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/err.h>
  17 #include <linux/slab.h>
  18 #include <linux/kmod.h>
  19 #include <linux/list.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/workqueue.h>
  22 #include <linux/notifier.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/netfilter.h>
  25 #include <linux/module.h>
  26 #include <linux/cache.h>
  27 #include <linux/audit.h>
  28 #include <net/dst.h>
  29 #include <net/xfrm.h>
  30 #include <net/ip.h>
  31 #ifdef CONFIG_XFRM_STATISTICS
  32 #include <net/snmp.h>
  33 #endif
  34
  35 #include "xfrm_hash.h"
  36
  37 int sysctl_xfrm_larval_drop __read_mostly = 1;
  38
  39 #ifdef CONFIG_XFRM_STATISTICS
  40 DEFINE_SNMP_STAT(struct linux_xfrm_mib, xfrm_statistics) __read_mostly;
  41 EXPORT_SYMBOL(xfrm_statistics);
  42 #endif
  43
  44 DEFINE_MUTEX(xfrm_cfg_mutex);
  45 EXPORT_SYMBOL(xfrm_cfg_mutex);
  46
  47 static DEFINE_RWLOCK(xfrm_policy_lock);
  48
  49 static struct list_head xfrm_policy_all;
  50 unsigned int xfrm_policy_count[XFRM_POLICY_MAX*2];
  51 EXPORT_SYMBOL(xfrm_policy_count);
  52
  53 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
  54 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
  55
  56 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  57
  58 static struct work_struct xfrm_policy_gc_work;
  59 static HLIST_HEAD(xfrm_policy_gc_list);
  60 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
  61
  62 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
  63 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
  64 static void xfrm_init_pmtu(struct dst_entry *dst);
  65
  66 static inline int
  67 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  68 {
  69         return  addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
  70                 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
  71                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  72                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  73                 (fl->proto == sel->proto || !sel->proto) &&
  74                 (fl->oif == sel->ifindex || !sel->ifindex);
  75 }
  76
  77 static inline int
  78 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  79 {
  80         return  addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
  81                 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
  82                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  83                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  84                 (fl->proto == sel->proto || !sel->proto) &&
  85                 (fl->oif == sel->ifindex || !sel->ifindex);
  86 }
  87
  88 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
  89                     unsigned short family)
  90 {
  91         switch (family) {
  92         case AF_INET:
  93                 return __xfrm4_selector_match(sel, fl);
  94         case AF_INET6:
  95                 return __xfrm6_selector_match(sel, fl);
  96         }
  97         return 0;
  98 }
  99
 100 static inline struct dst_entry *__xfrm_dst_lookup(int tos,
 101                                                   xfrm_address_t *saddr,
 102                                                   xfrm_address_t *daddr,
 103                                                   int family)
 104 {
 105         struct xfrm_policy_afinfo *afinfo;
 106         struct dst_entry *dst;
 107
 108         afinfo = xfrm_policy_get_afinfo(family);
 109         if (unlikely(afinfo == NULL))
 110                 return ERR_PTR(-EAFNOSUPPORT);
 111
 112         dst = afinfo->dst_lookup(tos, saddr, daddr);
 113
 114         xfrm_policy_put_afinfo(afinfo);
 115
 116         return dst;
 117 }
 118
 119 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
 120                                                 xfrm_address_t *prev_saddr,
 121                                                 xfrm_address_t *prev_daddr,
 122                                                 int family)
 123 {
 124         xfrm_address_t *saddr = &x->props.saddr;
 125         xfrm_address_t *daddr = &x->id.daddr;
 126         struct dst_entry *dst;
 127
 128         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 129                 saddr = x->coaddr;
 130                 daddr = prev_daddr;
 131         }
 132         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 133                 saddr = prev_saddr;
 134                 daddr = x->coaddr;
 135         }
 136
 137         dst = __xfrm_dst_lookup(tos, saddr, daddr, family);
 138
 139         if (!IS_ERR(dst)) {
 140                 if (prev_saddr != saddr)
 141                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 142                 if (prev_daddr != daddr)
 143                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 144         }
 145
 146         return dst;
 147 }
 148
 149 static inline unsigned long make_jiffies(long secs)
 150 {
 151         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 152                 return MAX_SCHEDULE_TIMEOUT-1;
 153         else
 154                 return secs*HZ;
 155 }
 156
 157 static void xfrm_policy_timer(unsigned long data)
 158 {
 159         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 160         unsigned long now = get_seconds();
 161         long next = LONG_MAX;
 162         int warn = 0;
 163         int dir;
 164
 165         read_lock(&xp->lock);
 166
 167         if (xp->walk.dead)
 168                 goto out;
 169
 170         dir = xfrm_policy_id2dir(xp->index);
 171
 172         if (xp->lft.hard_add_expires_seconds) {
 173                 long tmo = xp->lft.hard_add_expires_seconds +
 174                         xp->curlft.add_time - now;
 175                 if (tmo <= 0)
 176                         goto expired;
 177                 if (tmo < next)
 178                         next = tmo;
 179         }
 180         if (xp->lft.hard_use_expires_seconds) {
 181                 long tmo = xp->lft.hard_use_expires_seconds +
 182                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 183                 if (tmo <= 0)
 184                         goto expired;
 185                 if (tmo < next)
 186                         next = tmo;
 187         }
 188         if (xp->lft.soft_add_expires_seconds) {
 189                 long tmo = xp->lft.soft_add_expires_seconds +
 190                         xp->curlft.add_time - now;
 191                 if (tmo <= 0) {
 192                         warn = 1;
 193                         tmo = XFRM_KM_TIMEOUT;
 194                 }
 195                 if (tmo < next)
 196                         next = tmo;
 197         }
 198         if (xp->lft.soft_use_expires_seconds) {
 199                 long tmo = xp->lft.soft_use_expires_seconds +
 200                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 201                 if (tmo <= 0) {
 202                         warn = 1;
 203                         tmo = XFRM_KM_TIMEOUT;
 204                 }
 205                 if (tmo < next)
 206                         next = tmo;
 207         }
 208
 209         if (warn)
 210                 km_policy_expired(xp, dir, 0, 0);
 211         if (next != LONG_MAX &&
 212             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 213                 xfrm_pol_hold(xp);
 214
 215 out:
 216         read_unlock(&xp->lock);
 217         xfrm_pol_put(xp);
 218         return;
 219
 220 expired:
 221         read_unlock(&xp->lock);
 222         if (!xfrm_policy_delete(xp, dir))
 223                 km_policy_expired(xp, dir, 1, 0);
 224         xfrm_pol_put(xp);
 225 }
 226
 227
 228 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 229  * SPD calls.
 230  */
 231
 232 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
 233 {
 234         struct xfrm_policy *policy;
 235
 236         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 237
 238         if (policy) {
 239                 INIT_LIST_HEAD(&policy->walk.all);
 240                 INIT_HLIST_NODE(&policy->bydst);
 241                 INIT_HLIST_NODE(&policy->byidx);
 242                 rwlock_init(&policy->lock);
 243                 atomic_set(&policy->refcnt, 1);
 244                 setup_timer(&policy->timer, xfrm_policy_timer,
 245                                 (unsigned long)policy);
 246         }
 247         return policy;
 248 }
 249 EXPORT_SYMBOL(xfrm_policy_alloc);
 250
 251 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 252
 253 void xfrm_policy_destroy(struct xfrm_policy *policy)
 254 {
 255         BUG_ON(!policy->walk.dead);
 256
 257         BUG_ON(policy->bundles);
 258
 259         if (del_timer(&policy->timer))
 260                 BUG();
 261
 262         security_xfrm_policy_free(policy->security);
 263         kfree(policy);
 264 }
 265 EXPORT_SYMBOL(xfrm_policy_destroy);
 266
 267 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
 268 {
 269         struct dst_entry *dst;
 270
 271         while ((dst = policy->bundles) != NULL) {
 272                 policy->bundles = dst->next;
 273                 dst_free(dst);
 274         }
 275
 276         if (del_timer(&policy->timer))
 277                 atomic_dec(&policy->refcnt);
 278
 279         if (atomic_read(&policy->refcnt) > 1)
 280                 flow_cache_flush();
 281
 282         xfrm_pol_put(policy);
 283 }
 284
 285 static void xfrm_policy_gc_task(struct work_struct *work)
 286 {
 287         struct xfrm_policy *policy;
 288         struct hlist_node *entry, *tmp;
 289         struct hlist_head gc_list;
 290
 291         spin_lock_bh(&xfrm_policy_gc_lock);
 292         gc_list.first = xfrm_policy_gc_list.first;
 293         INIT_HLIST_HEAD(&xfrm_policy_gc_list);
 294         spin_unlock_bh(&xfrm_policy_gc_lock);
 295
 296         hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
 297                 xfrm_policy_gc_kill(policy);
 298 }
 299
 300 /* Rule must be locked. Release descentant resources, announce
 301  * entry dead. The rule must be unlinked from lists to the moment.
 302  */
 303
 304 static void xfrm_policy_kill(struct xfrm_policy *policy)
 305 {
 306         int dead;
 307
 308         write_lock_bh(&policy->lock);
 309         dead = policy->walk.dead;
 310         policy->walk.dead = 1;
 311         write_unlock_bh(&policy->lock);
 312
 313         if (unlikely(dead)) {
 314                 WARN_ON(1);
 315                 return;
 316         }
 317
 318         spin_lock(&xfrm_policy_gc_lock);
 319         hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
 320         spin_unlock(&xfrm_policy_gc_lock);
 321
 322         schedule_work(&xfrm_policy_gc_work);
 323 }
 324
 325 struct xfrm_policy_hash {
 326         struct hlist_head       *table;
 327         unsigned int            hmask;
 328 };
 329
 330 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
 331 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
 332 static struct hlist_head *xfrm_policy_byidx __read_mostly;
 333 static unsigned int xfrm_idx_hmask __read_mostly;
 334 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 335
 336 static inline unsigned int idx_hash(u32 index)
 337 {
 338         return __idx_hash(index, xfrm_idx_hmask);
 339 }
 340
 341 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
 342 {
 343         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 344         unsigned int hash = __sel_hash(sel, family, hmask);
 345
 346         return (hash == hmask + 1 ?
