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 bool tcf_proto_check_delete(struct tcf_proto *tp)
313 if (tp->ops->delete_empty)
314 return tp->ops->delete_empty(tp);
320 static void tcf_proto_mark_delete(struct tcf_proto *tp)
322 spin_lock(&tp->lock);
324 spin_unlock(&tp->lock);
327 static bool tcf_proto_is_deleting(struct tcf_proto *tp)
331 spin_lock(&tp->lock);
332 deleting = tp->deleting;
333 spin_unlock(&tp->lock);
338 #define ASSERT_BLOCK_LOCKED(block) \
339 lockdep_assert_held(&(block)->lock)
341 struct tcf_filter_chain_list_item {
342 struct list_head list;
343 tcf_chain_head_change_t *chain_head_change;
344 void *chain_head_change_priv;
347 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
350 struct tcf_chain *chain;
352 ASSERT_BLOCK_LOCKED(block);
354 chain = kzalloc(sizeof(*chain), GFP_KERNEL);
357 list_add_tail(&chain->list, &block->chain_list);
358 mutex_init(&chain->filter_chain_lock);
359 chain->block = block;
360 chain->index = chain_index;
363 block->chain0.chain = chain;
367 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
368 struct tcf_proto *tp_head)
370 if (item->chain_head_change)
371 item->chain_head_change(tp_head, item->chain_head_change_priv);
374 static void tcf_chain0_head_change(struct tcf_chain *chain,
375 struct tcf_proto *tp_head)
377 struct tcf_filter_chain_list_item *item;
378 struct tcf_block *block = chain->block;
383 mutex_lock(&block->lock);
384 list_for_each_entry(item, &block->chain0.filter_chain_list, list)
385 tcf_chain_head_change_item(item, tp_head);
386 mutex_unlock(&block->lock);
389 /* Returns true if block can be safely freed. */
391 static bool tcf_chain_detach(struct tcf_chain *chain)
393 struct tcf_block *block = chain->block;
395 ASSERT_BLOCK_LOCKED(block);
397 list_del(&chain->list);
399 block->chain0.chain = NULL;
401 if (list_empty(&block->chain_list) &&
402 refcount_read(&block->refcnt) == 0)
408 static void tcf_block_destroy(struct tcf_block *block)
410 mutex_destroy(&block->lock);
411 mutex_destroy(&block->proto_destroy_lock);
412 kfree_rcu(block, rcu);
415 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
417 struct tcf_block *block = chain->block;
419 mutex_destroy(&chain->filter_chain_lock);
420 kfree_rcu(chain, rcu);
422 tcf_block_destroy(block);
425 static void tcf_chain_hold(struct tcf_chain *chain)
427 ASSERT_BLOCK_LOCKED(chain->block);
432 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
434 ASSERT_BLOCK_LOCKED(chain->block);
436 /* In case all the references are action references, this
437 * chain should not be shown to the user.
439 return chain->refcnt == chain->action_refcnt;
442 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
445 struct tcf_chain *chain;
447 ASSERT_BLOCK_LOCKED(block);
449 list_for_each_entry(chain, &block->chain_list, list) {
450 if (chain->index == chain_index)
456 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
457 u32 seq, u16 flags, int event, bool unicast);
459 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
460 u32 chain_index, bool create,
463 struct tcf_chain *chain = NULL;
464 bool is_first_reference;
466 mutex_lock(&block->lock);
467 chain = tcf_chain_lookup(block, chain_index);
469 tcf_chain_hold(chain);
473 chain = tcf_chain_create(block, chain_index);
479 ++chain->action_refcnt;
480 is_first_reference = chain->refcnt - chain->action_refcnt == 1;
481 mutex_unlock(&block->lock);
483 /* Send notification only in case we got the first
484 * non-action reference. Until then, the chain acts only as
485 * a placeholder for actions pointing to it and user ought
486 * not know about them.
488 if (is_first_reference && !by_act)
489 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
490 RTM_NEWCHAIN, false);
495 mutex_unlock(&block->lock);
499 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
502 return __tcf_chain_get(block, chain_index, create, false);
505 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
507 return __tcf_chain_get(block, chain_index, true, true);
509 EXPORT_SYMBOL(tcf_chain_get_by_act);
511 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
513 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
514 void *tmplt_priv, u32 chain_index,
515 struct tcf_block *block, struct sk_buff *oskb,
516 u32 seq, u16 flags, bool unicast);
518 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
519 bool explicitly_created)
521 struct tcf_block *block = chain->block;
522 const struct tcf_proto_ops *tmplt_ops;
523 bool free_block = false;
527 mutex_lock(&block->lock);
528 if (explicitly_created) {
529 if (!chain->explicitly_created) {
530 mutex_unlock(&block->lock);
533 chain->explicitly_created = false;
537 chain->action_refcnt--;
539 /* tc_chain_notify_delete can't be called while holding block lock.
540 * However, when block is unlocked chain can be changed concurrently, so
541 * save these to temporary variables.
543 refcnt = --chain->refcnt;
544 tmplt_ops = chain->tmplt_ops;
545 tmplt_priv = chain->tmplt_priv;
547 /* The last dropped non-action reference will trigger notification. */
548 if (refcnt - chain->action_refcnt == 0 && !by_act) {
549 tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
550 block, NULL, 0, 0, false);
551 /* Last reference to chain, no need to lock. */
552 chain->flushing = false;
556 free_block = tcf_chain_detach(chain);
557 mutex_unlock(&block->lock);
560 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
561 tcf_chain_destroy(chain, free_block);
565 static void tcf_chain_put(struct tcf_chain *chain)
567 __tcf_chain_put(chain, false, false);
570 void tcf_chain_put_by_act(struct tcf_chain *chain)
572 __tcf_chain_put(chain, true, false);
574 EXPORT_SYMBOL(tcf_chain_put_by_act);
576 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
578 __tcf_chain_put(chain, false, true);
581 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
583 struct tcf_proto *tp, *tp_next;
585 mutex_lock(&chain->filter_chain_lock);
586 tp = tcf_chain_dereference(chain->filter_chain, chain);
588 tp_next = rcu_dereference_protected(tp->next, 1);
589 tcf_proto_signal_destroying(chain, tp);
592 tp = tcf_chain_dereference(chain->filter_chain, chain);
593 RCU_INIT_POINTER(chain->filter_chain, NULL);
594 tcf_chain0_head_change(chain, NULL);
595 chain->flushing = true;
596 mutex_unlock(&chain->filter_chain_lock);
599 tp_next = rcu_dereference_protected(tp->next, 1);
600 tcf_proto_put(tp, rtnl_held, NULL);
605 static int tcf_block_setup(struct tcf_block *block,
606 struct flow_block_offload *bo);
608 static void tc_indr_block_cmd(struct net_device *dev, struct tcf_block *block,
609 flow_indr_block_bind_cb_t *cb, void *cb_priv,
610 enum flow_block_command command, bool ingress)
612 struct flow_block_offload bo = {
614 .binder_type = ingress ?
615 FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS :
616 FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS,
618 .block_shared = tcf_block_non_null_shared(block),
620 INIT_LIST_HEAD(&bo.cb_list);
625 bo.block = &block->flow_block;
627 down_write(&block->cb_lock);
628 cb(dev, cb_priv, TC_SETUP_BLOCK, &bo);
630 tcf_block_setup(block, &bo);
631 up_write(&block->cb_lock);
634 static struct tcf_block *tc_dev_block(struct net_device *dev, bool ingress)
636 const struct Qdisc_class_ops *cops;
637 const struct Qdisc_ops *ops;
640 if (!dev_ingress_queue(dev))
643 qdisc = dev_ingress_queue(dev)->qdisc_sleeping;
651 if (!ingress && !strcmp("ingress", ops->id))
658 if (!cops->tcf_block)
661 return cops->tcf_block(qdisc,
662 ingress ? TC_H_MIN_INGRESS : TC_H_MIN_EGRESS,
666 static void tc_indr_block_get_and_cmd(struct net_device *dev,
667 flow_indr_block_bind_cb_t *cb,
669 enum flow_block_command command)
671 struct tcf_block *block;
673 block = tc_dev_block(dev, true);
674 tc_indr_block_cmd(dev, block, cb, cb_priv, command, true);
676 block = tc_dev_block(dev, false);
677 tc_indr_block_cmd(dev, block, cb, cb_priv, command, false);
680 static void tc_indr_block_call(struct tcf_block *block,
681 struct net_device *dev,
682 struct tcf_block_ext_info *ei,
683 enum flow_block_command command,
684 struct netlink_ext_ack *extack)
686 struct flow_block_offload bo = {
688 .binder_type = ei->binder_type,
690 .block = &block->flow_block,
691 .block_shared = tcf_block_shared(block),
694 INIT_LIST_HEAD(&bo.cb_list);
696 flow_indr_block_call(dev, &bo, command);
697 tcf_block_setup(block, &bo);
700 static bool tcf_block_offload_in_use(struct tcf_block *block)
702 return atomic_read(&block->offloadcnt);
705 static int tcf_block_offload_cmd(struct tcf_block *block,
706 struct net_device *dev,
707 struct tcf_block_ext_info *ei,
708 enum flow_block_command command,
709 struct netlink_ext_ack *extack)
711 struct flow_block_offload bo = {};
714 bo.net = dev_net(dev);
715 bo.command = command;
716 bo.binder_type = ei->binder_type;
717 bo.block = &block->flow_block;
718 bo.block_shared = tcf_block_shared(block);
720 INIT_LIST_HEAD(&bo.cb_list);
722 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
726 return tcf_block_setup(block, &bo);
729 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
730 struct tcf_block_ext_info *ei,
731 struct netlink_ext_ack *extack)
733 struct net_device *dev = q->dev_queue->dev;
736 down_write(&block->cb_lock);
737 if (!dev->netdev_ops->ndo_setup_tc)
738 goto no_offload_dev_inc;
740 /* If tc offload feature is disabled and the block we try to bind
741 * to already has some offloaded filters, forbid to bind.
