fix a kmap leak in virtio_console
[linux/fpc-iii.git] / net / sched / cls_api.c
blob29a30a14c31596cc51028be8db46eb4df9756f6e
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 <linux/module.h>
18 #include <linux/types.h>
19 #include <linux/kernel.h>
20 #include <linux/string.h>
21 #include <linux/errno.h>
22 #include <linux/skbuff.h>
23 #include <linux/init.h>
24 #include <linux/kmod.h>
25 #include <linux/err.h>
26 #include <linux/slab.h>
27 #include <net/net_namespace.h>
28 #include <net/sock.h>
29 #include <net/netlink.h>
30 #include <net/pkt_sched.h>
31 #include <net/pkt_cls.h>
33 /* The list of all installed classifier types */
34 static LIST_HEAD(tcf_proto_base);
36 /* Protects list of registered TC modules. It is pure SMP lock. */
37 static DEFINE_RWLOCK(cls_mod_lock);
39 /* Find classifier type by string name */
41 static const struct tcf_proto_ops *tcf_proto_lookup_ops(struct nlattr *kind)
43 const struct tcf_proto_ops *t, *res = NULL;
45 if (kind) {
46 read_lock(&cls_mod_lock);
47 list_for_each_entry(t, &tcf_proto_base, head) {
48 if (nla_strcmp(kind, t->kind) == 0) {
49 if (try_module_get(t->owner))
50 res = t;
51 break;
54 read_unlock(&cls_mod_lock);
56 return res;
59 /* Register(unregister) new classifier type */
61 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
63 struct tcf_proto_ops *t;
64 int rc = -EEXIST;
66 write_lock(&cls_mod_lock);
67 list_for_each_entry(t, &tcf_proto_base, head)
68 if (!strcmp(ops->kind, t->kind))
69 goto out;
71 list_add_tail(&ops->head, &tcf_proto_base);
72 rc = 0;
73 out:
74 write_unlock(&cls_mod_lock);
75 return rc;
77 EXPORT_SYMBOL(register_tcf_proto_ops);
79 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
81 struct tcf_proto_ops *t;
82 int rc = -ENOENT;
84 write_lock(&cls_mod_lock);
85 list_for_each_entry(t, &tcf_proto_base, head) {
86 if (t == ops) {
87 list_del(&t->head);
88 rc = 0;
89 break;
92 write_unlock(&cls_mod_lock);
93 return rc;
95 EXPORT_SYMBOL(unregister_tcf_proto_ops);
97 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
98 struct nlmsghdr *n, struct tcf_proto *tp,
99 unsigned long fh, int event);
102 /* Select new prio value from the range, managed by kernel. */
104 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
106 u32 first = TC_H_MAKE(0xC0000000U, 0U);
108 if (tp)
109 first = tp->prio - 1;
111 return first;
114 /* Add/change/delete/get a filter node */
116 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n)
118 struct net *net = sock_net(skb->sk);
119 struct nlattr *tca[TCA_MAX + 1];
120 spinlock_t *root_lock;
121 struct tcmsg *t;
122 u32 protocol;
123 u32 prio;
124 u32 nprio;
125 u32 parent;
126 struct net_device *dev;
127 struct Qdisc *q;
128 struct tcf_proto **back, **chain;
129 struct tcf_proto *tp;
130 const struct tcf_proto_ops *tp_ops;
131 const struct Qdisc_class_ops *cops;
132 unsigned long cl;
133 unsigned long fh;
134 int err;
135 int tp_created = 0;
137 if ((n->nlmsg_type != RTM_GETTFILTER) && !capable(CAP_NET_ADMIN))
138 return -EPERM;
140 replay:
141 err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
142 if (err < 0)
143 return err;
145 t = nlmsg_data(n);
146 protocol = TC_H_MIN(t->tcm_info);
147 prio = TC_H_MAJ(t->tcm_info);
148 nprio = prio;
149 parent = t->tcm_parent;
150 cl = 0;
152 if (prio == 0) {
153 /* If no priority is given, user wants we allocated it. */
154 if (n->nlmsg_type != RTM_NEWTFILTER ||
155 !(n->nlmsg_flags & NLM_F_CREATE))
156 return -ENOENT;
157 prio = TC_H_MAKE(0x80000000U, 0U);
160 /* Find head of filter chain. */
162 /* Find link */
163 dev = __dev_get_by_index(net, t->tcm_ifindex);
164 if (dev == NULL)
165 return -ENODEV;
