xref: /linux/net/sched/cls_api.c (revision b24413180f5600bcb3bb70fbed5cf186b60864bd)
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/err.h>
27 #include <linux/slab.h>
28 #include <net/net_namespace.h>
29 #include <net/sock.h>
30 #include <net/netlink.h>
31 #include <net/pkt_sched.h>
32 #include <net/pkt_cls.h>
33 
34 /* The list of all installed classifier types */
35 static LIST_HEAD(tcf_proto_base);
36 
37 /* Protects list of registered TC modules. It is pure SMP lock. */
38 static DEFINE_RWLOCK(cls_mod_lock);
39 
40 /* Find classifier type by string name */
41 
42 static const struct tcf_proto_ops *tcf_proto_lookup_ops(const char *kind)
43 {
44 	const struct tcf_proto_ops *t, *res = NULL;
45 
46 	if (kind) {
47 		read_lock(&cls_mod_lock);
48 		list_for_each_entry(t, &tcf_proto_base, head) {
49 			if (strcmp(kind, t->kind) == 0) {
50 				if (try_module_get(t->owner))
51 					res = t;
52 				break;
53 			}
54 		}
55 		read_unlock(&cls_mod_lock);
56 	}
57 	return res;
58 }
59 
60 /* Register(unregister) new classifier type */
61 
62 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
63 {
64 	struct tcf_proto_ops *t;
65 	int rc = -EEXIST;
66 
67 	write_lock(&cls_mod_lock);
68 	list_for_each_entry(t, &tcf_proto_base, head)
69 		if (!strcmp(ops->kind, t->kind))
70 			goto out;
71 
72 	list_add_tail(&ops->head, &tcf_proto_base);
73 	rc = 0;
74 out:
75 	write_unlock(&cls_mod_lock);
76 	return rc;
77 }
78 EXPORT_SYMBOL(register_tcf_proto_ops);
79 
80 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
81 {
82 	struct tcf_proto_ops *t;
83 	int rc = -ENOENT;
84 
85 	/* Wait for outstanding call_rcu()s, if any, from a
86 	 * tcf_proto_ops's destroy() handler.
87 	 */
88 	rcu_barrier();
89 
90 	write_lock(&cls_mod_lock);
91 	list_for_each_entry(t, &tcf_proto_base, head) {
92 		if (t == ops) {
93 			list_del(&t->head);
94 			rc = 0;
95 			break;
96 		}
97 	}
98 	write_unlock(&cls_mod_lock);
99 	return rc;
100 }
101 EXPORT_SYMBOL(unregister_tcf_proto_ops);
102 
103 /* Select new prio value from the range, managed by kernel. */
104 
105 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
106 {
107 	u32 first = TC_H_MAKE(0xC0000000U, 0U);
108 
109 	if (tp)
110 		first = tp->prio - 1;
111 
112 	return TC_H_MAJ(first);
113 }
114 
115 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
116 					  u32 prio, u32 parent, struct Qdisc *q,
117 					  struct tcf_chain *chain)
118 {
119 	struct tcf_proto *tp;
120 	int err;
121 
122 	tp = kzalloc(sizeof(*tp), GFP_KERNEL);
123 	if (!tp)
124 		return ERR_PTR(-ENOBUFS);
125 
126 	err = -ENOENT;
127 	tp->ops = tcf_proto_lookup_ops(kind);
128 	if (!tp->ops) {
129 #ifdef CONFIG_MODULES
130 		rtnl_unlock();
131 		request_module("cls_%s", kind);
132 		rtnl_lock();
133 		tp->ops = tcf_proto_lookup_ops(kind);
134 		/* We dropped the RTNL semaphore in order to perform
135 		 * the module load. So, even if we succeeded in loading
136 		 * the module we have to replay the request. We indicate
137 		 * this using -EAGAIN.
