net/xfrm/xfrm_state.c

   1 /*
   2  * xfrm_state.c
   3  *
   4  * Changes:
   5  *      Mitsuru KANDA @USAGI
   6  *      Kazunori MIYAZAWA @USAGI
   7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
   8  *              IPv6 support
   9  *      YOSHIFUJI Hideaki @USAGI
  10  *              Split up af-specific functions
  11  *      Derek Atkins <derek@ihtfp.com>
  12  *              Add UDP Encapsulation
  13  *
  14  */
  15
  16 #include <linux/workqueue.h>
  17 #include <net/xfrm.h>
  18 #include <linux/pfkeyv2.h>
  19 #include <linux/ipsec.h>
  20 #include <linux/module.h>
  21 #include <linux/cache.h>
  22 #include <linux/audit.h>
  23 #include <linux/uaccess.h>
  24 #include <linux/ktime.h>
  25 #include <linux/slab.h>
  26 #include <linux/interrupt.h>
  27 #include <linux/kernel.h>
  28
  29 #include "xfrm_hash.h"
  30
  31 #define xfrm_state_deref_prot(table, net) \
  32         rcu_dereference_protected((table), lockdep_is_held(&(net)->xfrm.xfrm_state_lock))
  33
  34 static void xfrm_state_gc_task(struct work_struct *work);
  35
  36 /* Each xfrm_state may be linked to two tables:
  37
  38    1. Hash table by (spi,daddr,ah/esp) to find SA by SPI. (input,ctl)
  39    2. Hash table by (daddr,family,reqid) to find what SAs exist for given
  40       destination/tunnel endpoint. (output)
  41  */
  42
  43 static unsigned int xfrm_state_hashmax __read_mostly = 1 * 1024 * 1024;
  44 static __read_mostly seqcount_t xfrm_state_hash_generation = SEQCNT_ZERO(xfrm_state_hash_generation);
  45
  46 static DECLARE_WORK(xfrm_state_gc_work, xfrm_state_gc_task);
  47 static HLIST_HEAD(xfrm_state_gc_list);
  48
  49 static inline bool xfrm_state_hold_rcu(struct xfrm_state __rcu *x)
  50 {
  51         return refcount_inc_not_zero(&x->refcnt);
  52 }
  53
  54 static inline unsigned int xfrm_dst_hash(struct net *net,
  55                                          const xfrm_address_t *daddr,
  56                                          const xfrm_address_t *saddr,
  57                                          u32 reqid,
  58                                          unsigned short family)
  59 {
  60         return __xfrm_dst_hash(daddr, saddr, reqid, family, net->xfrm.state_hmask);
  61 }
  62
  63 static inline unsigned int xfrm_src_hash(struct net *net,
  64                                          const xfrm_address_t *daddr,
  65                                          const xfrm_address_t *saddr,
  66                                          unsigned short family)
  67 {
  68         return __xfrm_src_hash(daddr, saddr, family, net->xfrm.state_hmask);
  69 }
  70
  71 static inline unsigned int
  72 xfrm_spi_hash(struct net *net, const xfrm_address_t *daddr,
  73               __be32 spi, u8 proto, unsigned short family)
  74 {
  75         return __xfrm_spi_hash(daddr, spi, proto, family, net->xfrm.state_hmask);
  76 }
  77
  78 static void xfrm_hash_transfer(struct hlist_head *list,
  79                                struct hlist_head *ndsttable,
  80                                struct hlist_head *nsrctable,
  81                                struct hlist_head *nspitable,
  82                                unsigned int nhashmask)
  83 {
  84         struct hlist_node *tmp;
  85         struct xfrm_state *x;
  86
  87         hlist_for_each_entry_safe(x, tmp, list, bydst) {
  88                 unsigned int h;
  89
  90                 h = __xfrm_dst_hash(&x->id.daddr, &x->props.saddr,
  91                                     x->props.reqid, x->props.family,
  92                                     nhashmask);
  93                 hlist_add_head_rcu(&x->bydst, ndsttable + h);
  94
  95                 h = __xfrm_src_hash(&x->id.daddr, &x->props.saddr,
  96                                     x->props.family,
  97                                     nhashmask);
  98                 hlist_add_head_rcu(&x->bysrc, nsrctable + h);
  99
 100                 if (x->id.spi) {
 101                         h = __xfrm_spi_hash(&x->id.daddr, x->id.spi,
 102                                             x->id.proto, x->props.family,
 103                                             nhashmask);
 104                         hlist_add_head_rcu(&x->byspi, nspitable + h);
 105                 }
 106         }
 107 }
 108
 109 static unsigned long xfrm_hash_new_size(unsigned int state_hmask)
 110 {
 111         return ((state_hmask + 1) << 1) * sizeof(struct hlist_head);
 112 }
 113
 114 static void xfrm_hash_resize(struct work_struct *work)
 115 {
 116         struct net *net = container_of(work, struct net, xfrm.state_hash_work);
 117         struct hlist_head *ndst, *nsrc, *nspi, *odst, *osrc, *ospi;
 118         unsigned long nsize, osize;
 119         unsigned int nhashmask, ohashmask;
 120         int i;
 121
 122         nsize = xfrm_hash_new_size(net->xfrm.state_hmask);
 123         ndst = xfrm_hash_alloc(nsize);
 124         if (!ndst)
 125                 return;
 126         nsrc = xfrm_hash_alloc(nsize);
 127         if (!nsrc) {
 128                 xfrm_hash_free(ndst, nsize);
 129                 return;
 130         }
 131         nspi = xfrm_hash_alloc(nsize);
 132         if (!nspi) {
 133                 xfrm_hash_free(ndst, nsize);
 134                 xfrm_hash_free(nsrc, nsize);
 135                 return;
 136         }
 137
 138         spin_lock_bh(&net->xfrm.xfrm_state_lock);
 139         write_seqcount_begin(&xfrm_state_hash_generation);
 140
 141         nhashmask = (nsize / sizeof(struct hlist_head)) - 1U;
 142         odst = xfrm_state_deref_prot(net->xfrm.state_bydst, net);
 143         for (i = net->xfrm.state_hmask; i >= 0; i--)
 144                 xfrm_hash_transfer(odst + i, ndst, nsrc, nspi, nhashmask);
 145
 146         osrc = xfrm_state_deref_prot(net->xfrm.state_bysrc, net);
 147         ospi = xfrm_state_deref_prot(net->xfrm.state_byspi, net);
 148         ohashmask = net->xfrm.state_hmask;
 149
 150         rcu_assign_pointer(net->xfrm.state_bydst, ndst);
 151         rcu_assign_pointer(net->xfrm.state_bysrc, nsrc);
 152         rcu_assign_pointer(net->xfrm.state_byspi, nspi);
 153         net->xfrm.state_hmask = nhashmask;
 154
 155         write_seqcount_end(&xfrm_state_hash_generation);
 156         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
 157
 158         osize = (ohashmask + 1) * sizeof(struct hlist_head);
 159
 160         synchronize_rcu();
 161
 162         xfrm_hash_free(odst, osize);
 163         xfrm_hash_free(osrc, osize);
 164         xfrm_hash_free(ospi, osize);
 165 }
 166
 167 static DEFINE_SPINLOCK(xfrm_state_afinfo_lock);
 168 static struct xfrm_state_afinfo __rcu *xfrm_state_afinfo[NPROTO];
 169
 170 static DEFINE_SPINLOCK(xfrm_state_gc_lock);
 171
 172 int __xfrm_state_delete(struct xfrm_state *x);
 173
 174 int km_query(struct xfrm_state *x, struct xfrm_tmpl *t, struct xfrm_policy *pol);
 175 bool km_is_alive(const struct km_event *c);
 176 void km_state_expired(struct xfrm_state *x, int hard, u32 portid);
 177
 178 static DEFINE_SPINLOCK(xfrm_type_lock);
 179 int xfrm_register_type(const struct xfrm_type *type, unsigned short family)
 180 {
 181         struct xfrm_state_afinfo *afinfo = xfrm_state_get_afinfo(family);
 182         const struct xfrm_type **typemap;
 183         int err = 0;
 184
 185         if (unlikely(afinfo == NULL))
 186                 return -EAFNOSUPPORT;
 187         typemap = afinfo->type_map;
 188         spin_lock_bh(&xfrm_type_lock);
 189
 190         if (likely(typemap[type->proto] == NULL))
 191                 typemap[type->proto] = type;
 192         else
 193                 err = -EEXIST;
 194         spin_unlock_bh(&xfrm_type_lock);
 195         rcu_read_unlock();
 196         return err;
 197 }
 198 EXPORT_SYMBOL(xfrm_register_type);
 199
 200 int xfrm_unregister_type(const struct xfrm_type *type, unsigned short family)
 201 {
 202         struct xfrm_state_afinfo *afinfo = xfrm_state_get_afinfo(family);
 203         const struct xfrm_type **typemap;
 204         int err = 0;
 205
 206         if (unlikely(afinfo == NULL))
 207                 return -EAFNOSUPPORT;
 208         typemap = afinfo->type_map;
 209         spin_lock_bh(&xfrm_type_lock);
 210
 211         if (unlikely(typemap[type->proto] != type))
 212                 err = -ENOENT;
 213         else
 214                 typemap[type->proto] = NULL;
 215         spin_unlock_bh(&xfrm_type_lock);
 216         rcu_read_unlock();
 217         return err;
 218 }
 219 EXPORT_SYMBOL(xfrm_unregister_type);
 220
 221 static const struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
 222 {
 223         struct xfrm_state_afinfo *afinfo;
 224         const struct xfrm_type **typemap;
 225         const struct xfrm_type *type;
 226         int modload_attempted = 0;
 227
 228 retry:
 229         afinfo = xfrm_state_get_afinfo(family);
 230         if (unlikely(afinfo == NULL))
 231                 return NULL;
 232         typemap = afinfo->type_map;
 233
 234         type = READ_ONCE(typemap[proto]);
 235         if (unlikely(type && !try_module_get(type->owner)))
 236                 type = NULL;
 237
 238         rcu_read_unlock();
 239
 240         if (!type && !modload_attempted) {
 241                 request_module("xfrm-type-%d-%d", family, proto);
 242                 modload_attempted = 1;
 243                 goto retry;
 244         }
 245
 246         return type;
 247 }
 248
 249 static void xfrm_put_type(const struct xfrm_type *type)
 250 {
 251         module_put(type->owner);
 252 }
 253
 254 static DEFINE_SPINLOCK(xfrm_type_offload_lock);
 255 int xfrm_register_type_offload(const struct xfrm_type_offload *type,
 256                                unsigned short family)
 257 {
 258         struct xfrm_state_afinfo *afinfo = xfrm_state_get_afinfo(family);
 259         const struct xfrm_type_offload **typemap;
 260         int err = 0;
 261
 262         if (unlikely(afinfo == NULL))
 263                 return -EAFNOSUPPORT;
 264         typemap = afinfo->type_offload_map;
 265         spin_lock_bh(&xfrm_type_offload_lock);
 266
 267         if (likely(typemap[type->proto] == NULL))
 268                 typemap[type->proto] = type;
 269         else
 270                 err = -EEXIST;
 271         spin_unlock_bh(&xfrm_type_offload_lock);
 272         rcu_read_unlock();
 273         return err;
 274 }
 275 EXPORT_SYMBOL(xfrm_register_type_offload);
 276
 277 int xfrm_unregister_type_offload(const struct xfrm_type_offload *type,
 278                                  unsigned short family)
 279 {
 280         struct xfrm_state_afinfo *afinfo = xfrm_state_get_afinfo(family);
 281         const struct xfrm_type_offload **typemap;
 282         int err = 0;
 283
 284         if (unlikely(afinfo == NULL))
 285                 return -EAFNOSUPPORT;
 286         typemap = afinfo->type_offload_map;
 287         spin_lock_bh(&xfrm_type_offload_lock);
 288
 289         if (unlikely(typemap[type->proto] != type))
 290                 err = -ENOENT;
 291         else
 292                 typemap[type->proto] = NULL;
 293         spin_unlock_bh(&xfrm_type_offload_lock);
 294         rcu_read_unlock();
 295         return err;
 296 }
 297 EXPORT_SYMBOL(xfrm_unregister_type_offload);
 298
 299 static const struct xfrm_type_offload *
 300 xfrm_get_type_offload(u8 proto, unsigned short family, bool try_load)
 301 {
 302         struct xfrm_state_afinfo *afinfo;
 303         const struct xfrm_type_offload **typemap;
