2 * net/sched/cls_api.c Packet classifier API.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
13 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
17 #include <linux/module.h>
18 #include <linux/types.h>
19 #include <linux/kernel.h>
20 #include <linux/string.h>
21 #include <linux/errno.h>
22 #include <linux/err.h>
23 #include <linux/skbuff.h>
24 #include <linux/init.h>
25 #include <linux/kmod.h>
26 #include <linux/slab.h>
27 #include <linux/idr.h>
28 #include <linux/rhashtable.h>
29 #include <net/net_namespace.h>
31 #include <net/netlink.h>
32 #include <net/pkt_sched.h>
33 #include <net/pkt_cls.h>
34 #include <net/tc_act/tc_pedit.h>
35 #include <net/tc_act/tc_mirred.h>
36 #include <net/tc_act/tc_vlan.h>
37 #include <net/tc_act/tc_tunnel_key.h>
38 #include <net/tc_act/tc_csum.h>
39 #include <net/tc_act/tc_gact.h>
40 #include <net/tc_act/tc_skbedit.h>
42 extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
44 /* The list of all installed classifier types */
45 static LIST_HEAD(tcf_proto_base);
47 /* Protects list of registered TC modules. It is pure SMP lock. */
48 static DEFINE_RWLOCK(cls_mod_lock);
50 /* Find classifier type by string name */
52 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
54 const struct tcf_proto_ops *t, *res = NULL;
57 read_lock(&cls_mod_lock);
58 list_for_each_entry(t, &tcf_proto_base, head) {
59 if (strcmp(kind, t->kind) == 0) {
60 if (try_module_get(t->owner))
65 read_unlock(&cls_mod_lock);
70 static const struct tcf_proto_ops *
71 tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
72 struct netlink_ext_ack *extack)
74 const struct tcf_proto_ops *ops;
76 ops = __tcf_proto_lookup_ops(kind);
82 request_module("cls_%s", kind);
85 ops = __tcf_proto_lookup_ops(kind);
86 /* We dropped the RTNL semaphore in order to perform
87 * the module load. So, even if we succeeded in loading
88 * the module we have to replay the request. We indicate
92 module_put(ops->owner);
93 return ERR_PTR(-EAGAIN);
96 NL_SET_ERR_MSG(extack, "TC classifier not found");
97 return ERR_PTR(-ENOENT);
100 /* Register(unregister) new classifier type */
102 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
104 struct tcf_proto_ops *t;
107 write_lock(&cls_mod_lock);
108 list_for_each_entry(t, &tcf_proto_base, head)
109 if (!strcmp(ops->kind, t->kind))
112 list_add_tail(&ops->head, &tcf_proto_base);
115 write_unlock(&cls_mod_lock);
118 EXPORT_SYMBOL(register_tcf_proto_ops);
120 static struct workqueue_struct *tc_filter_wq;
122 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
124 struct tcf_proto_ops *t;
127 /* Wait for outstanding call_rcu()s, if any, from a
128 * tcf_proto_ops's destroy() handler.
131 flush_workqueue(tc_filter_wq);
133 write_lock(&cls_mod_lock);
134 list_for_each_entry(t, &tcf_proto_base, head) {
141 write_unlock(&cls_mod_lock);
144 EXPORT_SYMBOL(unregister_tcf_proto_ops);
146 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
148 INIT_RCU_WORK(rwork, func);
149 return queue_rcu_work(tc_filter_wq, rwork);
151 EXPORT_SYMBOL(tcf_queue_work);
153 /* Select new prio value from the range, managed by kernel. */
155 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
157 u32 first = TC_H_MAKE(0xC0000000U, 0U);
160 first = tp->prio - 1;
162 return TC_H_MAJ(first);
165 static bool tcf_proto_is_unlocked(const char *kind)
167 const struct tcf_proto_ops *ops;
170 ops = tcf_proto_lookup_ops(kind, false, NULL);
171 /* On error return false to take rtnl lock. Proto lookup/create
172 * functions will perform lookup again and properly handle errors.
177 ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
178 module_put(ops->owner);
182 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
183 u32 prio, struct tcf_chain *chain,
185 struct netlink_ext_ack *extack)
187 struct tcf_proto *tp;
190 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
192 return ERR_PTR(-ENOBUFS);
194 tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
195 if (IS_ERR(tp->ops)) {
196 err = PTR_ERR(tp->ops);
199 tp->classify = tp->ops->classify;
200 tp->protocol = protocol;
203 spin_lock_init(&tp->lock);
204 refcount_set(&tp->refcnt, 1);
206 err = tp->ops->init(tp);
208 module_put(tp->ops->owner);
218 static void tcf_proto_get(struct tcf_proto *tp)
220 refcount_inc(&tp->refcnt);
223 static void tcf_chain_put(struct tcf_chain *chain);
225 static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
226 struct netlink_ext_ack *extack)
228 tp->ops->destroy(tp, rtnl_held, extack);
229 tcf_chain_put(tp->chain);
230 module_put(tp->ops->owner);
234 static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
235 struct netlink_ext_ack *extack)
237 if (refcount_dec_and_test(&tp->refcnt))
238 tcf_proto_destroy(tp, rtnl_held, extack);
241 static int walker_check_empty(struct tcf_proto *tp, void *fh,
242 struct tcf_walker *arg)
245 arg->nonempty = true;
251 static bool tcf_proto_is_empty(struct tcf_proto *tp, bool rtnl_held)
253 struct tcf_walker walker = { .fn = walker_check_empty, };
256 tp->ops->walk(tp, &walker, rtnl_held);
257 return !walker.nonempty;
262 static bool tcf_proto_check_delete(struct tcf_proto *tp, bool rtnl_held)
264 spin_lock(&tp->lock);
265 if (tcf_proto_is_empty(tp, rtnl_held))
267 spin_unlock(&tp->lock);
271 static void tcf_proto_mark_delete(struct tcf_proto *tp)
273 spin_lock(&tp->lock);
275 spin_unlock(&tp->lock);
278 static bool tcf_proto_is_deleting(struct tcf_proto *tp)
282 spin_lock(&tp->lock);
283 deleting = tp->deleting;
284 spin_unlock(&tp->lock);
289 #define ASSERT_BLOCK_LOCKED(block) \
290 lockdep_assert_held(&(block)->lock)
292 struct tcf_filter_chain_list_item {
293 struct list_head list;
294 tcf_chain_head_change_t *chain_head_change;
295 void *chain_head_change_priv;
298 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
301 struct tcf_chain *chain;
303 ASSERT_BLOCK_LOCKED(block);
305 chain = kzalloc(sizeof(*chain), GFP_KERNEL);
308 list_add_tail(&chain->list, &block->chain_list);
309 mutex_init(&chain->filter_chain_lock);
310 chain->block = block;
311 chain->index = chain_index;
314 block->chain0.chain = chain;
318 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
319 struct tcf_proto *tp_head)
321 if (item->chain_head_change)
322 item->chain_head_change(tp_head, item->chain_head_change_priv);
325 static void tcf_chain0_head_change(struct tcf_chain *chain,
326 struct tcf_proto *tp_head)
328 struct tcf_filter_chain_list_item *item;
329 struct tcf_block *block = chain->block;
334 mutex_lock(&block->lock);
335 list_for_each_entry(item, &block->chain0.filter_chain_list, list)
336 tcf_chain_head_change_item(item, tp_head);
337 mutex_unlock(&block->lock);
340 /* Returns true if block can be safely freed. */
342 static bool tcf_chain_detach(struct tcf_chain *chain)
344 struct tcf_block *block = chain->block;
346 ASSERT_BLOCK_LOCKED(block);
348 list_del(&chain->list);
350 block->chain0.chain = NULL;
352 if (list_empty(&block->chain_list) &&
353 refcount_read(&block->refcnt) == 0)
359 static void tcf_block_destroy(struct tcf_block *block)
361 mutex_destroy(&block->lock);
362 kfree_rcu(block, rcu);
365 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
367 struct tcf_block *block = chain->block;
369 mutex_destroy(&chain->filter_chain_lock);
372 tcf_block_destroy(block);
375 static void tcf_chain_hold(struct tcf_chain *chain)
377 ASSERT_BLOCK_LOCKED(chain->block);
382 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
384 ASSERT_BLOCK_LOCKED(chain->block);
386 /* In case all the references are action references, this
387 * chain should not be shown to the user.
389 return chain->refcnt == chain->action_refcnt;
392 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
395 struct tcf_chain *chain;
397 ASSERT_BLOCK_LOCKED(block);
399 list_for_each_entry(chain, &block->chain_list, list) {
400 if (chain->index == chain_index)
406 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
407 u32 seq, u16 flags, int event, bool unicast);
409 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
410 u32 chain_index, bool create,
413 struct tcf_chain *chain = NULL;
414 bool is_first_reference;
416 mutex_lock(&block->lock);
417 chain = tcf_chain_lookup(block, chain_index);
419 tcf_chain_hold(chain);
423 chain = tcf_chain_create(block, chain_index);
429 ++chain->action_refcnt;
430 is_first_reference = chain->refcnt - chain->action_refcnt == 1;
431 mutex_unlock(&block->lock);
433 /* Send notification only in case we got the first
434 * non-action reference. Until then, the chain acts only as
435 * a placeholder for actions pointing to it and user ought
436 * not know about them.
