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         if (xp->dead)
 152                 goto out;
 153
 154         dir = xp->index & 7;
 155
 156         if (xp->lft.hard_add_expires_seconds) {
 157                 long tmo = xp->lft.hard_add_expires_seconds +
 158                         xp->curlft.add_time - now;
 159                 if (tmo <= 0)
 160                         goto expired;
 161                 if (tmo < next)
 162                         next = tmo;
 163         }
 164         if (xp->lft.hard_use_expires_seconds) {
 165                 long tmo = xp->lft.hard_use_expires_seconds +
 166                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 167                 if (tmo <= 0)
 168                         goto expired;
 169                 if (tmo < next)
 170                         next = tmo;
 171         }
 172         if (xp->lft.soft_add_expires_seconds) {
 173                 long tmo = xp->lft.soft_add_expires_seconds +
 174                         xp->curlft.add_time - now;
 175                 if (tmo <= 0) {
 176                         warn = 1;
 177                         tmo = XFRM_KM_TIMEOUT;
 178                 }
 179                 if (tmo < next)
 180                         next = tmo;
 181         }
 182         if (xp->lft.soft_use_expires_seconds) {
 183                 long tmo = xp->lft.soft_use_expires_seconds +
 184                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 185                 if (tmo <= 0) {
 186                         warn = 1;
 187                         tmo = XFRM_KM_TIMEOUT;
 188                 }
 189                 if (tmo < next)
 190                         next = tmo;
 191         }
 192
 193         if (warn)
 194                 km_policy_expired(xp, dir, 0);
 195         if (next != LONG_MAX &&
 196             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 197                 xfrm_pol_hold(xp);
 198
 199 out:
 200         xfrm_pol_put(xp);
 201         return;
 202
 203 expired:
 204         km_policy_expired(xp, dir, 1);
 205         xfrm_policy_delete(xp, dir);
 206         xfrm_pol_put(xp);
 207 }
 208
 209
 210 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 211  * SPD calls.
 212  */
 213
 214 struct xfrm_policy *xfrm_policy_alloc(int gfp)
 215 {
 216         struct xfrm_policy *policy;
 217
 218         policy = kmalloc(sizeof(struct xfrm_policy), gfp);
 219
 220         if (policy) {
 221                 memset(policy, 0, sizeof(struct xfrm_policy));
 222                 atomic_set(&policy->refcnt, 1);
 223                 policy->lock = RW_LOCK_UNLOCKED;
 224                 init_timer(&policy->timer);
 225                 policy->timer.data = (unsigned long)policy;
 226                 policy->timer.function = xfrm_policy_timer;
 227         }
 228         return policy;
 229 }
 230
 231 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 232
 233 void __xfrm_policy_destroy(struct xfrm_policy *policy)
 234 {
 235         if (!policy->dead)
 236                 BUG();
 237
 238         if (policy->bundles)
 239                 BUG();
 240
 241         if (del_timer(&policy->timer))
 242                 BUG();
 243
 244         kfree(policy);
 245 }
 246
 247 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
 248 {
 249         struct dst_entry *dst;
 250
 251         while ((dst = policy->bundles) != NULL) {
 252                 policy->bundles = dst->next;
 253                 dst_free(dst);
 254         }
 255
 256         if (del_timer(&policy->timer))
 257                 atomic_dec(&policy->refcnt);
 258
 259         if (atomic_read(&policy->refcnt) > 1)
 260                 flow_cache_flush();
 261
 262         xfrm_pol_put(policy);
 263 }
 264
 265 static void xfrm_policy_gc_task(void *data)
 266 {
 267         struct xfrm_policy *policy;
 268         struct list_head *entry, *tmp;
 269         struct list_head gc_list = LIST_HEAD_INIT(gc_list);
 270
 271         spin_lock_bh(&xfrm_policy_gc_lock);
 272         list_splice_init(&xfrm_policy_gc_list, &gc_list);
 273         spin_unlock_bh(&xfrm_policy_gc_lock);
 274
 275         list_for_each_safe(entry, tmp, &gc_list) {
 276                 policy = list_entry(entry, struct xfrm_policy, list);
 277                 xfrm_policy_gc_kill(policy);
 278         }
 279 }
 280
 281 /* Rule must be locked. Release descentant resources, announce
 282  * entry dead. The rule must be unlinked from lists to the moment.
