xref: /linux/net/sched/cls_api.c (revision fcc79e1714e8c2b8e216dc3149812edd37884eef)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * net/sched/cls_api.c	Packet classifier API.
4  *
5  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6  *
7  * Changes:
8  *
9  * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
10  */
11 
12 #include <linux/module.h>
13 #include <linux/types.h>
14 #include <linux/kernel.h>
15 #include <linux/string.h>
16 #include <linux/errno.h>
17 #include <linux/err.h>
18 #include <linux/skbuff.h>
19 #include <linux/init.h>
20 #include <linux/kmod.h>
21 #include <linux/slab.h>
22 #include <linux/idr.h>
23 #include <linux/jhash.h>
24 #include <linux/rculist.h>
25 #include <linux/rhashtable.h>
26 #include <net/net_namespace.h>
27 #include <net/sock.h>
28 #include <net/netlink.h>
29 #include <net/pkt_sched.h>
30 #include <net/pkt_cls.h>
31 #include <net/tc_act/tc_pedit.h>
32 #include <net/tc_act/tc_mirred.h>
33 #include <net/tc_act/tc_vlan.h>
34 #include <net/tc_act/tc_tunnel_key.h>
35 #include <net/tc_act/tc_csum.h>
36 #include <net/tc_act/tc_gact.h>
37 #include <net/tc_act/tc_police.h>
38 #include <net/tc_act/tc_sample.h>
39 #include <net/tc_act/tc_skbedit.h>
40 #include <net/tc_act/tc_ct.h>
41 #include <net/tc_act/tc_mpls.h>
42 #include <net/tc_act/tc_gate.h>
43 #include <net/flow_offload.h>
44 #include <net/tc_wrapper.h>
45 
46 /* The list of all installed classifier types */
47 static LIST_HEAD(tcf_proto_base);
48 
49 /* Protects list of registered TC modules. It is pure SMP lock. */
50 static DEFINE_RWLOCK(cls_mod_lock);
51 
52 static struct xarray tcf_exts_miss_cookies_xa;
53 struct tcf_exts_miss_cookie_node {
54 	const struct tcf_chain *chain;
55 	const struct tcf_proto *tp;
56 	const struct tcf_exts *exts;
57 	u32 chain_index;
58 	u32 tp_prio;
59 	u32 handle;
60 	u32 miss_cookie_base;
61 	struct rcu_head rcu;
62 };
63 
64 /* Each tc action entry cookie will be comprised of 32bit miss_cookie_base +
65  * action index in the exts tc actions array.
66  */
67 union tcf_exts_miss_cookie {
68 	struct {
69 		u32 miss_cookie_base;
70 		u32 act_index;
71 	};
72 	u64 miss_cookie;
73 };
74 
75 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
76 static int
77 tcf_exts_miss_cookie_base_alloc(struct tcf_exts *exts, struct tcf_proto *tp,
78 				u32 handle)
79 {
80 	struct tcf_exts_miss_cookie_node *n;
81 	static u32 next;
82 	int err;
83 
84 	if (WARN_ON(!handle || !tp->ops->get_exts))
85 		return -EINVAL;
86 
87 	n = kzalloc(sizeof(*n), GFP_KERNEL);
88 	if (!n)
89 		return -ENOMEM;
90 
91 	n->chain_index = tp->chain->index;
92 	n->chain = tp->chain;
93 	n->tp_prio = tp->prio;
94 	n->tp = tp;
95 	n->exts = exts;
96 	n->handle = handle;
97 
98 	err = xa_alloc_cyclic(&tcf_exts_miss_cookies_xa, &n->miss_cookie_base,
99 			      n, xa_limit_32b, &next, GFP_KERNEL);
100 	if (err)
101 		goto err_xa_alloc;
102 
103 	exts->miss_cookie_node = n;
104 	return 0;
105 
106 err_xa_alloc:
107 	kfree(n);
108 	return err;
109 }
110 
111 static void tcf_exts_miss_cookie_base_destroy(struct tcf_exts *exts)
112 {
113 	struct tcf_exts_miss_cookie_node *n;
114 
115 	if (!exts->miss_cookie_node)
116 		return;
117 
118 	n = exts->miss_cookie_node;
119 	xa_erase(&tcf_exts_miss_cookies_xa, n->miss_cookie_base);
120 	kfree_rcu(n, rcu);
121 }
122 
123 static struct tcf_exts_miss_cookie_node *
124 tcf_exts_miss_cookie_lookup(u64 miss_cookie, int *act_index)
125 {
126 	union tcf_exts_miss_cookie mc = { .miss_cookie = miss_cookie, };
127 
128 	*act_index = mc.act_index;
129 	return xa_load(&tcf_exts_miss_cookies_xa, mc.miss_cookie_base);
130 }
131 #else /* IS_ENABLED(CONFIG_NET_TC_SKB_EXT) */
132 static int
133 tcf_exts_miss_cookie_base_alloc(struct tcf_exts *exts, struct tcf_proto *tp,
134 				u32 handle)
135 {
136 	return 0;
137 }
138 
139 static void tcf_exts_miss_cookie_base_destroy(struct tcf_exts *exts)
140 {
141 }
142 #endif /* IS_ENABLED(CONFIG_NET_TC_SKB_EXT) */
143 
144 static u64 tcf_exts_miss_cookie_get(u32 miss_cookie_base, int act_index)
145 {
146 	union tcf_exts_miss_cookie mc = { .act_index = act_index, };
147 
148 	if (!miss_cookie_base)
149 		return 0;
150 
151 	mc.miss_cookie_base = miss_cookie_base;
152 	return mc.miss_cookie;
153 }
154 
155 #ifdef CONFIG_NET_CLS_ACT
156 DEFINE_STATIC_KEY_FALSE(tc_skb_ext_tc);
157 EXPORT_SYMBOL(tc_skb_ext_tc);
158 
159 void tc_skb_ext_tc_enable(void)
160 {
161 	static_branch_inc(&tc_skb_ext_tc);
162 }
163 EXPORT_SYMBOL(tc_skb_ext_tc_enable);
164 
165 void tc_skb_ext_tc_disable(void)
166 {
167 	static_branch_dec(&tc_skb_ext_tc);
168 }
169 EXPORT_SYMBOL(tc_skb_ext_tc_disable);
170 #endif
171 
172 static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
173 {
174 	return jhash_3words(tp->chain->index, tp->prio,
175 			    (__force __u32)tp->protocol, 0);
176 }
177 
178 static void tcf_proto_signal_destroying(struct tcf_chain *chain,
179 					struct tcf_proto *tp)
180 {
181 	struct tcf_block *block = chain->block;
182 
183 	mutex_lock(&block->proto_destroy_lock);
184 	hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
185 		     destroy_obj_hashfn(tp));
186 	mutex_unlock(&block->proto_destroy_lock);
187 }
188 
189 static bool tcf_proto_cmp(const struct tcf_proto *tp1,
190 			  const struct tcf_proto *tp2)
191 {
192 	return tp1->chain->index == tp2->chain->index &&
193 	       tp1->prio == tp2->prio &&
194 	       tp1->protocol == tp2->protocol;
195 }
196 
197 static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
198 					struct tcf_proto *tp)
199 {
200 	u32 hash = destroy_obj_hashfn(tp);
201 	struct tcf_proto *iter;
202 	bool found = false;
203 
204 	rcu_read_lock();
205 	hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
206 				   destroy_ht_node, hash) {
207 		if (tcf_proto_cmp(tp, iter)) {
208 			found = true;
209 			break;
210 		}
211 	}
212 	rcu_read_unlock();
213 
214 	return found;
215 }
216 
217 static void
218 tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
219 {
220 	struct tcf_block *block = chain->block;
221 
222 	mutex_lock(&block->proto_destroy_lock);
223 	if (hash_hashed(&tp->destroy_ht_node))
224 		hash_del_rcu(&tp->destroy_ht_node);
225 	mutex_unlock(&block->proto_destroy_lock);
226 }
227 
228 /* Find classifier type by string name */
229 
230 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
231 {
232 	const struct tcf_proto_ops *t, *res = NULL;
233 
234 	if (kind) {
235 		read_lock(&cls_mod_lock);
236 		list_for_each_entry(t, &tcf_proto_base, head) {
237 			if (strcmp(kind, t->kind) == 0) {
238 				if (try_module_get(t->owner))
239 					res = t;
240 				break;
241 			}
242 		}
243 		read_unlock(&cls_mod_lock);
244 	}
245 	return res;
246 }
247 
248 static const struct tcf_proto_ops *
249 tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
250 		     struct netlink_ext_ack *extack)
251 {
252 	const struct tcf_proto_ops *ops;
253 
254 	ops = __tcf_proto_lookup_ops(kind);
255 	if (ops)
256 		return ops;
257 #ifdef CONFIG_MODULES
258 	if (rtnl_held)
259 		rtnl_unlock();
260 	request_module(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 
280 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
281 {
282 	struct tcf_proto_ops *t;
283 	int rc = -EEXIST;
284 
285 	write_lock(&cls_mod_lock);
286 	list_for_each_entry(t, &tcf_proto_base, head)
287 		if (!strcmp(ops->kind, t->kind))
288 			goto out;
289 
290 	list_add_tail(&ops->head, &tcf_proto_base);
291 	rc = 0;
292 out:
293 	write_unlock(&cls_mod_lock);
294 	return rc;
295 }
296 EXPORT_SYMBOL(register_tcf_proto_ops);
297 
298 static struct workqueue_struct *tc_filter_wq;
299 
300 void unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
301 {
302 	struct tcf_proto_ops *t;
303 	int rc = -ENOENT;
304 
305 	/* Wait for outstanding call_rcu()s, if any, from a
306 	 * tcf_proto_ops's destroy() handler.
307 	 */
308 	rcu_barrier();
309 	flush_workqueue(tc_filter_wq);
310 
311 	write_lock(&cls_mod_lock);
312 	list_for_each_entry(t, &tcf_proto_base, head) {
313 		if (t == ops) {
314 			list_del(&t->head);
315 			rc = 0;
316 			break;
317 		}
318 	}
319 	write_unlock(&cls_mod_lock);
320 
321 	WARN(rc, "unregister tc filter kind(%s) failed %d\n", ops->kind, rc);
322 }
323 EXPORT_SYMBOL(unregister_tcf_proto_ops);
324 
325 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
326 {
327 	INIT_RCU_WORK(rwork, func);
328 	return queue_rcu_work(tc_filter_wq, rwork);
329 }
330 EXPORT_SYMBOL(tcf_queue_work);
331 
332 /* Select new prio value from the range, managed by kernel. */
333 
334 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
335 {
336 	u32 first = TC_H_MAKE(0xC0000000U, 0U);
337 
338 	if (tp)
339 		first = tp->prio - 1;
340 
341 	return TC_H_MAJ(first);
342 }
343 
344 static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
345 {
346 	if (kind)
347 		return nla_strscpy(name, kind, IFNAMSIZ) < 0;
348 	memset(name, 0, IFNAMSIZ);
349 	return false;
350 }
351 
352 static bool tcf_proto_is_unlocked(const char *kind)
353 {
354 	const struct tcf_proto_ops *ops;
355 	bool ret;
356 
357 	if (strlen(kind) == 0)
358 		return false;
359 
360 	ops = tcf_proto_lookup_ops(kind, false, NULL);
361 	/* On error return false to take rtnl lock. Proto lookup/create
362 	 * functions will perform lookup again and properly handle errors.
363 	 */
364 	if (IS_ERR(ops))
365 		return false;
366 
367 	ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
368 	module_put(ops->owner);
369 	return ret;
370 }
371 
372 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
373 					  u32 prio, struct tcf_chain *chain,
374 					  bool rtnl_held,
375 					  struct netlink_ext_ack *extack)
376 {
377 	struct tcf_proto *tp;
378 	int err;
379 
380 	tp = kzalloc(sizeof(*tp), GFP_KERNEL);
381 	if (!tp)
382 		return ERR_PTR(-ENOBUFS);
383 
384 	tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
385 	if (IS_ERR(tp->ops)) {
386 		err = PTR_ERR(tp->ops);
387 		goto errout;
388 	}
389 	tp->classify = tp->ops->classify;
390 	tp->protocol = protocol;
391 	tp->prio = prio;
392 	tp->chain = chain;
393 	spin_lock_init(&tp->lock);
394 	refcount_set(&tp->refcnt, 1);
395 
396 	err = tp->ops->init(tp);
397 	if (err) {
398 		module_put(tp->ops->owner);
399 		goto errout;
400 	}
401 	return tp;
402 
403 errout:
404 	kfree(tp);
405 	return ERR_PTR(err);
406 }
407 
408 static void tcf_proto_get(struct tcf_proto *tp)
409 {
410 	refcount_inc(&tp->refcnt);
411 }
412 
413 static void tcf_maintain_bypass(struct tcf_block *block)
414 {
415 	int filtercnt = atomic_read(&block->filtercnt);
416 	int skipswcnt = atomic_read(&block->skipswcnt);
417 	bool bypass_wanted = filtercnt > 0 && filtercnt == skipswcnt;
418 
419 	if (bypass_wanted != block->bypass_wanted) {
420 #ifdef CONFIG_NET_CLS_ACT
421 		if (bypass_wanted)
422 			static_branch_inc(&tcf_bypass_check_needed_key);
423 		else
424 			static_branch_dec(&tcf_bypass_check_needed_key);
425 #endif
426 		block->bypass_wanted = bypass_wanted;
427 	}
428 }
429 
430 static void tcf_block_filter_cnt_update(struct tcf_block *block, bool *counted, bool add)
431 {
432 	lockdep_assert_not_held(&block->cb_lock);
433 
434 	down_write(&block->cb_lock);
435 	if (*counted != add) {
436 		if (add) {
437 			atomic_inc(&block->filtercnt);
438 			*counted = true;
439 		} else {
440 			atomic_dec(&block->filtercnt);
441 			*counted = false;
442 		}
443 	}
444 	tcf_maintain_bypass(block);
445 	up_write(&block->cb_lock);
446 }
447 
448 static void tcf_chain_put(struct tcf_chain *chain);
449 
450 static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
451 			      bool sig_destroy, struct netlink_ext_ack *extack)
452 {
453 	tp->ops->destroy(tp, rtnl_held, extack);
454 	tcf_block_filter_cnt_update(tp->chain->block, &tp->counted, false);
455 	if (sig_destroy)
456 		tcf_proto_signal_destroyed(tp->chain, tp);
457 	tcf_chain_put(tp->chain);
458 	module_put(tp->ops->owner);
459 	kfree_rcu(tp, rcu);
460 }
461 
462 static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
463 			  struct netlink_ext_ack *extack)
464 {
465 	if (refcount_dec_and_test(&tp->refcnt))
466 		tcf_proto_destroy(tp, rtnl_held, true, extack);
467 }
468 
469 static bool tcf_proto_check_delete(struct tcf_proto *tp)
470 {
471 	if (tp->ops->delete_empty)
472 		return tp->ops->delete_empty(tp);
473 
474 	tp->deleting = true;
475 	return tp->deleting;
476 }
477 
478 static void tcf_proto_mark_delete(struct tcf_proto *tp)
479 {
480 	spin_lock(&tp->lock);
481 	tp->deleting = true;
482 	spin_unlock(&tp->lock);
483 }
484 
485 static bool tcf_proto_is_deleting(struct tcf_proto *tp)
486 {
487 	bool deleting;
488 
489 	spin_lock(&tp->lock);
490 	deleting = tp->deleting;
491 	spin_unlock(&tp->lock);
492 
493 	return deleting;
494 }
495 
496 #define ASSERT_BLOCK_LOCKED(block)					\
497 	lockdep_assert_held(&(block)->lock)
498 
499 struct tcf_filter_chain_list_item {
500 	struct list_head list;
501 	tcf_chain_head_change_t *chain_head_change;
502 	void *chain_head_change_priv;
503 };
504 
505 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
506 					  u32 chain_index)
507 {
508 	struct tcf_chain *chain;
509 
510 	ASSERT_BLOCK_LOCKED(block);
511 
512 	chain = kzalloc(sizeof(*chain), GFP_KERNEL);
513 	if (!chain)
514 		return NULL;
515 	list_add_tail_rcu(&chain->list, &block->chain_list);
516 	mutex_init(&chain->filter_chain_lock);
517 	chain->block = block;
518 	chain->index = chain_index;
519 	chain->refcnt = 1;
520 	if (!chain->index)
521 		block->chain0.chain = chain;
522 	return chain;
523 }
524 
525 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
526 				       struct tcf_proto *tp_head)
527 {
528 	if (item->chain_head_change)
529 		item->chain_head_change(tp_head, item->chain_head_change_priv);
530 }
531 
532 static void tcf_chain0_head_change(struct tcf_chain *chain,
533 				   struct tcf_proto *tp_head)
534 {
535 	struct tcf_filter_chain_list_item *item;
536 	struct tcf_block *block = chain->block;
537 
538 	if (chain->index)
539 		return;
540 
541 	mutex_lock(&block->lock);
542 	list_for_each_entry(item, &block->chain0.filter_chain_list, list)
543 		tcf_chain_head_change_item(item, tp_head);
544 	mutex_unlock(&block->lock);
545 }
546 
547 /* Returns true if block can be safely freed. */
548 
549 static bool tcf_chain_detach(struct tcf_chain *chain)
550 {
551 	struct tcf_block *block = chain->block;
552 
553 	ASSERT_BLOCK_LOCKED(block);
554 
555 	list_del_rcu(&chain->list);
556 	if (!chain->index)
557 		block->chain0.chain = NULL;
558 
559 	if (list_empty(&block->chain_list) &&
560 	    refcount_read(&block->refcnt) == 0)
561 		return true;
562 
563 	return false;
564 }
565 
566 static void tcf_block_destroy(struct tcf_block *block)
567 {
568 	mutex_destroy(&block->lock);
569 	mutex_destroy(&block->proto_destroy_lock);
570 	xa_destroy(&block->ports);
571 	kfree_rcu(block, rcu);
572 }
573 
574 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
575 {
576 	struct tcf_block *block = chain->block;
577 
578 	mutex_destroy(&chain->filter_chain_lock);
579 	kfree_rcu(chain, rcu);
580 	if (free_block)
581 		tcf_block_destroy(block);
582 }
583 
584 static void tcf_chain_hold(struct tcf_chain *chain)
585 {
586 	ASSERT_BLOCK_LOCKED(chain->block);
587 
588 	++chain->refcnt;
589 }
590 
591 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
592 {
593 	ASSERT_BLOCK_LOCKED(chain->block);
594 
595 	/* In case all the references are action references, this
596 	 * chain should not be shown to the user.