138 		 */
139 		if (tp->ops) {
140 			module_put(tp->ops->owner);
141 			err = -EAGAIN;
142 		} else {
143 			err = -ENOENT;
144 		}
145 		goto errout;
146 #endif
147 	}
148 	tp->classify = tp->ops->classify;
149 	tp->protocol = protocol;
150 	tp->prio = prio;
151 	tp->classid = parent;
152 	tp->q = q;
153 	tp->chain = chain;
154 
155 	err = tp->ops->init(tp);
156 	if (err) {
157 		module_put(tp->ops->owner);
158 		goto errout;
159 	}
160 	return tp;
161 
162 errout:
163 	kfree(tp);
164 	return ERR_PTR(err);
165 }
166 
167 static void tcf_proto_destroy(struct tcf_proto *tp)
168 {
169 	tp->ops->destroy(tp);
170 	module_put(tp->ops->owner);
171 	kfree_rcu(tp, rcu);
172 }
173 
174 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
175 					  u32 chain_index)
176 {
177 	struct tcf_chain *chain;
178 
179 	chain = kzalloc(sizeof(*chain), GFP_KERNEL);
180 	if (!chain)
181 		return NULL;
182 	list_add_tail(&chain->list, &block->chain_list);
183 	chain->block = block;
184 	chain->index = chain_index;
185 	chain->refcnt = 1;
186 	return chain;
187 }
188 
189 static void tcf_chain_flush(struct tcf_chain *chain)
190 {
191 	struct tcf_proto *tp;
192 
193 	if (chain->p_filter_chain)
194 		RCU_INIT_POINTER(*chain->p_filter_chain, NULL);
195 	while ((tp = rtnl_dereference(chain->filter_chain)) != NULL) {
196 		RCU_INIT_POINTER(chain->filter_chain, tp->next);
197 		tcf_chain_put(chain);
198 		tcf_proto_destroy(tp);
199 	}
200 }
201 
202 static void tcf_chain_destroy(struct tcf_chain *chain)
203 {
204 	list_del(&chain->list);
205 	kfree(chain);
206 }
207 
208 static void tcf_chain_hold(struct tcf_chain *chain)
209 {
210 	++chain->refcnt;
211 }
212 
213 struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
214 				bool create)
215 {
216 	struct tcf_chain *chain;
217 
218 	list_for_each_entry(chain, &block->chain_list, list) {
219 		if (chain->index == chain_index) {
220 			tcf_chain_hold(chain);
221 			return chain;
222 		}
223 	}
224 
225 	return create ? tcf_chain_create(block, chain_index) : NULL;
226 }
227 EXPORT_SYMBOL(tcf_chain_get);
228 
229 void tcf_chain_put(struct tcf_chain *chain)
230 {
231 	if (--chain->refcnt == 0)
232 		tcf_chain_destroy(chain);
233 }
234 EXPORT_SYMBOL(tcf_chain_put);
235 
236 static void
237 tcf_chain_filter_chain_ptr_set(struct tcf_chain *chain,
238 			       struct tcf_proto __rcu **p_filter_chain)
239 {
240 	chain->p_filter_chain = p_filter_chain;
241 }
242 
243 int tcf_block_get(struct tcf_block **p_block,
244 		  struct tcf_proto __rcu **p_filter_chain)
245 {
246 	struct tcf_block *block = kzalloc(sizeof(*block), GFP_KERNEL);
247 	struct tcf_chain *chain;
248 	int err;
249 
250 	if (!block)
251 		return -ENOMEM;
252 	INIT_LIST_HEAD(&block->chain_list);
253 	/* Create chain 0 by default, it has to be always present. */
254 	chain = tcf_chain_create(block, 0);
255 	if (!chain) {
256 		err = -ENOMEM;
257 		goto err_chain_create;
258 	}
259 	tcf_chain_filter_chain_ptr_set(chain, p_filter_chain);
260 	*p_block = block;
261 	return 0;
262 
263 err_chain_create:
264 	kfree(block);
265 	return err;
266 }
267 EXPORT_SYMBOL(tcf_block_get);
268 
269 void tcf_block_put(struct tcf_block *block)
270 {
271 	struct tcf_chain *chain, *tmp;
272 
273 	if (!block)
274 		return;
275 
276 	/* XXX: Standalone actions are not allowed to jump to any chain, and
277 	 * bound actions should be all removed after flushing. However,
278 	 * filters are destroyed in RCU callbacks, we have to hold the chains
279 	 * first, otherwise we would always race with RCU callbacks on this list
280 	 * without proper locking.
