xref: /openbmc/linux/net/sched/cls_api.c (revision adb57164)
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 		last_executed_chain = fchain->index;
1671 	}
1672 
1673 	ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode,
1674 			     &last_executed_chain);
1675 
1676 	/* If we missed on some chain */
1677 	if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1678 		ext = skb_ext_add(skb, TC_SKB_EXT);
1679 		if (WARN_ON_ONCE(!ext))
1680 			return TC_ACT_SHOT;
1681 		ext->chain = last_executed_chain;
1682 	}
1683 
1684 	return ret;
1685 #endif
1686 }
1687 EXPORT_SYMBOL(tcf_classify_ingress);
1688 
1689 struct tcf_chain_info {
1690 	struct tcf_proto __rcu **pprev;
1691 	struct tcf_proto __rcu *next;
1692 };
1693 
1694 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1695 					   struct tcf_chain_info *chain_info)
1696 {
1697 	return tcf_chain_dereference(*chain_info->pprev, chain);
1698 }
1699 
1700 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1701 			       struct tcf_chain_info *chain_info,
1702 			       struct tcf_proto *tp)
1703 {
1704 	if (chain->flushing)
1705 		return -EAGAIN;
1706 
1707 	if (*chain_info->pprev == chain->filter_chain)
1708 		tcf_chain0_head_change(chain, tp);
1709 	tcf_proto_get(tp);
1710 	RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1711 	rcu_assign_pointer(*chain_info->pprev, tp);
1712 
1713 	return 0;
1714 }
1715 
1716 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1717 				struct tcf_chain_info *chain_info,
1718 				struct tcf_proto *tp)
1719 {
1720 	struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1721 
1722 	tcf_proto_mark_delete(tp);
1723 	if (tp == chain->filter_chain)
1724 		tcf_chain0_head_change(chain, next);
1725 	RCU_INIT_POINTER(*chain_info->pprev, next);
1726 }
1727 
1728 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1729 					   struct tcf_chain_info *chain_info,
1730 					   u32 protocol, u32 prio,
1731 					   bool prio_allocate);
1732 
1733 /* Try to insert new proto.
1734  * If proto with specified priority already exists, free new proto
1735  * and return existing one.
1736  */
1737 
1738 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1739 						    struct tcf_proto *tp_new,
1740 						    u32 protocol, u32 prio,
1741 						    bool rtnl_held)
1742 {
1743 	struct tcf_chain_info chain_info;
1744 	struct tcf_proto *tp;
1745 	int err = 0;
1746 
1747 	mutex_lock(&chain->filter_chain_lock);
1748 
1749 	if (tcf_proto_exists_destroying(chain, tp_new)) {
1750 		mutex_unlock(&chain->filter_chain_lock);
1751 		tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1752 		return ERR_PTR(-EAGAIN);
1753 	}
1754 
1755 	tp = tcf_chain_tp_find(chain, &chain_info,
1756 			       protocol, prio, false);
1757 	if (!tp)
1758 		err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1759 	mutex_unlock(&chain->filter_chain_lock);
1760 
1761 	if (tp) {
1762 		tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1763 		tp_new = tp;
1764 	} else if (err) {
1765 		tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1766 		tp_new = ERR_PTR(err);
1767 	}
1768 
1769 	return tp_new;
1770 }
1771 
1772 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1773 				      struct tcf_proto *tp, bool rtnl_held,
1774 				      struct netlink_ext_ack *extack)
1775 {
1776 	struct tcf_chain_info chain_info;
1777 	struct tcf_proto *tp_iter;
1778 	struct tcf_proto **pprev;
1779 	struct tcf_proto *next;
1780 
1781 	mutex_lock(&chain->filter_chain_lock);
1782 
1783 	/* Atomically find and remove tp from chain. */
1784 	for (pprev = &chain->filter_chain;
1785 	     (tp_iter = tcf_chain_dereference(*pprev, chain));
1786 	     pprev = &tp_iter->next) {
1787 		if (tp_iter == tp) {
1788 			chain_info.pprev = pprev;
1789 			chain_info.next = tp_iter->next;
1790 			WARN_ON(tp_iter->deleting);
1791 			break;
1792 		}
1793 	}
1794 	/* Verify that tp still exists and no new filters were inserted
1795 	 * concurrently.
1796 	 * Mark tp for deletion if it is empty.
1797 	 */
1798 	if (!tp_iter || !tcf_proto_check_delete(tp)) {
1799 		mutex_unlock(&chain->filter_chain_lock);
1800 		return;
1801 	}
1802 
1803 	tcf_proto_signal_destroying(chain, tp);
1804 	next = tcf_chain_dereference(chain_info.next, chain);
1805 	if (tp == chain->filter_chain)
1806 		tcf_chain0_head_change(chain, next);
1807 	RCU_INIT_POINTER(*chain_info.pprev, next);
1808 	mutex_unlock(&chain->filter_chain_lock);
1809 
1810 	tcf_proto_put(tp, rtnl_held, extack);
1811 }
1812 
1813 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1814 					   struct tcf_chain_info *chain_info,
1815 					   u32 protocol, u32 prio,
1816 					   bool prio_allocate)
1817 {
1818 	struct tcf_proto **pprev;
1819 	struct tcf_proto *tp;
1820 
1821 	/* Check the chain for existence of proto-tcf with this priority */
1822 	for (pprev = &chain->filter_chain;
1823 	     (tp = tcf_chain_dereference(*pprev, chain));
1824 	     pprev = &tp->next) {
1825 		if (tp->prio >= prio) {
1826 			if (tp->prio == prio) {
1827 				if (prio_allocate ||
1828 				    (tp->protocol != protocol && protocol))
1829 					return ERR_PTR(-EINVAL);
1830 			} else {
1831 				tp = NULL;
1832 			}
1833 			break;
1834 		}
1835 	}
1836 	chain_info->pprev = pprev;
1837 	if (tp) {
1838 		chain_info->next = tp->next;
1839 		tcf_proto_get(tp);
1840 	} else {
1841 		chain_info->next = NULL;
1842 	}
1843 	return tp;
1844 }
1845 
1846 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1847 			 struct tcf_proto *tp, struct tcf_block *block,
1848 			 struct Qdisc *q, u32 parent, void *fh,
1849 			 u32 portid, u32 seq, u16 flags, int event,
1850 			 bool rtnl_held)
1851 {
1852 	struct tcmsg *tcm;
1853 	struct nlmsghdr  *nlh;
1854 	unsigned char *b = skb_tail_pointer(skb);
1855 
1856 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1857 	if (!nlh)
1858 		goto out_nlmsg_trim;
1859 	tcm = nlmsg_data(nlh);
1860 	tcm->tcm_family = AF_UNSPEC;
1861 	tcm->tcm__pad1 = 0;
1862 	tcm->tcm__pad2 = 0;
1863 	if (q) {
1864 		tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1865 		tcm->tcm_parent = parent;
1866 	} else {
1867 		tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1868 		tcm->tcm_block_index = block->index;
1869 	}
1870 	tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1871 	if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1872 		goto nla_put_failure;
1873 	if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1874 		goto nla_put_failure;
1875 	if (!fh) {
1876 		tcm->tcm_handle = 0;
1877 	} else {
1878 		if (tp->ops->dump &&
1879 		    tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
1880 			goto nla_put_failure;
1881 	}
1882 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1883 	return skb->len;
1884 
1885 out_nlmsg_trim:
1886 nla_put_failure:
1887 	nlmsg_trim(skb, b);
1888 	return -1;
1889 }
1890 
1891 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1892 			  struct nlmsghdr *n, struct tcf_proto *tp,
1893 			  struct tcf_block *block, struct Qdisc *q,
1894 			  u32 parent, void *fh, int event, bool unicast,
1895 			  bool rtnl_held)
1896 {
1897 	struct sk_buff *skb;
1898 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1899 	int err = 0;
1900 
1901 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1902 	if (!skb)
1903 		return -ENOBUFS;
1904 
1905 	if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1906 			  n->nlmsg_seq, n->nlmsg_flags, event,
1907 			  rtnl_held) <= 0) {
1908 		kfree_skb(skb);
1909 		return -EINVAL;
1910 	}
1911 
1912 	if (unicast)
1913 		err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1914 	else
1915 		err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1916 				     n->nlmsg_flags & NLM_F_ECHO);
1917 
1918 	if (err > 0)
1919 		err = 0;
1920 	return err;
1921 }
1922 
1923 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1924 			      struct nlmsghdr *n, struct tcf_proto *tp,
