xref: /linux/net/sched/cls_api.c (revision 1375dc4a4579d5e767dd8c2d2abcd929ff59d0a7)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * net/sched/cls_api.c	Packet classifier API.
4  *
5  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6  *
7  * Changes:
8  *
9  * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
10  */
11 
12 #include <linux/module.h>
13 #include <linux/types.h>
14 #include <linux/kernel.h>
15 #include <linux/string.h>
16 #include <linux/errno.h>
17 #include <linux/err.h>
18 #include <linux/skbuff.h>
19 #include <linux/init.h>
20 #include <linux/kmod.h>
21 #include <linux/slab.h>
22 #include <linux/idr.h>
23 #include <linux/rhashtable.h>
24 #include <net/net_namespace.h>
25 #include <net/sock.h>
26 #include <net/netlink.h>
27 #include <net/pkt_sched.h>
28 #include <net/pkt_cls.h>
29 #include <net/tc_act/tc_pedit.h>
30 #include <net/tc_act/tc_mirred.h>
31 #include <net/tc_act/tc_vlan.h>
32 #include <net/tc_act/tc_tunnel_key.h>
33 #include <net/tc_act/tc_csum.h>
34 #include <net/tc_act/tc_gact.h>
35 #include <net/tc_act/tc_police.h>
36 #include <net/tc_act/tc_sample.h>
37 #include <net/tc_act/tc_skbedit.h>
38 #include <net/tc_act/tc_ct.h>
39 #include <net/tc_act/tc_mpls.h>
40 
41 extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
42 
43 /* The list of all installed classifier types */
44 static LIST_HEAD(tcf_proto_base);
45 
46 /* Protects list of registered TC modules. It is pure SMP lock. */
47 static DEFINE_RWLOCK(cls_mod_lock);
48 
49 /* Find classifier type by string name */
50 
51 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
52 {
53 	const struct tcf_proto_ops *t, *res = NULL;
54 
55 	if (kind) {
56 		read_lock(&cls_mod_lock);
57 		list_for_each_entry(t, &tcf_proto_base, head) {
58 			if (strcmp(kind, t->kind) == 0) {
59 				if (try_module_get(t->owner))
60 					res = t;
61 				break;
62 			}
63 		}
64 		read_unlock(&cls_mod_lock);
65 	}
66 	return res;
67 }
68 
69 static const struct tcf_proto_ops *
70 tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
71 		     struct netlink_ext_ack *extack)
72 {
73 	const struct tcf_proto_ops *ops;
74 
75 	ops = __tcf_proto_lookup_ops(kind);
76 	if (ops)
77 		return ops;
78 #ifdef CONFIG_MODULES
79 	if (rtnl_held)
80 		rtnl_unlock();
81 	request_module("cls_%s", kind);
82 	if (rtnl_held)
83 		rtnl_lock();
84 	ops = __tcf_proto_lookup_ops(kind);
85 	/* We dropped the RTNL semaphore in order to perform
86 	 * the module load. So, even if we succeeded in loading
87 	 * the module we have to replay the request. We indicate
88 	 * this using -EAGAIN.
89 	 */
90 	if (ops) {
91 		module_put(ops->owner);
92 		return ERR_PTR(-EAGAIN);
93 	}
94 #endif
95 	NL_SET_ERR_MSG(extack, "TC classifier not found");
96 	return ERR_PTR(-ENOENT);
97 }
98 
99 /* Register(unregister) new classifier type */
100 
101 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
102 {
103 	struct tcf_proto_ops *t;
104 	int rc = -EEXIST;
105 
106 	write_lock(&cls_mod_lock);
107 	list_for_each_entry(t, &tcf_proto_base, head)
108 		if (!strcmp(ops->kind, t->kind))
109 			goto out;
110 
111 	list_add_tail(&ops->head, &tcf_proto_base);
112 	rc = 0;
113 out:
114 	write_unlock(&cls_mod_lock);
115 	return rc;
116 }
117 EXPORT_SYMBOL(register_tcf_proto_ops);
118 
119 static struct workqueue_struct *tc_filter_wq;
120 
121 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
122 {
123 	struct tcf_proto_ops *t;
124 	int rc = -ENOENT;
125 
126 	/* Wait for outstanding call_rcu()s, if any, from a
127 	 * tcf_proto_ops's destroy() handler.
128 	 */
129 	rcu_barrier();
130 	flush_workqueue(tc_filter_wq);
131 
132 	write_lock(&cls_mod_lock);
133 	list_for_each_entry(t, &tcf_proto_base, head) {
134 		if (t == ops) {
135 			list_del(&t->head);
136 			rc = 0;
137 			break;
138 		}
139 	}
140 	write_unlock(&cls_mod_lock);
141 	return rc;
142 }
143 EXPORT_SYMBOL(unregister_tcf_proto_ops);
144 
145 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
146 {
147 	INIT_RCU_WORK(rwork, func);
148 	return queue_rcu_work(tc_filter_wq, rwork);
149 }
150 EXPORT_SYMBOL(tcf_queue_work);
151 
152 /* Select new prio value from the range, managed by kernel. */
153 
154 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
155 {
156 	u32 first = TC_H_MAKE(0xC0000000U, 0U);
157 
158 	if (tp)
159 		first = tp->prio - 1;
160 
161 	return TC_H_MAJ(first);
162 }
163 
164 static bool tcf_proto_is_unlocked(const char *kind)
165 {
166 	const struct tcf_proto_ops *ops;
167 	bool ret;
168 
169 	ops = tcf_proto_lookup_ops(kind, false, NULL);
170 	/* On error return false to take rtnl lock. Proto lookup/create
171 	 * functions will perform lookup again and properly handle errors.
172 	 */
173 	if (IS_ERR(ops))
174 		return false;
175 
176 	ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
177 	module_put(ops->owner);
178 	return ret;
179 }
180 
181 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
182 					  u32 prio, struct tcf_chain *chain,
183 					  bool rtnl_held,
184 					  struct netlink_ext_ack *extack)
185 {
186 	struct tcf_proto *tp;
187 	int err;
188 
189 	tp = kzalloc(sizeof(*tp), GFP_KERNEL);
190 	if (!tp)
191 		return ERR_PTR(-ENOBUFS);
192 
193 	tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
194 	if (IS_ERR(tp->ops)) {
195 		err = PTR_ERR(tp->ops);
196 		goto errout;
197 	}
198 	tp->classify = tp->ops->classify;
199 	tp->protocol = protocol;
200 	tp->prio = prio;
201 	tp->chain = chain;
202 	spin_lock_init(&tp->lock);
203 	refcount_set(&tp->refcnt, 1);
204 
205 	err = tp->ops->init(tp);
206 	if (err) {
207 		module_put(tp->ops->owner);
208 		goto errout;
209 	}
210 	return tp;
211 
212 errout:
213 	kfree(tp);
214 	return ERR_PTR(err);
215 }
216 
217 static void tcf_proto_get(struct tcf_proto *tp)
218 {
219 	refcount_inc(&tp->refcnt);
220 }
221 
222 static void tcf_chain_put(struct tcf_chain *chain);
223 
224 static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
225 			      struct netlink_ext_ack *extack)
226 {
227 	tp->ops->destroy(tp, rtnl_held, extack);
228 	tcf_chain_put(tp->chain);
229 	module_put(tp->ops->owner);
230 	kfree_rcu(tp, rcu);
231 }
232 
233 static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
234 			  struct netlink_ext_ack *extack)
235 {
236 	if (refcount_dec_and_test(&tp->refcnt))
237 		tcf_proto_destroy(tp, rtnl_held, extack);
238 }
239 
240 static int walker_check_empty(struct tcf_proto *tp, void *fh,
241 			      struct tcf_walker *arg)
242 {
243 	if (fh) {
244 		arg->nonempty = true;
245 		return -1;
246 	}
247 	return 0;
248 }
249 
250 static bool tcf_proto_is_empty(struct tcf_proto *tp, bool rtnl_held)
251 {
252 	struct tcf_walker walker = { .fn = walker_check_empty, };
253 
254 	if (tp->ops->walk) {
255 		tp->ops->walk(tp, &walker, rtnl_held);
256 		return !walker.nonempty;
257 	}
258 	return true;
259 }
260 
261 static bool tcf_proto_check_delete(struct tcf_proto *tp, bool rtnl_held)
262 {
263 	spin_lock(&tp->lock);
264 	if (tcf_proto_is_empty(tp, rtnl_held))
265 		tp->deleting = true;
266 	spin_unlock(&tp->lock);
267 	return tp->deleting;
268 }
269 
270 static void tcf_proto_mark_delete(struct tcf_proto *tp)
271 {
272 	spin_lock(&tp->lock);
273 	tp->deleting = true;
274 	spin_unlock(&tp->lock);
275 }
276 
277 static bool tcf_proto_is_deleting(struct tcf_proto *tp)
278 {
279 	bool deleting;
280 
281 	spin_lock(&tp->lock);
282 	deleting = tp->deleting;
283 	spin_unlock(&tp->lock);
284 
285 	return deleting;
286 }
287 
288 #define ASSERT_BLOCK_LOCKED(block)					\
289 	lockdep_assert_held(&(block)->lock)
290 
291 struct tcf_filter_chain_list_item {
292 	struct list_head list;
293 	tcf_chain_head_change_t *chain_head_change;
294 	void *chain_head_change_priv;
295 };
296 
297 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
298 					  u32 chain_index)
299 {
300 	struct tcf_chain *chain;
301 
302 	ASSERT_BLOCK_LOCKED(block);
303 
304 	chain = kzalloc(sizeof(*chain), GFP_KERNEL);
305 	if (!chain)
306 		return NULL;
307 	list_add_tail(&chain->list, &block->chain_list);
308 	mutex_init(&chain->filter_chain_lock);
309 	chain->block = block;
310 	chain->index = chain_index;
311 	chain->refcnt = 1;
312 	if (!chain->index)
313 		block->chain0.chain = chain;
314 	return chain;
315 }
316 
317 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
318 				       struct tcf_proto *tp_head)
319 {
320 	if (item->chain_head_change)
321 		item->chain_head_change(tp_head, item->chain_head_change_priv);
322 }
323 
324 static void tcf_chain0_head_change(struct tcf_chain *chain,
325 				   struct tcf_proto *tp_head)
326 {
327 	struct tcf_filter_chain_list_item *item;
328 	struct tcf_block *block = chain->block;
329 
330 	if (chain->index)
331 		return;
332 
333 	mutex_lock(&block->lock);
334 	list_for_each_entry(item, &block->chain0.filter_chain_list, list)
335 		tcf_chain_head_change_item(item, tp_head);
336 	mutex_unlock(&block->lock);
337 }
338 
339 /* Returns true if block can be safely freed. */
340 
341 static bool tcf_chain_detach(struct tcf_chain *chain)
342 {
343 	struct tcf_block *block = chain->block;
344 
345 	ASSERT_BLOCK_LOCKED(block);
346 
347 	list_del(&chain->list);
348 	if (!chain->index)
349 		block->chain0.chain = NULL;
350 
351 	if (list_empty(&block->chain_list) &&
352 	    refcount_read(&block->refcnt) == 0)
353 		return true;
354 
355 	return false;
356 }
357 
358 static void tcf_block_destroy(struct tcf_block *block)
359 {
360 	mutex_destroy(&block->lock);
361 	kfree_rcu(block, rcu);
362 }
363 
364 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
365 {
366 	struct tcf_block *block = chain->block;
367 
368 	mutex_destroy(&chain->filter_chain_lock);
369 	kfree_rcu(chain, rcu);
370 	if (free_block)
371 		tcf_block_destroy(block);
372 }
373 
374 static void tcf_chain_hold(struct tcf_chain *chain)
375 {
376 	ASSERT_BLOCK_LOCKED(chain->block);
377 
378 	++chain->refcnt;
379 }
380 
381 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
382 {
383 	ASSERT_BLOCK_LOCKED(chain->block);
384 
385 	/* In case all the references are action references, this
386 	 * chain should not be shown to the user.
387 	 */
388 	return chain->refcnt == chain->action_refcnt;
389 }
390 
391 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
392 					  u32 chain_index)
393 {
394 	struct tcf_chain *chain;
395 
396 	ASSERT_BLOCK_LOCKED(block);
397 
398 	list_for_each_entry(chain, &block->chain_list, list) {
399 		if (chain->index == chain_index)
400 			return chain;
401 	}
402 	return NULL;
403 }
404 
405 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
406 			   u32 seq, u16 flags, int event, bool unicast);
407 
408 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
409 					 u32 chain_index, bool create,
410 					 bool by_act)
411 {
412 	struct tcf_chain *chain = NULL;
413 	bool is_first_reference;
414 
415 	mutex_lock(&block->lock);
416 	chain = tcf_chain_lookup(block, chain_index);
417 	if (chain) {
418 		tcf_chain_hold(chain);
419 	} else {
420 		if (!create)
421 			goto errout;
422 		chain = tcf_chain_create(block, chain_index);
423 		if (!chain)
424 			goto errout;
425 	}
426 
427 	if (by_act)
428 		++chain->action_refcnt;
429 	is_first_reference = chain->refcnt - chain->action_refcnt == 1;
430 	mutex_unlock(&block->lock);
431 
432 	/* Send notification only in case we got the first
433 	 * non-action reference. Until then, the chain acts only as
434 	 * a placeholder for actions pointing to it and user ought
435 	 * not know about them.
