[linux.git] / net / sched / cls_api.c

/*
 * net/sched/cls_api.c	Packet classifier API.
 *
 *		This program is free software; you can redistribute it and/or
 *		modify it under the terms of the GNU General Public License
 *		as published by the Free Software Foundation; either version
 *		2 of the License, or (at your option) any later version.
 *
 * Authors:	Alexey Kuznetsov, <[email protected]>
 *
 * Changes:
 *
 * Eduardo J. Blanco <[email protected]> :990222: kmod support
 *
 */

#include <linux/module.h>
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/string.h>
#include <linux/errno.h>
#include <linux/skbuff.h>
#include <linux/init.h>
#include <linux/kmod.h>
#include <linux/netlink.h>
#include <linux/err.h>
#include <linux/slab.h>
#include <net/net_namespace.h>
#include <net/sock.h>
#include <net/netlink.h>
#include <net/pkt_sched.h>
#include <net/pkt_cls.h>

/* The list of all installed classifier types */

static struct tcf_proto_ops *tcf_proto_base __read_mostly;

/* Protects list of registered TC modules. It is pure SMP lock. */
static DEFINE_RWLOCK(cls_mod_lock);

/* Find classifier type by string name */

static const struct tcf_proto_ops *tcf_proto_lookup_ops(struct nlattr *kind)
{
	const struct tcf_proto_ops *t = NULL;

	if (kind) {
		read_lock(&cls_mod_lock);
		for (t = tcf_proto_base; t; t = t->next) {
			if (nla_strcmp(kind, t->kind) == 0) {
				if (!try_module_get(t->owner))
					t = NULL;
				break;
			}
		}
		read_unlock(&cls_mod_lock);
	}
	return t;
}

/* Register(unregister) new classifier type */

int register_tcf_proto_ops(struct tcf_proto_ops *ops)
{
	struct tcf_proto_ops *t, **tp;
	int rc = -EEXIST;

	write_lock(&cls_mod_lock);
	for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
		if (!strcmp(ops->kind, t->kind))
			goto out;

	ops->next = NULL;
	*tp = ops;
	rc = 0;
out:
	write_unlock(&cls_mod_lock);
	return rc;
}
EXPORT_SYMBOL(register_tcf_proto_ops);

int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
{
	struct tcf_proto_ops *t, **tp;
	int rc = -ENOENT;

	write_lock(&cls_mod_lock);
	for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
		if (t == ops)
			break;

	if (!t)
		goto out;
	*tp = t->next;
	rc = 0;
out:
	write_unlock(&cls_mod_lock);
	return rc;
}
EXPORT_SYMBOL(unregister_tcf_proto_ops);

static int tfilter_notify(struct net *net, struct sk_buff *oskb,
			  struct nlmsghdr *n, struct tcf_proto *tp,
			  unsigned long fh, int event);


/* Select new prio value from the range, managed by kernel. */

static inline u32 tcf_auto_prio(struct tcf_proto *tp)
{
	u32 first = TC_H_MAKE(0xC0000000U, 0U);

	if (tp)
		first = tp->prio - 1;

	return first;
}

/* Add/change/delete/get a filter node */

static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
{
	struct net *net = sock_net(skb->sk);
	struct nlattr *tca[TCA_MAX + 1];
	spinlock_t *root_lock;
	struct tcmsg *t;
	u32 protocol;
	u32 prio;
	u32 nprio;
	u32 parent;
	struct net_device *dev;
	struct Qdisc  *q;
	struct tcf_proto **back, **chain;
	struct tcf_proto *tp;
	const struct tcf_proto_ops *tp_ops;
	const struct Qdisc_class_ops *cops;
	unsigned long cl;
	unsigned long fh;
	int err;
	int tp_created = 0;

replay:
	t = nlmsg_data(n);
	protocol = TC_H_MIN(t->tcm_info);
	prio = TC_H_MAJ(t->tcm_info);
	nprio = prio;
	parent = t->tcm_parent;
	cl = 0;

	if (prio == 0) {
		/* If no priority is given, user wants we allocated it. */
		if (n->nlmsg_type != RTM_NEWTFILTER ||
		    !(n->nlmsg_flags & NLM_F_CREATE))
			return -ENOENT;
		prio = TC_H_MAKE(0x80000000U, 0U);
	}

	/* Find head of filter chain. */

	/* Find link */
	dev = __dev_get_by_index(net, t->tcm_ifindex);
	if (dev == NULL)
		return -ENODEV;

	err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
	if (err < 0)
		return err;

	/* Find qdisc */
	if (!parent) {
		q = dev->qdisc;
		parent = q->handle;
	} else {
		q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
		if (q == NULL)
			return -EINVAL;
	}

	/* Is it classful? */
	cops = q->ops->cl_ops;
	if (!cops)
		return -EINVAL;

	if (cops->tcf_chain == NULL)
		return -EOPNOTSUPP;

