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 <net/net_namespace.h>
27 #include <net/netlink.h>
28 #include <net/pkt_sched.h>
29 #include <net/pkt_cls.h>
30 #include <net/tc_act/tc_pedit.h>
31 #include <net/tc_act/tc_mirred.h>
32 #include <net/tc_act/tc_vlan.h>
33 #include <net/tc_act/tc_tunnel_key.h>
34 #include <net/tc_act/tc_csum.h>
35 #include <net/tc_act/tc_gact.h>
36 #include <net/tc_act/tc_police.h>
37 #include <net/tc_act/tc_sample.h>
38 #include <net/tc_act/tc_skbedit.h>
39 #include <net/tc_act/tc_ct.h>
40 #include <net/tc_act/tc_mpls.h>
41 #include <net/flow_offload.h>
43 extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
45 /* The list of all installed classifier types */
46 static LIST_HEAD(tcf_proto_base);
48 /* Protects list of registered TC modules. It is pure SMP lock. */
49 static DEFINE_RWLOCK(cls_mod_lock);
51 static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
53 return jhash_3words(tp->chain->index, tp->prio,
54 (__force __u32)tp->protocol, 0);
57 static void tcf_proto_signal_destroying(struct tcf_chain *chain,
60 struct tcf_block *block = chain->block;
62 mutex_lock(&block->proto_destroy_lock);
63 hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
64 destroy_obj_hashfn(tp));
65 mutex_unlock(&block->proto_destroy_lock);
68 static bool tcf_proto_cmp(const struct tcf_proto *tp1,
69 const struct tcf_proto *tp2)
71 return tp1->chain->index == tp2->chain->index &&
72 tp1->prio == tp2->prio &&
73 tp1->protocol == tp2->protocol;
76 static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
79 u32 hash = destroy_obj_hashfn(tp);
80 struct tcf_proto *iter;
84 hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
85 destroy_ht_node, hash) {
86 if (tcf_proto_cmp(tp, iter)) {
97 tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
99 struct tcf_block *block = chain->block;
101 mutex_lock(&block->proto_destroy_lock);
102 if (hash_hashed(&tp->destroy_ht_node))
103 hash_del_rcu(&tp->destroy_ht_node);
104 mutex_unlock(&block->proto_destroy_lock);
107 /* Find classifier type by string name */
109 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
111 const struct tcf_proto_ops *t, *res = NULL;
114 read_lock(&cls_mod_lock);
115 list_for_each_entry(t, &tcf_proto_base, head) {
116 if (strcmp(kind, t->kind) == 0) {
117 if (try_module_get(t->owner))
122 read_unlock(&cls_mod_lock);
127 static const struct tcf_proto_ops *
128 tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
129 struct netlink_ext_ack *extack)
131 const struct tcf_proto_ops *ops;
133 ops = __tcf_proto_lookup_ops(kind);
136 #ifdef CONFIG_MODULES
139 request_module("cls_%s", kind);
142 ops = __tcf_proto_lookup_ops(kind);
143 /* We dropped the RTNL semaphore in order to perform
144 * the module load. So, even if we succeeded in loading
145 * the module we have to replay the request. We indicate
146 * this using -EAGAIN.
149 module_put(ops->owner);
150 return ERR_PTR(-EAGAIN);
153 NL_SET_ERR_MSG(extack, "TC classifier not found");
154 return ERR_PTR(-ENOENT);
157 /* Register(unregister) new classifier type */
159 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
161 struct tcf_proto_ops *t;
164 write_lock(&cls_mod_lock);
165 list_for_each_entry(t, &tcf_proto_base, head)
166 if (!strcmp(ops->kind, t->kind))
169 list_add_tail(&ops->head, &tcf_proto_base);
172 write_unlock(&cls_mod_lock);
175 EXPORT_SYMBOL(register_tcf_proto_ops);
177 static struct workqueue_struct *tc_filter_wq;
179 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
181 struct tcf_proto_ops *t;
184 /* Wait for outstanding call_rcu()s, if any, from a
185 * tcf_proto_ops's destroy() handler.
188 flush_workqueue(tc_filter_wq);
190 write_lock(&cls_mod_lock);
191 list_for_each_entry(t, &tcf_proto_base, head) {
198 write_unlock(&cls_mod_lock);
201 EXPORT_SYMBOL(unregister_tcf_proto_ops);
203 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
205 INIT_RCU_WORK(rwork, func);
206 return queue_rcu_work(tc_filter_wq, rwork);
208 EXPORT_SYMBOL(tcf_queue_work);
210 /* Select new prio value from the range, managed by kernel. */
212 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
214 u32 first = TC_H_MAKE(0xC0000000U, 0U);
217 first = tp->prio - 1;
219 return TC_H_MAJ(first);
222 static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
225 return nla_strlcpy(name, kind, IFNAMSIZ) >= IFNAMSIZ;
226 memset(name, 0, IFNAMSIZ);
230 static bool tcf_proto_is_unlocked(const char *kind)
232 const struct tcf_proto_ops *ops;
235 if (strlen(kind) == 0)
238 ops = tcf_proto_lookup_ops(kind, false, NULL);
239 /* On error return false to take rtnl lock. Proto lookup/create
240 * functions will perform lookup again and properly handle errors.
245 ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
246 module_put(ops->owner);
250 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
251 u32 prio, struct tcf_chain *chain,
253 struct netlink_ext_ack *extack)
255 struct tcf_proto *tp;
258 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
260 return ERR_PTR(-ENOBUFS);
262 tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
263 if (IS_ERR(tp->ops)) {
264 err = PTR_ERR(tp->ops);
267 tp->classify = tp->ops->classify;
268 tp->protocol = protocol;
271 spin_lock_init(&tp->lock);
272 refcount_set(&tp->refcnt, 1);
274 err = tp->ops->init(tp);
276 module_put(tp->ops->owner);
286 static void tcf_proto_get(struct tcf_proto *tp)
288 refcount_inc(&tp->refcnt);
291 static void tcf_chain_put(struct tcf_chain *chain);
293 static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
294 bool sig_destroy, struct netlink_ext_ack *extack)
296 tp->ops->destroy(tp, rtnl_held, extack);
298 tcf_proto_signal_destroyed(tp->chain, tp);
299 tcf_chain_put(tp->chain);
300 module_put(tp->ops->owner);
304 static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
305 struct netlink_ext_ack *extack)
307 if (refcount_dec_and_test(&tp->refcnt))
308 tcf_proto_destroy(tp, rtnl_held, true, extack);
311 static int walker_check_empty(struct tcf_proto *tp, void *fh,
312 struct tcf_walker *arg)
315 arg->nonempty = true;
321 static bool tcf_proto_is_empty(struct tcf_proto *tp, bool rtnl_held)
323 struct tcf_walker walker = { .fn = walker_check_empty, };
326 tp->ops->walk(tp, &walker, rtnl_held);
327 return !walker.nonempty;
332 static bool tcf_proto_check_delete(struct tcf_proto *tp, bool rtnl_held)
334 spin_lock(&tp->lock);
335 if (tcf_proto_is_empty(tp, rtnl_held))
337 spin_unlock(&tp->lock);
341 static void tcf_proto_mark_delete(struct tcf_proto *tp)
343 spin_lock(&tp->lock);
345 spin_unlock(&tp->lock);
348 static bool tcf_proto_is_deleting(struct tcf_proto *tp)
352 spin_lock(&tp->lock);
353 deleting = tp->deleting;
354 spin_unlock(&tp->lock);
359 #define ASSERT_BLOCK_LOCKED(block) \
360 lockdep_assert_held(&(block)->lock)
362 struct tcf_filter_chain_list_item {
363 struct list_head list;
364 tcf_chain_head_change_t *chain_head_change;
365 void *chain_head_change_priv;
368 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
371 struct tcf_chain *chain;
373 ASSERT_BLOCK_LOCKED(block);
375 chain = kzalloc(sizeof(*chain), GFP_KERNEL);
378 list_add_tail(&chain->list, &block->chain_list);
379 mutex_init(&chain->filter_chain_lock);
380 chain->block = block;
381 chain->index = chain_index;
384 block->chain0.chain = chain;
388 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
389 struct tcf_proto *tp_head)
391 if (item->chain_head_change)
392 item->chain_head_change(tp_head, item->chain_head_change_priv);
395 static void tcf_chain0_head_change(struct tcf_chain *chain,
396 struct tcf_proto *tp_head)
398 struct tcf_filter_chain_list_item *item;
399 struct tcf_block *block = chain->block;
404 mutex_lock(&block->lock);
405 list_for_each_entry(item, &block->chain0.filter_chain_list, list)
406 tcf_chain_head_change_item(item, tp_head);
407 mutex_unlock(&block->lock);
410 /* Returns true if block can be safely freed. */
412 static bool tcf_chain_detach(struct tcf_chain *chain)
414 struct tcf_block *block = chain->block;
416 ASSERT_BLOCK_LOCKED(block);
418 list_del(&chain->list);
420 block->chain0.chain = NULL;
422 if (list_empty(&block->chain_list) &&
423 refcount_read(&block->refcnt) == 0)
429 static void tcf_block_destroy(struct tcf_block *block)
431 mutex_destroy(&block->lock);
432 mutex_destroy(&block->proto_destroy_lock);
433 kfree_rcu(block, rcu);
436 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
438 struct tcf_block *block = chain->block;
440 mutex_destroy(&chain->filter_chain_lock);
441 kfree_rcu(chain, rcu);
443 tcf_block_destroy(block);
446 static void tcf_chain_hold(struct tcf_chain *chain)
448 ASSERT_BLOCK_LOCKED(chain->block);
453 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
455 ASSERT_BLOCK_LOCKED(chain->block);
457 /* In case all the references are action references, this
458 * chain should not be shown to the user.