436 	 */
437 	if (is_first_reference && !by_act)
438 		tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
439 				RTM_NEWCHAIN, false);
440 
441 	return chain;
442 
443 errout:
444 	mutex_unlock(&block->lock);
445 	return chain;
446 }
447 
448 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
449 				       bool create)
450 {
451 	return __tcf_chain_get(block, chain_index, create, false);
452 }
453 
454 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
455 {
456 	return __tcf_chain_get(block, chain_index, true, true);
457 }
458 EXPORT_SYMBOL(tcf_chain_get_by_act);
459 
460 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
461 			       void *tmplt_priv);
462 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
463 				  void *tmplt_priv, u32 chain_index,
464 				  struct tcf_block *block, struct sk_buff *oskb,
465 				  u32 seq, u16 flags, bool unicast);
466 
467 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
468 			    bool explicitly_created)
469 {
470 	struct tcf_block *block = chain->block;
471 	const struct tcf_proto_ops *tmplt_ops;
472 	bool free_block = false;
473 	unsigned int refcnt;
474 	void *tmplt_priv;
475 
476 	mutex_lock(&block->lock);
477 	if (explicitly_created) {
478 		if (!chain->explicitly_created) {
479 			mutex_unlock(&block->lock);
480 			return;
481 		}
482 		chain->explicitly_created = false;
483 	}
484 
485 	if (by_act)
486 		chain->action_refcnt--;
487 
488 	/* tc_chain_notify_delete can't be called while holding block lock.
489 	 * However, when block is unlocked chain can be changed concurrently, so
490 	 * save these to temporary variables.
491 	 */
492 	refcnt = --chain->refcnt;
493 	tmplt_ops = chain->tmplt_ops;
494 	tmplt_priv = chain->tmplt_priv;
495 
496 	/* The last dropped non-action reference will trigger notification. */
497 	if (refcnt - chain->action_refcnt == 0 && !by_act) {
498 		tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
499 				       block, NULL, 0, 0, false);
500 		/* Last reference to chain, no need to lock. */
501 		chain->flushing = false;
502 	}
503 
504 	if (refcnt == 0)
505 		free_block = tcf_chain_detach(chain);
506 	mutex_unlock(&block->lock);
507 
508 	if (refcnt == 0) {
509 		tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
510 		tcf_chain_destroy(chain, free_block);
511 	}
512 }
513 
514 static void tcf_chain_put(struct tcf_chain *chain)
515 {
516 	__tcf_chain_put(chain, false, false);
517 }
518 
519 void tcf_chain_put_by_act(struct tcf_chain *chain)
520 {
521 	__tcf_chain_put(chain, true, false);
522 }
523 EXPORT_SYMBOL(tcf_chain_put_by_act);
524 
525 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
526 {
527 	__tcf_chain_put(chain, false, true);
528 }
529 
530 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
531 {
532 	struct tcf_proto *tp, *tp_next;
533 
534 	mutex_lock(&chain->filter_chain_lock);
535 	tp = tcf_chain_dereference(chain->filter_chain, chain);
536 	RCU_INIT_POINTER(chain->filter_chain, NULL);
537 	tcf_chain0_head_change(chain, NULL);
538 	chain->flushing = true;
539 	mutex_unlock(&chain->filter_chain_lock);
540 
541 	while (tp) {
542 		tp_next = rcu_dereference_protected(tp->next, 1);
543 		tcf_proto_put(tp, rtnl_held, NULL);
544 		tp = tp_next;
545 	}
546 }
547 
548 static struct tcf_block *tc_dev_ingress_block(struct net_device *dev)
549 {
550 	const struct Qdisc_class_ops *cops;
551 	struct Qdisc *qdisc;
552 
553 	if (!dev_ingress_queue(dev))
554 		return NULL;
555 
556 	qdisc = dev_ingress_queue(dev)->qdisc_sleeping;
557 	if (!qdisc)
558 		return NULL;
559 
560 	cops = qdisc->ops->cl_ops;
561 	if (!cops)
562 		return NULL;
563 
564 	if (!cops->tcf_block)
565 		return NULL;
566 
567 	return cops->tcf_block(qdisc, TC_H_MIN_INGRESS, NULL);
568 }
569 
570 static struct rhashtable indr_setup_block_ht;
571 
572 struct tc_indr_block_dev {
573 	struct rhash_head ht_node;
574 	struct net_device *dev;
575 	unsigned int refcnt;
576 	struct list_head cb_list;
577 	struct tcf_block *block;
578 };
579 
580 struct tc_indr_block_cb {
581 	struct list_head list;
582 	void *cb_priv;
583 	tc_indr_block_bind_cb_t *cb;
584 	void *cb_ident;
585 };
586 
587 static const struct rhashtable_params tc_indr_setup_block_ht_params = {
588 	.key_offset	= offsetof(struct tc_indr_block_dev, dev),
589 	.head_offset	= offsetof(struct tc_indr_block_dev, ht_node),
590 	.key_len	= sizeof(struct net_device *),
591 };
592 
593 static struct tc_indr_block_dev *
594 tc_indr_block_dev_lookup(struct net_device *dev)
595 {
596 	return rhashtable_lookup_fast(&indr_setup_block_ht, &dev,
597 				      tc_indr_setup_block_ht_params);
598 }
599 
600 static struct tc_indr_block_dev *tc_indr_block_dev_get(struct net_device *dev)
601 {
602 	struct tc_indr_block_dev *indr_dev;
603 
604 	indr_dev = tc_indr_block_dev_lookup(dev);
605 	if (indr_dev)
606 		goto inc_ref;
607 
608 	indr_dev = kzalloc(sizeof(*indr_dev), GFP_KERNEL);
609 	if (!indr_dev)
610 		return NULL;
611 
612 	INIT_LIST_HEAD(&indr_dev->cb_list);
613 	indr_dev->dev = dev;
614 	indr_dev->block = tc_dev_ingress_block(dev);
615 	if (rhashtable_insert_fast(&indr_setup_block_ht, &indr_dev->ht_node,
616 				   tc_indr_setup_block_ht_params)) {
617 		kfree(indr_dev);
618 		return NULL;
619 	}
620 
621 inc_ref:
622 	indr_dev->refcnt++;
623 	return indr_dev;
624 }
625 
626 static void tc_indr_block_dev_put(struct tc_indr_block_dev *indr_dev)
627 {
628 	if (--indr_dev->refcnt)
629 		return;
630 
631 	rhashtable_remove_fast(&indr_setup_block_ht, &indr_dev->ht_node,
632 			       tc_indr_setup_block_ht_params);
633 	kfree(indr_dev);
634 }
635 
636 static struct tc_indr_block_cb *
637 tc_indr_block_cb_lookup(struct tc_indr_block_dev *indr_dev,
638 			tc_indr_block_bind_cb_t *cb, void *cb_ident)
639 {
640 	struct tc_indr_block_cb *indr_block_cb;
641 
642 	list_for_each_entry(indr_block_cb, &indr_dev->cb_list, list)
643 		if (indr_block_cb->cb == cb &&
644 		    indr_block_cb->cb_ident == cb_ident)
645 			return indr_block_cb;
646 	return NULL;
647 }
648 
649 static struct tc_indr_block_cb *
650 tc_indr_block_cb_add(struct tc_indr_block_dev *indr_dev, void *cb_priv,
651 		     tc_indr_block_bind_cb_t *cb, void *cb_ident)
652 {
653 	struct tc_indr_block_cb *indr_block_cb;
654 
655 	indr_block_cb = tc_indr_block_cb_lookup(indr_dev, cb, cb_ident);
656 	if (indr_block_cb)
657 		return ERR_PTR(-EEXIST);
658 
659 	indr_block_cb = kzalloc(sizeof(*indr_block_cb), GFP_KERNEL);
660 	if (!indr_block_cb)
661 		return ERR_PTR(-ENOMEM);
662 
663 	indr_block_cb->cb_priv = cb_priv;
664 	indr_block_cb->cb = cb;
665 	indr_block_cb->cb_ident = cb_ident;
666 	list_add(&indr_block_cb->list, &indr_dev->cb_list);
667 
668 	return indr_block_cb;
669 }
670 
671 static void tc_indr_block_cb_del(struct tc_indr_block_cb *indr_block_cb)
672 {
673 	list_del(&indr_block_cb->list);
674 	kfree(indr_block_cb);
675 }
676 
677 static int tcf_block_setup(struct tcf_block *block,
678 			   struct flow_block_offload *bo);
679 
680 static void tc_indr_block_ing_cmd(struct tc_indr_block_dev *indr_dev,
681 				  struct tc_indr_block_cb *indr_block_cb,
682 				  enum flow_block_command command)
683 {
684 	struct flow_block_offload bo = {
685 		.command	= command,
686 		.binder_type	= FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS,
687 		.net		= dev_net(indr_dev->dev),
688 		.block_shared	= tcf_block_non_null_shared(indr_dev->block),
689 	};
690 	INIT_LIST_HEAD(&bo.cb_list);
691 
692 	if (!indr_dev->block)
693 		return;
694 
695 	bo.block = &indr_dev->block->flow_block;
696 
697 	indr_block_cb->cb(indr_dev->dev, indr_block_cb->cb_priv, TC_SETUP_BLOCK,
698 			  &bo);
699 	tcf_block_setup(indr_dev->block, &bo);
700 }
701 
702 int __tc_indr_block_cb_register(struct net_device *dev, void *cb_priv,
703 				tc_indr_block_bind_cb_t *cb, void *cb_ident)
704 {
705 	struct tc_indr_block_cb *indr_block_cb;
706 	struct tc_indr_block_dev *indr_dev;
707 	int err;
708 
709 	indr_dev = tc_indr_block_dev_get(dev);
710 	if (!indr_dev)
711 		return -ENOMEM;
712 
713 	indr_block_cb = tc_indr_block_cb_add(indr_dev, cb_priv, cb, cb_ident);
714 	err = PTR_ERR_OR_ZERO(indr_block_cb);
715 	if (err)
716 		goto err_dev_put;
717 
718 	tc_indr_block_ing_cmd(indr_dev, indr_block_cb, FLOW_BLOCK_BIND);
719 	return 0;
720 
721 err_dev_put:
722 	tc_indr_block_dev_put(indr_dev);
723 	return err;
724 }
725 EXPORT_SYMBOL_GPL(__tc_indr_block_cb_register);
726 
727 int tc_indr_block_cb_register(struct net_device *dev, void *cb_priv,
728 			      tc_indr_block_bind_cb_t *cb, void *cb_ident)
729 {
730 	int err;
731 
732 	rtnl_lock();
733 	err = __tc_indr_block_cb_register(dev, cb_priv, cb, cb_ident);
734 	rtnl_unlock();
735 
736 	return err;
737 }
738 EXPORT_SYMBOL_GPL(tc_indr_block_cb_register);
739 
740 void __tc_indr_block_cb_unregister(struct net_device *dev,
741 				   tc_indr_block_bind_cb_t *cb, void *cb_ident)
742 {
743 	struct tc_indr_block_cb *indr_block_cb;
744 	struct tc_indr_block_dev *indr_dev;
745 
746 	indr_dev = tc_indr_block_dev_lookup(dev);
747 	if (!indr_dev)
748 		return;
749 
750 	indr_block_cb = tc_indr_block_cb_lookup(indr_dev, cb, cb_ident);
751 	if (!indr_block_cb)
752 		return;
753 
754 	/* Send unbind message if required to free any block cbs. */
755 	tc_indr_block_ing_cmd(indr_dev, indr_block_cb, FLOW_BLOCK_UNBIND);
756 	tc_indr_block_cb_del(indr_block_cb);
757 	tc_indr_block_dev_put(indr_dev);
758 }
759 EXPORT_SYMBOL_GPL(__tc_indr_block_cb_unregister);
760 
761 void tc_indr_block_cb_unregister(struct net_device *dev,
762 				 tc_indr_block_bind_cb_t *cb, void *cb_ident)
763 {
764 	rtnl_lock();
765 	__tc_indr_block_cb_unregister(dev, cb, cb_ident);
766 	rtnl_unlock();
767 }
768 EXPORT_SYMBOL_GPL(tc_indr_block_cb_unregister);
769 
770 static void tc_indr_block_call(struct tcf_block *block, struct net_device *dev,
771 			       struct tcf_block_ext_info *ei,
772 			       enum flow_block_command command,
773 			       struct netlink_ext_ack *extack)
774 {
775 	struct tc_indr_block_cb *indr_block_cb;
776 	struct tc_indr_block_dev *indr_dev;
777 	struct flow_block_offload bo = {
778 		.command	= command,
779 		.binder_type	= ei->binder_type,
780 		.net		= dev_net(dev),
781 		.block		= &block->flow_block,
782 		.block_shared	= tcf_block_shared(block),
783 		.extack		= extack,
784 	};
785 	INIT_LIST_HEAD(&bo.cb_list);
786 
787 	indr_dev = tc_indr_block_dev_lookup(dev);
788 	if (!indr_dev)
789 		return;
790 
791 	indr_dev->block = command == FLOW_BLOCK_BIND ? block : NULL;
792 
793 	list_for_each_entry(indr_block_cb, &indr_dev->cb_list, list)
794 		indr_block_cb->cb(dev, indr_block_cb->cb_priv, TC_SETUP_BLOCK,
795 				  &bo);
796 
797 	tcf_block_setup(block, &bo);
798 }
799 
800 static bool tcf_block_offload_in_use(struct tcf_block *block)
801 {
802 	return block->offloadcnt;
803 }
804 
805 static int tcf_block_offload_cmd(struct tcf_block *block,
806 				 struct net_device *dev,
807 				 struct tcf_block_ext_info *ei,
808 				 enum flow_block_command command,
809 				 struct netlink_ext_ack *extack)
810 {
811 	struct flow_block_offload bo = {};
812 	int err;
813 
814 	bo.net = dev_net(dev);
815 	bo.command = command;
816 	bo.binder_type = ei->binder_type;
817 	bo.block = &block->flow_block;
818 	bo.block_shared = tcf_block_shared(block);
819 	bo.extack = extack;
820 	INIT_LIST_HEAD(&bo.cb_list);
821 
822 	err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
823 	if (err < 0)
824 		return err;
825 
826 	return tcf_block_setup(block, &bo);
827 }
828 
829 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
830 				  struct tcf_block_ext_info *ei,
831 				  struct netlink_ext_ack *extack)
832 {
833 	struct net_device *dev = q->dev_queue->dev;
834 	int err;
835 
836 	if (!dev->netdev_ops->ndo_setup_tc)
837 		goto no_offload_dev_inc;
838 
839 	/* If tc offload feature is disabled and the block we try to bind
840 	 * to already has some offloaded filters, forbid to bind.
