xref: /linux/net/sched/cls_api.c (revision ead751507de86d90fa250431e9990a8b881f713c)
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 static struct workqueue_struct *tc_filter_wq;
81 
82 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
83 {
84 	struct tcf_proto_ops *t;
85 	int rc = -ENOENT;
86 
87 	/* Wait for outstanding call_rcu()s, if any, from a
88 	 * tcf_proto_ops's destroy() handler.
89 	 */
90 	rcu_barrier();
91 	flush_workqueue(tc_filter_wq);
92 
93 	write_lock(&cls_mod_lock);
94 	list_for_each_entry(t, &tcf_proto_base, head) {
95 		if (t == ops) {
96 			list_del(&t->head);
97 			rc = 0;
98 			break;
99 		}
100 	}
101 	write_unlock(&cls_mod_lock);
102 	return rc;
103 }
104 EXPORT_SYMBOL(unregister_tcf_proto_ops);
105 
106 bool tcf_queue_work(struct work_struct *work)
107 {
108 	return queue_work(tc_filter_wq, work);
109 }
110 EXPORT_SYMBOL(tcf_queue_work);
111 
112 /* Select new prio value from the range, managed by kernel. */
113 
114 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
115 {
116 	u32 first = TC_H_MAKE(0xC0000000U, 0U);
117 
118 	if (tp)
119 		first = tp->prio - 1;
120 
121 	return TC_H_MAJ(first);
122 }
123 
124 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
125 					  u32 prio, u32 parent, struct Qdisc *q,
126 					  struct tcf_chain *chain)
127 {
128 	struct tcf_proto *tp;
129 	int err;
130 
131 	tp = kzalloc(sizeof(*tp), GFP_KERNEL);
132 	if (!tp)
133 		return ERR_PTR(-ENOBUFS);
134 
135 	err = -ENOENT;
136 	tp->ops = tcf_proto_lookup_ops(kind);
137 	if (!tp->ops) {
138 #ifdef CONFIG_MODULES
139 		rtnl_unlock();
140 		request_module("cls_%s", kind);
141 		rtnl_lock();
142 		tp->ops = tcf_proto_lookup_ops(kind);
143 		/* We dropped the RTNL semaphore in order to perform
144 		 * the module load. So, even if we succeeded in loading
145 		 * the module we have to replay the request. We indicate
146 		 * this using -EAGAIN.
147 		 */
148 		if (tp->ops) {
149 			module_put(tp->ops->owner);
150 			err = -EAGAIN;
151 		} else {
152 			err = -ENOENT;
153 		}
154 		goto errout;
155 #endif
156 	}
157 	tp->classify = tp->ops->classify;
158 	tp->protocol = protocol;
159 	tp->prio = prio;
160 	tp->classid = parent;
161 	tp->q = q;
162 	tp->chain = chain;
163 
164 	err = tp->ops->init(tp);
165 	if (err) {
166 		module_put(tp->ops->owner);
167 		goto errout;
168 	}
169 	return tp;
170 
171 errout:
172 	kfree(tp);
173 	return ERR_PTR(err);
174 }
175 
176 static void tcf_proto_destroy(struct tcf_proto *tp)
177 {
178 	tp->ops->destroy(tp);
179 	module_put(tp->ops->owner);
180 	kfree_rcu(tp, rcu);
181 }
182 
183 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
184 					  u32 chain_index)
185 {
186 	struct tcf_chain *chain;
187 
188 	chain = kzalloc(sizeof(*chain), GFP_KERNEL);
189 	if (!chain)
190 		return NULL;
191 	list_add_tail(&chain->list, &block->chain_list);
192 	chain->block = block;
193 	chain->index = chain_index;
194 	chain->refcnt = 1;
195 	return chain;
196 }
197 
198 static void tcf_chain_flush(struct tcf_chain *chain)
199 {
200 	struct tcf_proto *tp;
201 
202 	if (chain->p_filter_chain)
203 		RCU_INIT_POINTER(*chain->p_filter_chain, NULL);
