xref: /openbmc/linux/net/sched/cls_api.c (revision 08283d30)
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 <net/net_namespace.h>
26 #include <net/sock.h>
27 #include <net/netlink.h>
28 #include <net/pkt_sched.h>
29 #include <net/pkt_cls.h>
30 #include <net/tc_act/tc_pedit.h>
31 #include <net/tc_act/tc_mirred.h>
32 #include <net/tc_act/tc_vlan.h>
33 #include <net/tc_act/tc_tunnel_key.h>
34 #include <net/tc_act/tc_csum.h>
35 #include <net/tc_act/tc_gact.h>
36 #include <net/tc_act/tc_police.h>
37 #include <net/tc_act/tc_sample.h>
38 #include <net/tc_act/tc_skbedit.h>
39 #include <net/tc_act/tc_ct.h>
40 #include <net/tc_act/tc_mpls.h>
41 #include <net/flow_offload.h>
42 
43 extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
44 
45 /* The list of all installed classifier types */
46 static LIST_HEAD(tcf_proto_base);
47 
48 /* Protects list of registered TC modules. It is pure SMP lock. */
49 static DEFINE_RWLOCK(cls_mod_lock);
50 
51 static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
52 {
53 	return jhash_3words(tp->chain->index, tp->prio,
54 			    (__force __u32)tp->protocol, 0);
55 }
56 
57 static void tcf_proto_signal_destroying(struct tcf_chain *chain,
58 					struct tcf_proto *tp)
59 {
60 	struct tcf_block *block = chain->block;
61 
62 	mutex_lock(&block->proto_destroy_lock);
63 	hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
64 		     destroy_obj_hashfn(tp));
65 	mutex_unlock(&block->proto_destroy_lock);
66 }
67 
68 static bool tcf_proto_cmp(const struct tcf_proto *tp1,
69 			  const struct tcf_proto *tp2)
70 {
71 	return tp1->chain->index == tp2->chain->index &&
72 	       tp1->prio == tp2->prio &&
73 	       tp1->protocol == tp2->protocol;
74 }
75 
76 static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
77 					struct tcf_proto *tp)
78 {
79 	u32 hash = destroy_obj_hashfn(tp);
80 	struct tcf_proto *iter;
81 	bool found = false;
82 
83 	rcu_read_lock();
84 	hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
85 				   destroy_ht_node, hash) {
86 		if (tcf_proto_cmp(tp, iter)) {
87 			found = true;
88 			break;
89 		}
90 	}
91 	rcu_read_unlock();
92 
93 	return found;
94 }
95 
96 static void
97 tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
98 {
99 	struct tcf_block *block = chain->block;
100 
101 	mutex_lock(&block->proto_destroy_lock);
102 	if (hash_hashed(&tp->destroy_ht_node))
103 		hash_del_rcu(&tp->destroy_ht_node);
104 	mutex_unlock(&block->proto_destroy_lock);
105 }
106 
107 /* Find classifier type by string name */
108 
109 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
110 {
111 	const struct tcf_proto_ops *t, *res = NULL;
112 
113 	if (kind) {
114 		read_lock(&cls_mod_lock);
115 		list_for_each_entry(t, &tcf_proto_base, head) {
116 			if (strcmp(kind, t->kind) == 0) {
117 				if (try_module_get(t->owner))
118 					res = t;
119 				break;
120 			}
121 		}
122 		read_unlock(&cls_mod_lock);
123 	}
124 	return res;
125 }
126 
127 static const struct tcf_proto_ops *
128 tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
129 		     struct netlink_ext_ack *extack)
130 {
131 	const struct tcf_proto_ops *ops;
132 
133 	ops = __tcf_proto_lookup_ops(kind);
134 	if (ops)
135 		return ops;
136 #ifdef CONFIG_MODULES
137 	if (rtnl_held)
138 		rtnl_unlock();
139 	request_module("cls_%s", kind);
140 	if (rtnl_held)
141 		rtnl_lock();
142 	ops = __tcf_proto_lookup_ops(kind);
143 	/* We dropped the RTNL semaphore in order to perform
144 	 * the module load. So, even if we succeeded in loading
145 	 * the module we have to replay the request. We indicate
146 	 * this using -EAGAIN.
147 	 */
148 	if (ops) {
149 		module_put(ops->owner);
150 		return ERR_PTR(-EAGAIN);
151 	}
152 #endif
153 	NL_SET_ERR_MSG(extack, "TC classifier not found");
154 	return ERR_PTR(-ENOENT);
155 }
156 
157 /* Register(unregister) new classifier type */
158 
159 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
160 {
161 	struct tcf_proto_ops *t;
162 	int rc = -EEXIST;
163 
164 	write_lock(&cls_mod_lock);
165 	list_for_each_entry(t, &tcf_proto_base, head)
166 		if (!strcmp(ops->kind, t->kind))
167 			goto out;
168 
169 	list_add_tail(&ops->head, &tcf_proto_base);
170 	rc = 0;
171 out:
172 	write_unlock(&cls_mod_lock);
173 	return rc;
174 }
175 EXPORT_SYMBOL(register_tcf_proto_ops);
176 
177 static struct workqueue_struct *tc_filter_wq;
178 
179 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
180 {
181 	struct tcf_proto_ops *t;
182 	int rc = -ENOENT;
183 
184 	/* Wait for outstanding call_rcu()s, if any, from a
185 	 * tcf_proto_ops's destroy() handler.
186 	 */
187 	rcu_barrier();
188 	flush_workqueue(tc_filter_wq);
189 
190 	write_lock(&cls_mod_lock);
191 	list_for_each_entry(t, &tcf_proto_base, head) {
192 		if (t == ops) {
193 			list_del(&t->head);
194 			rc = 0;
195 			break;
196 		}
197 	}
198 	write_unlock(&cls_mod_lock);
199 	return rc;
200 }
201 EXPORT_SYMBOL(unregister_tcf_proto_ops);
202 
203 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
204 {
205 	INIT_RCU_WORK(rwork, func);
206 	return queue_rcu_work(tc_filter_wq, rwork);
207 }
208 EXPORT_SYMBOL(tcf_queue_work);
209 
210 /* Select new prio value from the range, managed by kernel. */
211 
212 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
213 {
214 	u32 first = TC_H_MAKE(0xC0000000U, 0U);
215 
216 	if (tp)
217 		first = tp->prio - 1;
218 
219 	return TC_H_MAJ(first);
220 }
221 
222 static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
223 {
224 	if (kind)
225 		return nla_strlcpy(name, kind, IFNAMSIZ) >= IFNAMSIZ;
226 	memset(name, 0, IFNAMSIZ);
227 	return false;
228 }
229 
230 static bool tcf_proto_is_unlocked(const char *kind)
231 {
232 	const struct tcf_proto_ops *ops;
233 	bool ret;
234 
235 	if (strlen(kind) == 0)
236 		return false;
237 
238 	ops = tcf_proto_lookup_ops(kind, false, NULL);
239 	/* On error return false to take rtnl lock. Proto lookup/create
240 	 * functions will perform lookup again and properly handle errors.
241 	 */
242 	if (IS_ERR(ops))
243 		return false;
244 
245 	ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
246 	module_put(ops->owner);
247 	return ret;
248 }
249 
250 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
251 					  u32 prio, struct tcf_chain *chain,
252 					  bool rtnl_held,
253 					  struct netlink_ext_ack *extack)
254 {
255 	struct tcf_proto *tp;
256 	int err;
257 
258 	tp = kzalloc(sizeof(*tp), GFP_KERNEL);
259 	if (!tp)
260 		return ERR_PTR(-ENOBUFS);
261 
262 	tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
263 	if (IS_ERR(tp->ops)) {
264 		err = PTR_ERR(tp->ops);
265 		goto errout;
266 	}
267 	tp->classify = tp->ops->classify;
268 	tp->protocol = protocol;
269 	tp->prio = prio;
270 	tp->chain = chain;
271 	spin_lock_init(&tp->lock);
272 	refcount_set(&tp->refcnt, 1);
273 
274 	err = tp->ops->init(tp);
275 	if (err) {
276 		module_put(tp->ops->owner);
277 		goto errout;
278 	}
279 	return tp;
280 
281 errout:
282 	kfree(tp);
283 	return ERR_PTR(err);
284 }
285 
286 static void tcf_proto_get(struct tcf_proto *tp)
287 {
288 	refcount_inc(&tp->refcnt);
289 }
290 
291 static void tcf_chain_put(struct tcf_chain *chain);
292 
293 static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
294 			      bool sig_destroy, struct netlink_ext_ack *extack)
295 {
296 	tp->ops->destroy(tp, rtnl_held, extack);
297 	if (sig_destroy)
298 		tcf_proto_signal_destroyed(tp->chain, tp);
299 	tcf_chain_put(tp->chain);
300 	module_put(tp->ops->owner);
301 	kfree_rcu(tp, rcu);
302 }
303 
304 static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
305 			  struct netlink_ext_ack *extack)
306 {
307 	if (refcount_dec_and_test(&tp->refcnt))
308 		tcf_proto_destroy(tp, rtnl_held, true, extack);
309 }
310 
311 static int walker_check_empty(struct tcf_proto *tp, void *fh,
312 			      struct tcf_walker *arg)
313 {
314 	if (fh) {
315 		arg->nonempty = true;
316 		return -1;
317 	}
318 	return 0;
319 }
320 
321 static bool tcf_proto_is_empty(struct tcf_proto *tp, bool rtnl_held)
322 {
323 	struct tcf_walker walker = { .fn = walker_check_empty, };
324 
325 	if (tp->ops->walk) {
326 		tp->ops->walk(tp, &walker, rtnl_held);
327 		return !walker.nonempty;
328 	}
329 	return true;
330 }
331 
332 static bool tcf_proto_check_delete(struct tcf_proto *tp, bool rtnl_held)
333 {
334 	spin_lock(&tp->lock);
335 	if (tcf_proto_is_empty(tp, rtnl_held))
336 		tp->deleting = true;
337 	spin_unlock(&tp->lock);
338 	return tp->deleting;
339 }
340 
341 static void tcf_proto_mark_delete(struct tcf_proto *tp)
342 {
343 	spin_lock(&tp->lock);
344 	tp->deleting = true;
345 	spin_unlock(&tp->lock);
346 }
347 
348 static bool tcf_proto_is_deleting(struct tcf_proto *tp)
349 {
350 	bool deleting;
351 
352 	spin_lock(&tp->lock);
353 	deleting = tp->deleting;
354 	spin_unlock(&tp->lock);
355 
356 	return deleting;
357 }
358 
359 #define ASSERT_BLOCK_LOCKED(block)					\
360 	lockdep_assert_held(&(block)->lock)
361 
362 struct tcf_filter_chain_list_item {
363 	struct list_head list;
364 	tcf_chain_head_change_t *chain_head_change;
365 	void *chain_head_change_priv;
366 };
367 
368 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
369 					  u32 chain_index)
370 {
371 	struct tcf_chain *chain;
372 
373 	ASSERT_BLOCK_LOCKED(block);
374 
375 	chain = kzalloc(sizeof(*chain), GFP_KERNEL);
376 	if (!chain)
377 		return NULL;
378 	list_add_tail(&chain->list, &block->chain_list);
379 	mutex_init(&chain->filter_chain_lock);
380 	chain->block = block;
381 	chain->index = chain_index;
382 	chain->refcnt = 1;
383 	if (!chain->index)
384 		block->chain0.chain = chain;
385 	return chain;
386 }
387 
388 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
389 				       struct tcf_proto *tp_head)
390 {
391 	if (item->chain_head_change)
392 		item->chain_head_change(tp_head, item->chain_head_change_priv);
393 }
394 
395 static void tcf_chain0_head_change(struct tcf_chain *chain,
396 				   struct tcf_proto *tp_head)
397 {
398 	struct tcf_filter_chain_list_item *item;
399 	struct tcf_block *block = chain->block;
400 
401 	if (chain->index)
402 		return;
403 
404 	mutex_lock(&block->lock);
405 	list_for_each_entry(item, &block->chain0.filter_chain_list, list)
406 		tcf_chain_head_change_item(item, tp_head);
407 	mutex_unlock(&block->lock);
408 }
409 
410 /* Returns true if block can be safely freed. */
411 
412 static bool tcf_chain_detach(struct tcf_chain *chain)
413 {
414 	struct tcf_block *block = chain->block;
415 
416 	ASSERT_BLOCK_LOCKED(block);
417 
418 	list_del(&chain->list);
419 	if (!chain->index)
420 		block->chain0.chain = NULL;
421 
422 	if (list_empty(&block->chain_list) &&
423 	    refcount_read(&block->refcnt) == 0)
424 		return true;
425 
426 	return false;
427 }
428 
429 static void tcf_block_destroy(struct tcf_block *block)
430 {
431 	mutex_destroy(&block->lock);
432 	mutex_destroy(&block->proto_destroy_lock);
433 	kfree_rcu(block, rcu);
434 }
435 
436 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
437 {
438 	struct tcf_block *block = chain->block;
439 
440 	mutex_destroy(&chain->filter_chain_lock);
441 	kfree_rcu(chain, rcu);
442 	if (free_block)
443 		tcf_block_destroy(block);
444 }
445 
446 static void tcf_chain_hold(struct tcf_chain *chain)
447 {
448 	ASSERT_BLOCK_LOCKED(chain->block);
449 
450 	++chain->refcnt;
451 }
452 
453 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
454 {
455 	ASSERT_BLOCK_LOCKED(chain->block);
456 
457 	/* In case all the references are action references, this
458 	 * chain should not be shown to the user.
459 	 */
460 	return chain->refcnt == chain->action_refcnt;
461 }
462 
463 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
464 					  u32 chain_index)
465 {
466 	struct tcf_chain *chain;
467 
468 	ASSERT_BLOCK_LOCKED(block);
469 
470 	list_for_each_entry(chain, &block->chain_list, list) {
471 		if (chain->index == chain_index)
472 			return chain;
473 	}
474 	return NULL;
475 }
476 
477 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
478 			   u32 seq, u16 flags, int event, bool unicast);
479 
480 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
481 					 u32 chain_index, bool create,
482 					 bool by_act)
483 {
484 	struct tcf_chain *chain = NULL;
485 	bool is_first_reference;
486 
487 	mutex_lock(&block->lock);
488 	chain = tcf_chain_lookup(block, chain_index);
489 	if (chain) {
490 		tcf_chain_hold(chain);
491 	} else {
492 		if (!create)
493 			goto errout;
494 		chain = tcf_chain_create(block, chain_index);
495 		if (!chain)
496 			goto errout;
497 	}
498 
499 	if (by_act)
500 		++chain->action_refcnt;
501 	is_first_reference = chain->refcnt - chain->action_refcnt == 1;
502 	mutex_unlock(&block->lock);
503 
504 	/* Send notification only in case we got the first
505 	 * non-action reference. Until then, the chain acts only as
506 	 * a placeholder for actions pointing to it and user ought
507 	 * not know about them.