 347                 &xfrm_policy_inexact[dir] :
 348                 xfrm_policy_bydst[dir].table + hash);
 349 }
 350
 351 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
 352 {
 353         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 354         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 355
 356         return xfrm_policy_bydst[dir].table + hash;
 357 }
 358
 359 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 360                                    struct hlist_head *ndsttable,
 361                                    unsigned int nhashmask)
 362 {
 363         struct hlist_node *entry, *tmp, *entry0 = NULL;
 364         struct xfrm_policy *pol;
 365         unsigned int h0 = 0;
 366
 367 redo:
 368         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
 369                 unsigned int h;
 370
 371                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 372                                 pol->family, nhashmask);
 373                 if (!entry0) {
 374                         hlist_del(entry);
 375                         hlist_add_head(&pol->bydst, ndsttable+h);
 376                         h0 = h;
 377                 } else {
 378                         if (h != h0)
 379                                 continue;
 380                         hlist_del(entry);
 381                         hlist_add_after(entry0, &pol->bydst);
 382                 }
 383                 entry0 = entry;
 384         }
 385         if (!hlist_empty(list)) {
 386                 entry0 = NULL;
 387                 goto redo;
 388         }
 389 }
 390
 391 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 392                                    struct hlist_head *nidxtable,
 393                                    unsigned int nhashmask)
 394 {
 395         struct hlist_node *entry, *tmp;
 396         struct xfrm_policy *pol;
 397
 398         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
 399                 unsigned int h;
 400
 401                 h = __idx_hash(pol->index, nhashmask);
 402                 hlist_add_head(&pol->byidx, nidxtable+h);
 403         }
 404 }
 405
 406 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 407 {
 408         return ((old_hmask + 1) << 1) - 1;
 409 }
 410
 411 static void xfrm_bydst_resize(int dir)
 412 {
 413         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 414         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 415         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 416         struct hlist_head *odst = xfrm_policy_bydst[dir].table;
 417         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 418         int i;
 419
 420         if (!ndst)
 421                 return;
 422
 423         write_lock_bh(&xfrm_policy_lock);
 424
 425         for (i = hmask; i >= 0; i--)
 426                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 427
 428         xfrm_policy_bydst[dir].table = ndst;
 429         xfrm_policy_bydst[dir].hmask = nhashmask;
 430
 431         write_unlock_bh(&xfrm_policy_lock);
 432
 433         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 434 }
 435
 436 static void xfrm_byidx_resize(int total)
 437 {
 438         unsigned int hmask = xfrm_idx_hmask;
 439         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 440         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 441         struct hlist_head *oidx = xfrm_policy_byidx;
 442         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 443         int i;
 444
 445         if (!nidx)
 446                 return;
 447
 448         write_lock_bh(&xfrm_policy_lock);
 449
 450         for (i = hmask; i >= 0; i--)
 451                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 452
 453         xfrm_policy_byidx = nidx;
 454         xfrm_idx_hmask = nhashmask;
 455
 456         write_unlock_bh(&xfrm_policy_lock);
 457
 458         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 459 }
 460
 461 static inline int xfrm_bydst_should_resize(int dir, int *total)
 462 {
 463         unsigned int cnt = xfrm_policy_count[dir];
 464         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 465
 466         if (total)
 467                 *total += cnt;
 468
 469         if ((hmask + 1) < xfrm_policy_hashmax &&
 470             cnt > hmask)
 471                 return 1;
 472
 473         return 0;
 474 }
 475
 476 static inline int xfrm_byidx_should_resize(int total)
 477 {
 478         unsigned int hmask = xfrm_idx_hmask;
 479
 480         if ((hmask + 1) < xfrm_policy_hashmax &&
 481             total > hmask)
 482                 return 1;
 483
 484         return 0;
 485 }
 486
 487 void xfrm_spd_getinfo(struct xfrmk_spdinfo *si)
 488 {
 489         read_lock_bh(&xfrm_policy_lock);
 490         si->incnt = xfrm_policy_count[XFRM_POLICY_IN];
 491         si->outcnt = xfrm_policy_count[XFRM_POLICY_OUT];
 492         si->fwdcnt = xfrm_policy_count[XFRM_POLICY_FWD];
 493         si->inscnt = xfrm_policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 494         si->outscnt = xfrm_policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 495         si->fwdscnt = xfrm_policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 496         si->spdhcnt = xfrm_idx_hmask;
 497         si->spdhmcnt = xfrm_policy_hashmax;
 498         read_unlock_bh(&xfrm_policy_lock);
 499 }
 500 EXPORT_SYMBOL(xfrm_spd_getinfo);
 501
 502 static DEFINE_MUTEX(hash_resize_mutex);
 503 static void xfrm_hash_resize(struct work_struct *__unused)
 504 {
 505         int dir, total;
 506
 507         mutex_lock(&hash_resize_mutex);
 508
 509         total = 0;
 510         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 511                 if (xfrm_bydst_should_resize(dir, &total))
 512                         xfrm_bydst_resize(dir);
 513         }
 514         if (xfrm_byidx_should_resize(total))
 515                 xfrm_byidx_resize(total);
 516
 517         mutex_unlock(&hash_resize_mutex);
 518 }
 519
 520 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize);
 521
 522 /* Generate new index... KAME seems to generate them ordered by cost
 523  * of an absolute inpredictability of ordering of rules. This will not pass. */
 524 static u32 xfrm_gen_index(u8 type, int dir)
 525 {
 526         static u32 idx_generator;
 527
 528         for (;;) {
 529                 struct hlist_node *entry;
 530                 struct hlist_head *list;
 531                 struct xfrm_policy *p;
 532                 u32 idx;
 533                 int found;
 534
 535                 idx = (idx_generator | dir);
 536                 idx_generator += 8;
 537                 if (idx == 0)
 538                         idx = 8;
 539                 list = xfrm_policy_byidx + idx_hash(idx);
 540                 found = 0;
 541                 hlist_for_each_entry(p, entry, list, byidx) {
 542                         if (p->index == idx) {
 543                                 found = 1;
 544                                 break;
 545                         }
 546                 }
 547                 if (!found)
 548                         return idx;
 549         }
 550 }
 551
 552 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 553 {
 554         u32 *p1 = (u32 *) s1;
 555         u32 *p2 = (u32 *) s2;
 556         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 557         int i;
 558
 559         for (i = 0; i < len; i++) {
 560                 if (p1[i] != p2[i])
 561                         return 1;
 562         }
 563
 564         return 0;
 565 }
 566
 567 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 568 {
 569         struct xfrm_policy *pol;
 570         struct xfrm_policy *delpol;
 571         struct hlist_head *chain;
 572         struct hlist_node *entry, *newpos;
 573         struct dst_entry *gc_list;
 574
 575         write_lock_bh(&xfrm_policy_lock);
 576         chain = policy_hash_bysel(&policy->selector, policy->family, dir);
 577         delpol = NULL;
 578         newpos = NULL;
 579         hlist_for_each_entry(pol, entry, chain, bydst) {
 580                 if (pol->type == policy->type &&
 581                     !selector_cmp(&pol->selector, &policy->selector) &&
 582                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 583                     !WARN_ON(delpol)) {
 584                         if (excl) {
 585                                 write_unlock_bh(&xfrm_policy_lock);
 586                                 return -EEXIST;
 587                         }
 588                         delpol = pol;
 589                         if (policy->priority > pol->priority)
 590                                 continue;
 591                 } else if (policy->priority >= pol->priority) {
 592                         newpos = &pol->bydst;
 593                         continue;
 594                 }
 595                 if (delpol)
 596                         break;
 597         }
 598         if (newpos)
 599                 hlist_add_after(newpos, &policy->bydst);
 600         else
 601                 hlist_add_head(&policy->bydst, chain);
 602         xfrm_pol_hold(policy);
 603         xfrm_policy_count[dir]++;
 604         atomic_inc(&flow_cache_genid);
 605         if (delpol) {
 606                 hlist_del(&delpol->bydst);
 607                 hlist_del(&delpol->byidx);
 608                 list_del(&delpol->walk.all);
 609                 xfrm_policy_count[dir]--;
 610         }
 611         policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
 612         hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
 613         policy->curlft.add_time = get_seconds();
 614         policy->curlft.use_time = 0;
 615         if (!mod_timer(&policy->timer, jiffies + HZ))
 616                 xfrm_pol_hold(policy);
 617         list_add(&policy->walk.all, &xfrm_policy_all);
 618         write_unlock_bh(&xfrm_policy_lock);
 619
 620         if (delpol)
 621                 xfrm_policy_kill(delpol);
 622         else if (xfrm_bydst_should_resize(dir, NULL))
 623                 schedule_work(&xfrm_hash_work);
 624
 625         read_lock_bh(&xfrm_policy_lock);
 626         gc_list = NULL;
 627         entry = &policy->bydst;
 628         hlist_for_each_entry_continue(policy, entry, bydst) {
 629                 struct dst_entry *dst;
 630
 631                 write_lock(&policy->lock);
 632                 dst = policy->bundles;
 633                 if (dst) {
 634                         struct dst_entry *tail = dst;
 635                         while (tail->next)
 636                                 tail = tail->next;
 637                         tail->next = gc_list;
 638                         gc_list = dst;
 639
 640                         policy->bundles = NULL;
 641                 }
 642                 write_unlock(&policy->lock);
 643         }
 644         read_unlock_bh(&xfrm_policy_lock);
 645
 646         while (gc_list) {
 647                 struct dst_entry *dst = gc_list;
 648
 649                 gc_list = dst->next;
 650                 dst_free(dst);
 651         }
 652
 653         return 0;
 654 }
 655 EXPORT_SYMBOL(xfrm_policy_insert);
 656
 657 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
 658                                           struct xfrm_selector *sel,
 659                                           struct xfrm_sec_ctx *ctx, int delete,
