net/xfrm/xfrm_state.c

   1 /*
   2  * xfrm_state.c
   3  *
   4  * Changes:
   5  *      Mitsuru KANDA @USAGI
   6  *      Kazunori MIYAZAWA @USAGI
   7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
   8  *              IPv6 support
   9  *      YOSHIFUJI Hideaki @USAGI
  10  *              Split up af-specific functions
  11  *      Derek Atkins <derek@ihtfp.com>
  12  *              Add UDP Encapsulation
  13  *
  14  */
  15
  16 #include <linux/workqueue.h>
  17 #include <net/xfrm.h>
  18 #include <linux/pfkeyv2.h>
  19 #include <linux/ipsec.h>
  20 #include <linux/module.h>
  21 #include <linux/cache.h>
  22 #include <linux/audit.h>
  23 #include <linux/uaccess.h>
  24 #include <linux/ktime.h>
  25 #include <linux/slab.h>
  26 #include <linux/interrupt.h>
  27 #include <linux/kernel.h>
  28
  29 #include "xfrm_hash.h"
  30
  31 #define xfrm_state_deref_prot(table, net) \
  32         rcu_dereference_protected((table), lockdep_is_held(&(net)->xfrm.xfrm_state_lock))
  33
  34 static void xfrm_state_gc_task(struct work_struct *work);
  35
  36 /* Each xfrm_state may be linked to two tables:
  37
  38    1. Hash table by (spi,daddr,ah/esp) to find SA by SPI. (input,ctl)
  39    2. Hash table by (daddr,family,reqid) to find what SAs exist for given
  40       destination/tunnel endpoint. (output)
  41  */
  42
  43 static unsigned int xfrm_state_hashmax __read_mostly = 1 * 1024 * 1024;
  44 static __read_mostly seqcount_t xfrm_state_hash_generation = SEQCNT_ZERO(xfrm_state_hash_generation);
  45
  46 static DECLARE_WORK(xfrm_state_gc_work, xfrm_state_gc_task);
  47 static HLIST_HEAD(xfrm_state_gc_list);
  48
  49 static inline bool xfrm_state_hold_rcu(struct xfrm_state __rcu *x)
  50 {
  51         return refcount_inc_not_zero(&x->refcnt);
  52 }
  53
  54 static inline unsigned int xfrm_dst_hash(struct net *net,
  55                                          const xfrm_address_t *daddr,
  56                                          const xfrm_address_t *saddr,
  57                                          u32 reqid,
  58                                          unsigned short family)
  59 {
  60         return __xfrm_dst_hash(daddr, saddr, reqid, family, net->xfrm.state_hmask);
  61 }
  62
  63 static inline unsigned int xfrm_src_hash(struct net *net,
  64                                          const xfrm_address_t *daddr,
  65                                          const xfrm_address_t *saddr,
  66                                          unsigned short family)
  67 {
  68         return __xfrm_src_hash(daddr, saddr, family, net->xfrm.state_hmask);
  69 }
  70
  71 static inline unsigned int
  72 xfrm_spi_hash(struct net *net, const xfrm_address_t *daddr,
  73               __be32 spi, u8 proto, unsigned short family)
  74 {
  75         return __xfrm_spi_hash(daddr, spi, proto, family, net->xfrm.state_hmask);
  76 }
  77
  78 static void xfrm_hash_transfer(struct hlist_head *list,
  79                                struct hlist_head *ndsttable,
  80                                struct hlist_head *nsrctable,
  81                                struct hlist_head *nspitable,
  82                                unsigned int nhashmask)
  83 {
  84         struct hlist_node *tmp;
  85         struct xfrm_state *x;
  86
  87         hlist_for_each_entry_safe(x, tmp, list, bydst) {
  88                 unsigned int h;
  89
  90                 h = __xfrm_dst_hash(&x->id.daddr, &x->props.saddr,
  91                                     x->props.reqid, x->props.family,
  92                                     nhashmask);
  93                 hlist_add_head_rcu(&x->bydst, ndsttable + h);
  94
  95                 h = __xfrm_src_hash(&x->id.daddr, &x->props.saddr,
  96                                     x->props.family,
  97                                     nhashmask);
  98                 hlist_add_head_rcu(&x->bysrc, nsrctable + h);
  99
 100                 if (x->id.spi) {
 101                         h = __xfrm_spi_hash(&x->id.daddr, x->id.spi,
 102                                             x->id.proto, x->props.family,
 103                                             nhashmask);
 104                         hlist_add_head_rcu(&x->byspi, nspitable + h);
 105                 }
 106         }
 107 }
 108
 109 static unsigned long xfrm_hash_new_size(unsigned int state_hmask)
 110 {
 111         return ((state_hmask + 1) << 1) * sizeof(struct hlist_head);
 112 }
 113
 114 static void xfrm_hash_resize(struct work_struct *work)
 115 {
 116         struct net *net = container_of(work, struct net, xfrm.state_hash_work);
 117         struct hlist_head *ndst, *nsrc, *nspi, *odst, *osrc, *ospi;
 118         unsigned long nsize, osize;
 119         unsigned int nhashmask, ohashmask;
 120         int i;
 121
 122         nsize = xfrm_hash_new_size(net->xfrm.state_hmask);
 123         ndst = xfrm_hash_alloc(nsize);
 124         if (!ndst)
 125                 return;
 126         nsrc = xfrm_hash_alloc(nsize);
 127         if (!nsrc) {
 128                 xfrm_hash_free(ndst, nsize);
 129                 return;
 130         }
 131         nspi = xfrm_hash_alloc(nsize);
 132         if (!nspi) {
 133                 xfrm_hash_free(ndst, nsize);
 134                 xfrm_hash_free(nsrc, nsize);
 135                 return;
 136         }
 137
 138         spin_lock_bh(&net->xfrm.xfrm_state_lock);
 139         write_seqcount_begin(&xfrm_state_hash_generation);
 140
 141         nhashmask = (nsize / sizeof(struct hlist_head)) - 1U;
 142         odst = xfrm_state_deref_prot(net->xfrm.state_bydst, net);
 143         for (i = net->xfrm.state_hmask; i >= 0; i--)
 144                 xfrm_hash_transfer(odst + i, ndst, nsrc, nspi, nhashmask);
 145
 146         osrc = xfrm_state_deref_prot(net->xfrm.state_bysrc, net);
 147         ospi = xfrm_state_deref_prot(net->xfrm.state_byspi, net);
 148         ohashmask = net->xfrm.state_hmask;
 149
 150         rcu_assign_pointer(net->xfrm.state_bydst, ndst);
 151         rcu_assign_pointer(net->xfrm.state_bysrc, nsrc);
 152         rcu_assign_pointer(net->xfrm.state_byspi, nspi);
 153         net->xfrm.state_hmask = nhashmask;
 154
 155         write_seqcount_end(&xfrm_state_hash_generation);
 156         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
 157
 158         osize = (ohashmask + 1) * sizeof(struct hlist_head);
 159
 160         synchronize_rcu();
 161
 162         xfrm_hash_free(odst, osize);
 163         xfrm_hash_free(osrc, osize);
 164         xfrm_hash_free(ospi, osize);
 165 }
 166
 167 static DEFINE_SPINLOCK(xfrm_state_afinfo_lock);