281 	 */
282 
283 	/* Wait for existing RCU callbacks to cool down. */
284 	rcu_barrier();
285 
286 	/* Hold a refcnt for all chains, except 0, in case they are gone. */
287 	list_for_each_entry(chain, &block->chain_list, list)
288 		if (chain->index)
289 			tcf_chain_hold(chain);
290 
291 	/* No race on the list, because no chain could be destroyed. */
292 	list_for_each_entry(chain, &block->chain_list, list)
293 		tcf_chain_flush(chain);
294 
295 	/* Wait for RCU callbacks to release the reference count. */
296 	rcu_barrier();
297 
298 	/* At this point, all the chains should have refcnt == 1. */
299 	list_for_each_entry_safe(chain, tmp, &block->chain_list, list)
300 		tcf_chain_put(chain);
301 	kfree(block);
302 }
303 EXPORT_SYMBOL(tcf_block_put);
304 
305 /* Main classifier routine: scans classifier chain attached
306  * to this qdisc, (optionally) tests for protocol and asks
307  * specific classifiers.
308  */
309 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
310 		 struct tcf_result *res, bool compat_mode)
311 {
312 	__be16 protocol = tc_skb_protocol(skb);
313 #ifdef CONFIG_NET_CLS_ACT
314 	const int max_reclassify_loop = 4;
315 	const struct tcf_proto *orig_tp = tp;
316 	const struct tcf_proto *first_tp;
317 	int limit = 0;
318 
319 reclassify:
320 #endif
321 	for (; tp; tp = rcu_dereference_bh(tp->next)) {
322 		int err;
323 
324 		if (tp->protocol != protocol &&
325 		    tp->protocol != htons(ETH_P_ALL))
326 			continue;
327 
328 		err = tp->classify(skb, tp, res);
329 #ifdef CONFIG_NET_CLS_ACT
330 		if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
331 			first_tp = orig_tp;
332 			goto reset;
333 		} else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
334 			first_tp = res->goto_tp;
335 			goto reset;
336 		}
337 #endif
338 		if (err >= 0)
339 			return err;
340 	}
341 
342 	return TC_ACT_UNSPEC; /* signal: continue lookup */
343 #ifdef CONFIG_NET_CLS_ACT
344 reset:
345 	if (unlikely(limit++ >= max_reclassify_loop)) {
346 		net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n",
347 				       tp->q->ops->id, tp->prio & 0xffff,
348 				       ntohs(tp->protocol));
349 		return TC_ACT_SHOT;
350 	}
351 
352 	tp = first_tp;
353 	protocol = tc_skb_protocol(skb);
354 	goto reclassify;
355 #endif
356 }
357 EXPORT_SYMBOL(tcf_classify);
358 
359 struct tcf_chain_info {
360 	struct tcf_proto __rcu **pprev;
361 	struct tcf_proto __rcu *next;
362 };
363 
364 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain_info *chain_info)
365 {
366 	return rtnl_dereference(*chain_info->pprev);
367 }
368 
369 static void tcf_chain_tp_insert(struct tcf_chain *chain,
370 				struct tcf_chain_info *chain_info,
371 				struct tcf_proto *tp)
372 {
373 	if (chain->p_filter_chain &&
374 	    *chain_info->pprev == chain->filter_chain)
375 		rcu_assign_pointer(*chain->p_filter_chain, tp);
376 	RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain_info));
377 	rcu_assign_pointer(*chain_info->pprev, tp);
378 	tcf_chain_hold(chain);
379 }
380 
381 static void tcf_chain_tp_remove(struct tcf_chain *chain,
382 				struct tcf_chain_info *chain_info,
383 				struct tcf_proto *tp)
384 {
385 	struct tcf_proto *next = rtnl_dereference(chain_info->next);
386 
387 	if (chain->p_filter_chain && tp == chain->filter_chain)
388 		RCU_INIT_POINTER(*chain->p_filter_chain, next);
389 	RCU_INIT_POINTER(*chain_info->pprev, next);
390 	tcf_chain_put(chain);
391 }
392 
393 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
