]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - net/sched/cls_api.c
net, sched: fix soft lockup in tc_classify
[mirror_ubuntu-bionic-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, 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 #ifdef CONFIG_MODULES
272 struct nlattr *kind = tca[TCA_KIND];
273 char name[IFNAMSIZ];
274
275 if (kind != NULL &&
276 nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
277 rtnl_unlock();
278 request_module("cls_%s", name);
279 rtnl_lock();
280 tp_ops = tcf_proto_lookup_ops(kind);
281 /* We dropped the RTNL semaphore in order to
282 * perform the module load. So, even if we
283 * succeeded in loading the module we have to
284 * replay the request. We indicate this using
285 * -EAGAIN.
286 */
287 if (tp_ops != NULL) {
288 module_put(tp_ops->owner);
289 err = -EAGAIN;
290 }
291 }
292 #endif
293 kfree(tp);
294 goto errout;
295 }
296 tp->ops = tp_ops;
297 tp->protocol = protocol;
298 tp->prio = nprio ? :
299 TC_H_MAJ(tcf_auto_prio(rtnl_dereference(*back)));
300 tp->q = q;
301 tp->classify = tp_ops->classify;
302 tp->classid = parent;
303
304 err = tp_ops->init(tp);
305 if (err != 0) {
306 module_put(tp_ops->owner);
307 kfree(tp);
308 goto errout;
309 }
310
311 tp_created = 1;
312
313 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
314 goto errout;
315
316 fh = tp->ops->get(tp, t->tcm_handle);
317
318 if (fh == 0) {
319 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
320 struct tcf_proto *next = rtnl_dereference(tp->next);
321
322 RCU_INIT_POINTER(*back, next);
323
324 tfilter_notify(net, skb, n, tp, fh,
325 RTM_DELTFILTER, false);
326 tcf_destroy(tp, true);
327 err = 0;
328 goto errout;
329 }
330
331 err = -ENOENT;
332 if (n->nlmsg_type != RTM_NEWTFILTER ||
333 !(n->nlmsg_flags & NLM_F_CREATE))
334 goto errout;
335 } else {
336 switch (n->nlmsg_type) {
337 case RTM_NEWTFILTER:
338 err = -EEXIST;
339 if (n->nlmsg_flags & NLM_F_EXCL) {
340 if (tp_created)
341 tcf_destroy(tp, true);
342 goto errout;
343 }
344 break;
345 case RTM_DELTFILTER:
346 err = tp->ops->delete(tp, fh);
347 if (err == 0) {
348 struct tcf_proto *next = rtnl_dereference(tp->next);
349
350 tfilter_notify(net, skb, n, tp,
351 t->tcm_handle,
352 RTM_DELTFILTER, false);
353 if (tcf_destroy(tp, false))
354 RCU_INIT_POINTER(*back, next);
355 }
356 goto errout;
357 case RTM_GETTFILTER:
358 err = tfilter_notify(net, skb, n, tp, fh,
359 RTM_NEWTFILTER, true);
360 goto errout;
361 default:
362 err = -EINVAL;
363 goto errout;
364 }
365 }
366
367 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
368 n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE);
369 if (err == 0) {
370 if (tp_created) {
371 RCU_INIT_POINTER(tp->next, rtnl_dereference(*back));
372 rcu_assign_pointer(*back, tp);
373 }
374 tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER, false);
375 } else {
376 if (tp_created)
377 tcf_destroy(tp, true);
378 }
379
380 errout:
381 if (cl)
382 cops->put(q, cl);
383 if (err == -EAGAIN)
384 /* Replay the request. */
385 goto replay;
386 return err;
387 }
388
389 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
390 struct tcf_proto *tp, unsigned long fh, u32 portid,
391 u32 seq, u16 flags, int event)
392 {
393 struct tcmsg *tcm;
394 struct nlmsghdr *nlh;
395 unsigned char *b = skb_tail_pointer(skb);
396
397 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
398 if (!nlh)
399 goto out_nlmsg_trim;
400 tcm = nlmsg_data(nlh);
401 tcm->tcm_family = AF_UNSPEC;
402 tcm->tcm__pad1 = 0;
403 tcm->tcm__pad2 = 0;
404 tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
405 tcm->tcm_parent = tp->classid;
406 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
407 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
408 goto nla_put_failure;
409 tcm->tcm_handle = fh;
410 if (RTM_DELTFILTER != event) {
411 tcm->tcm_handle = 0;
