net/sched/cls_api.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * net/sched/cls_api.c  Packet classifier API.
   4  *
   5  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
   6  *
   7  * Changes:
   8  *
   9  * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
  10  */
  11
  12 #include <linux/module.h>
  13 #include <linux/types.h>
  14 #include <linux/kernel.h>
  15 #include <linux/string.h>
  16 #include <linux/errno.h>
  17 #include <linux/err.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/init.h>
  20 #include <linux/kmod.h>
  21 #include <linux/slab.h>
  22 #include <linux/idr.h>
  23 #include <linux/rhashtable.h>
  24 #include <linux/jhash.h>
  25 #include <linux/rculist.h>
  26 #include <net/net_namespace.h>
  27 #include <net/sock.h>
  28 #include <net/netlink.h>
  29 #include <net/pkt_sched.h>
  30 #include <net/pkt_cls.h>
  31 #include <net/tc_act/tc_pedit.h>
  32 #include <net/tc_act/tc_mirred.h>
  33 #include <net/tc_act/tc_vlan.h>
  34 #include <net/tc_act/tc_tunnel_key.h>
  35 #include <net/tc_act/tc_csum.h>
  36 #include <net/tc_act/tc_gact.h>
  37 #include <net/tc_act/tc_police.h>
  38 #include <net/tc_act/tc_sample.h>
  39 #include <net/tc_act/tc_skbedit.h>
  40 #include <net/tc_act/tc_ct.h>
  41 #include <net/tc_act/tc_mpls.h>
  42 #include <net/tc_act/tc_gate.h>
  43 #include <net/flow_offload.h>
  44
  45 extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
  46
  47 /* The list of all installed classifier types */
  48 static LIST_HEAD(tcf_proto_base);
  49
  50 /* Protects list of registered TC modules. It is pure SMP lock. */
  51 static DEFINE_RWLOCK(cls_mod_lock);
  52
  53 static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
  54 {
  55         return jhash_3words(tp->chain->index, tp->prio,
  56                             (__force __u32)tp->protocol, 0);
  57 }
  58
  59 static void tcf_proto_signal_destroying(struct tcf_chain *chain,
  60                                         struct tcf_proto *tp)
  61 {
  62         struct tcf_block *block = chain->block;
  63
  64         mutex_lock(&block->proto_destroy_lock);
  65         hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
  66                      destroy_obj_hashfn(tp));
  67         mutex_unlock(&block->proto_destroy_lock);
  68 }
  69
  70 static bool tcf_proto_cmp(const struct tcf_proto *tp1,
  71                           const struct tcf_proto *tp2)
  72 {
  73         return tp1->chain->index == tp2->chain->index &&
  74                tp1->prio == tp2->prio &&
  75                tp1->protocol == tp2->protocol;
  76 }
  77
  78 static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
  79                                         struct tcf_proto *tp)
  80 {
  81         u32 hash = destroy_obj_hashfn(tp);
  82         struct tcf_proto *iter;
  83         bool found = false;
  84
  85         rcu_read_lock();
  86         hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
  87                                    destroy_ht_node, hash) {
  88                 if (tcf_proto_cmp(tp, iter)) {
  89                         found = true;
  90                         break;
  91                 }
  92         }
  93         rcu_read_unlock();
  94
  95         return found;
  96 }
  97
  98 static void
  99 tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
 100 {
 101         struct tcf_block *block = chain->block;
 102
 103         mutex_lock(&block->proto_destroy_lock);
 104         if (hash_hashed(&tp->destroy_ht_node))
 105                 hash_del_rcu(&tp->destroy_ht_node);
 106         mutex_unlock(&block->proto_destroy_lock);
 107 }
 108
 109 /* Find classifier type by string name */
 110
 111 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
 112 {
 113         const struct tcf_proto_ops *t, *res = NULL;
 114
 115         if (kind) {
 116                 read_lock(&cls_mod_lock);
 117                 list_for_each_entry(t, &tcf_proto_base, head) {
 118                         if (strcmp(kind, t->kind) == 0) {
 119                                 if (try_module_get(t->owner))
 120                                         res = t;
 121                                 break;
 122                         }
 123                 }
 124                 read_unlock(&cls_mod_lock);
 125         }
 126         return res;
 127 }
 128
 129 static const struct tcf_proto_ops *
 130 tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
 131                      struct netlink_ext_ack *extack)
 132 {
 133         const struct tcf_proto_ops *ops;
 134
 135         ops = __tcf_proto_lookup_ops(kind);
 136         if (ops)
 137                 return ops;
 138 #ifdef CONFIG_MODULES
 139         if (rtnl_held)
 140                 rtnl_unlock();
 141         request_module("cls_%s", kind);
 142         if (rtnl_held)
 143                 rtnl_lock();
 144         ops = __tcf_proto_lookup_ops(kind);
 145         /* We dropped the RTNL semaphore in order to perform
 146          * the module load. So, even if we succeeded in loading
 147          * the module we have to replay the request. We indicate
 148          * this using -EAGAIN.
 149          */
 150         if (ops) {
 151                 module_put(ops->owner);
 152                 return ERR_PTR(-EAGAIN);
 153         }
 154 #endif
 155         NL_SET_ERR_MSG(extack, "TC classifier not found");
 156         return ERR_PTR(-ENOENT);
 157 }
 158
 159 /* Register(unregister) new classifier type */
 160
 161 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
 162 {
 163         struct tcf_proto_ops *t;
 164         int rc = -EEXIST;
 165
 166         write_lock(&cls_mod_lock);
 167         list_for_each_entry(t, &tcf_proto_base, head)
 168                 if (!strcmp(ops->kind, t->kind))
 169                         goto out;
 170
 171         list_add_tail(&ops->head, &tcf_proto_base);
 172         rc = 0;
 173 out:
 174         write_unlock(&cls_mod_lock);
 175         return rc;
 176 }
 177 EXPORT_SYMBOL(register_tcf_proto_ops);
 178
 179 static struct workqueue_struct *tc_filter_wq;
 180
 181 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
 182 {
 183         struct tcf_proto_ops *t;
 184         int rc = -ENOENT;
 185
 186         /* Wait for outstanding call_rcu()s, if any, from a
 187          * tcf_proto_ops's destroy() handler.
 188          */
 189         rcu_barrier();
 190         flush_workqueue(tc_filter_wq);
 191
 192         write_lock(&cls_mod_lock);
 193         list_for_each_entry(t, &tcf_proto_base, head) {
 194                 if (t == ops) {
 195                         list_del(&t->head);
 196                         rc = 0;
 197                         break;
 198                 }
 199         }
 200         write_unlock(&cls_mod_lock);
 201         return rc;
 202 }
 203 EXPORT_SYMBOL(unregister_tcf_proto_ops);
 204
 205 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
 206 {
 207         INIT_RCU_WORK(rwork, func);
 208         return queue_rcu_work(tc_filter_wq, rwork);
 209 }
 210 EXPORT_SYMBOL(tcf_queue_work);
 211
 212 /* Select new prio value from the range, managed by kernel. */
 213
 214 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
 215 {
 216         u32 first = TC_H_MAKE(0xC0000000U, 0U);
 217
 218         if (tp)
 219                 first = tp->prio - 1;
 220
 221         return TC_H_MAJ(first);
 222 }
 223
 224 static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
 225 {
 226         if (kind)
 227                 return nla_strlcpy(name, kind, IFNAMSIZ) >= IFNAMSIZ;
 228         memset(name, 0, IFNAMSIZ);
 229         return false;
 230 }
 231
 232 static bool tcf_proto_is_unlocked(const char *kind)
 233 {
 234         const struct tcf_proto_ops *ops;
 235         bool ret;
 236
 237         if (strlen(kind) == 0)
 238                 return false;
 239
 240         ops = tcf_proto_lookup_ops(kind, false, NULL);
 241         /* On error return false to take rtnl lock. Proto lookup/create
 242          * functions will perform lookup again and properly handle errors.
 243          */
 244         if (IS_ERR(ops))
 245                 return false;
 246
 247         ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
 248         module_put(ops->owner);
 249         return ret;
 250 }
 251
 252 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
 253                                           u32 prio, struct tcf_chain *chain,
 254                                           bool rtnl_held,
 255                                           struct netlink_ext_ack *extack)
 256 {
 257         struct tcf_proto *tp;
 258         int err;
 259
 260         tp = kzalloc(sizeof(*tp), GFP_KERNEL);
 261         if (!tp)
 262                 return ERR_PTR(-ENOBUFS);
 263
 264         tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
 265         if (IS_ERR(tp->ops)) {
 266                 err = PTR_ERR(tp->ops);
 267                 goto errout;
 268         }
 269         tp->classify = tp->ops->classify;
 270         tp->protocol = protocol;
 271         tp->prio = prio;
 272         tp->chain = chain;
 273         spin_lock_init(&tp->lock);
 274         refcount_set(&tp->refcnt, 1);
 275
 276         err = tp->ops->init(tp);
 277         if (err) {
 278                 module_put(tp->ops->owner);
 279                 goto errout;
 280         }
 281         return tp;
 282
 283 errout:
 284         kfree(tp);
 285         return ERR_PTR(err);
 286 }
 287
 288 static void tcf_proto_get(struct tcf_proto *tp)
 289 {
 290         refcount_inc(&tp->refcnt);
 291 }
 292
 293 static void tcf_chain_put(struct tcf_chain *chain);
 294
 295 static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
 296                               bool sig_destroy, struct netlink_ext_ack *extack)
 297 {
 298         tp->ops->destroy(tp, rtnl_held, extack);
 299         if (sig_destroy)
 300                 tcf_proto_signal_destroyed(tp->chain, tp);
 301         tcf_chain_put(tp->chain);
 302         module_put(tp->ops->owner);
 303         kfree_rcu(tp, rcu);
 304 }
 305
 306 static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
 307                           struct netlink_ext_ack *extack)
 308 {
 309         if (refcount_dec_and_test(&tp->refcnt))
 310                 tcf_proto_destroy(tp, rtnl_held, true, extack);
 311 }
 312
 313 static bool tcf_proto_check_delete(struct tcf_proto *tp)
 314 {
 315         if (tp->ops->delete_empty)
 316                 return tp->ops->delete_empty(tp);
 317
 318         tp->deleting = true;
 319         return tp->deleting;
 320 }
 321
 322 static void tcf_proto_mark_delete(struct tcf_proto *tp)
 323 {
 324         spin_lock(&tp->lock);
 325         tp->deleting = true;
 326         spin_unlock(&tp->lock);
 327 }
 328
 329 static bool tcf_proto_is_deleting(struct tcf_proto *tp)
 330 {
 331         bool deleting;
 332
 333         spin_lock(&tp->lock);
 334         deleting = tp->deleting;
 335         spin_unlock(&tp->lock);
 336
 337         return deleting;
 338 }
 339
 340 #define ASSERT_BLOCK_LOCKED(block)                                      \
 341         lockdep_assert_held(&(block)->lock)
 342
 343 struct tcf_filter_chain_list_item {
 344         struct list_head list;
 345         tcf_chain_head_change_t *chain_head_change;
 346         void *chain_head_change_priv;
 347 };
 348
 349 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
 350                                           u32 chain_index)
 351 {
 352         struct tcf_chain *chain;
 353
 354         ASSERT_BLOCK_LOCKED(block);
 355
 356         chain = kzalloc(sizeof(*chain), GFP_KERNEL);
 357         if (!chain)
 358                 return NULL;
 359         list_add_tail_rcu(&chain->list, &block->chain_list);
 360         mutex_init(&chain->filter_chain_lock);
 361         chain->block = block;
 362         chain->index = chain_index;
 363         chain->refcnt = 1;
 364         if (!chain->index)
 365                 block->chain0.chain = chain;
 366         return chain;
 367 }
 368
 369 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
 370                                        struct tcf_proto *tp_head)
 371 {
 372         if (item->chain_head_change)
 373                 item->chain_head_change(tp_head, item->chain_head_change_priv);
 374 }
 375
 376 static void tcf_chain0_head_change(struct tcf_chain *chain,
 377                                    struct tcf_proto *tp_head)
 378 {
 379         struct tcf_filter_chain_list_item *item;
 380         struct tcf_block *block = chain->block;
 381
 382         if (chain->index)
 383                 return;
 384
 385         mutex_lock(&block->lock);
 386         list_for_each_entry(item, &block->chain0.filter_chain_list, list)
 387                 tcf_chain_head_change_item(item, tp_head);
 388         mutex_unlock(&block->lock);
 389 }
 390
 391 /* Returns true if block can be safely freed. */
 392
 393 static bool tcf_chain_detach(struct tcf_chain *chain)
 394 {
 395         struct tcf_block *block = chain->block;
 396
 397         ASSERT_BLOCK_LOCKED(block);
 398
 399         list_del_rcu(&chain->list);
 400         if (!chain->index)
 401                 block->chain0.chain = NULL;
 402
 403         if (list_empty(&block->chain_list) &&
 404             refcount_read(&block->refcnt) == 0)
 405                 return true;
 406
 407         return false;
 408 }
 409
 410 static void tcf_block_destroy(struct tcf_block *block)
 411 {
 412         mutex_destroy(&block->lock);
 413         mutex_destroy(&block->proto_destroy_lock);
 414         kfree_rcu(block, rcu);
 415 }
 416
 417 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
 418 {
 419         struct tcf_block *block = chain->block;
 420
 421         mutex_destroy(&chain->filter_chain_lock);
 422         kfree_rcu(chain, rcu);
 423         if (free_block)
 424                 tcf_block_destroy(block);
 425 }
 426
 427 static void tcf_chain_hold(struct tcf_chain *chain)
 428 {
 429         ASSERT_BLOCK_LOCKED(chain->block);
 430
 431         ++chain->refcnt;
 432 }
 433
 434 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
 435 {
 436         ASSERT_BLOCK_LOCKED(chain->block);
 437
 438         /* In case all the references are action references, this
 439          * chain should not be shown to the user.
 440          */
 441         return chain->refcnt == chain->action_refcnt;
 442 }
 443
 444 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
 445                                           u32 chain_index)
 446 {
 447         struct tcf_chain *chain;
 448
 449         ASSERT_BLOCK_LOCKED(block);
 450
 451         list_for_each_entry(chain, &block->chain_list, list) {
 452                 if (chain->index == chain_index)
 453                         return chain;
 454         }
 455         return NULL;
 456 }
 457
 458 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
 459 static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
 460                                               u32 chain_index)
 461 {
 462         struct tcf_chain *chain;
 463
 464         list_for_each_entry_rcu(chain, &block->chain_list, list) {
 465                 if (chain->index == chain_index)
 466                         return chain;
 467         }
 468         return NULL;
 469 }
 470 #endif
 471
 472 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
 473                            u32 seq, u16 flags, int event, bool unicast);
 474
 475 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
 476                                          u32 chain_index, bool create,
 477                                          bool by_act)
 478 {
 479         struct tcf_chain *chain = NULL;
 480         bool is_first_reference;
 481
 482         mutex_lock(&block->lock);
 483         chain = tcf_chain_lookup(block, chain_index);
 484         if (chain) {
 485                 tcf_chain_hold(chain);
 486         } else {
 487                 if (!create)
 488                         goto errout;
 489                 chain = tcf_chain_create(block, chain_index);
 490                 if (!chain)
 491                         goto errout;
 492         }
 493
 494         if (by_act)
 495                 ++chain->action_refcnt;
 496         is_first_reference = chain->refcnt - chain->action_refcnt == 1;
 497         mutex_unlock(&block->lock);
 498
 499         /* Send notification only in case we got the first
 500          * non-action reference. Until then, the chain acts only as
 501          * a placeholder for actions pointing to it and user ought
 502          * not know about them.
