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