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 struct kmem_cache *secpath_cachep __ro_after_init;
  42
  43 static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
  44 static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[AF_INET6 + 1];
  45
  46 static struct gro_cells gro_cells;
  47 static struct net_device xfrm_napi_dev;
  48
  49 static DEFINE_PER_CPU(struct xfrm_trans_tasklet, xfrm_trans_tasklet);
  50
  51 int xfrm_input_register_afinfo(const struct xfrm_input_afinfo *afinfo)
  52 {
  53         int err = 0;
  54
  55         if (WARN_ON(afinfo->family >= ARRAY_SIZE(xfrm_input_afinfo)))
  56                 return -EAFNOSUPPORT;
  57
  58         spin_lock_bh(&xfrm_input_afinfo_lock);
  59         if (unlikely(xfrm_input_afinfo[afinfo->family] != NULL))
  60                 err = -EEXIST;
  61         else
  62                 rcu_assign_pointer(xfrm_input_afinfo[afinfo->family], afinfo);
  63         spin_unlock_bh(&xfrm_input_afinfo_lock);
  64         return err;
  65 }
  66 EXPORT_SYMBOL(xfrm_input_register_afinfo);
  67
  68 int xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo *afinfo)
  69 {
  70         int err = 0;
  71
  72         spin_lock_bh(&xfrm_input_afinfo_lock);
  73         if (likely(xfrm_input_afinfo[afinfo->family] != NULL)) {
  74                 if (unlikely(xfrm_input_afinfo[afinfo->family] != afinfo))
  75                         err = -EINVAL;
  76                 else
  77                         RCU_INIT_POINTER(xfrm_input_afinfo[afinfo->family], NULL);
  78         }
  79         spin_unlock_bh(&xfrm_input_afinfo_lock);
  80         synchronize_rcu();
  81         return err;
  82 }
  83 EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
  84
  85 static const struct xfrm_input_afinfo *xfrm_input_get_afinfo(unsigned int family)
  86 {
  87         const struct xfrm_input_afinfo *afinfo;
  88
  89         if (WARN_ON_ONCE(family >= ARRAY_SIZE(xfrm_input_afinfo)))
  90                 return NULL;
  91
  92         rcu_read_lock();
  93         afinfo = rcu_dereference(xfrm_input_afinfo[family]);
  94         if (unlikely(!afinfo))
  95                 rcu_read_unlock();
  96         return afinfo;
  97 }
  98
  99 static int xfrm_rcv_cb(struct sk_buff *skb, unsigned int family, u8 protocol,
 100                        int err)
 101 {
 102         int ret;
 103         const struct xfrm_input_afinfo *afinfo = xfrm_input_get_afinfo(family);
 104
 105         if (!afinfo)
 106                 return -EAFNOSUPPORT;
 107
 108         ret = afinfo->callback(skb, protocol, err);
 109         rcu_read_unlock();
 110
 111         return ret;
 112 }
 113
 114 void __secpath_destroy(struct sec_path *sp)
 115 {
 116         int i;
 117         for (i = 0; i < sp->len; i++)
 118                 xfrm_state_put(sp->xvec[i]);
 119         kmem_cache_free(secpath_cachep, sp);
 120 }
 121 EXPORT_SYMBOL(__secpath_destroy);
 122
 123 struct sec_path *secpath_dup(struct sec_path *src)
 124 {
 125         struct sec_path *sp;
 126
 127         sp = kmem_cache_alloc(secpath_cachep, GFP_ATOMIC);
 128         if (!sp)
 129                 return NULL;
 130
 131         sp->len = 0;
 132         sp->olen = 0;
 133
 134         memset(sp->ovec, 0, sizeof(sp->ovec));
 135
 136         if (src) {
 137                 int i;
 138
 139                 memcpy(sp, src, sizeof(*sp));
 140                 for (i = 0; i < sp->len; i++)
 141                         xfrm_state_hold(sp->xvec[i]);
 142         }
 143         refcount_set(&sp->refcnt, 1);
 144         return sp;
 145 }
 146 EXPORT_SYMBOL(secpath_dup);
 147
 148 int secpath_set(struct sk_buff *skb)
 149 {
 150         struct sec_path *sp;
 151
 152         /* Allocate new secpath or COW existing one. */
 153         if (!skb->sp || refcount_read(&skb->sp->refcnt) != 1) {
 154                 sp = secpath_dup(skb->sp);
 155                 if (!sp)
 156                         return -ENOMEM;
 157
 158                 if (skb->sp)
 159                         secpath_put(skb->sp);
 160                 skb->sp = sp;
 161         }
 162         return 0;
 163 }
 164 EXPORT_SYMBOL(secpath_set);
 165
 166 /* Fetch spi and seq from ipsec header */
 167
 168 int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
 169 {
 170         int offset, offset_seq;
 171         int hlen;
 172
 173         switch (nexthdr) {
 174         case IPPROTO_AH:
 175                 hlen = sizeof(struct ip_auth_hdr);
 176                 offset = offsetof(struct ip_auth_hdr, spi);
 177                 offset_seq = offsetof(struct ip_auth_hdr, seq_no);
 178                 break;
 179         case IPPROTO_ESP:
 180                 hlen = sizeof(struct ip_esp_hdr);
 181                 offset = offsetof(struct ip_esp_hdr, spi);
 182                 offset_seq = offsetof(struct ip_esp_hdr, seq_no);
 183                 break;
 184         case IPPROTO_COMP:
 185                 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
 186                         return -EINVAL;
 187                 *spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
 188                 *seq = 0;
 189                 return 0;
 190         default:
 191                 return 1;
 192         }
 193
 194         if (!pskb_may_pull(skb, hlen))
 195                 return -EINVAL;
 196
 197         *spi = *(__be32 *)(skb_transport_header(skb) + offset);