1925 			      struct tcf_block *block, struct Qdisc *q,
1926 			      u32 parent, void *fh, bool unicast, bool *last,
1927 			      bool rtnl_held, struct netlink_ext_ack *extack)
1928 {
1929 	struct sk_buff *skb;
1930 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1931 	int err;
1932 
1933 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1934 	if (!skb)
1935 		return -ENOBUFS;
1936 
1937 	if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1938 			  n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
1939 			  rtnl_held) <= 0) {
1940 		NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1941 		kfree_skb(skb);
1942 		return -EINVAL;
1943 	}
1944 
1945 	err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
1946 	if (err) {
1947 		kfree_skb(skb);
1948 		return err;
1949 	}
1950 
1951 	if (unicast)
1952 		err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1953 	else
1954 		err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1955 				     n->nlmsg_flags & NLM_F_ECHO);
1956 	if (err < 0)
1957 		NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1958 
1959 	if (err > 0)
1960 		err = 0;
1961 	return err;
1962 }
1963 
1964 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1965 				 struct tcf_block *block, struct Qdisc *q,
1966 				 u32 parent, struct nlmsghdr *n,
1967 				 struct tcf_chain *chain, int event,
1968 				 bool rtnl_held)
1969 {
1970 	struct tcf_proto *tp;
1971 
1972 	for (tp = tcf_get_next_proto(chain, NULL, rtnl_held);
1973 	     tp; tp = tcf_get_next_proto(chain, tp, rtnl_held))
1974 		tfilter_notify(net, oskb, n, tp, block,
1975 			       q, parent, NULL, event, false, rtnl_held);
1976 }
1977 
1978 static void tfilter_put(struct tcf_proto *tp, void *fh)
1979 {
1980 	if (tp->ops->put && fh)
1981 		tp->ops->put(tp, fh);
1982 }
1983 
1984 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1985 			  struct netlink_ext_ack *extack)
1986 {
1987 	struct net *net = sock_net(skb->sk);
1988 	struct nlattr *tca[TCA_MAX + 1];
1989 	char name[IFNAMSIZ];
1990 	struct tcmsg *t;
1991 	u32 protocol;
1992 	u32 prio;
1993 	bool prio_allocate;
1994 	u32 parent;
1995 	u32 chain_index;
1996 	struct Qdisc *q = NULL;
1997 	struct tcf_chain_info chain_info;
1998 	struct tcf_chain *chain = NULL;
1999 	struct tcf_block *block;
2000 	struct tcf_proto *tp;
2001 	unsigned long cl;
2002 	void *fh;
2003 	int err;
2004 	int tp_created;
2005 	bool rtnl_held = false;
2006 
2007 	if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2008 		return -EPERM;
2009 
2010 replay:
2011 	tp_created = 0;
2012 
2013 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2014 				     rtm_tca_policy, extack);
2015 	if (err < 0)
2016 		return err;
2017 
2018 	t = nlmsg_data(n);
2019 	protocol = TC_H_MIN(t->tcm_info);
2020 	prio = TC_H_MAJ(t->tcm_info);
2021 	prio_allocate = false;
2022 	parent = t->tcm_parent;
2023 	tp = NULL;
2024 	cl = 0;
2025 	block = NULL;
2026 
2027 	if (prio == 0) {
2028 		/* If no priority is provided by the user,
2029 		 * we allocate one.
2030 		 */
2031 		if (n->nlmsg_flags & NLM_F_CREATE) {
2032 			prio = TC_H_MAKE(0x80000000U, 0U);
2033 			prio_allocate = true;
2034 		} else {
2035 			NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2036 			return -ENOENT;
2037 		}
2038 	}
2039 
2040 	/* Find head of filter chain. */
2041 
2042 	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2043 	if (err)
2044 		return err;
2045 
2046 	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2047 		NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2048 		err = -EINVAL;
2049 		goto errout;
2050 	}
2051 
2052 	/* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2053 	 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2054 	 * type is not specified, classifier is not unlocked.
2055 	 */
2056 	if (rtnl_held ||
2057 	    (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2058 	    !tcf_proto_is_unlocked(name)) {
2059 		rtnl_held = true;
2060 		rtnl_lock();
2061 	}
2062 
2063 	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2064 	if (err)
2065 		goto errout;
2066 
2067 	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2068 				 extack);
2069 	if (IS_ERR(block)) {
2070 		err = PTR_ERR(block);
2071 		goto errout;
2072 	}
2073 
2074 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2075 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2076 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2077 		err = -EINVAL;
2078 		goto errout;
2079 	}
2080 	chain = tcf_chain_get(block, chain_index, true);
2081 	if (!chain) {
2082 		NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2083 		err = -ENOMEM;
2084 		goto errout;
2085 	}
2086 
2087 	mutex_lock(&chain->filter_chain_lock);
2088 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2089 			       prio, prio_allocate);
2090 	if (IS_ERR(tp)) {
2091 		NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2092 		err = PTR_ERR(tp);
2093 		goto errout_locked;
2094 	}
2095 
2096 	if (tp == NULL) {
2097 		struct tcf_proto *tp_new = NULL;
2098 
2099 		if (chain->flushing) {
2100 			err = -EAGAIN;
2101 			goto errout_locked;
2102 		}
2103 
2104 		/* Proto-tcf does not exist, create new one */
2105 
2106 		if (tca[TCA_KIND] == NULL || !protocol) {
2107 			NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2108 			err = -EINVAL;
2109 			goto errout_locked;
2110 		}
2111 
2112 		if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2113 			NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2114 			err = -ENOENT;
2115 			goto errout_locked;
2116 		}
2117 
2118 		if (prio_allocate)
2119 			prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2120 							       &chain_info));
2121 
2122 		mutex_unlock(&chain->filter_chain_lock);
2123 		tp_new = tcf_proto_create(name, protocol, prio, chain,
2124 					  rtnl_held, extack);
2125 		if (IS_ERR(tp_new)) {
2126 			err = PTR_ERR(tp_new);
2127 			goto errout_tp;
2128 		}
2129 
2130 		tp_created = 1;
2131 		tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2132 						rtnl_held);
2133 		if (IS_ERR(tp)) {
2134 			err = PTR_ERR(tp);
2135 			goto errout_tp;
2136 		}
2137 	} else {
2138 		mutex_unlock(&chain->filter_chain_lock);
2139 	}
2140 
2141 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2142 		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2143 		err = -EINVAL;
2144 		goto errout;
2145 	}
2146 
2147 	fh = tp->ops->get(tp, t->tcm_handle);
2148 
2149 	if (!fh) {
2150 		if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2151 			NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2152 			err = -ENOENT;
2153 			goto errout;
2154 		}
2155 	} else if (n->nlmsg_flags & NLM_F_EXCL) {
2156 		tfilter_put(tp, fh);
2157 		NL_SET_ERR_MSG(extack, "Filter already exists");
2158 		err = -EEXIST;
2159 		goto errout;
2160 	}
2161 
2162 	if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2163 		NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2164 		err = -EINVAL;
2165 		goto errout;
2166 	}
2167 
2168 	err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2169 			      n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
2170 			      rtnl_held, extack);
2171 	if (err == 0) {
2172 		tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2173 			       RTM_NEWTFILTER, false, rtnl_held);
2174 		tfilter_put(tp, fh);
2175 		/* q pointer is NULL for shared blocks */
2176 		if (q)
2177 			q->flags &= ~TCQ_F_CAN_BYPASS;
2178 	}
2179 
2180 errout:
2181 	if (err && tp_created)
2182 		tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2183 errout_tp:
2184 	if (chain) {
2185 		if (tp && !IS_ERR(tp))
2186 			tcf_proto_put(tp, rtnl_held, NULL);
2187 		if (!tp_created)
2188 			tcf_chain_put(chain);
2189 	}
2190 	tcf_block_release(q, block, rtnl_held);
2191 
2192 	if (rtnl_held)
2193 		rtnl_unlock();
2194 
2195 	if (err == -EAGAIN) {
2196 		/* Take rtnl lock in case EAGAIN is caused by concurrent flush
2197 		 * of target chain.
