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/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>
29 #include <net/netlink.h>
30 #include <net/pkt_sched.h>
31 #include <net/pkt_cls.h>
33 /* The list of all installed classifier types */
34 static LIST_HEAD(tcf_proto_base);
36 /* Protects list of registered TC modules. It is pure SMP lock. */
37 static DEFINE_RWLOCK(cls_mod_lock);
39 /* Find classifier type by string name */
41 static const struct tcf_proto_ops *tcf_proto_lookup_ops(struct nlattr *kind)
43 const struct tcf_proto_ops *t, *res = NULL;
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))
54 read_unlock(&cls_mod_lock);
59 /* Register(unregister) new classifier type */
61 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
63 struct tcf_proto_ops *t;
66 write_lock(&cls_mod_lock);
67 list_for_each_entry(t, &tcf_proto_base, head)
68 if (!strcmp(ops->kind, t->kind))
71 list_add_tail(&ops->head, &tcf_proto_base);
74 write_unlock(&cls_mod_lock);
77 EXPORT_SYMBOL(register_tcf_proto_ops);
79 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
81 struct tcf_proto_ops *t;
84 /* Wait for outstanding call_rcu()s, if any, from a
85 * tcf_proto_ops's destroy() handler.
89 write_lock(&cls_mod_lock);
90 list_for_each_entry(t, &tcf_proto_base, head) {
97 write_unlock(&cls_mod_lock);
100 EXPORT_SYMBOL(unregister_tcf_proto_ops);
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, bool unicast);
106 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
108 struct tcf_proto __rcu **chain, int event)
110 struct tcf_proto __rcu **it_chain;
111 struct tcf_proto *tp;
113 for (it_chain = chain; (tp = rtnl_dereference(*it_chain)) != NULL;
114 it_chain = &tp->next)
115 tfilter_notify(net, oskb, n, tp, 0, event, false);
118 /* Select new prio value from the range, managed by kernel. */
120 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
122 u32 first = TC_H_MAKE(0xC0000000U, 0U);
125 first = tp->prio - 1;
130 /* Add/change/delete/get a filter node */
132 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n)
134 struct net *net = sock_net(skb->sk);
135 struct nlattr *tca[TCA_MAX + 1];
141 struct net_device *dev;
143 struct tcf_proto __rcu **back;
144 struct tcf_proto __rcu **chain;
145 struct tcf_proto *tp;
146 const struct tcf_proto_ops *tp_ops;
147 const struct Qdisc_class_ops *cops;
153 if ((n->nlmsg_type != RTM_GETTFILTER) &&
154 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
160 err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
165 protocol = TC_H_MIN(t->tcm_info);
166 prio = TC_H_MAJ(t->tcm_info);
168 parent = t->tcm_parent;
172 switch (n->nlmsg_type) {
174 if (protocol || t->tcm_handle || tca[TCA_KIND])
178 /* If no priority is provided by the user,
181 if (n->nlmsg_flags & NLM_F_CREATE) {
182 prio = TC_H_MAKE(0x80000000U, 0U);
191 /* Find head of filter chain. */
194 dev = __dev_get_by_index(net, t->tcm_ifindex);
203 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
208 /* Is it classful? */
209 cops = q->ops->cl_ops;
213 if (cops->tcf_chain == NULL)
216 /* Do we search for filter, attached to class? */
217 if (TC_H_MIN(parent)) {
218 cl = cops->get(q, parent);
223 /* And the last stroke */
224 chain = cops->tcf_chain(q, cl);
228 if (n->nlmsg_type == RTM_DELTFILTER && prio == 0) {
229 tfilter_notify_chain(net, skb, n, chain, RTM_DELTFILTER);
230 tcf_destroy_chain(chain);
235 /* Check the chain for existence of proto-tcf with this priority */
237 (tp = rtnl_dereference(*back)) != NULL;
239 if (tp->prio >= prio) {
240 if (tp->prio == prio) {
242 (tp->protocol != protocol && protocol))
251 /* Proto-tcf does not exist, create new one */
253 if (tca[TCA_KIND] == NULL || !protocol)
257 if (n->nlmsg_type != RTM_NEWTFILTER ||
258 !(n->nlmsg_flags & NLM_F_CREATE))
262 /* Create new proto tcf */
265 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
269 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
270 if (tp_ops == NULL) {
271 #ifdef CONFIG_MODULES
272 struct nlattr *kind = tca[TCA_KIND];