 198         *seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
 199         return 0;
 200 }
 201 EXPORT_SYMBOL(xfrm_parse_spi);
 202
 203 int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
 204 {
 205         struct xfrm_mode *inner_mode = x->inner_mode;
 206         int err;
 207
 208         err = x->outer_mode->afinfo->extract_input(x, skb);
 209         if (err)
 210                 return err;
 211
 212         if (x->sel.family == AF_UNSPEC) {
 213                 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
 214                 if (inner_mode == NULL)
 215                         return -EAFNOSUPPORT;
 216         }
 217
 218         skb->protocol = inner_mode->afinfo->eth_proto;
 219         return inner_mode->input2(x, skb);
 220 }
 221 EXPORT_SYMBOL(xfrm_prepare_input);
 222
 223 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
 224 {
 225         struct net *net = dev_net(skb->dev);
 226         int err;
 227         __be32 seq;
 228         __be32 seq_hi;
 229         struct xfrm_state *x = NULL;
 230         xfrm_address_t *daddr;
 231         struct xfrm_mode *inner_mode;
 232         u32 mark = skb->mark;
 233         unsigned int family = AF_UNSPEC;
 234         int decaps = 0;
 235         int async = 0;
 236         bool xfrm_gro = false;
 237         bool crypto_done = false;
 238         struct xfrm_offload *xo = xfrm_offload(skb);
 239
 240         if (encap_type < 0) {
 241                 x = xfrm_input_state(skb);
 242
 243                 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
 244                         if (x->km.state == XFRM_STATE_ACQ)
 245                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
 246                         else
 247                                 XFRM_INC_STATS(net,
 248                                                LINUX_MIB_XFRMINSTATEINVALID);
 249
 250                         if (encap_type == -1)
 251                                 dev_put(skb->dev);
 252                         goto drop;
 253                 }
 254
 255                 family = x->outer_mode->afinfo->family;
 256
 257                 /* An encap_type of -1 indicates async resumption. */
 258                 if (encap_type == -1) {
 259                         async = 1;
 260                         seq = XFRM_SKB_CB(skb)->seq.input.low;
 261                         goto resume;
 262                 }
 263
 264                 /* encap_type < -1 indicates a GRO call. */
 265                 encap_type = 0;
 266                 seq = XFRM_SPI_SKB_CB(skb)->seq;
 267
 268                 if (xo && (xo->flags & CRYPTO_DONE)) {
 269                         crypto_done = true;
 270                         family = XFRM_SPI_SKB_CB(skb)->family;
 271
 272                         if (!(xo->status & CRYPTO_SUCCESS)) {
 273                                 if (xo->status &
 274                                     (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
 275                                      CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
 276                                      CRYPTO_TUNNEL_AH_AUTH_FAILED |
 277                                      CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
 278
 279                                         xfrm_audit_state_icvfail(x, skb,
 280                                                                  x->type->proto);
 281                                         x->stats.integrity_failed++;
 282                                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
 283                                         goto drop;
 284                                 }
 285
 286                                 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
 287                                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
 288                                         goto drop;
 289                                 }
 290
 291                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
 292                                 goto drop;
 293                         }
 294
 295                         if ((err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
 296                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
 297                                 goto drop;
 298                         }
 299                 }
 300
 301                 goto lock;
 302         }
 303
 304         family = XFRM_SPI_SKB_CB(skb)->family;
 305
 306         /* if tunnel is present override skb->mark value with tunnel i_key */
 307         switch (family) {
 308         case AF_INET:
 309                 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
 310                         mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
 311                 break;
 312         case AF_INET6:
 313                 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
 314                         mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
 315                 break;
 316         }
 317
 318         err = secpath_set(skb);
 319         if (err) {
 320                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
 321                 goto drop;
 322         }
 323
 324         seq = 0;
 325         if (!spi && (err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
 326                 secpath_reset(skb);
