net/sched/cls_api.c

   1 /*
   2  * net/sched/cls_api.c  Packet classifier API.
   3  *
   4  *              This program is free software; you can redistribute it and/or
   5  *              modify it under the terms of the GNU General Public License
   6  *              as published by the Free Software Foundation; either version
   7  *              2 of the License, or (at your option) any later version.
   8  *
   9  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  10  *
  11  * Changes:
  12  *
  13  * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
  14  *
  15  */
  16
  17 #include <linux/module.h>
  18 #include <linux/types.h>
  19 #include <linux/kernel.h>
  20 #include <linux/string.h>
  21 #include <linux/errno.h>
  22 #include <linux/err.h>
  23 #include <linux/skbuff.h>
  24 #include <linux/init.h>
  25 #include <linux/kmod.h>
  26 #include <linux/slab.h>
  27 #include <linux/idr.h>
  28 #include <linux/rhashtable.h>
  29 #include <net/net_namespace.h>
  30 #include <net/sock.h>
  31 #include <net/netlink.h>
  32 #include <net/pkt_sched.h>
  33 #include <net/pkt_cls.h>
  34 #include <net/tc_act/tc_pedit.h>
  35 #include <net/tc_act/tc_mirred.h>
  36 #include <net/tc_act/tc_vlan.h>
  37 #include <net/tc_act/tc_tunnel_key.h>
  38 #include <net/tc_act/tc_csum.h>
  39 #include <net/tc_act/tc_gact.h>
  40 #include <net/tc_act/tc_police.h>
  41 #include <net/tc_act/tc_sample.h>
  42 #include <net/tc_act/tc_skbedit.h>
  43
  44 extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
  45
  46 /* The list of all installed classifier types */
  47 static LIST_HEAD(tcf_proto_base);
  48
  49 /* Protects list of registered TC modules. It is pure SMP lock. */
  50 static DEFINE_RWLOCK(cls_mod_lock);
  51
  52 /* Find classifier type by string name */
  53
  54 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
  55 {
  56         const struct tcf_proto_ops *t, *res = NULL;
  57
  58         if (kind) {
  59                 read_lock(&cls_mod_lock);
  60                 list_for_each_entry(t, &tcf_proto_base, head) {
  61                         if (strcmp(kind, t->kind) == 0) {
  62                                 if (try_module_get(t->owner))
  63                                         res = t;
  64                                 break;
  65                         }
  66                 }
  67                 read_unlock(&cls_mod_lock);
  68         }
  69         return res;
  70 }
  71
  72 static const struct tcf_proto_ops *
  73 tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
  74                      struct netlink_ext_ack *extack)
  75 {
  76         const struct tcf_proto_ops *ops;
  77
  78         ops = __tcf_proto_lookup_ops(kind);
  79         if (ops)
  80                 return ops;
  81 #ifdef CONFIG_MODULES
  82         if (rtnl_held)
  83                 rtnl_unlock();
  84         request_module("cls_%s", kind);
  85         if (rtnl_held)
  86                 rtnl_lock();
  87         ops = __tcf_proto_lookup_ops(kind);
  88         /* We dropped the RTNL semaphore in order to perform
  89          * the module load. So, even if we succeeded in loading
  90          * the module we have to replay the request. We indicate
  91          * this using -EAGAIN.
  92          */
  93         if (ops) {
  94                 module_put(ops->owner);
  95                 return ERR_PTR(-EAGAIN);
  96         }
  97 #endif
  98         NL_SET_ERR_MSG(extack, "TC classifier not found");
  99         return ERR_PTR(-ENOENT);
 100 }
 101
 102 /* Register(unregister) new classifier type */
 103
 104 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
 105 {
 106         struct tcf_proto_ops *t;
 107         int rc = -EEXIST;
 108
 109         write_lock(&cls_mod_lock);
 110         list_for_each_entry(t, &tcf_proto_base, head)
 111                 if (!strcmp(ops->kind, t->kind))
 112                         goto out;
 113
 114         list_add_tail(&ops->head, &tcf_proto_base);
 115         rc = 0;
 116 out:
 117         write_unlock(&cls_mod_lock);
 118         return rc;
 119 }
 120 EXPORT_SYMBOL(register_tcf_proto_ops);
 121
 122 static struct workqueue_struct *tc_filter_wq;
 123
 124 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
 125 {
 126         struct tcf_proto_ops *t;
 127         int rc = -ENOENT;
 128
 129         /* Wait for outstanding call_rcu()s, if any, from a
 130          * tcf_proto_ops's destroy() handler.
 131          */
 132         rcu_barrier();
 133         flush_workqueue(tc_filter_wq);
 134
 135         write_lock(&cls_mod_lock);
 136         list_for_each_entry(t, &tcf_proto_base, head) {
 137                 if (t == ops) {
 138                         list_del(&t->head);
 139                         rc = 0;
 140                         break;
 141                 }
 142         }
 143         write_unlock(&cls_mod_lock);
 144         return rc;
 145 }
 146 EXPORT_SYMBOL(unregister_tcf_proto_ops);
 147
 148 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
 149 {
 150         INIT_RCU_WORK(rwork, func);
 151         return queue_rcu_work(tc_filter_wq, rwork);
 152 }
 153 EXPORT_SYMBOL(tcf_queue_work);
 154
 155 /* Select new prio value from the range, managed by kernel. */
 156
 157 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
 158 {
 159         u32 first = TC_H_MAKE(0xC0000000U, 0U);
 160
 161         if (tp)
 162                 first = tp->prio - 1;
 163
 164         return TC_H_MAJ(first);
 165 }
 166
 167 static bool tcf_proto_is_unlocked(const char *kind)
 168 {
 169         const struct tcf_proto_ops *ops;
 170         bool ret;
 171
 172         ops = tcf_proto_lookup_ops(kind, false, NULL);
 173         /* On error return false to take rtnl lock. Proto lookup/create
 174          * functions will perform lookup again and properly handle errors.
 175          */
 176         if (IS_ERR(ops))
 177                 return false;
 178
 179         ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
 180         module_put(ops->owner);
 181         return ret;
 182 }
 183
 184 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
 185                                           u32 prio, struct tcf_chain *chain,
 186                                           bool rtnl_held,
 187                                           struct netlink_ext_ack *extack)
 188 {
 189         struct tcf_proto *tp;
 190         int err;
 191
 192         tp = kzalloc(sizeof(*tp), GFP_KERNEL);
 193         if (!tp)
 194                 return ERR_PTR(-ENOBUFS);
 195
 196         tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
 197         if (IS_ERR(tp->ops)) {
 198                 err = PTR_ERR(tp->ops);
 199                 goto errout;
 200         }
 201         tp->classify = tp->ops->classify;
 202         tp->protocol = protocol;
 203         tp->prio = prio;
 204         tp->chain = chain;
 205         spin_lock_init(&tp->lock);
 206         refcount_set(&tp->refcnt, 1);
 207
 208         err = tp->ops->init(tp);
 209         if (err) {
 210                 module_put(tp->ops->owner);
 211                 goto errout;
 212         }
 213         return tp;
 214
 215 errout:
 216         kfree(tp);
 217         return ERR_PTR(err);
 218 }
 219
 220 static void tcf_proto_get(struct tcf_proto *tp)
 221 {
 222         refcount_inc(&tp->refcnt);
 223 }
 224
 225 static void tcf_chain_put(struct tcf_chain *chain);
 226
 227 static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
 228                               struct netlink_ext_ack *extack)
 229 {
 230         tp->ops->destroy(tp, rtnl_held, extack);
 231         tcf_chain_put(tp->chain);
 232         module_put(tp->ops->owner);
 233         kfree_rcu(tp, rcu);
 234 }
 235
 236 static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
 237                           struct netlink_ext_ack *extack)
 238 {
 239         if (refcount_dec_and_test(&tp->refcnt))
 240                 tcf_proto_destroy(tp, rtnl_held, extack);
 241 }
 242
 243 static int walker_check_empty(struct tcf_proto *tp, void *fh,
 244                               struct tcf_walker *arg)
 245 {
 246         if (fh) {
 247                 arg->nonempty = true;
 248                 return -1;
 249         }
 250         return 0;
 251 }
 252
 253 static bool tcf_proto_is_empty(struct tcf_proto *tp, bool rtnl_held)
 254 {
 255         struct tcf_walker walker = { .fn = walker_check_empty, };
 256
 257         if (tp->ops->walk) {
 258                 tp->ops->walk(tp, &walker, rtnl_held);
 259                 return !walker.nonempty;
 260         }
 261         return true;
 262 }
 263
 264 static bool tcf_proto_check_delete(struct tcf_proto *tp, bool rtnl_held)
 265 {
 266         spin_lock(&tp->lock);
 267         if (tcf_proto_is_empty(tp, rtnl_held))
 268                 tp->deleting = true;
 269         spin_unlock(&tp->lock);
 270         return tp->deleting;
 271 }
 272
 273 static void tcf_proto_mark_delete(struct tcf_proto *tp)
 274 {
 275         spin_lock(&tp->lock);
 276         tp->deleting = true;
 277         spin_unlock(&tp->lock);
 278 }
 279
 280 static bool tcf_proto_is_deleting(struct tcf_proto *tp)
 281 {
 282         bool deleting;
 283
 284         spin_lock(&tp->lock);
 285         deleting = tp->deleting;
 286         spin_unlock(&tp->lock);
 287
 288         return deleting;
 289 }
 290
 291 #define ASSERT_BLOCK_LOCKED(block)                                      \
 292         lockdep_assert_held(&(block)->lock)
 293
 294 struct tcf_filter_chain_list_item {
 295         struct list_head list;
 296         tcf_chain_head_change_t *chain_head_change;
 297         void *chain_head_change_priv;
 298 };
 299
 300 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
 301                                           u32 chain_index)
 302 {
 303         struct tcf_chain *chain;
 304
 305         ASSERT_BLOCK_LOCKED(block);
 306
 307         chain = kzalloc(sizeof(*chain), GFP_KERNEL);
 308         if (!chain)
 309                 return NULL;
 310         list_add_tail(&chain->list, &block->chain_list);
 311         mutex_init(&chain->filter_chain_lock);
 312         chain->block = block;
 313         chain->index = chain_index;
 314         chain->refcnt = 1;
 315         if (!chain->index)
 316                 block->chain0.chain = chain;
 317         return chain;
 318 }
 319
 320 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
 321                                        struct tcf_proto *tp_head)
 322 {
 323         if (item->chain_head_change)
 324                 item->chain_head_change(tp_head, item->chain_head_change_priv);
 325 }
 326
 327 static void tcf_chain0_head_change(struct tcf_chain *chain,
 328                                    struct tcf_proto *tp_head)
 329 {
 330         struct tcf_filter_chain_list_item *item;
 331         struct tcf_block *block = chain->block;
 332
 333         if (chain->index)
 334                 return;
 335
 336         mutex_lock(&block->lock);
 337         list_for_each_entry(item, &block->chain0.filter_chain_list, list)
 338                 tcf_chain_head_change_item(item, tp_head);
 339         mutex_unlock(&block->lock);
 340 }
 341
 342 /* Returns true if block can be safely freed. */
 343
 344 static bool tcf_chain_detach(struct tcf_chain *chain)
 345 {
 346         struct tcf_block *block = chain->block;
 347
 348         ASSERT_BLOCK_LOCKED(block);
 349
 350         list_del(&chain->list);
 351         if (!chain->index)
 352                 block->chain0.chain = NULL;
 353
 354         if (list_empty(&block->chain_list) &&
 355             refcount_read(&block->refcnt) == 0)
 356                 return true;
 357
 358         return false;
 359 }
 360
 361 static void tcf_block_destroy(struct tcf_block *block)
 362 {
 363         mutex_destroy(&block->lock);
 364         kfree_rcu(block, rcu);
 365 }
 366
 367 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
 368 {
 369         struct tcf_block *block = chain->block;
 370
 371         mutex_destroy(&chain->filter_chain_lock);
 372         kfree_rcu(chain, rcu);
 373         if (free_block)
 374                 tcf_block_destroy(block);
 375 }
 376
 377 static void tcf_chain_hold(struct tcf_chain *chain)
 378 {
 379         ASSERT_BLOCK_LOCKED(chain->block);
 380
 381         ++chain->refcnt;
 382 }
 383
 384 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
 385 {
 386         ASSERT_BLOCK_LOCKED(chain->block);
 387
 388         /* In case all the references are action references, this
 389          * chain should not be shown to the user.
 390          */
 391         return chain->refcnt == chain->action_refcnt;
 392 }
 393
 394 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
 395                                           u32 chain_index)
 396 {
 397         struct tcf_chain *chain;
 398
 399         ASSERT_BLOCK_LOCKED(block);
 400
 401         list_for_each_entry(chain, &block->chain_list, list) {
 402                 if (chain->index == chain_index)
 403                         return chain;
 404         }
 405         return NULL;
 406 }
 407
 408 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
 409                            u32 seq, u16 flags, int event, bool unicast);
 410
 411 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
 412                                          u32 chain_index, bool create,
 413                                          bool by_act)
 414 {
 415         struct tcf_chain *chain = NULL;
 416         bool is_first_reference;
 417
 418         mutex_lock(&block->lock);
 419         chain = tcf_chain_lookup(block, chain_index);
 420         if (chain) {
 421                 tcf_chain_hold(chain);
 422         } else {
 423                 if (!create)
 424                         goto errout;
 425                 chain = tcf_chain_create(block, chain_index);
 426                 if (!chain)
 427                         goto errout;
 428         }
 429
 430         if (by_act)
 431                 ++chain->action_refcnt;
 432         is_first_reference = chain->refcnt - chain->action_refcnt == 1;
 433         mutex_unlock(&block->lock);
 434
 435         /* Send notification only in case we got the first
 436          * non-action reference. Until then, the chain acts only as
 437          * a placeholder for actions pointing to it and user ought
 438          * not know about them.
