Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6] / 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 sk_buff *oskb, struct nlmsghdr *n,
102                           struct tcf_proto *tp, unsigned long fh, int event);
103
104
105 /* Select new prio value from the range, managed by kernel. */
106
107 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
108 {
109         u32 first = TC_H_MAKE(0xC0000000U, 0U);
110
111         if (tp)
112                 first = tp->prio-1;
113
114         return first;
115 }
116
117 /* Add/change/delete/get a filter node */
118
119 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
120 {
121         struct net *net = sock_net(skb->sk);
122         struct nlattr *tca[TCA_MAX + 1];
123         struct tcmsg *t;
124         u32 protocol;
125         u32 prio;
126         u32 nprio;
127         u32 parent;
128         struct net_device *dev;
129         struct Qdisc  *q;
130         struct tcf_proto **back, **chain;
131         struct tcf_proto *tp;
132         struct tcf_proto_ops *tp_ops;
133         const struct Qdisc_class_ops *cops;
134         unsigned long cl;
135         unsigned long fh;
136         int err;
137
138         if (net != &init_net)
139                 return -EINVAL;
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(&init_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                 struct netdev_queue *dev_queue = &dev->tx_queue;
170                 q = dev_queue->qdisc_sleeping;
171                 parent = q->handle;
172         } else {
173                 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
174                 if (q == NULL)
175                         return -EINVAL;
176         }
177
178         /* Is it classful? */
179         if ((cops = q->ops->cl_ops) == NULL)
180                 return -EINVAL;
181
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;
187         }
188
189         /* And the last stroke */
190         chain = cops->tcf_chain(q, cl);
191         err = -EINVAL;
192         if (chain == NULL)
193                 goto errout;
194
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;
204                 }
205         }
206
207         if (tp == NULL) {
208                 /* Proto-tcf does not exist, create new one */
209
210                 if (tca[TCA_KIND] == NULL || !protocol)
211                         goto errout;
212
213                 err = -ENOENT;
214                 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
215                         goto errout;
216
217
218                 /* Create new proto tcf */
219
220                 err = -ENOBUFS;
221                 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
222                 if (tp == NULL)
223                         goto errout;
224                 err = -ENOENT;
225                 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
226                 if (tp_ops == NULL) {
227 #ifdef CONFIG_KMOD
228                         struct nlattr *kind = tca[TCA_KIND];
229                         char name[IFNAMSIZ];
230
231                         if (kind != NULL &&
232                             nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
233                                 rtnl_unlock();
234                                 request_module("cls_%s", name);
235                                 rtnl_lock();
236                                 tp_ops = tcf_proto_lookup_ops(kind);
237                                 /* We dropped the RTNL semaphore in order to
238                                  * perform the module load.  So, even if we
239                                  * succeeded in loading the module we have to
240                                  * replay the request.  We indicate this using
241                                  * -EAGAIN.
242                                  */
243                                 if (tp_ops != NULL) {
244                                         module_put(tp_ops->owner);
245                                         err = -EAGAIN;
246                                 }
247                         }
248 #endif
249                         kfree(tp);
250                         goto errout;
251                 }
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
259                 err = tp_ops->init(tp);
260                 if (err != 0) {
261                         module_put(tp_ops->owner);
262                         kfree(tp);
263                         goto errout;
264                 }
265
266                 qdisc_lock_tree(dev);
267                 tp->next = *back;
268                 *back = tp;
269                 qdisc_unlock_tree(dev);
270
271         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
272                 goto errout;
273
274         fh = tp->ops->get(tp, t->tcm_handle);
275
276         if (fh == 0) {
277                 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
278                         qdisc_lock_tree(dev);
279                         *back = tp->next;
280                         qdisc_unlock_tree(dev);
281
282                         tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
283                         tcf_destroy(tp);
284                         err = 0;
285                         goto errout;
286                 }
287
288                 err = -ENOENT;
289                 if (n->nlmsg_type != RTM_NEWTFILTER ||
290                     !(n->nlmsg_flags & NLM_F_CREATE))
291                         goto errout;
292         } else {
293                 switch (n->nlmsg_type) {
294                 case RTM_NEWTFILTER:
