net/netfilter/nft_set_rbtree.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2008-2009 Patrick McHardy <kaber@trash.net>
   4  *
   5  * Development of this code funded by Astaro AG (http://www.astaro.com/)
   6  */
   7
   8 #include <linux/kernel.h>
   9 #include <linux/init.h>
  10 #include <linux/module.h>
  11 #include <linux/list.h>
  12 #include <linux/rbtree.h>
  13 #include <linux/netlink.h>
  14 #include <linux/netfilter.h>
  15 #include <linux/netfilter/nf_tables.h>
  16 #include <net/netfilter/nf_tables_core.h>
  17
  18 struct nft_rbtree {
  19         struct rb_root          root;
  20         rwlock_t                lock;
  21         seqcount_rwlock_t       count;
  22         struct delayed_work     gc_work;
  23 };
  24
  25 struct nft_rbtree_elem {
  26         struct rb_node          node;
  27         struct nft_set_ext      ext;
  28 };
  29
  30 static bool nft_rbtree_interval_end(const struct nft_rbtree_elem *rbe)
  31 {
  32         return nft_set_ext_exists(&rbe->ext, NFT_SET_EXT_FLAGS) &&
  33                (*nft_set_ext_flags(&rbe->ext) & NFT_SET_ELEM_INTERVAL_END);
  34 }
  35
  36 static bool nft_rbtree_interval_start(const struct nft_rbtree_elem *rbe)
  37 {
  38         return !nft_rbtree_interval_end(rbe);
  39 }
  40
  41 static int nft_rbtree_cmp(const struct nft_set *set,
  42                           const struct nft_rbtree_elem *e1,
  43                           const struct nft_rbtree_elem *e2)
  44 {
  45         return memcmp(nft_set_ext_key(&e1->ext), nft_set_ext_key(&e2->ext),
  46                       set->klen);
  47 }
  48
  49 static bool nft_rbtree_elem_expired(const struct nft_rbtree_elem *rbe)
  50 {
  51         return nft_set_elem_expired(&rbe->ext) ||
  52                nft_set_elem_is_dead(&rbe->ext);
  53 }
  54
  55 static bool __nft_rbtree_lookup(const struct net *net, const struct nft_set *set,
  56                                 const u32 *key, const struct nft_set_ext **ext,
  57                                 unsigned int seq)
  58 {
  59         struct nft_rbtree *priv = nft_set_priv(set);
  60         const struct nft_rbtree_elem *rbe, *interval = NULL;
  61         u8 genmask = nft_genmask_cur(net);
  62         const struct rb_node *parent;
  63         int d;
  64
  65         parent = rcu_dereference_raw(priv->root.rb_node);
  66         while (parent != NULL) {
  67                 if (read_seqcount_retry(&priv->count, seq))
  68                         return false;
  69
  70                 rbe = rb_entry(parent, struct nft_rbtree_elem, node);
  71
  72                 d = memcmp(nft_set_ext_key(&rbe->ext), key, set->klen);
  73                 if (d < 0) {
  74                         parent = rcu_dereference_raw(parent->rb_left);
  75                         if (interval &&
  76                             !nft_rbtree_cmp(set, rbe, interval) &&
  77                             nft_rbtree_interval_end(rbe) &&
  78                             nft_rbtree_interval_start(interval))
  79                                 continue;
  80                         interval = rbe;
  81                 } else if (d > 0)
  82                         parent = rcu_dereference_raw(parent->rb_right);
  83                 else {
  84                         if (!nft_set_elem_active(&rbe->ext, genmask)) {
  85                                 parent = rcu_dereference_raw(parent->rb_left);
  86                                 continue;
  87                         }
  88
  89                         if (nft_rbtree_elem_expired(rbe))
  90                                 return false;
  91
  92                         if (nft_rbtree_interval_end(rbe)) {
  93                                 if (nft_set_is_anonymous(set))
  94                                         return false;
