Loading...
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 <linux/rhashtable.h>
26#include <net/net_namespace.h>
27#include <net/sock.h>
28#include <net/netlink.h>
29#include <net/pkt_sched.h>
30#include <net/pkt_cls.h>
31#include <net/tc_act/tc_pedit.h>
32#include <net/tc_act/tc_mirred.h>
33#include <net/tc_act/tc_vlan.h>
34#include <net/tc_act/tc_tunnel_key.h>
35#include <net/tc_act/tc_csum.h>
36#include <net/tc_act/tc_gact.h>
37#include <net/tc_act/tc_police.h>
38#include <net/tc_act/tc_sample.h>
39#include <net/tc_act/tc_skbedit.h>
40#include <net/tc_act/tc_ct.h>
41#include <net/tc_act/tc_mpls.h>
42#include <net/tc_act/tc_gate.h>
43#include <net/flow_offload.h>
44#include <net/tc_wrapper.h>
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 struct xarray tcf_exts_miss_cookies_xa;
53struct tcf_exts_miss_cookie_node {
54 const struct tcf_chain *chain;
55 const struct tcf_proto *tp;
56 const struct tcf_exts *exts;
57 u32 chain_index;
58 u32 tp_prio;
59 u32 handle;
60 u32 miss_cookie_base;
61 struct rcu_head rcu;
62};
63
64/* Each tc action entry cookie will be comprised of 32bit miss_cookie_base +
65 * action index in the exts tc actions array.
66 */
67union tcf_exts_miss_cookie {
68 struct {
69 u32 miss_cookie_base;
70 u32 act_index;
71 };
72 u64 miss_cookie;
73};
74
75#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
76static int
77tcf_exts_miss_cookie_base_alloc(struct tcf_exts *exts, struct tcf_proto *tp,
78 u32 handle)
79{
80 struct tcf_exts_miss_cookie_node *n;
81 static u32 next;
82 int err;
83
84 if (WARN_ON(!handle || !tp->ops->get_exts))
85 return -EINVAL;
86
87 n = kzalloc(sizeof(*n), GFP_KERNEL);
88 if (!n)
89 return -ENOMEM;
90
91 n->chain_index = tp->chain->index;
92 n->chain = tp->chain;
93 n->tp_prio = tp->prio;
94 n->tp = tp;
95 n->exts = exts;
96 n->handle = handle;
97
98 err = xa_alloc_cyclic(&tcf_exts_miss_cookies_xa, &n->miss_cookie_base,
99 n, xa_limit_32b, &next, GFP_KERNEL);
100 if (err < 0)
101 goto err_xa_alloc;
102
103 exts->miss_cookie_node = n;
104 return 0;
105
106err_xa_alloc:
107 kfree(n);
108 return err;
109}
110
111static void tcf_exts_miss_cookie_base_destroy(struct tcf_exts *exts)
112{
113 struct tcf_exts_miss_cookie_node *n;
114
115 if (!exts->miss_cookie_node)
116 return;
117
118 n = exts->miss_cookie_node;
119 xa_erase(&tcf_exts_miss_cookies_xa, n->miss_cookie_base);
120 kfree_rcu(n, rcu);
121}
122
123static struct tcf_exts_miss_cookie_node *
124tcf_exts_miss_cookie_lookup(u64 miss_cookie, int *act_index)
125{
126 union tcf_exts_miss_cookie mc = { .miss_cookie = miss_cookie, };
127
128 *act_index = mc.act_index;
129 return xa_load(&tcf_exts_miss_cookies_xa, mc.miss_cookie_base);
130}
131#else /* IS_ENABLED(CONFIG_NET_TC_SKB_EXT) */
132static int
133tcf_exts_miss_cookie_base_alloc(struct tcf_exts *exts, struct tcf_proto *tp,
134 u32 handle)
135{
136 return 0;
137}
138
139static void tcf_exts_miss_cookie_base_destroy(struct tcf_exts *exts)
140{
141}
142#endif /* IS_ENABLED(CONFIG_NET_TC_SKB_EXT) */
143
144static u64 tcf_exts_miss_cookie_get(u32 miss_cookie_base, int act_index)
145{
146 union tcf_exts_miss_cookie mc = { .act_index = act_index, };
147
148 if (!miss_cookie_base)
149 return 0;
150
151 mc.miss_cookie_base = miss_cookie_base;
152 return mc.miss_cookie;
153}
154
155#ifdef CONFIG_NET_CLS_ACT
156DEFINE_STATIC_KEY_FALSE(tc_skb_ext_tc);
157EXPORT_SYMBOL(tc_skb_ext_tc);
158
159void tc_skb_ext_tc_enable(void)
160{
161 static_branch_inc(&tc_skb_ext_tc);
162}
163EXPORT_SYMBOL(tc_skb_ext_tc_enable);
164
165void tc_skb_ext_tc_disable(void)
166{
167 static_branch_dec(&tc_skb_ext_tc);
168}
169EXPORT_SYMBOL(tc_skb_ext_tc_disable);
170#endif
171
172static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
173{
174 return jhash_3words(tp->chain->index, tp->prio,
175 (__force __u32)tp->protocol, 0);
176}
177
178static void tcf_proto_signal_destroying(struct tcf_chain *chain,
179 struct tcf_proto *tp)
180{
181 struct tcf_block *block = chain->block;
182
183 mutex_lock(&block->proto_destroy_lock);
184 hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
185 destroy_obj_hashfn(tp));
186 mutex_unlock(&block->proto_destroy_lock);
187}
188
189static bool tcf_proto_cmp(const struct tcf_proto *tp1,
190 const struct tcf_proto *tp2)
191{
192 return tp1->chain->index == tp2->chain->index &&
193 tp1->prio == tp2->prio &&
194 tp1->protocol == tp2->protocol;
195}
196
197static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
198 struct tcf_proto *tp)
199{
200 u32 hash = destroy_obj_hashfn(tp);
201 struct tcf_proto *iter;
202 bool found = false;
203
204 rcu_read_lock();
205 hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
206 destroy_ht_node, hash) {
207 if (tcf_proto_cmp(tp, iter)) {
208 found = true;
209 break;
210 }
211 }
212 rcu_read_unlock();
213
214 return found;
215}
216
217static void
218tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
219{
220 struct tcf_block *block = chain->block;
221
222 mutex_lock(&block->proto_destroy_lock);
223 if (hash_hashed(&tp->destroy_ht_node))
224 hash_del_rcu(&tp->destroy_ht_node);
225 mutex_unlock(&block->proto_destroy_lock);
226}
227
228/* Find classifier type by string name */
229
230static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
231{
232 const struct tcf_proto_ops *t, *res = NULL;
233
234 if (kind) {
235 read_lock(&cls_mod_lock);
236 list_for_each_entry(t, &tcf_proto_base, head) {
237 if (strcmp(kind, t->kind) == 0) {
238 if (try_module_get(t->owner))
239 res = t;
240 break;
241 }
242 }
243 read_unlock(&cls_mod_lock);
244 }
245 return res;
246}
247
248static const struct tcf_proto_ops *
249tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
250 struct netlink_ext_ack *extack)
251{
252 const struct tcf_proto_ops *ops;
253
254 ops = __tcf_proto_lookup_ops(kind);
255 if (ops)
256 return ops;
257#ifdef CONFIG_MODULES
258 if (rtnl_held)
259 rtnl_unlock();
260 request_module(NET_CLS_ALIAS_PREFIX "%s", kind);
261 if (rtnl_held)
262 rtnl_lock();
263 ops = __tcf_proto_lookup_ops(kind);
264 /* We dropped the RTNL semaphore in order to perform
265 * the module load. So, even if we succeeded in loading
266 * the module we have to replay the request. We indicate
267 * this using -EAGAIN.
268 */
269 if (ops) {
270 module_put(ops->owner);
271 return ERR_PTR(-EAGAIN);
272 }
273#endif
274 NL_SET_ERR_MSG(extack, "TC classifier not found");
275 return ERR_PTR(-ENOENT);
276}
277
278/* Register(unregister) new classifier type */
279
280int register_tcf_proto_ops(struct tcf_proto_ops *ops)
281{
282 struct tcf_proto_ops *t;
283 int rc = -EEXIST;
284
285 write_lock(&cls_mod_lock);
286 list_for_each_entry(t, &tcf_proto_base, head)
287 if (!strcmp(ops->kind, t->kind))
288 goto out;
289
290 list_add_tail(&ops->head, &tcf_proto_base);
291 rc = 0;
292out:
293 write_unlock(&cls_mod_lock);
294 return rc;
295}
296EXPORT_SYMBOL(register_tcf_proto_ops);
297
298static struct workqueue_struct *tc_filter_wq;
299
300void unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
301{
302 struct tcf_proto_ops *t;
303 int rc = -ENOENT;
304
305 /* Wait for outstanding call_rcu()s, if any, from a
306 * tcf_proto_ops's destroy() handler.
307 */
308 rcu_barrier();
309 flush_workqueue(tc_filter_wq);
310
311 write_lock(&cls_mod_lock);
312 list_for_each_entry(t, &tcf_proto_base, head) {
313 if (t == ops) {
314 list_del(&t->head);
315 rc = 0;
316 break;
317 }
318 }
319 write_unlock(&cls_mod_lock);
320
321 WARN(rc, "unregister tc filter kind(%s) failed %d\n", ops->kind, rc);
322}
323EXPORT_SYMBOL(unregister_tcf_proto_ops);
324
325bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
326{
327 INIT_RCU_WORK(rwork, func);
328 return queue_rcu_work(tc_filter_wq, rwork);
329}
330EXPORT_SYMBOL(tcf_queue_work);
331
332/* Select new prio value from the range, managed by kernel. */
333
334static inline u32 tcf_auto_prio(struct tcf_proto *tp)
335{
336 u32 first = TC_H_MAKE(0xC0000000U, 0U);
337
338 if (tp)
339 first = tp->prio - 1;
340
341 return TC_H_MAJ(first);
342}
343
344static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
345{
346 if (kind)
347 return nla_strscpy(name, kind, IFNAMSIZ) < 0;
348 memset(name, 0, IFNAMSIZ);
349 return false;
350}
351
352static bool tcf_proto_is_unlocked(const char *kind)
353{
354 const struct tcf_proto_ops *ops;
355 bool ret;
356
357 if (strlen(kind) == 0)
358 return false;
359
360 ops = tcf_proto_lookup_ops(kind, false, NULL);
361 /* On error return false to take rtnl lock. Proto lookup/create
362 * functions will perform lookup again and properly handle errors.
363 */
364 if (IS_ERR(ops))
365 return false;
366
367 ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
368 module_put(ops->owner);
369 return ret;
370}
371
372static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
373 u32 prio, struct tcf_chain *chain,
374 bool rtnl_held,
375 struct netlink_ext_ack *extack)
376{
377 struct tcf_proto *tp;
378 int err;
379
380 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
381 if (!tp)
382 return ERR_PTR(-ENOBUFS);
383
384 tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
385 if (IS_ERR(tp->ops)) {
386 err = PTR_ERR(tp->ops);
387 goto errout;
388 }
389 tp->classify = tp->ops->classify;
390 tp->protocol = protocol;
391 tp->prio = prio;
392 tp->chain = chain;
393 tp->usesw = !tp->ops->reoffload;
394 spin_lock_init(&tp->lock);
395 refcount_set(&tp->refcnt, 1);
396
397 err = tp->ops->init(tp);
398 if (err) {
399 module_put(tp->ops->owner);
400 goto errout;
401 }
402 return tp;
403
404errout:
405 kfree(tp);
406 return ERR_PTR(err);
407}
408
409static void tcf_proto_get(struct tcf_proto *tp)
410{
411 refcount_inc(&tp->refcnt);
412}
413
414static void tcf_proto_count_usesw(struct tcf_proto *tp, bool add)
415{
416#ifdef CONFIG_NET_CLS_ACT
417 struct tcf_block *block = tp->chain->block;
418 bool counted = false;
419
420 if (!add) {
421 if (tp->usesw && tp->counted) {
422 if (!atomic_dec_return(&block->useswcnt))
423 static_branch_dec(&tcf_sw_enabled_key);
424 tp->counted = false;
425 }
426 return;
427 }
428
429 spin_lock(&tp->lock);
430 if (tp->usesw && !tp->counted) {
431 counted = true;
432 tp->counted = true;
433 }
434 spin_unlock(&tp->lock);
435
436 if (counted && atomic_inc_return(&block->useswcnt) == 1)
437 static_branch_inc(&tcf_sw_enabled_key);
438#endif
439}
440
441static void tcf_chain_put(struct tcf_chain *chain);
442
443static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
444 bool sig_destroy, struct netlink_ext_ack *extack)
445{
446 tp->ops->destroy(tp, rtnl_held, extack);
447 tcf_proto_count_usesw(tp, false);
448 if (sig_destroy)
449 tcf_proto_signal_destroyed(tp->chain, tp);
450 tcf_chain_put(tp->chain);
451 module_put(tp->ops->owner);
452 kfree_rcu(tp, rcu);
453}
454
455static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
456 struct netlink_ext_ack *extack)
457{
458 if (refcount_dec_and_test(&tp->refcnt))
459 tcf_proto_destroy(tp, rtnl_held, true, extack);
460}
461
462static bool tcf_proto_check_delete(struct tcf_proto *tp)
463{
464 if (tp->ops->delete_empty)
465 return tp->ops->delete_empty(tp);
466
467 tp->deleting = true;
468 return tp->deleting;
469}
470
471static void tcf_proto_mark_delete(struct tcf_proto *tp)
472{
473 spin_lock(&tp->lock);
474 tp->deleting = true;
475 spin_unlock(&tp->lock);
476}
477
478static bool tcf_proto_is_deleting(struct tcf_proto *tp)
479{
480 bool deleting;
481
482 spin_lock(&tp->lock);
483 deleting = tp->deleting;
484 spin_unlock(&tp->lock);
485
486 return deleting;
487}
488
489#define ASSERT_BLOCK_LOCKED(block) \
490 lockdep_assert_held(&(block)->lock)
491
492struct tcf_filter_chain_list_item {
493 struct list_head list;
494 tcf_chain_head_change_t *chain_head_change;
495 void *chain_head_change_priv;
496};
497
498static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
499 u32 chain_index)
500{
501 struct tcf_chain *chain;
502
503 ASSERT_BLOCK_LOCKED(block);
504
505 chain = kzalloc(sizeof(*chain), GFP_KERNEL);
506 if (!chain)
507 return NULL;
508 list_add_tail_rcu(&chain->list, &block->chain_list);
509 mutex_init(&chain->filter_chain_lock);
510 chain->block = block;
511 chain->index = chain_index;
512 chain->refcnt = 1;
513 if (!chain->index)
514 block->chain0.chain = chain;
515 return chain;
516}
517
518static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
519 struct tcf_proto *tp_head)
520{
521 if (item->chain_head_change)
522 item->chain_head_change(tp_head, item->chain_head_change_priv);
523}
524
525static void tcf_chain0_head_change(struct tcf_chain *chain,
526 struct tcf_proto *tp_head)
527{
528 struct tcf_filter_chain_list_item *item;
529 struct tcf_block *block = chain->block;
530
531 if (chain->index)
532 return;
533
534 mutex_lock(&block->lock);
535 list_for_each_entry(item, &block->chain0.filter_chain_list, list)
536 tcf_chain_head_change_item(item, tp_head);
537 mutex_unlock(&block->lock);
538}
539
540/* Returns true if block can be safely freed. */
541
542static bool tcf_chain_detach(struct tcf_chain *chain)
543{
544 struct tcf_block *block = chain->block;
545
546 ASSERT_BLOCK_LOCKED(block);
547
548 list_del_rcu(&chain->list);
549 if (!chain->index)
550 block->chain0.chain = NULL;
551
552 if (list_empty(&block->chain_list) &&
553 refcount_read(&block->refcnt) == 0)
554 return true;
555
556 return false;
557}
558
559static void tcf_block_destroy(struct tcf_block *block)
560{
561 mutex_destroy(&block->lock);
562 mutex_destroy(&block->proto_destroy_lock);
563 xa_destroy(&block->ports);
564 kfree_rcu(block, rcu);
565}
566
567static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
568{
569 struct tcf_block *block = chain->block;
570
571 mutex_destroy(&chain->filter_chain_lock);
572 kfree_rcu(chain, rcu);
573 if (free_block)
574 tcf_block_destroy(block);
575}
576
577static void tcf_chain_hold(struct tcf_chain *chain)
578{
579 ASSERT_BLOCK_LOCKED(chain->block);
580
581 ++chain->refcnt;
582}
583
584static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
585{
586 ASSERT_BLOCK_LOCKED(chain->block);
587
588 /* In case all the references are action references, this
589 * chain should not be shown to the user.
590 */
591 return chain->refcnt == chain->action_refcnt;
592}
593
594static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
595 u32 chain_index)
596{
597 struct tcf_chain *chain;
598
599 ASSERT_BLOCK_LOCKED(block);
600
601 list_for_each_entry(chain, &block->chain_list, list) {
602 if (chain->index == chain_index)
603 return chain;
604 }
605 return NULL;
606}
607
608#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
609static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
610 u32 chain_index)
611{
612 struct tcf_chain *chain;
613
614 list_for_each_entry_rcu(chain, &block->chain_list, list) {
615 if (chain->index == chain_index)
616 return chain;
617 }
618 return NULL;
619}
620#endif
621
622static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
623 u32 seq, u16 flags, int event, bool unicast,
624 struct netlink_ext_ack *extack);
625
626static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
627 u32 chain_index, bool create,
628 bool by_act)
629{
630 struct tcf_chain *chain = NULL;
631 bool is_first_reference;
632
633 mutex_lock(&block->lock);
634 chain = tcf_chain_lookup(block, chain_index);
635 if (chain) {
636 tcf_chain_hold(chain);
637 } else {
638 if (!create)
639 goto errout;
640 chain = tcf_chain_create(block, chain_index);
641 if (!chain)
642 goto errout;
643 }
644
645 if (by_act)
646 ++chain->action_refcnt;
647 is_first_reference = chain->refcnt - chain->action_refcnt == 1;
648 mutex_unlock(&block->lock);
649
650 /* Send notification only in case we got the first
651 * non-action reference. Until then, the chain acts only as
652 * a placeholder for actions pointing to it and user ought
653 * not know about them.
654 */
655 if (is_first_reference && !by_act)
656 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
657 RTM_NEWCHAIN, false, NULL);
658
659 return chain;
660
661errout:
662 mutex_unlock(&block->lock);
663 return chain;
664}
665
666static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
667 bool create)
668{
669 return __tcf_chain_get(block, chain_index, create, false);
670}
671
672struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
673{
674 return __tcf_chain_get(block, chain_index, true, true);
675}
676EXPORT_SYMBOL(tcf_chain_get_by_act);
677
678static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
679 void *tmplt_priv);
680static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
681 void *tmplt_priv, u32 chain_index,
682 struct tcf_block *block, struct sk_buff *oskb,
683 u32 seq, u16 flags);
684
685static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
686 bool explicitly_created)
687{
688 struct tcf_block *block = chain->block;
689 const struct tcf_proto_ops *tmplt_ops;
690 unsigned int refcnt, non_act_refcnt;
691 bool free_block = false;
692 void *tmplt_priv;
693
694 mutex_lock(&block->lock);
695 if (explicitly_created) {
696 if (!chain->explicitly_created) {
697 mutex_unlock(&block->lock);
698 return;
699 }
700 chain->explicitly_created = false;
701 }
702
703 if (by_act)
704 chain->action_refcnt--;
705
706 /* tc_chain_notify_delete can't be called while holding block lock.
707 * However, when block is unlocked chain can be changed concurrently, so
708 * save these to temporary variables.
