net/sched/cls_api.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * net/sched/cls_api.c  Packet classifier API.
   4  *
   5  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
   6  *
   7  * Changes:
   8  *
   9  * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
  10  */
  11
  12 #include <linux/module.h>
  13 #include <linux/types.h>
  14 #include <linux/kernel.h>
  15 #include <linux/string.h>
  16 #include <linux/errno.h>
  17 #include <linux/err.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/init.h>
  20 #include <linux/kmod.h>
  21 #include <linux/slab.h>
  22 #include <linux/idr.h>
  23 #include <linux/jhash.h>
  24 #include <linux/rculist.h>
  25 #include <linux/rhashtable.h>
  26 #include <net/net_namespace.h>
  27 #include <net/sock.h>
  28 #include <net/netlink.h>
  29 #include <net/pkt_sched.h>
  30 #include <net/pkt_cls.h>
  31 #include <net/tc_act/tc_pedit.h>
  32 #include <net/tc_act/tc_mirred.h>
  33 #include <net/tc_act/tc_vlan.h>
  34 #include <net/tc_act/tc_tunnel_key.h>
  35 #include <net/tc_act/tc_csum.h>
  36 #include <net/tc_act/tc_gact.h>
  37 #include <net/tc_act/tc_police.h>
  38 #include <net/tc_act/tc_sample.h>
  39 #include <net/tc_act/tc_skbedit.h>
  40 #include <net/tc_act/tc_ct.h>
  41 #include <net/tc_act/tc_mpls.h>
  42 #include <net/tc_act/tc_gate.h>
  43 #include <net/flow_offload.h>
  44 #include <net/tc_wrapper.h>
  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 struct xarray tcf_exts_miss_cookies_xa;
  53 struct tcf_exts_miss_cookie_node {
  54         const struct tcf_chain *chain;
  55         const struct tcf_proto *tp;
  56         const struct tcf_exts *exts;
  57         u32 chain_index;
  58         u32 tp_prio;
  59         u32 handle;
  60         u32 miss_cookie_base;
  61         struct rcu_head rcu;
  62 };
  63
  64 /* Each tc action entry cookie will be comprised of 32bit miss_cookie_base +
  65  * action index in the exts tc actions array.
  66  */
  67 union tcf_exts_miss_cookie {
  68         struct {
  69                 u32 miss_cookie_base;
  70                 u32 act_index;
  71         };
  72         u64 miss_cookie;
  73 };
  74
  75 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
  76 static int
  77 tcf_exts_miss_cookie_base_alloc(struct tcf_exts *exts, struct tcf_proto *tp,
  78                                 u32 handle)
  79 {
  80         struct tcf_exts_miss_cookie_node *n;
  81         static u32 next;
  82         int err;
  83
  84         if (WARN_ON(!handle || !tp->ops->get_exts))
  85                 return -EINVAL;
  86
  87         n = kzalloc(sizeof(*n), GFP_KERNEL);
  88         if (!n)
  89                 return -ENOMEM;
  90
  91         n->chain_index = tp->chain->index;
  92         n->chain = tp->chain;
  93         n->tp_prio = tp->prio;
  94         n->tp = tp;
  95         n->exts = exts;
  96         n->handle = handle;
  97
  98         err = xa_alloc_cyclic(&tcf_exts_miss_cookies_xa, &n->miss_cookie_base,
  99                               n, xa_limit_32b, &next, GFP_KERNEL);
 100         if (err)
 101                 goto err_xa_alloc;
 102
 103         exts->miss_cookie_node = n;
 104         return 0;
 105
 106 err_xa_alloc:
 107         kfree(n);
 108         return err;
 109 }
 110
 111 static void tcf_exts_miss_cookie_base_destroy(struct tcf_exts *exts)
 112 {
 113         struct tcf_exts_miss_cookie_node *n;
 114
 115         if (!exts->miss_cookie_node)
 116                 return;
 117
 118         n = exts->miss_cookie_node;
 119         xa_erase(&tcf_exts_miss_cookies_xa, n->miss_cookie_base);
 120         kfree_rcu(n, rcu);
 121 }
 122
 123 static struct tcf_exts_miss_cookie_node *
 124 tcf_exts_miss_cookie_lookup(u64 miss_cookie, int *act_index)
 125 {
 126         union tcf_exts_miss_cookie mc = { .miss_cookie = miss_cookie, };
 127
 128         *act_index = mc.act_index;
 129         return xa_load(&tcf_exts_miss_cookies_xa, mc.miss_cookie_base);
 130 }
 131 #else /* IS_ENABLED(CONFIG_NET_TC_SKB_EXT) */
 132 static int
 133 tcf_exts_miss_cookie_base_alloc(struct tcf_exts *exts, struct tcf_proto *tp,
 134                                 u32 handle)
 135 {
 136         return 0;
 137 }
 138
 139 static void tcf_exts_miss_cookie_base_destroy(struct tcf_exts *exts)
 140 {
 141 }
 142 #endif /* IS_ENABLED(CONFIG_NET_TC_SKB_EXT) */
 143
 144 static u64 tcf_exts_miss_cookie_get(u32 miss_cookie_base, int act_index)
 145 {
 146         union tcf_exts_miss_cookie mc = { .act_index = act_index, };
 147
 148         if (!miss_cookie_base)
 149                 return 0;
 150
 151         mc.miss_cookie_base = miss_cookie_base;
 152         return mc.miss_cookie;
 153 }
 154
 155 #ifdef CONFIG_NET_CLS_ACT
 156 DEFINE_STATIC_KEY_FALSE(tc_skb_ext_tc);
 157 EXPORT_SYMBOL(tc_skb_ext_tc);
 158
 159 void tc_skb_ext_tc_enable(void)
 160 {
 161         static_branch_inc(&tc_skb_ext_tc);
 162 }
 163 EXPORT_SYMBOL(tc_skb_ext_tc_enable);
 164
 165 void tc_skb_ext_tc_disable(void)
 166 {
 167         static_branch_dec(&tc_skb_ext_tc);
 168 }
 169 EXPORT_SYMBOL(tc_skb_ext_tc_disable);
 170 #endif
 171
 172 static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
 173 {
 174         return jhash_3words(tp->chain->index, tp->prio,
 175                             (__force __u32)tp->protocol, 0);
 176 }
 177
 178 static void tcf_proto_signal_destroying(struct tcf_chain *chain,
 179                                         struct tcf_proto *tp)
 180 {
 181         struct tcf_block *block = chain->block;
 182
 183         mutex_lock(&block->proto_destroy_lock);
 184         hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
 185                      destroy_obj_hashfn(tp));
 186         mutex_unlock(&block->proto_destroy_lock);
 187 }
 188
 189 static bool tcf_proto_cmp(const struct tcf_proto *tp1,
 190                           const struct tcf_proto *tp2)
 191 {
 192         return tp1->chain->index == tp2->chain->index &&
 193                tp1->prio == tp2->prio &&
 194                tp1->protocol == tp2->protocol;
 195 }
 196
 197 static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
 198                                         struct tcf_proto *tp)
 199 {
 200         u32 hash = destroy_obj_hashfn(tp);
 201         struct tcf_proto *iter;
 202         bool found = false;
 203
 204         rcu_read_lock();
 205         hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
 206                                    destroy_ht_node, hash) {
 207                 if (tcf_proto_cmp(tp, iter)) {
 208                         found = true;
 209                         break;
 210                 }
 211         }
 212         rcu_read_unlock();
 213
 214         return found;
 215 }
 216
 217 static void
 218 tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
 219 {
 220         struct tcf_block *block = chain->block;
 221
 222         mutex_lock(&block->proto_destroy_lock);
 223         if (hash_hashed(&tp->destroy_ht_node))
 224                 hash_del_rcu(&tp->destroy_ht_node);
 225         mutex_unlock(&block->proto_destroy_lock);
 226 }
 227
 228 /* Find classifier type by string name */
 229
 230 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
 231 {
 232         const struct tcf_proto_ops *t, *res = NULL;
 233
 234         if (kind) {
 235                 read_lock(&cls_mod_lock);
 236                 list_for_each_entry(t, &tcf_proto_base, head) {
 237                         if (strcmp(kind, t->kind) == 0) {
 238                                 if (try_module_get(t->owner))
 239                                         res = t;
 240                                 break;
 241                         }
 242                 }
 243                 read_unlock(&cls_mod_lock);
 244         }
 245         return res;
 246 }
 247
 248 static const struct tcf_proto_ops *
 249 tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
 250                      struct netlink_ext_ack *extack)
 251 {
 252         const struct tcf_proto_ops *ops;
 253
 254         ops = __tcf_proto_lookup_ops(kind);
 255         if (ops)
 256                 return ops;
 257 #ifdef CONFIG_MODULES
 258         if (rtnl_held)
 259                 rtnl_unlock();
 260         request_module("cls_%s", kind);
 261         if (rtnl_held)
 262                 rtnl_lock();
 263         ops = __tcf_proto_lookup_ops(kind);
 264         /* We dropped the RTNL semaphore in order to perform
 265          * the module load. So, even if we succeeded in loading
 266          * the module we have to replay the request. We indicate
 267          * this using -EAGAIN.
 268          */
 269         if (ops) {
 270                 module_put(ops->owner);
 271                 return ERR_PTR(-EAGAIN);
 272         }
 273 #endif
 274         NL_SET_ERR_MSG(extack, "TC classifier not found");
 275         return ERR_PTR(-ENOENT);
 276 }
 277
 278 /* Register(unregister) new classifier type */
 279
 280 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
 281 {
 282         struct tcf_proto_ops *t;
 283         int rc = -EEXIST;
 284
 285         write_lock(&cls_mod_lock);
 286         list_for_each_entry(t, &tcf_proto_base, head)
 287                 if (!strcmp(ops->kind, t->kind))
 288                         goto out;
 289
 290         list_add_tail(&ops->head, &tcf_proto_base);
 291         rc = 0;
 292 out:
 293         write_unlock(&cls_mod_lock);
 294         return rc;
 295 }
 296 EXPORT_SYMBOL(register_tcf_proto_ops);
 297
 298 static struct workqueue_struct *tc_filter_wq;
 299
 300 void unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
 301 {
 302         struct tcf_proto_ops *t;
 303         int rc = -ENOENT;
 304
 305         /* Wait for outstanding call_rcu()s, if any, from a
 306          * tcf_proto_ops's destroy() handler.
 307          */
 308         rcu_barrier();
 309         flush_workqueue(tc_filter_wq);
 310
 311         write_lock(&cls_mod_lock);
 312         list_for_each_entry(t, &tcf_proto_base, head) {
 313                 if (t == ops) {
 314                         list_del(&t->head);
 315                         rc = 0;
 316                         break;
 317                 }
 318         }
 319         write_unlock(&cls_mod_lock);
 320
 321         WARN(rc, "unregister tc filter kind(%s) failed %d\n", ops->kind, rc);
 322 }
 323 EXPORT_SYMBOL(unregister_tcf_proto_ops);
 324
 325 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
 326 {
 327         INIT_RCU_WORK(rwork, func);
 328         return queue_rcu_work(tc_filter_wq, rwork);
 329 }
 330 EXPORT_SYMBOL(tcf_queue_work);
 331
 332 /* Select new prio value from the range, managed by kernel. */
 333
 334 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
 335 {
 336         u32 first = TC_H_MAKE(0xC0000000U, 0U);
 337
 338         if (tp)
 339                 first = tp->prio - 1;
 340
 341         return TC_H_MAJ(first);
 342 }
 343
 344 static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
 345 {
 346         if (kind)
 347                 return nla_strscpy(name, kind, IFNAMSIZ) < 0;
 348         memset(name, 0, IFNAMSIZ);
 349         return false;
 350 }
 351
 352 static bool tcf_proto_is_unlocked(const char *kind)
 353 {
 354         const struct tcf_proto_ops *ops;
 355         bool ret;
 356
 357         if (strlen(kind) == 0)
 358                 return false;
 359
 360         ops = tcf_proto_lookup_ops(kind, false, NULL);
 361         /* On error return false to take rtnl lock. Proto lookup/create
 362          * functions will perform lookup again and properly handle errors.
 363          */
 364         if (IS_ERR(ops))
 365                 return false;
 366
 367         ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
 368         module_put(ops->owner);
 369         return ret;
 370 }
 371
 372 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
 373                                           u32 prio, struct tcf_chain *chain,
 374                                           bool rtnl_held,
 375                                           struct netlink_ext_ack *extack)
 376 {
 377         struct tcf_proto *tp;
 378         int err;
 379
 380         tp = kzalloc(sizeof(*tp), GFP_KERNEL);
 381         if (!tp)
 382                 return ERR_PTR(-ENOBUFS);
 383
 384         tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
 385         if (IS_ERR(tp->ops)) {
 386                 err = PTR_ERR(tp->ops);
 387                 goto errout;
 388         }
 389         tp->classify = tp->ops->classify;
 390         tp->protocol = protocol;
 391         tp->prio = prio;
 392         tp->chain = chain;
 393         spin_lock_init(&tp->lock);
 394         refcount_set(&tp->refcnt, 1);
 395
 396         err = tp->ops->init(tp);
 397         if (err) {
 398                 module_put(tp->ops->owner);
 399                 goto errout;
 400         }
 401         return tp;
 402
 403 errout:
 404         kfree(tp);
 405         return ERR_PTR(err);
 406 }
 407
 408 static void tcf_proto_get(struct tcf_proto *tp)
 409 {
 410         refcount_inc(&tp->refcnt);
 411 }
 412
 413 static void tcf_chain_put(struct tcf_chain *chain);
 414
 415 static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
 416                               bool sig_destroy, struct netlink_ext_ack *extack)
 417 {
 418         tp->ops->destroy(tp, rtnl_held, extack);
 419         if (sig_destroy)
 420                 tcf_proto_signal_destroyed(tp->chain, tp);
 421         tcf_chain_put(tp->chain);
 422         module_put(tp->ops->owner);
 423         kfree_rcu(tp, rcu);
 424 }
 425
 426 static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
 427                           struct netlink_ext_ack *extack)
 428 {
 429         if (refcount_dec_and_test(&tp->refcnt))
 430                 tcf_proto_destroy(tp, rtnl_held, true, extack);
 431 }
 432
 433 static bool tcf_proto_check_delete(struct tcf_proto *tp)
 434 {
 435         if (tp->ops->delete_empty)
 436                 return tp->ops->delete_empty(tp);
 437
 438         tp->deleting = true;
 439         return tp->deleting;
 440 }
 441
 442 static void tcf_proto_mark_delete(struct tcf_proto *tp)
 443 {
 444         spin_lock(&tp->lock);
 445         tp->deleting = true;
 446         spin_unlock(&tp->lock);
 447 }
 448
 449 static bool tcf_proto_is_deleting(struct tcf_proto *tp)
 450 {
 451         bool deleting;
 452
 453         spin_lock(&tp->lock);
 454         deleting = tp->deleting;
 455         spin_unlock(&tp->lock);
 456
 457         return deleting;
 458 }
 459
 460 #define ASSERT_BLOCK_LOCKED(block)                                      \
 461         lockdep_assert_held(&(block)->lock)
 462
 463 struct tcf_filter_chain_list_item {
 464         struct list_head list;
 465         tcf_chain_head_change_t *chain_head_change;
 466         void *chain_head_change_priv;
 467 };
 468
 469 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
 470                                           u32 chain_index)
 471 {
 472         struct tcf_chain *chain;
 473
 474         ASSERT_BLOCK_LOCKED(block);
 475
 476         chain = kzalloc(sizeof(*chain), GFP_KERNEL);
 477         if (!chain)
 478                 return NULL;
 479         list_add_tail_rcu(&chain->list, &block->chain_list);
 480         mutex_init(&chain->filter_chain_lock);
 481         chain->block = block;
 482         chain->index = chain_index;
 483         chain->refcnt = 1;
 484         if (!chain->index)
 485                 block->chain0.chain = chain;
 486         return chain;
 487 }
 488
 489 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
 490                                        struct tcf_proto *tp_head)
 491 {
 492         if (item->chain_head_change)
 493                 item->chain_head_change(tp_head, item->chain_head_change_priv);
 494 }
 495
 496 static void tcf_chain0_head_change(struct tcf_chain *chain,
 497                                    struct tcf_proto *tp_head)
 498 {
 499         struct tcf_filter_chain_list_item *item;
 500         struct tcf_block *block = chain->block;
 501
 502         if (chain->index)
 503                 return;
 504
 505         mutex_lock(&block->lock);
 506         list_for_each_entry(item, &block->chain0.filter_chain_list, list)
 507                 tcf_chain_head_change_item(item, tp_head);
 508         mutex_unlock(&block->lock);
 509 }
 510
 511 /* Returns true if block can be safely freed. */
 512
 513 static bool tcf_chain_detach(struct tcf_chain *chain)
 514 {
 515         struct tcf_block *block = chain->block;
 516
 517         ASSERT_BLOCK_LOCKED(block);
 518
 519         list_del_rcu(&chain->list);
 520         if (!chain->index)
 521                 block->chain0.chain = NULL;
 522
 523         if (list_empty(&block->chain_list) &&
 524             refcount_read(&block->refcnt) == 0)
 525                 return true;
 526
 527         return false;
 528 }
 529
 530 static void tcf_block_destroy(struct tcf_block *block)
 531 {
 532         mutex_destroy(&block->lock);
 533         mutex_destroy(&block->proto_destroy_lock);
 534         kfree_rcu(block, rcu);
 535 }
 536
 537 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
 538 {
 539         struct tcf_block *block = chain->block;
 540
 541         mutex_destroy(&chain->filter_chain_lock);
 542         kfree_rcu(chain, rcu);
 543         if (free_block)
 544                 tcf_block_destroy(block);
 545 }
 546
 547 static void tcf_chain_hold(struct tcf_chain *chain)
 548 {
 549         ASSERT_BLOCK_LOCKED(chain->block);
 550
 551         ++chain->refcnt;
 552 }
 553
 554 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
 555 {
 556         ASSERT_BLOCK_LOCKED(chain->block);
 557
 558         /* In case all the references are action references, this
 559          * chain should not be shown to the user.
