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 <asm/bitops.h>
20 #include <linux/config.h>
21 #include <linux/module.h>
22 #include <linux/types.h>
23 #include <linux/kernel.h>
24 #include <linux/sched.h>
25 #include <linux/string.h>
27 #include <linux/socket.h>
28 #include <linux/sockios.h>
30 #include <linux/errno.h>
31 #include <linux/interrupt.h>
32 #include <linux/netdevice.h>
33 #include <linux/skbuff.h>
34 #include <linux/rtnetlink.h>
35 #include <linux/init.h>
36 #include <linux/kmod.h>
38 #include <net/pkt_sched.h>
41 #define DPRINTK(format,args...) printk(KERN_DEBUG format,##args)
43 #define DPRINTK(format,args...)
46 /* The list of all installed classifier types */
48 static struct tcf_proto_ops *tcf_proto_base;
50 /* Protects list of registered TC modules. It is pure SMP lock. */
51 static rwlock_t cls_mod_lock = RW_LOCK_UNLOCKED;
53 /* Find classifier type by string name */
55 struct tcf_proto_ops * tcf_proto_lookup_ops(struct rtattr *kind)
57 struct tcf_proto_ops *t = NULL;
60 read_lock(&cls_mod_lock);
61 for (t = tcf_proto_base; t; t = t->next) {
62 if (rtattr_strcmp(kind, t->kind) == 0)
65 read_unlock(&cls_mod_lock);
70 /* Register(unregister) new classifier type */
72 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
74 struct tcf_proto_ops *t, **tp;
77 write_lock(&cls_mod_lock);
78 for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
79 if (!strcmp(ops->kind, t->kind))
86 write_unlock(&cls_mod_lock);
90 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
92 struct tcf_proto_ops *t, **tp;
95 write_lock(&cls_mod_lock);
96 for (tp = &tcf_proto_base; (t=*tp) != NULL; tp = &t->next)
105 write_unlock(&cls_mod_lock);
109 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
110 struct tcf_proto *tp, unsigned long fh, int event);
113 /* Select new prio value from the range, managed by kernel. */
115 static __inline__ u32 tcf_auto_prio(struct tcf_proto *tp)
117 u32 first = TC_H_MAKE(0xC0000000U,0U);
125 /* Add/change/delete/get a filter node */
127 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
129 struct rtattr **tca = arg;
130 struct tcmsg *t = NLMSG_DATA(n);
131 u32 protocol = TC_H_MIN(t->tcm_info);
132 u32 prio = TC_H_MAJ(t->tcm_info);
134 u32 parent = t->tcm_parent;
135 struct net_device *dev;
137 struct tcf_proto **back, **chain;
138 struct tcf_proto *tp = NULL;
139 struct tcf_proto_ops *tp_ops;
140 struct Qdisc_class_ops *cops;
141 unsigned long cl = 0;
146 /* If no priority is given, user wants we allocated it. */
147 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
149 prio = TC_H_MAKE(0x80000000U,0U);
152 /* Find head of filter chain. */
155 if ((dev = __dev_get_by_index(t->tcm_ifindex)) == NULL)
160 q = dev->qdisc_sleeping;
162 } else if ((q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent))) == NULL)
165 /* Is it classful? */
166 if ((cops = q->ops->cl_ops) == NULL)
169 /* Do we search for filter, attached to class? */
170 if (TC_H_MIN(parent)) {
171 cl = cops->get(q, parent);
176 /* And the last stroke */
177 chain = cops->tcf_chain(q, cl);
182 /* Check the chain for existence of proto-tcf with this priority */
183 for (back = chain; (tp=*back) != NULL; back = &tp->next) {
184 if (tp->prio >= prio) {
185 if (tp->prio == prio) {
186 if (!nprio || (tp->protocol != protocol && protocol))
195 /* Proto-tcf does not exist, create new one */
197 if (tca[TCA_KIND-1] == NULL || !protocol)
201 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
205 /* Create new proto tcf */
208 if ((tp = kmalloc(sizeof(*tp), GFP_KERNEL)) == NULL)
210 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND-1]);
212 if (tp_ops==NULL && tca[TCA_KIND-1] != NULL) {
213 struct rtattr *kind = tca[TCA_KIND-1];
215 if (RTA_PAYLOAD(kind) <= IFNAMSIZ) {
216 request_module("cls_%s", (char*)RTA_DATA(kind));
217 tp_ops = tcf_proto_lookup_ops(kind);
221 if (tp_ops == NULL) {
226 memset(tp, 0, sizeof(*tp));
228 tp->protocol = protocol;
229 tp->prio = nprio ? : tcf_auto_prio(*back);
231 tp->classify = tp_ops->classify;
232 tp->classid = parent;
234 if (!try_module_get(tp_ops->owner)) {
238 if ((err = tp_ops->init(tp)) != 0) {
239 module_put(tp_ops->owner);
244 qdisc_lock_tree(dev);