709 */
710 refcnt = --chain->refcnt;
711 non_act_refcnt = refcnt - chain->action_refcnt;
712 tmplt_ops = chain->tmplt_ops;
713 tmplt_priv = chain->tmplt_priv;
714
715 if (non_act_refcnt == chain->explicitly_created && !by_act) {
716 if (non_act_refcnt == 0)
717 tc_chain_notify_delete(tmplt_ops, tmplt_priv,
718 chain->index, block, NULL, 0, 0);
719 /* Last reference to chain, no need to lock. */
720 chain->flushing = false;
721 }
722
723 if (refcnt == 0)
724 free_block = tcf_chain_detach(chain);
725 mutex_unlock(&block->lock);
726
727 if (refcnt == 0) {
728 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
729 tcf_chain_destroy(chain, free_block);
730 }
731}
732
733static void tcf_chain_put(struct tcf_chain *chain)
734{
735 __tcf_chain_put(chain, false, false);
736}
737
738void tcf_chain_put_by_act(struct tcf_chain *chain)
739{
740 __tcf_chain_put(chain, true, false);
741}
742EXPORT_SYMBOL(tcf_chain_put_by_act);
743
744static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
745{
746 __tcf_chain_put(chain, false, true);
747}
748
749static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
750{
751 struct tcf_proto *tp, *tp_next;
752
753 mutex_lock(&chain->filter_chain_lock);
754 tp = tcf_chain_dereference(chain->filter_chain, chain);
755 while (tp) {
756 tp_next = rcu_dereference_protected(tp->next, 1);
757 tcf_proto_signal_destroying(chain, tp);
758 tp = tp_next;
759 }
760 tp = tcf_chain_dereference(chain->filter_chain, chain);
761 RCU_INIT_POINTER(chain->filter_chain, NULL);
762 tcf_chain0_head_change(chain, NULL);
763 chain->flushing = true;
764 mutex_unlock(&chain->filter_chain_lock);
765
766 while (tp) {
767 tp_next = rcu_dereference_protected(tp->next, 1);
768 tcf_proto_put(tp, rtnl_held, NULL);
769 tp = tp_next;
770 }
771}
772
773static int tcf_block_setup(struct tcf_block *block,
774 struct flow_block_offload *bo);
775
776static void tcf_block_offload_init(struct flow_block_offload *bo,
777 struct net_device *dev, struct Qdisc *sch,
778 enum flow_block_command command,
779 enum flow_block_binder_type binder_type,
780 struct flow_block *flow_block,
781 bool shared, struct netlink_ext_ack *extack)
782{
783 bo->net = dev_net(dev);
784 bo->command = command;
785 bo->binder_type = binder_type;
786 bo->block = flow_block;
787 bo->block_shared = shared;
788 bo->extack = extack;
789 bo->sch = sch;
790 bo->cb_list_head = &flow_block->cb_list;
791 INIT_LIST_HEAD(&bo->cb_list);
792}
793
794static void tcf_block_unbind(struct tcf_block *block,
795 struct flow_block_offload *bo);
796
797static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
798{
799 struct tcf_block *block = block_cb->indr.data;
800 struct net_device *dev = block_cb->indr.dev;
801 struct Qdisc *sch = block_cb->indr.sch;
802 struct netlink_ext_ack extack = {};
803 struct flow_block_offload bo = {};
804
805 tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
806 block_cb->indr.binder_type,
807 &block->flow_block, tcf_block_shared(block),
808 &extack);
809 rtnl_lock();
810 down_write(&block->cb_lock);
811 list_del(&block_cb->driver_list);
812 list_move(&block_cb->list, &bo.cb_list);
813 tcf_block_unbind(block, &bo);
814 up_write(&block->cb_lock);
815 rtnl_unlock();
816}
817
818static bool tcf_block_offload_in_use(struct tcf_block *block)
819{
820 return atomic_read(&block->offloadcnt);
821}
822
823static int tcf_block_offload_cmd(struct tcf_block *block,
824 struct net_device *dev, struct Qdisc *sch,
825 struct tcf_block_ext_info *ei,
826 enum flow_block_command command,
827 struct netlink_ext_ack *extack)
828{
829 struct flow_block_offload bo = {};
830
831 tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
832 &block->flow_block, tcf_block_shared(block),
833 extack);
834
835 if (dev->netdev_ops->ndo_setup_tc) {
836 int err;
837
838 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
839 if (err < 0) {
840 if (err != -EOPNOTSUPP)
841 NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
842 return err;
843 }
844
845 return tcf_block_setup(block, &bo);
846 }
847
848 flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
849 tc_block_indr_cleanup);
850 tcf_block_setup(block, &bo);
851
852 return -EOPNOTSUPP;
853}
854
855static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
856 struct tcf_block_ext_info *ei,
857 struct netlink_ext_ack *extack)
858{
859 struct net_device *dev = q->dev_queue->dev;
860 int err;
861
862 down_write(&block->cb_lock);
863
864 /* If tc offload feature is disabled and the block we try to bind
865 * to already has some offloaded filters, forbid to bind.
866 */
867 if (dev->netdev_ops->ndo_setup_tc &&
868 !tc_can_offload(dev) &&
869 tcf_block_offload_in_use(block)) {
870 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
871 err = -EOPNOTSUPP;
872 goto err_unlock;
873 }
874
875 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
876 if (err == -EOPNOTSUPP)
877 goto no_offload_dev_inc;
878 if (err)
879 goto err_unlock;
880
881 up_write(&block->cb_lock);
882 return 0;
883
884no_offload_dev_inc:
885 if (tcf_block_offload_in_use(block))
886 goto err_unlock;
887
888 err = 0;
889 block->nooffloaddevcnt++;
890err_unlock:
891 up_write(&block->cb_lock);
892 return err;
893}
894
895static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
896 struct tcf_block_ext_info *ei)
897{
898 struct net_device *dev = q->dev_queue->dev;
899 int err;
900
901 down_write(&block->cb_lock);
902 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
903 if (err == -EOPNOTSUPP)
904 goto no_offload_dev_dec;
905 up_write(&block->cb_lock);
906 return;
907
908no_offload_dev_dec:
909 WARN_ON(block->nooffloaddevcnt-- == 0);
910 up_write(&block->cb_lock);
911}
912
913static int
914tcf_chain0_head_change_cb_add(struct tcf_block *block,
915 struct tcf_block_ext_info *ei,
916 struct netlink_ext_ack *extack)
917{
918 struct tcf_filter_chain_list_item *item;
919 struct tcf_chain *chain0;
920
921 item = kmalloc(sizeof(*item), GFP_KERNEL);
922 if (!item) {
923 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
924 return -ENOMEM;
925 }
926 item->chain_head_change = ei->chain_head_change;
927 item->chain_head_change_priv = ei->chain_head_change_priv;
928
929 mutex_lock(&block->lock);
930 chain0 = block->chain0.chain;
931 if (chain0)
932 tcf_chain_hold(chain0);
933 else
934 list_add(&item->list, &block->chain0.filter_chain_list);
935 mutex_unlock(&block->lock);
936
937 if (chain0) {
938 struct tcf_proto *tp_head;
939
940 mutex_lock(&chain0->filter_chain_lock);
941
942 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
943 if (tp_head)
944 tcf_chain_head_change_item(item, tp_head);
945
946 mutex_lock(&block->lock);
947 list_add(&item->list, &block->chain0.filter_chain_list);
948 mutex_unlock(&block->lock);
949
950 mutex_unlock(&chain0->filter_chain_lock);
951 tcf_chain_put(chain0);
952 }
953
954 return 0;
955}
956
957static void
958tcf_chain0_head_change_cb_del(struct tcf_block *block,
959 struct tcf_block_ext_info *ei)
960{
961 struct tcf_filter_chain_list_item *item;
962
963 mutex_lock(&block->lock);
964 list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
965 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
966 (item->chain_head_change == ei->chain_head_change &&
967 item->chain_head_change_priv == ei->chain_head_change_priv)) {
968 if (block->chain0.chain)
969 tcf_chain_head_change_item(item, NULL);
970 list_del(&item->list);
971 mutex_unlock(&block->lock);
972
973 kfree(item);
974 return;
975 }
976 }
977 mutex_unlock(&block->lock);
978 WARN_ON(1);
979}
980
981struct tcf_net {
982 spinlock_t idr_lock; /* Protects idr */
983 struct idr idr;
984};
985
986static unsigned int tcf_net_id;
987
988static int tcf_block_insert(struct tcf_block *block, struct net *net,
989 struct netlink_ext_ack *extack)
990{
991 struct tcf_net *tn = net_generic(net, tcf_net_id);
992 int err;
993
994 idr_preload(GFP_KERNEL);
995 spin_lock(&tn->idr_lock);
996 err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
997 GFP_NOWAIT);
998 spin_unlock(&tn->idr_lock);
999 idr_preload_end();
1000
1001 return err;
1002}
1003
1004static void tcf_block_remove(struct tcf_block *block, struct net *net)
1005{
1006 struct tcf_net *tn = net_generic(net, tcf_net_id);
1007
1008 spin_lock(&tn->idr_lock);
1009 idr_remove(&tn->idr, block->index);
1010 spin_unlock(&tn->idr_lock);
1011}
1012
1013static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
1014 u32 block_index,
1015 struct netlink_ext_ack *extack)
1016{
1017 struct tcf_block *block;
1018
1019 block = kzalloc(sizeof(*block), GFP_KERNEL);
1020 if (!block) {
1021 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
1022 return ERR_PTR(-ENOMEM);
1023 }
1024 mutex_init(&block->lock);
1025 mutex_init(&block->proto_destroy_lock);
1026 init_rwsem(&block->cb_lock);
1027 flow_block_init(&block->flow_block);
1028 INIT_LIST_HEAD(&block->chain_list);
1029 INIT_LIST_HEAD(&block->owner_list);
1030 INIT_LIST_HEAD(&block->chain0.filter_chain_list);
1031
1032 refcount_set(&block->refcnt, 1);
1033 block->net = net;
1034 block->index = block_index;
1035 xa_init(&block->ports);
1036
1037 /* Don't store q pointer for blocks which are shared */
1038 if (!tcf_block_shared(block))
1039 block->q = q;
1040 return block;
1041}
1042
1043struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
1044{
1045 struct tcf_net *tn = net_generic(net, tcf_net_id);
1046
1047 return idr_find(&tn->idr, block_index);
1048}
1049EXPORT_SYMBOL(tcf_block_lookup);
1050
1051static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
1052{
1053 struct tcf_block *block;
1054
1055 rcu_read_lock();
1056 block = tcf_block_lookup(net, block_index);
1057 if (block && !refcount_inc_not_zero(&block->refcnt))
1058 block = NULL;
1059 rcu_read_unlock();
1060
1061 return block;
1062}
1063
1064static struct tcf_chain *
1065__tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1066{
1067 mutex_lock(&block->lock);
1068 if (chain)
1069 chain = list_is_last(&chain->list, &block->chain_list) ?
1070 NULL : list_next_entry(chain, list);
1071 else
1072 chain = list_first_entry_or_null(&block->chain_list,
1073 struct tcf_chain, list);
1074
1075 /* skip all action-only chains */
1076 while (chain && tcf_chain_held_by_acts_only(chain))
1077 chain = list_is_last(&chain->list, &block->chain_list) ?
1078 NULL : list_next_entry(chain, list);
1079
1080 if (chain)
1081 tcf_chain_hold(chain);
1082 mutex_unlock(&block->lock);
1083
1084 return chain;
1085}
1086
1087/* Function to be used by all clients that want to iterate over all chains on
1088 * block. It properly obtains block->lock and takes reference to chain before
1089 * returning it. Users of this function must be tolerant to concurrent chain
1090 * insertion/deletion or ensure that no concurrent chain modification is
1091 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1092 * consistent dump because rtnl lock is released each time skb is filled with
1093 * data and sent to user-space.
1094 */
1095
1096struct tcf_chain *
1097tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1098{
1099 struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
1100
1101 if (chain)
1102 tcf_chain_put(chain);
1103
1104 return chain_next;
1105}
1106EXPORT_SYMBOL(tcf_get_next_chain);
1107
1108static struct tcf_proto *
1109__tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1110{
1111 u32 prio = 0;
1112
1113 ASSERT_RTNL();
1114 mutex_lock(&chain->filter_chain_lock);
1115
1116 if (!tp) {
1117 tp = tcf_chain_dereference(chain->filter_chain, chain);
1118 } else if (tcf_proto_is_deleting(tp)) {
1119 /* 'deleting' flag is set and chain->filter_chain_lock was
1120 * unlocked, which means next pointer could be invalid. Restart
1121 * search.
1122 */
1123 prio = tp->prio + 1;
1124 tp = tcf_chain_dereference(chain->filter_chain, chain);
1125
1126 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1127 if (!tp->deleting && tp->prio >= prio)
1128 break;
1129 } else {
1130 tp = tcf_chain_dereference(tp->next, chain);
1131 }
1132
1133 if (tp)
1134 tcf_proto_get(tp);
1135
1136 mutex_unlock(&chain->filter_chain_lock);
1137
1138 return tp;
1139}
1140
1141/* Function to be used by all clients that want to iterate over all tp's on
1142 * chain. Users of this function must be tolerant to concurrent tp
1143 * insertion/deletion or ensure that no concurrent chain modification is
1144 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1145 * consistent dump because rtnl lock is released each time skb is filled with
1146 * data and sent to user-space.
1147 */
1148
1149struct tcf_proto *
1150tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1151{
1152 struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1153
1154 if (tp)
1155 tcf_proto_put(tp, true, NULL);
1156
1157 return tp_next;
1158}
1159EXPORT_SYMBOL(tcf_get_next_proto);
1160
1161static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1162{
1163 struct tcf_chain *chain;
1164
1165 /* Last reference to block. At this point chains cannot be added or
1166 * removed concurrently.
1167 */
1168 for (chain = tcf_get_next_chain(block, NULL);
1169 chain;
1170 chain = tcf_get_next_chain(block, chain)) {
1171 tcf_chain_put_explicitly_created(chain);
1172 tcf_chain_flush(chain, rtnl_held);
1173 }
1174}
1175
1176/* Lookup Qdisc and increments its reference counter.
1177 * Set parent, if necessary.
1178 */
1179
1180static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1181 u32 *parent, int ifindex, bool rtnl_held,
1182 struct netlink_ext_ack *extack)
1183{
1184 const struct Qdisc_class_ops *cops;
1185 struct net_device *dev;
1186 int err = 0;
1187
1188 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1189 return 0;
1190
1191 rcu_read_lock();
1192
1193 /* Find link */
1194 dev = dev_get_by_index_rcu(net, ifindex);
1195 if (!dev) {
1196 rcu_read_unlock();
1197 return -ENODEV;
1198 }
1199
1200 /* Find qdisc */
1201 if (!*parent) {
1202 *q = rcu_dereference(dev->qdisc);
1203 *parent = (*q)->handle;
1204 } else {
1205 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1206 if (!*q) {
1207 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1208 err = -EINVAL;
1209 goto errout_rcu;
1210 }
1211 }
1212
1213 *q = qdisc_refcount_inc_nz(*q);
1214 if (!*q) {
1215 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1216 err = -EINVAL;
1217 goto errout_rcu;
1218 }
1219
1220 /* Is it classful? */
1221 cops = (*q)->ops->cl_ops;
1222 if (!cops) {
1223 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1224 err = -EINVAL;
1225 goto errout_qdisc;
1226 }
1227
1228 if (!cops->tcf_block) {
1229 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1230 err = -EOPNOTSUPP;
1231 goto errout_qdisc;
1232 }
1233
1234errout_rcu:
1235 /* At this point we know that qdisc is not noop_qdisc,
1236 * which means that qdisc holds a reference to net_device
1237 * and we hold a reference to qdisc, so it is safe to release
1238 * rcu read lock.
1239 */
1240 rcu_read_unlock();
1241 return err;
1242
1243errout_qdisc:
1244 rcu_read_unlock();
1245
1246 if (rtnl_held)
1247 qdisc_put(*q);
1248 else
1249 qdisc_put_unlocked(*q);
1250 *q = NULL;
1251
1252 return err;
1253}
1254
1255static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1256 int ifindex, struct netlink_ext_ack *extack)
1257{
1258 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1259 return 0;
1260
1261 /* Do we search for filter, attached to class? */
1262 if (TC_H_MIN(parent)) {
1263 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1264
1265 *cl = cops->find(q, parent);
1266 if (*cl == 0) {
1267 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1268 return -ENOENT;
1269 }
1270 }
1271
1272 return 0;
1273}
1274
1275static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1276 unsigned long cl, int ifindex,
1277 u32 block_index,
1278 struct netlink_ext_ack *extack)
1279{
1280 struct tcf_block *block;
1281
1282 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1283 block = tcf_block_refcnt_get(net, block_index);
1284 if (!block) {
1285 NL_SET_ERR_MSG(extack, "Block of given index was not found");
1286 return ERR_PTR(-EINVAL);
1287 }
1288 } else {
1289 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1290
1291 block = cops->tcf_block(q, cl, extack);
1292 if (!block)
1293 return ERR_PTR(-EINVAL);
1294
1295 if (tcf_block_shared(block)) {
1296 NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1297 return ERR_PTR(-EOPNOTSUPP);
1298 }
1299
1300 /* Always take reference to block in order to support execution
1301 * of rules update path of cls API without rtnl lock. Caller
1302 * must release block when it is finished using it. 'if' block
1303 * of this conditional obtain reference to block by calling
1304 * tcf_block_refcnt_get().
1305 */
1306 refcount_inc(&block->refcnt);
1307 }
1308
1309 return block;
1310}
1311
1312static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1313 struct tcf_block_ext_info *ei, bool rtnl_held)
1314{
1315 if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1316 /* Flushing/putting all chains will cause the block to be
1317 * deallocated when last chain is freed. However, if chain_list
1318 * is empty, block has to be manually deallocated. After block
1319 * reference counter reached 0, it is no longer possible to
1320 * increment it or add new chains to block.
1321 */
1322 bool free_block = list_empty(&block->chain_list);
1323
1324 mutex_unlock(&block->lock);
1325 if (tcf_block_shared(block))
1326 tcf_block_remove(block, block->net);
1327
1328 if (q)
1329 tcf_block_offload_unbind(block, q, ei);
1330
1331 if (free_block)
1332 tcf_block_destroy(block);
1333 else
1334 tcf_block_flush_all_chains(block, rtnl_held);
1335 } else if (q) {
1336 tcf_block_offload_unbind(block, q, ei);
1337 }
1338}
1339
1340static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1341{
1342 __tcf_block_put(block, NULL, NULL, rtnl_held);
1343}
1344
1345/* Find tcf block.
1346 * Set q, parent, cl when appropriate.
1347 */
1348
1349static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1350 u32 *parent, unsigned long *cl,
1351 int ifindex, u32 block_index,
1352 struct netlink_ext_ack *extack)
1353{
1354 struct tcf_block *block;
1355 int err = 0;
1356
1357 ASSERT_RTNL();
1358
1359 err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1360 if (err)
1361 goto errout;
1362
1363 err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1364 if (err)
1365 goto errout_qdisc;
1366
1367 block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1368 if (IS_ERR(block)) {
1369 err = PTR_ERR(block);
1370 goto errout_qdisc;
1371 }
1372
1373 return block;
1374
1375errout_qdisc:
1376 if (*q)
1377 qdisc_put(*q);
1378errout:
1379 *q = NULL;
1380 return ERR_PTR(err);
1381}
1382
1383static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1384 bool rtnl_held)
1385{
1386 if (!IS_ERR_OR_NULL(block))
1387 tcf_block_refcnt_put(block, rtnl_held);
1388
1389 if (q) {
1390 if (rtnl_held)
1391 qdisc_put(q);
1392 else
1393 qdisc_put_unlocked(q);
1394 }
1395}
1396
1397struct tcf_block_owner_item {
1398 struct list_head list;
1399 struct Qdisc *q;
1400 enum flow_block_binder_type binder_type;
1401};
1402
1403static void
1404tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1405 struct Qdisc *q,
1406 enum flow_block_binder_type binder_type)
1407{
1408 if (block->keep_dst &&
1409 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1410 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1411 netif_keep_dst(qdisc_dev(q));
1412}
1413
1414void tcf_block_netif_keep_dst(struct tcf_block *block)
1415{
1416 struct tcf_block_owner_item *item;
1417
1418 block->keep_dst = true;
1419 list_for_each_entry(item, &block->owner_list, list)
1420 tcf_block_owner_netif_keep_dst(block, item->q,
1421 item->binder_type);
1422}
1423EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1424
1425static int tcf_block_owner_add(struct tcf_block *block,
1426 struct Qdisc *q,
1427 enum flow_block_binder_type binder_type)
1428{
1429 struct tcf_block_owner_item *item;
1430
1431 item = kmalloc(sizeof(*item), GFP_KERNEL);
1432 if (!item)
1433 return -ENOMEM;
1434 item->q = q;
1435 item->binder_type = binder_type;
1436 list_add(&item->list, &block->owner_list);
1437 return 0;
1438}
1439
1440static void tcf_block_owner_del(struct tcf_block *block,
1441 struct Qdisc *q,
1442 enum flow_block_binder_type binder_type)
1443{
1444 struct tcf_block_owner_item *item;
1445
1446 list_for_each_entry(item, &block->owner_list, list) {
1447 if (item->q == q && item->binder_type == binder_type) {
1448 list_del(&item->list);
1449 kfree(item);
1450 return;
1451 }
1452 }
1453 WARN_ON(1);
1454}
1455
1456static bool tcf_block_tracks_dev(struct tcf_block *block,
1457 struct tcf_block_ext_info *ei)
1458{
1459 return tcf_block_shared(block) &&
1460 (ei->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS ||
1461 ei->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS);
1462}
1463
1464int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1465 struct tcf_block_ext_info *ei,
1466 struct netlink_ext_ack *extack)
1467{
1468 struct net_device *dev = qdisc_dev(q);
1469 struct net *net = qdisc_net(q);
1470 struct tcf_block *block = NULL;
1471 int err;
1472
1473 if (ei->block_index)
1474 /* block_index not 0 means the shared block is requested */
1475 block = tcf_block_refcnt_get(net, ei->block_index);
1476
1477 if (!block) {
1478 block = tcf_block_create(net, q, ei->block_index, extack);
1479 if (IS_ERR(block))
1480 return PTR_ERR(block);
1481 if (tcf_block_shared(block)) {
1482 err = tcf_block_insert(block, net, extack);
1483 if (err)
1484 goto err_block_insert;
1485 }
1486 }
1487
1488 err = tcf_block_owner_add(block, q, ei->binder_type);
1489 if (err)
1490 goto err_block_owner_add;
1491
1492 tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1493
1494 err = tcf_chain0_head_change_cb_add(block, ei, extack);
1495 if (err)
1496 goto err_chain0_head_change_cb_add;
1497
1498 err = tcf_block_offload_bind(block, q, ei, extack);
1499 if (err)
1500 goto err_block_offload_bind;
1501
1502 if (tcf_block_tracks_dev(block, ei)) {
1503 err = xa_insert(&block->ports, dev->ifindex, dev, GFP_KERNEL);
1504 if (err) {
1505 NL_SET_ERR_MSG(extack, "block dev insert failed");
1506 goto err_dev_insert;
1507 }
1508 }
1509
1510 *p_block = block;
1511 return 0;
1512
1513err_dev_insert:
1514 tcf_block_offload_unbind(block, q, ei);
1515err_block_offload_bind:
1516 tcf_chain0_head_change_cb_del(block, ei);
1517err_chain0_head_change_cb_add:
1518 tcf_block_owner_del(block, q, ei->binder_type);
1519err_block_owner_add:
1520err_block_insert:
1521 tcf_block_refcnt_put(block, true);
1522 return err;
1523}
1524EXPORT_SYMBOL(tcf_block_get_ext);
1525
1526static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1527{
1528 struct tcf_proto __rcu **p_filter_chain = priv;
1529
1530 rcu_assign_pointer(*p_filter_chain, tp_head);
1531}
1532
1533int tcf_block_get(struct tcf_block **p_block,
1534 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1535 struct netlink_ext_ack *extack)
1536{
1537 struct tcf_block_ext_info ei = {
1538 .chain_head_change = tcf_chain_head_change_dflt,
1539 .chain_head_change_priv = p_filter_chain,
1540 };
1541
1542 WARN_ON(!p_filter_chain);
1543 return tcf_block_get_ext(p_block, q, &ei, extack);
1544}
1545EXPORT_SYMBOL(tcf_block_get);
1546
1547/* XXX: Standalone actions are not allowed to jump to any chain, and bound
1548 * actions should be all removed after flushing.