204 	while ((tp = rtnl_dereference(chain->filter_chain)) != NULL) {
205 		RCU_INIT_POINTER(chain->filter_chain, tp->next);
206 		tcf_chain_put(chain);
207 		tcf_proto_destroy(tp);
208 	}
209 }
210 
211 static void tcf_chain_destroy(struct tcf_chain *chain)
212 {
213 	list_del(&chain->list);
214 	kfree(chain);
215 }
216 
217 static void tcf_chain_hold(struct tcf_chain *chain)
218 {
219 	++chain->refcnt;
220 }
221 
222 struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
223 				bool create)
224 {
225 	struct tcf_chain *chain;
226 
227 	list_for_each_entry(chain, &block->chain_list, list) {
228 		if (chain->index == chain_index) {
229 			tcf_chain_hold(chain);
230 			return chain;
231 		}
232 	}
233 
234 	return create ? tcf_chain_create(block, chain_index) : NULL;
235 }
236 EXPORT_SYMBOL(tcf_chain_get);
237 
238 void tcf_chain_put(struct tcf_chain *chain)
239 {
240 	if (--chain->refcnt == 0)
241 		tcf_chain_destroy(chain);
242 }
243 EXPORT_SYMBOL(tcf_chain_put);
244 
245 static void
246 tcf_chain_filter_chain_ptr_set(struct tcf_chain *chain,
247 			       struct tcf_proto __rcu **p_filter_chain)
248 {
249 	chain->p_filter_chain = p_filter_chain;
250 }
251 
252 int tcf_block_get(struct tcf_block **p_block,
253 		  struct tcf_proto __rcu **p_filter_chain)
254 {
255 	struct tcf_block *block = kzalloc(sizeof(*block), GFP_KERNEL);
256 	struct tcf_chain *chain;
257 	int err;
258 
259 	if (!block)
260 		return -ENOMEM;
261 	INIT_LIST_HEAD(&block->chain_list);
262 	/* Create chain 0 by default, it has to be always present. */
263 	chain = tcf_chain_create(block, 0);
264 	if (!chain) {
265 		err = -ENOMEM;
266 		goto err_chain_create;
267 	}
268 	tcf_chain_filter_chain_ptr_set(chain, p_filter_chain);
269 	*p_block = block;
270 	return 0;
271 
272 err_chain_create:
273 	kfree(block);
274 	return err;
275 }
276 EXPORT_SYMBOL(tcf_block_get);
277 
278 static void tcf_block_put_final(struct work_struct *work)
279 {
280 	struct tcf_block *block = container_of(work, struct tcf_block, work);
281 	struct tcf_chain *chain, *tmp;
282 
283 	rtnl_lock();
284 	/* Only chain 0 should be still here. */
285 	list_for_each_entry_safe(chain, tmp, &block->chain_list, list)
286 		tcf_chain_put(chain);
287 	rtnl_unlock();
288 	kfree(block);
289 }
290 
291 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
292  * actions should be all removed after flushing. However, filters are now
293  * destroyed in tc filter workqueue with RTNL lock, they can not race here.
294  */
295 void tcf_block_put(struct tcf_block *block)
296 {
297 	struct tcf_chain *chain, *tmp;
298 
299 	if (!block)
300 		return;
301 
302 	list_for_each_entry_safe(chain, tmp, &block->chain_list, list)
303 		tcf_chain_flush(chain);
304 
305 	INIT_WORK(&block->work, tcf_block_put_final);
306 	/* Wait for RCU callbacks to release the reference count and make
307 	 * sure their works have been queued before this.
308 	 */
309 	rcu_barrier();
310 	tcf_queue_work(&block->work);
311 }
312 EXPORT_SYMBOL(tcf_block_put);
313 
314 /* Main classifier routine: scans classifier chain attached
315  * to this qdisc, (optionally) tests for protocol and asks
316  * specific classifiers.