 327                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
 328                 goto drop;
 329         }
 330
 331         daddr = (xfrm_address_t *)(skb_network_header(skb) +
 332                                    XFRM_SPI_SKB_CB(skb)->daddroff);
 333         do {
 334                 if (skb->sp->len == XFRM_MAX_DEPTH) {
 335                         secpath_reset(skb);
 336                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
 337                         goto drop;
 338                 }
 339
 340                 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
 341                 if (x == NULL) {
 342                         secpath_reset(skb);
 343                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
 344                         xfrm_audit_state_notfound(skb, family, spi, seq);
 345                         goto drop;
 346                 }
 347
 348                 skb->mark = xfrm_smark_get(skb->mark, x);
 349
 350                 skb->sp->xvec[skb->sp->len++] = x;
 351
 352                 skb_dst_force(skb);
 353                 if (!skb_dst(skb)) {
 354                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
 355                         goto drop;
 356                 }
 357
 358 lock:
 359                 spin_lock(&x->lock);
 360
 361                 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
 362                         if (x->km.state == XFRM_STATE_ACQ)
 363                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
 364                         else
 365                                 XFRM_INC_STATS(net,
 366                                                LINUX_MIB_XFRMINSTATEINVALID);
 367                         goto drop_unlock;
 368                 }
 369
 370                 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
 371                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
 372                         goto drop_unlock;
 373                 }
 374
 375                 if (x->repl->check(x, skb, seq)) {
 376                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
 377                         goto drop_unlock;
 378                 }
 379
 380                 if (xfrm_state_check_expire(x)) {
 381                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
 382                         goto drop_unlock;
 383                 }
 384
 385                 spin_unlock(&x->lock);
 386
 387                 if (xfrm_tunnel_check(skb, x, family)) {
 388                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
 389                         goto drop;
 390                 }
 391
 392                 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
 393
 394                 XFRM_SKB_CB(skb)->seq.input.low = seq;
 395                 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
 396
 397                 dev_hold(skb->dev);
 398
 399                 if (crypto_done)
 400                         nexthdr = x->type_offload->input_tail(x, skb);
 401                 else
 402                         nexthdr = x->type->input(x, skb);
 403
 404                 if (nexthdr == -EINPROGRESS)
 405                         return 0;
 406 resume:
 407                 dev_put(skb->dev);
 408
 409                 spin_lock(&x->lock);
 410                 if (nexthdr < 0) {
 411                         if (nexthdr == -EBADMSG) {
 412                                 xfrm_audit_state_icvfail(x, skb,
 413                                                          x->type->proto);
 414                                 x->stats.integrity_failed++;
 415                         }
 416                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
 417                         goto drop_unlock;
 418                 }
 419
 420                 /* only the first xfrm gets the encap type */
 421                 encap_type = 0;
 422
 423                 if (x->repl->recheck(x, skb, seq)) {
 424                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
 425                         goto drop_unlock;
 426                 }
 427
 428                 x->repl->advance(x, seq);
 429
 430                 x->curlft.bytes += skb->len;
 431                 x->curlft.packets++;
 432
 433                 spin_unlock(&x->lock);
 434
 435                 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
 436
 437                 inner_mode = x->inner_mode;
 438
 439                 if (x->sel.family == AF_UNSPEC) {
 440                         inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
 441                         if (inner_mode == NULL) {
 442                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
 443                                 goto drop;
 444                         }
 445                 }
 446
 447                 if (inner_mode->input(x, skb)) {
 448                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
 449                         goto drop;
 450                 }
 451
 452                 if (x->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) {
 453                         decaps = 1;
 454                         break;
 455                 }
 456
 457                 /*
 458                  * We need the inner address.  However, we only get here for
 459                  * transport mode so the outer address is identical.
