mm: always lock the root (oldest) anon_vma
[linux-2.6/libata-dev.git] / net / sched / cls_api.c
blob5fd0c28ef79a319ef1ab54474f6db98eb3b274cd
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/netlink.h>
26 #include <linux/err.h>
27 #include <linux/slab.h>
28 #include <net/net_namespace.h>
29 #include <net/sock.h>
30 #include <net/netlink.h>
31 #include <net/pkt_sched.h>
32 #include <net/pkt_cls.h>
34 /* The list of all installed classifier types */
36 static struct tcf_proto_ops *tcf_proto_base __read_mostly;
38 /* Protects list of registered TC modules. It is pure SMP lock. */
39 static DEFINE_RWLOCK(cls_mod_lock);
41 /* Find classifier type by string name */
43 static struct tcf_proto_ops *tcf_proto_lookup_ops(struct nlattr *kind)
45 struct tcf_proto_ops *t = NULL;
47 if (kind) {
48 read_lock(&cls_mod_lock);
49 for (t = tcf_proto_base; t; t = t->next) {
50 if (nla_strcmp(kind, t->kind) == 0) {
51 if (!try_module_get(t->owner))
52 t = NULL;
53 break;
56 read_unlock(&cls_mod_lock);
58 return t;
61 /* Register(unregister) new classifier type */
63 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
65 struct tcf_proto_ops *t, **tp;
66 int rc = -EEXIST;
68 write_lock(&cls_mod_lock);
69 for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
70 if (!strcmp(ops->kind, t->kind))
71 goto out;
73 ops->next = NULL;
74 *tp = ops;
75 rc = 0;
76 out:
77 write_unlock(&cls_mod_lock);
78 return rc;
80 EXPORT_SYMBOL(register_tcf_proto_ops);
82 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
84 struct tcf_proto_ops *t, **tp;
85 int rc = -ENOENT;
87 write_lock(&cls_mod_lock);
88 for (tp = &tcf_proto_base; (t=*tp) != NULL; tp = &t->next)
89 if (t == ops)
90 break;
92 if (!t)
93 goto out;
94 *tp = t->next;
95 rc = 0;
96 out:
97 write_unlock(&cls_mod_lock);
98 return rc;
100 EXPORT_SYMBOL(unregister_tcf_proto_ops);
102 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
103 struct nlmsghdr *n, struct tcf_proto *tp,
104 unsigned long fh, int event);
107 /* Select new prio value from the range, managed by kernel. */
109 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
111 u32 first = TC_H_MAKE(0xC0000000U, 0U);
113 if (tp)
114 first = tp->prio-1;
116 return first;
119 /* Add/change/delete/get a filter node */
121 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
123 struct net *net = sock_net(skb->sk);
124 struct nlattr *tca[TCA_MAX + 1];
125 spinlock_t *root_lock;
126 struct tcmsg *t;
127 u32 protocol;
128 u32 prio;
129 u32 nprio;
130 u32 parent;
131 struct net_device *dev;
132 struct Qdisc *q;
133 struct tcf_proto **back, **chain;
134 struct tcf_proto *tp;
135 struct tcf_proto_ops *tp_ops;
136 const struct Qdisc_class_ops *cops;
137 unsigned long cl;
138 unsigned long fh;
139 int err;
140 int tp_created = 0;
142 replay:
143 t = NLMSG_DATA(n);
144 protocol = TC_H_MIN(t->tcm_info);
145 prio = TC_H_MAJ(t->tcm_info);
146 nprio = prio;
147 parent = t->tcm_parent;
148 cl = 0;
150 if (prio == 0) {
151 /* If no priority is given, user wants we allocated it. */
152 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
153 return -ENOENT;
154 prio = TC_H_MAKE(0x80000000U, 0U);
157 /* Find head of filter chain. */
159 /* Find link */
160 dev = __dev_get_by_index(net, t->tcm_ifindex);
161 if (dev == NULL)
162 return -ENODEV;
164 err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
165 if (err < 0)
166 return err;