 503          */
 504         if (is_first_reference && !by_act)
 505                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
 506                                 RTM_NEWCHAIN, false);
 507
 508         return chain;
 509
 510 errout:
 511         mutex_unlock(&block->lock);
 512         return chain;
 513 }
 514
 515 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
 516                                        bool create)
 517 {
 518         return __tcf_chain_get(block, chain_index, create, false);
 519 }
 520
 521 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
 522 {
 523         return __tcf_chain_get(block, chain_index, true, true);
 524 }
 525 EXPORT_SYMBOL(tcf_chain_get_by_act);
 526
 527 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
 528                                void *tmplt_priv);
 529 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
 530                                   void *tmplt_priv, u32 chain_index,
 531                                   struct tcf_block *block, struct sk_buff *oskb,
 532                                   u32 seq, u16 flags, bool unicast);
 533
 534 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
 535                             bool explicitly_created)
 536 {
 537         struct tcf_block *block = chain->block;
 538         const struct tcf_proto_ops *tmplt_ops;
 539         bool free_block = false;
 540         unsigned int refcnt;
 541         void *tmplt_priv;
 542
 543         mutex_lock(&block->lock);
 544         if (explicitly_created) {
 545                 if (!chain->explicitly_created) {
 546                         mutex_unlock(&block->lock);
 547                         return;
 548                 }
 549                 chain->explicitly_created = false;
 550         }
 551
 552         if (by_act)
 553                 chain->action_refcnt--;
 554
 555         /* tc_chain_notify_delete can't be called while holding block lock.
 556          * However, when block is unlocked chain can be changed concurrently, so
 557          * save these to temporary variables.
 558          */
 559         refcnt = --chain->refcnt;
 560         tmplt_ops = chain->tmplt_ops;
 561         tmplt_priv = chain->tmplt_priv;
 562
 563         /* The last dropped non-action reference will trigger notification. */
 564         if (refcnt - chain->action_refcnt == 0 && !by_act) {
 565                 tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
 566                                        block, NULL, 0, 0, false);
 567                 /* Last reference to chain, no need to lock. */
 568                 chain->flushing = false;
 569         }
 570
 571         if (refcnt == 0)
 572                 free_block = tcf_chain_detach(chain);
 573         mutex_unlock(&block->lock);
 574
 575         if (refcnt == 0) {
 576                 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
 577                 tcf_chain_destroy(chain, free_block);
 578         }
 579 }
 580
 581 static void tcf_chain_put(struct tcf_chain *chain)
 582 {
 583         __tcf_chain_put(chain, false, false);
 584 }
 585
 586 void tcf_chain_put_by_act(struct tcf_chain *chain)
 587 {
 588         __tcf_chain_put(chain, true, false);
 589 }
 590 EXPORT_SYMBOL(tcf_chain_put_by_act);
 591
 592 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
 593 {
 594         __tcf_chain_put(chain, false, true);
 595 }
 596
 597 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
 598 {
 599         struct tcf_proto *tp, *tp_next;
 600
 601         mutex_lock(&chain->filter_chain_lock);
 602         tp = tcf_chain_dereference(chain->filter_chain, chain);
 603         while (tp) {
 604                 tp_next = rcu_dereference_protected(tp->next, 1);
 605                 tcf_proto_signal_destroying(chain, tp);
 606                 tp = tp_next;
 607         }
 608         tp = tcf_chain_dereference(chain->filter_chain, chain);
 609         RCU_INIT_POINTER(chain->filter_chain, NULL);
 610         tcf_chain0_head_change(chain, NULL);
 611         chain->flushing = true;
 612         mutex_unlock(&chain->filter_chain_lock);
 613
 614         while (tp) {
 615                 tp_next = rcu_dereference_protected(tp->next, 1);
 616                 tcf_proto_put(tp, rtnl_held, NULL);
 617                 tp = tp_next;
 618         }
 619 }
 620
 621 static int tcf_block_setup(struct tcf_block *block,
 622                            struct flow_block_offload *bo);
 623
 624 static void tcf_block_offload_init(struct flow_block_offload *bo,
 625                                    struct net_device *dev, struct Qdisc *sch,
 626                                    enum flow_block_command command,
 627                                    enum flow_block_binder_type binder_type,
 628                                    struct flow_block *flow_block,
 629                                    bool shared, struct netlink_ext_ack *extack)
 630 {
 631         bo->net = dev_net(dev);
 632         bo->command = command;
 633         bo->binder_type = binder_type;
 634         bo->block = flow_block;
 635         bo->block_shared = shared;
 636         bo->extack = extack;
 637         bo->sch = sch;
 638         INIT_LIST_HEAD(&bo->cb_list);
 639 }
 640
 641 static void tcf_block_unbind(struct tcf_block *block,
 642                              struct flow_block_offload *bo);
 643
 644 static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
 645 {
 646         struct tcf_block *block = block_cb->indr.data;
 647         struct net_device *dev = block_cb->indr.dev;
 648         struct Qdisc *sch = block_cb->indr.sch;
 649         struct netlink_ext_ack extack = {};
 650         struct flow_block_offload bo;
 651
 652         tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
 653                                block_cb->indr.binder_type,
 654                                &block->flow_block, tcf_block_shared(block),
 655                                &extack);
 656         down_write(&block->cb_lock);
 657         list_del(&block_cb->driver_list);
 658         list_move(&block_cb->list, &bo.cb_list);
 659         up_write(&block->cb_lock);
 660         rtnl_lock();
 661         tcf_block_unbind(block, &bo);
 662         rtnl_unlock();
 663 }
 664
 665 static bool tcf_block_offload_in_use(struct tcf_block *block)
 666 {
 667         return atomic_read(&block->offloadcnt);
 668 }
 669
 670 static int tcf_block_offload_cmd(struct tcf_block *block,
 671                                  struct net_device *dev, struct Qdisc *sch,
 672                                  struct tcf_block_ext_info *ei,
 673                                  enum flow_block_command command,
 674                                  struct netlink_ext_ack *extack)
 675 {
 676         struct flow_block_offload bo = {};
 677
 678         tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
 679                                &block->flow_block, tcf_block_shared(block),
 680                                extack);
 681
 682         if (dev->netdev_ops->ndo_setup_tc) {
 683                 int err;
 684
 685                 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
 686                 if (err < 0) {
 687                         if (err != -EOPNOTSUPP)
 688                                 NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
 689                         return err;
 690                 }
 691
 692                 return tcf_block_setup(block, &bo);
 693         }
 694
 695         flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
 696                                     tc_block_indr_cleanup);
 697         tcf_block_setup(block, &bo);
 698
 699         return -EOPNOTSUPP;
 700 }
 701
 702 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
 703                                   struct tcf_block_ext_info *ei,
 704                                   struct netlink_ext_ack *extack)
 705 {
 706         struct net_device *dev = q->dev_queue->dev;
 707         int err;
 708
 709         down_write(&block->cb_lock);
 710
 711         /* If tc offload feature is disabled and the block we try to bind
 712          * to already has some offloaded filters, forbid to bind.
 713          */
 714         if (dev->netdev_ops->ndo_setup_tc &&
 715             !tc_can_offload(dev) &&
 716             tcf_block_offload_in_use(block)) {
 717                 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
 718                 err = -EOPNOTSUPP;
 719                 goto err_unlock;
 720         }
 721
 722         err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
 723         if (err == -EOPNOTSUPP)
 724                 goto no_offload_dev_inc;
 725         if (err)
 726                 goto err_unlock;
 727
 728         up_write(&block->cb_lock);
 729         return 0;
 730
 731 no_offload_dev_inc:
 732         if (tcf_block_offload_in_use(block))
 733                 goto err_unlock;
 734
 735         err = 0;
 736         block->nooffloaddevcnt++;
 737 err_unlock:
 738         up_write(&block->cb_lock);
 739         return err;
 740 }
 741
 742 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
 743                                      struct tcf_block_ext_info *ei)
 744 {
 745         struct net_device *dev = q->dev_queue->dev;
 746         int err;
 747
 748         down_write(&block->cb_lock);
 749         err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
 750         if (err == -EOPNOTSUPP)
 751                 goto no_offload_dev_dec;
 752         up_write(&block->cb_lock);
 753         return;
 754
 755 no_offload_dev_dec:
 756         WARN_ON(block->nooffloaddevcnt-- == 0);
 757         up_write(&block->cb_lock);
 758 }
 759
 760 static int
 761 tcf_chain0_head_change_cb_add(struct tcf_block *block,
 762                               struct tcf_block_ext_info *ei,
 763                               struct netlink_ext_ack *extack)
 764 {
 765         struct tcf_filter_chain_list_item *item;
 766         struct tcf_chain *chain0;
 767
 768         item = kmalloc(sizeof(*item), GFP_KERNEL);
 769         if (!item) {
 770                 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
 771                 return -ENOMEM;
 772         }
 773         item->chain_head_change = ei->chain_head_change;
 774         item->chain_head_change_priv = ei->chain_head_change_priv;
 775
 776         mutex_lock(&block->lock);
 777         chain0 = block->chain0.chain;
 778         if (chain0)
 779                 tcf_chain_hold(chain0);
 780         else
 781                 list_add(&item->list, &block->chain0.filter_chain_list);
 782         mutex_unlock(&block->lock);
 783
 784         if (chain0) {
 785                 struct tcf_proto *tp_head;
 786
 787                 mutex_lock(&chain0->filter_chain_lock);
 788
 789                 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
 790                 if (tp_head)
 791                         tcf_chain_head_change_item(item, tp_head);
 792
 793                 mutex_lock(&block->lock);
 794                 list_add(&item->list, &block->chain0.filter_chain_list);
 795                 mutex_unlock(&block->lock);
 796
 797                 mutex_unlock(&chain0->filter_chain_lock);
 798                 tcf_chain_put(chain0);
 799         }
 800
 801         return 0;
 802 }
 803
 804 static void
 805 tcf_chain0_head_change_cb_del(struct tcf_block *block,
 806                               struct tcf_block_ext_info *ei)
 807 {
 808         struct tcf_filter_chain_list_item *item;
 809
 810         mutex_lock(&block->lock);
 811         list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
 812                 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
 813                     (item->chain_head_change == ei->chain_head_change &&
 814                      item->chain_head_change_priv == ei->chain_head_change_priv)) {
 815                         if (block->chain0.chain)
 816                                 tcf_chain_head_change_item(item, NULL);
 817                         list_del(&item->list);
 818                         mutex_unlock(&block->lock);
 819
 820                         kfree(item);
 821                         return;
 822                 }
 823         }
 824         mutex_unlock(&block->lock);
 825         WARN_ON(1);
 826 }
 827
 828 struct tcf_net {
 829         spinlock_t idr_lock; /* Protects idr */
 830         struct idr idr;
 831 };
 832
 833 static unsigned int tcf_net_id;
 834
 835 static int tcf_block_insert(struct tcf_block *block, struct net *net,
 836                             struct netlink_ext_ack *extack)
 837 {
 838         struct tcf_net *tn = net_generic(net, tcf_net_id);
 839         int err;
 840
 841         idr_preload(GFP_KERNEL);
 842         spin_lock(&tn->idr_lock);
 843         err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
 844                             GFP_NOWAIT);
 845         spin_unlock(&tn->idr_lock);
 846         idr_preload_end();
 847
 848         return err;
 849 }
 850
 851 static void tcf_block_remove(struct tcf_block *block, struct net *net)
 852 {
 853         struct tcf_net *tn = net_generic(net, tcf_net_id);
 854
 855         spin_lock(&tn->idr_lock);
 856         idr_remove(&tn->idr, block->index);
 857         spin_unlock(&tn->idr_lock);
 858 }
 859
 860 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
 861                                           u32 block_index,
 862                                           struct netlink_ext_ack *extack)
 863 {
 864         struct tcf_block *block;
 865
 866         block = kzalloc(sizeof(*block), GFP_KERNEL);
 867         if (!block) {
 868                 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
 869                 return ERR_PTR(-ENOMEM);
 870         }
 871         mutex_init(&block->lock);
 872         mutex_init(&block->proto_destroy_lock);
 873         init_rwsem(&block->cb_lock);
 874         flow_block_init(&block->flow_block);
 875         INIT_LIST_HEAD(&block->chain_list);
 876         INIT_LIST_HEAD(&block->owner_list);
 877         INIT_LIST_HEAD(&block->chain0.filter_chain_list);
 878
 879         refcount_set(&block->refcnt, 1);
 880         block->net = net;
 881         block->index = block_index;
 882
 883         /* Don't store q pointer for blocks which are shared */
 884         if (!tcf_block_shared(block))
 885                 block->q = q;
 886         return block;
 887 }
 888
 889 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
 890 {
 891         struct tcf_net *tn = net_generic(net, tcf_net_id);
 892
 893         return idr_find(&tn->idr, block_index);
 894 }
 895
 896 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
 897 {
 898         struct tcf_block *block;
 899
 900         rcu_read_lock();
 901         block = tcf_block_lookup(net, block_index);
 902         if (block && !refcount_inc_not_zero(&block->refcnt))
 903                 block = NULL;
 904         rcu_read_unlock();
 905
 906         return block;
 907 }
 908
 909 static struct tcf_chain *
 910 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
 911 {
 912         mutex_lock(&block->lock);
 913         if (chain)
 914                 chain = list_is_last(&chain->list, &block->chain_list) ?
 915                         NULL : list_next_entry(chain, list);
 916         else
 917                 chain = list_first_entry_or_null(&block->chain_list,
 918                                                  struct tcf_chain, list);
 919
 920         /* skip all action-only chains */
 921         while (chain && tcf_chain_held_by_acts_only(chain))
 922                 chain = list_is_last(&chain->list, &block->chain_list) ?
 923                         NULL : list_next_entry(chain, list);
 924
 925         if (chain)
 926                 tcf_chain_hold(chain);
 927         mutex_unlock(&block->lock);
 928
 929         return chain;
 930 }
 931
 932 /* Function to be used by all clients that want to iterate over all chains on
 933  * block. It properly obtains block->lock and takes reference to chain before
 934  * returning it. Users of this function must be tolerant to concurrent chain
 935  * insertion/deletion or ensure that no concurrent chain modification is
 936  * possible. Note that all netlink dump callbacks cannot guarantee to provide
 937  * consistent dump because rtnl lock is released each time skb is filled with
 938  * data and sent to user-space.
 939  */
 940
 941 struct tcf_chain *
 942 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
 943 {
 944         struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
 945
 946         if (chain)
 947                 tcf_chain_put(chain);
 948
 949         return chain_next;
 950 }
 951 EXPORT_SYMBOL(tcf_get_next_chain);
 952
 953 static struct tcf_proto *
 954 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
 955 {
 956         u32 prio = 0;
 957
 958         ASSERT_RTNL();
 959         mutex_lock(&chain->filter_chain_lock);
 960
 961         if (!tp) {
 962                 tp = tcf_chain_dereference(chain->filter_chain, chain);
 963         } else if (tcf_proto_is_deleting(tp)) {
 964                 /* 'deleting' flag is set and chain->filter_chain_lock was
 965                  * unlocked, which means next pointer could be invalid. Restart
 966                  * search.
