Linux v2.6.17-rc2
[linux-2.6/next.git] / net / sched / cls_api.c
blobb4d89fbb378212966b93d9cbc25596ce8add064c
1 /*
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>
11 * Changes:
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/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>
26 #include <linux/mm.h>
27 #include <linux/socket.h>
28 #include <linux/sockios.h>
29 #include <linux/in.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>
37 #include <net/sock.h>
38 #include <net/pkt_sched.h>
39 #include <net/pkt_cls.h>
41 #if 0 /* control */
42 #define DPRINTK(format,args...) printk(KERN_DEBUG format,##args)
43 #else
44 #define DPRINTK(format,args...)
45 #endif
47 /* The list of all installed classifier types */
49 static struct tcf_proto_ops *tcf_proto_base;
51 /* Protects list of registered TC modules. It is pure SMP lock. */
52 static DEFINE_RWLOCK(cls_mod_lock);
54 /* Find classifier type by string name */
56 static struct tcf_proto_ops * tcf_proto_lookup_ops(struct rtattr *kind)
58 struct tcf_proto_ops *t = NULL;
60 if (kind) {
61 read_lock(&cls_mod_lock);
62 for (t = tcf_proto_base; t; t = t->next) {
63 if (rtattr_strcmp(kind, t->kind) == 0) {
64 if (!try_module_get(t->owner))
65 t = NULL;
66 break;
69 read_unlock(&cls_mod_lock);
71 return t;
74 /* Register(unregister) new classifier type */
76 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
78 struct tcf_proto_ops *t, **tp;
79 int rc = -EEXIST;
81 write_lock(&cls_mod_lock);
82 for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
83 if (!strcmp(ops->kind, t->kind))
84 goto out;
86 ops->next = NULL;
87 *tp = ops;
88 rc = 0;
89 out:
90 write_unlock(&cls_mod_lock);
91 return rc;
94 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
96 struct tcf_proto_ops *t, **tp;
97 int rc = -ENOENT;
99 write_lock(&cls_mod_lock);
100 for (tp = &tcf_proto_base; (t=*tp) != NULL; tp = &t->next)
101 if (t == ops)
102 break;
104 if (!t)
105 goto out;
106 *tp = t->next;
107 rc = 0;
108 out:
109 write_unlock(&cls_mod_lock);
110 return rc;
113 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
114 struct tcf_proto *tp, unsigned long fh, int event);
117 /* Select new prio value from the range, managed by kernel. */
119 static __inline__ u32 tcf_auto_prio(struct tcf_proto *tp)
121 u32 first = TC_H_MAKE(0xC0000000U,0U);
123 if (tp)
124 first = tp->prio-1;
126 return first;
129 /* Add/change/delete/get a filter node */
131 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
133 struct rtattr **tca;
134 struct tcmsg *t;
135 u32 protocol;
136 u32 prio;
137 u32 nprio;
138 u32 parent;
139 struct net_device *dev;
140 struct Qdisc *q;
141 struct tcf_proto **back, **chain;
142 struct tcf_proto *tp;
143 struct tcf_proto_ops *tp_ops;
144 struct Qdisc_class_ops *cops;
145 unsigned long cl;
146 unsigned long fh;
147 int err;
149 replay:
150 tca = arg;
151 t = NLMSG_DATA(n);
152 protocol = TC_H_MIN(t->tcm_info);
153 prio = TC_H_MAJ(t->tcm_info);
154 nprio = prio;
155 parent = t->tcm_parent;
156 cl = 0;
158 if (prio == 0) {
159 /* If no priority is given, user wants we allocated it. */
160 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
161 return -ENOENT;
162 prio = TC_H_MAKE(0x80000000U,0U);
165 /* Find head of filter chain. */
167 /* Find link */
168 if ((dev = __dev_get_by_index(t->tcm_ifindex)) == NULL)
169 return -ENODEV;
171 /* Find qdisc */
172 if (!parent) {
173 q = dev->qdisc_sleeping;
174 parent = q->handle;
175 } else if ((q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent))) == NULL)
176 return -EINVAL;
178 /* Is it classful? */
179 if ((cops = q->ops->cl_ops) == NULL)
180 return -EINVAL;
182 /* Do we search for filter, attached to class? */
183 if (TC_H_MIN(parent)) {
184 cl = cops->get(q, parent);
185 if (cl == 0)
