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 <asm/uaccess.h>
15 #include <asm/system.h>
16 #include <linux/bitops.h>
17 #include <linux/config.h>
18 #include <linux/types.h>
19 #include <linux/kernel.h>
20 #include <linux/sched.h>
21 #include <linux/string.h>
23 #include <linux/socket.h>
24 #include <linux/sockios.h>
26 #include <linux/errno.h>
27 #include <linux/interrupt.h>
28 #include <linux/netdevice.h>
29 #include <linux/skbuff.h>
30 #include <linux/rtnetlink.h>
31 #include <linux/init.h>
32 #include <linux/kmod.h>
34 #include <net/sch_generic.h>
35 #include <net/act_api.h>
38 #define DPRINTK(format, args...) printk(KERN_DEBUG format, ##args)
40 #define DPRINTK(format, args...)
43 #define D2PRINTK(format, args...) printk(KERN_DEBUG format, ##args)
45 #define D2PRINTK(format, args...)
48 static struct tc_action_ops *act_base = NULL;
49 static DEFINE_RWLOCK(act_mod_lock);
51 int tcf_register_action(struct tc_action_ops *act)
53 struct tc_action_ops *a, **ap;
55 write_lock(&act_mod_lock);
56 for (ap = &act_base; (a = *ap) != NULL; ap = &a->next) {
57 if (act->type == a->type || (strcmp(act->kind, a->kind) == 0)) {
58 write_unlock(&act_mod_lock);
64 write_unlock(&act_mod_lock);
68 int tcf_unregister_action(struct tc_action_ops *act)
70 struct tc_action_ops *a, **ap;
73 write_lock(&act_mod_lock);
74 for (ap = &act_base; (a = *ap) != NULL; ap = &a->next)
82 write_unlock(&act_mod_lock);
87 static struct tc_action_ops *tc_lookup_action_n(char *kind)
89 struct tc_action_ops *a = NULL;
92 read_lock(&act_mod_lock);
93 for (a = act_base; a; a = a->next) {
94 if (strcmp(kind, a->kind) == 0) {
95 if (!try_module_get(a->owner)) {
96 read_unlock(&act_mod_lock);
102 read_unlock(&act_mod_lock);
107 /* lookup by rtattr */
108 static struct tc_action_ops *tc_lookup_action(struct rtattr *kind)
110 struct tc_action_ops *a = NULL;
113 read_lock(&act_mod_lock);
114 for (a = act_base; a; a = a->next) {
115 if (rtattr_strcmp(kind, a->kind) == 0) {
116 if (!try_module_get(a->owner)) {
117 read_unlock(&act_mod_lock);
123 read_unlock(&act_mod_lock);
130 static struct tc_action_ops *tc_lookup_action_id(u32 type)
132 struct tc_action_ops *a = NULL;
135 read_lock(&act_mod_lock);
136 for (a = act_base; a; a = a->next) {
137 if (a->type == type) {
138 if (!try_module_get(a->owner)) {
139 read_unlock(&act_mod_lock);
145 read_unlock(&act_mod_lock);
151 int tcf_action_exec(struct sk_buff *skb, struct tc_action *act,
152 struct tcf_result *res)
157 if (skb->tc_verd & TC_NCLS) {
158 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
159 D2PRINTK("(%p)tcf_action_exec: cleared TC_NCLS in %s out %s\n",
160 skb, skb->input_dev ? skb->input_dev->name : "xxx",
165 while ((a = act) != NULL) {
167 if (a->ops && a->ops->act) {
168 ret = a->ops->act(&skb, a);
169 if (TC_MUNGED & skb->tc_verd) {
170 /* copied already, allow trampling */
171 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
172 skb->tc_verd = CLR_TC_MUNGED(skb->tc_verd);
174 if (ret != TC_ACT_PIPE)
176 if (ret == TC_ACT_REPEAT)
177 goto repeat; /* we need a ttl - JHS */
182 if (skb->tc_classid > 0) {
183 res->classid = skb->tc_classid;
190 void tcf_action_destroy(struct tc_action *act, int bind)
194 for (a = act; a; a = act) {
