net/xfrm/xfrm_policy.c

   1 /*
   2  * xfrm_policy.c
   3  *
   4  * Changes:
   5  *      Mitsuru KANDA @USAGI
   6  *      Kazunori MIYAZAWA @USAGI
   7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
   8  *              IPv6 support
   9  *      Kazunori MIYAZAWA @USAGI
  10  *      YOSHIFUJI Hideaki
  11  *              Split up af-specific portion
  12  *      Derek Atkins <derek@ihtfp.com>          Add the post_input processor
  13  *
  14  */
  15
  16 #include <linux/config.h>
  17 #include <linux/slab.h>
  18 #include <linux/kmod.h>
  19 #include <linux/list.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/workqueue.h>
  22 #include <linux/notifier.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/module.h>
  25 #include <net/xfrm.h>
  26 #include <net/ip.h>
  27
  28 DECLARE_MUTEX(xfrm_cfg_sem);
  29 EXPORT_SYMBOL(xfrm_cfg_sem);
  30
  31 static DEFINE_RWLOCK(xfrm_policy_lock);
  32
  33 struct xfrm_policy *xfrm_policy_list[XFRM_POLICY_MAX*2];
  34 EXPORT_SYMBOL(xfrm_policy_list);
  35
  36 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
  37 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
  38
  39 static kmem_cache_t *xfrm_dst_cache;
  40
  41 static struct work_struct xfrm_policy_gc_work;
  42 static struct list_head xfrm_policy_gc_list =
  43         LIST_HEAD_INIT(xfrm_policy_gc_list);
  44 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
  45
  46 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
  47 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
  48
  49 int xfrm_register_type(struct xfrm_type *type, unsigned short family)
  50 {
  51         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
  52         struct xfrm_type_map *typemap;
  53         int err = 0;
  54
  55         if (unlikely(afinfo == NULL))
  56                 return -EAFNOSUPPORT;
  57         typemap = afinfo->type_map;
  58
  59         write_lock(&typemap->lock);
  60         if (likely(typemap->map[type->proto] == NULL))
  61                 typemap->map[type->proto] = type;
  62         else
  63                 err = -EEXIST;
  64         write_unlock(&typemap->lock);
  65         xfrm_policy_put_afinfo(afinfo);
  66         return err;
  67 }
  68 EXPORT_SYMBOL(xfrm_register_type);
  69
  70 int xfrm_unregister_type(struct xfrm_type *type, unsigned short family)
  71 {
  72         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
  73         struct xfrm_type_map *typemap;
  74         int err = 0;
  75
  76         if (unlikely(afinfo == NULL))
  77                 return -EAFNOSUPPORT;
  78         typemap = afinfo->type_map;
  79
  80         write_lock(&typemap->lock);
  81         if (unlikely(typemap->map[type->proto] != type))
  82                 err = -ENOENT;
  83         else
  84                 typemap->map[type->proto] = NULL;
  85         write_unlock(&typemap->lock);
  86         xfrm_policy_put_afinfo(afinfo);
  87         return err;
  88 }
  89 EXPORT_SYMBOL(xfrm_unregister_type);
  90
  91 struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
  92 {
  93         struct xfrm_policy_afinfo *afinfo;
  94         struct xfrm_type_map *typemap;
  95         struct xfrm_type *type;
  96         int modload_attempted = 0;
  97
  98 retry:
  99         afinfo = xfrm_policy_get_afinfo(family);
 100         if (unlikely(afinfo == NULL))
 101                 return NULL;
 102         typemap = afinfo->type_map;
 103
 104         read_lock(&typemap->lock);
 105         type = typemap->map[proto];
 106         if (unlikely(type && !try_module_get(type->owner)))
 107                 type = NULL;
 108         read_unlock(&typemap->lock);
 109         if (!type && !modload_attempted) {
 110                 xfrm_policy_put_afinfo(afinfo);
 111                 request_module("xfrm-type-%d-%d",
 112                                (int) family, (int) proto);
 113                 modload_attempted = 1;
 114                 goto retry;
 115         }
 116
 117         xfrm_policy_put_afinfo(afinfo);
 118         return type;
 119 }
 120 EXPORT_SYMBOL(xfrm_get_type);
 121
 122 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl,
 123                     unsigned short family)
 124 {
 125         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
 126         int err = 0;
 127
 128         if (unlikely(afinfo == NULL))
 129                 return -EAFNOSUPPORT;
 130
 131         if (likely(afinfo->dst_lookup != NULL))
 132                 err = afinfo->dst_lookup(dst, fl);
 133         else
 134                 err = -EINVAL;
 135         xfrm_policy_put_afinfo(afinfo);
 136         return err;
 137 }
 138 EXPORT_SYMBOL(xfrm_dst_lookup);
 139
 140 void xfrm_put_type(struct xfrm_type *type)
 141 {
 142         module_put(type->owner);
 143 }
 144
 145 static inline unsigned long make_jiffies(long secs)
 146 {
 147         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 148                 return MAX_SCHEDULE_TIMEOUT-1;
