net/xfrm/xfrm_input.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * xfrm_input.c
   4  *
   5  * Changes:
   6  *      YOSHIFUJI Hideaki @USAGI
   7  *              Split up af-specific portion
   8  *
   9  */
  10
  11 #include <linux/slab.h>
  12 #include <linux/module.h>
  13 #include <linux/netdevice.h>
  14 #include <net/dst.h>
  15 #include <net/ip.h>
  16 #include <net/xfrm.h>
  17 #include <net/ip_tunnels.h>
  18 #include <net/ip6_tunnel.h>
  19
  20 static struct kmem_cache *secpath_cachep __read_mostly;
  21
  22 static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
  23 static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[AF_INET6 + 1];
  24
  25 static struct gro_cells gro_cells;
  26 static struct net_device xfrm_napi_dev;
  27
  28 int xfrm_input_register_afinfo(const struct xfrm_input_afinfo *afinfo)
  29 {
  30         int err = 0;
  31
  32         if (WARN_ON(afinfo->family >= ARRAY_SIZE(xfrm_input_afinfo)))
  33                 return -EAFNOSUPPORT;
  34
  35         spin_lock_bh(&xfrm_input_afinfo_lock);
  36         if (unlikely(xfrm_input_afinfo[afinfo->family] != NULL))
  37                 err = -EEXIST;
  38         else
  39                 rcu_assign_pointer(xfrm_input_afinfo[afinfo->family], afinfo);
  40         spin_unlock_bh(&xfrm_input_afinfo_lock);
  41         return err;
  42 }
  43 EXPORT_SYMBOL(xfrm_input_register_afinfo);
  44
  45 int xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo *afinfo)
  46 {
  47         int err = 0;
  48
  49         spin_lock_bh(&xfrm_input_afinfo_lock);
  50         if (likely(xfrm_input_afinfo[afinfo->family] != NULL)) {
  51                 if (unlikely(xfrm_input_afinfo[afinfo->family] != afinfo))
  52                         err = -EINVAL;
  53                 else
  54                         RCU_INIT_POINTER(xfrm_input_afinfo[afinfo->family], NULL);
  55         }
  56         spin_unlock_bh(&xfrm_input_afinfo_lock);
  57         synchronize_rcu();
  58         return err;
  59 }
  60 EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
  61
  62 static const struct xfrm_input_afinfo *xfrm_input_get_afinfo(unsigned int family)
  63 {
  64         const struct xfrm_input_afinfo *afinfo;
  65
  66         if (WARN_ON_ONCE(family >= ARRAY_SIZE(xfrm_input_afinfo)))
  67                 return NULL;
  68
  69         rcu_read_lock();
  70         afinfo = rcu_dereference(xfrm_input_afinfo[family]);
  71         if (unlikely(!afinfo))
  72                 rcu_read_unlock();
  73         return afinfo;
  74 }
  75
  76 static int xfrm_rcv_cb(struct sk_buff *skb, unsigned int family, u8 protocol,
  77                        int err)
  78 {
  79         int ret;
  80         const struct xfrm_input_afinfo *afinfo = xfrm_input_get_afinfo(family);
  81
  82         if (!afinfo)
  83                 return -EAFNOSUPPORT;
  84
  85         ret = afinfo->callback(skb, protocol, err);
  86         rcu_read_unlock();
  87
  88         return ret;
  89 }
  90
  91 void __secpath_destroy(struct sec_path *sp)
  92 {
  93         int i;
  94         for (i = 0; i < sp->len; i++)
  95                 xfrm_state_put(sp->xvec[i]);
  96         kmem_cache_free(secpath_cachep, sp);
  97 }
  98 EXPORT_SYMBOL(__secpath_destroy);
  99
 100 struct sec_path *secpath_dup(struct sec_path *src)
 101 {
 102         struct sec_path *sp;
 103
 104         sp = kmem_cache_alloc(secpath_cachep, GFP_ATOMIC);
 105         if (!sp)
 106                 return NULL;
 107
 108         sp->len = 0;
 109         sp->olen = 0;
 110
 111         memset(sp->ovec, 0, sizeof(sp->ovec[XFRM_MAX_OFFLOAD_DEPTH]));
 112
 113         if (src) {
 114                 int i;
 115
 116                 memcpy(sp, src, sizeof(*sp));
 117                 for (i = 0; i < sp->len; i++)
 118                         xfrm_state_hold(sp->xvec[i]);
 119         }
 120         refcount_set(&sp->refcnt, 1);
 121         return sp;
 122 }
 123 EXPORT_SYMBOL(secpath_dup);
 124
 125 int secpath_set(struct sk_buff *skb)
 126 {
 127         struct sec_path *sp;
 128
 129         /* Allocate new secpath or COW existing one. */
 130         if (!skb->sp || refcount_read(&skb->sp->refcnt) != 1) {
 131                 sp = secpath_dup(skb->sp);
 132                 if (!sp)
 133                         return -ENOMEM;
 134
 135                 if (skb->sp)
 136                         secpath_put(skb->sp);
 137                 skb->sp = sp;
 138         }
 139         return 0;
 140 }
 141 EXPORT_SYMBOL(secpath_set);
 142
 143 /* Fetch spi and seq from ipsec header */
 144
 145 int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
 146 {
 147         int offset, offset_seq;
 148         int hlen;
 149
 150         switch (nexthdr) {
 151         case IPPROTO_AH:
 152                 hlen = sizeof(struct ip_auth_hdr);
 153                 offset = offsetof(struct ip_auth_hdr, spi);
 154                 offset_seq = offsetof(struct ip_auth_hdr, seq_no);
 155                 break;
 156         case IPPROTO_ESP:
 157                 hlen = sizeof(struct ip_esp_hdr);
 158                 offset = offsetof(struct ip_esp_hdr, spi);
 159                 offset_seq = offsetof(struct ip_esp_hdr, seq_no);
 160                 break;
 161         case IPPROTO_COMP:
 162                 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
 163                         return -EINVAL;
 164                 *spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
 165                 *seq = 0;
 166                 return 0;
 167         default:
 168                 return 1;
 169         }
 170
 171         if (!pskb_may_pull(skb, hlen))
 172                 return -EINVAL;
 173
 174         *spi = *(__be32 *)(skb_transport_header(skb) + offset);
 175         *seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
 176         return 0;
 177 }
 178 EXPORT_SYMBOL(xfrm_parse_spi);
 179
 180 int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
 181 {
 182         struct xfrm_mode *inner_mode = x->inner_mode;