 168 static struct xfrm_state_afinfo __rcu *xfrm_state_afinfo[NPROTO];
 169
 170 static DEFINE_SPINLOCK(xfrm_state_gc_lock);
 171
 172 int __xfrm_state_delete(struct xfrm_state *x);
 173
 174 int km_query(struct xfrm_state *x, struct xfrm_tmpl *t, struct xfrm_policy *pol);
 175 bool km_is_alive(const struct km_event *c);
 176 void km_state_expired(struct xfrm_state *x, int hard, u32 portid);
 177
 178 static DEFINE_SPINLOCK(xfrm_type_lock);
 179 int xfrm_register_type(const struct xfrm_type *type, unsigned short family)
 180 {
 181         struct xfrm_state_afinfo *afinfo = xfrm_state_get_afinfo(family);
 182         const struct xfrm_type **typemap;
 183         int err = 0;
 184
 185         if (unlikely(afinfo == NULL))
 186                 return -EAFNOSUPPORT;
 187         typemap = afinfo->type_map;
 188         spin_lock_bh(&xfrm_type_lock);
 189
 190         if (likely(typemap[type->proto] == NULL))
 191                 typemap[type->proto] = type;
 192         else
 193                 err = -EEXIST;
 194         spin_unlock_bh(&xfrm_type_lock);
 195         rcu_read_unlock();
 196         return err;
 197 }
 198 EXPORT_SYMBOL(xfrm_register_type);
 199
 200 int xfrm_unregister_type(const struct xfrm_type *type, unsigned short family)
 201 {
 202         struct xfrm_state_afinfo *afinfo = xfrm_state_get_afinfo(family);
 203         const struct xfrm_type **typemap;
 204         int err = 0;
 205
 206         if (unlikely(afinfo == NULL))
 207                 return -EAFNOSUPPORT;
 208         typemap = afinfo->type_map;
 209         spin_lock_bh(&xfrm_type_lock);
 210
 211         if (unlikely(typemap[type->proto] != type))
 212                 err = -ENOENT;
 213         else
 214                 typemap[type->proto] = NULL;
 215         spin_unlock_bh(&xfrm_type_lock);
 216         rcu_read_unlock();
 217         return err;
 218 }
 219 EXPORT_SYMBOL(xfrm_unregister_type);
 220
 221 static const struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
 222 {
 223         struct xfrm_state_afinfo *afinfo;
 224         const struct xfrm_type **typemap;
 225         const struct xfrm_type *type;
 226         int modload_attempted = 0;
 227
 228 retry:
 229         afinfo = xfrm_state_get_afinfo(family);
 230         if (unlikely(afinfo == NULL))
 231                 return NULL;
 232         typemap = afinfo->type_map;
 233
 234         type = READ_ONCE(typemap[proto]);
 235         if (unlikely(type && !try_module_get(type->owner)))
 236                 type = NULL;
 237
 238         rcu_read_unlock();
 239
 240         if (!type && !modload_attempted) {
 241                 request_module("xfrm-type-%d-%d", family, proto);
 242                 modload_attempted = 1;
 243                 goto retry;
 244         }
 245
 246         return type;
 247 }
 248
 249 static void xfrm_put_type(const struct xfrm_type *type)
 250 {
 251         module_put(type->owner);
 252 }
 253
 254 static DEFINE_SPINLOCK(xfrm_type_offload_lock);
 255 int xfrm_register_type_offload(const struct xfrm_type_offload *type,
 256                                unsigned short family)
 257 {
 258         struct xfrm_state_afinfo *afinfo = xfrm_state_get_afinfo(family);
 259         const struct xfrm_type_offload **typemap;
 260         int err = 0;
 261
 262         if (unlikely(afinfo == NULL))
 263                 return -EAFNOSUPPORT;
 264         typemap = afinfo->type_offload_map;
 265         spin_lock_bh(&xfrm_type_offload_lock);
 266
 267         if (likely(typemap[type->proto] == NULL))
 268                 typemap[type->proto] = type;
 269         else
 270                 err = -EEXIST;
 271         spin_unlock_bh(&xfrm_type_offload_lock);
 272         rcu_read_unlock();
 273         return err;
 274 }
 275 EXPORT_SYMBOL(xfrm_register_type_offload);
 276
 277 int xfrm_unregister_type_offload(const struct xfrm_type_offload *type,
 278                                  unsigned short family)
 279 {
 280         struct xfrm_state_afinfo *afinfo = xfrm_state_get_afinfo(family);
 281         const struct xfrm_type_offload **typemap;
 282         int err = 0;
 283
 284         if (unlikely(afinfo == NULL))
 285                 return -EAFNOSUPPORT;
 286         typemap = afinfo->type_offload_map;
 287         spin_lock_bh(&xfrm_type_offload_lock);
 288
 289         if (unlikely(typemap[type->proto] != type))
 290                 err = -ENOENT;
 291         else
 292                 typemap[type->proto] = NULL;
 293         spin_unlock_bh(&xfrm_type_offload_lock);
 294         rcu_read_unlock();
 295         return err;
 296 }
 297 EXPORT_SYMBOL(xfrm_unregister_type_offload);
 298
 299 static const struct xfrm_type_offload *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                 refcount_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                                            struct xfrm_encap_tmpl *encap)
1314 {
1315         struct net *net = xs_net(orig);
1316         struct xfrm_state *x = xfrm_state_alloc(net);
1317         if (!x)
1318                 goto out;
1319
1320         memcpy(&x->id, &orig->id, sizeof(x->id));
1321         memcpy(&x->sel, &orig->sel, sizeof(x->sel));
1322         memcpy(&x->lft, &orig->lft, sizeof(x->lft));
1323         x->props.mode = orig->props.mode;
1324         x->props.replay_window = orig->props.replay_window;
1325         x->props.reqid = orig->props.reqid;
1326         x->props.family = orig->props.family;
1327         x->props.saddr = orig->props.saddr;
1328
1329         if (orig->aalg) {
1330                 x->aalg = xfrm_algo_auth_clone(orig->aalg);
1331                 if (!x->aalg)
1332                         goto error;
1333         }
1334         x->props.aalgo = orig->props.aalgo;
1335
1336         if (orig->aead) {
1337                 x->aead = xfrm_algo_aead_clone(orig->aead);
1338                 if (!x->aead)
1339                         goto error;
1340         }
1341         if (orig->ealg) {
1342                 x->ealg = xfrm_algo_clone(orig->ealg);
1343                 if (!x->ealg)
1344                         goto error;
1345         }
1346         x->props.ealgo = orig->props.ealgo;
1347
1348         if (orig->calg) {
1349                 x->calg = xfrm_algo_clone(orig->calg);
1350                 if (!x->calg)
1351                         goto error;
1352         }
1353         x->props.calgo = orig->props.calgo;
1354
1355         if (encap || orig->encap) {
1356                 if (encap)
1357                         x->encap = kmemdup(encap, sizeof(*x->encap),
1358                                         GFP_KERNEL);
