xref: /linux/net/sched/cls_api.c (revision fd1557923b2e45a7ff4664bd88311d8239d69ca9)
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/skbuff.h>
23 #include <linux/init.h>
24 #include <linux/kmod.h>
25 #include <linux/err.h>
26 #include <linux/slab.h>
27 #include <net/net_namespace.h>
28 #include <net/sock.h>
29 #include <net/netlink.h>
30 #include <net/pkt_sched.h>
31 #include <net/pkt_cls.h>
32 
33 /* The list of all installed classifier types */
34 static LIST_HEAD(tcf_proto_base);
35 
36 /* Protects list of registered TC modules. It is pure SMP lock. */
37 static DEFINE_RWLOCK(cls_mod_lock);
38 
39 /* Find classifier type by string name */
40 
41 static const struct tcf_proto_ops *tcf_proto_lookup_ops(struct nlattr *kind)
42 {
43 	const struct tcf_proto_ops *t, *res = NULL;
44 
45 	if (kind) {
46 		read_lock(&cls_mod_lock);
47 		list_for_each_entry(t, &tcf_proto_base, head) {
48 			if (nla_strcmp(kind, t->kind) == 0) {
49 				if (try_module_get(t->owner))
50 					res = t;
51 				break;
52 			}
53 		}
54 		read_unlock(&cls_mod_lock);
55 	}
56 	return res;
57 }
58 
59 /* Register(unregister) new classifier type */
60 
61 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
62 {
63 	struct tcf_proto_ops *t;
64 	int rc = -EEXIST;
65 
66 	write_lock(&cls_mod_lock);
67 	list_for_each_entry(t, &tcf_proto_base, head)
68 		if (!strcmp(ops->kind, t->kind))
69 			goto out;
70 
71 	list_add_tail(&ops->head, &tcf_proto_base);
72 	rc = 0;
73 out:
74 	write_unlock(&cls_mod_lock);
75 	return rc;
76 }
77 EXPORT_SYMBOL(register_tcf_proto_ops);
78 
79 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
80 {
81 	struct tcf_proto_ops *t;
82 	int rc = -ENOENT;
83 
84 	write_lock(&cls_mod_lock);
85 	list_for_each_entry(t, &tcf_proto_base, head) {
86 		if (t == ops) {
87 			list_del(&t->head);
88 			rc = 0;
89 			break;
90 		}
91 	}
92 	write_unlock(&cls_mod_lock);
93 	return rc;
94 }
95 EXPORT_SYMBOL(unregister_tcf_proto_ops);
96 
97 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
98 			  struct nlmsghdr *n, struct tcf_proto *tp,
99 			  unsigned long fh, int event);
100 
101 
102 /* Select new prio value from the range, managed by kernel. */
103 
104 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
105 {
106 	u32 first = TC_H_MAKE(0xC0000000U, 0U);
107 
108 	if (tp)
109 		first = tp->prio - 1;
110 
111 	return first;
112 }
113 
114 /* Add/change/delete/get a filter node */
115 
116 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n)
117 {
118 	struct net *net = sock_net(skb->sk);
119 	struct nlattr *tca[TCA_MAX + 1];
120 	struct tcmsg *t;
121 	u32 protocol;
122 	u32 prio;
123 	u32 nprio;
124 	u32 parent;
125 	struct net_device *dev;
126 	struct Qdisc  *q;
127 	struct tcf_proto __rcu **back;
128 	struct tcf_proto __rcu **chain;
129 	struct tcf_proto *tp;
130 	const struct tcf_proto_ops *tp_ops;
131 	const struct Qdisc_class_ops *cops;
132 	unsigned long cl;
133 	unsigned long fh;
134 	int err;
135 	int tp_created = 0;
136 
