xref: /linux/net/sched/cls_api.c (revision 83a37b3292f4aca799b355179ad6fbdd78a08e10)
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, struct Qdisc *q)
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 	block->net = qdisc_net(q);
261 	block->q = q;
262 	*p_block = block;
263 	return 0;
264 
265 err_chain_create:
266 	kfree(block);
267 	return err;
268 }
269 EXPORT_SYMBOL(tcf_block_get);
270 
271 void tcf_block_put(struct tcf_block *block)
272 {
273 	struct tcf_chain *chain, *tmp;
274 
275 	if (!block)
276 		return;
277 
278 	/* XXX: Standalone actions are not allowed to jump to any chain, and
279 	 * bound actions should be all removed after flushing. However,
280 	 * filters are destroyed in RCU callbacks, we have to hold the chains
281 	 * first, otherwise we would always race with RCU callbacks on this list
282 	 * without proper locking.
283 	 */
284 
285 	/* Wait for existing RCU callbacks to cool down. */
286 	rcu_barrier();
287 
288 	/* Hold a refcnt for all chains, except 0, in case they are gone. */
289 	list_for_each_entry(chain, &block->chain_list, list)
290 		if (chain->index)
291 			tcf_chain_hold(chain);
292 
293 	/* No race on the list, because no chain could be destroyed. */
294 	list_for_each_entry(chain, &block->chain_list, list)
295 		tcf_chain_flush(chain);
296 
297 	/* Wait for RCU callbacks to release the reference count. */
298 	rcu_barrier();
299 
300 	/* At this point, all the chains should have refcnt == 1. */
301 	list_for_each_entry_safe(chain, tmp, &block->chain_list, list)
302 		tcf_chain_put(chain);
303 	kfree(block);
304 }
305 EXPORT_SYMBOL(tcf_block_put);
306 
307 /* Main classifier routine: scans classifier chain attached
308  * to this qdisc, (optionally) tests for protocol and asks
309  * specific classifiers.
310  */
311 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
312 		 struct tcf_result *res, bool compat_mode)
313 {
314 	__be16 protocol = tc_skb_protocol(skb);
315 #ifdef CONFIG_NET_CLS_ACT
316 	const int max_reclassify_loop = 4;
317 	const struct tcf_proto *orig_tp = tp;
318 	const struct tcf_proto *first_tp;
319 	int limit = 0;
320 
321 reclassify:
322 #endif
323 	for (; tp; tp = rcu_dereference_bh(tp->next)) {
324 		int err;
325 
326 		if (tp->protocol != protocol &&
327 		    tp->protocol != htons(ETH_P_ALL))
328 			continue;
329 
330 		err = tp->classify(skb, tp, res);
331 #ifdef CONFIG_NET_CLS_ACT
332 		if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
333 			first_tp = orig_tp;
334 			goto reset;
335 		} else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
336 			first_tp = res->goto_tp;
337 			goto reset;
338 		}
339 #endif
340 		if (err >= 0)
341 			return err;
342 	}
343 
344 	return TC_ACT_UNSPEC; /* signal: continue lookup */
345 #ifdef CONFIG_NET_CLS_ACT
346 reset:
347 	if (unlikely(limit++ >= max_reclassify_loop)) {
348 		net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n",
349 				       tp->q->ops->id, tp->prio & 0xffff,
350 				       ntohs(tp->protocol));
351 		return TC_ACT_SHOT;
352 	}
353 
354 	tp = first_tp;
355 	protocol = tc_skb_protocol(skb);
356 	goto reclassify;
357 #endif
358 }
359 EXPORT_SYMBOL(tcf_classify);
360 
361 struct tcf_chain_info {
362 	struct tcf_proto __rcu **pprev;
363 	struct tcf_proto __rcu *next;
364 };
365 
366 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain_info *chain_info)
367 {
368 	return rtnl_dereference(*chain_info->pprev);
369 }
370 
371 static void tcf_chain_tp_insert(struct tcf_chain *chain,
372 				struct tcf_chain_info *chain_info,
373 				struct tcf_proto *tp)
374 {
375 	if (chain->p_filter_chain &&
376 	    *chain_info->pprev == chain->filter_chain)
377 		rcu_assign_pointer(*chain->p_filter_chain, tp);
378 	RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain_info));
