8eb79e92e94cb82348a7cf76e8399f46f86426a7
[linux-2.6.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 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         spinlock_t *root_lock;
124         struct tcmsg *t;
125         u32 protocol;
126         u32 prio;
127         u32 nprio;
128         u32 parent;
129         struct net_device *dev;
130         struct Qdisc  *q;
131         struct tcf_proto **back, **chain;
132         struct tcf_proto *tp;
133         struct tcf_proto_ops *tp_ops;
134         const struct Qdisc_class_ops *cops;
135         unsigned long cl;
136         unsigned long fh;
137         int err;
138
139         if (net != &init_net)
140                 return -EINVAL;
141
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;
149
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);
155         }
156
157         /* Find head of filter chain. */
158
159         /* Find link */
160         dev = __dev_get_by_index(&init_net, t->tcm_ifindex);
161         if (dev == NULL)
162                 return -ENODEV;
163
164         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
165         if (err < 0)
166                 return err;
167
168         /* Find qdisc */
169         if (!parent) {
170                 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, 0);
171                 q = dev_queue->qdisc_sleeping;
172                 parent = q->handle;
173         } else {
174                 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
175                 if (q == NULL)
176                         return -EINVAL;
177         }
178
179         /* Is it classful? */
180         if ((cops = q->ops->cl_ops) == NULL)
181                 return -EINVAL;
182
183         /* Do we search for filter, attached to class? */
184         if (TC_H_MIN(parent)) {
185                 cl = cops->get(q, parent);
186                 if (cl == 0)
187                         return -ENOENT;
188         }
189
190         /* And the last stroke */
191         chain = cops->tcf_chain(q, cl);
192         err = -EINVAL;
193         if (chain == NULL)
194                 goto errout;
195
196         /* Check the chain for existence of proto-tcf with this priority */
197         for (back = chain; (tp=*back) != NULL; back = &tp->next) {
198                 if (tp->prio >= prio) {
199                         if (tp->prio == prio) {
200                                 if (!nprio || (tp->protocol != protocol && protocol))
201                                         goto errout;
202                         } else
203                                 tp = NULL;
204                         break;
205                 }
206         }
207
208         root_lock = qdisc_root_sleeping_lock(q);
209
210         if (tp == NULL) {
211                 /* Proto-tcf does not exist, create new one */
212
213                 if (tca[TCA_KIND] == NULL || !protocol)
214                         goto errout;
215
216                 err = -ENOENT;
217                 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
218                         goto errout;
219
220
221                 /* Create new proto tcf */
222
223                 err = -ENOBUFS;
224                 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
225                 if (tp == NULL)
226                         goto errout;
227                 err = -ENOENT;
228                 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
229                 if (tp_ops == NULL) {
230 #ifdef CONFIG_KMOD
231                         struct nlattr *kind = tca[TCA_KIND];
232                         char name[IFNAMSIZ];
233
234                         if (kind != NULL &&
235                             nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
236                                 rtnl_unlock();
237                                 request_module("cls_%s", name);
238                                 rtnl_lock();
239                                 tp_ops = tcf_proto_lookup_ops(kind);
240                                 /* We dropped the RTNL semaphore in order to
241                                  * perform the module load.  So, even if we
242                                  * succeeded in loading the module we have to
243                                  * replay the request.  We indicate this using
244                                  * -EAGAIN.
