1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * net/sched/cls_api.c Packet classifier API.
5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
9 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
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/rhashtable.h>
24 #include <linux/jhash.h>
25 #include <linux/rculist.h>
26 #include <net/net_namespace.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/flow_offload.h>
44 extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
46 /* The list of all installed classifier types */
47 static LIST_HEAD(tcf_proto_base);
49 /* Protects list of registered TC modules. It is pure SMP lock. */
50 static DEFINE_RWLOCK(cls_mod_lock);
52 static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
54 return jhash_3words(tp->chain->index, tp->prio,
55 (__force __u32)tp->protocol, 0);
58 static void tcf_proto_signal_destroying(struct tcf_chain *chain,
61 struct tcf_block *block = chain->block;
63 mutex_lock(&block->proto_destroy_lock);
64 hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
65 destroy_obj_hashfn(tp));
66 mutex_unlock(&block->proto_destroy_lock);
69 static bool tcf_proto_cmp(const struct tcf_proto *tp1,
70 const struct tcf_proto *tp2)
72 return tp1->chain->index == tp2->chain->index &&
73 tp1->prio == tp2->prio &&
74 tp1->protocol == tp2->protocol;
77 static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
80 u32 hash = destroy_obj_hashfn(tp);
81 struct tcf_proto *iter;
85 hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
86 destroy_ht_node, hash) {
87 if (tcf_proto_cmp(tp, iter)) {
98 tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
100 struct tcf_block *block = chain->block;
102 mutex_lock(&block->proto_destroy_lock);
103 if (hash_hashed(&tp->destroy_ht_node))
104 hash_del_rcu(&tp->destroy_ht_node);
105 mutex_unlock(&block->proto_destroy_lock);
108 /* Find classifier type by string name */
110 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
112 const struct tcf_proto_ops *t, *res = NULL;
115 read_lock(&cls_mod_lock);
116 list_for_each_entry(t, &tcf_proto_base, head) {
117 if (strcmp(kind, t->kind) == 0) {
118 if (try_module_get(t->owner))
123 read_unlock(&cls_mod_lock);
128 static const struct tcf_proto_ops *
129 tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
130 struct netlink_ext_ack *extack)
132 const struct tcf_proto_ops *ops;
134 ops = __tcf_proto_lookup_ops(kind);
137 #ifdef CONFIG_MODULES
140 request_module("cls_%s", kind);
143 ops = __tcf_proto_lookup_ops(kind);
144 /* We dropped the RTNL semaphore in order to perform
145 * the module load. So, even if we succeeded in loading
146 * the module we have to replay the request. We indicate
147 * this using -EAGAIN.
150 module_put(ops->owner);
151 return ERR_PTR(-EAGAIN);
154 NL_SET_ERR_MSG(extack, "TC classifier not found");
155 return ERR_PTR(-ENOENT);
158 /* Register(unregister) new classifier type */
160 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
162 struct tcf_proto_ops *t;
165 write_lock(&cls_mod_lock);
166 list_for_each_entry(t, &tcf_proto_base, head)
167 if (!strcmp(ops->kind, t->kind))
170 list_add_tail(&ops->head, &tcf_proto_base);
173 write_unlock(&cls_mod_lock);
176 EXPORT_SYMBOL(register_tcf_proto_ops);
178 static struct workqueue_struct *tc_filter_wq;
180 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
182 struct tcf_proto_ops *t;
185 /* Wait for outstanding call_rcu()s, if any, from a
186 * tcf_proto_ops's destroy() handler.
189 flush_workqueue(tc_filter_wq);
191 write_lock(&cls_mod_lock);
192 list_for_each_entry(t, &tcf_proto_base, head) {
199 write_unlock(&cls_mod_lock);
202 EXPORT_SYMBOL(unregister_tcf_proto_ops);
204 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
206 INIT_RCU_WORK(rwork, func);
207 return queue_rcu_work(tc_filter_wq, rwork);
209 EXPORT_SYMBOL(tcf_queue_work);
211 /* Select new prio value from the range, managed by kernel. */
213 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
215 u32 first = TC_H_MAKE(0xC0000000U, 0U);
218 first = tp->prio - 1;
220 return TC_H_MAJ(first);
223 static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
226 return nla_strlcpy(name, kind, IFNAMSIZ) >= IFNAMSIZ;
227 memset(name, 0, IFNAMSIZ);
231 static bool tcf_proto_is_unlocked(const char *kind)
233 const struct tcf_proto_ops *ops;
236 if (strlen(kind) == 0)
239 ops = tcf_proto_lookup_ops(kind, false, NULL);
240 /* On error return false to take rtnl lock. Proto lookup/create
241 * functions will perform lookup again and properly handle errors.
246 ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
247 module_put(ops->owner);
251 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
252 u32 prio, struct tcf_chain *chain,
254 struct netlink_ext_ack *extack)
256 struct tcf_proto *tp;
259 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
261 return ERR_PTR(-ENOBUFS);
263 tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
264 if (IS_ERR(tp->ops)) {
265 err = PTR_ERR(tp->ops);
268 tp->classify = tp->ops->classify;
269 tp->protocol = protocol;
272 spin_lock_init(&tp->lock);
273 refcount_set(&tp->refcnt, 1);
275 err = tp->ops->init(tp);
277 module_put(tp->ops->owner);
287 static void tcf_proto_get(struct tcf_proto *tp)
289 refcount_inc(&tp->refcnt);
292 static void tcf_chain_put(struct tcf_chain *chain);
294 static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
295 bool sig_destroy, struct netlink_ext_ack *extack)
297 tp->ops->destroy(tp, rtnl_held, extack);
299 tcf_proto_signal_destroyed(tp->chain, tp);
300 tcf_chain_put(tp->chain);
301 module_put(tp->ops->owner);
305 static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
306 struct netlink_ext_ack *extack)
308 if (refcount_dec_and_test(&tp->refcnt))
309 tcf_proto_destroy(tp, rtnl_held, true, extack);
312 static bool tcf_proto_check_delete(struct tcf_proto *tp)
314 if (tp->ops->delete_empty)
315 return tp->ops->delete_empty(tp);
321 static void tcf_proto_mark_delete(struct tcf_proto *tp)
323 spin_lock(&tp->lock);
325 spin_unlock(&tp->lock);
328 static bool tcf_proto_is_deleting(struct tcf_proto *tp)
332 spin_lock(&tp->lock);
333 deleting = tp->deleting;
334 spin_unlock(&tp->lock);
339 #define ASSERT_BLOCK_LOCKED(block) \
340 lockdep_assert_held(&(block)->lock)
342 struct tcf_filter_chain_list_item {
343 struct list_head list;
344 tcf_chain_head_change_t *chain_head_change;
345 void *chain_head_change_priv;
348 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
351 struct tcf_chain *chain;
353 ASSERT_BLOCK_LOCKED(block);
355 chain = kzalloc(sizeof(*chain), GFP_KERNEL);
358 list_add_tail_rcu(&chain->list, &block->chain_list);
359 mutex_init(&chain->filter_chain_lock);
360 chain->block = block;
361 chain->index = chain_index;
364 block->chain0.chain = chain;
368 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
369 struct tcf_proto *tp_head)
371 if (item->chain_head_change)
372 item->chain_head_change(tp_head, item->chain_head_change_priv);
375 static void tcf_chain0_head_change(struct tcf_chain *chain,
376 struct tcf_proto *tp_head)
378 struct tcf_filter_chain_list_item *item;
379 struct tcf_block *block = chain->block;
384 mutex_lock(&block->lock);
385 list_for_each_entry(item, &block->chain0.filter_chain_list, list)
386 tcf_chain_head_change_item(item, tp_head);
387 mutex_unlock(&block->lock);
390 /* Returns true if block can be safely freed. */
392 static bool tcf_chain_detach(struct tcf_chain *chain)
394 struct tcf_block *block = chain->block;
396 ASSERT_BLOCK_LOCKED(block);
398 list_del_rcu(&chain->list);
400 block->chain0.chain = NULL;
402 if (list_empty(&block->chain_list) &&
403 refcount_read(&block->refcnt) == 0)
409 static void tcf_block_destroy(struct tcf_block *block)
411 mutex_destroy(&block->lock);
412 mutex_destroy(&block->proto_destroy_lock);
413 kfree_rcu(block, rcu);
416 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
418 struct tcf_block *block = chain->block;
420 mutex_destroy(&chain->filter_chain_lock);
421 kfree_rcu(chain, rcu);
423 tcf_block_destroy(block);
426 static void tcf_chain_hold(struct tcf_chain *chain)
428 ASSERT_BLOCK_LOCKED(chain->block);
433 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
435 ASSERT_BLOCK_LOCKED(chain->block);
437 /* In case all the references are action references, this
438 * chain should not be shown to the user.
440 return chain->refcnt == chain->action_refcnt;
443 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
446 struct tcf_chain *chain;
448 ASSERT_BLOCK_LOCKED(block);
450 list_for_each_entry(chain, &block->chain_list, list) {
451 if (chain->index == chain_index)
457 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
458 static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
461 struct tcf_chain *chain;
463 list_for_each_entry_rcu(chain, &block->chain_list, list) {
464 if (chain->index == chain_index)
471 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
472 u32 seq, u16 flags, int event, bool unicast);
474 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
475 u32 chain_index, bool create,
478 struct tcf_chain *chain = NULL;
479 bool is_first_reference;
481 mutex_lock(&block->lock);
482 chain = tcf_chain_lookup(block, chain_index);
484 tcf_chain_hold(chain);
488 chain = tcf_chain_create(block, chain_index);
494 ++chain->action_refcnt;
495 is_first_reference = chain->refcnt - chain->action_refcnt == 1;
496 mutex_unlock(&block->lock);
498 /* Send notification only in case we got the first
499 * non-action reference. Until then, the chain acts only as
500 * a placeholder for actions pointing to it and user ought
501 * not know about them.