 967                  */
 968                 prio = tp->prio + 1;
 969                 tp = tcf_chain_dereference(chain->filter_chain, chain);
 970
 971                 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
 972                         if (!tp->deleting && tp->prio >= prio)
 973                                 break;
 974         } else {
 975                 tp = tcf_chain_dereference(tp->next, chain);
 976         }
 977
 978         if (tp)
 979                 tcf_proto_get(tp);
 980
 981         mutex_unlock(&chain->filter_chain_lock);
 982
 983         return tp;
 984 }
 985
 986 /* Function to be used by all clients that want to iterate over all tp's on
 987  * chain. Users of this function must be tolerant to concurrent tp
 988  * insertion/deletion or ensure that no concurrent chain modification is
 989  * possible. Note that all netlink dump callbacks cannot guarantee to provide
 990  * consistent dump because rtnl lock is released each time skb is filled with
 991  * data and sent to user-space.
 992  */
 993
 994 struct tcf_proto *
 995 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp,
 996                    bool rtnl_held)
 997 {
 998         struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
 999
1000         if (tp)
1001                 tcf_proto_put(tp, rtnl_held, NULL);
1002
1003         return tp_next;
1004 }
1005 EXPORT_SYMBOL(tcf_get_next_proto);
1006
1007 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1008 {
1009         struct tcf_chain *chain;
1010
1011         /* Last reference to block. At this point chains cannot be added or
1012          * removed concurrently.
1013          */
1014         for (chain = tcf_get_next_chain(block, NULL);
1015              chain;
1016              chain = tcf_get_next_chain(block, chain)) {
1017                 tcf_chain_put_explicitly_created(chain);
1018                 tcf_chain_flush(chain, rtnl_held);
1019         }
1020 }
1021
1022 /* Lookup Qdisc and increments its reference counter.
1023  * Set parent, if necessary.
1024  */
1025
1026 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1027                             u32 *parent, int ifindex, bool rtnl_held,
1028                             struct netlink_ext_ack *extack)
1029 {
1030         const struct Qdisc_class_ops *cops;
1031         struct net_device *dev;
1032         int err = 0;
1033
1034         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1035                 return 0;
1036
1037         rcu_read_lock();
1038
1039         /* Find link */
1040         dev = dev_get_by_index_rcu(net, ifindex);
1041         if (!dev) {
1042                 rcu_read_unlock();
1043                 return -ENODEV;
1044         }
1045
1046         /* Find qdisc */
1047         if (!*parent) {
1048                 *q = dev->qdisc;
1049                 *parent = (*q)->handle;
1050         } else {
1051                 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1052                 if (!*q) {
1053                         NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1054                         err = -EINVAL;
1055                         goto errout_rcu;
1056                 }
1057         }
1058
1059         *q = qdisc_refcount_inc_nz(*q);
1060         if (!*q) {
1061                 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1062                 err = -EINVAL;
1063                 goto errout_rcu;
1064         }
1065
1066         /* Is it classful? */
1067         cops = (*q)->ops->cl_ops;
1068         if (!cops) {
1069                 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1070                 err = -EINVAL;
1071                 goto errout_qdisc;
1072         }
1073
1074         if (!cops->tcf_block) {
1075                 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1076                 err = -EOPNOTSUPP;
1077                 goto errout_qdisc;
1078         }
1079
1080 errout_rcu:
1081         /* At this point we know that qdisc is not noop_qdisc,
1082          * which means that qdisc holds a reference to net_device
1083          * and we hold a reference to qdisc, so it is safe to release
1084          * rcu read lock.
1085          */
1086         rcu_read_unlock();
1087         return err;
1088
1089 errout_qdisc:
1090         rcu_read_unlock();
1091
1092         if (rtnl_held)
1093                 qdisc_put(*q);
1094         else
1095                 qdisc_put_unlocked(*q);
1096         *q = NULL;
1097
1098         return err;
1099 }
1100
1101 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1102                                int ifindex, struct netlink_ext_ack *extack)
1103 {
1104         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1105                 return 0;
1106
1107         /* Do we search for filter, attached to class? */
1108         if (TC_H_MIN(parent)) {
1109                 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1110
1111                 *cl = cops->find(q, parent);
1112                 if (*cl == 0) {
1113                         NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1114                         return -ENOENT;
1115                 }
1116         }
1117
1118         return 0;
1119 }
1120
1121 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1122                                           unsigned long cl, int ifindex,
1123                                           u32 block_index,
1124                                           struct netlink_ext_ack *extack)
1125 {
1126         struct tcf_block *block;
1127
1128         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1129                 block = tcf_block_refcnt_get(net, block_index);
1130                 if (!block) {
1131                         NL_SET_ERR_MSG(extack, "Block of given index was not found");
1132                         return ERR_PTR(-EINVAL);
1133                 }
1134         } else {
1135                 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1136
1137                 block = cops->tcf_block(q, cl, extack);
1138                 if (!block)
1139                         return ERR_PTR(-EINVAL);
1140
1141                 if (tcf_block_shared(block)) {
1142                         NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1143                         return ERR_PTR(-EOPNOTSUPP);
1144                 }
1145
1146                 /* Always take reference to block in order to support execution
1147                  * of rules update path of cls API without rtnl lock. Caller
1148                  * must release block when it is finished using it. 'if' block
1149                  * of this conditional obtain reference to block by calling
1150                  * tcf_block_refcnt_get().
1151                  */
1152                 refcount_inc(&block->refcnt);
1153         }
1154
1155         return block;
1156 }
1157
1158 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1159                             struct tcf_block_ext_info *ei, bool rtnl_held)
1160 {
1161         if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1162                 /* Flushing/putting all chains will cause the block to be
1163                  * deallocated when last chain is freed. However, if chain_list
1164                  * is empty, block has to be manually deallocated. After block
1165                  * reference counter reached 0, it is no longer possible to
1166                  * increment it or add new chains to block.
1167                  */
1168                 bool free_block = list_empty(&block->chain_list);
1169
1170                 mutex_unlock(&block->lock);
1171                 if (tcf_block_shared(block))
1172                         tcf_block_remove(block, block->net);
1173
1174                 if (q)
1175                         tcf_block_offload_unbind(block, q, ei);
1176
1177                 if (free_block)
1178                         tcf_block_destroy(block);
1179                 else
1180                         tcf_block_flush_all_chains(block, rtnl_held);
1181         } else if (q) {
1182                 tcf_block_offload_unbind(block, q, ei);
1183         }
1184 }
1185
1186 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1187 {
1188         __tcf_block_put(block, NULL, NULL, rtnl_held);
1189 }
1190
1191 /* Find tcf block.
1192  * Set q, parent, cl when appropriate.
1193  */
1194
1195 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1196                                         u32 *parent, unsigned long *cl,
1197                                         int ifindex, u32 block_index,
1198                                         struct netlink_ext_ack *extack)
1199 {
1200         struct tcf_block *block;
1201         int err = 0;
1202
1203         ASSERT_RTNL();
1204
1205         err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1206         if (err)
1207                 goto errout;
1208
1209         err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1210         if (err)
1211                 goto errout_qdisc;
1212
1213         block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1214         if (IS_ERR(block)) {
1215                 err = PTR_ERR(block);
1216                 goto errout_qdisc;
1217         }
1218
1219         return block;
1220
1221 errout_qdisc:
1222         if (*q)
1223                 qdisc_put(*q);
1224 errout:
1225         *q = NULL;
1226         return ERR_PTR(err);
1227 }
1228
1229 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1230                               bool rtnl_held)
1231 {
1232         if (!IS_ERR_OR_NULL(block))
1233                 tcf_block_refcnt_put(block, rtnl_held);
1234
1235         if (q) {
1236                 if (rtnl_held)
1237                         qdisc_put(q);
1238                 else
1239                         qdisc_put_unlocked(q);
1240         }
1241 }
1242
1243 struct tcf_block_owner_item {
1244         struct list_head list;
1245         struct Qdisc *q;
1246         enum flow_block_binder_type binder_type;
1247 };
1248
1249 static void
1250 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1251                                struct Qdisc *q,
1252                                enum flow_block_binder_type binder_type)
1253 {
1254         if (block->keep_dst &&
1255             binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1256             binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1257                 netif_keep_dst(qdisc_dev(q));
1258 }
1259
1260 void tcf_block_netif_keep_dst(struct tcf_block *block)
1261 {
1262         struct tcf_block_owner_item *item;
1263
1264         block->keep_dst = true;
1265         list_for_each_entry(item, &block->owner_list, list)
1266                 tcf_block_owner_netif_keep_dst(block, item->q,
1267                                                item->binder_type);
1268 }
1269 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1270
1271 static int tcf_block_owner_add(struct tcf_block *block,
1272                                struct Qdisc *q,
1273                                enum flow_block_binder_type binder_type)
1274 {
1275         struct tcf_block_owner_item *item;
1276
1277         item = kmalloc(sizeof(*item), GFP_KERNEL);
1278         if (!item)
1279                 return -ENOMEM;
1280         item->q = q;
1281         item->binder_type = binder_type;
1282         list_add(&item->list, &block->owner_list);
1283         return 0;
1284 }
1285
1286 static void tcf_block_owner_del(struct tcf_block *block,
1287                                 struct Qdisc *q,
1288                                 enum flow_block_binder_type binder_type)
1289 {
1290         struct tcf_block_owner_item *item;
1291
1292         list_for_each_entry(item, &block->owner_list, list) {
1293                 if (item->q == q && item->binder_type == binder_type) {
1294                         list_del(&item->list);
1295                         kfree(item);
1296                         return;
1297                 }
1298         }
1299         WARN_ON(1);
1300 }
1301
1302 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1303                       struct tcf_block_ext_info *ei,
1304                       struct netlink_ext_ack *extack)
1305 {
1306         struct net *net = qdisc_net(q);
1307         struct tcf_block *block = NULL;
1308         int err;
1309
1310         if (ei->block_index)
1311                 /* block_index not 0 means the shared block is requested */
1312                 block = tcf_block_refcnt_get(net, ei->block_index);
1313
1314         if (!block) {
1315                 block = tcf_block_create(net, q, ei->block_index, extack);
1316                 if (IS_ERR(block))
1317                         return PTR_ERR(block);
1318                 if (tcf_block_shared(block)) {
1319                         err = tcf_block_insert(block, net, extack);
1320                         if (err)
1321                                 goto err_block_insert;
1322                 }
1323         }
1324
1325         err = tcf_block_owner_add(block, q, ei->binder_type);
1326         if (err)
1327                 goto err_block_owner_add;
1328
1329         tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1330
1331         err = tcf_chain0_head_change_cb_add(block, ei, extack);
1332         if (err)
1333                 goto err_chain0_head_change_cb_add;
1334
1335         err = tcf_block_offload_bind(block, q, ei, extack);
1336         if (err)
1337                 goto err_block_offload_bind;
1338
1339         *p_block = block;
1340         return 0;
1341
1342 err_block_offload_bind:
1343         tcf_chain0_head_change_cb_del(block, ei);
1344 err_chain0_head_change_cb_add:
1345         tcf_block_owner_del(block, q, ei->binder_type);
1346 err_block_owner_add:
1347 err_block_insert:
1348         tcf_block_refcnt_put(block, true);
1349         return err;
1350 }
1351 EXPORT_SYMBOL(tcf_block_get_ext);
1352
1353 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1354 {
1355         struct tcf_proto __rcu **p_filter_chain = priv;
1356
1357         rcu_assign_pointer(*p_filter_chain, tp_head);
1358 }
1359
1360 int tcf_block_get(struct tcf_block **p_block,
1361                   struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1362                   struct netlink_ext_ack *extack)
1363 {
1364         struct tcf_block_ext_info ei = {
1365                 .chain_head_change = tcf_chain_head_change_dflt,
1366                 .chain_head_change_priv = p_filter_chain,
1367         };
1368
1369         WARN_ON(!p_filter_chain);
1370         return tcf_block_get_ext(p_block, q, &ei, extack);
1371 }
1372 EXPORT_SYMBOL(tcf_block_get);
1373
1374 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1375  * actions should be all removed after flushing.
1376  */
1377 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1378                        struct tcf_block_ext_info *ei)
1379 {
1380         if (!block)
1381                 return;
1382         tcf_chain0_head_change_cb_del(block, ei);
1383         tcf_block_owner_del(block, q, ei->binder_type);
1384
1385         __tcf_block_put(block, q, ei, true);
1386 }
1387 EXPORT_SYMBOL(tcf_block_put_ext);
1388
1389 void tcf_block_put(struct tcf_block *block)
1390 {
1391         struct tcf_block_ext_info ei = {0, };
1392
1393         if (!block)
1394                 return;
1395         tcf_block_put_ext(block, block->q, &ei);
1396 }
1397
1398 EXPORT_SYMBOL(tcf_block_put);
1399
1400 static int
1401 tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1402                             void *cb_priv, bool add, bool offload_in_use,
1403                             struct netlink_ext_ack *extack)
1404 {
1405         struct tcf_chain *chain, *chain_prev;
1406         struct tcf_proto *tp, *tp_prev;
1407         int err;
1408
1409         lockdep_assert_held(&block->cb_lock);
1410
1411         for (chain = __tcf_get_next_chain(block, NULL);
1412              chain;
1413              chain_prev = chain,
1414                      chain = __tcf_get_next_chain(block, chain),
1415                      tcf_chain_put(chain_prev)) {
1416                 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1417                      tp_prev = tp,
1418                              tp = __tcf_get_next_proto(chain, tp),
1419                              tcf_proto_put(tp_prev, true, NULL)) {
1420                         if (tp->ops->reoffload) {
1421                                 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1422                                                          extack);
1423                                 if (err && add)
1424                                         goto err_playback_remove;
1425                         } else if (add && offload_in_use) {
1426                                 err = -EOPNOTSUPP;
1427                                 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1428                                 goto err_playback_remove;
1429                         }
1430                 }
1431         }
1432
1433         return 0;
1434
1435 err_playback_remove:
1436         tcf_proto_put(tp, true, NULL);
1437         tcf_chain_put(chain);
1438         tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1439                                     extack);
1440         return err;
1441 }
1442
1443 static int tcf_block_bind(struct tcf_block *block,
1444                           struct flow_block_offload *bo)
1445 {
1446         struct flow_block_cb *block_cb, *next;
1447         int err, i = 0;
1448
1449         lockdep_assert_held(&block->cb_lock);
1450
1451         list_for_each_entry(block_cb, &bo->cb_list, list) {
1452                 err = tcf_block_playback_offloads(block, block_cb->cb,
1453                                                   block_cb->cb_priv, true,
1454                                                   tcf_block_offload_in_use(block),
1455                                                   bo->extack);
1456                 if (err)
1457                         goto err_unroll;
1458                 if (!bo->unlocked_driver_cb)
1459                         block->lockeddevcnt++;
1460
1461                 i++;
1462         }
1463         list_splice(&bo->cb_list, &block->flow_block.cb_list);
1464
1465         return 0;
1466
1467 err_unroll:
1468         list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1469                 if (i-- > 0) {
1470                         list_del(&block_cb->list);
1471                         tcf_block_playback_offloads(block, block_cb->cb,
1472                                                     block_cb->cb_priv, false,
1473                                                     tcf_block_offload_in_use(block),
1474                                                     NULL);
1475                         if (!bo->unlocked_driver_cb)
1476                                 block->lockeddevcnt--;
1477                 }
1478                 flow_block_cb_free(block_cb);
1479         }
1480
1481         return err;
1482 }
1483
1484 static void tcf_block_unbind(struct tcf_block *block,
1485                              struct flow_block_offload *bo)
1486 {
1487         struct flow_block_cb *block_cb, *next;
1488
1489         lockdep_assert_held(&block->cb_lock);
1490
1491         list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1492                 tcf_block_playback_offloads(block, block_cb->cb,
1493                                             block_cb->cb_priv, false,
1494                                             tcf_block_offload_in_use(block),
1495                                             NULL);
1496                 list_del(&block_cb->list);
1497                 flow_block_cb_free(block_cb);
1498                 if (!bo->unlocked_driver_cb)
1499                         block->lockeddevcnt--;
1500         }
1501 }
1502
1503 static int tcf_block_setup(struct tcf_block *block,
1504                            struct flow_block_offload *bo)
1505 {
1506         int err;
1507
1508         switch (bo->command) {
1509         case FLOW_BLOCK_BIND:
1510                 err = tcf_block_bind(block, bo);
1511                 break;
1512         case FLOW_BLOCK_UNBIND:
1513                 err = 0;
1514                 tcf_block_unbind(block, bo);
1515                 break;
1516         default:
1517                 WARN_ON_ONCE(1);
1518                 err = -EOPNOTSUPP;
1519         }
1520
1521         return err;
1522 }
1523
1524 /* Main classifier routine: scans classifier chain attached
1525  * to this qdisc, (optionally) tests for protocol and asks
1526  * specific classifiers.