 149         else
 150                 return secs*HZ;
 151 }
 152
 153 static void xfrm_policy_timer(unsigned long data)
 154 {
 155         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 156         unsigned long now = (unsigned long)xtime.tv_sec;
 157         long next = LONG_MAX;
 158         int warn = 0;
 159         int dir;
 160
 161         read_lock(&xp->lock);
 162
 163         if (xp->dead)
 164                 goto out;
 165
 166         dir = xp->index & 7;
 167
 168         if (xp->lft.hard_add_expires_seconds) {
 169                 long tmo = xp->lft.hard_add_expires_seconds +
 170                         xp->curlft.add_time - now;
 171                 if (tmo <= 0)
 172                         goto expired;
 173                 if (tmo < next)
 174                         next = tmo;
 175         }
 176         if (xp->lft.hard_use_expires_seconds) {
 177                 long tmo = xp->lft.hard_use_expires_seconds +
 178                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 179                 if (tmo <= 0)
 180                         goto expired;
 181                 if (tmo < next)
 182                         next = tmo;
 183         }
 184         if (xp->lft.soft_add_expires_seconds) {
 185                 long tmo = xp->lft.soft_add_expires_seconds +
 186                         xp->curlft.add_time - now;
 187                 if (tmo <= 0) {
 188                         warn = 1;
 189                         tmo = XFRM_KM_TIMEOUT;
 190                 }
 191                 if (tmo < next)
 192                         next = tmo;
 193         }
 194         if (xp->lft.soft_use_expires_seconds) {
 195                 long tmo = xp->lft.soft_use_expires_seconds +
 196                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 197                 if (tmo <= 0) {
 198                         warn = 1;
 199                         tmo = XFRM_KM_TIMEOUT;
 200                 }
 201                 if (tmo < next)
 202                         next = tmo;
 203         }
 204
 205         if (warn)
 206                 km_policy_expired(xp, dir, 0);
 207         if (next != LONG_MAX &&
 208             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 209                 xfrm_pol_hold(xp);
 210
 211 out:
 212         read_unlock(&xp->lock);
 213         xfrm_pol_put(xp);
 214         return;
 215
 216 expired:
 217         read_unlock(&xp->lock);
 218         km_policy_expired(xp, dir, 1);
 219         xfrm_policy_delete(xp, dir);
 220         xfrm_pol_put(xp);
 221 }
 222
 223
 224 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 225  * SPD calls.
 226  */
 227
 228 struct xfrm_policy *xfrm_policy_alloc(int gfp)
 229 {
 230         struct xfrm_policy *policy;
 231
 232         policy = kmalloc(sizeof(struct xfrm_policy), gfp);
 233
 234         if (policy) {
 235                 memset(policy, 0, sizeof(struct xfrm_policy));
 236                 atomic_set(&policy->refcnt, 1);
 237                 rwlock_init(&policy->lock);
 238                 init_timer(&policy->timer);
 239                 policy->timer.data = (unsigned long)policy;
 240                 policy->timer.function = xfrm_policy_timer;
 241         }
 242         return policy;
 243 }
 244 EXPORT_SYMBOL(xfrm_policy_alloc);
 245
 246 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 247
 248 void __xfrm_policy_destroy(struct xfrm_policy *policy)
 249 {
 250         if (!policy->dead)
 251                 BUG();
 252
 253         if (policy->bundles)
 254                 BUG();
 255
 256         if (del_timer(&policy->timer))
 257                 BUG();
 258
 259         kfree(policy);
 260 }
 261 EXPORT_SYMBOL(__xfrm_policy_destroy);
 262
 263 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
 264 {
 265         struct dst_entry *dst;
 266
 267         while ((dst = policy->bundles) != NULL) {
 268                 policy->bundles = dst->next;
 269                 dst_free(dst);
 270         }
 271
 272         if (del_timer(&policy->timer))
 273                 atomic_dec(&policy->refcnt);
 274
 275         if (atomic_read(&policy->refcnt) > 1)
 276                 flow_cache_flush();
 277
 278         xfrm_pol_put(policy);
 279 }
 280
 281 static void xfrm_policy_gc_task(void *data)
 282 {
 283         struct xfrm_policy *policy;
 284         struct list_head *entry, *tmp;
 285         struct list_head gc_list = LIST_HEAD_INIT(gc_list);
 286
 287         spin_lock_bh(&xfrm_policy_gc_lock);
 288         list_splice_init(&xfrm_policy_gc_list, &gc_list);
 289         spin_unlock_bh(&xfrm_policy_gc_lock);
 290
 291         list_for_each_safe(entry, tmp, &gc_list) {
 292                 policy = list_entry(entry, struct xfrm_policy, list);
 293                 xfrm_policy_gc_kill(policy);
 294         }
 295 }
 296
 297 /* Rule must be locked. Release descentant resources, announce
 298  * entry dead. The rule must be unlinked from lists to the moment.