460 return chain->refcnt == chain->action_refcnt;
463 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
466 struct tcf_chain *chain;
468 ASSERT_BLOCK_LOCKED(block);
470 list_for_each_entry(chain, &block->chain_list, list) {
471 if (chain->index == chain_index)
477 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
478 u32 seq, u16 flags, int event, bool unicast);
480 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
481 u32 chain_index, bool create,
484 struct tcf_chain *chain = NULL;
485 bool is_first_reference;
487 mutex_lock(&block->lock);
488 chain = tcf_chain_lookup(block, chain_index);
490 tcf_chain_hold(chain);
494 chain = tcf_chain_create(block, chain_index);
500 ++chain->action_refcnt;
501 is_first_reference = chain->refcnt - chain->action_refcnt == 1;
502 mutex_unlock(&block->lock);
504 /* Send notification only in case we got the first
505 * non-action reference. Until then, the chain acts only as
506 * a placeholder for actions pointing to it and user ought
507 * not know about them.
509 if (is_first_reference && !by_act)
510 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
511 RTM_NEWCHAIN, false);
516 mutex_unlock(&block->lock);
520 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
523 return __tcf_chain_get(block, chain_index, create, false);
526 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
528 return __tcf_chain_get(block, chain_index, true, true);
530 EXPORT_SYMBOL(tcf_chain_get_by_act);
532 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
534 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
535 void *tmplt_priv, u32 chain_index,
536 struct tcf_block *block, struct sk_buff *oskb,
537 u32 seq, u16 flags, bool unicast);
539 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
540 bool explicitly_created)
542 struct tcf_block *block = chain->block;
543 const struct tcf_proto_ops *tmplt_ops;
544 bool free_block = false;
548 mutex_lock(&block->lock);
549 if (explicitly_created) {
550 if (!chain->explicitly_created) {
551 mutex_unlock(&block->lock);
554 chain->explicitly_created = false;
558 chain->action_refcnt--;
560 /* tc_chain_notify_delete can't be called while holding block lock.
561 * However, when block is unlocked chain can be changed concurrently, so
562 * save these to temporary variables.
564 refcnt = --chain->refcnt;
565 tmplt_ops = chain->tmplt_ops;
566 tmplt_priv = chain->tmplt_priv;
568 /* The last dropped non-action reference will trigger notification. */
569 if (refcnt - chain->action_refcnt == 0 && !by_act) {
570 tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
571 block, NULL, 0, 0, false);
572 /* Last reference to chain, no need to lock. */
573 chain->flushing = false;
577 free_block = tcf_chain_detach(chain);
578 mutex_unlock(&block->lock);
581 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
582 tcf_chain_destroy(chain, free_block);
586 static void tcf_chain_put(struct tcf_chain *chain)
588 __tcf_chain_put(chain, false, false);
591 void tcf_chain_put_by_act(struct tcf_chain *chain)
593 __tcf_chain_put(chain, true, false);
595 EXPORT_SYMBOL(tcf_chain_put_by_act);
597 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
599 __tcf_chain_put(chain, false, true);
602 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
604 struct tcf_proto *tp, *tp_next;
606 mutex_lock(&chain->filter_chain_lock);
607 tp = tcf_chain_dereference(chain->filter_chain, chain);
609 tp_next = rcu_dereference_protected(tp->next, 1);
610 tcf_proto_signal_destroying(chain, tp);
613 tp = tcf_chain_dereference(chain->filter_chain, chain);
614 RCU_INIT_POINTER(chain->filter_chain, NULL);
615 tcf_chain0_head_change(chain, NULL);
616 chain->flushing = true;
617 mutex_unlock(&chain->filter_chain_lock);
620 tp_next = rcu_dereference_protected(tp->next, 1);
621 tcf_proto_put(tp, rtnl_held, NULL);
626 static int tcf_block_setup(struct tcf_block *block,
627 struct flow_block_offload *bo);
629 static void tc_indr_block_cmd(struct net_device *dev, struct tcf_block *block,
630 flow_indr_block_bind_cb_t *cb, void *cb_priv,
631 enum flow_block_command command, bool ingress)
633 struct flow_block_offload bo = {
635 .binder_type = ingress ?
636 FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS :
637 FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS,
639 .block_shared = tcf_block_non_null_shared(block),
641 INIT_LIST_HEAD(&bo.cb_list);
646 bo.block = &block->flow_block;
648 down_write(&block->cb_lock);
649 cb(dev, cb_priv, TC_SETUP_BLOCK, &bo);
651 tcf_block_setup(block, &bo);
652 up_write(&block->cb_lock);
655 static struct tcf_block *tc_dev_block(struct net_device *dev, bool ingress)
657 const struct Qdisc_class_ops *cops;
658 const struct Qdisc_ops *ops;
661 if (!dev_ingress_queue(dev))
664 qdisc = dev_ingress_queue(dev)->qdisc_sleeping;
672 if (!ingress && !strcmp("ingress", ops->id))
679 if (!cops->tcf_block)
682 return cops->tcf_block(qdisc,
683 ingress ? TC_H_MIN_INGRESS : TC_H_MIN_EGRESS,
687 static void tc_indr_block_get_and_cmd(struct net_device *dev,
688 flow_indr_block_bind_cb_t *cb,
690 enum flow_block_command command)
692 struct tcf_block *block;
694 block = tc_dev_block(dev, true);
695 tc_indr_block_cmd(dev, block, cb, cb_priv, command, true);
697 block = tc_dev_block(dev, false);
698 tc_indr_block_cmd(dev, block, cb, cb_priv, command, false);
701 static void tc_indr_block_call(struct tcf_block *block,
702 struct net_device *dev,
703 struct tcf_block_ext_info *ei,
704 enum flow_block_command command,
705 struct netlink_ext_ack *extack)
707 struct flow_block_offload bo = {
709 .binder_type = ei->binder_type,
711 .block = &block->flow_block,
712 .block_shared = tcf_block_shared(block),
715 INIT_LIST_HEAD(&bo.cb_list);
717 flow_indr_block_call(dev, &bo, command);
718 tcf_block_setup(block, &bo);
721 static bool tcf_block_offload_in_use(struct tcf_block *block)
723 return atomic_read(&block->offloadcnt);
726 static int tcf_block_offload_cmd(struct tcf_block *block,
727 struct net_device *dev,
728 struct tcf_block_ext_info *ei,
729 enum flow_block_command command,
730 struct netlink_ext_ack *extack)
732 struct flow_block_offload bo = {};
735 bo.net = dev_net(dev);
736 bo.command = command;
737 bo.binder_type = ei->binder_type;
738 bo.block = &block->flow_block;
739 bo.block_shared = tcf_block_shared(block);
741 INIT_LIST_HEAD(&bo.cb_list);
743 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
747 return tcf_block_setup(block, &bo);
750 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
751 struct tcf_block_ext_info *ei,
752 struct netlink_ext_ack *extack)
754 struct net_device *dev = q->dev_queue->dev;
757 down_write(&block->cb_lock);
758 if (!dev->netdev_ops->ndo_setup_tc)
759 goto no_offload_dev_inc;
761 /* If tc offload feature is disabled and the block we try to bind
762 * to already has some offloaded filters, forbid to bind.
764 if (!tc_can_offload(dev) && tcf_block_offload_in_use(block)) {
765 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
770 err = tcf_block_offload_cmd(block, dev, ei, FLOW_BLOCK_BIND, extack);
771 if (err == -EOPNOTSUPP)
772 goto no_offload_dev_inc;
776 tc_indr_block_call(block, dev, ei, FLOW_BLOCK_BIND, extack);
777 up_write(&block->cb_lock);
781 if (tcf_block_offload_in_use(block)) {
786 block->nooffloaddevcnt++;
787 tc_indr_block_call(block, dev, ei, FLOW_BLOCK_BIND, extack);
789 up_write(&block->cb_lock);
793 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
794 struct tcf_block_ext_info *ei)
796 struct net_device *dev = q->dev_queue->dev;
799 down_write(&block->cb_lock);
800 tc_indr_block_call(block, dev, ei, FLOW_BLOCK_UNBIND, NULL);
802 if (!dev->netdev_ops->ndo_setup_tc)
803 goto no_offload_dev_dec;
804 err = tcf_block_offload_cmd(block, dev, ei, FLOW_BLOCK_UNBIND, NULL);
805 if (err == -EOPNOTSUPP)
806 goto no_offload_dev_dec;
807 up_write(&block->cb_lock);
811 WARN_ON(block->nooffloaddevcnt-- == 0);
812 up_write(&block->cb_lock);
816 tcf_chain0_head_change_cb_add(struct tcf_block *block,
817 struct tcf_block_ext_info *ei,
818 struct netlink_ext_ack *extack)
820 struct tcf_filter_chain_list_item *item;
821 struct tcf_chain *chain0;
823 item = kmalloc(sizeof(*item), GFP_KERNEL);
825 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
828 item->chain_head_change = ei->chain_head_change;
829 item->chain_head_change_priv = ei->chain_head_change_priv;
831 mutex_lock(&block->lock);
832 chain0 = block->chain0.chain;
834 tcf_chain_hold(chain0);
836 list_add(&item->list, &block->chain0.filter_chain_list);
837 mutex_unlock(&block->lock);
840 struct tcf_proto *tp_head;
842 mutex_lock(&chain0->filter_chain_lock);
844 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
846 tcf_chain_head_change_item(item, tp_head);
848 mutex_lock(&block->lock);
849 list_add(&item->list, &block->chain0.filter_chain_list);
850 mutex_unlock(&block->lock);
852 mutex_unlock(&chain0->filter_chain_lock);
853 tcf_chain_put(chain0);
860 tcf_chain0_head_change_cb_del(struct tcf_block *block,
861 struct tcf_block_ext_info *ei)
863 struct tcf_filter_chain_list_item *item;
865 mutex_lock(&block->lock);
866 list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
867 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
868 (item->chain_head_change == ei->chain_head_change &&
869 item->chain_head_change_priv == ei->chain_head_change_priv)) {
870 if (block->chain0.chain)
871 tcf_chain_head_change_item(item, NULL);
872 list_del(&item->list);
873 mutex_unlock(&block->lock);
879 mutex_unlock(&block->lock);
884 spinlock_t idr_lock; /* Protects idr */
888 static unsigned int tcf_net_id;
890 static int tcf_block_insert(struct tcf_block *block, struct net *net,
891 struct netlink_ext_ack *extack)
893 struct tcf_net *tn = net_generic(net, tcf_net_id);
896 idr_preload(GFP_KERNEL);
897 spin_lock(&tn->idr_lock);
898 err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
900 spin_unlock(&tn->idr_lock);
906 static void tcf_block_remove(struct tcf_block *block, struct net *net)
908 struct tcf_net *tn = net_generic(net, tcf_net_id);
910 spin_lock(&tn->idr_lock);
911 idr_remove(&tn->idr, block->index);
912 spin_unlock(&tn->idr_lock);
915 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
917 struct netlink_ext_ack *extack)
919 struct tcf_block *block;
921 block = kzalloc(sizeof(*block), GFP_KERNEL);
923 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
924 return ERR_PTR(-ENOMEM);
926 mutex_init(&block->lock);
927 mutex_init(&block->proto_destroy_lock);
928 init_rwsem(&block->cb_lock);
929 flow_block_init(&block->flow_block);
930 INIT_LIST_HEAD(&block->chain_list);
931 INIT_LIST_HEAD(&block->owner_list);
932 INIT_LIST_HEAD(&block->chain0.filter_chain_list);
934 refcount_set(&block->refcnt, 1);
936 block->index = block_index;
938 /* Don't store q pointer for blocks which are shared */
939 if (!tcf_block_shared(block))
944 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
946 struct tcf_net *tn = net_generic(net, tcf_net_id);
948 return idr_find(&tn->idr, block_index);
951 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
953 struct tcf_block *block;
956 block = tcf_block_lookup(net, block_index);
957 if (block && !refcount_inc_not_zero(&block->refcnt))
964 static struct tcf_chain *
965 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
967 mutex_lock(&block->lock);
969 chain = list_is_last(&chain->list, &block->chain_list) ?