	/* Do we search for filter, attached to class? */
	if (TC_H_MIN(parent)) {
		cl = cops->get(q, parent);
		if (cl == 0)
			return -ENOENT;
	}

	/* And the last stroke */
	chain = cops->tcf_chain(q, cl);
	err = -EINVAL;
	if (chain == NULL)
		goto errout;

	/* Check the chain for existence of proto-tcf with this priority */
	for (back = chain; (tp = *back) != NULL; back = &tp->next) {
		if (tp->prio >= prio) {
			if (tp->prio == prio) {
				if (!nprio ||
				    (tp->protocol != protocol && protocol))
					goto errout;
			} else
				tp = NULL;
			break;
		}
	}

	root_lock = qdisc_root_sleeping_lock(q);

	if (tp == NULL) {
		/* Proto-tcf does not exist, create new one */

		if (tca[TCA_KIND] == NULL || !protocol)
			goto errout;

		err = -ENOENT;
		if (n->nlmsg_type != RTM_NEWTFILTER ||
		    !(n->nlmsg_flags & NLM_F_CREATE))
			goto errout;


		/* Create new proto tcf */

		err = -ENOBUFS;
		tp = kzalloc(sizeof(*tp), GFP_KERNEL);
		if (tp == NULL)
			goto errout;
		err = -ENOENT;
		tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
		if (tp_ops == NULL) {
#ifdef CONFIG_MODULES
			struct nlattr *kind = tca[TCA_KIND];
			char name[IFNAMSIZ];

			if (kind != NULL &&
			    nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
				rtnl_unlock();
				request_module("cls_%s", name);
				rtnl_lock();
				tp_ops = tcf_proto_lookup_ops(kind);
				/* We dropped the RTNL semaphore in order to
				 * perform the module load.  So, even if we
				 * succeeded in loading the module we have to
				 * replay the request.  We indicate this using
				 * -EAGAIN.
				 */
				if (tp_ops != NULL) {
					module_put(tp_ops->owner);
					err = -EAGAIN;
				}
			}
#endif
			kfree(tp);
			goto errout;
		}
		tp->ops = tp_ops;
		tp->protocol = protocol;
		tp->prio = nprio ? : TC_H_MAJ(tcf_auto_prio(*back));
		tp->q = q;
		tp->classify = tp_ops->classify;
		tp->classid = parent;

		err = tp_ops->init(tp);
		if (err != 0) {
			module_put(tp_ops->owner);
			kfree(tp);
			goto errout;
		}

		tp_created = 1;

	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
		goto errout;

	fh = tp->ops->get(tp, t->tcm_handle);

	if (fh == 0) {
		if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
			spin_lock_bh(root_lock);
			*back = tp->next;
			spin_unlock_bh(root_lock);

			tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
			tcf_destroy(tp);
			err = 0;
			goto errout;
		}

		err = -ENOENT;
		if (n->nlmsg_type != RTM_NEWTFILTER ||
		    !(n->nlmsg_flags & NLM_F_CREATE))
			goto errout;
	} else {
		switch (n->nlmsg_type) {
		case RTM_NEWTFILTER:
			err = -EEXIST;
			if (n->nlmsg_flags & NLM_F_EXCL) {
				if (tp_created)
					tcf_destroy(tp);
				goto errout;
			}
			break;
		case RTM_DELTFILTER:
			err = tp->ops->delete(tp, fh);
			if (err == 0)
				tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
			goto errout;
		case RTM_GETTFILTER:
			err = tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
			goto errout;
		default:
			err = -EINVAL;
			goto errout;
		}
	}

	err = tp->ops->change(tp, cl, t->tcm_handle, tca, &fh);
	if (err == 0) {
		if (tp_created) {
			spin_lock_bh(root_lock);
			tp->next = *back;
			*back = tp;
			spin_unlock_bh(root_lock);
		}
		tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
	} else {
		if (tp_created)
			tcf_destroy(tp);
	}

errout:
	if (cl)
		cops->put(q, cl);
	if (err == -EAGAIN)
		/* Replay the request. */
		goto replay;
	return err;
}

static int tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp,
			 unsigned long fh, u32 pid, u32 seq, u16 flags, int event)
{
	struct tcmsg *tcm;
	struct nlmsghdr  *nlh;
	unsigned char *b = skb_tail_pointer(skb);

	nlh = nlmsg_put(skb, pid, seq, event, sizeof(*tcm), flags);
	if (!nlh)
		goto out_nlmsg_trim;
	tcm = nlmsg_data(nlh);
	tcm->tcm_family = AF_UNSPEC;
	tcm->tcm__pad1 = 0;
	tcm->tcm__pad2 = 0;
	tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
	tcm->tcm_parent = tp->classid;
	tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
	if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
		goto nla_put_failure;
	tcm->tcm_handle = fh;
	if (RTM_DELTFILTER != event) {
		tcm->tcm_handle = 0;
		if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
			goto nla_put_failure;
	}
	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
	return skb->len;

out_nlmsg_trim:
nla_put_failure:
	nlmsg_trim(skb, b);
	return -1;
}

static int tfilter_notify(struct net *net, struct sk_buff *oskb,
			  struct nlmsghdr *n, struct tcf_proto *tp,
			  unsigned long fh, int event)
{
	struct sk_buff *skb;
	u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;

