1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * net/sched/cls_api.c Packet classifier API.
5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
9 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
12 #include <linux/module.h>
13 #include <linux/types.h>
14 #include <linux/kernel.h>
15 #include <linux/string.h>
16 #include <linux/errno.h>
17 #include <linux/err.h>
18 #include <linux/skbuff.h>
19 #include <linux/init.h>
20 #include <linux/kmod.h>
21 #include <linux/slab.h>
22 #include <linux/idr.h>
23 #include <linux/rhashtable.h>
24 #include <net/net_namespace.h>
26 #include <net/netlink.h>
27 #include <net/pkt_sched.h>
28 #include <net/pkt_cls.h>
29 #include <net/tc_act/tc_pedit.h>
30 #include <net/tc_act/tc_mirred.h>
31 #include <net/tc_act/tc_vlan.h>
32 #include <net/tc_act/tc_tunnel_key.h>
33 #include <net/tc_act/tc_csum.h>
34 #include <net/tc_act/tc_gact.h>
35 #include <net/tc_act/tc_police.h>
36 #include <net/tc_act/tc_sample.h>
37 #include <net/tc_act/tc_skbedit.h>
38 #include <net/tc_act/tc_ct.h>
40 extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
42 /* The list of all installed classifier types */
43 static LIST_HEAD(tcf_proto_base);
45 /* Protects list of registered TC modules. It is pure SMP lock. */
46 static DEFINE_RWLOCK(cls_mod_lock);
48 /* Find classifier type by string name */
50 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
52 const struct tcf_proto_ops *t, *res = NULL;
55 read_lock(&cls_mod_lock);
56 list_for_each_entry(t, &tcf_proto_base, head) {
57 if (strcmp(kind, t->kind) == 0) {
58 if (try_module_get(t->owner))
63 read_unlock(&cls_mod_lock);
68 static const struct tcf_proto_ops *
69 tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
70 struct netlink_ext_ack *extack)
72 const struct tcf_proto_ops *ops;
74 ops = __tcf_proto_lookup_ops(kind);
80 request_module("cls_%s", kind);
83 ops = __tcf_proto_lookup_ops(kind);
84 /* We dropped the RTNL semaphore in order to perform
85 * the module load. So, even if we succeeded in loading
86 * the module we have to replay the request. We indicate
90 module_put(ops->owner);
91 return ERR_PTR(-EAGAIN);
94 NL_SET_ERR_MSG(extack, "TC classifier not found");
95 return ERR_PTR(-ENOENT);
98 /* Register(unregister) new classifier type */
100 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
102 struct tcf_proto_ops *t;
105 write_lock(&cls_mod_lock);
106 list_for_each_entry(t, &tcf_proto_base, head)
107 if (!strcmp(ops->kind, t->kind))
110 list_add_tail(&ops->head, &tcf_proto_base);
113 write_unlock(&cls_mod_lock);
116 EXPORT_SYMBOL(register_tcf_proto_ops);
118 static struct workqueue_struct *tc_filter_wq;
120 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
122 struct tcf_proto_ops *t;
125 /* Wait for outstanding call_rcu()s, if any, from a
126 * tcf_proto_ops's destroy() handler.
129 flush_workqueue(tc_filter_wq);
131 write_lock(&cls_mod_lock);
132 list_for_each_entry(t, &tcf_proto_base, head) {
139 write_unlock(&cls_mod_lock);
142 EXPORT_SYMBOL(unregister_tcf_proto_ops);
144 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
146 INIT_RCU_WORK(rwork, func);
147 return queue_rcu_work(tc_filter_wq, rwork);
149 EXPORT_SYMBOL(tcf_queue_work);
151 /* Select new prio value from the range, managed by kernel. */
153 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
155 u32 first = TC_H_MAKE(0xC0000000U, 0U);
158 first = tp->prio - 1;
160 return TC_H_MAJ(first);
163 static bool tcf_proto_is_unlocked(const char *kind)
165 const struct tcf_proto_ops *ops;
168 ops = tcf_proto_lookup_ops(kind, false, NULL);
169 /* On error return false to take rtnl lock. Proto lookup/create
170 * functions will perform lookup again and properly handle errors.