1359                 else
1360                         x->encap = kmemdup(orig->encap, sizeof(*x->encap),
1361                                         GFP_KERNEL);
1362
1363                 if (!x->encap)
1364                         goto error;
1365         }
1366
1367         if (orig->coaddr) {
1368                 x->coaddr = kmemdup(orig->coaddr, sizeof(*x->coaddr),
1369                                     GFP_KERNEL);
1370                 if (!x->coaddr)
1371                         goto error;
1372         }
1373
1374         if (orig->replay_esn) {
1375                 if (xfrm_replay_clone(x, orig))
1376                         goto error;
1377         }
1378
1379         memcpy(&x->mark, &orig->mark, sizeof(x->mark));
1380
1381         if (xfrm_init_state(x) < 0)
1382                 goto error;
1383
1384         x->props.flags = orig->props.flags;
1385         x->props.extra_flags = orig->props.extra_flags;
1386
1387         x->tfcpad = orig->tfcpad;
1388         x->replay_maxdiff = orig->replay_maxdiff;
1389         x->replay_maxage = orig->replay_maxage;
1390         x->curlft.add_time = orig->curlft.add_time;
1391         x->km.state = orig->km.state;
1392         x->km.seq = orig->km.seq;
1393         x->replay = orig->replay;
1394         x->preplay = orig->preplay;
1395
1396         return x;
1397
1398  error:
1399         xfrm_state_put(x);
1400 out:
1401         return NULL;
1402 }
1403
1404 struct xfrm_state *xfrm_migrate_state_find(struct xfrm_migrate *m, struct net *net)
1405 {
1406         unsigned int h;
1407         struct xfrm_state *x = NULL;
1408
1409         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1410
1411         if (m->reqid) {
1412                 h = xfrm_dst_hash(net, &m->old_daddr, &m->old_saddr,
1413                                   m->reqid, m->old_family);
1414                 hlist_for_each_entry(x, net->xfrm.state_bydst+h, bydst) {
1415                         if (x->props.mode != m->mode ||
1416                             x->id.proto != m->proto)
1417                                 continue;
1418                         if (m->reqid && x->props.reqid != m->reqid)
1419                                 continue;
1420                         if (!xfrm_addr_equal(&x->id.daddr, &m->old_daddr,
1421                                              m->old_family) ||
1422                             !xfrm_addr_equal(&x->props.saddr, &m->old_saddr,
1423                                              m->old_family))
1424                                 continue;
1425                         xfrm_state_hold(x);
1426                         break;
1427                 }
1428         } else {
1429                 h = xfrm_src_hash(net, &m->old_daddr, &m->old_saddr,
1430                                   m->old_family);
1431                 hlist_for_each_entry(x, net->xfrm.state_bysrc+h, bysrc) {
1432                         if (x->props.mode != m->mode ||
1433                             x->id.proto != m->proto)
1434                                 continue;
1435                         if (!xfrm_addr_equal(&x->id.daddr, &m->old_daddr,
1436                                              m->old_family) ||
1437                             !xfrm_addr_equal(&x->props.saddr, &m->old_saddr,
1438                                              m->old_family))
1439                                 continue;
1440                         xfrm_state_hold(x);
1441                         break;
1442                 }
1443         }
1444
1445         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1446
1447         return x;
1448 }
1449 EXPORT_SYMBOL(xfrm_migrate_state_find);
1450
1451 struct xfrm_state *xfrm_state_migrate(struct xfrm_state *x,
1452                                       struct xfrm_migrate *m,
1453                                       struct xfrm_encap_tmpl *encap)
1454 {
1455         struct xfrm_state *xc;
1456
1457         xc = xfrm_state_clone(x, encap);
1458         if (!xc)
1459                 return NULL;
1460
1461         memcpy(&xc->id.daddr, &m->new_daddr, sizeof(xc->id.daddr));
1462         memcpy(&xc->props.saddr, &m->new_saddr, sizeof(xc->props.saddr));
1463
1464         /* add state */
1465         if (xfrm_addr_equal(&x->id.daddr, &m->new_daddr, m->new_family)) {
1466                 /* a care is needed when the destination address of the
1467                    state is to be updated as it is a part of triplet */
1468                 xfrm_state_insert(xc);
1469         } else {
1470                 if (xfrm_state_add(xc) < 0)
1471                         goto error;
1472         }
1473
1474         return xc;
1475 error:
1476         xfrm_state_put(xc);
1477         return NULL;
1478 }
1479 EXPORT_SYMBOL(xfrm_state_migrate);
1480 #endif
1481
1482 int xfrm_state_update(struct xfrm_state *x)
1483 {
1484         struct xfrm_state *x1, *to_put;
1485         int err;
1486         int use_spi = xfrm_id_proto_match(x->id.proto, IPSEC_PROTO_ANY);
1487         struct net *net = xs_net(x);
1488
1489         to_put = NULL;
1490
1491         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1492         x1 = __xfrm_state_locate(x, use_spi, x->props.family);
1493
1494         err = -ESRCH;
1495         if (!x1)
1496                 goto out;
1497
1498         if (xfrm_state_kern(x1)) {
1499                 to_put = x1;
1500                 err = -EEXIST;
1501                 goto out;
1502         }
1503
1504         if (x1->km.state == XFRM_STATE_ACQ) {
1505                 __xfrm_state_insert(x);
1506                 x = NULL;
1507         }
1508         err = 0;
1509
1510 out:
1511         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1512
1513         if (to_put)
1514                 xfrm_state_put(to_put);
1515
1516         if (err)
1517                 return err;
1518
1519         if (!x) {
1520                 xfrm_state_delete(x1);
1521                 xfrm_state_put(x1);
1522                 return 0;
1523         }
1524
1525         err = -EINVAL;
1526         spin_lock_bh(&x1->lock);
1527         if (likely(x1->km.state == XFRM_STATE_VALID)) {
1528                 if (x->encap && x1->encap)
1529                         memcpy(x1->encap, x->encap, sizeof(*x1->encap));
1530                 if (x->coaddr && x1->coaddr) {
1531                         memcpy(x1->coaddr, x->coaddr, sizeof(*x1->coaddr));
1532                 }
1533                 if (!use_spi && memcmp(&x1->sel, &x->sel, sizeof(x1->sel)))
1534                         memcpy(&x1->sel, &x->sel, sizeof(x1->sel));
1535                 memcpy(&x1->lft, &x->lft, sizeof(x1->lft));