841 	 */
842 	if (!tc_can_offload(dev) && tcf_block_offload_in_use(block)) {
843 		NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
844 		return -EOPNOTSUPP;
845 	}
846 
847 	err = tcf_block_offload_cmd(block, dev, ei, FLOW_BLOCK_BIND, extack);
848 	if (err == -EOPNOTSUPP)
849 		goto no_offload_dev_inc;
850 	if (err)
851 		return err;
852 
853 	tc_indr_block_call(block, dev, ei, FLOW_BLOCK_BIND, extack);
854 	return 0;
855 
856 no_offload_dev_inc:
857 	if (tcf_block_offload_in_use(block))
858 		return -EOPNOTSUPP;
859 	block->nooffloaddevcnt++;
860 	tc_indr_block_call(block, dev, ei, FLOW_BLOCK_BIND, extack);
861 	return 0;
862 }
863 
864 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
865 				     struct tcf_block_ext_info *ei)
866 {
867 	struct net_device *dev = q->dev_queue->dev;
868 	int err;
869 
870 	tc_indr_block_call(block, dev, ei, FLOW_BLOCK_UNBIND, NULL);
871 
872 	if (!dev->netdev_ops->ndo_setup_tc)
873 		goto no_offload_dev_dec;
874 	err = tcf_block_offload_cmd(block, dev, ei, FLOW_BLOCK_UNBIND, NULL);
875 	if (err == -EOPNOTSUPP)
876 		goto no_offload_dev_dec;
877 	return;
878 
879 no_offload_dev_dec:
880 	WARN_ON(block->nooffloaddevcnt-- == 0);
881 }
882 
883 static int
884 tcf_chain0_head_change_cb_add(struct tcf_block *block,
885 			      struct tcf_block_ext_info *ei,
886 			      struct netlink_ext_ack *extack)
887 {
888 	struct tcf_filter_chain_list_item *item;
889 	struct tcf_chain *chain0;
890 
891 	item = kmalloc(sizeof(*item), GFP_KERNEL);
892 	if (!item) {
893 		NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
894 		return -ENOMEM;
895 	}
896 	item->chain_head_change = ei->chain_head_change;
897 	item->chain_head_change_priv = ei->chain_head_change_priv;
898 
899 	mutex_lock(&block->lock);
900 	chain0 = block->chain0.chain;
901 	if (chain0)
902 		tcf_chain_hold(chain0);
903 	else
904 		list_add(&item->list, &block->chain0.filter_chain_list);
905 	mutex_unlock(&block->lock);
906 
907 	if (chain0) {
908 		struct tcf_proto *tp_head;
909 
910 		mutex_lock(&chain0->filter_chain_lock);
911 
912 		tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
913 		if (tp_head)
914 			tcf_chain_head_change_item(item, tp_head);
915 
916 		mutex_lock(&block->lock);
917 		list_add(&item->list, &block->chain0.filter_chain_list);
918 		mutex_unlock(&block->lock);
919 
920 		mutex_unlock(&chain0->filter_chain_lock);
921 		tcf_chain_put(chain0);
922 	}
923 
924 	return 0;
925 }
926 
927 static void
928 tcf_chain0_head_change_cb_del(struct tcf_block *block,
929 			      struct tcf_block_ext_info *ei)
930 {
931 	struct tcf_filter_chain_list_item *item;
932 
933 	mutex_lock(&block->lock);
934 	list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
935 		if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
936 		    (item->chain_head_change == ei->chain_head_change &&
937 		     item->chain_head_change_priv == ei->chain_head_change_priv)) {
938 			if (block->chain0.chain)
939 				tcf_chain_head_change_item(item, NULL);
940 			list_del(&item->list);
941 			mutex_unlock(&block->lock);
942 
943 			kfree(item);
944 			return;
945 		}
946 	}
947 	mutex_unlock(&block->lock);
948 	WARN_ON(1);
949 }
950 
951 struct tcf_net {
952 	spinlock_t idr_lock; /* Protects idr */
953 	struct idr idr;
954 };
955 
956 static unsigned int tcf_net_id;
957 
958 static int tcf_block_insert(struct tcf_block *block, struct net *net,
959 			    struct netlink_ext_ack *extack)
960 {
961 	struct tcf_net *tn = net_generic(net, tcf_net_id);
962 	int err;
963 
964 	idr_preload(GFP_KERNEL);
965 	spin_lock(&tn->idr_lock);
966 	err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
967 			    GFP_NOWAIT);
968 	spin_unlock(&tn->idr_lock);
969 	idr_preload_end();
970 
971 	return err;
972 }
973 
974 static void tcf_block_remove(struct tcf_block *block, struct net *net)
975 {
976 	struct tcf_net *tn = net_generic(net, tcf_net_id);
977 
978 	spin_lock(&tn->idr_lock);
979 	idr_remove(&tn->idr, block->index);
980 	spin_unlock(&tn->idr_lock);
981 }
982 
983 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
984 					  u32 block_index,
985 					  struct netlink_ext_ack *extack)
986 {
987 	struct tcf_block *block;
988 
989 	block = kzalloc(sizeof(*block), GFP_KERNEL);
990 	if (!block) {
991 		NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
992 		return ERR_PTR(-ENOMEM);
993 	}
994 	mutex_init(&block->lock);
995 	flow_block_init(&block->flow_block);
996 	INIT_LIST_HEAD(&block->chain_list);
997 	INIT_LIST_HEAD(&block->owner_list);
998 	INIT_LIST_HEAD(&block->chain0.filter_chain_list);
999 
1000 	refcount_set(&block->refcnt, 1);
1001 	block->net = net;
1002 	block->index = block_index;
1003 
1004 	/* Don't store q pointer for blocks which are shared */
1005 	if (!tcf_block_shared(block))
1006 		block->q = q;
1007 	return block;
1008 }
1009 
1010 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
1011 {
1012 	struct tcf_net *tn = net_generic(net, tcf_net_id);
1013 
1014 	return idr_find(&tn->idr, block_index);
1015 }
1016 
1017 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
1018 {
1019 	struct tcf_block *block;
1020 
1021 	rcu_read_lock();
1022 	block = tcf_block_lookup(net, block_index);
1023 	if (block && !refcount_inc_not_zero(&block->refcnt))
1024 		block = NULL;
1025 	rcu_read_unlock();
1026 
1027 	return block;
1028 }
1029 
1030 static struct tcf_chain *
1031 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1032 {
1033 	mutex_lock(&block->lock);
1034 	if (chain)
1035 		chain = list_is_last(&chain->list, &block->chain_list) ?
1036 			NULL : list_next_entry(chain, list);
1037 	else
1038 		chain = list_first_entry_or_null(&block->chain_list,
1039 						 struct tcf_chain, list);
1040 
1041 	/* skip all action-only chains */
1042 	while (chain && tcf_chain_held_by_acts_only(chain))
1043 		chain = list_is_last(&chain->list, &block->chain_list) ?
1044 			NULL : list_next_entry(chain, list);
1045 
1046 	if (chain)
1047 		tcf_chain_hold(chain);
1048 	mutex_unlock(&block->lock);
1049 
1050 	return chain;
1051 }
1052 
1053 /* Function to be used by all clients that want to iterate over all chains on
1054  * block. It properly obtains block->lock and takes reference to chain before
1055  * returning it. Users of this function must be tolerant to concurrent chain
1056  * insertion/deletion or ensure that no concurrent chain modification is
1057  * possible. Note that all netlink dump callbacks cannot guarantee to provide
1058  * consistent dump because rtnl lock is released each time skb is filled with
1059  * data and sent to user-space.
1060  */
1061 
1062 struct tcf_chain *
1063 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1064 {
1065 	struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
1066 
1067 	if (chain)
1068 		tcf_chain_put(chain);
1069 
1070 	return chain_next;
1071 }
1072 EXPORT_SYMBOL(tcf_get_next_chain);
1073 
1074 static struct tcf_proto *
1075 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1076 {
1077 	u32 prio = 0;
1078 
1079 	ASSERT_RTNL();
1080 	mutex_lock(&chain->filter_chain_lock);
1081 
1082 	if (!tp) {
1083 		tp = tcf_chain_dereference(chain->filter_chain, chain);
1084 	} else if (tcf_proto_is_deleting(tp)) {
1085 		/* 'deleting' flag is set and chain->filter_chain_lock was
1086 		 * unlocked, which means next pointer could be invalid. Restart
1087 		 * search.
1088 		 */
1089 		prio = tp->prio + 1;
1090 		tp = tcf_chain_dereference(chain->filter_chain, chain);
1091 
1092 		for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1093 			if (!tp->deleting && tp->prio >= prio)
1094 				break;
1095 	} else {
1096 		tp = tcf_chain_dereference(tp->next, chain);
1097 	}
1098 
1099 	if (tp)
1100 		tcf_proto_get(tp);
1101 
1102 	mutex_unlock(&chain->filter_chain_lock);
1103 
1104 	return tp;
1105 }
1106 
1107 /* Function to be used by all clients that want to iterate over all tp's on
1108  * chain. Users of this function must be tolerant to concurrent tp
1109  * insertion/deletion or ensure that no concurrent chain modification is
1110  * possible. Note that all netlink dump callbacks cannot guarantee to provide
1111  * consistent dump because rtnl lock is released each time skb is filled with
1112  * data and sent to user-space.
1113  */
1114 
1115 struct tcf_proto *
1116 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp,
1117 		   bool rtnl_held)
1118 {
1119 	struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1120 
1121 	if (tp)
1122 		tcf_proto_put(tp, rtnl_held, NULL);
1123 
1124 	return tp_next;
1125 }
1126 EXPORT_SYMBOL(tcf_get_next_proto);
1127 
1128 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1129 {
1130 	struct tcf_chain *chain;
1131 
1132 	/* Last reference to block. At this point chains cannot be added or
1133 	 * removed concurrently.
1134 	 */
1135 	for (chain = tcf_get_next_chain(block, NULL);
1136 	     chain;
1137 	     chain = tcf_get_next_chain(block, chain)) {
1138 		tcf_chain_put_explicitly_created(chain);
1139 		tcf_chain_flush(chain, rtnl_held);
1140 	}
1141 }
1142 
1143 /* Lookup Qdisc and increments its reference counter.
1144  * Set parent, if necessary.
1145  */
1146 
1147 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1148 			    u32 *parent, int ifindex, bool rtnl_held,
1149 			    struct netlink_ext_ack *extack)
1150 {
1151 	const struct Qdisc_class_ops *cops;
1152 	struct net_device *dev;
1153 	int err = 0;
1154 
1155 	if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1156 		return 0;
1157 
1158 	rcu_read_lock();
1159 
1160 	/* Find link */
1161 	dev = dev_get_by_index_rcu(net, ifindex);
1162 	if (!dev) {
1163 		rcu_read_unlock();
1164 		return -ENODEV;
1165 	}
1166 
1167 	/* Find qdisc */
1168 	if (!*parent) {
1169 		*q = dev->qdisc;
1170 		*parent = (*q)->handle;
1171 	} else {
1172 		*q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1173 		if (!*q) {
1174 			NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1175 			err = -EINVAL;
1176 			goto errout_rcu;
1177 		}
1178 	}
1179 
1180 	*q = qdisc_refcount_inc_nz(*q);
1181 	if (!*q) {
1182 		NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1183 		err = -EINVAL;
1184 		goto errout_rcu;
1185 	}
1186 
1187 	/* Is it classful? */
1188 	cops = (*q)->ops->cl_ops;
1189 	if (!cops) {
1190 		NL_SET_ERR_MSG(extack, "Qdisc not classful");
1191 		err = -EINVAL;
1192 		goto errout_qdisc;
1193 	}
1194 
1195 	if (!cops->tcf_block) {
1196 		NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1197 		err = -EOPNOTSUPP;
1198 		goto errout_qdisc;
1199 	}
1200 
1201 errout_rcu:
1202 	/* At this point we know that qdisc is not noop_qdisc,
1203 	 * which means that qdisc holds a reference to net_device
1204 	 * and we hold a reference to qdisc, so it is safe to release
1205 	 * rcu read lock.
1206 	 */
1207 	rcu_read_unlock();
1208 	return err;
1209 
1210 errout_qdisc:
1211 	rcu_read_unlock();
1212 
1213 	if (rtnl_held)
1214 		qdisc_put(*q);
1215 	else
1216 		qdisc_put_unlocked(*q);
1217 	*q = NULL;
1218 
1219 	return err;
1220 }
1221 
1222 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1223 			       int ifindex, struct netlink_ext_ack *extack)
1224 {
1225 	if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1226 		return 0;
1227 
1228 	/* Do we search for filter, attached to class? */
1229 	if (TC_H_MIN(parent)) {
1230 		const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1231 
1232 		*cl = cops->find(q, parent);
1233 		if (*cl == 0) {
1234 			NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1235 			return -ENOENT;
1236 		}
1237 	}
1238 
1239 	return 0;
1240 }
1241 
1242 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1243 					  unsigned long cl, int ifindex,
1244 					  u32 block_index,
1245 					  struct netlink_ext_ack *extack)
1246 {
1247 	struct tcf_block *block;
1248 
1249 	if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1250 		block = tcf_block_refcnt_get(net, block_index);
1251 		if (!block) {
1252 			NL_SET_ERR_MSG(extack, "Block of given index was not found");
1253 			return ERR_PTR(-EINVAL);
1254 		}
1255 	} else {
1256 		const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1257 
1258 		block = cops->tcf_block(q, cl, extack);
1259 		if (!block)
1260 			return ERR_PTR(-EINVAL);
1261 
1262 		if (tcf_block_shared(block)) {
1263 			NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1264 			return ERR_PTR(-EOPNOTSUPP);
1265 		}
1266 
1267 		/* Always take reference to block in order to support execution
1268 		 * of rules update path of cls API without rtnl lock. Caller
1269 		 * must release block when it is finished using it. 'if' block
1270 		 * of this conditional obtain reference to block by calling
1271 		 * tcf_block_refcnt_get().