317  */
318 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
319 		 struct tcf_result *res, bool compat_mode)
320 {
321 	__be16 protocol = tc_skb_protocol(skb);
322 #ifdef CONFIG_NET_CLS_ACT
323 	const int max_reclassify_loop = 4;
324 	const struct tcf_proto *orig_tp = tp;
325 	const struct tcf_proto *first_tp;
326 	int limit = 0;
327 
328 reclassify:
329 #endif
330 	for (; tp; tp = rcu_dereference_bh(tp->next)) {
331 		int err;
332 
333 		if (tp->protocol != protocol &&
334 		    tp->protocol != htons(ETH_P_ALL))
335 			continue;
336 
337 		err = tp->classify(skb, tp, res);
338 #ifdef CONFIG_NET_CLS_ACT
339 		if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
340 			first_tp = orig_tp;
341 			goto reset;
342 		} else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
343 			first_tp = res->goto_tp;
344 			goto reset;
345 		}
346 #endif
347 		if (err >= 0)
348 			return err;
349 	}
350 
351 	return TC_ACT_UNSPEC; /* signal: continue lookup */
352 #ifdef CONFIG_NET_CLS_ACT
353 reset:
354 	if (unlikely(limit++ >= max_reclassify_loop)) {
355 		net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n",
356 				       tp->q->ops->id, tp->prio & 0xffff,
357 				       ntohs(tp->protocol));
358 		return TC_ACT_SHOT;
359 	}
360 
361 	tp = first_tp;
362 	protocol = tc_skb_protocol(skb);
363 	goto reclassify;
364 #endif
365 }
366 EXPORT_SYMBOL(tcf_classify);
367 
368 struct tcf_chain_info {
369 	struct tcf_proto __rcu **pprev;
370 	struct tcf_proto __rcu *next;
371 };
372 
373 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain_info *chain_info)
374 {
375 	return rtnl_dereference(*chain_info->pprev);
376 }
377 
378 static void tcf_chain_tp_insert(struct tcf_chain *chain,
379 				struct tcf_chain_info *chain_info,
380 				struct tcf_proto *tp)
381 {
382 	if (chain->p_filter_chain &&
383 	    *chain_info->pprev == chain->filter_chain)
384 		rcu_assign_pointer(*chain->p_filter_chain, tp);
385 	RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain_info));
386 	rcu_assign_pointer(*chain_info->pprev, tp);
387 	tcf_chain_hold(chain);
388 }
389 
390 static void tcf_chain_tp_remove(struct tcf_chain *chain,
391 				struct tcf_chain_info *chain_info,
392 				struct tcf_proto *tp)
393 {
394 	struct tcf_proto *next = rtnl_dereference(chain_info->next);
395 
396 	if (chain->p_filter_chain && tp == chain->filter_chain)
397 		RCU_INIT_POINTER(*chain->p_filter_chain, next);
398 	RCU_INIT_POINTER(*chain_info->pprev, next);
399 	tcf_chain_put(chain);
400 }
401 
402 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
403 					   struct tcf_chain_info *chain_info,
404 					   u32 protocol, u32 prio,
405 					   bool prio_allocate)
406 {
407 	struct tcf_proto **pprev;
408 	struct tcf_proto *tp;
409 
410 	/* Check the chain for existence of proto-tcf with this priority */
411 	for (pprev = &chain->filter_chain;
412 	     (tp = rtnl_dereference(*pprev)); pprev = &tp->next) {
413 		if (tp->prio >= prio) {
414 			if (tp->prio == prio) {
415 				if (prio_allocate ||
416 				    (tp->protocol != protocol && protocol))
417 					return ERR_PTR(-EINVAL);
418 			} else {
419 				tp = NULL;
420 			}
421 			break;
422 		}
423 	}
424 	chain_info->pprev = pprev;
425 	chain_info->next = tp ? tp->next : NULL;
426 	return tp;
427 }
428 
429 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
430 			 struct tcf_proto *tp, void *fh, u32 portid,
431 			 u32 seq, u16 flags, int event)
432 {
433 	struct tcmsg *tcm;
434 	struct nlmsghdr  *nlh;
435 	unsigned char *b = skb_tail_pointer(skb);
436 
437 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
438 	if (!nlh)
439 		goto out_nlmsg_trim;
440 	tcm = nlmsg_data(nlh);
441 	tcm->tcm_family = AF_UNSPEC;
442 	tcm->tcm__pad1 = 0;
443 	tcm->tcm__pad2 = 0;
444 	tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
445 	tcm->tcm_parent = tp->classid;
