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 <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 */
35 static struct tcf_proto_ops *tcf_proto_base __read_mostly;
37 /* Protects list of registered TC modules. It is pure SMP lock. */
38 static DEFINE_RWLOCK(cls_mod_lock);
40 /* Find classifier type by string name */
42 static struct tcf_proto_ops *tcf_proto_lookup_ops(struct nlattr *kind)
44 struct tcf_proto_ops *t = NULL;
47 read_lock(&cls_mod_lock);
48 for (t = tcf_proto_base; t; t = t->next) {
49 if (nla_strcmp(kind, t->kind) == 0) {
50 if (!try_module_get(t->owner))
55 read_unlock(&cls_mod_lock);
60 /* Register(unregister) new classifier type */
62 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
64 struct tcf_proto_ops *t, **tp;
67 write_lock(&cls_mod_lock);
68 for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
69 if (!strcmp(ops->kind, t->kind))
76 write_unlock(&cls_mod_lock);
79 EXPORT_SYMBOL(register_tcf_proto_ops);
81 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
83 struct tcf_proto_ops *t, **tp;
86 write_lock(&cls_mod_lock);
87 for (tp = &tcf_proto_base; (t=*tp) != NULL; tp = &t->next)
96 write_unlock(&cls_mod_lock);
99 EXPORT_SYMBOL(unregister_tcf_proto_ops);
101 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
102 struct tcf_proto *tp, unsigned long fh, int event);
105 /* Select new prio value from the range, managed by kernel. */
107 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
109 u32 first = TC_H_MAKE(0xC0000000U, 0U);
117 /* Add/change/delete/get a filter node */
119 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
121 struct net *net = sock_net(skb->sk);
122 struct nlattr *tca[TCA_MAX + 1];
128 struct net_device *dev;
130 struct tcf_proto **back, **chain;
131 struct tcf_proto *tp;
132 struct tcf_proto_ops *tp_ops;
133 const struct Qdisc_class_ops *cops;
138 if (net != &init_net)
143 protocol = TC_H_MIN(t->tcm_info);
144 prio = TC_H_MAJ(t->tcm_info);
146 parent = t->tcm_parent;
150 /* If no priority is given, user wants we allocated it. */
151 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
153 prio = TC_H_MAKE(0x80000000U, 0U);
156 /* Find head of filter chain. */
159 dev = __dev_get_by_index(&init_net, t->tcm_ifindex);
163 err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
169 q = dev->qdisc_sleeping;
172 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
177 /* Is it classful? */
178 if ((cops = q->ops->cl_ops) == NULL)
181 /* Do we search for filter, attached to class? */
182 if (TC_H_MIN(parent)) {
183 cl = cops->get(q, parent);
188 /* And the last stroke */
189 chain = cops->tcf_chain(q, cl);
194 /* Check the chain for existence of proto-tcf with this priority */
195 for (back = chain; (tp=*back) != NULL; back = &tp->next) {
196 if (tp->prio >= prio) {
197 if (tp->prio == prio) {
198 if (!nprio || (tp->protocol != protocol && protocol))
207 /* Proto-tcf does not exist, create new one */
209 if (tca[TCA_KIND] == NULL || !protocol)
213 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
217 /* Create new proto tcf */
220 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
224 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
225 if (tp_ops == NULL) {
227 struct nlattr *kind = tca[TCA_KIND];
231 nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
233 request_module("cls_%s", name);
235 tp_ops = tcf_proto_lookup_ops(kind);
236 /* We dropped the RTNL semaphore in order to
237 * perform the module load. So, even if we
238 * succeeded in loading the module we have to
239 * replay the request. We indicate this using
242 if (tp_ops != NULL) {
243 module_put(tp_ops->owner);
252 tp->protocol = protocol;
253 tp->prio = nprio ? : tcf_auto_prio(*back);
255 tp->classify = tp_ops->classify;
256 tp->classid = parent;
258 err = tp_ops->init(tp);
260 module_put(tp_ops->owner);
265 qdisc_lock_tree(dev);
268 qdisc_unlock_tree(dev);
270 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
273 fh = tp->ops->get(tp, t->tcm_handle);
276 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
277 qdisc_lock_tree(dev);
279 qdisc_unlock_tree(dev);
281 tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
288 if (n->nlmsg_type != RTM_NEWTFILTER ||
289 !(n->nlmsg_flags & NLM_F_CREATE))
292 switch (n->nlmsg_type) {
295 if (n->nlmsg_flags & NLM_F_EXCL)
299 err = tp->ops->delete(tp, fh);
301 tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
304 err = tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
312 err = tp->ops->change(tp, cl, t->tcm_handle, tca, &fh);
314 tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
320 /* Replay the request. */
325 static int tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp,
326 unsigned long fh, u32 pid, u32 seq, u16 flags, int event)
329 struct nlmsghdr *nlh;
330 unsigned char *b = skb_tail_pointer(skb);
332 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
333 tcm = NLMSG_DATA(nlh);
334 tcm->tcm_family = AF_UNSPEC;