1536                 x1->km.dying = 0;
1537
1538                 tasklet_hrtimer_start(&x1->mtimer, ktime_set(1, 0), HRTIMER_MODE_REL);
1539                 if (x1->curlft.use_time)
1540                         xfrm_state_check_expire(x1);
1541
1542                 err = 0;
1543                 x->km.state = XFRM_STATE_DEAD;
1544                 __xfrm_state_put(x);
1545         }
1546         spin_unlock_bh(&x1->lock);
1547
1548         xfrm_state_put(x1);
1549
1550         return err;
1551 }
1552 EXPORT_SYMBOL(xfrm_state_update);
1553
1554 int xfrm_state_check_expire(struct xfrm_state *x)
1555 {
1556         if (!x->curlft.use_time)
1557                 x->curlft.use_time = get_seconds();
1558
1559         if (x->curlft.bytes >= x->lft.hard_byte_limit ||
1560             x->curlft.packets >= x->lft.hard_packet_limit) {
1561                 x->km.state = XFRM_STATE_EXPIRED;
1562                 tasklet_hrtimer_start(&x->mtimer, 0, HRTIMER_MODE_REL);
1563                 return -EINVAL;
1564         }
1565
1566         if (!x->km.dying &&
1567             (x->curlft.bytes >= x->lft.soft_byte_limit ||
1568              x->curlft.packets >= x->lft.soft_packet_limit)) {
1569                 x->km.dying = 1;
1570                 km_state_expired(x, 0, 0);
1571         }
1572         return 0;
1573 }
1574 EXPORT_SYMBOL(xfrm_state_check_expire);
1575
1576 struct xfrm_state *
1577 xfrm_state_lookup(struct net *net, u32 mark, const xfrm_address_t *daddr, __be32 spi,
1578                   u8 proto, unsigned short family)
1579 {
1580         struct xfrm_state *x;
1581
1582         rcu_read_lock();
1583         x = __xfrm_state_lookup(net, mark, daddr, spi, proto, family);
1584         rcu_read_unlock();
1585         return x;
1586 }
1587 EXPORT_SYMBOL(xfrm_state_lookup);
1588
1589 struct xfrm_state *
1590 xfrm_state_lookup_byaddr(struct net *net, u32 mark,
1591                          const xfrm_address_t *daddr, const xfrm_address_t *saddr,
1592                          u8 proto, unsigned short family)
1593 {
1594         struct xfrm_state *x;
1595
1596         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1597         x = __xfrm_state_lookup_byaddr(net, mark, daddr, saddr, proto, family);
1598         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1599         return x;
1600 }
1601 EXPORT_SYMBOL(xfrm_state_lookup_byaddr);
1602
1603 struct xfrm_state *
1604 xfrm_find_acq(struct net *net, const struct xfrm_mark *mark, u8 mode, u32 reqid,
1605               u8 proto, const xfrm_address_t *daddr,
1606               const xfrm_address_t *saddr, int create, unsigned short family)
1607 {
1608         struct xfrm_state *x;
1609
1610         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1611         x = __find_acq_core(net, mark, family, mode, reqid, proto, daddr, saddr, create);
1612         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1613
1614         return x;
1615 }
1616 EXPORT_SYMBOL(xfrm_find_acq);
1617
1618 #ifdef CONFIG_XFRM_SUB_POLICY
1619 int
1620 xfrm_tmpl_sort(struct xfrm_tmpl **dst, struct xfrm_tmpl **src, int n,
1621                unsigned short family, struct net *net)
1622 {
1623         int err = 0;
1624         struct xfrm_state_afinfo *afinfo = xfrm_state_get_afinfo(family);
1625         if (!afinfo)
1626                 return -EAFNOSUPPORT;
1627
1628         spin_lock_bh(&net->xfrm.xfrm_state_lock); /*FIXME*/
1629         if (afinfo->tmpl_sort)
1630                 err = afinfo->tmpl_sort(dst, src, n);
1631         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1632         rcu_read_unlock();
1633         return err;
1634 }
1635 EXPORT_SYMBOL(xfrm_tmpl_sort);
1636
1637 int
1638 xfrm_state_sort(struct xfrm_state **dst, struct xfrm_state **src, int n,
1639                 unsigned short family)
1640 {
1641         int err = 0;
1642         struct xfrm_state_afinfo *afinfo = xfrm_state_get_afinfo(family);
1643         struct net *net = xs_net(*src);
1644
1645         if (!afinfo)
1646                 return -EAFNOSUPPORT;
1647
1648         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1649         if (afinfo->state_sort)
1650                 err = afinfo->state_sort(dst, src, n);
1651         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1652         rcu_read_unlock();
1653         return err;
1654 }
1655 EXPORT_SYMBOL(xfrm_state_sort);
1656 #endif
1657
1658 /* Silly enough, but I'm lazy to build resolution list */
1659
1660 static struct xfrm_state *__xfrm_find_acq_byseq(struct net *net, u32 mark, u32 seq)
1661 {
1662         int i;
1663
1664         for (i = 0; i <= net->xfrm.state_hmask; i++) {
1665                 struct xfrm_state *x;
1666
1667                 hlist_for_each_entry(x, net->xfrm.state_bydst+i, bydst) {
1668                         if (x->km.seq == seq &&
1669                             (mark & x->mark.m) == x->mark.v &&
1670                             x->km.state == XFRM_STATE_ACQ) {
1671                                 xfrm_state_hold(x);
1672                                 return x;
1673                         }
1674                 }
1675         }
1676         return NULL;
1677 }
1678
1679 struct xfrm_state *xfrm_find_acq_byseq(struct net *net, u32 mark, u32 seq)
1680 {
1681         struct xfrm_state *x;
1682
1683         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1684         x = __xfrm_find_acq_byseq(net, mark, seq);
1685         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1686         return x;
1687 }
1688 EXPORT_SYMBOL(xfrm_find_acq_byseq);
1689
1690 u32 xfrm_get_acqseq(void)
1691 {
1692         u32 res;
1693         static atomic_t acqseq;
1694
1695         do {
1696                 res = atomic_inc_return(&acqseq);
1697         } while (!res);
1698
1699         return res;
1700 }
1701 EXPORT_SYMBOL(xfrm_get_acqseq);
1702
1703 int verify_spi_info(u8 proto, u32 min, u32 max)
1704 {
1705         switch (proto) {
1706         case IPPROTO_AH:
1707         case IPPROTO_ESP:
1708                 break;
1709
1710         case IPPROTO_COMP:
1711                 /* IPCOMP spi is 16-bits. */
1712                 if (max >= 0x10000)
1713                         return -EINVAL;
1714                 break;
1715
1716         default:
1717                 return -EINVAL;
1718         }
1719
1720         if (min > max)
1721                 return -EINVAL;
1722
1723         return 0;
1724 }
1725 EXPORT_SYMBOL(verify_spi_info);
1726
1727 int xfrm_alloc_spi(struct xfrm_state *x, u32 low, u32 high)
1728 {
1729         struct net *net = xs_net(x);
1730         unsigned int h;
