]> git.ipfire.org Git - thirdparty/kernel/stable.git/blame - net/sched/cls_api.c
net: sched: introduce multichain support for filters
[thirdparty/kernel/stable.git] / net / sched / cls_api.c
CommitLineData
1da177e4
LT
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
1da177e4
LT
17#include <linux/module.h>
18#include <linux/types.h>
19#include <linux/kernel.h>
1da177e4 20#include <linux/string.h>
1da177e4 21#include <linux/errno.h>
33a48927 22#include <linux/err.h>
1da177e4 23#include <linux/skbuff.h>
1da177e4
LT
24#include <linux/init.h>
25#include <linux/kmod.h>
ab27cfb8 26#include <linux/err.h>
5a0e3ad6 27#include <linux/slab.h>
b854272b
DL
28#include <net/net_namespace.h>
29#include <net/sock.h>
dc5fc579 30#include <net/netlink.h>
1da177e4
LT
31#include <net/pkt_sched.h>
32#include <net/pkt_cls.h>
33
1da177e4 34/* The list of all installed classifier types */
36272874 35static LIST_HEAD(tcf_proto_base);
1da177e4
LT
36
37/* Protects list of registered TC modules. It is pure SMP lock. */
38static DEFINE_RWLOCK(cls_mod_lock);
39
40/* Find classifier type by string name */
41
33a48927 42static const struct tcf_proto_ops *tcf_proto_lookup_ops(const char *kind)
1da177e4 43{
dcd76081 44 const struct tcf_proto_ops *t, *res = NULL;
1da177e4
LT
45
46 if (kind) {
47 read_lock(&cls_mod_lock);
36272874 48 list_for_each_entry(t, &tcf_proto_base, head) {
33a48927 49 if (strcmp(kind, t->kind) == 0) {
dcd76081
ED
50 if (try_module_get(t->owner))
51 res = t;
1da177e4
LT
52 break;
53 }
54 }
55 read_unlock(&cls_mod_lock);
56 }
dcd76081 57 return res;
1da177e4
LT
58}
59
60/* Register(unregister) new classifier type */
61
62int register_tcf_proto_ops(struct tcf_proto_ops *ops)
63{
36272874 64 struct tcf_proto_ops *t;
1da177e4
LT
65 int rc = -EEXIST;
66
67 write_lock(&cls_mod_lock);
36272874 68 list_for_each_entry(t, &tcf_proto_base, head)
1da177e4
LT
69 if (!strcmp(ops->kind, t->kind))
70 goto out;
71
36272874 72 list_add_tail(&ops->head, &tcf_proto_base);
1da177e4
LT
73 rc = 0;
74out:
75 write_unlock(&cls_mod_lock);
76 return rc;
77}
aa767bfe 78EXPORT_SYMBOL(register_tcf_proto_ops);
1da177e4
LT
79
80int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
81{
36272874 82 struct tcf_proto_ops *t;
1da177e4
LT
83 int rc = -ENOENT;
84
c78e1746
DB
85 /* Wait for outstanding call_rcu()s, if any, from a
86 * tcf_proto_ops's destroy() handler.
87 */
88 rcu_barrier();
89
1da177e4 90 write_lock(&cls_mod_lock);
dcd76081
ED
91 list_for_each_entry(t, &tcf_proto_base, head) {
92 if (t == ops) {
93 list_del(&t->head);
94 rc = 0;
1da177e4 95 break;
dcd76081
ED
96 }
97 }
1da177e4
LT
98 write_unlock(&cls_mod_lock);
99 return rc;
100}
aa767bfe 101EXPORT_SYMBOL(unregister_tcf_proto_ops);
1da177e4 102
7316ae88
TG
103static int tfilter_notify(struct net *net, struct sk_buff *oskb,
104 struct nlmsghdr *n, struct tcf_proto *tp,
fa59b27c 105 unsigned long fh, int event, bool unicast);
1da177e4 106
ea7f8277
DB
107static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
108 struct nlmsghdr *n,
2190d1d0 109 struct tcf_chain *chain, int event)
ea7f8277 110{
ea7f8277
DB
111 struct tcf_proto *tp;
112
2190d1d0
JP
113 for (tp = rtnl_dereference(chain->filter_chain);
114 tp; tp = rtnl_dereference(tp->next))
19a8bb28 115 tfilter_notify(net, oskb, n, tp, 0, event, false);
ea7f8277 116}
1da177e4
LT
117
118/* Select new prio value from the range, managed by kernel. */
119
aa767bfe 120static inline u32 tcf_auto_prio(struct tcf_proto *tp)
1da177e4 121{
aa767bfe 122 u32 first = TC_H_MAKE(0xC0000000U, 0U);
1da177e4
LT
123
124 if (tp)
cc7ec456 125 first = tp->prio - 1;
1da177e4 126
7961973a 127 return TC_H_MAJ(first);
1da177e4
LT
128}
129
33a48927 130static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
6529eaba 131 u32 prio, u32 parent, struct Qdisc *q,
5bc17018 132 struct tcf_chain *chain)
33a48927
JP
133{
134 struct tcf_proto *tp;
135 int err;
136
137 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
138 if (!tp)
139 return ERR_PTR(-ENOBUFS);
140
141 err = -ENOENT;
142 tp->ops = tcf_proto_lookup_ops(kind);
143 if (!tp->ops) {
144#ifdef CONFIG_MODULES
145 rtnl_unlock();
146 request_module("cls_%s", kind);
147 rtnl_lock();
148 tp->ops = tcf_proto_lookup_ops(kind);
149 /* We dropped the RTNL semaphore in order to perform
150 * the module load. So, even if we succeeded in loading
151 * the module we have to replay the request. We indicate
152 * this using -EAGAIN.