394 					   struct tcf_chain_info *chain_info,
395 					   u32 protocol, u32 prio,
396 					   bool prio_allocate)
397 {
398 	struct tcf_proto **pprev;
399 	struct tcf_proto *tp;
400 
401 	/* Check the chain for existence of proto-tcf with this priority */
402 	for (pprev = &chain->filter_chain;
403 	     (tp = rtnl_dereference(*pprev)); pprev = &tp->next) {
404 		if (tp->prio >= prio) {
405 			if (tp->prio == prio) {
406 				if (prio_allocate ||
407 				    (tp->protocol != protocol && protocol))
408 					return ERR_PTR(-EINVAL);
409 			} else {
410 				tp = NULL;
411 			}
412 			break;
413 		}
414 	}
415 	chain_info->pprev = pprev;
416 	chain_info->next = tp ? tp->next : NULL;
417 	return tp;
418 }
419 
420 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
421 			 struct tcf_proto *tp, void *fh, u32 portid,
422 			 u32 seq, u16 flags, int event)
423 {
424 	struct tcmsg *tcm;
425 	struct nlmsghdr  *nlh;
426 	unsigned char *b = skb_tail_pointer(skb);
427 
428 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
429 	if (!nlh)
430 		goto out_nlmsg_trim;
431 	tcm = nlmsg_data(nlh);
432 	tcm->tcm_family = AF_UNSPEC;
433 	tcm->tcm__pad1 = 0;
434 	tcm->tcm__pad2 = 0;
435 	tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
436 	tcm->tcm_parent = tp->classid;
437 	tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
438 	if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
439 		goto nla_put_failure;
440 	if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
441 		goto nla_put_failure;
442 	if (!fh) {
443 		tcm->tcm_handle = 0;
444 	} else {
445 		if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
446 			goto nla_put_failure;
447 	}
448 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
449 	return skb->len;
450 
451 out_nlmsg_trim:
452 nla_put_failure:
453 	nlmsg_trim(skb, b);
454 	return -1;
455 }
456 
457 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
458 			  struct nlmsghdr *n, struct tcf_proto *tp,
459 			  void *fh, int event, bool unicast)
460 {
461 	struct sk_buff *skb;
462 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
463 
464 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
465 	if (!skb)
466 		return -ENOBUFS;
467 
468 	if (tcf_fill_node(net, skb, tp, fh, portid, n->nlmsg_seq,
469 			  n->nlmsg_flags, event) <= 0) {
470 		kfree_skb(skb);
471 		return -EINVAL;
472 	}
473 
474 	if (unicast)
475 		return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
476 
477 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
478 			      n->nlmsg_flags & NLM_F_ECHO);
479 }
480 
481 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
482 			      struct nlmsghdr *n, struct tcf_proto *tp,
483 			      void *fh, bool unicast, bool *last)
484 {
485 	struct sk_buff *skb;
486 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
487 	int err;
488 
489 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
490 	if (!skb)
491 		return -ENOBUFS;
492 
493 	if (tcf_fill_node(net, skb, tp, fh, portid, n->nlmsg_seq,
494 			  n->nlmsg_flags, RTM_DELTFILTER) <= 0) {
495 		kfree_skb(skb);
496 		return -EINVAL;
497 	}
498 
499 	err = tp->ops->delete(tp, fh, last);
500 	if (err) {
501 		kfree_skb(skb);
502 		return err;
503 	}
504 
505 	if (unicast)
506 		return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
507 
508 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
509 			      n->nlmsg_flags & NLM_F_ECHO);
510 }
511 
512 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
