2 * net/sched/act_api.c Packet action 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 * Author: Jamal Hadi Salim
14 #include <linux/types.h>
15 #include <linux/kernel.h>
16 #include <linux/string.h>
17 #include <linux/errno.h>
18 #include <linux/slab.h>
19 #include <linux/skbuff.h>
20 #include <linux/init.h>
21 #include <linux/kmod.h>
22 #include <linux/err.h>
23 #include <linux/module.h>
24 #include <net/net_namespace.h>
26 #include <net/sch_generic.h>
27 #include <net/act_api.h>
28 #include <net/netlink.h>
30 void tcf_hash_destroy(struct tcf_common *p, struct tcf_hashinfo *hinfo)
32 spin_lock_bh(&hinfo->lock);
33 hlist_del(&p->tcfc_head);
34 spin_unlock_bh(&hinfo->lock);
35 gen_kill_estimator(&p->tcfc_bstats,
38 * gen_estimator est_timer() might access p->tcfc_lock
39 * or bstats, wait a RCU grace period before freeing p
41 kfree_rcu(p, tcfc_rcu);
43 EXPORT_SYMBOL(tcf_hash_destroy);
45 int tcf_hash_release(struct tcf_common *p, int bind,
46 struct tcf_hashinfo *hinfo)
55 if (p->tcfc_bindcnt <= 0 && p->tcfc_refcnt <= 0) {
56 tcf_hash_destroy(p, hinfo);
62 EXPORT_SYMBOL(tcf_hash_release);
64 static int tcf_dump_walker(struct sk_buff *skb, struct netlink_callback *cb,
65 struct tc_action *a, struct tcf_hashinfo *hinfo)
67 struct hlist_head *head;
69 int err = 0, index = -1, i = 0, s_i = 0, n_i = 0;
72 spin_lock_bh(&hinfo->lock);
76 for (i = 0; i < (hinfo->hmask + 1); i++) {
77 head = &hinfo->htab[tcf_hash(i, hinfo->hmask)];
79 hlist_for_each_entry_rcu(p, head, tcfc_head) {
86 nest = nla_nest_start(skb, a->order);
89 err = tcf_action_dump_1(skb, a, 0, 0);
92 nlmsg_trim(skb, nest);
95 nla_nest_end(skb, nest);
97 if (n_i >= TCA_ACT_MAX_PRIO)
102 spin_unlock_bh(&hinfo->lock);
108 nla_nest_cancel(skb, nest);
112 static int tcf_del_walker(struct sk_buff *skb, struct tc_action *a,
113 struct tcf_hashinfo *hinfo)
115 struct hlist_head *head;
116 struct hlist_node *n;
117 struct tcf_common *p;
121 nest = nla_nest_start(skb, a->order);
123 goto nla_put_failure;
124 if (nla_put_string(skb, TCA_KIND, a->ops->kind))
125 goto nla_put_failure;
126 for (i = 0; i < (hinfo->hmask + 1); i++) {
127 head = &hinfo->htab[tcf_hash(i, hinfo->hmask)];
128 hlist_for_each_entry_safe(p, n, head, tcfc_head) {
129 if (ACT_P_DELETED == tcf_hash_release(p, 0, hinfo))
130 module_put(a->ops->owner);
134 if (nla_put_u32(skb, TCA_FCNT, n_i))
135 goto nla_put_failure;
136 nla_nest_end(skb, nest);
140 nla_nest_cancel(skb, nest);
144 static int tcf_generic_walker(struct sk_buff *skb, struct netlink_callback *cb,
145 int type, struct tc_action *a)
147 struct tcf_hashinfo *hinfo = a->ops->hinfo;
149 if (type == RTM_DELACTION) {
150 return tcf_del_walker(skb, a, hinfo);
151 } else if (type == RTM_GETACTION) {
152 return tcf_dump_walker(skb, cb, a, hinfo);
154 WARN(1, "tcf_generic_walker: unknown action %d\n", type);
159 struct tcf_common *tcf_hash_lookup(u32 index, struct tcf_hashinfo *hinfo)
161 struct tcf_common *p = NULL;
162 struct hlist_head *head;
164 spin_lock_bh(&hinfo->lock);
165 head = &hinfo->htab[tcf_hash(index, hinfo->hmask)];
166 hlist_for_each_entry_rcu(p, head, tcfc_head)
167 if (p->tcfc_index == index)
169 spin_unlock_bh(&hinfo->lock);
173 EXPORT_SYMBOL(tcf_hash_lookup);
175 u32 tcf_hash_new_index(u32 *idx_gen, struct tcf_hashinfo *hinfo)
182 } while (tcf_hash_lookup(val, hinfo));
187 EXPORT_SYMBOL(tcf_hash_new_index);
