Linux Audio

Check our new training course

Yocto / OpenEmbedded training

Feb 10-13, 2025
Register
Loading...
v6.8
   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 */
  47static LIST_HEAD(tcf_proto_base);
  48
  49/* Protects list of registered TC modules. It is pure SMP lock. */
  50static DEFINE_RWLOCK(cls_mod_lock);
  51
  52static struct xarray tcf_exts_miss_cookies_xa;
  53struct 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 */
  67union 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)
  76static int
  77tcf_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
 106err_xa_alloc:
 107	kfree(n);
 108	return err;
 109}
 110
 111static 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
 123static struct tcf_exts_miss_cookie_node *
 124tcf_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) */
 132static int
 133tcf_exts_miss_cookie_base_alloc(struct tcf_exts *exts, struct tcf_proto *tp,
 134				u32 handle)
 135{
 136	return 0;
 137}
 138
 139static void tcf_exts_miss_cookie_base_destroy(struct tcf_exts *exts)
 140{
 141}
 142#endif /* IS_ENABLED(CONFIG_NET_TC_SKB_EXT) */
 143
 144static 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
 156DEFINE_STATIC_KEY_FALSE(tc_skb_ext_tc);
 157EXPORT_SYMBOL(tc_skb_ext_tc);
 158
 159void tc_skb_ext_tc_enable(void)
 160{
 161	static_branch_inc(&tc_skb_ext_tc);
 162}
 163EXPORT_SYMBOL(tc_skb_ext_tc_enable);
 164
 165void tc_skb_ext_tc_disable(void)
 166{
 167	static_branch_dec(&tc_skb_ext_tc);
 168}
 169EXPORT_SYMBOL(tc_skb_ext_tc_disable);
 170#endif
 171
 172static 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
 178static 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
 189static 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
 197static 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
 217static void
 218tcf_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
 230static 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
 248static const struct tcf_proto_ops *
 249tcf_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
 280int 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;
 292out:
 293	write_unlock(&cls_mod_lock);
 294	return rc;
 295}
 296EXPORT_SYMBOL(register_tcf_proto_ops);
 297
 298static struct workqueue_struct *tc_filter_wq;
 299
 300void 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}
 323EXPORT_SYMBOL(unregister_tcf_proto_ops);
 324
 325bool 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}
 330EXPORT_SYMBOL(tcf_queue_work);
 331
 332/* Select new prio value from the range, managed by kernel. */
 333
 334static 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
 344static 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
 352static 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
 372static 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
 403errout:
 404	kfree(tp);
 405	return ERR_PTR(err);
 406}
 407
 408static void tcf_proto_get(struct tcf_proto *tp)
 409{
 410	refcount_inc(&tp->refcnt);
 411}
 412
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 413static void tcf_chain_put(struct tcf_chain *chain);
 414
 415static 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
 426static 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
 433static 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
 442static 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
 449static 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
 463struct 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
 469static 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
 489static 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
 496static 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
 513static 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
 530static void tcf_block_destroy(struct tcf_block *block)
 531{
 532	mutex_destroy(&block->lock);
 533	mutex_destroy(&block->proto_destroy_lock);
 534	xa_destroy(&block->ports);
 535	kfree_rcu(block, rcu);
 536}
 537
 538static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
 539{
 540	struct tcf_block *block = chain->block;
 541
 542	mutex_destroy(&chain->filter_chain_lock);
 543	kfree_rcu(chain, rcu);
 544	if (free_block)
 545		tcf_block_destroy(block);
 546}
 547
 548static void tcf_chain_hold(struct tcf_chain *chain)
 549{
 550	ASSERT_BLOCK_LOCKED(chain->block);
 551
 552	++chain->refcnt;
 553}
 554
 555static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
 556{
 557	ASSERT_BLOCK_LOCKED(chain->block);
 558
 559	/* In case all the references are action references, this
 560	 * chain should not be shown to the user.
 561	 */
 562	return chain->refcnt == chain->action_refcnt;
 563}
 564
 565static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
 566					  u32 chain_index)
 567{
 568	struct tcf_chain *chain;
 569
 570	ASSERT_BLOCK_LOCKED(block);
 571
 572	list_for_each_entry(chain, &block->chain_list, list) {
 573		if (chain->index == chain_index)
 574			return chain;
 575	}
 576	return NULL;
 577}
 578
 579#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
 580static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
 581					      u32 chain_index)
 582{
 583	struct tcf_chain *chain;
 584
 585	list_for_each_entry_rcu(chain, &block->chain_list, list) {
 586		if (chain->index == chain_index)
 587			return chain;
 588	}
 589	return NULL;
 590}
 591#endif
 592
 593static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
 594			   u32 seq, u16 flags, int event, bool unicast,
 595			   struct netlink_ext_ack *extack);
 596
 597static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
 598					 u32 chain_index, bool create,
 599					 bool by_act)
 600{
 601	struct tcf_chain *chain = NULL;
 602	bool is_first_reference;
 603
 604	mutex_lock(&block->lock);
 605	chain = tcf_chain_lookup(block, chain_index);
 606	if (chain) {
 607		tcf_chain_hold(chain);
 608	} else {
 609		if (!create)
 610			goto errout;
 611		chain = tcf_chain_create(block, chain_index);
 612		if (!chain)
 613			goto errout;
 614	}
 615
 616	if (by_act)
 617		++chain->action_refcnt;
 618	is_first_reference = chain->refcnt - chain->action_refcnt == 1;
 619	mutex_unlock(&block->lock);
 620
 621	/* Send notification only in case we got the first
 622	 * non-action reference. Until then, the chain acts only as
 623	 * a placeholder for actions pointing to it and user ought
 624	 * not know about them.
 625	 */
 626	if (is_first_reference && !by_act)
 627		tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
 628				RTM_NEWCHAIN, false, NULL);
 629
 630	return chain;
 631
 632errout:
 633	mutex_unlock(&block->lock);
 634	return chain;
 635}
 636
 637static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
 638				       bool create)
 639{
 640	return __tcf_chain_get(block, chain_index, create, false);
 641}
 642
 643struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
 644{
 645	return __tcf_chain_get(block, chain_index, true, true);
 646}
 647EXPORT_SYMBOL(tcf_chain_get_by_act);
 648
 649static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
 650			       void *tmplt_priv);
 651static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
 652				  void *tmplt_priv, u32 chain_index,
 653				  struct tcf_block *block, struct sk_buff *oskb,
 654				  u32 seq, u16 flags);
 655
 656static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
 657			    bool explicitly_created)
 658{
 659	struct tcf_block *block = chain->block;
 660	const struct tcf_proto_ops *tmplt_ops;
 661	unsigned int refcnt, non_act_refcnt;
 662	bool free_block = false;
 663	void *tmplt_priv;
 664
 665	mutex_lock(&block->lock);
 666	if (explicitly_created) {
 667		if (!chain->explicitly_created) {
 668			mutex_unlock(&block->lock);
 669			return;
 670		}
 671		chain->explicitly_created = false;
 672	}
 673
 674	if (by_act)
 675		chain->action_refcnt--;
 676
 677	/* tc_chain_notify_delete can't be called while holding block lock.
 678	 * However, when block is unlocked chain can be changed concurrently, so
 679	 * save these to temporary variables.
 680	 */
 681	refcnt = --chain->refcnt;
 682	non_act_refcnt = refcnt - chain->action_refcnt;
 683	tmplt_ops = chain->tmplt_ops;
 684	tmplt_priv = chain->tmplt_priv;
 685
 686	if (non_act_refcnt == chain->explicitly_created && !by_act) {
 687		if (non_act_refcnt == 0)
 688			tc_chain_notify_delete(tmplt_ops, tmplt_priv,
 689					       chain->index, block, NULL, 0, 0);
 690		/* Last reference to chain, no need to lock. */
 691		chain->flushing = false;
 692	}
 693
 694	if (refcnt == 0)
 695		free_block = tcf_chain_detach(chain);
 696	mutex_unlock(&block->lock);
 697
 698	if (refcnt == 0) {
 699		tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
 700		tcf_chain_destroy(chain, free_block);
 701	}
 702}
 703
 704static void tcf_chain_put(struct tcf_chain *chain)
 705{
 706	__tcf_chain_put(chain, false, false);
 707}
 708
 709void tcf_chain_put_by_act(struct tcf_chain *chain)
 710{
 711	__tcf_chain_put(chain, true, false);
 712}
 713EXPORT_SYMBOL(tcf_chain_put_by_act);
 714
 715static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
 716{
 717	__tcf_chain_put(chain, false, true);
 718}
 719
 720static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
 721{
 722	struct tcf_proto *tp, *tp_next;
 723
 724	mutex_lock(&chain->filter_chain_lock);
 725	tp = tcf_chain_dereference(chain->filter_chain, chain);
 726	while (tp) {
 727		tp_next = rcu_dereference_protected(tp->next, 1);
 728		tcf_proto_signal_destroying(chain, tp);
 729		tp = tp_next;
 730	}
 731	tp = tcf_chain_dereference(chain->filter_chain, chain);
 732	RCU_INIT_POINTER(chain->filter_chain, NULL);
 733	tcf_chain0_head_change(chain, NULL);
 734	chain->flushing = true;
 735	mutex_unlock(&chain->filter_chain_lock);
 736
 737	while (tp) {
 738		tp_next = rcu_dereference_protected(tp->next, 1);
 739		tcf_proto_put(tp, rtnl_held, NULL);
 740		tp = tp_next;
 741	}
 742}
 743
 744static int tcf_block_setup(struct tcf_block *block,
 745			   struct flow_block_offload *bo);
 746
 747static void tcf_block_offload_init(struct flow_block_offload *bo,
 748				   struct net_device *dev, struct Qdisc *sch,
 749				   enum flow_block_command command,
 750				   enum flow_block_binder_type binder_type,
 751				   struct flow_block *flow_block,
 752				   bool shared, struct netlink_ext_ack *extack)
 753{
 754	bo->net = dev_net(dev);
 755	bo->command = command;
 756	bo->binder_type = binder_type;
 757	bo->block = flow_block;
 758	bo->block_shared = shared;
 759	bo->extack = extack;
 760	bo->sch = sch;
 761	bo->cb_list_head = &flow_block->cb_list;
 762	INIT_LIST_HEAD(&bo->cb_list);
 763}
 764
 765static void tcf_block_unbind(struct tcf_block *block,
 766			     struct flow_block_offload *bo);
 767
 768static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
 769{
 770	struct tcf_block *block = block_cb->indr.data;
 771	struct net_device *dev = block_cb->indr.dev;
 772	struct Qdisc *sch = block_cb->indr.sch;
 773	struct netlink_ext_ack extack = {};
 774	struct flow_block_offload bo = {};
 775
 776	tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
 777			       block_cb->indr.binder_type,
 778			       &block->flow_block, tcf_block_shared(block),
 779			       &extack);
 780	rtnl_lock();
 781	down_write(&block->cb_lock);
 782	list_del(&block_cb->driver_list);
 783	list_move(&block_cb->list, &bo.cb_list);
 784	tcf_block_unbind(block, &bo);
 785	up_write(&block->cb_lock);
 786	rtnl_unlock();
 787}
 788
 789static bool tcf_block_offload_in_use(struct tcf_block *block)
 790{
 791	return atomic_read(&block->offloadcnt);
 792}
 793
 794static int tcf_block_offload_cmd(struct tcf_block *block,
 795				 struct net_device *dev, struct Qdisc *sch,
 796				 struct tcf_block_ext_info *ei,
 797				 enum flow_block_command command,
 798				 struct netlink_ext_ack *extack)
 799{
 800	struct flow_block_offload bo = {};
 801
 802	tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
 803			       &block->flow_block, tcf_block_shared(block),
 804			       extack);
 805
 806	if (dev->netdev_ops->ndo_setup_tc) {
 807		int err;
 808
 809		err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
 810		if (err < 0) {
 811			if (err != -EOPNOTSUPP)
 812				NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
 813			return err;
 814		}
 815
 816		return tcf_block_setup(block, &bo);
 817	}
 818
 819	flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
 820				    tc_block_indr_cleanup);
 821	tcf_block_setup(block, &bo);
 822
 823	return -EOPNOTSUPP;
 824}
 825
 826static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
 827				  struct tcf_block_ext_info *ei,
 828				  struct netlink_ext_ack *extack)
 829{
 830	struct net_device *dev = q->dev_queue->dev;
 831	int err;
 832
 833	down_write(&block->cb_lock);
 834
 835	/* If tc offload feature is disabled and the block we try to bind
 836	 * to already has some offloaded filters, forbid to bind.
 837	 */
 838	if (dev->netdev_ops->ndo_setup_tc &&
 839	    !tc_can_offload(dev) &&
 840	    tcf_block_offload_in_use(block)) {
 841		NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
 842		err = -EOPNOTSUPP;
 843		goto err_unlock;
 844	}
 845
 846	err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
 847	if (err == -EOPNOTSUPP)
 848		goto no_offload_dev_inc;
 849	if (err)
 850		goto err_unlock;
 851
 852	up_write(&block->cb_lock);
 853	return 0;
 854
 855no_offload_dev_inc:
 856	if (tcf_block_offload_in_use(block))
 857		goto err_unlock;
 858
 859	err = 0;
 860	block->nooffloaddevcnt++;
 861err_unlock:
 862	up_write(&block->cb_lock);
 863	return err;
 864}
 865
 866static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
 867				     struct tcf_block_ext_info *ei)
 868{
 869	struct net_device *dev = q->dev_queue->dev;
 870	int err;
 871
 872	down_write(&block->cb_lock);
 873	err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
 874	if (err == -EOPNOTSUPP)
 875		goto no_offload_dev_dec;
 876	up_write(&block->cb_lock);
 877	return;
 878
 879no_offload_dev_dec:
 880	WARN_ON(block->nooffloaddevcnt-- == 0);
 881	up_write(&block->cb_lock);
 882}
 883
 884static int
 885tcf_chain0_head_change_cb_add(struct tcf_block *block,
 886			      struct tcf_block_ext_info *ei,
 887			      struct netlink_ext_ack *extack)
 888{
 889	struct tcf_filter_chain_list_item *item;
 890	struct tcf_chain *chain0;
 891
 892	item = kmalloc(sizeof(*item), GFP_KERNEL);
 893	if (!item) {
 894		NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
 895		return -ENOMEM;
 896	}
 897	item->chain_head_change = ei->chain_head_change;
 898	item->chain_head_change_priv = ei->chain_head_change_priv;
 899
 900	mutex_lock(&block->lock);
 901	chain0 = block->chain0.chain;
 902	if (chain0)
 903		tcf_chain_hold(chain0);
 904	else
 905		list_add(&item->list, &block->chain0.filter_chain_list);
 906	mutex_unlock(&block->lock);
 907
 908	if (chain0) {
 909		struct tcf_proto *tp_head;
 910
 911		mutex_lock(&chain0->filter_chain_lock);
 912
 913		tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
 914		if (tp_head)
 915			tcf_chain_head_change_item(item, tp_head);
 916
 917		mutex_lock(&block->lock);
 918		list_add(&item->list, &block->chain0.filter_chain_list);
 919		mutex_unlock(&block->lock);
 920
 921		mutex_unlock(&chain0->filter_chain_lock);
 922		tcf_chain_put(chain0);
 923	}
 924
 925	return 0;
 926}
 927
 928static void
 929tcf_chain0_head_change_cb_del(struct tcf_block *block,
 930			      struct tcf_block_ext_info *ei)
 931{
 932	struct tcf_filter_chain_list_item *item;
 933
 934	mutex_lock(&block->lock);
 935	list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
 936		if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
 937		    (item->chain_head_change == ei->chain_head_change &&
 938		     item->chain_head_change_priv == ei->chain_head_change_priv)) {
 939			if (block->chain0.chain)
 940				tcf_chain_head_change_item(item, NULL);
 941			list_del(&item->list);
 942			mutex_unlock(&block->lock);
 943
 944			kfree(item);
 945			return;
 946		}
 947	}
 948	mutex_unlock(&block->lock);
 949	WARN_ON(1);
 950}
 951
 952struct tcf_net {
 953	spinlock_t idr_lock; /* Protects idr */
 954	struct idr idr;
 955};
 956
 957static unsigned int tcf_net_id;
 958
 959static int tcf_block_insert(struct tcf_block *block, struct net *net,
 960			    struct netlink_ext_ack *extack)
 961{
 962	struct tcf_net *tn = net_generic(net, tcf_net_id);
 963	int err;
 964
 965	idr_preload(GFP_KERNEL);
 966	spin_lock(&tn->idr_lock);
 967	err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
 968			    GFP_NOWAIT);
 969	spin_unlock(&tn->idr_lock);
 970	idr_preload_end();
 971
 972	return err;
 973}
 974
 975static void tcf_block_remove(struct tcf_block *block, struct net *net)
 976{
 977	struct tcf_net *tn = net_generic(net, tcf_net_id);
 978
 979	spin_lock(&tn->idr_lock);
 980	idr_remove(&tn->idr, block->index);
 981	spin_unlock(&tn->idr_lock);
 982}
 983
 984static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
 985					  u32 block_index,
 986					  struct netlink_ext_ack *extack)
 987{
 988	struct tcf_block *block;
 989
 990	block = kzalloc(sizeof(*block), GFP_KERNEL);
 991	if (!block) {
 992		NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
 993		return ERR_PTR(-ENOMEM);
 994	}
 995	mutex_init(&block->lock);
 996	mutex_init(&block->proto_destroy_lock);
 997	init_rwsem(&block->cb_lock);
 998	flow_block_init(&block->flow_block);
 999	INIT_LIST_HEAD(&block->chain_list);
1000	INIT_LIST_HEAD(&block->owner_list);
1001	INIT_LIST_HEAD(&block->chain0.filter_chain_list);
1002
1003	refcount_set(&block->refcnt, 1);
1004	block->net = net;
1005	block->index = block_index;
1006	xa_init(&block->ports);
1007
1008	/* Don't store q pointer for blocks which are shared */
1009	if (!tcf_block_shared(block))
1010		block->q = q;
1011	return block;
1012}
1013
1014struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
1015{
1016	struct tcf_net *tn = net_generic(net, tcf_net_id);
1017
1018	return idr_find(&tn->idr, block_index);
1019}
1020EXPORT_SYMBOL(tcf_block_lookup);
1021
1022static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
1023{
1024	struct tcf_block *block;
1025
1026	rcu_read_lock();
1027	block = tcf_block_lookup(net, block_index);
1028	if (block && !refcount_inc_not_zero(&block->refcnt))
1029		block = NULL;
1030	rcu_read_unlock();
1031
1032	return block;
1033}
1034
1035static struct tcf_chain *
1036__tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1037{
1038	mutex_lock(&block->lock);
1039	if (chain)
1040		chain = list_is_last(&chain->list, &block->chain_list) ?
1041			NULL : list_next_entry(chain, list);
1042	else
1043		chain = list_first_entry_or_null(&block->chain_list,
1044						 struct tcf_chain, list);
1045
1046	/* skip all action-only chains */
1047	while (chain && tcf_chain_held_by_acts_only(chain))
1048		chain = list_is_last(&chain->list, &block->chain_list) ?
1049			NULL : list_next_entry(chain, list);
1050
1051	if (chain)
1052		tcf_chain_hold(chain);
1053	mutex_unlock(&block->lock);
1054
1055	return chain;
1056}
1057
1058/* Function to be used by all clients that want to iterate over all chains on
1059 * block. It properly obtains block->lock and takes reference to chain before
1060 * returning it. Users of this function must be tolerant to concurrent chain
1061 * insertion/deletion or ensure that no concurrent chain modification is
1062 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1063 * consistent dump because rtnl lock is released each time skb is filled with
1064 * data and sent to user-space.
1065 */
1066
1067struct tcf_chain *
1068tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1069{
1070	struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
1071
1072	if (chain)
1073		tcf_chain_put(chain);
1074
1075	return chain_next;
1076}
1077EXPORT_SYMBOL(tcf_get_next_chain);
1078
1079static struct tcf_proto *
1080__tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1081{
1082	u32 prio = 0;
1083
1084	ASSERT_RTNL();
1085	mutex_lock(&chain->filter_chain_lock);
1086
1087	if (!tp) {
1088		tp = tcf_chain_dereference(chain->filter_chain, chain);
1089	} else if (tcf_proto_is_deleting(tp)) {
1090		/* 'deleting' flag is set and chain->filter_chain_lock was
1091		 * unlocked, which means next pointer could be invalid. Restart
1092		 * search.
1093		 */
1094		prio = tp->prio + 1;
1095		tp = tcf_chain_dereference(chain->filter_chain, chain);
1096
1097		for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1098			if (!tp->deleting && tp->prio >= prio)
1099				break;
1100	} else {
1101		tp = tcf_chain_dereference(tp->next, chain);
1102	}
1103
1104	if (tp)
1105		tcf_proto_get(tp);
1106
1107	mutex_unlock(&chain->filter_chain_lock);
1108
1109	return tp;
1110}
1111
1112/* Function to be used by all clients that want to iterate over all tp's on
1113 * chain. Users of this function must be tolerant to concurrent tp
1114 * insertion/deletion or ensure that no concurrent chain modification is
1115 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1116 * consistent dump because rtnl lock is released each time skb is filled with
1117 * data and sent to user-space.
1118 */
1119
1120struct tcf_proto *
1121tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1122{
1123	struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1124
1125	if (tp)
1126		tcf_proto_put(tp, true, NULL);
1127
1128	return tp_next;
1129}
1130EXPORT_SYMBOL(tcf_get_next_proto);
1131
1132static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1133{
1134	struct tcf_chain *chain;
1135
1136	/* Last reference to block. At this point chains cannot be added or
1137	 * removed concurrently.
1138	 */
1139	for (chain = tcf_get_next_chain(block, NULL);
1140	     chain;
1141	     chain = tcf_get_next_chain(block, chain)) {
1142		tcf_chain_put_explicitly_created(chain);
1143		tcf_chain_flush(chain, rtnl_held);
1144	}
1145}
1146
1147/* Lookup Qdisc and increments its reference counter.
1148 * Set parent, if necessary.
1149 */
1150
1151static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1152			    u32 *parent, int ifindex, bool rtnl_held,
1153			    struct netlink_ext_ack *extack)
1154{
1155	const struct Qdisc_class_ops *cops;
1156	struct net_device *dev;
1157	int err = 0;
1158
1159	if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1160		return 0;
1161
1162	rcu_read_lock();
1163
1164	/* Find link */
1165	dev = dev_get_by_index_rcu(net, ifindex);
1166	if (!dev) {
1167		rcu_read_unlock();
1168		return -ENODEV;
1169	}
1170
1171	/* Find qdisc */
1172	if (!*parent) {
1173		*q = rcu_dereference(dev->qdisc);
1174		*parent = (*q)->handle;
1175	} else {
1176		*q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1177		if (!*q) {
1178			NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1179			err = -EINVAL;
1180			goto errout_rcu;
1181		}
1182	}
1183
1184	*q = qdisc_refcount_inc_nz(*q);
1185	if (!*q) {
1186		NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1187		err = -EINVAL;
1188		goto errout_rcu;
1189	}
1190
1191	/* Is it classful? */
1192	cops = (*q)->ops->cl_ops;
1193	if (!cops) {
1194		NL_SET_ERR_MSG(extack, "Qdisc not classful");
1195		err = -EINVAL;
1196		goto errout_qdisc;
1197	}
1198
1199	if (!cops->tcf_block) {
1200		NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1201		err = -EOPNOTSUPP;
1202		goto errout_qdisc;
1203	}
1204
1205errout_rcu:
1206	/* At this point we know that qdisc is not noop_qdisc,
1207	 * which means that qdisc holds a reference to net_device
1208	 * and we hold a reference to qdisc, so it is safe to release
1209	 * rcu read lock.
1210	 */
1211	rcu_read_unlock();
1212	return err;
1213
1214errout_qdisc:
1215	rcu_read_unlock();
1216
1217	if (rtnl_held)
1218		qdisc_put(*q);
1219	else
1220		qdisc_put_unlocked(*q);
1221	*q = NULL;
1222
1223	return err;
1224}
1225
1226static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1227			       int ifindex, struct netlink_ext_ack *extack)
1228{
1229	if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1230		return 0;
1231
1232	/* Do we search for filter, attached to class? */
1233	if (TC_H_MIN(parent)) {
1234		const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1235
1236		*cl = cops->find(q, parent);
1237		if (*cl == 0) {
1238			NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1239			return -ENOENT;
1240		}
1241	}
1242
1243	return 0;
1244}
1245
1246static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1247					  unsigned long cl, int ifindex,
1248					  u32 block_index,
1249					  struct netlink_ext_ack *extack)
1250{
1251	struct tcf_block *block;
1252
1253	if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1254		block = tcf_block_refcnt_get(net, block_index);
1255		if (!block) {
1256			NL_SET_ERR_MSG(extack, "Block of given index was not found");
1257			return ERR_PTR(-EINVAL);
1258		}
1259	} else {
1260		const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1261
1262		block = cops->tcf_block(q, cl, extack);
1263		if (!block)
1264			return ERR_PTR(-EINVAL);
1265
1266		if (tcf_block_shared(block)) {
1267			NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1268			return ERR_PTR(-EOPNOTSUPP);
1269		}
1270
1271		/* Always take reference to block in order to support execution
1272		 * of rules update path of cls API without rtnl lock. Caller
1273		 * must release block when it is finished using it. 'if' block
1274		 * of this conditional obtain reference to block by calling
1275		 * tcf_block_refcnt_get().
1276		 */
1277		refcount_inc(&block->refcnt);
1278	}
1279
1280	return block;
1281}
1282
1283static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1284			    struct tcf_block_ext_info *ei, bool rtnl_held)
1285{
1286	if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1287		/* Flushing/putting all chains will cause the block to be
1288		 * deallocated when last chain is freed. However, if chain_list
1289		 * is empty, block has to be manually deallocated. After block
1290		 * reference counter reached 0, it is no longer possible to
1291		 * increment it or add new chains to block.
1292		 */
1293		bool free_block = list_empty(&block->chain_list);
1294
1295		mutex_unlock(&block->lock);
1296		if (tcf_block_shared(block))
1297			tcf_block_remove(block, block->net);
1298
1299		if (q)
1300			tcf_block_offload_unbind(block, q, ei);
1301
1302		if (free_block)
1303			tcf_block_destroy(block);
1304		else
1305			tcf_block_flush_all_chains(block, rtnl_held);
1306	} else if (q) {
1307		tcf_block_offload_unbind(block, q, ei);
1308	}
1309}
1310
1311static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1312{
1313	__tcf_block_put(block, NULL, NULL, rtnl_held);
1314}
1315
1316/* Find tcf block.
1317 * Set q, parent, cl when appropriate.
1318 */
1319
1320static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1321					u32 *parent, unsigned long *cl,
1322					int ifindex, u32 block_index,
1323					struct netlink_ext_ack *extack)
1324{
1325	struct tcf_block *block;
1326	int err = 0;
1327
1328	ASSERT_RTNL();
1329
1330	err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1331	if (err)
1332		goto errout;
1333
1334	err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1335	if (err)
1336		goto errout_qdisc;
1337
1338	block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1339	if (IS_ERR(block)) {
1340		err = PTR_ERR(block);
1341		goto errout_qdisc;
1342	}
1343
1344	return block;
1345
1346errout_qdisc:
1347	if (*q)
1348		qdisc_put(*q);
1349errout:
1350	*q = NULL;
1351	return ERR_PTR(err);
1352}
1353
1354static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1355			      bool rtnl_held)
1356{
1357	if (!IS_ERR_OR_NULL(block))
1358		tcf_block_refcnt_put(block, rtnl_held);
1359
1360	if (q) {
1361		if (rtnl_held)
1362			qdisc_put(q);
1363		else
1364			qdisc_put_unlocked(q);
1365	}
1366}
1367
1368struct tcf_block_owner_item {
1369	struct list_head list;
1370	struct Qdisc *q;
1371	enum flow_block_binder_type binder_type;
1372};
1373
1374static void
1375tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1376			       struct Qdisc *q,
1377			       enum flow_block_binder_type binder_type)
1378{
1379	if (block->keep_dst &&
1380	    binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1381	    binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1382		netif_keep_dst(qdisc_dev(q));
1383}
1384
1385void tcf_block_netif_keep_dst(struct tcf_block *block)
1386{
1387	struct tcf_block_owner_item *item;
1388
1389	block->keep_dst = true;
1390	list_for_each_entry(item, &block->owner_list, list)
1391		tcf_block_owner_netif_keep_dst(block, item->q,
1392					       item->binder_type);
1393}
1394EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1395
1396static int tcf_block_owner_add(struct tcf_block *block,
1397			       struct Qdisc *q,
1398			       enum flow_block_binder_type binder_type)
1399{
1400	struct tcf_block_owner_item *item;
1401
1402	item = kmalloc(sizeof(*item), GFP_KERNEL);
1403	if (!item)
1404		return -ENOMEM;
1405	item->q = q;
1406	item->binder_type = binder_type;
1407	list_add(&item->list, &block->owner_list);
1408	return 0;
1409}
1410
1411static void tcf_block_owner_del(struct tcf_block *block,
1412				struct Qdisc *q,
1413				enum flow_block_binder_type binder_type)
1414{
1415	struct tcf_block_owner_item *item;
1416
1417	list_for_each_entry(item, &block->owner_list, list) {
1418		if (item->q == q && item->binder_type == binder_type) {
1419			list_del(&item->list);
1420			kfree(item);
1421			return;
1422		}
1423	}
1424	WARN_ON(1);
1425}
1426
1427static bool tcf_block_tracks_dev(struct tcf_block *block,
1428				 struct tcf_block_ext_info *ei)
1429{
1430	return tcf_block_shared(block) &&
1431	       (ei->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS ||
1432		ei->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS);
1433}
1434
1435int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1436		      struct tcf_block_ext_info *ei,
1437		      struct netlink_ext_ack *extack)
1438{
1439	struct net_device *dev = qdisc_dev(q);
1440	struct net *net = qdisc_net(q);
1441	struct tcf_block *block = NULL;
1442	int err;
1443
1444	if (ei->block_index)
1445		/* block_index not 0 means the shared block is requested */
1446		block = tcf_block_refcnt_get(net, ei->block_index);
1447
1448	if (!block) {
1449		block = tcf_block_create(net, q, ei->block_index, extack);
1450		if (IS_ERR(block))
1451			return PTR_ERR(block);
1452		if (tcf_block_shared(block)) {
1453			err = tcf_block_insert(block, net, extack);
1454			if (err)
1455				goto err_block_insert;
1456		}
1457	}
1458
1459	err = tcf_block_owner_add(block, q, ei->binder_type);
1460	if (err)
1461		goto err_block_owner_add;
1462
1463	tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1464
1465	err = tcf_chain0_head_change_cb_add(block, ei, extack);
1466	if (err)
1467		goto err_chain0_head_change_cb_add;
1468
1469	err = tcf_block_offload_bind(block, q, ei, extack);
1470	if (err)
1471		goto err_block_offload_bind;
1472
1473	if (tcf_block_tracks_dev(block, ei)) {
1474		err = xa_insert(&block->ports, dev->ifindex, dev, GFP_KERNEL);
1475		if (err) {
1476			NL_SET_ERR_MSG(extack, "block dev insert failed");
1477			goto err_dev_insert;
1478		}
1479	}
1480
1481	*p_block = block;
1482	return 0;
1483
1484err_dev_insert:
 
