xref: /openbmc/linux/net/sched/cls_api.c (revision 68198dca)
1 /*
2  * net/sched/cls_api.c	Packet classifier API.
3  *
4  *		This program is free software; you can redistribute it and/or
5  *		modify it under the terms of the GNU General Public License
6  *		as published by the Free Software Foundation; either version
7  *		2 of the License, or (at your option) any later version.
8  *
9  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10  *
11  * Changes:
12  *
13  * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
14  *
15  */
16 
17 #include <linux/module.h>
18 #include <linux/types.h>
19 #include <linux/kernel.h>
20 #include <linux/string.h>
21 #include <linux/errno.h>
22 #include <linux/err.h>
23 #include <linux/skbuff.h>
24 #include <linux/init.h>
25 #include <linux/kmod.h>
26 #include <linux/slab.h>
27 #include <net/net_namespace.h>
28 #include <net/sock.h>
29 #include <net/netlink.h>
30 #include <net/pkt_sched.h>
31 #include <net/pkt_cls.h>
32 
33 /* The list of all installed classifier types */
34 static LIST_HEAD(tcf_proto_base);
35 
36 /* Protects list of registered TC modules. It is pure SMP lock. */
37 static DEFINE_RWLOCK(cls_mod_lock);
38 
39 /* Find classifier type by string name */
40 
41 static const struct tcf_proto_ops *tcf_proto_lookup_ops(const char *kind)
42 {
43 	const struct tcf_proto_ops *t, *res = NULL;
44 
45 	if (kind) {
46 		read_lock(&cls_mod_lock);
47 		list_for_each_entry(t, &tcf_proto_base, head) {
48 			if (strcmp(kind, t->kind) == 0) {
49 				if (try_module_get(t->owner))
50 					res = t;
51 				break;
52 			}
53 		}
54 		read_unlock(&cls_mod_lock);
55 	}
56 	return res;
57 }
58 
59 /* Register(unregister) new classifier type */
60 
61 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
62 {
63 	struct tcf_proto_ops *t;
64 	int rc = -EEXIST;
65 
66 	write_lock(&cls_mod_lock);
67 	list_for_each_entry(t, &tcf_proto_base, head)
68 		if (!strcmp(ops->kind, t->kind))
69 			goto out;
70 
71 	list_add_tail(&ops->head, &tcf_proto_base);
72 	rc = 0;
73 out:
74 	write_unlock(&cls_mod_lock);
75 	return rc;
76 }
77 EXPORT_SYMBOL(register_tcf_proto_ops);
78 
79 static struct workqueue_struct *tc_filter_wq;
80 
81 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
82 {
83 	struct tcf_proto_ops *t;
84 	int rc = -ENOENT;
85 
86 	/* Wait for outstanding call_rcu()s, if any, from a
87 	 * tcf_proto_ops's destroy() handler.
88 	 */
89 	rcu_barrier();
90 	flush_workqueue(tc_filter_wq);
91 
92 	write_lock(&cls_mod_lock);
93 	list_for_each_entry(t, &tcf_proto_base, head) {
94 		if (t == ops) {
95 			list_del(&t->head);
96 			rc = 0;
97 			break;
98 		}
99 	}
100 	write_unlock(&cls_mod_lock);
101 	return rc;
102 }
103 EXPORT_SYMBOL(unregister_tcf_proto_ops);
104 
105 bool tcf_queue_work(struct work_struct *work)
106 {
107 	return queue_work(tc_filter_wq, work);
108 }
109 EXPORT_SYMBOL(tcf_queue_work);
110 
111 /* Select new prio value from the range, managed by kernel. */
112 
113 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
114 {
115 	u32 first = TC_H_MAKE(0xC0000000U, 0U);
116 
117 	if (tp)
118 		first = tp->prio - 1;
119 
120 	return TC_H_MAJ(first);
121 }
122 
123 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
124 					  u32 prio, u32 parent, struct Qdisc *q,
125 					  struct tcf_chain *chain)
126 {
127 	struct tcf_proto *tp;
128 	int err;
129 
130 	tp = kzalloc(sizeof(*tp), GFP_KERNEL);
131 	if (!tp)
132 		return ERR_PTR(-ENOBUFS);
133 
134 	err = -ENOENT;
135 	tp->ops = tcf_proto_lookup_ops(kind);
136 	if (!tp->ops) {
137 #ifdef CONFIG_MODULES
138 		rtnl_unlock();
139 		request_module("cls_%s", kind);
140 		rtnl_lock();
141 		tp->ops = tcf_proto_lookup_ops(kind);
142 		/* We dropped the RTNL semaphore in order to perform
143 		 * the module load. So, even if we succeeded in loading
144 		 * the module we have to replay the request. We indicate
145 		 * this using -EAGAIN.
