net/xfrm/xfrm_input.c

   1 /*
   2  * xfrm_input.c
   3  *
   4  * Changes:
   5  *      YOSHIFUJI Hideaki @USAGI
   6  *              Split up af-specific portion
   7  *
   8  */
   9
  10 #include <linux/slab.h>
  11 #include <linux/module.h>
  12 #include <linux/netdevice.h>
  13 #include <net/dst.h>
  14 #include <net/ip.h>
  15 #include <net/xfrm.h>
  16 #include <net/ip_tunnels.h>
  17 #include <net/ip6_tunnel.h>
  18
  19 static struct kmem_cache *secpath_cachep __read_mostly;
  20
  21 static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
  22 static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[AF_INET6 + 1];
  23
  24 static struct gro_cells gro_cells;
  25 static struct net_device xfrm_napi_dev;
  26
  27 int xfrm_input_register_afinfo(const struct xfrm_input_afinfo *afinfo)
  28 {
  29         int err = 0;
  30
  31         if (WARN_ON(afinfo->family >= ARRAY_SIZE(xfrm_input_afinfo)))
  32                 return -EAFNOSUPPORT;
  33
  34         spin_lock_bh(&xfrm_input_afinfo_lock);
  35         if (unlikely(xfrm_input_afinfo[afinfo->family] != NULL))
  36                 err = -EEXIST;
  37         else
  38                 rcu_assign_pointer(xfrm_input_afinfo[afinfo->family], afinfo);
  39         spin_unlock_bh(&xfrm_input_afinfo_lock);
  40         return err;
  41 }
  42 EXPORT_SYMBOL(xfrm_input_register_afinfo);
  43
  44 int xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo *afinfo)
  45 {
  46         int err = 0;
  47
  48         spin_lock_bh(&xfrm_input_afinfo_lock);
  49         if (likely(xfrm_input_afinfo[afinfo->family] != NULL)) {
  50                 if (unlikely(xfrm_input_afinfo[afinfo->family] != afinfo))
  51                         err = -EINVAL;
  52                 else
  53                         RCU_INIT_POINTER(xfrm_input_afinfo[afinfo->family], NULL);
  54         }
  55         spin_unlock_bh(&xfrm_input_afinfo_lock);
  56         synchronize_rcu();
  57         return err;
  58 }
  59 EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
  60
  61 static const struct xfrm_input_afinfo *xfrm_input_get_afinfo(unsigned int family)
  62 {
  63         const struct xfrm_input_afinfo *afinfo;
  64
  65         if (WARN_ON_ONCE(family >= ARRAY_SIZE(xfrm_input_afinfo)))
  66                 return NULL;
  67
  68         rcu_read_lock();
  69         afinfo = rcu_dereference(xfrm_input_afinfo[family]);
  70         if (unlikely(!afinfo))
  71                 rcu_read_unlock();
  72         return afinfo;
  73 }
  74
  75 static int xfrm_rcv_cb(struct sk_buff *skb, unsigned int family, u8 protocol,
  76                        int err)
  77 {
  78         int ret;
  79         const struct xfrm_input_afinfo *afinfo = xfrm_input_get_afinfo(family);
  80
  81         if (!afinfo)
  82                 return -EAFNOSUPPORT;
  83
  84         ret = afinfo->callback(skb, protocol, err);
  85         rcu_read_unlock();
  86
  87         return ret;
  88 }
  89
  90 void __secpath_destroy(struct sec_path *sp)
  91 {
  92         int i;
  93         for (i = 0; i < sp->len; i++)
  94                 xfrm_state_put(sp->xvec[i]);
  95         kmem_cache_free(secpath_cachep, sp);
  96 }
  97 EXPORT_SYMBOL(__secpath_destroy);
  98
  99 struct sec_path *secpath_dup(struct sec_path *src)
 100 {
 101         struct sec_path *sp;
 102
 103         sp = kmem_cache_alloc(secpath_cachep, GFP_ATOMIC);
 104         if (!sp)
 105                 return NULL;
 106
 107         sp->len = 0;
 108         sp->olen = 0;
 109
 110         if (src) {
 111                 int i;
 112
 113                 memcpy(sp, src, sizeof(*sp));
 114                 for (i = 0; i < sp->len; i++)
 115                         xfrm_state_hold(sp->xvec[i]);
 116         }
 117         atomic_set(&sp->refcnt, 1);
 118         return sp;
 119 }
 120 EXPORT_SYMBOL(secpath_dup);
 121
 122 int secpath_set(struct sk_buff *skb)
 123 {
 124         struct sec_path *sp;
 125
 126         /* Allocate new secpath or COW existing one. */
 127         if (!skb->sp || atomic_read(&skb->sp->refcnt) != 1) {
 128                 sp = secpath_dup(skb->sp);
 129                 if (!sp)
 130                         return -ENOMEM;
 131
 132                 if (skb->sp)
 133                         secpath_put(skb->sp);
 134                 skb->sp = sp;
 135         }
 136         return 0;
 137 }
 138 EXPORT_SYMBOL(secpath_set);
 139
 140 /* Fetch spi and seq from ipsec header */
 141
 142 int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
 143 {
 144         int offset, offset_seq;
 145         int hlen;
 146
 147         switch (nexthdr) {
 148         case IPPROTO_AH:
 149                 hlen = sizeof(struct ip_auth_hdr);
 150                 offset = offsetof(struct ip_auth_hdr, spi);
 151                 offset_seq = offsetof(struct ip_auth_hdr, seq_no);
 152                 break;
 153         case IPPROTO_ESP:
 154                 hlen = sizeof(struct ip_esp_hdr);
 155                 offset = offsetof(struct ip_esp_hdr, spi);
 156                 offset_seq = offsetof(struct ip_esp_hdr, seq_no);
 157                 break;
 158         case IPPROTO_COMP:
 159                 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