1485err_block_offload_bind:
1486	tcf_chain0_head_change_cb_del(block, ei);
1487err_chain0_head_change_cb_add:
1488	tcf_block_owner_del(block, q, ei->binder_type);
1489err_block_owner_add:
1490err_block_insert:
1491	tcf_block_refcnt_put(block, true);
1492	return err;
1493}
1494EXPORT_SYMBOL(tcf_block_get_ext);
1495
1496static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1497{
1498	struct tcf_proto __rcu **p_filter_chain = priv;
1499
1500	rcu_assign_pointer(*p_filter_chain, tp_head);
1501}
1502
1503int tcf_block_get(struct tcf_block **p_block,
1504		  struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1505		  struct netlink_ext_ack *extack)
1506{
1507	struct tcf_block_ext_info ei = {
1508		.chain_head_change = tcf_chain_head_change_dflt,
1509		.chain_head_change_priv = p_filter_chain,
1510	};
1511
1512	WARN_ON(!p_filter_chain);
1513	return tcf_block_get_ext(p_block, q, &ei, extack);
1514}
1515EXPORT_SYMBOL(tcf_block_get);
1516
1517/* XXX: Standalone actions are not allowed to jump to any chain, and bound
1518 * actions should be all removed after flushing.
1519 */
1520void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1521		       struct tcf_block_ext_info *ei)
1522{
1523	struct net_device *dev = qdisc_dev(q);
1524
1525	if (!block)
1526		return;
1527	if (tcf_block_tracks_dev(block, ei))
1528		xa_erase(&block->ports, dev->ifindex);
1529	tcf_chain0_head_change_cb_del(block, ei);
1530	tcf_block_owner_del(block, q, ei->binder_type);
1531
1532	__tcf_block_put(block, q, ei, true);
1533}
1534EXPORT_SYMBOL(tcf_block_put_ext);
1535
1536void tcf_block_put(struct tcf_block *block)
1537{
1538	struct tcf_block_ext_info ei = {0, };
1539
1540	if (!block)
1541		return;
1542	tcf_block_put_ext(block, block->q, &ei);
1543}
1544
1545EXPORT_SYMBOL(tcf_block_put);
1546
1547static int
1548tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1549			    void *cb_priv, bool add, bool offload_in_use,
1550			    struct netlink_ext_ack *extack)
1551{
1552	struct tcf_chain *chain, *chain_prev;
1553	struct tcf_proto *tp, *tp_prev;
1554	int err;
1555
1556	lockdep_assert_held(&block->cb_lock);
1557
1558	for (chain = __tcf_get_next_chain(block, NULL);
1559	     chain;
1560	     chain_prev = chain,
1561		     chain = __tcf_get_next_chain(block, chain),
1562		     tcf_chain_put(chain_prev)) {
1563		if (chain->tmplt_ops && add)
1564			chain->tmplt_ops->tmplt_reoffload(chain, true, cb,
1565							  cb_priv);
1566		for (tp = __tcf_get_next_proto(chain, NULL); tp;
1567		     tp_prev = tp,
1568			     tp = __tcf_get_next_proto(chain, tp),
1569			     tcf_proto_put(tp_prev, true, NULL)) {
1570			if (tp->ops->reoffload) {
1571				err = tp->ops->reoffload(tp, add, cb, cb_priv,
1572							 extack);
1573				if (err && add)
1574					goto err_playback_remove;
1575			} else if (add && offload_in_use) {
1576				err = -EOPNOTSUPP;
1577				NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1578				goto err_playback_remove;
1579			}
1580		}
1581		if (chain->tmplt_ops && !add)
1582			chain->tmplt_ops->tmplt_reoffload(chain, false, cb,
1583							  cb_priv);
1584	}
1585
1586	return 0;
1587
1588err_playback_remove:
1589	tcf_proto_put(tp, true, NULL);
1590	tcf_chain_put(chain);
1591	tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1592				    extack);
1593	return err;
1594}
1595
1596static int tcf_block_bind(struct tcf_block *block,
1597			  struct flow_block_offload *bo)
1598{
1599	struct flow_block_cb *block_cb, *next;
1600	int err, i = 0;
1601
1602	lockdep_assert_held(&block->cb_lock);
1603
1604	list_for_each_entry(block_cb, &bo->cb_list, list) {
1605		err = tcf_block_playback_offloads(block, block_cb->cb,
1606						  block_cb->cb_priv, true,
1607						  tcf_block_offload_in_use(block),
1608						  bo->extack);
1609		if (err)
1610			goto err_unroll;
1611		if (!bo->unlocked_driver_cb)
1612			block->lockeddevcnt++;
1613
1614		i++;
1615	}
1616	list_splice(&bo->cb_list, &block->flow_block.cb_list);
1617
1618	return 0;
1619
1620err_unroll:
1621	list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1622		list_del(&block_cb->driver_list);
1623		if (i-- > 0) {
1624			list_del(&block_cb->list);
1625			tcf_block_playback_offloads(block, block_cb->cb,
1626						    block_cb->cb_priv, false,
1627						    tcf_block_offload_in_use(block),
1628						    NULL);
1629			if (!bo->unlocked_driver_cb)
1630				block->lockeddevcnt--;
1631		}
1632		flow_block_cb_free(block_cb);
1633	}
1634
1635	return err;
1636}
1637
1638static void tcf_block_unbind(struct tcf_block *block,
1639			     struct flow_block_offload *bo)
1640{
1641	struct flow_block_cb *block_cb, *next;
1642
1643	lockdep_assert_held(&block->cb_lock);
1644
1645	list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1646		tcf_block_playback_offloads(block, block_cb->cb,
1647					    block_cb->cb_priv, false,
1648					    tcf_block_offload_in_use(block),
1649					    NULL);
1650		list_del(&block_cb->list);
1651		flow_block_cb_free(block_cb);
1652		if (!bo->unlocked_driver_cb)
1653			block->lockeddevcnt--;
1654	}
1655}
1656
1657static int tcf_block_setup(struct tcf_block *block,
1658			   struct flow_block_offload *bo)
1659{
1660	int err;
1661
1662	switch (bo->command) {
1663	case FLOW_BLOCK_BIND:
1664		err = tcf_block_bind(block, bo);
1665		break;
1666	case FLOW_BLOCK_UNBIND:
1667		err = 0;
1668		tcf_block_unbind(block, bo);
1669		break;
1670	default:
1671		WARN_ON_ONCE(1);
1672		err = -EOPNOTSUPP;
1673	}
1674
1675	return err;
1676}
1677
1678/* Main classifier routine: scans classifier chain attached
1679 * to this qdisc, (optionally) tests for protocol and asks
1680 * specific classifiers.
1681 */
1682static inline int __tcf_classify(struct sk_buff *skb,
1683				 const struct tcf_proto *tp,
1684				 const struct tcf_proto *orig_tp,
1685				 struct tcf_result *res,
1686				 bool compat_mode,
1687				 struct tcf_exts_miss_cookie_node *n,
1688				 int act_index,
1689				 u32 *last_executed_chain)
1690{
1691#ifdef CONFIG_NET_CLS_ACT
1692	const int max_reclassify_loop = 16;
1693	const struct tcf_proto *first_tp;
1694	int limit = 0;
1695
1696reclassify:
1697#endif
1698	for (; tp; tp = rcu_dereference_bh(tp->next)) {
1699		__be16 protocol = skb_protocol(skb, false);
1700		int err = 0;
1701
1702		if (n) {
1703			struct tcf_exts *exts;
1704
1705			if (n->tp_prio != tp->prio)
1706				continue;
1707
1708			/* We re-lookup the tp and chain based on index instead
1709			 * of having hard refs and locks to them, so do a sanity
1710			 * check if any of tp,chain,exts was replaced by the
1711			 * time we got here with a cookie from hardware.
1712			 */
1713			if (unlikely(n->tp != tp || n->tp->chain != n->chain ||
1714				     !tp->ops->get_exts)) {
1715				tcf_set_drop_reason(skb,
1716						    SKB_DROP_REASON_TC_COOKIE_ERROR);
1717				return TC_ACT_SHOT;
1718			}
1719
1720			exts = tp->ops->get_exts(tp, n->handle);
1721			if (unlikely(!exts || n->exts != exts)) {
1722				tcf_set_drop_reason(skb,
1723						    SKB_DROP_REASON_TC_COOKIE_ERROR);
1724				return TC_ACT_SHOT;
1725			}
1726
1727			n = NULL;
1728			err = tcf_exts_exec_ex(skb, exts, act_index, res);
1729		} else {
1730			if (tp->protocol != protocol &&
1731			    tp->protocol != htons(ETH_P_ALL))
1732				continue;
1733
1734			err = tc_classify(skb, tp, res);
1735		}
1736#ifdef CONFIG_NET_CLS_ACT
1737		if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1738			first_tp = orig_tp;
1739			*last_executed_chain = first_tp->chain->index;
1740			goto reset;
1741		} else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1742			first_tp = res->goto_tp;
1743			*last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1744			goto reset;
1745		}
1746#endif
1747		if (err >= 0)
1748			return err;
1749	}
1750
1751	if (unlikely(n)) {
1752		tcf_set_drop_reason(skb,
1753				    SKB_DROP_REASON_TC_COOKIE_ERROR);
1754		return TC_ACT_SHOT;
1755	}
1756
1757	return TC_ACT_UNSPEC; /* signal: continue lookup */
1758#ifdef CONFIG_NET_CLS_ACT
1759reset:
1760	if (unlikely(limit++ >= max_reclassify_loop)) {
1761		net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1762				       tp->chain->block->index,
1763				       tp->prio & 0xffff,
1764				       ntohs(tp->protocol));
1765		tcf_set_drop_reason(skb,
1766				    SKB_DROP_REASON_TC_RECLASSIFY_LOOP);
1767		return TC_ACT_SHOT;
1768	}
1769
1770	tp = first_tp;
1771	goto reclassify;
1772#endif
1773}
1774
1775int tcf_classify(struct sk_buff *skb,
1776		 const struct tcf_block *block,
1777		 const struct tcf_proto *tp,
1778		 struct tcf_result *res, bool compat_mode)
1779{
1780#if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1781	u32 last_executed_chain = 0;
1782
1783	return __tcf_classify(skb, tp, tp, res, compat_mode, NULL, 0,
1784			      &last_executed_chain);
1785#else
1786	u32 last_executed_chain = tp ? tp->chain->index : 0;
1787	struct tcf_exts_miss_cookie_node *n = NULL;
1788	const struct tcf_proto *orig_tp = tp;
1789	struct tc_skb_ext *ext;
1790	int act_index = 0;
1791	int ret;
1792
1793	if (block) {
1794		ext = skb_ext_find(skb, TC_SKB_EXT);
1795
1796		if (ext && (ext->chain || ext->act_miss)) {
1797			struct tcf_chain *fchain;
1798			u32 chain;
1799
1800			if (ext->act_miss) {
1801				n = tcf_exts_miss_cookie_lookup(ext->act_miss_cookie,
1802								&act_index);
1803				if (!n) {
1804					tcf_set_drop_reason(skb,
1805							    SKB_DROP_REASON_TC_COOKIE_ERROR);
1806					return TC_ACT_SHOT;
1807				}
1808
1809				chain = n->chain_index;
1810			} else {
1811				chain = ext->chain;
1812			}
1813
1814			fchain = tcf_chain_lookup_rcu(block, chain);
1815			if (!fchain) {
1816				tcf_set_drop_reason(skb,
1817						    SKB_DROP_REASON_TC_CHAIN_NOTFOUND);
1818
1819				return TC_ACT_SHOT;
1820			}
1821
1822			/* Consume, so cloned/redirect skbs won't inherit ext */
1823			skb_ext_del(skb, TC_SKB_EXT);
1824
1825			tp = rcu_dereference_bh(fchain->filter_chain);
1826			last_executed_chain = fchain->index;
1827		}
1828	}
1829
1830	ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode, n, act_index,
1831			     &last_executed_chain);
1832
1833	if (tc_skb_ext_tc_enabled()) {
1834		/* If we missed on some chain */
1835		if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1836			struct tc_skb_cb *cb = tc_skb_cb(skb);
1837
1838			ext = tc_skb_ext_alloc(skb);
1839			if (WARN_ON_ONCE(!ext)) {
1840				tcf_set_drop_reason(skb, SKB_DROP_REASON_NOMEM);
1841				return TC_ACT_SHOT;
1842			}
1843			ext->chain = last_executed_chain;
1844			ext->mru = cb->mru;
1845			ext->post_ct = cb->post_ct;
1846			ext->post_ct_snat = cb->post_ct_snat;
1847			ext->post_ct_dnat = cb->post_ct_dnat;
1848			ext->zone = cb->zone;
1849		}
1850	}
1851
1852	return ret;
1853#endif
1854}
1855EXPORT_SYMBOL(tcf_classify);
1856
1857struct tcf_chain_info {
1858	struct tcf_proto __rcu **pprev;
1859	struct tcf_proto __rcu *next;
1860};
1861
1862static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1863					   struct tcf_chain_info *chain_info)
1864{
1865	return tcf_chain_dereference(*chain_info->pprev, chain);
1866}
1867
1868static int tcf_chain_tp_insert(struct tcf_chain *chain,
1869			       struct tcf_chain_info *chain_info,
1870			       struct tcf_proto *tp)
1871{
1872	if (chain->flushing)
1873		return -EAGAIN;
1874
1875	RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1876	if (*chain_info->pprev == chain->filter_chain)
1877		tcf_chain0_head_change(chain, tp);
1878	tcf_proto_get(tp);
1879	rcu_assign_pointer(*chain_info->pprev, tp);
1880
1881	return 0;
1882}
1883
1884static void tcf_chain_tp_remove(struct tcf_chain *chain,
1885				struct tcf_chain_info *chain_info,
1886				struct tcf_proto *tp)
1887{
1888	struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1889
1890	tcf_proto_mark_delete(tp);
1891	if (tp == chain->filter_chain)
1892		tcf_chain0_head_change(chain, next);
1893	RCU_INIT_POINTER(*chain_info->pprev, next);
1894}
1895
1896static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1897					   struct tcf_chain_info *chain_info,
1898					   u32 protocol, u32 prio,
1899					   bool prio_allocate);
 
1900
1901/* Try to insert new proto.
1902 * If proto with specified priority already exists, free new proto
1903 * and return existing one.
1904 */
1905
1906static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1907						    struct tcf_proto *tp_new,
1908						    u32 protocol, u32 prio,
1909						    bool rtnl_held)
1910{
1911	struct tcf_chain_info chain_info;
1912	struct tcf_proto *tp;
1913	int err = 0;
1914
1915	mutex_lock(&chain->filter_chain_lock);
1916
1917	if (tcf_proto_exists_destroying(chain, tp_new)) {
1918		mutex_unlock(&chain->filter_chain_lock);
1919		tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1920		return ERR_PTR(-EAGAIN);
1921	}
1922
1923	tp = tcf_chain_tp_find(chain, &chain_info,
1924			       protocol, prio, false);
1925	if (!tp)
1926		err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1927	mutex_unlock(&chain->filter_chain_lock);
1928
1929	if (tp) {
1930		tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1931		tp_new = tp;
1932	} else if (err) {
1933		tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1934		tp_new = ERR_PTR(err);
1935	}
1936
1937	return tp_new;
1938}
1939
1940static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1941				      struct tcf_proto *tp, bool rtnl_held,
1942				      struct netlink_ext_ack *extack)
1943{
1944	struct tcf_chain_info chain_info;
1945	struct tcf_proto *tp_iter;
1946	struct tcf_proto **pprev;
1947	struct tcf_proto *next;
1948
1949	mutex_lock(&chain->filter_chain_lock);
1950
1951	/* Atomically find and remove tp from chain. */
1952	for (pprev = &chain->filter_chain;
1953	     (tp_iter = tcf_chain_dereference(*pprev, chain));
1954	     pprev = &tp_iter->next) {
1955		if (tp_iter == tp) {
1956			chain_info.pprev = pprev;
1957			chain_info.next = tp_iter->next;
1958			WARN_ON(tp_iter->deleting);
1959			break;
1960		}
1961	}
1962	/* Verify that tp still exists and no new filters were inserted
1963	 * concurrently.
1964	 * Mark tp for deletion if it is empty.
1965	 */
1966	if (!tp_iter || !tcf_proto_check_delete(tp)) {
1967		mutex_unlock(&chain->filter_chain_lock);
1968		return;
1969	}
1970
1971	tcf_proto_signal_destroying(chain, tp);
1972	next = tcf_chain_dereference(chain_info.next, chain);
1973	if (tp == chain->filter_chain)
1974		tcf_chain0_head_change(chain, next);
1975	RCU_INIT_POINTER(*chain_info.pprev, next);
1976	mutex_unlock(&chain->filter_chain_lock);
1977
1978	tcf_proto_put(tp, rtnl_held, extack);
1979}
1980
1981static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1982					   struct tcf_chain_info *chain_info,
1983					   u32 protocol, u32 prio,
1984					   bool prio_allocate)
 
1985{
1986	struct tcf_proto **pprev;
1987	struct tcf_proto *tp;
1988
1989	/* Check the chain for existence of proto-tcf with this priority */
1990	for (pprev = &chain->filter_chain;
1991	     (tp = tcf_chain_dereference(*pprev, chain));
1992	     pprev = &tp->next) {
1993		if (tp->prio >= prio) {
1994			if (tp->prio == prio) {
1995				if (prio_allocate ||
1996				    (tp->protocol != protocol && protocol))
 
 
 
 
1997					return ERR_PTR(-EINVAL);
 
1998			} else {
1999				tp = NULL;
2000			}
2001			break;
2002		}
2003	}
2004	chain_info->pprev = pprev;
2005	if (tp) {
2006		chain_info->next = tp->next;
2007		tcf_proto_get(tp);
2008	} else {
2009		chain_info->next = NULL;
2010	}
2011	return tp;
2012}
2013
2014static int tcf_fill_node(struct net *net, struct sk_buff *skb,
2015			 struct tcf_proto *tp, struct tcf_block *block,
2016			 struct Qdisc *q, u32 parent, void *fh,
2017			 u32 portid, u32 seq, u16 flags, int event,
2018			 bool terse_dump, bool rtnl_held,
2019			 struct netlink_ext_ack *extack)
2020{
2021	struct tcmsg *tcm;
2022	struct nlmsghdr  *nlh;
2023	unsigned char *b = skb_tail_pointer(skb);
2024
2025	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2026	if (!nlh)
2027		goto out_nlmsg_trim;
2028	tcm = nlmsg_data(nlh);
2029	tcm->tcm_family = AF_UNSPEC;
2030	tcm->tcm__pad1 = 0;
2031	tcm->tcm__pad2 = 0;
2032	if (q) {
2033		tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
2034		tcm->tcm_parent = parent;
2035	} else {
2036		tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2037		tcm->tcm_block_index = block->index;
2038	}
2039	tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
2040	if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
2041		goto nla_put_failure;
2042	if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
2043		goto nla_put_failure;
2044	if (!fh) {
2045		tcm->tcm_handle = 0;
2046	} else if (terse_dump) {
2047		if (tp->ops->terse_dump) {
2048			if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
2049						rtnl_held) < 0)
2050				goto nla_put_failure;
2051		} else {
2052			goto cls_op_not_supp;
2053		}
2054	} else {
2055		if (tp->ops->dump &&
2056		    tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
2057			goto nla_put_failure;
2058	}
2059
2060	if (extack && extack->_msg &&
2061	    nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2062		goto nla_put_failure;
2063
2064	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2065
2066	return skb->len;
2067
2068out_nlmsg_trim:
2069nla_put_failure:
2070cls_op_not_supp:
2071	nlmsg_trim(skb, b);
2072	return -1;
2073}
2074
2075static int tfilter_notify(struct net *net, struct sk_buff *oskb,
2076			  struct nlmsghdr *n, struct tcf_proto *tp,
2077			  struct tcf_block *block, struct Qdisc *q,
2078			  u32 parent, void *fh, int event, bool unicast,
2079			  bool rtnl_held, struct netlink_ext_ack *extack)
2080{
2081	struct sk_buff *skb;
2082	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2083	int err = 0;
2084
2085	if (!unicast && !rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2086		return 0;
2087
2088	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2089	if (!skb)
2090		return -ENOBUFS;
2091
2092	if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2093			  n->nlmsg_seq, n->nlmsg_flags, event,
2094			  false, rtnl_held, extack) <= 0) {
2095		kfree_skb(skb);
2096		return -EINVAL;
2097	}
2098
2099	if (unicast)
2100		err = rtnl_unicast(skb, net, portid);
2101	else
2102		err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2103				     n->nlmsg_flags & NLM_F_ECHO);
2104	return err;
2105}
2106
2107static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
2108			      struct nlmsghdr *n, struct tcf_proto *tp,
2109			      struct tcf_block *block, struct Qdisc *q,
2110			      u32 parent, void *fh, bool *last, bool rtnl_held,
2111			      struct netlink_ext_ack *extack)
2112{
2113	struct sk_buff *skb;
2114	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2115	int err;
2116
2117	if (!rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2118		return tp->ops->delete(tp, fh, last, rtnl_held, extack);
2119
2120	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2121	if (!skb)
2122		return -ENOBUFS;
2123
2124	if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2125			  n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
2126			  false, rtnl_held, extack) <= 0) {
2127		NL_SET_ERR_MSG(extack, "Failed to build del event notification");
2128		kfree_skb(skb);
2129		return -EINVAL;
2130	}
2131
2132	err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
2133	if (err) {
2134		kfree_skb(skb);
2135		return err;
2136	}
2137
2138	err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2139			     n->nlmsg_flags & NLM_F_ECHO);
2140	if (err < 0)
2141		NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
2142
2143	return err;
2144}
2145
2146static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
2147				 struct tcf_block *block, struct Qdisc *q,
2148				 u32 parent, struct nlmsghdr *n,
2149				 struct tcf_chain *chain, int event,
2150				 struct netlink_ext_ack *extack)
2151{
2152	struct tcf_proto *tp;
2153
2154	for (tp = tcf_get_next_proto(chain, NULL);
2155	     tp; tp = tcf_get_next_proto(chain, tp))
2156		tfilter_notify(net, oskb, n, tp, block, q, parent, NULL,
2157			       event, false, true, extack);
2158}
2159
2160static void tfilter_put(struct tcf_proto *tp, void *fh)
2161{
2162	if (tp->ops->put && fh)
2163		tp->ops->put(tp, fh);
2164}
2165
2166static bool is_qdisc_ingress(__u32 classid)
2167{
2168	return (TC_H_MIN(classid) == TC_H_MIN(TC_H_MIN_INGRESS));
2169}
2170
2171static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2172			  struct netlink_ext_ack *extack)
2173{
2174	struct net *net = sock_net(skb->sk);
2175	struct nlattr *tca[TCA_MAX + 1];
2176	char name[IFNAMSIZ];
2177	struct tcmsg *t;
2178	u32 protocol;
2179	u32 prio;
2180	bool prio_allocate;
2181	u32 parent;
2182	u32 chain_index;
2183	struct Qdisc *q;
2184	struct tcf_chain_info chain_info;
2185	struct tcf_chain *chain;
2186	struct tcf_block *block;
2187	struct tcf_proto *tp;
2188	unsigned long cl;
2189	void *fh;
2190	int err;
2191	int tp_created;
2192	bool rtnl_held = false;
2193	u32 flags;
2194
2195replay:
2196	tp_created = 0;
2197
2198	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2199				     rtm_tca_policy, extack);
2200	if (err < 0)
2201		return err;
2202
2203	t = nlmsg_data(n);
2204	protocol = TC_H_MIN(t->tcm_info);
2205	prio = TC_H_MAJ(t->tcm_info);
2206	prio_allocate = false;
2207	parent = t->tcm_parent;
2208	tp = NULL;
2209	cl = 0;
2210	block = NULL;
2211	q = NULL;
2212	chain = NULL;
2213	flags = 0;
2214
2215	if (prio == 0) {
2216		/* If no priority is provided by the user,
2217		 * we allocate one.
2218		 */
2219		if (n->nlmsg_flags & NLM_F_CREATE) {
2220			prio = TC_H_MAKE(0x80000000U, 0U);
2221			prio_allocate = true;
2222		} else {
2223			NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2224			return -ENOENT;
2225		}
2226	}
2227
2228	/* Find head of filter chain. */
2229
2230	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2231	if (err)
2232		return err;
2233
2234	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2235		NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2236		err = -EINVAL;
2237		goto errout;
2238	}
2239
2240	/* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2241	 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2242	 * type is not specified, classifier is not unlocked.
2243	 */
2244	if (rtnl_held ||
2245	    (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2246	    !tcf_proto_is_unlocked(name)) {
2247		rtnl_held = true;
2248		rtnl_lock();
2249	}
2250
2251	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2252	if (err)
2253		goto errout;
2254
2255	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2256				 extack);
2257	if (IS_ERR(block)) {
2258		err = PTR_ERR(block);
2259		goto errout;
2260	}
2261	block->classid = parent;
2262
2263	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2264	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2265		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2266		err = -EINVAL;
2267		goto errout;
2268	}
2269	chain = tcf_chain_get(block, chain_index, true);
2270	if (!chain) {
2271		NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2272		err = -ENOMEM;
2273		goto errout;
2274	}
2275
2276	mutex_lock(&chain->filter_chain_lock);
2277	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2278			       prio, prio_allocate);
2279	if (IS_ERR(tp)) {
2280		NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2281		err = PTR_ERR(tp);
2282		goto errout_locked;
2283	}
2284
2285	if (tp == NULL) {
2286		struct tcf_proto *tp_new = NULL;
2287
2288		if (chain->flushing) {
2289			err = -EAGAIN;
2290			goto errout_locked;
2291		}
2292
2293		/* Proto-tcf does not exist, create new one */
2294
2295		if (tca[TCA_KIND] == NULL || !protocol) {
2296			NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2297			err = -EINVAL;
2298			goto errout_locked;
2299		}
2300
2301		if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2302			NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2303			err = -ENOENT;
2304			goto errout_locked;
2305		}
2306
2307		if (prio_allocate)
2308			prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2309							       &chain_info));
2310
2311		mutex_unlock(&chain->filter_chain_lock);
2312		tp_new = tcf_proto_create(name, protocol, prio, chain,
2313					  rtnl_held, extack);
2314		if (IS_ERR(tp_new)) {
2315			err = PTR_ERR(tp_new);
2316			goto errout_tp;
2317		}
2318
2319		tp_created = 1;
2320		tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2321						rtnl_held);
2322		if (IS_ERR(tp)) {
2323			err = PTR_ERR(tp);
2324			goto errout_tp;
2325		}
2326	} else {
2327		mutex_unlock(&chain->filter_chain_lock);
2328	}
2329
2330	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2331		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2332		err = -EINVAL;
2333		goto errout;
2334	}
2335
2336	fh = tp->ops->get(tp, t->tcm_handle);
2337
2338	if (!fh) {
2339		if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2340			NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2341			err = -ENOENT;
2342			goto errout;
2343		}
2344	} else if (n->nlmsg_flags & NLM_F_EXCL) {
2345		tfilter_put(tp, fh);
2346		NL_SET_ERR_MSG(extack, "Filter already exists");
2347		err = -EEXIST;
2348		goto errout;
2349	}
2350
2351	if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2352		tfilter_put(tp, fh);
2353		NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2354		err = -EINVAL;
2355		goto errout;
2356	}
2357
2358	if (!(n->nlmsg_flags & NLM_F_CREATE))
2359		flags |= TCA_ACT_FLAGS_REPLACE;
2360	if (!rtnl_held)
2361		flags |= TCA_ACT_FLAGS_NO_RTNL;
2362	if (is_qdisc_ingress(parent))
2363		flags |= TCA_ACT_FLAGS_AT_INGRESS;
2364	err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2365			      flags, extack);
2366	if (err == 0) {
2367		tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2368			       RTM_NEWTFILTER, false, rtnl_held, extack);
2369		tfilter_put(tp, fh);
 
