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