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/jhash.h>
24 #include <linux/rculist.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/tc_act/tc_gate.h>
42 #include <net/flow_offload.h>
44 extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
46 /* The list of all installed classifier types */
47 static LIST_HEAD(tcf_proto_base);
49 /* Protects list of registered TC modules. It is pure SMP lock. */
50 static DEFINE_RWLOCK(cls_mod_lock);
52 static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
54 return jhash_3words(tp->chain->index, tp->prio,
55 (__force __u32)tp->protocol, 0);
58 static void tcf_proto_signal_destroying(struct tcf_chain *chain,
61 struct tcf_block *block = chain->block;
63 mutex_lock(&block->proto_destroy_lock);
64 hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
65 destroy_obj_hashfn(tp));
66 mutex_unlock(&block->proto_destroy_lock);
69 static bool tcf_proto_cmp(const struct tcf_proto *tp1,
70 const struct tcf_proto *tp2)
72 return tp1->chain->index == tp2->chain->index &&
73 tp1->prio == tp2->prio &&
74 tp1->protocol == tp2->protocol;
77 static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
80 u32 hash = destroy_obj_hashfn(tp);
81 struct tcf_proto *iter;
85 hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
86 destroy_ht_node, hash) {
87 if (tcf_proto_cmp(tp, iter)) {
98 tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
100 struct tcf_block *block = chain->block;
102 mutex_lock(&block->proto_destroy_lock);
103 if (hash_hashed(&tp->destroy_ht_node))
104 hash_del_rcu(&tp->destroy_ht_node);
105 mutex_unlock(&block->proto_destroy_lock);
108 /* Find classifier type by string name */
110 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
112 const struct tcf_proto_ops *t, *res = NULL;
115 read_lock(&cls_mod_lock);
116 list_for_each_entry(t, &tcf_proto_base, head) {
117 if (strcmp(kind, t->kind) == 0) {
118 if (try_module_get(t->owner))
123 read_unlock(&cls_mod_lock);
128 static const struct tcf_proto_ops *
129 tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
130 struct netlink_ext_ack *extack)
132 const struct tcf_proto_ops *ops;
134 ops = __tcf_proto_lookup_ops(kind);
137 #ifdef CONFIG_MODULES
140 request_module("cls_%s", kind);
143 ops = __tcf_proto_lookup_ops(kind);
144 /* We dropped the RTNL semaphore in order to perform
145 * the module load. So, even if we succeeded in loading
146 * the module we have to replay the request. We indicate
147 * this using -EAGAIN.
150 module_put(ops->owner);
151 return ERR_PTR(-EAGAIN);
154 NL_SET_ERR_MSG(extack, "TC classifier not found");
155 return ERR_PTR(-ENOENT);
158 /* Register(unregister) new classifier type */
160 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
162 struct tcf_proto_ops *t;
165 write_lock(&cls_mod_lock);
166 list_for_each_entry(t, &tcf_proto_base, head)
167 if (!strcmp(ops->kind, t->kind))
170 list_add_tail(&ops->head, &tcf_proto_base);
173 write_unlock(&cls_mod_lock);
176 EXPORT_SYMBOL(register_tcf_proto_ops);
178 static struct workqueue_struct *tc_filter_wq;
180 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
182 struct tcf_proto_ops *t;
185 /* Wait for outstanding call_rcu()s, if any, from a
186 * tcf_proto_ops's destroy() handler.
189 flush_workqueue(tc_filter_wq);
191 write_lock(&cls_mod_lock);
192 list_for_each_entry(t, &tcf_proto_base, head) {
199 write_unlock(&cls_mod_lock);
202 EXPORT_SYMBOL(unregister_tcf_proto_ops);
204 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
206 INIT_RCU_WORK(rwork, func);
207 return queue_rcu_work(tc_filter_wq, rwork);
209 EXPORT_SYMBOL(tcf_queue_work);
211 /* Select new prio value from the range, managed by kernel. */
213 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
215 u32 first = TC_H_MAKE(0xC0000000U, 0U);
218 first = tp->prio - 1;
220 return TC_H_MAJ(first);
223 static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
226 return nla_strscpy(name, kind, IFNAMSIZ) < 0;
227 memset(name, 0, IFNAMSIZ);
231 static bool tcf_proto_is_unlocked(const char *kind)
233 const struct tcf_proto_ops *ops;
236 if (strlen(kind) == 0)
239 ops = tcf_proto_lookup_ops(kind, false, NULL);
240 /* On error return false to take rtnl lock. Proto lookup/create
241 * functions will perform lookup again and properly handle errors.
246 ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
247 module_put(ops->owner);
251 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
252 u32 prio, struct tcf_chain *chain,
254 struct netlink_ext_ack *extack)
256 struct tcf_proto *tp;
259 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
261 return ERR_PTR(-ENOBUFS);
263 tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
264 if (IS_ERR(tp->ops)) {
265 err = PTR_ERR(tp->ops);
268 tp->classify = tp->ops->classify;
269 tp->protocol = protocol;
272 spin_lock_init(&tp->lock);
273 refcount_set(&tp->refcnt, 1);
275 err = tp->ops->init(tp);
277 module_put(tp->ops->owner);
287 static void tcf_proto_get(struct tcf_proto *tp)
289 refcount_inc(&tp->refcnt);
292 static void tcf_chain_put(struct tcf_chain *chain);
294 static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
295 bool sig_destroy, struct netlink_ext_ack *extack)
297 tp->ops->destroy(tp, rtnl_held, extack);
299 tcf_proto_signal_destroyed(tp->chain, tp);
300 tcf_chain_put(tp->chain);
301 module_put(tp->ops->owner);
305 static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
306 struct netlink_ext_ack *extack)
308 if (refcount_dec_and_test(&tp->refcnt))
309 tcf_proto_destroy(tp, rtnl_held, true, extack);
312 static bool tcf_proto_check_delete(struct tcf_proto *tp)
314 if (tp->ops->delete_empty)
315 return tp->ops->delete_empty(tp);
321 static void tcf_proto_mark_delete(struct tcf_proto *tp)
323 spin_lock(&tp->lock);
325 spin_unlock(&tp->lock);
328 static bool tcf_proto_is_deleting(struct tcf_proto *tp)
332 spin_lock(&tp->lock);
333 deleting = tp->deleting;
334 spin_unlock(&tp->lock);
339 #define ASSERT_BLOCK_LOCKED(block) \
340 lockdep_assert_held(&(block)->lock)
342 struct tcf_filter_chain_list_item {
343 struct list_head list;
344 tcf_chain_head_change_t *chain_head_change;
345 void *chain_head_change_priv;
348 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
351 struct tcf_chain *chain;
353 ASSERT_BLOCK_LOCKED(block);
355 chain = kzalloc(sizeof(*chain), GFP_KERNEL);
358 list_add_tail_rcu(&chain->list, &block->chain_list);
359 mutex_init(&chain->filter_chain_lock);
360 chain->block = block;
361 chain->index = chain_index;
364 block->chain0.chain = chain;
368 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
369 struct tcf_proto *tp_head)
371 if (item->chain_head_change)
372 item->chain_head_change(tp_head, item->chain_head_change_priv);
375 static void tcf_chain0_head_change(struct tcf_chain *chain,
376 struct tcf_proto *tp_head)
378 struct tcf_filter_chain_list_item *item;
379 struct tcf_block *block = chain->block;
384 mutex_lock(&block->lock);
385 list_for_each_entry(item, &block->chain0.filter_chain_list, list)
386 tcf_chain_head_change_item(item, tp_head);
387 mutex_unlock(&block->lock);
390 /* Returns true if block can be safely freed. */
392 static bool tcf_chain_detach(struct tcf_chain *chain)
394 struct tcf_block *block = chain->block;
396 ASSERT_BLOCK_LOCKED(block);
398 list_del_rcu(&chain->list);
400 block->chain0.chain = NULL;
402 if (list_empty(&block->chain_list) &&
403 refcount_read(&block->refcnt) == 0)
409 static void tcf_block_destroy(struct tcf_block *block)
411 mutex_destroy(&block->lock);
412 mutex_destroy(&block->proto_destroy_lock);
413 kfree_rcu(block, rcu);
416 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
418 struct tcf_block *block = chain->block;
420 mutex_destroy(&chain->filter_chain_lock);
421 kfree_rcu(chain, rcu);
423 tcf_block_destroy(block);
426 static void tcf_chain_hold(struct tcf_chain *chain)
428 ASSERT_BLOCK_LOCKED(chain->block);
433 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
435 ASSERT_BLOCK_LOCKED(chain->block);
437 /* In case all the references are action references, this
438 * chain should not be shown to the user.
440 return chain->refcnt == chain->action_refcnt;
443 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
446 struct tcf_chain *chain;
448 ASSERT_BLOCK_LOCKED(block);
450 list_for_each_entry(chain, &block->chain_list, list) {
451 if (chain->index == chain_index)
457 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
458 static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
461 struct tcf_chain *chain;
463 list_for_each_entry_rcu(chain, &block->chain_list, list) {
464 if (chain->index == chain_index)
471 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
472 u32 seq, u16 flags, int event, bool unicast);
474 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
475 u32 chain_index, bool create,
478 struct tcf_chain *chain = NULL;
479 bool is_first_reference;
481 mutex_lock(&block->lock);
482 chain = tcf_chain_lookup(block, chain_index);
484 tcf_chain_hold(chain);
488 chain = tcf_chain_create(block, chain_index);
494 ++chain->action_refcnt;
495 is_first_reference = chain->refcnt - chain->action_refcnt == 1;
496 mutex_unlock(&block->lock);
498 /* Send notification only in case we got the first
499 * non-action reference. Until then, the chain acts only as
500 * a placeholder for actions pointing to it and user ought
501 * not know about them.
503 if (is_first_reference && !by_act)
504 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
505 RTM_NEWCHAIN, false);
510 mutex_unlock(&block->lock);
514 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
517 return __tcf_chain_get(block, chain_index, create, false);
520 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
522 return __tcf_chain_get(block, chain_index, true, true);
524 EXPORT_SYMBOL(tcf_chain_get_by_act);
526 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
528 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
529 void *tmplt_priv, u32 chain_index,
530 struct tcf_block *block, struct sk_buff *oskb,
531 u32 seq, u16 flags, bool unicast);
533 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
534 bool explicitly_created)
536 struct tcf_block *block = chain->block;
537 const struct tcf_proto_ops *tmplt_ops;
538 bool free_block = false;
542 mutex_lock(&block->lock);
543 if (explicitly_created) {
544 if (!chain->explicitly_created) {
545 mutex_unlock(&block->lock);
548 chain->explicitly_created = false;
552 chain->action_refcnt--;
554 /* tc_chain_notify_delete can't be called while holding block lock.
