net_sched: make cls_ops->tcf_chain() optional
[safe/jmp/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         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         int tp_created = 0;
139
140         if (net != &init_net)
141                 return -EINVAL;
142
143 replay:
144         t = NLMSG_DATA(n);
145         protocol = TC_H_MIN(t->tcm_info);
146         prio = TC_H_MAJ(t->tcm_info);
147         nprio = prio;
148         parent = t->tcm_parent;
149         cl = 0;
150
151         if (prio == 0) {
152                 /* If no priority is given, user wants we allocated it. */
153                 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
154                         return -ENOENT;
155                 prio = TC_H_MAKE(0x80000000U, 0U);
156         }
157
158         /* Find head of filter chain. */
159
160         /* Find link */
161         dev = __dev_get_by_index(&init_net, t->tcm_ifindex);
162         if (dev == NULL)
163                 return -ENODEV;
164
165         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
166         if (err < 0)
167                 return err;
168
169         /* Find qdisc */
170         if (!parent) {
171                 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, 0);
172                 q = dev_queue->qdisc_sleeping;
173                 parent = q->handle;
174         } else {
175                 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
176                 if (q == NULL)
177                         return -EINVAL;
178         }
179
180         /* Is it classful? */
181         if ((cops = q->ops->cl_ops) == NULL)
182                 return -EINVAL;
183
184         if (cops->tcf_chain == NULL)
185                 return -EOPNOTSUPP;
186
187         /* Do we search for filter, attached to class? */
188         if (TC_H_MIN(parent)) {
189                 cl = cops->get(q, parent);
190                 if (cl == 0)
191                         return -ENOENT;
192         }
193
194         /* And the last stroke */
195         chain = cops->tcf_chain(q, cl);
196         err = -EINVAL;
197         if (chain == NULL)
198                 goto errout;
199
200         /* Check the chain for existence of proto-tcf with this priority */
201         for (back = chain; (tp=*back) != NULL; back = &tp->next) {
202                 if (tp->prio >= prio) {
203                         if (tp->prio == prio) {
204                                 if (!nprio || (tp->protocol != protocol && protocol))
205                                         goto errout;
206                         } else
207                                 tp = NULL;
208                         break;
209                 }
210         }
211
212         root_lock = qdisc_root_sleeping_lock(q);
213
214         if (tp == NULL) {
215                 /* Proto-tcf does not exist, create new one */
216
217                 if (tca[TCA_KIND] == NULL || !protocol)
218                         goto errout;
219
220                 err = -ENOENT;
221                 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
222                         goto errout;
223
224
225                 /* Create new proto tcf */
226
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];
237
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.
249                                  */
250                                 if (tp_ops != NULL) {
251                                         module_put(tp_ops->owner);
252                                         err = -EAGAIN;
253                                 }
254                         }
255 #endif
256                         kfree(tp);
257                         goto errout;
258                 }
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;
265
266                 err = tp_ops->init(tp);
267                 if (err != 0) {
268                         module_put(tp_ops->owner);
269                         kfree(tp);
270                         goto errout;
271                 }
272
273                 tp_created = 1;
274
275         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
276                 goto errout;
277
278         fh = tp->ops->get(tp, t->tcm_handle);
279
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);
285
286                         tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
287                         tcf_destroy(tp);
288                         err = 0;
289                         goto errout;
290                 }
291
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;
304                         }
305                         break;
306                 case RTM_DELTFILTER:
307                         err = tp->ops->delete(tp, fh);
308                         if (err == 0)
309                                 tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
310                         goto errout;
311                 case RTM_GETTFILTER:
312                         err = tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
313                         goto errout;
314                 default:
315                         err = -EINVAL;
316                         goto errout;
317                 }
318         }
319
320         err = tp->ops->change(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);
