Linux Audio

Check our new training course

Loading...
v3.1
  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/netlink.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
 36static struct tcf_proto_ops *tcf_proto_base __read_mostly;
 37
 38/* Protects list of registered TC modules. It is pure SMP lock. */
 39static DEFINE_RWLOCK(cls_mod_lock);
 40
 41/* Find classifier type by string name */
 42
 43static const struct tcf_proto_ops *tcf_proto_lookup_ops(struct nlattr *kind)
 44{
 45	const struct tcf_proto_ops *t = NULL;
 46
 47	if (kind) {
 48		read_lock(&cls_mod_lock);
 49		for (t = tcf_proto_base; t; t = t->next) {
 50			if (nla_strcmp(kind, t->kind) == 0) {
 51				if (!try_module_get(t->owner))
 52					t = NULL;
 53				break;
 54			}
 55		}
 56		read_unlock(&cls_mod_lock);
 57	}
 58	return t;
 59}
 60
 61/* Register(unregister) new classifier type */
 62
 63int register_tcf_proto_ops(struct tcf_proto_ops *ops)
 64{
 65	struct tcf_proto_ops *t, **tp;
 66	int rc = -EEXIST;
 67
 68	write_lock(&cls_mod_lock);
 69	for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
 70		if (!strcmp(ops->kind, t->kind))
 71			goto out;
 72
 73	ops->next = NULL;
 74	*tp = ops;
 75	rc = 0;
 76out:
 77	write_unlock(&cls_mod_lock);
 78	return rc;
 79}
 80EXPORT_SYMBOL(register_tcf_proto_ops);
 81
 82int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
 83{
 84	struct tcf_proto_ops *t, **tp;
 85	int rc = -ENOENT;
 86
 87	write_lock(&cls_mod_lock);
 88	for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
 89		if (t == ops)
 
 
 90			break;
 91
 92	if (!t)
 93		goto out;
 94	*tp = t->next;
 95	rc = 0;
 96out:
 97	write_unlock(&cls_mod_lock);
 98	return rc;
 99}
100EXPORT_SYMBOL(unregister_tcf_proto_ops);
101
102static int tfilter_notify(struct net *net, struct sk_buff *oskb,
103			  struct nlmsghdr *n, struct tcf_proto *tp,
104			  unsigned long fh, int event);
105
106
107/* Select new prio value from the range, managed by kernel. */
108
109static inline u32 tcf_auto_prio(struct tcf_proto *tp)
110{
111	u32 first = TC_H_MAKE(0xC0000000U, 0U);
112
113	if (tp)
114		first = tp->prio - 1;
115
116	return first;
117}
118
119/* Add/change/delete/get a filter node */
120
121static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
122{
123	struct net *net = sock_net(skb->sk);
124	struct nlattr *tca[TCA_MAX + 1];
125	spinlock_t *root_lock;
126	struct tcmsg *t;
127	u32 protocol;
128	u32 prio;
129	u32 nprio;
130	u32 parent;
131	struct net_device *dev;
132	struct Qdisc  *q;
133	struct tcf_proto **back, **chain;
134	struct tcf_proto *tp;
135	const struct tcf_proto_ops *tp_ops;
136	const struct Qdisc_class_ops *cops;
137	unsigned long cl;
138	unsigned long fh;
139	int err;
140	int tp_created = 0;
141
 
 
 
142replay:
143	t = NLMSG_DATA(n);
 
 
 