597 	 */
598 	return chain->refcnt == chain->action_refcnt;
599 }
600 
601 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
602 					  u32 chain_index)
603 {
604 	struct tcf_chain *chain;
605 
606 	ASSERT_BLOCK_LOCKED(block);
607 
608 	list_for_each_entry(chain, &block->chain_list, list) {
609 		if (chain->index == chain_index)
610 			return chain;
611 	}
612 	return NULL;
613 }
614 
615 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
616 static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
617 					      u32 chain_index)
618 {
619 	struct tcf_chain *chain;
620 
621 	list_for_each_entry_rcu(chain, &block->chain_list, list) {
622 		if (chain->index == chain_index)
623 			return chain;
624 	}
625 	return NULL;
626 }
627 #endif
628 
629 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
630 			   u32 seq, u16 flags, int event, bool unicast,
631 			   struct netlink_ext_ack *extack);
632 
633 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
634 					 u32 chain_index, bool create,
635 					 bool by_act)
636 {
637 	struct tcf_chain *chain = NULL;
638 	bool is_first_reference;
639 
640 	mutex_lock(&block->lock);
641 	chain = tcf_chain_lookup(block, chain_index);
642 	if (chain) {
643 		tcf_chain_hold(chain);
644 	} else {
645 		if (!create)
646 			goto errout;
647 		chain = tcf_chain_create(block, chain_index);
648 		if (!chain)
649 			goto errout;
650 	}
651 
652 	if (by_act)
653 		++chain->action_refcnt;
654 	is_first_reference = chain->refcnt - chain->action_refcnt == 1;
655 	mutex_unlock(&block->lock);
656 
657 	/* Send notification only in case we got the first
658 	 * non-action reference. Until then, the chain acts only as
659 	 * a placeholder for actions pointing to it and user ought
660 	 * not know about them.
661 	 */
662 	if (is_first_reference && !by_act)
663 		tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
664 				RTM_NEWCHAIN, false, NULL);
665 
666 	return chain;
667 
668 errout:
669 	mutex_unlock(&block->lock);
670 	return chain;
671 }
672 
673 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
674 				       bool create)
675 {
676 	return __tcf_chain_get(block, chain_index, create, false);
677 }
678 
679 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
680 {
681 	return __tcf_chain_get(block, chain_index, true, true);
682 }
683 EXPORT_SYMBOL(tcf_chain_get_by_act);
684 
685 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
686 			       void *tmplt_priv);
687 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
688 				  void *tmplt_priv, u32 chain_index,
689 				  struct tcf_block *block, struct sk_buff *oskb,
690 				  u32 seq, u16 flags);
691 
692 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
693 			    bool explicitly_created)
694 {
695 	struct tcf_block *block = chain->block;
696 	const struct tcf_proto_ops *tmplt_ops;
697 	unsigned int refcnt, non_act_refcnt;
698 	bool free_block = false;
699 	void *tmplt_priv;
700 
701 	mutex_lock(&block->lock);
702 	if (explicitly_created) {
703 		if (!chain->explicitly_created) {
704 			mutex_unlock(&block->lock);
705 			return;
706 		}
707 		chain->explicitly_created = false;
708 	}
709 
710 	if (by_act)
711 		chain->action_refcnt--;
712 
713 	/* tc_chain_notify_delete can't be called while holding block lock.
714 	 * However, when block is unlocked chain can be changed concurrently, so
715 	 * save these to temporary variables.
716 	 */
717 	refcnt = --chain->refcnt;
718 	non_act_refcnt = refcnt - chain->action_refcnt;
719 	tmplt_ops = chain->tmplt_ops;
720 	tmplt_priv = chain->tmplt_priv;
721 
722 	if (non_act_refcnt == chain->explicitly_created && !by_act) {
723 		if (non_act_refcnt == 0)
724 			tc_chain_notify_delete(tmplt_ops, tmplt_priv,
725 					       chain->index, block, NULL, 0, 0);
726 		/* Last reference to chain, no need to lock. */
727 		chain->flushing = false;
728 	}
729 
730 	if (refcnt == 0)
731 		free_block = tcf_chain_detach(chain);
732 	mutex_unlock(&block->lock);
733 
734 	if (refcnt == 0) {
735 		tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
736 		tcf_chain_destroy(chain, free_block);
737 	}
738 }
739 
740 static void tcf_chain_put(struct tcf_chain *chain)
741 {
742 	__tcf_chain_put(chain, false, false);
743 }
744 
745 void tcf_chain_put_by_act(struct tcf_chain *chain)
746 {
747 	__tcf_chain_put(chain, true, false);
748 }
749 EXPORT_SYMBOL(tcf_chain_put_by_act);
750 
751 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
752 {
753 	__tcf_chain_put(chain, false, true);
754 }
755 
756 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
757 {
758 	struct tcf_proto *tp, *tp_next;
759 
760 	mutex_lock(&chain->filter_chain_lock);
761 	tp = tcf_chain_dereference(chain->filter_chain, chain);
762 	while (tp) {
763 		tp_next = rcu_dereference_protected(tp->next, 1);
764 		tcf_proto_signal_destroying(chain, tp);
765 		tp = tp_next;
766 	}
767 	tp = tcf_chain_dereference(chain->filter_chain, chain);
768 	RCU_INIT_POINTER(chain->filter_chain, NULL);
769 	tcf_chain0_head_change(chain, NULL);
770 	chain->flushing = true;
771 	mutex_unlock(&chain->filter_chain_lock);
772 
773 	while (tp) {
774 		tp_next = rcu_dereference_protected(tp->next, 1);
775 		tcf_proto_put(tp, rtnl_held, NULL);
776 		tp = tp_next;
777 	}
778 }
779 
780 static int tcf_block_setup(struct tcf_block *block,
781 			   struct flow_block_offload *bo);
782 
783 static void tcf_block_offload_init(struct flow_block_offload *bo,
784 				   struct net_device *dev, struct Qdisc *sch,
785 				   enum flow_block_command command,
786 				   enum flow_block_binder_type binder_type,
787 				   struct flow_block *flow_block,
788 				   bool shared, struct netlink_ext_ack *extack)
789 {
790 	bo->net = dev_net(dev);
791 	bo->command = command;
792 	bo->binder_type = binder_type;
793 	bo->block = flow_block;
794 	bo->block_shared = shared;
795 	bo->extack = extack;
796 	bo->sch = sch;
797 	bo->cb_list_head = &flow_block->cb_list;
798 	INIT_LIST_HEAD(&bo->cb_list);
799 }
800 
801 static void tcf_block_unbind(struct tcf_block *block,
802 			     struct flow_block_offload *bo);
803 
804 static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
805 {
806 	struct tcf_block *block = block_cb->indr.data;
807 	struct net_device *dev = block_cb->indr.dev;
808 	struct Qdisc *sch = block_cb->indr.sch;
809 	struct netlink_ext_ack extack = {};
810 	struct flow_block_offload bo = {};
811 
812 	tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
813 			       block_cb->indr.binder_type,
814 			       &block->flow_block, tcf_block_shared(block),
815 			       &extack);
816 	rtnl_lock();
817 	down_write(&block->cb_lock);
818 	list_del(&block_cb->driver_list);
819 	list_move(&block_cb->list, &bo.cb_list);
820 	tcf_block_unbind(block, &bo);
821 	up_write(&block->cb_lock);
822 	rtnl_unlock();
823 }
824 
825 static bool tcf_block_offload_in_use(struct tcf_block *block)
826 {
827 	return atomic_read(&block->offloadcnt);
828 }
829 
830 static int tcf_block_offload_cmd(struct tcf_block *block,
831 				 struct net_device *dev, struct Qdisc *sch,
832 				 struct tcf_block_ext_info *ei,
833 				 enum flow_block_command command,
834 				 struct netlink_ext_ack *extack)
835 {
836 	struct flow_block_offload bo = {};
837 
838 	tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
839 			       &block->flow_block, tcf_block_shared(block),
840 			       extack);
841 
842 	if (dev->netdev_ops->ndo_setup_tc) {
843 		int err;
844 
845 		err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
846 		if (err < 0) {
847 			if (err != -EOPNOTSUPP)
848 				NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
849 			return err;
850 		}
851 
852 		return tcf_block_setup(block, &bo);
853 	}
854 
855 	flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
856 				    tc_block_indr_cleanup);
857 	tcf_block_setup(block, &bo);
858 
859 	return -EOPNOTSUPP;
860 }
861 
862 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
863 				  struct tcf_block_ext_info *ei,
864 				  struct netlink_ext_ack *extack)
865 {
866 	struct net_device *dev = q->dev_queue->dev;
867 	int err;
868 
869 	down_write(&block->cb_lock);
870 
871 	/* If tc offload feature is disabled and the block we try to bind
872 	 * to already has some offloaded filters, forbid to bind.
873 	 */
874 	if (dev->netdev_ops->ndo_setup_tc &&
875 	    !tc_can_offload(dev) &&
876 	    tcf_block_offload_in_use(block)) {
877 		NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
878 		err = -EOPNOTSUPP;
879 		goto err_unlock;
880 	}
881 
882 	err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
883 	if (err == -EOPNOTSUPP)
884 		goto no_offload_dev_inc;
885 	if (err)
886 		goto err_unlock;
887 
888 	up_write(&block->cb_lock);
889 	return 0;
890 
891 no_offload_dev_inc:
892 	if (tcf_block_offload_in_use(block))
893 		goto err_unlock;
894 
895 	err = 0;
896 	block->nooffloaddevcnt++;
897 err_unlock:
898 	up_write(&block->cb_lock);
899 	return err;
900 }
901 
902 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
903 				     struct tcf_block_ext_info *ei)
904 {
905 	struct net_device *dev = q->dev_queue->dev;
906 	int err;
907 
908 	down_write(&block->cb_lock);
909 	err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
910 	if (err == -EOPNOTSUPP)
911 		goto no_offload_dev_dec;
912 	up_write(&block->cb_lock);
913 	return;
914 
915 no_offload_dev_dec:
916 	WARN_ON(block->nooffloaddevcnt-- == 0);
917 	up_write(&block->cb_lock);
918 }
919 
920 static int
921 tcf_chain0_head_change_cb_add(struct tcf_block *block,
922 			      struct tcf_block_ext_info *ei,
923 			      struct netlink_ext_ack *extack)
924 {
925 	struct tcf_filter_chain_list_item *item;
926 	struct tcf_chain *chain0;
927 
928 	item = kmalloc(sizeof(*item), GFP_KERNEL);
929 	if (!item) {
930 		NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
931 		return -ENOMEM;
932 	}
933 	item->chain_head_change = ei->chain_head_change;
934 	item->chain_head_change_priv = ei->chain_head_change_priv;
935 
936 	mutex_lock(&block->lock);
937 	chain0 = block->chain0.chain;
938 	if (chain0)
939 		tcf_chain_hold(chain0);
940 	else
941 		list_add(&item->list, &block->chain0.filter_chain_list);
942 	mutex_unlock(&block->lock);
943 
944 	if (chain0) {
945 		struct tcf_proto *tp_head;
946 
947 		mutex_lock(&chain0->filter_chain_lock);
948 
949 		tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
950 		if (tp_head)
951 			tcf_chain_head_change_item(item, tp_head);
952 
953 		mutex_lock(&block->lock);
954 		list_add(&item->list, &block->chain0.filter_chain_list);
955 		mutex_unlock(&block->lock);
956 
957 		mutex_unlock(&chain0->filter_chain_lock);
958 		tcf_chain_put(chain0);
959 	}
960 
961 	return 0;
962 }
963 
964 static void
965 tcf_chain0_head_change_cb_del(struct tcf_block *block,
966 			      struct tcf_block_ext_info *ei)
967 {
968 	struct tcf_filter_chain_list_item *item;
969 
970 	mutex_lock(&block->lock);
971 	list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
972 		if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
973 		    (item->chain_head_change == ei->chain_head_change &&
974 		     item->chain_head_change_priv == ei->chain_head_change_priv)) {
975 			if (block->chain0.chain)
976 				tcf_chain_head_change_item(item, NULL);
977 			list_del(&item->list);
978 			mutex_unlock(&block->lock);
979 
980 			kfree(item);
981 			return;
982 		}
983 	}
984 	mutex_unlock(&block->lock);
985 	WARN_ON(1);
986 }
987 
988 struct tcf_net {
989 	spinlock_t idr_lock; /* Protects idr */
990 	struct idr idr;
991 };
992 
993 static unsigned int tcf_net_id;
994 
995 static int tcf_block_insert(struct tcf_block *block, struct net *net,
996 			    struct netlink_ext_ack *extack)
997 {
998 	struct tcf_net *tn = net_generic(net, tcf_net_id);
999 	int err;
1000 
1001 	idr_preload(GFP_KERNEL);
1002 	spin_lock(&tn->idr_lock);
1003 	err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
1004 			    GFP_NOWAIT);
1005 	spin_unlock(&tn->idr_lock);
1006 	idr_preload_end();
1007 
1008 	return err;
1009 }
1010 
1011 static void tcf_block_remove(struct tcf_block *block, struct net *net)
1012 {
1013 	struct tcf_net *tn = net_generic(net, tcf_net_id);
1014 
1015 	spin_lock(&tn->idr_lock);
1016 	idr_remove(&tn->idr, block->index);
1017 	spin_unlock(&tn->idr_lock);
1018 }
1019 
1020 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
1021 					  u32 block_index,
1022 					  struct netlink_ext_ack *extack)
1023 {
1024 	struct tcf_block *block;
1025 
1026 	block = kzalloc(sizeof(*block), GFP_KERNEL);
1027 	if (!block) {
1028 		NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
1029 		return ERR_PTR(-ENOMEM);
1030 	}
1031 	mutex_init(&block->lock);
1032 	mutex_init(&block->proto_destroy_lock);
1033 	init_rwsem(&block->cb_lock);
1034 	flow_block_init(&block->flow_block);
1035 	INIT_LIST_HEAD(&block->chain_list);
1036 	INIT_LIST_HEAD(&block->owner_list);
1037 	INIT_LIST_HEAD(&block->chain0.filter_chain_list);
1038 
1039 	refcount_set(&block->refcnt, 1);
1040 	block->net = net;
1041 	block->index = block_index;
1042 	xa_init(&block->ports);
1043 
1044 	/* Don't store q pointer for blocks which are shared */
1045 	if (!tcf_block_shared(block))
1046 		block->q = q;
1047 	return block;
1048 }
1049 
1050 struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
1051 {
1052 	struct tcf_net *tn = net_generic(net, tcf_net_id);
1053 
1054 	return idr_find(&tn->idr, block_index);
1055 }
1056 EXPORT_SYMBOL(tcf_block_lookup);
1057 
1058 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
1059 {
1060 	struct tcf_block *block;
1061 
1062 	rcu_read_lock();
1063 	block = tcf_block_lookup(net, block_index);
1064 	if (block && !refcount_inc_not_zero(&block->refcnt))
1065 		block = NULL;
1066 	rcu_read_unlock();
1067 
1068 	return block;
1069 }
1070 
1071 static struct tcf_chain *
1072 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1073 {
1074 	mutex_lock(&block->lock);
1075 	if (chain)
1076 		chain = list_is_last(&chain->list, &block->chain_list) ?
