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/netlink.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 */
36 static struct tcf_proto_ops *tcf_proto_base __read_mostly;
38 /* Protects list of registered TC modules. It is pure SMP lock. */
39 static DEFINE_RWLOCK(cls_mod_lock);
41 /* Find classifier type by string name */
43 static struct tcf_proto_ops *tcf_proto_lookup_ops(struct nlattr *kind)
45 struct tcf_proto_ops *t = NULL;
48 read_lock(&cls_mod_lock);
49 for (t = tcf_proto_base; t; t = t->next) {
50 if (nla_strcmp(kind, t->kind) == 0) {
51 if (!try_module_get(t->owner))
56 read_unlock(&cls_mod_lock);
61 /* Register(unregister) new classifier type */
63 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
65 struct tcf_proto_ops *t, **tp;
68 write_lock(&cls_mod_lock);
69 for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
70 if (!strcmp(ops->kind, t->kind))
77 write_unlock(&cls_mod_lock);
80 EXPORT_SYMBOL(register_tcf_proto_ops);
82 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
84 struct tcf_proto_ops *t, **tp;
87 write_lock(&cls_mod_lock);
88 for (tp = &tcf_proto_base; (t=*tp) != NULL; tp = &t->next)
97 write_unlock(&cls_mod_lock);
100 EXPORT_SYMBOL(unregister_tcf_proto_ops);
102 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
103 struct tcf_proto *tp, unsigned long fh, int event);
106 /* Select new prio value from the range, managed by kernel. */
108 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
110 u32 first = TC_H_MAKE(0xC0000000U, 0U);
118 /* Add/change/delete/get a filter node */
120 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
122 struct net *net = sock_net(skb->sk);
123 struct nlattr *tca[TCA_MAX + 1];
124 spinlock_t *root_lock;
130 struct net_device *dev;
132 struct tcf_proto **back, **chain;
133 struct tcf_proto *tp;
134 struct tcf_proto_ops *tp_ops;
135 const struct Qdisc_class_ops *cops;
141 if (!net_eq(net, &init_net))
146 protocol = TC_H_MIN(t->tcm_info);
147 prio = TC_H_MAJ(t->tcm_info);
149 parent = t->tcm_parent;
153 /* If no priority is given, user wants we allocated it. */
154 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
156 prio = TC_H_MAKE(0x80000000U, 0U);
159 /* Find head of filter chain. */
162 dev = __dev_get_by_index(&init_net, t->tcm_ifindex);
166 err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
175 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
180 /* Is it classful? */
181 if ((cops = q->ops->cl_ops) == NULL)
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) {
204 if (!nprio || (tp->protocol != protocol && protocol))
212 root_lock = qdisc_root_sleeping_lock(q);
215 /* Proto-tcf does not exist, create new one */
217 if (tca[TCA_KIND] == NULL || !protocol)
221 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
225 /* Create new proto tcf */
228 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
232 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
233 if (tp_ops == NULL) {
234 #ifdef CONFIG_MODULES
235 struct nlattr *kind = tca[TCA_KIND];
239 nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
241 request_module("cls_%s", name);
243 tp_ops = tcf_proto_lookup_ops(kind);
244 /* We dropped the RTNL semaphore in order to
245 * perform the module load. So, even if we
246 * succeeded in loading the module we have to
247 * replay the request. We indicate this using
250 if (tp_ops != NULL) {
251 module_put(tp_ops->owner);
260 tp->protocol = protocol;
261 tp->prio = nprio ? : TC_H_MAJ(tcf_auto_prio(*back));
263 tp->classify = tp_ops->classify;
264 tp->classid = parent;
266 err = tp_ops->init(tp);
268 module_put(tp_ops->owner);
275 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
278 fh = tp->ops->get(tp, t->tcm_handle);
281 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
282 spin_lock_bh(root_lock);
284 spin_unlock_bh(root_lock);
286 tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
293 if (n->nlmsg_type != RTM_NEWTFILTER ||
294 !(n->nlmsg_flags & NLM_F_CREATE))
297 switch (n->nlmsg_type) {
300 if (n->nlmsg_flags & NLM_F_EXCL) {
307 err = tp->ops->delete(tp, fh);
309 tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
312 err = tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
320 err = tp->ops->change(tp, cl, t->tcm_handle, tca, &fh);
323 spin_lock_bh(root_lock);
326 spin_unlock_bh(root_lock);
328 tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
338 /* Replay the request. */
343 static int tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp,
344 unsigned long fh, u32 pid, u32 seq, u16 flags, int event)
347 struct nlmsghdr *nlh;
348 unsigned char *b = skb_tail_pointer(skb);
350 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
351 tcm = NLMSG_DATA(nlh);
352 tcm->tcm_family = AF_UNSPEC;
355 tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
356 tcm->tcm_parent = tp->classid;
