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         memset(sp->ovec, 0, sizeof(sp->ovec[XFRM_MAX_OFFLOAD_DEPTH]));
 111
 112         if (src) {
 113                 int i;
 114
 115                 memcpy(sp, src, sizeof(*sp));
 116                 for (i = 0; i < sp->len; i++)
 117                         xfrm_state_hold(sp->xvec[i]);
 118         }
 119         refcount_set(&sp->refcnt, 1);
 120         return sp;
 121 }
 122 EXPORT_SYMBOL(secpath_dup);
 123
 124 int secpath_set(struct sk_buff *skb)
 125 {
 126         struct sec_path *sp;
 127
 128         /* Allocate new secpath or COW existing one. */
 129         if (!skb->sp || refcount_read(&skb->sp->refcnt) != 1) {
 130                 sp = secpath_dup(skb->sp);
 131                 if (!sp)
 132                         return -ENOMEM;
 133
 134                 if (skb->sp)
 135                         secpath_put(skb->sp);
 136                 skb->sp = sp;
 137         }
 138         return 0;
 139 }
 140 EXPORT_SYMBOL(secpath_set);
 141
 142 /* Fetch spi and seq from ipsec header */
 143
 144 int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
 145 {
 146         int offset, offset_seq;
 147         int hlen;
 148
 149         switch (nexthdr) {
 150         case IPPROTO_AH:
 151                 hlen = sizeof(struct ip_auth_hdr);
 152                 offset = offsetof(struct ip_auth_hdr, spi);
 153                 offset_seq = offsetof(struct ip_auth_hdr, seq_no);
 154                 break;
 155         case IPPROTO_ESP:
 156                 hlen = sizeof(struct ip_esp_hdr);
 157                 offset = offsetof(struct ip_esp_hdr, spi);
 158                 offset_seq = offsetof(struct ip_esp_hdr, seq_no);
 159                 break;
 160         case IPPROTO_COMP:
 161                 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
 162                         return -EINVAL;
 163                 *spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
 164                 *seq = 0;
 165                 return 0;
 166         default:
 167                 return 1;
 168         }
 169
 170         if (!pskb_may_pull(skb, hlen))
 171                 return -EINVAL;
 172
 173         *spi = *(__be32 *)(skb_transport_header(skb) + offset);
 174         *seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
 175         return 0;
 176 }
 177 EXPORT_SYMBOL(xfrm_parse_spi);
 178
 179 int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
 180 {
 181         struct xfrm_mode *inner_mode = x->inner_mode;
 182         int err;
 183
 184         err = x->outer_mode->afinfo->extract_input(x, skb);
 185         if (err)
 186                 return err;
 187
 188         if (x->sel.family == AF_UNSPEC) {
 189                 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
 190                 if (inner_mode == NULL)
 191                         return -EAFNOSUPPORT;
 192         }
 193
 194         skb->protocol = inner_mode->afinfo->eth_proto;
 195         return inner_mode->input2(x, skb);
 196 }
 197 EXPORT_SYMBOL(xfrm_prepare_input);
 198
 199 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
 200 {
 201         struct net *net = dev_net(skb->dev);
 202         int err;
 203         __be32 seq;
 204         __be32 seq_hi;
 205         struct xfrm_state *x = NULL;
 206         xfrm_address_t *daddr;
 207         struct xfrm_mode *inner_mode;
 208         u32 mark = skb->mark;
 209         unsigned int family;
 210         int decaps = 0;
 211         int async = 0;
 212         bool xfrm_gro = false;
 213         bool crypto_done = false;
 214         struct xfrm_offload *xo = xfrm_offload(skb);
 215
 216         if (encap_type < 0) {
 217                 x = xfrm_input_state(skb);
 218                 family = x->outer_mode->afinfo->family;
 219
 220                 /* An encap_type of -1 indicates async resumption. */
 221                 if (encap_type == -1) {
 222                         async = 1;
 223                         seq = XFRM_SKB_CB(skb)->seq.input.low;
 224                         goto resume;
 225                 }
 226
 227                 /* encap_type < -1 indicates a GRO call. */
 228                 encap_type = 0;
 229                 seq = XFRM_SPI_SKB_CB(skb)->seq;
 230
 231                 if (xo && (xo->flags & CRYPTO_DONE)) {
 232                         crypto_done = true;
 233                         x = xfrm_input_state(skb);
 234                         family = XFRM_SPI_SKB_CB(skb)->family;
 235
 236                         if (!(xo->status & CRYPTO_SUCCESS)) {
 237                                 if (xo->status &
 238                                     (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
 239                                      CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
 240                                      CRYPTO_TUNNEL_AH_AUTH_FAILED |
 241                                      CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
 242
 243                                         xfrm_audit_state_icvfail(x, skb,
 244                                                                  x->type->proto);
 245                                         x->stats.integrity_failed++;
 246                                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
 247                                         goto drop;
 248                                 }
 249
 250                                 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
 251                                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
 252                                         goto drop;
 253                                 }
 254
 255                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
 256                                 goto drop;
 257                         }
 258
 259                         if ((err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
 260                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
 261                                 goto drop;
 262                         }
 263                 }
 264
 265                 goto lock;
 266         }
 267
 268         daddr = (xfrm_address_t *)(skb_network_header(skb) +
 269                                    XFRM_SPI_SKB_CB(skb)->daddroff);
 270         family = XFRM_SPI_SKB_CB(skb)->family;
 271
 272         /* if tunnel is present override skb->mark value with tunnel i_key */
 273         switch (family) {
 274         case AF_INET:
 275                 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
 276                         mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
 277                 break;
 278         case AF_INET6:
 279                 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
 280                         mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
 281                 break;
 282         }
 283
 284         err = secpath_set(skb);
 285         if (err) {
 286                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
 287                 goto drop;
 288         }
 289
 290         seq = 0;
 291         if (!spi && (err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