513 				 struct nlmsghdr *n,
514 				 struct tcf_chain *chain, int event)
515 {
516 	struct tcf_proto *tp;
517 
518 	for (tp = rtnl_dereference(chain->filter_chain);
519 	     tp; tp = rtnl_dereference(tp->next))
520 		tfilter_notify(net, oskb, n, tp, 0, event, false);
521 }
522 
523 /* Add/change/delete/get a filter node */
524 
525 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
526 			  struct netlink_ext_ack *extack)
527 {
528 	struct net *net = sock_net(skb->sk);
529 	struct nlattr *tca[TCA_MAX + 1];
530 	struct tcmsg *t;
531 	u32 protocol;
532 	u32 prio;
533 	bool prio_allocate;
534 	u32 parent;
535 	u32 chain_index;
536 	struct net_device *dev;
537 	struct Qdisc  *q;
538 	struct tcf_chain_info chain_info;
539 	struct tcf_chain *chain = NULL;
540 	struct tcf_block *block;
541 	struct tcf_proto *tp;
542 	const struct Qdisc_class_ops *cops;
543 	unsigned long cl;
544 	void *fh;
545 	int err;
546 	int tp_created;
547 
548 	if ((n->nlmsg_type != RTM_GETTFILTER) &&
549 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
550 		return -EPERM;
551 
552 replay:
553 	tp_created = 0;
554 
555 	err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack);
556 	if (err < 0)
557 		return err;
558 
559 	t = nlmsg_data(n);
560 	protocol = TC_H_MIN(t->tcm_info);
561 	prio = TC_H_MAJ(t->tcm_info);
562 	prio_allocate = false;
563 	parent = t->tcm_parent;
564 	cl = 0;
565 
566 	if (prio == 0) {
567 		switch (n->nlmsg_type) {
568 		case RTM_DELTFILTER:
569 			if (protocol || t->tcm_handle || tca[TCA_KIND])
570 				return -ENOENT;
571 			break;
572 		case RTM_NEWTFILTER:
573 			/* If no priority is provided by the user,
574 			 * we allocate one.
575 			 */
576 			if (n->nlmsg_flags & NLM_F_CREATE) {
577 				prio = TC_H_MAKE(0x80000000U, 0U);
578 				prio_allocate = true;
579 				break;
580 			}
581 			/* fall-through */
582 		default:
583 			return -ENOENT;
584 		}
585 	}
586 
587 	/* Find head of filter chain. */
588 
589 	/* Find link */
590 	dev = __dev_get_by_index(net, t->tcm_ifindex);
591 	if (dev == NULL)
592 		return -ENODEV;
593 
594 	/* Find qdisc */
595 	if (!parent) {
596 		q = dev->qdisc;
597 		parent = q->handle;
598 	} else {
599 		q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
600 		if (q == NULL)
601 			return -EINVAL;
602 	}
603 
604 	/* Is it classful? */
605 	cops = q->ops->cl_ops;
606 	if (!cops)
607 		return -EINVAL;
608 
609 	if (!cops->tcf_block)
610 		return -EOPNOTSUPP;
611 
612 	/* Do we search for filter, attached to class? */
613 	if (TC_H_MIN(parent)) {
614 		cl = cops->find(q, parent);
615 		if (cl == 0)
616 			return -ENOENT;
617 	}
618 
619 	/* And the last stroke */
620 	block = cops->tcf_block(q, cl);
621 	if (!block) {
622 		err = -EINVAL;
623 		goto errout;
624 	}
625 
626 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
627 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
628 		err = -EINVAL;
629 		goto errout;
630 	}
631 	chain = tcf_chain_get(block, chain_index,
632 			      n->nlmsg_type == RTM_NEWTFILTER);
633 	if (!chain) {
634 		err = n->nlmsg_type == RTM_NEWTFILTER ? -ENOMEM : -EINVAL;
635 		goto errout;
636 	}
637 
638 	if (n->nlmsg_type == RTM_DELTFILTER && prio == 0) {
639 		tfilter_notify_chain(net, skb, n, chain, RTM_DELTFILTER);
640 		tcf_chain_flush(chain);
641 		err = 0;
642 		goto errout;
643 	}
644 
645 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
646 			       prio, prio_allocate);
647 	if (IS_ERR(tp)) {