2370		/* q pointer is NULL for shared blocks */
2371		if (q)
2372			q->flags &= ~TCQ_F_CAN_BYPASS;
2373	}
2374
2375errout:
2376	if (err && tp_created)
2377		tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2378errout_tp:
2379	if (chain) {
2380		if (tp && !IS_ERR(tp))
2381			tcf_proto_put(tp, rtnl_held, NULL);
2382		if (!tp_created)
2383			tcf_chain_put(chain);
2384	}
2385	tcf_block_release(q, block, rtnl_held);
2386
2387	if (rtnl_held)
2388		rtnl_unlock();
2389
2390	if (err == -EAGAIN) {
2391		/* Take rtnl lock in case EAGAIN is caused by concurrent flush
2392		 * of target chain.
2393		 */
2394		rtnl_held = true;
2395		/* Replay the request. */
2396		goto replay;
2397	}
2398	return err;
2399
2400errout_locked:
2401	mutex_unlock(&chain->filter_chain_lock);
2402	goto errout;
2403}
2404
2405static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2406			  struct netlink_ext_ack *extack)
2407{
2408	struct net *net = sock_net(skb->sk);
2409	struct nlattr *tca[TCA_MAX + 1];
2410	char name[IFNAMSIZ];
2411	struct tcmsg *t;
2412	u32 protocol;
2413	u32 prio;
2414	u32 parent;
2415	u32 chain_index;
2416	struct Qdisc *q = NULL;
2417	struct tcf_chain_info chain_info;
2418	struct tcf_chain *chain = NULL;
2419	struct tcf_block *block = NULL;
2420	struct tcf_proto *tp = NULL;
2421	unsigned long cl = 0;
2422	void *fh = NULL;
2423	int err;
2424	bool rtnl_held = false;
2425
2426	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2427				     rtm_tca_policy, extack);
2428	if (err < 0)
2429		return err;
2430
2431	t = nlmsg_data(n);
2432	protocol = TC_H_MIN(t->tcm_info);
2433	prio = TC_H_MAJ(t->tcm_info);
2434	parent = t->tcm_parent;
2435
2436	if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2437		NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2438		return -ENOENT;
2439	}
2440
2441	/* Find head of filter chain. */
2442
2443	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2444	if (err)
2445		return err;
2446
2447	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2448		NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2449		err = -EINVAL;
2450		goto errout;
2451	}
2452	/* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2453	 * found), qdisc is not unlocked, classifier type is not specified,
2454	 * classifier is not unlocked.
2455	 */
2456	if (!prio ||
2457	    (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2458	    !tcf_proto_is_unlocked(name)) {
2459		rtnl_held = true;
2460		rtnl_lock();
2461	}
2462
2463	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2464	if (err)
2465		goto errout;
2466
2467	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2468				 extack);
2469	if (IS_ERR(block)) {
2470		err = PTR_ERR(block);
2471		goto errout;
2472	}
2473
2474	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2475	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2476		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2477		err = -EINVAL;
2478		goto errout;
2479	}
2480	chain = tcf_chain_get(block, chain_index, false);
2481	if (!chain) {
2482		/* User requested flush on non-existent chain. Nothing to do,
2483		 * so just return success.
2484		 */
2485		if (prio == 0) {
2486			err = 0;
2487			goto errout;
2488		}
2489		NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2490		err = -ENOENT;
2491		goto errout;
2492	}
2493
2494	if (prio == 0) {
2495		tfilter_notify_chain(net, skb, block, q, parent, n,
2496				     chain, RTM_DELTFILTER, extack);
2497		tcf_chain_flush(chain, rtnl_held);
2498		err = 0;
2499		goto errout;
2500	}
2501
2502	mutex_lock(&chain->filter_chain_lock);
2503	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2504			       prio, false);
2505	if (!tp || IS_ERR(tp)) {
 
2506		NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2507		err = tp ? PTR_ERR(tp) : -ENOENT;
 
 
2508		goto errout_locked;
2509	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2510		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2511		err = -EINVAL;
2512		goto errout_locked;
2513	} else if (t->tcm_handle == 0) {
2514		tcf_proto_signal_destroying(chain, tp);
2515		tcf_chain_tp_remove(chain, &chain_info, tp);
2516		mutex_unlock(&chain->filter_chain_lock);
2517
2518		tcf_proto_put(tp, rtnl_held, NULL);
2519		tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2520			       RTM_DELTFILTER, false, rtnl_held, extack);
2521		err = 0;
2522		goto errout;
2523	}
2524	mutex_unlock(&chain->filter_chain_lock);
2525
2526	fh = tp->ops->get(tp, t->tcm_handle);
2527
2528	if (!fh) {
2529		NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2530		err = -ENOENT;
2531	} else {
2532		bool last;
2533
2534		err = tfilter_del_notify(net, skb, n, tp, block, q, parent, fh,
2535					 &last, rtnl_held, extack);
2536
2537		if (err)
2538			goto errout;
2539		if (last)
2540			tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2541	}
2542
2543errout:
2544	if (chain) {
2545		if (tp && !IS_ERR(tp))
2546			tcf_proto_put(tp, rtnl_held, NULL);
2547		tcf_chain_put(chain);
2548	}
2549	tcf_block_release(q, block, rtnl_held);
2550
2551	if (rtnl_held)
2552		rtnl_unlock();
2553
2554	return err;
2555
2556errout_locked:
2557	mutex_unlock(&chain->filter_chain_lock);
2558	goto errout;
2559}
2560
2561static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2562			  struct netlink_ext_ack *extack)
2563{
2564	struct net *net = sock_net(skb->sk);
2565	struct nlattr *tca[TCA_MAX + 1];
2566	char name[IFNAMSIZ];
2567	struct tcmsg *t;
2568	u32 protocol;
2569	u32 prio;
2570	u32 parent;
2571	u32 chain_index;
2572	struct Qdisc *q = NULL;
2573	struct tcf_chain_info chain_info;
2574	struct tcf_chain *chain = NULL;
2575	struct tcf_block *block = NULL;
2576	struct tcf_proto *tp = NULL;
2577	unsigned long cl = 0;
2578	void *fh = NULL;
2579	int err;
2580	bool rtnl_held = false;
2581
2582	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2583				     rtm_tca_policy, extack);
2584	if (err < 0)
2585		return err;
2586
2587	t = nlmsg_data(n);
2588	protocol = TC_H_MIN(t->tcm_info);
2589	prio = TC_H_MAJ(t->tcm_info);
2590	parent = t->tcm_parent;
2591
2592	if (prio == 0) {
2593		NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2594		return -ENOENT;
2595	}
2596
2597	/* Find head of filter chain. */
2598
2599	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2600	if (err)
2601		return err;
2602
2603	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2604		NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2605		err = -EINVAL;
2606		goto errout;
2607	}
2608	/* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2609	 * unlocked, classifier type is not specified, classifier is not
2610	 * unlocked.
2611	 */
2612	if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2613	    !tcf_proto_is_unlocked(name)) {
2614		rtnl_held = true;
2615		rtnl_lock();
2616	}
2617
2618	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2619	if (err)
2620		goto errout;
2621
2622	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2623				 extack);
2624	if (IS_ERR(block)) {
2625		err = PTR_ERR(block);
2626		goto errout;
2627	}
2628
2629	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2630	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2631		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2632		err = -EINVAL;
2633		goto errout;
2634	}
2635	chain = tcf_chain_get(block, chain_index, false);
2636	if (!chain) {
2637		NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2638		err = -EINVAL;
2639		goto errout;
2640	}
2641
2642	mutex_lock(&chain->filter_chain_lock);
2643	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2644			       prio, false);
2645	mutex_unlock(&chain->filter_chain_lock);
2646	if (!tp || IS_ERR(tp)) {
 
