Loading...
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/netlink.h>
26#include <linux/err.h>
27#include <linux/slab.h>
28#include <net/net_namespace.h>
29#include <net/sock.h>
30#include <net/netlink.h>
31#include <net/pkt_sched.h>
32#include <net/pkt_cls.h>
33
34/* The list of all installed classifier types */
35
36static struct tcf_proto_ops *tcf_proto_base __read_mostly;
37
38/* Protects list of registered TC modules. It is pure SMP lock. */
39static DEFINE_RWLOCK(cls_mod_lock);
40
41/* Find classifier type by string name */
42
43static const struct tcf_proto_ops *tcf_proto_lookup_ops(struct nlattr *kind)
44{
45 const struct tcf_proto_ops *t = NULL;
46
47 if (kind) {
48 read_lock(&cls_mod_lock);
49 for (t = tcf_proto_base; t; t = t->next) {
50 if (nla_strcmp(kind, t->kind) == 0) {
51 if (!try_module_get(t->owner))
52 t = NULL;
53 break;
54 }
55 }
56 read_unlock(&cls_mod_lock);
57 }
58 return t;
59}
60
61/* Register(unregister) new classifier type */
62
63int register_tcf_proto_ops(struct tcf_proto_ops *ops)
64{
65 struct tcf_proto_ops *t, **tp;
66 int rc = -EEXIST;
67
68 write_lock(&cls_mod_lock);
69 for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
70 if (!strcmp(ops->kind, t->kind))
71 goto out;
72
73 ops->next = NULL;
74 *tp = ops;
75 rc = 0;
76out:
77 write_unlock(&cls_mod_lock);
78 return rc;
79}
80EXPORT_SYMBOL(register_tcf_proto_ops);
81
82int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
83{
84 struct tcf_proto_ops *t, **tp;
85 int rc = -ENOENT;
86
87 write_lock(&cls_mod_lock);
88 for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
89 if (t == ops)
90 break;
91
92 if (!t)
93 goto out;
94 *tp = t->next;
95 rc = 0;
96out:
97 write_unlock(&cls_mod_lock);
98 return rc;
99}
100EXPORT_SYMBOL(unregister_tcf_proto_ops);
101
102static 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
109static 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
121static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
122{
123 struct net *net = sock_net(skb->sk);
124 struct nlattr *tca[TCA_MAX + 1];
125 spinlock_t *root_lock;
126 struct tcmsg *t;
127 u32 protocol;
128 u32 prio;
129 u32 nprio;
130 u32 parent;
131 struct net_device *dev;
132 struct Qdisc *q;
133 struct tcf_proto **back, **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 = 0;
141
142replay:
143 t = NLMSG_DATA(n);
144 protocol = TC_H_MIN(t->tcm_info);
145 prio = TC_H_MAJ(t->tcm_info);
146 nprio = prio;
147 parent = t->tcm_parent;
148 cl = 0;
149
150 if (prio == 0) {
151 /* If no priority is given, user wants we allocated it. */
152 if (n->nlmsg_type != RTM_NEWTFILTER ||
153 !(n->nlmsg_flags & NLM_F_CREATE))
154 return -ENOENT;
155 prio = TC_H_MAKE(0x80000000U, 0U);
156 }
157
158 /* Find head of filter chain. */
159
160 /* Find link */
161 dev = __dev_get_by_index(net, t->tcm_ifindex);
162 if (dev == NULL)
163 return -ENODEV;
164
165 err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
166 if (err < 0)
167 return err;
168
169 /* Find qdisc */
170 if (!parent) {
171 q = dev->qdisc;
172 parent = q->handle;
173 } else {
174 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
175 if (q == NULL)
176 return -EINVAL;
177 }
178
179 /* Is it classful? */
180 cops = q->ops->cl_ops;
181 if (!cops)
182 return -EINVAL;
183
184 if (cops->tcf_chain == NULL)
185 return -EOPNOTSUPP;
186
187 /* Do we search for filter, attached to class? */
188 if (TC_H_MIN(parent)) {
189 cl = cops->get(q, parent);
190 if (cl == 0)
191 return -ENOENT;
192 }
193
194 /* And the last stroke */
195 chain = cops->tcf_chain(q, cl);
196 err = -EINVAL;
197 if (chain == NULL)
198 goto errout;
199
200 /* Check the chain for existence of proto-tcf with this priority */
201 for (back = chain; (tp = *back) != NULL; back = &tp->next) {
202 if (tp->prio >= prio) {
203 if (tp->prio == prio) {
204 if (!nprio ||
205 (tp->protocol != protocol && protocol))
206 goto errout;
207 } else
208 tp = NULL;
209 break;
210 }
211 }
212
213 root_lock = qdisc_root_sleeping_lock(q);
214
215 if (tp == NULL) {
216 /* Proto-tcf does not exist, create new one */
217
218 if (tca[TCA_KIND] == NULL || !protocol)
219 goto errout;
220
221 err = -ENOENT;
222 if (n->nlmsg_type != RTM_NEWTFILTER ||
223 !(n->nlmsg_flags & NLM_F_CREATE))
224 goto errout;
225
226
227 /* Create new proto tcf */
228
229 err = -ENOBUFS;
230 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
231 if (tp == NULL)
232 goto errout;
233 err = -ENOENT;
234 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
235 if (tp_ops == NULL) {
236#ifdef CONFIG_MODULES
237 struct nlattr *kind = tca[TCA_KIND];
238 char name[IFNAMSIZ];
239
240 if (kind != NULL &&
241 nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
242 rtnl_unlock();
243 request_module("cls_%s", name);
244 rtnl_lock();
245 tp_ops = tcf_proto_lookup_ops(kind);
246 /* We dropped the RTNL semaphore in order to
247 * perform the module load. So, even if we
248 * succeeded in loading the module we have to
249 * replay the request. We indicate this using
250 * -EAGAIN.
251 */
252 if (tp_ops != NULL) {
253 module_put(tp_ops->owner);
254 err = -EAGAIN;
255 }
256 }
257#endif
258 kfree(tp);
259 goto errout;
260 }
261 tp->ops = tp_ops;
262 tp->protocol = protocol;
263 tp->prio = nprio ? : TC_H_MAJ(tcf_auto_prio(*back));
264 tp->q = q;
265 tp->classify = tp_ops->classify;
266 tp->classid = parent;
267
268 err = tp_ops->init(tp);
269 if (err != 0) {
270 module_put(tp_ops->owner);
271 kfree(tp);
272 goto errout;
273 }
274
275 tp_created = 1;
276
277 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
278 goto errout;
279
280 fh = tp->ops->get(tp, t->tcm_handle);
281
282 if (fh == 0) {
283 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
284 spin_lock_bh(root_lock);
285 *back = tp->next;
286 spin_unlock_bh(root_lock);
287
288 tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
289 tcf_destroy(tp);
290 err = 0;
291 goto errout;
292 }
293
294 err = -ENOENT;
295 if (n->nlmsg_type != RTM_NEWTFILTER ||
296 !(n->nlmsg_flags & NLM_F_CREATE))
297 goto errout;
298 } else {
299 switch (n->nlmsg_type) {
300 case RTM_NEWTFILTER:
301 err = -EEXIST;
302 if (n->nlmsg_flags & NLM_F_EXCL) {
303 if (tp_created)
304 tcf_destroy(tp);
305 goto errout;
306 }
307 break;
308 case RTM_DELTFILTER:
309 err = tp->ops->delete(tp, fh);
310 if (err == 0)
311 tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
312 goto errout;
313 case RTM_GETTFILTER:
314 err = tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
315 goto errout;
316 default:
317 err = -EINVAL;
318 goto errout;
319 }
320 }
321
322 err = tp->ops->change(tp, cl, t->tcm_handle, tca, &fh);
323 if (err == 0) {
324 if (tp_created) {
325 spin_lock_bh(root_lock);
326 tp->next = *back;
327 *back = tp;
328 spin_unlock_bh(root_lock);
329 }
330 tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
331 } else {
332 if (tp_created)
333 tcf_destroy(tp);
334 }
335
336errout:
337 if (cl)
338 cops->put(q, cl);
339 if (err == -EAGAIN)
340 /* Replay the request. */
341 goto replay;
342 return err;
343}
344
345static int tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp,
346 unsigned long fh, u32 pid, u32 seq, u16 flags, int event)
347{
348 struct tcmsg *tcm;
349 struct nlmsghdr *nlh;
350 unsigned char *b = skb_tail_pointer(skb);
351
352 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
353 tcm = NLMSG_DATA(nlh);
354 tcm->tcm_family = AF_UNSPEC;
355 tcm->tcm__pad1 = 0;
356 tcm->tcm__pad2 = 0;
357 tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
358 tcm->tcm_parent = tp->classid;
359 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
360 NLA_PUT_STRING(skb, TCA_KIND, tp->ops->kind);
361 tcm->tcm_handle = fh;
362 if (RTM_DELTFILTER != event) {
363 tcm->tcm_handle = 0;
364 if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
365 goto nla_put_failure;
366 }
367 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
368 return skb->len;
369
370nlmsg_failure:
371nla_put_failure:
372 nlmsg_trim(skb, b);
373 return -1;
374}
375
376static int tfilter_notify(struct net *net, struct sk_buff *oskb,
377 struct nlmsghdr *n, struct tcf_proto *tp,
378 unsigned long fh, int event)
379{
380 struct sk_buff *skb;
381 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
382
383 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
384 if (!skb)
385 return -ENOBUFS;
386
387 if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) {
388 kfree_skb(skb);
389 return -EINVAL;
390 }
391
392 return rtnetlink_send(skb, net, pid, RTNLGRP_TC,
393 n->nlmsg_flags & NLM_F_ECHO);
394}
395
396struct tcf_dump_args {
397 struct tcf_walker w;
398 struct sk_buff *skb;
399 struct netlink_callback *cb;
400};
401
402static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
403 struct tcf_walker *arg)
404{
405 struct tcf_dump_args *a = (void *)arg;
406
407 return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid,
408 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
409}
410
411/* called with RTNL */
412static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
413{
414 struct net *net = sock_net(skb->sk);
415 int t;
416 int s_t;
417 struct net_device *dev;
418 struct Qdisc *q;
419 struct tcf_proto *tp, **chain;
420 struct tcmsg *tcm = (struct tcmsg *)NLMSG_DATA(cb->nlh);
421 unsigned long cl = 0;
422 const struct Qdisc_class_ops *cops;
423 struct tcf_dump_args arg;
424
425 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
426 return skb->len;
427 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
428 if (!dev)
429 return skb->len;
430
431 if (!tcm->tcm_parent)
432 q = dev->qdisc;
433 else
434 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
435 if (!q)
436 goto out;
437 cops = q->ops->cl_ops;
438 if (!cops)
439 goto errout;
440 if (cops->tcf_chain == NULL)
441 goto errout;
442 if (TC_H_MIN(tcm->tcm_parent)) {
443 cl = cops->get(q, tcm->tcm_parent);
444 if (cl == 0)
445 goto errout;
446 }
447 chain = cops->tcf_chain(q, cl);
448 if (chain == NULL)
449 goto errout;
450
451 s_t = cb->args[0];
452
453 for (tp = *chain, t = 0; tp; tp = tp->next, t++) {
454 if (t < s_t)
455 continue;
456 if (TC_H_MAJ(tcm->tcm_info) &&
457 TC_H_MAJ(tcm->tcm_info) != tp->prio)
458 continue;
459 if (TC_H_MIN(tcm->tcm_info) &&
460 TC_H_MIN(tcm->tcm_info) != tp->protocol)
461 continue;
462 if (t > s_t)
463 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
464 if (cb->args[1] == 0) {
465 if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid,
466 cb->nlh->nlmsg_seq, NLM_F_MULTI,
467 RTM_NEWTFILTER) <= 0)
468 break;
469
470 cb->args[1] = 1;
471 }
472 if (tp->ops->walk == NULL)
473 continue;
474 arg.w.fn = tcf_node_dump;
475 arg.skb = skb;
476 arg.cb = cb;
477 arg.w.stop = 0;
478 arg.w.skip = cb->args[1] - 1;
479 arg.w.count = 0;
480 tp->ops->walk(tp, &arg.w);
481 cb->args[1] = arg.w.count + 1;
482 if (arg.w.stop)
483 break;
484 }
485
486 cb->args[0] = t;
487
488errout:
489 if (cl)
490 cops->put(q, cl);
491out:
492 return skb->len;
493}
494
495void tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
496{
497#ifdef CONFIG_NET_CLS_ACT
498 if (exts->action) {
499 tcf_action_destroy(exts->action, TCA_ACT_UNBIND);
500 exts->action = NULL;
501 }
502#endif
503}
504EXPORT_SYMBOL(tcf_exts_destroy);
505
506int tcf_exts_validate(struct tcf_proto *tp, struct nlattr **tb,
507 struct nlattr *rate_tlv, struct tcf_exts *exts,
508 const struct tcf_ext_map *map)
509{
510 memset(exts, 0, sizeof(*exts));
511
512#ifdef CONFIG_NET_CLS_ACT
513 {
514 struct tc_action *act;
515
516 if (map->police && tb[map->police]) {
517 act = tcf_action_init_1(tb[map->police], rate_tlv,
518 "police", TCA_ACT_NOREPLACE,
519 TCA_ACT_BIND);
520 if (IS_ERR(act))
521 return PTR_ERR(act);
522
523 act->type = TCA_OLD_COMPAT;
524 exts->action = act;
525 } else if (map->action && tb[map->action]) {
526 act = tcf_action_init(tb[map->action], rate_tlv, NULL,
527 TCA_ACT_NOREPLACE, TCA_ACT_BIND);
528 if (IS_ERR(act))
529 return PTR_ERR(act);
530
531 exts->action = act;
532 }
533 }
534#else
535 if ((map->action && tb[map->action]) ||
536 (map->police && tb[map->police]))
537 return -EOPNOTSUPP;
538#endif
539
540 return 0;
541}
542EXPORT_SYMBOL(tcf_exts_validate);
543
544void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
545 struct tcf_exts *src)
546{
547#ifdef CONFIG_NET_CLS_ACT
548 if (src->action) {
549 struct tc_action *act;
550 tcf_tree_lock(tp);
551 act = dst->action;
552 dst->action = src->action;
553 tcf_tree_unlock(tp);
554 if (act)
555 tcf_action_destroy(act, TCA_ACT_UNBIND);
556 }
557#endif
558}
559EXPORT_SYMBOL(tcf_exts_change);
560
561int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts,
562 const struct tcf_ext_map *map)
563{
564#ifdef CONFIG_NET_CLS_ACT
565 if (map->action && exts->action) {
566 /*
567 * again for backward compatible mode - we want
568 * to work with both old and new modes of entering
569 * tc data even if iproute2 was newer - jhs
570 */
571 struct nlattr *nest;
572
573 if (exts->action->type != TCA_OLD_COMPAT) {
574 nest = nla_nest_start(skb, map->action);
575 if (nest == NULL)
576 goto nla_put_failure;
577 if (tcf_action_dump(skb, exts->action, 0, 0) < 0)
578 goto nla_put_failure;
579 nla_nest_end(skb, nest);
580 } else if (map->police) {
581 nest = nla_nest_start(skb, map->police);
582 if (nest == NULL)
583 goto nla_put_failure;
584 if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0)
585 goto nla_put_failure;
586 nla_nest_end(skb, nest);
587 }
588 }
589#endif
590 return 0;
591nla_put_failure: __attribute__ ((unused))
592 return -1;
593}
594EXPORT_SYMBOL(tcf_exts_dump);
595
596
597int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts,
598 const struct tcf_ext_map *map)
599{
600#ifdef CONFIG_NET_CLS_ACT
601 if (exts->action)
602 if (tcf_action_copy_stats(skb, exts->action, 1) < 0)
603 goto nla_put_failure;
604#endif
605 return 0;
606nla_put_failure: __attribute__ ((unused))
607 return -1;
608}
609EXPORT_SYMBOL(tcf_exts_dump_stats);
610
611static int __init tc_filter_init(void)
612{
613 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL);
614 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL);
615 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
616 tc_dump_tfilter, NULL);
617
618 return 0;
619}
620
621subsys_initcall(tc_filter_init);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * net/sched/cls_api.c Packet classifier API.