175 ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
176 module_put(ops->owner);
180 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
181 u32 prio, struct tcf_chain *chain,
183 struct netlink_ext_ack *extack)
185 struct tcf_proto *tp;
188 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
190 return ERR_PTR(-ENOBUFS);
192 tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
193 if (IS_ERR(tp->ops)) {
194 err = PTR_ERR(tp->ops);
197 tp->classify = tp->ops->classify;
198 tp->protocol = protocol;
201 spin_lock_init(&tp->lock);
202 refcount_set(&tp->refcnt, 1);
204 err = tp->ops->init(tp);
206 module_put(tp->ops->owner);
216 static void tcf_proto_get(struct tcf_proto *tp)
218 refcount_inc(&tp->refcnt);
221 static void tcf_chain_put(struct tcf_chain *chain);
223 static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
224 struct netlink_ext_ack *extack)
226 tp->ops->destroy(tp, rtnl_held, extack);
227 tcf_chain_put(tp->chain);
228 module_put(tp->ops->owner);
232 static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
233 struct netlink_ext_ack *extack)
235 if (refcount_dec_and_test(&tp->refcnt))
236 tcf_proto_destroy(tp, rtnl_held, extack);
239 static int walker_check_empty(struct tcf_proto *tp, void *fh,
240 struct tcf_walker *arg)
243 arg->nonempty = true;
249 static bool tcf_proto_is_empty(struct tcf_proto *tp, bool rtnl_held)
251 struct tcf_walker walker = { .fn = walker_check_empty, };
254 tp->ops->walk(tp, &walker, rtnl_held);
255 return !walker.nonempty;
260 static bool tcf_proto_check_delete(struct tcf_proto *tp, bool rtnl_held)
262 spin_lock(&tp->lock);
263 if (tcf_proto_is_empty(tp, rtnl_held))
265 spin_unlock(&tp->lock);
269 static void tcf_proto_mark_delete(struct tcf_proto *tp)
271 spin_lock(&tp->lock);
273 spin_unlock(&tp->lock);
276 static bool tcf_proto_is_deleting(struct tcf_proto *tp)
280 spin_lock(&tp->lock);
281 deleting = tp->deleting;
282 spin_unlock(&tp->lock);
287 #define ASSERT_BLOCK_LOCKED(block) \
288 lockdep_assert_held(&(block)->lock)
290 struct tcf_filter_chain_list_item {
291 struct list_head list;
292 tcf_chain_head_change_t *chain_head_change;
293 void *chain_head_change_priv;
296 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
299 struct tcf_chain *chain;
301 ASSERT_BLOCK_LOCKED(block);
303 chain = kzalloc(sizeof(*chain), GFP_KERNEL);
306 list_add_tail(&chain->list, &block->chain_list);
307 mutex_init(&chain->filter_chain_lock);
308 chain->block = block;
309 chain->index = chain_index;
312 block->chain0.chain = chain;
316 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
317 struct tcf_proto *tp_head)
319 if (item->chain_head_change)
320 item->chain_head_change(tp_head, item->chain_head_change_priv);
323 static void tcf_chain0_head_change(struct tcf_chain *chain,
324 struct tcf_proto *tp_head)
326 struct tcf_filter_chain_list_item *item;
327 struct tcf_block *block = chain->block;
332 mutex_lock(&block->lock);
333 list_for_each_entry(item, &block->chain0.filter_chain_list, list)
334 tcf_chain_head_change_item(item, tp_head);
335 mutex_unlock(&block->lock);
338 /* Returns true if block can be safely freed. */
340 static bool tcf_chain_detach(struct tcf_chain *chain)
342 struct tcf_block *block = chain->block;
344 ASSERT_BLOCK_LOCKED(block);
346 list_del(&chain->list);
348 block->chain0.chain = NULL;
350 if (list_empty(&block->chain_list) &&
351 refcount_read(&block->refcnt) == 0)
357 static void tcf_block_destroy(struct tcf_block *block)
359 mutex_destroy(&block->lock);
360 kfree_rcu(block, rcu);
363 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
365 struct tcf_block *block = chain->block;
367 mutex_destroy(&chain->filter_chain_lock);
368 kfree_rcu(chain, rcu);
370 tcf_block_destroy(block);
373 static void tcf_chain_hold(struct tcf_chain *chain)
375 ASSERT_BLOCK_LOCKED(chain->block);
380 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
382 ASSERT_BLOCK_LOCKED(chain->block);
384 /* In case all the references are action references, this
385 * chain should not be shown to the user.
387 return chain->refcnt == chain->action_refcnt;
390 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
393 struct tcf_chain *chain;
395 ASSERT_BLOCK_LOCKED(block);
397 list_for_each_entry(chain, &block->chain_list, list) {
398 if (chain->index == chain_index)
404 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
405 u32 seq, u16 flags, int event, bool unicast);
407 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
408 u32 chain_index, bool create,
411 struct tcf_chain *chain = NULL;
412 bool is_first_reference;
414 mutex_lock(&block->lock);
415 chain = tcf_chain_lookup(block, chain_index);
417 tcf_chain_hold(chain);
421 chain = tcf_chain_create(block, chain_index);
427 ++chain->action_refcnt;
428 is_first_reference = chain->refcnt - chain->action_refcnt == 1;
429 mutex_unlock(&block->lock);
431 /* Send notification only in case we got the first
432 * non-action reference. Until then, the chain acts only as
433 * a placeholder for actions pointing to it and user ought
434 * not know about them.
436 if (is_first_reference && !by_act)
437 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
438 RTM_NEWCHAIN, false);
443 mutex_unlock(&block->lock);
447 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
450 return __tcf_chain_get(block, chain_index, create, false);
453 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
455 return __tcf_chain_get(block, chain_index, true, true);
457 EXPORT_SYMBOL(tcf_chain_get_by_act);
459 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
461 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
462 void *tmplt_priv, u32 chain_index,
463 struct tcf_block *block, struct sk_buff *oskb,
464 u32 seq, u16 flags, bool unicast);
466 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
467 bool explicitly_created)
469 struct tcf_block *block = chain->block;
470 const struct tcf_proto_ops *tmplt_ops;
471 bool free_block = false;
475 mutex_lock(&block->lock);
476 if (explicitly_created) {
477 if (!chain->explicitly_created) {
478 mutex_unlock(&block->lock);
481 chain->explicitly_created = false;
485 chain->action_refcnt--;
487 /* tc_chain_notify_delete can't be called while holding block lock.
488 * However, when block is unlocked chain can be changed concurrently, so
489 * save these to temporary variables.