1731         struct xfrm_state *x0;
1732         int err = -ENOENT;
1733         __be32 minspi = htonl(low);
1734         __be32 maxspi = htonl(high);
1735         u32 mark = x->mark.v & x->mark.m;
1736
1737         spin_lock_bh(&x->lock);
1738         if (x->km.state == XFRM_STATE_DEAD)
1739                 goto unlock;
1740
1741         err = 0;
1742         if (x->id.spi)
1743                 goto unlock;
1744
1745         err = -ENOENT;
1746
1747         if (minspi == maxspi) {
1748                 x0 = xfrm_state_lookup(net, mark, &x->id.daddr, minspi, x->id.proto, x->props.family);
1749                 if (x0) {
1750                         xfrm_state_put(x0);
1751                         goto unlock;
1752                 }
1753                 x->id.spi = minspi;
1754         } else {
1755                 u32 spi = 0;
1756                 for (h = 0; h < high-low+1; h++) {
1757                         spi = low + prandom_u32()%(high-low+1);
1758                         x0 = xfrm_state_lookup(net, mark, &x->id.daddr, htonl(spi), x->id.proto, x->props.family);
1759                         if (x0 == NULL) {
1760                                 x->id.spi = htonl(spi);
1761                                 break;
1762                         }
1763                         xfrm_state_put(x0);
1764                 }
1765         }
1766         if (x->id.spi) {
1767                 spin_lock_bh(&net->xfrm.xfrm_state_lock);
1768                 h = xfrm_spi_hash(net, &x->id.daddr, x->id.spi, x->id.proto, x->props.family);
1769                 hlist_add_head_rcu(&x->byspi, net->xfrm.state_byspi + h);
1770                 spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1771
1772                 err = 0;
1773         }
1774
1775 unlock:
1776         spin_unlock_bh(&x->lock);
1777
1778         return err;
1779 }
1780 EXPORT_SYMBOL(xfrm_alloc_spi);
1781
1782 static bool __xfrm_state_filter_match(struct xfrm_state *x,
1783                                       struct xfrm_address_filter *filter)
1784 {
1785         if (filter) {
1786                 if ((filter->family == AF_INET ||
1787                      filter->family == AF_INET6) &&
1788                     x->props.family != filter->family)
1789                         return false;
1790
1791                 return addr_match(&x->props.saddr, &filter->saddr,
1792                                   filter->splen) &&
1793                        addr_match(&x->id.daddr, &filter->daddr,
1794                                   filter->dplen);
1795         }
1796         return true;
1797 }
1798
1799 int xfrm_state_walk(struct net *net, struct xfrm_state_walk *walk,
1800                     int (*func)(struct xfrm_state *, int, void*),
1801                     void *data)
1802 {
1803         struct xfrm_state *state;
1804         struct xfrm_state_walk *x;
1805         int err = 0;
1806
1807         if (walk->seq != 0 && list_empty(&walk->all))
1808                 return 0;
1809
1810         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1811         if (list_empty(&walk->all))
1812                 x = list_first_entry(&net->xfrm.state_all, struct xfrm_state_walk, all);
1813         else
1814                 x = list_first_entry(&walk->all, struct xfrm_state_walk, all);
1815         list_for_each_entry_from(x, &net->xfrm.state_all, all) {
1816                 if (x->state == XFRM_STATE_DEAD)
1817                         continue;
1818                 state = container_of(x, struct xfrm_state, km);
1819                 if (!xfrm_id_proto_match(state->id.proto, walk->proto))
1820                         continue;
1821                 if (!__xfrm_state_filter_match(state, walk->filter))
1822                         continue;
1823                 err = func(state, walk->seq, data);
1824                 if (err) {
1825                         list_move_tail(&walk->all, &x->all);
1826                         goto out;
1827                 }
1828                 walk->seq++;
1829         }
1830         if (walk->seq == 0) {
1831                 err = -ENOENT;
1832                 goto out;
1833         }
1834         list_del_init(&walk->all);
1835 out:
1836         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1837         return err;
1838 }
1839 EXPORT_SYMBOL(xfrm_state_walk);
1840
1841 void xfrm_state_walk_init(struct xfrm_state_walk *walk, u8 proto,
1842                           struct xfrm_address_filter *filter)
1843 {
1844         INIT_LIST_HEAD(&walk->all);
1845         walk->proto = proto;
1846         walk->state = XFRM_STATE_DEAD;
1847         walk->seq = 0;
1848         walk->filter = filter;
1849 }
1850 EXPORT_SYMBOL(xfrm_state_walk_init);
1851
1852 void xfrm_state_walk_done(struct xfrm_state_walk *walk, struct net *net)
1853 {
1854         kfree(walk->filter);
1855
1856         if (list_empty(&walk->all))
1857                 return;
1858
1859         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1860         list_del(&walk->all);
1861         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1862 }
1863 EXPORT_SYMBOL(xfrm_state_walk_done);
1864
1865 static void xfrm_replay_timer_handler(unsigned long data)
1866 {
1867         struct xfrm_state *x = (struct xfrm_state *)data;
1868
1869         spin_lock(&x->lock);
1870
1871         if (x->km.state == XFRM_STATE_VALID) {
1872                 if (xfrm_aevent_is_on(xs_net(x)))
1873                         x->repl->notify(x, XFRM_REPLAY_TIMEOUT);
1874                 else
1875                         x->xflags |= XFRM_TIME_DEFER;
1876         }
1877
1878         spin_unlock(&x->lock);
1879 }
1880
1881 static LIST_HEAD(xfrm_km_list);
1882
1883 void km_policy_notify(struct xfrm_policy *xp, int dir, const struct km_event *c)
1884 {
1885         struct xfrm_mgr *km;
1886
1887         rcu_read_lock();
1888         list_for_each_entry_rcu(km, &xfrm_km_list, list)
1889                 if (km->notify_policy)
1890                         km->notify_policy(xp, dir, c);
1891         rcu_read_unlock();
1892 }
1893
1894 void km_state_notify(struct xfrm_state *x, const struct km_event *c)
1895 {
1896         struct xfrm_mgr *km;
1897         rcu_read_lock();
1898         list_for_each_entry_rcu(km, &xfrm_km_list, list)
1899                 if (km->notify)
1900                         km->notify(x, c);
1901         rcu_read_unlock();
1902 }
1903
1904 EXPORT_SYMBOL(km_policy_notify);
1905 EXPORT_SYMBOL(km_state_notify);
1906
1907 void km_state_expired(struct xfrm_state *x, int hard, u32 portid)
1908 {
1909         struct km_event c;
1910
1911         c.data.hard = hard;