4 *
5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6 *
7 * Changes:
8 *
9 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
10 */
11
12#include <linux/module.h>
13#include <linux/types.h>
14#include <linux/kernel.h>
15#include <linux/string.h>
16#include <linux/errno.h>
17#include <linux/err.h>
18#include <linux/skbuff.h>
19#include <linux/init.h>
20#include <linux/kmod.h>
21#include <linux/slab.h>
22#include <linux/idr.h>
23#include <linux/jhash.h>
24#include <linux/rculist.h>
25#include <net/net_namespace.h>
26#include <net/sock.h>
27#include <net/netlink.h>
28#include <net/pkt_sched.h>
29#include <net/pkt_cls.h>
30#include <net/tc_act/tc_pedit.h>
31#include <net/tc_act/tc_mirred.h>
32#include <net/tc_act/tc_vlan.h>
33#include <net/tc_act/tc_tunnel_key.h>
34#include <net/tc_act/tc_csum.h>
35#include <net/tc_act/tc_gact.h>
36#include <net/tc_act/tc_police.h>
37#include <net/tc_act/tc_sample.h>
38#include <net/tc_act/tc_skbedit.h>
39#include <net/tc_act/tc_ct.h>
40#include <net/tc_act/tc_mpls.h>
41#include <net/tc_act/tc_gate.h>
42#include <net/flow_offload.h>
43
44extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
45
46/* The list of all installed classifier types */
47static LIST_HEAD(tcf_proto_base);
48
49/* Protects list of registered TC modules. It is pure SMP lock. */
50static DEFINE_RWLOCK(cls_mod_lock);
51
52static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
53{
54 return jhash_3words(tp->chain->index, tp->prio,
55 (__force __u32)tp->protocol, 0);
56}
57
58static void tcf_proto_signal_destroying(struct tcf_chain *chain,
59 struct tcf_proto *tp)
60{
61 struct tcf_block *block = chain->block;
62
63 mutex_lock(&block->proto_destroy_lock);
64 hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
65 destroy_obj_hashfn(tp));
66 mutex_unlock(&block->proto_destroy_lock);
67}
68
69static bool tcf_proto_cmp(const struct tcf_proto *tp1,
70 const struct tcf_proto *tp2)
71{
72 return tp1->chain->index == tp2->chain->index &&
73 tp1->prio == tp2->prio &&
74 tp1->protocol == tp2->protocol;
75}
76
77static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
78 struct tcf_proto *tp)
79{
80 u32 hash = destroy_obj_hashfn(tp);
81 struct tcf_proto *iter;
82 bool found = false;
83
84 rcu_read_lock();
85 hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
86 destroy_ht_node, hash) {
87 if (tcf_proto_cmp(tp, iter)) {
88 found = true;
89 break;
90 }
91 }
92 rcu_read_unlock();
93
94 return found;
95}
96
97static void
98tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
99{
100 struct tcf_block *block = chain->block;
101
102 mutex_lock(&block->proto_destroy_lock);
103 if (hash_hashed(&tp->destroy_ht_node))
104 hash_del_rcu(&tp->destroy_ht_node);
105 mutex_unlock(&block->proto_destroy_lock);
106}
107
108/* Find classifier type by string name */
109
110static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
111{
112 const struct tcf_proto_ops *t, *res = NULL;
113
114 if (kind) {
115 read_lock(&cls_mod_lock);
116 list_for_each_entry(t, &tcf_proto_base, head) {
117 if (strcmp(kind, t->kind) == 0) {
118 if (try_module_get(t->owner))
119 res = t;
120 break;
121 }
122 }
123 read_unlock(&cls_mod_lock);
124 }
125 return res;
126}
127
128static const struct tcf_proto_ops *
129tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
130 struct netlink_ext_ack *extack)
131{
132 const struct tcf_proto_ops *ops;
133
134 ops = __tcf_proto_lookup_ops(kind);
135 if (ops)
136 return ops;
137#ifdef CONFIG_MODULES
138 if (rtnl_held)
139 rtnl_unlock();
140 request_module("cls_%s", kind);
141 if (rtnl_held)
142 rtnl_lock();
143 ops = __tcf_proto_lookup_ops(kind);
144 /* We dropped the RTNL semaphore in order to perform
145 * the module load. So, even if we succeeded in loading
146 * the module we have to replay the request. We indicate
147 * this using -EAGAIN.
148 */
149 if (ops) {
150 module_put(ops->owner);
151 return ERR_PTR(-EAGAIN);
152 }
153#endif
154 NL_SET_ERR_MSG(extack, "TC classifier not found");
155 return ERR_PTR(-ENOENT);
156}
157
158/* Register(unregister) new classifier type */
159
160int register_tcf_proto_ops(struct tcf_proto_ops *ops)
161{
162 struct tcf_proto_ops *t;
163 int rc = -EEXIST;
164
165 write_lock(&cls_mod_lock);
166 list_for_each_entry(t, &tcf_proto_base, head)
167 if (!strcmp(ops->kind, t->kind))
168 goto out;
169
170 list_add_tail(&ops->head, &tcf_proto_base);
171 rc = 0;
172out:
173 write_unlock(&cls_mod_lock);
174 return rc;
175}
176EXPORT_SYMBOL(register_tcf_proto_ops);
177
178static struct workqueue_struct *tc_filter_wq;
179
180int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
181{
182 struct tcf_proto_ops *t;
183 int rc = -ENOENT;
184
185 /* Wait for outstanding call_rcu()s, if any, from a
186 * tcf_proto_ops's destroy() handler.
187 */
188 rcu_barrier();
189 flush_workqueue(tc_filter_wq);
190
191 write_lock(&cls_mod_lock);
192 list_for_each_entry(t, &tcf_proto_base, head) {
193 if (t == ops) {
194 list_del(&t->head);
195 rc = 0;
196 break;
197 }
198 }
199 write_unlock(&cls_mod_lock);
200 return rc;
201}
202EXPORT_SYMBOL(unregister_tcf_proto_ops);
203
204bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
205{
206 INIT_RCU_WORK(rwork, func);
207 return queue_rcu_work(tc_filter_wq, rwork);
208}
209EXPORT_SYMBOL(tcf_queue_work);
210
211/* Select new prio value from the range, managed by kernel. */
212
213static inline u32 tcf_auto_prio(struct tcf_proto *tp)
214{
215 u32 first = TC_H_MAKE(0xC0000000U, 0U);
216
217 if (tp)
218 first = tp->prio - 1;
219
220 return TC_H_MAJ(first);
221}
222
223static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
224{
225 if (kind)
226 return nla_strscpy(name, kind, IFNAMSIZ) < 0;
227 memset(name, 0, IFNAMSIZ);
228 return false;
229}
230
231static bool tcf_proto_is_unlocked(const char *kind)
232{
233 const struct tcf_proto_ops *ops;
234 bool ret;
235
236 if (strlen(kind) == 0)
237 return false;
238
239 ops = tcf_proto_lookup_ops(kind, false, NULL);
240 /* On error return false to take rtnl lock. Proto lookup/create
241 * functions will perform lookup again and properly handle errors.
242 */
243 if (IS_ERR(ops))
244 return false;
245
246 ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
247 module_put(ops->owner);
248 return ret;
249}
250
251static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
252 u32 prio, struct tcf_chain *chain,
253 bool rtnl_held,
254 struct netlink_ext_ack *extack)
255{
256 struct tcf_proto *tp;
257 int err;
258
259 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
260 if (!tp)
261 return ERR_PTR(-ENOBUFS);
262
263 tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
264 if (IS_ERR(tp->ops)) {
265 err = PTR_ERR(tp->ops);
266 goto errout;
267 }
268 tp->classify = tp->ops->classify;
269 tp->protocol = protocol;
270 tp->prio = prio;
271 tp->chain = chain;
272 spin_lock_init(&tp->lock);
273 refcount_set(&tp->refcnt, 1);
274
275 err = tp->ops->init(tp);
276 if (err) {
277 module_put(tp->ops->owner);
278 goto errout;
279 }
280 return tp;
281
282errout:
283 kfree(tp);
284 return ERR_PTR(err);
285}
286
287static void tcf_proto_get(struct tcf_proto *tp)
288{
289 refcount_inc(&tp->refcnt);
290}
291
292static void tcf_chain_put(struct tcf_chain *chain);
293
294static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
295 bool sig_destroy, struct netlink_ext_ack *extack)
296{
297 tp->ops->destroy(tp, rtnl_held, extack);
298 if (sig_destroy)
299 tcf_proto_signal_destroyed(tp->chain, tp);
300 tcf_chain_put(tp->chain);
301 module_put(tp->ops->owner);
302 kfree_rcu(tp, rcu);
303}
304
305static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
306 struct netlink_ext_ack *extack)
307{
308 if (refcount_dec_and_test(&tp->refcnt))
309 tcf_proto_destroy(tp, rtnl_held, true, extack);
310}
311
312static bool tcf_proto_check_delete(struct tcf_proto *tp)
313{
314 if (tp->ops->delete_empty)
315 return tp->ops->delete_empty(tp);
316
317 tp->deleting = true;
318 return tp->deleting;
319}
320
321static void tcf_proto_mark_delete(struct tcf_proto *tp)
322{
323 spin_lock(&tp->lock);
324 tp->deleting = true;
325 spin_unlock(&tp->lock);
326}
327
328static bool tcf_proto_is_deleting(struct tcf_proto *tp)
329{
330 bool deleting;
331
332 spin_lock(&tp->lock);
333 deleting = tp->deleting;
334 spin_unlock(&tp->lock);
335
336 return deleting;
337}
338
339#define ASSERT_BLOCK_LOCKED(block) \
340 lockdep_assert_held(&(block)->lock)
341
342struct tcf_filter_chain_list_item {
343 struct list_head list;
344 tcf_chain_head_change_t *chain_head_change;
345 void *chain_head_change_priv;
346};
347
348static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
349 u32 chain_index)
350{
351 struct tcf_chain *chain;
352
353 ASSERT_BLOCK_LOCKED(block);
354
355 chain = kzalloc(sizeof(*chain), GFP_KERNEL);
356 if (!chain)
357 return NULL;
358 list_add_tail_rcu(&chain->list, &block->chain_list);
359 mutex_init(&chain->filter_chain_lock);
360 chain->block = block;
361 chain->index = chain_index;
362 chain->refcnt = 1;
363 if (!chain->index)
364 block->chain0.chain = chain;
365 return chain;
366}
367
368static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
369 struct tcf_proto *tp_head)
370{
371 if (item->chain_head_change)
372 item->chain_head_change(tp_head, item->chain_head_change_priv);
373}
374
375static void tcf_chain0_head_change(struct tcf_chain *chain,
376 struct tcf_proto *tp_head)
377{
378 struct tcf_filter_chain_list_item *item;
379 struct tcf_block *block = chain->block;
380
381 if (chain->index)
382 return;
383
384 mutex_lock(&block->lock);
385 list_for_each_entry(item, &block->chain0.filter_chain_list, list)
386 tcf_chain_head_change_item(item, tp_head);
387 mutex_unlock(&block->lock);
388}
389
390/* Returns true if block can be safely freed. */
391
392static bool tcf_chain_detach(struct tcf_chain *chain)
393{
394 struct tcf_block *block = chain->block;
395
396 ASSERT_BLOCK_LOCKED(block);
397
398 list_del_rcu(&chain->list);
399 if (!chain->index)
400 block->chain0.chain = NULL;
401
402 if (list_empty(&block->chain_list) &&
403 refcount_read(&block->refcnt) == 0)
404 return true;
405
406 return false;
407}
408
409static void tcf_block_destroy(struct tcf_block *block)
410{
411 mutex_destroy(&block->lock);
412 mutex_destroy(&block->proto_destroy_lock);
413 kfree_rcu(block, rcu);
414}
415
416static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
417{
418 struct tcf_block *block = chain->block;
419
420 mutex_destroy(&chain->filter_chain_lock);
421 kfree_rcu(chain, rcu);
422 if (free_block)
423 tcf_block_destroy(block);
424}
425
426static void tcf_chain_hold(struct tcf_chain *chain)
427{
428 ASSERT_BLOCK_LOCKED(chain->block);
429
430 ++chain->refcnt;
431}
432
433static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
434{
435 ASSERT_BLOCK_LOCKED(chain->block);
436
437 /* In case all the references are action references, this
438 * chain should not be shown to the user.
439 */
440 return chain->refcnt == chain->action_refcnt;
441}
442
443static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
444 u32 chain_index)
445{
446 struct tcf_chain *chain;
447
448 ASSERT_BLOCK_LOCKED(block);
449
450 list_for_each_entry(chain, &block->chain_list, list) {
451 if (chain->index == chain_index)
452 return chain;
453 }
454 return NULL;
455}
456
457#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
458static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
459 u32 chain_index)
460{
461 struct tcf_chain *chain;
462
463 list_for_each_entry_rcu(chain, &block->chain_list, list) {
464 if (chain->index == chain_index)
465 return chain;
466 }
467 return NULL;
468}
469#endif
470
471static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
472 u32 seq, u16 flags, int event, bool unicast);
473
474static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
475 u32 chain_index, bool create,
476 bool by_act)
477{
478 struct tcf_chain *chain = NULL;
479 bool is_first_reference;
480
481 mutex_lock(&block->lock);
482 chain = tcf_chain_lookup(block, chain_index);
483 if (chain) {
484 tcf_chain_hold(chain);
485 } else {
486 if (!create)
487 goto errout;
488 chain = tcf_chain_create(block, chain_index);
489 if (!chain)
490 goto errout;
491 }
492
493 if (by_act)
494 ++chain->action_refcnt;
495 is_first_reference = chain->refcnt - chain->action_refcnt == 1;
496 mutex_unlock(&block->lock);
497
498 /* Send notification only in case we got the first
499 * non-action reference. Until then, the chain acts only as
500 * a placeholder for actions pointing to it and user ought
501 * not know about them.
502 */
503 if (is_first_reference && !by_act)
504 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
505 RTM_NEWCHAIN, false);
506
507 return chain;
508
509errout:
510 mutex_unlock(&block->lock);
511 return chain;
512}
513
514static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
515 bool create)
516{
517 return __tcf_chain_get(block, chain_index, create, false);
518}
519
520struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
521{
522 return __tcf_chain_get(block, chain_index, true, true);
523}
524EXPORT_SYMBOL(tcf_chain_get_by_act);
525
526static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
527 void *tmplt_priv);
528static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
529 void *tmplt_priv, u32 chain_index,
530 struct tcf_block *block, struct sk_buff *oskb,
531 u32 seq, u16 flags, bool unicast);
532
533static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
534 bool explicitly_created)
535{
536 struct tcf_block *block = chain->block;
537 const struct tcf_proto_ops *tmplt_ops;
538 bool free_block = false;
539 unsigned int refcnt;
540 void *tmplt_priv;
541
542 mutex_lock(&block->lock);
543 if (explicitly_created) {
544 if (!chain->explicitly_created) {
545 mutex_unlock(&block->lock);
546 return;
547 }
548 chain->explicitly_created = false;
549 }
550
551 if (by_act)
552 chain->action_refcnt--;
553
554 /* tc_chain_notify_delete can't be called while holding block lock.
555 * However, when block is unlocked chain can be changed concurrently, so
556 * save these to temporary variables.
