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, <[email protected]>
   6  *
   7  * Changes:
   8  *
   9  * Eduardo J. Blanco <[email protected]> :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/jhash.h>
  24 #include <linux/rculist.h>
  25 #include <net/net_namespace.h>
  26 #include <net/sock.h>
  27 #include <net/netlink.h>
  28 #include <net/pkt_sched.h>
  29 #include <net/pkt_cls.h>
  30 #include <net/tc_act/tc_pedit.h>
  31 #include <net/tc_act/tc_mirred.h>
  32 #include <net/tc_act/tc_vlan.h>
  33 #include <net/tc_act/tc_tunnel_key.h>
  34 #include <net/tc_act/tc_csum.h>
  35 #include <net/tc_act/tc_gact.h>
  36 #include <net/tc_act/tc_police.h>
  37 #include <net/tc_act/tc_sample.h>
  38 #include <net/tc_act/tc_skbedit.h>
  39 #include <net/tc_act/tc_ct.h>
  40 #include <net/tc_act/tc_mpls.h>
  41 #include <net/tc_act/tc_gate.h>
  42 #include <net/flow_offload.h>
  43
  44 extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
  45
  46 /* The list of all installed classifier types */
  47 static LIST_HEAD(tcf_proto_base);
  48
  49 /* Protects list of registered TC modules. It is pure SMP lock. */
  50 static DEFINE_RWLOCK(cls_mod_lock);
  51
  52 static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
  53 {
  54         return jhash_3words(tp->chain->index, tp->prio,
  55                             (__force __u32)tp->protocol, 0);
  56 }
  57
  58 static void tcf_proto_signal_destroying(struct tcf_chain *chain,
  59                                         struct tcf_proto *tp)
  60 {
  61         struct tcf_block *block = chain->block;
  62
  63         mutex_lock(&block->proto_destroy_lock);
  64         hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
  65                      destroy_obj_hashfn(tp));
  66         mutex_unlock(&block->proto_destroy_lock);
  67 }
  68
  69 static bool tcf_proto_cmp(const struct tcf_proto *tp1,
  70                           const struct tcf_proto *tp2)
  71 {
  72         return tp1->chain->index == tp2->chain->index &&
  73                tp1->prio == tp2->prio &&
  74                tp1->protocol == tp2->protocol;
  75 }
  76
  77 static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
  78                                         struct tcf_proto *tp)
  79 {
  80         u32 hash = destroy_obj_hashfn(tp);
  81         struct tcf_proto *iter;
  82         bool found = false;
  83
  84         rcu_read_lock();
  85         hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
  86                                    destroy_ht_node, hash) {
  87                 if (tcf_proto_cmp(tp, iter)) {
  88                         found = true;
  89                         break;
  90                 }
  91         }
  92         rcu_read_unlock();
  93
  94         return found;
  95 }
  96
  97 static void
  98 tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
  99 {
 100         struct tcf_block *block = chain->block;
 101
 102         mutex_lock(&block->proto_destroy_lock);
 103         if (hash_hashed(&tp->destroy_ht_node))
 104                 hash_del_rcu(&tp->destroy_ht_node);
 105         mutex_unlock(&block->proto_destroy_lock);
 106 }
 107
 108 /* Find classifier type by string name */
 109
 110 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
 111 {
 112         const struct tcf_proto_ops *t, *res = NULL;
 113
 114         if (kind) {
 115                 read_lock(&cls_mod_lock);
 116                 list_for_each_entry(t, &tcf_proto_base, head) {
 117                         if (strcmp(kind, t->kind) == 0) {
 118                                 if (try_module_get(t->owner))
 119                                         res = t;
 120                                 break;
 121                         }
 122                 }
 123                 read_unlock(&cls_mod_lock);
 124         }
 125         return res;
 126 }
 127
 128 static const struct tcf_proto_ops *
 129 tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
 130                      struct netlink_ext_ack *extack)
 131 {
 132         const struct tcf_proto_ops *ops;
 133
 134         ops = __tcf_proto_lookup_ops(kind);
 135         if (ops)
 136                 return ops;
 137 #ifdef CONFIG_MODULES
 138         if (rtnl_held)
 139                 rtnl_unlock();
 140         request_module("cls_%s", kind);
 141         if (rtnl_held)
 142                 rtnl_lock();
 143         ops = __tcf_proto_lookup_ops(kind);
 144         /* We dropped the RTNL semaphore in order to perform
 145          * the module load. So, even if we succeeded in loading
 146          * the module we have to replay the request. We indicate
 147          * this using -EAGAIN.
 148          */
 149         if (ops) {
 150                 module_put(ops->owner);
 151                 return ERR_PTR(-EAGAIN);
 152         }
 153 #endif
 154         NL_SET_ERR_MSG(extack, "TC classifier not found");
 155         return ERR_PTR(-ENOENT);
 156 }
 157
 158 /* Register(unregister) new classifier type */
 159
 160 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
 161 {
 162         struct tcf_proto_ops *t;
 163         int rc = -EEXIST;
 164
 165         write_lock(&cls_mod_lock);
 166         list_for_each_entry(t, &tcf_proto_base, head)
 167                 if (!strcmp(ops->kind, t->kind))
 168                         goto out;
 169
 170         list_add_tail(&ops->head, &tcf_proto_base);
 171         rc = 0;
 172 out:
 173         write_unlock(&cls_mod_lock);
 174         return rc;
 175 }
 176 EXPORT_SYMBOL(register_tcf_proto_ops);
 177
 178 static struct workqueue_struct *tc_filter_wq;
 179
 180 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
 181 {
 182         struct tcf_proto_ops *t;
 183         int rc = -ENOENT;
 184
 185         /* Wait for outstanding call_rcu()s, if any, from a
 186          * tcf_proto_ops's destroy() handler.
 187          */
 188         rcu_barrier();
 189         flush_workqueue(tc_filter_wq);
 190
 191         write_lock(&cls_mod_lock);
 192         list_for_each_entry(t, &tcf_proto_base, head) {
 193                 if (t == ops) {
 194                         list_del(&t->head);
 195                         rc = 0;
 196                         break;
 197                 }
 198         }
 199         write_unlock(&cls_mod_lock);
 200         return rc;
 201 }
 202 EXPORT_SYMBOL(unregister_tcf_proto_ops);
 203
 204 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
 205 {
 206         INIT_RCU_WORK(rwork, func);
 207         return queue_rcu_work(tc_filter_wq, rwork);
 208 }
 209 EXPORT_SYMBOL(tcf_queue_work);
 210
 211 /* Select new prio value from the range, managed by kernel. */
 212
 213 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
 214 {
 215         u32 first = TC_H_MAKE(0xC0000000U, 0U);
 216
 217         if (tp)
 218                 first = tp->prio - 1;
 219
 220         return TC_H_MAJ(first);
 221 }
 222
 223 static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
 224 {
 225         if (kind)
 226                 return nla_strscpy(name, kind, IFNAMSIZ) < 0;
 227         memset(name, 0, IFNAMSIZ);
 228         return false;
 229 }
 230
 231 static bool tcf_proto_is_unlocked(const char *kind)
 232 {
 233         const struct tcf_proto_ops *ops;
 234         bool ret;
 235
 236         if (strlen(kind) == 0)
 237                 return false;
 238
 239         ops = tcf_proto_lookup_ops(kind, false, NULL);
 240         /* On error return false to take rtnl lock. Proto lookup/create
 241          * functions will perform lookup again and properly handle errors.
 242          */
 243         if (IS_ERR(ops))
 244                 return false;
 245
 246         ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
 247         module_put(ops->owner);
 248         return ret;
 249 }
 250
 251 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
 252                                           u32 prio, struct tcf_chain *chain,
 253                                           bool rtnl_held,
 254                                           struct netlink_ext_ack *extack)
 255 {
 256         struct tcf_proto *tp;
 257         int err;
 258
 259         tp = kzalloc(sizeof(*tp), GFP_KERNEL);
 260         if (!tp)
 261                 return ERR_PTR(-ENOBUFS);
 262
 263         tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
 264         if (IS_ERR(tp->ops)) {
 265                 err = PTR_ERR(tp->ops);
 266                 goto errout;
 267         }
 268         tp->classify = tp->ops->classify;
 269         tp->protocol = protocol;
 270         tp->prio = prio;
 271         tp->chain = chain;
 272         spin_lock_init(&tp->lock);
 273         refcount_set(&tp->refcnt, 1);
 274
 275         err = tp->ops->init(tp);
 276         if (err) {
 277                 module_put(tp->ops->owner);
 278                 goto errout;
 279         }
 280         return tp;
 281
 282 errout:
 283         kfree(tp);
 284         return ERR_PTR(err);
 285 }
 286
 287 static void tcf_proto_get(struct tcf_proto *tp)
 288 {
 289         refcount_inc(&tp->refcnt);
 290 }
 291
 292 static void tcf_chain_put(struct tcf_chain *chain);
 293
 294 static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
 295                               bool sig_destroy, struct netlink_ext_ack *extack)
 296 {
 297         tp->ops->destroy(tp, rtnl_held, extack);
 298         if (sig_destroy)
 299                 tcf_proto_signal_destroyed(tp->chain, tp);
 300         tcf_chain_put(tp->chain);
 301         module_put(tp->ops->owner);
 302         kfree_rcu(tp, rcu);
 303 }
 304
 305 static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
 306                           struct netlink_ext_ack *extack)
 307 {
 308         if (refcount_dec_and_test(&tp->refcnt))
 309                 tcf_proto_destroy(tp, rtnl_held, true, extack);
 310 }
 311
 312 static bool tcf_proto_check_delete(struct tcf_proto *tp)
 313 {
 314         if (tp->ops->delete_empty)
 315                 return tp->ops->delete_empty(tp);
 316
 317         tp->deleting = true;
 318         return tp->deleting;
 319 }
 320
 321 static void tcf_proto_mark_delete(struct tcf_proto *tp)
 322 {
 323         spin_lock(&tp->lock);
 324         tp->deleting = true;
 325         spin_unlock(&tp->lock);
 326 }
 327
 328 static bool tcf_proto_is_deleting(struct tcf_proto *tp)
 329 {
 330         bool deleting;
 331
 332         spin_lock(&tp->lock);
 333         deleting = tp->deleting;
 334         spin_unlock(&tp->lock);
 335
 336         return deleting;
 337 }
 338
 339 #define ASSERT_BLOCK_LOCKED(block)                                      \
 340         lockdep_assert_held(&(block)->lock)
 341
 342 struct tcf_filter_chain_list_item {
 343         struct list_head list;
 344         tcf_chain_head_change_t *chain_head_change;
 345         void *chain_head_change_priv;
 346 };
 347
 348 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
 349                                           u32 chain_index)
 350 {
 351         struct tcf_chain *chain;
 352
 353         ASSERT_BLOCK_LOCKED(block);
 354
 355         chain = kzalloc(sizeof(*chain), GFP_KERNEL);
 356         if (!chain)
 357                 return NULL;
 358         list_add_tail_rcu(&chain->list, &block->chain_list);
 359         mutex_init(&chain->filter_chain_lock);
 360         chain->block = block;
 361         chain->index = chain_index;
 362         chain->refcnt = 1;
 363         if (!chain->index)
 364                 block->chain0.chain = chain;
 365         return chain;
 366 }
 367
 368 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
 369                                        struct tcf_proto *tp_head)
 370 {
 371         if (item->chain_head_change)
 372                 item->chain_head_change(tp_head, item->chain_head_change_priv);
 373 }
 374
 375 static void tcf_chain0_head_change(struct tcf_chain *chain,
 376                                    struct tcf_proto *tp_head)
 377 {
 378         struct tcf_filter_chain_list_item *item;
 379         struct tcf_block *block = chain->block;
 380
 381         if (chain->index)
 382                 return;
 383
 384         mutex_lock(&block->lock);
 385         list_for_each_entry(item, &block->chain0.filter_chain_list, list)
 386                 tcf_chain_head_change_item(item, tp_head);
 387         mutex_unlock(&block->lock);
 388 }
 389
 390 /* Returns true if block can be safely freed. */
 391
 392 static bool tcf_chain_detach(struct tcf_chain *chain)
 393 {
 394         struct tcf_block *block = chain->block;
 395
 396         ASSERT_BLOCK_LOCKED(block);
 397
 398         list_del_rcu(&chain->list);
 399         if (!chain->index)
 400                 block->chain0.chain = NULL;
 401
 402         if (list_empty(&block->chain_list) &&
 403             refcount_read(&block->refcnt) == 0)
 404                 return true;
 405
 406         return false;
 407 }
 408
 409 static void tcf_block_destroy(struct tcf_block *block)
 410 {
 411         mutex_destroy(&block->lock);
 412         mutex_destroy(&block->proto_destroy_lock);
 413         kfree_rcu(block, rcu);
 414 }
 415
 416 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
 417 {
 418         struct tcf_block *block = chain->block;
 419
 420         mutex_destroy(&chain->filter_chain_lock);
 421         kfree_rcu(chain, rcu);
 422         if (free_block)
 423                 tcf_block_destroy(block);
 424 }
 425
 426 static void tcf_chain_hold(struct tcf_chain *chain)
 427 {
 428         ASSERT_BLOCK_LOCKED(chain->block);
 429
 430         ++chain->refcnt;
 431 }
 432
 433 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
 434 {
 435         ASSERT_BLOCK_LOCKED(chain->block);
 436
 437         /* In case all the references are action references, this
 438          * chain should not be shown to the user.
 439          */
 440         return chain->refcnt == chain->action_refcnt;
 441 }
 442
 443 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
 444                                           u32 chain_index)
 445 {
 446         struct tcf_chain *chain;
 447
 448         ASSERT_BLOCK_LOCKED(block);
 449
 450         list_for_each_entry(chain, &block->chain_list, list) {
 451                 if (chain->index == chain_index)
 452                         return chain;
 453         }
 454         return NULL;
 455 }
 456
 457 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
 458 static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
 459                                               u32 chain_index)
 460 {
 461         struct tcf_chain *chain;
 462
 463         list_for_each_entry_rcu(chain, &block->chain_list, list) {
 464                 if (chain->index == chain_index)
 465                         return chain;
 466         }
 467         return NULL;
 468 }
 469 #endif
 470
 471 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
 472                            u32 seq, u16 flags, int event, bool unicast);
 473
 474 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
 475                                          u32 chain_index, bool create,
 476                                          bool by_act)
 477 {
 478         struct tcf_chain *chain = NULL;
 479         bool is_first_reference;
 480
 481         mutex_lock(&block->lock);
 482         chain = tcf_chain_lookup(block, chain_index);
 483         if (chain) {
 484                 tcf_chain_hold(chain);
 485         } else {
 486                 if (!create)
 487                         goto errout;
 488                 chain = tcf_chain_create(block, chain_index);
 489                 if (!chain)
 490                         goto errout;
 491         }
 492
 493         if (by_act)
 494                 ++chain->action_refcnt;
 495         is_first_reference = chain->refcnt - chain->action_refcnt == 1;
 496         mutex_unlock(&block->lock);
 497
 498         /* Send notification only in case we got the first
 499          * non-action reference. Until then, the chain acts only as
 500          * a placeholder for actions pointing to it and user ought
 501          * not know about them.
