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 <net/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 const struct tcf_proto_ops *tcf_proto_lookup_ops(struct nlattr *kind)
45 const 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 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];
125 spinlock_t *root_lock;
131 struct net_device *dev;
133 struct tcf_proto **back, **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) && !capable(CAP_NET_ADMIN))
146 err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
151 protocol = TC_H_MIN(t->tcm_info);
152 prio = TC_H_MAJ(t->tcm_info);
154 parent = t->tcm_parent;
158 /* If no priority is given, user wants we allocated it. */
159 if (n->nlmsg_type != RTM_NEWTFILTER ||
160 !(n->nlmsg_flags & NLM_F_CREATE))
162 prio = TC_H_MAKE(0x80000000U, 0U);
165 /* Find head of filter chain. */
168 dev = __dev_get_by_index(net, t->tcm_ifindex);
177 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
182 /* Is it classful? */
183 cops = q->ops->cl_ops;
187 if (cops->tcf_chain == NULL)
190 /* Do we search for filter, attached to class? */
191 if (TC_H_MIN(parent)) {
192 cl = cops->get(q, parent);
197 /* And the last stroke */
198 chain = cops->tcf_chain(q, cl);
203 /* Check the chain for existence of proto-tcf with this priority */
204 for (back = chain; (tp = *back) != NULL; back = &tp->next) {
205 if (tp->prio >= prio) {
206 if (tp->prio == prio) {
208 (tp->protocol != protocol && protocol))
216 root_lock = qdisc_root_sleeping_lock(q);
219 /* Proto-tcf does not exist, create new one */
221 if (tca[TCA_KIND] == NULL || !protocol)
225 if (n->nlmsg_type != RTM_NEWTFILTER ||
226 !(n->nlmsg_flags & NLM_F_CREATE))
230 /* Create new proto tcf */
233 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
237 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
238 if (tp_ops == NULL) {
239 #ifdef CONFIG_MODULES
240 struct nlattr *kind = tca[TCA_KIND];
244 nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
246 request_module("cls_%s", name);
248 tp_ops = tcf_proto_lookup_ops(kind);
249 /* We dropped the RTNL semaphore in order to
250 * perform the module load. So, even if we
251 * succeeded in loading the module we have to
252 * replay the request. We indicate this using
255 if (tp_ops != NULL) {
256 module_put(tp_ops->owner);
265 tp->protocol = protocol;
266 tp->prio = nprio ? : TC_H_MAJ(tcf_auto_prio(*back));
268 tp->classify = tp_ops->classify;
269 tp->classid = parent;
271 err = tp_ops->init(tp);
273 module_put(tp_ops->owner);
280 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
283 fh = tp->ops->get(tp, t->tcm_handle);
286 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
287 spin_lock_bh(root_lock);
289 spin_unlock_bh(root_lock);
291 tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
298 if (n->nlmsg_type != RTM_NEWTFILTER ||
299 !(n->nlmsg_flags & NLM_F_CREATE))
302 switch (n->nlmsg_type) {
305 if (n->nlmsg_flags & NLM_F_EXCL) {
312 err = tp->ops->delete(tp, fh);
314 tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
317 err = tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
325 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh);
328 spin_lock_bh(root_lock);
331 spin_unlock_bh(root_lock);
333 tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
343 /* Replay the request. */
348 static int tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp,
349 unsigned long fh, u32 portid, u32 seq, u16 flags, int event)
352 struct nlmsghdr *nlh;
353 unsigned char *b = skb_tail_pointer(skb);
355 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
358 tcm = nlmsg_data(nlh);
359 tcm->tcm_family = AF_UNSPEC;
362 tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
363 tcm->tcm_parent = tp->classid;
364 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
365 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
366 goto nla_put_failure;
367 tcm->tcm_handle = fh;
368 if (RTM_DELTFILTER != event) {
370 if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
371 goto nla_put_failure;
373 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
382 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
383 struct nlmsghdr *n, struct tcf_proto *tp,
384 unsigned long fh, int event)
387 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
389 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