 
144	protocol = TC_H_MIN(t->tcm_info);
145	prio = TC_H_MAJ(t->tcm_info);
146	nprio = prio;
147	parent = t->tcm_parent;
148	cl = 0;
149
150	if (prio == 0) {
151		/* If no priority is given, user wants we allocated it. */
152		if (n->nlmsg_type != RTM_NEWTFILTER ||
153		    !(n->nlmsg_flags & NLM_F_CREATE))
154			return -ENOENT;
155		prio = TC_H_MAKE(0x80000000U, 0U);
156	}
157
158	/* Find head of filter chain. */
159
160	/* Find link */
161	dev = __dev_get_by_index(net, t->tcm_ifindex);
162	if (dev == NULL)
163		return -ENODEV;
164
165	err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
166	if (err < 0)
167		return err;
168
169	/* Find qdisc */
170	if (!parent) {
171		q = dev->qdisc;
172		parent = q->handle;
173	} else {
174		q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
175		if (q == NULL)
176			return -EINVAL;
177	}
178
179	/* Is it classful? */
180	cops = q->ops->cl_ops;
181	if (!cops)
182		return -EINVAL;
183
184	if (cops->tcf_chain == NULL)
185		return -EOPNOTSUPP;
186
187	/* Do we search for filter, attached to class? */
188	if (TC_H_MIN(parent)) {
189		cl = cops->get(q, parent);
190		if (cl == 0)
191			return -ENOENT;
192	}
193
194	/* And the last stroke */
195	chain = cops->tcf_chain(q, cl);
196	err = -EINVAL;
197	if (chain == NULL)
198		goto errout;
199
200	/* Check the chain for existence of proto-tcf with this priority */
201	for (back = chain; (tp = *back) != NULL; back = &tp->next) {
202		if (tp->prio >= prio) {
203			if (tp->prio == prio) {
204				if (!nprio ||
205				    (tp->protocol != protocol && protocol))
206					goto errout;
207			} else
208				tp = NULL;
209			break;
210		}
211	}
212
213	root_lock = qdisc_root_sleeping_lock(q);
214
215	if (tp == NULL) {
216		/* Proto-tcf does not exist, create new one */
217
218		if (tca[TCA_KIND] == NULL || !protocol)
219			goto errout;
220
221		err = -ENOENT;
222		if (n->nlmsg_type != RTM_NEWTFILTER ||
223		    !(n->nlmsg_flags & NLM_F_CREATE))
224			goto errout;
225
226
227		/* Create new proto tcf */
228
229		err = -ENOBUFS;
230		tp = kzalloc(sizeof(*tp), GFP_KERNEL);
231		if (tp == NULL)
232			goto errout;
233		err = -ENOENT;
234		tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
235		if (tp_ops == NULL) {
236#ifdef CONFIG_MODULES
237			struct nlattr *kind = tca[TCA_KIND];
238			char name[IFNAMSIZ];
239
240			if (kind != NULL &&
241			    nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
242				rtnl_unlock();
243				request_module("cls_%s", name);
244				rtnl_lock();
245				tp_ops = tcf_proto_lookup_ops(kind);
246				/* We dropped the RTNL semaphore in order to
247				 * perform the module load.  So, even if we
248				 * succeeded in loading the module we have to
249				 * replay the request.  We indicate this using
250				 * -EAGAIN.
251				 */
252				if (tp_ops != NULL) {
253					module_put(tp_ops->owner);
254					err = -EAGAIN;
255				}
256			}
257#endif
258			kfree(tp);
259			goto errout;
260		}
261		tp->ops = tp_ops;
262		tp->protocol = protocol;
263		tp->prio = nprio ? : TC_H_MAJ(tcf_auto_prio(*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			spin_lock_bh(root_lock);
285			*back = tp->next;
286			spin_unlock_bh(root_lock);
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(tp, cl, t->tcm_handle, tca, &fh);
323	if (err == 0) {
324		if (tp_created) {
325			spin_lock_bh(root_lock);
326			tp->next = *back;
327			*back = tp;
328			spin_unlock_bh(root_lock);
329		}
330		tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
331	} else {
332		if (tp_created)
333			tcf_destroy(tp);
334	}
335
336errout:
337	if (cl)
338		cops->put(q, cl);
339	if (err == -EAGAIN)
340		/* Replay the request. */
341		goto replay;
342	return err;
343}
344
345static int tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp,
346			 unsigned long fh, u32 pid, u32 seq, u16 flags, int event)
347{
348	struct tcmsg *tcm;
349	struct nlmsghdr  *nlh;
350	unsigned char *b = skb_tail_pointer(skb);
351
352	nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
353	tcm = NLMSG_DATA(nlh);
 