 283  */
 284
 285 void xfrm_policy_kill(struct xfrm_policy *policy)
 286 {
 287         write_lock_bh(&policy->lock);
 288         if (policy->dead)
 289                 goto out;
 290
 291         policy->dead = 1;
 292
 293         spin_lock(&xfrm_policy_gc_lock);
 294         list_add(&policy->list, &xfrm_policy_gc_list);
 295         spin_unlock(&xfrm_policy_gc_lock);
 296         schedule_work(&xfrm_policy_gc_work);
 297
 298 out:
 299         write_unlock_bh(&policy->lock);
 300 }
 301
 302 /* Generate new index... KAME seems to generate them ordered by cost
 303  * of an absolute inpredictability of ordering of rules. This will not pass. */
 304 static u32 xfrm_gen_index(int dir)
 305 {
 306         u32 idx;
 307         struct xfrm_policy *p;
 308         static u32 idx_generator;
 309
 310         for (;;) {
 311                 idx = (idx_generator | dir);
 312                 idx_generator += 8;
 313                 if (idx == 0)
 314                         idx = 8;
 315                 for (p = xfrm_policy_list[dir]; p; p = p->next) {
 316                         if (p->index == idx)
 317                                 break;
 318                 }
 319                 if (!p)
 320                         return idx;
 321         }
 322 }
 323
 324 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 325 {
 326         struct xfrm_policy *pol, **p;
 327         struct xfrm_policy *delpol = NULL;
 328         struct xfrm_policy **newpos = NULL;
 329
 330         write_lock_bh(&xfrm_policy_lock);
 331         for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) {
 332                 if (!delpol && memcmp(&policy->selector, &pol->selector, sizeof(pol->selector)) == 0) {
 333                         if (excl) {
 334                                 write_unlock_bh(&xfrm_policy_lock);
 335                                 return -EEXIST;
 336                         }
 337                         *p = pol->next;
 338                         delpol = pol;
 339                         if (policy->priority > pol->priority)
 340                                 continue;
 341                 } else if (policy->priority >= pol->priority)
 342                         continue;
 343                 if (!newpos)
 344                         newpos = p;
 345                 if (delpol)
 346                         break;
 347         }
 348         if (newpos)
 349                 p = newpos;
 350         xfrm_pol_hold(policy);
 351         policy->next = *p;
 352         *p = policy;
 353         atomic_inc(&flow_cache_genid);
 354         policy->index = delpol ? delpol->index : xfrm_gen_index(dir);
 355         policy->curlft.add_time = (unsigned long)xtime.tv_sec;
 356         policy->curlft.use_time = 0;
 357         if (!mod_timer(&policy->timer, jiffies + HZ))
 358                 xfrm_pol_hold(policy);
 359         write_unlock_bh(&xfrm_policy_lock);
 360
 361         if (delpol) {
 362                 xfrm_policy_kill(delpol);
 363         }
 364         return 0;
 365 }
 366
 367 struct xfrm_policy *xfrm_policy_bysel(int dir, struct xfrm_selector *sel,
 368                                       int delete)
 369 {
 370         struct xfrm_policy *pol, **p;
 371
 372         write_lock_bh(&xfrm_policy_lock);
 373         for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) {
 374                 if (memcmp(sel, &pol->selector, sizeof(*sel)) == 0) {
 375                         xfrm_pol_hold(pol);
 376                         if (delete)
 377                                 *p = pol->next;
 378                         break;
 379                 }
 380         }
 381         write_unlock_bh(&xfrm_policy_lock);
 382
 383         if (pol && delete) {
 384                 atomic_inc(&flow_cache_genid);
 385                 xfrm_policy_kill(pol);
 386         }
 387         return pol;
 388 }
 389
 390 struct xfrm_policy *xfrm_policy_byid(int dir, u32 id, int delete)
 391 {
 392         struct xfrm_policy *pol, **p;
 393
 394         write_lock_bh(&xfrm_policy_lock);
 395         for (p = &xfrm_policy_list[id & 7]; (pol=*p)!=NULL; p = &pol->next) {
 396                 if (pol->index == id) {
 397                         xfrm_pol_hold(pol);
 398                         if (delete)
 399                                 *p = pol->next;
 400                         break;
 401                 }
 402         }
 403         write_unlock_bh(&xfrm_policy_lock);
 404
 405         if (pol && delete) {
 406                 atomic_inc(&flow_cache_genid);
 407                 xfrm_policy_kill(pol);
 408         }
 409         return pol;
 410 }
 411
 412 void xfrm_policy_flush(void)
 413 {
 414         struct xfrm_policy *xp;
 415         int dir;
 416
 417         write_lock_bh(&xfrm_policy_lock);
 418         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 419                 while ((xp = xfrm_policy_list[dir]) != NULL) {