 299  */
 300
 301 static void xfrm_policy_kill(struct xfrm_policy *policy)
 302 {
 303         write_lock_bh(&policy->lock);
 304         if (policy->dead)
 305                 goto out;
 306
 307         policy->dead = 1;
 308
 309         spin_lock(&xfrm_policy_gc_lock);
 310         list_add(&policy->list, &xfrm_policy_gc_list);
 311         spin_unlock(&xfrm_policy_gc_lock);
 312         schedule_work(&xfrm_policy_gc_work);
 313
 314 out:
 315         write_unlock_bh(&policy->lock);
 316 }
 317
 318 /* Generate new index... KAME seems to generate them ordered by cost
 319  * of an absolute inpredictability of ordering of rules. This will not pass. */
 320 static u32 xfrm_gen_index(int dir)
 321 {
 322         u32 idx;
 323         struct xfrm_policy *p;
 324         static u32 idx_generator;
 325
 326         for (;;) {
 327                 idx = (idx_generator | dir);
 328                 idx_generator += 8;
 329                 if (idx == 0)
 330                         idx = 8;
 331                 for (p = xfrm_policy_list[dir]; p; p = p->next) {
 332                         if (p->index == idx)
 333                                 break;
 334                 }
 335                 if (!p)
 336                         return idx;
 337         }
 338 }
 339
 340 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 341 {
 342         struct xfrm_policy *pol, **p;
 343         struct xfrm_policy *delpol = NULL;
 344         struct xfrm_policy **newpos = NULL;
 345
 346         write_lock_bh(&xfrm_policy_lock);
 347         for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL;) {
 348                 if (!delpol && memcmp(&policy->selector, &pol->selector, sizeof(pol->selector)) == 0) {
 349                         if (excl) {
 350                                 write_unlock_bh(&xfrm_policy_lock);
 351                                 return -EEXIST;
 352                         }
 353                         *p = pol->next;
 354                         delpol = pol;
 355                         if (policy->priority > pol->priority)
 356                                 continue;
 357                 } else if (policy->priority >= pol->priority) {
 358                         p = &pol->next;
 359                         continue;
 360                 }
 361                 if (!newpos)
 362                         newpos = p;
 363                 if (delpol)
 364                         break;
 365                 p = &pol->next;
 366         }
 367         if (newpos)
 368                 p = newpos;
 369         xfrm_pol_hold(policy);
 370         policy->next = *p;
 371         *p = policy;
 372         atomic_inc(&flow_cache_genid);
 373         policy->index = delpol ? delpol->index : xfrm_gen_index(dir);
 374         policy->curlft.add_time = (unsigned long)xtime.tv_sec;
 375         policy->curlft.use_time = 0;
 376         if (!mod_timer(&policy->timer, jiffies + HZ))
 377                 xfrm_pol_hold(policy);
 378         write_unlock_bh(&xfrm_policy_lock);
 379
 380         if (delpol) {
 381                 xfrm_policy_kill(delpol);
 382         }
 383         return 0;
 384 }
 385 EXPORT_SYMBOL(xfrm_policy_insert);
 386
 387 struct xfrm_policy *xfrm_policy_bysel(int dir, struct xfrm_selector *sel,
 388                                       int delete)
 389 {
 390         struct xfrm_policy *pol, **p;
 391
 392         write_lock_bh(&xfrm_policy_lock);
 393         for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) {
 394                 if (memcmp(sel, &pol->selector, sizeof(*sel)) == 0) {
 395                         xfrm_pol_hold(pol);
 396                         if (delete)
 397                                 *p = pol->next;
 398                         break;
 399                 }
 400         }
 401         write_unlock_bh(&xfrm_policy_lock);
 402
 403         if (pol && delete) {
 404                 atomic_inc(&flow_cache_genid);
 405                 xfrm_policy_kill(pol);
 406         }
 407         return pol;
 408 }
 409 EXPORT_SYMBOL(xfrm_policy_bysel);
 410
 411 struct xfrm_policy *xfrm_policy_byid(int dir, u32 id, int delete)
 412 {
 413         struct xfrm_policy *pol, **p;
 414
 415         write_lock_bh(&xfrm_policy_lock);
 416         for (p = &xfrm_policy_list[id & 7]; (pol=*p)!=NULL; p = &pol->next) {
 417                 if (pol->index == id) {
 418                         xfrm_pol_hold(pol);
 419                         if (delete)
 420                                 *p = pol->next;
 421                         break;
 422                 }
 423         }
 424         write_unlock_bh(&xfrm_policy_lock);
 425
 426         if (pol && delete) {
 427                 atomic_inc(&flow_cache_genid);
 428                 xfrm_policy_kill(pol);
 429         }
 430         return pol;
 431 }
 432 EXPORT_SYMBOL(xfrm_policy_byid);
 433
 434 void xfrm_policy_flush(void)
 435 {
 436         struct xfrm_policy *xp;
 437         int dir;
 438
 439         write_lock_bh(&xfrm_policy_lock);
 440         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 441                 while ((xp = xfrm_policy_list[dir]) != NULL) {
 442                         xfrm_policy_list[dir] = xp->next;
 443                         write_unlock_bh(&xfrm_policy_lock);
 444
 445                         xfrm_policy_kill(xp);
 446
 447                         write_lock_bh(&xfrm_policy_lock);
 448                 }
 449         }
 450         atomic_inc(&flow_cache_genid);
 451         write_unlock_bh(&xfrm_policy_lock);
 452 }
 453 EXPORT_SYMBOL(xfrm_policy_flush);
 454
 455 int xfrm_policy_walk(int (*func)(struct xfrm_policy *, int, int, void*),
 456                      void *data)
 457 {
 458         struct xfrm_policy *xp;
 459         int dir;
 460         int count = 0;
 461         int error = 0;
 462
 463         read_lock_bh(&xfrm_policy_lock);
 464         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
 465                 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next)
 466                         count++;
 467         }
 468
 469         if (count == 0) {
 470                 error = -ENOENT;
 471                 goto out;
 472         }
 473
 474         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
 475                 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next) {
 476                         error = func(xp, dir%XFRM_POLICY_MAX, --count, data);
