skb)
 329 {
 330         struct net_device *dev = skb->dev;
 331         int ret = NETDEV_TX_BUSY;
 332         struct netdev_queue *txq;
 333         struct softnet_data *sd;
 334         unsigned long flags;
 335
 336         rcu_read_lock();
 337         txq = netdev_core_pick_tx(dev, skb, NULL);
 338
 339         HARD_TX_LOCK(dev, txq, smp_processor_id());
 340         if (!netif_xmit_frozen_or_stopped(txq))
 341                 skb = dev_hard_start_xmit(skb, dev, txq, &ret);
 342         HARD_TX_UNLOCK(dev, txq);
 343
 344         if (!dev_xmit_complete(ret)) {
 345                 local_irq_save(flags);
 346                 sd = this_cpu_ptr(&softnet_data);
 347                 skb_queue_tail(&sd->xfrm_backlog, skb);
 348                 raise_softirq_irqoff(NET_TX_SOFTIRQ);
 349                 local_irq_restore(flags);
 350         }
 351         rcu_read_unlock();
 352 }
 353 EXPORT_SYMBOL_GPL(xfrm_dev_resume);
 354
 355 void xfrm_dev_backlog(struct softnet_data *sd)
 356 {
 357         struct sk_buff_head *xfrm_backlog = &sd->xfrm_backlog;
 358         struct sk_buff_head list;
 359         struct sk_buff *skb;
 360
 361         if (skb_queue_empty(xfrm_backlog))
 362                 return;
 363
 364         __skb_queue_head_init(&list);
 365
 366         spin_lock(&xfrm_backlog->lock);
 367         skb_queue_splice_init(xfrm_backlog, &list);
 368         spin_unlock(&xfrm_backlog->lock);
 369
 370         while (!skb_queue_empty(&list)) {
 371                 skb = __skb_dequeue(&list);
 372                 xfrm_dev_resume(skb);
 373         }
 374
 375 }
 376 #endif
 377
 378 static int xfrm_api_check(struct net_device *dev)
 379 {
 380 #ifdef CONFIG_XFRM_OFFLOAD
 381         if ((dev->features & NETIF_F_HW_ESP_TX_CSUM) &&
 382             !(dev->features & NETIF_F_HW_ESP))
 383                 return NOTIFY_BAD;
 384
 385         if ((dev->features & NETIF_F_HW_ESP) &&
 386             (!(dev->xfrmdev_ops &&
 387                dev->xfrmdev_ops->xdo_dev_state_add &&
 388                dev->xfrmdev_ops->xdo_dev_state_delete)))
 389                 return NOTIFY_BAD;
 390 #else
 391         if (dev->features & (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM))
 392                 return NOTIFY_BAD;
 393 #endif
 394
 395         return NOTIFY_DONE;
 396 }
 397
 398 static int xfrm_dev_down(struct net_device *dev)
 399 {
 400         if (dev->features & NETIF_F_HW_ESP)
 401                 xfrm_dev_state_flush(dev_net(dev), dev, true);
 402
 403         return NOTIFY_DONE;
 404 }
 405
 406 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
 407 {
 408         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 409
 410         switch (event) {
 411         case NETDEV_REGISTER:
 412                 return xfrm_api_check(dev);
 413
 414         case NETDEV_FEAT_CHANGE:
 415                 return xfrm_api_check(dev);
 416
 417         case NETDEV_DOWN:
 418         case NETDEV_UNREGISTER:
 419                 return xfrm_dev_down(dev);
 420         }
 421         return NOTIFY_DONE;
 422 }
 423
 424 static struct notifier_block xfrm_dev_notifier = {
 425         .notifier_call  = xfrm_dev_event,
 426 };
 427
 428 void __init xfrm_dev_init(void)
 429 {
 430         register_netdevice_notifier(&xfrm_dev_notifier);
 431 }