1527  */
1528 static inline int __tcf_classify(struct sk_buff *skb,
1529                                  const struct tcf_proto *tp,
1530                                  const struct tcf_proto *orig_tp,
1531                                  struct tcf_result *res,
1532                                  bool compat_mode,
1533                                  u32 *last_executed_chain)
1534 {
1535 #ifdef CONFIG_NET_CLS_ACT
1536         const int max_reclassify_loop = 4;
1537         const struct tcf_proto *first_tp;
1538         int limit = 0;
1539
1540 reclassify:
1541 #endif
1542         for (; tp; tp = rcu_dereference_bh(tp->next)) {
1543                 __be16 protocol = skb_protocol(skb, false);
1544                 int err;
1545
1546                 if (tp->protocol != protocol &&
1547                     tp->protocol != htons(ETH_P_ALL))
1548                         continue;
1549
1550                 err = tp->classify(skb, tp, res);
1551 #ifdef CONFIG_NET_CLS_ACT
1552                 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1553                         first_tp = orig_tp;
1554                         *last_executed_chain = first_tp->chain->index;
1555                         goto reset;
1556                 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1557                         first_tp = res->goto_tp;
1558                         *last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1559                         goto reset;
1560                 }
1561 #endif
1562                 if (err >= 0)
1563                         return err;
1564         }
1565
1566         return TC_ACT_UNSPEC; /* signal: continue lookup */
1567 #ifdef CONFIG_NET_CLS_ACT
1568 reset:
1569         if (unlikely(limit++ >= max_reclassify_loop)) {
1570                 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1571                                        tp->chain->block->index,
1572                                        tp->prio & 0xffff,
1573                                        ntohs(tp->protocol));
1574                 return TC_ACT_SHOT;
1575         }
1576
1577         tp = first_tp;
1578         goto reclassify;
1579 #endif
1580 }
1581
1582 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1583                  struct tcf_result *res, bool compat_mode)
1584 {
1585         u32 last_executed_chain = 0;
1586
1587         return __tcf_classify(skb, tp, tp, res, compat_mode,
1588                               &last_executed_chain);
1589 }
1590 EXPORT_SYMBOL(tcf_classify);
1591
1592 int tcf_classify_ingress(struct sk_buff *skb,
1593                          const struct tcf_block *ingress_block,
1594                          const struct tcf_proto *tp,
1595                          struct tcf_result *res, bool compat_mode)
1596 {
1597 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1598         u32 last_executed_chain = 0;
1599
1600         return __tcf_classify(skb, tp, tp, res, compat_mode,
1601                               &last_executed_chain);
1602 #else
1603         u32 last_executed_chain = tp ? tp->chain->index : 0;
1604         const struct tcf_proto *orig_tp = tp;
1605         struct tc_skb_ext *ext;
1606         int ret;
1607
1608         ext = skb_ext_find(skb, TC_SKB_EXT);
1609
1610         if (ext && ext->chain) {
1611                 struct tcf_chain *fchain;
1612
1613                 fchain = tcf_chain_lookup_rcu(ingress_block, ext->chain);
1614                 if (!fchain)
1615                         return TC_ACT_SHOT;
1616
1617                 /* Consume, so cloned/redirect skbs won't inherit ext */
1618                 skb_ext_del(skb, TC_SKB_EXT);
1619
1620                 tp = rcu_dereference_bh(fchain->filter_chain);
1621                 last_executed_chain = fchain->index;
1622         }
1623
1624         ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode,
1625                              &last_executed_chain);
1626
1627         /* If we missed on some chain */
1628         if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1629                 ext = skb_ext_add(skb, TC_SKB_EXT);
1630                 if (WARN_ON_ONCE(!ext))
1631                         return TC_ACT_SHOT;
1632                 ext->chain = last_executed_chain;
1633         }
1634
1635         return ret;
1636 #endif
1637 }
1638 EXPORT_SYMBOL(tcf_classify_ingress);
1639
1640 struct tcf_chain_info {
1641         struct tcf_proto __rcu **pprev;
1642         struct tcf_proto __rcu *next;
1643 };
1644
1645 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1646                                            struct tcf_chain_info *chain_info)
1647 {
1648         return tcf_chain_dereference(*chain_info->pprev, chain);
1649 }
1650
1651 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1652                                struct tcf_chain_info *chain_info,
1653                                struct tcf_proto *tp)
1654 {
1655         if (chain->flushing)
1656                 return -EAGAIN;
1657
1658         if (*chain_info->pprev == chain->filter_chain)
1659                 tcf_chain0_head_change(chain, tp);
1660         tcf_proto_get(tp);
1661         RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1662         rcu_assign_pointer(*chain_info->pprev, tp);
1663
1664         return 0;
1665 }
1666
1667 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1668                                 struct tcf_chain_info *chain_info,
1669                                 struct tcf_proto *tp)
1670 {
1671         struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1672
1673         tcf_proto_mark_delete(tp);
1674         if (tp == chain->filter_chain)
1675                 tcf_chain0_head_change(chain, next);
1676         RCU_INIT_POINTER(*chain_info->pprev, next);
1677 }
1678
1679 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1680                                            struct tcf_chain_info *chain_info,
1681                                            u32 protocol, u32 prio,
1682                                            bool prio_allocate);
1683
1684 /* Try to insert new proto.
1685  * If proto with specified priority already exists, free new proto
1686  * and return existing one.
1687  */
1688
1689 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1690                                                     struct tcf_proto *tp_new,
1691                                                     u32 protocol, u32 prio,
1692                                                     bool rtnl_held)
1693 {
1694         struct tcf_chain_info chain_info;
1695         struct tcf_proto *tp;
1696         int err = 0;
1697
1698         mutex_lock(&chain->filter_chain_lock);
1699
1700         if (tcf_proto_exists_destroying(chain, tp_new)) {
1701                 mutex_unlock(&chain->filter_chain_lock);
1702                 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1703                 return ERR_PTR(-EAGAIN);
1704         }
1705
1706         tp = tcf_chain_tp_find(chain, &chain_info,
1707                                protocol, prio, false);
1708         if (!tp)
1709                 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1710         mutex_unlock(&chain->filter_chain_lock);
1711
1712         if (tp) {
1713                 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1714                 tp_new = tp;
1715         } else if (err) {
1716                 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1717                 tp_new = ERR_PTR(err);
1718         }
1719
1720         return tp_new;
1721 }
1722
1723 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1724                                       struct tcf_proto *tp, bool rtnl_held,
1725                                       struct netlink_ext_ack *extack)
1726 {
1727         struct tcf_chain_info chain_info;
1728         struct tcf_proto *tp_iter;
1729         struct tcf_proto **pprev;
1730         struct tcf_proto *next;
1731
1732         mutex_lock(&chain->filter_chain_lock);
1733
1734         /* Atomically find and remove tp from chain. */
1735         for (pprev = &chain->filter_chain;
1736              (tp_iter = tcf_chain_dereference(*pprev, chain));
1737              pprev = &tp_iter->next) {
1738                 if (tp_iter == tp) {
1739                         chain_info.pprev = pprev;
1740                         chain_info.next = tp_iter->next;
1741                         WARN_ON(tp_iter->deleting);
1742                         break;
1743                 }
1744         }
1745         /* Verify that tp still exists and no new filters were inserted
1746          * concurrently.
1747          * Mark tp for deletion if it is empty.
1748          */
1749         if (!tp_iter || !tcf_proto_check_delete(tp)) {
1750                 mutex_unlock(&chain->filter_chain_lock);
1751                 return;
1752         }
1753
1754         tcf_proto_signal_destroying(chain, tp);
1755         next = tcf_chain_dereference(chain_info.next, chain);
1756         if (tp == chain->filter_chain)
1757                 tcf_chain0_head_change(chain, next);
1758         RCU_INIT_POINTER(*chain_info.pprev, next);
1759         mutex_unlock(&chain->filter_chain_lock);
1760
1761         tcf_proto_put(tp, rtnl_held, extack);
1762 }
1763
1764 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1765                                            struct tcf_chain_info *chain_info,
1766                                            u32 protocol, u32 prio,
1767                                            bool prio_allocate)
1768 {
1769         struct tcf_proto **pprev;
1770         struct tcf_proto *tp;
1771
1772         /* Check the chain for existence of proto-tcf with this priority */
1773         for (pprev = &chain->filter_chain;
1774              (tp = tcf_chain_dereference(*pprev, chain));
1775              pprev = &tp->next) {
1776                 if (tp->prio >= prio) {
1777                         if (tp->prio == prio) {
1778                                 if (prio_allocate ||
1779                                     (tp->protocol != protocol && protocol))
1780                                         return ERR_PTR(-EINVAL);
1781                         } else {
1782                                 tp = NULL;
1783                         }
1784                         break;
1785                 }
1786         }
1787         chain_info->pprev = pprev;
1788         if (tp) {
1789                 chain_info->next = tp->next;
1790                 tcf_proto_get(tp);
1791         } else {
1792                 chain_info->next = NULL;
1793         }
1794         return tp;
1795 }
1796
1797 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1798                          struct tcf_proto *tp, struct tcf_block *block,
1799                          struct Qdisc *q, u32 parent, void *fh,
1800                          u32 portid, u32 seq, u16 flags, int event,
1801                          bool terse_dump, bool rtnl_held)
1802 {
1803         struct tcmsg *tcm;
1804         struct nlmsghdr  *nlh;
1805         unsigned char *b = skb_tail_pointer(skb);
1806
1807         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1808         if (!nlh)
1809                 goto out_nlmsg_trim;
1810         tcm = nlmsg_data(nlh);
1811         tcm->tcm_family = AF_UNSPEC;
1812         tcm->tcm__pad1 = 0;
1813         tcm->tcm__pad2 = 0;
1814         if (q) {
1815                 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1816                 tcm->tcm_parent = parent;
1817         } else {
1818                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1819                 tcm->tcm_block_index = block->index;
1820         }
1821         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1822         if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1823                 goto nla_put_failure;
1824         if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1825                 goto nla_put_failure;
1826         if (!fh) {
1827                 tcm->tcm_handle = 0;
1828         } else if (terse_dump) {
1829                 if (tp->ops->terse_dump) {
1830                         if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
1831                                                 rtnl_held) < 0)
1832                                 goto nla_put_failure;
1833                 } else {
1834                         goto cls_op_not_supp;
1835                 }
1836         } else {
1837                 if (tp->ops->dump &&
1838                     tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
1839                         goto nla_put_failure;
1840         }
1841         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1842         return skb->len;
1843
1844 out_nlmsg_trim:
1845 nla_put_failure:
1846 cls_op_not_supp:
1847         nlmsg_trim(skb, b);
1848         return -1;
1849 }
1850
1851 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1852                           struct nlmsghdr *n, struct tcf_proto *tp,
1853                           struct tcf_block *block, struct Qdisc *q,
1854                           u32 parent, void *fh, int event, bool unicast,
1855                           bool rtnl_held)
1856 {
1857         struct sk_buff *skb;
1858         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1859         int err = 0;
1860
1861         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1862         if (!skb)
1863                 return -ENOBUFS;
1864
1865         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1866                           n->nlmsg_seq, n->nlmsg_flags, event,
1867                           false, rtnl_held) <= 0) {
1868                 kfree_skb(skb);
1869                 return -EINVAL;
1870         }
1871
1872         if (unicast)
1873                 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1874         else
1875                 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1876                                      n->nlmsg_flags & NLM_F_ECHO);
1877
1878         if (err > 0)
1879                 err = 0;
1880         return err;
1881 }
1882
1883 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1884                               struct nlmsghdr *n, struct tcf_proto *tp,
1885                               struct tcf_block *block, struct Qdisc *q,
1886                               u32 parent, void *fh, bool unicast, bool *last,
1887                               bool rtnl_held, struct netlink_ext_ack *extack)
1888 {
1889         struct sk_buff *skb;
1890         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1891         int err;
1892
1893         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1894         if (!skb)
1895                 return -ENOBUFS;
1896
1897         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1898                           n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
1899                           false, rtnl_held) <= 0) {
1900                 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1901                 kfree_skb(skb);
1902                 return -EINVAL;
1903         }
1904
1905         err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
1906         if (err) {
1907                 kfree_skb(skb);
1908                 return err;
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         if (err < 0)
1917                 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1918
1919         if (err > 0)
1920                 err = 0;
1921         return err;
1922 }
1923
1924 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1925                                  struct tcf_block *block, struct Qdisc *q,
1926                                  u32 parent, struct nlmsghdr *n,
1927                                  struct tcf_chain *chain, int event,
1928                                  bool rtnl_held)
1929 {
1930         struct tcf_proto *tp;
1931
1932         for (tp = tcf_get_next_proto(chain, NULL, rtnl_held);
1933              tp; tp = tcf_get_next_proto(chain, tp, rtnl_held))
1934                 tfilter_notify(net, oskb, n, tp, block,
1935                                q, parent, NULL, event, false, rtnl_held);
1936 }
1937
1938 static void tfilter_put(struct tcf_proto *tp, void *fh)
1939 {
1940         if (tp->ops->put && fh)
1941                 tp->ops->put(tp, fh);
1942 }
1943
1944 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1945                           struct netlink_ext_ack *extack)
1946 {
1947         struct net *net = sock_net(skb->sk);
1948         struct nlattr *tca[TCA_MAX + 1];
1949         char name[IFNAMSIZ];
1950         struct tcmsg *t;
1951         u32 protocol;
1952         u32 prio;
1953         bool prio_allocate;
1954         u32 parent;
1955         u32 chain_index;
1956         struct Qdisc *q = NULL;
1957         struct tcf_chain_info chain_info;
1958         struct tcf_chain *chain = NULL;
1959         struct tcf_block *block;
1960         struct tcf_proto *tp;
1961         unsigned long cl;
1962         void *fh;
1963         int err;
1964         int tp_created;
1965         bool rtnl_held = false;
1966
1967         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1968                 return -EPERM;
1969
1970 replay:
1971         tp_created = 0;
1972
1973         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
1974                                      rtm_tca_policy, extack);
1975         if (err < 0)
1976                 return err;
1977
1978         t = nlmsg_data(n);
1979         protocol = TC_H_MIN(t->tcm_info);
1980         prio = TC_H_MAJ(t->tcm_info);
1981         prio_allocate = false;
1982         parent = t->tcm_parent;
1983         tp = NULL;
1984         cl = 0;
1985         block = NULL;
1986
1987         if (prio == 0) {
1988                 /* If no priority is provided by the user,
1989                  * we allocate one.
