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 = 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)
94 write_unlock(&cls_mod_lock);
97 EXPORT_SYMBOL(unregister_tcf_proto_ops);
99 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
100 struct nlmsghdr *n, struct tcf_proto *tp,
101 unsigned long fh, int event);
104 /* Select new prio value from the range, managed by kernel. */
106 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
108 u32 first = TC_H_MAKE(0xC0000000U, 0U);
111 first = tp->prio - 1;
116 /* Add/change/delete/get a filter node */
118 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n)
120 struct net *net = sock_net(skb->sk);
121 struct nlattr *tca[TCA_MAX + 1];
122 spinlock_t *root_lock;
128 struct net_device *dev;
130 struct tcf_proto **back, **chain;
131 struct tcf_proto *tp;
132 const struct tcf_proto_ops *tp_ops;
133 const struct Qdisc_class_ops *cops;
139 if ((n->nlmsg_type != RTM_GETTFILTER) && !capable(CAP_NET_ADMIN))
143 err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
148 protocol = TC_H_MIN(t->tcm_info);
149 prio = TC_H_MAJ(t->tcm_info);
151 parent = t->tcm_parent;
155 /* If no priority is given, user wants we allocated it. */
156 if (n->nlmsg_type != RTM_NEWTFILTER ||
157 !(n->nlmsg_flags & NLM_F_CREATE))
159 prio = TC_H_MAKE(0x80000000U, 0U);
162 /* Find head of filter chain. */
165 dev = __dev_get_by_index(net, t->tcm_ifindex);
174 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
179 /* Is it classful? */
180 cops = q->ops->cl_ops;
184 if (cops->tcf_chain == NULL)
187 /* Do we search for filter, attached to class? */
188 if (TC_H_MIN(parent)) {
189 cl = cops->get(q, parent);
194 /* And the last stroke */
195 chain = cops->tcf_chain(q, cl);
200 /* Check the chain for existence of proto-tcf with this priority */
201 for (back = chain; (tp = *back) != NULL; back = &tp->next) {
202 if (tp->prio >= prio) {
203 if (tp->prio == prio) {
205 (tp->protocol != protocol && protocol))
213 root_lock = qdisc_root_sleeping_lock(q);
216 /* Proto-tcf does not exist, create new one */
218 if (tca[TCA_KIND] == NULL || !protocol)
222 if (n->nlmsg_type != RTM_NEWTFILTER ||
223 !(n->nlmsg_flags & NLM_F_CREATE))
227 /* Create new proto tcf */
230 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
234 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
235 if (tp_ops == NULL) {
236 #ifdef CONFIG_MODULES
237 struct nlattr *kind = tca[TCA_KIND];
241 nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
243 request_module("cls_%s", name);
245 tp_ops = tcf_proto_lookup_ops(kind);
246 /* We dropped the RTNL semaphore in order to
247 * perform the module load. So, even if we
248 * succeeded in loading the module we have to
249 * replay the request. We indicate this using
252 if (tp_ops != NULL) {
253 module_put(tp_ops->owner);
262 tp->protocol = protocol;
263 tp->prio = nprio ? : TC_H_MAJ(tcf_auto_prio(*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 spin_lock_bh(root_lock);
286 spin_unlock_bh(root_lock);
288 tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
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) {
309 err = tp->ops->delete(tp, fh);
311 tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
314 err = tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
322 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh);
325 spin_lock_bh(root_lock);
328 spin_unlock_bh(root_lock);
330 tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
340 /* Replay the request. */
345 static int tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp,
346 unsigned long fh, u32 portid, u32 seq, u16 flags, int event)
349 struct nlmsghdr *nlh;
350 unsigned char *b = skb_tail_pointer(skb);
352 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
355 tcm = nlmsg_data(nlh);
356 tcm->tcm_family = AF_UNSPEC;
359 tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
360 tcm->tcm_parent = tp->classid;
361 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
362 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
363 goto nla_put_failure;
364 tcm->tcm_handle = fh;
365 if (RTM_DELTFILTER != event) {
367 if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
368 goto nla_put_failure;
370 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
379 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
380 struct nlmsghdr *n, struct tcf_proto *tp,
381 unsigned long fh, int event)
384 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
386 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