168 /* Find qdisc */
169 if (!parent) {
170 q = dev->qdisc;
171 parent = q->handle;
172 } else {
173 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
174 if (q == NULL)
175 return -EINVAL;
178 /* Is it classful? */
179 if ((cops = q->ops->cl_ops) == NULL)
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 || (tp->protocol != protocol && protocol))
203 goto errout;
204 } else
205 tp = NULL;
206 break;
210 root_lock = qdisc_root_sleeping_lock(q);
212 if (tp == NULL) {
213 /* Proto-tcf does not exist, create new one */
215 if (tca[TCA_KIND] == NULL || !protocol)
216 goto errout;
218 err = -ENOENT;
219 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
220 goto errout;
223 /* Create new proto tcf */
225 err = -ENOBUFS;
226 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
227 if (tp == NULL)
228 goto errout;
229 err = -ENOENT;
230 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
231 if (tp_ops == NULL) {
232 #ifdef CONFIG_MODULES
233 struct nlattr *kind = tca[TCA_KIND];
234 char name[IFNAMSIZ];
236 if (kind != NULL &&
237 nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
238 rtnl_unlock();
239 request_module("cls_%s", name);
240 rtnl_lock();
241 tp_ops = tcf_proto_lookup_ops(kind);
242 /* We dropped the RTNL semaphore in order to
243 * perform the module load. So, even if we
244 * succeeded in loading the module we have to
245 * replay the request. We indicate this using
246 * -EAGAIN.
248 if (tp_ops != NULL) {
249 module_put(tp_ops->owner);
250 err = -EAGAIN;
253 #endif
254 kfree(tp);
255 goto errout;
257 tp->ops = tp_ops;
258 tp->protocol = protocol;
259 tp->prio = nprio ? : TC_H_MAJ(tcf_auto_prio(*back));
260 tp->q = q;
261 tp->classify = tp_ops->classify;
262 tp->classid = parent;
264 err = tp_ops->init(tp);
265 if (err != 0) {
266 module_put(tp_ops->owner);
267 kfree(tp);
268 goto errout;
271 tp_created = 1;
273 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
274 goto errout;
276 fh = tp->ops->get(tp, t->tcm_handle);
278 if (fh == 0) {
279 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
280 spin_lock_bh(root_lock);
281 *back = tp->next;
282 spin_unlock_bh(root_lock);
284 tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
285 tcf_destroy(tp);
286 err = 0;
287 goto errout;
290 err = -ENOENT;
291 if (n->nlmsg_type != RTM_NEWTFILTER ||
292 !(n->nlmsg_flags & NLM_F_CREATE))
293 goto errout;
294 } else {
295 switch (n->nlmsg_type) {
296 case RTM_NEWTFILTER:
297 err = -EEXIST;
298 if (n->nlmsg_flags & NLM_F_EXCL) {
299 if (tp_created)
300 tcf_destroy(tp);
301 goto errout;
303 break;
304 case RTM_DELTFILTER:
305 err = tp->ops->delete(tp, fh);
306 if (err == 0)
307 tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
308 goto errout;
309 case RTM_GETTFILTER:
310 err = tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
311 goto errout;
312 default:
313 err = -EINVAL;
314 goto errout;
318 err = tp->ops->change(tp, cl, t->tcm_handle, tca, &fh);
319 if (err == 0) {
320 if (tp_created) {
321 spin_lock_bh(root_lock);
322 tp->next = *back;
323 *back = tp;
324 spin_unlock_bh(root_lock);
326 tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
327 } else {
328 if (tp_created)
329 tcf_destroy(tp);
332 errout:
333 if (cl)
334 cops->put(q, cl);
335 if (err == -EAGAIN)
336 /* Replay the request. */
337 goto replay;
338 return err;
341 static int tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp,