 502          */
 503         if (is_first_reference && !by_act)
 504                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
 505                                 RTM_NEWCHAIN, false);
 506
 507         return chain;
 508
 509 errout:
 510         mutex_unlock(&block->lock);
 511         return chain;
 512 }
 513
 514 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
 515                                        bool create)
 516 {
 517         return __tcf_chain_get(block, chain_index, create, false);
 518 }
 519
 520 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
 521 {
 522         return __tcf_chain_get(block, chain_index, true, true);
 523 }
 524 EXPORT_SYMBOL(tcf_chain_get_by_act);
 525
 526 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
 527                                void *tmplt_priv);
 528 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
 529                                   void *tmplt_priv, u32 chain_index,
 530                                   struct tcf_block *block, struct sk_buff *oskb,
 531                                   u32 seq, u16 flags, bool unicast);
 532
 533 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
 534                             bool explicitly_created)
 535 {
 536         struct tcf_block *block = chain->block;
 537         const struct tcf_proto_ops *tmplt_ops;
 538         bool free_block = false;
 539         unsigned int refcnt;
 540         void *tmplt_priv;
 541
 542         mutex_lock(&block->lock);
 543         if (explicitly_created) {
 544                 if (!chain->explicitly_created) {
 545                         mutex_unlock(&block->lock);
 546                         return;
 547                 }
 548                 chain->explicitly_created = false;
 549         }
 550
 551         if (by_act)
 552                 chain->action_refcnt--;
 553
 554         /* tc_chain_notify_delete can't be called while holding block lock.
 555          * However, when block is unlocked chain can be changed concurrently, so
 556          * save these to temporary variables.
 557          */
 558         refcnt = --chain->refcnt;
 559         tmplt_ops = chain->tmplt_ops;
 560         tmplt_priv = chain->tmplt_priv;
 561
 562         /* The last dropped non-action reference will trigger notification. */
 563         if (refcnt - chain->action_refcnt == 0 && !by_act) {
 564                 tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
 565                                        block, NULL, 0, 0, false);
 566                 /* Last reference to chain, no need to lock. */
 567                 chain->flushing = false;
 568         }
 569
 570         if (refcnt == 0)
 571                 free_block = tcf_chain_detach(chain);
 572         mutex_unlock(&block->lock);
 573
 574         if (refcnt == 0) {
 575                 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
 576                 tcf_chain_destroy(chain, free_block);
 577         }
 578 }
 579
 580 static void tcf_chain_put(struct tcf_chain *chain)
 581 {
 582         __tcf_chain_put(chain, false, false);
 583 }
 584
 585 void tcf_chain_put_by_act(struct tcf_chain *chain)
 586 {
 587         __tcf_chain_put(chain, true, false);
 588 }
 589 EXPORT_SYMBOL(tcf_chain_put_by_act);
 590
 591 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
 592 {
 593         __tcf_chain_put(chain, false, true);
 594 }
 595
 596 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
 597 {
 598         struct tcf_proto *tp, *tp_next;
 599
 600         mutex_lock(&chain->filter_chain_lock);
 601         tp = tcf_chain_dereference(chain->filter_chain, chain);
 602         while (tp) {
 603                 tp_next = rcu_dereference_protected(tp->next, 1);
 604                 tcf_proto_signal_destroying(chain, tp);
 605                 tp = tp_next;
 606         }
 607         tp = tcf_chain_dereference(chain->filter_chain, chain);
 608         RCU_INIT_POINTER(chain->filter_chain, NULL);
 609         tcf_chain0_head_change(chain, NULL);
 610         chain->flushing = true;
 611         mutex_unlock(&chain->filter_chain_lock);
 612
 613         while (tp) {
 614                 tp_next = rcu_dereference_protected(tp->next, 1);
 615                 tcf_proto_put(tp, rtnl_held, NULL);
 616                 tp = tp_next;
 617         }
 618 }
 619
 620 static int tcf_block_setup(struct tcf_block *block,
 621                            struct flow_block_offload *bo);
 622
 623 static void tcf_block_offload_init(struct flow_block_offload *bo,
 624                                    struct net_device *dev, struct Qdisc *sch,
 625                                    enum flow_block_command command,
 626                                    enum flow_block_binder_type binder_type,
 627                                    struct flow_block *flow_block,
 628                                    bool shared, struct netlink_ext_ack *extack)
 629 {
 630         bo->net = dev_net(dev);
 631         bo->command = command;
 632         bo->binder_type = binder_type;
 633         bo->block = flow_block;
 634         bo->block_shared = shared;
 635         bo->extack = extack;
 636         bo->sch = sch;
 637         INIT_LIST_HEAD(&bo->cb_list);
 638 }
 639
 640 static void tcf_block_unbind(struct tcf_block *block,
 641                              struct flow_block_offload *bo);
 642
 643 static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
 644 {
 645         struct tcf_block *block = block_cb->indr.data;
 646         struct net_device *dev = block_cb->indr.dev;
 647         struct Qdisc *sch = block_cb->indr.sch;
 648         struct netlink_ext_ack extack = {};
 649         struct flow_block_offload bo;
 650
 651         tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
 652                                block_cb->indr.binder_type,
 653                                &block->flow_block, tcf_block_shared(block),
 654                                &extack);
 655         rtnl_lock();
 656         down_write(&block->cb_lock);
 657         list_del(&block_cb->driver_list);
 658         list_move(&block_cb->list, &bo.cb_list);
 659         tcf_block_unbind(block, &bo);
 660         up_write(&block->cb_lock);
 661         rtnl_unlock();
 662 }
 663
 664 static bool tcf_block_offload_in_use(struct tcf_block *block)
 665 {
 666         return atomic_read(&block->offloadcnt);
 667 }
 668
 669 static int tcf_block_offload_cmd(struct tcf_block *block,
 670                                  struct net_device *dev, struct Qdisc *sch,
 671                                  struct tcf_block_ext_info *ei,
 672                                  enum flow_block_command command,
 673                                  struct netlink_ext_ack *extack)
 674 {
 675         struct flow_block_offload bo = {};
 676
 677         tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
 678                                &block->flow_block, tcf_block_shared(block),
 679                                extack);
 680
 681         if (dev->netdev_ops->ndo_setup_tc) {
 682                 int err;
 683
 684                 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
 685                 if (err < 0) {
 686                         if (err != -EOPNOTSUPP)
 687                                 NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
 688                         return err;
 689                 }
 690
 691                 return tcf_block_setup(block, &bo);
 692         }
 693
 694         flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
 695                                     tc_block_indr_cleanup);
 696         tcf_block_setup(block, &bo);
 697
 698         return -EOPNOTSUPP;
 699 }
 700
 701 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
 702                                   struct tcf_block_ext_info *ei,
 703                                   struct netlink_ext_ack *extack)
 704 {
 705         struct net_device *dev = q->dev_queue->dev;
 706         int err;
 707
 708         down_write(&block->cb_lock);
 709
 710         /* If tc offload feature is disabled and the block we try to bind
 711          * to already has some offloaded filters, forbid to bind.
 712          */
 713         if (dev->netdev_ops->ndo_setup_tc &&
 714             !tc_can_offload(dev) &&
 715             tcf_block_offload_in_use(block)) {
 716                 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
 717                 err = -EOPNOTSUPP;
 718                 goto err_unlock;
 719         }
 720
 721         err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
 722         if (err == -EOPNOTSUPP)
 723                 goto no_offload_dev_inc;
 724         if (err)
 725                 goto err_unlock;
 726
 727         up_write(&block->cb_lock);
 728         return 0;
 729
 730 no_offload_dev_inc:
 731         if (tcf_block_offload_in_use(block))
 732                 goto err_unlock;
 733
 734         err = 0;
 735         block->nooffloaddevcnt++;
 736 err_unlock:
 737         up_write(&block->cb_lock);
 738         return err;
 739 }
 740
 741 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
 742                                      struct tcf_block_ext_info *ei)
 743 {
 744         struct net_device *dev = q->dev_queue->dev;
 745         int err;
 746
 747         down_write(&block->cb_lock);
 748         err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
 749         if (err == -EOPNOTSUPP)
 750                 goto no_offload_dev_dec;
 751         up_write(&block->cb_lock);
 752         return;
 753
 754 no_offload_dev_dec:
 755         WARN_ON(block->nooffloaddevcnt-- == 0);
 756         up_write(&block->cb_lock);
 757 }
 758
 759 static int
 760 tcf_chain0_head_change_cb_add(struct tcf_block *block,
 761                               struct tcf_block_ext_info *ei,
 762                               struct netlink_ext_ack *extack)
 763 {
 764         struct tcf_filter_chain_list_item *item;
 765         struct tcf_chain *chain0;
 766
 767         item = kmalloc(sizeof(*item), GFP_KERNEL);
 768         if (!item) {
 769                 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
 770                 return -ENOMEM;
 771         }
 772         item->chain_head_change = ei->chain_head_change;
 773         item->chain_head_change_priv = ei->chain_head_change_priv;
 774
 775         mutex_lock(&block->lock);
 776         chain0 = block->chain0.chain;
 777         if (chain0)
 778                 tcf_chain_hold(chain0);
 779         else
 780                 list_add(&item->list, &block->chain0.filter_chain_list);
 781         mutex_unlock(&block->lock);
 782
 783         if (chain0) {
 784                 struct tcf_proto *tp_head;
 785
 786                 mutex_lock(&chain0->filter_chain_lock);
 787
 788                 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
 789                 if (tp_head)
 790                         tcf_chain_head_change_item(item, tp_head);
 791
 792                 mutex_lock(&block->lock);
 793                 list_add(&item->list, &block->chain0.filter_chain_list);
 794                 mutex_unlock(&block->lock);
 795
 796                 mutex_unlock(&chain0->filter_chain_lock);
 797                 tcf_chain_put(chain0);
 798         }
 799
 800         return 0;
 801 }
 802
 803 static void
 804 tcf_chain0_head_change_cb_del(struct tcf_block *block,
 805                               struct tcf_block_ext_info *ei)
 806 {
 807         struct tcf_filter_chain_list_item *item;
 808
 809         mutex_lock(&block->lock);
 810         list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
 811                 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
 812                     (item->chain_head_change == ei->chain_head_change &&
 813                      item->chain_head_change_priv == ei->chain_head_change_priv)) {
 814                         if (block->chain0.chain)
 815                                 tcf_chain_head_change_item(item, NULL);
 816                         list_del(&item->list);
 817                         mutex_unlock(&block->lock);
 818
 819                         kfree(item);
 820                         return;
 821                 }
 822         }
 823         mutex_unlock(&block->lock);
 824         WARN_ON(1);
 825 }
 826
 827 struct tcf_net {
 828         spinlock_t idr_lock; /* Protects idr */
 829         struct idr idr;
 830 };
 831
 832 static unsigned int tcf_net_id;
 833
 834 static int tcf_block_insert(struct tcf_block *block, struct net *net,
 835                             struct netlink_ext_ack *extack)
 836 {
 837         struct tcf_net *tn = net_generic(net, tcf_net_id);
 838         int err;
 839
 840         idr_preload(GFP_KERNEL);
 841         spin_lock(&tn->idr_lock);
 842         err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
 843                             GFP_NOWAIT);
 844         spin_unlock(&tn->idr_lock);
 845         idr_preload_end();
 846
 847         return err;
 848 }
 849
 850 static void tcf_block_remove(struct tcf_block *block, struct net *net)
 851 {
 852         struct tcf_net *tn = net_generic(net, tcf_net_id);
 853
 854         spin_lock(&tn->idr_lock);
 855         idr_remove(&tn->idr, block->index);
 856         spin_unlock(&tn->idr_lock);
 857 }
 858
 859 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
 860                                           u32 block_index,
 861                                           struct netlink_ext_ack *extack)
 862 {
 863         struct tcf_block *block;
 864
 865         block = kzalloc(sizeof(*block), GFP_KERNEL);
 866         if (!block) {
 867                 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
 868                 return ERR_PTR(-ENOMEM);
 869         }
 870         mutex_init(&block->lock);
 871         mutex_init(&block->proto_destroy_lock);
 872         init_rwsem(&block->cb_lock);
 873         flow_block_init(&block->flow_block);
 874         INIT_LIST_HEAD(&block->chain_list);
 875         INIT_LIST_HEAD(&block->owner_list);
 876         INIT_LIST_HEAD(&block->chain0.filter_chain_list);
 877
 878         refcount_set(&block->refcnt, 1);
 879         block->net = net;
 880         block->index = block_index;
 881
 882         /* Don't store q pointer for blocks which are shared */
 883         if (!tcf_block_shared(block))
 884                 block->q = q;
 885         return block;
 886 }
 887
 888 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
 889 {
 890         struct tcf_net *tn = net_generic(net, tcf_net_id);
 891
 892         return idr_find(&tn->idr, block_index);
 893 }
 894
 895 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
 896 {
 897         struct tcf_block *block;
 898
 899         rcu_read_lock();
 900         block = tcf_block_lookup(net, block_index);
 901         if (block && !refcount_inc_not_zero(&block->refcnt))
 902                 block = NULL;
 903         rcu_read_unlock();
 904
 905         return block;
 906 }
 907
 908 static struct tcf_chain *
 909 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
 910 {
 911         mutex_lock(&block->lock);
 912         if (chain)
 913                 chain = list_is_last(&chain->list, &block->chain_list) ?
 914                         NULL : list_next_entry(chain, list);
 915         else
 916                 chain = list_first_entry_or_null(&block->chain_list,
 917                                                  struct tcf_chain, list);
 918
 919         /* skip all action-only chains */
 920         while (chain && tcf_chain_held_by_acts_only(chain))
 921                 chain = list_is_last(&chain->list, &block->chain_list) ?
 922                         NULL : list_next_entry(chain, list);
 923
 924         if (chain)
 925                 tcf_chain_hold(chain);
 926         mutex_unlock(&block->lock);
 927
 928         return chain;
 929 }
 930
 931 /* Function to be used by all clients that want to iterate over all chains on
 932  * block. It properly obtains block->lock and takes reference to chain before
 933  * returning it. Users of this function must be tolerant to concurrent chain
 934  * insertion/deletion or ensure that no concurrent chain modification is
 935  * possible. Note that all netlink dump callbacks cannot guarantee to provide
 936  * consistent dump because rtnl lock is released each time skb is filled with
 937  * data and sent to user-space.
