xref: /openbmc/linux/net/sched/cls_api.c (revision 15e3ae36)
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 <linux/jhash.h>
25 #include <linux/rculist.h>
26 #include <net/net_namespace.h>
27 #include <net/sock.h>
28 #include <net/netlink.h>
29 #include <net/pkt_sched.h>
30 #include <net/pkt_cls.h>
31 #include <net/tc_act/tc_pedit.h>
32 #include <net/tc_act/tc_mirred.h>
33 #include <net/tc_act/tc_vlan.h>
34 #include <net/tc_act/tc_tunnel_key.h>
35 #include <net/tc_act/tc_csum.h>
36 #include <net/tc_act/tc_gact.h>
37 #include <net/tc_act/tc_police.h>
38 #include <net/tc_act/tc_sample.h>
39 #include <net/tc_act/tc_skbedit.h>
40 #include <net/tc_act/tc_ct.h>
41 #include <net/tc_act/tc_mpls.h>
42 #include <net/flow_offload.h>
43 
44 extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
45 
46 /* The list of all installed classifier types */
47 static LIST_HEAD(tcf_proto_base);
48 
49 /* Protects list of registered TC modules. It is pure SMP lock. */
50 static DEFINE_RWLOCK(cls_mod_lock);
51 
52 static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
53 {
54 	return jhash_3words(tp->chain->index, tp->prio,
55 			    (__force __u32)tp->protocol, 0);
56 }
57 
58 static void tcf_proto_signal_destroying(struct tcf_chain *chain,
59 					struct tcf_proto *tp)
60 {
61 	struct tcf_block *block = chain->block;
62 
63 	mutex_lock(&block->proto_destroy_lock);
64 	hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
65 		     destroy_obj_hashfn(tp));
66 	mutex_unlock(&block->proto_destroy_lock);
67 }
68 
69 static bool tcf_proto_cmp(const struct tcf_proto *tp1,
70 			  const struct tcf_proto *tp2)
71 {
72 	return tp1->chain->index == tp2->chain->index &&
73 	       tp1->prio == tp2->prio &&
74 	       tp1->protocol == tp2->protocol;
75 }
76 
77 static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
78 					struct tcf_proto *tp)
79 {
80 	u32 hash = destroy_obj_hashfn(tp);
81 	struct tcf_proto *iter;
82 	bool found = false;
83 
84 	rcu_read_lock();
85 	hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
86 				   destroy_ht_node, hash) {
87 		if (tcf_proto_cmp(tp, iter)) {
88 			found = true;
89 			break;
90 		}
91 	}
92 	rcu_read_unlock();
93 
94 	return found;
95 }
96 
97 static void
98 tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
99 {
100 	struct tcf_block *block = chain->block;
101 
102 	mutex_lock(&block->proto_destroy_lock);
103 	if (hash_hashed(&tp->destroy_ht_node))
104 		hash_del_rcu(&tp->destroy_ht_node);
105 	mutex_unlock(&block->proto_destroy_lock);
106 }
107 
108 /* Find classifier type by string name */
109 
110 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
111 {
112 	const struct tcf_proto_ops *t, *res = NULL;
113 
114 	if (kind) {
115 		read_lock(&cls_mod_lock);
116 		list_for_each_entry(t, &tcf_proto_base, head) {
117 			if (strcmp(kind, t->kind) == 0) {
118 				if (try_module_get(t->owner))
119 					res = t;
120 				break;
121 			}
122 		}
123 		read_unlock(&cls_mod_lock);
124 	}
125 	return res;
126 }
127 
128 static const struct tcf_proto_ops *
129 tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
130 		     struct netlink_ext_ack *extack)
131 {
132 	const struct tcf_proto_ops *ops;
133 
134 	ops = __tcf_proto_lookup_ops(kind);
135 	if (ops)
136 		return ops;
137 #ifdef CONFIG_MODULES
138 	if (rtnl_held)
139 		rtnl_unlock();
140 	request_module("cls_%s", kind);
141 	if (rtnl_held)
142 		rtnl_lock();
143 	ops = __tcf_proto_lookup_ops(kind);
144 	/* We dropped the RTNL semaphore in order to perform
145 	 * the module load. So, even if we succeeded in loading
146 	 * the module we have to replay the request. We indicate
147 	 * this using -EAGAIN.
148 	 */
149 	if (ops) {
150 		module_put(ops->owner);
151 		return ERR_PTR(-EAGAIN);
152 	}
153 #endif
154 	NL_SET_ERR_MSG(extack, "TC classifier not found");
155 	return ERR_PTR(-ENOENT);
156 }
157 
158 /* Register(unregister) new classifier type */
159 
160 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
161 {
162 	struct tcf_proto_ops *t;
163 	int rc = -EEXIST;
164 
165 	write_lock(&cls_mod_lock);
166 	list_for_each_entry(t, &tcf_proto_base, head)
167 		if (!strcmp(ops->kind, t->kind))
168 			goto out;
169 
170 	list_add_tail(&ops->head, &tcf_proto_base);
171 	rc = 0;
172 out:
173 	write_unlock(&cls_mod_lock);
174 	return rc;
175 }
176 EXPORT_SYMBOL(register_tcf_proto_ops);
177 
178 static struct workqueue_struct *tc_filter_wq;
179 
180 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
181 {
182 	struct tcf_proto_ops *t;
183 	int rc = -ENOENT;
184 
185 	/* Wait for outstanding call_rcu()s, if any, from a
186 	 * tcf_proto_ops's destroy() handler.
187 	 */
188 	rcu_barrier();
189 	flush_workqueue(tc_filter_wq);
190 
191 	write_lock(&cls_mod_lock);
192 	list_for_each_entry(t, &tcf_proto_base, head) {
193 		if (t == ops) {
194 			list_del(&t->head);
195 			rc = 0;
196 			break;
197 		}
198 	}
199 	write_unlock(&cls_mod_lock);
200 	return rc;
201 }
202 EXPORT_SYMBOL(unregister_tcf_proto_ops);
203 
204 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
205 {
206 	INIT_RCU_WORK(rwork, func);
207 	return queue_rcu_work(tc_filter_wq, rwork);
208 }
209 EXPORT_SYMBOL(tcf_queue_work);
210 
211 /* Select new prio value from the range, managed by kernel. */
212 
213 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
214 {
215 	u32 first = TC_H_MAKE(0xC0000000U, 0U);
216 
217 	if (tp)
218 		first = tp->prio - 1;
219 
220 	return TC_H_MAJ(first);
221 }
222 
223 static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
224 {
225 	if (kind)
226 		return nla_strlcpy(name, kind, IFNAMSIZ) >= IFNAMSIZ;
227 	memset(name, 0, IFNAMSIZ);
228 	return false;
229 }
230 
231 static bool tcf_proto_is_unlocked(const char *kind)
232 {
233 	const struct tcf_proto_ops *ops;
234 	bool ret;
235 
236 	if (strlen(kind) == 0)
237 		return false;
238 
239 	ops = tcf_proto_lookup_ops(kind, false, NULL);
240 	/* On error return false to take rtnl lock. Proto lookup/create
241 	 * functions will perform lookup again and properly handle errors.
242 	 */
243 	if (IS_ERR(ops))
244 		return false;
245 
246 	ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
247 	module_put(ops->owner);
248 	return ret;
249 }
250 
251 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
252 					  u32 prio, struct tcf_chain *chain,
253 					  bool rtnl_held,
254 					  struct netlink_ext_ack *extack)
255 {
256 	struct tcf_proto *tp;
257 	int err;
258 
259 	tp = kzalloc(sizeof(*tp), GFP_KERNEL);
260 	if (!tp)
261 		return ERR_PTR(-ENOBUFS);
262 
263 	tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
264 	if (IS_ERR(tp->ops)) {
265 		err = PTR_ERR(tp->ops);
266 		goto errout;
267 	}
268 	tp->classify = tp->ops->classify;
269 	tp->protocol = protocol;
270 	tp->prio = prio;
271 	tp->chain = chain;
272 	spin_lock_init(&tp->lock);
273 	refcount_set(&tp->refcnt, 1);
274 
275 	err = tp->ops->init(tp);
276 	if (err) {
277 		module_put(tp->ops->owner);
278 		goto errout;
279 	}
280 	return tp;
281 
282 errout:
283 	kfree(tp);
284 	return ERR_PTR(err);
285 }
286 
287 static void tcf_proto_get(struct tcf_proto *tp)
288 {
289 	refcount_inc(&tp->refcnt);
290 }
291 
292 static void tcf_chain_put(struct tcf_chain *chain);
293 
294 static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
295 			      bool sig_destroy, struct netlink_ext_ack *extack)
296 {
297 	tp->ops->destroy(tp, rtnl_held, extack);
298 	if (sig_destroy)
299 		tcf_proto_signal_destroyed(tp->chain, tp);
300 	tcf_chain_put(tp->chain);
301 	module_put(tp->ops->owner);
302 	kfree_rcu(tp, rcu);
303 }
304 
305 static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
306 			  struct netlink_ext_ack *extack)
307 {
308 	if (refcount_dec_and_test(&tp->refcnt))
309 		tcf_proto_destroy(tp, rtnl_held, true, extack);
310 }
311 
312 static bool tcf_proto_check_delete(struct tcf_proto *tp)
313 {
314 	if (tp->ops->delete_empty)
315 		return tp->ops->delete_empty(tp);
316 
317 	tp->deleting = true;
318 	return tp->deleting;
319 }
320 
321 static void tcf_proto_mark_delete(struct tcf_proto *tp)
322 {
323 	spin_lock(&tp->lock);
324 	tp->deleting = true;
325 	spin_unlock(&tp->lock);
326 }
327 
328 static bool tcf_proto_is_deleting(struct tcf_proto *tp)
329 {
330 	bool deleting;
331 
332 	spin_lock(&tp->lock);
333 	deleting = tp->deleting;
334 	spin_unlock(&tp->lock);
335 
336 	return deleting;
337 }
338 
339 #define ASSERT_BLOCK_LOCKED(block)					\
340 	lockdep_assert_held(&(block)->lock)
341 
342 struct tcf_filter_chain_list_item {
343 	struct list_head list;
344 	tcf_chain_head_change_t *chain_head_change;
345 	void *chain_head_change_priv;
346 };
347 
348 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
349 					  u32 chain_index)
350 {
351 	struct tcf_chain *chain;
352 
353 	ASSERT_BLOCK_LOCKED(block);
354 
355 	chain = kzalloc(sizeof(*chain), GFP_KERNEL);
356 	if (!chain)
357 		return NULL;
358 	list_add_tail_rcu(&chain->list, &block->chain_list);
359 	mutex_init(&chain->filter_chain_lock);
360 	chain->block = block;
361 	chain->index = chain_index;
362 	chain->refcnt = 1;
363 	if (!chain->index)
364 		block->chain0.chain = chain;
365 	return chain;
366 }
367 
368 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
369 				       struct tcf_proto *tp_head)
370 {
371 	if (item->chain_head_change)
372 		item->chain_head_change(tp_head, item->chain_head_change_priv);
373 }
374 
375 static void tcf_chain0_head_change(struct tcf_chain *chain,
376 				   struct tcf_proto *tp_head)
377 {
378 	struct tcf_filter_chain_list_item *item;
379 	struct tcf_block *block = chain->block;
380 
381 	if (chain->index)
382 		return;
383 
384 	mutex_lock(&block->lock);
385 	list_for_each_entry(item, &block->chain0.filter_chain_list, list)
386 		tcf_chain_head_change_item(item, tp_head);
387 	mutex_unlock(&block->lock);
388 }
389 
390 /* Returns true if block can be safely freed. */
391 
392 static bool tcf_chain_detach(struct tcf_chain *chain)
393 {
394 	struct tcf_block *block = chain->block;
395 
396 	ASSERT_BLOCK_LOCKED(block);
397 
398 	list_del_rcu(&chain->list);
399 	if (!chain->index)
400 		block->chain0.chain = NULL;
401 
402 	if (list_empty(&block->chain_list) &&
403 	    refcount_read(&block->refcnt) == 0)
404 		return true;
405 
406 	return false;
407 }
408 
409 static void tcf_block_destroy(struct tcf_block *block)
410 {
411 	mutex_destroy(&block->lock);
412 	mutex_destroy(&block->proto_destroy_lock);
413 	kfree_rcu(block, rcu);
414 }
415 
416 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
417 {
418 	struct tcf_block *block = chain->block;
419 
420 	mutex_destroy(&chain->filter_chain_lock);
421 	kfree_rcu(chain, rcu);
422 	if (free_block)
423 		tcf_block_destroy(block);
424 }
425 
426 static void tcf_chain_hold(struct tcf_chain *chain)
427 {
428 	ASSERT_BLOCK_LOCKED(chain->block);
429 
430 	++chain->refcnt;
431 }
432 
433 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
434 {
435 	ASSERT_BLOCK_LOCKED(chain->block);
436 
437 	/* In case all the references are action references, this
438 	 * chain should not be shown to the user.
439 	 */
440 	return chain->refcnt == chain->action_refcnt;
441 }
442 
443 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
444 					  u32 chain_index)
445 {
446 	struct tcf_chain *chain;
447 
448 	ASSERT_BLOCK_LOCKED(block);
449 
450 	list_for_each_entry(chain, &block->chain_list, list) {
451 		if (chain->index == chain_index)
452 			return chain;
453 	}
454 	return NULL;
455 }
456 
457 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
458 static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
459 					      u32 chain_index)
460 {
461 	struct tcf_chain *chain;
462 
463 	list_for_each_entry_rcu(chain, &block->chain_list, list) {
464 		if (chain->index == chain_index)
465 			return chain;
466 	}
467 	return NULL;
468 }
469 #endif
470 
471 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
472 			   u32 seq, u16 flags, int event, bool unicast);
473 
474 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
475 					 u32 chain_index, bool create,
476 					 bool by_act)
477 {
478 	struct tcf_chain *chain = NULL;
479 	bool is_first_reference;
480 
481 	mutex_lock(&block->lock);
482 	chain = tcf_chain_lookup(block, chain_index);
483 	if (chain) {
484 		tcf_chain_hold(chain);
485 	} else {
486 		if (!create)
487 			goto errout;
488 		chain = tcf_chain_create(block, chain_index);
489 		if (!chain)
490 			goto errout;
491 	}
492 
493 	if (by_act)
494 		++chain->action_refcnt;
495 	is_first_reference = chain->refcnt - chain->action_refcnt == 1;
496 	mutex_unlock(&block->lock);
497 
498 	/* Send notification only in case we got the first
499 	 * non-action reference. Until then, the chain acts only as
500 	 * a placeholder for actions pointing to it and user ought
501 	 * not know about them.
