git.oblomov.eu Git - linux-2.6/blob - 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/slab.h>
  17 #include <linux/kmod.h>
  18 #include <linux/list.h>
  19 #include <linux/spinlock.h>
  20 #include <linux/workqueue.h>
  21 #include <linux/notifier.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/netfilter.h>
  24 #include <linux/module.h>
  25 #include <linux/cache.h>
  26 #include <net/xfrm.h>
  27 #include <net/ip.h>
  28
  29 #include "xfrm_hash.h"
  30
  31 int sysctl_xfrm_larval_drop __read_mostly;
  32
  33 DEFINE_MUTEX(xfrm_cfg_mutex);
  34 EXPORT_SYMBOL(xfrm_cfg_mutex);
  35
  36 static DEFINE_RWLOCK(xfrm_policy_lock);
  37
  38 unsigned int xfrm_policy_count[XFRM_POLICY_MAX*2];
  39 EXPORT_SYMBOL(xfrm_policy_count);
  40
  41 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
  42 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
  43
  44 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  45
  46 static struct work_struct xfrm_policy_gc_work;
  47 static HLIST_HEAD(xfrm_policy_gc_list);
  48 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
  49
  50 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
  51 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
  52
  53 static inline int
  54 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  55 {
  56         return  addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
  57                 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
  58                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  59                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  60                 (fl->proto == sel->proto || !sel->proto) &&
  61                 (fl->oif == sel->ifindex || !sel->ifindex);
  62 }
  63
  64 static inline int
  65 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  66 {
  67         return  addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
  68                 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
  69                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  70                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  71                 (fl->proto == sel->proto || !sel->proto) &&
  72                 (fl->oif == sel->ifindex || !sel->ifindex);
  73 }
  74
  75 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
  76                     unsigned short family)
  77 {
  78         switch (family) {
  79         case AF_INET:
  80                 return __xfrm4_selector_match(sel, fl);
  81         case AF_INET6:
  82                 return __xfrm6_selector_match(sel, fl);
  83         }
  84         return 0;
  85 }
  86
  87 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl,
  88                     unsigned short family)
  89 {
  90         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
  91         int err = 0;
  92
  93         if (unlikely(afinfo == NULL))
  94                 return -EAFNOSUPPORT;
  95
  96         if (likely(afinfo->dst_lookup != NULL))
  97                 err = afinfo->dst_lookup(dst, fl);
  98         else
  99                 err = -EINVAL;
 100         xfrm_policy_put_afinfo(afinfo);
 101         return err;
 102 }
 103 EXPORT_SYMBOL(xfrm_dst_lookup);
 104
 105 static inline unsigned long make_jiffies(long secs)
 106 {
 107         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 108                 return MAX_SCHEDULE_TIMEOUT-1;
 109         else
 110                 return secs*HZ;
 111 }
 112
 113 static void xfrm_policy_timer(unsigned long data)
 114 {
 115         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 116         unsigned long now = get_seconds();
 117         long next = LONG_MAX;
 118         int warn = 0;
 119         int dir;
 120
 121         read_lock(&xp->lock);
 122
 123         if (xp->dead)
 124                 goto out;
 125
 126         dir = xfrm_policy_id2dir(xp->index);
 127
 128         if (xp->lft.hard_add_expires_seconds) {
 129                 long tmo = xp->lft.hard_add_expires_seconds +
 130                         xp->curlft.add_time - now;
 131                 if (tmo <= 0)
 132                         goto expired;
 133                 if (tmo < next)
 134                         next = tmo;
 135         }
 136         if (xp->lft.hard_use_expires_seconds) {
 137                 long tmo = xp->lft.hard_use_expires_seconds +
 138                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 139                 if (tmo <= 0)
 140                         goto expired;
 141                 if (tmo < next)
 142                         next = tmo;
 143         }
 144         if (xp->lft.soft_add_expires_seconds) {
 145                 long tmo = xp->lft.soft_add_expires_seconds +
 146                         xp->curlft.add_time - now;
 147                 if (tmo <= 0) {
 148                         warn = 1;
 149                         tmo = XFRM_KM_TIMEOUT;
 150                 }
 151                 if (tmo < next)
 152                         next = tmo;
 153         }
 154         if (xp->lft.soft_use_expires_seconds) {
 155                 long tmo = xp->lft.soft_use_expires_seconds +
 156                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 157                 if (tmo <= 0) {
 158                         warn = 1;
 159                         tmo = XFRM_KM_TIMEOUT;
 160                 }
 161                 if (tmo < next)
 162                         next = tmo;
 163         }
 164
 165         if (warn)
 166                 km_policy_expired(xp, dir, 0, 0);
 167         if (next != LONG_MAX &&
 168             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 169                 xfrm_pol_hold(xp);
 170
 171 out:
 172         read_unlock(&xp->lock);
 173         xfrm_pol_put(xp);
 174         return;
 175
 176 expired:
 177         read_unlock(&xp->lock);
 178         if (!xfrm_policy_delete(xp, dir))
 179                 km_policy_expired(xp, dir, 1, 0);
 180         xfrm_pol_put(xp);
 181 }
 182
 183
 184 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 185  * SPD calls.
 186  */
 187
 188 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
 189 {
 190         struct xfrm_policy *policy;
 191
 192         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 193
 194         if (policy) {
 195                 INIT_HLIST_NODE(&policy->bydst);
 196                 INIT_HLIST_NODE(&policy->byidx);
 197                 rwlock_init(&policy->lock);
 198                 atomic_set(&policy->refcnt, 1);
 199                 init_timer(&policy->timer);
 200                 policy->timer.data = (unsigned long)policy;
 201                 policy->timer.function = xfrm_policy_timer;
 202         }
 203         return policy;
 204 }
 205 EXPORT_SYMBOL(xfrm_policy_alloc);
 206
 207 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 208
 209 void __xfrm_policy_destroy(struct xfrm_policy *policy)
 210 {
 211         BUG_ON(!policy->dead);
 212
 213         BUG_ON(policy->bundles);
 214
 215         if (del_timer(&policy->timer))
 216                 BUG();
 217
 218         security_xfrm_policy_free(policy);
 219         kfree(policy);
 220 }
 221 EXPORT_SYMBOL(__xfrm_policy_destroy);
 222
 223 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
 224 {
 225         struct dst_entry *dst;
 226
 227         while ((dst = policy->bundles) != NULL) {
 228                 policy->bundles = dst->next;
 229                 dst_free(dst);
 230         }
 231
 232         if (del_timer(&policy->timer))
 233                 atomic_dec(&policy->refcnt);
 234
 235         if (atomic_read(&policy->refcnt) > 1)
 236                 flow_cache_flush();
 237
 238         xfrm_pol_put(policy);
 239 }
 240
 241 static void xfrm_policy_gc_task(struct work_struct *work)
 242 {
 243         struct xfrm_policy *policy;
 244         struct hlist_node *entry, *tmp;
 245         struct hlist_head gc_list;
 246
 247         spin_lock_bh(&xfrm_policy_gc_lock);
 248         gc_list.first = xfrm_policy_gc_list.first;
 249         INIT_HLIST_HEAD(&xfrm_policy_gc_list);
 250         spin_unlock_bh(&xfrm_policy_gc_lock);
 251
 252         hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
 253                 xfrm_policy_gc_kill(policy);
 254 }
 255
 256 /* Rule must be locked. Release descentant resources, announce
 257  * entry dead. The rule must be unlinked from lists to the moment.
 258  */
 259
 260 static void xfrm_policy_kill(struct xfrm_policy *policy)
 261 {
 262         int dead;
 263
 264         write_lock_bh(&policy->lock);
 265         dead = policy->dead;
 266         policy->dead = 1;
 267         write_unlock_bh(&policy->lock);
 268
 269         if (unlikely(dead)) {
 270                 WARN_ON(1);
 271                 return;
 272         }
 273
 274         spin_lock(&xfrm_policy_gc_lock);
 275         hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
 276         spin_unlock(&xfrm_policy_gc_lock);
 277
 278         schedule_work(&xfrm_policy_gc_work);
 279 }
 280
 281 struct xfrm_policy_hash {
 282         struct hlist_head       *table;
 283         unsigned int            hmask;
 284 };
 285
 286 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
 287 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
 288 static struct hlist_head *xfrm_policy_byidx __read_mostly;
 289 static unsigned int xfrm_idx_hmask __read_mostly;
 290 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 291
 292 static inline unsigned int idx_hash(u32 index)
 293 {
 294         return __idx_hash(index, xfrm_idx_hmask);
 295 }
 296
 297 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
 298 {
 299         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 300         unsigned int hash = __sel_hash(sel, family, hmask);
 301
 302         return (hash == hmask + 1 ?
