GNU Linux-libre 4.9.330-gnu1
[releases.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/err.h>
26 #include <linux/slab.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 static LIST_HEAD(tcf_proto_base);
35
36 /* Protects list of registered TC modules. It is pure SMP lock. */
37 static DEFINE_RWLOCK(cls_mod_lock);
38
39 /* Find classifier type by string name */
40
41 static const struct tcf_proto_ops *tcf_proto_lookup_ops(struct nlattr *kind)
42 {
43         const struct tcf_proto_ops *t, *res = NULL;
44
45         if (kind) {
46                 read_lock(&cls_mod_lock);
47                 list_for_each_entry(t, &tcf_proto_base, head) {
48                         if (nla_strcmp(kind, t->kind) == 0) {
49                                 if (try_module_get(t->owner))
50                                         res = t;
51                                 break;
52                         }
53                 }
54                 read_unlock(&cls_mod_lock);
55         }
56         return res;
57 }
58
59 /* Register(unregister) new classifier type */
60
61 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
62 {
63         struct tcf_proto_ops *t;
64         int rc = -EEXIST;
65
66         write_lock(&cls_mod_lock);
67         list_for_each_entry(t, &tcf_proto_base, head)
68                 if (!strcmp(ops->kind, t->kind))
69                         goto out;
70
71         list_add_tail(&ops->head, &tcf_proto_base);
72         rc = 0;
73 out:
74         write_unlock(&cls_mod_lock);
75         return rc;
76 }
77 EXPORT_SYMBOL(register_tcf_proto_ops);
78
79 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
80 {
81         struct tcf_proto_ops *t;
82         int rc = -ENOENT;
83
84         /* Wait for outstanding call_rcu()s, if any, from a
85          * tcf_proto_ops's destroy() handler.
86          */
87         rcu_barrier();
88
89         write_lock(&cls_mod_lock);
90         list_for_each_entry(t, &tcf_proto_base, head) {
91                 if (t == ops) {
92                         list_del(&t->head);
93                         rc = 0;
94                         break;
95                 }
96         }
97         write_unlock(&cls_mod_lock);
98         return rc;
99 }
100 EXPORT_SYMBOL(unregister_tcf_proto_ops);
101
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, bool unicast);
105
106 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
107                                  struct nlmsghdr *n,
108                                  struct tcf_proto __rcu **chain, int event)
109 {
110         struct tcf_proto __rcu **it_chain;
111         struct tcf_proto *tp;
112
113         for (it_chain = chain; (tp = rtnl_dereference(*it_chain)) != NULL;
114              it_chain = &tp->next)
115                 tfilter_notify(net, oskb, n, tp, 0, event, false);
116 }
117
118 /* Select new prio value from the range, managed by kernel. */
119
120 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
121 {
122         u32 first = TC_H_MAKE(0xC0000000U, 0U);
123
124         if (tp)
125                 first = tp->prio - 1;
126
127         return first;
128 }
129
130 /* Add/change/delete/get a filter node */
131
132 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n)
133 {
134         struct net *net = sock_net(skb->sk);
135         struct nlattr *tca[TCA_MAX + 1];
136         struct tcmsg *t;
137         u32 protocol;
138         u32 prio;
139         u32 nprio;
140         u32 parent;
141         struct net_device *dev;
142         struct Qdisc  *q;
143         struct tcf_proto __rcu **back;
144         struct tcf_proto __rcu **chain;
145         struct tcf_proto *tp;
146         const struct tcf_proto_ops *tp_ops;
147         const struct Qdisc_class_ops *cops;
148         unsigned long cl;
149         unsigned long fh;
150         int err;
151         int tp_created;
152
153         if ((n->nlmsg_type != RTM_GETTFILTER) &&
154             !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
155                 return -EPERM;
156
157 replay:
158         tp_created = 0;
159
160         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
161         if (err < 0)
162                 return err;
163
164         t = nlmsg_data(n);
165         protocol = TC_H_MIN(t->tcm_info);
166         prio = TC_H_MAJ(t->tcm_info);
167         nprio = prio;
168         parent = t->tcm_parent;
169         cl = 0;
170
171         if (prio == 0) {
172                 switch (n->nlmsg_type) {
173                 case RTM_DELTFILTER:
174                         if (protocol || t->tcm_handle || tca[TCA_KIND])
175                                 return -ENOENT;
176                         break;
177                 case RTM_NEWTFILTER:
178                         /* If no priority is provided by the user,
179                          * we allocate one.