557 */
558 refcnt = --chain->refcnt;
559 tmplt_ops = chain->tmplt_ops;
560 tmplt_priv = chain->tmplt_priv;
561
562 /* The last dropped non-action reference will trigger notification. */
563 if (refcnt - chain->action_refcnt == 0 && !by_act) {
564 tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
565 block, NULL, 0, 0, false);
566 /* Last reference to chain, no need to lock. */
567 chain->flushing = false;
568 }
569
570 if (refcnt == 0)
571 free_block = tcf_chain_detach(chain);
572 mutex_unlock(&block->lock);
573
574 if (refcnt == 0) {
575 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
576 tcf_chain_destroy(chain, free_block);
577 }
578}
579
580static void tcf_chain_put(struct tcf_chain *chain)
581{
582 __tcf_chain_put(chain, false, false);
583}
584
585void tcf_chain_put_by_act(struct tcf_chain *chain)
586{
587 __tcf_chain_put(chain, true, false);
588}
589EXPORT_SYMBOL(tcf_chain_put_by_act);
590
591static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
592{
593 __tcf_chain_put(chain, false, true);
594}
595
596static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
597{
598 struct tcf_proto *tp, *tp_next;
599
600 mutex_lock(&chain->filter_chain_lock);
601 tp = tcf_chain_dereference(chain->filter_chain, chain);
602 while (tp) {
603 tp_next = rcu_dereference_protected(tp->next, 1);
604 tcf_proto_signal_destroying(chain, tp);
605 tp = tp_next;
606 }
607 tp = tcf_chain_dereference(chain->filter_chain, chain);
608 RCU_INIT_POINTER(chain->filter_chain, NULL);
609 tcf_chain0_head_change(chain, NULL);
610 chain->flushing = true;
611 mutex_unlock(&chain->filter_chain_lock);
612
613 while (tp) {
614 tp_next = rcu_dereference_protected(tp->next, 1);
615 tcf_proto_put(tp, rtnl_held, NULL);
616 tp = tp_next;
617 }
618}
619
620static int tcf_block_setup(struct tcf_block *block,
621 struct flow_block_offload *bo);
622
623static void tcf_block_offload_init(struct flow_block_offload *bo,
624 struct net_device *dev, struct Qdisc *sch,
625 enum flow_block_command command,
626 enum flow_block_binder_type binder_type,
627 struct flow_block *flow_block,
628 bool shared, struct netlink_ext_ack *extack)
629{
630 bo->net = dev_net(dev);
631 bo->command = command;
632 bo->binder_type = binder_type;
633 bo->block = flow_block;
634 bo->block_shared = shared;
635 bo->extack = extack;
636 bo->sch = sch;
637 bo->cb_list_head = &flow_block->cb_list;
638 INIT_LIST_HEAD(&bo->cb_list);
639}
640
641static void tcf_block_unbind(struct tcf_block *block,
642 struct flow_block_offload *bo);
643
644static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
645{
646 struct tcf_block *block = block_cb->indr.data;
647 struct net_device *dev = block_cb->indr.dev;
648 struct Qdisc *sch = block_cb->indr.sch;
649 struct netlink_ext_ack extack = {};
650 struct flow_block_offload bo = {};
651
652 tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
653 block_cb->indr.binder_type,
654 &block->flow_block, tcf_block_shared(block),
655 &extack);
656 rtnl_lock();
657 down_write(&block->cb_lock);
658 list_del(&block_cb->driver_list);
659 list_move(&block_cb->list, &bo.cb_list);
660 tcf_block_unbind(block, &bo);
661 up_write(&block->cb_lock);
662 rtnl_unlock();
663}
664
665static bool tcf_block_offload_in_use(struct tcf_block *block)
666{
667 return atomic_read(&block->offloadcnt);
668}
669
670static int tcf_block_offload_cmd(struct tcf_block *block,
671 struct net_device *dev, struct Qdisc *sch,
672 struct tcf_block_ext_info *ei,
673 enum flow_block_command command,
674 struct netlink_ext_ack *extack)
675{
676 struct flow_block_offload bo = {};
677
678 tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
679 &block->flow_block, tcf_block_shared(block),
680 extack);
681
682 if (dev->netdev_ops->ndo_setup_tc) {
683 int err;
684
685 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
686 if (err < 0) {
687 if (err != -EOPNOTSUPP)
688 NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
689 return err;
690 }
691
692 return tcf_block_setup(block, &bo);
693 }
694
695 flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
696 tc_block_indr_cleanup);
697 tcf_block_setup(block, &bo);
698
699 return -EOPNOTSUPP;
700}
701
702static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
703 struct tcf_block_ext_info *ei,
704 struct netlink_ext_ack *extack)
705{
706 struct net_device *dev = q->dev_queue->dev;
707 int err;
708
709 down_write(&block->cb_lock);
710
711 /* If tc offload feature is disabled and the block we try to bind
712 * to already has some offloaded filters, forbid to bind.
713 */
714 if (dev->netdev_ops->ndo_setup_tc &&
715 !tc_can_offload(dev) &&
716 tcf_block_offload_in_use(block)) {
717 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
718 err = -EOPNOTSUPP;
719 goto err_unlock;
720 }
721
722 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
723 if (err == -EOPNOTSUPP)
724 goto no_offload_dev_inc;
725 if (err)
726 goto err_unlock;
727
728 up_write(&block->cb_lock);
729 return 0;
730
731no_offload_dev_inc:
732 if (tcf_block_offload_in_use(block))
733 goto err_unlock;
734
735 err = 0;
736 block->nooffloaddevcnt++;
737err_unlock:
738 up_write(&block->cb_lock);
739 return err;
740}
741
742static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
743 struct tcf_block_ext_info *ei)
744{
745 struct net_device *dev = q->dev_queue->dev;
746 int err;
747
748 down_write(&block->cb_lock);
749 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
750 if (err == -EOPNOTSUPP)
751 goto no_offload_dev_dec;
752 up_write(&block->cb_lock);
753 return;
754
755no_offload_dev_dec:
756 WARN_ON(block->nooffloaddevcnt-- == 0);
757 up_write(&block->cb_lock);
758}
759
760static int
761tcf_chain0_head_change_cb_add(struct tcf_block *block,
762 struct tcf_block_ext_info *ei,
763 struct netlink_ext_ack *extack)
764{
765 struct tcf_filter_chain_list_item *item;
766 struct tcf_chain *chain0;
767
768 item = kmalloc(sizeof(*item), GFP_KERNEL);
769 if (!item) {
770 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
771 return -ENOMEM;
772 }
773 item->chain_head_change = ei->chain_head_change;
774 item->chain_head_change_priv = ei->chain_head_change_priv;
775
776 mutex_lock(&block->lock);
777 chain0 = block->chain0.chain;
778 if (chain0)
779 tcf_chain_hold(chain0);
780 else
781 list_add(&item->list, &block->chain0.filter_chain_list);
782 mutex_unlock(&block->lock);
783
784 if (chain0) {
785 struct tcf_proto *tp_head;
786
787 mutex_lock(&chain0->filter_chain_lock);
788
789 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
790 if (tp_head)
791 tcf_chain_head_change_item(item, tp_head);
792
793 mutex_lock(&block->lock);
794 list_add(&item->list, &block->chain0.filter_chain_list);
795 mutex_unlock(&block->lock);
796
797 mutex_unlock(&chain0->filter_chain_lock);
798 tcf_chain_put(chain0);
799 }
800
801 return 0;
802}
803
804static void
805tcf_chain0_head_change_cb_del(struct tcf_block *block,
806 struct tcf_block_ext_info *ei)
807{
808 struct tcf_filter_chain_list_item *item;
809
810 mutex_lock(&block->lock);
811 list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
812 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
813 (item->chain_head_change == ei->chain_head_change &&
814 item->chain_head_change_priv == ei->chain_head_change_priv)) {
815 if (block->chain0.chain)
816 tcf_chain_head_change_item(item, NULL);
817 list_del(&item->list);
818 mutex_unlock(&block->lock);
819
820 kfree(item);
821 return;
822 }
823 }
824 mutex_unlock(&block->lock);
825 WARN_ON(1);
826}
827
828struct tcf_net {
829 spinlock_t idr_lock; /* Protects idr */
830 struct idr idr;
831};
832
833static unsigned int tcf_net_id;
834
835static int tcf_block_insert(struct tcf_block *block, struct net *net,
836 struct netlink_ext_ack *extack)
837{
838 struct tcf_net *tn = net_generic(net, tcf_net_id);
839 int err;
840
841 idr_preload(GFP_KERNEL);
842 spin_lock(&tn->idr_lock);
843 err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
844 GFP_NOWAIT);
845 spin_unlock(&tn->idr_lock);
846 idr_preload_end();
847
848 return err;
849}
850
851static void tcf_block_remove(struct tcf_block *block, struct net *net)
852{
853 struct tcf_net *tn = net_generic(net, tcf_net_id);
854
855 spin_lock(&tn->idr_lock);
856 idr_remove(&tn->idr, block->index);
857 spin_unlock(&tn->idr_lock);
858}
859
860static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
861 u32 block_index,
862 struct netlink_ext_ack *extack)
863{
864 struct tcf_block *block;
865
866 block = kzalloc(sizeof(*block), GFP_KERNEL);
867 if (!block) {
868 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
869 return ERR_PTR(-ENOMEM);
870 }
871 mutex_init(&block->lock);
872 mutex_init(&block->proto_destroy_lock);
873 init_rwsem(&block->cb_lock);
874 flow_block_init(&block->flow_block);
875 INIT_LIST_HEAD(&block->chain_list);
876 INIT_LIST_HEAD(&block->owner_list);
877 INIT_LIST_HEAD(&block->chain0.filter_chain_list);
878
879 refcount_set(&block->refcnt, 1);
880 block->net = net;
881 block->index = block_index;
882
883 /* Don't store q pointer for blocks which are shared */
884 if (!tcf_block_shared(block))
885 block->q = q;
886 return block;
887}
888
889static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
890{
891 struct tcf_net *tn = net_generic(net, tcf_net_id);
892
893 return idr_find(&tn->idr, block_index);
894}
895
896static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
897{
898 struct tcf_block *block;
899
900 rcu_read_lock();
901 block = tcf_block_lookup(net, block_index);
902 if (block && !refcount_inc_not_zero(&block->refcnt))
903 block = NULL;
904 rcu_read_unlock();
905
906 return block;
907}
908
909static struct tcf_chain *
910__tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
911{
912 mutex_lock(&block->lock);
913 if (chain)
914 chain = list_is_last(&chain->list, &block->chain_list) ?
915 NULL : list_next_entry(chain, list);
916 else
917 chain = list_first_entry_or_null(&block->chain_list,
918 struct tcf_chain, list);
919
920 /* skip all action-only chains */
921 while (chain && tcf_chain_held_by_acts_only(chain))
922 chain = list_is_last(&chain->list, &block->chain_list) ?
923 NULL : list_next_entry(chain, list);
924
925 if (chain)
926 tcf_chain_hold(chain);
927 mutex_unlock(&block->lock);
928
929 return chain;
930}
931
932/* Function to be used by all clients that want to iterate over all chains on
933 * block. It properly obtains block->lock and takes reference to chain before
934 * returning it. Users of this function must be tolerant to concurrent chain
935 * insertion/deletion or ensure that no concurrent chain modification is
936 * possible. Note that all netlink dump callbacks cannot guarantee to provide
937 * consistent dump because rtnl lock is released each time skb is filled with
938 * data and sent to user-space.
939 */
940
941struct tcf_chain *
942tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
943{
944 struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
945
946 if (chain)
947 tcf_chain_put(chain);
948
949 return chain_next;
950}
951EXPORT_SYMBOL(tcf_get_next_chain);
952
953static struct tcf_proto *
954__tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
955{
956 u32 prio = 0;
957
958 ASSERT_RTNL();
959 mutex_lock(&chain->filter_chain_lock);
960
961 if (!tp) {
962 tp = tcf_chain_dereference(chain->filter_chain, chain);
963 } else if (tcf_proto_is_deleting(tp)) {
964 /* 'deleting' flag is set and chain->filter_chain_lock was
965 * unlocked, which means next pointer could be invalid. Restart
966 * search.
967 */
968 prio = tp->prio + 1;
969 tp = tcf_chain_dereference(chain->filter_chain, chain);
970
971 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
972 if (!tp->deleting && tp->prio >= prio)
973 break;
974 } else {
975 tp = tcf_chain_dereference(tp->next, chain);
976 }
977
978 if (tp)
979 tcf_proto_get(tp);
980
981 mutex_unlock(&chain->filter_chain_lock);
982
983 return tp;
984}
985
986/* Function to be used by all clients that want to iterate over all tp's on
987 * chain. Users of this function must be tolerant to concurrent tp
988 * insertion/deletion or ensure that no concurrent chain modification is
989 * possible. Note that all netlink dump callbacks cannot guarantee to provide
990 * consistent dump because rtnl lock is released each time skb is filled with
991 * data and sent to user-space.
992 */
993
994struct tcf_proto *
995tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
996{
997 struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
998
999 if (tp)
1000 tcf_proto_put(tp, true, NULL);
1001
1002 return tp_next;
1003}
1004EXPORT_SYMBOL(tcf_get_next_proto);
1005
1006static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1007{
1008 struct tcf_chain *chain;
1009
1010 /* Last reference to block. At this point chains cannot be added or
1011 * removed concurrently.
1012 */
1013 for (chain = tcf_get_next_chain(block, NULL);
1014 chain;
1015 chain = tcf_get_next_chain(block, chain)) {
1016 tcf_chain_put_explicitly_created(chain);
1017 tcf_chain_flush(chain, rtnl_held);
1018 }
1019}
1020
1021/* Lookup Qdisc and increments its reference counter.
1022 * Set parent, if necessary.
1023 */
1024
1025static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1026 u32 *parent, int ifindex, bool rtnl_held,
1027 struct netlink_ext_ack *extack)
1028{
1029 const struct Qdisc_class_ops *cops;
1030 struct net_device *dev;
1031 int err = 0;
1032
1033 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1034 return 0;
1035
1036 rcu_read_lock();
1037
1038 /* Find link */
1039 dev = dev_get_by_index_rcu(net, ifindex);
1040 if (!dev) {
1041 rcu_read_unlock();
1042 return -ENODEV;
1043 }
1044
1045 /* Find qdisc */
1046 if (!*parent) {
1047 *q = dev->qdisc;
1048 *parent = (*q)->handle;
1049 } else {
1050 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1051 if (!*q) {
1052 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1053 err = -EINVAL;
1054 goto errout_rcu;
1055 }
1056 }
1057
1058 *q = qdisc_refcount_inc_nz(*q);
1059 if (!*q) {
1060 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1061 err = -EINVAL;
1062 goto errout_rcu;
1063 }
1064
1065 /* Is it classful? */
1066 cops = (*q)->ops->cl_ops;
1067 if (!cops) {
1068 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1069 err = -EINVAL;
1070 goto errout_qdisc;
1071 }
1072
1073 if (!cops->tcf_block) {
1074 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1075 err = -EOPNOTSUPP;
1076 goto errout_qdisc;
1077 }
1078
1079errout_rcu:
1080 /* At this point we know that qdisc is not noop_qdisc,
1081 * which means that qdisc holds a reference to net_device
1082 * and we hold a reference to qdisc, so it is safe to release
1083 * rcu read lock.
1084 */
1085 rcu_read_unlock();
1086 return err;
1087
1088errout_qdisc:
1089 rcu_read_unlock();
1090
1091 if (rtnl_held)
1092 qdisc_put(*q);
1093 else
1094 qdisc_put_unlocked(*q);
1095 *q = NULL;
1096
1097 return err;
1098}
1099
1100static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1101 int ifindex, struct netlink_ext_ack *extack)
1102{
1103 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1104 return 0;
1105
1106 /* Do we search for filter, attached to class? */
1107 if (TC_H_MIN(parent)) {
1108 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1109
1110 *cl = cops->find(q, parent);
1111 if (*cl == 0) {
1112 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1113 return -ENOENT;
1114 }
1115 }
1116
1117 return 0;
1118}
1119
1120static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1121 unsigned long cl, int ifindex,
1122 u32 block_index,
1123 struct netlink_ext_ack *extack)
1124{
1125 struct tcf_block *block;
1126
1127 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1128 block = tcf_block_refcnt_get(net, block_index);
1129 if (!block) {
1130 NL_SET_ERR_MSG(extack, "Block of given index was not found");
1131 return ERR_PTR(-EINVAL);
1132 }
1133 } else {
1134 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1135
1136 block = cops->tcf_block(q, cl, extack);
1137 if (!block)
1138 return ERR_PTR(-EINVAL);
1139
1140 if (tcf_block_shared(block)) {
1141 NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1142 return ERR_PTR(-EOPNOTSUPP);
1143 }
1144
1145 /* Always take reference to block in order to support execution
1146 * of rules update path of cls API without rtnl lock. Caller
1147 * must release block when it is finished using it. 'if' block
1148 * of this conditional obtain reference to block by calling
1149 * tcf_block_refcnt_get().
1150 */
1151 refcount_inc(&block->refcnt);
1152 }
1153
1154 return block;
1155}
1156
1157static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1158 struct tcf_block_ext_info *ei, bool rtnl_held)
1159{
1160 if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1161 /* Flushing/putting all chains will cause the block to be
1162 * deallocated when last chain is freed. However, if chain_list
1163 * is empty, block has to be manually deallocated. After block
1164 * reference counter reached 0, it is no longer possible to
1165 * increment it or add new chains to block.
1166 */
1167 bool free_block = list_empty(&block->chain_list);
1168
1169 mutex_unlock(&block->lock);
1170 if (tcf_block_shared(block))
1171 tcf_block_remove(block, block->net);
1172
1173 if (q)
1174 tcf_block_offload_unbind(block, q, ei);
1175
1176 if (free_block)
1177 tcf_block_destroy(block);
1178 else
1179 tcf_block_flush_all_chains(block, rtnl_held);
1180 } else if (q) {
1181 tcf_block_offload_unbind(block, q, ei);
1182 }
1183}
1184
1185static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1186{
1187 __tcf_block_put(block, NULL, NULL, rtnl_held);
1188}
1189
1190/* Find tcf block.
