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 */
47 static LIST_HEAD(tcf_proto_base);
48
49 /* Protects list of registered TC modules. It is pure SMP lock. */
50 static DEFINE_RWLOCK(cls_mod_lock);
51
52 static struct xarray tcf_exts_miss_cookies_xa;
53 struct 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 */
67 union 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)
76 static int
77 tcf_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)
101 goto err_xa_alloc;
102
103 exts->miss_cookie_node = n;
104 return 0;
105
106 err_xa_alloc:
107 kfree(n);
108 return err;
109 }
110
111 static 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
123 static struct tcf_exts_miss_cookie_node *
124 tcf_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) */
132 static int
133 tcf_exts_miss_cookie_base_alloc(struct tcf_exts *exts, struct tcf_proto *tp,
134 u32 handle)
135 {
136 return 0;
137 }
138
139 static void tcf_exts_miss_cookie_base_destroy(struct tcf_exts *exts)
140 {
141 }
142 #endif /* IS_ENABLED(CONFIG_NET_TC_SKB_EXT) */
143
144 static 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
156 DEFINE_STATIC_KEY_FALSE(tc_skb_ext_tc);
157 EXPORT_SYMBOL(tc_skb_ext_tc);
158
159 void tc_skb_ext_tc_enable(void)
160 {
161 static_branch_inc(&tc_skb_ext_tc);
162 }
163 EXPORT_SYMBOL(tc_skb_ext_tc_enable);
164
165 void tc_skb_ext_tc_disable(void)
166 {
167 static_branch_dec(&tc_skb_ext_tc);
168 }
169 EXPORT_SYMBOL(tc_skb_ext_tc_disable);
170 #endif
171
172 static 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
178 static 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
189 static 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
197 static 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
217 static void
218 tcf_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
230 static 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
248 static const struct tcf_proto_ops *
249 tcf_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
280 int 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;
292 out:
293 write_unlock(&cls_mod_lock);
294 return rc;
295 }
296 EXPORT_SYMBOL(register_tcf_proto_ops);
297
298 static struct workqueue_struct *tc_filter_wq;
299
300 void 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 }
323 EXPORT_SYMBOL(unregister_tcf_proto_ops);
324
325 bool 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 }
330 EXPORT_SYMBOL(tcf_queue_work);
331
332 /* Select new prio value from the range, managed by kernel. */
333
334 static 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
344 static 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
352 static 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
372 static 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 spin_lock_init(&tp->lock);
394 refcount_set(&tp->refcnt, 1);
395
396 err = tp->ops->init(tp);
397 if (err) {
398 module_put(tp->ops->owner);
399 goto errout;
400 }
401 return tp;
402
403 errout:
404 kfree(tp);
405 return ERR_PTR(err);
406 }
407
408 static void tcf_proto_get(struct tcf_proto *tp)
409 {
410 refcount_inc(&tp->refcnt);
411 }
412
413 static void tcf_maintain_bypass(struct tcf_block *block)
414 {
415 int filtercnt = atomic_read(&block->filtercnt);
416 int skipswcnt = atomic_read(&block->skipswcnt);
417 bool bypass_wanted = filtercnt > 0 && filtercnt == skipswcnt;
418
419 if (bypass_wanted != block->bypass_wanted) {
420 #ifdef CONFIG_NET_CLS_ACT
421 if (bypass_wanted)
422 static_branch_inc(&tcf_bypass_check_needed_key);
423 else
424 static_branch_dec(&tcf_bypass_check_needed_key);
425 #endif
426 block->bypass_wanted = bypass_wanted;
427 }
428 }
429
430 static void tcf_block_filter_cnt_update(struct tcf_block *block, bool *counted, bool add)
431 {
432 lockdep_assert_not_held(&block->cb_lock);
433
434 down_write(&block->cb_lock);
435 if (*counted != add) {
436 if (add) {
437 atomic_inc(&block->filtercnt);
438 *counted = true;
439 } else {
440 atomic_dec(&block->filtercnt);
441 *counted = false;
442 }
443 }
444 tcf_maintain_bypass(block);
445 up_write(&block->cb_lock);
446 }
447
448 static void tcf_chain_put(struct tcf_chain *chain);
449
450 static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
451 bool sig_destroy, struct netlink_ext_ack *extack)
452 {
453 tp->ops->destroy(tp, rtnl_held, extack);
454 tcf_block_filter_cnt_update(tp->chain->block, &tp->counted, false);
455 if (sig_destroy)
456 tcf_proto_signal_destroyed(tp->chain, tp);
457 tcf_chain_put(tp->chain);
458 module_put(tp->ops->owner);
459 kfree_rcu(tp, rcu);
460 }
461
462 static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
463 struct netlink_ext_ack *extack)
464 {
465 if (refcount_dec_and_test(&tp->refcnt))
466 tcf_proto_destroy(tp, rtnl_held, true, extack);
467 }
468
469 static bool tcf_proto_check_delete(struct tcf_proto *tp)
470 {
471 if (tp->ops->delete_empty)
472 return tp->ops->delete_empty(tp);
473
474 tp->deleting = true;
475 return tp->deleting;
476 }
477
478 static void tcf_proto_mark_delete(struct tcf_proto *tp)
479 {
480 spin_lock(&tp->lock);
481 tp->deleting = true;
482 spin_unlock(&tp->lock);
483 }
484
485 static bool tcf_proto_is_deleting(struct tcf_proto *tp)
486 {
487 bool deleting;
488
489 spin_lock(&tp->lock);
490 deleting = tp->deleting;
491 spin_unlock(&tp->lock);
492
493 return deleting;
494 }
495
496 #define ASSERT_BLOCK_LOCKED(block) \
497 lockdep_assert_held(&(block)->lock)
498
499 struct tcf_filter_chain_list_item {
500 struct list_head list;
501 tcf_chain_head_change_t *chain_head_change;
502 void *chain_head_change_priv;
503 };
504
505 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
506 u32 chain_index)
507 {
508 struct tcf_chain *chain;
509
510 ASSERT_BLOCK_LOCKED(block);
511
512 chain = kzalloc(sizeof(*chain), GFP_KERNEL);
513 if (!chain)
514 return NULL;
515 list_add_tail_rcu(&chain->list, &block->chain_list);
516 mutex_init(&chain->filter_chain_lock);
517 chain->block = block;
518 chain->index = chain_index;
519 chain->refcnt = 1;
520 if (!chain->index)
521 block->chain0.chain = chain;
522 return chain;
523 }
524
525 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
526 struct tcf_proto *tp_head)
527 {
528 if (item->chain_head_change)
529 item->chain_head_change(tp_head, item->chain_head_change_priv);
530 }
531
532 static void tcf_chain0_head_change(struct tcf_chain *chain,
533 struct tcf_proto *tp_head)
534 {
535 struct tcf_filter_chain_list_item *item;
536 struct tcf_block *block = chain->block;
537
538 if (chain->index)
539 return;
540
541 mutex_lock(&block->lock);
542 list_for_each_entry(item, &block->chain0.filter_chain_list, list)
543 tcf_chain_head_change_item(item, tp_head);
544 mutex_unlock(&block->lock);
545 }
546
547 /* Returns true if block can be safely freed. */
548
549 static bool tcf_chain_detach(struct tcf_chain *chain)
550 {
551 struct tcf_block *block = chain->block;
552
553 ASSERT_BLOCK_LOCKED(block);
554
555 list_del_rcu(&chain->list);
556 if (!chain->index)
557 block->chain0.chain = NULL;
558
559 if (list_empty(&block->chain_list) &&
560 refcount_read(&block->refcnt) == 0)
561 return true;
562
563 return false;
564 }
565
566 static void tcf_block_destroy(struct tcf_block *block)
567 {
568 mutex_destroy(&block->lock);
569 mutex_destroy(&block->proto_destroy_lock);
570 xa_destroy(&block->ports);
571 kfree_rcu(block, rcu);
572 }
573
574 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
575 {
576 struct tcf_block *block = chain->block;
577
578 mutex_destroy(&chain->filter_chain_lock);
579 kfree_rcu(chain, rcu);
580 if (free_block)
581 tcf_block_destroy(block);
582 }
583
584 static void tcf_chain_hold(struct tcf_chain *chain)
585 {
586 ASSERT_BLOCK_LOCKED(chain->block);
587
588 ++chain->refcnt;
589 }
590
591 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
592 {
593 ASSERT_BLOCK_LOCKED(chain->block);
594
595 /* In case all the references are action references, this
596 * chain should not be shown to the user.
597 */
598 return chain->refcnt == chain->action_refcnt;
599 }
600
601 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
602 u32 chain_index)
603 {
604 struct tcf_chain *chain;
605
606 ASSERT_BLOCK_LOCKED(block);
607
608 list_for_each_entry(chain, &block->chain_list, list) {
609 if (chain->index == chain_index)
610 return chain;
611 }
612 return NULL;
613 }
614
615 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
616 static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
617 u32 chain_index)
618 {
619 struct tcf_chain *chain;
620
621 list_for_each_entry_rcu(chain, &block->chain_list, list) {
622 if (chain->index == chain_index)
623 return chain;
624 }
625 return NULL;
626 }
627 #endif
628
629 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
630 u32 seq, u16 flags, int event, bool unicast,
631 struct netlink_ext_ack *extack);
632
633 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
634 u32 chain_index, bool create,
635 bool by_act)
636 {
637 struct tcf_chain *chain = NULL;
638 bool is_first_reference;
639
640 mutex_lock(&block->lock);
641 chain = tcf_chain_lookup(block, chain_index);
642 if (chain) {
643 tcf_chain_hold(chain);
644 } else {
645 if (!create)
646 goto errout;
647 chain = tcf_chain_create(block, chain_index);
648 if (!chain)
649 goto errout;
650 }
651
652 if (by_act)
653 ++chain->action_refcnt;
654 is_first_reference = chain->refcnt - chain->action_refcnt == 1;
655 mutex_unlock(&block->lock);
656
657 /* Send notification only in case we got the first
658 * non-action reference. Until then, the chain acts only as
659 * a placeholder for actions pointing to it and user ought
660 * not know about them.
661 */
662 if (is_first_reference && !by_act)
663 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
664 RTM_NEWCHAIN, false, NULL);
665
666 return chain;
667
668 errout:
669 mutex_unlock(&block->lock);
670 return chain;
671 }
672
673 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
674 bool create)
675 {
676 return __tcf_chain_get(block, chain_index, create, false);
677 }
678
679 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
680 {
681 return __tcf_chain_get(block, chain_index, true, true);
682 }
683 EXPORT_SYMBOL(tcf_chain_get_by_act);
684
685 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
686 void *tmplt_priv);
687 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
688 void *tmplt_priv, u32 chain_index,
689 struct tcf_block *block, struct sk_buff *oskb,
690 u32 seq, u16 flags);
691
692 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
693 bool explicitly_created)
694 {
695 struct tcf_block *block = chain->block;
696 const struct tcf_proto_ops *tmplt_ops;
697 unsigned int refcnt, non_act_refcnt;
698 bool free_block = false;
699 void *tmplt_priv;
700
701 mutex_lock(&block->lock);
702 if (explicitly_created) {
703 if (!chain->explicitly_created) {
704 mutex_unlock(&block->lock);
705 return;
706 }
707 chain->explicitly_created = false;
708 }
709
710 if (by_act)
711 chain->action_refcnt--;
712
713 /* tc_chain_notify_delete can't be called while holding block lock.
714 * However, when block is unlocked chain can be changed concurrently, so
715 * save these to temporary variables.