 560          */
 561         return chain->refcnt == chain->action_refcnt;
 562 }
 563
 564 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
 565                                           u32 chain_index)
 566 {
 567         struct tcf_chain *chain;
 568
 569         ASSERT_BLOCK_LOCKED(block);
 570
 571         list_for_each_entry(chain, &block->chain_list, list) {
 572                 if (chain->index == chain_index)
 573                         return chain;
 574         }
 575         return NULL;
 576 }
 577
 578 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
 579 static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
 580                                               u32 chain_index)
 581 {
 582         struct tcf_chain *chain;
 583
 584         list_for_each_entry_rcu(chain, &block->chain_list, list) {
 585                 if (chain->index == chain_index)
 586                         return chain;
 587         }
 588         return NULL;
 589 }
 590 #endif
 591
 592 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
 593                            u32 seq, u16 flags, int event, bool unicast,
 594                            struct netlink_ext_ack *extack);
 595
 596 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
 597                                          u32 chain_index, bool create,
 598                                          bool by_act)
 599 {
 600         struct tcf_chain *chain = NULL;
 601         bool is_first_reference;
 602
 603         mutex_lock(&block->lock);
 604         chain = tcf_chain_lookup(block, chain_index);
 605         if (chain) {
 606                 tcf_chain_hold(chain);
 607         } else {
 608                 if (!create)
 609                         goto errout;
 610                 chain = tcf_chain_create(block, chain_index);
 611                 if (!chain)
 612                         goto errout;
 613         }
 614
 615         if (by_act)
 616                 ++chain->action_refcnt;
 617         is_first_reference = chain->refcnt - chain->action_refcnt == 1;
 618         mutex_unlock(&block->lock);
 619
 620         /* Send notification only in case we got the first
 621          * non-action reference. Until then, the chain acts only as
 622          * a placeholder for actions pointing to it and user ought
 623          * not know about them.
 624          */
 625         if (is_first_reference && !by_act)
 626                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
 627                                 RTM_NEWCHAIN, false, NULL);
 628
 629         return chain;
 630
 631 errout:
 632         mutex_unlock(&block->lock);
 633         return chain;
 634 }
 635
 636 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
 637                                        bool create)
 638 {
 639         return __tcf_chain_get(block, chain_index, create, false);
 640 }
 641
 642 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
 643 {
 644         return __tcf_chain_get(block, chain_index, true, true);
 645 }
 646 EXPORT_SYMBOL(tcf_chain_get_by_act);
 647
 648 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
 649                                void *tmplt_priv);
 650 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
 651                                   void *tmplt_priv, u32 chain_index,
 652                                   struct tcf_block *block, struct sk_buff *oskb,
 653                                   u32 seq, u16 flags, bool unicast);
 654
 655 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
 656                             bool explicitly_created)
 657 {
 658         struct tcf_block *block = chain->block;
 659         const struct tcf_proto_ops *tmplt_ops;
 660         bool free_block = false;
 661         unsigned int refcnt;
 662         void *tmplt_priv;
 663
 664         mutex_lock(&block->lock);
 665         if (explicitly_created) {
 666                 if (!chain->explicitly_created) {
 667                         mutex_unlock(&block->lock);
 668                         return;
 669                 }
 670                 chain->explicitly_created = false;
 671         }
 672
 673         if (by_act)
 674                 chain->action_refcnt--;
 675
 676         /* tc_chain_notify_delete can't be called while holding block lock.
 677          * However, when block is unlocked chain can be changed concurrently, so
 678          * save these to temporary variables.
 679          */
 680         refcnt = --chain->refcnt;
 681         tmplt_ops = chain->tmplt_ops;
 682         tmplt_priv = chain->tmplt_priv;
 683
 684         /* The last dropped non-action reference will trigger notification. */
 685         if (refcnt - chain->action_refcnt == 0 && !by_act) {
 686                 tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
 687                                        block, NULL, 0, 0, false);
 688                 /* Last reference to chain, no need to lock. */
 689                 chain->flushing = false;
 690         }
 691
 692         if (refcnt == 0)
 693                 free_block = tcf_chain_detach(chain);
 694         mutex_unlock(&block->lock);
 695
 696         if (refcnt == 0) {
 697                 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
 698                 tcf_chain_destroy(chain, free_block);
 699         }
 700 }
 701
 702 static void tcf_chain_put(struct tcf_chain *chain)
 703 {
 704         __tcf_chain_put(chain, false, false);
 705 }
 706
 707 void tcf_chain_put_by_act(struct tcf_chain *chain)
 708 {
 709         __tcf_chain_put(chain, true, false);
 710 }
 711 EXPORT_SYMBOL(tcf_chain_put_by_act);
 712
 713 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
 714 {
 715         __tcf_chain_put(chain, false, true);
 716 }
 717
 718 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
 719 {
 720         struct tcf_proto *tp, *tp_next;
 721
 722         mutex_lock(&chain->filter_chain_lock);
 723         tp = tcf_chain_dereference(chain->filter_chain, chain);
 724         while (tp) {
 725                 tp_next = rcu_dereference_protected(tp->next, 1);
 726                 tcf_proto_signal_destroying(chain, tp);
 727                 tp = tp_next;
 728         }
 729         tp = tcf_chain_dereference(chain->filter_chain, chain);
 730         RCU_INIT_POINTER(chain->filter_chain, NULL);
 731         tcf_chain0_head_change(chain, NULL);
 732         chain->flushing = true;
 733         mutex_unlock(&chain->filter_chain_lock);
 734
 735         while (tp) {
 736                 tp_next = rcu_dereference_protected(tp->next, 1);
 737                 tcf_proto_put(tp, rtnl_held, NULL);
 738                 tp = tp_next;
 739         }
 740 }
 741
 742 static int tcf_block_setup(struct tcf_block *block,
 743                            struct flow_block_offload *bo);
 744
 745 static void tcf_block_offload_init(struct flow_block_offload *bo,
 746                                    struct net_device *dev, struct Qdisc *sch,
 747                                    enum flow_block_command command,
 748                                    enum flow_block_binder_type binder_type,
 749                                    struct flow_block *flow_block,
 750                                    bool shared, struct netlink_ext_ack *extack)
 751 {
 752         bo->net = dev_net(dev);
 753         bo->command = command;
 754         bo->binder_type = binder_type;
 755         bo->block = flow_block;
 756         bo->block_shared = shared;
 757         bo->extack = extack;
 758         bo->sch = sch;
 759         bo->cb_list_head = &flow_block->cb_list;
 760         INIT_LIST_HEAD(&bo->cb_list);
 761 }
 762
 763 static void tcf_block_unbind(struct tcf_block *block,
 764                              struct flow_block_offload *bo);
 765
 766 static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
 767 {
 768         struct tcf_block *block = block_cb->indr.data;
 769         struct net_device *dev = block_cb->indr.dev;
 770         struct Qdisc *sch = block_cb->indr.sch;
 771         struct netlink_ext_ack extack = {};
 772         struct flow_block_offload bo = {};
 773
 774         tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
 775                                block_cb->indr.binder_type,
 776                                &block->flow_block, tcf_block_shared(block),
 777                                &extack);
 778         rtnl_lock();
 779         down_write(&block->cb_lock);
 780         list_del(&block_cb->driver_list);
 781         list_move(&block_cb->list, &bo.cb_list);
 782         tcf_block_unbind(block, &bo);
 783         up_write(&block->cb_lock);
 784         rtnl_unlock();
 785 }
 786
 787 static bool tcf_block_offload_in_use(struct tcf_block *block)
 788 {
 789         return atomic_read(&block->offloadcnt);
 790 }
 791
 792 static int tcf_block_offload_cmd(struct tcf_block *block,
 793                                  struct net_device *dev, struct Qdisc *sch,
 794                                  struct tcf_block_ext_info *ei,
 795                                  enum flow_block_command command,
 796                                  struct netlink_ext_ack *extack)
 797 {
 798         struct flow_block_offload bo = {};
 799
 800         tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
 801                                &block->flow_block, tcf_block_shared(block),
 802                                extack);
 803
 804         if (dev->netdev_ops->ndo_setup_tc) {
 805                 int err;
 806
 807                 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
 808                 if (err < 0) {
 809                         if (err != -EOPNOTSUPP)
 810                                 NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
 811                         return err;
 812                 }
 813
 814                 return tcf_block_setup(block, &bo);
 815         }
 816
 817         flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
 818                                     tc_block_indr_cleanup);
 819         tcf_block_setup(block, &bo);
 820
 821         return -EOPNOTSUPP;
 822 }
 823
 824 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
 825                                   struct tcf_block_ext_info *ei,
 826                                   struct netlink_ext_ack *extack)
 827 {
 828         struct net_device *dev = q->dev_queue->dev;
 829         int err;
 830
 831         down_write(&block->cb_lock);
 832
 833         /* If tc offload feature is disabled and the block we try to bind
 834          * to already has some offloaded filters, forbid to bind.
 835          */
 836         if (dev->netdev_ops->ndo_setup_tc &&
 837             !tc_can_offload(dev) &&
 838             tcf_block_offload_in_use(block)) {
 839                 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
 840                 err = -EOPNOTSUPP;
 841                 goto err_unlock;
 842         }
 843
 844         err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
 845         if (err == -EOPNOTSUPP)
 846                 goto no_offload_dev_inc;
 847         if (err)
 848                 goto err_unlock;
 849
 850         up_write(&block->cb_lock);
 851         return 0;
 852
 853 no_offload_dev_inc:
 854         if (tcf_block_offload_in_use(block))
 855                 goto err_unlock;
 856
 857         err = 0;
 858         block->nooffloaddevcnt++;
 859 err_unlock:
 860         up_write(&block->cb_lock);
 861         return err;
 862 }
 863
 864 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
 865                                      struct tcf_block_ext_info *ei)
 866 {
 867         struct net_device *dev = q->dev_queue->dev;
 868         int err;
 869
 870         down_write(&block->cb_lock);
 871         err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
 872         if (err == -EOPNOTSUPP)
 873                 goto no_offload_dev_dec;
 874         up_write(&block->cb_lock);
 875         return;
 876
 877 no_offload_dev_dec:
 878         WARN_ON(block->nooffloaddevcnt-- == 0);
 879         up_write(&block->cb_lock);
 880 }
 881
 882 static int
 883 tcf_chain0_head_change_cb_add(struct tcf_block *block,
 884                               struct tcf_block_ext_info *ei,
 885                               struct netlink_ext_ack *extack)
 886 {
 887         struct tcf_filter_chain_list_item *item;
 888         struct tcf_chain *chain0;
 889
 890         item = kmalloc(sizeof(*item), GFP_KERNEL);
 891         if (!item) {
 892                 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
 893                 return -ENOMEM;
 894         }
 895         item->chain_head_change = ei->chain_head_change;
 896         item->chain_head_change_priv = ei->chain_head_change_priv;
 897
 898         mutex_lock(&block->lock);
 899         chain0 = block->chain0.chain;
 900         if (chain0)
 901                 tcf_chain_hold(chain0);
 902         else
 903                 list_add(&item->list, &block->chain0.filter_chain_list);
 904         mutex_unlock(&block->lock);
 905
 906         if (chain0) {
 907                 struct tcf_proto *tp_head;
 908
 909                 mutex_lock(&chain0->filter_chain_lock);
 910
 911                 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
 912                 if (tp_head)
 913                         tcf_chain_head_change_item(item, tp_head);
 914
 915                 mutex_lock(&block->lock);
 916                 list_add(&item->list, &block->chain0.filter_chain_list);
 917                 mutex_unlock(&block->lock);
 918
 919                 mutex_unlock(&chain0->filter_chain_lock);
 920                 tcf_chain_put(chain0);
 921         }
 922
 923         return 0;
 924 }
 925
 926 static void
 927 tcf_chain0_head_change_cb_del(struct tcf_block *block,
 928                               struct tcf_block_ext_info *ei)
 929 {
 930         struct tcf_filter_chain_list_item *item;
 931
 932         mutex_lock(&block->lock);
 933         list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
 934                 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
 935                     (item->chain_head_change == ei->chain_head_change &&
 936                      item->chain_head_change_priv == ei->chain_head_change_priv)) {
 937                         if (block->chain0.chain)
 938                                 tcf_chain_head_change_item(item, NULL);
 939                         list_del(&item->list);
 940                         mutex_unlock(&block->lock);
 941
 942                         kfree(item);
 943                         return;
 944                 }
 945         }
 946         mutex_unlock(&block->lock);
 947         WARN_ON(1);
 948 }
 949
 950 struct tcf_net {
 951         spinlock_t idr_lock; /* Protects idr */
 952         struct idr idr;
 953 };
 954
 955 static unsigned int tcf_net_id;
 956
 957 static int tcf_block_insert(struct tcf_block *block, struct net *net,
 958                             struct netlink_ext_ack *extack)
 959 {
 960         struct tcf_net *tn = net_generic(net, tcf_net_id);
 961         int err;
 962
 963         idr_preload(GFP_KERNEL);
 964         spin_lock(&tn->idr_lock);
 965         err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
 966                             GFP_NOWAIT);
 967         spin_unlock(&tn->idr_lock);
 968         idr_preload_end();
 969
 970         return err;
 971 }
 972
 973 static void tcf_block_remove(struct tcf_block *block, struct net *net)
 974 {
 975         struct tcf_net *tn = net_generic(net, tcf_net_id);
 976
 977         spin_lock(&tn->idr_lock);
 978         idr_remove(&tn->idr, block->index);
 979         spin_unlock(&tn->idr_lock);
 980 }
 981
 982 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
 983                                           u32 block_index,
 984                                           struct netlink_ext_ack *extack)
 985 {
 986         struct tcf_block *block;
 987
 988         block = kzalloc(sizeof(*block), GFP_KERNEL);
 989         if (!block) {
 990                 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
 991                 return ERR_PTR(-ENOMEM);
 992         }
 993         mutex_init(&block->lock);
 994         mutex_init(&block->proto_destroy_lock);
 995         init_rwsem(&block->cb_lock);
 996         flow_block_init(&block->flow_block);
 997         INIT_LIST_HEAD(&block->chain_list);
 998         INIT_LIST_HEAD(&block->owner_list);
 999         INIT_LIST_HEAD(&block->chain0.filter_chain_list);
1000
1001         refcount_set(&block->refcnt, 1);
1002         block->net = net;
1003         block->index = block_index;
1004
1005         /* Don't store q pointer for blocks which are shared */
1006         if (!tcf_block_shared(block))
1007                 block->q = q;
1008         return block;
1009 }
1010
1011 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
1012 {
1013         struct tcf_net *tn = net_generic(net, tcf_net_id);
1014
1015         return idr_find(&tn->idr, block_index);
1016 }
1017
1018 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
1019 {
1020         struct tcf_block *block;
1021
1022         rcu_read_lock();
1023         block = tcf_block_lookup(net, block_index);
1024         if (block && !refcount_inc_not_zero(&block->refcnt))
1025                 block = NULL;
1026         rcu_read_unlock();
1027
1028         return block;
1029 }
1030
1031 static struct tcf_chain *
1032 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1033 {
1034         mutex_lock(&block->lock);
1035         if (chain)
1036                 chain = list_is_last(&chain->list, &block->chain_list) ?