648 		err = PTR_ERR(tp);
649 		goto errout;
650 	}
651 
652 	if (tp == NULL) {
653 		/* Proto-tcf does not exist, create new one */
654 
655 		if (tca[TCA_KIND] == NULL || !protocol) {
656 			err = -EINVAL;
657 			goto errout;
658 		}
659 
660 		if (n->nlmsg_type != RTM_NEWTFILTER ||
661 		    !(n->nlmsg_flags & NLM_F_CREATE)) {
662 			err = -ENOENT;
663 			goto errout;
664 		}
665 
666 		if (prio_allocate)
667 			prio = tcf_auto_prio(tcf_chain_tp_prev(&chain_info));
668 
669 		tp = tcf_proto_create(nla_data(tca[TCA_KIND]),
670 				      protocol, prio, parent, q, chain);
671 		if (IS_ERR(tp)) {
672 			err = PTR_ERR(tp);
673 			goto errout;
674 		}
675 		tp_created = 1;
676 	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
677 		err = -EINVAL;
678 		goto errout;
679 	}
680 
681 	fh = tp->ops->get(tp, t->tcm_handle);
682 
683 	if (!fh) {
684 		if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
685 			tcf_chain_tp_remove(chain, &chain_info, tp);
686 			tfilter_notify(net, skb, n, tp, fh,
687 				       RTM_DELTFILTER, false);
688 			tcf_proto_destroy(tp);
689 			err = 0;
690 			goto errout;
691 		}
692 
693 		if (n->nlmsg_type != RTM_NEWTFILTER ||
694 		    !(n->nlmsg_flags & NLM_F_CREATE)) {
695 			err = -ENOENT;
696 			goto errout;
697 		}
698 	} else {
699 		bool last;
700 
701 		switch (n->nlmsg_type) {
702 		case RTM_NEWTFILTER:
703 			if (n->nlmsg_flags & NLM_F_EXCL) {
704 				if (tp_created)
705 					tcf_proto_destroy(tp);
706 				err = -EEXIST;
707 				goto errout;
708 			}
709 			break;
710 		case RTM_DELTFILTER:
711 			err = tfilter_del_notify(net, skb, n, tp, fh, false,
712 						 &last);
713 			if (err)
714 				goto errout;
715 			if (last) {
716 				tcf_chain_tp_remove(chain, &chain_info, tp);
717 				tcf_proto_destroy(tp);
718 			}
719 			goto errout;
720 		case RTM_GETTFILTER:
721 			err = tfilter_notify(net, skb, n, tp, fh,
722 					     RTM_NEWTFILTER, true);
723 			goto errout;
724 		default:
725 			err = -EINVAL;
726 			goto errout;
727 		}
728 	}
729 
730 	err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
731 			      n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE);
732 	if (err == 0) {
733 		if (tp_created)
734 			tcf_chain_tp_insert(chain, &chain_info, tp);
735 		tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER, false);
736 	} else {
737 		if (tp_created)
738 			tcf_proto_destroy(tp);
739 	}
740 
741 errout:
742 	if (chain)
743 		tcf_chain_put(chain);
744 	if (err == -EAGAIN)
745 		/* Replay the request. */
746 		goto replay;
747 	return err;
748 }
749 
750 struct tcf_dump_args {
751 	struct tcf_walker w;
752 	struct sk_buff *skb;
753 	struct netlink_callback *cb;
754 };
755 
756 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
757 {
758 	struct tcf_dump_args *a = (void *)arg;
759 	struct net *net = sock_net(a->skb->sk);
760 
761 	return tcf_fill_node(net, a->skb, tp, n, NETLINK_CB(a->cb->skb).portid,
762 			     a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
763 			     RTM_NEWTFILTER);
764 }
765 
766 static bool tcf_chain_dump(struct tcf_chain *chain, struct sk_buff *skb,
767 			   struct netlink_callback *cb,
768 			   long index_start, long *p_index)
769 {
770 	struct net *net = sock_net(skb->sk);
771 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
772 	struct tcf_dump_args arg;
773 	struct tcf_proto *tp;
774 
775 	for (tp = rtnl_dereference(chain->filter_chain);
776 	     tp; tp = rtnl_dereference(tp->next), (*p_index)++) {