2198 		 */
2199 		rtnl_held = true;
2200 		/* Replay the request. */
2201 		goto replay;
2202 	}
2203 	return err;
2204 
2205 errout_locked:
2206 	mutex_unlock(&chain->filter_chain_lock);
2207 	goto errout;
2208 }
2209 
2210 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2211 			  struct netlink_ext_ack *extack)
2212 {
2213 	struct net *net = sock_net(skb->sk);
2214 	struct nlattr *tca[TCA_MAX + 1];
2215 	char name[IFNAMSIZ];
2216 	struct tcmsg *t;
2217 	u32 protocol;
2218 	u32 prio;
2219 	u32 parent;
2220 	u32 chain_index;
2221 	struct Qdisc *q = NULL;
2222 	struct tcf_chain_info chain_info;
2223 	struct tcf_chain *chain = NULL;
2224 	struct tcf_block *block = NULL;
2225 	struct tcf_proto *tp = NULL;
2226 	unsigned long cl = 0;
2227 	void *fh = NULL;
2228 	int err;
2229 	bool rtnl_held = false;
2230 
2231 	if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2232 		return -EPERM;
2233 
2234 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2235 				     rtm_tca_policy, extack);
2236 	if (err < 0)
2237 		return err;
2238 
2239 	t = nlmsg_data(n);
2240 	protocol = TC_H_MIN(t->tcm_info);
2241 	prio = TC_H_MAJ(t->tcm_info);
2242 	parent = t->tcm_parent;
2243 
2244 	if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2245 		NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2246 		return -ENOENT;
2247 	}
2248 
2249 	/* Find head of filter chain. */
2250 
2251 	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2252 	if (err)
2253 		return err;
2254 
2255 	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2256 		NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2257 		err = -EINVAL;
2258 		goto errout;
2259 	}
2260 	/* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2261 	 * found), qdisc is not unlocked, classifier type is not specified,
2262 	 * classifier is not unlocked.
2263 	 */
2264 	if (!prio ||
2265 	    (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2266 	    !tcf_proto_is_unlocked(name)) {
2267 		rtnl_held = true;
2268 		rtnl_lock();
2269 	}
2270 
2271 	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2272 	if (err)
2273 		goto errout;
2274 
2275 	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2276 				 extack);
2277 	if (IS_ERR(block)) {
2278 		err = PTR_ERR(block);
2279 		goto errout;
2280 	}
2281 
2282 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2283 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2284 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2285 		err = -EINVAL;
2286 		goto errout;
2287 	}
2288 	chain = tcf_chain_get(block, chain_index, false);
2289 	if (!chain) {
2290 		/* User requested flush on non-existent chain. Nothing to do,
2291 		 * so just return success.
2292 		 */
2293 		if (prio == 0) {
2294 			err = 0;
2295 			goto errout;
2296 		}
2297 		NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2298 		err = -ENOENT;
2299 		goto errout;
2300 	}
2301 
2302 	if (prio == 0) {
2303 		tfilter_notify_chain(net, skb, block, q, parent, n,
2304 				     chain, RTM_DELTFILTER, rtnl_held);
2305 		tcf_chain_flush(chain, rtnl_held);
2306 		err = 0;
2307 		goto errout;
2308 	}
2309 
2310 	mutex_lock(&chain->filter_chain_lock);
2311 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2312 			       prio, false);
2313 	if (!tp || IS_ERR(tp)) {
2314 		NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2315 		err = tp ? PTR_ERR(tp) : -ENOENT;
2316 		goto errout_locked;
2317 	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2318 		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2319 		err = -EINVAL;
2320 		goto errout_locked;
2321 	} else if (t->tcm_handle == 0) {
2322 		tcf_proto_signal_destroying(chain, tp);
2323 		tcf_chain_tp_remove(chain, &chain_info, tp);
2324 		mutex_unlock(&chain->filter_chain_lock);
2325 
2326 		tcf_proto_put(tp, rtnl_held, NULL);
2327 		tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2328 			       RTM_DELTFILTER, false, rtnl_held);
2329 		err = 0;
2330 		goto errout;
2331 	}
2332 	mutex_unlock(&chain->filter_chain_lock);
2333 
2334 	fh = tp->ops->get(tp, t->tcm_handle);
2335 
2336 	if (!fh) {
2337 		NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2338 		err = -ENOENT;
2339 	} else {
2340 		bool last;
2341 
2342 		err = tfilter_del_notify(net, skb, n, tp, block,
2343 					 q, parent, fh, false, &last,
2344 					 rtnl_held, extack);
2345 
2346 		if (err)
2347 			goto errout;
2348 		if (last)
2349 			tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2350 	}
2351 
2352 errout:
2353 	if (chain) {
2354 		if (tp && !IS_ERR(tp))
2355 			tcf_proto_put(tp, rtnl_held, NULL);
2356 		tcf_chain_put(chain);
2357 	}
2358 	tcf_block_release(q, block, rtnl_held);
2359 
2360 	if (rtnl_held)
2361 		rtnl_unlock();
2362 
2363 	return err;
2364 
2365 errout_locked:
2366 	mutex_unlock(&chain->filter_chain_lock);
2367 	goto errout;
2368 }
2369 
2370 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2371 			  struct netlink_ext_ack *extack)
2372 {
2373 	struct net *net = sock_net(skb->sk);
2374 	struct nlattr *tca[TCA_MAX + 1];
2375 	char name[IFNAMSIZ];
2376 	struct tcmsg *t;
2377 	u32 protocol;
2378 	u32 prio;
2379 	u32 parent;
2380 	u32 chain_index;
2381 	struct Qdisc *q = NULL;
2382 	struct tcf_chain_info chain_info;
2383 	struct tcf_chain *chain = NULL;
2384 	struct tcf_block *block = NULL;
2385 	struct tcf_proto *tp = NULL;
2386 	unsigned long cl = 0;
2387 	void *fh = NULL;
2388 	int err;
2389 	bool rtnl_held = false;
2390 
2391 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2392 				     rtm_tca_policy, extack);
2393 	if (err < 0)
2394 		return err;
2395 
2396 	t = nlmsg_data(n);
2397 	protocol = TC_H_MIN(t->tcm_info);
2398 	prio = TC_H_MAJ(t->tcm_info);
2399 	parent = t->tcm_parent;
2400 
2401 	if (prio == 0) {
2402 		NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2403 		return -ENOENT;
2404 	}
2405 
2406 	/* Find head of filter chain. */
2407 
2408 	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2409 	if (err)
2410 		return err;
2411 
2412 	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2413 		NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2414 		err = -EINVAL;
2415 		goto errout;
2416 	}
2417 	/* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2418 	 * unlocked, classifier type is not specified, classifier is not
2419 	 * unlocked.
2420 	 */
2421 	if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2422 	    !tcf_proto_is_unlocked(name)) {
2423 		rtnl_held = true;
2424 		rtnl_lock();
2425 	}
2426 
2427 	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2428 	if (err)
2429 		goto errout;
2430 
2431 	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2432 				 extack);
2433 	if (IS_ERR(block)) {
2434 		err = PTR_ERR(block);
2435 		goto errout;
2436 	}
2437 
2438 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2439 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2440 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2441 		err = -EINVAL;
2442 		goto errout;
2443 	}
2444 	chain = tcf_chain_get(block, chain_index, false);
2445 	if (!chain) {
2446 		NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2447 		err = -EINVAL;
2448 		goto errout;
2449 	}
2450 
2451 	mutex_lock(&chain->filter_chain_lock);
2452 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2453 			       prio, false);
2454 	mutex_unlock(&chain->filter_chain_lock);
2455 	if (!tp || IS_ERR(tp)) {
2456 		NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2457 		err = tp ? PTR_ERR(tp) : -ENOENT;
2458 		goto errout;
2459 	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2460 		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2461 		err = -EINVAL;
2462 		goto errout;
2463 	}
2464 
2465 	fh = tp->ops->get(tp, t->tcm_handle);
2466 
2467 	if (!fh) {
2468 		NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2469 		err = -ENOENT;
2470 	} else {
2471 		err = tfilter_notify(net, skb, n, tp, block, q, parent,
2472 				     fh, RTM_NEWTFILTER, true, rtnl_held);
2473 		if (err < 0)
2474 			NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2475 	}
2476 
2477 	tfilter_put(tp, fh);
2478 errout:
2479 	if (chain) {
2480 		if (tp && !IS_ERR(tp))
2481 			tcf_proto_put(tp, rtnl_held, NULL);
2482 		tcf_chain_put(chain);
2483 	}
2484 	tcf_block_release(q, block, rtnl_held);
2485 
2486 	if (rtnl_held)
2487 		rtnl_unlock();
2488 
2489 	return err;
2490 }
2491 
2492 struct tcf_dump_args {
2493 	struct tcf_walker w;
2494 	struct sk_buff *skb;
2495 	struct netlink_callback *cb;
2496 	struct tcf_block *block;
2497 	struct Qdisc *q;
2498 	u32 parent;
2499 };
2500 
2501 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2502 {
2503 	struct tcf_dump_args *a = (void *)arg;
2504 	struct net *net = sock_net(a->skb->sk);
2505 
2506 	return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2507 			     n, NETLINK_CB(a->cb->skb).portid,
2508 			     a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2509 			     RTM_NEWTFILTER, true);
2510 }
2511 
2512 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2513 			   struct sk_buff *skb, struct netlink_callback *cb,
2514 			   long index_start, long *p_index)
2515 {
2516 	struct net *net = sock_net(skb->sk);
2517 	struct tcf_block *block = chain->block;
2518 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2519 	struct tcf_proto *tp, *tp_prev;
2520 	struct tcf_dump_args arg;
2521 
2522 	for (tp = __tcf_get_next_proto(chain, NULL);
2523 	     tp;
2524 	     tp_prev = tp,
2525 		     tp = __tcf_get_next_proto(chain, tp),
2526 		     tcf_proto_put(tp_prev, true, NULL),
2527 		     (*p_index)++) {
2528 		if (*p_index < index_start)
2529 			continue;
2530 		if (TC_H_MAJ(tcm->tcm_info) &&
2531 		    TC_H_MAJ(tcm->tcm_info) != tp->prio)
2532 			continue;
2533 		if (TC_H_MIN(tcm->tcm_info) &&
2534 		    TC_H_MIN(tcm->tcm_info) != tp->protocol)
2535 			continue;
