xref: /linux/net/sched/cls_api.c (revision 995231c820e3bd3633cb38bf4ea6f2541e1da331)
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 static void tcf_block_offload_cmd(struct tcf_block *block, struct Qdisc *q,
244 				  struct tcf_block_ext_info *ei,
245 				  enum tc_block_command command)
246 {
247 	struct net_device *dev = q->dev_queue->dev;
248 	struct tc_block_offload bo = {};
249 
250 	if (!tc_can_offload(dev))
251 		return;
252 	bo.command = command;
253 	bo.binder_type = ei->binder_type;
254 	bo.block = block;
255 	dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
256 }
257 
258 static void tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
259 				   struct tcf_block_ext_info *ei)
260 {
261 	tcf_block_offload_cmd(block, q, ei, TC_BLOCK_BIND);
262 }
263 
264 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
265 				     struct tcf_block_ext_info *ei)
266 {
267 	tcf_block_offload_cmd(block, q, ei, TC_BLOCK_UNBIND);
268 }
269 
270 int tcf_block_get_ext(struct tcf_block **p_block,
271 		      struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
272 		      struct tcf_block_ext_info *ei)
273 {
274 	struct tcf_block *block = kzalloc(sizeof(*block), GFP_KERNEL);
275 	struct tcf_chain *chain;
276 	int err;
277 
278 	if (!block)
279 		return -ENOMEM;
280 	INIT_LIST_HEAD(&block->chain_list);
281 	INIT_LIST_HEAD(&block->cb_list);
282 
283 	/* Create chain 0 by default, it has to be always present. */
284 	chain = tcf_chain_create(block, 0);
285 	if (!chain) {
286 		err = -ENOMEM;
287 		goto err_chain_create;
288 	}
289 	tcf_chain_filter_chain_ptr_set(chain, p_filter_chain);
290 	block->net = qdisc_net(q);
291 	block->q = q;
292 	tcf_block_offload_bind(block, q, ei);
293 	*p_block = block;
294 	return 0;
295 
296 err_chain_create:
297 	kfree(block);
298 	return err;
299 }
300 EXPORT_SYMBOL(tcf_block_get_ext);
301 
302 int tcf_block_get(struct tcf_block **p_block,
303 		  struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q)
304 {
305 	struct tcf_block_ext_info ei = {0, };
306 
307 	return tcf_block_get_ext(p_block, p_filter_chain, q, &ei);
308 }
309 EXPORT_SYMBOL(tcf_block_get);
310 
311 void tcf_block_put_ext(struct tcf_block *block,
312 		       struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
313 		       struct tcf_block_ext_info *ei)
314 {
315 	struct tcf_chain *chain, *tmp;
316 
317 	if (!block)
318 		return;
319 
320 	tcf_block_offload_unbind(block, q, ei);
321 
322 	/* XXX: Standalone actions are not allowed to jump to any chain, and
323 	 * bound actions should be all removed after flushing. However,
324 	 * filters are destroyed in RCU callbacks, we have to hold the chains
325 	 * first, otherwise we would always race with RCU callbacks on this list
326 	 * without proper locking.
