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