 660                                           int *err)
 661 {
 662         struct xfrm_policy *pol, *ret;
 663         struct hlist_head *chain;
 664         struct hlist_node *entry;
 665
 666         *err = 0;
 667         write_lock_bh(&xfrm_policy_lock);
 668         chain = policy_hash_bysel(sel, sel->family, dir);
 669         ret = NULL;
 670         hlist_for_each_entry(pol, entry, chain, bydst) {
 671                 if (pol->type == type &&
 672                     !selector_cmp(sel, &pol->selector) &&
 673                     xfrm_sec_ctx_match(ctx, pol->security)) {
 674                         xfrm_pol_hold(pol);
 675                         if (delete) {
 676                                 *err = security_xfrm_policy_delete(
 677                                                                 pol->security);
 678                                 if (*err) {
 679                                         write_unlock_bh(&xfrm_policy_lock);
 680                                         return pol;
 681                                 }
 682                                 hlist_del(&pol->bydst);
 683                                 hlist_del(&pol->byidx);
 684                                 list_del(&pol->walk.all);
 685                                 xfrm_policy_count[dir]--;
 686                         }
 687                         ret = pol;
 688                         break;
 689                 }
 690         }
 691         write_unlock_bh(&xfrm_policy_lock);
 692
 693         if (ret && delete) {
 694                 atomic_inc(&flow_cache_genid);
 695                 xfrm_policy_kill(ret);
 696         }
 697         return ret;
 698 }
 699 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 700
 701 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete,
 702                                      int *err)
 703 {
 704         struct xfrm_policy *pol, *ret;
 705         struct hlist_head *chain;
 706         struct hlist_node *entry;
 707
 708         *err = -ENOENT;
 709         if (xfrm_policy_id2dir(id) != dir)
 710                 return NULL;
 711
 712         *err = 0;
 713         write_lock_bh(&xfrm_policy_lock);
 714         chain = xfrm_policy_byidx + idx_hash(id);
 715         ret = NULL;
 716         hlist_for_each_entry(pol, entry, chain, byidx) {
 717                 if (pol->type == type && pol->index == id) {
 718                         xfrm_pol_hold(pol);
 719                         if (delete) {
 720                                 *err = security_xfrm_policy_delete(
 721                                                                 pol->security);
 722                                 if (*err) {
 723                                         write_unlock_bh(&xfrm_policy_lock);
 724                                         return pol;
 725                                 }
 726                                 hlist_del(&pol->bydst);
 727                                 hlist_del(&pol->byidx);
 728                                 list_del(&pol->walk.all);
 729                                 xfrm_policy_count[dir]--;
 730                         }
 731                         ret = pol;
 732                         break;
 733                 }
 734         }
 735         write_unlock_bh(&xfrm_policy_lock);
 736
 737         if (ret && delete) {
 738                 atomic_inc(&flow_cache_genid);
 739                 xfrm_policy_kill(ret);
 740         }
 741         return ret;
 742 }
 743 EXPORT_SYMBOL(xfrm_policy_byid);
 744
 745 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 746 static inline int
 747 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
 748 {
 749         int dir, err = 0;
 750
 751         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 752                 struct xfrm_policy *pol;
 753                 struct hlist_node *entry;
 754                 int i;
 755
 756                 hlist_for_each_entry(pol, entry,
 757                                      &xfrm_policy_inexact[dir], bydst) {
 758                         if (pol->type != type)
 759                                 continue;
 760                         err = security_xfrm_policy_delete(pol->security);
 761                         if (err) {
 762                                 xfrm_audit_policy_delete(pol, 0,
 763                                                          audit_info->loginuid,
 764                                                          audit_info->sessionid,
 765                                                          audit_info->secid);
 766                                 return err;
 767                         }
 768                 }
 769                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 770                         hlist_for_each_entry(pol, entry,
 771                                              xfrm_policy_bydst[dir].table + i,
 772                                              bydst) {
 773                                 if (pol->type != type)
 774                                         continue;
 775                                 err = security_xfrm_policy_delete(
 776                                                                 pol->security);
 777                                 if (err) {
 778                                         xfrm_audit_policy_delete(pol, 0,
 779                                                         audit_info->loginuid,
 780                                                         audit_info->sessionid,
 781                                                         audit_info->secid);
 782                                         return err;
 783                                 }
 784                         }
 785                 }
 786         }
 787         return err;
 788 }
 789 #else
 790 static inline int
 791 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
 792 {
 793         return 0;
 794 }
 795 #endif
 796
 797 int xfrm_policy_flush(u8 type, struct xfrm_audit *audit_info)
 798 {
 799         int dir, err = 0;
 800
 801         write_lock_bh(&xfrm_policy_lock);
 802
 803         err = xfrm_policy_flush_secctx_check(type, audit_info);
 804         if (err)
 805                 goto out;
 806
 807         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 808                 struct xfrm_policy *pol;
 809                 struct hlist_node *entry;
 810                 int i, killed;
 811
 812                 killed = 0;
 813         again1:
 814                 hlist_for_each_entry(pol, entry,
 815                                      &xfrm_policy_inexact[dir], bydst) {
 816                         if (pol->type != type)
 817                                 continue;
 818                         hlist_del(&pol->bydst);
 819                         hlist_del(&pol->byidx);
 820                         write_unlock_bh(&xfrm_policy_lock);
 821
 822                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
 823                                                  audit_info->sessionid,
 824                                                  audit_info->secid);
 825
 826                         xfrm_policy_kill(pol);
 827                         killed++;
 828
 829                         write_lock_bh(&xfrm_policy_lock);
 830                         goto again1;
 831                 }
 832
 833                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 834         again2:
 835                         hlist_for_each_entry(pol, entry,
 836                                              xfrm_policy_bydst[dir].table + i,
 837                                              bydst) {
 838                                 if (pol->type != type)
 839                                         continue;
 840                                 hlist_del(&pol->bydst);
 841                                 hlist_del(&pol->byidx);
 842                                 list_del(&pol->walk.all);
 843                                 write_unlock_bh(&xfrm_policy_lock);
 844
 845                                 xfrm_audit_policy_delete(pol, 1,
 846                                                          audit_info->loginuid,
 847                                                          audit_info->sessionid,
 848                                                          audit_info->secid);
 849                                 xfrm_policy_kill(pol);
 850                                 killed++;
 851
 852                                 write_lock_bh(&xfrm_policy_lock);
 853                                 goto again2;
 854                         }
 855                 }
 856
 857                 xfrm_policy_count[dir] -= killed;
 858         }
 859         atomic_inc(&flow_cache_genid);
 860 out:
 861         write_unlock_bh(&xfrm_policy_lock);
 862         return err;
 863 }
 864 EXPORT_SYMBOL(xfrm_policy_flush);
 865
 866 int xfrm_policy_walk(struct xfrm_policy_walk *walk,
 867                      int (*func)(struct xfrm_policy *, int, int, void*),
 868                      void *data)
 869 {
 870         struct xfrm_policy *pol;
 871         struct xfrm_policy_walk_entry *x;
 872         int error = 0;
 873
 874         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
 875             walk->type != XFRM_POLICY_TYPE_ANY)
 876                 return -EINVAL;
 877
 878         if (list_empty(&walk->walk.all) && walk->seq != 0)
 879                 return 0;
 880
 881         write_lock_bh(&xfrm_policy_lock);
 882         if (list_empty(&walk->walk.all))
 883                 x = list_first_entry(&xfrm_policy_all, struct xfrm_policy_walk_entry, all);
 884         else
 885                 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
 886         list_for_each_entry_from(x, &xfrm_policy_all, all) {
 887                 if (x->dead)
 888                         continue;
 889                 pol = container_of(x, struct xfrm_policy, walk);
 890                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
 891                     walk->type != pol->type)
 892                         continue;
 893                 error = func(pol, xfrm_policy_id2dir(pol->index),
 894                              walk->seq, data);
 895                 if (error) {
 896                         list_move_tail(&walk->walk.all, &x->all);
 897                         goto out;
 898                 }
 899                 walk->seq++;
 900         }
 901         if (walk->seq == 0) {
 902                 error = -ENOENT;
 903                 goto out;
 904         }
 905         list_del_init(&walk->walk.all);
 906 out:
 907         write_unlock_bh(&xfrm_policy_lock);
 908         return error;
 909 }
 910 EXPORT_SYMBOL(xfrm_policy_walk);
 911
 912 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
 913 {
 914         INIT_LIST_HEAD(&walk->walk.all);
 915         walk->walk.dead = 1;
 916         walk->type = type;
 917         walk->seq = 0;
 918 }
 919 EXPORT_SYMBOL(xfrm_policy_walk_init);
 920
 921 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
 922 {
 923         if (list_empty(&walk->walk.all))
 924                 return;
 925
 926         write_lock_bh(&xfrm_policy_lock);
 927         list_del(&walk->walk.all);
 928         write_unlock_bh(&xfrm_policy_lock);
 929 }
 930 EXPORT_SYMBOL(xfrm_policy_walk_done);
 931
 932 /*
 933  * Find policy to apply to this flow.
 934  *
 935  * Returns 0 if policy found, else an -errno.