1037                         NULL : list_next_entry(chain, list);
1038         else
1039                 chain = list_first_entry_or_null(&block->chain_list,
1040                                                  struct tcf_chain, list);
1041
1042         /* skip all action-only chains */
1043         while (chain && tcf_chain_held_by_acts_only(chain))
1044                 chain = list_is_last(&chain->list, &block->chain_list) ?
1045                         NULL : list_next_entry(chain, list);
1046
1047         if (chain)
1048                 tcf_chain_hold(chain);
1049         mutex_unlock(&block->lock);
1050
1051         return chain;
1052 }
1053
1054 /* Function to be used by all clients that want to iterate over all chains on
1055  * block. It properly obtains block->lock and takes reference to chain before
1056  * returning it. Users of this function must be tolerant to concurrent chain
1057  * insertion/deletion or ensure that no concurrent chain modification is
1058  * possible. Note that all netlink dump callbacks cannot guarantee to provide
1059  * consistent dump because rtnl lock is released each time skb is filled with
1060  * data and sent to user-space.
1061  */
1062
1063 struct tcf_chain *
1064 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1065 {
1066         struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
1067
1068         if (chain)
1069                 tcf_chain_put(chain);
1070
1071         return chain_next;
1072 }
1073 EXPORT_SYMBOL(tcf_get_next_chain);
1074
1075 static struct tcf_proto *
1076 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1077 {
1078         u32 prio = 0;
1079
1080         ASSERT_RTNL();
1081         mutex_lock(&chain->filter_chain_lock);
1082
1083         if (!tp) {
1084                 tp = tcf_chain_dereference(chain->filter_chain, chain);
1085         } else if (tcf_proto_is_deleting(tp)) {
1086                 /* 'deleting' flag is set and chain->filter_chain_lock was
1087                  * unlocked, which means next pointer could be invalid. Restart
1088                  * search.
1089                  */
1090                 prio = tp->prio + 1;
1091                 tp = tcf_chain_dereference(chain->filter_chain, chain);
1092
1093                 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1094                         if (!tp->deleting && tp->prio >= prio)
1095                                 break;
1096         } else {
1097                 tp = tcf_chain_dereference(tp->next, chain);
1098         }
1099
1100         if (tp)
1101                 tcf_proto_get(tp);
1102
1103         mutex_unlock(&chain->filter_chain_lock);
1104
1105         return tp;
1106 }
1107
1108 /* Function to be used by all clients that want to iterate over all tp's on
1109  * chain. Users of this function must be tolerant to concurrent tp
1110  * insertion/deletion or ensure that no concurrent chain modification is
1111  * possible. Note that all netlink dump callbacks cannot guarantee to provide
1112  * consistent dump because rtnl lock is released each time skb is filled with
1113  * data and sent to user-space.
1114  */
1115
1116 struct tcf_proto *
1117 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1118 {
1119         struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1120
1121         if (tp)
1122                 tcf_proto_put(tp, true, NULL);
1123
1124         return tp_next;
1125 }
1126 EXPORT_SYMBOL(tcf_get_next_proto);
1127
1128 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1129 {
1130         struct tcf_chain *chain;
1131
1132         /* Last reference to block. At this point chains cannot be added or
1133          * removed concurrently.
1134          */
1135         for (chain = tcf_get_next_chain(block, NULL);
1136              chain;
1137              chain = tcf_get_next_chain(block, chain)) {
1138                 tcf_chain_put_explicitly_created(chain);
1139                 tcf_chain_flush(chain, rtnl_held);
1140         }
1141 }
1142
1143 /* Lookup Qdisc and increments its reference counter.
1144  * Set parent, if necessary.
1145  */
1146
1147 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1148                             u32 *parent, int ifindex, bool rtnl_held,
1149                             struct netlink_ext_ack *extack)
1150 {
1151         const struct Qdisc_class_ops *cops;
1152         struct net_device *dev;
1153         int err = 0;
1154
1155         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1156                 return 0;
1157
1158         rcu_read_lock();
1159
1160         /* Find link */
1161         dev = dev_get_by_index_rcu(net, ifindex);
1162         if (!dev) {
1163                 rcu_read_unlock();
1164                 return -ENODEV;
1165         }
1166
1167         /* Find qdisc */
1168         if (!*parent) {
1169                 *q = rcu_dereference(dev->qdisc);
1170                 *parent = (*q)->handle;
1171         } else {
1172                 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1173                 if (!*q) {
1174                         NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1175                         err = -EINVAL;
1176                         goto errout_rcu;
1177                 }
1178         }
1179
1180         *q = qdisc_refcount_inc_nz(*q);
1181         if (!*q) {
1182                 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1183                 err = -EINVAL;
1184                 goto errout_rcu;
1185         }
1186
1187         /* Is it classful? */
1188         cops = (*q)->ops->cl_ops;
1189         if (!cops) {
1190                 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1191                 err = -EINVAL;
1192                 goto errout_qdisc;
1193         }
1194
1195         if (!cops->tcf_block) {
1196                 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1197                 err = -EOPNOTSUPP;
1198                 goto errout_qdisc;
1199         }
1200
1201 errout_rcu:
1202         /* At this point we know that qdisc is not noop_qdisc,
1203          * which means that qdisc holds a reference to net_device
1204          * and we hold a reference to qdisc, so it is safe to release
1205          * rcu read lock.
1206          */
1207         rcu_read_unlock();
1208         return err;
1209
1210 errout_qdisc:
1211         rcu_read_unlock();
1212
1213         if (rtnl_held)
1214                 qdisc_put(*q);
1215         else
1216                 qdisc_put_unlocked(*q);
1217         *q = NULL;
1218
1219         return err;
1220 }
1221
1222 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1223                                int ifindex, struct netlink_ext_ack *extack)
1224 {
1225         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1226                 return 0;
1227
1228         /* Do we search for filter, attached to class? */
1229         if (TC_H_MIN(parent)) {
1230                 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1231
1232                 *cl = cops->find(q, parent);
1233                 if (*cl == 0) {
1234                         NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1235                         return -ENOENT;
1236                 }
1237         }
1238
1239         return 0;
1240 }
1241
1242 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1243                                           unsigned long cl, int ifindex,
1244                                           u32 block_index,
1245                                           struct netlink_ext_ack *extack)
1246 {
1247         struct tcf_block *block;
1248
1249         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1250                 block = tcf_block_refcnt_get(net, block_index);
1251                 if (!block) {
1252                         NL_SET_ERR_MSG(extack, "Block of given index was not found");
1253                         return ERR_PTR(-EINVAL);
1254                 }
1255         } else {
1256                 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1257
1258                 block = cops->tcf_block(q, cl, extack);
1259                 if (!block)
1260                         return ERR_PTR(-EINVAL);
1261
1262                 if (tcf_block_shared(block)) {
1263                         NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1264                         return ERR_PTR(-EOPNOTSUPP);
1265                 }
1266
1267                 /* Always take reference to block in order to support execution
1268                  * of rules update path of cls API without rtnl lock. Caller
1269                  * must release block when it is finished using it. 'if' block
1270                  * of this conditional obtain reference to block by calling
1271                  * tcf_block_refcnt_get().
1272                  */
1273                 refcount_inc(&block->refcnt);
1274         }
1275
1276         return block;
1277 }
1278
1279 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1280                             struct tcf_block_ext_info *ei, bool rtnl_held)
1281 {
1282         if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1283                 /* Flushing/putting all chains will cause the block to be
1284                  * deallocated when last chain is freed. However, if chain_list
1285                  * is empty, block has to be manually deallocated. After block
1286                  * reference counter reached 0, it is no longer possible to
1287                  * increment it or add new chains to block.
1288                  */
1289                 bool free_block = list_empty(&block->chain_list);
1290
1291                 mutex_unlock(&block->lock);
1292                 if (tcf_block_shared(block))
1293                         tcf_block_remove(block, block->net);
1294
1295                 if (q)
1296                         tcf_block_offload_unbind(block, q, ei);
1297
1298                 if (free_block)
1299                         tcf_block_destroy(block);
1300                 else
1301                         tcf_block_flush_all_chains(block, rtnl_held);
1302         } else if (q) {
1303                 tcf_block_offload_unbind(block, q, ei);
1304         }
1305 }
1306
1307 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1308 {
1309         __tcf_block_put(block, NULL, NULL, rtnl_held);
1310 }
1311
1312 /* Find tcf block.
1313  * Set q, parent, cl when appropriate.
1314  */
1315
1316 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1317                                         u32 *parent, unsigned long *cl,
1318                                         int ifindex, u32 block_index,
1319                                         struct netlink_ext_ack *extack)
1320 {
1321         struct tcf_block *block;
1322         int err = 0;
1323
1324         ASSERT_RTNL();
1325
1326         err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1327         if (err)
1328                 goto errout;
1329
1330         err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1331         if (err)
1332                 goto errout_qdisc;
1333
1334         block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1335         if (IS_ERR(block)) {
1336                 err = PTR_ERR(block);
1337                 goto errout_qdisc;
1338         }
1339
1340         return block;
1341
1342 errout_qdisc:
1343         if (*q)
1344                 qdisc_put(*q);
1345 errout:
1346         *q = NULL;
1347         return ERR_PTR(err);
1348 }
1349
1350 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1351                               bool rtnl_held)
1352 {
1353         if (!IS_ERR_OR_NULL(block))
1354                 tcf_block_refcnt_put(block, rtnl_held);
1355
1356         if (q) {
1357                 if (rtnl_held)
1358                         qdisc_put(q);
1359                 else
1360                         qdisc_put_unlocked(q);
1361         }
1362 }
1363
1364 struct tcf_block_owner_item {
1365         struct list_head list;
1366         struct Qdisc *q;
1367         enum flow_block_binder_type binder_type;
1368 };
1369
1370 static void
1371 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1372                                struct Qdisc *q,
1373                                enum flow_block_binder_type binder_type)
1374 {
1375         if (block->keep_dst &&
1376             binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1377             binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1378                 netif_keep_dst(qdisc_dev(q));
1379 }
1380
1381 void tcf_block_netif_keep_dst(struct tcf_block *block)
1382 {
1383         struct tcf_block_owner_item *item;
1384
1385         block->keep_dst = true;
1386         list_for_each_entry(item, &block->owner_list, list)
1387                 tcf_block_owner_netif_keep_dst(block, item->q,
1388                                                item->binder_type);
1389 }
1390 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1391
1392 static int tcf_block_owner_add(struct tcf_block *block,
1393                                struct Qdisc *q,
1394                                enum flow_block_binder_type binder_type)
1395 {
1396         struct tcf_block_owner_item *item;
1397
1398         item = kmalloc(sizeof(*item), GFP_KERNEL);
1399         if (!item)
1400                 return -ENOMEM;
1401         item->q = q;
1402         item->binder_type = binder_type;
1403         list_add(&item->list, &block->owner_list);
1404         return 0;
1405 }
1406
1407 static void tcf_block_owner_del(struct tcf_block *block,
1408                                 struct Qdisc *q,
1409                                 enum flow_block_binder_type binder_type)
1410 {
1411         struct tcf_block_owner_item *item;
1412
1413         list_for_each_entry(item, &block->owner_list, list) {
1414                 if (item->q == q && item->binder_type == binder_type) {
1415                         list_del(&item->list);
1416                         kfree(item);
1417                         return;
1418                 }
1419         }
1420         WARN_ON(1);
1421 }
1422
1423 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1424                       struct tcf_block_ext_info *ei,
1425                       struct netlink_ext_ack *extack)
1426 {
1427         struct net *net = qdisc_net(q);
1428         struct tcf_block *block = NULL;
1429         int err;
1430
1431         if (ei->block_index)
1432                 /* block_index not 0 means the shared block is requested */
1433                 block = tcf_block_refcnt_get(net, ei->block_index);
1434
1435         if (!block) {
1436                 block = tcf_block_create(net, q, ei->block_index, extack);
1437                 if (IS_ERR(block))
1438                         return PTR_ERR(block);
1439                 if (tcf_block_shared(block)) {
1440                         err = tcf_block_insert(block, net, extack);
1441                         if (err)
1442                                 goto err_block_insert;
1443                 }
1444         }
1445
1446         err = tcf_block_owner_add(block, q, ei->binder_type);
1447         if (err)
1448                 goto err_block_owner_add;
1449
1450         tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1451
1452         err = tcf_chain0_head_change_cb_add(block, ei, extack);
1453         if (err)
1454                 goto err_chain0_head_change_cb_add;
1455
1456         err = tcf_block_offload_bind(block, q, ei, extack);
1457         if (err)
1458                 goto err_block_offload_bind;
1459
1460         *p_block = block;
1461         return 0;
1462
1463 err_block_offload_bind:
1464         tcf_chain0_head_change_cb_del(block, ei);
1465 err_chain0_head_change_cb_add:
1466         tcf_block_owner_del(block, q, ei->binder_type);
1467 err_block_owner_add:
1468 err_block_insert:
1469         tcf_block_refcnt_put(block, true);
1470         return err;
1471 }
1472 EXPORT_SYMBOL(tcf_block_get_ext);
1473
1474 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1475 {
1476         struct tcf_proto __rcu **p_filter_chain = priv;
1477
1478         rcu_assign_pointer(*p_filter_chain, tp_head);
1479 }
1480
1481 int tcf_block_get(struct tcf_block **p_block,
1482                   struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1483                   struct netlink_ext_ack *extack)
1484 {
1485         struct tcf_block_ext_info ei = {
1486                 .chain_head_change = tcf_chain_head_change_dflt,
1487                 .chain_head_change_priv = p_filter_chain,
1488         };
1489
1490         WARN_ON(!p_filter_chain);
1491         return tcf_block_get_ext(p_block, q, &ei, extack);
1492 }
1493 EXPORT_SYMBOL(tcf_block_get);
1494
1495 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1496  * actions should be all removed after flushing.
