net_sched: make traffic control network namespace aware
[pandora-kernel.git] / net / sched / cls_api.c
1 /*
2  * net/sched/cls_api.c  Packet classifier API.
3  *
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.
8  *
9  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10  *
11  * Changes:
12  *
13  * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
14  *
15  */
16
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 <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>
32
33 /* The list of all installed classifier types */
34
35 static struct tcf_proto_ops *tcf_proto_base __read_mostly;
36
37 /* Protects list of registered TC modules. It is pure SMP lock. */
38 static DEFINE_RWLOCK(cls_mod_lock);
39
40 /* Find classifier type by string name */
41
42 static struct tcf_proto_ops *tcf_proto_lookup_ops(struct nlattr *kind)
43 {
44         struct tcf_proto_ops *t = NULL;
45
46         if (kind) {
47                 read_lock(&cls_mod_lock);
48                 for (t = tcf_proto_base; t; t = t->next) {
49                         if (nla_strcmp(kind, t->kind) == 0) {
50                                 if (!try_module_get(t->owner))
51                                         t = NULL;
52                                 break;
53                         }
54                 }
55                 read_unlock(&cls_mod_lock);
56         }
57         return t;
58 }
59
60 /* Register(unregister) new classifier type */
61
62 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
63 {
64         struct tcf_proto_ops *t, **tp;
65         int rc = -EEXIST;
66
67         write_lock(&cls_mod_lock);
68         for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
69                 if (!strcmp(ops->kind, t->kind))
70                         goto out;
71
72         ops->next = NULL;
73         *tp = ops;
74         rc = 0;
75 out:
76         write_unlock(&cls_mod_lock);
77         return rc;
78 }
79 EXPORT_SYMBOL(register_tcf_proto_ops);
80
81 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
82 {
83         struct tcf_proto_ops *t, **tp;
84         int rc = -ENOENT;
85
86         write_lock(&cls_mod_lock);
87         for (tp = &tcf_proto_base; (t=*tp) != NULL; tp = &t->next)
88                 if (t == ops)
89                         break;
90
91         if (!t)
92                 goto out;
93         *tp = t->next;
94         rc = 0;
95 out:
96         write_unlock(&cls_mod_lock);
97         return rc;
98 }
99 EXPORT_SYMBOL(unregister_tcf_proto_ops);
100
101 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
102                           struct nlmsghdr *n, struct tcf_proto *tp,
103                           unsigned long fh, int event);
104
105
106 /* Select new prio value from the range, managed by kernel. */
107
108 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
109 {
110         u32 first = TC_H_MAKE(0xC0000000U, 0U);
111
112         if (tp)
113                 first = tp->prio-1;
114
115         return first;
116 }
117
118 /* Add/change/delete/get a filter node */
119
120 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
121 {
122         struct net *net = sock_net(skb->sk);
123         struct nlattr *tca[TCA_MAX + 1];
124         spinlock_t *root_lock;
125         struct tcmsg *t;
126         u32 protocol;
127         u32 prio;
128         u32 nprio;
129         u32 parent;
130         struct net_device *dev;
131         struct Qdisc  *q;
132         struct tcf_proto **back, **chain;
133         struct tcf_proto *tp;
134         struct tcf_proto_ops *tp_ops;
135         const struct Qdisc_class_ops *cops;
136         unsigned long cl;
137         unsigned long fh;
138         int err;
139         int tp_created = 0;
140
141 replay:
142         t = NLMSG_DATA(n);
143         protocol = TC_H_MIN(t->tcm_info);
144         prio = TC_H_MAJ(t->tcm_info);
145         nprio = prio;
146         parent = t->tcm_parent;
147         cl = 0;
148
149         if (prio == 0) {
150                 /* If no priority is given, user wants we allocated it. */
151                 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
152                         return -ENOENT;
153                 prio = TC_H_MAKE(0x80000000U, 0U);
154         }
155
156         /* Find head of filter chain. */
157
158         /* Find link */
159         dev = __dev_get_by_index(net, t->tcm_ifindex);
160         if (dev == NULL)
161                 return -ENODEV;
162
163         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
164         if (err < 0)
165                 return err;
166
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;
175         }
176
177         /* Is it classful? */
178         if ((cops = q->ops->cl_ops) == NULL)
179                 return -EINVAL;
180
181         if (cops->tcf_chain == NULL)
182                 return -EOPNOTSUPP;
183
184         /* Do we search for filter, attached to class? */
185         if (TC_H_MIN(parent)) {
186                 cl = cops->get(q, parent);
187                 if (cl == 0)
188                         return -ENOENT;
189         }
190
191         /* And the last stroke */
192         chain = cops->tcf_chain(q, cl);
193         err = -EINVAL;
194         if (chain == NULL)
195                 goto errout;
196
197         /* Check the chain for existence of proto-tcf with this priority */