 303                 &xfrm_policy_inexact[dir] :
 304                 xfrm_policy_bydst[dir].table + hash);
 305 }
 306
 307 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
 308 {
 309         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 310         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 311
 312         return xfrm_policy_bydst[dir].table + hash;
 313 }
 314
 315 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 316                                    struct hlist_head *ndsttable,
 317                                    unsigned int nhashmask)
 318 {
 319         struct hlist_node *entry, *tmp;
 320         struct xfrm_policy *pol;
 321
 322         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
 323                 unsigned int h;
 324
 325                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 326                                 pol->family, nhashmask);
 327                 hlist_add_head(&pol->bydst, ndsttable+h);
 328         }
 329 }
 330
 331 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 332                                    struct hlist_head *nidxtable,
 333                                    unsigned int nhashmask)
 334 {
 335         struct hlist_node *entry, *tmp;
 336         struct xfrm_policy *pol;
 337
 338         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
 339                 unsigned int h;
 340
 341                 h = __idx_hash(pol->index, nhashmask);
 342                 hlist_add_head(&pol->byidx, nidxtable+h);
 343         }
 344 }
 345
 346 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 347 {
 348         return ((old_hmask + 1) << 1) - 1;
 349 }
 350
 351 static void xfrm_bydst_resize(int dir)
 352 {
 353         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 354         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 355         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 356         struct hlist_head *odst = xfrm_policy_bydst[dir].table;
 357         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 358         int i;
 359
 360         if (!ndst)
 361                 return;
 362
 363         write_lock_bh(&xfrm_policy_lock);
 364
 365         for (i = hmask; i >= 0; i--)
 366                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 367
 368         xfrm_policy_bydst[dir].table = ndst;
 369         xfrm_policy_bydst[dir].hmask = nhashmask;
 370
 371         write_unlock_bh(&xfrm_policy_lock);
 372
 373         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 374 }
 375
 376 static void xfrm_byidx_resize(int total)
 377 {
 378         unsigned int hmask = xfrm_idx_hmask;
 379         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 380         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 381         struct hlist_head *oidx = xfrm_policy_byidx;
 382         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 383         int i;
 384
 385         if (!nidx)
 386                 return;
 387
 388         write_lock_bh(&xfrm_policy_lock);
 389
 390         for (i = hmask; i >= 0; i--)
 391                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 392
 393         xfrm_policy_byidx = nidx;
 394         xfrm_idx_hmask = nhashmask;
 395
 396         write_unlock_bh(&xfrm_policy_lock);
 397
 398         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 399 }
 400
 401 static inline int xfrm_bydst_should_resize(int dir, int *total)
 402 {
 403         unsigned int cnt = xfrm_policy_count[dir];
 404         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 405
 406         if (total)
 407                 *total += cnt;
 408
 409         if ((hmask + 1) < xfrm_policy_hashmax &&
 410             cnt > hmask)
 411                 return 1;
 412
 413         return 0;
 414 }
 415
 416 static inline int xfrm_byidx_should_resize(int total)
 417 {
 418         unsigned int hmask = xfrm_idx_hmask;
 419
 420         if ((hmask + 1) < xfrm_policy_hashmax &&
 421             total > hmask)
 422                 return 1;
 423
 424         return 0;
 425 }
 426
 427 void xfrm_spd_getinfo(struct xfrmk_spdinfo *si)
 428 {
 429         read_lock_bh(&xfrm_policy_lock);
 430         si->incnt = xfrm_policy_count[XFRM_POLICY_IN];
 431         si->outcnt = xfrm_policy_count[XFRM_POLICY_OUT];
 432         si->fwdcnt = xfrm_policy_count[XFRM_POLICY_FWD];
 433         si->inscnt = xfrm_policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 434         si->outscnt = xfrm_policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 435         si->fwdscnt = xfrm_policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 436         si->spdhcnt = xfrm_idx_hmask;
 437         si->spdhmcnt = xfrm_policy_hashmax;
 438         read_unlock_bh(&xfrm_policy_lock);
 439 }
 440 EXPORT_SYMBOL(xfrm_spd_getinfo);
 441
 442 static DEFINE_MUTEX(hash_resize_mutex);
 443 static void xfrm_hash_resize(struct work_struct *__unused)
 444 {
 445         int dir, total;
 446
 447         mutex_lock(&hash_resize_mutex);
 448
 449         total = 0;
 450         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 451                 if (xfrm_bydst_should_resize(dir, &total))
 452                         xfrm_bydst_resize(dir);
 453         }
 454         if (xfrm_byidx_should_resize(total))
 455                 xfrm_byidx_resize(total);
 456
 457         mutex_unlock(&hash_resize_mutex);
 458 }
 459
 460 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize);
 461
 462 /* Generate new index... KAME seems to generate them ordered by cost
 463  * of an absolute inpredictability of ordering of rules. This will not pass. */
 464 static u32 xfrm_gen_index(u8 type, int dir)
 465 {
 466         static u32 idx_generator;
 467
 468         for (;;) {
 469                 struct hlist_node *entry;
 470                 struct hlist_head *list;
 471                 struct xfrm_policy *p;
 472                 u32 idx;
 473                 int found;
 474
 475                 idx = (idx_generator | dir);
 476                 idx_generator += 8;
 477                 if (idx == 0)
 478                         idx = 8;
 479                 list = xfrm_policy_byidx + idx_hash(idx);
 480                 found = 0;
 481                 hlist_for_each_entry(p, entry, list, byidx) {
 482                         if (p->index == idx) {
 483                                 found = 1;
 484                                 break;
 485                         }
 486                 }
 487                 if (!found)
 488                         return idx;
 489         }
 490 }
 491
 492 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 493 {
 494         u32 *p1 = (u32 *) s1;
 495         u32 *p2 = (u32 *) s2;
 496         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 497         int i;
 498
 499         for (i = 0; i < len; i++) {
 500                 if (p1[i] != p2[i])
 501                         return 1;
 502         }
 503
 504         return 0;
 505 }
 506
 507 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 508 {
 509         struct xfrm_policy *pol;
 510         struct xfrm_policy *delpol;
 511         struct hlist_head *chain;
 512         struct hlist_node *entry, *newpos;
 513         struct dst_entry *gc_list;
 514
 515         write_lock_bh(&xfrm_policy_lock);
 516         chain = policy_hash_bysel(&policy->selector, policy->family, dir);
 517         delpol = NULL;
 518         newpos = NULL;
 519         hlist_for_each_entry(pol, entry, chain, bydst) {
 520                 if (pol->type == policy->type &&
 521                     !selector_cmp(&pol->selector, &policy->selector) &&
 522                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 523                     !WARN_ON(delpol)) {
 524                         if (excl) {
 525                                 write_unlock_bh(&xfrm_policy_lock);
 526                                 return -EEXIST;
 527                         }
 528                         delpol = pol;
 529                         if (policy->priority > pol->priority)
 530                                 continue;
 531                 } else if (policy->priority >= pol->priority) {
 532                         newpos = &pol->bydst;
 533                         continue;
 534                 }
 535                 if (delpol)
 536                         break;
 537         }
 538         if (newpos)
 539                 hlist_add_after(newpos, &policy->bydst);
 540         else
 541                 hlist_add_head(&policy->bydst, chain);
 542         xfrm_pol_hold(policy);
 543         xfrm_policy_count[dir]++;
 544         atomic_inc(&flow_cache_genid);
 545         if (delpol) {
 546                 hlist_del(&delpol->bydst);
 547                 hlist_del(&delpol->byidx);
 548                 xfrm_policy_count[dir]--;
 549         }
 550         policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
 551         hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
 552         policy->curlft.add_time = get_seconds();
 553         policy->curlft.use_time = 0;
 554         if (!mod_timer(&policy->timer, jiffies + HZ))
 555                 xfrm_pol_hold(policy);
 556         write_unlock_bh(&xfrm_policy_lock);
 557
 558         if (delpol)
 559                 xfrm_policy_kill(delpol);
 560         else if (xfrm_bydst_should_resize(dir, NULL))
 561                 schedule_work(&xfrm_hash_work);
 562
 563         read_lock_bh(&xfrm_policy_lock);
 564         gc_list = NULL;
 565         entry = &policy->bydst;
 566         hlist_for_each_entry_continue(policy, entry, bydst) {
 567                 struct dst_entry *dst;
 568
 569                 write_lock(&policy->lock);
 570                 dst = policy->bundles;
 571                 if (dst) {
 572                         struct dst_entry *tail = dst;
 573                         while (tail->next)
 574                                 tail = tail->next;
 575                         tail->next = gc_list;
 576                         gc_list = dst;
 577
 578                         policy->bundles = NULL;
 579                 }
 580                 write_unlock(&policy->lock);
 581         }
 582         read_unlock_bh(&xfrm_policy_lock);
 583
 584         while (gc_list) {
 585                 struct dst_entry *dst = gc_list;
 586
 587                 gc_list = dst->next;
 588                 dst_free(dst);
 589         }
 590
 591         return 0;
 592 }
 593 EXPORT_SYMBOL(xfrm_policy_insert);
 594
 595 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
 596                                           struct xfrm_selector *sel,