167 /* Find qdisc */
168 if (!parent) {
169 q = dev->qdisc;
170 parent = q->handle;
171 } else {
172 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
173 if (q == NULL)
174 return -EINVAL;
177 /* Is it classful? */
178 cops = q->ops->cl_ops;
179 if (!cops)
180 return -EINVAL;
182 if (cops->tcf_chain == NULL)
183 return -EOPNOTSUPP;
185 /* Do we search for filter, attached to class? */
186 if (TC_H_MIN(parent)) {
187 cl = cops->get(q, parent);
188 if (cl == 0)
189 return -ENOENT;
192 /* And the last stroke */
193 chain = cops->tcf_chain(q, cl);
194 err = -EINVAL;
195 if (chain == NULL)
196 goto errout;
198 /* Check the chain for existence of proto-tcf with this priority */
199 for (back = chain; (tp = *back) != NULL; back = &tp->next) {
200 if (tp->prio >= prio) {
201 if (tp->prio == prio) {
202 if (!nprio ||
203 (tp->protocol != protocol && protocol))
204 goto errout;
205 } else
206 tp = NULL;
207 break;
211 root_lock = qdisc_root_sleeping_lock(q);
213 if (tp == NULL) {
214 /* Proto-tcf does not exist, create new one */
216 if (tca[TCA_KIND] == NULL || !protocol)
217 goto errout;
219 err = -ENOENT;
220 if (n->nlmsg_type != RTM_NEWTFILTER ||
221 !(n->nlmsg_flags & NLM_F_CREATE))
222 goto errout;
225 /* Create new proto tcf */
227 err = -ENOBUFS;
228 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
229 if (tp == NULL)
230 goto errout;
231 err = -ENOENT;
232 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
233 if (tp_ops == NULL) {
234 #ifdef CONFIG_MODULES
235 struct nlattr *kind = tca[TCA_KIND];
236 char name[IFNAMSIZ];
238 if (kind != NULL &&
239 nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
240 rtnl_unlock();
241 request_module("cls_%s", name);
242 rtnl_lock();
243 tp_ops = tcf_proto_lookup_ops(kind);
244 /* We dropped the RTNL semaphore in order to
245 * perform the module load. So, even if we
246 * succeeded in loading the module we have to
247 * replay the request. We indicate this using
248 * -EAGAIN.
250 if (tp_ops != NULL) {
251 module_put(tp_ops->owner);
252 err = -EAGAIN;
255 #endif
256 kfree(tp);
257 goto errout;
259 tp->ops = tp_ops;
260 tp->protocol = protocol;
261 tp->prio = nprio ? : TC_H_MAJ(tcf_auto_prio(*back));
262 tp->q = q;
263 tp->classify = tp_ops->classify;
264 tp->classid = parent;
266 err = tp_ops->init(tp);
267 if (err != 0) {
268 module_put(tp_ops->owner);
269 kfree(tp);
270 goto errout;
273 tp_created = 1;
275 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
276 goto errout;
278 fh = tp->ops->get(tp, t->tcm_handle);
280 if (fh == 0) {
281 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
282 spin_lock_bh(root_lock);
283 *back = tp->next;
284 spin_unlock_bh(root_lock);
286 tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
287 tcf_destroy(tp);
288 err = 0;
289 goto errout;
292 err = -ENOENT;
293 if (n->nlmsg_type != RTM_NEWTFILTER ||
294 !(n->nlmsg_flags & NLM_F_CREATE))
295 goto errout;
296 } else {
297 switch (n->nlmsg_type) {
298 case RTM_NEWTFILTER:
299 err = -EEXIST;
300 if (n->nlmsg_flags & NLM_F_EXCL) {
301 if (tp_created)
302 tcf_destroy(tp);
303 goto errout;
305 break;
306 case RTM_DELTFILTER:
307 err = tp->ops->delete(tp, fh);
308 if (err == 0)
309 tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
310 goto errout;
311 case RTM_GETTFILTER:
312 err = tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
313 goto errout;
314 default:
315 err = -EINVAL;
316 goto errout;
320 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh);
321 if (err == 0) {
322 if (tp_created) {
323 spin_lock_bh(root_lock);
324 tp->next = *back;
325 *back = tp;
326 spin_unlock_bh(root_lock);
328 tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