438 if (is_first_reference && !by_act)
439 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
440 RTM_NEWCHAIN, false);
445 mutex_unlock(&block->lock);
449 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
452 return __tcf_chain_get(block, chain_index, create, false);
455 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
457 return __tcf_chain_get(block, chain_index, true, true);
459 EXPORT_SYMBOL(tcf_chain_get_by_act);
461 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
463 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
464 void *tmplt_priv, u32 chain_index,
465 struct tcf_block *block, struct sk_buff *oskb,
466 u32 seq, u16 flags, bool unicast);
468 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
469 bool explicitly_created)
471 struct tcf_block *block = chain->block;
472 const struct tcf_proto_ops *tmplt_ops;
473 bool free_block = false;
477 mutex_lock(&block->lock);
478 if (explicitly_created) {
479 if (!chain->explicitly_created) {
480 mutex_unlock(&block->lock);
483 chain->explicitly_created = false;
487 chain->action_refcnt--;
489 /* tc_chain_notify_delete can't be called while holding block lock.
490 * However, when block is unlocked chain can be changed concurrently, so
491 * save these to temporary variables.
493 refcnt = --chain->refcnt;
494 tmplt_ops = chain->tmplt_ops;
495 tmplt_priv = chain->tmplt_priv;
497 /* The last dropped non-action reference will trigger notification. */
498 if (refcnt - chain->action_refcnt == 0 && !by_act) {
499 tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
500 block, NULL, 0, 0, false);
501 /* Last reference to chain, no need to lock. */
502 chain->flushing = false;
506 free_block = tcf_chain_detach(chain);
507 mutex_unlock(&block->lock);
510 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
511 tcf_chain_destroy(chain, free_block);
515 static void tcf_chain_put(struct tcf_chain *chain)
517 __tcf_chain_put(chain, false, false);
520 void tcf_chain_put_by_act(struct tcf_chain *chain)
522 __tcf_chain_put(chain, true, false);
524 EXPORT_SYMBOL(tcf_chain_put_by_act);
526 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
528 __tcf_chain_put(chain, false, true);
531 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
533 struct tcf_proto *tp, *tp_next;
535 mutex_lock(&chain->filter_chain_lock);
536 tp = tcf_chain_dereference(chain->filter_chain, chain);
537 RCU_INIT_POINTER(chain->filter_chain, NULL);
538 tcf_chain0_head_change(chain, NULL);
539 chain->flushing = true;
540 mutex_unlock(&chain->filter_chain_lock);
543 tp_next = rcu_dereference_protected(tp->next, 1);
544 tcf_proto_put(tp, rtnl_held, NULL);
549 static struct tcf_block *tc_dev_ingress_block(struct net_device *dev)
551 const struct Qdisc_class_ops *cops;
554 if (!dev_ingress_queue(dev))
557 qdisc = dev_ingress_queue(dev)->qdisc_sleeping;
561 cops = qdisc->ops->cl_ops;
565 if (!cops->tcf_block)
568 return cops->tcf_block(qdisc, TC_H_MIN_INGRESS, NULL);
571 static struct rhashtable indr_setup_block_ht;
573 struct tc_indr_block_dev {
574 struct rhash_head ht_node;
575 struct net_device *dev;
577 struct list_head cb_list;
578 struct tcf_block *block;
581 struct tc_indr_block_cb {
582 struct list_head list;
584 tc_indr_block_bind_cb_t *cb;
588 static const struct rhashtable_params tc_indr_setup_block_ht_params = {
589 .key_offset = offsetof(struct tc_indr_block_dev, dev),
590 .head_offset = offsetof(struct tc_indr_block_dev, ht_node),
591 .key_len = sizeof(struct net_device *),
594 static struct tc_indr_block_dev *
595 tc_indr_block_dev_lookup(struct net_device *dev)
597 return rhashtable_lookup_fast(&indr_setup_block_ht, &dev,
598 tc_indr_setup_block_ht_params);
601 static struct tc_indr_block_dev *tc_indr_block_dev_get(struct net_device *dev)
603 struct tc_indr_block_dev *indr_dev;
605 indr_dev = tc_indr_block_dev_lookup(dev);
609 indr_dev = kzalloc(sizeof(*indr_dev), GFP_KERNEL);
613 INIT_LIST_HEAD(&indr_dev->cb_list);
615 indr_dev->block = tc_dev_ingress_block(dev);
616 if (rhashtable_insert_fast(&indr_setup_block_ht, &indr_dev->ht_node,
617 tc_indr_setup_block_ht_params)) {
627 static void tc_indr_block_dev_put(struct tc_indr_block_dev *indr_dev)
629 if (--indr_dev->refcnt)
632 rhashtable_remove_fast(&indr_setup_block_ht, &indr_dev->ht_node,
633 tc_indr_setup_block_ht_params);
637 static struct tc_indr_block_cb *
638 tc_indr_block_cb_lookup(struct tc_indr_block_dev *indr_dev,
639 tc_indr_block_bind_cb_t *cb, void *cb_ident)
641 struct tc_indr_block_cb *indr_block_cb;
643 list_for_each_entry(indr_block_cb, &indr_dev->cb_list, list)
644 if (indr_block_cb->cb == cb &&
645 indr_block_cb->cb_ident == cb_ident)
646 return indr_block_cb;
650 static struct tc_indr_block_cb *
651 tc_indr_block_cb_add(struct tc_indr_block_dev *indr_dev, void *cb_priv,
652 tc_indr_block_bind_cb_t *cb, void *cb_ident)
654 struct tc_indr_block_cb *indr_block_cb;
656 indr_block_cb = tc_indr_block_cb_lookup(indr_dev, cb, cb_ident);
658 return ERR_PTR(-EEXIST);
660 indr_block_cb = kzalloc(sizeof(*indr_block_cb), GFP_KERNEL);
662 return ERR_PTR(-ENOMEM);
664 indr_block_cb->cb_priv = cb_priv;
665 indr_block_cb->cb = cb;
666 indr_block_cb->cb_ident = cb_ident;
667 list_add(&indr_block_cb->list, &indr_dev->cb_list);
669 return indr_block_cb;
672 static void tc_indr_block_cb_del(struct tc_indr_block_cb *indr_block_cb)
674 list_del(&indr_block_cb->list);
675 kfree(indr_block_cb);
678 static void tc_indr_block_ing_cmd(struct tc_indr_block_dev *indr_dev,
679 struct tc_indr_block_cb *indr_block_cb,
680 enum tc_block_command command)
682 struct tc_block_offload bo = {
684 .binder_type = TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS,
685 .block = indr_dev->block,
688 if (!indr_dev->block)
691 indr_block_cb->cb(indr_dev->dev, indr_block_cb->cb_priv, TC_SETUP_BLOCK,
695 int __tc_indr_block_cb_register(struct net_device *dev, void *cb_priv,
696 tc_indr_block_bind_cb_t *cb, void *cb_ident)
698 struct tc_indr_block_cb *indr_block_cb;
699 struct tc_indr_block_dev *indr_dev;
702 indr_dev = tc_indr_block_dev_get(dev);
706 indr_block_cb = tc_indr_block_cb_add(indr_dev, cb_priv, cb, cb_ident);
707 err = PTR_ERR_OR_ZERO(indr_block_cb);
711 tc_indr_block_ing_cmd(indr_dev, indr_block_cb, TC_BLOCK_BIND);
715 tc_indr_block_dev_put(indr_dev);
718 EXPORT_SYMBOL_GPL(__tc_indr_block_cb_register);
720 int tc_indr_block_cb_register(struct net_device *dev, void *cb_priv,
721 tc_indr_block_bind_cb_t *cb, void *cb_ident)
726 err = __tc_indr_block_cb_register(dev, cb_priv, cb, cb_ident);
731 EXPORT_SYMBOL_GPL(tc_indr_block_cb_register);
733 void __tc_indr_block_cb_unregister(struct net_device *dev,
734 tc_indr_block_bind_cb_t *cb, void *cb_ident)
736 struct tc_indr_block_cb *indr_block_cb;
737 struct tc_indr_block_dev *indr_dev;
739 indr_dev = tc_indr_block_dev_lookup(dev);
743 indr_block_cb = tc_indr_block_cb_lookup(indr_dev, cb, cb_ident);
747 /* Send unbind message if required to free any block cbs. */
748 tc_indr_block_ing_cmd(indr_dev, indr_block_cb, TC_BLOCK_UNBIND);
749 tc_indr_block_cb_del(indr_block_cb);
750 tc_indr_block_dev_put(indr_dev);
752 EXPORT_SYMBOL_GPL(__tc_indr_block_cb_unregister);
754 void tc_indr_block_cb_unregister(struct net_device *dev,
755 tc_indr_block_bind_cb_t *cb, void *cb_ident)
758 __tc_indr_block_cb_unregister(dev, cb, cb_ident);
761 EXPORT_SYMBOL_GPL(tc_indr_block_cb_unregister);
763 static void tc_indr_block_call(struct tcf_block *block, struct net_device *dev,
764 struct tcf_block_ext_info *ei,
765 enum tc_block_command command,
766 struct netlink_ext_ack *extack)
768 struct tc_indr_block_cb *indr_block_cb;
769 struct tc_indr_block_dev *indr_dev;
770 struct tc_block_offload bo = {
772 .binder_type = ei->binder_type,
777 indr_dev = tc_indr_block_dev_lookup(dev);
781 indr_dev->block = command == TC_BLOCK_BIND ? block : NULL;
783 list_for_each_entry(indr_block_cb, &indr_dev->cb_list, list)
784 indr_block_cb->cb(dev, indr_block_cb->cb_priv, TC_SETUP_BLOCK,
788 static bool tcf_block_offload_in_use(struct tcf_block *block)
790 return block->offloadcnt;
793 static int tcf_block_offload_cmd(struct tcf_block *block,
794 struct net_device *dev,
795 struct tcf_block_ext_info *ei,
796 enum tc_block_command command,
797 struct netlink_ext_ack *extack)
799 struct tc_block_offload bo = {};
801 bo.command = command;
802 bo.binder_type = ei->binder_type;
805 return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
808 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
809 struct tcf_block_ext_info *ei,
810 struct netlink_ext_ack *extack)
812 struct net_device *dev = q->dev_queue->dev;
815 if (!dev->netdev_ops->ndo_setup_tc)
816 goto no_offload_dev_inc;
818 /* If tc offload feature is disabled and the block we try to bind
819 * to already has some offloaded filters, forbid to bind.