1272 		 */
1273 		refcount_inc(&block->refcnt);
1274 	}
1275 
1276 	return block;
1277 }
1278 
1279 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1280 			    struct tcf_block_ext_info *ei, bool rtnl_held)
1281 {
1282 	if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1283 		/* Flushing/putting all chains will cause the block to be
1284 		 * deallocated when last chain is freed. However, if chain_list
1285 		 * is empty, block has to be manually deallocated. After block
1286 		 * reference counter reached 0, it is no longer possible to
1287 		 * increment it or add new chains to block.
1288 		 */
1289 		bool free_block = list_empty(&block->chain_list);
1290 
1291 		mutex_unlock(&block->lock);
1292 		if (tcf_block_shared(block))
1293 			tcf_block_remove(block, block->net);
1294 
1295 		if (q)
1296 			tcf_block_offload_unbind(block, q, ei);
1297 
1298 		if (free_block)
1299 			tcf_block_destroy(block);
1300 		else
1301 			tcf_block_flush_all_chains(block, rtnl_held);
1302 	} else if (q) {
1303 		tcf_block_offload_unbind(block, q, ei);
1304 	}
1305 }
1306 
1307 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1308 {
1309 	__tcf_block_put(block, NULL, NULL, rtnl_held);
1310 }
1311 
1312 /* Find tcf block.
1313  * Set q, parent, cl when appropriate.
1314  */
1315 
1316 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1317 					u32 *parent, unsigned long *cl,
1318 					int ifindex, u32 block_index,
1319 					struct netlink_ext_ack *extack)
1320 {
1321 	struct tcf_block *block;
1322 	int err = 0;
1323 
1324 	ASSERT_RTNL();
1325 
1326 	err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1327 	if (err)
1328 		goto errout;
1329 
1330 	err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1331 	if (err)
1332 		goto errout_qdisc;
1333 
1334 	block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1335 	if (IS_ERR(block)) {
1336 		err = PTR_ERR(block);
1337 		goto errout_qdisc;
1338 	}
1339 
1340 	return block;
1341 
1342 errout_qdisc:
1343 	if (*q)
1344 		qdisc_put(*q);
1345 errout:
1346 	*q = NULL;
1347 	return ERR_PTR(err);
1348 }
1349 
1350 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1351 			      bool rtnl_held)
1352 {
1353 	if (!IS_ERR_OR_NULL(block))
1354 		tcf_block_refcnt_put(block, rtnl_held);
1355 
1356 	if (q) {
1357 		if (rtnl_held)
1358 			qdisc_put(q);
1359 		else
1360 			qdisc_put_unlocked(q);
1361 	}
1362 }
1363 
1364 struct tcf_block_owner_item {
1365 	struct list_head list;
1366 	struct Qdisc *q;
1367 	enum flow_block_binder_type binder_type;
1368 };
1369 
1370 static void
1371 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1372 			       struct Qdisc *q,
1373 			       enum flow_block_binder_type binder_type)
1374 {
1375 	if (block->keep_dst &&
1376 	    binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1377 	    binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1378 		netif_keep_dst(qdisc_dev(q));
1379 }
1380 
1381 void tcf_block_netif_keep_dst(struct tcf_block *block)
1382 {
1383 	struct tcf_block_owner_item *item;
1384 
1385 	block->keep_dst = true;
1386 	list_for_each_entry(item, &block->owner_list, list)
1387 		tcf_block_owner_netif_keep_dst(block, item->q,
1388 					       item->binder_type);
1389 }
1390 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1391 
1392 static int tcf_block_owner_add(struct tcf_block *block,
1393 			       struct Qdisc *q,
1394 			       enum flow_block_binder_type binder_type)
1395 {
1396 	struct tcf_block_owner_item *item;
1397 
1398 	item = kmalloc(sizeof(*item), GFP_KERNEL);
1399 	if (!item)
1400 		return -ENOMEM;
1401 	item->q = q;
1402 	item->binder_type = binder_type;
1403 	list_add(&item->list, &block->owner_list);
1404 	return 0;
1405 }
1406 
1407 static void tcf_block_owner_del(struct tcf_block *block,
1408 				struct Qdisc *q,
1409 				enum flow_block_binder_type binder_type)
1410 {
1411 	struct tcf_block_owner_item *item;
1412 
1413 	list_for_each_entry(item, &block->owner_list, list) {
1414 		if (item->q == q && item->binder_type == binder_type) {
1415 			list_del(&item->list);
1416 			kfree(item);
1417 			return;
1418 		}
1419 	}
1420 	WARN_ON(1);
1421 }
1422 
1423 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1424 		      struct tcf_block_ext_info *ei,
1425 		      struct netlink_ext_ack *extack)
1426 {
1427 	struct net *net = qdisc_net(q);
1428 	struct tcf_block *block = NULL;
1429 	int err;
1430 
1431 	if (ei->block_index)
1432 		/* block_index not 0 means the shared block is requested */
1433 		block = tcf_block_refcnt_get(net, ei->block_index);
1434 
1435 	if (!block) {
1436 		block = tcf_block_create(net, q, ei->block_index, extack);
1437 		if (IS_ERR(block))
1438 			return PTR_ERR(block);
1439 		if (tcf_block_shared(block)) {
1440 			err = tcf_block_insert(block, net, extack);
1441 			if (err)
1442 				goto err_block_insert;
1443 		}
1444 	}
1445 
1446 	err = tcf_block_owner_add(block, q, ei->binder_type);
1447 	if (err)
1448 		goto err_block_owner_add;
1449 
1450 	tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1451 
1452 	err = tcf_chain0_head_change_cb_add(block, ei, extack);
1453 	if (err)
1454 		goto err_chain0_head_change_cb_add;
1455 
1456 	err = tcf_block_offload_bind(block, q, ei, extack);
1457 	if (err)
1458 		goto err_block_offload_bind;
1459 
1460 	*p_block = block;
1461 	return 0;
1462 
1463 err_block_offload_bind:
1464 	tcf_chain0_head_change_cb_del(block, ei);
1465 err_chain0_head_change_cb_add:
1466 	tcf_block_owner_del(block, q, ei->binder_type);
1467 err_block_owner_add:
1468 err_block_insert:
1469 	tcf_block_refcnt_put(block, true);
1470 	return err;
1471 }
1472 EXPORT_SYMBOL(tcf_block_get_ext);
1473 
1474 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1475 {
1476 	struct tcf_proto __rcu **p_filter_chain = priv;
1477 
1478 	rcu_assign_pointer(*p_filter_chain, tp_head);
1479 }
1480 
1481 int tcf_block_get(struct tcf_block **p_block,
1482 		  struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1483 		  struct netlink_ext_ack *extack)
1484 {
1485 	struct tcf_block_ext_info ei = {
1486 		.chain_head_change = tcf_chain_head_change_dflt,
1487 		.chain_head_change_priv = p_filter_chain,
1488 	};
1489 
1490 	WARN_ON(!p_filter_chain);
1491 	return tcf_block_get_ext(p_block, q, &ei, extack);
1492 }
1493 EXPORT_SYMBOL(tcf_block_get);
1494 
1495 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1496  * actions should be all removed after flushing.
1497  */
1498 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1499 		       struct tcf_block_ext_info *ei)
1500 {
1501 	if (!block)
1502 		return;
1503 	tcf_chain0_head_change_cb_del(block, ei);
1504 	tcf_block_owner_del(block, q, ei->binder_type);
1505 
1506 	__tcf_block_put(block, q, ei, true);
1507 }
1508 EXPORT_SYMBOL(tcf_block_put_ext);
1509 
1510 void tcf_block_put(struct tcf_block *block)
1511 {
1512 	struct tcf_block_ext_info ei = {0, };
1513 
1514 	if (!block)
1515 		return;
1516 	tcf_block_put_ext(block, block->q, &ei);
1517 }
1518 
1519 EXPORT_SYMBOL(tcf_block_put);
1520 
1521 static int
1522 tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1523 			    void *cb_priv, bool add, bool offload_in_use,
1524 			    struct netlink_ext_ack *extack)
1525 {
1526 	struct tcf_chain *chain, *chain_prev;
1527 	struct tcf_proto *tp, *tp_prev;
1528 	int err;
1529 
1530 	for (chain = __tcf_get_next_chain(block, NULL);
1531 	     chain;
1532 	     chain_prev = chain,
1533 		     chain = __tcf_get_next_chain(block, chain),
1534 		     tcf_chain_put(chain_prev)) {
1535 		for (tp = __tcf_get_next_proto(chain, NULL); tp;
1536 		     tp_prev = tp,
1537 			     tp = __tcf_get_next_proto(chain, tp),
1538 			     tcf_proto_put(tp_prev, true, NULL)) {
1539 			if (tp->ops->reoffload) {
1540 				err = tp->ops->reoffload(tp, add, cb, cb_priv,
1541 							 extack);
1542 				if (err && add)
1543 					goto err_playback_remove;
1544 			} else if (add && offload_in_use) {
1545 				err = -EOPNOTSUPP;
1546 				NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1547 				goto err_playback_remove;
1548 			}
1549 		}
1550 	}
1551 
1552 	return 0;
1553 
1554 err_playback_remove:
1555 	tcf_proto_put(tp, true, NULL);
1556 	tcf_chain_put(chain);
1557 	tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1558 				    extack);
1559 	return err;
1560 }
1561 
1562 static int tcf_block_bind(struct tcf_block *block,
1563 			  struct flow_block_offload *bo)
1564 {
1565 	struct flow_block_cb *block_cb, *next;
1566 	int err, i = 0;
1567 
1568 	list_for_each_entry(block_cb, &bo->cb_list, list) {
1569 		err = tcf_block_playback_offloads(block, block_cb->cb,
1570 						  block_cb->cb_priv, true,
1571 						  tcf_block_offload_in_use(block),
1572 						  bo->extack);
1573 		if (err)
1574 			goto err_unroll;
1575 
1576 		i++;
1577 	}
1578 	list_splice(&bo->cb_list, &block->flow_block.cb_list);
1579 
1580 	return 0;
1581 
1582 err_unroll:
1583 	list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1584 		if (i-- > 0) {
1585 			list_del(&block_cb->list);
1586 			tcf_block_playback_offloads(block, block_cb->cb,
1587 						    block_cb->cb_priv, false,
1588 						    tcf_block_offload_in_use(block),
1589 						    NULL);
1590 		}
1591 		flow_block_cb_free(block_cb);
1592 	}
1593 
1594 	return err;
1595 }
1596 
1597 static void tcf_block_unbind(struct tcf_block *block,
1598 			     struct flow_block_offload *bo)
1599 {
1600 	struct flow_block_cb *block_cb, *next;
1601 
1602 	list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1603 		tcf_block_playback_offloads(block, block_cb->cb,
1604 					    block_cb->cb_priv, false,
1605 					    tcf_block_offload_in_use(block),
1606 					    NULL);
1607 		list_del(&block_cb->list);
1608 		flow_block_cb_free(block_cb);
1609 	}
1610 }
1611 
1612 static int tcf_block_setup(struct tcf_block *block,
1613 			   struct flow_block_offload *bo)
1614 {
1615 	int err;
1616 
1617 	switch (bo->command) {
1618 	case FLOW_BLOCK_BIND:
1619 		err = tcf_block_bind(block, bo);
1620 		break;
1621 	case FLOW_BLOCK_UNBIND:
1622 		err = 0;
1623 		tcf_block_unbind(block, bo);
1624 		break;
1625 	default:
1626 		WARN_ON_ONCE(1);
1627 		err = -EOPNOTSUPP;
1628 	}
1629 
1630 	return err;
1631 }
1632 
1633 /* Main classifier routine: scans classifier chain attached
1634  * to this qdisc, (optionally) tests for protocol and asks
1635  * specific classifiers.
1636  */
1637 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1638 		 struct tcf_result *res, bool compat_mode)
1639 {
1640 #ifdef CONFIG_NET_CLS_ACT
1641 	const int max_reclassify_loop = 4;
1642 	const struct tcf_proto *orig_tp = tp;
1643 	const struct tcf_proto *first_tp;
1644 	int limit = 0;
1645 
1646 reclassify:
1647 #endif
1648 	for (; tp; tp = rcu_dereference_bh(tp->next)) {
1649 		__be16 protocol = tc_skb_protocol(skb);
1650 		int err;
1651 
1652 		if (tp->protocol != protocol &&
1653 		    tp->protocol != htons(ETH_P_ALL))
1654 			continue;
1655 
1656 		err = tp->classify(skb, tp, res);
1657 #ifdef CONFIG_NET_CLS_ACT
1658 		if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1659 			first_tp = orig_tp;
1660 			goto reset;
1661 		} else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1662 			first_tp = res->goto_tp;
1663 			goto reset;
1664 		}
1665 #endif
1666 		if (err >= 0)
1667 			return err;
1668 	}
1669 
1670 	return TC_ACT_UNSPEC; /* signal: continue lookup */
1671 #ifdef CONFIG_NET_CLS_ACT
1672 reset:
1673 	if (unlikely(limit++ >= max_reclassify_loop)) {
1674 		net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1675 				       tp->chain->block->index,
1676 				       tp->prio & 0xffff,
1677 				       ntohs(tp->protocol));
1678 		return TC_ACT_SHOT;
1679 	}
1680 
1681 	tp = first_tp;
1682 	goto reclassify;
1683 #endif
1684 }
1685 EXPORT_SYMBOL(tcf_classify);
1686 
1687 struct tcf_chain_info {
1688 	struct tcf_proto __rcu **pprev;
1689 	struct tcf_proto __rcu *next;
1690 };
1691 
1692 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1693 					   struct tcf_chain_info *chain_info)
1694 {
1695 	return tcf_chain_dereference(*chain_info->pprev, chain);
1696 }
1697 
1698 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1699 			       struct tcf_chain_info *chain_info,
1700 			       struct tcf_proto *tp)
1701 {
1702 	if (chain->flushing)
1703 		return -EAGAIN;
1704 
1705 	if (*chain_info->pprev == chain->filter_chain)
1706 		tcf_chain0_head_change(chain, tp);
1707 	tcf_proto_get(tp);
1708 	RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1709 	rcu_assign_pointer(*chain_info->pprev, tp);
1710 
1711 	return 0;
1712 }
1713 
1714 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1715 				struct tcf_chain_info *chain_info,
1716 				struct tcf_proto *tp)
1717 {
1718 	struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1719 
1720 	tcf_proto_mark_delete(tp);
1721 	if (tp == chain->filter_chain)
1722 		tcf_chain0_head_change(chain, next);
1723 	RCU_INIT_POINTER(*chain_info->pprev, next);
1724 }
1725 
1726 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1727 					   struct tcf_chain_info *chain_info,
1728 					   u32 protocol, u32 prio,
1729 					   bool prio_allocate);
1730 
1731 /* Try to insert new proto.