247 qdisc_unlock_tree(dev);
249 } else if (tca[TCA_KIND-1] && rtattr_strcmp(tca[TCA_KIND-1], tp->ops->kind))
252 fh = tp->ops->get(tp, t->tcm_handle);
255 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
256 qdisc_lock_tree(dev);
258 qdisc_unlock_tree(dev);
260 tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
267 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
270 switch (n->nlmsg_type) {
273 if (n->nlmsg_flags&NLM_F_EXCL)
277 err = tp->ops->delete(tp, fh);
279 tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
282 err = tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
290 err = tp->ops->change(tp, cl, t->tcm_handle, tca, &fh);
292 tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
300 unsigned long tcf_set_class(struct tcf_proto *tp, unsigned long *clp,
303 unsigned long old_cl;
306 old_cl = __cls_set_class(clp, cl);
314 tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp, unsigned long fh,
315 u32 pid, u32 seq, unsigned flags, int event)
318 struct nlmsghdr *nlh;
319 unsigned char *b = skb->tail;
321 nlh = NLMSG_PUT(skb, pid, seq, event, sizeof(*tcm));
322 nlh->nlmsg_flags = flags;
323 tcm = NLMSG_DATA(nlh);
324 tcm->tcm_family = AF_UNSPEC;
325 tcm->tcm_ifindex = tp->q->dev->ifindex;
326 tcm->tcm_parent = tp->classid;
327 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
328 RTA_PUT(skb, TCA_KIND, IFNAMSIZ, tp->ops->kind);
329 tcm->tcm_handle = fh;
330 if (RTM_DELTFILTER != event) {
332 if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
335 nlh->nlmsg_len = skb->tail - b;
340 skb_trim(skb, b - skb->data);
344 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
345 struct tcf_proto *tp, unsigned long fh, int event)
348 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
350 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
354 if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) {
359 return rtnetlink_send(skb, pid, RTMGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
366 struct netlink_callback *cb;
369 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n, struct tcf_walker *arg)
371 struct tcf_dump_args *a = (void*)arg;
373 return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid,
374 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
377 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
381 struct net_device *dev;
383 struct tcf_proto *tp, **chain;
384 struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh);
385 unsigned long cl = 0;
386 struct Qdisc_class_ops *cops;
387 struct tcf_dump_args arg;
389 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
391 if ((dev = dev_get_by_index(tcm->tcm_ifindex)) == NULL)
394 read_lock_bh(&qdisc_tree_lock);
395 if (!tcm->tcm_parent)
396 q = dev->qdisc_sleeping;
398 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
401 if ((cops = q->ops->cl_ops) == NULL)
403 if (TC_H_MIN(tcm->tcm_parent)) {
404 cl = cops->get(q, tcm->tcm_parent);
408 chain = cops->tcf_chain(q, cl);
414 for (tp=*chain, t=0; tp; tp = tp->next, t++) {
415 if (t < s_t) continue;
416 if (TC_H_MAJ(tcm->tcm_info) &&
417 TC_H_MAJ(tcm->tcm_info) != tp->prio)
419 if (TC_H_MIN(tcm->tcm_info) &&
420 TC_H_MIN(tcm->tcm_info) != tp->protocol)
423 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
424 if (cb->args[1] == 0) {
425 if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid,
426 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER) <= 0) {
431 if (tp->ops->walk == NULL)
433 arg.w.fn = tcf_node_dump;
437 arg.w.skip = cb->args[1]-1;
439 tp->ops->walk(tp, &arg.w);
440 cb->args[1] = arg.w.count+1;
451 read_unlock_bh(&qdisc_tree_lock);
457 static int __init tc_filter_init(void)
459 struct rtnetlink_link *link_p = rtnetlink_links[PF_UNSPEC];
461 /* Setup rtnetlink links. It is made here to avoid
462 exporting large number of public symbols.
466 link_p[RTM_NEWTFILTER-RTM_BASE].doit = tc_ctl_tfilter;
467 link_p[RTM_DELTFILTER-RTM_BASE].doit = tc_ctl_tfilter;
468 link_p[RTM_GETTFILTER-RTM_BASE].doit = tc_ctl_tfilter;
469 link_p[RTM_GETTFILTER-RTM_BASE].dumpit = tc_dump_tfilter;
474 subsys_initcall(tc_filter_init);
476 EXPORT_SYMBOL(register_tcf_proto_ops);
477 EXPORT_SYMBOL(unregister_tcf_proto_ops);
478 EXPORT_SYMBOL(tcf_set_class);