1990                  */
1991                 if (n->nlmsg_flags & NLM_F_CREATE) {
1992                         prio = TC_H_MAKE(0x80000000U, 0U);
1993                         prio_allocate = true;
1994                 } else {
1995                         NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1996                         return -ENOENT;
1997                 }
1998         }
1999
2000         /* Find head of filter chain. */
2001
2002         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2003         if (err)
2004                 return err;
2005
2006         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2007                 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2008                 err = -EINVAL;
2009                 goto errout;
2010         }
2011
2012         /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2013          * block is shared (no qdisc found), qdisc is not unlocked, classifier
2014          * type is not specified, classifier is not unlocked.
2015          */
2016         if (rtnl_held ||
2017             (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2018             !tcf_proto_is_unlocked(name)) {
2019                 rtnl_held = true;
2020                 rtnl_lock();
2021         }
2022
2023         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2024         if (err)
2025                 goto errout;
2026
2027         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2028                                  extack);
2029         if (IS_ERR(block)) {
2030                 err = PTR_ERR(block);
2031                 goto errout;
2032         }
2033         block->classid = parent;
2034
2035         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2036         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2037                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2038                 err = -EINVAL;
2039                 goto errout;
2040         }
2041         chain = tcf_chain_get(block, chain_index, true);
2042         if (!chain) {
2043                 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2044                 err = -ENOMEM;
2045                 goto errout;
2046         }
2047
2048         mutex_lock(&chain->filter_chain_lock);
2049         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2050                                prio, prio_allocate);
2051         if (IS_ERR(tp)) {
2052                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2053                 err = PTR_ERR(tp);
2054                 goto errout_locked;
2055         }
2056
2057         if (tp == NULL) {
2058                 struct tcf_proto *tp_new = NULL;
2059
2060                 if (chain->flushing) {
2061                         err = -EAGAIN;
2062                         goto errout_locked;
2063                 }
2064
2065                 /* Proto-tcf does not exist, create new one */
2066
2067                 if (tca[TCA_KIND] == NULL || !protocol) {
2068                         NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2069                         err = -EINVAL;
2070                         goto errout_locked;
2071                 }
2072
2073                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2074                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2075                         err = -ENOENT;
2076                         goto errout_locked;
2077                 }
2078
2079                 if (prio_allocate)
2080                         prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2081                                                                &chain_info));
2082
2083                 mutex_unlock(&chain->filter_chain_lock);
2084                 tp_new = tcf_proto_create(name, protocol, prio, chain,
2085                                           rtnl_held, extack);
2086                 if (IS_ERR(tp_new)) {
2087                         err = PTR_ERR(tp_new);
2088                         goto errout_tp;
2089                 }
2090
2091                 tp_created = 1;
2092                 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2093                                                 rtnl_held);
2094                 if (IS_ERR(tp)) {
2095                         err = PTR_ERR(tp);
2096                         goto errout_tp;
2097                 }
2098         } else {
2099                 mutex_unlock(&chain->filter_chain_lock);
2100         }
2101
2102         if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2103                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2104                 err = -EINVAL;
2105                 goto errout;
2106         }
2107
2108         fh = tp->ops->get(tp, t->tcm_handle);
2109
2110         if (!fh) {
2111                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2112                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2113                         err = -ENOENT;
2114                         goto errout;
2115                 }
2116         } else if (n->nlmsg_flags & NLM_F_EXCL) {
2117                 tfilter_put(tp, fh);
2118                 NL_SET_ERR_MSG(extack, "Filter already exists");
2119                 err = -EEXIST;
2120                 goto errout;
2121         }
2122
2123         if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2124                 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2125                 err = -EINVAL;
2126                 goto errout;
2127         }
2128
2129         err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2130                               n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
2131                               rtnl_held, extack);
2132         if (err == 0) {
2133                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2134                                RTM_NEWTFILTER, false, rtnl_held);
2135                 tfilter_put(tp, fh);
2136                 /* q pointer is NULL for shared blocks */
2137                 if (q)
2138                         q->flags &= ~TCQ_F_CAN_BYPASS;
2139         }
2140
2141 errout:
2142         if (err && tp_created)
2143                 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2144 errout_tp:
2145         if (chain) {
2146                 if (tp && !IS_ERR(tp))
2147                         tcf_proto_put(tp, rtnl_held, NULL);
2148                 if (!tp_created)
2149                         tcf_chain_put(chain);
2150         }
2151         tcf_block_release(q, block, rtnl_held);
2152
2153         if (rtnl_held)
2154                 rtnl_unlock();
2155
2156         if (err == -EAGAIN) {
2157                 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2158                  * of target chain.
2159                  */
2160                 rtnl_held = true;
2161                 /* Replay the request. */
2162                 goto replay;
2163         }
2164         return err;
2165
2166 errout_locked:
2167         mutex_unlock(&chain->filter_chain_lock);
2168         goto errout;
2169 }
2170
2171 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2172                           struct netlink_ext_ack *extack)
2173 {
2174         struct net *net = sock_net(skb->sk);
2175         struct nlattr *tca[TCA_MAX + 1];
2176         char name[IFNAMSIZ];
2177         struct tcmsg *t;
2178         u32 protocol;
2179         u32 prio;
2180         u32 parent;
2181         u32 chain_index;
2182         struct Qdisc *q = NULL;
2183         struct tcf_chain_info chain_info;
2184         struct tcf_chain *chain = NULL;
2185         struct tcf_block *block = NULL;
2186         struct tcf_proto *tp = NULL;
2187         unsigned long cl = 0;
2188         void *fh = NULL;
2189         int err;
2190         bool rtnl_held = false;
2191
2192         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2193                 return -EPERM;
2194
2195         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2196                                      rtm_tca_policy, extack);
2197         if (err < 0)
2198                 return err;
2199
2200         t = nlmsg_data(n);
2201         protocol = TC_H_MIN(t->tcm_info);
2202         prio = TC_H_MAJ(t->tcm_info);
2203         parent = t->tcm_parent;
2204
2205         if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2206                 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2207                 return -ENOENT;
2208         }
2209
2210         /* Find head of filter chain. */
2211
2212         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2213         if (err)
2214                 return err;
2215
2216         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2217                 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2218                 err = -EINVAL;
2219                 goto errout;
2220         }
2221         /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2222          * found), qdisc is not unlocked, classifier type is not specified,
2223          * classifier is not unlocked.
2224          */
2225         if (!prio ||
2226             (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2227             !tcf_proto_is_unlocked(name)) {
2228                 rtnl_held = true;
2229                 rtnl_lock();
2230         }
2231
2232         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2233         if (err)
2234                 goto errout;
2235
2236         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2237                                  extack);
2238         if (IS_ERR(block)) {
2239                 err = PTR_ERR(block);
2240                 goto errout;
2241         }
2242
2243         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2244         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2245                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2246                 err = -EINVAL;
2247                 goto errout;
2248         }
2249         chain = tcf_chain_get(block, chain_index, false);
2250         if (!chain) {
2251                 /* User requested flush on non-existent chain. Nothing to do,
2252                  * so just return success.
2253                  */
2254                 if (prio == 0) {
2255                         err = 0;
2256                         goto errout;
2257                 }
2258                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2259                 err = -ENOENT;
2260                 goto errout;
2261         }
2262
2263         if (prio == 0) {
2264                 tfilter_notify_chain(net, skb, block, q, parent, n,
2265                                      chain, RTM_DELTFILTER, rtnl_held);
2266                 tcf_chain_flush(chain, rtnl_held);
2267                 err = 0;
2268                 goto errout;
2269         }
2270
2271         mutex_lock(&chain->filter_chain_lock);
2272         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2273                                prio, false);
2274         if (!tp || IS_ERR(tp)) {
2275                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2276                 err = tp ? PTR_ERR(tp) : -ENOENT;
2277                 goto errout_locked;
2278         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2279                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2280                 err = -EINVAL;
2281                 goto errout_locked;
2282         } else if (t->tcm_handle == 0) {
2283                 tcf_proto_signal_destroying(chain, tp);
2284                 tcf_chain_tp_remove(chain, &chain_info, tp);
2285                 mutex_unlock(&chain->filter_chain_lock);
2286
2287                 tcf_proto_put(tp, rtnl_held, NULL);
2288                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2289                                RTM_DELTFILTER, false, rtnl_held);
2290                 err = 0;
2291                 goto errout;
2292         }
2293         mutex_unlock(&chain->filter_chain_lock);
2294
2295         fh = tp->ops->get(tp, t->tcm_handle);
2296
2297         if (!fh) {
2298                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2299                 err = -ENOENT;
2300         } else {
2301                 bool last;
2302
2303                 err = tfilter_del_notify(net, skb, n, tp, block,
2304                                          q, parent, fh, false, &last,
2305                                          rtnl_held, extack);
2306
2307                 if (err)
2308                         goto errout;
2309                 if (last)
2310                         tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2311         }
2312
2313 errout:
2314         if (chain) {
2315                 if (tp && !IS_ERR(tp))
2316                         tcf_proto_put(tp, rtnl_held, NULL);
2317                 tcf_chain_put(chain);
2318         }
2319         tcf_block_release(q, block, rtnl_held);
2320
2321         if (rtnl_held)
2322                 rtnl_unlock();
2323
2324         return err;
2325
2326 errout_locked:
2327         mutex_unlock(&chain->filter_chain_lock);
2328         goto errout;
2329 }
2330
2331 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2332                           struct netlink_ext_ack *extack)
2333 {
2334         struct net *net = sock_net(skb->sk);
2335         struct nlattr *tca[TCA_MAX + 1];
2336         char name[IFNAMSIZ];
2337         struct tcmsg *t;
2338         u32 protocol;
2339         u32 prio;
2340         u32 parent;
2341         u32 chain_index;
2342         struct Qdisc *q = NULL;
2343         struct tcf_chain_info chain_info;
2344         struct tcf_chain *chain = NULL;
2345         struct tcf_block *block = NULL;
2346         struct tcf_proto *tp = NULL;
2347         unsigned long cl = 0;
2348         void *fh = NULL;
2349         int err;
2350         bool rtnl_held = false;
2351
2352         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2353                                      rtm_tca_policy, extack);
2354         if (err < 0)
2355                 return err;
2356
2357         t = nlmsg_data(n);
2358         protocol = TC_H_MIN(t->tcm_info);
2359         prio = TC_H_MAJ(t->tcm_info);
2360         parent = t->tcm_parent;
2361
2362         if (prio == 0) {
2363                 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2364                 return -ENOENT;
2365         }
2366
2367         /* Find head of filter chain. */
2368
2369         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2370         if (err)
2371                 return err;
2372
2373         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2374                 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2375                 err = -EINVAL;
2376                 goto errout;
2377         }
2378         /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2379          * unlocked, classifier type is not specified, classifier is not
2380          * unlocked.
2381          */
2382         if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2383             !tcf_proto_is_unlocked(name)) {
2384                 rtnl_held = true;
2385                 rtnl_lock();
2386         }
2387
2388         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2389         if (err)
2390                 goto errout;
2391
2392         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2393                                  extack);
2394         if (IS_ERR(block)) {
2395                 err = PTR_ERR(block);
2396                 goto errout;
2397         }
2398
2399         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2400         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2401                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2402                 err = -EINVAL;
2403                 goto errout;
2404         }
2405         chain = tcf_chain_get(block, chain_index, false);
2406         if (!chain) {
2407                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2408                 err = -EINVAL;
2409                 goto errout;
2410         }
2411
2412         mutex_lock(&chain->filter_chain_lock);
2413         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2414                                prio, false);
2415         mutex_unlock(&chain->filter_chain_lock);
2416         if (!tp || IS_ERR(tp)) {
2417                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2418                 err = tp ? PTR_ERR(tp) : -ENOENT;
2419                 goto errout;
2420         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2421                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2422                 err = -EINVAL;
2423                 goto errout;
2424         }
2425
2426         fh = tp->ops->get(tp, t->tcm_handle);
2427
2428         if (!fh) {
2429                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2430                 err = -ENOENT;
2431         } else {
2432                 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2433                                      fh, RTM_NEWTFILTER, true, rtnl_held);
2434                 if (err < 0)
2435                         NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2436         }
2437
2438         tfilter_put(tp, fh);
2439 errout:
2440         if (chain) {
2441                 if (tp && !IS_ERR(tp))
2442                         tcf_proto_put(tp, rtnl_held, NULL);
2443                 tcf_chain_put(chain);
2444         }
2445         tcf_block_release(q, block, rtnl_held);
2446
2447         if (rtnl_held)
2448                 rtnl_unlock();
2449
2450         return err;
2451 }
2452
2453 struct tcf_dump_args {
2454         struct tcf_walker w;
2455         struct sk_buff *skb;
2456         struct netlink_callback *cb;
2457         struct tcf_block *block;
2458         struct Qdisc *q;
2459         u32 parent;
2460         bool terse_dump;
2461 };
2462
2463 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2464 {
2465         struct tcf_dump_args *a = (void *)arg;
2466         struct net *net = sock_net(a->skb->sk);
2467
2468         return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2469                              n, NETLINK_CB(a->cb->skb).portid,
2470                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2471                              RTM_NEWTFILTER, a->terse_dump, true);
2472 }
2473
2474 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2475                            struct sk_buff *skb, struct netlink_callback *cb,
2476                            long index_start, long *p_index, bool terse)
2477 {
2478         struct net *net = sock_net(skb->sk);
2479         struct tcf_block *block = chain->block;
2480         struct tcmsg *tcm = nlmsg_data(cb->nlh);
2481         struct tcf_proto *tp, *tp_prev;
2482         struct tcf_dump_args arg;
2483
2484         for (tp = __tcf_get_next_proto(chain, NULL);
2485              tp;
2486              tp_prev = tp,
2487                      tp = __tcf_get_next_proto(chain, tp),
2488                      tcf_proto_put(tp_prev, true, NULL),
2489                      (*p_index)++) {
2490                 if (*p_index < index_start)
2491                         continue;
2492                 if (TC_H_MAJ(tcm->tcm_info) &&
2493                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
2494                         continue;
2495                 if (TC_H_MIN(tcm->tcm_info) &&
2496                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
2497                         continue;
2498                 if (*p_index > index_start)
2499                         memset(&cb->args[1], 0,
2500                                sizeof(cb->args) - sizeof(cb->args[0]));
2501                 if (cb->args[1] == 0) {
2502                         if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2503                                           NETLINK_CB(cb->skb).portid,
2504                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
2505                                           RTM_NEWTFILTER, false, true) <= 0)
2506                                 goto errout;
2507                         cb->args[1] = 1;
2508                 }
2509                 if (!tp->ops->walk)
2510                         continue;
2511                 arg.w.fn = tcf_node_dump;
2512                 arg.skb = skb;
2513                 arg.cb = cb;
2514                 arg.block = block;
2515                 arg.q = q;
2516                 arg.parent = parent;
2517                 arg.w.stop = 0;
2518                 arg.w.skip = cb->args[1] - 1;
2519                 arg.w.count = 0;
2520                 arg.w.cookie = cb->args[2];
2521                 arg.terse_dump = terse;
2522                 tp->ops->walk(tp, &arg.w, true);
2523                 cb->args[2] = arg.w.cookie;
2524                 cb->args[1] = arg.w.count + 1;
2525                 if (arg.w.stop)
2526                         goto errout;
2527         }
2528         return true;
2529
2530 errout:
2531         tcf_proto_put(tp, true, NULL);
2532         return false;
2533 }
2534
2535 static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2536         [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2537 };
2538
2539 /* called with RTNL */
2540 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2541 {
2542         struct tcf_chain *chain, *chain_prev;
2543         struct net *net = sock_net(skb->sk);
2544         struct nlattr *tca[TCA_MAX + 1];
2545         struct Qdisc *q = NULL;
2546         struct tcf_block *block;
2547         struct tcmsg *tcm = nlmsg_data(cb->nlh);
2548         bool terse_dump = false;
2549         long index_start;
2550         long index;
2551         u32 parent;
2552         int err;
2553
2554         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2555                 return skb->len;
2556
2557         err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2558                                      tcf_tfilter_dump_policy, cb->extack);
2559         if (err)
2560                 return err;
2561
2562         if (tca[TCA_DUMP_FLAGS]) {
2563                 struct nla_bitfield32 flags =
2564                         nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2565
2566                 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2567         }
2568
2569         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2570                 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2571                 if (!block)
2572                         goto out;
2573                 /* If we work with block index, q is NULL and parent value
2574                  * will never be used in the following code. The check
2575                  * in tcf_fill_node prevents it. However, compiler does not
2576                  * see that far, so set parent to zero to silence the warning
2577                  * about parent being uninitialized.