 304         const struct xfrm_type_offload *type;
 305
 306 retry:
 307         afinfo = xfrm_state_get_afinfo(family);
 308         if (unlikely(afinfo == NULL))
 309                 return NULL;
 310         typemap = afinfo->type_offload_map;
 311
 312         type = typemap[proto];
 313         if ((type && !try_module_get(type->owner)))
 314                 type = NULL;
 315
 316         if (!type && try_load) {
 317                 request_module("xfrm-offload-%d-%d", family, proto);
 318                 try_load = 0;
 319                 goto retry;
 320         }
 321
 322         rcu_read_unlock();
 323         return type;
 324 }
 325
 326 static void xfrm_put_type_offload(const struct xfrm_type_offload *type)
 327 {
 328         module_put(type->owner);
 329 }
 330
 331 static DEFINE_SPINLOCK(xfrm_mode_lock);
 332 int xfrm_register_mode(struct xfrm_mode *mode, int family)
 333 {
 334         struct xfrm_state_afinfo *afinfo;
 335         struct xfrm_mode **modemap;
 336         int err;
 337
 338         if (unlikely(mode->encap >= XFRM_MODE_MAX))
 339                 return -EINVAL;
 340
 341         afinfo = xfrm_state_get_afinfo(family);
 342         if (unlikely(afinfo == NULL))
 343                 return -EAFNOSUPPORT;
 344
 345         err = -EEXIST;
 346         modemap = afinfo->mode_map;
 347         spin_lock_bh(&xfrm_mode_lock);
 348         if (modemap[mode->encap])
 349                 goto out;
 350
 351         err = -ENOENT;
 352         if (!try_module_get(afinfo->owner))
 353                 goto out;
 354
 355         mode->afinfo = afinfo;
 356         modemap[mode->encap] = mode;
 357         err = 0;
 358
 359 out:
 360         spin_unlock_bh(&xfrm_mode_lock);
 361         rcu_read_unlock();
 362         return err;
 363 }
 364 EXPORT_SYMBOL(xfrm_register_mode);
 365
 366 int xfrm_unregister_mode(struct xfrm_mode *mode, int family)
 367 {
 368         struct xfrm_state_afinfo *afinfo;
 369         struct xfrm_mode **modemap;
 370         int err;
 371
 372         if (unlikely(mode->encap >= XFRM_MODE_MAX))
 373                 return -EINVAL;
 374
 375         afinfo = xfrm_state_get_afinfo(family);
 376         if (unlikely(afinfo == NULL))
 377                 return -EAFNOSUPPORT;
 378
 379         err = -ENOENT;
 380         modemap = afinfo->mode_map;
 381         spin_lock_bh(&xfrm_mode_lock);
 382         if (likely(modemap[mode->encap] == mode)) {
 383                 modemap[mode->encap] = NULL;
 384                 module_put(mode->afinfo->owner);
 385                 err = 0;
 386         }
 387
 388         spin_unlock_bh(&xfrm_mode_lock);
 389         rcu_read_unlock();
 390         return err;
 391 }
 392 EXPORT_SYMBOL(xfrm_unregister_mode);
 393
 394 static struct xfrm_mode *xfrm_get_mode(unsigned int encap, int family)
 395 {
 396         struct xfrm_state_afinfo *afinfo;
 397         struct xfrm_mode *mode;
 398         int modload_attempted = 0;
 399
 400         if (unlikely(encap >= XFRM_MODE_MAX))
 401                 return NULL;
 402
 403 retry:
 404         afinfo = xfrm_state_get_afinfo(family);
 405         if (unlikely(afinfo == NULL))
 406                 return NULL;
 407
 408         mode = READ_ONCE(afinfo->mode_map[encap]);
 409         if (unlikely(mode && !try_module_get(mode->owner)))
 410                 mode = NULL;
 411
 412         rcu_read_unlock();
 413         if (!mode && !modload_attempted) {
 414                 request_module("xfrm-mode-%d-%d", family, encap);
 415                 modload_attempted = 1;
 416                 goto retry;
 417         }
 418
 419         return mode;
 420 }
 421
 422 static void xfrm_put_mode(struct xfrm_mode *mode)
 423 {
 424         module_put(mode->owner);
 425 }
 426
 427 static void xfrm_state_gc_destroy(struct xfrm_state *x)
 428 {
 429         tasklet_hrtimer_cancel(&x->mtimer);
 430         del_timer_sync(&x->rtimer);
 431         kfree(x->aead);
 432         kfree(x->aalg);
 433         kfree(x->ealg);
 434         kfree(x->calg);
 435         kfree(x->encap);
 436         kfree(x->coaddr);
 437         kfree(x->replay_esn);
 438         kfree(x->preplay_esn);
 439         if (x->inner_mode)
 440                 xfrm_put_mode(x->inner_mode);
 441         if (x->inner_mode_iaf)
 442                 xfrm_put_mode(x->inner_mode_iaf);
 443         if (x->outer_mode)
 444                 xfrm_put_mode(x->outer_mode);
 445         if (x->type_offload)
 446                 xfrm_put_type_offload(x->type_offload);
 447         if (x->type) {
 448                 x->type->destructor(x);
 449                 xfrm_put_type(x->type);
 450         }
 451         xfrm_dev_state_free(x);
 452         security_xfrm_state_free(x);
 453         kfree(x);
 454 }
 455
 456 static void xfrm_state_gc_task(struct work_struct *work)
 457 {
 458         struct xfrm_state *x;
 459         struct hlist_node *tmp;
 460         struct hlist_head gc_list;
 461
 462         spin_lock_bh(&xfrm_state_gc_lock);
 463         hlist_move_list(&xfrm_state_gc_list, &gc_list);
 464         spin_unlock_bh(&xfrm_state_gc_lock);
 465
 466         synchronize_rcu();
 467
 468         hlist_for_each_entry_safe(x, tmp, &gc_list, gclist)
 469                 xfrm_state_gc_destroy(x);
 470 }
 471
 472 static enum hrtimer_restart xfrm_timer_handler(struct hrtimer *me)
 473 {
 474         struct tasklet_hrtimer *thr = container_of(me, struct tasklet_hrtimer, timer);
 475         struct xfrm_state *x = container_of(thr, struct xfrm_state, mtimer);
 476         unsigned long now = get_seconds();
 477         long next = LONG_MAX;
 478         int warn = 0;
 479         int err = 0;
 480
 481         spin_lock(&x->lock);
 482         if (x->km.state == XFRM_STATE_DEAD)
 483                 goto out;
 484         if (x->km.state == XFRM_STATE_EXPIRED)
 485                 goto expired;
 486         if (x->lft.hard_add_expires_seconds) {
 487                 long tmo = x->lft.hard_add_expires_seconds +
 488                         x->curlft.add_time - now;
 489                 if (tmo <= 0) {
 490                         if (x->xflags & XFRM_SOFT_EXPIRE) {
 491                                 /* enter hard expire without soft expire first?!
 492                                  * setting a new date could trigger this.
 493                                  * workaround: fix x->curflt.add_time by below:
 494                                  */
 495                                 x->curlft.add_time = now - x->saved_tmo - 1;
 496                                 tmo = x->lft.hard_add_expires_seconds - x->saved_tmo;
 497                         } else
 498                                 goto expired;
 499                 }
 500                 if (tmo < next)
 501                         next = tmo;
 502         }
 503         if (x->lft.hard_use_expires_seconds) {
 504                 long tmo = x->lft.hard_use_expires_seconds +
 505                         (x->curlft.use_time ? : now) - now;
 506                 if (tmo <= 0)
 507                         goto expired;
 508                 if (tmo < next)
 509                         next = tmo;
 510         }
 511         if (x->km.dying)
 512                 goto resched;
 513         if (x->lft.soft_add_expires_seconds) {
 514                 long tmo = x->lft.soft_add_expires_seconds +
 515                         x->curlft.add_time - now;
 516                 if (tmo <= 0) {
 517                         warn = 1;
 518                         x->xflags &= ~XFRM_SOFT_EXPIRE;
 519                 } else if (tmo < next) {
 520                         next = tmo;
 521                         x->xflags |= XFRM_SOFT_EXPIRE;
 522                         x->saved_tmo = tmo;
 523                 }
 524         }
 525         if (x->lft.soft_use_expires_seconds) {
 526                 long tmo = x->lft.soft_use_expires_seconds +
 527                         (x->curlft.use_time ? : now) - now;
 528                 if (tmo <= 0)
 529                         warn = 1;
 530                 else if (tmo < next)
 531                         next = tmo;
 532         }
 533
 534         x->km.dying = warn;
 535         if (warn)
 536                 km_state_expired(x, 0, 0);
 537 resched:
 538         if (next != LONG_MAX) {
 539                 tasklet_hrtimer_start(&x->mtimer, ktime_set(next, 0), HRTIMER_MODE_REL);
 540         }
 541
 542         goto out;
 543
 544 expired:
 545         if (x->km.state == XFRM_STATE_ACQ && x->id.spi == 0)
 546                 x->km.state = XFRM_STATE_EXPIRED;
 547
 548         err = __xfrm_state_delete(x);
 549         if (!err)
 550                 km_state_expired(x, 1, 0);
 551
 552         xfrm_audit_state_delete(x, err ? 0 : 1, true);
 553
 554 out:
 555         spin_unlock(&x->lock);
 556         return HRTIMER_NORESTART;
 557 }
 558
 559 static void xfrm_replay_timer_handler(struct timer_list *t);
 560
 561 struct xfrm_state *xfrm_state_alloc(struct net *net)
 562 {
 563         struct xfrm_state *x;
 564
 565         x = kzalloc(sizeof(struct xfrm_state), GFP_ATOMIC);
 566
 567         if (x) {
 568                 write_pnet(&x->xs_net, net);
 569                 refcount_set(&x->refcnt, 1);
 570                 atomic_set(&x->tunnel_users, 0);
 571                 INIT_LIST_HEAD(&x->km.all);
 572                 INIT_HLIST_NODE(&x->bydst);
 573                 INIT_HLIST_NODE(&x->bysrc);
 574                 INIT_HLIST_NODE(&x->byspi);
 575                 tasklet_hrtimer_init(&x->mtimer, xfrm_timer_handler,
 576                                         CLOCK_BOOTTIME, HRTIMER_MODE_ABS);
 577                 timer_setup(&x->rtimer, xfrm_replay_timer_handler, 0);
 578                 x->curlft.add_time = get_seconds();
 579                 x->lft.soft_byte_limit = XFRM_INF;
 580                 x->lft.soft_packet_limit = XFRM_INF;
 581                 x->lft.hard_byte_limit = XFRM_INF;
 582                 x->lft.hard_packet_limit = XFRM_INF;
 583                 x->replay_maxage = 0;
 584                 x->replay_maxdiff = 0;
 585                 x->inner_mode = NULL;
 586                 x->inner_mode_iaf = NULL;
 587                 spin_lock_init(&x->lock);
 588         }
 589         return x;
 590 }
 591 EXPORT_SYMBOL(xfrm_state_alloc);
 592
 593 void __xfrm_state_destroy(struct xfrm_state *x)
 594 {
 595         WARN_ON(x->km.state != XFRM_STATE_DEAD);
 596
 597         spin_lock_bh(&xfrm_state_gc_lock);
 598         hlist_add_head(&x->gclist, &xfrm_state_gc_list);
 599         spin_unlock_bh(&xfrm_state_gc_lock);
 600         schedule_work(&xfrm_state_gc_work);
 601 }
 602 EXPORT_SYMBOL(__xfrm_state_destroy);
 603
 604 int __xfrm_state_delete(struct xfrm_state *x)
 605 {
 606         struct net *net = xs_net(x);
 607         int err = -ESRCH;
 608
 609         if (x->km.state != XFRM_STATE_DEAD) {
 610                 x->km.state = XFRM_STATE_DEAD;
 611                 spin_lock(&net->xfrm.xfrm_state_lock);
 612                 list_del(&x->km.all);
 613                 hlist_del_rcu(&x->bydst);
 614                 hlist_del_rcu(&x->bysrc);
 615                 if (x->id.spi)
 616                         hlist_del_rcu(&x->byspi);
 617                 net->xfrm.state_num--;
 618                 spin_unlock(&net->xfrm.xfrm_state_lock);
 619
 620                 xfrm_dev_state_delete(x);
 621
 622                 /* All xfrm_state objects are created by xfrm_state_alloc.