1077 			NULL : list_next_entry(chain, list);
1078 	else
1079 		chain = list_first_entry_or_null(&block->chain_list,
1080 						 struct tcf_chain, list);
1081 
1082 	/* skip all action-only chains */
1083 	while (chain && tcf_chain_held_by_acts_only(chain))
1084 		chain = list_is_last(&chain->list, &block->chain_list) ?
1085 			NULL : list_next_entry(chain, list);
1086 
1087 	if (chain)
1088 		tcf_chain_hold(chain);
1089 	mutex_unlock(&block->lock);
1090 
1091 	return chain;
1092 }
1093 
1094 /* Function to be used by all clients that want to iterate over all chains on
1095  * block. It properly obtains block->lock and takes reference to chain before
1096  * returning it. Users of this function must be tolerant to concurrent chain
1097  * insertion/deletion or ensure that no concurrent chain modification is
1098  * possible. Note that all netlink dump callbacks cannot guarantee to provide
1099  * consistent dump because rtnl lock is released each time skb is filled with
1100  * data and sent to user-space.
1101  */
1102 
1103 struct tcf_chain *
1104 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1105 {
1106 	struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
1107 
1108 	if (chain)
1109 		tcf_chain_put(chain);
1110 
1111 	return chain_next;
1112 }
1113 EXPORT_SYMBOL(tcf_get_next_chain);
1114 
1115 static struct tcf_proto *
1116 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1117 {
1118 	u32 prio = 0;
1119 
1120 	ASSERT_RTNL();
1121 	mutex_lock(&chain->filter_chain_lock);
1122 
1123 	if (!tp) {
1124 		tp = tcf_chain_dereference(chain->filter_chain, chain);
1125 	} else if (tcf_proto_is_deleting(tp)) {
1126 		/* 'deleting' flag is set and chain->filter_chain_lock was
1127 		 * unlocked, which means next pointer could be invalid. Restart
1128 		 * search.
1129 		 */
1130 		prio = tp->prio + 1;
1131 		tp = tcf_chain_dereference(chain->filter_chain, chain);
1132 
1133 		for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1134 			if (!tp->deleting && tp->prio >= prio)
1135 				break;
1136 	} else {
1137 		tp = tcf_chain_dereference(tp->next, chain);
1138 	}
1139 
1140 	if (tp)
1141 		tcf_proto_get(tp);
1142 
1143 	mutex_unlock(&chain->filter_chain_lock);
1144 
1145 	return tp;
1146 }
1147 
1148 /* Function to be used by all clients that want to iterate over all tp's on
1149  * chain. Users of this function must be tolerant to concurrent tp
1150  * insertion/deletion or ensure that no concurrent chain modification is
1151  * possible. Note that all netlink dump callbacks cannot guarantee to provide
1152  * consistent dump because rtnl lock is released each time skb is filled with
1153  * data and sent to user-space.
1154  */
1155 
1156 struct tcf_proto *
1157 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1158 {
1159 	struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1160 
1161 	if (tp)
1162 		tcf_proto_put(tp, true, NULL);
1163 
1164 	return tp_next;
1165 }
1166 EXPORT_SYMBOL(tcf_get_next_proto);
1167 
1168 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1169 {
1170 	struct tcf_chain *chain;
1171 
1172 	/* Last reference to block. At this point chains cannot be added or
1173 	 * removed concurrently.
1174 	 */
1175 	for (chain = tcf_get_next_chain(block, NULL);
1176 	     chain;
1177 	     chain = tcf_get_next_chain(block, chain)) {
1178 		tcf_chain_put_explicitly_created(chain);
1179 		tcf_chain_flush(chain, rtnl_held);
1180 	}
1181 }
1182 
1183 /* Lookup Qdisc and increments its reference counter.
1184  * Set parent, if necessary.
1185  */
1186 
1187 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1188 			    u32 *parent, int ifindex, bool rtnl_held,
1189 			    struct netlink_ext_ack *extack)
1190 {
1191 	const struct Qdisc_class_ops *cops;
1192 	struct net_device *dev;
1193 	int err = 0;
1194 
1195 	if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1196 		return 0;
1197 
1198 	rcu_read_lock();
1199 
1200 	/* Find link */
1201 	dev = dev_get_by_index_rcu(net, ifindex);
1202 	if (!dev) {
1203 		rcu_read_unlock();
1204 		return -ENODEV;
1205 	}
1206 
1207 	/* Find qdisc */
1208 	if (!*parent) {
1209 		*q = rcu_dereference(dev->qdisc);
1210 		*parent = (*q)->handle;
1211 	} else {
1212 		*q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1213 		if (!*q) {
1214 			NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1215 			err = -EINVAL;
1216 			goto errout_rcu;
1217 		}
1218 	}
1219 
1220 	*q = qdisc_refcount_inc_nz(*q);
1221 	if (!*q) {
1222 		NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1223 		err = -EINVAL;
1224 		goto errout_rcu;
1225 	}
1226 
1227 	/* Is it classful? */
1228 	cops = (*q)->ops->cl_ops;
1229 	if (!cops) {
1230 		NL_SET_ERR_MSG(extack, "Qdisc not classful");
1231 		err = -EINVAL;
1232 		goto errout_qdisc;
1233 	}
1234 
1235 	if (!cops->tcf_block) {
1236 		NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1237 		err = -EOPNOTSUPP;
1238 		goto errout_qdisc;
1239 	}
1240 
1241 errout_rcu:
1242 	/* At this point we know that qdisc is not noop_qdisc,
1243 	 * which means that qdisc holds a reference to net_device
1244 	 * and we hold a reference to qdisc, so it is safe to release
1245 	 * rcu read lock.
1246 	 */
1247 	rcu_read_unlock();
1248 	return err;
1249 
1250 errout_qdisc:
1251 	rcu_read_unlock();
1252 
1253 	if (rtnl_held)
1254 		qdisc_put(*q);
1255 	else
1256 		qdisc_put_unlocked(*q);
1257 	*q = NULL;
1258 
1259 	return err;
1260 }
1261 
1262 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1263 			       int ifindex, struct netlink_ext_ack *extack)
1264 {
1265 	if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1266 		return 0;
1267 
1268 	/* Do we search for filter, attached to class? */
1269 	if (TC_H_MIN(parent)) {
1270 		const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1271 
1272 		*cl = cops->find(q, parent);
1273 		if (*cl == 0) {
1274 			NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1275 			return -ENOENT;
1276 		}
1277 	}
1278 
1279 	return 0;
1280 }
1281 
1282 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1283 					  unsigned long cl, int ifindex,
1284 					  u32 block_index,
1285 					  struct netlink_ext_ack *extack)
1286 {
1287 	struct tcf_block *block;
1288 
1289 	if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1290 		block = tcf_block_refcnt_get(net, block_index);
1291 		if (!block) {
1292 			NL_SET_ERR_MSG(extack, "Block of given index was not found");
1293 			return ERR_PTR(-EINVAL);
1294 		}
1295 	} else {
1296 		const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1297 
1298 		block = cops->tcf_block(q, cl, extack);
1299 		if (!block)
1300 			return ERR_PTR(-EINVAL);
1301 
1302 		if (tcf_block_shared(block)) {
1303 			NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1304 			return ERR_PTR(-EOPNOTSUPP);
1305 		}
1306 
1307 		/* Always take reference to block in order to support execution
1308 		 * of rules update path of cls API without rtnl lock. Caller
1309 		 * must release block when it is finished using it. 'if' block
1310 		 * of this conditional obtain reference to block by calling
1311 		 * tcf_block_refcnt_get().
1312 		 */
1313 		refcount_inc(&block->refcnt);
1314 	}
1315 
1316 	return block;
1317 }
1318 
1319 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1320 			    struct tcf_block_ext_info *ei, bool rtnl_held)
1321 {
1322 	if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1323 		/* Flushing/putting all chains will cause the block to be
1324 		 * deallocated when last chain is freed. However, if chain_list
1325 		 * is empty, block has to be manually deallocated. After block
1326 		 * reference counter reached 0, it is no longer possible to
1327 		 * increment it or add new chains to block.
1328 		 */
1329 		bool free_block = list_empty(&block->chain_list);
1330 
1331 		mutex_unlock(&block->lock);
1332 		if (tcf_block_shared(block))
1333 			tcf_block_remove(block, block->net);
1334 
1335 		if (q)
1336 			tcf_block_offload_unbind(block, q, ei);
1337 
1338 		if (free_block)
1339 			tcf_block_destroy(block);
1340 		else
1341 			tcf_block_flush_all_chains(block, rtnl_held);
1342 	} else if (q) {
1343 		tcf_block_offload_unbind(block, q, ei);
1344 	}
1345 }
1346 
1347 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1348 {
1349 	__tcf_block_put(block, NULL, NULL, rtnl_held);
1350 }
1351 
1352 /* Find tcf block.
1353  * Set q, parent, cl when appropriate.
1354  */
1355 
1356 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1357 					u32 *parent, unsigned long *cl,
1358 					int ifindex, u32 block_index,
1359 					struct netlink_ext_ack *extack)
1360 {
1361 	struct tcf_block *block;
1362 	int err = 0;
1363 
1364 	ASSERT_RTNL();
1365 
1366 	err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1367 	if (err)
1368 		goto errout;
1369 
1370 	err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1371 	if (err)
1372 		goto errout_qdisc;
1373 
1374 	block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1375 	if (IS_ERR(block)) {
1376 		err = PTR_ERR(block);
1377 		goto errout_qdisc;
1378 	}
1379 
1380 	return block;
1381 
1382 errout_qdisc:
1383 	if (*q)
1384 		qdisc_put(*q);
1385 errout:
1386 	*q = NULL;
1387 	return ERR_PTR(err);
1388 }
1389 
1390 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1391 			      bool rtnl_held)
1392 {
1393 	if (!IS_ERR_OR_NULL(block))
1394 		tcf_block_refcnt_put(block, rtnl_held);
1395 
1396 	if (q) {
1397 		if (rtnl_held)
1398 			qdisc_put(q);
1399 		else
1400 			qdisc_put_unlocked(q);
1401 	}
1402 }
1403 
1404 struct tcf_block_owner_item {
1405 	struct list_head list;
1406 	struct Qdisc *q;
1407 	enum flow_block_binder_type binder_type;
1408 };
1409 
1410 static void
1411 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1412 			       struct Qdisc *q,
1413 			       enum flow_block_binder_type binder_type)
1414 {
1415 	if (block->keep_dst &&
1416 	    binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1417 	    binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1418 		netif_keep_dst(qdisc_dev(q));
1419 }
1420 
1421 void tcf_block_netif_keep_dst(struct tcf_block *block)
1422 {
1423 	struct tcf_block_owner_item *item;
1424 
1425 	block->keep_dst = true;
1426 	list_for_each_entry(item, &block->owner_list, list)
1427 		tcf_block_owner_netif_keep_dst(block, item->q,
1428 					       item->binder_type);
1429 }
1430 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1431 
1432 static int tcf_block_owner_add(struct tcf_block *block,
1433 			       struct Qdisc *q,
1434 			       enum flow_block_binder_type binder_type)
1435 {
1436 	struct tcf_block_owner_item *item;
1437 
1438 	item = kmalloc(sizeof(*item), GFP_KERNEL);
1439 	if (!item)
1440 		return -ENOMEM;
1441 	item->q = q;
1442 	item->binder_type = binder_type;
1443 	list_add(&item->list, &block->owner_list);
1444 	return 0;
1445 }
1446 
1447 static void tcf_block_owner_del(struct tcf_block *block,
1448 				struct Qdisc *q,
1449 				enum flow_block_binder_type binder_type)
1450 {
1451 	struct tcf_block_owner_item *item;
1452 
1453 	list_for_each_entry(item, &block->owner_list, list) {
1454 		if (item->q == q && item->binder_type == binder_type) {
1455 			list_del(&item->list);
1456 			kfree(item);
1457 			return;
1458 		}
1459 	}
1460 	WARN_ON(1);
1461 }
1462 
1463 static bool tcf_block_tracks_dev(struct tcf_block *block,
1464 				 struct tcf_block_ext_info *ei)
1465 {
1466 	return tcf_block_shared(block) &&
1467 	       (ei->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS ||
1468 		ei->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS);
1469 }
1470 
1471 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1472 		      struct tcf_block_ext_info *ei,
1473 		      struct netlink_ext_ack *extack)
1474 {
1475 	struct net_device *dev = qdisc_dev(q);
1476 	struct net *net = qdisc_net(q);
1477 	struct tcf_block *block = NULL;
1478 	int err;
1479 
1480 	if (ei->block_index)
1481 		/* block_index not 0 means the shared block is requested */
1482 		block = tcf_block_refcnt_get(net, ei->block_index);
1483 
1484 	if (!block) {
1485 		block = tcf_block_create(net, q, ei->block_index, extack);
1486 		if (IS_ERR(block))
1487 			return PTR_ERR(block);
1488 		if (tcf_block_shared(block)) {
1489 			err = tcf_block_insert(block, net, extack);
1490 			if (err)
1491 				goto err_block_insert;
1492 		}
1493 	}
1494 
1495 	err = tcf_block_owner_add(block, q, ei->binder_type);
1496 	if (err)
1497 		goto err_block_owner_add;
1498 
1499 	tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1500 
1501 	err = tcf_chain0_head_change_cb_add(block, ei, extack);
1502 	if (err)
1503 		goto err_chain0_head_change_cb_add;
1504 
1505 	err = tcf_block_offload_bind(block, q, ei, extack);
1506 	if (err)
1507 		goto err_block_offload_bind;
1508 
1509 	if (tcf_block_tracks_dev(block, ei)) {
1510 		err = xa_insert(&block->ports, dev->ifindex, dev, GFP_KERNEL);
1511 		if (err) {
1512 			NL_SET_ERR_MSG(extack, "block dev insert failed");
1513 			goto err_dev_insert;
1514 		}
1515 	}
1516 
1517 	*p_block = block;
1518 	return 0;
1519 
1520 err_dev_insert:
1521 	tcf_block_offload_unbind(block, q, ei);
1522 err_block_offload_bind:
1523 	tcf_chain0_head_change_cb_del(block, ei);
1524 err_chain0_head_change_cb_add:
1525 	tcf_block_owner_del(block, q, ei->binder_type);
1526 err_block_owner_add:
1527 err_block_insert:
1528 	tcf_block_refcnt_put(block, true);
1529 	return err;
1530 }
1531 EXPORT_SYMBOL(tcf_block_get_ext);
1532 
1533 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1534 {
1535 	struct tcf_proto __rcu **p_filter_chain = priv;
1536 
1537 	rcu_assign_pointer(*p_filter_chain, tp_head);
1538 }
1539 
1540 int tcf_block_get(struct tcf_block **p_block,
1541 		  struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1542 		  struct netlink_ext_ack *extack)
1543 {
1544 	struct tcf_block_ext_info ei = {
1545 		.chain_head_change = tcf_chain_head_change_dflt,
1546 		.chain_head_change_priv = p_filter_chain,
1547 	};
1548 
1549 	WARN_ON(!p_filter_chain);
1550 	return tcf_block_get_ext(p_block, q, &ei, extack);
1551 }
1552 EXPORT_SYMBOL(tcf_block_get);
1553 
1554 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1555  * actions should be all removed after flushing.