716 */
717 refcnt = --chain->refcnt;
718 non_act_refcnt = refcnt - chain->action_refcnt;
719 tmplt_ops = chain->tmplt_ops;
720 tmplt_priv = chain->tmplt_priv;
721
722 if (non_act_refcnt == chain->explicitly_created && !by_act) {
723 if (non_act_refcnt == 0)
724 tc_chain_notify_delete(tmplt_ops, tmplt_priv,
725 chain->index, block, NULL, 0, 0);
726 /* Last reference to chain, no need to lock. */
727 chain->flushing = false;
728 }
729
730 if (refcnt == 0)
731 free_block = tcf_chain_detach(chain);
732 mutex_unlock(&block->lock);
733
734 if (refcnt == 0) {
735 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
736 tcf_chain_destroy(chain, free_block);
737 }
738 }
739
740 static void tcf_chain_put(struct tcf_chain *chain)
741 {
742 __tcf_chain_put(chain, false, false);
743 }
744
745 void tcf_chain_put_by_act(struct tcf_chain *chain)
746 {
747 __tcf_chain_put(chain, true, false);
748 }
749 EXPORT_SYMBOL(tcf_chain_put_by_act);
750
751 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
752 {
753 __tcf_chain_put(chain, false, true);
754 }
755
756 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
757 {
758 struct tcf_proto *tp, *tp_next;
759
760 mutex_lock(&chain->filter_chain_lock);
761 tp = tcf_chain_dereference(chain->filter_chain, chain);
762 while (tp) {
763 tp_next = rcu_dereference_protected(tp->next, 1);
764 tcf_proto_signal_destroying(chain, tp);
765 tp = tp_next;
766 }
767 tp = tcf_chain_dereference(chain->filter_chain, chain);
768 RCU_INIT_POINTER(chain->filter_chain, NULL);
769 tcf_chain0_head_change(chain, NULL);
770 chain->flushing = true;
771 mutex_unlock(&chain->filter_chain_lock);
772
773 while (tp) {
774 tp_next = rcu_dereference_protected(tp->next, 1);
775 tcf_proto_put(tp, rtnl_held, NULL);
776 tp = tp_next;
777 }
778 }
779
780 static int tcf_block_setup(struct tcf_block *block,
781 struct flow_block_offload *bo);
782
783 static void tcf_block_offload_init(struct flow_block_offload *bo,
784 struct net_device *dev, struct Qdisc *sch,
785 enum flow_block_command command,
786 enum flow_block_binder_type binder_type,
787 struct flow_block *flow_block,
788 bool shared, struct netlink_ext_ack *extack)
789 {
790 bo->net = dev_net(dev);
791 bo->command = command;
792 bo->binder_type = binder_type;
793 bo->block = flow_block;
794 bo->block_shared = shared;
795 bo->extack = extack;
796 bo->sch = sch;
797 bo->cb_list_head = &flow_block->cb_list;
798 INIT_LIST_HEAD(&bo->cb_list);
799 }
800
801 static void tcf_block_unbind(struct tcf_block *block,
802 struct flow_block_offload *bo);
803
804 static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
805 {
806 struct tcf_block *block = block_cb->indr.data;
807 struct net_device *dev = block_cb->indr.dev;
808 struct Qdisc *sch = block_cb->indr.sch;
809 struct netlink_ext_ack extack = {};
810 struct flow_block_offload bo = {};
811
812 tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
813 block_cb->indr.binder_type,
814 &block->flow_block, tcf_block_shared(block),
815 &extack);
816 rtnl_lock();
817 down_write(&block->cb_lock);
818 list_del(&block_cb->driver_list);
819 list_move(&block_cb->list, &bo.cb_list);
820 tcf_block_unbind(block, &bo);
821 up_write(&block->cb_lock);
822 rtnl_unlock();
823 }
824
825 static bool tcf_block_offload_in_use(struct tcf_block *block)
826 {
827 return atomic_read(&block->offloadcnt);
828 }
829
830 static int tcf_block_offload_cmd(struct tcf_block *block,
831 struct net_device *dev, struct Qdisc *sch,
832 struct tcf_block_ext_info *ei,
833 enum flow_block_command command,
834 struct netlink_ext_ack *extack)
835 {
836 struct flow_block_offload bo = {};
837
838 tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
839 &block->flow_block, tcf_block_shared(block),
840 extack);
841
842 if (dev->netdev_ops->ndo_setup_tc) {
843 int err;
844
845 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
846 if (err < 0) {
847 if (err != -EOPNOTSUPP)
848 NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
849 return err;
850 }
851
852 return tcf_block_setup(block, &bo);
853 }
854
855 flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
856 tc_block_indr_cleanup);
857 tcf_block_setup(block, &bo);
858
859 return -EOPNOTSUPP;
860 }
861
862 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
863 struct tcf_block_ext_info *ei,
864 struct netlink_ext_ack *extack)
865 {
866 struct net_device *dev = q->dev_queue->dev;
867 int err;
868
869 down_write(&block->cb_lock);
870
871 /* If tc offload feature is disabled and the block we try to bind
872 * to already has some offloaded filters, forbid to bind.
873 */
874 if (dev->netdev_ops->ndo_setup_tc &&
875 !tc_can_offload(dev) &&
876 tcf_block_offload_in_use(block)) {
877 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
878 err = -EOPNOTSUPP;
879 goto err_unlock;
880 }
881
882 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
883 if (err == -EOPNOTSUPP)
884 goto no_offload_dev_inc;
885 if (err)
886 goto err_unlock;
887
888 up_write(&block->cb_lock);
889 return 0;
890
891 no_offload_dev_inc:
892 if (tcf_block_offload_in_use(block))
893 goto err_unlock;
894
895 err = 0;
896 block->nooffloaddevcnt++;
897 err_unlock:
898 up_write(&block->cb_lock);
899 return err;
900 }
901
902 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
903 struct tcf_block_ext_info *ei)
904 {
905 struct net_device *dev = q->dev_queue->dev;
906 int err;
907
908 down_write(&block->cb_lock);
909 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
910 if (err == -EOPNOTSUPP)
911 goto no_offload_dev_dec;
912 up_write(&block->cb_lock);
913 return;
914
915 no_offload_dev_dec:
916 WARN_ON(block->nooffloaddevcnt-- == 0);
917 up_write(&block->cb_lock);
918 }
919
920 static int
921 tcf_chain0_head_change_cb_add(struct tcf_block *block,
922 struct tcf_block_ext_info *ei,
923 struct netlink_ext_ack *extack)
924 {
925 struct tcf_filter_chain_list_item *item;
926 struct tcf_chain *chain0;
927
928 item = kmalloc(sizeof(*item), GFP_KERNEL);
929 if (!item) {
930 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
931 return -ENOMEM;
932 }
933 item->chain_head_change = ei->chain_head_change;
934 item->chain_head_change_priv = ei->chain_head_change_priv;
935
936 mutex_lock(&block->lock);
937 chain0 = block->chain0.chain;
938 if (chain0)
939 tcf_chain_hold(chain0);
940 else
941 list_add(&item->list, &block->chain0.filter_chain_list);
942 mutex_unlock(&block->lock);
943
944 if (chain0) {
945 struct tcf_proto *tp_head;
946
947 mutex_lock(&chain0->filter_chain_lock);
948
949 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
950 if (tp_head)
951 tcf_chain_head_change_item(item, tp_head);
952
953 mutex_lock(&block->lock);
954 list_add(&item->list, &block->chain0.filter_chain_list);
955 mutex_unlock(&block->lock);
956
957 mutex_unlock(&chain0->filter_chain_lock);
958 tcf_chain_put(chain0);
959 }
960
961 return 0;
962 }
963
964 static void
965 tcf_chain0_head_change_cb_del(struct tcf_block *block,
966 struct tcf_block_ext_info *ei)
967 {
968 struct tcf_filter_chain_list_item *item;
969
970 mutex_lock(&block->lock);
971 list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
972 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
973 (item->chain_head_change == ei->chain_head_change &&
974 item->chain_head_change_priv == ei->chain_head_change_priv)) {
975 if (block->chain0.chain)
976 tcf_chain_head_change_item(item, NULL);
977 list_del(&item->list);
978 mutex_unlock(&block->lock);
979
980 kfree(item);
981 return;
982 }
983 }
984 mutex_unlock(&block->lock);
985 WARN_ON(1);
986 }
987
988 struct tcf_net {
989 spinlock_t idr_lock; /* Protects idr */
990 struct idr idr;
991 };
992
993 static unsigned int tcf_net_id;
994
995 static int tcf_block_insert(struct tcf_block *block, struct net *net,
996 struct netlink_ext_ack *extack)
997 {
998 struct tcf_net *tn = net_generic(net, tcf_net_id);
999 int err;
1000
1001 idr_preload(GFP_KERNEL);
1002 spin_lock(&tn->idr_lock);
1003 err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
1004 GFP_NOWAIT);
1005 spin_unlock(&tn->idr_lock);
1006 idr_preload_end();
1007
1008 return err;
1009 }
1010
1011 static void tcf_block_remove(struct tcf_block *block, struct net *net)
1012 {
1013 struct tcf_net *tn = net_generic(net, tcf_net_id);
1014
1015 spin_lock(&tn->idr_lock);
1016 idr_remove(&tn->idr, block->index);
1017 spin_unlock(&tn->idr_lock);
1018 }
1019
1020 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
1021 u32 block_index,
1022 struct netlink_ext_ack *extack)
1023 {
1024 struct tcf_block *block;
1025
1026 block = kzalloc(sizeof(*block), GFP_KERNEL);
1027 if (!block) {
1028 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
1029 return ERR_PTR(-ENOMEM);
1030 }
1031 mutex_init(&block->lock);
1032 mutex_init(&block->proto_destroy_lock);
1033 init_rwsem(&block->cb_lock);
1034 flow_block_init(&block->flow_block);
1035 INIT_LIST_HEAD(&block->chain_list);
1036 INIT_LIST_HEAD(&block->owner_list);
1037 INIT_LIST_HEAD(&block->chain0.filter_chain_list);
1038
1039 refcount_set(&block->refcnt, 1);
1040 block->net = net;
1041 block->index = block_index;
1042 xa_init(&block->ports);
1043
1044 /* Don't store q pointer for blocks which are shared */
1045 if (!tcf_block_shared(block))
1046 block->q = q;
1047 return block;
1048 }
1049
1050 struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
1051 {
1052 struct tcf_net *tn = net_generic(net, tcf_net_id);
1053
1054 return idr_find(&tn->idr, block_index);
1055 }
1056 EXPORT_SYMBOL(tcf_block_lookup);
1057
1058 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
1059 {
1060 struct tcf_block *block;
1061
1062 rcu_read_lock();
1063 block = tcf_block_lookup(net, block_index);
1064 if (block && !refcount_inc_not_zero(&block->refcnt))
1065 block = NULL;
1066 rcu_read_unlock();
1067
1068 return block;
1069 }
1070
1071 static struct tcf_chain *
1072 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1073 {
1074 mutex_lock(&block->lock);
1075 if (chain)
1076 chain = list_is_last(&chain->list, &block->chain_list) ?
1077 NULL : list_next_entry(chain, list);
1078 else
1079 chain = list_first_entry_or_null(&block->chain_list,
1080 struct tcf_chain, list);
1081
1082 /* skip all action-only chains */
1083 while (chain && tcf_chain_held_by_acts_only(chain))
1084 chain = list_is_last(&chain->list, &block->chain_list) ?
1085 NULL : list_next_entry(chain, list);
1086
1087 if (chain)
1088 tcf_chain_hold(chain);
1089 mutex_unlock(&block->lock);
1090
1091 return chain;
1092 }
1093
1094 /* Function to be used by all clients that want to iterate over all chains on
1095 * block. It properly obtains block->lock and takes reference to chain before
1096 * returning it. Users of this function must be tolerant to concurrent chain
1097 * insertion/deletion or ensure that no concurrent chain modification is
1098 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1099 * consistent dump because rtnl lock is released each time skb is filled with
1100 * data and sent to user-space.
1101 */
1102
1103 struct tcf_chain *
1104 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1105 {
1106 struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
1107
1108 if (chain)
1109 tcf_chain_put(chain);
1110
1111 return chain_next;
1112 }
1113 EXPORT_SYMBOL(tcf_get_next_chain);
1114
1115 static struct tcf_proto *
1116 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1117 {
1118 u32 prio = 0;
1119
1120 ASSERT_RTNL();
1121 mutex_lock(&chain->filter_chain_lock);
1122
1123 if (!tp) {
1124 tp = tcf_chain_dereference(chain->filter_chain, chain);
1125 } else if (tcf_proto_is_deleting(tp)) {
1126 /* 'deleting' flag is set and chain->filter_chain_lock was
1127 * unlocked, which means next pointer could be invalid. Restart
1128 * search.
1129 */
1130 prio = tp->prio + 1;
1131 tp = tcf_chain_dereference(chain->filter_chain, chain);
1132
1133 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1134 if (!tp->deleting && tp->prio >= prio)
1135 break;
1136 } else {
1137 tp = tcf_chain_dereference(tp->next, chain);
1138 }
1139
1140 if (tp)
1141 tcf_proto_get(tp);
1142
1143 mutex_unlock(&chain->filter_chain_lock);
1144
1145 return tp;
1146 }
1147
1148 /* Function to be used by all clients that want to iterate over all tp's on
1149 * chain. Users of this function must be tolerant to concurrent tp
1150 * insertion/deletion or ensure that no concurrent chain modification is
1151 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1152 * consistent dump because rtnl lock is released each time skb is filled with
1153 * data and sent to user-space.
1154 */
1155
1156 struct tcf_proto *
1157 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1158 {
1159 struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1160
1161 if (tp)
1162 tcf_proto_put(tp, true, NULL);
1163
1164 return tp_next;
1165 }
1166 EXPORT_SYMBOL(tcf_get_next_proto);
1167
1168 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1169 {
1170 struct tcf_chain *chain;
1171
1172 /* Last reference to block. At this point chains cannot be added or
1173 * removed concurrently.
1174 */
1175 for (chain = tcf_get_next_chain(block, NULL);
1176 chain;
1177 chain = tcf_get_next_chain(block, chain)) {
1178 tcf_chain_put_explicitly_created(chain);
1179 tcf_chain_flush(chain, rtnl_held);
1180 }
1181 }
1182
1183 /* Lookup Qdisc and increments its reference counter.
1184 * Set parent, if necessary.