390 if (tcf_fill_node(skb, tp, fh, portid, n->nlmsg_seq, 0, event) <= 0) {
395 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
396 n->nlmsg_flags & NLM_F_ECHO);
399 struct tcf_dump_args {
402 struct netlink_callback *cb;
405 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
406 struct tcf_walker *arg)
408 struct tcf_dump_args *a = (void *)arg;
410 return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).portid,
411 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
414 /* called with RTNL */
415 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
417 struct net *net = sock_net(skb->sk);
420 struct net_device *dev;
422 struct tcf_proto *tp, **chain;
423 struct tcmsg *tcm = nlmsg_data(cb->nlh);
424 unsigned long cl = 0;
425 const struct Qdisc_class_ops *cops;
426 struct tcf_dump_args arg;
428 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
430 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
434 if (!tcm->tcm_parent)
437 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
440 cops = q->ops->cl_ops;
443 if (cops->tcf_chain == NULL)
445 if (TC_H_MIN(tcm->tcm_parent)) {
446 cl = cops->get(q, tcm->tcm_parent);
450 chain = cops->tcf_chain(q, cl);
456 for (tp = *chain, t = 0; tp; tp = tp->next, t++) {
459 if (TC_H_MAJ(tcm->tcm_info) &&
460 TC_H_MAJ(tcm->tcm_info) != tp->prio)
462 if (TC_H_MIN(tcm->tcm_info) &&
463 TC_H_MIN(tcm->tcm_info) != tp->protocol)
466 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
467 if (cb->args[1] == 0) {
468 if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).portid,
469 cb->nlh->nlmsg_seq, NLM_F_MULTI,
470 RTM_NEWTFILTER) <= 0)
475 if (tp->ops->walk == NULL)
477 arg.w.fn = tcf_node_dump;
481 arg.w.skip = cb->args[1] - 1;
483 tp->ops->walk(tp, &arg.w);
484 cb->args[1] = arg.w.count + 1;
498 void tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
500 #ifdef CONFIG_NET_CLS_ACT
501 tcf_action_destroy(&exts->actions, TCA_ACT_UNBIND);
502 INIT_LIST_HEAD(&exts->actions);
505 EXPORT_SYMBOL(tcf_exts_destroy);
507 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
508 struct nlattr *rate_tlv, struct tcf_exts *exts)
510 #ifdef CONFIG_NET_CLS_ACT
512 struct tc_action *act;
514 INIT_LIST_HEAD(&exts->actions);
515 if (exts->police && tb[exts->police]) {
516 act = tcf_action_init_1(net, tb[exts->police], rate_tlv,
517 "police", TCA_ACT_NOREPLACE,
522 act->type = exts->type = TCA_OLD_COMPAT;
523 list_add(&act->list, &exts->actions);
524 } else if (exts->action && tb[exts->action]) {
526 err = tcf_action_init(net, tb[exts->action], rate_tlv,
527 NULL, TCA_ACT_NOREPLACE,
528 TCA_ACT_BIND, &exts->actions);
534 if ((exts->action && tb[exts->action]) ||
535 (exts->police && tb[exts->police]))
541 EXPORT_SYMBOL(tcf_exts_validate);
543 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
544 struct tcf_exts *src)
546 #ifdef CONFIG_NET_CLS_ACT
547 if (!list_empty(&src->actions)) {
550 list_splice_init(&dst->actions, &tmp);
551 list_splice(&src->actions, &dst->actions);
553 tcf_action_destroy(&tmp, TCA_ACT_UNBIND);
557 EXPORT_SYMBOL(tcf_exts_change);
559 #define tcf_exts_first_act(ext) \
560 list_first_entry(&(exts)->actions, struct tc_action, list)
562 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
564 #ifdef CONFIG_NET_CLS_ACT
565 if (exts->action && !list_empty(&exts->actions)) {
567 * again for backward compatible mode - we want
568 * to work with both old and new modes of entering
569 * tc data even if iproute2 was newer - jhs
572 if (exts->type != TCA_OLD_COMPAT) {
573 nest = nla_nest_start(skb, exts->action);
575 goto nla_put_failure;
576 if (tcf_action_dump(skb, &exts->actions, 0, 0) < 0)
577 goto nla_put_failure;
578 nla_nest_end(skb, nest);
579 } else if (exts->police) {
580 struct tc_action *act = tcf_exts_first_act(exts);
581 nest = nla_nest_start(skb, exts->police);
583 goto nla_put_failure;
584 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
585 goto nla_put_failure;
586 nla_nest_end(skb, nest);
591 nla_put_failure: __attribute__ ((unused))
594 EXPORT_SYMBOL(tcf_exts_dump);
597 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
599 #ifdef CONFIG_NET_CLS_ACT
600 struct tc_action *a = tcf_exts_first_act(exts);
601 if (tcf_action_copy_stats(skb, a, 1) < 0)
606 EXPORT_SYMBOL(tcf_exts_dump_stats);
608 static int __init tc_filter_init(void)
610 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL);
611 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL);
612 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
613 tc_dump_tfilter, NULL);
618 subsys_initcall(tc_filter_init);