2647		NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2648		err = tp ? PTR_ERR(tp) : -ENOENT;
 
 
2649		goto errout;
2650	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2651		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2652		err = -EINVAL;
2653		goto errout;
2654	}
2655
2656	fh = tp->ops->get(tp, t->tcm_handle);
2657
2658	if (!fh) {
2659		NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2660		err = -ENOENT;
2661	} else {
2662		err = tfilter_notify(net, skb, n, tp, block, q, parent,
2663				     fh, RTM_NEWTFILTER, true, rtnl_held, NULL);
2664		if (err < 0)
2665			NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2666	}
2667
2668	tfilter_put(tp, fh);
2669errout:
2670	if (chain) {
2671		if (tp && !IS_ERR(tp))
2672			tcf_proto_put(tp, rtnl_held, NULL);
2673		tcf_chain_put(chain);
2674	}
2675	tcf_block_release(q, block, rtnl_held);
2676
2677	if (rtnl_held)
2678		rtnl_unlock();
2679
2680	return err;
2681}
2682
2683struct tcf_dump_args {
2684	struct tcf_walker w;
2685	struct sk_buff *skb;
2686	struct netlink_callback *cb;
2687	struct tcf_block *block;
2688	struct Qdisc *q;
2689	u32 parent;
2690	bool terse_dump;
2691};
2692
2693static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2694{
2695	struct tcf_dump_args *a = (void *)arg;
2696	struct net *net = sock_net(a->skb->sk);
2697
2698	return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2699			     n, NETLINK_CB(a->cb->skb).portid,
2700			     a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2701			     RTM_NEWTFILTER, a->terse_dump, true, NULL);
2702}
2703
2704static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2705			   struct sk_buff *skb, struct netlink_callback *cb,
2706			   long index_start, long *p_index, bool terse)
2707{
2708	struct net *net = sock_net(skb->sk);
2709	struct tcf_block *block = chain->block;
2710	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2711	struct tcf_proto *tp, *tp_prev;
2712	struct tcf_dump_args arg;
2713
2714	for (tp = __tcf_get_next_proto(chain, NULL);
2715	     tp;
2716	     tp_prev = tp,
2717		     tp = __tcf_get_next_proto(chain, tp),
2718		     tcf_proto_put(tp_prev, true, NULL),
2719		     (*p_index)++) {
2720		if (*p_index < index_start)
2721			continue;
2722		if (TC_H_MAJ(tcm->tcm_info) &&
2723		    TC_H_MAJ(tcm->tcm_info) != tp->prio)
2724			continue;
2725		if (TC_H_MIN(tcm->tcm_info) &&
2726		    TC_H_MIN(tcm->tcm_info) != tp->protocol)
2727			continue;
2728		if (*p_index > index_start)
2729			memset(&cb->args[1], 0,
2730			       sizeof(cb->args) - sizeof(cb->args[0]));
2731		if (cb->args[1] == 0) {
2732			if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2733					  NETLINK_CB(cb->skb).portid,
2734					  cb->nlh->nlmsg_seq, NLM_F_MULTI,
2735					  RTM_NEWTFILTER, false, true, NULL) <= 0)
2736				goto errout;
2737			cb->args[1] = 1;
2738		}
2739		if (!tp->ops->walk)
2740			continue;
2741		arg.w.fn = tcf_node_dump;
2742		arg.skb = skb;
2743		arg.cb = cb;
2744		arg.block = block;
2745		arg.q = q;
2746		arg.parent = parent;
2747		arg.w.stop = 0;
2748		arg.w.skip = cb->args[1] - 1;
2749		arg.w.count = 0;
2750		arg.w.cookie = cb->args[2];
2751		arg.terse_dump = terse;
2752		tp->ops->walk(tp, &arg.w, true);
2753		cb->args[2] = arg.w.cookie;
2754		cb->args[1] = arg.w.count + 1;
2755		if (arg.w.stop)
2756			goto errout;
2757	}
2758	return true;
2759
2760errout:
2761	tcf_proto_put(tp, true, NULL);
2762	return false;
2763}
2764
2765static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2766	[TCA_CHAIN]      = { .type = NLA_U32 },
2767	[TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2768};
2769
2770/* called with RTNL */
2771static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2772{
2773	struct tcf_chain *chain, *chain_prev;
2774	struct net *net = sock_net(skb->sk);
2775	struct nlattr *tca[TCA_MAX + 1];
2776	struct Qdisc *q = NULL;
2777	struct tcf_block *block;
2778	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2779	bool terse_dump = false;
2780	long index_start;
2781	long index;
2782	u32 parent;
2783	int err;
2784
2785	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2786		return skb->len;
2787
2788	err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2789				     tcf_tfilter_dump_policy, cb->extack);
2790	if (err)
2791		return err;
2792
2793	if (tca[TCA_DUMP_FLAGS]) {
2794		struct nla_bitfield32 flags =
2795			nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2796
2797		terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2798	}
2799
2800	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2801		block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2802		if (!block)
2803			goto out;
2804		/* If we work with block index, q is NULL and parent value
2805		 * will never be used in the following code. The check
2806		 * in tcf_fill_node prevents it. However, compiler does not
2807		 * see that far, so set parent to zero to silence the warning
2808		 * about parent being uninitialized.
2809		 */
2810		parent = 0;
2811	} else {
2812		const struct Qdisc_class_ops *cops;
2813		struct net_device *dev;
2814		unsigned long cl = 0;
2815
2816		dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2817		if (!dev)
2818			return skb->len;
2819
2820		parent = tcm->tcm_parent;
2821		if (!parent)
2822			q = rtnl_dereference(dev->qdisc);
2823		else
2824			q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2825		if (!q)
2826			goto out;
2827		cops = q->ops->cl_ops;
2828		if (!cops)
2829			goto out;
2830		if (!cops->tcf_block)
2831			goto out;
2832		if (TC_H_MIN(tcm->tcm_parent)) {
2833			cl = cops->find(q, tcm->tcm_parent);
2834			if (cl == 0)
2835				goto out;
2836		}
2837		block = cops->tcf_block(q, cl, NULL);
2838		if (!block)
2839			goto out;
2840		parent = block->classid;
2841		if (tcf_block_shared(block))
2842			q = NULL;
2843	}
2844
2845	index_start = cb->args[0];
2846	index = 0;
2847
2848	for (chain = __tcf_get_next_chain(block, NULL);
2849	     chain;
2850	     chain_prev = chain,
2851		     chain = __tcf_get_next_chain(block, chain),
2852		     tcf_chain_put(chain_prev)) {
2853		if (tca[TCA_CHAIN] &&
2854		    nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2855			continue;
2856		if (!tcf_chain_dump(chain, q, parent, skb, cb,
2857				    index_start, &index, terse_dump)) {
2858			tcf_chain_put(chain);
2859			err = -EMSGSIZE;
2860			break;
2861		}
2862	}
2863
2864	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2865		tcf_block_refcnt_put(block, true);
2866	cb->args[0] = index;
2867
2868out:
2869	/* If we did no progress, the error (EMSGSIZE) is real */
2870	if (skb->len == 0 && err)
2871		return err;
2872	return skb->len;
2873}
2874
2875static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2876			      void *tmplt_priv, u32 chain_index,
2877			      struct net *net, struct sk_buff *skb,
2878			      struct tcf_block *block,
2879			      u32 portid, u32 seq, u16 flags, int event,
2880			      struct netlink_ext_ack *extack)
2881{
2882	unsigned char *b = skb_tail_pointer(skb);
2883	const struct tcf_proto_ops *ops;
2884	struct nlmsghdr *nlh;
2885	struct tcmsg *tcm;
2886	void *priv;
2887
2888	ops = tmplt_ops;
2889	priv = tmplt_priv;
2890
2891	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2892	if (!nlh)
2893		goto out_nlmsg_trim;
2894	tcm = nlmsg_data(nlh);
2895	tcm->tcm_family = AF_UNSPEC;
2896	tcm->tcm__pad1 = 0;
2897	tcm->tcm__pad2 = 0;
2898	tcm->tcm_handle = 0;
2899	if (block->q) {
2900		tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2901		tcm->tcm_parent = block->q->handle;
2902	} else {
2903		tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2904		tcm->tcm_block_index = block->index;
2905	}
2906
2907	if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2908		goto nla_put_failure;
2909
2910	if (ops) {
2911		if (nla_put_string(skb, TCA_KIND, ops->kind))
2912			goto nla_put_failure;
2913		if (ops->tmplt_dump(skb, net, priv) < 0)
2914			goto nla_put_failure;
2915	}
2916
2917	if (extack && extack->_msg &&
2918	    nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2919		goto out_nlmsg_trim;
2920
2921	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2922
2923	return skb->len;
2924
2925out_nlmsg_trim:
2926nla_put_failure:
2927	nlmsg_trim(skb, b);
2928	return -EMSGSIZE;
2929}
2930
2931static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2932			   u32 seq, u16 flags, int event, bool unicast,
2933			   struct netlink_ext_ack *extack)
2934{
2935	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2936	struct tcf_block *block = chain->block;
2937	struct net *net = block->net;
2938	struct sk_buff *skb;
2939	int err = 0;
2940
2941	if (!unicast && !rtnl_notify_needed(net, flags, RTNLGRP_TC))
2942		return 0;
2943
2944	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2945	if (!skb)
2946		return -ENOBUFS;
2947
2948	if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2949			       chain->index, net, skb, block, portid,
2950			       seq, flags, event, extack) <= 0) {
2951		kfree_skb(skb);
2952		return -EINVAL;
2953	}
2954
2955	if (unicast)
2956		err = rtnl_unicast(skb, net, portid);
2957	else
2958		err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2959				     flags & NLM_F_ECHO);
2960
2961	return err;
2962}
2963
2964static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2965				  void *tmplt_priv, u32 chain_index,
2966				  struct tcf_block *block, struct sk_buff *oskb,
2967				  u32 seq, u16 flags)
2968{
2969	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2970	struct net *net = block->net;
2971	struct sk_buff *skb;
2972
2973	if (!rtnl_notify_needed(net, flags, RTNLGRP_TC))
2974		return 0;
2975
2976	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2977	if (!skb)
2978		return -ENOBUFS;
2979
2980	if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2981			       block, portid, seq, flags, RTM_DELCHAIN, NULL) <= 0) {
2982		kfree_skb(skb);
2983		return -EINVAL;
2984	}
2985
2986	return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2987}
2988
2989static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2990			      struct nlattr **tca,
2991			      struct netlink_ext_ack *extack)
2992{
2993	const struct tcf_proto_ops *ops;
2994	char name[IFNAMSIZ];
2995	void *tmplt_priv;
2996
2997	/* If kind is not set, user did not specify template. */
2998	if (!tca[TCA_KIND])
2999		return 0;
3000
3001	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
3002		NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
3003		return -EINVAL;
3004	}
3005
3006	ops = tcf_proto_lookup_ops(name, true, extack);
3007	if (IS_ERR(ops))
3008		return PTR_ERR(ops);
3009	if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump ||
3010	    !ops->tmplt_reoffload) {
3011		NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
3012		module_put(ops->owner);
3013		return -EOPNOTSUPP;
3014	}
3015
3016	tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
3017	if (IS_ERR(tmplt_priv)) {
3018		module_put(ops->owner);
3019		return PTR_ERR(tmplt_priv);
3020	}
3021	chain->tmplt_ops = ops;
3022	chain->tmplt_priv = tmplt_priv;
3023	return 0;
3024}
3025
3026static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
3027			       void *tmplt_priv)
3028{
3029	/* If template ops are set, no work to do for us. */
3030	if (!tmplt_ops)
3031		return;
3032
3033	tmplt_ops->tmplt_destroy(tmplt_priv);
3034	module_put(tmplt_ops->owner);
3035}
3036
3037/* Add/delete/get a chain */
3038
3039static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
3040			struct netlink_ext_ack *extack)
3041{
3042	struct net *net = sock_net(skb->sk);
3043	struct nlattr *tca[TCA_MAX + 1];
3044	struct tcmsg *t;
3045	u32 parent;
3046	u32 chain_index;
3047	struct Qdisc *q;
3048	struct tcf_chain *chain;
3049	struct tcf_block *block;
3050	unsigned long cl;
3051	int err;
3052
3053replay:
3054	q = NULL;
3055	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
3056				     rtm_tca_policy, extack);
3057	if (err < 0)
3058		return err;
3059
3060	t = nlmsg_data(n);
3061	parent = t->tcm_parent;
3062	cl = 0;
3063
3064	block = tcf_block_find(net, &q, &parent, &cl,
3065			       t->tcm_ifindex, t->tcm_block_index, extack);
3066	if (IS_ERR(block))
3067		return PTR_ERR(block);
3068
3069	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
3070	if (chain_index > TC_ACT_EXT_VAL_MASK) {
3071		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
3072		err = -EINVAL;
3073		goto errout_block;
3074	}
3075
3076	mutex_lock(&block->lock);
3077	chain = tcf_chain_lookup(block, chain_index);
3078	if (n->nlmsg_type == RTM_NEWCHAIN) {
3079		if (chain) {
3080			if (tcf_chain_held_by_acts_only(chain)) {
3081				/* The chain exists only because there is
3082				 * some action referencing it.
3083				 */
3084				tcf_chain_hold(chain);
3085			} else {
3086				NL_SET_ERR_MSG(extack, "Filter chain already exists");
3087				err = -EEXIST;
3088				goto errout_block_locked;
3089			}
3090		} else {
3091			if (!(n->nlmsg_flags & NLM_F_CREATE)) {
3092				NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
3093				err = -ENOENT;
3094				goto errout_block_locked;
3095			}
3096			chain = tcf_chain_create(block, chain_index);
3097			if (!chain) {
3098				NL_SET_ERR_MSG(extack, "Failed to create filter chain");
3099				err = -ENOMEM;
3100				goto errout_block_locked;
3101			}
3102		}
3103	} else {
3104		if (!chain || tcf_chain_held_by_acts_only(chain)) {
3105			NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
3106			err = -EINVAL;
3107			goto errout_block_locked;
3108		}
3109		tcf_chain_hold(chain);
3110	}
3111
3112	if (n->nlmsg_type == RTM_NEWCHAIN) {
3113		/* Modifying chain requires holding parent block lock. In case
3114		 * the chain was successfully added, take a reference to the
3115		 * chain. This ensures that an empty chain does not disappear at
3116		 * the end of this function.
3117		 */
3118		tcf_chain_hold(chain);
3119		chain->explicitly_created = true;
3120	}
3121	mutex_unlock(&block->lock);
3122
3123	switch (n->nlmsg_type) {
3124	case RTM_NEWCHAIN:
3125		err = tc_chain_tmplt_add(chain, net, tca, extack);
3126		if (err) {
3127			tcf_chain_put_explicitly_created(chain);
3128			goto errout;
3129		}
3130
3131		tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
3132				RTM_NEWCHAIN, false, extack);
3133		break;
3134	case RTM_DELCHAIN:
3135		tfilter_notify_chain(net, skb, block, q, parent, n,
3136				     chain, RTM_DELTFILTER, extack);
3137		/* Flush the chain first as the user requested chain removal. */
3138		tcf_chain_flush(chain, true);
3139		/* In case the chain was successfully deleted, put a reference
3140		 * to the chain previously taken during addition.
3141		 */
3142		tcf_chain_put_explicitly_created(chain);
3143		break;
3144	case RTM_GETCHAIN:
3145		err = tc_chain_notify(chain, skb, n->nlmsg_seq,
3146				      n->nlmsg_flags, n->nlmsg_type, true, extack);
3147		if (err < 0)
3148			NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
3149		break;
3150	default:
3151		err = -EOPNOTSUPP;
3152		NL_SET_ERR_MSG(extack, "Unsupported message type");
3153		goto errout;
3154	}
3155
3156errout:
3157	tcf_chain_put(chain);
3158errout_block:
3159	tcf_block_release(q, block, true);
3160	if (err == -EAGAIN)
3161		/* Replay the request. */
3162		goto replay;
3163	return err;
3164
3165errout_block_locked:
3166	mutex_unlock(&block->lock);
3167	goto errout_block;
3168}
3169
3170/* called with RTNL */
3171static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
3172{
3173	struct net *net = sock_net(skb->sk);
3174	struct nlattr *tca[TCA_MAX + 1];
3175	struct Qdisc *q = NULL;
3176	struct tcf_block *block;
3177	struct tcmsg *tcm = nlmsg_data(cb->nlh);
3178	struct tcf_chain *chain;
3179	long index_start;
3180	long index;
3181	int err;
3182
3183	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
3184		return skb->len;
3185
3186	err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
3187				     rtm_tca_policy, cb->extack);
3188	if (err)
3189		return err;
3190
3191	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
3192		block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
3193		if (!block)
3194			goto out;
3195	} else {
3196		const struct Qdisc_class_ops *cops;
3197		struct net_device *dev;
3198		unsigned long cl = 0;
3199
3200		dev = __dev_get_by_index(net, tcm->tcm_ifindex);
3201		if (!dev)
3202			return skb->len;
3203
3204		if (!tcm->tcm_parent)
3205			q = rtnl_dereference(dev->qdisc);
3206		else
3207			q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
3208
3209		if (!q)
3210			goto out;
3211		cops = q->ops->cl_ops;
3212		if (!cops)
3213			goto out;
3214		if (!cops->tcf_block)
3215			goto out;
3216		if (TC_H_MIN(tcm->tcm_parent)) {
3217			cl = cops->find(q, tcm->tcm_parent);
3218			if (cl == 0)
3219				goto out;
3220		}
3221		block = cops->tcf_block(q, cl, NULL);
3222		if (!block)
3223			goto out;
3224		if (tcf_block_shared(block))
3225			q = NULL;
3226	}
3227
3228	index_start = cb->args[0];
3229	index = 0;
3230
3231	mutex_lock(&block->lock);
3232	list_for_each_entry(chain, &block->chain_list, list) {
3233		if ((tca[TCA_CHAIN] &&
3234		     nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3235			continue;
3236		if (index < index_start) {
3237			index++;
3238			continue;
3239		}
3240		if (tcf_chain_held_by_acts_only(chain))
3241			continue;
3242		err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3243					 chain->index, net, skb, block,
3244					 NETLINK_CB(cb->skb).portid,
3245					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3246					 RTM_NEWCHAIN, NULL);
3247		if (err <= 0)
3248			break;
3249		index++;
3250	}
3251	mutex_unlock(&block->lock);
3252
3253	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3254		tcf_block_refcnt_put(block, true);
3255	cb->args[0] = index;
3256
3257out:
3258	/* If we did no progress, the error (EMSGSIZE) is real */
3259	if (skb->len == 0 && err)
3260		return err;
3261	return skb->len;
3262}
3263
3264int tcf_exts_init_ex(struct tcf_exts *exts, struct net *net, int action,
3265		     int police, struct tcf_proto *tp, u32 handle,
3266		     bool use_action_miss)
3267{
3268	int err = 0;
3269
3270#ifdef CONFIG_NET_CLS_ACT
3271	exts->type = 0;
3272	exts->nr_actions = 0;
3273	exts->miss_cookie_node = NULL;
3274	/* Note: we do not own yet a reference on net.
3275	 * This reference might be taken later from tcf_exts_get_net().
3276	 */
3277	exts->net = net;
3278	exts->actions = kcalloc(TCA_ACT_MAX_PRIO, sizeof(struct tc_action *),
3279				GFP_KERNEL);
3280	if (!exts->actions)
3281		return -ENOMEM;
3282#endif
3283
3284	exts->action = action;
3285	exts->police = police;
3286
3287	if (!use_action_miss)
3288		return 0;
3289
3290	err = tcf_exts_miss_cookie_base_alloc(exts, tp, handle);
3291	if (err)
3292		goto err_miss_alloc;
3293
3294	return 0;
3295
3296err_miss_alloc:
3297	tcf_exts_destroy(exts);
3298#ifdef CONFIG_NET_CLS_ACT
3299	exts->actions = NULL;
3300#endif
3301	return err;
3302}
3303EXPORT_SYMBOL(tcf_exts_init_ex);
3304
3305void tcf_exts_destroy(struct tcf_exts *exts)
3306{
3307	tcf_exts_miss_cookie_base_destroy(exts);
3308
3309#ifdef CONFIG_NET_CLS_ACT
3310	if (exts->actions) {
3311		tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3312		kfree(exts->actions);
3313	}
3314	exts->nr_actions = 0;
3315#endif
3316}
3317EXPORT_SYMBOL(tcf_exts_destroy);
3318
3319int tcf_exts_validate_ex(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3320			 struct nlattr *rate_tlv, struct tcf_exts *exts,
3321			 u32 flags, u32 fl_flags, struct netlink_ext_ack *extack)
3322{
3323#ifdef CONFIG_NET_CLS_ACT
3324	{
3325		int init_res[TCA_ACT_MAX_PRIO] = {};
3326		struct tc_action *act;
3327		size_t attr_size = 0;
3328
3329		if (exts->police && tb[exts->police]) {
3330			struct tc_action_ops *a_o;
3331
3332			flags |= TCA_ACT_FLAGS_POLICE | TCA_ACT_FLAGS_BIND;
3333			a_o = tc_action_load_ops(tb[exts->police], flags,
3334						 extack);
3335			if (IS_ERR(a_o))
3336				return PTR_ERR(a_o);
3337			act = tcf_action_init_1(net, tp, tb[exts->police],
3338						rate_tlv, a_o, init_res, flags,
3339						extack);
3340			module_put(a_o->owner);
3341			if (IS_ERR(act))
3342				return PTR_ERR(act);
3343
3344			act->type = exts->type = TCA_OLD_COMPAT;
3345			exts->actions[0] = act;
3346			exts->nr_actions = 1;
3347			tcf_idr_insert_many(exts->actions, init_res);
3348		} else if (exts->action && tb[exts->action]) {
3349			int err;
3350
3351			flags |= TCA_ACT_FLAGS_BIND;
3352			err = tcf_action_init(net, tp, tb[exts->action],
3353					      rate_tlv, exts->actions, init_res,
3354					      &attr_size, flags, fl_flags,
3355					      extack);
3356			if (err < 0)
3357				return err;
3358			exts->nr_actions = err;
3359		}
3360	}
3361#else
3362	if ((exts->action && tb[exts->action]) ||
3363	    (exts->police && tb[exts->police])) {
3364		NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3365		return -EOPNOTSUPP;
3366	}
3367#endif
3368
3369	return 0;
3370}
3371EXPORT_SYMBOL(tcf_exts_validate_ex);
3372
3373int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3374		      struct nlattr *rate_tlv, struct tcf_exts *exts,
3375		      u32 flags, struct netlink_ext_ack *extack)
3376{
3377	return tcf_exts_validate_ex(net, tp, tb, rate_tlv, exts,
3378				    flags, 0, extack);
3379}
3380EXPORT_SYMBOL(tcf_exts_validate);
3381
3382void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3383{
3384#ifdef CONFIG_NET_CLS_ACT
3385	struct tcf_exts old = *dst;
3386
3387	*dst = *src;
3388	tcf_exts_destroy(&old);
3389#endif
3390}
3391EXPORT_SYMBOL(tcf_exts_change);
3392
3393#ifdef CONFIG_NET_CLS_ACT
3394static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3395{
3396	if (exts->nr_actions == 0)
3397		return NULL;
3398	else
3399		return exts->actions[0];
3400}
3401#endif
3402
3403int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3404{
3405#ifdef CONFIG_NET_CLS_ACT
3406	struct nlattr *nest;
3407
3408	if (exts->action && tcf_exts_has_actions(exts)) {
3409		/*
3410		 * again for backward compatible mode - we want
3411		 * to work with both old and new modes of entering
3412		 * tc data even if iproute2  was newer - jhs
3413		 */
3414		if (exts->type != TCA_OLD_COMPAT) {
3415			nest = nla_nest_start_noflag(skb, exts->action);
3416			if (nest == NULL)
3417				goto nla_put_failure;
3418
3419			if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3420			    < 0)
3421				goto nla_put_failure;
3422			nla_nest_end(skb, nest);
3423		} else if (exts->police) {
3424			struct tc_action *act = tcf_exts_first_act(exts);
3425			nest = nla_nest_start_noflag(skb, exts->police);
3426			if (nest == NULL || !act)
3427				goto nla_put_failure;
3428			if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3429				goto nla_put_failure;
3430			nla_nest_end(skb, nest);
3431		}
3432	}
3433	return 0;
3434
3435nla_put_failure:
3436	nla_nest_cancel(skb, nest);
3437	return -1;
3438#else
3439	return 0;
3440#endif
3441}
3442EXPORT_SYMBOL(tcf_exts_dump);
3443
3444int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3445{
3446#ifdef CONFIG_NET_CLS_ACT
3447	struct nlattr *nest;
3448
3449	if (!exts->action || !tcf_exts_has_actions(exts))
3450		return 0;
3451
3452	nest = nla_nest_start_noflag(skb, exts->action);
3453	if (!nest)
3454		goto nla_put_failure;
3455
3456	if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3457		goto nla_put_failure;
3458	nla_nest_end(skb, nest);
3459	return 0;
3460
3461nla_put_failure:
3462	nla_nest_cancel(skb, nest);
3463	return -1;
3464#else
3465	return 0;
3466#endif
3467}
3468EXPORT_SYMBOL(tcf_exts_terse_dump);
3469
3470int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3471{
3472#ifdef CONFIG_NET_CLS_ACT
3473	struct tc_action *a = tcf_exts_first_act(exts);
3474	if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3475		return -1;
3476#endif
3477	return 0;
3478}
3479EXPORT_SYMBOL(tcf_exts_dump_stats);
3480
3481static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3482{
3483	if (*flags & TCA_CLS_FLAGS_IN_HW)
3484		return;
3485	*flags |= TCA_CLS_FLAGS_IN_HW;
3486	atomic_inc(&block->offloadcnt);
3487}
3488
3489static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3490{
3491	if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3492		return;
3493	*flags &= ~TCA_CLS_FLAGS_IN_HW;
3494	atomic_dec(&block->offloadcnt);
3495}
3496
3497static void tc_cls_offload_cnt_update(struct tcf_block *block,
3498				      struct tcf_proto *tp, u32 *cnt,
3499				      u32 *flags, u32 diff, bool add)
3500{
3501	lockdep_assert_held(&block->cb_lock);
3502
3503	spin_lock(&tp->lock);
3504	if (add) {
3505		if (!*cnt)
3506			tcf_block_offload_inc(block, flags);
3507		*cnt += diff;
3508	} else {
3509		*cnt -= diff;
3510		if (!*cnt)
3511			tcf_block_offload_dec(block, flags);
3512	}
3513	spin_unlock(&tp->lock);
3514}
3515
3516static void
3517tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3518			 u32 *cnt, u32 *flags)
3519{
3520	lockdep_assert_held(&block->cb_lock);
3521
3522	spin_lock(&tp->lock);
3523	tcf_block_offload_dec(block, flags);
3524	*cnt = 0;
3525	spin_unlock(&tp->lock);
3526}
3527
3528static int
3529__tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3530		   void *type_data, bool err_stop)
3531{
3532	struct flow_block_cb *block_cb;
3533	int ok_count = 0;
3534	int err;
3535
3536	list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3537		err = block_cb->cb(type, type_data, block_cb->cb_priv);
3538		if (err) {
3539			if (err_stop)
3540				return err;
3541		} else {
3542			ok_count++;
3543		}
3544	}
3545	return ok_count;
3546}
3547
3548int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3549		     void *type_data, bool err_stop, bool rtnl_held)
3550{
3551	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3552	int ok_count;
3553
3554retry:
3555	if (take_rtnl)
3556		rtnl_lock();
3557	down_read(&block->cb_lock);
3558	/* Need to obtain rtnl lock if block is bound to devs that require it.
3559	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3560	 * obtain the locks in same order here.
3561	 */
3562	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3563		up_read(&block->cb_lock);
3564		take_rtnl = true;
3565		goto retry;
3566	}
3567
3568	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3569
3570	up_read(&block->cb_lock);
3571	if (take_rtnl)
3572		rtnl_unlock();
3573	return ok_count;
3574}
3575EXPORT_SYMBOL(tc_setup_cb_call);
3576
3577/* Non-destructive filter add. If filter that wasn't already in hardware is
3578 * successfully offloaded, increment block offloads counter. On failure,
3579 * previously offloaded filter is considered to be intact and offloads counter
3580 * is not decremented.
3581 */
3582
3583int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3584		    enum tc_setup_type type, void *type_data, bool err_stop,
3585		    u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3586{
3587	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3588	int ok_count;
3589
3590retry:
3591	if (take_rtnl)
3592		rtnl_lock();
3593	down_read(&block->cb_lock);
3594	/* Need to obtain rtnl lock if block is bound to devs that require it.
3595	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3596	 * obtain the locks in same order here.
3597	 */
3598	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3599		up_read(&block->cb_lock);
3600		take_rtnl = true;
3601		goto retry;
3602	}
3603
3604	/* Make sure all netdevs sharing this block are offload-capable. */
3605	if (block->nooffloaddevcnt && err_stop) {
3606		ok_count = -EOPNOTSUPP;
3607		goto err_unlock;
3608	}
3609
3610	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3611	if (ok_count < 0)
3612		goto err_unlock;
3613
3614	if (tp->ops->hw_add)
3615		tp->ops->hw_add(tp, type_data);
3616	if (ok_count > 0)
3617		tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3618					  ok_count, true);
3619err_unlock:
3620	up_read(&block->cb_lock);
3621	if (take_rtnl)
3622		rtnl_unlock();
3623	return min(ok_count, 0);
3624}
3625EXPORT_SYMBOL(tc_setup_cb_add);
3626
3627/* Destructive filter replace. If filter that wasn't already in hardware is
3628 * successfully offloaded, increment block offload counter. On failure,
3629 * previously offloaded filter is considered to be destroyed and offload counter
3630 * is decremented.
3631 */
3632
3633int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3634			enum tc_setup_type type, void *type_data, bool err_stop,
3635			u32 *old_flags, unsigned int *old_in_hw_count,
3636			u32 *new_flags, unsigned int *new_in_hw_count,
3637			bool rtnl_held)
3638{
3639	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3640	int ok_count;
3641
3642retry:
3643	if (take_rtnl)
3644		rtnl_lock();
3645	down_read(&block->cb_lock);
3646	/* Need to obtain rtnl lock if block is bound to devs that require it.
3647	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3648	 * obtain the locks in same order here.
3649	 */
3650	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3651		up_read(&block->cb_lock);
3652		take_rtnl = true;
3653		goto retry;
3654	}
3655
3656	/* Make sure all netdevs sharing this block are offload-capable. */
3657	if (block->nooffloaddevcnt && err_stop) {
3658		ok_count = -EOPNOTSUPP;
3659		goto err_unlock;
3660	}
3661
3662	tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3663	if (tp->ops->hw_del)
3664		tp->ops->hw_del(tp, type_data);
3665
3666	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3667	if (ok_count < 0)
3668		goto err_unlock;
3669
3670	if (tp->ops->hw_add)
3671		tp->ops->hw_add(tp, type_data);
3672	if (ok_count > 0)
3673		tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3674					  new_flags, ok_count, true);
3675err_unlock:
3676	up_read(&block->cb_lock);
3677	if (take_rtnl)
3678		rtnl_unlock();
3679	return min(ok_count, 0);
3680}
3681EXPORT_SYMBOL(tc_setup_cb_replace);
3682
3683/* Destroy filter and decrement block offload counter, if filter was previously
3684 * offloaded.
3685 */
3686
3687int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3688			enum tc_setup_type type, void *type_data, bool err_stop,
3689			u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3690{
3691	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3692	int ok_count;
3693
3694retry:
3695	if (take_rtnl)
3696		rtnl_lock();
3697	down_read(&block->cb_lock);
3698	/* Need to obtain rtnl lock if block is bound to devs that require it.
3699	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3700	 * obtain the locks in same order here.
3701	 */
3702	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3703		up_read(&block->cb_lock);
3704		take_rtnl = true;
3705		goto retry;
3706	}
3707
3708	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3709
3710	tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3711	if (tp->ops->hw_del)
3712		tp->ops->hw_del(tp, type_data);
3713
3714	up_read(&block->cb_lock);
3715	if (take_rtnl)
3716		rtnl_unlock();
3717	return min(ok_count, 0);
3718}
3719EXPORT_SYMBOL(tc_setup_cb_destroy);
3720
3721int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3722			  bool add, flow_setup_cb_t *cb,
3723			  enum tc_setup_type type, void *type_data,
3724			  void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3725{
3726	int err = cb(type, type_data, cb_priv);
3727
3728	if (err) {
3729		if (add && tc_skip_sw(*flags))
3730			return err;
3731	} else {
3732		tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3733					  add);
3734	}
3735
3736	return 0;
3737}
3738EXPORT_SYMBOL(tc_setup_cb_reoffload);
3739
3740static int tcf_act_get_user_cookie(struct flow_action_entry *entry,
3741				   const struct tc_action *act)
3742{
3743	struct tc_cookie *user_cookie;
3744	int err = 0;
3745
3746	rcu_read_lock();
3747	user_cookie = rcu_dereference(act->user_cookie);
3748	if (user_cookie) {
3749		entry->user_cookie = flow_action_cookie_create(user_cookie->data,
3750							       user_cookie->len,
3751							       GFP_ATOMIC);
3752		if (!entry->user_cookie)
3753			err = -ENOMEM;
3754	}
3755	rcu_read_unlock();
3756	return err;
3757}
3758
3759static void tcf_act_put_user_cookie(struct flow_action_entry *entry)
3760{
3761	flow_action_cookie_destroy(entry->user_cookie);
3762}
3763
3764void tc_cleanup_offload_action(struct flow_action *flow_action)
3765{
3766	struct flow_action_entry *entry;
3767	int i;
3768
3769	flow_action_for_each(i, entry, flow_action) {
3770		tcf_act_put_user_cookie(entry);
3771		if (entry->destructor)
3772			entry->destructor(entry->destructor_priv);
3773	}
3774}
3775EXPORT_SYMBOL(tc_cleanup_offload_action);
3776
3777static int tc_setup_offload_act(struct tc_action *act,
3778				struct flow_action_entry *entry,
3779				u32 *index_inc,
3780				struct netlink_ext_ack *extack)
3781{
3782#ifdef CONFIG_NET_CLS_ACT
3783	if (act->ops->offload_act_setup) {
3784		return act->ops->offload_act_setup(act, entry, index_inc, true,
3785						   extack);
3786	} else {
3787		NL_SET_ERR_MSG(extack, "Action does not support offload");
3788		return -EOPNOTSUPP;
3789	}
3790#else
3791	return 0;
3792#endif
3793}
3794
3795int tc_setup_action(struct flow_action *flow_action,
3796		    struct tc_action *actions[],
3797		    u32 miss_cookie_base,
3798		    struct netlink_ext_ack *extack)
3799{
3800	int i, j, k, index, err = 0;
3801	struct tc_action *act;
3802
3803	BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3804	BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3805	BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3806
3807	if (!actions)
3808		return 0;
3809
3810	j = 0;
3811	tcf_act_for_each_action(i, act, actions) {
3812		struct flow_action_entry *entry;
3813
3814		entry = &flow_action->entries[j];
3815		spin_lock_bh(&act->tcfa_lock);
3816		err = tcf_act_get_user_cookie(entry, act);
3817		if (err)
3818			goto err_out_locked;
3819
3820		index = 0;
3821		err = tc_setup_offload_act(act, entry, &index, extack);
3822		if (err)
3823			goto err_out_locked;
3824
3825		for (k = 0; k < index ; k++) {
3826			entry[k].hw_stats = tc_act_hw_stats(act->hw_stats);
3827			entry[k].hw_index = act->tcfa_index;
3828			entry[k].cookie = (unsigned long)act;
3829			entry[k].miss_cookie =
3830				tcf_exts_miss_cookie_get(miss_cookie_base, i);
3831		}
3832
3833		j += index;
3834
3835		spin_unlock_bh(&act->tcfa_lock);
3836	}
3837
3838err_out:
3839	if (err)
3840		tc_cleanup_offload_action(flow_action);
3841
3842	return err;
3843err_out_locked:
3844	spin_unlock_bh(&act->tcfa_lock);
3845	goto err_out;
3846}
3847
3848int tc_setup_offload_action(struct flow_action *flow_action,
3849			    const struct tcf_exts *exts,
3850			    struct netlink_ext_ack *extack)
3851{
3852#ifdef CONFIG_NET_CLS_ACT
3853	u32 miss_cookie_base;
3854
3855	if (!exts)
3856		return 0;
3857
3858	miss_cookie_base = exts->miss_cookie_node ?
3859			   exts->miss_cookie_node->miss_cookie_base : 0;
3860	return tc_setup_action(flow_action, exts->actions, miss_cookie_base,
3861			       extack);
3862#else
3863	return 0;
3864#endif
3865}
3866EXPORT_SYMBOL(tc_setup_offload_action);
3867
3868unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3869{
3870	unsigned int num_acts = 0;
3871	struct tc_action *act;
3872	int i;
3873
3874	tcf_exts_for_each_action(i, act, exts) {
3875		if (is_tcf_pedit(act))
3876			num_acts += tcf_pedit_nkeys(act);
3877		else
3878			num_acts++;
3879	}
3880	return num_acts;
3881}
3882EXPORT_SYMBOL(tcf_exts_num_actions);
3883
3884#ifdef CONFIG_NET_CLS_ACT
3885static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3886					u32 *p_block_index,
3887					struct netlink_ext_ack *extack)
3888{
3889	*p_block_index = nla_get_u32(block_index_attr);
3890	if (!*p_block_index) {
3891		NL_SET_ERR_MSG(extack, "Block number may not be zero");
3892		return -EINVAL;
3893	}
3894
3895	return 0;
3896}
3897
3898int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3899		    enum flow_block_binder_type binder_type,
3900		    struct nlattr *block_index_attr,
3901		    struct netlink_ext_ack *extack)
3902{
3903	u32 block_index;
3904	int err;
3905
3906	if (!block_index_attr)
3907		return 0;
3908
3909	err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3910	if (err)
3911		return err;
3912
3913	qe->info.binder_type = binder_type;
3914	qe->info.chain_head_change = tcf_chain_head_change_dflt;
3915	qe->info.chain_head_change_priv = &qe->filter_chain;
3916	qe->info.block_index = block_index;
3917
3918	return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3919}
3920EXPORT_SYMBOL(tcf_qevent_init);
3921
3922void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3923{
3924	if (qe->info.block_index)
3925		tcf_block_put_ext(qe->block, sch, &qe->info);
3926}
3927EXPORT_SYMBOL(tcf_qevent_destroy);
3928
3929int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3930			       struct netlink_ext_ack *extack)
3931{
3932	u32 block_index;
3933	int err;
3934
3935	if (!block_index_attr)
3936		return 0;
3937
3938	err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3939	if (err)
3940		return err;
3941
3942	/* Bounce newly-configured block or change in block. */
3943	if (block_index != qe->info.block_index) {
3944		NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3945		return -EINVAL;
3946	}
3947
3948	return 0;
3949}
3950EXPORT_SYMBOL(tcf_qevent_validate_change);
3951
3952struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
3953				  struct sk_buff **to_free, int *ret)
3954{
3955	struct tcf_result cl_res;
3956	struct tcf_proto *fl;
3957
3958	if (!qe->info.block_index)
3959		return skb;
3960
3961	fl = rcu_dereference_bh(qe->filter_chain);
3962
3963	switch (tcf_classify(skb, NULL, fl, &cl_res, false)) {
3964	case TC_ACT_SHOT:
3965		qdisc_qstats_drop(sch);
3966		__qdisc_drop(skb, to_free);
3967		*ret = __NET_XMIT_BYPASS;
3968		return NULL;
3969	case TC_ACT_STOLEN:
3970	case TC_ACT_QUEUED:
3971	case TC_ACT_TRAP:
3972		__qdisc_drop(skb, to_free);
3973		*ret = __NET_XMIT_STOLEN;
3974		return NULL;
3975	case TC_ACT_REDIRECT:
3976		skb_do_redirect(skb);
3977		*ret = __NET_XMIT_STOLEN;
3978		return NULL;
3979	}
3980
3981	return skb;
3982}
3983EXPORT_SYMBOL(tcf_qevent_handle);
3984
3985int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
3986{
3987	if (!qe->info.block_index)
3988		return 0;
3989	return nla_put_u32(skb, attr_name, qe->info.block_index);
3990}
3991EXPORT_SYMBOL(tcf_qevent_dump);
3992#endif
3993
3994static __net_init int tcf_net_init(struct net *net)
3995{
3996	struct tcf_net *tn = net_generic(net, tcf_net_id);
3997
3998	spin_lock_init(&tn->idr_lock);
3999	idr_init(&tn->idr);
4000	return 0;
4001}
4002
4003static void __net_exit tcf_net_exit(struct net *net)
4004{
4005	struct tcf_net *tn = net_generic(net, tcf_net_id);
4006
4007	idr_destroy(&tn->idr);
4008}
4009
4010static struct pernet_operations tcf_net_ops = {
4011	.init = tcf_net_init,
4012	.exit = tcf_net_exit,
4013	.id   = &tcf_net_id,
4014	.size = sizeof(struct tcf_net),
4015};
4016
 
 
 
 
 
 
 
 
 
 
 
 
 