186 return -ENOENT;
189 /* And the last stroke */
190 chain = cops->tcf_chain(q, cl);
191 err = -EINVAL;
192 if (chain == NULL)
193 goto errout;
195 /* Check the chain for existence of proto-tcf with this priority */
196 for (back = chain; (tp=*back) != NULL; back = &tp->next) {
197 if (tp->prio >= prio) {
198 if (tp->prio == prio) {
199 if (!nprio || (tp->protocol != protocol && protocol))
200 goto errout;
201 } else
202 tp = NULL;
203 break;
207 if (tp == NULL) {
208 /* Proto-tcf does not exist, create new one */
210 if (tca[TCA_KIND-1] == NULL || !protocol)
211 goto errout;
213 err = -ENOENT;
214 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
215 goto errout;
218 /* Create new proto tcf */
220 err = -ENOBUFS;
221 if ((tp = kmalloc(sizeof(*tp), GFP_KERNEL)) == NULL)
222 goto errout;
223 err = -EINVAL;
224 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND-1]);
225 if (tp_ops == NULL) {
226 #ifdef CONFIG_KMOD
227 struct rtattr *kind = tca[TCA_KIND-1];
228 char name[IFNAMSIZ];
230 if (kind != NULL &&
231 rtattr_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
232 rtnl_unlock();
233 request_module("cls_%s", name);
234 rtnl_lock();
235 tp_ops = tcf_proto_lookup_ops(kind);
236 /* We dropped the RTNL semaphore in order to
237 * perform the module load. So, even if we
238 * succeeded in loading the module we have to
239 * replay the request. We indicate this using
240 * -EAGAIN.
242 if (tp_ops != NULL) {
243 module_put(tp_ops->owner);
244 err = -EAGAIN;
247 #endif
248 kfree(tp);
249 goto errout;
251 memset(tp, 0, sizeof(*tp));
252 tp->ops = tp_ops;
253 tp->protocol = protocol;
254 tp->prio = nprio ? : tcf_auto_prio(*back);
255 tp->q = q;
256 tp->classify = tp_ops->classify;
257 tp->classid = parent;
258 if ((err = tp_ops->init(tp)) != 0) {
259 module_put(tp_ops->owner);
260 kfree(tp);
261 goto errout;
264 qdisc_lock_tree(dev);
265 tp->next = *back;
266 *back = tp;
267 qdisc_unlock_tree(dev);
269 } else if (tca[TCA_KIND-1] && rtattr_strcmp(tca[TCA_KIND-1], tp->ops->kind))
270 goto errout;
272 fh = tp->ops->get(tp, t->tcm_handle);
274 if (fh == 0) {
275 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
276 qdisc_lock_tree(dev);
277 *back = tp->next;
278 qdisc_unlock_tree(dev);
280 tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
281 tcf_destroy(tp);
282 err = 0;
283 goto errout;
286 err = -ENOENT;
287 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
288 goto errout;
289 } else {
290 switch (n->nlmsg_type) {
291 case RTM_NEWTFILTER:
292 err = -EEXIST;
293 if (n->nlmsg_flags&NLM_F_EXCL)
294 goto errout;
295 break;
296 case RTM_DELTFILTER:
297 err = tp->ops->delete(tp, fh);
298 if (err == 0)
299 tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
300 goto errout;
301 case RTM_GETTFILTER:
302 err = tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
303 goto errout;
304 default:
305 err = -EINVAL;
306 goto errout;
310 err = tp->ops->change(tp, cl, t->tcm_handle, tca, &fh);
311 if (err == 0)
312 tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
314 errout:
315 if (cl)
316 cops->put(q, cl);
317 if (err == -EAGAIN)
318 /* Replay the request. */
319 goto replay;
320 return err;
323 static int
324 tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp, unsigned long fh,
325 u32 pid, u32 seq, u16 flags, int event)
327 struct tcmsg *tcm;
328 struct nlmsghdr *nlh;
329 unsigned char *b = skb->tail;
331 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
332 tcm = NLMSG_DATA(nlh);
333 tcm->tcm_family = AF_UNSPEC;
334 tcm->tcm__pad1 = 0;
335 tcm->tcm__pad1 = 0;
336 tcm->tcm_ifindex = tp->q->dev->ifindex;
337 tcm->tcm_parent = tp->classid;
338 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
339 RTA_PUT(skb, TCA_KIND, IFNAMSIZ, tp->ops->kind);
340 tcm->tcm_handle = fh;