245                                  */
246                                 if (tp_ops != NULL) {
247                                         module_put(tp_ops->owner);
248                                         err = -EAGAIN;
249                                 }
250                         }
251 #endif
252                         kfree(tp);
253                         goto errout;
254                 }
255                 tp->ops = tp_ops;
256                 tp->protocol = protocol;
257                 tp->prio = nprio ? : tcf_auto_prio(*back);
258                 tp->q = q;
259                 tp->classify = tp_ops->classify;
260                 tp->classid = parent;
261
262                 err = tp_ops->init(tp);
263                 if (err != 0) {
264                         module_put(tp_ops->owner);
265                         kfree(tp);
266                         goto errout;
267                 }
268
269                 spin_lock_bh(root_lock);
270                 tp->next = *back;
271                 *back = tp;
272                 spin_unlock_bh(root_lock);
273
274         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
275                 goto errout;
276
277         fh = tp->ops->get(tp, t->tcm_handle);
278
279         if (fh == 0) {
280                 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
281                         spin_lock_bh(root_lock);
282                         *back = tp->next;
283                         spin_unlock_bh(root_lock);
284
285                         tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
286                         tcf_destroy(tp);
287                         err = 0;
288                         goto errout;
289                 }
290
291                 err = -ENOENT;
292                 if (n->nlmsg_type != RTM_NEWTFILTER ||
293                     !(n->nlmsg_flags & NLM_F_CREATE))
294                         goto errout;
295         } else {
296                 switch (n->nlmsg_type) {
297                 case RTM_NEWTFILTER:
298                         err = -EEXIST;
299                         if (n->nlmsg_flags & NLM_F_EXCL)
300                                 goto errout;
301                         break;
302                 case RTM_DELTFILTER:
303                         err = tp->ops->delete(tp, fh);
304                         if (err == 0)
305                                 tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
306                         goto errout;
307                 case RTM_GETTFILTER:
308                         err = tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
309                         goto errout;
310                 default:
311                         err = -EINVAL;
312                         goto errout;
313                 }
314         }
315
316         err = tp->ops->change(tp, cl, t->tcm_handle, tca, &fh);
317         if (err == 0)
318                 tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
319
320 errout:
321         if (cl)
322                 cops->put(q, cl);
323         if (err == -EAGAIN)
324                 /* Replay the request. */
325                 goto replay;
326         return err;
327 }
328
329 static int tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp,
330                          unsigned long fh, u32 pid, u32 seq, u16 flags, int event)
331 {
332         struct tcmsg *tcm;
333         struct nlmsghdr  *nlh;
334         unsigned char *b = skb_tail_pointer(skb);
335
336         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
337         tcm = NLMSG_DATA(nlh);
338         tcm->tcm_family = AF_UNSPEC;
339         tcm->tcm__pad1 = 0;
340         tcm->tcm__pad1 = 0;
341         tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
342         tcm->tcm_parent = tp->classid;
343         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
344         NLA_PUT_STRING(skb, TCA_KIND, tp->ops->kind);
345         tcm->tcm_handle = fh;
346         if (RTM_DELTFILTER != event) {
347                 tcm->tcm_handle = 0;
348                 if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
349                         goto nla_put_failure;
350         }
351         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
352         return skb->len;
353
354 nlmsg_failure:
355 nla_put_failure:
356         nlmsg_trim(skb, b);
357         return -1;
358 }
359
360 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
361                           struct tcf_proto *tp, unsigned long fh, int event)
362 {
363         struct sk_buff *skb;
364         u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
365
366         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
367         if (!skb)
368                 return -ENOBUFS;
369
370         if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) {
371                 kfree_skb(skb);
372                 return -EINVAL;
373         }
374
375         return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC,
376                               n->nlmsg_flags & NLM_F_ECHO);
377 }
378
379 struct tcf_dump_args {
380         struct tcf_walker w;
381         struct sk_buff *skb;
382         struct netlink_callback *cb;
383 };
384
385 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
386                          struct tcf_walker *arg)
387 {
388         struct tcf_dump_args *a = (void *)arg;
389
390         return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid,
391                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
392 }
393
394 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
395 {
396         struct net *net = sock_net(skb->sk);
397         struct netdev_queue *dev_queue;
398         int t;
399         int s_t;
400         struct net_device *dev;
401         struct Qdisc *q;
402         struct tcf_proto *tp, **chain;
403         struct tcmsg *tcm = (struct tcmsg *)NLMSG_DATA(cb->nlh);
404         unsigned long cl = 0;
405         const struct Qdisc_class_ops *cops;
406         struct tcf_dump_args arg;
407
408         if (net != &init_net)
409                 return 0;
410
411         if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
412                 return skb->len;
413         if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
414                 return skb->len;
415
416         dev_queue = netdev_get_tx_queue(dev, 0);
417         if (!tcm->tcm_parent)
418                 q = dev_queue->qdisc_sleeping;
419         else
420                 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
421         if (!q)
422                 goto out;
423         if ((cops = q->ops->cl_ops) == NULL)