821 if (!tc_can_offload(dev) && tcf_block_offload_in_use(block)) {
822 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
826 err = tcf_block_offload_cmd(block, dev, ei, TC_BLOCK_BIND, extack);
827 if (err == -EOPNOTSUPP)
828 goto no_offload_dev_inc;
832 tc_indr_block_call(block, dev, ei, TC_BLOCK_BIND, extack);
836 if (tcf_block_offload_in_use(block))
838 block->nooffloaddevcnt++;
839 tc_indr_block_call(block, dev, ei, TC_BLOCK_BIND, extack);
843 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
844 struct tcf_block_ext_info *ei)
846 struct net_device *dev = q->dev_queue->dev;
849 tc_indr_block_call(block, dev, ei, TC_BLOCK_UNBIND, NULL);
851 if (!dev->netdev_ops->ndo_setup_tc)
852 goto no_offload_dev_dec;
853 err = tcf_block_offload_cmd(block, dev, ei, TC_BLOCK_UNBIND, NULL);
854 if (err == -EOPNOTSUPP)
855 goto no_offload_dev_dec;
859 WARN_ON(block->nooffloaddevcnt-- == 0);
863 tcf_chain0_head_change_cb_add(struct tcf_block *block,
864 struct tcf_block_ext_info *ei,
865 struct netlink_ext_ack *extack)
867 struct tcf_filter_chain_list_item *item;
868 struct tcf_chain *chain0;
870 item = kmalloc(sizeof(*item), GFP_KERNEL);
872 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
875 item->chain_head_change = ei->chain_head_change;
876 item->chain_head_change_priv = ei->chain_head_change_priv;
878 mutex_lock(&block->lock);
879 chain0 = block->chain0.chain;
881 tcf_chain_hold(chain0);
883 list_add(&item->list, &block->chain0.filter_chain_list);
884 mutex_unlock(&block->lock);
887 struct tcf_proto *tp_head;
889 mutex_lock(&chain0->filter_chain_lock);
891 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
893 tcf_chain_head_change_item(item, tp_head);
895 mutex_lock(&block->lock);
896 list_add(&item->list, &block->chain0.filter_chain_list);
897 mutex_unlock(&block->lock);
899 mutex_unlock(&chain0->filter_chain_lock);
900 tcf_chain_put(chain0);
907 tcf_chain0_head_change_cb_del(struct tcf_block *block,
908 struct tcf_block_ext_info *ei)
910 struct tcf_filter_chain_list_item *item;
912 mutex_lock(&block->lock);
913 list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
914 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
915 (item->chain_head_change == ei->chain_head_change &&
916 item->chain_head_change_priv == ei->chain_head_change_priv)) {
917 if (block->chain0.chain)
918 tcf_chain_head_change_item(item, NULL);
919 list_del(&item->list);
920 mutex_unlock(&block->lock);
926 mutex_unlock(&block->lock);
931 spinlock_t idr_lock; /* Protects idr */
935 static unsigned int tcf_net_id;
937 static int tcf_block_insert(struct tcf_block *block, struct net *net,
938 struct netlink_ext_ack *extack)
940 struct tcf_net *tn = net_generic(net, tcf_net_id);
943 idr_preload(GFP_KERNEL);
944 spin_lock(&tn->idr_lock);
945 err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
947 spin_unlock(&tn->idr_lock);
953 static void tcf_block_remove(struct tcf_block *block, struct net *net)
955 struct tcf_net *tn = net_generic(net, tcf_net_id);
957 spin_lock(&tn->idr_lock);
958 idr_remove(&tn->idr, block->index);
959 spin_unlock(&tn->idr_lock);
962 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
964 struct netlink_ext_ack *extack)
966 struct tcf_block *block;
968 block = kzalloc(sizeof(*block), GFP_KERNEL);
970 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
971 return ERR_PTR(-ENOMEM);
973 mutex_init(&block->lock);
974 INIT_LIST_HEAD(&block->chain_list);
975 INIT_LIST_HEAD(&block->cb_list);
976 INIT_LIST_HEAD(&block->owner_list);
977 INIT_LIST_HEAD(&block->chain0.filter_chain_list);
979 refcount_set(&block->refcnt, 1);
981 block->index = block_index;
983 /* Don't store q pointer for blocks which are shared */
984 if (!tcf_block_shared(block))
989 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
991 struct tcf_net *tn = net_generic(net, tcf_net_id);
993 return idr_find(&tn->idr, block_index);
996 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
998 struct tcf_block *block;
1001 block = tcf_block_lookup(net, block_index);
1002 if (block && !refcount_inc_not_zero(&block->refcnt))
1009 static struct tcf_chain *
1010 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1012 mutex_lock(&block->lock);
1014 chain = list_is_last(&chain->list, &block->chain_list) ?
1015 NULL : list_next_entry(chain, list);
1017 chain = list_first_entry_or_null(&block->chain_list,
1018 struct tcf_chain, list);
1020 /* skip all action-only chains */
1021 while (chain && tcf_chain_held_by_acts_only(chain))
1022 chain = list_is_last(&chain->list, &block->chain_list) ?
1023 NULL : list_next_entry(chain, list);
1026 tcf_chain_hold(chain);
1027 mutex_unlock(&block->lock);
1032 /* Function to be used by all clients that want to iterate over all chains on
1033 * block. It properly obtains block->lock and takes reference to chain before
1034 * returning it. Users of this function must be tolerant to concurrent chain
1035 * insertion/deletion or ensure that no concurrent chain modification is
1036 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1037 * consistent dump because rtnl lock is released each time skb is filled with
1038 * data and sent to user-space.
1042 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1044 struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
1047 tcf_chain_put(chain);
1051 EXPORT_SYMBOL(tcf_get_next_chain);
1053 static struct tcf_proto *
1054 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1059 mutex_lock(&chain->filter_chain_lock);
1062 tp = tcf_chain_dereference(chain->filter_chain, chain);
1063 } else if (tcf_proto_is_deleting(tp)) {
1064 /* 'deleting' flag is set and chain->filter_chain_lock was
1065 * unlocked, which means next pointer could be invalid. Restart
1068 prio = tp->prio + 1;
1069 tp = tcf_chain_dereference(chain->filter_chain, chain);
1071 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1072 if (!tp->deleting && tp->prio >= prio)
1075 tp = tcf_chain_dereference(tp->next, chain);
1081 mutex_unlock(&chain->filter_chain_lock);
1086 /* Function to be used by all clients that want to iterate over all tp's on
1087 * chain. Users of this function must be tolerant to concurrent tp
1088 * insertion/deletion or ensure that no concurrent chain modification is
1089 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1090 * consistent dump because rtnl lock is released each time skb is filled with
1091 * data and sent to user-space.
1095 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp,
1098 struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1101 tcf_proto_put(tp, rtnl_held, NULL);
1105 EXPORT_SYMBOL(tcf_get_next_proto);
1107 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1109 struct tcf_chain *chain;
1111 /* Last reference to block. At this point chains cannot be added or
1112 * removed concurrently.
1114 for (chain = tcf_get_next_chain(block, NULL);
1116 chain = tcf_get_next_chain(block, chain)) {
1117 tcf_chain_put_explicitly_created(chain);
1118 tcf_chain_flush(chain, rtnl_held);
1122 /* Lookup Qdisc and increments its reference counter.
1123 * Set parent, if necessary.