508 	 */
509 	if (is_first_reference && !by_act)
510 		tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
511 				RTM_NEWCHAIN, false);
512 
513 	return chain;
514 
515 errout:
516 	mutex_unlock(&block->lock);
517 	return chain;
518 }
519 
520 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
521 				       bool create)
522 {
523 	return __tcf_chain_get(block, chain_index, create, false);
524 }
525 
526 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
527 {
528 	return __tcf_chain_get(block, chain_index, true, true);
529 }
530 EXPORT_SYMBOL(tcf_chain_get_by_act);
531 
532 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
533 			       void *tmplt_priv);
534 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
535 				  void *tmplt_priv, u32 chain_index,
536 				  struct tcf_block *block, struct sk_buff *oskb,
537 				  u32 seq, u16 flags, bool unicast);
538 
539 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
540 			    bool explicitly_created)
541 {
542 	struct tcf_block *block = chain->block;
543 	const struct tcf_proto_ops *tmplt_ops;
544 	bool free_block = false;
545 	unsigned int refcnt;
546 	void *tmplt_priv;
547 
548 	mutex_lock(&block->lock);
549 	if (explicitly_created) {
550 		if (!chain->explicitly_created) {
551 			mutex_unlock(&block->lock);
552 			return;
553 		}
554 		chain->explicitly_created = false;
555 	}
556 
557 	if (by_act)
558 		chain->action_refcnt--;
559 
560 	/* tc_chain_notify_delete can't be called while holding block lock.
561 	 * However, when block is unlocked chain can be changed concurrently, so
562 	 * save these to temporary variables.
563 	 */
564 	refcnt = --chain->refcnt;
565 	tmplt_ops = chain->tmplt_ops;
566 	tmplt_priv = chain->tmplt_priv;
567 
568 	/* The last dropped non-action reference will trigger notification. */
569 	if (refcnt - chain->action_refcnt == 0 && !by_act) {
570 		tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
571 				       block, NULL, 0, 0, false);
572 		/* Last reference to chain, no need to lock. */
573 		chain->flushing = false;
574 	}
575 
576 	if (refcnt == 0)
577 		free_block = tcf_chain_detach(chain);
578 	mutex_unlock(&block->lock);
579 
580 	if (refcnt == 0) {
581 		tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
582 		tcf_chain_destroy(chain, free_block);
583 	}
584 }
585 
586 static void tcf_chain_put(struct tcf_chain *chain)
587 {
588 	__tcf_chain_put(chain, false, false);
589 }
590 
591 void tcf_chain_put_by_act(struct tcf_chain *chain)
592 {
593 	__tcf_chain_put(chain, true, false);
594 }
595 EXPORT_SYMBOL(tcf_chain_put_by_act);
596 
597 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
598 {
599 	__tcf_chain_put(chain, false, true);
600 }
601 
602 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
603 {
604 	struct tcf_proto *tp, *tp_next;
605 
606 	mutex_lock(&chain->filter_chain_lock);
607 	tp = tcf_chain_dereference(chain->filter_chain, chain);
608 	while (tp) {
609 		tp_next = rcu_dereference_protected(tp->next, 1);
610 		tcf_proto_signal_destroying(chain, tp);
611 		tp = tp_next;
612 	}
613 	tp = tcf_chain_dereference(chain->filter_chain, chain);
614 	RCU_INIT_POINTER(chain->filter_chain, NULL);
615 	tcf_chain0_head_change(chain, NULL);
616 	chain->flushing = true;
617 	mutex_unlock(&chain->filter_chain_lock);
618 
619 	while (tp) {
620 		tp_next = rcu_dereference_protected(tp->next, 1);
621 		tcf_proto_put(tp, rtnl_held, NULL);
622 		tp = tp_next;
623 	}
624 }
625 
626 static int tcf_block_setup(struct tcf_block *block,
627 			   struct flow_block_offload *bo);
628 
629 static void tc_indr_block_ing_cmd(struct net_device *dev,
630 				  struct tcf_block *block,
631 				  flow_indr_block_bind_cb_t *cb,
632 				  void *cb_priv,
633 				  enum flow_block_command command)
634 {
635 	struct flow_block_offload bo = {
636 		.command	= command,
637 		.binder_type	= FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS,
638 		.net		= dev_net(dev),
639 		.block_shared	= tcf_block_non_null_shared(block),
640 	};
641 	INIT_LIST_HEAD(&bo.cb_list);
642 
643 	if (!block)
644 		return;
645 
646 	bo.block = &block->flow_block;
647 
648 	down_write(&block->cb_lock);
649 	cb(dev, cb_priv, TC_SETUP_BLOCK, &bo);
650 
651 	tcf_block_setup(block, &bo);
652 	up_write(&block->cb_lock);
653 }
654 
655 static struct tcf_block *tc_dev_ingress_block(struct net_device *dev)
656 {
657 	const struct Qdisc_class_ops *cops;
658 	struct Qdisc *qdisc;
659 
660 	if (!dev_ingress_queue(dev))
661 		return NULL;
662 
663 	qdisc = dev_ingress_queue(dev)->qdisc_sleeping;
664 	if (!qdisc)
665 		return NULL;
666 
667 	cops = qdisc->ops->cl_ops;
668 	if (!cops)
669 		return NULL;
670 
671 	if (!cops->tcf_block)
672 		return NULL;
673 
674 	return cops->tcf_block(qdisc, TC_H_MIN_INGRESS, NULL);
675 }
676 
677 static void tc_indr_block_get_and_ing_cmd(struct net_device *dev,
678 					  flow_indr_block_bind_cb_t *cb,
679 					  void *cb_priv,
680 					  enum flow_block_command command)
681 {
682 	struct tcf_block *block = tc_dev_ingress_block(dev);
683 
684 	tc_indr_block_ing_cmd(dev, block, cb, cb_priv, command);
685 }
686 
687 static void tc_indr_block_call(struct tcf_block *block,
688 			       struct net_device *dev,
689 			       struct tcf_block_ext_info *ei,
690 			       enum flow_block_command command,
691 			       struct netlink_ext_ack *extack)
692 {
693 	struct flow_block_offload bo = {
694 		.command	= command,
695 		.binder_type	= ei->binder_type,
696 		.net		= dev_net(dev),
697 		.block		= &block->flow_block,
698 		.block_shared	= tcf_block_shared(block),
699 		.extack		= extack,
700 	};
701 	INIT_LIST_HEAD(&bo.cb_list);
702 
703 	flow_indr_block_call(dev, &bo, command);
704 	tcf_block_setup(block, &bo);
705 }
706 
707 static bool tcf_block_offload_in_use(struct tcf_block *block)
708 {
709 	return atomic_read(&block->offloadcnt);
710 }
711 
712 static int tcf_block_offload_cmd(struct tcf_block *block,
713 				 struct net_device *dev,
714 				 struct tcf_block_ext_info *ei,
715 				 enum flow_block_command command,
716 				 struct netlink_ext_ack *extack)
717 {
718 	struct flow_block_offload bo = {};
719 	int err;
720 
721 	bo.net = dev_net(dev);
722 	bo.command = command;
723 	bo.binder_type = ei->binder_type;
724 	bo.block = &block->flow_block;
725 	bo.block_shared = tcf_block_shared(block);
726 	bo.extack = extack;
727 	INIT_LIST_HEAD(&bo.cb_list);
728 
729 	err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
730 	if (err < 0)
731 		return err;
732 
733 	return tcf_block_setup(block, &bo);
734 }
735 
736 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
737 				  struct tcf_block_ext_info *ei,
738 				  struct netlink_ext_ack *extack)
739 {
740 	struct net_device *dev = q->dev_queue->dev;
741 	int err;
742 
743 	down_write(&block->cb_lock);
744 	if (!dev->netdev_ops->ndo_setup_tc)
745 		goto no_offload_dev_inc;
746 
747 	/* If tc offload feature is disabled and the block we try to bind
748 	 * to already has some offloaded filters, forbid to bind.
749 	 */
750 	if (!tc_can_offload(dev) && tcf_block_offload_in_use(block)) {
751 		NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
752 		err = -EOPNOTSUPP;
753 		goto err_unlock;
754 	}
755 
756 	err = tcf_block_offload_cmd(block, dev, ei, FLOW_BLOCK_BIND, extack);
757 	if (err == -EOPNOTSUPP)
758 		goto no_offload_dev_inc;
759 	if (err)
760 		goto err_unlock;
761 
762 	tc_indr_block_call(block, dev, ei, FLOW_BLOCK_BIND, extack);
763 	up_write(&block->cb_lock);
764 	return 0;
765 
766 no_offload_dev_inc:
767 	if (tcf_block_offload_in_use(block)) {
768 		err = -EOPNOTSUPP;
769 		goto err_unlock;
770 	}
771 	err = 0;
772 	block->nooffloaddevcnt++;
773 	tc_indr_block_call(block, dev, ei, FLOW_BLOCK_BIND, extack);
774 err_unlock:
775 	up_write(&block->cb_lock);
776 	return err;
777 }
778 
779 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
780 				     struct tcf_block_ext_info *ei)
781 {
782 	struct net_device *dev = q->dev_queue->dev;
783 	int err;
784 
785 	down_write(&block->cb_lock);
786 	tc_indr_block_call(block, dev, ei, FLOW_BLOCK_UNBIND, NULL);
787 
788 	if (!dev->netdev_ops->ndo_setup_tc)
789 		goto no_offload_dev_dec;
790 	err = tcf_block_offload_cmd(block, dev, ei, FLOW_BLOCK_UNBIND, NULL);
791 	if (err == -EOPNOTSUPP)
792 		goto no_offload_dev_dec;
793 	up_write(&block->cb_lock);
794 	return;
795 
796 no_offload_dev_dec:
797 	WARN_ON(block->nooffloaddevcnt-- == 0);
798 	up_write(&block->cb_lock);
799 }
800 
801 static int
802 tcf_chain0_head_change_cb_add(struct tcf_block *block,
803 			      struct tcf_block_ext_info *ei,
804 			      struct netlink_ext_ack *extack)
805 {
806 	struct tcf_filter_chain_list_item *item;
807 	struct tcf_chain *chain0;
808 
809 	item = kmalloc(sizeof(*item), GFP_KERNEL);
810 	if (!item) {
811 		NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
812 		return -ENOMEM;
813 	}
814 	item->chain_head_change = ei->chain_head_change;
815 	item->chain_head_change_priv = ei->chain_head_change_priv;
816 
817 	mutex_lock(&block->lock);
818 	chain0 = block->chain0.chain;
819 	if (chain0)
820 		tcf_chain_hold(chain0);
821 	else
822 		list_add(&item->list, &block->chain0.filter_chain_list);
823 	mutex_unlock(&block->lock);
824 
825 	if (chain0) {
826 		struct tcf_proto *tp_head;
827 
828 		mutex_lock(&chain0->filter_chain_lock);
829 
830 		tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
831 		if (tp_head)
832 			tcf_chain_head_change_item(item, tp_head);
833 
834 		mutex_lock(&block->lock);
835 		list_add(&item->list, &block->chain0.filter_chain_list);
836 		mutex_unlock(&block->lock);
837 
838 		mutex_unlock(&chain0->filter_chain_lock);
839 		tcf_chain_put(chain0);
840 	}
841 
842 	return 0;
843 }
844 
845 static void
846 tcf_chain0_head_change_cb_del(struct tcf_block *block,
847 			      struct tcf_block_ext_info *ei)
848 {
849 	struct tcf_filter_chain_list_item *item;
850 
851 	mutex_lock(&block->lock);
852 	list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
853 		if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
854 		    (item->chain_head_change == ei->chain_head_change &&
855 		     item->chain_head_change_priv == ei->chain_head_change_priv)) {
856 			if (block->chain0.chain)
857 				tcf_chain_head_change_item(item, NULL);
858 			list_del(&item->list);
859 			mutex_unlock(&block->lock);
860 
861 			kfree(item);
862 			return;
863 		}
864 	}
865 	mutex_unlock(&block->lock);
866 	WARN_ON(1);
867 }
868 
869 struct tcf_net {
870 	spinlock_t idr_lock; /* Protects idr */
871 	struct idr idr;
872 };
873 
874 static unsigned int tcf_net_id;
875 
876 static int tcf_block_insert(struct tcf_block *block, struct net *net,
877 			    struct netlink_ext_ack *extack)
878 {
879 	struct tcf_net *tn = net_generic(net, tcf_net_id);
880 	int err;
881 
882 	idr_preload(GFP_KERNEL);
883 	spin_lock(&tn->idr_lock);
884 	err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
885 			    GFP_NOWAIT);
886 	spin_unlock(&tn->idr_lock);
887 	idr_preload_end();
888 
889 	return err;
890 }
891 
892 static void tcf_block_remove(struct tcf_block *block, struct net *net)
893 {
894 	struct tcf_net *tn = net_generic(net, tcf_net_id);
895 
896 	spin_lock(&tn->idr_lock);
897 	idr_remove(&tn->idr, block->index);
898 	spin_unlock(&tn->idr_lock);
899 }
900 
901 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
902 					  u32 block_index,
903 					  struct netlink_ext_ack *extack)
904 {
905 	struct tcf_block *block;
906 
907 	block = kzalloc(sizeof(*block), GFP_KERNEL);
908 	if (!block) {
909 		NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
910 		return ERR_PTR(-ENOMEM);
911 	}
912 	mutex_init(&block->lock);
913 	mutex_init(&block->proto_destroy_lock);
914 	init_rwsem(&block->cb_lock);
915 	flow_block_init(&block->flow_block);
916 	INIT_LIST_HEAD(&block->chain_list);
917 	INIT_LIST_HEAD(&block->owner_list);
918 	INIT_LIST_HEAD(&block->chain0.filter_chain_list);
919 
920 	refcount_set(&block->refcnt, 1);
921 	block->net = net;
922 	block->index = block_index;
923 
924 	/* Don't store q pointer for blocks which are shared */
925 	if (!tcf_block_shared(block))
926 		block->q = q;
927 	return block;
928 }
929 
930 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
931 {
932 	struct tcf_net *tn = net_generic(net, tcf_net_id);
933 
934 	return idr_find(&tn->idr, block_index);
935 }
936 
937 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
938 {
939 	struct tcf_block *block;
940 
941 	rcu_read_lock();
942 	block = tcf_block_lookup(net, block_index);
943 	if (block && !refcount_inc_not_zero(&block->refcnt))
944 		block = NULL;
945 	rcu_read_unlock();
946 
947 	return block;
948 }
949 
950 static struct tcf_chain *
951 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
952 {
953 	mutex_lock(&block->lock);
954 	if (chain)
955 		chain = list_is_last(&chain->list, &block->chain_list) ?