502 	 */
503 	if (is_first_reference && !by_act)
504 		tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
505 				RTM_NEWCHAIN, false);
506 
507 	return chain;
508 
509 errout:
510 	mutex_unlock(&block->lock);
511 	return chain;
512 }
513 
514 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
515 				       bool create)
516 {
517 	return __tcf_chain_get(block, chain_index, create, false);
518 }
519 
520 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
521 {
522 	return __tcf_chain_get(block, chain_index, true, true);
523 }
524 EXPORT_SYMBOL(tcf_chain_get_by_act);
525 
526 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
527 			       void *tmplt_priv);
528 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
529 				  void *tmplt_priv, u32 chain_index,
530 				  struct tcf_block *block, struct sk_buff *oskb,
531 				  u32 seq, u16 flags, bool unicast);
532 
533 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
534 			    bool explicitly_created)
535 {
536 	struct tcf_block *block = chain->block;
537 	const struct tcf_proto_ops *tmplt_ops;
538 	bool free_block = false;
539 	unsigned int refcnt;
540 	void *tmplt_priv;
541 
542 	mutex_lock(&block->lock);
543 	if (explicitly_created) {
544 		if (!chain->explicitly_created) {
545 			mutex_unlock(&block->lock);
546 			return;
547 		}
548 		chain->explicitly_created = false;
549 	}
550 
551 	if (by_act)
552 		chain->action_refcnt--;
553 
554 	/* tc_chain_notify_delete can't be called while holding block lock.
555 	 * However, when block is unlocked chain can be changed concurrently, so
556 	 * save these to temporary variables.
557 	 */
558 	refcnt = --chain->refcnt;
559 	tmplt_ops = chain->tmplt_ops;
560 	tmplt_priv = chain->tmplt_priv;
561 
562 	/* The last dropped non-action reference will trigger notification. */
563 	if (refcnt - chain->action_refcnt == 0 && !by_act) {
564 		tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
565 				       block, NULL, 0, 0, false);
566 		/* Last reference to chain, no need to lock. */
567 		chain->flushing = false;
568 	}
569 
570 	if (refcnt == 0)
571 		free_block = tcf_chain_detach(chain);
572 	mutex_unlock(&block->lock);
573 
574 	if (refcnt == 0) {
575 		tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
576 		tcf_chain_destroy(chain, free_block);
577 	}
578 }
579 
580 static void tcf_chain_put(struct tcf_chain *chain)
581 {
582 	__tcf_chain_put(chain, false, false);
583 }
584 
585 void tcf_chain_put_by_act(struct tcf_chain *chain)
586 {
587 	__tcf_chain_put(chain, true, false);
588 }
589 EXPORT_SYMBOL(tcf_chain_put_by_act);
590 
591 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
592 {
593 	__tcf_chain_put(chain, false, true);
594 }
595 
596 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
597 {
598 	struct tcf_proto *tp, *tp_next;
599 
600 	mutex_lock(&chain->filter_chain_lock);
601 	tp = tcf_chain_dereference(chain->filter_chain, chain);
602 	while (tp) {
603 		tp_next = rcu_dereference_protected(tp->next, 1);
604 		tcf_proto_signal_destroying(chain, tp);
605 		tp = tp_next;
606 	}
607 	tp = tcf_chain_dereference(chain->filter_chain, chain);
608 	RCU_INIT_POINTER(chain->filter_chain, NULL);
609 	tcf_chain0_head_change(chain, NULL);
610 	chain->flushing = true;
611 	mutex_unlock(&chain->filter_chain_lock);
612 
613 	while (tp) {
614 		tp_next = rcu_dereference_protected(tp->next, 1);
615 		tcf_proto_put(tp, rtnl_held, NULL);
616 		tp = tp_next;
617 	}
618 }
619 
620 static int tcf_block_setup(struct tcf_block *block,
621 			   struct flow_block_offload *bo);
622 
623 static void tc_indr_block_cmd(struct net_device *dev, struct tcf_block *block,
624 			      flow_indr_block_bind_cb_t *cb, void *cb_priv,
625 			      enum flow_block_command command, bool ingress)
626 {
627 	struct flow_block_offload bo = {
628 		.command	= command,
629 		.binder_type	= ingress ?
630 				  FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS :
631 				  FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS,
632 		.net		= dev_net(dev),
633 		.block_shared	= tcf_block_non_null_shared(block),
634 	};
635 	INIT_LIST_HEAD(&bo.cb_list);
636 
637 	if (!block)
638 		return;
639 
640 	bo.block = &block->flow_block;
641 
642 	down_write(&block->cb_lock);
643 	cb(dev, cb_priv, TC_SETUP_BLOCK, &bo);
644 
645 	tcf_block_setup(block, &bo);
646 	up_write(&block->cb_lock);
647 }
648 
649 static struct tcf_block *tc_dev_block(struct net_device *dev, bool ingress)
650 {
651 	const struct Qdisc_class_ops *cops;
652 	const struct Qdisc_ops *ops;
653 	struct Qdisc *qdisc;
654 
655 	if (!dev_ingress_queue(dev))
656 		return NULL;
657 
658 	qdisc = dev_ingress_queue(dev)->qdisc_sleeping;
659 	if (!qdisc)
660 		return NULL;
661 
662 	ops = qdisc->ops;
663 	if (!ops)
664 		return NULL;
665 
666 	if (!ingress && !strcmp("ingress", ops->id))
667 		return NULL;
668 
669 	cops = ops->cl_ops;
670 	if (!cops)
671 		return NULL;
672 
673 	if (!cops->tcf_block)
674 		return NULL;
675 
676 	return cops->tcf_block(qdisc,
677 			       ingress ? TC_H_MIN_INGRESS : TC_H_MIN_EGRESS,
678 			       NULL);
679 }
680 
681 static void tc_indr_block_get_and_cmd(struct net_device *dev,
682 				      flow_indr_block_bind_cb_t *cb,
683 				      void *cb_priv,
684 				      enum flow_block_command command)
685 {
686 	struct tcf_block *block;
687 
688 	block = tc_dev_block(dev, true);
689 	tc_indr_block_cmd(dev, block, cb, cb_priv, command, true);
690 
691 	block = tc_dev_block(dev, false);
692 	tc_indr_block_cmd(dev, block, cb, cb_priv, command, false);
693 }
694 
695 static void tc_indr_block_call(struct tcf_block *block,
696 			       struct net_device *dev,
697 			       struct tcf_block_ext_info *ei,
698 			       enum flow_block_command command,
699 			       struct netlink_ext_ack *extack)
700 {
701 	struct flow_block_offload bo = {
702 		.command	= command,
703 		.binder_type	= ei->binder_type,
704 		.net		= dev_net(dev),
705 		.block		= &block->flow_block,
706 		.block_shared	= tcf_block_shared(block),
707 		.extack		= extack,
708 	};
709 	INIT_LIST_HEAD(&bo.cb_list);
710 
711 	flow_indr_block_call(dev, &bo, command, TC_SETUP_BLOCK);
712 	tcf_block_setup(block, &bo);
713 }
714 
715 static bool tcf_block_offload_in_use(struct tcf_block *block)
716 {
717 	return atomic_read(&block->offloadcnt);
718 }
719 
720 static int tcf_block_offload_cmd(struct tcf_block *block,
721 				 struct net_device *dev,
722 				 struct tcf_block_ext_info *ei,
723 				 enum flow_block_command command,
724 				 struct netlink_ext_ack *extack)
725 {
726 	struct flow_block_offload bo = {};
727 	int err;
728 
729 	bo.net = dev_net(dev);
730 	bo.command = command;
731 	bo.binder_type = ei->binder_type;
732 	bo.block = &block->flow_block;
733 	bo.block_shared = tcf_block_shared(block);
734 	bo.extack = extack;
735 	INIT_LIST_HEAD(&bo.cb_list);
736 
737 	err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
738 	if (err < 0)
739 		return err;
740 
741 	return tcf_block_setup(block, &bo);
742 }
743 
744 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
745 				  struct tcf_block_ext_info *ei,
746 				  struct netlink_ext_ack *extack)
747 {
748 	struct net_device *dev = q->dev_queue->dev;
749 	int err;
750 
751 	down_write(&block->cb_lock);
752 	if (!dev->netdev_ops->ndo_setup_tc)
753 		goto no_offload_dev_inc;
754 
755 	/* If tc offload feature is disabled and the block we try to bind
756 	 * to already has some offloaded filters, forbid to bind.
757 	 */
758 	if (!tc_can_offload(dev) && tcf_block_offload_in_use(block)) {
759 		NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
760 		err = -EOPNOTSUPP;
761 		goto err_unlock;
762 	}
763 
764 	err = tcf_block_offload_cmd(block, dev, ei, FLOW_BLOCK_BIND, extack);
765 	if (err == -EOPNOTSUPP)
766 		goto no_offload_dev_inc;
767 	if (err)
768 		goto err_unlock;
769 
770 	tc_indr_block_call(block, dev, ei, FLOW_BLOCK_BIND, extack);
771 	up_write(&block->cb_lock);
772 	return 0;
773 
774 no_offload_dev_inc:
775 	if (tcf_block_offload_in_use(block)) {
776 		err = -EOPNOTSUPP;
777 		goto err_unlock;
778 	}
779 	err = 0;
780 	block->nooffloaddevcnt++;
781 	tc_indr_block_call(block, dev, ei, FLOW_BLOCK_BIND, extack);
782 err_unlock:
783 	up_write(&block->cb_lock);
784 	return err;
785 }
786 
787 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
788 				     struct tcf_block_ext_info *ei)
789 {
790 	struct net_device *dev = q->dev_queue->dev;
791 	int err;
792 
793 	down_write(&block->cb_lock);
794 	tc_indr_block_call(block, dev, ei, FLOW_BLOCK_UNBIND, NULL);
795 
796 	if (!dev->netdev_ops->ndo_setup_tc)
797 		goto no_offload_dev_dec;
798 	err = tcf_block_offload_cmd(block, dev, ei, FLOW_BLOCK_UNBIND, NULL);
799 	if (err == -EOPNOTSUPP)
800 		goto no_offload_dev_dec;
801 	up_write(&block->cb_lock);
802 	return;
803 
804 no_offload_dev_dec:
805 	WARN_ON(block->nooffloaddevcnt-- == 0);
806 	up_write(&block->cb_lock);
807 }
808 
809 static int
810 tcf_chain0_head_change_cb_add(struct tcf_block *block,
811 			      struct tcf_block_ext_info *ei,
812 			      struct netlink_ext_ack *extack)
813 {
814 	struct tcf_filter_chain_list_item *item;
815 	struct tcf_chain *chain0;
816 
817 	item = kmalloc(sizeof(*item), GFP_KERNEL);
818 	if (!item) {
819 		NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
820 		return -ENOMEM;
821 	}
822 	item->chain_head_change = ei->chain_head_change;
823 	item->chain_head_change_priv = ei->chain_head_change_priv;
824 
825 	mutex_lock(&block->lock);
826 	chain0 = block->chain0.chain;
827 	if (chain0)
828 		tcf_chain_hold(chain0);
829 	else
830 		list_add(&item->list, &block->chain0.filter_chain_list);
831 	mutex_unlock(&block->lock);
832 
833 	if (chain0) {
834 		struct tcf_proto *tp_head;
835 
836 		mutex_lock(&chain0->filter_chain_lock);
837 
838 		tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
839 		if (tp_head)
840 			tcf_chain_head_change_item(item, tp_head);
841 
842 		mutex_lock(&block->lock);
843 		list_add(&item->list, &block->chain0.filter_chain_list);
844 		mutex_unlock(&block->lock);
845 
846 		mutex_unlock(&chain0->filter_chain_lock);
847 		tcf_chain_put(chain0);
848 	}
849 
850 	return 0;
851 }
852 
853 static void
854 tcf_chain0_head_change_cb_del(struct tcf_block *block,
855 			      struct tcf_block_ext_info *ei)
856 {
857 	struct tcf_filter_chain_list_item *item;
858 
859 	mutex_lock(&block->lock);
860 	list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
861 		if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
862 		    (item->chain_head_change == ei->chain_head_change &&
863 		     item->chain_head_change_priv == ei->chain_head_change_priv)) {
864 			if (block->chain0.chain)
865 				tcf_chain_head_change_item(item, NULL);
866 			list_del(&item->list);
867 			mutex_unlock(&block->lock);
868 
869 			kfree(item);
870 			return;
871 		}
872 	}
873 	mutex_unlock(&block->lock);
874 	WARN_ON(1);
875 }
876 
877 struct tcf_net {
878 	spinlock_t idr_lock; /* Protects idr */
879 	struct idr idr;
880 };
881 
882 static unsigned int tcf_net_id;
883 
884 static int tcf_block_insert(struct tcf_block *block, struct net *net,
885 			    struct netlink_ext_ack *extack)
886 {
887 	struct tcf_net *tn = net_generic(net, tcf_net_id);
888 	int err;
889 
890 	idr_preload(GFP_KERNEL);
891 	spin_lock(&tn->idr_lock);
892 	err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
893 			    GFP_NOWAIT);
894 	spin_unlock(&tn->idr_lock);
895 	idr_preload_end();
896 
897 	return err;
898 }
899 
900 static void tcf_block_remove(struct tcf_block *block, struct net *net)
901 {
902 	struct tcf_net *tn = net_generic(net, tcf_net_id);
903 
904 	spin_lock(&tn->idr_lock);
905 	idr_remove(&tn->idr, block->index);
906 	spin_unlock(&tn->idr_lock);
907 }
908 
909 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
910 					  u32 block_index,
911 					  struct netlink_ext_ack *extack)
912 {
913 	struct tcf_block *block;
914 
915 	block = kzalloc(sizeof(*block), GFP_KERNEL);
916 	if (!block) {
917 		NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
918 		return ERR_PTR(-ENOMEM);
919 	}
920 	mutex_init(&block->lock);
921 	mutex_init(&block->proto_destroy_lock);
922 	init_rwsem(&block->cb_lock);
923 	flow_block_init(&block->flow_block);
924 	INIT_LIST_HEAD(&block->chain_list);
925 	INIT_LIST_HEAD(&block->owner_list);
926 	INIT_LIST_HEAD(&block->chain0.filter_chain_list);
927 
928 	refcount_set(&block->refcnt, 1);
929 	block->net = net;
930 	block->index = block_index;
931 
932 	/* Don't store q pointer for blocks which are shared */
933 	if (!tcf_block_shared(block))
934 		block->q = q;
935 	return block;
936 }
937 
938 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
939 {
940 	struct tcf_net *tn = net_generic(net, tcf_net_id);
941 
942 	return idr_find(&tn->idr, block_index);
943 }
944 
945 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
946 {
947 	struct tcf_block *block;
948 
949 	rcu_read_lock();
950 	block = tcf_block_lookup(net, block_index);
951 	if (block && !refcount_inc_not_zero(&block->refcnt))
952 		block = NULL;
953 	rcu_read_unlock();
954 
955 	return block;
956 }
957 
958 static struct tcf_chain *
959 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
960 {
961 	mutex_lock(&block->lock);
962 	if (chain)
963 		chain = list_is_last(&chain->list, &block->chain_list) ?
964 			NULL : list_next_entry(chain, list);
965 	else
966 		chain = list_first_entry_or_null(&block->chain_list,
967 						 struct tcf_chain, list);
968 
969 	/* skip all action-only chains */
970 	while (chain && tcf_chain_held_by_acts_only(chain))
971 		chain = list_is_last(&chain->list, &block->chain_list) ?