777 		if (*p_index < index_start)
778 			continue;
779 		if (TC_H_MAJ(tcm->tcm_info) &&
780 		    TC_H_MAJ(tcm->tcm_info) != tp->prio)
781 			continue;
782 		if (TC_H_MIN(tcm->tcm_info) &&
783 		    TC_H_MIN(tcm->tcm_info) != tp->protocol)
784 			continue;
785 		if (*p_index > index_start)
786 			memset(&cb->args[1], 0,
787 			       sizeof(cb->args) - sizeof(cb->args[0]));
788 		if (cb->args[1] == 0) {
789 			if (tcf_fill_node(net, skb, tp, 0,
790 					  NETLINK_CB(cb->skb).portid,
791 					  cb->nlh->nlmsg_seq, NLM_F_MULTI,
792 					  RTM_NEWTFILTER) <= 0)
793 				return false;
794 
795 			cb->args[1] = 1;
796 		}
797 		if (!tp->ops->walk)
798 			continue;
799 		arg.w.fn = tcf_node_dump;
800 		arg.skb = skb;
801 		arg.cb = cb;
802 		arg.w.stop = 0;
803 		arg.w.skip = cb->args[1] - 1;
804 		arg.w.count = 0;
805 		tp->ops->walk(tp, &arg.w);
806 		cb->args[1] = arg.w.count + 1;
807 		if (arg.w.stop)
808 			return false;
809 	}
810 	return true;
811 }
812 
813 /* called with RTNL */
814 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
815 {
816 	struct net *net = sock_net(skb->sk);
817 	struct nlattr *tca[TCA_MAX + 1];
818 	struct net_device *dev;
819 	struct Qdisc *q;
820 	struct tcf_block *block;
821 	struct tcf_chain *chain;
822 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
823 	unsigned long cl = 0;
824 	const struct Qdisc_class_ops *cops;
825 	long index_start;
826 	long index;
827 	int err;
828 
829 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
830 		return skb->len;
831 
832 	err = nlmsg_parse(cb->nlh, sizeof(*tcm), tca, TCA_MAX, NULL, NULL);
833 	if (err)
834 		return err;
835 
836 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
837 	if (!dev)
838 		return skb->len;
839 
840 	if (!tcm->tcm_parent)
841 		q = dev->qdisc;
842 	else
843 		q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
844 	if (!q)
845 		goto out;
846 	cops = q->ops->cl_ops;
847 	if (!cops)
848 		goto out;
849 	if (!cops->tcf_block)
850 		goto out;
851 	if (TC_H_MIN(tcm->tcm_parent)) {
852 		cl = cops->find(q, tcm->tcm_parent);
853 		if (cl == 0)
854 			goto out;
855 	}
856 	block = cops->tcf_block(q, cl);
857 	if (!block)
858 		goto out;
859 
860 	index_start = cb->args[0];
861 	index = 0;
862 
863 	list_for_each_entry(chain, &block->chain_list, list) {
864 		if (tca[TCA_CHAIN] &&
865 		    nla_get_u32(tca[TCA_CHAIN]) != chain->index)
866 			continue;
867 		if (!tcf_chain_dump(chain, skb, cb, index_start, &index))
868 			break;
869 	}
870 
871 	cb->args[0] = index;
872 
873 out:
874 	return skb->len;
875 }
876 
877 void tcf_exts_destroy(struct tcf_exts *exts)
878 {
879 #ifdef CONFIG_NET_CLS_ACT
880 	LIST_HEAD(actions);
881 
882 	tcf_exts_to_list(exts, &actions);
883 	tcf_action_destroy(&actions, TCA_ACT_UNBIND);
884 	kfree(exts->actions);
885 	exts->nr_actions = 0;
886 #endif
887 }
888 EXPORT_SYMBOL(tcf_exts_destroy);
889 
890 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
891 		      struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr)
892 {
893 #ifdef CONFIG_NET_CLS_ACT
894 	{
895 		struct tc_action *act;
896 
897 		if (exts->police && tb[exts->police]) {
898 			act = tcf_action_init_1(net, tp, tb[exts->police],
899 						rate_tlv, "police", ovr,
900 						TCA_ACT_BIND);
901 			if (IS_ERR(act))
902 				return PTR_ERR(act);
903 
904 			act->type = exts->type = TCA_OLD_COMPAT;
905 			exts->actions[0] = act;
906 			exts->nr_actions = 1;
907 		} else if (exts->action && tb[exts->action]) {