424                 goto errout;
425         if (TC_H_MIN(tcm->tcm_parent)) {
426                 cl = cops->get(q, tcm->tcm_parent);
427                 if (cl == 0)
428                         goto errout;
429         }
430         chain = cops->tcf_chain(q, cl);
431         if (chain == NULL)
432                 goto errout;
433
434         s_t = cb->args[0];
435
436         for (tp=*chain, t=0; tp; tp = tp->next, t++) {
437                 if (t < s_t) continue;
438                 if (TC_H_MAJ(tcm->tcm_info) &&
439                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
440                         continue;
441                 if (TC_H_MIN(tcm->tcm_info) &&
442                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
443                         continue;
444                 if (t > s_t)
445                         memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
446                 if (cb->args[1] == 0) {
447                         if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid,
448                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
449                                           RTM_NEWTFILTER) <= 0)
450                                 break;
451
452                         cb->args[1] = 1;
453                 }
454                 if (tp->ops->walk == NULL)
455                         continue;
456                 arg.w.fn = tcf_node_dump;
457                 arg.skb = skb;
458                 arg.cb = cb;
459                 arg.w.stop = 0;
460                 arg.w.skip = cb->args[1]-1;
461                 arg.w.count = 0;
462                 tp->ops->walk(tp, &arg.w);
463                 cb->args[1] = arg.w.count+1;
464                 if (arg.w.stop)
465                         break;
466         }
467
468         cb->args[0] = t;
469
470 errout:
471         if (cl)
472                 cops->put(q, cl);
473 out:
474         dev_put(dev);
475         return skb->len;
476 }
477
478 void tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
479 {
480 #ifdef CONFIG_NET_CLS_ACT
481         if (exts->action) {
482                 tcf_action_destroy(exts->action, TCA_ACT_UNBIND);
483                 exts->action = NULL;
484         }
485 #endif
486 }
487 EXPORT_SYMBOL(tcf_exts_destroy);
488
489 int tcf_exts_validate(struct tcf_proto *tp, struct nlattr **tb,
490                   struct nlattr *rate_tlv, struct tcf_exts *exts,
491                   const struct tcf_ext_map *map)
492 {
493         memset(exts, 0, sizeof(*exts));
494
495 #ifdef CONFIG_NET_CLS_ACT
496         {
497                 struct tc_action *act;
498
499                 if (map->police && tb[map->police]) {
500                         act = tcf_action_init_1(tb[map->police], rate_tlv,
501                                                 "police", TCA_ACT_NOREPLACE,
502                                                 TCA_ACT_BIND);
503                         if (IS_ERR(act))
504                                 return PTR_ERR(act);
505
506                         act->type = TCA_OLD_COMPAT;
507                         exts->action = act;
508                 } else if (map->action && tb[map->action]) {
509                         act = tcf_action_init(tb[map->action], rate_tlv, NULL,
510                                               TCA_ACT_NOREPLACE, TCA_ACT_BIND);
511                         if (IS_ERR(act))
512                                 return PTR_ERR(act);
513
514                         exts->action = act;
515                 }
516         }
517 #else
518         if ((map->action && tb[map->action]) ||
519             (map->police && tb[map->police]))
520                 return -EOPNOTSUPP;
521 #endif
522
523         return 0;
524 }
525 EXPORT_SYMBOL(tcf_exts_validate);
526
527 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
528                      struct tcf_exts *src)
529 {
530 #ifdef CONFIG_NET_CLS_ACT
531         if (src->action) {
532                 struct tc_action *act;
533                 tcf_tree_lock(tp);
534                 act = xchg(&dst->action, src->action);
535                 tcf_tree_unlock(tp);
536                 if (act)
537                         tcf_action_destroy(act, TCA_ACT_UNBIND);
538         }
539 #endif
540 }
541 EXPORT_SYMBOL(tcf_exts_change);
542
543 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts,
544                   const struct tcf_ext_map *map)
545 {
546 #ifdef CONFIG_NET_CLS_ACT
547         if (map->action && exts->action) {
548                 /*
549                  * again for backward compatible mode - we want
550                  * to work with both old and new modes of entering
551                  * tc data even if iproute2  was newer - jhs
552                  */
553                 struct nlattr *nest;
554
555                 if (exts->action->type != TCA_OLD_COMPAT) {
556                         nest = nla_nest_start(skb, map->action);
557                         if (nest == NULL)
558                                 goto nla_put_failure;
559                         if (tcf_action_dump(skb, exts->action, 0, 0) < 0)
560                                 goto nla_put_failure;
561                         nla_nest_end(skb, nest);
562                 } else if (map->police) {
563                         nest = nla_nest_start(skb, map->police);
564                         if (nest == NULL)
565                                 goto nla_put_failure;
566                         if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0)
567                                 goto nla_put_failure;
568                         nla_nest_end(skb, nest);
569                 }
570         }
571 #endif
572         return 0;
573 nla_put_failure: __attribute__ ((unused))
574         return -1;
575 }
576 EXPORT_SYMBOL(tcf_exts_dump);
577
578
579 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts,
580                         const struct tcf_ext_map *map)
581 {
582 #ifdef CONFIG_NET_CLS_ACT
583         if (exts->action)
584                 if (tcf_action_copy_stats(skb, exts->action, 1) < 0)
585                         goto nla_put_failure;
586 #endif
587         return 0;
588 nla_put_failure: __attribute__ ((unused))
589         return -1;
590 }
591 EXPORT_SYMBOL(tcf_exts_dump_stats);
592
593 static int __init tc_filter_init(void)
594 {
595         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL);
596         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL);
597         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
598                                                  tc_dump_tfilter);
599
600         return 0;
601 }
602
603 subsys_initcall(tc_filter_init);