	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
	if (!skb)
		return -ENOBUFS;

	if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) {
		kfree_skb(skb);
		return -EINVAL;
	}

	return rtnetlink_send(skb, net, pid, RTNLGRP_TC,
			      n->nlmsg_flags & NLM_F_ECHO);
}

struct tcf_dump_args {
	struct tcf_walker w;
	struct sk_buff *skb;
	struct netlink_callback *cb;
};

static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
			 struct tcf_walker *arg)
{
	struct tcf_dump_args *a = (void *)arg;

	return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid,
			     a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
}

/* called with RTNL */
static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
{
	struct net *net = sock_net(skb->sk);
	int t;
	int s_t;
	struct net_device *dev;
	struct Qdisc *q;
	struct tcf_proto *tp, **chain;
	struct tcmsg *tcm = nlmsg_data(cb->nlh);
	unsigned long cl = 0;
	const struct Qdisc_class_ops *cops;
	struct tcf_dump_args arg;

	if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
		return skb->len;
	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
	if (!dev)
		return skb->len;

	if (!tcm->tcm_parent)
		q = dev->qdisc;
	else
		q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
	if (!q)
		goto out;
	cops = q->ops->cl_ops;
	if (!cops)
		goto errout;
	if (cops->tcf_chain == NULL)
		goto errout;
	if (TC_H_MIN(tcm->tcm_parent)) {
		cl = cops->get(q, tcm->tcm_parent);
		if (cl == 0)
			goto errout;
	}
	chain = cops->tcf_chain(q, cl);
	if (chain == NULL)
		goto errout;

	s_t = cb->args[0];

	for (tp = *chain, t = 0; tp; tp = tp->next, t++) {
		if (t < s_t)
			continue;
		if (TC_H_MAJ(tcm->tcm_info) &&
		    TC_H_MAJ(tcm->tcm_info) != tp->prio)
			continue;
		if (TC_H_MIN(tcm->tcm_info) &&
		    TC_H_MIN(tcm->tcm_info) != tp->protocol)
			continue;
		if (t > s_t)
			memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
		if (cb->args[1] == 0) {
			if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid,
					  cb->nlh->nlmsg_seq, NLM_F_MULTI,
					  RTM_NEWTFILTER) <= 0)
				break;

			cb->args[1] = 1;
		}
		if (tp->ops->walk == NULL)
			continue;
		arg.w.fn = tcf_node_dump;
		arg.skb = skb;
		arg.cb = cb;
		arg.w.stop = 0;
		arg.w.skip = cb->args[1] - 1;
		arg.w.count = 0;
		tp->ops->walk(tp, &arg.w);
		cb->args[1] = arg.w.count + 1;
		if (arg.w.stop)
			break;
	}

	cb->args[0] = t;

errout:
	if (cl)
		cops->put(q, cl);
out:
	return skb->len;
}

void tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
{
#ifdef CONFIG_NET_CLS_ACT
	if (exts->action) {
		tcf_action_destroy(exts->action, TCA_ACT_UNBIND);
		exts->action = NULL;
	}
#endif
}
EXPORT_SYMBOL(tcf_exts_destroy);

int tcf_exts_validate(struct tcf_proto *tp, struct nlattr **tb,
		  struct nlattr *rate_tlv, struct tcf_exts *exts,
		  const struct tcf_ext_map *map)
{
	memset(exts, 0, sizeof(*exts));

#ifdef CONFIG_NET_CLS_ACT
	{
		struct tc_action *act;

		if (map->police && tb[map->police]) {
			act = tcf_action_init_1(tb[map->police], rate_tlv,
						"police", TCA_ACT_NOREPLACE,
						TCA_ACT_BIND);
			if (IS_ERR(act))
				return PTR_ERR(act);

			act->type = TCA_OLD_COMPAT;
			exts->action = act;
		} else if (map->action && tb[map->action]) {
			act = tcf_action_init(tb[map->action], rate_tlv, NULL,
					      TCA_ACT_NOREPLACE, TCA_ACT_BIND);
			if (IS_ERR(act))
				return PTR_ERR(act);

			exts->action = act;
		}
	}
#else
	if ((map->action && tb[map->action]) ||
	    (map->police && tb[map->police]))
		return -EOPNOTSUPP;
#endif