 439          */
 440         if (is_first_reference && !by_act)
 441                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
 442                                 RTM_NEWCHAIN, false);
 443
 444         return chain;
 445
 446 errout:
 447         mutex_unlock(&block->lock);
 448         return chain;
 449 }
 450
 451 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
 452                                        bool create)
 453 {
 454         return __tcf_chain_get(block, chain_index, create, false);
 455 }
 456
 457 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
 458 {
 459         return __tcf_chain_get(block, chain_index, true, true);
 460 }
 461 EXPORT_SYMBOL(tcf_chain_get_by_act);
 462
 463 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
 464                                void *tmplt_priv);
 465 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
 466                                   void *tmplt_priv, u32 chain_index,
 467                                   struct tcf_block *block, struct sk_buff *oskb,
 468                                   u32 seq, u16 flags, bool unicast);
 469
 470 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
 471                             bool explicitly_created)
 472 {
 473         struct tcf_block *block = chain->block;
 474         const struct tcf_proto_ops *tmplt_ops;
 475         bool free_block = false;
 476         unsigned int refcnt;
 477         void *tmplt_priv;
 478
 479         mutex_lock(&block->lock);
 480         if (explicitly_created) {
 481                 if (!chain->explicitly_created) {
 482                         mutex_unlock(&block->lock);
 483                         return;
 484                 }
 485                 chain->explicitly_created = false;
 486         }
 487
 488         if (by_act)
 489                 chain->action_refcnt--;
 490
 491         /* tc_chain_notify_delete can't be called while holding block lock.
 492          * However, when block is unlocked chain can be changed concurrently, so
 493          * save these to temporary variables.
 494          */
 495         refcnt = --chain->refcnt;
 496         tmplt_ops = chain->tmplt_ops;
 497         tmplt_priv = chain->tmplt_priv;
 498
 499         /* The last dropped non-action reference will trigger notification. */
 500         if (refcnt - chain->action_refcnt == 0 && !by_act) {
 501                 tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
 502                                        block, NULL, 0, 0, false);
 503                 /* Last reference to chain, no need to lock. */
 504                 chain->flushing = false;
 505         }
 506
 507         if (refcnt == 0)
 508                 free_block = tcf_chain_detach(chain);
 509         mutex_unlock(&block->lock);
 510
 511         if (refcnt == 0) {
 512                 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
 513                 tcf_chain_destroy(chain, free_block);
 514         }
 515 }
 516
 517 static void tcf_chain_put(struct tcf_chain *chain)
 518 {
 519         __tcf_chain_put(chain, false, false);
 520 }
 521
 522 void tcf_chain_put_by_act(struct tcf_chain *chain)
 523 {
 524         __tcf_chain_put(chain, true, false);
 525 }
 526 EXPORT_SYMBOL(tcf_chain_put_by_act);
 527
 528 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
 529 {
 530         __tcf_chain_put(chain, false, true);
 531 }
 532
 533 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
 534 {
 535         struct tcf_proto *tp, *tp_next;
 536
 537         mutex_lock(&chain->filter_chain_lock);
 538         tp = tcf_chain_dereference(chain->filter_chain, chain);
 539         RCU_INIT_POINTER(chain->filter_chain, NULL);
 540         tcf_chain0_head_change(chain, NULL);
 541         chain->flushing = true;
 542         mutex_unlock(&chain->filter_chain_lock);
 543
 544         while (tp) {
 545                 tp_next = rcu_dereference_protected(tp->next, 1);
 546                 tcf_proto_put(tp, rtnl_held, NULL);
 547                 tp = tp_next;
 548         }
 549 }
 550
 551 static struct tcf_block *tc_dev_ingress_block(struct net_device *dev)
 552 {
 553         const struct Qdisc_class_ops *cops;
 554         struct Qdisc *qdisc;
 555
 556         if (!dev_ingress_queue(dev))
 557                 return NULL;
 558
 559         qdisc = dev_ingress_queue(dev)->qdisc_sleeping;
 560         if (!qdisc)
 561                 return NULL;
 562
 563         cops = qdisc->ops->cl_ops;
 564         if (!cops)
 565                 return NULL;
 566
 567         if (!cops->tcf_block)
 568                 return NULL;
 569
 570         return cops->tcf_block(qdisc, TC_H_MIN_INGRESS, NULL);
 571 }
 572
 573 static struct rhashtable indr_setup_block_ht;
 574
 575 struct tc_indr_block_dev {
 576         struct rhash_head ht_node;
 577         struct net_device *dev;
 578         unsigned int refcnt;
 579         struct list_head cb_list;
 580         struct tcf_block *block;
 581 };
 582
 583 struct tc_indr_block_cb {
 584         struct list_head list;
 585         void *cb_priv;
 586         tc_indr_block_bind_cb_t *cb;
 587         void *cb_ident;
 588 };
 589
 590 static const struct rhashtable_params tc_indr_setup_block_ht_params = {
 591         .key_offset     = offsetof(struct tc_indr_block_dev, dev),
 592         .head_offset    = offsetof(struct tc_indr_block_dev, ht_node),
 593         .key_len        = sizeof(struct net_device *),
 594 };
 595
 596 static struct tc_indr_block_dev *
 597 tc_indr_block_dev_lookup(struct net_device *dev)
 598 {
 599         return rhashtable_lookup_fast(&indr_setup_block_ht, &dev,
 600                                       tc_indr_setup_block_ht_params);
 601 }
 602
 603 static struct tc_indr_block_dev *tc_indr_block_dev_get(struct net_device *dev)
 604 {
 605         struct tc_indr_block_dev *indr_dev;
 606
 607         indr_dev = tc_indr_block_dev_lookup(dev);
 608         if (indr_dev)
 609                 goto inc_ref;
 610
 611         indr_dev = kzalloc(sizeof(*indr_dev), GFP_KERNEL);
 612         if (!indr_dev)
 613                 return NULL;
 614
 615         INIT_LIST_HEAD(&indr_dev->cb_list);
 616         indr_dev->dev = dev;
 617         indr_dev->block = tc_dev_ingress_block(dev);
 618         if (rhashtable_insert_fast(&indr_setup_block_ht, &indr_dev->ht_node,
 619                                    tc_indr_setup_block_ht_params)) {
 620                 kfree(indr_dev);
 621                 return NULL;
 622         }
 623
 624 inc_ref:
 625         indr_dev->refcnt++;
 626         return indr_dev;
 627 }
 628
 629 static void tc_indr_block_dev_put(struct tc_indr_block_dev *indr_dev)
 630 {
 631         if (--indr_dev->refcnt)
 632                 return;
 633
 634         rhashtable_remove_fast(&indr_setup_block_ht, &indr_dev->ht_node,
 635                                tc_indr_setup_block_ht_params);
 636         kfree(indr_dev);
 637 }
 638
 639 static struct tc_indr_block_cb *
 640 tc_indr_block_cb_lookup(struct tc_indr_block_dev *indr_dev,
 641                         tc_indr_block_bind_cb_t *cb, void *cb_ident)
 642 {
 643         struct tc_indr_block_cb *indr_block_cb;
 644
 645         list_for_each_entry(indr_block_cb, &indr_dev->cb_list, list)
 646                 if (indr_block_cb->cb == cb &&
 647                     indr_block_cb->cb_ident == cb_ident)
 648                         return indr_block_cb;
 649         return NULL;
 650 }
 651
 652 static struct tc_indr_block_cb *
 653 tc_indr_block_cb_add(struct tc_indr_block_dev *indr_dev, void *cb_priv,
 654                      tc_indr_block_bind_cb_t *cb, void *cb_ident)
 655 {
 656         struct tc_indr_block_cb *indr_block_cb;
 657
 658         indr_block_cb = tc_indr_block_cb_lookup(indr_dev, cb, cb_ident);
 659         if (indr_block_cb)
 660                 return ERR_PTR(-EEXIST);
 661
 662         indr_block_cb = kzalloc(sizeof(*indr_block_cb), GFP_KERNEL);
 663         if (!indr_block_cb)
 664                 return ERR_PTR(-ENOMEM);
 665
 666         indr_block_cb->cb_priv = cb_priv;
 667         indr_block_cb->cb = cb;
 668         indr_block_cb->cb_ident = cb_ident;
 669         list_add(&indr_block_cb->list, &indr_dev->cb_list);
 670
 671         return indr_block_cb;
 672 }
 673
 674 static void tc_indr_block_cb_del(struct tc_indr_block_cb *indr_block_cb)
 675 {
 676         list_del(&indr_block_cb->list);
 677         kfree(indr_block_cb);
 678 }
 679
 680 static void tc_indr_block_ing_cmd(struct tc_indr_block_dev *indr_dev,
 681                                   struct tc_indr_block_cb *indr_block_cb,
 682                                   enum tc_block_command command)
 683 {
 684         struct tc_block_offload bo = {
 685                 .command        = command,
 686                 .binder_type    = TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS,
 687                 .block          = indr_dev->block,
 688         };
 689
 690         if (!indr_dev->block)
 691                 return;
 692
 693         indr_block_cb->cb(indr_dev->dev, indr_block_cb->cb_priv, TC_SETUP_BLOCK,
 694                           &bo);
 695 }
 696
 697 int __tc_indr_block_cb_register(struct net_device *dev, void *cb_priv,
 698                                 tc_indr_block_bind_cb_t *cb, void *cb_ident)
 699 {
 700         struct tc_indr_block_cb *indr_block_cb;
 701         struct tc_indr_block_dev *indr_dev;
 702         int err;
 703
 704         indr_dev = tc_indr_block_dev_get(dev);
 705         if (!indr_dev)
 706                 return -ENOMEM;
 707
 708         indr_block_cb = tc_indr_block_cb_add(indr_dev, cb_priv, cb, cb_ident);
 709         err = PTR_ERR_OR_ZERO(indr_block_cb);
 710         if (err)
 711                 goto err_dev_put;
 712
 713         tc_indr_block_ing_cmd(indr_dev, indr_block_cb, TC_BLOCK_BIND);
 714         return 0;
 715
 716 err_dev_put:
 717         tc_indr_block_dev_put(indr_dev);
 718         return err;
 719 }
 720 EXPORT_SYMBOL_GPL(__tc_indr_block_cb_register);
 721
 722 int tc_indr_block_cb_register(struct net_device *dev, void *cb_priv,
 723                               tc_indr_block_bind_cb_t *cb, void *cb_ident)
 724 {
 725         int err;
 726
 727         rtnl_lock();
 728         err = __tc_indr_block_cb_register(dev, cb_priv, cb, cb_ident);
 729         rtnl_unlock();
 730
 731         return err;
 732 }
 733 EXPORT_SYMBOL_GPL(tc_indr_block_cb_register);
 734
 735 void __tc_indr_block_cb_unregister(struct net_device *dev,
 736                                    tc_indr_block_bind_cb_t *cb, void *cb_ident)
 737 {
 738         struct tc_indr_block_cb *indr_block_cb;
 739         struct tc_indr_block_dev *indr_dev;
 740
 741         indr_dev = tc_indr_block_dev_lookup(dev);
 742         if (!indr_dev)
 743                 return;
 744
 745         indr_block_cb = tc_indr_block_cb_lookup(indr_dev, cb, cb_ident);
 746         if (!indr_block_cb)
 747                 return;
 748
 749         /* Send unbind message if required to free any block cbs. */
 750         tc_indr_block_ing_cmd(indr_dev, indr_block_cb, TC_BLOCK_UNBIND);
 751         tc_indr_block_cb_del(indr_block_cb);
 752         tc_indr_block_dev_put(indr_dev);
 753 }
 754 EXPORT_SYMBOL_GPL(__tc_indr_block_cb_unregister);
 755
 756 void tc_indr_block_cb_unregister(struct net_device *dev,
 757                                  tc_indr_block_bind_cb_t *cb, void *cb_ident)
 758 {
 759         rtnl_lock();
 760         __tc_indr_block_cb_unregister(dev, cb, cb_ident);
 761         rtnl_unlock();
 762 }
 763 EXPORT_SYMBOL_GPL(tc_indr_block_cb_unregister);
 764
 765 static void tc_indr_block_call(struct tcf_block *block, struct net_device *dev,
 766                                struct tcf_block_ext_info *ei,
 767                                enum tc_block_command command,
 768                                struct netlink_ext_ack *extack)
 769 {
 770         struct tc_indr_block_cb *indr_block_cb;
 771         struct tc_indr_block_dev *indr_dev;
 772         struct tc_block_offload bo = {
 773                 .command        = command,
 774                 .binder_type    = ei->binder_type,
 775                 .block          = block,
 776                 .extack         = extack,
 777         };
 778
 779         indr_dev = tc_indr_block_dev_lookup(dev);
 780         if (!indr_dev)
 781                 return;
 782
 783         indr_dev->block = command == TC_BLOCK_BIND ? block : NULL;
 784
 785         list_for_each_entry(indr_block_cb, &indr_dev->cb_list, list)
 786                 indr_block_cb->cb(dev, indr_block_cb->cb_priv, TC_SETUP_BLOCK,
 787                                   &bo);
 788 }
 789
 790 static bool tcf_block_offload_in_use(struct tcf_block *block)
 791 {
 792         return block->offloadcnt;
 793 }
 794
 795 static int tcf_block_offload_cmd(struct tcf_block *block,
 796                                  struct net_device *dev,
 797                                  struct tcf_block_ext_info *ei,
 798                                  enum tc_block_command command,
 799                                  struct netlink_ext_ack *extack)
 800 {
 801         struct tc_block_offload bo = {};
 802
 803         bo.command = command;
 804         bo.binder_type = ei->binder_type;
 805         bo.block = block;
 806         bo.extack = extack;
 807         return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
 808 }
 809
 810 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
 811                                   struct tcf_block_ext_info *ei,
 812                                   struct netlink_ext_ack *extack)
 813 {
 814         struct net_device *dev = q->dev_queue->dev;
 815         int err;
 816
 817         if (!dev->netdev_ops->ndo_setup_tc)
 818                 goto no_offload_dev_inc;
 819
 820         /* If tc offload feature is disabled and the block we try to bind
 821          * to already has some offloaded filters, forbid to bind.