503 if (is_first_reference && !by_act)
504 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
505 RTM_NEWCHAIN, false);
510 mutex_unlock(&block->lock);
514 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
517 return __tcf_chain_get(block, chain_index, create, false);
520 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
522 return __tcf_chain_get(block, chain_index, true, true);
524 EXPORT_SYMBOL(tcf_chain_get_by_act);
526 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
528 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
529 void *tmplt_priv, u32 chain_index,
530 struct tcf_block *block, struct sk_buff *oskb,
531 u32 seq, u16 flags, bool unicast);
533 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
534 bool explicitly_created)
536 struct tcf_block *block = chain->block;
537 const struct tcf_proto_ops *tmplt_ops;
538 bool free_block = false;
542 mutex_lock(&block->lock);
543 if (explicitly_created) {
544 if (!chain->explicitly_created) {
545 mutex_unlock(&block->lock);
548 chain->explicitly_created = false;
552 chain->action_refcnt--;
554 /* tc_chain_notify_delete can't be called while holding block lock.
555 * However, when block is unlocked chain can be changed concurrently, so
556 * save these to temporary variables.
558 refcnt = --chain->refcnt;
559 tmplt_ops = chain->tmplt_ops;
560 tmplt_priv = chain->tmplt_priv;
562 /* The last dropped non-action reference will trigger notification. */
563 if (refcnt - chain->action_refcnt == 0 && !by_act) {
564 tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
565 block, NULL, 0, 0, false);
566 /* Last reference to chain, no need to lock. */
567 chain->flushing = false;
571 free_block = tcf_chain_detach(chain);
572 mutex_unlock(&block->lock);
575 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
576 tcf_chain_destroy(chain, free_block);
580 static void tcf_chain_put(struct tcf_chain *chain)
582 __tcf_chain_put(chain, false, false);
585 void tcf_chain_put_by_act(struct tcf_chain *chain)
587 __tcf_chain_put(chain, true, false);
589 EXPORT_SYMBOL(tcf_chain_put_by_act);
591 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
593 __tcf_chain_put(chain, false, true);
596 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
598 struct tcf_proto *tp, *tp_next;
600 mutex_lock(&chain->filter_chain_lock);
601 tp = tcf_chain_dereference(chain->filter_chain, chain);
603 tp_next = rcu_dereference_protected(tp->next, 1);
604 tcf_proto_signal_destroying(chain, tp);
607 tp = tcf_chain_dereference(chain->filter_chain, chain);
608 RCU_INIT_POINTER(chain->filter_chain, NULL);
609 tcf_chain0_head_change(chain, NULL);
610 chain->flushing = true;
611 mutex_unlock(&chain->filter_chain_lock);
614 tp_next = rcu_dereference_protected(tp->next, 1);
615 tcf_proto_put(tp, rtnl_held, NULL);
620 static int tcf_block_setup(struct tcf_block *block,
621 struct flow_block_offload *bo);
623 static void tc_indr_block_cmd(struct net_device *dev, struct tcf_block *block,
624 flow_indr_block_bind_cb_t *cb, void *cb_priv,
625 enum flow_block_command command, bool ingress)
627 struct flow_block_offload bo = {
629 .binder_type = ingress ?
630 FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS :
631 FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS,
633 .block_shared = tcf_block_non_null_shared(block),
635 INIT_LIST_HEAD(&bo.cb_list);
640 bo.block = &block->flow_block;
642 down_write(&block->cb_lock);
643 cb(dev, cb_priv, TC_SETUP_BLOCK, &bo);
645 tcf_block_setup(block, &bo);
646 up_write(&block->cb_lock);
649 static struct tcf_block *tc_dev_block(struct net_device *dev, bool ingress)
651 const struct Qdisc_class_ops *cops;
652 const struct Qdisc_ops *ops;
655 if (!dev_ingress_queue(dev))
658 qdisc = dev_ingress_queue(dev)->qdisc_sleeping;
666 if (!ingress && !strcmp("ingress", ops->id))
673 if (!cops->tcf_block)
676 return cops->tcf_block(qdisc,
677 ingress ? TC_H_MIN_INGRESS : TC_H_MIN_EGRESS,
681 static void tc_indr_block_get_and_cmd(struct net_device *dev,
682 flow_indr_block_bind_cb_t *cb,
684 enum flow_block_command command)
686 struct tcf_block *block;
688 block = tc_dev_block(dev, true);
689 tc_indr_block_cmd(dev, block, cb, cb_priv, command, true);
691 block = tc_dev_block(dev, false);
692 tc_indr_block_cmd(dev, block, cb, cb_priv, command, false);
695 static void tc_indr_block_call(struct tcf_block *block,
696 struct net_device *dev,
697 struct tcf_block_ext_info *ei,
698 enum flow_block_command command,
699 struct netlink_ext_ack *extack)
701 struct flow_block_offload bo = {
703 .binder_type = ei->binder_type,
705 .block = &block->flow_block,
706 .block_shared = tcf_block_shared(block),
709 INIT_LIST_HEAD(&bo.cb_list);
711 flow_indr_block_call(dev, &bo, command, TC_SETUP_BLOCK);
712 tcf_block_setup(block, &bo);
715 static bool tcf_block_offload_in_use(struct tcf_block *block)
717 return atomic_read(&block->offloadcnt);
720 static int tcf_block_offload_cmd(struct tcf_block *block,
721 struct net_device *dev,
722 struct tcf_block_ext_info *ei,
723 enum flow_block_command command,
724 struct netlink_ext_ack *extack)
726 struct flow_block_offload bo = {};
729 bo.net = dev_net(dev);
730 bo.command = command;
731 bo.binder_type = ei->binder_type;
732 bo.block = &block->flow_block;
733 bo.block_shared = tcf_block_shared(block);
735 INIT_LIST_HEAD(&bo.cb_list);
737 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
741 return tcf_block_setup(block, &bo);
744 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
745 struct tcf_block_ext_info *ei,
746 struct netlink_ext_ack *extack)
748 struct net_device *dev = q->dev_queue->dev;
751 down_write(&block->cb_lock);
752 if (!dev->netdev_ops->ndo_setup_tc)
753 goto no_offload_dev_inc;
755 /* If tc offload feature is disabled and the block we try to bind
756 * to already has some offloaded filters, forbid to bind.
758 if (!tc_can_offload(dev) && tcf_block_offload_in_use(block)) {
759 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
764 err = tcf_block_offload_cmd(block, dev, ei, FLOW_BLOCK_BIND, extack);
765 if (err == -EOPNOTSUPP)
766 goto no_offload_dev_inc;
770 tc_indr_block_call(block, dev, ei, FLOW_BLOCK_BIND, extack);
771 up_write(&block->cb_lock);
775 if (tcf_block_offload_in_use(block)) {
780 block->nooffloaddevcnt++;
781 tc_indr_block_call(block, dev, ei, FLOW_BLOCK_BIND, extack);
783 up_write(&block->cb_lock);
787 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
788 struct tcf_block_ext_info *ei)
790 struct net_device *dev = q->dev_queue->dev;
793 down_write(&block->cb_lock);
794 tc_indr_block_call(block, dev, ei, FLOW_BLOCK_UNBIND, NULL);
796 if (!dev->netdev_ops->ndo_setup_tc)
797 goto no_offload_dev_dec;
798 err = tcf_block_offload_cmd(block, dev, ei, FLOW_BLOCK_UNBIND, NULL);
799 if (err == -EOPNOTSUPP)
800 goto no_offload_dev_dec;
801 up_write(&block->cb_lock);
805 WARN_ON(block->nooffloaddevcnt-- == 0);
806 up_write(&block->cb_lock);
810 tcf_chain0_head_change_cb_add(struct tcf_block *block,
811 struct tcf_block_ext_info *ei,
812 struct netlink_ext_ack *extack)
814 struct tcf_filter_chain_list_item *item;
815 struct tcf_chain *chain0;
817 item = kmalloc(sizeof(*item), GFP_KERNEL);
819 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
822 item->chain_head_change = ei->chain_head_change;
823 item->chain_head_change_priv = ei->chain_head_change_priv;
825 mutex_lock(&block->lock);
826 chain0 = block->chain0.chain;
828 tcf_chain_hold(chain0);
830 list_add(&item->list, &block->chain0.filter_chain_list);
831 mutex_unlock(&block->lock);
834 struct tcf_proto *tp_head;
836 mutex_lock(&chain0->filter_chain_lock);
838 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
840 tcf_chain_head_change_item(item, tp_head);
842 mutex_lock(&block->lock);
843 list_add(&item->list, &block->chain0.filter_chain_list);
844 mutex_unlock(&block->lock);
846 mutex_unlock(&chain0->filter_chain_lock);
847 tcf_chain_put(chain0);
854 tcf_chain0_head_change_cb_del(struct tcf_block *block,
855 struct tcf_block_ext_info *ei)
857 struct tcf_filter_chain_list_item *item;
859 mutex_lock(&block->lock);
860 list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
861 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
862 (item->chain_head_change == ei->chain_head_change &&
863 item->chain_head_change_priv == ei->chain_head_change_priv)) {
864 if (block->chain0.chain)
865 tcf_chain_head_change_item(item, NULL);
866 list_del(&item->list);
867 mutex_unlock(&block->lock);
873 mutex_unlock(&block->lock);
878 spinlock_t idr_lock; /* Protects idr */
882 static unsigned int tcf_net_id;
884 static int tcf_block_insert(struct tcf_block *block, struct net *net,
885 struct netlink_ext_ack *extack)
887 struct tcf_net *tn = net_generic(net, tcf_net_id);
890 idr_preload(GFP_KERNEL);
891 spin_lock(&tn->idr_lock);
892 err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
894 spin_unlock(&tn->idr_lock);
900 static void tcf_block_remove(struct tcf_block *block, struct net *net)
902 struct tcf_net *tn = net_generic(net, tcf_net_id);
904 spin_lock(&tn->idr_lock);
905 idr_remove(&tn->idr, block->index);
906 spin_unlock(&tn->idr_lock);
909 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
911 struct netlink_ext_ack *extack)
913 struct tcf_block *block;
915 block = kzalloc(sizeof(*block), GFP_KERNEL);
917 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
918 return ERR_PTR(-ENOMEM);
920 mutex_init(&block->lock);
921 mutex_init(&block->proto_destroy_lock);
922 init_rwsem(&block->cb_lock);
923 flow_block_init(&block->flow_block);
924 INIT_LIST_HEAD(&block->chain_list);
925 INIT_LIST_HEAD(&block->owner_list);
926 INIT_LIST_HEAD(&block->chain0.filter_chain_list);
928 refcount_set(&block->refcnt, 1);
930 block->index = block_index;
932 /* Don't store q pointer for blocks which are shared */
933 if (!tcf_block_shared(block))
938 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
940 struct tcf_net *tn = net_generic(net, tcf_net_id);
942 return idr_find(&tn->idr, block_index);
945 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
947 struct tcf_block *block;
950 block = tcf_block_lookup(net, block_index);
951 if (block && !refcount_inc_not_zero(&block->refcnt))
958 static struct tcf_chain *
959 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
961 mutex_lock(&block->lock);
963 chain = list_is_last(&chain->list, &block->chain_list) ?