 477                         if (error)
 478                                 goto out;
 479                 }
 480         }
 481
 482 out:
 483         read_unlock_bh(&xfrm_policy_lock);
 484         return error;
 485 }
 486 EXPORT_SYMBOL(xfrm_policy_walk);
 487
 488 /* Find policy to apply to this flow. */
 489
 490 static void xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
 491                                void **objp, atomic_t **obj_refp)
 492 {
 493         struct xfrm_policy *pol;
 494
 495         read_lock_bh(&xfrm_policy_lock);
 496         for (pol = xfrm_policy_list[dir]; pol; pol = pol->next) {
 497                 struct xfrm_selector *sel = &pol->selector;
 498                 int match;
 499
 500                 if (pol->family != family)
 501                         continue;
 502
 503                 match = xfrm_selector_match(sel, fl, family);
 504                 if (match) {
 505                         xfrm_pol_hold(pol);
 506                         break;
 507                 }
 508         }
 509         read_unlock_bh(&xfrm_policy_lock);
 510         if ((*objp = (void *) pol) != NULL)
 511                 *obj_refp = &pol->refcnt;
 512 }
 513
 514 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
 515 {
 516         struct xfrm_policy *pol;
 517
 518         read_lock_bh(&xfrm_policy_lock);
 519         if ((pol = sk->sk_policy[dir]) != NULL) {
 520                 int match = xfrm_selector_match(&pol->selector, fl,
 521                                                 sk->sk_family);
 522                 if (match)
 523                         xfrm_pol_hold(pol);
 524                 else
 525                         pol = NULL;
 526         }
 527         read_unlock_bh(&xfrm_policy_lock);
 528         return pol;
 529 }
 530
 531 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
 532 {
 533         pol->next = xfrm_policy_list[dir];
 534         xfrm_policy_list[dir] = pol;
 535         xfrm_pol_hold(pol);
 536 }
 537
 538 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
 539                                                 int dir)
 540 {
 541         struct xfrm_policy **polp;
 542
 543         for (polp = &xfrm_policy_list[dir];
 544              *polp != NULL; polp = &(*polp)->next) {
 545                 if (*polp == pol) {
 546                         *polp = pol->next;
 547                         return pol;
 548                 }
 549         }
 550         return NULL;
 551 }
 552
 553 void xfrm_policy_delete(struct xfrm_policy *pol, int dir)
 554 {
 555         write_lock_bh(&xfrm_policy_lock);
 556         pol = __xfrm_policy_unlink(pol, dir);
 557         write_unlock_bh(&xfrm_policy_lock);
 558         if (pol) {
 559                 if (dir < XFRM_POLICY_MAX)
 560                         atomic_inc(&flow_cache_genid);
 561                 xfrm_policy_kill(pol);
 562         }
 563 }
 564
 565 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
 566 {
 567         struct xfrm_policy *old_pol;
 568
 569         write_lock_bh(&xfrm_policy_lock);
 570         old_pol = sk->sk_policy[dir];
 571         sk->sk_policy[dir] = pol;
 572         if (pol) {
 573                 pol->curlft.add_time = (unsigned long)xtime.tv_sec;
 574                 pol->index = xfrm_gen_index(XFRM_POLICY_MAX+dir);
 575                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
 576         }
 577         if (old_pol)
 578                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
 579         write_unlock_bh(&xfrm_policy_lock);
 580
 581         if (old_pol) {
 582                 xfrm_policy_kill(old_pol);
 583         }
 584         return 0;
 585 }
 586
 587 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
 588 {
 589         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
 590
 591         if (newp) {
 592                 newp->selector = old->selector;
 593                 newp->lft = old->lft;
 594                 newp->curlft = old->curlft;
 595                 newp->action = old->action;
 596                 newp->flags = old->flags;
 597                 newp->xfrm_nr = old->xfrm_nr;
 598                 newp->index = old->index;
 599                 memcpy(newp->xfrm_vec, old->xfrm_vec,
 600                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
 601                 write_lock_bh(&xfrm_policy_lock);
 602                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
 603                 write_unlock_bh(&xfrm_policy_lock);
 604                 xfrm_pol_put(newp);
 605         }
 606         return newp;
 607 }
 608
 609 int __xfrm_sk_clone_policy(struct sock *sk)
 610 {
 611         struct xfrm_policy *p0 = sk->sk_policy[0],
 612                            *p1 = sk->sk_policy[1];
 613
 614         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
 615         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
 616                 return -ENOMEM;
 617         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
 618                 return -ENOMEM;
 619         return 0;
 620 }
 621
 622 /* Resolve list of templates for the flow, given policy. */
 623
 624 static int
 625 xfrm_tmpl_resolve(struct xfrm_policy *policy, struct flowi *fl,
 626                   struct xfrm_state **xfrm,
 627                   unsigned short family)
 628 {
 629         int nx;
 630         int i, error;
 631         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
 632         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
 633
 634         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
 635                 struct xfrm_state *x;
 636                 xfrm_address_t *remote = daddr;
 637                 xfrm_address_t *local  = saddr;
 638                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
 639
 640                 if (tmpl->mode) {
 641                         remote = &tmpl->id.daddr;
 642                         local = &tmpl->saddr;
 643                 }
 644
 645                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
 646
 647                 if (x && x->km.state == XFRM_STATE_VALID) {
 648                         xfrm[nx++] = x;
 649                         daddr = remote;
 650                         saddr = local;
 651                         continue;
 652                 }
 653                 if (x) {
 654                         error = (x->km.state == XFRM_STATE_ERROR ?