276 nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
278 request_module("cls_%s", name);
280 tp_ops = tcf_proto_lookup_ops(kind);
281 /* We dropped the RTNL semaphore in order to
282 * perform the module load. So, even if we
283 * succeeded in loading the module we have to
284 * replay the request. We indicate this using
287 if (tp_ops != NULL) {
288 module_put(tp_ops->owner);
297 tp->protocol = protocol;
299 TC_H_MAJ(tcf_auto_prio(rtnl_dereference(*back)));
301 tp->classify = tp_ops->classify;
302 tp->classid = parent;
304 err = tp_ops->init(tp);
306 module_put(tp_ops->owner);
313 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
316 fh = tp->ops->get(tp, t->tcm_handle);
319 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
320 struct tcf_proto *next = rtnl_dereference(tp->next);
322 RCU_INIT_POINTER(*back, next);
324 tfilter_notify(net, skb, n, tp, fh,
325 RTM_DELTFILTER, false);
326 tcf_destroy(tp, true);
332 if (n->nlmsg_type != RTM_NEWTFILTER ||
333 !(n->nlmsg_flags & NLM_F_CREATE))
336 switch (n->nlmsg_type) {
339 if (n->nlmsg_flags & NLM_F_EXCL) {
341 tcf_destroy(tp, true);
346 err = tp->ops->delete(tp, fh);
348 struct tcf_proto *next = rtnl_dereference(tp->next);
350 tfilter_notify(net, skb, n, tp,
352 RTM_DELTFILTER, false);
353 if (tcf_destroy(tp, false))
354 RCU_INIT_POINTER(*back, next);
358 err = tfilter_notify(net, skb, n, tp, fh,
359 RTM_NEWTFILTER, true);
367 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
368 n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE);
371 RCU_INIT_POINTER(tp->next, rtnl_dereference(*back));
372 rcu_assign_pointer(*back, tp);
374 tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER, false);
377 tcf_destroy(tp, true);
384 /* Replay the request. */
389 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
390 struct tcf_proto *tp, unsigned long fh, u32 portid,
391 u32 seq, u16 flags, int event)
394 struct nlmsghdr *nlh;
395 unsigned char *b = skb_tail_pointer(skb);
397 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
400 tcm = nlmsg_data(nlh);
401 tcm->tcm_family = AF_UNSPEC;
404 tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
405 tcm->tcm_parent = tp->classid;
406 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
407 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
408 goto nla_put_failure;
409 tcm->tcm_handle = fh;
410 if (RTM_DELTFILTER != event) {
412 if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
413 goto nla_put_failure;
415 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
424 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
425 struct nlmsghdr *n, struct tcf_proto *tp,
426 unsigned long fh, int event, bool unicast)
429 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
431 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
435 if (tcf_fill_node(net, skb, tp, fh, portid, n->nlmsg_seq,
436 n->nlmsg_flags, event) <= 0) {
442 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
444 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
445 n->nlmsg_flags & NLM_F_ECHO);
448 struct tcf_dump_args {
451 struct netlink_callback *cb;
454 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
455 struct tcf_walker *arg)
457 struct tcf_dump_args *a = (void *)arg;
458 struct net *net = sock_net(a->skb->sk);
460 return tcf_fill_node(net, a->skb, tp, n, NETLINK_CB(a->cb->skb).portid,
461 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
465 /* called with RTNL */
466 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
468 struct net *net = sock_net(skb->sk);
471 struct net_device *dev;
473 struct tcf_proto *tp, __rcu **chain;
474 struct tcmsg *tcm = nlmsg_data(cb->nlh);
475 unsigned long cl = 0;
476 const struct Qdisc_class_ops *cops;
477 struct tcf_dump_args arg;
479 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
481 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
485 if (!tcm->tcm_parent)
488 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
491 cops = q->ops->cl_ops;
494 if (cops->tcf_chain == NULL)
496 if (TC_H_MIN(tcm->tcm_parent)) {