137 	if ((n->nlmsg_type != RTM_GETTFILTER) &&
138 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
139 		return -EPERM;
140 
141 replay:
142 	err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
143 	if (err < 0)
144 		return err;
145 
146 	t = nlmsg_data(n);
147 	protocol = TC_H_MIN(t->tcm_info);
148 	prio = TC_H_MAJ(t->tcm_info);
149 	nprio = prio;
150 	parent = t->tcm_parent;
151 	cl = 0;
152 
153 	if (prio == 0) {
154 		/* If no priority is given, user wants we allocated it. */
155 		if (n->nlmsg_type != RTM_NEWTFILTER ||
156 		    !(n->nlmsg_flags & NLM_F_CREATE))
157 			return -ENOENT;
158 		prio = TC_H_MAKE(0x80000000U, 0U);
159 	}
160 
161 	/* Find head of filter chain. */
162 
163 	/* Find link */
164 	dev = __dev_get_by_index(net, t->tcm_ifindex);
165 	if (dev == NULL)
166 		return -ENODEV;
167 
168 	/* Find qdisc */
169 	if (!parent) {
170 		q = dev->qdisc;
171 		parent = q->handle;
172 	} else {
173 		q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
174 		if (q == NULL)
175 			return -EINVAL;
176 	}
177 
178 	/* Is it classful? */
179 	cops = q->ops->cl_ops;
180 	if (!cops)
181 		return -EINVAL;
182 
183 	if (cops->tcf_chain == NULL)
184 		return -EOPNOTSUPP;
185 
186 	/* Do we search for filter, attached to class? */
187 	if (TC_H_MIN(parent)) {
188 		cl = cops->get(q, parent);
189 		if (cl == 0)
190 			return -ENOENT;
191 	}
192 
193 	/* And the last stroke */
194 	chain = cops->tcf_chain(q, cl);
195 	err = -EINVAL;
196 	if (chain == NULL)
197 		goto errout;
198 
199 	/* Check the chain for existence of proto-tcf with this priority */
200 	for (back = chain;
201 	     (tp = rtnl_dereference(*back)) != NULL;
202 	     back = &tp->next) {
203 		if (tp->prio >= prio) {
204 			if (tp->prio == prio) {
205 				if (!nprio ||
206 				    (tp->protocol != protocol && protocol))
207 					goto errout;
208 			} else
209 				tp = NULL;
210 			break;
211 		}
212 	}
213 
214 	if (tp == NULL) {
215 		/* Proto-tcf does not exist, create new one */
216 
217 		if (tca[TCA_KIND] == NULL || !protocol)
218 			goto errout;
219 
220 		err = -ENOENT;
221 		if (n->nlmsg_type != RTM_NEWTFILTER ||
222 		    !(n->nlmsg_flags & NLM_F_CREATE))
223 			goto errout;
224 
225 
226 		/* Create new proto tcf */
227 
228 		err = -ENOBUFS;
229 		tp = kzalloc(sizeof(*tp), GFP_KERNEL);
230 		if (tp == NULL)
231 			goto errout;
232 		err = -ENOENT;
233 		tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
234 		if (tp_ops == NULL) {
235 #ifdef CONFIG_MODULES
236 			struct nlattr *kind = tca[TCA_KIND];
237 			char name[IFNAMSIZ];
238 
239 			if (kind != NULL &&
240 			    nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
241 				rtnl_unlock();
242 				request_module("cls_%s", name);
243 				rtnl_lock();
244 				tp_ops = tcf_proto_lookup_ops(kind);
245 				/* We dropped the RTNL semaphore in order to
246 				 * perform the module load.  So, even if we
247 				 * succeeded in loading the module we have to
248 				 * replay the request.  We indicate this using
249 				 * -EAGAIN.