1191 * Set q, parent, cl when appropriate.
1192 */
1193
1194static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1195 u32 *parent, unsigned long *cl,
1196 int ifindex, u32 block_index,
1197 struct netlink_ext_ack *extack)
1198{
1199 struct tcf_block *block;
1200 int err = 0;
1201
1202 ASSERT_RTNL();
1203
1204 err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1205 if (err)
1206 goto errout;
1207
1208 err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1209 if (err)
1210 goto errout_qdisc;
1211
1212 block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1213 if (IS_ERR(block)) {
1214 err = PTR_ERR(block);
1215 goto errout_qdisc;
1216 }
1217
1218 return block;
1219
1220errout_qdisc:
1221 if (*q)
1222 qdisc_put(*q);
1223errout:
1224 *q = NULL;
1225 return ERR_PTR(err);
1226}
1227
1228static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1229 bool rtnl_held)
1230{
1231 if (!IS_ERR_OR_NULL(block))
1232 tcf_block_refcnt_put(block, rtnl_held);
1233
1234 if (q) {
1235 if (rtnl_held)
1236 qdisc_put(q);
1237 else
1238 qdisc_put_unlocked(q);
1239 }
1240}
1241
1242struct tcf_block_owner_item {
1243 struct list_head list;
1244 struct Qdisc *q;
1245 enum flow_block_binder_type binder_type;
1246};
1247
1248static void
1249tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1250 struct Qdisc *q,
1251 enum flow_block_binder_type binder_type)
1252{
1253 if (block->keep_dst &&
1254 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1255 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1256 netif_keep_dst(qdisc_dev(q));
1257}
1258
1259void tcf_block_netif_keep_dst(struct tcf_block *block)
1260{
1261 struct tcf_block_owner_item *item;
1262
1263 block->keep_dst = true;
1264 list_for_each_entry(item, &block->owner_list, list)
1265 tcf_block_owner_netif_keep_dst(block, item->q,
1266 item->binder_type);
1267}
1268EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1269
1270static int tcf_block_owner_add(struct tcf_block *block,
1271 struct Qdisc *q,
1272 enum flow_block_binder_type binder_type)
1273{
1274 struct tcf_block_owner_item *item;
1275
1276 item = kmalloc(sizeof(*item), GFP_KERNEL);
1277 if (!item)
1278 return -ENOMEM;
1279 item->q = q;
1280 item->binder_type = binder_type;
1281 list_add(&item->list, &block->owner_list);
1282 return 0;
1283}
1284
1285static void tcf_block_owner_del(struct tcf_block *block,
1286 struct Qdisc *q,
1287 enum flow_block_binder_type binder_type)
1288{
1289 struct tcf_block_owner_item *item;
1290
1291 list_for_each_entry(item, &block->owner_list, list) {
1292 if (item->q == q && item->binder_type == binder_type) {
1293 list_del(&item->list);
1294 kfree(item);
1295 return;
1296 }
1297 }
1298 WARN_ON(1);
1299}
1300
1301int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1302 struct tcf_block_ext_info *ei,
1303 struct netlink_ext_ack *extack)
1304{
1305 struct net *net = qdisc_net(q);
1306 struct tcf_block *block = NULL;
1307 int err;
1308
1309 if (ei->block_index)
1310 /* block_index not 0 means the shared block is requested */
1311 block = tcf_block_refcnt_get(net, ei->block_index);
1312
1313 if (!block) {
1314 block = tcf_block_create(net, q, ei->block_index, extack);
1315 if (IS_ERR(block))
1316 return PTR_ERR(block);
1317 if (tcf_block_shared(block)) {
1318 err = tcf_block_insert(block, net, extack);
1319 if (err)
1320 goto err_block_insert;
1321 }
1322 }
1323
1324 err = tcf_block_owner_add(block, q, ei->binder_type);
1325 if (err)
1326 goto err_block_owner_add;
1327
1328 tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1329
1330 err = tcf_chain0_head_change_cb_add(block, ei, extack);
1331 if (err)
1332 goto err_chain0_head_change_cb_add;
1333
1334 err = tcf_block_offload_bind(block, q, ei, extack);
1335 if (err)
1336 goto err_block_offload_bind;
1337
1338 *p_block = block;
1339 return 0;
1340
1341err_block_offload_bind:
1342 tcf_chain0_head_change_cb_del(block, ei);
1343err_chain0_head_change_cb_add:
1344 tcf_block_owner_del(block, q, ei->binder_type);
1345err_block_owner_add:
1346err_block_insert:
1347 tcf_block_refcnt_put(block, true);
1348 return err;
1349}
1350EXPORT_SYMBOL(tcf_block_get_ext);
1351
1352static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1353{
1354 struct tcf_proto __rcu **p_filter_chain = priv;
1355
1356 rcu_assign_pointer(*p_filter_chain, tp_head);
1357}
1358
1359int tcf_block_get(struct tcf_block **p_block,
1360 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1361 struct netlink_ext_ack *extack)
1362{
1363 struct tcf_block_ext_info ei = {
1364 .chain_head_change = tcf_chain_head_change_dflt,
1365 .chain_head_change_priv = p_filter_chain,
1366 };
1367
1368 WARN_ON(!p_filter_chain);
1369 return tcf_block_get_ext(p_block, q, &ei, extack);
1370}
1371EXPORT_SYMBOL(tcf_block_get);
1372
1373/* XXX: Standalone actions are not allowed to jump to any chain, and bound
1374 * actions should be all removed after flushing.
1375 */
1376void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1377 struct tcf_block_ext_info *ei)
1378{
1379 if (!block)
1380 return;
1381 tcf_chain0_head_change_cb_del(block, ei);
1382 tcf_block_owner_del(block, q, ei->binder_type);
1383
1384 __tcf_block_put(block, q, ei, true);
1385}
1386EXPORT_SYMBOL(tcf_block_put_ext);
1387
1388void tcf_block_put(struct tcf_block *block)
1389{
1390 struct tcf_block_ext_info ei = {0, };
1391
1392 if (!block)
1393 return;
1394 tcf_block_put_ext(block, block->q, &ei);
1395}
1396
1397EXPORT_SYMBOL(tcf_block_put);
1398
1399static int
1400tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1401 void *cb_priv, bool add, bool offload_in_use,
1402 struct netlink_ext_ack *extack)
1403{
1404 struct tcf_chain *chain, *chain_prev;
1405 struct tcf_proto *tp, *tp_prev;
1406 int err;
1407
1408 lockdep_assert_held(&block->cb_lock);
1409
1410 for (chain = __tcf_get_next_chain(block, NULL);
1411 chain;
1412 chain_prev = chain,
1413 chain = __tcf_get_next_chain(block, chain),
1414 tcf_chain_put(chain_prev)) {
1415 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1416 tp_prev = tp,
1417 tp = __tcf_get_next_proto(chain, tp),
1418 tcf_proto_put(tp_prev, true, NULL)) {
1419 if (tp->ops->reoffload) {
1420 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1421 extack);
1422 if (err && add)
1423 goto err_playback_remove;
1424 } else if (add && offload_in_use) {
1425 err = -EOPNOTSUPP;
1426 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1427 goto err_playback_remove;
1428 }
1429 }
1430 }
1431
1432 return 0;
1433
1434err_playback_remove:
1435 tcf_proto_put(tp, true, NULL);
1436 tcf_chain_put(chain);
1437 tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1438 extack);
1439 return err;
1440}
1441
1442static int tcf_block_bind(struct tcf_block *block,
1443 struct flow_block_offload *bo)
1444{
1445 struct flow_block_cb *block_cb, *next;
1446 int err, i = 0;
1447
1448 lockdep_assert_held(&block->cb_lock);
1449
1450 list_for_each_entry(block_cb, &bo->cb_list, list) {
1451 err = tcf_block_playback_offloads(block, block_cb->cb,
1452 block_cb->cb_priv, true,
1453 tcf_block_offload_in_use(block),
1454 bo->extack);
1455 if (err)
1456 goto err_unroll;
1457 if (!bo->unlocked_driver_cb)
1458 block->lockeddevcnt++;
1459
1460 i++;
1461 }
1462 list_splice(&bo->cb_list, &block->flow_block.cb_list);
1463
1464 return 0;
1465
1466err_unroll:
1467 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1468 if (i-- > 0) {
1469 list_del(&block_cb->list);
1470 tcf_block_playback_offloads(block, block_cb->cb,
1471 block_cb->cb_priv, false,
1472 tcf_block_offload_in_use(block),
1473 NULL);
1474 if (!bo->unlocked_driver_cb)
1475 block->lockeddevcnt--;
1476 }
1477 flow_block_cb_free(block_cb);
1478 }
1479
1480 return err;
1481}
1482
1483static void tcf_block_unbind(struct tcf_block *block,
1484 struct flow_block_offload *bo)
1485{
1486 struct flow_block_cb *block_cb, *next;
1487
1488 lockdep_assert_held(&block->cb_lock);
1489
1490 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1491 tcf_block_playback_offloads(block, block_cb->cb,
1492 block_cb->cb_priv, false,
1493 tcf_block_offload_in_use(block),
1494 NULL);
1495 list_del(&block_cb->list);
1496 flow_block_cb_free(block_cb);
1497 if (!bo->unlocked_driver_cb)
1498 block->lockeddevcnt--;
1499 }
1500}
1501
1502static int tcf_block_setup(struct tcf_block *block,
1503 struct flow_block_offload *bo)
1504{
1505 int err;
1506
1507 switch (bo->command) {
1508 case FLOW_BLOCK_BIND:
1509 err = tcf_block_bind(block, bo);
1510 break;
1511 case FLOW_BLOCK_UNBIND:
1512 err = 0;
1513 tcf_block_unbind(block, bo);
1514 break;
1515 default:
1516 WARN_ON_ONCE(1);
1517 err = -EOPNOTSUPP;
1518 }
1519
1520 return err;
1521}
1522
1523/* Main classifier routine: scans classifier chain attached
1524 * to this qdisc, (optionally) tests for protocol and asks
1525 * specific classifiers.
1526 */
1527static inline int __tcf_classify(struct sk_buff *skb,
1528 const struct tcf_proto *tp,
1529 const struct tcf_proto *orig_tp,
1530 struct tcf_result *res,
1531 bool compat_mode,
1532 u32 *last_executed_chain)
1533{
1534#ifdef CONFIG_NET_CLS_ACT
1535 const int max_reclassify_loop = 16;
1536 const struct tcf_proto *first_tp;
1537 int limit = 0;
1538
1539reclassify:
1540#endif
1541 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1542 __be16 protocol = skb_protocol(skb, false);
1543 int err;
1544
1545 if (tp->protocol != protocol &&
1546 tp->protocol != htons(ETH_P_ALL))
1547 continue;
1548
1549 err = tp->classify(skb, tp, res);
1550#ifdef CONFIG_NET_CLS_ACT
1551 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1552 first_tp = orig_tp;
1553 *last_executed_chain = first_tp->chain->index;
1554 goto reset;
1555 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1556 first_tp = res->goto_tp;
1557 *last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1558 goto reset;
1559 }
1560#endif
1561 if (err >= 0)
1562 return err;
1563 }
1564
1565 return TC_ACT_UNSPEC; /* signal: continue lookup */
1566#ifdef CONFIG_NET_CLS_ACT
1567reset:
1568 if (unlikely(limit++ >= max_reclassify_loop)) {
1569 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1570 tp->chain->block->index,
1571 tp->prio & 0xffff,
1572 ntohs(tp->protocol));
1573 return TC_ACT_SHOT;
1574 }
1575
1576 tp = first_tp;
1577 goto reclassify;
1578#endif
1579}
1580
1581int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1582 struct tcf_result *res, bool compat_mode)
1583{
1584 u32 last_executed_chain = 0;
1585
1586 return __tcf_classify(skb, tp, tp, res, compat_mode,
1587 &last_executed_chain);
1588}
1589EXPORT_SYMBOL(tcf_classify);
1590
1591int tcf_classify_ingress(struct sk_buff *skb,
1592 const struct tcf_block *ingress_block,
1593 const struct tcf_proto *tp,
1594 struct tcf_result *res, bool compat_mode)
1595{
1596#if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1597 u32 last_executed_chain = 0;
1598
1599 return __tcf_classify(skb, tp, tp, res, compat_mode,
1600 &last_executed_chain);
1601#else
1602 u32 last_executed_chain = tp ? tp->chain->index : 0;
1603 const struct tcf_proto *orig_tp = tp;
1604 struct tc_skb_ext *ext;
1605 int ret;
1606
1607 ext = skb_ext_find(skb, TC_SKB_EXT);
1608
1609 if (ext && ext->chain) {
1610 struct tcf_chain *fchain;
1611
1612 fchain = tcf_chain_lookup_rcu(ingress_block, ext->chain);
1613 if (!fchain)
1614 return TC_ACT_SHOT;
1615
1616 /* Consume, so cloned/redirect skbs won't inherit ext */
1617 skb_ext_del(skb, TC_SKB_EXT);
1618
1619 tp = rcu_dereference_bh(fchain->filter_chain);
1620 last_executed_chain = fchain->index;
1621 }
1622
1623 ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode,
1624 &last_executed_chain);
1625
1626 /* If we missed on some chain */
1627 if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1628 ext = tc_skb_ext_alloc(skb);
1629 if (WARN_ON_ONCE(!ext))
1630 return TC_ACT_SHOT;
1631 ext->chain = last_executed_chain;
1632 ext->mru = qdisc_skb_cb(skb)->mru;
1633 ext->post_ct = qdisc_skb_cb(skb)->post_ct;
1634 }
1635
1636 return ret;
1637#endif
1638}
1639EXPORT_SYMBOL(tcf_classify_ingress);
1640
1641struct tcf_chain_info {
1642 struct tcf_proto __rcu **pprev;
1643 struct tcf_proto __rcu *next;
1644};
1645
1646static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1647 struct tcf_chain_info *chain_info)
1648{
1649 return tcf_chain_dereference(*chain_info->pprev, chain);
1650}
1651
1652static int tcf_chain_tp_insert(struct tcf_chain *chain,
1653 struct tcf_chain_info *chain_info,
1654 struct tcf_proto *tp)
1655{
1656 if (chain->flushing)
1657 return -EAGAIN;
1658
1659 if (*chain_info->pprev == chain->filter_chain)
1660 tcf_chain0_head_change(chain, tp);
1661 tcf_proto_get(tp);
1662 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1663 rcu_assign_pointer(*chain_info->pprev, tp);
1664
1665 return 0;
1666}
1667
1668static void tcf_chain_tp_remove(struct tcf_chain *chain,
1669 struct tcf_chain_info *chain_info,
1670 struct tcf_proto *tp)
1671{
1672 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1673
1674 tcf_proto_mark_delete(tp);
1675 if (tp == chain->filter_chain)
1676 tcf_chain0_head_change(chain, next);
1677 RCU_INIT_POINTER(*chain_info->pprev, next);
1678}
1679
1680static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1681 struct tcf_chain_info *chain_info,
1682 u32 protocol, u32 prio,
1683 bool prio_allocate);
1684
1685/* Try to insert new proto.
1686 * If proto with specified priority already exists, free new proto
1687 * and return existing one.
1688 */
1689
1690static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1691 struct tcf_proto *tp_new,
1692 u32 protocol, u32 prio,
1693 bool rtnl_held)
1694{
1695 struct tcf_chain_info chain_info;
1696 struct tcf_proto *tp;
1697 int err = 0;
1698
1699 mutex_lock(&chain->filter_chain_lock);
1700
1701 if (tcf_proto_exists_destroying(chain, tp_new)) {
1702 mutex_unlock(&chain->filter_chain_lock);
1703 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1704 return ERR_PTR(-EAGAIN);
1705 }
1706
1707 tp = tcf_chain_tp_find(chain, &chain_info,
1708 protocol, prio, false);
1709 if (!tp)
1710 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1711 mutex_unlock(&chain->filter_chain_lock);
1712
1713 if (tp) {
1714 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1715 tp_new = tp;
1716 } else if (err) {
1717 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1718 tp_new = ERR_PTR(err);
1719 }
1720
1721 return tp_new;
1722}
1723
1724static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1725 struct tcf_proto *tp, bool rtnl_held,
1726 struct netlink_ext_ack *extack)
1727{
1728 struct tcf_chain_info chain_info;
1729 struct tcf_proto *tp_iter;
1730 struct tcf_proto **pprev;
1731 struct tcf_proto *next;
1732
1733 mutex_lock(&chain->filter_chain_lock);
1734
1735 /* Atomically find and remove tp from chain. */
1736 for (pprev = &chain->filter_chain;
1737 (tp_iter = tcf_chain_dereference(*pprev, chain));
1738 pprev = &tp_iter->next) {
1739 if (tp_iter == tp) {
1740 chain_info.pprev = pprev;
1741 chain_info.next = tp_iter->next;
1742 WARN_ON(tp_iter->deleting);
1743 break;
1744 }
1745 }
1746 /* Verify that tp still exists and no new filters were inserted
1747 * concurrently.