1549 */
1550void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1551 struct tcf_block_ext_info *ei)
1552{
1553 struct net_device *dev = qdisc_dev(q);
1554
1555 if (!block)
1556 return;
1557 if (tcf_block_tracks_dev(block, ei))
1558 xa_erase(&block->ports, dev->ifindex);
1559 tcf_chain0_head_change_cb_del(block, ei);
1560 tcf_block_owner_del(block, q, ei->binder_type);
1561
1562 __tcf_block_put(block, q, ei, true);
1563}
1564EXPORT_SYMBOL(tcf_block_put_ext);
1565
1566void tcf_block_put(struct tcf_block *block)
1567{
1568 struct tcf_block_ext_info ei = {0, };
1569
1570 if (!block)
1571 return;
1572 tcf_block_put_ext(block, block->q, &ei);
1573}
1574
1575EXPORT_SYMBOL(tcf_block_put);
1576
1577static int
1578tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1579 void *cb_priv, bool add, bool offload_in_use,
1580 struct netlink_ext_ack *extack)
1581{
1582 struct tcf_chain *chain, *chain_prev;
1583 struct tcf_proto *tp, *tp_prev;
1584 int err;
1585
1586 lockdep_assert_held(&block->cb_lock);
1587
1588 for (chain = __tcf_get_next_chain(block, NULL);
1589 chain;
1590 chain_prev = chain,
1591 chain = __tcf_get_next_chain(block, chain),
1592 tcf_chain_put(chain_prev)) {
1593 if (chain->tmplt_ops && add)
1594 chain->tmplt_ops->tmplt_reoffload(chain, true, cb,
1595 cb_priv);
1596 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1597 tp_prev = tp,
1598 tp = __tcf_get_next_proto(chain, tp),
1599 tcf_proto_put(tp_prev, true, NULL)) {
1600 if (tp->ops->reoffload) {
1601 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1602 extack);
1603 if (err && add)
1604 goto err_playback_remove;
1605 } else if (add && offload_in_use) {
1606 err = -EOPNOTSUPP;
1607 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1608 goto err_playback_remove;
1609 }
1610 }
1611 if (chain->tmplt_ops && !add)
1612 chain->tmplt_ops->tmplt_reoffload(chain, false, cb,
1613 cb_priv);
1614 }
1615
1616 return 0;
1617
1618err_playback_remove:
1619 tcf_proto_put(tp, true, NULL);
1620 tcf_chain_put(chain);
1621 tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1622 extack);
1623 return err;
1624}
1625
1626static int tcf_block_bind(struct tcf_block *block,
1627 struct flow_block_offload *bo)
1628{
1629 struct flow_block_cb *block_cb, *next;
1630 int err, i = 0;
1631
1632 lockdep_assert_held(&block->cb_lock);
1633
1634 list_for_each_entry(block_cb, &bo->cb_list, list) {
1635 err = tcf_block_playback_offloads(block, block_cb->cb,
1636 block_cb->cb_priv, true,
1637 tcf_block_offload_in_use(block),
1638 bo->extack);
1639 if (err)
1640 goto err_unroll;
1641 if (!bo->unlocked_driver_cb)
1642 block->lockeddevcnt++;
1643
1644 i++;
1645 }
1646 list_splice(&bo->cb_list, &block->flow_block.cb_list);
1647
1648 return 0;
1649
1650err_unroll:
1651 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1652 list_del(&block_cb->driver_list);
1653 if (i-- > 0) {
1654 list_del(&block_cb->list);
1655 tcf_block_playback_offloads(block, block_cb->cb,
1656 block_cb->cb_priv, false,
1657 tcf_block_offload_in_use(block),
1658 NULL);
1659 if (!bo->unlocked_driver_cb)
1660 block->lockeddevcnt--;
1661 }
1662 flow_block_cb_free(block_cb);
1663 }
1664
1665 return err;
1666}
1667
1668static void tcf_block_unbind(struct tcf_block *block,
1669 struct flow_block_offload *bo)
1670{
1671 struct flow_block_cb *block_cb, *next;
1672
1673 lockdep_assert_held(&block->cb_lock);
1674
1675 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1676 tcf_block_playback_offloads(block, block_cb->cb,
1677 block_cb->cb_priv, false,
1678 tcf_block_offload_in_use(block),
1679 NULL);
1680 list_del(&block_cb->list);
1681 flow_block_cb_free(block_cb);
1682 if (!bo->unlocked_driver_cb)
1683 block->lockeddevcnt--;
1684 }
1685}
1686
1687static int tcf_block_setup(struct tcf_block *block,
1688 struct flow_block_offload *bo)
1689{
1690 int err;
1691
1692 switch (bo->command) {
1693 case FLOW_BLOCK_BIND:
1694 err = tcf_block_bind(block, bo);
1695 break;
1696 case FLOW_BLOCK_UNBIND:
1697 err = 0;
1698 tcf_block_unbind(block, bo);
1699 break;
1700 default:
1701 WARN_ON_ONCE(1);
1702 err = -EOPNOTSUPP;
1703 }
1704
1705 return err;
1706}
1707
1708/* Main classifier routine: scans classifier chain attached
1709 * to this qdisc, (optionally) tests for protocol and asks
1710 * specific classifiers.
1711 */
1712static inline int __tcf_classify(struct sk_buff *skb,
1713 const struct tcf_proto *tp,
1714 const struct tcf_proto *orig_tp,
1715 struct tcf_result *res,
1716 bool compat_mode,
1717 struct tcf_exts_miss_cookie_node *n,
1718 int act_index,
1719 u32 *last_executed_chain)
1720{
1721#ifdef CONFIG_NET_CLS_ACT
1722 const int max_reclassify_loop = 16;
1723 const struct tcf_proto *first_tp;
1724 int limit = 0;
1725
1726reclassify:
1727#endif
1728 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1729 __be16 protocol = skb_protocol(skb, false);
1730 int err = 0;
1731
1732 if (n) {
1733 struct tcf_exts *exts;
1734
1735 if (n->tp_prio != tp->prio)
1736 continue;
1737
1738 /* We re-lookup the tp and chain based on index instead
1739 * of having hard refs and locks to them, so do a sanity
1740 * check if any of tp,chain,exts was replaced by the
1741 * time we got here with a cookie from hardware.
1742 */
1743 if (unlikely(n->tp != tp || n->tp->chain != n->chain ||
1744 !tp->ops->get_exts)) {
1745 tcf_set_drop_reason(skb,
1746 SKB_DROP_REASON_TC_COOKIE_ERROR);
1747 return TC_ACT_SHOT;
1748 }
1749
1750 exts = tp->ops->get_exts(tp, n->handle);
1751 if (unlikely(!exts || n->exts != exts)) {
1752 tcf_set_drop_reason(skb,
1753 SKB_DROP_REASON_TC_COOKIE_ERROR);
1754 return TC_ACT_SHOT;
1755 }
1756
1757 n = NULL;
1758 err = tcf_exts_exec_ex(skb, exts, act_index, res);
1759 } else {
1760 if (tp->protocol != protocol &&
1761 tp->protocol != htons(ETH_P_ALL))
1762 continue;
1763
1764 err = tc_classify(skb, tp, res);
1765 }
1766#ifdef CONFIG_NET_CLS_ACT
1767 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1768 first_tp = orig_tp;
1769 *last_executed_chain = first_tp->chain->index;
1770 goto reset;
1771 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1772 first_tp = res->goto_tp;
1773 *last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1774 goto reset;
1775 }
1776#endif
1777 if (err >= 0)
1778 return err;
1779 }
1780
1781 if (unlikely(n)) {
1782 tcf_set_drop_reason(skb,
1783 SKB_DROP_REASON_TC_COOKIE_ERROR);
1784 return TC_ACT_SHOT;
1785 }
1786
1787 return TC_ACT_UNSPEC; /* signal: continue lookup */
1788#ifdef CONFIG_NET_CLS_ACT
1789reset:
1790 if (unlikely(limit++ >= max_reclassify_loop)) {
1791 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1792 tp->chain->block->index,
1793 tp->prio & 0xffff,
1794 ntohs(tp->protocol));
1795 tcf_set_drop_reason(skb,
1796 SKB_DROP_REASON_TC_RECLASSIFY_LOOP);
1797 return TC_ACT_SHOT;
1798 }
1799
1800 tp = first_tp;
1801 goto reclassify;
1802#endif
1803}
1804
1805int tcf_classify(struct sk_buff *skb,
1806 const struct tcf_block *block,
1807 const struct tcf_proto *tp,
1808 struct tcf_result *res, bool compat_mode)
1809{
1810#if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1811 u32 last_executed_chain = 0;
1812
1813 return __tcf_classify(skb, tp, tp, res, compat_mode, NULL, 0,
1814 &last_executed_chain);
1815#else
1816 u32 last_executed_chain = tp ? tp->chain->index : 0;
1817 struct tcf_exts_miss_cookie_node *n = NULL;
1818 const struct tcf_proto *orig_tp = tp;
1819 struct tc_skb_ext *ext;
1820 int act_index = 0;
1821 int ret;
1822
1823 if (block) {
1824 ext = skb_ext_find(skb, TC_SKB_EXT);
1825
1826 if (ext && (ext->chain || ext->act_miss)) {
1827 struct tcf_chain *fchain;
1828 u32 chain;
1829
1830 if (ext->act_miss) {
1831 n = tcf_exts_miss_cookie_lookup(ext->act_miss_cookie,
1832 &act_index);
1833 if (!n) {
1834 tcf_set_drop_reason(skb,
1835 SKB_DROP_REASON_TC_COOKIE_ERROR);
1836 return TC_ACT_SHOT;
1837 }
1838
1839 chain = n->chain_index;
1840 } else {
1841 chain = ext->chain;
1842 }
1843
1844 fchain = tcf_chain_lookup_rcu(block, chain);
1845 if (!fchain) {
1846 tcf_set_drop_reason(skb,
1847 SKB_DROP_REASON_TC_CHAIN_NOTFOUND);
1848
1849 return TC_ACT_SHOT;
1850 }
1851
1852 /* Consume, so cloned/redirect skbs won't inherit ext */
1853 skb_ext_del(skb, TC_SKB_EXT);
1854
1855 tp = rcu_dereference_bh(fchain->filter_chain);
1856 last_executed_chain = fchain->index;
1857 }
1858 }
1859
1860 ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode, n, act_index,
1861 &last_executed_chain);
1862
1863 if (tc_skb_ext_tc_enabled()) {
1864 /* If we missed on some chain */
1865 if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1866 struct tc_skb_cb *cb = tc_skb_cb(skb);
1867
1868 ext = tc_skb_ext_alloc(skb);
1869 if (WARN_ON_ONCE(!ext)) {
1870 tcf_set_drop_reason(skb, SKB_DROP_REASON_NOMEM);
1871 return TC_ACT_SHOT;
1872 }
1873 ext->chain = last_executed_chain;
1874 ext->mru = cb->mru;
1875 ext->post_ct = cb->post_ct;
1876 ext->post_ct_snat = cb->post_ct_snat;
1877 ext->post_ct_dnat = cb->post_ct_dnat;
1878 ext->zone = cb->zone;
1879 }
1880 }
1881
1882 return ret;
1883#endif
1884}
1885EXPORT_SYMBOL(tcf_classify);
1886
1887struct tcf_chain_info {
1888 struct tcf_proto __rcu **pprev;
1889 struct tcf_proto __rcu *next;
1890};
1891
1892static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1893 struct tcf_chain_info *chain_info)
1894{
1895 return tcf_chain_dereference(*chain_info->pprev, chain);
1896}
1897
1898static int tcf_chain_tp_insert(struct tcf_chain *chain,
1899 struct tcf_chain_info *chain_info,
1900 struct tcf_proto *tp)
1901{
1902 if (chain->flushing)
1903 return -EAGAIN;
1904
1905 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1906 if (*chain_info->pprev == chain->filter_chain)
1907 tcf_chain0_head_change(chain, tp);
1908 tcf_proto_get(tp);
1909 rcu_assign_pointer(*chain_info->pprev, tp);
1910
1911 return 0;
1912}
1913
1914static void tcf_chain_tp_remove(struct tcf_chain *chain,
1915 struct tcf_chain_info *chain_info,
1916 struct tcf_proto *tp)
1917{
1918 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1919
1920 tcf_proto_mark_delete(tp);
1921 if (tp == chain->filter_chain)
1922 tcf_chain0_head_change(chain, next);
1923 RCU_INIT_POINTER(*chain_info->pprev, next);
1924}
1925
1926static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1927 struct tcf_chain_info *chain_info,
1928 u32 protocol, u32 prio,
1929 bool prio_allocate,
1930 struct netlink_ext_ack *extack);
1931
1932/* Try to insert new proto.
1933 * If proto with specified priority already exists, free new proto
1934 * and return existing one.
1935 */
1936
1937static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1938 struct tcf_proto *tp_new,
1939 u32 protocol, u32 prio,
1940 bool rtnl_held)
1941{
1942 struct tcf_chain_info chain_info;
1943 struct tcf_proto *tp;
1944 int err = 0;
1945
1946 mutex_lock(&chain->filter_chain_lock);
1947
1948 if (tcf_proto_exists_destroying(chain, tp_new)) {
1949 mutex_unlock(&chain->filter_chain_lock);
1950 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1951 return ERR_PTR(-EAGAIN);
1952 }
1953
1954 tp = tcf_chain_tp_find(chain, &chain_info, protocol, prio, false, NULL);
1955 if (!tp)
1956 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1957 mutex_unlock(&chain->filter_chain_lock);
1958
1959 if (tp) {
1960 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1961 tp_new = tp;
1962 } else if (err) {
1963 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1964 tp_new = ERR_PTR(err);
1965 }
1966
1967 return tp_new;
1968}
1969
1970static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1971 struct tcf_proto *tp, bool rtnl_held,
1972 struct netlink_ext_ack *extack)
1973{
1974 struct tcf_chain_info chain_info;
1975 struct tcf_proto *tp_iter;
1976 struct tcf_proto **pprev;
1977 struct tcf_proto *next;
1978
1979 mutex_lock(&chain->filter_chain_lock);
1980
1981 /* Atomically find and remove tp from chain. */
1982 for (pprev = &chain->filter_chain;
1983 (tp_iter = tcf_chain_dereference(*pprev, chain));
1984 pprev = &tp_iter->next) {
1985 if (tp_iter == tp) {
1986 chain_info.pprev = pprev;
1987 chain_info.next = tp_iter->next;
1988 WARN_ON(tp_iter->deleting);
1989 break;
1990 }
1991 }
1992 /* Verify that tp still exists and no new filters were inserted
1993 * concurrently.
1994 * Mark tp for deletion if it is empty.
1995 */
1996 if (!tp_iter || !tcf_proto_check_delete(tp)) {
1997 mutex_unlock(&chain->filter_chain_lock);
1998 return;
1999 }
2000
2001 tcf_proto_signal_destroying(chain, tp);
2002 next = tcf_chain_dereference(chain_info.next, chain);
2003 if (tp == chain->filter_chain)
2004 tcf_chain0_head_change(chain, next);
2005 RCU_INIT_POINTER(*chain_info.pprev, next);
2006 mutex_unlock(&chain->filter_chain_lock);
2007
2008 tcf_proto_put(tp, rtnl_held, extack);
2009}
2010
2011static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
2012 struct tcf_chain_info *chain_info,
2013 u32 protocol, u32 prio,
2014 bool prio_allocate,
2015 struct netlink_ext_ack *extack)
2016{
2017 struct tcf_proto **pprev;
2018 struct tcf_proto *tp;
2019
2020 /* Check the chain for existence of proto-tcf with this priority */
2021 for (pprev = &chain->filter_chain;
2022 (tp = tcf_chain_dereference(*pprev, chain));
2023 pprev = &tp->next) {
2024 if (tp->prio >= prio) {
2025 if (tp->prio == prio) {
2026 if (prio_allocate) {
2027 NL_SET_ERR_MSG(extack, "Lowest ID from auto-alloc range already in use");
2028 return ERR_PTR(-ENOSPC);
2029 }
2030 if (tp->protocol != protocol && protocol) {
2031 NL_SET_ERR_MSG(extack, "Protocol mismatch for filter with specified priority");
2032 return ERR_PTR(-EINVAL);
2033 }
2034 } else {
2035 tp = NULL;
2036 }
2037 break;
2038 }
2039 }
2040 chain_info->pprev = pprev;
2041 if (tp) {
2042 chain_info->next = tp->next;
2043 tcf_proto_get(tp);
2044 } else {
2045 chain_info->next = NULL;
2046 }
2047 return tp;
2048}
2049
2050static int tcf_fill_node(struct net *net, struct sk_buff *skb,
2051 struct tcf_proto *tp, struct tcf_block *block,
2052 struct Qdisc *q, u32 parent, void *fh,
2053 u32 portid, u32 seq, u16 flags, int event,
2054 bool terse_dump, bool rtnl_held,
2055 struct netlink_ext_ack *extack)
2056{
2057 struct tcmsg *tcm;
2058 struct nlmsghdr *nlh;
2059 unsigned char *b = skb_tail_pointer(skb);
2060
2061 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2062 if (!nlh)
2063 goto out_nlmsg_trim;
2064 tcm = nlmsg_data(nlh);
2065 tcm->tcm_family = AF_UNSPEC;
2066 tcm->tcm__pad1 = 0;
2067 tcm->tcm__pad2 = 0;
2068 if (q) {
2069 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
2070 tcm->tcm_parent = parent;
2071 } else {
2072 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2073 tcm->tcm_block_index = block->index;
2074 }
2075 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
2076 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
2077 goto nla_put_failure;
2078 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
2079 goto nla_put_failure;
2080 if (!fh) {
2081 tcm->tcm_handle = 0;
2082 } else if (terse_dump) {
2083 if (tp->ops->terse_dump) {
2084 if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
2085 rtnl_held) < 0)
2086 goto nla_put_failure;
2087 } else {
2088 goto cls_op_not_supp;
2089 }
2090 } else {
2091 if (tp->ops->dump &&
2092 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
2093 goto nla_put_failure;
2094 }
2095
2096 if (extack && extack->_msg &&
2097 nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2098 goto nla_put_failure;
2099
2100 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2101
2102 return skb->len;
2103
2104out_nlmsg_trim:
2105nla_put_failure:
2106cls_op_not_supp:
2107 nlmsg_trim(skb, b);
2108 return -1;
2109}
2110
2111static int tfilter_notify(struct net *net, struct sk_buff *oskb,
2112 struct nlmsghdr *n, struct tcf_proto *tp,
2113 struct tcf_block *block, struct Qdisc *q,
2114 u32 parent, void *fh, int event, bool unicast,
2115 bool rtnl_held, struct netlink_ext_ack *extack)
2116{
2117 struct sk_buff *skb;
2118 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2119 int err = 0;
2120
2121 if (!unicast && !rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2122 return 0;
2123
2124 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2125 if (!skb)
2126 return -ENOBUFS;
2127
2128 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2129 n->nlmsg_seq, n->nlmsg_flags, event,
2130 false, rtnl_held, extack) <= 0) {
2131 kfree_skb(skb);
2132 return -EINVAL;
2133 }
2134
2135 if (unicast)
2136 err = rtnl_unicast(skb, net, portid);
2137 else
2138 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2139 n->nlmsg_flags & NLM_F_ECHO);
2140 return err;
2141}
2142
2143static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
2144 struct nlmsghdr *n, struct tcf_proto *tp,
2145 struct tcf_block *block, struct Qdisc *q,
2146 u32 parent, void *fh, bool *last, bool rtnl_held,
2147 struct netlink_ext_ack *extack)
2148{
2149 struct sk_buff *skb;
2150 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2151 int err;
2152
2153 if (!rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2154 return tp->ops->delete(tp, fh, last, rtnl_held, extack);
2155
2156 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2157 if (!skb)
2158 return -ENOBUFS;
2159
2160 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2161 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
2162 false, rtnl_held, extack) <= 0) {
2163 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
2164 kfree_skb(skb);
2165 return -EINVAL;
2166 }
2167
2168 err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
2169 if (err) {
2170 kfree_skb(skb);
2171 return err;
2172 }
2173
2174 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2175 n->nlmsg_flags & NLM_F_ECHO);
2176 if (err < 0)
2177 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
2178
2179 return err;
2180}
2181
2182static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
2183 struct tcf_block *block, struct Qdisc *q,
2184 u32 parent, struct nlmsghdr *n,
2185 struct tcf_chain *chain, int event,
2186 struct netlink_ext_ack *extack)
2187{
2188 struct tcf_proto *tp;
2189
2190 for (tp = tcf_get_next_proto(chain, NULL);
2191 tp; tp = tcf_get_next_proto(chain, tp))
2192 tfilter_notify(net, oskb, n, tp, block, q, parent, NULL,
2193 event, false, true, extack);
2194}
2195
2196static void tfilter_put(struct tcf_proto *tp, void *fh)
2197{
2198 if (tp->ops->put && fh)
2199 tp->ops->put(tp, fh);
2200}
2201
2202static bool is_qdisc_ingress(__u32 classid)
2203{
2204 return (TC_H_MIN(classid) == TC_H_MIN(TC_H_MIN_INGRESS));
2205}
2206
2207static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2208 struct netlink_ext_ack *extack)
2209{
2210 struct net *net = sock_net(skb->sk);
2211 struct nlattr *tca[TCA_MAX + 1];
2212 char name[IFNAMSIZ];
2213 struct tcmsg *t;
2214 u32 protocol;
2215 u32 prio;
2216 bool prio_allocate;
2217 u32 parent;
2218 u32 chain_index;
2219 struct Qdisc *q;
2220 struct tcf_chain_info chain_info;
2221 struct tcf_chain *chain;
2222 struct tcf_block *block;
2223 struct tcf_proto *tp;
2224 unsigned long cl;
2225 void *fh;
2226 int err;
2227 int tp_created;
2228 bool rtnl_held = false;
2229 u32 flags;
2230
2231replay:
2232 tp_created = 0;
2233
2234 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2235 rtm_tca_policy, extack);
2236 if (err < 0)
2237 return err;
2238
2239 t = nlmsg_data(n);
2240 protocol = TC_H_MIN(t->tcm_info);
2241 prio = TC_H_MAJ(t->tcm_info);
2242 prio_allocate = false;
2243 parent = t->tcm_parent;
2244 tp = NULL;
2245 cl = 0;
2246 block = NULL;
2247 q = NULL;
2248 chain = NULL;
2249 flags = 0;
2250
2251 if (prio == 0) {
2252 /* If no priority is provided by the user,
2253 * we allocate one.