 655                                  -EINVAL : -EAGAIN);
 656                         xfrm_state_put(x);
 657                 }
 658
 659                 if (!tmpl->optional)
 660                         goto fail;
 661         }
 662         return nx;
 663
 664 fail:
 665         for (nx--; nx>=0; nx--)
 666                 xfrm_state_put(xfrm[nx]);
 667         return error;
 668 }
 669
 670 /* Check that the bundle accepts the flow and its components are
 671  * still valid.
 672  */
 673
 674 static struct dst_entry *
 675 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
 676 {
 677         struct dst_entry *x;
 678         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
 679         if (unlikely(afinfo == NULL))
 680                 return ERR_PTR(-EINVAL);
 681         x = afinfo->find_bundle(fl, policy);
 682         xfrm_policy_put_afinfo(afinfo);
 683         return x;
 684 }
 685
 686 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
 687  * all the metrics... Shortly, bundle a bundle.
 688  */
 689
 690 static int
 691 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
 692                    struct flowi *fl, struct dst_entry **dst_p,
 693                    unsigned short family)
 694 {
 695         int err;
 696         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
 697         if (unlikely(afinfo == NULL))
 698                 return -EINVAL;
 699         err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
 700         xfrm_policy_put_afinfo(afinfo);
 701         return err;
 702 }
 703
 704 static inline int policy_to_flow_dir(int dir)
 705 {
 706         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
 707             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
 708             XFRM_POLICY_FWD == FLOW_DIR_FWD)
 709                 return dir;
 710         switch (dir) {
 711         default:
 712         case XFRM_POLICY_IN:
 713                 return FLOW_DIR_IN;
 714         case XFRM_POLICY_OUT:
 715                 return FLOW_DIR_OUT;
 716         case XFRM_POLICY_FWD:
 717                 return FLOW_DIR_FWD;
 718         };
 719 }
 720
 721 static int stale_bundle(struct dst_entry *dst);
 722
 723 /* Main function: finds/creates a bundle for given flow.
 724  *
 725  * At the moment we eat a raw IP route. Mostly to speed up lookups
 726  * on interfaces with disabled IPsec.
 727  */
 728 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
 729                 struct sock *sk, int flags)
 730 {
 731         struct xfrm_policy *policy;
 732         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
 733         struct dst_entry *dst, *dst_orig = *dst_p;
 734         int nx = 0;
 735         int err;
 736         u32 genid;
 737         u16 family = dst_orig->ops->family;
 738 restart:
 739         genid = atomic_read(&flow_cache_genid);
 740         policy = NULL;
 741         if (sk && sk->sk_policy[1])
 742                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
 743
 744         if (!policy) {
 745                 /* To accelerate a bit...  */
 746                 if ((dst_orig->flags & DST_NOXFRM) || !xfrm_policy_list[XFRM_POLICY_OUT])
 747                         return 0;
 748
 749                 policy = flow_cache_lookup(fl, family,
 750                                            policy_to_flow_dir(XFRM_POLICY_OUT),
 751                                            xfrm_policy_lookup);
 752         }
 753
 754         if (!policy)
 755                 return 0;
 756
 757         policy->curlft.use_time = (unsigned long)xtime.tv_sec;
 758
 759         switch (policy->action) {
 760         case XFRM_POLICY_BLOCK:
 761                 /* Prohibit the flow */
 762                 xfrm_pol_put(policy);
 763                 return -EPERM;
 764
 765         case XFRM_POLICY_ALLOW:
 766                 if (policy->xfrm_nr == 0) {
 767                         /* Flow passes not transformed. */
 768                         xfrm_pol_put(policy);
 769                         return 0;
 770                 }
 771
 772                 /* Try to find matching bundle.
 773                  *
 774                  * LATER: help from flow cache. It is optional, this
 775                  * is required only for output policy.