2536 		if (*p_index > index_start)
2537 			memset(&cb->args[1], 0,
2538 			       sizeof(cb->args) - sizeof(cb->args[0]));
2539 		if (cb->args[1] == 0) {
2540 			if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2541 					  NETLINK_CB(cb->skb).portid,
2542 					  cb->nlh->nlmsg_seq, NLM_F_MULTI,
2543 					  RTM_NEWTFILTER, true) <= 0)
2544 				goto errout;
2545 			cb->args[1] = 1;
2546 		}
2547 		if (!tp->ops->walk)
2548 			continue;
2549 		arg.w.fn = tcf_node_dump;
2550 		arg.skb = skb;
2551 		arg.cb = cb;
2552 		arg.block = block;
2553 		arg.q = q;
2554 		arg.parent = parent;
2555 		arg.w.stop = 0;
2556 		arg.w.skip = cb->args[1] - 1;
2557 		arg.w.count = 0;
2558 		arg.w.cookie = cb->args[2];
2559 		tp->ops->walk(tp, &arg.w, true);
2560 		cb->args[2] = arg.w.cookie;
2561 		cb->args[1] = arg.w.count + 1;
2562 		if (arg.w.stop)
2563 			goto errout;
2564 	}
2565 	return true;
2566 
2567 errout:
2568 	tcf_proto_put(tp, true, NULL);
2569 	return false;
2570 }
2571 
2572 /* called with RTNL */
2573 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2574 {
2575 	struct tcf_chain *chain, *chain_prev;
2576 	struct net *net = sock_net(skb->sk);
2577 	struct nlattr *tca[TCA_MAX + 1];
2578 	struct Qdisc *q = NULL;
2579 	struct tcf_block *block;
2580 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2581 	long index_start;
2582 	long index;
2583 	u32 parent;
2584 	int err;
2585 
2586 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2587 		return skb->len;
2588 
2589 	err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2590 				     NULL, cb->extack);
2591 	if (err)
2592 		return err;
2593 
2594 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2595 		block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2596 		if (!block)
2597 			goto out;
2598 		/* If we work with block index, q is NULL and parent value
2599 		 * will never be used in the following code. The check
2600 		 * in tcf_fill_node prevents it. However, compiler does not
2601 		 * see that far, so set parent to zero to silence the warning
2602 		 * about parent being uninitialized.
2603 		 */
2604 		parent = 0;
2605 	} else {
2606 		const struct Qdisc_class_ops *cops;
2607 		struct net_device *dev;
2608 		unsigned long cl = 0;
2609 
2610 		dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2611 		if (!dev)
2612 			return skb->len;
2613 
2614 		parent = tcm->tcm_parent;
2615 		if (!parent) {
2616 			q = dev->qdisc;
2617 			parent = q->handle;
2618 		} else {
2619 			q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2620 		}
2621 		if (!q)
2622 			goto out;
2623 		cops = q->ops->cl_ops;
2624 		if (!cops)
2625 			goto out;
2626 		if (!cops->tcf_block)
2627 			goto out;
2628 		if (TC_H_MIN(tcm->tcm_parent)) {
2629 			cl = cops->find(q, tcm->tcm_parent);
2630 			if (cl == 0)
2631 				goto out;
2632 		}
2633 		block = cops->tcf_block(q, cl, NULL);
2634 		if (!block)
2635 			goto out;
2636 		if (tcf_block_shared(block))
2637 			q = NULL;
2638 	}
2639 
2640 	index_start = cb->args[0];
2641 	index = 0;
2642 
2643 	for (chain = __tcf_get_next_chain(block, NULL);
2644 	     chain;
2645 	     chain_prev = chain,
2646 		     chain = __tcf_get_next_chain(block, chain),
2647 		     tcf_chain_put(chain_prev)) {
2648 		if (tca[TCA_CHAIN] &&
2649 		    nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2650 			continue;
2651 		if (!tcf_chain_dump(chain, q, parent, skb, cb,
2652 				    index_start, &index)) {
2653 			tcf_chain_put(chain);
2654 			err = -EMSGSIZE;
2655 			break;
2656 		}
2657 	}
2658 
2659 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2660 		tcf_block_refcnt_put(block, true);
2661 	cb->args[0] = index;
2662 
2663 out:
2664 	/* If we did no progress, the error (EMSGSIZE) is real */
2665 	if (skb->len == 0 && err)
2666 		return err;
2667 	return skb->len;
2668 }
2669 
2670 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2671 			      void *tmplt_priv, u32 chain_index,
2672 			      struct net *net, struct sk_buff *skb,
2673 			      struct tcf_block *block,
2674 			      u32 portid, u32 seq, u16 flags, int event)
2675 {
2676 	unsigned char *b = skb_tail_pointer(skb);
2677 	const struct tcf_proto_ops *ops;
2678 	struct nlmsghdr *nlh;
2679 	struct tcmsg *tcm;
2680 	void *priv;
2681 
2682 	ops = tmplt_ops;
2683 	priv = tmplt_priv;
2684 
2685 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2686 	if (!nlh)
2687 		goto out_nlmsg_trim;
2688 	tcm = nlmsg_data(nlh);
2689 	tcm->tcm_family = AF_UNSPEC;
2690 	tcm->tcm__pad1 = 0;
2691 	tcm->tcm__pad2 = 0;
2692 	tcm->tcm_handle = 0;
2693 	if (block->q) {
2694 		tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2695 		tcm->tcm_parent = block->q->handle;
2696 	} else {
2697 		tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2698 		tcm->tcm_block_index = block->index;
2699 	}
2700 
2701 	if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2702 		goto nla_put_failure;
2703 
2704 	if (ops) {
2705 		if (nla_put_string(skb, TCA_KIND, ops->kind))
2706 			goto nla_put_failure;
2707 		if (ops->tmplt_dump(skb, net, priv) < 0)
2708 			goto nla_put_failure;
2709 	}
2710 
2711 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2712 	return skb->len;
2713 
2714 out_nlmsg_trim:
2715 nla_put_failure:
2716 	nlmsg_trim(skb, b);
2717 	return -EMSGSIZE;
2718 }
2719 
2720 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2721 			   u32 seq, u16 flags, int event, bool unicast)
2722 {
2723 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2724 	struct tcf_block *block = chain->block;
2725 	struct net *net = block->net;
2726 	struct sk_buff *skb;
2727 	int err = 0;
2728 
2729 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2730 	if (!skb)
2731 		return -ENOBUFS;
2732 
2733 	if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2734 			       chain->index, net, skb, block, portid,
2735 			       seq, flags, event) <= 0) {
2736 		kfree_skb(skb);
2737 		return -EINVAL;
2738 	}
2739 
2740 	if (unicast)
2741 		err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2742 	else
2743 		err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2744 				     flags & NLM_F_ECHO);
2745 
2746 	if (err > 0)
2747 		err = 0;
2748 	return err;
2749 }
2750 
2751 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2752 				  void *tmplt_priv, u32 chain_index,
2753 				  struct tcf_block *block, struct sk_buff *oskb,
2754 				  u32 seq, u16 flags, bool unicast)
2755 {
2756 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2757 	struct net *net = block->net;
2758 	struct sk_buff *skb;
2759 
2760 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2761 	if (!skb)
2762 		return -ENOBUFS;
2763 
2764 	if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2765 			       block, portid, seq, flags, RTM_DELCHAIN) <= 0) {
2766 		kfree_skb(skb);
2767 		return -EINVAL;
2768 	}
2769 
2770 	if (unicast)
2771 		return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2772 
2773 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2774 }
2775 
2776 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2777 			      struct nlattr **tca,
2778 			      struct netlink_ext_ack *extack)
2779 {
2780 	const struct tcf_proto_ops *ops;
2781 	char name[IFNAMSIZ];
2782 	void *tmplt_priv;
2783 
2784 	/* If kind is not set, user did not specify template. */
2785 	if (!tca[TCA_KIND])
2786 		return 0;
2787 
2788 	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2789 		NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
2790 		return -EINVAL;
2791 	}
2792 
2793 	ops = tcf_proto_lookup_ops(name, true, extack);
2794 	if (IS_ERR(ops))
2795 		return PTR_ERR(ops);
2796 	if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2797 		NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2798 		return -EOPNOTSUPP;
2799 	}
2800 
2801 	tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2802 	if (IS_ERR(tmplt_priv)) {
2803 		module_put(ops->owner);
2804 		return PTR_ERR(tmplt_priv);
2805 	}
2806 	chain->tmplt_ops = ops;
2807 	chain->tmplt_priv = tmplt_priv;
2808 	return 0;
2809 }
2810 
2811 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2812 			       void *tmplt_priv)
2813 {
2814 	/* If template ops are set, no work to do for us. */
2815 	if (!tmplt_ops)
2816 		return;
2817 
2818 	tmplt_ops->tmplt_destroy(tmplt_priv);
2819 	module_put(tmplt_ops->owner);
2820 }
2821 
2822 /* Add/delete/get a chain */
2823 
2824 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2825 			struct netlink_ext_ack *extack)
2826 {
2827 	struct net *net = sock_net(skb->sk);
2828 	struct nlattr *tca[TCA_MAX + 1];
2829 	struct tcmsg *t;
2830 	u32 parent;
2831 	u32 chain_index;
2832 	struct Qdisc *q = NULL;
2833 	struct tcf_chain *chain = NULL;
2834 	struct tcf_block *block;
2835 	unsigned long cl;
2836 	int err;
2837 
2838 	if (n->nlmsg_type != RTM_GETCHAIN &&
2839 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2840 		return -EPERM;
2841 
2842 replay:
2843 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2844 				     rtm_tca_policy, extack);
2845 	if (err < 0)
2846 		return err;
2847 
2848 	t = nlmsg_data(n);
2849 	parent = t->tcm_parent;
2850 	cl = 0;
2851 
2852 	block = tcf_block_find(net, &q, &parent, &cl,
2853 			       t->tcm_ifindex, t->tcm_block_index, extack);
2854 	if (IS_ERR(block))
2855 		return PTR_ERR(block);
2856 
2857 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2858 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2859 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2860 		err = -EINVAL;
2861 		goto errout_block;
2862 	}
2863 
2864 	mutex_lock(&block->lock);
2865 	chain = tcf_chain_lookup(block, chain_index);
2866 	if (n->nlmsg_type == RTM_NEWCHAIN) {
2867 		if (chain) {
2868 			if (tcf_chain_held_by_acts_only(chain)) {
2869 				/* The chain exists only because there is
2870 				 * some action referencing it.