1497  */
1498 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1499                        struct tcf_block_ext_info *ei)
1500 {
1501         if (!block)
1502                 return;
1503         tcf_chain0_head_change_cb_del(block, ei);
1504         tcf_block_owner_del(block, q, ei->binder_type);
1505
1506         __tcf_block_put(block, q, ei, true);
1507 }
1508 EXPORT_SYMBOL(tcf_block_put_ext);
1509
1510 void tcf_block_put(struct tcf_block *block)
1511 {
1512         struct tcf_block_ext_info ei = {0, };
1513
1514         if (!block)
1515                 return;
1516         tcf_block_put_ext(block, block->q, &ei);
1517 }
1518
1519 EXPORT_SYMBOL(tcf_block_put);
1520
1521 static int
1522 tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1523                             void *cb_priv, bool add, bool offload_in_use,
1524                             struct netlink_ext_ack *extack)
1525 {
1526         struct tcf_chain *chain, *chain_prev;
1527         struct tcf_proto *tp, *tp_prev;
1528         int err;
1529
1530         lockdep_assert_held(&block->cb_lock);
1531
1532         for (chain = __tcf_get_next_chain(block, NULL);
1533              chain;
1534              chain_prev = chain,
1535                      chain = __tcf_get_next_chain(block, chain),
1536                      tcf_chain_put(chain_prev)) {
1537                 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1538                      tp_prev = tp,
1539                              tp = __tcf_get_next_proto(chain, tp),
1540                              tcf_proto_put(tp_prev, true, NULL)) {
1541                         if (tp->ops->reoffload) {
1542                                 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1543                                                          extack);
1544                                 if (err && add)
1545                                         goto err_playback_remove;
1546                         } else if (add && offload_in_use) {
1547                                 err = -EOPNOTSUPP;
1548                                 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1549                                 goto err_playback_remove;
1550                         }
1551                 }
1552         }
1553
1554         return 0;
1555
1556 err_playback_remove:
1557         tcf_proto_put(tp, true, NULL);
1558         tcf_chain_put(chain);
1559         tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1560                                     extack);
1561         return err;
1562 }
1563
1564 static int tcf_block_bind(struct tcf_block *block,
1565                           struct flow_block_offload *bo)
1566 {
1567         struct flow_block_cb *block_cb, *next;
1568         int err, i = 0;
1569
1570         lockdep_assert_held(&block->cb_lock);
1571
1572         list_for_each_entry(block_cb, &bo->cb_list, list) {
1573                 err = tcf_block_playback_offloads(block, block_cb->cb,
1574                                                   block_cb->cb_priv, true,
1575                                                   tcf_block_offload_in_use(block),
1576                                                   bo->extack);
1577                 if (err)
1578                         goto err_unroll;
1579                 if (!bo->unlocked_driver_cb)
1580                         block->lockeddevcnt++;
1581
1582                 i++;
1583         }
1584         list_splice(&bo->cb_list, &block->flow_block.cb_list);
1585
1586         return 0;
1587
1588 err_unroll:
1589         list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1590                 list_del(&block_cb->driver_list);
1591                 if (i-- > 0) {
1592                         list_del(&block_cb->list);
1593                         tcf_block_playback_offloads(block, block_cb->cb,
1594                                                     block_cb->cb_priv, false,
1595                                                     tcf_block_offload_in_use(block),
1596                                                     NULL);
1597                         if (!bo->unlocked_driver_cb)
1598                                 block->lockeddevcnt--;
1599                 }
1600                 flow_block_cb_free(block_cb);
1601         }
1602
1603         return err;
1604 }
1605
1606 static void tcf_block_unbind(struct tcf_block *block,
1607                              struct flow_block_offload *bo)
1608 {
1609         struct flow_block_cb *block_cb, *next;
1610
1611         lockdep_assert_held(&block->cb_lock);
1612
1613         list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1614                 tcf_block_playback_offloads(block, block_cb->cb,
1615                                             block_cb->cb_priv, false,
1616                                             tcf_block_offload_in_use(block),
1617                                             NULL);
1618                 list_del(&block_cb->list);
1619                 flow_block_cb_free(block_cb);
1620                 if (!bo->unlocked_driver_cb)
1621                         block->lockeddevcnt--;
1622         }
1623 }
1624
1625 static int tcf_block_setup(struct tcf_block *block,
1626                            struct flow_block_offload *bo)
1627 {
1628         int err;
1629
1630         switch (bo->command) {
1631         case FLOW_BLOCK_BIND:
1632                 err = tcf_block_bind(block, bo);
1633                 break;
1634         case FLOW_BLOCK_UNBIND:
1635                 err = 0;
1636                 tcf_block_unbind(block, bo);
1637                 break;
1638         default:
1639                 WARN_ON_ONCE(1);
1640                 err = -EOPNOTSUPP;
1641         }
1642
1643         return err;
1644 }
1645
1646 /* Main classifier routine: scans classifier chain attached
1647  * to this qdisc, (optionally) tests for protocol and asks
1648  * specific classifiers.
1649  */
1650 static inline int __tcf_classify(struct sk_buff *skb,
1651                                  const struct tcf_proto *tp,
1652                                  const struct tcf_proto *orig_tp,
1653                                  struct tcf_result *res,
1654                                  bool compat_mode,
1655                                  struct tcf_exts_miss_cookie_node *n,
1656                                  int act_index,
1657                                  u32 *last_executed_chain)
1658 {
1659 #ifdef CONFIG_NET_CLS_ACT
1660         const int max_reclassify_loop = 16;
1661         const struct tcf_proto *first_tp;
1662         int limit = 0;
1663
1664 reclassify:
1665 #endif
1666         for (; tp; tp = rcu_dereference_bh(tp->next)) {
1667                 __be16 protocol = skb_protocol(skb, false);
1668                 int err = 0;
1669
1670                 if (n) {
1671                         struct tcf_exts *exts;
1672
1673                         if (n->tp_prio != tp->prio)
1674                                 continue;
1675
1676                         /* We re-lookup the tp and chain based on index instead
1677                          * of having hard refs and locks to them, so do a sanity
1678                          * check if any of tp,chain,exts was replaced by the
1679                          * time we got here with a cookie from hardware.
1680                          */
1681                         if (unlikely(n->tp != tp || n->tp->chain != n->chain ||
1682                                      !tp->ops->get_exts))
1683                                 return TC_ACT_SHOT;
1684
1685                         exts = tp->ops->get_exts(tp, n->handle);
1686                         if (unlikely(!exts || n->exts != exts))
1687                                 return TC_ACT_SHOT;
1688
1689                         n = NULL;
1690                         err = tcf_exts_exec_ex(skb, exts, act_index, res);
1691                 } else {
1692                         if (tp->protocol != protocol &&
1693                             tp->protocol != htons(ETH_P_ALL))
1694                                 continue;
1695
1696                         err = tc_classify(skb, tp, res);
1697                 }
1698 #ifdef CONFIG_NET_CLS_ACT
1699                 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1700                         first_tp = orig_tp;
1701                         *last_executed_chain = first_tp->chain->index;
1702                         goto reset;
1703                 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1704                         first_tp = res->goto_tp;
1705                         *last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1706                         goto reset;
1707                 }
1708 #endif
1709                 if (err >= 0)
1710                         return err;
1711         }
1712
1713         if (unlikely(n))
1714                 return TC_ACT_SHOT;
1715
1716         return TC_ACT_UNSPEC; /* signal: continue lookup */
1717 #ifdef CONFIG_NET_CLS_ACT
1718 reset:
1719         if (unlikely(limit++ >= max_reclassify_loop)) {
1720                 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1721                                        tp->chain->block->index,
1722                                        tp->prio & 0xffff,
1723                                        ntohs(tp->protocol));
1724                 return TC_ACT_SHOT;
1725         }
1726
1727         tp = first_tp;
1728         goto reclassify;
1729 #endif
1730 }
1731
1732 int tcf_classify(struct sk_buff *skb,
1733                  const struct tcf_block *block,
1734                  const struct tcf_proto *tp,
1735                  struct tcf_result *res, bool compat_mode)
1736 {
1737 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1738         u32 last_executed_chain = 0;
1739
1740         return __tcf_classify(skb, tp, tp, res, compat_mode, NULL, 0,
1741                               &last_executed_chain);
1742 #else
1743         u32 last_executed_chain = tp ? tp->chain->index : 0;
1744         struct tcf_exts_miss_cookie_node *n = NULL;
1745         const struct tcf_proto *orig_tp = tp;
1746         struct tc_skb_ext *ext;
1747         int act_index = 0;
1748         int ret;
1749
1750         if (block) {
1751                 ext = skb_ext_find(skb, TC_SKB_EXT);
1752
1753                 if (ext && (ext->chain || ext->act_miss)) {
1754                         struct tcf_chain *fchain;
1755                         u32 chain;
1756
1757                         if (ext->act_miss) {
1758                                 n = tcf_exts_miss_cookie_lookup(ext->act_miss_cookie,
1759                                                                 &act_index);
1760                                 if (!n)
1761                                         return TC_ACT_SHOT;
1762
1763                                 chain = n->chain_index;
1764                         } else {
1765                                 chain = ext->chain;
1766                         }
1767
1768                         fchain = tcf_chain_lookup_rcu(block, chain);
1769                         if (!fchain)
1770                                 return TC_ACT_SHOT;
1771
1772                         /* Consume, so cloned/redirect skbs won't inherit ext */
1773                         skb_ext_del(skb, TC_SKB_EXT);
1774
1775                         tp = rcu_dereference_bh(fchain->filter_chain);
1776                         last_executed_chain = fchain->index;
1777                 }
1778         }
1779
1780         ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode, n, act_index,
1781                              &last_executed_chain);
1782
1783         if (tc_skb_ext_tc_enabled()) {
1784                 /* If we missed on some chain */
1785                 if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1786                         struct tc_skb_cb *cb = tc_skb_cb(skb);
1787
1788                         ext = tc_skb_ext_alloc(skb);
1789                         if (WARN_ON_ONCE(!ext))
1790                                 return TC_ACT_SHOT;
1791                         ext->chain = last_executed_chain;
1792                         ext->mru = cb->mru;
1793                         ext->post_ct = cb->post_ct;
1794                         ext->post_ct_snat = cb->post_ct_snat;
1795                         ext->post_ct_dnat = cb->post_ct_dnat;
1796                         ext->zone = cb->zone;
1797                 }
1798         }
1799
1800         return ret;
1801 #endif
1802 }
1803 EXPORT_SYMBOL(tcf_classify);
1804
1805 struct tcf_chain_info {
1806         struct tcf_proto __rcu **pprev;
1807         struct tcf_proto __rcu *next;
1808 };
1809
1810 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1811                                            struct tcf_chain_info *chain_info)
1812 {
1813         return tcf_chain_dereference(*chain_info->pprev, chain);
1814 }
1815
1816 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1817                                struct tcf_chain_info *chain_info,
1818                                struct tcf_proto *tp)
1819 {
1820         if (chain->flushing)
1821                 return -EAGAIN;
1822
1823         RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1824         if (*chain_info->pprev == chain->filter_chain)
1825                 tcf_chain0_head_change(chain, tp);
1826         tcf_proto_get(tp);
1827         rcu_assign_pointer(*chain_info->pprev, tp);
1828
1829         return 0;
1830 }
1831
1832 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1833                                 struct tcf_chain_info *chain_info,
1834                                 struct tcf_proto *tp)
1835 {
1836         struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1837
1838         tcf_proto_mark_delete(tp);
1839         if (tp == chain->filter_chain)
1840                 tcf_chain0_head_change(chain, next);
1841         RCU_INIT_POINTER(*chain_info->pprev, next);
1842 }
1843
1844 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1845                                            struct tcf_chain_info *chain_info,
1846                                            u32 protocol, u32 prio,
1847                                            bool prio_allocate);
1848
1849 /* Try to insert new proto.
1850  * If proto with specified priority already exists, free new proto
1851  * and return existing one.
1852  */
1853
1854 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1855                                                     struct tcf_proto *tp_new,
1856                                                     u32 protocol, u32 prio,
1857                                                     bool rtnl_held)
1858 {
1859         struct tcf_chain_info chain_info;
1860         struct tcf_proto *tp;
1861         int err = 0;
1862
1863         mutex_lock(&chain->filter_chain_lock);
1864
1865         if (tcf_proto_exists_destroying(chain, tp_new)) {
1866                 mutex_unlock(&chain->filter_chain_lock);
1867                 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1868                 return ERR_PTR(-EAGAIN);
1869         }
1870
1871         tp = tcf_chain_tp_find(chain, &chain_info,
1872                                protocol, prio, false);
1873         if (!tp)
1874                 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1875         mutex_unlock(&chain->filter_chain_lock);
1876
1877         if (tp) {
1878                 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1879                 tp_new = tp;
1880         } else if (err) {
1881                 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1882                 tp_new = ERR_PTR(err);
1883         }
1884
1885         return tp_new;
1886 }
1887
1888 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1889                                       struct tcf_proto *tp, bool rtnl_held,
1890                                       struct netlink_ext_ack *extack)
1891 {
1892         struct tcf_chain_info chain_info;
1893         struct tcf_proto *tp_iter;
1894         struct tcf_proto **pprev;
1895         struct tcf_proto *next;
1896
1897         mutex_lock(&chain->filter_chain_lock);
1898
1899         /* Atomically find and remove tp from chain. */
1900         for (pprev = &chain->filter_chain;
1901              (tp_iter = tcf_chain_dereference(*pprev, chain));
1902              pprev = &tp_iter->next) {
1903                 if (tp_iter == tp) {
1904                         chain_info.pprev = pprev;
1905                         chain_info.next = tp_iter->next;
1906                         WARN_ON(tp_iter->deleting);
1907                         break;
1908                 }
1909         }
1910         /* Verify that tp still exists and no new filters were inserted
1911          * concurrently.
1912          * Mark tp for deletion if it is empty.