743 if (!tc_can_offload(dev) && tcf_block_offload_in_use(block)) {
744 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
749 err = tcf_block_offload_cmd(block, dev, ei, FLOW_BLOCK_BIND, extack);
750 if (err == -EOPNOTSUPP)
751 goto no_offload_dev_inc;
755 tc_indr_block_call(block, dev, ei, FLOW_BLOCK_BIND, extack);
756 up_write(&block->cb_lock);
760 if (tcf_block_offload_in_use(block)) {
765 block->nooffloaddevcnt++;
766 tc_indr_block_call(block, dev, ei, FLOW_BLOCK_BIND, extack);
768 up_write(&block->cb_lock);
772 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
773 struct tcf_block_ext_info *ei)
775 struct net_device *dev = q->dev_queue->dev;
778 down_write(&block->cb_lock);
779 tc_indr_block_call(block, dev, ei, FLOW_BLOCK_UNBIND, NULL);
781 if (!dev->netdev_ops->ndo_setup_tc)
782 goto no_offload_dev_dec;
783 err = tcf_block_offload_cmd(block, dev, ei, FLOW_BLOCK_UNBIND, NULL);
784 if (err == -EOPNOTSUPP)
785 goto no_offload_dev_dec;
786 up_write(&block->cb_lock);
790 WARN_ON(block->nooffloaddevcnt-- == 0);
791 up_write(&block->cb_lock);
795 tcf_chain0_head_change_cb_add(struct tcf_block *block,
796 struct tcf_block_ext_info *ei,
797 struct netlink_ext_ack *extack)
799 struct tcf_filter_chain_list_item *item;
800 struct tcf_chain *chain0;
802 item = kmalloc(sizeof(*item), GFP_KERNEL);
804 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
807 item->chain_head_change = ei->chain_head_change;
808 item->chain_head_change_priv = ei->chain_head_change_priv;
810 mutex_lock(&block->lock);
811 chain0 = block->chain0.chain;
813 tcf_chain_hold(chain0);
815 list_add(&item->list, &block->chain0.filter_chain_list);
816 mutex_unlock(&block->lock);
819 struct tcf_proto *tp_head;
821 mutex_lock(&chain0->filter_chain_lock);
823 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
825 tcf_chain_head_change_item(item, tp_head);
827 mutex_lock(&block->lock);
828 list_add(&item->list, &block->chain0.filter_chain_list);
829 mutex_unlock(&block->lock);
831 mutex_unlock(&chain0->filter_chain_lock);
832 tcf_chain_put(chain0);
839 tcf_chain0_head_change_cb_del(struct tcf_block *block,
840 struct tcf_block_ext_info *ei)
842 struct tcf_filter_chain_list_item *item;
844 mutex_lock(&block->lock);
845 list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
846 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
847 (item->chain_head_change == ei->chain_head_change &&
848 item->chain_head_change_priv == ei->chain_head_change_priv)) {
849 if (block->chain0.chain)
850 tcf_chain_head_change_item(item, NULL);
851 list_del(&item->list);
852 mutex_unlock(&block->lock);
858 mutex_unlock(&block->lock);
863 spinlock_t idr_lock; /* Protects idr */
867 static unsigned int tcf_net_id;
869 static int tcf_block_insert(struct tcf_block *block, struct net *net,
870 struct netlink_ext_ack *extack)
872 struct tcf_net *tn = net_generic(net, tcf_net_id);
875 idr_preload(GFP_KERNEL);
876 spin_lock(&tn->idr_lock);
877 err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
879 spin_unlock(&tn->idr_lock);
885 static void tcf_block_remove(struct tcf_block *block, struct net *net)
887 struct tcf_net *tn = net_generic(net, tcf_net_id);
889 spin_lock(&tn->idr_lock);
890 idr_remove(&tn->idr, block->index);
891 spin_unlock(&tn->idr_lock);
894 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
896 struct netlink_ext_ack *extack)
898 struct tcf_block *block;
900 block = kzalloc(sizeof(*block), GFP_KERNEL);
902 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
903 return ERR_PTR(-ENOMEM);
905 mutex_init(&block->lock);
906 mutex_init(&block->proto_destroy_lock);
907 init_rwsem(&block->cb_lock);
908 flow_block_init(&block->flow_block);
909 INIT_LIST_HEAD(&block->chain_list);
910 INIT_LIST_HEAD(&block->owner_list);
911 INIT_LIST_HEAD(&block->chain0.filter_chain_list);
913 refcount_set(&block->refcnt, 1);
915 block->index = block_index;
917 /* Don't store q pointer for blocks which are shared */
918 if (!tcf_block_shared(block))
923 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
925 struct tcf_net *tn = net_generic(net, tcf_net_id);
927 return idr_find(&tn->idr, block_index);
930 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
932 struct tcf_block *block;
935 block = tcf_block_lookup(net, block_index);
936 if (block && !refcount_inc_not_zero(&block->refcnt))
943 static struct tcf_chain *
944 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
946 mutex_lock(&block->lock);
948 chain = list_is_last(&chain->list, &block->chain_list) ?
949 NULL : list_next_entry(chain, list);
951 chain = list_first_entry_or_null(&block->chain_list,
952 struct tcf_chain, list);
954 /* skip all action-only chains */
955 while (chain && tcf_chain_held_by_acts_only(chain))
956 chain = list_is_last(&chain->list, &block->chain_list) ?
957 NULL : list_next_entry(chain, list);
960 tcf_chain_hold(chain);
961 mutex_unlock(&block->lock);
966 /* Function to be used by all clients that want to iterate over all chains on
967 * block. It properly obtains block->lock and takes reference to chain before
968 * returning it. Users of this function must be tolerant to concurrent chain
969 * insertion/deletion or ensure that no concurrent chain modification is
970 * possible. Note that all netlink dump callbacks cannot guarantee to provide
971 * consistent dump because rtnl lock is released each time skb is filled with
972 * data and sent to user-space.
976 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
978 struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
981 tcf_chain_put(chain);
985 EXPORT_SYMBOL(tcf_get_next_chain);
987 static struct tcf_proto *
988 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
993 mutex_lock(&chain->filter_chain_lock);
996 tp = tcf_chain_dereference(chain->filter_chain, chain);
997 } else if (tcf_proto_is_deleting(tp)) {
998 /* 'deleting' flag is set and chain->filter_chain_lock was
999 * unlocked, which means next pointer could be invalid. Restart
1002 prio = tp->prio + 1;
1003 tp = tcf_chain_dereference(chain->filter_chain, chain);
1005 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1006 if (!tp->deleting && tp->prio >= prio)
1009 tp = tcf_chain_dereference(tp->next, chain);
1015 mutex_unlock(&chain->filter_chain_lock);
1020 /* Function to be used by all clients that want to iterate over all tp's on
1021 * chain. Users of this function must be tolerant to concurrent tp
1022 * insertion/deletion or ensure that no concurrent chain modification is
1023 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1024 * consistent dump because rtnl lock is released each time skb is filled with
1025 * data and sent to user-space.
1029 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp,
1032 struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1035 tcf_proto_put(tp, rtnl_held, NULL);
1039 EXPORT_SYMBOL(tcf_get_next_proto);
1041 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1043 struct tcf_chain *chain;
1045 /* Last reference to block. At this point chains cannot be added or
1046 * removed concurrently.
1048 for (chain = tcf_get_next_chain(block, NULL);
1050 chain = tcf_get_next_chain(block, chain)) {
1051 tcf_chain_put_explicitly_created(chain);
1052 tcf_chain_flush(chain, rtnl_held);
1056 /* Lookup Qdisc and increments its reference counter.
1057 * Set parent, if necessary.