2871 				 */
2872 				tcf_chain_hold(chain);
2873 			} else {
2874 				NL_SET_ERR_MSG(extack, "Filter chain already exists");
2875 				err = -EEXIST;
2876 				goto errout_block_locked;
2877 			}
2878 		} else {
2879 			if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2880 				NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
2881 				err = -ENOENT;
2882 				goto errout_block_locked;
2883 			}
2884 			chain = tcf_chain_create(block, chain_index);
2885 			if (!chain) {
2886 				NL_SET_ERR_MSG(extack, "Failed to create filter chain");
2887 				err = -ENOMEM;
2888 				goto errout_block_locked;
2889 			}
2890 		}
2891 	} else {
2892 		if (!chain || tcf_chain_held_by_acts_only(chain)) {
2893 			NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2894 			err = -EINVAL;
2895 			goto errout_block_locked;
2896 		}
2897 		tcf_chain_hold(chain);
2898 	}
2899 
2900 	if (n->nlmsg_type == RTM_NEWCHAIN) {
2901 		/* Modifying chain requires holding parent block lock. In case
2902 		 * the chain was successfully added, take a reference to the
2903 		 * chain. This ensures that an empty chain does not disappear at
2904 		 * the end of this function.
2905 		 */
2906 		tcf_chain_hold(chain);
2907 		chain->explicitly_created = true;
2908 	}
2909 	mutex_unlock(&block->lock);
2910 
2911 	switch (n->nlmsg_type) {
2912 	case RTM_NEWCHAIN:
2913 		err = tc_chain_tmplt_add(chain, net, tca, extack);
2914 		if (err) {
2915 			tcf_chain_put_explicitly_created(chain);
2916 			goto errout;
2917 		}
2918 
2919 		tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2920 				RTM_NEWCHAIN, false);
2921 		break;
2922 	case RTM_DELCHAIN:
2923 		tfilter_notify_chain(net, skb, block, q, parent, n,
2924 				     chain, RTM_DELTFILTER, true);
2925 		/* Flush the chain first as the user requested chain removal. */
2926 		tcf_chain_flush(chain, true);
2927 		/* In case the chain was successfully deleted, put a reference
2928 		 * to the chain previously taken during addition.
2929 		 */
2930 		tcf_chain_put_explicitly_created(chain);
2931 		break;
2932 	case RTM_GETCHAIN:
2933 		err = tc_chain_notify(chain, skb, n->nlmsg_seq,
2934 				      n->nlmsg_seq, n->nlmsg_type, true);
2935 		if (err < 0)
2936 			NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2937 		break;
2938 	default:
2939 		err = -EOPNOTSUPP;
2940 		NL_SET_ERR_MSG(extack, "Unsupported message type");
2941 		goto errout;
2942 	}
2943 
2944 errout:
2945 	tcf_chain_put(chain);
2946 errout_block:
2947 	tcf_block_release(q, block, true);
2948 	if (err == -EAGAIN)
2949 		/* Replay the request. */
2950 		goto replay;
2951 	return err;
2952 
2953 errout_block_locked:
2954 	mutex_unlock(&block->lock);
2955 	goto errout_block;
2956 }
2957 
2958 /* called with RTNL */
2959 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2960 {
2961 	struct net *net = sock_net(skb->sk);
2962 	struct nlattr *tca[TCA_MAX + 1];
2963 	struct Qdisc *q = NULL;
2964 	struct tcf_block *block;
2965 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2966 	struct tcf_chain *chain;
2967 	long index_start;
2968 	long index;
2969 	u32 parent;
2970 	int err;
2971 
2972 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2973 		return skb->len;
2974 
2975 	err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2976 				     rtm_tca_policy, cb->extack);
2977 	if (err)
2978 		return err;
2979 
2980 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2981 		block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2982 		if (!block)
2983 			goto out;
2984 		/* If we work with block index, q is NULL and parent value
2985 		 * will never be used in the following code. The check
2986 		 * in tcf_fill_node prevents it. However, compiler does not
2987 		 * see that far, so set parent to zero to silence the warning
2988 		 * about parent being uninitialized.