2254 */
2255 if (n->nlmsg_flags & NLM_F_CREATE) {
2256 prio = TC_H_MAKE(0x80000000U, 0U);
2257 prio_allocate = true;
2258 } else {
2259 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2260 return -ENOENT;
2261 }
2262 }
2263
2264 /* Find head of filter chain. */
2265
2266 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2267 if (err)
2268 return err;
2269
2270 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2271 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2272 err = -EINVAL;
2273 goto errout;
2274 }
2275
2276 /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2277 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2278 * type is not specified, classifier is not unlocked.
2279 */
2280 if (rtnl_held ||
2281 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2282 !tcf_proto_is_unlocked(name)) {
2283 rtnl_held = true;
2284 rtnl_lock();
2285 }
2286
2287 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2288 if (err)
2289 goto errout;
2290
2291 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2292 extack);
2293 if (IS_ERR(block)) {
2294 err = PTR_ERR(block);
2295 goto errout;
2296 }
2297 block->classid = parent;
2298
2299 chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
2300 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2301 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2302 err = -EINVAL;
2303 goto errout;
2304 }
2305 chain = tcf_chain_get(block, chain_index, true);
2306 if (!chain) {
2307 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2308 err = -ENOMEM;
2309 goto errout;
2310 }
2311
2312 mutex_lock(&chain->filter_chain_lock);
2313 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2314 prio, prio_allocate, extack);
2315 if (IS_ERR(tp)) {
2316 err = PTR_ERR(tp);
2317 goto errout_locked;
2318 }
2319
2320 if (tp == NULL) {
2321 struct tcf_proto *tp_new = NULL;
2322
2323 if (chain->flushing) {
2324 err = -EAGAIN;
2325 goto errout_locked;
2326 }
2327
2328 /* Proto-tcf does not exist, create new one */
2329
2330 if (tca[TCA_KIND] == NULL || !protocol) {
2331 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2332 err = -EINVAL;
2333 goto errout_locked;
2334 }
2335
2336 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2337 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2338 err = -ENOENT;
2339 goto errout_locked;
2340 }
2341
2342 if (prio_allocate)
2343 prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2344 &chain_info));
2345
2346 mutex_unlock(&chain->filter_chain_lock);
2347 tp_new = tcf_proto_create(name, protocol, prio, chain,
2348 rtnl_held, extack);
2349 if (IS_ERR(tp_new)) {
2350 err = PTR_ERR(tp_new);
2351 goto errout_tp;
2352 }
2353
2354 tp_created = 1;
2355 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2356 rtnl_held);
2357 if (IS_ERR(tp)) {
2358 err = PTR_ERR(tp);
2359 goto errout_tp;
2360 }
2361 } else {
2362 mutex_unlock(&chain->filter_chain_lock);
2363 }
2364
2365 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2366 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2367 err = -EINVAL;
2368 goto errout;
2369 }
2370
2371 fh = tp->ops->get(tp, t->tcm_handle);
2372
2373 if (!fh) {
2374 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2375 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2376 err = -ENOENT;
2377 goto errout;
2378 }
2379 } else if (n->nlmsg_flags & NLM_F_EXCL) {
2380 tfilter_put(tp, fh);
2381 NL_SET_ERR_MSG(extack, "Filter already exists");
2382 err = -EEXIST;
2383 goto errout;
2384 }
2385
2386 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2387 tfilter_put(tp, fh);
2388 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2389 err = -EINVAL;
2390 goto errout;
2391 }
2392
2393 if (!(n->nlmsg_flags & NLM_F_CREATE))
2394 flags |= TCA_ACT_FLAGS_REPLACE;
2395 if (!rtnl_held)
2396 flags |= TCA_ACT_FLAGS_NO_RTNL;
2397 if (is_qdisc_ingress(parent))
2398 flags |= TCA_ACT_FLAGS_AT_INGRESS;
2399 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2400 flags, extack);
2401 if (err == 0) {
2402 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2403 RTM_NEWTFILTER, false, rtnl_held, extack);
2404 tfilter_put(tp, fh);
2405 tcf_proto_count_usesw(tp, true);
2406 /* q pointer is NULL for shared blocks */
2407 if (q)
2408 q->flags &= ~TCQ_F_CAN_BYPASS;
2409 }
2410
2411errout:
2412 if (err && tp_created)
2413 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2414errout_tp:
2415 if (chain) {
2416 if (tp && !IS_ERR(tp))
2417 tcf_proto_put(tp, rtnl_held, NULL);
2418 if (!tp_created)
2419 tcf_chain_put(chain);
2420 }
2421 tcf_block_release(q, block, rtnl_held);
2422
2423 if (rtnl_held)
2424 rtnl_unlock();
2425
2426 if (err == -EAGAIN) {
2427 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2428 * of target chain.
2429 */
2430 rtnl_held = true;
2431 /* Replay the request. */
2432 goto replay;
2433 }
2434 return err;
2435
2436errout_locked:
2437 mutex_unlock(&chain->filter_chain_lock);
2438 goto errout;
2439}
2440
2441static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2442 struct netlink_ext_ack *extack)
2443{
2444 struct net *net = sock_net(skb->sk);
2445 struct nlattr *tca[TCA_MAX + 1];
2446 char name[IFNAMSIZ];
2447 struct tcmsg *t;
2448 u32 protocol;
2449 u32 prio;
2450 u32 parent;
2451 u32 chain_index;
2452 struct Qdisc *q = NULL;
2453 struct tcf_chain_info chain_info;
2454 struct tcf_chain *chain = NULL;
2455 struct tcf_block *block = NULL;
2456 struct tcf_proto *tp = NULL;
2457 unsigned long cl = 0;
2458 void *fh = NULL;
2459 int err;
2460 bool rtnl_held = false;
2461
2462 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2463 rtm_tca_policy, extack);
2464 if (err < 0)
2465 return err;
2466
2467 t = nlmsg_data(n);
2468 protocol = TC_H_MIN(t->tcm_info);
2469 prio = TC_H_MAJ(t->tcm_info);
2470 parent = t->tcm_parent;
2471
2472 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2473 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2474 return -ENOENT;
2475 }
2476
2477 /* Find head of filter chain. */
2478
2479 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2480 if (err)
2481 return err;
2482
2483 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2484 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2485 err = -EINVAL;
2486 goto errout;
2487 }
2488 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2489 * found), qdisc is not unlocked, classifier type is not specified,
2490 * classifier is not unlocked.
2491 */
2492 if (!prio ||
2493 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2494 !tcf_proto_is_unlocked(name)) {
2495 rtnl_held = true;
2496 rtnl_lock();
2497 }
2498
2499 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2500 if (err)
2501 goto errout;
2502
2503 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2504 extack);
2505 if (IS_ERR(block)) {
2506 err = PTR_ERR(block);
2507 goto errout;
2508 }
2509
2510 chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
2511 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2512 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2513 err = -EINVAL;
2514 goto errout;
2515 }
2516 chain = tcf_chain_get(block, chain_index, false);
2517 if (!chain) {
2518 /* User requested flush on non-existent chain. Nothing to do,
2519 * so just return success.
2520 */
2521 if (prio == 0) {
2522 err = 0;
2523 goto errout;
2524 }
2525 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2526 err = -ENOENT;
2527 goto errout;
2528 }
2529
2530 if (prio == 0) {
2531 tfilter_notify_chain(net, skb, block, q, parent, n,
2532 chain, RTM_DELTFILTER, extack);
2533 tcf_chain_flush(chain, rtnl_held);
2534 err = 0;
2535 goto errout;
2536 }
2537
2538 mutex_lock(&chain->filter_chain_lock);
2539 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2540 prio, false, extack);
2541 if (!tp) {
2542 err = -ENOENT;
2543 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2544 goto errout_locked;
2545 } else if (IS_ERR(tp)) {
2546 err = PTR_ERR(tp);
2547 goto errout_locked;
2548 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2549 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2550 err = -EINVAL;
2551 goto errout_locked;
2552 } else if (t->tcm_handle == 0) {
2553 tcf_proto_signal_destroying(chain, tp);
2554 tcf_chain_tp_remove(chain, &chain_info, tp);
2555 mutex_unlock(&chain->filter_chain_lock);
2556
2557 tcf_proto_put(tp, rtnl_held, NULL);
2558 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2559 RTM_DELTFILTER, false, rtnl_held, extack);
2560 err = 0;
2561 goto errout;
2562 }
2563 mutex_unlock(&chain->filter_chain_lock);
2564
2565 fh = tp->ops->get(tp, t->tcm_handle);
2566
2567 if (!fh) {
2568 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2569 err = -ENOENT;
2570 } else {
2571 bool last;
2572
2573 err = tfilter_del_notify(net, skb, n, tp, block, q, parent, fh,
2574 &last, rtnl_held, extack);
2575
2576 if (err)
2577 goto errout;
2578 if (last)
2579 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2580 }
2581
2582errout:
2583 if (chain) {
2584 if (tp && !IS_ERR(tp))
2585 tcf_proto_put(tp, rtnl_held, NULL);
2586 tcf_chain_put(chain);
2587 }
2588 tcf_block_release(q, block, rtnl_held);
2589
2590 if (rtnl_held)
2591 rtnl_unlock();
2592
2593 return err;
2594
2595errout_locked:
2596 mutex_unlock(&chain->filter_chain_lock);
2597 goto errout;
2598}
2599
2600static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2601 struct netlink_ext_ack *extack)
2602{
2603 struct net *net = sock_net(skb->sk);
2604 struct nlattr *tca[TCA_MAX + 1];
2605 char name[IFNAMSIZ];
2606 struct tcmsg *t;
2607 u32 protocol;
2608 u32 prio;
2609 u32 parent;
2610 u32 chain_index;
2611 struct Qdisc *q = NULL;
2612 struct tcf_chain_info chain_info;
2613 struct tcf_chain *chain = NULL;
2614 struct tcf_block *block = NULL;
2615 struct tcf_proto *tp = NULL;
2616 unsigned long cl = 0;
2617 void *fh = NULL;
2618 int err;
2619 bool rtnl_held = false;
2620
2621 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2622 rtm_tca_policy, extack);
2623 if (err < 0)
2624 return err;
2625
2626 t = nlmsg_data(n);
2627 protocol = TC_H_MIN(t->tcm_info);
2628 prio = TC_H_MAJ(t->tcm_info);
2629 parent = t->tcm_parent;
2630
2631 if (prio == 0) {
2632 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2633 return -ENOENT;
2634 }
2635
2636 /* Find head of filter chain. */
2637
2638 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2639 if (err)
2640 return err;
2641
2642 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2643 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2644 err = -EINVAL;
2645 goto errout;
2646 }
2647 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2648 * unlocked, classifier type is not specified, classifier is not
2649 * unlocked.
2650 */
2651 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2652 !tcf_proto_is_unlocked(name)) {
2653 rtnl_held = true;
2654 rtnl_lock();
2655 }
2656
2657 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2658 if (err)
2659 goto errout;
2660
2661 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2662 extack);
2663 if (IS_ERR(block)) {
2664 err = PTR_ERR(block);
2665 goto errout;
2666 }
2667
2668 chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
2669 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2670 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2671 err = -EINVAL;
2672 goto errout;
2673 }
2674 chain = tcf_chain_get(block, chain_index, false);
2675 if (!chain) {
2676 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2677 err = -EINVAL;
2678 goto errout;
2679 }
2680
2681 mutex_lock(&chain->filter_chain_lock);
2682 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2683 prio, false, extack);
2684 mutex_unlock(&chain->filter_chain_lock);
2685 if (!tp) {
2686 err = -ENOENT;
2687 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2688 goto errout;
2689 } else if (IS_ERR(tp)) {
2690 err = PTR_ERR(tp);
2691 goto errout;
2692 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2693 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2694 err = -EINVAL;
2695 goto errout;
2696 }
2697
2698 fh = tp->ops->get(tp, t->tcm_handle);
2699
2700 if (!fh) {
2701 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2702 err = -ENOENT;
2703 } else {
2704 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2705 fh, RTM_NEWTFILTER, true, rtnl_held, NULL);
2706 if (err < 0)
2707 NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2708 }
2709
2710 tfilter_put(tp, fh);
2711errout:
2712 if (chain) {
2713 if (tp && !IS_ERR(tp))
2714 tcf_proto_put(tp, rtnl_held, NULL);
2715 tcf_chain_put(chain);
2716 }
2717 tcf_block_release(q, block, rtnl_held);
2718
2719 if (rtnl_held)
2720 rtnl_unlock();
2721
2722 return err;
2723}
2724
2725struct tcf_dump_args {
2726 struct tcf_walker w;
2727 struct sk_buff *skb;
2728 struct netlink_callback *cb;
2729 struct tcf_block *block;
2730 struct Qdisc *q;
2731 u32 parent;
2732 bool terse_dump;
2733};
2734
2735static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2736{
2737 struct tcf_dump_args *a = (void *)arg;
2738 struct net *net = sock_net(a->skb->sk);
2739
2740 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2741 n, NETLINK_CB(a->cb->skb).portid,
2742 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2743 RTM_NEWTFILTER, a->terse_dump, true, NULL);
2744}
2745
2746static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2747 struct sk_buff *skb, struct netlink_callback *cb,
2748 long index_start, long *p_index, bool terse)
2749{
2750 struct net *net = sock_net(skb->sk);
2751 struct tcf_block *block = chain->block;
2752 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2753 struct tcf_proto *tp, *tp_prev;
2754 struct tcf_dump_args arg;
2755
2756 for (tp = __tcf_get_next_proto(chain, NULL);
2757 tp;
2758 tp_prev = tp,
2759 tp = __tcf_get_next_proto(chain, tp),
2760 tcf_proto_put(tp_prev, true, NULL),
2761 (*p_index)++) {
2762 if (*p_index < index_start)
2763 continue;
2764 if (TC_H_MAJ(tcm->tcm_info) &&
2765 TC_H_MAJ(tcm->tcm_info) != tp->prio)
2766 continue;
2767 if (TC_H_MIN(tcm->tcm_info) &&
2768 TC_H_MIN(tcm->tcm_info) != tp->protocol)
2769 continue;
2770 if (*p_index > index_start)
2771 memset(&cb->args[1], 0,
2772 sizeof(cb->args) - sizeof(cb->args[0]));
2773 if (cb->args[1] == 0) {
2774 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2775 NETLINK_CB(cb->skb).portid,
2776 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2777 RTM_NEWTFILTER, false, true, NULL) <= 0)
2778 goto errout;
2779 cb->args[1] = 1;
2780 }
2781 if (!tp->ops->walk)
2782 continue;
2783 arg.w.fn = tcf_node_dump;
2784 arg.skb = skb;
2785 arg.cb = cb;
2786 arg.block = block;
2787 arg.q = q;
2788 arg.parent = parent;
2789 arg.w.stop = 0;
2790 arg.w.skip = cb->args[1] - 1;
2791 arg.w.count = 0;
2792 arg.w.cookie = cb->args[2];
2793 arg.terse_dump = terse;
2794 tp->ops->walk(tp, &arg.w, true);
2795 cb->args[2] = arg.w.cookie;
2796 cb->args[1] = arg.w.count + 1;
2797 if (arg.w.stop)
2798 goto errout;
2799 }
2800 return true;
2801
2802errout:
2803 tcf_proto_put(tp, true, NULL);
2804 return false;
2805}
2806
2807static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2808 [TCA_CHAIN] = { .type = NLA_U32 },
2809 [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2810};
2811
2812/* called with RTNL */
2813static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2814{
2815 struct tcf_chain *chain, *chain_prev;
2816 struct net *net = sock_net(skb->sk);
2817 struct nlattr *tca[TCA_MAX + 1];
2818 struct Qdisc *q = NULL;
2819 struct tcf_block *block;
2820 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2821 bool terse_dump = false;
2822 long index_start;
2823 long index;
2824 u32 parent;
2825 int err;
2826
2827 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2828 return skb->len;
2829
2830 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2831 tcf_tfilter_dump_policy, cb->extack);
2832 if (err)
2833 return err;
2834
2835 if (tca[TCA_DUMP_FLAGS]) {
2836 struct nla_bitfield32 flags =
2837 nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2838
2839 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2840 }
2841
2842 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2843 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2844 if (!block)
2845 goto out;
2846 /* If we work with block index, q is NULL and parent value
2847 * will never be used in the following code. The check
2848 * in tcf_fill_node prevents it. However, compiler does not
2849 * see that far, so set parent to zero to silence the warning
2850 * about parent being uninitialized.