 822          */
 823         if (!tc_can_offload(dev) && tcf_block_offload_in_use(block)) {
 824                 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
 825                 return -EOPNOTSUPP;
 826         }
 827
 828         err = tcf_block_offload_cmd(block, dev, ei, TC_BLOCK_BIND, extack);
 829         if (err == -EOPNOTSUPP)
 830                 goto no_offload_dev_inc;
 831         if (err)
 832                 return err;
 833
 834         tc_indr_block_call(block, dev, ei, TC_BLOCK_BIND, extack);
 835         return 0;
 836
 837 no_offload_dev_inc:
 838         if (tcf_block_offload_in_use(block))
 839                 return -EOPNOTSUPP;
 840         block->nooffloaddevcnt++;
 841         tc_indr_block_call(block, dev, ei, TC_BLOCK_BIND, extack);
 842         return 0;
 843 }
 844
 845 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
 846                                      struct tcf_block_ext_info *ei)
 847 {
 848         struct net_device *dev = q->dev_queue->dev;
 849         int err;
 850
 851         tc_indr_block_call(block, dev, ei, TC_BLOCK_UNBIND, NULL);
 852
 853         if (!dev->netdev_ops->ndo_setup_tc)
 854                 goto no_offload_dev_dec;
 855         err = tcf_block_offload_cmd(block, dev, ei, TC_BLOCK_UNBIND, NULL);
 856         if (err == -EOPNOTSUPP)
 857                 goto no_offload_dev_dec;
 858         return;
 859
 860 no_offload_dev_dec:
 861         WARN_ON(block->nooffloaddevcnt-- == 0);
 862 }
 863
 864 static int
 865 tcf_chain0_head_change_cb_add(struct tcf_block *block,
 866                               struct tcf_block_ext_info *ei,
 867                               struct netlink_ext_ack *extack)
 868 {
 869         struct tcf_filter_chain_list_item *item;
 870         struct tcf_chain *chain0;
 871
 872         item = kmalloc(sizeof(*item), GFP_KERNEL);
 873         if (!item) {
 874                 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
 875                 return -ENOMEM;
 876         }
 877         item->chain_head_change = ei->chain_head_change;
 878         item->chain_head_change_priv = ei->chain_head_change_priv;
 879
 880         mutex_lock(&block->lock);
 881         chain0 = block->chain0.chain;
 882         if (chain0)
 883                 tcf_chain_hold(chain0);
 884         else
 885                 list_add(&item->list, &block->chain0.filter_chain_list);
 886         mutex_unlock(&block->lock);
 887
 888         if (chain0) {
 889                 struct tcf_proto *tp_head;
 890
 891                 mutex_lock(&chain0->filter_chain_lock);
 892
 893                 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
 894                 if (tp_head)
 895                         tcf_chain_head_change_item(item, tp_head);
 896
 897                 mutex_lock(&block->lock);
 898                 list_add(&item->list, &block->chain0.filter_chain_list);
 899                 mutex_unlock(&block->lock);
 900
 901                 mutex_unlock(&chain0->filter_chain_lock);
 902                 tcf_chain_put(chain0);
 903         }
 904
 905         return 0;
 906 }
 907
 908 static void
 909 tcf_chain0_head_change_cb_del(struct tcf_block *block,
 910                               struct tcf_block_ext_info *ei)
 911 {
 912         struct tcf_filter_chain_list_item *item;
 913
 914         mutex_lock(&block->lock);
 915         list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
 916                 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
 917                     (item->chain_head_change == ei->chain_head_change &&
 918                      item->chain_head_change_priv == ei->chain_head_change_priv)) {
 919                         if (block->chain0.chain)
 920                                 tcf_chain_head_change_item(item, NULL);
 921                         list_del(&item->list);
 922                         mutex_unlock(&block->lock);
 923
 924                         kfree(item);
 925                         return;
 926                 }
 927         }
 928         mutex_unlock(&block->lock);
 929         WARN_ON(1);
 930 }
 931
 932 struct tcf_net {
 933         spinlock_t idr_lock; /* Protects idr */
 934         struct idr idr;
 935 };
 936
 937 static unsigned int tcf_net_id;
 938
 939 static int tcf_block_insert(struct tcf_block *block, struct net *net,
 940                             struct netlink_ext_ack *extack)
 941 {
 942         struct tcf_net *tn = net_generic(net, tcf_net_id);
 943         int err;
 944
 945         idr_preload(GFP_KERNEL);
 946         spin_lock(&tn->idr_lock);
 947         err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
 948                             GFP_NOWAIT);
 949         spin_unlock(&tn->idr_lock);
 950         idr_preload_end();
 951
 952         return err;
 953 }
 954
 955 static void tcf_block_remove(struct tcf_block *block, struct net *net)
 956 {
 957         struct tcf_net *tn = net_generic(net, tcf_net_id);
 958
 959         spin_lock(&tn->idr_lock);
 960         idr_remove(&tn->idr, block->index);
 961         spin_unlock(&tn->idr_lock);
 962 }
 963
 964 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
 965                                           u32 block_index,
 966                                           struct netlink_ext_ack *extack)
 967 {
 968         struct tcf_block *block;
 969
 970         block = kzalloc(sizeof(*block), GFP_KERNEL);
 971         if (!block) {
 972                 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
 973                 return ERR_PTR(-ENOMEM);
 974         }
 975         mutex_init(&block->lock);
 976         INIT_LIST_HEAD(&block->chain_list);
 977         INIT_LIST_HEAD(&block->cb_list);
 978         INIT_LIST_HEAD(&block->owner_list);
 979         INIT_LIST_HEAD(&block->chain0.filter_chain_list);
 980
 981         refcount_set(&block->refcnt, 1);
 982         block->net = net;
 983         block->index = block_index;
 984
 985         /* Don't store q pointer for blocks which are shared */
 986         if (!tcf_block_shared(block))
 987                 block->q = q;
 988         return block;
 989 }
 990
 991 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
 992 {
 993         struct tcf_net *tn = net_generic(net, tcf_net_id);
 994
 995         return idr_find(&tn->idr, block_index);
 996 }
 997
 998 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
 999 {
1000         struct tcf_block *block;
1001
1002         rcu_read_lock();
1003         block = tcf_block_lookup(net, block_index);
1004         if (block && !refcount_inc_not_zero(&block->refcnt))
1005                 block = NULL;
1006         rcu_read_unlock();
1007
1008         return block;
1009 }
1010
1011 static struct tcf_chain *
1012 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1013 {
1014         mutex_lock(&block->lock);
1015         if (chain)
1016                 chain = list_is_last(&chain->list, &block->chain_list) ?
1017                         NULL : list_next_entry(chain, list);
1018         else
1019                 chain = list_first_entry_or_null(&block->chain_list,
1020                                                  struct tcf_chain, list);
1021
1022         /* skip all action-only chains */
1023         while (chain && tcf_chain_held_by_acts_only(chain))
1024                 chain = list_is_last(&chain->list, &block->chain_list) ?
1025                         NULL : list_next_entry(chain, list);
1026
1027         if (chain)
1028                 tcf_chain_hold(chain);
1029         mutex_unlock(&block->lock);
1030
1031         return chain;
1032 }
1033
1034 /* Function to be used by all clients that want to iterate over all chains on
1035  * block. It properly obtains block->lock and takes reference to chain before
1036  * returning it. Users of this function must be tolerant to concurrent chain
1037  * insertion/deletion or ensure that no concurrent chain modification is
1038  * possible. Note that all netlink dump callbacks cannot guarantee to provide
1039  * consistent dump because rtnl lock is released each time skb is filled with
1040  * data and sent to user-space.
1041  */
1042
1043 struct tcf_chain *
1044 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1045 {
1046         struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
1047
1048         if (chain)
1049                 tcf_chain_put(chain);
1050
1051         return chain_next;
1052 }
1053 EXPORT_SYMBOL(tcf_get_next_chain);
1054
1055 static struct tcf_proto *
1056 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1057 {
1058         u32 prio = 0;
1059
1060         ASSERT_RTNL();
1061         mutex_lock(&chain->filter_chain_lock);
1062
1063         if (!tp) {
1064                 tp = tcf_chain_dereference(chain->filter_chain, chain);
1065         } else if (tcf_proto_is_deleting(tp)) {
1066                 /* 'deleting' flag is set and chain->filter_chain_lock was
1067                  * unlocked, which means next pointer could be invalid. Restart
1068                  * search.
1069                  */
1070                 prio = tp->prio + 1;
1071                 tp = tcf_chain_dereference(chain->filter_chain, chain);
1072
1073                 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1074                         if (!tp->deleting && tp->prio >= prio)
1075                                 break;
1076         } else {
1077                 tp = tcf_chain_dereference(tp->next, chain);
1078         }
1079
1080         if (tp)
1081                 tcf_proto_get(tp);
1082
1083         mutex_unlock(&chain->filter_chain_lock);
1084
1085         return tp;
1086 }
1087
1088 /* Function to be used by all clients that want to iterate over all tp's on
1089  * chain. Users of this function must be tolerant to concurrent tp
1090  * insertion/deletion or ensure that no concurrent chain modification is
1091  * possible. Note that all netlink dump callbacks cannot guarantee to provide
1092  * consistent dump because rtnl lock is released each time skb is filled with
1093  * data and sent to user-space.
1094  */
1095
1096 struct tcf_proto *
1097 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp,
1098                    bool rtnl_held)
1099 {
1100         struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1101
1102         if (tp)
1103                 tcf_proto_put(tp, rtnl_held, NULL);
1104
1105         return tp_next;
1106 }
1107 EXPORT_SYMBOL(tcf_get_next_proto);
1108
1109 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1110 {
1111         struct tcf_chain *chain;
1112
1113         /* Last reference to block. At this point chains cannot be added or
1114          * removed concurrently.
1115          */
1116         for (chain = tcf_get_next_chain(block, NULL);
1117              chain;
1118              chain = tcf_get_next_chain(block, chain)) {
1119                 tcf_chain_put_explicitly_created(chain);
1120                 tcf_chain_flush(chain, rtnl_held);
1121         }
1122 }
1123
1124 /* Lookup Qdisc and increments its reference counter.
1125  * Set parent, if necessary.
1126  */
1127
1128 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1129                             u32 *parent, int ifindex, bool rtnl_held,
1130                             struct netlink_ext_ack *extack)
1131 {
1132         const struct Qdisc_class_ops *cops;
1133         struct net_device *dev;
1134         int err = 0;
1135
1136         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1137                 return 0;
1138
1139         rcu_read_lock();
1140
1141         /* Find link */
1142         dev = dev_get_by_index_rcu(net, ifindex);
1143         if (!dev) {
1144                 rcu_read_unlock();
1145                 return -ENODEV;
1146         }
1147
1148         /* Find qdisc */
1149         if (!*parent) {
1150                 *q = dev->qdisc;
1151                 *parent = (*q)->handle;
1152         } else {
1153                 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1154                 if (!*q) {
1155                         NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1156                         err = -EINVAL;
1157                         goto errout_rcu;
1158                 }
1159         }
1160
1161         *q = qdisc_refcount_inc_nz(*q);
1162         if (!*q) {
1163                 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1164                 err = -EINVAL;
1165                 goto errout_rcu;
1166         }
1167
1168         /* Is it classful? */
1169         cops = (*q)->ops->cl_ops;
1170         if (!cops) {
1171                 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1172                 err = -EINVAL;
1173                 goto errout_qdisc;
1174         }
1175
1176         if (!cops->tcf_block) {
1177                 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1178                 err = -EOPNOTSUPP;
1179                 goto errout_qdisc;
1180         }
1181
1182 errout_rcu:
1183         /* At this point we know that qdisc is not noop_qdisc,
1184          * which means that qdisc holds a reference to net_device
1185          * and we hold a reference to qdisc, so it is safe to release
1186          * rcu read lock.
