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