956 			NULL : list_next_entry(chain, list);
957 	else
958 		chain = list_first_entry_or_null(&block->chain_list,
959 						 struct tcf_chain, list);
960 
961 	/* skip all action-only chains */
962 	while (chain && tcf_chain_held_by_acts_only(chain))
963 		chain = list_is_last(&chain->list, &block->chain_list) ?
964 			NULL : list_next_entry(chain, list);
965 
966 	if (chain)
967 		tcf_chain_hold(chain);
968 	mutex_unlock(&block->lock);
969 
970 	return chain;
971 }
972 
973 /* Function to be used by all clients that want to iterate over all chains on
974  * block. It properly obtains block->lock and takes reference to chain before
975  * returning it. Users of this function must be tolerant to concurrent chain
976  * insertion/deletion or ensure that no concurrent chain modification is
977  * possible. Note that all netlink dump callbacks cannot guarantee to provide
978  * consistent dump because rtnl lock is released each time skb is filled with
979  * data and sent to user-space.
980  */
981 
982 struct tcf_chain *
983 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
984 {
985 	struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
986 
987 	if (chain)
988 		tcf_chain_put(chain);
989 
990 	return chain_next;
991 }
992 EXPORT_SYMBOL(tcf_get_next_chain);
993 
994 static struct tcf_proto *
995 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
996 {
997 	u32 prio = 0;
998 
999 	ASSERT_RTNL();
1000 	mutex_lock(&chain->filter_chain_lock);
1001 
1002 	if (!tp) {
1003 		tp = tcf_chain_dereference(chain->filter_chain, chain);
1004 	} else if (tcf_proto_is_deleting(tp)) {
1005 		/* 'deleting' flag is set and chain->filter_chain_lock was
1006 		 * unlocked, which means next pointer could be invalid. Restart
1007 		 * search.
1008 		 */
1009 		prio = tp->prio + 1;
1010 		tp = tcf_chain_dereference(chain->filter_chain, chain);
1011 
1012 		for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1013 			if (!tp->deleting && tp->prio >= prio)
1014 				break;
1015 	} else {
1016 		tp = tcf_chain_dereference(tp->next, chain);
1017 	}
1018 
1019 	if (tp)
1020 		tcf_proto_get(tp);
1021 
1022 	mutex_unlock(&chain->filter_chain_lock);
1023 
1024 	return tp;
1025 }
1026 
1027 /* Function to be used by all clients that want to iterate over all tp's on
1028  * chain. Users of this function must be tolerant to concurrent tp
1029  * insertion/deletion or ensure that no concurrent chain modification is
1030  * possible. Note that all netlink dump callbacks cannot guarantee to provide
1031  * consistent dump because rtnl lock is released each time skb is filled with
1032  * data and sent to user-space.
1033  */
1034 
1035 struct tcf_proto *
1036 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp,
1037 		   bool rtnl_held)
1038 {
1039 	struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1040 
1041 	if (tp)
1042 		tcf_proto_put(tp, rtnl_held, NULL);
1043 
1044 	return tp_next;
1045 }
1046 EXPORT_SYMBOL(tcf_get_next_proto);
1047 
1048 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1049 {
1050 	struct tcf_chain *chain;
1051 
1052 	/* Last reference to block. At this point chains cannot be added or
1053 	 * removed concurrently.
1054 	 */
1055 	for (chain = tcf_get_next_chain(block, NULL);
1056 	     chain;
1057 	     chain = tcf_get_next_chain(block, chain)) {
1058 		tcf_chain_put_explicitly_created(chain);
1059 		tcf_chain_flush(chain, rtnl_held);
1060 	}
1061 }
1062 
1063 /* Lookup Qdisc and increments its reference counter.
1064  * Set parent, if necessary.
1065  */
1066 
1067 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1068 			    u32 *parent, int ifindex, bool rtnl_held,
1069 			    struct netlink_ext_ack *extack)
1070 {
1071 	const struct Qdisc_class_ops *cops;
1072 	struct net_device *dev;
1073 	int err = 0;
1074 
1075 	if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1076 		return 0;
1077 
1078 	rcu_read_lock();
1079 
1080 	/* Find link */
1081 	dev = dev_get_by_index_rcu(net, ifindex);
1082 	if (!dev) {
1083 		rcu_read_unlock();
1084 		return -ENODEV;
1085 	}
1086 
1087 	/* Find qdisc */
1088 	if (!*parent) {
1089 		*q = dev->qdisc;
1090 		*parent = (*q)->handle;
1091 	} else {
1092 		*q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1093 		if (!*q) {
1094 			NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1095 			err = -EINVAL;
1096 			goto errout_rcu;
1097 		}
1098 	}
1099 
1100 	*q = qdisc_refcount_inc_nz(*q);
1101 	if (!*q) {
1102 		NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1103 		err = -EINVAL;
1104 		goto errout_rcu;
1105 	}
1106 
1107 	/* Is it classful? */
1108 	cops = (*q)->ops->cl_ops;
1109 	if (!cops) {
1110 		NL_SET_ERR_MSG(extack, "Qdisc not classful");
1111 		err = -EINVAL;
1112 		goto errout_qdisc;
1113 	}
1114 
1115 	if (!cops->tcf_block) {
1116 		NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1117 		err = -EOPNOTSUPP;
1118 		goto errout_qdisc;
1119 	}
1120 
1121 errout_rcu:
1122 	/* At this point we know that qdisc is not noop_qdisc,
1123 	 * which means that qdisc holds a reference to net_device
1124 	 * and we hold a reference to qdisc, so it is safe to release
1125 	 * rcu read lock.
1126 	 */
1127 	rcu_read_unlock();
1128 	return err;
1129 
1130 errout_qdisc:
1131 	rcu_read_unlock();
1132 
1133 	if (rtnl_held)
1134 		qdisc_put(*q);
1135 	else
1136 		qdisc_put_unlocked(*q);
1137 	*q = NULL;
1138 
1139 	return err;
1140 }
1141 
1142 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1143 			       int ifindex, struct netlink_ext_ack *extack)
1144 {
1145 	if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1146 		return 0;
1147 
1148 	/* Do we search for filter, attached to class? */
1149 	if (TC_H_MIN(parent)) {
1150 		const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1151 
1152 		*cl = cops->find(q, parent);
1153 		if (*cl == 0) {
1154 			NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1155 			return -ENOENT;
1156 		}
1157 	}
1158 
1159 	return 0;
1160 }
1161 
1162 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1163 					  unsigned long cl, int ifindex,
1164 					  u32 block_index,
1165 					  struct netlink_ext_ack *extack)
1166 {
1167 	struct tcf_block *block;
1168 
1169 	if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1170 		block = tcf_block_refcnt_get(net, block_index);
1171 		if (!block) {
1172 			NL_SET_ERR_MSG(extack, "Block of given index was not found");
1173 			return ERR_PTR(-EINVAL);
1174 		}
1175 	} else {
1176 		const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1177 
1178 		block = cops->tcf_block(q, cl, extack);
1179 		if (!block)
1180 			return ERR_PTR(-EINVAL);
1181 
1182 		if (tcf_block_shared(block)) {
1183 			NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1184 			return ERR_PTR(-EOPNOTSUPP);
1185 		}
1186 
1187 		/* Always take reference to block in order to support execution
1188 		 * of rules update path of cls API without rtnl lock. Caller
1189 		 * must release block when it is finished using it. 'if' block
1190 		 * of this conditional obtain reference to block by calling
1191 		 * tcf_block_refcnt_get().
1192 		 */
1193 		refcount_inc(&block->refcnt);
1194 	}
1195 
1196 	return block;
1197 }
1198 
1199 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1200 			    struct tcf_block_ext_info *ei, bool rtnl_held)
1201 {
1202 	if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1203 		/* Flushing/putting all chains will cause the block to be
1204 		 * deallocated when last chain is freed. However, if chain_list
1205 		 * is empty, block has to be manually deallocated. After block
1206 		 * reference counter reached 0, it is no longer possible to
1207 		 * increment it or add new chains to block.
1208 		 */
1209 		bool free_block = list_empty(&block->chain_list);
1210 
1211 		mutex_unlock(&block->lock);
1212 		if (tcf_block_shared(block))
1213 			tcf_block_remove(block, block->net);
1214 
1215 		if (q)
1216 			tcf_block_offload_unbind(block, q, ei);
1217 
1218 		if (free_block)
1219 			tcf_block_destroy(block);
1220 		else
1221 			tcf_block_flush_all_chains(block, rtnl_held);
1222 	} else if (q) {
1223 		tcf_block_offload_unbind(block, q, ei);
1224 	}
1225 }
1226 
1227 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1228 {
1229 	__tcf_block_put(block, NULL, NULL, rtnl_held);
1230 }
1231 
1232 /* Find tcf block.
1233  * Set q, parent, cl when appropriate.
1234  */
1235 
1236 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1237 					u32 *parent, unsigned long *cl,
1238 					int ifindex, u32 block_index,
1239 					struct netlink_ext_ack *extack)
1240 {
1241 	struct tcf_block *block;
1242 	int err = 0;
1243 
1244 	ASSERT_RTNL();
1245 
1246 	err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1247 	if (err)
1248 		goto errout;
1249 
1250 	err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1251 	if (err)
1252 		goto errout_qdisc;
1253 
1254 	block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1255 	if (IS_ERR(block)) {
1256 		err = PTR_ERR(block);
1257 		goto errout_qdisc;
1258 	}
1259 
1260 	return block;
1261 
1262 errout_qdisc:
1263 	if (*q)
1264 		qdisc_put(*q);
1265 errout:
1266 	*q = NULL;
1267 	return ERR_PTR(err);
1268 }
1269 
1270 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1271 			      bool rtnl_held)
1272 {
1273 	if (!IS_ERR_OR_NULL(block))
1274 		tcf_block_refcnt_put(block, rtnl_held);
1275 
1276 	if (q) {
1277 		if (rtnl_held)
1278 			qdisc_put(q);
1279 		else
1280 			qdisc_put_unlocked(q);
1281 	}
1282 }
1283 
1284 struct tcf_block_owner_item {
1285 	struct list_head list;
1286 	struct Qdisc *q;
1287 	enum flow_block_binder_type binder_type;
1288 };
1289 
1290 static void
1291 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1292 			       struct Qdisc *q,
1293 			       enum flow_block_binder_type binder_type)
1294 {
1295 	if (block->keep_dst &&
1296 	    binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1297 	    binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1298 		netif_keep_dst(qdisc_dev(q));
1299 }
1300 
1301 void tcf_block_netif_keep_dst(struct tcf_block *block)
1302 {
1303 	struct tcf_block_owner_item *item;
1304 
1305 	block->keep_dst = true;
1306 	list_for_each_entry(item, &block->owner_list, list)
1307 		tcf_block_owner_netif_keep_dst(block, item->q,
1308 					       item->binder_type);
1309 }
1310 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1311 
1312 static int tcf_block_owner_add(struct tcf_block *block,
1313 			       struct Qdisc *q,
1314 			       enum flow_block_binder_type binder_type)
1315 {
1316 	struct tcf_block_owner_item *item;
1317 
1318 	item = kmalloc(sizeof(*item), GFP_KERNEL);
1319 	if (!item)
1320 		return -ENOMEM;
1321 	item->q = q;
1322 	item->binder_type = binder_type;
1323 	list_add(&item->list, &block->owner_list);
1324 	return 0;
1325 }
1326 
1327 static void tcf_block_owner_del(struct tcf_block *block,
1328 				struct Qdisc *q,
1329 				enum flow_block_binder_type binder_type)
1330 {
1331 	struct tcf_block_owner_item *item;
1332 
1333 	list_for_each_entry(item, &block->owner_list, list) {
1334 		if (item->q == q && item->binder_type == binder_type) {
1335 			list_del(&item->list);
1336 			kfree(item);
1337 			return;
1338 		}
1339 	}
1340 	WARN_ON(1);
1341 }
1342 
1343 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1344 		      struct tcf_block_ext_info *ei,
1345 		      struct netlink_ext_ack *extack)
1346 {
1347 	struct net *net = qdisc_net(q);
1348 	struct tcf_block *block = NULL;
1349 	int err;
1350 
1351 	if (ei->block_index)
1352 		/* block_index not 0 means the shared block is requested */
1353 		block = tcf_block_refcnt_get(net, ei->block_index);
1354 
1355 	if (!block) {
1356 		block = tcf_block_create(net, q, ei->block_index, extack);
1357 		if (IS_ERR(block))
1358 			return PTR_ERR(block);
1359 		if (tcf_block_shared(block)) {
1360 			err = tcf_block_insert(block, net, extack);
1361 			if (err)
1362 				goto err_block_insert;
1363 		}
1364 	}
1365 
1366 	err = tcf_block_owner_add(block, q, ei->binder_type);
1367 	if (err)
1368 		goto err_block_owner_add;
1369 
1370 	tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1371 
1372 	err = tcf_chain0_head_change_cb_add(block, ei, extack);
1373 	if (err)
1374 		goto err_chain0_head_change_cb_add;
1375 
1376 	err = tcf_block_offload_bind(block, q, ei, extack);
1377 	if (err)
1378 		goto err_block_offload_bind;
1379 
1380 	*p_block = block;
1381 	return 0;
1382 
1383 err_block_offload_bind:
1384 	tcf_chain0_head_change_cb_del(block, ei);
1385 err_chain0_head_change_cb_add:
1386 	tcf_block_owner_del(block, q, ei->binder_type);
1387 err_block_owner_add:
1388 err_block_insert:
1389 	tcf_block_refcnt_put(block, true);
1390 	return err;
1391 }
1392 EXPORT_SYMBOL(tcf_block_get_ext);
1393 
1394 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1395 {
1396 	struct tcf_proto __rcu **p_filter_chain = priv;
1397 
1398 	rcu_assign_pointer(*p_filter_chain, tp_head);
1399 }
1400 
1401 int tcf_block_get(struct tcf_block **p_block,
1402 		  struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1403 		  struct netlink_ext_ack *extack)
1404 {
1405 	struct tcf_block_ext_info ei = {
1406 		.chain_head_change = tcf_chain_head_change_dflt,
1407 		.chain_head_change_priv = p_filter_chain,
1408 	};
1409 
1410 	WARN_ON(!p_filter_chain);
1411 	return tcf_block_get_ext(p_block, q, &ei, extack);
1412 }
1413 EXPORT_SYMBOL(tcf_block_get);
1414 
1415 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1416  * actions should be all removed after flushing.