1912         c.portid = portid;
1913         c.event = XFRM_MSG_EXPIRE;
1914         km_state_notify(x, &c);
1915 }
1916
1917 EXPORT_SYMBOL(km_state_expired);
1918 /*
1919  * We send to all registered managers regardless of failure
1920  * We are happy with one success
1921 */
1922 int km_query(struct xfrm_state *x, struct xfrm_tmpl *t, struct xfrm_policy *pol)
1923 {
1924         int err = -EINVAL, acqret;
1925         struct xfrm_mgr *km;
1926
1927         rcu_read_lock();
1928         list_for_each_entry_rcu(km, &xfrm_km_list, list) {
1929                 acqret = km->acquire(x, t, pol);
1930                 if (!acqret)
1931                         err = acqret;
1932         }
1933         rcu_read_unlock();
1934         return err;
1935 }
1936 EXPORT_SYMBOL(km_query);
1937
1938 int km_new_mapping(struct xfrm_state *x, xfrm_address_t *ipaddr, __be16 sport)
1939 {
1940         int err = -EINVAL;
1941         struct xfrm_mgr *km;
1942
1943         rcu_read_lock();
1944         list_for_each_entry_rcu(km, &xfrm_km_list, list) {
1945                 if (km->new_mapping)
1946                         err = km->new_mapping(x, ipaddr, sport);
1947                 if (!err)
1948                         break;
1949         }
1950         rcu_read_unlock();
1951         return err;
1952 }
1953 EXPORT_SYMBOL(km_new_mapping);
1954
1955 void km_policy_expired(struct xfrm_policy *pol, int dir, int hard, u32 portid)
1956 {
1957         struct km_event c;
1958
1959         c.data.hard = hard;
1960         c.portid = portid;
1961         c.event = XFRM_MSG_POLEXPIRE;
1962         km_policy_notify(pol, dir, &c);
1963 }
1964 EXPORT_SYMBOL(km_policy_expired);
1965
1966 #ifdef CONFIG_XFRM_MIGRATE
1967 int km_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
1968                const struct xfrm_migrate *m, int num_migrate,
1969                const struct xfrm_kmaddress *k,
1970                const struct xfrm_encap_tmpl *encap)
1971 {
1972         int err = -EINVAL;
1973         int ret;
1974         struct xfrm_mgr *km;
1975
1976         rcu_read_lock();
1977         list_for_each_entry_rcu(km, &xfrm_km_list, list) {
1978                 if (km->migrate) {
1979                         ret = km->migrate(sel, dir, type, m, num_migrate, k,
1980                                           encap);
1981                         if (!ret)
1982                                 err = ret;
1983                 }
1984         }
1985         rcu_read_unlock();
1986         return err;
1987 }
1988 EXPORT_SYMBOL(km_migrate);
1989 #endif
1990
1991 int km_report(struct net *net, u8 proto, struct xfrm_selector *sel, xfrm_address_t *addr)
1992 {
1993         int err = -EINVAL;
1994         int ret;
1995         struct xfrm_mgr *km;
1996
1997         rcu_read_lock();
1998         list_for_each_entry_rcu(km, &xfrm_km_list, list) {
1999                 if (km->report) {
2000                         ret = km->report(net, proto, sel, addr);
2001                         if (!ret)
2002                                 err = ret;
2003                 }
2004         }
2005         rcu_read_unlock();
2006         return err;
2007 }
2008 EXPORT_SYMBOL(km_report);
2009
2010 bool km_is_alive(const struct km_event *c)
2011 {
2012         struct xfrm_mgr *km;
2013         bool is_alive = false;
2014
2015         rcu_read_lock();
2016         list_for_each_entry_rcu(km, &xfrm_km_list, list) {
2017                 if (km->is_alive && km->is_alive(c)) {
2018                         is_alive = true;
2019                         break;
2020                 }
2021         }
2022         rcu_read_unlock();
2023
2024         return is_alive;
2025 }
2026 EXPORT_SYMBOL(km_is_alive);
2027
2028 int xfrm_user_policy(struct sock *sk, int optname, u8 __user *optval, int optlen)
2029 {
2030         int err;
2031         u8 *data;
2032         struct xfrm_mgr *km;
2033         struct xfrm_policy *pol = NULL;
2034
2035         if (optlen <= 0 || optlen > PAGE_SIZE)
2036                 return -EMSGSIZE;
2037
2038         data = memdup_user(optval, optlen);
2039         if (IS_ERR(data))
2040                 return PTR_ERR(data);
2041
2042         err = -EINVAL;
2043         rcu_read_lock();
2044         list_for_each_entry_rcu(km, &xfrm_km_list, list) {
2045                 pol = km->compile_policy(sk, optname, data,
2046                                          optlen, &err);
2047                 if (err >= 0)
2048                         break;
2049         }
2050         rcu_read_unlock();
2051
2052         if (err >= 0) {
2053                 xfrm_sk_policy_insert(sk, err, pol);
2054                 xfrm_pol_put(pol);
2055                 err = 0;
2056         }
2057
2058         kfree(data);
2059         return err;
2060 }
2061 EXPORT_SYMBOL(xfrm_user_policy);
2062
2063 static DEFINE_SPINLOCK(xfrm_km_lock);
2064
2065 int xfrm_register_km(struct xfrm_mgr *km)
2066 {
2067         spin_lock_bh(&xfrm_km_lock);
2068         list_add_tail_rcu(&km->list, &xfrm_km_list);
2069         spin_unlock_bh(&xfrm_km_lock);
2070         return 0;
2071 }
2072 EXPORT_SYMBOL(xfrm_register_km);
2073
2074 int xfrm_unregister_km(struct xfrm_mgr *km)
2075 {
2076         spin_lock_bh(&xfrm_km_lock);
2077         list_del_rcu(&km->list);
2078         spin_unlock_bh(&xfrm_km_lock);
2079         synchronize_rcu();
2080         return 0;
2081 }
2082 EXPORT_SYMBOL(xfrm_unregister_km);
2083
2084 int xfrm_state_register_afinfo(struct xfrm_state_afinfo *afinfo)
2085 {
2086         int err = 0;
2087
2088         if (WARN_ON(afinfo->family >= NPROTO))
2089                 return -EAFNOSUPPORT;
2090
2091         spin_lock_bh(&xfrm_state_afinfo_lock);
2092         if (unlikely(xfrm_state_afinfo[afinfo->family] != NULL))
2093                 err = -EEXIST;
2094         else
2095                 rcu_assign_pointer(xfrm_state_afinfo[afinfo->family], afinfo);
2096         spin_unlock_bh(&xfrm_state_afinfo_lock);
2097         return err;
2098 }
2099 EXPORT_SYMBOL(xfrm_state_register_afinfo);
2100
2101 int xfrm_state_unregister_afinfo(struct xfrm_state_afinfo *afinfo)
2102 {
2103         int err = 0, family = afinfo->family;
2104
2105         if (WARN_ON(family >= NPROTO))
2106                 return -EAFNOSUPPORT;
2107
2108         spin_lock_bh(&xfrm_state_afinfo_lock);
2109         if (likely(xfrm_state_afinfo[afinfo->family] != NULL)) {
2110                 if (rcu_access_pointer(xfrm_state_afinfo[family]) != afinfo)
2111                         err = -EINVAL;
2112                 else
2113                         RCU_INIT_POINTER(xfrm_state_afinfo[afinfo->family], NULL);
2114         }