153 */
154 if (tp->ops) {
155 module_put(tp->ops->owner);
156 err = -EAGAIN;
157 } else {
158 err = -ENOENT;
159 }
160 goto errout;
161#endif
162 }
163 tp->classify = tp->ops->classify;
164 tp->protocol = protocol;
165 tp->prio = prio;
166 tp->classid = parent;
167 tp->q = q;
5bc17018 168 tp->chain = chain;
33a48927
JP
169
170 err = tp->ops->init(tp);
171 if (err) {
172 module_put(tp->ops->owner);
173 goto errout;
174 }
175 return tp;
176
177errout:
178 kfree(tp);
179 return ERR_PTR(err);
180}
181
763dbf63 182static void tcf_proto_destroy(struct tcf_proto *tp)
cf1facda 183{
763dbf63
WC
184 tp->ops->destroy(tp);
185 module_put(tp->ops->owner);
186 kfree_rcu(tp, rcu);
cf1facda
JP
187}
188
5bc17018
JP
189static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
190 u32 chain_index)
2190d1d0 191{
5bc17018
JP
192 struct tcf_chain *chain;
193
194 chain = kzalloc(sizeof(*chain), GFP_KERNEL);
195 if (!chain)
196 return NULL;
197 list_add_tail(&chain->list, &block->chain_list);
198 chain->block = block;
199 chain->index = chain_index;
200 chain->refcnt = 1;
201 return chain;
2190d1d0
JP
202}
203
204static void tcf_chain_destroy(struct tcf_chain *chain)
cf1facda
JP
205{
206 struct tcf_proto *tp;
207
5bc17018 208 list_del(&chain->list);
2190d1d0
JP
209 while ((tp = rtnl_dereference(chain->filter_chain)) != NULL) {
210 RCU_INIT_POINTER(chain->filter_chain, tp->next);
763dbf63 211 tcf_proto_destroy(tp);
cf1facda 212 }
2190d1d0
JP
213 kfree(chain);
214}
215
5bc17018
JP
216struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index)
217{
218 struct tcf_chain *chain;
219
220 list_for_each_entry(chain, &block->chain_list, list) {
221 if (chain->index == chain_index) {
222 chain->refcnt++;
223 return chain;
224 }
225 }
226 return tcf_chain_create(block, chain_index);
227}
228EXPORT_SYMBOL(tcf_chain_get);
229
230void tcf_chain_put(struct tcf_chain *chain)
231{
232 /* Destroy unused chain, with exception of chain 0, which is the
233 * default one and has to be always present.