970 NULL : list_next_entry(chain, list);
972 chain = list_first_entry_or_null(&block->chain_list,
973 struct tcf_chain, list);
975 /* skip all action-only chains */
976 while (chain && tcf_chain_held_by_acts_only(chain))
977 chain = list_is_last(&chain->list, &block->chain_list) ?
978 NULL : list_next_entry(chain, list);
981 tcf_chain_hold(chain);
982 mutex_unlock(&block->lock);
987 /* Function to be used by all clients that want to iterate over all chains on
988 * block. It properly obtains block->lock and takes reference to chain before
989 * returning it. Users of this function must be tolerant to concurrent chain
990 * insertion/deletion or ensure that no concurrent chain modification is
991 * possible. Note that all netlink dump callbacks cannot guarantee to provide
992 * consistent dump because rtnl lock is released each time skb is filled with
993 * data and sent to user-space.
997 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
999 struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
1002 tcf_chain_put(chain);
1006 EXPORT_SYMBOL(tcf_get_next_chain);
1008 static struct tcf_proto *
1009 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1014 mutex_lock(&chain->filter_chain_lock);
1017 tp = tcf_chain_dereference(chain->filter_chain, chain);
1018 } else if (tcf_proto_is_deleting(tp)) {
1019 /* 'deleting' flag is set and chain->filter_chain_lock was
1020 * unlocked, which means next pointer could be invalid. Restart
1023 prio = tp->prio + 1;
1024 tp = tcf_chain_dereference(chain->filter_chain, chain);
1026 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1027 if (!tp->deleting && tp->prio >= prio)
1030 tp = tcf_chain_dereference(tp->next, chain);
1036 mutex_unlock(&chain->filter_chain_lock);
1041 /* Function to be used by all clients that want to iterate over all tp's on
1042 * chain. Users of this function must be tolerant to concurrent tp
1043 * insertion/deletion or ensure that no concurrent chain modification is
1044 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1045 * consistent dump because rtnl lock is released each time skb is filled with
1046 * data and sent to user-space.
1050 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp,
1053 struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1056 tcf_proto_put(tp, rtnl_held, NULL);
1060 EXPORT_SYMBOL(tcf_get_next_proto);
1062 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1064 struct tcf_chain *chain;
1066 /* Last reference to block. At this point chains cannot be added or
1067 * removed concurrently.
1069 for (chain = tcf_get_next_chain(block, NULL);
1071 chain = tcf_get_next_chain(block, chain)) {
1072 tcf_chain_put_explicitly_created(chain);
1073 tcf_chain_flush(chain, rtnl_held);
1077 /* Lookup Qdisc and increments its reference counter.
1078 * Set parent, if necessary.
1081 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1082 u32 *parent, int ifindex, bool rtnl_held,
1083 struct netlink_ext_ack *extack)
1085 const struct Qdisc_class_ops *cops;
1086 struct net_device *dev;
1089 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1095 dev = dev_get_by_index_rcu(net, ifindex);
1104 *parent = (*q)->handle;
1106 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1108 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1114 *q = qdisc_refcount_inc_nz(*q);
1116 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1121 /* Is it classful? */
1122 cops = (*q)->ops->cl_ops;
1124 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1129 if (!cops->tcf_block) {
1130 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1136 /* At this point we know that qdisc is not noop_qdisc,
1137 * which means that qdisc holds a reference to net_device
1138 * and we hold a reference to qdisc, so it is safe to release
1150 qdisc_put_unlocked(*q);
1156 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1157 int ifindex, struct netlink_ext_ack *extack)
1159 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1162 /* Do we search for filter, attached to class? */
1163 if (TC_H_MIN(parent)) {
1164 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1166 *cl = cops->find(q, parent);
1168 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1176 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1177 unsigned long cl, int ifindex,
1179 struct netlink_ext_ack *extack)
1181 struct tcf_block *block;
1183 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1184 block = tcf_block_refcnt_get(net, block_index);
1186 NL_SET_ERR_MSG(extack, "Block of given index was not found");
1187 return ERR_PTR(-EINVAL);
1190 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1192 block = cops->tcf_block(q, cl, extack);
1194 return ERR_PTR(-EINVAL);
1196 if (tcf_block_shared(block)) {
1197 NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1198 return ERR_PTR(-EOPNOTSUPP);
1201 /* Always take reference to block in order to support execution
1202 * of rules update path of cls API without rtnl lock. Caller
1203 * must release block when it is finished using it. 'if' block
1204 * of this conditional obtain reference to block by calling
1205 * tcf_block_refcnt_get().
1207 refcount_inc(&block->refcnt);
1213 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1214 struct tcf_block_ext_info *ei, bool rtnl_held)
1216 if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1217 /* Flushing/putting all chains will cause the block to be
1218 * deallocated when last chain is freed. However, if chain_list
1219 * is empty, block has to be manually deallocated. After block
1220 * reference counter reached 0, it is no longer possible to
1221 * increment it or add new chains to block.
1223 bool free_block = list_empty(&block->chain_list);
1225 mutex_unlock(&block->lock);
1226 if (tcf_block_shared(block))
1227 tcf_block_remove(block, block->net);
1230 tcf_block_offload_unbind(block, q, ei);
1233 tcf_block_destroy(block);
1235 tcf_block_flush_all_chains(block, rtnl_held);
1237 tcf_block_offload_unbind(block, q, ei);
1241 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1243 __tcf_block_put(block, NULL, NULL, rtnl_held);
1247 * Set q, parent, cl when appropriate.
1250 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1251 u32 *parent, unsigned long *cl,
1252 int ifindex, u32 block_index,
1253 struct netlink_ext_ack *extack)
1255 struct tcf_block *block;
1260 err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1264 err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1268 block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1269 if (IS_ERR(block)) {
1270 err = PTR_ERR(block);
1281 return ERR_PTR(err);
1284 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1287 if (!IS_ERR_OR_NULL(block))
1288 tcf_block_refcnt_put(block, rtnl_held);
1294 qdisc_put_unlocked(q);
1298 struct tcf_block_owner_item {
1299 struct list_head list;
1301 enum flow_block_binder_type binder_type;
1305 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1307 enum flow_block_binder_type binder_type)
1309 if (block->keep_dst &&
1310 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1311 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1312 netif_keep_dst(qdisc_dev(q));
1315 void tcf_block_netif_keep_dst(struct tcf_block *block)
1317 struct tcf_block_owner_item *item;
1319 block->keep_dst = true;
1320 list_for_each_entry(item, &block->owner_list, list)
1321 tcf_block_owner_netif_keep_dst(block, item->q,
1324 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1326 static int tcf_block_owner_add(struct tcf_block *block,
1328 enum flow_block_binder_type binder_type)
1330 struct tcf_block_owner_item *item;
1332 item = kmalloc(sizeof(*item), GFP_KERNEL);
1336 item->binder_type = binder_type;
1337 list_add(&item->list, &block->owner_list);
1341 static void tcf_block_owner_del(struct tcf_block *block,
1343 enum flow_block_binder_type binder_type)
1345 struct tcf_block_owner_item *item;
1347 list_for_each_entry(item, &block->owner_list, list) {
1348 if (item->q == q && item->binder_type == binder_type) {
1349 list_del(&item->list);
1357 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1358 struct tcf_block_ext_info *ei,
1359 struct netlink_ext_ack *extack)
1361 struct net *net = qdisc_net(q);
1362 struct tcf_block *block = NULL;
1365 if (ei->block_index)
1366 /* block_index not 0 means the shared block is requested */
1367 block = tcf_block_refcnt_get(net, ei->block_index);
1370 block = tcf_block_create(net, q, ei->block_index, extack);
1372 return PTR_ERR(block);
1373 if (tcf_block_shared(block)) {
1374 err = tcf_block_insert(block, net, extack);
1376 goto err_block_insert;
1380 err = tcf_block_owner_add(block, q, ei->binder_type);
1382 goto err_block_owner_add;
1384 tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1386 err = tcf_chain0_head_change_cb_add(block, ei, extack);
1388 goto err_chain0_head_change_cb_add;
1390 err = tcf_block_offload_bind(block, q, ei, extack);
1392 goto err_block_offload_bind;
1397 err_block_offload_bind:
1398 tcf_chain0_head_change_cb_del(block, ei);
1399 err_chain0_head_change_cb_add:
1400 tcf_block_owner_del(block, q, ei->binder_type);
1401 err_block_owner_add:
1403 tcf_block_refcnt_put(block, true);
1406 EXPORT_SYMBOL(tcf_block_get_ext);
1408 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1410 struct tcf_proto __rcu **p_filter_chain = priv;
1412 rcu_assign_pointer(*p_filter_chain, tp_head);
1415 int tcf_block_get(struct tcf_block **p_block,
1416 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1417 struct netlink_ext_ack *extack)
1419 struct tcf_block_ext_info ei = {
1420 .chain_head_change = tcf_chain_head_change_dflt,
1421 .chain_head_change_priv = p_filter_chain,
1424 WARN_ON(!p_filter_chain);
1425 return tcf_block_get_ext(p_block, q, &ei, extack);
1427 EXPORT_SYMBOL(tcf_block_get);
1429 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1430 * actions should be all removed after flushing.