146 		 */
147 		if (tp->ops) {
148 			module_put(tp->ops->owner);
149 			err = -EAGAIN;
150 		} else {
151 			err = -ENOENT;
152 		}
153 		goto errout;
154 #endif
155 	}
156 	tp->classify = tp->ops->classify;
157 	tp->protocol = protocol;
158 	tp->prio = prio;
159 	tp->classid = parent;
160 	tp->q = q;
161 	tp->chain = chain;
162 
163 	err = tp->ops->init(tp);
164 	if (err) {
165 		module_put(tp->ops->owner);
166 		goto errout;
167 	}
168 	return tp;
169 
170 errout:
171 	kfree(tp);
172 	return ERR_PTR(err);
173 }
174 
175 static void tcf_proto_destroy(struct tcf_proto *tp)
176 {
177 	tp->ops->destroy(tp);
178 	module_put(tp->ops->owner);
179 	kfree_rcu(tp, rcu);
180 }
181 
182 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
183 					  u32 chain_index)
184 {
185 	struct tcf_chain *chain;
186 
187 	chain = kzalloc(sizeof(*chain), GFP_KERNEL);
188 	if (!chain)
189 		return NULL;
190 	list_add_tail(&chain->list, &block->chain_list);
191 	chain->block = block;
192 	chain->index = chain_index;
193 	chain->refcnt = 1;
194 	return chain;
195 }
196 
197 static void tcf_chain_head_change(struct tcf_chain *chain,
198 				  struct tcf_proto *tp_head)
199 {
200 	if (chain->chain_head_change)
201 		chain->chain_head_change(tp_head,
202 					 chain->chain_head_change_priv);
203 }
204 
205 static void tcf_chain_flush(struct tcf_chain *chain)
206 {
207 	struct tcf_proto *tp = rtnl_dereference(chain->filter_chain);
208 
209 	tcf_chain_head_change(chain, NULL);
210 	while (tp) {
211 		RCU_INIT_POINTER(chain->filter_chain, tp->next);
212 		tcf_proto_destroy(tp);
213 		tp = rtnl_dereference(chain->filter_chain);
214 		tcf_chain_put(chain);
215 	}
216 }
217 
218 static void tcf_chain_destroy(struct tcf_chain *chain)
219 {
220 	list_del(&chain->list);
221 	kfree(chain);
222 }
223 
224 static void tcf_chain_hold(struct tcf_chain *chain)
225 {
226 	++chain->refcnt;
227 }
228 
229 struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
230 				bool create)
231 {
232 	struct tcf_chain *chain;
233 
234 	list_for_each_entry(chain, &block->chain_list, list) {
235 		if (chain->index == chain_index) {
236 			tcf_chain_hold(chain);
237 			return chain;
238 		}
239 	}
240 
241 	return create ? tcf_chain_create(block, chain_index) : NULL;
242 }
243 EXPORT_SYMBOL(tcf_chain_get);
244 
245 void tcf_chain_put(struct tcf_chain *chain)
246 {
247 	if (--chain->refcnt == 0)
248 		tcf_chain_destroy(chain);
249 }
250 EXPORT_SYMBOL(tcf_chain_put);
251 
252 static void tcf_block_offload_cmd(struct tcf_block *block, struct Qdisc *q,
253 				  struct tcf_block_ext_info *ei,
254 				  enum tc_block_command command)
255 {
256 	struct net_device *dev = q->dev_queue->dev;
257 	struct tc_block_offload bo = {};
258 
259 	if (!dev->netdev_ops->ndo_setup_tc)
260 		return;
261 	bo.command = command;
262 	bo.binder_type = ei->binder_type;
263 	bo.block = block;
264 	dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
265 }
266 
267 static void tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
268 				   struct tcf_block_ext_info *ei)
269 {
270 	tcf_block_offload_cmd(block, q, ei, TC_BLOCK_BIND);
271 }
272 
273 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
274 				     struct tcf_block_ext_info *ei)
275 {
276 	tcf_block_offload_cmd(block, q, ei, TC_BLOCK_UNBIND);
277 }
278 
279 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
280 		      struct tcf_block_ext_info *ei)
281 {
282 	struct tcf_block *block = kzalloc(sizeof(*block), GFP_KERNEL);
283 	struct tcf_chain *chain;
284 	int err;
285 
286 	if (!block)
287 		return -ENOMEM;
288 	INIT_LIST_HEAD(&block->chain_list);
289 	INIT_LIST_HEAD(&block->cb_list);
290 
291 	/* Create chain 0 by default, it has to be always present. */
292 	chain = tcf_chain_create(block, 0);
293 	if (!chain) {
294 		err = -ENOMEM;
295 		goto err_chain_create;
296 	}
297 	WARN_ON(!ei->chain_head_change);
298 	chain->chain_head_change = ei->chain_head_change;
299 	chain->chain_head_change_priv = ei->chain_head_change_priv;
300 	block->net = qdisc_net(q);
301 	block->q = q;
302 	tcf_block_offload_bind(block, q, ei);
303 	*p_block = block;
304 	return 0;
305 
306 err_chain_create:
307 	kfree(block);
308 	return err;
309 }
310 EXPORT_SYMBOL(tcf_block_get_ext);
311 
312 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
313 {
314 	struct tcf_proto __rcu **p_filter_chain = priv;
315 
316 	rcu_assign_pointer(*p_filter_chain, tp_head);
317 }
318 
319 int tcf_block_get(struct tcf_block **p_block,
320 		  struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q)
321 {
322 	struct tcf_block_ext_info ei = {
323 		.chain_head_change = tcf_chain_head_change_dflt,
324 		.chain_head_change_priv = p_filter_chain,
325 	};
326 
327 	WARN_ON(!p_filter_chain);
328 	return tcf_block_get_ext(p_block, q, &ei);
329 }
330 EXPORT_SYMBOL(tcf_block_get);
331 
332 static void tcf_block_put_final(struct work_struct *work)
333 {
334 	struct tcf_block *block = container_of(work, struct tcf_block, work);
335 	struct tcf_chain *chain, *tmp;
336 
337 	rtnl_lock();
338 
339 	/* At this point, all the chains should have refcnt == 1. */
340 	list_for_each_entry_safe(chain, tmp, &block->chain_list, list)
341 		tcf_chain_put(chain);
342 	rtnl_unlock();
343 	kfree(block);
344 }
345 
346 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
347  * actions should be all removed after flushing.