 623                  * The xfrm_state_alloc call gives a reference, and that
 624                  * is what we are dropping here.
 625                  */
 626                 xfrm_state_put(x);
 627                 err = 0;
 628         }
 629
 630         return err;
 631 }
 632 EXPORT_SYMBOL(__xfrm_state_delete);
 633
 634 int xfrm_state_delete(struct xfrm_state *x)
 635 {
 636         int err;
 637
 638         spin_lock_bh(&x->lock);
 639         err = __xfrm_state_delete(x);
 640         spin_unlock_bh(&x->lock);
 641
 642         return err;
 643 }
 644 EXPORT_SYMBOL(xfrm_state_delete);
 645
 646 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 647 static inline int
 648 xfrm_state_flush_secctx_check(struct net *net, u8 proto, bool task_valid)
 649 {
 650         int i, err = 0;
 651
 652         for (i = 0; i <= net->xfrm.state_hmask; i++) {
 653                 struct xfrm_state *x;
 654
 655                 hlist_for_each_entry(x, net->xfrm.state_bydst+i, bydst) {
 656                         if (xfrm_id_proto_match(x->id.proto, proto) &&
 657                            (err = security_xfrm_state_delete(x)) != 0) {
 658                                 xfrm_audit_state_delete(x, 0, task_valid);
 659                                 return err;
 660                         }
 661                 }
 662         }
 663
 664         return err;
 665 }
 666
 667 static inline int
 668 xfrm_dev_state_flush_secctx_check(struct net *net, struct net_device *dev, bool task_valid)
 669 {
 670         int i, err = 0;
 671
 672         for (i = 0; i <= net->xfrm.state_hmask; i++) {
 673                 struct xfrm_state *x;
 674                 struct xfrm_state_offload *xso;
 675
 676                 hlist_for_each_entry(x, net->xfrm.state_bydst+i, bydst) {
 677                         xso = &x->xso;
 678
 679                         if (xso->dev == dev &&
 680                            (err = security_xfrm_state_delete(x)) != 0) {
 681                                 xfrm_audit_state_delete(x, 0, task_valid);
 682                                 return err;
 683                         }
 684                 }
 685         }
 686
 687         return err;
 688 }
 689 #else
 690 static inline int
 691 xfrm_state_flush_secctx_check(struct net *net, u8 proto, bool task_valid)
 692 {
 693         return 0;
 694 }
 695
 696 static inline int
 697 xfrm_dev_state_flush_secctx_check(struct net *net, struct net_device *dev, bool task_valid)
 698 {
 699         return 0;
 700 }
 701 #endif
 702
 703 int xfrm_state_flush(struct net *net, u8 proto, bool task_valid)
 704 {
 705         int i, err = 0, cnt = 0;
 706
 707         spin_lock_bh(&net->xfrm.xfrm_state_lock);
 708         err = xfrm_state_flush_secctx_check(net, proto, task_valid);
 709         if (err)
 710                 goto out;
 711
 712         err = -ESRCH;
 713         for (i = 0; i <= net->xfrm.state_hmask; i++) {
 714                 struct xfrm_state *x;
 715 restart:
 716                 hlist_for_each_entry(x, net->xfrm.state_bydst+i, bydst) {
 717                         if (!xfrm_state_kern(x) &&
 718                             xfrm_id_proto_match(x->id.proto, proto)) {
 719                                 xfrm_state_hold(x);
 720                                 spin_unlock_bh(&net->xfrm.xfrm_state_lock);
 721
 722                                 err = xfrm_state_delete(x);
 723                                 xfrm_audit_state_delete(x, err ? 0 : 1,
 724                                                         task_valid);
 725                                 xfrm_state_put(x);
 726                                 if (!err)
 727                                         cnt++;
 728
 729                                 spin_lock_bh(&net->xfrm.xfrm_state_lock);
 730                                 goto restart;
 731                         }
 732                 }
 733         }
 734 out:
 735         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
 736         if (cnt) {
 737                 err = 0;
 738                 xfrm_policy_cache_flush();
 739         }
 740         return err;
 741 }
 742 EXPORT_SYMBOL(xfrm_state_flush);
 743
 744 int xfrm_dev_state_flush(struct net *net, struct net_device *dev, bool task_valid)
 745 {
 746         int i, err = 0, cnt = 0;
 747
 748         spin_lock_bh(&net->xfrm.xfrm_state_lock);
 749         err = xfrm_dev_state_flush_secctx_check(net, dev, task_valid);
 750         if (err)
 751                 goto out;
 752
 753         err = -ESRCH;
 754         for (i = 0; i <= net->xfrm.state_hmask; i++) {
 755                 struct xfrm_state *x;
 756                 struct xfrm_state_offload *xso;
 757 restart:
 758                 hlist_for_each_entry(x, net->xfrm.state_bydst+i, bydst) {
 759                         xso = &x->xso;
 760
 761                         if (!xfrm_state_kern(x) && xso->dev == dev) {
 762                                 xfrm_state_hold(x);
 763                                 spin_unlock_bh(&net->xfrm.xfrm_state_lock);
 764
 765                                 err = xfrm_state_delete(x);
 766                                 xfrm_audit_state_delete(x, err ? 0 : 1,
 767                                                         task_valid);
 768                                 xfrm_state_put(x);
 769                                 if (!err)
 770                                         cnt++;
 771
 772                                 spin_lock_bh(&net->xfrm.xfrm_state_lock);
 773                                 goto restart;
 774                         }
 775                 }
 776         }
 777         if (cnt)
 778                 err = 0;
 779
 780 out:
 781         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
 782         return err;
 783 }
 784 EXPORT_SYMBOL(xfrm_dev_state_flush);
 785
 786 void xfrm_sad_getinfo(struct net *net, struct xfrmk_sadinfo *si)
 787 {
 788         spin_lock_bh(&net->xfrm.xfrm_state_lock);
 789         si->sadcnt = net->xfrm.state_num;
 790         si->sadhcnt = net->xfrm.state_hmask;
 791         si->sadhmcnt = xfrm_state_hashmax;
 792         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
 793 }
 794 EXPORT_SYMBOL(xfrm_sad_getinfo);
 795
 796 static void
 797 xfrm_init_tempstate(struct xfrm_state *x, const struct flowi *fl,
 798                     const struct xfrm_tmpl *tmpl,
 799                     const xfrm_address_t *daddr, const xfrm_address_t *saddr,
 800                     unsigned short family)
 801 {
 802         struct xfrm_state_afinfo *afinfo = xfrm_state_afinfo_get_rcu(family);
 803
 804         if (!afinfo)
 805                 return;
 806
 807         afinfo->init_tempsel(&x->sel, fl);
 808
 809         if (family != tmpl->encap_family) {
 810                 afinfo = xfrm_state_afinfo_get_rcu(tmpl->encap_family);
 811                 if (!afinfo)
 812                         return;
 813         }
 814         afinfo->init_temprop(x, tmpl, daddr, saddr);
 815 }
 816
 817 static struct xfrm_state *__xfrm_state_lookup(struct net *net, u32 mark,
 818                                               const xfrm_address_t *daddr,
 819                                               __be32 spi, u8 proto,
 820                                               unsigned short family)
 821 {
 822         unsigned int h = xfrm_spi_hash(net, daddr, spi, proto, family);
 823         struct xfrm_state *x;
 824
 825         hlist_for_each_entry_rcu(x, net->xfrm.state_byspi + h, byspi) {
 826                 if (x->props.family != family ||
 827                     x->id.spi       != spi ||
 828                     x->id.proto     != proto ||
 829                     !xfrm_addr_equal(&x->id.daddr, daddr, family))
 830                         continue;
 831
 832                 if ((mark & x->mark.m) != x->mark.v)
 833                         continue;
 834                 if (!xfrm_state_hold_rcu(x))
 835                         continue;
 836                 return x;
 837         }
 838
 839         return NULL;
 840 }
 841
 842 static struct xfrm_state *__xfrm_state_lookup_byaddr(struct net *net, u32 mark,
 843                                                      const xfrm_address_t *daddr,
 844                                                      const xfrm_address_t *saddr,
 845                                                      u8 proto, unsigned short family)
 846 {
 847         unsigned int h = xfrm_src_hash(net, daddr, saddr, family);
 848         struct xfrm_state *x;
 849
 850         hlist_for_each_entry_rcu(x, net->xfrm.state_bysrc + h, bysrc) {
 851                 if (x->props.family != family ||
 852                     x->id.proto     != proto ||
 853                     !xfrm_addr_equal(&x->id.daddr, daddr, family) ||
 854                     !xfrm_addr_equal(&x->props.saddr, saddr, family))
 855                         continue;
 856
 857                 if ((mark & x->mark.m) != x->mark.v)
 858                         continue;
 859                 if (!xfrm_state_hold_rcu(x))
 860                         continue;
 861                 return x;
 862         }
 863
 864         return NULL;
 865 }
 866
 867 static inline struct xfrm_state *
 868 __xfrm_state_locate(struct xfrm_state *x, int use_spi, int family)
 869 {
 870         struct net *net = xs_net(x);
 871         u32 mark = x->mark.v & x->mark.m;
 872
 873         if (use_spi)
 874                 return __xfrm_state_lookup(net, mark, &x->id.daddr,
 875                                            x->id.spi, x->id.proto, family);
 876         else
 877                 return __xfrm_state_lookup_byaddr(net, mark,
 878                                                   &x->id.daddr,
 879                                                   &x->props.saddr,
 880                                                   x->id.proto, family);
 881 }
 882
 883 static void xfrm_hash_grow_check(struct net *net, int have_hash_collision)
 884 {
 885         if (have_hash_collision &&
 886             (net->xfrm.state_hmask + 1) < xfrm_state_hashmax &&
 887             net->xfrm.state_num > net->xfrm.state_hmask)
 888                 schedule_work(&net->xfrm.state_hash_work);
 889 }
 890
 891 static void xfrm_state_look_at(struct xfrm_policy *pol, struct xfrm_state *x,
 892                                const struct flowi *fl, unsigned short family,
 893                                struct xfrm_state **best, int *acq_in_progress,
 894                                int *error)
 895 {
 896         /* Resolution logic:
 897          * 1. There is a valid state with matching selector. Done.
 898          * 2. Valid state with inappropriate selector. Skip.
 899          *
 900          * Entering area of "sysdeps".
 901          *
 902          * 3. If state is not valid, selector is temporary, it selects
 903          *    only session which triggered previous resolution. Key
 904          *    manager will do something to install a state with proper
 905          *    selector.
