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,
133 struct tcf_block *block)
135 struct tcf_proto *tp;
138 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
140 return ERR_PTR(-ENOBUFS);
143 tp->ops = tcf_proto_lookup_ops(kind);
145 #ifdef CONFIG_MODULES
147 request_module("cls_%s", kind);
149 tp->ops = tcf_proto_lookup_ops(kind);
150 /* We dropped the RTNL semaphore in order to perform
151 * the module load. So, even if we succeeded in loading
152 * the module we have to replay the request. We indicate
153 * this using -EAGAIN.
156 module_put(tp->ops->owner);
164 tp->classify = tp->ops->classify;
165 tp->protocol = protocol;
167 tp->classid = parent;
171 err = tp->ops->init(tp);
173 module_put(tp->ops->owner);
183 static void tcf_proto_destroy(struct tcf_proto *tp)
185 tp->ops->destroy(tp);
186 module_put(tp->ops->owner);
190 static void tcf_chain_destroy(struct tcf_proto __rcu **fl)
192 struct tcf_proto *tp;
194 while ((tp = rtnl_dereference(*fl)) != NULL) {
195 RCU_INIT_POINTER(*fl, tp->next);
196 tcf_proto_destroy(tp);
200 int tcf_block_get(struct tcf_block **p_block,
201 struct tcf_proto __rcu **p_filter_chain)
203 struct tcf_block *block = kzalloc(sizeof(*block), GFP_KERNEL);
207 block->p_filter_chain = p_filter_chain;
211 EXPORT_SYMBOL(tcf_block_get);
213 void tcf_block_put(struct tcf_block *block)
217 tcf_chain_destroy(block->p_filter_chain);
220 EXPORT_SYMBOL(tcf_block_put);
222 /* Main classifier routine: scans classifier chain attached
223 * to this qdisc, (optionally) tests for protocol and asks
224 * specific classifiers.
226 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
227 struct tcf_result *res, bool compat_mode)
229 __be16 protocol = tc_skb_protocol(skb);
230 #ifdef CONFIG_NET_CLS_ACT
231 const int max_reclassify_loop = 4;
232 const struct tcf_proto *old_tp = tp;
237 for (; tp; tp = rcu_dereference_bh(tp->next)) {
240 if (tp->protocol != protocol &&
241 tp->protocol != htons(ETH_P_ALL))
244 err = tp->classify(skb, tp, res);
245 #ifdef CONFIG_NET_CLS_ACT
246 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode))
253 return TC_ACT_UNSPEC; /* signal: continue lookup */
254 #ifdef CONFIG_NET_CLS_ACT
256 if (unlikely(limit++ >= max_reclassify_loop)) {
257 net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n",
258 tp->q->ops->id, tp->prio & 0xffff,
259 ntohs(tp->protocol));
264 protocol = tc_skb_protocol(skb);
268 EXPORT_SYMBOL(tcf_classify);
270 /* Add/change/delete/get a filter node */
272 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
273 struct netlink_ext_ack *extack)
275 struct net *net = sock_net(skb->sk);
276 struct nlattr *tca[TCA_MAX + 1];
282 struct net_device *dev;
284 struct tcf_proto __rcu **back;
285 struct tcf_proto __rcu **chain;
286 struct tcf_block *block;
287 struct tcf_proto *next;
288 struct tcf_proto *tp;
289 const struct Qdisc_class_ops *cops;
295 if ((n->nlmsg_type != RTM_GETTFILTER) &&
296 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
302 err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack);
307 protocol = TC_H_MIN(t->tcm_info);
308 prio = TC_H_MAJ(t->tcm_info);
310 parent = t->tcm_parent;
314 switch (n->nlmsg_type) {
316 if (protocol || t->tcm_handle || tca[TCA_KIND])
320 /* If no priority is provided by the user,
323 if (n->nlmsg_flags & NLM_F_CREATE) {
324 prio = TC_H_MAKE(0x80000000U, 0U);
333 /* Find head of filter chain. */
336 dev = __dev_get_by_index(net, t->tcm_ifindex);
345 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
350 /* Is it classful? */
351 cops = q->ops->cl_ops;
355 if (!cops->tcf_block)
358 /* Do we search for filter, attached to class? */
359 if (TC_H_MIN(parent)) {
360 cl = cops->get(q, parent);
365 /* And the last stroke */
366 block = cops->tcf_block(q, cl);
371 chain = block->p_filter_chain;
373 if (n->nlmsg_type == RTM_DELTFILTER && prio == 0) {
374 tfilter_notify_chain(net, skb, n, chain, RTM_DELTFILTER);
375 tcf_chain_destroy(chain);
380 /* Check the chain for existence of proto-tcf with this priority */