189 static int tcf_hash_search(struct tc_action *a, u32 index)
191 struct tcf_hashinfo *hinfo = a->ops->hinfo;
192 struct tcf_common *p = tcf_hash_lookup(index, hinfo);
201 struct tcf_common *tcf_hash_check(u32 index, struct tc_action *a, int bind,
202 struct tcf_hashinfo *hinfo)
204 struct tcf_common *p = NULL;
205 if (index && (p = tcf_hash_lookup(index, hinfo)) != NULL) {
213 EXPORT_SYMBOL(tcf_hash_check);
215 struct tcf_common *tcf_hash_create(u32 index, struct nlattr *est,
216 struct tc_action *a, int size, int bind,
217 u32 *idx_gen, struct tcf_hashinfo *hinfo)
219 struct tcf_common *p = kzalloc(size, GFP_KERNEL);
222 return ERR_PTR(-ENOMEM);
227 spin_lock_init(&p->tcfc_lock);
228 INIT_HLIST_NODE(&p->tcfc_head);
229 p->tcfc_index = index ? index : tcf_hash_new_index(idx_gen, hinfo);
230 p->tcfc_tm.install = jiffies;
231 p->tcfc_tm.lastuse = jiffies;
233 int err = gen_new_estimator(&p->tcfc_bstats, &p->tcfc_rate_est,
241 a->priv = (void *) p;
244 EXPORT_SYMBOL(tcf_hash_create);
246 void tcf_hash_insert(struct tcf_common *p, struct tcf_hashinfo *hinfo)
248 unsigned int h = tcf_hash(p->tcfc_index, hinfo->hmask);
250 spin_lock_bh(&hinfo->lock);
251 hlist_add_head(&p->tcfc_head, &hinfo->htab[h]);
252 spin_unlock_bh(&hinfo->lock);
254 EXPORT_SYMBOL(tcf_hash_insert);
256 static LIST_HEAD(act_base);
257 static DEFINE_RWLOCK(act_mod_lock);
259 int tcf_register_action(struct tc_action_ops *act)
261 struct tc_action_ops *a;
263 /* Must supply act, dump, cleanup and init */
264 if (!act->act || !act->dump || !act->cleanup || !act->init)
267 /* Supply defaults */
269 act->lookup = tcf_hash_search;
271 act->walk = tcf_generic_walker;
273 write_lock(&act_mod_lock);
274 list_for_each_entry(a, &act_base, head) {
275 if (act->type == a->type || (strcmp(act->kind, a->kind) == 0)) {
276 write_unlock(&act_mod_lock);
280 list_add_tail(&act->head, &act_base);
281 write_unlock(&act_mod_lock);
284 EXPORT_SYMBOL(tcf_register_action);
286 int tcf_unregister_action(struct tc_action_ops *act)
288 struct tc_action_ops *a;
291 write_lock(&act_mod_lock);
292 list_for_each_entry(a, &act_base, head) {
294 list_del(&act->head);
299 write_unlock(&act_mod_lock);
302 EXPORT_SYMBOL(tcf_unregister_action);
305 static struct tc_action_ops *tc_lookup_action_n(char *kind)
307 struct tc_action_ops *a, *res = NULL;
310 read_lock(&act_mod_lock);
311 list_for_each_entry(a, &act_base, head) {
312 if (strcmp(kind, a->kind) == 0) {
313 if (try_module_get(a->owner))
318 read_unlock(&act_mod_lock);
323 /* lookup by nlattr */
324 static struct tc_action_ops *tc_lookup_action(struct nlattr *kind)
326 struct tc_action_ops *a, *res = NULL;
329 read_lock(&act_mod_lock);
330 list_for_each_entry(a, &act_base, head) {
331 if (nla_strcmp(kind, a->kind) == 0) {
332 if (try_module_get(a->owner))
337 read_unlock(&act_mod_lock);
342 int tcf_action_exec(struct sk_buff *skb, const struct list_head *actions,
343 struct tcf_result *res)
345 const struct tc_action *a;
348 if (skb->tc_verd & TC_NCLS) {
349 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
353 list_for_each_entry(a, actions, list) {
356 ret = a->ops->act(skb, a, res);
357 if (TC_MUNGED & skb->tc_verd) {
358 /* copied already, allow trampling */
359 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
360 skb->tc_verd = CLR_TC_MUNGED(skb->tc_verd);