972 			NULL : list_next_entry(chain, list);
973 
974 	if (chain)
975 		tcf_chain_hold(chain);
976 	mutex_unlock(&block->lock);
977 
978 	return chain;
979 }
980 
981 /* Function to be used by all clients that want to iterate over all chains on
982  * block. It properly obtains block->lock and takes reference to chain before
983  * returning it. Users of this function must be tolerant to concurrent chain
984  * insertion/deletion or ensure that no concurrent chain modification is
985  * possible. Note that all netlink dump callbacks cannot guarantee to provide
986  * consistent dump because rtnl lock is released each time skb is filled with
987  * data and sent to user-space.
988  */
989 
990 struct tcf_chain *
991 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
992 {
993 	struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
994 
995 	if (chain)
996 		tcf_chain_put(chain);
997 
998 	return chain_next;
999 }
1000 EXPORT_SYMBOL(tcf_get_next_chain);
1001 
1002 static struct tcf_proto *
1003 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1004 {
1005 	u32 prio = 0;
1006 
1007 	ASSERT_RTNL();
1008 	mutex_lock(&chain->filter_chain_lock);
1009 
1010 	if (!tp) {
1011 		tp = tcf_chain_dereference(chain->filter_chain, chain);
1012 	} else if (tcf_proto_is_deleting(tp)) {
1013 		/* 'deleting' flag is set and chain->filter_chain_lock was
1014 		 * unlocked, which means next pointer could be invalid. Restart
1015 		 * search.
1016 		 */
1017 		prio = tp->prio + 1;
1018 		tp = tcf_chain_dereference(chain->filter_chain, chain);
1019 
1020 		for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1021 			if (!tp->deleting && tp->prio >= prio)
1022 				break;
1023 	} else {
1024 		tp = tcf_chain_dereference(tp->next, chain);
1025 	}
1026 
1027 	if (tp)
1028 		tcf_proto_get(tp);
1029 
1030 	mutex_unlock(&chain->filter_chain_lock);
1031 
1032 	return tp;
1033 }
1034 
1035 /* Function to be used by all clients that want to iterate over all tp's on
1036  * chain. Users of this function must be tolerant to concurrent tp
1037  * insertion/deletion or ensure that no concurrent chain modification is
1038  * possible. Note that all netlink dump callbacks cannot guarantee to provide
1039  * consistent dump because rtnl lock is released each time skb is filled with
1040  * data and sent to user-space.
1041  */
1042 
1043 struct tcf_proto *
1044 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp,
1045 		   bool rtnl_held)
1046 {
1047 	struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1048 
1049 	if (tp)
1050 		tcf_proto_put(tp, rtnl_held, NULL);
1051 
1052 	return tp_next;
1053 }
1054 EXPORT_SYMBOL(tcf_get_next_proto);
1055 
1056 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1057 {
1058 	struct tcf_chain *chain;
1059 
1060 	/* Last reference to block. At this point chains cannot be added or
1061 	 * removed concurrently.
1062 	 */
1063 	for (chain = tcf_get_next_chain(block, NULL);
1064 	     chain;
1065 	     chain = tcf_get_next_chain(block, chain)) {
1066 		tcf_chain_put_explicitly_created(chain);
1067 		tcf_chain_flush(chain, rtnl_held);
1068 	}
1069 }
1070 
1071 /* Lookup Qdisc and increments its reference counter.
1072  * Set parent, if necessary.
1073  */
1074 
1075 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1076 			    u32 *parent, int ifindex, bool rtnl_held,
1077 			    struct netlink_ext_ack *extack)
1078 {
1079 	const struct Qdisc_class_ops *cops;
1080 	struct net_device *dev;
1081 	int err = 0;
1082 
1083 	if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1084 		return 0;
1085 
1086 	rcu_read_lock();
1087 
1088 	/* Find link */
1089 	dev = dev_get_by_index_rcu(net, ifindex);
1090 	if (!dev) {
1091 		rcu_read_unlock();
1092 		return -ENODEV;
1093 	}
1094 
1095 	/* Find qdisc */
1096 	if (!*parent) {
1097 		*q = dev->qdisc;
1098 		*parent = (*q)->handle;
1099 	} else {
1100 		*q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1101 		if (!*q) {
1102 			NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1103 			err = -EINVAL;
1104 			goto errout_rcu;
1105 		}
1106 	}
1107 
1108 	*q = qdisc_refcount_inc_nz(*q);
1109 	if (!*q) {
1110 		NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1111 		err = -EINVAL;
1112 		goto errout_rcu;
1113 	}
1114 
1115 	/* Is it classful? */
1116 	cops = (*q)->ops->cl_ops;
1117 	if (!cops) {
1118 		NL_SET_ERR_MSG(extack, "Qdisc not classful");
1119 		err = -EINVAL;
1120 		goto errout_qdisc;
1121 	}
1122 
1123 	if (!cops->tcf_block) {
1124 		NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1125 		err = -EOPNOTSUPP;
1126 		goto errout_qdisc;
1127 	}
1128 
1129 errout_rcu:
1130 	/* At this point we know that qdisc is not noop_qdisc,
1131 	 * which means that qdisc holds a reference to net_device
1132 	 * and we hold a reference to qdisc, so it is safe to release
1133 	 * rcu read lock.
1134 	 */
1135 	rcu_read_unlock();
1136 	return err;
1137 
1138 errout_qdisc:
1139 	rcu_read_unlock();
1140 
1141 	if (rtnl_held)
1142 		qdisc_put(*q);
1143 	else
1144 		qdisc_put_unlocked(*q);
1145 	*q = NULL;
1146 
1147 	return err;
1148 }
1149 
1150 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1151 			       int ifindex, struct netlink_ext_ack *extack)
1152 {
1153 	if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1154 		return 0;
1155 
1156 	/* Do we search for filter, attached to class? */
1157 	if (TC_H_MIN(parent)) {
1158 		const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1159 
1160 		*cl = cops->find(q, parent);
1161 		if (*cl == 0) {
1162 			NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1163 			return -ENOENT;
1164 		}
1165 	}
1166 
1167 	return 0;
1168 }
1169 
1170 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1171 					  unsigned long cl, int ifindex,
1172 					  u32 block_index,
1173 					  struct netlink_ext_ack *extack)
1174 {
1175 	struct tcf_block *block;
1176 
1177 	if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1178 		block = tcf_block_refcnt_get(net, block_index);
1179 		if (!block) {
1180 			NL_SET_ERR_MSG(extack, "Block of given index was not found");
1181 			return ERR_PTR(-EINVAL);
1182 		}
1183 	} else {
1184 		const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1185 
1186 		block = cops->tcf_block(q, cl, extack);
1187 		if (!block)
1188 			return ERR_PTR(-EINVAL);
1189 
1190 		if (tcf_block_shared(block)) {
1191 			NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1192 			return ERR_PTR(-EOPNOTSUPP);
1193 		}
1194 
1195 		/* Always take reference to block in order to support execution
1196 		 * of rules update path of cls API without rtnl lock. Caller
1197 		 * must release block when it is finished using it. 'if' block
1198 		 * of this conditional obtain reference to block by calling
1199 		 * tcf_block_refcnt_get().
1200 		 */
1201 		refcount_inc(&block->refcnt);
1202 	}
1203 
1204 	return block;
1205 }
1206 
1207 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1208 			    struct tcf_block_ext_info *ei, bool rtnl_held)
1209 {
1210 	if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1211 		/* Flushing/putting all chains will cause the block to be
1212 		 * deallocated when last chain is freed. However, if chain_list
1213 		 * is empty, block has to be manually deallocated. After block
1214 		 * reference counter reached 0, it is no longer possible to
1215 		 * increment it or add new chains to block.
1216 		 */
1217 		bool free_block = list_empty(&block->chain_list);
1218 
1219 		mutex_unlock(&block->lock);
1220 		if (tcf_block_shared(block))
1221 			tcf_block_remove(block, block->net);
1222 
1223 		if (q)
1224 			tcf_block_offload_unbind(block, q, ei);
1225 
1226 		if (free_block)
1227 			tcf_block_destroy(block);
1228 		else
1229 			tcf_block_flush_all_chains(block, rtnl_held);
1230 	} else if (q) {
1231 		tcf_block_offload_unbind(block, q, ei);
1232 	}
1233 }
1234 
1235 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1236 {
1237 	__tcf_block_put(block, NULL, NULL, rtnl_held);
1238 }
1239 
1240 /* Find tcf block.
1241  * Set q, parent, cl when appropriate.
1242  */
1243 
1244 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1245 					u32 *parent, unsigned long *cl,
1246 					int ifindex, u32 block_index,
1247 					struct netlink_ext_ack *extack)
1248 {
1249 	struct tcf_block *block;
1250 	int err = 0;
1251 
1252 	ASSERT_RTNL();
1253 
1254 	err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1255 	if (err)
1256 		goto errout;
1257 
1258 	err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1259 	if (err)
1260 		goto errout_qdisc;
1261 
1262 	block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1263 	if (IS_ERR(block)) {
1264 		err = PTR_ERR(block);
1265 		goto errout_qdisc;
1266 	}
1267 
1268 	return block;
1269 
1270 errout_qdisc:
1271 	if (*q)
1272 		qdisc_put(*q);
1273 errout:
1274 	*q = NULL;
1275 	return ERR_PTR(err);
1276 }
1277 
1278 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1279 			      bool rtnl_held)
1280 {
1281 	if (!IS_ERR_OR_NULL(block))
1282 		tcf_block_refcnt_put(block, rtnl_held);
1283 
1284 	if (q) {
1285 		if (rtnl_held)
1286 			qdisc_put(q);
1287 		else
1288 			qdisc_put_unlocked(q);
1289 	}
1290 }
1291 
1292 struct tcf_block_owner_item {
1293 	struct list_head list;
1294 	struct Qdisc *q;
1295 	enum flow_block_binder_type binder_type;
1296 };
1297 
1298 static void
1299 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1300 			       struct Qdisc *q,
1301 			       enum flow_block_binder_type binder_type)
1302 {
1303 	if (block->keep_dst &&
1304 	    binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1305 	    binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1306 		netif_keep_dst(qdisc_dev(q));
1307 }
1308 
1309 void tcf_block_netif_keep_dst(struct tcf_block *block)
1310 {
1311 	struct tcf_block_owner_item *item;
1312 
1313 	block->keep_dst = true;
1314 	list_for_each_entry(item, &block->owner_list, list)
1315 		tcf_block_owner_netif_keep_dst(block, item->q,
1316 					       item->binder_type);
1317 }
1318 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1319 
1320 static int tcf_block_owner_add(struct tcf_block *block,
1321 			       struct Qdisc *q,
1322 			       enum flow_block_binder_type binder_type)
1323 {
1324 	struct tcf_block_owner_item *item;
1325 
1326 	item = kmalloc(sizeof(*item), GFP_KERNEL);
1327 	if (!item)
1328 		return -ENOMEM;
1329 	item->q = q;
1330 	item->binder_type = binder_type;
1331 	list_add(&item->list, &block->owner_list);
1332 	return 0;
1333 }
1334 
1335 static void tcf_block_owner_del(struct tcf_block *block,
1336 				struct Qdisc *q,
1337 				enum flow_block_binder_type binder_type)
1338 {
1339 	struct tcf_block_owner_item *item;
1340 
1341 	list_for_each_entry(item, &block->owner_list, list) {
1342 		if (item->q == q && item->binder_type == binder_type) {
1343 			list_del(&item->list);
1344 			kfree(item);
1345 			return;
1346 		}
1347 	}
1348 	WARN_ON(1);
1349 }
1350 
1351 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1352 		      struct tcf_block_ext_info *ei,
1353 		      struct netlink_ext_ack *extack)
1354 {
1355 	struct net *net = qdisc_net(q);
1356 	struct tcf_block *block = NULL;
1357 	int err;
1358 
1359 	if (ei->block_index)
1360 		/* block_index not 0 means the shared block is requested */
1361 		block = tcf_block_refcnt_get(net, ei->block_index);
1362 
1363 	if (!block) {
1364 		block = tcf_block_create(net, q, ei->block_index, extack);
1365 		if (IS_ERR(block))
1366 			return PTR_ERR(block);
1367 		if (tcf_block_shared(block)) {
1368 			err = tcf_block_insert(block, net, extack);
1369 			if (err)
1370 				goto err_block_insert;
1371 		}
1372 	}
1373 
1374 	err = tcf_block_owner_add(block, q, ei->binder_type);
1375 	if (err)
1376 		goto err_block_owner_add;
1377 
1378 	tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1379 
1380 	err = tcf_chain0_head_change_cb_add(block, ei, extack);
1381 	if (err)
1382 		goto err_chain0_head_change_cb_add;
1383 
1384 	err = tcf_block_offload_bind(block, q, ei, extack);
1385 	if (err)
1386 		goto err_block_offload_bind;
1387 
1388 	*p_block = block;
1389 	return 0;
1390 
1391 err_block_offload_bind:
1392 	tcf_chain0_head_change_cb_del(block, ei);
1393 err_chain0_head_change_cb_add:
1394 	tcf_block_owner_del(block, q, ei->binder_type);
1395 err_block_owner_add:
1396 err_block_insert:
1397 	tcf_block_refcnt_put(block, true);
1398 	return err;
1399 }
1400 EXPORT_SYMBOL(tcf_block_get_ext);
1401 
1402 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1403 {
1404 	struct tcf_proto __rcu **p_filter_chain = priv;
1405 
1406 	rcu_assign_pointer(*p_filter_chain, tp_head);
1407 }
1408 
1409 int tcf_block_get(struct tcf_block **p_block,
1410 		  struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1411 		  struct netlink_ext_ack *extack)
1412 {
1413 	struct tcf_block_ext_info ei = {
1414 		.chain_head_change = tcf_chain_head_change_dflt,
1415 		.chain_head_change_priv = p_filter_chain,
1416 	};
1417 
1418 	WARN_ON(!p_filter_chain);
1419 	return tcf_block_get_ext(p_block, q, &ei, extack);
1420 }
1421 EXPORT_SYMBOL(tcf_block_get);
1422 
1423 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1424  * actions should be all removed after flushing.