555 * However, when block is unlocked chain can be changed concurrently, so
556 * save these to temporary variables.
558 refcnt = --chain->refcnt;
559 tmplt_ops = chain->tmplt_ops;
560 tmplt_priv = chain->tmplt_priv;
562 /* The last dropped non-action reference will trigger notification. */
563 if (refcnt - chain->action_refcnt == 0 && !by_act) {
564 tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
565 block, NULL, 0, 0, false);
566 /* Last reference to chain, no need to lock. */
567 chain->flushing = false;
571 free_block = tcf_chain_detach(chain);
572 mutex_unlock(&block->lock);
575 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
576 tcf_chain_destroy(chain, free_block);
580 static void tcf_chain_put(struct tcf_chain *chain)
582 __tcf_chain_put(chain, false, false);
585 void tcf_chain_put_by_act(struct tcf_chain *chain)
587 __tcf_chain_put(chain, true, false);
589 EXPORT_SYMBOL(tcf_chain_put_by_act);
591 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
593 __tcf_chain_put(chain, false, true);
596 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
598 struct tcf_proto *tp, *tp_next;
600 mutex_lock(&chain->filter_chain_lock);
601 tp = tcf_chain_dereference(chain->filter_chain, chain);
603 tp_next = rcu_dereference_protected(tp->next, 1);
604 tcf_proto_signal_destroying(chain, tp);
607 tp = tcf_chain_dereference(chain->filter_chain, chain);
608 RCU_INIT_POINTER(chain->filter_chain, NULL);
609 tcf_chain0_head_change(chain, NULL);
610 chain->flushing = true;
611 mutex_unlock(&chain->filter_chain_lock);
614 tp_next = rcu_dereference_protected(tp->next, 1);
615 tcf_proto_put(tp, rtnl_held, NULL);
620 static int tcf_block_setup(struct tcf_block *block,
621 struct flow_block_offload *bo);
623 static void tcf_block_offload_init(struct flow_block_offload *bo,
624 struct net_device *dev, struct Qdisc *sch,
625 enum flow_block_command command,
626 enum flow_block_binder_type binder_type,
627 struct flow_block *flow_block,
628 bool shared, struct netlink_ext_ack *extack)
630 bo->net = dev_net(dev);
631 bo->command = command;
632 bo->binder_type = binder_type;
633 bo->block = flow_block;
634 bo->block_shared = shared;
637 bo->cb_list_head = &flow_block->cb_list;
638 INIT_LIST_HEAD(&bo->cb_list);
641 static void tcf_block_unbind(struct tcf_block *block,
642 struct flow_block_offload *bo);
644 static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
646 struct tcf_block *block = block_cb->indr.data;
647 struct net_device *dev = block_cb->indr.dev;
648 struct Qdisc *sch = block_cb->indr.sch;
649 struct netlink_ext_ack extack = {};
650 struct flow_block_offload bo = {};
652 tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
653 block_cb->indr.binder_type,
654 &block->flow_block, tcf_block_shared(block),
657 down_write(&block->cb_lock);
658 list_del(&block_cb->driver_list);
659 list_move(&block_cb->list, &bo.cb_list);
660 tcf_block_unbind(block, &bo);
661 up_write(&block->cb_lock);
665 static bool tcf_block_offload_in_use(struct tcf_block *block)
667 return atomic_read(&block->offloadcnt);
670 static int tcf_block_offload_cmd(struct tcf_block *block,
671 struct net_device *dev, struct Qdisc *sch,
672 struct tcf_block_ext_info *ei,
673 enum flow_block_command command,
674 struct netlink_ext_ack *extack)
676 struct flow_block_offload bo = {};
678 tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
679 &block->flow_block, tcf_block_shared(block),
682 if (dev->netdev_ops->ndo_setup_tc) {
685 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
687 if (err != -EOPNOTSUPP)
688 NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
692 return tcf_block_setup(block, &bo);
695 flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
696 tc_block_indr_cleanup);
697 tcf_block_setup(block, &bo);
702 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
703 struct tcf_block_ext_info *ei,
704 struct netlink_ext_ack *extack)
706 struct net_device *dev = q->dev_queue->dev;
709 down_write(&block->cb_lock);
711 /* If tc offload feature is disabled and the block we try to bind
712 * to already has some offloaded filters, forbid to bind.
714 if (dev->netdev_ops->ndo_setup_tc &&
715 !tc_can_offload(dev) &&
716 tcf_block_offload_in_use(block)) {
717 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
722 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
723 if (err == -EOPNOTSUPP)
724 goto no_offload_dev_inc;
728 up_write(&block->cb_lock);
732 if (tcf_block_offload_in_use(block))
736 block->nooffloaddevcnt++;
738 up_write(&block->cb_lock);
742 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
743 struct tcf_block_ext_info *ei)
745 struct net_device *dev = q->dev_queue->dev;
748 down_write(&block->cb_lock);
749 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
750 if (err == -EOPNOTSUPP)
751 goto no_offload_dev_dec;
752 up_write(&block->cb_lock);
756 WARN_ON(block->nooffloaddevcnt-- == 0);
757 up_write(&block->cb_lock);
761 tcf_chain0_head_change_cb_add(struct tcf_block *block,
762 struct tcf_block_ext_info *ei,
763 struct netlink_ext_ack *extack)
765 struct tcf_filter_chain_list_item *item;
766 struct tcf_chain *chain0;
768 item = kmalloc(sizeof(*item), GFP_KERNEL);
770 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
773 item->chain_head_change = ei->chain_head_change;
774 item->chain_head_change_priv = ei->chain_head_change_priv;
776 mutex_lock(&block->lock);
777 chain0 = block->chain0.chain;
779 tcf_chain_hold(chain0);
781 list_add(&item->list, &block->chain0.filter_chain_list);
782 mutex_unlock(&block->lock);
785 struct tcf_proto *tp_head;
787 mutex_lock(&chain0->filter_chain_lock);
789 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
791 tcf_chain_head_change_item(item, tp_head);
793 mutex_lock(&block->lock);
794 list_add(&item->list, &block->chain0.filter_chain_list);
795 mutex_unlock(&block->lock);
797 mutex_unlock(&chain0->filter_chain_lock);
798 tcf_chain_put(chain0);
805 tcf_chain0_head_change_cb_del(struct tcf_block *block,
806 struct tcf_block_ext_info *ei)
808 struct tcf_filter_chain_list_item *item;
810 mutex_lock(&block->lock);
811 list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
812 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
813 (item->chain_head_change == ei->chain_head_change &&
814 item->chain_head_change_priv == ei->chain_head_change_priv)) {
815 if (block->chain0.chain)
816 tcf_chain_head_change_item(item, NULL);
817 list_del(&item->list);
818 mutex_unlock(&block->lock);
824 mutex_unlock(&block->lock);
829 spinlock_t idr_lock; /* Protects idr */
833 static unsigned int tcf_net_id;
835 static int tcf_block_insert(struct tcf_block *block, struct net *net,
836 struct netlink_ext_ack *extack)
838 struct tcf_net *tn = net_generic(net, tcf_net_id);
841 idr_preload(GFP_KERNEL);
842 spin_lock(&tn->idr_lock);
843 err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
845 spin_unlock(&tn->idr_lock);
851 static void tcf_block_remove(struct tcf_block *block, struct net *net)
853 struct tcf_net *tn = net_generic(net, tcf_net_id);
855 spin_lock(&tn->idr_lock);
856 idr_remove(&tn->idr, block->index);
857 spin_unlock(&tn->idr_lock);
860 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
862 struct netlink_ext_ack *extack)
864 struct tcf_block *block;
866 block = kzalloc(sizeof(*block), GFP_KERNEL);
868 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
869 return ERR_PTR(-ENOMEM);
871 mutex_init(&block->lock);
872 mutex_init(&block->proto_destroy_lock);
873 init_rwsem(&block->cb_lock);
874 flow_block_init(&block->flow_block);
875 INIT_LIST_HEAD(&block->chain_list);
876 INIT_LIST_HEAD(&block->owner_list);
877 INIT_LIST_HEAD(&block->chain0.filter_chain_list);
879 refcount_set(&block->refcnt, 1);
881 block->index = block_index;
883 /* Don't store q pointer for blocks which are shared */
884 if (!tcf_block_shared(block))
889 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
891 struct tcf_net *tn = net_generic(net, tcf_net_id);
893 return idr_find(&tn->idr, block_index);
896 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
898 struct tcf_block *block;
901 block = tcf_block_lookup(net, block_index);
902 if (block && !refcount_inc_not_zero(&block->refcnt))
909 static struct tcf_chain *
910 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
912 mutex_lock(&block->lock);
914 chain = list_is_last(&chain->list, &block->chain_list) ?
915 NULL : list_next_entry(chain, list);
917 chain = list_first_entry_or_null(&block->chain_list,
918 struct tcf_chain, list);
920 /* skip all action-only chains */
921 while (chain && tcf_chain_held_by_acts_only(chain))
922 chain = list_is_last(&chain->list, &block->chain_list) ?
923 NULL : list_next_entry(chain, list);
926 tcf_chain_hold(chain);
927 mutex_unlock(&block->lock);
932 /* Function to be used by all clients that want to iterate over all chains on
933 * block. It properly obtains block->lock and takes reference to chain before
934 * returning it. Users of this function must be tolerant to concurrent chain
935 * insertion/deletion or ensure that no concurrent chain modification is
936 * possible. Note that all netlink dump callbacks cannot guarantee to provide
937 * consistent dump because rtnl lock is released each time skb is filled with
938 * data and sent to user-space.
942 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
944 struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
947 tcf_chain_put(chain);
951 EXPORT_SYMBOL(tcf_get_next_chain);
953 static struct tcf_proto *
954 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
959 mutex_lock(&chain->filter_chain_lock);
962 tp = tcf_chain_dereference(chain->filter_chain, chain);
963 } else if (tcf_proto_is_deleting(tp)) {
964 /* 'deleting' flag is set and chain->filter_chain_lock was
965 * unlocked, which means next pointer could be invalid. Restart
969 tp = tcf_chain_dereference(chain->filter_chain, chain);
971 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
972 if (!tp->deleting && tp->prio >= prio)
975 tp = tcf_chain_dereference(tp->next, chain);
981 mutex_unlock(&chain->filter_chain_lock);
986 /* Function to be used by all clients that want to iterate over all tp's on
987 * chain. Users of this function must be tolerant to concurrent tp
988 * insertion/deletion or ensure that no concurrent chain modification is
989 * possible. Note that all netlink dump callbacks cannot guarantee to provide
990 * consistent dump because rtnl lock is released each time skb is filled with
991 * data and sent to user-space.