2989 		 */
2990 		parent = 0;
2991 	} else {
2992 		const struct Qdisc_class_ops *cops;
2993 		struct net_device *dev;
2994 		unsigned long cl = 0;
2995 
2996 		dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2997 		if (!dev)
2998 			return skb->len;
2999 
3000 		parent = tcm->tcm_parent;
3001 		if (!parent) {
3002 			q = dev->qdisc;
3003 			parent = q->handle;
3004 		} else {
3005 			q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
3006 		}
3007 		if (!q)
3008 			goto out;
3009 		cops = q->ops->cl_ops;
3010 		if (!cops)
3011 			goto out;
3012 		if (!cops->tcf_block)
3013 			goto out;
3014 		if (TC_H_MIN(tcm->tcm_parent)) {
3015 			cl = cops->find(q, tcm->tcm_parent);
3016 			if (cl == 0)
3017 				goto out;
3018 		}
3019 		block = cops->tcf_block(q, cl, NULL);
3020 		if (!block)
3021 			goto out;
3022 		if (tcf_block_shared(block))
3023 			q = NULL;
3024 	}
3025 
3026 	index_start = cb->args[0];
3027 	index = 0;
3028 
3029 	mutex_lock(&block->lock);
3030 	list_for_each_entry(chain, &block->chain_list, list) {
3031 		if ((tca[TCA_CHAIN] &&
3032 		     nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3033 			continue;
3034 		if (index < index_start) {
3035 			index++;
3036 			continue;
3037 		}
3038 		if (tcf_chain_held_by_acts_only(chain))
3039 			continue;
3040 		err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3041 					 chain->index, net, skb, block,
3042 					 NETLINK_CB(cb->skb).portid,
3043 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3044 					 RTM_NEWCHAIN);
3045 		if (err <= 0)
3046 			break;
3047 		index++;
3048 	}
3049 	mutex_unlock(&block->lock);
3050 
3051 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3052 		tcf_block_refcnt_put(block, true);
3053 	cb->args[0] = index;
3054 
3055 out:
3056 	/* If we did no progress, the error (EMSGSIZE) is real */
3057 	if (skb->len == 0 && err)
3058 		return err;
3059 	return skb->len;
3060 }
3061 
3062 void tcf_exts_destroy(struct tcf_exts *exts)
3063 {
3064 #ifdef CONFIG_NET_CLS_ACT
3065 	if (exts->actions) {
3066 		tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3067 		kfree(exts->actions);
3068 	}
3069 	exts->nr_actions = 0;
3070 #endif
3071 }
3072 EXPORT_SYMBOL(tcf_exts_destroy);
3073 
3074 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3075 		      struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
3076 		      bool rtnl_held, struct netlink_ext_ack *extack)
3077 {
3078 #ifdef CONFIG_NET_CLS_ACT
3079 	{
3080 		struct tc_action *act;
3081 		size_t attr_size = 0;
3082 
3083 		if (exts->police && tb[exts->police]) {
3084 			act = tcf_action_init_1(net, tp, tb[exts->police],
3085 						rate_tlv, "police", ovr,
3086 						TCA_ACT_BIND, rtnl_held,
3087 						extack);
3088 			if (IS_ERR(act))
3089 				return PTR_ERR(act);
3090 
3091 			act->type = exts->type = TCA_OLD_COMPAT;
3092 			exts->actions[0] = act;
3093 			exts->nr_actions = 1;
3094 		} else if (exts->action && tb[exts->action]) {
3095 			int err;
3096 
3097 			err = tcf_action_init(net, tp, tb[exts->action],
3098 					      rate_tlv, NULL, ovr, TCA_ACT_BIND,
3099 					      exts->actions, &attr_size,
3100 					      rtnl_held, extack);
3101 			if (err < 0)
3102 				return err;
3103 			exts->nr_actions = err;
3104 		}
3105 	}
3106 #else
3107 	if ((exts->action && tb[exts->action]) ||
3108 	    (exts->police && tb[exts->police])) {
3109 		NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3110 		return -EOPNOTSUPP;
3111 	}
3112 #endif
3113 
3114 	return 0;
3115 }
3116 EXPORT_SYMBOL(tcf_exts_validate);
3117 
3118 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3119 {
3120 #ifdef CONFIG_NET_CLS_ACT
3121 	struct tcf_exts old = *dst;
3122 
3123 	*dst = *src;
3124 	tcf_exts_destroy(&old);
3125 #endif
3126 }
3127 EXPORT_SYMBOL(tcf_exts_change);
3128 
3129 #ifdef CONFIG_NET_CLS_ACT
3130 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3131 {
3132 	if (exts->nr_actions == 0)
3133 		return NULL;
3134 	else
3135 		return exts->actions[0];
3136 }
3137 #endif
3138 
3139 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3140 {
3141 #ifdef CONFIG_NET_CLS_ACT
3142 	struct nlattr *nest;
3143 
3144 	if (exts->action && tcf_exts_has_actions(exts)) {
3145 		/*
3146 		 * again for backward compatible mode - we want
3147 		 * to work with both old and new modes of entering
3148 		 * tc data even if iproute2  was newer - jhs
3149 		 */
3150 		if (exts->type != TCA_OLD_COMPAT) {
3151 			nest = nla_nest_start_noflag(skb, exts->action);
3152 			if (nest == NULL)
3153 				goto nla_put_failure;
3154 
3155 			if (tcf_action_dump(skb, exts->actions, 0, 0) < 0)
3156 				goto nla_put_failure;
3157 			nla_nest_end(skb, nest);
3158 		} else if (exts->police) {
3159 			struct tc_action *act = tcf_exts_first_act(exts);
3160 			nest = nla_nest_start_noflag(skb, exts->police);
3161 			if (nest == NULL || !act)
3162 				goto nla_put_failure;
3163 			if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3164 				goto nla_put_failure;
3165 			nla_nest_end(skb, nest);
3166 		}
3167 	}
3168 	return 0;
3169 
3170 nla_put_failure:
3171 	nla_nest_cancel(skb, nest);
3172 	return -1;
3173 #else
3174 	return 0;
3175 #endif
3176 }
3177 EXPORT_SYMBOL(tcf_exts_dump);
3178 
3179 
3180 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3181 {
3182 #ifdef CONFIG_NET_CLS_ACT
3183 	struct tc_action *a = tcf_exts_first_act(exts);
3184 	if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3185 		return -1;
3186 #endif
3187 	return 0;
3188 }
3189 EXPORT_SYMBOL(tcf_exts_dump_stats);
3190 
3191 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3192 {
3193 	if (*flags & TCA_CLS_FLAGS_IN_HW)
3194 		return;
3195 	*flags |= TCA_CLS_FLAGS_IN_HW;
3196 	atomic_inc(&block->offloadcnt);
3197 }
3198 
3199 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3200 {
3201 	if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3202 		return;
3203 	*flags &= ~TCA_CLS_FLAGS_IN_HW;
3204 	atomic_dec(&block->offloadcnt);
3205 }
3206 
3207 static void tc_cls_offload_cnt_update(struct tcf_block *block,
3208 				      struct tcf_proto *tp, u32 *cnt,
3209 				      u32 *flags, u32 diff, bool add)
3210 {
3211 	lockdep_assert_held(&block->cb_lock);
3212 
3213 	spin_lock(&tp->lock);
3214 	if (add) {
3215 		if (!*cnt)
3216 			tcf_block_offload_inc(block, flags);
3217 		*cnt += diff;
3218 	} else {
3219 		*cnt -= diff;
3220 		if (!*cnt)
3221 			tcf_block_offload_dec(block, flags);
3222 	}
3223 	spin_unlock(&tp->lock);
3224 }
3225 
3226 static void
3227 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3228 			 u32 *cnt, u32 *flags)
3229 {
3230 	lockdep_assert_held(&block->cb_lock);
3231 
3232 	spin_lock(&tp->lock);
3233 	tcf_block_offload_dec(block, flags);
3234 	*cnt = 0;
3235 	spin_unlock(&tp->lock);
3236 }
3237 
3238 static int
3239 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3240 		   void *type_data, bool err_stop)
3241 {
3242 	struct flow_block_cb *block_cb;
3243 	int ok_count = 0;
3244 	int err;
3245 
3246 	list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3247 		err = block_cb->cb(type, type_data, block_cb->cb_priv);
3248 		if (err) {
3249 			if (err_stop)
3250 				return err;
3251 		} else {
3252 			ok_count++;
3253 		}
3254 	}
3255 	return ok_count;
3256 }
3257 
3258 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3259 		     void *type_data, bool err_stop, bool rtnl_held)
3260 {
3261 	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3262 	int ok_count;
3263 
3264 retry:
3265 	if (take_rtnl)
3266 		rtnl_lock();
3267 	down_read(&block->cb_lock);
3268 	/* Need to obtain rtnl lock if block is bound to devs that require it.
3269 	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3270 	 * obtain the locks in same order here.
3271 	 */
3272 	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3273 		up_read(&block->cb_lock);
3274 		take_rtnl = true;
3275 		goto retry;
3276 	}
3277 
3278 	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3279 
3280 	up_read(&block->cb_lock);
3281 	if (take_rtnl)
3282 		rtnl_unlock();
3283 	return ok_count;
3284 }
3285 EXPORT_SYMBOL(tc_setup_cb_call);
3286 
3287 /* Non-destructive filter add. If filter that wasn't already in hardware is
3288  * successfully offloaded, increment block offloads counter. On failure,
3289  * previously offloaded filter is considered to be intact and offloads counter
3290  * is not decremented.
3291  */
3292 
3293 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3294 		    enum tc_setup_type type, void *type_data, bool err_stop,
3295 		    u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3296 {
3297 	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3298 	int ok_count;
3299 
3300 retry:
3301 	if (take_rtnl)
3302 		rtnl_lock();
3303 	down_read(&block->cb_lock);
3304 	/* Need to obtain rtnl lock if block is bound to devs that require it.
3305 	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3306 	 * obtain the locks in same order here.
3307 	 */
3308 	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3309 		up_read(&block->cb_lock);
3310 		take_rtnl = true;
3311 		goto retry;
3312 	}
3313 
3314 	/* Make sure all netdevs sharing this block are offload-capable. */
3315 	if (block->nooffloaddevcnt && err_stop) {
3316 		ok_count = -EOPNOTSUPP;
3317 		goto err_unlock;
3318 	}
3319 
3320 	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3321 	if (ok_count < 0)
3322 		goto err_unlock;
3323 
3324 	if (tp->ops->hw_add)
3325 		tp->ops->hw_add(tp, type_data);
3326 	if (ok_count > 0)
3327 		tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3328 					  ok_count, true);
3329 err_unlock:
3330 	up_read(&block->cb_lock);
3331 	if (take_rtnl)
3332 		rtnl_unlock();
3333 	return ok_count < 0 ? ok_count : 0;
3334 }
3335 EXPORT_SYMBOL(tc_setup_cb_add);
3336 
3337 /* Destructive filter replace. If filter that wasn't already in hardware is
3338  * successfully offloaded, increment block offload counter. On failure,
3339  * previously offloaded filter is considered to be destroyed and offload counter
3340  * is decremented.