1432 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1433 struct tcf_block_ext_info *ei)
1437 tcf_chain0_head_change_cb_del(block, ei);
1438 tcf_block_owner_del(block, q, ei->binder_type);
1440 __tcf_block_put(block, q, ei, true);
1442 EXPORT_SYMBOL(tcf_block_put_ext);
1444 void tcf_block_put(struct tcf_block *block)
1446 struct tcf_block_ext_info ei = {0, };
1450 tcf_block_put_ext(block, block->q, &ei);
1453 EXPORT_SYMBOL(tcf_block_put);
1456 tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1457 void *cb_priv, bool add, bool offload_in_use,
1458 struct netlink_ext_ack *extack)
1460 struct tcf_chain *chain, *chain_prev;
1461 struct tcf_proto *tp, *tp_prev;
1464 lockdep_assert_held(&block->cb_lock);
1466 for (chain = __tcf_get_next_chain(block, NULL);
1469 chain = __tcf_get_next_chain(block, chain),
1470 tcf_chain_put(chain_prev)) {
1471 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1473 tp = __tcf_get_next_proto(chain, tp),
1474 tcf_proto_put(tp_prev, true, NULL)) {
1475 if (tp->ops->reoffload) {
1476 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1479 goto err_playback_remove;
1480 } else if (add && offload_in_use) {
1482 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1483 goto err_playback_remove;
1490 err_playback_remove:
1491 tcf_proto_put(tp, true, NULL);
1492 tcf_chain_put(chain);
1493 tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1498 static int tcf_block_bind(struct tcf_block *block,
1499 struct flow_block_offload *bo)
1501 struct flow_block_cb *block_cb, *next;
1504 lockdep_assert_held(&block->cb_lock);
1506 list_for_each_entry(block_cb, &bo->cb_list, list) {
1507 err = tcf_block_playback_offloads(block, block_cb->cb,
1508 block_cb->cb_priv, true,
1509 tcf_block_offload_in_use(block),
1513 if (!bo->unlocked_driver_cb)
1514 block->lockeddevcnt++;
1518 list_splice(&bo->cb_list, &block->flow_block.cb_list);
1523 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1525 list_del(&block_cb->list);
1526 tcf_block_playback_offloads(block, block_cb->cb,
1527 block_cb->cb_priv, false,
1528 tcf_block_offload_in_use(block),
1530 if (!bo->unlocked_driver_cb)
1531 block->lockeddevcnt--;
1533 flow_block_cb_free(block_cb);
1539 static void tcf_block_unbind(struct tcf_block *block,
1540 struct flow_block_offload *bo)
1542 struct flow_block_cb *block_cb, *next;
1544 lockdep_assert_held(&block->cb_lock);
1546 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1547 tcf_block_playback_offloads(block, block_cb->cb,
1548 block_cb->cb_priv, false,
1549 tcf_block_offload_in_use(block),
1551 list_del(&block_cb->list);
1552 flow_block_cb_free(block_cb);
1553 if (!bo->unlocked_driver_cb)
1554 block->lockeddevcnt--;
1558 static int tcf_block_setup(struct tcf_block *block,
1559 struct flow_block_offload *bo)
1563 switch (bo->command) {
1564 case FLOW_BLOCK_BIND:
1565 err = tcf_block_bind(block, bo);
1567 case FLOW_BLOCK_UNBIND:
1569 tcf_block_unbind(block, bo);
1579 /* Main classifier routine: scans classifier chain attached
1580 * to this qdisc, (optionally) tests for protocol and asks
1581 * specific classifiers.
1583 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1584 struct tcf_result *res, bool compat_mode)
1586 #ifdef CONFIG_NET_CLS_ACT
1587 const int max_reclassify_loop = 4;
1588 const struct tcf_proto *orig_tp = tp;
1589 const struct tcf_proto *first_tp;
1594 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1595 __be16 protocol = tc_skb_protocol(skb);
1598 if (tp->protocol != protocol &&
1599 tp->protocol != htons(ETH_P_ALL))
1602 err = tp->classify(skb, tp, res);
1603 #ifdef CONFIG_NET_CLS_ACT
1604 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1607 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1608 first_tp = res->goto_tp;
1610 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1612 struct tc_skb_ext *ext;
1614 ext = skb_ext_add(skb, TC_SKB_EXT);
1615 if (WARN_ON_ONCE(!ext))
1618 ext->chain = err & TC_ACT_EXT_VAL_MASK;
1628 return TC_ACT_UNSPEC; /* signal: continue lookup */
1629 #ifdef CONFIG_NET_CLS_ACT
1631 if (unlikely(limit++ >= max_reclassify_loop)) {
1632 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1633 tp->chain->block->index,
1635 ntohs(tp->protocol));
1643 EXPORT_SYMBOL(tcf_classify);
1645 struct tcf_chain_info {
1646 struct tcf_proto __rcu **pprev;
1647 struct tcf_proto __rcu *next;
1650 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1651 struct tcf_chain_info *chain_info)
1653 return tcf_chain_dereference(*chain_info->pprev, chain);
1656 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1657 struct tcf_chain_info *chain_info,
1658 struct tcf_proto *tp)
1660 if (chain->flushing)
1663 if (*chain_info->pprev == chain->filter_chain)
1664 tcf_chain0_head_change(chain, tp);
1666 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1667 rcu_assign_pointer(*chain_info->pprev, tp);
1672 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1673 struct tcf_chain_info *chain_info,
1674 struct tcf_proto *tp)
1676 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1678 tcf_proto_mark_delete(tp);
1679 if (tp == chain->filter_chain)
1680 tcf_chain0_head_change(chain, next);
1681 RCU_INIT_POINTER(*chain_info->pprev, next);
1684 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1685 struct tcf_chain_info *chain_info,
1686 u32 protocol, u32 prio,
1687 bool prio_allocate);
1689 /* Try to insert new proto.
1690 * If proto with specified priority already exists, free new proto
1691 * and return existing one.
1694 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1695 struct tcf_proto *tp_new,
1696 u32 protocol, u32 prio,
1699 struct tcf_chain_info chain_info;
1700 struct tcf_proto *tp;
1703 mutex_lock(&chain->filter_chain_lock);
1705 if (tcf_proto_exists_destroying(chain, tp_new)) {
1706 mutex_unlock(&chain->filter_chain_lock);
1707 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1708 return ERR_PTR(-EAGAIN);
1711 tp = tcf_chain_tp_find(chain, &chain_info,
1712 protocol, prio, false);
1714 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1715 mutex_unlock(&chain->filter_chain_lock);
1718 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1721 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1722 tp_new = ERR_PTR(err);
1728 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1729 struct tcf_proto *tp, bool rtnl_held,
1730 struct netlink_ext_ack *extack)
1732 struct tcf_chain_info chain_info;
1733 struct tcf_proto *tp_iter;
1734 struct tcf_proto **pprev;
1735 struct tcf_proto *next;
1737 mutex_lock(&chain->filter_chain_lock);
1739 /* Atomically find and remove tp from chain. */
1740 for (pprev = &chain->filter_chain;
1741 (tp_iter = tcf_chain_dereference(*pprev, chain));
1742 pprev = &tp_iter->next) {
1743 if (tp_iter == tp) {
1744 chain_info.pprev = pprev;
1745 chain_info.next = tp_iter->next;
1746 WARN_ON(tp_iter->deleting);
1750 /* Verify that tp still exists and no new filters were inserted
1752 * Mark tp for deletion if it is empty.