180                          */
181                         if (n->nlmsg_flags & NLM_F_CREATE) {
182                                 prio = TC_H_MAKE(0x80000000U, 0U);
183                                 break;
184                         }
185                         /* fall-through */
186                 default:
187                         return -ENOENT;
188                 }
189         }
190
191         /* Find head of filter chain. */
192
193         /* Find link */
194         dev = __dev_get_by_index(net, t->tcm_ifindex);
195         if (dev == NULL)
196                 return -ENODEV;
197
198         /* Find qdisc */
199         if (!parent) {
200                 q = dev->qdisc;
201                 parent = q->handle;
202         } else {
203                 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
204                 if (q == NULL)
205                         return -EINVAL;
206         }
207
208         /* Is it classful? */
209         cops = q->ops->cl_ops;
210         if (!cops)
211                 return -EINVAL;
212
213         if (cops->tcf_chain == NULL)
214                 return -EOPNOTSUPP;
215
216         /* Do we search for filter, attached to class? */
217         if (TC_H_MIN(parent)) {
218                 cl = cops->get(q, parent);
219                 if (cl == 0)
220                         return -ENOENT;
221         }
222
223         /* And the last stroke */
224         chain = cops->tcf_chain(q, cl);
225         err = -EINVAL;
226         if (chain == NULL)
227                 goto errout;
228         if (n->nlmsg_type == RTM_DELTFILTER && prio == 0) {
229                 tfilter_notify_chain(net, skb, n, chain, RTM_DELTFILTER);
230                 tcf_destroy_chain(chain);
231                 err = 0;
232                 goto errout;
233         }
234
235         /* Check the chain for existence of proto-tcf with this priority */
236         for (back = chain;
237              (tp = rtnl_dereference(*back)) != NULL;
238              back = &tp->next) {
239                 if (tp->prio >= prio) {
240                         if (tp->prio == prio) {
241                                 if (!nprio ||
242                                     (tp->protocol != protocol && protocol))
243                                         goto errout;
244                         } else
245                                 tp = NULL;
246                         break;
247                 }
248         }
249
250         if (tp == NULL) {
251                 /* Proto-tcf does not exist, create new one */
252
253                 if (tca[TCA_KIND] == NULL || !protocol)
254                         goto errout;
255
256                 err = -ENOENT;
257                 if (n->nlmsg_type != RTM_NEWTFILTER ||
258                     !(n->nlmsg_flags & NLM_F_CREATE))
259                         goto errout;
260
261
262                 /* Create new proto tcf */
263
264                 err = -ENOBUFS;
265                 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
266                 if (tp == NULL)
267                         goto errout;
268                 err = -ENOENT;
269                 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
270                 if (tp_ops == NULL) {
271                         struct nlattr *kind = tca[TCA_KIND];
272                         char name[IFNAMSIZ];
273
274                         if (cl)
275                                 cops->put(q, cl);
276                         cl = 0;
277 #ifdef CONFIG_MODULES
278                         if (kind != NULL &&
279                             nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
280                                 rtnl_unlock();
281                                 request_module("cls_%s", name);
282                                 rtnl_lock();
283                                 tp_ops = tcf_proto_lookup_ops(kind);
284                                 /* We dropped the RTNL semaphore in order to
285                                  * perform the module load.  So, even if we
286                                  * succeeded in loading the module we have to
287                                  * replay the request.  We indicate this using
288                                  * -EAGAIN.