1060 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1061 u32 *parent, int ifindex, bool rtnl_held,
1062 struct netlink_ext_ack *extack)
1064 const struct Qdisc_class_ops *cops;
1065 struct net_device *dev;
1068 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1074 dev = dev_get_by_index_rcu(net, ifindex);
1083 *parent = (*q)->handle;
1085 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1087 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1093 *q = qdisc_refcount_inc_nz(*q);
1095 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1100 /* Is it classful? */
1101 cops = (*q)->ops->cl_ops;
1103 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1108 if (!cops->tcf_block) {
1109 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1115 /* At this point we know that qdisc is not noop_qdisc,
1116 * which means that qdisc holds a reference to net_device
1117 * and we hold a reference to qdisc, so it is safe to release
1129 qdisc_put_unlocked(*q);
1135 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1136 int ifindex, struct netlink_ext_ack *extack)
1138 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1141 /* Do we search for filter, attached to class? */
1142 if (TC_H_MIN(parent)) {
1143 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1145 *cl = cops->find(q, parent);
1147 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1155 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1156 unsigned long cl, int ifindex,
1158 struct netlink_ext_ack *extack)
1160 struct tcf_block *block;
1162 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1163 block = tcf_block_refcnt_get(net, block_index);
1165 NL_SET_ERR_MSG(extack, "Block of given index was not found");
1166 return ERR_PTR(-EINVAL);
1169 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1171 block = cops->tcf_block(q, cl, extack);
1173 return ERR_PTR(-EINVAL);
1175 if (tcf_block_shared(block)) {
1176 NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1177 return ERR_PTR(-EOPNOTSUPP);
1180 /* Always take reference to block in order to support execution
1181 * of rules update path of cls API without rtnl lock. Caller
1182 * must release block when it is finished using it. 'if' block
1183 * of this conditional obtain reference to block by calling
1184 * tcf_block_refcnt_get().
1186 refcount_inc(&block->refcnt);
1192 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1193 struct tcf_block_ext_info *ei, bool rtnl_held)
1195 if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1196 /* Flushing/putting all chains will cause the block to be
1197 * deallocated when last chain is freed. However, if chain_list
1198 * is empty, block has to be manually deallocated. After block
1199 * reference counter reached 0, it is no longer possible to
1200 * increment it or add new chains to block.
1202 bool free_block = list_empty(&block->chain_list);
1204 mutex_unlock(&block->lock);
1205 if (tcf_block_shared(block))
1206 tcf_block_remove(block, block->net);
1209 tcf_block_offload_unbind(block, q, ei);
1212 tcf_block_destroy(block);
1214 tcf_block_flush_all_chains(block, rtnl_held);
1216 tcf_block_offload_unbind(block, q, ei);
1220 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1222 __tcf_block_put(block, NULL, NULL, rtnl_held);
1226 * Set q, parent, cl when appropriate.
1229 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1230 u32 *parent, unsigned long *cl,
1231 int ifindex, u32 block_index,
1232 struct netlink_ext_ack *extack)
1234 struct tcf_block *block;
1239 err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1243 err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1247 block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1248 if (IS_ERR(block)) {
1249 err = PTR_ERR(block);
1260 return ERR_PTR(err);
1263 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1266 if (!IS_ERR_OR_NULL(block))
1267 tcf_block_refcnt_put(block, rtnl_held);
1273 qdisc_put_unlocked(q);
1277 struct tcf_block_owner_item {
1278 struct list_head list;
1280 enum flow_block_binder_type binder_type;
1284 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1286 enum flow_block_binder_type binder_type)
1288 if (block->keep_dst &&
1289 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1290 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1291 netif_keep_dst(qdisc_dev(q));
1294 void tcf_block_netif_keep_dst(struct tcf_block *block)
1296 struct tcf_block_owner_item *item;
1298 block->keep_dst = true;
1299 list_for_each_entry(item, &block->owner_list, list)
1300 tcf_block_owner_netif_keep_dst(block, item->q,
1303 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1305 static int tcf_block_owner_add(struct tcf_block *block,
1307 enum flow_block_binder_type binder_type)
1309 struct tcf_block_owner_item *item;
1311 item = kmalloc(sizeof(*item), GFP_KERNEL);
1315 item->binder_type = binder_type;
1316 list_add(&item->list, &block->owner_list);
1320 static void tcf_block_owner_del(struct tcf_block *block,
1322 enum flow_block_binder_type binder_type)
1324 struct tcf_block_owner_item *item;
1326 list_for_each_entry(item, &block->owner_list, list) {
1327 if (item->q == q && item->binder_type == binder_type) {
1328 list_del(&item->list);
1336 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1337 struct tcf_block_ext_info *ei,
1338 struct netlink_ext_ack *extack)
1340 struct net *net = qdisc_net(q);
1341 struct tcf_block *block = NULL;
1344 if (ei->block_index)
1345 /* block_index not 0 means the shared block is requested */
1346 block = tcf_block_refcnt_get(net, ei->block_index);
1349 block = tcf_block_create(net, q, ei->block_index, extack);
1351 return PTR_ERR(block);
1352 if (tcf_block_shared(block)) {
1353 err = tcf_block_insert(block, net, extack);
1355 goto err_block_insert;
1359 err = tcf_block_owner_add(block, q, ei->binder_type);
1361 goto err_block_owner_add;
1363 tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1365 err = tcf_chain0_head_change_cb_add(block, ei, extack);
1367 goto err_chain0_head_change_cb_add;
1369 err = tcf_block_offload_bind(block, q, ei, extack);
1371 goto err_block_offload_bind;
1376 err_block_offload_bind:
1377 tcf_chain0_head_change_cb_del(block, ei);
1378 err_chain0_head_change_cb_add:
1379 tcf_block_owner_del(block, q, ei->binder_type);
1380 err_block_owner_add:
1382 tcf_block_refcnt_put(block, true);
1385 EXPORT_SYMBOL(tcf_block_get_ext);
1387 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1389 struct tcf_proto __rcu **p_filter_chain = priv;
1391 rcu_assign_pointer(*p_filter_chain, tp_head);
1394 int tcf_block_get(struct tcf_block **p_block,
1395 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1396 struct netlink_ext_ack *extack)
1398 struct tcf_block_ext_info ei = {
1399 .chain_head_change = tcf_chain_head_change_dflt,
1400 .chain_head_change_priv = p_filter_chain,
1403 WARN_ON(!p_filter_chain);
1404 return tcf_block_get_ext(p_block, q, &ei, extack);
1406 EXPORT_SYMBOL(tcf_block_get);
1408 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1409 * actions should be all removed after flushing.
1411 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1412 struct tcf_block_ext_info *ei)
1416 tcf_chain0_head_change_cb_del(block, ei);
1417 tcf_block_owner_del(block, q, ei->binder_type);
1419 __tcf_block_put(block, q, ei, true);
1421 EXPORT_SYMBOL(tcf_block_put_ext);
1423 void tcf_block_put(struct tcf_block *block)
1425 struct tcf_block_ext_info ei = {0, };
1429 tcf_block_put_ext(block, block->q, &ei);
1432 EXPORT_SYMBOL(tcf_block_put);
1435 tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1436 void *cb_priv, bool add, bool offload_in_use,
1437 struct netlink_ext_ack *extack)
1439 struct tcf_chain *chain, *chain_prev;
1440 struct tcf_proto *tp, *tp_prev;
1443 lockdep_assert_held(&block->cb_lock);
1445 for (chain = __tcf_get_next_chain(block, NULL);
1448 chain = __tcf_get_next_chain(block, chain),
1449 tcf_chain_put(chain_prev)) {
1450 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1452 tp = __tcf_get_next_proto(chain, tp),
1453 tcf_proto_put(tp_prev, true, NULL)) {
1454 if (tp->ops->reoffload) {
1455 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1458 goto err_playback_remove;
1459 } else if (add && offload_in_use) {
1461 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1462 goto err_playback_remove;
1469 err_playback_remove:
1470 tcf_proto_put(tp, true, NULL);
1471 tcf_chain_put(chain);
1472 tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1477 static int tcf_block_bind(struct tcf_block *block,
1478 struct flow_block_offload *bo)
1480 struct flow_block_cb *block_cb, *next;
1483 lockdep_assert_held(&block->cb_lock);
1485 list_for_each_entry(block_cb, &bo->cb_list, list) {
1486 err = tcf_block_playback_offloads(block, block_cb->cb,
1487 block_cb->cb_priv, true,
1488 tcf_block_offload_in_use(block),
1492 if (!bo->unlocked_driver_cb)
1493 block->lockeddevcnt++;
1497 list_splice(&bo->cb_list, &block->flow_block.cb_list);
1502 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1504 list_del(&block_cb->list);
1505 tcf_block_playback_offloads(block, block_cb->cb,
1506 block_cb->cb_priv, false,
1507 tcf_block_offload_in_use(block),
1509 if (!bo->unlocked_driver_cb)
1510 block->lockeddevcnt--;
1512 flow_block_cb_free(block_cb);
1518 static void tcf_block_unbind(struct tcf_block *block,
1519 struct flow_block_offload *bo)
1521 struct flow_block_cb *block_cb, *next;
1523 lockdep_assert_held(&block->cb_lock);
1525 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1526 tcf_block_playback_offloads(block, block_cb->cb,
1527 block_cb->cb_priv, false,
1528 tcf_block_offload_in_use(block),
1530 list_del(&block_cb->list);
1531 flow_block_cb_free(block_cb);
1532 if (!bo->unlocked_driver_cb)
1533 block->lockeddevcnt--;
1537 static int tcf_block_setup(struct tcf_block *block,
1538 struct flow_block_offload *bo)
1542 switch (bo->command) {
1543 case FLOW_BLOCK_BIND:
1544 err = tcf_block_bind(block, bo);
1546 case FLOW_BLOCK_UNBIND:
1548 tcf_block_unbind(block, bo);
1558 /* Main classifier routine: scans classifier chain attached
1559 * to this qdisc, (optionally) tests for protocol and asks
1560 * specific classifiers.