382 (tp = rtnl_dereference(*back)) != NULL;
384 if (tp->prio >= prio) {
385 if (tp->prio == prio) {
387 (tp->protocol != protocol && protocol)) {
399 /* Proto-tcf does not exist, create new one */
401 if (tca[TCA_KIND] == NULL || !protocol) {
406 if (n->nlmsg_type != RTM_NEWTFILTER ||
407 !(n->nlmsg_flags & NLM_F_CREATE)) {
413 nprio = TC_H_MAJ(tcf_auto_prio(rtnl_dereference(*back)));
415 tp = tcf_proto_create(nla_data(tca[TCA_KIND]),
416 protocol, nprio, parent, q, block);
422 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
427 fh = tp->ops->get(tp, t->tcm_handle);
430 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
431 next = rtnl_dereference(tp->next);
432 RCU_INIT_POINTER(*back, next);
433 tfilter_notify(net, skb, n, tp, fh,
434 RTM_DELTFILTER, false);
435 tcf_proto_destroy(tp);
440 if (n->nlmsg_type != RTM_NEWTFILTER ||
441 !(n->nlmsg_flags & NLM_F_CREATE)) {
448 switch (n->nlmsg_type) {
450 if (n->nlmsg_flags & NLM_F_EXCL) {
452 tcf_proto_destroy(tp);
458 err = tp->ops->delete(tp, fh, &last);
461 next = rtnl_dereference(tp->next);
462 tfilter_notify(net, skb, n, tp, t->tcm_handle,
463 RTM_DELTFILTER, false);
465 RCU_INIT_POINTER(*back, next);
466 tcf_proto_destroy(tp);
470 err = tfilter_notify(net, skb, n, tp, fh,
471 RTM_NEWTFILTER, true);
479 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
480 n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE);
483 RCU_INIT_POINTER(tp->next, rtnl_dereference(*back));
484 rcu_assign_pointer(*back, tp);
486 tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER, false);
489 tcf_proto_destroy(tp);
496 /* Replay the request. */
501 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
502 struct tcf_proto *tp, unsigned long fh, u32 portid,
503 u32 seq, u16 flags, int event)
506 struct nlmsghdr *nlh;
507 unsigned char *b = skb_tail_pointer(skb);
509 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
512 tcm = nlmsg_data(nlh);
513 tcm->tcm_family = AF_UNSPEC;
516 tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
517 tcm->tcm_parent = tp->classid;
518 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
519 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
520 goto nla_put_failure;
521 tcm->tcm_handle = fh;
522 if (RTM_DELTFILTER != event) {
524 if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
525 goto nla_put_failure;
527 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
536 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
537 struct nlmsghdr *n, struct tcf_proto *tp,
538 unsigned long fh, int event, bool unicast)
541 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
543 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
547 if (tcf_fill_node(net, skb, tp, fh, portid, n->nlmsg_seq,
548 n->nlmsg_flags, event) <= 0) {
554 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
556 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
557 n->nlmsg_flags & NLM_F_ECHO);
560 struct tcf_dump_args {
563 struct netlink_callback *cb;
566 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
567 struct tcf_walker *arg)
569 struct tcf_dump_args *a = (void *)arg;
570 struct net *net = sock_net(a->skb->sk);
572 return tcf_fill_node(net, a->skb, tp, n, NETLINK_CB(a->cb->skb).portid,
573 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
577 /* called with RTNL */
578 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
580 struct net *net = sock_net(skb->sk);
583 struct net_device *dev;
585 struct tcf_block *block;
586 struct tcf_proto *tp, __rcu **chain;
587 struct tcmsg *tcm = nlmsg_data(cb->nlh);
588 unsigned long cl = 0;
589 const struct Qdisc_class_ops *cops;
590 struct tcf_dump_args arg;
592 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
594 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
598 if (!tcm->tcm_parent)
601 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
604 cops = q->ops->cl_ops;
607 if (!cops->tcf_block)
609 if (TC_H_MIN(tcm->tcm_parent)) {
610 cl = cops->get(q, tcm->tcm_parent);
614 block = cops->tcf_block(q, cl);