491 refcnt = --chain->refcnt;
492 tmplt_ops = chain->tmplt_ops;
493 tmplt_priv = chain->tmplt_priv;
495 /* The last dropped non-action reference will trigger notification. */
496 if (refcnt - chain->action_refcnt == 0 && !by_act) {
497 tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
498 block, NULL, 0, 0, false);
499 /* Last reference to chain, no need to lock. */
500 chain->flushing = false;
504 free_block = tcf_chain_detach(chain);
505 mutex_unlock(&block->lock);
508 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
509 tcf_chain_destroy(chain, free_block);
513 static void tcf_chain_put(struct tcf_chain *chain)
515 __tcf_chain_put(chain, false, false);
518 void tcf_chain_put_by_act(struct tcf_chain *chain)
520 __tcf_chain_put(chain, true, false);
522 EXPORT_SYMBOL(tcf_chain_put_by_act);
524 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
526 __tcf_chain_put(chain, false, true);
529 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
531 struct tcf_proto *tp, *tp_next;
533 mutex_lock(&chain->filter_chain_lock);
534 tp = tcf_chain_dereference(chain->filter_chain, chain);
535 RCU_INIT_POINTER(chain->filter_chain, NULL);
536 tcf_chain0_head_change(chain, NULL);
537 chain->flushing = true;
538 mutex_unlock(&chain->filter_chain_lock);
541 tp_next = rcu_dereference_protected(tp->next, 1);
542 tcf_proto_put(tp, rtnl_held, NULL);
547 static struct tcf_block *tc_dev_ingress_block(struct net_device *dev)
549 const struct Qdisc_class_ops *cops;
552 if (!dev_ingress_queue(dev))
555 qdisc = dev_ingress_queue(dev)->qdisc_sleeping;
559 cops = qdisc->ops->cl_ops;
563 if (!cops->tcf_block)
566 return cops->tcf_block(qdisc, TC_H_MIN_INGRESS, NULL);
569 static struct rhashtable indr_setup_block_ht;
571 struct tc_indr_block_dev {
572 struct rhash_head ht_node;
573 struct net_device *dev;
575 struct list_head cb_list;
576 struct tcf_block *block;
579 struct tc_indr_block_cb {
580 struct list_head list;
582 tc_indr_block_bind_cb_t *cb;
586 static const struct rhashtable_params tc_indr_setup_block_ht_params = {
587 .key_offset = offsetof(struct tc_indr_block_dev, dev),
588 .head_offset = offsetof(struct tc_indr_block_dev, ht_node),
589 .key_len = sizeof(struct net_device *),
592 static struct tc_indr_block_dev *
593 tc_indr_block_dev_lookup(struct net_device *dev)
595 return rhashtable_lookup_fast(&indr_setup_block_ht, &dev,
596 tc_indr_setup_block_ht_params);
599 static struct tc_indr_block_dev *tc_indr_block_dev_get(struct net_device *dev)
601 struct tc_indr_block_dev *indr_dev;
603 indr_dev = tc_indr_block_dev_lookup(dev);
607 indr_dev = kzalloc(sizeof(*indr_dev), GFP_KERNEL);
611 INIT_LIST_HEAD(&indr_dev->cb_list);
613 indr_dev->block = tc_dev_ingress_block(dev);
614 if (rhashtable_insert_fast(&indr_setup_block_ht, &indr_dev->ht_node,
615 tc_indr_setup_block_ht_params)) {
625 static void tc_indr_block_dev_put(struct tc_indr_block_dev *indr_dev)
627 if (--indr_dev->refcnt)
630 rhashtable_remove_fast(&indr_setup_block_ht, &indr_dev->ht_node,
631 tc_indr_setup_block_ht_params);
635 static struct tc_indr_block_cb *
636 tc_indr_block_cb_lookup(struct tc_indr_block_dev *indr_dev,
637 tc_indr_block_bind_cb_t *cb, void *cb_ident)
639 struct tc_indr_block_cb *indr_block_cb;
641 list_for_each_entry(indr_block_cb, &indr_dev->cb_list, list)
642 if (indr_block_cb->cb == cb &&
643 indr_block_cb->cb_ident == cb_ident)
644 return indr_block_cb;
648 static struct tc_indr_block_cb *
649 tc_indr_block_cb_add(struct tc_indr_block_dev *indr_dev, void *cb_priv,
650 tc_indr_block_bind_cb_t *cb, void *cb_ident)
652 struct tc_indr_block_cb *indr_block_cb;
654 indr_block_cb = tc_indr_block_cb_lookup(indr_dev, cb, cb_ident);
656 return ERR_PTR(-EEXIST);
658 indr_block_cb = kzalloc(sizeof(*indr_block_cb), GFP_KERNEL);
660 return ERR_PTR(-ENOMEM);
662 indr_block_cb->cb_priv = cb_priv;
663 indr_block_cb->cb = cb;
664 indr_block_cb->cb_ident = cb_ident;
665 list_add(&indr_block_cb->list, &indr_dev->cb_list);
667 return indr_block_cb;
670 static void tc_indr_block_cb_del(struct tc_indr_block_cb *indr_block_cb)
672 list_del(&indr_block_cb->list);
673 kfree(indr_block_cb);
676 static int tcf_block_setup(struct tcf_block *block,
677 struct flow_block_offload *bo);
679 static void tc_indr_block_ing_cmd(struct tc_indr_block_dev *indr_dev,
680 struct tc_indr_block_cb *indr_block_cb,
681 enum flow_block_command command)
683 struct flow_block_offload bo = {
685 .binder_type = FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS,
686 .net = dev_net(indr_dev->dev),
687 .block_shared = tcf_block_non_null_shared(indr_dev->block),
689 INIT_LIST_HEAD(&bo.cb_list);
691 if (!indr_dev->block)
694 bo.block = &indr_dev->block->flow_block;
696 indr_block_cb->cb(indr_dev->dev, indr_block_cb->cb_priv, TC_SETUP_BLOCK,
698 tcf_block_setup(indr_dev->block, &bo);
701 int __tc_indr_block_cb_register(struct net_device *dev, void *cb_priv,
702 tc_indr_block_bind_cb_t *cb, void *cb_ident)
704 struct tc_indr_block_cb *indr_block_cb;
705 struct tc_indr_block_dev *indr_dev;
708 indr_dev = tc_indr_block_dev_get(dev);