 597                                           struct xfrm_sec_ctx *ctx, int delete,
 598                                           int *err)
 599 {
 600         struct xfrm_policy *pol, *ret;
 601         struct hlist_head *chain;
 602         struct hlist_node *entry;
 603
 604         *err = 0;
 605         write_lock_bh(&xfrm_policy_lock);
 606         chain = policy_hash_bysel(sel, sel->family, dir);
 607         ret = NULL;
 608         hlist_for_each_entry(pol, entry, chain, bydst) {
 609                 if (pol->type == type &&
 610                     !selector_cmp(sel, &pol->selector) &&
 611                     xfrm_sec_ctx_match(ctx, pol->security)) {
 612                         xfrm_pol_hold(pol);
 613                         if (delete) {
 614                                 *err = security_xfrm_policy_delete(pol);
 615                                 if (*err) {
 616                                         write_unlock_bh(&xfrm_policy_lock);
 617                                         return pol;
 618                                 }
 619                                 hlist_del(&pol->bydst);
 620                                 hlist_del(&pol->byidx);
 621                                 xfrm_policy_count[dir]--;
 622                         }
 623                         ret = pol;
 624                         break;
 625                 }
 626         }
 627         write_unlock_bh(&xfrm_policy_lock);
 628
 629         if (ret && delete) {
 630                 atomic_inc(&flow_cache_genid);
 631                 xfrm_policy_kill(ret);
 632         }
 633         return ret;
 634 }
 635 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 636
 637 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete,
 638                                      int *err)
 639 {
 640         struct xfrm_policy *pol, *ret;
 641         struct hlist_head *chain;
 642         struct hlist_node *entry;
 643
 644         *err = -ENOENT;
 645         if (xfrm_policy_id2dir(id) != dir)
 646                 return NULL;
 647
 648         *err = 0;
 649         write_lock_bh(&xfrm_policy_lock);
 650         chain = xfrm_policy_byidx + idx_hash(id);
 651         ret = NULL;
 652         hlist_for_each_entry(pol, entry, chain, byidx) {
 653                 if (pol->type == type && pol->index == id) {
 654                         xfrm_pol_hold(pol);
 655                         if (delete) {
 656                                 *err = security_xfrm_policy_delete(pol);
 657                                 if (*err) {
 658                                         write_unlock_bh(&xfrm_policy_lock);
 659                                         return pol;
 660                                 }
 661                                 hlist_del(&pol->bydst);
 662                                 hlist_del(&pol->byidx);
 663                                 xfrm_policy_count[dir]--;
 664                         }
 665                         ret = pol;
 666                         break;
 667                 }
 668         }
 669         write_unlock_bh(&xfrm_policy_lock);
 670
 671         if (ret && delete) {
 672                 atomic_inc(&flow_cache_genid);
 673                 xfrm_policy_kill(ret);
 674         }
 675         return ret;
 676 }
 677 EXPORT_SYMBOL(xfrm_policy_byid);
 678
 679 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 680 static inline int
 681 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
 682 {
 683         int dir, err = 0;
 684
 685         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 686                 struct xfrm_policy *pol;
 687                 struct hlist_node *entry;
 688                 int i;
 689
 690                 hlist_for_each_entry(pol, entry,
 691                                      &xfrm_policy_inexact[dir], bydst) {
 692                         if (pol->type != type)
 693                                 continue;
 694                         err = security_xfrm_policy_delete(pol);
 695                         if (err) {
 696                                 xfrm_audit_policy_delete(pol, 0,
 697                                                          audit_info->loginuid,
 698                                                          audit_info->secid);
 699                                 return err;
 700                         }
 701                 }
 702                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 703                         hlist_for_each_entry(pol, entry,
 704                                              xfrm_policy_bydst[dir].table + i,
 705                                              bydst) {
 706                                 if (pol->type != type)
 707                                         continue;
 708                                 err = security_xfrm_policy_delete(pol);
 709                                 if (err) {
 710                                         xfrm_audit_policy_delete(pol, 0,
 711                                                         audit_info->loginuid,
 712                                                         audit_info->secid);
 713                                         return err;
 714                                 }
 715                         }
 716                 }
 717         }
 718         return err;
 719 }
 720 #else
 721 static inline int
 722 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
 723 {
 724         return 0;
 725 }
 726 #endif
 727
 728 int xfrm_policy_flush(u8 type, struct xfrm_audit *audit_info)
 729 {
 730         int dir, err = 0;
 731
 732         write_lock_bh(&xfrm_policy_lock);
 733
 734         err = xfrm_policy_flush_secctx_check(type, audit_info);
 735         if (err)
 736                 goto out;
 737
 738         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 739                 struct xfrm_policy *pol;
 740                 struct hlist_node *entry;
 741                 int i, killed;
 742
 743                 killed = 0;
 744         again1:
 745                 hlist_for_each_entry(pol, entry,
 746                                      &xfrm_policy_inexact[dir], bydst) {
 747                         if (pol->type != type)
 748                                 continue;
 749                         hlist_del(&pol->bydst);
 750                         hlist_del(&pol->byidx);
 751                         write_unlock_bh(&xfrm_policy_lock);
 752
 753                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
 754                                                  audit_info->secid);
 755
 756                         xfrm_policy_kill(pol);
 757                         killed++;
 758
 759                         write_lock_bh(&xfrm_policy_lock);
 760                         goto again1;
 761                 }
 762
 763                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 764         again2:
 765                         hlist_for_each_entry(pol, entry,
 766                                              xfrm_policy_bydst[dir].table + i,
 767                                              bydst) {
 768                                 if (pol->type != type)
 769                                         continue;
 770                                 hlist_del(&pol->bydst);
 771                                 hlist_del(&pol->byidx);
 772                                 write_unlock_bh(&xfrm_policy_lock);
 773
 774                                 xfrm_audit_policy_delete(pol, 1,
 775                                                          audit_info->loginuid,
 776                                                          audit_info->secid);
 777                                 xfrm_policy_kill(pol);
 778                                 killed++;
 779
 780                                 write_lock_bh(&xfrm_policy_lock);
 781                                 goto again2;
 782                         }
 783                 }
 784
 785                 xfrm_policy_count[dir] -= killed;
 786         }
 787         atomic_inc(&flow_cache_genid);
 788 out:
 789         write_unlock_bh(&xfrm_policy_lock);
 790         return err;
 791 }
 792 EXPORT_SYMBOL(xfrm_policy_flush);
 793
 794 int xfrm_policy_walk(u8 type, int (*func)(struct xfrm_policy *, int, int, void*),
 795                      void *data)
 796 {
 797         struct xfrm_policy *pol, *last = NULL;
 798         struct hlist_node *entry;
 799         int dir, last_dir = 0, count, error;
 800
 801         read_lock_bh(&xfrm_policy_lock);
 802         count = 0;
 803
 804         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
 805                 struct hlist_head *table = xfrm_policy_bydst[dir].table;
 806                 int i;
 807
 808                 hlist_for_each_entry(pol, entry,
 809                                      &xfrm_policy_inexact[dir], bydst) {
 810                         if (pol->type != type)
 811                                 continue;
 812                         if (last) {
 813                                 error = func(last, last_dir % XFRM_POLICY_MAX,
 814                                              count, data);
 815                                 if (error)
 816                                         goto out;
 817                         }
 818                         last = pol;
 819                         last_dir = dir;
 820                         count++;
 821                 }
 822                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 823                         hlist_for_each_entry(pol, entry, table + i, bydst) {
 824                                 if (pol->type != type)
 825                                         continue;
 826                                 if (last) {
 827                                         error = func(last, last_dir % XFRM_POLICY_MAX,
 828                                                      count, data);
 829                                         if (error)
 830                                                 goto out;
 831                                 }
 832                                 last = pol;
 833                                 last_dir = dir;
 834                                 count++;
 835                         }
 836                 }
 837         }
 838         if (count == 0) {
 839                 error = -ENOENT;
 840                 goto out;
 841         }
 842         error = func(last, last_dir % XFRM_POLICY_MAX, 0, data);
 843 out:
 844         read_unlock_bh(&xfrm_policy_lock);
 845         return error;
 846 }
 847 EXPORT_SYMBOL(xfrm_policy_walk);
 848
 849 /*
 850  * Find policy to apply to this flow.
 851  *
 852  * Returns 0 if policy found, else an -errno.