2851 */
2852 parent = 0;
2853 } else {
2854 const struct Qdisc_class_ops *cops;
2855 struct net_device *dev;
2856 unsigned long cl = 0;
2857
2858 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2859 if (!dev)
2860 return skb->len;
2861
2862 parent = tcm->tcm_parent;
2863 if (!parent)
2864 q = rtnl_dereference(dev->qdisc);
2865 else
2866 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2867 if (!q)
2868 goto out;
2869 cops = q->ops->cl_ops;
2870 if (!cops)
2871 goto out;
2872 if (!cops->tcf_block)
2873 goto out;
2874 if (TC_H_MIN(tcm->tcm_parent)) {
2875 cl = cops->find(q, tcm->tcm_parent);
2876 if (cl == 0)
2877 goto out;
2878 }
2879 block = cops->tcf_block(q, cl, NULL);
2880 if (!block)
2881 goto out;
2882 parent = block->classid;
2883 if (tcf_block_shared(block))
2884 q = NULL;
2885 }
2886
2887 index_start = cb->args[0];
2888 index = 0;
2889
2890 for (chain = __tcf_get_next_chain(block, NULL);
2891 chain;
2892 chain_prev = chain,
2893 chain = __tcf_get_next_chain(block, chain),
2894 tcf_chain_put(chain_prev)) {
2895 if (tca[TCA_CHAIN] &&
2896 nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2897 continue;
2898 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2899 index_start, &index, terse_dump)) {
2900 tcf_chain_put(chain);
2901 err = -EMSGSIZE;
2902 break;
2903 }
2904 }
2905
2906 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2907 tcf_block_refcnt_put(block, true);
2908 cb->args[0] = index;
2909
2910out:
2911 /* If we did no progress, the error (EMSGSIZE) is real */
2912 if (skb->len == 0 && err)
2913 return err;
2914 return skb->len;
2915}
2916
2917static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2918 void *tmplt_priv, u32 chain_index,
2919 struct net *net, struct sk_buff *skb,
2920 struct tcf_block *block,
2921 u32 portid, u32 seq, u16 flags, int event,
2922 struct netlink_ext_ack *extack)
2923{
2924 unsigned char *b = skb_tail_pointer(skb);
2925 const struct tcf_proto_ops *ops;
2926 struct nlmsghdr *nlh;
2927 struct tcmsg *tcm;
2928 void *priv;
2929
2930 ops = tmplt_ops;
2931 priv = tmplt_priv;
2932
2933 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2934 if (!nlh)
2935 goto out_nlmsg_trim;
2936 tcm = nlmsg_data(nlh);
2937 tcm->tcm_family = AF_UNSPEC;
2938 tcm->tcm__pad1 = 0;
2939 tcm->tcm__pad2 = 0;
2940 tcm->tcm_handle = 0;
2941 if (block->q) {
2942 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2943 tcm->tcm_parent = block->q->handle;
2944 } else {
2945 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2946 tcm->tcm_block_index = block->index;
2947 }
2948
2949 if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2950 goto nla_put_failure;
2951
2952 if (ops) {
2953 if (nla_put_string(skb, TCA_KIND, ops->kind))
2954 goto nla_put_failure;
2955 if (ops->tmplt_dump(skb, net, priv) < 0)
2956 goto nla_put_failure;
2957 }
2958
2959 if (extack && extack->_msg &&
2960 nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2961 goto out_nlmsg_trim;
2962
2963 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2964
2965 return skb->len;
2966
2967out_nlmsg_trim:
2968nla_put_failure:
2969 nlmsg_trim(skb, b);
2970 return -EMSGSIZE;
2971}
2972
2973static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2974 u32 seq, u16 flags, int event, bool unicast,
2975 struct netlink_ext_ack *extack)
2976{
2977 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2978 struct tcf_block *block = chain->block;
2979 struct net *net = block->net;
2980 struct sk_buff *skb;
2981 int err = 0;
2982
2983 if (!unicast && !rtnl_notify_needed(net, flags, RTNLGRP_TC))
2984 return 0;
2985
2986 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2987 if (!skb)
2988 return -ENOBUFS;
2989
2990 if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2991 chain->index, net, skb, block, portid,
2992 seq, flags, event, extack) <= 0) {
2993 kfree_skb(skb);
2994 return -EINVAL;
2995 }
2996
2997 if (unicast)
2998 err = rtnl_unicast(skb, net, portid);
2999 else
3000 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
3001 flags & NLM_F_ECHO);
3002
3003 return err;
3004}
3005
3006static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
3007 void *tmplt_priv, u32 chain_index,
3008 struct tcf_block *block, struct sk_buff *oskb,
3009 u32 seq, u16 flags)
3010{
3011 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
3012 struct net *net = block->net;
3013 struct sk_buff *skb;
3014
3015 if (!rtnl_notify_needed(net, flags, RTNLGRP_TC))
3016 return 0;
3017
3018 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3019 if (!skb)
3020 return -ENOBUFS;
3021
3022 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
3023 block, portid, seq, flags, RTM_DELCHAIN, NULL) <= 0) {
3024 kfree_skb(skb);
3025 return -EINVAL;
3026 }
3027
3028 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
3029}
3030
3031static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
3032 struct nlattr **tca,
3033 struct netlink_ext_ack *extack)
3034{
3035 const struct tcf_proto_ops *ops;
3036 char name[IFNAMSIZ];
3037 void *tmplt_priv;
3038
3039 /* If kind is not set, user did not specify template. */
3040 if (!tca[TCA_KIND])
3041 return 0;
3042
3043 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
3044 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
3045 return -EINVAL;
3046 }
3047
3048 ops = tcf_proto_lookup_ops(name, true, extack);
3049 if (IS_ERR(ops))
3050 return PTR_ERR(ops);
3051 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump ||
3052 !ops->tmplt_reoffload) {
3053 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
3054 module_put(ops->owner);
3055 return -EOPNOTSUPP;
3056 }
3057
3058 tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
3059 if (IS_ERR(tmplt_priv)) {
3060 module_put(ops->owner);
3061 return PTR_ERR(tmplt_priv);
3062 }
3063 chain->tmplt_ops = ops;
3064 chain->tmplt_priv = tmplt_priv;
3065 return 0;
3066}
3067
3068static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
3069 void *tmplt_priv)
3070{
3071 /* If template ops are set, no work to do for us. */
3072 if (!tmplt_ops)
3073 return;
3074
3075 tmplt_ops->tmplt_destroy(tmplt_priv);
3076 module_put(tmplt_ops->owner);
3077}
3078
3079/* Add/delete/get a chain */
3080
3081static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
3082 struct netlink_ext_ack *extack)
3083{
3084 struct net *net = sock_net(skb->sk);
3085 struct nlattr *tca[TCA_MAX + 1];
3086 struct tcmsg *t;
3087 u32 parent;
3088 u32 chain_index;
3089 struct Qdisc *q;
3090 struct tcf_chain *chain;
3091 struct tcf_block *block;
3092 unsigned long cl;
3093 int err;
3094
3095replay:
3096 q = NULL;
3097 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
3098 rtm_tca_policy, extack);
3099 if (err < 0)
3100 return err;
3101
3102 t = nlmsg_data(n);
3103 parent = t->tcm_parent;
3104 cl = 0;
3105
3106 block = tcf_block_find(net, &q, &parent, &cl,
3107 t->tcm_ifindex, t->tcm_block_index, extack);
3108 if (IS_ERR(block))
3109 return PTR_ERR(block);
3110
3111 chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
3112 if (chain_index > TC_ACT_EXT_VAL_MASK) {
3113 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
3114 err = -EINVAL;
3115 goto errout_block;
3116 }
3117
3118 mutex_lock(&block->lock);
3119 chain = tcf_chain_lookup(block, chain_index);
3120 if (n->nlmsg_type == RTM_NEWCHAIN) {
3121 if (chain) {
3122 if (tcf_chain_held_by_acts_only(chain)) {
3123 /* The chain exists only because there is
3124 * some action referencing it.
3125 */
3126 tcf_chain_hold(chain);
3127 } else {
3128 NL_SET_ERR_MSG(extack, "Filter chain already exists");
3129 err = -EEXIST;
3130 goto errout_block_locked;
3131 }
3132 } else {
3133 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
3134 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
3135 err = -ENOENT;
3136 goto errout_block_locked;
3137 }
3138 chain = tcf_chain_create(block, chain_index);
3139 if (!chain) {
3140 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
3141 err = -ENOMEM;
3142 goto errout_block_locked;
3143 }
3144 }
3145 } else {
3146 if (!chain || tcf_chain_held_by_acts_only(chain)) {
3147 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
3148 err = -EINVAL;
3149 goto errout_block_locked;
3150 }
3151 tcf_chain_hold(chain);
3152 }
3153
3154 if (n->nlmsg_type == RTM_NEWCHAIN) {
3155 /* Modifying chain requires holding parent block lock. In case
3156 * the chain was successfully added, take a reference to the
3157 * chain. This ensures that an empty chain does not disappear at
3158 * the end of this function.
3159 */
3160 tcf_chain_hold(chain);
3161 chain->explicitly_created = true;
3162 }
3163 mutex_unlock(&block->lock);
3164
3165 switch (n->nlmsg_type) {
3166 case RTM_NEWCHAIN:
3167 err = tc_chain_tmplt_add(chain, net, tca, extack);
3168 if (err) {
3169 tcf_chain_put_explicitly_created(chain);
3170 goto errout;
3171 }
3172
3173 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
3174 RTM_NEWCHAIN, false, extack);
3175 break;
3176 case RTM_DELCHAIN:
3177 tfilter_notify_chain(net, skb, block, q, parent, n,
3178 chain, RTM_DELTFILTER, extack);
3179 /* Flush the chain first as the user requested chain removal. */
3180 tcf_chain_flush(chain, true);
3181 /* In case the chain was successfully deleted, put a reference
3182 * to the chain previously taken during addition.
3183 */
3184 tcf_chain_put_explicitly_created(chain);
3185 break;
3186 case RTM_GETCHAIN:
3187 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
3188 n->nlmsg_flags, n->nlmsg_type, true, extack);
3189 if (err < 0)
3190 NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
3191 break;
3192 default:
3193 err = -EOPNOTSUPP;
3194 NL_SET_ERR_MSG(extack, "Unsupported message type");
3195 goto errout;
3196 }
3197
3198errout:
3199 tcf_chain_put(chain);
3200errout_block:
3201 tcf_block_release(q, block, true);
3202 if (err == -EAGAIN)
3203 /* Replay the request. */
3204 goto replay;
3205 return err;
3206
3207errout_block_locked:
3208 mutex_unlock(&block->lock);
3209 goto errout_block;
3210}
3211
3212/* called with RTNL */
3213static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
3214{
3215 struct net *net = sock_net(skb->sk);
3216 struct nlattr *tca[TCA_MAX + 1];
3217 struct Qdisc *q = NULL;
3218 struct tcf_block *block;
3219 struct tcmsg *tcm = nlmsg_data(cb->nlh);
3220 struct tcf_chain *chain;
3221 long index_start;
3222 long index;
3223 int err;
3224
3225 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
3226 return skb->len;
3227
3228 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
3229 rtm_tca_policy, cb->extack);
3230 if (err)
3231 return err;
3232
3233 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
3234 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
3235 if (!block)
3236 goto out;
3237 } else {
3238 const struct Qdisc_class_ops *cops;
3239 struct net_device *dev;
3240 unsigned long cl = 0;
3241
3242 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
3243 if (!dev)
3244 return skb->len;
3245
3246 if (!tcm->tcm_parent)
3247 q = rtnl_dereference(dev->qdisc);
3248 else
3249 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
3250
3251 if (!q)
3252 goto out;
3253 cops = q->ops->cl_ops;
3254 if (!cops)
3255 goto out;
3256 if (!cops->tcf_block)
3257 goto out;
3258 if (TC_H_MIN(tcm->tcm_parent)) {
3259 cl = cops->find(q, tcm->tcm_parent);
3260 if (cl == 0)
3261 goto out;
3262 }
3263 block = cops->tcf_block(q, cl, NULL);
3264 if (!block)
3265 goto out;
3266 if (tcf_block_shared(block))
3267 q = NULL;
3268 }
3269
3270 index_start = cb->args[0];
3271 index = 0;
3272
3273 mutex_lock(&block->lock);
3274 list_for_each_entry(chain, &block->chain_list, list) {
3275 if ((tca[TCA_CHAIN] &&
3276 nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3277 continue;
3278 if (index < index_start) {
3279 index++;
3280 continue;
3281 }
3282 if (tcf_chain_held_by_acts_only(chain))
3283 continue;
3284 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3285 chain->index, net, skb, block,
3286 NETLINK_CB(cb->skb).portid,
3287 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3288 RTM_NEWCHAIN, NULL);
3289 if (err <= 0)
3290 break;
3291 index++;
3292 }
3293 mutex_unlock(&block->lock);
3294
3295 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3296 tcf_block_refcnt_put(block, true);
3297 cb->args[0] = index;
3298
3299out:
3300 /* If we did no progress, the error (EMSGSIZE) is real */
3301 if (skb->len == 0 && err)
3302 return err;
3303 return skb->len;
3304}
3305
3306int tcf_exts_init_ex(struct tcf_exts *exts, struct net *net, int action,
3307 int police, struct tcf_proto *tp, u32 handle,
3308 bool use_action_miss)
3309{
3310 int err = 0;
3311
3312#ifdef CONFIG_NET_CLS_ACT
3313 exts->type = 0;
3314 exts->nr_actions = 0;
3315 exts->miss_cookie_node = NULL;
3316 /* Note: we do not own yet a reference on net.
3317 * This reference might be taken later from tcf_exts_get_net().
3318 */
3319 exts->net = net;
3320 exts->actions = kcalloc(TCA_ACT_MAX_PRIO, sizeof(struct tc_action *),
3321 GFP_KERNEL);
3322 if (!exts->actions)
3323 return -ENOMEM;
3324#endif
3325
3326 exts->action = action;
3327 exts->police = police;
3328
3329 if (!use_action_miss)
3330 return 0;
3331
3332 err = tcf_exts_miss_cookie_base_alloc(exts, tp, handle);
3333 if (err)
3334 goto err_miss_alloc;
3335
3336 return 0;
3337
3338err_miss_alloc:
3339 tcf_exts_destroy(exts);
3340#ifdef CONFIG_NET_CLS_ACT
3341 exts->actions = NULL;
3342#endif
3343 return err;
3344}
3345EXPORT_SYMBOL(tcf_exts_init_ex);
3346
3347void tcf_exts_destroy(struct tcf_exts *exts)
3348{
3349 tcf_exts_miss_cookie_base_destroy(exts);
3350
3351#ifdef CONFIG_NET_CLS_ACT
3352 if (exts->actions) {
3353 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3354 kfree(exts->actions);
3355 }
3356 exts->nr_actions = 0;
3357#endif
3358}
3359EXPORT_SYMBOL(tcf_exts_destroy);
3360
3361int tcf_exts_validate_ex(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3362 struct nlattr *rate_tlv, struct tcf_exts *exts,
3363 u32 flags, u32 fl_flags, struct netlink_ext_ack *extack)
3364{
3365#ifdef CONFIG_NET_CLS_ACT
3366 {
3367 int init_res[TCA_ACT_MAX_PRIO] = {};
3368 struct tc_action *act;
3369 size_t attr_size = 0;
3370
3371 if (exts->police && tb[exts->police]) {
3372 struct tc_action_ops *a_o;
3373
3374 flags |= TCA_ACT_FLAGS_POLICE | TCA_ACT_FLAGS_BIND;
3375 a_o = tc_action_load_ops(tb[exts->police], flags,
3376 extack);
3377 if (IS_ERR(a_o))
3378 return PTR_ERR(a_o);
3379 act = tcf_action_init_1(net, tp, tb[exts->police],
3380 rate_tlv, a_o, init_res, flags,
3381 extack);
3382 module_put(a_o->owner);
3383 if (IS_ERR(act))
3384 return PTR_ERR(act);
3385
3386 act->type = exts->type = TCA_OLD_COMPAT;
3387 exts->actions[0] = act;
3388 exts->nr_actions = 1;
3389 tcf_idr_insert_many(exts->actions, init_res);
3390 } else if (exts->action && tb[exts->action]) {
3391 int err;
3392
3393 flags |= TCA_ACT_FLAGS_BIND;
3394 err = tcf_action_init(net, tp, tb[exts->action],
3395 rate_tlv, exts->actions, init_res,
3396 &attr_size, flags, fl_flags,
3397 extack);
3398 if (err < 0)
3399 return err;
3400 exts->nr_actions = err;
3401 }
3402 }
3403#else
3404 if ((exts->action && tb[exts->action]) ||
3405 (exts->police && tb[exts->police])) {
3406 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3407 return -EOPNOTSUPP;
3408 }
3409#endif
3410
3411 return 0;
3412}
3413EXPORT_SYMBOL(tcf_exts_validate_ex);
3414
3415int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3416 struct nlattr *rate_tlv, struct tcf_exts *exts,
3417 u32 flags, struct netlink_ext_ack *extack)
3418{
3419 return tcf_exts_validate_ex(net, tp, tb, rate_tlv, exts,
3420 flags, 0, extack);
3421}
3422EXPORT_SYMBOL(tcf_exts_validate);
3423
3424void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3425{
3426#ifdef CONFIG_NET_CLS_ACT
3427 struct tcf_exts old = *dst;
3428
3429 *dst = *src;
3430 tcf_exts_destroy(&old);
3431#endif
3432}
3433EXPORT_SYMBOL(tcf_exts_change);
3434
3435#ifdef CONFIG_NET_CLS_ACT
3436static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3437{
3438 if (exts->nr_actions == 0)
3439 return NULL;
3440 else
3441 return exts->actions[0];
3442}
3443#endif
3444
3445int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3446{
3447#ifdef CONFIG_NET_CLS_ACT
3448 struct nlattr *nest;
3449
3450 if (exts->action && tcf_exts_has_actions(exts)) {
3451 /*
3452 * again for backward compatible mode - we want
3453 * to work with both old and new modes of entering
3454 * tc data even if iproute2 was newer - jhs
3455 */
3456 if (exts->type != TCA_OLD_COMPAT) {
3457 nest = nla_nest_start_noflag(skb, exts->action);
3458 if (nest == NULL)
3459 goto nla_put_failure;
3460
3461 if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3462 < 0)
3463 goto nla_put_failure;
3464 nla_nest_end(skb, nest);
3465 } else if (exts->police) {
3466 struct tc_action *act = tcf_exts_first_act(exts);
3467 nest = nla_nest_start_noflag(skb, exts->police);
3468 if (nest == NULL || !act)
3469 goto nla_put_failure;
3470 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3471 goto nla_put_failure;
3472 nla_nest_end(skb, nest);
3473 }
3474 }
3475 return 0;
3476
3477nla_put_failure:
3478 nla_nest_cancel(skb, nest);
3479 return -1;
3480#else
3481 return 0;
3482#endif
3483}
3484EXPORT_SYMBOL(tcf_exts_dump);
3485
3486int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3487{
3488#ifdef CONFIG_NET_CLS_ACT
3489 struct nlattr *nest;
3490
3491 if (!exts->action || !tcf_exts_has_actions(exts))
3492 return 0;
3493
3494 nest = nla_nest_start_noflag(skb, exts->action);
3495 if (!nest)
3496 goto nla_put_failure;
3497
3498 if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3499 goto nla_put_failure;
3500 nla_nest_end(skb, nest);
3501 return 0;
3502
3503nla_put_failure:
3504 nla_nest_cancel(skb, nest);
3505 return -1;
3506#else
3507 return 0;
3508#endif
3509}
3510EXPORT_SYMBOL(tcf_exts_terse_dump);
3511
3512int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3513{
3514#ifdef CONFIG_NET_CLS_ACT
3515 struct tc_action *a = tcf_exts_first_act(exts);
3516 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3517 return -1;
3518#endif
3519 return 0;
3520}
3521EXPORT_SYMBOL(tcf_exts_dump_stats);
3522
3523static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3524{
3525 if (*flags & TCA_CLS_FLAGS_IN_HW)
3526 return;
3527 *flags |= TCA_CLS_FLAGS_IN_HW;
3528 atomic_inc(&block->offloadcnt);
3529}
3530
3531static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3532{
3533 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3534 return;
3535 *flags &= ~TCA_CLS_FLAGS_IN_HW;
3536 atomic_dec(&block->offloadcnt);
3537}
3538
3539static void tc_cls_offload_cnt_update(struct tcf_block *block,
3540 struct tcf_proto *tp, u32 *cnt,
3541 u32 *flags, u32 diff, bool add)
3542{
3543 lockdep_assert_held(&block->cb_lock);
3544
3545 spin_lock(&tp->lock);
3546 if (add) {
3547 if (!*cnt)
3548 tcf_block_offload_inc(block, flags);
3549 *cnt += diff;
3550 } else {
3551 *cnt -= diff;
3552 if (!*cnt)
3553 tcf_block_offload_dec(block, flags);
3554 }
3555 spin_unlock(&tp->lock);
3556}
3557
3558static void
3559tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3560 u32 *cnt, u32 *flags)
3561{
3562 lockdep_assert_held(&block->cb_lock);
3563
3564 spin_lock(&tp->lock);
3565 tcf_block_offload_dec(block, flags);
3566 *cnt = 0;
3567 spin_unlock(&tp->lock);
3568}
3569
3570static int
3571__tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3572 void *type_data, bool err_stop)
3573{
3574 struct flow_block_cb *block_cb;
3575 int ok_count = 0;
3576 int err;
3577
3578 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3579 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3580 if (err) {
3581 if (err_stop)
3582 return err;
3583 } else {
3584 ok_count++;
3585 }
3586 }
3587 return ok_count;
3588}
3589
3590int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3591 void *type_data, bool err_stop, bool rtnl_held)
3592{
3593 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3594 int ok_count;
3595
3596retry:
3597 if (take_rtnl)
3598 rtnl_lock();
3599 down_read(&block->cb_lock);
3600 /* Need to obtain rtnl lock if block is bound to devs that require it.
3601 * In block bind code cb_lock is obtained while holding rtnl, so we must
3602 * obtain the locks in same order here.
3603 */
3604 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3605 up_read(&block->cb_lock);
3606 take_rtnl = true;
3607 goto retry;
3608 }
3609
3610 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3611
3612 up_read(&block->cb_lock);
3613 if (take_rtnl)
3614 rtnl_unlock();
3615 return ok_count;
3616}
3617EXPORT_SYMBOL(tc_setup_cb_call);
3618
3619/* Non-destructive filter add. If filter that wasn't already in hardware is
3620 * successfully offloaded, increment block offloads counter. On failure,
3621 * previously offloaded filter is considered to be intact and offloads counter
3622 * is not decremented.
3623 */
3624
3625int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3626 enum tc_setup_type type, void *type_data, bool err_stop,
3627 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3628{
3629 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3630 int ok_count;
3631
3632retry:
3633 if (take_rtnl)
3634 rtnl_lock();
3635 down_read(&block->cb_lock);
3636 /* Need to obtain rtnl lock if block is bound to devs that require it.
3637 * In block bind code cb_lock is obtained while holding rtnl, so we must
3638 * obtain the locks in same order here.
3639 */
3640 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3641 up_read(&block->cb_lock);
3642 take_rtnl = true;
3643 goto retry;
3644 }
3645
3646 /* Make sure all netdevs sharing this block are offload-capable. */
3647 if (block->nooffloaddevcnt && err_stop) {
3648 ok_count = -EOPNOTSUPP;
3649 goto err_unlock;
3650 }
3651
3652 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3653 if (ok_count < 0)
3654 goto err_unlock;
3655
3656 if (tp->ops->hw_add)
3657 tp->ops->hw_add(tp, type_data);
3658 if (ok_count > 0)
3659 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3660 ok_count, true);
3661err_unlock:
3662 up_read(&block->cb_lock);
3663 if (take_rtnl)
3664 rtnl_unlock();
3665 return min(ok_count, 0);
3666}
3667EXPORT_SYMBOL(tc_setup_cb_add);
3668
3669/* Destructive filter replace. If filter that wasn't already in hardware is
3670 * successfully offloaded, increment block offload counter. On failure,
3671 * previously offloaded filter is considered to be destroyed and offload counter
3672 * is decremented.