1425  */
1426 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1427 		       struct tcf_block_ext_info *ei)
1428 {
1429 	if (!block)
1430 		return;
1431 	tcf_chain0_head_change_cb_del(block, ei);
1432 	tcf_block_owner_del(block, q, ei->binder_type);
1433 
1434 	__tcf_block_put(block, q, ei, true);
1435 }
1436 EXPORT_SYMBOL(tcf_block_put_ext);
1437 
1438 void tcf_block_put(struct tcf_block *block)
1439 {
1440 	struct tcf_block_ext_info ei = {0, };
1441 
1442 	if (!block)
1443 		return;
1444 	tcf_block_put_ext(block, block->q, &ei);
1445 }
1446 
1447 EXPORT_SYMBOL(tcf_block_put);
1448 
1449 static int
1450 tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1451 			    void *cb_priv, bool add, bool offload_in_use,
1452 			    struct netlink_ext_ack *extack)
1453 {
1454 	struct tcf_chain *chain, *chain_prev;
1455 	struct tcf_proto *tp, *tp_prev;
1456 	int err;
1457 
1458 	lockdep_assert_held(&block->cb_lock);
1459 
1460 	for (chain = __tcf_get_next_chain(block, NULL);
1461 	     chain;
1462 	     chain_prev = chain,
1463 		     chain = __tcf_get_next_chain(block, chain),
1464 		     tcf_chain_put(chain_prev)) {
1465 		for (tp = __tcf_get_next_proto(chain, NULL); tp;
1466 		     tp_prev = tp,
1467 			     tp = __tcf_get_next_proto(chain, tp),
1468 			     tcf_proto_put(tp_prev, true, NULL)) {
1469 			if (tp->ops->reoffload) {
1470 				err = tp->ops->reoffload(tp, add, cb, cb_priv,
1471 							 extack);
1472 				if (err && add)
1473 					goto err_playback_remove;
1474 			} else if (add && offload_in_use) {
1475 				err = -EOPNOTSUPP;
1476 				NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1477 				goto err_playback_remove;
1478 			}
1479 		}
1480 	}
1481 
1482 	return 0;
1483 
1484 err_playback_remove:
1485 	tcf_proto_put(tp, true, NULL);
1486 	tcf_chain_put(chain);
1487 	tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1488 				    extack);
1489 	return err;
1490 }
1491 
1492 static int tcf_block_bind(struct tcf_block *block,
1493 			  struct flow_block_offload *bo)
1494 {
1495 	struct flow_block_cb *block_cb, *next;
1496 	int err, i = 0;
1497 
1498 	lockdep_assert_held(&block->cb_lock);
1499 
1500 	list_for_each_entry(block_cb, &bo->cb_list, list) {
1501 		err = tcf_block_playback_offloads(block, block_cb->cb,
1502 						  block_cb->cb_priv, true,
1503 						  tcf_block_offload_in_use(block),
1504 						  bo->extack);
1505 		if (err)
1506 			goto err_unroll;
1507 		if (!bo->unlocked_driver_cb)
1508 			block->lockeddevcnt++;
1509 
1510 		i++;
1511 	}
1512 	list_splice(&bo->cb_list, &block->flow_block.cb_list);
1513 
1514 	return 0;
1515 
1516 err_unroll:
1517 	list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1518 		if (i-- > 0) {
1519 			list_del(&block_cb->list);
1520 			tcf_block_playback_offloads(block, block_cb->cb,
1521 						    block_cb->cb_priv, false,
1522 						    tcf_block_offload_in_use(block),
1523 						    NULL);
1524 			if (!bo->unlocked_driver_cb)
1525 				block->lockeddevcnt--;
1526 		}
1527 		flow_block_cb_free(block_cb);
1528 	}
1529 
1530 	return err;
1531 }
1532 
1533 static void tcf_block_unbind(struct tcf_block *block,
1534 			     struct flow_block_offload *bo)
1535 {
1536 	struct flow_block_cb *block_cb, *next;
1537 
1538 	lockdep_assert_held(&block->cb_lock);
1539 
1540 	list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1541 		tcf_block_playback_offloads(block, block_cb->cb,
1542 					    block_cb->cb_priv, false,
1543 					    tcf_block_offload_in_use(block),
1544 					    NULL);
1545 		list_del(&block_cb->list);
1546 		flow_block_cb_free(block_cb);
1547 		if (!bo->unlocked_driver_cb)
1548 			block->lockeddevcnt--;
1549 	}
1550 }
1551 
1552 static int tcf_block_setup(struct tcf_block *block,
1553 			   struct flow_block_offload *bo)
1554 {
1555 	int err;
1556 
1557 	switch (bo->command) {
1558 	case FLOW_BLOCK_BIND:
1559 		err = tcf_block_bind(block, bo);
1560 		break;
1561 	case FLOW_BLOCK_UNBIND:
1562 		err = 0;
1563 		tcf_block_unbind(block, bo);
1564 		break;
1565 	default:
1566 		WARN_ON_ONCE(1);
1567 		err = -EOPNOTSUPP;
1568 	}
1569 
1570 	return err;
1571 }
1572 
1573 /* Main classifier routine: scans classifier chain attached
1574  * to this qdisc, (optionally) tests for protocol and asks
1575  * specific classifiers.
1576  */
1577 static inline int __tcf_classify(struct sk_buff *skb,
1578 				 const struct tcf_proto *tp,
1579 				 const struct tcf_proto *orig_tp,
1580 				 struct tcf_result *res,
1581 				 bool compat_mode,
1582 				 u32 *last_executed_chain)
1583 {
1584 #ifdef CONFIG_NET_CLS_ACT
1585 	const int max_reclassify_loop = 4;
1586 	const struct tcf_proto *first_tp;
1587 	int limit = 0;
1588 
1589 reclassify:
1590 #endif
1591 	for (; tp; tp = rcu_dereference_bh(tp->next)) {
1592 		__be16 protocol = tc_skb_protocol(skb);
1593 		int err;
1594 
1595 		if (tp->protocol != protocol &&
1596 		    tp->protocol != htons(ETH_P_ALL))
1597 			continue;
1598 
1599 		err = tp->classify(skb, tp, res);
1600 #ifdef CONFIG_NET_CLS_ACT
1601 		if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1602 			first_tp = orig_tp;
1603 			*last_executed_chain = first_tp->chain->index;
1604 			goto reset;
1605 		} else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1606 			first_tp = res->goto_tp;
1607 			*last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1608 			goto reset;
1609 		}
1610 #endif
1611 		if (err >= 0)
1612 			return err;
1613 	}
1614 
1615 	return TC_ACT_UNSPEC; /* signal: continue lookup */
1616 #ifdef CONFIG_NET_CLS_ACT
1617 reset:
1618 	if (unlikely(limit++ >= max_reclassify_loop)) {
1619 		net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1620 				       tp->chain->block->index,
1621 				       tp->prio & 0xffff,
1622 				       ntohs(tp->protocol));
1623 		return TC_ACT_SHOT;
1624 	}
1625 
1626 	tp = first_tp;
1627 	goto reclassify;
1628 #endif
1629 }
1630 
1631 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1632 		 struct tcf_result *res, bool compat_mode)
1633 {
1634 	u32 last_executed_chain = 0;
1635 
1636 	return __tcf_classify(skb, tp, tp, res, compat_mode,
1637 			      &last_executed_chain);
1638 }
1639 EXPORT_SYMBOL(tcf_classify);
1640 
1641 int tcf_classify_ingress(struct sk_buff *skb,
1642 			 const struct tcf_block *ingress_block,
1643 			 const struct tcf_proto *tp,
1644 			 struct tcf_result *res, bool compat_mode)
1645 {
1646 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1647 	u32 last_executed_chain = 0;
1648 
1649 	return __tcf_classify(skb, tp, tp, res, compat_mode,
1650 			      &last_executed_chain);
1651 #else
1652 	u32 last_executed_chain = tp ? tp->chain->index : 0;
1653 	const struct tcf_proto *orig_tp = tp;
1654 	struct tc_skb_ext *ext;
1655 	int ret;
1656 
1657 	ext = skb_ext_find(skb, TC_SKB_EXT);
1658 
1659 	if (ext && ext->chain) {
1660 		struct tcf_chain *fchain;
1661 
1662 		fchain = tcf_chain_lookup_rcu(ingress_block, ext->chain);
1663 		if (!fchain)
1664 			return TC_ACT_SHOT;
1665 
1666 		/* Consume, so cloned/redirect skbs won't inherit ext */
1667 		skb_ext_del(skb, TC_SKB_EXT);
1668 
1669 		tp = rcu_dereference_bh(fchain->filter_chain);
1670 	}
1671 
1672 	ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode,
1673 			     &last_executed_chain);
1674 
1675 	/* If we missed on some chain */
1676 	if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1677 		ext = skb_ext_add(skb, TC_SKB_EXT);
1678 		if (WARN_ON_ONCE(!ext))
1679 			return TC_ACT_SHOT;
1680 		ext->chain = last_executed_chain;
1681 	}
1682 
1683 	return ret;
1684 #endif
1685 }
1686 EXPORT_SYMBOL(tcf_classify_ingress);
1687 
1688 struct tcf_chain_info {
1689 	struct tcf_proto __rcu **pprev;
1690 	struct tcf_proto __rcu *next;
1691 };
1692 
1693 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1694 					   struct tcf_chain_info *chain_info)
1695 {
1696 	return tcf_chain_dereference(*chain_info->pprev, chain);
1697 }
1698 
1699 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1700 			       struct tcf_chain_info *chain_info,
1701 			       struct tcf_proto *tp)
1702 {
1703 	if (chain->flushing)
1704 		return -EAGAIN;
1705 
1706 	if (*chain_info->pprev == chain->filter_chain)
1707 		tcf_chain0_head_change(chain, tp);
1708 	tcf_proto_get(tp);
1709 	RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1710 	rcu_assign_pointer(*chain_info->pprev, tp);
1711 
1712 	return 0;
1713 }
1714 
1715 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1716 				struct tcf_chain_info *chain_info,
1717 				struct tcf_proto *tp)
1718 {
1719 	struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1720 
1721 	tcf_proto_mark_delete(tp);
1722 	if (tp == chain->filter_chain)
1723 		tcf_chain0_head_change(chain, next);
1724 	RCU_INIT_POINTER(*chain_info->pprev, next);
1725 }
1726 
1727 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1728 					   struct tcf_chain_info *chain_info,
1729 					   u32 protocol, u32 prio,
1730 					   bool prio_allocate);
1731 
1732 /* Try to insert new proto.
1733  * If proto with specified priority already exists, free new proto
1734  * and return existing one.
1735  */
1736 
1737 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1738 						    struct tcf_proto *tp_new,
1739 						    u32 protocol, u32 prio,
1740 						    bool rtnl_held)
1741 {
1742 	struct tcf_chain_info chain_info;
1743 	struct tcf_proto *tp;
1744 	int err = 0;
1745 
1746 	mutex_lock(&chain->filter_chain_lock);
1747 
1748 	if (tcf_proto_exists_destroying(chain, tp_new)) {
1749 		mutex_unlock(&chain->filter_chain_lock);
1750 		tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1751 		return ERR_PTR(-EAGAIN);
1752 	}
1753 
1754 	tp = tcf_chain_tp_find(chain, &chain_info,
1755 			       protocol, prio, false);
1756 	if (!tp)
1757 		err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1758 	mutex_unlock(&chain->filter_chain_lock);
1759 
1760 	if (tp) {
1761 		tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1762 		tp_new = tp;
1763 	} else if (err) {
1764 		tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1765 		tp_new = ERR_PTR(err);
1766 	}
1767 
1768 	return tp_new;
1769 }
1770 
1771 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1772 				      struct tcf_proto *tp, bool rtnl_held,
1773 				      struct netlink_ext_ack *extack)
1774 {
1775 	struct tcf_chain_info chain_info;
1776 	struct tcf_proto *tp_iter;
1777 	struct tcf_proto **pprev;
1778 	struct tcf_proto *next;
1779 
1780 	mutex_lock(&chain->filter_chain_lock);
1781 
1782 	/* Atomically find and remove tp from chain. */
1783 	for (pprev = &chain->filter_chain;
1784 	     (tp_iter = tcf_chain_dereference(*pprev, chain));
1785 	     pprev = &tp_iter->next) {
1786 		if (tp_iter == tp) {
1787 			chain_info.pprev = pprev;
1788 			chain_info.next = tp_iter->next;
1789 			WARN_ON(tp_iter->deleting);
1790 			break;
1791 		}
1792 	}
1793 	/* Verify that tp still exists and no new filters were inserted
1794 	 * concurrently.
1795 	 * Mark tp for deletion if it is empty.
1796 	 */
1797 	if (!tp_iter || !tcf_proto_check_delete(tp)) {
1798 		mutex_unlock(&chain->filter_chain_lock);
1799 		return;
1800 	}
1801 
1802 	tcf_proto_signal_destroying(chain, tp);
1803 	next = tcf_chain_dereference(chain_info.next, chain);
1804 	if (tp == chain->filter_chain)
1805 		tcf_chain0_head_change(chain, next);
1806 	RCU_INIT_POINTER(*chain_info.pprev, next);
1807 	mutex_unlock(&chain->filter_chain_lock);
1808 
1809 	tcf_proto_put(tp, rtnl_held, extack);
1810 }
1811 
1812 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1813 					   struct tcf_chain_info *chain_info,
1814 					   u32 protocol, u32 prio,
1815 					   bool prio_allocate)
1816 {
1817 	struct tcf_proto **pprev;
1818 	struct tcf_proto *tp;
1819 
1820 	/* Check the chain for existence of proto-tcf with this priority */
1821 	for (pprev = &chain->filter_chain;
1822 	     (tp = tcf_chain_dereference(*pprev, chain));
1823 	     pprev = &tp->next) {
1824 		if (tp->prio >= prio) {
1825 			if (tp->prio == prio) {
1826 				if (prio_allocate ||
1827 				    (tp->protocol != protocol && protocol))
1828 					return ERR_PTR(-EINVAL);
1829 			} else {
1830 				tp = NULL;
1831 			}
1832 			break;
1833 		}
1834 	}
1835 	chain_info->pprev = pprev;
1836 	if (tp) {
1837 		chain_info->next = tp->next;
1838 		tcf_proto_get(tp);
1839 	} else {
1840 		chain_info->next = NULL;
1841 	}
1842 	return tp;
1843 }
1844 
1845 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1846 			 struct tcf_proto *tp, struct tcf_block *block,
1847 			 struct Qdisc *q, u32 parent, void *fh,
1848 			 u32 portid, u32 seq, u16 flags, int event,
1849 			 bool rtnl_held)
1850 {
1851 	struct tcmsg *tcm;
1852 	struct nlmsghdr  *nlh;
1853 	unsigned char *b = skb_tail_pointer(skb);
1854 
1855 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1856 	if (!nlh)
1857 		goto out_nlmsg_trim;
1858 	tcm = nlmsg_data(nlh);
1859 	tcm->tcm_family = AF_UNSPEC;
1860 	tcm->tcm__pad1 = 0;
1861 	tcm->tcm__pad2 = 0;
1862 	if (q) {
1863 		tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1864 		tcm->tcm_parent = parent;
1865 	} else {
1866 		tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1867 		tcm->tcm_block_index = block->index;
1868 	}
1869 	tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1870 	if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1871 		goto nla_put_failure;
1872 	if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1873 		goto nla_put_failure;
1874 	if (!fh) {
1875 		tcm->tcm_handle = 0;
1876 	} else {
1877 		if (tp->ops->dump &&
1878 		    tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
1879 			goto nla_put_failure;
1880 	}
1881 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1882 	return skb->len;
1883 
1884 out_nlmsg_trim:
1885 nla_put_failure:
1886 	nlmsg_trim(skb, b);
1887 	return -1;
1888 }
1889 
1890 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1891 			  struct nlmsghdr *n, struct tcf_proto *tp,
1892 			  struct tcf_block *block, struct Qdisc *q,
1893 			  u32 parent, void *fh, int event, bool unicast,
1894 			  bool rtnl_held)
1895 {
1896 	struct sk_buff *skb;
1897 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1898 	int err = 0;
1899 
1900 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1901 	if (!skb)
1902 		return -ENOBUFS;
1903 
1904 	if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1905 			  n->nlmsg_seq, n->nlmsg_flags, event,
1906 			  rtnl_held) <= 0) {
1907 		kfree_skb(skb);
1908 		return -EINVAL;
1909 	}
1910 
1911 	if (unicast)
1912 		err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1913 	else
1914 		err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1915 				     n->nlmsg_flags & NLM_F_ECHO);
1916 
1917 	if (err > 0)
1918 		err = 0;
1919 	return err;
1920 }
1921 
1922 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1923 			      struct nlmsghdr *n, struct tcf_proto *tp,
1924 			      struct tcf_block *block, struct Qdisc *q,
1925 			      u32 parent, void *fh, bool unicast, bool *last,
1926 			      bool rtnl_held, struct netlink_ext_ack *extack)
1927 {
1928 	struct sk_buff *skb;
1929 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1930 	int err;
1931 
1932 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1933 	if (!skb)
1934 		return -ENOBUFS;
1935 
1936 	if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1937 			  n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
1938 			  rtnl_held) <= 0) {
1939 		NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1940 		kfree_skb(skb);
1941 		return -EINVAL;
1942 	}
1943 
1944 	err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
1945 	if (err) {
1946 		kfree_skb(skb);
1947 		return err;
1948 	}
1949 
1950 	if (unicast)
1951 		err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1952 	else
1953 		err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1954 				     n->nlmsg_flags & NLM_F_ECHO);
1955 	if (err < 0)
1956 		NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1957 
1958 	if (err > 0)
1959 		err = 0;
1960 	return err;
1961 }
1962 
1963 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1964 				 struct tcf_block *block, struct Qdisc *q,
1965 				 u32 parent, struct nlmsghdr *n,
1966 				 struct tcf_chain *chain, int event,
1967 				 bool rtnl_held)
1968 {
1969 	struct tcf_proto *tp;
1970 
1971 	for (tp = tcf_get_next_proto(chain, NULL, rtnl_held);
1972 	     tp; tp = tcf_get_next_proto(chain, tp, rtnl_held))
1973 		tfilter_notify(net, oskb, n, tp, block,
1974 			       q, parent, NULL, event, false, rtnl_held);
1975 }
1976 
1977 static void tfilter_put(struct tcf_proto *tp, void *fh)
1978 {
1979 	if (tp->ops->put && fh)
1980 		tp->ops->put(tp, fh);
1981 }
1982 
1983 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1984 			  struct netlink_ext_ack *extack)
1985 {
1986 	struct net *net = sock_net(skb->sk);
1987 	struct nlattr *tca[TCA_MAX + 1];
1988 	char name[IFNAMSIZ];
1989 	struct tcmsg *t;
1990 	u32 protocol;
1991 	u32 prio;
1992 	bool prio_allocate;
1993 	u32 parent;
1994 	u32 chain_index;
1995 	struct Qdisc *q = NULL;
1996 	struct tcf_chain_info chain_info;
1997 	struct tcf_chain *chain = NULL;
1998 	struct tcf_block *block;
1999 	struct tcf_proto *tp;
2000 	unsigned long cl;
2001 	void *fh;
2002 	int err;
2003 	int tp_created;
2004 	bool rtnl_held = false;
2005 
2006 	if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2007 		return -EPERM;
2008 
2009 replay:
2010 	tp_created = 0;
2011 
2012 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2013 				     rtm_tca_policy, extack);
2014 	if (err < 0)
2015 		return err;
2016 
2017 	t = nlmsg_data(n);
2018 	protocol = TC_H_MIN(t->tcm_info);
2019 	prio = TC_H_MAJ(t->tcm_info);
2020 	prio_allocate = false;
2021 	parent = t->tcm_parent;
2022 	tp = NULL;
2023 	cl = 0;
2024 	block = NULL;
2025 
2026 	if (prio == 0) {
2027 		/* If no priority is provided by the user,
2028 		 * we allocate one.