1913          */
1914         if (!tp_iter || !tcf_proto_check_delete(tp)) {
1915                 mutex_unlock(&chain->filter_chain_lock);
1916                 return;
1917         }
1918
1919         tcf_proto_signal_destroying(chain, tp);
1920         next = tcf_chain_dereference(chain_info.next, chain);
1921         if (tp == chain->filter_chain)
1922                 tcf_chain0_head_change(chain, next);
1923         RCU_INIT_POINTER(*chain_info.pprev, next);
1924         mutex_unlock(&chain->filter_chain_lock);
1925
1926         tcf_proto_put(tp, rtnl_held, extack);
1927 }
1928
1929 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1930                                            struct tcf_chain_info *chain_info,
1931                                            u32 protocol, u32 prio,
1932                                            bool prio_allocate)
1933 {
1934         struct tcf_proto **pprev;
1935         struct tcf_proto *tp;
1936
1937         /* Check the chain for existence of proto-tcf with this priority */
1938         for (pprev = &chain->filter_chain;
1939              (tp = tcf_chain_dereference(*pprev, chain));
1940              pprev = &tp->next) {
1941                 if (tp->prio >= prio) {
1942                         if (tp->prio == prio) {
1943                                 if (prio_allocate ||
1944                                     (tp->protocol != protocol && protocol))
1945                                         return ERR_PTR(-EINVAL);
1946                         } else {
1947                                 tp = NULL;
1948                         }
1949                         break;
1950                 }
1951         }
1952         chain_info->pprev = pprev;
1953         if (tp) {
1954                 chain_info->next = tp->next;
1955                 tcf_proto_get(tp);
1956         } else {
1957                 chain_info->next = NULL;
1958         }
1959         return tp;
1960 }
1961
1962 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1963                          struct tcf_proto *tp, struct tcf_block *block,
1964                          struct Qdisc *q, u32 parent, void *fh,
1965                          u32 portid, u32 seq, u16 flags, int event,
1966                          bool terse_dump, bool rtnl_held,
1967                          struct netlink_ext_ack *extack)
1968 {
1969         struct tcmsg *tcm;
1970         struct nlmsghdr  *nlh;
1971         unsigned char *b = skb_tail_pointer(skb);
1972
1973         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1974         if (!nlh)
1975                 goto out_nlmsg_trim;
1976         tcm = nlmsg_data(nlh);
1977         tcm->tcm_family = AF_UNSPEC;
1978         tcm->tcm__pad1 = 0;
1979         tcm->tcm__pad2 = 0;
1980         if (q) {
1981                 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1982                 tcm->tcm_parent = parent;
1983         } else {
1984                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1985                 tcm->tcm_block_index = block->index;
1986         }
1987         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1988         if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1989                 goto nla_put_failure;
1990         if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1991                 goto nla_put_failure;
1992         if (!fh) {
1993                 tcm->tcm_handle = 0;
1994         } else if (terse_dump) {
1995                 if (tp->ops->terse_dump) {
1996                         if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
1997                                                 rtnl_held) < 0)
1998                                 goto nla_put_failure;
1999                 } else {
2000                         goto cls_op_not_supp;
2001                 }
2002         } else {
2003                 if (tp->ops->dump &&
2004                     tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
2005                         goto nla_put_failure;
2006         }
2007
2008         if (extack && extack->_msg &&
2009             nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2010                 goto nla_put_failure;
2011
2012         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2013
2014         return skb->len;
2015
2016 out_nlmsg_trim:
2017 nla_put_failure:
2018 cls_op_not_supp:
2019         nlmsg_trim(skb, b);
2020         return -1;
2021 }
2022
2023 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
2024                           struct nlmsghdr *n, struct tcf_proto *tp,
2025                           struct tcf_block *block, struct Qdisc *q,
2026                           u32 parent, void *fh, int event, bool unicast,
2027                           bool rtnl_held, struct netlink_ext_ack *extack)
2028 {
2029         struct sk_buff *skb;
2030         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2031         int err = 0;
2032
2033         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2034         if (!skb)
2035                 return -ENOBUFS;
2036
2037         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2038                           n->nlmsg_seq, n->nlmsg_flags, event,
2039                           false, rtnl_held, extack) <= 0) {
2040                 kfree_skb(skb);
2041                 return -EINVAL;
2042         }
2043
2044         if (unicast)
2045                 err = rtnl_unicast(skb, net, portid);
2046         else
2047                 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2048                                      n->nlmsg_flags & NLM_F_ECHO);
2049         return err;
2050 }
2051
2052 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
2053                               struct nlmsghdr *n, struct tcf_proto *tp,
2054                               struct tcf_block *block, struct Qdisc *q,
2055                               u32 parent, void *fh, bool unicast, bool *last,
2056                               bool rtnl_held, struct netlink_ext_ack *extack)
2057 {
2058         struct sk_buff *skb;
2059         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2060         int err;
2061
2062         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2063         if (!skb)
2064                 return -ENOBUFS;
2065
2066         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2067                           n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
2068                           false, rtnl_held, extack) <= 0) {
2069                 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
2070                 kfree_skb(skb);
2071                 return -EINVAL;
2072         }
2073
2074         err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
2075         if (err) {
2076                 kfree_skb(skb);
2077                 return err;
2078         }
2079
2080         if (unicast)
2081                 err = rtnl_unicast(skb, net, portid);
2082         else
2083                 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2084                                      n->nlmsg_flags & NLM_F_ECHO);
2085         if (err < 0)
2086                 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
2087
2088         return err;
2089 }
2090
2091 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
2092                                  struct tcf_block *block, struct Qdisc *q,
2093                                  u32 parent, struct nlmsghdr *n,
2094                                  struct tcf_chain *chain, int event,
2095                                  struct netlink_ext_ack *extack)
2096 {
2097         struct tcf_proto *tp;
2098
2099         for (tp = tcf_get_next_proto(chain, NULL);
2100              tp; tp = tcf_get_next_proto(chain, tp))
2101                 tfilter_notify(net, oskb, n, tp, block, q, parent, NULL,
2102                                event, false, true, extack);
2103 }
2104
2105 static void tfilter_put(struct tcf_proto *tp, void *fh)
2106 {
2107         if (tp->ops->put && fh)
2108                 tp->ops->put(tp, fh);
2109 }
2110
2111 static bool is_qdisc_ingress(__u32 classid)
2112 {
2113         return (TC_H_MIN(classid) == TC_H_MIN(TC_H_MIN_INGRESS));
2114 }
2115
2116 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2117                           struct netlink_ext_ack *extack)
2118 {
2119         struct net *net = sock_net(skb->sk);
2120         struct nlattr *tca[TCA_MAX + 1];
2121         char name[IFNAMSIZ];
2122         struct tcmsg *t;
2123         u32 protocol;
2124         u32 prio;
2125         bool prio_allocate;
2126         u32 parent;
2127         u32 chain_index;
2128         struct Qdisc *q;
2129         struct tcf_chain_info chain_info;
2130         struct tcf_chain *chain;
2131         struct tcf_block *block;
2132         struct tcf_proto *tp;
2133         unsigned long cl;
2134         void *fh;
2135         int err;
2136         int tp_created;
2137         bool rtnl_held = false;
2138         u32 flags;
2139
2140 replay:
2141         tp_created = 0;
2142
2143         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2144                                      rtm_tca_policy, extack);
2145         if (err < 0)
2146                 return err;
2147
2148         t = nlmsg_data(n);
2149         protocol = TC_H_MIN(t->tcm_info);
2150         prio = TC_H_MAJ(t->tcm_info);
2151         prio_allocate = false;
2152         parent = t->tcm_parent;
2153         tp = NULL;
2154         cl = 0;
2155         block = NULL;
2156         q = NULL;
2157         chain = NULL;
2158         flags = 0;
2159
2160         if (prio == 0) {
2161                 /* If no priority is provided by the user,
2162                  * we allocate one.
2163                  */
2164                 if (n->nlmsg_flags & NLM_F_CREATE) {
2165                         prio = TC_H_MAKE(0x80000000U, 0U);
2166                         prio_allocate = true;
2167                 } else {
2168                         NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2169                         return -ENOENT;
2170                 }
2171         }
2172
2173         /* Find head of filter chain. */
2174
2175         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2176         if (err)
2177                 return err;
2178
2179         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2180                 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2181                 err = -EINVAL;
2182                 goto errout;
2183         }
2184
2185         /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2186          * block is shared (no qdisc found), qdisc is not unlocked, classifier
2187          * type is not specified, classifier is not unlocked.
2188          */
2189         if (rtnl_held ||
2190             (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2191             !tcf_proto_is_unlocked(name)) {
2192                 rtnl_held = true;
2193                 rtnl_lock();
2194         }
2195
2196         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2197         if (err)
2198                 goto errout;
2199
2200         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2201                                  extack);
2202         if (IS_ERR(block)) {
2203                 err = PTR_ERR(block);
2204                 goto errout;
2205         }
2206         block->classid = parent;
2207
2208         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2209         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2210                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2211                 err = -EINVAL;
2212                 goto errout;
2213         }
2214         chain = tcf_chain_get(block, chain_index, true);
2215         if (!chain) {
2216                 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2217                 err = -ENOMEM;
2218                 goto errout;
2219         }
2220
2221         mutex_lock(&chain->filter_chain_lock);
2222         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2223                                prio, prio_allocate);
2224         if (IS_ERR(tp)) {
2225                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2226                 err = PTR_ERR(tp);
2227                 goto errout_locked;
2228         }
2229
2230         if (tp == NULL) {
2231                 struct tcf_proto *tp_new = NULL;
2232
2233                 if (chain->flushing) {
2234                         err = -EAGAIN;
2235                         goto errout_locked;
2236                 }
2237
2238                 /* Proto-tcf does not exist, create new one */
2239
2240                 if (tca[TCA_KIND] == NULL || !protocol) {
2241                         NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2242                         err = -EINVAL;
2243                         goto errout_locked;
2244                 }
2245
2246                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2247                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2248                         err = -ENOENT;
2249                         goto errout_locked;
2250                 }
2251
2252                 if (prio_allocate)
2253                         prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2254                                                                &chain_info));
2255
2256                 mutex_unlock(&chain->filter_chain_lock);
2257                 tp_new = tcf_proto_create(name, protocol, prio, chain,
2258                                           rtnl_held, extack);
2259                 if (IS_ERR(tp_new)) {
2260                         err = PTR_ERR(tp_new);
2261                         goto errout_tp;
2262                 }
2263
2264                 tp_created = 1;
2265                 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2266                                                 rtnl_held);
2267                 if (IS_ERR(tp)) {
2268                         err = PTR_ERR(tp);
2269                         goto errout_tp;
2270                 }
2271         } else {
2272                 mutex_unlock(&chain->filter_chain_lock);
2273         }
2274
2275         if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2276                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2277                 err = -EINVAL;
2278                 goto errout;
2279         }
2280
2281         fh = tp->ops->get(tp, t->tcm_handle);
2282
2283         if (!fh) {
2284                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2285                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2286                         err = -ENOENT;
2287                         goto errout;
2288                 }
2289         } else if (n->nlmsg_flags & NLM_F_EXCL) {
2290                 tfilter_put(tp, fh);
2291                 NL_SET_ERR_MSG(extack, "Filter already exists");
2292                 err = -EEXIST;
2293                 goto errout;
2294         }
2295
2296         if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2297                 tfilter_put(tp, fh);
2298                 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2299                 err = -EINVAL;
2300                 goto errout;
2301         }
2302
2303         if (!(n->nlmsg_flags & NLM_F_CREATE))
2304                 flags |= TCA_ACT_FLAGS_REPLACE;
2305         if (!rtnl_held)
2306                 flags |= TCA_ACT_FLAGS_NO_RTNL;
2307         if (is_qdisc_ingress(parent))
2308                 flags |= TCA_ACT_FLAGS_AT_INGRESS;
2309         err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2310                               flags, extack);
2311         if (err == 0) {
2312                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2313                                RTM_NEWTFILTER, false, rtnl_held, extack);
2314                 tfilter_put(tp, fh);
2315                 /* q pointer is NULL for shared blocks */
2316                 if (q)
2317                         q->flags &= ~TCQ_F_CAN_BYPASS;
2318         }
2319
2320 errout:
2321         if (err && tp_created)
2322                 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2323 errout_tp:
2324         if (chain) {
2325                 if (tp && !IS_ERR(tp))
2326                         tcf_proto_put(tp, rtnl_held, NULL);
2327                 if (!tp_created)
2328                         tcf_chain_put(chain);
2329         }
2330         tcf_block_release(q, block, rtnl_held);
2331
2332         if (rtnl_held)
2333                 rtnl_unlock();
2334
2335         if (err == -EAGAIN) {
2336                 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2337                  * of target chain.
2338                  */
2339                 rtnl_held = true;
2340                 /* Replay the request. */
2341                 goto replay;
2342         }
2343         return err;
2344
2345 errout_locked:
2346         mutex_unlock(&chain->filter_chain_lock);
2347         goto errout;
2348 }
2349
2350 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2351                           struct netlink_ext_ack *extack)
2352 {
2353         struct net *net = sock_net(skb->sk);
2354         struct nlattr *tca[TCA_MAX + 1];
2355         char name[IFNAMSIZ];
2356         struct tcmsg *t;
2357         u32 protocol;
2358         u32 prio;
2359         u32 parent;
2360         u32 chain_index;
2361         struct Qdisc *q = NULL;
2362         struct tcf_chain_info chain_info;
2363         struct tcf_chain *chain = NULL;
2364         struct tcf_block *block = NULL;
2365         struct tcf_proto *tp = NULL;
2366         unsigned long cl = 0;
2367         void *fh = NULL;
2368         int err;
2369         bool rtnl_held = false;
2370
2371         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2372                                      rtm_tca_policy, extack);
2373         if (err < 0)
2374                 return err;
2375
2376         t = nlmsg_data(n);
2377         protocol = TC_H_MIN(t->tcm_info);
2378         prio = TC_H_MAJ(t->tcm_info);
2379         parent = t->tcm_parent;
2380
2381         if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2382                 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2383                 return -ENOENT;
2384         }
2385
2386         /* Find head of filter chain. */
2387
2388         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2389         if (err)
2390                 return err;
2391
2392         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2393                 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2394                 err = -EINVAL;
2395                 goto errout;
2396         }
2397         /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2398          * found), qdisc is not unlocked, classifier type is not specified,
2399          * classifier is not unlocked.
2400          */
2401         if (!prio ||
2402             (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2403             !tcf_proto_is_unlocked(name)) {
2404                 rtnl_held = true;
2405                 rtnl_lock();
2406         }
2407
2408         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2409         if (err)
2410                 goto errout;
2411
2412         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2413                                  extack);
2414         if (IS_ERR(block)) {
2415                 err = PTR_ERR(block);
2416                 goto errout;
2417         }
2418
2419         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2420         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2421                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2422                 err = -EINVAL;
2423                 goto errout;
2424         }
2425         chain = tcf_chain_get(block, chain_index, false);
2426         if (!chain) {
2427                 /* User requested flush on non-existent chain. Nothing to do,
2428                  * so just return success.