195 if (a->ops && a->ops->cleanup) {
196 DPRINTK("tcf_action_destroy destroying %p next %p\n",
198 if (a->ops->cleanup(a, bind) == ACT_P_DELETED)
199 module_put(a->ops->owner);
202 } else { /*FIXME: Remove later - catch insertion bugs*/
203 printk("tcf_action_destroy: BUG? destroying NULL ops\n");
211 tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
215 if (a->ops == NULL || a->ops->dump == NULL)
217 return a->ops->dump(skb, a, bind, ref);
221 tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
224 unsigned char *b = skb->tail;
227 if (a->ops == NULL || a->ops->dump == NULL)
230 RTA_PUT(skb, TCA_KIND, IFNAMSIZ, a->ops->kind);
231 if (tcf_action_copy_stats(skb, a, 0))
233 r = (struct rtattr*) skb->tail;
234 RTA_PUT(skb, TCA_OPTIONS, 0, NULL);
235 if ((err = tcf_action_dump_old(skb, a, bind, ref)) > 0) {
236 r->rta_len = skb->tail - (u8*)r;
241 skb_trim(skb, b - skb->data);
246 tcf_action_dump(struct sk_buff *skb, struct tc_action *act, int bind, int ref)
250 unsigned char *b = skb->tail;
253 while ((a = act) != NULL) {
254 r = (struct rtattr*) skb->tail;
256 RTA_PUT(skb, a->order, 0, NULL);
257 err = tcf_action_dump_1(skb, a, bind, ref);
260 r->rta_len = skb->tail - (u8*)r;
266 skb_trim(skb, b - skb->data);
270 struct tc_action *tcf_action_init_1(struct rtattr *rta, struct rtattr *est,
271 char *name, int ovr, int bind, int *err)
274 struct tc_action_ops *a_o;
275 char act_name[IFNAMSIZ];
276 struct rtattr *tb[TCA_ACT_MAX+1];
282 if (rtattr_parse_nested(tb, TCA_ACT_MAX, rta) < 0)
284 kind = tb[TCA_ACT_KIND-1];
287 if (rtattr_strlcpy(act_name, kind, IFNAMSIZ) >= IFNAMSIZ)
290 if (strlcpy(act_name, name, IFNAMSIZ) >= IFNAMSIZ)
294 a_o = tc_lookup_action_n(act_name);
298 request_module(act_name);
301 a_o = tc_lookup_action_n(act_name);
303 /* We dropped the RTNL semaphore in order to
304 * perform the module load. So, even if we
305 * succeeded in loading the module we have to
306 * tell the caller to replay the request. We
307 * indicate this using -EAGAIN.
318 a = kmalloc(sizeof(*a), GFP_KERNEL);
321 memset(a, 0, sizeof(*a));
323 /* backward compatibility for policer */
325 *err = a_o->init(tb[TCA_ACT_OPTIONS-1], est, a, ovr, bind);
327 *err = a_o->init(rta, est, a, ovr, bind);
331 /* module count goes up only when brand new policy is created
332 if it exists and is only bound to in a_o->init() then
333 ACT_P_CREATED is not returned (a zero is).
335 if (*err != ACT_P_CREATED)
336 module_put(a_o->owner);
338 DPRINTK("tcf_action_init_1: successfull %s\n", act_name);
346 module_put(a_o->owner);
351 struct tc_action *tcf_action_init(struct rtattr *rta, struct rtattr *est,
352 char *name, int ovr, int bind, int *err)
354 struct rtattr *tb[TCA_ACT_MAX_PRIO+1];
355 struct tc_action *head = NULL, *act, *act_prev = NULL;
358 if (rtattr_parse_nested(tb, TCA_ACT_MAX_PRIO, rta) < 0) {
363 for (i=0; i < TCA_ACT_MAX_PRIO && tb[i]; i++) {
364 act = tcf_action_init_1(tb[i], est, name, ovr, bind, err);
372 act_prev->next = act;
379 tcf_action_destroy(head, bind);
383 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *a,
388 struct tcf_act_hdr *h = a->priv;
393 /* compat_mode being true specifies a call that is supposed
394 * to add additional backward compatiblity statistic TLVs.