1185 */
1186
1187 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1188 u32 *parent, int ifindex, bool rtnl_held,
1189 struct netlink_ext_ack *extack)
1190 {
1191 const struct Qdisc_class_ops *cops;
1192 struct net_device *dev;
1193 int err = 0;
1194
1195 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1196 return 0;
1197
1198 rcu_read_lock();
1199
1200 /* Find link */
1201 dev = dev_get_by_index_rcu(net, ifindex);
1202 if (!dev) {
1203 rcu_read_unlock();
1204 return -ENODEV;
1205 }
1206
1207 /* Find qdisc */
1208 if (!*parent) {
1209 *q = rcu_dereference(dev->qdisc);
1210 *parent = (*q)->handle;
1211 } else {
1212 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1213 if (!*q) {
1214 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1215 err = -EINVAL;
1216 goto errout_rcu;
1217 }
1218 }
1219
1220 *q = qdisc_refcount_inc_nz(*q);
1221 if (!*q) {
1222 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1223 err = -EINVAL;
1224 goto errout_rcu;
1225 }
1226
1227 /* Is it classful? */
1228 cops = (*q)->ops->cl_ops;
1229 if (!cops) {
1230 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1231 err = -EINVAL;
1232 goto errout_qdisc;
1233 }
1234
1235 if (!cops->tcf_block) {
1236 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1237 err = -EOPNOTSUPP;
1238 goto errout_qdisc;
1239 }
1240
1241 errout_rcu:
1242 /* At this point we know that qdisc is not noop_qdisc,
1243 * which means that qdisc holds a reference to net_device
1244 * and we hold a reference to qdisc, so it is safe to release
1245 * rcu read lock.
1246 */
1247 rcu_read_unlock();
1248 return err;
1249
1250 errout_qdisc:
1251 rcu_read_unlock();
1252
1253 if (rtnl_held)
1254 qdisc_put(*q);
1255 else
1256 qdisc_put_unlocked(*q);
1257 *q = NULL;
1258
1259 return err;
1260 }
1261
1262 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1263 int ifindex, struct netlink_ext_ack *extack)
1264 {
1265 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1266 return 0;
1267
1268 /* Do we search for filter, attached to class? */
1269 if (TC_H_MIN(parent)) {
1270 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1271
1272 *cl = cops->find(q, parent);
1273 if (*cl == 0) {
1274 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1275 return -ENOENT;
1276 }
1277 }
1278
1279 return 0;
1280 }
1281
1282 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1283 unsigned long cl, int ifindex,
1284 u32 block_index,
1285 struct netlink_ext_ack *extack)
1286 {
1287 struct tcf_block *block;
1288
1289 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1290 block = tcf_block_refcnt_get(net, block_index);
1291 if (!block) {
1292 NL_SET_ERR_MSG(extack, "Block of given index was not found");
1293 return ERR_PTR(-EINVAL);
1294 }
1295 } else {
1296 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1297
1298 block = cops->tcf_block(q, cl, extack);
1299 if (!block)
1300 return ERR_PTR(-EINVAL);
1301
1302 if (tcf_block_shared(block)) {
1303 NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1304 return ERR_PTR(-EOPNOTSUPP);
1305 }
1306
1307 /* Always take reference to block in order to support execution
1308 * of rules update path of cls API without rtnl lock. Caller
1309 * must release block when it is finished using it. 'if' block
1310 * of this conditional obtain reference to block by calling
1311 * tcf_block_refcnt_get().
1312 */
1313 refcount_inc(&block->refcnt);
1314 }
1315
1316 return block;
1317 }
1318
1319 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1320 struct tcf_block_ext_info *ei, bool rtnl_held)
1321 {
1322 if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1323 /* Flushing/putting all chains will cause the block to be
1324 * deallocated when last chain is freed. However, if chain_list
1325 * is empty, block has to be manually deallocated. After block
1326 * reference counter reached 0, it is no longer possible to
1327 * increment it or add new chains to block.
1328 */
1329 bool free_block = list_empty(&block->chain_list);
1330
1331 mutex_unlock(&block->lock);
1332 if (tcf_block_shared(block))
1333 tcf_block_remove(block, block->net);
1334
1335 if (q)
1336 tcf_block_offload_unbind(block, q, ei);
1337
1338 if (free_block)
1339 tcf_block_destroy(block);
1340 else
1341 tcf_block_flush_all_chains(block, rtnl_held);
1342 } else if (q) {
1343 tcf_block_offload_unbind(block, q, ei);
1344 }
1345 }
1346
1347 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1348 {
1349 __tcf_block_put(block, NULL, NULL, rtnl_held);
1350 }
1351
1352 /* Find tcf block.
1353 * Set q, parent, cl when appropriate.
1354 */
1355
1356 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1357 u32 *parent, unsigned long *cl,
1358 int ifindex, u32 block_index,
1359 struct netlink_ext_ack *extack)
1360 {
1361 struct tcf_block *block;
1362 int err = 0;
1363
1364 ASSERT_RTNL();
1365
1366 err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1367 if (err)
1368 goto errout;
1369
1370 err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1371 if (err)
1372 goto errout_qdisc;
1373
1374 block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1375 if (IS_ERR(block)) {
1376 err = PTR_ERR(block);
1377 goto errout_qdisc;
1378 }
1379
1380 return block;
1381
1382 errout_qdisc:
1383 if (*q)
1384 qdisc_put(*q);
1385 errout:
1386 *q = NULL;
1387 return ERR_PTR(err);
1388 }
1389
1390 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1391 bool rtnl_held)
1392 {
1393 if (!IS_ERR_OR_NULL(block))
1394 tcf_block_refcnt_put(block, rtnl_held);
1395
1396 if (q) {
1397 if (rtnl_held)
1398 qdisc_put(q);
1399 else
1400 qdisc_put_unlocked(q);
1401 }
1402 }
1403
1404 struct tcf_block_owner_item {
1405 struct list_head list;
1406 struct Qdisc *q;
1407 enum flow_block_binder_type binder_type;
1408 };
1409
1410 static void
1411 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1412 struct Qdisc *q,
1413 enum flow_block_binder_type binder_type)
1414 {
1415 if (block->keep_dst &&
1416 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1417 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1418 netif_keep_dst(qdisc_dev(q));
1419 }
1420
1421 void tcf_block_netif_keep_dst(struct tcf_block *block)
1422 {
1423 struct tcf_block_owner_item *item;
1424
1425 block->keep_dst = true;
1426 list_for_each_entry(item, &block->owner_list, list)
1427 tcf_block_owner_netif_keep_dst(block, item->q,
1428 item->binder_type);
1429 }
1430 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1431
1432 static int tcf_block_owner_add(struct tcf_block *block,
1433 struct Qdisc *q,
1434 enum flow_block_binder_type binder_type)
1435 {
1436 struct tcf_block_owner_item *item;
1437
1438 item = kmalloc(sizeof(*item), GFP_KERNEL);
1439 if (!item)
1440 return -ENOMEM;
1441 item->q = q;
1442 item->binder_type = binder_type;
1443 list_add(&item->list, &block->owner_list);
1444 return 0;
1445 }
1446
1447 static void tcf_block_owner_del(struct tcf_block *block,
1448 struct Qdisc *q,
1449 enum flow_block_binder_type binder_type)
1450 {
1451 struct tcf_block_owner_item *item;
1452
1453 list_for_each_entry(item, &block->owner_list, list) {
1454 if (item->q == q && item->binder_type == binder_type) {
1455 list_del(&item->list);
1456 kfree(item);
1457 return;
1458 }
1459 }
1460 WARN_ON(1);
1461 }
1462
1463 static bool tcf_block_tracks_dev(struct tcf_block *block,
1464 struct tcf_block_ext_info *ei)
1465 {
1466 return tcf_block_shared(block) &&
1467 (ei->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS ||
1468 ei->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS);
1469 }
1470
1471 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1472 struct tcf_block_ext_info *ei,
1473 struct netlink_ext_ack *extack)
1474 {
1475 struct net_device *dev = qdisc_dev(q);
1476 struct net *net = qdisc_net(q);
1477 struct tcf_block *block = NULL;
1478 int err;
1479
1480 if (ei->block_index)
1481 /* block_index not 0 means the shared block is requested */
1482 block = tcf_block_refcnt_get(net, ei->block_index);
1483
1484 if (!block) {
1485 block = tcf_block_create(net, q, ei->block_index, extack);
1486 if (IS_ERR(block))
1487 return PTR_ERR(block);
1488 if (tcf_block_shared(block)) {
1489 err = tcf_block_insert(block, net, extack);
1490 if (err)
1491 goto err_block_insert;
1492 }
1493 }
1494
1495 err = tcf_block_owner_add(block, q, ei->binder_type);
1496 if (err)
1497 goto err_block_owner_add;
1498
1499 tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1500
1501 err = tcf_chain0_head_change_cb_add(block, ei, extack);
1502 if (err)
1503 goto err_chain0_head_change_cb_add;
1504
1505 err = tcf_block_offload_bind(block, q, ei, extack);
1506 if (err)
1507 goto err_block_offload_bind;
1508
1509 if (tcf_block_tracks_dev(block, ei)) {
1510 err = xa_insert(&block->ports, dev->ifindex, dev, GFP_KERNEL);
1511 if (err) {
1512 NL_SET_ERR_MSG(extack, "block dev insert failed");
1513 goto err_dev_insert;
1514 }
1515 }
1516
1517 *p_block = block;
1518 return 0;
1519
1520 err_dev_insert:
1521 err_block_offload_bind:
1522 tcf_chain0_head_change_cb_del(block, ei);
1523 err_chain0_head_change_cb_add:
1524 tcf_block_owner_del(block, q, ei->binder_type);
1525 err_block_owner_add:
1526 err_block_insert:
1527 tcf_block_refcnt_put(block, true);
1528 return err;
1529 }
1530 EXPORT_SYMBOL(tcf_block_get_ext);
1531
1532 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1533 {
1534 struct tcf_proto __rcu **p_filter_chain = priv;
1535
1536 rcu_assign_pointer(*p_filter_chain, tp_head);
1537 }
1538
1539 int tcf_block_get(struct tcf_block **p_block,
1540 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1541 struct netlink_ext_ack *extack)
1542 {
1543 struct tcf_block_ext_info ei = {
1544 .chain_head_change = tcf_chain_head_change_dflt,
1545 .chain_head_change_priv = p_filter_chain,
1546 };
1547
1548 WARN_ON(!p_filter_chain);
1549 return tcf_block_get_ext(p_block, q, &ei, extack);
1550 }
1551 EXPORT_SYMBOL(tcf_block_get);
1552
1553 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1554 * actions should be all removed after flushing.
1555 */
1556 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1557 struct tcf_block_ext_info *ei)
1558 {
1559 struct net_device *dev = qdisc_dev(q);
1560
1561 if (!block)
1562 return;
1563 if (tcf_block_tracks_dev(block, ei))
1564 xa_erase(&block->ports, dev->ifindex);
1565 tcf_chain0_head_change_cb_del(block, ei);
1566 tcf_block_owner_del(block, q, ei->binder_type);
1567
1568 __tcf_block_put(block, q, ei, true);
1569 }
1570 EXPORT_SYMBOL(tcf_block_put_ext);
1571
1572 void tcf_block_put(struct tcf_block *block)
1573 {
1574 struct tcf_block_ext_info ei = {0, };
1575
1576 if (!block)
1577 return;
1578 tcf_block_put_ext(block, block->q, &ei);
1579 }
1580
1581 EXPORT_SYMBOL(tcf_block_put);
1582
1583 static int
1584 tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1585 void *cb_priv, bool add, bool offload_in_use,
1586 struct netlink_ext_ack *extack)
1587 {
1588 struct tcf_chain *chain, *chain_prev;
1589 struct tcf_proto *tp, *tp_prev;
1590 int err;
1591
1592 lockdep_assert_held(&block->cb_lock);
1593
1594 for (chain = __tcf_get_next_chain(block, NULL);
1595 chain;
1596 chain_prev = chain,
1597 chain = __tcf_get_next_chain(block, chain),
1598 tcf_chain_put(chain_prev)) {
1599 if (chain->tmplt_ops && add)
1600 chain->tmplt_ops->tmplt_reoffload(chain, true, cb,
1601 cb_priv);
1602 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1603 tp_prev = tp,
1604 tp = __tcf_get_next_proto(chain, tp),
1605 tcf_proto_put(tp_prev, true, NULL)) {
1606 if (tp->ops->reoffload) {
1607 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1608 extack);
1609 if (err && add)
1610 goto err_playback_remove;
1611 } else if (add && offload_in_use) {
1612 err = -EOPNOTSUPP;
1613 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1614 goto err_playback_remove;
1615 }
1616 }
1617 if (chain->tmplt_ops && !add)
1618 chain->tmplt_ops->tmplt_reoffload(chain, false, cb,
1619 cb_priv);
1620 }
1621
1622 return 0;
1623
1624 err_playback_remove:
1625 tcf_proto_put(tp, true, NULL);
1626 tcf_chain_put(chain);
1627 tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1628 extack);
1629 return err;
1630 }
1631
1632 static int tcf_block_bind(struct tcf_block *block,
1633 struct flow_block_offload *bo)
1634 {
1635 struct flow_block_cb *block_cb, *next;
1636 int err, i = 0;
1637
1638 lockdep_assert_held(&block->cb_lock);
1639
1640 list_for_each_entry(block_cb, &bo->cb_list, list) {
1641 err = tcf_block_playback_offloads(block, block_cb->cb,
1642 block_cb->cb_priv, true,
1643 tcf_block_offload_in_use(block),
1644 bo->extack);
1645 if (err)
1646 goto err_unroll;
1647 if (!bo->unlocked_driver_cb)
1648 block->lockeddevcnt++;
1649
1650 i++;
1651 }
1652 list_splice(&bo->cb_list, &block->flow_block.cb_list);
1653
1654 return 0;
1655
1656 err_unroll:
1657 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1658 list_del(&block_cb->driver_list);
1659 if (i-- > 0) {
1660 list_del(&block_cb->list);
1661 tcf_block_playback_offloads(block, block_cb->cb,
1662 block_cb->cb_priv, false,
1663 tcf_block_offload_in_use(block),
1664 NULL);
1665 if (!bo->unlocked_driver_cb)
1666 block->lockeddevcnt--;
1667 }
1668 flow_block_cb_free(block_cb);
1669 }
1670
1671 return err;
1672 }
1673
1674 static void tcf_block_unbind(struct tcf_block *block,
1675 struct flow_block_offload *bo)
1676 {
1677 struct flow_block_cb *block_cb, *next;
1678
1679 lockdep_assert_held(&block->cb_lock);
1680
1681 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1682 tcf_block_playback_offloads(block, block_cb->cb,
1683 block_cb->cb_priv, false,
1684 tcf_block_offload_in_use(block),
1685 NULL);
1686 list_del(&block_cb->list);
1687 flow_block_cb_free(block_cb);
1688 if (!bo->unlocked_driver_cb)
1689 block->lockeddevcnt--;
1690 }
1691 }
1692
1693 static int tcf_block_setup(struct tcf_block *block,
1694 struct flow_block_offload *bo)
1695 {
1696 int err;
1697
1698 switch (bo->command) {
1699 case FLOW_BLOCK_BIND:
1700 err = tcf_block_bind(block, bo);
1701 break;
1702 case FLOW_BLOCK_UNBIND:
1703 err = 0;
1704 tcf_block_unbind(block, bo);
1705 break;
1706 default:
1707 WARN_ON_ONCE(1);
1708 err = -EOPNOTSUPP;
1709 }
1710
1711 return err;
1712 }
1713
1714 /* Main classifier routine: scans classifier chain attached
1715 * to this qdisc, (optionally) tests for protocol and asks
1716 * specific classifiers.