2429                  */
2430                 if (prio == 0) {
2431                         err = 0;
2432                         goto errout;
2433                 }
2434                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2435                 err = -ENOENT;
2436                 goto errout;
2437         }
2438
2439         if (prio == 0) {
2440                 tfilter_notify_chain(net, skb, block, q, parent, n,
2441                                      chain, RTM_DELTFILTER, extack);
2442                 tcf_chain_flush(chain, rtnl_held);
2443                 err = 0;
2444                 goto errout;
2445         }
2446
2447         mutex_lock(&chain->filter_chain_lock);
2448         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2449                                prio, false);
2450         if (!tp || IS_ERR(tp)) {
2451                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2452                 err = tp ? PTR_ERR(tp) : -ENOENT;
2453                 goto errout_locked;
2454         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2455                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2456                 err = -EINVAL;
2457                 goto errout_locked;
2458         } else if (t->tcm_handle == 0) {
2459                 tcf_proto_signal_destroying(chain, tp);
2460                 tcf_chain_tp_remove(chain, &chain_info, tp);
2461                 mutex_unlock(&chain->filter_chain_lock);
2462
2463                 tcf_proto_put(tp, rtnl_held, NULL);
2464                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2465                                RTM_DELTFILTER, false, rtnl_held, extack);
2466                 err = 0;
2467                 goto errout;
2468         }
2469         mutex_unlock(&chain->filter_chain_lock);
2470
2471         fh = tp->ops->get(tp, t->tcm_handle);
2472
2473         if (!fh) {
2474                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2475                 err = -ENOENT;
2476         } else {
2477                 bool last;
2478
2479                 err = tfilter_del_notify(net, skb, n, tp, block,
2480                                          q, parent, fh, false, &last,
2481                                          rtnl_held, extack);
2482
2483                 if (err)
2484                         goto errout;
2485                 if (last)
2486                         tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2487         }
2488
2489 errout:
2490         if (chain) {
2491                 if (tp && !IS_ERR(tp))
2492                         tcf_proto_put(tp, rtnl_held, NULL);
2493                 tcf_chain_put(chain);
2494         }
2495         tcf_block_release(q, block, rtnl_held);
2496
2497         if (rtnl_held)
2498                 rtnl_unlock();
2499
2500         return err;
2501
2502 errout_locked:
2503         mutex_unlock(&chain->filter_chain_lock);
2504         goto errout;
2505 }
2506
2507 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2508                           struct netlink_ext_ack *extack)
2509 {
2510         struct net *net = sock_net(skb->sk);
2511         struct nlattr *tca[TCA_MAX + 1];
2512         char name[IFNAMSIZ];
2513         struct tcmsg *t;
2514         u32 protocol;
2515         u32 prio;
2516         u32 parent;
2517         u32 chain_index;
2518         struct Qdisc *q = NULL;
2519         struct tcf_chain_info chain_info;
2520         struct tcf_chain *chain = NULL;
2521         struct tcf_block *block = NULL;
2522         struct tcf_proto *tp = NULL;
2523         unsigned long cl = 0;
2524         void *fh = NULL;
2525         int err;
2526         bool rtnl_held = false;
2527
2528         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2529                                      rtm_tca_policy, extack);
2530         if (err < 0)
2531                 return err;
2532
2533         t = nlmsg_data(n);
2534         protocol = TC_H_MIN(t->tcm_info);
2535         prio = TC_H_MAJ(t->tcm_info);
2536         parent = t->tcm_parent;
2537
2538         if (prio == 0) {
2539                 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2540                 return -ENOENT;
2541         }
2542
2543         /* Find head of filter chain. */
2544
2545         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2546         if (err)
2547                 return err;
2548
2549         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2550                 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2551                 err = -EINVAL;
2552                 goto errout;
2553         }
2554         /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2555          * unlocked, classifier type is not specified, classifier is not
2556          * unlocked.
2557          */
2558         if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2559             !tcf_proto_is_unlocked(name)) {
2560                 rtnl_held = true;
2561                 rtnl_lock();
2562         }
2563
2564         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2565         if (err)
2566                 goto errout;
2567
2568         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2569                                  extack);
2570         if (IS_ERR(block)) {
2571                 err = PTR_ERR(block);
2572                 goto errout;
2573         }
2574
2575         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2576         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2577                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2578                 err = -EINVAL;
2579                 goto errout;
2580         }
2581         chain = tcf_chain_get(block, chain_index, false);
2582         if (!chain) {
2583                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2584                 err = -EINVAL;
2585                 goto errout;
2586         }
2587
2588         mutex_lock(&chain->filter_chain_lock);
2589         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2590                                prio, false);
2591         mutex_unlock(&chain->filter_chain_lock);
2592         if (!tp || IS_ERR(tp)) {
2593                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2594                 err = tp ? PTR_ERR(tp) : -ENOENT;
2595                 goto errout;
2596         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2597                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2598                 err = -EINVAL;
2599                 goto errout;
2600         }
2601
2602         fh = tp->ops->get(tp, t->tcm_handle);
2603
2604         if (!fh) {
2605                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2606                 err = -ENOENT;
2607         } else {
2608                 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2609                                      fh, RTM_NEWTFILTER, true, rtnl_held, NULL);
2610                 if (err < 0)
2611                         NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2612         }
2613
2614         tfilter_put(tp, fh);
2615 errout:
2616         if (chain) {
2617                 if (tp && !IS_ERR(tp))
2618                         tcf_proto_put(tp, rtnl_held, NULL);
2619                 tcf_chain_put(chain);
2620         }
2621         tcf_block_release(q, block, rtnl_held);
2622
2623         if (rtnl_held)
2624                 rtnl_unlock();
2625
2626         return err;
2627 }
2628
2629 struct tcf_dump_args {
2630         struct tcf_walker w;
2631         struct sk_buff *skb;
2632         struct netlink_callback *cb;
2633         struct tcf_block *block;
2634         struct Qdisc *q;
2635         u32 parent;
2636         bool terse_dump;
2637 };
2638
2639 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2640 {
2641         struct tcf_dump_args *a = (void *)arg;
2642         struct net *net = sock_net(a->skb->sk);
2643
2644         return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2645                              n, NETLINK_CB(a->cb->skb).portid,
2646                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2647                              RTM_NEWTFILTER, a->terse_dump, true, NULL);
2648 }
2649
2650 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2651                            struct sk_buff *skb, struct netlink_callback *cb,
2652                            long index_start, long *p_index, bool terse)
2653 {
2654         struct net *net = sock_net(skb->sk);
2655         struct tcf_block *block = chain->block;
2656         struct tcmsg *tcm = nlmsg_data(cb->nlh);
2657         struct tcf_proto *tp, *tp_prev;
2658         struct tcf_dump_args arg;
2659
2660         for (tp = __tcf_get_next_proto(chain, NULL);
2661              tp;
2662              tp_prev = tp,
2663                      tp = __tcf_get_next_proto(chain, tp),
2664                      tcf_proto_put(tp_prev, true, NULL),
2665                      (*p_index)++) {
2666                 if (*p_index < index_start)
2667                         continue;
2668                 if (TC_H_MAJ(tcm->tcm_info) &&
2669                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
2670                         continue;
2671                 if (TC_H_MIN(tcm->tcm_info) &&
2672                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
2673                         continue;
2674                 if (*p_index > index_start)
2675                         memset(&cb->args[1], 0,
2676                                sizeof(cb->args) - sizeof(cb->args[0]));
2677                 if (cb->args[1] == 0) {
2678                         if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2679                                           NETLINK_CB(cb->skb).portid,
2680                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
2681                                           RTM_NEWTFILTER, false, true, NULL) <= 0)
2682                                 goto errout;
2683                         cb->args[1] = 1;
2684                 }
2685                 if (!tp->ops->walk)
2686                         continue;
2687                 arg.w.fn = tcf_node_dump;
2688                 arg.skb = skb;
2689                 arg.cb = cb;
2690                 arg.block = block;
2691                 arg.q = q;
2692                 arg.parent = parent;
2693                 arg.w.stop = 0;
2694                 arg.w.skip = cb->args[1] - 1;
2695                 arg.w.count = 0;
2696                 arg.w.cookie = cb->args[2];
2697                 arg.terse_dump = terse;
2698                 tp->ops->walk(tp, &arg.w, true);
2699                 cb->args[2] = arg.w.cookie;
2700                 cb->args[1] = arg.w.count + 1;
2701                 if (arg.w.stop)
2702                         goto errout;
2703         }
2704         return true;
2705
2706 errout:
2707         tcf_proto_put(tp, true, NULL);
2708         return false;
2709 }
2710
2711 static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2712         [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2713 };
2714
2715 /* called with RTNL */
2716 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2717 {
2718         struct tcf_chain *chain, *chain_prev;
2719         struct net *net = sock_net(skb->sk);
2720         struct nlattr *tca[TCA_MAX + 1];
2721         struct Qdisc *q = NULL;
2722         struct tcf_block *block;
2723         struct tcmsg *tcm = nlmsg_data(cb->nlh);
2724         bool terse_dump = false;
2725         long index_start;
2726         long index;
2727         u32 parent;
2728         int err;
2729
2730         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2731                 return skb->len;
2732
2733         err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2734                                      tcf_tfilter_dump_policy, cb->extack);
2735         if (err)
2736                 return err;
2737
2738         if (tca[TCA_DUMP_FLAGS]) {
2739                 struct nla_bitfield32 flags =
2740                         nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2741
2742                 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2743         }
2744
2745         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2746                 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2747                 if (!block)
2748                         goto out;
2749                 /* If we work with block index, q is NULL and parent value
2750                  * will never be used in the following code. The check
2751                  * in tcf_fill_node prevents it. However, compiler does not
2752                  * see that far, so set parent to zero to silence the warning
2753                  * about parent being uninitialized.
2754                  */
2755                 parent = 0;
2756         } else {
2757                 const struct Qdisc_class_ops *cops;
2758                 struct net_device *dev;
2759                 unsigned long cl = 0;
2760
2761                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2762                 if (!dev)
2763                         return skb->len;
2764
2765                 parent = tcm->tcm_parent;
2766                 if (!parent)
2767                         q = rtnl_dereference(dev->qdisc);
2768                 else
2769                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2770                 if (!q)
2771                         goto out;
2772                 cops = q->ops->cl_ops;
2773                 if (!cops)
2774                         goto out;
2775                 if (!cops->tcf_block)
2776                         goto out;
2777                 if (TC_H_MIN(tcm->tcm_parent)) {
2778                         cl = cops->find(q, tcm->tcm_parent);
2779                         if (cl == 0)
2780                                 goto out;
2781                 }
2782                 block = cops->tcf_block(q, cl, NULL);
2783                 if (!block)
2784                         goto out;
2785                 parent = block->classid;
2786                 if (tcf_block_shared(block))
2787                         q = NULL;
2788         }
2789
2790         index_start = cb->args[0];
2791         index = 0;
2792
2793         for (chain = __tcf_get_next_chain(block, NULL);
2794              chain;
2795              chain_prev = chain,
2796                      chain = __tcf_get_next_chain(block, chain),
2797                      tcf_chain_put(chain_prev)) {
2798                 if (tca[TCA_CHAIN] &&
2799                     nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2800                         continue;
2801                 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2802                                     index_start, &index, terse_dump)) {
2803                         tcf_chain_put(chain);
2804                         err = -EMSGSIZE;
2805                         break;
2806                 }
2807         }
2808
2809         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2810                 tcf_block_refcnt_put(block, true);
2811         cb->args[0] = index;
2812
2813 out:
2814         /* If we did no progress, the error (EMSGSIZE) is real */
2815         if (skb->len == 0 && err)
2816                 return err;
2817         return skb->len;
2818 }
2819
2820 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2821                               void *tmplt_priv, u32 chain_index,
2822                               struct net *net, struct sk_buff *skb,
2823                               struct tcf_block *block,
2824                               u32 portid, u32 seq, u16 flags, int event,
2825                               struct netlink_ext_ack *extack)
2826 {
2827         unsigned char *b = skb_tail_pointer(skb);
2828         const struct tcf_proto_ops *ops;
2829         struct nlmsghdr *nlh;
2830         struct tcmsg *tcm;
2831         void *priv;
2832
2833         ops = tmplt_ops;
2834         priv = tmplt_priv;
2835
2836         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2837         if (!nlh)
2838                 goto out_nlmsg_trim;
2839         tcm = nlmsg_data(nlh);
2840         tcm->tcm_family = AF_UNSPEC;
2841         tcm->tcm__pad1 = 0;
2842         tcm->tcm__pad2 = 0;
2843         tcm->tcm_handle = 0;
2844         if (block->q) {
2845                 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2846                 tcm->tcm_parent = block->q->handle;
2847         } else {
2848                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2849                 tcm->tcm_block_index = block->index;
2850         }
2851
2852         if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2853                 goto nla_put_failure;
2854
2855         if (ops) {
2856                 if (nla_put_string(skb, TCA_KIND, ops->kind))
2857                         goto nla_put_failure;
2858                 if (ops->tmplt_dump(skb, net, priv) < 0)
2859                         goto nla_put_failure;
2860         }
2861
2862         if (extack && extack->_msg &&
2863             nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2864                 goto out_nlmsg_trim;
2865
2866         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2867
2868         return skb->len;
2869
2870 out_nlmsg_trim:
2871 nla_put_failure:
2872         nlmsg_trim(skb, b);
2873         return -EMSGSIZE;
2874 }
2875
2876 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2877                            u32 seq, u16 flags, int event, bool unicast,
2878                            struct netlink_ext_ack *extack)
2879 {
2880         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2881         struct tcf_block *block = chain->block;
2882         struct net *net = block->net;
2883         struct sk_buff *skb;
2884         int err = 0;
2885
2886         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2887         if (!skb)
2888                 return -ENOBUFS;
2889
2890         if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2891                                chain->index, net, skb, block, portid,
2892                                seq, flags, event, extack) <= 0) {
2893                 kfree_skb(skb);
2894                 return -EINVAL;
2895         }
2896
2897         if (unicast)
2898                 err = rtnl_unicast(skb, net, portid);
2899         else
2900                 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2901                                      flags & NLM_F_ECHO);
2902
2903         return err;
2904 }
2905
2906 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2907                                   void *tmplt_priv, u32 chain_index,
2908                                   struct tcf_block *block, struct sk_buff *oskb,
2909                                   u32 seq, u16 flags, bool unicast)
2910 {
2911         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2912         struct net *net = block->net;
2913         struct sk_buff *skb;
2914
2915         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2916         if (!skb)
2917                 return -ENOBUFS;
2918
2919         if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2920                                block, portid, seq, flags, RTM_DELCHAIN, NULL) <= 0) {
2921                 kfree_skb(skb);
2922                 return -EINVAL;
2923         }
2924
2925         if (unicast)
2926                 return rtnl_unicast(skb, net, portid);
2927
2928         return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2929 }
2930
2931 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2932                               struct nlattr **tca,
2933                               struct netlink_ext_ack *extack)
2934 {
2935         const struct tcf_proto_ops *ops;
2936         char name[IFNAMSIZ];
2937         void *tmplt_priv;
2938
2939         /* If kind is not set, user did not specify template. */
2940         if (!tca[TCA_KIND])
2941                 return 0;
2942
2943         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2944                 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
2945                 return -EINVAL;
2946         }
2947
2948         ops = tcf_proto_lookup_ops(name, true, extack);
2949         if (IS_ERR(ops))
2950                 return PTR_ERR(ops);
2951         if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2952                 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2953                 return -EOPNOTSUPP;
2954         }
2955
2956         tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2957         if (IS_ERR(tmplt_priv)) {
2958                 module_put(ops->owner);
2959                 return PTR_ERR(tmplt_priv);
2960         }
2961         chain->tmplt_ops = ops;
2962         chain->tmplt_priv = tmplt_priv;
2963         return 0;
2964 }
2965
2966 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2967                                void *tmplt_priv)
2968 {
2969         /* If template ops are set, no work to do for us. */
2970         if (!tmplt_ops)
2971                 return;
2972
2973         tmplt_ops->tmplt_destroy(tmplt_priv);
2974         module_put(tmplt_ops->owner);
2975 }
2976
2977 /* Add/delete/get a chain */
2978
2979 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2980                         struct netlink_ext_ack *extack)
2981 {
2982         struct net *net = sock_net(skb->sk);
2983         struct nlattr *tca[TCA_MAX + 1];
2984         struct tcmsg *t;
2985         u32 parent;
2986         u32 chain_index;
2987         struct Qdisc *q;
2988         struct tcf_chain *chain;
2989         struct tcf_block *block;
2990         unsigned long cl;
2991         int err;
2992
2993 replay:
2994         q = NULL;
2995         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2996                                      rtm_tca_policy, extack);
2997         if (err < 0)
2998                 return err;
2999
3000         t = nlmsg_data(n);
3001         parent = t->tcm_parent;
3002         cl = 0;
3003
3004         block = tcf_block_find(net, &q, &parent, &cl,
3005                                t->tcm_ifindex, t->tcm_block_index, extack);
3006         if (IS_ERR(block))
3007                 return PTR_ERR(block);
3008
3009         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
3010         if (chain_index > TC_ACT_EXT_VAL_MASK) {
3011                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
3012                 err = -EINVAL;
3013                 goto errout_block;
3014         }
3015
3016         mutex_lock(&block->lock);
3017         chain = tcf_chain_lookup(block, chain_index);
3018         if (n->nlmsg_type == RTM_NEWCHAIN) {
3019                 if (chain) {
3020                         if (tcf_chain_held_by_acts_only(chain)) {
3021                                 /* The chain exists only because there is
3022                                  * some action referencing it.