712 indr_block_cb = tc_indr_block_cb_add(indr_dev, cb_priv, cb, cb_ident);
713 err = PTR_ERR_OR_ZERO(indr_block_cb);
717 tc_indr_block_ing_cmd(indr_dev, indr_block_cb, FLOW_BLOCK_BIND);
721 tc_indr_block_dev_put(indr_dev);
724 EXPORT_SYMBOL_GPL(__tc_indr_block_cb_register);
726 int tc_indr_block_cb_register(struct net_device *dev, void *cb_priv,
727 tc_indr_block_bind_cb_t *cb, void *cb_ident)
732 err = __tc_indr_block_cb_register(dev, cb_priv, cb, cb_ident);
737 EXPORT_SYMBOL_GPL(tc_indr_block_cb_register);
739 void __tc_indr_block_cb_unregister(struct net_device *dev,
740 tc_indr_block_bind_cb_t *cb, void *cb_ident)
742 struct tc_indr_block_cb *indr_block_cb;
743 struct tc_indr_block_dev *indr_dev;
745 indr_dev = tc_indr_block_dev_lookup(dev);
749 indr_block_cb = tc_indr_block_cb_lookup(indr_dev, cb, cb_ident);
753 /* Send unbind message if required to free any block cbs. */
754 tc_indr_block_ing_cmd(indr_dev, indr_block_cb, FLOW_BLOCK_UNBIND);
755 tc_indr_block_cb_del(indr_block_cb);
756 tc_indr_block_dev_put(indr_dev);
758 EXPORT_SYMBOL_GPL(__tc_indr_block_cb_unregister);
760 void tc_indr_block_cb_unregister(struct net_device *dev,
761 tc_indr_block_bind_cb_t *cb, void *cb_ident)
764 __tc_indr_block_cb_unregister(dev, cb, cb_ident);
767 EXPORT_SYMBOL_GPL(tc_indr_block_cb_unregister);
769 static void tc_indr_block_call(struct tcf_block *block, struct net_device *dev,
770 struct tcf_block_ext_info *ei,
771 enum flow_block_command command,
772 struct netlink_ext_ack *extack)
774 struct tc_indr_block_cb *indr_block_cb;
775 struct tc_indr_block_dev *indr_dev;
776 struct flow_block_offload bo = {
778 .binder_type = ei->binder_type,
780 .block = &block->flow_block,
781 .block_shared = tcf_block_shared(block),
784 INIT_LIST_HEAD(&bo.cb_list);
786 indr_dev = tc_indr_block_dev_lookup(dev);
790 indr_dev->block = command == FLOW_BLOCK_BIND ? block : NULL;
792 list_for_each_entry(indr_block_cb, &indr_dev->cb_list, list)
793 indr_block_cb->cb(dev, indr_block_cb->cb_priv, TC_SETUP_BLOCK,
796 tcf_block_setup(block, &bo);
799 static bool tcf_block_offload_in_use(struct tcf_block *block)
801 return block->offloadcnt;
804 static int tcf_block_offload_cmd(struct tcf_block *block,
805 struct net_device *dev,
806 struct tcf_block_ext_info *ei,
807 enum flow_block_command command,
808 struct netlink_ext_ack *extack)
810 struct flow_block_offload bo = {};
813 bo.net = dev_net(dev);
814 bo.command = command;
815 bo.binder_type = ei->binder_type;
816 bo.block = &block->flow_block;
817 bo.block_shared = tcf_block_shared(block);
819 INIT_LIST_HEAD(&bo.cb_list);
821 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
825 return tcf_block_setup(block, &bo);
828 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
829 struct tcf_block_ext_info *ei,
830 struct netlink_ext_ack *extack)
832 struct net_device *dev = q->dev_queue->dev;
835 if (!dev->netdev_ops->ndo_setup_tc)
836 goto no_offload_dev_inc;
838 /* If tc offload feature is disabled and the block we try to bind
839 * to already has some offloaded filters, forbid to bind.
841 if (!tc_can_offload(dev) && tcf_block_offload_in_use(block)) {
842 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
846 err = tcf_block_offload_cmd(block, dev, ei, FLOW_BLOCK_BIND, extack);
847 if (err == -EOPNOTSUPP)
848 goto no_offload_dev_inc;
852 tc_indr_block_call(block, dev, ei, FLOW_BLOCK_BIND, extack);
856 if (tcf_block_offload_in_use(block))
858 block->nooffloaddevcnt++;
859 tc_indr_block_call(block, dev, ei, FLOW_BLOCK_BIND, extack);
863 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
864 struct tcf_block_ext_info *ei)
866 struct net_device *dev = q->dev_queue->dev;
869 tc_indr_block_call(block, dev, ei, FLOW_BLOCK_UNBIND, NULL);
871 if (!dev->netdev_ops->ndo_setup_tc)
872 goto no_offload_dev_dec;
873 err = tcf_block_offload_cmd(block, dev, ei, FLOW_BLOCK_UNBIND, NULL);
874 if (err == -EOPNOTSUPP)
875 goto no_offload_dev_dec;
879 WARN_ON(block->nooffloaddevcnt-- == 0);
883 tcf_chain0_head_change_cb_add(struct tcf_block *block,
884 struct tcf_block_ext_info *ei,
885 struct netlink_ext_ack *extack)
887 struct tcf_filter_chain_list_item *item;
888 struct tcf_chain *chain0;
890 item = kmalloc(sizeof(*item), GFP_KERNEL);
892 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
895 item->chain_head_change = ei->chain_head_change;
896 item->chain_head_change_priv = ei->chain_head_change_priv;
898 mutex_lock(&block->lock);
899 chain0 = block->chain0.chain;
901 tcf_chain_hold(chain0);
903 list_add(&item->list, &block->chain0.filter_chain_list);
904 mutex_unlock(&block->lock);
907 struct tcf_proto *tp_head;
909 mutex_lock(&chain0->filter_chain_lock);
911 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
913 tcf_chain_head_change_item(item, tp_head);
915 mutex_lock(&block->lock);
916 list_add(&item->list, &block->chain0.filter_chain_list);
917 mutex_unlock(&block->lock);
919 mutex_unlock(&chain0->filter_chain_lock);
920 tcf_chain_put(chain0);
927 tcf_chain0_head_change_cb_del(struct tcf_block *block,
928 struct tcf_block_ext_info *ei)
930 struct tcf_filter_chain_list_item *item;