342 unsigned long fh, u32 pid, u32 seq, u16 flags, int event)
344 struct tcmsg *tcm;
345 struct nlmsghdr *nlh;
346 unsigned char *b = skb_tail_pointer(skb);
348 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
349 tcm = NLMSG_DATA(nlh);
350 tcm->tcm_family = AF_UNSPEC;
351 tcm->tcm__pad1 = 0;
352 tcm->tcm__pad2 = 0;
353 tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
354 tcm->tcm_parent = tp->classid;
355 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
356 NLA_PUT_STRING(skb, TCA_KIND, tp->ops->kind);
357 tcm->tcm_handle = fh;
358 if (RTM_DELTFILTER != event) {
359 tcm->tcm_handle = 0;
360 if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
361 goto nla_put_failure;
363 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
364 return skb->len;
366 nlmsg_failure:
367 nla_put_failure:
368 nlmsg_trim(skb, b);
369 return -1;
372 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
373 struct nlmsghdr *n, struct tcf_proto *tp,
374 unsigned long fh, int event)
376 struct sk_buff *skb;
377 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
379 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
380 if (!skb)
381 return -ENOBUFS;
383 if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) {
384 kfree_skb(skb);
385 return -EINVAL;
388 return rtnetlink_send(skb, net, pid, RTNLGRP_TC,
389 n->nlmsg_flags & NLM_F_ECHO);
392 struct tcf_dump_args {
393 struct tcf_walker w;
394 struct sk_buff *skb;
395 struct netlink_callback *cb;
398 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
399 struct tcf_walker *arg)
401 struct tcf_dump_args *a = (void *)arg;
403 return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid,
404 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
407 /* called with RTNL */
408 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
410 struct net *net = sock_net(skb->sk);
411 int t;
412 int s_t;
413 struct net_device *dev;
414 struct Qdisc *q;
415 struct tcf_proto *tp, **chain;
416 struct tcmsg *tcm = (struct tcmsg *)NLMSG_DATA(cb->nlh);
417 unsigned long cl = 0;
418 const struct Qdisc_class_ops *cops;
419 struct tcf_dump_args arg;
421 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
422 return skb->len;
423 if ((dev = __dev_get_by_index(net, tcm->tcm_ifindex)) == NULL)
424 return skb->len;
426 if (!tcm->tcm_parent)
427 q = dev->qdisc;
428 else
429 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
430 if (!q)
431 goto out;
432 if ((cops = q->ops->cl_ops) == NULL)
433 goto errout;
434 if (cops->tcf_chain == NULL)
435 goto errout;
436 if (TC_H_MIN(tcm->tcm_parent)) {
437 cl = cops->get(q, tcm->tcm_parent);
438 if (cl == 0)
439 goto errout;
441 chain = cops->tcf_chain(q, cl);
442 if (chain == NULL)
443 goto errout;
445 s_t = cb->args[0];
447 for (tp=*chain, t=0; tp; tp = tp->next, t++) {
448 if (t < s_t) continue;
449 if (TC_H_MAJ(tcm->tcm_info) &&
450 TC_H_MAJ(tcm->tcm_info) != tp->prio)
451 continue;
452 if (TC_H_MIN(tcm->tcm_info) &&
453 TC_H_MIN(tcm->tcm_info) != tp->protocol)
454 continue;
455 if (t > s_t)
456 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
457 if (cb->args[1] == 0) {
458 if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid,
459 cb->nlh->nlmsg_seq, NLM_F_MULTI,
460 RTM_NEWTFILTER) <= 0)
461 break;
463 cb->args[1] = 1;
465 if (tp->ops->walk == NULL)
466 continue;
467 arg.w.fn = tcf_node_dump;
468 arg.skb = skb;
469 arg.cb = cb;