412 if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
413 goto nla_put_failure;
414 }
415 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
416 return skb->len;
417
418 out_nlmsg_trim:
419 nla_put_failure:
420 nlmsg_trim(skb, b);
421 return -1;
422 }
423
424 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
425 struct nlmsghdr *n, struct tcf_proto *tp,
426 unsigned long fh, int event, bool unicast)
427 {
428 struct sk_buff *skb;
429 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
430
431 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
432 if (!skb)
433 return -ENOBUFS;
434
435 if (tcf_fill_node(net, skb, tp, fh, portid, n->nlmsg_seq,
436 n->nlmsg_flags, event) <= 0) {
437 kfree_skb(skb);
438 return -EINVAL;
439 }
440
441 if (unicast)
442 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
443
444 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
445 n->nlmsg_flags & NLM_F_ECHO);
446 }
447
448 struct tcf_dump_args {
449 struct tcf_walker w;
450 struct sk_buff *skb;
451 struct netlink_callback *cb;
452 };
453
454 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
455 struct tcf_walker *arg)
456 {
457 struct tcf_dump_args *a = (void *)arg;
458 struct net *net = sock_net(a->skb->sk);
459
460 return tcf_fill_node(net, a->skb, tp, n, NETLINK_CB(a->cb->skb).portid,
461 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
462 RTM_NEWTFILTER);
463 }
464
465 /* called with RTNL */
466 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
467 {
468 struct net *net = sock_net(skb->sk);
469 int t;
470 int s_t;
471 struct net_device *dev;
472 struct Qdisc *q;
473 struct tcf_proto *tp, __rcu **chain;
474 struct tcmsg *tcm = nlmsg_data(cb->nlh);
475 unsigned long cl = 0;
476 const struct Qdisc_class_ops *cops;
477 struct tcf_dump_args arg;
478
479 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
480 return skb->len;
481 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
482 if (!dev)
483 return skb->len;
484
485 if (!tcm->tcm_parent)
486 q = dev->qdisc;
487 else
488 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
489 if (!q)
490 goto out;
491 cops = q->ops->cl_ops;
492 if (!cops)
493 goto errout;
494 if (cops->tcf_chain == NULL)
495 goto errout;
496 if (TC_H_MIN(tcm->tcm_parent)) {
497 cl = cops->get(q, tcm->tcm_parent);
498 if (cl == 0)
499 goto errout;
500 }
501 chain = cops->tcf_chain(q, cl);
502 if (chain == NULL)
503 goto errout;
504
505 s_t = cb->args[0];
506
507 for (tp = rtnl_dereference(*chain), t = 0;
508 tp; tp = rtnl_dereference(tp->next), t++) {
509 if (t < s_t)
510 continue;
511 if (TC_H_MAJ(tcm->tcm_info) &&
512 TC_H_MAJ(tcm->tcm_info) != tp->prio)
513 continue;
514 if (TC_H_MIN(tcm->tcm_info) &&
515 TC_H_MIN(tcm->tcm_info) != tp->protocol)
516 continue;
517 if (t > s_t)
518 memset(&cb->args[1], 0,
519 sizeof(cb->args)-sizeof(cb->args[0]));
520 if (cb->args[1] == 0) {
521 if (tcf_fill_node(net, skb, tp, 0,
522 NETLINK_CB(cb->skb).portid,
523 cb->nlh->nlmsg_seq, NLM_F_MULTI,
524 RTM_NEWTFILTER) <= 0)
525 break;
526
527 cb->args[1] = 1;
528 }
529 if (tp->ops->walk == NULL)
530 continue;
531 arg.w.fn = tcf_node_dump;
532 arg.skb = skb;
533 arg.cb = cb;
534 arg.w.stop = 0;
535 arg.w.skip = cb->args[1] - 1;
536 arg.w.count = 0;
537 tp->ops->walk(tp, &arg.w);
538 cb->args[1] = arg.w.count + 1;
539 if (arg.w.stop)
540 break;
541 }
542
543 cb->args[0] = t;
544
545 errout:
546 if (cl)
547 cops->put(q, cl);
548 out:
549 return skb->len;
550 }
551
552 void tcf_exts_destroy(struct tcf_exts *exts)
553 {
554 #ifdef CONFIG_NET_CLS_ACT
555 LIST_HEAD(actions);
556
557 tcf_exts_to_list(exts, &actions);
558 tcf_action_destroy(&actions, TCA_ACT_UNBIND);
559 kfree(exts->actions);
560 exts->nr_actions = 0;
561 #endif
562 }
563 EXPORT_SYMBOL(tcf_exts_destroy);
564
565 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