 183         int err;
 184
 185         err = x->outer_mode->afinfo->extract_input(x, skb);
 186         if (err)
 187                 return err;
 188
 189         if (x->sel.family == AF_UNSPEC) {
 190                 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
 191                 if (inner_mode == NULL)
 192                         return -EAFNOSUPPORT;
 193         }
 194
 195         skb->protocol = inner_mode->afinfo->eth_proto;
 196         return inner_mode->input2(x, skb);
 197 }
 198 EXPORT_SYMBOL(xfrm_prepare_input);
 199
 200 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
 201 {
 202         struct net *net = dev_net(skb->dev);
 203         int err;
 204         __be32 seq;
 205         __be32 seq_hi;
 206         struct xfrm_state *x = NULL;
 207         xfrm_address_t *daddr;
 208         struct xfrm_mode *inner_mode;
 209         u32 mark = skb->mark;
 210         unsigned int family;
 211         int decaps = 0;
 212         int async = 0;
 213         bool xfrm_gro = false;
 214         bool crypto_done = false;
 215         struct xfrm_offload *xo = xfrm_offload(skb);
 216
 217         if (encap_type < 0) {
 218                 x = xfrm_input_state(skb);
 219                 family = x->outer_mode->afinfo->family;
 220
 221                 /* An encap_type of -1 indicates async resumption. */
 222                 if (encap_type == -1) {
 223                         async = 1;
 224                         seq = XFRM_SKB_CB(skb)->seq.input.low;
 225                         goto resume;
 226                 }
 227
 228                 /* encap_type < -1 indicates a GRO call. */
 229                 encap_type = 0;
 230                 seq = XFRM_SPI_SKB_CB(skb)->seq;
 231
 232                 if (xo && (xo->flags & CRYPTO_DONE)) {
 233                         crypto_done = true;
 234                         x = xfrm_input_state(skb);
 235                         family = XFRM_SPI_SKB_CB(skb)->family;
 236
 237                         if (!(xo->status & CRYPTO_SUCCESS)) {
 238                                 if (xo->status &
 239                                     (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
 240                                      CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
 241                                      CRYPTO_TUNNEL_AH_AUTH_FAILED |
 242                                      CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
 243
 244                                         xfrm_audit_state_icvfail(x, skb,
 245                                                                  x->type->proto);
 246                                         x->stats.integrity_failed++;
 247                                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
 248                                         goto drop;
 249                                 }
 250
 251                                 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
 252                                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
 253                                         goto drop;
 254                                 }
 255
 256                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
 257                                 goto drop;
 258                         }
 259
 260                         if ((err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
 261                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
 262                                 goto drop;
 263                         }
 264                 }
 265
 266                 goto lock;
 267         }
 268
 269         daddr = (xfrm_address_t *)(skb_network_header(skb) +
 270                                    XFRM_SPI_SKB_CB(skb)->daddroff);
 271         family = XFRM_SPI_SKB_CB(skb)->family;
 272
 273         /* if tunnel is present override skb->mark value with tunnel i_key */
 274         switch (family) {
 275         case AF_INET:
 276                 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
 277                         mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
 278                 break;
 279         case AF_INET6:
 280                 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
 281                         mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
 282                 break;
 283         }
 284
 285         err = secpath_set(skb);
 286         if (err) {
 287                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
 288                 goto drop;
 289         }
 290
 291         seq = 0;
 292         if (!spi && (err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
 293                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
 294                 goto drop;
 295         }
 296
 297         do {
 298                 if (skb->sp->len == XFRM_MAX_DEPTH) {
 299                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
 300                         goto drop;
 301                 }
 302
 303                 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
 304                 if (x == NULL) {
 305                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
 306                         xfrm_audit_state_notfound(skb, family, spi, seq);
 307                         goto drop;
 308                 }
 309
 310                 skb->sp->xvec[skb->sp->len++] = x;
 311
 312 lock:
 313                 spin_lock(&x->lock);
 314
 315                 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
 316                         if (x->km.state == XFRM_STATE_ACQ)
 317                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
 318                         else
 319                                 XFRM_INC_STATS(net,
 320                                                LINUX_MIB_XFRMINSTATEINVALID);
 321                         goto drop_unlock;
 322                 }
 323
 324                 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