198         for (back = chain; (tp=*back) != NULL; back = &tp->next) {
199                 if (tp->prio >= prio) {
200                         if (tp->prio == prio) {
201                                 if (!nprio || (tp->protocol != protocol && protocol))
202                                         goto errout;
203                         } else
204                                 tp = NULL;
205                         break;
206                 }
207         }
208
209         root_lock = qdisc_root_sleeping_lock(q);
210
211         if (tp == NULL) {
212                 /* Proto-tcf does not exist, create new one */
213
214                 if (tca[TCA_KIND] == NULL || !protocol)
215                         goto errout;
216
217                 err = -ENOENT;
218                 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
219                         goto errout;
220
221
222                 /* Create new proto tcf */
223
224                 err = -ENOBUFS;
225                 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
226                 if (tp == NULL)
227                         goto errout;
228                 err = -ENOENT;
229                 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
230                 if (tp_ops == NULL) {
231 #ifdef CONFIG_MODULES
232                         struct nlattr *kind = tca[TCA_KIND];
233                         char name[IFNAMSIZ];
234
235                         if (kind != NULL &&
236                             nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
237                                 rtnl_unlock();
238                                 request_module("cls_%s", name);
239                                 rtnl_lock();
240                                 tp_ops = tcf_proto_lookup_ops(kind);
241                                 /* We dropped the RTNL semaphore in order to
242                                  * perform the module load.  So, even if we
243                                  * succeeded in loading the module we have to
244                                  * replay the request.  We indicate this using
245                                  * -EAGAIN.
246                                  */
247                                 if (tp_ops != NULL) {
248                                         module_put(tp_ops->owner);
249                                         err = -EAGAIN;
250                                 }
251                         }
252 #endif
253                         kfree(tp);
254                         goto errout;
255                 }
256                 tp->ops = tp_ops;
257                 tp->protocol = protocol;
258                 tp->prio = nprio ? : TC_H_MAJ(tcf_auto_prio(*back));
259                 tp->q = q;
260                 tp->classify = tp_ops->classify;
261                 tp->classid = parent;
262
263                 err = tp_ops->init(tp);
264                 if (err != 0) {
265                         module_put(tp_ops->owner);
266                         kfree(tp);
267                         goto errout;
268                 }
269
270                 tp_created = 1;
271
272         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
273                 goto errout;
274
275         fh = tp->ops->get(tp, t->tcm_handle);
276
277         if (fh == 0) {
278                 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
279                         spin_lock_bh(root_lock);
280                         *back = tp->next;
281                         spin_unlock_bh(root_lock);
282
283                         tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
284                         tcf_destroy(tp);
285                         err = 0;
286                         goto errout;
287                 }
288
289                 err = -ENOENT;
290                 if (n->nlmsg_type != RTM_NEWTFILTER ||
291                     !(n->nlmsg_flags & NLM_F_CREATE))
292                         goto errout;
293         } else {
294                 switch (n->nlmsg_type) {
295                 case RTM_NEWTFILTER:
296                         err = -EEXIST;
297                         if (n->nlmsg_flags & NLM_F_EXCL) {
298                                 if (tp_created)
299                                         tcf_destroy(tp);
300                                 goto errout;
301                         }
302                         break;
303                 case RTM_DELTFILTER:
304                         err = tp->ops->delete(tp, fh);
305                         if (err == 0)
306                                 tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
307                         goto errout;
308                 case RTM_GETTFILTER:
309                         err = tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
310                         goto errout;
311                 default:
312                         err = -EINVAL;
313                         goto errout;
314                 }
315         }
316
317         err = tp->ops->change(tp, cl, t->tcm_handle, tca, &fh);
318         if (err == 0) {
319                 if (tp_created) {
320                         spin_lock_bh(root_lock);
321                         tp->next = *back;
322                         *back = tp;