348  */
349 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
350 		       struct tcf_block_ext_info *ei)
351 {
352 	struct tcf_chain *chain;
353 
354 	if (!block)
355 		return;
356 	/* Hold a refcnt for all chains, except 0, so that they don't disappear
357 	 * while we are iterating.
358 	 */
359 	list_for_each_entry(chain, &block->chain_list, list)
360 		if (chain->index)
361 			tcf_chain_hold(chain);
362 
363 	list_for_each_entry(chain, &block->chain_list, list)
364 		tcf_chain_flush(chain);
365 
366 	tcf_block_offload_unbind(block, q, ei);
367 
368 	INIT_WORK(&block->work, tcf_block_put_final);
369 	/* Wait for existing RCU callbacks to cool down, make sure their works
370 	 * have been queued before this. We can not flush pending works here
371 	 * because we are holding the RTNL lock.
372 	 */
373 	rcu_barrier();
374 	tcf_queue_work(&block->work);
375 }
376 EXPORT_SYMBOL(tcf_block_put_ext);
377 
378 void tcf_block_put(struct tcf_block *block)
379 {
380 	struct tcf_block_ext_info ei = {0, };
381 
382 	tcf_block_put_ext(block, block->q, &ei);
383 }
384 
385 EXPORT_SYMBOL(tcf_block_put);
386 
387 struct tcf_block_cb {
388 	struct list_head list;
389 	tc_setup_cb_t *cb;
390 	void *cb_ident;
391 	void *cb_priv;
392 	unsigned int refcnt;
393 };
394 
395 void *tcf_block_cb_priv(struct tcf_block_cb *block_cb)
396 {
397 	return block_cb->cb_priv;
398 }
399 EXPORT_SYMBOL(tcf_block_cb_priv);
400 
401 struct tcf_block_cb *tcf_block_cb_lookup(struct tcf_block *block,
402 					 tc_setup_cb_t *cb, void *cb_ident)
403 {	struct tcf_block_cb *block_cb;
404 
405 	list_for_each_entry(block_cb, &block->cb_list, list)
406 		if (block_cb->cb == cb && block_cb->cb_ident == cb_ident)
407 			return block_cb;
408 	return NULL;
409 }
410 EXPORT_SYMBOL(tcf_block_cb_lookup);
411 
412 void tcf_block_cb_incref(struct tcf_block_cb *block_cb)
413 {
414 	block_cb->refcnt++;
415 }
416 EXPORT_SYMBOL(tcf_block_cb_incref);
417 
418 unsigned int tcf_block_cb_decref(struct tcf_block_cb *block_cb)
419 {
420 	return --block_cb->refcnt;
421 }
422 EXPORT_SYMBOL(tcf_block_cb_decref);
423 
424 struct tcf_block_cb *__tcf_block_cb_register(struct tcf_block *block,
425 					     tc_setup_cb_t *cb, void *cb_ident,
426 					     void *cb_priv)
427 {
428 	struct tcf_block_cb *block_cb;
429 
430 	block_cb = kzalloc(sizeof(*block_cb), GFP_KERNEL);
431 	if (!block_cb)
432 		return NULL;
433 	block_cb->cb = cb;
434 	block_cb->cb_ident = cb_ident;
435 	block_cb->cb_priv = cb_priv;
436 	list_add(&block_cb->list, &block->cb_list);
437 	return block_cb;
438 }
439 EXPORT_SYMBOL(__tcf_block_cb_register);
440 
441 int tcf_block_cb_register(struct tcf_block *block,
442 			  tc_setup_cb_t *cb, void *cb_ident,
443 			  void *cb_priv)
444 {
445 	struct tcf_block_cb *block_cb;
446 
447 	block_cb = __tcf_block_cb_register(block, cb, cb_ident, cb_priv);
448 	return block_cb ? 0 : -ENOMEM;
449 }
450 EXPORT_SYMBOL(tcf_block_cb_register);
451 
452 void __tcf_block_cb_unregister(struct tcf_block_cb *block_cb)
453 {
454 	list_del(&block_cb->list);
455 	kfree(block_cb);
456 }
457 EXPORT_SYMBOL(__tcf_block_cb_unregister);
458 
459 void tcf_block_cb_unregister(struct tcf_block *block,
460 			     tc_setup_cb_t *cb, void *cb_ident)
461 {
462 	struct tcf_block_cb *block_cb;
463 
464 	block_cb = tcf_block_cb_lookup(block, cb, cb_ident);
465 	if (!block_cb)
466 		return;
467 	__tcf_block_cb_unregister(block_cb);
468 }
469 EXPORT_SYMBOL(tcf_block_cb_unregister);
470 
471 static int tcf_block_cb_call(struct tcf_block *block, enum tc_setup_type type,
472 			     void *type_data, bool err_stop)
473 {
474 	struct tcf_block_cb *block_cb;
475 	int ok_count = 0;
476 	int err;
477 
478 	list_for_each_entry(block_cb, &block->cb_list, list) {
479 		err = block_cb->cb(type, type_data, block_cb->cb_priv);
480 		if (err) {
481 			if (err_stop)
482 				return err;
483 		} else {
484 			ok_count++;
485 		}
486 	}
487 	return ok_count;
488 }
489 
490 /* Main classifier routine: scans classifier chain attached
491  * to this qdisc, (optionally) tests for protocol and asks
492  * specific classifiers.