234 */
235 if (--chain->refcnt == 0 && !chain->filter_chain && chain->index != 0)
236 tcf_chain_destroy(chain);
237}
238EXPORT_SYMBOL(tcf_chain_put);
239
2190d1d0
JP
240static void
241tcf_chain_filter_chain_ptr_set(struct tcf_chain *chain,
242 struct tcf_proto __rcu **p_filter_chain)
243{
244 chain->p_filter_chain = p_filter_chain;
cf1facda 245}
6529eaba
JP
246
247int tcf_block_get(struct tcf_block **p_block,
248 struct tcf_proto __rcu **p_filter_chain)
249{
250 struct tcf_block *block = kzalloc(sizeof(*block), GFP_KERNEL);
5bc17018 251 struct tcf_chain *chain;
2190d1d0 252 int err;
6529eaba
JP
253
254 if (!block)
255 return -ENOMEM;
5bc17018
JP
256 INIT_LIST_HEAD(&block->chain_list);
257 /* Create chain 0 by default, it has to be always present. */
258 chain = tcf_chain_create(block, 0);
259 if (!chain) {
2190d1d0
JP
260 err = -ENOMEM;
261 goto err_chain_create;
262 }
5bc17018 263 tcf_chain_filter_chain_ptr_set(chain, p_filter_chain);
6529eaba
JP
264 *p_block = block;
265 return 0;
2190d1d0
JP
266
267err_chain_create:
268 kfree(block);
269 return err;
6529eaba
JP
270}
271EXPORT_SYMBOL(tcf_block_get);
272
273void tcf_block_put(struct tcf_block *block)
274{
5bc17018
JP
275 struct tcf_chain *chain, *tmp;
276
6529eaba
JP
277 if (!block)
278 return;
5bc17018
JP
279
280 list_for_each_entry_safe(chain, tmp, &block->chain_list, list)
281 tcf_chain_destroy(chain);
6529eaba
JP
282 kfree(block);
283}
284EXPORT_SYMBOL(tcf_block_put);
cf1facda 285
87d83093
JP
286/* Main classifier routine: scans classifier chain attached
287 * to this qdisc, (optionally) tests for protocol and asks
288 * specific classifiers.
289 */
290int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
291 struct tcf_result *res, bool compat_mode)
292{
293 __be16 protocol = tc_skb_protocol(skb);
294#ifdef CONFIG_NET_CLS_ACT
295 const int max_reclassify_loop = 4;
296 const struct tcf_proto *old_tp = tp;
297 int limit = 0;
298
299reclassify:
300#endif
301 for (; tp; tp = rcu_dereference_bh(tp->next)) {
302 int err;
303
304 if (tp->protocol != protocol &&
305 tp->protocol != htons(ETH_P_ALL))
306 continue;
307
308 err = tp->classify(skb, tp, res);
309#ifdef CONFIG_NET_CLS_ACT
310 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode))
311 goto reset;
312#endif
313 if (err >= 0)
314 return err;
315 }
316
317 return TC_ACT_UNSPEC; /* signal: continue lookup */
318#ifdef CONFIG_NET_CLS_ACT
319reset:
320 if (unlikely(limit++ >= max_reclassify_loop)) {
321 net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n",
322 tp->q->ops->id, tp->prio & 0xffff,
323 ntohs(tp->protocol));
324 return TC_ACT_SHOT;
325 }
326
327 tp = old_tp;
328 protocol = tc_skb_protocol(skb);
329 goto reclassify;
330#endif
331}
332EXPORT_SYMBOL(tcf_classify);
333
2190d1d0
JP
334struct tcf_chain_info {
335 struct tcf_proto __rcu **pprev;
336 struct tcf_proto __rcu *next;
337};
338
339static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain_info *chain_info)
340{
341 return rtnl_dereference(*chain_info->pprev);
342}
343
344static void tcf_chain_tp_insert(struct tcf_chain *chain,
345 struct tcf_chain_info *chain_info,
346 struct tcf_proto *tp)
347{
348 if (chain->p_filter_chain &&
349 *chain_info->pprev == chain->filter_chain)
350 *chain->p_filter_chain = tp;
351 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain_info));
352 rcu_assign_pointer(*chain_info->pprev, tp);
353}
354
355static void tcf_chain_tp_remove(struct tcf_chain *chain,
356 struct tcf_chain_info *chain_info,
357 struct tcf_proto *tp)
358{
359 struct tcf_proto *next = rtnl_dereference(chain_info->next);
360
361 if (chain->p_filter_chain && tp == chain->filter_chain)
362 *chain->p_filter_chain = next;
