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);
107 /* Select new prio value from the range, managed by kernel. */
109 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
111 u32 first = TC_H_MAKE(0xC0000000U, 0U);
114 first = tp->prio - 1;
119 /* Add/change/delete/get a filter node */
121 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n)
123 struct net *net = sock_net(skb->sk);
124 struct nlattr *tca[TCA_MAX + 1];
130 struct net_device *dev;
132 struct tcf_proto __rcu **back;
133 struct tcf_proto __rcu **chain;
134 struct tcf_proto *tp;
135 const struct tcf_proto_ops *tp_ops;
136 const struct Qdisc_class_ops *cops;
142 if ((n->nlmsg_type != RTM_GETTFILTER) &&
143 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
149 err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
154 protocol = TC_H_MIN(t->tcm_info);
155 prio = TC_H_MAJ(t->tcm_info);
157 parent = t->tcm_parent;
161 /* If no priority is given, user wants we allocated it. */
162 if (n->nlmsg_type != RTM_NEWTFILTER ||
163 !(n->nlmsg_flags & NLM_F_CREATE))
165 prio = TC_H_MAKE(0x80000000U, 0U);
168 /* Find head of filter chain. */
171 dev = __dev_get_by_index(net, t->tcm_ifindex);
180 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
185 /* Is it classful? */
186 cops = q->ops->cl_ops;
190 if (cops->tcf_chain == NULL)
193 /* Do we search for filter, attached to class? */
194 if (TC_H_MIN(parent)) {
195 cl = cops->get(q, parent);
200 /* And the last stroke */
201 chain = cops->tcf_chain(q, cl);
206 /* Check the chain for existence of proto-tcf with this priority */
208 (tp = rtnl_dereference(*back)) != NULL;
210 if (tp->prio >= prio) {
211 if (tp->prio == prio) {
213 (tp->protocol != protocol && protocol))
222 /* Proto-tcf does not exist, create new one */
224 if (tca[TCA_KIND] == NULL || !protocol)
228 if (n->nlmsg_type != RTM_NEWTFILTER ||
229 !(n->nlmsg_flags & NLM_F_CREATE))
233 /* Create new proto tcf */
236 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
240 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
241 if (tp_ops == NULL) {
242 #ifdef CONFIG_MODULES
243 struct nlattr *kind = tca[TCA_KIND];
247 nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
249 request_module("cls_%s", name);
251 tp_ops = tcf_proto_lookup_ops(kind);
252 /* We dropped the RTNL semaphore in order to
253 * perform the module load. So, even if we
254 * succeeded in loading the module we have to
255 * replay the request. We indicate this using
258 if (tp_ops != NULL) {
259 module_put(tp_ops->owner);
268 tp->protocol = protocol;
270 TC_H_MAJ(tcf_auto_prio(rtnl_dereference(*back)));
272 tp->classify = tp_ops->classify;
273 tp->classid = parent;
275 err = tp_ops->init(tp);
277 module_put(tp_ops->owner);
284 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
287 fh = tp->ops->get(tp, t->tcm_handle);
290 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
291 struct tcf_proto *next = rtnl_dereference(tp->next);
293 RCU_INIT_POINTER(*back, next);
295 tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
296 tcf_destroy(tp, true);
302 if (n->nlmsg_type != RTM_NEWTFILTER ||
303 !(n->nlmsg_flags & NLM_F_CREATE))
306 switch (n->nlmsg_type) {
309 if (n->nlmsg_flags & NLM_F_EXCL) {
311 tcf_destroy(tp, true);
316 err = tp->ops->delete(tp, fh);
318 struct tcf_proto *next = rtnl_dereference(tp->next);
320 tfilter_notify(net, skb, n, tp,
321 t->tcm_handle, RTM_DELTFILTER);
322 if (tcf_destroy(tp, false))
323 RCU_INIT_POINTER(*back, next);
327 err = tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
335 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
336 n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE);
339 RCU_INIT_POINTER(tp->next, rtnl_dereference(*back));
340 rcu_assign_pointer(*back, tp);
342 tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
345 tcf_destroy(tp, true);
352 /* Replay the request. */
357 static int tcf_fill_node(struct net *net, struct sk_buff *skb, struct tcf_proto *tp,
358 unsigned long fh, u32 portid, u32 seq, u16 flags, int event)
361 struct nlmsghdr *nlh;
362 unsigned char *b = skb_tail_pointer(skb);
364 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
367 tcm = nlmsg_data(nlh);
368 tcm->tcm_family = AF_UNSPEC;
371 tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
372 tcm->tcm_parent = tp->classid;
373 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
374 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
375 goto nla_put_failure;
376 tcm->tcm_handle = fh;
377 if (RTM_DELTFILTER != event) {
379 if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
380 goto nla_put_failure;
382 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
391 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
392 struct nlmsghdr *n, struct tcf_proto *tp,