446 	tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
447 	if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
448 		goto nla_put_failure;
449 	if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
450 		goto nla_put_failure;
451 	if (!fh) {
452 		tcm->tcm_handle = 0;
453 	} else {
454 		if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
455 			goto nla_put_failure;
456 	}
457 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
458 	return skb->len;
459 
460 out_nlmsg_trim:
461 nla_put_failure:
462 	nlmsg_trim(skb, b);
463 	return -1;
464 }
465 
466 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
467 			  struct nlmsghdr *n, struct tcf_proto *tp,
468 			  void *fh, int event, bool unicast)
469 {
470 	struct sk_buff *skb;
471 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
472 
473 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
474 	if (!skb)
475 		return -ENOBUFS;
476 
477 	if (tcf_fill_node(net, skb, tp, fh, portid, n->nlmsg_seq,
478 			  n->nlmsg_flags, event) <= 0) {
479 		kfree_skb(skb);
480 		return -EINVAL;
481 	}
482 
483 	if (unicast)
484 		return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
485 
486 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
487 			      n->nlmsg_flags & NLM_F_ECHO);
488 }
489 
490 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
491 			      struct nlmsghdr *n, struct tcf_proto *tp,
492 			      void *fh, bool unicast, bool *last)
493 {
494 	struct sk_buff *skb;
495 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
496 	int err;
497 
498 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
499 	if (!skb)
500 		return -ENOBUFS;
501 
502 	if (tcf_fill_node(net, skb, tp, fh, portid, n->nlmsg_seq,
503 			  n->nlmsg_flags, RTM_DELTFILTER) <= 0) {
504 		kfree_skb(skb);
505 		return -EINVAL;
506 	}
507 
508 	err = tp->ops->delete(tp, fh, last);
509 	if (err) {
510 		kfree_skb(skb);
511 		return err;
512 	}
513 
514 	if (unicast)
515 		return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
516 
517 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
518 			      n->nlmsg_flags & NLM_F_ECHO);
519 }
520 
521 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
522 				 struct nlmsghdr *n,
523 				 struct tcf_chain *chain, int event)
524 {
525 	struct tcf_proto *tp;
526 
527 	for (tp = rtnl_dereference(chain->filter_chain);
528 	     tp; tp = rtnl_dereference(tp->next))
529 		tfilter_notify(net, oskb, n, tp, 0, event, false);
530 }
531 
532 /* Add/change/delete/get a filter node */
533 
534 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
535 			  struct netlink_ext_ack *extack)
536 {
537 	struct net *net = sock_net(skb->sk);
538 	struct nlattr *tca[TCA_MAX + 1];
539 	struct tcmsg *t;
540 	u32 protocol;
541 	u32 prio;
542 	bool prio_allocate;
543 	u32 parent;
544 	u32 chain_index;
545 	struct net_device *dev;
546 	struct Qdisc  *q;
547 	struct tcf_chain_info chain_info;
548 	struct tcf_chain *chain = NULL;
549 	struct tcf_block *block;
550 	struct tcf_proto *tp;
551 	const struct Qdisc_class_ops *cops;
552 	unsigned long cl;
553 	void *fh;
554 	int err;
555 	int tp_created;
556 
557 	if ((n->nlmsg_type != RTM_GETTFILTER) &&
558 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
559 		return -EPERM;
560 
561 replay:
562 	tp_created = 0;
563 
564 	err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack);
565 	if (err < 0)
566 		return err;
567 
568 	t = nlmsg_data(n);
569 	protocol = TC_H_MIN(t->tcm_info);
570 	prio = TC_H_MAJ(t->tcm_info);
571 	prio_allocate = false;
572 	parent = t->tcm_parent;
573 	cl = 0;
574 
575 	if (prio == 0) {
576 		switch (n->nlmsg_type) {
577 		case RTM_DELTFILTER:
578 			if (protocol || t->tcm_handle || tca[TCA_KIND])
579 				return -ENOENT;
580 			break;
581 		case RTM_NEWTFILTER:
582 			/* If no priority is provided by the user,
583 			 * we allocate one.