493  */
494 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
495 		 struct tcf_result *res, bool compat_mode)
496 {
497 	__be16 protocol = tc_skb_protocol(skb);
498 #ifdef CONFIG_NET_CLS_ACT
499 	const int max_reclassify_loop = 4;
500 	const struct tcf_proto *orig_tp = tp;
501 	const struct tcf_proto *first_tp;
502 	int limit = 0;
503 
504 reclassify:
505 #endif
506 	for (; tp; tp = rcu_dereference_bh(tp->next)) {
507 		int err;
508 
509 		if (tp->protocol != protocol &&
510 		    tp->protocol != htons(ETH_P_ALL))
511 			continue;
512 
513 		err = tp->classify(skb, tp, res);
514 #ifdef CONFIG_NET_CLS_ACT
515 		if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
516 			first_tp = orig_tp;
517 			goto reset;
518 		} else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
519 			first_tp = res->goto_tp;
520 			goto reset;
521 		}
522 #endif
523 		if (err >= 0)
524 			return err;
525 	}
526 
527 	return TC_ACT_UNSPEC; /* signal: continue lookup */
528 #ifdef CONFIG_NET_CLS_ACT
529 reset:
530 	if (unlikely(limit++ >= max_reclassify_loop)) {
531 		net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n",
532 				       tp->q->ops->id, tp->prio & 0xffff,
533 				       ntohs(tp->protocol));
534 		return TC_ACT_SHOT;
535 	}
536 
537 	tp = first_tp;
538 	protocol = tc_skb_protocol(skb);
539 	goto reclassify;
540 #endif
541 }
542 EXPORT_SYMBOL(tcf_classify);
543 
544 struct tcf_chain_info {
545 	struct tcf_proto __rcu **pprev;
546 	struct tcf_proto __rcu *next;
547 };
548 
549 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain_info *chain_info)
550 {
551 	return rtnl_dereference(*chain_info->pprev);
552 }
553 
554 static void tcf_chain_tp_insert(struct tcf_chain *chain,
555 				struct tcf_chain_info *chain_info,
556 				struct tcf_proto *tp)
557 {
558 	if (*chain_info->pprev == chain->filter_chain)
559 		tcf_chain_head_change(chain, tp);
560 	RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain_info));
561 	rcu_assign_pointer(*chain_info->pprev, tp);
562 	tcf_chain_hold(chain);
563 }
564 
565 static void tcf_chain_tp_remove(struct tcf_chain *chain,
566 				struct tcf_chain_info *chain_info,
567 				struct tcf_proto *tp)
568 {
569 	struct tcf_proto *next = rtnl_dereference(chain_info->next);
570 
571 	if (tp == chain->filter_chain)
572 		tcf_chain_head_change(chain, next);
573 	RCU_INIT_POINTER(*chain_info->pprev, next);
574 	tcf_chain_put(chain);
575 }
576 
577 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
578 					   struct tcf_chain_info *chain_info,
579 					   u32 protocol, u32 prio,
580 					   bool prio_allocate)
581 {
582 	struct tcf_proto **pprev;
583 	struct tcf_proto *tp;
584 
585 	/* Check the chain for existence of proto-tcf with this priority */
586 	for (pprev = &chain->filter_chain;
587 	     (tp = rtnl_dereference(*pprev)); pprev = &tp->next) {
588 		if (tp->prio >= prio) {
589 			if (tp->prio == prio) {
590 				if (prio_allocate ||
591 				    (tp->protocol != protocol && protocol))
592 					return ERR_PTR(-EINVAL);
593 			} else {
594 				tp = NULL;
595 			}
596 			break;
597 		}
598 	}
599 	chain_info->pprev = pprev;
600 	chain_info->next = tp ? tp->next : NULL;
601 	return tp;
602 }
603 
604 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
605 			 struct tcf_proto *tp, struct Qdisc *q, u32 parent,
606 			 void *fh, u32 portid, u32 seq, u16 flags, int event)
607 {
608 	struct tcmsg *tcm;
609 	struct nlmsghdr  *nlh;
610 	unsigned char *b = skb_tail_pointer(skb);
611 
612 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
613 	if (!nlh)
614 		goto out_nlmsg_trim;
615 	tcm = nlmsg_data(nlh);
616 	tcm->tcm_family = AF_UNSPEC;
617 	tcm->tcm__pad1 = 0;
618 	tcm->tcm__pad2 = 0;
619 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
620 	tcm->tcm_parent = parent;
621 	tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
622 	if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
