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 <asm/uaccess.h>
18 #include <asm/system.h>
19 #include <linux/bitops.h>
20 #include <linux/module.h>
21 #include <linux/types.h>
22 #include <linux/kernel.h>
23 #include <linux/string.h>
25 #include <linux/socket.h>
26 #include <linux/sockios.h>
28 #include <linux/errno.h>
29 #include <linux/interrupt.h>
30 #include <linux/netdevice.h>
31 #include <linux/skbuff.h>
32 #include <linux/rtnetlink.h>
33 #include <linux/init.h>
34 #include <linux/kmod.h>
36 #include <net/pkt_sched.h>
37 #include <net/pkt_cls.h>
40 #define DPRINTK(format,args...) printk(KERN_DEBUG format,##args)
42 #define DPRINTK(format,args...)
45 /* The list of all installed classifier types */
47 static struct tcf_proto_ops *tcf_proto_base;
49 /* Protects list of registered TC modules. It is pure SMP lock. */
50 static DEFINE_RWLOCK(cls_mod_lock);
52 /* Find classifier type by string name */
54 static struct tcf_proto_ops * tcf_proto_lookup_ops(struct rtattr *kind)
56 struct tcf_proto_ops *t = NULL;
59 read_lock(&cls_mod_lock);
60 for (t = tcf_proto_base; t; t = t->next) {
61 if (rtattr_strcmp(kind, t->kind) == 0) {
62 if (!try_module_get(t->owner))
67 read_unlock(&cls_mod_lock);
72 /* Register(unregister) new classifier type */
74 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
76 struct tcf_proto_ops *t, **tp;
79 write_lock(&cls_mod_lock);
80 for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
81 if (!strcmp(ops->kind, t->kind))
88 write_unlock(&cls_mod_lock);
92 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
94 struct tcf_proto_ops *t, **tp;
97 write_lock(&cls_mod_lock);
98 for (tp = &tcf_proto_base; (t=*tp) != NULL; tp = &t->next)
107 write_unlock(&cls_mod_lock);
111 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
112 struct tcf_proto *tp, unsigned long fh, int event);
115 /* Select new prio value from the range, managed by kernel. */
117 static __inline__ u32 tcf_auto_prio(struct tcf_proto *tp)
119 u32 first = TC_H_MAKE(0xC0000000U,0U);
127 /* Add/change/delete/get a filter node */
129 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
137 struct net_device *dev;
139 struct tcf_proto **back, **chain;
140 struct tcf_proto *tp;
141 struct tcf_proto_ops *tp_ops;
142 struct Qdisc_class_ops *cops;
150 protocol = TC_H_MIN(t->tcm_info);
151 prio = TC_H_MAJ(t->tcm_info);
153 parent = t->tcm_parent;
157 /* If no priority is given, user wants we allocated it. */
158 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
160 prio = TC_H_MAKE(0x80000000U,0U);
163 /* Find head of filter chain. */
166 if ((dev = __dev_get_by_index(t->tcm_ifindex)) == NULL)
171 q = dev->qdisc_sleeping;
173 } else if ((q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent))) == NULL)
176 /* Is it classful? */
177 if ((cops = q->ops->cl_ops) == NULL)
180 /* Do we search for filter, attached to class? */
181 if (TC_H_MIN(parent)) {
182 cl = cops->get(q, parent);
187 /* And the last stroke */
188 chain = cops->tcf_chain(q, cl);
193 /* Check the chain for existence of proto-tcf with this priority */
194 for (back = chain; (tp=*back) != NULL; back = &tp->next) {
195 if (tp->prio >= prio) {
196 if (tp->prio == prio) {
197 if (!nprio || (tp->protocol != protocol && protocol))
206 /* Proto-tcf does not exist, create new one */
208 if (tca[TCA_KIND-1] == NULL || !protocol)
212 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
216 /* Create new proto tcf */
219 if ((tp = kzalloc(sizeof(*tp), GFP_KERNEL)) == NULL)
222 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND-1]);
223 if (tp_ops == NULL) {
225 struct rtattr *kind = tca[TCA_KIND-1];
229 rtattr_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
231 request_module("cls_%s", name);
233 tp_ops = tcf_proto_lookup_ops(kind);