617 chain = block->p_filter_chain;
621 for (tp = rtnl_dereference(*chain), t = 0;
622 tp; tp = rtnl_dereference(tp->next), t++) {
625 if (TC_H_MAJ(tcm->tcm_info) &&
626 TC_H_MAJ(tcm->tcm_info) != tp->prio)
628 if (TC_H_MIN(tcm->tcm_info) &&
629 TC_H_MIN(tcm->tcm_info) != tp->protocol)
632 memset(&cb->args[1], 0,
633 sizeof(cb->args)-sizeof(cb->args[0]));
634 if (cb->args[1] == 0) {
635 if (tcf_fill_node(net, skb, tp, 0,
636 NETLINK_CB(cb->skb).portid,
637 cb->nlh->nlmsg_seq, NLM_F_MULTI,
638 RTM_NEWTFILTER) <= 0)
643 if (tp->ops->walk == NULL)
645 arg.w.fn = tcf_node_dump;
649 arg.w.skip = cb->args[1] - 1;
651 tp->ops->walk(tp, &arg.w);
652 cb->args[1] = arg.w.count + 1;
666 void tcf_exts_destroy(struct tcf_exts *exts)
668 #ifdef CONFIG_NET_CLS_ACT
671 tcf_exts_to_list(exts, &actions);
672 tcf_action_destroy(&actions, TCA_ACT_UNBIND);
673 kfree(exts->actions);
674 exts->nr_actions = 0;
677 EXPORT_SYMBOL(tcf_exts_destroy);
679 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
680 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr)
682 #ifdef CONFIG_NET_CLS_ACT
684 struct tc_action *act;
686 if (exts->police && tb[exts->police]) {
687 act = tcf_action_init_1(net, tb[exts->police], rate_tlv,
688 "police", ovr, TCA_ACT_BIND);
692 act->type = exts->type = TCA_OLD_COMPAT;
693 exts->actions[0] = act;
694 exts->nr_actions = 1;
695 } else if (exts->action && tb[exts->action]) {
699 err = tcf_action_init(net, tb[exts->action], rate_tlv,
700 NULL, ovr, TCA_ACT_BIND,
704 list_for_each_entry(act, &actions, list)
705 exts->actions[i++] = act;
706 exts->nr_actions = i;
710 if ((exts->action && tb[exts->action]) ||
711 (exts->police && tb[exts->police]))
717 EXPORT_SYMBOL(tcf_exts_validate);
719 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
720 struct tcf_exts *src)
722 #ifdef CONFIG_NET_CLS_ACT
723 struct tcf_exts old = *dst;
726 dst->nr_actions = src->nr_actions;
727 dst->actions = src->actions;
728 dst->type = src->type;
731 tcf_exts_destroy(&old);
734 EXPORT_SYMBOL(tcf_exts_change);
736 #ifdef CONFIG_NET_CLS_ACT
737 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
739 if (exts->nr_actions == 0)
742 return exts->actions[0];
746 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
748 #ifdef CONFIG_NET_CLS_ACT
751 if (exts->action && exts->nr_actions) {
753 * again for backward compatible mode - we want
754 * to work with both old and new modes of entering
755 * tc data even if iproute2 was newer - jhs
757 if (exts->type != TCA_OLD_COMPAT) {
760 nest = nla_nest_start(skb, exts->action);
762 goto nla_put_failure;
764 tcf_exts_to_list(exts, &actions);
765 if (tcf_action_dump(skb, &actions, 0, 0) < 0)
766 goto nla_put_failure;
767 nla_nest_end(skb, nest);
768 } else if (exts->police) {
769 struct tc_action *act = tcf_exts_first_act(exts);
770 nest = nla_nest_start(skb, exts->police);
771 if (nest == NULL || !act)
772 goto nla_put_failure;
773 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
774 goto nla_put_failure;
775 nla_nest_end(skb, nest);
781 nla_nest_cancel(skb, nest);
787 EXPORT_SYMBOL(tcf_exts_dump);
790 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
792 #ifdef CONFIG_NET_CLS_ACT
793 struct tc_action *a = tcf_exts_first_act(exts);
794 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
799 EXPORT_SYMBOL(tcf_exts_dump_stats);
801 int tcf_exts_get_dev(struct net_device *dev, struct tcf_exts *exts,
802 struct net_device **hw_dev)
804 #ifdef CONFIG_NET_CLS_ACT
805 const struct tc_action *a;
808 if (tc_no_actions(exts))
811 tcf_exts_to_list(exts, &actions);
812 list_for_each_entry(a, &actions, list) {
813 if (a->ops->get_dev) {
814 a->ops->get_dev(a, dev_net(dev), hw_dev);
823 EXPORT_SYMBOL(tcf_exts_get_dev);
825 static int __init tc_filter_init(void)
827 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL);
828 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL);
829 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
830 tc_dump_tfilter, NULL);
835 subsys_initcall(tc_filter_init);