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/bottom_half.h>
  12 #include <linux/cache.h>
  13 #include <linux/interrupt.h>
  14 #include <linux/slab.h>
  15 #include <linux/module.h>
  16 #include <linux/netdevice.h>
  17 #include <linux/percpu.h>
  18 #include <net/dst.h>
  19 #include <net/ip.h>
  20 #include <net/xfrm.h>
  21 #include <net/ip_tunnels.h>
  22 #include <net/ip6_tunnel.h>
  23
  24 struct xfrm_trans_tasklet {
  25         struct tasklet_struct tasklet;
  26         struct sk_buff_head queue;
  27 };
  28
  29 struct xfrm_trans_cb {
  30         union {
  31                 struct inet_skb_parm    h4;
  32 #if IS_ENABLED(CONFIG_IPV6)
  33                 struct inet6_skb_parm   h6;
  34 #endif
  35         } header;
  36         int (*finish)(struct net *net, struct sock *sk, struct sk_buff *skb);
  37 };
  38
  39 #define XFRM_TRANS_SKB_CB(__skb) ((struct xfrm_trans_cb *)&((__skb)->cb[0]))
  40
  41 static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
  42 static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[AF_INET6 + 1];
  43
  44 static struct gro_cells gro_cells;
  45 static struct net_device xfrm_napi_dev;
  46
  47 static DEFINE_PER_CPU(struct xfrm_trans_tasklet, xfrm_trans_tasklet);
  48
  49 int xfrm_input_register_afinfo(const struct xfrm_input_afinfo *afinfo)
  50 {
  51         int err = 0;
  52
  53         if (WARN_ON(afinfo->family >= ARRAY_SIZE(xfrm_input_afinfo)))
  54                 return -EAFNOSUPPORT;
  55
  56         spin_lock_bh(&xfrm_input_afinfo_lock);
  57         if (unlikely(xfrm_input_afinfo[afinfo->family] != NULL))
  58                 err = -EEXIST;
  59         else
  60                 rcu_assign_pointer(xfrm_input_afinfo[afinfo->family], afinfo);
  61         spin_unlock_bh(&xfrm_input_afinfo_lock);
  62         return err;
  63 }
  64 EXPORT_SYMBOL(xfrm_input_register_afinfo);
  65
  66 int xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo *afinfo)
  67 {
  68         int err = 0;
  69
  70         spin_lock_bh(&xfrm_input_afinfo_lock);
  71         if (likely(xfrm_input_afinfo[afinfo->family] != NULL)) {
  72                 if (unlikely(xfrm_input_afinfo[afinfo->family] != afinfo))
  73                         err = -EINVAL;
  74                 else
  75                         RCU_INIT_POINTER(xfrm_input_afinfo[afinfo->family], NULL);
  76         }
  77         spin_unlock_bh(&xfrm_input_afinfo_lock);
  78         synchronize_rcu();
  79         return err;
  80 }
  81 EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
  82
  83 static const struct xfrm_input_afinfo *xfrm_input_get_afinfo(unsigned int family)
  84 {
  85         const struct xfrm_input_afinfo *afinfo;
  86
  87         if (WARN_ON_ONCE(family >= ARRAY_SIZE(xfrm_input_afinfo)))
  88                 return NULL;
  89
  90         rcu_read_lock();
  91         afinfo = rcu_dereference(xfrm_input_afinfo[family]);
  92         if (unlikely(!afinfo))
  93                 rcu_read_unlock();
  94         return afinfo;
  95 }
  96
  97 static int xfrm_rcv_cb(struct sk_buff *skb, unsigned int family, u8 protocol,
  98                        int err)
  99 {
 100         int ret;
 101         const struct xfrm_input_afinfo *afinfo = xfrm_input_get_afinfo(family);
 102
 103         if (!afinfo)
 104                 return -EAFNOSUPPORT;
 105
 106         ret = afinfo->callback(skb, protocol, err);
 107         rcu_read_unlock();
 108
 109         return ret;
 110 }
 111
 112 struct sec_path *secpath_set(struct sk_buff *skb)
 113 {
 114         struct sec_path *sp, *tmp = skb_ext_find(skb, SKB_EXT_SEC_PATH);
 115
 116         sp = skb_ext_add(skb, SKB_EXT_SEC_PATH);
 117         if (!sp)
 118                 return NULL;
 119
 120         if (tmp) /* reused existing one (was COW'd if needed) */
 121                 return sp;
 122
 123         /* allocated new secpath */
 124         memset(sp->ovec, 0, sizeof(sp->ovec));
 125         sp->olen = 0;
 126         sp->len = 0;
 127
 128         return sp;
 129 }
 130 EXPORT_SYMBOL(secpath_set);
 131
 132 /* Fetch spi and seq from ipsec header */
 133
 134 int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
 135 {
 136         int offset, offset_seq;
 137         int hlen;
 138
 139         switch (nexthdr) {
 140         case IPPROTO_AH:
 141                 hlen = sizeof(struct ip_auth_hdr);
 142                 offset = offsetof(struct ip_auth_hdr, spi);
 143                 offset_seq = offsetof(struct ip_auth_hdr, seq_no);
 144                 break;
 145         case IPPROTO_ESP:
 146                 hlen = sizeof(struct ip_esp_hdr);