327                 }
328                 tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
329         } else {
330                 if (tp_created)
331                         tcf_destroy(tp);
332         }
333
334 errout:
335         if (cl)
336                 cops->put(q, cl);
337         if (err == -EAGAIN)
338                 /* Replay the request. */
339                 goto replay;
340         return err;
341 }
342
343 static int tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp,
344                          unsigned long fh, u32 pid, u32 seq, u16 flags, int event)
345 {
346         struct tcmsg *tcm;
347         struct nlmsghdr  *nlh;
348         unsigned char *b = skb_tail_pointer(skb);
349
350         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
351         tcm = NLMSG_DATA(nlh);
352         tcm->tcm_family = AF_UNSPEC;
353         tcm->tcm__pad1 = 0;
354         tcm->tcm__pad1 = 0;
355         tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
356         tcm->tcm_parent = tp->classid;
357         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
358         NLA_PUT_STRING(skb, TCA_KIND, tp->ops->kind);
359         tcm->tcm_handle = fh;
360         if (RTM_DELTFILTER != event) {
361                 tcm->tcm_handle = 0;
362                 if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
363                         goto nla_put_failure;
364         }
365         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
366         return skb->len;
367
368 nlmsg_failure:
369 nla_put_failure:
370         nlmsg_trim(skb, b);
371         return -1;
372 }
373
374 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
375                           struct tcf_proto *tp, unsigned long fh, int event)
376 {
377         struct sk_buff *skb;
378         u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
379
380         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
381         if (!skb)
382                 return -ENOBUFS;
383
384         if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) {
385                 kfree_skb(skb);
386                 return -EINVAL;
387         }
388
389         return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC,
390                               n->nlmsg_flags & NLM_F_ECHO);
391 }
392
393 struct tcf_dump_args {
394         struct tcf_walker w;
395         struct sk_buff *skb;
396         struct netlink_callback *cb;
397 };
398
399 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
400                          struct tcf_walker *arg)
401 {
402         struct tcf_dump_args *a = (void *)arg;
403
404         return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid,
405                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
406 }
407
408 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
409 {
410         struct net *net = sock_net(skb->sk);
411         struct netdev_queue *dev_queue;
412         int t;
413         int s_t;
414         struct net_device *dev;
415         struct Qdisc *q;
416         struct tcf_proto *tp, **chain;
417         struct tcmsg *tcm = (struct tcmsg *)NLMSG_DATA(cb->nlh);
418         unsigned long cl = 0;
419         const struct Qdisc_class_ops *cops;
420         struct tcf_dump_args arg;
421
422         if (net != &init_net)
423                 return 0;
424
425         if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
426                 return skb->len;
427         if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
428                 return skb->len;
429
430         dev_queue = netdev_get_tx_queue(dev, 0);
431         if (!tcm->tcm_parent)
432                 q = dev_queue->qdisc_sleeping;
433         else
434                 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
435         if (!q)
436                 goto out;
437         if ((cops = q->ops->cl_ops) == NULL)
438                 goto errout;
439         if (cops->tcf_chain == NULL)
440                 goto errout;
441         if (TC_H_MIN(tcm->tcm_parent)) {
442                 cl = cops->get(q, tcm->tcm_parent);
443                 if (cl == 0)
444                         goto errout;
445         }
446         chain = cops->tcf_chain(q, cl);
447         if (chain == NULL)
448                 goto errout;
449
450         s_t = cb->args[0];
451
452         for (tp=*chain, t=0; tp; tp = tp->next, t++) {
453                 if (t < s_t) continue;
454                 if (TC_H_MAJ(tcm->tcm_info) &&
455                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
456                         continue;
457                 if (TC_H_MIN(tcm->tcm_info) &&
458                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
459                         continue;
460                 if (t > s_t)
461                         memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
462                 if (cb->args[1] == 0) {
463                         if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid,
464                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
465                                           RTM_NEWTFILTER) <= 0)
466                                 break;
467
468                         cb->args[1] = 1;