964 NULL : list_next_entry(chain, list);
966 chain = list_first_entry_or_null(&block->chain_list,
967 struct tcf_chain, list);
969 /* skip all action-only chains */
970 while (chain && tcf_chain_held_by_acts_only(chain))
971 chain = list_is_last(&chain->list, &block->chain_list) ?
972 NULL : list_next_entry(chain, list);
975 tcf_chain_hold(chain);
976 mutex_unlock(&block->lock);
981 /* Function to be used by all clients that want to iterate over all chains on
982 * block. It properly obtains block->lock and takes reference to chain before
983 * returning it. Users of this function must be tolerant to concurrent chain
984 * insertion/deletion or ensure that no concurrent chain modification is
985 * possible. Note that all netlink dump callbacks cannot guarantee to provide
986 * consistent dump because rtnl lock is released each time skb is filled with
987 * data and sent to user-space.
991 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
993 struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
996 tcf_chain_put(chain);
1000 EXPORT_SYMBOL(tcf_get_next_chain);
1002 static struct tcf_proto *
1003 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1008 mutex_lock(&chain->filter_chain_lock);
1011 tp = tcf_chain_dereference(chain->filter_chain, chain);
1012 } else if (tcf_proto_is_deleting(tp)) {
1013 /* 'deleting' flag is set and chain->filter_chain_lock was
1014 * unlocked, which means next pointer could be invalid. Restart
1017 prio = tp->prio + 1;
1018 tp = tcf_chain_dereference(chain->filter_chain, chain);
1020 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1021 if (!tp->deleting && tp->prio >= prio)
1024 tp = tcf_chain_dereference(tp->next, chain);
1030 mutex_unlock(&chain->filter_chain_lock);
1035 /* Function to be used by all clients that want to iterate over all tp's on
1036 * chain. Users of this function must be tolerant to concurrent tp
1037 * insertion/deletion or ensure that no concurrent chain modification is
1038 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1039 * consistent dump because rtnl lock is released each time skb is filled with
1040 * data and sent to user-space.
1044 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp,
1047 struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1050 tcf_proto_put(tp, rtnl_held, NULL);
1054 EXPORT_SYMBOL(tcf_get_next_proto);
1056 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1058 struct tcf_chain *chain;
1060 /* Last reference to block. At this point chains cannot be added or
1061 * removed concurrently.
1063 for (chain = tcf_get_next_chain(block, NULL);
1065 chain = tcf_get_next_chain(block, chain)) {
1066 tcf_chain_put_explicitly_created(chain);
1067 tcf_chain_flush(chain, rtnl_held);
1071 /* Lookup Qdisc and increments its reference counter.
1072 * Set parent, if necessary.
1075 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1076 u32 *parent, int ifindex, bool rtnl_held,
1077 struct netlink_ext_ack *extack)
1079 const struct Qdisc_class_ops *cops;
1080 struct net_device *dev;
1083 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1089 dev = dev_get_by_index_rcu(net, ifindex);
1098 *parent = (*q)->handle;
1100 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1102 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1108 *q = qdisc_refcount_inc_nz(*q);
1110 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1115 /* Is it classful? */
1116 cops = (*q)->ops->cl_ops;
1118 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1123 if (!cops->tcf_block) {
1124 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1130 /* At this point we know that qdisc is not noop_qdisc,
1131 * which means that qdisc holds a reference to net_device
1132 * and we hold a reference to qdisc, so it is safe to release
1144 qdisc_put_unlocked(*q);
1150 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1151 int ifindex, struct netlink_ext_ack *extack)
1153 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1156 /* Do we search for filter, attached to class? */
1157 if (TC_H_MIN(parent)) {
1158 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1160 *cl = cops->find(q, parent);
1162 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1170 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1171 unsigned long cl, int ifindex,
1173 struct netlink_ext_ack *extack)
1175 struct tcf_block *block;
1177 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1178 block = tcf_block_refcnt_get(net, block_index);
1180 NL_SET_ERR_MSG(extack, "Block of given index was not found");
1181 return ERR_PTR(-EINVAL);
1184 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1186 block = cops->tcf_block(q, cl, extack);
1188 return ERR_PTR(-EINVAL);
1190 if (tcf_block_shared(block)) {
1191 NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1192 return ERR_PTR(-EOPNOTSUPP);
1195 /* Always take reference to block in order to support execution
1196 * of rules update path of cls API without rtnl lock. Caller
1197 * must release block when it is finished using it. 'if' block
1198 * of this conditional obtain reference to block by calling
1199 * tcf_block_refcnt_get().
1201 refcount_inc(&block->refcnt);
1207 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1208 struct tcf_block_ext_info *ei, bool rtnl_held)
1210 if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1211 /* Flushing/putting all chains will cause the block to be
1212 * deallocated when last chain is freed. However, if chain_list
1213 * is empty, block has to be manually deallocated. After block
1214 * reference counter reached 0, it is no longer possible to
1215 * increment it or add new chains to block.
1217 bool free_block = list_empty(&block->chain_list);
1219 mutex_unlock(&block->lock);
1220 if (tcf_block_shared(block))
1221 tcf_block_remove(block, block->net);
1224 tcf_block_offload_unbind(block, q, ei);
1227 tcf_block_destroy(block);
1229 tcf_block_flush_all_chains(block, rtnl_held);
1231 tcf_block_offload_unbind(block, q, ei);
1235 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1237 __tcf_block_put(block, NULL, NULL, rtnl_held);
1241 * Set q, parent, cl when appropriate.
1244 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1245 u32 *parent, unsigned long *cl,
1246 int ifindex, u32 block_index,
1247 struct netlink_ext_ack *extack)
1249 struct tcf_block *block;
1254 err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1258 err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1262 block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1263 if (IS_ERR(block)) {
1264 err = PTR_ERR(block);
1275 return ERR_PTR(err);
1278 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1281 if (!IS_ERR_OR_NULL(block))
1282 tcf_block_refcnt_put(block, rtnl_held);
1288 qdisc_put_unlocked(q);
1292 struct tcf_block_owner_item {
1293 struct list_head list;
1295 enum flow_block_binder_type binder_type;
1299 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1301 enum flow_block_binder_type binder_type)
1303 if (block->keep_dst &&
1304 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1305 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1306 netif_keep_dst(qdisc_dev(q));
1309 void tcf_block_netif_keep_dst(struct tcf_block *block)
1311 struct tcf_block_owner_item *item;
1313 block->keep_dst = true;
1314 list_for_each_entry(item, &block->owner_list, list)
1315 tcf_block_owner_netif_keep_dst(block, item->q,
1318 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1320 static int tcf_block_owner_add(struct tcf_block *block,
1322 enum flow_block_binder_type binder_type)
1324 struct tcf_block_owner_item *item;
1326 item = kmalloc(sizeof(*item), GFP_KERNEL);
1330 item->binder_type = binder_type;
1331 list_add(&item->list, &block->owner_list);
1335 static void tcf_block_owner_del(struct tcf_block *block,
1337 enum flow_block_binder_type binder_type)
1339 struct tcf_block_owner_item *item;
1341 list_for_each_entry(item, &block->owner_list, list) {
1342 if (item->q == q && item->binder_type == binder_type) {
1343 list_del(&item->list);
1351 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1352 struct tcf_block_ext_info *ei,
1353 struct netlink_ext_ack *extack)
1355 struct net *net = qdisc_net(q);
1356 struct tcf_block *block = NULL;
1359 if (ei->block_index)
1360 /* block_index not 0 means the shared block is requested */
1361 block = tcf_block_refcnt_get(net, ei->block_index);
1364 block = tcf_block_create(net, q, ei->block_index, extack);
1366 return PTR_ERR(block);
1367 if (tcf_block_shared(block)) {
1368 err = tcf_block_insert(block, net, extack);
1370 goto err_block_insert;
1374 err = tcf_block_owner_add(block, q, ei->binder_type);
1376 goto err_block_owner_add;
1378 tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1380 err = tcf_chain0_head_change_cb_add(block, ei, extack);
1382 goto err_chain0_head_change_cb_add;
1384 err = tcf_block_offload_bind(block, q, ei, extack);
1386 goto err_block_offload_bind;
1391 err_block_offload_bind:
1392 tcf_chain0_head_change_cb_del(block, ei);
1393 err_chain0_head_change_cb_add:
1394 tcf_block_owner_del(block, q, ei->binder_type);
1395 err_block_owner_add:
1397 tcf_block_refcnt_put(block, true);
1400 EXPORT_SYMBOL(tcf_block_get_ext);
1402 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1404 struct tcf_proto __rcu **p_filter_chain = priv;
1406 rcu_assign_pointer(*p_filter_chain, tp_head);
1409 int tcf_block_get(struct tcf_block **p_block,
1410 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1411 struct netlink_ext_ack *extack)
1413 struct tcf_block_ext_info ei = {
1414 .chain_head_change = tcf_chain_head_change_dflt,
1415 .chain_head_change_priv = p_filter_chain,
1418 WARN_ON(!p_filter_chain);
1419 return tcf_block_get_ext(p_block, q, &ei, extack);
1421 EXPORT_SYMBOL(tcf_block_get);
1423 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1424 * actions should be all removed after flushing.