250 				 */
251 				if (tp_ops != NULL) {
252 					module_put(tp_ops->owner);
253 					err = -EAGAIN;
254 				}
255 			}
256 #endif
257 			kfree(tp);
258 			goto errout;
259 		}
260 		tp->ops = tp_ops;
261 		tp->protocol = protocol;
262 		tp->prio = nprio ? :
263 			       TC_H_MAJ(tcf_auto_prio(rtnl_dereference(*back)));
264 		tp->q = q;
265 		tp->classify = tp_ops->classify;
266 		tp->classid = parent;
267 
268 		err = tp_ops->init(tp);
269 		if (err != 0) {
270 			module_put(tp_ops->owner);
271 			kfree(tp);
272 			goto errout;
273 		}
274 
275 		tp_created = 1;
276 
277 	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
278 		goto errout;
279 
280 	fh = tp->ops->get(tp, t->tcm_handle);
281 
282 	if (fh == 0) {
283 		if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
284 			struct tcf_proto *next = rtnl_dereference(tp->next);
285 
286 			RCU_INIT_POINTER(*back, next);
287 
288 			tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
289 			tcf_destroy(tp, true);
290 			err = 0;
291 			goto errout;
292 		}
293 
294 		err = -ENOENT;
295 		if (n->nlmsg_type != RTM_NEWTFILTER ||
296 		    !(n->nlmsg_flags & NLM_F_CREATE))
297 			goto errout;
298 	} else {
299 		switch (n->nlmsg_type) {
300 		case RTM_NEWTFILTER:
301 			err = -EEXIST;
302 			if (n->nlmsg_flags & NLM_F_EXCL) {
303 				if (tp_created)
304 					tcf_destroy(tp, true);
305 				goto errout;
306 			}
307 			break;
308 		case RTM_DELTFILTER:
309 			err = tp->ops->delete(tp, fh);
310 			if (err == 0) {
311 				struct tcf_proto *next = rtnl_dereference(tp->next);
312 
313 				tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
314 				if (tcf_destroy(tp, false))
315 					RCU_INIT_POINTER(*back, next);
316 			}
317 			goto errout;
318 		case RTM_GETTFILTER:
319 			err = tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
320 			goto errout;
321 		default:
322 			err = -EINVAL;
323 			goto errout;
324 		}
325 	}
326 
327 	err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
328 			      n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE);
329 	if (err == 0) {
330 		if (tp_created) {
331 			RCU_INIT_POINTER(tp->next, rtnl_dereference(*back));
332 			rcu_assign_pointer(*back, tp);
333 		}
334 		tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
335 	} else {
336 		if (tp_created)
337 			tcf_destroy(tp, true);
338 	}
339 
340 errout:
341 	if (cl)
342 		cops->put(q, cl);
343 	if (err == -EAGAIN)
344 		/* Replay the request. */
345 		goto replay;
346 	return err;
347 }
348 
349 static int tcf_fill_node(struct net *net, struct sk_buff *skb, struct tcf_proto *tp,
350 			 unsigned long fh, u32 portid, u32 seq, u16 flags, int event)
351 {
352 	struct tcmsg *tcm;
353 	struct nlmsghdr  *nlh;
354 	unsigned char *b = skb_tail_pointer(skb);
355 
356 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
357 	if (!nlh)
358 		goto out_nlmsg_trim;
359 	tcm = nlmsg_data(nlh);
360 	tcm->tcm_family = AF_UNSPEC;
361 	tcm->tcm__pad1 = 0;
362 	tcm->tcm__pad2 = 0;
363 	tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
364 	tcm->tcm_parent = tp->classid;
365 	tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
366 	if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
367 		goto nla_put_failure;
368 	tcm->tcm_handle = fh;
369 	if (RTM_DELTFILTER != event) {
370 		tcm->tcm_handle = 0;
371 		if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
372 			goto nla_put_failure;
373 	}
374 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