 853  */
 854 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
 855                              u8 type, u16 family, int dir)
 856 {
 857         struct xfrm_selector *sel = &pol->selector;
 858         int match, ret = -ESRCH;
 859
 860         if (pol->family != family ||
 861             pol->type != type)
 862                 return ret;
 863
 864         match = xfrm_selector_match(sel, fl, family);
 865         if (match)
 866                 ret = security_xfrm_policy_lookup(pol, fl->secid, dir);
 867
 868         return ret;
 869 }
 870
 871 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
 872                                                      u16 family, u8 dir)
 873 {
 874         int err;
 875         struct xfrm_policy *pol, *ret;
 876         xfrm_address_t *daddr, *saddr;
 877         struct hlist_node *entry;
 878         struct hlist_head *chain;
 879         u32 priority = ~0U;
 880
 881         daddr = xfrm_flowi_daddr(fl, family);
 882         saddr = xfrm_flowi_saddr(fl, family);
 883         if (unlikely(!daddr || !saddr))
 884                 return NULL;
 885
 886         read_lock_bh(&xfrm_policy_lock);
 887         chain = policy_hash_direct(daddr, saddr, family, dir);
 888         ret = NULL;
 889         hlist_for_each_entry(pol, entry, chain, bydst) {
 890                 err = xfrm_policy_match(pol, fl, type, family, dir);
 891                 if (err) {
 892                         if (err == -ESRCH)
 893                                 continue;
 894                         else {
 895                                 ret = ERR_PTR(err);
 896                                 goto fail;
 897                         }
 898                 } else {
 899                         ret = pol;
 900                         priority = ret->priority;
 901                         break;
 902                 }
 903         }
 904         chain = &xfrm_policy_inexact[dir];
 905         hlist_for_each_entry(pol, entry, chain, bydst) {
 906                 err = xfrm_policy_match(pol, fl, type, family, dir);
 907                 if (err) {
 908                         if (err == -ESRCH)
 909                                 continue;
 910                         else {
 911                                 ret = ERR_PTR(err);
 912                                 goto fail;
 913                         }
 914                 } else if (pol->priority < priority) {
 915                         ret = pol;
 916                         break;
 917                 }
 918         }
 919         if (ret)
 920                 xfrm_pol_hold(ret);
 921 fail:
 922         read_unlock_bh(&xfrm_policy_lock);
 923
 924         return ret;
 925 }
 926
 927 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
 928                                void **objp, atomic_t **obj_refp)
 929 {
 930         struct xfrm_policy *pol;
 931         int err = 0;
 932
 933 #ifdef CONFIG_XFRM_SUB_POLICY
 934         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
 935         if (IS_ERR(pol)) {
 936                 err = PTR_ERR(pol);
 937                 pol = NULL;
 938         }
 939         if (pol || err)
 940                 goto end;
 941 #endif
 942         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
 943         if (IS_ERR(pol)) {
 944                 err = PTR_ERR(pol);
 945                 pol = NULL;
 946         }
 947 #ifdef CONFIG_XFRM_SUB_POLICY
 948 end:
 949 #endif
 950         if ((*objp = (void *) pol) != NULL)
 951                 *obj_refp = &pol->refcnt;
 952         return err;
 953 }
 954
 955 static inline int policy_to_flow_dir(int dir)
 956 {
 957         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
 958             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
 959             XFRM_POLICY_FWD == FLOW_DIR_FWD)
 960                 return dir;
 961         switch (dir) {
 962         default:
 963         case XFRM_POLICY_IN:
 964                 return FLOW_DIR_IN;
 965         case XFRM_POLICY_OUT:
 966                 return FLOW_DIR_OUT;
 967         case XFRM_POLICY_FWD:
 968                 return FLOW_DIR_FWD;
 969         }
 970 }
 971
 972 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
 973 {
 974         struct xfrm_policy *pol;
 975
 976         read_lock_bh(&xfrm_policy_lock);
 977         if ((pol = sk->sk_policy[dir]) != NULL) {
 978                 int match = xfrm_selector_match(&pol->selector, fl,
 979                                                 sk->sk_family);
 980                 int err = 0;
 981
 982                 if (match) {
 983                         err = security_xfrm_policy_lookup(pol, fl->secid,
 984                                         policy_to_flow_dir(dir));
 985                         if (!err)
 986                                 xfrm_pol_hold(pol);
 987                         else if (err == -ESRCH)
 988                                 pol = NULL;
 989                         else
 990                                 pol = ERR_PTR(err);
 991                 } else
 992                         pol = NULL;
 993         }
 994         read_unlock_bh(&xfrm_policy_lock);
 995         return pol;
 996 }
 997
 998 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
 999 {
1000         struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1001                                                      pol->family, dir);
1002
1003         hlist_add_head(&pol->bydst, chain);
1004         hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1005         xfrm_policy_count[dir]++;
1006         xfrm_pol_hold(pol);
1007
1008         if (xfrm_bydst_should_resize(dir, NULL))
1009                 schedule_work(&xfrm_hash_work);
1010 }
1011
1012 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1013                                                 int dir)
1014 {
1015         if (hlist_unhashed(&pol->bydst))
1016                 return NULL;
1017
1018         hlist_del(&pol->bydst);
1019         hlist_del(&pol->byidx);
1020         xfrm_policy_count[dir]--;
1021
1022         return pol;
1023 }
1024
1025 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1026 {
1027         write_lock_bh(&xfrm_policy_lock);
1028         pol = __xfrm_policy_unlink(pol, dir);
1029         write_unlock_bh(&xfrm_policy_lock);
1030         if (pol) {
1031                 if (dir < XFRM_POLICY_MAX)
1032                         atomic_inc(&flow_cache_genid);
1033                 xfrm_policy_kill(pol);
1034                 return 0;
1035         }
1036         return -ENOENT;
1037 }
1038 EXPORT_SYMBOL(xfrm_policy_delete);
1039
1040 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1041 {
1042         struct xfrm_policy *old_pol;
1043
1044 #ifdef CONFIG_XFRM_SUB_POLICY
1045         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1046                 return -EINVAL;
1047 #endif
1048
1049         write_lock_bh(&xfrm_policy_lock);
1050         old_pol = sk->sk_policy[dir];
1051         sk->sk_policy[dir] = pol;
1052         if (pol) {
1053                 pol->curlft.add_time = get_seconds();
1054                 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1055                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1056         }
1057         if (old_pol)
1058                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1059         write_unlock_bh(&xfrm_policy_lock);
1060
1061         if (old_pol) {
1062                 xfrm_policy_kill(old_pol);
1063         }
1064         return 0;
1065 }
1066
1067 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1068 {
1069         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1070
1071         if (newp) {
1072                 newp->selector = old->selector;
1073                 if (security_xfrm_policy_clone(old, newp)) {
1074                         kfree(newp);
1075                         return NULL;  /* ENOMEM */
1076                 }
1077                 newp->lft = old->lft;
1078                 newp->curlft = old->curlft;
1079                 newp->action = old->action;
1080                 newp->flags = old->flags;
1081                 newp->xfrm_nr = old->xfrm_nr;
1082                 newp->index = old->index;
1083                 newp->type = old->type;
1084                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1085                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1086                 write_lock_bh(&xfrm_policy_lock);
1087                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1088                 write_unlock_bh(&xfrm_policy_lock);
1089                 xfrm_pol_put(newp);
1090         }
1091         return newp;
1092 }
1093
1094 int __xfrm_sk_clone_policy(struct sock *sk)
1095 {
1096         struct xfrm_policy *p0 = sk->sk_policy[0],
1097                            *p1 = sk->sk_policy[1];
1098
1099         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1100         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1101                 return -ENOMEM;
1102         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1103                 return -ENOMEM;
1104         return 0;
1105 }
1106
1107 static int
1108 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1109                unsigned short family)
1110 {
1111         int err;
1112         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1113
1114         if (unlikely(afinfo == NULL))
1115                 return -EINVAL;
1116         err = afinfo->get_saddr(local, remote);
1117         xfrm_policy_put_afinfo(afinfo);
1118         return err;
1119 }
1120
1121 /* Resolve list of templates for the flow, given policy. */
1122
1123 static int
1124 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1125                       struct xfrm_state **xfrm,
1126                       unsigned short family)
1127 {
1128         int nx;
1129         int i, error;
1130         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1131         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1132         xfrm_address_t tmp;
1133
1134         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1135                 struct xfrm_state *x;
1136                 xfrm_address_t *remote = daddr;
1137                 xfrm_address_t *local  = saddr;
1138                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1139
1140                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1141                     tmpl->mode == XFRM_MODE_BEET) {
1142                         remote = &tmpl->id.daddr;
1143                         local = &tmpl->saddr;
1144                         family = tmpl->encap_family;
1145                         if (xfrm_addr_any(local, family)) {
1146                                 error = xfrm_get_saddr(&tmp, remote, family);
1147                                 if (error)
1148                                         goto fail;
1149                                 local = &tmp;
1150                         }
1151                 }
1152
1153                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1154
1155                 if (x && x->km.state == XFRM_STATE_VALID) {
1156                         xfrm[nx++] = x;
1157                         daddr = remote;
1158                         saddr = local;
1159                         continue;
1160                 }
1161                 if (x) {
1162                         error = (x->km.state == XFRM_STATE_ERROR ?
1163                                  -EINVAL : -EAGAIN);
1164                         xfrm_state_put(x);
1165                 }
1166
1167                 if (!tmpl->optional)
1168                         goto fail;
1169         }
1170         return nx;
1171
1172 fail:
1173         for (nx--; nx>=0; nx--)
1174                 xfrm_state_put(xfrm[nx]);
1175         return error;
1176 }
1177
1178 static int
1179 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1180                   struct xfrm_state **xfrm,
1181                   unsigned short family)
1182 {
1183         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1184         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1185         int cnx = 0;
1186         int error;
1187         int ret;
1188         int i;
1189
1190         for (i = 0; i < npols; i++) {
1191                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1192                         error = -ENOBUFS;
1193                         goto fail;
1194                 }
1195
1196                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1197                 if (ret < 0) {
1198                         error = ret;
1199                         goto fail;
1200                 } else
1201                         cnx += ret;
1202         }
1203
1204         /* found states are sorted for outbound processing */
1205         if (npols > 1)
1206                 xfrm_state_sort(xfrm, tpp, cnx, family);
1207
1208         return cnx;
1209
1210  fail:
1211         for (cnx--; cnx>=0; cnx--)
1212                 xfrm_state_put(tpp[cnx]);
1213         return error;
1214
1215 }
1216
1217 /* Check that the bundle accepts the flow and its components are
1218  * still valid.