2578                  */
2579                 parent = 0;
2580         } else {
2581                 const struct Qdisc_class_ops *cops;
2582                 struct net_device *dev;
2583                 unsigned long cl = 0;
2584
2585                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2586                 if (!dev)
2587                         return skb->len;
2588
2589                 parent = tcm->tcm_parent;
2590                 if (!parent)
2591                         q = dev->qdisc;
2592                 else
2593                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2594                 if (!q)
2595                         goto out;
2596                 cops = q->ops->cl_ops;
2597                 if (!cops)
2598                         goto out;
2599                 if (!cops->tcf_block)
2600                         goto out;
2601                 if (TC_H_MIN(tcm->tcm_parent)) {
2602                         cl = cops->find(q, tcm->tcm_parent);
2603                         if (cl == 0)
2604                                 goto out;
2605                 }
2606                 block = cops->tcf_block(q, cl, NULL);
2607                 if (!block)
2608                         goto out;
2609                 parent = block->classid;
2610                 if (tcf_block_shared(block))
2611                         q = NULL;
2612         }
2613
2614         index_start = cb->args[0];
2615         index = 0;
2616
2617         for (chain = __tcf_get_next_chain(block, NULL);
2618              chain;
2619              chain_prev = chain,
2620                      chain = __tcf_get_next_chain(block, chain),
2621                      tcf_chain_put(chain_prev)) {
2622                 if (tca[TCA_CHAIN] &&
2623                     nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2624                         continue;
2625                 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2626                                     index_start, &index, terse_dump)) {
2627                         tcf_chain_put(chain);
2628                         err = -EMSGSIZE;
2629                         break;
2630                 }
2631         }
2632
2633         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2634                 tcf_block_refcnt_put(block, true);
2635         cb->args[0] = index;
2636
2637 out:
2638         /* If we did no progress, the error (EMSGSIZE) is real */
2639         if (skb->len == 0 && err)
2640                 return err;
2641         return skb->len;
2642 }
2643
2644 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2645                               void *tmplt_priv, u32 chain_index,
2646                               struct net *net, struct sk_buff *skb,
2647                               struct tcf_block *block,
2648                               u32 portid, u32 seq, u16 flags, int event)
2649 {
2650         unsigned char *b = skb_tail_pointer(skb);
2651         const struct tcf_proto_ops *ops;
2652         struct nlmsghdr *nlh;
2653         struct tcmsg *tcm;
2654         void *priv;
2655
2656         ops = tmplt_ops;
2657         priv = tmplt_priv;
2658
2659         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2660         if (!nlh)
2661                 goto out_nlmsg_trim;
2662         tcm = nlmsg_data(nlh);
2663         tcm->tcm_family = AF_UNSPEC;
2664         tcm->tcm__pad1 = 0;
2665         tcm->tcm__pad2 = 0;
2666         tcm->tcm_handle = 0;
2667         if (block->q) {
2668                 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2669                 tcm->tcm_parent = block->q->handle;
2670         } else {
2671                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2672                 tcm->tcm_block_index = block->index;
2673         }
2674
2675         if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2676                 goto nla_put_failure;
2677
2678         if (ops) {
2679                 if (nla_put_string(skb, TCA_KIND, ops->kind))
2680                         goto nla_put_failure;
2681                 if (ops->tmplt_dump(skb, net, priv) < 0)
2682                         goto nla_put_failure;
2683         }
2684
2685         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2686         return skb->len;
2687
2688 out_nlmsg_trim:
2689 nla_put_failure:
2690         nlmsg_trim(skb, b);
2691         return -EMSGSIZE;
2692 }
2693
2694 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2695                            u32 seq, u16 flags, int event, bool unicast)
2696 {
2697         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2698         struct tcf_block *block = chain->block;
2699         struct net *net = block->net;
2700         struct sk_buff *skb;
2701         int err = 0;
2702
2703         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2704         if (!skb)
2705                 return -ENOBUFS;
2706
2707         if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2708                                chain->index, net, skb, block, portid,
2709                                seq, flags, event) <= 0) {
2710                 kfree_skb(skb);
2711                 return -EINVAL;
2712         }
2713
2714         if (unicast)
2715                 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2716         else
2717                 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2718                                      flags & NLM_F_ECHO);
2719
2720         if (err > 0)
2721                 err = 0;
2722         return err;
2723 }
2724
2725 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2726                                   void *tmplt_priv, u32 chain_index,
2727                                   struct tcf_block *block, struct sk_buff *oskb,
2728                                   u32 seq, u16 flags, bool unicast)
2729 {
2730         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2731         struct net *net = block->net;
2732         struct sk_buff *skb;
2733
2734         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2735         if (!skb)
2736                 return -ENOBUFS;
2737
2738         if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2739                                block, portid, seq, flags, RTM_DELCHAIN) <= 0) {
2740                 kfree_skb(skb);
2741                 return -EINVAL;
2742         }
2743
2744         if (unicast)
2745                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2746
2747         return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2748 }
2749
2750 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2751                               struct nlattr **tca,
2752                               struct netlink_ext_ack *extack)
2753 {
2754         const struct tcf_proto_ops *ops;
2755         char name[IFNAMSIZ];
2756         void *tmplt_priv;
2757
2758         /* If kind is not set, user did not specify template. */
2759         if (!tca[TCA_KIND])
2760                 return 0;
2761
2762         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2763                 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
2764                 return -EINVAL;
2765         }
2766
2767         ops = tcf_proto_lookup_ops(name, true, extack);
2768         if (IS_ERR(ops))
2769                 return PTR_ERR(ops);
2770         if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2771                 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2772                 return -EOPNOTSUPP;
2773         }
2774
2775         tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2776         if (IS_ERR(tmplt_priv)) {
2777                 module_put(ops->owner);
2778                 return PTR_ERR(tmplt_priv);
2779         }
2780         chain->tmplt_ops = ops;
2781         chain->tmplt_priv = tmplt_priv;
2782         return 0;
2783 }
2784
2785 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2786                                void *tmplt_priv)
2787 {
2788         /* If template ops are set, no work to do for us. */
2789         if (!tmplt_ops)
2790                 return;
2791
2792         tmplt_ops->tmplt_destroy(tmplt_priv);
2793         module_put(tmplt_ops->owner);
2794 }
2795
2796 /* Add/delete/get a chain */
2797
2798 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2799                         struct netlink_ext_ack *extack)
2800 {
2801         struct net *net = sock_net(skb->sk);
2802         struct nlattr *tca[TCA_MAX + 1];
2803         struct tcmsg *t;
2804         u32 parent;
2805         u32 chain_index;
2806         struct Qdisc *q = NULL;
2807         struct tcf_chain *chain = NULL;
2808         struct tcf_block *block;
2809         unsigned long cl;
2810         int err;
2811
2812         if (n->nlmsg_type != RTM_GETCHAIN &&
2813             !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2814                 return -EPERM;
2815
2816 replay:
2817         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2818                                      rtm_tca_policy, extack);
2819         if (err < 0)
2820                 return err;
2821
2822         t = nlmsg_data(n);
2823         parent = t->tcm_parent;
2824         cl = 0;
2825
2826         block = tcf_block_find(net, &q, &parent, &cl,
2827                                t->tcm_ifindex, t->tcm_block_index, extack);
2828         if (IS_ERR(block))
2829                 return PTR_ERR(block);
2830
2831         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2832         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2833                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2834                 err = -EINVAL;
2835                 goto errout_block;
2836         }
2837
2838         mutex_lock(&block->lock);
2839         chain = tcf_chain_lookup(block, chain_index);
2840         if (n->nlmsg_type == RTM_NEWCHAIN) {
2841                 if (chain) {
2842                         if (tcf_chain_held_by_acts_only(chain)) {
2843                                 /* The chain exists only because there is
2844                                  * some action referencing it.
2845                                  */
2846                                 tcf_chain_hold(chain);
2847                         } else {
2848                                 NL_SET_ERR_MSG(extack, "Filter chain already exists");
2849                                 err = -EEXIST;
2850                                 goto errout_block_locked;
2851                         }
2852                 } else {
2853                         if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2854                                 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
2855                                 err = -ENOENT;
2856                                 goto errout_block_locked;
2857                         }
2858                         chain = tcf_chain_create(block, chain_index);
2859                         if (!chain) {
2860                                 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
2861                                 err = -ENOMEM;
2862                                 goto errout_block_locked;
2863                         }
2864                 }
2865         } else {
2866                 if (!chain || tcf_chain_held_by_acts_only(chain)) {
2867                         NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2868                         err = -EINVAL;
2869                         goto errout_block_locked;
2870                 }
2871                 tcf_chain_hold(chain);
2872         }
2873
2874         if (n->nlmsg_type == RTM_NEWCHAIN) {
2875                 /* Modifying chain requires holding parent block lock. In case
2876                  * the chain was successfully added, take a reference to the
2877                  * chain. This ensures that an empty chain does not disappear at
2878                  * the end of this function.
2879                  */
2880                 tcf_chain_hold(chain);
2881                 chain->explicitly_created = true;
2882         }
2883         mutex_unlock(&block->lock);
2884
2885         switch (n->nlmsg_type) {
2886         case RTM_NEWCHAIN:
2887                 err = tc_chain_tmplt_add(chain, net, tca, extack);
2888                 if (err) {
2889                         tcf_chain_put_explicitly_created(chain);
2890                         goto errout;
2891                 }
2892
2893                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2894                                 RTM_NEWCHAIN, false);
2895                 break;
2896         case RTM_DELCHAIN:
2897                 tfilter_notify_chain(net, skb, block, q, parent, n,
2898                                      chain, RTM_DELTFILTER, true);
2899                 /* Flush the chain first as the user requested chain removal. */
2900                 tcf_chain_flush(chain, true);
2901                 /* In case the chain was successfully deleted, put a reference
2902                  * to the chain previously taken during addition.
2903                  */
2904                 tcf_chain_put_explicitly_created(chain);
2905                 break;
2906         case RTM_GETCHAIN:
2907                 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
2908                                       n->nlmsg_seq, n->nlmsg_type, true);
2909                 if (err < 0)
2910                         NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2911                 break;
2912         default:
2913                 err = -EOPNOTSUPP;
2914                 NL_SET_ERR_MSG(extack, "Unsupported message type");
2915                 goto errout;
2916         }
2917
2918 errout:
2919         tcf_chain_put(chain);
2920 errout_block:
2921         tcf_block_release(q, block, true);
2922         if (err == -EAGAIN)
2923                 /* Replay the request. */
2924                 goto replay;
2925         return err;
2926
2927 errout_block_locked:
2928         mutex_unlock(&block->lock);
2929         goto errout_block;
2930 }
2931
2932 /* called with RTNL */
2933 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2934 {
2935         struct net *net = sock_net(skb->sk);
2936         struct nlattr *tca[TCA_MAX + 1];
2937         struct Qdisc *q = NULL;
2938         struct tcf_block *block;
2939         struct tcmsg *tcm = nlmsg_data(cb->nlh);
2940         struct tcf_chain *chain;
2941         long index_start;
2942         long index;
2943         u32 parent;
2944         int err;
2945
2946         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2947                 return skb->len;
2948
2949         err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2950                                      rtm_tca_policy, cb->extack);
2951         if (err)
2952                 return err;
2953
2954         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2955                 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2956                 if (!block)
2957                         goto out;
2958                 /* If we work with block index, q is NULL and parent value
2959                  * will never be used in the following code. The check
2960                  * in tcf_fill_node prevents it. However, compiler does not
2961                  * see that far, so set parent to zero to silence the warning
2962                  * about parent being uninitialized.