1562 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1563 struct tcf_result *res, bool compat_mode)
1565 #ifdef CONFIG_NET_CLS_ACT
1566 const int max_reclassify_loop = 4;
1567 const struct tcf_proto *orig_tp = tp;
1568 const struct tcf_proto *first_tp;
1573 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1574 __be16 protocol = tc_skb_protocol(skb);
1577 if (tp->protocol != protocol &&
1578 tp->protocol != htons(ETH_P_ALL))
1581 err = tp->classify(skb, tp, res);
1582 #ifdef CONFIG_NET_CLS_ACT
1583 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1586 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1587 first_tp = res->goto_tp;
1589 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1591 struct tc_skb_ext *ext;
1593 ext = skb_ext_add(skb, TC_SKB_EXT);
1594 if (WARN_ON_ONCE(!ext))
1597 ext->chain = err & TC_ACT_EXT_VAL_MASK;
1607 return TC_ACT_UNSPEC; /* signal: continue lookup */
1608 #ifdef CONFIG_NET_CLS_ACT
1610 if (unlikely(limit++ >= max_reclassify_loop)) {
1611 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1612 tp->chain->block->index,
1614 ntohs(tp->protocol));
1622 EXPORT_SYMBOL(tcf_classify);
1624 struct tcf_chain_info {
1625 struct tcf_proto __rcu **pprev;
1626 struct tcf_proto __rcu *next;
1629 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1630 struct tcf_chain_info *chain_info)
1632 return tcf_chain_dereference(*chain_info->pprev, chain);
1635 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1636 struct tcf_chain_info *chain_info,
1637 struct tcf_proto *tp)
1639 if (chain->flushing)
1642 if (*chain_info->pprev == chain->filter_chain)
1643 tcf_chain0_head_change(chain, tp);
1645 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1646 rcu_assign_pointer(*chain_info->pprev, tp);
1651 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1652 struct tcf_chain_info *chain_info,
1653 struct tcf_proto *tp)
1655 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1657 tcf_proto_mark_delete(tp);
1658 if (tp == chain->filter_chain)
1659 tcf_chain0_head_change(chain, next);
1660 RCU_INIT_POINTER(*chain_info->pprev, next);
1663 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1664 struct tcf_chain_info *chain_info,
1665 u32 protocol, u32 prio,
1666 bool prio_allocate);
1668 /* Try to insert new proto.
1669 * If proto with specified priority already exists, free new proto
1670 * and return existing one.
1673 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1674 struct tcf_proto *tp_new,
1675 u32 protocol, u32 prio,
1678 struct tcf_chain_info chain_info;
1679 struct tcf_proto *tp;
1682 mutex_lock(&chain->filter_chain_lock);
1684 if (tcf_proto_exists_destroying(chain, tp_new)) {
1685 mutex_unlock(&chain->filter_chain_lock);
1686 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1687 return ERR_PTR(-EAGAIN);
1690 tp = tcf_chain_tp_find(chain, &chain_info,
1691 protocol, prio, false);
1693 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1694 mutex_unlock(&chain->filter_chain_lock);
1697 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1700 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1701 tp_new = ERR_PTR(err);
1707 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1708 struct tcf_proto *tp, bool rtnl_held,
1709 struct netlink_ext_ack *extack)
1711 struct tcf_chain_info chain_info;
1712 struct tcf_proto *tp_iter;
1713 struct tcf_proto **pprev;
1714 struct tcf_proto *next;
1716 mutex_lock(&chain->filter_chain_lock);
1718 /* Atomically find and remove tp from chain. */
1719 for (pprev = &chain->filter_chain;
1720 (tp_iter = tcf_chain_dereference(*pprev, chain));
1721 pprev = &tp_iter->next) {
1722 if (tp_iter == tp) {
1723 chain_info.pprev = pprev;
1724 chain_info.next = tp_iter->next;
1725 WARN_ON(tp_iter->deleting);
1729 /* Verify that tp still exists and no new filters were inserted
1731 * Mark tp for deletion if it is empty.
1733 if (!tp_iter || !tcf_proto_check_delete(tp)) {
1734 mutex_unlock(&chain->filter_chain_lock);
1738 tcf_proto_signal_destroying(chain, tp);
1739 next = tcf_chain_dereference(chain_info.next, chain);
1740 if (tp == chain->filter_chain)
1741 tcf_chain0_head_change(chain, next);
1742 RCU_INIT_POINTER(*chain_info.pprev, next);
1743 mutex_unlock(&chain->filter_chain_lock);
1745 tcf_proto_put(tp, rtnl_held, extack);
1748 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1749 struct tcf_chain_info *chain_info,
1750 u32 protocol, u32 prio,
1753 struct tcf_proto **pprev;
1754 struct tcf_proto *tp;
1756 /* Check the chain for existence of proto-tcf with this priority */
1757 for (pprev = &chain->filter_chain;
1758 (tp = tcf_chain_dereference(*pprev, chain));
1759 pprev = &tp->next) {
1760 if (tp->prio >= prio) {
1761 if (tp->prio == prio) {
1762 if (prio_allocate ||
1763 (tp->protocol != protocol && protocol))
1764 return ERR_PTR(-EINVAL);
1771 chain_info->pprev = pprev;
1773 chain_info->next = tp->next;
1776 chain_info->next = NULL;
1781 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1782 struct tcf_proto *tp, struct tcf_block *block,
1783 struct Qdisc *q, u32 parent, void *fh,
1784 u32 portid, u32 seq, u16 flags, int event,
1788 struct nlmsghdr *nlh;
1789 unsigned char *b = skb_tail_pointer(skb);
1791 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1793 goto out_nlmsg_trim;
1794 tcm = nlmsg_data(nlh);
1795 tcm->tcm_family = AF_UNSPEC;
1799 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1800 tcm->tcm_parent = parent;
1802 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1803 tcm->tcm_block_index = block->index;
1805 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1806 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1807 goto nla_put_failure;
1808 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1809 goto nla_put_failure;
1811 tcm->tcm_handle = 0;
1813 if (tp->ops->dump &&
1814 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
1815 goto nla_put_failure;
1817 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1826 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1827 struct nlmsghdr *n, struct tcf_proto *tp,
1828 struct tcf_block *block, struct Qdisc *q,
1829 u32 parent, void *fh, int event, bool unicast,
1832 struct sk_buff *skb;
1833 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1836 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1840 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1841 n->nlmsg_seq, n->nlmsg_flags, event,
1848 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1850 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1851 n->nlmsg_flags & NLM_F_ECHO);
1858 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1859 struct nlmsghdr *n, struct tcf_proto *tp,
1860 struct tcf_block *block, struct Qdisc *q,
1861 u32 parent, void *fh, bool unicast, bool *last,
1862 bool rtnl_held, struct netlink_ext_ack *extack)
1864 struct sk_buff *skb;
1865 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1868 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1872 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1873 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
1875 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1880 err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
1887 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1889 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1890 n->nlmsg_flags & NLM_F_ECHO);
1892 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1899 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1900 struct tcf_block *block, struct Qdisc *q,
1901 u32 parent, struct nlmsghdr *n,
1902 struct tcf_chain *chain, int event,
1905 struct tcf_proto *tp;
1907 for (tp = tcf_get_next_proto(chain, NULL, rtnl_held);
1908 tp; tp = tcf_get_next_proto(chain, tp, rtnl_held))
1909 tfilter_notify(net, oskb, n, tp, block,
1910 q, parent, NULL, event, false, rtnl_held);
1913 static void tfilter_put(struct tcf_proto *tp, void *fh)
1915 if (tp->ops->put && fh)
1916 tp->ops->put(tp, fh);
1919 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1920 struct netlink_ext_ack *extack)
1922 struct net *net = sock_net(skb->sk);
1923 struct nlattr *tca[TCA_MAX + 1];
1924 char name[IFNAMSIZ];
1931 struct Qdisc *q = NULL;
1932 struct tcf_chain_info chain_info;
1933 struct tcf_chain *chain = NULL;
1934 struct tcf_block *block;
1935 struct tcf_proto *tp;
1940 bool rtnl_held = false;
1942 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1948 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
1949 rtm_tca_policy, extack);
1954 protocol = TC_H_MIN(t->tcm_info);
1955 prio = TC_H_MAJ(t->tcm_info);
1956 prio_allocate = false;
1957 parent = t->tcm_parent;
1963 /* If no priority is provided by the user,
1966 if (n->nlmsg_flags & NLM_F_CREATE) {
1967 prio = TC_H_MAKE(0x80000000U, 0U);
1968 prio_allocate = true;
1970 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1975 /* Find head of filter chain. */
1977 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
1981 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
1982 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
1987 /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
1988 * block is shared (no qdisc found), qdisc is not unlocked, classifier
1989 * type is not specified, classifier is not unlocked.