	return 0;
}
EXPORT_SYMBOL(tcf_exts_validate);

void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
		     struct tcf_exts *src)
{
#ifdef CONFIG_NET_CLS_ACT
	if (src->action) {
		struct tc_action *act;
		tcf_tree_lock(tp);
		act = dst->action;
		dst->action = src->action;
		tcf_tree_unlock(tp);
		if (act)
			tcf_action_destroy(act, TCA_ACT_UNBIND);
	}
#endif
}
EXPORT_SYMBOL(tcf_exts_change);

int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts,
		  const struct tcf_ext_map *map)
{
#ifdef CONFIG_NET_CLS_ACT
	if (map->action && exts->action) {
		/*
		 * again for backward compatible mode - we want
		 * to work with both old and new modes of entering
		 * tc data even if iproute2  was newer - jhs
		 */
		struct nlattr *nest;

		if (exts->action->type != TCA_OLD_COMPAT) {
			nest = nla_nest_start(skb, map->action);
			if (nest == NULL)
				goto nla_put_failure;
			if (tcf_action_dump(skb, exts->action, 0, 0) < 0)
				goto nla_put_failure;
			nla_nest_end(skb, nest);
		} else if (map->police) {
			nest = nla_nest_start(skb, map->police);
			if (nest == NULL)
				goto nla_put_failure;
			if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0)
				goto nla_put_failure;
			nla_nest_end(skb, nest);
		}
	}
#endif
	return 0;
nla_put_failure: __attribute__ ((unused))
	return -1;
}
EXPORT_SYMBOL(tcf_exts_dump);


int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts,
			const struct tcf_ext_map *map)
{
#ifdef CONFIG_NET_CLS_ACT
	if (exts->action)
		if (tcf_action_copy_stats(skb, exts->action, 1) < 0)
			goto nla_put_failure;
#endif
	return 0;
nla_put_failure: __attribute__ ((unused))
	return -1;
}
EXPORT_SYMBOL(tcf_exts_dump_stats);

static int __init tc_filter_init(void)
{
	rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL);
	rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL);
	rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
		      tc_dump_tfilter, NULL);