 420                         xfrm_policy_list[dir] = xp->next;
 421                         write_unlock_bh(&xfrm_policy_lock);
 422
 423                         xfrm_policy_kill(xp);
 424
 425                         write_lock_bh(&xfrm_policy_lock);
 426                 }
 427         }
 428         atomic_inc(&flow_cache_genid);
 429         write_unlock_bh(&xfrm_policy_lock);
 430 }
 431
 432 int xfrm_policy_walk(int (*func)(struct xfrm_policy *, int, int, void*),
 433                      void *data)
 434 {
 435         struct xfrm_policy *xp;
 436         int dir;
 437         int count = 0;
 438         int error = 0;
 439
 440         read_lock_bh(&xfrm_policy_lock);
 441         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
 442                 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next)
 443                         count++;
 444         }
 445
 446         if (count == 0) {
 447                 error = -ENOENT;
 448                 goto out;
 449         }
 450
 451         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
 452                 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next) {
 453                         error = func(xp, dir%XFRM_POLICY_MAX, --count, data);
 454                         if (error)
 455                                 goto out;
 456                 }
 457         }
 458
 459 out:
 460         read_unlock_bh(&xfrm_policy_lock);
 461         return error;
 462 }
 463
 464
 465 /* Find policy to apply to this flow. */
 466
 467 static void xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
 468                                void **objp, atomic_t **obj_refp)
 469 {
 470         struct xfrm_policy *pol;
 471
 472         read_lock_bh(&xfrm_policy_lock);
 473         for (pol = xfrm_policy_list[dir]; pol; pol = pol->next) {
 474                 struct xfrm_selector *sel = &pol->selector;
 475                 int match;
 476
 477                 if (pol->family != family)
 478                         continue;
 479
 480                 match = xfrm_selector_match(sel, fl, family);
 481                 if (match) {
 482                         xfrm_pol_hold(pol);
 483                         break;
 484                 }
 485         }
 486         read_unlock_bh(&xfrm_policy_lock);
 487         if ((*objp = (void *) pol) != NULL)
 488                 *obj_refp = &pol->refcnt;
 489 }
 490
 491 struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
 492 {
 493         struct xfrm_policy *pol;
 494
 495         read_lock_bh(&xfrm_policy_lock);
 496         if ((pol = sk->sk_policy[dir]) != NULL) {
 497                 int match = xfrm_selector_match(&pol->selector, fl,
 498                                                 sk->sk_family);
 499                 if (match)
 500                         xfrm_pol_hold(pol);
 501                 else
 502                         pol = NULL;
 503         }
 504         read_unlock_bh(&xfrm_policy_lock);
 505         return pol;
 506 }
 507
 508 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
 509 {
 510         pol->next = xfrm_policy_list[dir];
 511         xfrm_policy_list[dir] = pol;
 512         xfrm_pol_hold(pol);
 513 }
 514
 515 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
 516                                                 int dir)
 517 {
 518         struct xfrm_policy **polp;
 519
 520         for (polp = &xfrm_policy_list[dir];
 521              *polp != NULL; polp = &(*polp)->next) {
 522                 if (*polp == pol) {
 523                         *polp = pol->next;
 524                         return pol;
 525                 }
 526         }
 527         return NULL;
 528 }
 529
 530 void xfrm_policy_delete(struct xfrm_policy *pol, int dir)
 531 {
 532         write_lock_bh(&xfrm_policy_lock);
 533         pol = __xfrm_policy_unlink(pol, dir);
 534         write_unlock_bh(&xfrm_policy_lock);
 535         if (pol)
 536                 xfrm_policy_kill(pol);
 537 }
 538
 539 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
 540 {
 541         struct xfrm_policy *old_pol;
 542
 543         write_lock_bh(&xfrm_policy_lock);
 544         old_pol = sk->sk_policy[dir];
 545         sk->sk_policy[dir] = pol;
 546         if (pol) {
 547                 pol->curlft.add_time = (unsigned long)xtime.tv_sec;
 548                 pol->index = xfrm_gen_index(XFRM_POLICY_MAX+dir);
 549                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
 550         }
 551         if (old_pol)
 552                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
 553         write_unlock_bh(&xfrm_policy_lock);
 554
 555         if (old_pol) {
 556                 xfrm_policy_kill(old_pol);
 557         }
 558         return 0;
 559 }
 560
 561 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
 562 {
 563         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
 564
 565         if (newp) {
 566                 newp->selector = old->selector;