995 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
997 struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1000 tcf_proto_put(tp, true, NULL);
1004 EXPORT_SYMBOL(tcf_get_next_proto);
1006 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1008 struct tcf_chain *chain;
1010 /* Last reference to block. At this point chains cannot be added or
1011 * removed concurrently.
1013 for (chain = tcf_get_next_chain(block, NULL);
1015 chain = tcf_get_next_chain(block, chain)) {
1016 tcf_chain_put_explicitly_created(chain);
1017 tcf_chain_flush(chain, rtnl_held);
1021 /* Lookup Qdisc and increments its reference counter.
1022 * Set parent, if necessary.
1025 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1026 u32 *parent, int ifindex, bool rtnl_held,
1027 struct netlink_ext_ack *extack)
1029 const struct Qdisc_class_ops *cops;
1030 struct net_device *dev;
1033 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1039 dev = dev_get_by_index_rcu(net, ifindex);
1048 *parent = (*q)->handle;
1050 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1052 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1058 *q = qdisc_refcount_inc_nz(*q);
1060 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1065 /* Is it classful? */
1066 cops = (*q)->ops->cl_ops;
1068 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1073 if (!cops->tcf_block) {
1074 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1080 /* At this point we know that qdisc is not noop_qdisc,
1081 * which means that qdisc holds a reference to net_device
1082 * and we hold a reference to qdisc, so it is safe to release
1094 qdisc_put_unlocked(*q);
1100 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1101 int ifindex, struct netlink_ext_ack *extack)
1103 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1106 /* Do we search for filter, attached to class? */
1107 if (TC_H_MIN(parent)) {
1108 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1110 *cl = cops->find(q, parent);
1112 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1120 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1121 unsigned long cl, int ifindex,
1123 struct netlink_ext_ack *extack)
1125 struct tcf_block *block;
1127 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1128 block = tcf_block_refcnt_get(net, block_index);
1130 NL_SET_ERR_MSG(extack, "Block of given index was not found");
1131 return ERR_PTR(-EINVAL);
1134 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1136 block = cops->tcf_block(q, cl, extack);
1138 return ERR_PTR(-EINVAL);
1140 if (tcf_block_shared(block)) {
1141 NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1142 return ERR_PTR(-EOPNOTSUPP);
1145 /* Always take reference to block in order to support execution
1146 * of rules update path of cls API without rtnl lock. Caller
1147 * must release block when it is finished using it. 'if' block
1148 * of this conditional obtain reference to block by calling
1149 * tcf_block_refcnt_get().
1151 refcount_inc(&block->refcnt);
1157 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1158 struct tcf_block_ext_info *ei, bool rtnl_held)
1160 if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1161 /* Flushing/putting all chains will cause the block to be
1162 * deallocated when last chain is freed. However, if chain_list
1163 * is empty, block has to be manually deallocated. After block
1164 * reference counter reached 0, it is no longer possible to
1165 * increment it or add new chains to block.
1167 bool free_block = list_empty(&block->chain_list);
1169 mutex_unlock(&block->lock);
1170 if (tcf_block_shared(block))
1171 tcf_block_remove(block, block->net);
1174 tcf_block_offload_unbind(block, q, ei);
1177 tcf_block_destroy(block);
1179 tcf_block_flush_all_chains(block, rtnl_held);
1181 tcf_block_offload_unbind(block, q, ei);
1185 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1187 __tcf_block_put(block, NULL, NULL, rtnl_held);
1191 * Set q, parent, cl when appropriate.
1194 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1195 u32 *parent, unsigned long *cl,
1196 int ifindex, u32 block_index,
1197 struct netlink_ext_ack *extack)
1199 struct tcf_block *block;
1204 err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1208 err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1212 block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1213 if (IS_ERR(block)) {
1214 err = PTR_ERR(block);
1225 return ERR_PTR(err);
1228 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1231 if (!IS_ERR_OR_NULL(block))
1232 tcf_block_refcnt_put(block, rtnl_held);
1238 qdisc_put_unlocked(q);
1242 struct tcf_block_owner_item {
1243 struct list_head list;
1245 enum flow_block_binder_type binder_type;
1249 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1251 enum flow_block_binder_type binder_type)
1253 if (block->keep_dst &&
1254 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1255 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1256 netif_keep_dst(qdisc_dev(q));
1259 void tcf_block_netif_keep_dst(struct tcf_block *block)
1261 struct tcf_block_owner_item *item;
1263 block->keep_dst = true;
1264 list_for_each_entry(item, &block->owner_list, list)
1265 tcf_block_owner_netif_keep_dst(block, item->q,
1268 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1270 static int tcf_block_owner_add(struct tcf_block *block,
1272 enum flow_block_binder_type binder_type)
1274 struct tcf_block_owner_item *item;
1276 item = kmalloc(sizeof(*item), GFP_KERNEL);
1280 item->binder_type = binder_type;
1281 list_add(&item->list, &block->owner_list);
1285 static void tcf_block_owner_del(struct tcf_block *block,
1287 enum flow_block_binder_type binder_type)
1289 struct tcf_block_owner_item *item;
1291 list_for_each_entry(item, &block->owner_list, list) {
1292 if (item->q == q && item->binder_type == binder_type) {
1293 list_del(&item->list);
1301 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1302 struct tcf_block_ext_info *ei,
1303 struct netlink_ext_ack *extack)
1305 struct net *net = qdisc_net(q);
1306 struct tcf_block *block = NULL;
1309 if (ei->block_index)
1310 /* block_index not 0 means the shared block is requested */
1311 block = tcf_block_refcnt_get(net, ei->block_index);
1314 block = tcf_block_create(net, q, ei->block_index, extack);
1316 return PTR_ERR(block);
1317 if (tcf_block_shared(block)) {
1318 err = tcf_block_insert(block, net, extack);
1320 goto err_block_insert;
1324 err = tcf_block_owner_add(block, q, ei->binder_type);
1326 goto err_block_owner_add;
1328 tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1330 err = tcf_chain0_head_change_cb_add(block, ei, extack);
1332 goto err_chain0_head_change_cb_add;
1334 err = tcf_block_offload_bind(block, q, ei, extack);
1336 goto err_block_offload_bind;
1341 err_block_offload_bind:
1342 tcf_chain0_head_change_cb_del(block, ei);
1343 err_chain0_head_change_cb_add:
1344 tcf_block_owner_del(block, q, ei->binder_type);
1345 err_block_owner_add:
1347 tcf_block_refcnt_put(block, true);
1350 EXPORT_SYMBOL(tcf_block_get_ext);
1352 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1354 struct tcf_proto __rcu **p_filter_chain = priv;
1356 rcu_assign_pointer(*p_filter_chain, tp_head);
1359 int tcf_block_get(struct tcf_block **p_block,
1360 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1361 struct netlink_ext_ack *extack)
1363 struct tcf_block_ext_info ei = {
1364 .chain_head_change = tcf_chain_head_change_dflt,
1365 .chain_head_change_priv = p_filter_chain,
1368 WARN_ON(!p_filter_chain);
1369 return tcf_block_get_ext(p_block, q, &ei, extack);
1371 EXPORT_SYMBOL(tcf_block_get);
1373 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1374 * actions should be all removed after flushing.
1376 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1377 struct tcf_block_ext_info *ei)
1381 tcf_chain0_head_change_cb_del(block, ei);
1382 tcf_block_owner_del(block, q, ei->binder_type);
1384 __tcf_block_put(block, q, ei, true);
1386 EXPORT_SYMBOL(tcf_block_put_ext);
1388 void tcf_block_put(struct tcf_block *block)
1390 struct tcf_block_ext_info ei = {0, };
1394 tcf_block_put_ext(block, block->q, &ei);
1397 EXPORT_SYMBOL(tcf_block_put);
1400 tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1401 void *cb_priv, bool add, bool offload_in_use,
1402 struct netlink_ext_ack *extack)
1404 struct tcf_chain *chain, *chain_prev;
1405 struct tcf_proto *tp, *tp_prev;
1408 lockdep_assert_held(&block->cb_lock);
1410 for (chain = __tcf_get_next_chain(block, NULL);
1413 chain = __tcf_get_next_chain(block, chain),
1414 tcf_chain_put(chain_prev)) {
1415 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1417 tp = __tcf_get_next_proto(chain, tp),
1418 tcf_proto_put(tp_prev, true, NULL)) {
1419 if (tp->ops->reoffload) {
1420 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1423 goto err_playback_remove;
1424 } else if (add && offload_in_use) {
1426 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1427 goto err_playback_remove;
1434 err_playback_remove:
1435 tcf_proto_put(tp, true, NULL);
1436 tcf_chain_put(chain);
1437 tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1442 static int tcf_block_bind(struct tcf_block *block,
1443 struct flow_block_offload *bo)
1445 struct flow_block_cb *block_cb, *next;
1448 lockdep_assert_held(&block->cb_lock);
1450 list_for_each_entry(block_cb, &bo->cb_list, list) {
1451 err = tcf_block_playback_offloads(block, block_cb->cb,
1452 block_cb->cb_priv, true,
1453 tcf_block_offload_in_use(block),
1457 if (!bo->unlocked_driver_cb)
1458 block->lockeddevcnt++;
1462 list_splice(&bo->cb_list, &block->flow_block.cb_list);
1467 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1469 list_del(&block_cb->list);
1470 tcf_block_playback_offloads(block, block_cb->cb,
1471 block_cb->cb_priv, false,
1472 tcf_block_offload_in_use(block),
1474 if (!bo->unlocked_driver_cb)
1475 block->lockeddevcnt--;
1477 flow_block_cb_free(block_cb);
1483 static void tcf_block_unbind(struct tcf_block *block,
1484 struct flow_block_offload *bo)
1486 struct flow_block_cb *block_cb, *next;
1488 lockdep_assert_held(&block->cb_lock);
1490 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1491 tcf_block_playback_offloads(block, block_cb->cb,
1492 block_cb->cb_priv, false,
1493 tcf_block_offload_in_use(block),
1495 list_del(&block_cb->list);
1496 flow_block_cb_free(block_cb);
1497 if (!bo->unlocked_driver_cb)
1498 block->lockeddevcnt--;
1502 static int tcf_block_setup(struct tcf_block *block,
1503 struct flow_block_offload *bo)
1507 switch (bo->command) {
1508 case FLOW_BLOCK_BIND:
1509 err = tcf_block_bind(block, bo);
1511 case FLOW_BLOCK_UNBIND:
1513 tcf_block_unbind(block, bo);
1523 /* Main classifier routine: scans classifier chain attached
1524 * to this qdisc, (optionally) tests for protocol and asks
1525 * specific classifiers.