289                                  */
290                                 if (tp_ops != NULL) {
291                                         module_put(tp_ops->owner);
292                                         err = -EAGAIN;
293                                 }
294                         }
295 #endif
296                         kfree(tp);
297                         goto errout;
298                 }
299                 tp->ops = tp_ops;
300                 tp->protocol = protocol;
301                 tp->prio = nprio ? :
302                                TC_H_MAJ(tcf_auto_prio(rtnl_dereference(*back)));
303                 tp->q = q;
304                 tp->classify = tp_ops->classify;
305                 tp->classid = parent;
306
307                 err = tp_ops->init(tp);
308                 if (err != 0) {
309                         module_put(tp_ops->owner);
310                         kfree(tp);
311                         goto errout;
312                 }
313
314                 tp_created = 1;
315
316         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
317                 goto errout;
318
319         fh = tp->ops->get(tp, t->tcm_handle);
320
321         if (fh == 0) {
322                 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
323                         struct tcf_proto *next = rtnl_dereference(tp->next);
324
325                         RCU_INIT_POINTER(*back, next);
326
327                         tfilter_notify(net, skb, n, tp, fh,
328                                        RTM_DELTFILTER, false);
329                         tcf_destroy(tp, true);
330                         err = 0;
331                         goto errout;
332                 }
333
334                 err = -ENOENT;
335                 if (n->nlmsg_type != RTM_NEWTFILTER ||
336                     !(n->nlmsg_flags & NLM_F_CREATE))
337                         goto errout;
338         } else {
339                 switch (n->nlmsg_type) {
340                 case RTM_NEWTFILTER:
341                         err = -EEXIST;
342                         if (n->nlmsg_flags & NLM_F_EXCL) {
343                                 if (tp_created)
344                                         tcf_destroy(tp, true);
345                                 goto errout;
346                         }
347                         break;
348                 case RTM_DELTFILTER:
349                         err = tp->ops->delete(tp, fh);
350                         if (err == 0) {
351                                 struct tcf_proto *next = rtnl_dereference(tp->next);
352
353                                 tfilter_notify(net, skb, n, tp,
354                                                t->tcm_handle,
355                                                RTM_DELTFILTER, false);
356                                 if (tcf_destroy(tp, false))
357                                         RCU_INIT_POINTER(*back, next);
358                         }
359                         goto errout;
360                 case RTM_GETTFILTER:
361                         err = tfilter_notify(net, skb, n, tp, fh,
362                                              RTM_NEWTFILTER, true);
363                         goto errout;
364                 default:
365                         err = -EINVAL;
366                         goto errout;
367                 }
368         }
369
370         err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
371                               n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE);
372         if (err == 0) {
373                 if (tp_created) {
374                         RCU_INIT_POINTER(tp->next, rtnl_dereference(*back));
375                         rcu_assign_pointer(*back, tp);
376                 }
377                 tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER, false);
378         } else {
379                 if (tp_created)
380                         tcf_destroy(tp, true);
381         }
382
383 errout:
384         if (cl)
385                 cops->put(q, cl);
386         if (err == -EAGAIN)
387                 /* Replay the request. */
388                 goto replay;
389         return err;
390 }
391
392 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
393                          struct tcf_proto *tp, unsigned long fh, u32 portid,
394                          u32 seq, u16 flags, int event)
395 {
396         struct tcmsg *tcm;
397         struct nlmsghdr  *nlh;
398         unsigned char *b = skb_tail_pointer(skb);
399
400         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
401         if (!nlh)
402                 goto out_nlmsg_trim;
403         tcm = nlmsg_data(nlh);
404         tcm->tcm_family = AF_UNSPEC;
405         tcm->tcm__pad1 = 0;
406         tcm->tcm__pad2 = 0;
407         tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
408         tcm->tcm_parent = tp->classid;
409         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
410         if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
411                 goto nla_put_failure;
412         tcm->tcm_handle = fh;