3341  */
3342 
3343 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3344 			enum tc_setup_type type, void *type_data, bool err_stop,
3345 			u32 *old_flags, unsigned int *old_in_hw_count,
3346 			u32 *new_flags, unsigned int *new_in_hw_count,
3347 			bool rtnl_held)
3348 {
3349 	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3350 	int ok_count;
3351 
3352 retry:
3353 	if (take_rtnl)
3354 		rtnl_lock();
3355 	down_read(&block->cb_lock);
3356 	/* Need to obtain rtnl lock if block is bound to devs that require it.
3357 	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3358 	 * obtain the locks in same order here.
3359 	 */
3360 	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3361 		up_read(&block->cb_lock);
3362 		take_rtnl = true;
3363 		goto retry;
3364 	}
3365 
3366 	/* Make sure all netdevs sharing this block are offload-capable. */
3367 	if (block->nooffloaddevcnt && err_stop) {
3368 		ok_count = -EOPNOTSUPP;
3369 		goto err_unlock;
3370 	}
3371 
3372 	tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3373 	if (tp->ops->hw_del)
3374 		tp->ops->hw_del(tp, type_data);
3375 
3376 	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3377 	if (ok_count < 0)
3378 		goto err_unlock;
3379 
3380 	if (tp->ops->hw_add)
3381 		tp->ops->hw_add(tp, type_data);
3382 	if (ok_count > 0)
3383 		tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3384 					  new_flags, ok_count, true);
3385 err_unlock:
3386 	up_read(&block->cb_lock);
3387 	if (take_rtnl)
3388 		rtnl_unlock();
3389 	return ok_count < 0 ? ok_count : 0;
3390 }
3391 EXPORT_SYMBOL(tc_setup_cb_replace);
3392 
3393 /* Destroy filter and decrement block offload counter, if filter was previously
3394  * offloaded.
3395  */
3396 
3397 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3398 			enum tc_setup_type type, void *type_data, bool err_stop,
3399 			u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3400 {
3401 	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3402 	int ok_count;
3403 
3404 retry:
3405 	if (take_rtnl)
3406 		rtnl_lock();
3407 	down_read(&block->cb_lock);
3408 	/* Need to obtain rtnl lock if block is bound to devs that require it.
3409 	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3410 	 * obtain the locks in same order here.
3411 	 */
3412 	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3413 		up_read(&block->cb_lock);
3414 		take_rtnl = true;
3415 		goto retry;
3416 	}
3417 
3418 	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3419 
3420 	tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3421 	if (tp->ops->hw_del)
3422 		tp->ops->hw_del(tp, type_data);
3423 
3424 	up_read(&block->cb_lock);
3425 	if (take_rtnl)
3426 		rtnl_unlock();
3427 	return ok_count < 0 ? ok_count : 0;
3428 }
3429 EXPORT_SYMBOL(tc_setup_cb_destroy);
3430 
3431 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3432 			  bool add, flow_setup_cb_t *cb,
3433 			  enum tc_setup_type type, void *type_data,
3434 			  void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3435 {
3436 	int err = cb(type, type_data, cb_priv);
3437 
3438 	if (err) {
3439 		if (add && tc_skip_sw(*flags))
3440 			return err;
3441 	} else {
3442 		tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3443 					  add);
3444 	}
3445 
3446 	return 0;
3447 }
3448 EXPORT_SYMBOL(tc_setup_cb_reoffload);
3449 
3450 static int tcf_act_get_cookie(struct flow_action_entry *entry,
3451 			      const struct tc_action *act)
3452 {
3453 	struct tc_cookie *cookie;
3454 	int err = 0;
3455 
3456 	rcu_read_lock();
3457 	cookie = rcu_dereference(act->act_cookie);
3458 	if (cookie) {
3459 		entry->cookie = flow_action_cookie_create(cookie->data,
3460 							  cookie->len,
3461 							  GFP_ATOMIC);
3462 		if (!entry->cookie)
3463 			err = -ENOMEM;
3464 	}
3465 	rcu_read_unlock();
3466 	return err;
3467 }
3468 
3469 static void tcf_act_put_cookie(struct flow_action_entry *entry)
3470 {
3471 	flow_action_cookie_destroy(entry->cookie);
3472 }
3473 
3474 void tc_cleanup_flow_action(struct flow_action *flow_action)
3475 {
3476 	struct flow_action_entry *entry;
3477 	int i;
3478 
3479 	flow_action_for_each(i, entry, flow_action) {
3480 		tcf_act_put_cookie(entry);
3481 		if (entry->destructor)
3482 			entry->destructor(entry->destructor_priv);
3483 	}
3484 }
3485 EXPORT_SYMBOL(tc_cleanup_flow_action);
3486 
3487 static void tcf_mirred_get_dev(struct flow_action_entry *entry,
3488 			       const struct tc_action *act)
3489 {
3490 #ifdef CONFIG_NET_CLS_ACT
3491 	entry->dev = act->ops->get_dev(act, &entry->destructor);
3492 	if (!entry->dev)
3493 		return;
3494 	entry->destructor_priv = entry->dev;
3495 #endif
3496 }
3497 
3498 static void tcf_tunnel_encap_put_tunnel(void *priv)
3499 {
3500 	struct ip_tunnel_info *tunnel = priv;
3501 
3502 	kfree(tunnel);
3503 }
3504 
3505 static int tcf_tunnel_encap_get_tunnel(struct flow_action_entry *entry,
3506 				       const struct tc_action *act)
3507 {
3508 	entry->tunnel = tcf_tunnel_info_copy(act);
3509 	if (!entry->tunnel)
3510 		return -ENOMEM;
3511 	entry->destructor = tcf_tunnel_encap_put_tunnel;
3512 	entry->destructor_priv = entry->tunnel;
3513 	return 0;
3514 }
3515 
3516 static void tcf_sample_get_group(struct flow_action_entry *entry,
3517 				 const struct tc_action *act)
3518 {
3519 #ifdef CONFIG_NET_CLS_ACT
3520 	entry->sample.psample_group =
3521 		act->ops->get_psample_group(act, &entry->destructor);
3522 	entry->destructor_priv = entry->sample.psample_group;
3523 #endif
3524 }
3525 
3526 int tc_setup_flow_action(struct flow_action *flow_action,
3527 			 const struct tcf_exts *exts)
3528 {
3529 	struct tc_action *act;
3530 	int i, j, k, err = 0;
3531 
3532 	BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3533 	BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3534 	BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3535 
3536 	if (!exts)
3537 		return 0;
3538 
3539 	j = 0;
3540 	tcf_exts_for_each_action(i, act, exts) {
3541 		struct flow_action_entry *entry;
3542 
3543 		entry = &flow_action->entries[j];
3544 		spin_lock_bh(&act->tcfa_lock);
3545 		err = tcf_act_get_cookie(entry, act);
3546 		if (err)
3547 			goto err_out_locked;
3548 
3549 		entry->hw_stats = act->hw_stats;
3550 
3551 		if (is_tcf_gact_ok(act)) {
3552 			entry->id = FLOW_ACTION_ACCEPT;
3553 		} else if (is_tcf_gact_shot(act)) {
3554 			entry->id = FLOW_ACTION_DROP;
3555 		} else if (is_tcf_gact_trap(act)) {
3556 			entry->id = FLOW_ACTION_TRAP;
3557 		} else if (is_tcf_gact_goto_chain(act)) {
3558 			entry->id = FLOW_ACTION_GOTO;
3559 			entry->chain_index = tcf_gact_goto_chain_index(act);
3560 		} else if (is_tcf_mirred_egress_redirect(act)) {
3561 			entry->id = FLOW_ACTION_REDIRECT;
3562 			tcf_mirred_get_dev(entry, act);
3563 		} else if (is_tcf_mirred_egress_mirror(act)) {
3564 			entry->id = FLOW_ACTION_MIRRED;
3565 			tcf_mirred_get_dev(entry, act);