1417  */
1418 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1419 		       struct tcf_block_ext_info *ei)
1420 {
1421 	if (!block)
1422 		return;
1423 	tcf_chain0_head_change_cb_del(block, ei);
1424 	tcf_block_owner_del(block, q, ei->binder_type);
1425 
1426 	__tcf_block_put(block, q, ei, true);
1427 }
1428 EXPORT_SYMBOL(tcf_block_put_ext);
1429 
1430 void tcf_block_put(struct tcf_block *block)
1431 {
1432 	struct tcf_block_ext_info ei = {0, };
1433 
1434 	if (!block)
1435 		return;
1436 	tcf_block_put_ext(block, block->q, &ei);
1437 }
1438 
1439 EXPORT_SYMBOL(tcf_block_put);
1440 
1441 static int
1442 tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1443 			    void *cb_priv, bool add, bool offload_in_use,
1444 			    struct netlink_ext_ack *extack)
1445 {
1446 	struct tcf_chain *chain, *chain_prev;
1447 	struct tcf_proto *tp, *tp_prev;
1448 	int err;
1449 
1450 	lockdep_assert_held(&block->cb_lock);
1451 
1452 	for (chain = __tcf_get_next_chain(block, NULL);
1453 	     chain;
1454 	     chain_prev = chain,
1455 		     chain = __tcf_get_next_chain(block, chain),
1456 		     tcf_chain_put(chain_prev)) {
1457 		for (tp = __tcf_get_next_proto(chain, NULL); tp;
1458 		     tp_prev = tp,
1459 			     tp = __tcf_get_next_proto(chain, tp),
1460 			     tcf_proto_put(tp_prev, true, NULL)) {
1461 			if (tp->ops->reoffload) {
1462 				err = tp->ops->reoffload(tp, add, cb, cb_priv,
1463 							 extack);
1464 				if (err && add)
1465 					goto err_playback_remove;
1466 			} else if (add && offload_in_use) {
1467 				err = -EOPNOTSUPP;
1468 				NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1469 				goto err_playback_remove;
1470 			}
1471 		}
1472 	}
1473 
1474 	return 0;
1475 
1476 err_playback_remove:
1477 	tcf_proto_put(tp, true, NULL);
1478 	tcf_chain_put(chain);
1479 	tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1480 				    extack);
1481 	return err;
1482 }
1483 
1484 static int tcf_block_bind(struct tcf_block *block,
1485 			  struct flow_block_offload *bo)
1486 {
1487 	struct flow_block_cb *block_cb, *next;
1488 	int err, i = 0;
1489 
1490 	lockdep_assert_held(&block->cb_lock);
1491 
1492 	list_for_each_entry(block_cb, &bo->cb_list, list) {
1493 		err = tcf_block_playback_offloads(block, block_cb->cb,
1494 						  block_cb->cb_priv, true,
1495 						  tcf_block_offload_in_use(block),
1496 						  bo->extack);
1497 		if (err)
1498 			goto err_unroll;
1499 		if (!bo->unlocked_driver_cb)
1500 			block->lockeddevcnt++;
1501 
1502 		i++;
1503 	}
1504 	list_splice(&bo->cb_list, &block->flow_block.cb_list);
1505 
1506 	return 0;
1507 
1508 err_unroll:
1509 	list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1510 		if (i-- > 0) {
1511 			list_del(&block_cb->list);
1512 			tcf_block_playback_offloads(block, block_cb->cb,
1513 						    block_cb->cb_priv, false,
1514 						    tcf_block_offload_in_use(block),
1515 						    NULL);
1516 			if (!bo->unlocked_driver_cb)
1517 				block->lockeddevcnt--;
1518 		}
1519 		flow_block_cb_free(block_cb);
1520 	}
1521 
1522 	return err;
1523 }
1524 
1525 static void tcf_block_unbind(struct tcf_block *block,
1526 			     struct flow_block_offload *bo)
1527 {
1528 	struct flow_block_cb *block_cb, *next;
1529 
1530 	lockdep_assert_held(&block->cb_lock);
1531 
1532 	list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1533 		tcf_block_playback_offloads(block, block_cb->cb,
1534 					    block_cb->cb_priv, false,
1535 					    tcf_block_offload_in_use(block),
1536 					    NULL);
1537 		list_del(&block_cb->list);
1538 		flow_block_cb_free(block_cb);
1539 		if (!bo->unlocked_driver_cb)
1540 			block->lockeddevcnt--;
1541 	}
1542 }
1543 
1544 static int tcf_block_setup(struct tcf_block *block,
1545 			   struct flow_block_offload *bo)
1546 {
1547 	int err;
1548 
1549 	switch (bo->command) {
1550 	case FLOW_BLOCK_BIND:
1551 		err = tcf_block_bind(block, bo);
1552 		break;
1553 	case FLOW_BLOCK_UNBIND:
1554 		err = 0;
1555 		tcf_block_unbind(block, bo);
1556 		break;
1557 	default:
1558 		WARN_ON_ONCE(1);
1559 		err = -EOPNOTSUPP;
1560 	}
1561 
1562 	return err;
1563 }
1564 
1565 /* Main classifier routine: scans classifier chain attached
1566  * to this qdisc, (optionally) tests for protocol and asks
1567  * specific classifiers.
1568  */
1569 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1570 		 struct tcf_result *res, bool compat_mode)
1571 {
1572 #ifdef CONFIG_NET_CLS_ACT
1573 	const int max_reclassify_loop = 4;
1574 	const struct tcf_proto *orig_tp = tp;
1575 	const struct tcf_proto *first_tp;
1576 	int limit = 0;
1577 
1578 reclassify:
1579 #endif
1580 	for (; tp; tp = rcu_dereference_bh(tp->next)) {
1581 		__be16 protocol = tc_skb_protocol(skb);
1582 		int err;
1583 
1584 		if (tp->protocol != protocol &&
1585 		    tp->protocol != htons(ETH_P_ALL))
1586 			continue;
1587 
1588 		err = tp->classify(skb, tp, res);
1589 #ifdef CONFIG_NET_CLS_ACT
1590 		if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1591 			first_tp = orig_tp;
1592 			goto reset;
1593 		} else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1594 			first_tp = res->goto_tp;
1595 
1596 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1597 			{
1598 				struct tc_skb_ext *ext;
1599 
1600 				ext = skb_ext_add(skb, TC_SKB_EXT);
1601 				if (WARN_ON_ONCE(!ext))
1602 					return TC_ACT_SHOT;
1603 
1604 				ext->chain = err & TC_ACT_EXT_VAL_MASK;
1605 			}
1606 #endif
1607 			goto reset;
1608 		}
1609 #endif
1610 		if (err >= 0)
1611 			return err;
1612 	}
1613 
1614 	return TC_ACT_UNSPEC; /* signal: continue lookup */
1615 #ifdef CONFIG_NET_CLS_ACT
1616 reset:
1617 	if (unlikely(limit++ >= max_reclassify_loop)) {
1618 		net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1619 				       tp->chain->block->index,
1620 				       tp->prio & 0xffff,
1621 				       ntohs(tp->protocol));
1622 		return TC_ACT_SHOT;
1623 	}
1624 
1625 	tp = first_tp;
1626 	goto reclassify;
1627 #endif
1628 }
1629 EXPORT_SYMBOL(tcf_classify);
1630 
1631 struct tcf_chain_info {
1632 	struct tcf_proto __rcu **pprev;
1633 	struct tcf_proto __rcu *next;
1634 };
1635 
1636 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1637 					   struct tcf_chain_info *chain_info)
1638 {
1639 	return tcf_chain_dereference(*chain_info->pprev, chain);
1640 }
1641 
1642 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1643 			       struct tcf_chain_info *chain_info,
1644 			       struct tcf_proto *tp)
1645 {
1646 	if (chain->flushing)
1647 		return -EAGAIN;
1648 
1649 	if (*chain_info->pprev == chain->filter_chain)
1650 		tcf_chain0_head_change(chain, tp);
1651 	tcf_proto_get(tp);
1652 	RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1653 	rcu_assign_pointer(*chain_info->pprev, tp);
1654 
1655 	return 0;
1656 }
1657 
1658 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1659 				struct tcf_chain_info *chain_info,
1660 				struct tcf_proto *tp)
1661 {
1662 	struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1663 
1664 	tcf_proto_mark_delete(tp);
1665 	if (tp == chain->filter_chain)
1666 		tcf_chain0_head_change(chain, next);
1667 	RCU_INIT_POINTER(*chain_info->pprev, next);
1668 }
1669 
1670 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1671 					   struct tcf_chain_info *chain_info,
1672 					   u32 protocol, u32 prio,
1673 					   bool prio_allocate);
1674 
1675 /* Try to insert new proto.
1676  * If proto with specified priority already exists, free new proto
1677  * and return existing one.
1678  */
1679 
1680 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1681 						    struct tcf_proto *tp_new,
1682 						    u32 protocol, u32 prio,
1683 						    bool rtnl_held)
1684 {
1685 	struct tcf_chain_info chain_info;
1686 	struct tcf_proto *tp;
1687 	int err = 0;
1688 
1689 	mutex_lock(&chain->filter_chain_lock);
1690 
1691 	if (tcf_proto_exists_destroying(chain, tp_new)) {
1692 		mutex_unlock(&chain->filter_chain_lock);
1693 		tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1694 		return ERR_PTR(-EAGAIN);
1695 	}
1696 
1697 	tp = tcf_chain_tp_find(chain, &chain_info,
1698 			       protocol, prio, false);
1699 	if (!tp)
1700 		err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1701 	mutex_unlock(&chain->filter_chain_lock);
1702 
1703 	if (tp) {
1704 		tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1705 		tp_new = tp;
1706 	} else if (err) {
1707 		tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1708 		tp_new = ERR_PTR(err);
1709 	}
1710 
1711 	return tp_new;
1712 }
1713 
1714 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1715 				      struct tcf_proto *tp, bool rtnl_held,
1716 				      struct netlink_ext_ack *extack)
1717 {
1718 	struct tcf_chain_info chain_info;
1719 	struct tcf_proto *tp_iter;
1720 	struct tcf_proto **pprev;
1721 	struct tcf_proto *next;
1722 
1723 	mutex_lock(&chain->filter_chain_lock);
1724 
1725 	/* Atomically find and remove tp from chain. */
1726 	for (pprev = &chain->filter_chain;
1727 	     (tp_iter = tcf_chain_dereference(*pprev, chain));
1728 	     pprev = &tp_iter->next) {
1729 		if (tp_iter == tp) {
1730 			chain_info.pprev = pprev;
1731 			chain_info.next = tp_iter->next;
1732 			WARN_ON(tp_iter->deleting);
1733 			break;
1734 		}
1735 	}
1736 	/* Verify that tp still exists and no new filters were inserted
1737 	 * concurrently.
1738 	 * Mark tp for deletion if it is empty.
1739 	 */
1740 	if (!tp_iter || !tcf_proto_check_delete(tp, rtnl_held)) {
1741 		mutex_unlock(&chain->filter_chain_lock);
1742 		return;
1743 	}
1744 
1745 	tcf_proto_signal_destroying(chain, tp);
1746 	next = tcf_chain_dereference(chain_info.next, chain);
1747 	if (tp == chain->filter_chain)
1748 		tcf_chain0_head_change(chain, next);
1749 	RCU_INIT_POINTER(*chain_info.pprev, next);
1750 	mutex_unlock(&chain->filter_chain_lock);
1751 
1752 	tcf_proto_put(tp, rtnl_held, extack);
1753 }
1754 
1755 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1756 					   struct tcf_chain_info *chain_info,
1757 					   u32 protocol, u32 prio,
1758 					   bool prio_allocate)
1759 {
1760 	struct tcf_proto **pprev;
1761 	struct tcf_proto *tp;
1762 
1763 	/* Check the chain for existence of proto-tcf with this priority */
1764 	for (pprev = &chain->filter_chain;
1765 	     (tp = tcf_chain_dereference(*pprev, chain));
1766 	     pprev = &tp->next) {
1767 		if (tp->prio >= prio) {
1768 			if (tp->prio == prio) {
1769 				if (prio_allocate ||
1770 				    (tp->protocol != protocol && protocol))
1771 					return ERR_PTR(-EINVAL);
1772 			} else {
1773 				tp = NULL;
1774 			}
1775 			break;
1776 		}
1777 	}
1778 	chain_info->pprev = pprev;
1779 	if (tp) {
1780 		chain_info->next = tp->next;
1781 		tcf_proto_get(tp);
1782 	} else {
1783 		chain_info->next = NULL;
1784 	}
1785 	return tp;
1786 }
1787 
1788 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1789 			 struct tcf_proto *tp, struct tcf_block *block,
1790 			 struct Qdisc *q, u32 parent, void *fh,
1791 			 u32 portid, u32 seq, u16 flags, int event,
1792 			 bool rtnl_held)
1793 {
1794 	struct tcmsg *tcm;
1795 	struct nlmsghdr  *nlh;
1796 	unsigned char *b = skb_tail_pointer(skb);
1797 
1798 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1799 	if (!nlh)
1800 		goto out_nlmsg_trim;
1801 	tcm = nlmsg_data(nlh);
1802 	tcm->tcm_family = AF_UNSPEC;
1803 	tcm->tcm__pad1 = 0;
1804 	tcm->tcm__pad2 = 0;
1805 	if (q) {
1806 		tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1807 		tcm->tcm_parent = parent;
1808 	} else {
1809 		tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1810 		tcm->tcm_block_index = block->index;
1811 	}
1812 	tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1813 	if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1814 		goto nla_put_failure;
1815 	if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1816 		goto nla_put_failure;
1817 	if (!fh) {
1818 		tcm->tcm_handle = 0;
1819 	} else {
1820 		if (tp->ops->dump &&
1821 		    tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
1822 			goto nla_put_failure;
1823 	}
1824 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1825 	return skb->len;
1826 
1827 out_nlmsg_trim:
1828 nla_put_failure:
1829 	nlmsg_trim(skb, b);
1830 	return -1;
1831 }
1832 
1833 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1834 			  struct nlmsghdr *n, struct tcf_proto *tp,
1835 			  struct tcf_block *block, struct Qdisc *q,
1836 			  u32 parent, void *fh, int event, bool unicast,
1837 			  bool rtnl_held)
1838 {
1839 	struct sk_buff *skb;
1840 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1841 	int err = 0;
1842 
1843 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1844 	if (!skb)
1845 		return -ENOBUFS;
1846 
1847 	if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1848 			  n->nlmsg_seq, n->nlmsg_flags, event,
1849 			  rtnl_held) <= 0) {
1850 		kfree_skb(skb);
1851 		return -EINVAL;
1852 	}
1853 
1854 	if (unicast)
1855 		err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1856 	else
1857 		err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1858 				     n->nlmsg_flags & NLM_F_ECHO);
1859 
1860 	if (err > 0)
1861 		err = 0;
1862 	return err;
1863 }
1864 
1865 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1866 			      struct nlmsghdr *n, struct tcf_proto *tp,
1867 			      struct tcf_block *block, struct Qdisc *q,
1868 			      u32 parent, void *fh, bool unicast, bool *last,
1869 			      bool rtnl_held, struct netlink_ext_ack *extack)
1870 {
1871 	struct sk_buff *skb;
1872 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1873 	int err;
1874 
1875 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1876 	if (!skb)
1877 		return -ENOBUFS;
1878 
1879 	if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1880 			  n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
1881 			  rtnl_held) <= 0) {
1882 		NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1883 		kfree_skb(skb);
1884 		return -EINVAL;
1885 	}
1886 
1887 	err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
1888 	if (err) {
1889 		kfree_skb(skb);
1890 		return err;
1891 	}
1892 
1893 	if (unicast)
1894 		err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1895 	else
1896 		err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1897 				     n->nlmsg_flags & NLM_F_ECHO);
1898 	if (err < 0)
1899 		NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1900 
1901 	if (err > 0)
1902 		err = 0;
1903 	return err;
1904 }
1905 
1906 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1907 				 struct tcf_block *block, struct Qdisc *q,
1908 				 u32 parent, struct nlmsghdr *n,
1909 				 struct tcf_chain *chain, int event,
1910 				 bool rtnl_held)
1911 {
1912 	struct tcf_proto *tp;
1913 
1914 	for (tp = tcf_get_next_proto(chain, NULL, rtnl_held);
1915 	     tp; tp = tcf_get_next_proto(chain, tp, rtnl_held))
1916 		tfilter_notify(net, oskb, n, tp, block,
1917 			       q, parent, NULL, event, false, rtnl_held);
1918 }
1919 
1920 static void tfilter_put(struct tcf_proto *tp, void *fh)
1921 {
1922 	if (tp->ops->put && fh)
1923 		tp->ops->put(tp, fh);
1924 }
1925 
1926 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1927 			  struct netlink_ext_ack *extack)
1928 {
1929 	struct net *net = sock_net(skb->sk);
1930 	struct nlattr *tca[TCA_MAX + 1];
1931 	char name[IFNAMSIZ];
1932 	struct tcmsg *t;
1933 	u32 protocol;
1934 	u32 prio;
1935 	bool prio_allocate;
1936 	u32 parent;
1937 	u32 chain_index;
1938 	struct Qdisc *q = NULL;
1939 	struct tcf_chain_info chain_info;
1940 	struct tcf_chain *chain = NULL;
1941 	struct tcf_block *block;
1942 	struct tcf_proto *tp;
1943 	unsigned long cl;
1944 	void *fh;
1945 	int err;
1946 	int tp_created;
1947 	bool rtnl_held = false;
1948 
1949 	if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1950 		return -EPERM;
1951 
1952 replay:
1953 	tp_created = 0;
1954 
1955 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
1956 				     rtm_tca_policy, extack);
1957 	if (err < 0)
1958 		return err;
1959 
1960 	t = nlmsg_data(n);
1961 	protocol = TC_H_MIN(t->tcm_info);
1962 	prio = TC_H_MAJ(t->tcm_info);
1963 	prio_allocate = false;
1964 	parent = t->tcm_parent;
1965 	tp = NULL;
1966 	cl = 0;
1967 	block = NULL;
1968 
1969 	if (prio == 0) {
1970 		/* If no priority is provided by the user,
1971 		 * we allocate one.