584 			 */
585 			if (n->nlmsg_flags & NLM_F_CREATE) {
586 				prio = TC_H_MAKE(0x80000000U, 0U);
587 				prio_allocate = true;
588 				break;
589 			}
590 			/* fall-through */
591 		default:
592 			return -ENOENT;
593 		}
594 	}
595 
596 	/* Find head of filter chain. */
597 
598 	/* Find link */
599 	dev = __dev_get_by_index(net, t->tcm_ifindex);
600 	if (dev == NULL)
601 		return -ENODEV;
602 
603 	/* Find qdisc */
604 	if (!parent) {
605 		q = dev->qdisc;
606 		parent = q->handle;
607 	} else {
608 		q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
609 		if (q == NULL)
610 			return -EINVAL;
611 	}
612 
613 	/* Is it classful? */
614 	cops = q->ops->cl_ops;
615 	if (!cops)
616 		return -EINVAL;
617 
618 	if (!cops->tcf_block)
619 		return -EOPNOTSUPP;
620 
621 	/* Do we search for filter, attached to class? */
622 	if (TC_H_MIN(parent)) {
623 		cl = cops->find(q, parent);
624 		if (cl == 0)
625 			return -ENOENT;
626 	}
627 
628 	/* And the last stroke */
629 	block = cops->tcf_block(q, cl);
630 	if (!block) {
631 		err = -EINVAL;
632 		goto errout;
633 	}
634 
635 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
636 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
637 		err = -EINVAL;
638 		goto errout;
639 	}
640 	chain = tcf_chain_get(block, chain_index,
641 			      n->nlmsg_type == RTM_NEWTFILTER);
642 	if (!chain) {
643 		err = n->nlmsg_type == RTM_NEWTFILTER ? -ENOMEM : -EINVAL;
644 		goto errout;
645 	}
646 
647 	if (n->nlmsg_type == RTM_DELTFILTER && prio == 0) {
648 		tfilter_notify_chain(net, skb, n, chain, RTM_DELTFILTER);
649 		tcf_chain_flush(chain);
650 		err = 0;
651 		goto errout;
652 	}
653 
654 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
655 			       prio, prio_allocate);
656 	if (IS_ERR(tp)) {
657 		err = PTR_ERR(tp);
658 		goto errout;
659 	}
660 
661 	if (tp == NULL) {
662 		/* Proto-tcf does not exist, create new one */
663 
664 		if (tca[TCA_KIND] == NULL || !protocol) {
665 			err = -EINVAL;
666 			goto errout;
667 		}
668 
669 		if (n->nlmsg_type != RTM_NEWTFILTER ||
670 		    !(n->nlmsg_flags & NLM_F_CREATE)) {
671 			err = -ENOENT;
672 			goto errout;
673 		}
674 
675 		if (prio_allocate)
676 			prio = tcf_auto_prio(tcf_chain_tp_prev(&chain_info));
677 
678 		tp = tcf_proto_create(nla_data(tca[TCA_KIND]),
679 				      protocol, prio, parent, q, chain);
680 		if (IS_ERR(tp)) {
681 			err = PTR_ERR(tp);
682 			goto errout;
683 		}
684 		tp_created = 1;
685 	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
686 		err = -EINVAL;
687 		goto errout;
688 	}
689 
690 	fh = tp->ops->get(tp, t->tcm_handle);
691 
692 	if (!fh) {
693 		if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
694 			tcf_chain_tp_remove(chain, &chain_info, tp);
695 			tfilter_notify(net, skb, n, tp, fh,
696 				       RTM_DELTFILTER, false);
697 			tcf_proto_destroy(tp);
698 			err = 0;
699 			goto errout;
700 		}
701 
702 		if (n->nlmsg_type != RTM_NEWTFILTER ||
703 		    !(n->nlmsg_flags & NLM_F_CREATE)) {
704 			err = -ENOENT;
705 			goto errout;
706 		}
