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/interrupt.h>
  13 #include <linux/slab.h>
  14 #include <linux/module.h>
  15 #include <linux/netdevice.h>
  16 #include <linux/percpu.h>
  17 #include <net/dst.h>
  18 #include <net/ip.h>
  19 #include <net/xfrm.h>
  20 #include <net/ip_tunnels.h>
  21 #include <net/ip6_tunnel.h>
  22
  23 struct xfrm_trans_tasklet {
  24         struct tasklet_struct tasklet;
  25         struct sk_buff_head queue;
  26 };
  27
  28 struct xfrm_trans_cb {
  29         int (*finish)(struct net *net, struct sock *sk, struct sk_buff *skb);
  30 };
  31
  32 #define XFRM_TRANS_SKB_CB(__skb) ((struct xfrm_trans_cb *)&((__skb)->cb[0]))
  33
  34 static struct kmem_cache *secpath_cachep __read_mostly;
  35
  36 static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
  37 static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[AF_INET6 + 1];
  38
  39 static struct gro_cells gro_cells;
  40 static struct net_device xfrm_napi_dev;
  41
  42 static DEFINE_PER_CPU(struct xfrm_trans_tasklet, xfrm_trans_tasklet);
  43
  44 int xfrm_input_register_afinfo(const struct xfrm_input_afinfo *afinfo)
  45 {
  46         int err = 0;
  47
  48         if (WARN_ON(afinfo->family >= ARRAY_SIZE(xfrm_input_afinfo)))
  49                 return -EAFNOSUPPORT;
  50
  51         spin_lock_bh(&xfrm_input_afinfo_lock);
  52         if (unlikely(xfrm_input_afinfo[afinfo->family] != NULL))
  53                 err = -EEXIST;
  54         else
  55                 rcu_assign_pointer(xfrm_input_afinfo[afinfo->family], afinfo);
  56         spin_unlock_bh(&xfrm_input_afinfo_lock);
  57         return err;
  58 }
  59 EXPORT_SYMBOL(xfrm_input_register_afinfo);
  60
  61 int xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo *afinfo)
  62 {
  63         int err = 0;
  64
  65         spin_lock_bh(&xfrm_input_afinfo_lock);
  66         if (likely(xfrm_input_afinfo[afinfo->family] != NULL)) {
  67                 if (unlikely(xfrm_input_afinfo[afinfo->family] != afinfo))
  68                         err = -EINVAL;
  69                 else
  70                         RCU_INIT_POINTER(xfrm_input_afinfo[afinfo->family], NULL);
  71         }
  72         spin_unlock_bh(&xfrm_input_afinfo_lock);
  73         synchronize_rcu();
  74         return err;
  75 }
  76 EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
  77
  78 static const struct xfrm_input_afinfo *xfrm_input_get_afinfo(unsigned int family)
  79 {
  80         const struct xfrm_input_afinfo *afinfo;
  81
  82         if (WARN_ON_ONCE(family >= ARRAY_SIZE(xfrm_input_afinfo)))
  83                 return NULL;
  84
  85         rcu_read_lock();
  86         afinfo = rcu_dereference(xfrm_input_afinfo[family]);
  87         if (unlikely(!afinfo))
  88                 rcu_read_unlock();
  89         return afinfo;
  90 }
  91
  92 static int xfrm_rcv_cb(struct sk_buff *skb, unsigned int family, u8 protocol,
  93                        int err)
  94 {
  95         int ret;
  96         const struct xfrm_input_afinfo *afinfo = xfrm_input_get_afinfo(family);
  97
  98         if (!afinfo)
  99                 return -EAFNOSUPPORT;
 100
 101         ret = afinfo->callback(skb, protocol, err);
 102         rcu_read_unlock();
 103
 104         return ret;
 105 }
 106
 107 void __secpath_destroy(struct sec_path *sp)
 108 {
 109         int i;
 110         for (i = 0; i < sp->len; i++)
 111                 xfrm_state_put(sp->xvec[i]);
 112         kmem_cache_free(secpath_cachep, sp);
 113 }
 114 EXPORT_SYMBOL(__secpath_destroy);
 115
 116 struct sec_path *secpath_dup(struct sec_path *src)
 117 {
 118         struct sec_path *sp;
 119
 120         sp = kmem_cache_alloc(secpath_cachep, GFP_ATOMIC);
 121         if (!sp)
 122                 return NULL;
 123
 124         sp->len = 0;
 125         sp->olen = 0;
 126
 127         memset(sp->ovec, 0, sizeof(sp->ovec[XFRM_MAX_OFFLOAD_DEPTH]));
 128
 129         if (src) {
 130                 int i;
 131
 132                 memcpy(sp, src, sizeof(*sp));
 133                 for (i = 0; i < sp->len; i++)
 134                         xfrm_state_hold(sp->xvec[i]);
 135         }
 136         refcount_set(&sp->refcnt, 1);
 137         return sp;
 138 }
 139 EXPORT_SYMBOL(secpath_dup);
 140
 141 int secpath_set(struct sk_buff *skb)
 142 {
 143         struct sec_path *sp;
 144
 145         /* Allocate new secpath or COW existing one. */
 146         if (!skb->sp || refcount_read(&skb->sp->refcnt) != 1) {
 147                 sp = secpath_dup(skb->sp);
 148                 if (!sp)
 149                         return -ENOMEM;
 150
 151                 if (skb->sp)
 152                         secpath_put(skb->sp);
 153                 skb->sp = sp;
 154         }
 155         return 0;
 156 }
 157 EXPORT_SYMBOL(secpath_set);
 158
 159 /* Fetch spi and seq from ipsec header */
 160
 161 int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
 162 {
 163         int offset, offset_seq;
 164         int hlen;
 165
 166         switch (nexthdr) {
 167         case IPPROTO_AH:
 168                 hlen = sizeof(struct ip_auth_hdr);
 169                 offset = offsetof(struct ip_auth_hdr, spi);
 170                 offset_seq = offsetof(struct ip_auth_hdr, seq_no);
 171                 break;
 172         case IPPROTO_ESP:
 173                 hlen = sizeof(struct ip_esp_hdr);
 174                 offset = offsetof(struct ip_esp_hdr, spi);