1972 		 */
1973 		if (n->nlmsg_flags & NLM_F_CREATE) {
1974 			prio = TC_H_MAKE(0x80000000U, 0U);
1975 			prio_allocate = true;
1976 		} else {
1977 			NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1978 			return -ENOENT;
1979 		}
1980 	}
1981 
1982 	/* Find head of filter chain. */
1983 
1984 	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
1985 	if (err)
1986 		return err;
1987 
1988 	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
1989 		NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
1990 		err = -EINVAL;
1991 		goto errout;
1992 	}
1993 
1994 	/* Take rtnl mutex if rtnl_held was set to true on previous iteration,
1995 	 * block is shared (no qdisc found), qdisc is not unlocked, classifier
1996 	 * type is not specified, classifier is not unlocked.
1997 	 */
1998 	if (rtnl_held ||
1999 	    (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2000 	    !tcf_proto_is_unlocked(name)) {
2001 		rtnl_held = true;
2002 		rtnl_lock();
2003 	}
2004 
2005 	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2006 	if (err)
2007 		goto errout;
2008 
2009 	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2010 				 extack);
2011 	if (IS_ERR(block)) {
2012 		err = PTR_ERR(block);
2013 		goto errout;
2014 	}
2015 
2016 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2017 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2018 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2019 		err = -EINVAL;
2020 		goto errout;
2021 	}
2022 	chain = tcf_chain_get(block, chain_index, true);
2023 	if (!chain) {
2024 		NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2025 		err = -ENOMEM;
2026 		goto errout;
2027 	}
2028 
2029 	mutex_lock(&chain->filter_chain_lock);
2030 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2031 			       prio, prio_allocate);
2032 	if (IS_ERR(tp)) {
2033 		NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2034 		err = PTR_ERR(tp);
2035 		goto errout_locked;
2036 	}
2037 
2038 	if (tp == NULL) {
2039 		struct tcf_proto *tp_new = NULL;
2040 
2041 		if (chain->flushing) {
2042 			err = -EAGAIN;
2043 			goto errout_locked;
2044 		}
2045 
2046 		/* Proto-tcf does not exist, create new one */
2047 
2048 		if (tca[TCA_KIND] == NULL || !protocol) {
2049 			NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2050 			err = -EINVAL;
2051 			goto errout_locked;
2052 		}
2053 
2054 		if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2055 			NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2056 			err = -ENOENT;
2057 			goto errout_locked;
2058 		}
2059 
2060 		if (prio_allocate)
2061 			prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2062 							       &chain_info));
2063 
2064 		mutex_unlock(&chain->filter_chain_lock);
2065 		tp_new = tcf_proto_create(nla_data(tca[TCA_KIND]),
2066 					  protocol, prio, chain, rtnl_held,
2067 					  extack);
2068 		if (IS_ERR(tp_new)) {
2069 			err = PTR_ERR(tp_new);
2070 			goto errout_tp;
2071 		}
2072 
2073 		tp_created = 1;
2074 		tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2075 						rtnl_held);
2076 		if (IS_ERR(tp)) {
2077 			err = PTR_ERR(tp);
2078 			goto errout_tp;
2079 		}
2080 	} else {
2081 		mutex_unlock(&chain->filter_chain_lock);
2082 	}
2083 
2084 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2085 		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2086 		err = -EINVAL;
2087 		goto errout;
2088 	}
2089 
2090 	fh = tp->ops->get(tp, t->tcm_handle);
2091 
2092 	if (!fh) {
2093 		if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2094 			NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2095 			err = -ENOENT;
2096 			goto errout;
2097 		}
2098 	} else if (n->nlmsg_flags & NLM_F_EXCL) {
2099 		tfilter_put(tp, fh);
2100 		NL_SET_ERR_MSG(extack, "Filter already exists");
2101 		err = -EEXIST;
2102 		goto errout;
2103 	}
2104 
2105 	if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2106 		NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2107 		err = -EINVAL;
2108 		goto errout;
2109 	}
2110 
2111 	err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2112 			      n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
2113 			      rtnl_held, extack);
2114 	if (err == 0) {
2115 		tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2116 			       RTM_NEWTFILTER, false, rtnl_held);
2117 		tfilter_put(tp, fh);
2118 		/* q pointer is NULL for shared blocks */
2119 		if (q)
2120 			q->flags &= ~TCQ_F_CAN_BYPASS;
2121 	}
2122 
2123 errout:
2124 	if (err && tp_created)
2125 		tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2126 errout_tp:
2127 	if (chain) {
2128 		if (tp && !IS_ERR(tp))
2129 			tcf_proto_put(tp, rtnl_held, NULL);
2130 		if (!tp_created)
2131 			tcf_chain_put(chain);
2132 	}
2133 	tcf_block_release(q, block, rtnl_held);
2134 
2135 	if (rtnl_held)
2136 		rtnl_unlock();
2137 
2138 	if (err == -EAGAIN) {
2139 		/* Take rtnl lock in case EAGAIN is caused by concurrent flush
2140 		 * of target chain.
2141 		 */
2142 		rtnl_held = true;
2143 		/* Replay the request. */
2144 		goto replay;
2145 	}
2146 	return err;
2147 
2148 errout_locked:
2149 	mutex_unlock(&chain->filter_chain_lock);
2150 	goto errout;
2151 }
2152 
2153 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2154 			  struct netlink_ext_ack *extack)
2155 {
2156 	struct net *net = sock_net(skb->sk);
2157 	struct nlattr *tca[TCA_MAX + 1];
2158 	char name[IFNAMSIZ];
2159 	struct tcmsg *t;
2160 	u32 protocol;
2161 	u32 prio;
2162 	u32 parent;
2163 	u32 chain_index;
2164 	struct Qdisc *q = NULL;
2165 	struct tcf_chain_info chain_info;
2166 	struct tcf_chain *chain = NULL;
2167 	struct tcf_block *block = NULL;
2168 	struct tcf_proto *tp = NULL;
2169 	unsigned long cl = 0;
2170 	void *fh = NULL;
2171 	int err;
2172 	bool rtnl_held = false;
2173 
2174 	if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2175 		return -EPERM;
2176 
2177 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2178 				     rtm_tca_policy, extack);
2179 	if (err < 0)
2180 		return err;
2181 
2182 	t = nlmsg_data(n);
2183 	protocol = TC_H_MIN(t->tcm_info);
2184 	prio = TC_H_MAJ(t->tcm_info);
2185 	parent = t->tcm_parent;
2186 
2187 	if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2188 		NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2189 		return -ENOENT;
2190 	}
2191 
2192 	/* Find head of filter chain. */
2193 
2194 	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2195 	if (err)
2196 		return err;
2197 
2198 	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2199 		NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2200 		err = -EINVAL;
2201 		goto errout;
2202 	}
2203 	/* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2204 	 * found), qdisc is not unlocked, classifier type is not specified,
2205 	 * classifier is not unlocked.
2206 	 */
2207 	if (!prio ||
2208 	    (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2209 	    !tcf_proto_is_unlocked(name)) {
2210 		rtnl_held = true;
2211 		rtnl_lock();
2212 	}
2213 
2214 	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2215 	if (err)
2216 		goto errout;
2217 
2218 	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2219 				 extack);
2220 	if (IS_ERR(block)) {
2221 		err = PTR_ERR(block);
2222 		goto errout;
2223 	}
2224 
2225 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2226 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2227 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2228 		err = -EINVAL;
2229 		goto errout;
2230 	}
2231 	chain = tcf_chain_get(block, chain_index, false);
2232 	if (!chain) {
2233 		/* User requested flush on non-existent chain. Nothing to do,
2234 		 * so just return success.
2235 		 */
2236 		if (prio == 0) {
2237 			err = 0;
2238 			goto errout;
2239 		}
2240 		NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2241 		err = -ENOENT;
2242 		goto errout;
2243 	}
2244 
2245 	if (prio == 0) {
2246 		tfilter_notify_chain(net, skb, block, q, parent, n,
2247 				     chain, RTM_DELTFILTER, rtnl_held);
2248 		tcf_chain_flush(chain, rtnl_held);
2249 		err = 0;
2250 		goto errout;
2251 	}
2252 
2253 	mutex_lock(&chain->filter_chain_lock);
2254 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2255 			       prio, false);
2256 	if (!tp || IS_ERR(tp)) {
2257 		NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2258 		err = tp ? PTR_ERR(tp) : -ENOENT;
2259 		goto errout_locked;
2260 	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2261 		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2262 		err = -EINVAL;
2263 		goto errout_locked;
2264 	} else if (t->tcm_handle == 0) {
2265 		tcf_proto_signal_destroying(chain, tp);
2266 		tcf_chain_tp_remove(chain, &chain_info, tp);
2267 		mutex_unlock(&chain->filter_chain_lock);
2268 
2269 		tcf_proto_put(tp, rtnl_held, NULL);
2270 		tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2271 			       RTM_DELTFILTER, false, rtnl_held);
2272 		err = 0;
2273 		goto errout;
2274 	}
2275 	mutex_unlock(&chain->filter_chain_lock);
2276 
2277 	fh = tp->ops->get(tp, t->tcm_handle);
2278 
2279 	if (!fh) {
2280 		NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2281 		err = -ENOENT;
2282 	} else {
2283 		bool last;
2284 
2285 		err = tfilter_del_notify(net, skb, n, tp, block,
2286 					 q, parent, fh, false, &last,
2287 					 rtnl_held, extack);
2288 
2289 		if (err)
2290 			goto errout;
2291 		if (last)
2292 			tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2293 	}
2294 
2295 errout:
2296 	if (chain) {
2297 		if (tp && !IS_ERR(tp))
2298 			tcf_proto_put(tp, rtnl_held, NULL);
2299 		tcf_chain_put(chain);
2300 	}
2301 	tcf_block_release(q, block, rtnl_held);
2302 
2303 	if (rtnl_held)
2304 		rtnl_unlock();
2305 
2306 	return err;
2307 
2308 errout_locked:
2309 	mutex_unlock(&chain->filter_chain_lock);
2310 	goto errout;
2311 }
2312 
2313 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2314 			  struct netlink_ext_ack *extack)
2315 {
2316 	struct net *net = sock_net(skb->sk);
2317 	struct nlattr *tca[TCA_MAX + 1];
2318 	char name[IFNAMSIZ];
2319 	struct tcmsg *t;
2320 	u32 protocol;
2321 	u32 prio;
2322 	u32 parent;
2323 	u32 chain_index;
2324 	struct Qdisc *q = NULL;
2325 	struct tcf_chain_info chain_info;
2326 	struct tcf_chain *chain = NULL;
2327 	struct tcf_block *block = NULL;
2328 	struct tcf_proto *tp = NULL;
2329 	unsigned long cl = 0;
2330 	void *fh = NULL;
2331 	int err;
2332 	bool rtnl_held = false;
2333 
2334 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2335 				     rtm_tca_policy, extack);
2336 	if (err < 0)
2337 		return err;
2338 
2339 	t = nlmsg_data(n);
2340 	protocol = TC_H_MIN(t->tcm_info);
2341 	prio = TC_H_MAJ(t->tcm_info);
2342 	parent = t->tcm_parent;
2343 
2344 	if (prio == 0) {
2345 		NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2346 		return -ENOENT;
2347 	}
2348 
2349 	/* Find head of filter chain. */
2350 
2351 	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2352 	if (err)
2353 		return err;
2354 
2355 	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2356 		NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2357 		err = -EINVAL;
2358 		goto errout;
2359 	}
2360 	/* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2361 	 * unlocked, classifier type is not specified, classifier is not
2362 	 * unlocked.