 567                 newp->lft = old->lft;
 568                 newp->curlft = old->curlft;
 569                 newp->action = old->action;
 570                 newp->flags = old->flags;
 571                 newp->xfrm_nr = old->xfrm_nr;
 572                 newp->index = old->index;
 573                 memcpy(newp->xfrm_vec, old->xfrm_vec,
 574                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
 575                 write_lock_bh(&xfrm_policy_lock);
 576                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
 577                 write_unlock_bh(&xfrm_policy_lock);
 578                 xfrm_pol_put(newp);
 579         }
 580         return newp;
 581 }
 582
 583 int __xfrm_sk_clone_policy(struct sock *sk)
 584 {
 585         struct xfrm_policy *p0 = sk->sk_policy[0],
 586                            *p1 = sk->sk_policy[1];
 587
 588         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
 589         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
 590                 return -ENOMEM;
 591         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
 592                 return -ENOMEM;
 593         return 0;
 594 }
 595
 596 /* Resolve list of templates for the flow, given policy. */
 597
 598 static int
 599 xfrm_tmpl_resolve(struct xfrm_policy *policy, struct flowi *fl,
 600                   struct xfrm_state **xfrm,
 601                   unsigned short family)
 602 {
 603         int nx;
 604         int i, error;
 605         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
 606         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
 607
 608         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
 609                 struct xfrm_state *x;
 610                 xfrm_address_t *remote = daddr;
 611                 xfrm_address_t *local  = saddr;
 612                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
 613
 614                 if (tmpl->mode) {
 615                         remote = &tmpl->id.daddr;
 616                         local = &tmpl->saddr;
 617                 }
 618
 619                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
 620
 621                 if (x && x->km.state == XFRM_STATE_VALID) {
 622                         xfrm[nx++] = x;
 623                         daddr = remote;
 624                         saddr = local;
 625                         continue;
 626                 }
 627                 if (x) {
 628                         error = (x->km.state == XFRM_STATE_ERROR ?
 629                                  -EINVAL : -EAGAIN);
 630                         xfrm_state_put(x);
 631                 }
 632
 633                 if (!tmpl->optional)
 634                         goto fail;
 635         }
 636         return nx;
 637
 638 fail:
 639         for (nx--; nx>=0; nx--)
 640                 xfrm_state_put(xfrm[nx]);
 641         return error;
 642 }
 643
 644 /* Check that the bundle accepts the flow and its components are
 645  * still valid.
 646  */
 647
 648 static struct dst_entry *
 649 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
 650 {
 651         struct dst_entry *x;
 652         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
 653         if (unlikely(afinfo == NULL))
 654                 return ERR_PTR(-EINVAL);
 655         x = afinfo->find_bundle(fl, policy);
 656         xfrm_policy_put_afinfo(afinfo);
 657         return x;
 658 }
 659
 660 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
 661  * all the metrics... Shortly, bundle a bundle.
 662  */
 663
 664 static int
 665 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
 666                    struct flowi *fl, struct dst_entry **dst_p,
 667                    unsigned short family)
 668 {
 669         int err;
 670         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
 671         if (unlikely(afinfo == NULL))
 672                 return -EINVAL;
 673         err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
 674         xfrm_policy_put_afinfo(afinfo);
 675         return err;
 676 }
 677
 678 static inline int policy_to_flow_dir(int dir)
 679 {
 680         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
 681             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
 682             XFRM_POLICY_FWD == FLOW_DIR_FWD)
 683                 return dir;
 684         switch (dir) {
 685         default:
 686         case XFRM_POLICY_IN:
 687                 return FLOW_DIR_IN;
 688         case XFRM_POLICY_OUT:
 689                 return FLOW_DIR_OUT;
 690         case XFRM_POLICY_FWD:
 691                 return FLOW_DIR_FWD;
 692         };
 693 }
 694
 695 static int stale_bundle(struct dst_entry *dst);
 696
 697 /* Main function: finds/creates a bundle for given flow.
 698  *
 699  * At the moment we eat a raw IP route. Mostly to speed up lookups
 700  * on interfaces with disabled IPsec.