  95                                 parent = rcu_dereference_raw(parent->rb_left);
  96                                 interval = NULL;
  97                                 continue;
  98                         }
  99
 100                         *ext = &rbe->ext;
 101                         return true;
 102                 }
 103         }
 104
 105         if (set->flags & NFT_SET_INTERVAL && interval != NULL &&
 106             nft_set_elem_active(&interval->ext, genmask) &&
 107             !nft_rbtree_elem_expired(interval) &&
 108             nft_rbtree_interval_start(interval)) {
 109                 *ext = &interval->ext;
 110                 return true;
 111         }
 112
 113         return false;
 114 }
 115
 116 INDIRECT_CALLABLE_SCOPE
 117 bool nft_rbtree_lookup(const struct net *net, const struct nft_set *set,
 118                        const u32 *key, const struct nft_set_ext **ext)
 119 {
 120         struct nft_rbtree *priv = nft_set_priv(set);
 121         unsigned int seq = read_seqcount_begin(&priv->count);
 122         bool ret;
 123
 124         ret = __nft_rbtree_lookup(net, set, key, ext, seq);
 125         if (ret || !read_seqcount_retry(&priv->count, seq))
 126                 return ret;
 127
 128         read_lock_bh(&priv->lock);
 129         seq = read_seqcount_begin(&priv->count);
 130         ret = __nft_rbtree_lookup(net, set, key, ext, seq);
 131         read_unlock_bh(&priv->lock);
 132
 133         return ret;
 134 }
 135
 136 static bool __nft_rbtree_get(const struct net *net, const struct nft_set *set,
 137                              const u32 *key, struct nft_rbtree_elem **elem,
 138                              unsigned int seq, unsigned int flags, u8 genmask)
 139 {
 140         struct nft_rbtree_elem *rbe, *interval = NULL;
 141         struct nft_rbtree *priv = nft_set_priv(set);
 142         const struct rb_node *parent;
 143         const void *this;
 144         int d;
 145
 146         parent = rcu_dereference_raw(priv->root.rb_node);
 147         while (parent != NULL) {
 148                 if (read_seqcount_retry(&priv->count, seq))
 149                         return false;
 150
 151                 rbe = rb_entry(parent, struct nft_rbtree_elem, node);
 152
 153                 this = nft_set_ext_key(&rbe->ext);
 154                 d = memcmp(this, key, set->klen);
 155                 if (d < 0) {
 156                         parent = rcu_dereference_raw(parent->rb_left);
 157                         if (!(flags & NFT_SET_ELEM_INTERVAL_END))
 158                                 interval = rbe;
 159                 } else if (d > 0) {
 160                         parent = rcu_dereference_raw(parent->rb_right);
 161                         if (flags & NFT_SET_ELEM_INTERVAL_END)
 162                                 interval = rbe;
 163                 } else {
 164                         if (!nft_set_elem_active(&rbe->ext, genmask)) {
 165                                 parent = rcu_dereference_raw(parent->rb_left);
 166                                 continue;
 167                         }
 168
 169                         if (nft_set_elem_expired(&rbe->ext))
 170                                 return false;
 171
 172                         if (!nft_set_ext_exists(&rbe->ext, NFT_SET_EXT_FLAGS) ||
 173                             (*nft_set_ext_flags(&rbe->ext) & NFT_SET_ELEM_INTERVAL_END) ==
 174                             (flags & NFT_SET_ELEM_INTERVAL_END)) {
 175                                 *elem = rbe;
 176                                 return true;
 177                         }
 178
 179                         if (nft_rbtree_interval_end(rbe))
 180                                 interval = NULL;
