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