327 	 */
328 
329 	/* Wait for existing RCU callbacks to cool down. */
330 	rcu_barrier();
331 
332 	/* Hold a refcnt for all chains, except 0, in case they are gone. */
333 	list_for_each_entry(chain, &block->chain_list, list)
334 		if (chain->index)
335 			tcf_chain_hold(chain);
336 
337 	/* No race on the list, because no chain could be destroyed. */
338 	list_for_each_entry(chain, &block->chain_list, list)
339 		tcf_chain_flush(chain);
340 
341 	/* Wait for RCU callbacks to release the reference count. */
342 	rcu_barrier();
343 
344 	/* At this point, all the chains should have refcnt == 1. */
345 	list_for_each_entry_safe(chain, tmp, &block->chain_list, list)
346 		tcf_chain_put(chain);
347 	kfree(block);
348 }
349 EXPORT_SYMBOL(tcf_block_put_ext);
350 
351 void tcf_block_put(struct tcf_block *block)
352 {
353 	struct tcf_block_ext_info ei = {0, };
354 
355 	tcf_block_put_ext(block, NULL, block->q, &ei);
356 }
357 EXPORT_SYMBOL(tcf_block_put);
358 
359 struct tcf_block_cb {
360 	struct list_head list;
361 	tc_setup_cb_t *cb;
362 	void *cb_ident;
363 	void *cb_priv;
364 	unsigned int refcnt;
365 };
366 
367 void *tcf_block_cb_priv(struct tcf_block_cb *block_cb)
368 {
369 	return block_cb->cb_priv;
370 }
371 EXPORT_SYMBOL(tcf_block_cb_priv);
372 
373 struct tcf_block_cb *tcf_block_cb_lookup(struct tcf_block *block,
374 					 tc_setup_cb_t *cb, void *cb_ident)
375 {	struct tcf_block_cb *block_cb;
376 
377 	list_for_each_entry(block_cb, &block->cb_list, list)
378 		if (block_cb->cb == cb && block_cb->cb_ident == cb_ident)
379 			return block_cb;
380 	return NULL;
381 }
382 EXPORT_SYMBOL(tcf_block_cb_lookup);
383 
384 void tcf_block_cb_incref(struct tcf_block_cb *block_cb)
385 {
386 	block_cb->refcnt++;
387 }
388 EXPORT_SYMBOL(tcf_block_cb_incref);
389 
390 unsigned int tcf_block_cb_decref(struct tcf_block_cb *block_cb)
391 {
392 	return --block_cb->refcnt;
393 }
394 EXPORT_SYMBOL(tcf_block_cb_decref);
395 
396 struct tcf_block_cb *__tcf_block_cb_register(struct tcf_block *block,
397 					     tc_setup_cb_t *cb, void *cb_ident,
398 					     void *cb_priv)
399 {
400 	struct tcf_block_cb *block_cb;
401 
402 	block_cb = kzalloc(sizeof(*block_cb), GFP_KERNEL);
403 	if (!block_cb)
404 		return NULL;
405 	block_cb->cb = cb;
406 	block_cb->cb_ident = cb_ident;
407 	block_cb->cb_priv = cb_priv;
408 	list_add(&block_cb->list, &block->cb_list);
409 	return block_cb;
410 }
411 EXPORT_SYMBOL(__tcf_block_cb_register);
412 
413 int tcf_block_cb_register(struct tcf_block *block,
414 			  tc_setup_cb_t *cb, void *cb_ident,
415 			  void *cb_priv)
416 {
417 	struct tcf_block_cb *block_cb;
418 
419 	block_cb = __tcf_block_cb_register(block, cb, cb_ident, cb_priv);
420 	return block_cb ? 0 : -ENOMEM;
421 }
422 EXPORT_SYMBOL(tcf_block_cb_register);
423 
424 void __tcf_block_cb_unregister(struct tcf_block_cb *block_cb)
425 {
426 	list_del(&block_cb->list);
427 	kfree(block_cb);
428 }
429 EXPORT_SYMBOL(__tcf_block_cb_unregister);
430 
431 void tcf_block_cb_unregister(struct tcf_block *block,
432 			     tc_setup_cb_t *cb, void *cb_ident)
433 {
434 	struct tcf_block_cb *block_cb;
435 
436 	block_cb = tcf_block_cb_lookup(block, cb, cb_ident);
437 	if (!block_cb)
438 		return;
439 	__tcf_block_cb_unregister(block_cb);
440 }
441 EXPORT_SYMBOL(tcf_block_cb_unregister);
442 
443 static int tcf_block_cb_call(struct tcf_block *block, enum tc_setup_type type,
444 			     void *type_data, bool err_stop)
445 {
446 	struct tcf_block_cb *block_cb;
447 	int ok_count = 0;
448 	int err;
449 
450 	list_for_each_entry(block_cb, &block->cb_list, list) {
451 		err = block_cb->cb(type, type_data, block_cb->cb_priv);
452 		if (err) {
453 			if (err_stop)
454 				return err;
455 		} else {
456 			ok_count++;
457 		}
458 	}
459 	return ok_count;
460 }
461 
462 /* Main classifier routine: scans classifier chain attached
463  * to this qdisc, (optionally) tests for protocol and asks
464  * specific classifiers.