2029 		 */
2030 		if (n->nlmsg_flags & NLM_F_CREATE) {
2031 			prio = TC_H_MAKE(0x80000000U, 0U);
2032 			prio_allocate = true;
2033 		} else {
2034 			NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2035 			return -ENOENT;
2036 		}
2037 	}
2038 
2039 	/* Find head of filter chain. */
2040 
2041 	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2042 	if (err)
2043 		return err;
2044 
2045 	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2046 		NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2047 		err = -EINVAL;
2048 		goto errout;
2049 	}
2050 
2051 	/* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2052 	 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2053 	 * type is not specified, classifier is not unlocked.
2054 	 */
2055 	if (rtnl_held ||
2056 	    (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2057 	    !tcf_proto_is_unlocked(name)) {
2058 		rtnl_held = true;
2059 		rtnl_lock();
2060 	}
2061 
2062 	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2063 	if (err)
2064 		goto errout;
2065 
2066 	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2067 				 extack);
2068 	if (IS_ERR(block)) {
2069 		err = PTR_ERR(block);
2070 		goto errout;
2071 	}
2072 
2073 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2074 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2075 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2076 		err = -EINVAL;
2077 		goto errout;
2078 	}
2079 	chain = tcf_chain_get(block, chain_index, true);
2080 	if (!chain) {
2081 		NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2082 		err = -ENOMEM;
2083 		goto errout;
2084 	}
2085 
2086 	mutex_lock(&chain->filter_chain_lock);
2087 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2088 			       prio, prio_allocate);
2089 	if (IS_ERR(tp)) {
2090 		NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2091 		err = PTR_ERR(tp);
2092 		goto errout_locked;
2093 	}
2094 
2095 	if (tp == NULL) {
2096 		struct tcf_proto *tp_new = NULL;
2097 
2098 		if (chain->flushing) {
2099 			err = -EAGAIN;
2100 			goto errout_locked;
2101 		}
2102 
2103 		/* Proto-tcf does not exist, create new one */
2104 
2105 		if (tca[TCA_KIND] == NULL || !protocol) {
2106 			NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2107 			err = -EINVAL;
2108 			goto errout_locked;
2109 		}
2110 
2111 		if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2112 			NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2113 			err = -ENOENT;
2114 			goto errout_locked;
2115 		}
2116 
2117 		if (prio_allocate)
2118 			prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2119 							       &chain_info));
2120 
2121 		mutex_unlock(&chain->filter_chain_lock);
2122 		tp_new = tcf_proto_create(name, protocol, prio, chain,
2123 					  rtnl_held, extack);
2124 		if (IS_ERR(tp_new)) {
2125 			err = PTR_ERR(tp_new);
2126 			goto errout_tp;
2127 		}
2128 
2129 		tp_created = 1;
2130 		tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2131 						rtnl_held);
2132 		if (IS_ERR(tp)) {
2133 			err = PTR_ERR(tp);
2134 			goto errout_tp;
2135 		}
2136 	} else {
2137 		mutex_unlock(&chain->filter_chain_lock);
2138 	}
2139 
2140 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2141 		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2142 		err = -EINVAL;
2143 		goto errout;
2144 	}
2145 
2146 	fh = tp->ops->get(tp, t->tcm_handle);
2147 
2148 	if (!fh) {
2149 		if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2150 			NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2151 			err = -ENOENT;
2152 			goto errout;
2153 		}
2154 	} else if (n->nlmsg_flags & NLM_F_EXCL) {
2155 		tfilter_put(tp, fh);
2156 		NL_SET_ERR_MSG(extack, "Filter already exists");
2157 		err = -EEXIST;
2158 		goto errout;
2159 	}
2160 
2161 	if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2162 		NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2163 		err = -EINVAL;
2164 		goto errout;
2165 	}
2166 
2167 	err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2168 			      n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
2169 			      rtnl_held, extack);
2170 	if (err == 0) {
2171 		tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2172 			       RTM_NEWTFILTER, false, rtnl_held);
2173 		tfilter_put(tp, fh);
2174 		/* q pointer is NULL for shared blocks */
2175 		if (q)
2176 			q->flags &= ~TCQ_F_CAN_BYPASS;
2177 	}
2178 
2179 errout:
2180 	if (err && tp_created)
2181 		tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2182 errout_tp:
2183 	if (chain) {
2184 		if (tp && !IS_ERR(tp))
2185 			tcf_proto_put(tp, rtnl_held, NULL);
2186 		if (!tp_created)
2187 			tcf_chain_put(chain);
2188 	}
2189 	tcf_block_release(q, block, rtnl_held);
2190 
2191 	if (rtnl_held)
2192 		rtnl_unlock();
2193 
2194 	if (err == -EAGAIN) {
2195 		/* Take rtnl lock in case EAGAIN is caused by concurrent flush
2196 		 * of target chain.
2197 		 */
2198 		rtnl_held = true;
2199 		/* Replay the request. */
2200 		goto replay;
2201 	}
2202 	return err;
2203 
2204 errout_locked:
2205 	mutex_unlock(&chain->filter_chain_lock);
2206 	goto errout;
2207 }
2208 
2209 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2210 			  struct netlink_ext_ack *extack)
2211 {
2212 	struct net *net = sock_net(skb->sk);
2213 	struct nlattr *tca[TCA_MAX + 1];
2214 	char name[IFNAMSIZ];
2215 	struct tcmsg *t;
2216 	u32 protocol;
2217 	u32 prio;
2218 	u32 parent;
2219 	u32 chain_index;
2220 	struct Qdisc *q = NULL;
2221 	struct tcf_chain_info chain_info;
2222 	struct tcf_chain *chain = NULL;
2223 	struct tcf_block *block = NULL;
2224 	struct tcf_proto *tp = NULL;
2225 	unsigned long cl = 0;
2226 	void *fh = NULL;
2227 	int err;
2228 	bool rtnl_held = false;
2229 
2230 	if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2231 		return -EPERM;
2232 
2233 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2234 				     rtm_tca_policy, extack);
2235 	if (err < 0)
2236 		return err;
2237 
2238 	t = nlmsg_data(n);
2239 	protocol = TC_H_MIN(t->tcm_info);
2240 	prio = TC_H_MAJ(t->tcm_info);
2241 	parent = t->tcm_parent;
2242 
2243 	if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2244 		NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2245 		return -ENOENT;
2246 	}
2247 
2248 	/* Find head of filter chain. */
2249 
2250 	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2251 	if (err)
2252 		return err;
2253 
2254 	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2255 		NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2256 		err = -EINVAL;
2257 		goto errout;
2258 	}
2259 	/* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2260 	 * found), qdisc is not unlocked, classifier type is not specified,
2261 	 * classifier is not unlocked.
2262 	 */
2263 	if (!prio ||
2264 	    (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2265 	    !tcf_proto_is_unlocked(name)) {
2266 		rtnl_held = true;
2267 		rtnl_lock();
2268 	}
2269 
2270 	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2271 	if (err)
2272 		goto errout;
2273 
2274 	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2275 				 extack);
2276 	if (IS_ERR(block)) {
2277 		err = PTR_ERR(block);
2278 		goto errout;
2279 	}
2280 
2281 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2282 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2283 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2284 		err = -EINVAL;
2285 		goto errout;
2286 	}
2287 	chain = tcf_chain_get(block, chain_index, false);
2288 	if (!chain) {
2289 		/* User requested flush on non-existent chain. Nothing to do,
2290 		 * so just return success.
2291 		 */
2292 		if (prio == 0) {
2293 			err = 0;
2294 			goto errout;
2295 		}
2296 		NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2297 		err = -ENOENT;
2298 		goto errout;
2299 	}
2300 
2301 	if (prio == 0) {
2302 		tfilter_notify_chain(net, skb, block, q, parent, n,
2303 				     chain, RTM_DELTFILTER, rtnl_held);
2304 		tcf_chain_flush(chain, rtnl_held);
2305 		err = 0;
2306 		goto errout;
2307 	}
2308 
2309 	mutex_lock(&chain->filter_chain_lock);
2310 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2311 			       prio, false);
2312 	if (!tp || IS_ERR(tp)) {
2313 		NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2314 		err = tp ? PTR_ERR(tp) : -ENOENT;
2315 		goto errout_locked;
2316 	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2317 		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2318 		err = -EINVAL;
2319 		goto errout_locked;
2320 	} else if (t->tcm_handle == 0) {
2321 		tcf_proto_signal_destroying(chain, tp);
2322 		tcf_chain_tp_remove(chain, &chain_info, tp);
2323 		mutex_unlock(&chain->filter_chain_lock);
2324 
2325 		tcf_proto_put(tp, rtnl_held, NULL);
2326 		tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2327 			       RTM_DELTFILTER, false, rtnl_held);
2328 		err = 0;
2329 		goto errout;
2330 	}
2331 	mutex_unlock(&chain->filter_chain_lock);
2332 
2333 	fh = tp->ops->get(tp, t->tcm_handle);
2334 
2335 	if (!fh) {
2336 		NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2337 		err = -ENOENT;
2338 	} else {
2339 		bool last;
2340 
2341 		err = tfilter_del_notify(net, skb, n, tp, block,
2342 					 q, parent, fh, false, &last,
2343 					 rtnl_held, extack);
2344 
2345 		if (err)
2346 			goto errout;
2347 		if (last)
2348 			tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2349 	}
2350 
2351 errout:
2352 	if (chain) {
2353 		if (tp && !IS_ERR(tp))
2354 			tcf_proto_put(tp, rtnl_held, NULL);
2355 		tcf_chain_put(chain);
2356 	}
2357 	tcf_block_release(q, block, rtnl_held);
2358 
2359 	if (rtnl_held)
2360 		rtnl_unlock();
2361 
2362 	return err;
2363 
2364 errout_locked:
2365 	mutex_unlock(&chain->filter_chain_lock);
2366 	goto errout;
2367 }
2368 
2369 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2370 			  struct netlink_ext_ack *extack)
2371 {
2372 	struct net *net = sock_net(skb->sk);
2373 	struct nlattr *tca[TCA_MAX + 1];
2374 	char name[IFNAMSIZ];
2375 	struct tcmsg *t;
2376 	u32 protocol;
2377 	u32 prio;
2378 	u32 parent;
2379 	u32 chain_index;
2380 	struct Qdisc *q = NULL;
2381 	struct tcf_chain_info chain_info;
2382 	struct tcf_chain *chain = NULL;
2383 	struct tcf_block *block = NULL;
2384 	struct tcf_proto *tp = NULL;
2385 	unsigned long cl = 0;
2386 	void *fh = NULL;
2387 	int err;
2388 	bool rtnl_held = false;
2389 
2390 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2391 				     rtm_tca_policy, extack);
2392 	if (err < 0)
2393 		return err;
2394 
2395 	t = nlmsg_data(n);
2396 	protocol = TC_H_MIN(t->tcm_info);
2397 	prio = TC_H_MAJ(t->tcm_info);
2398 	parent = t->tcm_parent;
2399 
2400 	if (prio == 0) {
2401 		NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2402 		return -ENOENT;
2403 	}
2404 
2405 	/* Find head of filter chain. */
2406 
2407 	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2408 	if (err)
2409 		return err;
2410 
2411 	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2412 		NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2413 		err = -EINVAL;
2414 		goto errout;
2415 	}
2416 	/* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2417 	 * unlocked, classifier type is not specified, classifier is not
2418 	 * unlocked.