 938  */
 939
 940 struct tcf_chain *
 941 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
 942 {
 943         struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
 944
 945         if (chain)
 946                 tcf_chain_put(chain);
 947
 948         return chain_next;
 949 }
 950 EXPORT_SYMBOL(tcf_get_next_chain);
 951
 952 static struct tcf_proto *
 953 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
 954 {
 955         u32 prio = 0;
 956
 957         ASSERT_RTNL();
 958         mutex_lock(&chain->filter_chain_lock);
 959
 960         if (!tp) {
 961                 tp = tcf_chain_dereference(chain->filter_chain, chain);
 962         } else if (tcf_proto_is_deleting(tp)) {
 963                 /* 'deleting' flag is set and chain->filter_chain_lock was
 964                  * unlocked, which means next pointer could be invalid. Restart
 965                  * search.
 966                  */
 967                 prio = tp->prio + 1;
 968                 tp = tcf_chain_dereference(chain->filter_chain, chain);
 969
 970                 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
 971                         if (!tp->deleting && tp->prio >= prio)
 972                                 break;
 973         } else {
 974                 tp = tcf_chain_dereference(tp->next, chain);
 975         }
 976
 977         if (tp)
 978                 tcf_proto_get(tp);
 979
 980         mutex_unlock(&chain->filter_chain_lock);
 981
 982         return tp;
 983 }
 984
 985 /* Function to be used by all clients that want to iterate over all tp's on
 986  * chain. Users of this function must be tolerant to concurrent tp
 987  * insertion/deletion or ensure that no concurrent chain modification is
 988  * possible. Note that all netlink dump callbacks cannot guarantee to provide
 989  * consistent dump because rtnl lock is released each time skb is filled with
 990  * data and sent to user-space.
 991  */
 992
 993 struct tcf_proto *
 994 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
 995 {
 996         struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
 997
 998         if (tp)
 999                 tcf_proto_put(tp, true, NULL);
1000
1001         return tp_next;
1002 }
1003 EXPORT_SYMBOL(tcf_get_next_proto);
1004
1005 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1006 {
1007         struct tcf_chain *chain;
1008
1009         /* Last reference to block. At this point chains cannot be added or
1010          * removed concurrently.
1011          */
1012         for (chain = tcf_get_next_chain(block, NULL);
1013              chain;
1014              chain = tcf_get_next_chain(block, chain)) {
1015                 tcf_chain_put_explicitly_created(chain);
1016                 tcf_chain_flush(chain, rtnl_held);
1017         }
1018 }
1019
1020 /* Lookup Qdisc and increments its reference counter.
1021  * Set parent, if necessary.
1022  */
1023
1024 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1025                             u32 *parent, int ifindex, bool rtnl_held,
1026                             struct netlink_ext_ack *extack)
1027 {
1028         const struct Qdisc_class_ops *cops;
1029         struct net_device *dev;
1030         int err = 0;
1031
1032         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1033                 return 0;
1034
1035         rcu_read_lock();
1036
1037         /* Find link */
1038         dev = dev_get_by_index_rcu(net, ifindex);
1039         if (!dev) {
1040                 rcu_read_unlock();
1041                 return -ENODEV;
1042         }
1043
1044         /* Find qdisc */
1045         if (!*parent) {
1046                 *q = dev->qdisc;
1047                 *parent = (*q)->handle;
1048         } else {
1049                 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1050                 if (!*q) {
1051                         NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1052                         err = -EINVAL;
1053                         goto errout_rcu;
1054                 }
1055         }
1056
1057         *q = qdisc_refcount_inc_nz(*q);
1058         if (!*q) {
1059                 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1060                 err = -EINVAL;
1061                 goto errout_rcu;
1062         }
1063
1064         /* Is it classful? */
1065         cops = (*q)->ops->cl_ops;
1066         if (!cops) {
1067                 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1068                 err = -EINVAL;
1069                 goto errout_qdisc;
1070         }
1071
1072         if (!cops->tcf_block) {
1073                 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1074                 err = -EOPNOTSUPP;
1075                 goto errout_qdisc;
1076         }
1077
1078 errout_rcu:
1079         /* At this point we know that qdisc is not noop_qdisc,
1080          * which means that qdisc holds a reference to net_device
1081          * and we hold a reference to qdisc, so it is safe to release
1082          * rcu read lock.
1083          */
1084         rcu_read_unlock();
1085         return err;
1086
1087 errout_qdisc:
1088         rcu_read_unlock();
1089
1090         if (rtnl_held)
1091                 qdisc_put(*q);
1092         else
1093                 qdisc_put_unlocked(*q);
1094         *q = NULL;
1095
1096         return err;
1097 }
1098
1099 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1100                                int ifindex, struct netlink_ext_ack *extack)
1101 {
1102         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1103                 return 0;
1104
1105         /* Do we search for filter, attached to class? */
1106         if (TC_H_MIN(parent)) {
1107                 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1108
1109                 *cl = cops->find(q, parent);
1110                 if (*cl == 0) {
1111                         NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1112                         return -ENOENT;
1113                 }
1114         }
1115
1116         return 0;
1117 }
1118
1119 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1120                                           unsigned long cl, int ifindex,
1121                                           u32 block_index,
1122                                           struct netlink_ext_ack *extack)
1123 {
1124         struct tcf_block *block;
1125
1126         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1127                 block = tcf_block_refcnt_get(net, block_index);
1128                 if (!block) {
1129                         NL_SET_ERR_MSG(extack, "Block of given index was not found");
1130                         return ERR_PTR(-EINVAL);
1131                 }
1132         } else {
1133                 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1134
1135                 block = cops->tcf_block(q, cl, extack);
1136                 if (!block)
1137                         return ERR_PTR(-EINVAL);
1138
1139                 if (tcf_block_shared(block)) {
1140                         NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1141                         return ERR_PTR(-EOPNOTSUPP);
1142                 }
1143
1144                 /* Always take reference to block in order to support execution
1145                  * of rules update path of cls API without rtnl lock. Caller
1146                  * must release block when it is finished using it. 'if' block
1147                  * of this conditional obtain reference to block by calling
1148                  * tcf_block_refcnt_get().
1149                  */
1150                 refcount_inc(&block->refcnt);
1151         }
1152
1153         return block;
1154 }
1155
1156 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1157                             struct tcf_block_ext_info *ei, bool rtnl_held)
1158 {
1159         if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1160                 /* Flushing/putting all chains will cause the block to be
1161                  * deallocated when last chain is freed. However, if chain_list
1162                  * is empty, block has to be manually deallocated. After block
1163                  * reference counter reached 0, it is no longer possible to
1164                  * increment it or add new chains to block.
1165                  */
1166                 bool free_block = list_empty(&block->chain_list);
1167
1168                 mutex_unlock(&block->lock);
1169                 if (tcf_block_shared(block))
1170                         tcf_block_remove(block, block->net);
1171
1172                 if (q)
1173                         tcf_block_offload_unbind(block, q, ei);
1174
1175                 if (free_block)
1176                         tcf_block_destroy(block);
1177                 else
1178                         tcf_block_flush_all_chains(block, rtnl_held);
1179         } else if (q) {
1180                 tcf_block_offload_unbind(block, q, ei);
1181         }
1182 }
1183
1184 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1185 {
1186         __tcf_block_put(block, NULL, NULL, rtnl_held);
1187 }
1188
1189 /* Find tcf block.
1190  * Set q, parent, cl when appropriate.
1191  */
1192
1193 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1194                                         u32 *parent, unsigned long *cl,
1195                                         int ifindex, u32 block_index,
1196                                         struct netlink_ext_ack *extack)
1197 {
1198         struct tcf_block *block;
1199         int err = 0;
1200
1201         ASSERT_RTNL();
1202
1203         err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1204         if (err)
1205                 goto errout;
1206
1207         err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1208         if (err)
1209                 goto errout_qdisc;
1210
1211         block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1212         if (IS_ERR(block)) {
1213                 err = PTR_ERR(block);
1214                 goto errout_qdisc;
1215         }
1216
1217         return block;
1218
1219 errout_qdisc:
1220         if (*q)
1221                 qdisc_put(*q);
1222 errout:
1223         *q = NULL;
1224         return ERR_PTR(err);
1225 }
1226
1227 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1228                               bool rtnl_held)
1229 {
1230         if (!IS_ERR_OR_NULL(block))
1231                 tcf_block_refcnt_put(block, rtnl_held);
1232
1233         if (q) {
1234                 if (rtnl_held)
1235                         qdisc_put(q);
1236                 else
1237                         qdisc_put_unlocked(q);
1238         }
1239 }
1240
1241 struct tcf_block_owner_item {
1242         struct list_head list;
1243         struct Qdisc *q;
1244         enum flow_block_binder_type binder_type;
1245 };
1246
1247 static void
1248 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1249                                struct Qdisc *q,
1250                                enum flow_block_binder_type binder_type)
1251 {
1252         if (block->keep_dst &&
1253             binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1254             binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1255                 netif_keep_dst(qdisc_dev(q));
1256 }
1257
1258 void tcf_block_netif_keep_dst(struct tcf_block *block)
1259 {
1260         struct tcf_block_owner_item *item;
1261
1262         block->keep_dst = true;
1263         list_for_each_entry(item, &block->owner_list, list)
1264                 tcf_block_owner_netif_keep_dst(block, item->q,
1265                                                item->binder_type);
1266 }
1267 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1268
1269 static int tcf_block_owner_add(struct tcf_block *block,
1270                                struct Qdisc *q,
1271                                enum flow_block_binder_type binder_type)
1272 {
1273         struct tcf_block_owner_item *item;
1274
1275         item = kmalloc(sizeof(*item), GFP_KERNEL);
1276         if (!item)
1277                 return -ENOMEM;
1278         item->q = q;
1279         item->binder_type = binder_type;
1280         list_add(&item->list, &block->owner_list);
1281         return 0;
1282 }
1283
1284 static void tcf_block_owner_del(struct tcf_block *block,
1285                                 struct Qdisc *q,
1286                                 enum flow_block_binder_type binder_type)
1287 {
1288         struct tcf_block_owner_item *item;
1289
1290         list_for_each_entry(item, &block->owner_list, list) {
1291                 if (item->q == q && item->binder_type == binder_type) {
1292                         list_del(&item->list);
1293                         kfree(item);
1294                         return;
1295                 }
1296         }
1297         WARN_ON(1);
1298 }
1299
1300 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1301                       struct tcf_block_ext_info *ei,
1302                       struct netlink_ext_ack *extack)
1303 {
1304         struct net *net = qdisc_net(q);
1305         struct tcf_block *block = NULL;
1306         int err;
1307
1308         if (ei->block_index)
1309                 /* block_index not 0 means the shared block is requested */
1310                 block = tcf_block_refcnt_get(net, ei->block_index);
1311
1312         if (!block) {
1313                 block = tcf_block_create(net, q, ei->block_index, extack);
1314                 if (IS_ERR(block))
1315                         return PTR_ERR(block);
1316                 if (tcf_block_shared(block)) {
1317                         err = tcf_block_insert(block, net, extack);
1318                         if (err)
1319                                 goto err_block_insert;
1320                 }
1321         }
1322
1323         err = tcf_block_owner_add(block, q, ei->binder_type);
1324         if (err)
1325                 goto err_block_owner_add;
1326
1327         tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1328
1329         err = tcf_chain0_head_change_cb_add(block, ei, extack);
1330         if (err)
1331                 goto err_chain0_head_change_cb_add;
1332
1333         err = tcf_block_offload_bind(block, q, ei, extack);
1334         if (err)
1335                 goto err_block_offload_bind;
1336
1337         *p_block = block;
1338         return 0;
1339
1340 err_block_offload_bind:
1341         tcf_chain0_head_change_cb_del(block, ei);
1342 err_chain0_head_change_cb_add:
1343         tcf_block_owner_del(block, q, ei->binder_type);
1344 err_block_owner_add:
1345 err_block_insert:
1346         tcf_block_refcnt_put(block, true);
1347         return err;
1348 }
1349 EXPORT_SYMBOL(tcf_block_get_ext);
1350
1351 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1352 {
1353         struct tcf_proto __rcu **p_filter_chain = priv;
1354
1355         rcu_assign_pointer(*p_filter_chain, tp_head);
1356 }
1357
1358 int tcf_block_get(struct tcf_block **p_block,
1359                   struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1360                   struct netlink_ext_ack *extack)
1361 {
1362         struct tcf_block_ext_info ei = {
1363                 .chain_head_change = tcf_chain_head_change_dflt,
1364                 .chain_head_change_priv = p_filter_chain,
1365         };
1366
1367         WARN_ON(!p_filter_chain);
1368         return tcf_block_get_ext(p_block, q, &ei, extack);
1369 }
1370 EXPORT_SYMBOL(tcf_block_get);
1371
1372 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1373  * actions should be all removed after flushing.
