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