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
16 #include <asm/uaccess.h>
17 #include <asm/system.h>
18 #include <asm/bitops.h>
19 #include <linux/config.h>
20 #include <linux/module.h>
21 #include <linux/types.h>
22 #include <linux/kernel.h>
23 #include <linux/sched.h>
24 #include <linux/string.h>
26 #include <linux/socket.h>
27 #include <linux/sockios.h>
29 #include <linux/errno.h>
30 #include <linux/interrupt.h>
31 #include <linux/netdevice.h>
32 #include <linux/skbuff.h>
33 #include <linux/rtnetlink.h>
34 #include <linux/init.h>
35 #include <linux/kmod.h>
37 #include <net/pkt_sched.h>
39 /* The list of all installed classifier types */
41 static struct tcf_proto_ops *tcf_proto_base;
43 /* Protects list of registered TC modules. It is pure SMP lock. */
44 static rwlock_t cls_mod_lock = RW_LOCK_UNLOCKED;
46 /* Find classifier type by string name */
48 struct tcf_proto_ops * tcf_proto_lookup_ops(struct rtattr *kind)
50 struct tcf_proto_ops *t = NULL;
53 read_lock(&cls_mod_lock);
54 for (t = tcf_proto_base; t; t = t->next) {
55 if (rtattr_strcmp(kind, t->kind) == 0)
58 read_unlock(&cls_mod_lock);
63 /* Register(unregister) new classifier type */
65 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
67 struct tcf_proto_ops *t, **tp;
70 write_lock(&cls_mod_lock);
71 for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
72 if (!strcmp(ops->kind, t->kind))
79 write_unlock(&cls_mod_lock);
83 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
85 struct tcf_proto_ops *t, **tp;
88 write_lock(&cls_mod_lock);
89 for (tp = &tcf_proto_base; (t=*tp) != NULL; tp = &t->next)
98 write_unlock(&cls_mod_lock);
102 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
103 struct tcf_proto *tp, unsigned long fh, int event);
106 /* Select new prio value from the range, managed by kernel. */
108 static __inline__ u32 tcf_auto_prio(struct tcf_proto *tp)
110 u32 first = TC_H_MAKE(0xC0000000U,0U);
118 /* Add/change/delete/get a filter node */
120 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
122 struct rtattr **tca = arg;
123 struct tcmsg *t = NLMSG_DATA(n);
124 u32 protocol = TC_H_MIN(t->tcm_info);
125 u32 prio = TC_H_MAJ(t->tcm_info);
127 u32 parent = t->tcm_parent;
128 struct net_device *dev;
130 struct tcf_proto **back, **chain;
131 struct tcf_proto *tp = NULL;
132 struct tcf_proto_ops *tp_ops;
133 struct Qdisc_class_ops *cops;
134 unsigned long cl = 0;
139 /* If no priority is given, user wants we allocated it. */
140 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
142 prio = TC_H_MAKE(0x80000000U,0U);
145 /* Find head of filter chain. */
148 if ((dev = __dev_get_by_index(t->tcm_ifindex)) == NULL)
153 q = dev->qdisc_sleeping;
155 } else if ((q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent))) == NULL)
158 /* Is it classful? */
159 if ((cops = q->ops->cl_ops) == NULL)
162 /* Do we search for filter, attached to class? */
163 if (TC_H_MIN(parent)) {
164 cl = cops->get(q, parent);
169 /* And the last stroke */
170 chain = cops->tcf_chain(q, cl);
175 /* Check the chain for existence of proto-tcf with this priority */
176 for (back = chain; (tp=*back) != NULL; back = &tp->next) {
177 if (tp->prio >= prio) {
178 if (tp->prio == prio) {
179 if (!nprio || (tp->protocol != protocol && protocol))
188 /* Proto-tcf does not exist, create new one */
190 if (tca[TCA_KIND-1] == NULL || !protocol)
194 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
198 /* Create new proto tcf */
201 if ((tp = kmalloc(sizeof(*tp), GFP_KERNEL)) == NULL)
203 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND-1]);
205 if (tp_ops==NULL && tca[TCA_KIND-1] != NULL) {
206 struct rtattr *kind = tca[TCA_KIND-1];
208 if (RTA_PAYLOAD(kind) <= IFNAMSIZ) {
209 request_module("cls_%s", (char*)RTA_DATA(kind));
210 tp_ops = tcf_proto_lookup_ops(kind);
214 if (tp_ops == NULL) {
219 memset(tp, 0, sizeof(*tp));
221 tp->protocol = protocol;
222 tp->prio = nprio ? : tcf_auto_prio(*back);
224 tp->classify = tp_ops->classify;
225 tp->classid = parent;
227 if (!try_module_get(tp_ops->owner) ||
228 (err = tp_ops->init(tp)) != 0) {