932 mutex_lock(&block->lock);
933 list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
934 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
935 (item->chain_head_change == ei->chain_head_change &&
936 item->chain_head_change_priv == ei->chain_head_change_priv)) {
937 if (block->chain0.chain)
938 tcf_chain_head_change_item(item, NULL);
939 list_del(&item->list);
940 mutex_unlock(&block->lock);
946 mutex_unlock(&block->lock);
951 spinlock_t idr_lock; /* Protects idr */
955 static unsigned int tcf_net_id;
957 static int tcf_block_insert(struct tcf_block *block, struct net *net,
958 struct netlink_ext_ack *extack)
960 struct tcf_net *tn = net_generic(net, tcf_net_id);
963 idr_preload(GFP_KERNEL);
964 spin_lock(&tn->idr_lock);
965 err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
967 spin_unlock(&tn->idr_lock);
973 static void tcf_block_remove(struct tcf_block *block, struct net *net)
975 struct tcf_net *tn = net_generic(net, tcf_net_id);
977 spin_lock(&tn->idr_lock);
978 idr_remove(&tn->idr, block->index);
979 spin_unlock(&tn->idr_lock);
982 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
984 struct netlink_ext_ack *extack)
986 struct tcf_block *block;
988 block = kzalloc(sizeof(*block), GFP_KERNEL);
990 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
991 return ERR_PTR(-ENOMEM);
993 mutex_init(&block->lock);
994 flow_block_init(&block->flow_block);
995 INIT_LIST_HEAD(&block->chain_list);
996 INIT_LIST_HEAD(&block->owner_list);
997 INIT_LIST_HEAD(&block->chain0.filter_chain_list);
999 refcount_set(&block->refcnt, 1);
1001 block->index = block_index;
1003 /* Don't store q pointer for blocks which are shared */
1004 if (!tcf_block_shared(block))
1009 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
1011 struct tcf_net *tn = net_generic(net, tcf_net_id);
1013 return idr_find(&tn->idr, block_index);
1016 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
1018 struct tcf_block *block;
1021 block = tcf_block_lookup(net, block_index);
1022 if (block && !refcount_inc_not_zero(&block->refcnt))
1029 static struct tcf_chain *
1030 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1032 mutex_lock(&block->lock);
1034 chain = list_is_last(&chain->list, &block->chain_list) ?
1035 NULL : list_next_entry(chain, list);
1037 chain = list_first_entry_or_null(&block->chain_list,
1038 struct tcf_chain, list);
1040 /* skip all action-only chains */
1041 while (chain && tcf_chain_held_by_acts_only(chain))
1042 chain = list_is_last(&chain->list, &block->chain_list) ?
1043 NULL : list_next_entry(chain, list);
1046 tcf_chain_hold(chain);
1047 mutex_unlock(&block->lock);
1052 /* Function to be used by all clients that want to iterate over all chains on
1053 * block. It properly obtains block->lock and takes reference to chain before
1054 * returning it. Users of this function must be tolerant to concurrent chain
1055 * insertion/deletion or ensure that no concurrent chain modification is
1056 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1057 * consistent dump because rtnl lock is released each time skb is filled with
1058 * data and sent to user-space.
1062 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1064 struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
1067 tcf_chain_put(chain);
1071 EXPORT_SYMBOL(tcf_get_next_chain);
1073 static struct tcf_proto *
1074 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1079 mutex_lock(&chain->filter_chain_lock);
1082 tp = tcf_chain_dereference(chain->filter_chain, chain);
1083 } else if (tcf_proto_is_deleting(tp)) {
1084 /* 'deleting' flag is set and chain->filter_chain_lock was
1085 * unlocked, which means next pointer could be invalid. Restart
1088 prio = tp->prio + 1;
1089 tp = tcf_chain_dereference(chain->filter_chain, chain);
1091 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1092 if (!tp->deleting && tp->prio >= prio)
1095 tp = tcf_chain_dereference(tp->next, chain);
1101 mutex_unlock(&chain->filter_chain_lock);
1106 /* Function to be used by all clients that want to iterate over all tp's on
1107 * chain. Users of this function must be tolerant to concurrent tp
1108 * insertion/deletion or ensure that no concurrent chain modification is
1109 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1110 * consistent dump because rtnl lock is released each time skb is filled with
1111 * data and sent to user-space.
1115 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp,
1118 struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1121 tcf_proto_put(tp, rtnl_held, NULL);
1125 EXPORT_SYMBOL(tcf_get_next_proto);
1127 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1129 struct tcf_chain *chain;
1131 /* Last reference to block. At this point chains cannot be added or
1132 * removed concurrently.
1134 for (chain = tcf_get_next_chain(block, NULL);
1136 chain = tcf_get_next_chain(block, chain)) {
1137 tcf_chain_put_explicitly_created(chain);
1138 tcf_chain_flush(chain, rtnl_held);
1142 /* Lookup Qdisc and increments its reference counter.
1143 * Set parent, if necessary.