341 if (RTM_DELTFILTER != event) {
342 tcm->tcm_handle = 0;
343 if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
344 goto rtattr_failure;
346 nlh->nlmsg_len = skb->tail - b;
347 return skb->len;
349 nlmsg_failure:
350 rtattr_failure:
351 skb_trim(skb, b - skb->data);
352 return -1;
355 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
356 struct tcf_proto *tp, unsigned long fh, int event)
358 struct sk_buff *skb;
359 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
361 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
362 if (!skb)
363 return -ENOBUFS;
365 if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) {
366 kfree_skb(skb);
367 return -EINVAL;
370 return rtnetlink_send(skb, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
373 struct tcf_dump_args
375 struct tcf_walker w;
376 struct sk_buff *skb;
377 struct netlink_callback *cb;
380 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n, struct tcf_walker *arg)
382 struct tcf_dump_args *a = (void*)arg;
384 return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid,
385 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
388 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
390 int t;
391 int s_t;
392 struct net_device *dev;
393 struct Qdisc *q;
394 struct tcf_proto *tp, **chain;
395 struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh);
396 unsigned long cl = 0;
397 struct Qdisc_class_ops *cops;
398 struct tcf_dump_args arg;
400 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
401 return skb->len;
402 if ((dev = dev_get_by_index(tcm->tcm_ifindex)) == NULL)
403 return skb->len;
405 read_lock_bh(&qdisc_tree_lock);
406 if (!tcm->tcm_parent)
407 q = dev->qdisc_sleeping;
408 else
409 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
410 if (!q)
411 goto out;
412 if ((cops = q->ops->cl_ops) == NULL)
413 goto errout;
414 if (TC_H_MIN(tcm->tcm_parent)) {
415 cl = cops->get(q, tcm->tcm_parent);
416 if (cl == 0)
417 goto errout;
419 chain = cops->tcf_chain(q, cl);
420 if (chain == NULL)
421 goto errout;
423 s_t = cb->args[0];
425 for (tp=*chain, t=0; tp; tp = tp->next, t++) {
426 if (t < s_t) continue;
427 if (TC_H_MAJ(tcm->tcm_info) &&
428 TC_H_MAJ(tcm->tcm_info) != tp->prio)
429 continue;
430 if (TC_H_MIN(tcm->tcm_info) &&
431 TC_H_MIN(tcm->tcm_info) != tp->protocol)
432 continue;
433 if (t > s_t)
434 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
435 if (cb->args[1] == 0) {
436 if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid,
437 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER) <= 0) {
438 break;
440 cb->args[1] = 1;
442 if (tp->ops->walk == NULL)
443 continue;
444 arg.w.fn = tcf_node_dump;
445 arg.skb = skb;
446 arg.cb = cb;
447 arg.w.stop = 0;
448 arg.w.skip = cb->args[1]-1;
449 arg.w.count = 0;
450 tp->ops->walk(tp, &arg.w);
451 cb->args[1] = arg.w.count+1;
452 if (arg.w.stop)
453 break;
456 cb->args[0] = t;
458 errout:
459 if (cl)
460 cops->put(q, cl);
461 out:
462 read_unlock_bh(&qdisc_tree_lock);
463 dev_put(dev);
464 return skb->len;
467 void
468 tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
470 #ifdef CONFIG_NET_CLS_ACT
471 if (exts->action) {
472 tcf_action_destroy(exts->action, TCA_ACT_UNBIND);
473 exts->action = NULL;
475 #elif defined CONFIG_NET_CLS_POLICE
476 if (exts->police) {
477 tcf_police_release(exts->police, TCA_ACT_UNBIND);
478 exts->police = NULL;
480 #endif
485 tcf_exts_validate(struct tcf_proto *tp, struct rtattr **tb,
486 struct rtattr *rate_tlv, struct tcf_exts *exts,
487 struct tcf_ext_map *map)
489 memset(exts, 0, sizeof(*exts));
491 #ifdef CONFIG_NET_CLS_ACT
493 int err;
494 struct tc_action *act;
496 if (map->police && tb[map->police-1]) {
497 act = tcf_action_init_1(tb[map->police-1], rate_tlv, "police",