 147                 offset = offsetof(struct ip_esp_hdr, spi);
 148                 offset_seq = offsetof(struct ip_esp_hdr, seq_no);
 149                 break;
 150         case IPPROTO_COMP:
 151                 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
 152                         return -EINVAL;
 153                 *spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
 154                 *seq = 0;
 155                 return 0;
 156         default:
 157                 return 1;
 158         }
 159
 160         if (!pskb_may_pull(skb, hlen))
 161                 return -EINVAL;
 162
 163         *spi = *(__be32 *)(skb_transport_header(skb) + offset);
 164         *seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
 165         return 0;
 166 }
 167 EXPORT_SYMBOL(xfrm_parse_spi);
 168
 169 int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
 170 {
 171         struct xfrm_mode *inner_mode = x->inner_mode;
 172         int err;
 173
 174         err = x->outer_mode->afinfo->extract_input(x, skb);
 175         if (err)
 176                 return err;
 177
 178         if (x->sel.family == AF_UNSPEC) {
 179                 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
 180                 if (inner_mode == NULL)
 181                         return -EAFNOSUPPORT;
 182         }
 183
 184         skb->protocol = inner_mode->afinfo->eth_proto;
 185         return inner_mode->input2(x, skb);
 186 }
 187 EXPORT_SYMBOL(xfrm_prepare_input);
 188
 189 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
 190 {
 191         struct net *net = dev_net(skb->dev);
 192         int err;
 193         __be32 seq;
 194         __be32 seq_hi;
 195         struct xfrm_state *x = NULL;
 196         xfrm_address_t *daddr;
 197         struct xfrm_mode *inner_mode;
 198         u32 mark = skb->mark;
 199         unsigned int family = AF_UNSPEC;
 200         int decaps = 0;
 201         int async = 0;
 202         bool xfrm_gro = false;
 203         bool crypto_done = false;
 204         struct xfrm_offload *xo = xfrm_offload(skb);
 205         struct sec_path *sp;
 206
 207         if (encap_type < 0) {
 208                 x = xfrm_input_state(skb);
 209
 210                 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
 211                         if (x->km.state == XFRM_STATE_ACQ)
 212                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
 213                         else
 214                                 XFRM_INC_STATS(net,
 215                                                LINUX_MIB_XFRMINSTATEINVALID);
 216                         goto drop;
 217                 }
 218
 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                         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         family = XFRM_SPI_SKB_CB(skb)->family;
 269
 270         /* if tunnel is present override skb->mark value with tunnel i_key */
 271         switch (family) {
 272         case AF_INET:
 273                 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
 274                         mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
 275                 break;
 276         case AF_INET6:
 277                 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
 278                         mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
 279                 break;
 280         }
 281
 282         sp = secpath_set(skb);
 283         if (!sp) {
 284                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
 285                 goto drop;
 286         }
 287
 288         seq = 0;
 289         if (!spi && (err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
 290                 secpath_reset(skb);
 291                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
 292                 goto drop;
 293         }
 294
 295         daddr = (xfrm_address_t *)(skb_network_header(skb) +
 296                                    XFRM_SPI_SKB_CB(skb)->daddroff);
 297         do {
 298                 sp = skb_sec_path(skb);
 299
 300                 if (sp->len == XFRM_MAX_DEPTH) {
 301                         secpath_reset(skb);
 302                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
 303                         goto drop;
 304                 }
 305
 306                 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
 307                 if (x == NULL) {
 308                         secpath_reset(skb);
 309                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
 310                         xfrm_audit_state_notfound(skb, family, spi, seq);