329 } else {
330 if (tp_created)
331 tcf_destroy(tp);
334 errout:
335 if (cl)
336 cops->put(q, cl);
337 if (err == -EAGAIN)
338 /* Replay the request. */
339 goto replay;
340 return err;
343 static int tcf_fill_node(struct net *net, struct sk_buff *skb, struct tcf_proto *tp,
344 unsigned long fh, u32 portid, u32 seq, u16 flags, int event)
346 struct tcmsg *tcm;
347 struct nlmsghdr *nlh;
348 unsigned char *b = skb_tail_pointer(skb);
350 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
351 if (!nlh)
352 goto out_nlmsg_trim;
353 tcm = nlmsg_data(nlh);
354 tcm->tcm_family = AF_UNSPEC;
355 tcm->tcm__pad1 = 0;
356 tcm->tcm__pad2 = 0;
357 tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
358 tcm->tcm_parent = tp->classid;
359 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
360 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
361 goto nla_put_failure;
362 tcm->tcm_handle = fh;
363 if (RTM_DELTFILTER != event) {
364 tcm->tcm_handle = 0;
365 if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
366 goto nla_put_failure;
368 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
369 return skb->len;
371 out_nlmsg_trim:
372 nla_put_failure:
373 nlmsg_trim(skb, b);
374 return -1;
377 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
378 struct nlmsghdr *n, struct tcf_proto *tp,
379 unsigned long fh, int event)
381 struct sk_buff *skb;
382 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
384 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
385 if (!skb)
386 return -ENOBUFS;
388 if (tcf_fill_node(net, skb, tp, fh, portid, n->nlmsg_seq, 0, event) <= 0) {
389 kfree_skb(skb);
390 return -EINVAL;
393 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
394 n->nlmsg_flags & NLM_F_ECHO);
397 struct tcf_dump_args {
398 struct tcf_walker w;
399 struct sk_buff *skb;
400 struct netlink_callback *cb;
403 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
404 struct tcf_walker *arg)
406 struct tcf_dump_args *a = (void *)arg;
407 struct net *net = sock_net(a->skb->sk);
409 return tcf_fill_node(net, a->skb, tp, n, NETLINK_CB(a->cb->skb).portid,
410 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
413 /* called with RTNL */
414 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
416 struct net *net = sock_net(skb->sk);
417 int t;
418 int s_t;
419 struct net_device *dev;
420 struct Qdisc *q;
421 struct tcf_proto *tp, **chain;
422 struct tcmsg *tcm = nlmsg_data(cb->nlh);
423 unsigned long cl = 0;
424 const struct Qdisc_class_ops *cops;
425 struct tcf_dump_args arg;
427 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
428 return skb->len;
429 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
430 if (!dev)
431 return skb->len;
433 if (!tcm->tcm_parent)
434 q = dev->qdisc;
435 else
436 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
437 if (!q)
438 goto out;
439 cops = q->ops->cl_ops;
440 if (!cops)
441 goto errout;
442 if (cops->tcf_chain == NULL)
443 goto errout;
444 if (TC_H_MIN(tcm->tcm_parent)) {
445 cl = cops->get(q, tcm->tcm_parent);
446 if (cl == 0)
447 goto errout;
449 chain = cops->tcf_chain(q, cl);
450 if (chain == NULL)
451 goto errout;
453 s_t = cb->args[0];
455 for (tp = *chain, t = 0; tp; tp = tp->next, t++) {
456 if (t < s_t)
457 continue;
458 if (TC_H_MAJ(tcm->tcm_info) &&
459 TC_H_MAJ(tcm->tcm_info) != tp->prio)
460 continue;
461 if (TC_H_MIN(tcm->tcm_info) &&
462 TC_H_MIN(tcm->tcm_info) != tp->protocol)
463 continue;
464 if (t > s_t)
465 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
466 if (cb->args[1] == 0) {
467 if (tcf_fill_node(net, skb, tp, 0, NETLINK_CB(cb->skb).portid,
468 cb->nlh->nlmsg_seq, NLM_F_MULTI,