465  */
466 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
467 		 struct tcf_result *res, bool compat_mode)
468 {
469 	__be16 protocol = tc_skb_protocol(skb);
470 #ifdef CONFIG_NET_CLS_ACT
471 	const int max_reclassify_loop = 4;
472 	const struct tcf_proto *orig_tp = tp;
473 	const struct tcf_proto *first_tp;
474 	int limit = 0;
475 
476 reclassify:
477 #endif
478 	for (; tp; tp = rcu_dereference_bh(tp->next)) {
479 		int err;
480 
481 		if (tp->protocol != protocol &&
482 		    tp->protocol != htons(ETH_P_ALL))
483 			continue;
484 
485 		err = tp->classify(skb, tp, res);
486 #ifdef CONFIG_NET_CLS_ACT
487 		if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
488 			first_tp = orig_tp;
489 			goto reset;
490 		} else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
491 			first_tp = res->goto_tp;
492 			goto reset;
493 		}
494 #endif
495 		if (err >= 0)
496 			return err;
497 	}
498 
499 	return TC_ACT_UNSPEC; /* signal: continue lookup */
500 #ifdef CONFIG_NET_CLS_ACT
501 reset:
502 	if (unlikely(limit++ >= max_reclassify_loop)) {
503 		net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n",
504 				       tp->q->ops->id, tp->prio & 0xffff,
505 				       ntohs(tp->protocol));
506 		return TC_ACT_SHOT;
507 	}
508 
509 	tp = first_tp;
510 	protocol = tc_skb_protocol(skb);
511 	goto reclassify;
512 #endif
513 }
514 EXPORT_SYMBOL(tcf_classify);
515 
516 struct tcf_chain_info {
517 	struct tcf_proto __rcu **pprev;
518 	struct tcf_proto __rcu *next;
519 };
520 
521 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain_info *chain_info)
522 {
523 	return rtnl_dereference(*chain_info->pprev);
524 }
525 
526 static void tcf_chain_tp_insert(struct tcf_chain *chain,
527 				struct tcf_chain_info *chain_info,
528 				struct tcf_proto *tp)
529 {
530 	if (chain->p_filter_chain &&
531 	    *chain_info->pprev == chain->filter_chain)
532 		rcu_assign_pointer(*chain->p_filter_chain, tp);
533 	RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain_info));
534 	rcu_assign_pointer(*chain_info->pprev, tp);
535 	tcf_chain_hold(chain);
536 }
537 
538 static void tcf_chain_tp_remove(struct tcf_chain *chain,
539 				struct tcf_chain_info *chain_info,
540 				struct tcf_proto *tp)
541 {
542 	struct tcf_proto *next = rtnl_dereference(chain_info->next);
543 
544 	if (chain->p_filter_chain && tp == chain->filter_chain)
545 		RCU_INIT_POINTER(*chain->p_filter_chain, next);
546 	RCU_INIT_POINTER(*chain_info->pprev, next);
547 	tcf_chain_put(chain);
548 }
549 
550 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
551 					   struct tcf_chain_info *chain_info,
552 					   u32 protocol, u32 prio,
553 					   bool prio_allocate)
554 {
555 	struct tcf_proto **pprev;
556 	struct tcf_proto *tp;
557 
558 	/* Check the chain for existence of proto-tcf with this priority */
559 	for (pprev = &chain->filter_chain;
560 	     (tp = rtnl_dereference(*pprev)); pprev = &tp->next) {
561 		if (tp->prio >= prio) {
562 			if (tp->prio == prio) {
563 				if (prio_allocate ||
564 				    (tp->protocol != protocol && protocol))
565 					return ERR_PTR(-EINVAL);
566 			} else {
567 				tp = NULL;
568 			}
569 			break;
570 		}
571 	}
572 	chain_info->pprev = pprev;
573 	chain_info->next = tp ? tp->next : NULL;
574 	return tp;
575 }
576 
577 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
578 			 struct tcf_proto *tp, struct Qdisc *q, u32 parent,
579 			 void *fh, u32 portid, u32 seq, u16 flags, int event)
580 {
581 	struct tcmsg *tcm;
582 	struct nlmsghdr  *nlh;
583 	unsigned char *b = skb_tail_pointer(skb);
584 
585 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
586 	if (!nlh)
587 		goto out_nlmsg_trim;
588 	tcm = nlmsg_data(nlh);
589 	tcm->tcm_family = AF_UNSPEC;
590 	tcm->tcm__pad1 = 0;
591 	tcm->tcm__pad2 = 0;
592 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
593 	tcm->tcm_parent = parent;
594 	tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