379 	rcu_assign_pointer(*chain_info->pprev, tp);
380 	tcf_chain_hold(chain);
381 }
382 
383 static void tcf_chain_tp_remove(struct tcf_chain *chain,
384 				struct tcf_chain_info *chain_info,
385 				struct tcf_proto *tp)
386 {
387 	struct tcf_proto *next = rtnl_dereference(chain_info->next);
388 
389 	if (chain->p_filter_chain && tp == chain->filter_chain)
390 		RCU_INIT_POINTER(*chain->p_filter_chain, next);
391 	RCU_INIT_POINTER(*chain_info->pprev, next);
392 	tcf_chain_put(chain);
393 }
394 
395 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
396 					   struct tcf_chain_info *chain_info,
397 					   u32 protocol, u32 prio,
398 					   bool prio_allocate)
399 {
400 	struct tcf_proto **pprev;
401 	struct tcf_proto *tp;
402 
403 	/* Check the chain for existence of proto-tcf with this priority */
404 	for (pprev = &chain->filter_chain;
405 	     (tp = rtnl_dereference(*pprev)); pprev = &tp->next) {
406 		if (tp->prio >= prio) {
407 			if (tp->prio == prio) {
408 				if (prio_allocate ||
409 				    (tp->protocol != protocol && protocol))
410 					return ERR_PTR(-EINVAL);
411 			} else {
412 				tp = NULL;
413 			}
414 			break;
415 		}
416 	}
417 	chain_info->pprev = pprev;
418 	chain_info->next = tp ? tp->next : NULL;
419 	return tp;
420 }
421 
422 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
423 			 struct tcf_proto *tp, struct Qdisc *q, u32 parent,
424 			 void *fh, u32 portid, u32 seq, u16 flags, int event)
425 {
426 	struct tcmsg *tcm;
427 	struct nlmsghdr  *nlh;
428 	unsigned char *b = skb_tail_pointer(skb);
429 
430 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
431 	if (!nlh)
432 		goto out_nlmsg_trim;
433 	tcm = nlmsg_data(nlh);
434 	tcm->tcm_family = AF_UNSPEC;
435 	tcm->tcm__pad1 = 0;
436 	tcm->tcm__pad2 = 0;
437 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
438 	tcm->tcm_parent = parent;
439 	tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
440 	if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
441 		goto nla_put_failure;
442 	if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
443 		goto nla_put_failure;
444 	if (!fh) {
445 		tcm->tcm_handle = 0;
446 	} else {
447 		if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
448 			goto nla_put_failure;
449 	}
450 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
451 	return skb->len;
452 
453 out_nlmsg_trim:
454 nla_put_failure:
455 	nlmsg_trim(skb, b);
456 	return -1;
457 }
458 
459 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
460 			  struct nlmsghdr *n, struct tcf_proto *tp,
461 			  struct Qdisc *q, u32 parent,
462 			  void *fh, int event, bool unicast)
463 {
464 	struct sk_buff *skb;
465 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
466 
467 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
468 	if (!skb)
469 		return -ENOBUFS;
470 
471 	if (tcf_fill_node(net, skb, tp, q, parent, fh, portid, n->nlmsg_seq,
472 			  n->nlmsg_flags, event) <= 0) {
473 		kfree_skb(skb);
474 		return -EINVAL;
475 	}
476 
477 	if (unicast)
478 		return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
479 
480 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
481 			      n->nlmsg_flags & NLM_F_ECHO);
482 }
483 
484 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
485 			      struct nlmsghdr *n, struct tcf_proto *tp,
486 			      struct Qdisc *q, u32 parent,
487 			      void *fh, bool unicast, bool *last)
488 {
489 	struct sk_buff *skb;
490 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
491 	int err;
492 
493 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
494 	if (!skb)
495 		return -ENOBUFS;
496 
497 	if (tcf_fill_node(net, skb, tp, q, parent, fh, portid, n->nlmsg_seq,
498 			  n->nlmsg_flags, RTM_DELTFILTER) <= 0) {
499 		kfree_skb(skb);