362 if (ret == TC_ACT_REPEAT)
363 goto repeat; /* we need a ttl - JHS */
364 if (ret != TC_ACT_PIPE)
371 EXPORT_SYMBOL(tcf_action_exec);
373 void tcf_action_destroy(struct list_head *actions, int bind)
375 struct tc_action *a, *tmp;
377 list_for_each_entry_safe(a, tmp, actions, list) {
379 if (a->ops->cleanup(a, bind) == ACT_P_DELETED)
380 module_put(a->ops->owner);
384 /*FIXME: Remove later - catch insertion bugs*/
385 WARN(1, "tcf_action_destroy: BUG? destroying NULL ops\n");
393 tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
399 return a->ops->dump(skb, a, bind, ref);
403 tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
406 unsigned char *b = skb_tail_pointer(skb);
412 if (nla_put_string(skb, TCA_KIND, a->ops->kind))
413 goto nla_put_failure;
414 if (tcf_action_copy_stats(skb, a, 0))
415 goto nla_put_failure;
416 nest = nla_nest_start(skb, TCA_OPTIONS);
418 goto nla_put_failure;
419 err = tcf_action_dump_old(skb, a, bind, ref);
421 nla_nest_end(skb, nest);
429 EXPORT_SYMBOL(tcf_action_dump_1);
432 tcf_action_dump(struct sk_buff *skb, struct list_head *actions, int bind, int ref)
438 list_for_each_entry(a, actions, list) {
439 nest = nla_nest_start(skb, a->order);
441 goto nla_put_failure;
442 err = tcf_action_dump_1(skb, a, bind, ref);
445 nla_nest_end(skb, nest);
453 nla_nest_cancel(skb, nest);
457 struct tc_action *tcf_action_init_1(struct net *net, struct nlattr *nla,
458 struct nlattr *est, char *name, int ovr,
462 struct tc_action_ops *a_o;
463 char act_name[IFNAMSIZ];
464 struct nlattr *tb[TCA_ACT_MAX + 1];
469 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL);
473 kind = tb[TCA_ACT_KIND];
476 if (nla_strlcpy(act_name, kind, IFNAMSIZ) >= IFNAMSIZ)
480 if (strlcpy(act_name, name, IFNAMSIZ) >= IFNAMSIZ)
484 a_o = tc_lookup_action_n(act_name);
486 #ifdef CONFIG_MODULES
488 request_module("act_%s", act_name);
491 a_o = tc_lookup_action_n(act_name);
493 /* We dropped the RTNL semaphore in order to
494 * perform the module load. So, even if we
495 * succeeded in loading the module we have to
496 * tell the caller to replay the request. We
497 * indicate this using -EAGAIN.
509 a = kzalloc(sizeof(*a), GFP_KERNEL);
513 INIT_LIST_HEAD(&a->list);
514 /* backward compatibility for policer */
516 err = a_o->init(net, tb[TCA_ACT_OPTIONS], est, a, ovr, bind);
518 err = a_o->init(net, nla, est, a, ovr, bind);
522 /* module count goes up only when brand new policy is created
523 * if it exists and is only bound to in a_o->init() then
524 * ACT_P_CREATED is not returned (a zero is).
526 if (err != ACT_P_CREATED)
527 module_put(a_o->owner);
535 module_put(a_o->owner);
540 int tcf_action_init(struct net *net, struct nlattr *nla,
541 struct nlattr *est, char *name, int ovr,
542 int bind, struct list_head *actions)
544 struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
545 struct tc_action *act;
549 err = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL);
553 for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
554 act = tcf_action_init_1(net, tb[i], est, name, ovr, bind);
560 list_add_tail(&act->list, actions);
565 tcf_action_destroy(actions, bind);
569 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *a,
574 struct tcf_act_hdr *h = a->priv;
579 /* compat_mode being true specifies a call that is supposed
580 * to add additional backward compatibility statistic TLVs.