 936  */
 937 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
 938                              u8 type, u16 family, int dir)
 939 {
 940         struct xfrm_selector *sel = &pol->selector;
 941         int match, ret = -ESRCH;
 942
 943         if (pol->family != family ||
 944             pol->type != type)
 945                 return ret;
 946
 947         match = xfrm_selector_match(sel, fl, family);
 948         if (match)
 949                 ret = security_xfrm_policy_lookup(pol->security, fl->secid,
 950                                                   dir);
 951
 952         return ret;
 953 }
 954
 955 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
 956                                                      u16 family, u8 dir)
 957 {
 958         int err;
 959         struct xfrm_policy *pol, *ret;
 960         xfrm_address_t *daddr, *saddr;
 961         struct hlist_node *entry;
 962         struct hlist_head *chain;
 963         u32 priority = ~0U;
 964
 965         daddr = xfrm_flowi_daddr(fl, family);
 966         saddr = xfrm_flowi_saddr(fl, family);
 967         if (unlikely(!daddr || !saddr))
 968                 return NULL;
 969
 970         read_lock_bh(&xfrm_policy_lock);
 971         chain = policy_hash_direct(daddr, saddr, family, dir);
 972         ret = NULL;
 973         hlist_for_each_entry(pol, entry, chain, bydst) {
 974                 err = xfrm_policy_match(pol, fl, type, family, dir);
 975                 if (err) {
 976                         if (err == -ESRCH)
 977                                 continue;
 978                         else {
 979                                 ret = ERR_PTR(err);
 980                                 goto fail;
 981                         }
 982                 } else {
 983                         ret = pol;
 984                         priority = ret->priority;
 985                         break;
 986                 }
 987         }
 988         chain = &xfrm_policy_inexact[dir];
 989         hlist_for_each_entry(pol, entry, chain, bydst) {
 990                 err = xfrm_policy_match(pol, fl, type, family, dir);
 991                 if (err) {
 992                         if (err == -ESRCH)
 993                                 continue;
 994                         else {
 995                                 ret = ERR_PTR(err);
 996                                 goto fail;
 997                         }
 998                 } else if (pol->priority < priority) {
 999                         ret = pol;
1000                         break;
1001                 }
1002         }
1003         if (ret)
1004                 xfrm_pol_hold(ret);
1005 fail:
1006         read_unlock_bh(&xfrm_policy_lock);
1007
1008         return ret;
1009 }
1010
1011 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
1012                                void **objp, atomic_t **obj_refp)
1013 {
1014         struct xfrm_policy *pol;
1015         int err = 0;
1016
1017 #ifdef CONFIG_XFRM_SUB_POLICY
1018         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
1019         if (IS_ERR(pol)) {
1020                 err = PTR_ERR(pol);
1021                 pol = NULL;
1022         }
1023         if (pol || err)
1024                 goto end;
1025 #endif
1026         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1027         if (IS_ERR(pol)) {
1028                 err = PTR_ERR(pol);
1029                 pol = NULL;
1030         }
1031 #ifdef CONFIG_XFRM_SUB_POLICY
1032 end:
1033 #endif
1034         if ((*objp = (void *) pol) != NULL)
1035                 *obj_refp = &pol->refcnt;
1036         return err;
1037 }
1038
1039 static inline int policy_to_flow_dir(int dir)
1040 {
1041         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1042             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1043             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1044                 return dir;
1045         switch (dir) {
1046         default:
1047         case XFRM_POLICY_IN:
1048                 return FLOW_DIR_IN;
1049         case XFRM_POLICY_OUT:
1050                 return FLOW_DIR_OUT;
1051         case XFRM_POLICY_FWD:
1052                 return FLOW_DIR_FWD;
1053         }
1054 }
1055
1056 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1057 {
1058         struct xfrm_policy *pol;
1059
1060         read_lock_bh(&xfrm_policy_lock);
1061         if ((pol = sk->sk_policy[dir]) != NULL) {
1062                 int match = xfrm_selector_match(&pol->selector, fl,
1063                                                 sk->sk_family);
1064                 int err = 0;
1065
1066                 if (match) {
1067                         err = security_xfrm_policy_lookup(pol->security,
1068                                                       fl->secid,
1069                                                       policy_to_flow_dir(dir));
1070                         if (!err)
1071                                 xfrm_pol_hold(pol);
1072                         else if (err == -ESRCH)
1073                                 pol = NULL;
1074                         else
1075                                 pol = ERR_PTR(err);
1076                 } else
1077                         pol = NULL;
1078         }
1079         read_unlock_bh(&xfrm_policy_lock);
1080         return pol;
1081 }
1082
1083 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1084 {
1085         struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1086                                                      pol->family, dir);
1087
1088         list_add(&pol->walk.all, &xfrm_policy_all);
1089         hlist_add_head(&pol->bydst, chain);
1090         hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1091         xfrm_policy_count[dir]++;
1092         xfrm_pol_hold(pol);
1093
1094         if (xfrm_bydst_should_resize(dir, NULL))
1095                 schedule_work(&xfrm_hash_work);
1096 }
1097
1098 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1099                                                 int dir)
1100 {
1101         if (hlist_unhashed(&pol->bydst))
1102                 return NULL;
1103
1104         hlist_del(&pol->bydst);
1105         hlist_del(&pol->byidx);
1106         list_del(&pol->walk.all);
1107         xfrm_policy_count[dir]--;
1108
1109         return pol;
1110 }
1111
1112 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1113 {
1114         write_lock_bh(&xfrm_policy_lock);
1115         pol = __xfrm_policy_unlink(pol, dir);
1116         write_unlock_bh(&xfrm_policy_lock);
1117         if (pol) {
1118                 if (dir < XFRM_POLICY_MAX)
1119                         atomic_inc(&flow_cache_genid);
1120                 xfrm_policy_kill(pol);
1121                 return 0;
1122         }
1123         return -ENOENT;
1124 }
1125 EXPORT_SYMBOL(xfrm_policy_delete);
1126
1127 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1128 {
1129         struct xfrm_policy *old_pol;
1130
1131 #ifdef CONFIG_XFRM_SUB_POLICY
1132         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1133                 return -EINVAL;
1134 #endif
1135
1136         write_lock_bh(&xfrm_policy_lock);
1137         old_pol = sk->sk_policy[dir];
1138         sk->sk_policy[dir] = pol;
1139         if (pol) {
1140                 pol->curlft.add_time = get_seconds();
1141                 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1142                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1143         }
1144         if (old_pol)
1145                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1146         write_unlock_bh(&xfrm_policy_lock);
1147
1148         if (old_pol) {
1149                 xfrm_policy_kill(old_pol);
1150         }
1151         return 0;
1152 }
1153
1154 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1155 {
1156         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1157
1158         if (newp) {
1159                 newp->selector = old->selector;
1160                 if (security_xfrm_policy_clone(old->security,
1161                                                &newp->security)) {
1162                         kfree(newp);
1163                         return NULL;  /* ENOMEM */
1164                 }
1165                 newp->lft = old->lft;
1166                 newp->curlft = old->curlft;
1167                 newp->action = old->action;
1168                 newp->flags = old->flags;
1169                 newp->xfrm_nr = old->xfrm_nr;
1170                 newp->index = old->index;
1171                 newp->type = old->type;
1172                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1173                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1174                 write_lock_bh(&xfrm_policy_lock);
1175                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1176                 write_unlock_bh(&xfrm_policy_lock);
1177                 xfrm_pol_put(newp);
1178         }
1179         return newp;
1180 }
1181
1182 int __xfrm_sk_clone_policy(struct sock *sk)
1183 {
1184         struct xfrm_policy *p0 = sk->sk_policy[0],
1185                            *p1 = sk->sk_policy[1];
1186
1187         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1188         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1189                 return -ENOMEM;
1190         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1191                 return -ENOMEM;
1192         return 0;
1193 }
1194
1195 static int
1196 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1197                unsigned short family)
1198 {
1199         int err;
1200         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1201
1202         if (unlikely(afinfo == NULL))
1203                 return -EINVAL;
1204         err = afinfo->get_saddr(local, remote);
1205         xfrm_policy_put_afinfo(afinfo);
1206         return err;
1207 }
1208
1209 /* Resolve list of templates for the flow, given policy. */
1210
1211 static int
1212 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1213                       struct xfrm_state **xfrm,
1214                       unsigned short family)
1215 {
1216         int nx;
1217         int i, error;
1218         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1219         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1220         xfrm_address_t tmp;
1221
1222         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1223                 struct xfrm_state *x;
1224                 xfrm_address_t *remote = daddr;
1225                 xfrm_address_t *local  = saddr;
1226                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1227
1228                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1229                     tmpl->mode == XFRM_MODE_BEET) {
1230                         remote = &tmpl->id.daddr;
1231                         local = &tmpl->saddr;
1232                         family = tmpl->encap_family;
1233                         if (xfrm_addr_any(local, family)) {
1234                                 error = xfrm_get_saddr(&tmp, remote, family);
1235                                 if (error)
1236                                         goto fail;
1237                                 local = &tmp;
1238                         }
1239                 }
1240
1241                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1242
1243                 if (x && x->km.state == XFRM_STATE_VALID) {
1244                         xfrm[nx++] = x;
1245                         daddr = remote;
1246                         saddr = local;
1247                         continue;
1248                 }
1249                 if (x) {
1250                         error = (x->km.state == XFRM_STATE_ERROR ?
1251                                  -EINVAL : -EAGAIN);
1252                         xfrm_state_put(x);
1253                 }
1254
1255                 if (!tmpl->optional)
1256                         goto fail;
1257         }
1258         return nx;
1259
1260 fail:
1261         for (nx--; nx>=0; nx--)
1262                 xfrm_state_put(xfrm[nx]);
1263         return error;
1264 }
1265
1266 static int
1267 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1268                   struct xfrm_state **xfrm,
1269                   unsigned short family)
1270 {
1271         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1272         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1273         int cnx = 0;
1274         int error;
1275         int ret;
1276         int i;
1277
1278         for (i = 0; i < npols; i++) {
1279                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1280                         error = -ENOBUFS;
1281                         goto fail;
1282                 }
1283
1284                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1285                 if (ret < 0) {
1286                         error = ret;
1287                         goto fail;
1288                 } else
1289                         cnx += ret;
1290         }
1291
1292         /* found states are sorted for outbound processing */
1293         if (npols > 1)
1294                 xfrm_state_sort(xfrm, tpp, cnx, family);
1295
1296         return cnx;
1297
1298  fail:
1299         for (cnx--; cnx>=0; cnx--)
1300                 xfrm_state_put(tpp[cnx]);
1301         return error;
1302
1303 }
1304
1305 /* Check that the bundle accepts the flow and its components are
1306  * still valid.
1307  */
1308
1309 static struct dst_entry *
1310 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1311 {
1312         struct dst_entry *x;
1313         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1314         if (unlikely(afinfo == NULL))
1315                 return ERR_PTR(-EINVAL);
1316         x = afinfo->find_bundle(fl, policy);
1317         xfrm_policy_put_afinfo(afinfo);
1318         return x;
1319 }
1320
1321 static inline int xfrm_get_tos(struct flowi *fl, int family)
1322 {
1323         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1324         int tos;
1325
1326         if (!afinfo)
1327                 return -EINVAL;
1328
1329         tos = afinfo->get_tos(fl);
1330
1331         xfrm_policy_put_afinfo(afinfo);
1332
1333         return tos;
1334 }
1335
1336 static inline struct xfrm_dst *xfrm_alloc_dst(int family)
1337 {
1338         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1339         struct xfrm_dst *xdst;
1340
1341         if (!afinfo)
1342                 return ERR_PTR(-EINVAL);
1343
1344         xdst = dst_alloc(afinfo->dst_ops) ?: ERR_PTR(-ENOBUFS);
1345
1346         xfrm_policy_put_afinfo(afinfo);
1347
1348         return xdst;
1349 }
1350
1351 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1352                                  int nfheader_len)
1353 {
1354         struct xfrm_policy_afinfo *afinfo =
1355                 xfrm_policy_get_afinfo(dst->ops->family);
1356         int err;
1357
1358         if (!afinfo)
1359                 return -EINVAL;
1360
1361         err = afinfo->init_path(path, dst, nfheader_len);
1362
1363         xfrm_policy_put_afinfo(afinfo);
1364
1365         return err;
1366 }
1367
1368 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev)
1369 {
1370         struct xfrm_policy_afinfo *afinfo =
1371                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1372         int err;
1373
1374         if (!afinfo)
1375                 return -EINVAL;
1376
1377         err = afinfo->fill_dst(xdst, dev);
1378
1379         xfrm_policy_put_afinfo(afinfo);
1380
1381         return err;
1382 }
1383
1384 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1385  * all the metrics... Shortly, bundle a bundle.