595 	if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
596 		goto nla_put_failure;
597 	if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
598 		goto nla_put_failure;
599 	if (!fh) {
600 		tcm->tcm_handle = 0;
601 	} else {
602 		if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
603 			goto nla_put_failure;
604 	}
605 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
606 	return skb->len;
607 
608 out_nlmsg_trim:
609 nla_put_failure:
610 	nlmsg_trim(skb, b);
611 	return -1;
612 }
613 
614 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
615 			  struct nlmsghdr *n, struct tcf_proto *tp,
616 			  struct Qdisc *q, u32 parent,
617 			  void *fh, int event, bool unicast)
618 {
619 	struct sk_buff *skb;
620 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
621 
622 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
623 	if (!skb)
624 		return -ENOBUFS;
625 
626 	if (tcf_fill_node(net, skb, tp, q, parent, fh, portid, n->nlmsg_seq,
627 			  n->nlmsg_flags, event) <= 0) {
628 		kfree_skb(skb);
629 		return -EINVAL;
630 	}
631 
632 	if (unicast)
633 		return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
634 
635 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
636 			      n->nlmsg_flags & NLM_F_ECHO);
637 }
638 
639 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
640 			      struct nlmsghdr *n, struct tcf_proto *tp,
641 			      struct Qdisc *q, u32 parent,
642 			      void *fh, bool unicast, bool *last)
643 {
644 	struct sk_buff *skb;
645 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
646 	int err;
647 
648 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
649 	if (!skb)
650 		return -ENOBUFS;
651 
652 	if (tcf_fill_node(net, skb, tp, q, parent, fh, portid, n->nlmsg_seq,
653 			  n->nlmsg_flags, RTM_DELTFILTER) <= 0) {
654 		kfree_skb(skb);
655 		return -EINVAL;
656 	}
657 
658 	err = tp->ops->delete(tp, fh, last);
659 	if (err) {
660 		kfree_skb(skb);
661 		return err;
662 	}
663 
664 	if (unicast)
665 		return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
666 
667 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
668 			      n->nlmsg_flags & NLM_F_ECHO);
669 }
670 
671 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
672 				 struct Qdisc *q, u32 parent,
673 				 struct nlmsghdr *n,
674 				 struct tcf_chain *chain, int event)
675 {
676 	struct tcf_proto *tp;
677 
678 	for (tp = rtnl_dereference(chain->filter_chain);
679 	     tp; tp = rtnl_dereference(tp->next))
680 		tfilter_notify(net, oskb, n, tp, q, parent, 0, event, false);
681 }
682 
683 /* Add/change/delete/get a filter node */
684 
685 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
686 			  struct netlink_ext_ack *extack)
687 {
688 	struct net *net = sock_net(skb->sk);
689 	struct nlattr *tca[TCA_MAX + 1];
690 	struct tcmsg *t;
691 	u32 protocol;
692 	u32 prio;
693 	bool prio_allocate;
694 	u32 parent;
695 	u32 chain_index;
696 	struct net_device *dev;
697 	struct Qdisc  *q;
698 	struct tcf_chain_info chain_info;
699 	struct tcf_chain *chain = NULL;
700 	struct tcf_block *block;
701 	struct tcf_proto *tp;
702 	const struct Qdisc_class_ops *cops;
703 	unsigned long cl;
704 	void *fh;
705 	int err;
706 	int tp_created;
707 
708 	if ((n->nlmsg_type != RTM_GETTFILTER) &&
709 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
710 		return -EPERM;
711 
712 replay:
713 	tp_created = 0;
714 
715 	err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack);
716 	if (err < 0)
717 		return err;
718 
719 	t = nlmsg_data(n);
720 	protocol = TC_H_MIN(t->tcm_info);
721 	prio = TC_H_MAJ(t->tcm_info);
722 	prio_allocate = false;
723 	parent = t->tcm_parent;
724 	cl = 0;
725 
726 	if (prio == 0) {
727 		switch (n->nlmsg_type) {
728 		case RTM_DELTFILTER:
729 			if (protocol || t->tcm_handle || tca[TCA_KIND])
730 				return -ENOENT;
731 			break;
732 		case RTM_NEWTFILTER:
733 			/* If no priority is provided by the user,
734 			 * we allocate one.