 906          */
 907         if (x->km.state == XFRM_STATE_VALID) {
 908                 if ((x->sel.family &&
 909                      !xfrm_selector_match(&x->sel, fl, x->sel.family)) ||
 910                     !security_xfrm_state_pol_flow_match(x, pol, fl))
 911                         return;
 912
 913                 if (!*best ||
 914                     (*best)->km.dying > x->km.dying ||
 915                     ((*best)->km.dying == x->km.dying &&
 916                      (*best)->curlft.add_time < x->curlft.add_time))
 917                         *best = x;
 918         } else if (x->km.state == XFRM_STATE_ACQ) {
 919                 *acq_in_progress = 1;
 920         } else if (x->km.state == XFRM_STATE_ERROR ||
 921                    x->km.state == XFRM_STATE_EXPIRED) {
 922                 if (xfrm_selector_match(&x->sel, fl, x->sel.family) &&
 923                     security_xfrm_state_pol_flow_match(x, pol, fl))
 924                         *error = -ESRCH;
 925         }
 926 }
 927
 928 struct xfrm_state *
 929 xfrm_state_find(const xfrm_address_t *daddr, const xfrm_address_t *saddr,
 930                 const struct flowi *fl, struct xfrm_tmpl *tmpl,
 931                 struct xfrm_policy *pol, int *err,
 932                 unsigned short family)
 933 {
 934         static xfrm_address_t saddr_wildcard = { };
 935         struct net *net = xp_net(pol);
 936         unsigned int h, h_wildcard;
 937         struct xfrm_state *x, *x0, *to_put;
 938         int acquire_in_progress = 0;
 939         int error = 0;
 940         struct xfrm_state *best = NULL;
 941         u32 mark = pol->mark.v & pol->mark.m;
 942         unsigned short encap_family = tmpl->encap_family;
 943         unsigned int sequence;
 944         struct km_event c;
 945
 946         to_put = NULL;
 947
 948         sequence = read_seqcount_begin(&xfrm_state_hash_generation);
 949
 950         rcu_read_lock();
 951         h = xfrm_dst_hash(net, daddr, saddr, tmpl->reqid, encap_family);
 952         hlist_for_each_entry_rcu(x, net->xfrm.state_bydst + h, bydst) {
 953                 if (x->props.family == encap_family &&
 954                     x->props.reqid == tmpl->reqid &&
 955                     (mark & x->mark.m) == x->mark.v &&
 956                     !(x->props.flags & XFRM_STATE_WILDRECV) &&
 957                     xfrm_state_addr_check(x, daddr, saddr, encap_family) &&
 958                     tmpl->mode == x->props.mode &&
 959                     tmpl->id.proto == x->id.proto &&
 960                     (tmpl->id.spi == x->id.spi || !tmpl->id.spi))
 961                         xfrm_state_look_at(pol, x, fl, encap_family,
 962                                            &best, &acquire_in_progress, &error);
 963         }
 964         if (best || acquire_in_progress)
 965                 goto found;
 966
 967         h_wildcard = xfrm_dst_hash(net, daddr, &saddr_wildcard, tmpl->reqid, encap_family);
 968         hlist_for_each_entry_rcu(x, net->xfrm.state_bydst + h_wildcard, bydst) {
 969                 if (x->props.family == encap_family &&
 970                     x->props.reqid == tmpl->reqid &&
 971                     (mark & x->mark.m) == x->mark.v &&
 972                     !(x->props.flags & XFRM_STATE_WILDRECV) &&
 973                     xfrm_addr_equal(&x->id.daddr, daddr, encap_family) &&
 974                     tmpl->mode == x->props.mode &&
 975                     tmpl->id.proto == x->id.proto &&
 976                     (tmpl->id.spi == x->id.spi || !tmpl->id.spi))
 977                         xfrm_state_look_at(pol, x, fl, encap_family,
 978                                            &best, &acquire_in_progress, &error);
 979         }
 980
 981 found:
 982         x = best;
 983         if (!x && !error && !acquire_in_progress) {
 984                 if (tmpl->id.spi &&
 985                     (x0 = __xfrm_state_lookup(net, mark, daddr, tmpl->id.spi,
 986                                               tmpl->id.proto, encap_family)) != NULL) {
 987                         to_put = x0;
 988                         error = -EEXIST;
 989                         goto out;
 990                 }
 991
 992                 c.net = net;
 993                 /* If the KMs have no listeners (yet...), avoid allocating an SA
 994                  * for each and every packet - garbage collection might not
 995                  * handle the flood.
 996                  */
 997                 if (!km_is_alive(&c)) {
 998                         error = -ESRCH;
 999                         goto out;
1000                 }
1001
1002                 x = xfrm_state_alloc(net);
1003                 if (x == NULL) {
1004                         error = -ENOMEM;
1005                         goto out;
1006                 }
1007                 /* Initialize temporary state matching only
1008                  * to current session. */
1009                 xfrm_init_tempstate(x, fl, tmpl, daddr, saddr, family);
1010                 memcpy(&x->mark, &pol->mark, sizeof(x->mark));
1011
1012                 error = security_xfrm_state_alloc_acquire(x, pol->security, fl->flowi_secid);
1013                 if (error) {
1014                         x->km.state = XFRM_STATE_DEAD;
1015                         to_put = x;
1016                         x = NULL;
1017                         goto out;
1018                 }
1019
1020                 if (km_query(x, tmpl, pol) == 0) {
1021                         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1022                         x->km.state = XFRM_STATE_ACQ;
1023                         list_add(&x->km.all, &net->xfrm.state_all);
1024                         hlist_add_head_rcu(&x->bydst, net->xfrm.state_bydst + h);
1025                         h = xfrm_src_hash(net, daddr, saddr, encap_family);
1026                         hlist_add_head_rcu(&x->bysrc, net->xfrm.state_bysrc + h);
1027                         if (x->id.spi) {
1028                                 h = xfrm_spi_hash(net, &x->id.daddr, x->id.spi, x->id.proto, encap_family);
1029                                 hlist_add_head_rcu(&x->byspi, net->xfrm.state_byspi + h);
1030                         }
1031                         x->lft.hard_add_expires_seconds = net->xfrm.sysctl_acq_expires;
1032                         tasklet_hrtimer_start(&x->mtimer, ktime_set(net->xfrm.sysctl_acq_expires, 0), HRTIMER_MODE_REL);
1033                         net->xfrm.state_num++;
1034                         xfrm_hash_grow_check(net, x->bydst.next != NULL);
1035                         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1036                 } else {
1037                         x->km.state = XFRM_STATE_DEAD;
1038                         to_put = x;
1039                         x = NULL;
1040                         error = -ESRCH;
1041                 }
1042         }
1043 out:
1044         if (x) {
1045                 if (!xfrm_state_hold_rcu(x)) {
1046                         *err = -EAGAIN;
1047                         x = NULL;
1048                 }
1049         } else {
1050                 *err = acquire_in_progress ? -EAGAIN : error;
1051         }
1052         rcu_read_unlock();
1053         if (to_put)
1054                 xfrm_state_put(to_put);
1055
1056         if (read_seqcount_retry(&xfrm_state_hash_generation, sequence)) {
1057                 *err = -EAGAIN;
1058                 if (x) {
1059                         xfrm_state_put(x);
1060                         x = NULL;
1061                 }
1062         }
1063
1064         return x;
1065 }
1066
1067 struct xfrm_state *
1068 xfrm_stateonly_find(struct net *net, u32 mark,
1069                     xfrm_address_t *daddr, xfrm_address_t *saddr,
1070                     unsigned short family, u8 mode, u8 proto, u32 reqid)
1071 {
1072         unsigned int h;
1073         struct xfrm_state *rx = NULL, *x = NULL;
1074
1075         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1076         h = xfrm_dst_hash(net, daddr, saddr, reqid, family);
1077         hlist_for_each_entry(x, net->xfrm.state_bydst+h, bydst) {
1078                 if (x->props.family == family &&
1079                     x->props.reqid == reqid &&
1080                     (mark & x->mark.m) == x->mark.v &&
1081                     !(x->props.flags & XFRM_STATE_WILDRECV) &&
1082                     xfrm_state_addr_check(x, daddr, saddr, family) &&
1083                     mode == x->props.mode &&
1084                     proto == x->id.proto &&
1085                     x->km.state == XFRM_STATE_VALID) {
1086                         rx = x;
1087                         break;
1088                 }
1089         }
1090
1091         if (rx)
1092                 xfrm_state_hold(rx);
1093         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1094
1095
1096         return rx;
1097 }
1098 EXPORT_SYMBOL(xfrm_stateonly_find);
1099
1100 struct xfrm_state *xfrm_state_lookup_byspi(struct net *net, __be32 spi,
1101                                               unsigned short family)
1102 {
1103         struct xfrm_state *x;
1104         struct xfrm_state_walk *w;
1105
1106         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1107         list_for_each_entry(w, &net->xfrm.state_all, all) {
1108                 x = container_of(w, struct xfrm_state, km);
1109                 if (x->props.family != family ||
1110                         x->id.spi != spi)
1111                         continue;
1112
1113                 xfrm_state_hold(x);
1114                 spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1115                 return x;
1116         }
1117         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1118         return NULL;
1119 }
1120 EXPORT_SYMBOL(xfrm_state_lookup_byspi);
1121
1122 static void __xfrm_state_insert(struct xfrm_state *x)
1123 {
1124         struct net *net = xs_net(x);
1125         unsigned int h;
1126
1127         list_add(&x->km.all, &net->xfrm.state_all);
1128
1129         h = xfrm_dst_hash(net, &x->id.daddr, &x->props.saddr,
1130                           x->props.reqid, x->props.family);
1131         hlist_add_head_rcu(&x->bydst, net->xfrm.state_bydst + h);
1132
1133         h = xfrm_src_hash(net, &x->id.daddr, &x->props.saddr, x->props.family);
1134         hlist_add_head_rcu(&x->bysrc, net->xfrm.state_bysrc + h);
1135
1136         if (x->id.spi) {
1137                 h = xfrm_spi_hash(net, &x->id.daddr, x->id.spi, x->id.proto,
1138                                   x->props.family);
1139
1140                 hlist_add_head_rcu(&x->byspi, net->xfrm.state_byspi + h);
1141         }
1142
1143         tasklet_hrtimer_start(&x->mtimer, ktime_set(1, 0), HRTIMER_MODE_REL);
1144         if (x->replay_maxage)
1145                 mod_timer(&x->rtimer, jiffies + x->replay_maxage);
1146
1147         net->xfrm.state_num++;
1148
1149         xfrm_hash_grow_check(net, x->bydst.next != NULL);
1150 }
1151
1152 /* net->xfrm.xfrm_state_lock is held */
1153 static void __xfrm_state_bump_genids(struct xfrm_state *xnew)
1154 {
1155         struct net *net = xs_net(xnew);
1156         unsigned short family = xnew->props.family;
1157         u32 reqid = xnew->props.reqid;
1158         struct xfrm_state *x;
1159         unsigned int h;
1160         u32 mark = xnew->mark.v & xnew->mark.m;
1161
1162         h = xfrm_dst_hash(net, &xnew->id.daddr, &xnew->props.saddr, reqid, family);
1163         hlist_for_each_entry(x, net->xfrm.state_bydst+h, bydst) {
1164                 if (x->props.family     == family &&
1165                     x->props.reqid      == reqid &&
1166                     (mark & x->mark.m) == x->mark.v &&
1167                     xfrm_addr_equal(&x->id.daddr, &xnew->id.daddr, family) &&
1168                     xfrm_addr_equal(&x->props.saddr, &xnew->props.saddr, family))
1169                         x->genid++;