 701  */
 702 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
 703                 struct sock *sk, int flags)
 704 {
 705         struct xfrm_policy *policy;
 706         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
 707         struct rtable *rt = (struct rtable*)*dst_p;
 708         struct dst_entry *dst;
 709         int nx = 0;
 710         int err;
 711         u32 genid;
 712         u16 family = (*dst_p)->ops->family;
 713
 714         switch (family) {
 715         case AF_INET:
 716                 if (!fl->fl4_src)
 717                         fl->fl4_src = rt->rt_src;
 718                 if (!fl->fl4_dst)
 719                         fl->fl4_dst = rt->rt_dst;
 720         case AF_INET6:
 721                 /* Still not clear... */
 722         default:
 723                 /* nothing */;
 724         }
 725
 726 restart:
 727         genid = atomic_read(&flow_cache_genid);
 728         policy = NULL;
 729         if (sk && sk->sk_policy[1])
 730                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
 731
 732         if (!policy) {
 733                 /* To accelerate a bit...  */
 734                 if ((rt->u.dst.flags & DST_NOXFRM) || !xfrm_policy_list[XFRM_POLICY_OUT])
 735                         return 0;
 736
 737                 policy = flow_cache_lookup(fl, family,
 738                                            policy_to_flow_dir(XFRM_POLICY_OUT),
 739                                            xfrm_policy_lookup);
 740         }
 741
 742         if (!policy)
 743                 return 0;
 744
 745         policy->curlft.use_time = (unsigned long)xtime.tv_sec;
 746
 747         switch (policy->action) {
 748         case XFRM_POLICY_BLOCK:
 749                 /* Prohibit the flow */
 750                 xfrm_pol_put(policy);
 751                 return -EPERM;
 752
 753         case XFRM_POLICY_ALLOW:
 754                 if (policy->xfrm_nr == 0) {
 755                         /* Flow passes not transformed. */
 756                         xfrm_pol_put(policy);
 757                         return 0;
 758                 }
 759
 760                 /* Try to find matching bundle.
 761                  *
 762                  * LATER: help from flow cache. It is optional, this
 763                  * is required only for output policy.
 764                  */
 765                 dst = xfrm_find_bundle(fl, policy, family);
 766                 if (IS_ERR(dst)) {
 767                         xfrm_pol_put(policy);
 768                         return PTR_ERR(dst);
 769                 }
 770
 771                 if (dst)
 772                         break;
 773
 774                 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
 775
 776                 if (unlikely(nx<0)) {
 777                         err = nx;
 778                         if (err == -EAGAIN && flags) {
 779                                 DECLARE_WAITQUEUE(wait, current);
 780
 781                                 add_wait_queue(&km_waitq, &wait);
 782                                 set_current_state(TASK_INTERRUPTIBLE);
 783                                 schedule();
 784                                 set_current_state(TASK_RUNNING);
 785                                 remove_wait_queue(&km_waitq, &wait);
 786
 787                                 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
 788
 789                                 if (nx == -EAGAIN && signal_pending(current)) {
 790                                         err = -ERESTART;
 791                                         goto error;
 792                                 }
 793                                 if (nx == -EAGAIN ||
 794                                     genid != atomic_read(&flow_cache_genid)) {
 795                                         xfrm_pol_put(policy);
 796                                         goto restart;
 797                                 }
 798                                 err = nx;
 799                         }
 800                         if (err < 0)
 801                                 goto error;
 802                 }
 803                 if (nx == 0) {
 804                         /* Flow passes not transformed. */
 805                         xfrm_pol_put(policy);
 806                         return 0;
 807                 }
 808
 809                 dst = &rt->u.dst;
 810                 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
 811
 812                 if (unlikely(err)) {
 813                         int i;
 814                         for (i=0; i<nx; i++)
 815                                 xfrm_state_put(xfrm[i]);
 816                         goto error;
 817                 }
 818
 819                 write_lock_bh(&policy->lock);
 820                 if (unlikely(policy->dead || stale_bundle(dst))) {
 821                         /* Wow! While we worked on resolving, this
 822                          * policy has gone. Retry. It is not paranoia,
 823                          * we just cannot enlist new bundle to dead object.
 824                          * We can't enlist stable bundles either.