735 			 */
736 			if (n->nlmsg_flags & NLM_F_CREATE) {
737 				prio = TC_H_MAKE(0x80000000U, 0U);
738 				prio_allocate = true;
739 				break;
740 			}
741 			/* fall-through */
742 		default:
743 			return -ENOENT;
744 		}
745 	}
746 
747 	/* Find head of filter chain. */
748 
749 	/* Find link */
750 	dev = __dev_get_by_index(net, t->tcm_ifindex);
751 	if (dev == NULL)
752 		return -ENODEV;
753 
754 	/* Find qdisc */
755 	if (!parent) {
756 		q = dev->qdisc;
757 		parent = q->handle;
758 	} else {
759 		q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
760 		if (q == NULL)
761 			return -EINVAL;
762 	}
763 
764 	/* Is it classful? */
765 	cops = q->ops->cl_ops;
766 	if (!cops)
767 		return -EINVAL;
768 
769 	if (!cops->tcf_block)
770 		return -EOPNOTSUPP;
771 
772 	/* Do we search for filter, attached to class? */
773 	if (TC_H_MIN(parent)) {
774 		cl = cops->find(q, parent);
775 		if (cl == 0)
776 			return -ENOENT;
777 	}
778 
779 	/* And the last stroke */
780 	block = cops->tcf_block(q, cl);
781 	if (!block) {
782 		err = -EINVAL;
783 		goto errout;
784 	}
785 
786 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
787 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
788 		err = -EINVAL;
789 		goto errout;
790 	}
791 	chain = tcf_chain_get(block, chain_index,
792 			      n->nlmsg_type == RTM_NEWTFILTER);
793 	if (!chain) {
794 		err = n->nlmsg_type == RTM_NEWTFILTER ? -ENOMEM : -EINVAL;
795 		goto errout;
796 	}
797 
798 	if (n->nlmsg_type == RTM_DELTFILTER && prio == 0) {
799 		tfilter_notify_chain(net, skb, q, parent, n,
800 				     chain, RTM_DELTFILTER);
801 		tcf_chain_flush(chain);
802 		err = 0;
803 		goto errout;
804 	}
805 
806 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
807 			       prio, prio_allocate);
808 	if (IS_ERR(tp)) {
809 		err = PTR_ERR(tp);
810 		goto errout;
811 	}
812 
813 	if (tp == NULL) {
814 		/* Proto-tcf does not exist, create new one */
815 
816 		if (tca[TCA_KIND] == NULL || !protocol) {
817 			err = -EINVAL;
818 			goto errout;
819 		}
820 
821 		if (n->nlmsg_type != RTM_NEWTFILTER ||
822 		    !(n->nlmsg_flags & NLM_F_CREATE)) {
823 			err = -ENOENT;
824 			goto errout;
825 		}
826 
827 		if (prio_allocate)
828 			prio = tcf_auto_prio(tcf_chain_tp_prev(&chain_info));
829 
830 		tp = tcf_proto_create(nla_data(tca[TCA_KIND]),
831 				      protocol, prio, parent, q, chain);
832 		if (IS_ERR(tp)) {
833 			err = PTR_ERR(tp);
834 			goto errout;
835 		}
836 		tp_created = 1;
837 	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
838 		err = -EINVAL;
839 		goto errout;
840 	}
841 
842 	fh = tp->ops->get(tp, t->tcm_handle);
843 
844 	if (!fh) {
845 		if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
846 			tcf_chain_tp_remove(chain, &chain_info, tp);
847 			tfilter_notify(net, skb, n, tp, q, parent, fh,
848 				       RTM_DELTFILTER, false);
849 			tcf_proto_destroy(tp);
850 			err = 0;
851 			goto errout;
852 		}
853 
854 		if (n->nlmsg_type != RTM_NEWTFILTER ||
855 		    !(n->nlmsg_flags & NLM_F_CREATE)) {
856 			err = -ENOENT;
857 			goto errout;
858 		}
859 	} else {
860 		bool last;
861 
862 		switch (n->nlmsg_type) {
863 		case RTM_NEWTFILTER:
864 			if (n->nlmsg_flags & NLM_F_EXCL) {
865 				if (tp_created)
866 					tcf_proto_destroy(tp);
867 				err = -EEXIST;
868 				goto errout;
869 			}
870 			break;