397 if (a->type == TCA_OLD_COMPAT)
398 err = gnet_stats_start_copy_compat(skb, 0,
399 TCA_STATS, TCA_XSTATS, h->stats_lock, &d);
403 err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
409 if (a->ops != NULL && a->ops->get_stats != NULL)
410 if (a->ops->get_stats(skb, a) < 0)
413 if (gnet_stats_copy_basic(&d, &h->bstats) < 0 ||
414 #ifdef CONFIG_NET_ESTIMATOR
415 gnet_stats_copy_rate_est(&d, &h->rate_est) < 0 ||
417 gnet_stats_copy_queue(&d, &h->qstats) < 0)
420 if (gnet_stats_finish_copy(&d) < 0)
430 tca_get_fill(struct sk_buff *skb, struct tc_action *a, u32 pid, u32 seq,
431 unsigned flags, int event, int bind, int ref)
434 struct nlmsghdr *nlh;
435 unsigned char *b = skb->tail;
438 nlh = NLMSG_PUT(skb, pid, seq, event, sizeof(*t));
439 nlh->nlmsg_flags = flags;
441 t->tca_family = AF_UNSPEC;
443 x = (struct rtattr*) skb->tail;
444 RTA_PUT(skb, TCA_ACT_TAB, 0, NULL);
446 if (tcf_action_dump(skb, a, bind, ref) < 0)
449 x->rta_len = skb->tail - (u8*)x;
451 nlh->nlmsg_len = skb->tail - b;
456 skb_trim(skb, b - skb->data);
461 act_get_notify(u32 pid, struct nlmsghdr *n, struct tc_action *a, int event)
466 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
469 if (tca_get_fill(skb, a, pid, n->nlmsg_seq, 0, event, 0, 0) <= 0) {
473 err = netlink_unicast(rtnl, skb, pid, MSG_DONTWAIT);
479 static struct tc_action *
480 tcf_action_get_1(struct rtattr *rta, struct nlmsghdr *n, u32 pid, int *err)
482 struct rtattr *tb[TCA_ACT_MAX+1];
487 if (rtattr_parse_nested(tb, TCA_ACT_MAX, rta) < 0)
490 if (tb[TCA_ACT_INDEX - 1] == NULL ||
491 RTA_PAYLOAD(tb[TCA_ACT_INDEX - 1]) < sizeof(index))
493 index = *(int *)RTA_DATA(tb[TCA_ACT_INDEX - 1]);
496 a = kmalloc(sizeof(struct tc_action), GFP_KERNEL);
499 memset(a, 0, sizeof(struct tc_action));
502 a->ops = tc_lookup_action(tb[TCA_ACT_KIND - 1]);
505 if (a->ops->lookup == NULL)
508 if (a->ops->lookup(a, index) == 0)
511 module_put(a->ops->owner);
515 module_put(a->ops->owner);
521 static void cleanup_a(struct tc_action *act)
525 for (a = act; a; a = act) {
531 static struct tc_action *create_a(int i)
533 struct tc_action *act;
535 act = kmalloc(sizeof(*act), GFP_KERNEL);
537 printk("create_a: failed to alloc!\n");
540 memset(act, 0, sizeof(*act));
545 static int tca_action_flush(struct rtattr *rta, struct nlmsghdr *n, u32 pid)
549 struct nlmsghdr *nlh;
551 struct netlink_callback dcb;
553 struct rtattr *tb[TCA_ACT_MAX+1];
555 struct tc_action *a = create_a(0);
559 printk("tca_action_flush: couldnt create tc_action\n");
563 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
565 printk("tca_action_flush: failed skb alloc\n");
570 b = (unsigned char *)skb->tail;
572 if (rtattr_parse_nested(tb, TCA_ACT_MAX, rta) < 0)
575 kind = tb[TCA_ACT_KIND-1];
576 a->ops = tc_lookup_action(kind);