2363 	 */
2364 	if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2365 	    !tcf_proto_is_unlocked(name)) {
2366 		rtnl_held = true;
2367 		rtnl_lock();
2368 	}
2369 
2370 	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2371 	if (err)
2372 		goto errout;
2373 
2374 	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2375 				 extack);
2376 	if (IS_ERR(block)) {
2377 		err = PTR_ERR(block);
2378 		goto errout;
2379 	}
2380 
2381 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2382 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2383 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2384 		err = -EINVAL;
2385 		goto errout;
2386 	}
2387 	chain = tcf_chain_get(block, chain_index, false);
2388 	if (!chain) {
2389 		NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2390 		err = -EINVAL;
2391 		goto errout;
2392 	}
2393 
2394 	mutex_lock(&chain->filter_chain_lock);
2395 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2396 			       prio, false);
2397 	mutex_unlock(&chain->filter_chain_lock);
2398 	if (!tp || IS_ERR(tp)) {
2399 		NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2400 		err = tp ? PTR_ERR(tp) : -ENOENT;
2401 		goto errout;
2402 	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2403 		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2404 		err = -EINVAL;
2405 		goto errout;
2406 	}
2407 
2408 	fh = tp->ops->get(tp, t->tcm_handle);
2409 
2410 	if (!fh) {
2411 		NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2412 		err = -ENOENT;
2413 	} else {
2414 		err = tfilter_notify(net, skb, n, tp, block, q, parent,
2415 				     fh, RTM_NEWTFILTER, true, rtnl_held);
2416 		if (err < 0)
2417 			NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2418 	}
2419 
2420 	tfilter_put(tp, fh);
2421 errout:
2422 	if (chain) {
2423 		if (tp && !IS_ERR(tp))
2424 			tcf_proto_put(tp, rtnl_held, NULL);
2425 		tcf_chain_put(chain);
2426 	}
2427 	tcf_block_release(q, block, rtnl_held);
2428 
2429 	if (rtnl_held)
2430 		rtnl_unlock();
2431 
2432 	return err;
2433 }
2434 
2435 struct tcf_dump_args {
2436 	struct tcf_walker w;
2437 	struct sk_buff *skb;
2438 	struct netlink_callback *cb;
2439 	struct tcf_block *block;
2440 	struct Qdisc *q;
2441 	u32 parent;
2442 };
2443 
2444 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2445 {
2446 	struct tcf_dump_args *a = (void *)arg;
2447 	struct net *net = sock_net(a->skb->sk);
2448 
2449 	return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2450 			     n, NETLINK_CB(a->cb->skb).portid,
2451 			     a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2452 			     RTM_NEWTFILTER, true);
2453 }
2454 
2455 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2456 			   struct sk_buff *skb, struct netlink_callback *cb,
2457 			   long index_start, long *p_index)
2458 {
2459 	struct net *net = sock_net(skb->sk);
2460 	struct tcf_block *block = chain->block;
2461 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2462 	struct tcf_proto *tp, *tp_prev;
2463 	struct tcf_dump_args arg;
2464 
2465 	for (tp = __tcf_get_next_proto(chain, NULL);
2466 	     tp;
2467 	     tp_prev = tp,
2468 		     tp = __tcf_get_next_proto(chain, tp),
2469 		     tcf_proto_put(tp_prev, true, NULL),
2470 		     (*p_index)++) {
2471 		if (*p_index < index_start)
2472 			continue;
2473 		if (TC_H_MAJ(tcm->tcm_info) &&
2474 		    TC_H_MAJ(tcm->tcm_info) != tp->prio)
2475 			continue;
2476 		if (TC_H_MIN(tcm->tcm_info) &&
2477 		    TC_H_MIN(tcm->tcm_info) != tp->protocol)
2478 			continue;
2479 		if (*p_index > index_start)
2480 			memset(&cb->args[1], 0,
2481 			       sizeof(cb->args) - sizeof(cb->args[0]));
2482 		if (cb->args[1] == 0) {
2483 			if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2484 					  NETLINK_CB(cb->skb).portid,
2485 					  cb->nlh->nlmsg_seq, NLM_F_MULTI,
2486 					  RTM_NEWTFILTER, true) <= 0)
2487 				goto errout;
2488 			cb->args[1] = 1;
2489 		}
2490 		if (!tp->ops->walk)
2491 			continue;
2492 		arg.w.fn = tcf_node_dump;
2493 		arg.skb = skb;
2494 		arg.cb = cb;
2495 		arg.block = block;
2496 		arg.q = q;
2497 		arg.parent = parent;
2498 		arg.w.stop = 0;
2499 		arg.w.skip = cb->args[1] - 1;
2500 		arg.w.count = 0;
2501 		arg.w.cookie = cb->args[2];
2502 		tp->ops->walk(tp, &arg.w, true);
2503 		cb->args[2] = arg.w.cookie;
2504 		cb->args[1] = arg.w.count + 1;
2505 		if (arg.w.stop)
2506 			goto errout;
2507 	}
2508 	return true;
2509 
2510 errout:
2511 	tcf_proto_put(tp, true, NULL);
2512 	return false;
2513 }
2514 
2515 /* called with RTNL */
2516 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2517 {
2518 	struct tcf_chain *chain, *chain_prev;
2519 	struct net *net = sock_net(skb->sk);
2520 	struct nlattr *tca[TCA_MAX + 1];
2521 	struct Qdisc *q = NULL;
2522 	struct tcf_block *block;
2523 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2524 	long index_start;
2525 	long index;
2526 	u32 parent;
2527 	int err;
2528 
2529 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2530 		return skb->len;
2531 
2532 	err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2533 				     NULL, cb->extack);
2534 	if (err)
2535 		return err;
2536 
2537 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2538 		block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2539 		if (!block)
2540 			goto out;
2541 		/* If we work with block index, q is NULL and parent value
2542 		 * will never be used in the following code. The check
2543 		 * in tcf_fill_node prevents it. However, compiler does not
2544 		 * see that far, so set parent to zero to silence the warning
2545 		 * about parent being uninitialized.
2546 		 */
2547 		parent = 0;
2548 	} else {
2549 		const struct Qdisc_class_ops *cops;
2550 		struct net_device *dev;
2551 		unsigned long cl = 0;
2552 
2553 		dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2554 		if (!dev)
2555 			return skb->len;
2556 
2557 		parent = tcm->tcm_parent;
2558 		if (!parent) {
2559 			q = dev->qdisc;
2560 			parent = q->handle;
2561 		} else {
2562 			q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2563 		}
2564 		if (!q)
2565 			goto out;
2566 		cops = q->ops->cl_ops;
2567 		if (!cops)
2568 			goto out;
2569 		if (!cops->tcf_block)
2570 			goto out;
2571 		if (TC_H_MIN(tcm->tcm_parent)) {
2572 			cl = cops->find(q, tcm->tcm_parent);
2573 			if (cl == 0)
2574 				goto out;
2575 		}
2576 		block = cops->tcf_block(q, cl, NULL);
2577 		if (!block)
2578 			goto out;
2579 		if (tcf_block_shared(block))
2580 			q = NULL;
2581 	}
2582 
2583 	index_start = cb->args[0];
2584 	index = 0;
2585 
2586 	for (chain = __tcf_get_next_chain(block, NULL);
2587 	     chain;
2588 	     chain_prev = chain,
2589 		     chain = __tcf_get_next_chain(block, chain),
2590 		     tcf_chain_put(chain_prev)) {
2591 		if (tca[TCA_CHAIN] &&
2592 		    nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2593 			continue;
2594 		if (!tcf_chain_dump(chain, q, parent, skb, cb,
2595 				    index_start, &index)) {
2596 			tcf_chain_put(chain);
2597 			err = -EMSGSIZE;
2598 			break;
2599 		}
2600 	}
2601 
2602 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2603 		tcf_block_refcnt_put(block, true);
2604 	cb->args[0] = index;
2605 
2606 out:
2607 	/* If we did no progress, the error (EMSGSIZE) is real */
2608 	if (skb->len == 0 && err)
2609 		return err;
2610 	return skb->len;
2611 }
2612 
2613 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2614 			      void *tmplt_priv, u32 chain_index,
2615 			      struct net *net, struct sk_buff *skb,
2616 			      struct tcf_block *block,
2617 			      u32 portid, u32 seq, u16 flags, int event)
2618 {
2619 	unsigned char *b = skb_tail_pointer(skb);
2620 	const struct tcf_proto_ops *ops;
2621 	struct nlmsghdr *nlh;
2622 	struct tcmsg *tcm;
2623 	void *priv;
2624 
2625 	ops = tmplt_ops;
2626 	priv = tmplt_priv;
2627 
2628 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2629 	if (!nlh)
2630 		goto out_nlmsg_trim;
2631 	tcm = nlmsg_data(nlh);
2632 	tcm->tcm_family = AF_UNSPEC;
2633 	tcm->tcm__pad1 = 0;
2634 	tcm->tcm__pad2 = 0;
2635 	tcm->tcm_handle = 0;
2636 	if (block->q) {
2637 		tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2638 		tcm->tcm_parent = block->q->handle;
2639 	} else {
2640 		tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2641 		tcm->tcm_block_index = block->index;
2642 	}
2643 
2644 	if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2645 		goto nla_put_failure;
2646 
2647 	if (ops) {
2648 		if (nla_put_string(skb, TCA_KIND, ops->kind))
2649 			goto nla_put_failure;
2650 		if (ops->tmplt_dump(skb, net, priv) < 0)
2651 			goto nla_put_failure;
2652 	}
2653 
2654 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2655 	return skb->len;
2656 
2657 out_nlmsg_trim:
2658 nla_put_failure:
2659 	nlmsg_trim(skb, b);
2660 	return -EMSGSIZE;
2661 }
2662 
2663 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2664 			   u32 seq, u16 flags, int event, bool unicast)
2665 {
2666 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2667 	struct tcf_block *block = chain->block;
2668 	struct net *net = block->net;
2669 	struct sk_buff *skb;
2670 	int err = 0;
2671 
2672 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2673 	if (!skb)
2674 		return -ENOBUFS;
2675 
2676 	if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2677 			       chain->index, net, skb, block, portid,
2678 			       seq, flags, event) <= 0) {
2679 		kfree_skb(skb);
2680 		return -EINVAL;
2681 	}
2682 
2683 	if (unicast)
2684 		err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2685 	else
2686 		err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2687 				     flags & NLM_F_ECHO);
2688 
2689 	if (err > 0)
2690 		err = 0;
2691 	return err;
2692 }
2693 
2694 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2695 				  void *tmplt_priv, u32 chain_index,
2696 				  struct tcf_block *block, struct sk_buff *oskb,
2697 				  u32 seq, u16 flags, bool unicast)
2698 {
2699 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2700 	struct net *net = block->net;
2701 	struct sk_buff *skb;
2702 
2703 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2704 	if (!skb)
2705 		return -ENOBUFS;
2706 
2707 	if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2708 			       block, portid, seq, flags, RTM_DELCHAIN) <= 0) {
2709 		kfree_skb(skb);
2710 		return -EINVAL;
2711 	}
2712 
2713 	if (unicast)
2714 		return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2715 
2716 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2717 }
2718 
2719 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2720 			      struct nlattr **tca,
2721 			      struct netlink_ext_ack *extack)
2722 {
2723 	const struct tcf_proto_ops *ops;
2724 	void *tmplt_priv;
2725 
2726 	/* If kind is not set, user did not specify template. */
2727 	if (!tca[TCA_KIND])
2728 		return 0;
2729 
2730 	ops = tcf_proto_lookup_ops(nla_data(tca[TCA_KIND]), true, extack);
2731 	if (IS_ERR(ops))
2732 		return PTR_ERR(ops);
2733 	if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2734 		NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2735 		return -EOPNOTSUPP;
2736 	}
2737 
2738 	tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2739 	if (IS_ERR(tmplt_priv)) {
2740 		module_put(ops->owner);
2741 		return PTR_ERR(tmplt_priv);
2742 	}
2743 	chain->tmplt_ops = ops;
2744 	chain->tmplt_priv = tmplt_priv;
2745 	return 0;
2746 }
2747 
2748 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2749 			       void *tmplt_priv)
2750 {
2751 	/* If template ops are set, no work to do for us. */
2752 	if (!tmplt_ops)
2753 		return;
2754 
2755 	tmplt_ops->tmplt_destroy(tmplt_priv);
2756 	module_put(tmplt_ops->owner);
2757 }
2758 
2759 /* Add/delete/get a chain */
2760 
2761 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2762 			struct netlink_ext_ack *extack)
2763 {
2764 	struct net *net = sock_net(skb->sk);
2765 	struct nlattr *tca[TCA_MAX + 1];
2766 	struct tcmsg *t;
2767 	u32 parent;
2768 	u32 chain_index;
2769 	struct Qdisc *q = NULL;
2770 	struct tcf_chain *chain = NULL;
2771 	struct tcf_block *block;
2772 	unsigned long cl;
2773 	int err;
2774 
2775 	if (n->nlmsg_type != RTM_GETCHAIN &&
2776 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2777 		return -EPERM;
2778 
2779 replay:
2780 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2781 				     rtm_tca_policy, extack);
2782 	if (err < 0)
2783 		return err;
2784 
2785 	t = nlmsg_data(n);
2786 	parent = t->tcm_parent;
2787 	cl = 0;
2788 
2789 	block = tcf_block_find(net, &q, &parent, &cl,
2790 			       t->tcm_ifindex, t->tcm_block_index, extack);
2791 	if (IS_ERR(block))
2792 		return PTR_ERR(block);
2793 
2794 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2795 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2796 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2797 		err = -EINVAL;
2798 		goto errout_block;
2799 	}
2800 
2801 	mutex_lock(&block->lock);
2802 	chain = tcf_chain_lookup(block, chain_index);
2803 	if (n->nlmsg_type == RTM_NEWCHAIN) {
2804 		if (chain) {
2805 			if (tcf_chain_held_by_acts_only(chain)) {
2806 				/* The chain exists only because there is
2807 				 * some action referencing it.
2808 				 */
2809 				tcf_chain_hold(chain);
2810 			} else {
2811 				NL_SET_ERR_MSG(extack, "Filter chain already exists");
2812 				err = -EEXIST;
2813 				goto errout_block_locked;
2814 			}
2815 		} else {
2816 			if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2817 				NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
2818 				err = -ENOENT;
2819 				goto errout_block_locked;
2820 			}
2821 			chain = tcf_chain_create(block, chain_index);
2822 			if (!chain) {
2823 				NL_SET_ERR_MSG(extack, "Failed to create filter chain");
2824 				err = -ENOMEM;
2825 				goto errout_block_locked;
2826 			}
2827 		}
2828 	} else {
2829 		if (!chain || tcf_chain_held_by_acts_only(chain)) {
2830 			NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2831 			err = -EINVAL;
2832 			goto errout_block_locked;
2833 		}
2834 		tcf_chain_hold(chain);
2835 	}
2836 
2837 	if (n->nlmsg_type == RTM_NEWCHAIN) {
2838 		/* Modifying chain requires holding parent block lock. In case
2839 		 * the chain was successfully added, take a reference to the
2840 		 * chain. This ensures that an empty chain does not disappear at
2841 		 * the end of this function.