1187          */
1188         rcu_read_unlock();
1189         return err;
1190
1191 errout_qdisc:
1192         rcu_read_unlock();
1193
1194         if (rtnl_held)
1195                 qdisc_put(*q);
1196         else
1197                 qdisc_put_unlocked(*q);
1198         *q = NULL;
1199
1200         return err;
1201 }
1202
1203 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1204                                int ifindex, struct netlink_ext_ack *extack)
1205 {
1206         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1207                 return 0;
1208
1209         /* Do we search for filter, attached to class? */
1210         if (TC_H_MIN(parent)) {
1211                 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1212
1213                 *cl = cops->find(q, parent);
1214                 if (*cl == 0) {
1215                         NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1216                         return -ENOENT;
1217                 }
1218         }
1219
1220         return 0;
1221 }
1222
1223 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1224                                           unsigned long cl, int ifindex,
1225                                           u32 block_index,
1226                                           struct netlink_ext_ack *extack)
1227 {
1228         struct tcf_block *block;
1229
1230         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1231                 block = tcf_block_refcnt_get(net, block_index);
1232                 if (!block) {
1233                         NL_SET_ERR_MSG(extack, "Block of given index was not found");
1234                         return ERR_PTR(-EINVAL);
1235                 }
1236         } else {
1237                 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1238
1239                 block = cops->tcf_block(q, cl, extack);
1240                 if (!block)
1241                         return ERR_PTR(-EINVAL);
1242
1243                 if (tcf_block_shared(block)) {
1244                         NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1245                         return ERR_PTR(-EOPNOTSUPP);
1246                 }
1247
1248                 /* Always take reference to block in order to support execution
1249                  * of rules update path of cls API without rtnl lock. Caller
1250                  * must release block when it is finished using it. 'if' block
1251                  * of this conditional obtain reference to block by calling
1252                  * tcf_block_refcnt_get().
1253                  */
1254                 refcount_inc(&block->refcnt);
1255         }
1256
1257         return block;
1258 }
1259
1260 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1261                             struct tcf_block_ext_info *ei, bool rtnl_held)
1262 {
1263         if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1264                 /* Flushing/putting all chains will cause the block to be
1265                  * deallocated when last chain is freed. However, if chain_list
1266                  * is empty, block has to be manually deallocated. After block
1267                  * reference counter reached 0, it is no longer possible to
1268                  * increment it or add new chains to block.
1269                  */
1270                 bool free_block = list_empty(&block->chain_list);
1271
1272                 mutex_unlock(&block->lock);
1273                 if (tcf_block_shared(block))
1274                         tcf_block_remove(block, block->net);
1275
1276                 if (q)
1277                         tcf_block_offload_unbind(block, q, ei);
1278
1279                 if (free_block)
1280                         tcf_block_destroy(block);
1281                 else
1282                         tcf_block_flush_all_chains(block, rtnl_held);
1283         } else if (q) {
1284                 tcf_block_offload_unbind(block, q, ei);
1285         }
1286 }
1287
1288 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1289 {
1290         __tcf_block_put(block, NULL, NULL, rtnl_held);
1291 }
1292
1293 /* Find tcf block.
1294  * Set q, parent, cl when appropriate.
1295  */
1296
1297 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1298                                         u32 *parent, unsigned long *cl,
1299                                         int ifindex, u32 block_index,
1300                                         struct netlink_ext_ack *extack)
1301 {
1302         struct tcf_block *block;
1303         int err = 0;
1304
1305         ASSERT_RTNL();
1306
1307         err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1308         if (err)
1309                 goto errout;
1310
1311         err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1312         if (err)
1313                 goto errout_qdisc;
1314
1315         block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1316         if (IS_ERR(block)) {
1317                 err = PTR_ERR(block);
1318                 goto errout_qdisc;
1319         }
1320
1321         return block;
1322
1323 errout_qdisc:
1324         if (*q)
1325                 qdisc_put(*q);
1326 errout:
1327         *q = NULL;
1328         return ERR_PTR(err);
1329 }
1330
1331 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1332                               bool rtnl_held)
1333 {
1334         if (!IS_ERR_OR_NULL(block))
1335                 tcf_block_refcnt_put(block, rtnl_held);
1336
1337         if (q) {
1338                 if (rtnl_held)
1339                         qdisc_put(q);
1340                 else
1341                         qdisc_put_unlocked(q);
1342         }
1343 }
1344
1345 struct tcf_block_owner_item {
1346         struct list_head list;
1347         struct Qdisc *q;
1348         enum tcf_block_binder_type binder_type;
1349 };
1350
1351 static void
1352 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1353                                struct Qdisc *q,
1354                                enum tcf_block_binder_type binder_type)
1355 {
1356         if (block->keep_dst &&
1357             binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1358             binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1359                 netif_keep_dst(qdisc_dev(q));
1360 }
1361
1362 void tcf_block_netif_keep_dst(struct tcf_block *block)
1363 {
1364         struct tcf_block_owner_item *item;
1365
1366         block->keep_dst = true;
1367         list_for_each_entry(item, &block->owner_list, list)
1368                 tcf_block_owner_netif_keep_dst(block, item->q,
1369                                                item->binder_type);
1370 }
1371 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1372
1373 static int tcf_block_owner_add(struct tcf_block *block,
1374                                struct Qdisc *q,
1375                                enum tcf_block_binder_type binder_type)
1376 {
1377         struct tcf_block_owner_item *item;
1378
1379         item = kmalloc(sizeof(*item), GFP_KERNEL);
1380         if (!item)
1381                 return -ENOMEM;
1382         item->q = q;
1383         item->binder_type = binder_type;
1384         list_add(&item->list, &block->owner_list);
1385         return 0;
1386 }
1387
1388 static void tcf_block_owner_del(struct tcf_block *block,
1389                                 struct Qdisc *q,
1390                                 enum tcf_block_binder_type binder_type)
1391 {
1392         struct tcf_block_owner_item *item;
1393
1394         list_for_each_entry(item, &block->owner_list, list) {
1395                 if (item->q == q && item->binder_type == binder_type) {
1396                         list_del(&item->list);
1397                         kfree(item);
1398                         return;
1399                 }
1400         }
1401         WARN_ON(1);
1402 }
1403
1404 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1405                       struct tcf_block_ext_info *ei,
1406                       struct netlink_ext_ack *extack)
1407 {
1408         struct net *net = qdisc_net(q);
1409         struct tcf_block *block = NULL;
1410         int err;
1411
1412         if (ei->block_index)
1413                 /* block_index not 0 means the shared block is requested */
1414                 block = tcf_block_refcnt_get(net, ei->block_index);
1415
1416         if (!block) {
1417                 block = tcf_block_create(net, q, ei->block_index, extack);
1418                 if (IS_ERR(block))
1419                         return PTR_ERR(block);
1420                 if (tcf_block_shared(block)) {
1421                         err = tcf_block_insert(block, net, extack);
1422                         if (err)
1423                                 goto err_block_insert;
1424                 }
1425         }
1426
1427         err = tcf_block_owner_add(block, q, ei->binder_type);
1428         if (err)
1429                 goto err_block_owner_add;
1430
1431         tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1432
1433         err = tcf_chain0_head_change_cb_add(block, ei, extack);
1434         if (err)
1435                 goto err_chain0_head_change_cb_add;
1436
1437         err = tcf_block_offload_bind(block, q, ei, extack);
1438         if (err)
1439                 goto err_block_offload_bind;
1440
1441         *p_block = block;
1442         return 0;
1443
1444 err_block_offload_bind:
1445         tcf_chain0_head_change_cb_del(block, ei);
1446 err_chain0_head_change_cb_add:
1447         tcf_block_owner_del(block, q, ei->binder_type);
1448 err_block_owner_add:
1449 err_block_insert:
1450         tcf_block_refcnt_put(block, true);
1451         return err;
1452 }
1453 EXPORT_SYMBOL(tcf_block_get_ext);
1454
1455 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1456 {
1457         struct tcf_proto __rcu **p_filter_chain = priv;
1458
1459         rcu_assign_pointer(*p_filter_chain, tp_head);
1460 }
1461
1462 int tcf_block_get(struct tcf_block **p_block,
1463                   struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1464                   struct netlink_ext_ack *extack)
1465 {
1466         struct tcf_block_ext_info ei = {
1467                 .chain_head_change = tcf_chain_head_change_dflt,
1468                 .chain_head_change_priv = p_filter_chain,
1469         };
1470
1471         WARN_ON(!p_filter_chain);
1472         return tcf_block_get_ext(p_block, q, &ei, extack);
1473 }
1474 EXPORT_SYMBOL(tcf_block_get);
1475
1476 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1477  * actions should be all removed after flushing.
1478  */
1479 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1480                        struct tcf_block_ext_info *ei)
1481 {
1482         if (!block)
1483                 return;
1484         tcf_chain0_head_change_cb_del(block, ei);
1485         tcf_block_owner_del(block, q, ei->binder_type);
1486
1487         __tcf_block_put(block, q, ei, true);
1488 }
1489 EXPORT_SYMBOL(tcf_block_put_ext);
1490
1491 void tcf_block_put(struct tcf_block *block)
1492 {
1493         struct tcf_block_ext_info ei = {0, };
1494
1495         if (!block)
1496                 return;
1497         tcf_block_put_ext(block, block->q, &ei);
1498 }
1499
1500 EXPORT_SYMBOL(tcf_block_put);
1501
1502 struct tcf_block_cb {
1503         struct list_head list;
1504         tc_setup_cb_t *cb;
1505         void *cb_ident;
1506         void *cb_priv;
1507         unsigned int refcnt;
1508 };
1509
1510 void *tcf_block_cb_priv(struct tcf_block_cb *block_cb)
1511 {
1512         return block_cb->cb_priv;
1513 }
1514 EXPORT_SYMBOL(tcf_block_cb_priv);
1515
1516 struct tcf_block_cb *tcf_block_cb_lookup(struct tcf_block *block,
1517                                          tc_setup_cb_t *cb, void *cb_ident)
1518 {       struct tcf_block_cb *block_cb;
1519
1520         list_for_each_entry(block_cb, &block->cb_list, list)
1521                 if (block_cb->cb == cb && block_cb->cb_ident == cb_ident)
1522                         return block_cb;
1523         return NULL;
1524 }
1525 EXPORT_SYMBOL(tcf_block_cb_lookup);
1526
1527 void tcf_block_cb_incref(struct tcf_block_cb *block_cb)
1528 {
1529         block_cb->refcnt++;
1530 }
1531 EXPORT_SYMBOL(tcf_block_cb_incref);
1532
1533 unsigned int tcf_block_cb_decref(struct tcf_block_cb *block_cb)
1534 {
1535         return --block_cb->refcnt;
1536 }
1537 EXPORT_SYMBOL(tcf_block_cb_decref);
1538
1539 static int
1540 tcf_block_playback_offloads(struct tcf_block *block, tc_setup_cb_t *cb,
1541                             void *cb_priv, bool add, bool offload_in_use,
1542                             struct netlink_ext_ack *extack)
1543 {
1544         struct tcf_chain *chain, *chain_prev;
1545         struct tcf_proto *tp, *tp_prev;
1546         int err;
1547
1548         for (chain = __tcf_get_next_chain(block, NULL);
1549              chain;
1550              chain_prev = chain,
1551                      chain = __tcf_get_next_chain(block, chain),
1552                      tcf_chain_put(chain_prev)) {
1553                 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1554                      tp_prev = tp,
1555                              tp = __tcf_get_next_proto(chain, tp),
1556                              tcf_proto_put(tp_prev, true, NULL)) {
1557                         if (tp->ops->reoffload) {
1558                                 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1559                                                          extack);
1560                                 if (err && add)
1561                                         goto err_playback_remove;
1562                         } else if (add && offload_in_use) {
1563                                 err = -EOPNOTSUPP;
1564                                 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1565                                 goto err_playback_remove;
1566                         }
1567                 }
1568         }
1569
1570         return 0;
1571
1572 err_playback_remove:
1573         tcf_proto_put(tp, true, NULL);
1574         tcf_chain_put(chain);
1575         tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1576                                     extack);
1577         return err;
1578 }
1579
1580 struct tcf_block_cb *__tcf_block_cb_register(struct tcf_block *block,
1581                                              tc_setup_cb_t *cb, void *cb_ident,
1582                                              void *cb_priv,
1583                                              struct netlink_ext_ack *extack)
1584 {
1585         struct tcf_block_cb *block_cb;
1586         int err;
1587
1588         /* Replay any already present rules */
1589         err = tcf_block_playback_offloads(block, cb, cb_priv, true,
1590                                           tcf_block_offload_in_use(block),
1591                                           extack);
1592         if (err)
1593                 return ERR_PTR(err);
1594
1595         block_cb = kzalloc(sizeof(*block_cb), GFP_KERNEL);
1596         if (!block_cb)
1597                 return ERR_PTR(-ENOMEM);
1598         block_cb->cb = cb;
1599         block_cb->cb_ident = cb_ident;
1600         block_cb->cb_priv = cb_priv;
1601         list_add(&block_cb->list, &block->cb_list);
1602         return block_cb;
1603 }
1604 EXPORT_SYMBOL(__tcf_block_cb_register);
1605
1606 int tcf_block_cb_register(struct tcf_block *block,
1607                           tc_setup_cb_t *cb, void *cb_ident,
1608                           void *cb_priv, struct netlink_ext_ack *extack)
1609 {
1610         struct tcf_block_cb *block_cb;
1611
1612         block_cb = __tcf_block_cb_register(block, cb, cb_ident, cb_priv,
1613                                            extack);
1614         return PTR_ERR_OR_ZERO(block_cb);
1615 }
1616 EXPORT_SYMBOL(tcf_block_cb_register);
1617
1618 void __tcf_block_cb_unregister(struct tcf_block *block,
1619                                struct tcf_block_cb *block_cb)
1620 {
1621         tcf_block_playback_offloads(block, block_cb->cb, block_cb->cb_priv,
1622                                     false, tcf_block_offload_in_use(block),
1623                                     NULL);
1624         list_del(&block_cb->list);
1625         kfree(block_cb);
1626 }
1627 EXPORT_SYMBOL(__tcf_block_cb_unregister);
1628
1629 void tcf_block_cb_unregister(struct tcf_block *block,
1630                              tc_setup_cb_t *cb, void *cb_ident)
1631 {
1632         struct tcf_block_cb *block_cb;
1633
1634         block_cb = tcf_block_cb_lookup(block, cb, cb_ident);
1635         if (!block_cb)
1636                 return;
1637         __tcf_block_cb_unregister(block, block_cb);
1638 }
1639 EXPORT_SYMBOL(tcf_block_cb_unregister);
1640
1641 /* Main classifier routine: scans classifier chain attached
1642  * to this qdisc, (optionally) tests for protocol and asks
1643  * specific classifiers.