 776                  */
 777                 dst = xfrm_find_bundle(fl, policy, family);
 778                 if (IS_ERR(dst)) {
 779                         xfrm_pol_put(policy);
 780                         return PTR_ERR(dst);
 781                 }
 782
 783                 if (dst)
 784                         break;
 785
 786                 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
 787
 788                 if (unlikely(nx<0)) {
 789                         err = nx;
 790                         if (err == -EAGAIN && flags) {
 791                                 DECLARE_WAITQUEUE(wait, current);
 792
 793                                 add_wait_queue(&km_waitq, &wait);
 794                                 set_current_state(TASK_INTERRUPTIBLE);
 795                                 schedule();
 796                                 set_current_state(TASK_RUNNING);
 797                                 remove_wait_queue(&km_waitq, &wait);
 798
 799                                 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
 800
 801                                 if (nx == -EAGAIN && signal_pending(current)) {
 802                                         err = -ERESTART;
 803                                         goto error;
 804                                 }
 805                                 if (nx == -EAGAIN ||
 806                                     genid != atomic_read(&flow_cache_genid)) {
 807                                         xfrm_pol_put(policy);
 808                                         goto restart;
 809                                 }
 810                                 err = nx;
 811                         }
 812                         if (err < 0)
 813                                 goto error;
 814                 }
 815                 if (nx == 0) {
 816                         /* Flow passes not transformed. */
 817                         xfrm_pol_put(policy);
 818                         return 0;
 819                 }
 820
 821                 dst = dst_orig;
 822                 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
 823
 824                 if (unlikely(err)) {
 825                         int i;
 826                         for (i=0; i<nx; i++)
 827                                 xfrm_state_put(xfrm[i]);
 828                         goto error;
 829                 }
 830
 831                 write_lock_bh(&policy->lock);
 832                 if (unlikely(policy->dead || stale_bundle(dst))) {
 833                         /* Wow! While we worked on resolving, this
 834                          * policy has gone. Retry. It is not paranoia,
 835                          * we just cannot enlist new bundle to dead object.
 836                          * We can't enlist stable bundles either.
 837                          */
 838                         write_unlock_bh(&policy->lock);
 839
 840                         xfrm_pol_put(policy);
 841                         if (dst)
 842                                 dst_free(dst);
 843                         goto restart;
 844                 }
 845                 dst->next = policy->bundles;
 846                 policy->bundles = dst;
 847                 dst_hold(dst);
 848                 write_unlock_bh(&policy->lock);
 849         }
 850         *dst_p = dst;
 851         dst_release(dst_orig);
 852         xfrm_pol_put(policy);
 853         return 0;
 854
 855 error:
 856         dst_release(dst_orig);
 857         xfrm_pol_put(policy);
 858         *dst_p = NULL;
 859         return err;
 860 }
 861 EXPORT_SYMBOL(xfrm_lookup);
 862
 863 /* When skb is transformed back to its "native" form, we have to
 864  * check policy restrictions. At the moment we make this in maximally
 865  * stupid way. Shame on me. :-) Of course, connected sockets must
 866  * have policy cached at them.
 867  */
 868
 869 static inline int
 870 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
 871               unsigned short family)
 872 {
 873         if (xfrm_state_kern(x))
 874                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, family);
 875         return  x->id.proto == tmpl->id.proto &&
 876                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
 877                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
 878                 x->props.mode == tmpl->mode &&
 879                 (tmpl->aalgos & (1<<x->props.aalgo)) &&
 880                 !(x->props.mode && xfrm_state_addr_cmp(tmpl, x, family));
 881 }
 882
 883 static inline int
 884 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
 885                unsigned short family)
 886 {
 887         int idx = start;
 888
 889         if (tmpl->optional) {
 890                 if (!tmpl->mode)
 891                         return start;
 892         } else
 893                 start = -1;
 894         for (; idx < sp->len; idx++) {
 895                 if (xfrm_state_ok(tmpl, sp->x[idx].xvec, family))
 896                         return ++idx;
 897                 if (sp->x[idx].xvec->props.mode)
 898                         break;
 899         }
 900         return start;
 901 }
 902
 903 static int
 904 _decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
 905 {
 906         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
 907
 908         if (unlikely(afinfo == NULL))
 909                 return -EAFNOSUPPORT;
 910
 911         afinfo->decode_session(skb, fl);
 912         xfrm_policy_put_afinfo(afinfo);
 913         return 0;
 914 }
 915
 916 static inline int secpath_has_tunnel(struct sec_path *sp, int k)
 917 {
 918         for (; k < sp->len; k++) {
 919                 if (sp->x[k].xvec->props.mode)
 920                         return 1;
 921         }
 922
 923         return 0;
 924 }
 925
 926 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
 927                         unsigned short family)
 928 {
 929         struct xfrm_policy *pol;
 930         struct flowi fl;
 931
 932         if (_decode_session(skb, &fl, family) < 0)
 933                 return 0;
 934
 935         /* First, check used SA against their selectors. */
 936         if (skb->sp) {
 937                 int i;
 938
 939                 for (i=skb->sp->len-1; i>=0; i--) {
 940                   struct sec_decap_state *xvec = &(skb->sp->x[i]);
 941                         if (!xfrm_selector_match(&xvec->xvec->sel, &fl, family))
 942                                 return 0;
 943
 944                         /* If there is a post_input processor, try running it */
 945                         if (xvec->xvec->type->post_input &&
 946                             (xvec->xvec->type->post_input)(xvec->xvec,
 947                                                            &(xvec->decap),
 948                                                            skb) != 0)
 949                                 return 0;
 950                 }
 951         }
 952
 953         pol = NULL;
 954         if (sk && sk->sk_policy[dir])
 955                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
 956
 957         if (!pol)
 958                 pol = flow_cache_lookup(&fl, family,
 959                                         policy_to_flow_dir(dir),
 960                                         xfrm_policy_lookup);
 961
 962         if (!pol)
 963                 return !skb->sp || !secpath_has_tunnel(skb->sp, 0);
 964
 965         pol->curlft.use_time = (unsigned long)xtime.tv_sec;
 966
 967         if (pol->action == XFRM_POLICY_ALLOW) {
 968                 struct sec_path *sp;
 969                 static struct sec_path dummy;
 970                 int i, k;
 971
 972                 if ((sp = skb->sp) == NULL)
 973                         sp = &dummy;
 974
 975                 /* For each tunnel xfrm, find the first matching tmpl.