1374  */
1375 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1376                        struct tcf_block_ext_info *ei)
1377 {
1378         if (!block)
1379                 return;
1380         tcf_chain0_head_change_cb_del(block, ei);
1381         tcf_block_owner_del(block, q, ei->binder_type);
1382
1383         __tcf_block_put(block, q, ei, true);
1384 }
1385 EXPORT_SYMBOL(tcf_block_put_ext);
1386
1387 void tcf_block_put(struct tcf_block *block)
1388 {
1389         struct tcf_block_ext_info ei = {0, };
1390
1391         if (!block)
1392                 return;
1393         tcf_block_put_ext(block, block->q, &ei);
1394 }
1395
1396 EXPORT_SYMBOL(tcf_block_put);
1397
1398 static int
1399 tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1400                             void *cb_priv, bool add, bool offload_in_use,
1401                             struct netlink_ext_ack *extack)
1402 {
1403         struct tcf_chain *chain, *chain_prev;
1404         struct tcf_proto *tp, *tp_prev;
1405         int err;
1406
1407         lockdep_assert_held(&block->cb_lock);
1408
1409         for (chain = __tcf_get_next_chain(block, NULL);
1410              chain;
1411              chain_prev = chain,
1412                      chain = __tcf_get_next_chain(block, chain),
1413                      tcf_chain_put(chain_prev)) {
1414                 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1415                      tp_prev = tp,
1416                              tp = __tcf_get_next_proto(chain, tp),
1417                              tcf_proto_put(tp_prev, true, NULL)) {
1418                         if (tp->ops->reoffload) {
1419                                 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1420                                                          extack);
1421                                 if (err && add)
1422                                         goto err_playback_remove;
1423                         } else if (add && offload_in_use) {
1424                                 err = -EOPNOTSUPP;
1425                                 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1426                                 goto err_playback_remove;
1427                         }
1428                 }
1429         }
1430
1431         return 0;
1432
1433 err_playback_remove:
1434         tcf_proto_put(tp, true, NULL);
1435         tcf_chain_put(chain);
1436         tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1437                                     extack);
1438         return err;
1439 }
1440
1441 static int tcf_block_bind(struct tcf_block *block,
1442                           struct flow_block_offload *bo)
1443 {
1444         struct flow_block_cb *block_cb, *next;
1445         int err, i = 0;
1446
1447         lockdep_assert_held(&block->cb_lock);
1448
1449         list_for_each_entry(block_cb, &bo->cb_list, list) {
1450                 err = tcf_block_playback_offloads(block, block_cb->cb,
1451                                                   block_cb->cb_priv, true,
1452                                                   tcf_block_offload_in_use(block),
1453                                                   bo->extack);
1454                 if (err)
1455                         goto err_unroll;
1456                 if (!bo->unlocked_driver_cb)
1457                         block->lockeddevcnt++;
1458
1459                 i++;
1460         }
1461         list_splice(&bo->cb_list, &block->flow_block.cb_list);
1462
1463         return 0;
1464
1465 err_unroll:
1466         list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1467                 if (i-- > 0) {
1468                         list_del(&block_cb->list);
1469                         tcf_block_playback_offloads(block, block_cb->cb,
1470                                                     block_cb->cb_priv, false,
1471                                                     tcf_block_offload_in_use(block),
1472                                                     NULL);
1473                         if (!bo->unlocked_driver_cb)
1474                                 block->lockeddevcnt--;
1475                 }
1476                 flow_block_cb_free(block_cb);
1477         }
1478
1479         return err;
1480 }
1481
1482 static void tcf_block_unbind(struct tcf_block *block,
1483                              struct flow_block_offload *bo)
1484 {
1485         struct flow_block_cb *block_cb, *next;
1486
1487         lockdep_assert_held(&block->cb_lock);
1488
1489         list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1490                 tcf_block_playback_offloads(block, block_cb->cb,
1491                                             block_cb->cb_priv, false,
1492                                             tcf_block_offload_in_use(block),
1493                                             NULL);
1494                 list_del(&block_cb->list);
1495                 flow_block_cb_free(block_cb);
1496                 if (!bo->unlocked_driver_cb)
1497                         block->lockeddevcnt--;
1498         }
1499 }
1500
1501 static int tcf_block_setup(struct tcf_block *block,
1502                            struct flow_block_offload *bo)
1503 {
1504         int err;
1505
1506         switch (bo->command) {
1507         case FLOW_BLOCK_BIND:
1508                 err = tcf_block_bind(block, bo);
1509                 break;
1510         case FLOW_BLOCK_UNBIND:
1511                 err = 0;
1512                 tcf_block_unbind(block, bo);
1513                 break;
1514         default:
1515                 WARN_ON_ONCE(1);
1516                 err = -EOPNOTSUPP;
1517         }
1518
1519         return err;
1520 }
1521
1522 /* Main classifier routine: scans classifier chain attached
1523  * to this qdisc, (optionally) tests for protocol and asks
1524  * specific classifiers.
1525  */
1526 static inline int __tcf_classify(struct sk_buff *skb,
1527                                  const struct tcf_proto *tp,
1528                                  const struct tcf_proto *orig_tp,
1529                                  struct tcf_result *res,
1530                                  bool compat_mode,
1531                                  u32 *last_executed_chain)
1532 {
1533 #ifdef CONFIG_NET_CLS_ACT
1534         const int max_reclassify_loop = 4;
1535         const struct tcf_proto *first_tp;
1536         int limit = 0;
1537
1538 reclassify:
1539 #endif
1540         for (; tp; tp = rcu_dereference_bh(tp->next)) {
1541                 __be16 protocol = skb_protocol(skb, false);
1542                 int err;
1543
1544                 if (tp->protocol != protocol &&
1545                     tp->protocol != htons(ETH_P_ALL))
1546                         continue;
1547
1548                 err = tp->classify(skb, tp, res);
1549 #ifdef CONFIG_NET_CLS_ACT
1550                 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1551                         first_tp = orig_tp;
1552                         *last_executed_chain = first_tp->chain->index;
1553                         goto reset;
1554                 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1555                         first_tp = res->goto_tp;
1556                         *last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1557                         goto reset;
1558                 }
1559 #endif
1560                 if (err >= 0)
1561                         return err;
1562         }
1563
1564         return TC_ACT_UNSPEC; /* signal: continue lookup */
1565 #ifdef CONFIG_NET_CLS_ACT
1566 reset:
1567         if (unlikely(limit++ >= max_reclassify_loop)) {
1568                 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1569                                        tp->chain->block->index,
1570                                        tp->prio & 0xffff,
1571                                        ntohs(tp->protocol));
1572                 return TC_ACT_SHOT;
1573         }
1574
1575         tp = first_tp;
1576         goto reclassify;
1577 #endif
1578 }
1579
1580 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1581                  struct tcf_result *res, bool compat_mode)
1582 {
1583         u32 last_executed_chain = 0;
1584
1585         return __tcf_classify(skb, tp, tp, res, compat_mode,
1586                               &last_executed_chain);
1587 }
1588 EXPORT_SYMBOL(tcf_classify);
1589
1590 int tcf_classify_ingress(struct sk_buff *skb,
1591                          const struct tcf_block *ingress_block,
1592                          const struct tcf_proto *tp,
1593                          struct tcf_result *res, bool compat_mode)
1594 {
1595 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1596         u32 last_executed_chain = 0;
1597
1598         return __tcf_classify(skb, tp, tp, res, compat_mode,
1599                               &last_executed_chain);
1600 #else
1601         u32 last_executed_chain = tp ? tp->chain->index : 0;
1602         const struct tcf_proto *orig_tp = tp;
1603         struct tc_skb_ext *ext;
1604         int ret;
1605
1606         ext = skb_ext_find(skb, TC_SKB_EXT);
1607
1608         if (ext && ext->chain) {
1609                 struct tcf_chain *fchain;
1610
1611                 fchain = tcf_chain_lookup_rcu(ingress_block, ext->chain);
1612                 if (!fchain)
1613                         return TC_ACT_SHOT;
1614
1615                 /* Consume, so cloned/redirect skbs won't inherit ext */
1616                 skb_ext_del(skb, TC_SKB_EXT);
1617
1618                 tp = rcu_dereference_bh(fchain->filter_chain);
1619                 last_executed_chain = fchain->index;
1620         }
1621
1622         ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode,
1623                              &last_executed_chain);
1624
1625         /* If we missed on some chain */
1626         if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1627                 ext = skb_ext_add(skb, TC_SKB_EXT);
1628                 if (WARN_ON_ONCE(!ext))
1629                         return TC_ACT_SHOT;
1630                 ext->chain = last_executed_chain;
1631                 ext->mru = qdisc_skb_cb(skb)->mru;
1632         }
1633
1634         return ret;
1635 #endif
1636 }
1637 EXPORT_SYMBOL(tcf_classify_ingress);
1638
1639 struct tcf_chain_info {
1640         struct tcf_proto __rcu **pprev;
1641         struct tcf_proto __rcu *next;
1642 };
1643
1644 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1645                                            struct tcf_chain_info *chain_info)
1646 {
1647         return tcf_chain_dereference(*chain_info->pprev, chain);
1648 }
1649
1650 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1651                                struct tcf_chain_info *chain_info,
1652                                struct tcf_proto *tp)
1653 {
1654         if (chain->flushing)
1655                 return -EAGAIN;
1656
1657         if (*chain_info->pprev == chain->filter_chain)
1658                 tcf_chain0_head_change(chain, tp);
1659         tcf_proto_get(tp);
1660         RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1661         rcu_assign_pointer(*chain_info->pprev, tp);
1662
1663         return 0;
1664 }
1665
1666 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1667                                 struct tcf_chain_info *chain_info,
1668                                 struct tcf_proto *tp)
1669 {
1670         struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1671
1672         tcf_proto_mark_delete(tp);
1673         if (tp == chain->filter_chain)
1674                 tcf_chain0_head_change(chain, next);
1675         RCU_INIT_POINTER(*chain_info->pprev, next);
1676 }
1677
1678 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1679                                            struct tcf_chain_info *chain_info,
1680                                            u32 protocol, u32 prio,
1681                                            bool prio_allocate);
1682
1683 /* Try to insert new proto.
1684  * If proto with specified priority already exists, free new proto
1685  * and return existing one.
1686  */
1687
1688 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1689                                                     struct tcf_proto *tp_new,
1690                                                     u32 protocol, u32 prio,
1691                                                     bool rtnl_held)
1692 {
1693         struct tcf_chain_info chain_info;
1694         struct tcf_proto *tp;
1695         int err = 0;
1696
1697         mutex_lock(&chain->filter_chain_lock);
1698
1699         if (tcf_proto_exists_destroying(chain, tp_new)) {
1700                 mutex_unlock(&chain->filter_chain_lock);
1701                 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1702                 return ERR_PTR(-EAGAIN);
1703         }
1704
1705         tp = tcf_chain_tp_find(chain, &chain_info,
1706                                protocol, prio, false);
1707         if (!tp)
1708                 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1709         mutex_unlock(&chain->filter_chain_lock);
1710
1711         if (tp) {
1712                 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1713                 tp_new = tp;
1714         } else if (err) {
1715                 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1716                 tp_new = ERR_PTR(err);
1717         }
1718
1719         return tp_new;
1720 }
1721
1722 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1723                                       struct tcf_proto *tp, bool rtnl_held,
1724                                       struct netlink_ext_ack *extack)
1725 {
1726         struct tcf_chain_info chain_info;
1727         struct tcf_proto *tp_iter;
1728         struct tcf_proto **pprev;
1729         struct tcf_proto *next;
1730
1731         mutex_lock(&chain->filter_chain_lock);
1732
1733         /* Atomically find and remove tp from chain. */
1734         for (pprev = &chain->filter_chain;
1735              (tp_iter = tcf_chain_dereference(*pprev, chain));
1736              pprev = &tp_iter->next) {
1737                 if (tp_iter == tp) {
1738                         chain_info.pprev = pprev;
1739                         chain_info.next = tp_iter->next;
1740                         WARN_ON(tp_iter->deleting);
1741                         break;
1742                 }
1743         }
1744         /* Verify that tp still exists and no new filters were inserted
1745          * concurrently.
1746          * Mark tp for deletion if it is empty.
1747          */
1748         if (!tp_iter || !tcf_proto_check_delete(tp)) {
1749                 mutex_unlock(&chain->filter_chain_lock);
1750                 return;
1751         }
1752
1753         tcf_proto_signal_destroying(chain, tp);
1754         next = tcf_chain_dereference(chain_info.next, chain);
1755         if (tp == chain->filter_chain)
1756                 tcf_chain0_head_change(chain, next);
1757         RCU_INIT_POINTER(*chain_info.pprev, next);
1758         mutex_unlock(&chain->filter_chain_lock);
1759
1760         tcf_proto_put(tp, rtnl_held, extack);
1761 }
1762
1763 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1764                                            struct tcf_chain_info *chain_info,
1765                                            u32 protocol, u32 prio,
1766                                            bool prio_allocate)
1767 {
1768         struct tcf_proto **pprev;
1769         struct tcf_proto *tp;
1770
1771         /* Check the chain for existence of proto-tcf with this priority */
1772         for (pprev = &chain->filter_chain;
1773              (tp = tcf_chain_dereference(*pprev, chain));
1774              pprev = &tp->next) {
1775                 if (tp->prio >= prio) {
1776                         if (tp->prio == prio) {
1777                                 if (prio_allocate ||
1778                                     (tp->protocol != protocol && protocol))
1779                                         return ERR_PTR(-EINVAL);
1780                         } else {
1781                                 tp = NULL;
1782                         }
1783                         break;
1784                 }
1785         }
1786         chain_info->pprev = pprev;
1787         if (tp) {
1788                 chain_info->next = tp->next;
1789                 tcf_proto_get(tp);
1790         } else {
1791                 chain_info->next = NULL;
1792         }
1793         return tp;
1794 }
1795
1796 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1797                          struct tcf_proto *tp, struct tcf_block *block,
1798                          struct Qdisc *q, u32 parent, void *fh,
1799                          u32 portid, u32 seq, u16 flags, int event,
1800                          bool terse_dump, bool rtnl_held)
1801 {
1802         struct tcmsg *tcm;
1803         struct nlmsghdr  *nlh;
1804         unsigned char *b = skb_tail_pointer(skb);
1805
1806         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1807         if (!nlh)
1808                 goto out_nlmsg_trim;
1809         tcm = nlmsg_data(nlh);
1810         tcm->tcm_family = AF_UNSPEC;
1811         tcm->tcm__pad1 = 0;
1812         tcm->tcm__pad2 = 0;
1813         if (q) {
1814                 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1815                 tcm->tcm_parent = parent;
1816         } else {
1817                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1818                 tcm->tcm_block_index = block->index;
1819         }
1820         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1821         if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1822                 goto nla_put_failure;
1823         if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1824                 goto nla_put_failure;
1825         if (!fh) {
1826                 tcm->tcm_handle = 0;
1827         } else if (terse_dump) {
1828                 if (tp->ops->terse_dump) {
1829                         if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
1830                                                 rtnl_held) < 0)
1831                                 goto nla_put_failure;
1832                 } else {
1833                         goto cls_op_not_supp;
1834                 }
1835         } else {
1836                 if (tp->ops->dump &&
1837                     tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
1838                         goto nla_put_failure;
1839         }
1840         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1841         return skb->len;
1842
1843 out_nlmsg_trim:
1844 nla_put_failure:
1845 cls_op_not_supp:
1846         nlmsg_trim(skb, b);
1847         return -1;
1848 }
1849
1850 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1851                           struct nlmsghdr *n, struct tcf_proto *tp,
1852                           struct tcf_block *block, struct Qdisc *q,
1853                           u32 parent, void *fh, int event, bool unicast,
1854                           bool rtnl_held)
1855 {
1856         struct sk_buff *skb;
1857         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1858         int err = 0;
1859
1860         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1861         if (!skb)
1862                 return -ENOBUFS;
1863
1864         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1865                           n->nlmsg_seq, n->nlmsg_flags, event,
1866                           false, rtnl_held) <= 0) {
1867                 kfree_skb(skb);
1868                 return -EINVAL;
1869         }
1870
1871         if (unicast)
1872                 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1873         else
1874                 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1875                                      n->nlmsg_flags & NLM_F_ECHO);
1876
1877         if (err > 0)
1878                 err = 0;
1879         return err;
1880 }
1881
1882 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1883                               struct nlmsghdr *n, struct tcf_proto *tp,
1884                               struct tcf_block *block, struct Qdisc *q,
1885                               u32 parent, void *fh, bool unicast, bool *last,
1886                               bool rtnl_held, struct netlink_ext_ack *extack)
1887 {
1888         struct sk_buff *skb;
1889         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1890         int err;
1891
1892         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1893         if (!skb)
1894                 return -ENOBUFS;
1895
1896         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1897                           n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
1898                           false, rtnl_held) <= 0) {
1899                 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1900                 kfree_skb(skb);
1901                 return -EINVAL;
1902         }
1903
1904         err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
1905         if (err) {
1906                 kfree_skb(skb);
1907                 return err;
1908         }
1909
1910         if (unicast)
1911                 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1912         else
1913                 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1914                                      n->nlmsg_flags & NLM_F_ECHO);
1915         if (err < 0)
1916                 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1917
1918         if (err > 0)
1919                 err = 0;
1920         return err;
1921 }
1922
1923 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1924                                  struct tcf_block *block, struct Qdisc *q,
1925                                  u32 parent, struct nlmsghdr *n,
1926                                  struct tcf_chain *chain, int event)
1927 {
1928         struct tcf_proto *tp;
1929
1930         for (tp = tcf_get_next_proto(chain, NULL);
1931              tp; tp = tcf_get_next_proto(chain, tp))
1932                 tfilter_notify(net, oskb, n, tp, block,
1933                                q, parent, NULL, event, false, true);
1934 }
1935
1936 static void tfilter_put(struct tcf_proto *tp, void *fh)
1937 {
1938         if (tp->ops->put && fh)
1939                 tp->ops->put(tp, fh);
1940 }
1941
1942 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1943                           struct netlink_ext_ack *extack)
1944 {
1945         struct net *net = sock_net(skb->sk);
1946         struct nlattr *tca[TCA_MAX + 1];
1947         char name[IFNAMSIZ];
1948         struct tcmsg *t;
1949         u32 protocol;
1950         u32 prio;
1951         bool prio_allocate;
1952         u32 parent;
1953         u32 chain_index;
1954         struct Qdisc *q = NULL;
1955         struct tcf_chain_info chain_info;
1956         struct tcf_chain *chain = NULL;
1957         struct tcf_block *block;
1958         struct tcf_proto *tp;
1959         unsigned long cl;
1960         void *fh;
1961         int err;
1962         int tp_created;
1963         bool rtnl_held = false;
1964
1965         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1966                 return -EPERM;
1967
1968 replay:
1969         tp_created = 0;
1970
1971         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
1972                                      rtm_tca_policy, extack);
1973         if (err < 0)
1974                 return err;
1975
1976         t = nlmsg_data(n);
1977         protocol = TC_H_MIN(t->tcm_info);
1978         prio = TC_H_MAJ(t->tcm_info);
1979         prio_allocate = false;
1980         parent = t->tcm_parent;
1981         tp = NULL;
1982         cl = 0;
1983         block = NULL;
1984
1985         if (prio == 0) {
1986                 /* If no priority is provided by the user,
1987                  * we allocate one.