413         if (RTM_DELTFILTER != event) {
414                 tcm->tcm_handle = 0;
415                 if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
416                         goto nla_put_failure;
417         }
418         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
419         return skb->len;
420
421 out_nlmsg_trim:
422 nla_put_failure:
423         nlmsg_trim(skb, b);
424         return -1;
425 }
426
427 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
428                           struct nlmsghdr *n, struct tcf_proto *tp,
429                           unsigned long fh, int event, bool unicast)
430 {
431         struct sk_buff *skb;
432         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
433
434         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
435         if (!skb)
436                 return -ENOBUFS;
437
438         if (tcf_fill_node(net, skb, tp, fh, portid, n->nlmsg_seq,
439                           n->nlmsg_flags, event) <= 0) {
440                 kfree_skb(skb);
441                 return -EINVAL;
442         }
443
444         if (unicast)
445                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
446
447         return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
448                               n->nlmsg_flags & NLM_F_ECHO);
449 }
450
451 struct tcf_dump_args {
452         struct tcf_walker w;
453         struct sk_buff *skb;
454         struct netlink_callback *cb;
455 };
456
457 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
458                          struct tcf_walker *arg)
459 {
460         struct tcf_dump_args *a = (void *)arg;
461         struct net *net = sock_net(a->skb->sk);
462
463         return tcf_fill_node(net, a->skb, tp, n, NETLINK_CB(a->cb->skb).portid,
464                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
465                              RTM_NEWTFILTER);
466 }
467
468 /* called with RTNL */
469 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
470 {
471         struct net *net = sock_net(skb->sk);
472         int t;
473         int s_t;
474         struct net_device *dev;
475         struct Qdisc *q;
476         struct tcf_proto *tp, __rcu **chain;
477         struct tcmsg *tcm = nlmsg_data(cb->nlh);
478         unsigned long cl = 0;
479         const struct Qdisc_class_ops *cops;
480         struct tcf_dump_args arg;
481
482         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
483                 return skb->len;
484         dev = __dev_get_by_index(net, tcm->tcm_ifindex);
485         if (!dev)
486                 return skb->len;
487
488         if (!tcm->tcm_parent)
489                 q = dev->qdisc;
490         else
491                 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
492         if (!q)
493                 goto out;
494         cops = q->ops->cl_ops;
495         if (!cops)
496                 goto errout;
497         if (cops->tcf_chain == NULL)
498                 goto errout;
499         if (TC_H_MIN(tcm->tcm_parent)) {
500                 cl = cops->get(q, tcm->tcm_parent);
501                 if (cl == 0)
502                         goto errout;
503         }
504         chain = cops->tcf_chain(q, cl);
505         if (chain == NULL)
506                 goto errout;
507
508         s_t = cb->args[0];
509
510         for (tp = rtnl_dereference(*chain), t = 0;
511              tp; tp = rtnl_dereference(tp->next), t++) {
512                 if (t < s_t)
513                         continue;
514                 if (TC_H_MAJ(tcm->tcm_info) &&
515                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
516                         continue;
517                 if (TC_H_MIN(tcm->tcm_info) &&
518                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
519                         continue;
520                 if (t > s_t)
521                         memset(&cb->args[1], 0,
522                                sizeof(cb->args)-sizeof(cb->args[0]));
523                 if (cb->args[1] == 0) {
524                         if (tcf_fill_node(net, skb, tp, 0,
525                                           NETLINK_CB(cb->skb).portid,
526                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
527                                           RTM_NEWTFILTER) <= 0)
528                                 break;
529
530                         cb->args[1] = 1;
531                 }
532                 if (tp->ops->walk == NULL)
533                         continue;
534                 arg.w.fn = tcf_node_dump;
535                 arg.skb = skb;
536                 arg.cb = cb;
537                 arg.w.stop = 0;
538                 arg.w.skip = cb->args[1] - 1;
539                 arg.w.count = 0;
540                 tp->ops->walk(tp, &arg.w);
541                 cb->args[1] = arg.w.count + 1;
542                 if (arg.w.stop)
543                         break;
544         }
545
546         cb->args[0] = t;
547
548 errout:
549         if (cl)
550                 cops->put(q, cl);
551 out:
552         return skb->len;
553 }
554