1126 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1127 u32 *parent, int ifindex, bool rtnl_held,
1128 struct netlink_ext_ack *extack)
1130 const struct Qdisc_class_ops *cops;
1131 struct net_device *dev;
1134 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1140 dev = dev_get_by_index_rcu(net, ifindex);
1149 *parent = (*q)->handle;
1151 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1153 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1159 *q = qdisc_refcount_inc_nz(*q);
1161 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1166 /* Is it classful? */
1167 cops = (*q)->ops->cl_ops;
1169 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1174 if (!cops->tcf_block) {
1175 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1181 /* At this point we know that qdisc is not noop_qdisc,
1182 * which means that qdisc holds a reference to net_device
1183 * and we hold a reference to qdisc, so it is safe to release
1195 qdisc_put_unlocked(*q);
1201 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1202 int ifindex, struct netlink_ext_ack *extack)
1204 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1207 /* Do we search for filter, attached to class? */
1208 if (TC_H_MIN(parent)) {
1209 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1211 *cl = cops->find(q, parent);
1213 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1221 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1222 unsigned long cl, int ifindex,
1224 struct netlink_ext_ack *extack)
1226 struct tcf_block *block;
1228 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1229 block = tcf_block_refcnt_get(net, block_index);
1231 NL_SET_ERR_MSG(extack, "Block of given index was not found");
1232 return ERR_PTR(-EINVAL);
1235 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1237 block = cops->tcf_block(q, cl, extack);
1239 return ERR_PTR(-EINVAL);
1241 if (tcf_block_shared(block)) {
1242 NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1243 return ERR_PTR(-EOPNOTSUPP);
1246 /* Always take reference to block in order to support execution
1247 * of rules update path of cls API without rtnl lock. Caller
1248 * must release block when it is finished using it. 'if' block
1249 * of this conditional obtain reference to block by calling
1250 * tcf_block_refcnt_get().
1252 refcount_inc(&block->refcnt);
1258 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1259 struct tcf_block_ext_info *ei, bool rtnl_held)
1261 if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1262 /* Flushing/putting all chains will cause the block to be
1263 * deallocated when last chain is freed. However, if chain_list
1264 * is empty, block has to be manually deallocated. After block
1265 * reference counter reached 0, it is no longer possible to
1266 * increment it or add new chains to block.
1268 bool free_block = list_empty(&block->chain_list);
1270 mutex_unlock(&block->lock);
1271 if (tcf_block_shared(block))
1272 tcf_block_remove(block, block->net);
1275 tcf_block_offload_unbind(block, q, ei);
1278 tcf_block_destroy(block);
1280 tcf_block_flush_all_chains(block, rtnl_held);
1282 tcf_block_offload_unbind(block, q, ei);
1286 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1288 __tcf_block_put(block, NULL, NULL, rtnl_held);
1292 * Set q, parent, cl when appropriate.
1295 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1296 u32 *parent, unsigned long *cl,
1297 int ifindex, u32 block_index,
1298 struct netlink_ext_ack *extack)
1300 struct tcf_block *block;
1305 err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1309 err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1313 block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1314 if (IS_ERR(block)) {
1315 err = PTR_ERR(block);
1326 return ERR_PTR(err);
1329 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1332 if (!IS_ERR_OR_NULL(block))
1333 tcf_block_refcnt_put(block, rtnl_held);
1339 qdisc_put_unlocked(q);
1343 struct tcf_block_owner_item {
1344 struct list_head list;
1346 enum tcf_block_binder_type binder_type;
1350 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1352 enum tcf_block_binder_type binder_type)
1354 if (block->keep_dst &&
1355 binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1356 binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1357 netif_keep_dst(qdisc_dev(q));
1360 void tcf_block_netif_keep_dst(struct tcf_block *block)
1362 struct tcf_block_owner_item *item;
1364 block->keep_dst = true;
1365 list_for_each_entry(item, &block->owner_list, list)
1366 tcf_block_owner_netif_keep_dst(block, item->q,
1369 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1371 static int tcf_block_owner_add(struct tcf_block *block,
1373 enum tcf_block_binder_type binder_type)
1375 struct tcf_block_owner_item *item;
1377 item = kmalloc(sizeof(*item), GFP_KERNEL);
1381 item->binder_type = binder_type;
1382 list_add(&item->list, &block->owner_list);
1386 static void tcf_block_owner_del(struct tcf_block *block,
1388 enum tcf_block_binder_type binder_type)
1390 struct tcf_block_owner_item *item;
1392 list_for_each_entry(item, &block->owner_list, list) {
1393 if (item->q == q && item->binder_type == binder_type) {
1394 list_del(&item->list);
1402 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1403 struct tcf_block_ext_info *ei,
1404 struct netlink_ext_ack *extack)
1406 struct net *net = qdisc_net(q);
1407 struct tcf_block *block = NULL;
1410 if (ei->block_index)
1411 /* block_index not 0 means the shared block is requested */
1412 block = tcf_block_refcnt_get(net, ei->block_index);
1415 block = tcf_block_create(net, q, ei->block_index, extack);
1417 return PTR_ERR(block);
1418 if (tcf_block_shared(block)) {
1419 err = tcf_block_insert(block, net, extack);
1421 goto err_block_insert;
1425 err = tcf_block_owner_add(block, q, ei->binder_type);
1427 goto err_block_owner_add;
1429 tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1431 err = tcf_chain0_head_change_cb_add(block, ei, extack);
1433 goto err_chain0_head_change_cb_add;
1435 err = tcf_block_offload_bind(block, q, ei, extack);
1437 goto err_block_offload_bind;
1442 err_block_offload_bind:
1443 tcf_chain0_head_change_cb_del(block, ei);
1444 err_chain0_head_change_cb_add:
1445 tcf_block_owner_del(block, q, ei->binder_type);
1446 err_block_owner_add:
1448 tcf_block_refcnt_put(block, true);
1451 EXPORT_SYMBOL(tcf_block_get_ext);
1453 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1455 struct tcf_proto __rcu **p_filter_chain = priv;
1457 rcu_assign_pointer(*p_filter_chain, tp_head);
1460 int tcf_block_get(struct tcf_block **p_block,
1461 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1462 struct netlink_ext_ack *extack)
1464 struct tcf_block_ext_info ei = {
1465 .chain_head_change = tcf_chain_head_change_dflt,
1466 .chain_head_change_priv = p_filter_chain,
1469 WARN_ON(!p_filter_chain);
1470 return tcf_block_get_ext(p_block, q, &ei, extack);
1472 EXPORT_SYMBOL(tcf_block_get);
1474 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1475 * actions should be all removed after flushing.
1477 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1478 struct tcf_block_ext_info *ei)
1482 tcf_chain0_head_change_cb_del(block, ei);
1483 tcf_block_owner_del(block, q, ei->binder_type);
1485 __tcf_block_put(block, q, ei, true);
1487 EXPORT_SYMBOL(tcf_block_put_ext);
1489 void tcf_block_put(struct tcf_block *block)
1491 struct tcf_block_ext_info ei = {0, };
1495 tcf_block_put_ext(block, block->q, &ei);
1498 EXPORT_SYMBOL(tcf_block_put);
1500 struct tcf_block_cb {
1501 struct list_head list;
1505 unsigned int refcnt;
1508 void *tcf_block_cb_priv(struct tcf_block_cb *block_cb)
1510 return block_cb->cb_priv;
1512 EXPORT_SYMBOL(tcf_block_cb_priv);
1514 struct tcf_block_cb *tcf_block_cb_lookup(struct tcf_block *block,
1515 tc_setup_cb_t *cb, void *cb_ident)
1516 { struct tcf_block_cb *block_cb;
1518 list_for_each_entry(block_cb, &block->cb_list, list)
1519 if (block_cb->cb == cb && block_cb->cb_ident == cb_ident)
1523 EXPORT_SYMBOL(tcf_block_cb_lookup);
1525 void tcf_block_cb_incref(struct tcf_block_cb *block_cb)
1529 EXPORT_SYMBOL(tcf_block_cb_incref);
1531 unsigned int tcf_block_cb_decref(struct tcf_block_cb *block_cb)
1533 return --block_cb->refcnt;
1535 EXPORT_SYMBOL(tcf_block_cb_decref);
1538 tcf_block_playback_offloads(struct tcf_block *block, tc_setup_cb_t *cb,
1539 void *cb_priv, bool add, bool offload_in_use,
1540 struct netlink_ext_ack *extack)
1542 struct tcf_chain *chain, *chain_prev;
1543 struct tcf_proto *tp, *tp_prev;
1546 for (chain = __tcf_get_next_chain(block, NULL);
1549 chain = __tcf_get_next_chain(block, chain),
1550 tcf_chain_put(chain_prev)) {
1551 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1553 tp = __tcf_get_next_proto(chain, tp),
1554 tcf_proto_put(tp_prev, true, NULL)) {
1555 if (tp->ops->reoffload) {
1556 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1559 goto err_playback_remove;
1560 } else if (add && offload_in_use) {
1562 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1563 goto err_playback_remove;
1570 err_playback_remove:
1571 tcf_proto_put(tp, true, NULL);
1572 tcf_chain_put(chain);
1573 tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1578 struct tcf_block_cb *__tcf_block_cb_register(struct tcf_block *block,
1579 tc_setup_cb_t *cb, void *cb_ident,
1581 struct netlink_ext_ack *extack)
1583 struct tcf_block_cb *block_cb;
1586 /* Replay any already present rules */
1587 err = tcf_block_playback_offloads(block, cb, cb_priv, true,
1588 tcf_block_offload_in_use(block),
1591 return ERR_PTR(err);
1593 block_cb = kzalloc(sizeof(*block_cb), GFP_KERNEL);
1595 return ERR_PTR(-ENOMEM);
1597 block_cb->cb_ident = cb_ident;
1598 block_cb->cb_priv = cb_priv;
1599 list_add(&block_cb->list, &block->cb_list);
1602 EXPORT_SYMBOL(__tcf_block_cb_register);
1604 int tcf_block_cb_register(struct tcf_block *block,
1605 tc_setup_cb_t *cb, void *cb_ident,
1606 void *cb_priv, struct netlink_ext_ack *extack)
1608 struct tcf_block_cb *block_cb;
1610 block_cb = __tcf_block_cb_register(block, cb, cb_ident, cb_priv,
1612 return PTR_ERR_OR_ZERO(block_cb);
1614 EXPORT_SYMBOL(tcf_block_cb_register);
1616 void __tcf_block_cb_unregister(struct tcf_block *block,
1617 struct tcf_block_cb *block_cb)
1619 tcf_block_playback_offloads(block, block_cb->cb, block_cb->cb_priv,
1620 false, tcf_block_offload_in_use(block),
1622 list_del(&block_cb->list);
1625 EXPORT_SYMBOL(__tcf_block_cb_unregister);
1627 void tcf_block_cb_unregister(struct tcf_block *block,
1628 tc_setup_cb_t *cb, void *cb_ident)
1630 struct tcf_block_cb *block_cb;
1632 block_cb = tcf_block_cb_lookup(block, cb, cb_ident);
1635 __tcf_block_cb_unregister(block, block_cb);
1637 EXPORT_SYMBOL(tcf_block_cb_unregister);
1639 /* Main classifier routine: scans classifier chain attached
1640 * to this qdisc, (optionally) tests for protocol and asks
1641 * specific classifiers.