	return 0;
}

subsys_initcall(tc_filter_init);
Commit	Line	Data
1da177e4 LT	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, <[email protected]>
	10	*
	11	* Changes:
	12	*
	13	* Eduardo J. Blanco <[email protected]> :990222: kmod support
	14	*
	15	*/
	16
1da177e4 LT	17	#include <linux/module.h>
	18	#include <linux/types.h>
	19	#include <linux/kernel.h>
1da177e4	20	#include <linux/string.h>
1da177e4	21	#include <linux/errno.h>
1da177e4	22	#include <linux/skbuff.h>
1da177e4 LT	23	#include <linux/init.h>
1da177e4 LT	24	#include <linux/kmod.h>
dc5fc579	25	#include <linux/netlink.h>
ab27cfb8	26	#include <linux/err.h>
5a0e3ad6	27	#include <linux/slab.h>
b854272b DL	28	#include <net/net_namespace.h>
b854272b DL	29	#include <net/sock.h>
dc5fc579	30	#include <net/netlink.h>
1da177e4 LT	31	#include <net/pkt_sched.h>
	32	#include <net/pkt_cls.h>
	33
1da177e4 LT	34	/* The list of all installed classifier types */
1da177e4 LT	35
2eb9d75c	36	static struct tcf_proto_ops *tcf_proto_base __read_mostly;
1da177e4 LT	37
	38	/* Protects list of registered TC modules. It is pure SMP lock. */
	39	static DEFINE_RWLOCK(cls_mod_lock);
	40
	41	/* Find classifier type by string name */
	42
dc7f9f6e	43	static const struct tcf_proto_ops tcf_proto_lookup_ops(struct nlattr kind)
1da177e4	44	{
dc7f9f6e	45	const struct tcf_proto_ops *t = NULL;
1da177e4 LT	46
	47	if (kind) {
	48	read_lock(&cls_mod_lock);
	49	for (t = tcf_proto_base; t; t = t->next) {
add93b61	50	if (nla_strcmp(kind, t->kind) == 0) {
1da177e4 LT	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
	63	int 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;
	76	out:
	77	write_unlock(&cls_mod_lock);
	78	return rc;
	79	}
aa767bfe	80	EXPORT_SYMBOL(register_tcf_proto_ops);
1da177e4 LT	81
	82	int 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);
cc7ec456	88	for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
1da177e4 LT	89	if (t == ops)
	90	break;
	91
	92	if (!t)
	93	goto out;
	94	*tp = t->next;
	95	rc = 0;
	96	out:
	97	write_unlock(&cls_mod_lock);
	98	return rc;
	99	}
aa767bfe	100	EXPORT_SYMBOL(unregister_tcf_proto_ops);
1da177e4	101
7316ae88 TG	102	static int tfilter_notify(struct net net, struct sk_buff oskb,
	103	struct nlmsghdr n, struct tcf_proto tp,
	104	unsigned long fh, int event);
1da177e4 LT	105
	106
	107	/* Select new prio value from the range, managed by kernel. */
	108
aa767bfe	109	static inline u32 tcf_auto_prio(struct tcf_proto *tp)
1da177e4	110	{
aa767bfe	111	u32 first = TC_H_MAKE(0xC0000000U, 0U);
1da177e4 LT	112
1da177e4 LT	113	if (tp)
cc7ec456	114	first = tp->prio - 1;
1da177e4 LT	115
	116	return first;
	117	}
	118
	119	/* Add/change/delete/get a filter node */
	120
	121	static int tc_ctl_tfilter(struct sk_buff skb, struct nlmsghdr n, void *arg)
	122	{
3b1e0a65	123	struct net *net = sock_net(skb->sk);
add93b61	124	struct nlattr *tca[TCA_MAX + 1];
55dbc640	125	spinlock_t *root_lock;
1da177e4 LT	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;
dc7f9f6e	135	const struct tcf_proto_ops *tp_ops;
20fea08b	136	const struct Qdisc_class_ops *cops;
1da177e4 LT	137	unsigned long cl;
	138	unsigned long fh;
	139	int err;
12186be7	140	int tp_created = 0;
1da177e4 LT	141
1da177e4 LT	142	replay:
942b8165	143	t = nlmsg_data(n);
1da177e4 LT	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. */
cc7ec456 ED	152	if (n->nlmsg_type != RTM_NEWTFILTER \|\|
cc7ec456 ED	153	!(n->nlmsg_flags & NLM_F_CREATE))
1da177e4	154	return -ENOENT;
aa767bfe	155	prio = TC_H_MAKE(0x80000000U, 0U);
1da177e4 LT	156	}
	157
	158	/* Find head of filter chain. */
	159
	160	/* Find link */
7316ae88	161	dev = __dev_get_by_index(net, t->tcm_ifindex);
aa767bfe	162	if (dev == NULL)
1da177e4 LT	163	return -ENODEV;
1da177e4 LT	164
add93b61 PM	165	err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
	166	if (err < 0)
	167	return err;
	168
1da177e4 LT	169	/* Find qdisc */
1da177e4 LT	170	if (!parent) {
af356afa	171	q = dev->qdisc;
1da177e4	172	parent = q->handle;
aa767bfe SH	173	} else {
	174	q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
	175	if (q == NULL)
	176	return -EINVAL;
	177	}
1da177e4 LT	178
1da177e4 LT	179	/* Is it classful? */