3023                                  */
3024                                 tcf_chain_hold(chain);
3025                         } else {
3026                                 NL_SET_ERR_MSG(extack, "Filter chain already exists");
3027                                 err = -EEXIST;
3028                                 goto errout_block_locked;
3029                         }
3030                 } else {
3031                         if (!(n->nlmsg_flags & NLM_F_CREATE)) {
3032                                 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
3033                                 err = -ENOENT;
3034                                 goto errout_block_locked;
3035                         }
3036                         chain = tcf_chain_create(block, chain_index);
3037                         if (!chain) {
3038                                 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
3039                                 err = -ENOMEM;
3040                                 goto errout_block_locked;
3041                         }
3042                 }
3043         } else {
3044                 if (!chain || tcf_chain_held_by_acts_only(chain)) {
3045                         NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
3046                         err = -EINVAL;
3047                         goto errout_block_locked;
3048                 }
3049                 tcf_chain_hold(chain);
3050         }
3051
3052         if (n->nlmsg_type == RTM_NEWCHAIN) {
3053                 /* Modifying chain requires holding parent block lock. In case
3054                  * the chain was successfully added, take a reference to the
3055                  * chain. This ensures that an empty chain does not disappear at
3056                  * the end of this function.
3057                  */
3058                 tcf_chain_hold(chain);
3059                 chain->explicitly_created = true;
3060         }
3061         mutex_unlock(&block->lock);
3062
3063         switch (n->nlmsg_type) {
3064         case RTM_NEWCHAIN:
3065                 err = tc_chain_tmplt_add(chain, net, tca, extack);
3066                 if (err) {
3067                         tcf_chain_put_explicitly_created(chain);
3068                         goto errout;
3069                 }
3070
3071                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
3072                                 RTM_NEWCHAIN, false, extack);
3073                 break;
3074         case RTM_DELCHAIN:
3075                 tfilter_notify_chain(net, skb, block, q, parent, n,
3076                                      chain, RTM_DELTFILTER, extack);
3077                 /* Flush the chain first as the user requested chain removal. */
3078                 tcf_chain_flush(chain, true);
3079                 /* In case the chain was successfully deleted, put a reference
3080                  * to the chain previously taken during addition.
3081                  */
3082                 tcf_chain_put_explicitly_created(chain);
3083                 break;
3084         case RTM_GETCHAIN:
3085                 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
3086                                       n->nlmsg_flags, n->nlmsg_type, true, extack);
3087                 if (err < 0)
3088                         NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
3089                 break;
3090         default:
3091                 err = -EOPNOTSUPP;
3092                 NL_SET_ERR_MSG(extack, "Unsupported message type");
3093                 goto errout;
3094         }
3095
3096 errout:
3097         tcf_chain_put(chain);
3098 errout_block:
3099         tcf_block_release(q, block, true);
3100         if (err == -EAGAIN)
3101                 /* Replay the request. */
3102                 goto replay;
3103         return err;
3104
3105 errout_block_locked:
3106         mutex_unlock(&block->lock);
3107         goto errout_block;
3108 }
3109
3110 /* called with RTNL */
3111 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
3112 {
3113         struct net *net = sock_net(skb->sk);
3114         struct nlattr *tca[TCA_MAX + 1];
3115         struct Qdisc *q = NULL;
3116         struct tcf_block *block;
3117         struct tcmsg *tcm = nlmsg_data(cb->nlh);
3118         struct tcf_chain *chain;
3119         long index_start;
3120         long index;
3121         int err;
3122
3123         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
3124                 return skb->len;
3125
3126         err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
3127                                      rtm_tca_policy, cb->extack);
3128         if (err)
3129                 return err;
3130
3131         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
3132                 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
3133                 if (!block)
3134                         goto out;
3135         } else {
3136                 const struct Qdisc_class_ops *cops;
3137                 struct net_device *dev;
3138                 unsigned long cl = 0;
3139
3140                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
3141                 if (!dev)
3142                         return skb->len;
3143
3144                 if (!tcm->tcm_parent)
3145                         q = rtnl_dereference(dev->qdisc);
3146                 else
3147                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
3148
3149                 if (!q)
3150                         goto out;
3151                 cops = q->ops->cl_ops;
3152                 if (!cops)
3153                         goto out;
3154                 if (!cops->tcf_block)
3155                         goto out;
3156                 if (TC_H_MIN(tcm->tcm_parent)) {
3157                         cl = cops->find(q, tcm->tcm_parent);
3158                         if (cl == 0)
3159                                 goto out;
3160                 }
3161                 block = cops->tcf_block(q, cl, NULL);
3162                 if (!block)
3163                         goto out;
3164                 if (tcf_block_shared(block))
3165                         q = NULL;
3166         }
3167
3168         index_start = cb->args[0];
3169         index = 0;
3170
3171         mutex_lock(&block->lock);
3172         list_for_each_entry(chain, &block->chain_list, list) {
3173                 if ((tca[TCA_CHAIN] &&
3174                      nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3175                         continue;
3176                 if (index < index_start) {
3177                         index++;
3178                         continue;
3179                 }
3180                 if (tcf_chain_held_by_acts_only(chain))
3181                         continue;
3182                 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3183                                          chain->index, net, skb, block,
3184                                          NETLINK_CB(cb->skb).portid,
3185                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
3186                                          RTM_NEWCHAIN, NULL);
3187                 if (err <= 0)
3188                         break;
3189                 index++;
3190         }
3191         mutex_unlock(&block->lock);
3192
3193         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3194                 tcf_block_refcnt_put(block, true);
3195         cb->args[0] = index;
3196
3197 out:
3198         /* If we did no progress, the error (EMSGSIZE) is real */
3199         if (skb->len == 0 && err)
3200                 return err;
3201         return skb->len;
3202 }
3203
3204 int tcf_exts_init_ex(struct tcf_exts *exts, struct net *net, int action,
3205                      int police, struct tcf_proto *tp, u32 handle,
3206                      bool use_action_miss)
3207 {
3208         int err = 0;
3209
3210 #ifdef CONFIG_NET_CLS_ACT
3211         exts->type = 0;
3212         exts->nr_actions = 0;
3213         exts->miss_cookie_node = NULL;
3214         /* Note: we do not own yet a reference on net.
3215          * This reference might be taken later from tcf_exts_get_net().
3216          */
3217         exts->net = net;
3218         exts->actions = kcalloc(TCA_ACT_MAX_PRIO, sizeof(struct tc_action *),
3219                                 GFP_KERNEL);
3220         if (!exts->actions)
3221                 return -ENOMEM;
3222 #endif
3223
3224         exts->action = action;
3225         exts->police = police;
3226
3227         if (!use_action_miss)
3228                 return 0;
3229
3230         err = tcf_exts_miss_cookie_base_alloc(exts, tp, handle);
3231         if (err)
3232                 goto err_miss_alloc;
3233
3234         return 0;
3235
3236 err_miss_alloc:
3237         tcf_exts_destroy(exts);
3238 #ifdef CONFIG_NET_CLS_ACT
3239         exts->actions = NULL;
3240 #endif
3241         return err;
3242 }
3243 EXPORT_SYMBOL(tcf_exts_init_ex);
3244
3245 void tcf_exts_destroy(struct tcf_exts *exts)
3246 {
3247         tcf_exts_miss_cookie_base_destroy(exts);
3248
3249 #ifdef CONFIG_NET_CLS_ACT
3250         if (exts->actions) {
3251                 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3252                 kfree(exts->actions);
3253         }
3254         exts->nr_actions = 0;
3255 #endif
3256 }
3257 EXPORT_SYMBOL(tcf_exts_destroy);
3258
3259 int tcf_exts_validate_ex(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3260                          struct nlattr *rate_tlv, struct tcf_exts *exts,
3261                          u32 flags, u32 fl_flags, struct netlink_ext_ack *extack)
3262 {
3263 #ifdef CONFIG_NET_CLS_ACT
3264         {
3265                 int init_res[TCA_ACT_MAX_PRIO] = {};
3266                 struct tc_action *act;
3267                 size_t attr_size = 0;
3268
3269                 if (exts->police && tb[exts->police]) {
3270                         struct tc_action_ops *a_o;
3271
3272                         a_o = tc_action_load_ops(tb[exts->police], true,
3273                                                  !(flags & TCA_ACT_FLAGS_NO_RTNL),
3274                                                  extack);
3275                         if (IS_ERR(a_o))
3276                                 return PTR_ERR(a_o);
3277                         flags |= TCA_ACT_FLAGS_POLICE | TCA_ACT_FLAGS_BIND;
3278                         act = tcf_action_init_1(net, tp, tb[exts->police],
3279                                                 rate_tlv, a_o, init_res, flags,
3280                                                 extack);
3281                         module_put(a_o->owner);
3282                         if (IS_ERR(act))
3283                                 return PTR_ERR(act);
3284
3285                         act->type = exts->type = TCA_OLD_COMPAT;
3286                         exts->actions[0] = act;
3287                         exts->nr_actions = 1;
3288                         tcf_idr_insert_many(exts->actions);
3289                 } else if (exts->action && tb[exts->action]) {
3290                         int err;
3291
3292                         flags |= TCA_ACT_FLAGS_BIND;
3293                         err = tcf_action_init(net, tp, tb[exts->action],
3294                                               rate_tlv, exts->actions, init_res,
3295                                               &attr_size, flags, fl_flags,
3296                                               extack);
3297                         if (err < 0)
3298                                 return err;
3299                         exts->nr_actions = err;
3300                 }
3301         }
3302 #else
3303         if ((exts->action && tb[exts->action]) ||
3304             (exts->police && tb[exts->police])) {
3305                 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3306                 return -EOPNOTSUPP;
3307         }
3308 #endif
3309
3310         return 0;
3311 }
3312 EXPORT_SYMBOL(tcf_exts_validate_ex);
3313
3314 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3315                       struct nlattr *rate_tlv, struct tcf_exts *exts,
3316                       u32 flags, struct netlink_ext_ack *extack)
3317 {
3318         return tcf_exts_validate_ex(net, tp, tb, rate_tlv, exts,
3319                                     flags, 0, extack);
3320 }
3321 EXPORT_SYMBOL(tcf_exts_validate);
3322
3323 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3324 {
3325 #ifdef CONFIG_NET_CLS_ACT
3326         struct tcf_exts old = *dst;
3327
3328         *dst = *src;
3329         tcf_exts_destroy(&old);
3330 #endif
3331 }
3332 EXPORT_SYMBOL(tcf_exts_change);
3333
3334 #ifdef CONFIG_NET_CLS_ACT
3335 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3336 {
3337         if (exts->nr_actions == 0)
3338                 return NULL;
3339         else
3340                 return exts->actions[0];
3341 }
3342 #endif
3343
3344 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3345 {
3346 #ifdef CONFIG_NET_CLS_ACT
3347         struct nlattr *nest;
3348
3349         if (exts->action && tcf_exts_has_actions(exts)) {
3350                 /*
3351                  * again for backward compatible mode - we want
3352                  * to work with both old and new modes of entering
3353                  * tc data even if iproute2  was newer - jhs
3354                  */
3355                 if (exts->type != TCA_OLD_COMPAT) {
3356                         nest = nla_nest_start_noflag(skb, exts->action);
3357                         if (nest == NULL)
3358                                 goto nla_put_failure;
3359
3360                         if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3361                             < 0)
3362                                 goto nla_put_failure;
3363                         nla_nest_end(skb, nest);
3364                 } else if (exts->police) {
3365                         struct tc_action *act = tcf_exts_first_act(exts);
3366                         nest = nla_nest_start_noflag(skb, exts->police);
3367                         if (nest == NULL || !act)
3368                                 goto nla_put_failure;
3369                         if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3370                                 goto nla_put_failure;
3371                         nla_nest_end(skb, nest);
3372                 }
3373         }
3374         return 0;
3375
3376 nla_put_failure:
3377         nla_nest_cancel(skb, nest);
3378         return -1;
3379 #else
3380         return 0;
3381 #endif
3382 }
3383 EXPORT_SYMBOL(tcf_exts_dump);
3384
3385 int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3386 {
3387 #ifdef CONFIG_NET_CLS_ACT
3388         struct nlattr *nest;
3389
3390         if (!exts->action || !tcf_exts_has_actions(exts))
3391                 return 0;
3392
3393         nest = nla_nest_start_noflag(skb, exts->action);
3394         if (!nest)
3395                 goto nla_put_failure;
3396
3397         if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3398                 goto nla_put_failure;
3399         nla_nest_end(skb, nest);
3400         return 0;
3401
3402 nla_put_failure:
3403         nla_nest_cancel(skb, nest);
3404         return -1;
3405 #else
3406         return 0;
3407 #endif
3408 }
3409 EXPORT_SYMBOL(tcf_exts_terse_dump);
3410
3411 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3412 {
3413 #ifdef CONFIG_NET_CLS_ACT
3414         struct tc_action *a = tcf_exts_first_act(exts);
3415         if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3416                 return -1;
3417 #endif
3418         return 0;
3419 }
3420 EXPORT_SYMBOL(tcf_exts_dump_stats);
3421
3422 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3423 {
3424         if (*flags & TCA_CLS_FLAGS_IN_HW)
3425                 return;
3426         *flags |= TCA_CLS_FLAGS_IN_HW;
3427         atomic_inc(&block->offloadcnt);
3428 }
3429
3430 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3431 {
3432         if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3433                 return;
3434         *flags &= ~TCA_CLS_FLAGS_IN_HW;
3435         atomic_dec(&block->offloadcnt);
3436 }
3437
3438 static void tc_cls_offload_cnt_update(struct tcf_block *block,
3439                                       struct tcf_proto *tp, u32 *cnt,
3440                                       u32 *flags, u32 diff, bool add)
3441 {
3442         lockdep_assert_held(&block->cb_lock);
3443
3444         spin_lock(&tp->lock);
3445         if (add) {
3446                 if (!*cnt)
3447                         tcf_block_offload_inc(block, flags);
3448                 *cnt += diff;
3449         } else {
3450                 *cnt -= diff;
3451                 if (!*cnt)
3452                         tcf_block_offload_dec(block, flags);
3453         }
3454         spin_unlock(&tp->lock);
3455 }
3456
3457 static void
3458 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3459                          u32 *cnt, u32 *flags)
3460 {
3461         lockdep_assert_held(&block->cb_lock);
3462
3463         spin_lock(&tp->lock);
3464         tcf_block_offload_dec(block, flags);
3465         *cnt = 0;
3466         spin_unlock(&tp->lock);
3467 }
3468
3469 static int
3470 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3471                    void *type_data, bool err_stop)
3472 {
3473         struct flow_block_cb *block_cb;
3474         int ok_count = 0;
3475         int err;
3476
3477         list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3478                 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3479                 if (err) {
3480                         if (err_stop)
3481                                 return err;
3482                 } else {
3483                         ok_count++;
3484                 }
3485         }
3486         return ok_count;
3487 }
3488
3489 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3490                      void *type_data, bool err_stop, bool rtnl_held)
3491 {
3492         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3493         int ok_count;
3494
3495 retry:
3496         if (take_rtnl)
3497                 rtnl_lock();
3498         down_read(&block->cb_lock);
3499         /* Need to obtain rtnl lock if block is bound to devs that require it.