1386  */
1387
1388 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1389                                             struct xfrm_state **xfrm, int nx,
1390                                             struct flowi *fl,
1391                                             struct dst_entry *dst)
1392 {
1393         unsigned long now = jiffies;
1394         struct net_device *dev;
1395         struct dst_entry *dst_prev = NULL;
1396         struct dst_entry *dst0 = NULL;
1397         int i = 0;
1398         int err;
1399         int header_len = 0;
1400         int nfheader_len = 0;
1401         int trailer_len = 0;
1402         int tos;
1403         int family = policy->selector.family;
1404         xfrm_address_t saddr, daddr;
1405
1406         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1407
1408         tos = xfrm_get_tos(fl, family);
1409         err = tos;
1410         if (tos < 0)
1411                 goto put_states;
1412
1413         dst_hold(dst);
1414
1415         for (; i < nx; i++) {
1416                 struct xfrm_dst *xdst = xfrm_alloc_dst(family);
1417                 struct dst_entry *dst1 = &xdst->u.dst;
1418
1419                 err = PTR_ERR(xdst);
1420                 if (IS_ERR(xdst)) {
1421                         dst_release(dst);
1422                         goto put_states;
1423                 }
1424
1425                 if (!dst_prev)
1426                         dst0 = dst1;
1427                 else {
1428                         dst_prev->child = dst_clone(dst1);
1429                         dst1->flags |= DST_NOHASH;
1430                 }
1431
1432                 xdst->route = dst;
1433                 memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1434
1435                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1436                         family = xfrm[i]->props.family;
1437                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1438                                               family);
1439                         err = PTR_ERR(dst);
1440                         if (IS_ERR(dst))
1441                                 goto put_states;
1442                 } else
1443                         dst_hold(dst);
1444
1445                 dst1->xfrm = xfrm[i];
1446                 xdst->genid = xfrm[i]->genid;
1447
1448                 dst1->obsolete = -1;
1449                 dst1->flags |= DST_HOST;
1450                 dst1->lastuse = now;
1451
1452                 dst1->input = dst_discard;
1453                 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1454
1455                 dst1->next = dst_prev;
1456                 dst_prev = dst1;
1457
1458                 header_len += xfrm[i]->props.header_len;
1459                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1460                         nfheader_len += xfrm[i]->props.header_len;
1461                 trailer_len += xfrm[i]->props.trailer_len;
1462         }
1463
1464         dst_prev->child = dst;
1465         dst0->path = dst;
1466
1467         err = -ENODEV;
1468         dev = dst->dev;
1469         if (!dev)
1470                 goto free_dst;
1471
1472         /* Copy neighbout for reachability confirmation */
1473         dst0->neighbour = neigh_clone(dst->neighbour);
1474
1475         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1476         xfrm_init_pmtu(dst_prev);
1477
1478         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1479                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1480
1481                 err = xfrm_fill_dst(xdst, dev);
1482                 if (err)
1483                         goto free_dst;
1484
1485                 dst_prev->header_len = header_len;
1486                 dst_prev->trailer_len = trailer_len;
1487                 header_len -= xdst->u.dst.xfrm->props.header_len;
1488                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1489         }
1490
1491 out:
1492         return dst0;
1493
1494 put_states:
1495         for (; i < nx; i++)
1496                 xfrm_state_put(xfrm[i]);
1497 free_dst:
1498         if (dst0)
1499                 dst_free(dst0);
1500         dst0 = ERR_PTR(err);
1501         goto out;
1502 }
1503
1504 static int inline
1505 xfrm_dst_alloc_copy(void **target, void *src, int size)
1506 {
1507         if (!*target) {
1508                 *target = kmalloc(size, GFP_ATOMIC);
1509                 if (!*target)
1510                         return -ENOMEM;
1511         }
1512         memcpy(*target, src, size);
1513         return 0;
1514 }
1515
1516 static int inline
1517 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1518 {
1519 #ifdef CONFIG_XFRM_SUB_POLICY
1520         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1521         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1522                                    sel, sizeof(*sel));
1523 #else
1524         return 0;
1525 #endif
1526 }
1527
1528 static int inline
1529 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1530 {
1531 #ifdef CONFIG_XFRM_SUB_POLICY
1532         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1533         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1534 #else
1535         return 0;
1536 #endif
1537 }
1538
1539 static int stale_bundle(struct dst_entry *dst);
1540
1541 /* Main function: finds/creates a bundle for given flow.
1542  *
1543  * At the moment we eat a raw IP route. Mostly to speed up lookups
1544  * on interfaces with disabled IPsec.
1545  */
1546 int __xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1547                   struct sock *sk, int flags)
1548 {
1549         struct xfrm_policy *policy;
1550         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1551         int npols;
1552         int pol_dead;
1553         int xfrm_nr;
1554         int pi;
1555         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1556         struct dst_entry *dst, *dst_orig = *dst_p;
1557         int nx = 0;
1558         int err;
1559         u32 genid;
1560         u16 family;
1561         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1562
1563 restart:
1564         genid = atomic_read(&flow_cache_genid);
1565         policy = NULL;
1566         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1567                 pols[pi] = NULL;
1568         npols = 0;
1569         pol_dead = 0;
1570         xfrm_nr = 0;
1571
1572         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1573                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1574                 err = PTR_ERR(policy);
1575                 if (IS_ERR(policy)) {
1576                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1577                         goto dropdst;
1578                 }
1579         }
1580
1581         if (!policy) {
1582                 /* To accelerate a bit...  */
1583                 if ((dst_orig->flags & DST_NOXFRM) ||
1584                     !xfrm_policy_count[XFRM_POLICY_OUT])
1585                         goto nopol;
1586
1587                 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1588                                            dir, xfrm_policy_lookup);
1589                 err = PTR_ERR(policy);
1590                 if (IS_ERR(policy)) {
1591                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1592                         goto dropdst;
1593                 }
1594         }
1595
1596         if (!policy)
1597                 goto nopol;
1598
1599         family = dst_orig->ops->family;
1600         pols[0] = policy;
1601         npols ++;
1602         xfrm_nr += pols[0]->xfrm_nr;
1603
1604         err = -ENOENT;
1605         if ((flags & XFRM_LOOKUP_ICMP) && !(policy->flags & XFRM_POLICY_ICMP))
1606                 goto error;
1607
1608         policy->curlft.use_time = get_seconds();
1609
1610         switch (policy->action) {
1611         default:
1612         case XFRM_POLICY_BLOCK:
1613                 /* Prohibit the flow */
1614                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1615                 err = -EPERM;
1616                 goto error;
1617
1618         case XFRM_POLICY_ALLOW:
1619 #ifndef CONFIG_XFRM_SUB_POLICY
1620                 if (policy->xfrm_nr == 0) {
1621                         /* Flow passes not transformed. */
1622                         xfrm_pol_put(policy);
1623                         return 0;
1624                 }
1625 #endif
1626
1627                 /* Try to find matching bundle.
1628                  *
1629                  * LATER: help from flow cache. It is optional, this
1630                  * is required only for output policy.
1631                  */
1632                 dst = xfrm_find_bundle(fl, policy, family);
1633                 if (IS_ERR(dst)) {
1634                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1635                         err = PTR_ERR(dst);
1636                         goto error;
1637                 }
1638
1639                 if (dst)
1640                         break;
1641
1642 #ifdef CONFIG_XFRM_SUB_POLICY
1643                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1644                         pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1645                                                             fl, family,
1646                                                             XFRM_POLICY_OUT);
1647                         if (pols[1]) {
1648                                 if (IS_ERR(pols[1])) {
1649                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1650                                         err = PTR_ERR(pols[1]);
1651                                         goto error;
1652                                 }
1653                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1654                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1655                                         err = -EPERM;
1656                                         goto error;
1657                                 }
1658                                 npols ++;
1659                                 xfrm_nr += pols[1]->xfrm_nr;
1660                         }
1661                 }
1662
1663                 /*
1664                  * Because neither flowi nor bundle information knows about
1665                  * transformation template size. On more than one policy usage
1666                  * we can realize whether all of them is bypass or not after
1667                  * they are searched. See above not-transformed bypass
1668                  * is surrounded by non-sub policy configuration, too.
1669                  */
1670                 if (xfrm_nr == 0) {
1671                         /* Flow passes not transformed. */
1672                         xfrm_pols_put(pols, npols);
1673                         return 0;
1674                 }
1675
1676 #endif
1677                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1678
1679                 if (unlikely(nx<0)) {
1680                         err = nx;
1681                         if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
1682                                 /* EREMOTE tells the caller to generate
1683                                  * a one-shot blackhole route.
1684                                  */
1685                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1686                                 xfrm_pol_put(policy);
1687                                 return -EREMOTE;
1688                         }
1689                         if (err == -EAGAIN && (flags & XFRM_LOOKUP_WAIT)) {
1690                                 DECLARE_WAITQUEUE(wait, current);
1691
1692                                 add_wait_queue(&km_waitq, &wait);
1693                                 set_current_state(TASK_INTERRUPTIBLE);
1694                                 schedule();
1695                                 set_current_state(TASK_RUNNING);
1696                                 remove_wait_queue(&km_waitq, &wait);
1697
1698                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1699
1700                                 if (nx == -EAGAIN && signal_pending(current)) {
1701                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1702                                         err = -ERESTART;
1703                                         goto error;
1704                                 }
1705                                 if (nx == -EAGAIN ||
1706                                     genid != atomic_read(&flow_cache_genid)) {
1707                                         xfrm_pols_put(pols, npols);
1708                                         goto restart;
1709                                 }
1710                                 err = nx;
1711                         }
1712                         if (err < 0) {
1713                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1714                                 goto error;
1715                         }
1716                 }
1717                 if (nx == 0) {
1718                         /* Flow passes not transformed. */
1719                         xfrm_pols_put(pols, npols);
1720                         return 0;
1721                 }
1722
1723                 dst = xfrm_bundle_create(policy, xfrm, nx, fl, dst_orig);
1724                 err = PTR_ERR(dst);
1725                 if (IS_ERR(dst)) {
1726                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1727                         goto error;
1728                 }
1729
1730                 for (pi = 0; pi < npols; pi++) {
1731                         read_lock_bh(&pols[pi]->lock);
1732                         pol_dead |= pols[pi]->walk.dead;
1733                         read_unlock_bh(&pols[pi]->lock);
1734                 }
1735
1736                 write_lock_bh(&policy->lock);
1737                 if (unlikely(pol_dead || stale_bundle(dst))) {
1738                         /* Wow! While we worked on resolving, this
1739                          * policy has gone. Retry. It is not paranoia,
1740                          * we just cannot enlist new bundle to dead object.