1754 if (!tp_iter || !tcf_proto_check_delete(tp, rtnl_held)) {
1755 mutex_unlock(&chain->filter_chain_lock);
1759 tcf_proto_signal_destroying(chain, tp);
1760 next = tcf_chain_dereference(chain_info.next, chain);
1761 if (tp == chain->filter_chain)
1762 tcf_chain0_head_change(chain, next);
1763 RCU_INIT_POINTER(*chain_info.pprev, next);
1764 mutex_unlock(&chain->filter_chain_lock);
1766 tcf_proto_put(tp, rtnl_held, extack);
1769 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1770 struct tcf_chain_info *chain_info,
1771 u32 protocol, u32 prio,
1774 struct tcf_proto **pprev;
1775 struct tcf_proto *tp;
1777 /* Check the chain for existence of proto-tcf with this priority */
1778 for (pprev = &chain->filter_chain;
1779 (tp = tcf_chain_dereference(*pprev, chain));
1780 pprev = &tp->next) {
1781 if (tp->prio >= prio) {
1782 if (tp->prio == prio) {
1783 if (prio_allocate ||
1784 (tp->protocol != protocol && protocol))
1785 return ERR_PTR(-EINVAL);
1792 chain_info->pprev = pprev;
1794 chain_info->next = tp->next;
1797 chain_info->next = NULL;
1802 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1803 struct tcf_proto *tp, struct tcf_block *block,
1804 struct Qdisc *q, u32 parent, void *fh,
1805 u32 portid, u32 seq, u16 flags, int event,
1809 struct nlmsghdr *nlh;
1810 unsigned char *b = skb_tail_pointer(skb);
1812 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1814 goto out_nlmsg_trim;
1815 tcm = nlmsg_data(nlh);
1816 tcm->tcm_family = AF_UNSPEC;
1820 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1821 tcm->tcm_parent = parent;
1823 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1824 tcm->tcm_block_index = block->index;
1826 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1827 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1828 goto nla_put_failure;
1829 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1830 goto nla_put_failure;
1832 tcm->tcm_handle = 0;
1834 if (tp->ops->dump &&
1835 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
1836 goto nla_put_failure;
1838 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1847 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1848 struct nlmsghdr *n, struct tcf_proto *tp,
1849 struct tcf_block *block, struct Qdisc *q,
1850 u32 parent, void *fh, int event, bool unicast,
1853 struct sk_buff *skb;
1854 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1857 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1861 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1862 n->nlmsg_seq, n->nlmsg_flags, event,
1869 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1871 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1872 n->nlmsg_flags & NLM_F_ECHO);
1879 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1880 struct nlmsghdr *n, struct tcf_proto *tp,
1881 struct tcf_block *block, struct Qdisc *q,
1882 u32 parent, void *fh, bool unicast, bool *last,
1883 bool rtnl_held, struct netlink_ext_ack *extack)
1885 struct sk_buff *skb;
1886 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1889 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1893 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1894 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
1896 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1901 err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
1908 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1910 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1911 n->nlmsg_flags & NLM_F_ECHO);
1913 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1920 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1921 struct tcf_block *block, struct Qdisc *q,
1922 u32 parent, struct nlmsghdr *n,
1923 struct tcf_chain *chain, int event,
1926 struct tcf_proto *tp;
1928 for (tp = tcf_get_next_proto(chain, NULL, rtnl_held);
1929 tp; tp = tcf_get_next_proto(chain, tp, rtnl_held))
1930 tfilter_notify(net, oskb, n, tp, block,
1931 q, parent, NULL, event, false, rtnl_held);
1934 static void tfilter_put(struct tcf_proto *tp, void *fh)
1936 if (tp->ops->put && fh)
1937 tp->ops->put(tp, fh);
1940 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1941 struct netlink_ext_ack *extack)
1943 struct net *net = sock_net(skb->sk);
1944 struct nlattr *tca[TCA_MAX + 1];
1945 char name[IFNAMSIZ];
1952 struct Qdisc *q = NULL;
1953 struct tcf_chain_info chain_info;
1954 struct tcf_chain *chain = NULL;
1955 struct tcf_block *block;
1956 struct tcf_proto *tp;
1961 bool rtnl_held = false;
1963 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1969 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
1970 rtm_tca_policy, extack);
1975 protocol = TC_H_MIN(t->tcm_info);
1976 prio = TC_H_MAJ(t->tcm_info);
1977 prio_allocate = false;
1978 parent = t->tcm_parent;
1984 /* If no priority is provided by the user,
1987 if (n->nlmsg_flags & NLM_F_CREATE) {
1988 prio = TC_H_MAKE(0x80000000U, 0U);
1989 prio_allocate = true;
1991 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1996 /* Find head of filter chain. */
1998 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2002 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2003 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2008 /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2009 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2010 * type is not specified, classifier is not unlocked.
2013 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2014 !tcf_proto_is_unlocked(name)) {
2019 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2023 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2025 if (IS_ERR(block)) {
2026 err = PTR_ERR(block);
2030 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2031 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2032 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2036 chain = tcf_chain_get(block, chain_index, true);
2038 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2043 mutex_lock(&chain->filter_chain_lock);
2044 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2045 prio, prio_allocate);
2047 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2053 struct tcf_proto *tp_new = NULL;
2055 if (chain->flushing) {
2060 /* Proto-tcf does not exist, create new one */
2062 if (tca[TCA_KIND] == NULL || !protocol) {
2063 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2068 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2069 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2075 prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2078 mutex_unlock(&chain->filter_chain_lock);
2079 tp_new = tcf_proto_create(nla_data(tca[TCA_KIND]),
2080 protocol, prio, chain, rtnl_held,
2082 if (IS_ERR(tp_new)) {
2083 err = PTR_ERR(tp_new);
2088 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2095 mutex_unlock(&chain->filter_chain_lock);
2098 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2099 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2104 fh = tp->ops->get(tp, t->tcm_handle);
2107 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2108 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2112 } else if (n->nlmsg_flags & NLM_F_EXCL) {
2113 tfilter_put(tp, fh);
2114 NL_SET_ERR_MSG(extack, "Filter already exists");
2119 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2120 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2125 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2126 n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
2129 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2130 RTM_NEWTFILTER, false, rtnl_held);
2131 tfilter_put(tp, fh);
2132 /* q pointer is NULL for shared blocks */
2134 q->flags &= ~TCQ_F_CAN_BYPASS;
2138 if (err && tp_created)
2139 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2142 if (tp && !IS_ERR(tp))
2143 tcf_proto_put(tp, rtnl_held, NULL);
2145 tcf_chain_put(chain);
2147 tcf_block_release(q, block, rtnl_held);
2152 if (err == -EAGAIN) {
2153 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2157 /* Replay the request. */
2163 mutex_unlock(&chain->filter_chain_lock);
2167 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2168 struct netlink_ext_ack *extack)
2170 struct net *net = sock_net(skb->sk);
2171 struct nlattr *tca[TCA_MAX + 1];
2172 char name[IFNAMSIZ];
2178 struct Qdisc *q = NULL;
2179 struct tcf_chain_info chain_info;
2180 struct tcf_chain *chain = NULL;
2181 struct tcf_block *block = NULL;
2182 struct tcf_proto *tp = NULL;
2183 unsigned long cl = 0;
2186 bool rtnl_held = false;
2188 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2191 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2192 rtm_tca_policy, extack);
2197 protocol = TC_H_MIN(t->tcm_info);
2198 prio = TC_H_MAJ(t->tcm_info);
2199 parent = t->tcm_parent;
2201 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2202 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2206 /* Find head of filter chain. */
2208 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2212 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2213 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2217 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2218 * found), qdisc is not unlocked, classifier type is not specified,
2219 * classifier is not unlocked.
2222 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2223 !tcf_proto_is_unlocked(name)) {
2228 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2232 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2234 if (IS_ERR(block)) {
2235 err = PTR_ERR(block);
2239 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2240 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2241 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2245 chain = tcf_chain_get(block, chain_index, false);
2247 /* User requested flush on non-existent chain. Nothing to do,
2248 * so just return success.