4017static int __init tc_filter_init(void)
4018{
4019	int err;
4020
4021	tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
4022	if (!tc_filter_wq)
4023		return -ENOMEM;
4024
4025	err = register_pernet_subsys(&tcf_net_ops);
4026	if (err)
4027		goto err_register_pernet_subsys;
4028
4029	xa_init_flags(&tcf_exts_miss_cookies_xa, XA_FLAGS_ALLOC1);
4030
4031	rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
4032		      RTNL_FLAG_DOIT_UNLOCKED);
4033	rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
4034		      RTNL_FLAG_DOIT_UNLOCKED);
4035	rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
4036		      tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
4037	rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
4038	rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
4039	rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
4040		      tc_dump_chain, 0);
4041
4042	return 0;
4043
4044err_register_pernet_subsys:
4045	destroy_workqueue(tc_filter_wq);
4046	return err;
4047}
4048
4049subsys_initcall(tc_filter_init);
v6.13.7
   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 */
  47static LIST_HEAD(tcf_proto_base);
  48
  49/* Protects list of registered TC modules. It is pure SMP lock. */
  50static DEFINE_RWLOCK(cls_mod_lock);
  51
  52static struct xarray tcf_exts_miss_cookies_xa;
  53struct 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 */
  67union 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)
  76static int
  77tcf_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 < 0)
 101		goto err_xa_alloc;
 102
 103	exts->miss_cookie_node = n;
 104	return 0;
 105
 106err_xa_alloc:
 107	kfree(n);
 108	return err;
 109}
 110
 111static 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
 123static struct tcf_exts_miss_cookie_node *
 124tcf_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) */
 132static int
 133tcf_exts_miss_cookie_base_alloc(struct tcf_exts *exts, struct tcf_proto *tp,
 134				u32 handle)
 135{
 136	return 0;
 137}
 138
 139static void tcf_exts_miss_cookie_base_destroy(struct tcf_exts *exts)
 140{
 141}
 142#endif /* IS_ENABLED(CONFIG_NET_TC_SKB_EXT) */
 143
 144static 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
 156DEFINE_STATIC_KEY_FALSE(tc_skb_ext_tc);
 157EXPORT_SYMBOL(tc_skb_ext_tc);
 158
 159void tc_skb_ext_tc_enable(void)
 160{
 161	static_branch_inc(&tc_skb_ext_tc);
 162}
 163EXPORT_SYMBOL(tc_skb_ext_tc_enable);
 164
 165void tc_skb_ext_tc_disable(void)
 166{
 167	static_branch_dec(&tc_skb_ext_tc);
 168}
 169EXPORT_SYMBOL(tc_skb_ext_tc_disable);
 170#endif
 171
 172static 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
 178static 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
 189static 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
 197static 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
 217static void
 218tcf_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
 230static 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
 248static const struct tcf_proto_ops *
 249tcf_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(NET_CLS_ALIAS_PREFIX "%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
 280int 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;
 292out:
 293	write_unlock(&cls_mod_lock);
 294	return rc;
 295}
 296EXPORT_SYMBOL(register_tcf_proto_ops);
 297
 298static struct workqueue_struct *tc_filter_wq;
 299
 300void 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}
 323EXPORT_SYMBOL(unregister_tcf_proto_ops);
 324
 325bool 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}
 330EXPORT_SYMBOL(tcf_queue_work);
 331
 332/* Select new prio value from the range, managed by kernel. */
 333
 334static 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
 344static 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
 352static 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
 372static 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	tp->usesw = !tp->ops->reoffload;
 394	spin_lock_init(&tp->lock);
 395	refcount_set(&tp->refcnt, 1);
 396
 397	err = tp->ops->init(tp);
 398	if (err) {
 399		module_put(tp->ops->owner);
 400		goto errout;
 401	}
 402	return tp;
 403
 404errout:
 405	kfree(tp);
 406	return ERR_PTR(err);
 407}
 408
 409static void tcf_proto_get(struct tcf_proto *tp)
 410{
 411	refcount_inc(&tp->refcnt);
 412}
 413
 414static void tcf_proto_count_usesw(struct tcf_proto *tp, bool add)
 415{
 416#ifdef CONFIG_NET_CLS_ACT
 417	struct tcf_block *block = tp->chain->block;
 418	bool counted = false;
 419
 420	if (!add) {
 421		if (tp->usesw && tp->counted) {
 422			if (!atomic_dec_return(&block->useswcnt))
 423				static_branch_dec(&tcf_sw_enabled_key);
 424			tp->counted = false;
 425		}
 426		return;
 427	}
 428
 429	spin_lock(&tp->lock);
 430	if (tp->usesw && !tp->counted) {
 431		counted = true;
 432		tp->counted = true;
 433	}
 434	spin_unlock(&tp->lock);
 435
 436	if (counted && atomic_inc_return(&block->useswcnt) == 1)
 437		static_branch_inc(&tcf_sw_enabled_key);
 438#endif
 439}
 440
 441static void tcf_chain_put(struct tcf_chain *chain);
 442
 443static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
 444			      bool sig_destroy, struct netlink_ext_ack *extack)
 445{
 446	tp->ops->destroy(tp, rtnl_held, extack);
 447	tcf_proto_count_usesw(tp, false);
 448	if (sig_destroy)
 449		tcf_proto_signal_destroyed(tp->chain, tp);
 450	tcf_chain_put(tp->chain);
 451	module_put(tp->ops->owner);
 452	kfree_rcu(tp, rcu);
 453}
 454
 455static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
 456			  struct netlink_ext_ack *extack)
 457{
 458	if (refcount_dec_and_test(&tp->refcnt))
 459		tcf_proto_destroy(tp, rtnl_held, true, extack);
 460}
 461
 462static bool tcf_proto_check_delete(struct tcf_proto *tp)
 463{
 464	if (tp->ops->delete_empty)
 465		return tp->ops->delete_empty(tp);
 466
 467	tp->deleting = true;
 468	return tp->deleting;
 469}
 470
 471static void tcf_proto_mark_delete(struct tcf_proto *tp)
 472{
 473	spin_lock(&tp->lock);
 474	tp->deleting = true;
 475	spin_unlock(&tp->lock);
 476}
 477
 478static bool tcf_proto_is_deleting(struct tcf_proto *tp)
 479{
 480	bool deleting;
 481
 482	spin_lock(&tp->lock);
 483	deleting = tp->deleting;
 484	spin_unlock(&tp->lock);
 485
 486	return deleting;
 487}
 488
 489#define ASSERT_BLOCK_LOCKED(block)					\
 490	lockdep_assert_held(&(block)->lock)
 491
 492struct tcf_filter_chain_list_item {
 493	struct list_head list;
 494	tcf_chain_head_change_t *chain_head_change;
 495	void *chain_head_change_priv;
 496};
 497
 498static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
 499					  u32 chain_index)
 500{
 501	struct tcf_chain *chain;
 502
 503	ASSERT_BLOCK_LOCKED(block);
 504
 505	chain = kzalloc(sizeof(*chain), GFP_KERNEL);
 506	if (!chain)
 507		return NULL;
 508	list_add_tail_rcu(&chain->list, &block->chain_list);
 509	mutex_init(&chain->filter_chain_lock);
 510	chain->block = block;
 511	chain->index = chain_index;
 512	chain->refcnt = 1;
 513	if (!chain->index)
 514		block->chain0.chain = chain;
 515	return chain;
 516}
 517
 518static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
 519				       struct tcf_proto *tp_head)
 520{
 521	if (item->chain_head_change)
 522		item->chain_head_change(tp_head, item->chain_head_change_priv);
 523}
 524
 525static void tcf_chain0_head_change(struct tcf_chain *chain,
 526				   struct tcf_proto *tp_head)
 527{
 528	struct tcf_filter_chain_list_item *item;
 529	struct tcf_block *block = chain->block;
 530
 531	if (chain->index)
 532		return;
 533
 534	mutex_lock(&block->lock);
 535	list_for_each_entry(item, &block->chain0.filter_chain_list, list)
 536		tcf_chain_head_change_item(item, tp_head);
 537	mutex_unlock(&block->lock);
 538}
 539
 540/* Returns true if block can be safely freed. */
 541
 542static bool tcf_chain_detach(struct tcf_chain *chain)
 543{
 544	struct tcf_block *block = chain->block;
 545
 546	ASSERT_BLOCK_LOCKED(block);
 547
 548	list_del_rcu(&chain->list);
 549	if (!chain->index)
 550		block->chain0.chain = NULL;
 551
 552	if (list_empty(&block->chain_list) &&
 553	    refcount_read(&block->refcnt) == 0)
 554		return true;
 555
 556	return false;
 557}
 558
 559static void tcf_block_destroy(struct tcf_block *block)
 560{
 561	mutex_destroy(&block->lock);
 562	mutex_destroy(&block->proto_destroy_lock);
 563	xa_destroy(&block->ports);
 564	kfree_rcu(block, rcu);
 565}
 566
 567static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
 568{
 569	struct tcf_block *block = chain->block;
 570
 571	mutex_destroy(&chain->filter_chain_lock);
 572	kfree_rcu(chain, rcu);
 573	if (free_block)
 574		tcf_block_destroy(block);
 575}
 576
 577static void tcf_chain_hold(struct tcf_chain *chain)
 578{
 579	ASSERT_BLOCK_LOCKED(chain->block);
 580
 581	++chain->refcnt;
 582}
 583
 584static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
 585{
 586	ASSERT_BLOCK_LOCKED(chain->block);
 587
 588	/* In case all the references are action references, this
 589	 * chain should not be shown to the user.
 590	 */
 591	return chain->refcnt == chain->action_refcnt;
 592}
 593
 594static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
 595					  u32 chain_index)
 596{
 597	struct tcf_chain *chain;
 598
 599	ASSERT_BLOCK_LOCKED(block);
 600
 601	list_for_each_entry(chain, &block->chain_list, list) {
 602		if (chain->index == chain_index)
 603			return chain;
 604	}
 605	return NULL;
 606}
 607
 608#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
 609static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
 610					      u32 chain_index)
 611{
 612	struct tcf_chain *chain;
 613
 614	list_for_each_entry_rcu(chain, &block->chain_list, list) {
 615		if (chain->index == chain_index)
 616			return chain;
 617	}
 618	return NULL;
 619}
 620#endif
 621
 622static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
 623			   u32 seq, u16 flags, int event, bool unicast,
 624			   struct netlink_ext_ack *extack);
 625
 626static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
 627					 u32 chain_index, bool create,
 628					 bool by_act)
 629{
 630	struct tcf_chain *chain = NULL;
 631	bool is_first_reference;
 632
 633	mutex_lock(&block->lock);
 634	chain = tcf_chain_lookup(block, chain_index);
 635	if (chain) {
 636		tcf_chain_hold(chain);
 637	} else {
 638		if (!create)
 639			goto errout;
 640		chain = tcf_chain_create(block, chain_index);
 641		if (!chain)
 642			goto errout;
 643	}
 644
 645	if (by_act)
 646		++chain->action_refcnt;
 647	is_first_reference = chain->refcnt - chain->action_refcnt == 1;
 648	mutex_unlock(&block->lock);
 649
 650	/* Send notification only in case we got the first
 651	 * non-action reference. Until then, the chain acts only as
 652	 * a placeholder for actions pointing to it and user ought
 653	 * not know about them.
 654	 */
 655	if (is_first_reference && !by_act)
 656		tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
 657				RTM_NEWCHAIN, false, NULL);
 658
 659	return chain;
 660
 661errout:
 662	mutex_unlock(&block->lock);
 663	return chain;
 664}
 665
 666static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
 667				       bool create)
 668{
 669	return __tcf_chain_get(block, chain_index, create, false);
 670}
 671
 672struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
 673{
 674	return __tcf_chain_get(block, chain_index, true, true);
 675}
 676EXPORT_SYMBOL(tcf_chain_get_by_act);
 677
 678static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
 679			       void *tmplt_priv);
 680static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
 681				  void *tmplt_priv, u32 chain_index,
 682				  struct tcf_block *block, struct sk_buff *oskb,
 683				  u32 seq, u16 flags);
 684
 685static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
 686			    bool explicitly_created)
 687{
 688	struct tcf_block *block = chain->block;
 689	const struct tcf_proto_ops *tmplt_ops;
 690	unsigned int refcnt, non_act_refcnt;
 691	bool free_block = false;
 692	void *tmplt_priv;
 693
 694	mutex_lock(&block->lock);
 695	if (explicitly_created) {
 696		if (!chain->explicitly_created) {
 697			mutex_unlock(&block->lock);
 698			return;
 699		}
 700		chain->explicitly_created = false;
 701	}
 702
 703	if (by_act)
 704		chain->action_refcnt--;
 705
 706	/* tc_chain_notify_delete can't be called while holding block lock.
 707	 * However, when block is unlocked chain can be changed concurrently, so
 708	 * save these to temporary variables.
 709	 */
 710	refcnt = --chain->refcnt;
 711	non_act_refcnt = refcnt - chain->action_refcnt;
 712	tmplt_ops = chain->tmplt_ops;
 713	tmplt_priv = chain->tmplt_priv;
 714
 715	if (non_act_refcnt == chain->explicitly_created && !by_act) {
 716		if (non_act_refcnt == 0)
 717			tc_chain_notify_delete(tmplt_ops, tmplt_priv,
 718					       chain->index, block, NULL, 0, 0);
 719		/* Last reference to chain, no need to lock. */
 720		chain->flushing = false;
 721	}
 722
 723	if (refcnt == 0)
 724		free_block = tcf_chain_detach(chain);
 725	mutex_unlock(&block->lock);
 726
 727	if (refcnt == 0) {
 728		tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
 729		tcf_chain_destroy(chain, free_block);
 730	}
 731}
 732
 733static void tcf_chain_put(struct tcf_chain *chain)
 734{
 735	__tcf_chain_put(chain, false, false);
 736}
 737
 738void tcf_chain_put_by_act(struct tcf_chain *chain)
 739{
 740	__tcf_chain_put(chain, true, false);
 741}
 742EXPORT_SYMBOL(tcf_chain_put_by_act);
 743
 744static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
 745{
 746	__tcf_chain_put(chain, false, true);
 747}
 748
 749static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
 750{
 751	struct tcf_proto *tp, *tp_next;
 752
 753	mutex_lock(&chain->filter_chain_lock);
 754	tp = tcf_chain_dereference(chain->filter_chain, chain);
 755	while (tp) {
 756		tp_next = rcu_dereference_protected(tp->next, 1);
 757		tcf_proto_signal_destroying(chain, tp);
 758		tp = tp_next;
 759	}
 760	tp = tcf_chain_dereference(chain->filter_chain, chain);
 761	RCU_INIT_POINTER(chain->filter_chain, NULL);
 762	tcf_chain0_head_change(chain, NULL);
 763	chain->flushing = true;
 764	mutex_unlock(&chain->filter_chain_lock);
 765
 766	while (tp) {
 767		tp_next = rcu_dereference_protected(tp->next, 1);
 768		tcf_proto_put(tp, rtnl_held, NULL);
 769		tp = tp_next;
 770	}
 771}
 772
 773static int tcf_block_setup(struct tcf_block *block,
 774			   struct flow_block_offload *bo);
 775
 776static void tcf_block_offload_init(struct flow_block_offload *bo,
 777				   struct net_device *dev, struct Qdisc *sch,
 778				   enum flow_block_command command,
 779				   enum flow_block_binder_type binder_type,
 780				   struct flow_block *flow_block,
 781				   bool shared, struct netlink_ext_ack *extack)
 782{
 783	bo->net = dev_net(dev);
 784	bo->command = command;
 785	bo->binder_type = binder_type;
 786	bo->block = flow_block;
 787	bo->block_shared = shared;
 788	bo->extack = extack;
 789	bo->sch = sch;
 790	bo->cb_list_head = &flow_block->cb_list;
 791	INIT_LIST_HEAD(&bo->cb_list);
 792}
 793
 794static void tcf_block_unbind(struct tcf_block *block,
 795			     struct flow_block_offload *bo);
 796
 797static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
 798{
 799	struct tcf_block *block = block_cb->indr.data;
 800	struct net_device *dev = block_cb->indr.dev;
 801	struct Qdisc *sch = block_cb->indr.sch;
 802	struct netlink_ext_ack extack = {};
 803	struct flow_block_offload bo = {};
 804
 805	tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
 806			       block_cb->indr.binder_type,
 807			       &block->flow_block, tcf_block_shared(block),
 808			       &extack);
 809	rtnl_lock();
 810	down_write(&block->cb_lock);
 811	list_del(&block_cb->driver_list);
 812	list_move(&block_cb->list, &bo.cb_list);
 813	tcf_block_unbind(block, &bo);
 814	up_write(&block->cb_lock);
 815	rtnl_unlock();
 816}
 817
 818static bool tcf_block_offload_in_use(struct tcf_block *block)
 819{
 820	return atomic_read(&block->offloadcnt);
 821}
 822
 823static int tcf_block_offload_cmd(struct tcf_block *block,
 824				 struct net_device *dev, struct Qdisc *sch,
 825				 struct tcf_block_ext_info *ei,
 826				 enum flow_block_command command,
 827				 struct netlink_ext_ack *extack)
 828{
 829	struct flow_block_offload bo = {};
 830
 831	tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
 832			       &block->flow_block, tcf_block_shared(block),
 833			       extack);
 834
 835	if (dev->netdev_ops->ndo_setup_tc) {
 836		int err;
 837
 838		err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
 839		if (err < 0) {
 840			if (err != -EOPNOTSUPP)
 841				NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
 842			return err;
 843		}
 844
 845		return tcf_block_setup(block, &bo);
 846	}
 847
 848	flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
 849				    tc_block_indr_cleanup);
 850	tcf_block_setup(block, &bo);
 851
 852	return -EOPNOTSUPP;
 853}
 854
 855static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
 856				  struct tcf_block_ext_info *ei,
 857				  struct netlink_ext_ack *extack)
 858{
 859	struct net_device *dev = q->dev_queue->dev;
 860	int err;
 861
 862	down_write(&block->cb_lock);
 863
 864	/* If tc offload feature is disabled and the block we try to bind
 865	 * to already has some offloaded filters, forbid to bind.
 866	 */
 867	if (dev->netdev_ops->ndo_setup_tc &&
 868	    !tc_can_offload(dev) &&
 869	    tcf_block_offload_in_use(block)) {
 870		NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
 871		err = -EOPNOTSUPP;
 872		goto err_unlock;
 873	}
 874
 875	err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
 876	if (err == -EOPNOTSUPP)
 877		goto no_offload_dev_inc;
 878	if (err)
 879		goto err_unlock;
 880
 881	up_write(&block->cb_lock);
 882	return 0;
 883
 884no_offload_dev_inc:
 885	if (tcf_block_offload_in_use(block))
 886		goto err_unlock;
 887
 888	err = 0;
 889	block->nooffloaddevcnt++;
 890err_unlock:
 891	up_write(&block->cb_lock);
 892	return err;
 893}
 894
 895static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
 896				     struct tcf_block_ext_info *ei)
 897{
 898	struct net_device *dev = q->dev_queue->dev;
 899	int err;
 900
 901	down_write(&block->cb_lock);
 902	err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
 903	if (err == -EOPNOTSUPP)
 904		goto no_offload_dev_dec;
 905	up_write(&block->cb_lock);
 906	return;
 907
 908no_offload_dev_dec:
 909	WARN_ON(block->nooffloaddevcnt-- == 0);
 910	up_write(&block->cb_lock);
 911}
 912
 913static int
 914tcf_chain0_head_change_cb_add(struct tcf_block *block,
 915			      struct tcf_block_ext_info *ei,
 916			      struct netlink_ext_ack *extack)
 917{
 918	struct tcf_filter_chain_list_item *item;
 919	struct tcf_chain *chain0;
 920
 921	item = kmalloc(sizeof(*item), GFP_KERNEL);
 922	if (!item) {
 923		NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
 924		return -ENOMEM;
 925	}
 926	item->chain_head_change = ei->chain_head_change;
 927	item->chain_head_change_priv = ei->chain_head_change_priv;
 928
 929	mutex_lock(&block->lock);
 930	chain0 = block->chain0.chain;
 931	if (chain0)
 932		tcf_chain_hold(chain0);
 933	else
 934		list_add(&item->list, &block->chain0.filter_chain_list);
 935	mutex_unlock(&block->lock);
 936
 937	if (chain0) {
 938		struct tcf_proto *tp_head;
 939
 940		mutex_lock(&chain0->filter_chain_lock);
 941
 942		tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
 943		if (tp_head)
 944			tcf_chain_head_change_item(item, tp_head);
 945
 946		mutex_lock(&block->lock);
 947		list_add(&item->list, &block->chain0.filter_chain_list);
 948		mutex_unlock(&block->lock);
 949
 950		mutex_unlock(&chain0->filter_chain_lock);
 951		tcf_chain_put(chain0);
 952	}
 953
 954	return 0;
 955}
 956
 957static void
 958tcf_chain0_head_change_cb_del(struct tcf_block *block,
 959			      struct tcf_block_ext_info *ei)
 960{
 961	struct tcf_filter_chain_list_item *item;
 962
 963	mutex_lock(&block->lock);
 964	list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
 965		if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
 966		    (item->chain_head_change == ei->chain_head_change &&
 967		     item->chain_head_change_priv == ei->chain_head_change_priv)) {
 968			if (block->chain0.chain)
 969				tcf_chain_head_change_item(item, NULL);
 970			list_del(&item->list);
 971			mutex_unlock(&block->lock);
 972
 973			kfree(item);
 974			return;
 975		}
 976	}
 977	mutex_unlock(&block->lock);
 978	WARN_ON(1);
 979}
 980
 981struct tcf_net {
 982	spinlock_t idr_lock; /* Protects idr */
 983	struct idr idr;
 984};
 985
 986static unsigned int tcf_net_id;
 987
 988static int tcf_block_insert(struct tcf_block *block, struct net *net,
 989			    struct netlink_ext_ack *extack)
 990{
 991	struct tcf_net *tn = net_generic(net, tcf_net_id);
 992	int err;
 993
 994	idr_preload(GFP_KERNEL);
 995	spin_lock(&tn->idr_lock);
 996	err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
 997			    GFP_NOWAIT);
 998	spin_unlock(&tn->idr_lock);
 999	idr_preload_end();
1000
1001	return err;
1002}
1003
1004static void tcf_block_remove(struct tcf_block *block, struct net *net)
1005{
1006	struct tcf_net *tn = net_generic(net, tcf_net_id);
1007
1008	spin_lock(&tn->idr_lock);
1009	idr_remove(&tn->idr, block->index);
1010	spin_unlock(&tn->idr_lock);
1011}
1012
1013static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
1014					  u32 block_index,
1015					  struct netlink_ext_ack *extack)
1016{
1017	struct tcf_block *block;
1018
1019	block = kzalloc(sizeof(*block), GFP_KERNEL);
1020	if (!block) {
1021		NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
1022		return ERR_PTR(-ENOMEM);
1023	}
1024	mutex_init(&block->lock);
1025	mutex_init(&block->proto_destroy_lock);
1026	init_rwsem(&block->cb_lock);
1027	flow_block_init(&block->flow_block);
1028	INIT_LIST_HEAD(&block->chain_list);
1029	INIT_LIST_HEAD(&block->owner_list);
1030	INIT_LIST_HEAD(&block->chain0.filter_chain_list);
1031
1032	refcount_set(&block->refcnt, 1);
1033	block->net = net;
1034	block->index = block_index;
1035	xa_init(&block->ports);
1036
1037	/* Don't store q pointer for blocks which are shared */
1038	if (!tcf_block_shared(block))
1039		block->q = q;
1040	return block;
1041}
1042
1043struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
1044{
1045	struct tcf_net *tn = net_generic(net, tcf_net_id);
1046
1047	return idr_find(&tn->idr, block_index);
1048}
1049EXPORT_SYMBOL(tcf_block_lookup);
1050
1051static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
1052{
1053	struct tcf_block *block;
1054
1055	rcu_read_lock();
1056	block = tcf_block_lookup(net, block_index);
1057	if (block && !refcount_inc_not_zero(&block->refcnt))
1058		block = NULL;
1059	rcu_read_unlock();
1060
1061	return block;
1062}
1063
1064static struct tcf_chain *
1065__tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1066{
1067	mutex_lock(&block->lock);
1068	if (chain)
1069		chain = list_is_last(&chain->list, &block->chain_list) ?
1070			NULL : list_next_entry(chain, list);
1071	else
1072		chain = list_first_entry_or_null(&block->chain_list,
1073						 struct tcf_chain, list);
1074
1075	/* skip all action-only chains */
1076	while (chain && tcf_chain_held_by_acts_only(chain))
1077		chain = list_is_last(&chain->list, &block->chain_list) ?
1078			NULL : list_next_entry(chain, list);
1079
1080	if (chain)
1081		tcf_chain_hold(chain);
1082	mutex_unlock(&block->lock);
1083
1084	return chain;
1085}
1086
1087/* Function to be used by all clients that want to iterate over all chains on
1088 * block. It properly obtains block->lock and takes reference to chain before
1089 * returning it. Users of this function must be tolerant to concurrent chain
1090 * insertion/deletion or ensure that no concurrent chain modification is
1091 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1092 * consistent dump because rtnl lock is released each time skb is filled with
1093 * data and sent to user-space.
1094 */
1095
1096struct tcf_chain *
1097tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1098{
1099	struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
1100
1101	if (chain)
1102		tcf_chain_put(chain);
1103
1104	return chain_next;
1105}
1106EXPORT_SYMBOL(tcf_get_next_chain);
1107
1108static struct tcf_proto *
1109__tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1110{
1111	u32 prio = 0;
1112
1113	ASSERT_RTNL();
1114	mutex_lock(&chain->filter_chain_lock);
1115
1116	if (!tp) {
1117		tp = tcf_chain_dereference(chain->filter_chain, chain);
1118	} else if (tcf_proto_is_deleting(tp)) {
1119		/* 'deleting' flag is set and chain->filter_chain_lock was
1120		 * unlocked, which means next pointer could be invalid. Restart
1121		 * search.
1122		 */
1123		prio = tp->prio + 1;
1124		tp = tcf_chain_dereference(chain->filter_chain, chain);
1125
1126		for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1127			if (!tp->deleting && tp->prio >= prio)
1128				break;
1129	} else {
1130		tp = tcf_chain_dereference(tp->next, chain);
1131	}
1132
1133	if (tp)
1134		tcf_proto_get(tp);
1135
1136	mutex_unlock(&chain->filter_chain_lock);
1137
1138	return tp;
1139}
1140
1141/* Function to be used by all clients that want to iterate over all tp's on
1142 * chain. Users of this function must be tolerant to concurrent tp
1143 * insertion/deletion or ensure that no concurrent chain modification is
1144 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1145 * consistent dump because rtnl lock is released each time skb is filled with
1146 * data and sent to user-space.
1147 */
1148
1149struct tcf_proto *
1150tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1151{
1152	struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1153
1154	if (tp)
1155		tcf_proto_put(tp, true, NULL);
1156
1157	return tp_next;
1158}
1159EXPORT_SYMBOL(tcf_get_next_proto);
1160
1161static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1162{
1163	struct tcf_chain *chain;
1164
1165	/* Last reference to block. At this point chains cannot be added or
1166	 * removed concurrently.
1167	 */
1168	for (chain = tcf_get_next_chain(block, NULL);
1169	     chain;
1170	     chain = tcf_get_next_chain(block, chain)) {
1171		tcf_chain_put_explicitly_created(chain);
1172		tcf_chain_flush(chain, rtnl_held);
1173	}
1174}
1175
1176/* Lookup Qdisc and increments its reference counter.
1177 * Set parent, if necessary.
1178 */
1179
1180static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1181			    u32 *parent, int ifindex, bool rtnl_held,
1182			    struct netlink_ext_ack *extack)
1183{
1184	const struct Qdisc_class_ops *cops;
1185	struct net_device *dev;
1186	int err = 0;
1187
1188	if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1189		return 0;
1190
1191	rcu_read_lock();
1192
1193	/* Find link */
1194	dev = dev_get_by_index_rcu(net, ifindex);
1195	if (!dev) {
1196		rcu_read_unlock();
1197		return -ENODEV;
1198	}
1199
1200	/* Find qdisc */
1201	if (!*parent) {
1202		*q = rcu_dereference(dev->qdisc);
1203		*parent = (*q)->handle;
1204	} else {
1205		*q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1206		if (!*q) {
1207			NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1208			err = -EINVAL;
1209			goto errout_rcu;
1210		}
1211	}
1212
1213	*q = qdisc_refcount_inc_nz(*q);
1214	if (!*q) {
1215		NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1216		err = -EINVAL;
1217		goto errout_rcu;
1218	}
1219
1220	/* Is it classful? */
1221	cops = (*q)->ops->cl_ops;
1222	if (!cops) {
1223		NL_SET_ERR_MSG(extack, "Qdisc not classful");
1224		err = -EINVAL;
1225		goto errout_qdisc;
1226	}
1227
1228	if (!cops->tcf_block) {
1229		NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1230		err = -EOPNOTSUPP;
1231		goto errout_qdisc;
1232	}
1233
1234errout_rcu:
1235	/* At this point we know that qdisc is not noop_qdisc,
1236	 * which means that qdisc holds a reference to net_device
1237	 * and we hold a reference to qdisc, so it is safe to release
1238	 * rcu read lock.
1239	 */
1240	rcu_read_unlock();
1241	return err;
1242
1243errout_qdisc:
1244	rcu_read_unlock();
1245
1246	if (rtnl_held)
1247		qdisc_put(*q);
1248	else
1249		qdisc_put_unlocked(*q);
1250	*q = NULL;
1251
1252	return err;
1253}
1254
1255static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1256			       int ifindex, struct netlink_ext_ack *extack)
1257{
1258	if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1259		return 0;
1260
1261	/* Do we search for filter, attached to class? */
1262	if (TC_H_MIN(parent)) {
1263		const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1264
1265		*cl = cops->find(q, parent);
1266		if (*cl == 0) {
1267			NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1268			return -ENOENT;
1269		}
1270	}
1271
1272	return 0;
1273}
1274
1275static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1276					  unsigned long cl, int ifindex,
1277					  u32 block_index,
1278					  struct netlink_ext_ack *extack)
1279{
1280	struct tcf_block *block;
1281
1282	if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1283		block = tcf_block_refcnt_get(net, block_index);
1284		if (!block) {
1285			NL_SET_ERR_MSG(extack, "Block of given index was not found");
1286			return ERR_PTR(-EINVAL);
1287		}
1288	} else {
1289		const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1290
1291		block = cops->tcf_block(q, cl, extack);
1292		if (!block)
1293			return ERR_PTR(-EINVAL);
1294
1295		if (tcf_block_shared(block)) {
1296			NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1297			return ERR_PTR(-EOPNOTSUPP);
1298		}
1299
1300		/* Always take reference to block in order to support execution
1301		 * of rules update path of cls API without rtnl lock. Caller
1302		 * must release block when it is finished using it. 'if' block
1303		 * of this conditional obtain reference to block by calling
1304		 * tcf_block_refcnt_get().
1305		 */
1306		refcount_inc(&block->refcnt);
1307	}
1308
1309	return block;
1310}
1311
1312static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1313			    struct tcf_block_ext_info *ei, bool rtnl_held)
1314{
1315	if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1316		/* Flushing/putting all chains will cause the block to be
1317		 * deallocated when last chain is freed. However, if chain_list
1318		 * is empty, block has to be manually deallocated. After block
1319		 * reference counter reached 0, it is no longer possible to
1320		 * increment it or add new chains to block.
1321		 */
1322		bool free_block = list_empty(&block->chain_list);
1323
1324		mutex_unlock(&block->lock);
1325		if (tcf_block_shared(block))
1326			tcf_block_remove(block, block->net);
1327
1328		if (q)
1329			tcf_block_offload_unbind(block, q, ei);
1330
1331		if (free_block)
1332			tcf_block_destroy(block);
1333		else
1334			tcf_block_flush_all_chains(block, rtnl_held);
1335	} else if (q) {
1336		tcf_block_offload_unbind(block, q, ei);
1337	}
1338}
1339
1340static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1341{
1342	__tcf_block_put(block, NULL, NULL, rtnl_held);
1343}
1344
1345/* Find tcf block.
1346 * Set q, parent, cl when appropriate.
1347 */
1348
1349static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1350					u32 *parent, unsigned long *cl,
1351					int ifindex, u32 block_index,
1352					struct netlink_ext_ack *extack)
1353{
1354	struct tcf_block *block;
1355	int err = 0;
1356
1357	ASSERT_RTNL();
1358
1359	err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1360	if (err)
1361		goto errout;
1362
1363	err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1364	if (err)
1365		goto errout_qdisc;
1366
1367	block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1368	if (IS_ERR(block)) {
1369		err = PTR_ERR(block);
1370		goto errout_qdisc;
1371	}
1372
1373	return block;
1374
1375errout_qdisc:
1376	if (*q)
1377		qdisc_put(*q);
1378errout:
1379	*q = NULL;
1380	return ERR_PTR(err);
1381}
1382
1383static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1384			      bool rtnl_held)
1385{
1386	if (!IS_ERR_OR_NULL(block))
1387		tcf_block_refcnt_put(block, rtnl_held);
1388
1389	if (q) {
1390		if (rtnl_held)
1391			qdisc_put(q);
1392		else
1393			qdisc_put_unlocked(q);
1394	}
1395}
1396
1397struct tcf_block_owner_item {
1398	struct list_head list;
1399	struct Qdisc *q;
1400	enum flow_block_binder_type binder_type;
1401};
1402
1403static void
1404tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1405			       struct Qdisc *q,
1406			       enum flow_block_binder_type binder_type)
1407{
1408	if (block->keep_dst &&
1409	    binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1410	    binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1411		netif_keep_dst(qdisc_dev(q));
1412}
1413
1414void tcf_block_netif_keep_dst(struct tcf_block *block)
1415{
1416	struct tcf_block_owner_item *item;
1417
1418	block->keep_dst = true;
1419	list_for_each_entry(item, &block->owner_list, list)
1420		tcf_block_owner_netif_keep_dst(block, item->q,
1421					       item->binder_type);
1422}
1423EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1424
1425static int tcf_block_owner_add(struct tcf_block *block,
1426			       struct Qdisc *q,
1427			       enum flow_block_binder_type binder_type)
1428{
1429	struct tcf_block_owner_item *item;
1430
1431	item = kmalloc(sizeof(*item), GFP_KERNEL);
1432	if (!item)
1433		return -ENOMEM;
1434	item->q = q;
1435	item->binder_type = binder_type;
1436	list_add(&item->list, &block->owner_list);
1437	return 0;
1438}
1439
1440static void tcf_block_owner_del(struct tcf_block *block,
1441				struct Qdisc *q,
1442				enum flow_block_binder_type binder_type)
1443{
1444	struct tcf_block_owner_item *item;
1445
1446	list_for_each_entry(item, &block->owner_list, list) {
1447		if (item->q == q && item->binder_type == binder_type) {
1448			list_del(&item->list);
1449			kfree(item);
1450			return;
1451		}
1452	}
1453	WARN_ON(1);
1454}
1455
1456static bool tcf_block_tracks_dev(struct tcf_block *block,
1457				 struct tcf_block_ext_info *ei)
1458{
1459	return tcf_block_shared(block) &&
1460	       (ei->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS ||
1461		ei->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS);
1462}
1463
1464int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1465		      struct tcf_block_ext_info *ei,
1466		      struct netlink_ext_ack *extack)
1467{
1468	struct net_device *dev = qdisc_dev(q);
1469	struct net *net = qdisc_net(q);
1470	struct tcf_block *block = NULL;
1471	int err;
1472
1473	if (ei->block_index)
1474		/* block_index not 0 means the shared block is requested */
1475		block = tcf_block_refcnt_get(net, ei->block_index);
1476
1477	if (!block) {
1478		block = tcf_block_create(net, q, ei->block_index, extack);
1479		if (IS_ERR(block))
1480			return PTR_ERR(block);
1481		if (tcf_block_shared(block)) {
1482			err = tcf_block_insert(block, net, extack);
1483			if (err)
1484				goto err_block_insert;
1485		}
1486	}
1487
1488	err = tcf_block_owner_add(block, q, ei->binder_type);
1489	if (err)
1490		goto err_block_owner_add;
1491
1492	tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1493
1494	err = tcf_chain0_head_change_cb_add(block, ei, extack);
1495	if (err)
1496		goto err_chain0_head_change_cb_add;
1497
1498	err = tcf_block_offload_bind(block, q, ei, extack);
1499	if (err)
1500		goto err_block_offload_bind;
1501
1502	if (tcf_block_tracks_dev(block, ei)) {
1503		err = xa_insert(&block->ports, dev->ifindex, dev, GFP_KERNEL);
1504		if (err) {
1505			NL_SET_ERR_MSG(extack, "block dev insert failed");
1506			goto err_dev_insert;
1507		}
1508	}
1509
1510	*p_block = block;
1511	return 0;
1512
1513err_dev_insert:
1514	tcf_block_offload_unbind(block, q, ei);
1515err_block_offload_bind:
1516	tcf_chain0_head_change_cb_del(block, ei);
1517err_chain0_head_change_cb_add:
1518	tcf_block_owner_del(block, q, ei->binder_type);
1519err_block_owner_add:
1520err_block_insert:
1521	tcf_block_refcnt_put(block, true);
1522	return err;
1523}
1524EXPORT_SYMBOL(tcf_block_get_ext);
1525
1526static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1527{
1528	struct tcf_proto __rcu **p_filter_chain = priv;
1529
1530	rcu_assign_pointer(*p_filter_chain, tp_head);
1531}
1532
1533int tcf_block_get(struct tcf_block **p_block,
1534		  struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1535		  struct netlink_ext_ack *extack)
1536{
1537	struct tcf_block_ext_info ei = {
1538		.chain_head_change = tcf_chain_head_change_dflt,
1539		.chain_head_change_priv = p_filter_chain,
1540	};
1541
1542	WARN_ON(!p_filter_chain);
1543	return tcf_block_get_ext(p_block, q, &ei, extack);
1544}
1545EXPORT_SYMBOL(tcf_block_get);
1546
1547/* XXX: Standalone actions are not allowed to jump to any chain, and bound
1548 * actions should be all removed after flushing.
1549 */
1550void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1551		       struct tcf_block_ext_info *ei)
1552{
1553	struct net_device *dev = qdisc_dev(q);
1554
1555	if (!block)
1556		return;
1557	if (tcf_block_tracks_dev(block, ei))
1558		xa_erase(&block->ports, dev->ifindex);
1559	tcf_chain0_head_change_cb_del(block, ei);
1560	tcf_block_owner_del(block, q, ei->binder_type);
1561
1562	__tcf_block_put(block, q, ei, true);
1563}
1564EXPORT_SYMBOL(tcf_block_put_ext);
1565
1566void tcf_block_put(struct tcf_block *block)
1567{
1568	struct tcf_block_ext_info ei = {0, };
1569
1570	if (!block)
1571		return;
1572	tcf_block_put_ext(block, block->q, &ei);
1573}
1574
1575EXPORT_SYMBOL(tcf_block_put);
1576
1577static int
1578tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1579			    void *cb_priv, bool add, bool offload_in_use,
1580			    struct netlink_ext_ack *extack)
1581{
1582	struct tcf_chain *chain, *chain_prev;
1583	struct tcf_proto *tp, *tp_prev;
1584	int err;
1585
1586	lockdep_assert_held(&block->cb_lock);
1587
1588	for (chain = __tcf_get_next_chain(block, NULL);
1589	     chain;
1590	     chain_prev = chain,
1591		     chain = __tcf_get_next_chain(block, chain),
1592		     tcf_chain_put(chain_prev)) {
1593		if (chain->tmplt_ops && add)
1594			chain->tmplt_ops->tmplt_reoffload(chain, true, cb,
1595							  cb_priv);
1596		for (tp = __tcf_get_next_proto(chain, NULL); tp;
1597		     tp_prev = tp,
1598			     tp = __tcf_get_next_proto(chain, tp),
1599			     tcf_proto_put(tp_prev, true, NULL)) {
1600			if (tp->ops->reoffload) {
1601				err = tp->ops->reoffload(tp, add, cb, cb_priv,
1602							 extack);
1603				if (err && add)
1604					goto err_playback_remove;
1605			} else if (add && offload_in_use) {
1606				err = -EOPNOTSUPP;
1607				NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1608				goto err_playback_remove;
1609			}
1610		}
1611		if (chain->tmplt_ops && !add)
1612			chain->tmplt_ops->tmplt_reoffload(chain, false, cb,
1613							  cb_priv);
1614	}
1615
1616	return 0;
1617
1618err_playback_remove:
1619	tcf_proto_put(tp, true, NULL);
1620	tcf_chain_put(chain);
1621	tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1622				    extack);
1623	return err;
1624}
1625
1626static int tcf_block_bind(struct tcf_block *block,
1627			  struct flow_block_offload *bo)
1628{
1629	struct flow_block_cb *block_cb, *next;
1630	int err, i = 0;
1631
1632	lockdep_assert_held(&block->cb_lock);
1633
1634	list_for_each_entry(block_cb, &bo->cb_list, list) {
1635		err = tcf_block_playback_offloads(block, block_cb->cb,
1636						  block_cb->cb_priv, true,
1637						  tcf_block_offload_in_use(block),
1638						  bo->extack);
1639		if (err)
1640			goto err_unroll;
1641		if (!bo->unlocked_driver_cb)
1642			block->lockeddevcnt++;
1643
1644		i++;
1645	}
1646	list_splice(&bo->cb_list, &block->flow_block.cb_list);
1647
1648	return 0;
1649
1650err_unroll:
1651	list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1652		list_del(&block_cb->driver_list);
1653		if (i-- > 0) {
1654			list_del(&block_cb->list);
1655			tcf_block_playback_offloads(block, block_cb->cb,
1656						    block_cb->cb_priv, false,
1657						    tcf_block_offload_in_use(block),
1658						    NULL);
1659			if (!bo->unlocked_driver_cb)
1660				block->lockeddevcnt--;
1661		}
1662		flow_block_cb_free(block_cb);
1663	}
1664
1665	return err;
1666}
1667
1668static void tcf_block_unbind(struct tcf_block *block,
1669			     struct flow_block_offload *bo)
1670{
1671	struct flow_block_cb *block_cb, *next;
1672
1673	lockdep_assert_held(&block->cb_lock);
1674
1675	list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1676		tcf_block_playback_offloads(block, block_cb->cb,
1677					    block_cb->cb_priv, false,
1678					    tcf_block_offload_in_use(block),
1679					    NULL);
1680		list_del(&block_cb->list);
1681		flow_block_cb_free(block_cb);
1682		if (!bo->unlocked_driver_cb)
1683			block->lockeddevcnt--;
1684	}
1685}
1686
1687static int tcf_block_setup(struct tcf_block *block,
1688			   struct flow_block_offload *bo)
1689{
1690	int err;
1691
1692	switch (bo->command) {
1693	case FLOW_BLOCK_BIND:
1694		err = tcf_block_bind(block, bo);
1695		break;
1696	case FLOW_BLOCK_UNBIND:
1697		err = 0;
1698		tcf_block_unbind(block, bo);
1699		break;
1700	default:
1701		WARN_ON_ONCE(1);
1702		err = -EOPNOTSUPP;
1703	}
1704
1705	return err;
1706}
1707
1708/* Main classifier routine: scans classifier chain attached
1709 * to this qdisc, (optionally) tests for protocol and asks
1710 * specific classifiers.
1711 */
1712static inline int __tcf_classify(struct sk_buff *skb,
1713				 const struct tcf_proto *tp,
1714				 const struct tcf_proto *orig_tp,
1715				 struct tcf_result *res,
1716				 bool compat_mode,
1717				 struct tcf_exts_miss_cookie_node *n,
1718				 int act_index,
1719				 u32 *last_executed_chain)
1720{
1721#ifdef CONFIG_NET_CLS_ACT
1722	const int max_reclassify_loop = 16;
1723	const struct tcf_proto *first_tp;
1724	int limit = 0;
1725
1726reclassify:
1727#endif
1728	for (; tp; tp = rcu_dereference_bh(tp->next)) {
1729		__be16 protocol = skb_protocol(skb, false);
1730		int err = 0;
1731
1732		if (n) {
1733			struct tcf_exts *exts;
1734
1735			if (n->tp_prio != tp->prio)
1736				continue;
1737
1738			/* We re-lookup the tp and chain based on index instead
1739			 * of having hard refs and locks to them, so do a sanity
1740			 * check if any of tp,chain,exts was replaced by the
1741			 * time we got here with a cookie from hardware.
1742			 */
1743			if (unlikely(n->tp != tp || n->tp->chain != n->chain ||
1744				     !tp->ops->get_exts)) {
1745				tcf_set_drop_reason(skb,
1746						    SKB_DROP_REASON_TC_COOKIE_ERROR);
1747				return TC_ACT_SHOT;
1748			}
1749
1750			exts = tp->ops->get_exts(tp, n->handle);
1751			if (unlikely(!exts || n->exts != exts)) {
1752				tcf_set_drop_reason(skb,
1753						    SKB_DROP_REASON_TC_COOKIE_ERROR);
1754				return TC_ACT_SHOT;
1755			}
1756
1757			n = NULL;
1758			err = tcf_exts_exec_ex(skb, exts, act_index, res);
1759		} else {
1760			if (tp->protocol != protocol &&
1761			    tp->protocol != htons(ETH_P_ALL))
1762				continue;
1763
1764			err = tc_classify(skb, tp, res);
1765		}
1766#ifdef CONFIG_NET_CLS_ACT
1767		if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1768			first_tp = orig_tp;
1769			*last_executed_chain = first_tp->chain->index;
1770			goto reset;
1771		} else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1772			first_tp = res->goto_tp;
1773			*last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1774			goto reset;
1775		}
1776#endif
1777		if (err >= 0)
1778			return err;
1779	}
1780
1781	if (unlikely(n)) {
1782		tcf_set_drop_reason(skb,
1783				    SKB_DROP_REASON_TC_COOKIE_ERROR);
1784		return TC_ACT_SHOT;
1785	}
1786
1787	return TC_ACT_UNSPEC; /* signal: continue lookup */
1788#ifdef CONFIG_NET_CLS_ACT
1789reset:
1790	if (unlikely(limit++ >= max_reclassify_loop)) {
1791		net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1792				       tp->chain->block->index,
1793				       tp->prio & 0xffff,
1794				       ntohs(tp->protocol));
1795		tcf_set_drop_reason(skb,
1796				    SKB_DROP_REASON_TC_RECLASSIFY_LOOP);
1797		return TC_ACT_SHOT;
1798	}
1799
1800	tp = first_tp;
1801	goto reclassify;
1802#endif
1803}
1804
1805int tcf_classify(struct sk_buff *skb,
1806		 const struct tcf_block *block,
1807		 const struct tcf_proto *tp,
1808		 struct tcf_result *res, bool compat_mode)
1809{
1810#if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1811	u32 last_executed_chain = 0;
1812
1813	return __tcf_classify(skb, tp, tp, res, compat_mode, NULL, 0,
1814			      &last_executed_chain);
1815#else
1816	u32 last_executed_chain = tp ? tp->chain->index : 0;
1817	struct tcf_exts_miss_cookie_node *n = NULL;
1818	const struct tcf_proto *orig_tp = tp;
1819	struct tc_skb_ext *ext;
1820	int act_index = 0;
1821	int ret;
1822
1823	if (block) {
1824		ext = skb_ext_find(skb, TC_SKB_EXT);
1825
1826		if (ext && (ext->chain || ext->act_miss)) {
1827			struct tcf_chain *fchain;
1828			u32 chain;
1829
1830			if (ext->act_miss) {
1831				n = tcf_exts_miss_cookie_lookup(ext->act_miss_cookie,
1832								&act_index);
1833				if (!n) {
1834					tcf_set_drop_reason(skb,
1835							    SKB_DROP_REASON_TC_COOKIE_ERROR);
1836					return TC_ACT_SHOT;
1837				}
1838
1839				chain = n->chain_index;
1840			} else {
1841				chain = ext->chain;
1842			}
1843
1844			fchain = tcf_chain_lookup_rcu(block, chain);
1845			if (!fchain) {
1846				tcf_set_drop_reason(skb,
1847						    SKB_DROP_REASON_TC_CHAIN_NOTFOUND);
1848
1849				return TC_ACT_SHOT;
1850			}
1851
1852			/* Consume, so cloned/redirect skbs won't inherit ext */
1853			skb_ext_del(skb, TC_SKB_EXT);
1854
1855			tp = rcu_dereference_bh(fchain->filter_chain);
1856			last_executed_chain = fchain->index;
1857		}
1858	}
1859
1860	ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode, n, act_index,
1861			     &last_executed_chain);
1862
1863	if (tc_skb_ext_tc_enabled()) {
1864		/* If we missed on some chain */
1865		if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1866			struct tc_skb_cb *cb = tc_skb_cb(skb);
1867
1868			ext = tc_skb_ext_alloc(skb);
1869			if (WARN_ON_ONCE(!ext)) {
1870				tcf_set_drop_reason(skb, SKB_DROP_REASON_NOMEM);
1871				return TC_ACT_SHOT;
1872			}
1873			ext->chain = last_executed_chain;
1874			ext->mru = cb->mru;
1875			ext->post_ct = cb->post_ct;
1876			ext->post_ct_snat = cb->post_ct_snat;
1877			ext->post_ct_dnat = cb->post_ct_dnat;
1878			ext->zone = cb->zone;
1879		}
1880	}
1881
1882	return ret;
1883#endif
1884}
1885EXPORT_SYMBOL(tcf_classify);
1886
1887struct tcf_chain_info {
1888	struct tcf_proto __rcu **pprev;
1889	struct tcf_proto __rcu *next;
1890};
1891
1892static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1893					   struct tcf_chain_info *chain_info)
1894{
1895	return tcf_chain_dereference(*chain_info->pprev, chain);
1896}
1897
1898static int tcf_chain_tp_insert(struct tcf_chain *chain,
1899			       struct tcf_chain_info *chain_info,
1900			       struct tcf_proto *tp)
1901{
1902	if (chain->flushing)
1903		return -EAGAIN;
1904
1905	RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1906	if (*chain_info->pprev == chain->filter_chain)
1907		tcf_chain0_head_change(chain, tp);
1908	tcf_proto_get(tp);
1909	rcu_assign_pointer(*chain_info->pprev, tp);
1910
1911	return 0;
1912}
1913
1914static void tcf_chain_tp_remove(struct tcf_chain *chain,
1915				struct tcf_chain_info *chain_info,
1916				struct tcf_proto *tp)
1917{
1918	struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1919
1920	tcf_proto_mark_delete(tp);
1921	if (tp == chain->filter_chain)
1922		tcf_chain0_head_change(chain, next);
1923	RCU_INIT_POINTER(*chain_info->pprev, next);
1924}
1925
1926static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1927					   struct tcf_chain_info *chain_info,
1928					   u32 protocol, u32 prio,
1929					   bool prio_allocate,
1930					   struct netlink_ext_ack *extack);
1931
1932/* Try to insert new proto.
1933 * If proto with specified priority already exists, free new proto
1934 * and return existing one.
1935 */
1936
1937static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1938						    struct tcf_proto *tp_new,
1939						    u32 protocol, u32 prio,
1940						    bool rtnl_held)
1941{
1942	struct tcf_chain_info chain_info;
1943	struct tcf_proto *tp;
1944	int err = 0;
1945
1946	mutex_lock(&chain->filter_chain_lock);
1947
1948	if (tcf_proto_exists_destroying(chain, tp_new)) {
1949		mutex_unlock(&chain->filter_chain_lock);
1950		tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1951		return ERR_PTR(-EAGAIN);
1952	}
1953
1954	tp = tcf_chain_tp_find(chain, &chain_info, protocol, prio, false, NULL);
 
