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 write_lock(&cls_mod_lock);
85 list_for_each_entry(t, &tcf_proto_base, head) {
92 write_unlock(&cls_mod_lock);
95 EXPORT_SYMBOL(unregister_tcf_proto_ops);
97 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
98 struct nlmsghdr *n, struct tcf_proto *tp,
99 unsigned long fh, int event);
102 /* Select new prio value from the range, managed by kernel. */
104 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
106 u32 first = TC_H_MAKE(0xC0000000U, 0U);
109 first = tp->prio - 1;
114 /* Add/change/delete/get a filter node */
116 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n)
118 struct net *net = sock_net(skb->sk);
119 struct nlattr *tca[TCA_MAX + 1];
125 struct net_device *dev;
127 struct tcf_proto __rcu **back;
128 struct tcf_proto __rcu **chain;
129 struct tcf_proto *tp;
130 const struct tcf_proto_ops *tp_ops;
131 const struct Qdisc_class_ops *cops;
137 if ((n->nlmsg_type != RTM_GETTFILTER) &&
138 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
142 err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
147 protocol = TC_H_MIN(t->tcm_info);
148 prio = TC_H_MAJ(t->tcm_info);
150 parent = t->tcm_parent;
154 /* If no priority is given, user wants we allocated it. */
155 if (n->nlmsg_type != RTM_NEWTFILTER ||
156 !(n->nlmsg_flags & NLM_F_CREATE))
158 prio = TC_H_MAKE(0x80000000U, 0U);
161 /* Find head of filter chain. */
164 dev = __dev_get_by_index(net, t->tcm_ifindex);
173 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
178 /* Is it classful? */
179 cops = q->ops->cl_ops;
183 if (cops->tcf_chain == NULL)
186 /* Do we search for filter, attached to class? */
187 if (TC_H_MIN(parent)) {
188 cl = cops->get(q, parent);
193 /* And the last stroke */
194 chain = cops->tcf_chain(q, cl);
199 /* Check the chain for existence of proto-tcf with this priority */
201 (tp = rtnl_dereference(*back)) != NULL;
203 if (tp->prio >= prio) {
204 if (tp->prio == prio) {
206 (tp->protocol != protocol && protocol))
215 /* Proto-tcf does not exist, create new one */
217 if (tca[TCA_KIND] == NULL || !protocol)
221 if (n->nlmsg_type != RTM_NEWTFILTER ||
222 !(n->nlmsg_flags & NLM_F_CREATE))
226 /* Create new proto tcf */
229 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
233 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
234 if (tp_ops == NULL) {
235 #ifdef CONFIG_MODULES
236 struct nlattr *kind = tca[TCA_KIND];
240 nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
242 request_module("cls_%s", name);
244 tp_ops = tcf_proto_lookup_ops(kind);
245 /* We dropped the RTNL semaphore in order to
246 * perform the module load. So, even if we
247 * succeeded in loading the module we have to
248 * replay the request. We indicate this using
251 if (tp_ops != NULL) {
252 module_put(tp_ops->owner);
261 tp->protocol = protocol;
263 TC_H_MAJ(tcf_auto_prio(rtnl_dereference(*back)));
265 tp->classify = tp_ops->classify;
266 tp->classid = parent;
268 err = tp_ops->init(tp);
270 module_put(tp_ops->owner);
277 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
280 fh = tp->ops->get(tp, t->tcm_handle);
283 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
284 struct tcf_proto *next = rtnl_dereference(tp->next);
286 RCU_INIT_POINTER(*back, next);
288 tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
289 tcf_destroy(tp, true);
295 if (n->nlmsg_type != RTM_NEWTFILTER ||
296 !(n->nlmsg_flags & NLM_F_CREATE))
299 switch (n->nlmsg_type) {
302 if (n->nlmsg_flags & NLM_F_EXCL) {
304 tcf_destroy(tp, true);
309 err = tp->ops->delete(tp, fh);
311 tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
312 if (tcf_destroy(tp, false)) {
313 struct tcf_proto *next = rtnl_dereference(tp->next);
315 RCU_INIT_POINTER(*back, next);
320 err = tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
328 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
329 n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE);
332 RCU_INIT_POINTER(tp->next, rtnl_dereference(*back));
333 rcu_assign_pointer(*back, tp);
335 tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
338 tcf_destroy(tp, true);
345 /* Replay the request. */
350 static int tcf_fill_node(struct net *net, struct sk_buff *skb, struct tcf_proto *tp,
351 unsigned long fh, u32 portid, u32 seq, u16 flags, int event)
354 struct nlmsghdr *nlh;
355 unsigned char *b = skb_tail_pointer(skb);
357 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
360 tcm = nlmsg_data(nlh);
361 tcm->tcm_family = AF_UNSPEC;
364 tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
365 tcm->tcm_parent = tp->classid;
366 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
367 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
368 goto nla_put_failure;
369 tcm->tcm_handle = fh;
370 if (RTM_DELTFILTER != event) {
372 if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
373 goto nla_put_failure;