2419 	 */
2420 	if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2421 	    !tcf_proto_is_unlocked(name)) {
2422 		rtnl_held = true;
2423 		rtnl_lock();
2424 	}
2425 
2426 	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2427 	if (err)
2428 		goto errout;
2429 
2430 	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2431 				 extack);
2432 	if (IS_ERR(block)) {
2433 		err = PTR_ERR(block);
2434 		goto errout;
2435 	}
2436 
2437 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2438 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2439 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2440 		err = -EINVAL;
2441 		goto errout;
2442 	}
2443 	chain = tcf_chain_get(block, chain_index, false);
2444 	if (!chain) {
2445 		NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2446 		err = -EINVAL;
2447 		goto errout;
2448 	}
2449 
2450 	mutex_lock(&chain->filter_chain_lock);
2451 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2452 			       prio, false);
2453 	mutex_unlock(&chain->filter_chain_lock);
2454 	if (!tp || IS_ERR(tp)) {
2455 		NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2456 		err = tp ? PTR_ERR(tp) : -ENOENT;
2457 		goto errout;
2458 	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2459 		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2460 		err = -EINVAL;
2461 		goto errout;
2462 	}
2463 
2464 	fh = tp->ops->get(tp, t->tcm_handle);
2465 
2466 	if (!fh) {
2467 		NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2468 		err = -ENOENT;
2469 	} else {
2470 		err = tfilter_notify(net, skb, n, tp, block, q, parent,
2471 				     fh, RTM_NEWTFILTER, true, rtnl_held);
2472 		if (err < 0)
2473 			NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2474 	}
2475 
2476 	tfilter_put(tp, fh);
2477 errout:
2478 	if (chain) {
2479 		if (tp && !IS_ERR(tp))
2480 			tcf_proto_put(tp, rtnl_held, NULL);
2481 		tcf_chain_put(chain);
2482 	}
2483 	tcf_block_release(q, block, rtnl_held);
2484 
2485 	if (rtnl_held)
2486 		rtnl_unlock();
2487 
2488 	return err;
2489 }
2490 
2491 struct tcf_dump_args {
2492 	struct tcf_walker w;
2493 	struct sk_buff *skb;
2494 	struct netlink_callback *cb;
2495 	struct tcf_block *block;
2496 	struct Qdisc *q;
2497 	u32 parent;
2498 };
2499 
2500 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2501 {
2502 	struct tcf_dump_args *a = (void *)arg;
2503 	struct net *net = sock_net(a->skb->sk);
2504 
2505 	return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2506 			     n, NETLINK_CB(a->cb->skb).portid,
2507 			     a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2508 			     RTM_NEWTFILTER, true);
2509 }
2510 
2511 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2512 			   struct sk_buff *skb, struct netlink_callback *cb,
2513 			   long index_start, long *p_index)
2514 {
2515 	struct net *net = sock_net(skb->sk);
2516 	struct tcf_block *block = chain->block;
2517 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2518 	struct tcf_proto *tp, *tp_prev;
2519 	struct tcf_dump_args arg;
2520 
2521 	for (tp = __tcf_get_next_proto(chain, NULL);
2522 	     tp;
2523 	     tp_prev = tp,
2524 		     tp = __tcf_get_next_proto(chain, tp),
2525 		     tcf_proto_put(tp_prev, true, NULL),
2526 		     (*p_index)++) {
2527 		if (*p_index < index_start)
2528 			continue;
2529 		if (TC_H_MAJ(tcm->tcm_info) &&
2530 		    TC_H_MAJ(tcm->tcm_info) != tp->prio)
2531 			continue;
2532 		if (TC_H_MIN(tcm->tcm_info) &&
2533 		    TC_H_MIN(tcm->tcm_info) != tp->protocol)
2534 			continue;
2535 		if (*p_index > index_start)
2536 			memset(&cb->args[1], 0,
2537 			       sizeof(cb->args) - sizeof(cb->args[0]));
2538 		if (cb->args[1] == 0) {
2539 			if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2540 					  NETLINK_CB(cb->skb).portid,
2541 					  cb->nlh->nlmsg_seq, NLM_F_MULTI,
2542 					  RTM_NEWTFILTER, true) <= 0)
2543 				goto errout;
2544 			cb->args[1] = 1;
2545 		}
2546 		if (!tp->ops->walk)
2547 			continue;
2548 		arg.w.fn = tcf_node_dump;
2549 		arg.skb = skb;
2550 		arg.cb = cb;
2551 		arg.block = block;
2552 		arg.q = q;
2553 		arg.parent = parent;
2554 		arg.w.stop = 0;
2555 		arg.w.skip = cb->args[1] - 1;
2556 		arg.w.count = 0;
2557 		arg.w.cookie = cb->args[2];
2558 		tp->ops->walk(tp, &arg.w, true);
2559 		cb->args[2] = arg.w.cookie;
2560 		cb->args[1] = arg.w.count + 1;
2561 		if (arg.w.stop)
2562 			goto errout;
2563 	}
2564 	return true;
2565 
2566 errout:
2567 	tcf_proto_put(tp, true, NULL);
2568 	return false;
2569 }
2570 
2571 /* called with RTNL */
2572 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2573 {
2574 	struct tcf_chain *chain, *chain_prev;
2575 	struct net *net = sock_net(skb->sk);
2576 	struct nlattr *tca[TCA_MAX + 1];
2577 	struct Qdisc *q = NULL;
2578 	struct tcf_block *block;
2579 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2580 	long index_start;
2581 	long index;
2582 	u32 parent;
2583 	int err;
2584 
2585 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2586 		return skb->len;
2587 
2588 	err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2589 				     NULL, cb->extack);
2590 	if (err)
2591 		return err;
2592 
2593 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2594 		block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2595 		if (!block)
2596 			goto out;
2597 		/* If we work with block index, q is NULL and parent value
2598 		 * will never be used in the following code. The check
2599 		 * in tcf_fill_node prevents it. However, compiler does not
2600 		 * see that far, so set parent to zero to silence the warning
2601 		 * about parent being uninitialized.
2602 		 */
2603 		parent = 0;
2604 	} else {
2605 		const struct Qdisc_class_ops *cops;
2606 		struct net_device *dev;
2607 		unsigned long cl = 0;
2608 
2609 		dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2610 		if (!dev)
2611 			return skb->len;
2612 
2613 		parent = tcm->tcm_parent;
2614 		if (!parent) {
2615 			q = dev->qdisc;
2616 			parent = q->handle;
2617 		} else {
2618 			q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2619 		}
2620 		if (!q)
2621 			goto out;
2622 		cops = q->ops->cl_ops;
2623 		if (!cops)
2624 			goto out;
2625 		if (!cops->tcf_block)
2626 			goto out;
2627 		if (TC_H_MIN(tcm->tcm_parent)) {
2628 			cl = cops->find(q, tcm->tcm_parent);
2629 			if (cl == 0)
2630 				goto out;
2631 		}
2632 		block = cops->tcf_block(q, cl, NULL);
2633 		if (!block)
2634 			goto out;
2635 		if (tcf_block_shared(block))
2636 			q = NULL;
2637 	}
2638 
2639 	index_start = cb->args[0];
2640 	index = 0;
2641 
2642 	for (chain = __tcf_get_next_chain(block, NULL);
2643 	     chain;
2644 	     chain_prev = chain,
2645 		     chain = __tcf_get_next_chain(block, chain),
2646 		     tcf_chain_put(chain_prev)) {
2647 		if (tca[TCA_CHAIN] &&
2648 		    nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2649 			continue;
2650 		if (!tcf_chain_dump(chain, q, parent, skb, cb,
2651 				    index_start, &index)) {
2652 			tcf_chain_put(chain);
2653 			err = -EMSGSIZE;
2654 			break;
2655 		}
2656 	}
2657 
2658 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2659 		tcf_block_refcnt_put(block, true);
2660 	cb->args[0] = index;
2661 
2662 out:
2663 	/* If we did no progress, the error (EMSGSIZE) is real */
2664 	if (skb->len == 0 && err)
2665 		return err;
2666 	return skb->len;
2667 }
2668 
2669 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2670 			      void *tmplt_priv, u32 chain_index,
2671 			      struct net *net, struct sk_buff *skb,
2672 			      struct tcf_block *block,
2673 			      u32 portid, u32 seq, u16 flags, int event)
2674 {
2675 	unsigned char *b = skb_tail_pointer(skb);
2676 	const struct tcf_proto_ops *ops;
2677 	struct nlmsghdr *nlh;
2678 	struct tcmsg *tcm;
2679 	void *priv;
2680 
2681 	ops = tmplt_ops;
2682 	priv = tmplt_priv;
2683 
2684 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2685 	if (!nlh)
2686 		goto out_nlmsg_trim;
2687 	tcm = nlmsg_data(nlh);
2688 	tcm->tcm_family = AF_UNSPEC;
2689 	tcm->tcm__pad1 = 0;
2690 	tcm->tcm__pad2 = 0;
2691 	tcm->tcm_handle = 0;
2692 	if (block->q) {
2693 		tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2694 		tcm->tcm_parent = block->q->handle;
2695 	} else {
2696 		tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2697 		tcm->tcm_block_index = block->index;
2698 	}
2699 
2700 	if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2701 		goto nla_put_failure;
2702 
2703 	if (ops) {
2704 		if (nla_put_string(skb, TCA_KIND, ops->kind))
2705 			goto nla_put_failure;
2706 		if (ops->tmplt_dump(skb, net, priv) < 0)
2707 			goto nla_put_failure;
2708 	}
2709 
2710 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2711 	return skb->len;
2712 
2713 out_nlmsg_trim:
2714 nla_put_failure:
2715 	nlmsg_trim(skb, b);
2716 	return -EMSGSIZE;
2717 }
2718 
2719 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2720 			   u32 seq, u16 flags, int event, bool unicast)
2721 {
2722 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2723 	struct tcf_block *block = chain->block;
2724 	struct net *net = block->net;
2725 	struct sk_buff *skb;
2726 	int err = 0;
2727 
2728 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2729 	if (!skb)
2730 		return -ENOBUFS;
2731 
2732 	if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2733 			       chain->index, net, skb, block, portid,
2734 			       seq, flags, event) <= 0) {
2735 		kfree_skb(skb);
2736 		return -EINVAL;
2737 	}
2738 
2739 	if (unicast)
2740 		err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2741 	else
2742 		err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2743 				     flags & NLM_F_ECHO);
2744 
2745 	if (err > 0)
2746 		err = 0;
2747 	return err;
2748 }
2749 
2750 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2751 				  void *tmplt_priv, u32 chain_index,
2752 				  struct tcf_block *block, struct sk_buff *oskb,
2753 				  u32 seq, u16 flags, bool unicast)
2754 {
2755 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2756 	struct net *net = block->net;
2757 	struct sk_buff *skb;
2758 
2759 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2760 	if (!skb)
2761 		return -ENOBUFS;
2762 
2763 	if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2764 			       block, portid, seq, flags, RTM_DELCHAIN) <= 0) {
2765 		kfree_skb(skb);
2766 		return -EINVAL;
2767 	}
2768 
2769 	if (unicast)
2770 		return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2771 
2772 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2773 }
2774 
2775 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2776 			      struct nlattr **tca,
2777 			      struct netlink_ext_ack *extack)
2778 {
2779 	const struct tcf_proto_ops *ops;
2780 	char name[IFNAMSIZ];
2781 	void *tmplt_priv;
2782 
2783 	/* If kind is not set, user did not specify template. */
2784 	if (!tca[TCA_KIND])
2785 		return 0;
2786 
2787 	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2788 		NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
2789 		return -EINVAL;
2790 	}
2791 
2792 	ops = tcf_proto_lookup_ops(name, true, extack);
2793 	if (IS_ERR(ops))
2794 		return PTR_ERR(ops);
2795 	if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2796 		NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2797 		return -EOPNOTSUPP;
2798 	}
2799 
2800 	tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2801 	if (IS_ERR(tmplt_priv)) {
2802 		module_put(ops->owner);
2803 		return PTR_ERR(tmplt_priv);
2804 	}
2805 	chain->tmplt_ops = ops;
2806 	chain->tmplt_priv = tmplt_priv;
2807 	return 0;
2808 }
2809 
2810 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2811 			       void *tmplt_priv)
2812 {
2813 	/* If template ops are set, no work to do for us. */
2814 	if (!tmplt_ops)
2815 		return;
2816 
2817 	tmplt_ops->tmplt_destroy(tmplt_priv);
2818 	module_put(tmplt_ops->owner);
2819 }
2820 
2821 /* Add/delete/get a chain */
2822 
2823 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2824 			struct netlink_ext_ack *extack)
2825 {
2826 	struct net *net = sock_net(skb->sk);
2827 	struct nlattr *tca[TCA_MAX + 1];
2828 	struct tcmsg *t;
2829 	u32 parent;
2830 	u32 chain_index;
2831 	struct Qdisc *q = NULL;
2832 	struct tcf_chain *chain = NULL;
2833 	struct tcf_block *block;
2834 	unsigned long cl;
2835 	int err;
2836 
2837 	if (n->nlmsg_type != RTM_GETCHAIN &&
2838 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2839 		return -EPERM;
2840 
2841 replay:
2842 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2843 				     rtm_tca_policy, extack);
2844 	if (err < 0)
2845 		return err;
2846 
2847 	t = nlmsg_data(n);
2848 	parent = t->tcm_parent;
2849 	cl = 0;
2850 
2851 	block = tcf_block_find(net, &q, &parent, &cl,
2852 			       t->tcm_ifindex, t->tcm_block_index, extack);
2853 	if (IS_ERR(block))
2854 		return PTR_ERR(block);
2855 
2856 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2857 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2858 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2859 		err = -EINVAL;
2860 		goto errout_block;
2861 	}
2862 
2863 	mutex_lock(&block->lock);
2864 	chain = tcf_chain_lookup(block, chain_index);
2865 	if (n->nlmsg_type == RTM_NEWCHAIN) {
2866 		if (chain) {
2867 			if (tcf_chain_held_by_acts_only(chain)) {
2868 				/* The chain exists only because there is
2869 				 * some action referencing it.
2870 				 */
2871 				tcf_chain_hold(chain);
2872 			} else {
2873 				NL_SET_ERR_MSG(extack, "Filter chain already exists");
2874 				err = -EEXIST;
2875 				goto errout_block_locked;
2876 			}
2877 		} else {
2878 			if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2879 				NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
2880 				err = -ENOENT;
2881 				goto errout_block_locked;
2882 			}
2883 			chain = tcf_chain_create(block, chain_index);
2884 			if (!chain) {
2885 				NL_SET_ERR_MSG(extack, "Failed to create filter chain");
2886 				err = -ENOMEM;
2887 				goto errout_block_locked;
2888 			}
2889 		}
2890 	} else {
2891 		if (!chain || tcf_chain_held_by_acts_only(chain)) {
2892 			NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2893 			err = -EINVAL;
2894 			goto errout_block_locked;
2895 		}
2896 		tcf_chain_hold(chain);
2897 	}
2898 
2899 	if (n->nlmsg_type == RTM_NEWCHAIN) {
2900 		/* Modifying chain requires holding parent block lock. In case
2901 		 * the chain was successfully added, take a reference to the
2902 		 * chain. This ensures that an empty chain does not disappear at
2903 		 * the end of this function.