 825                          */
 826                         write_unlock_bh(&policy->lock);
 827
 828                         xfrm_pol_put(policy);
 829                         if (dst)
 830                                 dst_free(dst);
 831                         goto restart;
 832                 }
 833                 dst->next = policy->bundles;
 834                 policy->bundles = dst;
 835                 dst_hold(dst);
 836                 write_unlock_bh(&policy->lock);
 837         }
 838         *dst_p = dst;
 839         ip_rt_put(rt);
 840         xfrm_pol_put(policy);
 841         return 0;
 842
 843 error:
 844         ip_rt_put(rt);
 845         xfrm_pol_put(policy);
 846         *dst_p = NULL;
 847         return err;
 848 }
 849
 850 /* When skb is transformed back to its "native" form, we have to
 851  * check policy restrictions. At the moment we make this in maximally
 852  * stupid way. Shame on me. :-) Of course, connected sockets must
 853  * have policy cached at them.
 854  */
 855
 856 static inline int
 857 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
 858               unsigned short family)
 859 {
 860         if (xfrm_state_kern(x))
 861                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, family);
 862         return  x->id.proto == tmpl->id.proto &&
 863                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
 864                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
 865                 x->props.mode == tmpl->mode &&
 866                 (tmpl->aalgos & (1<<x->props.aalgo)) &&
 867                 !(x->props.mode && xfrm_state_addr_cmp(tmpl, x, family));
 868 }
 869
 870 static inline int
 871 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
 872                unsigned short family)
 873 {
 874         int idx = start;
 875
 876         if (tmpl->optional) {
 877                 if (!tmpl->mode)
 878                         return start;
 879         } else
 880                 start = -1;
 881         for (; idx < sp->len; idx++) {
 882                 if (xfrm_state_ok(tmpl, sp->x[idx].xvec, family))
 883                         return ++idx;
 884                 if (sp->x[idx].xvec->props.mode)
 885                         break;
 886         }
 887         return start;
 888 }
 889
 890 static int
 891 _decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
 892 {
 893         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
 894
 895         if (unlikely(afinfo == NULL))
 896                 return -EAFNOSUPPORT;
 897
 898         afinfo->decode_session(skb, fl);
 899         xfrm_policy_put_afinfo(afinfo);
 900         return 0;
 901 }
 902
 903 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
 904                         unsigned short family)
 905 {
 906         struct xfrm_policy *pol;
 907         struct flowi fl;
 908
 909         if (_decode_session(skb, &fl, family) < 0)
 910                 return 0;
 911
 912         /* First, check used SA against their selectors. */
 913         if (skb->sp) {
 914                 int i;
 915
 916                 for (i=skb->sp->len-1; i>=0; i--) {
 917                   struct sec_decap_state *xvec = &(skb->sp->x[i]);
 918                         if (!xfrm_selector_match(&xvec->xvec->sel, &fl, family))
 919                                 return 0;
 920
 921                         /* If there is a post_input processor, try running it */
 922                         if (xvec->xvec->type->post_input &&
 923                             (xvec->xvec->type->post_input)(xvec->xvec,
 924                                                            &(xvec->decap),
 925                                                            skb) != 0)
 926                                 return 0;
 927                 }
 928         }
 929
 930         pol = NULL;
 931         if (sk && sk->sk_policy[dir])
 932                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
 933
 934         if (!pol)
 935                 pol = flow_cache_lookup(&fl, family,
 936                                         policy_to_flow_dir(dir),
 937                                         xfrm_policy_lookup);
 938
 939         if (!pol)
 940                 return !skb->sp;
 941
 942         pol->curlft.use_time = (unsigned long)xtime.tv_sec;
 943
 944         if (pol->action == XFRM_POLICY_ALLOW) {
 945                 struct sec_path *sp;
 946                 static struct sec_path dummy;
 947                 int i, k;
 948
 949                 if ((sp = skb->sp) == NULL)
 950                         sp = &dummy;
 951
 952                 /* For each tunnel xfrm, find the first matching tmpl.
 953                  * For each tmpl before that, find corresponding xfrm.
 954                  * Order is _important_. Later we will implement
 955                  * some barriers, but at the moment barriers
 956                  * are implied between each two transformations.