871 		case RTM_DELTFILTER:
872 			err = tfilter_del_notify(net, skb, n, tp, q, parent,
873 						 fh, false, &last);
874 			if (err)
875 				goto errout;
876 			if (last) {
877 				tcf_chain_tp_remove(chain, &chain_info, tp);
878 				tcf_proto_destroy(tp);
879 			}
880 			goto errout;
881 		case RTM_GETTFILTER:
882 			err = tfilter_notify(net, skb, n, tp, q, parent, fh,
883 					     RTM_NEWTFILTER, true);
884 			goto errout;
885 		default:
886 			err = -EINVAL;
887 			goto errout;
888 		}
889 	}
890 
891 	err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
892 			      n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE);
893 	if (err == 0) {
894 		if (tp_created)
895 			tcf_chain_tp_insert(chain, &chain_info, tp);
896 		tfilter_notify(net, skb, n, tp, q, parent, fh,
897 			       RTM_NEWTFILTER, false);
898 	} else {
899 		if (tp_created)
900 			tcf_proto_destroy(tp);
901 	}
902 
903 errout:
904 	if (chain)
905 		tcf_chain_put(chain);
906 	if (err == -EAGAIN)
907 		/* Replay the request. */
908 		goto replay;
909 	return err;
910 }
911 
912 struct tcf_dump_args {
913 	struct tcf_walker w;
914 	struct sk_buff *skb;
915 	struct netlink_callback *cb;
916 	struct Qdisc *q;
917 	u32 parent;
918 };
919 
920 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
921 {
922 	struct tcf_dump_args *a = (void *)arg;
923 	struct net *net = sock_net(a->skb->sk);
924 
925 	return tcf_fill_node(net, a->skb, tp, a->q, a->parent,
926 			     n, NETLINK_CB(a->cb->skb).portid,
927 			     a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
928 			     RTM_NEWTFILTER);
929 }
930 
931 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
932 			   struct sk_buff *skb, struct netlink_callback *cb,
933 			   long index_start, long *p_index)
934 {
935 	struct net *net = sock_net(skb->sk);
936 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
937 	struct tcf_dump_args arg;
938 	struct tcf_proto *tp;
939 
940 	for (tp = rtnl_dereference(chain->filter_chain);
941 	     tp; tp = rtnl_dereference(tp->next), (*p_index)++) {
942 		if (*p_index < index_start)
943 			continue;
944 		if (TC_H_MAJ(tcm->tcm_info) &&
945 		    TC_H_MAJ(tcm->tcm_info) != tp->prio)
946 			continue;
947 		if (TC_H_MIN(tcm->tcm_info) &&
948 		    TC_H_MIN(tcm->tcm_info) != tp->protocol)
949 			continue;
950 		if (*p_index > index_start)
951 			memset(&cb->args[1], 0,
952 			       sizeof(cb->args) - sizeof(cb->args[0]));
953 		if (cb->args[1] == 0) {
954 			if (tcf_fill_node(net, skb, tp, q, parent, 0,
955 					  NETLINK_CB(cb->skb).portid,
956 					  cb->nlh->nlmsg_seq, NLM_F_MULTI,
957 					  RTM_NEWTFILTER) <= 0)
958 				return false;
959 
960 			cb->args[1] = 1;
961 		}
962 		if (!tp->ops->walk)
963 			continue;
964 		arg.w.fn = tcf_node_dump;
965 		arg.skb = skb;
966 		arg.cb = cb;
967 		arg.q = q;
968 		arg.parent = parent;
969 		arg.w.stop = 0;
970 		arg.w.skip = cb->args[1] - 1;
971 		arg.w.count = 0;
972 		tp->ops->walk(tp, &arg.w);
973 		cb->args[1] = arg.w.count + 1;
974 		if (arg.w.stop)
975 			return false;
976 	}
977 	return true;
978 }
979 
980 /* called with RTNL */
981 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
982 {
983 	struct net *net = sock_net(skb->sk);
984 	struct nlattr *tca[TCA_MAX + 1];
985 	struct net_device *dev;
986 	struct Qdisc *q;
987 	struct tcf_block *block;
988 	struct tcf_chain *chain;