3673 */
3674
3675int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3676 enum tc_setup_type type, void *type_data, bool err_stop,
3677 u32 *old_flags, unsigned int *old_in_hw_count,
3678 u32 *new_flags, unsigned int *new_in_hw_count,
3679 bool rtnl_held)
3680{
3681 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3682 int ok_count;
3683
3684retry:
3685 if (take_rtnl)
3686 rtnl_lock();
3687 down_read(&block->cb_lock);
3688 /* Need to obtain rtnl lock if block is bound to devs that require it.
3689 * In block bind code cb_lock is obtained while holding rtnl, so we must
3690 * obtain the locks in same order here.
3691 */
3692 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3693 up_read(&block->cb_lock);
3694 take_rtnl = true;
3695 goto retry;
3696 }
3697
3698 /* Make sure all netdevs sharing this block are offload-capable. */
3699 if (block->nooffloaddevcnt && err_stop) {
3700 ok_count = -EOPNOTSUPP;
3701 goto err_unlock;
3702 }
3703
3704 tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3705 if (tp->ops->hw_del)
3706 tp->ops->hw_del(tp, type_data);
3707
3708 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3709 if (ok_count < 0)
3710 goto err_unlock;
3711
3712 if (tp->ops->hw_add)
3713 tp->ops->hw_add(tp, type_data);
3714 if (ok_count > 0)
3715 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3716 new_flags, ok_count, true);
3717err_unlock:
3718 up_read(&block->cb_lock);
3719 if (take_rtnl)
3720 rtnl_unlock();
3721 return min(ok_count, 0);
3722}
3723EXPORT_SYMBOL(tc_setup_cb_replace);
3724
3725/* Destroy filter and decrement block offload counter, if filter was previously
3726 * offloaded.
3727 */
3728
3729int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3730 enum tc_setup_type type, void *type_data, bool err_stop,
3731 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3732{
3733 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3734 int ok_count;
3735
3736retry:
3737 if (take_rtnl)
3738 rtnl_lock();
3739 down_read(&block->cb_lock);
3740 /* Need to obtain rtnl lock if block is bound to devs that require it.
3741 * In block bind code cb_lock is obtained while holding rtnl, so we must
3742 * obtain the locks in same order here.
3743 */
3744 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3745 up_read(&block->cb_lock);
3746 take_rtnl = true;
3747 goto retry;
3748 }
3749
3750 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3751
3752 tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3753 if (tp->ops->hw_del)
3754 tp->ops->hw_del(tp, type_data);
3755
3756 up_read(&block->cb_lock);
3757 if (take_rtnl)
3758 rtnl_unlock();
3759 return min(ok_count, 0);
3760}
3761EXPORT_SYMBOL(tc_setup_cb_destroy);
3762
3763int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3764 bool add, flow_setup_cb_t *cb,
3765 enum tc_setup_type type, void *type_data,
3766 void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3767{
3768 int err = cb(type, type_data, cb_priv);
3769
3770 if (err) {
3771 if (add && tc_skip_sw(*flags))
3772 return err;
3773 } else {
3774 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3775 add);
3776 }
3777
3778 return 0;
3779}
3780EXPORT_SYMBOL(tc_setup_cb_reoffload);
3781
3782static int tcf_act_get_user_cookie(struct flow_action_entry *entry,
3783 const struct tc_action *act)
3784{
3785 struct tc_cookie *user_cookie;
3786 int err = 0;
3787
3788 rcu_read_lock();
3789 user_cookie = rcu_dereference(act->user_cookie);
3790 if (user_cookie) {
3791 entry->user_cookie = flow_action_cookie_create(user_cookie->data,
3792 user_cookie->len,
3793 GFP_ATOMIC);
3794 if (!entry->user_cookie)
3795 err = -ENOMEM;
3796 }
3797 rcu_read_unlock();
3798 return err;
3799}
3800
3801static void tcf_act_put_user_cookie(struct flow_action_entry *entry)
3802{
3803 flow_action_cookie_destroy(entry->user_cookie);
3804}
3805
3806void tc_cleanup_offload_action(struct flow_action *flow_action)
3807{
3808 struct flow_action_entry *entry;
3809 int i;
3810
3811 flow_action_for_each(i, entry, flow_action) {
3812 tcf_act_put_user_cookie(entry);
3813 if (entry->destructor)
3814 entry->destructor(entry->destructor_priv);
3815 }
3816}
3817EXPORT_SYMBOL(tc_cleanup_offload_action);
3818
3819static int tc_setup_offload_act(struct tc_action *act,
3820 struct flow_action_entry *entry,
3821 u32 *index_inc,
3822 struct netlink_ext_ack *extack)
3823{
3824#ifdef CONFIG_NET_CLS_ACT
3825 if (act->ops->offload_act_setup) {
3826 return act->ops->offload_act_setup(act, entry, index_inc, true,
3827 extack);
3828 } else {
3829 NL_SET_ERR_MSG(extack, "Action does not support offload");
3830 return -EOPNOTSUPP;
3831 }
3832#else
3833 return 0;
3834#endif
3835}
3836
3837int tc_setup_action(struct flow_action *flow_action,
3838 struct tc_action *actions[],
3839 u32 miss_cookie_base,
3840 struct netlink_ext_ack *extack)
3841{
3842 int i, j, k, index, err = 0;
3843 struct tc_action *act;
3844
3845 BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3846 BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3847 BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3848
3849 if (!actions)
3850 return 0;
3851
3852 j = 0;
3853 tcf_act_for_each_action(i, act, actions) {
3854 struct flow_action_entry *entry;
3855
3856 entry = &flow_action->entries[j];
3857 spin_lock_bh(&act->tcfa_lock);
3858 err = tcf_act_get_user_cookie(entry, act);
3859 if (err)
3860 goto err_out_locked;
3861
3862 index = 0;
3863 err = tc_setup_offload_act(act, entry, &index, extack);
3864 if (err)
3865 goto err_out_locked;
3866
3867 for (k = 0; k < index ; k++) {
3868 entry[k].hw_stats = tc_act_hw_stats(act->hw_stats);
3869 entry[k].hw_index = act->tcfa_index;
3870 entry[k].cookie = (unsigned long)act;
3871 entry[k].miss_cookie =
3872 tcf_exts_miss_cookie_get(miss_cookie_base, i);
3873 }
3874
3875 j += index;
3876
3877 spin_unlock_bh(&act->tcfa_lock);
3878 }
3879
3880err_out:
3881 if (err)
3882 tc_cleanup_offload_action(flow_action);
3883
3884 return err;
3885err_out_locked:
3886 spin_unlock_bh(&act->tcfa_lock);
3887 goto err_out;
3888}
3889
3890int tc_setup_offload_action(struct flow_action *flow_action,
3891 const struct tcf_exts *exts,
3892 struct netlink_ext_ack *extack)
3893{
3894#ifdef CONFIG_NET_CLS_ACT
3895 u32 miss_cookie_base;
3896
3897 if (!exts)
3898 return 0;
3899
3900 miss_cookie_base = exts->miss_cookie_node ?
3901 exts->miss_cookie_node->miss_cookie_base : 0;
3902 return tc_setup_action(flow_action, exts->actions, miss_cookie_base,
3903 extack);
3904#else
3905 return 0;
3906#endif
3907}
3908EXPORT_SYMBOL(tc_setup_offload_action);
3909
3910unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3911{
3912 unsigned int num_acts = 0;
3913 struct tc_action *act;
3914 int i;
3915
3916 tcf_exts_for_each_action(i, act, exts) {
3917 if (is_tcf_pedit(act))
3918 num_acts += tcf_pedit_nkeys(act);
3919 else
3920 num_acts++;
3921 }
3922 return num_acts;
3923}
3924EXPORT_SYMBOL(tcf_exts_num_actions);
3925
3926#ifdef CONFIG_NET_CLS_ACT
3927static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3928 u32 *p_block_index,
3929 struct netlink_ext_ack *extack)
3930{
3931 *p_block_index = nla_get_u32(block_index_attr);
3932 if (!*p_block_index) {
3933 NL_SET_ERR_MSG(extack, "Block number may not be zero");
3934 return -EINVAL;
3935 }
3936
3937 return 0;
3938}
3939
3940int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3941 enum flow_block_binder_type binder_type,
3942 struct nlattr *block_index_attr,
3943 struct netlink_ext_ack *extack)
3944{
3945 u32 block_index;
3946 int err;
3947
3948 if (!block_index_attr)
3949 return 0;
3950
3951 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3952 if (err)
3953 return err;
3954
3955 qe->info.binder_type = binder_type;
3956 qe->info.chain_head_change = tcf_chain_head_change_dflt;
3957 qe->info.chain_head_change_priv = &qe->filter_chain;
3958 qe->info.block_index = block_index;
3959
3960 return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3961}
3962EXPORT_SYMBOL(tcf_qevent_init);
3963
3964void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3965{
3966 if (qe->info.block_index)
3967 tcf_block_put_ext(qe->block, sch, &qe->info);
3968}
3969EXPORT_SYMBOL(tcf_qevent_destroy);
3970
3971int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3972 struct netlink_ext_ack *extack)
3973{
3974 u32 block_index;
3975 int err;
3976
3977 if (!block_index_attr)
3978 return 0;
3979
3980 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3981 if (err)
3982 return err;
3983
3984 /* Bounce newly-configured block or change in block. */
3985 if (block_index != qe->info.block_index) {
3986 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3987 return -EINVAL;
3988 }
3989
3990 return 0;
3991}
3992EXPORT_SYMBOL(tcf_qevent_validate_change);
3993
3994struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
3995 struct sk_buff **to_free, int *ret)
3996{
3997 struct tcf_result cl_res;
3998 struct tcf_proto *fl;
3999
4000 if (!qe->info.block_index)
4001 return skb;
4002
4003 fl = rcu_dereference_bh(qe->filter_chain);
4004
4005 switch (tcf_classify(skb, NULL, fl, &cl_res, false)) {
4006 case TC_ACT_SHOT:
4007 qdisc_qstats_drop(sch);
4008 __qdisc_drop(skb, to_free);
4009 *ret = __NET_XMIT_BYPASS;
4010 return NULL;
4011 case TC_ACT_STOLEN:
4012 case TC_ACT_QUEUED:
4013 case TC_ACT_TRAP:
4014 __qdisc_drop(skb, to_free);
4015 *ret = __NET_XMIT_STOLEN;
4016 return NULL;
4017 case TC_ACT_REDIRECT:
4018 skb_do_redirect(skb);
4019 *ret = __NET_XMIT_STOLEN;
4020 return NULL;
4021 }
4022
4023 return skb;
4024}
4025EXPORT_SYMBOL(tcf_qevent_handle);
4026
4027int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
4028{
4029 if (!qe->info.block_index)
4030 return 0;
4031 return nla_put_u32(skb, attr_name, qe->info.block_index);
4032}
4033EXPORT_SYMBOL(tcf_qevent_dump);
4034#endif
4035
4036static __net_init int tcf_net_init(struct net *net)
4037{
4038 struct tcf_net *tn = net_generic(net, tcf_net_id);
4039
4040 spin_lock_init(&tn->idr_lock);
4041 idr_init(&tn->idr);
4042 return 0;
4043}
4044
4045static void __net_exit tcf_net_exit(struct net *net)
4046{
4047 struct tcf_net *tn = net_generic(net, tcf_net_id);
4048
4049 idr_destroy(&tn->idr);
4050}
4051
4052static struct pernet_operations tcf_net_ops = {
4053 .init = tcf_net_init,
4054 .exit = tcf_net_exit,
4055 .id = &tcf_net_id,
4056 .size = sizeof(struct tcf_net),
4057};
4058
4059static const struct rtnl_msg_handler tc_filter_rtnl_msg_handlers[] __initconst = {
4060 {.msgtype = RTM_NEWTFILTER, .doit = tc_new_tfilter,
4061 .flags = RTNL_FLAG_DOIT_UNLOCKED},
4062 {.msgtype = RTM_DELTFILTER, .doit = tc_del_tfilter,
4063 .flags = RTNL_FLAG_DOIT_UNLOCKED},
4064 {.msgtype = RTM_GETTFILTER, .doit = tc_get_tfilter,
4065 .dumpit = tc_dump_tfilter, .flags = RTNL_FLAG_DOIT_UNLOCKED},
4066 {.msgtype = RTM_NEWCHAIN, .doit = tc_ctl_chain},
4067 {.msgtype = RTM_DELCHAIN, .doit = tc_ctl_chain},
4068 {.msgtype = RTM_GETCHAIN, .doit = tc_ctl_chain,
4069 .dumpit = tc_dump_chain},
4070};
4071
4072static int __init tc_filter_init(void)
4073{
4074 int err;
4075
4076 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
4077 if (!tc_filter_wq)
4078 return -ENOMEM;
4079
4080 err = register_pernet_subsys(&tcf_net_ops);
4081 if (err)
4082 goto err_register_pernet_subsys;
4083
4084 xa_init_flags(&tcf_exts_miss_cookies_xa, XA_FLAGS_ALLOC1);
4085 rtnl_register_many(tc_filter_rtnl_msg_handlers);
4086
4087 return 0;
4088
4089err_register_pernet_subsys:
4090 destroy_workqueue(tc_filter_wq);
4091 return err;
4092}
4093
4094subsys_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_strlcpy(name, kind, IFNAMSIZ) >= IFNAMSIZ;
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 INIT_LIST_HEAD(&bo->cb_list);
638}
639
640static void tcf_block_unbind(struct tcf_block *block,
641 struct flow_block_offload *bo);
642
643static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
644{
645 struct tcf_block *block = block_cb->indr.data;
646 struct net_device *dev = block_cb->indr.dev;
647 struct Qdisc *sch = block_cb->indr.sch;
648 struct netlink_ext_ack extack = {};
649 struct flow_block_offload bo;
650
651 tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
652 block_cb->indr.binder_type,
653 &block->flow_block, tcf_block_shared(block),
654 &extack);
655 down_write(&block->cb_lock);
656 list_del(&block_cb->driver_list);
657 list_move(&block_cb->list, &bo.cb_list);
658 up_write(&block->cb_lock);
659 rtnl_lock();
660 tcf_block_unbind(block, &bo);
661 rtnl_unlock();
662}
663
664static bool tcf_block_offload_in_use(struct tcf_block *block)
665{
666 return atomic_read(&block->offloadcnt);
667}
668
669static int tcf_block_offload_cmd(struct tcf_block *block,
670 struct net_device *dev, struct Qdisc *sch,
671 struct tcf_block_ext_info *ei,
672 enum flow_block_command command,
673 struct netlink_ext_ack *extack)
674{
675 struct flow_block_offload bo = {};
676
677 tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
678 &block->flow_block, tcf_block_shared(block),
679 extack);
680
681 if (dev->netdev_ops->ndo_setup_tc) {
682 int err;
683
684 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
685 if (err < 0) {
686 if (err != -EOPNOTSUPP)
687 NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
688 return err;
689 }
690
691 return tcf_block_setup(block, &bo);
692 }
693
694 flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
695 tc_block_indr_cleanup);
696 tcf_block_setup(block, &bo);
697
698 return -EOPNOTSUPP;
699}
700
701static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
702 struct tcf_block_ext_info *ei,
703 struct netlink_ext_ack *extack)
704{
705 struct net_device *dev = q->dev_queue->dev;
706 int err;
707
708 down_write(&block->cb_lock);
709
710 /* If tc offload feature is disabled and the block we try to bind
711 * to already has some offloaded filters, forbid to bind.
712 */
713 if (dev->netdev_ops->ndo_setup_tc &&
714 !tc_can_offload(dev) &&
715 tcf_block_offload_in_use(block)) {
716 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
717 err = -EOPNOTSUPP;
718 goto err_unlock;
719 }
720
721 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
722 if (err == -EOPNOTSUPP)
723 goto no_offload_dev_inc;
724 if (err)
725 goto err_unlock;
726
727 up_write(&block->cb_lock);
728 return 0;
729
730no_offload_dev_inc:
731 if (tcf_block_offload_in_use(block))
732 goto err_unlock;
733
734 err = 0;
735 block->nooffloaddevcnt++;
736err_unlock:
737 up_write(&block->cb_lock);
738 return err;
739}
740
741static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
742 struct tcf_block_ext_info *ei)
743{
744 struct net_device *dev = q->dev_queue->dev;
745 int err;
746
747 down_write(&block->cb_lock);
748 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
749 if (err == -EOPNOTSUPP)
750 goto no_offload_dev_dec;
751 up_write(&block->cb_lock);
752 return;
753
754no_offload_dev_dec:
755 WARN_ON(block->nooffloaddevcnt-- == 0);
756 up_write(&block->cb_lock);
757}
758
759static int
760tcf_chain0_head_change_cb_add(struct tcf_block *block,
761 struct tcf_block_ext_info *ei,
762 struct netlink_ext_ack *extack)
763{
764 struct tcf_filter_chain_list_item *item;
765 struct tcf_chain *chain0;
766
767 item = kmalloc(sizeof(*item), GFP_KERNEL);
768 if (!item) {
769 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
770 return -ENOMEM;
771 }
772 item->chain_head_change = ei->chain_head_change;
773 item->chain_head_change_priv = ei->chain_head_change_priv;
774
775 mutex_lock(&block->lock);
776 chain0 = block->chain0.chain;
777 if (chain0)
778 tcf_chain_hold(chain0);
779 else
780 list_add(&item->list, &block->chain0.filter_chain_list);
781 mutex_unlock(&block->lock);
782
783 if (chain0) {
784 struct tcf_proto *tp_head;
785
786 mutex_lock(&chain0->filter_chain_lock);
787
788 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
789 if (tp_head)
790 tcf_chain_head_change_item(item, tp_head);
791
792 mutex_lock(&block->lock);
793 list_add(&item->list, &block->chain0.filter_chain_list);
794 mutex_unlock(&block->lock);
795
796 mutex_unlock(&chain0->filter_chain_lock);
797 tcf_chain_put(chain0);
798 }
799
800 return 0;
801}
802
803static void
804tcf_chain0_head_change_cb_del(struct tcf_block *block,
805 struct tcf_block_ext_info *ei)
806{
807 struct tcf_filter_chain_list_item *item;
808
809 mutex_lock(&block->lock);
810 list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
811 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
812 (item->chain_head_change == ei->chain_head_change &&
813 item->chain_head_change_priv == ei->chain_head_change_priv)) {
814 if (block->chain0.chain)
815 tcf_chain_head_change_item(item, NULL);
816 list_del(&item->list);
817 mutex_unlock(&block->lock);
818
819 kfree(item);
820 return;
821 }
822 }
823 mutex_unlock(&block->lock);
824 WARN_ON(1);
825}
826
827struct tcf_net {
828 spinlock_t idr_lock; /* Protects idr */
829 struct idr idr;
830};
831
832static unsigned int tcf_net_id;
833
834static int tcf_block_insert(struct tcf_block *block, struct net *net,
835 struct netlink_ext_ack *extack)
836{
837 struct tcf_net *tn = net_generic(net, tcf_net_id);
838 int err;
839
840 idr_preload(GFP_KERNEL);
841 spin_lock(&tn->idr_lock);
842 err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
843 GFP_NOWAIT);
844 spin_unlock(&tn->idr_lock);
845 idr_preload_end();
846
847 return err;
848}
849
850static void tcf_block_remove(struct tcf_block *block, struct net *net)
851{
852 struct tcf_net *tn = net_generic(net, tcf_net_id);
853
854 spin_lock(&tn->idr_lock);
855 idr_remove(&tn->idr, block->index);
856 spin_unlock(&tn->idr_lock);
857}
858
859static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
860 u32 block_index,
861 struct netlink_ext_ack *extack)
862{
863 struct tcf_block *block;
864
865 block = kzalloc(sizeof(*block), GFP_KERNEL);
866 if (!block) {
867 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
868 return ERR_PTR(-ENOMEM);
869 }
870 mutex_init(&block->lock);
871 mutex_init(&block->proto_destroy_lock);
872 init_rwsem(&block->cb_lock);
873 flow_block_init(&block->flow_block);
874 INIT_LIST_HEAD(&block->chain_list);
875 INIT_LIST_HEAD(&block->owner_list);
876 INIT_LIST_HEAD(&block->chain0.filter_chain_list);
877
878 refcount_set(&block->refcnt, 1);
879 block->net = net;
880 block->index = block_index;
881
882 /* Don't store q pointer for blocks which are shared */
883 if (!tcf_block_shared(block))
884 block->q = q;
885 return block;
886}
887
888static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
889{
890 struct tcf_net *tn = net_generic(net, tcf_net_id);
891
892 return idr_find(&tn->idr, block_index);
893}
894
895static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
896{
897 struct tcf_block *block;
898
899 rcu_read_lock();
900 block = tcf_block_lookup(net, block_index);
901 if (block && !refcount_inc_not_zero(&block->refcnt))
902 block = NULL;
903 rcu_read_unlock();
904
905 return block;
906}
907
908static struct tcf_chain *
909__tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
910{
911 mutex_lock(&block->lock);
912 if (chain)
913 chain = list_is_last(&chain->list, &block->chain_list) ?
914 NULL : list_next_entry(chain, list);
915 else
916 chain = list_first_entry_or_null(&block->chain_list,
917 struct tcf_chain, list);
918
919 /* skip all action-only chains */
920 while (chain && tcf_chain_held_by_acts_only(chain))
921 chain = list_is_last(&chain->list, &block->chain_list) ?