 976                  * For each tmpl before that, find corresponding xfrm.
 977                  * Order is _important_. Later we will implement
 978                  * some barriers, but at the moment barriers
 979                  * are implied between each two transformations.
 980                  */
 981                 for (i = pol->xfrm_nr-1, k = 0; i >= 0; i--) {
 982                         k = xfrm_policy_ok(pol->xfrm_vec+i, sp, k, family);
 983                         if (k < 0)
 984                                 goto reject;
 985                 }
 986
 987                 if (secpath_has_tunnel(sp, k))
 988                         goto reject;
 989
 990                 xfrm_pol_put(pol);
 991                 return 1;
 992         }
 993
 994 reject:
 995         xfrm_pol_put(pol);
 996         return 0;
 997 }
 998 EXPORT_SYMBOL(__xfrm_policy_check);
 999
1000 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1001 {
1002         struct flowi fl;
1003
1004         if (_decode_session(skb, &fl, family) < 0)
1005                 return 0;
1006
1007         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1008 }
1009 EXPORT_SYMBOL(__xfrm_route_forward);
1010
1011 /* Optimize later using cookies and generation ids. */
1012
1013 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1014 {
1015         if (!stale_bundle(dst))
1016                 return dst;
1017
1018         dst_release(dst);
1019         return NULL;
1020 }
1021
1022 static int stale_bundle(struct dst_entry *dst)
1023 {
1024         struct dst_entry *child = dst;
1025
1026         while (child) {
1027                 if (child->obsolete > 0 ||
1028                     (child->dev && !netif_running(child->dev)) ||
1029                     (child->xfrm && child->xfrm->km.state != XFRM_STATE_VALID)) {
1030                         return 1;
1031                 }
1032                 child = child->child;
1033         }
1034
1035         return 0;
1036 }
1037
1038 static void xfrm_dst_destroy(struct dst_entry *dst)
1039 {
1040         if (!dst->xfrm)
1041                 return;
1042         xfrm_state_put(dst->xfrm);
1043         dst->xfrm = NULL;
1044 }
1045
1046 static void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
1047                             int unregister)
1048 {
1049         if (!unregister)
1050                 return;
1051
1052         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1053                 dst->dev = &loopback_dev;
1054                 dev_hold(&loopback_dev);
1055                 dev_put(dev);
1056         }
1057 }
1058
1059 static void xfrm_link_failure(struct sk_buff *skb)
1060 {
1061         /* Impossible. Such dst must be popped before reaches point of failure. */
1062         return;
1063 }
1064
1065 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1066 {
1067         if (dst) {
1068                 if (dst->obsolete) {
1069                         dst_release(dst);
1070                         dst = NULL;
1071                 }
1072         }
1073         return dst;
1074 }
1075
1076 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1077 {
1078         int i;
1079         struct xfrm_policy *pol;
1080         struct dst_entry *dst, **dstp, *gc_list = NULL;
1081
1082         read_lock_bh(&xfrm_policy_lock);
1083         for (i=0; i<2*XFRM_POLICY_MAX; i++) {
1084                 for (pol = xfrm_policy_list[i]; pol; pol = pol->next) {
1085                         write_lock(&pol->lock);
1086                         dstp = &pol->bundles;
1087                         while ((dst=*dstp) != NULL) {
1088                                 if (func(dst)) {
1089                                         *dstp = dst->next;
1090                                         dst->next = gc_list;
1091                                         gc_list = dst;
1092                                 } else {
1093                                         dstp = &dst->next;
1094                                 }
1095                         }
1096                         write_unlock(&pol->lock);
1097                 }
1098         }
1099         read_unlock_bh(&xfrm_policy_lock);
1100
1101         while (gc_list) {
1102                 dst = gc_list;
1103                 gc_list = dst->next;
1104                 dst_free(dst);
1105         }
1106 }
1107
1108 static int unused_bundle(struct dst_entry *dst)
1109 {
1110         return !atomic_read(&dst->__refcnt);
1111 }
1112
1113 static void __xfrm_garbage_collect(void)
1114 {
1115         xfrm_prune_bundles(unused_bundle);
1116 }
1117
1118 int xfrm_flush_bundles(void)
1119 {
1120         xfrm_prune_bundles(stale_bundle);
1121         return 0;
1122 }
1123
1124 /* Well... that's _TASK_. We need to scan through transformation
1125  * list and figure out what mss tcp should generate in order to
1126  * final datagram fit to mtu. Mama mia... :-)
1127  *
1128  * Apparently, some easy way exists, but we used to choose the most
1129  * bizarre ones. :-) So, raising Kalashnikov... tra-ta-ta.