1146 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1147 u32 *parent, int ifindex, bool rtnl_held,
1148 struct netlink_ext_ack *extack)
1150 const struct Qdisc_class_ops *cops;
1151 struct net_device *dev;
1154 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1160 dev = dev_get_by_index_rcu(net, ifindex);
1169 *parent = (*q)->handle;
1171 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1173 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1179 *q = qdisc_refcount_inc_nz(*q);
1181 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1186 /* Is it classful? */
1187 cops = (*q)->ops->cl_ops;
1189 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1194 if (!cops->tcf_block) {
1195 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1201 /* At this point we know that qdisc is not noop_qdisc,
1202 * which means that qdisc holds a reference to net_device
1203 * and we hold a reference to qdisc, so it is safe to release
1215 qdisc_put_unlocked(*q);
1221 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1222 int ifindex, struct netlink_ext_ack *extack)
1224 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1227 /* Do we search for filter, attached to class? */
1228 if (TC_H_MIN(parent)) {
1229 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1231 *cl = cops->find(q, parent);
1233 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1241 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1242 unsigned long cl, int ifindex,
1244 struct netlink_ext_ack *extack)
1246 struct tcf_block *block;
1248 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1249 block = tcf_block_refcnt_get(net, block_index);
1251 NL_SET_ERR_MSG(extack, "Block of given index was not found");
1252 return ERR_PTR(-EINVAL);
1255 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1257 block = cops->tcf_block(q, cl, extack);
1259 return ERR_PTR(-EINVAL);
1261 if (tcf_block_shared(block)) {
1262 NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1263 return ERR_PTR(-EOPNOTSUPP);
1266 /* Always take reference to block in order to support execution
1267 * of rules update path of cls API without rtnl lock. Caller
1268 * must release block when it is finished using it. 'if' block
1269 * of this conditional obtain reference to block by calling
1270 * tcf_block_refcnt_get().
1272 refcount_inc(&block->refcnt);
1278 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1279 struct tcf_block_ext_info *ei, bool rtnl_held)
1281 if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1282 /* Flushing/putting all chains will cause the block to be
1283 * deallocated when last chain is freed. However, if chain_list
1284 * is empty, block has to be manually deallocated. After block
1285 * reference counter reached 0, it is no longer possible to
1286 * increment it or add new chains to block.
1288 bool free_block = list_empty(&block->chain_list);
1290 mutex_unlock(&block->lock);
1291 if (tcf_block_shared(block))
1292 tcf_block_remove(block, block->net);
1295 tcf_block_offload_unbind(block, q, ei);
1298 tcf_block_destroy(block);
1300 tcf_block_flush_all_chains(block, rtnl_held);
1302 tcf_block_offload_unbind(block, q, ei);
1306 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1308 __tcf_block_put(block, NULL, NULL, rtnl_held);
1312 * Set q, parent, cl when appropriate.
1315 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1316 u32 *parent, unsigned long *cl,
1317 int ifindex, u32 block_index,
1318 struct netlink_ext_ack *extack)
1320 struct tcf_block *block;
1325 err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1329 err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1333 block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1334 if (IS_ERR(block)) {
1335 err = PTR_ERR(block);
1346 return ERR_PTR(err);
1349 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1352 if (!IS_ERR_OR_NULL(block))
1353 tcf_block_refcnt_put(block, rtnl_held);
1359 qdisc_put_unlocked(q);
1363 struct tcf_block_owner_item {
1364 struct list_head list;
1366 enum flow_block_binder_type binder_type;
1370 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1372 enum flow_block_binder_type binder_type)
1374 if (block->keep_dst &&
1375 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1376 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1377 netif_keep_dst(qdisc_dev(q));
1380 void tcf_block_netif_keep_dst(struct tcf_block *block)
1382 struct tcf_block_owner_item *item;
1384 block->keep_dst = true;
1385 list_for_each_entry(item, &block->owner_list, list)
1386 tcf_block_owner_netif_keep_dst(block, item->q,
1389 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1391 static int tcf_block_owner_add(struct tcf_block *block,
1393 enum flow_block_binder_type binder_type)
1395 struct tcf_block_owner_item *item;
1397 item = kmalloc(sizeof(*item), GFP_KERNEL);
1401 item->binder_type = binder_type;
1402 list_add(&item->list, &block->owner_list);
1406 static void tcf_block_owner_del(struct tcf_block *block,
1408 enum flow_block_binder_type binder_type)
1410 struct tcf_block_owner_item *item;
1412 list_for_each_entry(item, &block->owner_list, list) {
1413 if (item->q == q && item->binder_type == binder_type) {
1414 list_del(&item->list);
1422 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1423 struct tcf_block_ext_info *ei,
1424 struct netlink_ext_ack *extack)
1426 struct net *net = qdisc_net(q);
1427 struct tcf_block *block = NULL;
1430 if (ei->block_index)
1431 /* block_index not 0 means the shared block is requested */
1432 block = tcf_block_refcnt_get(net, ei->block_index);
1435 block = tcf_block_create(net, q, ei->block_index, extack);
1437 return PTR_ERR(block);
1438 if (tcf_block_shared(block)) {
1439 err = tcf_block_insert(block, net, extack);
1441 goto err_block_insert;
1445 err = tcf_block_owner_add(block, q, ei->binder_type);
1447 goto err_block_owner_add;
1449 tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1451 err = tcf_chain0_head_change_cb_add(block, ei, extack);
1453 goto err_chain0_head_change_cb_add;
1455 err = tcf_block_offload_bind(block, q, ei, extack);
1457 goto err_block_offload_bind;
1462 err_block_offload_bind:
1463 tcf_chain0_head_change_cb_del(block, ei);
1464 err_chain0_head_change_cb_add:
1465 tcf_block_owner_del(block, q, ei->binder_type);
1466 err_block_owner_add:
1468 tcf_block_refcnt_put(block, true);
1471 EXPORT_SYMBOL(tcf_block_get_ext);
1473 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1475 struct tcf_proto __rcu **p_filter_chain = priv;
1477 rcu_assign_pointer(*p_filter_chain, tp_head);
1480 int tcf_block_get(struct tcf_block **p_block,
1481 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1482 struct netlink_ext_ack *extack)
1484 struct tcf_block_ext_info ei = {
1485 .chain_head_change = tcf_chain_head_change_dflt,
1486 .chain_head_change_priv = p_filter_chain,
1489 WARN_ON(!p_filter_chain);
1490 return tcf_block_get_ext(p_block, q, &ei, extack);
1492 EXPORT_SYMBOL(tcf_block_get);
1494 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1495 * actions should be all removed after flushing.