2115         spin_unlock_bh(&xfrm_state_afinfo_lock);
2116         synchronize_rcu();
2117         return err;
2118 }
2119 EXPORT_SYMBOL(xfrm_state_unregister_afinfo);
2120
2121 struct xfrm_state_afinfo *xfrm_state_afinfo_get_rcu(unsigned int family)
2122 {
2123         if (unlikely(family >= NPROTO))
2124                 return NULL;
2125
2126         return rcu_dereference(xfrm_state_afinfo[family]);
2127 }
2128
2129 struct xfrm_state_afinfo *xfrm_state_get_afinfo(unsigned int family)
2130 {
2131         struct xfrm_state_afinfo *afinfo;
2132         if (unlikely(family >= NPROTO))
2133                 return NULL;
2134         rcu_read_lock();
2135         afinfo = rcu_dereference(xfrm_state_afinfo[family]);
2136         if (unlikely(!afinfo))
2137                 rcu_read_unlock();
2138         return afinfo;
2139 }
2140
2141 /* Temporarily located here until net/xfrm/xfrm_tunnel.c is created */
2142 void xfrm_state_delete_tunnel(struct xfrm_state *x)
2143 {
2144         if (x->tunnel) {
2145                 struct xfrm_state *t = x->tunnel;
2146
2147                 if (atomic_read(&t->tunnel_users) == 2)
2148                         xfrm_state_delete(t);
2149                 atomic_dec(&t->tunnel_users);
2150                 xfrm_state_put(t);
2151                 x->tunnel = NULL;
2152         }
2153 }
2154 EXPORT_SYMBOL(xfrm_state_delete_tunnel);
2155
2156 int xfrm_state_mtu(struct xfrm_state *x, int mtu)
2157 {
2158         const struct xfrm_type *type = READ_ONCE(x->type);
2159
2160         if (x->km.state == XFRM_STATE_VALID &&
2161             type && type->get_mtu)
2162                 return type->get_mtu(x, mtu);
2163
2164         return mtu - x->props.header_len;
2165 }
2166
2167 int __xfrm_init_state(struct xfrm_state *x, bool init_replay)
2168 {
2169         struct xfrm_state_afinfo *afinfo;
2170         struct xfrm_mode *inner_mode;
2171         int family = x->props.family;
2172         int err;
2173
2174         err = -EAFNOSUPPORT;
2175         afinfo = xfrm_state_get_afinfo(family);
2176         if (!afinfo)
2177                 goto error;
2178
2179         err = 0;
2180         if (afinfo->init_flags)
2181                 err = afinfo->init_flags(x);
2182
2183         rcu_read_unlock();
2184
2185         if (err)
2186                 goto error;
2187
2188         err = -EPROTONOSUPPORT;
2189
2190         if (x->sel.family != AF_UNSPEC) {
2191                 inner_mode = xfrm_get_mode(x->props.mode, x->sel.family);
2192                 if (inner_mode == NULL)
2193                         goto error;
2194
2195                 if (!(inner_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2196                     family != x->sel.family) {
2197                         xfrm_put_mode(inner_mode);
2198                         goto error;
2199                 }
2200
2201                 x->inner_mode = inner_mode;
2202         } else {
2203                 struct xfrm_mode *inner_mode_iaf;
2204                 int iafamily = AF_INET;
2205
2206                 inner_mode = xfrm_get_mode(x->props.mode, x->props.family);
2207                 if (inner_mode == NULL)
2208                         goto error;
2209
2210                 if (!(inner_mode->flags & XFRM_MODE_FLAG_TUNNEL)) {
2211                         xfrm_put_mode(inner_mode);
2212                         goto error;
2213                 }
2214                 x->inner_mode = inner_mode;
2215
2216                 if (x->props.family == AF_INET)
2217                         iafamily = AF_INET6;
2218
2219                 inner_mode_iaf = xfrm_get_mode(x->props.mode, iafamily);
2220                 if (inner_mode_iaf) {
2221                         if (inner_mode_iaf->flags & XFRM_MODE_FLAG_TUNNEL)
2222                                 x->inner_mode_iaf = inner_mode_iaf;
2223                         else
2224                                 xfrm_put_mode(inner_mode_iaf);
2225                 }
2226         }
2227
2228         x->type = xfrm_get_type(x->id.proto, family);
2229         if (x->type == NULL)
2230                 goto error;
2231
2232         x->type_offload = xfrm_get_type_offload(x->id.proto, family);
2233
2234         err = x->type->init_state(x);
2235         if (err)
2236                 goto error;
2237
2238         x->outer_mode = xfrm_get_mode(x->props.mode, family);
2239         if (x->outer_mode == NULL) {
2240                 err = -EPROTONOSUPPORT;
2241                 goto error;
2242         }
2243
2244         if (init_replay) {
2245                 err = xfrm_init_replay(x);
2246                 if (err)
2247                         goto error;
2248         }
2249
2250         x->km.state = XFRM_STATE_VALID;
2251
2252 error:
2253         return err;
2254 }
2255
2256 EXPORT_SYMBOL(__xfrm_init_state);
2257
2258 int xfrm_init_state(struct xfrm_state *x)
2259 {
2260         return __xfrm_init_state(x, true);
2261 }
2262
2263 EXPORT_SYMBOL(xfrm_init_state);
2264
2265 int __net_init xfrm_state_init(struct net *net)
2266 {
2267         unsigned int sz;
2268
2269         INIT_LIST_HEAD(&net->xfrm.state_all);
2270
2271         sz = sizeof(struct hlist_head) * 8;
2272
2273         net->xfrm.state_bydst = xfrm_hash_alloc(sz);
2274         if (!net->xfrm.state_bydst)
2275                 goto out_bydst;
2276         net->xfrm.state_bysrc = xfrm_hash_alloc(sz);
2277         if (!net->xfrm.state_bysrc)
2278                 goto out_bysrc;
2279         net->xfrm.state_byspi = xfrm_hash_alloc(sz);
2280         if (!net->xfrm.state_byspi)
2281                 goto out_byspi;
2282         net->xfrm.state_hmask = ((sz / sizeof(struct hlist_head)) - 1);
2283
2284         net->xfrm.state_num = 0;
2285         INIT_WORK(&net->xfrm.state_hash_work, xfrm_hash_resize);
2286         spin_lock_init(&net->xfrm.xfrm_state_lock);
2287         return 0;
2288
2289 out_byspi:
2290         xfrm_hash_free(net->xfrm.state_bysrc, sz);
2291 out_bysrc:
2292         xfrm_hash_free(net->xfrm.state_bydst, sz);
2293 out_bydst:
2294         return -ENOMEM;
2295 }
2296
2297 void xfrm_state_fini(struct net *net)
2298 {
2299         unsigned int sz;
2300
2301         flush_work(&net->xfrm.state_hash_work);
2302         xfrm_state_flush(net, IPSEC_PROTO_ANY, false);
2303         flush_work(&xfrm_state_gc_work);
2304
2305         WARN_ON(!list_empty(&net->xfrm.state_all));
2306
2307         sz = (net->xfrm.state_hmask + 1) * sizeof(struct hlist_head);
2308         WARN_ON(!hlist_empty(net->xfrm.state_byspi));
2309         xfrm_hash_free(net->xfrm.state_byspi, sz);
2310         WARN_ON(!hlist_empty(net->xfrm.state_bysrc));