1527 static inline int __tcf_classify(struct sk_buff *skb,
1528 const struct tcf_proto *tp,
1529 const struct tcf_proto *orig_tp,
1530 struct tcf_result *res,
1532 u32 *last_executed_chain)
1534 #ifdef CONFIG_NET_CLS_ACT
1535 const int max_reclassify_loop = 16;
1536 const struct tcf_proto *first_tp;
1541 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1542 __be16 protocol = skb_protocol(skb, false);
1545 if (tp->protocol != protocol &&
1546 tp->protocol != htons(ETH_P_ALL))
1549 err = tp->classify(skb, tp, res);
1550 #ifdef CONFIG_NET_CLS_ACT
1551 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1553 *last_executed_chain = first_tp->chain->index;
1555 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1556 first_tp = res->goto_tp;
1557 *last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1565 return TC_ACT_UNSPEC; /* signal: continue lookup */
1566 #ifdef CONFIG_NET_CLS_ACT
1568 if (unlikely(limit++ >= max_reclassify_loop)) {
1569 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1570 tp->chain->block->index,
1572 ntohs(tp->protocol));
1581 int tcf_classify(struct sk_buff *skb,
1582 const struct tcf_block *block,
1583 const struct tcf_proto *tp,
1584 struct tcf_result *res, bool compat_mode)
1586 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1587 u32 last_executed_chain = 0;
1589 return __tcf_classify(skb, tp, tp, res, compat_mode,
1590 &last_executed_chain);
1592 u32 last_executed_chain = tp ? tp->chain->index : 0;
1593 const struct tcf_proto *orig_tp = tp;
1594 struct tc_skb_ext *ext;
1598 ext = skb_ext_find(skb, TC_SKB_EXT);
1600 if (ext && ext->chain) {
1601 struct tcf_chain *fchain;
1603 fchain = tcf_chain_lookup_rcu(block, ext->chain);
1607 /* Consume, so cloned/redirect skbs won't inherit ext */
1608 skb_ext_del(skb, TC_SKB_EXT);
1610 tp = rcu_dereference_bh(fchain->filter_chain);
1611 last_executed_chain = fchain->index;
1615 ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode,
1616 &last_executed_chain);
1618 /* If we missed on some chain */
1619 if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1620 struct tc_skb_cb *cb = tc_skb_cb(skb);
1622 ext = tc_skb_ext_alloc(skb);
1623 if (WARN_ON_ONCE(!ext))
1625 ext->chain = last_executed_chain;
1627 ext->post_ct = cb->post_ct;
1628 ext->post_ct_snat = cb->post_ct_snat;
1629 ext->post_ct_dnat = cb->post_ct_dnat;
1630 ext->zone = cb->zone;
1636 EXPORT_SYMBOL(tcf_classify);
1638 struct tcf_chain_info {
1639 struct tcf_proto __rcu **pprev;
1640 struct tcf_proto __rcu *next;
1643 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1644 struct tcf_chain_info *chain_info)
1646 return tcf_chain_dereference(*chain_info->pprev, chain);
1649 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1650 struct tcf_chain_info *chain_info,
1651 struct tcf_proto *tp)
1653 if (chain->flushing)
1656 if (*chain_info->pprev == chain->filter_chain)
1657 tcf_chain0_head_change(chain, tp);
1659 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1660 rcu_assign_pointer(*chain_info->pprev, tp);
1665 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1666 struct tcf_chain_info *chain_info,
1667 struct tcf_proto *tp)
1669 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1671 tcf_proto_mark_delete(tp);
1672 if (tp == chain->filter_chain)
1673 tcf_chain0_head_change(chain, next);
1674 RCU_INIT_POINTER(*chain_info->pprev, next);
1677 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1678 struct tcf_chain_info *chain_info,
1679 u32 protocol, u32 prio,
1680 bool prio_allocate);
1682 /* Try to insert new proto.
1683 * If proto with specified priority already exists, free new proto
1684 * and return existing one.
1687 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1688 struct tcf_proto *tp_new,
1689 u32 protocol, u32 prio,
1692 struct tcf_chain_info chain_info;
1693 struct tcf_proto *tp;
1696 mutex_lock(&chain->filter_chain_lock);
1698 if (tcf_proto_exists_destroying(chain, tp_new)) {
1699 mutex_unlock(&chain->filter_chain_lock);
1700 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1701 return ERR_PTR(-EAGAIN);
1704 tp = tcf_chain_tp_find(chain, &chain_info,
1705 protocol, prio, false);
1707 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1708 mutex_unlock(&chain->filter_chain_lock);
1711 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1714 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1715 tp_new = ERR_PTR(err);
1721 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1722 struct tcf_proto *tp, bool rtnl_held,
1723 struct netlink_ext_ack *extack)
1725 struct tcf_chain_info chain_info;
1726 struct tcf_proto *tp_iter;
1727 struct tcf_proto **pprev;
1728 struct tcf_proto *next;
1730 mutex_lock(&chain->filter_chain_lock);
1732 /* Atomically find and remove tp from chain. */
1733 for (pprev = &chain->filter_chain;
1734 (tp_iter = tcf_chain_dereference(*pprev, chain));
1735 pprev = &tp_iter->next) {
1736 if (tp_iter == tp) {
1737 chain_info.pprev = pprev;
1738 chain_info.next = tp_iter->next;
1739 WARN_ON(tp_iter->deleting);
1743 /* Verify that tp still exists and no new filters were inserted
1745 * Mark tp for deletion if it is empty.
1747 if (!tp_iter || !tcf_proto_check_delete(tp)) {
1748 mutex_unlock(&chain->filter_chain_lock);
1752 tcf_proto_signal_destroying(chain, tp);
1753 next = tcf_chain_dereference(chain_info.next, chain);
1754 if (tp == chain->filter_chain)
1755 tcf_chain0_head_change(chain, next);
1756 RCU_INIT_POINTER(*chain_info.pprev, next);
1757 mutex_unlock(&chain->filter_chain_lock);
1759 tcf_proto_put(tp, rtnl_held, extack);
1762 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1763 struct tcf_chain_info *chain_info,
1764 u32 protocol, u32 prio,
1767 struct tcf_proto **pprev;
1768 struct tcf_proto *tp;
1770 /* Check the chain for existence of proto-tcf with this priority */
1771 for (pprev = &chain->filter_chain;
1772 (tp = tcf_chain_dereference(*pprev, chain));
1773 pprev = &tp->next) {
1774 if (tp->prio >= prio) {
1775 if (tp->prio == prio) {
1776 if (prio_allocate ||
1777 (tp->protocol != protocol && protocol))
1778 return ERR_PTR(-EINVAL);
1785 chain_info->pprev = pprev;
1787 chain_info->next = tp->next;
1790 chain_info->next = NULL;
1795 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1796 struct tcf_proto *tp, struct tcf_block *block,
1797 struct Qdisc *q, u32 parent, void *fh,
1798 u32 portid, u32 seq, u16 flags, int event,
1799 bool terse_dump, bool rtnl_held)
1802 struct nlmsghdr *nlh;
1803 unsigned char *b = skb_tail_pointer(skb);
1805 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1807 goto out_nlmsg_trim;
1808 tcm = nlmsg_data(nlh);
1809 tcm->tcm_family = AF_UNSPEC;
1813 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1814 tcm->tcm_parent = parent;
1816 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1817 tcm->tcm_block_index = block->index;
1819 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1820 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1821 goto nla_put_failure;
1822 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1823 goto nla_put_failure;
1825 tcm->tcm_handle = 0;
1826 } else if (terse_dump) {
1827 if (tp->ops->terse_dump) {
1828 if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
1830 goto nla_put_failure;
1832 goto cls_op_not_supp;
1835 if (tp->ops->dump &&
1836 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
1837 goto nla_put_failure;
1839 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1849 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1850 struct nlmsghdr *n, struct tcf_proto *tp,
1851 struct tcf_block *block, struct Qdisc *q,
1852 u32 parent, void *fh, int event, bool unicast,
1855 struct sk_buff *skb;
1856 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1859 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1863 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1864 n->nlmsg_seq, n->nlmsg_flags, event,
1865 false, rtnl_held) <= 0) {
1871 err = rtnl_unicast(skb, net, portid);
1873 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1874 n->nlmsg_flags & NLM_F_ECHO);
1878 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1879 struct nlmsghdr *n, struct tcf_proto *tp,
1880 struct tcf_block *block, struct Qdisc *q,
1881 u32 parent, void *fh, bool unicast, bool *last,
1882 bool rtnl_held, struct netlink_ext_ack *extack)
1884 struct sk_buff *skb;
1885 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1888 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1892 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1893 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
1894 false, rtnl_held) <= 0) {
1895 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1900 err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
1907 err = rtnl_unicast(skb, net, portid);
1909 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1910 n->nlmsg_flags & NLM_F_ECHO);
1912 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1917 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1918 struct tcf_block *block, struct Qdisc *q,
1919 u32 parent, struct nlmsghdr *n,
1920 struct tcf_chain *chain, int event)
1922 struct tcf_proto *tp;
1924 for (tp = tcf_get_next_proto(chain, NULL);
1925 tp; tp = tcf_get_next_proto(chain, tp))
1926 tfilter_notify(net, oskb, n, tp, block,
1927 q, parent, NULL, event, false, true);
1930 static void tfilter_put(struct tcf_proto *tp, void *fh)
1932 if (tp->ops->put && fh)
1933 tp->ops->put(tp, fh);
1936 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1937 struct netlink_ext_ack *extack)
1939 struct net *net = sock_net(skb->sk);
1940 struct nlattr *tca[TCA_MAX + 1];
1941 char name[IFNAMSIZ];
1948 struct Qdisc *q = NULL;
1949 struct tcf_chain_info chain_info;
1950 struct tcf_chain *chain = NULL;
1951 struct tcf_block *block;
1952 struct tcf_proto *tp;
1957 bool rtnl_held = false;
1960 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1966 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
1967 rtm_tca_policy, extack);
1972 protocol = TC_H_MIN(t->tcm_info);
1973 prio = TC_H_MAJ(t->tcm_info);
1974 prio_allocate = false;
1975 parent = t->tcm_parent;
1982 /* If no priority is provided by the user,
1985 if (n->nlmsg_flags & NLM_F_CREATE) {
1986 prio = TC_H_MAKE(0x80000000U, 0U);
1987 prio_allocate = true;
1989 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1994 /* Find head of filter chain. */
1996 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2000 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2001 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2006 /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2007 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2008 * type is not specified, classifier is not unlocked.