1219  */
1220
1221 static struct dst_entry *
1222 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1223 {
1224         struct dst_entry *x;
1225         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1226         if (unlikely(afinfo == NULL))
1227                 return ERR_PTR(-EINVAL);
1228         x = afinfo->find_bundle(fl, policy);
1229         xfrm_policy_put_afinfo(afinfo);
1230         return x;
1231 }
1232
1233 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1234  * all the metrics... Shortly, bundle a bundle.
1235  */
1236
1237 static int
1238 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
1239                    struct flowi *fl, struct dst_entry **dst_p,
1240                    unsigned short family)
1241 {
1242         int err;
1243         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1244         if (unlikely(afinfo == NULL))
1245                 return -EINVAL;
1246         err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
1247         xfrm_policy_put_afinfo(afinfo);
1248         return err;
1249 }
1250
1251 static int inline
1252 xfrm_dst_alloc_copy(void **target, void *src, int size)
1253 {
1254         if (!*target) {
1255                 *target = kmalloc(size, GFP_ATOMIC);
1256                 if (!*target)
1257                         return -ENOMEM;
1258         }
1259         memcpy(*target, src, size);
1260         return 0;
1261 }
1262
1263 static int inline
1264 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1265 {
1266 #ifdef CONFIG_XFRM_SUB_POLICY
1267         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1268         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1269                                    sel, sizeof(*sel));
1270 #else
1271         return 0;
1272 #endif
1273 }
1274
1275 static int inline
1276 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1277 {
1278 #ifdef CONFIG_XFRM_SUB_POLICY
1279         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1280         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1281 #else
1282         return 0;
1283 #endif
1284 }
1285
1286 static int stale_bundle(struct dst_entry *dst);
1287
1288 /* Main function: finds/creates a bundle for given flow.
1289  *
1290  * At the moment we eat a raw IP route. Mostly to speed up lookups
1291  * on interfaces with disabled IPsec.
1292  */
1293 int __xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1294                   struct sock *sk, int flags)
1295 {
1296         struct xfrm_policy *policy;
1297         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1298         int npols;
1299         int pol_dead;
1300         int xfrm_nr;
1301         int pi;
1302         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1303         struct dst_entry *dst, *dst_orig = *dst_p;
1304         int nx = 0;
1305         int err;
1306         u32 genid;
1307         u16 family;
1308         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1309
1310 restart:
1311         genid = atomic_read(&flow_cache_genid);
1312         policy = NULL;
1313         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1314                 pols[pi] = NULL;
1315         npols = 0;
1316         pol_dead = 0;
1317         xfrm_nr = 0;
1318
1319         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1320                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1321                 if (IS_ERR(policy))
1322                         return PTR_ERR(policy);
1323         }
1324
1325         if (!policy) {
1326                 /* To accelerate a bit...  */
1327                 if ((dst_orig->flags & DST_NOXFRM) ||
1328                     !xfrm_policy_count[XFRM_POLICY_OUT])
1329                         return 0;
1330
1331                 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1332                                            dir, xfrm_policy_lookup);
1333                 if (IS_ERR(policy))
1334                         return PTR_ERR(policy);
1335         }
1336
1337         if (!policy)
1338                 return 0;
1339
1340         family = dst_orig->ops->family;
1341         policy->curlft.use_time = get_seconds();
1342         pols[0] = policy;
1343         npols ++;
1344         xfrm_nr += pols[0]->xfrm_nr;
1345
1346         switch (policy->action) {
1347         case XFRM_POLICY_BLOCK:
1348                 /* Prohibit the flow */
1349                 err = -EPERM;
1350                 goto error;
1351
1352         case XFRM_POLICY_ALLOW:
1353 #ifndef CONFIG_XFRM_SUB_POLICY
1354                 if (policy->xfrm_nr == 0) {
1355                         /* Flow passes not transformed. */
1356                         xfrm_pol_put(policy);
1357                         return 0;
1358                 }
1359 #endif
1360
1361                 /* Try to find matching bundle.
1362                  *
1363                  * LATER: help from flow cache. It is optional, this
1364                  * is required only for output policy.
1365                  */
1366                 dst = xfrm_find_bundle(fl, policy, family);
1367                 if (IS_ERR(dst)) {
1368                         err = PTR_ERR(dst);
1369                         goto error;
1370                 }
1371
1372                 if (dst)
1373                         break;
1374
1375 #ifdef CONFIG_XFRM_SUB_POLICY
1376                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1377                         pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1378                                                             fl, family,
1379                                                             XFRM_POLICY_OUT);
1380                         if (pols[1]) {
1381                                 if (IS_ERR(pols[1])) {
1382                                         err = PTR_ERR(pols[1]);
1383                                         goto error;
1384                                 }
1385                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1386                                         err = -EPERM;
1387                                         goto error;
1388                                 }
1389                                 npols ++;
1390                                 xfrm_nr += pols[1]->xfrm_nr;
1391                         }
1392                 }
1393
1394                 /*
1395                  * Because neither flowi nor bundle information knows about
1396                  * transformation template size. On more than one policy usage
1397                  * we can realize whether all of them is bypass or not after
1398                  * they are searched. See above not-transformed bypass
1399                  * is surrounded by non-sub policy configuration, too.
1400                  */
1401                 if (xfrm_nr == 0) {
1402                         /* Flow passes not transformed. */
1403                         xfrm_pols_put(pols, npols);
1404                         return 0;
1405                 }
1406
1407 #endif
1408                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1409
1410                 if (unlikely(nx<0)) {
1411                         err = nx;
1412                         if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
1413                                 /* EREMOTE tells the caller to generate
1414                                  * a one-shot blackhole route.
1415                                  */
1416                                 xfrm_pol_put(policy);
1417                                 return -EREMOTE;
1418                         }
1419                         if (err == -EAGAIN && flags) {
1420                                 DECLARE_WAITQUEUE(wait, current);
1421
1422                                 add_wait_queue(&km_waitq, &wait);
1423                                 set_current_state(TASK_INTERRUPTIBLE);
1424                                 schedule();
1425                                 set_current_state(TASK_RUNNING);
1426                                 remove_wait_queue(&km_waitq, &wait);
1427
1428                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1429
1430                                 if (nx == -EAGAIN && signal_pending(current)) {
1431                                         err = -ERESTART;
1432                                         goto error;
1433                                 }
1434                                 if (nx == -EAGAIN ||
1435                                     genid != atomic_read(&flow_cache_genid)) {
1436                                         xfrm_pols_put(pols, npols);
1437                                         goto restart;
1438                                 }
1439                                 err = nx;
1440                         }
1441                         if (err < 0)
1442                                 goto error;
1443                 }
1444                 if (nx == 0) {
1445                         /* Flow passes not transformed. */
1446                         xfrm_pols_put(pols, npols);
1447                         return 0;
1448                 }
1449
1450                 dst = dst_orig;
1451                 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
1452
1453                 if (unlikely(err)) {
1454                         int i;
1455                         for (i=0; i<nx; i++)
1456                                 xfrm_state_put(xfrm[i]);
1457                         goto error;
1458                 }
1459
1460                 for (pi = 0; pi < npols; pi++) {
1461                         read_lock_bh(&pols[pi]->lock);
1462                         pol_dead |= pols[pi]->dead;
1463                         read_unlock_bh(&pols[pi]->lock);
1464                 }
1465
1466                 write_lock_bh(&policy->lock);
1467                 if (unlikely(pol_dead || stale_bundle(dst))) {
1468                         /* Wow! While we worked on resolving, this
1469                          * policy has gone. Retry. It is not paranoia,
1470                          * we just cannot enlist new bundle to dead object.
1471                          * We can't enlist stable bundles either.
1472                          */
1473                         write_unlock_bh(&policy->lock);
1474                         if (dst)
1475                                 dst_free(dst);
1476
1477                         err = -EHOSTUNREACH;
1478                         goto error;
1479                 }
1480
1481                 if (npols > 1)
1482                         err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1483                 else
1484                         err = xfrm_dst_update_origin(dst, fl);
1485                 if (unlikely(err)) {
1486                         write_unlock_bh(&policy->lock);
1487                         if (dst)
1488                                 dst_free(dst);
1489                         goto error;
1490                 }
1491
1492                 dst->next = policy->bundles;
1493                 policy->bundles = dst;
1494                 dst_hold(dst);
1495                 write_unlock_bh(&policy->lock);
1496         }
1497         *dst_p = dst;
1498         dst_release(dst_orig);
1499         xfrm_pols_put(pols, npols);
1500         return 0;
1501
1502 error:
1503         dst_release(dst_orig);
1504         xfrm_pols_put(pols, npols);
1505         *dst_p = NULL;
1506         return err;
1507 }
1508 EXPORT_SYMBOL(__xfrm_lookup);
1509
1510 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1511                 struct sock *sk, int flags)
1512 {
1513         int err = __xfrm_lookup(dst_p, fl, sk, flags);
1514
1515         if (err == -EREMOTE) {
1516                 dst_release(*dst_p);
1517                 *dst_p = NULL;
1518                 err = -EAGAIN;
1519         }
1520
1521         return err;
1522 }
1523 EXPORT_SYMBOL(xfrm_lookup);
1524
1525 static inline int
1526 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1527 {
1528         struct xfrm_state *x;
1529
1530         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1531                 return 0;
1532         x = skb->sp->xvec[idx];
1533         if (!x->type->reject)
1534                 return 0;
1535         return x->type->reject(x, skb, fl);
1536 }
1537
1538 /* When skb is transformed back to its "native" form, we have to
1539  * check policy restrictions. At the moment we make this in maximally
1540  * stupid way. Shame on me. :-) Of course, connected sockets must
1541  * have policy cached at them.