989 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
990 	unsigned long cl = 0;
991 	const struct Qdisc_class_ops *cops;
992 	long index_start;
993 	long index;
994 	u32 parent;
995 	int err;
996 
997 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
998 		return skb->len;
999 
1000 	err = nlmsg_parse(cb->nlh, sizeof(*tcm), tca, TCA_MAX, NULL, NULL);
1001 	if (err)
1002 		return err;
1003 
1004 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1005 	if (!dev)
1006 		return skb->len;
1007 
1008 	parent = tcm->tcm_parent;
1009 	if (!parent) {
1010 		q = dev->qdisc;
1011 		parent = q->handle;
1012 	} else {
1013 		q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
1014 	}
1015 	if (!q)
1016 		goto out;
1017 	cops = q->ops->cl_ops;
1018 	if (!cops)
1019 		goto out;
1020 	if (!cops->tcf_block)
1021 		goto out;
1022 	if (TC_H_MIN(tcm->tcm_parent)) {
1023 		cl = cops->find(q, tcm->tcm_parent);
1024 		if (cl == 0)
1025 			goto out;
1026 	}
1027 	block = cops->tcf_block(q, cl);
1028 	if (!block)
1029 		goto out;
1030 
1031 	index_start = cb->args[0];
1032 	index = 0;
1033 
1034 	list_for_each_entry(chain, &block->chain_list, list) {
1035 		if (tca[TCA_CHAIN] &&
1036 		    nla_get_u32(tca[TCA_CHAIN]) != chain->index)
1037 			continue;
1038 		if (!tcf_chain_dump(chain, q, parent, skb, cb,
1039 				    index_start, &index))
1040 			break;
1041 	}
1042 
1043 	cb->args[0] = index;
1044 
1045 out:
1046 	return skb->len;
1047 }
1048 
1049 void tcf_exts_destroy(struct tcf_exts *exts)
1050 {
1051 #ifdef CONFIG_NET_CLS_ACT
1052 	LIST_HEAD(actions);
1053 
1054 	tcf_exts_to_list(exts, &actions);
1055 	tcf_action_destroy(&actions, TCA_ACT_UNBIND);
1056 	kfree(exts->actions);
1057 	exts->nr_actions = 0;
1058 #endif
1059 }
1060 EXPORT_SYMBOL(tcf_exts_destroy);
1061 
1062 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
1063 		      struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr)
1064 {
1065 #ifdef CONFIG_NET_CLS_ACT
1066 	{
1067 		struct tc_action *act;
1068 
1069 		if (exts->police && tb[exts->police]) {
1070 			act = tcf_action_init_1(net, tp, tb[exts->police],
1071 						rate_tlv, "police", ovr,
1072 						TCA_ACT_BIND);
1073 			if (IS_ERR(act))
1074 				return PTR_ERR(act);
1075 
1076 			act->type = exts->type = TCA_OLD_COMPAT;
1077 			exts->actions[0] = act;
1078 			exts->nr_actions = 1;
1079 		} else if (exts->action && tb[exts->action]) {
1080 			LIST_HEAD(actions);
1081 			int err, i = 0;
1082 
1083 			err = tcf_action_init(net, tp, tb[exts->action],
1084 					      rate_tlv, NULL, ovr, TCA_ACT_BIND,
1085 					      &actions);
1086 			if (err)
1087 				return err;
1088 			list_for_each_entry(act, &actions, list)
1089 				exts->actions[i++] = act;
1090 			exts->nr_actions = i;
1091 		}
1092 	}
1093 #else
1094 	if ((exts->action && tb[exts->action]) ||
1095 	    (exts->police && tb[exts->police]))
1096 		return -EOPNOTSUPP;
1097 #endif
1098 
1099 	return 0;
1100 }
1101 EXPORT_SYMBOL(tcf_exts_validate);
1102 
1103 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
1104 {
1105 #ifdef CONFIG_NET_CLS_ACT
1106 	struct tcf_exts old = *dst;
1107 
1108 	*dst = *src;
1109 	tcf_exts_destroy(&old);
1110 #endif
1111 }
1112 EXPORT_SYMBOL(tcf_exts_change);
1113 
1114 #ifdef CONFIG_NET_CLS_ACT
1115 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
1116 {
1117 	if (exts->nr_actions == 0)