393 if (tcf_fill_node(skb, tp, fh, portid, n->nlmsg_seq, 0, event) <= 0) {
398 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
399 n->nlmsg_flags & NLM_F_ECHO);
402 struct tcf_dump_args {
405 struct netlink_callback *cb;
408 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
409 struct tcf_walker *arg)
411 struct tcf_dump_args *a = (void *)arg;
413 return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).portid,
414 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
417 /* called with RTNL */
418 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
420 struct net *net = sock_net(skb->sk);
423 struct net_device *dev;
425 struct tcf_proto *tp, **chain;
426 struct tcmsg *tcm = nlmsg_data(cb->nlh);
427 unsigned long cl = 0;
428 const struct Qdisc_class_ops *cops;
429 struct tcf_dump_args arg;
431 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
433 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
437 if (!tcm->tcm_parent)
440 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
443 cops = q->ops->cl_ops;
446 if (cops->tcf_chain == NULL)
448 if (TC_H_MIN(tcm->tcm_parent)) {
449 cl = cops->get(q, tcm->tcm_parent);
453 chain = cops->tcf_chain(q, cl);
459 for (tp = *chain, t = 0; tp; tp = tp->next, t++) {
462 if (TC_H_MAJ(tcm->tcm_info) &&
463 TC_H_MAJ(tcm->tcm_info) != tp->prio)
465 if (TC_H_MIN(tcm->tcm_info) &&
466 TC_H_MIN(tcm->tcm_info) != tp->protocol)
469 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
470 if (cb->args[1] == 0) {
471 if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).portid,
472 cb->nlh->nlmsg_seq, NLM_F_MULTI,
473 RTM_NEWTFILTER) <= 0)
478 if (tp->ops->walk == NULL)
480 arg.w.fn = tcf_node_dump;
484 arg.w.skip = cb->args[1] - 1;
486 tp->ops->walk(tp, &arg.w);
487 cb->args[1] = arg.w.count + 1;
501 void tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
503 #ifdef CONFIG_NET_CLS_ACT
505 tcf_action_destroy(exts->action, TCA_ACT_UNBIND);
510 EXPORT_SYMBOL(tcf_exts_destroy);
512 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
513 struct nlattr *rate_tlv, struct tcf_exts *exts,
514 const struct tcf_ext_map *map)
516 memset(exts, 0, sizeof(*exts));
518 #ifdef CONFIG_NET_CLS_ACT
520 struct tc_action *act;
522 if (map->police && tb[map->police]) {
523 act = tcf_action_init_1(net, tb[map->police], rate_tlv,
524 "police", TCA_ACT_NOREPLACE,
529 act->type = TCA_OLD_COMPAT;
531 } else if (map->action && tb[map->action]) {
532 act = tcf_action_init(net, tb[map->action], rate_tlv,
533 NULL, TCA_ACT_NOREPLACE,
542 if ((map->action && tb[map->action]) ||
543 (map->police && tb[map->police]))
549 EXPORT_SYMBOL(tcf_exts_validate);
551 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
552 struct tcf_exts *src)
554 #ifdef CONFIG_NET_CLS_ACT
556 struct tc_action *act;
559 dst->action = src->action;
562 tcf_action_destroy(act, TCA_ACT_UNBIND);
566 EXPORT_SYMBOL(tcf_exts_change);
568 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts,
569 const struct tcf_ext_map *map)
571 #ifdef CONFIG_NET_CLS_ACT
572 if (map->action && exts->action) {
574 * again for backward compatible mode - we want
575 * to work with both old and new modes of entering
576 * tc data even if iproute2 was newer - jhs
580 if (exts->action->type != TCA_OLD_COMPAT) {
581 nest = nla_nest_start(skb, map->action);
583 goto nla_put_failure;
584 if (tcf_action_dump(skb, exts->action, 0, 0) < 0)
585 goto nla_put_failure;
586 nla_nest_end(skb, nest);
587 } else if (map->police) {
588 nest = nla_nest_start(skb, map->police);
590 goto nla_put_failure;
591 if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0)
592 goto nla_put_failure;
593 nla_nest_end(skb, nest);
598 nla_put_failure: __attribute__ ((unused))
601 EXPORT_SYMBOL(tcf_exts_dump);
604 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts,
605 const struct tcf_ext_map *map)
607 #ifdef CONFIG_NET_CLS_ACT
609 if (tcf_action_copy_stats(skb, exts->action, 1) < 0)
610 goto nla_put_failure;
613 nla_put_failure: __attribute__ ((unused))
616 EXPORT_SYMBOL(tcf_exts_dump_stats);
618 static int __init tc_filter_init(void)
620 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL);
621 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL);
622 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
623 tc_dump_tfilter, NULL);
628 subsys_initcall(tc_filter_init);