 957                  */
 958                 for (i = pol->xfrm_nr-1, k = 0; i >= 0; i--) {
 959                         k = xfrm_policy_ok(pol->xfrm_vec+i, sp, k, family);
 960                         if (k < 0)
 961                                 goto reject;
 962                 }
 963
 964                 for (; k < sp->len; k++) {
 965                         if (sp->x[k].xvec->props.mode)
 966                                 goto reject;
 967                 }
 968
 969                 xfrm_pol_put(pol);
 970                 return 1;
 971         }
 972
 973 reject:
 974         xfrm_pol_put(pol);
 975         return 0;
 976 }
 977
 978 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
 979 {
 980         struct flowi fl;
 981
 982         if (_decode_session(skb, &fl, family) < 0)
 983                 return 0;
 984
 985         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
 986 }
 987
 988 /* Optimize later using cookies and generation ids. */
 989
 990 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
 991 {
 992         if (!stale_bundle(dst))
 993                 return dst;
 994
 995         dst_release(dst);
 996         return NULL;
 997 }
 998
 999 static int stale_bundle(struct dst_entry *dst)
1000 {
1001         struct dst_entry *child = dst;
1002
1003         while (child) {
1004                 if (child->obsolete > 0 ||
1005                     (child->dev && !netif_running(child->dev)) ||
1006                     (child->xfrm && child->xfrm->km.state != XFRM_STATE_VALID)) {
1007                         return 1;
1008                 }
1009                 child = child->child;
1010         }
1011
1012         return 0;
1013 }
1014
1015 static void xfrm_dst_destroy(struct dst_entry *dst)
1016 {
1017         xfrm_state_put(dst->xfrm);
1018         dst->xfrm = NULL;
1019 }
1020
1021 static void xfrm_link_failure(struct sk_buff *skb)
1022 {
1023         /* Impossible. Such dst must be popped before reaches point of failure. */
1024         return;
1025 }
1026
1027 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1028 {
1029         if (dst) {
1030                 if (dst->obsolete) {
1031                         dst_release(dst);
1032                         dst = NULL;
1033                 }
1034         }
1035         return dst;
1036 }
1037
1038 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1039 {
1040         int i;
1041         struct xfrm_policy *pol;
1042         struct dst_entry *dst, **dstp, *gc_list = NULL;
1043
1044         read_lock_bh(&xfrm_policy_lock);
1045         for (i=0; i<2*XFRM_POLICY_MAX; i++) {
1046                 for (pol = xfrm_policy_list[i]; pol; pol = pol->next) {
1047                         write_lock(&pol->lock);
1048                         dstp = &pol->bundles;
1049                         while ((dst=*dstp) != NULL) {
1050                                 if (func(dst)) {
1051                                         *dstp = dst->next;
1052                                         dst->next = gc_list;
1053                                         gc_list = dst;
1054                                 } else {
1055                                         dstp = &dst->next;
1056                                 }
1057                         }
1058                         write_unlock(&pol->lock);
1059                 }
1060         }
1061         read_unlock_bh(&xfrm_policy_lock);
1062
1063         while (gc_list) {
1064                 dst = gc_list;
1065                 gc_list = dst->next;
1066                 dst_free(dst);
1067         }
1068 }
1069
1070 static int unused_bundle(struct dst_entry *dst)
1071 {
1072         return !atomic_read(&dst->__refcnt);
1073 }
1074
1075 static void __xfrm_garbage_collect(void)
1076 {
1077         xfrm_prune_bundles(unused_bundle);
1078 }
1079
1080 int xfrm_flush_bundles(void)
1081 {
1082         xfrm_prune_bundles(stale_bundle);
1083         return 0;
1084 }
1085
1086 /* Well... that's _TASK_. We need to scan through transformation
1087  * list and figure out what mss tcp should generate in order to
1088  * final datagram fit to mtu. Mama mia... :-)
1089  *
1090  * Apparently, some easy way exists, but we used to choose the most
1091  * bizarre ones. :-) So, raising Kalashnikov... tra-ta-ta.