 
354	tcm->tcm_family = AF_UNSPEC;
355	tcm->tcm__pad1 = 0;
356	tcm->tcm__pad2 = 0;
357	tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
358	tcm->tcm_parent = tp->classid;
359	tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
360	NLA_PUT_STRING(skb, TCA_KIND, tp->ops->kind);
 
361	tcm->tcm_handle = fh;
362	if (RTM_DELTFILTER != event) {
363		tcm->tcm_handle = 0;
364		if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
365			goto nla_put_failure;
366	}
367	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
368	return skb->len;
369
370nlmsg_failure:
371nla_put_failure:
372	nlmsg_trim(skb, b);
373	return -1;
374}
375
376static int tfilter_notify(struct net *net, struct sk_buff *oskb,
377			  struct nlmsghdr *n, struct tcf_proto *tp,
378			  unsigned long fh, int event)
379{
380	struct sk_buff *skb;
381	u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
382
383	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
384	if (!skb)
385		return -ENOBUFS;
386
387	if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) {
388		kfree_skb(skb);
389		return -EINVAL;
390	}
391
392	return rtnetlink_send(skb, net, pid, RTNLGRP_TC,
393			      n->nlmsg_flags & NLM_F_ECHO);
394}
395
396struct tcf_dump_args {
397	struct tcf_walker w;
398	struct sk_buff *skb;
399	struct netlink_callback *cb;
400};
401
402static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
403			 struct tcf_walker *arg)
404{
405	struct tcf_dump_args *a = (void *)arg;
 
406
407	return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid,
408			     a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
409}
410
411/* called with RTNL */
412static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
413{
414	struct net *net = sock_net(skb->sk);
415	int t;
416	int s_t;
417	struct net_device *dev;
418	struct Qdisc *q;
419	struct tcf_proto *tp, **chain;
420	struct tcmsg *tcm = (struct tcmsg *)NLMSG_DATA(cb->nlh);
421	unsigned long cl = 0;
422	const struct Qdisc_class_ops *cops;
423	struct tcf_dump_args arg;
424
425	if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
426		return skb->len;
427	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
428	if (!dev)
429		return skb->len;
430
431	if (!tcm->tcm_parent)
432		q = dev->qdisc;
433	else
434		q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
435	if (!q)
436		goto out;
437	cops = q->ops->cl_ops;
438	if (!cops)
439		goto errout;
440	if (cops->tcf_chain == NULL)
441		goto errout;
442	if (TC_H_MIN(tcm->tcm_parent)) {
443		cl = cops->get(q, tcm->tcm_parent);
444		if (cl == 0)
445			goto errout;
446	}
447	chain = cops->tcf_chain(q, cl);
448	if (chain == NULL)
449		goto errout;
450
451	s_t = cb->args[0];
452
453	for (tp = *chain, t = 0; tp; tp = tp->next, t++) {
454		if (t < s_t)
455			continue;
456		if (TC_H_MAJ(tcm->tcm_info) &&
457		    TC_H_MAJ(tcm->tcm_info) != tp->prio)
458			continue;
459		if (TC_H_MIN(tcm->tcm_info) &&
460		    TC_H_MIN(tcm->tcm_info) != tp->protocol)
461			continue;
462		if (t > s_t)
463			memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
464		if (cb->args[1] == 0) {
465			if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid,
466					  cb->nlh->nlmsg_seq, NLM_F_MULTI,
467					  RTM_NEWTFILTER) <= 0)
468				break;
469
470			cb->args[1] = 1;
471		}
472		if (tp->ops->walk == NULL)
473			continue;
474		arg.w.fn = tcf_node_dump;
475		arg.skb = skb;
476		arg.cb = cb;
477		arg.w.stop = 0;
478		arg.w.skip = cb->args[1] - 1;
479		arg.w.count = 0;
480		tp->ops->walk(tp, &arg.w);
481		cb->args[1] = arg.w.count + 1;
482		if (arg.w.stop)
483			break;
484	}
485
486	cb->args[0] = t;
487
488errout:
489	if (cl)
490		cops->put(q, cl);
491out:
492	return skb->len;
493}
494
495void tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
496{
497#ifdef CONFIG_NET_CLS_ACT
498	if (exts->action) {
499		tcf_action_destroy(exts->action, TCA_ACT_UNBIND);
500		exts->action = NULL;
501	}
502#endif
503}
504EXPORT_SYMBOL(tcf_exts_destroy);
505
506int tcf_exts_validate(struct tcf_proto *tp, struct nlattr **tb,
507		  struct nlattr *rate_tlv, struct tcf_exts *exts,
508		  const struct tcf_ext_map *map)
509{
510	memset(exts, 0, sizeof(*exts));
511
512#ifdef CONFIG_NET_CLS_ACT
513	{
514		struct tc_action *act;
515
516		if (map->police && tb[map->police]) {
517			act = tcf_action_init_1(tb[map->police], rate_tlv,
 