 160                         return -EINVAL;
 161                 *spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
 162                 *seq = 0;
 163                 return 0;
 164         default:
 165                 return 1;
 166         }
 167
 168         if (!pskb_may_pull(skb, hlen))
 169                 return -EINVAL;
 170
 171         *spi = *(__be32 *)(skb_transport_header(skb) + offset);
 172         *seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
 173         return 0;
 174 }
 175 EXPORT_SYMBOL(xfrm_parse_spi);
 176
 177 int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
 178 {
 179         struct xfrm_mode *inner_mode = x->inner_mode;
 180         int err;
 181
 182         err = x->outer_mode->afinfo->extract_input(x, skb);
 183         if (err)
 184                 return err;
 185
 186         if (x->sel.family == AF_UNSPEC) {
 187                 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
 188                 if (inner_mode == NULL)
 189                         return -EAFNOSUPPORT;
 190         }
 191
 192         skb->protocol = inner_mode->afinfo->eth_proto;
 193         return inner_mode->input2(x, skb);
 194 }
 195 EXPORT_SYMBOL(xfrm_prepare_input);
 196
 197 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
 198 {
 199         struct net *net = dev_net(skb->dev);
 200         int err;
 201         __be32 seq;
 202         __be32 seq_hi;
 203         struct xfrm_state *x = NULL;
 204         xfrm_address_t *daddr;
 205         struct xfrm_mode *inner_mode;
 206         u32 mark = skb->mark;
 207         unsigned int family;
 208         int decaps = 0;
 209         int async = 0;
 210         struct xfrm_offload *xo;
 211         bool xfrm_gro = false;
 212
 213         if (encap_type < 0) {
 214                 x = xfrm_input_state(skb);
 215                 family = x->outer_mode->afinfo->family;
 216
 217                 /* An encap_type of -1 indicates async resumption. */
 218                 if (encap_type == -1) {
 219                         async = 1;
 220                         seq = XFRM_SKB_CB(skb)->seq.input.low;
 221                         goto resume;
 222                 }
 223                 /* encap_type < -1 indicates a GRO call. */
 224                 encap_type = 0;
 225                 seq = XFRM_SPI_SKB_CB(skb)->seq;
 226                 goto lock;
 227         }
 228
 229         daddr = (xfrm_address_t *)(skb_network_header(skb) +
 230                                    XFRM_SPI_SKB_CB(skb)->daddroff);
 231         family = XFRM_SPI_SKB_CB(skb)->family;
 232
 233         /* if tunnel is present override skb->mark value with tunnel i_key */
 234         switch (family) {
 235         case AF_INET:
 236                 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
 237                         mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
 238                 break;
 239         case AF_INET6:
 240                 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
 241                         mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
 242                 break;
 243         }
 244
 245         err = secpath_set(skb);
 246         if (err) {
 247                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
 248                 goto drop;
 249         }
 250
 251         seq = 0;
 252         if (!spi && (err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
 253                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
 254                 goto drop;
 255         }
 256
 257         do {
 258                 if (skb->sp->len == XFRM_MAX_DEPTH) {
 259                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
 260                         goto drop;
 261                 }
 262
 263                 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
 264                 if (x == NULL) {
 265                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
 266                         xfrm_audit_state_notfound(skb, family, spi, seq);
 267                         goto drop;
 268                 }
 269
 270                 skb->sp->xvec[skb->sp->len++] = x;
 271
 272 lock:
 273                 spin_lock(&x->lock);
 274
 275                 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
 276                         if (x->km.state == XFRM_STATE_ACQ)
 277                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
 278                         else
 279                                 XFRM_INC_STATS(net,
 280                                                LINUX_MIB_XFRMINSTATEINVALID);
 281                         goto drop_unlock;
 282                 }
 283
 284                 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
 285                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
 286                         goto drop_unlock;
 287                 }
 288
 289                 if (x->repl->check(x, skb, seq)) {
 290                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
 291                         goto drop_unlock;
 292                 }
 293
 294                 if (xfrm_state_check_expire(x)) {