375 	return skb->len;
376 
377 out_nlmsg_trim:
378 nla_put_failure:
379 	nlmsg_trim(skb, b);
380 	return -1;
381 }
382 
383 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
384 			  struct nlmsghdr *n, struct tcf_proto *tp,
385 			  unsigned long fh, int event)
386 {
387 	struct sk_buff *skb;
388 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
389 
390 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
391 	if (!skb)
392 		return -ENOBUFS;
393 
394 	if (tcf_fill_node(net, skb, tp, fh, portid, n->nlmsg_seq, 0, event) <= 0) {
395 		kfree_skb(skb);
396 		return -EINVAL;
397 	}
398 
399 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
400 			      n->nlmsg_flags & NLM_F_ECHO);
401 }
402 
403 struct tcf_dump_args {
404 	struct tcf_walker w;
405 	struct sk_buff *skb;
406 	struct netlink_callback *cb;
407 };
408 
409 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
410 			 struct tcf_walker *arg)
411 {
412 	struct tcf_dump_args *a = (void *)arg;
413 	struct net *net = sock_net(a->skb->sk);
414 
415 	return tcf_fill_node(net, a->skb, tp, n, NETLINK_CB(a->cb->skb).portid,
416 			     a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
417 }
418 
419 /* called with RTNL */
420 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
421 {
422 	struct net *net = sock_net(skb->sk);
423 	int t;
424 	int s_t;
425 	struct net_device *dev;
426 	struct Qdisc *q;
427 	struct tcf_proto *tp, __rcu **chain;
428 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
429 	unsigned long cl = 0;
430 	const struct Qdisc_class_ops *cops;
431 	struct tcf_dump_args arg;
432 
433 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
434 		return skb->len;
435 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
436 	if (!dev)
437 		return skb->len;
438 
439 	if (!tcm->tcm_parent)
440 		q = dev->qdisc;
441 	else
442 		q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
443 	if (!q)
444 		goto out;
445 	cops = q->ops->cl_ops;
446 	if (!cops)
447 		goto errout;
448 	if (cops->tcf_chain == NULL)
449 		goto errout;
450 	if (TC_H_MIN(tcm->tcm_parent)) {
451 		cl = cops->get(q, tcm->tcm_parent);
452 		if (cl == 0)
453 			goto errout;
454 	}
455 	chain = cops->tcf_chain(q, cl);
456 	if (chain == NULL)
457 		goto errout;
458 
459 	s_t = cb->args[0];
460 
461 	for (tp = rtnl_dereference(*chain), t = 0;
462 	     tp; tp = rtnl_dereference(tp->next), t++) {
463 		if (t < s_t)
464 			continue;
465 		if (TC_H_MAJ(tcm->tcm_info) &&
466 		    TC_H_MAJ(tcm->tcm_info) != tp->prio)
467 			continue;
468 		if (TC_H_MIN(tcm->tcm_info) &&
469 		    TC_H_MIN(tcm->tcm_info) != tp->protocol)
470 			continue;
471 		if (t > s_t)
472 			memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
473 		if (cb->args[1] == 0) {
474 			if (tcf_fill_node(net, skb, tp, 0, NETLINK_CB(cb->skb).portid,
475 					  cb->nlh->nlmsg_seq, NLM_F_MULTI,
476 					  RTM_NEWTFILTER) <= 0)
477 				break;
478 
479 			cb->args[1] = 1;
480 		}
481 		if (tp->ops->walk == NULL)
482 			continue;
483 		arg.w.fn = tcf_node_dump;
484 		arg.skb = skb;
485 		arg.cb = cb;
486 		arg.w.stop = 0;
487 		arg.w.skip = cb->args[1] - 1;
488 		arg.w.count = 0;
489 		tp->ops->walk(tp, &arg.w);
490 		cb->args[1] = arg.w.count + 1;
491 		if (arg.w.stop)
492 			break;
493 	}
494 
495 	cb->args[0] = t;
496 
497 errout:
498 	if (cl)
499 		cops->put(q, cl);
500 out:
501 	return skb->len;
502 }
503 
504 void tcf_exts_destroy(struct tcf_exts *exts)
505 {
506 #ifdef CONFIG_NET_CLS_ACT