1717 */
1718 static inline int __tcf_classify(struct sk_buff *skb,
1719 const struct tcf_proto *tp,
1720 const struct tcf_proto *orig_tp,
1721 struct tcf_result *res,
1722 bool compat_mode,
1723 struct tcf_exts_miss_cookie_node *n,
1724 int act_index,
1725 u32 *last_executed_chain)
1726 {
1727 #ifdef CONFIG_NET_CLS_ACT
1728 const int max_reclassify_loop = 16;
1729 const struct tcf_proto *first_tp;
1730 int limit = 0;
1731
1732 reclassify:
1733 #endif
1734 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1735 __be16 protocol = skb_protocol(skb, false);
1736 int err = 0;
1737
1738 if (n) {
1739 struct tcf_exts *exts;
1740
1741 if (n->tp_prio != tp->prio)
1742 continue;
1743
1744 /* We re-lookup the tp and chain based on index instead
1745 * of having hard refs and locks to them, so do a sanity
1746 * check if any of tp,chain,exts was replaced by the
1747 * time we got here with a cookie from hardware.
1748 */
1749 if (unlikely(n->tp != tp || n->tp->chain != n->chain ||
1750 !tp->ops->get_exts)) {
1751 tcf_set_drop_reason(skb,
1752 SKB_DROP_REASON_TC_COOKIE_ERROR);
1753 return TC_ACT_SHOT;
1754 }
1755
1756 exts = tp->ops->get_exts(tp, n->handle);
1757 if (unlikely(!exts || n->exts != exts)) {
1758 tcf_set_drop_reason(skb,
1759 SKB_DROP_REASON_TC_COOKIE_ERROR);
1760 return TC_ACT_SHOT;
1761 }
1762
1763 n = NULL;
1764 err = tcf_exts_exec_ex(skb, exts, act_index, res);
1765 } else {
1766 if (tp->protocol != protocol &&
1767 tp->protocol != htons(ETH_P_ALL))
1768 continue;
1769
1770 err = tc_classify(skb, tp, res);
1771 }
1772 #ifdef CONFIG_NET_CLS_ACT
1773 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1774 first_tp = orig_tp;
1775 *last_executed_chain = first_tp->chain->index;
1776 goto reset;
1777 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1778 first_tp = res->goto_tp;
1779 *last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1780 goto reset;
1781 }
1782 #endif
1783 if (err >= 0)
1784 return err;
1785 }
1786
1787 if (unlikely(n)) {
1788 tcf_set_drop_reason(skb,
1789 SKB_DROP_REASON_TC_COOKIE_ERROR);
1790 return TC_ACT_SHOT;
1791 }
1792
1793 return TC_ACT_UNSPEC; /* signal: continue lookup */
1794 #ifdef CONFIG_NET_CLS_ACT
1795 reset:
1796 if (unlikely(limit++ >= max_reclassify_loop)) {
1797 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1798 tp->chain->block->index,
1799 tp->prio & 0xffff,
1800 ntohs(tp->protocol));
1801 tcf_set_drop_reason(skb,
1802 SKB_DROP_REASON_TC_RECLASSIFY_LOOP);
1803 return TC_ACT_SHOT;
1804 }
1805
1806 tp = first_tp;
1807 goto reclassify;
1808 #endif
1809 }
1810
1811 int tcf_classify(struct sk_buff *skb,
1812 const struct tcf_block *block,
1813 const struct tcf_proto *tp,
1814 struct tcf_result *res, bool compat_mode)
1815 {
1816 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1817 u32 last_executed_chain = 0;
1818
1819 return __tcf_classify(skb, tp, tp, res, compat_mode, NULL, 0,
1820 &last_executed_chain);
1821 #else
1822 u32 last_executed_chain = tp ? tp->chain->index : 0;
1823 struct tcf_exts_miss_cookie_node *n = NULL;
1824 const struct tcf_proto *orig_tp = tp;
1825 struct tc_skb_ext *ext;
1826 int act_index = 0;
1827 int ret;
1828
1829 if (block) {
1830 ext = skb_ext_find(skb, TC_SKB_EXT);
1831
1832 if (ext && (ext->chain || ext->act_miss)) {
1833 struct tcf_chain *fchain;
1834 u32 chain;
1835
1836 if (ext->act_miss) {
1837 n = tcf_exts_miss_cookie_lookup(ext->act_miss_cookie,
1838 &act_index);
1839 if (!n) {
1840 tcf_set_drop_reason(skb,
1841 SKB_DROP_REASON_TC_COOKIE_ERROR);
1842 return TC_ACT_SHOT;
1843 }
1844
1845 chain = n->chain_index;
1846 } else {
1847 chain = ext->chain;
1848 }
1849
1850 fchain = tcf_chain_lookup_rcu(block, chain);
1851 if (!fchain) {
1852 tcf_set_drop_reason(skb,
1853 SKB_DROP_REASON_TC_CHAIN_NOTFOUND);
1854
1855 return TC_ACT_SHOT;
1856 }
1857
1858 /* Consume, so cloned/redirect skbs won't inherit ext */
1859 skb_ext_del(skb, TC_SKB_EXT);
1860
1861 tp = rcu_dereference_bh(fchain->filter_chain);
1862 last_executed_chain = fchain->index;
1863 }
1864 }
1865
1866 ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode, n, act_index,
1867 &last_executed_chain);
1868
1869 if (tc_skb_ext_tc_enabled()) {
1870 /* If we missed on some chain */
1871 if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1872 struct tc_skb_cb *cb = tc_skb_cb(skb);
1873
1874 ext = tc_skb_ext_alloc(skb);
1875 if (WARN_ON_ONCE(!ext)) {
1876 tcf_set_drop_reason(skb, SKB_DROP_REASON_NOMEM);
1877 return TC_ACT_SHOT;
1878 }
1879 ext->chain = last_executed_chain;
1880 ext->mru = cb->mru;
1881 ext->post_ct = cb->post_ct;
1882 ext->post_ct_snat = cb->post_ct_snat;
1883 ext->post_ct_dnat = cb->post_ct_dnat;
1884 ext->zone = cb->zone;
1885 }
1886 }
1887
1888 return ret;
1889 #endif
1890 }
1891 EXPORT_SYMBOL(tcf_classify);
1892
1893 struct tcf_chain_info {
1894 struct tcf_proto __rcu **pprev;
1895 struct tcf_proto __rcu *next;
1896 };
1897
1898 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1899 struct tcf_chain_info *chain_info)
1900 {
1901 return tcf_chain_dereference(*chain_info->pprev, chain);
1902 }
1903
1904 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1905 struct tcf_chain_info *chain_info,
1906 struct tcf_proto *tp)
1907 {
1908 if (chain->flushing)
1909 return -EAGAIN;
1910
1911 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1912 if (*chain_info->pprev == chain->filter_chain)
1913 tcf_chain0_head_change(chain, tp);
1914 tcf_proto_get(tp);
1915 rcu_assign_pointer(*chain_info->pprev, tp);
1916
1917 return 0;
1918 }
1919
1920 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1921 struct tcf_chain_info *chain_info,
1922 struct tcf_proto *tp)
1923 {
1924 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1925
1926 tcf_proto_mark_delete(tp);
1927 if (tp == chain->filter_chain)
1928 tcf_chain0_head_change(chain, next);
1929 RCU_INIT_POINTER(*chain_info->pprev, next);
1930 }
1931
1932 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1933 struct tcf_chain_info *chain_info,
1934 u32 protocol, u32 prio,
1935 bool prio_allocate);
1936
1937 /* Try to insert new proto.
1938 * If proto with specified priority already exists, free new proto
1939 * and return existing one.
1940 */
1941
1942 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1943 struct tcf_proto *tp_new,
1944 u32 protocol, u32 prio,
1945 bool rtnl_held)
1946 {
1947 struct tcf_chain_info chain_info;
1948 struct tcf_proto *tp;
1949 int err = 0;
1950
1951 mutex_lock(&chain->filter_chain_lock);
1952
1953 if (tcf_proto_exists_destroying(chain, tp_new)) {
1954 mutex_unlock(&chain->filter_chain_lock);
1955 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1956 return ERR_PTR(-EAGAIN);
1957 }
1958
1959 tp = tcf_chain_tp_find(chain, &chain_info,
1960 protocol, prio, false);
1961 if (!tp)
1962 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1963 mutex_unlock(&chain->filter_chain_lock);
1964
1965 if (tp) {
1966 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1967 tp_new = tp;
1968 } else if (err) {
1969 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1970 tp_new = ERR_PTR(err);
1971 }
1972
1973 return tp_new;
1974 }
1975
1976 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1977 struct tcf_proto *tp, bool rtnl_held,
1978 struct netlink_ext_ack *extack)
1979 {
1980 struct tcf_chain_info chain_info;
1981 struct tcf_proto *tp_iter;
1982 struct tcf_proto **pprev;
1983 struct tcf_proto *next;
1984
1985 mutex_lock(&chain->filter_chain_lock);
1986
1987 /* Atomically find and remove tp from chain. */
1988 for (pprev = &chain->filter_chain;
1989 (tp_iter = tcf_chain_dereference(*pprev, chain));
1990 pprev = &tp_iter->next) {
1991 if (tp_iter == tp) {
1992 chain_info.pprev = pprev;
1993 chain_info.next = tp_iter->next;
1994 WARN_ON(tp_iter->deleting);
1995 break;
1996 }
1997 }
1998 /* Verify that tp still exists and no new filters were inserted
1999 * concurrently.
2000 * Mark tp for deletion if it is empty.
