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 struct nlattr *kind = tca[TCA_KIND];
277 #ifdef CONFIG_MODULES
279 nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
281 request_module("cls_%s", name);
283 tp_ops = tcf_proto_lookup_ops(kind);
284 /* We dropped the RTNL semaphore in order to
285 * perform the module load. So, even if we
286 * succeeded in loading the module we have to
287 * replay the request. We indicate this using
290 if (tp_ops != NULL) {
291 module_put(tp_ops->owner);
300 tp->protocol = protocol;
302 TC_H_MAJ(tcf_auto_prio(rtnl_dereference(*back)));
304 tp->classify = tp_ops->classify;
305 tp->classid = parent;
307 err = tp_ops->init(tp);
309 module_put(tp_ops->owner);
316 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
319 fh = tp->ops->get(tp, t->tcm_handle);
322 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
323 struct tcf_proto *next = rtnl_dereference(tp->next);
325 RCU_INIT_POINTER(*back, next);
327 tfilter_notify(net, skb, n, tp, fh,
328 RTM_DELTFILTER, false);
329 tcf_destroy(tp, true);
335 if (n->nlmsg_type != RTM_NEWTFILTER ||
336 !(n->nlmsg_flags & NLM_F_CREATE))
339 switch (n->nlmsg_type) {
342 if (n->nlmsg_flags & NLM_F_EXCL) {
344 tcf_destroy(tp, true);
349 err = tp->ops->delete(tp, fh);
351 struct tcf_proto *next = rtnl_dereference(tp->next);
353 tfilter_notify(net, skb, n, tp,
355 RTM_DELTFILTER, false);
356 if (tcf_destroy(tp, false))
357 RCU_INIT_POINTER(*back, next);
361 err = tfilter_notify(net, skb, n, tp, fh,
362 RTM_NEWTFILTER, true);
370 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
371 n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE);
374 RCU_INIT_POINTER(tp->next, rtnl_dereference(*back));
375 rcu_assign_pointer(*back, tp);
377 tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER, false);
380 tcf_destroy(tp, true);
387 /* Replay the request. */
392 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
393 struct tcf_proto *tp, unsigned long fh, u32 portid,
394 u32 seq, u16 flags, int event)
397 struct nlmsghdr *nlh;
398 unsigned char *b = skb_tail_pointer(skb);
400 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
403 tcm = nlmsg_data(nlh);
404 tcm->tcm_family = AF_UNSPEC;
407 tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
408 tcm->tcm_parent = tp->classid;
409 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
410 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
411 goto nla_put_failure;
412 tcm->tcm_handle = fh;
413 if (RTM_DELTFILTER != event) {
415 if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
416 goto nla_put_failure;
418 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
427 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
428 struct nlmsghdr *n, struct tcf_proto *tp,
429 unsigned long fh, int event, bool unicast)
432 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
434 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
438 if (tcf_fill_node(net, skb, tp, fh, portid, n->nlmsg_seq,
439 n->nlmsg_flags, event) <= 0) {
445 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
447 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
448 n->nlmsg_flags & NLM_F_ECHO);
451 struct tcf_dump_args {
454 struct netlink_callback *cb;
457 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
458 struct tcf_walker *arg)
460 struct tcf_dump_args *a = (void *)arg;
461 struct net *net = sock_net(a->skb->sk);
463 return tcf_fill_node(net, a->skb, tp, n, NETLINK_CB(a->cb->skb).portid,
464 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
468 /* called with RTNL */
469 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
471 struct net *net = sock_net(skb->sk);
474 struct net_device *dev;
476 struct tcf_proto *tp, __rcu **chain;
477 struct tcmsg *tcm = nlmsg_data(cb->nlh);
478 unsigned long cl = 0;
479 const struct Qdisc_class_ops *cops;
480 struct tcf_dump_args arg;
482 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
