2 * net/sched/cls_api.c Packet classifier API.
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.
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
13 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
17 #include <linux/module.h>
18 #include <linux/types.h>
19 #include <linux/kernel.h>
20 #include <linux/string.h>
21 #include <linux/errno.h>
22 #include <linux/err.h>
23 #include <linux/skbuff.h>
24 #include <linux/init.h>
25 #include <linux/kmod.h>
26 #include <linux/err.h>
27 #include <linux/slab.h>
28 #include <net/net_namespace.h>
30 #include <net/netlink.h>
31 #include <net/pkt_sched.h>
32 #include <net/pkt_cls.h>
34 /* The list of all installed classifier types */
35 static LIST_HEAD(tcf_proto_base);
37 /* Protects list of registered TC modules. It is pure SMP lock. */
38 static DEFINE_RWLOCK(cls_mod_lock);
40 /* Find classifier type by string name */
42 static const struct tcf_proto_ops *tcf_proto_lookup_ops(const char *kind)
44 const struct tcf_proto_ops *t, *res = NULL;
47 read_lock(&cls_mod_lock);
48 list_for_each_entry(t, &tcf_proto_base, head) {
49 if (strcmp(kind, t->kind) == 0) {
50 if (try_module_get(t->owner))
55 read_unlock(&cls_mod_lock);
60 /* Register(unregister) new classifier type */
62 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
64 struct tcf_proto_ops *t;
67 write_lock(&cls_mod_lock);
68 list_for_each_entry(t, &tcf_proto_base, head)
69 if (!strcmp(ops->kind, t->kind))
72 list_add_tail(&ops->head, &tcf_proto_base);
75 write_unlock(&cls_mod_lock);
78 EXPORT_SYMBOL(register_tcf_proto_ops);
80 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
82 struct tcf_proto_ops *t;
85 /* Wait for outstanding call_rcu()s, if any, from a
86 * tcf_proto_ops's destroy() handler.
90 write_lock(&cls_mod_lock);
91 list_for_each_entry(t, &tcf_proto_base, head) {
98 write_unlock(&cls_mod_lock);
101 EXPORT_SYMBOL(unregister_tcf_proto_ops);
103 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
104 struct nlmsghdr *n, struct tcf_proto *tp,
105 unsigned long fh, int event, bool unicast);
107 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
109 struct tcf_proto __rcu **chain, int event)
111 struct tcf_proto __rcu **it_chain;
112 struct tcf_proto *tp;
114 for (it_chain = chain; (tp = rtnl_dereference(*it_chain)) != NULL;
115 it_chain = &tp->next)
116 tfilter_notify(net, oskb, n, tp, 0, event, false);
119 /* Select new prio value from the range, managed by kernel. */
121 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
123 u32 first = TC_H_MAKE(0xC0000000U, 0U);
126 first = tp->prio - 1;
131 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
132 u32 prio, u32 parent, struct Qdisc *q)
134 struct tcf_proto *tp;
137 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
139 return ERR_PTR(-ENOBUFS);
142 tp->ops = tcf_proto_lookup_ops(kind);
144 #ifdef CONFIG_MODULES
146 request_module("cls_%s", kind);
148 tp->ops = tcf_proto_lookup_ops(kind);
149 /* We dropped the RTNL semaphore in order to perform
150 * the module load. So, even if we succeeded in loading
151 * the module we have to replay the request. We indicate
152 * this using -EAGAIN.
155 module_put(tp->ops->owner);
163 tp->classify = tp->ops->classify;
164 tp->protocol = protocol;
166 tp->classid = parent;
169 err = tp->ops->init(tp);
171 module_put(tp->ops->owner);
181 static bool tcf_proto_destroy(struct tcf_proto *tp, bool force)
183 if (tp->ops->destroy(tp, force)) {
184 module_put(tp->ops->owner);
191 void tcf_destroy_chain(struct tcf_proto __rcu **fl)
193 struct tcf_proto *tp;
195 while ((tp = rtnl_dereference(*fl)) != NULL) {
196 RCU_INIT_POINTER(*fl, tp->next);
197 tcf_proto_destroy(tp, true);
200 EXPORT_SYMBOL(tcf_destroy_chain);
202 /* Add/change/delete/get a filter node */
204 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n)
206 struct net *net = sock_net(skb->sk);
207 struct nlattr *tca[TCA_MAX + 1];
213 struct net_device *dev;
215 struct tcf_proto __rcu **back;
216 struct tcf_proto __rcu **chain;
217 struct tcf_proto *next;
218 struct tcf_proto *tp;
219 const struct Qdisc_class_ops *cops;
225 if ((n->nlmsg_type != RTM_GETTFILTER) &&
226 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
232 err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
237 protocol = TC_H_MIN(t->tcm_info);
238 prio = TC_H_MAJ(t->tcm_info);
240 parent = t->tcm_parent;
244 switch (n->nlmsg_type) {