1556  */
1557 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1558 		       struct tcf_block_ext_info *ei)
1559 {
1560 	struct net_device *dev = qdisc_dev(q);
1561 
1562 	if (!block)
1563 		return;
1564 	if (tcf_block_tracks_dev(block, ei))
1565 		xa_erase(&block->ports, dev->ifindex);
1566 	tcf_chain0_head_change_cb_del(block, ei);
1567 	tcf_block_owner_del(block, q, ei->binder_type);
1568 
1569 	__tcf_block_put(block, q, ei, true);
1570 }
1571 EXPORT_SYMBOL(tcf_block_put_ext);
1572 
1573 void tcf_block_put(struct tcf_block *block)
1574 {
1575 	struct tcf_block_ext_info ei = {0, };
1576 
1577 	if (!block)
1578 		return;
1579 	tcf_block_put_ext(block, block->q, &ei);
1580 }
1581 
1582 EXPORT_SYMBOL(tcf_block_put);
1583 
1584 static int
1585 tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1586 			    void *cb_priv, bool add, bool offload_in_use,
1587 			    struct netlink_ext_ack *extack)
1588 {
1589 	struct tcf_chain *chain, *chain_prev;
1590 	struct tcf_proto *tp, *tp_prev;
1591 	int err;
1592 
1593 	lockdep_assert_held(&block->cb_lock);
1594 
1595 	for (chain = __tcf_get_next_chain(block, NULL);
1596 	     chain;
1597 	     chain_prev = chain,
1598 		     chain = __tcf_get_next_chain(block, chain),
1599 		     tcf_chain_put(chain_prev)) {
1600 		if (chain->tmplt_ops && add)
1601 			chain->tmplt_ops->tmplt_reoffload(chain, true, cb,
1602 							  cb_priv);
1603 		for (tp = __tcf_get_next_proto(chain, NULL); tp;
1604 		     tp_prev = tp,
1605 			     tp = __tcf_get_next_proto(chain, tp),
1606 			     tcf_proto_put(tp_prev, true, NULL)) {
1607 			if (tp->ops->reoffload) {
1608 				err = tp->ops->reoffload(tp, add, cb, cb_priv,
1609 							 extack);
1610 				if (err && add)
1611 					goto err_playback_remove;
1612 			} else if (add && offload_in_use) {
1613 				err = -EOPNOTSUPP;
1614 				NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1615 				goto err_playback_remove;
1616 			}
1617 		}
1618 		if (chain->tmplt_ops && !add)
1619 			chain->tmplt_ops->tmplt_reoffload(chain, false, cb,
1620 							  cb_priv);
1621 	}
1622 
1623 	return 0;
1624 
1625 err_playback_remove:
1626 	tcf_proto_put(tp, true, NULL);
1627 	tcf_chain_put(chain);
1628 	tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1629 				    extack);
1630 	return err;
1631 }
1632 
1633 static int tcf_block_bind(struct tcf_block *block,
1634 			  struct flow_block_offload *bo)
1635 {
1636 	struct flow_block_cb *block_cb, *next;
1637 	int err, i = 0;
1638 
1639 	lockdep_assert_held(&block->cb_lock);
1640 
1641 	list_for_each_entry(block_cb, &bo->cb_list, list) {
1642 		err = tcf_block_playback_offloads(block, block_cb->cb,
1643 						  block_cb->cb_priv, true,
1644 						  tcf_block_offload_in_use(block),
1645 						  bo->extack);
1646 		if (err)
1647 			goto err_unroll;
1648 		if (!bo->unlocked_driver_cb)
1649 			block->lockeddevcnt++;
1650 
1651 		i++;
1652 	}
1653 	list_splice(&bo->cb_list, &block->flow_block.cb_list);
1654 
1655 	return 0;
1656 
1657 err_unroll:
1658 	list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1659 		list_del(&block_cb->driver_list);
1660 		if (i-- > 0) {
1661 			list_del(&block_cb->list);
1662 			tcf_block_playback_offloads(block, block_cb->cb,
1663 						    block_cb->cb_priv, false,
1664 						    tcf_block_offload_in_use(block),
1665 						    NULL);
1666 			if (!bo->unlocked_driver_cb)
1667 				block->lockeddevcnt--;
1668 		}
1669 		flow_block_cb_free(block_cb);
1670 	}
1671 
1672 	return err;
1673 }
1674 
1675 static void tcf_block_unbind(struct tcf_block *block,
1676 			     struct flow_block_offload *bo)
1677 {
1678 	struct flow_block_cb *block_cb, *next;
1679 
1680 	lockdep_assert_held(&block->cb_lock);
1681 
1682 	list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1683 		tcf_block_playback_offloads(block, block_cb->cb,
1684 					    block_cb->cb_priv, false,
1685 					    tcf_block_offload_in_use(block),
1686 					    NULL);
1687 		list_del(&block_cb->list);
1688 		flow_block_cb_free(block_cb);
1689 		if (!bo->unlocked_driver_cb)
1690 			block->lockeddevcnt--;
1691 	}
1692 }
1693 
1694 static int tcf_block_setup(struct tcf_block *block,
1695 			   struct flow_block_offload *bo)
1696 {
1697 	int err;
1698 
1699 	switch (bo->command) {
1700 	case FLOW_BLOCK_BIND:
1701 		err = tcf_block_bind(block, bo);
1702 		break;
1703 	case FLOW_BLOCK_UNBIND:
1704 		err = 0;
1705 		tcf_block_unbind(block, bo);
1706 		break;
1707 	default:
1708 		WARN_ON_ONCE(1);
1709 		err = -EOPNOTSUPP;
1710 	}
1711 
1712 	return err;
1713 }
1714 
1715 /* Main classifier routine: scans classifier chain attached
1716  * to this qdisc, (optionally) tests for protocol and asks
1717  * specific classifiers.
1718  */
1719 static inline int __tcf_classify(struct sk_buff *skb,
1720 				 const struct tcf_proto *tp,
1721 				 const struct tcf_proto *orig_tp,
1722 				 struct tcf_result *res,
1723 				 bool compat_mode,
1724 				 struct tcf_exts_miss_cookie_node *n,
1725 				 int act_index,
1726 				 u32 *last_executed_chain)
1727 {
1728 #ifdef CONFIG_NET_CLS_ACT
1729 	const int max_reclassify_loop = 16;
1730 	const struct tcf_proto *first_tp;
1731 	int limit = 0;
1732 
1733 reclassify:
1734 #endif
1735 	for (; tp; tp = rcu_dereference_bh(tp->next)) {
1736 		__be16 protocol = skb_protocol(skb, false);
1737 		int err = 0;
1738 
1739 		if (n) {
1740 			struct tcf_exts *exts;
1741 
1742 			if (n->tp_prio != tp->prio)
1743 				continue;
1744 
1745 			/* We re-lookup the tp and chain based on index instead
1746 			 * of having hard refs and locks to them, so do a sanity
1747 			 * check if any of tp,chain,exts was replaced by the
1748 			 * time we got here with a cookie from hardware.
1749 			 */
1750 			if (unlikely(n->tp != tp || n->tp->chain != n->chain ||
1751 				     !tp->ops->get_exts)) {
1752 				tcf_set_drop_reason(skb,
1753 						    SKB_DROP_REASON_TC_COOKIE_ERROR);
1754 				return TC_ACT_SHOT;
1755 			}
1756 
1757 			exts = tp->ops->get_exts(tp, n->handle);
1758 			if (unlikely(!exts || n->exts != exts)) {
1759 				tcf_set_drop_reason(skb,
1760 						    SKB_DROP_REASON_TC_COOKIE_ERROR);
1761 				return TC_ACT_SHOT;
1762 			}
1763 
1764 			n = NULL;
1765 			err = tcf_exts_exec_ex(skb, exts, act_index, res);
1766 		} else {
1767 			if (tp->protocol != protocol &&
1768 			    tp->protocol != htons(ETH_P_ALL))
1769 				continue;
1770 
1771 			err = tc_classify(skb, tp, res);
1772 		}
1773 #ifdef CONFIG_NET_CLS_ACT
1774 		if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1775 			first_tp = orig_tp;
1776 			*last_executed_chain = first_tp->chain->index;
1777 			goto reset;
1778 		} else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1779 			first_tp = res->goto_tp;
1780 			*last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1781 			goto reset;
1782 		}
1783 #endif
1784 		if (err >= 0)
1785 			return err;
1786 	}
1787 
1788 	if (unlikely(n)) {
1789 		tcf_set_drop_reason(skb,
1790 				    SKB_DROP_REASON_TC_COOKIE_ERROR);
1791 		return TC_ACT_SHOT;
1792 	}
1793 
1794 	return TC_ACT_UNSPEC; /* signal: continue lookup */
1795 #ifdef CONFIG_NET_CLS_ACT
1796 reset:
1797 	if (unlikely(limit++ >= max_reclassify_loop)) {
1798 		net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1799 				       tp->chain->block->index,
1800 				       tp->prio & 0xffff,
1801 				       ntohs(tp->protocol));
1802 		tcf_set_drop_reason(skb,
1803 				    SKB_DROP_REASON_TC_RECLASSIFY_LOOP);
1804 		return TC_ACT_SHOT;
1805 	}
1806 
1807 	tp = first_tp;
1808 	goto reclassify;
1809 #endif
1810 }
1811 
1812 int tcf_classify(struct sk_buff *skb,
1813 		 const struct tcf_block *block,
1814 		 const struct tcf_proto *tp,
1815 		 struct tcf_result *res, bool compat_mode)
1816 {
1817 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1818 	u32 last_executed_chain = 0;
1819 
1820 	return __tcf_classify(skb, tp, tp, res, compat_mode, NULL, 0,
1821 			      &last_executed_chain);
1822 #else
1823 	u32 last_executed_chain = tp ? tp->chain->index : 0;
1824 	struct tcf_exts_miss_cookie_node *n = NULL;
1825 	const struct tcf_proto *orig_tp = tp;
1826 	struct tc_skb_ext *ext;
1827 	int act_index = 0;
1828 	int ret;
1829 
1830 	if (block) {
1831 		ext = skb_ext_find(skb, TC_SKB_EXT);
1832 
1833 		if (ext && (ext->chain || ext->act_miss)) {
1834 			struct tcf_chain *fchain;
1835 			u32 chain;
1836 
1837 			if (ext->act_miss) {
1838 				n = tcf_exts_miss_cookie_lookup(ext->act_miss_cookie,
1839 								&act_index);
1840 				if (!n) {
1841 					tcf_set_drop_reason(skb,
1842 							    SKB_DROP_REASON_TC_COOKIE_ERROR);
1843 					return TC_ACT_SHOT;
1844 				}
1845 
1846 				chain = n->chain_index;
1847 			} else {
1848 				chain = ext->chain;
1849 			}
1850 
1851 			fchain = tcf_chain_lookup_rcu(block, chain);
1852 			if (!fchain) {
1853 				tcf_set_drop_reason(skb,
1854 						    SKB_DROP_REASON_TC_CHAIN_NOTFOUND);
1855 
1856 				return TC_ACT_SHOT;
1857 			}
1858 
1859 			/* Consume, so cloned/redirect skbs won't inherit ext */
1860 			skb_ext_del(skb, TC_SKB_EXT);
1861 
1862 			tp = rcu_dereference_bh(fchain->filter_chain);
1863 			last_executed_chain = fchain->index;
1864 		}
1865 	}
1866 
1867 	ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode, n, act_index,
1868 			     &last_executed_chain);
1869 
1870 	if (tc_skb_ext_tc_enabled()) {
1871 		/* If we missed on some chain */
1872 		if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1873 			struct tc_skb_cb *cb = tc_skb_cb(skb);
1874 
1875 			ext = tc_skb_ext_alloc(skb);
1876 			if (WARN_ON_ONCE(!ext)) {
1877 				tcf_set_drop_reason(skb, SKB_DROP_REASON_NOMEM);
1878 				return TC_ACT_SHOT;
1879 			}
1880 			ext->chain = last_executed_chain;
1881 			ext->mru = cb->mru;
1882 			ext->post_ct = cb->post_ct;
1883 			ext->post_ct_snat = cb->post_ct_snat;
1884 			ext->post_ct_dnat = cb->post_ct_dnat;
1885 			ext->zone = cb->zone;
1886 		}
1887 	}
1888 
1889 	return ret;
1890 #endif
1891 }
1892 EXPORT_SYMBOL(tcf_classify);
1893 
1894 struct tcf_chain_info {
1895 	struct tcf_proto __rcu **pprev;
1896 	struct tcf_proto __rcu *next;
1897 };
1898 
1899 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1900 					   struct tcf_chain_info *chain_info)
1901 {
1902 	return tcf_chain_dereference(*chain_info->pprev, chain);
1903 }
1904 
1905 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1906 			       struct tcf_chain_info *chain_info,
1907 			       struct tcf_proto *tp)
1908 {
1909 	if (chain->flushing)
1910 		return -EAGAIN;
1911 
1912 	RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1913 	if (*chain_info->pprev == chain->filter_chain)
1914 		tcf_chain0_head_change(chain, tp);
1915 	tcf_proto_get(tp);
1916 	rcu_assign_pointer(*chain_info->pprev, tp);
1917 
1918 	return 0;
1919 }
1920 
1921 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1922 				struct tcf_chain_info *chain_info,
1923 				struct tcf_proto *tp)
1924 {
1925 	struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1926 
1927 	tcf_proto_mark_delete(tp);
1928 	if (tp == chain->filter_chain)
1929 		tcf_chain0_head_change(chain, next);
1930 	RCU_INIT_POINTER(*chain_info->pprev, next);
1931 }
1932 
1933 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1934 					   struct tcf_chain_info *chain_info,
1935 					   u32 protocol, u32 prio,
1936 					   bool prio_allocate,
1937 					   struct netlink_ext_ack *extack);
1938 
1939 /* Try to insert new proto.
1940  * If proto with specified priority already exists, free new proto
1941  * and return existing one.
1942  */
1943 
1944 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1945 						    struct tcf_proto *tp_new,
1946 						    u32 protocol, u32 prio,
1947 						    bool rtnl_held)
1948 {
1949 	struct tcf_chain_info chain_info;
1950 	struct tcf_proto *tp;
1951 	int err = 0;
1952 
1953 	mutex_lock(&chain->filter_chain_lock);
1954 
1955 	if (tcf_proto_exists_destroying(chain, tp_new)) {
1956 		mutex_unlock(&chain->filter_chain_lock);
1957 		tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1958 		return ERR_PTR(-EAGAIN);
1959 	}
1960 
1961 	tp = tcf_chain_tp_find(chain, &chain_info, protocol, prio, false, NULL);
1962 	if (!tp)
1963 		err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1964 	mutex_unlock(&chain->filter_chain_lock);
1965 
1966 	if (tp) {
1967 		tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1968 		tp_new = tp;
1969 	} else if (err) {
1970 		tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1971 		tp_new = ERR_PTR(err);
1972 	}
1973 
1974 	return tp_new;
1975 }
1976 
1977 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1978 				      struct tcf_proto *tp, bool rtnl_held,
1979 				      struct netlink_ext_ack *extack)
1980 {
1981 	struct tcf_chain_info chain_info;
1982 	struct tcf_proto *tp_iter;
1983 	struct tcf_proto **pprev;
1984 	struct tcf_proto *next;
1985 
1986 	mutex_lock(&chain->filter_chain_lock);
1987 
1988 	/* Atomically find and remove tp from chain. */
1989 	for (pprev = &chain->filter_chain;
1990 	     (tp_iter = tcf_chain_dereference(*pprev, chain));
1991 	     pprev = &tp_iter->next) {
1992 		if (tp_iter == tp) {
1993 			chain_info.pprev = pprev;
1994 			chain_info.next = tp_iter->next;
1995 			WARN_ON(tp_iter->deleting);
1996 			break;
1997 		}
1998 	}
1999 	/* Verify that tp still exists and no new filters were inserted
2000 	 * concurrently.
2001 	 * Mark tp for deletion if it is empty.
