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))
158 err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
163 protocol = TC_H_MIN(t->tcm_info);
164 prio = TC_H_MAJ(t->tcm_info);
166 parent = t->tcm_parent;
170 switch (n->nlmsg_type) {
172 if (protocol || t->tcm_handle || tca[TCA_KIND])
176 /* If no priority is provided by the user,
179 if (n->nlmsg_flags & NLM_F_CREATE) {
180 prio = TC_H_MAKE(0x80000000U, 0U);
189 /* Find head of filter chain. */
192 dev = __dev_get_by_index(net, t->tcm_ifindex);
201 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
206 /* Is it classful? */
207 cops = q->ops->cl_ops;
211 if (cops->tcf_chain == NULL)
214 /* Do we search for filter, attached to class? */
215 if (TC_H_MIN(parent)) {
216 cl = cops->get(q, parent);
221 /* And the last stroke */
222 chain = cops->tcf_chain(q, cl);
226 if (n->nlmsg_type == RTM_DELTFILTER && prio == 0) {
227 tfilter_notify_chain(net, skb, n, chain, RTM_DELTFILTER);
228 tcf_destroy_chain(chain);
233 /* Check the chain for existence of proto-tcf with this priority */
235 (tp = rtnl_dereference(*back)) != NULL;
237 if (tp->prio >= prio) {
238 if (tp->prio == prio) {
240 (tp->protocol != protocol && protocol))
249 /* Proto-tcf does not exist, create new one */
251 if (tca[TCA_KIND] == NULL || !protocol)
255 if (n->nlmsg_type != RTM_NEWTFILTER ||
256 !(n->nlmsg_flags & NLM_F_CREATE))
260 /* Create new proto tcf */
263 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
267 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
268 if (tp_ops == NULL) {
269 #ifdef CONFIG_MODULES
270 struct nlattr *kind = tca[TCA_KIND];
274 nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
276 request_module("cls_%s", name);
278 tp_ops = tcf_proto_lookup_ops(kind);
279 /* We dropped the RTNL semaphore in order to
280 * perform the module load. So, even if we
281 * succeeded in loading the module we have to
282 * replay the request. We indicate this using
285 if (tp_ops != NULL) {
286 module_put(tp_ops->owner);
295 tp->protocol = protocol;
297 TC_H_MAJ(tcf_auto_prio(rtnl_dereference(*back)));
299 tp->classify = tp_ops->classify;
300 tp->classid = parent;
302 err = tp_ops->init(tp);
304 module_put(tp_ops->owner);
311 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
314 fh = tp->ops->get(tp, t->tcm_handle);
317 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
318 struct tcf_proto *next = rtnl_dereference(tp->next);
320 RCU_INIT_POINTER(*back, next);
322 tfilter_notify(net, skb, n, tp, fh,
323 RTM_DELTFILTER, false);
324 tcf_destroy(tp, true);
330 if (n->nlmsg_type != RTM_NEWTFILTER ||
331 !(n->nlmsg_flags & NLM_F_CREATE))
334 switch (n->nlmsg_type) {
337 if (n->nlmsg_flags & NLM_F_EXCL) {
339 tcf_destroy(tp, true);
344 err = tp->ops->delete(tp, fh);
346 struct tcf_proto *next = rtnl_dereference(tp->next);
348 tfilter_notify(net, skb, n, tp,
350 RTM_DELTFILTER, false);
351 if (tcf_destroy(tp, false))
352 RCU_INIT_POINTER(*back, next);
356 err = tfilter_notify(net, skb, n, tp, fh,
357 RTM_NEWTFILTER, true);
365 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
366 n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE);
369 RCU_INIT_POINTER(tp->next, rtnl_dereference(*back));
370 rcu_assign_pointer(*back, tp);
372 tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER, false);
375 tcf_destroy(tp, true);
382 /* Replay the request. */
387 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
388 struct tcf_proto *tp, unsigned long fh, u32 portid,
389 u32 seq, u16 flags, int event)
392 struct nlmsghdr *nlh;
393 unsigned char *b = skb_tail_pointer(skb);
395 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
398 tcm = nlmsg_data(nlh);
399 tcm->tcm_family = AF_UNSPEC;
402 tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
403 tcm->tcm_parent = tp->classid;
404 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
405 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
406 goto nla_put_failure;
407 tcm->tcm_handle = fh;
408 if (RTM_DELTFILTER != event) {
410 if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
411 goto nla_put_failure;
413 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
422 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
423 struct nlmsghdr *n, struct tcf_proto *tp,
424 unsigned long fh, int event, bool unicast)
427 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
429 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
433 if (tcf_fill_node(net, skb, tp, fh, portid, n->nlmsg_seq,
434 n->nlmsg_flags, event) <= 0) {
440 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
442 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
443 n->nlmsg_flags & NLM_F_ECHO);
446 struct tcf_dump_args {
449 struct netlink_callback *cb;
452 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
453 struct tcf_walker *arg)
455 struct tcf_dump_args *a = (void *)arg;