 325                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
 326                         goto drop_unlock;
 327                 }
 328
 329                 if (x->repl->check(x, skb, seq)) {
 330                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
 331                         goto drop_unlock;
 332                 }
 333
 334                 if (xfrm_state_check_expire(x)) {
 335                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
 336                         goto drop_unlock;
 337                 }
 338
 339                 spin_unlock(&x->lock);
 340
 341                 if (xfrm_tunnel_check(skb, x, family)) {
 342                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
 343                         goto drop;
 344                 }
 345
 346                 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
 347
 348                 XFRM_SKB_CB(skb)->seq.input.low = seq;
 349                 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
 350
 351                 skb_dst_force(skb);
 352                 dev_hold(skb->dev);
 353
 354                 if (crypto_done)
 355                         nexthdr = x->type_offload->input_tail(x, skb);
 356                 else
 357                         nexthdr = x->type->input(x, skb);
 358
 359                 if (nexthdr == -EINPROGRESS)
 360                         return 0;
 361 resume:
 362                 dev_put(skb->dev);
 363
 364                 spin_lock(&x->lock);
 365                 if (nexthdr <= 0) {
 366                         if (nexthdr == -EBADMSG) {
 367                                 xfrm_audit_state_icvfail(x, skb,
 368                                                          x->type->proto);
 369                                 x->stats.integrity_failed++;
 370                         }
 371                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
 372                         goto drop_unlock;
 373                 }
 374
 375                 /* only the first xfrm gets the encap type */
 376                 encap_type = 0;
 377
 378                 if (async && x->repl->recheck(x, skb, seq)) {
 379                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
 380                         goto drop_unlock;
 381                 }
 382
 383                 x->repl->advance(x, seq);
 384
 385                 x->curlft.bytes += skb->len;
 386                 x->curlft.packets++;
 387
 388                 spin_unlock(&x->lock);
 389
 390                 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
 391
 392                 inner_mode = x->inner_mode;
 393
 394                 if (x->sel.family == AF_UNSPEC) {
 395                         inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
 396                         if (inner_mode == NULL) {
 397                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
 398                                 goto drop;
 399                         }
 400                 }
 401
 402                 if (inner_mode->input(x, skb)) {
 403                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
 404                         goto drop;
 405                 }
 406
 407                 if (x->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) {
 408                         decaps = 1;
 409                         break;
 410                 }
 411
 412                 /*
 413                  * We need the inner address.  However, we only get here for
 414                  * transport mode so the outer address is identical.
 415                  */
 416                 daddr = &x->id.daddr;
 417                 family = x->outer_mode->afinfo->family;
 418
 419                 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
 420                 if (err < 0) {
 421                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
 422                         goto drop;
 423                 }
 424         } while (!err);
 425
 426         err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
 427         if (err)
 428                 goto drop;
 429
 430         nf_reset(skb);
 431
 432         if (decaps) {
 433                 if (skb->sp)
 434                         skb->sp->olen = 0;
 435                 skb_dst_drop(skb);
 436                 gro_cells_receive(&gro_cells, skb);
 437                 return 0;
 438         } else {
 439                 xo = xfrm_offload(skb);
 440                 if (xo)
 441                         xfrm_gro = xo->flags & XFRM_GRO;
 442
 443                 err = x->inner_mode->afinfo->transport_finish(skb, xfrm_gro || async);
 444                 if (xfrm_gro) {
 445                         if (skb->sp)
 446                                 skb->sp->olen = 0;
 447                         skb_dst_drop(skb);
 448                         gro_cells_receive(&gro_cells, skb);
 449                         return err;
 450                 }
 451
 452                 return err;
 453         }
 454
 455 drop_unlock:
 456         spin_unlock(&x->lock);
 457 drop:
 458         xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
 459         kfree_skb(skb);
 460         return 0;
 461 }
 462 EXPORT_SYMBOL(xfrm_input);
 463
 464 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
 465 {
 466         return xfrm_input(skb, nexthdr, 0, -1);
 467 }
 468 EXPORT_SYMBOL(xfrm_input_resume);
 469
 470 void __init xfrm_input_init(void)
 471 {
 472         int err;
 473
 474         init_dummy_netdev(&xfrm_napi_dev);
 475         err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
 476         if (err)
 477                 gro_cells.cells = NULL;
 478
 479         secpath_cachep = kmem_cache_create("secpath_cache",
 480                                            sizeof(struct sec_path),
 481                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
 482                                            NULL);
 483 }