1542  */
1543
1544 static inline int
1545 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1546               unsigned short family)
1547 {
1548         if (xfrm_state_kern(x))
1549                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1550         return  x->id.proto == tmpl->id.proto &&
1551                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1552                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1553                 x->props.mode == tmpl->mode &&
1554                 ((tmpl->aalgos & (1<<x->props.aalgo)) ||
1555                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1556                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1557                   xfrm_state_addr_cmp(tmpl, x, family));
1558 }
1559
1560 /*
1561  * 0 or more than 0 is returned when validation is succeeded (either bypass
1562  * because of optional transport mode, or next index of the mathced secpath
1563  * state with the template.
1564  * -1 is returned when no matching template is found.
1565  * Otherwise "-2 - errored_index" is returned.
1566  */
1567 static inline int
1568 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1569                unsigned short family)
1570 {
1571         int idx = start;
1572
1573         if (tmpl->optional) {
1574                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1575                         return start;
1576         } else
1577                 start = -1;
1578         for (; idx < sp->len; idx++) {
1579                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1580                         return ++idx;
1581                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1582                         if (start == -1)
1583                                 start = -2-idx;
1584                         break;
1585                 }
1586         }
1587         return start;
1588 }
1589
1590 int
1591 xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
1592 {
1593         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1594         int err;
1595
1596         if (unlikely(afinfo == NULL))
1597                 return -EAFNOSUPPORT;
1598
1599         afinfo->decode_session(skb, fl);
1600         err = security_xfrm_decode_session(skb, &fl->secid);
1601         xfrm_policy_put_afinfo(afinfo);
1602         return err;
1603 }
1604 EXPORT_SYMBOL(xfrm_decode_session);
1605
1606 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1607 {
1608         for (; k < sp->len; k++) {
1609                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1610                         *idxp = k;
1611                         return 1;
1612                 }
1613         }
1614
1615         return 0;
1616 }
1617
1618 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1619                         unsigned short family)
1620 {
1621         struct xfrm_policy *pol;
1622         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1623         int npols = 0;
1624         int xfrm_nr;
1625         int pi;
1626         struct flowi fl;
1627         u8 fl_dir = policy_to_flow_dir(dir);
1628         int xerr_idx = -1;
1629
1630         if (xfrm_decode_session(skb, &fl, family) < 0)
1631                 return 0;
1632         nf_nat_decode_session(skb, &fl, family);
1633
1634         /* First, check used SA against their selectors. */
1635         if (skb->sp) {
1636                 int i;
1637
1638                 for (i=skb->sp->len-1; i>=0; i--) {
1639                         struct xfrm_state *x = skb->sp->xvec[i];
1640                         if (!xfrm_selector_match(&x->sel, &fl, family))
1641                                 return 0;
1642                 }
1643         }
1644
1645         pol = NULL;
1646         if (sk && sk->sk_policy[dir]) {
1647                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1648                 if (IS_ERR(pol))
1649                         return 0;
1650         }
1651
1652         if (!pol)
1653                 pol = flow_cache_lookup(&fl, family, fl_dir,
1654                                         xfrm_policy_lookup);
1655
1656         if (IS_ERR(pol))
1657                 return 0;
1658
1659         if (!pol) {
1660                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1661                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1662                         return 0;
1663                 }
1664                 return 1;
1665         }
1666
1667         pol->curlft.use_time = get_seconds();
1668
1669         pols[0] = pol;
1670         npols ++;
1671 #ifdef CONFIG_XFRM_SUB_POLICY
1672         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1673                 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1674                                                     &fl, family,
1675                                                     XFRM_POLICY_IN);
1676                 if (pols[1]) {
1677                         if (IS_ERR(pols[1]))
1678                                 return 0;
1679                         pols[1]->curlft.use_time = get_seconds();
1680                         npols ++;
1681                 }
1682         }
1683 #endif
1684
1685         if (pol->action == XFRM_POLICY_ALLOW) {
1686                 struct sec_path *sp;
1687                 static struct sec_path dummy;
1688                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1689                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1690                 struct xfrm_tmpl **tpp = tp;
1691                 int ti = 0;
1692                 int i, k;
1693
1694                 if ((sp = skb->sp) == NULL)
1695                         sp = &dummy;
1696
1697                 for (pi = 0; pi < npols; pi++) {
1698                         if (pols[pi] != pol &&
1699                             pols[pi]->action != XFRM_POLICY_ALLOW)
1700                                 goto reject;
1701                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH)
1702                                 goto reject_error;
1703                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
1704                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1705                 }
1706                 xfrm_nr = ti;
1707                 if (npols > 1) {
1708                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1709                         tpp = stp;
1710                 }
1711
1712                 /* For each tunnel xfrm, find the first matching tmpl.
1713                  * For each tmpl before that, find corresponding xfrm.
1714                  * Order is _important_. Later we will implement
1715                  * some barriers, but at the moment barriers
1716                  * are implied between each two transformations.
1717                  */
1718                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
1719                         k = xfrm_policy_ok(tpp[i], sp, k, family);
1720                         if (k < 0) {
1721                                 if (k < -1)
1722                                         /* "-2 - errored_index" returned */
1723                                         xerr_idx = -(2+k);
1724                                 goto reject;
1725                         }
1726                 }
1727
1728                 if (secpath_has_nontransport(sp, k, &xerr_idx))
1729                         goto reject;
1730
1731                 xfrm_pols_put(pols, npols);
1732                 return 1;
1733         }
1734
1735 reject:
1736         xfrm_secpath_reject(xerr_idx, skb, &fl);
1737 reject_error:
1738         xfrm_pols_put(pols, npols);
1739         return 0;
1740 }
1741 EXPORT_SYMBOL(__xfrm_policy_check);
1742
1743 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1744 {
1745         struct flowi fl;
1746
1747         if (xfrm_decode_session(skb, &fl, family) < 0)
1748                 return 0;
1749
1750         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1751 }
1752 EXPORT_SYMBOL(__xfrm_route_forward);
1753
1754 /* Optimize later using cookies and generation ids. */
1755
1756 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1757 {
1758         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
1759          * to "-1" to force all XFRM destinations to get validated by
1760          * dst_ops->check on every use.  We do this because when a
1761          * normal route referenced by an XFRM dst is obsoleted we do
1762          * not go looking around for all parent referencing XFRM dsts
1763          * so that we can invalidate them.  It is just too much work.
1764          * Instead we make the checks here on every use.  For example:
1765          *
1766          *      XFRM dst A --> IPv4 dst X
1767          *
1768          * X is the "xdst->route" of A (X is also the "dst->path" of A
1769          * in this example).  If X is marked obsolete, "A" will not
1770          * notice.  That's what we are validating here via the
1771          * stale_bundle() check.
1772          *
1773          * When a policy's bundle is pruned, we dst_free() the XFRM
1774          * dst which causes it's ->obsolete field to be set to a
1775          * positive non-zero integer.  If an XFRM dst has been pruned
1776          * like this, we want to force a new route lookup.