470 arg.w.stop = 0;
471 arg.w.skip = cb->args[1]-1;
472 arg.w.count = 0;
473 tp->ops->walk(tp, &arg.w);
474 cb->args[1] = arg.w.count+1;
475 if (arg.w.stop)
476 break;
479 cb->args[0] = t;
481 errout:
482 if (cl)
483 cops->put(q, cl);
484 out:
485 return skb->len;
488 void tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
490 #ifdef CONFIG_NET_CLS_ACT
491 if (exts->action) {
492 tcf_action_destroy(exts->action, TCA_ACT_UNBIND);
493 exts->action = NULL;
495 #endif
497 EXPORT_SYMBOL(tcf_exts_destroy);
499 int tcf_exts_validate(struct tcf_proto *tp, struct nlattr **tb,
500 struct nlattr *rate_tlv, struct tcf_exts *exts,
501 const struct tcf_ext_map *map)
503 memset(exts, 0, sizeof(*exts));
505 #ifdef CONFIG_NET_CLS_ACT
507 struct tc_action *act;
509 if (map->police && tb[map->police]) {
510 act = tcf_action_init_1(tb[map->police], rate_tlv,
511 "police", TCA_ACT_NOREPLACE,
512 TCA_ACT_BIND);
513 if (IS_ERR(act))
514 return PTR_ERR(act);
516 act->type = TCA_OLD_COMPAT;
517 exts->action = act;
518 } else if (map->action && tb[map->action]) {
519 act = tcf_action_init(tb[map->action], rate_tlv, NULL,
520 TCA_ACT_NOREPLACE, TCA_ACT_BIND);
521 if (IS_ERR(act))
522 return PTR_ERR(act);
524 exts->action = act;
527 #else
528 if ((map->action && tb[map->action]) ||
529 (map->police && tb[map->police]))
530 return -EOPNOTSUPP;
531 #endif
533 return 0;
535 EXPORT_SYMBOL(tcf_exts_validate);
537 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
538 struct tcf_exts *src)
540 #ifdef CONFIG_NET_CLS_ACT
541 if (src->action) {
542 struct tc_action *act;
543 tcf_tree_lock(tp);
544 act = dst->action;
545 dst->action = src->action;
546 tcf_tree_unlock(tp);
547 if (act)
548 tcf_action_destroy(act, TCA_ACT_UNBIND);
550 #endif
552 EXPORT_SYMBOL(tcf_exts_change);
554 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts,
555 const struct tcf_ext_map *map)
557 #ifdef CONFIG_NET_CLS_ACT
558 if (map->action && exts->action) {
560 * again for backward compatible mode - we want
561 * to work with both old and new modes of entering
562 * tc data even if iproute2 was newer - jhs
564 struct nlattr *nest;
566 if (exts->action->type != TCA_OLD_COMPAT) {
567 nest = nla_nest_start(skb, map->action);
568 if (nest == NULL)
569 goto nla_put_failure;
570 if (tcf_action_dump(skb, exts->action, 0, 0) < 0)
571 goto nla_put_failure;
572 nla_nest_end(skb, nest);
573 } else if (map->police) {
574 nest = nla_nest_start(skb, map->police);
575 if (nest == NULL)
576 goto nla_put_failure;
577 if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0)
578 goto nla_put_failure;
579 nla_nest_end(skb, nest);
582 #endif
583 return 0;
584 nla_put_failure: __attribute__ ((unused))
585 return -1;
587 EXPORT_SYMBOL(tcf_exts_dump);
590 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts,
591 const struct tcf_ext_map *map)
593 #ifdef CONFIG_NET_CLS_ACT
594 if (exts->action)
595 if (tcf_action_copy_stats(skb, exts->action, 1) < 0)
596 goto nla_put_failure;
597 #endif
598 return 0;
599 nla_put_failure: __attribute__ ((unused))
600 return -1;
602 EXPORT_SYMBOL(tcf_exts_dump_stats);
604 static int __init tc_filter_init(void)
606 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL);
607 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL);
608 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
609 tc_dump_tfilter);
611 return 0;
614 subsys_initcall(tc_filter_init);