507 	tcf_action_destroy(&exts->actions, TCA_ACT_UNBIND);
508 	INIT_LIST_HEAD(&exts->actions);
509 #endif
510 }
511 EXPORT_SYMBOL(tcf_exts_destroy);
512 
513 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
514 		  struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr)
515 {
516 #ifdef CONFIG_NET_CLS_ACT
517 	{
518 		struct tc_action *act;
519 
520 		INIT_LIST_HEAD(&exts->actions);
521 		if (exts->police && tb[exts->police]) {
522 			act = tcf_action_init_1(net, tb[exts->police], rate_tlv,
523 						"police", ovr,
524 						TCA_ACT_BIND);
525 			if (IS_ERR(act))
526 				return PTR_ERR(act);
527 
528 			act->type = exts->type = TCA_OLD_COMPAT;
529 			list_add(&act->list, &exts->actions);
530 		} else if (exts->action && tb[exts->action]) {
531 			int err;
532 			err = tcf_action_init(net, tb[exts->action], rate_tlv,
533 					      NULL, ovr,
534 					      TCA_ACT_BIND, &exts->actions);
535 			if (err)
536 				return err;
537 		}
538 	}
539 #else
540 	if ((exts->action && tb[exts->action]) ||
541 	    (exts->police && tb[exts->police]))
542 		return -EOPNOTSUPP;
543 #endif
544 
545 	return 0;
546 }
547 EXPORT_SYMBOL(tcf_exts_validate);
548 
549 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
550 		     struct tcf_exts *src)
551 {
552 #ifdef CONFIG_NET_CLS_ACT
553 	LIST_HEAD(tmp);
554 	tcf_tree_lock(tp);
555 	list_splice_init(&dst->actions, &tmp);
556 	list_splice(&src->actions, &dst->actions);
557 	dst->type = src->type;
558 	tcf_tree_unlock(tp);
559 	tcf_action_destroy(&tmp, TCA_ACT_UNBIND);
560 #endif
561 }
562 EXPORT_SYMBOL(tcf_exts_change);
563 
564 #define tcf_exts_first_act(ext)					\
565 	list_first_entry_or_null(&(exts)->actions,		\
566 				 struct tc_action, list)
567 
568 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
569 {
570 #ifdef CONFIG_NET_CLS_ACT
571 	struct nlattr *nest;
572 
573 	if (exts->action && !list_empty(&exts->actions)) {
574 		/*
575 		 * again for backward compatible mode - we want
576 		 * to work with both old and new modes of entering
577 		 * tc data even if iproute2  was newer - jhs
578 		 */
579 		if (exts->type != TCA_OLD_COMPAT) {
580 			nest = nla_nest_start(skb, exts->action);
581 			if (nest == NULL)
582 				goto nla_put_failure;
583 			if (tcf_action_dump(skb, &exts->actions, 0, 0) < 0)
584 				goto nla_put_failure;
585 			nla_nest_end(skb, nest);
586 		} else if (exts->police) {
587 			struct tc_action *act = tcf_exts_first_act(exts);
588 			nest = nla_nest_start(skb, exts->police);
589 			if (nest == NULL || !act)
590 				goto nla_put_failure;
591 			if (tcf_action_dump_old(skb, act, 0, 0) < 0)
592 				goto nla_put_failure;
593 			nla_nest_end(skb, nest);
594 		}
595 	}
596 	return 0;
597 
598 nla_put_failure:
599 	nla_nest_cancel(skb, nest);
600 	return -1;
601 #else
602 	return 0;
603 #endif
604 }
605 EXPORT_SYMBOL(tcf_exts_dump);
606 
607 
608 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
609 {
610 #ifdef CONFIG_NET_CLS_ACT
611 	struct tc_action *a = tcf_exts_first_act(exts);
612 	if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
613 		return -1;
614 #endif
615 	return 0;
616 }
617 EXPORT_SYMBOL(tcf_exts_dump_stats);
618 
619 static int __init tc_filter_init(void)
620 {
621 	rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL);
622 	rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL);
623 	rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
624 		      tc_dump_tfilter, NULL);
625 
626 	return 0;
627 }
628 
629 subsys_initcall(tc_filter_init);
630