337 tcm->tcm_ifindex = tp->q->dev->ifindex;
338 tcm->tcm_parent = tp->classid;
339 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
340 NLA_PUT_STRING(skb, TCA_KIND, tp->ops->kind);
341 tcm->tcm_handle = fh;
342 if (RTM_DELTFILTER != event) {
344 if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
345 goto nla_put_failure;
347 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
356 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
357 struct tcf_proto *tp, unsigned long fh, int event)
360 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
362 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
366 if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) {
371 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC,
372 n->nlmsg_flags & NLM_F_ECHO);
375 struct tcf_dump_args {
378 struct netlink_callback *cb;
381 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
382 struct tcf_walker *arg)
384 struct tcf_dump_args *a = (void *)arg;
386 return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid,
387 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
390 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
392 struct net *net = sock_net(skb->sk);
395 struct net_device *dev;
397 struct tcf_proto *tp, **chain;
398 struct tcmsg *tcm = (struct tcmsg *)NLMSG_DATA(cb->nlh);
399 unsigned long cl = 0;
400 const struct Qdisc_class_ops *cops;
401 struct tcf_dump_args arg;
403 if (net != &init_net)
406 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
408 if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
411 if (!tcm->tcm_parent)
412 q = dev->qdisc_sleeping;
414 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
417 if ((cops = q->ops->cl_ops) == NULL)
419 if (TC_H_MIN(tcm->tcm_parent)) {
420 cl = cops->get(q, tcm->tcm_parent);
424 chain = cops->tcf_chain(q, cl);
430 for (tp=*chain, t=0; tp; tp = tp->next, t++) {
431 if (t < s_t) continue;
432 if (TC_H_MAJ(tcm->tcm_info) &&
433 TC_H_MAJ(tcm->tcm_info) != tp->prio)
435 if (TC_H_MIN(tcm->tcm_info) &&
436 TC_H_MIN(tcm->tcm_info) != tp->protocol)
439 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
440 if (cb->args[1] == 0) {
441 if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid,
442 cb->nlh->nlmsg_seq, NLM_F_MULTI,
443 RTM_NEWTFILTER) <= 0)
448 if (tp->ops->walk == NULL)
450 arg.w.fn = tcf_node_dump;
454 arg.w.skip = cb->args[1]-1;
456 tp->ops->walk(tp, &arg.w);
457 cb->args[1] = arg.w.count+1;
472 void tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
474 #ifdef CONFIG_NET_CLS_ACT
476 tcf_action_destroy(exts->action, TCA_ACT_UNBIND);
481 EXPORT_SYMBOL(tcf_exts_destroy);
483 int tcf_exts_validate(struct tcf_proto *tp, struct nlattr **tb,
484 struct nlattr *rate_tlv, struct tcf_exts *exts,
485 const struct tcf_ext_map *map)
487 memset(exts, 0, sizeof(*exts));
489 #ifdef CONFIG_NET_CLS_ACT
491 struct tc_action *act;
493 if (map->police && tb[map->police]) {
494 act = tcf_action_init_1(tb[map->police], rate_tlv,
495 "police", TCA_ACT_NOREPLACE,
500 act->type = TCA_OLD_COMPAT;
502 } else if (map->action && tb[map->action]) {
503 act = tcf_action_init(tb[map->action], rate_tlv, NULL,
504 TCA_ACT_NOREPLACE, TCA_ACT_BIND);
512 if ((map->action && tb[map->action]) ||
513 (map->police && tb[map->police]))
519 EXPORT_SYMBOL(tcf_exts_validate);
521 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
522 struct tcf_exts *src)
524 #ifdef CONFIG_NET_CLS_ACT
526 struct tc_action *act;
528 act = xchg(&dst->action, src->action);
531 tcf_action_destroy(act, TCA_ACT_UNBIND);
535 EXPORT_SYMBOL(tcf_exts_change);
537 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts,
538 const struct tcf_ext_map *map)
540 #ifdef CONFIG_NET_CLS_ACT
541 if (map->action && exts->action) {
543 * again for backward compatible mode - we want
544 * to work with both old and new modes of entering
545 * tc data even if iproute2 was newer - jhs
549 if (exts->action->type != TCA_OLD_COMPAT) {
550 nest = nla_nest_start(skb, map->action);
552 goto nla_put_failure;
553 if (tcf_action_dump(skb, exts->action, 0, 0) < 0)
554 goto nla_put_failure;
555 nla_nest_end(skb, nest);
556 } else if (map->police) {
557 nest = nla_nest_start(skb, map->police);
559 goto nla_put_failure;
560 if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0)
561 goto nla_put_failure;
562 nla_nest_end(skb, nest);
567 nla_put_failure: __attribute__ ((unused))
570 EXPORT_SYMBOL(tcf_exts_dump);
573 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts,
574 const struct tcf_ext_map *map)
576 #ifdef CONFIG_NET_CLS_ACT
578 if (tcf_action_copy_stats(skb, exts->action, 1) < 0)
579 goto nla_put_failure;
582 nla_put_failure: __attribute__ ((unused))
585 EXPORT_SYMBOL(tcf_exts_dump_stats);
587 static int __init tc_filter_init(void)
589 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL);
590 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL);
591 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
597 subsys_initcall(tc_filter_init);