1170         }
1171 }
1172
1173 void xfrm_state_insert(struct xfrm_state *x)
1174 {
1175         struct net *net = xs_net(x);
1176
1177         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1178         __xfrm_state_bump_genids(x);
1179         __xfrm_state_insert(x);
1180         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1181 }
1182 EXPORT_SYMBOL(xfrm_state_insert);
1183
1184 /* net->xfrm.xfrm_state_lock is held */
1185 static struct xfrm_state *__find_acq_core(struct net *net,
1186                                           const struct xfrm_mark *m,
1187                                           unsigned short family, u8 mode,
1188                                           u32 reqid, u8 proto,
1189                                           const xfrm_address_t *daddr,
1190                                           const xfrm_address_t *saddr,
1191                                           int create)
1192 {
1193         unsigned int h = xfrm_dst_hash(net, daddr, saddr, reqid, family);
1194         struct xfrm_state *x;
1195         u32 mark = m->v & m->m;
1196
1197         hlist_for_each_entry(x, net->xfrm.state_bydst+h, bydst) {
1198                 if (x->props.reqid  != reqid ||
1199                     x->props.mode   != mode ||
1200                     x->props.family != family ||
1201                     x->km.state     != XFRM_STATE_ACQ ||
1202                     x->id.spi       != 0 ||
1203                     x->id.proto     != proto ||
1204                     (mark & x->mark.m) != x->mark.v ||
1205                     !xfrm_addr_equal(&x->id.daddr, daddr, family) ||
1206                     !xfrm_addr_equal(&x->props.saddr, saddr, family))
1207                         continue;
1208
1209                 xfrm_state_hold(x);
1210                 return x;
1211         }
1212
1213         if (!create)
1214                 return NULL;
1215
1216         x = xfrm_state_alloc(net);
1217         if (likely(x)) {
1218                 switch (family) {
1219                 case AF_INET:
1220                         x->sel.daddr.a4 = daddr->a4;
1221                         x->sel.saddr.a4 = saddr->a4;
1222                         x->sel.prefixlen_d = 32;
1223                         x->sel.prefixlen_s = 32;
1224                         x->props.saddr.a4 = saddr->a4;
1225                         x->id.daddr.a4 = daddr->a4;
1226                         break;
1227
1228                 case AF_INET6:
1229                         x->sel.daddr.in6 = daddr->in6;
1230                         x->sel.saddr.in6 = saddr->in6;
1231                         x->sel.prefixlen_d = 128;
1232                         x->sel.prefixlen_s = 128;
1233                         x->props.saddr.in6 = saddr->in6;
1234                         x->id.daddr.in6 = daddr->in6;
1235                         break;
1236                 }
1237
1238                 x->km.state = XFRM_STATE_ACQ;
1239                 x->id.proto = proto;
1240                 x->props.family = family;
1241                 x->props.mode = mode;
1242                 x->props.reqid = reqid;
1243                 x->mark.v = m->v;
1244                 x->mark.m = m->m;
1245                 x->lft.hard_add_expires_seconds = net->xfrm.sysctl_acq_expires;
1246                 xfrm_state_hold(x);
1247                 tasklet_hrtimer_start(&x->mtimer, ktime_set(net->xfrm.sysctl_acq_expires, 0), HRTIMER_MODE_REL);
1248                 list_add(&x->km.all, &net->xfrm.state_all);
1249                 hlist_add_head_rcu(&x->bydst, net->xfrm.state_bydst + h);
1250                 h = xfrm_src_hash(net, daddr, saddr, family);
1251                 hlist_add_head_rcu(&x->bysrc, net->xfrm.state_bysrc + h);
1252
1253                 net->xfrm.state_num++;
1254
1255                 xfrm_hash_grow_check(net, x->bydst.next != NULL);
1256         }
1257
1258         return x;
1259 }
1260
1261 static struct xfrm_state *__xfrm_find_acq_byseq(struct net *net, u32 mark, u32 seq);
1262
1263 int xfrm_state_add(struct xfrm_state *x)
1264 {
1265         struct net *net = xs_net(x);
1266         struct xfrm_state *x1, *to_put;
1267         int family;
1268         int err;
1269         u32 mark = x->mark.v & x->mark.m;
1270         int use_spi = xfrm_id_proto_match(x->id.proto, IPSEC_PROTO_ANY);
1271
1272         family = x->props.family;
1273
1274         to_put = NULL;
1275
1276         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1277
1278         x1 = __xfrm_state_locate(x, use_spi, family);
1279         if (x1) {
1280                 to_put = x1;
1281                 x1 = NULL;
1282                 err = -EEXIST;
1283                 goto out;
1284         }
1285
1286         if (use_spi && x->km.seq) {
1287                 x1 = __xfrm_find_acq_byseq(net, mark, x->km.seq);
1288                 if (x1 && ((x1->id.proto != x->id.proto) ||
1289                     !xfrm_addr_equal(&x1->id.daddr, &x->id.daddr, family))) {
1290                         to_put = x1;
1291                         x1 = NULL;
1292                 }
1293         }
1294
1295         if (use_spi && !x1)
1296                 x1 = __find_acq_core(net, &x->mark, family, x->props.mode,
1297                                      x->props.reqid, x->id.proto,
1298                                      &x->id.daddr, &x->props.saddr, 0);
1299
1300         __xfrm_state_bump_genids(x);
1301         __xfrm_state_insert(x);
1302         err = 0;
1303
1304 out:
1305         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1306
1307         if (x1) {
1308                 xfrm_state_delete(x1);
1309                 xfrm_state_put(x1);
1310         }
1311
1312         if (to_put)
1313                 xfrm_state_put(to_put);
1314
1315         return err;
1316 }
1317 EXPORT_SYMBOL(xfrm_state_add);
1318
1319 #ifdef CONFIG_XFRM_MIGRATE
1320 static struct xfrm_state *xfrm_state_clone(struct xfrm_state *orig,
1321                                            struct xfrm_encap_tmpl *encap)
1322 {
1323         struct net *net = xs_net(orig);
1324         struct xfrm_state *x = xfrm_state_alloc(net);
1325         if (!x)
1326                 goto out;
1327
1328         memcpy(&x->id, &orig->id, sizeof(x->id));
1329         memcpy(&x->sel, &orig->sel, sizeof(x->sel));
1330         memcpy(&x->lft, &orig->lft, sizeof(x->lft));
1331         x->props.mode = orig->props.mode;
1332         x->props.replay_window = orig->props.replay_window;
1333         x->props.reqid = orig->props.reqid;
1334         x->props.family = orig->props.family;
1335         x->props.saddr = orig->props.saddr;
1336
1337         if (orig->aalg) {
1338                 x->aalg = xfrm_algo_auth_clone(orig->aalg);
1339                 if (!x->aalg)
1340                         goto error;
1341         }
1342         x->props.aalgo = orig->props.aalgo;
1343
1344         if (orig->aead) {
1345                 x->aead = xfrm_algo_aead_clone(orig->aead);
1346                 if (!x->aead)
1347                         goto error;
1348         }
1349         if (orig->ealg) {
1350                 x->ealg = xfrm_algo_clone(orig->ealg);
1351                 if (!x->ealg)
1352                         goto error;
1353         }
1354         x->props.ealgo = orig->props.ealgo;
1355
1356         if (orig->calg) {
1357                 x->calg = xfrm_algo_clone(orig->calg);
1358                 if (!x->calg)
1359                         goto error;
1360         }
1361         x->props.calgo = orig->props.calgo;
1362
1363         if (encap || orig->encap) {
1364                 if (encap)
1365                         x->encap = kmemdup(encap, sizeof(*x->encap),
1366                                         GFP_KERNEL);
1367                 else
1368                         x->encap = kmemdup(orig->encap, sizeof(*x->encap),
1369                                         GFP_KERNEL);
1370
1371                 if (!x->encap)
1372                         goto error;
1373         }
1374
1375         if (orig->coaddr) {
1376                 x->coaddr = kmemdup(orig->coaddr, sizeof(*x->coaddr),
1377                                     GFP_KERNEL);
1378                 if (!x->coaddr)
1379                         goto error;
1380         }
1381
1382         if (orig->replay_esn) {
1383                 if (xfrm_replay_clone(x, orig))
1384                         goto error;
1385         }
1386
1387         memcpy(&x->mark, &orig->mark, sizeof(x->mark));
1388
1389         if (xfrm_init_state(x) < 0)
1390                 goto error;
1391
1392         x->props.flags = orig->props.flags;
1393         x->props.extra_flags = orig->props.extra_flags;
1394
1395         x->tfcpad = orig->tfcpad;
1396         x->replay_maxdiff = orig->replay_maxdiff;
1397         x->replay_maxage = orig->replay_maxage;
1398         x->curlft.add_time = orig->curlft.add_time;
1399         x->km.state = orig->km.state;
1400         x->km.seq = orig->km.seq;
1401         x->replay = orig->replay;
1402         x->preplay = orig->preplay;
1403
1404         return x;
1405
1406  error:
1407         xfrm_state_put(x);
1408 out:
1409         return NULL;
1410 }
1411
1412 struct xfrm_state *xfrm_migrate_state_find(struct xfrm_migrate *m, struct net *net)
1413 {
1414         unsigned int h;
1415         struct xfrm_state *x = NULL;
1416
1417         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1418
1419         if (m->reqid) {
1420                 h = xfrm_dst_hash(net, &m->old_daddr, &m->old_saddr,
1421                                   m->reqid, m->old_family);
1422                 hlist_for_each_entry(x, net->xfrm.state_bydst+h, bydst) {
1423                         if (x->props.mode != m->mode ||
1424                             x->id.proto != m->proto)
1425                                 continue;
1426                         if (m->reqid && x->props.reqid != m->reqid)
1427                                 continue;
1428                         if (!xfrm_addr_equal(&x->id.daddr, &m->old_daddr,
1429                                              m->old_family) ||
1430                             !xfrm_addr_equal(&x->props.saddr, &m->old_saddr,
1431                                              m->old_family))
1432                                 continue;
1433                         xfrm_state_hold(x);
1434                         break;
1435                 }
1436         } else {
1437                 h = xfrm_src_hash(net, &m->old_daddr, &m->old_saddr,
1438                                   m->old_family);
1439                 hlist_for_each_entry(x, net->xfrm.state_bysrc+h, bysrc) {
1440                         if (x->props.mode != m->mode ||
1441                             x->id.proto != m->proto)
1442                                 continue;
1443                         if (!xfrm_addr_equal(&x->id.daddr, &m->old_daddr,
1444                                              m->old_family) ||
1445                             !xfrm_addr_equal(&x->props.saddr, &m->old_saddr,
1446                                              m->old_family))
1447                                 continue;
1448                         xfrm_state_hold(x);
1449                         break;
1450                 }
1451         }
1452
1453         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1454
1455         return x;
1456 }
1457 EXPORT_SYMBOL(xfrm_migrate_state_find);
1458
1459 struct xfrm_state *xfrm_state_migrate(struct xfrm_state *x,
1460                                       struct xfrm_migrate *m,
1461                                       struct xfrm_encap_tmpl *encap)
1462 {
1463         struct xfrm_state *xc;
1464
1465         xc = xfrm_state_clone(x, encap);
1466         if (!xc)
1467                 return NULL;
1468
1469         memcpy(&xc->id.daddr, &m->new_daddr, sizeof(xc->id.daddr));
1470         memcpy(&xc->props.saddr, &m->new_saddr, sizeof(xc->props.saddr));
1471
1472         /* add state */