1497 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1498 struct tcf_block_ext_info *ei)
1502 tcf_chain0_head_change_cb_del(block, ei);
1503 tcf_block_owner_del(block, q, ei->binder_type);
1505 __tcf_block_put(block, q, ei, true);
1507 EXPORT_SYMBOL(tcf_block_put_ext);
1509 void tcf_block_put(struct tcf_block *block)
1511 struct tcf_block_ext_info ei = {0, };
1515 tcf_block_put_ext(block, block->q, &ei);
1518 EXPORT_SYMBOL(tcf_block_put);
1521 tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1522 void *cb_priv, bool add, bool offload_in_use,
1523 struct netlink_ext_ack *extack)
1525 struct tcf_chain *chain, *chain_prev;
1526 struct tcf_proto *tp, *tp_prev;
1529 for (chain = __tcf_get_next_chain(block, NULL);
1532 chain = __tcf_get_next_chain(block, chain),
1533 tcf_chain_put(chain_prev)) {
1534 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1536 tp = __tcf_get_next_proto(chain, tp),
1537 tcf_proto_put(tp_prev, true, NULL)) {
1538 if (tp->ops->reoffload) {
1539 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1542 goto err_playback_remove;
1543 } else if (add && offload_in_use) {
1545 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1546 goto err_playback_remove;
1553 err_playback_remove:
1554 tcf_proto_put(tp, true, NULL);
1555 tcf_chain_put(chain);
1556 tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1561 static int tcf_block_bind(struct tcf_block *block,
1562 struct flow_block_offload *bo)
1564 struct flow_block_cb *block_cb, *next;
1567 list_for_each_entry(block_cb, &bo->cb_list, list) {
1568 err = tcf_block_playback_offloads(block, block_cb->cb,
1569 block_cb->cb_priv, true,
1570 tcf_block_offload_in_use(block),
1577 list_splice(&bo->cb_list, &block->flow_block.cb_list);
1582 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1584 list_del(&block_cb->list);
1585 tcf_block_playback_offloads(block, block_cb->cb,
1586 block_cb->cb_priv, false,
1587 tcf_block_offload_in_use(block),
1590 flow_block_cb_free(block_cb);
1596 static void tcf_block_unbind(struct tcf_block *block,
1597 struct flow_block_offload *bo)
1599 struct flow_block_cb *block_cb, *next;
1601 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1602 tcf_block_playback_offloads(block, block_cb->cb,
1603 block_cb->cb_priv, false,
1604 tcf_block_offload_in_use(block),
1606 list_del(&block_cb->list);
1607 flow_block_cb_free(block_cb);
1611 static int tcf_block_setup(struct tcf_block *block,
1612 struct flow_block_offload *bo)
1616 switch (bo->command) {
1617 case FLOW_BLOCK_BIND:
1618 err = tcf_block_bind(block, bo);
1620 case FLOW_BLOCK_UNBIND:
1622 tcf_block_unbind(block, bo);
1632 /* Main classifier routine: scans classifier chain attached
1633 * to this qdisc, (optionally) tests for protocol and asks
1634 * specific classifiers.
1636 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1637 struct tcf_result *res, bool compat_mode)
1639 #ifdef CONFIG_NET_CLS_ACT
1640 const int max_reclassify_loop = 4;
1641 const struct tcf_proto *orig_tp = tp;
1642 const struct tcf_proto *first_tp;
1647 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1648 __be16 protocol = tc_skb_protocol(skb);
1651 if (tp->protocol != protocol &&
1652 tp->protocol != htons(ETH_P_ALL))
1655 err = tp->classify(skb, tp, res);
1656 #ifdef CONFIG_NET_CLS_ACT
1657 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1660 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1661 first_tp = res->goto_tp;
1669 return TC_ACT_UNSPEC; /* signal: continue lookup */
1670 #ifdef CONFIG_NET_CLS_ACT
1672 if (unlikely(limit++ >= max_reclassify_loop)) {
1673 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1674 tp->chain->block->index,
1676 ntohs(tp->protocol));
1684 EXPORT_SYMBOL(tcf_classify);
1686 struct tcf_chain_info {
1687 struct tcf_proto __rcu **pprev;
1688 struct tcf_proto __rcu *next;
1691 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1692 struct tcf_chain_info *chain_info)
1694 return tcf_chain_dereference(*chain_info->pprev, chain);
1697 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1698 struct tcf_chain_info *chain_info,
1699 struct tcf_proto *tp)
1701 if (chain->flushing)
1704 if (*chain_info->pprev == chain->filter_chain)
1705 tcf_chain0_head_change(chain, tp);
1707 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1708 rcu_assign_pointer(*chain_info->pprev, tp);
1713 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1714 struct tcf_chain_info *chain_info,
1715 struct tcf_proto *tp)
1717 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1719 tcf_proto_mark_delete(tp);
1720 if (tp == chain->filter_chain)
1721 tcf_chain0_head_change(chain, next);
1722 RCU_INIT_POINTER(*chain_info->pprev, next);
1725 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1726 struct tcf_chain_info *chain_info,
1727 u32 protocol, u32 prio,
1728 bool prio_allocate);
1730 /* Try to insert new proto.