1955	if (!tp)
1956		err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1957	mutex_unlock(&chain->filter_chain_lock);
1958
1959	if (tp) {
1960		tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1961		tp_new = tp;
1962	} else if (err) {
1963		tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1964		tp_new = ERR_PTR(err);
1965	}
1966
1967	return tp_new;
1968}
1969
1970static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1971				      struct tcf_proto *tp, bool rtnl_held,
1972				      struct netlink_ext_ack *extack)
1973{
1974	struct tcf_chain_info chain_info;
1975	struct tcf_proto *tp_iter;
1976	struct tcf_proto **pprev;
1977	struct tcf_proto *next;
1978
1979	mutex_lock(&chain->filter_chain_lock);
1980
1981	/* Atomically find and remove tp from chain. */
1982	for (pprev = &chain->filter_chain;
1983	     (tp_iter = tcf_chain_dereference(*pprev, chain));
1984	     pprev = &tp_iter->next) {
1985		if (tp_iter == tp) {
1986			chain_info.pprev = pprev;
1987			chain_info.next = tp_iter->next;
1988			WARN_ON(tp_iter->deleting);
1989			break;
1990		}
1991	}
1992	/* Verify that tp still exists and no new filters were inserted
1993	 * concurrently.
1994	 * Mark tp for deletion if it is empty.
1995	 */
1996	if (!tp_iter || !tcf_proto_check_delete(tp)) {
1997		mutex_unlock(&chain->filter_chain_lock);
1998		return;
1999	}
2000
2001	tcf_proto_signal_destroying(chain, tp);
2002	next = tcf_chain_dereference(chain_info.next, chain);
2003	if (tp == chain->filter_chain)
2004		tcf_chain0_head_change(chain, next);
2005	RCU_INIT_POINTER(*chain_info.pprev, next);
2006	mutex_unlock(&chain->filter_chain_lock);
2007
2008	tcf_proto_put(tp, rtnl_held, extack);
2009}
2010
2011static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
2012					   struct tcf_chain_info *chain_info,
2013					   u32 protocol, u32 prio,
2014					   bool prio_allocate,
2015					   struct netlink_ext_ack *extack)
2016{
2017	struct tcf_proto **pprev;
2018	struct tcf_proto *tp;
2019
2020	/* Check the chain for existence of proto-tcf with this priority */
2021	for (pprev = &chain->filter_chain;
2022	     (tp = tcf_chain_dereference(*pprev, chain));
2023	     pprev = &tp->next) {
2024		if (tp->prio >= prio) {
2025			if (tp->prio == prio) {
2026				if (prio_allocate) {
2027					NL_SET_ERR_MSG(extack, "Lowest ID from auto-alloc range already in use");
2028					return ERR_PTR(-ENOSPC);
2029				}
2030				if (tp->protocol != protocol && protocol) {
2031					NL_SET_ERR_MSG(extack, "Protocol mismatch for filter with specified priority");
2032					return ERR_PTR(-EINVAL);
2033				}
2034			} else {
2035				tp = NULL;
2036			}
2037			break;
2038		}
2039	}
2040	chain_info->pprev = pprev;
2041	if (tp) {
2042		chain_info->next = tp->next;
2043		tcf_proto_get(tp);
2044	} else {
2045		chain_info->next = NULL;
2046	}
2047	return tp;
2048}
2049
2050static int tcf_fill_node(struct net *net, struct sk_buff *skb,
2051			 struct tcf_proto *tp, struct tcf_block *block,
2052			 struct Qdisc *q, u32 parent, void *fh,
2053			 u32 portid, u32 seq, u16 flags, int event,
2054			 bool terse_dump, bool rtnl_held,
2055			 struct netlink_ext_ack *extack)
2056{
2057	struct tcmsg *tcm;
2058	struct nlmsghdr  *nlh;
2059	unsigned char *b = skb_tail_pointer(skb);
2060
2061	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2062	if (!nlh)
2063		goto out_nlmsg_trim;
2064	tcm = nlmsg_data(nlh);
2065	tcm->tcm_family = AF_UNSPEC;
2066	tcm->tcm__pad1 = 0;
2067	tcm->tcm__pad2 = 0;
2068	if (q) {
2069		tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
2070		tcm->tcm_parent = parent;
2071	} else {
2072		tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2073		tcm->tcm_block_index = block->index;
2074	}
2075	tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
2076	if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
2077		goto nla_put_failure;
2078	if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
2079		goto nla_put_failure;
2080	if (!fh) {
2081		tcm->tcm_handle = 0;
2082	} else if (terse_dump) {
2083		if (tp->ops->terse_dump) {
2084			if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
2085						rtnl_held) < 0)
2086				goto nla_put_failure;
2087		} else {
2088			goto cls_op_not_supp;
2089		}
2090	} else {
2091		if (tp->ops->dump &&
2092		    tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
2093			goto nla_put_failure;
2094	}
2095
2096	if (extack && extack->_msg &&
2097	    nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2098		goto nla_put_failure;
2099
2100	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2101
2102	return skb->len;
2103
2104out_nlmsg_trim:
2105nla_put_failure:
2106cls_op_not_supp:
2107	nlmsg_trim(skb, b);
2108	return -1;
2109}
2110
2111static int tfilter_notify(struct net *net, struct sk_buff *oskb,
2112			  struct nlmsghdr *n, struct tcf_proto *tp,
2113			  struct tcf_block *block, struct Qdisc *q,
2114			  u32 parent, void *fh, int event, bool unicast,
2115			  bool rtnl_held, struct netlink_ext_ack *extack)
2116{
2117	struct sk_buff *skb;
2118	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2119	int err = 0;
2120
2121	if (!unicast && !rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2122		return 0;
2123
2124	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2125	if (!skb)
2126		return -ENOBUFS;
2127
2128	if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2129			  n->nlmsg_seq, n->nlmsg_flags, event,
2130			  false, rtnl_held, extack) <= 0) {
2131		kfree_skb(skb);
2132		return -EINVAL;
2133	}
2134
2135	if (unicast)
2136		err = rtnl_unicast(skb, net, portid);
2137	else
2138		err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2139				     n->nlmsg_flags & NLM_F_ECHO);
2140	return err;
2141}
2142
2143static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
2144			      struct nlmsghdr *n, struct tcf_proto *tp,
2145			      struct tcf_block *block, struct Qdisc *q,
2146			      u32 parent, void *fh, bool *last, bool rtnl_held,
2147			      struct netlink_ext_ack *extack)
2148{
2149	struct sk_buff *skb;
2150	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2151	int err;
2152
2153	if (!rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2154		return tp->ops->delete(tp, fh, last, rtnl_held, extack);
2155
2156	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2157	if (!skb)
2158		return -ENOBUFS;
2159
2160	if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2161			  n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
2162			  false, rtnl_held, extack) <= 0) {
2163		NL_SET_ERR_MSG(extack, "Failed to build del event notification");
2164		kfree_skb(skb);
2165		return -EINVAL;
2166	}
2167
2168	err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
2169	if (err) {
2170		kfree_skb(skb);
2171		return err;
2172	}
2173
2174	err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2175			     n->nlmsg_flags & NLM_F_ECHO);
2176	if (err < 0)
2177		NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
2178
2179	return err;
2180}
2181
2182static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
2183				 struct tcf_block *block, struct Qdisc *q,
2184				 u32 parent, struct nlmsghdr *n,
2185				 struct tcf_chain *chain, int event,
2186				 struct netlink_ext_ack *extack)
2187{
2188	struct tcf_proto *tp;
2189
2190	for (tp = tcf_get_next_proto(chain, NULL);
2191	     tp; tp = tcf_get_next_proto(chain, tp))
2192		tfilter_notify(net, oskb, n, tp, block, q, parent, NULL,
2193			       event, false, true, extack);
2194}
2195
2196static void tfilter_put(struct tcf_proto *tp, void *fh)
2197{
2198	if (tp->ops->put && fh)
2199		tp->ops->put(tp, fh);
2200}
2201
2202static bool is_qdisc_ingress(__u32 classid)
2203{
2204	return (TC_H_MIN(classid) == TC_H_MIN(TC_H_MIN_INGRESS));
2205}
2206
2207static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2208			  struct netlink_ext_ack *extack)
2209{
2210	struct net *net = sock_net(skb->sk);
2211	struct nlattr *tca[TCA_MAX + 1];
2212	char name[IFNAMSIZ];
2213	struct tcmsg *t;
2214	u32 protocol;
2215	u32 prio;
2216	bool prio_allocate;
2217	u32 parent;
2218	u32 chain_index;
2219	struct Qdisc *q;
2220	struct tcf_chain_info chain_info;
2221	struct tcf_chain *chain;
2222	struct tcf_block *block;
2223	struct tcf_proto *tp;
2224	unsigned long cl;
2225	void *fh;
2226	int err;
2227	int tp_created;
2228	bool rtnl_held = false;
2229	u32 flags;
2230
2231replay:
2232	tp_created = 0;
2233
2234	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2235				     rtm_tca_policy, extack);
2236	if (err < 0)
2237		return err;
2238
2239	t = nlmsg_data(n);
2240	protocol = TC_H_MIN(t->tcm_info);
2241	prio = TC_H_MAJ(t->tcm_info);
2242	prio_allocate = false;
2243	parent = t->tcm_parent;
2244	tp = NULL;
2245	cl = 0;
2246	block = NULL;
2247	q = NULL;
2248	chain = NULL;
2249	flags = 0;
2250
2251	if (prio == 0) {
2252		/* If no priority is provided by the user,
2253		 * we allocate one.
2254		 */
2255		if (n->nlmsg_flags & NLM_F_CREATE) {
2256			prio = TC_H_MAKE(0x80000000U, 0U);
2257			prio_allocate = true;
2258		} else {
2259			NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2260			return -ENOENT;
2261		}
2262	}
2263
2264	/* Find head of filter chain. */
2265
2266	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2267	if (err)
2268		return err;
2269
2270	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2271		NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2272		err = -EINVAL;
2273		goto errout;
2274	}
2275
2276	/* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2277	 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2278	 * type is not specified, classifier is not unlocked.
2279	 */
2280	if (rtnl_held ||
2281	    (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2282	    !tcf_proto_is_unlocked(name)) {
2283		rtnl_held = true;
2284		rtnl_lock();
2285	}
2286
2287	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2288	if (err)
2289		goto errout;
2290
2291	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2292				 extack);
2293	if (IS_ERR(block)) {
2294		err = PTR_ERR(block);
2295		goto errout;
2296	}
2297	block->classid = parent;
2298
2299	chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
2300	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2301		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2302		err = -EINVAL;
2303		goto errout;
2304	}
2305	chain = tcf_chain_get(block, chain_index, true);
2306	if (!chain) {
2307		NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2308		err = -ENOMEM;
2309		goto errout;
2310	}
2311
2312	mutex_lock(&chain->filter_chain_lock);
2313	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2314			       prio, prio_allocate, extack);
2315	if (IS_ERR(tp)) {
 
