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 */
34static LIST_HEAD(tcf_proto_base);
35
36/* Protects list of registered TC modules. It is pure SMP lock. */
37static DEFINE_RWLOCK(cls_mod_lock);
38
39/* Find classifier type by string name */
40
41static 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
61int 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;
73out:
74	write_unlock(&cls_mod_lock);
75	return rc;
76}
77EXPORT_SYMBOL(register_tcf_proto_ops);
78
79int 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}
95EXPORT_SYMBOL(unregister_tcf_proto_ops);
96
97static 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
104static 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
116static 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
141replay:
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);
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);
305				goto errout;
306			}
307			break;
308		case RTM_DELTFILTER:
309			err = tp->ops->delete(tp, fh);
310			if (err == 0)
311				tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
312			goto errout;
313		case RTM_GETTFILTER:
314			err = tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
315			goto errout;
316		default:
317			err = -EINVAL;
318			goto errout;
319		}
320	}
321
322	err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
323			      n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE);
324	if (err == 0) {
325		if (tp_created) {
326			RCU_INIT_POINTER(tp->next, rtnl_dereference(*back));
327			rcu_assign_pointer(*back, tp);
328		}
329		tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
330	} else {
331		if (tp_created)
332			tcf_destroy(tp);
333	}
334
335errout:
336	if (cl)
337		cops->put(q, cl);
338	if (err == -EAGAIN)
339		/* Replay the request. */
340		goto replay;
341	return err;
342}
343
344static int tcf_fill_node(struct net *net, struct sk_buff *skb, struct tcf_proto *tp,
345			 unsigned long fh, u32 portid, u32 seq, u16 flags, int event)
346{
347	struct tcmsg *tcm;
348	struct nlmsghdr  *nlh;
349	unsigned char *b = skb_tail_pointer(skb);
350
351	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
352	if (!nlh)
353		goto out_nlmsg_trim;
354	tcm = nlmsg_data(nlh);
355	tcm->tcm_family = AF_UNSPEC;
356	tcm->tcm__pad1 = 0;
357	tcm->tcm__pad2 = 0;
358	tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
359	tcm->tcm_parent = tp->classid;
360	tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
361	if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
362		goto nla_put_failure;
363	tcm->tcm_handle = fh;
364	if (RTM_DELTFILTER != event) {
365		tcm->tcm_handle = 0;
366		if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
367			goto nla_put_failure;
368	}
369	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
370	return skb->len;
371
372out_nlmsg_trim:
373nla_put_failure:
374	nlmsg_trim(skb, b);
375	return -1;
376}
377
378static int tfilter_notify(struct net *net, struct sk_buff *oskb,
379			  struct nlmsghdr *n, struct tcf_proto *tp,
380			  unsigned long fh, int event)
381{
382	struct sk_buff *skb;
383	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
384
385	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
386	if (!skb)
387		return -ENOBUFS;
388
389	if (tcf_fill_node(net, skb, tp, fh, portid, n->nlmsg_seq, 0, event) <= 0) {
390		kfree_skb(skb);
391		return -EINVAL;
392	}
393
394	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
395			      n->nlmsg_flags & NLM_F_ECHO);
396}
397
398struct tcf_dump_args {
399	struct tcf_walker w;
400	struct sk_buff *skb;
401	struct netlink_callback *cb;
402};
403
404static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
405			 struct tcf_walker *arg)
406{
407	struct tcf_dump_args *a = (void *)arg;
408	struct net *net = sock_net(a->skb->sk);
409
410	return tcf_fill_node(net, a->skb, tp, n, NETLINK_CB(a->cb->skb).portid,
411			     a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
412}
413
414/* called with RTNL */
415static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
416{
417	struct net *net = sock_net(skb->sk);
418	int t;
419	int s_t;
420	struct net_device *dev;
421	struct Qdisc *q;
422	struct tcf_proto *tp, __rcu **chain;
423	struct tcmsg *tcm = nlmsg_data(cb->nlh);
424	unsigned long cl = 0;
425	const struct Qdisc_class_ops *cops;
426	struct tcf_dump_args arg;
427
428	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
429		return skb->len;
430	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
431	if (!dev)
432		return skb->len;
433
434	if (!tcm->tcm_parent)
435		q = dev->qdisc;
436	else
437		q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
438	if (!q)
439		goto out;
440	cops = q->ops->cl_ops;
441	if (!cops)
442		goto errout;
443	if (cops->tcf_chain == NULL)