1130  *
1131  * Consider this function as something like dark humour. :-)
1132  */
1133 static int xfrm_get_mss(struct dst_entry *dst, u32 mtu)
1134 {
1135         int res = mtu - dst->header_len;
1136
1137         for (;;) {
1138                 struct dst_entry *d = dst;
1139                 int m = res;
1140
1141                 do {
1142                         struct xfrm_state *x = d->xfrm;
1143                         if (x) {
1144                                 spin_lock_bh(&x->lock);
1145                                 if (x->km.state == XFRM_STATE_VALID &&
1146                                     x->type && x->type->get_max_size)
1147                                         m = x->type->get_max_size(d->xfrm, m);
1148                                 else
1149                                         m += x->props.header_len;
1150                                 spin_unlock_bh(&x->lock);
1151                         }
1152                 } while ((d = d->child) != NULL);
1153
1154                 if (m <= mtu)
1155                         break;
1156                 res -= (m - mtu);
1157                 if (res < 88)
1158                         return mtu;
1159         }
1160
1161         return res + dst->header_len;
1162 }
1163
1164 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
1165 {
1166         int err = 0;
1167         if (unlikely(afinfo == NULL))
1168                 return -EINVAL;
1169         if (unlikely(afinfo->family >= NPROTO))
1170                 return -EAFNOSUPPORT;
1171         write_lock(&xfrm_policy_afinfo_lock);
1172         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
1173                 err = -ENOBUFS;
1174         else {
1175                 struct dst_ops *dst_ops = afinfo->dst_ops;
1176                 if (likely(dst_ops->kmem_cachep == NULL))
1177                         dst_ops->kmem_cachep = xfrm_dst_cache;
1178                 if (likely(dst_ops->check == NULL))
1179                         dst_ops->check = xfrm_dst_check;
1180                 if (likely(dst_ops->destroy == NULL))
1181                         dst_ops->destroy = xfrm_dst_destroy;
1182                 if (likely(dst_ops->ifdown == NULL))
1183                         dst_ops->ifdown = xfrm_dst_ifdown;
1184                 if (likely(dst_ops->negative_advice == NULL))
1185                         dst_ops->negative_advice = xfrm_negative_advice;
1186                 if (likely(dst_ops->link_failure == NULL))
1187                         dst_ops->link_failure = xfrm_link_failure;
1188                 if (likely(dst_ops->get_mss == NULL))
1189                         dst_ops->get_mss = xfrm_get_mss;
1190                 if (likely(afinfo->garbage_collect == NULL))
1191                         afinfo->garbage_collect = __xfrm_garbage_collect;
1192                 xfrm_policy_afinfo[afinfo->family] = afinfo;
1193         }
1194         write_unlock(&xfrm_policy_afinfo_lock);
1195         return err;
1196 }
1197 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
1198
1199 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
1200 {
1201         int err = 0;
1202         if (unlikely(afinfo == NULL))
1203                 return -EINVAL;
1204         if (unlikely(afinfo->family >= NPROTO))
1205                 return -EAFNOSUPPORT;
1206         write_lock(&xfrm_policy_afinfo_lock);
1207         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
1208                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
1209                         err = -EINVAL;
1210                 else {
1211                         struct dst_ops *dst_ops = afinfo->dst_ops;
1212                         xfrm_policy_afinfo[afinfo->family] = NULL;
1213                         dst_ops->kmem_cachep = NULL;
1214                         dst_ops->check = NULL;
1215                         dst_ops->destroy = NULL;
1216                         dst_ops->ifdown = NULL;
1217                         dst_ops->negative_advice = NULL;
1218                         dst_ops->link_failure = NULL;
1219                         dst_ops->get_mss = NULL;
1220                         afinfo->garbage_collect = NULL;
1221                 }
1222         }
1223         write_unlock(&xfrm_policy_afinfo_lock);
1224         return err;
1225 }
1226 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
1227
1228 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
1229 {
1230         struct xfrm_policy_afinfo *afinfo;
1231         if (unlikely(family >= NPROTO))
1232                 return NULL;
1233         read_lock(&xfrm_policy_afinfo_lock);
1234         afinfo = xfrm_policy_afinfo[family];
1235         if (likely(afinfo != NULL))
1236                 read_lock(&afinfo->lock);
1237         read_unlock(&xfrm_policy_afinfo_lock);
1238         return afinfo;
1239 }
1240
1241 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
1242 {
1243         if (unlikely(afinfo == NULL))
1244                 return;
1245         read_unlock(&afinfo->lock);
1246 }
1247
1248 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
1249 {
1250         switch (event) {
1251         case NETDEV_DOWN:
1252                 xfrm_flush_bundles();
1253         }
1254         return NOTIFY_DONE;
1255 }
1256
1257 static struct notifier_block xfrm_dev_notifier = {
1258         xfrm_dev_event,
1259         NULL,
1260         0
1261 };
1262
1263 static void __init xfrm_policy_init(void)
1264 {
1265         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
1266                                            sizeof(struct xfrm_dst),
1267                                            0, SLAB_HWCACHE_ALIGN,
1268                                            NULL, NULL);
1269         if (!xfrm_dst_cache)
1270                 panic("XFRM: failed to allocate xfrm_dst_cache\n");
1271
1272         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task, NULL);
1273         register_netdevice_notifier(&xfrm_dev_notifier);
1274 }
1275
1276 void __init xfrm_init(void)
1277 {
1278         xfrm_state_init();
1279         xfrm_policy_init();
1280         xfrm_input_init();
1281 }
1282