 292                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
 293                 goto drop;
 294         }
 295
 296         do {
 297                 if (skb->sp->len == XFRM_MAX_DEPTH) {
 298                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
 299                         goto drop;
 300                 }
 301
 302                 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
 303                 if (x == NULL) {
 304                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
 305                         xfrm_audit_state_notfound(skb, family, spi, seq);
 306                         goto drop;
 307                 }
 308
 309                 skb->sp->xvec[skb->sp->len++] = x;
 310
 311 lock:
 312                 spin_lock(&x->lock);
 313
 314                 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
 315                         if (x->km.state == XFRM_STATE_ACQ)
 316                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
 317                         else
 318                                 XFRM_INC_STATS(net,
 319                                                LINUX_MIB_XFRMINSTATEINVALID);
 320                         goto drop_unlock;
 321                 }
 322
 323                 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
 324                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
 325                         goto drop_unlock;
 326                 }
 327
 328                 if (x->repl->check(x, skb, seq)) {
 329                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
 330                         goto drop_unlock;
 331                 }
 332
 333                 if (xfrm_state_check_expire(x)) {
 334                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
 335                         goto drop_unlock;
 336                 }
 337
 338                 spin_unlock(&x->lock);
 339
 340                 if (xfrm_tunnel_check(skb, x, family)) {
 341                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
 342                         goto drop;
 343                 }
 344
 345                 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
 346
 347                 XFRM_SKB_CB(skb)->seq.input.low = seq;
 348                 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
 349
 350                 skb_dst_force(skb);
 351                 dev_hold(skb->dev);
 352
 353                 if (crypto_done)
 354                         nexthdr = x->type_offload->input_tail(x, skb);
 355                 else
 356                         nexthdr = x->type->input(x, skb);
 357
 358                 if (nexthdr == -EINPROGRESS)
 359                         return 0;
 360 resume:
 361                 dev_put(skb->dev);
 362
 363                 spin_lock(&x->lock);
 364                 if (nexthdr <= 0) {
 365                         if (nexthdr == -EBADMSG) {
 366                                 xfrm_audit_state_icvfail(x, skb,
 367                                                          x->type->proto);
 368                                 x->stats.integrity_failed++;
 369                         }
 370                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
 371                         goto drop_unlock;
 372                 }
 373
 374                 /* only the first xfrm gets the encap type */
 375                 encap_type = 0;
 376
 377                 if (async && x->repl->recheck(x, skb, seq)) {
 378                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
 379                         goto drop_unlock;
 380                 }
 381
 382                 x->repl->advance(x, seq);
 383
 384                 x->curlft.bytes += skb->len;
 385                 x->curlft.packets++;
 386
 387                 spin_unlock(&x->lock);
 388
 389                 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
 390
 391                 inner_mode = x->inner_mode;
 392
 393                 if (x->sel.family == AF_UNSPEC) {
 394                         inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
 395                         if (inner_mode == NULL) {
 396                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
 397                                 goto drop;
 398                         }
 399                 }
 400
 401                 if (inner_mode->input(x, skb)) {
 402                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
 403                         goto drop;
 404                 }
 405
 406                 if (x->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) {
 407                         decaps = 1;
 408                         break;
 409                 }
 410
 411                 /*
 412                  * We need the inner address.  However, we only get here for
 413                  * transport mode so the outer address is identical.
 414                  */
 415                 daddr = &x->id.daddr;
 416                 family = x->outer_mode->afinfo->family;
 417
 418                 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
 419                 if (err < 0) {
 420                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
 421                         goto drop;
 422                 }
 423         } while (!err);
 424
 425         err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
 426         if (err)
 427                 goto drop;
 428
 429         nf_reset(skb);
 430
 431         if (decaps) {
 432                 skb->sp->olen = 0;
 433                 skb_dst_drop(skb);
 434                 gro_cells_receive(&gro_cells, skb);
 435                 return 0;
 436         } else {
 437                 xo = xfrm_offload(skb);
 438                 if (xo)
 439                         xfrm_gro = xo->flags & XFRM_GRO;
 440
 441                 err = x->inner_mode->afinfo->transport_finish(skb, xfrm_gro || async);
 442                 if (xfrm_gro) {
 443                         skb->sp->olen = 0;
 444                         skb_dst_drop(skb);
 445                         gro_cells_receive(&gro_cells, skb);
 446                         return err;
 447                 }
 448
 449                 return err;
 450         }
 451
 452 drop_unlock:
 453         spin_unlock(&x->lock);
 454 drop:
 455         xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
 456         kfree_skb(skb);
 457         return 0;
 458 }
 459 EXPORT_SYMBOL(xfrm_input);
 460
 461 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
 462 {
 463         return xfrm_input(skb, nexthdr, 0, -1);
 464 }
 465 EXPORT_SYMBOL(xfrm_input_resume);
 466
 467 void __init xfrm_input_init(void)
 468 {
 469         int err;
 470
 471         init_dummy_netdev(&xfrm_napi_dev);
 472         err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
 473         if (err)
 474                 gro_cells.cells = NULL;
 475
 476         secpath_cachep = kmem_cache_create("secpath_cache",
 477                                            sizeof(struct sec_path),
 478                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
 479                                            NULL);
 480 }