444		goto errout;
445	if (TC_H_MIN(tcm->tcm_parent)) {
446		cl = cops->get(q, tcm->tcm_parent);
447		if (cl == 0)
448			goto errout;
449	}
450	chain = cops->tcf_chain(q, cl);
451	if (chain == NULL)
452		goto errout;
453
454	s_t = cb->args[0];
455
456	for (tp = rtnl_dereference(*chain), t = 0;
457	     tp; tp = rtnl_dereference(tp->next), t++) {
458		if (t < s_t)
459			continue;
460		if (TC_H_MAJ(tcm->tcm_info) &&
461		    TC_H_MAJ(tcm->tcm_info) != tp->prio)
462			continue;
463		if (TC_H_MIN(tcm->tcm_info) &&
464		    TC_H_MIN(tcm->tcm_info) != tp->protocol)
465			continue;
466		if (t > s_t)
467			memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
468		if (cb->args[1] == 0) {
469			if (tcf_fill_node(net, skb, tp, 0, NETLINK_CB(cb->skb).portid,
470					  cb->nlh->nlmsg_seq, NLM_F_MULTI,
471					  RTM_NEWTFILTER) <= 0)
472				break;
473
474			cb->args[1] = 1;
475		}
476		if (tp->ops->walk == NULL)
477			continue;
478		arg.w.fn = tcf_node_dump;
479		arg.skb = skb;
480		arg.cb = cb;
481		arg.w.stop = 0;
482		arg.w.skip = cb->args[1] - 1;
483		arg.w.count = 0;
484		tp->ops->walk(tp, &arg.w);
485		cb->args[1] = arg.w.count + 1;
486		if (arg.w.stop)
487			break;
488	}
489
490	cb->args[0] = t;
491
492errout:
493	if (cl)
494		cops->put(q, cl);
495out:
496	return skb->len;
497}
498
499void tcf_exts_destroy(struct tcf_exts *exts)
500{
501#ifdef CONFIG_NET_CLS_ACT
502	tcf_action_destroy(&exts->actions, TCA_ACT_UNBIND);
503	INIT_LIST_HEAD(&exts->actions);
504#endif
505}
506EXPORT_SYMBOL(tcf_exts_destroy);
507
508int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
509		  struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr)
510{
511#ifdef CONFIG_NET_CLS_ACT
512	{
513		struct tc_action *act;
514
515		INIT_LIST_HEAD(&exts->actions);
516		if (exts->police && tb[exts->police]) {
517			act = tcf_action_init_1(net, tb[exts->police], rate_tlv,
518						"police", ovr,
519						TCA_ACT_BIND);
520			if (IS_ERR(act))
521				return PTR_ERR(act);
522
523			act->type = exts->type = TCA_OLD_COMPAT;
524			list_add(&act->list, &exts->actions);
525		} else if (exts->action && tb[exts->action]) {
526			int err;
527			err = tcf_action_init(net, tb[exts->action], rate_tlv,
528					      NULL, ovr,
529					      TCA_ACT_BIND, &exts->actions);
530			if (err)
531				return err;
532		}
533	}
534#else
535	if ((exts->action && tb[exts->action]) ||
536	    (exts->police && tb[exts->police]))
537		return -EOPNOTSUPP;
538#endif
539
540	return 0;
541}
542EXPORT_SYMBOL(tcf_exts_validate);
543
544void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
545		     struct tcf_exts *src)
546{
547#ifdef CONFIG_NET_CLS_ACT
548	LIST_HEAD(tmp);
549	tcf_tree_lock(tp);
550	list_splice_init(&dst->actions, &tmp);
551	list_splice(&src->actions, &dst->actions);
552	dst->type = src->type;
553	tcf_tree_unlock(tp);
554	tcf_action_destroy(&tmp, TCA_ACT_UNBIND);
555#endif
556}
557EXPORT_SYMBOL(tcf_exts_change);
558
559#define tcf_exts_first_act(ext) \
560		list_first_entry(&(exts)->actions, struct tc_action, list)
561
562int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
563{
564#ifdef CONFIG_NET_CLS_ACT
565	struct nlattr *nest;
566
567	if (exts->action && !list_empty(&exts->actions)) {
568		/*
569		 * again for backward compatible mode - we want
570		 * to work with both old and new modes of entering
571		 * tc data even if iproute2  was newer - jhs
572		 */
573		if (exts->type != TCA_OLD_COMPAT) {
574			nest = nla_nest_start(skb, exts->action);
575			if (nest == NULL)
576				goto nla_put_failure;
577			if (tcf_action_dump(skb, &exts->actions, 0, 0) < 0)
578				goto nla_put_failure;
579			nla_nest_end(skb, nest);
580		} else if (exts->police) {
581			struct tc_action *act = tcf_exts_first_act(exts);
582			nest = nla_nest_start(skb, exts->police);
583			if (nest == NULL || !act)
584				goto nla_put_failure;
585			if (tcf_action_dump_old(skb, act, 0, 0) < 0)
586				goto nla_put_failure;
587			nla_nest_end(skb, nest);
588		}
589	}
590	return 0;
591
592nla_put_failure:
593	nla_nest_cancel(skb, nest);
594	return -1;
595#else
596	return 0;
597#endif
598}
599EXPORT_SYMBOL(tcf_exts_dump);
600
601
602int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
603{
604#ifdef CONFIG_NET_CLS_ACT
605	struct tc_action *a = tcf_exts_first_act(exts);
606	if (tcf_action_copy_stats(skb, a, 1) < 0)
607		return -1;
608#endif
609	return 0;
610}
611EXPORT_SYMBOL(tcf_exts_dump_stats);
612
613static int __init tc_filter_init(void)
614{
615	rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL);
616	rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL);
617	rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
618		      tc_dump_tfilter, NULL);
619
620	return 0;
621}
622
623subsys_initcall(tc_filter_init);
624