2311         xfrm_hash_free(net->xfrm.state_bysrc, sz);
2312         WARN_ON(!hlist_empty(net->xfrm.state_bydst));
2313         xfrm_hash_free(net->xfrm.state_bydst, sz);
2314 }
2315
2316 #ifdef CONFIG_AUDITSYSCALL
2317 static void xfrm_audit_helper_sainfo(struct xfrm_state *x,
2318                                      struct audit_buffer *audit_buf)
2319 {
2320         struct xfrm_sec_ctx *ctx = x->security;
2321         u32 spi = ntohl(x->id.spi);
2322
2323         if (ctx)
2324                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2325                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2326
2327         switch (x->props.family) {
2328         case AF_INET:
2329                 audit_log_format(audit_buf, " src=%pI4 dst=%pI4",
2330                                  &x->props.saddr.a4, &x->id.daddr.a4);
2331                 break;
2332         case AF_INET6:
2333                 audit_log_format(audit_buf, " src=%pI6 dst=%pI6",
2334                                  x->props.saddr.a6, x->id.daddr.a6);
2335                 break;
2336         }
2337
2338         audit_log_format(audit_buf, " spi=%u(0x%x)", spi, spi);
2339 }
2340
2341 static void xfrm_audit_helper_pktinfo(struct sk_buff *skb, u16 family,
2342                                       struct audit_buffer *audit_buf)
2343 {
2344         const struct iphdr *iph4;
2345         const struct ipv6hdr *iph6;
2346
2347         switch (family) {
2348         case AF_INET:
2349                 iph4 = ip_hdr(skb);
2350                 audit_log_format(audit_buf, " src=%pI4 dst=%pI4",
2351                                  &iph4->saddr, &iph4->daddr);
2352                 break;
2353         case AF_INET6:
2354                 iph6 = ipv6_hdr(skb);
2355                 audit_log_format(audit_buf,
2356                                  " src=%pI6 dst=%pI6 flowlbl=0x%x%02x%02x",
2357                                  &iph6->saddr, &iph6->daddr,
2358                                  iph6->flow_lbl[0] & 0x0f,
2359                                  iph6->flow_lbl[1],
2360                                  iph6->flow_lbl[2]);
2361                 break;
2362         }
2363 }
2364
2365 void xfrm_audit_state_add(struct xfrm_state *x, int result, bool task_valid)
2366 {
2367         struct audit_buffer *audit_buf;
2368
2369         audit_buf = xfrm_audit_start("SAD-add");
2370         if (audit_buf == NULL)
2371                 return;
2372         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
2373         xfrm_audit_helper_sainfo(x, audit_buf);
2374         audit_log_format(audit_buf, " res=%u", result);
2375         audit_log_end(audit_buf);
2376 }
2377 EXPORT_SYMBOL_GPL(xfrm_audit_state_add);
2378
2379 void xfrm_audit_state_delete(struct xfrm_state *x, int result, bool task_valid)
2380 {
2381         struct audit_buffer *audit_buf;
2382
2383         audit_buf = xfrm_audit_start("SAD-delete");
2384         if (audit_buf == NULL)
2385                 return;
2386         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
2387         xfrm_audit_helper_sainfo(x, audit_buf);
2388         audit_log_format(audit_buf, " res=%u", result);
2389         audit_log_end(audit_buf);
2390 }
2391 EXPORT_SYMBOL_GPL(xfrm_audit_state_delete);
2392
2393 void xfrm_audit_state_replay_overflow(struct xfrm_state *x,
2394                                       struct sk_buff *skb)
2395 {
2396         struct audit_buffer *audit_buf;
2397         u32 spi;
2398
2399         audit_buf = xfrm_audit_start("SA-replay-overflow");
2400         if (audit_buf == NULL)
2401                 return;
2402         xfrm_audit_helper_pktinfo(skb, x->props.family, audit_buf);
2403         /* don't record the sequence number because it's inherent in this kind
2404          * of audit message */
2405         spi = ntohl(x->id.spi);
2406         audit_log_format(audit_buf, " spi=%u(0x%x)", spi, spi);
2407         audit_log_end(audit_buf);
2408 }
2409 EXPORT_SYMBOL_GPL(xfrm_audit_state_replay_overflow);
2410
2411 void xfrm_audit_state_replay(struct xfrm_state *x,
2412                              struct sk_buff *skb, __be32 net_seq)
2413 {
2414         struct audit_buffer *audit_buf;
2415         u32 spi;
2416
2417         audit_buf = xfrm_audit_start("SA-replayed-pkt");
2418         if (audit_buf == NULL)
2419                 return;
2420         xfrm_audit_helper_pktinfo(skb, x->props.family, audit_buf);
2421         spi = ntohl(x->id.spi);
2422         audit_log_format(audit_buf, " spi=%u(0x%x) seqno=%u",
2423                          spi, spi, ntohl(net_seq));
2424         audit_log_end(audit_buf);
2425 }
2426 EXPORT_SYMBOL_GPL(xfrm_audit_state_replay);
2427
2428 void xfrm_audit_state_notfound_simple(struct sk_buff *skb, u16 family)
2429 {
2430         struct audit_buffer *audit_buf;
2431
2432         audit_buf = xfrm_audit_start("SA-notfound");
2433         if (audit_buf == NULL)
2434                 return;
2435         xfrm_audit_helper_pktinfo(skb, family, audit_buf);
2436         audit_log_end(audit_buf);
2437 }
2438 EXPORT_SYMBOL_GPL(xfrm_audit_state_notfound_simple);
2439
2440 void xfrm_audit_state_notfound(struct sk_buff *skb, u16 family,
2441                                __be32 net_spi, __be32 net_seq)
2442 {
2443         struct audit_buffer *audit_buf;
2444         u32 spi;
2445
2446         audit_buf = xfrm_audit_start("SA-notfound");
2447         if (audit_buf == NULL)
2448                 return;
2449         xfrm_audit_helper_pktinfo(skb, family, audit_buf);
2450         spi = ntohl(net_spi);
2451         audit_log_format(audit_buf, " spi=%u(0x%x) seqno=%u",
2452                          spi, spi, ntohl(net_seq));
2453         audit_log_end(audit_buf);
2454 }
2455 EXPORT_SYMBOL_GPL(xfrm_audit_state_notfound);
2456
2457 void xfrm_audit_state_icvfail(struct xfrm_state *x,
2458                               struct sk_buff *skb, u8 proto)
2459 {
2460         struct audit_buffer *audit_buf;
2461         __be32 net_spi;
2462         __be32 net_seq;
2463
2464         audit_buf = xfrm_audit_start("SA-icv-failure");
2465         if (audit_buf == NULL)
2466                 return;
2467         xfrm_audit_helper_pktinfo(skb, x->props.family, audit_buf);
2468         if (xfrm_parse_spi(skb, proto, &net_spi, &net_seq) == 0) {
2469                 u32 spi = ntohl(net_spi);
2470                 audit_log_format(audit_buf, " spi=%u(0x%x) seqno=%u",
2471                                  spi, spi, ntohl(net_seq));
2472         }
2473         audit_log_end(audit_buf);
2474 }
2475 EXPORT_SYMBOL_GPL(xfrm_audit_state_icvfail);
2476 #endif /* CONFIG_AUDITSYSCALL */