1426 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1427 struct tcf_block_ext_info *ei)
1431 tcf_chain0_head_change_cb_del(block, ei);
1432 tcf_block_owner_del(block, q, ei->binder_type);
1434 __tcf_block_put(block, q, ei, true);
1436 EXPORT_SYMBOL(tcf_block_put_ext);
1438 void tcf_block_put(struct tcf_block *block)
1440 struct tcf_block_ext_info ei = {0, };
1444 tcf_block_put_ext(block, block->q, &ei);
1447 EXPORT_SYMBOL(tcf_block_put);
1450 tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1451 void *cb_priv, bool add, bool offload_in_use,
1452 struct netlink_ext_ack *extack)
1454 struct tcf_chain *chain, *chain_prev;
1455 struct tcf_proto *tp, *tp_prev;
1458 lockdep_assert_held(&block->cb_lock);
1460 for (chain = __tcf_get_next_chain(block, NULL);
1463 chain = __tcf_get_next_chain(block, chain),
1464 tcf_chain_put(chain_prev)) {
1465 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1467 tp = __tcf_get_next_proto(chain, tp),
1468 tcf_proto_put(tp_prev, true, NULL)) {
1469 if (tp->ops->reoffload) {
1470 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1473 goto err_playback_remove;
1474 } else if (add && offload_in_use) {
1476 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1477 goto err_playback_remove;
1484 err_playback_remove:
1485 tcf_proto_put(tp, true, NULL);
1486 tcf_chain_put(chain);
1487 tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1492 static int tcf_block_bind(struct tcf_block *block,
1493 struct flow_block_offload *bo)
1495 struct flow_block_cb *block_cb, *next;
1498 lockdep_assert_held(&block->cb_lock);
1500 list_for_each_entry(block_cb, &bo->cb_list, list) {
1501 err = tcf_block_playback_offloads(block, block_cb->cb,
1502 block_cb->cb_priv, true,
1503 tcf_block_offload_in_use(block),
1507 if (!bo->unlocked_driver_cb)
1508 block->lockeddevcnt++;
1512 list_splice(&bo->cb_list, &block->flow_block.cb_list);
1517 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1519 list_del(&block_cb->list);
1520 tcf_block_playback_offloads(block, block_cb->cb,
1521 block_cb->cb_priv, false,
1522 tcf_block_offload_in_use(block),
1524 if (!bo->unlocked_driver_cb)
1525 block->lockeddevcnt--;
1527 flow_block_cb_free(block_cb);
1533 static void tcf_block_unbind(struct tcf_block *block,
1534 struct flow_block_offload *bo)
1536 struct flow_block_cb *block_cb, *next;
1538 lockdep_assert_held(&block->cb_lock);
1540 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1541 tcf_block_playback_offloads(block, block_cb->cb,
1542 block_cb->cb_priv, false,
1543 tcf_block_offload_in_use(block),
1545 list_del(&block_cb->list);
1546 flow_block_cb_free(block_cb);
1547 if (!bo->unlocked_driver_cb)
1548 block->lockeddevcnt--;
1552 static int tcf_block_setup(struct tcf_block *block,
1553 struct flow_block_offload *bo)
1557 switch (bo->command) {
1558 case FLOW_BLOCK_BIND:
1559 err = tcf_block_bind(block, bo);
1561 case FLOW_BLOCK_UNBIND:
1563 tcf_block_unbind(block, bo);
1573 /* Main classifier routine: scans classifier chain attached
1574 * to this qdisc, (optionally) tests for protocol and asks
1575 * specific classifiers.
1577 static inline int __tcf_classify(struct sk_buff *skb,
1578 const struct tcf_proto *tp,
1579 const struct tcf_proto *orig_tp,
1580 struct tcf_result *res,
1582 u32 *last_executed_chain)
1584 #ifdef CONFIG_NET_CLS_ACT
1585 const int max_reclassify_loop = 4;
1586 const struct tcf_proto *first_tp;
1591 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1592 __be16 protocol = tc_skb_protocol(skb);
1595 if (tp->protocol != protocol &&
1596 tp->protocol != htons(ETH_P_ALL))
1599 err = tp->classify(skb, tp, res);
1600 #ifdef CONFIG_NET_CLS_ACT
1601 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1603 *last_executed_chain = first_tp->chain->index;
1605 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1606 first_tp = res->goto_tp;
1607 *last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1615 return TC_ACT_UNSPEC; /* signal: continue lookup */
1616 #ifdef CONFIG_NET_CLS_ACT
1618 if (unlikely(limit++ >= max_reclassify_loop)) {
1619 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1620 tp->chain->block->index,
1622 ntohs(tp->protocol));
1631 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1632 struct tcf_result *res, bool compat_mode)
1634 u32 last_executed_chain = 0;
1636 return __tcf_classify(skb, tp, tp, res, compat_mode,
1637 &last_executed_chain);
1639 EXPORT_SYMBOL(tcf_classify);
1641 int tcf_classify_ingress(struct sk_buff *skb,
1642 const struct tcf_block *ingress_block,
1643 const struct tcf_proto *tp,
1644 struct tcf_result *res, bool compat_mode)
1646 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1647 u32 last_executed_chain = 0;
1649 return __tcf_classify(skb, tp, tp, res, compat_mode,
1650 &last_executed_chain);
1652 u32 last_executed_chain = tp ? tp->chain->index : 0;
1653 const struct tcf_proto *orig_tp = tp;
1654 struct tc_skb_ext *ext;
1657 ext = skb_ext_find(skb, TC_SKB_EXT);
1659 if (ext && ext->chain) {
1660 struct tcf_chain *fchain;
1662 fchain = tcf_chain_lookup_rcu(ingress_block, ext->chain);
1666 /* Consume, so cloned/redirect skbs won't inherit ext */
1667 skb_ext_del(skb, TC_SKB_EXT);
1669 tp = rcu_dereference_bh(fchain->filter_chain);
1672 ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode,
1673 &last_executed_chain);
1675 /* If we missed on some chain */
1676 if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1677 ext = skb_ext_add(skb, TC_SKB_EXT);
1678 if (WARN_ON_ONCE(!ext))
1680 ext->chain = last_executed_chain;
1686 EXPORT_SYMBOL(tcf_classify_ingress);
1688 struct tcf_chain_info {
1689 struct tcf_proto __rcu **pprev;
1690 struct tcf_proto __rcu *next;
1693 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1694 struct tcf_chain_info *chain_info)
1696 return tcf_chain_dereference(*chain_info->pprev, chain);
1699 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1700 struct tcf_chain_info *chain_info,
1701 struct tcf_proto *tp)
1703 if (chain->flushing)
1706 if (*chain_info->pprev == chain->filter_chain)
1707 tcf_chain0_head_change(chain, tp);
1709 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1710 rcu_assign_pointer(*chain_info->pprev, tp);
1715 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1716 struct tcf_chain_info *chain_info,
1717 struct tcf_proto *tp)
1719 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1721 tcf_proto_mark_delete(tp);
1722 if (tp == chain->filter_chain)
1723 tcf_chain0_head_change(chain, next);
1724 RCU_INIT_POINTER(*chain_info->pprev, next);
1727 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1728 struct tcf_chain_info *chain_info,
1729 u32 protocol, u32 prio,
1730 bool prio_allocate);
1732 /* Try to insert new proto.
1733 * If proto with specified priority already exists, free new proto
1734 * and return existing one.
1737 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1738 struct tcf_proto *tp_new,
1739 u32 protocol, u32 prio,
1742 struct tcf_chain_info chain_info;
1743 struct tcf_proto *tp;
1746 mutex_lock(&chain->filter_chain_lock);
1748 if (tcf_proto_exists_destroying(chain, tp_new)) {
1749 mutex_unlock(&chain->filter_chain_lock);
1750 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1751 return ERR_PTR(-EAGAIN);
1754 tp = tcf_chain_tp_find(chain, &chain_info,
1755 protocol, prio, false);
1757 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1758 mutex_unlock(&chain->filter_chain_lock);
1761 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1764 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1765 tp_new = ERR_PTR(err);
1771 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1772 struct tcf_proto *tp, bool rtnl_held,
1773 struct netlink_ext_ack *extack)
1775 struct tcf_chain_info chain_info;
1776 struct tcf_proto *tp_iter;
1777 struct tcf_proto **pprev;
1778 struct tcf_proto *next;
1780 mutex_lock(&chain->filter_chain_lock);
1782 /* Atomically find and remove tp from chain. */
1783 for (pprev = &chain->filter_chain;
1784 (tp_iter = tcf_chain_dereference(*pprev, chain));
1785 pprev = &tp_iter->next) {
1786 if (tp_iter == tp) {
1787 chain_info.pprev = pprev;
1788 chain_info.next = tp_iter->next;
1789 WARN_ON(tp_iter->deleting);
1793 /* Verify that tp still exists and no new filters were inserted
1795 * Mark tp for deletion if it is empty.