cc7ec456 ED	180	cops = q->ops->cl_ops;
cc7ec456 ED	181	if (!cops)
1da177e4 LT	182	return -EINVAL;
1da177e4 LT	183
71ebe5e9 PM	184	if (cops->tcf_chain == NULL)
	185	return -EOPNOTSUPP;
	186
1da177e4 LT	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 */
cc7ec456	201	for (back = chain; (tp = *back) != NULL; back = &tp->next) {
1da177e4 LT	202	if (tp->prio >= prio) {
1da177e4 LT	203	if (tp->prio == prio) {
cc7ec456 ED	204	if (!nprio \|\|
cc7ec456 ED	205	(tp->protocol != protocol && protocol))
1da177e4 LT	206	goto errout;
	207	} else
	208	tp = NULL;
	209	break;
	210	}
	211	}
	212
102396ae	213	root_lock = qdisc_root_sleeping_lock(q);
55dbc640	214
1da177e4 LT	215	if (tp == NULL) {
	216	/* Proto-tcf does not exist, create new one */
	217
add93b61	218	if (tca[TCA_KIND] == NULL \|\| !protocol)
1da177e4 LT	219	goto errout;
	220
	221	err = -ENOENT;
cc7ec456 ED	222	if (n->nlmsg_type != RTM_NEWTFILTER \|\|
cc7ec456 ED	223	!(n->nlmsg_flags & NLM_F_CREATE))
1da177e4 LT	224	goto errout;
	225
	226
	227	/* Create new proto tcf */
	228
	229	err = -ENOBUFS;
aa767bfe SH	230	tp = kzalloc(sizeof(*tp), GFP_KERNEL);
aa767bfe SH	231	if (tp == NULL)
1da177e4	232	goto errout;
f2df8249	233	err = -ENOENT;
add93b61	234	tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
1da177e4	235	if (tp_ops == NULL) {
95a5afca	236	#ifdef CONFIG_MODULES
add93b61	237	struct nlattr *kind = tca[TCA_KIND];
1da177e4 LT	238	char name[IFNAMSIZ];
	239
	240	if (kind != NULL &&
add93b61	241	nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
1da177e4 LT	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	}
1da177e4 LT	261	tp->ops = tp_ops;
1da177e4 LT	262	tp->protocol = protocol;
d0ab8ff8	263	tp->prio = nprio ? : TC_H_MAJ(tcf_auto_prio(*back));
1da177e4 LT	264	tp->q = q;
	265	tp->classify = tp_ops->classify;
	266	tp->classid = parent;
aa767bfe SH	267
	268	err = tp_ops->init(tp);
	269	if (err != 0) {
1da177e4 LT	270	module_put(tp_ops->owner);
	271	kfree(tp);
	272	goto errout;
	273	}
	274
12186be7	275	tp_created = 1;
1da177e4	276
add93b61	277	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
1da177e4 LT	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) {
55dbc640	284	spin_lock_bh(root_lock);
1da177e4	285	*back = tp->next;
323c0488	286	spin_unlock_bh(root_lock);
1da177e4	287
7316ae88	288	tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
1da177e4 LT	289	tcf_destroy(tp);
	290	err = 0;
	291	goto errout;
	292	}
	293
	294	err = -ENOENT;
aa767bfe SH	295	if (n->nlmsg_type != RTM_NEWTFILTER \|\|
aa767bfe SH	296	!(n->nlmsg_flags & NLM_F_CREATE))
1da177e4 LT	297	goto errout;
	298	} else {
	299	switch (n->nlmsg_type) {
10297b99	300	case RTM_NEWTFILTER:
1da177e4	301	err = -EEXIST;
12186be7 MU	302	if (n->nlmsg_flags & NLM_F_EXCL) {
	303	if (tp_created)
	304	tcf_destroy(tp);
1da177e4	305	goto errout;
12186be7	306	}
1da177e4 LT	307	break;
	308	case RTM_DELTFILTER:
	309	err = tp->ops->delete(tp, fh);
	310	if (err == 0)
7316ae88	311	tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
1da177e4 LT	312	goto errout;
1da177e4 LT	313	case RTM_GETTFILTER:
7316ae88	314	err = tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
1da177e4 LT	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);
12186be7 MU	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	}
7316ae88	330	tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
12186be7 MU	331	} else {
	332	if (tp_created)
	333	tcf_destroy(tp);
	334	}
1da177e4 LT	335
	336	errout:
	337	if (cl)
	338	cops->put(q, cl);
	339	if (err == -EAGAIN)
	340	/* Replay the request. */
	341	goto replay;
	342	return err;
	343	}
	344
aa767bfe SH	345	static int tcf_fill_node(struct sk_buff skb, struct tcf_proto tp,
aa767bfe SH	346	unsigned long fh, u32 pid, u32 seq, u16 flags, int event)
1da177e4 LT	347	{
	348	struct tcmsg *tcm;
	349	struct nlmsghdr *nlh;
27a884dc	350	unsigned char *b = skb_tail_pointer(skb);
1da177e4	351
942b8165 DM	352	nlh = nlmsg_put(skb, pid, seq, event, sizeof(*tcm), flags);
	353	if (!nlh)
	354	goto out_nlmsg_trim;
	355	tcm = nlmsg_data(nlh);
1da177e4	356	tcm->tcm_family = AF_UNSPEC;
9ef1d4c7	357	tcm->tcm__pad1 = 0;
ad61df91	358	tcm->tcm__pad2 = 0;
5ce2d488	359	tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