1748 * Mark tp for deletion if it is empty.
1749 */
1750 if (!tp_iter || !tcf_proto_check_delete(tp)) {
1751 mutex_unlock(&chain->filter_chain_lock);
1752 return;
1753 }
1754
1755 tcf_proto_signal_destroying(chain, tp);
1756 next = tcf_chain_dereference(chain_info.next, chain);
1757 if (tp == chain->filter_chain)
1758 tcf_chain0_head_change(chain, next);
1759 RCU_INIT_POINTER(*chain_info.pprev, next);
1760 mutex_unlock(&chain->filter_chain_lock);
1761
1762 tcf_proto_put(tp, rtnl_held, extack);
1763}
1764
1765static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1766 struct tcf_chain_info *chain_info,
1767 u32 protocol, u32 prio,
1768 bool prio_allocate)
1769{
1770 struct tcf_proto **pprev;
1771 struct tcf_proto *tp;
1772
1773 /* Check the chain for existence of proto-tcf with this priority */
1774 for (pprev = &chain->filter_chain;
1775 (tp = tcf_chain_dereference(*pprev, chain));
1776 pprev = &tp->next) {
1777 if (tp->prio >= prio) {
1778 if (tp->prio == prio) {
1779 if (prio_allocate ||
1780 (tp->protocol != protocol && protocol))
1781 return ERR_PTR(-EINVAL);
1782 } else {
1783 tp = NULL;
1784 }
1785 break;
1786 }
1787 }
1788 chain_info->pprev = pprev;
1789 if (tp) {
1790 chain_info->next = tp->next;
1791 tcf_proto_get(tp);
1792 } else {
1793 chain_info->next = NULL;
1794 }
1795 return tp;
1796}
1797
1798static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1799 struct tcf_proto *tp, struct tcf_block *block,
1800 struct Qdisc *q, u32 parent, void *fh,
1801 u32 portid, u32 seq, u16 flags, int event,
1802 bool terse_dump, bool rtnl_held)
1803{
1804 struct tcmsg *tcm;
1805 struct nlmsghdr *nlh;
1806 unsigned char *b = skb_tail_pointer(skb);
1807
1808 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1809 if (!nlh)
1810 goto out_nlmsg_trim;
1811 tcm = nlmsg_data(nlh);
1812 tcm->tcm_family = AF_UNSPEC;
1813 tcm->tcm__pad1 = 0;
1814 tcm->tcm__pad2 = 0;
1815 if (q) {
1816 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1817 tcm->tcm_parent = parent;
1818 } else {
1819 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1820 tcm->tcm_block_index = block->index;
1821 }
1822 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1823 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1824 goto nla_put_failure;
1825 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1826 goto nla_put_failure;
1827 if (!fh) {
1828 tcm->tcm_handle = 0;
1829 } else if (terse_dump) {
1830 if (tp->ops->terse_dump) {
1831 if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
1832 rtnl_held) < 0)
1833 goto nla_put_failure;
1834 } else {
1835 goto cls_op_not_supp;
1836 }
1837 } else {
1838 if (tp->ops->dump &&
1839 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
1840 goto nla_put_failure;
1841 }
1842 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1843 return skb->len;
1844
1845out_nlmsg_trim:
1846nla_put_failure:
1847cls_op_not_supp:
1848 nlmsg_trim(skb, b);
1849 return -1;
1850}
1851
1852static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1853 struct nlmsghdr *n, struct tcf_proto *tp,
1854 struct tcf_block *block, struct Qdisc *q,
1855 u32 parent, void *fh, int event, bool unicast,
1856 bool rtnl_held)
1857{
1858 struct sk_buff *skb;
1859 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1860 int err = 0;
1861
1862 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1863 if (!skb)
1864 return -ENOBUFS;
1865
1866 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1867 n->nlmsg_seq, n->nlmsg_flags, event,
1868 false, rtnl_held) <= 0) {
1869 kfree_skb(skb);
1870 return -EINVAL;
1871 }
1872
1873 if (unicast)
1874 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1875 else
1876 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1877 n->nlmsg_flags & NLM_F_ECHO);
1878
1879 if (err > 0)
1880 err = 0;
1881 return err;
1882}
1883
1884static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1885 struct nlmsghdr *n, struct tcf_proto *tp,
1886 struct tcf_block *block, struct Qdisc *q,
1887 u32 parent, void *fh, bool unicast, bool *last,
1888 bool rtnl_held, struct netlink_ext_ack *extack)
1889{
1890 struct sk_buff *skb;
1891 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1892 int err;
1893
1894 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1895 if (!skb)
1896 return -ENOBUFS;
1897
1898 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1899 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
1900 false, rtnl_held) <= 0) {
1901 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1902 kfree_skb(skb);
1903 return -EINVAL;
1904 }
1905
1906 err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
1907 if (err) {
1908 kfree_skb(skb);
1909 return err;
1910 }
1911
1912 if (unicast)
1913 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1914 else
1915 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1916 n->nlmsg_flags & NLM_F_ECHO);
1917 if (err < 0)
1918 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1919
1920 if (err > 0)
1921 err = 0;
1922 return err;
1923}
1924
1925static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1926 struct tcf_block *block, struct Qdisc *q,
1927 u32 parent, struct nlmsghdr *n,
1928 struct tcf_chain *chain, int event)
1929{
1930 struct tcf_proto *tp;
1931
1932 for (tp = tcf_get_next_proto(chain, NULL);
1933 tp; tp = tcf_get_next_proto(chain, tp))
1934 tfilter_notify(net, oskb, n, tp, block,
1935 q, parent, NULL, event, false, true);
1936}
1937
1938static void tfilter_put(struct tcf_proto *tp, void *fh)
1939{
1940 if (tp->ops->put && fh)
1941 tp->ops->put(tp, fh);
1942}
1943
1944static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1945 struct netlink_ext_ack *extack)
1946{
1947 struct net *net = sock_net(skb->sk);
1948 struct nlattr *tca[TCA_MAX + 1];
1949 char name[IFNAMSIZ];
1950 struct tcmsg *t;
1951 u32 protocol;
1952 u32 prio;
1953 bool prio_allocate;
1954 u32 parent;
1955 u32 chain_index;
1956 struct Qdisc *q = NULL;
1957 struct tcf_chain_info chain_info;
1958 struct tcf_chain *chain = NULL;
1959 struct tcf_block *block;
1960 struct tcf_proto *tp;
1961 unsigned long cl;
1962 void *fh;
1963 int err;
1964 int tp_created;
1965 bool rtnl_held = false;
1966
1967 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1968 return -EPERM;
1969
1970replay:
1971 tp_created = 0;
1972
1973 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
1974 rtm_tca_policy, extack);
1975 if (err < 0)
1976 return err;
1977
1978 t = nlmsg_data(n);
1979 protocol = TC_H_MIN(t->tcm_info);
1980 prio = TC_H_MAJ(t->tcm_info);
1981 prio_allocate = false;
1982 parent = t->tcm_parent;
1983 tp = NULL;
1984 cl = 0;
1985 block = NULL;
1986
1987 if (prio == 0) {
1988 /* If no priority is provided by the user,
1989 * we allocate one.
1990 */
1991 if (n->nlmsg_flags & NLM_F_CREATE) {
1992 prio = TC_H_MAKE(0x80000000U, 0U);
1993 prio_allocate = true;
1994 } else {
1995 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1996 return -ENOENT;
1997 }
1998 }
1999
2000 /* Find head of filter chain. */
2001
2002 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2003 if (err)
2004 return err;
2005
2006 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2007 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2008 err = -EINVAL;
2009 goto errout;
2010 }
2011
2012 /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2013 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2014 * type is not specified, classifier is not unlocked.
2015 */
2016 if (rtnl_held ||
2017 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2018 !tcf_proto_is_unlocked(name)) {
2019 rtnl_held = true;
2020 rtnl_lock();
2021 }
2022
2023 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2024 if (err)
2025 goto errout;
2026
2027 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2028 extack);
2029 if (IS_ERR(block)) {
2030 err = PTR_ERR(block);
2031 goto errout;
2032 }
2033 block->classid = parent;
2034
2035 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2036 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2037 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2038 err = -EINVAL;
2039 goto errout;
2040 }
2041 chain = tcf_chain_get(block, chain_index, true);
2042 if (!chain) {
2043 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2044 err = -ENOMEM;
2045 goto errout;
2046 }
2047
2048 mutex_lock(&chain->filter_chain_lock);
2049 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2050 prio, prio_allocate);
2051 if (IS_ERR(tp)) {
2052 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2053 err = PTR_ERR(tp);
2054 goto errout_locked;
2055 }
2056
2057 if (tp == NULL) {
2058 struct tcf_proto *tp_new = NULL;
2059
2060 if (chain->flushing) {
2061 err = -EAGAIN;
2062 goto errout_locked;
2063 }
2064
2065 /* Proto-tcf does not exist, create new one */
2066
2067 if (tca[TCA_KIND] == NULL || !protocol) {
2068 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2069 err = -EINVAL;
2070 goto errout_locked;
2071 }
2072
2073 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2074 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2075 err = -ENOENT;
2076 goto errout_locked;
2077 }
2078
2079 if (prio_allocate)
2080 prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2081 &chain_info));
2082
2083 mutex_unlock(&chain->filter_chain_lock);
2084 tp_new = tcf_proto_create(name, protocol, prio, chain,
2085 rtnl_held, extack);
2086 if (IS_ERR(tp_new)) {
2087 err = PTR_ERR(tp_new);
2088 goto errout_tp;
2089 }
2090
2091 tp_created = 1;
2092 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2093 rtnl_held);
2094 if (IS_ERR(tp)) {
2095 err = PTR_ERR(tp);
2096 goto errout_tp;
2097 }
2098 } else {
2099 mutex_unlock(&chain->filter_chain_lock);
2100 }
2101
2102 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2103 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2104 err = -EINVAL;
2105 goto errout;
2106 }
2107
2108 fh = tp->ops->get(tp, t->tcm_handle);
2109
2110 if (!fh) {
2111 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2112 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2113 err = -ENOENT;
2114 goto errout;
2115 }
2116 } else if (n->nlmsg_flags & NLM_F_EXCL) {
2117 tfilter_put(tp, fh);
2118 NL_SET_ERR_MSG(extack, "Filter already exists");
2119 err = -EEXIST;
2120 goto errout;
2121 }
2122
2123 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2124 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2125 err = -EINVAL;
2126 goto errout;
2127 }
2128
2129 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2130 n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
2131 rtnl_held, extack);
2132 if (err == 0) {
2133 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2134 RTM_NEWTFILTER, false, rtnl_held);
2135 tfilter_put(tp, fh);
2136 /* q pointer is NULL for shared blocks */
2137 if (q)
2138 q->flags &= ~TCQ_F_CAN_BYPASS;
2139 }
2140
2141errout:
2142 if (err && tp_created)
2143 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2144errout_tp:
2145 if (chain) {
2146 if (tp && !IS_ERR(tp))
2147 tcf_proto_put(tp, rtnl_held, NULL);
2148 if (!tp_created)
2149 tcf_chain_put(chain);
2150 }
2151 tcf_block_release(q, block, rtnl_held);
2152
2153 if (rtnl_held)
2154 rtnl_unlock();
2155
2156 if (err == -EAGAIN) {
2157 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2158 * of target chain.
2159 */
2160 rtnl_held = true;
2161 /* Replay the request. */
2162 goto replay;
2163 }
2164 return err;
2165
2166errout_locked:
2167 mutex_unlock(&chain->filter_chain_lock);
2168 goto errout;
2169}
2170
2171static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2172 struct netlink_ext_ack *extack)
2173{
2174 struct net *net = sock_net(skb->sk);
2175 struct nlattr *tca[TCA_MAX + 1];
2176 char name[IFNAMSIZ];
2177 struct tcmsg *t;
2178 u32 protocol;
2179 u32 prio;
2180 u32 parent;
2181 u32 chain_index;
2182 struct Qdisc *q = NULL;
2183 struct tcf_chain_info chain_info;
2184 struct tcf_chain *chain = NULL;
2185 struct tcf_block *block = NULL;
2186 struct tcf_proto *tp = NULL;
2187 unsigned long cl = 0;
2188 void *fh = NULL;
2189 int err;
2190 bool rtnl_held = false;
2191
2192 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2193 return -EPERM;
2194
2195 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2196 rtm_tca_policy, extack);
2197 if (err < 0)
2198 return err;
2199
2200 t = nlmsg_data(n);
2201 protocol = TC_H_MIN(t->tcm_info);
2202 prio = TC_H_MAJ(t->tcm_info);
2203 parent = t->tcm_parent;
2204
2205 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2206 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2207 return -ENOENT;
2208 }
2209
2210 /* Find head of filter chain. */
2211
2212 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2213 if (err)
2214 return err;
2215
2216 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2217 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2218 err = -EINVAL;
2219 goto errout;
2220 }
2221 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2222 * found), qdisc is not unlocked, classifier type is not specified,
2223 * classifier is not unlocked.
2224 */
2225 if (!prio ||
2226 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2227 !tcf_proto_is_unlocked(name)) {
2228 rtnl_held = true;
2229 rtnl_lock();
2230 }
2231
2232 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2233 if (err)
2234 goto errout;
2235
2236 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2237 extack);
2238 if (IS_ERR(block)) {
2239 err = PTR_ERR(block);
2240 goto errout;
2241 }
2242
2243 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2244 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2245 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2246 err = -EINVAL;
2247 goto errout;
2248 }
2249 chain = tcf_chain_get(block, chain_index, false);
2250 if (!chain) {
2251 /* User requested flush on non-existent chain. Nothing to do,
2252 * so just return success.
2253 */
2254 if (prio == 0) {
2255 err = 0;
2256 goto errout;
2257 }
2258 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2259 err = -ENOENT;
2260 goto errout;
2261 }
2262
2263 if (prio == 0) {
2264 tfilter_notify_chain(net, skb, block, q, parent, n,
2265 chain, RTM_DELTFILTER);
2266 tcf_chain_flush(chain, rtnl_held);
2267 err = 0;
2268 goto errout;
2269 }
2270
2271 mutex_lock(&chain->filter_chain_lock);
2272 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2273 prio, false);
2274 if (!tp || IS_ERR(tp)) {
2275 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2276 err = tp ? PTR_ERR(tp) : -ENOENT;
2277 goto errout_locked;
2278 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2279 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2280 err = -EINVAL;
2281 goto errout_locked;
2282 } else if (t->tcm_handle == 0) {
2283 tcf_proto_signal_destroying(chain, tp);
2284 tcf_chain_tp_remove(chain, &chain_info, tp);
2285 mutex_unlock(&chain->filter_chain_lock);
2286
2287 tcf_proto_put(tp, rtnl_held, NULL);
2288 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2289 RTM_DELTFILTER, false, rtnl_held);
2290 err = 0;
2291 goto errout;
2292 }
2293 mutex_unlock(&chain->filter_chain_lock);
2294
2295 fh = tp->ops->get(tp, t->tcm_handle);
2296
2297 if (!fh) {
2298 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2299 err = -ENOENT;
2300 } else {
2301 bool last;
2302
2303 err = tfilter_del_notify(net, skb, n, tp, block,
2304 q, parent, fh, false, &last,
2305 rtnl_held, extack);
2306
2307 if (err)
2308 goto errout;
2309 if (last)
2310 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2311 }
2312
2313errout:
2314 if (chain) {
2315 if (tp && !IS_ERR(tp))
2316 tcf_proto_put(tp, rtnl_held, NULL);
2317 tcf_chain_put(chain);
2318 }
2319 tcf_block_release(q, block, rtnl_held);
2320
2321 if (rtnl_held)
2322 rtnl_unlock();
2323
2324 return err;
2325
2326errout_locked:
2327 mutex_unlock(&chain->filter_chain_lock);
2328 goto errout;
2329}
2330
2331static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2332 struct netlink_ext_ack *extack)
2333{
2334 struct net *net = sock_net(skb->sk);
2335 struct nlattr *tca[TCA_MAX + 1];
2336 char name[IFNAMSIZ];
2337 struct tcmsg *t;
2338 u32 protocol;
2339 u32 prio;
2340 u32 parent;
2341 u32 chain_index;
2342 struct Qdisc *q = NULL;
2343 struct tcf_chain_info chain_info;
2344 struct tcf_chain *chain = NULL;
2345 struct tcf_block *block = NULL;
2346 struct tcf_proto *tp = NULL;
2347 unsigned long cl = 0;
2348 void *fh = NULL;
2349 int err;
2350 bool rtnl_held = false;
2351
2352 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2353 rtm_tca_policy, extack);
2354 if (err < 0)
2355 return err;
2356
2357 t = nlmsg_data(n);
2358 protocol = TC_H_MIN(t->tcm_info);
2359 prio = TC_H_MAJ(t->tcm_info);
2360 parent = t->tcm_parent;
2361
2362 if (prio == 0) {
2363 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2364 return -ENOENT;
2365 }
2366
2367 /* Find head of filter chain. */
2368
2369 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2370 if (err)
2371 return err;
2372
2373 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2374 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2375 err = -EINVAL;
2376 goto errout;
2377 }
2378 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2379 * unlocked, classifier type is not specified, classifier is not
2380 * unlocked.