2011 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2012 !tcf_proto_is_unlocked(name)) {
2017 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2021 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2023 if (IS_ERR(block)) {
2024 err = PTR_ERR(block);
2027 block->classid = parent;
2029 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2030 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2031 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2035 chain = tcf_chain_get(block, chain_index, true);
2037 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2042 mutex_lock(&chain->filter_chain_lock);
2043 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2044 prio, prio_allocate);
2046 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2052 struct tcf_proto *tp_new = NULL;
2054 if (chain->flushing) {
2059 /* Proto-tcf does not exist, create new one */
2061 if (tca[TCA_KIND] == NULL || !protocol) {
2062 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2067 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2068 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2074 prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2077 mutex_unlock(&chain->filter_chain_lock);
2078 tp_new = tcf_proto_create(name, protocol, prio, chain,
2080 if (IS_ERR(tp_new)) {
2081 err = PTR_ERR(tp_new);
2086 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2093 mutex_unlock(&chain->filter_chain_lock);
2096 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2097 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2102 fh = tp->ops->get(tp, t->tcm_handle);
2105 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2106 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2110 } else if (n->nlmsg_flags & NLM_F_EXCL) {
2111 tfilter_put(tp, fh);
2112 NL_SET_ERR_MSG(extack, "Filter already exists");
2117 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2118 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2123 if (!(n->nlmsg_flags & NLM_F_CREATE))
2124 flags |= TCA_ACT_FLAGS_REPLACE;
2126 flags |= TCA_ACT_FLAGS_NO_RTNL;
2127 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2130 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2131 RTM_NEWTFILTER, false, rtnl_held);
2132 tfilter_put(tp, fh);
2133 /* q pointer is NULL for shared blocks */
2135 q->flags &= ~TCQ_F_CAN_BYPASS;
2139 if (err && tp_created)
2140 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2143 if (tp && !IS_ERR(tp))
2144 tcf_proto_put(tp, rtnl_held, NULL);
2146 tcf_chain_put(chain);
2148 tcf_block_release(q, block, rtnl_held);
2153 if (err == -EAGAIN) {
2154 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2158 /* Replay the request. */
2164 mutex_unlock(&chain->filter_chain_lock);
2168 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2169 struct netlink_ext_ack *extack)
2171 struct net *net = sock_net(skb->sk);
2172 struct nlattr *tca[TCA_MAX + 1];
2173 char name[IFNAMSIZ];
2179 struct Qdisc *q = NULL;
2180 struct tcf_chain_info chain_info;
2181 struct tcf_chain *chain = NULL;
2182 struct tcf_block *block = NULL;
2183 struct tcf_proto *tp = NULL;
2184 unsigned long cl = 0;
2187 bool rtnl_held = false;
2189 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2192 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2193 rtm_tca_policy, extack);
2198 protocol = TC_H_MIN(t->tcm_info);
2199 prio = TC_H_MAJ(t->tcm_info);
2200 parent = t->tcm_parent;
2202 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2203 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2207 /* Find head of filter chain. */
2209 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2213 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2214 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2218 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2219 * found), qdisc is not unlocked, classifier type is not specified,
2220 * classifier is not unlocked.
2223 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2224 !tcf_proto_is_unlocked(name)) {
2229 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2233 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2235 if (IS_ERR(block)) {
2236 err = PTR_ERR(block);
2240 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2241 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2242 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2246 chain = tcf_chain_get(block, chain_index, false);
2248 /* User requested flush on non-existent chain. Nothing to do,
2249 * so just return success.
2255 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2261 tfilter_notify_chain(net, skb, block, q, parent, n,
2262 chain, RTM_DELTFILTER);
2263 tcf_chain_flush(chain, rtnl_held);
2268 mutex_lock(&chain->filter_chain_lock);
2269 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2271 if (!tp || IS_ERR(tp)) {
2272 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2273 err = tp ? PTR_ERR(tp) : -ENOENT;
2275 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2276 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2279 } else if (t->tcm_handle == 0) {
2280 tcf_proto_signal_destroying(chain, tp);
2281 tcf_chain_tp_remove(chain, &chain_info, tp);
2282 mutex_unlock(&chain->filter_chain_lock);
2284 tcf_proto_put(tp, rtnl_held, NULL);
2285 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2286 RTM_DELTFILTER, false, rtnl_held);
2290 mutex_unlock(&chain->filter_chain_lock);
2292 fh = tp->ops->get(tp, t->tcm_handle);
2295 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2300 err = tfilter_del_notify(net, skb, n, tp, block,
2301 q, parent, fh, false, &last,
2307 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2312 if (tp && !IS_ERR(tp))
2313 tcf_proto_put(tp, rtnl_held, NULL);
2314 tcf_chain_put(chain);
2316 tcf_block_release(q, block, rtnl_held);
2324 mutex_unlock(&chain->filter_chain_lock);
2328 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2329 struct netlink_ext_ack *extack)
2331 struct net *net = sock_net(skb->sk);
2332 struct nlattr *tca[TCA_MAX + 1];
2333 char name[IFNAMSIZ];
2339 struct Qdisc *q = NULL;
2340 struct tcf_chain_info chain_info;
2341 struct tcf_chain *chain = NULL;
2342 struct tcf_block *block = NULL;
2343 struct tcf_proto *tp = NULL;
2344 unsigned long cl = 0;
2347 bool rtnl_held = false;
2349 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2350 rtm_tca_policy, extack);
2355 protocol = TC_H_MIN(t->tcm_info);
2356 prio = TC_H_MAJ(t->tcm_info);
2357 parent = t->tcm_parent;
2360 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2364 /* Find head of filter chain. */
2366 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2370 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2371 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2375 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2376 * unlocked, classifier type is not specified, classifier is not
2379 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2380 !tcf_proto_is_unlocked(name)) {
2385 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2389 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2391 if (IS_ERR(block)) {
2392 err = PTR_ERR(block);
2396 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2397 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2398 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2402 chain = tcf_chain_get(block, chain_index, false);
2404 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2409 mutex_lock(&chain->filter_chain_lock);
2410 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2412 mutex_unlock(&chain->filter_chain_lock);
2413 if (!tp || IS_ERR(tp)) {
2414 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2415 err = tp ? PTR_ERR(tp) : -ENOENT;
2417 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2418 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2423 fh = tp->ops->get(tp, t->tcm_handle);
2426 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2429 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2430 fh, RTM_NEWTFILTER, true, rtnl_held);
2432 NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2435 tfilter_put(tp, fh);
2438 if (tp && !IS_ERR(tp))
2439 tcf_proto_put(tp, rtnl_held, NULL);
2440 tcf_chain_put(chain);
2442 tcf_block_release(q, block, rtnl_held);
2450 struct tcf_dump_args {
2451 struct tcf_walker w;
2452 struct sk_buff *skb;
2453 struct netlink_callback *cb;
2454 struct tcf_block *block;
2460 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2462 struct tcf_dump_args *a = (void *)arg;
2463 struct net *net = sock_net(a->skb->sk);
2465 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2466 n, NETLINK_CB(a->cb->skb).portid,
2467 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2468 RTM_NEWTFILTER, a->terse_dump, true);
2471 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2472 struct sk_buff *skb, struct netlink_callback *cb,
2473 long index_start, long *p_index, bool terse)
2475 struct net *net = sock_net(skb->sk);
2476 struct tcf_block *block = chain->block;
2477 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2478 struct tcf_proto *tp, *tp_prev;
2479 struct tcf_dump_args arg;
2481 for (tp = __tcf_get_next_proto(chain, NULL);
2484 tp = __tcf_get_next_proto(chain, tp),
2485 tcf_proto_put(tp_prev, true, NULL),
2487 if (*p_index < index_start)
2489 if (TC_H_MAJ(tcm->tcm_info) &&
2490 TC_H_MAJ(tcm->tcm_info) != tp->prio)
2492 if (TC_H_MIN(tcm->tcm_info) &&
2493 TC_H_MIN(tcm->tcm_info) != tp->protocol)
2495 if (*p_index > index_start)
2496 memset(&cb->args[1], 0,
2497 sizeof(cb->args) - sizeof(cb->args[0]));
2498 if (cb->args[1] == 0) {
2499 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2500 NETLINK_CB(cb->skb).portid,
2501 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2502 RTM_NEWTFILTER, false, true) <= 0)
2508 arg.w.fn = tcf_node_dump;
2513 arg.parent = parent;
2515 arg.w.skip = cb->args[1] - 1;
2517 arg.w.cookie = cb->args[2];
2518 arg.terse_dump = terse;
2519 tp->ops->walk(tp, &arg.w, true);
2520 cb->args[2] = arg.w.cookie;
2521 cb->args[1] = arg.w.count + 1;
2528 tcf_proto_put(tp, true, NULL);
2532 static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2533 [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2536 /* called with RTNL */
2537 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2539 struct tcf_chain *chain, *chain_prev;
2540 struct net *net = sock_net(skb->sk);
2541 struct nlattr *tca[TCA_MAX + 1];
2542 struct Qdisc *q = NULL;
2543 struct tcf_block *block;
2544 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2545 bool terse_dump = false;
2551 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2554 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2555 tcf_tfilter_dump_policy, cb->extack);
2559 if (tca[TCA_DUMP_FLAGS]) {
2560 struct nla_bitfield32 flags =
2561 nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2563 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2566 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2567 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2570 /* If we work with block index, q is NULL and parent value
2571 * will never be used in the following code. The check
2572 * in tcf_fill_node prevents it. However, compiler does not
2573 * see that far, so set parent to zero to silence the warning
2574 * about parent being uninitialized.