2254 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2260 tfilter_notify_chain(net, skb, block, q, parent, n,
2261 chain, RTM_DELTFILTER, rtnl_held);
2262 tcf_chain_flush(chain, rtnl_held);
2267 mutex_lock(&chain->filter_chain_lock);
2268 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2270 if (!tp || IS_ERR(tp)) {
2271 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2272 err = tp ? PTR_ERR(tp) : -ENOENT;
2274 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2275 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2278 } else if (t->tcm_handle == 0) {
2279 tcf_proto_signal_destroying(chain, tp);
2280 tcf_chain_tp_remove(chain, &chain_info, tp);
2281 mutex_unlock(&chain->filter_chain_lock);
2283 tcf_proto_put(tp, rtnl_held, NULL);
2284 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2285 RTM_DELTFILTER, false, rtnl_held);
2289 mutex_unlock(&chain->filter_chain_lock);
2291 fh = tp->ops->get(tp, t->tcm_handle);
2294 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2299 err = tfilter_del_notify(net, skb, n, tp, block,
2300 q, parent, fh, false, &last,
2306 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2311 if (tp && !IS_ERR(tp))
2312 tcf_proto_put(tp, rtnl_held, NULL);
2313 tcf_chain_put(chain);
2315 tcf_block_release(q, block, rtnl_held);
2323 mutex_unlock(&chain->filter_chain_lock);
2327 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2328 struct netlink_ext_ack *extack)
2330 struct net *net = sock_net(skb->sk);
2331 struct nlattr *tca[TCA_MAX + 1];
2332 char name[IFNAMSIZ];
2338 struct Qdisc *q = NULL;
2339 struct tcf_chain_info chain_info;
2340 struct tcf_chain *chain = NULL;
2341 struct tcf_block *block = NULL;
2342 struct tcf_proto *tp = NULL;
2343 unsigned long cl = 0;
2346 bool rtnl_held = false;
2348 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2349 rtm_tca_policy, extack);
2354 protocol = TC_H_MIN(t->tcm_info);
2355 prio = TC_H_MAJ(t->tcm_info);
2356 parent = t->tcm_parent;
2359 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2363 /* Find head of filter chain. */
2365 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2369 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2370 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2374 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2375 * unlocked, classifier type is not specified, classifier is not
2378 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2379 !tcf_proto_is_unlocked(name)) {
2384 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2388 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2390 if (IS_ERR(block)) {
2391 err = PTR_ERR(block);
2395 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2396 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2397 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2401 chain = tcf_chain_get(block, chain_index, false);
2403 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2408 mutex_lock(&chain->filter_chain_lock);
2409 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2411 mutex_unlock(&chain->filter_chain_lock);
2412 if (!tp || IS_ERR(tp)) {
2413 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2414 err = tp ? PTR_ERR(tp) : -ENOENT;
2416 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2417 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2422 fh = tp->ops->get(tp, t->tcm_handle);
2425 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2428 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2429 fh, RTM_NEWTFILTER, true, rtnl_held);
2431 NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2434 tfilter_put(tp, fh);
2437 if (tp && !IS_ERR(tp))
2438 tcf_proto_put(tp, rtnl_held, NULL);
2439 tcf_chain_put(chain);
2441 tcf_block_release(q, block, rtnl_held);
2449 struct tcf_dump_args {
2450 struct tcf_walker w;
2451 struct sk_buff *skb;
2452 struct netlink_callback *cb;
2453 struct tcf_block *block;
2458 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2460 struct tcf_dump_args *a = (void *)arg;
2461 struct net *net = sock_net(a->skb->sk);
2463 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2464 n, NETLINK_CB(a->cb->skb).portid,
2465 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2466 RTM_NEWTFILTER, true);
2469 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2470 struct sk_buff *skb, struct netlink_callback *cb,
2471 long index_start, long *p_index)
2473 struct net *net = sock_net(skb->sk);
2474 struct tcf_block *block = chain->block;
2475 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2476 struct tcf_proto *tp, *tp_prev;
2477 struct tcf_dump_args arg;
2479 for (tp = __tcf_get_next_proto(chain, NULL);
2482 tp = __tcf_get_next_proto(chain, tp),
2483 tcf_proto_put(tp_prev, true, NULL),
2485 if (*p_index < index_start)
2487 if (TC_H_MAJ(tcm->tcm_info) &&
2488 TC_H_MAJ(tcm->tcm_info) != tp->prio)
2490 if (TC_H_MIN(tcm->tcm_info) &&
2491 TC_H_MIN(tcm->tcm_info) != tp->protocol)
2493 if (*p_index > index_start)
2494 memset(&cb->args[1], 0,
2495 sizeof(cb->args) - sizeof(cb->args[0]));
2496 if (cb->args[1] == 0) {
2497 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2498 NETLINK_CB(cb->skb).portid,
2499 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2500 RTM_NEWTFILTER, true) <= 0)
2506 arg.w.fn = tcf_node_dump;
2511 arg.parent = parent;
2513 arg.w.skip = cb->args[1] - 1;
2515 arg.w.cookie = cb->args[2];
2516 tp->ops->walk(tp, &arg.w, true);
2517 cb->args[2] = arg.w.cookie;
2518 cb->args[1] = arg.w.count + 1;
2525 tcf_proto_put(tp, true, NULL);
2529 /* called with RTNL */
2530 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2532 struct tcf_chain *chain, *chain_prev;
2533 struct net *net = sock_net(skb->sk);
2534 struct nlattr *tca[TCA_MAX + 1];
2535 struct Qdisc *q = NULL;
2536 struct tcf_block *block;
2537 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2543 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2546 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2551 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2552 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2555 /* If we work with block index, q is NULL and parent value
2556 * will never be used in the following code. The check
2557 * in tcf_fill_node prevents it. However, compiler does not
2558 * see that far, so set parent to zero to silence the warning
2559 * about parent being uninitialized.
2563 const struct Qdisc_class_ops *cops;
2564 struct net_device *dev;
2565 unsigned long cl = 0;
2567 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2571 parent = tcm->tcm_parent;
2576 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2580 cops = q->ops->cl_ops;
2583 if (!cops->tcf_block)
2585 if (TC_H_MIN(tcm->tcm_parent)) {
2586 cl = cops->find(q, tcm->tcm_parent);
2590 block = cops->tcf_block(q, cl, NULL);
2593 if (tcf_block_shared(block))
2597 index_start = cb->args[0];
2600 for (chain = __tcf_get_next_chain(block, NULL);
2603 chain = __tcf_get_next_chain(block, chain),
2604 tcf_chain_put(chain_prev)) {
2605 if (tca[TCA_CHAIN] &&
2606 nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2608 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2609 index_start, &index)) {
2610 tcf_chain_put(chain);
2616 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2617 tcf_block_refcnt_put(block, true);
2618 cb->args[0] = index;
2621 /* If we did no progress, the error (EMSGSIZE) is real */
2622 if (skb->len == 0 && err)
2627 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2628 void *tmplt_priv, u32 chain_index,
2629 struct net *net, struct sk_buff *skb,
2630 struct tcf_block *block,
2631 u32 portid, u32 seq, u16 flags, int event)
2633 unsigned char *b = skb_tail_pointer(skb);
2634 const struct tcf_proto_ops *ops;
2635 struct nlmsghdr *nlh;
2642 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2644 goto out_nlmsg_trim;
2645 tcm = nlmsg_data(nlh);
2646 tcm->tcm_family = AF_UNSPEC;
2649 tcm->tcm_handle = 0;
2651 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2652 tcm->tcm_parent = block->q->handle;
2654 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2655 tcm->tcm_block_index = block->index;
2658 if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2659 goto nla_put_failure;
2662 if (nla_put_string(skb, TCA_KIND, ops->kind))
2663 goto nla_put_failure;
2664 if (ops->tmplt_dump(skb, net, priv) < 0)
2665 goto nla_put_failure;
2668 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2677 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2678 u32 seq, u16 flags, int event, bool unicast)
2680 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2681 struct tcf_block *block = chain->block;
2682 struct net *net = block->net;
2683 struct sk_buff *skb;
2686 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2690 if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2691 chain->index, net, skb, block, portid,
2692 seq, flags, event) <= 0) {
2698 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2700 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2701 flags & NLM_F_ECHO);
2708 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2709 void *tmplt_priv, u32 chain_index,
2710 struct tcf_block *block, struct sk_buff *oskb,
2711 u32 seq, u16 flags, bool unicast)
2713 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2714 struct net *net = block->net;
2715 struct sk_buff *skb;
2717 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2721 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2722 block, portid, seq, flags, RTM_DELCHAIN) <= 0) {
2728 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2730 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2733 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2734 struct nlattr **tca,
2735 struct netlink_ext_ack *extack)
2737 const struct tcf_proto_ops *ops;
2738 char name[IFNAMSIZ];
2741 /* If kind is not set, user did not specify template. */
2745 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2746 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
2750 ops = tcf_proto_lookup_ops(name, true, extack);
2752 return PTR_ERR(ops);
2753 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2754 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2758 tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2759 if (IS_ERR(tmplt_priv)) {
2760 module_put(ops->owner);
2761 return PTR_ERR(tmplt_priv);
2763 chain->tmplt_ops = ops;
2764 chain->tmplt_priv = tmplt_priv;
2768 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2771 /* If template ops are set, no work to do for us. */
2775 tmplt_ops->tmplt_destroy(tmplt_priv);
2776 module_put(tmplt_ops->owner);
2779 /* Add/delete/get a chain */
2781 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2782 struct netlink_ext_ack *extack)
2784 struct net *net = sock_net(skb->sk);
2785 struct nlattr *tca[TCA_MAX + 1];
2789 struct Qdisc *q = NULL;
2790 struct tcf_chain *chain = NULL;
2791 struct tcf_block *block;
2795 if (n->nlmsg_type != RTM_GETCHAIN &&
2796 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2800 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2801 rtm_tca_policy, extack);
2806 parent = t->tcm_parent;
2809 block = tcf_block_find(net, &q, &parent, &cl,
2810 t->tcm_ifindex, t->tcm_block_index, extack);
2812 return PTR_ERR(block);
2814 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2815 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2816 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2821 mutex_lock(&block->lock);
2822 chain = tcf_chain_lookup(block, chain_index);
2823 if (n->nlmsg_type == RTM_NEWCHAIN) {
2825 if (tcf_chain_held_by_acts_only(chain)) {
2826 /* The chain exists only because there is
2827 * some action referencing it.
2829 tcf_chain_hold(chain);
2831 NL_SET_ERR_MSG(extack, "Filter chain already exists");
2833 goto errout_block_locked;
2836 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2837 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
2839 goto errout_block_locked;
2841 chain = tcf_chain_create(block, chain_index);
2843 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
2845 goto errout_block_locked;
2849 if (!chain || tcf_chain_held_by_acts_only(chain)) {
2850 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2852 goto errout_block_locked;
2854 tcf_chain_hold(chain);
2857 if (n->nlmsg_type == RTM_NEWCHAIN) {
2858 /* Modifying chain requires holding parent block lock. In case
2859 * the chain was successfully added, take a reference to the
2860 * chain. This ensures that an empty chain does not disappear at
2861 * the end of this function.
2863 tcf_chain_hold(chain);
2864 chain->explicitly_created = true;
2866 mutex_unlock(&block->lock);
2868 switch (n->nlmsg_type) {
2870 err = tc_chain_tmplt_add(chain, net, tca, extack);
2872 tcf_chain_put_explicitly_created(chain);
2876 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2877 RTM_NEWCHAIN, false);
2880 tfilter_notify_chain(net, skb, block, q, parent, n,
2881 chain, RTM_DELTFILTER, true);
2882 /* Flush the chain first as the user requested chain removal. */
2883 tcf_chain_flush(chain, true);
2884 /* In case the chain was successfully deleted, put a reference
2885 * to the chain previously taken during addition.