2316		err = PTR_ERR(tp);
2317		goto errout_locked;
2318	}
2319
2320	if (tp == NULL) {
2321		struct tcf_proto *tp_new = NULL;
2322
2323		if (chain->flushing) {
2324			err = -EAGAIN;
2325			goto errout_locked;
2326		}
2327
2328		/* Proto-tcf does not exist, create new one */
2329
2330		if (tca[TCA_KIND] == NULL || !protocol) {
2331			NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2332			err = -EINVAL;
2333			goto errout_locked;
2334		}
2335
2336		if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2337			NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2338			err = -ENOENT;
2339			goto errout_locked;
2340		}
2341
2342		if (prio_allocate)
2343			prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2344							       &chain_info));
2345
2346		mutex_unlock(&chain->filter_chain_lock);
2347		tp_new = tcf_proto_create(name, protocol, prio, chain,
2348					  rtnl_held, extack);
2349		if (IS_ERR(tp_new)) {
2350			err = PTR_ERR(tp_new);
2351			goto errout_tp;
2352		}
2353
2354		tp_created = 1;
2355		tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2356						rtnl_held);
2357		if (IS_ERR(tp)) {
2358			err = PTR_ERR(tp);
2359			goto errout_tp;
2360		}
2361	} else {
2362		mutex_unlock(&chain->filter_chain_lock);
2363	}
2364
2365	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2366		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2367		err = -EINVAL;
2368		goto errout;
2369	}
2370
2371	fh = tp->ops->get(tp, t->tcm_handle);
2372
2373	if (!fh) {
2374		if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2375			NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2376			err = -ENOENT;
2377			goto errout;
2378		}
2379	} else if (n->nlmsg_flags & NLM_F_EXCL) {
2380		tfilter_put(tp, fh);
2381		NL_SET_ERR_MSG(extack, "Filter already exists");
2382		err = -EEXIST;
2383		goto errout;
2384	}
2385
2386	if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2387		tfilter_put(tp, fh);
2388		NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2389		err = -EINVAL;
2390		goto errout;
2391	}
2392
2393	if (!(n->nlmsg_flags & NLM_F_CREATE))
2394		flags |= TCA_ACT_FLAGS_REPLACE;
2395	if (!rtnl_held)
2396		flags |= TCA_ACT_FLAGS_NO_RTNL;
2397	if (is_qdisc_ingress(parent))
2398		flags |= TCA_ACT_FLAGS_AT_INGRESS;
2399	err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2400			      flags, extack);
2401	if (err == 0) {
2402		tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2403			       RTM_NEWTFILTER, false, rtnl_held, extack);
2404		tfilter_put(tp, fh);
2405		tcf_proto_count_usesw(tp, true);
2406		/* q pointer is NULL for shared blocks */
2407		if (q)
2408			q->flags &= ~TCQ_F_CAN_BYPASS;
2409	}
2410
2411errout:
2412	if (err && tp_created)
2413		tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2414errout_tp:
2415	if (chain) {
2416		if (tp && !IS_ERR(tp))
2417			tcf_proto_put(tp, rtnl_held, NULL);
2418		if (!tp_created)
2419			tcf_chain_put(chain);
2420	}
2421	tcf_block_release(q, block, rtnl_held);
2422
2423	if (rtnl_held)
2424		rtnl_unlock();
2425
2426	if (err == -EAGAIN) {
2427		/* Take rtnl lock in case EAGAIN is caused by concurrent flush
2428		 * of target chain.
2429		 */
2430		rtnl_held = true;
2431		/* Replay the request. */
2432		goto replay;
2433	}
2434	return err;
2435
2436errout_locked:
2437	mutex_unlock(&chain->filter_chain_lock);
2438	goto errout;
2439}
2440
2441static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2442			  struct netlink_ext_ack *extack)
2443{
2444	struct net *net = sock_net(skb->sk);
2445	struct nlattr *tca[TCA_MAX + 1];
2446	char name[IFNAMSIZ];
2447	struct tcmsg *t;
2448	u32 protocol;
2449	u32 prio;
2450	u32 parent;
2451	u32 chain_index;
2452	struct Qdisc *q = NULL;
2453	struct tcf_chain_info chain_info;
2454	struct tcf_chain *chain = NULL;
2455	struct tcf_block *block = NULL;
2456	struct tcf_proto *tp = NULL;
2457	unsigned long cl = 0;
2458	void *fh = NULL;
2459	int err;
2460	bool rtnl_held = false;
2461
2462	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2463				     rtm_tca_policy, extack);
2464	if (err < 0)
2465		return err;
2466
2467	t = nlmsg_data(n);
2468	protocol = TC_H_MIN(t->tcm_info);
2469	prio = TC_H_MAJ(t->tcm_info);
2470	parent = t->tcm_parent;
2471
2472	if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2473		NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2474		return -ENOENT;
2475	}
2476
2477	/* Find head of filter chain. */
2478
2479	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2480	if (err)
2481		return err;
2482
2483	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2484		NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2485		err = -EINVAL;
2486		goto errout;
2487	}
2488	/* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2489	 * found), qdisc is not unlocked, classifier type is not specified,
2490	 * classifier is not unlocked.
2491	 */
2492	if (!prio ||
2493	    (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2494	    !tcf_proto_is_unlocked(name)) {
2495		rtnl_held = true;
2496		rtnl_lock();
2497	}
2498
2499	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2500	if (err)
2501		goto errout;
2502
2503	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2504				 extack);
2505	if (IS_ERR(block)) {
2506		err = PTR_ERR(block);
2507		goto errout;
2508	}
2509
2510	chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
2511	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2512		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2513		err = -EINVAL;
2514		goto errout;
2515	}
2516	chain = tcf_chain_get(block, chain_index, false);
2517	if (!chain) {
2518		/* User requested flush on non-existent chain. Nothing to do,
2519		 * so just return success.
2520		 */
2521		if (prio == 0) {
2522			err = 0;
2523			goto errout;
2524		}
2525		NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2526		err = -ENOENT;
2527		goto errout;
2528	}
2529
2530	if (prio == 0) {
2531		tfilter_notify_chain(net, skb, block, q, parent, n,
2532				     chain, RTM_DELTFILTER, extack);
2533		tcf_chain_flush(chain, rtnl_held);
2534		err = 0;
2535		goto errout;
2536	}
2537
2538	mutex_lock(&chain->filter_chain_lock);
2539	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2540			       prio, false, extack);
2541	if (!tp) {
2542		err = -ENOENT;
2543		NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2544		goto errout_locked;
2545	} else if (IS_ERR(tp)) {
2546		err = PTR_ERR(tp);
2547		goto errout_locked;
2548	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2549		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2550		err = -EINVAL;
2551		goto errout_locked;
2552	} else if (t->tcm_handle == 0) {
2553		tcf_proto_signal_destroying(chain, tp);
2554		tcf_chain_tp_remove(chain, &chain_info, tp);
2555		mutex_unlock(&chain->filter_chain_lock);
2556
2557		tcf_proto_put(tp, rtnl_held, NULL);
2558		tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2559			       RTM_DELTFILTER, false, rtnl_held, extack);
2560		err = 0;
2561		goto errout;
2562	}
2563	mutex_unlock(&chain->filter_chain_lock);
2564
2565	fh = tp->ops->get(tp, t->tcm_handle);
2566
2567	if (!fh) {
2568		NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2569		err = -ENOENT;
2570	} else {
2571		bool last;
2572
2573		err = tfilter_del_notify(net, skb, n, tp, block, q, parent, fh,
2574					 &last, rtnl_held, extack);
2575
2576		if (err)
2577			goto errout;
2578		if (last)
2579			tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2580	}
2581
2582errout:
2583	if (chain) {
2584		if (tp && !IS_ERR(tp))
2585			tcf_proto_put(tp, rtnl_held, NULL);
2586		tcf_chain_put(chain);
2587	}
2588	tcf_block_release(q, block, rtnl_held);
2589
2590	if (rtnl_held)
2591		rtnl_unlock();
2592
2593	return err;
2594
2595errout_locked:
2596	mutex_unlock(&chain->filter_chain_lock);
2597	goto errout;
2598}
2599
2600static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2601			  struct netlink_ext_ack *extack)
2602{
2603	struct net *net = sock_net(skb->sk);
2604	struct nlattr *tca[TCA_MAX + 1];
2605	char name[IFNAMSIZ];
2606	struct tcmsg *t;
2607	u32 protocol;
2608	u32 prio;
2609	u32 parent;
2610	u32 chain_index;
2611	struct Qdisc *q = NULL;
2612	struct tcf_chain_info chain_info;
2613	struct tcf_chain *chain = NULL;
2614	struct tcf_block *block = NULL;
2615	struct tcf_proto *tp = NULL;
2616	unsigned long cl = 0;
2617	void *fh = NULL;
2618	int err;
2619	bool rtnl_held = false;
2620
2621	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2622				     rtm_tca_policy, extack);
2623	if (err < 0)
2624		return err;
2625
2626	t = nlmsg_data(n);
2627	protocol = TC_H_MIN(t->tcm_info);
2628	prio = TC_H_MAJ(t->tcm_info);
2629	parent = t->tcm_parent;
2630
2631	if (prio == 0) {
2632		NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2633		return -ENOENT;
2634	}
2635
2636	/* Find head of filter chain. */
2637
2638	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2639	if (err)
2640		return err;
2641
2642	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2643		NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2644		err = -EINVAL;
2645		goto errout;
2646	}
2647	/* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2648	 * unlocked, classifier type is not specified, classifier is not
2649	 * unlocked.
2650	 */
2651	if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2652	    !tcf_proto_is_unlocked(name)) {
2653		rtnl_held = true;
2654		rtnl_lock();
2655	}
2656
2657	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2658	if (err)
2659		goto errout;
2660
2661	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2662				 extack);
2663	if (IS_ERR(block)) {
2664		err = PTR_ERR(block);
2665		goto errout;
2666	}
2667
2668	chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
2669	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2670		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2671		err = -EINVAL;
2672		goto errout;
2673	}
2674	chain = tcf_chain_get(block, chain_index, false);
2675	if (!chain) {
2676		NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2677		err = -EINVAL;
2678		goto errout;
2679	}
2680
2681	mutex_lock(&chain->filter_chain_lock);
2682	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2683			       prio, false, extack);
2684	mutex_unlock(&chain->filter_chain_lock);
2685	if (!tp) {
2686		err = -ENOENT;
2687		NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2688		goto errout;
2689	} else if (IS_ERR(tp)) {
2690		err = PTR_ERR(tp);
2691		goto errout;
2692	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2693		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2694		err = -EINVAL;
2695		goto errout;
2696	}
2697
2698	fh = tp->ops->get(tp, t->tcm_handle);
2699
2700	if (!fh) {
2701		NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2702		err = -ENOENT;
2703	} else {
2704		err = tfilter_notify(net, skb, n, tp, block, q, parent,
2705				     fh, RTM_NEWTFILTER, true, rtnl_held, NULL);
2706		if (err < 0)
2707			NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2708	}
2709
2710	tfilter_put(tp, fh);
2711errout:
2712	if (chain) {
2713		if (tp && !IS_ERR(tp))
2714			tcf_proto_put(tp, rtnl_held, NULL);
2715		tcf_chain_put(chain);
2716	}
2717	tcf_block_release(q, block, rtnl_held);
2718
2719	if (rtnl_held)
2720		rtnl_unlock();
2721
2722	return err;
2723}
2724
2725struct tcf_dump_args {
2726	struct tcf_walker w;
2727	struct sk_buff *skb;
2728	struct netlink_callback *cb;
2729	struct tcf_block *block;
2730	struct Qdisc *q;
2731	u32 parent;
2732	bool terse_dump;
2733};
2734
2735static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2736{
2737	struct tcf_dump_args *a = (void *)arg;
2738	struct net *net = sock_net(a->skb->sk);
2739
2740	return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2741			     n, NETLINK_CB(a->cb->skb).portid,
2742			     a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2743			     RTM_NEWTFILTER, a->terse_dump, true, NULL);
2744}
2745
2746static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2747			   struct sk_buff *skb, struct netlink_callback *cb,
2748			   long index_start, long *p_index, bool terse)
2749{
2750	struct net *net = sock_net(skb->sk);
2751	struct tcf_block *block = chain->block;
2752	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2753	struct tcf_proto *tp, *tp_prev;
2754	struct tcf_dump_args arg;
2755
2756	for (tp = __tcf_get_next_proto(chain, NULL);
2757	     tp;
2758	     tp_prev = tp,
2759		     tp = __tcf_get_next_proto(chain, tp),
2760		     tcf_proto_put(tp_prev, true, NULL),
2761		     (*p_index)++) {
2762		if (*p_index < index_start)
2763			continue;
2764		if (TC_H_MAJ(tcm->tcm_info) &&
2765		    TC_H_MAJ(tcm->tcm_info) != tp->prio)
2766			continue;
2767		if (TC_H_MIN(tcm->tcm_info) &&
2768		    TC_H_MIN(tcm->tcm_info) != tp->protocol)
2769			continue;
2770		if (*p_index > index_start)
2771			memset(&cb->args[1], 0,
2772			       sizeof(cb->args) - sizeof(cb->args[0]));
2773		if (cb->args[1] == 0) {
2774			if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2775					  NETLINK_CB(cb->skb).portid,
2776					  cb->nlh->nlmsg_seq, NLM_F_MULTI,
2777					  RTM_NEWTFILTER, false, true, NULL) <= 0)
2778				goto errout;
2779			cb->args[1] = 1;
2780		}
2781		if (!tp->ops->walk)
2782			continue;
2783		arg.w.fn = tcf_node_dump;
2784		arg.skb = skb;
2785		arg.cb = cb;
2786		arg.block = block;
2787		arg.q = q;
2788		arg.parent = parent;
2789		arg.w.stop = 0;
2790		arg.w.skip = cb->args[1] - 1;
2791		arg.w.count = 0;
2792		arg.w.cookie = cb->args[2];
2793		arg.terse_dump = terse;
2794		tp->ops->walk(tp, &arg.w, true);
2795		cb->args[2] = arg.w.cookie;
2796		cb->args[1] = arg.w.count + 1;
2797		if (arg.w.stop)
2798			goto errout;
2799	}
2800	return true;
2801
2802errout:
2803	tcf_proto_put(tp, true, NULL);
2804	return false;
2805}
2806
2807static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2808	[TCA_CHAIN]      = { .type = NLA_U32 },
2809	[TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2810};
2811
2812/* called with RTNL */
2813static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2814{
2815	struct tcf_chain *chain, *chain_prev;
2816	struct net *net = sock_net(skb->sk);
2817	struct nlattr *tca[TCA_MAX + 1];
2818	struct Qdisc *q = NULL;
2819	struct tcf_block *block;
2820	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2821	bool terse_dump = false;
2822	long index_start;
2823	long index;
2824	u32 parent;
2825	int err;
2826
2827	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2828		return skb->len;
2829
2830	err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2831				     tcf_tfilter_dump_policy, cb->extack);
2832	if (err)
2833		return err;
2834
2835	if (tca[TCA_DUMP_FLAGS]) {
2836		struct nla_bitfield32 flags =
2837			nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2838
2839		terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2840	}
2841
2842	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2843		block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2844		if (!block)
2845			goto out;
2846		/* If we work with block index, q is NULL and parent value
2847		 * will never be used in the following code. The check
2848		 * in tcf_fill_node prevents it. However, compiler does not
2849		 * see that far, so set parent to zero to silence the warning
2850		 * about parent being uninitialized.
2851		 */
2852		parent = 0;
2853	} else {
2854		const struct Qdisc_class_ops *cops;
2855		struct net_device *dev;
2856		unsigned long cl = 0;
2857
2858		dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2859		if (!dev)
2860			return skb->len;
2861
2862		parent = tcm->tcm_parent;
2863		if (!parent)
2864			q = rtnl_dereference(dev->qdisc);
2865		else
2866			q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2867		if (!q)
2868			goto out;
2869		cops = q->ops->cl_ops;
2870		if (!cops)
2871			goto out;
2872		if (!cops->tcf_block)
2873			goto out;
2874		if (TC_H_MIN(tcm->tcm_parent)) {
2875			cl = cops->find(q, tcm->tcm_parent);
2876			if (cl == 0)
2877				goto out;
2878		}
2879		block = cops->tcf_block(q, cl, NULL);
2880		if (!block)
2881			goto out;
2882		parent = block->classid;
2883		if (tcf_block_shared(block))
2884			q = NULL;
2885	}
2886
2887	index_start = cb->args[0];
2888	index = 0;
2889
2890	for (chain = __tcf_get_next_chain(block, NULL);
2891	     chain;
2892	     chain_prev = chain,
2893		     chain = __tcf_get_next_chain(block, chain),
2894		     tcf_chain_put(chain_prev)) {
2895		if (tca[TCA_CHAIN] &&
2896		    nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2897			continue;
2898		if (!tcf_chain_dump(chain, q, parent, skb, cb,
2899				    index_start, &index, terse_dump)) {
2900			tcf_chain_put(chain);
2901			err = -EMSGSIZE;
2902			break;
2903		}
2904	}
2905
2906	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2907		tcf_block_refcnt_put(block, true);
2908	cb->args[0] = index;
2909
2910out:
2911	/* If we did no progress, the error (EMSGSIZE) is real */
2912	if (skb->len == 0 && err)
2913		return err;
2914	return skb->len;
2915}
2916
2917static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2918			      void *tmplt_priv, u32 chain_index,
2919			      struct net *net, struct sk_buff *skb,
2920			      struct tcf_block *block,
2921			      u32 portid, u32 seq, u16 flags, int event,
2922			      struct netlink_ext_ack *extack)
2923{
2924	unsigned char *b = skb_tail_pointer(skb);
2925	const struct tcf_proto_ops *ops;
2926	struct nlmsghdr *nlh;
2927	struct tcmsg *tcm;
2928	void *priv;
2929
2930	ops = tmplt_ops;
2931	priv = tmplt_priv;
2932
2933	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2934	if (!nlh)
2935		goto out_nlmsg_trim;
2936	tcm = nlmsg_data(nlh);
2937	tcm->tcm_family = AF_UNSPEC;
2938	tcm->tcm__pad1 = 0;
2939	tcm->tcm__pad2 = 0;
2940	tcm->tcm_handle = 0;
2941	if (block->q) {
2942		tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2943		tcm->tcm_parent = block->q->handle;
2944	} else {
2945		tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2946		tcm->tcm_block_index = block->index;
2947	}
2948
2949	if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2950		goto nla_put_failure;
2951
2952	if (ops) {
2953		if (nla_put_string(skb, TCA_KIND, ops->kind))
2954			goto nla_put_failure;
2955		if (ops->tmplt_dump(skb, net, priv) < 0)
2956			goto nla_put_failure;
2957	}
2958
2959	if (extack && extack->_msg &&
2960	    nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2961		goto out_nlmsg_trim;
2962
2963	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2964
2965	return skb->len;
2966
2967out_nlmsg_trim:
2968nla_put_failure:
2969	nlmsg_trim(skb, b);
2970	return -EMSGSIZE;
2971}
2972
2973static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2974			   u32 seq, u16 flags, int event, bool unicast,
2975			   struct netlink_ext_ack *extack)
2976{
2977	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2978	struct tcf_block *block = chain->block;
2979	struct net *net = block->net;
2980	struct sk_buff *skb;
2981	int err = 0;
2982
2983	if (!unicast && !rtnl_notify_needed(net, flags, RTNLGRP_TC))
2984		return 0;
2985
2986	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2987	if (!skb)
2988		return -ENOBUFS;
2989
2990	if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2991			       chain->index, net, skb, block, portid,
2992			       seq, flags, event, extack) <= 0) {
2993		kfree_skb(skb);
2994		return -EINVAL;
2995	}
2996
2997	if (unicast)
2998		err = rtnl_unicast(skb, net, portid);
2999	else
3000		err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
3001				     flags & NLM_F_ECHO);
3002
3003	return err;
3004}
3005
3006static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
3007				  void *tmplt_priv, u32 chain_index,
3008				  struct tcf_block *block, struct sk_buff *oskb,
3009				  u32 seq, u16 flags)
3010{
3011	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
3012	struct net *net = block->net;
3013	struct sk_buff *skb;
3014
3015	if (!rtnl_notify_needed(net, flags, RTNLGRP_TC))
3016		return 0;
3017
3018	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3019	if (!skb)
3020		return -ENOBUFS;
3021
3022	if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
3023			       block, portid, seq, flags, RTM_DELCHAIN, NULL) <= 0) {
3024		kfree_skb(skb);
3025		return -EINVAL;
3026	}
3027
3028	return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
3029}
3030
3031static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
3032			      struct nlattr **tca,
3033			      struct netlink_ext_ack *extack)
3034{
3035	const struct tcf_proto_ops *ops;
3036	char name[IFNAMSIZ];
3037	void *tmplt_priv;
3038
3039	/* If kind is not set, user did not specify template. */
3040	if (!tca[TCA_KIND])
3041		return 0;
3042
3043	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
3044		NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
3045		return -EINVAL;
3046	}
3047
3048	ops = tcf_proto_lookup_ops(name, true, extack);
3049	if (IS_ERR(ops))
3050		return PTR_ERR(ops);
3051	if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump ||
3052	    !ops->tmplt_reoffload) {
3053		NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
3054		module_put(ops->owner);
3055		return -EOPNOTSUPP;
3056	}
3057
3058	tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
3059	if (IS_ERR(tmplt_priv)) {
3060		module_put(ops->owner);
3061		return PTR_ERR(tmplt_priv);
3062	}
3063	chain->tmplt_ops = ops;
3064	chain->tmplt_priv = tmplt_priv;
3065	return 0;
3066}
3067
3068static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
3069			       void *tmplt_priv)
3070{
3071	/* If template ops are set, no work to do for us. */
3072	if (!tmplt_ops)
3073		return;
3074
3075	tmplt_ops->tmplt_destroy(tmplt_priv);
3076	module_put(tmplt_ops->owner);
3077}
3078
3079/* Add/delete/get a chain */
3080
3081static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
3082			struct netlink_ext_ack *extack)
3083{
3084	struct net *net = sock_net(skb->sk);
3085	struct nlattr *tca[TCA_MAX + 1];
3086	struct tcmsg *t;
3087	u32 parent;
3088	u32 chain_index;
3089	struct Qdisc *q;
3090	struct tcf_chain *chain;
3091	struct tcf_block *block;
3092	unsigned long cl;
3093	int err;
3094
3095replay:
3096	q = NULL;
3097	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
3098				     rtm_tca_policy, extack);
3099	if (err < 0)
3100		return err;
3101
3102	t = nlmsg_data(n);
3103	parent = t->tcm_parent;
3104	cl = 0;
3105
3106	block = tcf_block_find(net, &q, &parent, &cl,
3107			       t->tcm_ifindex, t->tcm_block_index, extack);
3108	if (IS_ERR(block))
3109		return PTR_ERR(block);
3110
3111	chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
3112	if (chain_index > TC_ACT_EXT_VAL_MASK) {
3113		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
3114		err = -EINVAL;
3115		goto errout_block;
3116	}
3117
3118	mutex_lock(&block->lock);
3119	chain = tcf_chain_lookup(block, chain_index);
3120	if (n->nlmsg_type == RTM_NEWCHAIN) {
3121		if (chain) {
3122			if (tcf_chain_held_by_acts_only(chain)) {
3123				/* The chain exists only because there is
3124				 * some action referencing it.
3125				 */
3126				tcf_chain_hold(chain);
3127			} else {
3128				NL_SET_ERR_MSG(extack, "Filter chain already exists");
3129				err = -EEXIST;
3130				goto errout_block_locked;
3131			}
3132		} else {
3133			if (!(n->nlmsg_flags & NLM_F_CREATE)) {
3134				NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
3135				err = -ENOENT;
3136				goto errout_block_locked;
3137			}
3138			chain = tcf_chain_create(block, chain_index);
3139			if (!chain) {
3140				NL_SET_ERR_MSG(extack, "Failed to create filter chain");
3141				err = -ENOMEM;
3142				goto errout_block_locked;
3143			}
3144		}
3145	} else {
3146		if (!chain || tcf_chain_held_by_acts_only(chain)) {
3147			NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
3148			err = -EINVAL;
3149			goto errout_block_locked;
3150		}
3151		tcf_chain_hold(chain);
3152	}
3153
3154	if (n->nlmsg_type == RTM_NEWCHAIN) {
3155		/* Modifying chain requires holding parent block lock. In case
3156		 * the chain was successfully added, take a reference to the
3157		 * chain. This ensures that an empty chain does not disappear at
3158		 * the end of this function.
3159		 */
3160		tcf_chain_hold(chain);
3161		chain->explicitly_created = true;
3162	}
3163	mutex_unlock(&block->lock);
3164
3165	switch (n->nlmsg_type) {
3166	case RTM_NEWCHAIN:
3167		err = tc_chain_tmplt_add(chain, net, tca, extack);
3168		if (err) {
3169			tcf_chain_put_explicitly_created(chain);
3170			goto errout;
3171		}
3172
3173		tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
3174				RTM_NEWCHAIN, false, extack);
3175		break;
3176	case RTM_DELCHAIN:
3177		tfilter_notify_chain(net, skb, block, q, parent, n,
3178				     chain, RTM_DELTFILTER, extack);
3179		/* Flush the chain first as the user requested chain removal. */
3180		tcf_chain_flush(chain, true);
3181		/* In case the chain was successfully deleted, put a reference
3182		 * to the chain previously taken during addition.
3183		 */
3184		tcf_chain_put_explicitly_created(chain);
3185		break;
3186	case RTM_GETCHAIN:
3187		err = tc_chain_notify(chain, skb, n->nlmsg_seq,