1992 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
1993 !tcf_proto_is_unlocked(name)) {
1998 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2002 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2004 if (IS_ERR(block)) {
2005 err = PTR_ERR(block);
2009 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2010 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2011 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2015 chain = tcf_chain_get(block, chain_index, true);
2017 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2022 mutex_lock(&chain->filter_chain_lock);
2023 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2024 prio, prio_allocate);
2026 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2032 struct tcf_proto *tp_new = NULL;
2034 if (chain->flushing) {
2039 /* Proto-tcf does not exist, create new one */
2041 if (tca[TCA_KIND] == NULL || !protocol) {
2042 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2047 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2048 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2054 prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2057 mutex_unlock(&chain->filter_chain_lock);
2058 tp_new = tcf_proto_create(nla_data(tca[TCA_KIND]),
2059 protocol, prio, chain, rtnl_held,
2061 if (IS_ERR(tp_new)) {
2062 err = PTR_ERR(tp_new);
2067 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2074 mutex_unlock(&chain->filter_chain_lock);
2077 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2078 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2083 fh = tp->ops->get(tp, t->tcm_handle);
2086 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2087 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2091 } else if (n->nlmsg_flags & NLM_F_EXCL) {
2092 tfilter_put(tp, fh);
2093 NL_SET_ERR_MSG(extack, "Filter already exists");
2098 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2099 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2104 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2105 n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
2108 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2109 RTM_NEWTFILTER, false, rtnl_held);
2110 tfilter_put(tp, fh);
2111 /* q pointer is NULL for shared blocks */
2113 q->flags &= ~TCQ_F_CAN_BYPASS;
2117 if (err && tp_created)
2118 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2121 if (tp && !IS_ERR(tp))
2122 tcf_proto_put(tp, rtnl_held, NULL);
2124 tcf_chain_put(chain);
2126 tcf_block_release(q, block, rtnl_held);
2131 if (err == -EAGAIN) {
2132 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2136 /* Replay the request. */
2142 mutex_unlock(&chain->filter_chain_lock);
2146 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2147 struct netlink_ext_ack *extack)
2149 struct net *net = sock_net(skb->sk);
2150 struct nlattr *tca[TCA_MAX + 1];
2151 char name[IFNAMSIZ];
2157 struct Qdisc *q = NULL;
2158 struct tcf_chain_info chain_info;
2159 struct tcf_chain *chain = NULL;
2160 struct tcf_block *block = NULL;
2161 struct tcf_proto *tp = NULL;
2162 unsigned long cl = 0;
2165 bool rtnl_held = false;
2167 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2170 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2171 rtm_tca_policy, extack);
2176 protocol = TC_H_MIN(t->tcm_info);
2177 prio = TC_H_MAJ(t->tcm_info);
2178 parent = t->tcm_parent;
2180 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2181 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2185 /* Find head of filter chain. */
2187 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2191 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2192 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2196 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2197 * found), qdisc is not unlocked, classifier type is not specified,
2198 * classifier is not unlocked.
2201 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2202 !tcf_proto_is_unlocked(name)) {
2207 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2211 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2213 if (IS_ERR(block)) {
2214 err = PTR_ERR(block);
2218 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2219 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2220 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2224 chain = tcf_chain_get(block, chain_index, false);
2226 /* User requested flush on non-existent chain. Nothing to do,
2227 * so just return success.
2233 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2239 tfilter_notify_chain(net, skb, block, q, parent, n,
2240 chain, RTM_DELTFILTER, rtnl_held);
2241 tcf_chain_flush(chain, rtnl_held);
2246 mutex_lock(&chain->filter_chain_lock);
2247 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2249 if (!tp || IS_ERR(tp)) {
2250 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2251 err = tp ? PTR_ERR(tp) : -ENOENT;
2253 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2254 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2257 } else if (t->tcm_handle == 0) {
2258 tcf_proto_signal_destroying(chain, tp);
2259 tcf_chain_tp_remove(chain, &chain_info, tp);
2260 mutex_unlock(&chain->filter_chain_lock);
2262 tcf_proto_put(tp, rtnl_held, NULL);
2263 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2264 RTM_DELTFILTER, false, rtnl_held);
2268 mutex_unlock(&chain->filter_chain_lock);
2270 fh = tp->ops->get(tp, t->tcm_handle);
2273 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2278 err = tfilter_del_notify(net, skb, n, tp, block,
2279 q, parent, fh, false, &last,
2285 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2290 if (tp && !IS_ERR(tp))
2291 tcf_proto_put(tp, rtnl_held, NULL);
2292 tcf_chain_put(chain);
2294 tcf_block_release(q, block, rtnl_held);
2302 mutex_unlock(&chain->filter_chain_lock);
2306 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2307 struct netlink_ext_ack *extack)
2309 struct net *net = sock_net(skb->sk);
2310 struct nlattr *tca[TCA_MAX + 1];
2311 char name[IFNAMSIZ];
2317 struct Qdisc *q = NULL;
2318 struct tcf_chain_info chain_info;
2319 struct tcf_chain *chain = NULL;
2320 struct tcf_block *block = NULL;
2321 struct tcf_proto *tp = NULL;
2322 unsigned long cl = 0;
2325 bool rtnl_held = false;
2327 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2328 rtm_tca_policy, extack);
2333 protocol = TC_H_MIN(t->tcm_info);
2334 prio = TC_H_MAJ(t->tcm_info);
2335 parent = t->tcm_parent;
2338 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2342 /* Find head of filter chain. */
2344 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2348 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2349 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2353 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2354 * unlocked, classifier type is not specified, classifier is not
2357 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2358 !tcf_proto_is_unlocked(name)) {
2363 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2367 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2369 if (IS_ERR(block)) {
2370 err = PTR_ERR(block);
2374 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2375 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2376 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2380 chain = tcf_chain_get(block, chain_index, false);
2382 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2387 mutex_lock(&chain->filter_chain_lock);
2388 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2390 mutex_unlock(&chain->filter_chain_lock);
2391 if (!tp || IS_ERR(tp)) {
2392 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2393 err = tp ? PTR_ERR(tp) : -ENOENT;
2395 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2396 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2401 fh = tp->ops->get(tp, t->tcm_handle);
2404 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2407 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2408 fh, RTM_NEWTFILTER, true, rtnl_held);
2410 NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2413 tfilter_put(tp, fh);
2416 if (tp && !IS_ERR(tp))
2417 tcf_proto_put(tp, rtnl_held, NULL);
2418 tcf_chain_put(chain);
2420 tcf_block_release(q, block, rtnl_held);
2428 struct tcf_dump_args {
2429 struct tcf_walker w;
2430 struct sk_buff *skb;
2431 struct netlink_callback *cb;
2432 struct tcf_block *block;
2437 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2439 struct tcf_dump_args *a = (void *)arg;
2440 struct net *net = sock_net(a->skb->sk);
2442 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2443 n, NETLINK_CB(a->cb->skb).portid,
2444 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2445 RTM_NEWTFILTER, true);
2448 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2449 struct sk_buff *skb, struct netlink_callback *cb,
2450 long index_start, long *p_index)
2452 struct net *net = sock_net(skb->sk);
2453 struct tcf_block *block = chain->block;
2454 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2455 struct tcf_proto *tp, *tp_prev;
2456 struct tcf_dump_args arg;
2458 for (tp = __tcf_get_next_proto(chain, NULL);
2461 tp = __tcf_get_next_proto(chain, tp),
2462 tcf_proto_put(tp_prev, true, NULL),
2464 if (*p_index < index_start)
2466 if (TC_H_MAJ(tcm->tcm_info) &&
2467 TC_H_MAJ(tcm->tcm_info) != tp->prio)
2469 if (TC_H_MIN(tcm->tcm_info) &&
2470 TC_H_MIN(tcm->tcm_info) != tp->protocol)
2472 if (*p_index > index_start)
2473 memset(&cb->args[1], 0,
2474 sizeof(cb->args) - sizeof(cb->args[0]));
2475 if (cb->args[1] == 0) {
2476 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2477 NETLINK_CB(cb->skb).portid,
2478 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2479 RTM_NEWTFILTER, true) <= 0)
2485 arg.w.fn = tcf_node_dump;
2490 arg.parent = parent;
2492 arg.w.skip = cb->args[1] - 1;
2494 arg.w.cookie = cb->args[2];
2495 tp->ops->walk(tp, &arg.w, true);
2496 cb->args[2] = arg.w.cookie;
2497 cb->args[1] = arg.w.count + 1;
2504 tcf_proto_put(tp, true, NULL);
2508 /* called with RTNL */
2509 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2511 struct tcf_chain *chain, *chain_prev;
2512 struct net *net = sock_net(skb->sk);
2513 struct nlattr *tca[TCA_MAX + 1];
2514 struct Qdisc *q = NULL;
2515 struct tcf_block *block;
2516 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2522 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2525 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2530 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2531 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2534 /* If we work with block index, q is NULL and parent value
2535 * will never be used in the following code. The check
2536 * in tcf_fill_node prevents it. However, compiler does not
2537 * see that far, so set parent to zero to silence the warning
2538 * about parent being uninitialized.