1797 if (!tp_iter || !tcf_proto_check_delete(tp)) {
1798 mutex_unlock(&chain->filter_chain_lock);
1802 tcf_proto_signal_destroying(chain, tp);
1803 next = tcf_chain_dereference(chain_info.next, chain);
1804 if (tp == chain->filter_chain)
1805 tcf_chain0_head_change(chain, next);
1806 RCU_INIT_POINTER(*chain_info.pprev, next);
1807 mutex_unlock(&chain->filter_chain_lock);
1809 tcf_proto_put(tp, rtnl_held, extack);
1812 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1813 struct tcf_chain_info *chain_info,
1814 u32 protocol, u32 prio,
1817 struct tcf_proto **pprev;
1818 struct tcf_proto *tp;
1820 /* Check the chain for existence of proto-tcf with this priority */
1821 for (pprev = &chain->filter_chain;
1822 (tp = tcf_chain_dereference(*pprev, chain));
1823 pprev = &tp->next) {
1824 if (tp->prio >= prio) {
1825 if (tp->prio == prio) {
1826 if (prio_allocate ||
1827 (tp->protocol != protocol && protocol))
1828 return ERR_PTR(-EINVAL);
1835 chain_info->pprev = pprev;
1837 chain_info->next = tp->next;
1840 chain_info->next = NULL;
1845 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1846 struct tcf_proto *tp, struct tcf_block *block,
1847 struct Qdisc *q, u32 parent, void *fh,
1848 u32 portid, u32 seq, u16 flags, int event,
1852 struct nlmsghdr *nlh;
1853 unsigned char *b = skb_tail_pointer(skb);
1855 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1857 goto out_nlmsg_trim;
1858 tcm = nlmsg_data(nlh);
1859 tcm->tcm_family = AF_UNSPEC;
1863 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1864 tcm->tcm_parent = parent;
1866 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1867 tcm->tcm_block_index = block->index;
1869 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1870 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1871 goto nla_put_failure;
1872 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1873 goto nla_put_failure;
1875 tcm->tcm_handle = 0;
1877 if (tp->ops->dump &&
1878 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
1879 goto nla_put_failure;
1881 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1890 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1891 struct nlmsghdr *n, struct tcf_proto *tp,
1892 struct tcf_block *block, struct Qdisc *q,
1893 u32 parent, void *fh, int event, bool unicast,
1896 struct sk_buff *skb;
1897 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1900 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1904 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1905 n->nlmsg_seq, n->nlmsg_flags, event,
1912 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1914 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1915 n->nlmsg_flags & NLM_F_ECHO);
1922 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1923 struct nlmsghdr *n, struct tcf_proto *tp,
1924 struct tcf_block *block, struct Qdisc *q,
1925 u32 parent, void *fh, bool unicast, bool *last,
1926 bool rtnl_held, struct netlink_ext_ack *extack)
1928 struct sk_buff *skb;
1929 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1932 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1936 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1937 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
1939 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1944 err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
1951 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1953 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1954 n->nlmsg_flags & NLM_F_ECHO);
1956 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1963 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1964 struct tcf_block *block, struct Qdisc *q,
1965 u32 parent, struct nlmsghdr *n,
1966 struct tcf_chain *chain, int event,
1969 struct tcf_proto *tp;
1971 for (tp = tcf_get_next_proto(chain, NULL, rtnl_held);
1972 tp; tp = tcf_get_next_proto(chain, tp, rtnl_held))
1973 tfilter_notify(net, oskb, n, tp, block,
1974 q, parent, NULL, event, false, rtnl_held);
1977 static void tfilter_put(struct tcf_proto *tp, void *fh)
1979 if (tp->ops->put && fh)
1980 tp->ops->put(tp, fh);
1983 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1984 struct netlink_ext_ack *extack)
1986 struct net *net = sock_net(skb->sk);
1987 struct nlattr *tca[TCA_MAX + 1];
1988 char name[IFNAMSIZ];
1995 struct Qdisc *q = NULL;
1996 struct tcf_chain_info chain_info;
1997 struct tcf_chain *chain = NULL;
1998 struct tcf_block *block;
1999 struct tcf_proto *tp;
2004 bool rtnl_held = false;
2006 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2012 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2013 rtm_tca_policy, extack);
2018 protocol = TC_H_MIN(t->tcm_info);
2019 prio = TC_H_MAJ(t->tcm_info);
2020 prio_allocate = false;
2021 parent = t->tcm_parent;
2027 /* If no priority is provided by the user,
2030 if (n->nlmsg_flags & NLM_F_CREATE) {
2031 prio = TC_H_MAKE(0x80000000U, 0U);
2032 prio_allocate = true;
2034 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2039 /* Find head of filter chain. */
2041 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2045 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2046 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2051 /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2052 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2053 * type is not specified, classifier is not unlocked.
2056 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2057 !tcf_proto_is_unlocked(name)) {
2062 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2066 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2068 if (IS_ERR(block)) {
2069 err = PTR_ERR(block);
2073 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2074 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2075 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2079 chain = tcf_chain_get(block, chain_index, true);
2081 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2086 mutex_lock(&chain->filter_chain_lock);
2087 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2088 prio, prio_allocate);
2090 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2096 struct tcf_proto *tp_new = NULL;
2098 if (chain->flushing) {
2103 /* Proto-tcf does not exist, create new one */
2105 if (tca[TCA_KIND] == NULL || !protocol) {
2106 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2111 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2112 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2118 prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2121 mutex_unlock(&chain->filter_chain_lock);
2122 tp_new = tcf_proto_create(name, protocol, prio, chain,
2124 if (IS_ERR(tp_new)) {
2125 err = PTR_ERR(tp_new);
2130 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2137 mutex_unlock(&chain->filter_chain_lock);
2140 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2141 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2146 fh = tp->ops->get(tp, t->tcm_handle);
2149 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2150 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2154 } else if (n->nlmsg_flags & NLM_F_EXCL) {
2155 tfilter_put(tp, fh);
2156 NL_SET_ERR_MSG(extack, "Filter already exists");
2161 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2162 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2167 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2168 n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
2171 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2172 RTM_NEWTFILTER, false, rtnl_held);
2173 tfilter_put(tp, fh);
2174 /* q pointer is NULL for shared blocks */
2176 q->flags &= ~TCQ_F_CAN_BYPASS;
2180 if (err && tp_created)
2181 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2184 if (tp && !IS_ERR(tp))
2185 tcf_proto_put(tp, rtnl_held, NULL);
2187 tcf_chain_put(chain);
2189 tcf_block_release(q, block, rtnl_held);
2194 if (err == -EAGAIN) {
2195 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2199 /* Replay the request. */
2205 mutex_unlock(&chain->filter_chain_lock);
2209 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2210 struct netlink_ext_ack *extack)
2212 struct net *net = sock_net(skb->sk);
2213 struct nlattr *tca[TCA_MAX + 1];
2214 char name[IFNAMSIZ];
2220 struct Qdisc *q = NULL;
2221 struct tcf_chain_info chain_info;
2222 struct tcf_chain *chain = NULL;
2223 struct tcf_block *block = NULL;
2224 struct tcf_proto *tp = NULL;
2225 unsigned long cl = 0;
2228 bool rtnl_held = false;
2230 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2233 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2234 rtm_tca_policy, extack);
2239 protocol = TC_H_MIN(t->tcm_info);
2240 prio = TC_H_MAJ(t->tcm_info);
2241 parent = t->tcm_parent;
2243 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2244 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2248 /* Find head of filter chain. */
2250 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2254 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2255 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2259 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2260 * found), qdisc is not unlocked, classifier type is not specified,
2261 * classifier is not unlocked.
2264 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2265 !tcf_proto_is_unlocked(name)) {
2270 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2274 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2276 if (IS_ERR(block)) {
2277 err = PTR_ERR(block);
2281 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2282 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2283 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2287 chain = tcf_chain_get(block, chain_index, false);
2289 /* User requested flush on non-existent chain. Nothing to do,
2290 * so just return success.
2296 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2302 tfilter_notify_chain(net, skb, block, q, parent, n,
2303 chain, RTM_DELTFILTER, rtnl_held);
2304 tcf_chain_flush(chain, rtnl_held);
2309 mutex_lock(&chain->filter_chain_lock);
2310 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2312 if (!tp || IS_ERR(tp)) {
2313 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2314 err = tp ? PTR_ERR(tp) : -ENOENT;
2316 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2317 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2320 } else if (t->tcm_handle == 0) {
2321 tcf_proto_signal_destroying(chain, tp);
2322 tcf_chain_tp_remove(chain, &chain_info, tp);
2323 mutex_unlock(&chain->filter_chain_lock);
2325 tcf_proto_put(tp, rtnl_held, NULL);
2326 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2327 RTM_DELTFILTER, false, rtnl_held);
2331 mutex_unlock(&chain->filter_chain_lock);
2333 fh = tp->ops->get(tp, t->tcm_handle);
2336 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2341 err = tfilter_del_notify(net, skb, n, tp, block,
2342 q, parent, fh, false, &last,
2348 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2353 if (tp && !IS_ERR(tp))
2354 tcf_proto_put(tp, rtnl_held, NULL);
2355 tcf_chain_put(chain);
2357 tcf_block_release(q, block, rtnl_held);
2365 mutex_unlock(&chain->filter_chain_lock);
2369 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2370 struct netlink_ext_ack *extack)
2372 struct net *net = sock_net(skb->sk);
2373 struct nlattr *tca[TCA_MAX + 1];
2374 char name[IFNAMSIZ];
2380 struct Qdisc *q = NULL;
2381 struct tcf_chain_info chain_info;
2382 struct tcf_chain *chain = NULL;
2383 struct tcf_block *block = NULL;
2384 struct tcf_proto *tp = NULL;
2385 unsigned long cl = 0;
2388 bool rtnl_held = false;
2390 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2391 rtm_tca_policy, extack);
2396 protocol = TC_H_MIN(t->tcm_info);
2397 prio = TC_H_MAJ(t->tcm_info);
2398 parent = t->tcm_parent;
2401 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2405 /* Find head of filter chain. */
2407 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2411 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2412 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2416 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2417 * unlocked, classifier type is not specified, classifier is not
2420 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2421 !tcf_proto_is_unlocked(name)) {
2426 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2430 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2432 if (IS_ERR(block)) {
2433 err = PTR_ERR(block);
2437 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2438 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2439 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2443 chain = tcf_chain_get(block, chain_index, false);
2445 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2450 mutex_lock(&chain->filter_chain_lock);
2451 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2453 mutex_unlock(&chain->filter_chain_lock);
2454 if (!tp || IS_ERR(tp)) {
2455 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2456 err = tp ? PTR_ERR(tp) : -ENOENT;
2458 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2459 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2464 fh = tp->ops->get(tp, t->tcm_handle);
2467 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2470 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2471 fh, RTM_NEWTFILTER, true, rtnl_held);
2473 NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2476 tfilter_put(tp, fh);
2479 if (tp && !IS_ERR(tp))
2480 tcf_proto_put(tp, rtnl_held, NULL);
2481 tcf_chain_put(chain);
2483 tcf_block_release(q, block, rtnl_held);
2491 struct tcf_dump_args {
2492 struct tcf_walker w;
2493 struct sk_buff *skb;
2494 struct netlink_callback *cb;
2495 struct tcf_block *block;
2500 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2502 struct tcf_dump_args *a = (void *)arg;
2503 struct net *net = sock_net(a->skb->sk);
2505 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2506 n, NETLINK_CB(a->cb->skb).portid,
2507 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2508 RTM_NEWTFILTER, true);
2511 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2512 struct sk_buff *skb, struct netlink_callback *cb,
2513 long index_start, long *p_index)
2515 struct net *net = sock_net(skb->sk);
2516 struct tcf_block *block = chain->block;
2517 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2518 struct tcf_proto *tp, *tp_prev;
2519 struct tcf_dump_args arg;
2521 for (tp = __tcf_get_next_proto(chain, NULL);
2524 tp = __tcf_get_next_proto(chain, tp),
2525 tcf_proto_put(tp_prev, true, NULL),
2527 if (*p_index < index_start)
2529 if (TC_H_MAJ(tcm->tcm_info) &&
2530 TC_H_MAJ(tcm->tcm_info) != tp->prio)
2532 if (TC_H_MIN(tcm->tcm_info) &&
2533 TC_H_MIN(tcm->tcm_info) != tp->protocol)
2535 if (*p_index > index_start)
2536 memset(&cb->args[1], 0,
2537 sizeof(cb->args) - sizeof(cb->args[0]));
2538 if (cb->args[1] == 0) {
2539 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2540 NETLINK_CB(cb->skb).portid,
2541 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2542 RTM_NEWTFILTER, true) <= 0)
2548 arg.w.fn = tcf_node_dump;
2553 arg.parent = parent;
2555 arg.w.skip = cb->args[1] - 1;
2557 arg.w.cookie = cb->args[2];
2558 tp->ops->walk(tp, &arg.w, true);
2559 cb->args[2] = arg.w.cookie;
2560 cb->args[1] = arg.w.count + 1;
2567 tcf_proto_put(tp, true, NULL);
2571 /* called with RTNL */
2572 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2574 struct tcf_chain *chain, *chain_prev;
2575 struct net *net = sock_net(skb->sk);
2576 struct nlattr *tca[TCA_MAX + 1];
2577 struct Qdisc *q = NULL;
2578 struct tcf_block *block;
2579 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2585 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2588 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2593 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2594 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2597 /* If we work with block index, q is NULL and parent value
2598 * will never be used in the following code. The check
2599 * in tcf_fill_node prevents it. However, compiler does not
2600 * see that far, so set parent to zero to silence the warning
2601 * about parent being uninitialized.