623 		goto nla_put_failure;
624 	if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
625 		goto nla_put_failure;
626 	if (!fh) {
627 		tcm->tcm_handle = 0;
628 	} else {
629 		if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
630 			goto nla_put_failure;
631 	}
632 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
633 	return skb->len;
634 
635 out_nlmsg_trim:
636 nla_put_failure:
637 	nlmsg_trim(skb, b);
638 	return -1;
639 }
640 
641 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
642 			  struct nlmsghdr *n, struct tcf_proto *tp,
643 			  struct Qdisc *q, u32 parent,
644 			  void *fh, int event, bool unicast)
645 {
646 	struct sk_buff *skb;
647 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
648 
649 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
650 	if (!skb)
651 		return -ENOBUFS;
652 
653 	if (tcf_fill_node(net, skb, tp, q, parent, fh, portid, n->nlmsg_seq,
654 			  n->nlmsg_flags, event) <= 0) {
655 		kfree_skb(skb);
656 		return -EINVAL;
657 	}
658 
659 	if (unicast)
660 		return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
661 
662 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
663 			      n->nlmsg_flags & NLM_F_ECHO);
664 }
665 
666 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
667 			      struct nlmsghdr *n, struct tcf_proto *tp,
668 			      struct Qdisc *q, u32 parent,
669 			      void *fh, bool unicast, bool *last)
670 {
671 	struct sk_buff *skb;
672 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
673 	int err;
674 
675 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
676 	if (!skb)
677 		return -ENOBUFS;
678 
679 	if (tcf_fill_node(net, skb, tp, q, parent, fh, portid, n->nlmsg_seq,
680 			  n->nlmsg_flags, RTM_DELTFILTER) <= 0) {
681 		kfree_skb(skb);
682 		return -EINVAL;
683 	}
684 
685 	err = tp->ops->delete(tp, fh, last);
686 	if (err) {
687 		kfree_skb(skb);
688 		return err;
689 	}
690 
691 	if (unicast)
692 		return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
693 
694 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
695 			      n->nlmsg_flags & NLM_F_ECHO);
696 }
697 
698 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
699 				 struct Qdisc *q, u32 parent,
700 				 struct nlmsghdr *n,
701 				 struct tcf_chain *chain, int event)
702 {
703 	struct tcf_proto *tp;
704 
705 	for (tp = rtnl_dereference(chain->filter_chain);
706 	     tp; tp = rtnl_dereference(tp->next))
707 		tfilter_notify(net, oskb, n, tp, q, parent, 0, event, false);
708 }
709 
710 /* Add/change/delete/get a filter node */
711 
712 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
713 			  struct netlink_ext_ack *extack)
714 {
715 	struct net *net = sock_net(skb->sk);
716 	struct nlattr *tca[TCA_MAX + 1];
717 	struct tcmsg *t;
718 	u32 protocol;
719 	u32 prio;
720 	bool prio_allocate;
721 	u32 parent;
722 	u32 chain_index;
723 	struct net_device *dev;
724 	struct Qdisc  *q;
725 	struct tcf_chain_info chain_info;
726 	struct tcf_chain *chain = NULL;
727 	struct tcf_block *block;
728 	struct tcf_proto *tp;
729 	const struct Qdisc_class_ops *cops;
730 	unsigned long cl;
731 	void *fh;
732 	int err;
733 	int tp_created;
734 
735 	if ((n->nlmsg_type != RTM_GETTFILTER) &&
736 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
737 		return -EPERM;
738 
739 replay:
740 	tp_created = 0;
741 
742 	err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack);
743 	if (err < 0)
744 		return err;
745 
746 	t = nlmsg_data(n);
747 	protocol = TC_H_MIN(t->tcm_info);
748 	prio = TC_H_MAJ(t->tcm_info);
749 	prio_allocate = false;
750 	parent = t->tcm_parent;
751 	cl = 0;
752 
753 	if (prio == 0) {
754 		switch (n->nlmsg_type) {
755 		case RTM_DELTFILTER:
756 			if (protocol || t->tcm_handle || tca[TCA_KIND])
757 				return -ENOENT;
758 			break;
759 		case RTM_NEWTFILTER:
760 			/* If no priority is provided by the user,
761 			 * we allocate one.