518						"police", TCA_ACT_NOREPLACE,
519						TCA_ACT_BIND);
520			if (IS_ERR(act))
521				return PTR_ERR(act);
522
523			act->type = TCA_OLD_COMPAT;
524			exts->action = act;
525		} else if (map->action && tb[map->action]) {
526			act = tcf_action_init(tb[map->action], rate_tlv, NULL,
527					      TCA_ACT_NOREPLACE, TCA_ACT_BIND);
528			if (IS_ERR(act))
529				return PTR_ERR(act);
530
531			exts->action = act;
532		}
533	}
534#else
535	if ((map->action && tb[map->action]) ||
536	    (map->police && tb[map->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	if (src->action) {
549		struct tc_action *act;
550		tcf_tree_lock(tp);
551		act = dst->action;
552		dst->action = src->action;
553		tcf_tree_unlock(tp);
554		if (act)
555			tcf_action_destroy(act, TCA_ACT_UNBIND);
556	}
557#endif
558}
559EXPORT_SYMBOL(tcf_exts_change);
560
561int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts,
562		  const struct tcf_ext_map *map)
 
 
563{
564#ifdef CONFIG_NET_CLS_ACT
565	if (map->action && exts->action) {
566		/*
567		 * again for backward compatible mode - we want
568		 * to work with both old and new modes of entering
569		 * tc data even if iproute2  was newer - jhs
570		 */
571		struct nlattr *nest;
572
573		if (exts->action->type != TCA_OLD_COMPAT) {
574			nest = nla_nest_start(skb, map->action);
575			if (nest == NULL)
576				goto nla_put_failure;
577			if (tcf_action_dump(skb, exts->action, 0, 0) < 0)
578				goto nla_put_failure;
579			nla_nest_end(skb, nest);
580		} else if (map->police) {
581			nest = nla_nest_start(skb, map->police);
582			if (nest == NULL)
 