2605 const struct Qdisc_class_ops *cops;
2606 struct net_device *dev;
2607 unsigned long cl = 0;
2609 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2613 parent = tcm->tcm_parent;
2618 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2622 cops = q->ops->cl_ops;
2625 if (!cops->tcf_block)
2627 if (TC_H_MIN(tcm->tcm_parent)) {
2628 cl = cops->find(q, tcm->tcm_parent);
2632 block = cops->tcf_block(q, cl, NULL);
2635 if (tcf_block_shared(block))
2639 index_start = cb->args[0];
2642 for (chain = __tcf_get_next_chain(block, NULL);
2645 chain = __tcf_get_next_chain(block, chain),
2646 tcf_chain_put(chain_prev)) {
2647 if (tca[TCA_CHAIN] &&
2648 nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2650 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2651 index_start, &index)) {
2652 tcf_chain_put(chain);
2658 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2659 tcf_block_refcnt_put(block, true);
2660 cb->args[0] = index;
2663 /* If we did no progress, the error (EMSGSIZE) is real */
2664 if (skb->len == 0 && err)
2669 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2670 void *tmplt_priv, u32 chain_index,
2671 struct net *net, struct sk_buff *skb,
2672 struct tcf_block *block,
2673 u32 portid, u32 seq, u16 flags, int event)
2675 unsigned char *b = skb_tail_pointer(skb);
2676 const struct tcf_proto_ops *ops;
2677 struct nlmsghdr *nlh;
2684 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2686 goto out_nlmsg_trim;
2687 tcm = nlmsg_data(nlh);
2688 tcm->tcm_family = AF_UNSPEC;
2691 tcm->tcm_handle = 0;
2693 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2694 tcm->tcm_parent = block->q->handle;
2696 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2697 tcm->tcm_block_index = block->index;
2700 if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2701 goto nla_put_failure;
2704 if (nla_put_string(skb, TCA_KIND, ops->kind))
2705 goto nla_put_failure;
2706 if (ops->tmplt_dump(skb, net, priv) < 0)
2707 goto nla_put_failure;
2710 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2719 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2720 u32 seq, u16 flags, int event, bool unicast)
2722 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2723 struct tcf_block *block = chain->block;
2724 struct net *net = block->net;
2725 struct sk_buff *skb;
2728 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2732 if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2733 chain->index, net, skb, block, portid,
2734 seq, flags, event) <= 0) {
2740 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2742 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2743 flags & NLM_F_ECHO);
2750 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2751 void *tmplt_priv, u32 chain_index,
2752 struct tcf_block *block, struct sk_buff *oskb,
2753 u32 seq, u16 flags, bool unicast)
2755 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2756 struct net *net = block->net;
2757 struct sk_buff *skb;
2759 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2763 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2764 block, portid, seq, flags, RTM_DELCHAIN) <= 0) {
2770 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2772 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2775 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2776 struct nlattr **tca,
2777 struct netlink_ext_ack *extack)
2779 const struct tcf_proto_ops *ops;
2780 char name[IFNAMSIZ];
2783 /* If kind is not set, user did not specify template. */
2787 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2788 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
2792 ops = tcf_proto_lookup_ops(name, true, extack);
2794 return PTR_ERR(ops);
2795 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2796 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2800 tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2801 if (IS_ERR(tmplt_priv)) {
2802 module_put(ops->owner);
2803 return PTR_ERR(tmplt_priv);
2805 chain->tmplt_ops = ops;
2806 chain->tmplt_priv = tmplt_priv;
2810 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2813 /* If template ops are set, no work to do for us. */
2817 tmplt_ops->tmplt_destroy(tmplt_priv);
2818 module_put(tmplt_ops->owner);
2821 /* Add/delete/get a chain */
2823 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2824 struct netlink_ext_ack *extack)
2826 struct net *net = sock_net(skb->sk);
2827 struct nlattr *tca[TCA_MAX + 1];
2831 struct Qdisc *q = NULL;
2832 struct tcf_chain *chain = NULL;
2833 struct tcf_block *block;
2837 if (n->nlmsg_type != RTM_GETCHAIN &&
2838 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2842 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2843 rtm_tca_policy, extack);
2848 parent = t->tcm_parent;
2851 block = tcf_block_find(net, &q, &parent, &cl,
2852 t->tcm_ifindex, t->tcm_block_index, extack);
2854 return PTR_ERR(block);
2856 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2857 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2858 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2863 mutex_lock(&block->lock);
2864 chain = tcf_chain_lookup(block, chain_index);
2865 if (n->nlmsg_type == RTM_NEWCHAIN) {
2867 if (tcf_chain_held_by_acts_only(chain)) {
2868 /* The chain exists only because there is
2869 * some action referencing it.
2871 tcf_chain_hold(chain);
2873 NL_SET_ERR_MSG(extack, "Filter chain already exists");
2875 goto errout_block_locked;
2878 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2879 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
2881 goto errout_block_locked;
2883 chain = tcf_chain_create(block, chain_index);
2885 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
2887 goto errout_block_locked;
2891 if (!chain || tcf_chain_held_by_acts_only(chain)) {
2892 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2894 goto errout_block_locked;
2896 tcf_chain_hold(chain);
2899 if (n->nlmsg_type == RTM_NEWCHAIN) {
2900 /* Modifying chain requires holding parent block lock. In case
2901 * the chain was successfully added, take a reference to the
2902 * chain. This ensures that an empty chain does not disappear at
2903 * the end of this function.
2905 tcf_chain_hold(chain);
2906 chain->explicitly_created = true;
2908 mutex_unlock(&block->lock);
2910 switch (n->nlmsg_type) {
2912 err = tc_chain_tmplt_add(chain, net, tca, extack);
2914 tcf_chain_put_explicitly_created(chain);
2918 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2919 RTM_NEWCHAIN, false);
2922 tfilter_notify_chain(net, skb, block, q, parent, n,
2923 chain, RTM_DELTFILTER, true);
2924 /* Flush the chain first as the user requested chain removal. */
2925 tcf_chain_flush(chain, true);
2926 /* In case the chain was successfully deleted, put a reference
2927 * to the chain previously taken during addition.
2929 tcf_chain_put_explicitly_created(chain);
2932 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
2933 n->nlmsg_seq, n->nlmsg_type, true);
2935 NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2939 NL_SET_ERR_MSG(extack, "Unsupported message type");
2944 tcf_chain_put(chain);
2946 tcf_block_release(q, block, true);
2948 /* Replay the request. */
2952 errout_block_locked:
2953 mutex_unlock(&block->lock);
2957 /* called with RTNL */
2958 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2960 struct net *net = sock_net(skb->sk);
2961 struct nlattr *tca[TCA_MAX + 1];
2962 struct Qdisc *q = NULL;
2963 struct tcf_block *block;
2964 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2965 struct tcf_chain *chain;
2971 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2974 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2975 rtm_tca_policy, cb->extack);
2979 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2980 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2983 /* If we work with block index, q is NULL and parent value
2984 * will never be used in the following code. The check
2985 * in tcf_fill_node prevents it. However, compiler does not
2986 * see that far, so set parent to zero to silence the warning
2987 * about parent being uninitialized.