323                         spin_unlock_bh(root_lock);
324                 }
325                 tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
326         } else {
327                 if (tp_created)
328                         tcf_destroy(tp);
329         }
330
331 errout:
332         if (cl)
333                 cops->put(q, cl);
334         if (err == -EAGAIN)
335                 /* Replay the request. */
336                 goto replay;
337         return err;
338 }
339
340 static int tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp,
341                          unsigned long fh, u32 pid, u32 seq, u16 flags, int event)
342 {
343         struct tcmsg *tcm;
344         struct nlmsghdr  *nlh;
345         unsigned char *b = skb_tail_pointer(skb);
346
347         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
348         tcm = NLMSG_DATA(nlh);
349         tcm->tcm_family = AF_UNSPEC;
350         tcm->tcm__pad1 = 0;
351         tcm->tcm__pad2 = 0;
352         tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
353         tcm->tcm_parent = tp->classid;
354         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
355         NLA_PUT_STRING(skb, TCA_KIND, tp->ops->kind);
356         tcm->tcm_handle = fh;
357         if (RTM_DELTFILTER != event) {
358                 tcm->tcm_handle = 0;
359                 if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
360                         goto nla_put_failure;
361         }
362         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
363         return skb->len;
364
365 nlmsg_failure:
366 nla_put_failure:
367         nlmsg_trim(skb, b);
368         return -1;
369 }
370
371 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
372                           struct nlmsghdr *n, struct tcf_proto *tp,
373                           unsigned long fh, int event)
374 {
375         struct sk_buff *skb;
376         u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
377
378         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
379         if (!skb)
380                 return -ENOBUFS;
381
382         if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) {
383                 kfree_skb(skb);
384                 return -EINVAL;
385         }
386
387         return rtnetlink_send(skb, net, pid, RTNLGRP_TC,
388                               n->nlmsg_flags & NLM_F_ECHO);
389 }
390
391 struct tcf_dump_args {
392         struct tcf_walker w;
393         struct sk_buff *skb;
394         struct netlink_callback *cb;
395 };
396
397 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
398                          struct tcf_walker *arg)
399 {
400         struct tcf_dump_args *a = (void *)arg;
401
402         return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid,
403                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
404 }
405
406 /* called with RTNL */
407 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
408 {
409         struct net *net = sock_net(skb->sk);
410         int t;
411         int s_t;
412         struct net_device *dev;
413         struct Qdisc *q;
414         struct tcf_proto *tp, **chain;
415         struct tcmsg *tcm = (struct tcmsg *)NLMSG_DATA(cb->nlh);
416         unsigned long cl = 0;
417         const struct Qdisc_class_ops *cops;
418         struct tcf_dump_args arg;
419
420         if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
421                 return skb->len;
422         if ((dev = __dev_get_by_index(net, tcm->tcm_ifindex)) == NULL)
423                 return skb->len;
424
425         if (!tcm->tcm_parent)
426                 q = dev->qdisc;
427         else
428                 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
429         if (!q)
430                 goto out;
431         if ((cops = q->ops->cl_ops) == NULL)
432                 goto errout;
433         if (cops->tcf_chain == NULL)
434                 goto errout;
435         if (TC_H_MIN(tcm->tcm_parent)) {
436                 cl = cops->get(q, tcm->tcm_parent);
437                 if (cl == 0)
438                         goto errout;
439         }
440         chain = cops->tcf_chain(q, cl);
441         if (chain == NULL)
442                 goto errout;
443
444         s_t = cb->args[0];
445
446         for (tp=*chain, t=0; tp; tp = tp->next, t++) {
447                 if (t < s_t) continue;
448                 if (TC_H_MAJ(tcm->tcm_info) &&
449                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
450                         continue;
451                 if (TC_H_MIN(tcm->tcm_info) &&
452                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
453                         continue;
454                 if (t > s_t)
455                         memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
456                 if (cb->args[1] == 0) {
457                         if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid,
458                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
459                                           RTM_NEWTFILTER) <= 0)
460                                 break;
461
462                         cb->args[1] = 1;
463                 }
464                 if (tp->ops->walk == NULL)
465                         continue;