2381 */
2382 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2383 !tcf_proto_is_unlocked(name)) {
2384 rtnl_held = true;
2385 rtnl_lock();
2386 }
2387
2388 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2389 if (err)
2390 goto errout;
2391
2392 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2393 extack);
2394 if (IS_ERR(block)) {
2395 err = PTR_ERR(block);
2396 goto errout;
2397 }
2398
2399 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2400 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2401 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2402 err = -EINVAL;
2403 goto errout;
2404 }
2405 chain = tcf_chain_get(block, chain_index, false);
2406 if (!chain) {
2407 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2408 err = -EINVAL;
2409 goto errout;
2410 }
2411
2412 mutex_lock(&chain->filter_chain_lock);
2413 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2414 prio, false);
2415 mutex_unlock(&chain->filter_chain_lock);
2416 if (!tp || IS_ERR(tp)) {
2417 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2418 err = tp ? PTR_ERR(tp) : -ENOENT;
2419 goto errout;
2420 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2421 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2422 err = -EINVAL;
2423 goto errout;
2424 }
2425
2426 fh = tp->ops->get(tp, t->tcm_handle);
2427
2428 if (!fh) {
2429 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2430 err = -ENOENT;
2431 } else {
2432 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2433 fh, RTM_NEWTFILTER, true, rtnl_held);
2434 if (err < 0)
2435 NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2436 }
2437
2438 tfilter_put(tp, fh);
2439errout:
2440 if (chain) {
2441 if (tp && !IS_ERR(tp))
2442 tcf_proto_put(tp, rtnl_held, NULL);
2443 tcf_chain_put(chain);
2444 }
2445 tcf_block_release(q, block, rtnl_held);
2446
2447 if (rtnl_held)
2448 rtnl_unlock();
2449
2450 return err;
2451}
2452
2453struct tcf_dump_args {
2454 struct tcf_walker w;
2455 struct sk_buff *skb;
2456 struct netlink_callback *cb;
2457 struct tcf_block *block;
2458 struct Qdisc *q;
2459 u32 parent;
2460 bool terse_dump;
2461};
2462
2463static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2464{
2465 struct tcf_dump_args *a = (void *)arg;
2466 struct net *net = sock_net(a->skb->sk);
2467
2468 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2469 n, NETLINK_CB(a->cb->skb).portid,
2470 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2471 RTM_NEWTFILTER, a->terse_dump, true);
2472}
2473
2474static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2475 struct sk_buff *skb, struct netlink_callback *cb,
2476 long index_start, long *p_index, bool terse)
2477{
2478 struct net *net = sock_net(skb->sk);
2479 struct tcf_block *block = chain->block;
2480 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2481 struct tcf_proto *tp, *tp_prev;
2482 struct tcf_dump_args arg;
2483
2484 for (tp = __tcf_get_next_proto(chain, NULL);
2485 tp;
2486 tp_prev = tp,
2487 tp = __tcf_get_next_proto(chain, tp),
2488 tcf_proto_put(tp_prev, true, NULL),
2489 (*p_index)++) {
2490 if (*p_index < index_start)
2491 continue;
2492 if (TC_H_MAJ(tcm->tcm_info) &&
2493 TC_H_MAJ(tcm->tcm_info) != tp->prio)
2494 continue;
2495 if (TC_H_MIN(tcm->tcm_info) &&
2496 TC_H_MIN(tcm->tcm_info) != tp->protocol)
2497 continue;
2498 if (*p_index > index_start)
2499 memset(&cb->args[1], 0,
2500 sizeof(cb->args) - sizeof(cb->args[0]));
2501 if (cb->args[1] == 0) {
2502 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2503 NETLINK_CB(cb->skb).portid,
2504 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2505 RTM_NEWTFILTER, false, true) <= 0)
2506 goto errout;
2507 cb->args[1] = 1;
2508 }
2509 if (!tp->ops->walk)
2510 continue;
2511 arg.w.fn = tcf_node_dump;
2512 arg.skb = skb;
2513 arg.cb = cb;
2514 arg.block = block;
2515 arg.q = q;
2516 arg.parent = parent;
2517 arg.w.stop = 0;
2518 arg.w.skip = cb->args[1] - 1;
2519 arg.w.count = 0;
2520 arg.w.cookie = cb->args[2];
2521 arg.terse_dump = terse;
2522 tp->ops->walk(tp, &arg.w, true);
2523 cb->args[2] = arg.w.cookie;
2524 cb->args[1] = arg.w.count + 1;
2525 if (arg.w.stop)
2526 goto errout;
2527 }
2528 return true;
2529
2530errout:
2531 tcf_proto_put(tp, true, NULL);
2532 return false;
2533}
2534
2535static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2536 [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2537};
2538
2539/* called with RTNL */
2540static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2541{
2542 struct tcf_chain *chain, *chain_prev;
2543 struct net *net = sock_net(skb->sk);
2544 struct nlattr *tca[TCA_MAX + 1];
2545 struct Qdisc *q = NULL;
2546 struct tcf_block *block;
2547 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2548 bool terse_dump = false;
2549 long index_start;
2550 long index;
2551 u32 parent;
2552 int err;
2553
2554 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2555 return skb->len;
2556
2557 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2558 tcf_tfilter_dump_policy, cb->extack);
2559 if (err)
2560 return err;
2561
2562 if (tca[TCA_DUMP_FLAGS]) {
2563 struct nla_bitfield32 flags =
2564 nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2565
2566 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2567 }
2568
2569 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2570 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2571 if (!block)
2572 goto out;
2573 /* If we work with block index, q is NULL and parent value
2574 * will never be used in the following code. The check
2575 * in tcf_fill_node prevents it. However, compiler does not
2576 * see that far, so set parent to zero to silence the warning
2577 * about parent being uninitialized.
2578 */
2579 parent = 0;
2580 } else {
2581 const struct Qdisc_class_ops *cops;
2582 struct net_device *dev;
2583 unsigned long cl = 0;
2584
2585 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2586 if (!dev)
2587 return skb->len;
2588
2589 parent = tcm->tcm_parent;
2590 if (!parent)
2591 q = dev->qdisc;
2592 else
2593 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2594 if (!q)
2595 goto out;
2596 cops = q->ops->cl_ops;
2597 if (!cops)
2598 goto out;
2599 if (!cops->tcf_block)
2600 goto out;
2601 if (TC_H_MIN(tcm->tcm_parent)) {
2602 cl = cops->find(q, tcm->tcm_parent);
2603 if (cl == 0)
2604 goto out;
2605 }
2606 block = cops->tcf_block(q, cl, NULL);
2607 if (!block)
2608 goto out;
2609 parent = block->classid;
2610 if (tcf_block_shared(block))
2611 q = NULL;
2612 }
2613
2614 index_start = cb->args[0];
2615 index = 0;
2616
2617 for (chain = __tcf_get_next_chain(block, NULL);
2618 chain;
2619 chain_prev = chain,
2620 chain = __tcf_get_next_chain(block, chain),
2621 tcf_chain_put(chain_prev)) {
2622 if (tca[TCA_CHAIN] &&
2623 nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2624 continue;
2625 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2626 index_start, &index, terse_dump)) {
2627 tcf_chain_put(chain);
2628 err = -EMSGSIZE;
2629 break;
2630 }
2631 }
2632
2633 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2634 tcf_block_refcnt_put(block, true);
2635 cb->args[0] = index;
2636
2637out:
2638 /* If we did no progress, the error (EMSGSIZE) is real */
2639 if (skb->len == 0 && err)
2640 return err;
2641 return skb->len;
2642}
2643
2644static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2645 void *tmplt_priv, u32 chain_index,
2646 struct net *net, struct sk_buff *skb,
2647 struct tcf_block *block,
2648 u32 portid, u32 seq, u16 flags, int event)
2649{
2650 unsigned char *b = skb_tail_pointer(skb);
2651 const struct tcf_proto_ops *ops;
2652 struct nlmsghdr *nlh;
2653 struct tcmsg *tcm;
2654 void *priv;
2655
2656 ops = tmplt_ops;
2657 priv = tmplt_priv;
2658
2659 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2660 if (!nlh)
2661 goto out_nlmsg_trim;
2662 tcm = nlmsg_data(nlh);
2663 tcm->tcm_family = AF_UNSPEC;
2664 tcm->tcm__pad1 = 0;
2665 tcm->tcm__pad2 = 0;
2666 tcm->tcm_handle = 0;
2667 if (block->q) {
2668 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2669 tcm->tcm_parent = block->q->handle;
2670 } else {
2671 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2672 tcm->tcm_block_index = block->index;
2673 }
2674
2675 if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2676 goto nla_put_failure;
2677
2678 if (ops) {
2679 if (nla_put_string(skb, TCA_KIND, ops->kind))
2680 goto nla_put_failure;
2681 if (ops->tmplt_dump(skb, net, priv) < 0)
2682 goto nla_put_failure;
2683 }
2684
2685 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2686 return skb->len;
2687
2688out_nlmsg_trim:
2689nla_put_failure:
2690 nlmsg_trim(skb, b);
2691 return -EMSGSIZE;
2692}
2693
2694static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2695 u32 seq, u16 flags, int event, bool unicast)
2696{
2697 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2698 struct tcf_block *block = chain->block;
2699 struct net *net = block->net;
2700 struct sk_buff *skb;
2701 int err = 0;
2702
2703 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2704 if (!skb)
2705 return -ENOBUFS;
2706
2707 if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2708 chain->index, net, skb, block, portid,
2709 seq, flags, event) <= 0) {
2710 kfree_skb(skb);
2711 return -EINVAL;
2712 }
2713
2714 if (unicast)
2715 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2716 else
2717 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2718 flags & NLM_F_ECHO);
2719
2720 if (err > 0)
2721 err = 0;
2722 return err;
2723}
2724
2725static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2726 void *tmplt_priv, u32 chain_index,
2727 struct tcf_block *block, struct sk_buff *oskb,
2728 u32 seq, u16 flags, bool unicast)
2729{
2730 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2731 struct net *net = block->net;
2732 struct sk_buff *skb;
2733
2734 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2735 if (!skb)
2736 return -ENOBUFS;
2737
2738 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2739 block, portid, seq, flags, RTM_DELCHAIN) <= 0) {
2740 kfree_skb(skb);
2741 return -EINVAL;
2742 }
2743
2744 if (unicast)
2745 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2746
2747 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2748}
2749
2750static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2751 struct nlattr **tca,
2752 struct netlink_ext_ack *extack)
2753{
2754 const struct tcf_proto_ops *ops;
2755 char name[IFNAMSIZ];
2756 void *tmplt_priv;
2757
2758 /* If kind is not set, user did not specify template. */
2759 if (!tca[TCA_KIND])
2760 return 0;
2761
2762 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2763 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
2764 return -EINVAL;
2765 }
2766
2767 ops = tcf_proto_lookup_ops(name, true, extack);
2768 if (IS_ERR(ops))
2769 return PTR_ERR(ops);
2770 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2771 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2772 return -EOPNOTSUPP;
2773 }
2774
2775 tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2776 if (IS_ERR(tmplt_priv)) {
2777 module_put(ops->owner);
2778 return PTR_ERR(tmplt_priv);
2779 }
2780 chain->tmplt_ops = ops;
2781 chain->tmplt_priv = tmplt_priv;
2782 return 0;
2783}
2784
2785static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2786 void *tmplt_priv)
2787{
2788 /* If template ops are set, no work to do for us. */
2789 if (!tmplt_ops)
2790 return;
2791
2792 tmplt_ops->tmplt_destroy(tmplt_priv);
2793 module_put(tmplt_ops->owner);
2794}
2795
2796/* Add/delete/get a chain */
2797
2798static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2799 struct netlink_ext_ack *extack)
2800{
2801 struct net *net = sock_net(skb->sk);
2802 struct nlattr *tca[TCA_MAX + 1];
2803 struct tcmsg *t;
2804 u32 parent;
2805 u32 chain_index;
2806 struct Qdisc *q = NULL;
2807 struct tcf_chain *chain = NULL;
2808 struct tcf_block *block;
2809 unsigned long cl;
2810 int err;
2811
2812 if (n->nlmsg_type != RTM_GETCHAIN &&
2813 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2814 return -EPERM;
2815
2816replay:
2817 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2818 rtm_tca_policy, extack);
2819 if (err < 0)
2820 return err;
2821
2822 t = nlmsg_data(n);
2823 parent = t->tcm_parent;
2824 cl = 0;
2825
2826 block = tcf_block_find(net, &q, &parent, &cl,
2827 t->tcm_ifindex, t->tcm_block_index, extack);
2828 if (IS_ERR(block))
2829 return PTR_ERR(block);
2830
2831 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2832 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2833 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2834 err = -EINVAL;
2835 goto errout_block;
2836 }
2837
2838 mutex_lock(&block->lock);
2839 chain = tcf_chain_lookup(block, chain_index);
2840 if (n->nlmsg_type == RTM_NEWCHAIN) {
2841 if (chain) {
2842 if (tcf_chain_held_by_acts_only(chain)) {
2843 /* The chain exists only because there is
2844 * some action referencing it.
2845 */
2846 tcf_chain_hold(chain);
2847 } else {
2848 NL_SET_ERR_MSG(extack, "Filter chain already exists");
2849 err = -EEXIST;
2850 goto errout_block_locked;
2851 }
2852 } else {
2853 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2854 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
2855 err = -ENOENT;
2856 goto errout_block_locked;
2857 }
2858 chain = tcf_chain_create(block, chain_index);
2859 if (!chain) {
2860 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
2861 err = -ENOMEM;
2862 goto errout_block_locked;
2863 }
2864 }
2865 } else {
2866 if (!chain || tcf_chain_held_by_acts_only(chain)) {
2867 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2868 err = -EINVAL;
2869 goto errout_block_locked;
2870 }
2871 tcf_chain_hold(chain);
2872 }
2873
2874 if (n->nlmsg_type == RTM_NEWCHAIN) {
2875 /* Modifying chain requires holding parent block lock. In case
2876 * the chain was successfully added, take a reference to the
2877 * chain. This ensures that an empty chain does not disappear at
2878 * the end of this function.
2879 */
2880 tcf_chain_hold(chain);
2881 chain->explicitly_created = true;
2882 }
2883 mutex_unlock(&block->lock);
2884
2885 switch (n->nlmsg_type) {
2886 case RTM_NEWCHAIN:
2887 err = tc_chain_tmplt_add(chain, net, tca, extack);
2888 if (err) {
2889 tcf_chain_put_explicitly_created(chain);
2890 goto errout;
2891 }
2892
2893 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2894 RTM_NEWCHAIN, false);
2895 break;
2896 case RTM_DELCHAIN:
2897 tfilter_notify_chain(net, skb, block, q, parent, n,
2898 chain, RTM_DELTFILTER);
2899 /* Flush the chain first as the user requested chain removal. */
2900 tcf_chain_flush(chain, true);
2901 /* In case the chain was successfully deleted, put a reference
2902 * to the chain previously taken during addition.