234 /* We dropped the RTNL semaphore in order to
235 * perform the module load. So, even if we
236 * succeeded in loading the module we have to
237 * replay the request. We indicate this using
240 if (tp_ops != NULL) {
241 module_put(tp_ops->owner);
250 tp->protocol = protocol;
251 tp->prio = nprio ? : tcf_auto_prio(*back);
253 tp->classify = tp_ops->classify;
254 tp->classid = parent;
255 if ((err = tp_ops->init(tp)) != 0) {
256 module_put(tp_ops->owner);
261 qdisc_lock_tree(dev);
264 qdisc_unlock_tree(dev);
266 } else if (tca[TCA_KIND-1] && rtattr_strcmp(tca[TCA_KIND-1], tp->ops->kind))
269 fh = tp->ops->get(tp, t->tcm_handle);
272 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
273 qdisc_lock_tree(dev);
275 qdisc_unlock_tree(dev);
277 tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
284 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
287 switch (n->nlmsg_type) {
290 if (n->nlmsg_flags&NLM_F_EXCL)
294 err = tp->ops->delete(tp, fh);
296 tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
299 err = tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
307 err = tp->ops->change(tp, cl, t->tcm_handle, tca, &fh);
309 tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
315 /* Replay the request. */
321 tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp, unsigned long fh,
322 u32 pid, u32 seq, u16 flags, int event)
325 struct nlmsghdr *nlh;
326 unsigned char *b = skb->tail;
328 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
329 tcm = NLMSG_DATA(nlh);
330 tcm->tcm_family = AF_UNSPEC;
333 tcm->tcm_ifindex = tp->q->dev->ifindex;
334 tcm->tcm_parent = tp->classid;
335 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
336 RTA_PUT(skb, TCA_KIND, IFNAMSIZ, tp->ops->kind);
337 tcm->tcm_handle = fh;
338 if (RTM_DELTFILTER != event) {
340 if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
343 nlh->nlmsg_len = skb->tail - b;
348 skb_trim(skb, b - skb->data);
352 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
353 struct tcf_proto *tp, unsigned long fh, int event)
356 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
358 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
362 if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) {
367 return rtnetlink_send(skb, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
374 struct netlink_callback *cb;
377 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n, struct tcf_walker *arg)
379 struct tcf_dump_args *a = (void*)arg;
381 return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid,
382 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
385 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
389 struct net_device *dev;
391 struct tcf_proto *tp, **chain;
392 struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh);
393 unsigned long cl = 0;
394 struct Qdisc_class_ops *cops;
395 struct tcf_dump_args arg;
397 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
399 if ((dev = dev_get_by_index(tcm->tcm_ifindex)) == NULL)
402 read_lock(&qdisc_tree_lock);
403 if (!tcm->tcm_parent)
404 q = dev->qdisc_sleeping;
406 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
409 if ((cops = q->ops->cl_ops) == NULL)
411 if (TC_H_MIN(tcm->tcm_parent)) {
412 cl = cops->get(q, tcm->tcm_parent);
416 chain = cops->tcf_chain(q, cl);
422 for (tp=*chain, t=0; tp; tp = tp->next, t++) {
423 if (t < s_t) continue;
424 if (TC_H_MAJ(tcm->tcm_info) &&
425 TC_H_MAJ(tcm->tcm_info) != tp->prio)
427 if (TC_H_MIN(tcm->tcm_info) &&
428 TC_H_MIN(tcm->tcm_info) != tp->protocol)
431 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
432 if (cb->args[1] == 0) {
433 if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid,
434 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER) <= 0) {
439 if (tp->ops->walk == NULL)