484 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
488 if (!tcm->tcm_parent)
491 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
494 cops = q->ops->cl_ops;
497 if (cops->tcf_chain == NULL)
499 if (TC_H_MIN(tcm->tcm_parent)) {
500 cl = cops->get(q, tcm->tcm_parent);
504 chain = cops->tcf_chain(q, cl);
510 for (tp = rtnl_dereference(*chain), t = 0;
511 tp; tp = rtnl_dereference(tp->next), t++) {
514 if (TC_H_MAJ(tcm->tcm_info) &&
515 TC_H_MAJ(tcm->tcm_info) != tp->prio)
517 if (TC_H_MIN(tcm->tcm_info) &&
518 TC_H_MIN(tcm->tcm_info) != tp->protocol)
521 memset(&cb->args[1], 0,
522 sizeof(cb->args)-sizeof(cb->args[0]));
523 if (cb->args[1] == 0) {
524 if (tcf_fill_node(net, skb, tp, 0,
525 NETLINK_CB(cb->skb).portid,
526 cb->nlh->nlmsg_seq, NLM_F_MULTI,
527 RTM_NEWTFILTER) <= 0)
532 if (tp->ops->walk == NULL)
534 arg.w.fn = tcf_node_dump;
538 arg.w.skip = cb->args[1] - 1;
540 tp->ops->walk(tp, &arg.w);
541 cb->args[1] = arg.w.count + 1;
555 void tcf_exts_destroy(struct tcf_exts *exts)
557 #ifdef CONFIG_NET_CLS_ACT
560 tcf_exts_to_list(exts, &actions);
561 tcf_action_destroy(&actions, TCA_ACT_UNBIND);
562 kfree(exts->actions);
563 exts->nr_actions = 0;
566 EXPORT_SYMBOL(tcf_exts_destroy);
568 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
569 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr)
571 #ifdef CONFIG_NET_CLS_ACT
573 struct tc_action *act;
575 if (exts->police && tb[exts->police]) {
576 act = tcf_action_init_1(net, tb[exts->police], rate_tlv,
577 "police", ovr, TCA_ACT_BIND);
581 act->type = exts->type = TCA_OLD_COMPAT;
582 exts->actions[0] = act;
583 exts->nr_actions = 1;
584 } else if (exts->action && tb[exts->action]) {
588 err = tcf_action_init(net, tb[exts->action], rate_tlv,
589 NULL, ovr, TCA_ACT_BIND,
593 list_for_each_entry(act, &actions, list)
594 exts->actions[i++] = act;
595 exts->nr_actions = i;
599 if ((exts->action && tb[exts->action]) ||
600 (exts->police && tb[exts->police]))
606 EXPORT_SYMBOL(tcf_exts_validate);
608 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
609 struct tcf_exts *src)
611 #ifdef CONFIG_NET_CLS_ACT
612 struct tcf_exts old = *dst;
615 dst->nr_actions = src->nr_actions;
616 dst->actions = src->actions;
617 dst->type = src->type;
620 tcf_exts_destroy(&old);
623 EXPORT_SYMBOL(tcf_exts_change);
625 #ifdef CONFIG_NET_CLS_ACT
626 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
628 if (exts->nr_actions == 0)
631 return exts->actions[0];
635 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
637 #ifdef CONFIG_NET_CLS_ACT
640 if (exts->action && exts->nr_actions) {
642 * again for backward compatible mode - we want
643 * to work with both old and new modes of entering
644 * tc data even if iproute2 was newer - jhs
646 if (exts->type != TCA_OLD_COMPAT) {
649 nest = nla_nest_start(skb, exts->action);
651 goto nla_put_failure;
653 tcf_exts_to_list(exts, &actions);
654 if (tcf_action_dump(skb, &actions, 0, 0) < 0)
655 goto nla_put_failure;
656 nla_nest_end(skb, nest);
657 } else if (exts->police) {
658 struct tc_action *act = tcf_exts_first_act(exts);
659 nest = nla_nest_start(skb, exts->police);
660 if (nest == NULL || !act)
661 goto nla_put_failure;
662 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
663 goto nla_put_failure;
664 nla_nest_end(skb, nest);
670 nla_nest_cancel(skb, nest);
676 EXPORT_SYMBOL(tcf_exts_dump);
679 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
681 #ifdef CONFIG_NET_CLS_ACT
682 struct tc_action *a = tcf_exts_first_act(exts);
683 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
688 EXPORT_SYMBOL(tcf_exts_dump_stats);
690 static int __init tc_filter_init(void)
692 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL);
693 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL);
694 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
695 tc_dump_tfilter, NULL);
700 subsys_initcall(tc_filter_init);