555 void tcf_exts_destroy(struct tcf_exts *exts)
556 {
557 #ifdef CONFIG_NET_CLS_ACT
558         LIST_HEAD(actions);
559
560         tcf_exts_to_list(exts, &actions);
561         tcf_action_destroy(&actions, TCA_ACT_UNBIND);
562         kfree(exts->actions);
563         exts->nr_actions = 0;
564 #endif
565 }
566 EXPORT_SYMBOL(tcf_exts_destroy);
567
568 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
569                       struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr)
570 {
571 #ifdef CONFIG_NET_CLS_ACT
572         {
573                 struct tc_action *act;
574
575                 if (exts->police && tb[exts->police]) {
576                         act = tcf_action_init_1(net, tb[exts->police], rate_tlv,
577                                                 "police", ovr, TCA_ACT_BIND);
578                         if (IS_ERR(act))
579                                 return PTR_ERR(act);
580
581                         act->type = exts->type = TCA_OLD_COMPAT;
582                         exts->actions[0] = act;
583                         exts->nr_actions = 1;
584                 } else if (exts->action && tb[exts->action]) {
585                         LIST_HEAD(actions);
586                         int err, i = 0;
587
588                         err = tcf_action_init(net, tb[exts->action], rate_tlv,
589                                               NULL, ovr, TCA_ACT_BIND,
590                                               &actions);
591                         if (err)
592                                 return err;
593                         list_for_each_entry(act, &actions, list)
594                                 exts->actions[i++] = act;
595                         exts->nr_actions = i;
596                 }
597         }
598 #else
599         if ((exts->action && tb[exts->action]) ||
600             (exts->police && tb[exts->police]))
601                 return -EOPNOTSUPP;
602 #endif
603
604         return 0;
605 }
606 EXPORT_SYMBOL(tcf_exts_validate);
607
608 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
609                      struct tcf_exts *src)
610 {
611 #ifdef CONFIG_NET_CLS_ACT
612         struct tcf_exts old = *dst;
613
614         tcf_tree_lock(tp);
615         dst->nr_actions = src->nr_actions;
616         dst->actions = src->actions;
617         dst->type = src->type;
618         tcf_tree_unlock(tp);
619
620         tcf_exts_destroy(&old);
621 #endif
622 }
623 EXPORT_SYMBOL(tcf_exts_change);
624
625 #ifdef CONFIG_NET_CLS_ACT
626 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
627 {
628         if (exts->nr_actions == 0)
629                 return NULL;
630         else
631                 return exts->actions[0];
632 }
633 #endif
634
635 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
636 {
637 #ifdef CONFIG_NET_CLS_ACT
638         struct nlattr *nest;
639
640         if (exts->action && exts->nr_actions) {
641                 /*
642                  * again for backward compatible mode - we want
643                  * to work with both old and new modes of entering
644                  * tc data even if iproute2  was newer - jhs
645                  */
646                 if (exts->type != TCA_OLD_COMPAT) {
647                         LIST_HEAD(actions);
648
649                         nest = nla_nest_start(skb, exts->action);
650                         if (nest == NULL)
651                                 goto nla_put_failure;
652
653                         tcf_exts_to_list(exts, &actions);
654                         if (tcf_action_dump(skb, &actions, 0, 0) < 0)
655                                 goto nla_put_failure;
656                         nla_nest_end(skb, nest);
657                 } else if (exts->police) {
658                         struct tc_action *act = tcf_exts_first_act(exts);
659                         nest = nla_nest_start(skb, exts->police);
660                         if (nest == NULL || !act)
661                                 goto nla_put_failure;
662                         if (tcf_action_dump_old(skb, act, 0, 0) < 0)
663                                 goto nla_put_failure;
664                         nla_nest_end(skb, nest);
665                 }
666         }
667         return 0;
668
669 nla_put_failure:
670         nla_nest_cancel(skb, nest);
671         return -1;
672 #else
673         return 0;
674 #endif
675 }
676 EXPORT_SYMBOL(tcf_exts_dump);
677
678
679 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
680 {
681 #ifdef CONFIG_NET_CLS_ACT
682         struct tc_action *a = tcf_exts_first_act(exts);
683         if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
684                 return -1;
685 #endif
686         return 0;
687 }
688 EXPORT_SYMBOL(tcf_exts_dump_stats);
689
690 static int __init tc_filter_init(void)
691 {
692         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL);
693         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL);
694         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
695                       tc_dump_tfilter, NULL);
696
697         return 0;
698 }
699
700 subsys_initcall(tc_filter_init);