2001 */
2002 if (!tp_iter || !tcf_proto_check_delete(tp)) {
2003 mutex_unlock(&chain->filter_chain_lock);
2004 return;
2005 }
2006
2007 tcf_proto_signal_destroying(chain, tp);
2008 next = tcf_chain_dereference(chain_info.next, chain);
2009 if (tp == chain->filter_chain)
2010 tcf_chain0_head_change(chain, next);
2011 RCU_INIT_POINTER(*chain_info.pprev, next);
2012 mutex_unlock(&chain->filter_chain_lock);
2013
2014 tcf_proto_put(tp, rtnl_held, extack);
2015 }
2016
2017 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
2018 struct tcf_chain_info *chain_info,
2019 u32 protocol, u32 prio,
2020 bool prio_allocate)
2021 {
2022 struct tcf_proto **pprev;
2023 struct tcf_proto *tp;
2024
2025 /* Check the chain for existence of proto-tcf with this priority */
2026 for (pprev = &chain->filter_chain;
2027 (tp = tcf_chain_dereference(*pprev, chain));
2028 pprev = &tp->next) {
2029 if (tp->prio >= prio) {
2030 if (tp->prio == prio) {
2031 if (prio_allocate ||
2032 (tp->protocol != protocol && protocol))
2033 return ERR_PTR(-EINVAL);
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
2050 static 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
2104 out_nlmsg_trim:
2105 nla_put_failure:
2106 cls_op_not_supp:
2107 nlmsg_trim(skb, b);
2108 return -1;
2109 }
2110
2111 static 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
2143 static 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
2182 static 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
2196 static void tfilter_put(struct tcf_proto *tp, void *fh)
2197 {
2198 if (tp->ops->put && fh)
2199 tp->ops->put(tp, fh);
2200 }
2201
2202 static bool is_qdisc_ingress(__u32 classid)
2203 {
2204 return (TC_H_MIN(classid) == TC_H_MIN(TC_H_MIN_INGRESS));
2205 }
2206
2207 static 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
2231 replay:
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 = tca[TCA_CHAIN] ? nla_get_u32(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);
2315 if (IS_ERR(tp)) {
2316 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2317 err = PTR_ERR(tp);
2318 goto errout_locked;
2319 }
2320
2321 if (tp == NULL) {
2322 struct tcf_proto *tp_new = NULL;
2323
2324 if (chain->flushing) {
2325 err = -EAGAIN;
2326 goto errout_locked;
2327 }
2328
2329 /* Proto-tcf does not exist, create new one */
2330
2331 if (tca[TCA_KIND] == NULL || !protocol) {
2332 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2333 err = -EINVAL;
2334 goto errout_locked;
2335 }
2336
2337 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2338 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2339 err = -ENOENT;
2340 goto errout_locked;
2341 }
2342
2343 if (prio_allocate)
2344 prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2345 &chain_info));
2346
2347 mutex_unlock(&chain->filter_chain_lock);
2348 tp_new = tcf_proto_create(name, protocol, prio, chain,
2349 rtnl_held, extack);
2350 if (IS_ERR(tp_new)) {
2351 err = PTR_ERR(tp_new);
2352 goto errout_tp;
2353 }
2354
2355 tp_created = 1;
2356 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2357 rtnl_held);
2358 if (IS_ERR(tp)) {
2359 err = PTR_ERR(tp);
2360 goto errout_tp;
2361 }
2362 } else {
2363 mutex_unlock(&chain->filter_chain_lock);
2364 }
2365
2366 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2367 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2368 err = -EINVAL;
2369 goto errout;
2370 }
2371
2372 fh = tp->ops->get(tp, t->tcm_handle);
2373
2374 if (!fh) {
2375 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2376 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2377 err = -ENOENT;
2378 goto errout;
2379 }
2380 } else if (n->nlmsg_flags & NLM_F_EXCL) {
2381 tfilter_put(tp, fh);
2382 NL_SET_ERR_MSG(extack, "Filter already exists");
2383 err = -EEXIST;
2384 goto errout;
2385 }
2386
2387 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2388 tfilter_put(tp, fh);
2389 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2390 err = -EINVAL;
2391 goto errout;
2392 }
2393
2394 if (!(n->nlmsg_flags & NLM_F_CREATE))
2395 flags |= TCA_ACT_FLAGS_REPLACE;
2396 if (!rtnl_held)
2397 flags |= TCA_ACT_FLAGS_NO_RTNL;
2398 if (is_qdisc_ingress(parent))
2399 flags |= TCA_ACT_FLAGS_AT_INGRESS;
2400 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2401 flags, extack);
2402 if (err == 0) {
2403 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2404 RTM_NEWTFILTER, false, rtnl_held, extack);
2405 tfilter_put(tp, fh);
2406 tcf_block_filter_cnt_update(block, &tp->counted, true);
2407 /* q pointer is NULL for shared blocks */
2408 if (q)
2409 q->flags &= ~TCQ_F_CAN_BYPASS;
2410 }
2411
2412 errout:
2413 if (err && tp_created)
2414 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2415 errout_tp:
2416 if (chain) {
2417 if (tp && !IS_ERR(tp))
2418 tcf_proto_put(tp, rtnl_held, NULL);
2419 if (!tp_created)
2420 tcf_chain_put(chain);
2421 }
2422 tcf_block_release(q, block, rtnl_held);
2423
2424 if (rtnl_held)
2425 rtnl_unlock();
2426
2427 if (err == -EAGAIN) {
2428 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2429 * of target chain.
2430 */
2431 rtnl_held = true;
2432 /* Replay the request. */
2433 goto replay;
2434 }
2435 return err;
2436
2437 errout_locked:
2438 mutex_unlock(&chain->filter_chain_lock);
2439 goto errout;
2440 }
2441
2442 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2443 struct netlink_ext_ack *extack)
2444 {
2445 struct net *net = sock_net(skb->sk);
2446 struct nlattr *tca[TCA_MAX + 1];
2447 char name[IFNAMSIZ];
2448 struct tcmsg *t;
2449 u32 protocol;
2450 u32 prio;
2451 u32 parent;
2452 u32 chain_index;
2453 struct Qdisc *q = NULL;
2454 struct tcf_chain_info chain_info;
2455 struct tcf_chain *chain = NULL;
2456 struct tcf_block *block = NULL;
2457 struct tcf_proto *tp = NULL;
2458 unsigned long cl = 0;
2459 void *fh = NULL;
2460 int err;
2461 bool rtnl_held = false;
2462
2463 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2464 rtm_tca_policy, extack);
2465 if (err < 0)
2466 return err;
2467
2468 t = nlmsg_data(n);
2469 protocol = TC_H_MIN(t->tcm_info);
2470 prio = TC_H_MAJ(t->tcm_info);
2471 parent = t->tcm_parent;
2472
2473 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2474 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2475 return -ENOENT;
2476 }
2477
2478 /* Find head of filter chain. */
2479
2480 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2481 if (err)
2482 return err;
2483
2484 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2485 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2486 err = -EINVAL;
2487 goto errout;
2488 }
2489 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2490 * found), qdisc is not unlocked, classifier type is not specified,
2491 * classifier is not unlocked.
2492 */
2493 if (!prio ||
2494 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2495 !tcf_proto_is_unlocked(name)) {
2496 rtnl_held = true;
2497 rtnl_lock();
2498 }
2499
2500 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2501 if (err)
2502 goto errout;
2503
2504 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2505 extack);
2506 if (IS_ERR(block)) {
2507 err = PTR_ERR(block);
2508 goto errout;
2509 }
2510
2511 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2512 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2513 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2514 err = -EINVAL;
2515 goto errout;
2516 }
2517 chain = tcf_chain_get(block, chain_index, false);
2518 if (!chain) {
2519 /* User requested flush on non-existent chain. Nothing to do,
2520 * so just return success.
2521 */
2522 if (prio == 0) {
2523 err = 0;
2524 goto errout;
2525 }
2526 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2527 err = -ENOENT;
2528 goto errout;
2529 }
2530
2531 if (prio == 0) {
2532 tfilter_notify_chain(net, skb, block, q, parent, n,
2533 chain, RTM_DELTFILTER, extack);
2534 tcf_chain_flush(chain, rtnl_held);
2535 err = 0;
2536 goto errout;
2537 }
2538
2539 mutex_lock(&chain->filter_chain_lock);
2540 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2541 prio, false);
2542 if (!tp || IS_ERR(tp)) {
2543 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2544 err = tp ? PTR_ERR(tp) : -ENOENT;
2545 goto errout_locked;
2546 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2547 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2548 err = -EINVAL;
2549 goto errout_locked;
2550 } else if (t->tcm_handle == 0) {
2551 tcf_proto_signal_destroying(chain, tp);
2552 tcf_chain_tp_remove(chain, &chain_info, tp);
2553 mutex_unlock(&chain->filter_chain_lock);
2554
2555 tcf_proto_put(tp, rtnl_held, NULL);
2556 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2557 RTM_DELTFILTER, false, rtnl_held, extack);
2558 err = 0;
2559 goto errout;
2560 }
2561 mutex_unlock(&chain->filter_chain_lock);
2562
2563 fh = tp->ops->get(tp, t->tcm_handle);
2564
2565 if (!fh) {
2566 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2567 err = -ENOENT;
2568 } else {
2569 bool last;
2570
2571 err = tfilter_del_notify(net, skb, n, tp, block, q, parent, fh,
2572 &last, rtnl_held, extack);
2573
2574 if (err)
2575 goto errout;
2576 if (last)
2577 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2578 }
2579
2580 errout:
2581 if (chain) {
2582 if (tp && !IS_ERR(tp))
2583 tcf_proto_put(tp, rtnl_held, NULL);
2584 tcf_chain_put(chain);
2585 }
2586 tcf_block_release(q, block, rtnl_held);
2587
2588 if (rtnl_held)
2589 rtnl_unlock();
2590
2591 return err;
2592
2593 errout_locked:
2594 mutex_unlock(&chain->filter_chain_lock);
2595 goto errout;
2596 }
2597
2598 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2599 struct netlink_ext_ack *extack)
2600 {
2601 struct net *net = sock_net(skb->sk);
2602 struct nlattr *tca[TCA_MAX + 1];
2603 char name[IFNAMSIZ];
2604 struct tcmsg *t;
2605 u32 protocol;
2606 u32 prio;
2607 u32 parent;
2608 u32 chain_index;
2609 struct Qdisc *q = NULL;
2610 struct tcf_chain_info chain_info;
2611 struct tcf_chain *chain = NULL;
2612 struct tcf_block *block = NULL;
2613 struct tcf_proto *tp = NULL;
2614 unsigned long cl = 0;
2615 void *fh = NULL;
2616 int err;
2617 bool rtnl_held = false;
2618
2619 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2620 rtm_tca_policy, extack);
2621 if (err < 0)
2622 return err;
2623
2624 t = nlmsg_data(n);
2625 protocol = TC_H_MIN(t->tcm_info);
2626 prio = TC_H_MAJ(t->tcm_info);
2627 parent = t->tcm_parent;
2628
2629 if (prio == 0) {
2630 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2631 return -ENOENT;
2632 }
2633
2634 /* Find head of filter chain. */
2635
2636 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2637 if (err)
2638 return err;
2639
2640 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2641 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2642 err = -EINVAL;
2643 goto errout;
2644 }
2645 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2646 * unlocked, classifier type is not specified, classifier is not
2647 * unlocked.
2648 */
2649 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2650 !tcf_proto_is_unlocked(name)) {
2651 rtnl_held = true;
2652 rtnl_lock();
2653 }
2654
2655 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2656 if (err)
2657 goto errout;
2658
2659 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2660 extack);
2661 if (IS_ERR(block)) {
2662 err = PTR_ERR(block);
2663 goto errout;
2664 }
2665
2666 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2667 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2668 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2669 err = -EINVAL;
2670 goto errout;
2671 }
2672 chain = tcf_chain_get(block, chain_index, false);
2673 if (!chain) {
2674 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2675 err = -EINVAL;
2676 goto errout;
2677 }
2678
2679 mutex_lock(&chain->filter_chain_lock);
2680 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2681 prio, false);
2682 mutex_unlock(&chain->filter_chain_lock);
2683 if (!tp || IS_ERR(tp)) {
2684 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2685 err = tp ? PTR_ERR(tp) : -ENOENT;
2686 goto errout;
2687 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2688 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2689 err = -EINVAL;
2690 goto errout;
2691 }
2692
2693 fh = tp->ops->get(tp, t->tcm_handle);
2694
2695 if (!fh) {
2696 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2697 err = -ENOENT;
2698 } else {
2699 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2700 fh, RTM_NEWTFILTER, true, rtnl_held, NULL);
2701 if (err < 0)
2702 NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2703 }
2704
2705 tfilter_put(tp, fh);
2706 errout:
2707 if (chain) {
2708 if (tp && !IS_ERR(tp))
2709 tcf_proto_put(tp, rtnl_held, NULL);
2710 tcf_chain_put(chain);
2711 }
2712 tcf_block_release(q, block, rtnl_held);
2713
2714 if (rtnl_held)
2715 rtnl_unlock();
2716
2717 return err;
2718 }
2719
2720 struct tcf_dump_args {
2721 struct tcf_walker w;
2722 struct sk_buff *skb;
2723 struct netlink_callback *cb;
2724 struct tcf_block *block;
2725 struct Qdisc *q;
2726 u32 parent;
2727 bool terse_dump;
2728 };
2729
2730 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2731 {
2732 struct tcf_dump_args *a = (void *)arg;
2733 struct net *net = sock_net(a->skb->sk);
2734
2735 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2736 n, NETLINK_CB(a->cb->skb).portid,
2737 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2738 RTM_NEWTFILTER, a->terse_dump, true, NULL);
2739 }
2740
2741 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2742 struct sk_buff *skb, struct netlink_callback *cb,
2743 long index_start, long *p_index, bool terse)
2744 {
2745 struct net *net = sock_net(skb->sk);
2746 struct tcf_block *block = chain->block;
2747 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2748 struct tcf_proto *tp, *tp_prev;
2749 struct tcf_dump_args arg;
2750
2751 for (tp = __tcf_get_next_proto(chain, NULL);
2752 tp;
2753 tp_prev = tp,
2754 tp = __tcf_get_next_proto(chain, tp),
2755 tcf_proto_put(tp_prev, true, NULL),
2756 (*p_index)++) {
2757 if (*p_index < index_start)
2758 continue;
2759 if (TC_H_MAJ(tcm->tcm_info) &&
2760 TC_H_MAJ(tcm->tcm_info) != tp->prio)
2761 continue;
2762 if (TC_H_MIN(tcm->tcm_info) &&
2763 TC_H_MIN(tcm->tcm_info) != tp->protocol)
2764 continue;
2765 if (*p_index > index_start)
2766 memset(&cb->args[1], 0,
2767 sizeof(cb->args) - sizeof(cb->args[0]));
2768 if (cb->args[1] == 0) {
2769 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2770 NETLINK_CB(cb->skb).portid,
2771 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2772 RTM_NEWTFILTER, false, true, NULL) <= 0)
2773 goto errout;
2774 cb->args[1] = 1;
2775 }
2776 if (!tp->ops->walk)
2777 continue;
2778 arg.w.fn = tcf_node_dump;
2779 arg.skb = skb;
2780 arg.cb = cb;
2781 arg.block = block;
2782 arg.q = q;
2783 arg.parent = parent;
2784 arg.w.stop = 0;
2785 arg.w.skip = cb->args[1] - 1;
2786 arg.w.count = 0;
2787 arg.w.cookie = cb->args[2];
2788 arg.terse_dump = terse;
2789 tp->ops->walk(tp, &arg.w, true);
2790 cb->args[2] = arg.w.cookie;
2791 cb->args[1] = arg.w.count + 1;
2792 if (arg.w.stop)
2793 goto errout;
2794 }
2795 return true;
2796
2797 errout:
2798 tcf_proto_put(tp, true, NULL);
2799 return false;
2800 }
2801
2802 static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2803 [TCA_CHAIN] = { .type = NLA_U32 },
2804 [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2805 };
2806
2807 /* called with RTNL */
2808 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2809 {
2810 struct tcf_chain *chain, *chain_prev;
2811 struct net *net = sock_net(skb->sk);
2812 struct nlattr *tca[TCA_MAX + 1];
2813 struct Qdisc *q = NULL;
2814 struct tcf_block *block;
2815 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2816 bool terse_dump = false;
2817 long index_start;
2818 long index;
2819 u32 parent;
2820 int err;
2821
2822 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2823 return skb->len;
2824
2825 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2826 tcf_tfilter_dump_policy, cb->extack);
2827 if (err)
2828 return err;
2829
2830 if (tca[TCA_DUMP_FLAGS]) {
2831 struct nla_bitfield32 flags =
2832 nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2833
2834 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2835 }
2836
2837 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2838 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2839 if (!block)
2840 goto out;
2841 /* If we work with block index, q is NULL and parent value
2842 * will never be used in the following code. The check
2843 * in tcf_fill_node prevents it. However, compiler does not
2844 * see that far, so set parent to zero to silence the warning
2845 * about parent being uninitialized.