375 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
384 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
385 struct nlmsghdr *n, struct tcf_proto *tp,
386 unsigned long fh, int event)
389 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
391 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
395 if (tcf_fill_node(net, skb, tp, fh, portid, n->nlmsg_seq, 0, event) <= 0) {
400 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
401 n->nlmsg_flags & NLM_F_ECHO);
404 struct tcf_dump_args {
407 struct netlink_callback *cb;
410 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
411 struct tcf_walker *arg)
413 struct tcf_dump_args *a = (void *)arg;
414 struct net *net = sock_net(a->skb->sk);
416 return tcf_fill_node(net, a->skb, tp, n, NETLINK_CB(a->cb->skb).portid,
417 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
420 /* called with RTNL */
421 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
423 struct net *net = sock_net(skb->sk);
426 struct net_device *dev;
428 struct tcf_proto *tp, __rcu **chain;
429 struct tcmsg *tcm = nlmsg_data(cb->nlh);
430 unsigned long cl = 0;
431 const struct Qdisc_class_ops *cops;
432 struct tcf_dump_args arg;
434 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
436 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
440 if (!tcm->tcm_parent)
443 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
446 cops = q->ops->cl_ops;
449 if (cops->tcf_chain == NULL)
451 if (TC_H_MIN(tcm->tcm_parent)) {
452 cl = cops->get(q, tcm->tcm_parent);
456 chain = cops->tcf_chain(q, cl);
462 for (tp = rtnl_dereference(*chain), t = 0;
463 tp; tp = rtnl_dereference(tp->next), t++) {
466 if (TC_H_MAJ(tcm->tcm_info) &&
467 TC_H_MAJ(tcm->tcm_info) != tp->prio)
469 if (TC_H_MIN(tcm->tcm_info) &&
470 TC_H_MIN(tcm->tcm_info) != tp->protocol)
473 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
474 if (cb->args[1] == 0) {
475 if (tcf_fill_node(net, skb, tp, 0, NETLINK_CB(cb->skb).portid,
476 cb->nlh->nlmsg_seq, NLM_F_MULTI,
477 RTM_NEWTFILTER) <= 0)
482 if (tp->ops->walk == NULL)
484 arg.w.fn = tcf_node_dump;
488 arg.w.skip = cb->args[1] - 1;
490 tp->ops->walk(tp, &arg.w);
491 cb->args[1] = arg.w.count + 1;
505 void tcf_exts_destroy(struct tcf_exts *exts)
507 #ifdef CONFIG_NET_CLS_ACT
508 tcf_action_destroy(&exts->actions, TCA_ACT_UNBIND);
509 INIT_LIST_HEAD(&exts->actions);
512 EXPORT_SYMBOL(tcf_exts_destroy);
514 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
515 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr)
517 #ifdef CONFIG_NET_CLS_ACT
519 struct tc_action *act;
521 INIT_LIST_HEAD(&exts->actions);
522 if (exts->police && tb[exts->police]) {
523 act = tcf_action_init_1(net, tb[exts->police], rate_tlv,
529 act->type = exts->type = TCA_OLD_COMPAT;
530 list_add(&act->list, &exts->actions);
531 } else if (exts->action && tb[exts->action]) {
533 err = tcf_action_init(net, tb[exts->action], rate_tlv,
535 TCA_ACT_BIND, &exts->actions);
541 if ((exts->action && tb[exts->action]) ||
542 (exts->police && tb[exts->police]))
548 EXPORT_SYMBOL(tcf_exts_validate);
550 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
551 struct tcf_exts *src)
553 #ifdef CONFIG_NET_CLS_ACT
556 list_splice_init(&dst->actions, &tmp);
557 list_splice(&src->actions, &dst->actions);
558 dst->type = src->type;
560 tcf_action_destroy(&tmp, TCA_ACT_UNBIND);
563 EXPORT_SYMBOL(tcf_exts_change);
565 #define tcf_exts_first_act(ext) \
566 list_first_entry_or_null(&(exts)->actions, \
567 struct tc_action, list)
569 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
571 #ifdef CONFIG_NET_CLS_ACT
574 if (exts->action && !list_empty(&exts->actions)) {
576 * again for backward compatible mode - we want
577 * to work with both old and new modes of entering
578 * tc data even if iproute2 was newer - jhs
580 if (exts->type != TCA_OLD_COMPAT) {
581 nest = nla_nest_start(skb, exts->action);
583 goto nla_put_failure;
584 if (tcf_action_dump(skb, &exts->actions, 0, 0) < 0)
585 goto nla_put_failure;
586 nla_nest_end(skb, nest);
587 } else if (exts->police) {
588 struct tc_action *act = tcf_exts_first_act(exts);
589 nest = nla_nest_start(skb, exts->police);
590 if (nest == NULL || !act)
591 goto nla_put_failure;
592 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
593 goto nla_put_failure;
594 nla_nest_end(skb, nest);
600 nla_nest_cancel(skb, nest);
606 EXPORT_SYMBOL(tcf_exts_dump);
609 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
611 #ifdef CONFIG_NET_CLS_ACT
612 struct tc_action *a = tcf_exts_first_act(exts);
613 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
618 EXPORT_SYMBOL(tcf_exts_dump_stats);
620 static int __init tc_filter_init(void)
622 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL);
623 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL);
624 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
625 tc_dump_tfilter, NULL);
630 subsys_initcall(tc_filter_init);