2903 */
2904 tcf_chain_put_explicitly_created(chain);
2905 break;
2906 case RTM_GETCHAIN:
2907 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
2908 n->nlmsg_flags, n->nlmsg_type, true);
2909 if (err < 0)
2910 NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2911 break;
2912 default:
2913 err = -EOPNOTSUPP;
2914 NL_SET_ERR_MSG(extack, "Unsupported message type");
2915 goto errout;
2916 }
2917
2918errout:
2919 tcf_chain_put(chain);
2920errout_block:
2921 tcf_block_release(q, block, true);
2922 if (err == -EAGAIN)
2923 /* Replay the request. */
2924 goto replay;
2925 return err;
2926
2927errout_block_locked:
2928 mutex_unlock(&block->lock);
2929 goto errout_block;
2930}
2931
2932/* called with RTNL */
2933static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2934{
2935 struct net *net = sock_net(skb->sk);
2936 struct nlattr *tca[TCA_MAX + 1];
2937 struct Qdisc *q = NULL;
2938 struct tcf_block *block;
2939 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2940 struct tcf_chain *chain;
2941 long index_start;
2942 long index;
2943 int err;
2944
2945 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2946 return skb->len;
2947
2948 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2949 rtm_tca_policy, cb->extack);
2950 if (err)
2951 return err;
2952
2953 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2954 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2955 if (!block)
2956 goto out;
2957 } else {
2958 const struct Qdisc_class_ops *cops;
2959 struct net_device *dev;
2960 unsigned long cl = 0;
2961
2962 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2963 if (!dev)
2964 return skb->len;
2965
2966 if (!tcm->tcm_parent)
2967 q = dev->qdisc;
2968 else
2969 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2970
2971 if (!q)
2972 goto out;
2973 cops = q->ops->cl_ops;
2974 if (!cops)
2975 goto out;
2976 if (!cops->tcf_block)
2977 goto out;
2978 if (TC_H_MIN(tcm->tcm_parent)) {
2979 cl = cops->find(q, tcm->tcm_parent);
2980 if (cl == 0)
2981 goto out;
2982 }
2983 block = cops->tcf_block(q, cl, NULL);
2984 if (!block)
2985 goto out;
2986 if (tcf_block_shared(block))
2987 q = NULL;
2988 }
2989
2990 index_start = cb->args[0];
2991 index = 0;
2992
2993 mutex_lock(&block->lock);
2994 list_for_each_entry(chain, &block->chain_list, list) {
2995 if ((tca[TCA_CHAIN] &&
2996 nla_get_u32(tca[TCA_CHAIN]) != chain->index))
2997 continue;
2998 if (index < index_start) {
2999 index++;
3000 continue;
3001 }
3002 if (tcf_chain_held_by_acts_only(chain))
3003 continue;
3004 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3005 chain->index, net, skb, block,
3006 NETLINK_CB(cb->skb).portid,
3007 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3008 RTM_NEWCHAIN);
3009 if (err <= 0)
3010 break;
3011 index++;
3012 }
3013 mutex_unlock(&block->lock);
3014
3015 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3016 tcf_block_refcnt_put(block, true);
3017 cb->args[0] = index;
3018
3019out:
3020 /* If we did no progress, the error (EMSGSIZE) is real */
3021 if (skb->len == 0 && err)
3022 return err;
3023 return skb->len;
3024}
3025
3026void tcf_exts_destroy(struct tcf_exts *exts)
3027{
3028#ifdef CONFIG_NET_CLS_ACT
3029 if (exts->actions) {
3030 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3031 kfree(exts->actions);
3032 }
3033 exts->nr_actions = 0;
3034#endif
3035}
3036EXPORT_SYMBOL(tcf_exts_destroy);
3037
3038int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3039 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
3040 bool rtnl_held, struct netlink_ext_ack *extack)
3041{
3042#ifdef CONFIG_NET_CLS_ACT
3043 {
3044 int init_res[TCA_ACT_MAX_PRIO] = {};
3045 struct tc_action *act;
3046 size_t attr_size = 0;
3047
3048 if (exts->police && tb[exts->police]) {
3049 struct tc_action_ops *a_o;
3050
3051 a_o = tc_action_load_ops("police", tb[exts->police], rtnl_held, extack);
3052 if (IS_ERR(a_o))
3053 return PTR_ERR(a_o);
3054 act = tcf_action_init_1(net, tp, tb[exts->police],
3055 rate_tlv, "police", ovr,
3056 TCA_ACT_BIND, a_o, init_res,
3057 rtnl_held, extack);
3058 module_put(a_o->owner);
3059 if (IS_ERR(act))
3060 return PTR_ERR(act);
3061
3062 act->type = exts->type = TCA_OLD_COMPAT;
3063 exts->actions[0] = act;
3064 exts->nr_actions = 1;
3065 tcf_idr_insert_many(exts->actions);
3066 } else if (exts->action && tb[exts->action]) {
3067 int err;
3068
3069 err = tcf_action_init(net, tp, tb[exts->action],
3070 rate_tlv, NULL, ovr, TCA_ACT_BIND,
3071 exts->actions, init_res,
3072 &attr_size, rtnl_held, extack);
3073 if (err < 0)
3074 return err;
3075 exts->nr_actions = err;
3076 }
3077 }
3078#else
3079 if ((exts->action && tb[exts->action]) ||
3080 (exts->police && tb[exts->police])) {
3081 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3082 return -EOPNOTSUPP;
3083 }
3084#endif
3085
3086 return 0;
3087}
3088EXPORT_SYMBOL(tcf_exts_validate);
3089
3090void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3091{
3092#ifdef CONFIG_NET_CLS_ACT
3093 struct tcf_exts old = *dst;
3094
3095 *dst = *src;
3096 tcf_exts_destroy(&old);
3097#endif
3098}
3099EXPORT_SYMBOL(tcf_exts_change);
3100
3101#ifdef CONFIG_NET_CLS_ACT
3102static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3103{
3104 if (exts->nr_actions == 0)
3105 return NULL;
3106 else
3107 return exts->actions[0];
3108}
3109#endif
3110
3111int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3112{
3113#ifdef CONFIG_NET_CLS_ACT
3114 struct nlattr *nest;
3115
3116 if (exts->action && tcf_exts_has_actions(exts)) {
3117 /*
3118 * again for backward compatible mode - we want
3119 * to work with both old and new modes of entering
3120 * tc data even if iproute2 was newer - jhs
3121 */
3122 if (exts->type != TCA_OLD_COMPAT) {
3123 nest = nla_nest_start_noflag(skb, exts->action);
3124 if (nest == NULL)
3125 goto nla_put_failure;
3126
3127 if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3128 < 0)
3129 goto nla_put_failure;
3130 nla_nest_end(skb, nest);
3131 } else if (exts->police) {
3132 struct tc_action *act = tcf_exts_first_act(exts);
3133 nest = nla_nest_start_noflag(skb, exts->police);
3134 if (nest == NULL || !act)
3135 goto nla_put_failure;
3136 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3137 goto nla_put_failure;
3138 nla_nest_end(skb, nest);
3139 }
3140 }
3141 return 0;
3142
3143nla_put_failure:
3144 nla_nest_cancel(skb, nest);
3145 return -1;
3146#else
3147 return 0;
3148#endif
3149}
3150EXPORT_SYMBOL(tcf_exts_dump);
3151
3152int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3153{
3154#ifdef CONFIG_NET_CLS_ACT
3155 struct nlattr *nest;
3156
3157 if (!exts->action || !tcf_exts_has_actions(exts))
3158 return 0;
3159
3160 nest = nla_nest_start_noflag(skb, exts->action);
3161 if (!nest)
3162 goto nla_put_failure;
3163
3164 if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3165 goto nla_put_failure;
3166 nla_nest_end(skb, nest);
3167 return 0;
3168
3169nla_put_failure:
3170 nla_nest_cancel(skb, nest);
3171 return -1;
3172#else
3173 return 0;
3174#endif
3175}
3176EXPORT_SYMBOL(tcf_exts_terse_dump);
3177
3178int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3179{
3180#ifdef CONFIG_NET_CLS_ACT
3181 struct tc_action *a = tcf_exts_first_act(exts);
3182 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3183 return -1;
3184#endif
3185 return 0;
3186}
3187EXPORT_SYMBOL(tcf_exts_dump_stats);
3188
3189static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3190{
3191 if (*flags & TCA_CLS_FLAGS_IN_HW)
3192 return;
3193 *flags |= TCA_CLS_FLAGS_IN_HW;
3194 atomic_inc(&block->offloadcnt);
3195}
3196
3197static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3198{
3199 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3200 return;
3201 *flags &= ~TCA_CLS_FLAGS_IN_HW;
3202 atomic_dec(&block->offloadcnt);
3203}
3204
3205static void tc_cls_offload_cnt_update(struct tcf_block *block,
3206 struct tcf_proto *tp, u32 *cnt,
3207 u32 *flags, u32 diff, bool add)
3208{
3209 lockdep_assert_held(&block->cb_lock);
3210
3211 spin_lock(&tp->lock);
3212 if (add) {
3213 if (!*cnt)
3214 tcf_block_offload_inc(block, flags);
3215 *cnt += diff;
3216 } else {
3217 *cnt -= diff;
3218 if (!*cnt)
3219 tcf_block_offload_dec(block, flags);
3220 }
3221 spin_unlock(&tp->lock);
3222}
3223
3224static void
3225tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3226 u32 *cnt, u32 *flags)
3227{
3228 lockdep_assert_held(&block->cb_lock);
3229
3230 spin_lock(&tp->lock);
3231 tcf_block_offload_dec(block, flags);
3232 *cnt = 0;
3233 spin_unlock(&tp->lock);
3234}
3235
3236static int
3237__tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3238 void *type_data, bool err_stop)
3239{
3240 struct flow_block_cb *block_cb;
3241 int ok_count = 0;
3242 int err;
3243
3244 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3245 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3246 if (err) {
3247 if (err_stop)
3248 return err;
3249 } else {
3250 ok_count++;
3251 }
3252 }
3253 return ok_count;
3254}
3255
3256int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3257 void *type_data, bool err_stop, bool rtnl_held)
3258{
3259 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3260 int ok_count;
3261
3262retry:
3263 if (take_rtnl)
3264 rtnl_lock();
3265 down_read(&block->cb_lock);
3266 /* Need to obtain rtnl lock if block is bound to devs that require it.
3267 * In block bind code cb_lock is obtained while holding rtnl, so we must
3268 * obtain the locks in same order here.
3269 */
3270 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3271 up_read(&block->cb_lock);
3272 take_rtnl = true;
3273 goto retry;
3274 }
3275
3276 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3277
3278 up_read(&block->cb_lock);
3279 if (take_rtnl)
3280 rtnl_unlock();
3281 return ok_count;
3282}
3283EXPORT_SYMBOL(tc_setup_cb_call);
3284
3285/* Non-destructive filter add. If filter that wasn't already in hardware is
3286 * successfully offloaded, increment block offloads counter. On failure,
3287 * previously offloaded filter is considered to be intact and offloads counter
3288 * is not decremented.
3289 */
3290
3291int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3292 enum tc_setup_type type, void *type_data, bool err_stop,
3293 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3294{
3295 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3296 int ok_count;
3297
3298retry:
3299 if (take_rtnl)
3300 rtnl_lock();
3301 down_read(&block->cb_lock);
3302 /* Need to obtain rtnl lock if block is bound to devs that require it.
3303 * In block bind code cb_lock is obtained while holding rtnl, so we must
3304 * obtain the locks in same order here.
3305 */
3306 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3307 up_read(&block->cb_lock);
3308 take_rtnl = true;
3309 goto retry;
3310 }
3311
3312 /* Make sure all netdevs sharing this block are offload-capable. */
3313 if (block->nooffloaddevcnt && err_stop) {
3314 ok_count = -EOPNOTSUPP;
3315 goto err_unlock;
3316 }
3317
3318 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3319 if (ok_count < 0)
3320 goto err_unlock;
3321
3322 if (tp->ops->hw_add)
3323 tp->ops->hw_add(tp, type_data);
3324 if (ok_count > 0)
3325 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3326 ok_count, true);
3327err_unlock:
3328 up_read(&block->cb_lock);
3329 if (take_rtnl)
3330 rtnl_unlock();
3331 return ok_count < 0 ? ok_count : 0;
3332}
3333EXPORT_SYMBOL(tc_setup_cb_add);
3334
3335/* Destructive filter replace. If filter that wasn't already in hardware is
3336 * successfully offloaded, increment block offload counter. On failure,
3337 * previously offloaded filter is considered to be destroyed and offload counter
3338 * is decremented.
3339 */
3340
3341int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3342 enum tc_setup_type type, void *type_data, bool err_stop,
3343 u32 *old_flags, unsigned int *old_in_hw_count,
3344 u32 *new_flags, unsigned int *new_in_hw_count,
3345 bool rtnl_held)
3346{
3347 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3348 int ok_count;
3349
3350retry:
3351 if (take_rtnl)
3352 rtnl_lock();
3353 down_read(&block->cb_lock);
3354 /* Need to obtain rtnl lock if block is bound to devs that require it.
3355 * In block bind code cb_lock is obtained while holding rtnl, so we must
3356 * obtain the locks in same order here.
3357 */
3358 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3359 up_read(&block->cb_lock);
3360 take_rtnl = true;
3361 goto retry;
3362 }
3363
3364 /* Make sure all netdevs sharing this block are offload-capable. */
3365 if (block->nooffloaddevcnt && err_stop) {
3366 ok_count = -EOPNOTSUPP;
3367 goto err_unlock;
3368 }
3369
3370 tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3371 if (tp->ops->hw_del)
3372 tp->ops->hw_del(tp, type_data);
3373
3374 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3375 if (ok_count < 0)
3376 goto err_unlock;
3377
3378 if (tp->ops->hw_add)
3379 tp->ops->hw_add(tp, type_data);
3380 if (ok_count > 0)
3381 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3382 new_flags, ok_count, true);
3383err_unlock:
3384 up_read(&block->cb_lock);
3385 if (take_rtnl)
3386 rtnl_unlock();
3387 return ok_count < 0 ? ok_count : 0;
3388}
3389EXPORT_SYMBOL(tc_setup_cb_replace);
3390
3391/* Destroy filter and decrement block offload counter, if filter was previously
3392 * offloaded.
3393 */
3394
3395int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3396 enum tc_setup_type type, void *type_data, bool err_stop,
3397 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3398{
3399 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3400 int ok_count;
3401
3402retry:
3403 if (take_rtnl)
3404 rtnl_lock();
3405 down_read(&block->cb_lock);
3406 /* Need to obtain rtnl lock if block is bound to devs that require it.
3407 * In block bind code cb_lock is obtained while holding rtnl, so we must
3408 * obtain the locks in same order here.