2846 */
2847 parent = 0;
2848 } else {
2849 const struct Qdisc_class_ops *cops;
2850 struct net_device *dev;
2851 unsigned long cl = 0;
2852
2853 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2854 if (!dev)
2855 return skb->len;
2856
2857 parent = tcm->tcm_parent;
2858 if (!parent)
2859 q = rtnl_dereference(dev->qdisc);
2860 else
2861 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2862 if (!q)
2863 goto out;
2864 cops = q->ops->cl_ops;
2865 if (!cops)
2866 goto out;
2867 if (!cops->tcf_block)
2868 goto out;
2869 if (TC_H_MIN(tcm->tcm_parent)) {
2870 cl = cops->find(q, tcm->tcm_parent);
2871 if (cl == 0)
2872 goto out;
2873 }
2874 block = cops->tcf_block(q, cl, NULL);
2875 if (!block)
2876 goto out;
2877 parent = block->classid;
2878 if (tcf_block_shared(block))
2879 q = NULL;
2880 }
2881
2882 index_start = cb->args[0];
2883 index = 0;
2884
2885 for (chain = __tcf_get_next_chain(block, NULL);
2886 chain;
2887 chain_prev = chain,
2888 chain = __tcf_get_next_chain(block, chain),
2889 tcf_chain_put(chain_prev)) {
2890 if (tca[TCA_CHAIN] &&
2891 nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2892 continue;
2893 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2894 index_start, &index, terse_dump)) {
2895 tcf_chain_put(chain);
2896 err = -EMSGSIZE;
2897 break;
2898 }
2899 }
2900
2901 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2902 tcf_block_refcnt_put(block, true);
2903 cb->args[0] = index;
2904
2905 out:
2906 /* If we did no progress, the error (EMSGSIZE) is real */
2907 if (skb->len == 0 && err)
2908 return err;
2909 return skb->len;
2910 }
2911
2912 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2913 void *tmplt_priv, u32 chain_index,
2914 struct net *net, struct sk_buff *skb,
2915 struct tcf_block *block,
2916 u32 portid, u32 seq, u16 flags, int event,
2917 struct netlink_ext_ack *extack)
2918 {
2919 unsigned char *b = skb_tail_pointer(skb);
2920 const struct tcf_proto_ops *ops;
2921 struct nlmsghdr *nlh;
2922 struct tcmsg *tcm;
2923 void *priv;
2924
2925 ops = tmplt_ops;
2926 priv = tmplt_priv;
2927
2928 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2929 if (!nlh)
2930 goto out_nlmsg_trim;
2931 tcm = nlmsg_data(nlh);
2932 tcm->tcm_family = AF_UNSPEC;
2933 tcm->tcm__pad1 = 0;
2934 tcm->tcm__pad2 = 0;
2935 tcm->tcm_handle = 0;
2936 if (block->q) {
2937 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2938 tcm->tcm_parent = block->q->handle;
2939 } else {
2940 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2941 tcm->tcm_block_index = block->index;
2942 }
2943
2944 if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2945 goto nla_put_failure;
2946
2947 if (ops) {
2948 if (nla_put_string(skb, TCA_KIND, ops->kind))
2949 goto nla_put_failure;
2950 if (ops->tmplt_dump(skb, net, priv) < 0)
2951 goto nla_put_failure;
2952 }
2953
2954 if (extack && extack->_msg &&
2955 nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2956 goto out_nlmsg_trim;
2957
2958 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2959
2960 return skb->len;
2961
2962 out_nlmsg_trim:
2963 nla_put_failure:
2964 nlmsg_trim(skb, b);
2965 return -EMSGSIZE;
2966 }
2967
2968 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2969 u32 seq, u16 flags, int event, bool unicast,
2970 struct netlink_ext_ack *extack)
2971 {
2972 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2973 struct tcf_block *block = chain->block;
2974 struct net *net = block->net;
2975 struct sk_buff *skb;
2976 int err = 0;
2977
2978 if (!unicast && !rtnl_notify_needed(net, flags, RTNLGRP_TC))
2979 return 0;
2980
2981 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2982 if (!skb)
2983 return -ENOBUFS;
2984
2985 if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2986 chain->index, net, skb, block, portid,
2987 seq, flags, event, extack) <= 0) {
2988 kfree_skb(skb);
2989 return -EINVAL;
2990 }
2991
2992 if (unicast)
2993 err = rtnl_unicast(skb, net, portid);
2994 else
2995 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2996 flags & NLM_F_ECHO);
2997
2998 return err;
2999 }
3000
3001 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
3002 void *tmplt_priv, u32 chain_index,
3003 struct tcf_block *block, struct sk_buff *oskb,
3004 u32 seq, u16 flags)
3005 {
3006 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
3007 struct net *net = block->net;
3008 struct sk_buff *skb;
3009
3010 if (!rtnl_notify_needed(net, flags, RTNLGRP_TC))
3011 return 0;
3012
3013 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3014 if (!skb)
3015 return -ENOBUFS;
3016
3017 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
3018 block, portid, seq, flags, RTM_DELCHAIN, NULL) <= 0) {
3019 kfree_skb(skb);
3020 return -EINVAL;
3021 }
3022
3023 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
3024 }
3025
3026 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
3027 struct nlattr **tca,
3028 struct netlink_ext_ack *extack)
3029 {
3030 const struct tcf_proto_ops *ops;
3031 char name[IFNAMSIZ];
3032 void *tmplt_priv;
3033
3034 /* If kind is not set, user did not specify template. */
3035 if (!tca[TCA_KIND])
3036 return 0;
3037
3038 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
3039 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
3040 return -EINVAL;
3041 }
3042
3043 ops = tcf_proto_lookup_ops(name, true, extack);
3044 if (IS_ERR(ops))
3045 return PTR_ERR(ops);
3046 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump ||
3047 !ops->tmplt_reoffload) {
3048 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
3049 module_put(ops->owner);
3050 return -EOPNOTSUPP;
3051 }
3052
3053 tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
3054 if (IS_ERR(tmplt_priv)) {
3055 module_put(ops->owner);
3056 return PTR_ERR(tmplt_priv);
3057 }
3058 chain->tmplt_ops = ops;
3059 chain->tmplt_priv = tmplt_priv;
3060 return 0;
3061 }
3062
3063 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
3064 void *tmplt_priv)
3065 {
3066 /* If template ops are set, no work to do for us. */
3067 if (!tmplt_ops)
3068 return;
3069
3070 tmplt_ops->tmplt_destroy(tmplt_priv);
3071 module_put(tmplt_ops->owner);
3072 }
3073
3074 /* Add/delete/get a chain */
3075
3076 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
3077 struct netlink_ext_ack *extack)
3078 {
3079 struct net *net = sock_net(skb->sk);
3080 struct nlattr *tca[TCA_MAX + 1];
3081 struct tcmsg *t;
3082 u32 parent;
3083 u32 chain_index;
3084 struct Qdisc *q;
3085 struct tcf_chain *chain;
3086 struct tcf_block *block;
3087 unsigned long cl;
3088 int err;
3089
3090 replay:
3091 q = NULL;
3092 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
3093 rtm_tca_policy, extack);
3094 if (err < 0)
3095 return err;
3096
3097 t = nlmsg_data(n);
3098 parent = t->tcm_parent;
3099 cl = 0;
3100
3101 block = tcf_block_find(net, &q, &parent, &cl,
3102 t->tcm_ifindex, t->tcm_block_index, extack);
3103 if (IS_ERR(block))
3104 return PTR_ERR(block);
3105
3106 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
3107 if (chain_index > TC_ACT_EXT_VAL_MASK) {
3108 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
3109 err = -EINVAL;
3110 goto errout_block;
3111 }
3112
3113 mutex_lock(&block->lock);
3114 chain = tcf_chain_lookup(block, chain_index);
3115 if (n->nlmsg_type == RTM_NEWCHAIN) {
3116 if (chain) {
3117 if (tcf_chain_held_by_acts_only(chain)) {
3118 /* The chain exists only because there is
3119 * some action referencing it.
3120 */
3121 tcf_chain_hold(chain);
3122 } else {
3123 NL_SET_ERR_MSG(extack, "Filter chain already exists");
3124 err = -EEXIST;
3125 goto errout_block_locked;
3126 }
3127 } else {
3128 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
3129 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
3130 err = -ENOENT;
3131 goto errout_block_locked;
3132 }
3133 chain = tcf_chain_create(block, chain_index);
3134 if (!chain) {
3135 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
3136 err = -ENOMEM;
3137 goto errout_block_locked;
3138 }
3139 }
3140 } else {
3141 if (!chain || tcf_chain_held_by_acts_only(chain)) {
3142 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
3143 err = -EINVAL;
3144 goto errout_block_locked;
3145 }
3146 tcf_chain_hold(chain);
3147 }
3148
3149 if (n->nlmsg_type == RTM_NEWCHAIN) {
3150 /* Modifying chain requires holding parent block lock. In case
3151 * the chain was successfully added, take a reference to the
3152 * chain. This ensures that an empty chain does not disappear at
3153 * the end of this function.
3154 */
3155 tcf_chain_hold(chain);
3156 chain->explicitly_created = true;
3157 }
3158 mutex_unlock(&block->lock);
3159
3160 switch (n->nlmsg_type) {
3161 case RTM_NEWCHAIN:
3162 err = tc_chain_tmplt_add(chain, net, tca, extack);
3163 if (err) {
3164 tcf_chain_put_explicitly_created(chain);
3165 goto errout;
3166 }
3167
3168 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
3169 RTM_NEWCHAIN, false, extack);
3170 break;
3171 case RTM_DELCHAIN:
3172 tfilter_notify_chain(net, skb, block, q, parent, n,
3173 chain, RTM_DELTFILTER, extack);
3174 /* Flush the chain first as the user requested chain removal. */
3175 tcf_chain_flush(chain, true);
3176 /* In case the chain was successfully deleted, put a reference
3177 * to the chain previously taken during addition.