500 		return -EINVAL;
501 	}
502 
503 	err = tp->ops->delete(tp, fh, last);
504 	if (err) {
505 		kfree_skb(skb);
506 		return err;
507 	}
508 
509 	if (unicast)
510 		return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
511 
512 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
513 			      n->nlmsg_flags & NLM_F_ECHO);
514 }
515 
516 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
517 				 struct Qdisc *q, u32 parent,
518 				 struct nlmsghdr *n,
519 				 struct tcf_chain *chain, int event)
520 {
521 	struct tcf_proto *tp;
522 
523 	for (tp = rtnl_dereference(chain->filter_chain);
524 	     tp; tp = rtnl_dereference(tp->next))
525 		tfilter_notify(net, oskb, n, tp, q, parent, 0, event, false);
526 }
527 
528 /* Add/change/delete/get a filter node */
529 
530 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
531 			  struct netlink_ext_ack *extack)
532 {
533 	struct net *net = sock_net(skb->sk);
534 	struct nlattr *tca[TCA_MAX + 1];
535 	struct tcmsg *t;
536 	u32 protocol;
537 	u32 prio;
538 	bool prio_allocate;
539 	u32 parent;
540 	u32 chain_index;
541 	struct net_device *dev;
542 	struct Qdisc  *q;
543 	struct tcf_chain_info chain_info;
544 	struct tcf_chain *chain = NULL;
545 	struct tcf_block *block;
546 	struct tcf_proto *tp;
547 	const struct Qdisc_class_ops *cops;
548 	unsigned long cl;
549 	void *fh;
550 	int err;
551 	int tp_created;
552 
553 	if ((n->nlmsg_type != RTM_GETTFILTER) &&
554 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
555 		return -EPERM;
556 
557 replay:
558 	tp_created = 0;
559 
560 	err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack);
561 	if (err < 0)
562 		return err;
563 
564 	t = nlmsg_data(n);
565 	protocol = TC_H_MIN(t->tcm_info);
566 	prio = TC_H_MAJ(t->tcm_info);
567 	prio_allocate = false;
568 	parent = t->tcm_parent;
569 	cl = 0;
570 
571 	if (prio == 0) {
572 		switch (n->nlmsg_type) {
573 		case RTM_DELTFILTER:
574 			if (protocol || t->tcm_handle || tca[TCA_KIND])
575 				return -ENOENT;
576 			break;
577 		case RTM_NEWTFILTER:
578 			/* If no priority is provided by the user,
579 			 * we allocate one.
580 			 */
581 			if (n->nlmsg_flags & NLM_F_CREATE) {
582 				prio = TC_H_MAKE(0x80000000U, 0U);
583 				prio_allocate = true;
584 				break;
585 			}
586 			/* fall-through */
587 		default:
588 			return -ENOENT;
589 		}
590 	}
591 
592 	/* Find head of filter chain. */
593 
594 	/* Find link */
595 	dev = __dev_get_by_index(net, t->tcm_ifindex);
596 	if (dev == NULL)
597 		return -ENODEV;
598 
599 	/* Find qdisc */
600 	if (!parent) {
601 		q = dev->qdisc;
602 		parent = q->handle;
603 	} else {
604 		q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
605 		if (q == NULL)
606 			return -EINVAL;
607 	}
608 
609 	/* Is it classful? */
610 	cops = q->ops->cl_ops;
611 	if (!cops)
612 		return -EINVAL;
613 
614 	if (!cops->tcf_block)
615 		return -EOPNOTSUPP;
616 
617 	/* Do we search for filter, attached to class? */
618 	if (TC_H_MIN(parent)) {
619 		cl = cops->find(q, parent);
620 		if (cl == 0)
621 			return -ENOENT;
622 	}
623 
624 	/* And the last stroke */
625 	block = cops->tcf_block(q, cl);
626 	if (!block) {
627 		err = -EINVAL;
628 		goto errout;
629 	}
630 
631 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
632 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
633 		err = -EINVAL;
634 		goto errout;
635 	}
636 	chain = tcf_chain_get(block, chain_index,
637 			      n->nlmsg_type == RTM_NEWTFILTER);
638 	if (!chain) {
639 		err = n->nlmsg_type == RTM_NEWTFILTER ? -ENOMEM : -EINVAL;
640 		goto errout;
641 	}
642 
643 	if (n->nlmsg_type == RTM_DELTFILTER && prio == 0) {
644 		tfilter_notify_chain(net, skb, q, parent, n,
645 				     chain, RTM_DELTFILTER);
646 		tcf_chain_flush(chain);
647 		err = 0;