3566 		} else if (is_tcf_mirred_ingress_redirect(act)) {
3567 			entry->id = FLOW_ACTION_REDIRECT_INGRESS;
3568 			tcf_mirred_get_dev(entry, act);
3569 		} else if (is_tcf_mirred_ingress_mirror(act)) {
3570 			entry->id = FLOW_ACTION_MIRRED_INGRESS;
3571 			tcf_mirred_get_dev(entry, act);
3572 		} else if (is_tcf_vlan(act)) {
3573 			switch (tcf_vlan_action(act)) {
3574 			case TCA_VLAN_ACT_PUSH:
3575 				entry->id = FLOW_ACTION_VLAN_PUSH;
3576 				entry->vlan.vid = tcf_vlan_push_vid(act);
3577 				entry->vlan.proto = tcf_vlan_push_proto(act);
3578 				entry->vlan.prio = tcf_vlan_push_prio(act);
3579 				break;
3580 			case TCA_VLAN_ACT_POP:
3581 				entry->id = FLOW_ACTION_VLAN_POP;
3582 				break;
3583 			case TCA_VLAN_ACT_MODIFY:
3584 				entry->id = FLOW_ACTION_VLAN_MANGLE;
3585 				entry->vlan.vid = tcf_vlan_push_vid(act);
3586 				entry->vlan.proto = tcf_vlan_push_proto(act);
3587 				entry->vlan.prio = tcf_vlan_push_prio(act);
3588 				break;
3589 			default:
3590 				err = -EOPNOTSUPP;
3591 				goto err_out_locked;
3592 			}
3593 		} else if (is_tcf_tunnel_set(act)) {
3594 			entry->id = FLOW_ACTION_TUNNEL_ENCAP;
3595 			err = tcf_tunnel_encap_get_tunnel(entry, act);
3596 			if (err)
3597 				goto err_out_locked;
3598 		} else if (is_tcf_tunnel_release(act)) {
3599 			entry->id = FLOW_ACTION_TUNNEL_DECAP;
3600 		} else if (is_tcf_pedit(act)) {
3601 			for (k = 0; k < tcf_pedit_nkeys(act); k++) {
3602 				switch (tcf_pedit_cmd(act, k)) {
3603 				case TCA_PEDIT_KEY_EX_CMD_SET:
3604 					entry->id = FLOW_ACTION_MANGLE;
3605 					break;
3606 				case TCA_PEDIT_KEY_EX_CMD_ADD:
3607 					entry->id = FLOW_ACTION_ADD;
3608 					break;
3609 				default:
3610 					err = -EOPNOTSUPP;
3611 					goto err_out_locked;
3612 				}
3613 				entry->mangle.htype = tcf_pedit_htype(act, k);
3614 				entry->mangle.mask = tcf_pedit_mask(act, k);
3615 				entry->mangle.val = tcf_pedit_val(act, k);
3616 				entry->mangle.offset = tcf_pedit_offset(act, k);
3617 				entry->hw_stats = act->hw_stats;
3618 				entry = &flow_action->entries[++j];
3619 			}
3620 		} else if (is_tcf_csum(act)) {
3621 			entry->id = FLOW_ACTION_CSUM;
3622 			entry->csum_flags = tcf_csum_update_flags(act);
3623 		} else if (is_tcf_skbedit_mark(act)) {
3624 			entry->id = FLOW_ACTION_MARK;
3625 			entry->mark = tcf_skbedit_mark(act);
3626 		} else if (is_tcf_sample(act)) {
3627 			entry->id = FLOW_ACTION_SAMPLE;
3628 			entry->sample.trunc_size = tcf_sample_trunc_size(act);
3629 			entry->sample.truncate = tcf_sample_truncate(act);
3630 			entry->sample.rate = tcf_sample_rate(act);
3631 			tcf_sample_get_group(entry, act);
3632 		} else if (is_tcf_police(act)) {
3633 			entry->id = FLOW_ACTION_POLICE;
3634 			entry->police.burst = tcf_police_tcfp_burst(act);
3635 			entry->police.rate_bytes_ps =
3636 				tcf_police_rate_bytes_ps(act);
3637 		} else if (is_tcf_ct(act)) {
3638 			entry->id = FLOW_ACTION_CT;
3639 			entry->ct.action = tcf_ct_action(act);
3640 			entry->ct.zone = tcf_ct_zone(act);
3641 			entry->ct.flow_table = tcf_ct_ft(act);
3642 		} else if (is_tcf_mpls(act)) {
3643 			switch (tcf_mpls_action(act)) {
3644 			case TCA_MPLS_ACT_PUSH:
3645 				entry->id = FLOW_ACTION_MPLS_PUSH;
3646 				entry->mpls_push.proto = tcf_mpls_proto(act);
3647 				entry->mpls_push.label = tcf_mpls_label(act);
3648 				entry->mpls_push.tc = tcf_mpls_tc(act);
3649 				entry->mpls_push.bos = tcf_mpls_bos(act);
3650 				entry->mpls_push.ttl = tcf_mpls_ttl(act);
3651 				break;
3652 			case TCA_MPLS_ACT_POP:
3653 				entry->id = FLOW_ACTION_MPLS_POP;
3654 				entry->mpls_pop.proto = tcf_mpls_proto(act);
3655 				break;
3656 			case TCA_MPLS_ACT_MODIFY:
3657 				entry->id = FLOW_ACTION_MPLS_MANGLE;
3658 				entry->mpls_mangle.label = tcf_mpls_label(act);
3659 				entry->mpls_mangle.tc = tcf_mpls_tc(act);
3660 				entry->mpls_mangle.bos = tcf_mpls_bos(act);
3661 				entry->mpls_mangle.ttl = tcf_mpls_ttl(act);
3662 				break;
3663 			default:
3664 				goto err_out_locked;
3665 			}
3666 		} else if (is_tcf_skbedit_ptype(act)) {
3667 			entry->id = FLOW_ACTION_PTYPE;
3668 			entry->ptype = tcf_skbedit_ptype(act);
3669 		} else if (is_tcf_skbedit_priority(act)) {
3670 			entry->id = FLOW_ACTION_PRIORITY;
3671 			entry->priority = tcf_skbedit_priority(act);
3672 		} else {
3673 			err = -EOPNOTSUPP;
3674 			goto err_out_locked;
3675 		}
3676 		spin_unlock_bh(&act->tcfa_lock);
3677 
3678 		if (!is_tcf_pedit(act))
3679 			j++;
3680 	}
3681 
3682 err_out:
3683 	if (err)
3684 		tc_cleanup_flow_action(flow_action);
3685 
3686 	return err;
3687 err_out_locked:
3688 	spin_unlock_bh(&act->tcfa_lock);
3689 	goto err_out;
3690 }
3691 EXPORT_SYMBOL(tc_setup_flow_action);
3692 
3693 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3694 {
3695 	unsigned int num_acts = 0;
3696 	struct tc_action *act;
3697 	int i;
3698 
3699 	tcf_exts_for_each_action(i, act, exts) {
3700 		if (is_tcf_pedit(act))
3701 			num_acts += tcf_pedit_nkeys(act);
3702 		else
3703 			num_acts++;
3704 	}
3705 	return num_acts;
3706 }
3707 EXPORT_SYMBOL(tcf_exts_num_actions);
3708 
3709 static __net_init int tcf_net_init(struct net *net)
3710 {
3711 	struct tcf_net *tn = net_generic(net, tcf_net_id);
3712 
3713 	spin_lock_init(&tn->idr_lock);
3714 	idr_init(&tn->idr);
3715 	return 0;
3716 }
3717 
3718 static void __net_exit tcf_net_exit(struct net *net)
3719 {
3720 	struct tcf_net *tn = net_generic(net, tcf_net_id);
3721 
3722 	idr_destroy(&tn->idr);
3723 }
3724 
3725 static struct pernet_operations tcf_net_ops = {
3726 	.init = tcf_net_init,
3727 	.exit = tcf_net_exit,
3728 	.id   = &tcf_net_id,
3729 	.size = sizeof(struct tcf_net),
3730 };
3731 
3732 static struct flow_indr_block_entry block_entry = {
3733 	.cb = tc_indr_block_get_and_cmd,
3734 	.list = LIST_HEAD_INIT(block_entry.list),
3735 };
3736 
3737 static int __init tc_filter_init(void)
3738 {
3739 	int err;
3740 
3741 	tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3742 	if (!tc_filter_wq)
3743 		return -ENOMEM;
3744 
3745 	err = register_pernet_subsys(&tcf_net_ops);
3746 	if (err)
3747 		goto err_register_pernet_subsys;
3748 
3749 	flow_indr_add_block_cb(&block_entry);
3750 
3751 	rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3752 		      RTNL_FLAG_DOIT_UNLOCKED);
3753 	rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3754 		      RTNL_FLAG_DOIT_UNLOCKED);
3755 	rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3756 		      tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3757 	rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3758 	rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3759 	rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3760 		      tc_dump_chain, 0);
3761 
3762 	return 0;
3763 
3764 err_register_pernet_subsys:
3765 	destroy_workqueue(tc_filter_wq);
3766 	return err;
3767 }
3768 
3769 subsys_initcall(tc_filter_init);
3770