393 unsigned long fh, int event)
396 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
398 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
402 if (tcf_fill_node(net, skb, tp, fh, portid, n->nlmsg_seq, 0, event) <= 0) {
407 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
408 n->nlmsg_flags & NLM_F_ECHO);
411 struct tcf_dump_args {
414 struct netlink_callback *cb;
417 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
418 struct tcf_walker *arg)
420 struct tcf_dump_args *a = (void *)arg;
421 struct net *net = sock_net(a->skb->sk);
423 return tcf_fill_node(net, a->skb, tp, n, NETLINK_CB(a->cb->skb).portid,
424 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
427 /* called with RTNL */
428 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
430 struct net *net = sock_net(skb->sk);
433 struct net_device *dev;
435 struct tcf_proto *tp, __rcu **chain;
436 struct tcmsg *tcm = nlmsg_data(cb->nlh);
437 unsigned long cl = 0;
438 const struct Qdisc_class_ops *cops;
439 struct tcf_dump_args arg;
441 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
443 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
447 if (!tcm->tcm_parent)
450 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
453 cops = q->ops->cl_ops;
456 if (cops->tcf_chain == NULL)
458 if (TC_H_MIN(tcm->tcm_parent)) {
459 cl = cops->get(q, tcm->tcm_parent);
463 chain = cops->tcf_chain(q, cl);
469 for (tp = rtnl_dereference(*chain), t = 0;
470 tp; tp = rtnl_dereference(tp->next), t++) {
473 if (TC_H_MAJ(tcm->tcm_info) &&
474 TC_H_MAJ(tcm->tcm_info) != tp->prio)
476 if (TC_H_MIN(tcm->tcm_info) &&
477 TC_H_MIN(tcm->tcm_info) != tp->protocol)
480 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
481 if (cb->args[1] == 0) {
482 if (tcf_fill_node(net, skb, tp, 0, NETLINK_CB(cb->skb).portid,
483 cb->nlh->nlmsg_seq, NLM_F_MULTI,
484 RTM_NEWTFILTER) <= 0)
489 if (tp->ops->walk == NULL)
491 arg.w.fn = tcf_node_dump;
495 arg.w.skip = cb->args[1] - 1;
497 tp->ops->walk(tp, &arg.w);
498 cb->args[1] = arg.w.count + 1;
512 void tcf_exts_destroy(struct tcf_exts *exts)
514 #ifdef CONFIG_NET_CLS_ACT
515 tcf_action_destroy(&exts->actions, TCA_ACT_UNBIND);
516 INIT_LIST_HEAD(&exts->actions);
519 EXPORT_SYMBOL(tcf_exts_destroy);
521 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
522 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr)
524 #ifdef CONFIG_NET_CLS_ACT
526 struct tc_action *act;
528 INIT_LIST_HEAD(&exts->actions);
529 if (exts->police && tb[exts->police]) {
530 act = tcf_action_init_1(net, tb[exts->police], rate_tlv,
536 act->type = exts->type = TCA_OLD_COMPAT;
537 list_add(&act->list, &exts->actions);
538 } else if (exts->action && tb[exts->action]) {
540 err = tcf_action_init(net, tb[exts->action], rate_tlv,
542 TCA_ACT_BIND, &exts->actions);
548 if ((exts->action && tb[exts->action]) ||
549 (exts->police && tb[exts->police]))
555 EXPORT_SYMBOL(tcf_exts_validate);
557 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
558 struct tcf_exts *src)
560 #ifdef CONFIG_NET_CLS_ACT
563 list_splice_init(&dst->actions, &tmp);
564 list_splice(&src->actions, &dst->actions);
565 dst->type = src->type;
567 tcf_action_destroy(&tmp, TCA_ACT_UNBIND);
570 EXPORT_SYMBOL(tcf_exts_change);
572 #define tcf_exts_first_act(ext) \
573 list_first_entry_or_null(&(exts)->actions, \
574 struct tc_action, list)
576 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
578 #ifdef CONFIG_NET_CLS_ACT
581 if (exts->action && !list_empty(&exts->actions)) {
583 * again for backward compatible mode - we want
584 * to work with both old and new modes of entering
585 * tc data even if iproute2 was newer - jhs
587 if (exts->type != TCA_OLD_COMPAT) {
588 nest = nla_nest_start(skb, exts->action);
590 goto nla_put_failure;
591 if (tcf_action_dump(skb, &exts->actions, 0, 0) < 0)
592 goto nla_put_failure;
593 nla_nest_end(skb, nest);
594 } else if (exts->police) {
595 struct tc_action *act = tcf_exts_first_act(exts);
596 nest = nla_nest_start(skb, exts->police);
597 if (nest == NULL || !act)
598 goto nla_put_failure;
599 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
600 goto nla_put_failure;
601 nla_nest_end(skb, nest);
607 nla_nest_cancel(skb, nest);
613 EXPORT_SYMBOL(tcf_exts_dump);
616 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
618 #ifdef CONFIG_NET_CLS_ACT
619 struct tc_action *a = tcf_exts_first_act(exts);
620 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
625 EXPORT_SYMBOL(tcf_exts_dump_stats);
627 static int __init tc_filter_init(void)
629 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL);
630 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL);
631 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
632 tc_dump_tfilter, NULL);
637 subsys_initcall(tc_filter_init);