583				goto nla_put_failure;
584			if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0)
585				goto nla_put_failure;
586			nla_nest_end(skb, nest);
587		}
588	}
589#endif
590	return 0;
591nla_put_failure: __attribute__ ((unused))
592	return -1;
593}
594EXPORT_SYMBOL(tcf_exts_dump);
595
596
597int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts,
598			const struct tcf_ext_map *map)
599{
600#ifdef CONFIG_NET_CLS_ACT
601	if (exts->action)
602		if (tcf_action_copy_stats(skb, exts->action, 1) < 0)
603			goto nla_put_failure;
604#endif
605	return 0;
606nla_put_failure: __attribute__ ((unused))
607	return -1;
608}
609EXPORT_SYMBOL(tcf_exts_dump_stats);
610
611static int __init tc_filter_init(void)
612{
613	rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL);
614	rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL);
615	rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
616		      tc_dump_tfilter, NULL);
617
618	return 0;
619}
620
621subsys_initcall(tc_filter_init);
v3.15
  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	spinlock_t *root_lock;
121	struct tcmsg *t;
122	u32 protocol;
123	u32 prio;
124	u32 nprio;
125	u32 parent;
126	struct net_device *dev;
127	struct Qdisc  *q;
128	struct tcf_proto **back, **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) && !netlink_capable(skb, CAP_NET_ADMIN))
138		return -EPERM;
139
140replay:
141	err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
142	if (err < 0)
143		return err;
144
145	t = nlmsg_data(n);
146	protocol = TC_H_MIN(t->tcm_info);
147	prio = TC_H_MAJ(t->tcm_info);
148	nprio = prio;
149	parent = t->tcm_parent;
150	cl = 0;
151
152	if (prio == 0) {
153		/* If no priority is given, user wants we allocated it. */
154		if (n->nlmsg_type != RTM_NEWTFILTER ||
155		    !(n->nlmsg_flags & NLM_F_CREATE))
156			return -ENOENT;
157		prio = TC_H_MAKE(0x80000000U, 0U);
158	}
159
160	/* Find head of filter chain. */
161
162	/* Find link */
163	dev = __dev_get_by_index(net, t->tcm_ifindex);
164	if (dev == NULL)
165		return -ENODEV;
166
 
 
 