922 NULL : list_next_entry(chain, list);
923
924 if (chain)
925 tcf_chain_hold(chain);
926 mutex_unlock(&block->lock);
927
928 return chain;
929}
930
931/* Function to be used by all clients that want to iterate over all chains on
932 * block. It properly obtains block->lock and takes reference to chain before
933 * returning it. Users of this function must be tolerant to concurrent chain
934 * insertion/deletion or ensure that no concurrent chain modification is
935 * possible. Note that all netlink dump callbacks cannot guarantee to provide
936 * consistent dump because rtnl lock is released each time skb is filled with
937 * data and sent to user-space.
938 */
939
940struct tcf_chain *
941tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
942{
943 struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
944
945 if (chain)
946 tcf_chain_put(chain);
947
948 return chain_next;
949}
950EXPORT_SYMBOL(tcf_get_next_chain);
951
952static struct tcf_proto *
953__tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
954{
955 u32 prio = 0;
956
957 ASSERT_RTNL();
958 mutex_lock(&chain->filter_chain_lock);
959
960 if (!tp) {
961 tp = tcf_chain_dereference(chain->filter_chain, chain);
962 } else if (tcf_proto_is_deleting(tp)) {
963 /* 'deleting' flag is set and chain->filter_chain_lock was
964 * unlocked, which means next pointer could be invalid. Restart
965 * search.
966 */
967 prio = tp->prio + 1;
968 tp = tcf_chain_dereference(chain->filter_chain, chain);
969
970 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
971 if (!tp->deleting && tp->prio >= prio)
972 break;
973 } else {
974 tp = tcf_chain_dereference(tp->next, chain);
975 }
976
977 if (tp)
978 tcf_proto_get(tp);
979
980 mutex_unlock(&chain->filter_chain_lock);
981
982 return tp;
983}
984
985/* Function to be used by all clients that want to iterate over all tp's on
986 * chain. Users of this function must be tolerant to concurrent tp
987 * insertion/deletion or ensure that no concurrent chain modification is
988 * possible. Note that all netlink dump callbacks cannot guarantee to provide
989 * consistent dump because rtnl lock is released each time skb is filled with
990 * data and sent to user-space.
991 */
992
993struct tcf_proto *
994tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp,
995 bool rtnl_held)
996{
997 struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
998
999 if (tp)
1000 tcf_proto_put(tp, rtnl_held, 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 = 4;
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 = skb_ext_add(skb, TC_SKB_EXT);
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 }
1634
1635 return ret;
1636#endif
1637}
1638EXPORT_SYMBOL(tcf_classify_ingress);
1639
1640struct tcf_chain_info {
1641 struct tcf_proto __rcu **pprev;
1642 struct tcf_proto __rcu *next;
1643};
1644
1645static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1646 struct tcf_chain_info *chain_info)
1647{
1648 return tcf_chain_dereference(*chain_info->pprev, chain);
1649}
1650
1651static int tcf_chain_tp_insert(struct tcf_chain *chain,
1652 struct tcf_chain_info *chain_info,
1653 struct tcf_proto *tp)
1654{
1655 if (chain->flushing)
1656 return -EAGAIN;
1657
1658 if (*chain_info->pprev == chain->filter_chain)
1659 tcf_chain0_head_change(chain, tp);
1660 tcf_proto_get(tp);
1661 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1662 rcu_assign_pointer(*chain_info->pprev, tp);
1663
1664 return 0;
1665}
1666
1667static void tcf_chain_tp_remove(struct tcf_chain *chain,
1668 struct tcf_chain_info *chain_info,
1669 struct tcf_proto *tp)
1670{
1671 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1672
1673 tcf_proto_mark_delete(tp);
1674 if (tp == chain->filter_chain)
1675 tcf_chain0_head_change(chain, next);
1676 RCU_INIT_POINTER(*chain_info->pprev, next);
1677}
1678
1679static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1680 struct tcf_chain_info *chain_info,
1681 u32 protocol, u32 prio,
1682 bool prio_allocate);
1683
1684/* Try to insert new proto.
1685 * If proto with specified priority already exists, free new proto
1686 * and return existing one.
1687 */
1688
1689static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1690 struct tcf_proto *tp_new,
1691 u32 protocol, u32 prio,
1692 bool rtnl_held)
1693{
1694 struct tcf_chain_info chain_info;
1695 struct tcf_proto *tp;
1696 int err = 0;
1697
1698 mutex_lock(&chain->filter_chain_lock);
1699
1700 if (tcf_proto_exists_destroying(chain, tp_new)) {
1701 mutex_unlock(&chain->filter_chain_lock);
1702 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1703 return ERR_PTR(-EAGAIN);
1704 }
1705
1706 tp = tcf_chain_tp_find(chain, &chain_info,
1707 protocol, prio, false);
1708 if (!tp)
1709 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1710 mutex_unlock(&chain->filter_chain_lock);
1711
1712 if (tp) {
1713 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1714 tp_new = tp;
1715 } else if (err) {
1716 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1717 tp_new = ERR_PTR(err);
1718 }
1719
1720 return tp_new;
1721}
1722
1723static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1724 struct tcf_proto *tp, bool rtnl_held,
1725 struct netlink_ext_ack *extack)
1726{
1727 struct tcf_chain_info chain_info;
1728 struct tcf_proto *tp_iter;
1729 struct tcf_proto **pprev;
1730 struct tcf_proto *next;
1731
1732 mutex_lock(&chain->filter_chain_lock);
1733
1734 /* Atomically find and remove tp from chain. */
1735 for (pprev = &chain->filter_chain;
1736 (tp_iter = tcf_chain_dereference(*pprev, chain));
1737 pprev = &tp_iter->next) {
1738 if (tp_iter == tp) {
1739 chain_info.pprev = pprev;
1740 chain_info.next = tp_iter->next;
1741 WARN_ON(tp_iter->deleting);
1742 break;
1743 }
1744 }
1745 /* Verify that tp still exists and no new filters were inserted
1746 * concurrently.
1747 * Mark tp for deletion if it is empty.
1748 */
1749 if (!tp_iter || !tcf_proto_check_delete(tp)) {
1750 mutex_unlock(&chain->filter_chain_lock);
1751 return;
1752 }
1753
1754 tcf_proto_signal_destroying(chain, tp);
1755 next = tcf_chain_dereference(chain_info.next, chain);
1756 if (tp == chain->filter_chain)
1757 tcf_chain0_head_change(chain, next);
1758 RCU_INIT_POINTER(*chain_info.pprev, next);
1759 mutex_unlock(&chain->filter_chain_lock);
1760
1761 tcf_proto_put(tp, rtnl_held, extack);
1762}
1763
1764static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1765 struct tcf_chain_info *chain_info,
1766 u32 protocol, u32 prio,
1767 bool prio_allocate)
1768{
1769 struct tcf_proto **pprev;
1770 struct tcf_proto *tp;
1771
1772 /* Check the chain for existence of proto-tcf with this priority */
1773 for (pprev = &chain->filter_chain;
1774 (tp = tcf_chain_dereference(*pprev, chain));
1775 pprev = &tp->next) {
1776 if (tp->prio >= prio) {
1777 if (tp->prio == prio) {
1778 if (prio_allocate ||
1779 (tp->protocol != protocol && protocol))
1780 return ERR_PTR(-EINVAL);
1781 } else {
1782 tp = NULL;
1783 }
1784 break;
1785 }
1786 }
1787 chain_info->pprev = pprev;
1788 if (tp) {
1789 chain_info->next = tp->next;
1790 tcf_proto_get(tp);
1791 } else {
1792 chain_info->next = NULL;
1793 }
1794 return tp;
1795}
1796
1797static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1798 struct tcf_proto *tp, struct tcf_block *block,
1799 struct Qdisc *q, u32 parent, void *fh,
1800 u32 portid, u32 seq, u16 flags, int event,
1801 bool terse_dump, bool rtnl_held)
1802{
1803 struct tcmsg *tcm;
1804 struct nlmsghdr *nlh;
1805 unsigned char *b = skb_tail_pointer(skb);
1806
1807 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1808 if (!nlh)
1809 goto out_nlmsg_trim;
1810 tcm = nlmsg_data(nlh);
1811 tcm->tcm_family = AF_UNSPEC;
1812 tcm->tcm__pad1 = 0;
1813 tcm->tcm__pad2 = 0;
1814 if (q) {
1815 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1816 tcm->tcm_parent = parent;
1817 } else {
1818 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1819 tcm->tcm_block_index = block->index;
1820 }
1821 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1822 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1823 goto nla_put_failure;
1824 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1825 goto nla_put_failure;
1826 if (!fh) {
1827 tcm->tcm_handle = 0;
1828 } else if (terse_dump) {
1829 if (tp->ops->terse_dump) {
1830 if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
1831 rtnl_held) < 0)
1832 goto nla_put_failure;
1833 } else {
1834 goto cls_op_not_supp;
1835 }
1836 } else {
1837 if (tp->ops->dump &&
1838 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
1839 goto nla_put_failure;
1840 }
1841 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1842 return skb->len;
1843
1844out_nlmsg_trim:
1845nla_put_failure:
1846cls_op_not_supp:
1847 nlmsg_trim(skb, b);
1848 return -1;
1849}
1850
1851static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1852 struct nlmsghdr *n, struct tcf_proto *tp,
1853 struct tcf_block *block, struct Qdisc *q,
1854 u32 parent, void *fh, int event, bool unicast,
1855 bool rtnl_held)
1856{
1857 struct sk_buff *skb;
1858 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1859 int err = 0;
1860
1861 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1862 if (!skb)
1863 return -ENOBUFS;
1864
1865 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1866 n->nlmsg_seq, n->nlmsg_flags, event,
1867 false, rtnl_held) <= 0) {
1868 kfree_skb(skb);
1869 return -EINVAL;
1870 }
1871
1872 if (unicast)
1873 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1874 else
1875 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1876 n->nlmsg_flags & NLM_F_ECHO);
1877
1878 if (err > 0)
1879 err = 0;
1880 return err;
1881}
1882
1883static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1884 struct nlmsghdr *n, struct tcf_proto *tp,
1885 struct tcf_block *block, struct Qdisc *q,
1886 u32 parent, void *fh, bool unicast, bool *last,
1887 bool rtnl_held, struct netlink_ext_ack *extack)
1888{
1889 struct sk_buff *skb;
1890 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1891 int err;
1892
1893 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1894 if (!skb)
1895 return -ENOBUFS;
1896
1897 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1898 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
1899 false, rtnl_held) <= 0) {
1900 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1901 kfree_skb(skb);
1902 return -EINVAL;
1903 }
1904
1905 err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
1906 if (err) {
1907 kfree_skb(skb);
1908 return err;
1909 }
1910
1911 if (unicast)
1912 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1913 else
1914 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1915 n->nlmsg_flags & NLM_F_ECHO);
1916 if (err < 0)
1917 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1918
1919 if (err > 0)
1920 err = 0;
1921 return err;
1922}
1923
1924static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1925 struct tcf_block *block, struct Qdisc *q,
1926 u32 parent, struct nlmsghdr *n,
1927 struct tcf_chain *chain, int event,
1928 bool rtnl_held)
1929{
1930 struct tcf_proto *tp;
1931
1932 for (tp = tcf_get_next_proto(chain, NULL, rtnl_held);
1933 tp; tp = tcf_get_next_proto(chain, tp, rtnl_held))
1934 tfilter_notify(net, oskb, n, tp, block,
1935 q, parent, NULL, event, false, rtnl_held);
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, rtnl_held);
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, true);
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_seq, 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 u32 parent;
2944 int err;
2945
2946 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2947 return skb->len;
2948
2949 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2950 rtm_tca_policy, cb->extack);
2951 if (err)
2952 return err;
2953
2954 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2955 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2956 if (!block)
2957 goto out;
2958 /* If we work with block index, q is NULL and parent value
2959 * will never be used in the following code. The check
2960 * in tcf_fill_node prevents it. However, compiler does not
2961 * see that far, so set parent to zero to silence the warning
2962 * about parent being uninitialized.
2963 */
2964 parent = 0;
2965 } else {
2966 const struct Qdisc_class_ops *cops;
2967 struct net_device *dev;
2968 unsigned long cl = 0;
2969
2970 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2971 if (!dev)
2972 return skb->len;
2973
2974 parent = tcm->tcm_parent;
2975 if (!parent) {
2976 q = dev->qdisc;
2977 parent = q->handle;
2978 } else {
2979 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2980 }
2981 if (!q)
2982 goto out;
2983 cops = q->ops->cl_ops;
2984 if (!cops)
2985 goto out;
2986 if (!cops->tcf_block)
2987 goto out;
2988 if (TC_H_MIN(tcm->tcm_parent)) {
2989 cl = cops->find(q, tcm->tcm_parent);
2990 if (cl == 0)
2991 goto out;
2992 }
2993 block = cops->tcf_block(q, cl, NULL);
2994 if (!block)
2995 goto out;
2996 if (tcf_block_shared(block))
2997 q = NULL;
2998 }
2999
3000 index_start = cb->args[0];
3001 index = 0;
3002
3003 mutex_lock(&block->lock);
3004 list_for_each_entry(chain, &block->chain_list, list) {
3005 if ((tca[TCA_CHAIN] &&
3006 nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3007 continue;
3008 if (index < index_start) {
3009 index++;
3010 continue;
3011 }
3012 if (tcf_chain_held_by_acts_only(chain))
3013 continue;
3014 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3015 chain->index, net, skb, block,
3016 NETLINK_CB(cb->skb).portid,
3017 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3018 RTM_NEWCHAIN);
3019 if (err <= 0)
3020 break;
3021 index++;
3022 }
3023 mutex_unlock(&block->lock);
3024
3025 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3026 tcf_block_refcnt_put(block, true);
3027 cb->args[0] = index;
3028
3029out:
3030 /* If we did no progress, the error (EMSGSIZE) is real */
3031 if (skb->len == 0 && err)
3032 return err;
3033 return skb->len;
3034}
3035
3036void tcf_exts_destroy(struct tcf_exts *exts)
3037{
3038#ifdef CONFIG_NET_CLS_ACT
3039 if (exts->actions) {
3040 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3041 kfree(exts->actions);
3042 }
3043 exts->nr_actions = 0;
3044#endif
3045}
3046EXPORT_SYMBOL(tcf_exts_destroy);
3047
3048int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3049 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
3050 bool rtnl_held, struct netlink_ext_ack *extack)
3051{
3052#ifdef CONFIG_NET_CLS_ACT
3053 {
3054 struct tc_action *act;
3055 size_t attr_size = 0;
3056
3057 if (exts->police && tb[exts->police]) {
3058 act = tcf_action_init_1(net, tp, tb[exts->police],
3059 rate_tlv, "police", ovr,
3060 TCA_ACT_BIND, rtnl_held,
3061 extack);
3062 if (IS_ERR(act))
3063 return PTR_ERR(act);
3064
3065 act->type = exts->type = TCA_OLD_COMPAT;
3066 exts->actions[0] = act;
3067 exts->nr_actions = 1;
3068 } else if (exts->action && tb[exts->action]) {
3069 int err;
3070
3071 err = tcf_action_init(net, tp, tb[exts->action],
3072 rate_tlv, NULL, ovr, TCA_ACT_BIND,
3073 exts->actions, &attr_size,
3074 rtnl_held, extack);
3075 if (err < 0)
3076 return err;
3077 exts->nr_actions = err;
3078 }
3079 }
3080#else
3081 if ((exts->action && tb[exts->action]) ||
3082 (exts->police && tb[exts->police])) {
3083 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3084 return -EOPNOTSUPP;
3085 }
3086#endif
3087
3088 return 0;
3089}
3090EXPORT_SYMBOL(tcf_exts_validate);
3091
3092void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3093{
3094#ifdef CONFIG_NET_CLS_ACT
3095 struct tcf_exts old = *dst;
3096
3097 *dst = *src;
3098 tcf_exts_destroy(&old);
3099#endif
3100}
3101EXPORT_SYMBOL(tcf_exts_change);
3102
3103#ifdef CONFIG_NET_CLS_ACT
3104static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3105{
3106 if (exts->nr_actions == 0)
3107 return NULL;
3108 else
3109 return exts->actions[0];
3110}
3111#endif
3112
3113int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3114{
3115#ifdef CONFIG_NET_CLS_ACT
3116 struct nlattr *nest;
3117
3118 if (exts->action && tcf_exts_has_actions(exts)) {
3119 /*
3120 * again for backward compatible mode - we want
3121 * to work with both old and new modes of entering
3122 * tc data even if iproute2 was newer - jhs
3123 */
3124 if (exts->type != TCA_OLD_COMPAT) {
3125 nest = nla_nest_start_noflag(skb, exts->action);
3126 if (nest == NULL)
3127 goto nla_put_failure;
3128
3129 if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3130 < 0)
3131 goto nla_put_failure;
3132 nla_nest_end(skb, nest);
3133 } else if (exts->police) {
3134 struct tc_action *act = tcf_exts_first_act(exts);
3135 nest = nla_nest_start_noflag(skb, exts->police);
3136 if (nest == NULL || !act)
3137 goto nla_put_failure;
3138 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3139 goto nla_put_failure;
3140 nla_nest_end(skb, nest);
3141 }
3142 }
3143 return 0;
3144
3145nla_put_failure:
3146 nla_nest_cancel(skb, nest);
3147 return -1;
3148#else
3149 return 0;
3150#endif
3151}
3152EXPORT_SYMBOL(tcf_exts_dump);
3153
3154int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3155{
3156#ifdef CONFIG_NET_CLS_ACT
3157 struct nlattr *nest;
3158
3159 if (!exts->action || !tcf_exts_has_actions(exts))
3160 return 0;
3161
3162 nest = nla_nest_start_noflag(skb, exts->action);
3163 if (!nest)
3164 goto nla_put_failure;
3165
3166 if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3167 goto nla_put_failure;
3168 nla_nest_end(skb, nest);
3169 return 0;
3170
3171nla_put_failure:
3172 nla_nest_cancel(skb, nest);
3173 return -1;
3174#else
3175 return 0;
3176#endif
3177}
3178EXPORT_SYMBOL(tcf_exts_terse_dump);
3179
3180int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3181{
3182#ifdef CONFIG_NET_CLS_ACT
3183 struct tc_action *a = tcf_exts_first_act(exts);
3184 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3185 return -1;
3186#endif
3187 return 0;
3188}
3189EXPORT_SYMBOL(tcf_exts_dump_stats);
3190
3191static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3192{
3193 if (*flags & TCA_CLS_FLAGS_IN_HW)
3194 return;
3195 *flags |= TCA_CLS_FLAGS_IN_HW;
3196 atomic_inc(&block->offloadcnt);
3197}
3198
3199static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3200{
3201 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3202 return;
3203 *flags &= ~TCA_CLS_FLAGS_IN_HW;
3204 atomic_dec(&block->offloadcnt);
3205}
3206
3207static void tc_cls_offload_cnt_update(struct tcf_block *block,
3208 struct tcf_proto *tp, u32 *cnt,
3209 u32 *flags, u32 diff, bool add)
3210{
3211 lockdep_assert_held(&block->cb_lock);
3212
3213 spin_lock(&tp->lock);
3214 if (add) {
3215 if (!*cnt)
3216 tcf_block_offload_inc(block, flags);
3217 *cnt += diff;
3218 } else {
3219 *cnt -= diff;
3220 if (!*cnt)
3221 tcf_block_offload_dec(block, flags);
3222 }
3223 spin_unlock(&tp->lock);
3224}
3225
3226static void
3227tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3228 u32 *cnt, u32 *flags)
3229{
3230 lockdep_assert_held(&block->cb_lock);
3231
3232 spin_lock(&tp->lock);
3233 tcf_block_offload_dec(block, flags);
3234 *cnt = 0;
3235 spin_unlock(&tp->lock);
3236}
3237
3238static int
3239__tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3240 void *type_data, bool err_stop)
3241{
3242 struct flow_block_cb *block_cb;
3243 int ok_count = 0;
3244 int err;
3245
3246 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3247 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3248 if (err) {
3249 if (err_stop)
3250 return err;
3251 } else {
3252 ok_count++;
3253 }
3254 }
3255 return ok_count;
3256}
3257
3258int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3259 void *type_data, bool err_stop, bool rtnl_held)
3260{
3261 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3262 int ok_count;
3263
3264retry:
3265 if (take_rtnl)
3266 rtnl_lock();
3267 down_read(&block->cb_lock);
3268 /* Need to obtain rtnl lock if block is bound to devs that require it.
3269 * In block bind code cb_lock is obtained while holding rtnl, so we must
3270 * obtain the locks in same order here.
3271 */
3272 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3273 up_read(&block->cb_lock);
3274 take_rtnl = true;
3275 goto retry;
3276 }
3277
3278 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3279
3280 up_read(&block->cb_lock);
3281 if (take_rtnl)
3282 rtnl_unlock();
3283 return ok_count;
3284}
3285EXPORT_SYMBOL(tc_setup_cb_call);
3286
3287/* Non-destructive filter add. If filter that wasn't already in hardware is
3288 * successfully offloaded, increment block offloads counter. On failure,
3289 * previously offloaded filter is considered to be intact and offloads counter
3290 * is not decremented.
3291 */
3292
3293int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3294 enum tc_setup_type type, void *type_data, bool err_stop,
3295 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3296{
3297 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3298 int ok_count;
3299
3300retry:
3301 if (take_rtnl)
3302 rtnl_lock();
3303 down_read(&block->cb_lock);
3304 /* Need to obtain rtnl lock if block is bound to devs that require it.
3305 * In block bind code cb_lock is obtained while holding rtnl, so we must
3306 * obtain the locks in same order here.