469                 }
470                 if (tp->ops->walk == NULL)
471                         continue;
472                 arg.w.fn = tcf_node_dump;
473                 arg.skb = skb;
474                 arg.cb = cb;
475                 arg.w.stop = 0;
476                 arg.w.skip = cb->args[1]-1;
477                 arg.w.count = 0;
478                 tp->ops->walk(tp, &arg.w);
479                 cb->args[1] = arg.w.count+1;
480                 if (arg.w.stop)
481                         break;
482         }
483
484         cb->args[0] = t;
485
486 errout:
487         if (cl)
488                 cops->put(q, cl);
489 out:
490         dev_put(dev);
491         return skb->len;
492 }
493
494 void tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
495 {
496 #ifdef CONFIG_NET_CLS_ACT
497         if (exts->action) {
498                 tcf_action_destroy(exts->action, TCA_ACT_UNBIND);
499                 exts->action = NULL;
500         }
501 #endif
502 }
503 EXPORT_SYMBOL(tcf_exts_destroy);
504
505 int tcf_exts_validate(struct tcf_proto *tp, struct nlattr **tb,
506                   struct nlattr *rate_tlv, struct tcf_exts *exts,
507                   const struct tcf_ext_map *map)
508 {
509         memset(exts, 0, sizeof(*exts));
510
511 #ifdef CONFIG_NET_CLS_ACT
512         {
513                 struct tc_action *act;
514
515                 if (map->police && tb[map->police]) {
516                         act = tcf_action_init_1(tb[map->police], rate_tlv,
517                                                 "police", TCA_ACT_NOREPLACE,
518                                                 TCA_ACT_BIND);
519                         if (IS_ERR(act))
520                                 return PTR_ERR(act);
521
522                         act->type = TCA_OLD_COMPAT;
523                         exts->action = act;
524                 } else if (map->action && tb[map->action]) {
525                         act = tcf_action_init(tb[map->action], rate_tlv, NULL,
526                                               TCA_ACT_NOREPLACE, TCA_ACT_BIND);
527                         if (IS_ERR(act))
528                                 return PTR_ERR(act);
529
530                         exts->action = act;
531                 }
532         }
533 #else
534         if ((map->action && tb[map->action]) ||
535             (map->police && tb[map->police]))
536                 return -EOPNOTSUPP;
537 #endif
538
539         return 0;
540 }
541 EXPORT_SYMBOL(tcf_exts_validate);
542
543 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
544                      struct tcf_exts *src)
545 {
546 #ifdef CONFIG_NET_CLS_ACT
547         if (src->action) {
548                 struct tc_action *act;
549                 tcf_tree_lock(tp);
550                 act = dst->action;
551                 dst->action = src->action;
552                 tcf_tree_unlock(tp);
553                 if (act)
554                         tcf_action_destroy(act, TCA_ACT_UNBIND);
555         }
556 #endif
557 }
558 EXPORT_SYMBOL(tcf_exts_change);
559
560 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts,
561                   const struct tcf_ext_map *map)
562 {
563 #ifdef CONFIG_NET_CLS_ACT
564         if (map->action && exts->action) {
565                 /*
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
569                  */
570                 struct nlattr *nest;
571
572                 if (exts->action->type != TCA_OLD_COMPAT) {
573                         nest = nla_nest_start(skb, map->action);
574                         if (nest == NULL)
575                                 goto nla_put_failure;
576                         if (tcf_action_dump(skb, exts->action, 0, 0) < 0)
577                                 goto nla_put_failure;
578                         nla_nest_end(skb, nest);
579                 } else if (map->police) {
580                         nest = nla_nest_start(skb, map->police);
581                         if (nest == NULL)
582                                 goto nla_put_failure;
583                         if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0)
584                                 goto nla_put_failure;
585                         nla_nest_end(skb, nest);
586                 }
587         }
588 #endif
589         return 0;
590 nla_put_failure: __attribute__ ((unused))
591         return -1;
592 }
593 EXPORT_SYMBOL(tcf_exts_dump);
594
595
596 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts,
597                         const struct tcf_ext_map *map)
598 {
599 #ifdef CONFIG_NET_CLS_ACT
600         if (exts->action)
601                 if (tcf_action_copy_stats(skb, exts->action, 1) < 0)
602                         goto nla_put_failure;
603 #endif
604         return 0;
605 nla_put_failure: __attribute__ ((unused))
606         return -1;
607 }
608 EXPORT_SYMBOL(tcf_exts_dump_stats);
609
610 static int __init tc_filter_init(void)
611 {
612         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL);
613         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL);
614         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
615                                                  tc_dump_tfilter);
616
617         return 0;
618 }
619
620 subsys_initcall(tc_filter_init);