1da177e4 LT	360	tcm->tcm_parent = tp->classid;
1da177e4 LT	361	tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1b34ec43 DM	362	if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1b34ec43 DM	363	goto nla_put_failure;
1da177e4 LT	364	tcm->tcm_handle = fh;
	365	if (RTM_DELTFILTER != event) {
	366	tcm->tcm_handle = 0;
	367	if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
add93b61	368	goto nla_put_failure;
1da177e4	369	}
27a884dc	370	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1da177e4 LT	371	return skb->len;
1da177e4 LT	372
942b8165	373	out_nlmsg_trim:
add93b61	374	nla_put_failure:
dc5fc579	375	nlmsg_trim(skb, b);
1da177e4 LT	376	return -1;
	377	}
	378
7316ae88 TG	379	static int tfilter_notify(struct net net, struct sk_buff oskb,
	380	struct nlmsghdr n, struct tcf_proto tp,
	381	unsigned long fh, int event)
1da177e4 LT	382	{
	383	struct sk_buff *skb;
	384	u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
	385
	386	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
	387	if (!skb)
	388	return -ENOBUFS;
	389
	390	if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) {
	391	kfree_skb(skb);
	392	return -EINVAL;
	393	}
	394
7316ae88	395	return rtnetlink_send(skb, net, pid, RTNLGRP_TC,
aa767bfe	396	n->nlmsg_flags & NLM_F_ECHO);
1da177e4 LT	397	}
1da177e4 LT	398
aa767bfe	399	struct tcf_dump_args {
1da177e4 LT	400	struct tcf_walker w;
	401	struct sk_buff *skb;
	402	struct netlink_callback *cb;
	403	};
	404
aa767bfe SH	405	static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
aa767bfe SH	406	struct tcf_walker *arg)
1da177e4	407	{
aa767bfe	408	struct tcf_dump_args a = (void )arg;
1da177e4 LT	409
	410	return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid,
	411	a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
	412	}
	413
bd27a875	414	/* called with RTNL */
1da177e4 LT	415	static int tc_dump_tfilter(struct sk_buff skb, struct netlink_callback cb)
1da177e4 LT	416	{
3b1e0a65	417	struct net *net = sock_net(skb->sk);
1da177e4 LT	418	int t;
	419	int s_t;
	420	struct net_device *dev;
	421	struct Qdisc *q;
	422	struct tcf_proto tp, *chain;
942b8165	423	struct tcmsg *tcm = nlmsg_data(cb->nlh);
1da177e4	424	unsigned long cl = 0;
20fea08b	425	const struct Qdisc_class_ops *cops;
1da177e4 LT	426	struct tcf_dump_args arg;
	427
	428	if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
	429	return skb->len;
cc7ec456 ED	430	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
cc7ec456 ED	431	if (!dev)
1da177e4 LT	432	return skb->len;
1da177e4 LT	433
1da177e4	434	if (!tcm->tcm_parent)
af356afa	435	q = dev->qdisc;
1da177e4 LT	436	else
	437	q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
	438	if (!q)
	439	goto out;
cc7ec456 ED	440	cops = q->ops->cl_ops;
cc7ec456 ED	441	if (!cops)
1da177e4	442	goto errout;
71ebe5e9 PM	443	if (cops->tcf_chain == NULL)
71ebe5e9 PM	444	goto errout;
1da177e4 LT	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
cc7ec456 ED	456	for (tp = *chain, t = 0; tp; tp = tp->next, t++) {
	457	if (t < s_t)
	458	continue;
1da177e4 LT	459	if (TC_H_MAJ(tcm->tcm_info) &&
	460	TC_H_MAJ(tcm->tcm_info) != tp->prio)
	461	continue;
	462	if (TC_H_MIN(tcm->tcm_info) &&
	463	TC_H_MIN(tcm->tcm_info) != tp->protocol)
	464	continue;
	465	if (t > s_t)
	466	memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
	467	if (cb->args[1] == 0) {
	468	if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid,
aa767bfe SH	469	cb->nlh->nlmsg_seq, NLM_F_MULTI,
aa767bfe SH	470	RTM_NEWTFILTER) <= 0)
1da177e4	471	break;
aa767bfe	472
1da177e4 LT	473	cb->args[1] = 1;
	474	}
	475	if (tp->ops->walk == NULL)
	476	continue;
	477	arg.w.fn = tcf_node_dump;
	478	arg.skb = skb;
	479	arg.cb = cb;
	480	arg.w.stop = 0;
cc7ec456	481	arg.w.skip = cb->args[1] - 1;
1da177e4 LT	482	arg.w.count = 0;
1da177e4 LT	483	tp->ops->walk(tp, &arg.w);
cc7ec456	484	cb->args[1] = arg.w.count + 1;
1da177e4 LT	485	if (arg.w.stop)
	486	break;
	487	}
	488
	489	cb->args[0] = t;
	490
	491	errout:
	492	if (cl)
	493	cops->put(q, cl);
	494	out:
1da177e4 LT	495	return skb->len;
	496	}
	497
aa767bfe	498	void tcf_exts_destroy(struct tcf_proto tp, struct tcf_exts exts)
1da177e4 LT	499	{