762 			 */
763 			if (n->nlmsg_flags & NLM_F_CREATE) {
764 				prio = TC_H_MAKE(0x80000000U, 0U);
765 				prio_allocate = true;
766 				break;
767 			}
768 			/* fall-through */
769 		default:
770 			return -ENOENT;
771 		}
772 	}
773 
774 	/* Find head of filter chain. */
775 
776 	/* Find link */
777 	dev = __dev_get_by_index(net, t->tcm_ifindex);
778 	if (dev == NULL)
779 		return -ENODEV;
780 
781 	/* Find qdisc */
782 	if (!parent) {
783 		q = dev->qdisc;
784 		parent = q->handle;
785 	} else {
786 		q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
787 		if (q == NULL)
788 			return -EINVAL;
789 	}
790 
791 	/* Is it classful? */
792 	cops = q->ops->cl_ops;
793 	if (!cops)
794 		return -EINVAL;
795 
796 	if (!cops->tcf_block)
797 		return -EOPNOTSUPP;
798 
799 	/* Do we search for filter, attached to class? */
800 	if (TC_H_MIN(parent)) {
801 		cl = cops->find(q, parent);
802 		if (cl == 0)
803 			return -ENOENT;
804 	}
805 
806 	/* And the last stroke */
807 	block = cops->tcf_block(q, cl);
808 	if (!block) {
809 		err = -EINVAL;
810 		goto errout;
811 	}
812 
813 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
814 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
815 		err = -EINVAL;
816 		goto errout;
817 	}
818 	chain = tcf_chain_get(block, chain_index,
819 			      n->nlmsg_type == RTM_NEWTFILTER);
820 	if (!chain) {
821 		err = n->nlmsg_type == RTM_NEWTFILTER ? -ENOMEM : -EINVAL;
822 		goto errout;
823 	}
824 
825 	if (n->nlmsg_type == RTM_DELTFILTER && prio == 0) {
826 		tfilter_notify_chain(net, skb, q, parent, n,
827 				     chain, RTM_DELTFILTER);
828 		tcf_chain_flush(chain);
829 		err = 0;
830 		goto errout;
831 	}
832 
833 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
834 			       prio, prio_allocate);
835 	if (IS_ERR(tp)) {
836 		err = PTR_ERR(tp);
837 		goto errout;
838 	}
839 
840 	if (tp == NULL) {
841 		/* Proto-tcf does not exist, create new one */
842 
843 		if (tca[TCA_KIND] == NULL || !protocol) {
844 			err = -EINVAL;
845 			goto errout;
846 		}
847 
848 		if (n->nlmsg_type != RTM_NEWTFILTER ||
849 		    !(n->nlmsg_flags & NLM_F_CREATE)) {
850 			err = -ENOENT;
851 			goto errout;
852 		}
853 
854 		if (prio_allocate)
855 			prio = tcf_auto_prio(tcf_chain_tp_prev(&chain_info));
856 
857 		tp = tcf_proto_create(nla_data(tca[TCA_KIND]),
858 				      protocol, prio, parent, q, chain);
859 		if (IS_ERR(tp)) {
860 			err = PTR_ERR(tp);
861 			goto errout;
862 		}
863 		tp_created = 1;
864 	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
865 		err = -EINVAL;
866 		goto errout;
867 	}
868 
869 	fh = tp->ops->get(tp, t->tcm_handle);
870 
871 	if (!fh) {
872 		if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
873 			tcf_chain_tp_remove(chain, &chain_info, tp);
874 			tfilter_notify(net, skb, n, tp, q, parent, fh,
875 				       RTM_DELTFILTER, false);
876 			tcf_proto_destroy(tp);
877 			err = 0;
878 			goto errout;
879 		}
880 
881 		if (n->nlmsg_type != RTM_NEWTFILTER ||
882 		    !(n->nlmsg_flags & NLM_F_CREATE)) {
883 			err = -ENOENT;
884 			goto errout;
885 		}
886 	} else {
887 		bool last;
888 
889 		switch (n->nlmsg_type) {
890 		case RTM_NEWTFILTER:
891 			if (n->nlmsg_flags & NLM_F_EXCL) {
892 				if (tp_created)
893 					tcf_proto_destroy(tp);
894 				err = -EEXIST;
895 				goto errout;
896 			}
897 			break;
898 		case RTM_DELTFILTER:
899 			err = tfilter_del_notify(net, skb, n, tp, q, parent,
900 						 fh, false, &last);
901 			if (err)
902 				goto errout;
903 			if (last) {
904 				tcf_chain_tp_remove(chain, &chain_info, tp);
905 				tcf_proto_destroy(tp);
906 			}
907 			goto errout;
908 		case RTM_GETTFILTER:
909 			err = tfilter_notify(net, skb, n, tp, q, parent, fh,
910 					     RTM_NEWTFILTER, true);
911 			goto errout;
912 		default:
913 			err = -EINVAL;
914 			goto errout;
915 		}
916 	}
917 
918 	err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
919 			      n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE);
920 	if (err == 0) {
921 		if (tp_created)
922 			tcf_chain_tp_insert(chain, &chain_info, tp);
923 		tfilter_notify(net, skb, n, tp, q, parent, fh,
924 			       RTM_NEWTFILTER, false);
925 	} else {
926 		if (tp_created)
927 			tcf_proto_destroy(tp);
928 	}
929 
930 errout:
931 	if (chain)
932 		tcf_chain_put(chain);
933 	if (err == -EAGAIN)
934 		/* Replay the request. */
935 		goto replay;
936 	return err;
937 }
938 
939 struct tcf_dump_args {
940 	struct tcf_walker w;
941 	struct sk_buff *skb;
942 	struct netlink_callback *cb;