2887 tcf_chain_put_explicitly_created(chain);
2890 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
2891 n->nlmsg_seq, n->nlmsg_type, true);
2893 NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2897 NL_SET_ERR_MSG(extack, "Unsupported message type");
2902 tcf_chain_put(chain);
2904 tcf_block_release(q, block, true);
2906 /* Replay the request. */
2910 errout_block_locked:
2911 mutex_unlock(&block->lock);
2915 /* called with RTNL */
2916 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2918 struct net *net = sock_net(skb->sk);
2919 struct nlattr *tca[TCA_MAX + 1];
2920 struct Qdisc *q = NULL;
2921 struct tcf_block *block;
2922 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2923 struct tcf_chain *chain;
2929 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2932 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2933 rtm_tca_policy, cb->extack);
2937 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2938 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2941 /* If we work with block index, q is NULL and parent value
2942 * will never be used in the following code. The check
2943 * in tcf_fill_node prevents it. However, compiler does not
2944 * see that far, so set parent to zero to silence the warning
2945 * about parent being uninitialized.
2949 const struct Qdisc_class_ops *cops;
2950 struct net_device *dev;
2951 unsigned long cl = 0;
2953 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2957 parent = tcm->tcm_parent;
2962 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2966 cops = q->ops->cl_ops;
2969 if (!cops->tcf_block)
2971 if (TC_H_MIN(tcm->tcm_parent)) {
2972 cl = cops->find(q, tcm->tcm_parent);
2976 block = cops->tcf_block(q, cl, NULL);
2979 if (tcf_block_shared(block))
2983 index_start = cb->args[0];
2986 mutex_lock(&block->lock);
2987 list_for_each_entry(chain, &block->chain_list, list) {
2988 if ((tca[TCA_CHAIN] &&
2989 nla_get_u32(tca[TCA_CHAIN]) != chain->index))
2991 if (index < index_start) {
2995 if (tcf_chain_held_by_acts_only(chain))
2997 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2998 chain->index, net, skb, block,
2999 NETLINK_CB(cb->skb).portid,
3000 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3006 mutex_unlock(&block->lock);
3008 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3009 tcf_block_refcnt_put(block, true);
3010 cb->args[0] = index;
3013 /* If we did no progress, the error (EMSGSIZE) is real */
3014 if (skb->len == 0 && err)
3019 void tcf_exts_destroy(struct tcf_exts *exts)
3021 #ifdef CONFIG_NET_CLS_ACT
3022 if (exts->actions) {
3023 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3024 kfree(exts->actions);
3026 exts->nr_actions = 0;
3029 EXPORT_SYMBOL(tcf_exts_destroy);
3031 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3032 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
3033 bool rtnl_held, struct netlink_ext_ack *extack)
3035 #ifdef CONFIG_NET_CLS_ACT
3037 struct tc_action *act;
3038 size_t attr_size = 0;
3040 if (exts->police && tb[exts->police]) {
3041 act = tcf_action_init_1(net, tp, tb[exts->police],
3042 rate_tlv, "police", ovr,
3043 TCA_ACT_BIND, rtnl_held,
3046 return PTR_ERR(act);
3048 act->type = exts->type = TCA_OLD_COMPAT;
3049 exts->actions[0] = act;
3050 exts->nr_actions = 1;
3051 } else if (exts->action && tb[exts->action]) {
3054 err = tcf_action_init(net, tp, tb[exts->action],
3055 rate_tlv, NULL, ovr, TCA_ACT_BIND,
3056 exts->actions, &attr_size,
3060 exts->nr_actions = err;
3064 if ((exts->action && tb[exts->action]) ||
3065 (exts->police && tb[exts->police])) {
3066 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3073 EXPORT_SYMBOL(tcf_exts_validate);
3075 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3077 #ifdef CONFIG_NET_CLS_ACT
3078 struct tcf_exts old = *dst;
3081 tcf_exts_destroy(&old);
3084 EXPORT_SYMBOL(tcf_exts_change);
3086 #ifdef CONFIG_NET_CLS_ACT
3087 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3089 if (exts->nr_actions == 0)
3092 return exts->actions[0];
3096 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3098 #ifdef CONFIG_NET_CLS_ACT
3099 struct nlattr *nest;
3101 if (exts->action && tcf_exts_has_actions(exts)) {
3103 * again for backward compatible mode - we want
3104 * to work with both old and new modes of entering
3105 * tc data even if iproute2 was newer - jhs
3107 if (exts->type != TCA_OLD_COMPAT) {
3108 nest = nla_nest_start_noflag(skb, exts->action);
3110 goto nla_put_failure;
3112 if (tcf_action_dump(skb, exts->actions, 0, 0) < 0)
3113 goto nla_put_failure;
3114 nla_nest_end(skb, nest);
3115 } else if (exts->police) {
3116 struct tc_action *act = tcf_exts_first_act(exts);
3117 nest = nla_nest_start_noflag(skb, exts->police);
3118 if (nest == NULL || !act)
3119 goto nla_put_failure;
3120 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3121 goto nla_put_failure;
3122 nla_nest_end(skb, nest);
3128 nla_nest_cancel(skb, nest);
3134 EXPORT_SYMBOL(tcf_exts_dump);
3137 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3139 #ifdef CONFIG_NET_CLS_ACT
3140 struct tc_action *a = tcf_exts_first_act(exts);
3141 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3146 EXPORT_SYMBOL(tcf_exts_dump_stats);
3148 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3150 if (*flags & TCA_CLS_FLAGS_IN_HW)
3152 *flags |= TCA_CLS_FLAGS_IN_HW;
3153 atomic_inc(&block->offloadcnt);
3156 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3158 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3160 *flags &= ~TCA_CLS_FLAGS_IN_HW;
3161 atomic_dec(&block->offloadcnt);
3164 static void tc_cls_offload_cnt_update(struct tcf_block *block,
3165 struct tcf_proto *tp, u32 *cnt,
3166 u32 *flags, u32 diff, bool add)
3168 lockdep_assert_held(&block->cb_lock);
3170 spin_lock(&tp->lock);
3173 tcf_block_offload_inc(block, flags);
3178 tcf_block_offload_dec(block, flags);
3180 spin_unlock(&tp->lock);
3184 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3185 u32 *cnt, u32 *flags)
3187 lockdep_assert_held(&block->cb_lock);
3189 spin_lock(&tp->lock);
3190 tcf_block_offload_dec(block, flags);
3192 spin_unlock(&tp->lock);
3196 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3197 void *type_data, bool err_stop)
3199 struct flow_block_cb *block_cb;
3203 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3204 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3215 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3216 void *type_data, bool err_stop, bool rtnl_held)
3218 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3224 down_read(&block->cb_lock);
3225 /* Need to obtain rtnl lock if block is bound to devs that require it.
3226 * In block bind code cb_lock is obtained while holding rtnl, so we must
3227 * obtain the locks in same order here.
3229 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3230 up_read(&block->cb_lock);
3235 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3237 up_read(&block->cb_lock);
3242 EXPORT_SYMBOL(tc_setup_cb_call);
3244 /* Non-destructive filter add. If filter that wasn't already in hardware is
3245 * successfully offloaded, increment block offloads counter. On failure,
3246 * previously offloaded filter is considered to be intact and offloads counter
3247 * is not decremented.
3250 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3251 enum tc_setup_type type, void *type_data, bool err_stop,
3252 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3254 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3260 down_read(&block->cb_lock);
3261 /* Need to obtain rtnl lock if block is bound to devs that require it.
3262 * In block bind code cb_lock is obtained while holding rtnl, so we must
3263 * obtain the locks in same order here.
3265 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3266 up_read(&block->cb_lock);
3271 /* Make sure all netdevs sharing this block are offload-capable. */
3272 if (block->nooffloaddevcnt && err_stop) {
3273 ok_count = -EOPNOTSUPP;
3277 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3281 if (tp->ops->hw_add)
3282 tp->ops->hw_add(tp, type_data);
3284 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3287 up_read(&block->cb_lock);
3290 return ok_count < 0 ? ok_count : 0;
3292 EXPORT_SYMBOL(tc_setup_cb_add);
3294 /* Destructive filter replace. If filter that wasn't already in hardware is
3295 * successfully offloaded, increment block offload counter. On failure,
3296 * previously offloaded filter is considered to be destroyed and offload counter
3300 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3301 enum tc_setup_type type, void *type_data, bool err_stop,
3302 u32 *old_flags, unsigned int *old_in_hw_count,
3303 u32 *new_flags, unsigned int *new_in_hw_count,
3306 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3312 down_read(&block->cb_lock);
3313 /* Need to obtain rtnl lock if block is bound to devs that require it.
3314 * In block bind code cb_lock is obtained while holding rtnl, so we must
3315 * obtain the locks in same order here.
3317 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3318 up_read(&block->cb_lock);
3323 /* Make sure all netdevs sharing this block are offload-capable. */
3324 if (block->nooffloaddevcnt && err_stop) {
3325 ok_count = -EOPNOTSUPP;
3329 tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3330 if (tp->ops->hw_del)
3331 tp->ops->hw_del(tp, type_data);
3333 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3337 if (tp->ops->hw_add)
3338 tp->ops->hw_add(tp, type_data);
3340 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3341 new_flags, ok_count, true);
3343 up_read(&block->cb_lock);
3346 return ok_count < 0 ? ok_count : 0;
3348 EXPORT_SYMBOL(tc_setup_cb_replace);
3350 /* Destroy filter and decrement block offload counter, if filter was previously
3354 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3355 enum tc_setup_type type, void *type_data, bool err_stop,
3356 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3358 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3364 down_read(&block->cb_lock);
3365 /* Need to obtain rtnl lock if block is bound to devs that require it.
3366 * In block bind code cb_lock is obtained while holding rtnl, so we must
3367 * obtain the locks in same order here.