1777          */
1778         if (dst->obsolete < 0 && !stale_bundle(dst))
1779                 return dst;
1780
1781         return NULL;
1782 }
1783
1784 static int stale_bundle(struct dst_entry *dst)
1785 {
1786         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
1787 }
1788
1789 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1790 {
1791         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1792                 dst->dev = init_net.loopback_dev;
1793                 dev_hold(dst->dev);
1794                 dev_put(dev);
1795         }
1796 }
1797 EXPORT_SYMBOL(xfrm_dst_ifdown);
1798
1799 static void xfrm_link_failure(struct sk_buff *skb)
1800 {
1801         /* Impossible. Such dst must be popped before reaches point of failure. */
1802         return;
1803 }
1804
1805 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1806 {
1807         if (dst) {
1808                 if (dst->obsolete) {
1809                         dst_release(dst);
1810                         dst = NULL;
1811                 }
1812         }
1813         return dst;
1814 }
1815
1816 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
1817 {
1818         struct dst_entry *dst, **dstp;
1819
1820         write_lock(&pol->lock);
1821         dstp = &pol->bundles;
1822         while ((dst=*dstp) != NULL) {
1823                 if (func(dst)) {
1824                         *dstp = dst->next;
1825                         dst->next = *gc_list_p;
1826                         *gc_list_p = dst;
1827                 } else {
1828                         dstp = &dst->next;
1829                 }
1830         }
1831         write_unlock(&pol->lock);
1832 }
1833
1834 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1835 {
1836         struct dst_entry *gc_list = NULL;
1837         int dir;
1838
1839         read_lock_bh(&xfrm_policy_lock);
1840         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
1841                 struct xfrm_policy *pol;
1842                 struct hlist_node *entry;
1843                 struct hlist_head *table;
1844                 int i;
1845
1846                 hlist_for_each_entry(pol, entry,
1847                                      &xfrm_policy_inexact[dir], bydst)
1848                         prune_one_bundle(pol, func, &gc_list);
1849
1850                 table = xfrm_policy_bydst[dir].table;
1851                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
1852                         hlist_for_each_entry(pol, entry, table + i, bydst)
1853                                 prune_one_bundle(pol, func, &gc_list);
1854                 }
1855         }
1856         read_unlock_bh(&xfrm_policy_lock);
1857
1858         while (gc_list) {
1859                 struct dst_entry *dst = gc_list;
1860                 gc_list = dst->next;
1861                 dst_free(dst);
1862         }
1863 }
1864
1865 static int unused_bundle(struct dst_entry *dst)
1866 {
1867         return !atomic_read(&dst->__refcnt);
1868 }
1869
1870 static void __xfrm_garbage_collect(void)
1871 {
1872         xfrm_prune_bundles(unused_bundle);
1873 }
1874
1875 static int xfrm_flush_bundles(void)
1876 {
1877         xfrm_prune_bundles(stale_bundle);
1878         return 0;
1879 }
1880
1881 void xfrm_init_pmtu(struct dst_entry *dst)
1882 {
1883         do {
1884                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1885                 u32 pmtu, route_mtu_cached;
1886
1887                 pmtu = dst_mtu(dst->child);
1888                 xdst->child_mtu_cached = pmtu;
1889
1890                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
1891
1892                 route_mtu_cached = dst_mtu(xdst->route);
1893                 xdst->route_mtu_cached = route_mtu_cached;
1894
1895                 if (pmtu > route_mtu_cached)
1896                         pmtu = route_mtu_cached;
1897
1898                 dst->metrics[RTAX_MTU-1] = pmtu;
1899         } while ((dst = dst->next));
1900 }
1901
1902 EXPORT_SYMBOL(xfrm_init_pmtu);
1903
1904 /* Check that the bundle accepts the flow and its components are
1905  * still valid.
1906  */
1907
1908 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
1909                 struct flowi *fl, int family, int strict)
1910 {
1911         struct dst_entry *dst = &first->u.dst;
1912         struct xfrm_dst *last;
1913         u32 mtu;
1914
1915         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
1916             (dst->dev && !netif_running(dst->dev)))
1917                 return 0;
1918 #ifdef CONFIG_XFRM_SUB_POLICY
1919         if (fl) {
1920                 if (first->origin && !flow_cache_uli_match(first->origin, fl))
1921                         return 0;
1922                 if (first->partner &&
1923                     !xfrm_selector_match(first->partner, fl, family))
1924                         return 0;
1925         }
1926 #endif
1927
1928         last = NULL;
1929
1930         do {
1931                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1932
1933                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
1934                         return 0;
1935                 if (fl && pol &&
1936                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
1937                         return 0;
1938                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
1939                         return 0;
1940                 if (xdst->genid != dst->xfrm->genid)
1941                         return 0;
1942
1943                 if (strict && fl &&
1944                     !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
1945                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
1946                         return 0;
1947
1948                 mtu = dst_mtu(dst->child);
1949                 if (xdst->child_mtu_cached != mtu) {
1950                         last = xdst;
1951                         xdst->child_mtu_cached = mtu;
1952                 }
1953
1954                 if (!dst_check(xdst->route, xdst->route_cookie))
1955                         return 0;
1956                 mtu = dst_mtu(xdst->route);
1957                 if (xdst->route_mtu_cached != mtu) {
1958                         last = xdst;
1959                         xdst->route_mtu_cached = mtu;
1960                 }
1961
1962                 dst = dst->child;
1963         } while (dst->xfrm);
1964
1965         if (likely(!last))
1966                 return 1;
1967
1968         mtu = last->child_mtu_cached;
1969         for (;;) {
1970                 dst = &last->u.dst;
1971
1972                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
1973                 if (mtu > last->route_mtu_cached)
1974                         mtu = last->route_mtu_cached;
1975                 dst->metrics[RTAX_MTU-1] = mtu;
1976
1977                 if (last == first)
1978                         break;
1979
1980                 last = (struct xfrm_dst *)last->u.dst.next;
1981                 last->child_mtu_cached = mtu;
1982         }
1983
1984         return 1;
1985 }
1986
1987 EXPORT_SYMBOL(xfrm_bundle_ok);
1988
1989 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
1990 {
1991         int err = 0;
1992         if (unlikely(afinfo == NULL))
1993                 return -EINVAL;
1994         if (unlikely(afinfo->family >= NPROTO))
1995                 return -EAFNOSUPPORT;
1996         write_lock_bh(&xfrm_policy_afinfo_lock);
1997         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
1998                 err = -ENOBUFS;
1999         else {
2000                 struct dst_ops *dst_ops = afinfo->dst_ops;
2001                 if (likely(dst_ops->kmem_cachep == NULL))
2002                         dst_ops->kmem_cachep = xfrm_dst_cache;
2003                 if (likely(dst_ops->check == NULL))
2004                         dst_ops->check = xfrm_dst_check;
2005                 if (likely(dst_ops->negative_advice == NULL))
2006                         dst_ops->negative_advice = xfrm_negative_advice;
2007                 if (likely(dst_ops->link_failure == NULL))
2008                         dst_ops->link_failure = xfrm_link_failure;
2009                 if (likely(afinfo->garbage_collect == NULL))
2010                         afinfo->garbage_collect = __xfrm_garbage_collect;
2011                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2012         }
2013         write_unlock_bh(&xfrm_policy_afinfo_lock);
2014         return err;
2015 }
2016 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2017
2018 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2019 {
2020         int err = 0;
2021         if (unlikely(afinfo == NULL))
2022                 return -EINVAL;
2023         if (unlikely(afinfo->family >= NPROTO))
2024                 return -EAFNOSUPPORT;
2025         write_lock_bh(&xfrm_policy_afinfo_lock);
2026         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2027                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2028                         err = -EINVAL;
2029                 else {
2030                         struct dst_ops *dst_ops = afinfo->dst_ops;
2031                         xfrm_policy_afinfo[afinfo->family] = NULL;
2032                         dst_ops->kmem_cachep = NULL;
2033                         dst_ops->check = NULL;
2034                         dst_ops->negative_advice = NULL;
2035                         dst_ops->link_failure = NULL;
2036                         afinfo->garbage_collect = NULL;
2037                 }
2038         }
2039         write_unlock_bh(&xfrm_policy_afinfo_lock);
2040         return err;
2041 }
2042 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2043
2044 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2045 {
2046         struct xfrm_policy_afinfo *afinfo;
2047         if (unlikely(family >= NPROTO))
2048                 return NULL;
2049         read_lock(&xfrm_policy_afinfo_lock);
2050         afinfo = xfrm_policy_afinfo[family];
2051         if (unlikely(!afinfo))
2052                 read_unlock(&xfrm_policy_afinfo_lock);
2053         return afinfo;
2054 }
2055
2056 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2057 {
2058         read_unlock(&xfrm_policy_afinfo_lock);
2059 }
2060
2061 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2062 {
2063         struct net_device *dev = ptr;
2064
2065         if (dev->nd_net != &init_net)
2066                 return NOTIFY_DONE;
2067
2068         switch (event) {
2069         case NETDEV_DOWN:
2070                 xfrm_flush_bundles();
2071         }
2072         return NOTIFY_DONE;
2073 }
2074
2075 static struct notifier_block xfrm_dev_notifier = {
2076         xfrm_dev_event,
2077         NULL,
2078         0
2079 };
2080
2081 static void __init xfrm_policy_init(void)
2082 {
2083         unsigned int hmask, sz;
2084         int dir;
2085
2086         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2087                                            sizeof(struct xfrm_dst),
2088                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2089                                            NULL);
2090
2091         hmask = 8 - 1;
2092         sz = (hmask+1) * sizeof(struct hlist_head);
2093
2094         xfrm_policy_byidx = xfrm_hash_alloc(sz);
2095         xfrm_idx_hmask = hmask;
2096         if (!xfrm_policy_byidx)
2097                 panic("XFRM: failed to allocate byidx hash\n");
2098
2099         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2100                 struct xfrm_policy_hash *htab;
2101
2102                 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2103
2104                 htab = &xfrm_policy_bydst[dir];
2105                 htab->table = xfrm_hash_alloc(sz);
2106                 htab->hmask = hmask;
2107                 if (!htab->table)
2108                         panic("XFRM: failed to allocate bydst hash\n");
2109         }
2110
2111         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task);
2112         register_netdevice_notifier(&xfrm_dev_notifier);
2113 }
2114