2542 const struct Qdisc_class_ops *cops;
2543 struct net_device *dev;
2544 unsigned long cl = 0;
2546 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2550 parent = tcm->tcm_parent;
2555 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2559 cops = q->ops->cl_ops;
2562 if (!cops->tcf_block)
2564 if (TC_H_MIN(tcm->tcm_parent)) {
2565 cl = cops->find(q, tcm->tcm_parent);
2569 block = cops->tcf_block(q, cl, NULL);
2572 if (tcf_block_shared(block))
2576 index_start = cb->args[0];
2579 for (chain = __tcf_get_next_chain(block, NULL);
2582 chain = __tcf_get_next_chain(block, chain),
2583 tcf_chain_put(chain_prev)) {
2584 if (tca[TCA_CHAIN] &&
2585 nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2587 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2588 index_start, &index)) {
2589 tcf_chain_put(chain);
2595 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2596 tcf_block_refcnt_put(block, true);
2597 cb->args[0] = index;
2600 /* If we did no progress, the error (EMSGSIZE) is real */
2601 if (skb->len == 0 && err)
2606 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2607 void *tmplt_priv, u32 chain_index,
2608 struct net *net, struct sk_buff *skb,
2609 struct tcf_block *block,
2610 u32 portid, u32 seq, u16 flags, int event)
2612 unsigned char *b = skb_tail_pointer(skb);
2613 const struct tcf_proto_ops *ops;
2614 struct nlmsghdr *nlh;
2621 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2623 goto out_nlmsg_trim;
2624 tcm = nlmsg_data(nlh);
2625 tcm->tcm_family = AF_UNSPEC;
2628 tcm->tcm_handle = 0;
2630 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2631 tcm->tcm_parent = block->q->handle;
2633 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2634 tcm->tcm_block_index = block->index;
2637 if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2638 goto nla_put_failure;
2641 if (nla_put_string(skb, TCA_KIND, ops->kind))
2642 goto nla_put_failure;
2643 if (ops->tmplt_dump(skb, net, priv) < 0)
2644 goto nla_put_failure;
2647 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2656 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2657 u32 seq, u16 flags, int event, bool unicast)
2659 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2660 struct tcf_block *block = chain->block;
2661 struct net *net = block->net;
2662 struct sk_buff *skb;
2665 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2669 if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2670 chain->index, net, skb, block, portid,
2671 seq, flags, event) <= 0) {
2677 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2679 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2680 flags & NLM_F_ECHO);
2687 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2688 void *tmplt_priv, u32 chain_index,
2689 struct tcf_block *block, struct sk_buff *oskb,
2690 u32 seq, u16 flags, bool unicast)
2692 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2693 struct net *net = block->net;
2694 struct sk_buff *skb;
2696 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2700 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2701 block, portid, seq, flags, RTM_DELCHAIN) <= 0) {
2707 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2709 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2712 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2713 struct nlattr **tca,
2714 struct netlink_ext_ack *extack)
2716 const struct tcf_proto_ops *ops;
2717 char name[IFNAMSIZ];
2720 /* If kind is not set, user did not specify template. */
2724 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2725 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
2729 ops = tcf_proto_lookup_ops(name, true, extack);
2731 return PTR_ERR(ops);
2732 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2733 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2737 tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2738 if (IS_ERR(tmplt_priv)) {
2739 module_put(ops->owner);
2740 return PTR_ERR(tmplt_priv);
2742 chain->tmplt_ops = ops;
2743 chain->tmplt_priv = tmplt_priv;
2747 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2750 /* If template ops are set, no work to do for us. */
2754 tmplt_ops->tmplt_destroy(tmplt_priv);
2755 module_put(tmplt_ops->owner);
2758 /* Add/delete/get a chain */
2760 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2761 struct netlink_ext_ack *extack)
2763 struct net *net = sock_net(skb->sk);
2764 struct nlattr *tca[TCA_MAX + 1];
2768 struct Qdisc *q = NULL;
2769 struct tcf_chain *chain = NULL;
2770 struct tcf_block *block;
2774 if (n->nlmsg_type != RTM_GETCHAIN &&
2775 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2779 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2780 rtm_tca_policy, extack);
2785 parent = t->tcm_parent;
2788 block = tcf_block_find(net, &q, &parent, &cl,
2789 t->tcm_ifindex, t->tcm_block_index, extack);
2791 return PTR_ERR(block);
2793 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2794 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2795 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2800 mutex_lock(&block->lock);
2801 chain = tcf_chain_lookup(block, chain_index);
2802 if (n->nlmsg_type == RTM_NEWCHAIN) {
2804 if (tcf_chain_held_by_acts_only(chain)) {
2805 /* The chain exists only because there is
2806 * some action referencing it.
2808 tcf_chain_hold(chain);
2810 NL_SET_ERR_MSG(extack, "Filter chain already exists");
2812 goto errout_block_locked;
2815 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2816 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
2818 goto errout_block_locked;
2820 chain = tcf_chain_create(block, chain_index);
2822 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
2824 goto errout_block_locked;
2828 if (!chain || tcf_chain_held_by_acts_only(chain)) {
2829 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2831 goto errout_block_locked;
2833 tcf_chain_hold(chain);
2836 if (n->nlmsg_type == RTM_NEWCHAIN) {
2837 /* Modifying chain requires holding parent block lock. In case
2838 * the chain was successfully added, take a reference to the
2839 * chain. This ensures that an empty chain does not disappear at
2840 * the end of this function.
2842 tcf_chain_hold(chain);
2843 chain->explicitly_created = true;
2845 mutex_unlock(&block->lock);
2847 switch (n->nlmsg_type) {
2849 err = tc_chain_tmplt_add(chain, net, tca, extack);
2851 tcf_chain_put_explicitly_created(chain);
2855 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2856 RTM_NEWCHAIN, false);
2859 tfilter_notify_chain(net, skb, block, q, parent, n,
2860 chain, RTM_DELTFILTER, true);
2861 /* Flush the chain first as the user requested chain removal. */
2862 tcf_chain_flush(chain, true);
2863 /* In case the chain was successfully deleted, put a reference
2864 * to the chain previously taken during addition.
2866 tcf_chain_put_explicitly_created(chain);
2869 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
2870 n->nlmsg_seq, n->nlmsg_type, true);
2872 NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2876 NL_SET_ERR_MSG(extack, "Unsupported message type");
2881 tcf_chain_put(chain);
2883 tcf_block_release(q, block, true);
2885 /* Replay the request. */
2889 errout_block_locked:
2890 mutex_unlock(&block->lock);
2894 /* called with RTNL */
2895 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2897 struct net *net = sock_net(skb->sk);
2898 struct nlattr *tca[TCA_MAX + 1];
2899 struct Qdisc *q = NULL;
2900 struct tcf_block *block;
2901 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2902 struct tcf_chain *chain;
2908 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2911 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2912 rtm_tca_policy, cb->extack);
2916 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2917 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2920 /* If we work with block index, q is NULL and parent value
2921 * will never be used in the following code. The check
2922 * in tcf_fill_node prevents it. However, compiler does not
2923 * see that far, so set parent to zero to silence the warning
2924 * about parent being uninitialized.