2002 	 */
2003 	if (!tp_iter || !tcf_proto_check_delete(tp)) {
2004 		mutex_unlock(&chain->filter_chain_lock);
2005 		return;
2006 	}
2007 
2008 	tcf_proto_signal_destroying(chain, tp);
2009 	next = tcf_chain_dereference(chain_info.next, chain);
2010 	if (tp == chain->filter_chain)
2011 		tcf_chain0_head_change(chain, next);
2012 	RCU_INIT_POINTER(*chain_info.pprev, next);
2013 	mutex_unlock(&chain->filter_chain_lock);
2014 
2015 	tcf_proto_put(tp, rtnl_held, extack);
2016 }
2017 
2018 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
2019 					   struct tcf_chain_info *chain_info,
2020 					   u32 protocol, u32 prio,
2021 					   bool prio_allocate,
2022 					   struct netlink_ext_ack *extack)
2023 {
2024 	struct tcf_proto **pprev;
2025 	struct tcf_proto *tp;
2026 
2027 	/* Check the chain for existence of proto-tcf with this priority */
2028 	for (pprev = &chain->filter_chain;
2029 	     (tp = tcf_chain_dereference(*pprev, chain));
2030 	     pprev = &tp->next) {
2031 		if (tp->prio >= prio) {
2032 			if (tp->prio == prio) {
2033 				if (prio_allocate) {
2034 					NL_SET_ERR_MSG(extack, "Lowest ID from auto-alloc range already in use");
2035 					return ERR_PTR(-ENOSPC);
2036 				}
2037 				if (tp->protocol != protocol && protocol) {
2038 					NL_SET_ERR_MSG(extack, "Protocol mismatch for filter with specified priority");
2039 					return ERR_PTR(-EINVAL);
2040 				}
2041 			} else {
2042 				tp = NULL;
2043 			}
2044 			break;
2045 		}
2046 	}
2047 	chain_info->pprev = pprev;
2048 	if (tp) {
2049 		chain_info->next = tp->next;
2050 		tcf_proto_get(tp);
2051 	} else {
2052 		chain_info->next = NULL;
2053 	}
2054 	return tp;
2055 }
2056 
2057 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
2058 			 struct tcf_proto *tp, struct tcf_block *block,
2059 			 struct Qdisc *q, u32 parent, void *fh,
2060 			 u32 portid, u32 seq, u16 flags, int event,
2061 			 bool terse_dump, bool rtnl_held,
2062 			 struct netlink_ext_ack *extack)
2063 {
2064 	struct tcmsg *tcm;
2065 	struct nlmsghdr  *nlh;
2066 	unsigned char *b = skb_tail_pointer(skb);
2067 
2068 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2069 	if (!nlh)
2070 		goto out_nlmsg_trim;
2071 	tcm = nlmsg_data(nlh);
2072 	tcm->tcm_family = AF_UNSPEC;
2073 	tcm->tcm__pad1 = 0;
2074 	tcm->tcm__pad2 = 0;
2075 	if (q) {
2076 		tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
2077 		tcm->tcm_parent = parent;
2078 	} else {
2079 		tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2080 		tcm->tcm_block_index = block->index;
2081 	}
2082 	tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
2083 	if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
2084 		goto nla_put_failure;
2085 	if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
2086 		goto nla_put_failure;
2087 	if (!fh) {
2088 		tcm->tcm_handle = 0;
2089 	} else if (terse_dump) {
2090 		if (tp->ops->terse_dump) {
2091 			if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
2092 						rtnl_held) < 0)
2093 				goto nla_put_failure;
2094 		} else {
2095 			goto cls_op_not_supp;
2096 		}
2097 	} else {
2098 		if (tp->ops->dump &&
2099 		    tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
2100 			goto nla_put_failure;
2101 	}
2102 
2103 	if (extack && extack->_msg &&
2104 	    nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2105 		goto nla_put_failure;
2106 
2107 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2108 
2109 	return skb->len;
2110 
2111 out_nlmsg_trim:
2112 nla_put_failure:
2113 cls_op_not_supp:
2114 	nlmsg_trim(skb, b);
2115 	return -1;
2116 }
2117 
2118 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
2119 			  struct nlmsghdr *n, struct tcf_proto *tp,
2120 			  struct tcf_block *block, struct Qdisc *q,
2121 			  u32 parent, void *fh, int event, bool unicast,
2122 			  bool rtnl_held, struct netlink_ext_ack *extack)
2123 {
2124 	struct sk_buff *skb;
2125 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2126 	int err = 0;
2127 
2128 	if (!unicast && !rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2129 		return 0;
2130 
2131 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2132 	if (!skb)
2133 		return -ENOBUFS;
2134 
2135 	if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2136 			  n->nlmsg_seq, n->nlmsg_flags, event,
2137 			  false, rtnl_held, extack) <= 0) {
2138 		kfree_skb(skb);
2139 		return -EINVAL;
2140 	}
2141 
2142 	if (unicast)
2143 		err = rtnl_unicast(skb, net, portid);
2144 	else
2145 		err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2146 				     n->nlmsg_flags & NLM_F_ECHO);
2147 	return err;
2148 }
2149 
2150 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
2151 			      struct nlmsghdr *n, struct tcf_proto *tp,
2152 			      struct tcf_block *block, struct Qdisc *q,
2153 			      u32 parent, void *fh, bool *last, bool rtnl_held,
2154 			      struct netlink_ext_ack *extack)
2155 {
2156 	struct sk_buff *skb;
2157 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2158 	int err;
2159 
2160 	if (!rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2161 		return tp->ops->delete(tp, fh, last, rtnl_held, extack);
2162 
2163 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2164 	if (!skb)
2165 		return -ENOBUFS;
2166 
2167 	if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2168 			  n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
2169 			  false, rtnl_held, extack) <= 0) {
2170 		NL_SET_ERR_MSG(extack, "Failed to build del event notification");
2171 		kfree_skb(skb);
2172 		return -EINVAL;
2173 	}
2174 
2175 	err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
2176 	if (err) {
2177 		kfree_skb(skb);
2178 		return err;
2179 	}
2180 
2181 	err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2182 			     n->nlmsg_flags & NLM_F_ECHO);
2183 	if (err < 0)
2184 		NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
2185 
2186 	return err;
2187 }
2188 
2189 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
2190 				 struct tcf_block *block, struct Qdisc *q,
2191 				 u32 parent, struct nlmsghdr *n,
2192 				 struct tcf_chain *chain, int event,
2193 				 struct netlink_ext_ack *extack)
2194 {
2195 	struct tcf_proto *tp;
2196 
2197 	for (tp = tcf_get_next_proto(chain, NULL);
2198 	     tp; tp = tcf_get_next_proto(chain, tp))
2199 		tfilter_notify(net, oskb, n, tp, block, q, parent, NULL,
2200 			       event, false, true, extack);
2201 }
2202 
2203 static void tfilter_put(struct tcf_proto *tp, void *fh)
2204 {
2205 	if (tp->ops->put && fh)
2206 		tp->ops->put(tp, fh);
2207 }
2208 
2209 static bool is_qdisc_ingress(__u32 classid)
2210 {
2211 	return (TC_H_MIN(classid) == TC_H_MIN(TC_H_MIN_INGRESS));
2212 }
2213 
2214 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2215 			  struct netlink_ext_ack *extack)
2216 {
2217 	struct net *net = sock_net(skb->sk);
2218 	struct nlattr *tca[TCA_MAX + 1];
2219 	char name[IFNAMSIZ];
2220 	struct tcmsg *t;
2221 	u32 protocol;
2222 	u32 prio;
2223 	bool prio_allocate;
2224 	u32 parent;
2225 	u32 chain_index;
2226 	struct Qdisc *q;
2227 	struct tcf_chain_info chain_info;
2228 	struct tcf_chain *chain;
2229 	struct tcf_block *block;
2230 	struct tcf_proto *tp;
2231 	unsigned long cl;
2232 	void *fh;
2233 	int err;
2234 	int tp_created;
2235 	bool rtnl_held = false;
2236 	u32 flags;
2237 
2238 replay:
2239 	tp_created = 0;
2240 
2241 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2242 				     rtm_tca_policy, extack);
2243 	if (err < 0)
2244 		return err;
2245 
2246 	t = nlmsg_data(n);
2247 	protocol = TC_H_MIN(t->tcm_info);
2248 	prio = TC_H_MAJ(t->tcm_info);
2249 	prio_allocate = false;
2250 	parent = t->tcm_parent;
2251 	tp = NULL;
2252 	cl = 0;
2253 	block = NULL;
2254 	q = NULL;
2255 	chain = NULL;
2256 	flags = 0;
2257 
2258 	if (prio == 0) {
2259 		/* If no priority is provided by the user,
2260 		 * we allocate one.
2261 		 */
2262 		if (n->nlmsg_flags & NLM_F_CREATE) {
2263 			prio = TC_H_MAKE(0x80000000U, 0U);
2264 			prio_allocate = true;
2265 		} else {
2266 			NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2267 			return -ENOENT;
2268 		}
2269 	}
2270 
2271 	/* Find head of filter chain. */
2272 
2273 	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2274 	if (err)
2275 		return err;
2276 
2277 	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2278 		NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2279 		err = -EINVAL;
2280 		goto errout;
2281 	}
2282 
2283 	/* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2284 	 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2285 	 * type is not specified, classifier is not unlocked.
2286 	 */
2287 	if (rtnl_held ||
2288 	    (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2289 	    !tcf_proto_is_unlocked(name)) {
2290 		rtnl_held = true;
2291 		rtnl_lock();
2292 	}
2293 
2294 	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2295 	if (err)
2296 		goto errout;
2297 
2298 	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2299 				 extack);
2300 	if (IS_ERR(block)) {
2301 		err = PTR_ERR(block);
2302 		goto errout;
2303 	}
2304 	block->classid = parent;
2305 
2306 	chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
2307 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2308 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2309 		err = -EINVAL;
2310 		goto errout;
2311 	}
2312 	chain = tcf_chain_get(block, chain_index, true);
2313 	if (!chain) {
2314 		NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2315 		err = -ENOMEM;
2316 		goto errout;
2317 	}
2318 
2319 	mutex_lock(&chain->filter_chain_lock);
2320 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2321 			       prio, prio_allocate, extack);
2322 	if (IS_ERR(tp)) {
2323 		err = PTR_ERR(tp);
2324 		goto errout_locked;
2325 	}
2326 
2327 	if (tp == NULL) {
2328 		struct tcf_proto *tp_new = NULL;
2329 
2330 		if (chain->flushing) {
2331 			err = -EAGAIN;
2332 			goto errout_locked;
2333 		}
2334 
2335 		/* Proto-tcf does not exist, create new one */
2336 
2337 		if (tca[TCA_KIND] == NULL || !protocol) {
2338 			NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2339 			err = -EINVAL;
2340 			goto errout_locked;
2341 		}
2342 
2343 		if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2344 			NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2345 			err = -ENOENT;
2346 			goto errout_locked;
2347 		}
2348 
2349 		if (prio_allocate)
2350 			prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2351 							       &chain_info));
2352 
2353 		mutex_unlock(&chain->filter_chain_lock);
2354 		tp_new = tcf_proto_create(name, protocol, prio, chain,
2355 					  rtnl_held, extack);
2356 		if (IS_ERR(tp_new)) {
2357 			err = PTR_ERR(tp_new);
2358 			goto errout_tp;
2359 		}
2360 
2361 		tp_created = 1;
2362 		tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2363 						rtnl_held);
2364 		if (IS_ERR(tp)) {
2365 			err = PTR_ERR(tp);
2366 			goto errout_tp;
2367 		}
2368 	} else {
2369 		mutex_unlock(&chain->filter_chain_lock);
2370 	}
2371 
2372 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2373 		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2374 		err = -EINVAL;
2375 		goto errout;
2376 	}
2377 
2378 	fh = tp->ops->get(tp, t->tcm_handle);
2379 
2380 	if (!fh) {
2381 		if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2382 			NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2383 			err = -ENOENT;
2384 			goto errout;
2385 		}
2386 	} else if (n->nlmsg_flags & NLM_F_EXCL) {
2387 		tfilter_put(tp, fh);
2388 		NL_SET_ERR_MSG(extack, "Filter already exists");
2389 		err = -EEXIST;
2390 		goto errout;
2391 	}
2392 
2393 	if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2394 		tfilter_put(tp, fh);
2395 		NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2396 		err = -EINVAL;
2397 		goto errout;
2398 	}
2399 
2400 	if (!(n->nlmsg_flags & NLM_F_CREATE))
2401 		flags |= TCA_ACT_FLAGS_REPLACE;
2402 	if (!rtnl_held)
2403 		flags |= TCA_ACT_FLAGS_NO_RTNL;
2404 	if (is_qdisc_ingress(parent))
2405 		flags |= TCA_ACT_FLAGS_AT_INGRESS;
2406 	err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2407 			      flags, extack);
2408 	if (err == 0) {
2409 		tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2410 			       RTM_NEWTFILTER, false, rtnl_held, extack);
2411 		tfilter_put(tp, fh);
2412 		tcf_block_filter_cnt_update(block, &tp->counted, true);
2413 		/* q pointer is NULL for shared blocks */
2414 		if (q)
2415 			q->flags &= ~TCQ_F_CAN_BYPASS;
2416 	}
2417 
2418 errout:
2419 	if (err && tp_created)
2420 		tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2421 errout_tp:
2422 	if (chain) {
2423 		if (tp && !IS_ERR(tp))
2424 			tcf_proto_put(tp, rtnl_held, NULL);
2425 		if (!tp_created)
2426 			tcf_chain_put(chain);
2427 	}
2428 	tcf_block_release(q, block, rtnl_held);
2429 
2430 	if (rtnl_held)
2431 		rtnl_unlock();
2432 
2433 	if (err == -EAGAIN) {
2434 		/* Take rtnl lock in case EAGAIN is caused by concurrent flush
2435 		 * of target chain.
2436 		 */
2437 		rtnl_held = true;
2438 		/* Replay the request. */
2439 		goto replay;
2440 	}
2441 	return err;
2442 
2443 errout_locked:
2444 	mutex_unlock(&chain->filter_chain_lock);
2445 	goto errout;
2446 }
2447 
2448 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2449 			  struct netlink_ext_ack *extack)
2450 {
2451 	struct net *net = sock_net(skb->sk);
2452 	struct nlattr *tca[TCA_MAX + 1];
2453 	char name[IFNAMSIZ];
2454 	struct tcmsg *t;
2455 	u32 protocol;
2456 	u32 prio;
2457 	u32 parent;
2458 	u32 chain_index;
2459 	struct Qdisc *q = NULL;
2460 	struct tcf_chain_info chain_info;
2461 	struct tcf_chain *chain = NULL;
2462 	struct tcf_block *block = NULL;
2463 	struct tcf_proto *tp = NULL;
2464 	unsigned long cl = 0;
2465 	void *fh = NULL;
2466 	int err;
2467 	bool rtnl_held = false;
2468 
2469 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2470 				     rtm_tca_policy, extack);
2471 	if (err < 0)
2472 		return err;
2473 
2474 	t = nlmsg_data(n);
2475 	protocol = TC_H_MIN(t->tcm_info);
2476 	prio = TC_H_MAJ(t->tcm_info);
2477 	parent = t->tcm_parent;
2478 
2479 	if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2480 		NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2481 		return -ENOENT;
2482 	}
2483 
2484 	/* Find head of filter chain. */
2485 
2486 	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2487 	if (err)
2488 		return err;
2489 
2490 	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2491 		NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2492 		err = -EINVAL;
2493 		goto errout;
2494 	}
2495 	/* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2496 	 * found), qdisc is not unlocked, classifier type is not specified,
2497 	 * classifier is not unlocked.
2498 	 */
2499 	if (!prio ||
2500 	    (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2501 	    !tcf_proto_is_unlocked(name)) {
2502 		rtnl_held = true;
2503 		rtnl_lock();
2504 	}
2505 
2506 	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2507 	if (err)
2508 		goto errout;
2509 
2510 	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2511 				 extack);
2512 	if (IS_ERR(block)) {
2513 		err = PTR_ERR(block);
2514 		goto errout;
2515 	}
2516 
2517 	chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
2518 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2519 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2520 		err = -EINVAL;
2521 		goto errout;
2522 	}
2523 	chain = tcf_chain_get(block, chain_index, false);
2524 	if (!chain) {
2525 		/* User requested flush on non-existent chain. Nothing to do,
2526 		 * so just return success.