3500          * In block bind code cb_lock is obtained while holding rtnl, so we must
3501          * obtain the locks in same order here.
3502          */
3503         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3504                 up_read(&block->cb_lock);
3505                 take_rtnl = true;
3506                 goto retry;
3507         }
3508
3509         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3510
3511         up_read(&block->cb_lock);
3512         if (take_rtnl)
3513                 rtnl_unlock();
3514         return ok_count;
3515 }
3516 EXPORT_SYMBOL(tc_setup_cb_call);
3517
3518 /* Non-destructive filter add. If filter that wasn't already in hardware is
3519  * successfully offloaded, increment block offloads counter. On failure,
3520  * previously offloaded filter is considered to be intact and offloads counter
3521  * is not decremented.
3522  */
3523
3524 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3525                     enum tc_setup_type type, void *type_data, bool err_stop,
3526                     u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3527 {
3528         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3529         int ok_count;
3530
3531 retry:
3532         if (take_rtnl)
3533                 rtnl_lock();
3534         down_read(&block->cb_lock);
3535         /* Need to obtain rtnl lock if block is bound to devs that require it.
3536          * In block bind code cb_lock is obtained while holding rtnl, so we must
3537          * obtain the locks in same order here.
3538          */
3539         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3540                 up_read(&block->cb_lock);
3541                 take_rtnl = true;
3542                 goto retry;
3543         }
3544
3545         /* Make sure all netdevs sharing this block are offload-capable. */
3546         if (block->nooffloaddevcnt && err_stop) {
3547                 ok_count = -EOPNOTSUPP;
3548                 goto err_unlock;
3549         }
3550
3551         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3552         if (ok_count < 0)
3553                 goto err_unlock;
3554
3555         if (tp->ops->hw_add)
3556                 tp->ops->hw_add(tp, type_data);
3557         if (ok_count > 0)
3558                 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3559                                           ok_count, true);
3560 err_unlock:
3561         up_read(&block->cb_lock);
3562         if (take_rtnl)
3563                 rtnl_unlock();
3564         return min(ok_count, 0);
3565 }
3566 EXPORT_SYMBOL(tc_setup_cb_add);
3567
3568 /* Destructive filter replace. If filter that wasn't already in hardware is
3569  * successfully offloaded, increment block offload counter. On failure,
3570  * previously offloaded filter is considered to be destroyed and offload counter
3571  * is decremented.
3572  */
3573
3574 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3575                         enum tc_setup_type type, void *type_data, bool err_stop,
3576                         u32 *old_flags, unsigned int *old_in_hw_count,
3577                         u32 *new_flags, unsigned int *new_in_hw_count,
3578                         bool rtnl_held)
3579 {
3580         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3581         int ok_count;
3582
3583 retry:
3584         if (take_rtnl)
3585                 rtnl_lock();
3586         down_read(&block->cb_lock);
3587         /* Need to obtain rtnl lock if block is bound to devs that require it.
3588          * In block bind code cb_lock is obtained while holding rtnl, so we must
3589          * obtain the locks in same order here.
3590          */
3591         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3592                 up_read(&block->cb_lock);
3593                 take_rtnl = true;
3594                 goto retry;
3595         }
3596
3597         /* Make sure all netdevs sharing this block are offload-capable. */
3598         if (block->nooffloaddevcnt && err_stop) {
3599                 ok_count = -EOPNOTSUPP;
3600                 goto err_unlock;
3601         }
3602
3603         tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3604         if (tp->ops->hw_del)
3605                 tp->ops->hw_del(tp, type_data);
3606
3607         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3608         if (ok_count < 0)
3609                 goto err_unlock;
3610
3611         if (tp->ops->hw_add)
3612                 tp->ops->hw_add(tp, type_data);
3613         if (ok_count > 0)
3614                 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3615                                           new_flags, ok_count, true);
3616 err_unlock:
3617         up_read(&block->cb_lock);
3618         if (take_rtnl)
3619                 rtnl_unlock();
3620         return min(ok_count, 0);
3621 }
3622 EXPORT_SYMBOL(tc_setup_cb_replace);
3623
3624 /* Destroy filter and decrement block offload counter, if filter was previously
3625  * offloaded.
3626  */
3627
3628 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3629                         enum tc_setup_type type, void *type_data, bool err_stop,
3630                         u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3631 {
3632         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3633         int ok_count;
3634
3635 retry:
3636         if (take_rtnl)
3637                 rtnl_lock();
3638         down_read(&block->cb_lock);
3639         /* Need to obtain rtnl lock if block is bound to devs that require it.
3640          * In block bind code cb_lock is obtained while holding rtnl, so we must
3641          * obtain the locks in same order here.
3642          */
3643         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3644                 up_read(&block->cb_lock);
3645                 take_rtnl = true;
3646                 goto retry;
3647         }
3648
3649         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3650
3651         tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3652         if (tp->ops->hw_del)
3653                 tp->ops->hw_del(tp, type_data);
3654
3655         up_read(&block->cb_lock);
3656         if (take_rtnl)
3657                 rtnl_unlock();
3658         return min(ok_count, 0);
3659 }
3660 EXPORT_SYMBOL(tc_setup_cb_destroy);
3661
3662 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3663                           bool add, flow_setup_cb_t *cb,
3664                           enum tc_setup_type type, void *type_data,
3665                           void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3666 {
3667         int err = cb(type, type_data, cb_priv);
3668
3669         if (err) {
3670                 if (add && tc_skip_sw(*flags))
3671                         return err;
3672         } else {
3673                 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3674                                           add);
3675         }
3676
3677         return 0;
3678 }
3679 EXPORT_SYMBOL(tc_setup_cb_reoffload);
3680
3681 static int tcf_act_get_user_cookie(struct flow_action_entry *entry,
3682                                    const struct tc_action *act)
3683 {
3684         struct tc_cookie *user_cookie;
3685         int err = 0;
3686
3687         rcu_read_lock();
3688         user_cookie = rcu_dereference(act->user_cookie);
3689         if (user_cookie) {
3690                 entry->user_cookie = flow_action_cookie_create(user_cookie->data,
3691                                                                user_cookie->len,
3692                                                                GFP_ATOMIC);
3693                 if (!entry->user_cookie)
3694                         err = -ENOMEM;
3695         }
3696         rcu_read_unlock();
3697         return err;
3698 }
3699
3700 static void tcf_act_put_user_cookie(struct flow_action_entry *entry)
3701 {
3702         flow_action_cookie_destroy(entry->user_cookie);
3703 }
3704
3705 void tc_cleanup_offload_action(struct flow_action *flow_action)
3706 {
3707         struct flow_action_entry *entry;
3708         int i;
3709
3710         flow_action_for_each(i, entry, flow_action) {
3711                 tcf_act_put_user_cookie(entry);
3712                 if (entry->destructor)
3713                         entry->destructor(entry->destructor_priv);
3714         }
3715 }
3716 EXPORT_SYMBOL(tc_cleanup_offload_action);
3717
3718 static int tc_setup_offload_act(struct tc_action *act,
3719                                 struct flow_action_entry *entry,
3720                                 u32 *index_inc,
3721                                 struct netlink_ext_ack *extack)
3722 {
3723 #ifdef CONFIG_NET_CLS_ACT
3724         if (act->ops->offload_act_setup) {
3725                 return act->ops->offload_act_setup(act, entry, index_inc, true,
3726                                                    extack);
3727         } else {
3728                 NL_SET_ERR_MSG(extack, "Action does not support offload");
3729                 return -EOPNOTSUPP;
3730         }
3731 #else
3732         return 0;
3733 #endif
3734 }
3735
3736 int tc_setup_action(struct flow_action *flow_action,
3737                     struct tc_action *actions[],
3738                     u32 miss_cookie_base,
3739                     struct netlink_ext_ack *extack)
3740 {
3741         int i, j, k, index, err = 0;
3742         struct tc_action *act;
3743
3744         BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3745         BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3746         BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3747
3748         if (!actions)
3749                 return 0;
3750
3751         j = 0;
3752         tcf_act_for_each_action(i, act, actions) {
3753                 struct flow_action_entry *entry;
3754
3755                 entry = &flow_action->entries[j];
3756                 spin_lock_bh(&act->tcfa_lock);
3757                 err = tcf_act_get_user_cookie(entry, act);
3758                 if (err)
3759                         goto err_out_locked;
3760
3761                 index = 0;
3762                 err = tc_setup_offload_act(act, entry, &index, extack);
3763                 if (err)
3764                         goto err_out_locked;
3765
3766                 for (k = 0; k < index ; k++) {
3767                         entry[k].hw_stats = tc_act_hw_stats(act->hw_stats);
3768                         entry[k].hw_index = act->tcfa_index;
3769                         entry[k].cookie = (unsigned long)act;
3770                         entry[k].miss_cookie =
3771                                 tcf_exts_miss_cookie_get(miss_cookie_base, i);
3772                 }
3773
3774                 j += index;
3775
3776                 spin_unlock_bh(&act->tcfa_lock);
3777         }
3778
3779 err_out:
3780         if (err)
3781                 tc_cleanup_offload_action(flow_action);
3782
3783         return err;
3784 err_out_locked:
3785         spin_unlock_bh(&act->tcfa_lock);
3786         goto err_out;
3787 }
3788
3789 int tc_setup_offload_action(struct flow_action *flow_action,
3790                             const struct tcf_exts *exts,
3791                             struct netlink_ext_ack *extack)
3792 {
3793 #ifdef CONFIG_NET_CLS_ACT
3794         u32 miss_cookie_base;
3795
3796         if (!exts)
3797                 return 0;
3798
3799         miss_cookie_base = exts->miss_cookie_node ?
3800                            exts->miss_cookie_node->miss_cookie_base : 0;
3801         return tc_setup_action(flow_action, exts->actions, miss_cookie_base,
3802                                extack);
3803 #else
3804         return 0;
3805 #endif
3806 }
3807 EXPORT_SYMBOL(tc_setup_offload_action);
3808
3809 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3810 {
3811         unsigned int num_acts = 0;
3812         struct tc_action *act;
3813         int i;
3814
3815         tcf_exts_for_each_action(i, act, exts) {
3816                 if (is_tcf_pedit(act))
3817                         num_acts += tcf_pedit_nkeys(act);
3818                 else
3819                         num_acts++;
3820         }
3821         return num_acts;
3822 }
3823 EXPORT_SYMBOL(tcf_exts_num_actions);
3824
3825 #ifdef CONFIG_NET_CLS_ACT
3826 static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3827                                         u32 *p_block_index,
3828                                         struct netlink_ext_ack *extack)
3829 {
3830         *p_block_index = nla_get_u32(block_index_attr);
3831         if (!*p_block_index) {
3832                 NL_SET_ERR_MSG(extack, "Block number may not be zero");
3833                 return -EINVAL;
3834         }
3835
3836         return 0;
3837 }
3838
3839 int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3840                     enum flow_block_binder_type binder_type,
3841                     struct nlattr *block_index_attr,
3842                     struct netlink_ext_ack *extack)
3843 {
3844         u32 block_index;
3845         int err;
3846
3847         if (!block_index_attr)
3848                 return 0;
3849
3850         err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3851         if (err)
3852                 return err;
3853
3854         qe->info.binder_type = binder_type;
3855         qe->info.chain_head_change = tcf_chain_head_change_dflt;
3856         qe->info.chain_head_change_priv = &qe->filter_chain;
3857         qe->info.block_index = block_index;
3858
3859         return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3860 }
3861 EXPORT_SYMBOL(tcf_qevent_init);
3862
3863 void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3864 {
3865         if (qe->info.block_index)
3866                 tcf_block_put_ext(qe->block, sch, &qe->info);
3867 }
3868 EXPORT_SYMBOL(tcf_qevent_destroy);
3869
3870 int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3871                                struct netlink_ext_ack *extack)
3872 {
3873         u32 block_index;
3874         int err;
3875
3876         if (!block_index_attr)
3877                 return 0;
3878
3879         err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3880         if (err)
3881                 return err;
3882
3883         /* Bounce newly-configured block or change in block. */
3884         if (block_index != qe->info.block_index) {
3885                 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3886                 return -EINVAL;
3887         }
3888
3889         return 0;
3890 }
3891 EXPORT_SYMBOL(tcf_qevent_validate_change);
3892
3893 struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
3894                                   struct sk_buff **to_free, int *ret)
3895 {
3896         struct tcf_result cl_res;
3897         struct tcf_proto *fl;
3898
3899         if (!qe->info.block_index)
3900                 return skb;
3901
3902         fl = rcu_dereference_bh(qe->filter_chain);
3903
3904         switch (tcf_classify(skb, NULL, fl, &cl_res, false)) {
3905         case TC_ACT_SHOT:
3906                 qdisc_qstats_drop(sch);
3907                 __qdisc_drop(skb, to_free);
3908                 *ret = __NET_XMIT_BYPASS;
3909                 return NULL;
3910         case TC_ACT_STOLEN:
3911         case TC_ACT_QUEUED:
3912         case TC_ACT_TRAP:
3913                 __qdisc_drop(skb, to_free);
3914                 *ret = __NET_XMIT_STOLEN;
3915                 return NULL;
3916         case TC_ACT_REDIRECT:
3917                 skb_do_redirect(skb);
3918                 *ret = __NET_XMIT_STOLEN;
3919                 return NULL;
3920         }
3921
3922         return skb;
3923 }
3924 EXPORT_SYMBOL(tcf_qevent_handle);
3925
3926 int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
3927 {
3928         if (!qe->info.block_index)
3929                 return 0;
3930         return nla_put_u32(skb, attr_name, qe->info.block_index);
3931 }
3932 EXPORT_SYMBOL(tcf_qevent_dump);
3933 #endif
3934
3935 static __net_init int tcf_net_init(struct net *net)
3936 {
3937         struct tcf_net *tn = net_generic(net, tcf_net_id);
3938
3939         spin_lock_init(&tn->idr_lock);
3940         idr_init(&tn->idr);
3941         return 0;
3942 }
3943
3944 static void __net_exit tcf_net_exit(struct net *net)
3945 {
3946         struct tcf_net *tn = net_generic(net, tcf_net_id);
3947
3948         idr_destroy(&tn->idr);
3949 }
3950
3951 static struct pernet_operations tcf_net_ops = {
3952         .init = tcf_net_init,
3953         .exit = tcf_net_exit,
3954         .id   = &tcf_net_id,
3955         .size = sizeof(struct tcf_net),
3956 };
3957
3958 static int __init tc_filter_init(void)
3959 {
3960         int err;
3961
3962         tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3963         if (!tc_filter_wq)
3964                 return -ENOMEM;
3965
3966         err = register_pernet_subsys(&tcf_net_ops);
3967         if (err)
3968                 goto err_register_pernet_subsys;
3969
3970         xa_init_flags(&tcf_exts_miss_cookies_xa, XA_FLAGS_ALLOC1);
3971
3972         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3973                       RTNL_FLAG_DOIT_UNLOCKED);
3974         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3975                       RTNL_FLAG_DOIT_UNLOCKED);
3976         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3977                       tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3978         rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3979         rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3980         rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3981                       tc_dump_chain, 0);
3982
3983         return 0;
3984
3985 err_register_pernet_subsys:
3986         destroy_workqueue(tc_filter_wq);
3987         return err;
3988 }
3989
3990 subsys_initcall(tc_filter_init);