583 if (a->type == TCA_OLD_COMPAT)
584 err = gnet_stats_start_copy_compat(skb, 0,
585 TCA_STATS, TCA_XSTATS, &h->tcf_lock, &d);
589 err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
595 if (gnet_stats_copy_basic(&d, &h->tcf_bstats) < 0 ||
596 gnet_stats_copy_rate_est(&d, &h->tcf_bstats,
597 &h->tcf_rate_est) < 0 ||
598 gnet_stats_copy_queue(&d, &h->tcf_qstats) < 0)
601 if (gnet_stats_finish_copy(&d) < 0)
611 tca_get_fill(struct sk_buff *skb, struct list_head *actions, u32 portid, u32 seq,
612 u16 flags, int event, int bind, int ref)
615 struct nlmsghdr *nlh;
616 unsigned char *b = skb_tail_pointer(skb);
619 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*t), flags);
623 t->tca_family = AF_UNSPEC;
627 nest = nla_nest_start(skb, TCA_ACT_TAB);
631 if (tcf_action_dump(skb, actions, bind, ref) < 0)
634 nla_nest_end(skb, nest);
636 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
645 act_get_notify(struct net *net, u32 portid, struct nlmsghdr *n,
646 struct list_head *actions, int event)
650 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
653 if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, event, 0, 0) <= 0) {
658 return rtnl_unicast(skb, net, portid);
661 static struct tc_action *
662 tcf_action_get_1(struct nlattr *nla, struct nlmsghdr *n, u32 portid)
664 struct nlattr *tb[TCA_ACT_MAX + 1];
669 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL);
674 if (tb[TCA_ACT_INDEX] == NULL ||
675 nla_len(tb[TCA_ACT_INDEX]) < sizeof(index))
677 index = nla_get_u32(tb[TCA_ACT_INDEX]);
680 a = kzalloc(sizeof(struct tc_action), GFP_KERNEL);
684 INIT_LIST_HEAD(&a->list);
686 a->ops = tc_lookup_action(tb[TCA_ACT_KIND]);
690 if (a->ops->lookup(a, index) == 0)
693 module_put(a->ops->owner);
697 module_put(a->ops->owner);
704 static void cleanup_a(struct list_head *actions)
706 struct tc_action *a, *tmp;
708 list_for_each_entry_safe(a, tmp, actions, list) {
714 static struct tc_action *create_a(int i)
716 struct tc_action *act;
718 act = kzalloc(sizeof(*act), GFP_KERNEL);
720 pr_debug("create_a: failed to alloc!\n");
724 INIT_LIST_HEAD(&act->list);
728 static int tca_action_flush(struct net *net, struct nlattr *nla,
729 struct nlmsghdr *n, u32 portid)
733 struct nlmsghdr *nlh;
735 struct netlink_callback dcb;
737 struct nlattr *tb[TCA_ACT_MAX + 1];
739 struct tc_action *a = create_a(0);
743 pr_debug("tca_action_flush: couldnt create tc_action\n");
747 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
749 pr_debug("tca_action_flush: failed skb alloc\n");
754 b = skb_tail_pointer(skb);
756 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL);
761 kind = tb[TCA_ACT_KIND];
762 a->ops = tc_lookup_action(kind);
766 nlh = nlmsg_put(skb, portid, n->nlmsg_seq, RTM_DELACTION, sizeof(*t), 0);
770 t->tca_family = AF_UNSPEC;
774 nest = nla_nest_start(skb, TCA_ACT_TAB);
778 err = a->ops->walk(skb, &dcb, RTM_DELACTION, a);
784 nla_nest_end(skb, nest);
786 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
787 nlh->nlmsg_flags |= NLM_F_ROOT;
788 module_put(a->ops->owner);
790 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
791 n->nlmsg_flags & NLM_F_ECHO);
798 module_put(a->ops->owner);
807 tca_action_gd(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
808 u32 portid, int event)
811 struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
812 struct tc_action *act;
815 ret = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL);
819 if (event == RTM_DELACTION && n->nlmsg_flags & NLM_F_ROOT) {
821 return tca_action_flush(net, tb[1], n, portid);
826 for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
827 act = tcf_action_get_1(tb[i], n, portid);
833 list_add_tail(&act->list, &actions);
836 if (event == RTM_GETACTION)
837 ret = act_get_notify(net, portid, n, &actions, event);
841 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
847 if (tca_get_fill(skb, &actions, portid, n->nlmsg_seq, 0, event,
854 /* now do the delete */
855 tcf_action_destroy(&actions, 0);