3188				      n->nlmsg_flags, n->nlmsg_type, true, extack);
3189		if (err < 0)
3190			NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
3191		break;
3192	default:
3193		err = -EOPNOTSUPP;
3194		NL_SET_ERR_MSG(extack, "Unsupported message type");
3195		goto errout;
3196	}
3197
3198errout:
3199	tcf_chain_put(chain);
3200errout_block:
3201	tcf_block_release(q, block, true);
3202	if (err == -EAGAIN)
3203		/* Replay the request. */
3204		goto replay;
3205	return err;
3206
3207errout_block_locked:
3208	mutex_unlock(&block->lock);
3209	goto errout_block;
3210}
3211
3212/* called with RTNL */
3213static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
3214{
3215	struct net *net = sock_net(skb->sk);
3216	struct nlattr *tca[TCA_MAX + 1];
3217	struct Qdisc *q = NULL;
3218	struct tcf_block *block;
3219	struct tcmsg *tcm = nlmsg_data(cb->nlh);
3220	struct tcf_chain *chain;
3221	long index_start;
3222	long index;
3223	int err;
3224
3225	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
3226		return skb->len;
3227
3228	err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
3229				     rtm_tca_policy, cb->extack);
3230	if (err)
3231		return err;
3232
3233	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
3234		block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
3235		if (!block)
3236			goto out;
3237	} else {
3238		const struct Qdisc_class_ops *cops;
3239		struct net_device *dev;
3240		unsigned long cl = 0;
3241
3242		dev = __dev_get_by_index(net, tcm->tcm_ifindex);
3243		if (!dev)
3244			return skb->len;
3245
3246		if (!tcm->tcm_parent)
3247			q = rtnl_dereference(dev->qdisc);
3248		else
3249			q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
3250
3251		if (!q)
3252			goto out;
3253		cops = q->ops->cl_ops;
3254		if (!cops)
3255			goto out;
3256		if (!cops->tcf_block)
3257			goto out;
3258		if (TC_H_MIN(tcm->tcm_parent)) {
3259			cl = cops->find(q, tcm->tcm_parent);
3260			if (cl == 0)
3261				goto out;
3262		}
3263		block = cops->tcf_block(q, cl, NULL);
3264		if (!block)
3265			goto out;
3266		if (tcf_block_shared(block))
3267			q = NULL;
3268	}
3269
3270	index_start = cb->args[0];
3271	index = 0;
3272
3273	mutex_lock(&block->lock);
3274	list_for_each_entry(chain, &block->chain_list, list) {
3275		if ((tca[TCA_CHAIN] &&
3276		     nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3277			continue;
3278		if (index < index_start) {
3279			index++;
3280			continue;
3281		}
3282		if (tcf_chain_held_by_acts_only(chain))
3283			continue;
3284		err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3285					 chain->index, net, skb, block,
3286					 NETLINK_CB(cb->skb).portid,
3287					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3288					 RTM_NEWCHAIN, NULL);
3289		if (err <= 0)
3290			break;
3291		index++;
3292	}
3293	mutex_unlock(&block->lock);
3294
3295	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3296		tcf_block_refcnt_put(block, true);
3297	cb->args[0] = index;
3298
3299out:
3300	/* If we did no progress, the error (EMSGSIZE) is real */
3301	if (skb->len == 0 && err)
3302		return err;
3303	return skb->len;
3304}
3305
3306int tcf_exts_init_ex(struct tcf_exts *exts, struct net *net, int action,
3307		     int police, struct tcf_proto *tp, u32 handle,
3308		     bool use_action_miss)
3309{
3310	int err = 0;
3311
3312#ifdef CONFIG_NET_CLS_ACT
3313	exts->type = 0;
3314	exts->nr_actions = 0;
3315	exts->miss_cookie_node = NULL;
3316	/* Note: we do not own yet a reference on net.
3317	 * This reference might be taken later from tcf_exts_get_net().
3318	 */
3319	exts->net = net;
3320	exts->actions = kcalloc(TCA_ACT_MAX_PRIO, sizeof(struct tc_action *),
3321				GFP_KERNEL);
3322	if (!exts->actions)
3323		return -ENOMEM;
3324#endif
3325
3326	exts->action = action;
3327	exts->police = police;
3328
3329	if (!use_action_miss)
3330		return 0;
3331
3332	err = tcf_exts_miss_cookie_base_alloc(exts, tp, handle);
3333	if (err)
3334		goto err_miss_alloc;
3335
3336	return 0;
3337
3338err_miss_alloc:
3339	tcf_exts_destroy(exts);
3340#ifdef CONFIG_NET_CLS_ACT
3341	exts->actions = NULL;
3342#endif
3343	return err;
3344}
3345EXPORT_SYMBOL(tcf_exts_init_ex);
3346
3347void tcf_exts_destroy(struct tcf_exts *exts)
3348{
3349	tcf_exts_miss_cookie_base_destroy(exts);
3350
3351#ifdef CONFIG_NET_CLS_ACT
3352	if (exts->actions) {
3353		tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3354		kfree(exts->actions);
3355	}
3356	exts->nr_actions = 0;
3357#endif
3358}
3359EXPORT_SYMBOL(tcf_exts_destroy);
3360
3361int tcf_exts_validate_ex(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3362			 struct nlattr *rate_tlv, struct tcf_exts *exts,
3363			 u32 flags, u32 fl_flags, struct netlink_ext_ack *extack)
3364{
3365#ifdef CONFIG_NET_CLS_ACT
3366	{
3367		int init_res[TCA_ACT_MAX_PRIO] = {};
3368		struct tc_action *act;
3369		size_t attr_size = 0;
3370
3371		if (exts->police && tb[exts->police]) {
3372			struct tc_action_ops *a_o;
3373
3374			flags |= TCA_ACT_FLAGS_POLICE | TCA_ACT_FLAGS_BIND;
3375			a_o = tc_action_load_ops(tb[exts->police], flags,
3376						 extack);
3377			if (IS_ERR(a_o))
3378				return PTR_ERR(a_o);
3379			act = tcf_action_init_1(net, tp, tb[exts->police],
3380						rate_tlv, a_o, init_res, flags,
3381						extack);
3382			module_put(a_o->owner);
3383			if (IS_ERR(act))
3384				return PTR_ERR(act);
3385
3386			act->type = exts->type = TCA_OLD_COMPAT;
3387			exts->actions[0] = act;
3388			exts->nr_actions = 1;
3389			tcf_idr_insert_many(exts->actions, init_res);
3390		} else if (exts->action && tb[exts->action]) {
3391			int err;
3392
3393			flags |= TCA_ACT_FLAGS_BIND;
3394			err = tcf_action_init(net, tp, tb[exts->action],
3395					      rate_tlv, exts->actions, init_res,
3396					      &attr_size, flags, fl_flags,
3397					      extack);
3398			if (err < 0)
3399				return err;
3400			exts->nr_actions = err;
3401		}
3402	}
3403#else
3404	if ((exts->action && tb[exts->action]) ||
3405	    (exts->police && tb[exts->police])) {
3406		NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3407		return -EOPNOTSUPP;
3408	}
3409#endif
3410
3411	return 0;
3412}
3413EXPORT_SYMBOL(tcf_exts_validate_ex);
3414
3415int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3416		      struct nlattr *rate_tlv, struct tcf_exts *exts,
3417		      u32 flags, struct netlink_ext_ack *extack)
3418{
3419	return tcf_exts_validate_ex(net, tp, tb, rate_tlv, exts,
3420				    flags, 0, extack);
3421}
3422EXPORT_SYMBOL(tcf_exts_validate);
3423
3424void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3425{
3426#ifdef CONFIG_NET_CLS_ACT
3427	struct tcf_exts old = *dst;
3428
3429	*dst = *src;
3430	tcf_exts_destroy(&old);
3431#endif
3432}
3433EXPORT_SYMBOL(tcf_exts_change);
3434
3435#ifdef CONFIG_NET_CLS_ACT
3436static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3437{
3438	if (exts->nr_actions == 0)
3439		return NULL;
3440	else
3441		return exts->actions[0];
3442}
3443#endif
3444
3445int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3446{
3447#ifdef CONFIG_NET_CLS_ACT
3448	struct nlattr *nest;
3449
3450	if (exts->action && tcf_exts_has_actions(exts)) {
3451		/*
3452		 * again for backward compatible mode - we want
3453		 * to work with both old and new modes of entering
3454		 * tc data even if iproute2  was newer - jhs
3455		 */
3456		if (exts->type != TCA_OLD_COMPAT) {
3457			nest = nla_nest_start_noflag(skb, exts->action);
3458			if (nest == NULL)
3459				goto nla_put_failure;
3460
3461			if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3462			    < 0)
3463				goto nla_put_failure;
3464			nla_nest_end(skb, nest);
3465		} else if (exts->police) {
3466			struct tc_action *act = tcf_exts_first_act(exts);
3467			nest = nla_nest_start_noflag(skb, exts->police);
3468			if (nest == NULL || !act)
3469				goto nla_put_failure;
3470			if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3471				goto nla_put_failure;
3472			nla_nest_end(skb, nest);
3473		}
3474	}
3475	return 0;
3476
3477nla_put_failure:
3478	nla_nest_cancel(skb, nest);
3479	return -1;
3480#else
3481	return 0;
3482#endif
3483}
3484EXPORT_SYMBOL(tcf_exts_dump);
3485
3486int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3487{
3488#ifdef CONFIG_NET_CLS_ACT
3489	struct nlattr *nest;
3490
3491	if (!exts->action || !tcf_exts_has_actions(exts))
3492		return 0;
3493
3494	nest = nla_nest_start_noflag(skb, exts->action);
3495	if (!nest)
3496		goto nla_put_failure;
3497
3498	if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3499		goto nla_put_failure;
3500	nla_nest_end(skb, nest);
3501	return 0;
3502
3503nla_put_failure:
3504	nla_nest_cancel(skb, nest);
3505	return -1;
3506#else
3507	return 0;
3508#endif
3509}
3510EXPORT_SYMBOL(tcf_exts_terse_dump);
3511
3512int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3513{
3514#ifdef CONFIG_NET_CLS_ACT
3515	struct tc_action *a = tcf_exts_first_act(exts);
3516	if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3517		return -1;
3518#endif
3519	return 0;
3520}
3521EXPORT_SYMBOL(tcf_exts_dump_stats);
3522
3523static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3524{
3525	if (*flags & TCA_CLS_FLAGS_IN_HW)
3526		return;
3527	*flags |= TCA_CLS_FLAGS_IN_HW;
3528	atomic_inc(&block->offloadcnt);
3529}
3530
3531static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3532{
3533	if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3534		return;
3535	*flags &= ~TCA_CLS_FLAGS_IN_HW;
3536	atomic_dec(&block->offloadcnt);
3537}
3538
3539static void tc_cls_offload_cnt_update(struct tcf_block *block,
3540				      struct tcf_proto *tp, u32 *cnt,
3541				      u32 *flags, u32 diff, bool add)
3542{
3543	lockdep_assert_held(&block->cb_lock);
3544
3545	spin_lock(&tp->lock);
3546	if (add) {
3547		if (!*cnt)
3548			tcf_block_offload_inc(block, flags);
3549		*cnt += diff;
3550	} else {
3551		*cnt -= diff;
3552		if (!*cnt)
3553			tcf_block_offload_dec(block, flags);
3554	}
3555	spin_unlock(&tp->lock);
3556}
3557
3558static void
3559tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3560			 u32 *cnt, u32 *flags)
3561{
3562	lockdep_assert_held(&block->cb_lock);
3563
3564	spin_lock(&tp->lock);
3565	tcf_block_offload_dec(block, flags);
3566	*cnt = 0;
3567	spin_unlock(&tp->lock);
3568}
3569
3570static int
3571__tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3572		   void *type_data, bool err_stop)
3573{
3574	struct flow_block_cb *block_cb;
3575	int ok_count = 0;
3576	int err;
3577
3578	list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3579		err = block_cb->cb(type, type_data, block_cb->cb_priv);
3580		if (err) {
3581			if (err_stop)
3582				return err;
3583		} else {
3584			ok_count++;
3585		}
3586	}
3587	return ok_count;
3588}
3589
3590int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3591		     void *type_data, bool err_stop, bool rtnl_held)
3592{
3593	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3594	int ok_count;
3595
3596retry:
3597	if (take_rtnl)
3598		rtnl_lock();
3599	down_read(&block->cb_lock);
3600	/* Need to obtain rtnl lock if block is bound to devs that require it.
3601	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3602	 * obtain the locks in same order here.
3603	 */
3604	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3605		up_read(&block->cb_lock);
3606		take_rtnl = true;
3607		goto retry;
3608	}
3609
3610	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3611
3612	up_read(&block->cb_lock);
3613	if (take_rtnl)
3614		rtnl_unlock();
3615	return ok_count;
3616}
3617EXPORT_SYMBOL(tc_setup_cb_call);
3618
3619/* Non-destructive filter add. If filter that wasn't already in hardware is
3620 * successfully offloaded, increment block offloads counter. On failure,
3621 * previously offloaded filter is considered to be intact and offloads counter
3622 * is not decremented.
3623 */
3624
3625int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3626		    enum tc_setup_type type, void *type_data, bool err_stop,
3627		    u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3628{
3629	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3630	int ok_count;
3631
3632retry:
3633	if (take_rtnl)
3634		rtnl_lock();
3635	down_read(&block->cb_lock);
3636	/* Need to obtain rtnl lock if block is bound to devs that require it.
3637	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3638	 * obtain the locks in same order here.
3639	 */
3640	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3641		up_read(&block->cb_lock);
3642		take_rtnl = true;
3643		goto retry;
3644	}
3645
3646	/* Make sure all netdevs sharing this block are offload-capable. */
3647	if (block->nooffloaddevcnt && err_stop) {
3648		ok_count = -EOPNOTSUPP;
3649		goto err_unlock;
3650	}
3651
3652	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3653	if (ok_count < 0)
3654		goto err_unlock;
3655
3656	if (tp->ops->hw_add)
3657		tp->ops->hw_add(tp, type_data);
3658	if (ok_count > 0)
3659		tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3660					  ok_count, true);
3661err_unlock:
3662	up_read(&block->cb_lock);
3663	if (take_rtnl)
3664		rtnl_unlock();
3665	return min(ok_count, 0);
3666}
3667EXPORT_SYMBOL(tc_setup_cb_add);
3668
3669/* Destructive filter replace. If filter that wasn't already in hardware is
3670 * successfully offloaded, increment block offload counter. On failure,
3671 * previously offloaded filter is considered to be destroyed and offload counter
3672 * is decremented.
3673 */
3674
3675int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3676			enum tc_setup_type type, void *type_data, bool err_stop,
3677			u32 *old_flags, unsigned int *old_in_hw_count,
3678			u32 *new_flags, unsigned int *new_in_hw_count,
3679			bool rtnl_held)
3680{
3681	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3682	int ok_count;
3683
3684retry:
3685	if (take_rtnl)
3686		rtnl_lock();
3687	down_read(&block->cb_lock);
3688	/* Need to obtain rtnl lock if block is bound to devs that require it.
3689	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3690	 * obtain the locks in same order here.
3691	 */
3692	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3693		up_read(&block->cb_lock);
3694		take_rtnl = true;
3695		goto retry;
3696	}
3697
3698	/* Make sure all netdevs sharing this block are offload-capable. */
3699	if (block->nooffloaddevcnt && err_stop) {
3700		ok_count = -EOPNOTSUPP;
3701		goto err_unlock;
3702	}
3703
3704	tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3705	if (tp->ops->hw_del)
3706		tp->ops->hw_del(tp, type_data);
3707
3708	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3709	if (ok_count < 0)
3710		goto err_unlock;
3711
3712	if (tp->ops->hw_add)
3713		tp->ops->hw_add(tp, type_data);
3714	if (ok_count > 0)
3715		tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3716					  new_flags, ok_count, true);
3717err_unlock:
3718	up_read(&block->cb_lock);
3719	if (take_rtnl)
3720		rtnl_unlock();
3721	return min(ok_count, 0);
3722}
3723EXPORT_SYMBOL(tc_setup_cb_replace);
3724
3725/* Destroy filter and decrement block offload counter, if filter was previously
3726 * offloaded.
3727 */
3728
3729int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3730			enum tc_setup_type type, void *type_data, bool err_stop,
3731			u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3732{
3733	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3734	int ok_count;
3735
3736retry:
3737	if (take_rtnl)
3738		rtnl_lock();
3739	down_read(&block->cb_lock);
3740	/* Need to obtain rtnl lock if block is bound to devs that require it.
3741	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3742	 * obtain the locks in same order here.
3743	 */
3744	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3745		up_read(&block->cb_lock);
3746		take_rtnl = true;
3747		goto retry;
3748	}
3749
3750	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3751
3752	tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3753	if (tp->ops->hw_del)
3754		tp->ops->hw_del(tp, type_data);
3755
3756	up_read(&block->cb_lock);
3757	if (take_rtnl)
3758		rtnl_unlock();
3759	return min(ok_count, 0);
3760}
3761EXPORT_SYMBOL(tc_setup_cb_destroy);
3762
3763int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3764			  bool add, flow_setup_cb_t *cb,
3765			  enum tc_setup_type type, void *type_data,
3766			  void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3767{
3768	int err = cb(type, type_data, cb_priv);
3769
3770	if (err) {
3771		if (add && tc_skip_sw(*flags))
3772			return err;
3773	} else {
3774		tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3775					  add);
3776	}
3777
3778	return 0;
3779}
3780EXPORT_SYMBOL(tc_setup_cb_reoffload);
3781
3782static int tcf_act_get_user_cookie(struct flow_action_entry *entry,
3783				   const struct tc_action *act)
3784{
3785	struct tc_cookie *user_cookie;
3786	int err = 0;
3787
3788	rcu_read_lock();
3789	user_cookie = rcu_dereference(act->user_cookie);
3790	if (user_cookie) {
3791		entry->user_cookie = flow_action_cookie_create(user_cookie->data,
3792							       user_cookie->len,
3793							       GFP_ATOMIC);
3794		if (!entry->user_cookie)
3795			err = -ENOMEM;
3796	}
3797	rcu_read_unlock();
3798	return err;
3799}
3800
3801static void tcf_act_put_user_cookie(struct flow_action_entry *entry)
3802{
3803	flow_action_cookie_destroy(entry->user_cookie);
3804}
3805
3806void tc_cleanup_offload_action(struct flow_action *flow_action)
3807{
3808	struct flow_action_entry *entry;
3809	int i;
3810
3811	flow_action_for_each(i, entry, flow_action) {
3812		tcf_act_put_user_cookie(entry);
3813		if (entry->destructor)
3814			entry->destructor(entry->destructor_priv);
3815	}
3816}
3817EXPORT_SYMBOL(tc_cleanup_offload_action);
3818
3819static int tc_setup_offload_act(struct tc_action *act,
3820				struct flow_action_entry *entry,
3821				u32 *index_inc,
3822				struct netlink_ext_ack *extack)
3823{
3824#ifdef CONFIG_NET_CLS_ACT
3825	if (act->ops->offload_act_setup) {
3826		return act->ops->offload_act_setup(act, entry, index_inc, true,
3827						   extack);
3828	} else {
3829		NL_SET_ERR_MSG(extack, "Action does not support offload");
3830		return -EOPNOTSUPP;
3831	}
3832#else
3833	return 0;
3834#endif
3835}
3836
3837int tc_setup_action(struct flow_action *flow_action,
3838		    struct tc_action *actions[],
3839		    u32 miss_cookie_base,
3840		    struct netlink_ext_ack *extack)
3841{
3842	int i, j, k, index, err = 0;
3843	struct tc_action *act;
3844
3845	BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3846	BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3847	BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3848
3849	if (!actions)
3850		return 0;
3851
3852	j = 0;
3853	tcf_act_for_each_action(i, act, actions) {
3854		struct flow_action_entry *entry;
3855
3856		entry = &flow_action->entries[j];
3857		spin_lock_bh(&act->tcfa_lock);
3858		err = tcf_act_get_user_cookie(entry, act);
3859		if (err)
3860			goto err_out_locked;
3861
3862		index = 0;
3863		err = tc_setup_offload_act(act, entry, &index, extack);
3864		if (err)
3865			goto err_out_locked;
3866
3867		for (k = 0; k < index ; k++) {
3868			entry[k].hw_stats = tc_act_hw_stats(act->hw_stats);
3869			entry[k].hw_index = act->tcfa_index;
3870			entry[k].cookie = (unsigned long)act;
3871			entry[k].miss_cookie =
3872				tcf_exts_miss_cookie_get(miss_cookie_base, i);
3873		}
3874
3875		j += index;
3876
3877		spin_unlock_bh(&act->tcfa_lock);
3878	}
3879
3880err_out:
3881	if (err)
3882		tc_cleanup_offload_action(flow_action);
3883
3884	return err;
3885err_out_locked:
3886	spin_unlock_bh(&act->tcfa_lock);
3887	goto err_out;
3888}
3889
3890int tc_setup_offload_action(struct flow_action *flow_action,
3891			    const struct tcf_exts *exts,
3892			    struct netlink_ext_ack *extack)
3893{
3894#ifdef CONFIG_NET_CLS_ACT
3895	u32 miss_cookie_base;
3896
3897	if (!exts)
3898		return 0;
3899
3900	miss_cookie_base = exts->miss_cookie_node ?
3901			   exts->miss_cookie_node->miss_cookie_base : 0;
3902	return tc_setup_action(flow_action, exts->actions, miss_cookie_base,
3903			       extack);
3904#else
3905	return 0;
3906#endif
3907}
3908EXPORT_SYMBOL(tc_setup_offload_action);
3909
3910unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3911{
3912	unsigned int num_acts = 0;
3913	struct tc_action *act;
3914	int i;
3915
3916	tcf_exts_for_each_action(i, act, exts) {
3917		if (is_tcf_pedit(act))
3918			num_acts += tcf_pedit_nkeys(act);
3919		else
3920			num_acts++;
3921	}
3922	return num_acts;
3923}
3924EXPORT_SYMBOL(tcf_exts_num_actions);
3925
3926#ifdef CONFIG_NET_CLS_ACT
3927static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3928					u32 *p_block_index,
3929					struct netlink_ext_ack *extack)
3930{
3931	*p_block_index = nla_get_u32(block_index_attr);
3932	if (!*p_block_index) {
3933		NL_SET_ERR_MSG(extack, "Block number may not be zero");
3934		return -EINVAL;
3935	}
3936
3937	return 0;
3938}
3939
3940int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3941		    enum flow_block_binder_type binder_type,
3942		    struct nlattr *block_index_attr,
3943		    struct netlink_ext_ack *extack)
3944{
3945	u32 block_index;
3946	int err;
3947
3948	if (!block_index_attr)
3949		return 0;
3950
3951	err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3952	if (err)
3953		return err;
3954
3955	qe->info.binder_type = binder_type;
3956	qe->info.chain_head_change = tcf_chain_head_change_dflt;
3957	qe->info.chain_head_change_priv = &qe->filter_chain;
3958	qe->info.block_index = block_index;
3959
3960	return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3961}
3962EXPORT_SYMBOL(tcf_qevent_init);
3963
3964void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3965{
3966	if (qe->info.block_index)
3967		tcf_block_put_ext(qe->block, sch, &qe->info);
3968}
3969EXPORT_SYMBOL(tcf_qevent_destroy);
3970
3971int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3972			       struct netlink_ext_ack *extack)
3973{
3974	u32 block_index;
3975	int err;
3976
3977	if (!block_index_attr)
3978		return 0;
3979
3980	err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3981	if (err)
3982		return err;
3983
3984	/* Bounce newly-configured block or change in block. */
3985	if (block_index != qe->info.block_index) {
3986		NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3987		return -EINVAL;
3988	}
3989
3990	return 0;
3991}
3992EXPORT_SYMBOL(tcf_qevent_validate_change);
3993
3994struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
3995				  struct sk_buff **to_free, int *ret)
3996{
3997	struct tcf_result cl_res;
3998	struct tcf_proto *fl;
3999
4000	if (!qe->info.block_index)
4001		return skb;
4002
4003	fl = rcu_dereference_bh(qe->filter_chain);
4004
4005	switch (tcf_classify(skb, NULL, fl, &cl_res, false)) {
4006	case TC_ACT_SHOT:
4007		qdisc_qstats_drop(sch);
4008		__qdisc_drop(skb, to_free);
4009		*ret = __NET_XMIT_BYPASS;
4010		return NULL;
4011	case TC_ACT_STOLEN:
4012	case TC_ACT_QUEUED:
4013	case TC_ACT_TRAP:
4014		__qdisc_drop(skb, to_free);
4015		*ret = __NET_XMIT_STOLEN;
4016		return NULL;
4017	case TC_ACT_REDIRECT:
4018		skb_do_redirect(skb);
4019		*ret = __NET_XMIT_STOLEN;
4020		return NULL;
4021	}
4022
4023	return skb;
4024}
4025EXPORT_SYMBOL(tcf_qevent_handle);
4026
4027int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
4028{
4029	if (!qe->info.block_index)
4030		return 0;
4031	return nla_put_u32(skb, attr_name, qe->info.block_index);
4032}
4033EXPORT_SYMBOL(tcf_qevent_dump);
4034#endif
4035
4036static __net_init int tcf_net_init(struct net *net)
4037{
4038	struct tcf_net *tn = net_generic(net, tcf_net_id);
4039
4040	spin_lock_init(&tn->idr_lock);
4041	idr_init(&tn->idr);
4042	return 0;
4043}
4044
4045static void __net_exit tcf_net_exit(struct net *net)
4046{
4047	struct tcf_net *tn = net_generic(net, tcf_net_id);
4048
4049	idr_destroy(&tn->idr);
4050}
4051
4052static struct pernet_operations tcf_net_ops = {
4053	.init = tcf_net_init,
4054	.exit = tcf_net_exit,
4055	.id   = &tcf_net_id,
4056	.size = sizeof(struct tcf_net),
4057};
4058
4059static const struct rtnl_msg_handler tc_filter_rtnl_msg_handlers[] __initconst = {
4060	{.msgtype = RTM_NEWTFILTER, .doit = tc_new_tfilter,
4061	 .flags = RTNL_FLAG_DOIT_UNLOCKED},
4062	{.msgtype = RTM_DELTFILTER, .doit = tc_del_tfilter,
4063	 .flags = RTNL_FLAG_DOIT_UNLOCKED},
4064	{.msgtype = RTM_GETTFILTER, .doit = tc_get_tfilter,
4065	 .dumpit = tc_dump_tfilter, .flags = RTNL_FLAG_DOIT_UNLOCKED},
4066	{.msgtype = RTM_NEWCHAIN, .doit = tc_ctl_chain},
4067	{.msgtype = RTM_DELCHAIN, .doit = tc_ctl_chain},
4068	{.msgtype = RTM_GETCHAIN, .doit = tc_ctl_chain,
4069	 .dumpit = tc_dump_chain},
4070};
4071
4072static int __init tc_filter_init(void)
4073{
4074	int err;
4075
4076	tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
4077	if (!tc_filter_wq)
4078		return -ENOMEM;
4079
4080	err = register_pernet_subsys(&tcf_net_ops);
4081	if (err)
4082		goto err_register_pernet_subsys;
4083
4084	xa_init_flags(&tcf_exts_miss_cookies_xa, XA_FLAGS_ALLOC1);
4085	rtnl_register_many(tc_filter_rtnl_msg_handlers);
 
 
 
 
 
 
 
 
 
 
4086
4087	return 0;
4088
4089err_register_pernet_subsys:
4090	destroy_workqueue(tc_filter_wq);
4091	return err;
4092}
4093
4094subsys_initcall(tc_filter_init);