1732  * If proto with specified priority already exists, free new proto
1733  * and return existing one.
1734  */
1735 
1736 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1737 						    struct tcf_proto *tp_new,
1738 						    u32 protocol, u32 prio,
1739 						    bool rtnl_held)
1740 {
1741 	struct tcf_chain_info chain_info;
1742 	struct tcf_proto *tp;
1743 	int err = 0;
1744 
1745 	mutex_lock(&chain->filter_chain_lock);
1746 
1747 	tp = tcf_chain_tp_find(chain, &chain_info,
1748 			       protocol, prio, false);
1749 	if (!tp)
1750 		err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1751 	mutex_unlock(&chain->filter_chain_lock);
1752 
1753 	if (tp) {
1754 		tcf_proto_destroy(tp_new, rtnl_held, NULL);
1755 		tp_new = tp;
1756 	} else if (err) {
1757 		tcf_proto_destroy(tp_new, rtnl_held, NULL);
1758 		tp_new = ERR_PTR(err);
1759 	}
1760 
1761 	return tp_new;
1762 }
1763 
1764 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1765 				      struct tcf_proto *tp, bool rtnl_held,
1766 				      struct netlink_ext_ack *extack)
1767 {
1768 	struct tcf_chain_info chain_info;
1769 	struct tcf_proto *tp_iter;
1770 	struct tcf_proto **pprev;
1771 	struct tcf_proto *next;
1772 
1773 	mutex_lock(&chain->filter_chain_lock);
1774 
1775 	/* Atomically find and remove tp from chain. */
1776 	for (pprev = &chain->filter_chain;
1777 	     (tp_iter = tcf_chain_dereference(*pprev, chain));
1778 	     pprev = &tp_iter->next) {
1779 		if (tp_iter == tp) {
1780 			chain_info.pprev = pprev;
1781 			chain_info.next = tp_iter->next;
1782 			WARN_ON(tp_iter->deleting);
1783 			break;
1784 		}
1785 	}
1786 	/* Verify that tp still exists and no new filters were inserted
1787 	 * concurrently.
1788 	 * Mark tp for deletion if it is empty.
1789 	 */
1790 	if (!tp_iter || !tcf_proto_check_delete(tp, rtnl_held)) {
1791 		mutex_unlock(&chain->filter_chain_lock);
1792 		return;
1793 	}
1794 
1795 	next = tcf_chain_dereference(chain_info.next, chain);
1796 	if (tp == chain->filter_chain)
1797 		tcf_chain0_head_change(chain, next);
1798 	RCU_INIT_POINTER(*chain_info.pprev, next);
1799 	mutex_unlock(&chain->filter_chain_lock);
1800 
1801 	tcf_proto_put(tp, rtnl_held, extack);
1802 }
1803 
1804 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1805 					   struct tcf_chain_info *chain_info,
1806 					   u32 protocol, u32 prio,
1807 					   bool prio_allocate)
1808 {
1809 	struct tcf_proto **pprev;
1810 	struct tcf_proto *tp;
1811 
1812 	/* Check the chain for existence of proto-tcf with this priority */
1813 	for (pprev = &chain->filter_chain;
1814 	     (tp = tcf_chain_dereference(*pprev, chain));
1815 	     pprev = &tp->next) {
1816 		if (tp->prio >= prio) {
1817 			if (tp->prio == prio) {
1818 				if (prio_allocate ||
1819 				    (tp->protocol != protocol && protocol))
1820 					return ERR_PTR(-EINVAL);
1821 			} else {
1822 				tp = NULL;
1823 			}
1824 			break;
1825 		}
1826 	}
1827 	chain_info->pprev = pprev;
1828 	if (tp) {
1829 		chain_info->next = tp->next;
1830 		tcf_proto_get(tp);
1831 	} else {
1832 		chain_info->next = NULL;
1833 	}
1834 	return tp;
1835 }
1836 
1837 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1838 			 struct tcf_proto *tp, struct tcf_block *block,
1839 			 struct Qdisc *q, u32 parent, void *fh,
1840 			 u32 portid, u32 seq, u16 flags, int event,
1841 			 bool rtnl_held)
1842 {
1843 	struct tcmsg *tcm;
1844 	struct nlmsghdr  *nlh;
1845 	unsigned char *b = skb_tail_pointer(skb);
1846 
1847 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1848 	if (!nlh)
1849 		goto out_nlmsg_trim;
1850 	tcm = nlmsg_data(nlh);
1851 	tcm->tcm_family = AF_UNSPEC;
1852 	tcm->tcm__pad1 = 0;
1853 	tcm->tcm__pad2 = 0;
1854 	if (q) {
1855 		tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1856 		tcm->tcm_parent = parent;
1857 	} else {
1858 		tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1859 		tcm->tcm_block_index = block->index;
1860 	}
1861 	tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1862 	if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1863 		goto nla_put_failure;
1864 	if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1865 		goto nla_put_failure;
1866 	if (!fh) {
1867 		tcm->tcm_handle = 0;
1868 	} else {
1869 		if (tp->ops->dump &&
1870 		    tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
1871 			goto nla_put_failure;
1872 	}
1873 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1874 	return skb->len;
1875 
1876 out_nlmsg_trim:
1877 nla_put_failure:
1878 	nlmsg_trim(skb, b);
1879 	return -1;
1880 }
1881 
1882 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1883 			  struct nlmsghdr *n, struct tcf_proto *tp,
1884 			  struct tcf_block *block, struct Qdisc *q,
1885 			  u32 parent, void *fh, int event, bool unicast,
1886 			  bool rtnl_held)
1887 {
1888 	struct sk_buff *skb;
1889 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1890 	int err = 0;
1891 
1892 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1893 	if (!skb)
1894 		return -ENOBUFS;
1895 
1896 	if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1897 			  n->nlmsg_seq, n->nlmsg_flags, event,
1898 			  rtnl_held) <= 0) {
1899 		kfree_skb(skb);
1900 		return -EINVAL;
1901 	}
1902 
1903 	if (unicast)
1904 		err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1905 	else
1906 		err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1907 				     n->nlmsg_flags & NLM_F_ECHO);
1908 
1909 	if (err > 0)
1910 		err = 0;
1911 	return err;
1912 }
1913 
1914 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1915 			      struct nlmsghdr *n, struct tcf_proto *tp,
1916 			      struct tcf_block *block, struct Qdisc *q,
1917 			      u32 parent, void *fh, bool unicast, bool *last,
1918 			      bool rtnl_held, struct netlink_ext_ack *extack)
1919 {
1920 	struct sk_buff *skb;
1921 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1922 	int err;
1923 
1924 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1925 	if (!skb)
1926 		return -ENOBUFS;
1927 
1928 	if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1929 			  n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
1930 			  rtnl_held) <= 0) {
1931 		NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1932 		kfree_skb(skb);
1933 		return -EINVAL;
1934 	}
1935 
1936 	err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
1937 	if (err) {
1938 		kfree_skb(skb);
1939 		return err;
1940 	}
1941 
1942 	if (unicast)
1943 		err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1944 	else
1945 		err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1946 				     n->nlmsg_flags & NLM_F_ECHO);
1947 	if (err < 0)
1948 		NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1949 
1950 	if (err > 0)
1951 		err = 0;
1952 	return err;
1953 }
1954 
1955 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1956 				 struct tcf_block *block, struct Qdisc *q,
1957 				 u32 parent, struct nlmsghdr *n,
1958 				 struct tcf_chain *chain, int event,
1959 				 bool rtnl_held)
1960 {
1961 	struct tcf_proto *tp;
1962 
1963 	for (tp = tcf_get_next_proto(chain, NULL, rtnl_held);
1964 	     tp; tp = tcf_get_next_proto(chain, tp, rtnl_held))
1965 		tfilter_notify(net, oskb, n, tp, block,
1966 			       q, parent, NULL, event, false, rtnl_held);
1967 }
1968 
1969 static void tfilter_put(struct tcf_proto *tp, void *fh)
1970 {
1971 	if (tp->ops->put && fh)
1972 		tp->ops->put(tp, fh);
1973 }
1974 
1975 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1976 			  struct netlink_ext_ack *extack)
1977 {
1978 	struct net *net = sock_net(skb->sk);
1979 	struct nlattr *tca[TCA_MAX + 1];
1980 	struct tcmsg *t;
1981 	u32 protocol;
1982 	u32 prio;
1983 	bool prio_allocate;
1984 	u32 parent;
1985 	u32 chain_index;
1986 	struct Qdisc *q = NULL;
1987 	struct tcf_chain_info chain_info;
1988 	struct tcf_chain *chain = NULL;
1989 	struct tcf_block *block;
1990 	struct tcf_proto *tp;
1991 	unsigned long cl;
1992 	void *fh;
1993 	int err;
1994 	int tp_created;
1995 	bool rtnl_held = false;
1996 
1997 	if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1998 		return -EPERM;
1999 
2000 replay:
2001 	tp_created = 0;
2002 
2003 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2004 				     rtm_tca_policy, extack);
2005 	if (err < 0)
2006 		return err;
2007 
2008 	t = nlmsg_data(n);
2009 	protocol = TC_H_MIN(t->tcm_info);
2010 	prio = TC_H_MAJ(t->tcm_info);
2011 	prio_allocate = false;
2012 	parent = t->tcm_parent;
2013 	tp = NULL;
2014 	cl = 0;
2015 	block = NULL;
2016 
2017 	if (prio == 0) {
2018 		/* If no priority is provided by the user,
2019 		 * we allocate one.
2020 		 */
2021 		if (n->nlmsg_flags & NLM_F_CREATE) {
2022 			prio = TC_H_MAKE(0x80000000U, 0U);
2023 			prio_allocate = true;
2024 		} else {
2025 			NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2026 			return -ENOENT;
2027 		}
2028 	}
2029 
2030 	/* Find head of filter chain. */
2031 
2032 	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2033 	if (err)
2034 		return err;
2035 
2036 	/* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2037 	 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2038 	 * type is not specified, classifier is not unlocked.
2039 	 */
2040 	if (rtnl_held ||
2041 	    (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2042 	    !tca[TCA_KIND] || !tcf_proto_is_unlocked(nla_data(tca[TCA_KIND]))) {
2043 		rtnl_held = true;
2044 		rtnl_lock();
2045 	}
2046 
2047 	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2048 	if (err)
2049 		goto errout;
2050 
2051 	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2052 				 extack);
2053 	if (IS_ERR(block)) {
2054 		err = PTR_ERR(block);
2055 		goto errout;
2056 	}
2057 
2058 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2059 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2060 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2061 		err = -EINVAL;
2062 		goto errout;
2063 	}
2064 	chain = tcf_chain_get(block, chain_index, true);
2065 	if (!chain) {
2066 		NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2067 		err = -ENOMEM;
2068 		goto errout;
2069 	}
2070 
2071 	mutex_lock(&chain->filter_chain_lock);
2072 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2073 			       prio, prio_allocate);
2074 	if (IS_ERR(tp)) {
2075 		NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2076 		err = PTR_ERR(tp);
2077 		goto errout_locked;
2078 	}
2079 
2080 	if (tp == NULL) {
2081 		struct tcf_proto *tp_new = NULL;
2082 
2083 		if (chain->flushing) {
2084 			err = -EAGAIN;
2085 			goto errout_locked;
2086 		}
2087 
2088 		/* Proto-tcf does not exist, create new one */
2089 
2090 		if (tca[TCA_KIND] == NULL || !protocol) {
2091 			NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2092 			err = -EINVAL;
2093 			goto errout_locked;
2094 		}
2095 
2096 		if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2097 			NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2098 			err = -ENOENT;
2099 			goto errout_locked;
2100 		}
2101 
2102 		if (prio_allocate)
2103 			prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2104 							       &chain_info));
2105 
2106 		mutex_unlock(&chain->filter_chain_lock);
2107 		tp_new = tcf_proto_create(nla_data(tca[TCA_KIND]),
2108 					  protocol, prio, chain, rtnl_held,
2109 					  extack);
2110 		if (IS_ERR(tp_new)) {
2111 			err = PTR_ERR(tp_new);
2112 			goto errout_tp;
2113 		}
2114 
2115 		tp_created = 1;
2116 		tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2117 						rtnl_held);
2118 		if (IS_ERR(tp)) {
2119 			err = PTR_ERR(tp);
2120 			goto errout_tp;
2121 		}
2122 	} else {
2123 		mutex_unlock(&chain->filter_chain_lock);
2124 	}
2125 
2126 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2127 		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2128 		err = -EINVAL;
2129 		goto errout;
2130 	}
2131 
2132 	fh = tp->ops->get(tp, t->tcm_handle);
2133 
2134 	if (!fh) {
2135 		if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2136 			NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2137 			err = -ENOENT;
2138 			goto errout;
2139 		}
2140 	} else if (n->nlmsg_flags & NLM_F_EXCL) {
2141 		tfilter_put(tp, fh);
2142 		NL_SET_ERR_MSG(extack, "Filter already exists");
2143 		err = -EEXIST;
2144 		goto errout;
2145 	}
2146 
2147 	if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2148 		NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2149 		err = -EINVAL;
2150 		goto errout;
2151 	}
2152 
2153 	err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2154 			      n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
2155 			      rtnl_held, extack);
2156 	if (err == 0) {
2157 		tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2158 			       RTM_NEWTFILTER, false, rtnl_held);
2159 		tfilter_put(tp, fh);
2160 		/* q pointer is NULL for shared blocks */
2161 		if (q)
2162 			q->flags &= ~TCQ_F_CAN_BYPASS;
2163 	}
2164 
2165 errout:
2166 	if (err && tp_created)
2167 		tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2168 errout_tp:
2169 	if (chain) {
2170 		if (tp && !IS_ERR(tp))
2171 			tcf_proto_put(tp, rtnl_held, NULL);
2172 		if (!tp_created)
2173 			tcf_chain_put(chain);
2174 	}
2175 	tcf_block_release(q, block, rtnl_held);
2176 
2177 	if (rtnl_held)
2178 		rtnl_unlock();
2179 
2180 	if (err == -EAGAIN) {
2181 		/* Take rtnl lock in case EAGAIN is caused by concurrent flush
2182 		 * of target chain.