295                         err = -EEXIST;
296                         if (n->nlmsg_flags & NLM_F_EXCL)
297                                 goto errout;
298                         break;
299                 case RTM_DELTFILTER:
300                         err = tp->ops->delete(tp, fh);
301                         if (err == 0)
302                                 tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
303                         goto errout;
304                 case RTM_GETTFILTER:
305                         err = tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
306                         goto errout;
307                 default:
308                         err = -EINVAL;
309                         goto errout;
310                 }
311         }
312
313         err = tp->ops->change(tp, cl, t->tcm_handle, tca, &fh);
314         if (err == 0)
315                 tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
316
317 errout:
318         if (cl)
319                 cops->put(q, cl);
320         if (err == -EAGAIN)
321                 /* Replay the request. */
322                 goto replay;
323         return err;
324 }
325
326 static int tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp,
327                          unsigned long fh, u32 pid, u32 seq, u16 flags, int event)
328 {
329         struct tcmsg *tcm;
330         struct nlmsghdr  *nlh;
331         unsigned char *b = skb_tail_pointer(skb);
332
333         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
334         tcm = NLMSG_DATA(nlh);
335         tcm->tcm_family = AF_UNSPEC;
336         tcm->tcm__pad1 = 0;
337         tcm->tcm__pad1 = 0;
338         tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
339         tcm->tcm_parent = tp->classid;
340         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
341         NLA_PUT_STRING(skb, TCA_KIND, tp->ops->kind);
342         tcm->tcm_handle = fh;
343         if (RTM_DELTFILTER != event) {
344                 tcm->tcm_handle = 0;
345                 if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
346                         goto nla_put_failure;
347         }
348         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
349         return skb->len;
350
351 nlmsg_failure:
352 nla_put_failure:
353         nlmsg_trim(skb, b);
354         return -1;
355 }
356
357 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
358                           struct tcf_proto *tp, unsigned long fh, int event)
359 {
360         struct sk_buff *skb;
361         u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
362
363         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
364         if (!skb)
365                 return -ENOBUFS;
366
367         if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) {
368                 kfree_skb(skb);
369                 return -EINVAL;
370         }
371
372         return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC,
373                               n->nlmsg_flags & NLM_F_ECHO);
374 }
375
376 struct tcf_dump_args {
377         struct tcf_walker w;
378         struct sk_buff *skb;
379         struct netlink_callback *cb;
380 };
381
382 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
383                          struct tcf_walker *arg)
384 {
385         struct tcf_dump_args *a = (void *)arg;
386
387         return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid,
388                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
389 }
390
391 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
392 {
393         struct net *net = sock_net(skb->sk);
394         struct netdev_queue *dev_queue;
395         int t;
396         int s_t;
397         struct net_device *dev;
398         struct Qdisc *q;
399         struct tcf_proto *tp, **chain;
400         struct tcmsg *tcm = (struct tcmsg *)NLMSG_DATA(cb->nlh);
401         unsigned long cl = 0;
402         const struct Qdisc_class_ops *cops;
403         struct tcf_dump_args arg;
404
405         if (net != &init_net)
406                 return 0;
407
408         if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
409                 return skb->len;
410         if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
411                 return skb->len;
412
413         dev_queue = &dev->tx_queue;
414         if (!tcm->tcm_parent)
415                 q = dev_queue->qdisc_sleeping;
416         else
417                 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
418         if (!q)
419                 goto out;
420         if ((cops = q->ops->cl_ops) == NULL)
421                 goto errout;
422         if (TC_H_MIN(tcm->tcm_parent)) {
423                 cl = cops->get(q, tcm->tcm_parent);
424                 if (cl == 0)
425                         goto errout;
426         }
427         chain = cops->tcf_chain(q, cl);
428         if (chain == NULL)
429                 goto errout;
430
431         s_t = cb->args[0];
432
433         for (tp=*chain, t=0; tp; tp = tp->next, t++) {
434                 if (t < s_t) continue;
435                 if (TC_H_MAJ(tcm->tcm_info) &&
436                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
437                         continue;
438                 if (TC_H_MIN(tcm->tcm_info) &&
439                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
440                         continue;
441                 if (t > s_t)
442                         memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
443                 if (cb->args[1] == 0) {
444                         if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid,