2578 const struct Qdisc_class_ops *cops;
2579 struct net_device *dev;
2580 unsigned long cl = 0;
2582 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2586 parent = tcm->tcm_parent;
2590 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2593 cops = q->ops->cl_ops;
2596 if (!cops->tcf_block)
2598 if (TC_H_MIN(tcm->tcm_parent)) {
2599 cl = cops->find(q, tcm->tcm_parent);
2603 block = cops->tcf_block(q, cl, NULL);
2606 parent = block->classid;
2607 if (tcf_block_shared(block))
2611 index_start = cb->args[0];
2614 for (chain = __tcf_get_next_chain(block, NULL);
2617 chain = __tcf_get_next_chain(block, chain),
2618 tcf_chain_put(chain_prev)) {
2619 if (tca[TCA_CHAIN] &&
2620 nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2622 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2623 index_start, &index, terse_dump)) {
2624 tcf_chain_put(chain);
2630 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2631 tcf_block_refcnt_put(block, true);
2632 cb->args[0] = index;
2635 /* If we did no progress, the error (EMSGSIZE) is real */
2636 if (skb->len == 0 && err)
2641 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2642 void *tmplt_priv, u32 chain_index,
2643 struct net *net, struct sk_buff *skb,
2644 struct tcf_block *block,
2645 u32 portid, u32 seq, u16 flags, int event)
2647 unsigned char *b = skb_tail_pointer(skb);
2648 const struct tcf_proto_ops *ops;
2649 struct nlmsghdr *nlh;
2656 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2658 goto out_nlmsg_trim;
2659 tcm = nlmsg_data(nlh);
2660 tcm->tcm_family = AF_UNSPEC;
2663 tcm->tcm_handle = 0;
2665 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2666 tcm->tcm_parent = block->q->handle;
2668 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2669 tcm->tcm_block_index = block->index;
2672 if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2673 goto nla_put_failure;
2676 if (nla_put_string(skb, TCA_KIND, ops->kind))
2677 goto nla_put_failure;
2678 if (ops->tmplt_dump(skb, net, priv) < 0)
2679 goto nla_put_failure;
2682 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2691 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2692 u32 seq, u16 flags, int event, bool unicast)
2694 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2695 struct tcf_block *block = chain->block;
2696 struct net *net = block->net;
2697 struct sk_buff *skb;
2700 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2704 if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2705 chain->index, net, skb, block, portid,
2706 seq, flags, event) <= 0) {
2712 err = rtnl_unicast(skb, net, portid);
2714 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2715 flags & NLM_F_ECHO);
2720 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2721 void *tmplt_priv, u32 chain_index,
2722 struct tcf_block *block, struct sk_buff *oskb,
2723 u32 seq, u16 flags, bool unicast)
2725 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2726 struct net *net = block->net;
2727 struct sk_buff *skb;
2729 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2733 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2734 block, portid, seq, flags, RTM_DELCHAIN) <= 0) {
2740 return rtnl_unicast(skb, net, portid);
2742 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2745 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2746 struct nlattr **tca,
2747 struct netlink_ext_ack *extack)
2749 const struct tcf_proto_ops *ops;
2750 char name[IFNAMSIZ];
2753 /* If kind is not set, user did not specify template. */
2757 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2758 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
2762 ops = tcf_proto_lookup_ops(name, true, extack);
2764 return PTR_ERR(ops);
2765 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2766 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2770 tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2771 if (IS_ERR(tmplt_priv)) {
2772 module_put(ops->owner);
2773 return PTR_ERR(tmplt_priv);
2775 chain->tmplt_ops = ops;
2776 chain->tmplt_priv = tmplt_priv;
2780 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2783 /* If template ops are set, no work to do for us. */
2787 tmplt_ops->tmplt_destroy(tmplt_priv);
2788 module_put(tmplt_ops->owner);
2791 /* Add/delete/get a chain */
2793 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2794 struct netlink_ext_ack *extack)
2796 struct net *net = sock_net(skb->sk);
2797 struct nlattr *tca[TCA_MAX + 1];
2801 struct Qdisc *q = NULL;
2802 struct tcf_chain *chain = NULL;
2803 struct tcf_block *block;
2807 if (n->nlmsg_type != RTM_GETCHAIN &&
2808 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2812 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2813 rtm_tca_policy, extack);
2818 parent = t->tcm_parent;
2821 block = tcf_block_find(net, &q, &parent, &cl,
2822 t->tcm_ifindex, t->tcm_block_index, extack);
2824 return PTR_ERR(block);
2826 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2827 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2828 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2833 mutex_lock(&block->lock);
2834 chain = tcf_chain_lookup(block, chain_index);
2835 if (n->nlmsg_type == RTM_NEWCHAIN) {
2837 if (tcf_chain_held_by_acts_only(chain)) {
2838 /* The chain exists only because there is
2839 * some action referencing it.
2841 tcf_chain_hold(chain);
2843 NL_SET_ERR_MSG(extack, "Filter chain already exists");
2845 goto errout_block_locked;
2848 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2849 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
2851 goto errout_block_locked;
2853 chain = tcf_chain_create(block, chain_index);
2855 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
2857 goto errout_block_locked;
2861 if (!chain || tcf_chain_held_by_acts_only(chain)) {
2862 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2864 goto errout_block_locked;
2866 tcf_chain_hold(chain);
2869 if (n->nlmsg_type == RTM_NEWCHAIN) {
2870 /* Modifying chain requires holding parent block lock. In case
2871 * the chain was successfully added, take a reference to the
2872 * chain. This ensures that an empty chain does not disappear at
2873 * the end of this function.
2875 tcf_chain_hold(chain);
2876 chain->explicitly_created = true;
2878 mutex_unlock(&block->lock);
2880 switch (n->nlmsg_type) {
2882 err = tc_chain_tmplt_add(chain, net, tca, extack);
2884 tcf_chain_put_explicitly_created(chain);
2888 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2889 RTM_NEWCHAIN, false);
2892 tfilter_notify_chain(net, skb, block, q, parent, n,
2893 chain, RTM_DELTFILTER);
2894 /* Flush the chain first as the user requested chain removal. */
2895 tcf_chain_flush(chain, true);
2896 /* In case the chain was successfully deleted, put a reference
2897 * to the chain previously taken during addition.
2899 tcf_chain_put_explicitly_created(chain);
2902 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
2903 n->nlmsg_flags, n->nlmsg_type, true);
2905 NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2909 NL_SET_ERR_MSG(extack, "Unsupported message type");
2914 tcf_chain_put(chain);
2916 tcf_block_release(q, block, true);
2918 /* Replay the request. */
2922 errout_block_locked:
2923 mutex_unlock(&block->lock);
2927 /* called with RTNL */
2928 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2930 struct net *net = sock_net(skb->sk);
2931 struct nlattr *tca[TCA_MAX + 1];
2932 struct Qdisc *q = NULL;
2933 struct tcf_block *block;
2934 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2935 struct tcf_chain *chain;
2940 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2943 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2944 rtm_tca_policy, cb->extack);
2948 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2949 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2953 const struct Qdisc_class_ops *cops;
2954 struct net_device *dev;
2955 unsigned long cl = 0;
2957 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2961 if (!tcm->tcm_parent)
2964 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2968 cops = q->ops->cl_ops;
2971 if (!cops->tcf_block)
2973 if (TC_H_MIN(tcm->tcm_parent)) {
2974 cl = cops->find(q, tcm->tcm_parent);
2978 block = cops->tcf_block(q, cl, NULL);
2981 if (tcf_block_shared(block))
2985 index_start = cb->args[0];
2988 mutex_lock(&block->lock);
2989 list_for_each_entry(chain, &block->chain_list, list) {
2990 if ((tca[TCA_CHAIN] &&
2991 nla_get_u32(tca[TCA_CHAIN]) != chain->index))
2993 if (index < index_start) {
2997 if (tcf_chain_held_by_acts_only(chain))
2999 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3000 chain->index, net, skb, block,
3001 NETLINK_CB(cb->skb).portid,
3002 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3008 mutex_unlock(&block->lock);
3010 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3011 tcf_block_refcnt_put(block, true);
3012 cb->args[0] = index;
3015 /* If we did no progress, the error (EMSGSIZE) is real */
3016 if (skb->len == 0 && err)
3021 void tcf_exts_destroy(struct tcf_exts *exts)
3023 #ifdef CONFIG_NET_CLS_ACT
3024 if (exts->actions) {
3025 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3026 kfree(exts->actions);
3028 exts->nr_actions = 0;
3031 EXPORT_SYMBOL(tcf_exts_destroy);
3033 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3034 struct nlattr *rate_tlv, struct tcf_exts *exts,
3035 u32 flags, struct netlink_ext_ack *extack)
3037 #ifdef CONFIG_NET_CLS_ACT
3039 int init_res[TCA_ACT_MAX_PRIO] = {};
3040 struct tc_action *act;
3041 size_t attr_size = 0;
3043 if (exts->police && tb[exts->police]) {
3044 struct tc_action_ops *a_o;
3046 a_o = tc_action_load_ops(tb[exts->police], true,
3047 !(flags & TCA_ACT_FLAGS_NO_RTNL),
3050 return PTR_ERR(a_o);
3051 flags |= TCA_ACT_FLAGS_POLICE | TCA_ACT_FLAGS_BIND;
3052 act = tcf_action_init_1(net, tp, tb[exts->police],
3053 rate_tlv, a_o, init_res, flags,
3055 module_put(a_o->owner);
3057 return PTR_ERR(act);
3059 act->type = exts->type = TCA_OLD_COMPAT;
3060 exts->actions[0] = act;
3061 exts->nr_actions = 1;
3062 tcf_idr_insert_many(exts->actions);
3063 } else if (exts->action && tb[exts->action]) {
3066 flags |= TCA_ACT_FLAGS_BIND;
3067 err = tcf_action_init(net, tp, tb[exts->action],
3068 rate_tlv, exts->actions, init_res,
3069 &attr_size, flags, extack);
3072 exts->nr_actions = err;
3076 if ((exts->action && tb[exts->action]) ||
3077 (exts->police && tb[exts->police])) {
3078 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3085 EXPORT_SYMBOL(tcf_exts_validate);
3087 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3089 #ifdef CONFIG_NET_CLS_ACT
3090 struct tcf_exts old = *dst;
3093 tcf_exts_destroy(&old);
3096 EXPORT_SYMBOL(tcf_exts_change);
3098 #ifdef CONFIG_NET_CLS_ACT
3099 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3101 if (exts->nr_actions == 0)
3104 return exts->actions[0];
3108 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3110 #ifdef CONFIG_NET_CLS_ACT
3111 struct nlattr *nest;
3113 if (exts->action && tcf_exts_has_actions(exts)) {
3115 * again for backward compatible mode - we want
3116 * to work with both old and new modes of entering
3117 * tc data even if iproute2 was newer - jhs
3119 if (exts->type != TCA_OLD_COMPAT) {
3120 nest = nla_nest_start_noflag(skb, exts->action);
3122 goto nla_put_failure;
3124 if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3126 goto nla_put_failure;
3127 nla_nest_end(skb, nest);
3128 } else if (exts->police) {
3129 struct tc_action *act = tcf_exts_first_act(exts);
3130 nest = nla_nest_start_noflag(skb, exts->police);
3131 if (nest == NULL || !act)
3132 goto nla_put_failure;
3133 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3134 goto nla_put_failure;
3135 nla_nest_end(skb, nest);
3141 nla_nest_cancel(skb, nest);
3147 EXPORT_SYMBOL(tcf_exts_dump);
3149 int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3151 #ifdef CONFIG_NET_CLS_ACT
3152 struct nlattr *nest;
3154 if (!exts->action || !tcf_exts_has_actions(exts))
3157 nest = nla_nest_start_noflag(skb, exts->action);
3159 goto nla_put_failure;
3161 if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3162 goto nla_put_failure;
3163 nla_nest_end(skb, nest);
3167 nla_nest_cancel(skb, nest);
3173 EXPORT_SYMBOL(tcf_exts_terse_dump);
3175 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3177 #ifdef CONFIG_NET_CLS_ACT
3178 struct tc_action *a = tcf_exts_first_act(exts);
3179 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3184 EXPORT_SYMBOL(tcf_exts_dump_stats);
3186 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3188 if (*flags & TCA_CLS_FLAGS_IN_HW)
3190 *flags |= TCA_CLS_FLAGS_IN_HW;
3191 atomic_inc(&block->offloadcnt);
3194 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3196 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3198 *flags &= ~TCA_CLS_FLAGS_IN_HW;
3199 atomic_dec(&block->offloadcnt);
3202 static void tc_cls_offload_cnt_update(struct tcf_block *block,
3203 struct tcf_proto *tp, u32 *cnt,
3204 u32 *flags, u32 diff, bool add)
3206 lockdep_assert_held(&block->cb_lock);
3208 spin_lock(&tp->lock);
3211 tcf_block_offload_inc(block, flags);
3216 tcf_block_offload_dec(block, flags);
3218 spin_unlock(&tp->lock);
3222 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3223 u32 *cnt, u32 *flags)
3225 lockdep_assert_held(&block->cb_lock);
3227 spin_lock(&tp->lock);
3228 tcf_block_offload_dec(block, flags);
3230 spin_unlock(&tp->lock);
3234 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3235 void *type_data, bool err_stop)
3237 struct flow_block_cb *block_cb;
3241 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3242 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3253 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3254 void *type_data, bool err_stop, bool rtnl_held)
3256 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3262 down_read(&block->cb_lock);
3263 /* Need to obtain rtnl lock if block is bound to devs that require it.