1988                  */
1989                 if (n->nlmsg_flags & NLM_F_CREATE) {
1990                         prio = TC_H_MAKE(0x80000000U, 0U);
1991                         prio_allocate = true;
1992                 } else {
1993                         NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1994                         return -ENOENT;
1995                 }
1996         }
1997
1998         /* Find head of filter chain. */
1999
2000         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2001         if (err)
2002                 return err;
2003
2004         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2005                 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2006                 err = -EINVAL;
2007                 goto errout;
2008         }
2009
2010         /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2011          * block is shared (no qdisc found), qdisc is not unlocked, classifier
2012          * type is not specified, classifier is not unlocked.
2013          */
2014         if (rtnl_held ||
2015             (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2016             !tcf_proto_is_unlocked(name)) {
2017                 rtnl_held = true;
2018                 rtnl_lock();
2019         }
2020
2021         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2022         if (err)
2023                 goto errout;
2024
2025         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2026                                  extack);
2027         if (IS_ERR(block)) {
2028                 err = PTR_ERR(block);
2029                 goto errout;
2030         }
2031         block->classid = parent;
2032
2033         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2034         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2035                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2036                 err = -EINVAL;
2037                 goto errout;
2038         }
2039         chain = tcf_chain_get(block, chain_index, true);
2040         if (!chain) {
2041                 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2042                 err = -ENOMEM;
2043                 goto errout;
2044         }
2045
2046         mutex_lock(&chain->filter_chain_lock);
2047         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2048                                prio, prio_allocate);
2049         if (IS_ERR(tp)) {
2050                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2051                 err = PTR_ERR(tp);
2052                 goto errout_locked;
2053         }
2054
2055         if (tp == NULL) {
2056                 struct tcf_proto *tp_new = NULL;
2057
2058                 if (chain->flushing) {
2059                         err = -EAGAIN;
2060                         goto errout_locked;
2061                 }
2062
2063                 /* Proto-tcf does not exist, create new one */
2064
2065                 if (tca[TCA_KIND] == NULL || !protocol) {
2066                         NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2067                         err = -EINVAL;
2068                         goto errout_locked;
2069                 }
2070
2071                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2072                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2073                         err = -ENOENT;
2074                         goto errout_locked;
2075                 }
2076
2077                 if (prio_allocate)
2078                         prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2079                                                                &chain_info));
2080
2081                 mutex_unlock(&chain->filter_chain_lock);
2082                 tp_new = tcf_proto_create(name, protocol, prio, chain,
2083                                           rtnl_held, extack);
2084                 if (IS_ERR(tp_new)) {
2085                         err = PTR_ERR(tp_new);
2086                         goto errout_tp;
2087                 }
2088
2089                 tp_created = 1;
2090                 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2091                                                 rtnl_held);
2092                 if (IS_ERR(tp)) {
2093                         err = PTR_ERR(tp);
2094                         goto errout_tp;
2095                 }
2096         } else {
2097                 mutex_unlock(&chain->filter_chain_lock);
2098         }
2099
2100         if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2101                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2102                 err = -EINVAL;
2103                 goto errout;
2104         }
2105
2106         fh = tp->ops->get(tp, t->tcm_handle);
2107
2108         if (!fh) {
2109                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2110                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2111                         err = -ENOENT;
2112                         goto errout;
2113                 }
2114         } else if (n->nlmsg_flags & NLM_F_EXCL) {
2115                 tfilter_put(tp, fh);
2116                 NL_SET_ERR_MSG(extack, "Filter already exists");
2117                 err = -EEXIST;
2118                 goto errout;
2119         }
2120
2121         if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2122                 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2123                 err = -EINVAL;
2124                 goto errout;
2125         }
2126
2127         err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2128                               n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
2129                               rtnl_held, extack);
2130         if (err == 0) {
2131                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2132                                RTM_NEWTFILTER, false, rtnl_held);
2133                 tfilter_put(tp, fh);
2134                 /* q pointer is NULL for shared blocks */
2135                 if (q)
2136                         q->flags &= ~TCQ_F_CAN_BYPASS;
2137         }
2138
2139 errout:
2140         if (err && tp_created)
2141                 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2142 errout_tp:
2143         if (chain) {
2144                 if (tp && !IS_ERR(tp))
2145                         tcf_proto_put(tp, rtnl_held, NULL);
2146                 if (!tp_created)
2147                         tcf_chain_put(chain);
2148         }
2149         tcf_block_release(q, block, rtnl_held);
2150
2151         if (rtnl_held)
2152                 rtnl_unlock();
2153
2154         if (err == -EAGAIN) {
2155                 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2156                  * of target chain.
2157                  */
2158                 rtnl_held = true;
2159                 /* Replay the request. */
2160                 goto replay;
2161         }
2162         return err;
2163
2164 errout_locked:
2165         mutex_unlock(&chain->filter_chain_lock);
2166         goto errout;
2167 }
2168
2169 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2170                           struct netlink_ext_ack *extack)
2171 {
2172         struct net *net = sock_net(skb->sk);
2173         struct nlattr *tca[TCA_MAX + 1];
2174         char name[IFNAMSIZ];
2175         struct tcmsg *t;
2176         u32 protocol;
2177         u32 prio;
2178         u32 parent;
2179         u32 chain_index;
2180         struct Qdisc *q = NULL;
2181         struct tcf_chain_info chain_info;
2182         struct tcf_chain *chain = NULL;
2183         struct tcf_block *block = NULL;
2184         struct tcf_proto *tp = NULL;
2185         unsigned long cl = 0;
2186         void *fh = NULL;
2187         int err;
2188         bool rtnl_held = false;
2189
2190         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2191                 return -EPERM;
2192
2193         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2194                                      rtm_tca_policy, extack);
2195         if (err < 0)
2196                 return err;
2197
2198         t = nlmsg_data(n);
2199         protocol = TC_H_MIN(t->tcm_info);
2200         prio = TC_H_MAJ(t->tcm_info);
2201         parent = t->tcm_parent;
2202
2203         if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2204                 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2205                 return -ENOENT;
2206         }
2207
2208         /* Find head of filter chain. */
2209
2210         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2211         if (err)
2212                 return err;
2213
2214         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2215                 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2216                 err = -EINVAL;
2217                 goto errout;
2218         }
2219         /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2220          * found), qdisc is not unlocked, classifier type is not specified,
2221          * classifier is not unlocked.
2222          */
2223         if (!prio ||
2224             (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2225             !tcf_proto_is_unlocked(name)) {
2226                 rtnl_held = true;
2227                 rtnl_lock();
2228         }
2229
2230         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2231         if (err)
2232                 goto errout;
2233
2234         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2235                                  extack);
2236         if (IS_ERR(block)) {
2237                 err = PTR_ERR(block);
2238                 goto errout;
2239         }
2240
2241         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2242         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2243                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2244                 err = -EINVAL;
2245                 goto errout;
2246         }
2247         chain = tcf_chain_get(block, chain_index, false);
2248         if (!chain) {
2249                 /* User requested flush on non-existent chain. Nothing to do,
2250                  * so just return success.
2251                  */
2252                 if (prio == 0) {
2253                         err = 0;
2254                         goto errout;
2255                 }
2256                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2257                 err = -ENOENT;
2258                 goto errout;
2259         }
2260
2261         if (prio == 0) {
2262                 tfilter_notify_chain(net, skb, block, q, parent, n,
2263                                      chain, RTM_DELTFILTER);
2264                 tcf_chain_flush(chain, rtnl_held);
2265                 err = 0;
2266                 goto errout;
2267         }
2268
2269         mutex_lock(&chain->filter_chain_lock);
2270         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2271                                prio, false);
2272         if (!tp || IS_ERR(tp)) {
2273                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2274                 err = tp ? PTR_ERR(tp) : -ENOENT;
2275                 goto errout_locked;
2276         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2277                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2278                 err = -EINVAL;
2279                 goto errout_locked;
2280         } else if (t->tcm_handle == 0) {
2281                 tcf_proto_signal_destroying(chain, tp);
2282                 tcf_chain_tp_remove(chain, &chain_info, tp);
2283                 mutex_unlock(&chain->filter_chain_lock);
2284
2285                 tcf_proto_put(tp, rtnl_held, NULL);
2286                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2287                                RTM_DELTFILTER, false, rtnl_held);
2288                 err = 0;
2289                 goto errout;
2290         }
2291         mutex_unlock(&chain->filter_chain_lock);
2292
2293         fh = tp->ops->get(tp, t->tcm_handle);
2294
2295         if (!fh) {
2296                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2297                 err = -ENOENT;
2298         } else {
2299                 bool last;
2300
2301                 err = tfilter_del_notify(net, skb, n, tp, block,
2302                                          q, parent, fh, false, &last,
2303                                          rtnl_held, extack);
2304
2305                 if (err)
2306                         goto errout;
2307                 if (last)
2308                         tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2309         }
2310
2311 errout:
2312         if (chain) {
2313                 if (tp && !IS_ERR(tp))
2314                         tcf_proto_put(tp, rtnl_held, NULL);
2315                 tcf_chain_put(chain);
2316         }
2317         tcf_block_release(q, block, rtnl_held);
2318
2319         if (rtnl_held)
2320                 rtnl_unlock();
2321
2322         return err;
2323
2324 errout_locked:
2325         mutex_unlock(&chain->filter_chain_lock);
2326         goto errout;
2327 }
2328
2329 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2330                           struct netlink_ext_ack *extack)
2331 {
2332         struct net *net = sock_net(skb->sk);
2333         struct nlattr *tca[TCA_MAX + 1];
2334         char name[IFNAMSIZ];
2335         struct tcmsg *t;
2336         u32 protocol;
2337         u32 prio;
2338         u32 parent;
2339         u32 chain_index;
2340         struct Qdisc *q = NULL;
2341         struct tcf_chain_info chain_info;
2342         struct tcf_chain *chain = NULL;
2343         struct tcf_block *block = NULL;
2344         struct tcf_proto *tp = NULL;
2345         unsigned long cl = 0;
2346         void *fh = NULL;
2347         int err;
2348         bool rtnl_held = false;
2349
2350         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2351                                      rtm_tca_policy, extack);
2352         if (err < 0)
2353                 return err;
2354
2355         t = nlmsg_data(n);
2356         protocol = TC_H_MIN(t->tcm_info);
2357         prio = TC_H_MAJ(t->tcm_info);
2358         parent = t->tcm_parent;
2359
2360         if (prio == 0) {
2361                 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2362                 return -ENOENT;
2363         }
2364
2365         /* Find head of filter chain. */
2366
2367         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2368         if (err)
2369                 return err;
2370
2371         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2372                 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2373                 err = -EINVAL;
2374                 goto errout;
2375         }
2376         /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2377          * unlocked, classifier type is not specified, classifier is not
2378          * unlocked.