707 	} else {
708 		bool last;
709 
710 		switch (n->nlmsg_type) {
711 		case RTM_NEWTFILTER:
712 			if (n->nlmsg_flags & NLM_F_EXCL) {
713 				if (tp_created)
714 					tcf_proto_destroy(tp);
715 				err = -EEXIST;
716 				goto errout;
717 			}
718 			break;
719 		case RTM_DELTFILTER:
720 			err = tfilter_del_notify(net, skb, n, tp, fh, false,
721 						 &last);
722 			if (err)
723 				goto errout;
724 			if (last) {
725 				tcf_chain_tp_remove(chain, &chain_info, tp);
726 				tcf_proto_destroy(tp);
727 			}
728 			goto errout;
729 		case RTM_GETTFILTER:
730 			err = tfilter_notify(net, skb, n, tp, fh,
731 					     RTM_NEWTFILTER, true);
732 			goto errout;
733 		default:
734 			err = -EINVAL;
735 			goto errout;
736 		}
737 	}
738 
739 	err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
740 			      n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE);
741 	if (err == 0) {
742 		if (tp_created)
743 			tcf_chain_tp_insert(chain, &chain_info, tp);
744 		tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER, false);
745 	} else {
746 		if (tp_created)
747 			tcf_proto_destroy(tp);
748 	}
749 
750 errout:
751 	if (chain)
752 		tcf_chain_put(chain);
753 	if (err == -EAGAIN)
754 		/* Replay the request. */
755 		goto replay;
756 	return err;
757 }
758 
759 struct tcf_dump_args {
760 	struct tcf_walker w;
761 	struct sk_buff *skb;
762 	struct netlink_callback *cb;
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, n, NETLINK_CB(a->cb->skb).portid,
771 			     a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
772 			     RTM_NEWTFILTER);
773 }
774 
775 static bool tcf_chain_dump(struct tcf_chain *chain, struct sk_buff *skb,
776 			   struct netlink_callback *cb,
777 			   long index_start, long *p_index)
778 {
779 	struct net *net = sock_net(skb->sk);
780 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
781 	struct tcf_dump_args arg;
782 	struct tcf_proto *tp;
783 
784 	for (tp = rtnl_dereference(chain->filter_chain);
785 	     tp; tp = rtnl_dereference(tp->next), (*p_index)++) {
786 		if (*p_index < index_start)
787 			continue;
788 		if (TC_H_MAJ(tcm->tcm_info) &&
789 		    TC_H_MAJ(tcm->tcm_info) != tp->prio)
790 			continue;
791 		if (TC_H_MIN(tcm->tcm_info) &&
792 		    TC_H_MIN(tcm->tcm_info) != tp->protocol)
793 			continue;
794 		if (*p_index > index_start)
795 			memset(&cb->args[1], 0,
796 			       sizeof(cb->args) - sizeof(cb->args[0]));
797 		if (cb->args[1] == 0) {
798 			if (tcf_fill_node(net, skb, tp, 0,
799 					  NETLINK_CB(cb->skb).portid,
800 					  cb->nlh->nlmsg_seq, NLM_F_MULTI,
801 					  RTM_NEWTFILTER) <= 0)
802 				return false;
803 
804 			cb->args[1] = 1;
805 		}
806 		if (!tp->ops->walk)
807 			continue;
808 		arg.w.fn = tcf_node_dump;
809 		arg.skb = skb;
810 		arg.cb = cb;
811 		arg.w.stop = 0;
812 		arg.w.skip = cb->args[1] - 1;
813 		arg.w.count = 0;
814 		tp->ops->walk(tp, &arg.w);
815 		cb->args[1] = arg.w.count + 1;
816 		if (arg.w.stop)
817 			return false;
818 	}
819 	return true;
820 }
821 
822 /* called with RTNL */