357 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
358 NLA_PUT_STRING(skb, TCA_KIND, tp->ops->kind);
359 tcm->tcm_handle = fh;
360 if (RTM_DELTFILTER != event) {
362 if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
363 goto nla_put_failure;
365 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
374 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
375 struct tcf_proto *tp, unsigned long fh, int event)
378 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
380 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
384 if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) {
389 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC,
390 n->nlmsg_flags & NLM_F_ECHO);
393 struct tcf_dump_args {
396 struct netlink_callback *cb;
399 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
400 struct tcf_walker *arg)
402 struct tcf_dump_args *a = (void *)arg;
404 return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid,
405 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
408 /* called with RTNL */
409 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
411 struct net *net = sock_net(skb->sk);
414 struct net_device *dev;
416 struct tcf_proto *tp, **chain;
417 struct tcmsg *tcm = (struct tcmsg *)NLMSG_DATA(cb->nlh);
418 unsigned long cl = 0;
419 const struct Qdisc_class_ops *cops;
420 struct tcf_dump_args arg;
422 if (!net_eq(net, &init_net))
425 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
427 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
430 if (!tcm->tcm_parent)
433 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
436 if ((cops = q->ops->cl_ops) == NULL)
438 if (cops->tcf_chain == NULL)
440 if (TC_H_MIN(tcm->tcm_parent)) {
441 cl = cops->get(q, tcm->tcm_parent);
445 chain = cops->tcf_chain(q, cl);
451 for (tp=*chain, t=0; tp; tp = tp->next, t++) {
452 if (t < s_t) continue;
453 if (TC_H_MAJ(tcm->tcm_info) &&
454 TC_H_MAJ(tcm->tcm_info) != tp->prio)
456 if (TC_H_MIN(tcm->tcm_info) &&
457 TC_H_MIN(tcm->tcm_info) != tp->protocol)
460 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
461 if (cb->args[1] == 0) {
462 if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid,
463 cb->nlh->nlmsg_seq, NLM_F_MULTI,
464 RTM_NEWTFILTER) <= 0)
469 if (tp->ops->walk == NULL)
471 arg.w.fn = tcf_node_dump;
475 arg.w.skip = cb->args[1]-1;
477 tp->ops->walk(tp, &arg.w);
478 cb->args[1] = arg.w.count+1;
492 void tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
494 #ifdef CONFIG_NET_CLS_ACT
496 tcf_action_destroy(exts->action, TCA_ACT_UNBIND);
501 EXPORT_SYMBOL(tcf_exts_destroy);
503 int tcf_exts_validate(struct tcf_proto *tp, struct nlattr **tb,
504 struct nlattr *rate_tlv, struct tcf_exts *exts,
505 const struct tcf_ext_map *map)
507 memset(exts, 0, sizeof(*exts));
509 #ifdef CONFIG_NET_CLS_ACT
511 struct tc_action *act;
513 if (map->police && tb[map->police]) {
514 act = tcf_action_init_1(tb[map->police], rate_tlv,
515 "police", TCA_ACT_NOREPLACE,
520 act->type = TCA_OLD_COMPAT;
522 } else if (map->action && tb[map->action]) {
523 act = tcf_action_init(tb[map->action], rate_tlv, NULL,
524 TCA_ACT_NOREPLACE, TCA_ACT_BIND);
532 if ((map->action && tb[map->action]) ||
533 (map->police && tb[map->police]))
539 EXPORT_SYMBOL(tcf_exts_validate);
541 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
542 struct tcf_exts *src)
544 #ifdef CONFIG_NET_CLS_ACT
546 struct tc_action *act;
549 dst->action = src->action;
552 tcf_action_destroy(act, TCA_ACT_UNBIND);
556 EXPORT_SYMBOL(tcf_exts_change);
558 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts,
559 const struct tcf_ext_map *map)
561 #ifdef CONFIG_NET_CLS_ACT
562 if (map->action && exts->action) {
564 * again for backward compatible mode - we want
565 * to work with both old and new modes of entering
566 * tc data even if iproute2 was newer - jhs
570 if (exts->action->type != TCA_OLD_COMPAT) {
571 nest = nla_nest_start(skb, map->action);
573 goto nla_put_failure;
574 if (tcf_action_dump(skb, exts->action, 0, 0) < 0)
575 goto nla_put_failure;
576 nla_nest_end(skb, nest);
577 } else if (map->police) {
578 nest = nla_nest_start(skb, map->police);
580 goto nla_put_failure;
581 if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0)
582 goto nla_put_failure;
583 nla_nest_end(skb, nest);
588 nla_put_failure: __attribute__ ((unused))
591 EXPORT_SYMBOL(tcf_exts_dump);
594 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts,
595 const struct tcf_ext_map *map)
597 #ifdef CONFIG_NET_CLS_ACT
599 if (tcf_action_copy_stats(skb, exts->action, 1) < 0)
600 goto nla_put_failure;
603 nla_put_failure: __attribute__ ((unused))
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);
611 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL);
612 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
618 subsys_initcall(tc_filter_init);