1473         if (xfrm_addr_equal(&x->id.daddr, &m->new_daddr, m->new_family)) {
1474                 /* a care is needed when the destination address of the
1475                    state is to be updated as it is a part of triplet */
1476                 xfrm_state_insert(xc);
1477         } else {
1478                 if (xfrm_state_add(xc) < 0)
1479                         goto error;
1480         }
1481
1482         return xc;
1483 error:
1484         xfrm_state_put(xc);
1485         return NULL;
1486 }
1487 EXPORT_SYMBOL(xfrm_state_migrate);
1488 #endif
1489
1490 int xfrm_state_update(struct xfrm_state *x)
1491 {
1492         struct xfrm_state *x1, *to_put;
1493         int err;
1494         int use_spi = xfrm_id_proto_match(x->id.proto, IPSEC_PROTO_ANY);
1495         struct net *net = xs_net(x);
1496
1497         to_put = NULL;
1498
1499         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1500         x1 = __xfrm_state_locate(x, use_spi, x->props.family);
1501
1502         err = -ESRCH;
1503         if (!x1)
1504                 goto out;
1505
1506         if (xfrm_state_kern(x1)) {
1507                 to_put = x1;
1508                 err = -EEXIST;
1509                 goto out;
1510         }
1511
1512         if (x1->km.state == XFRM_STATE_ACQ) {
1513                 __xfrm_state_insert(x);
1514                 x = NULL;
1515         }
1516         err = 0;
1517
1518 out:
1519         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1520
1521         if (to_put)
1522                 xfrm_state_put(to_put);
1523
1524         if (err)
1525                 return err;
1526
1527         if (!x) {
1528                 xfrm_state_delete(x1);
1529                 xfrm_state_put(x1);
1530                 return 0;
1531         }
1532
1533         err = -EINVAL;
1534         spin_lock_bh(&x1->lock);
1535         if (likely(x1->km.state == XFRM_STATE_VALID)) {
1536                 if (x->encap && x1->encap)
1537                         memcpy(x1->encap, x->encap, sizeof(*x1->encap));
1538                 if (x->coaddr && x1->coaddr) {
1539                         memcpy(x1->coaddr, x->coaddr, sizeof(*x1->coaddr));
1540                 }
1541                 if (!use_spi && memcmp(&x1->sel, &x->sel, sizeof(x1->sel)))
1542                         memcpy(&x1->sel, &x->sel, sizeof(x1->sel));
1543                 memcpy(&x1->lft, &x->lft, sizeof(x1->lft));
1544                 x1->km.dying = 0;
1545
1546                 tasklet_hrtimer_start(&x1->mtimer, ktime_set(1, 0), HRTIMER_MODE_REL);
1547                 if (x1->curlft.use_time)
1548                         xfrm_state_check_expire(x1);
1549
1550                 err = 0;
1551                 x->km.state = XFRM_STATE_DEAD;
1552                 __xfrm_state_put(x);
1553         }
1554         spin_unlock_bh(&x1->lock);
1555
1556         xfrm_state_put(x1);
1557
1558         return err;
1559 }
1560 EXPORT_SYMBOL(xfrm_state_update);
1561
1562 int xfrm_state_check_expire(struct xfrm_state *x)
1563 {
1564         if (!x->curlft.use_time)
1565                 x->curlft.use_time = get_seconds();
1566
1567         if (x->curlft.bytes >= x->lft.hard_byte_limit ||
1568             x->curlft.packets >= x->lft.hard_packet_limit) {
1569                 x->km.state = XFRM_STATE_EXPIRED;
1570                 tasklet_hrtimer_start(&x->mtimer, 0, HRTIMER_MODE_REL);
1571                 return -EINVAL;
1572         }
1573
1574         if (!x->km.dying &&
1575             (x->curlft.bytes >= x->lft.soft_byte_limit ||
1576              x->curlft.packets >= x->lft.soft_packet_limit)) {
1577                 x->km.dying = 1;
1578                 km_state_expired(x, 0, 0);
1579         }
1580         return 0;
1581 }
1582 EXPORT_SYMBOL(xfrm_state_check_expire);
1583
1584 struct xfrm_state *
1585 xfrm_state_lookup(struct net *net, u32 mark, const xfrm_address_t *daddr, __be32 spi,
1586                   u8 proto, unsigned short family)
1587 {
1588         struct xfrm_state *x;
1589
1590         rcu_read_lock();
1591         x = __xfrm_state_lookup(net, mark, daddr, spi, proto, family);
1592         rcu_read_unlock();
1593         return x;
1594 }
1595 EXPORT_SYMBOL(xfrm_state_lookup);
1596
1597 struct xfrm_state *
1598 xfrm_state_lookup_byaddr(struct net *net, u32 mark,
1599                          const xfrm_address_t *daddr, const xfrm_address_t *saddr,
1600                          u8 proto, unsigned short family)
1601 {
1602         struct xfrm_state *x;
1603
1604         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1605         x = __xfrm_state_lookup_byaddr(net, mark, daddr, saddr, proto, family);
1606         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1607         return x;
1608 }
1609 EXPORT_SYMBOL(xfrm_state_lookup_byaddr);
1610
1611 struct xfrm_state *
1612 xfrm_find_acq(struct net *net, const struct xfrm_mark *mark, u8 mode, u32 reqid,
1613               u8 proto, const xfrm_address_t *daddr,
1614               const xfrm_address_t *saddr, int create, unsigned short family)
1615 {
1616         struct xfrm_state *x;
1617
1618         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1619         x = __find_acq_core(net, mark, family, mode, reqid, proto, daddr, saddr, create);
1620         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1621
1622         return x;
1623 }
1624 EXPORT_SYMBOL(xfrm_find_acq);
1625
1626 #ifdef CONFIG_XFRM_SUB_POLICY
1627 int
1628 xfrm_tmpl_sort(struct xfrm_tmpl **dst, struct xfrm_tmpl **src, int n,
1629                unsigned short family, struct net *net)
1630 {
1631         int i;
1632         int err = 0;
1633         struct xfrm_state_afinfo *afinfo = xfrm_state_get_afinfo(family);
1634         if (!afinfo)
1635                 return -EAFNOSUPPORT;
1636
1637         spin_lock_bh(&net->xfrm.xfrm_state_lock); /*FIXME*/
1638         if (afinfo->tmpl_sort)
1639                 err = afinfo->tmpl_sort(dst, src, n);
1640         else
1641                 for (i = 0; i < n; i++)
1642                         dst[i] = src[i];
1643         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1644         rcu_read_unlock();
1645         return err;
1646 }
1647 EXPORT_SYMBOL(xfrm_tmpl_sort);
1648
1649 int
1650 xfrm_state_sort(struct xfrm_state **dst, struct xfrm_state **src, int n,
1651                 unsigned short family)
1652 {
1653         int i;
1654         int err = 0;
1655         struct xfrm_state_afinfo *afinfo = xfrm_state_get_afinfo(family);
1656         struct net *net = xs_net(*src);
1657
1658         if (!afinfo)
1659                 return -EAFNOSUPPORT;
1660
1661         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1662         if (afinfo->state_sort)
1663                 err = afinfo->state_sort(dst, src, n);
1664         else
1665                 for (i = 0; i < n; i++)
1666                         dst[i] = src[i];
1667         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1668         rcu_read_unlock();
1669         return err;
1670 }
1671 EXPORT_SYMBOL(xfrm_state_sort);
1672 #endif
1673
1674 /* Silly enough, but I'm lazy to build resolution list */
1675
1676 static struct xfrm_state *__xfrm_find_acq_byseq(struct net *net, u32 mark, u32 seq)
1677 {
1678         int i;
1679
1680         for (i = 0; i <= net->xfrm.state_hmask; i++) {
1681                 struct xfrm_state *x;
1682
1683                 hlist_for_each_entry(x, net->xfrm.state_bydst+i, bydst) {
1684                         if (x->km.seq == seq &&
1685                             (mark & x->mark.m) == x->mark.v &&
1686                             x->km.state == XFRM_STATE_ACQ) {
1687                                 xfrm_state_hold(x);
1688                                 return x;
1689                         }
1690                 }
1691         }
1692         return NULL;
1693 }
1694
1695 struct xfrm_state *xfrm_find_acq_byseq(struct net *net, u32 mark, u32 seq)
1696 {
1697         struct xfrm_state *x;
1698
1699         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1700         x = __xfrm_find_acq_byseq(net, mark, seq);
1701         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1702         return x;
1703 }
1704 EXPORT_SYMBOL(xfrm_find_acq_byseq);
1705
1706 u32 xfrm_get_acqseq(void)
1707 {
1708         u32 res;
1709         static atomic_t acqseq;
1710
1711         do {
1712                 res = atomic_inc_return(&acqseq);
1713         } while (!res);
1714
1715         return res;
1716 }
1717 EXPORT_SYMBOL(xfrm_get_acqseq);
1718
1719 int verify_spi_info(u8 proto, u32 min, u32 max)
1720 {
1721         switch (proto) {
1722         case IPPROTO_AH:
1723         case IPPROTO_ESP:
1724                 break;
1725
1726         case IPPROTO_COMP:
1727                 /* IPCOMP spi is 16-bits. */
1728                 if (max >= 0x10000)
1729                         return -EINVAL;
1730                 break;
1731
1732         default:
1733                 return -EINVAL;
1734         }
1735
1736         if (min > max)
1737                 return -EINVAL;
1738
1739         return 0;
1740 }
1741 EXPORT_SYMBOL(verify_spi_info);
1742
1743 int xfrm_alloc_spi(struct xfrm_state *x, u32 low, u32 high)
1744 {
1745         struct net *net = xs_net(x);
1746         unsigned int h;
1747         struct xfrm_state *x0;
1748         int err = -ENOENT;
1749         __be32 minspi = htonl(low);
1750         __be32 maxspi = htonl(high);
1751         u32 mark = x->mark.v & x->mark.m;
1752
1753         spin_lock_bh(&x->lock);
1754         if (x->km.state == XFRM_STATE_DEAD)
1755                 goto unlock;
1756
1757         err = 0;
1758         if (x->id.spi)
1759                 goto unlock;
1760
1761         err = -ENOENT;
1762
1763         if (minspi == maxspi) {
1764                 x0 = xfrm_state_lookup(net, mark, &x->id.daddr, minspi, x->id.proto, x->props.family);
1765                 if (x0) {
1766                         xfrm_state_put(x0);
1767                         goto unlock;
1768                 }
1769                 x->id.spi = minspi;
1770         } else {
1771                 u32 spi = 0;
1772                 for (h = 0; h < high-low+1; h++) {
1773                         spi = low + prandom_u32()%(high-low+1);
1774                         x0 = xfrm_state_lookup(net, mark, &x->id.daddr, htonl(spi), x->id.proto, x->props.family);
1775                         if (x0 == NULL) {
1776                                 x->id.spi = htonl(spi);
1777                                 break;
1778                         }
1779                         xfrm_state_put(x0);
1780                 }
1781         }
1782         if (x->id.spi) {
1783                 spin_lock_bh(&net->xfrm.xfrm_state_lock);
1784                 h = xfrm_spi_hash(net, &x->id.daddr, x->id.spi, x->id.proto, x->props.family);
1785                 hlist_add_head_rcu(&x->byspi, net->xfrm.state_byspi + h);
1786                 spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1787
1788                 err = 0;
1789         }
1790
1791 unlock:
1792         spin_unlock_bh(&x->lock);
1793
1794         return err;
1795 }
1796 EXPORT_SYMBOL(xfrm_alloc_spi);
1797
1798 static bool __xfrm_state_filter_match(struct xfrm_state *x,
1799                                       struct xfrm_address_filter *filter)
1800 {
1801         if (filter) {
1802                 if ((filter->family == AF_INET ||
1803                      filter->family == AF_INET6) &&
1804                     x->props.family != filter->family)
1805                         return false;
1806
1807                 return addr_match(&x->props.saddr, &filter->saddr,
1808                                   filter->splen) &&
1809                        addr_match(&x->id.daddr, &filter->daddr,
1810                                   filter->dplen);
1811         }
1812         return true;
1813 }
1814
1815 int xfrm_state_walk(struct net *net, struct xfrm_state_walk *walk,