 175                 offset_seq = offsetof(struct ip_esp_hdr, seq_no);
 176                 break;
 177         case IPPROTO_COMP:
 178                 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
 179                         return -EINVAL;
 180                 *spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
 181                 *seq = 0;
 182                 return 0;
 183         default:
 184                 return 1;
 185         }
 186
 187         if (!pskb_may_pull(skb, hlen))
 188                 return -EINVAL;
 189
 190         *spi = *(__be32 *)(skb_transport_header(skb) + offset);
 191         *seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
 192         return 0;
 193 }
 194 EXPORT_SYMBOL(xfrm_parse_spi);
 195
 196 int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
 197 {
 198         struct xfrm_mode *inner_mode = x->inner_mode;
 199         int err;
 200
 201         err = x->outer_mode->afinfo->extract_input(x, skb);
 202         if (err)
 203                 return err;
 204
 205         if (x->sel.family == AF_UNSPEC) {
 206                 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
 207                 if (inner_mode == NULL)
 208                         return -EAFNOSUPPORT;
 209         }
 210
 211         skb->protocol = inner_mode->afinfo->eth_proto;
 212         return inner_mode->input2(x, skb);
 213 }
 214 EXPORT_SYMBOL(xfrm_prepare_input);
 215
 216 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
 217 {
 218         struct net *net = dev_net(skb->dev);
 219         int err;
 220         __be32 seq;
 221         __be32 seq_hi;
 222         struct xfrm_state *x = NULL;
 223         xfrm_address_t *daddr;
 224         struct xfrm_mode *inner_mode;
 225         u32 mark = skb->mark;
 226         unsigned int family = AF_UNSPEC;
 227         int decaps = 0;
 228         int async = 0;
 229         bool xfrm_gro = false;
 230         bool crypto_done = false;
 231         struct xfrm_offload *xo = xfrm_offload(skb);
 232
 233         if (encap_type < 0) {
 234                 x = xfrm_input_state(skb);
 235
 236                 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
 237                         if (x->km.state == XFRM_STATE_ACQ)
 238                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
 239                         else
 240                                 XFRM_INC_STATS(net,
 241                                                LINUX_MIB_XFRMINSTATEINVALID);
 242                         goto drop;
 243                 }
 244
 245                 family = x->outer_mode->afinfo->family;
 246
 247                 /* An encap_type of -1 indicates async resumption. */
 248                 if (encap_type == -1) {
 249                         async = 1;
 250                         seq = XFRM_SKB_CB(skb)->seq.input.low;
 251                         goto resume;
 252                 }
 253
 254                 /* encap_type < -1 indicates a GRO call. */
 255                 encap_type = 0;
 256                 seq = XFRM_SPI_SKB_CB(skb)->seq;
 257
 258                 if (xo && (xo->flags & CRYPTO_DONE)) {
 259                         crypto_done = true;
 260                         family = XFRM_SPI_SKB_CB(skb)->family;
 261
 262                         if (!(xo->status & CRYPTO_SUCCESS)) {
 263                                 if (xo->status &
 264                                     (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
 265                                      CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
 266                                      CRYPTO_TUNNEL_AH_AUTH_FAILED |
 267                                      CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
 268
 269                                         xfrm_audit_state_icvfail(x, skb,
 270                                                                  x->type->proto);
 271                                         x->stats.integrity_failed++;
 272                                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
 273                                         goto drop;
 274                                 }
 275
 276                                 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
 277                                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
 278                                         goto drop;
 279                                 }
 280
 281                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
 282                                 goto drop;
 283                         }
 284
 285                         if ((err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
 286                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
 287                                 goto drop;
 288                         }
 289                 }
 290
 291                 goto lock;
 292         }
 293
 294         family = XFRM_SPI_SKB_CB(skb)->family;
 295
 296         /* if tunnel is present override skb->mark value with tunnel i_key */
 297         switch (family) {
 298         case AF_INET:
 299                 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
 300                         mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
 301                 break;
 302         case AF_INET6:
 303                 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