908 			LIST_HEAD(actions);
909 			int err, i = 0;
910 
911 			err = tcf_action_init(net, tp, tb[exts->action],
912 					      rate_tlv, NULL, ovr, TCA_ACT_BIND,
913 					      &actions);
914 			if (err)
915 				return err;
916 			list_for_each_entry(act, &actions, list)
917 				exts->actions[i++] = act;
918 			exts->nr_actions = i;
919 		}
920 	}
921 #else
922 	if ((exts->action && tb[exts->action]) ||
923 	    (exts->police && tb[exts->police]))
924 		return -EOPNOTSUPP;
925 #endif
926 
927 	return 0;
928 }
929 EXPORT_SYMBOL(tcf_exts_validate);
930 
931 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
932 {
933 #ifdef CONFIG_NET_CLS_ACT
934 	struct tcf_exts old = *dst;
935 
936 	*dst = *src;
937 	tcf_exts_destroy(&old);
938 #endif
939 }
940 EXPORT_SYMBOL(tcf_exts_change);
941 
942 #ifdef CONFIG_NET_CLS_ACT
943 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
944 {
945 	if (exts->nr_actions == 0)
946 		return NULL;
947 	else
948 		return exts->actions[0];
949 }
950 #endif
951 
952 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
953 {
954 #ifdef CONFIG_NET_CLS_ACT
955 	struct nlattr *nest;
956 
957 	if (exts->action && tcf_exts_has_actions(exts)) {
958 		/*
959 		 * again for backward compatible mode - we want
960 		 * to work with both old and new modes of entering
961 		 * tc data even if iproute2  was newer - jhs
962 		 */
963 		if (exts->type != TCA_OLD_COMPAT) {
964 			LIST_HEAD(actions);
965 
966 			nest = nla_nest_start(skb, exts->action);
967 			if (nest == NULL)
968 				goto nla_put_failure;
969 
970 			tcf_exts_to_list(exts, &actions);
971 			if (tcf_action_dump(skb, &actions, 0, 0) < 0)
972 				goto nla_put_failure;
973 			nla_nest_end(skb, nest);
974 		} else if (exts->police) {
975 			struct tc_action *act = tcf_exts_first_act(exts);
976 			nest = nla_nest_start(skb, exts->police);
977 			if (nest == NULL || !act)
978 				goto nla_put_failure;
979 			if (tcf_action_dump_old(skb, act, 0, 0) < 0)
980 				goto nla_put_failure;
981 			nla_nest_end(skb, nest);
982 		}
983 	}
984 	return 0;
985 
986 nla_put_failure:
987 	nla_nest_cancel(skb, nest);
988 	return -1;
989 #else
990 	return 0;
991 #endif
992 }
993 EXPORT_SYMBOL(tcf_exts_dump);
994 
995 
996 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
997 {
998 #ifdef CONFIG_NET_CLS_ACT
999 	struct tc_action *a = tcf_exts_first_act(exts);
1000 	if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
1001 		return -1;
1002 #endif
1003 	return 0;
1004 }
1005 EXPORT_SYMBOL(tcf_exts_dump_stats);
1006 
1007 int tcf_exts_get_dev(struct net_device *dev, struct tcf_exts *exts,
1008 		     struct net_device **hw_dev)
1009 {
1010 #ifdef CONFIG_NET_CLS_ACT
1011 	const struct tc_action *a;
1012 	LIST_HEAD(actions);
1013 
1014 	if (!tcf_exts_has_actions(exts))
1015 		return -EINVAL;
1016 
1017 	tcf_exts_to_list(exts, &actions);
1018 	list_for_each_entry(a, &actions, list) {
1019 		if (a->ops->get_dev) {
1020 			a->ops->get_dev(a, dev_net(dev), hw_dev);
1021 			break;
1022 		}
1023 	}
1024 	if (*hw_dev)
1025 		return 0;
1026 #endif
1027 	return -EOPNOTSUPP;
1028 }
1029 EXPORT_SYMBOL(tcf_exts_get_dev);
1030 
1031 static int __init tc_filter_init(void)
1032 {
1033 	rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, 0);
1034 	rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, 0);
1035 	rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
1036 		      tc_dump_tfilter, 0);
1037 
1038 	return 0;
1039 }
1040 
1041 subsys_initcall(tc_filter_init);
1042