466                 arg.w.fn = tcf_node_dump;
467                 arg.skb = skb;
468                 arg.cb = cb;
469                 arg.w.stop = 0;
470                 arg.w.skip = cb->args[1]-1;
471                 arg.w.count = 0;
472                 tp->ops->walk(tp, &arg.w);
473                 cb->args[1] = arg.w.count+1;
474                 if (arg.w.stop)
475                         break;
476         }
477
478         cb->args[0] = t;
479
480 errout:
481         if (cl)
482                 cops->put(q, cl);
483 out:
484         return skb->len;
485 }
486
487 void tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
488 {
489 #ifdef CONFIG_NET_CLS_ACT
490         if (exts->action) {
491                 tcf_action_destroy(exts->action, TCA_ACT_UNBIND);
492                 exts->action = NULL;
493         }
494 #endif
495 }
496 EXPORT_SYMBOL(tcf_exts_destroy);
497
498 int tcf_exts_validate(struct tcf_proto *tp, struct nlattr **tb,
499                   struct nlattr *rate_tlv, struct tcf_exts *exts,
500                   const struct tcf_ext_map *map)
501 {
502         memset(exts, 0, sizeof(*exts));
503
504 #ifdef CONFIG_NET_CLS_ACT
505         {
506                 struct tc_action *act;
507
508                 if (map->police && tb[map->police]) {
509                         act = tcf_action_init_1(tb[map->police], rate_tlv,
510                                                 "police", TCA_ACT_NOREPLACE,
511                                                 TCA_ACT_BIND);
512                         if (IS_ERR(act))
513                                 return PTR_ERR(act);
514
515                         act->type = TCA_OLD_COMPAT;
516                         exts->action = act;
517                 } else if (map->action && tb[map->action]) {
518                         act = tcf_action_init(tb[map->action], rate_tlv, NULL,
519                                               TCA_ACT_NOREPLACE, TCA_ACT_BIND);
520                         if (IS_ERR(act))
521                                 return PTR_ERR(act);
522
523                         exts->action = act;
524                 }
525         }
526 #else
527         if ((map->action && tb[map->action]) ||
528             (map->police && tb[map->police]))
529                 return -EOPNOTSUPP;
530 #endif
531
532         return 0;
533 }
534 EXPORT_SYMBOL(tcf_exts_validate);
535
536 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
537                      struct tcf_exts *src)
538 {
539 #ifdef CONFIG_NET_CLS_ACT
540         if (src->action) {
541                 struct tc_action *act;
542                 tcf_tree_lock(tp);
543                 act = dst->action;
544                 dst->action = src->action;
545                 tcf_tree_unlock(tp);
546                 if (act)
547                         tcf_action_destroy(act, TCA_ACT_UNBIND);
548         }
549 #endif
550 }
551 EXPORT_SYMBOL(tcf_exts_change);
552
553 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts,
554                   const struct tcf_ext_map *map)
555 {
556 #ifdef CONFIG_NET_CLS_ACT
557         if (map->action && exts->action) {
558                 /*
559                  * again for backward compatible mode - we want
560                  * to work with both old and new modes of entering
561                  * tc data even if iproute2  was newer - jhs
562                  */
563                 struct nlattr *nest;
564
565                 if (exts->action->type != TCA_OLD_COMPAT) {
566                         nest = nla_nest_start(skb, map->action);
567                         if (nest == NULL)
568                                 goto nla_put_failure;
569                         if (tcf_action_dump(skb, exts->action, 0, 0) < 0)
570                                 goto nla_put_failure;
571                         nla_nest_end(skb, nest);
572                 } else if (map->police) {
573                         nest = nla_nest_start(skb, map->police);
574                         if (nest == NULL)
575                                 goto nla_put_failure;
576                         if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0)
577                                 goto nla_put_failure;
578                         nla_nest_end(skb, nest);
579                 }
580         }
581 #endif
582         return 0;
583 nla_put_failure: __attribute__ ((unused))
584         return -1;
585 }
586 EXPORT_SYMBOL(tcf_exts_dump);
587
588
589 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts,
590                         const struct tcf_ext_map *map)
591 {
592 #ifdef CONFIG_NET_CLS_ACT
593         if (exts->action)
594                 if (tcf_action_copy_stats(skb, exts->action, 1) < 0)
595                         goto nla_put_failure;
596 #endif
597         return 0;
598 nla_put_failure: __attribute__ ((unused))
599         return -1;
600 }
601 EXPORT_SYMBOL(tcf_exts_dump_stats);
602
603 static int __init tc_filter_init(void)
604 {
605         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL);
606         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL);
607         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
608                                                  tc_dump_tfilter);
609
610         return 0;
611 }
612
613 subsys_initcall(tc_filter_init);