566 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr)
567 {
568 #ifdef CONFIG_NET_CLS_ACT
569 {
570 struct tc_action *act;
571
572 if (exts->police && tb[exts->police]) {
573 act = tcf_action_init_1(net, tb[exts->police], rate_tlv,
574 "police", ovr, TCA_ACT_BIND);
575 if (IS_ERR(act))
576 return PTR_ERR(act);
577
578 act->type = exts->type = TCA_OLD_COMPAT;
579 exts->actions[0] = act;
580 exts->nr_actions = 1;
581 } else if (exts->action && tb[exts->action]) {
582 LIST_HEAD(actions);
583 int err, i = 0;
584
585 err = tcf_action_init(net, tb[exts->action], rate_tlv,
586 NULL, ovr, TCA_ACT_BIND,
587 &actions);
588 if (err)
589 return err;
590 list_for_each_entry(act, &actions, list)
591 exts->actions[i++] = act;
592 exts->nr_actions = i;
593 }
594 }
595 #else
596 if ((exts->action && tb[exts->action]) ||
597 (exts->police && tb[exts->police]))
598 return -EOPNOTSUPP;
599 #endif
600
601 return 0;
602 }
603 EXPORT_SYMBOL(tcf_exts_validate);
604
605 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
606 struct tcf_exts *src)
607 {
608 #ifdef CONFIG_NET_CLS_ACT
609 struct tcf_exts old = *dst;
610
611 tcf_tree_lock(tp);
612 dst->nr_actions = src->nr_actions;
613 dst->actions = src->actions;
614 dst->type = src->type;
615 tcf_tree_unlock(tp);
616
617 tcf_exts_destroy(&old);
618 #endif
619 }
620 EXPORT_SYMBOL(tcf_exts_change);
621
622 #ifdef CONFIG_NET_CLS_ACT
623 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
624 {
625 if (exts->nr_actions == 0)
626 return NULL;
627 else
628 return exts->actions[0];
629 }
630 #endif
631
632 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
633 {
634 #ifdef CONFIG_NET_CLS_ACT
635 struct nlattr *nest;
636
637 if (exts->action && exts->nr_actions) {
638 /*
639 * again for backward compatible mode - we want
640 * to work with both old and new modes of entering
641 * tc data even if iproute2 was newer - jhs
642 */
643 if (exts->type != TCA_OLD_COMPAT) {
644 LIST_HEAD(actions);
645
646 nest = nla_nest_start(skb, exts->action);
647 if (nest == NULL)
648 goto nla_put_failure;
649
650 tcf_exts_to_list(exts, &actions);
651 if (tcf_action_dump(skb, &actions, 0, 0) < 0)
652 goto nla_put_failure;
653 nla_nest_end(skb, nest);
654 } else if (exts->police) {
655 struct tc_action *act = tcf_exts_first_act(exts);
656 nest = nla_nest_start(skb, exts->police);
657 if (nest == NULL || !act)
658 goto nla_put_failure;
659 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
660 goto nla_put_failure;
661 nla_nest_end(skb, nest);
662 }
663 }
664 return 0;
665
666 nla_put_failure:
667 nla_nest_cancel(skb, nest);
668 return -1;
669 #else
670 return 0;
671 #endif
672 }
673 EXPORT_SYMBOL(tcf_exts_dump);
674
675
676 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
677 {
678 #ifdef CONFIG_NET_CLS_ACT
679 struct tc_action *a = tcf_exts_first_act(exts);
680 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
681 return -1;
682 #endif
683 return 0;
684 }
685 EXPORT_SYMBOL(tcf_exts_dump_stats);
686
687 int tcf_exts_get_dev(struct net_device *dev, struct tcf_exts *exts,
688 struct net_device **hw_dev)
689 {
690 #ifdef CONFIG_NET_CLS_ACT
691 const struct tc_action *a;
692 LIST_HEAD(actions);
693
694 if (tc_no_actions(exts))
695 return -EINVAL;
696
697 tcf_exts_to_list(exts, &actions);
698 list_for_each_entry(a, &actions, list) {
699 if (a->ops->get_dev) {
700 a->ops->get_dev(a, dev_net(dev), hw_dev);
701 break;
702 }
703 }
704 if (*hw_dev)
705 return 0;
706 #endif
707 return -EOPNOTSUPP;
708 }
709 EXPORT_SYMBOL(tcf_exts_get_dev);
710
711 static int __init tc_filter_init(void)
712 {
713 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL);
714 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL);
715 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
716 tc_dump_tfilter, NULL);
717
718 return 0;
719 }
720
721 subsys_initcall(tc_filter_init);