1643 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1644 struct tcf_result *res, bool compat_mode)
1646 #ifdef CONFIG_NET_CLS_ACT
1647 const int max_reclassify_loop = 4;
1648 const struct tcf_proto *orig_tp = tp;
1649 const struct tcf_proto *first_tp;
1654 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1655 __be16 protocol = tc_skb_protocol(skb);
1658 if (tp->protocol != protocol &&
1659 tp->protocol != htons(ETH_P_ALL))
1662 err = tp->classify(skb, tp, res);
1663 #ifdef CONFIG_NET_CLS_ACT
1664 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1667 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1668 first_tp = res->goto_tp;
1676 return TC_ACT_UNSPEC; /* signal: continue lookup */
1677 #ifdef CONFIG_NET_CLS_ACT
1679 if (unlikely(limit++ >= max_reclassify_loop)) {
1680 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1681 tp->chain->block->index,
1683 ntohs(tp->protocol));
1691 EXPORT_SYMBOL(tcf_classify);
1693 struct tcf_chain_info {
1694 struct tcf_proto __rcu **pprev;
1695 struct tcf_proto __rcu *next;
1698 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1699 struct tcf_chain_info *chain_info)
1701 return tcf_chain_dereference(*chain_info->pprev, chain);
1704 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1705 struct tcf_chain_info *chain_info,
1706 struct tcf_proto *tp)
1708 if (chain->flushing)
1711 if (*chain_info->pprev == chain->filter_chain)
1712 tcf_chain0_head_change(chain, tp);
1714 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1715 rcu_assign_pointer(*chain_info->pprev, tp);
1720 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1721 struct tcf_chain_info *chain_info,
1722 struct tcf_proto *tp)
1724 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1726 tcf_proto_mark_delete(tp);
1727 if (tp == chain->filter_chain)
1728 tcf_chain0_head_change(chain, next);
1729 RCU_INIT_POINTER(*chain_info->pprev, next);
1732 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1733 struct tcf_chain_info *chain_info,
1734 u32 protocol, u32 prio,
1735 bool prio_allocate);
1737 /* Try to insert new proto.
1738 * If proto with specified priority already exists, free new proto
1739 * and return existing one.
1742 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1743 struct tcf_proto *tp_new,
1744 u32 protocol, u32 prio,
1747 struct tcf_chain_info chain_info;
1748 struct tcf_proto *tp;
1751 mutex_lock(&chain->filter_chain_lock);
1753 tp = tcf_chain_tp_find(chain, &chain_info,
1754 protocol, prio, false);
1756 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1757 mutex_unlock(&chain->filter_chain_lock);
1760 tcf_proto_destroy(tp_new, rtnl_held, NULL);
1763 tcf_proto_destroy(tp_new, rtnl_held, NULL);
1764 tp_new = ERR_PTR(err);
1770 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1771 struct tcf_proto *tp, bool rtnl_held,
1772 struct netlink_ext_ack *extack)
1774 struct tcf_chain_info chain_info;
1775 struct tcf_proto *tp_iter;
1776 struct tcf_proto **pprev;
1777 struct tcf_proto *next;
1779 mutex_lock(&chain->filter_chain_lock);
1781 /* Atomically find and remove tp from chain. */
1782 for (pprev = &chain->filter_chain;
1783 (tp_iter = tcf_chain_dereference(*pprev, chain));
1784 pprev = &tp_iter->next) {
1785 if (tp_iter == tp) {
1786 chain_info.pprev = pprev;
1787 chain_info.next = tp_iter->next;
1788 WARN_ON(tp_iter->deleting);
1792 /* Verify that tp still exists and no new filters were inserted
1794 * Mark tp for deletion if it is empty.
1796 if (!tp_iter || !tcf_proto_check_delete(tp, rtnl_held)) {
1797 mutex_unlock(&chain->filter_chain_lock);
1801 next = tcf_chain_dereference(chain_info.next, chain);
1802 if (tp == chain->filter_chain)
1803 tcf_chain0_head_change(chain, next);
1804 RCU_INIT_POINTER(*chain_info.pprev, next);
1805 mutex_unlock(&chain->filter_chain_lock);
1807 tcf_proto_put(tp, rtnl_held, extack);
1810 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1811 struct tcf_chain_info *chain_info,
1812 u32 protocol, u32 prio,
1815 struct tcf_proto **pprev;
1816 struct tcf_proto *tp;
1818 /* Check the chain for existence of proto-tcf with this priority */
1819 for (pprev = &chain->filter_chain;
1820 (tp = tcf_chain_dereference(*pprev, chain));
1821 pprev = &tp->next) {
1822 if (tp->prio >= prio) {
1823 if (tp->prio == prio) {
1824 if (prio_allocate ||
1825 (tp->protocol != protocol && protocol))
1826 return ERR_PTR(-EINVAL);
1833 chain_info->pprev = pprev;
1835 chain_info->next = tp->next;
1838 chain_info->next = NULL;
1843 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1844 struct tcf_proto *tp, struct tcf_block *block,
1845 struct Qdisc *q, u32 parent, void *fh,
1846 u32 portid, u32 seq, u16 flags, int event,
1850 struct nlmsghdr *nlh;
1851 unsigned char *b = skb_tail_pointer(skb);
1853 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1855 goto out_nlmsg_trim;
1856 tcm = nlmsg_data(nlh);
1857 tcm->tcm_family = AF_UNSPEC;
1861 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1862 tcm->tcm_parent = parent;
1864 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1865 tcm->tcm_block_index = block->index;
1867 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1868 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1869 goto nla_put_failure;
1870 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1871 goto nla_put_failure;
1873 tcm->tcm_handle = 0;
1875 if (tp->ops->dump &&
1876 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
1877 goto nla_put_failure;
1879 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1888 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1889 struct nlmsghdr *n, struct tcf_proto *tp,
1890 struct tcf_block *block, struct Qdisc *q,
1891 u32 parent, void *fh, int event, bool unicast,
1894 struct sk_buff *skb;
1895 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1897 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1901 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1902 n->nlmsg_seq, n->nlmsg_flags, event,
1909 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1911 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1912 n->nlmsg_flags & NLM_F_ECHO);
1915 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1916 struct nlmsghdr *n, struct tcf_proto *tp,
1917 struct tcf_block *block, struct Qdisc *q,
1918 u32 parent, void *fh, bool unicast, bool *last,
1919 bool rtnl_held, struct netlink_ext_ack *extack)
1921 struct sk_buff *skb;
1922 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1925 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1929 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1930 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
1932 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1937 err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
1944 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1946 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1947 n->nlmsg_flags & NLM_F_ECHO);
1949 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1953 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1954 struct tcf_block *block, struct Qdisc *q,
1955 u32 parent, struct nlmsghdr *n,
1956 struct tcf_chain *chain, int event,
1959 struct tcf_proto *tp;
1961 for (tp = tcf_get_next_proto(chain, NULL, rtnl_held);
1962 tp; tp = tcf_get_next_proto(chain, tp, rtnl_held))
1963 tfilter_notify(net, oskb, n, tp, block,
1964 q, parent, NULL, event, false, rtnl_held);
1967 static void tfilter_put(struct tcf_proto *tp, void *fh)
1969 if (tp->ops->put && fh)
1970 tp->ops->put(tp, fh);
1973 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1974 struct netlink_ext_ack *extack)
1976 struct net *net = sock_net(skb->sk);
1977 struct nlattr *tca[TCA_MAX + 1];
1984 struct Qdisc *q = NULL;
1985 struct tcf_chain_info chain_info;
1986 struct tcf_chain *chain = NULL;
1987 struct tcf_block *block;
1988 struct tcf_proto *tp;
1993 bool rtnl_held = false;
1995 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2001 err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, rtm_tca_policy, extack);
2006 protocol = TC_H_MIN(t->tcm_info);
2007 prio = TC_H_MAJ(t->tcm_info);
2008 prio_allocate = false;
2009 parent = t->tcm_parent;
2015 /* If no priority is provided by the user,
2018 if (n->nlmsg_flags & NLM_F_CREATE) {
2019 prio = TC_H_MAKE(0x80000000U, 0U);
2020 prio_allocate = true;
2022 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2027 /* Find head of filter chain. */
2029 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2033 /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2034 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2035 * type is not specified, classifier is not unlocked.