2963                  */
2964                 parent = 0;
2965         } else {
2966                 const struct Qdisc_class_ops *cops;
2967                 struct net_device *dev;
2968                 unsigned long cl = 0;
2969
2970                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2971                 if (!dev)
2972                         return skb->len;
2973
2974                 parent = tcm->tcm_parent;
2975                 if (!parent) {
2976                         q = dev->qdisc;
2977                         parent = q->handle;
2978                 } else {
2979                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2980                 }
2981                 if (!q)
2982                         goto out;
2983                 cops = q->ops->cl_ops;
2984                 if (!cops)
2985                         goto out;
2986                 if (!cops->tcf_block)
2987                         goto out;
2988                 if (TC_H_MIN(tcm->tcm_parent)) {
2989                         cl = cops->find(q, tcm->tcm_parent);
2990                         if (cl == 0)
2991                                 goto out;
2992                 }
2993                 block = cops->tcf_block(q, cl, NULL);
2994                 if (!block)
2995                         goto out;
2996                 if (tcf_block_shared(block))
2997                         q = NULL;
2998         }
2999
3000         index_start = cb->args[0];
3001         index = 0;
3002
3003         mutex_lock(&block->lock);
3004         list_for_each_entry(chain, &block->chain_list, list) {
3005                 if ((tca[TCA_CHAIN] &&
3006                      nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3007                         continue;
3008                 if (index < index_start) {
3009                         index++;
3010                         continue;
3011                 }
3012                 if (tcf_chain_held_by_acts_only(chain))
3013                         continue;
3014                 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3015                                          chain->index, net, skb, block,
3016                                          NETLINK_CB(cb->skb).portid,
3017                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
3018                                          RTM_NEWCHAIN);
3019                 if (err <= 0)
3020                         break;
3021                 index++;
3022         }
3023         mutex_unlock(&block->lock);
3024
3025         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3026                 tcf_block_refcnt_put(block, true);
3027         cb->args[0] = index;
3028
3029 out:
3030         /* If we did no progress, the error (EMSGSIZE) is real */
3031         if (skb->len == 0 && err)
3032                 return err;
3033         return skb->len;
3034 }
3035
3036 void tcf_exts_destroy(struct tcf_exts *exts)
3037 {
3038 #ifdef CONFIG_NET_CLS_ACT
3039         if (exts->actions) {
3040                 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3041                 kfree(exts->actions);
3042         }
3043         exts->nr_actions = 0;
3044 #endif
3045 }
3046 EXPORT_SYMBOL(tcf_exts_destroy);
3047
3048 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3049                       struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
3050                       bool rtnl_held, struct netlink_ext_ack *extack)
3051 {
3052 #ifdef CONFIG_NET_CLS_ACT
3053         {
3054                 struct tc_action *act;
3055                 size_t attr_size = 0;
3056
3057                 if (exts->police && tb[exts->police]) {
3058                         act = tcf_action_init_1(net, tp, tb[exts->police],
3059                                                 rate_tlv, "police", ovr,
3060                                                 TCA_ACT_BIND, rtnl_held,
3061                                                 extack);
3062                         if (IS_ERR(act))
3063                                 return PTR_ERR(act);
3064
3065                         act->type = exts->type = TCA_OLD_COMPAT;
3066                         exts->actions[0] = act;
3067                         exts->nr_actions = 1;
3068                 } else if (exts->action && tb[exts->action]) {
3069                         int err;
3070
3071                         err = tcf_action_init(net, tp, tb[exts->action],
3072                                               rate_tlv, NULL, ovr, TCA_ACT_BIND,
3073                                               exts->actions, &attr_size,
3074                                               rtnl_held, extack);
3075                         if (err < 0)
3076                                 return err;
3077                         exts->nr_actions = err;
3078                 }
3079         }
3080 #else
3081         if ((exts->action && tb[exts->action]) ||
3082             (exts->police && tb[exts->police])) {
3083                 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3084                 return -EOPNOTSUPP;
3085         }
3086 #endif
3087
3088         return 0;
3089 }
3090 EXPORT_SYMBOL(tcf_exts_validate);
3091
3092 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3093 {
3094 #ifdef CONFIG_NET_CLS_ACT
3095         struct tcf_exts old = *dst;
3096
3097         *dst = *src;
3098         tcf_exts_destroy(&old);
3099 #endif
3100 }
3101 EXPORT_SYMBOL(tcf_exts_change);
3102
3103 #ifdef CONFIG_NET_CLS_ACT
3104 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3105 {
3106         if (exts->nr_actions == 0)
3107                 return NULL;
3108         else
3109                 return exts->actions[0];
3110 }
3111 #endif
3112
3113 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3114 {
3115 #ifdef CONFIG_NET_CLS_ACT
3116         struct nlattr *nest;
3117
3118         if (exts->action && tcf_exts_has_actions(exts)) {
3119                 /*
3120                  * again for backward compatible mode - we want
3121                  * to work with both old and new modes of entering
3122                  * tc data even if iproute2  was newer - jhs
3123                  */
3124                 if (exts->type != TCA_OLD_COMPAT) {
3125                         nest = nla_nest_start_noflag(skb, exts->action);
3126                         if (nest == NULL)
3127                                 goto nla_put_failure;
3128
3129                         if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3130                             < 0)
3131                                 goto nla_put_failure;
3132                         nla_nest_end(skb, nest);
3133                 } else if (exts->police) {
3134                         struct tc_action *act = tcf_exts_first_act(exts);
3135                         nest = nla_nest_start_noflag(skb, exts->police);
3136                         if (nest == NULL || !act)
3137                                 goto nla_put_failure;
3138                         if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3139                                 goto nla_put_failure;
3140                         nla_nest_end(skb, nest);
3141                 }
3142         }
3143         return 0;
3144
3145 nla_put_failure:
3146         nla_nest_cancel(skb, nest);
3147         return -1;
3148 #else
3149         return 0;
3150 #endif
3151 }
3152 EXPORT_SYMBOL(tcf_exts_dump);
3153
3154 int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3155 {
3156 #ifdef CONFIG_NET_CLS_ACT
3157         struct nlattr *nest;
3158
3159         if (!exts->action || !tcf_exts_has_actions(exts))
3160                 return 0;
3161
3162         nest = nla_nest_start_noflag(skb, exts->action);
3163         if (!nest)
3164                 goto nla_put_failure;
3165
3166         if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3167                 goto nla_put_failure;
3168         nla_nest_end(skb, nest);
3169         return 0;
3170
3171 nla_put_failure:
3172         nla_nest_cancel(skb, nest);
3173         return -1;
3174 #else
3175         return 0;
3176 #endif
3177 }
3178 EXPORT_SYMBOL(tcf_exts_terse_dump);
3179
3180 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3181 {
3182 #ifdef CONFIG_NET_CLS_ACT
3183         struct tc_action *a = tcf_exts_first_act(exts);
3184         if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3185                 return -1;
3186 #endif
3187         return 0;
3188 }
3189 EXPORT_SYMBOL(tcf_exts_dump_stats);
3190
3191 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3192 {
3193         if (*flags & TCA_CLS_FLAGS_IN_HW)
3194                 return;
3195         *flags |= TCA_CLS_FLAGS_IN_HW;
3196         atomic_inc(&block->offloadcnt);
3197 }
3198
3199 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3200 {
3201         if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3202                 return;
3203         *flags &= ~TCA_CLS_FLAGS_IN_HW;
3204         atomic_dec(&block->offloadcnt);
3205 }
3206
3207 static void tc_cls_offload_cnt_update(struct tcf_block *block,
3208                                       struct tcf_proto *tp, u32 *cnt,
3209                                       u32 *flags, u32 diff, bool add)
3210 {
3211         lockdep_assert_held(&block->cb_lock);
3212
3213         spin_lock(&tp->lock);
3214         if (add) {
3215                 if (!*cnt)
3216                         tcf_block_offload_inc(block, flags);
3217                 *cnt += diff;
3218         } else {
3219                 *cnt -= diff;
3220                 if (!*cnt)
3221                         tcf_block_offload_dec(block, flags);
3222         }
3223         spin_unlock(&tp->lock);
3224 }
3225
3226 static void
3227 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3228                          u32 *cnt, u32 *flags)
3229 {
3230         lockdep_assert_held(&block->cb_lock);
3231
3232         spin_lock(&tp->lock);
3233         tcf_block_offload_dec(block, flags);
3234         *cnt = 0;
3235         spin_unlock(&tp->lock);
3236 }
3237
3238 static int
3239 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3240                    void *type_data, bool err_stop)
3241 {
3242         struct flow_block_cb *block_cb;
3243         int ok_count = 0;
3244         int err;
3245
3246         list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3247                 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3248                 if (err) {
3249                         if (err_stop)
3250                                 return err;
3251                 } else {
3252                         ok_count++;
3253                 }
3254         }
3255         return ok_count;
3256 }
3257
3258 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3259                      void *type_data, bool err_stop, bool rtnl_held)
3260 {
3261         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3262         int ok_count;
3263
3264 retry:
3265         if (take_rtnl)
3266                 rtnl_lock();
3267         down_read(&block->cb_lock);
3268         /* Need to obtain rtnl lock if block is bound to devs that require it.
3269          * In block bind code cb_lock is obtained while holding rtnl, so we must
3270          * obtain the locks in same order here.
3271          */
3272         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3273                 up_read(&block->cb_lock);
3274                 take_rtnl = true;
3275                 goto retry;
3276         }
3277
3278         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3279
3280         up_read(&block->cb_lock);
3281         if (take_rtnl)
3282                 rtnl_unlock();
3283         return ok_count;
3284 }
3285 EXPORT_SYMBOL(tc_setup_cb_call);
3286
3287 /* Non-destructive filter add. If filter that wasn't already in hardware is
3288  * successfully offloaded, increment block offloads counter. On failure,
3289  * previously offloaded filter is considered to be intact and offloads counter
3290  * is not decremented.
3291  */
3292
3293 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3294                     enum tc_setup_type type, void *type_data, bool err_stop,
3295                     u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3296 {
3297         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3298         int ok_count;
3299
3300 retry:
3301         if (take_rtnl)
3302                 rtnl_lock();
3303         down_read(&block->cb_lock);
3304         /* Need to obtain rtnl lock if block is bound to devs that require it.
3305          * In block bind code cb_lock is obtained while holding rtnl, so we must
3306          * obtain the locks in same order here.
3307          */
3308         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3309                 up_read(&block->cb_lock);
3310                 take_rtnl = true;
3311                 goto retry;
3312         }
3313
3314         /* Make sure all netdevs sharing this block are offload-capable. */
3315         if (block->nooffloaddevcnt && err_stop) {
3316                 ok_count = -EOPNOTSUPP;
3317                 goto err_unlock;
3318         }
3319
3320         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3321         if (ok_count < 0)
3322                 goto err_unlock;
3323
3324         if (tp->ops->hw_add)
3325                 tp->ops->hw_add(tp, type_data);
3326         if (ok_count > 0)
3327                 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3328                                           ok_count, true);
3329 err_unlock:
3330         up_read(&block->cb_lock);
3331         if (take_rtnl)
3332                 rtnl_unlock();
3333         return ok_count < 0 ? ok_count : 0;
3334 }
3335 EXPORT_SYMBOL(tc_setup_cb_add);
3336
3337 /* Destructive filter replace. If filter that wasn't already in hardware is
3338  * successfully offloaded, increment block offload counter. On failure,
3339  * previously offloaded filter is considered to be destroyed and offload counter
3340  * is decremented.
3341  */
3342
3343 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3344                         enum tc_setup_type type, void *type_data, bool err_stop,
3345                         u32 *old_flags, unsigned int *old_in_hw_count,
3346                         u32 *new_flags, unsigned int *new_in_hw_count,
3347                         bool rtnl_held)
3348 {
3349         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3350         int ok_count;
3351
3352 retry:
3353         if (take_rtnl)
3354                 rtnl_lock();
3355         down_read(&block->cb_lock);
3356         /* Need to obtain rtnl lock if block is bound to devs that require it.
3357          * In block bind code cb_lock is obtained while holding rtnl, so we must
3358          * obtain the locks in same order here.
3359          */
3360         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3361                 up_read(&block->cb_lock);
3362                 take_rtnl = true;
3363                 goto retry;
3364         }
3365
3366         /* Make sure all netdevs sharing this block are offload-capable. */
3367         if (block->nooffloaddevcnt && err_stop) {
3368                 ok_count = -EOPNOTSUPP;
3369                 goto err_unlock;
3370         }
3371
3372         tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3373         if (tp->ops->hw_del)
3374                 tp->ops->hw_del(tp, type_data);
3375
3376         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3377         if (ok_count < 0)
3378                 goto err_unlock;
3379
3380         if (tp->ops->hw_add)
3381                 tp->ops->hw_add(tp, type_data);
3382         if (ok_count > 0)
3383                 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3384                                           new_flags, ok_count, true);
3385 err_unlock:
3386         up_read(&block->cb_lock);
3387         if (take_rtnl)
3388                 rtnl_unlock();
3389         return ok_count < 0 ? ok_count : 0;
3390 }
3391 EXPORT_SYMBOL(tc_setup_cb_replace);
3392
3393 /* Destroy filter and decrement block offload counter, if filter was previously
3394  * offloaded.
3395  */
3396
3397 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3398                         enum tc_setup_type type, void *type_data, bool err_stop,
3399                         u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3400 {
3401         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3402         int ok_count;
3403
3404 retry:
3405         if (take_rtnl)
3406                 rtnl_lock();
3407         down_read(&block->cb_lock);
3408         /* Need to obtain rtnl lock if block is bound to devs that require it.
3409          * In block bind code cb_lock is obtained while holding rtnl, so we must
3410          * obtain the locks in same order here.