2527 		 */
2528 		if (prio == 0) {
2529 			err = 0;
2530 			goto errout;
2531 		}
2532 		NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2533 		err = -ENOENT;
2534 		goto errout;
2535 	}
2536 
2537 	if (prio == 0) {
2538 		tfilter_notify_chain(net, skb, block, q, parent, n,
2539 				     chain, RTM_DELTFILTER, extack);
2540 		tcf_chain_flush(chain, rtnl_held);
2541 		err = 0;
2542 		goto errout;
2543 	}
2544 
2545 	mutex_lock(&chain->filter_chain_lock);
2546 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2547 			       prio, false, extack);
2548 	if (!tp) {
2549 		err = -ENOENT;
2550 		NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2551 		goto errout_locked;
2552 	} else if (IS_ERR(tp)) {
2553 		err = PTR_ERR(tp);
2554 		goto errout_locked;
2555 	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2556 		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2557 		err = -EINVAL;
2558 		goto errout_locked;
2559 	} else if (t->tcm_handle == 0) {
2560 		tcf_proto_signal_destroying(chain, tp);
2561 		tcf_chain_tp_remove(chain, &chain_info, tp);
2562 		mutex_unlock(&chain->filter_chain_lock);
2563 
2564 		tcf_proto_put(tp, rtnl_held, NULL);
2565 		tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2566 			       RTM_DELTFILTER, false, rtnl_held, extack);
2567 		err = 0;
2568 		goto errout;
2569 	}
2570 	mutex_unlock(&chain->filter_chain_lock);
2571 
2572 	fh = tp->ops->get(tp, t->tcm_handle);
2573 
2574 	if (!fh) {
2575 		NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2576 		err = -ENOENT;
2577 	} else {
2578 		bool last;
2579 
2580 		err = tfilter_del_notify(net, skb, n, tp, block, q, parent, fh,
2581 					 &last, rtnl_held, extack);
2582 
2583 		if (err)
2584 			goto errout;
2585 		if (last)
2586 			tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2587 	}
2588 
2589 errout:
2590 	if (chain) {
2591 		if (tp && !IS_ERR(tp))
2592 			tcf_proto_put(tp, rtnl_held, NULL);
2593 		tcf_chain_put(chain);
2594 	}
2595 	tcf_block_release(q, block, rtnl_held);
2596 
2597 	if (rtnl_held)
2598 		rtnl_unlock();
2599 
2600 	return err;
2601 
2602 errout_locked:
2603 	mutex_unlock(&chain->filter_chain_lock);
2604 	goto errout;
2605 }
2606 
2607 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2608 			  struct netlink_ext_ack *extack)
2609 {
2610 	struct net *net = sock_net(skb->sk);
2611 	struct nlattr *tca[TCA_MAX + 1];
2612 	char name[IFNAMSIZ];
2613 	struct tcmsg *t;
2614 	u32 protocol;
2615 	u32 prio;
2616 	u32 parent;
2617 	u32 chain_index;
2618 	struct Qdisc *q = NULL;
2619 	struct tcf_chain_info chain_info;
2620 	struct tcf_chain *chain = NULL;
2621 	struct tcf_block *block = NULL;
2622 	struct tcf_proto *tp = NULL;
2623 	unsigned long cl = 0;
2624 	void *fh = NULL;
2625 	int err;
2626 	bool rtnl_held = false;
2627 
2628 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2629 				     rtm_tca_policy, extack);
2630 	if (err < 0)
2631 		return err;
2632 
2633 	t = nlmsg_data(n);
2634 	protocol = TC_H_MIN(t->tcm_info);
2635 	prio = TC_H_MAJ(t->tcm_info);
2636 	parent = t->tcm_parent;
2637 
2638 	if (prio == 0) {
2639 		NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2640 		return -ENOENT;
2641 	}
2642 
2643 	/* Find head of filter chain. */
2644 
2645 	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2646 	if (err)
2647 		return err;
2648 
2649 	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2650 		NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2651 		err = -EINVAL;
2652 		goto errout;
2653 	}
2654 	/* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2655 	 * unlocked, classifier type is not specified, classifier is not
2656 	 * unlocked.
2657 	 */
2658 	if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2659 	    !tcf_proto_is_unlocked(name)) {
2660 		rtnl_held = true;
2661 		rtnl_lock();
2662 	}
2663 
2664 	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2665 	if (err)
2666 		goto errout;
2667 
2668 	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2669 				 extack);
2670 	if (IS_ERR(block)) {
2671 		err = PTR_ERR(block);
2672 		goto errout;
2673 	}
2674 
2675 	chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
2676 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2677 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2678 		err = -EINVAL;
2679 		goto errout;
2680 	}
2681 	chain = tcf_chain_get(block, chain_index, false);
2682 	if (!chain) {
2683 		NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2684 		err = -EINVAL;
2685 		goto errout;
2686 	}
2687 
2688 	mutex_lock(&chain->filter_chain_lock);
2689 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2690 			       prio, false, extack);
2691 	mutex_unlock(&chain->filter_chain_lock);
2692 	if (!tp) {
2693 		err = -ENOENT;
2694 		NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2695 		goto errout;
2696 	} else if (IS_ERR(tp)) {
2697 		err = PTR_ERR(tp);
2698 		goto errout;
2699 	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2700 		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2701 		err = -EINVAL;
2702 		goto errout;
2703 	}
2704 
2705 	fh = tp->ops->get(tp, t->tcm_handle);
2706 
2707 	if (!fh) {
2708 		NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2709 		err = -ENOENT;
2710 	} else {
2711 		err = tfilter_notify(net, skb, n, tp, block, q, parent,
2712 				     fh, RTM_NEWTFILTER, true, rtnl_held, NULL);
2713 		if (err < 0)
2714 			NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2715 	}
2716 
2717 	tfilter_put(tp, fh);
2718 errout:
2719 	if (chain) {
2720 		if (tp && !IS_ERR(tp))
2721 			tcf_proto_put(tp, rtnl_held, NULL);
2722 		tcf_chain_put(chain);
2723 	}
2724 	tcf_block_release(q, block, rtnl_held);
2725 
2726 	if (rtnl_held)
2727 		rtnl_unlock();
2728 
2729 	return err;
2730 }
2731 
2732 struct tcf_dump_args {
2733 	struct tcf_walker w;
2734 	struct sk_buff *skb;
2735 	struct netlink_callback *cb;
2736 	struct tcf_block *block;
2737 	struct Qdisc *q;
2738 	u32 parent;
2739 	bool terse_dump;
2740 };
2741 
2742 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2743 {
2744 	struct tcf_dump_args *a = (void *)arg;
2745 	struct net *net = sock_net(a->skb->sk);
2746 
2747 	return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2748 			     n, NETLINK_CB(a->cb->skb).portid,
2749 			     a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2750 			     RTM_NEWTFILTER, a->terse_dump, true, NULL);
2751 }
2752 
2753 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2754 			   struct sk_buff *skb, struct netlink_callback *cb,
2755 			   long index_start, long *p_index, bool terse)
2756 {
2757 	struct net *net = sock_net(skb->sk);
2758 	struct tcf_block *block = chain->block;
2759 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2760 	struct tcf_proto *tp, *tp_prev;
2761 	struct tcf_dump_args arg;
2762 
2763 	for (tp = __tcf_get_next_proto(chain, NULL);
2764 	     tp;
2765 	     tp_prev = tp,
2766 		     tp = __tcf_get_next_proto(chain, tp),
2767 		     tcf_proto_put(tp_prev, true, NULL),
2768 		     (*p_index)++) {
2769 		if (*p_index < index_start)
2770 			continue;
2771 		if (TC_H_MAJ(tcm->tcm_info) &&
2772 		    TC_H_MAJ(tcm->tcm_info) != tp->prio)
2773 			continue;
2774 		if (TC_H_MIN(tcm->tcm_info) &&
2775 		    TC_H_MIN(tcm->tcm_info) != tp->protocol)
2776 			continue;
2777 		if (*p_index > index_start)
2778 			memset(&cb->args[1], 0,
2779 			       sizeof(cb->args) - sizeof(cb->args[0]));
2780 		if (cb->args[1] == 0) {
2781 			if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2782 					  NETLINK_CB(cb->skb).portid,
2783 					  cb->nlh->nlmsg_seq, NLM_F_MULTI,
2784 					  RTM_NEWTFILTER, false, true, NULL) <= 0)
2785 				goto errout;
2786 			cb->args[1] = 1;
2787 		}
2788 		if (!tp->ops->walk)
2789 			continue;
2790 		arg.w.fn = tcf_node_dump;
2791 		arg.skb = skb;
2792 		arg.cb = cb;
2793 		arg.block = block;
2794 		arg.q = q;
2795 		arg.parent = parent;
2796 		arg.w.stop = 0;
2797 		arg.w.skip = cb->args[1] - 1;
2798 		arg.w.count = 0;
2799 		arg.w.cookie = cb->args[2];
2800 		arg.terse_dump = terse;
2801 		tp->ops->walk(tp, &arg.w, true);
2802 		cb->args[2] = arg.w.cookie;
2803 		cb->args[1] = arg.w.count + 1;
2804 		if (arg.w.stop)
2805 			goto errout;
2806 	}
2807 	return true;
2808 
2809 errout:
2810 	tcf_proto_put(tp, true, NULL);
2811 	return false;
2812 }
2813 
2814 static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2815 	[TCA_CHAIN]      = { .type = NLA_U32 },
2816 	[TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2817 };
2818 
2819 /* called with RTNL */
2820 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2821 {
2822 	struct tcf_chain *chain, *chain_prev;
2823 	struct net *net = sock_net(skb->sk);
2824 	struct nlattr *tca[TCA_MAX + 1];
2825 	struct Qdisc *q = NULL;
2826 	struct tcf_block *block;
2827 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2828 	bool terse_dump = false;
2829 	long index_start;
2830 	long index;
2831 	u32 parent;
2832 	int err;
2833 
2834 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2835 		return skb->len;
2836 
2837 	err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2838 				     tcf_tfilter_dump_policy, cb->extack);
2839 	if (err)
2840 		return err;
2841 
2842 	if (tca[TCA_DUMP_FLAGS]) {
2843 		struct nla_bitfield32 flags =
2844 			nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2845 
2846 		terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2847 	}
2848 
2849 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2850 		block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2851 		if (!block)
2852 			goto out;
2853 		/* If we work with block index, q is NULL and parent value
2854 		 * will never be used in the following code. The check
2855 		 * in tcf_fill_node prevents it. However, compiler does not
2856 		 * see that far, so set parent to zero to silence the warning
2857 		 * about parent being uninitialized.
2858 		 */
2859 		parent = 0;
2860 	} else {
2861 		const struct Qdisc_class_ops *cops;
2862 		struct net_device *dev;
2863 		unsigned long cl = 0;
2864 
2865 		dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2866 		if (!dev)
2867 			return skb->len;
2868 
2869 		parent = tcm->tcm_parent;
2870 		if (!parent)
2871 			q = rtnl_dereference(dev->qdisc);
2872 		else
2873 			q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2874 		if (!q)
2875 			goto out;
2876 		cops = q->ops->cl_ops;
2877 		if (!cops)
2878 			goto out;
2879 		if (!cops->tcf_block)
2880 			goto out;
2881 		if (TC_H_MIN(tcm->tcm_parent)) {
2882 			cl = cops->find(q, tcm->tcm_parent);
2883 			if (cl == 0)
2884 				goto out;
2885 		}
2886 		block = cops->tcf_block(q, cl, NULL);
2887 		if (!block)
2888 			goto out;
2889 		parent = block->classid;
2890 		if (tcf_block_shared(block))
2891 			q = NULL;
2892 	}
2893 
2894 	index_start = cb->args[0];
2895 	index = 0;
2896 
2897 	for (chain = __tcf_get_next_chain(block, NULL);
2898 	     chain;
2899 	     chain_prev = chain,
2900 		     chain = __tcf_get_next_chain(block, chain),
2901 		     tcf_chain_put(chain_prev)) {
2902 		if (tca[TCA_CHAIN] &&
2903 		    nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2904 			continue;
2905 		if (!tcf_chain_dump(chain, q, parent, skb, cb,
2906 				    index_start, &index, terse_dump)) {
2907 			tcf_chain_put(chain);
2908 			err = -EMSGSIZE;
2909 			break;
2910 		}
2911 	}
2912 
2913 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2914 		tcf_block_refcnt_put(block, true);
2915 	cb->args[0] = index;
2916 
2917 out:
2918 	/* If we did no progress, the error (EMSGSIZE) is real */
2919 	if (skb->len == 0 && err)
2920 		return err;
2921 	return skb->len;
2922 }
2923 
2924 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2925 			      void *tmplt_priv, u32 chain_index,
2926 			      struct net *net, struct sk_buff *skb,
2927 			      struct tcf_block *block,
2928 			      u32 portid, u32 seq, u16 flags, int event,
2929 			      struct netlink_ext_ack *extack)
2930 {
2931 	unsigned char *b = skb_tail_pointer(skb);
2932 	const struct tcf_proto_ops *ops;
2933 	struct nlmsghdr *nlh;
2934 	struct tcmsg *tcm;
2935 	void *priv;
2936 
2937 	ops = tmplt_ops;
2938 	priv = tmplt_priv;
2939 
2940 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2941 	if (!nlh)
2942 		goto out_nlmsg_trim;
2943 	tcm = nlmsg_data(nlh);
2944 	tcm->tcm_family = AF_UNSPEC;
2945 	tcm->tcm__pad1 = 0;
2946 	tcm->tcm__pad2 = 0;
2947 	tcm->tcm_handle = 0;
2948 	if (block->q) {
2949 		tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2950 		tcm->tcm_parent = block->q->handle;
2951 	} else {
2952 		tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2953 		tcm->tcm_block_index = block->index;
2954 	}
2955 
2956 	if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2957 		goto nla_put_failure;
2958 
2959 	if (ops) {
2960 		if (nla_put_string(skb, TCA_KIND, ops->kind))
2961 			goto nla_put_failure;
2962 		if (ops->tmplt_dump(skb, net, priv) < 0)
2963 			goto nla_put_failure;
2964 	}
2965 
2966 	if (extack && extack->_msg &&
2967 	    nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2968 		goto out_nlmsg_trim;
2969 
2970 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2971 
2972 	return skb->len;
2973 
2974 out_nlmsg_trim:
2975 nla_put_failure:
2976 	nlmsg_trim(skb, b);
2977 	return -EMSGSIZE;
2978 }
2979 
2980 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2981 			   u32 seq, u16 flags, int event, bool unicast,
2982 			   struct netlink_ext_ack *extack)
2983 {
2984 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2985 	struct tcf_block *block = chain->block;
2986 	struct net *net = block->net;
2987 	struct sk_buff *skb;
2988 	int err = 0;
2989 
2990 	if (!unicast && !rtnl_notify_needed(net, flags, RTNLGRP_TC))
2991 		return 0;
2992 
2993 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2994 	if (!skb)
2995 		return -ENOBUFS;
2996 
2997 	if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2998 			       chain->index, net, skb, block, portid,
2999 			       seq, flags, event, extack) <= 0) {
3000 		kfree_skb(skb);
3001 		return -EINVAL;
3002 	}
3003 
3004 	if (unicast)
3005 		err = rtnl_unicast(skb, net, portid);
3006 	else
3007 		err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
3008 				     flags & NLM_F_ECHO);
3009 
3010 	return err;
3011 }
3012 
3013 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
3014 				  void *tmplt_priv, u32 chain_index,
3015 				  struct tcf_block *block, struct sk_buff *oskb,
3016 				  u32 seq, u16 flags)
3017 {
3018 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
3019 	struct net *net = block->net;
3020 	struct sk_buff *skb;
3021 
3022 	if (!rtnl_notify_needed(net, flags, RTNLGRP_TC))
3023 		return 0;
3024 
3025 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3026 	if (!skb)
3027 		return -ENOBUFS;
3028 
3029 	if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
3030 			       block, portid, seq, flags, RTM_DELCHAIN, NULL) <= 0) {
3031 		kfree_skb(skb);
3032 		return -EINVAL;
3033 	}
3034 
3035 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
3036 }
3037 
3038 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
3039 			      struct nlattr **tca,
3040 			      struct netlink_ext_ack *extack)
3041 {
3042 	const struct tcf_proto_ops *ops;
3043 	char name[IFNAMSIZ];
3044 	void *tmplt_priv;
3045 
3046 	/* If kind is not set, user did not specify template. */
3047 	if (!tca[TCA_KIND])
3048 		return 0;
3049 
3050 	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
3051 		NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
3052 		return -EINVAL;
3053 	}
3054 
3055 	ops = tcf_proto_lookup_ops(name, true, extack);
3056 	if (IS_ERR(ops))
3057 		return PTR_ERR(ops);
3058 	if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump ||
3059 	    !ops->tmplt_reoffload) {
3060 		NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
3061 		module_put(ops->owner);
3062 		return -EOPNOTSUPP;
3063 	}
3064 
3065 	tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
3066 	if (IS_ERR(tmplt_priv)) {
3067 		module_put(ops->owner);
3068 		return PTR_ERR(tmplt_priv);
3069 	}
3070 	chain->tmplt_ops = ops;
3071 	chain->tmplt_priv = tmplt_priv;
3072 	return 0;
3073 }
3074 
3075 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
3076 			       void *tmplt_priv)
3077 {
3078 	/* If template ops are set, no work to do for us. */
3079 	if (!tmplt_ops)
3080 		return;
3081 
3082 	tmplt_ops->tmplt_destroy(tmplt_priv);
3083 	module_put(tmplt_ops->owner);
3084 }
3085 
3086 /* Add/delete/get a chain */
3087 
3088 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
3089 			struct netlink_ext_ack *extack)
3090 {
3091 	struct net *net = sock_net(skb->sk);
3092 	struct nlattr *tca[TCA_MAX + 1];
3093 	struct tcmsg *t;
3094 	u32 parent;
3095 	u32 chain_index;
3096 	struct Qdisc *q;
3097 	struct tcf_chain *chain;
3098 	struct tcf_block *block;
3099 	unsigned long cl;
3100 	int err;
3101 
3102 replay:
3103 	q = NULL;
3104 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
3105 				     rtm_tca_policy, extack);
3106 	if (err < 0)
3107 		return err;
3108 
3109 	t = nlmsg_data(n);
3110 	parent = t->tcm_parent;
3111 	cl = 0;
3112 
3113 	block = tcf_block_find(net, &q, &parent, &cl,
3114 			       t->tcm_ifindex, t->tcm_block_index, extack);
3115 	if (IS_ERR(block))
3116 		return PTR_ERR(block);
3117 
3118 	chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
3119 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
3120 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
3121 		err = -EINVAL;
3122 		goto errout_block;
3123 	}
3124 
3125 	mutex_lock(&block->lock);
3126 	chain = tcf_chain_lookup(block, chain_index);
3127 	if (n->nlmsg_type == RTM_NEWCHAIN) {
3128 		if (chain) {
3129 			if (tcf_chain_held_by_acts_only(chain)) {
3130 				/* The chain exists only because there is
3131 				 * some action referencing it.