 181
 182                         parent = rcu_dereference_raw(parent->rb_left);
 183                 }
 184         }
 185
 186         if (set->flags & NFT_SET_INTERVAL && interval != NULL &&
 187             nft_set_elem_active(&interval->ext, genmask) &&
 188             !nft_set_elem_expired(&interval->ext) &&
 189             ((!nft_rbtree_interval_end(interval) &&
 190               !(flags & NFT_SET_ELEM_INTERVAL_END)) ||
 191              (nft_rbtree_interval_end(interval) &&
 192               (flags & NFT_SET_ELEM_INTERVAL_END)))) {
 193                 *elem = interval;
 194                 return true;
 195         }
 196
 197         return false;
 198 }
 199
 200 static void *nft_rbtree_get(const struct net *net, const struct nft_set *set,
 201                             const struct nft_set_elem *elem, unsigned int flags)
 202 {
 203         struct nft_rbtree *priv = nft_set_priv(set);
 204         unsigned int seq = read_seqcount_begin(&priv->count);
 205         struct nft_rbtree_elem *rbe = ERR_PTR(-ENOENT);
 206         const u32 *key = (const u32 *)&elem->key.val;
 207         u8 genmask = nft_genmask_cur(net);
 208         bool ret;
 209
 210         ret = __nft_rbtree_get(net, set, key, &rbe, seq, flags, genmask);
 211         if (ret || !read_seqcount_retry(&priv->count, seq))
 212                 return rbe;
 213
 214         read_lock_bh(&priv->lock);
 215         seq = read_seqcount_begin(&priv->count);
 216         ret = __nft_rbtree_get(net, set, key, &rbe, seq, flags, genmask);
 217         if (!ret)
 218                 rbe = ERR_PTR(-ENOENT);
 219         read_unlock_bh(&priv->lock);
 220
 221         return rbe;
 222 }
 223
 224 static void nft_rbtree_gc_remove(struct net *net, struct nft_set *set,
 225                                  struct nft_rbtree *priv,
 226                                  struct nft_rbtree_elem *rbe)
 227 {
 228         struct nft_set_elem elem = {
 229                 .priv   = rbe,
 230         };
 231
 232         nft_setelem_data_deactivate(net, set, &elem);
 233         rb_erase(&rbe->node, &priv->root);
 234 }
 235
 236 static const struct nft_rbtree_elem *
 237 nft_rbtree_gc_elem(const struct nft_set *__set, struct nft_rbtree *priv,
 238                    struct nft_rbtree_elem *rbe, u8 genmask)
 239 {
 240         struct nft_set *set = (struct nft_set *)__set;
 241         struct rb_node *prev = rb_prev(&rbe->node);
 242         struct net *net = read_pnet(&set->net);
 243         struct nft_rbtree_elem *rbe_prev;
 244         struct nft_trans_gc *gc;
 245
 246         gc = nft_trans_gc_alloc(set, 0, GFP_ATOMIC);
 247         if (!gc)
 248                 return ERR_PTR(-ENOMEM);
 249
 250         /* search for end interval coming before this element.
 251          * end intervals don't carry a timeout extension, they
 252          * are coupled with the interval start element.
 253          */
 254         while (prev) {
 255                 rbe_prev = rb_entry(prev, struct nft_rbtree_elem, node);
 256                 if (nft_rbtree_interval_end(rbe_prev) &&
 257                     nft_set_elem_active(&rbe_prev->ext, genmask))
 258                         break;
 259
 260                 prev = rb_prev(prev);
 261         }
 262
 263         rbe_prev = NULL;
 264         if (prev) {
 265                 rbe_prev = rb_entry(prev, struct nft_rbtree_elem, node);
 266                 nft_rbtree_gc_remove(net, set, priv, rbe_prev);
 267
 268                 /* There is always room in this trans gc for this element,
 269                  * memory allocation never actually happens, hence, the warning
 270                  * splat in such case. No need to set NFT_SET_ELEM_DEAD_BIT,
 271                  * this is synchronous gc which never fails.