 
167	/* Find qdisc */
168	if (!parent) {
169		q = dev->qdisc;
170		parent = q->handle;
171	} else {
172		q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
173		if (q == NULL)
174			return -EINVAL;
175	}
176
177	/* Is it classful? */
178	cops = q->ops->cl_ops;
179	if (!cops)
180		return -EINVAL;
181
182	if (cops->tcf_chain == NULL)
183		return -EOPNOTSUPP;
184
185	/* Do we search for filter, attached to class? */
186	if (TC_H_MIN(parent)) {
187		cl = cops->get(q, parent);
188		if (cl == 0)
189			return -ENOENT;
190	}
191
192	/* And the last stroke */
193	chain = cops->tcf_chain(q, cl);
194	err = -EINVAL;
195	if (chain == NULL)
196		goto errout;
197
198	/* Check the chain for existence of proto-tcf with this priority */
199	for (back = chain; (tp = *back) != NULL; back = &tp->next) {
200		if (tp->prio >= prio) {
201			if (tp->prio == prio) {
202				if (!nprio ||
203				    (tp->protocol != protocol && protocol))
204					goto errout;
205			} else
206				tp = NULL;
207			break;
208		}
209	}
210
211	root_lock = qdisc_root_sleeping_lock(q);
212
213	if (tp == NULL) {
214		/* Proto-tcf does not exist, create new one */
215
216		if (tca[TCA_KIND] == NULL || !protocol)
217			goto errout;
218
219		err = -ENOENT;
220		if (n->nlmsg_type != RTM_NEWTFILTER ||
221		    !(n->nlmsg_flags & NLM_F_CREATE))
222			goto errout;
223
224
225		/* Create new proto tcf */
226
227		err = -ENOBUFS;
228		tp = kzalloc(sizeof(*tp), GFP_KERNEL);
229		if (tp == NULL)
230			goto errout;
231		err = -ENOENT;
232		tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
233		if (tp_ops == NULL) {
234#ifdef CONFIG_MODULES
235			struct nlattr *kind = tca[TCA_KIND];
236			char name[IFNAMSIZ];
237
238			if (kind != NULL &&
239			    nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
240				rtnl_unlock();
241				request_module("cls_%s", name);
242				rtnl_lock();
243				tp_ops = tcf_proto_lookup_ops(kind);
244				/* We dropped the RTNL semaphore in order to
245				 * perform the module load.  So, even if we
246				 * succeeded in loading the module we have to
247				 * replay the request.  We indicate this using
248				 * -EAGAIN.
249				 */
250				if (tp_ops != NULL) {
251					module_put(tp_ops->owner);
252					err = -EAGAIN;
253				}
254			}
255#endif
256			kfree(tp);
257			goto errout;
258		}
259		tp->ops = tp_ops;
260		tp->protocol = protocol;
261		tp->prio = nprio ? : TC_H_MAJ(tcf_auto_prio(*back));
262		tp->q = q;
263		tp->classify = tp_ops->classify;
264		tp->classid = parent;
265
266		err = tp_ops->init(tp);
267		if (err != 0) {
268			module_put(tp_ops->owner);
269			kfree(tp);
270			goto errout;
271		}
272
273		tp_created = 1;
274
275	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
276		goto errout;
277
278	fh = tp->ops->get(tp, t->tcm_handle);
279
280	if (fh == 0) {
281		if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
282			spin_lock_bh(root_lock);
283			*back = tp->next;
284			spin_unlock_bh(root_lock);
285
286			tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
287			tcf_destroy(tp);
288			err = 0;
289			goto errout;
290		}
291
292		err = -ENOENT;
293		if (n->nlmsg_type != RTM_NEWTFILTER ||
294		    !(n->nlmsg_flags & NLM_F_CREATE))
295			goto errout;
296	} else {
297		switch (n->nlmsg_type) {
298		case RTM_NEWTFILTER:
299			err = -EEXIST;
300			if (n->nlmsg_flags & NLM_F_EXCL) {
301				if (tp_created)
302					tcf_destroy(tp);
303				goto errout;
304			}
305			break;
306		case RTM_DELTFILTER:
307			err = tp->ops->delete(tp, fh);
308			if (err == 0)
309				tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
310			goto errout;
311		case RTM_GETTFILTER:
312			err = tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
313			goto errout;
314		default:
315			err = -EINVAL;
316			goto errout;
317		}
318	}
319
320	err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh);
321	if (err == 0) {
322		if (tp_created) {
323			spin_lock_bh(root_lock);
324			tp->next = *back;
325			*back = tp;
326			spin_unlock_bh(root_lock);
327		}
328		tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
329	} else {
330		if (tp_created)
331			tcf_destroy(tp);
332	}
333
334errout:
335	if (cl)
336		cops->put(q, cl);
337	if (err == -EAGAIN)
338		/* Replay the request. */
339		goto replay;
340	return err;
341}
342
343static int tcf_fill_node(struct net *net, struct sk_buff *skb, struct tcf_proto *tp,