3132 				 */
3133 				tcf_chain_hold(chain);
3134 			} else {
3135 				NL_SET_ERR_MSG(extack, "Filter chain already exists");
3136 				err = -EEXIST;
3137 				goto errout_block_locked;
3138 			}
3139 		} else {
3140 			if (!(n->nlmsg_flags & NLM_F_CREATE)) {
3141 				NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
3142 				err = -ENOENT;
3143 				goto errout_block_locked;
3144 			}
3145 			chain = tcf_chain_create(block, chain_index);
3146 			if (!chain) {
3147 				NL_SET_ERR_MSG(extack, "Failed to create filter chain");
3148 				err = -ENOMEM;
3149 				goto errout_block_locked;
3150 			}
3151 		}
3152 	} else {
3153 		if (!chain || tcf_chain_held_by_acts_only(chain)) {
3154 			NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
3155 			err = -EINVAL;
3156 			goto errout_block_locked;
3157 		}
3158 		tcf_chain_hold(chain);
3159 	}
3160 
3161 	if (n->nlmsg_type == RTM_NEWCHAIN) {
3162 		/* Modifying chain requires holding parent block lock. In case
3163 		 * the chain was successfully added, take a reference to the
3164 		 * chain. This ensures that an empty chain does not disappear at
3165 		 * the end of this function.
3166 		 */
3167 		tcf_chain_hold(chain);
3168 		chain->explicitly_created = true;
3169 	}
3170 	mutex_unlock(&block->lock);
3171 
3172 	switch (n->nlmsg_type) {
3173 	case RTM_NEWCHAIN:
3174 		err = tc_chain_tmplt_add(chain, net, tca, extack);
3175 		if (err) {
3176 			tcf_chain_put_explicitly_created(chain);
3177 			goto errout;
3178 		}
3179 
3180 		tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
3181 				RTM_NEWCHAIN, false, extack);
3182 		break;
3183 	case RTM_DELCHAIN:
3184 		tfilter_notify_chain(net, skb, block, q, parent, n,
3185 				     chain, RTM_DELTFILTER, extack);
3186 		/* Flush the chain first as the user requested chain removal. */
3187 		tcf_chain_flush(chain, true);
3188 		/* In case the chain was successfully deleted, put a reference
3189 		 * to the chain previously taken during addition.
3190 		 */
3191 		tcf_chain_put_explicitly_created(chain);
3192 		break;
3193 	case RTM_GETCHAIN:
3194 		err = tc_chain_notify(chain, skb, n->nlmsg_seq,
3195 				      n->nlmsg_flags, n->nlmsg_type, true, extack);
3196 		if (err < 0)
3197 			NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
3198 		break;
3199 	default:
3200 		err = -EOPNOTSUPP;
3201 		NL_SET_ERR_MSG(extack, "Unsupported message type");
3202 		goto errout;
3203 	}
3204 
3205 errout:
3206 	tcf_chain_put(chain);
3207 errout_block:
3208 	tcf_block_release(q, block, true);
3209 	if (err == -EAGAIN)
3210 		/* Replay the request. */
3211 		goto replay;
3212 	return err;
3213 
3214 errout_block_locked:
3215 	mutex_unlock(&block->lock);
3216 	goto errout_block;
3217 }
3218 
3219 /* called with RTNL */
3220 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
3221 {
3222 	struct net *net = sock_net(skb->sk);
3223 	struct nlattr *tca[TCA_MAX + 1];
3224 	struct Qdisc *q = NULL;
3225 	struct tcf_block *block;
3226 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
3227 	struct tcf_chain *chain;
3228 	long index_start;
3229 	long index;
3230 	int err;
3231 
3232 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
3233 		return skb->len;
3234 
3235 	err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
3236 				     rtm_tca_policy, cb->extack);
3237 	if (err)
3238 		return err;
3239 
3240 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
3241 		block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
3242 		if (!block)
3243 			goto out;
3244 	} else {
3245 		const struct Qdisc_class_ops *cops;
3246 		struct net_device *dev;
3247 		unsigned long cl = 0;
3248 
3249 		dev = __dev_get_by_index(net, tcm->tcm_ifindex);
3250 		if (!dev)
3251 			return skb->len;
3252 
3253 		if (!tcm->tcm_parent)
3254 			q = rtnl_dereference(dev->qdisc);
3255 		else
3256 			q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
3257 
3258 		if (!q)
3259 			goto out;
3260 		cops = q->ops->cl_ops;
3261 		if (!cops)
3262 			goto out;
3263 		if (!cops->tcf_block)
3264 			goto out;
3265 		if (TC_H_MIN(tcm->tcm_parent)) {
3266 			cl = cops->find(q, tcm->tcm_parent);
3267 			if (cl == 0)
3268 				goto out;
3269 		}
3270 		block = cops->tcf_block(q, cl, NULL);
3271 		if (!block)
3272 			goto out;
3273 		if (tcf_block_shared(block))
3274 			q = NULL;
3275 	}
3276 
3277 	index_start = cb->args[0];
3278 	index = 0;
3279 
3280 	mutex_lock(&block->lock);
3281 	list_for_each_entry(chain, &block->chain_list, list) {
3282 		if ((tca[TCA_CHAIN] &&
3283 		     nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3284 			continue;
3285 		if (index < index_start) {
3286 			index++;
3287 			continue;
3288 		}
3289 		if (tcf_chain_held_by_acts_only(chain))
3290 			continue;
3291 		err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3292 					 chain->index, net, skb, block,
3293 					 NETLINK_CB(cb->skb).portid,
3294 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3295 					 RTM_NEWCHAIN, NULL);
3296 		if (err <= 0)
3297 			break;
3298 		index++;
3299 	}
3300 	mutex_unlock(&block->lock);
3301 
3302 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3303 		tcf_block_refcnt_put(block, true);
3304 	cb->args[0] = index;
3305 
3306 out:
3307 	/* If we did no progress, the error (EMSGSIZE) is real */
3308 	if (skb->len == 0 && err)
3309 		return err;
3310 	return skb->len;
3311 }
3312 
3313 int tcf_exts_init_ex(struct tcf_exts *exts, struct net *net, int action,
3314 		     int police, struct tcf_proto *tp, u32 handle,
3315 		     bool use_action_miss)
3316 {
3317 	int err = 0;
3318 
3319 #ifdef CONFIG_NET_CLS_ACT
3320 	exts->type = 0;
3321 	exts->nr_actions = 0;
3322 	exts->miss_cookie_node = NULL;
3323 	/* Note: we do not own yet a reference on net.
3324 	 * This reference might be taken later from tcf_exts_get_net().
3325 	 */
3326 	exts->net = net;
3327 	exts->actions = kcalloc(TCA_ACT_MAX_PRIO, sizeof(struct tc_action *),
3328 				GFP_KERNEL);
3329 	if (!exts->actions)
3330 		return -ENOMEM;
3331 #endif
3332 
3333 	exts->action = action;
3334 	exts->police = police;
3335 
3336 	if (!use_action_miss)
3337 		return 0;
3338 
3339 	err = tcf_exts_miss_cookie_base_alloc(exts, tp, handle);
3340 	if (err)
3341 		goto err_miss_alloc;
3342 
3343 	return 0;
3344 
3345 err_miss_alloc:
3346 	tcf_exts_destroy(exts);
3347 #ifdef CONFIG_NET_CLS_ACT
3348 	exts->actions = NULL;
3349 #endif
3350 	return err;
3351 }
3352 EXPORT_SYMBOL(tcf_exts_init_ex);
3353 
3354 void tcf_exts_destroy(struct tcf_exts *exts)
3355 {
3356 	tcf_exts_miss_cookie_base_destroy(exts);
3357 
3358 #ifdef CONFIG_NET_CLS_ACT
3359 	if (exts->actions) {
3360 		tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3361 		kfree(exts->actions);
3362 	}
3363 	exts->nr_actions = 0;
3364 #endif
3365 }
3366 EXPORT_SYMBOL(tcf_exts_destroy);
3367 
3368 int tcf_exts_validate_ex(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3369 			 struct nlattr *rate_tlv, struct tcf_exts *exts,
3370 			 u32 flags, u32 fl_flags, struct netlink_ext_ack *extack)
3371 {
3372 #ifdef CONFIG_NET_CLS_ACT
3373 	{
3374 		int init_res[TCA_ACT_MAX_PRIO] = {};
3375 		struct tc_action *act;
3376 		size_t attr_size = 0;
3377 
3378 		if (exts->police && tb[exts->police]) {
3379 			struct tc_action_ops *a_o;
3380 
3381 			flags |= TCA_ACT_FLAGS_POLICE | TCA_ACT_FLAGS_BIND;
3382 			a_o = tc_action_load_ops(tb[exts->police], flags,
3383 						 extack);
3384 			if (IS_ERR(a_o))
3385 				return PTR_ERR(a_o);
3386 			act = tcf_action_init_1(net, tp, tb[exts->police],
3387 						rate_tlv, a_o, init_res, flags,
3388 						extack);
3389 			module_put(a_o->owner);
3390 			if (IS_ERR(act))
3391 				return PTR_ERR(act);
3392 
3393 			act->type = exts->type = TCA_OLD_COMPAT;
3394 			exts->actions[0] = act;
3395 			exts->nr_actions = 1;
3396 			tcf_idr_insert_many(exts->actions, init_res);
3397 		} else if (exts->action && tb[exts->action]) {
3398 			int err;
3399 
3400 			flags |= TCA_ACT_FLAGS_BIND;
3401 			err = tcf_action_init(net, tp, tb[exts->action],
3402 					      rate_tlv, exts->actions, init_res,
3403 					      &attr_size, flags, fl_flags,
3404 					      extack);
3405 			if (err < 0)
3406 				return err;
3407 			exts->nr_actions = err;
3408 		}
3409 	}
3410 #else
3411 	if ((exts->action && tb[exts->action]) ||
3412 	    (exts->police && tb[exts->police])) {
3413 		NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3414 		return -EOPNOTSUPP;
3415 	}
3416 #endif
3417 
3418 	return 0;
3419 }
3420 EXPORT_SYMBOL(tcf_exts_validate_ex);
3421 
3422 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3423 		      struct nlattr *rate_tlv, struct tcf_exts *exts,
3424 		      u32 flags, struct netlink_ext_ack *extack)
3425 {
3426 	return tcf_exts_validate_ex(net, tp, tb, rate_tlv, exts,
3427 				    flags, 0, extack);
3428 }
3429 EXPORT_SYMBOL(tcf_exts_validate);
3430 
3431 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3432 {
3433 #ifdef CONFIG_NET_CLS_ACT
3434 	struct tcf_exts old = *dst;
3435 
3436 	*dst = *src;
3437 	tcf_exts_destroy(&old);
3438 #endif
3439 }
3440 EXPORT_SYMBOL(tcf_exts_change);
3441 
3442 #ifdef CONFIG_NET_CLS_ACT
3443 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3444 {
3445 	if (exts->nr_actions == 0)
3446 		return NULL;
3447 	else
3448 		return exts->actions[0];
3449 }
3450 #endif
3451 
3452 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3453 {
3454 #ifdef CONFIG_NET_CLS_ACT
3455 	struct nlattr *nest;
3456 
3457 	if (exts->action && tcf_exts_has_actions(exts)) {
3458 		/*
3459 		 * again for backward compatible mode - we want
3460 		 * to work with both old and new modes of entering
3461 		 * tc data even if iproute2  was newer - jhs
3462 		 */
3463 		if (exts->type != TCA_OLD_COMPAT) {
3464 			nest = nla_nest_start_noflag(skb, exts->action);
3465 			if (nest == NULL)
3466 				goto nla_put_failure;
3467 
3468 			if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3469 			    < 0)
3470 				goto nla_put_failure;
3471 			nla_nest_end(skb, nest);
3472 		} else if (exts->police) {
3473 			struct tc_action *act = tcf_exts_first_act(exts);
3474 			nest = nla_nest_start_noflag(skb, exts->police);
3475 			if (nest == NULL || !act)
3476 				goto nla_put_failure;
3477 			if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3478 				goto nla_put_failure;
3479 			nla_nest_end(skb, nest);
3480 		}
3481 	}
3482 	return 0;
3483 
3484 nla_put_failure:
3485 	nla_nest_cancel(skb, nest);
3486 	return -1;
3487 #else
3488 	return 0;
3489 #endif
3490 }
3491 EXPORT_SYMBOL(tcf_exts_dump);
3492 
3493 int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3494 {
3495 #ifdef CONFIG_NET_CLS_ACT
3496 	struct nlattr *nest;
3497 
3498 	if (!exts->action || !tcf_exts_has_actions(exts))
3499 		return 0;
3500 
3501 	nest = nla_nest_start_noflag(skb, exts->action);
3502 	if (!nest)
3503 		goto nla_put_failure;
3504 
3505 	if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3506 		goto nla_put_failure;
3507 	nla_nest_end(skb, nest);
3508 	return 0;
3509 
3510 nla_put_failure:
3511 	nla_nest_cancel(skb, nest);
3512 	return -1;
3513 #else
3514 	return 0;
3515 #endif
3516 }
3517 EXPORT_SYMBOL(tcf_exts_terse_dump);
3518 
3519 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3520 {
3521 #ifdef CONFIG_NET_CLS_ACT
3522 	struct tc_action *a = tcf_exts_first_act(exts);
3523 	if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3524 		return -1;
3525 #endif
3526 	return 0;
3527 }
3528 EXPORT_SYMBOL(tcf_exts_dump_stats);
3529 
3530 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3531 {
3532 	if (*flags & TCA_CLS_FLAGS_IN_HW)
3533 		return;
3534 	*flags |= TCA_CLS_FLAGS_IN_HW;
3535 	if (tc_skip_sw(*flags))
3536 		atomic_inc(&block->skipswcnt);
3537 	atomic_inc(&block->offloadcnt);
3538 }
3539 
3540 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3541 {
3542 	if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3543 		return;
3544 	*flags &= ~TCA_CLS_FLAGS_IN_HW;
3545 	if (tc_skip_sw(*flags))
3546 		atomic_dec(&block->skipswcnt);
3547 	atomic_dec(&block->offloadcnt);
3548 }
3549 
3550 static void tc_cls_offload_cnt_update(struct tcf_block *block,
3551 				      struct tcf_proto *tp, u32 *cnt,
3552 				      u32 *flags, u32 diff, bool add)
3553 {
3554 	lockdep_assert_held(&block->cb_lock);
3555 
3556 	spin_lock(&tp->lock);
3557 	if (add) {
3558 		if (!*cnt)
3559 			tcf_block_offload_inc(block, flags);
3560 		*cnt += diff;
3561 	} else {
3562 		*cnt -= diff;
3563 		if (!*cnt)
3564 			tcf_block_offload_dec(block, flags);
3565 	}
3566 	spin_unlock(&tp->lock);
3567 }
3568 
3569 static void
3570 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3571 			 u32 *cnt, u32 *flags)
3572 {
3573 	lockdep_assert_held(&block->cb_lock);
3574 
3575 	spin_lock(&tp->lock);
3576 	tcf_block_offload_dec(block, flags);
3577 	*cnt = 0;
3578 	spin_unlock(&tp->lock);
3579 }
3580 
3581 static int
3582 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3583 		   void *type_data, bool err_stop)
3584 {
3585 	struct flow_block_cb *block_cb;
3586 	int ok_count = 0;
3587 	int err;
3588 
3589 	list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3590 		err = block_cb->cb(type, type_data, block_cb->cb_priv);
3591 		if (err) {
3592 			if (err_stop)
3593 				return err;
3594 		} else {
3595 			ok_count++;
3596 		}
3597 	}
3598 	return ok_count;
3599 }
3600 
3601 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3602 		     void *type_data, bool err_stop, bool rtnl_held)
3603 {
3604 	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3605 	int ok_count;
3606 
3607 retry:
3608 	if (take_rtnl)
3609 		rtnl_lock();
3610 	down_read(&block->cb_lock);
3611 	/* Need to obtain rtnl lock if block is bound to devs that require it.