363 RCU_INIT_POINTER(*chain_info->pprev, next);
364}
365
366static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
367 struct tcf_chain_info *chain_info,
368 u32 protocol, u32 prio,
369 bool prio_allocate)
370{
371 struct tcf_proto **pprev;
372 struct tcf_proto *tp;
373
374 /* Check the chain for existence of proto-tcf with this priority */
375 for (pprev = &chain->filter_chain;
376 (tp = rtnl_dereference(*pprev)); pprev = &tp->next) {
377 if (tp->prio >= prio) {
378 if (tp->prio == prio) {
379 if (prio_allocate ||
380 (tp->protocol != protocol && protocol))
381 return ERR_PTR(-EINVAL);
382 } else {
383 tp = NULL;
384 }
385 break;
386 }
387 }
388 chain_info->pprev = pprev;
389 chain_info->next = tp ? tp->next : NULL;
390 return tp;
391}
392
1da177e4
LT
393/* Add/change/delete/get a filter node */
394
c21ef3e3
DA
395static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
396 struct netlink_ext_ack *extack)
1da177e4 397{
3b1e0a65 398 struct net *net = sock_net(skb->sk);
add93b61 399 struct nlattr *tca[TCA_MAX + 1];
1da177e4
LT
400 struct tcmsg *t;
401 u32 protocol;
402 u32 prio;
9d36d9e5 403 bool prio_allocate;
1da177e4 404 u32 parent;
5bc17018 405 u32 chain_index;
1da177e4
LT
406 struct net_device *dev;
407 struct Qdisc *q;
2190d1d0 408 struct tcf_chain_info chain_info;
5bc17018 409 struct tcf_chain *chain = NULL;
6529eaba 410 struct tcf_block *block;
1da177e4 411 struct tcf_proto *tp;
20fea08b 412 const struct Qdisc_class_ops *cops;
1da177e4
LT
413 unsigned long cl;
414 unsigned long fh;
415 int err;
628185cf 416 int tp_created;
1da177e4 417
4e8bbb81 418 if ((n->nlmsg_type != RTM_GETTFILTER) &&
5f013c9b 419 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
dfc47ef8 420 return -EPERM;
de179c8c 421
1da177e4 422replay:
628185cf
DB
423 tp_created = 0;
424
c21ef3e3 425 err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack);
de179c8c
H
426 if (err < 0)
427 return err;
428
942b8165 429 t = nlmsg_data(n);
1da177e4
LT
430 protocol = TC_H_MIN(t->tcm_info);
431 prio = TC_H_MAJ(t->tcm_info);
9d36d9e5 432 prio_allocate = false;
1da177e4
LT
433 parent = t->tcm_parent;
434 cl = 0;
435
436 if (prio == 0) {
ea7f8277
DB
437 switch (n->nlmsg_type) {
438 case RTM_DELTFILTER:
9f6ed032 439 if (protocol || t->tcm_handle || tca[TCA_KIND])
ea7f8277
DB
440 return -ENOENT;
441 break;
442 case RTM_NEWTFILTER:
443 /* If no priority is provided by the user,
444 * we allocate one.
445 */
446 if (n->nlmsg_flags & NLM_F_CREATE) {
447 prio = TC_H_MAKE(0x80000000U, 0U);
9d36d9e5 448 prio_allocate = true;
ea7f8277
DB
449 break;
450 }
451 /* fall-through */
452 default:
1da177e4 453 return -ENOENT;
ea7f8277 454 }
1da177e4
LT
455 }
456
457 /* Find head of filter chain. */
458
459 /* Find link */
7316ae88 460 dev = __dev_get_by_index(net, t->tcm_ifindex);
aa767bfe 461 if (dev == NULL)
1da177e4
LT
462 return -ENODEV;
463
464 /* Find qdisc */
465 if (!parent) {
af356afa 466 q = dev->qdisc;
1da177e4 467 parent = q->handle;
aa767bfe
SH
468 } else {
469 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
470 if (q == NULL)
471 return -EINVAL;
472 }
1da177e4
LT
473
474 /* Is it classful? */
cc7ec456
ED
475 cops = q->ops->cl_ops;
476 if (!cops)
1da177e4
LT
477 return -EINVAL;
478
6529eaba 479 if (!cops->tcf_block)
71ebe5e9
PM
480 return -EOPNOTSUPP;
481
1da177e4
LT
482 /* Do we search for filter, attached to class? */
483 if (TC_H_MIN(parent)) {
484 cl = cops->get(q, parent);
485 if (cl == 0)
486 return -ENOENT;
487 }
488
489 /* And the last stroke */
6529eaba
JP
490 block = cops->tcf_block(q, cl);
491 if (!block) {
6bb16e7a 492 err = -EINVAL;
1da177e4 493 goto errout;
6bb16e7a 494 }
5bc17018
JP
495