 295                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
 296                         goto drop_unlock;
 297                 }
 298
 299                 spin_unlock(&x->lock);
 300
 301                 if (xfrm_tunnel_check(skb, x, family)) {
 302                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
 303                         goto drop;
 304                 }
 305
 306                 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
 307
 308                 XFRM_SKB_CB(skb)->seq.input.low = seq;
 309                 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
 310
 311                 skb_dst_force(skb);
 312                 dev_hold(skb->dev);
 313
 314                 nexthdr = x->type->input(x, skb);
 315
 316                 if (nexthdr == -EINPROGRESS)
 317                         return 0;
 318 resume:
 319                 dev_put(skb->dev);
 320
 321                 spin_lock(&x->lock);
 322                 if (nexthdr <= 0) {
 323                         if (nexthdr == -EBADMSG) {
 324                                 xfrm_audit_state_icvfail(x, skb,
 325                                                          x->type->proto);
 326                                 x->stats.integrity_failed++;
 327                         }
 328                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
 329                         goto drop_unlock;
 330                 }
 331
 332                 /* only the first xfrm gets the encap type */
 333                 encap_type = 0;
 334
 335                 if (async && x->repl->recheck(x, skb, seq)) {
 336                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
 337                         goto drop_unlock;
 338                 }
 339
 340                 x->repl->advance(x, seq);
 341
 342                 x->curlft.bytes += skb->len;
 343                 x->curlft.packets++;
 344
 345                 spin_unlock(&x->lock);
 346
 347                 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
 348
 349                 inner_mode = x->inner_mode;
 350
 351                 if (x->sel.family == AF_UNSPEC) {
 352                         inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
 353                         if (inner_mode == NULL) {
 354                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
 355                                 goto drop;
 356                         }
 357                 }
 358
 359                 if (inner_mode->input(x, skb)) {
 360                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
 361                         goto drop;
 362                 }
 363
 364                 if (x->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) {
 365                         decaps = 1;
 366                         break;
 367                 }
 368
 369                 /*
 370                  * We need the inner address.  However, we only get here for
 371                  * transport mode so the outer address is identical.
 372                  */
 373                 daddr = &x->id.daddr;
 374                 family = x->outer_mode->afinfo->family;
 375
 376                 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
 377                 if (err < 0) {
 378                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
 379                         goto drop;
 380                 }
 381         } while (!err);
 382
 383         err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
 384         if (err)
 385                 goto drop;
 386
 387         nf_reset(skb);
 388
 389         if (decaps) {
 390                 skb_dst_drop(skb);
 391                 gro_cells_receive(&gro_cells, skb);
 392                 return 0;
 393         } else {
 394                 xo = xfrm_offload(skb);
 395                 if (xo)
 396                         xfrm_gro = xo->flags & XFRM_GRO;
 397
 398                 err = x->inner_mode->afinfo->transport_finish(skb, async);
 399                 if (xfrm_gro) {
 400                         skb_dst_drop(skb);
 401                         gro_cells_receive(&gro_cells, skb);
 402                         return err;
 403                 }
 404
 405                 return err;
 406         }
 407
 408 drop_unlock:
 409         spin_unlock(&x->lock);
 410 drop:
 411         xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
 412         kfree_skb(skb);
 413         return 0;
 414 }
 415 EXPORT_SYMBOL(xfrm_input);
 416
 417 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
 418 {
 419         return xfrm_input(skb, nexthdr, 0, -1);
 420 }
 421 EXPORT_SYMBOL(xfrm_input_resume);
 422
 423 void __init xfrm_input_init(void)
 424 {
 425         int err;
 426
 427         init_dummy_netdev(&xfrm_napi_dev);
 428         err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
 429         if (err)
 430                 gro_cells.cells = NULL;
 431
 432         secpath_cachep = kmem_cache_create("secpath_cache",
 433                                            sizeof(struct sec_path),
 434                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
 435                                            NULL);
 436 }