2183 		 */
2184 		rtnl_held = true;
2185 		/* Replay the request. */
2186 		goto replay;
2187 	}
2188 	return err;
2189 
2190 errout_locked:
2191 	mutex_unlock(&chain->filter_chain_lock);
2192 	goto errout;
2193 }
2194 
2195 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2196 			  struct netlink_ext_ack *extack)
2197 {
2198 	struct net *net = sock_net(skb->sk);
2199 	struct nlattr *tca[TCA_MAX + 1];
2200 	struct tcmsg *t;
2201 	u32 protocol;
2202 	u32 prio;
2203 	u32 parent;
2204 	u32 chain_index;
2205 	struct Qdisc *q = NULL;
2206 	struct tcf_chain_info chain_info;
2207 	struct tcf_chain *chain = NULL;
2208 	struct tcf_block *block = NULL;
2209 	struct tcf_proto *tp = NULL;
2210 	unsigned long cl = 0;
2211 	void *fh = NULL;
2212 	int err;
2213 	bool rtnl_held = false;
2214 
2215 	if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2216 		return -EPERM;
2217 
2218 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2219 				     rtm_tca_policy, extack);
2220 	if (err < 0)
2221 		return err;
2222 
2223 	t = nlmsg_data(n);
2224 	protocol = TC_H_MIN(t->tcm_info);
2225 	prio = TC_H_MAJ(t->tcm_info);
2226 	parent = t->tcm_parent;
2227 
2228 	if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2229 		NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2230 		return -ENOENT;
2231 	}
2232 
2233 	/* Find head of filter chain. */
2234 
2235 	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2236 	if (err)
2237 		return err;
2238 
2239 	/* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2240 	 * found), qdisc is not unlocked, classifier type is not specified,
2241 	 * classifier is not unlocked.
2242 	 */
2243 	if (!prio ||
2244 	    (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2245 	    !tca[TCA_KIND] || !tcf_proto_is_unlocked(nla_data(tca[TCA_KIND]))) {
2246 		rtnl_held = true;
2247 		rtnl_lock();
2248 	}
2249 
2250 	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2251 	if (err)
2252 		goto errout;
2253 
2254 	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2255 				 extack);
2256 	if (IS_ERR(block)) {
2257 		err = PTR_ERR(block);
2258 		goto errout;
2259 	}
2260 
2261 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2262 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2263 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2264 		err = -EINVAL;
2265 		goto errout;
2266 	}
2267 	chain = tcf_chain_get(block, chain_index, false);
2268 	if (!chain) {
2269 		/* User requested flush on non-existent chain. Nothing to do,
2270 		 * so just return success.
2271 		 */
2272 		if (prio == 0) {
2273 			err = 0;
2274 			goto errout;
2275 		}
2276 		NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2277 		err = -ENOENT;
2278 		goto errout;
2279 	}
2280 
2281 	if (prio == 0) {
2282 		tfilter_notify_chain(net, skb, block, q, parent, n,
2283 				     chain, RTM_DELTFILTER, rtnl_held);
2284 		tcf_chain_flush(chain, rtnl_held);
2285 		err = 0;
2286 		goto errout;
2287 	}
2288 
2289 	mutex_lock(&chain->filter_chain_lock);
2290 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2291 			       prio, false);
2292 	if (!tp || IS_ERR(tp)) {
2293 		NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2294 		err = tp ? PTR_ERR(tp) : -ENOENT;
2295 		goto errout_locked;
2296 	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2297 		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2298 		err = -EINVAL;
2299 		goto errout_locked;
2300 	} else if (t->tcm_handle == 0) {
2301 		tcf_chain_tp_remove(chain, &chain_info, tp);
2302 		mutex_unlock(&chain->filter_chain_lock);
2303 
2304 		tcf_proto_put(tp, rtnl_held, NULL);
2305 		tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2306 			       RTM_DELTFILTER, false, rtnl_held);
2307 		err = 0;
2308 		goto errout;
2309 	}
2310 	mutex_unlock(&chain->filter_chain_lock);
2311 
2312 	fh = tp->ops->get(tp, t->tcm_handle);
2313 
2314 	if (!fh) {
2315 		NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2316 		err = -ENOENT;
2317 	} else {
2318 		bool last;
2319 
2320 		err = tfilter_del_notify(net, skb, n, tp, block,
2321 					 q, parent, fh, false, &last,
2322 					 rtnl_held, extack);
2323 
2324 		if (err)
2325 			goto errout;
2326 		if (last)
2327 			tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2328 	}
2329 
2330 errout:
2331 	if (chain) {
2332 		if (tp && !IS_ERR(tp))
2333 			tcf_proto_put(tp, rtnl_held, NULL);
2334 		tcf_chain_put(chain);
2335 	}
2336 	tcf_block_release(q, block, rtnl_held);
2337 
2338 	if (rtnl_held)
2339 		rtnl_unlock();
2340 
2341 	return err;
2342 
2343 errout_locked:
2344 	mutex_unlock(&chain->filter_chain_lock);
2345 	goto errout;
2346 }
2347 
2348 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2349 			  struct netlink_ext_ack *extack)
2350 {
2351 	struct net *net = sock_net(skb->sk);
2352 	struct nlattr *tca[TCA_MAX + 1];
2353 	struct tcmsg *t;
2354 	u32 protocol;
2355 	u32 prio;
2356 	u32 parent;
2357 	u32 chain_index;
2358 	struct Qdisc *q = NULL;
2359 	struct tcf_chain_info chain_info;
2360 	struct tcf_chain *chain = NULL;
2361 	struct tcf_block *block = NULL;
2362 	struct tcf_proto *tp = NULL;
2363 	unsigned long cl = 0;
2364 	void *fh = NULL;
2365 	int err;
2366 	bool rtnl_held = false;
2367 
2368 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2369 				     rtm_tca_policy, extack);
2370 	if (err < 0)
2371 		return err;
2372 
2373 	t = nlmsg_data(n);
2374 	protocol = TC_H_MIN(t->tcm_info);
2375 	prio = TC_H_MAJ(t->tcm_info);
2376 	parent = t->tcm_parent;
2377 
2378 	if (prio == 0) {
2379 		NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2380 		return -ENOENT;
2381 	}
2382 
2383 	/* Find head of filter chain. */
2384 
2385 	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2386 	if (err)
2387 		return err;
2388 
2389 	/* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2390 	 * unlocked, classifier type is not specified, classifier is not
2391 	 * unlocked.
2392 	 */
2393 	if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2394 	    !tca[TCA_KIND] || !tcf_proto_is_unlocked(nla_data(tca[TCA_KIND]))) {
2395 		rtnl_held = true;
2396 		rtnl_lock();
2397 	}
2398 
2399 	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2400 	if (err)
2401 		goto errout;
2402 
2403 	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2404 				 extack);
2405 	if (IS_ERR(block)) {
2406 		err = PTR_ERR(block);
2407 		goto errout;
2408 	}
2409 
2410 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2411 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2412 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2413 		err = -EINVAL;
2414 		goto errout;
2415 	}
2416 	chain = tcf_chain_get(block, chain_index, false);
2417 	if (!chain) {
2418 		NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2419 		err = -EINVAL;
2420 		goto errout;
2421 	}
2422 
2423 	mutex_lock(&chain->filter_chain_lock);
2424 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2425 			       prio, false);
2426 	mutex_unlock(&chain->filter_chain_lock);
2427 	if (!tp || IS_ERR(tp)) {
2428 		NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2429 		err = tp ? PTR_ERR(tp) : -ENOENT;
2430 		goto errout;
2431 	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2432 		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2433 		err = -EINVAL;
2434 		goto errout;
2435 	}
2436 
2437 	fh = tp->ops->get(tp, t->tcm_handle);
2438 
2439 	if (!fh) {
2440 		NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2441 		err = -ENOENT;
2442 	} else {
2443 		err = tfilter_notify(net, skb, n, tp, block, q, parent,
2444 				     fh, RTM_NEWTFILTER, true, rtnl_held);
2445 		if (err < 0)
2446 			NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2447 	}
2448 
2449 	tfilter_put(tp, fh);
2450 errout:
2451 	if (chain) {
2452 		if (tp && !IS_ERR(tp))
2453 			tcf_proto_put(tp, rtnl_held, NULL);
2454 		tcf_chain_put(chain);
2455 	}
2456 	tcf_block_release(q, block, rtnl_held);
2457 
2458 	if (rtnl_held)
2459 		rtnl_unlock();
2460 
2461 	return err;
2462 }
2463 
2464 struct tcf_dump_args {
2465 	struct tcf_walker w;
2466 	struct sk_buff *skb;
2467 	struct netlink_callback *cb;
2468 	struct tcf_block *block;
2469 	struct Qdisc *q;
2470 	u32 parent;
2471 };
2472 
2473 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2474 {
2475 	struct tcf_dump_args *a = (void *)arg;
2476 	struct net *net = sock_net(a->skb->sk);
2477 
2478 	return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2479 			     n, NETLINK_CB(a->cb->skb).portid,
2480 			     a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2481 			     RTM_NEWTFILTER, true);
2482 }
2483 
2484 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2485 			   struct sk_buff *skb, struct netlink_callback *cb,
2486 			   long index_start, long *p_index)
2487 {
2488 	struct net *net = sock_net(skb->sk);
2489 	struct tcf_block *block = chain->block;
2490 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2491 	struct tcf_proto *tp, *tp_prev;
2492 	struct tcf_dump_args arg;
2493 
2494 	for (tp = __tcf_get_next_proto(chain, NULL);
2495 	     tp;
2496 	     tp_prev = tp,
2497 		     tp = __tcf_get_next_proto(chain, tp),
2498 		     tcf_proto_put(tp_prev, true, NULL),
2499 		     (*p_index)++) {
2500 		if (*p_index < index_start)
2501 			continue;
2502 		if (TC_H_MAJ(tcm->tcm_info) &&
2503 		    TC_H_MAJ(tcm->tcm_info) != tp->prio)
2504 			continue;
2505 		if (TC_H_MIN(tcm->tcm_info) &&
2506 		    TC_H_MIN(tcm->tcm_info) != tp->protocol)
2507 			continue;
2508 		if (*p_index > index_start)
2509 			memset(&cb->args[1], 0,
2510 			       sizeof(cb->args) - sizeof(cb->args[0]));
2511 		if (cb->args[1] == 0) {
2512 			if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2513 					  NETLINK_CB(cb->skb).portid,
2514 					  cb->nlh->nlmsg_seq, NLM_F_MULTI,
2515 					  RTM_NEWTFILTER, true) <= 0)
2516 				goto errout;
2517 			cb->args[1] = 1;
2518 		}
2519 		if (!tp->ops->walk)
2520 			continue;
2521 		arg.w.fn = tcf_node_dump;
2522 		arg.skb = skb;
2523 		arg.cb = cb;
2524 		arg.block = block;
2525 		arg.q = q;
2526 		arg.parent = parent;
2527 		arg.w.stop = 0;
2528 		arg.w.skip = cb->args[1] - 1;
2529 		arg.w.count = 0;
2530 		arg.w.cookie = cb->args[2];
2531 		tp->ops->walk(tp, &arg.w, true);
2532 		cb->args[2] = arg.w.cookie;
2533 		cb->args[1] = arg.w.count + 1;
2534 		if (arg.w.stop)
2535 			goto errout;
2536 	}
2537 	return true;
2538 
2539 errout:
2540 	tcf_proto_put(tp, true, NULL);
2541 	return false;
2542 }
2543 
2544 /* called with RTNL */
2545 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2546 {
2547 	struct tcf_chain *chain, *chain_prev;
2548 	struct net *net = sock_net(skb->sk);
2549 	struct nlattr *tca[TCA_MAX + 1];
2550 	struct Qdisc *q = NULL;
2551 	struct tcf_block *block;
2552 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2553 	long index_start;
2554 	long index;
2555 	u32 parent;
2556 	int err;
2557 
2558 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2559 		return skb->len;
2560 
2561 	err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2562 				     NULL, cb->extack);
2563 	if (err)
2564 		return err;
2565 
2566 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2567 		block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2568 		if (!block)
2569 			goto out;
2570 		/* If we work with block index, q is NULL and parent value
2571 		 * will never be used in the following code. The check
2572 		 * in tcf_fill_node prevents it. However, compiler does not
2573 		 * see that far, so set parent to zero to silence the warning
2574 		 * about parent being uninitialized.