496 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
497 if (chain_index > TC_ACT_EXT_VAL_MASK) {
498 err = -EINVAL;
499 goto errout;
500 }
501 chain = tcf_chain_get(block, chain_index);
502 if (!chain) {
503 err = -ENOMEM;
504 goto errout;
505 }
6529eaba 506
ea7f8277
DB
507 if (n->nlmsg_type == RTM_DELTFILTER && prio == 0) {
508 tfilter_notify_chain(net, skb, n, chain, RTM_DELTFILTER);
fbe9c5b0 509 tcf_chain_destroy(chain);
ea7f8277
DB
510 err = 0;
511 goto errout;
512 }
1da177e4 513
2190d1d0
JP
514 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
515 prio, prio_allocate);
516 if (IS_ERR(tp)) {
517 err = PTR_ERR(tp);
518 goto errout;
1da177e4
LT
519 }
520
521 if (tp == NULL) {
522 /* Proto-tcf does not exist, create new one */
523
6bb16e7a
JP
524 if (tca[TCA_KIND] == NULL || !protocol) {
525 err = -EINVAL;
1da177e4 526 goto errout;
6bb16e7a 527 }
1da177e4 528
cc7ec456 529 if (n->nlmsg_type != RTM_NEWTFILTER ||
6bb16e7a
JP
530 !(n->nlmsg_flags & NLM_F_CREATE)) {
531 err = -ENOENT;
1da177e4 532 goto errout;
6bb16e7a 533 }
1da177e4 534
9d36d9e5 535 if (prio_allocate)
2190d1d0 536 prio = tcf_auto_prio(tcf_chain_tp_prev(&chain_info));
1da177e4 537
33a48927 538 tp = tcf_proto_create(nla_data(tca[TCA_KIND]),
5bc17018 539 protocol, prio, parent, q, chain);
33a48927
JP
540 if (IS_ERR(tp)) {
541 err = PTR_ERR(tp);
1da177e4
LT
542 goto errout;
543 }
12186be7 544 tp_created = 1;
6bb16e7a
JP
545 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
546 err = -EINVAL;
1da177e4 547 goto errout;
6bb16e7a 548 }
1da177e4
LT
549
550 fh = tp->ops->get(tp, t->tcm_handle);
551
552 if (fh == 0) {
553 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
2190d1d0 554 tcf_chain_tp_remove(chain, &chain_info, tp);
fa59b27c
ED
555 tfilter_notify(net, skb, n, tp, fh,
556 RTM_DELTFILTER, false);
763dbf63 557 tcf_proto_destroy(tp);
1da177e4
LT
558 err = 0;
559 goto errout;
560 }
561
aa767bfe 562 if (n->nlmsg_type != RTM_NEWTFILTER ||
6bb16e7a
JP
563 !(n->nlmsg_flags & NLM_F_CREATE)) {
564 err = -ENOENT;
1da177e4 565 goto errout;
6bb16e7a 566 }
1da177e4 567 } else {
763dbf63
WC
568 bool last;
569
1da177e4 570 switch (n->nlmsg_type) {
10297b99 571 case RTM_NEWTFILTER:
12186be7
MU
572 if (n->nlmsg_flags & NLM_F_EXCL) {
573 if (tp_created)
763dbf63 574 tcf_proto_destroy(tp);
6bb16e7a 575 err = -EEXIST;
1da177e4 576 goto errout;
12186be7 577 }
1da177e4
LT
578 break;
579 case RTM_DELTFILTER:
763dbf63 580 err = tp->ops->delete(tp, fh, &last);
40c81b25
JP
581 if (err)
582 goto errout;
40c81b25
JP
583 tfilter_notify(net, skb, n, tp, t->tcm_handle,
584 RTM_DELTFILTER, false);
763dbf63 585 if (last) {
2190d1d0 586 tcf_chain_tp_remove(chain, &chain_info, tp);
763dbf63
WC
587 tcf_proto_destroy(tp);
588 }
d7cf52c2 589 goto errout;
1da177e4 590 case RTM_GETTFILTER:
5a7a5555 591 err = tfilter_notify(net, skb, n, tp, fh,
fa59b27c 592 RTM_NEWTFILTER, true);
1da177e4
LT
593 goto errout;
594 default:
595 err = -EINVAL;
596 goto errout;
597 }
598 }
599
2f7ef2f8
CW
600 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
601 n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE);
12186be7 602 if (err == 0) {
2190d1d0
JP
603 if (tp_created)
604 tcf_chain_tp_insert(chain, &chain_info, tp);
fa59b27c 605 tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER, false);
12186be7
MU
606 } else {
607 if (tp_created)
763dbf63 608 tcf_proto_destroy(tp);
12186be7 609 }
1da177e4
LT
610
611errout:
5bc17018
JP
612 if (chain)
613 tcf_chain_put(chain);
1da177e4
LT
614 if (cl)
615 cops->put(q, cl);
616 if (err == -EAGAIN)
617 /* Replay the request. */
618 goto replay;
619 return err;
620}
621
0b0f43fe
JHS
622static int tcf_fill_node(struct net *net, struct sk_buff *skb,