445                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
446                                           RTM_NEWTFILTER) <= 0)
447                                 break;
448
449                         cb->args[1] = 1;
450                 }
451                 if (tp->ops->walk == NULL)
452                         continue;
453                 arg.w.fn = tcf_node_dump;
454                 arg.skb = skb;
455                 arg.cb = cb;
456                 arg.w.stop = 0;
457                 arg.w.skip = cb->args[1]-1;
458                 arg.w.count = 0;
459                 tp->ops->walk(tp, &arg.w);
460                 cb->args[1] = arg.w.count+1;
461                 if (arg.w.stop)
462                         break;
463         }
464
465         cb->args[0] = t;
466
467 errout:
468         if (cl)
469                 cops->put(q, cl);
470 out:
471         dev_put(dev);
472         return skb->len;
473 }
474
475 void tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
476 {
477 #ifdef CONFIG_NET_CLS_ACT
478         if (exts->action) {
479                 tcf_action_destroy(exts->action, TCA_ACT_UNBIND);
480                 exts->action = NULL;
481         }
482 #endif
483 }
484 EXPORT_SYMBOL(tcf_exts_destroy);
485
486 int tcf_exts_validate(struct tcf_proto *tp, struct nlattr **tb,
487                   struct nlattr *rate_tlv, struct tcf_exts *exts,
488                   const struct tcf_ext_map *map)
489 {
490         memset(exts, 0, sizeof(*exts));
491
492 #ifdef CONFIG_NET_CLS_ACT
493         {
494                 struct tc_action *act;
495
496                 if (map->police && tb[map->police]) {
497                         act = tcf_action_init_1(tb[map->police], rate_tlv,
498                                                 "police", TCA_ACT_NOREPLACE,
499                                                 TCA_ACT_BIND);
500                         if (IS_ERR(act))
501                                 return PTR_ERR(act);
502
503                         act->type = TCA_OLD_COMPAT;
504                         exts->action = act;
505                 } else if (map->action && tb[map->action]) {
506                         act = tcf_action_init(tb[map->action], rate_tlv, NULL,
507                                               TCA_ACT_NOREPLACE, TCA_ACT_BIND);
508                         if (IS_ERR(act))
509                                 return PTR_ERR(act);
510
511                         exts->action = act;
512                 }
513         }
514 #else
515         if ((map->action && tb[map->action]) ||
516             (map->police && tb[map->police]))
517                 return -EOPNOTSUPP;
518 #endif
519
520         return 0;
521 }
522 EXPORT_SYMBOL(tcf_exts_validate);
523
524 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
525                      struct tcf_exts *src)
526 {
527 #ifdef CONFIG_NET_CLS_ACT
528         if (src->action) {
529                 struct tc_action *act;
530                 tcf_tree_lock(tp);
531                 act = xchg(&dst->action, src->action);
532                 tcf_tree_unlock(tp);
533                 if (act)
534                         tcf_action_destroy(act, TCA_ACT_UNBIND);
535         }
536 #endif
537 }
538 EXPORT_SYMBOL(tcf_exts_change);
539
540 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts,
541                   const struct tcf_ext_map *map)
542 {
543 #ifdef CONFIG_NET_CLS_ACT
544         if (map->action && exts->action) {
545                 /*
546                  * again for backward compatible mode - we want
547                  * to work with both old and new modes of entering
548                  * tc data even if iproute2  was newer - jhs
549                  */
550                 struct nlattr *nest;
551
552                 if (exts->action->type != TCA_OLD_COMPAT) {
553                         nest = nla_nest_start(skb, map->action);
554                         if (nest == NULL)
555                                 goto nla_put_failure;
556                         if (tcf_action_dump(skb, exts->action, 0, 0) < 0)
557                                 goto nla_put_failure;
558                         nla_nest_end(skb, nest);
559                 } else if (map->police) {
560                         nest = nla_nest_start(skb, map->police);
561                         if (nest == NULL)
562                                 goto nla_put_failure;
563                         if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0)
564                                 goto nla_put_failure;
565                         nla_nest_end(skb, nest);
566                 }
567         }
568 #endif
569         return 0;
570 nla_put_failure: __attribute__ ((unused))
571         return -1;
572 }
573 EXPORT_SYMBOL(tcf_exts_dump);
574
575
576 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts,
577                         const struct tcf_ext_map *map)
578 {
579 #ifdef CONFIG_NET_CLS_ACT
580         if (exts->action)
581                 if (tcf_action_copy_stats(skb, exts->action, 1) < 0)
582                         goto nla_put_failure;
583 #endif
584         return 0;
585 nla_put_failure: __attribute__ ((unused))
586         return -1;
587 }
588 EXPORT_SYMBOL(tcf_exts_dump_stats);
589
590 static int __init tc_filter_init(void)
591 {
592         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL);
593         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL);
594         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
595                                                  tc_dump_tfilter);
596
597         return 0;
598 }
599
600 subsys_initcall(tc_filter_init);