3369 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3370 up_read(&block->cb_lock);
3375 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3377 tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3378 if (tp->ops->hw_del)
3379 tp->ops->hw_del(tp, type_data);
3381 up_read(&block->cb_lock);
3384 return ok_count < 0 ? ok_count : 0;
3386 EXPORT_SYMBOL(tc_setup_cb_destroy);
3388 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3389 bool add, flow_setup_cb_t *cb,
3390 enum tc_setup_type type, void *type_data,
3391 void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3393 int err = cb(type, type_data, cb_priv);
3396 if (add && tc_skip_sw(*flags))
3399 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3405 EXPORT_SYMBOL(tc_setup_cb_reoffload);
3407 void tc_cleanup_flow_action(struct flow_action *flow_action)
3409 struct flow_action_entry *entry;
3412 flow_action_for_each(i, entry, flow_action)
3413 if (entry->destructor)
3414 entry->destructor(entry->destructor_priv);
3416 EXPORT_SYMBOL(tc_cleanup_flow_action);
3418 static void tcf_mirred_get_dev(struct flow_action_entry *entry,
3419 const struct tc_action *act)
3421 #ifdef CONFIG_NET_CLS_ACT
3422 entry->dev = act->ops->get_dev(act, &entry->destructor);
3425 entry->destructor_priv = entry->dev;
3429 static void tcf_tunnel_encap_put_tunnel(void *priv)
3431 struct ip_tunnel_info *tunnel = priv;
3436 static int tcf_tunnel_encap_get_tunnel(struct flow_action_entry *entry,
3437 const struct tc_action *act)
3439 entry->tunnel = tcf_tunnel_info_copy(act);
3442 entry->destructor = tcf_tunnel_encap_put_tunnel;
3443 entry->destructor_priv = entry->tunnel;
3447 static void tcf_sample_get_group(struct flow_action_entry *entry,
3448 const struct tc_action *act)
3450 #ifdef CONFIG_NET_CLS_ACT
3451 entry->sample.psample_group =
3452 act->ops->get_psample_group(act, &entry->destructor);
3453 entry->destructor_priv = entry->sample.psample_group;
3457 int tc_setup_flow_action(struct flow_action *flow_action,
3458 const struct tcf_exts *exts, bool rtnl_held)
3460 const struct tc_action *act;
3461 int i, j, k, err = 0;
3470 tcf_exts_for_each_action(i, act, exts) {
3471 struct flow_action_entry *entry;
3473 entry = &flow_action->entries[j];
3474 if (is_tcf_gact_ok(act)) {
3475 entry->id = FLOW_ACTION_ACCEPT;
3476 } else if (is_tcf_gact_shot(act)) {
3477 entry->id = FLOW_ACTION_DROP;
3478 } else if (is_tcf_gact_trap(act)) {
3479 entry->id = FLOW_ACTION_TRAP;
3480 } else if (is_tcf_gact_goto_chain(act)) {
3481 entry->id = FLOW_ACTION_GOTO;
3482 entry->chain_index = tcf_gact_goto_chain_index(act);
3483 } else if (is_tcf_mirred_egress_redirect(act)) {
3484 entry->id = FLOW_ACTION_REDIRECT;
3485 tcf_mirred_get_dev(entry, act);
3486 } else if (is_tcf_mirred_egress_mirror(act)) {
3487 entry->id = FLOW_ACTION_MIRRED;
3488 tcf_mirred_get_dev(entry, act);
3489 } else if (is_tcf_mirred_ingress_redirect(act)) {
3490 entry->id = FLOW_ACTION_REDIRECT_INGRESS;
3491 tcf_mirred_get_dev(entry, act);
3492 } else if (is_tcf_mirred_ingress_mirror(act)) {
3493 entry->id = FLOW_ACTION_MIRRED_INGRESS;
3494 tcf_mirred_get_dev(entry, act);
3495 } else if (is_tcf_vlan(act)) {
3496 switch (tcf_vlan_action(act)) {
3497 case TCA_VLAN_ACT_PUSH:
3498 entry->id = FLOW_ACTION_VLAN_PUSH;
3499 entry->vlan.vid = tcf_vlan_push_vid(act);
3500 entry->vlan.proto = tcf_vlan_push_proto(act);
3501 entry->vlan.prio = tcf_vlan_push_prio(act);
3503 case TCA_VLAN_ACT_POP:
3504 entry->id = FLOW_ACTION_VLAN_POP;
3506 case TCA_VLAN_ACT_MODIFY:
3507 entry->id = FLOW_ACTION_VLAN_MANGLE;
3508 entry->vlan.vid = tcf_vlan_push_vid(act);
3509 entry->vlan.proto = tcf_vlan_push_proto(act);
3510 entry->vlan.prio = tcf_vlan_push_prio(act);
3516 } else if (is_tcf_tunnel_set(act)) {
3517 entry->id = FLOW_ACTION_TUNNEL_ENCAP;
3518 err = tcf_tunnel_encap_get_tunnel(entry, act);
3521 } else if (is_tcf_tunnel_release(act)) {
3522 entry->id = FLOW_ACTION_TUNNEL_DECAP;
3523 } else if (is_tcf_pedit(act)) {
3524 for (k = 0; k < tcf_pedit_nkeys(act); k++) {
3525 switch (tcf_pedit_cmd(act, k)) {
3526 case TCA_PEDIT_KEY_EX_CMD_SET:
3527 entry->id = FLOW_ACTION_MANGLE;
3529 case TCA_PEDIT_KEY_EX_CMD_ADD:
3530 entry->id = FLOW_ACTION_ADD;
3536 entry->mangle.htype = tcf_pedit_htype(act, k);
3537 entry->mangle.mask = tcf_pedit_mask(act, k);
3538 entry->mangle.val = tcf_pedit_val(act, k);
3539 entry->mangle.offset = tcf_pedit_offset(act, k);
3540 entry = &flow_action->entries[++j];
3542 } else if (is_tcf_csum(act)) {
3543 entry->id = FLOW_ACTION_CSUM;
3544 entry->csum_flags = tcf_csum_update_flags(act);
3545 } else if (is_tcf_skbedit_mark(act)) {
3546 entry->id = FLOW_ACTION_MARK;
3547 entry->mark = tcf_skbedit_mark(act);
3548 } else if (is_tcf_sample(act)) {
3549 entry->id = FLOW_ACTION_SAMPLE;
3550 entry->sample.trunc_size = tcf_sample_trunc_size(act);
3551 entry->sample.truncate = tcf_sample_truncate(act);
3552 entry->sample.rate = tcf_sample_rate(act);
3553 tcf_sample_get_group(entry, act);
3554 } else if (is_tcf_police(act)) {
3555 entry->id = FLOW_ACTION_POLICE;
3556 entry->police.burst = tcf_police_tcfp_burst(act);
3557 entry->police.rate_bytes_ps =
3558 tcf_police_rate_bytes_ps(act);
3559 } else if (is_tcf_ct(act)) {
3560 entry->id = FLOW_ACTION_CT;
3561 entry->ct.action = tcf_ct_action(act);
3562 entry->ct.zone = tcf_ct_zone(act);
3563 } else if (is_tcf_mpls(act)) {
3564 switch (tcf_mpls_action(act)) {
3565 case TCA_MPLS_ACT_PUSH:
3566 entry->id = FLOW_ACTION_MPLS_PUSH;
3567 entry->mpls_push.proto = tcf_mpls_proto(act);
3568 entry->mpls_push.label = tcf_mpls_label(act);
3569 entry->mpls_push.tc = tcf_mpls_tc(act);
3570 entry->mpls_push.bos = tcf_mpls_bos(act);
3571 entry->mpls_push.ttl = tcf_mpls_ttl(act);
3573 case TCA_MPLS_ACT_POP:
3574 entry->id = FLOW_ACTION_MPLS_POP;
3575 entry->mpls_pop.proto = tcf_mpls_proto(act);
3577 case TCA_MPLS_ACT_MODIFY:
3578 entry->id = FLOW_ACTION_MPLS_MANGLE;
3579 entry->mpls_mangle.label = tcf_mpls_label(act);
3580 entry->mpls_mangle.tc = tcf_mpls_tc(act);
3581 entry->mpls_mangle.bos = tcf_mpls_bos(act);
3582 entry->mpls_mangle.ttl = tcf_mpls_ttl(act);
3587 } else if (is_tcf_skbedit_ptype(act)) {
3588 entry->id = FLOW_ACTION_PTYPE;
3589 entry->ptype = tcf_skbedit_ptype(act);
3595 if (!is_tcf_pedit(act))
3604 tc_cleanup_flow_action(flow_action);
3608 EXPORT_SYMBOL(tc_setup_flow_action);
3610 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3612 unsigned int num_acts = 0;
3613 struct tc_action *act;
3616 tcf_exts_for_each_action(i, act, exts) {
3617 if (is_tcf_pedit(act))
3618 num_acts += tcf_pedit_nkeys(act);
3624 EXPORT_SYMBOL(tcf_exts_num_actions);
3626 static __net_init int tcf_net_init(struct net *net)
3628 struct tcf_net *tn = net_generic(net, tcf_net_id);
3630 spin_lock_init(&tn->idr_lock);
3635 static void __net_exit tcf_net_exit(struct net *net)
3637 struct tcf_net *tn = net_generic(net, tcf_net_id);
3639 idr_destroy(&tn->idr);
3642 static struct pernet_operations tcf_net_ops = {
3643 .init = tcf_net_init,
3644 .exit = tcf_net_exit,
3646 .size = sizeof(struct tcf_net),
3649 static struct flow_indr_block_entry block_entry = {
3650 .cb = tc_indr_block_get_and_cmd,
3651 .list = LIST_HEAD_INIT(block_entry.list),
3654 static int __init tc_filter_init(void)
3658 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3662 err = register_pernet_subsys(&tcf_net_ops);
3664 goto err_register_pernet_subsys;
3666 flow_indr_add_block_cb(&block_entry);
3668 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3669 RTNL_FLAG_DOIT_UNLOCKED);
3670 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3671 RTNL_FLAG_DOIT_UNLOCKED);
3672 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3673 tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3674 rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3675 rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3676 rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3681 err_register_pernet_subsys:
3682 destroy_workqueue(tc_filter_wq);
3686 subsys_initcall(tc_filter_init);