623 struct tcf_proto *tp, unsigned long fh, u32 portid,
624 u32 seq, u16 flags, int event)
1da177e4
LT
625{
626 struct tcmsg *tcm;
627 struct nlmsghdr *nlh;
27a884dc 628 unsigned char *b = skb_tail_pointer(skb);
1da177e4 629
15e47304 630 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
942b8165
DM
631 if (!nlh)
632 goto out_nlmsg_trim;
633 tcm = nlmsg_data(nlh);
1da177e4 634 tcm->tcm_family = AF_UNSPEC;
9ef1d4c7 635 tcm->tcm__pad1 = 0;
ad61df91 636 tcm->tcm__pad2 = 0;
5ce2d488 637 tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
1da177e4
LT
638 tcm->tcm_parent = tp->classid;
639 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1b34ec43
DM
640 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
641 goto nla_put_failure;
5bc17018
JP
642 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
643 goto nla_put_failure;
1da177e4
LT
644 tcm->tcm_handle = fh;
645 if (RTM_DELTFILTER != event) {
646 tcm->tcm_handle = 0;
832d1d5b 647 if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
add93b61 648 goto nla_put_failure;
1da177e4 649 }
27a884dc 650 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1da177e4
LT
651 return skb->len;
652
942b8165 653out_nlmsg_trim:
add93b61 654nla_put_failure:
dc5fc579 655 nlmsg_trim(skb, b);
1da177e4
LT
656 return -1;
657}
658
7316ae88
TG
659static int tfilter_notify(struct net *net, struct sk_buff *oskb,
660 struct nlmsghdr *n, struct tcf_proto *tp,
fa59b27c 661 unsigned long fh, int event, bool unicast)
1da177e4
LT
662{
663 struct sk_buff *skb;
15e47304 664 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1da177e4
LT
665
666 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
667 if (!skb)
668 return -ENOBUFS;
669
30a391a1
RM
670 if (tcf_fill_node(net, skb, tp, fh, portid, n->nlmsg_seq,
671 n->nlmsg_flags, event) <= 0) {
1da177e4
LT
672 kfree_skb(skb);
673 return -EINVAL;
674 }
675
fa59b27c
ED
676 if (unicast)
677 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
678
15e47304 679 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
aa767bfe 680 n->nlmsg_flags & NLM_F_ECHO);
1da177e4
LT
681}
682
aa767bfe 683struct tcf_dump_args {
1da177e4
LT
684 struct tcf_walker w;
685 struct sk_buff *skb;
686 struct netlink_callback *cb;
687};
688
aa767bfe
SH
689static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
690 struct tcf_walker *arg)
1da177e4 691{
aa767bfe 692 struct tcf_dump_args *a = (void *)arg;
832d1d5b 693 struct net *net = sock_net(a->skb->sk);
1da177e4 694
832d1d5b 695 return tcf_fill_node(net, a->skb, tp, n, NETLINK_CB(a->cb->skb).portid,
5a7a5555
JHS
696 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
697 RTM_NEWTFILTER);
1da177e4
LT
698}
699
5bc17018 700static bool tcf_chain_dump(struct tcf_chain *chain, struct sk_buff *skb,
acb31fae
JP
701 struct netlink_callback *cb,
702 long index_start, long *p_index)
703{
704 struct net *net = sock_net(skb->sk);
705 struct tcmsg *tcm = nlmsg_data(cb->nlh);
706 struct tcf_dump_args arg;
707 struct tcf_proto *tp;
708
709 for (tp = rtnl_dereference(chain->filter_chain);
710 tp; tp = rtnl_dereference(tp->next), (*p_index)++) {
711 if (*p_index < index_start)
712 continue;
713 if (TC_H_MAJ(tcm->tcm_info) &&
714 TC_H_MAJ(tcm->tcm_info) != tp->prio)
715 continue;
716 if (TC_H_MIN(tcm->tcm_info) &&
717 TC_H_MIN(tcm->tcm_info) != tp->protocol)
718 continue;
719 if (*p_index > index_start)
720 memset(&cb->args[1], 0,
721 sizeof(cb->args) - sizeof(cb->args[0]));
722 if (cb->args[1] == 0) {
723 if (tcf_fill_node(net, skb, tp, 0,
724 NETLINK_CB(cb->skb).portid,
725 cb->nlh->nlmsg_seq, NLM_F_MULTI,
726 RTM_NEWTFILTER) <= 0)
5bc17018 727 return false;
acb31fae
JP
728
729 cb->args[1] = 1;
730 }