2904 		 */
2905 		tcf_chain_hold(chain);
2906 		chain->explicitly_created = true;
2907 	}
2908 	mutex_unlock(&block->lock);
2909 
2910 	switch (n->nlmsg_type) {
2911 	case RTM_NEWCHAIN:
2912 		err = tc_chain_tmplt_add(chain, net, tca, extack);
2913 		if (err) {
2914 			tcf_chain_put_explicitly_created(chain);
2915 			goto errout;
2916 		}
2917 
2918 		tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2919 				RTM_NEWCHAIN, false);
2920 		break;
2921 	case RTM_DELCHAIN:
2922 		tfilter_notify_chain(net, skb, block, q, parent, n,
2923 				     chain, RTM_DELTFILTER, true);
2924 		/* Flush the chain first as the user requested chain removal. */
2925 		tcf_chain_flush(chain, true);
2926 		/* In case the chain was successfully deleted, put a reference
2927 		 * to the chain previously taken during addition.
2928 		 */
2929 		tcf_chain_put_explicitly_created(chain);
2930 		break;
2931 	case RTM_GETCHAIN:
2932 		err = tc_chain_notify(chain, skb, n->nlmsg_seq,
2933 				      n->nlmsg_seq, n->nlmsg_type, true);
2934 		if (err < 0)
2935 			NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2936 		break;
2937 	default:
2938 		err = -EOPNOTSUPP;
2939 		NL_SET_ERR_MSG(extack, "Unsupported message type");
2940 		goto errout;
2941 	}
2942 
2943 errout:
2944 	tcf_chain_put(chain);
2945 errout_block:
2946 	tcf_block_release(q, block, true);
2947 	if (err == -EAGAIN)
2948 		/* Replay the request. */
2949 		goto replay;
2950 	return err;
2951 
2952 errout_block_locked:
2953 	mutex_unlock(&block->lock);
2954 	goto errout_block;
2955 }
2956 
2957 /* called with RTNL */
2958 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2959 {
2960 	struct net *net = sock_net(skb->sk);
2961 	struct nlattr *tca[TCA_MAX + 1];
2962 	struct Qdisc *q = NULL;
2963 	struct tcf_block *block;
2964 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2965 	struct tcf_chain *chain;
2966 	long index_start;
2967 	long index;
2968 	u32 parent;
2969 	int err;
2970 
2971 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2972 		return skb->len;
2973 
2974 	err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2975 				     rtm_tca_policy, cb->extack);
2976 	if (err)
2977 		return err;
2978 
2979 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2980 		block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2981 		if (!block)
2982 			goto out;
2983 		/* If we work with block index, q is NULL and parent value
2984 		 * will never be used in the following code. The check
2985 		 * in tcf_fill_node prevents it. However, compiler does not
2986 		 * see that far, so set parent to zero to silence the warning
2987 		 * about parent being uninitialized.
2988 		 */
2989 		parent = 0;
2990 	} else {
2991 		const struct Qdisc_class_ops *cops;
2992 		struct net_device *dev;
2993 		unsigned long cl = 0;
2994 
2995 		dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2996 		if (!dev)
2997 			return skb->len;
2998 
2999 		parent = tcm->tcm_parent;
3000 		if (!parent) {
3001 			q = dev->qdisc;
3002 			parent = q->handle;
3003 		} else {
3004 			q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
3005 		}
3006 		if (!q)
3007 			goto out;
3008 		cops = q->ops->cl_ops;
3009 		if (!cops)
3010 			goto out;
3011 		if (!cops->tcf_block)
3012 			goto out;
3013 		if (TC_H_MIN(tcm->tcm_parent)) {
3014 			cl = cops->find(q, tcm->tcm_parent);
3015 			if (cl == 0)
3016 				goto out;
3017 		}
3018 		block = cops->tcf_block(q, cl, NULL);
3019 		if (!block)
3020 			goto out;
3021 		if (tcf_block_shared(block))
3022 			q = NULL;
3023 	}
3024 
3025 	index_start = cb->args[0];
3026 	index = 0;
3027 
3028 	mutex_lock(&block->lock);
3029 	list_for_each_entry(chain, &block->chain_list, list) {
3030 		if ((tca[TCA_CHAIN] &&
3031 		     nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3032 			continue;
3033 		if (index < index_start) {
3034 			index++;
3035 			continue;
3036 		}
3037 		if (tcf_chain_held_by_acts_only(chain))
3038 			continue;
3039 		err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3040 					 chain->index, net, skb, block,
3041 					 NETLINK_CB(cb->skb).portid,
3042 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3043 					 RTM_NEWCHAIN);
3044 		if (err <= 0)
3045 			break;
3046 		index++;
3047 	}
3048 	mutex_unlock(&block->lock);
3049 
3050 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3051 		tcf_block_refcnt_put(block, true);
3052 	cb->args[0] = index;
3053 
3054 out:
3055 	/* If we did no progress, the error (EMSGSIZE) is real */
3056 	if (skb->len == 0 && err)
3057 		return err;
3058 	return skb->len;
3059 }
3060 
3061 void tcf_exts_destroy(struct tcf_exts *exts)
3062 {
3063 #ifdef CONFIG_NET_CLS_ACT
3064 	if (exts->actions) {
3065 		tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3066 		kfree(exts->actions);
3067 	}
3068 	exts->nr_actions = 0;
3069 #endif
3070 }
3071 EXPORT_SYMBOL(tcf_exts_destroy);
3072 
3073 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3074 		      struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
3075 		      bool rtnl_held, struct netlink_ext_ack *extack)
3076 {
3077 #ifdef CONFIG_NET_CLS_ACT
3078 	{
3079 		struct tc_action *act;
3080 		size_t attr_size = 0;
3081 
3082 		if (exts->police && tb[exts->police]) {
3083 			act = tcf_action_init_1(net, tp, tb[exts->police],
3084 						rate_tlv, "police", ovr,
3085 						TCA_ACT_BIND, rtnl_held,
3086 						extack);
3087 			if (IS_ERR(act))
3088 				return PTR_ERR(act);
3089 
3090 			act->type = exts->type = TCA_OLD_COMPAT;
3091 			exts->actions[0] = act;
3092 			exts->nr_actions = 1;
3093 		} else if (exts->action && tb[exts->action]) {
3094 			int err;
3095 
3096 			err = tcf_action_init(net, tp, tb[exts->action],
3097 					      rate_tlv, NULL, ovr, TCA_ACT_BIND,
3098 					      exts->actions, &attr_size,
3099 					      rtnl_held, extack);
3100 			if (err < 0)
3101 				return err;
3102 			exts->nr_actions = err;
3103 		}
3104 	}
3105 #else
3106 	if ((exts->action && tb[exts->action]) ||
3107 	    (exts->police && tb[exts->police])) {
3108 		NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3109 		return -EOPNOTSUPP;
3110 	}
3111 #endif
3112 
3113 	return 0;
3114 }
3115 EXPORT_SYMBOL(tcf_exts_validate);
3116 
3117 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3118 {
3119 #ifdef CONFIG_NET_CLS_ACT
3120 	struct tcf_exts old = *dst;
3121 
3122 	*dst = *src;
3123 	tcf_exts_destroy(&old);
3124 #endif
3125 }
3126 EXPORT_SYMBOL(tcf_exts_change);
3127 
3128 #ifdef CONFIG_NET_CLS_ACT
3129 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3130 {
3131 	if (exts->nr_actions == 0)
3132 		return NULL;
3133 	else
3134 		return exts->actions[0];
3135 }
3136 #endif
3137 
3138 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3139 {
3140 #ifdef CONFIG_NET_CLS_ACT
3141 	struct nlattr *nest;
3142 
3143 	if (exts->action && tcf_exts_has_actions(exts)) {
3144 		/*
3145 		 * again for backward compatible mode - we want
3146 		 * to work with both old and new modes of entering
3147 		 * tc data even if iproute2  was newer - jhs
3148 		 */
3149 		if (exts->type != TCA_OLD_COMPAT) {
3150 			nest = nla_nest_start_noflag(skb, exts->action);
3151 			if (nest == NULL)
3152 				goto nla_put_failure;
3153 
3154 			if (tcf_action_dump(skb, exts->actions, 0, 0) < 0)
3155 				goto nla_put_failure;
3156 			nla_nest_end(skb, nest);
3157 		} else if (exts->police) {
3158 			struct tc_action *act = tcf_exts_first_act(exts);
3159 			nest = nla_nest_start_noflag(skb, exts->police);
3160 			if (nest == NULL || !act)
3161 				goto nla_put_failure;
3162 			if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3163 				goto nla_put_failure;
3164 			nla_nest_end(skb, nest);
3165 		}
3166 	}
3167 	return 0;
3168 
3169 nla_put_failure:
3170 	nla_nest_cancel(skb, nest);
3171 	return -1;
3172 #else
3173 	return 0;
3174 #endif
3175 }
3176 EXPORT_SYMBOL(tcf_exts_dump);
3177 
3178 
3179 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3180 {
3181 #ifdef CONFIG_NET_CLS_ACT
3182 	struct tc_action *a = tcf_exts_first_act(exts);
3183 	if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3184 		return -1;
3185 #endif
3186 	return 0;
3187 }
3188 EXPORT_SYMBOL(tcf_exts_dump_stats);
3189 
3190 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3191 {
3192 	if (*flags & TCA_CLS_FLAGS_IN_HW)
3193 		return;
3194 	*flags |= TCA_CLS_FLAGS_IN_HW;
3195 	atomic_inc(&block->offloadcnt);
3196 }
3197 
3198 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3199 {
3200 	if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3201 		return;
3202 	*flags &= ~TCA_CLS_FLAGS_IN_HW;
3203 	atomic_dec(&block->offloadcnt);
3204 }
3205 
3206 static void tc_cls_offload_cnt_update(struct tcf_block *block,
3207 				      struct tcf_proto *tp, u32 *cnt,
3208 				      u32 *flags, u32 diff, bool add)
3209 {
3210 	lockdep_assert_held(&block->cb_lock);
3211 
3212 	spin_lock(&tp->lock);
3213 	if (add) {
3214 		if (!*cnt)
3215 			tcf_block_offload_inc(block, flags);
3216 		*cnt += diff;
3217 	} else {
3218 		*cnt -= diff;
3219 		if (!*cnt)
3220 			tcf_block_offload_dec(block, flags);
3221 	}
3222 	spin_unlock(&tp->lock);
3223 }
3224 
3225 static void
3226 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3227 			 u32 *cnt, u32 *flags)
3228 {
3229 	lockdep_assert_held(&block->cb_lock);
3230 
3231 	spin_lock(&tp->lock);
3232 	tcf_block_offload_dec(block, flags);
3233 	*cnt = 0;
3234 	spin_unlock(&tp->lock);
3235 }
3236 
3237 static int
3238 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3239 		   void *type_data, bool err_stop)
3240 {
3241 	struct flow_block_cb *block_cb;
3242 	int ok_count = 0;
3243 	int err;
3244 
3245 	list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3246 		err = block_cb->cb(type, type_data, block_cb->cb_priv);
3247 		if (err) {
3248 			if (err_stop)
3249 				return err;
3250 		} else {
3251 			ok_count++;
3252 		}
3253 	}
3254 	return ok_count;
3255 }
3256 
3257 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3258 		     void *type_data, bool err_stop, bool rtnl_held)
3259 {
3260 	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3261 	int ok_count;
3262 
3263 retry:
3264 	if (take_rtnl)
3265 		rtnl_lock();
3266 	down_read(&block->cb_lock);
3267 	/* Need to obtain rtnl lock if block is bound to devs that require it.
3268 	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3269 	 * obtain the locks in same order here.
3270 	 */
3271 	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3272 		up_read(&block->cb_lock);
3273 		take_rtnl = true;
3274 		goto retry;
3275 	}
3276 
3277 	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3278 
3279 	up_read(&block->cb_lock);
3280 	if (take_rtnl)
3281 		rtnl_unlock();
3282 	return ok_count;
3283 }
3284 EXPORT_SYMBOL(tc_setup_cb_call);
3285 
3286 /* Non-destructive filter add. If filter that wasn't already in hardware is
3287  * successfully offloaded, increment block offloads counter. On failure,
3288  * previously offloaded filter is considered to be intact and offloads counter
3289  * is not decremented.
3290  */
3291 
3292 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3293 		    enum tc_setup_type type, void *type_data, bool err_stop,
3294 		    u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3295 {
3296 	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3297 	int ok_count;
3298 
3299 retry:
3300 	if (take_rtnl)
3301 		rtnl_lock();
3302 	down_read(&block->cb_lock);
3303 	/* Need to obtain rtnl lock if block is bound to devs that require it.
3304 	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3305 	 * obtain the locks in same order here.
3306 	 */
3307 	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3308 		up_read(&block->cb_lock);
3309 		take_rtnl = true;
3310 		goto retry;
3311 	}
3312 
3313 	/* Make sure all netdevs sharing this block are offload-capable. */
3314 	if (block->nooffloaddevcnt && err_stop) {
3315 		ok_count = -EOPNOTSUPP;
3316 		goto err_unlock;
3317 	}
3318 
3319 	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3320 	if (ok_count < 0)
3321 		goto err_unlock;
3322 
3323 	if (tp->ops->hw_add)
3324 		tp->ops->hw_add(tp, type_data);
3325 	if (ok_count > 0)
3326 		tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3327 					  ok_count, true);
3328 err_unlock:
3329 	up_read(&block->cb_lock);
3330 	if (take_rtnl)
3331 		rtnl_unlock();
3332 	return ok_count < 0 ? ok_count : 0;
3333 }
3334 EXPORT_SYMBOL(tc_setup_cb_add);
3335 
3336 /* Destructive filter replace. If filter that wasn't already in hardware is
3337  * successfully offloaded, increment block offload counter. On failure,
3338  * previously offloaded filter is considered to be destroyed and offload counter
3339  * is decremented.
3340  */
3341 
3342 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3343 			enum tc_setup_type type, void *type_data, bool err_stop,
3344 			u32 *old_flags, unsigned int *old_in_hw_count,
3345 			u32 *new_flags, unsigned int *new_in_hw_count,
3346 			bool rtnl_held)
3347 {
3348 	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3349 	int ok_count;
3350 
3351 retry:
3352 	if (take_rtnl)
3353 		rtnl_lock();
3354 	down_read(&block->cb_lock);
3355 	/* Need to obtain rtnl lock if block is bound to devs that require it.
3356 	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3357 	 * obtain the locks in same order here.
3358 	 */
3359 	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3360 		up_read(&block->cb_lock);
3361 		take_rtnl = true;
3362 		goto retry;
3363 	}
3364 
3365 	/* Make sure all netdevs sharing this block are offload-capable. */
3366 	if (block->nooffloaddevcnt && err_stop) {
3367 		ok_count = -EOPNOTSUPP;
3368 		goto err_unlock;
3369 	}
3370 
3371 	tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3372 	if (tp->ops->hw_del)
3373 		tp->ops->hw_del(tp, type_data);
3374 
3375 	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3376 	if (ok_count < 0)
3377 		goto err_unlock;
3378 
3379 	if (tp->ops->hw_add)
3380 		tp->ops->hw_add(tp, type_data);
3381 	if (ok_count > 0)
3382 		tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3383 					  new_flags, ok_count, true);
3384 err_unlock:
3385 	up_read(&block->cb_lock);
3386 	if (take_rtnl)
3387 		rtnl_unlock();
3388 	return ok_count < 0 ? ok_count : 0;
3389 }
3390 EXPORT_SYMBOL(tc_setup_cb_replace);
3391 
3392 /* Destroy filter and decrement block offload counter, if filter was previously
3393  * offloaded.