 311                         goto drop;
 312                 }
 313
 314                 skb->mark = xfrm_smark_get(skb->mark, x);
 315
 316                 sp->xvec[sp->len++] = x;
 317
 318                 skb_dst_force(skb);
 319                 if (!skb_dst(skb)) {
 320                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
 321                         goto drop;
 322                 }
 323
 324 lock:
 325                 spin_lock(&x->lock);
 326
 327                 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
 328                         if (x->km.state == XFRM_STATE_ACQ)
 329                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
 330                         else
 331                                 XFRM_INC_STATS(net,
 332                                                LINUX_MIB_XFRMINSTATEINVALID);
 333                         goto drop_unlock;
 334                 }
 335
 336                 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
 337                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
 338                         goto drop_unlock;
 339                 }
 340
 341                 if (x->repl->check(x, skb, seq)) {
 342                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
 343                         goto drop_unlock;
 344                 }
 345
 346                 if (xfrm_state_check_expire(x)) {
 347                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
 348                         goto drop_unlock;
 349                 }
 350
 351                 spin_unlock(&x->lock);
 352
 353                 if (xfrm_tunnel_check(skb, x, family)) {
 354                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
 355                         goto drop;
 356                 }
 357
 358                 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
 359
 360                 XFRM_SKB_CB(skb)->seq.input.low = seq;
 361                 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
 362
 363                 dev_hold(skb->dev);
 364
 365                 if (crypto_done)
 366                         nexthdr = x->type_offload->input_tail(x, skb);
 367                 else
 368                         nexthdr = x->type->input(x, skb);
 369
 370                 if (nexthdr == -EINPROGRESS)
 371                         return 0;
 372 resume:
 373                 dev_put(skb->dev);
 374
 375                 spin_lock(&x->lock);
 376                 if (nexthdr <= 0) {
 377                         if (nexthdr == -EBADMSG) {
 378                                 xfrm_audit_state_icvfail(x, skb,
 379                                                          x->type->proto);
 380                                 x->stats.integrity_failed++;
 381                         }
 382                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
 383                         goto drop_unlock;
 384                 }
 385
 386                 /* only the first xfrm gets the encap type */
 387                 encap_type = 0;
 388
 389                 if (async && x->repl->recheck(x, skb, seq)) {
 390                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
 391                         goto drop_unlock;
 392                 }
 393
 394                 x->repl->advance(x, seq);
 395
 396                 x->curlft.bytes += skb->len;
 397                 x->curlft.packets++;
 398
 399                 spin_unlock(&x->lock);
 400
 401                 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
 402
 403                 inner_mode = x->inner_mode;
 404
 405                 if (x->sel.family == AF_UNSPEC) {
 406                         inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
 407                         if (inner_mode == NULL) {
 408                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
 409                                 goto drop;
 410                         }
 411                 }
 412
 413                 if (inner_mode->input(x, skb)) {
 414                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
 415                         goto drop;
 416                 }
 417
 418                 if (x->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) {
 419                         decaps = 1;
 420                         break;
 421                 }
 422
 423                 /*
 424                  * We need the inner address.  However, we only get here for
 425                  * transport mode so the outer address is identical.