1741                          * We can't enlist stable bundles either.
1742                          */
1743                         write_unlock_bh(&policy->lock);
1744                         dst_free(dst);
1745
1746                         if (pol_dead)
1747                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLDEAD);
1748                         else
1749                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1750                         err = -EHOSTUNREACH;
1751                         goto error;
1752                 }
1753
1754                 if (npols > 1)
1755                         err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1756                 else
1757                         err = xfrm_dst_update_origin(dst, fl);
1758                 if (unlikely(err)) {
1759                         write_unlock_bh(&policy->lock);
1760                         dst_free(dst);
1761                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1762                         goto error;
1763                 }
1764
1765                 dst->next = policy->bundles;
1766                 policy->bundles = dst;
1767                 dst_hold(dst);
1768                 write_unlock_bh(&policy->lock);
1769         }
1770         *dst_p = dst;
1771         dst_release(dst_orig);
1772         xfrm_pols_put(pols, npols);
1773         return 0;
1774
1775 error:
1776         xfrm_pols_put(pols, npols);
1777 dropdst:
1778         dst_release(dst_orig);
1779         *dst_p = NULL;
1780         return err;
1781
1782 nopol:
1783         err = -ENOENT;
1784         if (flags & XFRM_LOOKUP_ICMP)
1785                 goto dropdst;
1786         return 0;
1787 }
1788 EXPORT_SYMBOL(__xfrm_lookup);
1789
1790 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1791                 struct sock *sk, int flags)
1792 {
1793         int err = __xfrm_lookup(dst_p, fl, sk, flags);
1794
1795         if (err == -EREMOTE) {
1796                 dst_release(*dst_p);
1797                 *dst_p = NULL;
1798                 err = -EAGAIN;
1799         }
1800
1801         return err;
1802 }
1803 EXPORT_SYMBOL(xfrm_lookup);
1804
1805 static inline int
1806 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1807 {
1808         struct xfrm_state *x;
1809
1810         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1811                 return 0;
1812         x = skb->sp->xvec[idx];
1813         if (!x->type->reject)
1814                 return 0;
1815         return x->type->reject(x, skb, fl);
1816 }
1817
1818 /* When skb is transformed back to its "native" form, we have to
1819  * check policy restrictions. At the moment we make this in maximally
1820  * stupid way. Shame on me. :-) Of course, connected sockets must
1821  * have policy cached at them.
1822  */
1823
1824 static inline int
1825 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1826               unsigned short family)
1827 {
1828         if (xfrm_state_kern(x))
1829                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1830         return  x->id.proto == tmpl->id.proto &&
1831                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1832                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1833                 x->props.mode == tmpl->mode &&
1834                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1835                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1836                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1837                   xfrm_state_addr_cmp(tmpl, x, family));
1838 }
1839
1840 /*
1841  * 0 or more than 0 is returned when validation is succeeded (either bypass
1842  * because of optional transport mode, or next index of the mathced secpath
1843  * state with the template.
1844  * -1 is returned when no matching template is found.
1845  * Otherwise "-2 - errored_index" is returned.
1846  */
1847 static inline int
1848 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1849                unsigned short family)
1850 {
1851         int idx = start;
1852
1853         if (tmpl->optional) {
1854                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1855                         return start;
1856         } else
1857                 start = -1;
1858         for (; idx < sp->len; idx++) {
1859                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1860                         return ++idx;
1861                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1862                         if (start == -1)
1863                                 start = -2-idx;
1864                         break;
1865                 }
1866         }
1867         return start;
1868 }
1869
1870 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1871                           unsigned int family, int reverse)
1872 {
1873         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1874         int err;
1875
1876         if (unlikely(afinfo == NULL))
1877                 return -EAFNOSUPPORT;
1878
1879         afinfo->decode_session(skb, fl, reverse);
1880         err = security_xfrm_decode_session(skb, &fl->secid);
1881         xfrm_policy_put_afinfo(afinfo);
1882         return err;
1883 }
1884 EXPORT_SYMBOL(__xfrm_decode_session);
1885
1886 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1887 {
1888         for (; k < sp->len; k++) {
1889                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1890                         *idxp = k;
1891                         return 1;
1892                 }
1893         }
1894
1895         return 0;
1896 }
1897
1898 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1899                         unsigned short family)
1900 {
1901         struct xfrm_policy *pol;
1902         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1903         int npols = 0;
1904         int xfrm_nr;
1905         int pi;
1906         int reverse;
1907         struct flowi fl;
1908         u8 fl_dir;
1909         int xerr_idx = -1;
1910
1911         reverse = dir & ~XFRM_POLICY_MASK;
1912         dir &= XFRM_POLICY_MASK;
1913         fl_dir = policy_to_flow_dir(dir);
1914
1915         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
1916                 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
1917                 return 0;
1918         }
1919
1920         nf_nat_decode_session(skb, &fl, family);
1921
1922         /* First, check used SA against their selectors. */
1923         if (skb->sp) {
1924                 int i;
1925
1926                 for (i=skb->sp->len-1; i>=0; i--) {
1927                         struct xfrm_state *x = skb->sp->xvec[i];
1928                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
1929                                 XFRM_INC_STATS(LINUX_MIB_XFRMINSTATEMISMATCH);
1930                                 return 0;
1931                         }
1932                 }
1933         }
1934
1935         pol = NULL;
1936         if (sk && sk->sk_policy[dir]) {
1937                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1938                 if (IS_ERR(pol)) {
1939                         XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1940                         return 0;
1941                 }
1942         }
1943
1944         if (!pol)
1945                 pol = flow_cache_lookup(&fl, family, fl_dir,
1946                                         xfrm_policy_lookup);
1947
1948         if (IS_ERR(pol)) {
1949                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1950                 return 0;
1951         }
1952
1953         if (!pol) {
1954                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1955                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1956                         XFRM_INC_STATS(LINUX_MIB_XFRMINNOPOLS);
1957                         return 0;
1958                 }
1959                 return 1;
1960         }
1961
1962         pol->curlft.use_time = get_seconds();
1963
1964         pols[0] = pol;
1965         npols ++;
1966 #ifdef CONFIG_XFRM_SUB_POLICY
1967         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1968                 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1969                                                     &fl, family,
1970                                                     XFRM_POLICY_IN);
1971                 if (pols[1]) {
1972                         if (IS_ERR(pols[1])) {
1973                                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1974                                 return 0;
1975                         }
1976                         pols[1]->curlft.use_time = get_seconds();
1977                         npols ++;
1978                 }
1979         }
1980 #endif
1981
1982         if (pol->action == XFRM_POLICY_ALLOW) {
1983                 struct sec_path *sp;
1984                 static struct sec_path dummy;
1985                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1986                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1987                 struct xfrm_tmpl **tpp = tp;
1988                 int ti = 0;
1989                 int i, k;
1990
1991                 if ((sp = skb->sp) == NULL)
1992                         sp = &dummy;
1993
1994                 for (pi = 0; pi < npols; pi++) {
1995                         if (pols[pi] != pol &&
1996                             pols[pi]->action != XFRM_POLICY_ALLOW) {
1997                                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
1998                                 goto reject;
1999                         }
2000                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2001                                 XFRM_INC_STATS(LINUX_MIB_XFRMINBUFFERERROR);
2002                                 goto reject_error;
2003                         }
2004                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2005                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2006                 }
2007                 xfrm_nr = ti;
2008                 if (npols > 1) {
2009                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2010                         tpp = stp;
2011                 }
2012
2013                 /* For each tunnel xfrm, find the first matching tmpl.
2014                  * For each tmpl before that, find corresponding xfrm.
2015                  * Order is _important_. Later we will implement
2016                  * some barriers, but at the moment barriers
2017                  * are implied between each two transformations.
2018                  */
2019                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2020                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2021                         if (k < 0) {
2022                                 if (k < -1)
2023                                         /* "-2 - errored_index" returned */
2024                                         xerr_idx = -(2+k);
2025                                 XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
2026                                 goto reject;
2027                         }
2028                 }
2029
2030                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2031                         XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
2032                         goto reject;
2033                 }
2034
2035                 xfrm_pols_put(pols, npols);
2036                 return 1;
2037         }
2038         XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
2039
2040 reject:
2041         xfrm_secpath_reject(xerr_idx, skb, &fl);
2042 reject_error:
2043         xfrm_pols_put(pols, npols);
2044         return 0;
2045 }
2046 EXPORT_SYMBOL(__xfrm_policy_check);
2047
2048 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2049 {
2050         struct flowi fl;
2051
2052         if (xfrm_decode_session(skb, &fl, family) < 0) {
2053                 /* XXX: we should have something like FWDHDRERROR here. */
2054                 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
2055                 return 0;
2056         }
2057
2058         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
2059 }
2060 EXPORT_SYMBOL(__xfrm_route_forward);
2061
2062 /* Optimize later using cookies and generation ids. */
2063
2064 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2065 {
2066         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2067          * to "-1" to force all XFRM destinations to get validated by
2068          * dst_ops->check on every use.  We do this because when a
2069          * normal route referenced by an XFRM dst is obsoleted we do
2070          * not go looking around for all parent referencing XFRM dsts
2071          * so that we can invalidate them.  It is just too much work.