3264 * In block bind code cb_lock is obtained while holding rtnl, so we must
3265 * obtain the locks in same order here.
3267 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3268 up_read(&block->cb_lock);
3273 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3275 up_read(&block->cb_lock);
3280 EXPORT_SYMBOL(tc_setup_cb_call);
3282 /* Non-destructive filter add. If filter that wasn't already in hardware is
3283 * successfully offloaded, increment block offloads counter. On failure,
3284 * previously offloaded filter is considered to be intact and offloads counter
3285 * is not decremented.
3288 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3289 enum tc_setup_type type, void *type_data, bool err_stop,
3290 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3292 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3298 down_read(&block->cb_lock);
3299 /* Need to obtain rtnl lock if block is bound to devs that require it.
3300 * In block bind code cb_lock is obtained while holding rtnl, so we must
3301 * obtain the locks in same order here.
3303 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3304 up_read(&block->cb_lock);
3309 /* Make sure all netdevs sharing this block are offload-capable. */
3310 if (block->nooffloaddevcnt && err_stop) {
3311 ok_count = -EOPNOTSUPP;
3315 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3319 if (tp->ops->hw_add)
3320 tp->ops->hw_add(tp, type_data);
3322 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3325 up_read(&block->cb_lock);
3328 return ok_count < 0 ? ok_count : 0;
3330 EXPORT_SYMBOL(tc_setup_cb_add);
3332 /* Destructive filter replace. If filter that wasn't already in hardware is
3333 * successfully offloaded, increment block offload counter. On failure,
3334 * previously offloaded filter is considered to be destroyed and offload counter
3338 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3339 enum tc_setup_type type, void *type_data, bool err_stop,
3340 u32 *old_flags, unsigned int *old_in_hw_count,
3341 u32 *new_flags, unsigned int *new_in_hw_count,
3344 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3350 down_read(&block->cb_lock);
3351 /* Need to obtain rtnl lock if block is bound to devs that require it.
3352 * In block bind code cb_lock is obtained while holding rtnl, so we must
3353 * obtain the locks in same order here.
3355 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3356 up_read(&block->cb_lock);
3361 /* Make sure all netdevs sharing this block are offload-capable. */
3362 if (block->nooffloaddevcnt && err_stop) {
3363 ok_count = -EOPNOTSUPP;
3367 tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3368 if (tp->ops->hw_del)
3369 tp->ops->hw_del(tp, type_data);
3371 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3375 if (tp->ops->hw_add)
3376 tp->ops->hw_add(tp, type_data);
3378 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3379 new_flags, ok_count, true);
3381 up_read(&block->cb_lock);
3384 return ok_count < 0 ? ok_count : 0;
3386 EXPORT_SYMBOL(tc_setup_cb_replace);
3388 /* Destroy filter and decrement block offload counter, if filter was previously
3392 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3393 enum tc_setup_type type, void *type_data, bool err_stop,
3394 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3396 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3402 down_read(&block->cb_lock);
3403 /* Need to obtain rtnl lock if block is bound to devs that require it.
3404 * In block bind code cb_lock is obtained while holding rtnl, so we must
3405 * obtain the locks in same order here.
3407 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3408 up_read(&block->cb_lock);
3413 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3415 tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3416 if (tp->ops->hw_del)
3417 tp->ops->hw_del(tp, type_data);
3419 up_read(&block->cb_lock);
3422 return ok_count < 0 ? ok_count : 0;
3424 EXPORT_SYMBOL(tc_setup_cb_destroy);
3426 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3427 bool add, flow_setup_cb_t *cb,
3428 enum tc_setup_type type, void *type_data,
3429 void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3431 int err = cb(type, type_data, cb_priv);
3434 if (add && tc_skip_sw(*flags))
3437 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3443 EXPORT_SYMBOL(tc_setup_cb_reoffload);
3445 static int tcf_act_get_cookie(struct flow_action_entry *entry,
3446 const struct tc_action *act)
3448 struct tc_cookie *cookie;
3452 cookie = rcu_dereference(act->act_cookie);
3454 entry->cookie = flow_action_cookie_create(cookie->data,
3464 static void tcf_act_put_cookie(struct flow_action_entry *entry)
3466 flow_action_cookie_destroy(entry->cookie);
3469 void tc_cleanup_flow_action(struct flow_action *flow_action)
3471 struct flow_action_entry *entry;
3474 flow_action_for_each(i, entry, flow_action) {
3475 tcf_act_put_cookie(entry);
3476 if (entry->destructor)
3477 entry->destructor(entry->destructor_priv);
3480 EXPORT_SYMBOL(tc_cleanup_flow_action);
3482 static void tcf_mirred_get_dev(struct flow_action_entry *entry,
3483 const struct tc_action *act)
3485 #ifdef CONFIG_NET_CLS_ACT
3486 entry->dev = act->ops->get_dev(act, &entry->destructor);
3489 entry->destructor_priv = entry->dev;
3493 static void tcf_tunnel_encap_put_tunnel(void *priv)
3495 struct ip_tunnel_info *tunnel = priv;
3500 static int tcf_tunnel_encap_get_tunnel(struct flow_action_entry *entry,
3501 const struct tc_action *act)
3503 entry->tunnel = tcf_tunnel_info_copy(act);
3506 entry->destructor = tcf_tunnel_encap_put_tunnel;
3507 entry->destructor_priv = entry->tunnel;
3511 static void tcf_sample_get_group(struct flow_action_entry *entry,
3512 const struct tc_action *act)
3514 #ifdef CONFIG_NET_CLS_ACT
3515 entry->sample.psample_group =
3516 act->ops->get_psample_group(act, &entry->destructor);
3517 entry->destructor_priv = entry->sample.psample_group;
3521 static void tcf_gate_entry_destructor(void *priv)
3523 struct action_gate_entry *oe = priv;
3528 static int tcf_gate_get_entries(struct flow_action_entry *entry,
3529 const struct tc_action *act)
3531 entry->gate.entries = tcf_gate_get_list(act);
3533 if (!entry->gate.entries)
3536 entry->destructor = tcf_gate_entry_destructor;
3537 entry->destructor_priv = entry->gate.entries;
3542 static enum flow_action_hw_stats tc_act_hw_stats(u8 hw_stats)
3544 if (WARN_ON_ONCE(hw_stats > TCA_ACT_HW_STATS_ANY))
3545 return FLOW_ACTION_HW_STATS_DONT_CARE;
3547 return FLOW_ACTION_HW_STATS_DISABLED;
3552 int tc_setup_flow_action(struct flow_action *flow_action,
3553 const struct tcf_exts *exts)
3555 struct tc_action *act;
3556 int i, j, k, err = 0;
3558 BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3559 BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3560 BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3566 tcf_exts_for_each_action(i, act, exts) {
3567 struct flow_action_entry *entry;
3569 entry = &flow_action->entries[j];
3570 spin_lock_bh(&act->tcfa_lock);
3571 err = tcf_act_get_cookie(entry, act);
3573 goto err_out_locked;
3575 entry->hw_stats = tc_act_hw_stats(act->hw_stats);
3577 if (is_tcf_gact_ok(act)) {
3578 entry->id = FLOW_ACTION_ACCEPT;
3579 } else if (is_tcf_gact_shot(act)) {
3580 entry->id = FLOW_ACTION_DROP;
3581 } else if (is_tcf_gact_trap(act)) {
3582 entry->id = FLOW_ACTION_TRAP;
3583 } else if (is_tcf_gact_goto_chain(act)) {
3584 entry->id = FLOW_ACTION_GOTO;
3585 entry->chain_index = tcf_gact_goto_chain_index(act);
3586 } else if (is_tcf_mirred_egress_redirect(act)) {
3587 entry->id = FLOW_ACTION_REDIRECT;
3588 tcf_mirred_get_dev(entry, act);
3589 } else if (is_tcf_mirred_egress_mirror(act)) {
3590 entry->id = FLOW_ACTION_MIRRED;
3591 tcf_mirred_get_dev(entry, act);
3592 } else if (is_tcf_mirred_ingress_redirect(act)) {
3593 entry->id = FLOW_ACTION_REDIRECT_INGRESS;
3594 tcf_mirred_get_dev(entry, act);
3595 } else if (is_tcf_mirred_ingress_mirror(act)) {
3596 entry->id = FLOW_ACTION_MIRRED_INGRESS;
3597 tcf_mirred_get_dev(entry, act);