3411          */
3412         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3413                 up_read(&block->cb_lock);
3414                 take_rtnl = true;
3415                 goto retry;
3416         }
3417
3418         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3419
3420         tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3421         if (tp->ops->hw_del)
3422                 tp->ops->hw_del(tp, type_data);
3423
3424         up_read(&block->cb_lock);
3425         if (take_rtnl)
3426                 rtnl_unlock();
3427         return ok_count < 0 ? ok_count : 0;
3428 }
3429 EXPORT_SYMBOL(tc_setup_cb_destroy);
3430
3431 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3432                           bool add, flow_setup_cb_t *cb,
3433                           enum tc_setup_type type, void *type_data,
3434                           void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3435 {
3436         int err = cb(type, type_data, cb_priv);
3437
3438         if (err) {
3439                 if (add && tc_skip_sw(*flags))
3440                         return err;
3441         } else {
3442                 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3443                                           add);
3444         }
3445
3446         return 0;
3447 }
3448 EXPORT_SYMBOL(tc_setup_cb_reoffload);
3449
3450 static int tcf_act_get_cookie(struct flow_action_entry *entry,
3451                               const struct tc_action *act)
3452 {
3453         struct tc_cookie *cookie;
3454         int err = 0;
3455
3456         rcu_read_lock();
3457         cookie = rcu_dereference(act->act_cookie);
3458         if (cookie) {
3459                 entry->cookie = flow_action_cookie_create(cookie->data,
3460                                                           cookie->len,
3461                                                           GFP_ATOMIC);
3462                 if (!entry->cookie)
3463                         err = -ENOMEM;
3464         }
3465         rcu_read_unlock();
3466         return err;
3467 }
3468
3469 static void tcf_act_put_cookie(struct flow_action_entry *entry)
3470 {
3471         flow_action_cookie_destroy(entry->cookie);
3472 }
3473
3474 void tc_cleanup_flow_action(struct flow_action *flow_action)
3475 {
3476         struct flow_action_entry *entry;
3477         int i;
3478
3479         flow_action_for_each(i, entry, flow_action) {
3480                 tcf_act_put_cookie(entry);
3481                 if (entry->destructor)
3482                         entry->destructor(entry->destructor_priv);
3483         }
3484 }
3485 EXPORT_SYMBOL(tc_cleanup_flow_action);
3486
3487 static void tcf_mirred_get_dev(struct flow_action_entry *entry,
3488                                const struct tc_action *act)
3489 {
3490 #ifdef CONFIG_NET_CLS_ACT
3491         entry->dev = act->ops->get_dev(act, &entry->destructor);
3492         if (!entry->dev)
3493                 return;
3494         entry->destructor_priv = entry->dev;
3495 #endif
3496 }
3497
3498 static void tcf_tunnel_encap_put_tunnel(void *priv)
3499 {
3500         struct ip_tunnel_info *tunnel = priv;
3501
3502         kfree(tunnel);
3503 }
3504
3505 static int tcf_tunnel_encap_get_tunnel(struct flow_action_entry *entry,
3506                                        const struct tc_action *act)
3507 {
3508         entry->tunnel = tcf_tunnel_info_copy(act);
3509         if (!entry->tunnel)
3510                 return -ENOMEM;
3511         entry->destructor = tcf_tunnel_encap_put_tunnel;
3512         entry->destructor_priv = entry->tunnel;
3513         return 0;
3514 }
3515
3516 static void tcf_sample_get_group(struct flow_action_entry *entry,
3517                                  const struct tc_action *act)
3518 {
3519 #ifdef CONFIG_NET_CLS_ACT
3520         entry->sample.psample_group =
3521                 act->ops->get_psample_group(act, &entry->destructor);
3522         entry->destructor_priv = entry->sample.psample_group;
3523 #endif
3524 }
3525
3526 static void tcf_gate_entry_destructor(void *priv)
3527 {
3528         struct action_gate_entry *oe = priv;
3529
3530         kfree(oe);
3531 }
3532
3533 static int tcf_gate_get_entries(struct flow_action_entry *entry,
3534                                 const struct tc_action *act)
3535 {
3536         entry->gate.entries = tcf_gate_get_list(act);
3537
3538         if (!entry->gate.entries)
3539                 return -EINVAL;
3540
3541         entry->destructor = tcf_gate_entry_destructor;
3542         entry->destructor_priv = entry->gate.entries;
3543
3544         return 0;
3545 }
3546
3547 static enum flow_action_hw_stats tc_act_hw_stats(u8 hw_stats)
3548 {
3549         if (WARN_ON_ONCE(hw_stats > TCA_ACT_HW_STATS_ANY))
3550                 return FLOW_ACTION_HW_STATS_DONT_CARE;
3551         else if (!hw_stats)
3552                 return FLOW_ACTION_HW_STATS_DISABLED;
3553
3554         return hw_stats;
3555 }
3556
3557 int tc_setup_flow_action(struct flow_action *flow_action,
3558                          const struct tcf_exts *exts)
3559 {
3560         struct tc_action *act;
3561         int i, j, k, err = 0;
3562
3563         BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3564         BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3565         BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3566
3567         if (!exts)
3568                 return 0;
3569
3570         j = 0;
3571         tcf_exts_for_each_action(i, act, exts) {
3572                 struct flow_action_entry *entry;
3573
3574                 entry = &flow_action->entries[j];
3575                 spin_lock_bh(&act->tcfa_lock);
3576                 err = tcf_act_get_cookie(entry, act);
3577                 if (err)
3578                         goto err_out_locked;
3579
3580                 entry->hw_stats = tc_act_hw_stats(act->hw_stats);
3581
3582                 if (is_tcf_gact_ok(act)) {
3583                         entry->id = FLOW_ACTION_ACCEPT;
3584                 } else if (is_tcf_gact_shot(act)) {
3585                         entry->id = FLOW_ACTION_DROP;
3586                 } else if (is_tcf_gact_trap(act)) {
3587                         entry->id = FLOW_ACTION_TRAP;
3588                 } else if (is_tcf_gact_goto_chain(act)) {
3589                         entry->id = FLOW_ACTION_GOTO;
3590                         entry->chain_index = tcf_gact_goto_chain_index(act);
3591                 } else if (is_tcf_mirred_egress_redirect(act)) {
3592                         entry->id = FLOW_ACTION_REDIRECT;
3593                         tcf_mirred_get_dev(entry, act);
3594                 } else if (is_tcf_mirred_egress_mirror(act)) {
3595                         entry->id = FLOW_ACTION_MIRRED;
3596                         tcf_mirred_get_dev(entry, act);
3597                 } else if (is_tcf_mirred_ingress_redirect(act)) {
3598                         entry->id = FLOW_ACTION_REDIRECT_INGRESS;
3599                         tcf_mirred_get_dev(entry, act);
3600                 } else if (is_tcf_mirred_ingress_mirror(act)) {
3601                         entry->id = FLOW_ACTION_MIRRED_INGRESS;
3602                         tcf_mirred_get_dev(entry, act);
3603                 } else if (is_tcf_vlan(act)) {
3604                         switch (tcf_vlan_action(act)) {
3605                         case TCA_VLAN_ACT_PUSH:
3606                                 entry->id = FLOW_ACTION_VLAN_PUSH;
3607                                 entry->vlan.vid = tcf_vlan_push_vid(act);
3608                                 entry->vlan.proto = tcf_vlan_push_proto(act);
3609                                 entry->vlan.prio = tcf_vlan_push_prio(act);
3610                                 break;
3611                         case TCA_VLAN_ACT_POP:
3612                                 entry->id = FLOW_ACTION_VLAN_POP;
3613                                 break;
3614                         case TCA_VLAN_ACT_MODIFY:
3615                                 entry->id = FLOW_ACTION_VLAN_MANGLE;
3616                                 entry->vlan.vid = tcf_vlan_push_vid(act);
3617                                 entry->vlan.proto = tcf_vlan_push_proto(act);
3618                                 entry->vlan.prio = tcf_vlan_push_prio(act);
3619                                 break;
3620                         default:
3621                                 err = -EOPNOTSUPP;
3622                                 goto err_out_locked;
3623                         }
3624                 } else if (is_tcf_tunnel_set(act)) {
3625                         entry->id = FLOW_ACTION_TUNNEL_ENCAP;
3626                         err = tcf_tunnel_encap_get_tunnel(entry, act);
3627                         if (err)
3628                                 goto err_out_locked;
3629                 } else if (is_tcf_tunnel_release(act)) {
3630                         entry->id = FLOW_ACTION_TUNNEL_DECAP;
3631                 } else if (is_tcf_pedit(act)) {
3632                         for (k = 0; k < tcf_pedit_nkeys(act); k++) {
3633                                 switch (tcf_pedit_cmd(act, k)) {
3634                                 case TCA_PEDIT_KEY_EX_CMD_SET:
3635                                         entry->id = FLOW_ACTION_MANGLE;
3636                                         break;
3637                                 case TCA_PEDIT_KEY_EX_CMD_ADD:
3638                                         entry->id = FLOW_ACTION_ADD;
3639                                         break;
3640                                 default:
3641                                         err = -EOPNOTSUPP;
3642                                         goto err_out_locked;
3643                                 }
3644                                 entry->mangle.htype = tcf_pedit_htype(act, k);
3645                                 entry->mangle.mask = tcf_pedit_mask(act, k);
3646                                 entry->mangle.val = tcf_pedit_val(act, k);
3647                                 entry->mangle.offset = tcf_pedit_offset(act, k);
3648                                 entry->hw_stats = tc_act_hw_stats(act->hw_stats);
3649                                 entry = &flow_action->entries[++j];
3650                         }
3651                 } else if (is_tcf_csum(act)) {
3652                         entry->id = FLOW_ACTION_CSUM;
3653                         entry->csum_flags = tcf_csum_update_flags(act);
3654                 } else if (is_tcf_skbedit_mark(act)) {
3655                         entry->id = FLOW_ACTION_MARK;
3656                         entry->mark = tcf_skbedit_mark(act);
3657                 } else if (is_tcf_sample(act)) {
3658                         entry->id = FLOW_ACTION_SAMPLE;
3659                         entry->sample.trunc_size = tcf_sample_trunc_size(act);
3660                         entry->sample.truncate = tcf_sample_truncate(act);
3661                         entry->sample.rate = tcf_sample_rate(act);
3662                         tcf_sample_get_group(entry, act);
3663                 } else if (is_tcf_police(act)) {
3664                         entry->id = FLOW_ACTION_POLICE;
3665                         entry->police.burst = tcf_police_burst(act);
3666                         entry->police.rate_bytes_ps =
3667                                 tcf_police_rate_bytes_ps(act);
3668                         entry->police.mtu = tcf_police_tcfp_mtu(act);
3669                         entry->police.index = act->tcfa_index;
3670                 } else if (is_tcf_ct(act)) {
3671                         entry->id = FLOW_ACTION_CT;
3672                         entry->ct.action = tcf_ct_action(act);
3673                         entry->ct.zone = tcf_ct_zone(act);
3674                         entry->ct.flow_table = tcf_ct_ft(act);
3675                 } else if (is_tcf_mpls(act)) {
3676                         switch (tcf_mpls_action(act)) {
3677                         case TCA_MPLS_ACT_PUSH:
3678                                 entry->id = FLOW_ACTION_MPLS_PUSH;
3679                                 entry->mpls_push.proto = tcf_mpls_proto(act);
3680                                 entry->mpls_push.label = tcf_mpls_label(act);
3681                                 entry->mpls_push.tc = tcf_mpls_tc(act);
3682                                 entry->mpls_push.bos = tcf_mpls_bos(act);
3683                                 entry->mpls_push.ttl = tcf_mpls_ttl(act);
3684                                 break;
3685                         case TCA_MPLS_ACT_POP:
3686                                 entry->id = FLOW_ACTION_MPLS_POP;
3687                                 entry->mpls_pop.proto = tcf_mpls_proto(act);
3688                                 break;
3689                         case TCA_MPLS_ACT_MODIFY:
3690                                 entry->id = FLOW_ACTION_MPLS_MANGLE;
3691                                 entry->mpls_mangle.label = tcf_mpls_label(act);
3692                                 entry->mpls_mangle.tc = tcf_mpls_tc(act);
3693                                 entry->mpls_mangle.bos = tcf_mpls_bos(act);
3694                                 entry->mpls_mangle.ttl = tcf_mpls_ttl(act);
3695                                 break;
3696                         default:
3697                                 goto err_out_locked;
3698                         }
3699                 } else if (is_tcf_skbedit_ptype(act)) {
3700                         entry->id = FLOW_ACTION_PTYPE;
3701                         entry->ptype = tcf_skbedit_ptype(act);
3702                 } else if (is_tcf_skbedit_priority(act)) {
3703                         entry->id = FLOW_ACTION_PRIORITY;
3704                         entry->priority = tcf_skbedit_priority(act);
3705                 } else if (is_tcf_gate(act)) {
3706                         entry->id = FLOW_ACTION_GATE;
3707                         entry->gate.index = tcf_gate_index(act);
3708                         entry->gate.prio = tcf_gate_prio(act);
3709                         entry->gate.basetime = tcf_gate_basetime(act);
3710                         entry->gate.cycletime = tcf_gate_cycletime(act);
3711                         entry->gate.cycletimeext = tcf_gate_cycletimeext(act);
3712                         entry->gate.num_entries = tcf_gate_num_entries(act);
3713                         err = tcf_gate_get_entries(entry, act);
3714                         if (err)
3715                                 goto err_out;
3716                 } else {
3717                         err = -EOPNOTSUPP;
3718                         goto err_out_locked;
3719                 }
3720                 spin_unlock_bh(&act->tcfa_lock);
3721
3722                 if (!is_tcf_pedit(act))
3723                         j++;
3724         }
3725
3726 err_out:
3727         if (err)
3728                 tc_cleanup_flow_action(flow_action);
3729
3730         return err;
3731 err_out_locked:
3732         spin_unlock_bh(&act->tcfa_lock);
3733         goto err_out;
3734 }
3735 EXPORT_SYMBOL(tc_setup_flow_action);
3736
3737 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3738 {
3739         unsigned int num_acts = 0;
3740         struct tc_action *act;
3741         int i;
3742
3743         tcf_exts_for_each_action(i, act, exts) {
3744                 if (is_tcf_pedit(act))
3745                         num_acts += tcf_pedit_nkeys(act);
3746                 else
3747                         num_acts++;
3748         }
3749         return num_acts;
3750 }
3751 EXPORT_SYMBOL(tcf_exts_num_actions);
3752
3753 #ifdef CONFIG_NET_CLS_ACT
3754 static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3755                                         u32 *p_block_index,
3756                                         struct netlink_ext_ack *extack)
3757 {
3758         *p_block_index = nla_get_u32(block_index_attr);
3759         if (!*p_block_index) {
3760                 NL_SET_ERR_MSG(extack, "Block number may not be zero");
3761                 return -EINVAL;
3762         }
3763
3764         return 0;
3765 }
3766
3767 int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3768                     enum flow_block_binder_type binder_type,
3769                     struct nlattr *block_index_attr,
3770                     struct netlink_ext_ack *extack)
3771 {
3772         u32 block_index;
3773         int err;
3774
3775         if (!block_index_attr)
3776                 return 0;
3777
3778         err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3779         if (err)
3780                 return err;
3781
3782         if (!block_index)
3783                 return 0;
3784
3785         qe->info.binder_type = binder_type;
3786         qe->info.chain_head_change = tcf_chain_head_change_dflt;
3787         qe->info.chain_head_change_priv = &qe->filter_chain;
3788         qe->info.block_index = block_index;
3789
3790         return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3791 }
3792 EXPORT_SYMBOL(tcf_qevent_init);
3793
3794 void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3795 {
3796         if (qe->info.block_index)
3797                 tcf_block_put_ext(qe->block, sch, &qe->info);
3798 }
3799 EXPORT_SYMBOL(tcf_qevent_destroy);
3800
3801 int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3802                                struct netlink_ext_ack *extack)
3803 {
3804         u32 block_index;
3805         int err;
3806
3807         if (!block_index_attr)
3808                 return 0;
3809
3810         err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3811         if (err)
3812                 return err;
3813
3814         /* Bounce newly-configured block or change in block. */
3815         if (block_index != qe->info.block_index) {
3816                 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3817                 return -EINVAL;
3818         }
3819
3820         return 0;
3821 }
3822 EXPORT_SYMBOL(tcf_qevent_validate_change);
3823
3824 struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
3825                                   struct sk_buff **to_free, int *ret)
3826 {
3827         struct tcf_result cl_res;
3828         struct tcf_proto *fl;
3829
3830         if (!qe->info.block_index)
3831                 return skb;
3832
3833         fl = rcu_dereference_bh(qe->filter_chain);
3834
3835         switch (tcf_classify(skb, fl, &cl_res, false)) {
3836         case TC_ACT_SHOT:
3837                 qdisc_qstats_drop(sch);
3838                 __qdisc_drop(skb, to_free);
3839                 *ret = __NET_XMIT_BYPASS;
3840                 return NULL;
3841         case TC_ACT_STOLEN:
3842         case TC_ACT_QUEUED:
3843         case TC_ACT_TRAP:
3844                 __qdisc_drop(skb, to_free);
3845                 *ret = __NET_XMIT_STOLEN;
3846                 return NULL;
3847         case TC_ACT_REDIRECT:
3848                 skb_do_redirect(skb);
3849                 *ret = __NET_XMIT_STOLEN;
3850                 return NULL;
3851         }
3852
3853         return skb;
3854 }
3855 EXPORT_SYMBOL(tcf_qevent_handle);
3856
3857 int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
3858 {
3859         if (!qe->info.block_index)
3860                 return 0;
3861         return nla_put_u32(skb, attr_name, qe->info.block_index);
3862 }
3863 EXPORT_SYMBOL(tcf_qevent_dump);
3864 #endif
3865
3866 static __net_init int tcf_net_init(struct net *net)
3867 {
3868         struct tcf_net *tn = net_generic(net, tcf_net_id);
3869
3870         spin_lock_init(&tn->idr_lock);
3871         idr_init(&tn->idr);
3872         return 0;
3873 }
3874
3875 static void __net_exit tcf_net_exit(struct net *net)
3876 {
3877         struct tcf_net *tn = net_generic(net, tcf_net_id);
3878
3879         idr_destroy(&tn->idr);
3880 }
3881
3882 static struct pernet_operations tcf_net_ops = {
3883         .init = tcf_net_init,
3884         .exit = tcf_net_exit,
3885         .id   = &tcf_net_id,
3886         .size = sizeof(struct tcf_net),
3887 };
3888
3889 static int __init tc_filter_init(void)
3890 {
3891         int err;
3892
3893         tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3894         if (!tc_filter_wq)
3895                 return -ENOMEM;
3896
3897         err = register_pernet_subsys(&tcf_net_ops);
3898         if (err)
3899                 goto err_register_pernet_subsys;
3900
3901         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3902                       RTNL_FLAG_DOIT_UNLOCKED);
3903         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3904                       RTNL_FLAG_DOIT_UNLOCKED);
3905         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3906                       tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3907         rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3908         rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3909         rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3910                       tc_dump_chain, 0);
3911
3912         return 0;
3913
3914 err_register_pernet_subsys:
3915         destroy_workqueue(tc_filter_wq);
3916         return err;
3917 }
3918
3919 subsys_initcall(tc_filter_init);