2928 const struct Qdisc_class_ops *cops;
2929 struct net_device *dev;
2930 unsigned long cl = 0;
2932 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2936 parent = tcm->tcm_parent;
2941 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2945 cops = q->ops->cl_ops;
2948 if (!cops->tcf_block)
2950 if (TC_H_MIN(tcm->tcm_parent)) {
2951 cl = cops->find(q, tcm->tcm_parent);
2955 block = cops->tcf_block(q, cl, NULL);
2958 if (tcf_block_shared(block))
2962 index_start = cb->args[0];
2965 mutex_lock(&block->lock);
2966 list_for_each_entry(chain, &block->chain_list, list) {
2967 if ((tca[TCA_CHAIN] &&
2968 nla_get_u32(tca[TCA_CHAIN]) != chain->index))
2970 if (index < index_start) {
2974 if (tcf_chain_held_by_acts_only(chain))
2976 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2977 chain->index, net, skb, block,
2978 NETLINK_CB(cb->skb).portid,
2979 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2985 mutex_unlock(&block->lock);
2987 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2988 tcf_block_refcnt_put(block, true);
2989 cb->args[0] = index;
2992 /* If we did no progress, the error (EMSGSIZE) is real */
2993 if (skb->len == 0 && err)
2998 void tcf_exts_destroy(struct tcf_exts *exts)
3000 #ifdef CONFIG_NET_CLS_ACT
3001 if (exts->actions) {
3002 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3003 kfree(exts->actions);
3005 exts->nr_actions = 0;
3008 EXPORT_SYMBOL(tcf_exts_destroy);
3010 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3011 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
3012 bool rtnl_held, struct netlink_ext_ack *extack)
3014 #ifdef CONFIG_NET_CLS_ACT
3016 struct tc_action *act;
3017 size_t attr_size = 0;
3019 if (exts->police && tb[exts->police]) {
3020 act = tcf_action_init_1(net, tp, tb[exts->police],
3021 rate_tlv, "police", ovr,
3022 TCA_ACT_BIND, rtnl_held,
3025 return PTR_ERR(act);
3027 act->type = exts->type = TCA_OLD_COMPAT;
3028 exts->actions[0] = act;
3029 exts->nr_actions = 1;
3030 } else if (exts->action && tb[exts->action]) {
3033 err = tcf_action_init(net, tp, tb[exts->action],
3034 rate_tlv, NULL, ovr, TCA_ACT_BIND,
3035 exts->actions, &attr_size,
3039 exts->nr_actions = err;
3043 if ((exts->action && tb[exts->action]) ||
3044 (exts->police && tb[exts->police])) {
3045 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3052 EXPORT_SYMBOL(tcf_exts_validate);
3054 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3056 #ifdef CONFIG_NET_CLS_ACT
3057 struct tcf_exts old = *dst;
3060 tcf_exts_destroy(&old);
3063 EXPORT_SYMBOL(tcf_exts_change);
3065 #ifdef CONFIG_NET_CLS_ACT
3066 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3068 if (exts->nr_actions == 0)
3071 return exts->actions[0];
3075 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3077 #ifdef CONFIG_NET_CLS_ACT
3078 struct nlattr *nest;
3080 if (exts->action && tcf_exts_has_actions(exts)) {
3082 * again for backward compatible mode - we want
3083 * to work with both old and new modes of entering
3084 * tc data even if iproute2 was newer - jhs
3086 if (exts->type != TCA_OLD_COMPAT) {
3087 nest = nla_nest_start_noflag(skb, exts->action);
3089 goto nla_put_failure;
3091 if (tcf_action_dump(skb, exts->actions, 0, 0) < 0)
3092 goto nla_put_failure;
3093 nla_nest_end(skb, nest);
3094 } else if (exts->police) {
3095 struct tc_action *act = tcf_exts_first_act(exts);
3096 nest = nla_nest_start_noflag(skb, exts->police);
3097 if (nest == NULL || !act)
3098 goto nla_put_failure;
3099 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3100 goto nla_put_failure;
3101 nla_nest_end(skb, nest);
3107 nla_nest_cancel(skb, nest);
3113 EXPORT_SYMBOL(tcf_exts_dump);
3116 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3118 #ifdef CONFIG_NET_CLS_ACT
3119 struct tc_action *a = tcf_exts_first_act(exts);
3120 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3125 EXPORT_SYMBOL(tcf_exts_dump_stats);
3127 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3129 if (*flags & TCA_CLS_FLAGS_IN_HW)
3131 *flags |= TCA_CLS_FLAGS_IN_HW;
3132 atomic_inc(&block->offloadcnt);
3135 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3137 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3139 *flags &= ~TCA_CLS_FLAGS_IN_HW;
3140 atomic_dec(&block->offloadcnt);
3143 static void tc_cls_offload_cnt_update(struct tcf_block *block,
3144 struct tcf_proto *tp, u32 *cnt,
3145 u32 *flags, u32 diff, bool add)
3147 lockdep_assert_held(&block->cb_lock);
3149 spin_lock(&tp->lock);
3152 tcf_block_offload_inc(block, flags);
3157 tcf_block_offload_dec(block, flags);
3159 spin_unlock(&tp->lock);
3163 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3164 u32 *cnt, u32 *flags)
3166 lockdep_assert_held(&block->cb_lock);
3168 spin_lock(&tp->lock);
3169 tcf_block_offload_dec(block, flags);
3171 spin_unlock(&tp->lock);
3175 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3176 void *type_data, bool err_stop)
3178 struct flow_block_cb *block_cb;
3182 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3183 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3194 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3195 void *type_data, bool err_stop, bool rtnl_held)
3197 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3203 down_read(&block->cb_lock);
3204 /* Need to obtain rtnl lock if block is bound to devs that require it.
3205 * In block bind code cb_lock is obtained while holding rtnl, so we must
3206 * obtain the locks in same order here.
3208 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3209 up_read(&block->cb_lock);
3214 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3216 up_read(&block->cb_lock);
3221 EXPORT_SYMBOL(tc_setup_cb_call);
3223 /* Non-destructive filter add. If filter that wasn't already in hardware is
3224 * successfully offloaded, increment block offloads counter. On failure,
3225 * previously offloaded filter is considered to be intact and offloads counter
3226 * is not decremented.
3229 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3230 enum tc_setup_type type, void *type_data, bool err_stop,
3231 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3233 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3239 down_read(&block->cb_lock);
3240 /* Need to obtain rtnl lock if block is bound to devs that require it.
3241 * In block bind code cb_lock is obtained while holding rtnl, so we must
3242 * obtain the locks in same order here.
3244 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3245 up_read(&block->cb_lock);
3250 /* Make sure all netdevs sharing this block are offload-capable. */
3251 if (block->nooffloaddevcnt && err_stop) {
3252 ok_count = -EOPNOTSUPP;
3256 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3260 if (tp->ops->hw_add)
3261 tp->ops->hw_add(tp, type_data);
3263 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3266 up_read(&block->cb_lock);
3269 return ok_count < 0 ? ok_count : 0;
3271 EXPORT_SYMBOL(tc_setup_cb_add);
3273 /* Destructive filter replace. If filter that wasn't already in hardware is
3274 * successfully offloaded, increment block offload counter. On failure,
3275 * previously offloaded filter is considered to be destroyed and offload counter
3279 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3280 enum tc_setup_type type, void *type_data, bool err_stop,
3281 u32 *old_flags, unsigned int *old_in_hw_count,
3282 u32 *new_flags, unsigned int *new_in_hw_count,
3285 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3291 down_read(&block->cb_lock);
3292 /* Need to obtain rtnl lock if block is bound to devs that require it.
3293 * In block bind code cb_lock is obtained while holding rtnl, so we must
3294 * obtain the locks in same order here.
3296 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3297 up_read(&block->cb_lock);
3302 /* Make sure all netdevs sharing this block are offload-capable. */
3303 if (block->nooffloaddevcnt && err_stop) {
3304 ok_count = -EOPNOTSUPP;
3308 tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3309 if (tp->ops->hw_del)
3310 tp->ops->hw_del(tp, type_data);
3312 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3316 if (tp->ops->hw_add)
3317 tp->ops->hw_add(tp, type_data);
3319 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3320 new_flags, ok_count, true);
3322 up_read(&block->cb_lock);
3325 return ok_count < 0 ? ok_count : 0;
3327 EXPORT_SYMBOL(tc_setup_cb_replace);
3329 /* Destroy filter and decrement block offload counter, if filter was previously
3333 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3334 enum tc_setup_type type, void *type_data, bool err_stop,
3335 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3337 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3343 down_read(&block->cb_lock);
3344 /* Need to obtain rtnl lock if block is bound to devs that require it.
3345 * In block bind code cb_lock is obtained while holding rtnl, so we must
3346 * obtain the locks in same order here.