469 RTM_NEWTFILTER) <= 0)
470 break;
472 cb->args[1] = 1;
474 if (tp->ops->walk == NULL)
475 continue;
476 arg.w.fn = tcf_node_dump;
477 arg.skb = skb;
478 arg.cb = cb;
479 arg.w.stop = 0;
480 arg.w.skip = cb->args[1] - 1;
481 arg.w.count = 0;
482 tp->ops->walk(tp, &arg.w);
483 cb->args[1] = arg.w.count + 1;
484 if (arg.w.stop)
485 break;
488 cb->args[0] = t;
490 errout:
491 if (cl)
492 cops->put(q, cl);
493 out:
494 return skb->len;
497 void tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
499 #ifdef CONFIG_NET_CLS_ACT
500 tcf_action_destroy(&exts->actions, TCA_ACT_UNBIND);
501 INIT_LIST_HEAD(&exts->actions);
502 #endif
504 EXPORT_SYMBOL(tcf_exts_destroy);
506 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
507 struct nlattr *rate_tlv, struct tcf_exts *exts)
509 #ifdef CONFIG_NET_CLS_ACT
511 struct tc_action *act;
513 INIT_LIST_HEAD(&exts->actions);
514 if (exts->police && tb[exts->police]) {
515 act = tcf_action_init_1(net, tb[exts->police], rate_tlv,
516 "police", TCA_ACT_NOREPLACE,
517 TCA_ACT_BIND);
518 if (IS_ERR(act))
519 return PTR_ERR(act);
521 act->type = exts->type = TCA_OLD_COMPAT;
522 list_add(&act->list, &exts->actions);
523 } else if (exts->action && tb[exts->action]) {
524 int err;
525 err = tcf_action_init(net, tb[exts->action], rate_tlv,
526 NULL, TCA_ACT_NOREPLACE,
527 TCA_ACT_BIND, &exts->actions);
528 if (err)
529 return err;
532 #else
533 if ((exts->action && tb[exts->action]) ||
534 (exts->police && tb[exts->police]))
535 return -EOPNOTSUPP;
536 #endif
538 return 0;
540 EXPORT_SYMBOL(tcf_exts_validate);
542 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
543 struct tcf_exts *src)
545 #ifdef CONFIG_NET_CLS_ACT
546 if (!list_empty(&src->actions)) {
547 LIST_HEAD(tmp);
548 tcf_tree_lock(tp);
549 list_splice_init(&dst->actions, &tmp);
550 list_splice(&src->actions, &dst->actions);
551 tcf_tree_unlock(tp);
552 tcf_action_destroy(&tmp, TCA_ACT_UNBIND);
554 #endif
556 EXPORT_SYMBOL(tcf_exts_change);
558 #define tcf_exts_first_act(ext) \
559 list_first_entry(&(exts)->actions, struct tc_action, list)
561 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
563 #ifdef CONFIG_NET_CLS_ACT
564 if (exts->action && !list_empty(&exts->actions)) {
566 * again for backward compatible mode - we want
567 * to work with both old and new modes of entering
568 * tc data even if iproute2 was newer - jhs
570 struct nlattr *nest;
571 if (exts->type != TCA_OLD_COMPAT) {
572 nest = nla_nest_start(skb, exts->action);
573 if (nest == NULL)
574 goto nla_put_failure;
575 if (tcf_action_dump(skb, &exts->actions, 0, 0) < 0)
576 goto nla_put_failure;
577 nla_nest_end(skb, nest);
578 } else if (exts->police) {
579 struct tc_action *act = tcf_exts_first_act(exts);
580 nest = nla_nest_start(skb, exts->police);
581 if (nest == NULL || !act)
582 goto nla_put_failure;
583 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
584 goto nla_put_failure;
585 nla_nest_end(skb, nest);
588 #endif
589 return 0;
590 nla_put_failure: __attribute__ ((unused))
591 return -1;
593 EXPORT_SYMBOL(tcf_exts_dump);
596 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
598 #ifdef CONFIG_NET_CLS_ACT
599 struct tc_action *a = tcf_exts_first_act(exts);
600 if (tcf_action_copy_stats(skb, a, 1) < 0)
601 return -1;
602 #endif
603 return 0;
605 EXPORT_SYMBOL(tcf_exts_dump_stats);
607 static int __init tc_filter_init(void)
609 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL);
610 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL);
611 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
612 tc_dump_tfilter, NULL);
614 return 0;
617 subsys_initcall(tc_filter_init);