2379          */
2380         if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2381             !tcf_proto_is_unlocked(name)) {
2382                 rtnl_held = true;
2383                 rtnl_lock();
2384         }
2385
2386         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2387         if (err)
2388                 goto errout;
2389
2390         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2391                                  extack);
2392         if (IS_ERR(block)) {
2393                 err = PTR_ERR(block);
2394                 goto errout;
2395         }
2396
2397         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2398         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2399                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2400                 err = -EINVAL;
2401                 goto errout;
2402         }
2403         chain = tcf_chain_get(block, chain_index, false);
2404         if (!chain) {
2405                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2406                 err = -EINVAL;
2407                 goto errout;
2408         }
2409
2410         mutex_lock(&chain->filter_chain_lock);
2411         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2412                                prio, false);
2413         mutex_unlock(&chain->filter_chain_lock);
2414         if (!tp || IS_ERR(tp)) {
2415                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2416                 err = tp ? PTR_ERR(tp) : -ENOENT;
2417                 goto errout;
2418         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2419                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2420                 err = -EINVAL;
2421                 goto errout;
2422         }
2423
2424         fh = tp->ops->get(tp, t->tcm_handle);
2425
2426         if (!fh) {
2427                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2428                 err = -ENOENT;
2429         } else {
2430                 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2431                                      fh, RTM_NEWTFILTER, true, rtnl_held);
2432                 if (err < 0)
2433                         NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2434         }
2435
2436         tfilter_put(tp, fh);
2437 errout:
2438         if (chain) {
2439                 if (tp && !IS_ERR(tp))
2440                         tcf_proto_put(tp, rtnl_held, NULL);
2441                 tcf_chain_put(chain);
2442         }
2443         tcf_block_release(q, block, rtnl_held);
2444
2445         if (rtnl_held)
2446                 rtnl_unlock();
2447
2448         return err;
2449 }
2450
2451 struct tcf_dump_args {
2452         struct tcf_walker w;
2453         struct sk_buff *skb;
2454         struct netlink_callback *cb;
2455         struct tcf_block *block;
2456         struct Qdisc *q;
2457         u32 parent;
2458         bool terse_dump;
2459 };
2460
2461 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2462 {
2463         struct tcf_dump_args *a = (void *)arg;
2464         struct net *net = sock_net(a->skb->sk);
2465
2466         return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2467                              n, NETLINK_CB(a->cb->skb).portid,
2468                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2469                              RTM_NEWTFILTER, a->terse_dump, true);
2470 }
2471
2472 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2473                            struct sk_buff *skb, struct netlink_callback *cb,
2474                            long index_start, long *p_index, bool terse)
2475 {
2476         struct net *net = sock_net(skb->sk);
2477         struct tcf_block *block = chain->block;
2478         struct tcmsg *tcm = nlmsg_data(cb->nlh);
2479         struct tcf_proto *tp, *tp_prev;
2480         struct tcf_dump_args arg;
2481
2482         for (tp = __tcf_get_next_proto(chain, NULL);
2483              tp;
2484              tp_prev = tp,
2485                      tp = __tcf_get_next_proto(chain, tp),
2486                      tcf_proto_put(tp_prev, true, NULL),
2487                      (*p_index)++) {
2488                 if (*p_index < index_start)
2489                         continue;
2490                 if (TC_H_MAJ(tcm->tcm_info) &&
2491                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
2492                         continue;
2493                 if (TC_H_MIN(tcm->tcm_info) &&
2494                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
2495                         continue;
2496                 if (*p_index > index_start)
2497                         memset(&cb->args[1], 0,
2498                                sizeof(cb->args) - sizeof(cb->args[0]));
2499                 if (cb->args[1] == 0) {
2500                         if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2501                                           NETLINK_CB(cb->skb).portid,
2502                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
2503                                           RTM_NEWTFILTER, false, true) <= 0)
2504                                 goto errout;
2505                         cb->args[1] = 1;
2506                 }
2507                 if (!tp->ops->walk)
2508                         continue;
2509                 arg.w.fn = tcf_node_dump;
2510                 arg.skb = skb;
2511                 arg.cb = cb;
2512                 arg.block = block;
2513                 arg.q = q;
2514                 arg.parent = parent;
2515                 arg.w.stop = 0;
2516                 arg.w.skip = cb->args[1] - 1;
2517                 arg.w.count = 0;
2518                 arg.w.cookie = cb->args[2];
2519                 arg.terse_dump = terse;
2520                 tp->ops->walk(tp, &arg.w, true);
2521                 cb->args[2] = arg.w.cookie;
2522                 cb->args[1] = arg.w.count + 1;
2523                 if (arg.w.stop)
2524                         goto errout;
2525         }
2526         return true;
2527
2528 errout:
2529         tcf_proto_put(tp, true, NULL);
2530         return false;
2531 }
2532
2533 static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2534         [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2535 };
2536
2537 /* called with RTNL */
2538 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2539 {
2540         struct tcf_chain *chain, *chain_prev;
2541         struct net *net = sock_net(skb->sk);
2542         struct nlattr *tca[TCA_MAX + 1];
2543         struct Qdisc *q = NULL;
2544         struct tcf_block *block;
2545         struct tcmsg *tcm = nlmsg_data(cb->nlh);
2546         bool terse_dump = false;
2547         long index_start;
2548         long index;
2549         u32 parent;
2550         int err;
2551
2552         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2553                 return skb->len;
2554
2555         err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2556                                      tcf_tfilter_dump_policy, cb->extack);
2557         if (err)
2558                 return err;
2559
2560         if (tca[TCA_DUMP_FLAGS]) {
2561                 struct nla_bitfield32 flags =
2562                         nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2563
2564                 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2565         }
2566
2567         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2568                 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2569                 if (!block)
2570                         goto out;
2571                 /* If we work with block index, q is NULL and parent value
2572                  * will never be used in the following code. The check
2573                  * in tcf_fill_node prevents it. However, compiler does not
2574                  * see that far, so set parent to zero to silence the warning
2575                  * about parent being uninitialized.
2576                  */
2577                 parent = 0;
2578         } else {
2579                 const struct Qdisc_class_ops *cops;
2580                 struct net_device *dev;
2581                 unsigned long cl = 0;
2582
2583                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2584                 if (!dev)
2585                         return skb->len;
2586
2587                 parent = tcm->tcm_parent;
2588                 if (!parent)
2589                         q = dev->qdisc;
2590                 else
2591                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2592                 if (!q)
2593                         goto out;
2594                 cops = q->ops->cl_ops;
2595                 if (!cops)
2596                         goto out;
2597                 if (!cops->tcf_block)
2598                         goto out;
2599                 if (TC_H_MIN(tcm->tcm_parent)) {
2600                         cl = cops->find(q, tcm->tcm_parent);
2601                         if (cl == 0)
2602                                 goto out;
2603                 }
2604                 block = cops->tcf_block(q, cl, NULL);
2605                 if (!block)
2606                         goto out;
2607                 parent = block->classid;
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, terse_dump)) {
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         char name[IFNAMSIZ];
2754         void *tmplt_priv;
2755
2756         /* If kind is not set, user did not specify template. */
2757         if (!tca[TCA_KIND])
2758                 return 0;
2759
2760         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2761                 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
2762                 return -EINVAL;
2763         }
2764
2765         ops = tcf_proto_lookup_ops(name, true, extack);
2766         if (IS_ERR(ops))
2767                 return PTR_ERR(ops);
2768         if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2769                 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2770                 return -EOPNOTSUPP;
2771         }
2772
2773         tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2774         if (IS_ERR(tmplt_priv)) {
2775                 module_put(ops->owner);
2776                 return PTR_ERR(tmplt_priv);
2777         }
2778         chain->tmplt_ops = ops;
2779         chain->tmplt_priv = tmplt_priv;
2780         return 0;
2781 }
2782
2783 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2784                                void *tmplt_priv)
2785 {
2786         /* If template ops are set, no work to do for us. */
2787         if (!tmplt_ops)
2788                 return;
2789
2790         tmplt_ops->tmplt_destroy(tmplt_priv);
2791         module_put(tmplt_ops->owner);
2792 }
2793
2794 /* Add/delete/get a chain */
2795
2796 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2797                         struct netlink_ext_ack *extack)
2798 {
2799         struct net *net = sock_net(skb->sk);
2800         struct nlattr *tca[TCA_MAX + 1];
2801         struct tcmsg *t;
2802         u32 parent;
2803         u32 chain_index;
2804         struct Qdisc *q = NULL;
2805         struct tcf_chain *chain = NULL;
2806         struct tcf_block *block;
2807         unsigned long cl;
2808         int err;
2809
2810         if (n->nlmsg_type != RTM_GETCHAIN &&
2811             !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2812                 return -EPERM;
2813
2814 replay:
2815         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2816                                      rtm_tca_policy, extack);
2817         if (err < 0)
2818                 return err;
2819
2820         t = nlmsg_data(n);
2821         parent = t->tcm_parent;
2822         cl = 0;
2823
2824         block = tcf_block_find(net, &q, &parent, &cl,
2825                                t->tcm_ifindex, t->tcm_block_index, extack);
2826         if (IS_ERR(block))
2827                 return PTR_ERR(block);
2828
2829         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2830         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2831                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2832                 err = -EINVAL;
2833                 goto errout_block;
2834         }
2835
2836         mutex_lock(&block->lock);
2837         chain = tcf_chain_lookup(block, chain_index);
2838         if (n->nlmsg_type == RTM_NEWCHAIN) {
2839                 if (chain) {
2840                         if (tcf_chain_held_by_acts_only(chain)) {
2841                                 /* The chain exists only because there is
2842                                  * some action referencing it.
2843                                  */
2844                                 tcf_chain_hold(chain);
2845                         } else {
2846                                 NL_SET_ERR_MSG(extack, "Filter chain already exists");
2847                                 err = -EEXIST;
2848                                 goto errout_block_locked;
2849                         }
2850                 } else {
2851                         if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2852                                 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
2853                                 err = -ENOENT;
2854                                 goto errout_block_locked;
2855                         }
2856                         chain = tcf_chain_create(block, chain_index);
2857                         if (!chain) {
2858                                 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
2859                                 err = -ENOMEM;
2860                                 goto errout_block_locked;
2861                         }
2862                 }
2863         } else {
2864                 if (!chain || tcf_chain_held_by_acts_only(chain)) {
2865                         NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2866                         err = -EINVAL;
2867                         goto errout_block_locked;
2868                 }
2869                 tcf_chain_hold(chain);
2870         }
2871
2872         if (n->nlmsg_type == RTM_NEWCHAIN) {
2873                 /* Modifying chain requires holding parent block lock. In case
2874                  * the chain was successfully added, take a reference to the
2875                  * chain. This ensures that an empty chain does not disappear at
2876                  * the end of this function.
2877                  */
2878                 tcf_chain_hold(chain);
2879                 chain->explicitly_created = true;
2880         }
2881         mutex_unlock(&block->lock);
2882
2883         switch (n->nlmsg_type) {
2884         case RTM_NEWCHAIN:
2885                 err = tc_chain_tmplt_add(chain, net, tca, extack);
2886                 if (err) {
2887                         tcf_chain_put_explicitly_created(chain);
2888                         goto errout;
2889                 }
2890
2891                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2892                                 RTM_NEWCHAIN, false);
2893                 break;
2894         case RTM_DELCHAIN:
2895                 tfilter_notify_chain(net, skb, block, q, parent, n,
2896                                      chain, RTM_DELTFILTER);
2897                 /* Flush the chain first as the user requested chain removal. */
2898                 tcf_chain_flush(chain, true);
2899                 /* In case the chain was successfully deleted, put a reference
2900                  * to the chain previously taken during addition.
2901                  */
2902                 tcf_chain_put_explicitly_created(chain);
2903                 break;
2904         case RTM_GETCHAIN:
2905                 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
2906                                       n->nlmsg_seq, n->nlmsg_type, true);
2907                 if (err < 0)
2908                         NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2909                 break;
2910         default:
2911                 err = -EOPNOTSUPP;
2912                 NL_SET_ERR_MSG(extack, "Unsupported message type");
2913                 goto errout;
2914         }
2915
2916 errout:
2917         tcf_chain_put(chain);
2918 errout_block:
2919         tcf_block_release(q, block, true);
2920         if (err == -EAGAIN)
2921                 /* Replay the request. */
2922                 goto replay;
2923         return err;
2924
2925 errout_block_locked:
2926         mutex_unlock(&block->lock);
2927         goto errout_block;
2928 }
2929
2930 /* called with RTNL */
2931 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2932 {
2933         struct net *net = sock_net(skb->sk);
2934         struct nlattr *tca[TCA_MAX + 1];
2935         struct Qdisc *q = NULL;
2936         struct tcf_block *block;
2937         struct tcmsg *tcm = nlmsg_data(cb->nlh);
2938         struct tcf_chain *chain;
2939         long index_start;
2940         long index;
2941         int err;
2942
2943         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2944                 return skb->len;
2945
2946         err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2947                                      rtm_tca_policy, cb->extack);
2948         if (err)
2949                 return err;
2950
2951         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2952                 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2953                 if (!block)
2954                         goto out;
2955         } else {
2956                 const struct Qdisc_class_ops *cops;
2957                 struct net_device *dev;
2958                 unsigned long cl = 0;
2959
2960                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2961                 if (!dev)
2962                         return skb->len;
2963
2964                 if (!tcm->tcm_parent)
2965                         q = dev->qdisc;
2966                 else
2967                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2968
2969                 if (!q)
2970                         goto out;
2971                 cops = q->ops->cl_ops;
2972                 if (!cops)
2973                         goto out;
2974                 if (!cops->tcf_block)
2975                         goto out;
2976                 if (TC_H_MIN(tcm->tcm_parent)) {