232 write_lock(&qdisc_tree_lock);
233 spin_lock_bh(&dev->queue_lock);
236 spin_unlock_bh(&dev->queue_lock);
237 write_unlock(&qdisc_tree_lock);
238 } else if (tca[TCA_KIND-1] && rtattr_strcmp(tca[TCA_KIND-1], tp->ops->kind))
241 fh = tp->ops->get(tp, t->tcm_handle);
244 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
245 write_lock(&qdisc_tree_lock);
246 spin_lock_bh(&dev->queue_lock);
248 spin_unlock_bh(&dev->queue_lock);
249 write_unlock(&qdisc_tree_lock);
256 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
259 switch (n->nlmsg_type) {
262 if (n->nlmsg_flags&NLM_F_EXCL)
266 err = tp->ops->delete(tp, fh);
269 err = tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
277 err = tp->ops->change(tp, cl, t->tcm_handle, tca, &fh);
279 tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
288 tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp, unsigned long fh,
289 u32 pid, u32 seq, unsigned flags, int event)
292 struct nlmsghdr *nlh;
293 unsigned char *b = skb->tail;
295 nlh = NLMSG_PUT(skb, pid, seq, event, sizeof(*tcm));
296 nlh->nlmsg_flags = flags;
297 tcm = NLMSG_DATA(nlh);
298 tcm->tcm_family = AF_UNSPEC;
299 tcm->tcm_ifindex = tp->q->dev->ifindex;
300 tcm->tcm_parent = tp->classid;
302 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
303 RTA_PUT(skb, TCA_KIND, IFNAMSIZ, tp->ops->kind);
304 if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
306 nlh->nlmsg_len = skb->tail - b;
311 skb_trim(skb, b - skb->data);
315 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
316 struct tcf_proto *tp, unsigned long fh, int event)
319 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
321 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
325 if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) {
330 return rtnetlink_send(skb, pid, RTMGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
337 struct netlink_callback *cb;
340 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n, struct tcf_walker *arg)
342 struct tcf_dump_args *a = (void*)arg;
344 return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid,
345 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
348 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
352 struct net_device *dev;
354 struct tcf_proto *tp, **chain;
355 struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh);
356 unsigned long cl = 0;
357 struct Qdisc_class_ops *cops;
358 struct tcf_dump_args arg;
360 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
362 if ((dev = dev_get_by_index(tcm->tcm_ifindex)) == NULL)
365 read_lock(&qdisc_tree_lock);
366 if (!tcm->tcm_parent)
367 q = dev->qdisc_sleeping;
369 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
372 if ((cops = q->ops->cl_ops) == NULL)
374 if (TC_H_MIN(tcm->tcm_parent)) {
375 cl = cops->get(q, tcm->tcm_parent);
379 chain = cops->tcf_chain(q, cl);
385 for (tp=*chain, t=0; tp; tp = tp->next, t++) {
386 if (t < s_t) continue;
387 if (TC_H_MAJ(tcm->tcm_info) &&
388 TC_H_MAJ(tcm->tcm_info) != tp->prio)
390 if (TC_H_MIN(tcm->tcm_info) &&
391 TC_H_MIN(tcm->tcm_info) != tp->protocol)
394 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
395 if (cb->args[1] == 0) {
396 if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid,
397 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER) <= 0) {
402 if (tp->ops->walk == NULL)
404 arg.w.fn = tcf_node_dump;
408 arg.w.skip = cb->args[1]-1;
410 tp->ops->walk(tp, &arg.w);
411 cb->args[1] = arg.w.count+1;
422 read_unlock(&qdisc_tree_lock);
428 static int __init tc_filter_init(void)
430 struct rtnetlink_link *link_p = rtnetlink_links[PF_UNSPEC];
432 /* Setup rtnetlink links. It is made here to avoid
433 exporting large number of public symbols.
437 link_p[RTM_NEWTFILTER-RTM_BASE].doit = tc_ctl_tfilter;
438 link_p[RTM_DELTFILTER-RTM_BASE].doit = tc_ctl_tfilter;
439 link_p[RTM_GETTFILTER-RTM_BASE].doit = tc_ctl_tfilter;
440 link_p[RTM_GETTFILTER-RTM_BASE].dumpit = tc_dump_tfilter;
445 subsys_initcall(tc_filter_init);
447 EXPORT_SYMBOL(register_tcf_proto_ops);
448 EXPORT_SYMBOL(unregister_tcf_proto_ops);