 272                  */
 273                 gc = nft_trans_gc_queue_sync(gc, GFP_ATOMIC);
 274                 if (WARN_ON_ONCE(!gc))
 275                         return ERR_PTR(-ENOMEM);
 276
 277                 nft_trans_gc_elem_add(gc, rbe_prev);
 278         }
 279
 280         nft_rbtree_gc_remove(net, set, priv, rbe);
 281         gc = nft_trans_gc_queue_sync(gc, GFP_ATOMIC);
 282         if (WARN_ON_ONCE(!gc))
 283                 return ERR_PTR(-ENOMEM);
 284
 285         nft_trans_gc_elem_add(gc, rbe);
 286
 287         nft_trans_gc_queue_sync_done(gc);
 288
 289         return rbe_prev;
 290 }
 291
 292 static bool nft_rbtree_update_first(const struct nft_set *set,
 293                                     struct nft_rbtree_elem *rbe,
 294                                     struct rb_node *first)
 295 {
 296         struct nft_rbtree_elem *first_elem;
 297
 298         first_elem = rb_entry(first, struct nft_rbtree_elem, node);
 299         /* this element is closest to where the new element is to be inserted:
 300          * update the first element for the node list path.
 301          */
 302         if (nft_rbtree_cmp(set, rbe, first_elem) < 0)
 303                 return true;
 304
 305         return false;
 306 }
 307
 308 static int __nft_rbtree_insert(const struct net *net, const struct nft_set *set,
 309                                struct nft_rbtree_elem *new,
 310                                struct nft_set_ext **ext)
 311 {
 312         struct nft_rbtree_elem *rbe, *rbe_le = NULL, *rbe_ge = NULL;
 313         struct rb_node *node, *next, *parent, **p, *first = NULL;
 314         struct nft_rbtree *priv = nft_set_priv(set);
 315         u8 cur_genmask = nft_genmask_cur(net);
 316         u8 genmask = nft_genmask_next(net);
 317         int d;
 318
 319         /* Descend the tree to search for an existing element greater than the
 320          * key value to insert that is greater than the new element. This is the
 321          * first element to walk the ordered elements to find possible overlap.
 322          */
 323         parent = NULL;
 324         p = &priv->root.rb_node;
 325         while (*p != NULL) {
 326                 parent = *p;
 327                 rbe = rb_entry(parent, struct nft_rbtree_elem, node);
 328                 d = nft_rbtree_cmp(set, rbe, new);
 329
 330                 if (d < 0) {
 331                         p = &parent->rb_left;
 332                 } else if (d > 0) {
 333                         if (!first ||
 334                             nft_rbtree_update_first(set, rbe, first))
 335                                 first = &rbe->node;
 336
 337                         p = &parent->rb_right;
 338                 } else {
 339                         if (nft_rbtree_interval_end(rbe))
 340                                 p = &parent->rb_left;
 341                         else
 342                                 p = &parent->rb_right;
 343                 }
 344         }
 345
 346         if (!first)
 347                 first = rb_first(&priv->root);
 348
 349         /* Detect overlap by going through the list of valid tree nodes.
 350          * Values stored in the tree are in reversed order, starting from
 351          * highest to lowest value.
 352          */
 353         for (node = first; node != NULL; node = next) {
 354                 next = rb_next(node);
 355
 356                 rbe = rb_entry(node, struct nft_rbtree_elem, node);
 357
 358                 if (!nft_set_elem_active(&rbe->ext, genmask))
 359                         continue;
 360
 361                 /* perform garbage collection to avoid bogus overlap reports
 362                  * but skip new elements in this transaction.
 363                  */
 364                 if (nft_set_elem_expired(&rbe->ext) &&
 365                     nft_set_elem_active(&rbe->ext, cur_genmask)) {
 366                         const struct nft_rbtree_elem *removed_end;
 367
 368                         removed_end = nft_rbtree_gc_elem(set, priv, rbe, genmask);
 369                         if (IS_ERR(removed_end))
 370                                 return PTR_ERR(removed_end);
 371
 372                         if (removed_end == rbe_le || removed_end == rbe_ge)
 373                                 return -EAGAIN;
 374
 375                         continue;
 376                 }
 377
 378                 d = nft_rbtree_cmp(set, rbe, new);
 379                 if (d == 0) {
 380                         /* Matching end element: no need to look for an
 381                          * overlapping greater or equal element.