344			 unsigned long fh, u32 portid, u32 seq, u16 flags, int event)
345{
346	struct tcmsg *tcm;
347	struct nlmsghdr  *nlh;
348	unsigned char *b = skb_tail_pointer(skb);
349
350	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
351	if (!nlh)
352		goto out_nlmsg_trim;
353	tcm = nlmsg_data(nlh);
354	tcm->tcm_family = AF_UNSPEC;
355	tcm->tcm__pad1 = 0;
356	tcm->tcm__pad2 = 0;
357	tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
358	tcm->tcm_parent = tp->classid;
359	tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
360	if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
361		goto nla_put_failure;
362	tcm->tcm_handle = fh;
363	if (RTM_DELTFILTER != event) {
364		tcm->tcm_handle = 0;
365		if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
366			goto nla_put_failure;
367	}
368	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
369	return skb->len;
370
371out_nlmsg_trim:
372nla_put_failure:
373	nlmsg_trim(skb, b);
374	return -1;
375}
376
377static int tfilter_notify(struct net *net, struct sk_buff *oskb,
378			  struct nlmsghdr *n, struct tcf_proto *tp,
379			  unsigned long fh, int event)
380{
381	struct sk_buff *skb;
382	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
383
384	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
385	if (!skb)
386		return -ENOBUFS;
387
388	if (tcf_fill_node(net, skb, tp, fh, portid, n->nlmsg_seq, 0, event) <= 0) {
389		kfree_skb(skb);
390		return -EINVAL;
391	}
392
393	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
394			      n->nlmsg_flags & NLM_F_ECHO);
395}
396
397struct tcf_dump_args {
398	struct tcf_walker w;
399	struct sk_buff *skb;
400	struct netlink_callback *cb;
401};
402
403static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
404			 struct tcf_walker *arg)
405{
406	struct tcf_dump_args *a = (void *)arg;
407	struct net *net = sock_net(a->skb->sk);
408
409	return tcf_fill_node(net, a->skb, tp, n, NETLINK_CB(a->cb->skb).portid,
410			     a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
411}
412
413/* called with RTNL */
414static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
415{
416	struct net *net = sock_net(skb->sk);
417	int t;
418	int s_t;
419	struct net_device *dev;
420	struct Qdisc *q;
421	struct tcf_proto *tp, **chain;
422	struct tcmsg *tcm = nlmsg_data(cb->nlh);
423	unsigned long cl = 0;
424	const struct Qdisc_class_ops *cops;
425	struct tcf_dump_args arg;
426
427	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
428		return skb->len;
429	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
430	if (!dev)
431		return skb->len;
432
433	if (!tcm->tcm_parent)
434		q = dev->qdisc;
435	else
436		q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
437	if (!q)
438		goto out;
439	cops = q->ops->cl_ops;
440	if (!cops)
441		goto errout;
442	if (cops->tcf_chain == NULL)
443		goto errout;
444	if (TC_H_MIN(tcm->tcm_parent)) {
445		cl = cops->get(q, tcm->tcm_parent);
446		if (cl == 0)
447			goto errout;
448	}
449	chain = cops->tcf_chain(q, cl);
450	if (chain == NULL)
451		goto errout;
452
453	s_t = cb->args[0];
454
455	for (tp = *chain, t = 0; tp; tp = tp->next, t++) {
456		if (t < s_t)
457			continue;
458		if (TC_H_MAJ(tcm->tcm_info) &&
459		    TC_H_MAJ(tcm->tcm_info) != tp->prio)
460			continue;
461		if (TC_H_MIN(tcm->tcm_info) &&
462		    TC_H_MIN(tcm->tcm_info) != tp->protocol)
463			continue;
464		if (t > s_t)
465			memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
466		if (cb->args[1] == 0) {
467			if (tcf_fill_node(net, skb, tp, 0, NETLINK_CB(cb->skb).portid,
468					  cb->nlh->nlmsg_seq, NLM_F_MULTI,
469					  RTM_NEWTFILTER) <= 0)
470				break;
471
472			cb->args[1] = 1;
473		}
474		if (tp->ops->walk == NULL)
475			continue;
476		arg.w.fn = tcf_node_dump;
477		arg.skb = skb;
478		arg.cb = cb;
479		arg.w.stop = 0;
480		arg.w.skip = cb->args[1] - 1;
481		arg.w.count = 0;
482		tp->ops->walk(tp, &arg.w);
483		cb->args[1] = arg.w.count + 1;
484		if (arg.w.stop)
485			break;
486	}
487
488	cb->args[0] = t;
489
490errout:
491	if (cl)
492		cops->put(q, cl);
493out:
494	return skb->len;
495}
496
497void tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
498{
499#ifdef CONFIG_NET_CLS_ACT
500	tcf_action_destroy(&exts->actions, TCA_ACT_UNBIND);
501	INIT_LIST_HEAD(&exts->actions);
 