731 if (!tp->ops->walk)
732 continue;
733 arg.w.fn = tcf_node_dump;
734 arg.skb = skb;
735 arg.cb = cb;
736 arg.w.stop = 0;
737 arg.w.skip = cb->args[1] - 1;
738 arg.w.count = 0;
739 tp->ops->walk(tp, &arg.w);
740 cb->args[1] = arg.w.count + 1;
741 if (arg.w.stop)
5bc17018 742 return false;
acb31fae 743 }
5bc17018 744 return true;
acb31fae
JP
745}
746
bd27a875 747/* called with RTNL */
1da177e4
LT
748static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
749{
3b1e0a65 750 struct net *net = sock_net(skb->sk);
5bc17018 751 struct nlattr *tca[TCA_MAX + 1];
1da177e4
LT
752 struct net_device *dev;
753 struct Qdisc *q;
6529eaba 754 struct tcf_block *block;
2190d1d0 755 struct tcf_chain *chain;
942b8165 756 struct tcmsg *tcm = nlmsg_data(cb->nlh);
1da177e4 757 unsigned long cl = 0;
20fea08b 758 const struct Qdisc_class_ops *cops;
acb31fae
JP
759 long index_start;
760 long index;
5bc17018 761 int err;
1da177e4 762
573ce260 763 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1da177e4 764 return skb->len;
5bc17018
JP
765
766 err = nlmsg_parse(cb->nlh, sizeof(*tcm), tca, TCA_MAX, NULL, NULL);
767 if (err)
768 return err;
769
cc7ec456
ED
770 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
771 if (!dev)
1da177e4
LT
772 return skb->len;
773
1da177e4 774 if (!tcm->tcm_parent)
af356afa 775 q = dev->qdisc;
1da177e4
LT
776 else
777 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
778 if (!q)
779 goto out;
cc7ec456
ED
780 cops = q->ops->cl_ops;
781 if (!cops)
1da177e4 782 goto errout;
6529eaba 783 if (!cops->tcf_block)
71ebe5e9 784 goto errout;
1da177e4
LT
785 if (TC_H_MIN(tcm->tcm_parent)) {
786 cl = cops->get(q, tcm->tcm_parent);
787 if (cl == 0)
788 goto errout;
789 }
6529eaba
JP
790 block = cops->tcf_block(q, cl);
791 if (!block)
1da177e4
LT
792 goto errout;
793
acb31fae
JP
794 index_start = cb->args[0];
795 index = 0;
5bc17018
JP
796
797 list_for_each_entry(chain, &block->chain_list, list) {
798 if (tca[TCA_CHAIN] &&
799 nla_get_u32(tca[TCA_CHAIN]) != chain->index)
800 continue;
801 if (!tcf_chain_dump(chain, skb, cb, index_start, &index))
802 break;
803 }
804
acb31fae 805 cb->args[0] = index;
1da177e4
LT
806
807errout:
808 if (cl)
809 cops->put(q, cl);
810out:
1da177e4
LT
811 return skb->len;
812}
813
18d0264f 814void tcf_exts_destroy(struct tcf_exts *exts)
1da177e4
LT
815{
816#ifdef CONFIG_NET_CLS_ACT
22dc13c8
WC
817 LIST_HEAD(actions);
818
819 tcf_exts_to_list(exts, &actions);
820 tcf_action_destroy(&actions, TCA_ACT_UNBIND);
821 kfree(exts->actions);
822 exts->nr_actions = 0;
1da177e4
LT
823#endif
824}
aa767bfe 825EXPORT_SYMBOL(tcf_exts_destroy);
1da177e4 826
c1b52739 827int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
5a7a5555 828 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr)
1da177e4 829{
1da177e4
LT
830#ifdef CONFIG_NET_CLS_ACT
831 {
1da177e4
LT
832 struct tc_action *act;
833
5da57f42
WC
834 if (exts->police && tb[exts->police]) {
835 act = tcf_action_init_1(net, tb[exts->police], rate_tlv,
5a7a5555 836 "police", ovr, TCA_ACT_BIND);
ab27cfb8
PM
837 if (IS_ERR(act))
838 return PTR_ERR(act);
1da177e4 839
33be6271 840 act->type = exts->type = TCA_OLD_COMPAT;
22dc13c8
WC
841 exts->actions[0] = act;
842 exts->nr_actions = 1;
5da57f42 843 } else if (exts->action && tb[exts->action]) {
22dc13c8
WC
844 LIST_HEAD(actions);
845 int err, i = 0;
846
5da57f42 847 err = tcf_action_init(net, tb[exts->action], rate_tlv,
5a7a5555
JHS
848 NULL, ovr, TCA_ACT_BIND,
849 &actions);
33be6271
WC
850 if (err)
851 return err;
22dc13c8
WC
852 list_for_each_entry(act, &actions, list)
853 exts->actions[i++] = act;
854 exts->nr_actions = i;
1da177e4
LT
855 }
856 }
1da177e4 857#else
5da57f42