1644  */
1645 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1646                  struct tcf_result *res, bool compat_mode)
1647 {
1648 #ifdef CONFIG_NET_CLS_ACT
1649         const int max_reclassify_loop = 4;
1650         const struct tcf_proto *orig_tp = tp;
1651         const struct tcf_proto *first_tp;
1652         int limit = 0;
1653
1654 reclassify:
1655 #endif
1656         for (; tp; tp = rcu_dereference_bh(tp->next)) {
1657                 __be16 protocol = tc_skb_protocol(skb);
1658                 int err;
1659
1660                 if (tp->protocol != protocol &&
1661                     tp->protocol != htons(ETH_P_ALL))
1662                         continue;
1663
1664                 err = tp->classify(skb, tp, res);
1665 #ifdef CONFIG_NET_CLS_ACT
1666                 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1667                         first_tp = orig_tp;
1668                         goto reset;
1669                 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1670                         first_tp = res->goto_tp;
1671                         goto reset;
1672                 }
1673 #endif
1674                 if (err >= 0)
1675                         return err;
1676         }
1677
1678         return TC_ACT_UNSPEC; /* signal: continue lookup */
1679 #ifdef CONFIG_NET_CLS_ACT
1680 reset:
1681         if (unlikely(limit++ >= max_reclassify_loop)) {
1682                 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1683                                        tp->chain->block->index,
1684                                        tp->prio & 0xffff,
1685                                        ntohs(tp->protocol));
1686                 return TC_ACT_SHOT;
1687         }
1688
1689         tp = first_tp;
1690         goto reclassify;
1691 #endif
1692 }
1693 EXPORT_SYMBOL(tcf_classify);
1694
1695 struct tcf_chain_info {
1696         struct tcf_proto __rcu **pprev;
1697         struct tcf_proto __rcu *next;
1698 };
1699
1700 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1701                                            struct tcf_chain_info *chain_info)
1702 {
1703         return tcf_chain_dereference(*chain_info->pprev, chain);
1704 }
1705
1706 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1707                                struct tcf_chain_info *chain_info,
1708                                struct tcf_proto *tp)
1709 {
1710         if (chain->flushing)
1711                 return -EAGAIN;
1712
1713         if (*chain_info->pprev == chain->filter_chain)
1714                 tcf_chain0_head_change(chain, tp);
1715         tcf_proto_get(tp);
1716         RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1717         rcu_assign_pointer(*chain_info->pprev, tp);
1718
1719         return 0;
1720 }
1721
1722 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1723                                 struct tcf_chain_info *chain_info,
1724                                 struct tcf_proto *tp)
1725 {
1726         struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1727
1728         tcf_proto_mark_delete(tp);
1729         if (tp == chain->filter_chain)
1730                 tcf_chain0_head_change(chain, next);
1731         RCU_INIT_POINTER(*chain_info->pprev, next);
1732 }
1733
1734 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1735                                            struct tcf_chain_info *chain_info,
1736                                            u32 protocol, u32 prio,
1737                                            bool prio_allocate);
1738
1739 /* Try to insert new proto.
1740  * If proto with specified priority already exists, free new proto
1741  * and return existing one.
1742  */
1743
1744 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1745                                                     struct tcf_proto *tp_new,
1746                                                     u32 protocol, u32 prio,
1747                                                     bool rtnl_held)
1748 {
1749         struct tcf_chain_info chain_info;
1750         struct tcf_proto *tp;
1751         int err = 0;
1752
1753         mutex_lock(&chain->filter_chain_lock);
1754
1755         tp = tcf_chain_tp_find(chain, &chain_info,
1756                                protocol, prio, false);
1757         if (!tp)
1758                 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1759         mutex_unlock(&chain->filter_chain_lock);
1760
1761         if (tp) {
1762                 tcf_proto_destroy(tp_new, rtnl_held, NULL);
1763                 tp_new = tp;
1764         } else if (err) {
1765                 tcf_proto_destroy(tp_new, rtnl_held, NULL);
1766                 tp_new = ERR_PTR(err);
1767         }
1768
1769         return tp_new;
1770 }
1771
1772 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1773                                       struct tcf_proto *tp, bool rtnl_held,
1774                                       struct netlink_ext_ack *extack)
1775 {
1776         struct tcf_chain_info chain_info;
1777         struct tcf_proto *tp_iter;
1778         struct tcf_proto **pprev;
1779         struct tcf_proto *next;
1780
1781         mutex_lock(&chain->filter_chain_lock);
1782
1783         /* Atomically find and remove tp from chain. */
1784         for (pprev = &chain->filter_chain;
1785              (tp_iter = tcf_chain_dereference(*pprev, chain));
1786              pprev = &tp_iter->next) {
1787                 if (tp_iter == tp) {
1788                         chain_info.pprev = pprev;
1789                         chain_info.next = tp_iter->next;
1790                         WARN_ON(tp_iter->deleting);
1791                         break;
1792                 }
1793         }
1794         /* Verify that tp still exists and no new filters were inserted
1795          * concurrently.
1796          * Mark tp for deletion if it is empty.
1797          */
1798         if (!tp_iter || !tcf_proto_check_delete(tp, rtnl_held)) {
1799                 mutex_unlock(&chain->filter_chain_lock);
1800                 return;
1801         }
1802
1803         next = tcf_chain_dereference(chain_info.next, chain);
1804         if (tp == chain->filter_chain)
1805                 tcf_chain0_head_change(chain, next);
1806         RCU_INIT_POINTER(*chain_info.pprev, next);
1807         mutex_unlock(&chain->filter_chain_lock);
1808
1809         tcf_proto_put(tp, rtnl_held, extack);
1810 }
1811
1812 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1813                                            struct tcf_chain_info *chain_info,
1814                                            u32 protocol, u32 prio,
1815                                            bool prio_allocate)
1816 {
1817         struct tcf_proto **pprev;
1818         struct tcf_proto *tp;
1819
1820         /* Check the chain for existence of proto-tcf with this priority */
1821         for (pprev = &chain->filter_chain;
1822              (tp = tcf_chain_dereference(*pprev, chain));
1823              pprev = &tp->next) {
1824                 if (tp->prio >= prio) {
1825                         if (tp->prio == prio) {
1826                                 if (prio_allocate ||
1827                                     (tp->protocol != protocol && protocol))
1828                                         return ERR_PTR(-EINVAL);
1829                         } else {
1830                                 tp = NULL;
1831                         }
1832                         break;
1833                 }
1834         }
1835         chain_info->pprev = pprev;
1836         if (tp) {
1837                 chain_info->next = tp->next;
1838                 tcf_proto_get(tp);
1839         } else {
1840                 chain_info->next = NULL;
1841         }
1842         return tp;
1843 }
1844
1845 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1846                          struct tcf_proto *tp, struct tcf_block *block,
1847                          struct Qdisc *q, u32 parent, void *fh,
1848                          u32 portid, u32 seq, u16 flags, int event,
1849                          bool rtnl_held)
1850 {
1851         struct tcmsg *tcm;
1852         struct nlmsghdr  *nlh;
1853         unsigned char *b = skb_tail_pointer(skb);
1854
1855         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1856         if (!nlh)
1857                 goto out_nlmsg_trim;
1858         tcm = nlmsg_data(nlh);
1859         tcm->tcm_family = AF_UNSPEC;
1860         tcm->tcm__pad1 = 0;
1861         tcm->tcm__pad2 = 0;
1862         if (q) {
1863                 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1864                 tcm->tcm_parent = parent;
1865         } else {
1866                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1867                 tcm->tcm_block_index = block->index;
1868         }
1869         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1870         if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1871                 goto nla_put_failure;
1872         if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1873                 goto nla_put_failure;
1874         if (!fh) {
1875                 tcm->tcm_handle = 0;
1876         } else {
1877                 if (tp->ops->dump &&
1878                     tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
1879                         goto nla_put_failure;
1880         }
1881         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1882         return skb->len;
1883
1884 out_nlmsg_trim:
1885 nla_put_failure:
1886         nlmsg_trim(skb, b);
1887         return -1;
1888 }
1889
1890 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1891                           struct nlmsghdr *n, struct tcf_proto *tp,
1892                           struct tcf_block *block, struct Qdisc *q,
1893                           u32 parent, void *fh, int event, bool unicast,
1894                           bool rtnl_held)
1895 {
1896         struct sk_buff *skb;
1897         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1898         int err = 0;
1899
1900         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1901         if (!skb)
1902                 return -ENOBUFS;
1903
1904         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1905                           n->nlmsg_seq, n->nlmsg_flags, event,
1906                           rtnl_held) <= 0) {
1907                 kfree_skb(skb);
1908                 return -EINVAL;
1909         }
1910
1911         if (unicast)
1912                 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1913         else
1914                 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1915                                      n->nlmsg_flags & NLM_F_ECHO);
1916
1917         if (err > 0)
1918                 err = 0;
1919         return err;
1920 }
1921
1922 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1923                               struct nlmsghdr *n, struct tcf_proto *tp,
1924                               struct tcf_block *block, struct Qdisc *q,
1925                               u32 parent, void *fh, bool unicast, bool *last,
1926                               bool rtnl_held, struct netlink_ext_ack *extack)
1927 {
1928         struct sk_buff *skb;
1929         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1930         int err;
1931
1932         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1933         if (!skb)
1934                 return -ENOBUFS;
1935
1936         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1937                           n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
1938                           rtnl_held) <= 0) {
1939                 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1940                 kfree_skb(skb);
1941                 return -EINVAL;
1942         }
1943
1944         err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
1945         if (err) {
1946                 kfree_skb(skb);
1947                 return err;
1948         }
1949
1950         if (unicast)
1951                 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1952         else
1953                 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1954                                      n->nlmsg_flags & NLM_F_ECHO);
1955         if (err < 0)
1956                 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1957
1958         if (err > 0)
1959                 err = 0;
1960         return err;
1961 }
1962
1963 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1964                                  struct tcf_block *block, struct Qdisc *q,
1965                                  u32 parent, struct nlmsghdr *n,
1966                                  struct tcf_chain *chain, int event,
1967                                  bool rtnl_held)
1968 {
1969         struct tcf_proto *tp;
1970
1971         for (tp = tcf_get_next_proto(chain, NULL, rtnl_held);
1972              tp; tp = tcf_get_next_proto(chain, tp, rtnl_held))
1973                 tfilter_notify(net, oskb, n, tp, block,
1974                                q, parent, NULL, event, false, rtnl_held);
1975 }
1976
1977 static void tfilter_put(struct tcf_proto *tp, void *fh)
1978 {
1979         if (tp->ops->put && fh)
1980                 tp->ops->put(tp, fh);
1981 }
1982
1983 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1984                           struct netlink_ext_ack *extack)
1985 {
1986         struct net *net = sock_net(skb->sk);
1987         struct nlattr *tca[TCA_MAX + 1];
1988         struct tcmsg *t;
1989         u32 protocol;
1990         u32 prio;
1991         bool prio_allocate;
1992         u32 parent;
1993         u32 chain_index;
1994         struct Qdisc *q = NULL;
1995         struct tcf_chain_info chain_info;
1996         struct tcf_chain *chain = NULL;
1997         struct tcf_block *block;
1998         struct tcf_proto *tp;
1999         unsigned long cl;
2000         void *fh;
2001         int err;
2002         int tp_created;
2003         bool rtnl_held = false;
2004
2005         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2006                 return -EPERM;
2007
2008 replay:
2009         tp_created = 0;
2010
2011         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2012                                      rtm_tca_policy, extack);
2013         if (err < 0)
2014                 return err;
2015
2016         t = nlmsg_data(n);
2017         protocol = TC_H_MIN(t->tcm_info);
2018         prio = TC_H_MAJ(t->tcm_info);
2019         prio_allocate = false;
2020         parent = t->tcm_parent;
2021         tp = NULL;
2022         cl = 0;
2023         block = NULL;
2024
2025         if (prio == 0) {
2026                 /* If no priority is provided by the user,
2027                  * we allocate one.
2028                  */
2029                 if (n->nlmsg_flags & NLM_F_CREATE) {
2030                         prio = TC_H_MAKE(0x80000000U, 0U);
2031                         prio_allocate = true;
2032                 } else {
2033                         NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2034                         return -ENOENT;
2035                 }
2036         }
2037
2038         /* Find head of filter chain. */
2039
2040         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2041         if (err)
2042                 return err;
2043
2044         /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2045          * block is shared (no qdisc found), qdisc is not unlocked, classifier
2046          * type is not specified, classifier is not unlocked.