2977                         cl = cops->find(q, tcm->tcm_parent);
2978                         if (cl == 0)
2979                                 goto out;
2980                 }
2981                 block = cops->tcf_block(q, cl, NULL);
2982                 if (!block)
2983                         goto out;
2984                 if (tcf_block_shared(block))
2985                         q = NULL;
2986         }
2987
2988         index_start = cb->args[0];
2989         index = 0;
2990
2991         mutex_lock(&block->lock);
2992         list_for_each_entry(chain, &block->chain_list, list) {
2993                 if ((tca[TCA_CHAIN] &&
2994                      nla_get_u32(tca[TCA_CHAIN]) != chain->index))
2995                         continue;
2996                 if (index < index_start) {
2997                         index++;
2998                         continue;
2999                 }
3000                 if (tcf_chain_held_by_acts_only(chain))
3001                         continue;
3002                 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3003                                          chain->index, net, skb, block,
3004                                          NETLINK_CB(cb->skb).portid,
3005                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
3006                                          RTM_NEWCHAIN);
3007                 if (err <= 0)
3008                         break;
3009                 index++;
3010         }
3011         mutex_unlock(&block->lock);
3012
3013         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3014                 tcf_block_refcnt_put(block, true);
3015         cb->args[0] = index;
3016
3017 out:
3018         /* If we did no progress, the error (EMSGSIZE) is real */
3019         if (skb->len == 0 && err)
3020                 return err;
3021         return skb->len;
3022 }
3023
3024 void tcf_exts_destroy(struct tcf_exts *exts)
3025 {
3026 #ifdef CONFIG_NET_CLS_ACT
3027         if (exts->actions) {
3028                 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3029                 kfree(exts->actions);
3030         }
3031         exts->nr_actions = 0;
3032 #endif
3033 }
3034 EXPORT_SYMBOL(tcf_exts_destroy);
3035
3036 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3037                       struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
3038                       bool rtnl_held, struct netlink_ext_ack *extack)
3039 {
3040 #ifdef CONFIG_NET_CLS_ACT
3041         {
3042                 struct tc_action *act;
3043                 size_t attr_size = 0;
3044
3045                 if (exts->police && tb[exts->police]) {
3046                         act = tcf_action_init_1(net, tp, tb[exts->police],
3047                                                 rate_tlv, "police", ovr,
3048                                                 TCA_ACT_BIND, rtnl_held,
3049                                                 extack);
3050                         if (IS_ERR(act))
3051                                 return PTR_ERR(act);
3052
3053                         act->type = exts->type = TCA_OLD_COMPAT;
3054                         exts->actions[0] = act;
3055                         exts->nr_actions = 1;
3056                 } else if (exts->action && tb[exts->action]) {
3057                         int err;
3058
3059                         err = tcf_action_init(net, tp, tb[exts->action],
3060                                               rate_tlv, NULL, ovr, TCA_ACT_BIND,
3061                                               exts->actions, &attr_size,
3062                                               rtnl_held, extack);
3063                         if (err < 0)
3064                                 return err;
3065                         exts->nr_actions = err;
3066                 }
3067         }
3068 #else
3069         if ((exts->action && tb[exts->action]) ||
3070             (exts->police && tb[exts->police])) {
3071                 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3072                 return -EOPNOTSUPP;
3073         }
3074 #endif
3075
3076         return 0;
3077 }
3078 EXPORT_SYMBOL(tcf_exts_validate);
3079
3080 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3081 {
3082 #ifdef CONFIG_NET_CLS_ACT
3083         struct tcf_exts old = *dst;
3084
3085         *dst = *src;
3086         tcf_exts_destroy(&old);
3087 #endif
3088 }
3089 EXPORT_SYMBOL(tcf_exts_change);
3090
3091 #ifdef CONFIG_NET_CLS_ACT
3092 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3093 {
3094         if (exts->nr_actions == 0)
3095                 return NULL;
3096         else
3097                 return exts->actions[0];
3098 }
3099 #endif
3100
3101 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3102 {
3103 #ifdef CONFIG_NET_CLS_ACT
3104         struct nlattr *nest;
3105
3106         if (exts->action && tcf_exts_has_actions(exts)) {
3107                 /*
3108                  * again for backward compatible mode - we want
3109                  * to work with both old and new modes of entering
3110                  * tc data even if iproute2  was newer - jhs
3111                  */
3112                 if (exts->type != TCA_OLD_COMPAT) {
3113                         nest = nla_nest_start_noflag(skb, exts->action);
3114                         if (nest == NULL)
3115                                 goto nla_put_failure;
3116
3117                         if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3118                             < 0)
3119                                 goto nla_put_failure;
3120                         nla_nest_end(skb, nest);
3121                 } else if (exts->police) {
3122                         struct tc_action *act = tcf_exts_first_act(exts);
3123                         nest = nla_nest_start_noflag(skb, exts->police);
3124                         if (nest == NULL || !act)
3125                                 goto nla_put_failure;
3126                         if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3127                                 goto nla_put_failure;
3128                         nla_nest_end(skb, nest);
3129                 }
3130         }
3131         return 0;
3132
3133 nla_put_failure:
3134         nla_nest_cancel(skb, nest);
3135         return -1;
3136 #else
3137         return 0;
3138 #endif
3139 }
3140 EXPORT_SYMBOL(tcf_exts_dump);
3141
3142 int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3143 {
3144 #ifdef CONFIG_NET_CLS_ACT
3145         struct nlattr *nest;
3146
3147         if (!exts->action || !tcf_exts_has_actions(exts))
3148                 return 0;
3149
3150         nest = nla_nest_start_noflag(skb, exts->action);
3151         if (!nest)
3152                 goto nla_put_failure;
3153
3154         if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3155                 goto nla_put_failure;
3156         nla_nest_end(skb, nest);
3157         return 0;
3158
3159 nla_put_failure:
3160         nla_nest_cancel(skb, nest);
3161         return -1;
3162 #else
3163         return 0;
3164 #endif
3165 }
3166 EXPORT_SYMBOL(tcf_exts_terse_dump);
3167
3168 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3169 {
3170 #ifdef CONFIG_NET_CLS_ACT
3171         struct tc_action *a = tcf_exts_first_act(exts);
3172         if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3173                 return -1;
3174 #endif
3175         return 0;
3176 }
3177 EXPORT_SYMBOL(tcf_exts_dump_stats);
3178
3179 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3180 {
3181         if (*flags & TCA_CLS_FLAGS_IN_HW)
3182                 return;
3183         *flags |= TCA_CLS_FLAGS_IN_HW;
3184         atomic_inc(&block->offloadcnt);
3185 }
3186
3187 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3188 {
3189         if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3190                 return;
3191         *flags &= ~TCA_CLS_FLAGS_IN_HW;
3192         atomic_dec(&block->offloadcnt);
3193 }
3194
3195 static void tc_cls_offload_cnt_update(struct tcf_block *block,
3196                                       struct tcf_proto *tp, u32 *cnt,
3197                                       u32 *flags, u32 diff, bool add)
3198 {
3199         lockdep_assert_held(&block->cb_lock);
3200
3201         spin_lock(&tp->lock);
3202         if (add) {
3203                 if (!*cnt)
3204                         tcf_block_offload_inc(block, flags);
3205                 *cnt += diff;
3206         } else {
3207                 *cnt -= diff;
3208                 if (!*cnt)
3209                         tcf_block_offload_dec(block, flags);
3210         }
3211         spin_unlock(&tp->lock);
3212 }
3213
3214 static void
3215 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3216                          u32 *cnt, u32 *flags)
3217 {
3218         lockdep_assert_held(&block->cb_lock);
3219
3220         spin_lock(&tp->lock);
3221         tcf_block_offload_dec(block, flags);
3222         *cnt = 0;
3223         spin_unlock(&tp->lock);
3224 }
3225
3226 static int
3227 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3228                    void *type_data, bool err_stop)
3229 {
3230         struct flow_block_cb *block_cb;
3231         int ok_count = 0;
3232         int err;
3233
3234         list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3235                 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3236                 if (err) {
3237                         if (err_stop)
3238                                 return err;
3239                 } else {
3240                         ok_count++;
3241                 }
3242         }
3243         return ok_count;
3244 }
3245
3246 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3247                      void *type_data, bool err_stop, bool rtnl_held)
3248 {
3249         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3250         int ok_count;
3251
3252 retry:
3253         if (take_rtnl)
3254                 rtnl_lock();
3255         down_read(&block->cb_lock);
3256         /* Need to obtain rtnl lock if block is bound to devs that require it.
3257          * In block bind code cb_lock is obtained while holding rtnl, so we must
3258          * obtain the locks in same order here.
3259          */
3260         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3261                 up_read(&block->cb_lock);
3262                 take_rtnl = true;
3263                 goto retry;
3264         }
3265
3266         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3267
3268         up_read(&block->cb_lock);
3269         if (take_rtnl)
3270                 rtnl_unlock();
3271         return ok_count;
3272 }
3273 EXPORT_SYMBOL(tc_setup_cb_call);
3274
3275 /* Non-destructive filter add. If filter that wasn't already in hardware is
3276  * successfully offloaded, increment block offloads counter. On failure,
3277  * previously offloaded filter is considered to be intact and offloads counter
3278  * is not decremented.
3279  */
3280
3281 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3282                     enum tc_setup_type type, void *type_data, bool err_stop,
3283                     u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3284 {
3285         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3286         int ok_count;
3287
3288 retry:
3289         if (take_rtnl)
3290                 rtnl_lock();
3291         down_read(&block->cb_lock);
3292         /* Need to obtain rtnl lock if block is bound to devs that require it.
3293          * In block bind code cb_lock is obtained while holding rtnl, so we must
3294          * obtain the locks in same order here.
3295          */
3296         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3297                 up_read(&block->cb_lock);
3298                 take_rtnl = true;
3299                 goto retry;
3300         }
3301
3302         /* Make sure all netdevs sharing this block are offload-capable. */
3303         if (block->nooffloaddevcnt && err_stop) {
3304                 ok_count = -EOPNOTSUPP;
3305                 goto err_unlock;
3306         }
3307
3308         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3309         if (ok_count < 0)
3310                 goto err_unlock;
3311
3312         if (tp->ops->hw_add)
3313                 tp->ops->hw_add(tp, type_data);
3314         if (ok_count > 0)
3315                 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3316                                           ok_count, true);
3317 err_unlock:
3318         up_read(&block->cb_lock);
3319         if (take_rtnl)
3320                 rtnl_unlock();
3321         return ok_count < 0 ? ok_count : 0;
3322 }
3323 EXPORT_SYMBOL(tc_setup_cb_add);
3324
3325 /* Destructive filter replace. If filter that wasn't already in hardware is
3326  * successfully offloaded, increment block offload counter. On failure,
3327  * previously offloaded filter is considered to be destroyed and offload counter
3328  * is decremented.
3329  */
3330
3331 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3332                         enum tc_setup_type type, void *type_data, bool err_stop,
3333                         u32 *old_flags, unsigned int *old_in_hw_count,
3334                         u32 *new_flags, unsigned int *new_in_hw_count,
3335                         bool rtnl_held)
3336 {
3337         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3338         int ok_count;
3339
3340 retry:
3341         if (take_rtnl)
3342                 rtnl_lock();
3343         down_read(&block->cb_lock);
3344         /* Need to obtain rtnl lock if block is bound to devs that require it.
3345          * In block bind code cb_lock is obtained while holding rtnl, so we must
3346          * obtain the locks in same order here.
3347          */
3348         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3349                 up_read(&block->cb_lock);
3350                 take_rtnl = true;
3351                 goto retry;
3352         }
3353
3354         /* Make sure all netdevs sharing this block are offload-capable. */
3355         if (block->nooffloaddevcnt && err_stop) {
3356                 ok_count = -EOPNOTSUPP;
3357                 goto err_unlock;
3358         }
3359
3360         tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3361         if (tp->ops->hw_del)
3362                 tp->ops->hw_del(tp, type_data);
3363
3364         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3365         if (ok_count < 0)
3366                 goto err_unlock;
3367
3368         if (tp->ops->hw_add)
3369                 tp->ops->hw_add(tp, type_data);
3370         if (ok_count > 0)
3371                 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3372                                           new_flags, ok_count, true);
3373 err_unlock:
3374         up_read(&block->cb_lock);
3375         if (take_rtnl)
3376                 rtnl_unlock();
3377         return ok_count < 0 ? ok_count : 0;
3378 }
3379 EXPORT_SYMBOL(tc_setup_cb_replace);
3380
3381 /* Destroy filter and decrement block offload counter, if filter was previously
3382  * offloaded.
3383  */
3384
3385 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3386                         enum tc_setup_type type, void *type_data, bool err_stop,
3387                         u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3388 {
3389         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3390         int ok_count;
3391
3392 retry:
3393         if (take_rtnl)
3394                 rtnl_lock();
3395         down_read(&block->cb_lock);
3396         /* Need to obtain rtnl lock if block is bound to devs that require it.
3397          * In block bind code cb_lock is obtained while holding rtnl, so we must
3398          * obtain the locks in same order here.
3399          */
3400         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3401                 up_read(&block->cb_lock);
3402                 take_rtnl = true;
3403                 goto retry;
3404         }
3405
3406         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3407
3408         tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3409         if (tp->ops->hw_del)
3410                 tp->ops->hw_del(tp, type_data);
3411
3412         up_read(&block->cb_lock);
3413         if (take_rtnl)
3414                 rtnl_unlock();
3415         return ok_count < 0 ? ok_count : 0;
3416 }
3417 EXPORT_SYMBOL(tc_setup_cb_destroy);
3418
3419 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3420                           bool add, flow_setup_cb_t *cb,
3421                           enum tc_setup_type type, void *type_data,
3422                           void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3423 {
3424         int err = cb(type, type_data, cb_priv);
3425
3426         if (err) {
3427                 if (add && tc_skip_sw(*flags))
3428                         return err;
3429         } else {
3430                 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3431                                           add);
3432         }
3433
3434         return 0;
3435 }
3436 EXPORT_SYMBOL(tc_setup_cb_reoffload);
3437
3438 static int tcf_act_get_cookie(struct flow_action_entry *entry,
3439                               const struct tc_action *act)
3440 {
3441         struct tc_cookie *cookie;
3442         int err = 0;
3443
3444         rcu_read_lock();
3445         cookie = rcu_dereference(act->act_cookie);
3446         if (cookie) {
3447                 entry->cookie = flow_action_cookie_create(cookie->data,
3448                                                           cookie->len,
3449                                                           GFP_ATOMIC);
3450                 if (!entry->cookie)
3451                         err = -ENOMEM;
3452         }
3453         rcu_read_unlock();
3454         return err;
3455 }
3456
3457 static void tcf_act_put_cookie(struct flow_action_entry *entry)
3458 {
3459         flow_action_cookie_destroy(entry->cookie);
3460 }
3461
3462 void tc_cleanup_flow_action(struct flow_action *flow_action)
3463 {
3464         struct flow_action_entry *entry;
3465         int i;
3466
3467         flow_action_for_each(i, entry, flow_action) {
3468                 tcf_act_put_cookie(entry);
3469                 if (entry->destructor)
3470                         entry->destructor(entry->destructor_priv);
3471         }
3472 }
3473 EXPORT_SYMBOL(tc_cleanup_flow_action);