456 struct net *net = sock_net(a->skb->sk);
458 return tcf_fill_node(net, a->skb, tp, n, NETLINK_CB(a->cb->skb).portid,
459 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
463 /* called with RTNL */
464 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
466 struct net *net = sock_net(skb->sk);
469 struct net_device *dev;
471 struct tcf_proto *tp, __rcu **chain;
472 struct tcmsg *tcm = nlmsg_data(cb->nlh);
473 unsigned long cl = 0;
474 const struct Qdisc_class_ops *cops;
475 struct tcf_dump_args arg;
477 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
479 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
483 if (!tcm->tcm_parent)
486 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
489 cops = q->ops->cl_ops;
492 if (cops->tcf_chain == NULL)
494 if (TC_H_MIN(tcm->tcm_parent)) {
495 cl = cops->get(q, tcm->tcm_parent);
499 chain = cops->tcf_chain(q, cl);
505 for (tp = rtnl_dereference(*chain), t = 0;
506 tp; tp = rtnl_dereference(tp->next), t++) {
509 if (TC_H_MAJ(tcm->tcm_info) &&
510 TC_H_MAJ(tcm->tcm_info) != tp->prio)
512 if (TC_H_MIN(tcm->tcm_info) &&
513 TC_H_MIN(tcm->tcm_info) != tp->protocol)
516 memset(&cb->args[1], 0,
517 sizeof(cb->args)-sizeof(cb->args[0]));
518 if (cb->args[1] == 0) {
519 if (tcf_fill_node(net, skb, tp, 0,
520 NETLINK_CB(cb->skb).portid,
521 cb->nlh->nlmsg_seq, NLM_F_MULTI,
522 RTM_NEWTFILTER) <= 0)
527 if (tp->ops->walk == NULL)
529 arg.w.fn = tcf_node_dump;
533 arg.w.skip = cb->args[1] - 1;
535 tp->ops->walk(tp, &arg.w);
536 cb->args[1] = arg.w.count + 1;
550 void tcf_exts_destroy(struct tcf_exts *exts)
552 #ifdef CONFIG_NET_CLS_ACT
555 tcf_exts_to_list(exts, &actions);
556 tcf_action_destroy(&actions, TCA_ACT_UNBIND);
557 kfree(exts->actions);
558 exts->nr_actions = 0;
561 EXPORT_SYMBOL(tcf_exts_destroy);
563 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
564 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr)
566 #ifdef CONFIG_NET_CLS_ACT
568 struct tc_action *act;
570 if (exts->police && tb[exts->police]) {
571 act = tcf_action_init_1(net, tb[exts->police], rate_tlv,
572 "police", ovr, TCA_ACT_BIND);
576 act->type = exts->type = TCA_OLD_COMPAT;
577 exts->actions[0] = act;
578 exts->nr_actions = 1;
579 } else if (exts->action && tb[exts->action]) {
583 err = tcf_action_init(net, tb[exts->action], rate_tlv,
584 NULL, ovr, TCA_ACT_BIND,
588 list_for_each_entry(act, &actions, list)
589 exts->actions[i++] = act;
590 exts->nr_actions = i;
594 if ((exts->action && tb[exts->action]) ||
595 (exts->police && tb[exts->police]))
601 EXPORT_SYMBOL(tcf_exts_validate);
603 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
604 struct tcf_exts *src)
606 #ifdef CONFIG_NET_CLS_ACT
607 struct tcf_exts old = *dst;
610 dst->nr_actions = src->nr_actions;
611 dst->actions = src->actions;
612 dst->type = src->type;
615 tcf_exts_destroy(&old);
618 EXPORT_SYMBOL(tcf_exts_change);
620 #ifdef CONFIG_NET_CLS_ACT
621 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
623 if (exts->nr_actions == 0)
626 return exts->actions[0];
630 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
632 #ifdef CONFIG_NET_CLS_ACT
635 if (exts->action && exts->nr_actions) {
637 * again for backward compatible mode - we want
638 * to work with both old and new modes of entering
639 * tc data even if iproute2 was newer - jhs
641 if (exts->type != TCA_OLD_COMPAT) {
644 nest = nla_nest_start(skb, exts->action);
646 goto nla_put_failure;
648 tcf_exts_to_list(exts, &actions);
649 if (tcf_action_dump(skb, &actions, 0, 0) < 0)
650 goto nla_put_failure;
651 nla_nest_end(skb, nest);
652 } else if (exts->police) {
653 struct tc_action *act = tcf_exts_first_act(exts);
654 nest = nla_nest_start(skb, exts->police);
655 if (nest == NULL || !act)
656 goto nla_put_failure;
657 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
658 goto nla_put_failure;
659 nla_nest_end(skb, nest);
665 nla_nest_cancel(skb, nest);
671 EXPORT_SYMBOL(tcf_exts_dump);
674 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
676 #ifdef CONFIG_NET_CLS_ACT
677 struct tc_action *a = tcf_exts_first_act(exts);
678 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
683 EXPORT_SYMBOL(tcf_exts_dump_stats);
685 int tcf_exts_get_dev(struct net_device *dev, struct tcf_exts *exts,
686 struct net_device **hw_dev)
688 #ifdef CONFIG_NET_CLS_ACT
689 const struct tc_action *a;
692 if (tc_no_actions(exts))
695 tcf_exts_to_list(exts, &actions);
696 list_for_each_entry(a, &actions, list) {
697 if (a->ops->get_dev) {
698 a->ops->get_dev(a, dev_net(dev), hw_dev);
707 EXPORT_SYMBOL(tcf_exts_get_dev);
709 static int __init tc_filter_init(void)
711 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL);
712 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL);
713 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
714 tc_dump_tfilter, NULL);
719 subsys_initcall(tc_filter_init);