2991 const struct Qdisc_class_ops *cops;
2992 struct net_device *dev;
2993 unsigned long cl = 0;
2995 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2999 parent = tcm->tcm_parent;
3004 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
3008 cops = q->ops->cl_ops;
3011 if (!cops->tcf_block)
3013 if (TC_H_MIN(tcm->tcm_parent)) {
3014 cl = cops->find(q, tcm->tcm_parent);
3018 block = cops->tcf_block(q, cl, NULL);
3021 if (tcf_block_shared(block))
3025 index_start = cb->args[0];
3028 mutex_lock(&block->lock);
3029 list_for_each_entry(chain, &block->chain_list, list) {
3030 if ((tca[TCA_CHAIN] &&
3031 nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3033 if (index < index_start) {
3037 if (tcf_chain_held_by_acts_only(chain))
3039 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3040 chain->index, net, skb, block,
3041 NETLINK_CB(cb->skb).portid,
3042 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3048 mutex_unlock(&block->lock);
3050 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3051 tcf_block_refcnt_put(block, true);
3052 cb->args[0] = index;
3055 /* If we did no progress, the error (EMSGSIZE) is real */
3056 if (skb->len == 0 && err)
3061 void tcf_exts_destroy(struct tcf_exts *exts)
3063 #ifdef CONFIG_NET_CLS_ACT
3064 if (exts->actions) {
3065 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3066 kfree(exts->actions);
3068 exts->nr_actions = 0;
3071 EXPORT_SYMBOL(tcf_exts_destroy);
3073 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3074 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
3075 bool rtnl_held, struct netlink_ext_ack *extack)
3077 #ifdef CONFIG_NET_CLS_ACT
3079 struct tc_action *act;
3080 size_t attr_size = 0;
3082 if (exts->police && tb[exts->police]) {
3083 act = tcf_action_init_1(net, tp, tb[exts->police],
3084 rate_tlv, "police", ovr,
3085 TCA_ACT_BIND, rtnl_held,
3088 return PTR_ERR(act);
3090 act->type = exts->type = TCA_OLD_COMPAT;
3091 exts->actions[0] = act;
3092 exts->nr_actions = 1;
3093 } else if (exts->action && tb[exts->action]) {
3096 err = tcf_action_init(net, tp, tb[exts->action],
3097 rate_tlv, NULL, ovr, TCA_ACT_BIND,
3098 exts->actions, &attr_size,
3102 exts->nr_actions = err;
3106 if ((exts->action && tb[exts->action]) ||
3107 (exts->police && tb[exts->police])) {
3108 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3115 EXPORT_SYMBOL(tcf_exts_validate);
3117 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3119 #ifdef CONFIG_NET_CLS_ACT
3120 struct tcf_exts old = *dst;
3123 tcf_exts_destroy(&old);
3126 EXPORT_SYMBOL(tcf_exts_change);
3128 #ifdef CONFIG_NET_CLS_ACT
3129 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3131 if (exts->nr_actions == 0)
3134 return exts->actions[0];
3138 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3140 #ifdef CONFIG_NET_CLS_ACT
3141 struct nlattr *nest;
3143 if (exts->action && tcf_exts_has_actions(exts)) {
3145 * again for backward compatible mode - we want
3146 * to work with both old and new modes of entering
3147 * tc data even if iproute2 was newer - jhs
3149 if (exts->type != TCA_OLD_COMPAT) {
3150 nest = nla_nest_start_noflag(skb, exts->action);
3152 goto nla_put_failure;
3154 if (tcf_action_dump(skb, exts->actions, 0, 0) < 0)
3155 goto nla_put_failure;
3156 nla_nest_end(skb, nest);
3157 } else if (exts->police) {
3158 struct tc_action *act = tcf_exts_first_act(exts);
3159 nest = nla_nest_start_noflag(skb, exts->police);
3160 if (nest == NULL || !act)
3161 goto nla_put_failure;
3162 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3163 goto nla_put_failure;
3164 nla_nest_end(skb, nest);
3170 nla_nest_cancel(skb, nest);
3176 EXPORT_SYMBOL(tcf_exts_dump);
3179 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3181 #ifdef CONFIG_NET_CLS_ACT
3182 struct tc_action *a = tcf_exts_first_act(exts);
3183 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3188 EXPORT_SYMBOL(tcf_exts_dump_stats);
3190 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3192 if (*flags & TCA_CLS_FLAGS_IN_HW)
3194 *flags |= TCA_CLS_FLAGS_IN_HW;
3195 atomic_inc(&block->offloadcnt);
3198 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3200 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3202 *flags &= ~TCA_CLS_FLAGS_IN_HW;
3203 atomic_dec(&block->offloadcnt);
3206 static void tc_cls_offload_cnt_update(struct tcf_block *block,
3207 struct tcf_proto *tp, u32 *cnt,
3208 u32 *flags, u32 diff, bool add)
3210 lockdep_assert_held(&block->cb_lock);
3212 spin_lock(&tp->lock);
3215 tcf_block_offload_inc(block, flags);
3220 tcf_block_offload_dec(block, flags);
3222 spin_unlock(&tp->lock);
3226 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3227 u32 *cnt, u32 *flags)
3229 lockdep_assert_held(&block->cb_lock);
3231 spin_lock(&tp->lock);
3232 tcf_block_offload_dec(block, flags);
3234 spin_unlock(&tp->lock);
3238 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3239 void *type_data, bool err_stop)
3241 struct flow_block_cb *block_cb;
3245 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3246 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3257 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3258 void *type_data, bool err_stop, bool rtnl_held)
3260 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3266 down_read(&block->cb_lock);
3267 /* Need to obtain rtnl lock if block is bound to devs that require it.
3268 * In block bind code cb_lock is obtained while holding rtnl, so we must
3269 * obtain the locks in same order here.
3271 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3272 up_read(&block->cb_lock);
3277 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3279 up_read(&block->cb_lock);
3284 EXPORT_SYMBOL(tc_setup_cb_call);
3286 /* Non-destructive filter add. If filter that wasn't already in hardware is
3287 * successfully offloaded, increment block offloads counter. On failure,
3288 * previously offloaded filter is considered to be intact and offloads counter
3289 * is not decremented.
3292 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3293 enum tc_setup_type type, void *type_data, bool err_stop,
3294 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3296 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3302 down_read(&block->cb_lock);
3303 /* Need to obtain rtnl lock if block is bound to devs that require it.
3304 * In block bind code cb_lock is obtained while holding rtnl, so we must
3305 * obtain the locks in same order here.
3307 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3308 up_read(&block->cb_lock);
3313 /* Make sure all netdevs sharing this block are offload-capable. */
3314 if (block->nooffloaddevcnt && err_stop) {
3315 ok_count = -EOPNOTSUPP;
3319 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3323 if (tp->ops->hw_add)
3324 tp->ops->hw_add(tp, type_data);
3326 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3329 up_read(&block->cb_lock);
3332 return ok_count < 0 ? ok_count : 0;
3334 EXPORT_SYMBOL(tc_setup_cb_add);
3336 /* Destructive filter replace. If filter that wasn't already in hardware is
3337 * successfully offloaded, increment block offload counter. On failure,
3338 * previously offloaded filter is considered to be destroyed and offload counter
3342 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3343 enum tc_setup_type type, void *type_data, bool err_stop,
3344 u32 *old_flags, unsigned int *old_in_hw_count,
3345 u32 *new_flags, unsigned int *new_in_hw_count,
3348 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3354 down_read(&block->cb_lock);
3355 /* Need to obtain rtnl lock if block is bound to devs that require it.
3356 * In block bind code cb_lock is obtained while holding rtnl, so we must
3357 * obtain the locks in same order here.
3359 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3360 up_read(&block->cb_lock);
3365 /* Make sure all netdevs sharing this block are offload-capable. */
3366 if (block->nooffloaddevcnt && err_stop) {
3367 ok_count = -EOPNOTSUPP;
3371 tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3372 if (tp->ops->hw_del)
3373 tp->ops->hw_del(tp, type_data);
3375 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3379 if (tp->ops->hw_add)
3380 tp->ops->hw_add(tp, type_data);
3382 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3383 new_flags, ok_count, true);
3385 up_read(&block->cb_lock);
3388 return ok_count < 0 ? ok_count : 0;
3390 EXPORT_SYMBOL(tc_setup_cb_replace);
3392 /* Destroy filter and decrement block offload counter, if filter was previously
3396 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3397 enum tc_setup_type type, void *type_data, bool err_stop,
3398 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3400 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3406 down_read(&block->cb_lock);
3407 /* Need to obtain rtnl lock if block is bound to devs that require it.
3408 * In block bind code cb_lock is obtained while holding rtnl, so we must
3409 * obtain the locks in same order here.