3474
3475 static void tcf_mirred_get_dev(struct flow_action_entry *entry,
3476                                const struct tc_action *act)
3477 {
3478 #ifdef CONFIG_NET_CLS_ACT
3479         entry->dev = act->ops->get_dev(act, &entry->destructor);
3480         if (!entry->dev)
3481                 return;
3482         entry->destructor_priv = entry->dev;
3483 #endif
3484 }
3485
3486 static void tcf_tunnel_encap_put_tunnel(void *priv)
3487 {
3488         struct ip_tunnel_info *tunnel = priv;
3489
3490         kfree(tunnel);
3491 }
3492
3493 static int tcf_tunnel_encap_get_tunnel(struct flow_action_entry *entry,
3494                                        const struct tc_action *act)
3495 {
3496         entry->tunnel = tcf_tunnel_info_copy(act);
3497         if (!entry->tunnel)
3498                 return -ENOMEM;
3499         entry->destructor = tcf_tunnel_encap_put_tunnel;
3500         entry->destructor_priv = entry->tunnel;
3501         return 0;
3502 }
3503
3504 static void tcf_sample_get_group(struct flow_action_entry *entry,
3505                                  const struct tc_action *act)
3506 {
3507 #ifdef CONFIG_NET_CLS_ACT
3508         entry->sample.psample_group =
3509                 act->ops->get_psample_group(act, &entry->destructor);
3510         entry->destructor_priv = entry->sample.psample_group;
3511 #endif
3512 }
3513
3514 static void tcf_gate_entry_destructor(void *priv)
3515 {
3516         struct action_gate_entry *oe = priv;
3517
3518         kfree(oe);
3519 }
3520
3521 static int tcf_gate_get_entries(struct flow_action_entry *entry,
3522                                 const struct tc_action *act)
3523 {
3524         entry->gate.entries = tcf_gate_get_list(act);
3525
3526         if (!entry->gate.entries)
3527                 return -EINVAL;
3528
3529         entry->destructor = tcf_gate_entry_destructor;
3530         entry->destructor_priv = entry->gate.entries;
3531
3532         return 0;
3533 }
3534
3535 static enum flow_action_hw_stats tc_act_hw_stats(u8 hw_stats)
3536 {
3537         if (WARN_ON_ONCE(hw_stats > TCA_ACT_HW_STATS_ANY))
3538                 return FLOW_ACTION_HW_STATS_DONT_CARE;
3539         else if (!hw_stats)
3540                 return FLOW_ACTION_HW_STATS_DISABLED;
3541
3542         return hw_stats;
3543 }
3544
3545 int tc_setup_flow_action(struct flow_action *flow_action,
3546                          const struct tcf_exts *exts)
3547 {
3548         struct tc_action *act;
3549         int i, j, k, err = 0;
3550
3551         BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3552         BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3553         BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3554
3555         if (!exts)
3556                 return 0;
3557
3558         j = 0;
3559         tcf_exts_for_each_action(i, act, exts) {
3560                 struct flow_action_entry *entry;
3561
3562                 entry = &flow_action->entries[j];
3563                 spin_lock_bh(&act->tcfa_lock);
3564                 err = tcf_act_get_cookie(entry, act);
3565                 if (err)
3566                         goto err_out_locked;
3567
3568                 entry->hw_stats = tc_act_hw_stats(act->hw_stats);
3569
3570                 if (is_tcf_gact_ok(act)) {
3571                         entry->id = FLOW_ACTION_ACCEPT;
3572                 } else if (is_tcf_gact_shot(act)) {
3573                         entry->id = FLOW_ACTION_DROP;
3574                 } else if (is_tcf_gact_trap(act)) {
3575                         entry->id = FLOW_ACTION_TRAP;
3576                 } else if (is_tcf_gact_goto_chain(act)) {
3577                         entry->id = FLOW_ACTION_GOTO;
3578                         entry->chain_index = tcf_gact_goto_chain_index(act);
3579                 } else if (is_tcf_mirred_egress_redirect(act)) {
3580                         entry->id = FLOW_ACTION_REDIRECT;
3581                         tcf_mirred_get_dev(entry, act);
3582                 } else if (is_tcf_mirred_egress_mirror(act)) {
3583                         entry->id = FLOW_ACTION_MIRRED;
3584                         tcf_mirred_get_dev(entry, act);
3585                 } else if (is_tcf_mirred_ingress_redirect(act)) {
3586                         entry->id = FLOW_ACTION_REDIRECT_INGRESS;
3587                         tcf_mirred_get_dev(entry, act);
3588                 } else if (is_tcf_mirred_ingress_mirror(act)) {
3589                         entry->id = FLOW_ACTION_MIRRED_INGRESS;
3590                         tcf_mirred_get_dev(entry, act);
3591                 } else if (is_tcf_vlan(act)) {
3592                         switch (tcf_vlan_action(act)) {
3593                         case TCA_VLAN_ACT_PUSH:
3594                                 entry->id = FLOW_ACTION_VLAN_PUSH;
3595                                 entry->vlan.vid = tcf_vlan_push_vid(act);
3596                                 entry->vlan.proto = tcf_vlan_push_proto(act);
3597                                 entry->vlan.prio = tcf_vlan_push_prio(act);
3598                                 break;
3599                         case TCA_VLAN_ACT_POP:
3600                                 entry->id = FLOW_ACTION_VLAN_POP;
3601                                 break;
3602                         case TCA_VLAN_ACT_MODIFY:
3603                                 entry->id = FLOW_ACTION_VLAN_MANGLE;
3604                                 entry->vlan.vid = tcf_vlan_push_vid(act);
3605                                 entry->vlan.proto = tcf_vlan_push_proto(act);
3606                                 entry->vlan.prio = tcf_vlan_push_prio(act);
3607                                 break;
3608                         default:
3609                                 err = -EOPNOTSUPP;
3610                                 goto err_out_locked;
3611                         }
3612                 } else if (is_tcf_tunnel_set(act)) {
3613                         entry->id = FLOW_ACTION_TUNNEL_ENCAP;
3614                         err = tcf_tunnel_encap_get_tunnel(entry, act);
3615                         if (err)
3616                                 goto err_out_locked;
3617                 } else if (is_tcf_tunnel_release(act)) {
3618                         entry->id = FLOW_ACTION_TUNNEL_DECAP;
3619                 } else if (is_tcf_pedit(act)) {
3620                         for (k = 0; k < tcf_pedit_nkeys(act); k++) {
3621                                 switch (tcf_pedit_cmd(act, k)) {
3622                                 case TCA_PEDIT_KEY_EX_CMD_SET:
3623                                         entry->id = FLOW_ACTION_MANGLE;
3624                                         break;
3625                                 case TCA_PEDIT_KEY_EX_CMD_ADD:
3626                                         entry->id = FLOW_ACTION_ADD;
3627                                         break;
3628                                 default:
3629                                         err = -EOPNOTSUPP;
3630                                         goto err_out_locked;
3631                                 }
3632                                 entry->mangle.htype = tcf_pedit_htype(act, k);
3633                                 entry->mangle.mask = tcf_pedit_mask(act, k);
3634                                 entry->mangle.val = tcf_pedit_val(act, k);
3635                                 entry->mangle.offset = tcf_pedit_offset(act, k);
3636                                 entry->hw_stats = tc_act_hw_stats(act->hw_stats);
3637                                 entry = &flow_action->entries[++j];
3638                         }
3639                 } else if (is_tcf_csum(act)) {
3640                         entry->id = FLOW_ACTION_CSUM;
3641                         entry->csum_flags = tcf_csum_update_flags(act);
3642                 } else if (is_tcf_skbedit_mark(act)) {
3643                         entry->id = FLOW_ACTION_MARK;
3644                         entry->mark = tcf_skbedit_mark(act);
3645                 } else if (is_tcf_sample(act)) {
3646                         entry->id = FLOW_ACTION_SAMPLE;
3647                         entry->sample.trunc_size = tcf_sample_trunc_size(act);
3648                         entry->sample.truncate = tcf_sample_truncate(act);
3649                         entry->sample.rate = tcf_sample_rate(act);
3650                         tcf_sample_get_group(entry, act);
3651                 } else if (is_tcf_police(act)) {
3652                         entry->id = FLOW_ACTION_POLICE;
3653                         entry->police.burst = tcf_police_burst(act);
3654                         entry->police.rate_bytes_ps =
3655                                 tcf_police_rate_bytes_ps(act);
3656                         entry->police.mtu = tcf_police_tcfp_mtu(act);
3657                         entry->police.index = act->tcfa_index;
3658                 } else if (is_tcf_ct(act)) {
3659                         entry->id = FLOW_ACTION_CT;
3660                         entry->ct.action = tcf_ct_action(act);
3661                         entry->ct.zone = tcf_ct_zone(act);
3662                         entry->ct.flow_table = tcf_ct_ft(act);
3663                 } else if (is_tcf_mpls(act)) {
3664                         switch (tcf_mpls_action(act)) {
3665                         case TCA_MPLS_ACT_PUSH:
3666                                 entry->id = FLOW_ACTION_MPLS_PUSH;
3667                                 entry->mpls_push.proto = tcf_mpls_proto(act);
3668                                 entry->mpls_push.label = tcf_mpls_label(act);
3669                                 entry->mpls_push.tc = tcf_mpls_tc(act);
3670                                 entry->mpls_push.bos = tcf_mpls_bos(act);
3671                                 entry->mpls_push.ttl = tcf_mpls_ttl(act);
3672                                 break;
3673                         case TCA_MPLS_ACT_POP:
3674                                 entry->id = FLOW_ACTION_MPLS_POP;
3675                                 entry->mpls_pop.proto = tcf_mpls_proto(act);
3676                                 break;
3677                         case TCA_MPLS_ACT_MODIFY:
3678                                 entry->id = FLOW_ACTION_MPLS_MANGLE;
3679                                 entry->mpls_mangle.label = tcf_mpls_label(act);
3680                                 entry->mpls_mangle.tc = tcf_mpls_tc(act);
3681                                 entry->mpls_mangle.bos = tcf_mpls_bos(act);
3682                                 entry->mpls_mangle.ttl = tcf_mpls_ttl(act);
3683                                 break;
3684                         default:
3685                                 goto err_out_locked;
3686                         }
3687                 } else if (is_tcf_skbedit_ptype(act)) {
3688                         entry->id = FLOW_ACTION_PTYPE;
3689                         entry->ptype = tcf_skbedit_ptype(act);
3690                 } else if (is_tcf_skbedit_priority(act)) {
3691                         entry->id = FLOW_ACTION_PRIORITY;
3692                         entry->priority = tcf_skbedit_priority(act);
3693                 } else if (is_tcf_gate(act)) {
3694                         entry->id = FLOW_ACTION_GATE;
3695                         entry->gate.index = tcf_gate_index(act);
3696                         entry->gate.prio = tcf_gate_prio(act);
3697                         entry->gate.basetime = tcf_gate_basetime(act);
3698                         entry->gate.cycletime = tcf_gate_cycletime(act);
3699                         entry->gate.cycletimeext = tcf_gate_cycletimeext(act);
3700                         entry->gate.num_entries = tcf_gate_num_entries(act);
3701                         err = tcf_gate_get_entries(entry, act);
3702                         if (err)
3703                                 goto err_out_locked;
3704                 } else {
3705                         err = -EOPNOTSUPP;
3706                         goto err_out_locked;
3707                 }
3708                 spin_unlock_bh(&act->tcfa_lock);
3709
3710                 if (!is_tcf_pedit(act))
3711                         j++;
3712         }
3713
3714 err_out:
3715         if (err)
3716                 tc_cleanup_flow_action(flow_action);
3717
3718         return err;
3719 err_out_locked:
3720         spin_unlock_bh(&act->tcfa_lock);
3721         goto err_out;
3722 }
3723 EXPORT_SYMBOL(tc_setup_flow_action);
3724
3725 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3726 {
3727         unsigned int num_acts = 0;
3728         struct tc_action *act;
3729         int i;
3730
3731         tcf_exts_for_each_action(i, act, exts) {
3732                 if (is_tcf_pedit(act))
3733                         num_acts += tcf_pedit_nkeys(act);
3734                 else
3735                         num_acts++;
3736         }
3737         return num_acts;
3738 }
3739 EXPORT_SYMBOL(tcf_exts_num_actions);
3740
3741 #ifdef CONFIG_NET_CLS_ACT
3742 static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3743                                         u32 *p_block_index,
3744                                         struct netlink_ext_ack *extack)
3745 {
3746         *p_block_index = nla_get_u32(block_index_attr);
3747         if (!*p_block_index) {
3748                 NL_SET_ERR_MSG(extack, "Block number may not be zero");
3749                 return -EINVAL;
3750         }
3751
3752         return 0;
3753 }
3754
3755 int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3756                     enum flow_block_binder_type binder_type,
3757                     struct nlattr *block_index_attr,
3758                     struct netlink_ext_ack *extack)
3759 {
3760         u32 block_index;
3761         int err;
3762
3763         if (!block_index_attr)
3764                 return 0;
3765
3766         err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3767         if (err)
3768                 return err;
3769
3770         if (!block_index)
3771                 return 0;
3772
3773         qe->info.binder_type = binder_type;
3774         qe->info.chain_head_change = tcf_chain_head_change_dflt;
3775         qe->info.chain_head_change_priv = &qe->filter_chain;
3776         qe->info.block_index = block_index;
3777
3778         return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3779 }
3780 EXPORT_SYMBOL(tcf_qevent_init);
3781
3782 void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3783 {
3784         if (qe->info.block_index)
3785                 tcf_block_put_ext(qe->block, sch, &qe->info);
3786 }
3787 EXPORT_SYMBOL(tcf_qevent_destroy);
3788
3789 int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3790                                struct netlink_ext_ack *extack)
3791 {
3792         u32 block_index;
3793         int err;
3794
3795         if (!block_index_attr)
3796                 return 0;
3797
3798         err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3799         if (err)
3800                 return err;
3801
3802         /* Bounce newly-configured block or change in block. */
3803         if (block_index != qe->info.block_index) {
3804                 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3805                 return -EINVAL;
3806         }
3807
3808         return 0;
3809 }
3810 EXPORT_SYMBOL(tcf_qevent_validate_change);
3811
3812 struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
3813                                   struct sk_buff **to_free, int *ret)
3814 {
3815         struct tcf_result cl_res;
3816         struct tcf_proto *fl;
3817
3818         if (!qe->info.block_index)
3819                 return skb;
3820
3821         fl = rcu_dereference_bh(qe->filter_chain);
3822
3823         switch (tcf_classify(skb, fl, &cl_res, false)) {
3824         case TC_ACT_SHOT:
3825                 qdisc_qstats_drop(sch);
3826                 __qdisc_drop(skb, to_free);
3827                 *ret = __NET_XMIT_BYPASS;
3828                 return NULL;
3829         case TC_ACT_STOLEN:
3830         case TC_ACT_QUEUED:
3831         case TC_ACT_TRAP:
3832                 __qdisc_drop(skb, to_free);
3833                 *ret = __NET_XMIT_STOLEN;
3834                 return NULL;
3835         case TC_ACT_REDIRECT:
3836                 skb_do_redirect(skb);
3837                 *ret = __NET_XMIT_STOLEN;
3838                 return NULL;
3839         }
3840
3841         return skb;
3842 }
3843 EXPORT_SYMBOL(tcf_qevent_handle);
3844
3845 int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
3846 {
3847         if (!qe->info.block_index)
3848                 return 0;
3849         return nla_put_u32(skb, attr_name, qe->info.block_index);
3850 }
3851 EXPORT_SYMBOL(tcf_qevent_dump);
3852 #endif
3853
3854 static __net_init int tcf_net_init(struct net *net)
3855 {
3856         struct tcf_net *tn = net_generic(net, tcf_net_id);
3857
3858         spin_lock_init(&tn->idr_lock);
3859         idr_init(&tn->idr);
3860         return 0;
3861 }
3862
3863 static void __net_exit tcf_net_exit(struct net *net)
3864 {
3865         struct tcf_net *tn = net_generic(net, tcf_net_id);
3866
3867         idr_destroy(&tn->idr);
3868 }
3869
3870 static struct pernet_operations tcf_net_ops = {
3871         .init = tcf_net_init,
3872         .exit = tcf_net_exit,
3873         .id   = &tcf_net_id,
3874         .size = sizeof(struct tcf_net),
3875 };
3876
3877 static int __init tc_filter_init(void)
3878 {
3879         int err;
3880
3881         tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3882         if (!tc_filter_wq)
3883                 return -ENOMEM;
3884
3885         err = register_pernet_subsys(&tcf_net_ops);
3886         if (err)
3887                 goto err_register_pernet_subsys;
3888
3889         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3890                       RTNL_FLAG_DOIT_UNLOCKED);
3891         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3892                       RTNL_FLAG_DOIT_UNLOCKED);
3893         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3894                       tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3895         rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3896         rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3897         rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3898                       tc_dump_chain, 0);
3899
3900         return 0;
3901
3902 err_register_pernet_subsys:
3903         destroy_workqueue(tc_filter_wq);
3904         return err;
3905 }
3906
3907 subsys_initcall(tc_filter_init);