580 nlh = NLMSG_PUT(skb, pid, n->nlmsg_seq, RTM_DELACTION, sizeof(*t));
582 t->tca_family = AF_UNSPEC;
584 x = (struct rtattr *) skb->tail;
585 RTA_PUT(skb, TCA_ACT_TAB, 0, NULL);
587 err = a->ops->walk(skb, &dcb, RTM_DELACTION, a);
591 x->rta_len = skb->tail - (u8 *) x;
593 nlh->nlmsg_len = skb->tail - b;
594 nlh->nlmsg_flags |= NLM_F_ROOT;
595 module_put(a->ops->owner);
597 err = rtnetlink_send(skb, pid, RTMGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
604 module_put(a->ops->owner);
613 tca_action_gd(struct rtattr *rta, struct nlmsghdr *n, u32 pid, int event)
616 struct rtattr *tb[TCA_ACT_MAX_PRIO+1];
617 struct tc_action *head = NULL, *act, *act_prev = NULL;
619 if (rtattr_parse_nested(tb, TCA_ACT_MAX_PRIO, rta) < 0)
622 if (event == RTM_DELACTION && n->nlmsg_flags&NLM_F_ROOT) {
623 if (tb[0] != NULL && tb[1] == NULL)
624 return tca_action_flush(tb[0], n, pid);
627 for (i=0; i < TCA_ACT_MAX_PRIO && tb[i]; i++) {
628 act = tcf_action_get_1(tb[i], n, pid, &ret);
636 act_prev->next = act;
640 if (event == RTM_GETACTION)
641 ret = act_get_notify(pid, n, head, event);
645 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
651 if (tca_get_fill(skb, head, pid, n->nlmsg_seq, 0, event,
658 /* now do the delete */
659 tcf_action_destroy(head, 0);
660 ret = rtnetlink_send(skb, pid, RTMGRP_TC,
661 n->nlmsg_flags&NLM_F_ECHO);
671 static int tcf_add_notify(struct tc_action *a, u32 pid, u32 seq, int event,
675 struct nlmsghdr *nlh;
681 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
685 b = (unsigned char *)skb->tail;
687 nlh = NLMSG_PUT(skb, pid, seq, event, sizeof(*t));
688 nlh->nlmsg_flags = flags;
690 t->tca_family = AF_UNSPEC;
692 x = (struct rtattr*) skb->tail;
693 RTA_PUT(skb, TCA_ACT_TAB, 0, NULL);
695 if (tcf_action_dump(skb, a, 0, 0) < 0)
698 x->rta_len = skb->tail - (u8*)x;
700 nlh->nlmsg_len = skb->tail - b;
701 NETLINK_CB(skb).dst_groups = RTMGRP_TC;
703 err = rtnetlink_send(skb, pid, RTMGRP_TC, flags&NLM_F_ECHO);
710 skb_trim(skb, b - skb->data);
716 tcf_action_add(struct rtattr *rta, struct nlmsghdr *n, u32 pid, int ovr)
719 struct tc_action *act;
721 u32 seq = n->nlmsg_seq;
723 act = tcf_action_init(rta, NULL, NULL, ovr, 0, &ret);
727 /* dump then free all the actions after update; inserted policy
730 ret = tcf_add_notify(act, pid, seq, RTM_NEWACTION, n->nlmsg_flags);
731 for (a = act; a; a = act) {
739 static int tc_ctl_action(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
741 struct rtattr **tca = arg;
742 u32 pid = skb ? NETLINK_CB(skb).pid : 0;
743 int ret = 0, ovr = 0;
745 if (tca[TCA_ACT_TAB-1] == NULL) {
746 printk("tc_ctl_action: received NO action attribs\n");
750 /* n->nlmsg_flags&NLM_F_CREATE
752 switch (n->nlmsg_type) {
754 /* we are going to assume all other flags