3178 */
3179 tcf_chain_put_explicitly_created(chain);
3180 break;
3181 case RTM_GETCHAIN:
3182 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
3183 n->nlmsg_flags, n->nlmsg_type, true, extack);
3184 if (err < 0)
3185 NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
3186 break;
3187 default:
3188 err = -EOPNOTSUPP;
3189 NL_SET_ERR_MSG(extack, "Unsupported message type");
3190 goto errout;
3191 }
3192
3193 errout:
3194 tcf_chain_put(chain);
3195 errout_block:
3196 tcf_block_release(q, block, true);
3197 if (err == -EAGAIN)
3198 /* Replay the request. */
3199 goto replay;
3200 return err;
3201
3202 errout_block_locked:
3203 mutex_unlock(&block->lock);
3204 goto errout_block;
3205 }
3206
3207 /* called with RTNL */
3208 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
3209 {
3210 struct net *net = sock_net(skb->sk);
3211 struct nlattr *tca[TCA_MAX + 1];
3212 struct Qdisc *q = NULL;
3213 struct tcf_block *block;
3214 struct tcmsg *tcm = nlmsg_data(cb->nlh);
3215 struct tcf_chain *chain;
3216 long index_start;
3217 long index;
3218 int err;
3219
3220 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
3221 return skb->len;
3222
3223 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
3224 rtm_tca_policy, cb->extack);
3225 if (err)
3226 return err;
3227
3228 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
3229 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
3230 if (!block)
3231 goto out;
3232 } else {
3233 const struct Qdisc_class_ops *cops;
3234 struct net_device *dev;
3235 unsigned long cl = 0;
3236
3237 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
3238 if (!dev)
3239 return skb->len;
3240
3241 if (!tcm->tcm_parent)
3242 q = rtnl_dereference(dev->qdisc);
3243 else
3244 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
3245
3246 if (!q)
3247 goto out;
3248 cops = q->ops->cl_ops;
3249 if (!cops)
3250 goto out;
3251 if (!cops->tcf_block)
3252 goto out;
3253 if (TC_H_MIN(tcm->tcm_parent)) {
3254 cl = cops->find(q, tcm->tcm_parent);
3255 if (cl == 0)
3256 goto out;
3257 }
3258 block = cops->tcf_block(q, cl, NULL);
3259 if (!block)
3260 goto out;
3261 if (tcf_block_shared(block))
3262 q = NULL;
3263 }
3264
3265 index_start = cb->args[0];
3266 index = 0;
3267
3268 mutex_lock(&block->lock);
3269 list_for_each_entry(chain, &block->chain_list, list) {
3270 if ((tca[TCA_CHAIN] &&
3271 nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3272 continue;
3273 if (index < index_start) {
3274 index++;
3275 continue;
3276 }
3277 if (tcf_chain_held_by_acts_only(chain))
3278 continue;
3279 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3280 chain->index, net, skb, block,
3281 NETLINK_CB(cb->skb).portid,
3282 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3283 RTM_NEWCHAIN, NULL);
3284 if (err <= 0)
3285 break;
3286 index++;
3287 }
3288 mutex_unlock(&block->lock);
3289
3290 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3291 tcf_block_refcnt_put(block, true);
3292 cb->args[0] = index;
3293
3294 out:
3295 /* If we did no progress, the error (EMSGSIZE) is real */
3296 if (skb->len == 0 && err)
3297 return err;
3298 return skb->len;
3299 }
3300
3301 int tcf_exts_init_ex(struct tcf_exts *exts, struct net *net, int action,
3302 int police, struct tcf_proto *tp, u32 handle,
3303 bool use_action_miss)
3304 {
3305 int err = 0;
3306
3307 #ifdef CONFIG_NET_CLS_ACT
3308 exts->type = 0;
3309 exts->nr_actions = 0;
3310 exts->miss_cookie_node = NULL;
3311 /* Note: we do not own yet a reference on net.
3312 * This reference might be taken later from tcf_exts_get_net().
3313 */
3314 exts->net = net;
3315 exts->actions = kcalloc(TCA_ACT_MAX_PRIO, sizeof(struct tc_action *),
3316 GFP_KERNEL);
3317 if (!exts->actions)
3318 return -ENOMEM;
3319 #endif
3320
3321 exts->action = action;
3322 exts->police = police;
3323
3324 if (!use_action_miss)
3325 return 0;
3326
3327 err = tcf_exts_miss_cookie_base_alloc(exts, tp, handle);
3328 if (err)
3329 goto err_miss_alloc;
3330
3331 return 0;
3332
3333 err_miss_alloc:
3334 tcf_exts_destroy(exts);
3335 #ifdef CONFIG_NET_CLS_ACT
3336 exts->actions = NULL;
3337 #endif
3338 return err;
3339 }
3340 EXPORT_SYMBOL(tcf_exts_init_ex);
3341
3342 void tcf_exts_destroy(struct tcf_exts *exts)
3343 {
3344 tcf_exts_miss_cookie_base_destroy(exts);
3345
3346 #ifdef CONFIG_NET_CLS_ACT
3347 if (exts->actions) {
3348 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3349 kfree(exts->actions);
3350 }
3351 exts->nr_actions = 0;
3352 #endif
3353 }
3354 EXPORT_SYMBOL(tcf_exts_destroy);
3355
3356 int tcf_exts_validate_ex(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3357 struct nlattr *rate_tlv, struct tcf_exts *exts,
3358 u32 flags, u32 fl_flags, struct netlink_ext_ack *extack)
3359 {
3360 #ifdef CONFIG_NET_CLS_ACT
3361 {
3362 int init_res[TCA_ACT_MAX_PRIO] = {};
3363 struct tc_action *act;
3364 size_t attr_size = 0;
3365
3366 if (exts->police && tb[exts->police]) {
3367 struct tc_action_ops *a_o;
3368
3369 flags |= TCA_ACT_FLAGS_POLICE | TCA_ACT_FLAGS_BIND;
3370 a_o = tc_action_load_ops(tb[exts->police], flags,
3371 extack);
3372 if (IS_ERR(a_o))
3373 return PTR_ERR(a_o);
3374 act = tcf_action_init_1(net, tp, tb[exts->police],
3375 rate_tlv, a_o, init_res, flags,
3376 extack);
3377 module_put(a_o->owner);
3378 if (IS_ERR(act))
3379 return PTR_ERR(act);
3380
3381 act->type = exts->type = TCA_OLD_COMPAT;
3382 exts->actions[0] = act;
3383 exts->nr_actions = 1;
3384 tcf_idr_insert_many(exts->actions, init_res);
3385 } else if (exts->action && tb[exts->action]) {
3386 int err;
3387
3388 flags |= TCA_ACT_FLAGS_BIND;
3389 err = tcf_action_init(net, tp, tb[exts->action],
3390 rate_tlv, exts->actions, init_res,
3391 &attr_size, flags, fl_flags,
3392 extack);
3393 if (err < 0)
3394 return err;
3395 exts->nr_actions = err;
3396 }
3397 }
3398 #else
3399 if ((exts->action && tb[exts->action]) ||
3400 (exts->police && tb[exts->police])) {
3401 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3402 return -EOPNOTSUPP;
3403 }
3404 #endif
3405
3406 return 0;
3407 }
3408 EXPORT_SYMBOL(tcf_exts_validate_ex);
3409
3410 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3411 struct nlattr *rate_tlv, struct tcf_exts *exts,
3412 u32 flags, struct netlink_ext_ack *extack)
3413 {
3414 return tcf_exts_validate_ex(net, tp, tb, rate_tlv, exts,
3415 flags, 0, extack);
3416 }
3417 EXPORT_SYMBOL(tcf_exts_validate);
3418
3419 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3420 {
3421 #ifdef CONFIG_NET_CLS_ACT
3422 struct tcf_exts old = *dst;
3423
3424 *dst = *src;
3425 tcf_exts_destroy(&old);
3426 #endif
3427 }
3428 EXPORT_SYMBOL(tcf_exts_change);
3429
3430 #ifdef CONFIG_NET_CLS_ACT
3431 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3432 {
3433 if (exts->nr_actions == 0)
3434 return NULL;
3435 else
3436 return exts->actions[0];
3437 }
3438 #endif
3439
3440 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3441 {
3442 #ifdef CONFIG_NET_CLS_ACT
3443 struct nlattr *nest;
3444
3445 if (exts->action && tcf_exts_has_actions(exts)) {
3446 /*
3447 * again for backward compatible mode - we want
3448 * to work with both old and new modes of entering
3449 * tc data even if iproute2 was newer - jhs
3450 */
3451 if (exts->type != TCA_OLD_COMPAT) {
3452 nest = nla_nest_start_noflag(skb, exts->action);
3453 if (nest == NULL)
3454 goto nla_put_failure;
3455
3456 if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3457 < 0)
3458 goto nla_put_failure;
3459 nla_nest_end(skb, nest);
3460 } else if (exts->police) {
3461 struct tc_action *act = tcf_exts_first_act(exts);
3462 nest = nla_nest_start_noflag(skb, exts->police);
3463 if (nest == NULL || !act)
3464 goto nla_put_failure;
3465 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3466 goto nla_put_failure;
3467 nla_nest_end(skb, nest);
3468 }
3469 }
3470 return 0;
3471
3472 nla_put_failure:
3473 nla_nest_cancel(skb, nest);
3474 return -1;
3475 #else
3476 return 0;
3477 #endif
3478 }
3479 EXPORT_SYMBOL(tcf_exts_dump);
3480
3481 int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3482 {
3483 #ifdef CONFIG_NET_CLS_ACT
3484 struct nlattr *nest;
3485
3486 if (!exts->action || !tcf_exts_has_actions(exts))
3487 return 0;
3488
3489 nest = nla_nest_start_noflag(skb, exts->action);
3490 if (!nest)
3491 goto nla_put_failure;
3492
3493 if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3494 goto nla_put_failure;
3495 nla_nest_end(skb, nest);
3496 return 0;
3497
3498 nla_put_failure:
3499 nla_nest_cancel(skb, nest);
3500 return -1;
3501 #else
3502 return 0;
3503 #endif
3504 }
3505 EXPORT_SYMBOL(tcf_exts_terse_dump);
3506
3507 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3508 {
3509 #ifdef CONFIG_NET_CLS_ACT
3510 struct tc_action *a = tcf_exts_first_act(exts);
3511 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3512 return -1;
3513 #endif
3514 return 0;
3515 }
3516 EXPORT_SYMBOL(tcf_exts_dump_stats);
3517
3518 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3519 {
3520 if (*flags & TCA_CLS_FLAGS_IN_HW)
3521 return;
3522 *flags |= TCA_CLS_FLAGS_IN_HW;
3523 if (tc_skip_sw(*flags))
3524 atomic_inc(&block->skipswcnt);
3525 atomic_inc(&block->offloadcnt);
3526 }
3527
3528 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3529 {
3530 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3531 return;
3532 *flags &= ~TCA_CLS_FLAGS_IN_HW;
3533 if (tc_skip_sw(*flags))
3534 atomic_dec(&block->skipswcnt);
3535 atomic_dec(&block->offloadcnt);
3536 }
3537
3538 static void tc_cls_offload_cnt_update(struct tcf_block *block,
3539 struct tcf_proto *tp, u32 *cnt,
3540 u32 *flags, u32 diff, bool add)
3541 {
3542 lockdep_assert_held(&block->cb_lock);
3543
3544 spin_lock(&tp->lock);
3545 if (add) {
3546 if (!*cnt)
3547 tcf_block_offload_inc(block, flags);
3548 *cnt += diff;
3549 } else {
3550 *cnt -= diff;
3551 if (!*cnt)
3552 tcf_block_offload_dec(block, flags);
3553 }
3554 spin_unlock(&tp->lock);
3555 }
3556
3557 static void
3558 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3559 u32 *cnt, u32 *flags)
3560 {
3561 lockdep_assert_held(&block->cb_lock);
3562
3563 spin_lock(&tp->lock);
3564 tcf_block_offload_dec(block, flags);
3565 *cnt = 0;
3566 spin_unlock(&tp->lock);
3567 }
3568
3569 static int
3570 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3571 void *type_data, bool err_stop)
3572 {
3573 struct flow_block_cb *block_cb;
3574 int ok_count = 0;
3575 int err;
3576
3577 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3578 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3579 if (err) {
3580 if (err_stop)
3581 return err;
3582 } else {
3583 ok_count++;
3584 }
3585 }
3586 return ok_count;
3587 }
3588
3589 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3590 void *type_data, bool err_stop, bool rtnl_held)
3591 {
3592 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3593 int ok_count;
3594
3595 retry:
3596 if (take_rtnl)
3597 rtnl_lock();
3598 down_read(&block->cb_lock);
3599 /* Need to obtain rtnl lock if block is bound to devs that require it.
3600 * In block bind code cb_lock is obtained while holding rtnl, so we must
3601 * obtain the locks in same order here.
3602 */
3603 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3604 up_read(&block->cb_lock);
3605 take_rtnl = true;
3606 goto retry;
3607 }
3608
3609 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3610
3611 up_read(&block->cb_lock);
3612 if (take_rtnl)
3613 rtnl_unlock();
3614 return ok_count;
3615 }
3616 EXPORT_SYMBOL(tc_setup_cb_call);
3617
3618 /* Non-destructive filter add. If filter that wasn't already in hardware is
3619 * successfully offloaded, increment block offloads counter. On failure,
3620 * previously offloaded filter is considered to be intact and offloads counter
3621 * is not decremented.
3622 */
3623
3624 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3625 enum tc_setup_type type, void *type_data, bool err_stop,
3626 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3627 {
3628 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3629 int ok_count;
3630
3631 retry:
3632 if (take_rtnl)
3633 rtnl_lock();
3634 down_read(&block->cb_lock);
3635 /* Need to obtain rtnl lock if block is bound to devs that require it.
3636 * In block bind code cb_lock is obtained while holding rtnl, so we must
3637 * obtain the locks in same order here.
3638 */
3639 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3640 up_read(&block->cb_lock);
3641 take_rtnl = true;
3642 goto retry;
3643 }
3644
3645 /* Make sure all netdevs sharing this block are offload-capable. */
3646 if (block->nooffloaddevcnt && err_stop) {
3647 ok_count = -EOPNOTSUPP;
3648 goto err_unlock;
3649 }
3650
3651 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3652 if (ok_count < 0)
3653 goto err_unlock;
3654
3655 if (tp->ops->hw_add)
3656 tp->ops->hw_add(tp, type_data);
3657 if (ok_count > 0)
3658 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3659 ok_count, true);
3660 err_unlock:
3661 up_read(&block->cb_lock);
3662 if (take_rtnl)
3663 rtnl_unlock();
3664 return min(ok_count, 0);
3665 }
3666 EXPORT_SYMBOL(tc_setup_cb_add);
3667
3668 /* Destructive filter replace. If filter that wasn't already in hardware is
3669 * successfully offloaded, increment block offload counter. On failure,
3670 * previously offloaded filter is considered to be destroyed and offload counter
3671 * is decremented.