943 	struct Qdisc *q;
944 	u32 parent;
945 };
946 
947 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
948 {
949 	struct tcf_dump_args *a = (void *)arg;
950 	struct net *net = sock_net(a->skb->sk);
951 
952 	return tcf_fill_node(net, a->skb, tp, a->q, a->parent,
953 			     n, NETLINK_CB(a->cb->skb).portid,
954 			     a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
955 			     RTM_NEWTFILTER);
956 }
957 
958 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
959 			   struct sk_buff *skb, struct netlink_callback *cb,
960 			   long index_start, long *p_index)
961 {
962 	struct net *net = sock_net(skb->sk);
963 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
964 	struct tcf_dump_args arg;
965 	struct tcf_proto *tp;
966 
967 	for (tp = rtnl_dereference(chain->filter_chain);
968 	     tp; tp = rtnl_dereference(tp->next), (*p_index)++) {
969 		if (*p_index < index_start)
970 			continue;
971 		if (TC_H_MAJ(tcm->tcm_info) &&
972 		    TC_H_MAJ(tcm->tcm_info) != tp->prio)
973 			continue;
974 		if (TC_H_MIN(tcm->tcm_info) &&
975 		    TC_H_MIN(tcm->tcm_info) != tp->protocol)
976 			continue;
977 		if (*p_index > index_start)
978 			memset(&cb->args[1], 0,
979 			       sizeof(cb->args) - sizeof(cb->args[0]));
980 		if (cb->args[1] == 0) {
981 			if (tcf_fill_node(net, skb, tp, q, parent, 0,
982 					  NETLINK_CB(cb->skb).portid,
983 					  cb->nlh->nlmsg_seq, NLM_F_MULTI,
984 					  RTM_NEWTFILTER) <= 0)
985 				return false;
986 
987 			cb->args[1] = 1;
988 		}
989 		if (!tp->ops->walk)
990 			continue;
991 		arg.w.fn = tcf_node_dump;
992 		arg.skb = skb;
993 		arg.cb = cb;
994 		arg.q = q;
995 		arg.parent = parent;
996 		arg.w.stop = 0;
997 		arg.w.skip = cb->args[1] - 1;
998 		arg.w.count = 0;
999 		tp->ops->walk(tp, &arg.w);
1000 		cb->args[1] = arg.w.count + 1;
1001 		if (arg.w.stop)
1002 			return false;
1003 	}
1004 	return true;
1005 }
1006 
1007 /* called with RTNL */
1008 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
1009 {
1010 	struct net *net = sock_net(skb->sk);
1011 	struct nlattr *tca[TCA_MAX + 1];
1012 	struct net_device *dev;
1013 	struct Qdisc *q;
1014 	struct tcf_block *block;
1015 	struct tcf_chain *chain;
1016 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
1017 	unsigned long cl = 0;
1018 	const struct Qdisc_class_ops *cops;
1019 	long index_start;
1020 	long index;
1021 	u32 parent;
1022 	int err;
1023 
1024 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1025 		return skb->len;
1026 
1027 	err = nlmsg_parse(cb->nlh, sizeof(*tcm), tca, TCA_MAX, NULL, NULL);
1028 	if (err)
1029 		return err;
1030 
1031 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1032 	if (!dev)
1033 		return skb->len;
1034 
1035 	parent = tcm->tcm_parent;
1036 	if (!parent) {
1037 		q = dev->qdisc;
1038 		parent = q->handle;
1039 	} else {
1040 		q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
1041 	}
1042 	if (!q)
1043 		goto out;
1044 	cops = q->ops->cl_ops;
1045 	if (!cops)
1046 		goto out;
1047 	if (!cops->tcf_block)
1048 		goto out;
1049 	if (TC_H_MIN(tcm->tcm_parent)) {
1050 		cl = cops->find(q, tcm->tcm_parent);
1051 		if (cl == 0)
1052 			goto out;
1053 	}
1054 	block = cops->tcf_block(q, cl);
1055 	if (!block)
1056 		goto out;
1057 
1058 	index_start = cb->args[0];
1059 	index = 0;
1060 
1061 	list_for_each_entry(chain, &block->chain_list, list) {
1062 		if (tca[TCA_CHAIN] &&
1063 		    nla_get_u32(tca[TCA_CHAIN]) != chain->index)
1064 			continue;
1065 		if (!tcf_chain_dump(chain, q, parent, skb, cb,
1066 				    index_start, &index))
1067 			break;
1068 	}
1069 
1070 	cb->args[0] = index;
1071 
1072 out:
1073 	return skb->len;
1074 }
1075 
1076 void tcf_exts_destroy(struct tcf_exts *exts)
1077 {
1078 #ifdef CONFIG_NET_CLS_ACT
1079 	LIST_HEAD(actions);
1080 
1081 	ASSERT_RTNL();
1082 	tcf_exts_to_list(exts, &actions);
1083 	tcf_action_destroy(&actions, TCA_ACT_UNBIND);
1084 	kfree(exts->actions);
1085 	exts->nr_actions = 0;
1086 #endif
1087 }
1088 EXPORT_SYMBOL(tcf_exts_destroy);
1089 
1090 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
1091 		      struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr)
1092 {
1093 #ifdef CONFIG_NET_CLS_ACT
1094 	{
1095 		struct tc_action *act;
1096 
1097 		if (exts->police && tb[exts->police]) {
1098 			act = tcf_action_init_1(net, tp, tb[exts->police],
1099 						rate_tlv, "police", ovr,
1100 						TCA_ACT_BIND);
1101 			if (IS_ERR(act))
1102 				return PTR_ERR(act);
1103 
1104 			act->type = exts->type = TCA_OLD_COMPAT;
1105 			exts->actions[0] = act;