648 		goto errout;
649 	}
650 
651 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
652 			       prio, prio_allocate);
653 	if (IS_ERR(tp)) {
654 		err = PTR_ERR(tp);
655 		goto errout;
656 	}
657 
658 	if (tp == NULL) {
659 		/* Proto-tcf does not exist, create new one */
660 
661 		if (tca[TCA_KIND] == NULL || !protocol) {
662 			err = -EINVAL;
663 			goto errout;
664 		}
665 
666 		if (n->nlmsg_type != RTM_NEWTFILTER ||
667 		    !(n->nlmsg_flags & NLM_F_CREATE)) {
668 			err = -ENOENT;
669 			goto errout;
670 		}
671 
672 		if (prio_allocate)
673 			prio = tcf_auto_prio(tcf_chain_tp_prev(&chain_info));
674 
675 		tp = tcf_proto_create(nla_data(tca[TCA_KIND]),
676 				      protocol, prio, parent, q, chain);
677 		if (IS_ERR(tp)) {
678 			err = PTR_ERR(tp);
679 			goto errout;
680 		}
681 		tp_created = 1;
682 	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
683 		err = -EINVAL;
684 		goto errout;
685 	}
686 
687 	fh = tp->ops->get(tp, t->tcm_handle);
688 
689 	if (!fh) {
690 		if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
691 			tcf_chain_tp_remove(chain, &chain_info, tp);
692 			tfilter_notify(net, skb, n, tp, q, parent, fh,
693 				       RTM_DELTFILTER, false);
694 			tcf_proto_destroy(tp);
695 			err = 0;
696 			goto errout;
697 		}
698 
699 		if (n->nlmsg_type != RTM_NEWTFILTER ||
700 		    !(n->nlmsg_flags & NLM_F_CREATE)) {
701 			err = -ENOENT;
702 			goto errout;
703 		}
704 	} else {
705 		bool last;
706 
707 		switch (n->nlmsg_type) {
708 		case RTM_NEWTFILTER:
709 			if (n->nlmsg_flags & NLM_F_EXCL) {
710 				if (tp_created)
711 					tcf_proto_destroy(tp);
712 				err = -EEXIST;
713 				goto errout;
714 			}
715 			break;
716 		case RTM_DELTFILTER:
717 			err = tfilter_del_notify(net, skb, n, tp, q, parent,
718 						 fh, false, &last);
719 			if (err)
720 				goto errout;
721 			if (last) {
722 				tcf_chain_tp_remove(chain, &chain_info, tp);
723 				tcf_proto_destroy(tp);
724 			}
725 			goto errout;
726 		case RTM_GETTFILTER:
727 			err = tfilter_notify(net, skb, n, tp, q, parent, fh,
728 					     RTM_NEWTFILTER, true);
729 			goto errout;
730 		default:
731 			err = -EINVAL;
732 			goto errout;
733 		}
734 	}
735 
736 	err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
737 			      n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE);
738 	if (err == 0) {
739 		if (tp_created)
740 			tcf_chain_tp_insert(chain, &chain_info, tp);
741 		tfilter_notify(net, skb, n, tp, q, parent, fh,
742 			       RTM_NEWTFILTER, false);
743 	} else {
744 		if (tp_created)
745 			tcf_proto_destroy(tp);
746 	}
747 
748 errout:
749 	if (chain)
750 		tcf_chain_put(chain);
751 	if (err == -EAGAIN)
752 		/* Replay the request. */
753 		goto replay;
754 	return err;
755 }
756 
757 struct tcf_dump_args {
758 	struct tcf_walker w;
759 	struct sk_buff *skb;
760 	struct netlink_callback *cb;
761 	struct Qdisc *q;
762 	u32 parent;
763 };
764 
765 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
766 {
767 	struct tcf_dump_args *a = (void *)arg;
768 	struct net *net = sock_net(a->skb->sk);
769 
770 	return tcf_fill_node(net, a->skb, tp, a->q, a->parent,
771 			     n, NETLINK_CB(a->cb->skb).portid,
772 			     a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
773 			     RTM_NEWTFILTER);
774 }
775 
776 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
777 			   struct sk_buff *skb, struct netlink_callback *cb,
778 			   long index_start, long *p_index)
779 {
780 	struct net *net = sock_net(skb->sk);
781 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
782 	struct tcf_dump_args arg;
783 	struct tcf_proto *tp;
784 
785 	for (tp = rtnl_dereference(chain->filter_chain);
786 	     tp; tp = rtnl_dereference(tp->next), (*p_index)++) {
787 		if (*p_index < index_start)