2072          * Instead we make the checks here on every use.  For example:
2073          *
2074          *      XFRM dst A --> IPv4 dst X
2075          *
2076          * X is the "xdst->route" of A (X is also the "dst->path" of A
2077          * in this example).  If X is marked obsolete, "A" will not
2078          * notice.  That's what we are validating here via the
2079          * stale_bundle() check.
2080          *
2081          * When a policy's bundle is pruned, we dst_free() the XFRM
2082          * dst which causes it's ->obsolete field to be set to a
2083          * positive non-zero integer.  If an XFRM dst has been pruned
2084          * like this, we want to force a new route lookup.
2085          */
2086         if (dst->obsolete < 0 && !stale_bundle(dst))
2087                 return dst;
2088
2089         return NULL;
2090 }
2091
2092 static int stale_bundle(struct dst_entry *dst)
2093 {
2094         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
2095 }
2096
2097 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2098 {
2099         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2100                 dst->dev = dev_net(dev)->loopback_dev;
2101                 dev_hold(dst->dev);
2102                 dev_put(dev);
2103         }
2104 }
2105 EXPORT_SYMBOL(xfrm_dst_ifdown);
2106
2107 static void xfrm_link_failure(struct sk_buff *skb)
2108 {
2109         /* Impossible. Such dst must be popped before reaches point of failure. */
2110         return;
2111 }
2112
2113 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2114 {
2115         if (dst) {
2116                 if (dst->obsolete) {
2117                         dst_release(dst);
2118                         dst = NULL;
2119                 }
2120         }
2121         return dst;
2122 }
2123
2124 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
2125 {
2126         struct dst_entry *dst, **dstp;
2127
2128         write_lock(&pol->lock);
2129         dstp = &pol->bundles;
2130         while ((dst=*dstp) != NULL) {
2131                 if (func(dst)) {
2132                         *dstp = dst->next;
2133                         dst->next = *gc_list_p;
2134                         *gc_list_p = dst;
2135                 } else {
2136                         dstp = &dst->next;
2137                 }
2138         }
2139         write_unlock(&pol->lock);
2140 }
2141
2142 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
2143 {
2144         struct dst_entry *gc_list = NULL;
2145         int dir;
2146
2147         read_lock_bh(&xfrm_policy_lock);
2148         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2149                 struct xfrm_policy *pol;
2150                 struct hlist_node *entry;
2151                 struct hlist_head *table;
2152                 int i;
2153
2154                 hlist_for_each_entry(pol, entry,
2155                                      &xfrm_policy_inexact[dir], bydst)
2156                         prune_one_bundle(pol, func, &gc_list);
2157
2158                 table = xfrm_policy_bydst[dir].table;
2159                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
2160                         hlist_for_each_entry(pol, entry, table + i, bydst)
2161                                 prune_one_bundle(pol, func, &gc_list);
2162                 }
2163         }
2164         read_unlock_bh(&xfrm_policy_lock);
2165
2166         while (gc_list) {
2167                 struct dst_entry *dst = gc_list;
2168                 gc_list = dst->next;
2169                 dst_free(dst);
2170         }
2171 }
2172
2173 static int unused_bundle(struct dst_entry *dst)
2174 {
2175         return !atomic_read(&dst->__refcnt);
2176 }
2177
2178 static void __xfrm_garbage_collect(void)
2179 {
2180         xfrm_prune_bundles(unused_bundle);
2181 }
2182
2183 static int xfrm_flush_bundles(void)
2184 {
2185         xfrm_prune_bundles(stale_bundle);
2186         return 0;
2187 }
2188
2189 static void xfrm_init_pmtu(struct dst_entry *dst)
2190 {
2191         do {
2192                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2193                 u32 pmtu, route_mtu_cached;
2194
2195                 pmtu = dst_mtu(dst->child);
2196                 xdst->child_mtu_cached = pmtu;
2197
2198                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2199
2200                 route_mtu_cached = dst_mtu(xdst->route);
2201                 xdst->route_mtu_cached = route_mtu_cached;
2202
2203                 if (pmtu > route_mtu_cached)
2204                         pmtu = route_mtu_cached;
2205
2206                 dst->metrics[RTAX_MTU-1] = pmtu;
2207         } while ((dst = dst->next));
2208 }
2209
2210 /* Check that the bundle accepts the flow and its components are
2211  * still valid.
2212  */
2213
2214 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2215                 struct flowi *fl, int family, int strict)
2216 {
2217         struct dst_entry *dst = &first->u.dst;
2218         struct xfrm_dst *last;
2219         u32 mtu;
2220
2221         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2222             (dst->dev && !netif_running(dst->dev)))
2223                 return 0;
2224 #ifdef CONFIG_XFRM_SUB_POLICY
2225         if (fl) {
2226                 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2227                         return 0;
2228                 if (first->partner &&
2229                     !xfrm_selector_match(first->partner, fl, family))
2230                         return 0;
2231         }
2232 #endif
2233
2234         last = NULL;
2235
2236         do {
2237                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2238
2239                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2240                         return 0;
2241                 if (fl && pol &&
2242                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2243                         return 0;
2244                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2245                         return 0;
2246                 if (xdst->genid != dst->xfrm->genid)
2247                         return 0;
2248
2249                 if (strict && fl &&
2250                     !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2251                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2252                         return 0;
2253
2254                 mtu = dst_mtu(dst->child);
2255                 if (xdst->child_mtu_cached != mtu) {
2256                         last = xdst;
2257                         xdst->child_mtu_cached = mtu;
2258                 }
2259
2260                 if (!dst_check(xdst->route, xdst->route_cookie))
2261                         return 0;
2262                 mtu = dst_mtu(xdst->route);
2263                 if (xdst->route_mtu_cached != mtu) {
2264                         last = xdst;
2265                         xdst->route_mtu_cached = mtu;
2266                 }
2267
2268                 dst = dst->child;
2269         } while (dst->xfrm);
2270
2271         if (likely(!last))
2272                 return 1;
2273
2274         mtu = last->child_mtu_cached;
2275         for (;;) {
2276                 dst = &last->u.dst;
2277
2278                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2279                 if (mtu > last->route_mtu_cached)
2280                         mtu = last->route_mtu_cached;
2281                 dst->metrics[RTAX_MTU-1] = mtu;
2282
2283                 if (last == first)
2284                         break;
2285
2286                 last = (struct xfrm_dst *)last->u.dst.next;
2287                 last->child_mtu_cached = mtu;
2288         }
2289
2290         return 1;
2291 }
2292
2293 EXPORT_SYMBOL(xfrm_bundle_ok);
2294
2295 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2296 {
2297         int err = 0;
2298         if (unlikely(afinfo == NULL))
2299                 return -EINVAL;
2300         if (unlikely(afinfo->family >= NPROTO))
2301                 return -EAFNOSUPPORT;
2302         write_lock_bh(&xfrm_policy_afinfo_lock);
2303         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2304                 err = -ENOBUFS;
2305         else {
2306                 struct dst_ops *dst_ops = afinfo->dst_ops;
2307                 if (likely(dst_ops->kmem_cachep == NULL))
2308                         dst_ops->kmem_cachep = xfrm_dst_cache;
2309                 if (likely(dst_ops->check == NULL))
2310                         dst_ops->check = xfrm_dst_check;
2311                 if (likely(dst_ops->negative_advice == NULL))
2312                         dst_ops->negative_advice = xfrm_negative_advice;
2313                 if (likely(dst_ops->link_failure == NULL))
2314                         dst_ops->link_failure = xfrm_link_failure;
2315                 if (likely(afinfo->garbage_collect == NULL))
2316                         afinfo->garbage_collect = __xfrm_garbage_collect;
2317                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2318         }
2319         write_unlock_bh(&xfrm_policy_afinfo_lock);
2320         return err;
2321 }
2322 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2323
2324 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2325 {
2326         int err = 0;
2327         if (unlikely(afinfo == NULL))
2328                 return -EINVAL;
2329         if (unlikely(afinfo->family >= NPROTO))
2330                 return -EAFNOSUPPORT;
2331         write_lock_bh(&xfrm_policy_afinfo_lock);
2332         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2333                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2334                         err = -EINVAL;
2335                 else {
2336                         struct dst_ops *dst_ops = afinfo->dst_ops;
2337                         xfrm_policy_afinfo[afinfo->family] = NULL;
2338                         dst_ops->kmem_cachep = NULL;
2339                         dst_ops->check = NULL;
2340                         dst_ops->negative_advice = NULL;
2341                         dst_ops->link_failure = NULL;
2342                         afinfo->garbage_collect = NULL;
2343                 }
2344         }
2345         write_unlock_bh(&xfrm_policy_afinfo_lock);
2346         return err;
2347 }
2348 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2349
2350 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2351 {
2352         struct xfrm_policy_afinfo *afinfo;
2353         if (unlikely(family >= NPROTO))
2354                 return NULL;
2355         read_lock(&xfrm_policy_afinfo_lock);
2356         afinfo = xfrm_policy_afinfo[family];
2357         if (unlikely(!afinfo))
2358                 read_unlock(&xfrm_policy_afinfo_lock);
2359         return afinfo;
2360 }
2361
2362 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2363 {
2364         read_unlock(&xfrm_policy_afinfo_lock);
2365 }
2366
2367 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2368 {
2369         struct net_device *dev = ptr;
2370
2371         if (!net_eq(dev_net(dev), &init_net))
2372                 return NOTIFY_DONE;
2373
2374         switch (event) {
2375         case NETDEV_DOWN:
2376                 xfrm_flush_bundles();
2377         }
2378         return NOTIFY_DONE;
2379 }
2380
2381 static struct notifier_block xfrm_dev_notifier = {
2382         xfrm_dev_event,
2383         NULL,
2384         0
2385 };
2386
2387 #ifdef CONFIG_XFRM_STATISTICS
2388 static int __init xfrm_statistics_init(void)
2389 {
2390         if (snmp_mib_init((void **)xfrm_statistics,
2391                           sizeof(struct linux_xfrm_mib)) < 0)
2392                 return -ENOMEM;
2393         return 0;
2394 }
2395 #endif
2396
2397 static void __init xfrm_policy_init(void)
2398 {
2399         unsigned int hmask, sz;
2400         int dir;
2401
2402         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2403                                            sizeof(struct xfrm_dst),
2404                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2405                                            NULL);
2406
2407         hmask = 8 - 1;
2408         sz = (hmask+1) * sizeof(struct hlist_head);
2409
2410         xfrm_policy_byidx = xfrm_hash_alloc(sz);
2411         xfrm_idx_hmask = hmask;
2412         if (!xfrm_policy_byidx)
2413                 panic("XFRM: failed to allocate byidx hash\n");
2414
2415         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2416                 struct xfrm_policy_hash *htab;
2417
2418                 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2419
2420                 htab = &xfrm_policy_bydst[dir];
2421                 htab->table = xfrm_hash_alloc(sz);
2422                 htab->hmask = hmask;
2423                 if (!htab->table)
2424                         panic("XFRM: failed to allocate bydst hash\n");
2425         }
2426
2427         INIT_LIST_HEAD(&xfrm_policy_all);