2842 		 */
2843 		tcf_chain_hold(chain);
2844 		chain->explicitly_created = true;
2845 	}
2846 	mutex_unlock(&block->lock);
2847 
2848 	switch (n->nlmsg_type) {
2849 	case RTM_NEWCHAIN:
2850 		err = tc_chain_tmplt_add(chain, net, tca, extack);
2851 		if (err) {
2852 			tcf_chain_put_explicitly_created(chain);
2853 			goto errout;
2854 		}
2855 
2856 		tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2857 				RTM_NEWCHAIN, false);
2858 		break;
2859 	case RTM_DELCHAIN:
2860 		tfilter_notify_chain(net, skb, block, q, parent, n,
2861 				     chain, RTM_DELTFILTER, true);
2862 		/* Flush the chain first as the user requested chain removal. */
2863 		tcf_chain_flush(chain, true);
2864 		/* In case the chain was successfully deleted, put a reference
2865 		 * to the chain previously taken during addition.
2866 		 */
2867 		tcf_chain_put_explicitly_created(chain);
2868 		break;
2869 	case RTM_GETCHAIN:
2870 		err = tc_chain_notify(chain, skb, n->nlmsg_seq,
2871 				      n->nlmsg_seq, n->nlmsg_type, true);
2872 		if (err < 0)
2873 			NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2874 		break;
2875 	default:
2876 		err = -EOPNOTSUPP;
2877 		NL_SET_ERR_MSG(extack, "Unsupported message type");
2878 		goto errout;
2879 	}
2880 
2881 errout:
2882 	tcf_chain_put(chain);
2883 errout_block:
2884 	tcf_block_release(q, block, true);
2885 	if (err == -EAGAIN)
2886 		/* Replay the request. */
2887 		goto replay;
2888 	return err;
2889 
2890 errout_block_locked:
2891 	mutex_unlock(&block->lock);
2892 	goto errout_block;
2893 }
2894 
2895 /* called with RTNL */
2896 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2897 {
2898 	struct net *net = sock_net(skb->sk);
2899 	struct nlattr *tca[TCA_MAX + 1];
2900 	struct Qdisc *q = NULL;
2901 	struct tcf_block *block;
2902 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2903 	struct tcf_chain *chain;
2904 	long index_start;
2905 	long index;
2906 	u32 parent;
2907 	int err;
2908 
2909 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2910 		return skb->len;
2911 
2912 	err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2913 				     rtm_tca_policy, cb->extack);
2914 	if (err)
2915 		return err;
2916 
2917 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2918 		block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2919 		if (!block)
2920 			goto out;
2921 		/* If we work with block index, q is NULL and parent value
2922 		 * will never be used in the following code. The check
2923 		 * in tcf_fill_node prevents it. However, compiler does not
2924 		 * see that far, so set parent to zero to silence the warning
2925 		 * about parent being uninitialized.
2926 		 */
2927 		parent = 0;
2928 	} else {
2929 		const struct Qdisc_class_ops *cops;
2930 		struct net_device *dev;
2931 		unsigned long cl = 0;
2932 
2933 		dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2934 		if (!dev)
2935 			return skb->len;
2936 
2937 		parent = tcm->tcm_parent;
2938 		if (!parent) {
2939 			q = dev->qdisc;
2940 			parent = q->handle;
2941 		} else {
2942 			q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2943 		}
2944 		if (!q)
2945 			goto out;
2946 		cops = q->ops->cl_ops;
2947 		if (!cops)
2948 			goto out;
2949 		if (!cops->tcf_block)
2950 			goto out;
2951 		if (TC_H_MIN(tcm->tcm_parent)) {
2952 			cl = cops->find(q, tcm->tcm_parent);
2953 			if (cl == 0)
2954 				goto out;
2955 		}
2956 		block = cops->tcf_block(q, cl, NULL);
2957 		if (!block)
2958 			goto out;
2959 		if (tcf_block_shared(block))
2960 			q = NULL;
2961 	}
2962 
2963 	index_start = cb->args[0];
2964 	index = 0;
2965 
2966 	mutex_lock(&block->lock);
2967 	list_for_each_entry(chain, &block->chain_list, list) {
2968 		if ((tca[TCA_CHAIN] &&
2969 		     nla_get_u32(tca[TCA_CHAIN]) != chain->index))
2970 			continue;
2971 		if (index < index_start) {
2972 			index++;
2973 			continue;
2974 		}
2975 		if (tcf_chain_held_by_acts_only(chain))
2976 			continue;
2977 		err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2978 					 chain->index, net, skb, block,
2979 					 NETLINK_CB(cb->skb).portid,
2980 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2981 					 RTM_NEWCHAIN);
2982 		if (err <= 0)
2983 			break;
2984 		index++;
2985 	}
2986 	mutex_unlock(&block->lock);
2987 
2988 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2989 		tcf_block_refcnt_put(block, true);
2990 	cb->args[0] = index;
2991 
2992 out:
2993 	/* If we did no progress, the error (EMSGSIZE) is real */
2994 	if (skb->len == 0 && err)
2995 		return err;
2996 	return skb->len;
2997 }
2998 
2999 void tcf_exts_destroy(struct tcf_exts *exts)
3000 {
3001 #ifdef CONFIG_NET_CLS_ACT
3002 	if (exts->actions) {
3003 		tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3004 		kfree(exts->actions);
3005 	}
3006 	exts->nr_actions = 0;
3007 #endif
3008 }
3009 EXPORT_SYMBOL(tcf_exts_destroy);
3010 
3011 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3012 		      struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
3013 		      bool rtnl_held, struct netlink_ext_ack *extack)
3014 {
3015 #ifdef CONFIG_NET_CLS_ACT
3016 	{
3017 		struct tc_action *act;
3018 		size_t attr_size = 0;
3019 
3020 		if (exts->police && tb[exts->police]) {
3021 			act = tcf_action_init_1(net, tp, tb[exts->police],
3022 						rate_tlv, "police", ovr,
3023 						TCA_ACT_BIND, rtnl_held,
3024 						extack);
3025 			if (IS_ERR(act))
3026 				return PTR_ERR(act);
3027 
3028 			act->type = exts->type = TCA_OLD_COMPAT;
3029 			exts->actions[0] = act;
3030 			exts->nr_actions = 1;
3031 		} else if (exts->action && tb[exts->action]) {
3032 			int err;
3033 
3034 			err = tcf_action_init(net, tp, tb[exts->action],
3035 					      rate_tlv, NULL, ovr, TCA_ACT_BIND,
3036 					      exts->actions, &attr_size,
3037 					      rtnl_held, extack);
3038 			if (err < 0)
3039 				return err;
3040 			exts->nr_actions = err;
3041 		}
3042 	}
3043 #else
3044 	if ((exts->action && tb[exts->action]) ||
3045 	    (exts->police && tb[exts->police])) {
3046 		NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3047 		return -EOPNOTSUPP;
3048 	}
3049 #endif
3050 
3051 	return 0;
3052 }
3053 EXPORT_SYMBOL(tcf_exts_validate);
3054 
3055 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3056 {
3057 #ifdef CONFIG_NET_CLS_ACT
3058 	struct tcf_exts old = *dst;
3059 
3060 	*dst = *src;
3061 	tcf_exts_destroy(&old);
3062 #endif
3063 }
3064 EXPORT_SYMBOL(tcf_exts_change);
3065 
3066 #ifdef CONFIG_NET_CLS_ACT
3067 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3068 {
3069 	if (exts->nr_actions == 0)
3070 		return NULL;
3071 	else
3072 		return exts->actions[0];
3073 }
3074 #endif
3075 
3076 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3077 {
3078 #ifdef CONFIG_NET_CLS_ACT
3079 	struct nlattr *nest;
3080 
3081 	if (exts->action && tcf_exts_has_actions(exts)) {
3082 		/*
3083 		 * again for backward compatible mode - we want
3084 		 * to work with both old and new modes of entering
3085 		 * tc data even if iproute2  was newer - jhs
3086 		 */
3087 		if (exts->type != TCA_OLD_COMPAT) {
3088 			nest = nla_nest_start_noflag(skb, exts->action);
3089 			if (nest == NULL)
3090 				goto nla_put_failure;
3091 
3092 			if (tcf_action_dump(skb, exts->actions, 0, 0) < 0)
3093 				goto nla_put_failure;
3094 			nla_nest_end(skb, nest);
3095 		} else if (exts->police) {
3096 			struct tc_action *act = tcf_exts_first_act(exts);
3097 			nest = nla_nest_start_noflag(skb, exts->police);
3098 			if (nest == NULL || !act)
3099 				goto nla_put_failure;
3100 			if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3101 				goto nla_put_failure;
3102 			nla_nest_end(skb, nest);
3103 		}
3104 	}
3105 	return 0;
3106 
3107 nla_put_failure:
3108 	nla_nest_cancel(skb, nest);
3109 	return -1;
3110 #else
3111 	return 0;
3112 #endif
3113 }
3114 EXPORT_SYMBOL(tcf_exts_dump);
3115 
3116 
3117 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3118 {
3119 #ifdef CONFIG_NET_CLS_ACT
3120 	struct tc_action *a = tcf_exts_first_act(exts);
3121 	if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3122 		return -1;
3123 #endif
3124 	return 0;
3125 }
3126 EXPORT_SYMBOL(tcf_exts_dump_stats);
3127 
3128 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3129 {
3130 	if (*flags & TCA_CLS_FLAGS_IN_HW)
3131 		return;
3132 	*flags |= TCA_CLS_FLAGS_IN_HW;
3133 	atomic_inc(&block->offloadcnt);
3134 }
3135 
3136 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3137 {
3138 	if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3139 		return;
3140 	*flags &= ~TCA_CLS_FLAGS_IN_HW;
3141 	atomic_dec(&block->offloadcnt);
3142 }
3143 
3144 static void tc_cls_offload_cnt_update(struct tcf_block *block,
3145 				      struct tcf_proto *tp, u32 *cnt,
3146 				      u32 *flags, u32 diff, bool add)
3147 {
3148 	lockdep_assert_held(&block->cb_lock);
3149 
3150 	spin_lock(&tp->lock);
3151 	if (add) {
3152 		if (!*cnt)
3153 			tcf_block_offload_inc(block, flags);
3154 		*cnt += diff;
3155 	} else {
3156 		*cnt -= diff;
3157 		if (!*cnt)
3158 			tcf_block_offload_dec(block, flags);
3159 	}
3160 	spin_unlock(&tp->lock);
3161 }
3162 
3163 static void
3164 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3165 			 u32 *cnt, u32 *flags)
3166 {
3167 	lockdep_assert_held(&block->cb_lock);
3168 
3169 	spin_lock(&tp->lock);
3170 	tcf_block_offload_dec(block, flags);
3171 	*cnt = 0;
3172 	spin_unlock(&tp->lock);
3173 }
3174 
3175 static int
3176 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3177 		   void *type_data, bool err_stop)
3178 {
3179 	struct flow_block_cb *block_cb;
3180 	int ok_count = 0;
3181 	int err;
3182 
3183 	list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3184 		err = block_cb->cb(type, type_data, block_cb->cb_priv);
3185 		if (err) {
3186 			if (err_stop)
3187 				return err;
3188 		} else {
3189 			ok_count++;
3190 		}
3191 	}
3192 	return ok_count;
3193 }
3194 
3195 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3196 		     void *type_data, bool err_stop, bool rtnl_held)
3197 {
3198 	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3199 	int ok_count;
3200 
3201 retry:
3202 	if (take_rtnl)
3203 		rtnl_lock();
3204 	down_read(&block->cb_lock);
3205 	/* Need to obtain rtnl lock if block is bound to devs that require it.
3206 	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3207 	 * obtain the locks in same order here.
3208 	 */
3209 	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3210 		up_read(&block->cb_lock);
3211 		take_rtnl = true;
3212 		goto retry;
3213 	}
3214 
3215 	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3216 
3217 	up_read(&block->cb_lock);
3218 	if (take_rtnl)
3219 		rtnl_unlock();
3220 	return ok_count;
3221 }
3222 EXPORT_SYMBOL(tc_setup_cb_call);
3223 
3224 /* Non-destructive filter add. If filter that wasn't already in hardware is
3225  * successfully offloaded, increment block offloads counter. On failure,
3226  * previously offloaded filter is considered to be intact and offloads counter
3227  * is not decremented.
3228  */
3229 
3230 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3231 		    enum tc_setup_type type, void *type_data, bool err_stop,
3232 		    u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3233 {
3234 	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3235 	int ok_count;
3236 
3237 retry:
3238 	if (take_rtnl)
3239 		rtnl_lock();
3240 	down_read(&block->cb_lock);
3241 	/* Need to obtain rtnl lock if block is bound to devs that require it.
3242 	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3243 	 * obtain the locks in same order here.
3244 	 */
3245 	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3246 		up_read(&block->cb_lock);
3247 		take_rtnl = true;
3248 		goto retry;
3249 	}
3250 
3251 	/* Make sure all netdevs sharing this block are offload-capable. */
3252 	if (block->nooffloaddevcnt && err_stop) {
3253 		ok_count = -EOPNOTSUPP;
3254 		goto err_unlock;
3255 	}
3256 
3257 	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3258 	if (ok_count < 0)
3259 		goto err_unlock;
3260 
3261 	if (tp->ops->hw_add)
3262 		tp->ops->hw_add(tp, type_data);
3263 	if (ok_count > 0)
3264 		tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3265 					  ok_count, true);
3266 err_unlock:
3267 	up_read(&block->cb_lock);
3268 	if (take_rtnl)
3269 		rtnl_unlock();
3270 	return ok_count < 0 ? ok_count : 0;
3271 }
3272 EXPORT_SYMBOL(tc_setup_cb_add);
3273 
3274 /* Destructive filter replace. If filter that wasn't already in hardware is
3275  * successfully offloaded, increment block offload counter. On failure,
3276  * previously offloaded filter is considered to be destroyed and offload counter
3277  * is decremented.
3278  */
3279 
3280 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3281 			enum tc_setup_type type, void *type_data, bool err_stop,
3282 			u32 *old_flags, unsigned int *old_in_hw_count,
3283 			u32 *new_flags, unsigned int *new_in_hw_count,
3284 			bool rtnl_held)
3285 {
3286 	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3287 	int ok_count;
3288 
3289 retry:
3290 	if (take_rtnl)
3291 		rtnl_lock();
3292 	down_read(&block->cb_lock);
3293 	/* Need to obtain rtnl lock if block is bound to devs that require it.
3294 	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3295 	 * obtain the locks in same order here.