3394  */
3395 
3396 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3397 			enum tc_setup_type type, void *type_data, bool err_stop,
3398 			u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3399 {
3400 	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3401 	int ok_count;
3402 
3403 retry:
3404 	if (take_rtnl)
3405 		rtnl_lock();
3406 	down_read(&block->cb_lock);
3407 	/* Need to obtain rtnl lock if block is bound to devs that require it.
3408 	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3409 	 * obtain the locks in same order here.
3410 	 */
3411 	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3412 		up_read(&block->cb_lock);
3413 		take_rtnl = true;
3414 		goto retry;
3415 	}
3416 
3417 	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3418 
3419 	tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3420 	if (tp->ops->hw_del)
3421 		tp->ops->hw_del(tp, type_data);
3422 
3423 	up_read(&block->cb_lock);
3424 	if (take_rtnl)
3425 		rtnl_unlock();
3426 	return ok_count < 0 ? ok_count : 0;
3427 }
3428 EXPORT_SYMBOL(tc_setup_cb_destroy);
3429 
3430 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3431 			  bool add, flow_setup_cb_t *cb,
3432 			  enum tc_setup_type type, void *type_data,
3433 			  void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3434 {
3435 	int err = cb(type, type_data, cb_priv);
3436 
3437 	if (err) {
3438 		if (add && tc_skip_sw(*flags))
3439 			return err;
3440 	} else {
3441 		tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3442 					  add);
3443 	}
3444 
3445 	return 0;
3446 }
3447 EXPORT_SYMBOL(tc_setup_cb_reoffload);
3448 
3449 static int tcf_act_get_cookie(struct flow_action_entry *entry,
3450 			      const struct tc_action *act)
3451 {
3452 	struct tc_cookie *cookie;
3453 	int err = 0;
3454 
3455 	rcu_read_lock();
3456 	cookie = rcu_dereference(act->act_cookie);
3457 	if (cookie) {
3458 		entry->cookie = flow_action_cookie_create(cookie->data,
3459 							  cookie->len,
3460 							  GFP_ATOMIC);
3461 		if (!entry->cookie)
3462 			err = -ENOMEM;
3463 	}
3464 	rcu_read_unlock();
3465 	return err;
3466 }
3467 
3468 static void tcf_act_put_cookie(struct flow_action_entry *entry)
3469 {
3470 	flow_action_cookie_destroy(entry->cookie);
3471 }
3472 
3473 void tc_cleanup_flow_action(struct flow_action *flow_action)
3474 {
3475 	struct flow_action_entry *entry;
3476 	int i;
3477 
3478 	flow_action_for_each(i, entry, flow_action) {
3479 		tcf_act_put_cookie(entry);
3480 		if (entry->destructor)
3481 			entry->destructor(entry->destructor_priv);
3482 	}
3483 }
3484 EXPORT_SYMBOL(tc_cleanup_flow_action);
3485 
3486 static void tcf_mirred_get_dev(struct flow_action_entry *entry,
3487 			       const struct tc_action *act)
3488 {
3489 #ifdef CONFIG_NET_CLS_ACT
3490 	entry->dev = act->ops->get_dev(act, &entry->destructor);
3491 	if (!entry->dev)
3492 		return;
3493 	entry->destructor_priv = entry->dev;
3494 #endif
3495 }
3496 
3497 static void tcf_tunnel_encap_put_tunnel(void *priv)
3498 {
3499 	struct ip_tunnel_info *tunnel = priv;
3500 
3501 	kfree(tunnel);
3502 }
3503 
3504 static int tcf_tunnel_encap_get_tunnel(struct flow_action_entry *entry,
3505 				       const struct tc_action *act)
3506 {
3507 	entry->tunnel = tcf_tunnel_info_copy(act);
3508 	if (!entry->tunnel)
3509 		return -ENOMEM;
3510 	entry->destructor = tcf_tunnel_encap_put_tunnel;
3511 	entry->destructor_priv = entry->tunnel;
3512 	return 0;
3513 }
3514 
3515 static void tcf_sample_get_group(struct flow_action_entry *entry,
3516 				 const struct tc_action *act)
3517 {
3518 #ifdef CONFIG_NET_CLS_ACT
3519 	entry->sample.psample_group =
3520 		act->ops->get_psample_group(act, &entry->destructor);
3521 	entry->destructor_priv = entry->sample.psample_group;
3522 #endif
3523 }
3524 
3525 int tc_setup_flow_action(struct flow_action *flow_action,
3526 			 const struct tcf_exts *exts)
3527 {
3528 	struct tc_action *act;
3529 	int i, j, k, err = 0;
3530 
3531 	BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3532 	BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3533 	BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3534 
3535 	if (!exts)
3536 		return 0;
3537 
3538 	j = 0;
3539 	tcf_exts_for_each_action(i, act, exts) {
3540 		struct flow_action_entry *entry;
3541 
3542 		entry = &flow_action->entries[j];
3543 		spin_lock_bh(&act->tcfa_lock);
3544 		err = tcf_act_get_cookie(entry, act);
3545 		if (err)
3546 			goto err_out_locked;
3547 
3548 		entry->hw_stats = act->hw_stats;
3549 
3550 		if (is_tcf_gact_ok(act)) {
3551 			entry->id = FLOW_ACTION_ACCEPT;
3552 		} else if (is_tcf_gact_shot(act)) {
3553 			entry->id = FLOW_ACTION_DROP;
3554 		} else if (is_tcf_gact_trap(act)) {
3555 			entry->id = FLOW_ACTION_TRAP;
3556 		} else if (is_tcf_gact_goto_chain(act)) {
3557 			entry->id = FLOW_ACTION_GOTO;
3558 			entry->chain_index = tcf_gact_goto_chain_index(act);
3559 		} else if (is_tcf_mirred_egress_redirect(act)) {
3560 			entry->id = FLOW_ACTION_REDIRECT;
3561 			tcf_mirred_get_dev(entry, act);
3562 		} else if (is_tcf_mirred_egress_mirror(act)) {
3563 			entry->id = FLOW_ACTION_MIRRED;
3564 			tcf_mirred_get_dev(entry, act);
3565 		} else if (is_tcf_mirred_ingress_redirect(act)) {
3566 			entry->id = FLOW_ACTION_REDIRECT_INGRESS;
3567 			tcf_mirred_get_dev(entry, act);
3568 		} else if (is_tcf_mirred_ingress_mirror(act)) {
3569 			entry->id = FLOW_ACTION_MIRRED_INGRESS;
3570 			tcf_mirred_get_dev(entry, act);
3571 		} else if (is_tcf_vlan(act)) {
3572 			switch (tcf_vlan_action(act)) {
3573 			case TCA_VLAN_ACT_PUSH:
3574 				entry->id = FLOW_ACTION_VLAN_PUSH;
3575 				entry->vlan.vid = tcf_vlan_push_vid(act);
3576 				entry->vlan.proto = tcf_vlan_push_proto(act);
3577 				entry->vlan.prio = tcf_vlan_push_prio(act);
3578 				break;
3579 			case TCA_VLAN_ACT_POP:
3580 				entry->id = FLOW_ACTION_VLAN_POP;
3581 				break;
3582 			case TCA_VLAN_ACT_MODIFY:
3583 				entry->id = FLOW_ACTION_VLAN_MANGLE;
3584 				entry->vlan.vid = tcf_vlan_push_vid(act);
3585 				entry->vlan.proto = tcf_vlan_push_proto(act);
3586 				entry->vlan.prio = tcf_vlan_push_prio(act);
3587 				break;
3588 			default:
3589 				err = -EOPNOTSUPP;
3590 				goto err_out_locked;
3591 			}
3592 		} else if (is_tcf_tunnel_set(act)) {
3593 			entry->id = FLOW_ACTION_TUNNEL_ENCAP;
3594 			err = tcf_tunnel_encap_get_tunnel(entry, act);
3595 			if (err)
3596 				goto err_out_locked;
3597 		} else if (is_tcf_tunnel_release(act)) {
3598 			entry->id = FLOW_ACTION_TUNNEL_DECAP;
3599 		} else if (is_tcf_pedit(act)) {
3600 			for (k = 0; k < tcf_pedit_nkeys(act); k++) {
3601 				switch (tcf_pedit_cmd(act, k)) {
3602 				case TCA_PEDIT_KEY_EX_CMD_SET:
3603 					entry->id = FLOW_ACTION_MANGLE;
3604 					break;
3605 				case TCA_PEDIT_KEY_EX_CMD_ADD:
3606 					entry->id = FLOW_ACTION_ADD;
3607 					break;
3608 				default:
3609 					err = -EOPNOTSUPP;
3610 					goto err_out_locked;
3611 				}
3612 				entry->mangle.htype = tcf_pedit_htype(act, k);
3613 				entry->mangle.mask = tcf_pedit_mask(act, k);
3614 				entry->mangle.val = tcf_pedit_val(act, k);
3615 				entry->mangle.offset = tcf_pedit_offset(act, k);
3616 				entry->hw_stats = act->hw_stats;
3617 				entry = &flow_action->entries[++j];
3618 			}
3619 		} else if (is_tcf_csum(act)) {
3620 			entry->id = FLOW_ACTION_CSUM;
3621 			entry->csum_flags = tcf_csum_update_flags(act);
3622 		} else if (is_tcf_skbedit_mark(act)) {
3623 			entry->id = FLOW_ACTION_MARK;
3624 			entry->mark = tcf_skbedit_mark(act);
3625 		} else if (is_tcf_sample(act)) {
3626 			entry->id = FLOW_ACTION_SAMPLE;
3627 			entry->sample.trunc_size = tcf_sample_trunc_size(act);
3628 			entry->sample.truncate = tcf_sample_truncate(act);
3629 			entry->sample.rate = tcf_sample_rate(act);
3630 			tcf_sample_get_group(entry, act);
3631 		} else if (is_tcf_police(act)) {
3632 			entry->id = FLOW_ACTION_POLICE;
3633 			entry->police.burst = tcf_police_tcfp_burst(act);
3634 			entry->police.rate_bytes_ps =
3635 				tcf_police_rate_bytes_ps(act);
3636 		} else if (is_tcf_ct(act)) {
3637 			entry->id = FLOW_ACTION_CT;
3638 			entry->ct.action = tcf_ct_action(act);
3639 			entry->ct.zone = tcf_ct_zone(act);
3640 			entry->ct.flow_table = tcf_ct_ft(act);
3641 		} else if (is_tcf_mpls(act)) {
3642 			switch (tcf_mpls_action(act)) {
3643 			case TCA_MPLS_ACT_PUSH:
3644 				entry->id = FLOW_ACTION_MPLS_PUSH;
3645 				entry->mpls_push.proto = tcf_mpls_proto(act);
3646 				entry->mpls_push.label = tcf_mpls_label(act);
3647 				entry->mpls_push.tc = tcf_mpls_tc(act);
3648 				entry->mpls_push.bos = tcf_mpls_bos(act);
3649 				entry->mpls_push.ttl = tcf_mpls_ttl(act);
3650 				break;
3651 			case TCA_MPLS_ACT_POP:
3652 				entry->id = FLOW_ACTION_MPLS_POP;
3653 				entry->mpls_pop.proto = tcf_mpls_proto(act);
3654 				break;
3655 			case TCA_MPLS_ACT_MODIFY:
3656 				entry->id = FLOW_ACTION_MPLS_MANGLE;
3657 				entry->mpls_mangle.label = tcf_mpls_label(act);
3658 				entry->mpls_mangle.tc = tcf_mpls_tc(act);
3659 				entry->mpls_mangle.bos = tcf_mpls_bos(act);
3660 				entry->mpls_mangle.ttl = tcf_mpls_ttl(act);
3661 				break;
3662 			default:
3663 				goto err_out_locked;
3664 			}
3665 		} else if (is_tcf_skbedit_ptype(act)) {
3666 			entry->id = FLOW_ACTION_PTYPE;
3667 			entry->ptype = tcf_skbedit_ptype(act);
3668 		} else if (is_tcf_skbedit_priority(act)) {
3669 			entry->id = FLOW_ACTION_PRIORITY;
3670 			entry->priority = tcf_skbedit_priority(act);
3671 		} else {
3672 			err = -EOPNOTSUPP;
3673 			goto err_out_locked;
3674 		}
3675 		spin_unlock_bh(&act->tcfa_lock);
3676 
3677 		if (!is_tcf_pedit(act))
3678 			j++;
3679 	}
3680 
3681 err_out:
3682 	if (err)
3683 		tc_cleanup_flow_action(flow_action);
3684 
3685 	return err;
3686 err_out_locked:
3687 	spin_unlock_bh(&act->tcfa_lock);
3688 	goto err_out;
3689 }
3690 EXPORT_SYMBOL(tc_setup_flow_action);
3691 
3692 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3693 {
3694 	unsigned int num_acts = 0;
3695 	struct tc_action *act;
3696 	int i;
3697 
3698 	tcf_exts_for_each_action(i, act, exts) {
3699 		if (is_tcf_pedit(act))
3700 			num_acts += tcf_pedit_nkeys(act);
3701 		else
3702 			num_acts++;
3703 	}
3704 	return num_acts;
3705 }
3706 EXPORT_SYMBOL(tcf_exts_num_actions);
3707 
3708 static __net_init int tcf_net_init(struct net *net)
3709 {
3710 	struct tcf_net *tn = net_generic(net, tcf_net_id);
3711 
3712 	spin_lock_init(&tn->idr_lock);
3713 	idr_init(&tn->idr);
3714 	return 0;
3715 }
3716 
3717 static void __net_exit tcf_net_exit(struct net *net)
3718 {
3719 	struct tcf_net *tn = net_generic(net, tcf_net_id);
3720 
3721 	idr_destroy(&tn->idr);
3722 }
3723 
3724 static struct pernet_operations tcf_net_ops = {
3725 	.init = tcf_net_init,
3726 	.exit = tcf_net_exit,
3727 	.id   = &tcf_net_id,
3728 	.size = sizeof(struct tcf_net),
3729 };
3730 
3731 static struct flow_indr_block_entry block_entry = {
3732 	.cb = tc_indr_block_get_and_cmd,
3733 	.list = LIST_HEAD_INIT(block_entry.list),
3734 };
3735 
3736 static int __init tc_filter_init(void)
3737 {
3738 	int err;
3739 
3740 	tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3741 	if (!tc_filter_wq)
3742 		return -ENOMEM;
3743 
3744 	err = register_pernet_subsys(&tcf_net_ops);
3745 	if (err)
3746 		goto err_register_pernet_subsys;
3747 
3748 	flow_indr_add_block_cb(&block_entry);
3749 
3750 	rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3751 		      RTNL_FLAG_DOIT_UNLOCKED);
3752 	rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3753 		      RTNL_FLAG_DOIT_UNLOCKED);
3754 	rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3755 		      tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3756 	rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3757 	rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3758 	rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3759 		      tc_dump_chain, 0);
3760 
3761 	return 0;
3762 
3763 err_register_pernet_subsys:
3764 	destroy_workqueue(tc_filter_wq);
3765 	return err;
3766 }
3767 
3768 subsys_initcall(tc_filter_init);
3769