788 			continue;
789 		if (TC_H_MAJ(tcm->tcm_info) &&
790 		    TC_H_MAJ(tcm->tcm_info) != tp->prio)
791 			continue;
792 		if (TC_H_MIN(tcm->tcm_info) &&
793 		    TC_H_MIN(tcm->tcm_info) != tp->protocol)
794 			continue;
795 		if (*p_index > index_start)
796 			memset(&cb->args[1], 0,
797 			       sizeof(cb->args) - sizeof(cb->args[0]));
798 		if (cb->args[1] == 0) {
799 			if (tcf_fill_node(net, skb, tp, q, parent, 0,
800 					  NETLINK_CB(cb->skb).portid,
801 					  cb->nlh->nlmsg_seq, NLM_F_MULTI,
802 					  RTM_NEWTFILTER) <= 0)
803 				return false;
804 
805 			cb->args[1] = 1;
806 		}
807 		if (!tp->ops->walk)
808 			continue;
809 		arg.w.fn = tcf_node_dump;
810 		arg.skb = skb;
811 		arg.cb = cb;
812 		arg.q = q;
813 		arg.parent = parent;
814 		arg.w.stop = 0;
815 		arg.w.skip = cb->args[1] - 1;
816 		arg.w.count = 0;
817 		tp->ops->walk(tp, &arg.w);
818 		cb->args[1] = arg.w.count + 1;
819 		if (arg.w.stop)
820 			return false;
821 	}
822 	return true;
823 }
824 
825 /* called with RTNL */
826 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
827 {
828 	struct net *net = sock_net(skb->sk);
829 	struct nlattr *tca[TCA_MAX + 1];
830 	struct net_device *dev;
831 	struct Qdisc *q;
832 	struct tcf_block *block;
833 	struct tcf_chain *chain;
834 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
835 	unsigned long cl = 0;
836 	const struct Qdisc_class_ops *cops;
837 	long index_start;
838 	long index;
839 	u32 parent;
840 	int err;
841 
842 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
843 		return skb->len;
844 
845 	err = nlmsg_parse(cb->nlh, sizeof(*tcm), tca, TCA_MAX, NULL, NULL);
846 	if (err)
847 		return err;
848 
849 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
850 	if (!dev)
851 		return skb->len;
852 
853 	parent = tcm->tcm_parent;
854 	if (!parent) {
855 		q = dev->qdisc;
856 		parent = q->handle;
857 	} else {
858 		q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
859 	}
860 	if (!q)
861 		goto out;
862 	cops = q->ops->cl_ops;
863 	if (!cops)
864 		goto out;
865 	if (!cops->tcf_block)
866 		goto out;
867 	if (TC_H_MIN(tcm->tcm_parent)) {
868 		cl = cops->find(q, tcm->tcm_parent);
869 		if (cl == 0)
870 			goto out;
871 	}
872 	block = cops->tcf_block(q, cl);
873 	if (!block)
874 		goto out;
875 
876 	index_start = cb->args[0];
877 	index = 0;
878 
879 	list_for_each_entry(chain, &block->chain_list, list) {
880 		if (tca[TCA_CHAIN] &&
881 		    nla_get_u32(tca[TCA_CHAIN]) != chain->index)
882 			continue;
883 		if (!tcf_chain_dump(chain, q, parent, skb, cb,
884 				    index_start, &index))
885 			break;
886 	}
887 
888 	cb->args[0] = index;
889 
890 out:
891 	return skb->len;
892 }
893 
894 void tcf_exts_destroy(struct tcf_exts *exts)
895 {
896 #ifdef CONFIG_NET_CLS_ACT
897 	LIST_HEAD(actions);
898 
899 	tcf_exts_to_list(exts, &actions);
900 	tcf_action_destroy(&actions, TCA_ACT_UNBIND);
901 	kfree(exts->actions);
902 	exts->nr_actions = 0;
903 #endif
904 }
905 EXPORT_SYMBOL(tcf_exts_destroy);
906 
907 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
908 		      struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr)
909 {
910 #ifdef CONFIG_NET_CLS_ACT
911 	{
912 		struct tc_action *act;
913 
914 		if (exts->police && tb[exts->police]) {
915 			act = tcf_action_init_1(net, tp, tb[exts->police],
916 						rate_tlv, "police", ovr,
917 						TCA_ACT_BIND);
918 			if (IS_ERR(act))
919 				return PTR_ERR(act);
920 
921 			act->type = exts->type = TCA_OLD_COMPAT;
922 			exts->actions[0] = act;
923 			exts->nr_actions = 1;
924 		} else if (exts->action && tb[exts->action]) {
925 			LIST_HEAD(actions);
926 			int err, i = 0;
927 
928 			err = tcf_action_init(net, tp, tb[exts->action],
929 					      rate_tlv, NULL, ovr, TCA_ACT_BIND,
930 					      &actions);