3296 	 */
3297 	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3298 		up_read(&block->cb_lock);
3299 		take_rtnl = true;
3300 		goto retry;
3301 	}
3302 
3303 	/* Make sure all netdevs sharing this block are offload-capable. */
3304 	if (block->nooffloaddevcnt && err_stop) {
3305 		ok_count = -EOPNOTSUPP;
3306 		goto err_unlock;
3307 	}
3308 
3309 	tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3310 	if (tp->ops->hw_del)
3311 		tp->ops->hw_del(tp, type_data);
3312 
3313 	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3314 	if (ok_count < 0)
3315 		goto err_unlock;
3316 
3317 	if (tp->ops->hw_add)
3318 		tp->ops->hw_add(tp, type_data);
3319 	if (ok_count > 0)
3320 		tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3321 					  new_flags, ok_count, true);
3322 err_unlock:
3323 	up_read(&block->cb_lock);
3324 	if (take_rtnl)
3325 		rtnl_unlock();
3326 	return ok_count < 0 ? ok_count : 0;
3327 }
3328 EXPORT_SYMBOL(tc_setup_cb_replace);
3329 
3330 /* Destroy filter and decrement block offload counter, if filter was previously
3331  * offloaded.
3332  */
3333 
3334 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3335 			enum tc_setup_type type, void *type_data, bool err_stop,
3336 			u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3337 {
3338 	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3339 	int ok_count;
3340 
3341 retry:
3342 	if (take_rtnl)
3343 		rtnl_lock();
3344 	down_read(&block->cb_lock);
3345 	/* Need to obtain rtnl lock if block is bound to devs that require it.
3346 	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3347 	 * obtain the locks in same order here.
3348 	 */
3349 	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3350 		up_read(&block->cb_lock);
3351 		take_rtnl = true;
3352 		goto retry;
3353 	}
3354 
3355 	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3356 
3357 	tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3358 	if (tp->ops->hw_del)
3359 		tp->ops->hw_del(tp, type_data);
3360 
3361 	up_read(&block->cb_lock);
3362 	if (take_rtnl)
3363 		rtnl_unlock();
3364 	return ok_count < 0 ? ok_count : 0;
3365 }
3366 EXPORT_SYMBOL(tc_setup_cb_destroy);
3367 
3368 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3369 			  bool add, flow_setup_cb_t *cb,
3370 			  enum tc_setup_type type, void *type_data,
3371 			  void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3372 {
3373 	int err = cb(type, type_data, cb_priv);
3374 
3375 	if (err) {
3376 		if (add && tc_skip_sw(*flags))
3377 			return err;
3378 	} else {
3379 		tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3380 					  add);
3381 	}
3382 
3383 	return 0;
3384 }
3385 EXPORT_SYMBOL(tc_setup_cb_reoffload);
3386 
3387 void tc_cleanup_flow_action(struct flow_action *flow_action)
3388 {
3389 	struct flow_action_entry *entry;
3390 	int i;
3391 
3392 	flow_action_for_each(i, entry, flow_action)
3393 		if (entry->destructor)
3394 			entry->destructor(entry->destructor_priv);
3395 }
3396 EXPORT_SYMBOL(tc_cleanup_flow_action);
3397 
3398 static void tcf_mirred_get_dev(struct flow_action_entry *entry,
3399 			       const struct tc_action *act)
3400 {
3401 #ifdef CONFIG_NET_CLS_ACT
3402 	entry->dev = act->ops->get_dev(act, &entry->destructor);
3403 	if (!entry->dev)
3404 		return;
3405 	entry->destructor_priv = entry->dev;
3406 #endif
3407 }
3408 
3409 static void tcf_tunnel_encap_put_tunnel(void *priv)
3410 {
3411 	struct ip_tunnel_info *tunnel = priv;
3412 
3413 	kfree(tunnel);
3414 }
3415 
3416 static int tcf_tunnel_encap_get_tunnel(struct flow_action_entry *entry,
3417 				       const struct tc_action *act)
3418 {
3419 	entry->tunnel = tcf_tunnel_info_copy(act);
3420 	if (!entry->tunnel)
3421 		return -ENOMEM;
3422 	entry->destructor = tcf_tunnel_encap_put_tunnel;
3423 	entry->destructor_priv = entry->tunnel;
3424 	return 0;
3425 }
3426 
3427 static void tcf_sample_get_group(struct flow_action_entry *entry,
3428 				 const struct tc_action *act)
3429 {
3430 #ifdef CONFIG_NET_CLS_ACT
3431 	entry->sample.psample_group =
3432 		act->ops->get_psample_group(act, &entry->destructor);
3433 	entry->destructor_priv = entry->sample.psample_group;
3434 #endif
3435 }
3436 
3437 int tc_setup_flow_action(struct flow_action *flow_action,
3438 			 const struct tcf_exts *exts, bool rtnl_held)
3439 {
3440 	const struct tc_action *act;
3441 	int i, j, k, err = 0;
3442 
3443 	if (!exts)
3444 		return 0;
3445 
3446 	if (!rtnl_held)
3447 		rtnl_lock();
3448 
3449 	j = 0;
3450 	tcf_exts_for_each_action(i, act, exts) {
3451 		struct flow_action_entry *entry;
3452 
3453 		entry = &flow_action->entries[j];
3454 		if (is_tcf_gact_ok(act)) {
3455 			entry->id = FLOW_ACTION_ACCEPT;
3456 		} else if (is_tcf_gact_shot(act)) {
3457 			entry->id = FLOW_ACTION_DROP;
3458 		} else if (is_tcf_gact_trap(act)) {
3459 			entry->id = FLOW_ACTION_TRAP;
3460 		} else if (is_tcf_gact_goto_chain(act)) {
3461 			entry->id = FLOW_ACTION_GOTO;
3462 			entry->chain_index = tcf_gact_goto_chain_index(act);
3463 		} else if (is_tcf_mirred_egress_redirect(act)) {
3464 			entry->id = FLOW_ACTION_REDIRECT;
3465 			tcf_mirred_get_dev(entry, act);
3466 		} else if (is_tcf_mirred_egress_mirror(act)) {
3467 			entry->id = FLOW_ACTION_MIRRED;
3468 			tcf_mirred_get_dev(entry, act);
3469 		} else if (is_tcf_mirred_ingress_redirect(act)) {
3470 			entry->id = FLOW_ACTION_REDIRECT_INGRESS;
3471 			tcf_mirred_get_dev(entry, act);
3472 		} else if (is_tcf_mirred_ingress_mirror(act)) {
3473 			entry->id = FLOW_ACTION_MIRRED_INGRESS;
3474 			tcf_mirred_get_dev(entry, act);
3475 		} else if (is_tcf_vlan(act)) {
3476 			switch (tcf_vlan_action(act)) {
3477 			case TCA_VLAN_ACT_PUSH:
3478 				entry->id = FLOW_ACTION_VLAN_PUSH;
3479 				entry->vlan.vid = tcf_vlan_push_vid(act);
3480 				entry->vlan.proto = tcf_vlan_push_proto(act);
3481 				entry->vlan.prio = tcf_vlan_push_prio(act);
3482 				break;
3483 			case TCA_VLAN_ACT_POP:
3484 				entry->id = FLOW_ACTION_VLAN_POP;
3485 				break;
3486 			case TCA_VLAN_ACT_MODIFY:
3487 				entry->id = FLOW_ACTION_VLAN_MANGLE;
3488 				entry->vlan.vid = tcf_vlan_push_vid(act);
3489 				entry->vlan.proto = tcf_vlan_push_proto(act);
3490 				entry->vlan.prio = tcf_vlan_push_prio(act);
3491 				break;
3492 			default:
3493 				err = -EOPNOTSUPP;
3494 				goto err_out;
3495 			}
3496 		} else if (is_tcf_tunnel_set(act)) {
3497 			entry->id = FLOW_ACTION_TUNNEL_ENCAP;
3498 			err = tcf_tunnel_encap_get_tunnel(entry, act);
3499 			if (err)
3500 				goto err_out;
3501 		} else if (is_tcf_tunnel_release(act)) {
3502 			entry->id = FLOW_ACTION_TUNNEL_DECAP;
3503 		} else if (is_tcf_pedit(act)) {
3504 			for (k = 0; k < tcf_pedit_nkeys(act); k++) {
3505 				switch (tcf_pedit_cmd(act, k)) {
3506 				case TCA_PEDIT_KEY_EX_CMD_SET:
3507 					entry->id = FLOW_ACTION_MANGLE;
3508 					break;
3509 				case TCA_PEDIT_KEY_EX_CMD_ADD:
3510 					entry->id = FLOW_ACTION_ADD;
3511 					break;
3512 				default:
3513 					err = -EOPNOTSUPP;
3514 					goto err_out;
3515 				}
3516 				entry->mangle.htype = tcf_pedit_htype(act, k);
3517 				entry->mangle.mask = tcf_pedit_mask(act, k);
3518 				entry->mangle.val = tcf_pedit_val(act, k);
3519 				entry->mangle.offset = tcf_pedit_offset(act, k);
3520 				entry = &flow_action->entries[++j];
3521 			}
3522 		} else if (is_tcf_csum(act)) {
3523 			entry->id = FLOW_ACTION_CSUM;
3524 			entry->csum_flags = tcf_csum_update_flags(act);
3525 		} else if (is_tcf_skbedit_mark(act)) {
3526 			entry->id = FLOW_ACTION_MARK;
3527 			entry->mark = tcf_skbedit_mark(act);
3528 		} else if (is_tcf_sample(act)) {
3529 			entry->id = FLOW_ACTION_SAMPLE;
3530 			entry->sample.trunc_size = tcf_sample_trunc_size(act);
3531 			entry->sample.truncate = tcf_sample_truncate(act);
3532 			entry->sample.rate = tcf_sample_rate(act);
3533 			tcf_sample_get_group(entry, act);
3534 		} else if (is_tcf_police(act)) {
3535 			entry->id = FLOW_ACTION_POLICE;
3536 			entry->police.burst = tcf_police_tcfp_burst(act);
3537 			entry->police.rate_bytes_ps =
3538 				tcf_police_rate_bytes_ps(act);
3539 		} else if (is_tcf_ct(act)) {
3540 			entry->id = FLOW_ACTION_CT;
3541 			entry->ct.action = tcf_ct_action(act);
3542 			entry->ct.zone = tcf_ct_zone(act);
3543 		} else if (is_tcf_mpls(act)) {
3544 			switch (tcf_mpls_action(act)) {
3545 			case TCA_MPLS_ACT_PUSH:
3546 				entry->id = FLOW_ACTION_MPLS_PUSH;
3547 				entry->mpls_push.proto = tcf_mpls_proto(act);
3548 				entry->mpls_push.label = tcf_mpls_label(act);
3549 				entry->mpls_push.tc = tcf_mpls_tc(act);
3550 				entry->mpls_push.bos = tcf_mpls_bos(act);
3551 				entry->mpls_push.ttl = tcf_mpls_ttl(act);
3552 				break;
3553 			case TCA_MPLS_ACT_POP:
3554 				entry->id = FLOW_ACTION_MPLS_POP;
3555 				entry->mpls_pop.proto = tcf_mpls_proto(act);
3556 				break;
3557 			case TCA_MPLS_ACT_MODIFY:
3558 				entry->id = FLOW_ACTION_MPLS_MANGLE;
3559 				entry->mpls_mangle.label = tcf_mpls_label(act);
3560 				entry->mpls_mangle.tc = tcf_mpls_tc(act);
3561 				entry->mpls_mangle.bos = tcf_mpls_bos(act);
3562 				entry->mpls_mangle.ttl = tcf_mpls_ttl(act);
3563 				break;
3564 			default:
3565 				goto err_out;
3566 			}
3567 		} else if (is_tcf_skbedit_ptype(act)) {
3568 			entry->id = FLOW_ACTION_PTYPE;
3569 			entry->ptype = tcf_skbedit_ptype(act);
3570 		} else {
3571 			err = -EOPNOTSUPP;
3572 			goto err_out;
3573 		}
3574 
3575 		if (!is_tcf_pedit(act))
3576 			j++;
3577 	}
3578 
3579 err_out:
3580 	if (!rtnl_held)
3581 		rtnl_unlock();
3582 
3583 	if (err)
3584 		tc_cleanup_flow_action(flow_action);
3585 
3586 	return err;
3587 }
3588 EXPORT_SYMBOL(tc_setup_flow_action);
3589 
3590 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3591 {
3592 	unsigned int num_acts = 0;
3593 	struct tc_action *act;
3594 	int i;
3595 
3596 	tcf_exts_for_each_action(i, act, exts) {
3597 		if (is_tcf_pedit(act))
3598 			num_acts += tcf_pedit_nkeys(act);
3599 		else
3600 			num_acts++;
3601 	}
3602 	return num_acts;
3603 }
3604 EXPORT_SYMBOL(tcf_exts_num_actions);
3605 
3606 static __net_init int tcf_net_init(struct net *net)
3607 {
3608 	struct tcf_net *tn = net_generic(net, tcf_net_id);
3609 
3610 	spin_lock_init(&tn->idr_lock);
3611 	idr_init(&tn->idr);
3612 	return 0;
3613 }
3614 
3615 static void __net_exit tcf_net_exit(struct net *net)
3616 {
3617 	struct tcf_net *tn = net_generic(net, tcf_net_id);
3618 
3619 	idr_destroy(&tn->idr);
3620 }
3621 
3622 static struct pernet_operations tcf_net_ops = {
3623 	.init = tcf_net_init,
3624 	.exit = tcf_net_exit,
3625 	.id   = &tcf_net_id,
3626 	.size = sizeof(struct tcf_net),
3627 };
3628 
3629 static struct flow_indr_block_ing_entry block_ing_entry = {
3630 	.cb = tc_indr_block_get_and_ing_cmd,
3631 	.list = LIST_HEAD_INIT(block_ing_entry.list),
3632 };
3633 
3634 static int __init tc_filter_init(void)
3635 {
3636 	int err;
3637 
3638 	tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3639 	if (!tc_filter_wq)
3640 		return -ENOMEM;
3641 
3642 	err = register_pernet_subsys(&tcf_net_ops);
3643 	if (err)
3644 		goto err_register_pernet_subsys;
3645 
3646 	flow_indr_add_block_ing_cb(&block_ing_entry);
3647 
3648 	rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3649 		      RTNL_FLAG_DOIT_UNLOCKED);
3650 	rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3651 		      RTNL_FLAG_DOIT_UNLOCKED);
3652 	rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3653 		      tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3654 	rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3655 	rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3656 	rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3657 		      tc_dump_chain, 0);
3658 
3659 	return 0;
3660 
3661 err_register_pernet_subsys:
3662 	destroy_workqueue(tc_filter_wq);
3663 	return err;
3664 }
3665 
3666 subsys_initcall(tc_filter_init);
3667