2038 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2039 !tca[TCA_KIND] || !tcf_proto_is_unlocked(nla_data(tca[TCA_KIND]))) {
2044 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2048 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2050 if (IS_ERR(block)) {
2051 err = PTR_ERR(block);
2055 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2056 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2057 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2061 chain = tcf_chain_get(block, chain_index, true);
2063 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2068 mutex_lock(&chain->filter_chain_lock);
2069 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2070 prio, prio_allocate);
2072 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2078 struct tcf_proto *tp_new = NULL;
2080 if (chain->flushing) {
2085 /* Proto-tcf does not exist, create new one */
2087 if (tca[TCA_KIND] == NULL || !protocol) {
2088 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2093 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2094 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2100 prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2103 mutex_unlock(&chain->filter_chain_lock);
2104 tp_new = tcf_proto_create(nla_data(tca[TCA_KIND]),
2105 protocol, prio, chain, rtnl_held,
2107 if (IS_ERR(tp_new)) {
2108 err = PTR_ERR(tp_new);
2113 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2120 mutex_unlock(&chain->filter_chain_lock);
2123 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2124 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2129 fh = tp->ops->get(tp, t->tcm_handle);
2132 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2133 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2137 } else if (n->nlmsg_flags & NLM_F_EXCL) {
2138 tfilter_put(tp, fh);
2139 NL_SET_ERR_MSG(extack, "Filter already exists");
2144 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2145 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2150 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2151 n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
2154 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2155 RTM_NEWTFILTER, false, rtnl_held);
2156 tfilter_put(tp, fh);
2160 if (err && tp_created)
2161 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2164 if (tp && !IS_ERR(tp))
2165 tcf_proto_put(tp, rtnl_held, NULL);
2167 tcf_chain_put(chain);
2169 tcf_block_release(q, block, rtnl_held);
2174 if (err == -EAGAIN) {
2175 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2179 /* Replay the request. */
2185 mutex_unlock(&chain->filter_chain_lock);
2189 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2190 struct netlink_ext_ack *extack)
2192 struct net *net = sock_net(skb->sk);
2193 struct nlattr *tca[TCA_MAX + 1];
2199 struct Qdisc *q = NULL;
2200 struct tcf_chain_info chain_info;
2201 struct tcf_chain *chain = NULL;
2202 struct tcf_block *block = NULL;
2203 struct tcf_proto *tp = NULL;
2204 unsigned long cl = 0;
2207 bool rtnl_held = false;
2209 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2212 err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, rtm_tca_policy, extack);
2217 protocol = TC_H_MIN(t->tcm_info);
2218 prio = TC_H_MAJ(t->tcm_info);
2219 parent = t->tcm_parent;
2221 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2222 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2226 /* Find head of filter chain. */
2228 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2232 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2233 * found), qdisc is not unlocked, classifier type is not specified,
2234 * classifier is not unlocked.
2237 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2238 !tca[TCA_KIND] || !tcf_proto_is_unlocked(nla_data(tca[TCA_KIND]))) {
2243 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2247 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2249 if (IS_ERR(block)) {
2250 err = PTR_ERR(block);
2254 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2255 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2256 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2260 chain = tcf_chain_get(block, chain_index, false);
2262 /* User requested flush on non-existent chain. Nothing to do,
2263 * so just return success.
2269 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2275 tfilter_notify_chain(net, skb, block, q, parent, n,
2276 chain, RTM_DELTFILTER, rtnl_held);
2277 tcf_chain_flush(chain, rtnl_held);
2282 mutex_lock(&chain->filter_chain_lock);
2283 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2285 if (!tp || IS_ERR(tp)) {
2286 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2287 err = tp ? PTR_ERR(tp) : -ENOENT;
2289 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2290 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2293 } else if (t->tcm_handle == 0) {
2294 tcf_chain_tp_remove(chain, &chain_info, tp);
2295 mutex_unlock(&chain->filter_chain_lock);
2297 tcf_proto_put(tp, rtnl_held, NULL);
2298 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2299 RTM_DELTFILTER, false, rtnl_held);
2303 mutex_unlock(&chain->filter_chain_lock);
2305 fh = tp->ops->get(tp, t->tcm_handle);
2308 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2313 err = tfilter_del_notify(net, skb, n, tp, block,
2314 q, parent, fh, false, &last,
2320 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2325 if (tp && !IS_ERR(tp))
2326 tcf_proto_put(tp, rtnl_held, NULL);
2327 tcf_chain_put(chain);
2329 tcf_block_release(q, block, rtnl_held);
2337 mutex_unlock(&chain->filter_chain_lock);
2341 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2342 struct netlink_ext_ack *extack)
2344 struct net *net = sock_net(skb->sk);
2345 struct nlattr *tca[TCA_MAX + 1];
2351 struct Qdisc *q = NULL;
2352 struct tcf_chain_info chain_info;
2353 struct tcf_chain *chain = NULL;
2354 struct tcf_block *block = NULL;
2355 struct tcf_proto *tp = NULL;
2356 unsigned long cl = 0;
2359 bool rtnl_held = false;
2361 err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, rtm_tca_policy, extack);
2366 protocol = TC_H_MIN(t->tcm_info);
2367 prio = TC_H_MAJ(t->tcm_info);
2368 parent = t->tcm_parent;
2371 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2375 /* Find head of filter chain. */
2377 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2381 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2382 * unlocked, classifier type is not specified, classifier is not
2385 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2386 !tca[TCA_KIND] || !tcf_proto_is_unlocked(nla_data(tca[TCA_KIND]))) {
2391 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2395 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2397 if (IS_ERR(block)) {
2398 err = PTR_ERR(block);
2402 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2403 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2404 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2408 chain = tcf_chain_get(block, chain_index, false);
2410 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2415 mutex_lock(&chain->filter_chain_lock);
2416 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2418 mutex_unlock(&chain->filter_chain_lock);
2419 if (!tp || IS_ERR(tp)) {
2420 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2421 err = tp ? PTR_ERR(tp) : -ENOENT;
2423 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2424 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2429 fh = tp->ops->get(tp, t->tcm_handle);
2432 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2435 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2436 fh, RTM_NEWTFILTER, true, rtnl_held);
2438 NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2441 tfilter_put(tp, fh);
2444 if (tp && !IS_ERR(tp))
2445 tcf_proto_put(tp, rtnl_held, NULL);
2446 tcf_chain_put(chain);
2448 tcf_block_release(q, block, rtnl_held);
2456 struct tcf_dump_args {
2457 struct tcf_walker w;
2458 struct sk_buff *skb;
2459 struct netlink_callback *cb;
2460 struct tcf_block *block;
2465 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2467 struct tcf_dump_args *a = (void *)arg;
2468 struct net *net = sock_net(a->skb->sk);
2470 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2471 n, NETLINK_CB(a->cb->skb).portid,
2472 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2473 RTM_NEWTFILTER, true);
2476 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2477 struct sk_buff *skb, struct netlink_callback *cb,
2478 long index_start, long *p_index)
2480 struct net *net = sock_net(skb->sk);
2481 struct tcf_block *block = chain->block;
2482 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2483 struct tcf_proto *tp, *tp_prev;
2484 struct tcf_dump_args arg;
2486 for (tp = __tcf_get_next_proto(chain, NULL);
2489 tp = __tcf_get_next_proto(chain, tp),
2490 tcf_proto_put(tp_prev, true, NULL),
2492 if (*p_index < index_start)
2494 if (TC_H_MAJ(tcm->tcm_info) &&
2495 TC_H_MAJ(tcm->tcm_info) != tp->prio)
2497 if (TC_H_MIN(tcm->tcm_info) &&
2498 TC_H_MIN(tcm->tcm_info) != tp->protocol)
2500 if (*p_index > index_start)
2501 memset(&cb->args[1], 0,
2502 sizeof(cb->args) - sizeof(cb->args[0]));
2503 if (cb->args[1] == 0) {
2504 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2505 NETLINK_CB(cb->skb).portid,
2506 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2507 RTM_NEWTFILTER, true) <= 0)
2513 arg.w.fn = tcf_node_dump;
2518 arg.parent = parent;
2520 arg.w.skip = cb->args[1] - 1;
2522 arg.w.cookie = cb->args[2];
2523 tp->ops->walk(tp, &arg.w, true);
2524 cb->args[2] = arg.w.cookie;
2525 cb->args[1] = arg.w.count + 1;
2532 tcf_proto_put(tp, true, NULL);
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);
2550 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2553 err = nlmsg_parse(cb->nlh, sizeof(*tcm), tca, TCA_MAX, NULL,
2558 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2559 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2562 /* If we work with block index, q is NULL and parent value
2563 * will never be used in the following code. The check
2564 * in tcf_fill_node prevents it. However, compiler does not
2565 * see that far, so set parent to zero to silence the warning
2566 * about parent being uninitialized.