WC
858 if ((exts->action && tb[exts->action]) ||
859 (exts->police && tb[exts->police]))
1da177e4
LT
860 return -EOPNOTSUPP;
861#endif
862
863 return 0;
864}
aa767bfe 865EXPORT_SYMBOL(tcf_exts_validate);
1da177e4 866
aa767bfe
SH
867void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
868 struct tcf_exts *src)
1da177e4
LT
869{
870#ifdef CONFIG_NET_CLS_ACT
22dc13c8
WC
871 struct tcf_exts old = *dst;
872
a49eb42a 873 tcf_tree_lock(tp);
22dc13c8
WC
874 dst->nr_actions = src->nr_actions;
875 dst->actions = src->actions;
5301e3e1 876 dst->type = src->type;
a49eb42a 877 tcf_tree_unlock(tp);
22dc13c8
WC
878
879 tcf_exts_destroy(&old);
1da177e4
LT
880#endif
881}
aa767bfe 882EXPORT_SYMBOL(tcf_exts_change);
1da177e4 883
22dc13c8
WC
884#ifdef CONFIG_NET_CLS_ACT
885static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
886{
887 if (exts->nr_actions == 0)
888 return NULL;
889 else
890 return exts->actions[0];
891}
892#endif
33be6271 893
5da57f42 894int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
1da177e4
LT
895{
896#ifdef CONFIG_NET_CLS_ACT
9cc63db5
CW
897 struct nlattr *nest;
898
22dc13c8 899 if (exts->action && exts->nr_actions) {
1da177e4
LT
900 /*
901 * again for backward compatible mode - we want
902 * to work with both old and new modes of entering
903 * tc data even if iproute2 was newer - jhs
904 */
33be6271 905 if (exts->type != TCA_OLD_COMPAT) {
22dc13c8
WC
906 LIST_HEAD(actions);
907
5da57f42 908 nest = nla_nest_start(skb, exts->action);
4b3550ef
PM
909 if (nest == NULL)
910 goto nla_put_failure;
22dc13c8
WC
911
912 tcf_exts_to_list(exts, &actions);
913 if (tcf_action_dump(skb, &actions, 0, 0) < 0)
add93b61 914 goto nla_put_failure;
4b3550ef 915 nla_nest_end(skb, nest);
5da57f42 916 } else if (exts->police) {
33be6271 917 struct tc_action *act = tcf_exts_first_act(exts);
5da57f42 918 nest = nla_nest_start(skb, exts->police);
63acd680 919 if (nest == NULL || !act)
4b3550ef 920 goto nla_put_failure;
33be6271 921 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
add93b61 922 goto nla_put_failure;
4b3550ef 923 nla_nest_end(skb, nest);
1da177e4
LT
924 }
925 }
1da177e4 926 return 0;
9cc63db5
CW
927
928nla_put_failure:
929 nla_nest_cancel(skb, nest);
1da177e4 930 return -1;
9cc63db5
CW
931#else
932 return 0;
933#endif
1da177e4 934}
aa767bfe 935EXPORT_SYMBOL(tcf_exts_dump);
1da177e4 936
aa767bfe 937
5da57f42 938int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
1da177e4
LT
939{
940#ifdef CONFIG_NET_CLS_ACT
33be6271 941 struct tc_action *a = tcf_exts_first_act(exts);
b057df24 942 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
33be6271 943 return -1;
1da177e4
LT
944#endif
945 return 0;
1da177e4 946}
aa767bfe 947EXPORT_SYMBOL(tcf_exts_dump_stats);
1da177e4 948
7091d8c7
HHZ
949int tcf_exts_get_dev(struct net_device *dev, struct tcf_exts *exts,
950 struct net_device **hw_dev)
951{
952#ifdef CONFIG_NET_CLS_ACT
953 const struct tc_action *a;
954 LIST_HEAD(actions);
955
956 if (tc_no_actions(exts))
957 return -EINVAL;
958
959 tcf_exts_to_list(exts, &actions);
960 list_for_each_entry(a, &actions, list) {
961 if (a->ops->get_dev) {
962 a->ops->get_dev(a, dev_net(dev), hw_dev);
963 break;
964 }
965 }
966 if (*hw_dev)
967 return 0;
968#endif
969 return -EOPNOTSUPP;
970}
971EXPORT_SYMBOL(tcf_exts_get_dev);
972
1da177e4
LT
973static int __init tc_filter_init(void)
974{
c7ac8679
GR
975 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL);
976 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL);
82623c0d 977 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
c7ac8679 978 tc_dump_tfilter, NULL);
1da177e4 979
1da177e4
LT
980 return 0;
981}
982
983subsys_initcall(tc_filter_init);