 304                         mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
 305                 break;
 306         }
 307
 308         err = secpath_set(skb);
 309         if (err) {
 310                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
 311                 goto drop;
 312         }
 313
 314         seq = 0;
 315         if (!spi && (err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
 316                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
 317                 goto drop;
 318         }
 319
 320         daddr = (xfrm_address_t *)(skb_network_header(skb) +
 321                                    XFRM_SPI_SKB_CB(skb)->daddroff);
 322         do {
 323                 if (skb->sp->len == XFRM_MAX_DEPTH) {
 324                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
 325                         goto drop;
 326                 }
 327
 328                 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
 329                 if (x == NULL) {
 330                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
 331                         xfrm_audit_state_notfound(skb, family, spi, seq);
 332                         goto drop;
 333                 }
 334
 335                 skb->sp->xvec[skb->sp->len++] = x;
 336
 337 lock:
 338                 spin_lock(&x->lock);
 339
 340                 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
 341                         if (x->km.state == XFRM_STATE_ACQ)
 342                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
 343                         else
 344                                 XFRM_INC_STATS(net,
 345                                                LINUX_MIB_XFRMINSTATEINVALID);
 346                         goto drop_unlock;
 347                 }
 348
 349                 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
 350                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
 351                         goto drop_unlock;
 352                 }
 353
 354                 if (x->repl->check(x, skb, seq)) {
 355                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
 356                         goto drop_unlock;
 357                 }
 358
 359                 if (xfrm_state_check_expire(x)) {
 360                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
 361                         goto drop_unlock;
 362                 }
 363
 364                 spin_unlock(&x->lock);
 365
 366                 if (xfrm_tunnel_check(skb, x, family)) {
 367                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
 368                         goto drop;
 369                 }
 370
 371                 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
 372
 373                 XFRM_SKB_CB(skb)->seq.input.low = seq;
 374                 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
 375
 376                 skb_dst_force(skb);
 377                 dev_hold(skb->dev);
 378
 379                 if (crypto_done)
 380                         nexthdr = x->type_offload->input_tail(x, skb);
 381                 else
 382                         nexthdr = x->type->input(x, skb);
 383
 384                 if (nexthdr == -EINPROGRESS)
 385                         return 0;
 386 resume:
 387                 dev_put(skb->dev);
 388
 389                 spin_lock(&x->lock);
 390                 if (nexthdr <= 0) {
 391                         if (nexthdr == -EBADMSG) {
 392                                 xfrm_audit_state_icvfail(x, skb,
 393                                                          x->type->proto);
 394                                 x->stats.integrity_failed++;
 395                         }
 396                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
 397                         goto drop_unlock;
 398                 }
 399
 400                 /* only the first xfrm gets the encap type */
 401                 encap_type = 0;
 402
 403                 if (async && x->repl->recheck(x, skb, seq)) {
 404                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
 405                         goto drop_unlock;
 406                 }
 407
 408                 x->repl->advance(x, seq);
 409
 410                 x->curlft.bytes += skb->len;
 411                 x->curlft.packets++;
 412
 413                 spin_unlock(&x->lock);
 414
 415                 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
 416
 417                 inner_mode = x->inner_mode;
 418
 419                 if (x->sel.family == AF_UNSPEC) {
 420                         inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
 421                         if (inner_mode == NULL) {
 422                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
 423                                 goto drop;
 424                         }
 425                 }
 426
 427                 if (inner_mode->input(x, skb)) {
 428                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
 429                         goto drop;
 430                 }
 431
 432                 if (x->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) {
 433                         decaps = 1;
 434                         break;
 435                 }
 436
 437                 /*
 438                  * We need the inner address.  However, we only get here for
 439                  * transport mode so the outer address is identical.