2047          */
2048         if (rtnl_held ||
2049             (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2050             !tca[TCA_KIND] || !tcf_proto_is_unlocked(nla_data(tca[TCA_KIND]))) {
2051                 rtnl_held = true;
2052                 rtnl_lock();
2053         }
2054
2055         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2056         if (err)
2057                 goto errout;
2058
2059         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2060                                  extack);
2061         if (IS_ERR(block)) {
2062                 err = PTR_ERR(block);
2063                 goto errout;
2064         }
2065
2066         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2067         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2068                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2069                 err = -EINVAL;
2070                 goto errout;
2071         }
2072         chain = tcf_chain_get(block, chain_index, true);
2073         if (!chain) {
2074                 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2075                 err = -ENOMEM;
2076                 goto errout;
2077         }
2078
2079         mutex_lock(&chain->filter_chain_lock);
2080         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2081                                prio, prio_allocate);
2082         if (IS_ERR(tp)) {
2083                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2084                 err = PTR_ERR(tp);
2085                 goto errout_locked;
2086         }
2087
2088         if (tp == NULL) {
2089                 struct tcf_proto *tp_new = NULL;
2090
2091                 if (chain->flushing) {
2092                         err = -EAGAIN;
2093                         goto errout_locked;
2094                 }
2095
2096                 /* Proto-tcf does not exist, create new one */
2097
2098                 if (tca[TCA_KIND] == NULL || !protocol) {
2099                         NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2100                         err = -EINVAL;
2101                         goto errout_locked;
2102                 }
2103
2104                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2105                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2106                         err = -ENOENT;
2107                         goto errout_locked;
2108                 }
2109
2110                 if (prio_allocate)
2111                         prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2112                                                                &chain_info));
2113
2114                 mutex_unlock(&chain->filter_chain_lock);
2115                 tp_new = tcf_proto_create(nla_data(tca[TCA_KIND]),
2116                                           protocol, prio, chain, rtnl_held,
2117                                           extack);
2118                 if (IS_ERR(tp_new)) {
2119                         err = PTR_ERR(tp_new);
2120                         goto errout_tp;
2121                 }
2122
2123                 tp_created = 1;
2124                 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2125                                                 rtnl_held);
2126                 if (IS_ERR(tp)) {
2127                         err = PTR_ERR(tp);
2128                         goto errout_tp;
2129                 }
2130         } else {
2131                 mutex_unlock(&chain->filter_chain_lock);
2132         }
2133
2134         if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2135                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2136                 err = -EINVAL;
2137                 goto errout;
2138         }
2139
2140         fh = tp->ops->get(tp, t->tcm_handle);
2141
2142         if (!fh) {
2143                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2144                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2145                         err = -ENOENT;
2146                         goto errout;
2147                 }
2148         } else if (n->nlmsg_flags & NLM_F_EXCL) {
2149                 tfilter_put(tp, fh);
2150                 NL_SET_ERR_MSG(extack, "Filter already exists");
2151                 err = -EEXIST;
2152                 goto errout;
2153         }
2154
2155         if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2156                 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2157                 err = -EINVAL;
2158                 goto errout;
2159         }
2160
2161         err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2162                               n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
2163                               rtnl_held, extack);
2164         if (err == 0) {
2165                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2166                                RTM_NEWTFILTER, false, rtnl_held);
2167                 tfilter_put(tp, fh);
2168         }
2169
2170 errout:
2171         if (err && tp_created)
2172                 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2173 errout_tp:
2174         if (chain) {
2175                 if (tp && !IS_ERR(tp))
2176                         tcf_proto_put(tp, rtnl_held, NULL);
2177                 if (!tp_created)
2178                         tcf_chain_put(chain);
2179         }
2180         tcf_block_release(q, block, rtnl_held);
2181
2182         if (rtnl_held)
2183                 rtnl_unlock();
2184
2185         if (err == -EAGAIN) {
2186                 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2187                  * of target chain.
2188                  */
2189                 rtnl_held = true;
2190                 /* Replay the request. */
2191                 goto replay;
2192         }
2193         return err;
2194
2195 errout_locked:
2196         mutex_unlock(&chain->filter_chain_lock);
2197         goto errout;
2198 }
2199
2200 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2201                           struct netlink_ext_ack *extack)
2202 {
2203         struct net *net = sock_net(skb->sk);
2204         struct nlattr *tca[TCA_MAX + 1];
2205         struct tcmsg *t;
2206         u32 protocol;
2207         u32 prio;
2208         u32 parent;
2209         u32 chain_index;
2210         struct Qdisc *q = NULL;
2211         struct tcf_chain_info chain_info;
2212         struct tcf_chain *chain = NULL;
2213         struct tcf_block *block = NULL;
2214         struct tcf_proto *tp = NULL;
2215         unsigned long cl = 0;
2216         void *fh = NULL;
2217         int err;
2218         bool rtnl_held = false;
2219
2220         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2221                 return -EPERM;
2222
2223         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2224                                      rtm_tca_policy, extack);
2225         if (err < 0)
2226                 return err;
2227
2228         t = nlmsg_data(n);
2229         protocol = TC_H_MIN(t->tcm_info);
2230         prio = TC_H_MAJ(t->tcm_info);
2231         parent = t->tcm_parent;
2232
2233         if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2234                 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2235                 return -ENOENT;
2236         }
2237
2238         /* Find head of filter chain. */
2239
2240         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2241         if (err)
2242                 return err;
2243
2244         /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2245          * found), qdisc is not unlocked, classifier type is not specified,
2246          * classifier is not unlocked.
2247          */
2248         if (!prio ||
2249             (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2250             !tca[TCA_KIND] || !tcf_proto_is_unlocked(nla_data(tca[TCA_KIND]))) {
2251                 rtnl_held = true;
2252                 rtnl_lock();
2253         }
2254
2255         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2256         if (err)
2257                 goto errout;
2258
2259         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2260                                  extack);
2261         if (IS_ERR(block)) {
2262                 err = PTR_ERR(block);
2263                 goto errout;
2264         }
2265
2266         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2267         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2268                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2269                 err = -EINVAL;
2270                 goto errout;
2271         }
2272         chain = tcf_chain_get(block, chain_index, false);
2273         if (!chain) {
2274                 /* User requested flush on non-existent chain. Nothing to do,
2275                  * so just return success.
2276                  */
2277                 if (prio == 0) {
2278                         err = 0;
2279                         goto errout;
2280                 }
2281                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2282                 err = -ENOENT;
2283                 goto errout;
2284         }
2285
2286         if (prio == 0) {
2287                 tfilter_notify_chain(net, skb, block, q, parent, n,
2288                                      chain, RTM_DELTFILTER, rtnl_held);
2289                 tcf_chain_flush(chain, rtnl_held);
2290                 err = 0;
2291                 goto errout;
2292         }
2293
2294         mutex_lock(&chain->filter_chain_lock);
2295         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2296                                prio, false);
2297         if (!tp || IS_ERR(tp)) {
2298                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2299                 err = tp ? PTR_ERR(tp) : -ENOENT;
2300                 goto errout_locked;
2301         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2302                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2303                 err = -EINVAL;
2304                 goto errout_locked;
2305         } else if (t->tcm_handle == 0) {
2306                 tcf_chain_tp_remove(chain, &chain_info, tp);
2307                 mutex_unlock(&chain->filter_chain_lock);
2308
2309                 tcf_proto_put(tp, rtnl_held, NULL);
2310                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2311                                RTM_DELTFILTER, false, rtnl_held);
2312                 err = 0;
2313                 goto errout;
2314         }
2315         mutex_unlock(&chain->filter_chain_lock);
2316
2317         fh = tp->ops->get(tp, t->tcm_handle);
2318
2319         if (!fh) {
2320                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2321                 err = -ENOENT;
2322         } else {
2323                 bool last;
2324
2325                 err = tfilter_del_notify(net, skb, n, tp, block,
2326                                          q, parent, fh, false, &last,
2327                                          rtnl_held, extack);
2328
2329                 if (err)
2330                         goto errout;
2331                 if (last)
2332                         tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2333         }
2334
2335 errout:
2336         if (chain) {
2337                 if (tp && !IS_ERR(tp))
2338                         tcf_proto_put(tp, rtnl_held, NULL);
2339                 tcf_chain_put(chain);
2340         }
2341         tcf_block_release(q, block, rtnl_held);
2342
2343         if (rtnl_held)
2344                 rtnl_unlock();
2345
2346         return err;
2347
2348 errout_locked:
2349         mutex_unlock(&chain->filter_chain_lock);
2350         goto errout;
2351 }
2352
2353 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2354                           struct netlink_ext_ack *extack)
2355 {
2356         struct net *net = sock_net(skb->sk);
2357         struct nlattr *tca[TCA_MAX + 1];
2358         struct tcmsg *t;
2359         u32 protocol;
2360         u32 prio;
2361         u32 parent;
2362         u32 chain_index;
2363         struct Qdisc *q = NULL;
2364         struct tcf_chain_info chain_info;
2365         struct tcf_chain *chain = NULL;
2366         struct tcf_block *block = NULL;
2367         struct tcf_proto *tp = NULL;
2368         unsigned long cl = 0;
2369         void *fh = NULL;
2370         int err;
2371         bool rtnl_held = false;
2372
2373         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2374                                      rtm_tca_policy, extack);
2375         if (err < 0)
2376                 return err;
2377
2378         t = nlmsg_data(n);
2379         protocol = TC_H_MIN(t->tcm_info);
2380         prio = TC_H_MAJ(t->tcm_info);
2381         parent = t->tcm_parent;
2382
2383         if (prio == 0) {
2384                 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2385                 return -ENOENT;
2386         }
2387
2388         /* Find head of filter chain. */
2389
2390         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2391         if (err)
2392                 return err;
2393
2394         /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2395          * unlocked, classifier type is not specified, classifier is not
2396          * unlocked.
2397          */
2398         if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2399             !tca[TCA_KIND] || !tcf_proto_is_unlocked(nla_data(tca[TCA_KIND]))) {
2400                 rtnl_held = true;
2401                 rtnl_lock();
2402         }
2403
2404         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2405         if (err)
2406                 goto errout;
2407
2408         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2409                                  extack);
2410         if (IS_ERR(block)) {
2411                 err = PTR_ERR(block);
2412                 goto errout;
2413         }
2414
2415         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2416         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2417                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2418                 err = -EINVAL;
2419                 goto errout;
2420         }
2421         chain = tcf_chain_get(block, chain_index, false);
2422         if (!chain) {
2423                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2424                 err = -EINVAL;
2425                 goto errout;
2426         }
2427
2428         mutex_lock(&chain->filter_chain_lock);
2429         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2430                                prio, false);
2431         mutex_unlock(&chain->filter_chain_lock);
2432         if (!tp || IS_ERR(tp)) {
2433                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2434                 err = tp ? PTR_ERR(tp) : -ENOENT;
2435                 goto errout;
2436         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2437                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2438                 err = -EINVAL;
2439                 goto errout;
2440         }
2441
2442         fh = tp->ops->get(tp, t->tcm_handle);
2443
2444         if (!fh) {
2445                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2446                 err = -ENOENT;
2447         } else {
2448                 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2449                                      fh, RTM_NEWTFILTER, true, rtnl_held);
2450                 if (err < 0)
2451                         NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2452         }
2453
2454         tfilter_put(tp, fh);
2455 errout:
2456         if (chain) {
2457                 if (tp && !IS_ERR(tp))
2458                         tcf_proto_put(tp, rtnl_held, NULL);
2459                 tcf_chain_put(chain);
2460         }
2461         tcf_block_release(q, block, rtnl_held);
2462
2463         if (rtnl_held)
2464                 rtnl_unlock();
2465
2466         return err;
2467 }
2468
2469 struct tcf_dump_args {
2470         struct tcf_walker w;
2471         struct sk_buff *skb;
2472         struct netlink_callback *cb;
2473         struct tcf_block *block;
2474         struct Qdisc *q;
2475         u32 parent;
2476 };
2477
2478 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2479 {
2480         struct tcf_dump_args *a = (void *)arg;
2481         struct net *net = sock_net(a->skb->sk);
2482
2483         return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2484                              n, NETLINK_CB(a->cb->skb).portid,
2485                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2486                              RTM_NEWTFILTER, true);
2487 }
2488
2489 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2490                            struct sk_buff *skb, struct netlink_callback *cb,
2491                            long index_start, long *p_index)
2492 {
2493         struct net *net = sock_net(skb->sk);
2494         struct tcf_block *block = chain->block;
2495         struct tcmsg *tcm = nlmsg_data(cb->nlh);
2496         struct tcf_proto *tp, *tp_prev;
2497         struct tcf_dump_args arg;
2498
2499         for (tp = __tcf_get_next_proto(chain, NULL);
2500              tp;
2501              tp_prev = tp,
2502                      tp = __tcf_get_next_proto(chain, tp),
2503                      tcf_proto_put(tp_prev, true, NULL),
2504                      (*p_index)++) {
2505                 if (*p_index < index_start)
2506                         continue;
2507                 if (TC_H_MAJ(tcm->tcm_info) &&
2508                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
2509                         continue;
2510                 if (TC_H_MIN(tcm->tcm_info) &&
2511                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
2512                         continue;
2513                 if (*p_index > index_start)
2514                         memset(&cb->args[1], 0,
2515                                sizeof(cb->args) - sizeof(cb->args[0]));
2516                 if (cb->args[1] == 0) {
2517                         if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2518                                           NETLINK_CB(cb->skb).portid,
2519                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
2520                                           RTM_NEWTFILTER, true) <= 0)
2521                                 goto errout;
2522                         cb->args[1] = 1;
2523                 }
2524                 if (!tp->ops->walk)
2525                         continue;
2526                 arg.w.fn = tcf_node_dump;
2527                 arg.skb = skb;
2528                 arg.cb = cb;
2529                 arg.block = block;
2530                 arg.q = q;
2531                 arg.parent = parent;
2532                 arg.w.stop = 0;
2533                 arg.w.skip = cb->args[1] - 1;
2534                 arg.w.count = 0;
2535                 arg.w.cookie = cb->args[2];
2536                 tp->ops->walk(tp, &arg.w, true);
2537                 cb->args[2] = arg.w.cookie;
2538                 cb->args[1] = arg.w.count + 1;
2539                 if (arg.w.stop)
2540                         goto errout;
2541         }
2542         return true;
2543
2544 errout:
2545         tcf_proto_put(tp, true, NULL);
2546         return false;
2547 }
2548
2549 /* called with RTNL */
2550 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2551 {
2552         struct tcf_chain *chain, *chain_prev;
2553         struct net *net = sock_net(skb->sk);
2554         struct nlattr *tca[TCA_MAX + 1];
2555         struct Qdisc *q = NULL;
2556         struct tcf_block *block;
2557         struct tcmsg *tcm = nlmsg_data(cb->nlh);
2558         long index_start;
2559         long index;
2560         u32 parent;
2561         int err;
2562
2563         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2564                 return skb->len;
2565
2566         err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2567                                      NULL, cb->extack);
2568         if (err)
2569                 return err;
2570
2571         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2572                 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2573                 if (!block)
2574                         goto out;
2575                 /* If we work with block index, q is NULL and parent value
2576                  * will never be used in the following code. The check
2577                  * in tcf_fill_node prevents it. However, compiler does not
2578                  * see that far, so set parent to zero to silence the warning
2579                  * about parent being uninitialized.