3672 */
3673
3674 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3675 enum tc_setup_type type, void *type_data, bool err_stop,
3676 u32 *old_flags, unsigned int *old_in_hw_count,
3677 u32 *new_flags, unsigned int *new_in_hw_count,
3678 bool rtnl_held)
3679 {
3680 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3681 int ok_count;
3682
3683 retry:
3684 if (take_rtnl)
3685 rtnl_lock();
3686 down_read(&block->cb_lock);
3687 /* Need to obtain rtnl lock if block is bound to devs that require it.
3688 * In block bind code cb_lock is obtained while holding rtnl, so we must
3689 * obtain the locks in same order here.
3690 */
3691 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3692 up_read(&block->cb_lock);
3693 take_rtnl = true;
3694 goto retry;
3695 }
3696
3697 /* Make sure all netdevs sharing this block are offload-capable. */
3698 if (block->nooffloaddevcnt && err_stop) {
3699 ok_count = -EOPNOTSUPP;
3700 goto err_unlock;
3701 }
3702
3703 tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3704 if (tp->ops->hw_del)
3705 tp->ops->hw_del(tp, type_data);
3706
3707 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3708 if (ok_count < 0)
3709 goto err_unlock;
3710
3711 if (tp->ops->hw_add)
3712 tp->ops->hw_add(tp, type_data);
3713 if (ok_count > 0)
3714 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3715 new_flags, ok_count, true);
3716 err_unlock:
3717 up_read(&block->cb_lock);
3718 if (take_rtnl)
3719 rtnl_unlock();
3720 return min(ok_count, 0);
3721 }
3722 EXPORT_SYMBOL(tc_setup_cb_replace);
3723
3724 /* Destroy filter and decrement block offload counter, if filter was previously
3725 * offloaded.
3726 */
3727
3728 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3729 enum tc_setup_type type, void *type_data, bool err_stop,
3730 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3731 {
3732 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3733 int ok_count;
3734
3735 retry:
3736 if (take_rtnl)
3737 rtnl_lock();
3738 down_read(&block->cb_lock);
3739 /* Need to obtain rtnl lock if block is bound to devs that require it.
3740 * In block bind code cb_lock is obtained while holding rtnl, so we must
3741 * obtain the locks in same order here.
3742 */
3743 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3744 up_read(&block->cb_lock);
3745 take_rtnl = true;
3746 goto retry;
3747 }
3748
3749 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3750
3751 tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3752 if (tp->ops->hw_del)
3753 tp->ops->hw_del(tp, type_data);
3754
3755 up_read(&block->cb_lock);
3756 if (take_rtnl)
3757 rtnl_unlock();
3758 return min(ok_count, 0);
3759 }
3760 EXPORT_SYMBOL(tc_setup_cb_destroy);
3761
3762 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3763 bool add, flow_setup_cb_t *cb,
3764 enum tc_setup_type type, void *type_data,
3765 void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3766 {
3767 int err = cb(type, type_data, cb_priv);
3768
3769 if (err) {
3770 if (add && tc_skip_sw(*flags))
3771 return err;
3772 } else {
3773 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3774 add);
3775 }
3776
3777 return 0;
3778 }
3779 EXPORT_SYMBOL(tc_setup_cb_reoffload);
3780
3781 static int tcf_act_get_user_cookie(struct flow_action_entry *entry,
3782 const struct tc_action *act)
3783 {
3784 struct tc_cookie *user_cookie;
3785 int err = 0;
3786
3787 rcu_read_lock();
3788 user_cookie = rcu_dereference(act->user_cookie);
3789 if (user_cookie) {
3790 entry->user_cookie = flow_action_cookie_create(user_cookie->data,
3791 user_cookie->len,
3792 GFP_ATOMIC);
3793 if (!entry->user_cookie)
3794 err = -ENOMEM;
3795 }
3796 rcu_read_unlock();
3797 return err;
3798 }
3799
3800 static void tcf_act_put_user_cookie(struct flow_action_entry *entry)
3801 {
3802 flow_action_cookie_destroy(entry->user_cookie);
3803 }
3804
3805 void tc_cleanup_offload_action(struct flow_action *flow_action)
3806 {
3807 struct flow_action_entry *entry;
3808 int i;
3809
3810 flow_action_for_each(i, entry, flow_action) {
3811 tcf_act_put_user_cookie(entry);
3812 if (entry->destructor)
3813 entry->destructor(entry->destructor_priv);
3814 }
3815 }
3816 EXPORT_SYMBOL(tc_cleanup_offload_action);
3817
3818 static int tc_setup_offload_act(struct tc_action *act,
3819 struct flow_action_entry *entry,
3820 u32 *index_inc,
3821 struct netlink_ext_ack *extack)
3822 {
3823 #ifdef CONFIG_NET_CLS_ACT
3824 if (act->ops->offload_act_setup) {
3825 return act->ops->offload_act_setup(act, entry, index_inc, true,
3826 extack);
3827 } else {
3828 NL_SET_ERR_MSG(extack, "Action does not support offload");
3829 return -EOPNOTSUPP;
3830 }
3831 #else
3832 return 0;
3833 #endif
3834 }
3835
3836 int tc_setup_action(struct flow_action *flow_action,
3837 struct tc_action *actions[],
3838 u32 miss_cookie_base,
3839 struct netlink_ext_ack *extack)
3840 {
3841 int i, j, k, index, err = 0;
3842 struct tc_action *act;
3843
3844 BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3845 BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3846 BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3847
3848 if (!actions)
3849 return 0;
3850
3851 j = 0;
3852 tcf_act_for_each_action(i, act, actions) {
3853 struct flow_action_entry *entry;
3854
3855 entry = &flow_action->entries[j];
3856 spin_lock_bh(&act->tcfa_lock);
3857 err = tcf_act_get_user_cookie(entry, act);
3858 if (err)
3859 goto err_out_locked;
3860
3861 index = 0;
3862 err = tc_setup_offload_act(act, entry, &index, extack);
3863 if (err)
3864 goto err_out_locked;
3865
3866 for (k = 0; k < index ; k++) {
3867 entry[k].hw_stats = tc_act_hw_stats(act->hw_stats);
3868 entry[k].hw_index = act->tcfa_index;
3869 entry[k].cookie = (unsigned long)act;
3870 entry[k].miss_cookie =
3871 tcf_exts_miss_cookie_get(miss_cookie_base, i);
3872 }
3873
3874 j += index;
3875
3876 spin_unlock_bh(&act->tcfa_lock);
3877 }
3878
3879 err_out:
3880 if (err)
3881 tc_cleanup_offload_action(flow_action);
3882
3883 return err;
3884 err_out_locked:
3885 spin_unlock_bh(&act->tcfa_lock);
3886 goto err_out;
3887 }
3888
3889 int tc_setup_offload_action(struct flow_action *flow_action,
3890 const struct tcf_exts *exts,
3891 struct netlink_ext_ack *extack)
3892 {
3893 #ifdef CONFIG_NET_CLS_ACT
3894 u32 miss_cookie_base;
3895
3896 if (!exts)
3897 return 0;
3898
3899 miss_cookie_base = exts->miss_cookie_node ?
3900 exts->miss_cookie_node->miss_cookie_base : 0;
3901 return tc_setup_action(flow_action, exts->actions, miss_cookie_base,
3902 extack);
3903 #else
3904 return 0;
3905 #endif
3906 }
3907 EXPORT_SYMBOL(tc_setup_offload_action);
3908
3909 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3910 {
3911 unsigned int num_acts = 0;
3912 struct tc_action *act;
3913 int i;
3914
3915 tcf_exts_for_each_action(i, act, exts) {
3916 if (is_tcf_pedit(act))
3917 num_acts += tcf_pedit_nkeys(act);
3918 else
3919 num_acts++;
3920 }
3921 return num_acts;
3922 }
3923 EXPORT_SYMBOL(tcf_exts_num_actions);
3924
3925 #ifdef CONFIG_NET_CLS_ACT
3926 static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3927 u32 *p_block_index,
3928 struct netlink_ext_ack *extack)
3929 {
3930 *p_block_index = nla_get_u32(block_index_attr);
3931 if (!*p_block_index) {
3932 NL_SET_ERR_MSG(extack, "Block number may not be zero");
3933 return -EINVAL;
3934 }
3935
3936 return 0;
3937 }
3938
3939 int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3940 enum flow_block_binder_type binder_type,
3941 struct nlattr *block_index_attr,
3942 struct netlink_ext_ack *extack)
3943 {
3944 u32 block_index;
3945 int err;
3946
3947 if (!block_index_attr)
3948 return 0;
3949
3950 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3951 if (err)
3952 return err;
3953
3954 qe->info.binder_type = binder_type;
3955 qe->info.chain_head_change = tcf_chain_head_change_dflt;
3956 qe->info.chain_head_change_priv = &qe->filter_chain;
3957 qe->info.block_index = block_index;
3958
3959 return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3960 }
3961 EXPORT_SYMBOL(tcf_qevent_init);
3962
3963 void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3964 {
3965 if (qe->info.block_index)
3966 tcf_block_put_ext(qe->block, sch, &qe->info);
3967 }
3968 EXPORT_SYMBOL(tcf_qevent_destroy);
3969
3970 int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3971 struct netlink_ext_ack *extack)
3972 {
3973 u32 block_index;
3974 int err;
3975
3976 if (!block_index_attr)
3977 return 0;
3978
3979 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3980 if (err)
3981 return err;
3982
3983 /* Bounce newly-configured block or change in block. */
3984 if (block_index != qe->info.block_index) {
3985 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3986 return -EINVAL;
3987 }
3988
3989 return 0;
3990 }
3991 EXPORT_SYMBOL(tcf_qevent_validate_change);
3992
3993 struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
3994 struct sk_buff **to_free, int *ret)
3995 {
3996 struct tcf_result cl_res;
3997 struct tcf_proto *fl;
3998
3999 if (!qe->info.block_index)
4000 return skb;
4001
4002 fl = rcu_dereference_bh(qe->filter_chain);
4003
4004 switch (tcf_classify(skb, NULL, fl, &cl_res, false)) {
4005 case TC_ACT_SHOT:
4006 qdisc_qstats_drop(sch);
4007 __qdisc_drop(skb, to_free);
4008 *ret = __NET_XMIT_BYPASS;
4009 return NULL;
4010 case TC_ACT_STOLEN:
4011 case TC_ACT_QUEUED:
4012 case TC_ACT_TRAP:
4013 __qdisc_drop(skb, to_free);
4014 *ret = __NET_XMIT_STOLEN;
4015 return NULL;
4016 case TC_ACT_REDIRECT:
4017 skb_do_redirect(skb);
4018 *ret = __NET_XMIT_STOLEN;
4019 return NULL;
4020 }
4021
4022 return skb;
4023 }
4024 EXPORT_SYMBOL(tcf_qevent_handle);
4025
4026 int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
4027 {
4028 if (!qe->info.block_index)
4029 return 0;
4030 return nla_put_u32(skb, attr_name, qe->info.block_index);
4031 }
4032 EXPORT_SYMBOL(tcf_qevent_dump);
4033 #endif
4034
4035 static __net_init int tcf_net_init(struct net *net)
4036 {
4037 struct tcf_net *tn = net_generic(net, tcf_net_id);
4038
4039 spin_lock_init(&tn->idr_lock);
4040 idr_init(&tn->idr);
4041 return 0;
4042 }
4043
4044 static void __net_exit tcf_net_exit(struct net *net)
4045 {
4046 struct tcf_net *tn = net_generic(net, tcf_net_id);
4047
4048 idr_destroy(&tn->idr);
4049 }
4050
4051 static struct pernet_operations tcf_net_ops = {
4052 .init = tcf_net_init,
4053 .exit = tcf_net_exit,
4054 .id = &tcf_net_id,
4055 .size = sizeof(struct tcf_net),
4056 };
4057
4058 static int __init tc_filter_init(void)
4059 {
4060 int err;
4061
4062 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
4063 if (!tc_filter_wq)
4064 return -ENOMEM;
4065
4066 err = register_pernet_subsys(&tcf_net_ops);
4067 if (err)
4068 goto err_register_pernet_subsys;
4069
4070 xa_init_flags(&tcf_exts_miss_cookies_xa, XA_FLAGS_ALLOC1);
4071
4072 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
4073 RTNL_FLAG_DOIT_UNLOCKED);
4074 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
4075 RTNL_FLAG_DOIT_UNLOCKED);
4076 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
4077 tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
4078 rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
4079 rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
4080 rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
4081 tc_dump_chain, 0);
4082
4083 return 0;
4084
4085 err_register_pernet_subsys:
4086 destroy_workqueue(tc_filter_wq);
4087 return err;
4088 }
4089
4090 subsys_initcall(tc_filter_init);
4091
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.