1118 		return NULL;
1119 	else
1120 		return exts->actions[0];
1121 }
1122 #endif
1123 
1124 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
1125 {
1126 #ifdef CONFIG_NET_CLS_ACT
1127 	struct nlattr *nest;
1128 
1129 	if (exts->action && tcf_exts_has_actions(exts)) {
1130 		/*
1131 		 * again for backward compatible mode - we want
1132 		 * to work with both old and new modes of entering
1133 		 * tc data even if iproute2  was newer - jhs
1134 		 */
1135 		if (exts->type != TCA_OLD_COMPAT) {
1136 			LIST_HEAD(actions);
1137 
1138 			nest = nla_nest_start(skb, exts->action);
1139 			if (nest == NULL)
1140 				goto nla_put_failure;
1141 
1142 			tcf_exts_to_list(exts, &actions);
1143 			if (tcf_action_dump(skb, &actions, 0, 0) < 0)
1144 				goto nla_put_failure;
1145 			nla_nest_end(skb, nest);
1146 		} else if (exts->police) {
1147 			struct tc_action *act = tcf_exts_first_act(exts);
1148 			nest = nla_nest_start(skb, exts->police);
1149 			if (nest == NULL || !act)
1150 				goto nla_put_failure;
1151 			if (tcf_action_dump_old(skb, act, 0, 0) < 0)
1152 				goto nla_put_failure;
1153 			nla_nest_end(skb, nest);
1154 		}
1155 	}
1156 	return 0;
1157 
1158 nla_put_failure:
1159 	nla_nest_cancel(skb, nest);
1160 	return -1;
1161 #else
1162 	return 0;
1163 #endif
1164 }
1165 EXPORT_SYMBOL(tcf_exts_dump);
1166 
1167 
1168 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
1169 {
1170 #ifdef CONFIG_NET_CLS_ACT
1171 	struct tc_action *a = tcf_exts_first_act(exts);
1172 	if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
1173 		return -1;
1174 #endif
1175 	return 0;
1176 }
1177 EXPORT_SYMBOL(tcf_exts_dump_stats);
1178 
1179 static int tc_exts_setup_cb_egdev_call(struct tcf_exts *exts,
1180 				       enum tc_setup_type type,
1181 				       void *type_data, bool err_stop)
1182 {
1183 	int ok_count = 0;
1184 #ifdef CONFIG_NET_CLS_ACT
1185 	const struct tc_action *a;
1186 	struct net_device *dev;
1187 	int i, ret;
1188 
1189 	if (!tcf_exts_has_actions(exts))
1190 		return 0;
1191 
1192 	for (i = 0; i < exts->nr_actions; i++) {
1193 		a = exts->actions[i];
1194 		if (!a->ops->get_dev)
1195 			continue;
1196 		dev = a->ops->get_dev(a);
1197 		if (!dev || !tc_can_offload(dev))
1198 			continue;
1199 		ret = tc_setup_cb_egdev_call(dev, type, type_data, err_stop);
1200 		if (ret < 0)
1201 			return ret;
1202 		ok_count += ret;
1203 	}
1204 #endif
1205 	return ok_count;
1206 }
1207 
1208 int tc_setup_cb_call(struct tcf_block *block, struct tcf_exts *exts,
1209 		     enum tc_setup_type type, void *type_data, bool err_stop)
1210 {
1211 	int ok_count;
1212 	int ret;
1213 
1214 	ret = tcf_block_cb_call(block, type, type_data, err_stop);
1215 	if (ret < 0)
1216 		return ret;
1217 	ok_count = ret;
1218 
1219 	if (!exts)
1220 		return ok_count;
1221 	ret = tc_exts_setup_cb_egdev_call(exts, type, type_data, err_stop);
1222 	if (ret < 0)
1223 		return ret;
1224 	ok_count += ret;
1225 
1226 	return ok_count;
1227 }
1228 EXPORT_SYMBOL(tc_setup_cb_call);
1229 
1230 static int __init tc_filter_init(void)
1231 {
1232 	rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, 0);
1233 	rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, 0);
1234 	rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
1235 		      tc_dump_tfilter, 0);
1236 
1237 	return 0;
1238 }
1239 
1240 subsys_initcall(tc_filter_init);
1241