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