 
502#endif
503}
504EXPORT_SYMBOL(tcf_exts_destroy);
505
506int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
507		  struct nlattr *rate_tlv, struct tcf_exts *exts)
 
508{
 
 
509#ifdef CONFIG_NET_CLS_ACT
510	{
511		struct tc_action *act;
512
513		INIT_LIST_HEAD(&exts->actions);
514		if (exts->police && tb[exts->police]) {
515			act = tcf_action_init_1(net, tb[exts->police], rate_tlv,
516						"police", TCA_ACT_NOREPLACE,
517						TCA_ACT_BIND);
518			if (IS_ERR(act))
519				return PTR_ERR(act);
520
521			act->type = exts->type = TCA_OLD_COMPAT;
522			list_add(&act->list, &exts->actions);
523		} else if (exts->action && tb[exts->action]) {
524			int err;
525			err = tcf_action_init(net, tb[exts->action], rate_tlv,
526					      NULL, TCA_ACT_NOREPLACE,
527					      TCA_ACT_BIND, &exts->actions);
528			if (err)
529				return err;
530		}
531	}
532#else
533	if ((exts->action && tb[exts->action]) ||
534	    (exts->police && tb[exts->police]))
535		return -EOPNOTSUPP;
536#endif
537
538	return 0;
539}
540EXPORT_SYMBOL(tcf_exts_validate);
541
542void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
543		     struct tcf_exts *src)
544{
545#ifdef CONFIG_NET_CLS_ACT
546	if (!list_empty(&src->actions)) {
547		LIST_HEAD(tmp);
548		tcf_tree_lock(tp);
549		list_splice_init(&dst->actions, &tmp);
550		list_splice(&src->actions, &dst->actions);
551		tcf_tree_unlock(tp);
552		tcf_action_destroy(&tmp, TCA_ACT_UNBIND);
 
553	}
554#endif
555}
556EXPORT_SYMBOL(tcf_exts_change);
557
558#define tcf_exts_first_act(ext) \
559		list_first_entry(&(exts)->actions, struct tc_action, list)
560
561int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
562{
563#ifdef CONFIG_NET_CLS_ACT
564	if (exts->action && !list_empty(&exts->actions)) {
565		/*
566		 * again for backward compatible mode - we want
567		 * to work with both old and new modes of entering
568		 * tc data even if iproute2  was newer - jhs
569		 */
570		struct nlattr *nest;
571		if (exts->type != TCA_OLD_COMPAT) {
572			nest = nla_nest_start(skb, exts->action);
 
573			if (nest == NULL)
574				goto nla_put_failure;
575			if (tcf_action_dump(skb, &exts->actions, 0, 0) < 0)
576				goto nla_put_failure;
577			nla_nest_end(skb, nest);
578		} else if (exts->police) {
579			struct tc_action *act = tcf_exts_first_act(exts);
580			nest = nla_nest_start(skb, exts->police);
581			if (nest == NULL || !act)
582				goto nla_put_failure;
583			if (tcf_action_dump_old(skb, act, 0, 0) < 0)
584				goto nla_put_failure;
585			nla_nest_end(skb, nest);
586		}
587	}
588#endif
589	return 0;
590nla_put_failure: __attribute__ ((unused))
591	return -1;
592}
593EXPORT_SYMBOL(tcf_exts_dump);
594
595
596int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
 
597{
598#ifdef CONFIG_NET_CLS_ACT
599	struct tc_action *a = tcf_exts_first_act(exts);
600	if (tcf_action_copy_stats(skb, a, 1) < 0)
601		return -1;
602#endif
603	return 0;
 
 
604}
605EXPORT_SYMBOL(tcf_exts_dump_stats);
606
607static int __init tc_filter_init(void)
608{
609	rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL);
610	rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL);
611	rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
612		      tc_dump_tfilter, NULL);
613
614	return 0;
615}
616
617subsys_initcall(tc_filter_init);