246 if (protocol || t->tcm_handle || tca[TCA_KIND])
250 /* If no priority is provided by the user,
253 if (n->nlmsg_flags & NLM_F_CREATE) {
254 prio = TC_H_MAKE(0x80000000U, 0U);
263 /* Find head of filter chain. */
266 dev = __dev_get_by_index(net, t->tcm_ifindex);
275 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
280 /* Is it classful? */
281 cops = q->ops->cl_ops;
285 if (cops->tcf_chain == NULL)
288 /* Do we search for filter, attached to class? */
289 if (TC_H_MIN(parent)) {
290 cl = cops->get(q, parent);
295 /* And the last stroke */
296 chain = cops->tcf_chain(q, cl);
301 if (n->nlmsg_type == RTM_DELTFILTER && prio == 0) {
302 tfilter_notify_chain(net, skb, n, chain, RTM_DELTFILTER);
303 tcf_destroy_chain(chain);
308 /* Check the chain for existence of proto-tcf with this priority */
310 (tp = rtnl_dereference(*back)) != NULL;
312 if (tp->prio >= prio) {
313 if (tp->prio == prio) {
315 (tp->protocol != protocol && protocol)) {
327 /* Proto-tcf does not exist, create new one */
329 if (tca[TCA_KIND] == NULL || !protocol) {
334 if (n->nlmsg_type != RTM_NEWTFILTER ||
335 !(n->nlmsg_flags & NLM_F_CREATE)) {
341 nprio = TC_H_MAJ(tcf_auto_prio(rtnl_dereference(*back)));
343 tp = tcf_proto_create(nla_data(tca[TCA_KIND]),
344 protocol, nprio, parent, q);
350 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
355 fh = tp->ops->get(tp, t->tcm_handle);
358 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
359 next = rtnl_dereference(tp->next);
360 RCU_INIT_POINTER(*back, next);
361 tfilter_notify(net, skb, n, tp, fh,
362 RTM_DELTFILTER, false);
363 tcf_proto_destroy(tp, true);
368 if (n->nlmsg_type != RTM_NEWTFILTER ||
369 !(n->nlmsg_flags & NLM_F_CREATE)) {
374 switch (n->nlmsg_type) {
376 if (n->nlmsg_flags & NLM_F_EXCL) {
378 tcf_proto_destroy(tp, true);
384 err = tp->ops->delete(tp, fh);
387 next = rtnl_dereference(tp->next);
388 tfilter_notify(net, skb, n, tp, t->tcm_handle,
389 RTM_DELTFILTER, false);
390 if (tcf_proto_destroy(tp, false))
391 RCU_INIT_POINTER(*back, next);
394 err = tfilter_notify(net, skb, n, tp, fh,
395 RTM_NEWTFILTER, true);
403 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
404 n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE);
407 RCU_INIT_POINTER(tp->next, rtnl_dereference(*back));
408 rcu_assign_pointer(*back, tp);
410 tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER, false);
413 tcf_proto_destroy(tp, true);
420 /* Replay the request. */
425 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
426 struct tcf_proto *tp, unsigned long fh, u32 portid,
427 u32 seq, u16 flags, int event)
430 struct nlmsghdr *nlh;
431 unsigned char *b = skb_tail_pointer(skb);
433 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
436 tcm = nlmsg_data(nlh);
437 tcm->tcm_family = AF_UNSPEC;
440 tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
441 tcm->tcm_parent = tp->classid;
442 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
443 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
444 goto nla_put_failure;
445 tcm->tcm_handle = fh;
446 if (RTM_DELTFILTER != event) {
448 if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
449 goto nla_put_failure;
451 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
460 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
461 struct nlmsghdr *n, struct tcf_proto *tp,
462 unsigned long fh, int event, bool unicast)
465 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
467 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
471 if (tcf_fill_node(net, skb, tp, fh, portid, n->nlmsg_seq,
472 n->nlmsg_flags, event) <= 0) {
478 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
480 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
481 n->nlmsg_flags & NLM_F_ECHO);
484 struct tcf_dump_args {
487 struct netlink_callback *cb;
490 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
491 struct tcf_walker *arg)
493 struct tcf_dump_args *a = (void *)arg;
494 struct net *net = sock_net(a->skb->sk);
496 return tcf_fill_node(net, a->skb, tp, n, NETLINK_CB(a->cb->skb).portid,
497 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
501 /* called with RTNL */
502 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
504 struct net *net = sock_net(skb->sk);
507 struct net_device *dev;
509 struct tcf_proto *tp, __rcu **chain;
510 struct tcmsg *tcm = nlmsg_data(cb->nlh);
511 unsigned long cl = 0;