 460                  */
 461                 daddr = &x->id.daddr;
 462                 family = x->outer_mode->afinfo->family;
 463
 464                 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
 465                 if (err < 0) {
 466                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
 467                         goto drop;
 468                 }
 469                 crypto_done = false;
 470         } while (!err);
 471
 472         err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
 473         if (err)
 474                 goto drop;
 475
 476         nf_reset(skb);
 477
 478         if (decaps) {
 479                 if (skb->sp)
 480                         skb->sp->olen = 0;
 481                 skb_dst_drop(skb);
 482                 gro_cells_receive(&gro_cells, skb);
 483                 return 0;
 484         } else {
 485                 xo = xfrm_offload(skb);
 486                 if (xo)
 487                         xfrm_gro = xo->flags & XFRM_GRO;
 488
 489                 err = x->inner_mode->afinfo->transport_finish(skb, xfrm_gro || async);
 490                 if (xfrm_gro) {
 491                         if (skb->sp)
 492                                 skb->sp->olen = 0;
 493                         skb_dst_drop(skb);
 494                         gro_cells_receive(&gro_cells, skb);
 495                         return err;
 496                 }
 497
 498                 return err;
 499         }
 500
 501 drop_unlock:
 502         spin_unlock(&x->lock);
 503 drop:
 504         xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
 505         kfree_skb(skb);
 506         return 0;
 507 }
 508 EXPORT_SYMBOL(xfrm_input);
 509
 510 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
 511 {
 512         return xfrm_input(skb, nexthdr, 0, -1);
 513 }
 514 EXPORT_SYMBOL(xfrm_input_resume);
 515
 516 static void xfrm_trans_reinject(unsigned long data)
 517 {
 518         struct xfrm_trans_tasklet *trans = (void *)data;
 519         struct sk_buff_head queue;
 520         struct sk_buff *skb;
 521
 522         __skb_queue_head_init(&queue);
 523         skb_queue_splice_init(&trans->queue, &queue);
 524
 525         while ((skb = __skb_dequeue(&queue)))
 526                 XFRM_TRANS_SKB_CB(skb)->finish(dev_net(skb->dev), NULL, skb);
 527 }
 528
 529 int xfrm_trans_queue(struct sk_buff *skb,
 530                      int (*finish)(struct net *, struct sock *,
 531                                    struct sk_buff *))
 532 {
 533         struct xfrm_trans_tasklet *trans;
 534
 535         trans = this_cpu_ptr(&xfrm_trans_tasklet);
 536
 537         if (skb_queue_len(&trans->queue) >= netdev_max_backlog)
 538                 return -ENOBUFS;
 539
 540         XFRM_TRANS_SKB_CB(skb)->finish = finish;
 541         __skb_queue_tail(&trans->queue, skb);
 542         tasklet_schedule(&trans->tasklet);
 543         return 0;
 544 }
 545 EXPORT_SYMBOL(xfrm_trans_queue);
 546
 547 void __init xfrm_input_init(void)
 548 {
 549         int err;
 550         int i;
 551
 552         init_dummy_netdev(&xfrm_napi_dev);
 553         err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
 554         if (err)
 555                 gro_cells.cells = NULL;
 556
 557         secpath_cachep = kmem_cache_create("secpath_cache",
 558                                            sizeof(struct sec_path),
 559                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
 560                                            NULL);
 561
 562         for_each_possible_cpu(i) {
 563                 struct xfrm_trans_tasklet *trans;
 564
 565                 trans = &per_cpu(xfrm_trans_tasklet, i);
 566                 __skb_queue_head_init(&trans->queue);
 567                 tasklet_init(&trans->tasklet, xfrm_trans_reinject,
 568                              (unsigned long)trans);
 569         }
 570 }