2570 const struct Qdisc_class_ops *cops;
2571 struct net_device *dev;
2572 unsigned long cl = 0;
2574 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2578 parent = tcm->tcm_parent;
2583 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2587 cops = q->ops->cl_ops;
2590 if (!cops->tcf_block)
2592 if (TC_H_MIN(tcm->tcm_parent)) {
2593 cl = cops->find(q, tcm->tcm_parent);
2597 block = cops->tcf_block(q, cl, NULL);
2600 if (tcf_block_shared(block))
2604 index_start = cb->args[0];
2607 for (chain = __tcf_get_next_chain(block, NULL);
2610 chain = __tcf_get_next_chain(block, chain),
2611 tcf_chain_put(chain_prev)) {
2612 if (tca[TCA_CHAIN] &&
2613 nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2615 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2616 index_start, &index)) {
2617 tcf_chain_put(chain);
2623 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2624 tcf_block_refcnt_put(block, true);
2625 cb->args[0] = index;
2628 /* If we did no progress, the error (EMSGSIZE) is real */
2629 if (skb->len == 0 && err)
2634 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2635 void *tmplt_priv, u32 chain_index,
2636 struct net *net, struct sk_buff *skb,
2637 struct tcf_block *block,
2638 u32 portid, u32 seq, u16 flags, int event)
2640 unsigned char *b = skb_tail_pointer(skb);
2641 const struct tcf_proto_ops *ops;
2642 struct nlmsghdr *nlh;
2649 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2651 goto out_nlmsg_trim;
2652 tcm = nlmsg_data(nlh);
2653 tcm->tcm_family = AF_UNSPEC;
2656 tcm->tcm_handle = 0;
2658 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2659 tcm->tcm_parent = block->q->handle;
2661 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2662 tcm->tcm_block_index = block->index;
2665 if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2666 goto nla_put_failure;
2669 if (nla_put_string(skb, TCA_KIND, ops->kind))
2670 goto nla_put_failure;
2671 if (ops->tmplt_dump(skb, net, priv) < 0)
2672 goto nla_put_failure;
2675 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2684 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2685 u32 seq, u16 flags, int event, bool unicast)
2687 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2688 struct tcf_block *block = chain->block;
2689 struct net *net = block->net;
2690 struct sk_buff *skb;
2692 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2696 if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2697 chain->index, net, skb, block, portid,
2698 seq, flags, event) <= 0) {
2704 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2706 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2709 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2710 void *tmplt_priv, u32 chain_index,
2711 struct tcf_block *block, struct sk_buff *oskb,
2712 u32 seq, u16 flags, bool unicast)
2714 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2715 struct net *net = block->net;
2716 struct sk_buff *skb;
2718 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2722 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2723 block, portid, seq, flags, RTM_DELCHAIN) <= 0) {
2729 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2731 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2734 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2735 struct nlattr **tca,
2736 struct netlink_ext_ack *extack)
2738 const struct tcf_proto_ops *ops;
2741 /* If kind is not set, user did not specify template. */
2745 ops = tcf_proto_lookup_ops(nla_data(tca[TCA_KIND]), true, extack);
2747 return PTR_ERR(ops);
2748 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2749 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2753 tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2754 if (IS_ERR(tmplt_priv)) {
2755 module_put(ops->owner);
2756 return PTR_ERR(tmplt_priv);
2758 chain->tmplt_ops = ops;
2759 chain->tmplt_priv = tmplt_priv;
2763 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2766 /* If template ops are set, no work to do for us. */
2770 tmplt_ops->tmplt_destroy(tmplt_priv);
2771 module_put(tmplt_ops->owner);
2774 /* Add/delete/get a chain */
2776 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2777 struct netlink_ext_ack *extack)
2779 struct net *net = sock_net(skb->sk);
2780 struct nlattr *tca[TCA_MAX + 1];
2784 struct Qdisc *q = NULL;
2785 struct tcf_chain *chain = NULL;
2786 struct tcf_block *block;
2790 if (n->nlmsg_type != RTM_GETCHAIN &&
2791 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2795 err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, rtm_tca_policy, extack);
2800 parent = t->tcm_parent;
2803 block = tcf_block_find(net, &q, &parent, &cl,
2804 t->tcm_ifindex, t->tcm_block_index, extack);
2806 return PTR_ERR(block);
2808 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2809 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2810 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2815 mutex_lock(&block->lock);
2816 chain = tcf_chain_lookup(block, chain_index);
2817 if (n->nlmsg_type == RTM_NEWCHAIN) {
2819 if (tcf_chain_held_by_acts_only(chain)) {
2820 /* The chain exists only because there is
2821 * some action referencing it.
2823 tcf_chain_hold(chain);
2825 NL_SET_ERR_MSG(extack, "Filter chain already exists");
2827 goto errout_block_locked;
2830 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2831 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
2833 goto errout_block_locked;
2835 chain = tcf_chain_create(block, chain_index);
2837 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
2839 goto errout_block_locked;
2843 if (!chain || tcf_chain_held_by_acts_only(chain)) {
2844 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2846 goto errout_block_locked;
2848 tcf_chain_hold(chain);
2851 if (n->nlmsg_type == RTM_NEWCHAIN) {
2852 /* Modifying chain requires holding parent block lock. In case
2853 * the chain was successfully added, take a reference to the
2854 * chain. This ensures that an empty chain does not disappear at
2855 * the end of this function.
2857 tcf_chain_hold(chain);
2858 chain->explicitly_created = true;
2860 mutex_unlock(&block->lock);
2862 switch (n->nlmsg_type) {
2864 err = tc_chain_tmplt_add(chain, net, tca, extack);
2866 tcf_chain_put_explicitly_created(chain);
2870 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2871 RTM_NEWCHAIN, false);
2874 tfilter_notify_chain(net, skb, block, q, parent, n,
2875 chain, RTM_DELTFILTER, true);
2876 /* Flush the chain first as the user requested chain removal. */
2877 tcf_chain_flush(chain, true);
2878 /* In case the chain was successfully deleted, put a reference
2879 * to the chain previously taken during addition.
2881 tcf_chain_put_explicitly_created(chain);
2884 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
2885 n->nlmsg_seq, n->nlmsg_type, true);
2887 NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2891 NL_SET_ERR_MSG(extack, "Unsupported message type");
2896 tcf_chain_put(chain);
2898 tcf_block_release(q, block, true);
2900 /* Replay the request. */
2904 errout_block_locked:
2905 mutex_unlock(&block->lock);
2909 /* called with RTNL */
2910 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2912 struct net *net = sock_net(skb->sk);
2913 struct nlattr *tca[TCA_MAX + 1];
2914 struct Qdisc *q = NULL;
2915 struct tcf_block *block;
2916 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2917 struct tcf_chain *chain;
2923 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2926 err = nlmsg_parse(cb->nlh, sizeof(*tcm), tca, TCA_MAX, rtm_tca_policy,
2931 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2932 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2935 /* If we work with block index, q is NULL and parent value
2936 * will never be used in the following code. The check
2937 * in tcf_fill_node prevents it. However, compiler does not
2938 * see that far, so set parent to zero to silence the warning
2939 * about parent being uninitialized.