823 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
824 {
825 	struct net *net = sock_net(skb->sk);
826 	struct nlattr *tca[TCA_MAX + 1];
827 	struct net_device *dev;
828 	struct Qdisc *q;
829 	struct tcf_block *block;
830 	struct tcf_chain *chain;
831 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
832 	unsigned long cl = 0;
833 	const struct Qdisc_class_ops *cops;
834 	long index_start;
835 	long index;
836 	int err;
837 
838 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
839 		return skb->len;
840 
841 	err = nlmsg_parse(cb->nlh, sizeof(*tcm), tca, TCA_MAX, NULL, NULL);
842 	if (err)
843 		return err;
844 
845 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
846 	if (!dev)
847 		return skb->len;
848 
849 	if (!tcm->tcm_parent)
850 		q = dev->qdisc;
851 	else
852 		q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
853 	if (!q)
854 		goto out;
855 	cops = q->ops->cl_ops;
856 	if (!cops)
857 		goto out;
858 	if (!cops->tcf_block)
859 		goto out;
860 	if (TC_H_MIN(tcm->tcm_parent)) {
861 		cl = cops->find(q, tcm->tcm_parent);
862 		if (cl == 0)
863 			goto out;
864 	}
865 	block = cops->tcf_block(q, cl);
866 	if (!block)
867 		goto out;
868 
869 	index_start = cb->args[0];
870 	index = 0;
871 
872 	list_for_each_entry(chain, &block->chain_list, list) {
873 		if (tca[TCA_CHAIN] &&
874 		    nla_get_u32(tca[TCA_CHAIN]) != chain->index)
875 			continue;
876 		if (!tcf_chain_dump(chain, skb, cb, index_start, &index))
877 			break;
878 	}
879 
880 	cb->args[0] = index;
881 
882 out:
883 	return skb->len;
884 }
885 
886 void tcf_exts_destroy(struct tcf_exts *exts)
887 {
888 #ifdef CONFIG_NET_CLS_ACT
889 	LIST_HEAD(actions);
890 
891 	ASSERT_RTNL();
892 	tcf_exts_to_list(exts, &actions);
893 	tcf_action_destroy(&actions, TCA_ACT_UNBIND);
894 	kfree(exts->actions);
895 	exts->nr_actions = 0;
896 #endif
897 }
898 EXPORT_SYMBOL(tcf_exts_destroy);
899 
900 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
901 		      struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr)
902 {
903 #ifdef CONFIG_NET_CLS_ACT
904 	{
905 		struct tc_action *act;
906 
907 		if (exts->police && tb[exts->police]) {
908 			act = tcf_action_init_1(net, tp, tb[exts->police],
909 						rate_tlv, "police", ovr,
910 						TCA_ACT_BIND);
911 			if (IS_ERR(act))
912 				return PTR_ERR(act);
913 
914 			act->type = exts->type = TCA_OLD_COMPAT;
915 			exts->actions[0] = act;
916 			exts->nr_actions = 1;
917 		} else if (exts->action && tb[exts->action]) {
918 			LIST_HEAD(actions);
919 			int err, i = 0;
920 
921 			err = tcf_action_init(net, tp, tb[exts->action],
922 					      rate_tlv, NULL, ovr, TCA_ACT_BIND,
923 					      &actions);
924 			if (err)
925 				return err;
926 			list_for_each_entry(act, &actions, list)
927 				exts->actions[i++] = act;
928 			exts->nr_actions = i;
929 		}
930 	}
931 #else
932 	if ((exts->action && tb[exts->action]) ||
933 	    (exts->police && tb[exts->police]))
934 		return -EOPNOTSUPP;
935 #endif
936 
937 	return 0;
938 }
939 EXPORT_SYMBOL(tcf_exts_validate);
940 