3598 } else if (is_tcf_vlan(act)) {
3599 switch (tcf_vlan_action(act)) {
3600 case TCA_VLAN_ACT_PUSH:
3601 entry->id = FLOW_ACTION_VLAN_PUSH;
3602 entry->vlan.vid = tcf_vlan_push_vid(act);
3603 entry->vlan.proto = tcf_vlan_push_proto(act);
3604 entry->vlan.prio = tcf_vlan_push_prio(act);
3606 case TCA_VLAN_ACT_POP:
3607 entry->id = FLOW_ACTION_VLAN_POP;
3609 case TCA_VLAN_ACT_MODIFY:
3610 entry->id = FLOW_ACTION_VLAN_MANGLE;
3611 entry->vlan.vid = tcf_vlan_push_vid(act);
3612 entry->vlan.proto = tcf_vlan_push_proto(act);
3613 entry->vlan.prio = tcf_vlan_push_prio(act);
3617 goto err_out_locked;
3619 } else if (is_tcf_tunnel_set(act)) {
3620 entry->id = FLOW_ACTION_TUNNEL_ENCAP;
3621 err = tcf_tunnel_encap_get_tunnel(entry, act);
3623 goto err_out_locked;
3624 } else if (is_tcf_tunnel_release(act)) {
3625 entry->id = FLOW_ACTION_TUNNEL_DECAP;
3626 } else if (is_tcf_pedit(act)) {
3627 for (k = 0; k < tcf_pedit_nkeys(act); k++) {
3628 switch (tcf_pedit_cmd(act, k)) {
3629 case TCA_PEDIT_KEY_EX_CMD_SET:
3630 entry->id = FLOW_ACTION_MANGLE;
3632 case TCA_PEDIT_KEY_EX_CMD_ADD:
3633 entry->id = FLOW_ACTION_ADD;
3637 goto err_out_locked;
3639 entry->mangle.htype = tcf_pedit_htype(act, k);
3640 entry->mangle.mask = tcf_pedit_mask(act, k);
3641 entry->mangle.val = tcf_pedit_val(act, k);
3642 entry->mangle.offset = tcf_pedit_offset(act, k);
3643 entry->hw_stats = tc_act_hw_stats(act->hw_stats);
3644 entry = &flow_action->entries[++j];
3646 } else if (is_tcf_csum(act)) {
3647 entry->id = FLOW_ACTION_CSUM;
3648 entry->csum_flags = tcf_csum_update_flags(act);
3649 } else if (is_tcf_skbedit_mark(act)) {
3650 entry->id = FLOW_ACTION_MARK;
3651 entry->mark = tcf_skbedit_mark(act);
3652 } else if (is_tcf_sample(act)) {
3653 entry->id = FLOW_ACTION_SAMPLE;
3654 entry->sample.trunc_size = tcf_sample_trunc_size(act);
3655 entry->sample.truncate = tcf_sample_truncate(act);
3656 entry->sample.rate = tcf_sample_rate(act);
3657 tcf_sample_get_group(entry, act);
3658 } else if (is_tcf_police(act)) {
3659 entry->id = FLOW_ACTION_POLICE;
3660 entry->police.burst = tcf_police_burst(act);
3661 entry->police.rate_bytes_ps =
3662 tcf_police_rate_bytes_ps(act);
3663 entry->police.burst_pkt = tcf_police_burst_pkt(act);
3664 entry->police.rate_pkt_ps =
3665 tcf_police_rate_pkt_ps(act);
3666 entry->police.mtu = tcf_police_tcfp_mtu(act);
3667 entry->police.index = act->tcfa_index;
3668 } else if (is_tcf_ct(act)) {
3669 entry->id = FLOW_ACTION_CT;
3670 entry->ct.action = tcf_ct_action(act);
3671 entry->ct.zone = tcf_ct_zone(act);
3672 entry->ct.flow_table = tcf_ct_ft(act);
3673 } else if (is_tcf_mpls(act)) {
3674 switch (tcf_mpls_action(act)) {
3675 case TCA_MPLS_ACT_PUSH:
3676 entry->id = FLOW_ACTION_MPLS_PUSH;
3677 entry->mpls_push.proto = tcf_mpls_proto(act);
3678 entry->mpls_push.label = tcf_mpls_label(act);
3679 entry->mpls_push.tc = tcf_mpls_tc(act);
3680 entry->mpls_push.bos = tcf_mpls_bos(act);
3681 entry->mpls_push.ttl = tcf_mpls_ttl(act);
3683 case TCA_MPLS_ACT_POP:
3684 entry->id = FLOW_ACTION_MPLS_POP;
3685 entry->mpls_pop.proto = tcf_mpls_proto(act);
3687 case TCA_MPLS_ACT_MODIFY:
3688 entry->id = FLOW_ACTION_MPLS_MANGLE;
3689 entry->mpls_mangle.label = tcf_mpls_label(act);
3690 entry->mpls_mangle.tc = tcf_mpls_tc(act);
3691 entry->mpls_mangle.bos = tcf_mpls_bos(act);
3692 entry->mpls_mangle.ttl = tcf_mpls_ttl(act);
3696 goto err_out_locked;
3698 } else if (is_tcf_skbedit_ptype(act)) {
3699 entry->id = FLOW_ACTION_PTYPE;
3700 entry->ptype = tcf_skbedit_ptype(act);
3701 } else if (is_tcf_skbedit_priority(act)) {
3702 entry->id = FLOW_ACTION_PRIORITY;
3703 entry->priority = tcf_skbedit_priority(act);
3704 } else if (is_tcf_gate(act)) {
3705 entry->id = FLOW_ACTION_GATE;
3706 entry->gate.index = tcf_gate_index(act);
3707 entry->gate.prio = tcf_gate_prio(act);
3708 entry->gate.basetime = tcf_gate_basetime(act);
3709 entry->gate.cycletime = tcf_gate_cycletime(act);
3710 entry->gate.cycletimeext = tcf_gate_cycletimeext(act);
3711 entry->gate.num_entries = tcf_gate_num_entries(act);
3712 err = tcf_gate_get_entries(entry, act);
3714 goto err_out_locked;
3717 goto err_out_locked;
3719 spin_unlock_bh(&act->tcfa_lock);
3721 if (!is_tcf_pedit(act))
3727 tc_cleanup_flow_action(flow_action);
3731 spin_unlock_bh(&act->tcfa_lock);
3734 EXPORT_SYMBOL(tc_setup_flow_action);
3736 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3738 unsigned int num_acts = 0;
3739 struct tc_action *act;
3742 tcf_exts_for_each_action(i, act, exts) {
3743 if (is_tcf_pedit(act))
3744 num_acts += tcf_pedit_nkeys(act);
3750 EXPORT_SYMBOL(tcf_exts_num_actions);
3752 #ifdef CONFIG_NET_CLS_ACT
3753 static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3755 struct netlink_ext_ack *extack)
3757 *p_block_index = nla_get_u32(block_index_attr);
3758 if (!*p_block_index) {
3759 NL_SET_ERR_MSG(extack, "Block number may not be zero");
3766 int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3767 enum flow_block_binder_type binder_type,
3768 struct nlattr *block_index_attr,
3769 struct netlink_ext_ack *extack)
3774 if (!block_index_attr)
3777 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3784 qe->info.binder_type = binder_type;
3785 qe->info.chain_head_change = tcf_chain_head_change_dflt;
3786 qe->info.chain_head_change_priv = &qe->filter_chain;
3787 qe->info.block_index = block_index;
3789 return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3791 EXPORT_SYMBOL(tcf_qevent_init);
3793 void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3795 if (qe->info.block_index)
3796 tcf_block_put_ext(qe->block, sch, &qe->info);
3798 EXPORT_SYMBOL(tcf_qevent_destroy);
3800 int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3801 struct netlink_ext_ack *extack)
3806 if (!block_index_attr)
3809 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3813 /* Bounce newly-configured block or change in block. */
3814 if (block_index != qe->info.block_index) {
3815 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3821 EXPORT_SYMBOL(tcf_qevent_validate_change);
3823 struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
3824 struct sk_buff **to_free, int *ret)
3826 struct tcf_result cl_res;
3827 struct tcf_proto *fl;
3829 if (!qe->info.block_index)
3832 fl = rcu_dereference_bh(qe->filter_chain);
3834 switch (tcf_classify(skb, NULL, fl, &cl_res, false)) {
3836 qdisc_qstats_drop(sch);
3837 __qdisc_drop(skb, to_free);
3838 *ret = __NET_XMIT_BYPASS;
3843 __qdisc_drop(skb, to_free);
3844 *ret = __NET_XMIT_STOLEN;
3846 case TC_ACT_REDIRECT:
3847 skb_do_redirect(skb);
3848 *ret = __NET_XMIT_STOLEN;
3854 EXPORT_SYMBOL(tcf_qevent_handle);
3856 int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
3858 if (!qe->info.block_index)
3860 return nla_put_u32(skb, attr_name, qe->info.block_index);
3862 EXPORT_SYMBOL(tcf_qevent_dump);
3865 static __net_init int tcf_net_init(struct net *net)
3867 struct tcf_net *tn = net_generic(net, tcf_net_id);
3869 spin_lock_init(&tn->idr_lock);
3874 static void __net_exit tcf_net_exit(struct net *net)
3876 struct tcf_net *tn = net_generic(net, tcf_net_id);
3878 idr_destroy(&tn->idr);
3881 static struct pernet_operations tcf_net_ops = {
3882 .init = tcf_net_init,
3883 .exit = tcf_net_exit,
3885 .size = sizeof(struct tcf_net),
3888 static int __init tc_filter_init(void)
3892 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3896 err = register_pernet_subsys(&tcf_net_ops);
3898 goto err_register_pernet_subsys;
3900 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3901 RTNL_FLAG_DOIT_UNLOCKED);
3902 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3903 RTNL_FLAG_DOIT_UNLOCKED);
3904 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3905 tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3906 rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3907 rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3908 rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3913 err_register_pernet_subsys:
3914 destroy_workqueue(tc_filter_wq);
3918 subsys_initcall(tc_filter_init);