 382                          */
 383                         if (nft_rbtree_interval_end(rbe)) {
 384                                 rbe_le = rbe;
 385                                 break;
 386                         }
 387
 388                         /* first element that is greater or equal to key value. */
 389                         if (!rbe_ge) {
 390                                 rbe_ge = rbe;
 391                                 continue;
 392                         }
 393
 394                         /* this is a closer more or equal element, update it. */
 395                         if (nft_rbtree_cmp(set, rbe_ge, new) != 0) {
 396                                 rbe_ge = rbe;
 397                                 continue;
 398                         }
 399
 400                         /* element is equal to key value, make sure flags are
 401                          * the same, an existing more or equal start element
 402                          * must not be replaced by more or equal end element.
 403                          */
 404                         if ((nft_rbtree_interval_start(new) &&
 405                              nft_rbtree_interval_start(rbe_ge)) ||
 406                             (nft_rbtree_interval_end(new) &&
 407                              nft_rbtree_interval_end(rbe_ge))) {
 408                                 rbe_ge = rbe;
 409                                 continue;
 410                         }
 411                 } else if (d > 0) {
 412                         /* annotate element greater than the new element. */
 413                         rbe_ge = rbe;
 414                         continue;
 415                 } else if (d < 0) {
 416                         /* annotate element less than the new element. */
 417                         rbe_le = rbe;
 418                         break;
 419                 }
 420         }
 421
 422         /* - new start element matching existing start element: full overlap
 423          *   reported as -EEXIST, cleared by caller if NLM_F_EXCL is not given.
 424          */
 425         if (rbe_ge && !nft_rbtree_cmp(set, new, rbe_ge) &&
 426             nft_rbtree_interval_start(rbe_ge) == nft_rbtree_interval_start(new)) {
 427                 *ext = &rbe_ge->ext;
 428                 return -EEXIST;
 429         }
 430
 431         /* - new end element matching existing end element: full overlap
 432          *   reported as -EEXIST, cleared by caller if NLM_F_EXCL is not given.
 433          */
 434         if (rbe_le && !nft_rbtree_cmp(set, new, rbe_le) &&
 435             nft_rbtree_interval_end(rbe_le) == nft_rbtree_interval_end(new)) {
 436                 *ext = &rbe_le->ext;
 437                 return -EEXIST;
 438         }
 439
 440         /* - new start element with existing closest, less or equal key value
 441          *   being a start element: partial overlap, reported as -ENOTEMPTY.
 442          *   Anonymous sets allow for two consecutive start element since they
 443          *   are constant, skip them to avoid bogus overlap reports.
 444          */
 445         if (!nft_set_is_anonymous(set) && rbe_le &&
 446             nft_rbtree_interval_start(rbe_le) && nft_rbtree_interval_start(new))
 447                 return -ENOTEMPTY;
 448
 449         /* - new end element with existing closest, less or equal key value
 450          *   being a end element: partial overlap, reported as -ENOTEMPTY.