3612 	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3613 	 * obtain the locks in same order here.
3614 	 */
3615 	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3616 		up_read(&block->cb_lock);
3617 		take_rtnl = true;
3618 		goto retry;
3619 	}
3620 
3621 	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3622 
3623 	up_read(&block->cb_lock);
3624 	if (take_rtnl)
3625 		rtnl_unlock();
3626 	return ok_count;
3627 }
3628 EXPORT_SYMBOL(tc_setup_cb_call);
3629 
3630 /* Non-destructive filter add. If filter that wasn't already in hardware is
3631  * successfully offloaded, increment block offloads counter. On failure,
3632  * previously offloaded filter is considered to be intact and offloads counter
3633  * is not decremented.
3634  */
3635 
3636 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3637 		    enum tc_setup_type type, void *type_data, bool err_stop,
3638 		    u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3639 {
3640 	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3641 	int ok_count;
3642 
3643 retry:
3644 	if (take_rtnl)
3645 		rtnl_lock();
3646 	down_read(&block->cb_lock);
3647 	/* Need to obtain rtnl lock if block is bound to devs that require it.
3648 	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3649 	 * obtain the locks in same order here.
3650 	 */
3651 	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3652 		up_read(&block->cb_lock);
3653 		take_rtnl = true;
3654 		goto retry;
3655 	}
3656 
3657 	/* Make sure all netdevs sharing this block are offload-capable. */
3658 	if (block->nooffloaddevcnt && err_stop) {
3659 		ok_count = -EOPNOTSUPP;
3660 		goto err_unlock;
3661 	}
3662 
3663 	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3664 	if (ok_count < 0)
3665 		goto err_unlock;
3666 
3667 	if (tp->ops->hw_add)
3668 		tp->ops->hw_add(tp, type_data);
3669 	if (ok_count > 0)
3670 		tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3671 					  ok_count, true);
3672 err_unlock:
3673 	up_read(&block->cb_lock);
3674 	if (take_rtnl)
3675 		rtnl_unlock();
3676 	return min(ok_count, 0);
3677 }
3678 EXPORT_SYMBOL(tc_setup_cb_add);
3679 
3680 /* Destructive filter replace. If filter that wasn't already in hardware is
3681  * successfully offloaded, increment block offload counter. On failure,
3682  * previously offloaded filter is considered to be destroyed and offload counter
3683  * is decremented.
3684  */
3685 
3686 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3687 			enum tc_setup_type type, void *type_data, bool err_stop,
3688 			u32 *old_flags, unsigned int *old_in_hw_count,
3689 			u32 *new_flags, unsigned int *new_in_hw_count,
3690 			bool rtnl_held)
3691 {
3692 	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3693 	int ok_count;
3694 
3695 retry:
3696 	if (take_rtnl)
3697 		rtnl_lock();
3698 	down_read(&block->cb_lock);
3699 	/* Need to obtain rtnl lock if block is bound to devs that require it.
3700 	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3701 	 * obtain the locks in same order here.
3702 	 */
3703 	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3704 		up_read(&block->cb_lock);
3705 		take_rtnl = true;
3706 		goto retry;
3707 	}
3708 
3709 	/* Make sure all netdevs sharing this block are offload-capable. */
3710 	if (block->nooffloaddevcnt && err_stop) {
3711 		ok_count = -EOPNOTSUPP;
3712 		goto err_unlock;
3713 	}
3714 
3715 	tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3716 	if (tp->ops->hw_del)
3717 		tp->ops->hw_del(tp, type_data);
3718 
3719 	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3720 	if (ok_count < 0)
3721 		goto err_unlock;
3722 
3723 	if (tp->ops->hw_add)
3724 		tp->ops->hw_add(tp, type_data);
3725 	if (ok_count > 0)
3726 		tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3727 					  new_flags, ok_count, true);
3728 err_unlock:
3729 	up_read(&block->cb_lock);
3730 	if (take_rtnl)
3731 		rtnl_unlock();
3732 	return min(ok_count, 0);
3733 }
3734 EXPORT_SYMBOL(tc_setup_cb_replace);
3735 
3736 /* Destroy filter and decrement block offload counter, if filter was previously
3737  * offloaded.
3738  */
3739 
3740 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3741 			enum tc_setup_type type, void *type_data, bool err_stop,
3742 			u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3743 {
3744 	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3745 	int ok_count;
3746 
3747 retry:
3748 	if (take_rtnl)
3749 		rtnl_lock();
3750 	down_read(&block->cb_lock);
3751 	/* Need to obtain rtnl lock if block is bound to devs that require it.
3752 	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3753 	 * obtain the locks in same order here.
3754 	 */
3755 	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3756 		up_read(&block->cb_lock);
3757 		take_rtnl = true;
3758 		goto retry;
3759 	}
3760 
3761 	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3762 
3763 	tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3764 	if (tp->ops->hw_del)
3765 		tp->ops->hw_del(tp, type_data);
3766 
3767 	up_read(&block->cb_lock);
3768 	if (take_rtnl)
3769 		rtnl_unlock();
3770 	return min(ok_count, 0);
3771 }
3772 EXPORT_SYMBOL(tc_setup_cb_destroy);
3773 
3774 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3775 			  bool add, flow_setup_cb_t *cb,
3776 			  enum tc_setup_type type, void *type_data,
3777 			  void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3778 {
3779 	int err = cb(type, type_data, cb_priv);
3780 
3781 	if (err) {
3782 		if (add && tc_skip_sw(*flags))
3783 			return err;
3784 	} else {
3785 		tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3786 					  add);
3787 	}
3788 
3789 	return 0;
3790 }
3791 EXPORT_SYMBOL(tc_setup_cb_reoffload);
3792 
3793 static int tcf_act_get_user_cookie(struct flow_action_entry *entry,
3794 				   const struct tc_action *act)
3795 {
3796 	struct tc_cookie *user_cookie;
3797 	int err = 0;
3798 
3799 	rcu_read_lock();
3800 	user_cookie = rcu_dereference(act->user_cookie);
3801 	if (user_cookie) {
3802 		entry->user_cookie = flow_action_cookie_create(user_cookie->data,
3803 							       user_cookie->len,
3804 							       GFP_ATOMIC);
3805 		if (!entry->user_cookie)
3806 			err = -ENOMEM;
3807 	}
3808 	rcu_read_unlock();
3809 	return err;
3810 }
3811 
3812 static void tcf_act_put_user_cookie(struct flow_action_entry *entry)
3813 {
3814 	flow_action_cookie_destroy(entry->user_cookie);
3815 }
3816 
3817 void tc_cleanup_offload_action(struct flow_action *flow_action)
3818 {
3819 	struct flow_action_entry *entry;
3820 	int i;
3821 
3822 	flow_action_for_each(i, entry, flow_action) {
3823 		tcf_act_put_user_cookie(entry);
3824 		if (entry->destructor)
3825 			entry->destructor(entry->destructor_priv);
3826 	}
3827 }
3828 EXPORT_SYMBOL(tc_cleanup_offload_action);
3829 
3830 static int tc_setup_offload_act(struct tc_action *act,
3831 				struct flow_action_entry *entry,
3832 				u32 *index_inc,
3833 				struct netlink_ext_ack *extack)
3834 {
3835 #ifdef CONFIG_NET_CLS_ACT
3836 	if (act->ops->offload_act_setup) {
3837 		return act->ops->offload_act_setup(act, entry, index_inc, true,
3838 						   extack);
3839 	} else {
3840 		NL_SET_ERR_MSG(extack, "Action does not support offload");
3841 		return -EOPNOTSUPP;
3842 	}
3843 #else
3844 	return 0;
3845 #endif
3846 }
3847 
3848 int tc_setup_action(struct flow_action *flow_action,
3849 		    struct tc_action *actions[],
3850 		    u32 miss_cookie_base,
3851 		    struct netlink_ext_ack *extack)
3852 {
3853 	int i, j, k, index, err = 0;
3854 	struct tc_action *act;
3855 
3856 	BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3857 	BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3858 	BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3859 
3860 	if (!actions)
3861 		return 0;
3862 
3863 	j = 0;
3864 	tcf_act_for_each_action(i, act, actions) {
3865 		struct flow_action_entry *entry;
3866 
3867 		entry = &flow_action->entries[j];
3868 		spin_lock_bh(&act->tcfa_lock);
3869 		err = tcf_act_get_user_cookie(entry, act);
3870 		if (err)
3871 			goto err_out_locked;
3872 
3873 		index = 0;
3874 		err = tc_setup_offload_act(act, entry, &index, extack);
3875 		if (err)
3876 			goto err_out_locked;
3877 
3878 		for (k = 0; k < index ; k++) {
3879 			entry[k].hw_stats = tc_act_hw_stats(act->hw_stats);
3880 			entry[k].hw_index = act->tcfa_index;
3881 			entry[k].cookie = (unsigned long)act;
3882 			entry[k].miss_cookie =
3883 				tcf_exts_miss_cookie_get(miss_cookie_base, i);
3884 		}
3885 
3886 		j += index;
3887 
3888 		spin_unlock_bh(&act->tcfa_lock);
3889 	}
3890 
3891 err_out:
3892 	if (err)
3893 		tc_cleanup_offload_action(flow_action);
3894 
3895 	return err;
3896 err_out_locked:
3897 	spin_unlock_bh(&act->tcfa_lock);
3898 	goto err_out;
3899 }
3900 
3901 int tc_setup_offload_action(struct flow_action *flow_action,
3902 			    const struct tcf_exts *exts,
3903 			    struct netlink_ext_ack *extack)
3904 {
3905 #ifdef CONFIG_NET_CLS_ACT
3906 	u32 miss_cookie_base;
3907 
3908 	if (!exts)
3909 		return 0;
3910 
3911 	miss_cookie_base = exts->miss_cookie_node ?
3912 			   exts->miss_cookie_node->miss_cookie_base : 0;
3913 	return tc_setup_action(flow_action, exts->actions, miss_cookie_base,
3914 			       extack);
3915 #else
3916 	return 0;
3917 #endif
3918 }
3919 EXPORT_SYMBOL(tc_setup_offload_action);
3920 
3921 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3922 {
3923 	unsigned int num_acts = 0;
3924 	struct tc_action *act;
3925 	int i;
3926 
3927 	tcf_exts_for_each_action(i, act, exts) {
3928 		if (is_tcf_pedit(act))
3929 			num_acts += tcf_pedit_nkeys(act);
3930 		else
3931 			num_acts++;
3932 	}
3933 	return num_acts;
3934 }
3935 EXPORT_SYMBOL(tcf_exts_num_actions);
3936 
3937 #ifdef CONFIG_NET_CLS_ACT
3938 static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3939 					u32 *p_block_index,
3940 					struct netlink_ext_ack *extack)
3941 {
3942 	*p_block_index = nla_get_u32(block_index_attr);
3943 	if (!*p_block_index) {
3944 		NL_SET_ERR_MSG(extack, "Block number may not be zero");
3945 		return -EINVAL;
3946 	}
3947 
3948 	return 0;
3949 }
3950 
3951 int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3952 		    enum flow_block_binder_type binder_type,
3953 		    struct nlattr *block_index_attr,
3954 		    struct netlink_ext_ack *extack)
3955 {
3956 	u32 block_index;
3957 	int err;
3958 
3959 	if (!block_index_attr)
3960 		return 0;
3961 
3962 	err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3963 	if (err)
3964 		return err;
3965 
3966 	qe->info.binder_type = binder_type;
3967 	qe->info.chain_head_change = tcf_chain_head_change_dflt;
3968 	qe->info.chain_head_change_priv = &qe->filter_chain;
3969 	qe->info.block_index = block_index;
3970 
3971 	return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3972 }
3973 EXPORT_SYMBOL(tcf_qevent_init);
3974 
3975 void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3976 {
3977 	if (qe->info.block_index)
3978 		tcf_block_put_ext(qe->block, sch, &qe->info);
3979 }
3980 EXPORT_SYMBOL(tcf_qevent_destroy);
3981 
3982 int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3983 			       struct netlink_ext_ack *extack)
3984 {
3985 	u32 block_index;
3986 	int err;
3987 
3988 	if (!block_index_attr)
3989 		return 0;
3990 
3991 	err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3992 	if (err)
3993 		return err;
3994 
3995 	/* Bounce newly-configured block or change in block. */
3996 	if (block_index != qe->info.block_index) {
3997 		NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3998 		return -EINVAL;
3999 	}
4000 
4001 	return 0;
4002 }
4003 EXPORT_SYMBOL(tcf_qevent_validate_change);
4004 
4005 struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
4006 				  struct sk_buff **to_free, int *ret)
4007 {
4008 	struct tcf_result cl_res;
4009 	struct tcf_proto *fl;
4010 
4011 	if (!qe->info.block_index)
4012 		return skb;
4013 
4014 	fl = rcu_dereference_bh(qe->filter_chain);
4015 
4016 	switch (tcf_classify(skb, NULL, fl, &cl_res, false)) {
4017 	case TC_ACT_SHOT:
4018 		qdisc_qstats_drop(sch);
4019 		__qdisc_drop(skb, to_free);
4020 		*ret = __NET_XMIT_BYPASS;
4021 		return NULL;
4022 	case TC_ACT_STOLEN:
4023 	case TC_ACT_QUEUED:
4024 	case TC_ACT_TRAP:
4025 		__qdisc_drop(skb, to_free);
4026 		*ret = __NET_XMIT_STOLEN;
4027 		return NULL;
4028 	case TC_ACT_REDIRECT:
4029 		skb_do_redirect(skb);
4030 		*ret = __NET_XMIT_STOLEN;
4031 		return NULL;
4032 	}
4033 
4034 	return skb;
4035 }
4036 EXPORT_SYMBOL(tcf_qevent_handle);
4037 
4038 int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
4039 {
4040 	if (!qe->info.block_index)
4041 		return 0;
4042 	return nla_put_u32(skb, attr_name, qe->info.block_index);
4043 }
4044 EXPORT_SYMBOL(tcf_qevent_dump);
4045 #endif
4046 
4047 static __net_init int tcf_net_init(struct net *net)
4048 {
4049 	struct tcf_net *tn = net_generic(net, tcf_net_id);
4050 
4051 	spin_lock_init(&tn->idr_lock);
4052 	idr_init(&tn->idr);
4053 	return 0;
4054 }
4055 
4056 static void __net_exit tcf_net_exit(struct net *net)
4057 {
4058 	struct tcf_net *tn = net_generic(net, tcf_net_id);
4059 
4060 	idr_destroy(&tn->idr);
4061 }
4062 
4063 static struct pernet_operations tcf_net_ops = {
4064 	.init = tcf_net_init,
4065 	.exit = tcf_net_exit,
4066 	.id   = &tcf_net_id,
4067 	.size = sizeof(struct tcf_net),
4068 };
4069 
4070 static const struct rtnl_msg_handler tc_filter_rtnl_msg_handlers[] __initconst = {
4071 	{.msgtype = RTM_NEWTFILTER, .doit = tc_new_tfilter,
4072 	 .flags = RTNL_FLAG_DOIT_UNLOCKED},
4073 	{.msgtype = RTM_DELTFILTER, .doit = tc_del_tfilter,
4074 	 .flags = RTNL_FLAG_DOIT_UNLOCKED},
4075 	{.msgtype = RTM_GETTFILTER, .doit = tc_get_tfilter,
4076 	 .dumpit = tc_dump_tfilter, .flags = RTNL_FLAG_DOIT_UNLOCKED},
4077 	{.msgtype = RTM_NEWCHAIN, .doit = tc_ctl_chain},
4078 	{.msgtype = RTM_DELCHAIN, .doit = tc_ctl_chain},
4079 	{.msgtype = RTM_GETCHAIN, .doit = tc_ctl_chain,
4080 	 .dumpit = tc_dump_chain},
4081 };
4082 
4083 static int __init tc_filter_init(void)
4084 {
4085 	int err;
4086 
4087 	tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
4088 	if (!tc_filter_wq)
4089 		return -ENOMEM;
4090 
4091 	err = register_pernet_subsys(&tcf_net_ops);
4092 	if (err)
4093 		goto err_register_pernet_subsys;
4094 
4095 	xa_init_flags(&tcf_exts_miss_cookies_xa, XA_FLAGS_ALLOC1);
4096 	rtnl_register_many(tc_filter_rtnl_msg_handlers);
4097 
4098 	return 0;
4099 
4100 err_register_pernet_subsys:
4101 	destroy_workqueue(tc_filter_wq);
4102 	return err;
4103 }
4104 
4105 subsys_initcall(tc_filter_init);
4106