441 arg.w.fn = tcf_node_dump;
445 arg.w.skip = cb->args[1]-1;
447 tp->ops->walk(tp, &arg.w);
448 cb->args[1] = arg.w.count+1;
459 read_unlock(&qdisc_tree_lock);
465 tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
467 #ifdef CONFIG_NET_CLS_ACT
469 tcf_action_destroy(exts->action, TCA_ACT_UNBIND);
472 #elif defined CONFIG_NET_CLS_POLICE
474 tcf_police_release(exts->police, TCA_ACT_UNBIND);
482 tcf_exts_validate(struct tcf_proto *tp, struct rtattr **tb,
483 struct rtattr *rate_tlv, struct tcf_exts *exts,
484 struct tcf_ext_map *map)
486 memset(exts, 0, sizeof(*exts));
488 #ifdef CONFIG_NET_CLS_ACT
491 struct tc_action *act;
493 if (map->police && tb[map->police-1]) {
494 act = tcf_action_init_1(tb[map->police-1], rate_tlv, "police",
495 TCA_ACT_NOREPLACE, TCA_ACT_BIND, &err);
499 act->type = TCA_OLD_COMPAT;
501 } else if (map->action && tb[map->action-1]) {
502 act = tcf_action_init(tb[map->action-1], rate_tlv, NULL,
503 TCA_ACT_NOREPLACE, TCA_ACT_BIND, &err);
510 #elif defined CONFIG_NET_CLS_POLICE
511 if (map->police && tb[map->police-1]) {
512 struct tcf_police *p;
514 p = tcf_police_locate(tb[map->police-1], rate_tlv);
519 } else if (map->action && tb[map->action-1])
522 if ((map->action && tb[map->action-1]) ||
523 (map->police && tb[map->police-1]))
531 tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
532 struct tcf_exts *src)
534 #ifdef CONFIG_NET_CLS_ACT
536 struct tc_action *act;
538 act = xchg(&dst->action, src->action);
541 tcf_action_destroy(act, TCA_ACT_UNBIND);
543 #elif defined CONFIG_NET_CLS_POLICE
545 struct tcf_police *p;
547 p = xchg(&dst->police, src->police);
550 tcf_police_release(p, TCA_ACT_UNBIND);
556 tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts,
557 struct tcf_ext_map *map)
559 #ifdef CONFIG_NET_CLS_ACT
560 if (map->action && exts->action) {
562 * again for backward compatible mode - we want
563 * to work with both old and new modes of entering
564 * tc data even if iproute2 was newer - jhs
566 struct rtattr * p_rta = (struct rtattr*) skb->tail;
568 if (exts->action->type != TCA_OLD_COMPAT) {
569 RTA_PUT(skb, map->action, 0, NULL);
570 if (tcf_action_dump(skb, exts->action, 0, 0) < 0)
572 p_rta->rta_len = skb->tail - (u8*)p_rta;
573 } else if (map->police) {
574 RTA_PUT(skb, map->police, 0, NULL);
575 if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0)
577 p_rta->rta_len = skb->tail - (u8*)p_rta;
580 #elif defined CONFIG_NET_CLS_POLICE
581 if (map->police && exts->police) {
582 struct rtattr * p_rta = (struct rtattr*) skb->tail;
584 RTA_PUT(skb, map->police, 0, NULL);
586 if (tcf_police_dump(skb, exts->police) < 0)
589 p_rta->rta_len = skb->tail - (u8*)p_rta;
593 rtattr_failure: __attribute__ ((unused))
598 tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts,
599 struct tcf_ext_map *map)
601 #ifdef CONFIG_NET_CLS_ACT
603 if (tcf_action_copy_stats(skb, exts->action, 1) < 0)
605 #elif defined CONFIG_NET_CLS_POLICE
607 if (tcf_police_dump_stats(skb, exts->police) < 0)
611 rtattr_failure: __attribute__ ((unused))
615 static int __init tc_filter_init(void)
617 struct rtnetlink_link *link_p = rtnetlink_links[PF_UNSPEC];
619 /* Setup rtnetlink links. It is made here to avoid
620 exporting large number of public symbols.
624 link_p[RTM_NEWTFILTER-RTM_BASE].doit = tc_ctl_tfilter;
625 link_p[RTM_DELTFILTER-RTM_BASE].doit = tc_ctl_tfilter;
626 link_p[RTM_GETTFILTER-RTM_BASE].doit = tc_ctl_tfilter;
627 link_p[RTM_GETTFILTER-RTM_BASE].dumpit = tc_dump_tfilter;
632 subsys_initcall(tc_filter_init);
634 EXPORT_SYMBOL(register_tcf_proto_ops);
635 EXPORT_SYMBOL(unregister_tcf_proto_ops);
636 EXPORT_SYMBOL(tcf_exts_validate);
637 EXPORT_SYMBOL(tcf_exts_destroy);
638 EXPORT_SYMBOL(tcf_exts_change);
639 EXPORT_SYMBOL(tcf_exts_dump);
640 EXPORT_SYMBOL(tcf_exts_dump_stats);