 451          */
 452         if (rbe_le &&
 453             nft_rbtree_interval_end(rbe_le) && nft_rbtree_interval_end(new))
 454                 return -ENOTEMPTY;
 455
 456         /* - new end element with existing closest, greater or equal key value
 457          *   being an end element: partial overlap, reported as -ENOTEMPTY
 458          */
 459         if (rbe_ge &&
 460             nft_rbtree_interval_end(rbe_ge) && nft_rbtree_interval_end(new))
 461                 return -ENOTEMPTY;
 462
 463         /* Accepted element: pick insertion point depending on key value */
 464         parent = NULL;
 465         p = &priv->root.rb_node;
 466         while (*p != NULL) {
 467                 parent = *p;
 468                 rbe = rb_entry(parent, struct nft_rbtree_elem, node);
 469                 d = nft_rbtree_cmp(set, rbe, new);
 470
 471                 if (d < 0)
 472                         p = &parent->rb_left;
 473                 else if (d > 0)
 474                         p = &parent->rb_right;
 475                 else if (nft_rbtree_interval_end(rbe))
 476                         p = &parent->rb_left;
 477                 else
 478                         p = &parent->rb_right;
 479         }
 480
 481         rb_link_node_rcu(&new->node, parent, p);
 482         rb_insert_color(&new->node, &priv->root);
 483         return 0;
 484 }
 485
 486 static int nft_rbtree_insert(const struct net *net, const struct nft_set *set,
 487                              const struct nft_set_elem *elem,
 488                              struct nft_set_ext **ext)
 489 {
 490         struct nft_rbtree *priv = nft_set_priv(set);
 491         struct nft_rbtree_elem *rbe = elem->priv;
 492         int err;
 493
 494         do {
 495                 if (fatal_signal_pending(current))
 496                         return -EINTR;
 497
 498                 cond_resched();
 499
 500                 write_lock_bh(&priv->lock);
 501                 write_seqcount_begin(&priv->count);
 502                 err = __nft_rbtree_insert(net, set, rbe, ext);
 503                 write_seqcount_end(&priv->count);
 504                 write_unlock_bh(&priv->lock);
 505         } while (err == -EAGAIN);
 506
 507         return err;
 508 }
 509
 510 static void nft_rbtree_remove(const struct net *net,
 511                               const struct nft_set *set,
 512                               const struct nft_set_elem *elem)
 513 {
 514         struct nft_rbtree *priv = nft_set_priv(set);
 515         struct nft_rbtree_elem *rbe = elem->priv;
 516
 517         write_lock_bh(&priv->lock);
 518         write_seqcount_begin(&priv->count);
 519         rb_erase(&rbe->node, &priv->root);
 520         write_seqcount_end(&priv->count);
 521         write_unlock_bh(&priv->lock);
 522 }
 523
 524 static void nft_rbtree_activate(const struct net *net,
 525                                 const struct nft_set *set,
 526                                 const struct nft_set_elem *elem)
 527 {
 528         struct nft_rbtree_elem *rbe = elem->priv;
 529
 530         nft_set_elem_change_active(net, set, &rbe->ext);
 531 }
 532
 533 static bool nft_rbtree_flush(const struct net *net,
 534                              const struct nft_set *set, void *priv)
 535 {
 536         struct nft_rbtree_elem *rbe = priv;
 537
 538         nft_set_elem_change_active(net, set, &rbe->ext);
 539
 540         return true;
 541 }
 542
 543 static void *nft_rbtree_deactivate(const struct net *net,
 544                                    const struct nft_set *set,
 545                                    const struct nft_set_elem *elem)
 546 {
 547         const struct nft_rbtree *priv = nft_set_priv(set);
 548         const struct rb_node *parent = priv->root.rb_node;
 549         struct nft_rbtree_elem *rbe, *this = elem->priv;
 550         u8 genmask = nft_genmask_next(net);
 551         int d;
 552
 553         while (parent != NULL) {
 554                 rbe = rb_entry(parent, struct nft_rbtree_elem, node);
 555