	500	#ifdef CONFIG_NET_CLS_ACT
	501	if (exts->action) {
	502	tcf_action_destroy(exts->action, TCA_ACT_UNBIND);
	503	exts->action = NULL;
	504	}
1da177e4 LT	505	#endif
1da177e4 LT	506	}
aa767bfe	507	EXPORT_SYMBOL(tcf_exts_destroy);
1da177e4	508
add93b61 PM	509	int tcf_exts_validate(struct tcf_proto tp, struct nlattr *tb,
add93b61 PM	510	struct nlattr rate_tlv, struct tcf_exts exts,
5239008b	511	const struct tcf_ext_map *map)
1da177e4 LT	512	{
1da177e4 LT	513	memset(exts, 0, sizeof(*exts));
10297b99	514
1da177e4 LT	515	#ifdef CONFIG_NET_CLS_ACT
1da177e4 LT	516	{
1da177e4 LT	517	struct tc_action *act;
1da177e4 LT	518
add93b61	519	if (map->police && tb[map->police]) {
7ba699c6	520	act = tcf_action_init_1(tb[map->police], rate_tlv,
aa767bfe	521	"police", TCA_ACT_NOREPLACE,
ab27cfb8 PM	522	TCA_ACT_BIND);
	523	if (IS_ERR(act))
	524	return PTR_ERR(act);
1da177e4 LT	525
	526	act->type = TCA_OLD_COMPAT;
	527	exts->action = act;
add93b61	528	} else if (map->action && tb[map->action]) {
7ba699c6	529	act = tcf_action_init(tb[map->action], rate_tlv, NULL,
ab27cfb8 PM	530	TCA_ACT_NOREPLACE, TCA_ACT_BIND);
	531	if (IS_ERR(act))
	532	return PTR_ERR(act);
1da177e4 LT	533
	534	exts->action = act;
	535	}
	536	}
1da177e4	537	#else
add93b61 PM	538	if ((map->action && tb[map->action]) \|\|
add93b61 PM	539	(map->police && tb[map->police]))
1da177e4 LT	540	return -EOPNOTSUPP;
	541	#endif
	542
	543	return 0;
	544	}
aa767bfe	545	EXPORT_SYMBOL(tcf_exts_validate);
1da177e4	546
aa767bfe SH	547	void tcf_exts_change(struct tcf_proto tp, struct tcf_exts dst,
aa767bfe SH	548	struct tcf_exts *src)
1da177e4 LT	549	{
	550	#ifdef CONFIG_NET_CLS_ACT
	551	if (src->action) {
	552	struct tc_action *act;
	553	tcf_tree_lock(tp);
47a1a1d4 PM	554	act = dst->action;
47a1a1d4 PM	555	dst->action = src->action;
1da177e4 LT	556	tcf_tree_unlock(tp);
	557	if (act)
	558	tcf_action_destroy(act, TCA_ACT_UNBIND);
	559	}
1da177e4 LT	560	#endif
1da177e4 LT	561	}
aa767bfe	562	EXPORT_SYMBOL(tcf_exts_change);
1da177e4	563
aa767bfe	564	int tcf_exts_dump(struct sk_buff skb, struct tcf_exts exts,
5239008b	565	const struct tcf_ext_map *map)
1da177e4 LT	566	{
	567	#ifdef CONFIG_NET_CLS_ACT
	568	if (map->action && exts->action) {
	569	/*
	570	* again for backward compatible mode - we want
	571	* to work with both old and new modes of entering
	572	* tc data even if iproute2 was newer - jhs
	573	*/
4b3550ef	574	struct nlattr *nest;
1da177e4 LT	575
1da177e4 LT	576	if (exts->action->type != TCA_OLD_COMPAT) {
4b3550ef PM	577	nest = nla_nest_start(skb, map->action);
	578	if (nest == NULL)
	579	goto nla_put_failure;
1da177e4	580	if (tcf_action_dump(skb, exts->action, 0, 0) < 0)
add93b61	581	goto nla_put_failure;
4b3550ef	582	nla_nest_end(skb, nest);
1da177e4	583	} else if (map->police) {
4b3550ef PM	584	nest = nla_nest_start(skb, map->police);
	585	if (nest == NULL)
	586	goto nla_put_failure;
1da177e4	587	if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0)
add93b61	588	goto nla_put_failure;
4b3550ef	589	nla_nest_end(skb, nest);
1da177e4 LT	590	}
1da177e4 LT	591	}
1da177e4 LT	592	#endif
1da177e4 LT	593	return 0;
add93b61	594	nla_put_failure: __attribute__ ((unused))
1da177e4 LT	595	return -1;
1da177e4 LT	596	}
aa767bfe	597	EXPORT_SYMBOL(tcf_exts_dump);
1da177e4	598
aa767bfe SH	599
aa767bfe SH	600	int tcf_exts_dump_stats(struct sk_buff skb, struct tcf_exts exts,
5239008b	601	const struct tcf_ext_map *map)
1da177e4 LT	602	{
	603	#ifdef CONFIG_NET_CLS_ACT
	604	if (exts->action)
	605	if (tcf_action_copy_stats(skb, exts->action, 1) < 0)
add93b61	606	goto nla_put_failure;
1da177e4 LT	607	#endif
1da177e4 LT	608	return 0;
add93b61	609	nla_put_failure: __attribute__ ((unused))
1da177e4 LT	610	return -1;
1da177e4 LT	611	}
aa767bfe	612	EXPORT_SYMBOL(tcf_exts_dump_stats);
1da177e4 LT	613
	614	static int __init tc_filter_init(void)
	615	{
c7ac8679 GR	616	rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL);
c7ac8679 GR	617	rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL);
82623c0d	618	rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
c7ac8679	619	tc_dump_tfilter, NULL);
1da177e4	620
1da177e4 LT	621	return 0;
	622	}
	623
	624	subsys_initcall(tc_filter_init);