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