 556                 d = memcmp(nft_set_ext_key(&rbe->ext), &elem->key.val,
 557                                            set->klen);
 558                 if (d < 0)
 559                         parent = parent->rb_left;
 560                 else if (d > 0)
 561                         parent = parent->rb_right;
 562                 else {
 563                         if (nft_rbtree_interval_end(rbe) &&
 564                             nft_rbtree_interval_start(this)) {
 565                                 parent = parent->rb_left;
 566                                 continue;
 567                         } else if (nft_rbtree_interval_start(rbe) &&
 568                                    nft_rbtree_interval_end(this)) {
 569                                 parent = parent->rb_right;
 570                                 continue;
 571                         } else if (!nft_set_elem_active(&rbe->ext, genmask)) {
 572                                 parent = parent->rb_left;
 573                                 continue;
 574                         }
 575                         nft_rbtree_flush(net, set, rbe);
 576                         return rbe;
 577                 }
 578         }
 579         return NULL;
 580 }
 581
 582 static void nft_rbtree_walk(const struct nft_ctx *ctx,
 583                             struct nft_set *set,
 584                             struct nft_set_iter *iter)
 585 {
 586         struct nft_rbtree *priv = nft_set_priv(set);
 587         struct nft_rbtree_elem *rbe;
 588         struct nft_set_elem elem;
 589         struct rb_node *node;
 590
 591         read_lock_bh(&priv->lock);
 592         for (node = rb_first(&priv->root); node != NULL; node = rb_next(node)) {
 593                 rbe = rb_entry(node, struct nft_rbtree_elem, node);
 594
 595                 if (iter->count < iter->skip)
 596                         goto cont;
 597                 if (!nft_set_elem_active(&rbe->ext, iter->genmask))
 598                         goto cont;
 599
 600                 elem.priv = rbe;
 601
 602                 iter->err = iter->fn(ctx, set, iter, &elem);
 603                 if (iter->err < 0) {
 604                         read_unlock_bh(&priv->lock);
 605                         return;
 606                 }
 607 cont:
 608                 iter->count++;
 609         }
 610         read_unlock_bh(&priv->lock);
 611 }
 612
 613 static void nft_rbtree_gc(struct work_struct *work)
 614 {
 615         struct nft_rbtree_elem *rbe, *rbe_end = NULL;
 616         struct nftables_pernet *nft_net;
 617         struct nft_rbtree *priv;
 618         struct nft_trans_gc *gc;
 619         struct rb_node *node;
 620         struct nft_set *set;
 621         unsigned int gc_seq;
 622         struct net *net;
 623
 624         priv = container_of(work, struct nft_rbtree, gc_work.work);
 625         set  = nft_set_container_of(priv);
 626         net  = read_pnet(&set->net);
 627         nft_net = nft_pernet(net);
 628         gc_seq  = READ_ONCE(nft_net->gc_seq);
 629
 630         if (nft_set_gc_is_pending(set))
 631                 goto done;
 632
 633         gc = nft_trans_gc_alloc(set, gc_seq, GFP_KERNEL);
 634         if (!gc)
 635                 goto done;
 636
 637         read_lock_bh(&priv->lock);
 638         for (node = rb_first(&priv->root); node != NULL; node = rb_next(node)) {
 639
 640                 /* Ruleset has been updated, try later. */
 641                 if (READ_ONCE(nft_net->gc_seq) != gc_seq) {
 642                         nft_trans_gc_destroy(gc);
 643                         gc = NULL;
 644                         goto try_later;
 645                 }
 646
 647                 rbe = rb_entry(node, struct nft_rbtree_elem, node);
 648
 649                 if (nft_set_elem_is_dead(&rbe->ext))
 650                         goto dead_elem;
 651
 652                 /* elements are reversed in the rbtree for historical reasons,
 653                  * from highest to lowest value, that is why end element is
 654                  * always visited before the start element.