2580                  */
2581                 parent = 0;
2582         } else {
2583                 const struct Qdisc_class_ops *cops;
2584                 struct net_device *dev;
2585                 unsigned long cl = 0;
2586
2587                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2588                 if (!dev)
2589                         return skb->len;
2590
2591                 parent = tcm->tcm_parent;
2592                 if (!parent) {
2593                         q = dev->qdisc;
2594                         parent = q->handle;
2595                 } else {
2596                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2597                 }
2598                 if (!q)
2599                         goto out;
2600                 cops = q->ops->cl_ops;
2601                 if (!cops)
2602                         goto out;
2603                 if (!cops->tcf_block)
2604                         goto out;
2605                 if (TC_H_MIN(tcm->tcm_parent)) {
2606                         cl = cops->find(q, tcm->tcm_parent);
2607                         if (cl == 0)
2608                                 goto out;
2609                 }
2610                 block = cops->tcf_block(q, cl, NULL);
2611                 if (!block)
2612                         goto out;
2613                 if (tcf_block_shared(block))
2614                         q = NULL;
2615         }
2616
2617         index_start = cb->args[0];
2618         index = 0;
2619
2620         for (chain = __tcf_get_next_chain(block, NULL);
2621              chain;
2622              chain_prev = chain,
2623                      chain = __tcf_get_next_chain(block, chain),
2624                      tcf_chain_put(chain_prev)) {
2625                 if (tca[TCA_CHAIN] &&
2626                     nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2627                         continue;
2628                 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2629                                     index_start, &index)) {
2630                         tcf_chain_put(chain);
2631                         err = -EMSGSIZE;
2632                         break;
2633                 }
2634         }
2635
2636         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2637                 tcf_block_refcnt_put(block, true);
2638         cb->args[0] = index;
2639
2640 out:
2641         /* If we did no progress, the error (EMSGSIZE) is real */
2642         if (skb->len == 0 && err)
2643                 return err;
2644         return skb->len;
2645 }
2646
2647 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2648                               void *tmplt_priv, u32 chain_index,
2649                               struct net *net, struct sk_buff *skb,
2650                               struct tcf_block *block,
2651                               u32 portid, u32 seq, u16 flags, int event)
2652 {
2653         unsigned char *b = skb_tail_pointer(skb);
2654         const struct tcf_proto_ops *ops;
2655         struct nlmsghdr *nlh;
2656         struct tcmsg *tcm;
2657         void *priv;
2658
2659         ops = tmplt_ops;
2660         priv = tmplt_priv;
2661
2662         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2663         if (!nlh)
2664                 goto out_nlmsg_trim;
2665         tcm = nlmsg_data(nlh);
2666         tcm->tcm_family = AF_UNSPEC;
2667         tcm->tcm__pad1 = 0;
2668         tcm->tcm__pad2 = 0;
2669         tcm->tcm_handle = 0;
2670         if (block->q) {
2671                 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2672                 tcm->tcm_parent = block->q->handle;
2673         } else {
2674                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2675                 tcm->tcm_block_index = block->index;
2676         }
2677
2678         if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2679                 goto nla_put_failure;
2680
2681         if (ops) {
2682                 if (nla_put_string(skb, TCA_KIND, ops->kind))
2683                         goto nla_put_failure;
2684                 if (ops->tmplt_dump(skb, net, priv) < 0)
2685                         goto nla_put_failure;
2686         }
2687
2688         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2689         return skb->len;
2690
2691 out_nlmsg_trim:
2692 nla_put_failure:
2693         nlmsg_trim(skb, b);
2694         return -EMSGSIZE;
2695 }
2696
2697 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2698                            u32 seq, u16 flags, int event, bool unicast)
2699 {
2700         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2701         struct tcf_block *block = chain->block;
2702         struct net *net = block->net;
2703         struct sk_buff *skb;
2704         int err = 0;
2705
2706         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2707         if (!skb)
2708                 return -ENOBUFS;
2709
2710         if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2711                                chain->index, net, skb, block, portid,
2712                                seq, flags, event) <= 0) {
2713                 kfree_skb(skb);
2714                 return -EINVAL;
2715         }
2716
2717         if (unicast)
2718                 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2719         else
2720                 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2721                                      flags & NLM_F_ECHO);
2722
2723         if (err > 0)
2724                 err = 0;
2725         return err;
2726 }
2727
2728 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2729                                   void *tmplt_priv, u32 chain_index,
2730                                   struct tcf_block *block, struct sk_buff *oskb,
2731                                   u32 seq, u16 flags, bool unicast)
2732 {
2733         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2734         struct net *net = block->net;
2735         struct sk_buff *skb;
2736
2737         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2738         if (!skb)
2739                 return -ENOBUFS;
2740
2741         if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2742                                block, portid, seq, flags, RTM_DELCHAIN) <= 0) {
2743                 kfree_skb(skb);
2744                 return -EINVAL;
2745         }
2746
2747         if (unicast)
2748                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2749
2750         return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2751 }
2752
2753 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2754                               struct nlattr **tca,
2755                               struct netlink_ext_ack *extack)
2756 {
2757         const struct tcf_proto_ops *ops;
2758         void *tmplt_priv;
2759
2760         /* If kind is not set, user did not specify template. */
2761         if (!tca[TCA_KIND])
2762                 return 0;
2763
2764         ops = tcf_proto_lookup_ops(nla_data(tca[TCA_KIND]), true, extack);
2765         if (IS_ERR(ops))
2766                 return PTR_ERR(ops);
2767         if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2768                 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2769                 return -EOPNOTSUPP;
2770         }
2771
2772         tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2773         if (IS_ERR(tmplt_priv)) {
2774                 module_put(ops->owner);
2775                 return PTR_ERR(tmplt_priv);
2776         }
2777         chain->tmplt_ops = ops;
2778         chain->tmplt_priv = tmplt_priv;
2779         return 0;
2780 }
2781
2782 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2783                                void *tmplt_priv)
2784 {
2785         /* If template ops are set, no work to do for us. */
2786         if (!tmplt_ops)
2787                 return;
2788
2789         tmplt_ops->tmplt_destroy(tmplt_priv);
2790         module_put(tmplt_ops->owner);
2791 }
2792
2793 /* Add/delete/get a chain */
2794
2795 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2796                         struct netlink_ext_ack *extack)
2797 {
2798         struct net *net = sock_net(skb->sk);
2799         struct nlattr *tca[TCA_MAX + 1];
2800         struct tcmsg *t;
2801         u32 parent;
2802         u32 chain_index;
2803         struct Qdisc *q = NULL;
2804         struct tcf_chain *chain = NULL;
2805         struct tcf_block *block;
2806         unsigned long cl;
2807         int err;
2808
2809         if (n->nlmsg_type != RTM_GETCHAIN &&
2810             !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2811                 return -EPERM;
2812
2813 replay:
2814         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2815                                      rtm_tca_policy, extack);
2816         if (err < 0)
2817                 return err;
2818
2819         t = nlmsg_data(n);
2820         parent = t->tcm_parent;
2821         cl = 0;
2822
2823         block = tcf_block_find(net, &q, &parent, &cl,
2824                                t->tcm_ifindex, t->tcm_block_index, extack);
2825         if (IS_ERR(block))
2826                 return PTR_ERR(block);
2827
2828         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2829         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2830                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2831                 err = -EINVAL;
2832                 goto errout_block;
2833         }
2834
2835         mutex_lock(&block->lock);
2836         chain = tcf_chain_lookup(block, chain_index);
2837         if (n->nlmsg_type == RTM_NEWCHAIN) {
2838                 if (chain) {
2839                         if (tcf_chain_held_by_acts_only(chain)) {
2840                                 /* The chain exists only because there is
2841                                  * some action referencing it.
2842                                  */
2843                                 tcf_chain_hold(chain);
2844                         } else {
2845                                 NL_SET_ERR_MSG(extack, "Filter chain already exists");
2846                                 err = -EEXIST;
2847                                 goto errout_block_locked;
2848                         }
2849                 } else {
2850                         if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2851                                 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
2852                                 err = -ENOENT;
2853                                 goto errout_block_locked;
2854                         }
2855                         chain = tcf_chain_create(block, chain_index);
2856                         if (!chain) {
2857                                 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
2858                                 err = -ENOMEM;
2859                                 goto errout_block_locked;
2860                         }
2861                 }
2862         } else {
2863                 if (!chain || tcf_chain_held_by_acts_only(chain)) {
2864                         NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2865                         err = -EINVAL;
2866                         goto errout_block_locked;
2867                 }
2868                 tcf_chain_hold(chain);
2869         }
2870
2871         if (n->nlmsg_type == RTM_NEWCHAIN) {
2872                 /* Modifying chain requires holding parent block lock. In case
2873                  * the chain was successfully added, take a reference to the
2874                  * chain. This ensures that an empty chain does not disappear at
2875                  * the end of this function.
2876                  */
2877                 tcf_chain_hold(chain);
2878                 chain->explicitly_created = true;
2879         }
2880         mutex_unlock(&block->lock);
2881
2882         switch (n->nlmsg_type) {
2883         case RTM_NEWCHAIN:
2884                 err = tc_chain_tmplt_add(chain, net, tca, extack);
2885                 if (err) {
2886                         tcf_chain_put_explicitly_created(chain);
2887                         goto errout;
2888                 }
2889
2890                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2891                                 RTM_NEWCHAIN, false);
2892                 break;
2893         case RTM_DELCHAIN:
2894                 tfilter_notify_chain(net, skb, block, q, parent, n,
2895                                      chain, RTM_DELTFILTER, true);
2896                 /* Flush the chain first as the user requested chain removal. */
2897                 tcf_chain_flush(chain, true);
2898                 /* In case the chain was successfully deleted, put a reference
2899                  * to the chain previously taken during addition.
2900                  */
2901                 tcf_chain_put_explicitly_created(chain);
2902                 break;
2903         case RTM_GETCHAIN:
2904                 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
2905                                       n->nlmsg_seq, n->nlmsg_type, true);
2906                 if (err < 0)
2907                         NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2908                 break;
2909         default:
2910                 err = -EOPNOTSUPP;
2911                 NL_SET_ERR_MSG(extack, "Unsupported message type");
2912                 goto errout;
2913         }
2914
2915 errout:
2916         tcf_chain_put(chain);
2917 errout_block:
2918         tcf_block_release(q, block, true);
2919         if (err == -EAGAIN)
2920                 /* Replay the request. */
2921                 goto replay;
2922         return err;
2923
2924 errout_block_locked:
2925         mutex_unlock(&block->lock);
2926         goto errout_block;
2927 }
2928
2929 /* called with RTNL */
2930 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2931 {
2932         struct net *net = sock_net(skb->sk);
2933         struct nlattr *tca[TCA_MAX + 1];
2934         struct Qdisc *q = NULL;
2935         struct tcf_block *block;
2936         struct tcmsg *tcm = nlmsg_data(cb->nlh);
2937         struct tcf_chain *chain;
2938         long index_start;
2939         long index;
2940         u32 parent;
2941         int err;
2942
2943         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2944                 return skb->len;
2945
2946         err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2947                                      rtm_tca_policy, cb->extack);
2948         if (err)
2949                 return err;
2950
2951         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2952                 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2953                 if (!block)
2954                         goto out;
2955                 /* If we work with block index, q is NULL and parent value
2956                  * will never be used in the following code. The check
2957                  * in tcf_fill_node prevents it. However, compiler does not
2958                  * see that far, so set parent to zero to silence the warning
2959                  * about parent being uninitialized.