1816                     int (*func)(struct xfrm_state *, int, void*),
1817                     void *data)
1818 {
1819         struct xfrm_state *state;
1820         struct xfrm_state_walk *x;
1821         int err = 0;
1822
1823         if (walk->seq != 0 && list_empty(&walk->all))
1824                 return 0;
1825
1826         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1827         if (list_empty(&walk->all))
1828                 x = list_first_entry(&net->xfrm.state_all, struct xfrm_state_walk, all);
1829         else
1830                 x = list_first_entry(&walk->all, struct xfrm_state_walk, all);
1831         list_for_each_entry_from(x, &net->xfrm.state_all, all) {
1832                 if (x->state == XFRM_STATE_DEAD)
1833                         continue;
1834                 state = container_of(x, struct xfrm_state, km);
1835                 if (!xfrm_id_proto_match(state->id.proto, walk->proto))
1836                         continue;
1837                 if (!__xfrm_state_filter_match(state, walk->filter))
1838                         continue;
1839                 err = func(state, walk->seq, data);
1840                 if (err) {
1841                         list_move_tail(&walk->all, &x->all);
1842                         goto out;
1843                 }
1844                 walk->seq++;
1845         }
1846         if (walk->seq == 0) {
1847                 err = -ENOENT;
1848                 goto out;
1849         }
1850         list_del_init(&walk->all);
1851 out:
1852         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1853         return err;
1854 }
1855 EXPORT_SYMBOL(xfrm_state_walk);
1856
1857 void xfrm_state_walk_init(struct xfrm_state_walk *walk, u8 proto,
1858                           struct xfrm_address_filter *filter)
1859 {
1860         INIT_LIST_HEAD(&walk->all);
1861         walk->proto = proto;
1862         walk->state = XFRM_STATE_DEAD;
1863         walk->seq = 0;
1864         walk->filter = filter;
1865 }
1866 EXPORT_SYMBOL(xfrm_state_walk_init);
1867
1868 void xfrm_state_walk_done(struct xfrm_state_walk *walk, struct net *net)
1869 {
1870         kfree(walk->filter);
1871
1872         if (list_empty(&walk->all))
1873                 return;
1874
1875         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1876         list_del(&walk->all);
1877         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1878 }
1879 EXPORT_SYMBOL(xfrm_state_walk_done);
1880
1881 static void xfrm_replay_timer_handler(struct timer_list *t)
1882 {
1883         struct xfrm_state *x = from_timer(x, t, rtimer);
1884
1885         spin_lock(&x->lock);
1886
1887         if (x->km.state == XFRM_STATE_VALID) {
1888                 if (xfrm_aevent_is_on(xs_net(x)))
1889                         x->repl->notify(x, XFRM_REPLAY_TIMEOUT);
1890                 else
1891                         x->xflags |= XFRM_TIME_DEFER;
1892         }
1893
1894         spin_unlock(&x->lock);
1895 }
1896
1897 static LIST_HEAD(xfrm_km_list);
1898
1899 void km_policy_notify(struct xfrm_policy *xp, int dir, const struct km_event *c)
1900 {
1901         struct xfrm_mgr *km;
1902
1903         rcu_read_lock();
1904         list_for_each_entry_rcu(km, &xfrm_km_list, list)
1905                 if (km->notify_policy)
1906                         km->notify_policy(xp, dir, c);
1907         rcu_read_unlock();
1908 }
1909
1910 void km_state_notify(struct xfrm_state *x, const struct km_event *c)
1911 {
1912         struct xfrm_mgr *km;
1913         rcu_read_lock();
1914         list_for_each_entry_rcu(km, &xfrm_km_list, list)
1915                 if (km->notify)
1916                         km->notify(x, c);
1917         rcu_read_unlock();
1918 }
1919
1920 EXPORT_SYMBOL(km_policy_notify);
1921 EXPORT_SYMBOL(km_state_notify);
1922
1923 void km_state_expired(struct xfrm_state *x, int hard, u32 portid)
1924 {
1925         struct km_event c;
1926
1927         c.data.hard = hard;
1928         c.portid = portid;
1929         c.event = XFRM_MSG_EXPIRE;
1930         km_state_notify(x, &c);
1931 }
1932
1933 EXPORT_SYMBOL(km_state_expired);
1934 /*
1935  * We send to all registered managers regardless of failure
1936  * We are happy with one success
1937 */
1938 int km_query(struct xfrm_state *x, struct xfrm_tmpl *t, struct xfrm_policy *pol)
1939 {
1940         int err = -EINVAL, acqret;
1941         struct xfrm_mgr *km;
1942
1943         rcu_read_lock();
1944         list_for_each_entry_rcu(km, &xfrm_km_list, list) {
1945                 acqret = km->acquire(x, t, pol);
1946                 if (!acqret)
1947                         err = acqret;
1948         }
1949         rcu_read_unlock();
1950         return err;
1951 }
1952 EXPORT_SYMBOL(km_query);
1953
1954 int km_new_mapping(struct xfrm_state *x, xfrm_address_t *ipaddr, __be16 sport)
1955 {
1956         int err = -EINVAL;
1957         struct xfrm_mgr *km;
1958
1959         rcu_read_lock();
1960         list_for_each_entry_rcu(km, &xfrm_km_list, list) {
1961                 if (km->new_mapping)
1962                         err = km->new_mapping(x, ipaddr, sport);
1963                 if (!err)
1964                         break;
1965         }
1966         rcu_read_unlock();
1967         return err;
1968 }
1969 EXPORT_SYMBOL(km_new_mapping);
1970
1971 void km_policy_expired(struct xfrm_policy *pol, int dir, int hard, u32 portid)
1972 {
1973         struct km_event c;
1974
1975         c.data.hard = hard;
1976         c.portid = portid;
1977         c.event = XFRM_MSG_POLEXPIRE;
1978         km_policy_notify(pol, dir, &c);
1979 }
1980 EXPORT_SYMBOL(km_policy_expired);
1981
1982 #ifdef CONFIG_XFRM_MIGRATE
1983 int km_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
1984                const struct xfrm_migrate *m, int num_migrate,
1985                const struct xfrm_kmaddress *k,
1986                const struct xfrm_encap_tmpl *encap)
1987 {
1988         int err = -EINVAL;
1989         int ret;
1990         struct xfrm_mgr *km;
1991
1992         rcu_read_lock();
1993         list_for_each_entry_rcu(km, &xfrm_km_list, list) {
1994                 if (km->migrate) {
1995                         ret = km->migrate(sel, dir, type, m, num_migrate, k,
1996                                           encap);
1997                         if (!ret)
1998                                 err = ret;
1999                 }
2000         }
2001         rcu_read_unlock();
2002         return err;
2003 }
2004 EXPORT_SYMBOL(km_migrate);
2005 #endif
2006
2007 int km_report(struct net *net, u8 proto, struct xfrm_selector *sel, xfrm_address_t *addr)
2008 {
2009         int err = -EINVAL;
2010         int ret;
2011         struct xfrm_mgr *km;
2012
2013         rcu_read_lock();
2014         list_for_each_entry_rcu(km, &xfrm_km_list, list) {
2015                 if (km->report) {
2016                         ret = km->report(net, proto, sel, addr);
2017                         if (!ret)
2018                                 err = ret;
2019                 }
2020         }
2021         rcu_read_unlock();
2022         return err;
2023 }
2024 EXPORT_SYMBOL(km_report);
2025
2026 bool km_is_alive(const struct km_event *c)
2027 {
2028         struct xfrm_mgr *km;
2029         bool is_alive = false;
2030
2031         rcu_read_lock();
2032         list_for_each_entry_rcu(km, &xfrm_km_list, list) {
2033                 if (km->is_alive && km->is_alive(c)) {
2034                         is_alive = true;
2035                         break;
2036                 }
2037         }
2038         rcu_read_unlock();
2039
2040         return is_alive;
2041 }
2042 EXPORT_SYMBOL(km_is_alive);
2043
2044 int xfrm_user_policy(struct sock *sk, int optname, u8 __user *optval, int optlen)
2045 {
2046         int err;
2047         u8 *data;
2048         struct xfrm_mgr *km;
2049         struct xfrm_policy *pol = NULL;
2050
2051         if (optlen <= 0 || optlen > PAGE_SIZE)
2052                 return -EMSGSIZE;
2053
2054         data = memdup_user(optval, optlen);
2055         if (IS_ERR(data))
2056                 return PTR_ERR(data);
2057
2058         err = -EINVAL;
2059         rcu_read_lock();
2060         list_for_each_entry_rcu(km, &xfrm_km_list, list) {
2061                 pol = km->compile_policy(sk, optname, data,
2062                                          optlen, &err);
2063                 if (err >= 0)
2064                         break;
2065         }
2066         rcu_read_unlock();
2067
2068         if (err >= 0) {
2069                 xfrm_sk_policy_insert(sk, err, pol);
2070                 xfrm_pol_put(pol);
2071                 __sk_dst_reset(sk);
2072                 err = 0;
2073         }
2074
2075         kfree(data);
2076         return err;
2077 }
2078 EXPORT_SYMBOL(xfrm_user_policy);
2079
2080 static DEFINE_SPINLOCK(xfrm_km_lock);
2081
2082 int xfrm_register_km(struct xfrm_mgr *km)
2083 {
2084         spin_lock_bh(&xfrm_km_lock);
2085         list_add_tail_rcu(&km->list, &xfrm_km_list);
2086         spin_unlock_bh(&xfrm_km_lock);
2087         return 0;
2088 }
2089 EXPORT_SYMBOL(xfrm_register_km);
2090
2091 int xfrm_unregister_km(struct xfrm_mgr *km)
2092 {
2093         spin_lock_bh(&xfrm_km_lock);
2094         list_del_rcu(&km->list);
2095         spin_unlock_bh(&xfrm_km_lock);
2096         synchronize_rcu();
2097         return 0;
2098 }
2099 EXPORT_SYMBOL(xfrm_unregister_km);
2100
2101 int xfrm_state_register_afinfo(struct xfrm_state_afinfo *afinfo)
2102 {
2103         int err = 0;
2104
2105         if (WARN_ON(afinfo->family >= NPROTO))
2106                 return -EAFNOSUPPORT;
2107
2108         spin_lock_bh(&xfrm_state_afinfo_lock);
2109         if (unlikely(xfrm_state_afinfo[afinfo->family] != NULL))
2110                 err = -EEXIST;
2111         else
2112                 rcu_assign_pointer(xfrm_state_afinfo[afinfo->family], afinfo);
2113         spin_unlock_bh(&xfrm_state_afinfo_lock);
2114         return err;
2115 }
2116 EXPORT_SYMBOL(xfrm_state_register_afinfo);
2117
2118 int xfrm_state_unregister_afinfo(struct xfrm_state_afinfo *afinfo)
2119 {
2120         int err = 0, family = afinfo->family;
2121
2122         if (WARN_ON(family >= NPROTO))
2123                 return -EAFNOSUPPORT;
2124
2125         spin_lock_bh(&xfrm_state_afinfo_lock);
2126         if (likely(xfrm_state_afinfo[afinfo->family] != NULL)) {
2127                 if (rcu_access_pointer(xfrm_state_afinfo[family]) != afinfo)
2128                         err = -EINVAL;
2129                 else
2130                         RCU_INIT_POINTER(xfrm_state_afinfo[afinfo->family], NULL);
2131         }
2132         spin_unlock_bh(&xfrm_state_afinfo_lock);
2133         synchronize_rcu();
2134         return err;
2135 }
2136 EXPORT_SYMBOL(xfrm_state_unregister_afinfo);
2137
2138 struct xfrm_state_afinfo *xfrm_state_afinfo_get_rcu(unsigned int family)
2139 {
2140         if (unlikely(family >= NPROTO))
2141                 return NULL;
2142
2143         return rcu_dereference(xfrm_state_afinfo[family]);
2144 }
2145
2146 struct xfrm_state_afinfo *xfrm_state_get_afinfo(unsigned int family)
2147 {
2148         struct xfrm_state_afinfo *afinfo;
2149         if (unlikely(family >= NPROTO))
2150                 return NULL;
2151         rcu_read_lock();
2152         afinfo = rcu_dereference(xfrm_state_afinfo[family]);
2153         if (unlikely(!afinfo))
2154                 rcu_read_unlock();
2155         return afinfo;
2156 }
2157
2158 /* Temporarily located here until net/xfrm/xfrm_tunnel.c is created */
2159 void xfrm_state_delete_tunnel(struct xfrm_state *x)
2160 {
2161         if (x->tunnel) {
2162                 struct xfrm_state *t = x->tunnel;
2163
2164                 if (atomic_read(&t->tunnel_users) == 2)
2165                         xfrm_state_delete(t);
2166                 atomic_dec(&t->tunnel_users);
2167                 xfrm_state_put(t);
2168                 x->tunnel = NULL;
2169         }