3411 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3412 up_read(&block->cb_lock);
3417 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3419 tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3420 if (tp->ops->hw_del)
3421 tp->ops->hw_del(tp, type_data);
3423 up_read(&block->cb_lock);
3426 return ok_count < 0 ? ok_count : 0;
3428 EXPORT_SYMBOL(tc_setup_cb_destroy);
3430 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3431 bool add, flow_setup_cb_t *cb,
3432 enum tc_setup_type type, void *type_data,
3433 void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3435 int err = cb(type, type_data, cb_priv);
3438 if (add && tc_skip_sw(*flags))
3441 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3447 EXPORT_SYMBOL(tc_setup_cb_reoffload);
3449 static int tcf_act_get_cookie(struct flow_action_entry *entry,
3450 const struct tc_action *act)
3452 struct tc_cookie *cookie;
3456 cookie = rcu_dereference(act->act_cookie);
3458 entry->cookie = flow_action_cookie_create(cookie->data,
3468 static void tcf_act_put_cookie(struct flow_action_entry *entry)
3470 flow_action_cookie_destroy(entry->cookie);
3473 void tc_cleanup_flow_action(struct flow_action *flow_action)
3475 struct flow_action_entry *entry;
3478 flow_action_for_each(i, entry, flow_action) {
3479 tcf_act_put_cookie(entry);
3480 if (entry->destructor)
3481 entry->destructor(entry->destructor_priv);
3484 EXPORT_SYMBOL(tc_cleanup_flow_action);
3486 static void tcf_mirred_get_dev(struct flow_action_entry *entry,
3487 const struct tc_action *act)
3489 #ifdef CONFIG_NET_CLS_ACT
3490 entry->dev = act->ops->get_dev(act, &entry->destructor);
3493 entry->destructor_priv = entry->dev;
3497 static void tcf_tunnel_encap_put_tunnel(void *priv)
3499 struct ip_tunnel_info *tunnel = priv;
3504 static int tcf_tunnel_encap_get_tunnel(struct flow_action_entry *entry,
3505 const struct tc_action *act)
3507 entry->tunnel = tcf_tunnel_info_copy(act);
3510 entry->destructor = tcf_tunnel_encap_put_tunnel;
3511 entry->destructor_priv = entry->tunnel;
3515 static void tcf_sample_get_group(struct flow_action_entry *entry,
3516 const struct tc_action *act)
3518 #ifdef CONFIG_NET_CLS_ACT
3519 entry->sample.psample_group =
3520 act->ops->get_psample_group(act, &entry->destructor);
3521 entry->destructor_priv = entry->sample.psample_group;
3525 int tc_setup_flow_action(struct flow_action *flow_action,
3526 const struct tcf_exts *exts)
3528 struct tc_action *act;
3529 int i, j, k, err = 0;
3531 BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3532 BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3533 BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3539 tcf_exts_for_each_action(i, act, exts) {
3540 struct flow_action_entry *entry;
3542 entry = &flow_action->entries[j];
3543 spin_lock_bh(&act->tcfa_lock);
3544 err = tcf_act_get_cookie(entry, act);
3546 goto err_out_locked;
3548 entry->hw_stats = act->hw_stats;
3550 if (is_tcf_gact_ok(act)) {
3551 entry->id = FLOW_ACTION_ACCEPT;
3552 } else if (is_tcf_gact_shot(act)) {
3553 entry->id = FLOW_ACTION_DROP;
3554 } else if (is_tcf_gact_trap(act)) {
3555 entry->id = FLOW_ACTION_TRAP;
3556 } else if (is_tcf_gact_goto_chain(act)) {
3557 entry->id = FLOW_ACTION_GOTO;
3558 entry->chain_index = tcf_gact_goto_chain_index(act);
3559 } else if (is_tcf_mirred_egress_redirect(act)) {
3560 entry->id = FLOW_ACTION_REDIRECT;
3561 tcf_mirred_get_dev(entry, act);
3562 } else if (is_tcf_mirred_egress_mirror(act)) {
3563 entry->id = FLOW_ACTION_MIRRED;
3564 tcf_mirred_get_dev(entry, act);
3565 } else if (is_tcf_mirred_ingress_redirect(act)) {
3566 entry->id = FLOW_ACTION_REDIRECT_INGRESS;
3567 tcf_mirred_get_dev(entry, act);
3568 } else if (is_tcf_mirred_ingress_mirror(act)) {
3569 entry->id = FLOW_ACTION_MIRRED_INGRESS;
3570 tcf_mirred_get_dev(entry, act);
3571 } else if (is_tcf_vlan(act)) {
3572 switch (tcf_vlan_action(act)) {
3573 case TCA_VLAN_ACT_PUSH:
3574 entry->id = FLOW_ACTION_VLAN_PUSH;
3575 entry->vlan.vid = tcf_vlan_push_vid(act);
3576 entry->vlan.proto = tcf_vlan_push_proto(act);
3577 entry->vlan.prio = tcf_vlan_push_prio(act);
3579 case TCA_VLAN_ACT_POP:
3580 entry->id = FLOW_ACTION_VLAN_POP;
3582 case TCA_VLAN_ACT_MODIFY:
3583 entry->id = FLOW_ACTION_VLAN_MANGLE;
3584 entry->vlan.vid = tcf_vlan_push_vid(act);
3585 entry->vlan.proto = tcf_vlan_push_proto(act);
3586 entry->vlan.prio = tcf_vlan_push_prio(act);
3590 goto err_out_locked;
3592 } else if (is_tcf_tunnel_set(act)) {
3593 entry->id = FLOW_ACTION_TUNNEL_ENCAP;
3594 err = tcf_tunnel_encap_get_tunnel(entry, act);
3596 goto err_out_locked;
3597 } else if (is_tcf_tunnel_release(act)) {
3598 entry->id = FLOW_ACTION_TUNNEL_DECAP;
3599 } else if (is_tcf_pedit(act)) {
3600 for (k = 0; k < tcf_pedit_nkeys(act); k++) {
3601 switch (tcf_pedit_cmd(act, k)) {
3602 case TCA_PEDIT_KEY_EX_CMD_SET:
3603 entry->id = FLOW_ACTION_MANGLE;
3605 case TCA_PEDIT_KEY_EX_CMD_ADD:
3606 entry->id = FLOW_ACTION_ADD;
3610 goto err_out_locked;
3612 entry->mangle.htype = tcf_pedit_htype(act, k);
3613 entry->mangle.mask = tcf_pedit_mask(act, k);
3614 entry->mangle.val = tcf_pedit_val(act, k);
3615 entry->mangle.offset = tcf_pedit_offset(act, k);
3616 entry->hw_stats = act->hw_stats;
3617 entry = &flow_action->entries[++j];
3619 } else if (is_tcf_csum(act)) {
3620 entry->id = FLOW_ACTION_CSUM;
3621 entry->csum_flags = tcf_csum_update_flags(act);
3622 } else if (is_tcf_skbedit_mark(act)) {
3623 entry->id = FLOW_ACTION_MARK;
3624 entry->mark = tcf_skbedit_mark(act);
3625 } else if (is_tcf_sample(act)) {
3626 entry->id = FLOW_ACTION_SAMPLE;
3627 entry->sample.trunc_size = tcf_sample_trunc_size(act);
3628 entry->sample.truncate = tcf_sample_truncate(act);
3629 entry->sample.rate = tcf_sample_rate(act);
3630 tcf_sample_get_group(entry, act);
3631 } else if (is_tcf_police(act)) {
3632 entry->id = FLOW_ACTION_POLICE;
3633 entry->police.burst = tcf_police_tcfp_burst(act);
3634 entry->police.rate_bytes_ps =
3635 tcf_police_rate_bytes_ps(act);
3636 } else if (is_tcf_ct(act)) {
3637 entry->id = FLOW_ACTION_CT;
3638 entry->ct.action = tcf_ct_action(act);
3639 entry->ct.zone = tcf_ct_zone(act);
3640 entry->ct.flow_table = tcf_ct_ft(act);
3641 } else if (is_tcf_mpls(act)) {
3642 switch (tcf_mpls_action(act)) {
3643 case TCA_MPLS_ACT_PUSH:
3644 entry->id = FLOW_ACTION_MPLS_PUSH;
3645 entry->mpls_push.proto = tcf_mpls_proto(act);
3646 entry->mpls_push.label = tcf_mpls_label(act);
3647 entry->mpls_push.tc = tcf_mpls_tc(act);
3648 entry->mpls_push.bos = tcf_mpls_bos(act);
3649 entry->mpls_push.ttl = tcf_mpls_ttl(act);
3651 case TCA_MPLS_ACT_POP:
3652 entry->id = FLOW_ACTION_MPLS_POP;
3653 entry->mpls_pop.proto = tcf_mpls_proto(act);
3655 case TCA_MPLS_ACT_MODIFY:
3656 entry->id = FLOW_ACTION_MPLS_MANGLE;
3657 entry->mpls_mangle.label = tcf_mpls_label(act);
3658 entry->mpls_mangle.tc = tcf_mpls_tc(act);
3659 entry->mpls_mangle.bos = tcf_mpls_bos(act);
3660 entry->mpls_mangle.ttl = tcf_mpls_ttl(act);
3663 goto err_out_locked;
3665 } else if (is_tcf_skbedit_ptype(act)) {
3666 entry->id = FLOW_ACTION_PTYPE;
3667 entry->ptype = tcf_skbedit_ptype(act);
3668 } else if (is_tcf_skbedit_priority(act)) {
3669 entry->id = FLOW_ACTION_PRIORITY;
3670 entry->priority = tcf_skbedit_priority(act);
3673 goto err_out_locked;
3675 spin_unlock_bh(&act->tcfa_lock);
3677 if (!is_tcf_pedit(act))
3683 tc_cleanup_flow_action(flow_action);
3687 spin_unlock_bh(&act->tcfa_lock);
3690 EXPORT_SYMBOL(tc_setup_flow_action);
3692 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3694 unsigned int num_acts = 0;
3695 struct tc_action *act;
3698 tcf_exts_for_each_action(i, act, exts) {
3699 if (is_tcf_pedit(act))
3700 num_acts += tcf_pedit_nkeys(act);
3706 EXPORT_SYMBOL(tcf_exts_num_actions);
3708 static __net_init int tcf_net_init(struct net *net)
3710 struct tcf_net *tn = net_generic(net, tcf_net_id);
3712 spin_lock_init(&tn->idr_lock);
3717 static void __net_exit tcf_net_exit(struct net *net)
3719 struct tcf_net *tn = net_generic(net, tcf_net_id);
3721 idr_destroy(&tn->idr);
3724 static struct pernet_operations tcf_net_ops = {
3725 .init = tcf_net_init,
3726 .exit = tcf_net_exit,
3728 .size = sizeof(struct tcf_net),
3731 static struct flow_indr_block_entry block_entry = {
3732 .cb = tc_indr_block_get_and_cmd,
3733 .list = LIST_HEAD_INIT(block_entry.list),
3736 static int __init tc_filter_init(void)
3740 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3744 err = register_pernet_subsys(&tcf_net_ops);
3746 goto err_register_pernet_subsys;
3748 flow_indr_add_block_cb(&block_entry);
3750 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3751 RTNL_FLAG_DOIT_UNLOCKED);
3752 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3753 RTNL_FLAG_DOIT_UNLOCKED);
3754 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3755 tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3756 rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3757 rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3758 rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3763 err_register_pernet_subsys:
3764 destroy_workqueue(tc_filter_wq);
3768 subsys_initcall(tc_filter_init);