931 			if (err)
932 				return err;
933 			list_for_each_entry(act, &actions, list)
934 				exts->actions[i++] = act;
935 			exts->nr_actions = i;
936 		}
937 	}
938 #else
939 	if ((exts->action && tb[exts->action]) ||
940 	    (exts->police && tb[exts->police]))
941 		return -EOPNOTSUPP;
942 #endif
943 
944 	return 0;
945 }
946 EXPORT_SYMBOL(tcf_exts_validate);
947 
948 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
949 {
950 #ifdef CONFIG_NET_CLS_ACT
951 	struct tcf_exts old = *dst;
952 
953 	*dst = *src;
954 	tcf_exts_destroy(&old);
955 #endif
956 }
957 EXPORT_SYMBOL(tcf_exts_change);
958 
959 #ifdef CONFIG_NET_CLS_ACT
960 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
961 {
962 	if (exts->nr_actions == 0)
963 		return NULL;
964 	else
965 		return exts->actions[0];
966 }
967 #endif
968 
969 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
970 {
971 #ifdef CONFIG_NET_CLS_ACT
972 	struct nlattr *nest;
973 
974 	if (exts->action && tcf_exts_has_actions(exts)) {
975 		/*
976 		 * again for backward compatible mode - we want
977 		 * to work with both old and new modes of entering
978 		 * tc data even if iproute2  was newer - jhs
979 		 */
980 		if (exts->type != TCA_OLD_COMPAT) {
981 			LIST_HEAD(actions);
982 
983 			nest = nla_nest_start(skb, exts->action);
984 			if (nest == NULL)
985 				goto nla_put_failure;
986 
987 			tcf_exts_to_list(exts, &actions);
988 			if (tcf_action_dump(skb, &actions, 0, 0) < 0)
989 				goto nla_put_failure;
990 			nla_nest_end(skb, nest);
991 		} else if (exts->police) {
992 			struct tc_action *act = tcf_exts_first_act(exts);
993 			nest = nla_nest_start(skb, exts->police);
994 			if (nest == NULL || !act)
995 				goto nla_put_failure;
996 			if (tcf_action_dump_old(skb, act, 0, 0) < 0)
997 				goto nla_put_failure;
998 			nla_nest_end(skb, nest);
999 		}
1000 	}
1001 	return 0;
1002 
1003 nla_put_failure:
1004 	nla_nest_cancel(skb, nest);
1005 	return -1;
1006 #else
1007 	return 0;
1008 #endif
1009 }
1010 EXPORT_SYMBOL(tcf_exts_dump);
1011 
1012 
1013 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
1014 {
1015 #ifdef CONFIG_NET_CLS_ACT
1016 	struct tc_action *a = tcf_exts_first_act(exts);
1017 	if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
1018 		return -1;
1019 #endif
1020 	return 0;
1021 }
1022 EXPORT_SYMBOL(tcf_exts_dump_stats);
1023 
1024 static int tc_exts_setup_cb_egdev_call(struct tcf_exts *exts,
1025 				       enum tc_setup_type type,
1026 				       void *type_data, bool err_stop)
1027 {
1028 	int ok_count = 0;
1029 #ifdef CONFIG_NET_CLS_ACT
1030 	const struct tc_action *a;
1031 	struct net_device *dev;
1032 	LIST_HEAD(actions);
1033 	int ret;
1034 
1035 	if (!tcf_exts_has_actions(exts))
1036 		return 0;
1037 
1038 	tcf_exts_to_list(exts, &actions);
1039 	list_for_each_entry(a, &actions, list) {
1040 		if (!a->ops->get_dev)
1041 			continue;
1042 		dev = a->ops->get_dev(a);
1043 		if (!dev || !tc_can_offload(dev))
1044 			continue;
1045 		ret = tc_setup_cb_egdev_call(dev, type, type_data, err_stop);
1046 		if (ret < 0)
1047 			return ret;
1048 		ok_count += ret;
1049 	}
1050 #endif
1051 	return ok_count;
1052 }
1053 
1054 int tc_setup_cb_call(struct tcf_exts *exts, enum tc_setup_type type,
1055 		     void *type_data, bool err_stop)
1056 {
1057 	return tc_exts_setup_cb_egdev_call(exts, type, type_data, err_stop);
1058 }
1059 EXPORT_SYMBOL(tc_setup_cb_call);
1060 
1061 static int __init tc_filter_init(void)
1062 {
1063 	rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, 0);
1064 	rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, 0);
1065 	rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
1066 		      tc_dump_tfilter, 0);
1067 
1068 	return 0;
1069 }
1070 
1071 subsys_initcall(tc_filter_init);
1072