3307 */
3308 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3309 up_read(&block->cb_lock);
3310 take_rtnl = true;
3311 goto retry;
3312 }
3313
3314 /* Make sure all netdevs sharing this block are offload-capable. */
3315 if (block->nooffloaddevcnt && err_stop) {
3316 ok_count = -EOPNOTSUPP;
3317 goto err_unlock;
3318 }
3319
3320 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3321 if (ok_count < 0)
3322 goto err_unlock;
3323
3324 if (tp->ops->hw_add)
3325 tp->ops->hw_add(tp, type_data);
3326 if (ok_count > 0)
3327 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3328 ok_count, true);
3329err_unlock:
3330 up_read(&block->cb_lock);
3331 if (take_rtnl)
3332 rtnl_unlock();
3333 return ok_count < 0 ? ok_count : 0;
3334}
3335EXPORT_SYMBOL(tc_setup_cb_add);
3336
3337/* Destructive filter replace. If filter that wasn't already in hardware is
3338 * successfully offloaded, increment block offload counter. On failure,
3339 * previously offloaded filter is considered to be destroyed and offload counter
3340 * is decremented.
3341 */
3342
3343int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3344 enum tc_setup_type type, void *type_data, bool err_stop,
3345 u32 *old_flags, unsigned int *old_in_hw_count,
3346 u32 *new_flags, unsigned int *new_in_hw_count,
3347 bool rtnl_held)
3348{
3349 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3350 int ok_count;
3351
3352retry:
3353 if (take_rtnl)
3354 rtnl_lock();
3355 down_read(&block->cb_lock);
3356 /* Need to obtain rtnl lock if block is bound to devs that require it.
3357 * In block bind code cb_lock is obtained while holding rtnl, so we must
3358 * obtain the locks in same order here.
3359 */
3360 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3361 up_read(&block->cb_lock);
3362 take_rtnl = true;
3363 goto retry;
3364 }
3365
3366 /* Make sure all netdevs sharing this block are offload-capable. */
3367 if (block->nooffloaddevcnt && err_stop) {
3368 ok_count = -EOPNOTSUPP;
3369 goto err_unlock;
3370 }
3371
3372 tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3373 if (tp->ops->hw_del)
3374 tp->ops->hw_del(tp, type_data);
3375
3376 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3377 if (ok_count < 0)
3378 goto err_unlock;
3379
3380 if (tp->ops->hw_add)
3381 tp->ops->hw_add(tp, type_data);
3382 if (ok_count > 0)
3383 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3384 new_flags, ok_count, true);
3385err_unlock:
3386 up_read(&block->cb_lock);
3387 if (take_rtnl)
3388 rtnl_unlock();
3389 return ok_count < 0 ? ok_count : 0;
3390}
3391EXPORT_SYMBOL(tc_setup_cb_replace);
3392
3393/* Destroy filter and decrement block offload counter, if filter was previously
3394 * offloaded.
3395 */
3396
3397int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3398 enum tc_setup_type type, void *type_data, bool err_stop,
3399 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3400{
3401 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3402 int ok_count;
3403
3404retry:
3405 if (take_rtnl)
3406 rtnl_lock();
3407 down_read(&block->cb_lock);
3408 /* Need to obtain rtnl lock if block is bound to devs that require it.
3409 * In block bind code cb_lock is obtained while holding rtnl, so we must
3410 * obtain the locks in same order here.
3411 */
3412 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3413 up_read(&block->cb_lock);
3414 take_rtnl = true;
3415 goto retry;
3416 }
3417
3418 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3419
3420 tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3421 if (tp->ops->hw_del)
3422 tp->ops->hw_del(tp, type_data);
3423
3424 up_read(&block->cb_lock);
3425 if (take_rtnl)
3426 rtnl_unlock();
3427 return ok_count < 0 ? ok_count : 0;
3428}
3429EXPORT_SYMBOL(tc_setup_cb_destroy);
3430
3431int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3432 bool add, flow_setup_cb_t *cb,
3433 enum tc_setup_type type, void *type_data,
3434 void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3435{
3436 int err = cb(type, type_data, cb_priv);
3437
3438 if (err) {
3439 if (add && tc_skip_sw(*flags))
3440 return err;
3441 } else {
3442 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3443 add);
3444 }
3445
3446 return 0;
3447}
3448EXPORT_SYMBOL(tc_setup_cb_reoffload);
3449
3450static int tcf_act_get_cookie(struct flow_action_entry *entry,
3451 const struct tc_action *act)
3452{
3453 struct tc_cookie *cookie;
3454 int err = 0;
3455
3456 rcu_read_lock();
3457 cookie = rcu_dereference(act->act_cookie);
3458 if (cookie) {
3459 entry->cookie = flow_action_cookie_create(cookie->data,
3460 cookie->len,
3461 GFP_ATOMIC);
3462 if (!entry->cookie)
3463 err = -ENOMEM;
3464 }
3465 rcu_read_unlock();
3466 return err;
3467}
3468
3469static void tcf_act_put_cookie(struct flow_action_entry *entry)
3470{
3471 flow_action_cookie_destroy(entry->cookie);
3472}
3473
3474void tc_cleanup_flow_action(struct flow_action *flow_action)
3475{
3476 struct flow_action_entry *entry;
3477 int i;
3478
3479 flow_action_for_each(i, entry, flow_action) {
3480 tcf_act_put_cookie(entry);
3481 if (entry->destructor)
3482 entry->destructor(entry->destructor_priv);
3483 }
3484}
3485EXPORT_SYMBOL(tc_cleanup_flow_action);
3486
3487static void tcf_mirred_get_dev(struct flow_action_entry *entry,
3488 const struct tc_action *act)
3489{
3490#ifdef CONFIG_NET_CLS_ACT
3491 entry->dev = act->ops->get_dev(act, &entry->destructor);
3492 if (!entry->dev)
3493 return;
3494 entry->destructor_priv = entry->dev;
3495#endif
3496}
3497
3498static void tcf_tunnel_encap_put_tunnel(void *priv)
3499{
3500 struct ip_tunnel_info *tunnel = priv;
3501
3502 kfree(tunnel);
3503}
3504
3505static int tcf_tunnel_encap_get_tunnel(struct flow_action_entry *entry,
3506 const struct tc_action *act)
3507{
3508 entry->tunnel = tcf_tunnel_info_copy(act);
3509 if (!entry->tunnel)
3510 return -ENOMEM;
3511 entry->destructor = tcf_tunnel_encap_put_tunnel;
3512 entry->destructor_priv = entry->tunnel;
3513 return 0;
3514}
3515
3516static void tcf_sample_get_group(struct flow_action_entry *entry,
3517 const struct tc_action *act)
3518{
3519#ifdef CONFIG_NET_CLS_ACT
3520 entry->sample.psample_group =
3521 act->ops->get_psample_group(act, &entry->destructor);
3522 entry->destructor_priv = entry->sample.psample_group;
3523#endif
3524}
3525
3526static void tcf_gate_entry_destructor(void *priv)
3527{
3528 struct action_gate_entry *oe = priv;
3529
3530 kfree(oe);
3531}
3532
3533static int tcf_gate_get_entries(struct flow_action_entry *entry,
3534 const struct tc_action *act)
3535{
3536 entry->gate.entries = tcf_gate_get_list(act);
3537
3538 if (!entry->gate.entries)
3539 return -EINVAL;
3540
3541 entry->destructor = tcf_gate_entry_destructor;
3542 entry->destructor_priv = entry->gate.entries;
3543
3544 return 0;
3545}
3546
3547static enum flow_action_hw_stats tc_act_hw_stats(u8 hw_stats)
3548{
3549 if (WARN_ON_ONCE(hw_stats > TCA_ACT_HW_STATS_ANY))
3550 return FLOW_ACTION_HW_STATS_DONT_CARE;
3551 else if (!hw_stats)
3552 return FLOW_ACTION_HW_STATS_DISABLED;
3553
3554 return hw_stats;
3555}
3556
3557int tc_setup_flow_action(struct flow_action *flow_action,
3558 const struct tcf_exts *exts)
3559{
3560 struct tc_action *act;
3561 int i, j, k, err = 0;
3562
3563 BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3564 BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3565 BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3566
3567 if (!exts)
3568 return 0;
3569
3570 j = 0;
3571 tcf_exts_for_each_action(i, act, exts) {
3572 struct flow_action_entry *entry;
3573
3574 entry = &flow_action->entries[j];
3575 spin_lock_bh(&act->tcfa_lock);
3576 err = tcf_act_get_cookie(entry, act);
3577 if (err)
3578 goto err_out_locked;
3579
3580 entry->hw_stats = tc_act_hw_stats(act->hw_stats);
3581
3582 if (is_tcf_gact_ok(act)) {
3583 entry->id = FLOW_ACTION_ACCEPT;
3584 } else if (is_tcf_gact_shot(act)) {
3585 entry->id = FLOW_ACTION_DROP;
3586 } else if (is_tcf_gact_trap(act)) {
3587 entry->id = FLOW_ACTION_TRAP;
3588 } else if (is_tcf_gact_goto_chain(act)) {
3589 entry->id = FLOW_ACTION_GOTO;
3590 entry->chain_index = tcf_gact_goto_chain_index(act);
3591 } else if (is_tcf_mirred_egress_redirect(act)) {
3592 entry->id = FLOW_ACTION_REDIRECT;
3593 tcf_mirred_get_dev(entry, act);
3594 } else if (is_tcf_mirred_egress_mirror(act)) {
3595 entry->id = FLOW_ACTION_MIRRED;
3596 tcf_mirred_get_dev(entry, act);
3597 } else if (is_tcf_mirred_ingress_redirect(act)) {
3598 entry->id = FLOW_ACTION_REDIRECT_INGRESS;
3599 tcf_mirred_get_dev(entry, act);
3600 } else if (is_tcf_mirred_ingress_mirror(act)) {
3601 entry->id = FLOW_ACTION_MIRRED_INGRESS;
3602 tcf_mirred_get_dev(entry, act);
3603 } else if (is_tcf_vlan(act)) {
3604 switch (tcf_vlan_action(act)) {
3605 case TCA_VLAN_ACT_PUSH:
3606 entry->id = FLOW_ACTION_VLAN_PUSH;
3607 entry->vlan.vid = tcf_vlan_push_vid(act);
3608 entry->vlan.proto = tcf_vlan_push_proto(act);
3609 entry->vlan.prio = tcf_vlan_push_prio(act);
3610 break;
3611 case TCA_VLAN_ACT_POP:
3612 entry->id = FLOW_ACTION_VLAN_POP;
3613 break;
3614 case TCA_VLAN_ACT_MODIFY:
3615 entry->id = FLOW_ACTION_VLAN_MANGLE;
3616 entry->vlan.vid = tcf_vlan_push_vid(act);
3617 entry->vlan.proto = tcf_vlan_push_proto(act);
3618 entry->vlan.prio = tcf_vlan_push_prio(act);
3619 break;
3620 default:
3621 err = -EOPNOTSUPP;
3622 goto err_out_locked;
3623 }
3624 } else if (is_tcf_tunnel_set(act)) {
3625 entry->id = FLOW_ACTION_TUNNEL_ENCAP;
3626 err = tcf_tunnel_encap_get_tunnel(entry, act);
3627 if (err)
3628 goto err_out_locked;
3629 } else if (is_tcf_tunnel_release(act)) {
3630 entry->id = FLOW_ACTION_TUNNEL_DECAP;
3631 } else if (is_tcf_pedit(act)) {
3632 for (k = 0; k < tcf_pedit_nkeys(act); k++) {
3633 switch (tcf_pedit_cmd(act, k)) {
3634 case TCA_PEDIT_KEY_EX_CMD_SET:
3635 entry->id = FLOW_ACTION_MANGLE;
3636 break;
3637 case TCA_PEDIT_KEY_EX_CMD_ADD:
3638 entry->id = FLOW_ACTION_ADD;
3639 break;
3640 default:
3641 err = -EOPNOTSUPP;
3642 goto err_out_locked;
3643 }
3644 entry->mangle.htype = tcf_pedit_htype(act, k);
3645 entry->mangle.mask = tcf_pedit_mask(act, k);
3646 entry->mangle.val = tcf_pedit_val(act, k);
3647 entry->mangle.offset = tcf_pedit_offset(act, k);
3648 entry->hw_stats = tc_act_hw_stats(act->hw_stats);
3649 entry = &flow_action->entries[++j];
3650 }
3651 } else if (is_tcf_csum(act)) {
3652 entry->id = FLOW_ACTION_CSUM;
3653 entry->csum_flags = tcf_csum_update_flags(act);
3654 } else if (is_tcf_skbedit_mark(act)) {
3655 entry->id = FLOW_ACTION_MARK;
3656 entry->mark = tcf_skbedit_mark(act);
3657 } else if (is_tcf_sample(act)) {
3658 entry->id = FLOW_ACTION_SAMPLE;
3659 entry->sample.trunc_size = tcf_sample_trunc_size(act);
3660 entry->sample.truncate = tcf_sample_truncate(act);
3661 entry->sample.rate = tcf_sample_rate(act);
3662 tcf_sample_get_group(entry, act);
3663 } else if (is_tcf_police(act)) {
3664 entry->id = FLOW_ACTION_POLICE;
3665 entry->police.burst = tcf_police_burst(act);
3666 entry->police.rate_bytes_ps =
3667 tcf_police_rate_bytes_ps(act);
3668 entry->police.mtu = tcf_police_tcfp_mtu(act);
3669 entry->police.index = act->tcfa_index;
3670 } else if (is_tcf_ct(act)) {
3671 entry->id = FLOW_ACTION_CT;
3672 entry->ct.action = tcf_ct_action(act);
3673 entry->ct.zone = tcf_ct_zone(act);
3674 entry->ct.flow_table = tcf_ct_ft(act);
3675 } else if (is_tcf_mpls(act)) {
3676 switch (tcf_mpls_action(act)) {
3677 case TCA_MPLS_ACT_PUSH:
3678 entry->id = FLOW_ACTION_MPLS_PUSH;
3679 entry->mpls_push.proto = tcf_mpls_proto(act);
3680 entry->mpls_push.label = tcf_mpls_label(act);
3681 entry->mpls_push.tc = tcf_mpls_tc(act);
3682 entry->mpls_push.bos = tcf_mpls_bos(act);
3683 entry->mpls_push.ttl = tcf_mpls_ttl(act);
3684 break;
3685 case TCA_MPLS_ACT_POP:
3686 entry->id = FLOW_ACTION_MPLS_POP;
3687 entry->mpls_pop.proto = tcf_mpls_proto(act);
3688 break;
3689 case TCA_MPLS_ACT_MODIFY:
3690 entry->id = FLOW_ACTION_MPLS_MANGLE;
3691 entry->mpls_mangle.label = tcf_mpls_label(act);
3692 entry->mpls_mangle.tc = tcf_mpls_tc(act);
3693 entry->mpls_mangle.bos = tcf_mpls_bos(act);
3694 entry->mpls_mangle.ttl = tcf_mpls_ttl(act);
3695 break;
3696 default:
3697 goto err_out_locked;
3698 }
3699 } else if (is_tcf_skbedit_ptype(act)) {
3700 entry->id = FLOW_ACTION_PTYPE;
3701 entry->ptype = tcf_skbedit_ptype(act);
3702 } else if (is_tcf_skbedit_priority(act)) {
3703 entry->id = FLOW_ACTION_PRIORITY;
3704 entry->priority = tcf_skbedit_priority(act);
3705 } else if (is_tcf_gate(act)) {
3706 entry->id = FLOW_ACTION_GATE;
3707 entry->gate.index = tcf_gate_index(act);
3708 entry->gate.prio = tcf_gate_prio(act);
3709 entry->gate.basetime = tcf_gate_basetime(act);
3710 entry->gate.cycletime = tcf_gate_cycletime(act);
3711 entry->gate.cycletimeext = tcf_gate_cycletimeext(act);
3712 entry->gate.num_entries = tcf_gate_num_entries(act);
3713 err = tcf_gate_get_entries(entry, act);
3714 if (err)
3715 goto err_out;
3716 } else {
3717 err = -EOPNOTSUPP;
3718 goto err_out_locked;
3719 }
3720 spin_unlock_bh(&act->tcfa_lock);
3721
3722 if (!is_tcf_pedit(act))
3723 j++;
3724 }
3725
3726err_out:
3727 if (err)
3728 tc_cleanup_flow_action(flow_action);
3729
3730 return err;
3731err_out_locked:
3732 spin_unlock_bh(&act->tcfa_lock);
3733 goto err_out;
3734}
3735EXPORT_SYMBOL(tc_setup_flow_action);
3736
3737unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3738{
3739 unsigned int num_acts = 0;
3740 struct tc_action *act;
3741 int i;
3742
3743 tcf_exts_for_each_action(i, act, exts) {
3744 if (is_tcf_pedit(act))
3745 num_acts += tcf_pedit_nkeys(act);
3746 else
3747 num_acts++;
3748 }
3749 return num_acts;
3750}
3751EXPORT_SYMBOL(tcf_exts_num_actions);
3752
3753#ifdef CONFIG_NET_CLS_ACT
3754static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3755 u32 *p_block_index,
3756 struct netlink_ext_ack *extack)
3757{
3758 *p_block_index = nla_get_u32(block_index_attr);
3759 if (!*p_block_index) {
3760 NL_SET_ERR_MSG(extack, "Block number may not be zero");
3761 return -EINVAL;
3762 }
3763
3764 return 0;
3765}
3766
3767int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3768 enum flow_block_binder_type binder_type,
3769 struct nlattr *block_index_attr,
3770 struct netlink_ext_ack *extack)
3771{
3772 u32 block_index;
3773 int err;
3774
3775 if (!block_index_attr)
3776 return 0;
3777
3778 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3779 if (err)
3780 return err;
3781
3782 if (!block_index)
3783 return 0;
3784
3785 qe->info.binder_type = binder_type;
3786 qe->info.chain_head_change = tcf_chain_head_change_dflt;
3787 qe->info.chain_head_change_priv = &qe->filter_chain;
3788 qe->info.block_index = block_index;
3789
3790 return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3791}
3792EXPORT_SYMBOL(tcf_qevent_init);
3793
3794void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3795{
3796 if (qe->info.block_index)
3797 tcf_block_put_ext(qe->block, sch, &qe->info);
3798}
3799EXPORT_SYMBOL(tcf_qevent_destroy);
3800
3801int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3802 struct netlink_ext_ack *extack)
3803{
3804 u32 block_index;
3805 int err;
3806
3807 if (!block_index_attr)
3808 return 0;
3809
3810 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3811 if (err)
3812 return err;
3813
3814 /* Bounce newly-configured block or change in block. */
3815 if (block_index != qe->info.block_index) {
3816 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3817 return -EINVAL;
3818 }
3819
3820 return 0;
3821}
3822EXPORT_SYMBOL(tcf_qevent_validate_change);
3823
3824struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
3825 struct sk_buff **to_free, int *ret)
3826{
3827 struct tcf_result cl_res;
3828 struct tcf_proto *fl;
3829
3830 if (!qe->info.block_index)
3831 return skb;
3832
3833 fl = rcu_dereference_bh(qe->filter_chain);
3834
3835 switch (tcf_classify(skb, fl, &cl_res, false)) {
3836 case TC_ACT_SHOT:
3837 qdisc_qstats_drop(sch);
3838 __qdisc_drop(skb, to_free);
3839 *ret = __NET_XMIT_BYPASS;
3840 return NULL;
3841 case TC_ACT_STOLEN:
3842 case TC_ACT_QUEUED:
3843 case TC_ACT_TRAP:
3844 __qdisc_drop(skb, to_free);
3845 *ret = __NET_XMIT_STOLEN;
3846 return NULL;
3847 case TC_ACT_REDIRECT:
3848 skb_do_redirect(skb);
3849 *ret = __NET_XMIT_STOLEN;
3850 return NULL;
3851 }
3852
3853 return skb;
3854}
3855EXPORT_SYMBOL(tcf_qevent_handle);
3856
3857int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
3858{
3859 if (!qe->info.block_index)
3860 return 0;
3861 return nla_put_u32(skb, attr_name, qe->info.block_index);
3862}
3863EXPORT_SYMBOL(tcf_qevent_dump);
3864#endif
3865
3866static __net_init int tcf_net_init(struct net *net)
3867{
3868 struct tcf_net *tn = net_generic(net, tcf_net_id);
3869
3870 spin_lock_init(&tn->idr_lock);
3871 idr_init(&tn->idr);
3872 return 0;
3873}
3874
3875static void __net_exit tcf_net_exit(struct net *net)
3876{
3877 struct tcf_net *tn = net_generic(net, tcf_net_id);
3878
3879 idr_destroy(&tn->idr);
3880}
3881
3882static struct pernet_operations tcf_net_ops = {
3883 .init = tcf_net_init,
3884 .exit = tcf_net_exit,
3885 .id = &tcf_net_id,
3886 .size = sizeof(struct tcf_net),
3887};
3888
3889static int __init tc_filter_init(void)
3890{
3891 int err;
3892
3893 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3894 if (!tc_filter_wq)
3895 return -ENOMEM;
3896
3897 err = register_pernet_subsys(&tcf_net_ops);
3898 if (err)
3899 goto err_register_pernet_subsys;
3900
3901 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3902 RTNL_FLAG_DOIT_UNLOCKED);
3903 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3904 RTNL_FLAG_DOIT_UNLOCKED);
3905 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3906 tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3907 rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3908 rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3909 rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3910 tc_dump_chain, 0);
3911
3912 return 0;
3913
3914err_register_pernet_subsys:
3915 destroy_workqueue(tc_filter_wq);
3916 return err;
3917}
3918
3919subsys_initcall(tc_filter_init);