2115 void __init xfrm_init(void)
2116 {
2117         xfrm_state_init();
2118         xfrm_policy_init();
2119         xfrm_input_init();
2120 }
2121
2122 #ifdef CONFIG_AUDITSYSCALL
2123 static inline void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2124                                                 struct audit_buffer *audit_buf)
2125 {
2126         if (xp->security)
2127                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2128                                  xp->security->ctx_alg, xp->security->ctx_doi,
2129                                  xp->security->ctx_str);
2130
2131         switch(xp->selector.family) {
2132         case AF_INET:
2133                 audit_log_format(audit_buf, " src=%u.%u.%u.%u dst=%u.%u.%u.%u",
2134                                  NIPQUAD(xp->selector.saddr.a4),
2135                                  NIPQUAD(xp->selector.daddr.a4));
2136                 break;
2137         case AF_INET6:
2138                 {
2139                         struct in6_addr saddr6, daddr6;
2140
2141                         memcpy(&saddr6, xp->selector.saddr.a6,
2142                                 sizeof(struct in6_addr));
2143                         memcpy(&daddr6, xp->selector.daddr.a6,
2144                                 sizeof(struct in6_addr));
2145                         audit_log_format(audit_buf,
2146                                 " src=" NIP6_FMT " dst=" NIP6_FMT,
2147                                 NIP6(saddr6), NIP6(daddr6));
2148                 }
2149                 break;
2150         }
2151 }
2152
2153 void
2154 xfrm_audit_policy_add(struct xfrm_policy *xp, int result, u32 auid, u32 sid)
2155 {
2156         struct audit_buffer *audit_buf;
2157         extern int audit_enabled;
2158
2159         if (audit_enabled == 0)
2160                 return;
2161         audit_buf = xfrm_audit_start(sid, auid);
2162         if (audit_buf == NULL)
2163                 return;
2164         audit_log_format(audit_buf, " op=SPD-add res=%u", result);
2165         xfrm_audit_common_policyinfo(xp, audit_buf);
2166         audit_log_end(audit_buf);
2167 }
2168 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2169
2170 void
2171 xfrm_audit_policy_delete(struct xfrm_policy *xp, int result, u32 auid, u32 sid)
2172 {
2173         struct audit_buffer *audit_buf;
2174         extern int audit_enabled;
2175
2176         if (audit_enabled == 0)
2177                 return;
2178         audit_buf = xfrm_audit_start(sid, auid);
2179         if (audit_buf == NULL)
2180                 return;
2181         audit_log_format(audit_buf, " op=SPD-delete res=%u", result);
2182         xfrm_audit_common_policyinfo(xp, audit_buf);
2183         audit_log_end(audit_buf);
2184 }
2185 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2186 #endif
2187
2188 #ifdef CONFIG_XFRM_MIGRATE
2189 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2190                                        struct xfrm_selector *sel_tgt)
2191 {
2192         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2193                 if (sel_tgt->family == sel_cmp->family &&
2194                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2195                                   sel_cmp->family) == 0 &&
2196                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2197                                   sel_cmp->family) == 0 &&
2198                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2199                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2200                         return 1;
2201                 }
2202         } else {
2203                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2204                         return 1;
2205                 }
2206         }
2207         return 0;
2208 }
2209
2210 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2211                                                      u8 dir, u8 type)
2212 {
2213         struct xfrm_policy *pol, *ret = NULL;
2214         struct hlist_node *entry;
2215         struct hlist_head *chain;
2216         u32 priority = ~0U;
2217
2218         read_lock_bh(&xfrm_policy_lock);
2219         chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2220         hlist_for_each_entry(pol, entry, chain, bydst) {
2221                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2222                     pol->type == type) {
2223                         ret = pol;
2224                         priority = ret->priority;
2225                         break;
2226                 }
2227         }
2228         chain = &xfrm_policy_inexact[dir];
2229         hlist_for_each_entry(pol, entry, chain, bydst) {
2230                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2231                     pol->type == type &&
2232                     pol->priority < priority) {
2233                         ret = pol;
2234                         break;
2235                 }
2236         }
2237
2238         if (ret)
2239                 xfrm_pol_hold(ret);
2240
2241         read_unlock_bh(&xfrm_policy_lock);
2242
2243         return ret;
2244 }
2245
2246 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2247 {
2248         int match = 0;
2249
2250         if (t->mode == m->mode && t->id.proto == m->proto &&
2251             (m->reqid == 0 || t->reqid == m->reqid)) {
2252                 switch (t->mode) {
2253                 case XFRM_MODE_TUNNEL:
2254                 case XFRM_MODE_BEET:
2255                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2256                                           m->old_family) == 0 &&
2257                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2258                                           m->old_family) == 0) {
2259                                 match = 1;
2260                         }
2261                         break;
2262                 case XFRM_MODE_TRANSPORT:
2263                         /* in case of transport mode, template does not store
2264                            any IP addresses, hence we just compare mode and
2265                            protocol */
2266                         match = 1;
2267                         break;
2268                 default:
2269                         break;
2270                 }
2271         }
2272         return match;
2273 }
2274
2275 /* update endpoint address(es) of template(s) */
2276 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2277                                struct xfrm_migrate *m, int num_migrate)
2278 {
2279         struct xfrm_migrate *mp;
2280         struct dst_entry *dst;
2281         int i, j, n = 0;
2282
2283         write_lock_bh(&pol->lock);
2284         if (unlikely(pol->dead)) {
2285                 /* target policy has been deleted */
2286                 write_unlock_bh(&pol->lock);
2287                 return -ENOENT;
2288         }
2289
2290         for (i = 0; i < pol->xfrm_nr; i++) {
2291                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2292                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2293                                 continue;
2294                         n++;
2295                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2296                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2297                                 continue;
2298                         /* update endpoints */
2299                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2300                                sizeof(pol->xfrm_vec[i].id.daddr));
2301                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2302                                sizeof(pol->xfrm_vec[i].saddr));
2303                         pol->xfrm_vec[i].encap_family = mp->new_family;
2304                         /* flush bundles */
2305                         while ((dst = pol->bundles) != NULL) {
2306                                 pol->bundles = dst->next;
2307                                 dst_free(dst);
2308                         }
2309                 }
2310         }
2311
2312         write_unlock_bh(&pol->lock);
2313
2314         if (!n)
2315                 return -ENODATA;
2316
2317         return 0;
2318 }
2319
2320 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2321 {
2322         int i, j;
2323
2324         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2325                 return -EINVAL;
2326
2327         for (i = 0; i < num_migrate; i++) {
2328                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2329                                    m[i].old_family) == 0) &&
2330                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2331                                    m[i].old_family) == 0))
2332                         return -EINVAL;
2333                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2334                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2335                         return -EINVAL;
2336
2337                 /* check if there is any duplicated entry */
2338                 for (j = i + 1; j < num_migrate; j++) {
2339                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2340                                     sizeof(m[i].old_daddr)) &&
2341                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2342                                     sizeof(m[i].old_saddr)) &&
2343                             m[i].proto == m[j].proto &&
2344                             m[i].mode == m[j].mode &&
2345                             m[i].reqid == m[j].reqid &&
2346                             m[i].old_family == m[j].old_family)
2347                                 return -EINVAL;
2348                 }
2349         }
2350
2351         return 0;
2352 }
2353
2354 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2355                  struct xfrm_migrate *m, int num_migrate)
2356 {
2357         int i, err, nx_cur = 0, nx_new = 0;
2358         struct xfrm_policy *pol = NULL;
2359         struct xfrm_state *x, *xc;
2360         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2361         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2362         struct xfrm_migrate *mp;
2363
2364         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2365                 goto out;
2366
2367         /* Stage 1 - find policy */
2368         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2369                 err = -ENOENT;
2370                 goto out;
2371         }
2372
2373         /* Stage 2 - find and update state(s) */
2374         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2375                 if ((x = xfrm_migrate_state_find(mp))) {
2376                         x_cur[nx_cur] = x;
2377                         nx_cur++;
2378                         if ((xc = xfrm_state_migrate(x, mp))) {
2379                                 x_new[nx_new] = xc;
2380                                 nx_new++;
2381                         } else {
2382                                 err = -ENODATA;
2383                                 goto restore_state;
2384                         }
2385                 }
2386         }
2387
2388         /* Stage 3 - update policy */
2389         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2390                 goto restore_state;
2391
2392         /* Stage 4 - delete old state(s) */
2393         if (nx_cur) {
2394                 xfrm_states_put(x_cur, nx_cur);
2395                 xfrm_states_delete(x_cur, nx_cur);
2396         }
2397
2398         /* Stage 5 - announce */
2399         km_migrate(sel, dir, type, m, num_migrate);
2400
2401         xfrm_pol_put(pol);
2402
2403         return 0;
2404 out:
2405         return err;
2406
2407 restore_state:
2408         if (pol)
2409                 xfrm_pol_put(pol);
2410         if (nx_cur)
2411                 xfrm_states_put(x_cur, nx_cur);
2412         if (nx_new)
2413                 xfrm_states_delete(x_new, nx_new);
2414
2415         return err;
2416 }
2417 EXPORT_SYMBOL(xfrm_migrate);
2418 #endif