3348 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3349 up_read(&block->cb_lock);
3354 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3356 tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3357 if (tp->ops->hw_del)
3358 tp->ops->hw_del(tp, type_data);
3360 up_read(&block->cb_lock);
3363 return ok_count < 0 ? ok_count : 0;
3365 EXPORT_SYMBOL(tc_setup_cb_destroy);
3367 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3368 bool add, flow_setup_cb_t *cb,
3369 enum tc_setup_type type, void *type_data,
3370 void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3372 int err = cb(type, type_data, cb_priv);
3375 if (add && tc_skip_sw(*flags))
3378 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3384 EXPORT_SYMBOL(tc_setup_cb_reoffload);
3386 void tc_cleanup_flow_action(struct flow_action *flow_action)
3388 struct flow_action_entry *entry;
3391 flow_action_for_each(i, entry, flow_action)
3392 if (entry->destructor)
3393 entry->destructor(entry->destructor_priv);
3395 EXPORT_SYMBOL(tc_cleanup_flow_action);
3397 static void tcf_mirred_get_dev(struct flow_action_entry *entry,
3398 const struct tc_action *act)
3400 #ifdef CONFIG_NET_CLS_ACT
3401 entry->dev = act->ops->get_dev(act, &entry->destructor);
3404 entry->destructor_priv = entry->dev;
3408 static void tcf_tunnel_encap_put_tunnel(void *priv)
3410 struct ip_tunnel_info *tunnel = priv;
3415 static int tcf_tunnel_encap_get_tunnel(struct flow_action_entry *entry,
3416 const struct tc_action *act)
3418 entry->tunnel = tcf_tunnel_info_copy(act);
3421 entry->destructor = tcf_tunnel_encap_put_tunnel;
3422 entry->destructor_priv = entry->tunnel;
3426 static void tcf_sample_get_group(struct flow_action_entry *entry,
3427 const struct tc_action *act)
3429 #ifdef CONFIG_NET_CLS_ACT
3430 entry->sample.psample_group =
3431 act->ops->get_psample_group(act, &entry->destructor);
3432 entry->destructor_priv = entry->sample.psample_group;
3436 int tc_setup_flow_action(struct flow_action *flow_action,
3437 const struct tcf_exts *exts, bool rtnl_held)
3439 const struct tc_action *act;
3440 int i, j, k, err = 0;
3449 tcf_exts_for_each_action(i, act, exts) {
3450 struct flow_action_entry *entry;
3452 entry = &flow_action->entries[j];
3453 if (is_tcf_gact_ok(act)) {
3454 entry->id = FLOW_ACTION_ACCEPT;
3455 } else if (is_tcf_gact_shot(act)) {
3456 entry->id = FLOW_ACTION_DROP;
3457 } else if (is_tcf_gact_trap(act)) {
3458 entry->id = FLOW_ACTION_TRAP;
3459 } else if (is_tcf_gact_goto_chain(act)) {
3460 entry->id = FLOW_ACTION_GOTO;
3461 entry->chain_index = tcf_gact_goto_chain_index(act);
3462 } else if (is_tcf_mirred_egress_redirect(act)) {
3463 entry->id = FLOW_ACTION_REDIRECT;
3464 tcf_mirred_get_dev(entry, act);
3465 } else if (is_tcf_mirred_egress_mirror(act)) {
3466 entry->id = FLOW_ACTION_MIRRED;
3467 tcf_mirred_get_dev(entry, act);
3468 } else if (is_tcf_mirred_ingress_redirect(act)) {
3469 entry->id = FLOW_ACTION_REDIRECT_INGRESS;
3470 tcf_mirred_get_dev(entry, act);
3471 } else if (is_tcf_mirred_ingress_mirror(act)) {
3472 entry->id = FLOW_ACTION_MIRRED_INGRESS;
3473 tcf_mirred_get_dev(entry, act);
3474 } else if (is_tcf_vlan(act)) {
3475 switch (tcf_vlan_action(act)) {
3476 case TCA_VLAN_ACT_PUSH:
3477 entry->id = FLOW_ACTION_VLAN_PUSH;
3478 entry->vlan.vid = tcf_vlan_push_vid(act);
3479 entry->vlan.proto = tcf_vlan_push_proto(act);
3480 entry->vlan.prio = tcf_vlan_push_prio(act);
3482 case TCA_VLAN_ACT_POP:
3483 entry->id = FLOW_ACTION_VLAN_POP;
3485 case TCA_VLAN_ACT_MODIFY:
3486 entry->id = FLOW_ACTION_VLAN_MANGLE;
3487 entry->vlan.vid = tcf_vlan_push_vid(act);
3488 entry->vlan.proto = tcf_vlan_push_proto(act);
3489 entry->vlan.prio = tcf_vlan_push_prio(act);
3495 } else if (is_tcf_tunnel_set(act)) {
3496 entry->id = FLOW_ACTION_TUNNEL_ENCAP;
3497 err = tcf_tunnel_encap_get_tunnel(entry, act);
3500 } else if (is_tcf_tunnel_release(act)) {
3501 entry->id = FLOW_ACTION_TUNNEL_DECAP;
3502 } else if (is_tcf_pedit(act)) {
3503 for (k = 0; k < tcf_pedit_nkeys(act); k++) {
3504 switch (tcf_pedit_cmd(act, k)) {
3505 case TCA_PEDIT_KEY_EX_CMD_SET:
3506 entry->id = FLOW_ACTION_MANGLE;
3508 case TCA_PEDIT_KEY_EX_CMD_ADD:
3509 entry->id = FLOW_ACTION_ADD;
3515 entry->mangle.htype = tcf_pedit_htype(act, k);
3516 entry->mangle.mask = tcf_pedit_mask(act, k);
3517 entry->mangle.val = tcf_pedit_val(act, k);
3518 entry->mangle.offset = tcf_pedit_offset(act, k);
3519 entry = &flow_action->entries[++j];
3521 } else if (is_tcf_csum(act)) {
3522 entry->id = FLOW_ACTION_CSUM;
3523 entry->csum_flags = tcf_csum_update_flags(act);
3524 } else if (is_tcf_skbedit_mark(act)) {
3525 entry->id = FLOW_ACTION_MARK;
3526 entry->mark = tcf_skbedit_mark(act);
3527 } else if (is_tcf_sample(act)) {
3528 entry->id = FLOW_ACTION_SAMPLE;
3529 entry->sample.trunc_size = tcf_sample_trunc_size(act);
3530 entry->sample.truncate = tcf_sample_truncate(act);
3531 entry->sample.rate = tcf_sample_rate(act);
3532 tcf_sample_get_group(entry, act);
3533 } else if (is_tcf_police(act)) {
3534 entry->id = FLOW_ACTION_POLICE;
3535 entry->police.burst = tcf_police_tcfp_burst(act);
3536 entry->police.rate_bytes_ps =
3537 tcf_police_rate_bytes_ps(act);
3538 } else if (is_tcf_ct(act)) {
3539 entry->id = FLOW_ACTION_CT;
3540 entry->ct.action = tcf_ct_action(act);
3541 entry->ct.zone = tcf_ct_zone(act);
3542 } else if (is_tcf_mpls(act)) {
3543 switch (tcf_mpls_action(act)) {
3544 case TCA_MPLS_ACT_PUSH:
3545 entry->id = FLOW_ACTION_MPLS_PUSH;
3546 entry->mpls_push.proto = tcf_mpls_proto(act);
3547 entry->mpls_push.label = tcf_mpls_label(act);
3548 entry->mpls_push.tc = tcf_mpls_tc(act);
3549 entry->mpls_push.bos = tcf_mpls_bos(act);
3550 entry->mpls_push.ttl = tcf_mpls_ttl(act);
3552 case TCA_MPLS_ACT_POP:
3553 entry->id = FLOW_ACTION_MPLS_POP;
3554 entry->mpls_pop.proto = tcf_mpls_proto(act);
3556 case TCA_MPLS_ACT_MODIFY:
3557 entry->id = FLOW_ACTION_MPLS_MANGLE;
3558 entry->mpls_mangle.label = tcf_mpls_label(act);
3559 entry->mpls_mangle.tc = tcf_mpls_tc(act);
3560 entry->mpls_mangle.bos = tcf_mpls_bos(act);
3561 entry->mpls_mangle.ttl = tcf_mpls_ttl(act);
3566 } else if (is_tcf_skbedit_ptype(act)) {
3567 entry->id = FLOW_ACTION_PTYPE;
3568 entry->ptype = tcf_skbedit_ptype(act);
3574 if (!is_tcf_pedit(act))
3583 tc_cleanup_flow_action(flow_action);
3587 EXPORT_SYMBOL(tc_setup_flow_action);
3589 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3591 unsigned int num_acts = 0;
3592 struct tc_action *act;
3595 tcf_exts_for_each_action(i, act, exts) {
3596 if (is_tcf_pedit(act))
3597 num_acts += tcf_pedit_nkeys(act);
3603 EXPORT_SYMBOL(tcf_exts_num_actions);
3605 static __net_init int tcf_net_init(struct net *net)
3607 struct tcf_net *tn = net_generic(net, tcf_net_id);
3609 spin_lock_init(&tn->idr_lock);
3614 static void __net_exit tcf_net_exit(struct net *net)
3616 struct tcf_net *tn = net_generic(net, tcf_net_id);
3618 idr_destroy(&tn->idr);
3621 static struct pernet_operations tcf_net_ops = {
3622 .init = tcf_net_init,
3623 .exit = tcf_net_exit,
3625 .size = sizeof(struct tcf_net),
3628 static struct flow_indr_block_entry block_entry = {
3629 .cb = tc_indr_block_get_and_cmd,
3630 .list = LIST_HEAD_INIT(block_entry.list),
3633 static int __init tc_filter_init(void)
3637 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3641 err = register_pernet_subsys(&tcf_net_ops);
3643 goto err_register_pernet_subsys;
3645 flow_indr_add_block_cb(&block_entry);
3647 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3648 RTNL_FLAG_DOIT_UNLOCKED);
3649 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3650 RTNL_FLAG_DOIT_UNLOCKED);
3651 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3652 tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3653 rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3654 rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3655 rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3660 err_register_pernet_subsys:
3661 destroy_workqueue(tc_filter_wq);
3665 subsys_initcall(tc_filter_init);