1092  *
1093  * Consider this function as something like dark humour. :-)
1094  */
1095 static int xfrm_get_mss(struct dst_entry *dst, u32 mtu)
1096 {
1097         int res = mtu - dst->header_len;
1098
1099         for (;;) {
1100                 struct dst_entry *d = dst;
1101                 int m = res;
1102
1103                 do {
1104                         struct xfrm_state *x = d->xfrm;
1105                         if (x) {
1106                                 spin_lock_bh(&x->lock);
1107                                 if (x->km.state == XFRM_STATE_VALID &&
1108                                     x->type && x->type->get_max_size)
1109                                         m = x->type->get_max_size(d->xfrm, m);
1110                                 else
1111                                         m += x->props.header_len;
1112                                 spin_unlock_bh(&x->lock);
1113                         }
1114                 } while ((d = d->child) != NULL);
1115
1116                 if (m <= mtu)
1117                         break;
1118                 res -= (m - mtu);
1119                 if (res < 88)
1120                         return mtu;
1121         }
1122
1123         return res + dst->header_len;
1124 }
1125
1126 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
1127 {
1128         int err = 0;
1129         if (unlikely(afinfo == NULL))
1130                 return -EINVAL;
1131         if (unlikely(afinfo->family >= NPROTO))
1132                 return -EAFNOSUPPORT;
1133         write_lock(&xfrm_policy_afinfo_lock);
1134         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
1135                 err = -ENOBUFS;
1136         else {
1137                 struct dst_ops *dst_ops = afinfo->dst_ops;
1138                 if (likely(dst_ops->kmem_cachep == NULL))
1139                         dst_ops->kmem_cachep = xfrm_dst_cache;
1140                 if (likely(dst_ops->check == NULL))
1141                         dst_ops->check = xfrm_dst_check;
1142                 if (likely(dst_ops->destroy == NULL))
1143                         dst_ops->destroy = xfrm_dst_destroy;
1144                 if (likely(dst_ops->negative_advice == NULL))
1145                         dst_ops->negative_advice = xfrm_negative_advice;
1146                 if (likely(dst_ops->link_failure == NULL))
1147                         dst_ops->link_failure = xfrm_link_failure;
1148                 if (likely(dst_ops->get_mss == NULL))
1149                         dst_ops->get_mss = xfrm_get_mss;
1150                 if (likely(afinfo->garbage_collect == NULL))
1151                         afinfo->garbage_collect = __xfrm_garbage_collect;
1152                 xfrm_policy_afinfo[afinfo->family] = afinfo;
1153         }
1154         write_unlock(&xfrm_policy_afinfo_lock);
1155         return err;
1156 }
1157
1158 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
1159 {
1160         int err = 0;
1161         if (unlikely(afinfo == NULL))
1162                 return -EINVAL;
1163         if (unlikely(afinfo->family >= NPROTO))
1164                 return -EAFNOSUPPORT;
1165         write_lock(&xfrm_policy_afinfo_lock);
1166         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
1167                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
1168                         err = -EINVAL;
1169                 else {
1170                         struct dst_ops *dst_ops = afinfo->dst_ops;
1171                         xfrm_policy_afinfo[afinfo->family] = NULL;
1172                         dst_ops->kmem_cachep = NULL;
1173                         dst_ops->check = NULL;
1174                         dst_ops->destroy = NULL;
1175                         dst_ops->negative_advice = NULL;
1176                         dst_ops->link_failure = NULL;
1177                         dst_ops->get_mss = NULL;
1178                         afinfo->garbage_collect = NULL;
1179                 }
1180         }
1181         write_unlock(&xfrm_policy_afinfo_lock);
1182         return err;
1183 }
1184
1185 struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
1186 {
1187         struct xfrm_policy_afinfo *afinfo;
1188         if (unlikely(family >= NPROTO))
1189                 return NULL;
1190         read_lock(&xfrm_policy_afinfo_lock);
1191         afinfo = xfrm_policy_afinfo[family];
1192         if (likely(afinfo != NULL))
1193                 read_lock(&afinfo->lock);
1194         read_unlock(&xfrm_policy_afinfo_lock);
1195         return afinfo;
1196 }
1197
1198 void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
1199 {
1200         if (unlikely(afinfo == NULL))
1201                 return;
1202         read_unlock(&afinfo->lock);
1203 }
1204
1205 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
1206 {
1207         switch (event) {
1208         case NETDEV_DOWN:
1209                 xfrm_flush_bundles();
1210         }
1211         return NOTIFY_DONE;
1212 }
1213
1214 struct notifier_block xfrm_dev_notifier = {
1215         xfrm_dev_event,
1216         NULL,
1217         0
1218 };
1219
1220 void __init xfrm_policy_init(void)
1221 {
1222         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
1223                                            sizeof(struct xfrm_dst),
1224                                            0, SLAB_HWCACHE_ALIGN,
1225                                            NULL, NULL);
1226         if (!xfrm_dst_cache)
1227                 panic("XFRM: failed to allocate xfrm_dst_cache\n");
1228
1229         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task, NULL);
1230         register_netdevice_notifier(&xfrm_dev_notifier);
1231 }
1232
1233 void __init xfrm_init(void)
1234 {
1235         xfrm_state_init();
1236         xfrm_policy_init();
1237         xfrm_input_init();
1238 }
1239