2575 		 */
2576 		parent = 0;
2577 	} else {
2578 		const struct Qdisc_class_ops *cops;
2579 		struct net_device *dev;
2580 		unsigned long cl = 0;
2581 
2582 		dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2583 		if (!dev)
2584 			return skb->len;
2585 
2586 		parent = tcm->tcm_parent;
2587 		if (!parent) {
2588 			q = dev->qdisc;
2589 			parent = q->handle;
2590 		} else {
2591 			q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2592 		}
2593 		if (!q)
2594 			goto out;
2595 		cops = q->ops->cl_ops;
2596 		if (!cops)
2597 			goto out;
2598 		if (!cops->tcf_block)
2599 			goto out;
2600 		if (TC_H_MIN(tcm->tcm_parent)) {
2601 			cl = cops->find(q, tcm->tcm_parent);
2602 			if (cl == 0)
2603 				goto out;
2604 		}
2605 		block = cops->tcf_block(q, cl, NULL);
2606 		if (!block)
2607 			goto out;
2608 		if (tcf_block_shared(block))
2609 			q = NULL;
2610 	}
2611 
2612 	index_start = cb->args[0];
2613 	index = 0;
2614 
2615 	for (chain = __tcf_get_next_chain(block, NULL);
2616 	     chain;
2617 	     chain_prev = chain,
2618 		     chain = __tcf_get_next_chain(block, chain),
2619 		     tcf_chain_put(chain_prev)) {
2620 		if (tca[TCA_CHAIN] &&
2621 		    nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2622 			continue;
2623 		if (!tcf_chain_dump(chain, q, parent, skb, cb,
2624 				    index_start, &index)) {
2625 			tcf_chain_put(chain);
2626 			err = -EMSGSIZE;
2627 			break;
2628 		}
2629 	}
2630 
2631 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2632 		tcf_block_refcnt_put(block, true);
2633 	cb->args[0] = index;
2634 
2635 out:
2636 	/* If we did no progress, the error (EMSGSIZE) is real */
2637 	if (skb->len == 0 && err)
2638 		return err;
2639 	return skb->len;
2640 }
2641 
2642 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2643 			      void *tmplt_priv, u32 chain_index,
2644 			      struct net *net, struct sk_buff *skb,
2645 			      struct tcf_block *block,
2646 			      u32 portid, u32 seq, u16 flags, int event)
2647 {
2648 	unsigned char *b = skb_tail_pointer(skb);
2649 	const struct tcf_proto_ops *ops;
2650 	struct nlmsghdr *nlh;
2651 	struct tcmsg *tcm;
2652 	void *priv;
2653 
2654 	ops = tmplt_ops;
2655 	priv = tmplt_priv;
2656 
2657 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2658 	if (!nlh)
2659 		goto out_nlmsg_trim;
2660 	tcm = nlmsg_data(nlh);
2661 	tcm->tcm_family = AF_UNSPEC;
2662 	tcm->tcm__pad1 = 0;
2663 	tcm->tcm__pad2 = 0;
2664 	tcm->tcm_handle = 0;
2665 	if (block->q) {
2666 		tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2667 		tcm->tcm_parent = block->q->handle;
2668 	} else {
2669 		tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2670 		tcm->tcm_block_index = block->index;
2671 	}
2672 
2673 	if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2674 		goto nla_put_failure;
2675 
2676 	if (ops) {
2677 		if (nla_put_string(skb, TCA_KIND, ops->kind))
2678 			goto nla_put_failure;
2679 		if (ops->tmplt_dump(skb, net, priv) < 0)
2680 			goto nla_put_failure;
2681 	}
2682 
2683 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2684 	return skb->len;
2685 
2686 out_nlmsg_trim:
2687 nla_put_failure:
2688 	nlmsg_trim(skb, b);
2689 	return -EMSGSIZE;
2690 }
2691 
2692 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2693 			   u32 seq, u16 flags, int event, bool unicast)
2694 {
2695 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2696 	struct tcf_block *block = chain->block;
2697 	struct net *net = block->net;
2698 	struct sk_buff *skb;
2699 	int err = 0;
2700 
2701 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2702 	if (!skb)
2703 		return -ENOBUFS;
2704 
2705 	if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2706 			       chain->index, net, skb, block, portid,
2707 			       seq, flags, event) <= 0) {
2708 		kfree_skb(skb);
2709 		return -EINVAL;
2710 	}
2711 
2712 	if (unicast)
2713 		err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2714 	else
2715 		err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2716 				     flags & NLM_F_ECHO);
2717 
2718 	if (err > 0)
2719 		err = 0;
2720 	return err;
2721 }
2722 
2723 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2724 				  void *tmplt_priv, u32 chain_index,
2725 				  struct tcf_block *block, struct sk_buff *oskb,
2726 				  u32 seq, u16 flags, bool unicast)
2727 {
2728 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2729 	struct net *net = block->net;
2730 	struct sk_buff *skb;
2731 
2732 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2733 	if (!skb)
2734 		return -ENOBUFS;
2735 
2736 	if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2737 			       block, portid, seq, flags, RTM_DELCHAIN) <= 0) {
2738 		kfree_skb(skb);
2739 		return -EINVAL;
2740 	}
2741 
2742 	if (unicast)
2743 		return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2744 
2745 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2746 }
2747 
2748 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2749 			      struct nlattr **tca,
2750 			      struct netlink_ext_ack *extack)
2751 {
2752 	const struct tcf_proto_ops *ops;
2753 	void *tmplt_priv;
2754 
2755 	/* If kind is not set, user did not specify template. */
2756 	if (!tca[TCA_KIND])
2757 		return 0;
2758 
2759 	ops = tcf_proto_lookup_ops(nla_data(tca[TCA_KIND]), true, extack);
2760 	if (IS_ERR(ops))
2761 		return PTR_ERR(ops);
2762 	if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2763 		NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2764 		return -EOPNOTSUPP;
2765 	}
2766 
2767 	tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2768 	if (IS_ERR(tmplt_priv)) {
2769 		module_put(ops->owner);
2770 		return PTR_ERR(tmplt_priv);
2771 	}
2772 	chain->tmplt_ops = ops;
2773 	chain->tmplt_priv = tmplt_priv;
2774 	return 0;
2775 }
2776 
2777 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2778 			       void *tmplt_priv)
2779 {
2780 	/* If template ops are set, no work to do for us. */
2781 	if (!tmplt_ops)
2782 		return;
2783 
2784 	tmplt_ops->tmplt_destroy(tmplt_priv);
2785 	module_put(tmplt_ops->owner);
2786 }
2787 
2788 /* Add/delete/get a chain */
2789 
2790 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2791 			struct netlink_ext_ack *extack)
2792 {
2793 	struct net *net = sock_net(skb->sk);
2794 	struct nlattr *tca[TCA_MAX + 1];
2795 	struct tcmsg *t;
2796 	u32 parent;
2797 	u32 chain_index;
2798 	struct Qdisc *q = NULL;
2799 	struct tcf_chain *chain = NULL;
2800 	struct tcf_block *block;
2801 	unsigned long cl;
2802 	int err;
2803 
2804 	if (n->nlmsg_type != RTM_GETCHAIN &&
2805 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2806 		return -EPERM;
2807 
2808 replay:
2809 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2810 				     rtm_tca_policy, extack);
2811 	if (err < 0)
2812 		return err;
2813 
2814 	t = nlmsg_data(n);
2815 	parent = t->tcm_parent;
2816 	cl = 0;
2817 
2818 	block = tcf_block_find(net, &q, &parent, &cl,
2819 			       t->tcm_ifindex, t->tcm_block_index, extack);
2820 	if (IS_ERR(block))
2821 		return PTR_ERR(block);
2822 
2823 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2824 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2825 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2826 		err = -EINVAL;
2827 		goto errout_block;
2828 	}
2829 
2830 	mutex_lock(&block->lock);
2831 	chain = tcf_chain_lookup(block, chain_index);
2832 	if (n->nlmsg_type == RTM_NEWCHAIN) {
2833 		if (chain) {
2834 			if (tcf_chain_held_by_acts_only(chain)) {
2835 				/* The chain exists only because there is
2836 				 * some action referencing it.
2837 				 */
2838 				tcf_chain_hold(chain);
2839 			} else {
2840 				NL_SET_ERR_MSG(extack, "Filter chain already exists");
2841 				err = -EEXIST;
2842 				goto errout_block_locked;
2843 			}
2844 		} else {
2845 			if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2846 				NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
2847 				err = -ENOENT;
2848 				goto errout_block_locked;
2849 			}
2850 			chain = tcf_chain_create(block, chain_index);
2851 			if (!chain) {
2852 				NL_SET_ERR_MSG(extack, "Failed to create filter chain");
2853 				err = -ENOMEM;
2854 				goto errout_block_locked;
2855 			}
2856 		}
2857 	} else {
2858 		if (!chain || tcf_chain_held_by_acts_only(chain)) {
2859 			NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2860 			err = -EINVAL;
2861 			goto errout_block_locked;
2862 		}
2863 		tcf_chain_hold(chain);
2864 	}
2865 
2866 	if (n->nlmsg_type == RTM_NEWCHAIN) {
2867 		/* Modifying chain requires holding parent block lock. In case
2868 		 * the chain was successfully added, take a reference to the
2869 		 * chain. This ensures that an empty chain does not disappear at
2870 		 * the end of this function.
2871 		 */
2872 		tcf_chain_hold(chain);
2873 		chain->explicitly_created = true;
2874 	}
2875 	mutex_unlock(&block->lock);
2876 
2877 	switch (n->nlmsg_type) {
2878 	case RTM_NEWCHAIN:
2879 		err = tc_chain_tmplt_add(chain, net, tca, extack);
2880 		if (err) {
2881 			tcf_chain_put_explicitly_created(chain);
2882 			goto errout;
2883 		}
2884 
2885 		tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2886 				RTM_NEWCHAIN, false);
2887 		break;
2888 	case RTM_DELCHAIN:
2889 		tfilter_notify_chain(net, skb, block, q, parent, n,
2890 				     chain, RTM_DELTFILTER, true);
2891 		/* Flush the chain first as the user requested chain removal. */
2892 		tcf_chain_flush(chain, true);
2893 		/* In case the chain was successfully deleted, put a reference
2894 		 * to the chain previously taken during addition.
2895 		 */
2896 		tcf_chain_put_explicitly_created(chain);
2897 		break;
2898 	case RTM_GETCHAIN:
2899 		err = tc_chain_notify(chain, skb, n->nlmsg_seq,
2900 				      n->nlmsg_seq, n->nlmsg_type, true);
2901 		if (err < 0)
2902 			NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2903 		break;
2904 	default:
2905 		err = -EOPNOTSUPP;
2906 		NL_SET_ERR_MSG(extack, "Unsupported message type");
2907 		goto errout;
2908 	}
2909 
2910 errout:
2911 	tcf_chain_put(chain);
2912 errout_block:
2913 	tcf_block_release(q, block, true);
2914 	if (err == -EAGAIN)
2915 		/* Replay the request. */
2916 		goto replay;
2917 	return err;
2918 
2919 errout_block_locked:
2920 	mutex_unlock(&block->lock);
2921 	goto errout_block;
2922 }
2923 
2924 /* called with RTNL */
2925 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2926 {
2927 	struct net *net = sock_net(skb->sk);
2928 	struct nlattr *tca[TCA_MAX + 1];
2929 	struct Qdisc *q = NULL;
2930 	struct tcf_block *block;
2931 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2932 	struct tcf_chain *chain;
2933 	long index_start;
2934 	long index;
2935 	u32 parent;
2936 	int err;
2937 
2938 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2939 		return skb->len;
2940 
2941 	err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2942 				     rtm_tca_policy, cb->extack);
2943 	if (err)
2944 		return err;
2945 
2946 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2947 		block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2948 		if (!block)
2949 			goto out;
2950 		/* If we work with block index, q is NULL and parent value
2951 		 * will never be used in the following code. The check
2952 		 * in tcf_fill_node prevents it. However, compiler does not
2953 		 * see that far, so set parent to zero to silence the warning
2954 		 * about parent being uninitialized.