2943 const struct Qdisc_class_ops *cops;
2944 struct net_device *dev;
2945 unsigned long cl = 0;
2947 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2951 parent = tcm->tcm_parent;
2956 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2960 cops = q->ops->cl_ops;
2963 if (!cops->tcf_block)
2965 if (TC_H_MIN(tcm->tcm_parent)) {
2966 cl = cops->find(q, tcm->tcm_parent);
2970 block = cops->tcf_block(q, cl, NULL);
2973 if (tcf_block_shared(block))
2977 index_start = cb->args[0];
2980 mutex_lock(&block->lock);
2981 list_for_each_entry(chain, &block->chain_list, list) {
2982 if ((tca[TCA_CHAIN] &&
2983 nla_get_u32(tca[TCA_CHAIN]) != chain->index))
2985 if (index < index_start) {
2989 if (tcf_chain_held_by_acts_only(chain))
2991 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2992 chain->index, net, skb, block,
2993 NETLINK_CB(cb->skb).portid,
2994 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3000 mutex_unlock(&block->lock);
3002 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3003 tcf_block_refcnt_put(block, true);
3004 cb->args[0] = index;
3007 /* If we did no progress, the error (EMSGSIZE) is real */
3008 if (skb->len == 0 && err)
3013 void tcf_exts_destroy(struct tcf_exts *exts)
3015 #ifdef CONFIG_NET_CLS_ACT
3016 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3017 kfree(exts->actions);
3018 exts->nr_actions = 0;
3021 EXPORT_SYMBOL(tcf_exts_destroy);
3023 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3024 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
3025 bool rtnl_held, struct netlink_ext_ack *extack)
3027 #ifdef CONFIG_NET_CLS_ACT
3029 struct tc_action *act;
3030 size_t attr_size = 0;
3032 if (exts->police && tb[exts->police]) {
3033 act = tcf_action_init_1(net, tp, tb[exts->police],
3034 rate_tlv, "police", ovr,
3035 TCA_ACT_BIND, rtnl_held,
3038 return PTR_ERR(act);
3040 act->type = exts->type = TCA_OLD_COMPAT;
3041 exts->actions[0] = act;
3042 exts->nr_actions = 1;
3043 } else if (exts->action && tb[exts->action]) {
3046 err = tcf_action_init(net, tp, tb[exts->action],
3047 rate_tlv, NULL, ovr, TCA_ACT_BIND,
3048 exts->actions, &attr_size,
3052 exts->nr_actions = err;
3056 if ((exts->action && tb[exts->action]) ||
3057 (exts->police && tb[exts->police])) {
3058 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3065 EXPORT_SYMBOL(tcf_exts_validate);
3067 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3069 #ifdef CONFIG_NET_CLS_ACT
3070 struct tcf_exts old = *dst;
3073 tcf_exts_destroy(&old);
3076 EXPORT_SYMBOL(tcf_exts_change);
3078 #ifdef CONFIG_NET_CLS_ACT
3079 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3081 if (exts->nr_actions == 0)
3084 return exts->actions[0];
3088 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3090 #ifdef CONFIG_NET_CLS_ACT
3091 struct nlattr *nest;
3093 if (exts->action && tcf_exts_has_actions(exts)) {
3095 * again for backward compatible mode - we want
3096 * to work with both old and new modes of entering
3097 * tc data even if iproute2 was newer - jhs
3099 if (exts->type != TCA_OLD_COMPAT) {
3100 nest = nla_nest_start(skb, exts->action);
3102 goto nla_put_failure;
3104 if (tcf_action_dump(skb, exts->actions, 0, 0) < 0)
3105 goto nla_put_failure;
3106 nla_nest_end(skb, nest);
3107 } else if (exts->police) {
3108 struct tc_action *act = tcf_exts_first_act(exts);
3109 nest = nla_nest_start(skb, exts->police);
3110 if (nest == NULL || !act)
3111 goto nla_put_failure;
3112 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3113 goto nla_put_failure;
3114 nla_nest_end(skb, nest);
3120 nla_nest_cancel(skb, nest);
3126 EXPORT_SYMBOL(tcf_exts_dump);
3129 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3131 #ifdef CONFIG_NET_CLS_ACT
3132 struct tc_action *a = tcf_exts_first_act(exts);
3133 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3138 EXPORT_SYMBOL(tcf_exts_dump_stats);
3140 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3141 void *type_data, bool err_stop)
3143 struct tcf_block_cb *block_cb;
3147 /* Make sure all netdevs sharing this block are offload-capable. */
3148 if (block->nooffloaddevcnt && err_stop)
3151 list_for_each_entry(block_cb, &block->cb_list, list) {
3152 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3162 EXPORT_SYMBOL(tc_setup_cb_call);
3164 int tc_setup_flow_action(struct flow_action *flow_action,
3165 const struct tcf_exts *exts)
3167 const struct tc_action *act;
3174 tcf_exts_for_each_action(i, act, exts) {
3175 struct flow_action_entry *entry;
3177 entry = &flow_action->entries[j];
3178 if (is_tcf_gact_ok(act)) {
3179 entry->id = FLOW_ACTION_ACCEPT;
3180 } else if (is_tcf_gact_shot(act)) {
3181 entry->id = FLOW_ACTION_DROP;
3182 } else if (is_tcf_gact_trap(act)) {
3183 entry->id = FLOW_ACTION_TRAP;
3184 } else if (is_tcf_gact_goto_chain(act)) {
3185 entry->id = FLOW_ACTION_GOTO;
3186 entry->chain_index = tcf_gact_goto_chain_index(act);
3187 } else if (is_tcf_mirred_egress_redirect(act)) {
3188 entry->id = FLOW_ACTION_REDIRECT;
3189 entry->dev = tcf_mirred_dev(act);
3190 } else if (is_tcf_mirred_egress_mirror(act)) {
3191 entry->id = FLOW_ACTION_MIRRED;
3192 entry->dev = tcf_mirred_dev(act);
3193 } else if (is_tcf_vlan(act)) {
3194 switch (tcf_vlan_action(act)) {
3195 case TCA_VLAN_ACT_PUSH:
3196 entry->id = FLOW_ACTION_VLAN_PUSH;
3197 entry->vlan.vid = tcf_vlan_push_vid(act);
3198 entry->vlan.proto = tcf_vlan_push_proto(act);
3199 entry->vlan.prio = tcf_vlan_push_prio(act);
3201 case TCA_VLAN_ACT_POP:
3202 entry->id = FLOW_ACTION_VLAN_POP;
3204 case TCA_VLAN_ACT_MODIFY:
3205 entry->id = FLOW_ACTION_VLAN_MANGLE;
3206 entry->vlan.vid = tcf_vlan_push_vid(act);
3207 entry->vlan.proto = tcf_vlan_push_proto(act);
3208 entry->vlan.prio = tcf_vlan_push_prio(act);
3213 } else if (is_tcf_tunnel_set(act)) {
3214 entry->id = FLOW_ACTION_TUNNEL_ENCAP;
3215 entry->tunnel = tcf_tunnel_info(act);
3216 } else if (is_tcf_tunnel_release(act)) {
3217 entry->id = FLOW_ACTION_TUNNEL_DECAP;
3218 entry->tunnel = tcf_tunnel_info(act);
3219 } else if (is_tcf_pedit(act)) {
3220 for (k = 0; k < tcf_pedit_nkeys(act); k++) {
3221 switch (tcf_pedit_cmd(act, k)) {
3222 case TCA_PEDIT_KEY_EX_CMD_SET:
3223 entry->id = FLOW_ACTION_MANGLE;
3225 case TCA_PEDIT_KEY_EX_CMD_ADD:
3226 entry->id = FLOW_ACTION_ADD;
3231 entry->mangle.htype = tcf_pedit_htype(act, k);
3232 entry->mangle.mask = tcf_pedit_mask(act, k);
3233 entry->mangle.val = tcf_pedit_val(act, k);
3234 entry->mangle.offset = tcf_pedit_offset(act, k);
3235 entry = &flow_action->entries[++j];
3237 } else if (is_tcf_csum(act)) {
3238 entry->id = FLOW_ACTION_CSUM;
3239 entry->csum_flags = tcf_csum_update_flags(act);
3240 } else if (is_tcf_skbedit_mark(act)) {
3241 entry->id = FLOW_ACTION_MARK;
3242 entry->mark = tcf_skbedit_mark(act);
3247 if (!is_tcf_pedit(act))
3254 EXPORT_SYMBOL(tc_setup_flow_action);
3256 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3258 unsigned int num_acts = 0;
3259 struct tc_action *act;
3262 tcf_exts_for_each_action(i, act, exts) {
3263 if (is_tcf_pedit(act))
3264 num_acts += tcf_pedit_nkeys(act);
3270 EXPORT_SYMBOL(tcf_exts_num_actions);
3272 static __net_init int tcf_net_init(struct net *net)
3274 struct tcf_net *tn = net_generic(net, tcf_net_id);
3276 spin_lock_init(&tn->idr_lock);
3281 static void __net_exit tcf_net_exit(struct net *net)
3283 struct tcf_net *tn = net_generic(net, tcf_net_id);
3285 idr_destroy(&tn->idr);
3288 static struct pernet_operations tcf_net_ops = {
3289 .init = tcf_net_init,
3290 .exit = tcf_net_exit,
3292 .size = sizeof(struct tcf_net),
3295 static int __init tc_filter_init(void)
3299 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3303 err = register_pernet_subsys(&tcf_net_ops);
3305 goto err_register_pernet_subsys;
3307 err = rhashtable_init(&indr_setup_block_ht,
3308 &tc_indr_setup_block_ht_params);
3310 goto err_rhash_setup_block_ht;
3312 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3313 RTNL_FLAG_DOIT_UNLOCKED);
3314 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3315 RTNL_FLAG_DOIT_UNLOCKED);
3316 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3317 tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3318 rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3319 rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3320 rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3325 err_rhash_setup_block_ht:
3326 unregister_pernet_subsys(&tcf_net_ops);
3327 err_register_pernet_subsys:
3328 destroy_workqueue(tc_filter_wq);
3332 subsys_initcall(tc_filter_init);