 426                  */
 427                 daddr = &x->id.daddr;
 428                 family = x->outer_mode->afinfo->family;
 429
 430                 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
 431                 if (err < 0) {
 432                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
 433                         goto drop;
 434                 }
 435                 crypto_done = false;
 436         } while (!err);
 437
 438         err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
 439         if (err)
 440                 goto drop;
 441
 442         nf_reset(skb);
 443
 444         if (decaps) {
 445                 sp = skb_sec_path(skb);
 446                 if (sp)
 447                         sp->olen = 0;
 448                 skb_dst_drop(skb);
 449                 gro_cells_receive(&gro_cells, skb);
 450                 return 0;
 451         } else {
 452                 xo = xfrm_offload(skb);
 453                 if (xo)
 454                         xfrm_gro = xo->flags & XFRM_GRO;
 455
 456                 err = x->inner_mode->afinfo->transport_finish(skb, xfrm_gro || async);
 457                 if (xfrm_gro) {
 458                         sp = skb_sec_path(skb);
 459                         if (sp)
 460                                 sp->olen = 0;
 461                         skb_dst_drop(skb);
 462                         gro_cells_receive(&gro_cells, skb);
 463                         return err;
 464                 }
 465
 466                 return err;
 467         }
 468
 469 drop_unlock:
 470         spin_unlock(&x->lock);
 471 drop:
 472         xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
 473         kfree_skb(skb);
 474         return 0;
 475 }
 476 EXPORT_SYMBOL(xfrm_input);
 477
 478 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
 479 {
 480         return xfrm_input(skb, nexthdr, 0, -1);
 481 }
 482 EXPORT_SYMBOL(xfrm_input_resume);
 483
 484 static void xfrm_trans_reinject(unsigned long data)
 485 {
 486         struct xfrm_trans_tasklet *trans = (void *)data;
 487         struct sk_buff_head queue;
 488         struct sk_buff *skb;
 489
 490         __skb_queue_head_init(&queue);
 491         skb_queue_splice_init(&trans->queue, &queue);
 492
 493         while ((skb = __skb_dequeue(&queue)))
 494                 XFRM_TRANS_SKB_CB(skb)->finish(dev_net(skb->dev), NULL, skb);
 495 }
 496
 497 int xfrm_trans_queue(struct sk_buff *skb,
 498                      int (*finish)(struct net *, struct sock *,
 499                                    struct sk_buff *))
 500 {
 501         struct xfrm_trans_tasklet *trans;
 502
 503         trans = this_cpu_ptr(&xfrm_trans_tasklet);
 504
 505         if (skb_queue_len(&trans->queue) >= netdev_max_backlog)
 506                 return -ENOBUFS;
 507
 508         XFRM_TRANS_SKB_CB(skb)->finish = finish;
 509         __skb_queue_tail(&trans->queue, skb);
 510         tasklet_schedule(&trans->tasklet);
 511         return 0;
 512 }
 513 EXPORT_SYMBOL(xfrm_trans_queue);
 514
 515 void __init xfrm_input_init(void)
 516 {
 517         int err;
 518         int i;
 519
 520         init_dummy_netdev(&xfrm_napi_dev);
 521         err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
 522         if (err)
 523                 gro_cells.cells = NULL;
 524
 525         for_each_possible_cpu(i) {
 526                 struct xfrm_trans_tasklet *trans;
 527
 528                 trans = &per_cpu(xfrm_trans_tasklet, i);
 529                 __skb_queue_head_init(&trans->queue);
 530                 tasklet_init(&trans->tasklet, xfrm_trans_reinject,
 531                              (unsigned long)trans);
 532         }
 533 }