941 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
942 {
943 #ifdef CONFIG_NET_CLS_ACT
944 	struct tcf_exts old = *dst;
945 
946 	*dst = *src;
947 	tcf_exts_destroy(&old);
948 #endif
949 }
950 EXPORT_SYMBOL(tcf_exts_change);
951 
952 #ifdef CONFIG_NET_CLS_ACT
953 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
954 {
955 	if (exts->nr_actions == 0)
956 		return NULL;
957 	else
958 		return exts->actions[0];
959 }
960 #endif
961 
962 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
963 {
964 #ifdef CONFIG_NET_CLS_ACT
965 	struct nlattr *nest;
966 
967 	if (exts->action && tcf_exts_has_actions(exts)) {
968 		/*
969 		 * again for backward compatible mode - we want
970 		 * to work with both old and new modes of entering
971 		 * tc data even if iproute2  was newer - jhs
972 		 */
973 		if (exts->type != TCA_OLD_COMPAT) {
974 			LIST_HEAD(actions);
975 
976 			nest = nla_nest_start(skb, exts->action);
977 			if (nest == NULL)
978 				goto nla_put_failure;
979 
980 			tcf_exts_to_list(exts, &actions);
981 			if (tcf_action_dump(skb, &actions, 0, 0) < 0)
982 				goto nla_put_failure;
983 			nla_nest_end(skb, nest);
984 		} else if (exts->police) {
985 			struct tc_action *act = tcf_exts_first_act(exts);
986 			nest = nla_nest_start(skb, exts->police);
987 			if (nest == NULL || !act)
988 				goto nla_put_failure;
989 			if (tcf_action_dump_old(skb, act, 0, 0) < 0)
990 				goto nla_put_failure;
991 			nla_nest_end(skb, nest);
992 		}
993 	}
994 	return 0;
995 
996 nla_put_failure:
997 	nla_nest_cancel(skb, nest);
998 	return -1;
999 #else
1000 	return 0;
1001 #endif
1002 }
1003 EXPORT_SYMBOL(tcf_exts_dump);
1004 
1005 
1006 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
1007 {
1008 #ifdef CONFIG_NET_CLS_ACT
1009 	struct tc_action *a = tcf_exts_first_act(exts);
1010 	if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
1011 		return -1;
1012 #endif
1013 	return 0;
1014 }
1015 EXPORT_SYMBOL(tcf_exts_dump_stats);
1016 
1017 int tcf_exts_get_dev(struct net_device *dev, struct tcf_exts *exts,
1018 		     struct net_device **hw_dev)
1019 {
1020 #ifdef CONFIG_NET_CLS_ACT
1021 	const struct tc_action *a;
1022 	LIST_HEAD(actions);
1023 
1024 	if (!tcf_exts_has_actions(exts))
1025 		return -EINVAL;
1026 
1027 	tcf_exts_to_list(exts, &actions);
1028 	list_for_each_entry(a, &actions, list) {
1029 		if (a->ops->get_dev) {
1030 			a->ops->get_dev(a, dev_net(dev), hw_dev);
1031 			break;
1032 		}
1033 	}
1034 	if (*hw_dev)
1035 		return 0;
1036 #endif
1037 	return -EOPNOTSUPP;
1038 }
1039 EXPORT_SYMBOL(tcf_exts_get_dev);
1040 
1041 static int __init tc_filter_init(void)
1042 {
1043 	tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
1044 	if (!tc_filter_wq)
1045 		return -ENOMEM;
1046 
1047 	rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, 0);
1048 	rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, 0);
1049 	rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
1050 		      tc_dump_tfilter, 0);
1051 
1052 	return 0;
1053 }
1054 
1055 subsys_initcall(tc_filter_init);
1056