3409 */
3410 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3411 up_read(&block->cb_lock);
3412 take_rtnl = true;
3413 goto retry;
3414 }
3415
3416 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3417
3418 tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3419 if (tp->ops->hw_del)
3420 tp->ops->hw_del(tp, type_data);
3421
3422 up_read(&block->cb_lock);
3423 if (take_rtnl)
3424 rtnl_unlock();
3425 return ok_count < 0 ? ok_count : 0;
3426}
3427EXPORT_SYMBOL(tc_setup_cb_destroy);
3428
3429int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3430 bool add, flow_setup_cb_t *cb,
3431 enum tc_setup_type type, void *type_data,
3432 void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3433{
3434 int err = cb(type, type_data, cb_priv);
3435
3436 if (err) {
3437 if (add && tc_skip_sw(*flags))
3438 return err;
3439 } else {
3440 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3441 add);
3442 }
3443
3444 return 0;
3445}
3446EXPORT_SYMBOL(tc_setup_cb_reoffload);
3447
3448static int tcf_act_get_cookie(struct flow_action_entry *entry,
3449 const struct tc_action *act)
3450{
3451 struct tc_cookie *cookie;
3452 int err = 0;
3453
3454 rcu_read_lock();
3455 cookie = rcu_dereference(act->act_cookie);
3456 if (cookie) {
3457 entry->cookie = flow_action_cookie_create(cookie->data,
3458 cookie->len,
3459 GFP_ATOMIC);
3460 if (!entry->cookie)
3461 err = -ENOMEM;
3462 }
3463 rcu_read_unlock();
3464 return err;
3465}
3466
3467static void tcf_act_put_cookie(struct flow_action_entry *entry)
3468{
3469 flow_action_cookie_destroy(entry->cookie);
3470}
3471
3472void tc_cleanup_flow_action(struct flow_action *flow_action)
3473{
3474 struct flow_action_entry *entry;
3475 int i;
3476
3477 flow_action_for_each(i, entry, flow_action) {
3478 tcf_act_put_cookie(entry);
3479 if (entry->destructor)
3480 entry->destructor(entry->destructor_priv);
3481 }
3482}
3483EXPORT_SYMBOL(tc_cleanup_flow_action);
3484
3485static void tcf_mirred_get_dev(struct flow_action_entry *entry,
3486 const struct tc_action *act)
3487{
3488#ifdef CONFIG_NET_CLS_ACT
3489 entry->dev = act->ops->get_dev(act, &entry->destructor);
3490 if (!entry->dev)
3491 return;
3492 entry->destructor_priv = entry->dev;
3493#endif
3494}
3495
3496static void tcf_tunnel_encap_put_tunnel(void *priv)
3497{
3498 struct ip_tunnel_info *tunnel = priv;
3499
3500 kfree(tunnel);
3501}
3502
3503static int tcf_tunnel_encap_get_tunnel(struct flow_action_entry *entry,
3504 const struct tc_action *act)
3505{
3506 entry->tunnel = tcf_tunnel_info_copy(act);
3507 if (!entry->tunnel)
3508 return -ENOMEM;
3509 entry->destructor = tcf_tunnel_encap_put_tunnel;
3510 entry->destructor_priv = entry->tunnel;
3511 return 0;
3512}
3513
3514static void tcf_sample_get_group(struct flow_action_entry *entry,
3515 const struct tc_action *act)
3516{
3517#ifdef CONFIG_NET_CLS_ACT
3518 entry->sample.psample_group =
3519 act->ops->get_psample_group(act, &entry->destructor);
3520 entry->destructor_priv = entry->sample.psample_group;
3521#endif
3522}
3523
3524static void tcf_gate_entry_destructor(void *priv)
3525{
3526 struct action_gate_entry *oe = priv;
3527
3528 kfree(oe);
3529}
3530
3531static int tcf_gate_get_entries(struct flow_action_entry *entry,
3532 const struct tc_action *act)
3533{
3534 entry->gate.entries = tcf_gate_get_list(act);
3535
3536 if (!entry->gate.entries)
3537 return -EINVAL;
3538
3539 entry->destructor = tcf_gate_entry_destructor;
3540 entry->destructor_priv = entry->gate.entries;
3541
3542 return 0;
3543}
3544
3545static enum flow_action_hw_stats tc_act_hw_stats(u8 hw_stats)
3546{
3547 if (WARN_ON_ONCE(hw_stats > TCA_ACT_HW_STATS_ANY))
3548 return FLOW_ACTION_HW_STATS_DONT_CARE;
3549 else if (!hw_stats)
3550 return FLOW_ACTION_HW_STATS_DISABLED;
3551
3552 return hw_stats;
3553}
3554
3555int tc_setup_flow_action(struct flow_action *flow_action,
3556 const struct tcf_exts *exts)
3557{
3558 struct tc_action *act;
3559 int i, j, k, err = 0;
3560
3561 BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3562 BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3563 BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3564
3565 if (!exts)
3566 return 0;
3567
3568 j = 0;
3569 tcf_exts_for_each_action(i, act, exts) {
3570 struct flow_action_entry *entry;
3571
3572 entry = &flow_action->entries[j];
3573 spin_lock_bh(&act->tcfa_lock);
3574 err = tcf_act_get_cookie(entry, act);
3575 if (err)
3576 goto err_out_locked;
3577
3578 entry->hw_stats = tc_act_hw_stats(act->hw_stats);
3579
3580 if (is_tcf_gact_ok(act)) {
3581 entry->id = FLOW_ACTION_ACCEPT;
3582 } else if (is_tcf_gact_shot(act)) {
3583 entry->id = FLOW_ACTION_DROP;
3584 } else if (is_tcf_gact_trap(act)) {
3585 entry->id = FLOW_ACTION_TRAP;
3586 } else if (is_tcf_gact_goto_chain(act)) {
3587 entry->id = FLOW_ACTION_GOTO;
3588 entry->chain_index = tcf_gact_goto_chain_index(act);
3589 } else if (is_tcf_mirred_egress_redirect(act)) {
3590 entry->id = FLOW_ACTION_REDIRECT;
3591 tcf_mirred_get_dev(entry, act);
3592 } else if (is_tcf_mirred_egress_mirror(act)) {
3593 entry->id = FLOW_ACTION_MIRRED;
3594 tcf_mirred_get_dev(entry, act);
3595 } else if (is_tcf_mirred_ingress_redirect(act)) {
3596 entry->id = FLOW_ACTION_REDIRECT_INGRESS;
3597 tcf_mirred_get_dev(entry, act);
3598 } else if (is_tcf_mirred_ingress_mirror(act)) {
3599 entry->id = FLOW_ACTION_MIRRED_INGRESS;
3600 tcf_mirred_get_dev(entry, act);
3601 } else if (is_tcf_vlan(act)) {
3602 switch (tcf_vlan_action(act)) {
3603 case TCA_VLAN_ACT_PUSH:
3604 entry->id = FLOW_ACTION_VLAN_PUSH;
3605 entry->vlan.vid = tcf_vlan_push_vid(act);
3606 entry->vlan.proto = tcf_vlan_push_proto(act);
3607 entry->vlan.prio = tcf_vlan_push_prio(act);
3608 break;
3609 case TCA_VLAN_ACT_POP:
3610 entry->id = FLOW_ACTION_VLAN_POP;
3611 break;
3612 case TCA_VLAN_ACT_MODIFY:
3613 entry->id = FLOW_ACTION_VLAN_MANGLE;
3614 entry->vlan.vid = tcf_vlan_push_vid(act);
3615 entry->vlan.proto = tcf_vlan_push_proto(act);
3616 entry->vlan.prio = tcf_vlan_push_prio(act);
3617 break;
3618 default:
3619 err = -EOPNOTSUPP;
3620 goto err_out_locked;
3621 }
3622 } else if (is_tcf_tunnel_set(act)) {
3623 entry->id = FLOW_ACTION_TUNNEL_ENCAP;
3624 err = tcf_tunnel_encap_get_tunnel(entry, act);
3625 if (err)
3626 goto err_out_locked;
3627 } else if (is_tcf_tunnel_release(act)) {
3628 entry->id = FLOW_ACTION_TUNNEL_DECAP;
3629 } else if (is_tcf_pedit(act)) {
3630 for (k = 0; k < tcf_pedit_nkeys(act); k++) {
3631 switch (tcf_pedit_cmd(act, k)) {
3632 case TCA_PEDIT_KEY_EX_CMD_SET:
3633 entry->id = FLOW_ACTION_MANGLE;
3634 break;
3635 case TCA_PEDIT_KEY_EX_CMD_ADD:
3636 entry->id = FLOW_ACTION_ADD;
3637 break;
3638 default:
3639 err = -EOPNOTSUPP;
3640 goto err_out_locked;
3641 }
3642 entry->mangle.htype = tcf_pedit_htype(act, k);
3643 entry->mangle.mask = tcf_pedit_mask(act, k);
3644 entry->mangle.val = tcf_pedit_val(act, k);
3645 entry->mangle.offset = tcf_pedit_offset(act, k);
3646 entry->hw_stats = tc_act_hw_stats(act->hw_stats);
3647 entry = &flow_action->entries[++j];
3648 }
3649 } else if (is_tcf_csum(act)) {
3650 entry->id = FLOW_ACTION_CSUM;
3651 entry->csum_flags = tcf_csum_update_flags(act);
3652 } else if (is_tcf_skbedit_mark(act)) {
3653 entry->id = FLOW_ACTION_MARK;
3654 entry->mark = tcf_skbedit_mark(act);
3655 } else if (is_tcf_sample(act)) {
3656 entry->id = FLOW_ACTION_SAMPLE;
3657 entry->sample.trunc_size = tcf_sample_trunc_size(act);
3658 entry->sample.truncate = tcf_sample_truncate(act);
3659 entry->sample.rate = tcf_sample_rate(act);
3660 tcf_sample_get_group(entry, act);
3661 } else if (is_tcf_police(act)) {
3662 entry->id = FLOW_ACTION_POLICE;
3663 entry->police.burst = tcf_police_burst(act);
3664 entry->police.rate_bytes_ps =
3665 tcf_police_rate_bytes_ps(act);
3666 entry->police.burst_pkt = tcf_police_burst_pkt(act);
3667 entry->police.rate_pkt_ps =
3668 tcf_police_rate_pkt_ps(act);
3669 entry->police.mtu = tcf_police_tcfp_mtu(act);
3670 entry->police.index = act->tcfa_index;
3671 } else if (is_tcf_ct(act)) {
3672 entry->id = FLOW_ACTION_CT;
3673 entry->ct.action = tcf_ct_action(act);
3674 entry->ct.zone = tcf_ct_zone(act);
3675 entry->ct.flow_table = tcf_ct_ft(act);
3676 } else if (is_tcf_mpls(act)) {
3677 switch (tcf_mpls_action(act)) {
3678 case TCA_MPLS_ACT_PUSH:
3679 entry->id = FLOW_ACTION_MPLS_PUSH;
3680 entry->mpls_push.proto = tcf_mpls_proto(act);
3681 entry->mpls_push.label = tcf_mpls_label(act);
3682 entry->mpls_push.tc = tcf_mpls_tc(act);
3683 entry->mpls_push.bos = tcf_mpls_bos(act);
3684 entry->mpls_push.ttl = tcf_mpls_ttl(act);
3685 break;
3686 case TCA_MPLS_ACT_POP:
3687 entry->id = FLOW_ACTION_MPLS_POP;
3688 entry->mpls_pop.proto = tcf_mpls_proto(act);
3689 break;
3690 case TCA_MPLS_ACT_MODIFY:
3691 entry->id = FLOW_ACTION_MPLS_MANGLE;
3692 entry->mpls_mangle.label = tcf_mpls_label(act);
3693 entry->mpls_mangle.tc = tcf_mpls_tc(act);
3694 entry->mpls_mangle.bos = tcf_mpls_bos(act);
3695 entry->mpls_mangle.ttl = tcf_mpls_ttl(act);
3696 break;
3697 default:
3698 goto err_out_locked;
3699 }
3700 } else if (is_tcf_skbedit_ptype(act)) {
3701 entry->id = FLOW_ACTION_PTYPE;
3702 entry->ptype = tcf_skbedit_ptype(act);
3703 } else if (is_tcf_skbedit_priority(act)) {
3704 entry->id = FLOW_ACTION_PRIORITY;
3705 entry->priority = tcf_skbedit_priority(act);
3706 } else if (is_tcf_gate(act)) {
3707 entry->id = FLOW_ACTION_GATE;
3708 entry->gate.index = tcf_gate_index(act);
3709 entry->gate.prio = tcf_gate_prio(act);
3710 entry->gate.basetime = tcf_gate_basetime(act);
3711 entry->gate.cycletime = tcf_gate_cycletime(act);
3712 entry->gate.cycletimeext = tcf_gate_cycletimeext(act);
3713 entry->gate.num_entries = tcf_gate_num_entries(act);
3714 err = tcf_gate_get_entries(entry, act);
3715 if (err)
3716 goto err_out_locked;
3717 } else {
3718 err = -EOPNOTSUPP;
3719 goto err_out_locked;
3720 }
3721 spin_unlock_bh(&act->tcfa_lock);
3722
3723 if (!is_tcf_pedit(act))
3724 j++;
3725 }
3726
3727err_out:
3728 if (err)
3729 tc_cleanup_flow_action(flow_action);
3730
3731 return err;
3732err_out_locked:
3733 spin_unlock_bh(&act->tcfa_lock);
3734 goto err_out;
3735}
3736EXPORT_SYMBOL(tc_setup_flow_action);
3737
3738unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3739{
3740 unsigned int num_acts = 0;
3741 struct tc_action *act;
3742 int i;
3743
3744 tcf_exts_for_each_action(i, act, exts) {
3745 if (is_tcf_pedit(act))
3746 num_acts += tcf_pedit_nkeys(act);
3747 else
3748 num_acts++;
3749 }
3750 return num_acts;
3751}
3752EXPORT_SYMBOL(tcf_exts_num_actions);
3753
3754#ifdef CONFIG_NET_CLS_ACT
3755static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3756 u32 *p_block_index,
3757 struct netlink_ext_ack *extack)
3758{
3759 *p_block_index = nla_get_u32(block_index_attr);
3760 if (!*p_block_index) {
3761 NL_SET_ERR_MSG(extack, "Block number may not be zero");
3762 return -EINVAL;
3763 }
3764
3765 return 0;
3766}
3767
3768int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3769 enum flow_block_binder_type binder_type,
3770 struct nlattr *block_index_attr,
3771 struct netlink_ext_ack *extack)
3772{
3773 u32 block_index;
3774 int err;
3775
3776 if (!block_index_attr)
3777 return 0;
3778
3779 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3780 if (err)
3781 return err;
3782
3783 if (!block_index)
3784 return 0;
3785
3786 qe->info.binder_type = binder_type;
3787 qe->info.chain_head_change = tcf_chain_head_change_dflt;
3788 qe->info.chain_head_change_priv = &qe->filter_chain;
3789 qe->info.block_index = block_index;
3790
3791 return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3792}
3793EXPORT_SYMBOL(tcf_qevent_init);
3794
3795void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3796{
3797 if (qe->info.block_index)
3798 tcf_block_put_ext(qe->block, sch, &qe->info);
3799}
3800EXPORT_SYMBOL(tcf_qevent_destroy);
3801
3802int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3803 struct netlink_ext_ack *extack)
3804{
3805 u32 block_index;
3806 int err;
3807
3808 if (!block_index_attr)
3809 return 0;
3810
3811 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3812 if (err)
3813 return err;
3814
3815 /* Bounce newly-configured block or change in block. */
3816 if (block_index != qe->info.block_index) {
3817 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3818 return -EINVAL;
3819 }
3820
3821 return 0;
3822}
3823EXPORT_SYMBOL(tcf_qevent_validate_change);
3824
3825struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
3826 struct sk_buff **to_free, int *ret)
3827{
3828 struct tcf_result cl_res;
3829 struct tcf_proto *fl;
3830
3831 if (!qe->info.block_index)
3832 return skb;
3833
3834 fl = rcu_dereference_bh(qe->filter_chain);
3835
3836 switch (tcf_classify(skb, fl, &cl_res, false)) {
3837 case TC_ACT_SHOT:
3838 qdisc_qstats_drop(sch);
3839 __qdisc_drop(skb, to_free);
3840 *ret = __NET_XMIT_BYPASS;
3841 return NULL;
3842 case TC_ACT_STOLEN:
3843 case TC_ACT_QUEUED:
3844 case TC_ACT_TRAP:
3845 __qdisc_drop(skb, to_free);
3846 *ret = __NET_XMIT_STOLEN;
3847 return NULL;
3848 case TC_ACT_REDIRECT:
3849 skb_do_redirect(skb);
3850 *ret = __NET_XMIT_STOLEN;
3851 return NULL;
3852 }
3853
3854 return skb;
3855}
3856EXPORT_SYMBOL(tcf_qevent_handle);
3857
3858int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
3859{
3860 if (!qe->info.block_index)
3861 return 0;
3862 return nla_put_u32(skb, attr_name, qe->info.block_index);
3863}
3864EXPORT_SYMBOL(tcf_qevent_dump);
3865#endif
3866
3867static __net_init int tcf_net_init(struct net *net)
3868{
3869 struct tcf_net *tn = net_generic(net, tcf_net_id);
3870
3871 spin_lock_init(&tn->idr_lock);
3872 idr_init(&tn->idr);
3873 return 0;
3874}
3875
3876static void __net_exit tcf_net_exit(struct net *net)
3877{
3878 struct tcf_net *tn = net_generic(net, tcf_net_id);
3879
3880 idr_destroy(&tn->idr);
3881}
3882
3883static struct pernet_operations tcf_net_ops = {
3884 .init = tcf_net_init,
3885 .exit = tcf_net_exit,
3886 .id = &tcf_net_id,
3887 .size = sizeof(struct tcf_net),
3888};
3889
3890static int __init tc_filter_init(void)
3891{
3892 int err;
3893
3894 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3895 if (!tc_filter_wq)
3896 return -ENOMEM;
3897
3898 err = register_pernet_subsys(&tcf_net_ops);
3899 if (err)
3900 goto err_register_pernet_subsys;
3901
3902 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3903 RTNL_FLAG_DOIT_UNLOCKED);
3904 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3905 RTNL_FLAG_DOIT_UNLOCKED);
3906 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3907 tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3908 rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3909 rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3910 rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3911 tc_dump_chain, 0);
3912
3913 return 0;
3914
3915err_register_pernet_subsys:
3916 destroy_workqueue(tc_filter_wq);
3917 return err;
3918}
3919
3920subsys_initcall(tc_filter_init);