1106 			exts->nr_actions = 1;
1107 		} else if (exts->action && tb[exts->action]) {
1108 			LIST_HEAD(actions);
1109 			int err, i = 0;
1110 
1111 			err = tcf_action_init(net, tp, tb[exts->action],
1112 					      rate_tlv, NULL, ovr, TCA_ACT_BIND,
1113 					      &actions);
1114 			if (err)
1115 				return err;
1116 			list_for_each_entry(act, &actions, list)
1117 				exts->actions[i++] = act;
1118 			exts->nr_actions = i;
1119 		}
1120 		exts->net = net;
1121 	}
1122 #else
1123 	if ((exts->action && tb[exts->action]) ||
1124 	    (exts->police && tb[exts->police]))
1125 		return -EOPNOTSUPP;
1126 #endif
1127 
1128 	return 0;
1129 }
1130 EXPORT_SYMBOL(tcf_exts_validate);
1131 
1132 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
1133 {
1134 #ifdef CONFIG_NET_CLS_ACT
1135 	struct tcf_exts old = *dst;
1136 
1137 	*dst = *src;
1138 	tcf_exts_destroy(&old);
1139 #endif
1140 }
1141 EXPORT_SYMBOL(tcf_exts_change);
1142 
1143 #ifdef CONFIG_NET_CLS_ACT
1144 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
1145 {
1146 	if (exts->nr_actions == 0)
1147 		return NULL;
1148 	else
1149 		return exts->actions[0];
1150 }
1151 #endif
1152 
1153 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
1154 {
1155 #ifdef CONFIG_NET_CLS_ACT
1156 	struct nlattr *nest;
1157 
1158 	if (exts->action && tcf_exts_has_actions(exts)) {
1159 		/*
1160 		 * again for backward compatible mode - we want
1161 		 * to work with both old and new modes of entering
1162 		 * tc data even if iproute2  was newer - jhs
1163 		 */
1164 		if (exts->type != TCA_OLD_COMPAT) {
1165 			LIST_HEAD(actions);
1166 
1167 			nest = nla_nest_start(skb, exts->action);
1168 			if (nest == NULL)
1169 				goto nla_put_failure;
1170 
1171 			tcf_exts_to_list(exts, &actions);
1172 			if (tcf_action_dump(skb, &actions, 0, 0) < 0)
1173 				goto nla_put_failure;
1174 			nla_nest_end(skb, nest);
1175 		} else if (exts->police) {
1176 			struct tc_action *act = tcf_exts_first_act(exts);
1177 			nest = nla_nest_start(skb, exts->police);
1178 			if (nest == NULL || !act)
1179 				goto nla_put_failure;
1180 			if (tcf_action_dump_old(skb, act, 0, 0) < 0)
1181 				goto nla_put_failure;
1182 			nla_nest_end(skb, nest);
1183 		}
1184 	}
1185 	return 0;
1186 
1187 nla_put_failure:
1188 	nla_nest_cancel(skb, nest);
1189 	return -1;
1190 #else
1191 	return 0;
1192 #endif
1193 }
1194 EXPORT_SYMBOL(tcf_exts_dump);
1195 
1196 
1197 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
1198 {
1199 #ifdef CONFIG_NET_CLS_ACT
1200 	struct tc_action *a = tcf_exts_first_act(exts);
1201 	if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
1202 		return -1;
1203 #endif
1204 	return 0;
1205 }
1206 EXPORT_SYMBOL(tcf_exts_dump_stats);
1207 
1208 static int tc_exts_setup_cb_egdev_call(struct tcf_exts *exts,
1209 				       enum tc_setup_type type,
1210 				       void *type_data, bool err_stop)
1211 {
1212 	int ok_count = 0;
1213 #ifdef CONFIG_NET_CLS_ACT
1214 	const struct tc_action *a;
1215 	struct net_device *dev;
1216 	int i, ret;
1217 
1218 	if (!tcf_exts_has_actions(exts))
1219 		return 0;
1220 
1221 	for (i = 0; i < exts->nr_actions; i++) {
1222 		a = exts->actions[i];
1223 		if (!a->ops->get_dev)
1224 			continue;
1225 		dev = a->ops->get_dev(a);
1226 		if (!dev)
1227 			continue;
1228 		ret = tc_setup_cb_egdev_call(dev, type, type_data, err_stop);
1229 		if (ret < 0)
1230 			return ret;
1231 		ok_count += ret;
1232 	}
1233 #endif
1234 	return ok_count;
1235 }
1236 
1237 int tc_setup_cb_call(struct tcf_block *block, struct tcf_exts *exts,
1238 		     enum tc_setup_type type, void *type_data, bool err_stop)
1239 {
1240 	int ok_count;
1241 	int ret;
1242 
1243 	ret = tcf_block_cb_call(block, type, type_data, err_stop);
1244 	if (ret < 0)
1245 		return ret;
1246 	ok_count = ret;
1247 
1248 	if (!exts)
1249 		return ok_count;
1250 	ret = tc_exts_setup_cb_egdev_call(exts, type, type_data, err_stop);
1251 	if (ret < 0)
1252 		return ret;
1253 	ok_count += ret;
1254 
1255 	return ok_count;
1256 }
1257 EXPORT_SYMBOL(tc_setup_cb_call);
1258 
1259 static int __init tc_filter_init(void)
1260 {
1261 	tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
1262 	if (!tc_filter_wq)
1263 		return -ENOMEM;
1264 
1265 	rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, 0);
1266 	rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, 0);
1267 	rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
1268 		      tc_dump_tfilter, 0);
1269 
1270 	return 0;
1271 }
1272 
1273 subsys_initcall(tc_filter_init);
1274