856 ret = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
857 n->nlmsg_flags & NLM_F_ECHO);
867 static int tcf_add_notify(struct net *net, struct list_head *actions,
868 u32 portid, u32 seq, int event, u16 flags)
871 struct nlmsghdr *nlh;
877 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
881 b = skb_tail_pointer(skb);
883 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*t), flags);
887 t->tca_family = AF_UNSPEC;
891 nest = nla_nest_start(skb, TCA_ACT_TAB);
895 if (tcf_action_dump(skb, actions, 0, 0) < 0)
898 nla_nest_end(skb, nest);
900 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
901 NETLINK_CB(skb).dst_group = RTNLGRP_TC;
903 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
915 tcf_action_add(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
920 u32 seq = n->nlmsg_seq;
922 ret = tcf_action_init(net, nla, NULL, NULL, ovr, 0, &actions);
926 /* dump then free all the actions after update; inserted policy
929 ret = tcf_add_notify(net, &actions, portid, seq, RTM_NEWACTION, n->nlmsg_flags);
935 static int tc_ctl_action(struct sk_buff *skb, struct nlmsghdr *n)
937 struct net *net = sock_net(skb->sk);
938 struct nlattr *tca[TCA_ACT_MAX + 1];
939 u32 portid = skb ? NETLINK_CB(skb).portid : 0;
940 int ret = 0, ovr = 0;
942 if ((n->nlmsg_type != RTM_GETACTION) && !capable(CAP_NET_ADMIN))
945 ret = nlmsg_parse(n, sizeof(struct tcamsg), tca, TCA_ACT_MAX, NULL);
949 if (tca[TCA_ACT_TAB] == NULL) {
950 pr_notice("tc_ctl_action: received NO action attribs\n");
954 /* n->nlmsg_flags & NLM_F_CREATE */
955 switch (n->nlmsg_type) {
957 /* we are going to assume all other flags
958 * imply create only if it doesn't exist
959 * Note that CREATE | EXCL implies that
960 * but since we want avoid ambiguity (eg when flags
961 * is zero) then just set this
963 if (n->nlmsg_flags & NLM_F_REPLACE)
966 ret = tcf_action_add(net, tca[TCA_ACT_TAB], n, portid, ovr);
971 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
972 portid, RTM_DELACTION);
975 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
976 portid, RTM_GETACTION);
985 static struct nlattr *
986 find_dump_kind(const struct nlmsghdr *n)
988 struct nlattr *tb1, *tb2[TCA_ACT_MAX + 1];
989 struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
990 struct nlattr *nla[TCAA_MAX + 1];
993 if (nlmsg_parse(n, sizeof(struct tcamsg), nla, TCAA_MAX, NULL) < 0)
995 tb1 = nla[TCA_ACT_TAB];
999 if (nla_parse(tb, TCA_ACT_MAX_PRIO, nla_data(tb1),
1000 NLMSG_ALIGN(nla_len(tb1)), NULL) < 0)
1005 if (nla_parse(tb2, TCA_ACT_MAX, nla_data(tb[1]),
1006 nla_len(tb[1]), NULL) < 0)
1008 kind = tb2[TCA_ACT_KIND];
1014 tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
1016 struct nlmsghdr *nlh;
1017 unsigned char *b = skb_tail_pointer(skb);
1018 struct nlattr *nest;
1019 struct tc_action_ops *a_o;
1022 struct tcamsg *t = (struct tcamsg *) nlmsg_data(cb->nlh);
1023 struct nlattr *kind = find_dump_kind(cb->nlh);
1026 pr_info("tc_dump_action: action bad kind\n");
1030 a_o = tc_lookup_action(kind);
1034 memset(&a, 0, sizeof(struct tc_action));
1037 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1038 cb->nlh->nlmsg_type, sizeof(*t), 0);
1040 goto out_module_put;
1041 t = nlmsg_data(nlh);
1042 t->tca_family = AF_UNSPEC;
1046 nest = nla_nest_start(skb, TCA_ACT_TAB);
1048 goto out_module_put;
1050 ret = a_o->walk(skb, cb, RTM_GETACTION, &a);
1052 goto out_module_put;
1055 nla_nest_end(skb, nest);
1058 nla_nest_cancel(skb, nest);
1060 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1061 if (NETLINK_CB(cb->skb).portid && ret)
1062 nlh->nlmsg_flags |= NLM_F_MULTI;
1063 module_put(a_o->owner);
1067 module_put(a_o->owner);
1072 static int __init tc_action_init(void)
1074 rtnl_register(PF_UNSPEC, RTM_NEWACTION, tc_ctl_action, NULL, NULL);
1075 rtnl_register(PF_UNSPEC, RTM_DELACTION, tc_ctl_action, NULL, NULL);
1076 rtnl_register(PF_UNSPEC, RTM_GETACTION, tc_ctl_action, tc_dump_action,
1082 subsys_initcall(tc_action_init);