2428         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task);
2429         register_netdevice_notifier(&xfrm_dev_notifier);
2430 }
2431
2432 void __init xfrm_init(void)
2433 {
2434 #ifdef CONFIG_XFRM_STATISTICS
2435         xfrm_statistics_init();
2436 #endif
2437         xfrm_state_init();
2438         xfrm_policy_init();
2439         xfrm_input_init();
2440 #ifdef CONFIG_XFRM_STATISTICS
2441         xfrm_proc_init();
2442 #endif
2443 }
2444
2445 #ifdef CONFIG_AUDITSYSCALL
2446 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2447                                          struct audit_buffer *audit_buf)
2448 {
2449         struct xfrm_sec_ctx *ctx = xp->security;
2450         struct xfrm_selector *sel = &xp->selector;
2451
2452         if (ctx)
2453                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2454                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2455
2456         switch(sel->family) {
2457         case AF_INET:
2458                 audit_log_format(audit_buf, " src=" NIPQUAD_FMT,
2459                                  NIPQUAD(sel->saddr.a4));
2460                 if (sel->prefixlen_s != 32)
2461                         audit_log_format(audit_buf, " src_prefixlen=%d",
2462                                          sel->prefixlen_s);
2463                 audit_log_format(audit_buf, " dst=" NIPQUAD_FMT,
2464                                  NIPQUAD(sel->daddr.a4));
2465                 if (sel->prefixlen_d != 32)
2466                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2467                                          sel->prefixlen_d);
2468                 break;
2469         case AF_INET6:
2470                 audit_log_format(audit_buf, " src=" NIP6_FMT,
2471                                  NIP6(*(struct in6_addr *)sel->saddr.a6));
2472                 if (sel->prefixlen_s != 128)
2473                         audit_log_format(audit_buf, " src_prefixlen=%d",
2474                                          sel->prefixlen_s);
2475                 audit_log_format(audit_buf, " dst=" NIP6_FMT,
2476                                  NIP6(*(struct in6_addr *)sel->daddr.a6));
2477                 if (sel->prefixlen_d != 128)
2478                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2479                                          sel->prefixlen_d);
2480                 break;
2481         }
2482 }
2483
2484 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2485                            uid_t auid, u32 sessionid, u32 secid)
2486 {
2487         struct audit_buffer *audit_buf;
2488
2489         audit_buf = xfrm_audit_start("SPD-add");
2490         if (audit_buf == NULL)
2491                 return;
2492         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2493         audit_log_format(audit_buf, " res=%u", result);
2494         xfrm_audit_common_policyinfo(xp, audit_buf);
2495         audit_log_end(audit_buf);
2496 }
2497 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2498
2499 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2500                               uid_t auid, u32 sessionid, u32 secid)
2501 {
2502         struct audit_buffer *audit_buf;
2503
2504         audit_buf = xfrm_audit_start("SPD-delete");
2505         if (audit_buf == NULL)
2506                 return;
2507         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2508         audit_log_format(audit_buf, " res=%u", result);
2509         xfrm_audit_common_policyinfo(xp, audit_buf);
2510         audit_log_end(audit_buf);
2511 }
2512 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2513 #endif
2514
2515 #ifdef CONFIG_XFRM_MIGRATE
2516 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2517                                        struct xfrm_selector *sel_tgt)
2518 {
2519         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2520                 if (sel_tgt->family == sel_cmp->family &&
2521                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2522                                   sel_cmp->family) == 0 &&
2523                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2524                                   sel_cmp->family) == 0 &&
2525                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2526                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2527                         return 1;
2528                 }
2529         } else {
2530                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2531                         return 1;
2532                 }
2533         }
2534         return 0;
2535 }
2536
2537 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2538                                                      u8 dir, u8 type)
2539 {
2540         struct xfrm_policy *pol, *ret = NULL;
2541         struct hlist_node *entry;
2542         struct hlist_head *chain;
2543         u32 priority = ~0U;
2544
2545         read_lock_bh(&xfrm_policy_lock);
2546         chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2547         hlist_for_each_entry(pol, entry, chain, bydst) {
2548                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2549                     pol->type == type) {
2550                         ret = pol;
2551                         priority = ret->priority;
2552                         break;
2553                 }
2554         }
2555         chain = &xfrm_policy_inexact[dir];
2556         hlist_for_each_entry(pol, entry, chain, bydst) {
2557                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2558                     pol->type == type &&
2559                     pol->priority < priority) {
2560                         ret = pol;
2561                         break;
2562                 }
2563         }
2564
2565         if (ret)
2566                 xfrm_pol_hold(ret);
2567
2568         read_unlock_bh(&xfrm_policy_lock);
2569
2570         return ret;
2571 }
2572
2573 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2574 {
2575         int match = 0;
2576
2577         if (t->mode == m->mode && t->id.proto == m->proto &&
2578             (m->reqid == 0 || t->reqid == m->reqid)) {
2579                 switch (t->mode) {
2580                 case XFRM_MODE_TUNNEL:
2581                 case XFRM_MODE_BEET:
2582                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2583                                           m->old_family) == 0 &&
2584                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2585                                           m->old_family) == 0) {
2586                                 match = 1;
2587                         }
2588                         break;
2589                 case XFRM_MODE_TRANSPORT:
2590                         /* in case of transport mode, template does not store
2591                            any IP addresses, hence we just compare mode and
2592                            protocol */
2593                         match = 1;
2594                         break;
2595                 default:
2596                         break;
2597                 }
2598         }
2599         return match;
2600 }
2601
2602 /* update endpoint address(es) of template(s) */
2603 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2604                                struct xfrm_migrate *m, int num_migrate)
2605 {
2606         struct xfrm_migrate *mp;
2607         struct dst_entry *dst;
2608         int i, j, n = 0;
2609
2610         write_lock_bh(&pol->lock);
2611         if (unlikely(pol->walk.dead)) {
2612                 /* target policy has been deleted */
2613                 write_unlock_bh(&pol->lock);
2614                 return -ENOENT;
2615         }
2616
2617         for (i = 0; i < pol->xfrm_nr; i++) {
2618                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2619                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2620                                 continue;
2621                         n++;
2622                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2623                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2624                                 continue;
2625                         /* update endpoints */
2626                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2627                                sizeof(pol->xfrm_vec[i].id.daddr));
2628                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2629                                sizeof(pol->xfrm_vec[i].saddr));
2630                         pol->xfrm_vec[i].encap_family = mp->new_family;
2631                         /* flush bundles */
2632                         while ((dst = pol->bundles) != NULL) {
2633                                 pol->bundles = dst->next;
2634                                 dst_free(dst);
2635                         }
2636                 }
2637         }
2638
2639         write_unlock_bh(&pol->lock);
2640
2641         if (!n)
2642                 return -ENODATA;
2643
2644         return 0;
2645 }
2646
2647 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2648 {
2649         int i, j;
2650
2651         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2652                 return -EINVAL;
2653
2654         for (i = 0; i < num_migrate; i++) {
2655                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2656                                    m[i].old_family) == 0) &&
2657                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2658                                    m[i].old_family) == 0))
2659                         return -EINVAL;
2660                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2661                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2662                         return -EINVAL;
2663
2664                 /* check if there is any duplicated entry */
2665                 for (j = i + 1; j < num_migrate; j++) {
2666                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2667                                     sizeof(m[i].old_daddr)) &&
2668                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2669                                     sizeof(m[i].old_saddr)) &&
2670                             m[i].proto == m[j].proto &&
2671                             m[i].mode == m[j].mode &&
2672                             m[i].reqid == m[j].reqid &&
2673                             m[i].old_family == m[j].old_family)
2674                                 return -EINVAL;
2675                 }
2676         }
2677
2678         return 0;
2679 }
2680
2681 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2682                  struct xfrm_migrate *m, int num_migrate,
2683                  struct xfrm_kmaddress *k)
2684 {
2685         int i, err, nx_cur = 0, nx_new = 0;
2686         struct xfrm_policy *pol = NULL;
2687         struct xfrm_state *x, *xc;
2688         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2689         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2690         struct xfrm_migrate *mp;
2691
2692         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2693                 goto out;
2694
2695         /* Stage 1 - find policy */
2696         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2697                 err = -ENOENT;
2698                 goto out;
2699         }
2700
2701         /* Stage 2 - find and update state(s) */
2702         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2703                 if ((x = xfrm_migrate_state_find(mp))) {
2704                         x_cur[nx_cur] = x;
2705                         nx_cur++;
2706                         if ((xc = xfrm_state_migrate(x, mp))) {
2707                                 x_new[nx_new] = xc;
2708                                 nx_new++;
2709                         } else {
2710                                 err = -ENODATA;
2711                                 goto restore_state;
2712                         }
2713                 }
2714         }
2715
2716         /* Stage 3 - update policy */
2717         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2718                 goto restore_state;
2719
2720         /* Stage 4 - delete old state(s) */
2721         if (nx_cur) {
2722                 xfrm_states_put(x_cur, nx_cur);
2723                 xfrm_states_delete(x_cur, nx_cur);
2724         }
2725
2726         /* Stage 5 - announce */
2727         km_migrate(sel, dir, type, m, num_migrate, k);
2728
2729         xfrm_pol_put(pol);
2730
2731         return 0;
2732 out:
2733         return err;
2734
2735 restore_state:
2736         if (pol)
2737                 xfrm_pol_put(pol);
2738         if (nx_cur)
2739                 xfrm_states_put(x_cur, nx_cur);
2740         if (nx_new)
2741                 xfrm_states_delete(x_new, nx_new);
2742
2743         return err;
2744 }
2745 EXPORT_SYMBOL(xfrm_migrate);
2746 #endif