497 cl = cops->get(q, tcm->tcm_parent);
501 chain = cops->tcf_chain(q, cl);
507 for (tp = rtnl_dereference(*chain), t = 0;
508 tp; tp = rtnl_dereference(tp->next), t++) {
511 if (TC_H_MAJ(tcm->tcm_info) &&
512 TC_H_MAJ(tcm->tcm_info) != tp->prio)
514 if (TC_H_MIN(tcm->tcm_info) &&
515 TC_H_MIN(tcm->tcm_info) != tp->protocol)
518 memset(&cb->args[1], 0,
519 sizeof(cb->args)-sizeof(cb->args[0]));
520 if (cb->args[1] == 0) {
521 if (tcf_fill_node(net, skb, tp, 0,
522 NETLINK_CB(cb->skb).portid,
523 cb->nlh->nlmsg_seq, NLM_F_MULTI,
524 RTM_NEWTFILTER) <= 0)
529 if (tp->ops->walk == NULL)
531 arg.w.fn = tcf_node_dump;
535 arg.w.skip = cb->args[1] - 1;
537 tp->ops->walk(tp, &arg.w);
538 cb->args[1] = arg.w.count + 1;
552 void tcf_exts_destroy(struct tcf_exts *exts)
554 #ifdef CONFIG_NET_CLS_ACT
557 tcf_exts_to_list(exts, &actions);
558 tcf_action_destroy(&actions, TCA_ACT_UNBIND);
559 kfree(exts->actions);
560 exts->nr_actions = 0;
563 EXPORT_SYMBOL(tcf_exts_destroy);
565 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
566 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr)
568 #ifdef CONFIG_NET_CLS_ACT
570 struct tc_action *act;
572 if (exts->police && tb[exts->police]) {
573 act = tcf_action_init_1(net, tb[exts->police], rate_tlv,
574 "police", ovr, TCA_ACT_BIND);
578 act->type = exts->type = TCA_OLD_COMPAT;
579 exts->actions[0] = act;
580 exts->nr_actions = 1;
581 } else if (exts->action && tb[exts->action]) {
585 err = tcf_action_init(net, tb[exts->action], rate_tlv,
586 NULL, ovr, TCA_ACT_BIND,
590 list_for_each_entry(act, &actions, list)
591 exts->actions[i++] = act;
592 exts->nr_actions = i;
596 if ((exts->action && tb[exts->action]) ||
597 (exts->police && tb[exts->police]))
603 EXPORT_SYMBOL(tcf_exts_validate);
605 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
606 struct tcf_exts *src)
608 #ifdef CONFIG_NET_CLS_ACT
609 struct tcf_exts old = *dst;
612 dst->nr_actions = src->nr_actions;
613 dst->actions = src->actions;
614 dst->type = src->type;
617 tcf_exts_destroy(&old);
620 EXPORT_SYMBOL(tcf_exts_change);
622 #ifdef CONFIG_NET_CLS_ACT
623 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
625 if (exts->nr_actions == 0)
628 return exts->actions[0];
632 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
634 #ifdef CONFIG_NET_CLS_ACT
637 if (exts->action && exts->nr_actions) {
639 * again for backward compatible mode - we want
640 * to work with both old and new modes of entering
641 * tc data even if iproute2 was newer - jhs
643 if (exts->type != TCA_OLD_COMPAT) {
646 nest = nla_nest_start(skb, exts->action);
648 goto nla_put_failure;
650 tcf_exts_to_list(exts, &actions);
651 if (tcf_action_dump(skb, &actions, 0, 0) < 0)
652 goto nla_put_failure;
653 nla_nest_end(skb, nest);
654 } else if (exts->police) {
655 struct tc_action *act = tcf_exts_first_act(exts);
656 nest = nla_nest_start(skb, exts->police);
657 if (nest == NULL || !act)
658 goto nla_put_failure;
659 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
660 goto nla_put_failure;
661 nla_nest_end(skb, nest);
667 nla_nest_cancel(skb, nest);
673 EXPORT_SYMBOL(tcf_exts_dump);
676 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
678 #ifdef CONFIG_NET_CLS_ACT
679 struct tc_action *a = tcf_exts_first_act(exts);
680 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
685 EXPORT_SYMBOL(tcf_exts_dump_stats);
687 int tcf_exts_get_dev(struct net_device *dev, struct tcf_exts *exts,
688 struct net_device **hw_dev)
690 #ifdef CONFIG_NET_CLS_ACT
691 const struct tc_action *a;
694 if (tc_no_actions(exts))
697 tcf_exts_to_list(exts, &actions);
698 list_for_each_entry(a, &actions, list) {
699 if (a->ops->get_dev) {
700 a->ops->get_dev(a, dev_net(dev), hw_dev);
709 EXPORT_SYMBOL(tcf_exts_get_dev);
711 static int __init tc_filter_init(void)
713 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL);
714 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL);
715 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
716 tc_dump_tfilter, NULL);
721 subsys_initcall(tc_filter_init);