755 * imply create only if it doesnt exist
756 * Note that CREATE | EXCL implies that
757 * but since we want avoid ambiguity (eg when flags
758 * is zero) then just set this
760 if (n->nlmsg_flags&NLM_F_REPLACE)
763 ret = tcf_action_add(tca[TCA_ACT_TAB-1], n, pid, ovr);
768 ret = tca_action_gd(tca[TCA_ACT_TAB-1], n, pid, RTM_DELACTION);
771 ret = tca_action_gd(tca[TCA_ACT_TAB-1], n, pid, RTM_GETACTION);
781 find_dump_kind(struct nlmsghdr *n)
783 struct rtattr *tb1, *tb2[TCA_ACT_MAX+1];
784 struct rtattr *tb[TCA_ACT_MAX_PRIO + 1];
785 struct rtattr *rta[TCAA_MAX + 1];
787 int min_len = NLMSG_LENGTH(sizeof(struct tcamsg));
788 int attrlen = n->nlmsg_len - NLMSG_ALIGN(min_len);
789 struct rtattr *attr = (void *) n + NLMSG_ALIGN(min_len);
791 if (rtattr_parse(rta, TCAA_MAX, attr, attrlen) < 0)
793 tb1 = rta[TCA_ACT_TAB - 1];
797 if (rtattr_parse(tb, TCA_ACT_MAX_PRIO, RTA_DATA(tb1),
798 NLMSG_ALIGN(RTA_PAYLOAD(tb1))) < 0)
803 if (rtattr_parse(tb2, TCA_ACT_MAX, RTA_DATA(tb[0]),
804 RTA_PAYLOAD(tb[0])) < 0)
806 kind = tb2[TCA_ACT_KIND-1];
808 return (char *) RTA_DATA(kind);
812 tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
814 struct nlmsghdr *nlh;
815 unsigned char *b = skb->tail;
817 struct tc_action_ops *a_o;
820 struct tcamsg *t = (struct tcamsg *) NLMSG_DATA(cb->nlh);
821 char *kind = find_dump_kind(cb->nlh);
824 printk("tc_dump_action: action bad kind\n");
828 a_o = tc_lookup_action_n(kind);
830 printk("failed to find %s\n", kind);
834 memset(&a, 0, sizeof(struct tc_action));
837 if (a_o->walk == NULL) {
838 printk("tc_dump_action: %s !capable of dumping table\n", kind);
842 nlh = NLMSG_PUT(skb, NETLINK_CB(cb->skb).pid, cb->nlh->nlmsg_seq,
843 cb->nlh->nlmsg_type, sizeof(*t));
845 t->tca_family = AF_UNSPEC;
847 x = (struct rtattr *) skb->tail;
848 RTA_PUT(skb, TCA_ACT_TAB, 0, NULL);
850 ret = a_o->walk(skb, cb, RTM_GETACTION, &a);
855 x->rta_len = skb->tail - (u8 *) x;
858 skb_trim(skb, (u8*)x - skb->data);
860 nlh->nlmsg_len = skb->tail - b;
861 if (NETLINK_CB(cb->skb).pid && ret)
862 nlh->nlmsg_flags |= NLM_F_MULTI;
863 module_put(a_o->owner);
868 module_put(a_o->owner);
869 skb_trim(skb, b - skb->data);
873 static int __init tc_action_init(void)
875 struct rtnetlink_link *link_p = rtnetlink_links[PF_UNSPEC];
878 link_p[RTM_NEWACTION-RTM_BASE].doit = tc_ctl_action;
879 link_p[RTM_DELACTION-RTM_BASE].doit = tc_ctl_action;
880 link_p[RTM_GETACTION-RTM_BASE].doit = tc_ctl_action;
881 link_p[RTM_GETACTION-RTM_BASE].dumpit = tc_dump_action;
884 printk("TC classifier action (bugs to netdev@oss.sgi.com cc "
885 "hadi@cyberus.ca)\n");
889 subsys_initcall(tc_action_init);
891 EXPORT_SYMBOL(tcf_register_action);
892 EXPORT_SYMBOL(tcf_unregister_action);
893 EXPORT_SYMBOL(tcf_action_exec);
894 EXPORT_SYMBOL(tcf_action_dump_1);