512 const struct Qdisc_class_ops *cops;
513 struct tcf_dump_args arg;
515 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
517 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
521 if (!tcm->tcm_parent)
524 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
527 cops = q->ops->cl_ops;
530 if (cops->tcf_chain == NULL)
532 if (TC_H_MIN(tcm->tcm_parent)) {
533 cl = cops->get(q, tcm->tcm_parent);
537 chain = cops->tcf_chain(q, cl);
543 for (tp = rtnl_dereference(*chain), t = 0;
544 tp; tp = rtnl_dereference(tp->next), t++) {
547 if (TC_H_MAJ(tcm->tcm_info) &&
548 TC_H_MAJ(tcm->tcm_info) != tp->prio)
550 if (TC_H_MIN(tcm->tcm_info) &&
551 TC_H_MIN(tcm->tcm_info) != tp->protocol)
554 memset(&cb->args[1], 0,
555 sizeof(cb->args)-sizeof(cb->args[0]));
556 if (cb->args[1] == 0) {
557 if (tcf_fill_node(net, skb, tp, 0,
558 NETLINK_CB(cb->skb).portid,
559 cb->nlh->nlmsg_seq, NLM_F_MULTI,
560 RTM_NEWTFILTER) <= 0)
565 if (tp->ops->walk == NULL)
567 arg.w.fn = tcf_node_dump;
571 arg.w.skip = cb->args[1] - 1;
573 tp->ops->walk(tp, &arg.w);
574 cb->args[1] = arg.w.count + 1;
588 void tcf_exts_destroy(struct tcf_exts *exts)
590 #ifdef CONFIG_NET_CLS_ACT
593 tcf_exts_to_list(exts, &actions);
594 tcf_action_destroy(&actions, TCA_ACT_UNBIND);
595 kfree(exts->actions);
596 exts->nr_actions = 0;
599 EXPORT_SYMBOL(tcf_exts_destroy);
601 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
602 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr)
604 #ifdef CONFIG_NET_CLS_ACT
606 struct tc_action *act;
608 if (exts->police && tb[exts->police]) {
609 act = tcf_action_init_1(net, tb[exts->police], rate_tlv,
610 "police", ovr, TCA_ACT_BIND);
614 act->type = exts->type = TCA_OLD_COMPAT;
615 exts->actions[0] = act;
616 exts->nr_actions = 1;
617 } else if (exts->action && tb[exts->action]) {
621 err = tcf_action_init(net, tb[exts->action], rate_tlv,
622 NULL, ovr, TCA_ACT_BIND,
626 list_for_each_entry(act, &actions, list)
627 exts->actions[i++] = act;
628 exts->nr_actions = i;
632 if ((exts->action && tb[exts->action]) ||
633 (exts->police && tb[exts->police]))
639 EXPORT_SYMBOL(tcf_exts_validate);
641 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
642 struct tcf_exts *src)
644 #ifdef CONFIG_NET_CLS_ACT
645 struct tcf_exts old = *dst;
648 dst->nr_actions = src->nr_actions;
649 dst->actions = src->actions;
650 dst->type = src->type;
653 tcf_exts_destroy(&old);
656 EXPORT_SYMBOL(tcf_exts_change);
658 #ifdef CONFIG_NET_CLS_ACT
659 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
661 if (exts->nr_actions == 0)
664 return exts->actions[0];
668 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
670 #ifdef CONFIG_NET_CLS_ACT
673 if (exts->action && exts->nr_actions) {
675 * again for backward compatible mode - we want
676 * to work with both old and new modes of entering
677 * tc data even if iproute2 was newer - jhs
679 if (exts->type != TCA_OLD_COMPAT) {
682 nest = nla_nest_start(skb, exts->action);
684 goto nla_put_failure;
686 tcf_exts_to_list(exts, &actions);
687 if (tcf_action_dump(skb, &actions, 0, 0) < 0)
688 goto nla_put_failure;
689 nla_nest_end(skb, nest);
690 } else if (exts->police) {
691 struct tc_action *act = tcf_exts_first_act(exts);
692 nest = nla_nest_start(skb, exts->police);
693 if (nest == NULL || !act)
694 goto nla_put_failure;
695 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
696 goto nla_put_failure;
697 nla_nest_end(skb, nest);
703 nla_nest_cancel(skb, nest);
709 EXPORT_SYMBOL(tcf_exts_dump);
712 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
714 #ifdef CONFIG_NET_CLS_ACT
715 struct tc_action *a = tcf_exts_first_act(exts);
716 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
721 EXPORT_SYMBOL(tcf_exts_dump_stats);
723 int tcf_exts_get_dev(struct net_device *dev, struct tcf_exts *exts,
724 struct net_device **hw_dev)
726 #ifdef CONFIG_NET_CLS_ACT
727 const struct tc_action *a;
730 if (tc_no_actions(exts))
733 tcf_exts_to_list(exts, &actions);
734 list_for_each_entry(a, &actions, list) {
735 if (a->ops->get_dev) {
736 a->ops->get_dev(a, dev_net(dev), hw_dev);
745 EXPORT_SYMBOL(tcf_exts_get_dev);
747 static int __init tc_filter_init(void)
749 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL);
750 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL);
751 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
752 tc_dump_tfilter, NULL);
757 subsys_initcall(tc_filter_init);