2170 }
2171 EXPORT_SYMBOL(xfrm_state_delete_tunnel);
2172
2173 int xfrm_state_mtu(struct xfrm_state *x, int mtu)
2174 {
2175         const struct xfrm_type *type = READ_ONCE(x->type);
2176
2177         if (x->km.state == XFRM_STATE_VALID &&
2178             type && type->get_mtu)
2179                 return type->get_mtu(x, mtu);
2180
2181         return mtu - x->props.header_len;
2182 }
2183
2184 int __xfrm_init_state(struct xfrm_state *x, bool init_replay, bool offload)
2185 {
2186         struct xfrm_state_afinfo *afinfo;
2187         struct xfrm_mode *inner_mode;
2188         int family = x->props.family;
2189         int err;
2190
2191         err = -EAFNOSUPPORT;
2192         afinfo = xfrm_state_get_afinfo(family);
2193         if (!afinfo)
2194                 goto error;
2195
2196         err = 0;
2197         if (afinfo->init_flags)
2198                 err = afinfo->init_flags(x);
2199
2200         rcu_read_unlock();
2201
2202         if (err)
2203                 goto error;
2204
2205         err = -EPROTONOSUPPORT;
2206
2207         if (x->sel.family != AF_UNSPEC) {
2208                 inner_mode = xfrm_get_mode(x->props.mode, x->sel.family);
2209                 if (inner_mode == NULL)
2210                         goto error;
2211
2212                 if (!(inner_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2213                     family != x->sel.family) {
2214                         xfrm_put_mode(inner_mode);
2215                         goto error;
2216                 }
2217
2218                 x->inner_mode = inner_mode;
2219         } else {
2220                 struct xfrm_mode *inner_mode_iaf;
2221                 int iafamily = AF_INET;
2222
2223                 inner_mode = xfrm_get_mode(x->props.mode, x->props.family);
2224                 if (inner_mode == NULL)
2225                         goto error;
2226
2227                 if (!(inner_mode->flags & XFRM_MODE_FLAG_TUNNEL)) {
2228                         xfrm_put_mode(inner_mode);
2229                         goto error;
2230                 }
2231                 x->inner_mode = inner_mode;
2232
2233                 if (x->props.family == AF_INET)
2234                         iafamily = AF_INET6;
2235
2236                 inner_mode_iaf = xfrm_get_mode(x->props.mode, iafamily);
2237                 if (inner_mode_iaf) {
2238                         if (inner_mode_iaf->flags & XFRM_MODE_FLAG_TUNNEL)
2239                                 x->inner_mode_iaf = inner_mode_iaf;
2240                         else
2241                                 xfrm_put_mode(inner_mode_iaf);
2242                 }
2243         }
2244
2245         x->type = xfrm_get_type(x->id.proto, family);
2246         if (x->type == NULL)
2247                 goto error;
2248
2249         x->type_offload = xfrm_get_type_offload(x->id.proto, family, offload);
2250
2251         err = x->type->init_state(x);
2252         if (err)
2253                 goto error;
2254
2255         x->outer_mode = xfrm_get_mode(x->props.mode, family);
2256         if (x->outer_mode == NULL) {
2257                 err = -EPROTONOSUPPORT;
2258                 goto error;
2259         }
2260
2261         if (init_replay) {
2262                 err = xfrm_init_replay(x);
2263                 if (err)
2264                         goto error;
2265         }
2266
2267         x->km.state = XFRM_STATE_VALID;
2268
2269 error:
2270         return err;
2271 }
2272
2273 EXPORT_SYMBOL(__xfrm_init_state);
2274
2275 int xfrm_init_state(struct xfrm_state *x)
2276 {
2277         return __xfrm_init_state(x, true, false);
2278 }
2279
2280 EXPORT_SYMBOL(xfrm_init_state);
2281
2282 int __net_init xfrm_state_init(struct net *net)
2283 {
2284         unsigned int sz;
2285
2286         INIT_LIST_HEAD(&net->xfrm.state_all);
2287
2288         sz = sizeof(struct hlist_head) * 8;
2289
2290         net->xfrm.state_bydst = xfrm_hash_alloc(sz);
2291         if (!net->xfrm.state_bydst)
2292                 goto out_bydst;
2293         net->xfrm.state_bysrc = xfrm_hash_alloc(sz);
2294         if (!net->xfrm.state_bysrc)
2295                 goto out_bysrc;
2296         net->xfrm.state_byspi = xfrm_hash_alloc(sz);
2297         if (!net->xfrm.state_byspi)
2298                 goto out_byspi;
2299         net->xfrm.state_hmask = ((sz / sizeof(struct hlist_head)) - 1);
2300
2301         net->xfrm.state_num = 0;
2302         INIT_WORK(&net->xfrm.state_hash_work, xfrm_hash_resize);
2303         spin_lock_init(&net->xfrm.xfrm_state_lock);
2304         return 0;
2305
2306 out_byspi:
2307         xfrm_hash_free(net->xfrm.state_bysrc, sz);
2308 out_bysrc:
2309         xfrm_hash_free(net->xfrm.state_bydst, sz);
2310 out_bydst:
2311         return -ENOMEM;
2312 }
2313
2314 void xfrm_state_fini(struct net *net)
2315 {
2316         unsigned int sz;
2317
2318         flush_work(&net->xfrm.state_hash_work);
2319         xfrm_state_flush(net, IPSEC_PROTO_ANY, false);
2320         flush_work(&xfrm_state_gc_work);
2321
2322         WARN_ON(!list_empty(&net->xfrm.state_all));
2323
2324         sz = (net->xfrm.state_hmask + 1) * sizeof(struct hlist_head);
2325         WARN_ON(!hlist_empty(net->xfrm.state_byspi));
2326         xfrm_hash_free(net->xfrm.state_byspi, sz);
2327         WARN_ON(!hlist_empty(net->xfrm.state_bysrc));
2328         xfrm_hash_free(net->xfrm.state_bysrc, sz);
2329         WARN_ON(!hlist_empty(net->xfrm.state_bydst));
2330         xfrm_hash_free(net->xfrm.state_bydst, sz);
2331 }
2332
2333 #ifdef CONFIG_AUDITSYSCALL
2334 static void xfrm_audit_helper_sainfo(struct xfrm_state *x,
2335                                      struct audit_buffer *audit_buf)
2336 {
2337         struct xfrm_sec_ctx *ctx = x->security;
2338         u32 spi = ntohl(x->id.spi);
2339
2340         if (ctx)
2341                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2342                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2343
2344         switch (x->props.family) {
2345         case AF_INET:
2346                 audit_log_format(audit_buf, " src=%pI4 dst=%pI4",
2347                                  &x->props.saddr.a4, &x->id.daddr.a4);
2348                 break;
2349         case AF_INET6:
2350                 audit_log_format(audit_buf, " src=%pI6 dst=%pI6",
2351                                  x->props.saddr.a6, x->id.daddr.a6);
2352                 break;
2353         }
2354
2355         audit_log_format(audit_buf, " spi=%u(0x%x)", spi, spi);
2356 }
2357
2358 static void xfrm_audit_helper_pktinfo(struct sk_buff *skb, u16 family,
2359                                       struct audit_buffer *audit_buf)
2360 {
2361         const struct iphdr *iph4;
2362         const struct ipv6hdr *iph6;
2363
2364         switch (family) {
2365         case AF_INET:
2366                 iph4 = ip_hdr(skb);
2367                 audit_log_format(audit_buf, " src=%pI4 dst=%pI4",
2368                                  &iph4->saddr, &iph4->daddr);
2369                 break;
2370         case AF_INET6:
2371                 iph6 = ipv6_hdr(skb);
2372                 audit_log_format(audit_buf,
2373                                  " src=%pI6 dst=%pI6 flowlbl=0x%x%02x%02x",
2374                                  &iph6->saddr, &iph6->daddr,
2375                                  iph6->flow_lbl[0] & 0x0f,
2376                                  iph6->flow_lbl[1],
2377                                  iph6->flow_lbl[2]);
2378                 break;
2379         }
2380 }
2381
2382 void xfrm_audit_state_add(struct xfrm_state *x, int result, bool task_valid)
2383 {
2384         struct audit_buffer *audit_buf;
2385
2386         audit_buf = xfrm_audit_start("SAD-add");
2387         if (audit_buf == NULL)
2388                 return;
2389         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
2390         xfrm_audit_helper_sainfo(x, audit_buf);
2391         audit_log_format(audit_buf, " res=%u", result);
2392         audit_log_end(audit_buf);
2393 }
2394 EXPORT_SYMBOL_GPL(xfrm_audit_state_add);
2395
2396 void xfrm_audit_state_delete(struct xfrm_state *x, int result, bool task_valid)
2397 {
2398         struct audit_buffer *audit_buf;
2399
2400         audit_buf = xfrm_audit_start("SAD-delete");
2401         if (audit_buf == NULL)
2402                 return;
2403         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
2404         xfrm_audit_helper_sainfo(x, audit_buf);
2405         audit_log_format(audit_buf, " res=%u", result);
2406         audit_log_end(audit_buf);
2407 }
2408 EXPORT_SYMBOL_GPL(xfrm_audit_state_delete);
2409
2410 void xfrm_audit_state_replay_overflow(struct xfrm_state *x,
2411                                       struct sk_buff *skb)
2412 {
2413         struct audit_buffer *audit_buf;
2414         u32 spi;
2415
2416         audit_buf = xfrm_audit_start("SA-replay-overflow");
2417         if (audit_buf == NULL)
2418                 return;
2419         xfrm_audit_helper_pktinfo(skb, x->props.family, audit_buf);
2420         /* don't record the sequence number because it's inherent in this kind
2421          * of audit message */
2422         spi = ntohl(x->id.spi);
2423         audit_log_format(audit_buf, " spi=%u(0x%x)", spi, spi);
2424         audit_log_end(audit_buf);
2425 }
2426 EXPORT_SYMBOL_GPL(xfrm_audit_state_replay_overflow);
2427
2428 void xfrm_audit_state_replay(struct xfrm_state *x,
2429                              struct sk_buff *skb, __be32 net_seq)
2430 {
2431         struct audit_buffer *audit_buf;
2432         u32 spi;
2433
2434         audit_buf = xfrm_audit_start("SA-replayed-pkt");
2435         if (audit_buf == NULL)
2436                 return;
2437         xfrm_audit_helper_pktinfo(skb, x->props.family, audit_buf);
2438         spi = ntohl(x->id.spi);
2439         audit_log_format(audit_buf, " spi=%u(0x%x) seqno=%u",
2440                          spi, spi, ntohl(net_seq));
2441         audit_log_end(audit_buf);
2442 }
2443 EXPORT_SYMBOL_GPL(xfrm_audit_state_replay);
2444
2445 void xfrm_audit_state_notfound_simple(struct sk_buff *skb, u16 family)
2446 {
2447         struct audit_buffer *audit_buf;
2448
2449         audit_buf = xfrm_audit_start("SA-notfound");
2450         if (audit_buf == NULL)
2451                 return;
2452         xfrm_audit_helper_pktinfo(skb, family, audit_buf);
2453         audit_log_end(audit_buf);
2454 }
2455 EXPORT_SYMBOL_GPL(xfrm_audit_state_notfound_simple);
2456
2457 void xfrm_audit_state_notfound(struct sk_buff *skb, u16 family,
2458                                __be32 net_spi, __be32 net_seq)
2459 {
2460         struct audit_buffer *audit_buf;
2461         u32 spi;
2462
2463         audit_buf = xfrm_audit_start("SA-notfound");
2464         if (audit_buf == NULL)
2465                 return;
2466         xfrm_audit_helper_pktinfo(skb, family, audit_buf);
2467         spi = ntohl(net_spi);
2468         audit_log_format(audit_buf, " spi=%u(0x%x) seqno=%u",
2469                          spi, spi, ntohl(net_seq));
2470         audit_log_end(audit_buf);
2471 }
2472 EXPORT_SYMBOL_GPL(xfrm_audit_state_notfound);
2473
2474 void xfrm_audit_state_icvfail(struct xfrm_state *x,
2475                               struct sk_buff *skb, u8 proto)
2476 {
2477         struct audit_buffer *audit_buf;
2478         __be32 net_spi;
2479         __be32 net_seq;
2480
2481         audit_buf = xfrm_audit_start("SA-icv-failure");
2482         if (audit_buf == NULL)
2483                 return;
2484         xfrm_audit_helper_pktinfo(skb, x->props.family, audit_buf);
2485         if (xfrm_parse_spi(skb, proto, &net_spi, &net_seq) == 0) {
2486                 u32 spi = ntohl(net_spi);
2487                 audit_log_format(audit_buf, " spi=%u(0x%x) seqno=%u",
2488                                  spi, spi, ntohl(net_seq));
2489         }
2490         audit_log_end(audit_buf);
2491 }
2492 EXPORT_SYMBOL_GPL(xfrm_audit_state_icvfail);
2493 #endif /* CONFIG_AUDITSYSCALL */