 655                  */
 656                 if (nft_rbtree_interval_end(rbe)) {
 657                         rbe_end = rbe;
 658                         continue;
 659                 }
 660                 if (!nft_set_elem_expired(&rbe->ext))
 661                         continue;
 662
 663                 nft_set_elem_dead(&rbe->ext);
 664
 665                 if (!rbe_end)
 666                         continue;
 667
 668                 nft_set_elem_dead(&rbe_end->ext);
 669
 670                 gc = nft_trans_gc_queue_async(gc, gc_seq, GFP_ATOMIC);
 671                 if (!gc)
 672                         goto try_later;
 673
 674                 nft_trans_gc_elem_add(gc, rbe_end);
 675                 rbe_end = NULL;
 676 dead_elem:
 677                 gc = nft_trans_gc_queue_async(gc, gc_seq, GFP_ATOMIC);
 678                 if (!gc)
 679                         goto try_later;
 680
 681                 nft_trans_gc_elem_add(gc, rbe);
 682         }
 683
 684         gc = nft_trans_gc_catchall_async(gc, gc_seq);
 685
 686 try_later:
 687         read_unlock_bh(&priv->lock);
 688
 689         if (gc)
 690                 nft_trans_gc_queue_async_done(gc);
 691 done:
 692         queue_delayed_work(system_power_efficient_wq, &priv->gc_work,
 693                            nft_set_gc_interval(set));
 694 }
 695
 696 static u64 nft_rbtree_privsize(const struct nlattr * const nla[],
 697                                const struct nft_set_desc *desc)
 698 {
 699         return sizeof(struct nft_rbtree);
 700 }
 701
 702 static int nft_rbtree_init(const struct nft_set *set,
 703                            const struct nft_set_desc *desc,
 704                            const struct nlattr * const nla[])
 705 {
 706         struct nft_rbtree *priv = nft_set_priv(set);
 707
 708         rwlock_init(&priv->lock);
 709         seqcount_rwlock_init(&priv->count, &priv->lock);
 710         priv->root = RB_ROOT;
 711
 712         INIT_DEFERRABLE_WORK(&priv->gc_work, nft_rbtree_gc);
 713         if (set->flags & NFT_SET_TIMEOUT)
 714                 queue_delayed_work(system_power_efficient_wq, &priv->gc_work,
 715                                    nft_set_gc_interval(set));
 716
 717         return 0;
 718 }
 719
 720 static void nft_rbtree_destroy(const struct nft_ctx *ctx,
 721                                const struct nft_set *set)
 722 {
 723         struct nft_rbtree *priv = nft_set_priv(set);
 724         struct nft_rbtree_elem *rbe;
 725         struct rb_node *node;
 726
 727         cancel_delayed_work_sync(&priv->gc_work);
 728         rcu_barrier();
 729         while ((node = priv->root.rb_node) != NULL) {
 730                 rb_erase(node, &priv->root);
 731                 rbe = rb_entry(node, struct nft_rbtree_elem, node);
 732                 nf_tables_set_elem_destroy(ctx, set, rbe);
 733         }
 734 }
 735
 736 static bool nft_rbtree_estimate(const struct nft_set_desc *desc, u32 features,
 737                                 struct nft_set_estimate *est)
 738 {
 739         if (desc->field_count > 1)
 740                 return false;
 741
 742         if (desc->size)
 743                 est->size = sizeof(struct nft_rbtree) +
 744                             desc->size * sizeof(struct nft_rbtree_elem);
 745         else
 746                 est->size = ~0;
 747
 748         est->lookup = NFT_SET_CLASS_O_LOG_N;
 749         est->space  = NFT_SET_CLASS_O_N;
 750
 751         return true;
 752 }
 753
 754 const struct nft_set_type nft_set_rbtree_type = {
 755         .features       = NFT_SET_INTERVAL | NFT_SET_MAP | NFT_SET_OBJECT | NFT_SET_TIMEOUT,
 756         .ops            = {
 757                 .privsize       = nft_rbtree_privsize,
 758                 .elemsize       = offsetof(struct nft_rbtree_elem, ext),
 759                 .estimate       = nft_rbtree_estimate,
 760                 .init           = nft_rbtree_init,
 761                 .destroy        = nft_rbtree_destroy,
 762                 .insert         = nft_rbtree_insert,
 763                 .remove         = nft_rbtree_remove,
 764                 .deactivate     = nft_rbtree_deactivate,
 765                 .flush          = nft_rbtree_flush,
 766                 .activate       = nft_rbtree_activate,
 767                 .lookup         = nft_rbtree_lookup,
 768                 .walk           = nft_rbtree_walk,
 769                 .get            = nft_rbtree_get,
 770         },
 771 };