 440                  */
 441                 daddr = &x->id.daddr;
 442                 family = x->outer_mode->afinfo->family;
 443
 444                 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
 445                 if (err < 0) {
 446                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
 447                         goto drop;
 448                 }
 449         } while (!err);
 450
 451         err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
 452         if (err)
 453                 goto drop;
 454
 455         nf_reset(skb);
 456
 457         if (decaps) {
 458                 if (skb->sp)
 459                         skb->sp->olen = 0;
 460                 skb_dst_drop(skb);
 461                 gro_cells_receive(&gro_cells, skb);
 462                 return 0;
 463         } else {
 464                 xo = xfrm_offload(skb);
 465                 if (xo)
 466                         xfrm_gro = xo->flags & XFRM_GRO;
 467
 468                 err = x->inner_mode->afinfo->transport_finish(skb, xfrm_gro || async);
 469                 if (xfrm_gro) {
 470                         if (skb->sp)
 471                                 skb->sp->olen = 0;
 472                         skb_dst_drop(skb);
 473                         gro_cells_receive(&gro_cells, skb);
 474                         return err;
 475                 }
 476
 477                 return err;
 478         }
 479
 480 drop_unlock:
 481         spin_unlock(&x->lock);
 482 drop:
 483         xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
 484         kfree_skb(skb);
 485         return 0;
 486 }
 487 EXPORT_SYMBOL(xfrm_input);
 488
 489 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
 490 {
 491         return xfrm_input(skb, nexthdr, 0, -1);
 492 }
 493 EXPORT_SYMBOL(xfrm_input_resume);
 494
 495 static void xfrm_trans_reinject(unsigned long data)
 496 {
 497         struct xfrm_trans_tasklet *trans = (void *)data;
 498         struct sk_buff_head queue;
 499         struct sk_buff *skb;
 500
 501         __skb_queue_head_init(&queue);
 502         skb_queue_splice_init(&trans->queue, &queue);
 503
 504         while ((skb = __skb_dequeue(&queue)))
 505                 XFRM_TRANS_SKB_CB(skb)->finish(dev_net(skb->dev), NULL, skb);
 506 }
 507
 508 int xfrm_trans_queue(struct sk_buff *skb,
 509                      int (*finish)(struct net *, struct sock *,
 510                                    struct sk_buff *))
 511 {
 512         struct xfrm_trans_tasklet *trans;
 513
 514         trans = this_cpu_ptr(&xfrm_trans_tasklet);
 515
 516         if (skb_queue_len(&trans->queue) >= netdev_max_backlog)
 517                 return -ENOBUFS;
 518
 519         XFRM_TRANS_SKB_CB(skb)->finish = finish;
 520         __skb_queue_tail(&trans->queue, skb);
 521         tasklet_schedule(&trans->tasklet);
 522         return 0;
 523 }
 524 EXPORT_SYMBOL(xfrm_trans_queue);
 525
 526 void __init xfrm_input_init(void)
 527 {
 528         int err;
 529         int i;
 530
 531         init_dummy_netdev(&xfrm_napi_dev);
 532         err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
 533         if (err)
 534                 gro_cells.cells = NULL;
 535
 536         secpath_cachep = kmem_cache_create("secpath_cache",
 537                                            sizeof(struct sec_path),
 538                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
 539                                            NULL);
 540
 541         for_each_possible_cpu(i) {
 542                 struct xfrm_trans_tasklet *trans;
 543
 544                 trans = &per_cpu(xfrm_trans_tasklet, i);
 545                 __skb_queue_head_init(&trans->queue);
 546                 tasklet_init(&trans->tasklet, xfrm_trans_reinject,
 547                              (unsigned long)trans);
 548         }
 549 }