2955 		 */
2956 		parent = 0;
2957 	} else {
2958 		const struct Qdisc_class_ops *cops;
2959 		struct net_device *dev;
2960 		unsigned long cl = 0;
2961 
2962 		dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2963 		if (!dev)
2964 			return skb->len;
2965 
2966 		parent = tcm->tcm_parent;
2967 		if (!parent) {
2968 			q = dev->qdisc;
2969 			parent = q->handle;
2970 		} else {
2971 			q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2972 		}
2973 		if (!q)
2974 			goto out;
2975 		cops = q->ops->cl_ops;
2976 		if (!cops)
2977 			goto out;
2978 		if (!cops->tcf_block)
2979 			goto out;
2980 		if (TC_H_MIN(tcm->tcm_parent)) {
2981 			cl = cops->find(q, tcm->tcm_parent);
2982 			if (cl == 0)
2983 				goto out;
2984 		}
2985 		block = cops->tcf_block(q, cl, NULL);
2986 		if (!block)
2987 			goto out;
2988 		if (tcf_block_shared(block))
2989 			q = NULL;
2990 	}
2991 
2992 	index_start = cb->args[0];
2993 	index = 0;
2994 
2995 	mutex_lock(&block->lock);
2996 	list_for_each_entry(chain, &block->chain_list, list) {
2997 		if ((tca[TCA_CHAIN] &&
2998 		     nla_get_u32(tca[TCA_CHAIN]) != chain->index))
2999 			continue;
3000 		if (index < index_start) {
3001 			index++;
3002 			continue;
3003 		}
3004 		if (tcf_chain_held_by_acts_only(chain))
3005 			continue;
3006 		err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3007 					 chain->index, net, skb, block,
3008 					 NETLINK_CB(cb->skb).portid,
3009 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3010 					 RTM_NEWCHAIN);
3011 		if (err <= 0)
3012 			break;
3013 		index++;
3014 	}
3015 	mutex_unlock(&block->lock);
3016 
3017 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3018 		tcf_block_refcnt_put(block, true);
3019 	cb->args[0] = index;
3020 
3021 out:
3022 	/* If we did no progress, the error (EMSGSIZE) is real */
3023 	if (skb->len == 0 && err)
3024 		return err;
3025 	return skb->len;
3026 }
3027 
3028 void tcf_exts_destroy(struct tcf_exts *exts)
3029 {
3030 #ifdef CONFIG_NET_CLS_ACT
3031 	tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3032 	kfree(exts->actions);
3033 	exts->nr_actions = 0;
3034 #endif
3035 }
3036 EXPORT_SYMBOL(tcf_exts_destroy);
3037 
3038 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3039 		      struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
3040 		      bool rtnl_held, struct netlink_ext_ack *extack)
3041 {
3042 #ifdef CONFIG_NET_CLS_ACT
3043 	{
3044 		struct tc_action *act;
3045 		size_t attr_size = 0;
3046 
3047 		if (exts->police && tb[exts->police]) {
3048 			act = tcf_action_init_1(net, tp, tb[exts->police],
3049 						rate_tlv, "police", ovr,
3050 						TCA_ACT_BIND, rtnl_held,
3051 						extack);
3052 			if (IS_ERR(act))
3053 				return PTR_ERR(act);
3054 
3055 			act->type = exts->type = TCA_OLD_COMPAT;
3056 			exts->actions[0] = act;
3057 			exts->nr_actions = 1;
3058 		} else if (exts->action && tb[exts->action]) {
3059 			int err;
3060 
3061 			err = tcf_action_init(net, tp, tb[exts->action],
3062 					      rate_tlv, NULL, ovr, TCA_ACT_BIND,
3063 					      exts->actions, &attr_size,
3064 					      rtnl_held, extack);
3065 			if (err < 0)
3066 				return err;
3067 			exts->nr_actions = err;
3068 		}
3069 	}
3070 #else
3071 	if ((exts->action && tb[exts->action]) ||
3072 	    (exts->police && tb[exts->police])) {
3073 		NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3074 		return -EOPNOTSUPP;
3075 	}
3076 #endif
3077 
3078 	return 0;
3079 }
3080 EXPORT_SYMBOL(tcf_exts_validate);
3081 
3082 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3083 {
3084 #ifdef CONFIG_NET_CLS_ACT
3085 	struct tcf_exts old = *dst;
3086 
3087 	*dst = *src;
3088 	tcf_exts_destroy(&old);
3089 #endif
3090 }
3091 EXPORT_SYMBOL(tcf_exts_change);
3092 
3093 #ifdef CONFIG_NET_CLS_ACT
3094 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3095 {
3096 	if (exts->nr_actions == 0)
3097 		return NULL;
3098 	else
3099 		return exts->actions[0];
3100 }
3101 #endif
3102 
3103 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3104 {
3105 #ifdef CONFIG_NET_CLS_ACT
3106 	struct nlattr *nest;
3107 
3108 	if (exts->action && tcf_exts_has_actions(exts)) {
3109 		/*
3110 		 * again for backward compatible mode - we want
3111 		 * to work with both old and new modes of entering
3112 		 * tc data even if iproute2  was newer - jhs
3113 		 */
3114 		if (exts->type != TCA_OLD_COMPAT) {
3115 			nest = nla_nest_start_noflag(skb, exts->action);
3116 			if (nest == NULL)
3117 				goto nla_put_failure;
3118 
3119 			if (tcf_action_dump(skb, exts->actions, 0, 0) < 0)
3120 				goto nla_put_failure;
3121 			nla_nest_end(skb, nest);
3122 		} else if (exts->police) {
3123 			struct tc_action *act = tcf_exts_first_act(exts);
3124 			nest = nla_nest_start_noflag(skb, exts->police);
3125 			if (nest == NULL || !act)
3126 				goto nla_put_failure;
3127 			if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3128 				goto nla_put_failure;
3129 			nla_nest_end(skb, nest);
3130 		}
3131 	}
3132 	return 0;
3133 
3134 nla_put_failure:
3135 	nla_nest_cancel(skb, nest);
3136 	return -1;
3137 #else
3138 	return 0;
3139 #endif
3140 }
3141 EXPORT_SYMBOL(tcf_exts_dump);
3142 
3143 
3144 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3145 {
3146 #ifdef CONFIG_NET_CLS_ACT
3147 	struct tc_action *a = tcf_exts_first_act(exts);
3148 	if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3149 		return -1;
3150 #endif
3151 	return 0;
3152 }
3153 EXPORT_SYMBOL(tcf_exts_dump_stats);
3154 
3155 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3156 		     void *type_data, bool err_stop)
3157 {
3158 	struct flow_block_cb *block_cb;
3159 	int ok_count = 0;
3160 	int err;
3161 
3162 	/* Make sure all netdevs sharing this block are offload-capable. */
3163 	if (block->nooffloaddevcnt && err_stop)
3164 		return -EOPNOTSUPP;
3165 
3166 	list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3167 		err = block_cb->cb(type, type_data, block_cb->cb_priv);
3168 		if (err) {
3169 			if (err_stop)
3170 				return err;
3171 		} else {
3172 			ok_count++;
3173 		}
3174 	}
3175 	return ok_count;
3176 }
3177 EXPORT_SYMBOL(tc_setup_cb_call);
3178 
3179 int tc_setup_flow_action(struct flow_action *flow_action,
3180 			 const struct tcf_exts *exts)
3181 {
3182 	const struct tc_action *act;
3183 	int i, j, k;
3184 
3185 	if (!exts)
3186 		return 0;
3187 
3188 	j = 0;
3189 	tcf_exts_for_each_action(i, act, exts) {
3190 		struct flow_action_entry *entry;
3191 
3192 		entry = &flow_action->entries[j];
3193 		if (is_tcf_gact_ok(act)) {
3194 			entry->id = FLOW_ACTION_ACCEPT;
3195 		} else if (is_tcf_gact_shot(act)) {
3196 			entry->id = FLOW_ACTION_DROP;
3197 		} else if (is_tcf_gact_trap(act)) {
3198 			entry->id = FLOW_ACTION_TRAP;
3199 		} else if (is_tcf_gact_goto_chain(act)) {
3200 			entry->id = FLOW_ACTION_GOTO;
3201 			entry->chain_index = tcf_gact_goto_chain_index(act);
3202 		} else if (is_tcf_mirred_egress_redirect(act)) {
3203 			entry->id = FLOW_ACTION_REDIRECT;
3204 			entry->dev = tcf_mirred_dev(act);
3205 		} else if (is_tcf_mirred_egress_mirror(act)) {
3206 			entry->id = FLOW_ACTION_MIRRED;
3207 			entry->dev = tcf_mirred_dev(act);
3208 		} else if (is_tcf_vlan(act)) {
3209 			switch (tcf_vlan_action(act)) {
3210 			case TCA_VLAN_ACT_PUSH:
3211 				entry->id = FLOW_ACTION_VLAN_PUSH;
3212 				entry->vlan.vid = tcf_vlan_push_vid(act);
3213 				entry->vlan.proto = tcf_vlan_push_proto(act);
3214 				entry->vlan.prio = tcf_vlan_push_prio(act);
3215 				break;
3216 			case TCA_VLAN_ACT_POP:
3217 				entry->id = FLOW_ACTION_VLAN_POP;
3218 				break;
3219 			case TCA_VLAN_ACT_MODIFY:
3220 				entry->id = FLOW_ACTION_VLAN_MANGLE;
3221 				entry->vlan.vid = tcf_vlan_push_vid(act);
3222 				entry->vlan.proto = tcf_vlan_push_proto(act);
3223 				entry->vlan.prio = tcf_vlan_push_prio(act);
3224 				break;
3225 			default:
3226 				goto err_out;
3227 			}
3228 		} else if (is_tcf_tunnel_set(act)) {
3229 			entry->id = FLOW_ACTION_TUNNEL_ENCAP;
3230 			entry->tunnel = tcf_tunnel_info(act);
3231 		} else if (is_tcf_tunnel_release(act)) {
3232 			entry->id = FLOW_ACTION_TUNNEL_DECAP;
3233 		} else if (is_tcf_pedit(act)) {
3234 			for (k = 0; k < tcf_pedit_nkeys(act); k++) {
3235 				switch (tcf_pedit_cmd(act, k)) {
3236 				case TCA_PEDIT_KEY_EX_CMD_SET:
3237 					entry->id = FLOW_ACTION_MANGLE;
3238 					break;
3239 				case TCA_PEDIT_KEY_EX_CMD_ADD:
3240 					entry->id = FLOW_ACTION_ADD;
3241 					break;
3242 				default:
3243 					goto err_out;
3244 				}
3245 				entry->mangle.htype = tcf_pedit_htype(act, k);
3246 				entry->mangle.mask = tcf_pedit_mask(act, k);
3247 				entry->mangle.val = tcf_pedit_val(act, k);
3248 				entry->mangle.offset = tcf_pedit_offset(act, k);
3249 				entry = &flow_action->entries[++j];
3250 			}
3251 		} else if (is_tcf_csum(act)) {
3252 			entry->id = FLOW_ACTION_CSUM;
3253 			entry->csum_flags = tcf_csum_update_flags(act);
3254 		} else if (is_tcf_skbedit_mark(act)) {
3255 			entry->id = FLOW_ACTION_MARK;
3256 			entry->mark = tcf_skbedit_mark(act);
3257 		} else if (is_tcf_sample(act)) {
3258 			entry->id = FLOW_ACTION_SAMPLE;
3259 			entry->sample.psample_group =
3260 				tcf_sample_psample_group(act);
3261 			entry->sample.trunc_size = tcf_sample_trunc_size(act);
3262 			entry->sample.truncate = tcf_sample_truncate(act);
3263 			entry->sample.rate = tcf_sample_rate(act);
3264 		} else if (is_tcf_police(act)) {
3265 			entry->id = FLOW_ACTION_POLICE;
3266 			entry->police.burst = tcf_police_tcfp_burst(act);
3267 			entry->police.rate_bytes_ps =
3268 				tcf_police_rate_bytes_ps(act);
3269 		} else if (is_tcf_ct(act)) {
3270 			entry->id = FLOW_ACTION_CT;
3271 			entry->ct.action = tcf_ct_action(act);
3272 			entry->ct.zone = tcf_ct_zone(act);
3273 		} else if (is_tcf_mpls(act)) {
3274 			switch (tcf_mpls_action(act)) {
3275 			case TCA_MPLS_ACT_PUSH:
3276 				entry->id = FLOW_ACTION_MPLS_PUSH;
3277 				entry->mpls_push.proto = tcf_mpls_proto(act);
3278 				entry->mpls_push.label = tcf_mpls_label(act);
3279 				entry->mpls_push.tc = tcf_mpls_tc(act);
3280 				entry->mpls_push.bos = tcf_mpls_bos(act);
3281 				entry->mpls_push.ttl = tcf_mpls_ttl(act);
3282 				break;
3283 			case TCA_MPLS_ACT_POP:
3284 				entry->id = FLOW_ACTION_MPLS_POP;
3285 				entry->mpls_pop.proto = tcf_mpls_proto(act);
3286 				break;
3287 			case TCA_MPLS_ACT_MODIFY:
3288 				entry->id = FLOW_ACTION_MPLS_MANGLE;
3289 				entry->mpls_mangle.label = tcf_mpls_label(act);
3290 				entry->mpls_mangle.tc = tcf_mpls_tc(act);
3291 				entry->mpls_mangle.bos = tcf_mpls_bos(act);
3292 				entry->mpls_mangle.ttl = tcf_mpls_ttl(act);
3293 				break;
3294 			default:
3295 				goto err_out;
3296 			}
3297 		} else {
3298 			goto err_out;
3299 		}
3300 
3301 		if (!is_tcf_pedit(act))
3302 			j++;
3303 	}
3304 	return 0;
3305 err_out:
3306 	return -EOPNOTSUPP;
3307 }
3308 EXPORT_SYMBOL(tc_setup_flow_action);
3309 
3310 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3311 {
3312 	unsigned int num_acts = 0;
3313 	struct tc_action *act;
3314 	int i;
3315 
3316 	tcf_exts_for_each_action(i, act, exts) {
3317 		if (is_tcf_pedit(act))
3318 			num_acts += tcf_pedit_nkeys(act);
3319 		else
3320 			num_acts++;
3321 	}
3322 	return num_acts;
3323 }
3324 EXPORT_SYMBOL(tcf_exts_num_actions);
3325 
3326 static __net_init int tcf_net_init(struct net *net)
3327 {
3328 	struct tcf_net *tn = net_generic(net, tcf_net_id);
3329 
3330 	spin_lock_init(&tn->idr_lock);
3331 	idr_init(&tn->idr);
3332 	return 0;
3333 }
3334 
3335 static void __net_exit tcf_net_exit(struct net *net)
3336 {
3337 	struct tcf_net *tn = net_generic(net, tcf_net_id);
3338 
3339 	idr_destroy(&tn->idr);
3340 }
3341 
3342 static struct pernet_operations tcf_net_ops = {
3343 	.init = tcf_net_init,
3344 	.exit = tcf_net_exit,
3345 	.id   = &tcf_net_id,
3346 	.size = sizeof(struct tcf_net),
3347 };
3348 
3349 static int __init tc_filter_init(void)
3350 {
3351 	int err;
3352 
3353 	tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3354 	if (!tc_filter_wq)
3355 		return -ENOMEM;
3356 
3357 	err = register_pernet_subsys(&tcf_net_ops);
3358 	if (err)
3359 		goto err_register_pernet_subsys;
3360 
3361 	err = rhashtable_init(&indr_setup_block_ht,
3362 			      &tc_indr_setup_block_ht_params);
3363 	if (err)
3364 		goto err_rhash_setup_block_ht;
3365 
3366 	rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3367 		      RTNL_FLAG_DOIT_UNLOCKED);
3368 	rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3369 		      RTNL_FLAG_DOIT_UNLOCKED);
3370 	rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3371 		      tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3372 	rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3373 	rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3374 	rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3375 		      tc_dump_chain, 0);
3376 
3377 	return 0;
3378 
3379 err_rhash_setup_block_ht:
3380 	unregister_pernet_subsys(&tcf_net_ops);
3381 err_register_pernet_subsys:
3382 	destroy_workqueue(tc_filter_wq);
3383 	return err;
3384 }
3385 
3386 subsys_initcall(tc_filter_init);
3387