498 TCA_ACT_NOREPLACE, TCA_ACT_BIND, &err);
499 if (act == NULL)
500 return err;
502 act->type = TCA_OLD_COMPAT;
503 exts->action = act;
504 } else if (map->action && tb[map->action-1]) {
505 act = tcf_action_init(tb[map->action-1], rate_tlv, NULL,
506 TCA_ACT_NOREPLACE, TCA_ACT_BIND, &err);
507 if (act == NULL)
508 return err;
510 exts->action = act;
513 #elif defined CONFIG_NET_CLS_POLICE
514 if (map->police && tb[map->police-1]) {
515 struct tcf_police *p;
517 p = tcf_police_locate(tb[map->police-1], rate_tlv);
518 if (p == NULL)
519 return -EINVAL;
521 exts->police = p;
522 } else if (map->action && tb[map->action-1])
523 return -EOPNOTSUPP;
524 #else
525 if ((map->action && tb[map->action-1]) ||
526 (map->police && tb[map->police-1]))
527 return -EOPNOTSUPP;
528 #endif
530 return 0;
533 void
534 tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
535 struct tcf_exts *src)
537 #ifdef CONFIG_NET_CLS_ACT
538 if (src->action) {
539 struct tc_action *act;
540 tcf_tree_lock(tp);
541 act = xchg(&dst->action, src->action);
542 tcf_tree_unlock(tp);
543 if (act)
544 tcf_action_destroy(act, TCA_ACT_UNBIND);
546 #elif defined CONFIG_NET_CLS_POLICE
547 if (src->police) {
548 struct tcf_police *p;
549 tcf_tree_lock(tp);
550 p = xchg(&dst->police, src->police);
551 tcf_tree_unlock(tp);
552 if (p)
553 tcf_police_release(p, TCA_ACT_UNBIND);
555 #endif
559 tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts,
560 struct tcf_ext_map *map)
562 #ifdef CONFIG_NET_CLS_ACT
563 if (map->action && exts->action) {
565 * again for backward compatible mode - we want
566 * to work with both old and new modes of entering
567 * tc data even if iproute2 was newer - jhs
569 struct rtattr * p_rta = (struct rtattr*) skb->tail;
571 if (exts->action->type != TCA_OLD_COMPAT) {
572 RTA_PUT(skb, map->action, 0, NULL);
573 if (tcf_action_dump(skb, exts->action, 0, 0) < 0)
574 goto rtattr_failure;
575 p_rta->rta_len = skb->tail - (u8*)p_rta;
576 } else if (map->police) {
577 RTA_PUT(skb, map->police, 0, NULL);
578 if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0)
579 goto rtattr_failure;
580 p_rta->rta_len = skb->tail - (u8*)p_rta;
583 #elif defined CONFIG_NET_CLS_POLICE
584 if (map->police && exts->police) {
585 struct rtattr * p_rta = (struct rtattr*) skb->tail;
587 RTA_PUT(skb, map->police, 0, NULL);
589 if (tcf_police_dump(skb, exts->police) < 0)
590 goto rtattr_failure;
592 p_rta->rta_len = skb->tail - (u8*)p_rta;
594 #endif
595 return 0;
596 rtattr_failure: __attribute__ ((unused))
597 return -1;
601 tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts,
602 struct tcf_ext_map *map)
604 #ifdef CONFIG_NET_CLS_ACT
605 if (exts->action)
606 if (tcf_action_copy_stats(skb, exts->action, 1) < 0)
607 goto rtattr_failure;
608 #elif defined CONFIG_NET_CLS_POLICE
609 if (exts->police)
610 if (tcf_police_dump_stats(skb, exts->police) < 0)
611 goto rtattr_failure;
612 #endif
613 return 0;
614 rtattr_failure: __attribute__ ((unused))
615 return -1;
618 static int __init tc_filter_init(void)
620 struct rtnetlink_link *link_p = rtnetlink_links[PF_UNSPEC];
622 /* Setup rtnetlink links. It is made here to avoid
623 exporting large number of public symbols.
626 if (link_p) {
627 link_p[RTM_NEWTFILTER-RTM_BASE].doit = tc_ctl_tfilter;
628 link_p[RTM_DELTFILTER-RTM_BASE].doit = tc_ctl_tfilter;
629 link_p[RTM_GETTFILTER-RTM_BASE].doit = tc_ctl_tfilter;
630 link_p[RTM_GETTFILTER-RTM_BASE].dumpit = tc_dump_tfilter;
632 return 0;
635 subsys_initcall(tc_filter_init);
637 EXPORT_SYMBOL(register_tcf_proto_ops);
638 EXPORT_SYMBOL(unregister_tcf_proto_ops);
639 EXPORT_SYMBOL(tcf_exts_validate);
640 EXPORT_SYMBOL(tcf_exts_destroy);
641 EXPORT_SYMBOL(tcf_exts_change);
642 EXPORT_SYMBOL(tcf_exts_dump);
643 EXPORT_SYMBOL(tcf_exts_dump_stats);