2960                  */
2961                 parent = 0;
2962         } else {
2963                 const struct Qdisc_class_ops *cops;
2964                 struct net_device *dev;
2965                 unsigned long cl = 0;
2966
2967                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2968                 if (!dev)
2969                         return skb->len;
2970
2971                 parent = tcm->tcm_parent;
2972                 if (!parent) {
2973                         q = dev->qdisc;
2974                         parent = q->handle;
2975                 } else {
2976                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2977                 }
2978                 if (!q)
2979                         goto out;
2980                 cops = q->ops->cl_ops;
2981                 if (!cops)
2982                         goto out;
2983                 if (!cops->tcf_block)
2984                         goto out;
2985                 if (TC_H_MIN(tcm->tcm_parent)) {
2986                         cl = cops->find(q, tcm->tcm_parent);
2987                         if (cl == 0)
2988                                 goto out;
2989                 }
2990                 block = cops->tcf_block(q, cl, NULL);
2991                 if (!block)
2992                         goto out;
2993                 if (tcf_block_shared(block))
2994                         q = NULL;
2995         }
2996
2997         index_start = cb->args[0];
2998         index = 0;
2999
3000         mutex_lock(&block->lock);
3001         list_for_each_entry(chain, &block->chain_list, list) {
3002                 if ((tca[TCA_CHAIN] &&
3003                      nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3004                         continue;
3005                 if (index < index_start) {
3006                         index++;
3007                         continue;
3008                 }
3009                 if (tcf_chain_held_by_acts_only(chain))
3010                         continue;
3011                 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3012                                          chain->index, net, skb, block,
3013                                          NETLINK_CB(cb->skb).portid,
3014                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
3015                                          RTM_NEWCHAIN);
3016                 if (err <= 0)
3017                         break;
3018                 index++;
3019         }
3020         mutex_unlock(&block->lock);
3021
3022         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3023                 tcf_block_refcnt_put(block, true);
3024         cb->args[0] = index;
3025
3026 out:
3027         /* If we did no progress, the error (EMSGSIZE) is real */
3028         if (skb->len == 0 && err)
3029                 return err;
3030         return skb->len;
3031 }
3032
3033 void tcf_exts_destroy(struct tcf_exts *exts)
3034 {
3035 #ifdef CONFIG_NET_CLS_ACT
3036         tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3037         kfree(exts->actions);
3038         exts->nr_actions = 0;
3039 #endif
3040 }
3041 EXPORT_SYMBOL(tcf_exts_destroy);
3042
3043 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3044                       struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
3045                       bool rtnl_held, struct netlink_ext_ack *extack)
3046 {
3047 #ifdef CONFIG_NET_CLS_ACT
3048         {
3049                 struct tc_action *act;
3050                 size_t attr_size = 0;
3051
3052                 if (exts->police && tb[exts->police]) {
3053                         act = tcf_action_init_1(net, tp, tb[exts->police],
3054                                                 rate_tlv, "police", ovr,
3055                                                 TCA_ACT_BIND, rtnl_held,
3056                                                 extack);
3057                         if (IS_ERR(act))
3058                                 return PTR_ERR(act);
3059
3060                         act->type = exts->type = TCA_OLD_COMPAT;
3061                         exts->actions[0] = act;
3062                         exts->nr_actions = 1;
3063                 } else if (exts->action && tb[exts->action]) {
3064                         int err;
3065
3066                         err = tcf_action_init(net, tp, tb[exts->action],
3067                                               rate_tlv, NULL, ovr, TCA_ACT_BIND,
3068                                               exts->actions, &attr_size,
3069                                               rtnl_held, extack);
3070                         if (err < 0)
3071                                 return err;
3072                         exts->nr_actions = err;
3073                 }
3074         }
3075 #else
3076         if ((exts->action && tb[exts->action]) ||
3077             (exts->police && tb[exts->police])) {
3078                 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3079                 return -EOPNOTSUPP;
3080         }
3081 #endif
3082
3083         return 0;
3084 }
3085 EXPORT_SYMBOL(tcf_exts_validate);
3086
3087 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3088 {
3089 #ifdef CONFIG_NET_CLS_ACT
3090         struct tcf_exts old = *dst;
3091
3092         *dst = *src;
3093         tcf_exts_destroy(&old);
3094 #endif
3095 }
3096 EXPORT_SYMBOL(tcf_exts_change);
3097
3098 #ifdef CONFIG_NET_CLS_ACT
3099 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3100 {
3101         if (exts->nr_actions == 0)
3102                 return NULL;
3103         else
3104                 return exts->actions[0];
3105 }
3106 #endif
3107
3108 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3109 {
3110 #ifdef CONFIG_NET_CLS_ACT
3111         struct nlattr *nest;
3112
3113         if (exts->action && tcf_exts_has_actions(exts)) {
3114                 /*
3115                  * again for backward compatible mode - we want
3116                  * to work with both old and new modes of entering
3117                  * tc data even if iproute2  was newer - jhs
3118                  */
3119                 if (exts->type != TCA_OLD_COMPAT) {
3120                         nest = nla_nest_start_noflag(skb, exts->action);
3121                         if (nest == NULL)
3122                                 goto nla_put_failure;
3123
3124                         if (tcf_action_dump(skb, exts->actions, 0, 0) < 0)
3125                                 goto nla_put_failure;
3126                         nla_nest_end(skb, nest);
3127                 } else if (exts->police) {
3128                         struct tc_action *act = tcf_exts_first_act(exts);
3129                         nest = nla_nest_start_noflag(skb, exts->police);
3130                         if (nest == NULL || !act)
3131                                 goto nla_put_failure;
3132                         if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3133                                 goto nla_put_failure;
3134                         nla_nest_end(skb, nest);
3135                 }
3136         }
3137         return 0;
3138
3139 nla_put_failure:
3140         nla_nest_cancel(skb, nest);
3141         return -1;
3142 #else
3143         return 0;
3144 #endif
3145 }
3146 EXPORT_SYMBOL(tcf_exts_dump);
3147
3148
3149 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3150 {
3151 #ifdef CONFIG_NET_CLS_ACT
3152         struct tc_action *a = tcf_exts_first_act(exts);
3153         if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3154                 return -1;
3155 #endif
3156         return 0;
3157 }
3158 EXPORT_SYMBOL(tcf_exts_dump_stats);
3159
3160 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3161                      void *type_data, bool err_stop)
3162 {
3163         struct tcf_block_cb *block_cb;
3164         int ok_count = 0;
3165         int err;
3166
3167         /* Make sure all netdevs sharing this block are offload-capable. */
3168         if (block->nooffloaddevcnt && err_stop)
3169                 return -EOPNOTSUPP;
3170
3171         list_for_each_entry(block_cb, &block->cb_list, list) {
3172                 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3173                 if (err) {
3174                         if (err_stop)
3175                                 return err;
3176                 } else {
3177                         ok_count++;
3178                 }
3179         }
3180         return ok_count;
3181 }
3182 EXPORT_SYMBOL(tc_setup_cb_call);
3183
3184 int tc_setup_flow_action(struct flow_action *flow_action,
3185                          const struct tcf_exts *exts)
3186 {
3187         const struct tc_action *act;
3188         int i, j, k;
3189
3190         if (!exts)
3191                 return 0;
3192
3193         j = 0;
3194         tcf_exts_for_each_action(i, act, exts) {
3195                 struct flow_action_entry *entry;
3196
3197                 entry = &flow_action->entries[j];
3198                 if (is_tcf_gact_ok(act)) {
3199                         entry->id = FLOW_ACTION_ACCEPT;
3200                 } else if (is_tcf_gact_shot(act)) {
3201                         entry->id = FLOW_ACTION_DROP;
3202                 } else if (is_tcf_gact_trap(act)) {
3203                         entry->id = FLOW_ACTION_TRAP;
3204                 } else if (is_tcf_gact_goto_chain(act)) {
3205                         entry->id = FLOW_ACTION_GOTO;
3206                         entry->chain_index = tcf_gact_goto_chain_index(act);
3207                 } else if (is_tcf_mirred_egress_redirect(act)) {
3208                         entry->id = FLOW_ACTION_REDIRECT;
3209                         entry->dev = tcf_mirred_dev(act);
3210                 } else if (is_tcf_mirred_egress_mirror(act)) {
3211                         entry->id = FLOW_ACTION_MIRRED;
3212                         entry->dev = tcf_mirred_dev(act);
3213                 } else if (is_tcf_vlan(act)) {
3214                         switch (tcf_vlan_action(act)) {
3215                         case TCA_VLAN_ACT_PUSH:
3216                                 entry->id = FLOW_ACTION_VLAN_PUSH;
3217                                 entry->vlan.vid = tcf_vlan_push_vid(act);
3218                                 entry->vlan.proto = tcf_vlan_push_proto(act);
3219                                 entry->vlan.prio = tcf_vlan_push_prio(act);
3220                                 break;
3221                         case TCA_VLAN_ACT_POP:
3222                                 entry->id = FLOW_ACTION_VLAN_POP;
3223                                 break;
3224                         case TCA_VLAN_ACT_MODIFY:
3225                                 entry->id = FLOW_ACTION_VLAN_MANGLE;
3226                                 entry->vlan.vid = tcf_vlan_push_vid(act);
3227                                 entry->vlan.proto = tcf_vlan_push_proto(act);
3228                                 entry->vlan.prio = tcf_vlan_push_prio(act);
3229                                 break;
3230                         default:
3231                                 goto err_out;
3232                         }
3233                 } else if (is_tcf_tunnel_set(act)) {
3234                         entry->id = FLOW_ACTION_TUNNEL_ENCAP;
3235                         entry->tunnel = tcf_tunnel_info(act);
3236                 } else if (is_tcf_tunnel_release(act)) {
3237                         entry->id = FLOW_ACTION_TUNNEL_DECAP;
3238                 } else if (is_tcf_pedit(act)) {
3239                         for (k = 0; k < tcf_pedit_nkeys(act); k++) {
3240                                 switch (tcf_pedit_cmd(act, k)) {
3241                                 case TCA_PEDIT_KEY_EX_CMD_SET:
3242                                         entry->id = FLOW_ACTION_MANGLE;
3243                                         break;
3244                                 case TCA_PEDIT_KEY_EX_CMD_ADD:
3245                                         entry->id = FLOW_ACTION_ADD;
3246                                         break;
3247                                 default:
3248                                         goto err_out;
3249                                 }
3250                                 entry->mangle.htype = tcf_pedit_htype(act, k);
3251                                 entry->mangle.mask = tcf_pedit_mask(act, k);
3252                                 entry->mangle.val = tcf_pedit_val(act, k);
3253                                 entry->mangle.offset = tcf_pedit_offset(act, k);
3254                                 entry = &flow_action->entries[++j];
3255                         }
3256                 } else if (is_tcf_csum(act)) {
3257                         entry->id = FLOW_ACTION_CSUM;
3258                         entry->csum_flags = tcf_csum_update_flags(act);
3259                 } else if (is_tcf_skbedit_mark(act)) {
3260                         entry->id = FLOW_ACTION_MARK;
3261                         entry->mark = tcf_skbedit_mark(act);
3262                 } else if (is_tcf_sample(act)) {
3263                         entry->id = FLOW_ACTION_SAMPLE;
3264                         entry->sample.psample_group =
3265                                 tcf_sample_psample_group(act);
3266                         entry->sample.trunc_size = tcf_sample_trunc_size(act);
3267                         entry->sample.truncate = tcf_sample_truncate(act);
3268                         entry->sample.rate = tcf_sample_rate(act);
3269                 } else if (is_tcf_police(act)) {
3270                         entry->id = FLOW_ACTION_POLICE;
3271                         entry->police.burst = tcf_police_tcfp_burst(act);
3272                         entry->police.rate_bytes_ps =
3273                                 tcf_police_rate_bytes_ps(act);
3274                 } else {
3275                         goto err_out;
3276                 }
3277
3278                 if (!is_tcf_pedit(act))
3279                         j++;
3280         }
3281         return 0;
3282 err_out:
3283         return -EOPNOTSUPP;
3284 }
3285 EXPORT_SYMBOL(tc_setup_flow_action);
3286
3287 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3288 {
3289         unsigned int num_acts = 0;
3290         struct tc_action *act;
3291         int i;
3292
3293         tcf_exts_for_each_action(i, act, exts) {
3294                 if (is_tcf_pedit(act))
3295                         num_acts += tcf_pedit_nkeys(act);
3296                 else
3297                         num_acts++;
3298         }
3299         return num_acts;
3300 }
3301 EXPORT_SYMBOL(tcf_exts_num_actions);
3302
3303 static __net_init int tcf_net_init(struct net *net)
3304 {
3305         struct tcf_net *tn = net_generic(net, tcf_net_id);
3306
3307         spin_lock_init(&tn->idr_lock);
3308         idr_init(&tn->idr);
3309         return 0;
3310 }
3311
3312 static void __net_exit tcf_net_exit(struct net *net)
3313 {
3314         struct tcf_net *tn = net_generic(net, tcf_net_id);
3315
3316         idr_destroy(&tn->idr);
3317 }
3318
3319 static struct pernet_operations tcf_net_ops = {
3320         .init = tcf_net_init,
3321         .exit = tcf_net_exit,
3322         .id   = &tcf_net_id,
3323         .size = sizeof(struct tcf_net),
3324 };
3325
3326 static int __init tc_filter_init(void)
3327 {
3328         int err;
3329
3330         tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3331         if (!tc_filter_wq)
3332                 return -ENOMEM;
3333
3334         err = register_pernet_subsys(&tcf_net_ops);
3335         if (err)
3336                 goto err_register_pernet_subsys;
3337
3338         err = rhashtable_init(&indr_setup_block_ht,
3339                               &tc_indr_setup_block_ht_params);
3340         if (err)
3341                 goto err_rhash_setup_block_ht;
3342
3343         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3344                       RTNL_FLAG_DOIT_UNLOCKED);
3345         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3346                       RTNL_FLAG_DOIT_UNLOCKED);
3347         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3348                       tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3349         rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3350         rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3351         rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3352                       tc_dump_chain, 0);
3353
3354         return 0;
3355
3356 err_rhash_setup_block_ht:
3357         unregister_pernet_subsys(&tcf_net_ops);
3358 err_register_pernet_subsys:
3359         destroy_workqueue(tc_filter_wq);
3360         return err;
3361 }
3362
3363 subsys_initcall(tc_filter_init);