1731 * If proto with specified priority already exists, free new proto
1732 * and return existing one.
1735 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1736 struct tcf_proto *tp_new,
1737 u32 protocol, u32 prio,
1740 struct tcf_chain_info chain_info;
1741 struct tcf_proto *tp;
1744 mutex_lock(&chain->filter_chain_lock);
1746 tp = tcf_chain_tp_find(chain, &chain_info,
1747 protocol, prio, false);
1749 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1750 mutex_unlock(&chain->filter_chain_lock);
1753 tcf_proto_destroy(tp_new, rtnl_held, NULL);
1756 tcf_proto_destroy(tp_new, rtnl_held, NULL);
1757 tp_new = ERR_PTR(err);
1763 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1764 struct tcf_proto *tp, bool rtnl_held,
1765 struct netlink_ext_ack *extack)
1767 struct tcf_chain_info chain_info;
1768 struct tcf_proto *tp_iter;
1769 struct tcf_proto **pprev;
1770 struct tcf_proto *next;
1772 mutex_lock(&chain->filter_chain_lock);
1774 /* Atomically find and remove tp from chain. */
1775 for (pprev = &chain->filter_chain;
1776 (tp_iter = tcf_chain_dereference(*pprev, chain));
1777 pprev = &tp_iter->next) {
1778 if (tp_iter == tp) {
1779 chain_info.pprev = pprev;
1780 chain_info.next = tp_iter->next;
1781 WARN_ON(tp_iter->deleting);
1785 /* Verify that tp still exists and no new filters were inserted
1787 * Mark tp for deletion if it is empty.
1789 if (!tp_iter || !tcf_proto_check_delete(tp, rtnl_held)) {
1790 mutex_unlock(&chain->filter_chain_lock);
1794 next = tcf_chain_dereference(chain_info.next, chain);
1795 if (tp == chain->filter_chain)
1796 tcf_chain0_head_change(chain, next);
1797 RCU_INIT_POINTER(*chain_info.pprev, next);
1798 mutex_unlock(&chain->filter_chain_lock);
1800 tcf_proto_put(tp, rtnl_held, extack);
1803 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1804 struct tcf_chain_info *chain_info,
1805 u32 protocol, u32 prio,
1808 struct tcf_proto **pprev;
1809 struct tcf_proto *tp;
1811 /* Check the chain for existence of proto-tcf with this priority */
1812 for (pprev = &chain->filter_chain;
1813 (tp = tcf_chain_dereference(*pprev, chain));
1814 pprev = &tp->next) {
1815 if (tp->prio >= prio) {
1816 if (tp->prio == prio) {
1817 if (prio_allocate ||
1818 (tp->protocol != protocol && protocol))
1819 return ERR_PTR(-EINVAL);
1826 chain_info->pprev = pprev;
1828 chain_info->next = tp->next;
1831 chain_info->next = NULL;
1836 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1837 struct tcf_proto *tp, struct tcf_block *block,
1838 struct Qdisc *q, u32 parent, void *fh,
1839 u32 portid, u32 seq, u16 flags, int event,
1843 struct nlmsghdr *nlh;
1844 unsigned char *b = skb_tail_pointer(skb);
1846 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1848 goto out_nlmsg_trim;
1849 tcm = nlmsg_data(nlh);
1850 tcm->tcm_family = AF_UNSPEC;
1854 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1855 tcm->tcm_parent = parent;
1857 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1858 tcm->tcm_block_index = block->index;
1860 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1861 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1862 goto nla_put_failure;
1863 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1864 goto nla_put_failure;
1866 tcm->tcm_handle = 0;
1868 if (tp->ops->dump &&
1869 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
1870 goto nla_put_failure;
1872 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1881 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1882 struct nlmsghdr *n, struct tcf_proto *tp,
1883 struct tcf_block *block, struct Qdisc *q,
1884 u32 parent, void *fh, int event, bool unicast,
1887 struct sk_buff *skb;
1888 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1891 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1895 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1896 n->nlmsg_seq, n->nlmsg_flags, event,
1903 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1905 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1906 n->nlmsg_flags & NLM_F_ECHO);
1913 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1914 struct nlmsghdr *n, struct tcf_proto *tp,
1915 struct tcf_block *block, struct Qdisc *q,
1916 u32 parent, void *fh, bool unicast, bool *last,
1917 bool rtnl_held, struct netlink_ext_ack *extack)
1919 struct sk_buff *skb;
1920 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1923 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1927 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1928 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
1930 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1935 err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
1942 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1944 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1945 n->nlmsg_flags & NLM_F_ECHO);
1947 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1954 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1955 struct tcf_block *block, struct Qdisc *q,
1956 u32 parent, struct nlmsghdr *n,
1957 struct tcf_chain *chain, int event,
1960 struct tcf_proto *tp;
1962 for (tp = tcf_get_next_proto(chain, NULL, rtnl_held);
1963 tp; tp = tcf_get_next_proto(chain, tp, rtnl_held))
1964 tfilter_notify(net, oskb, n, tp, block,
1965 q, parent, NULL, event, false, rtnl_held);
1968 static void tfilter_put(struct tcf_proto *tp, void *fh)
1970 if (tp->ops->put && fh)
1971 tp->ops->put(tp, fh);
1974 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1975 struct netlink_ext_ack *extack)
1977 struct net *net = sock_net(skb->sk);
1978 struct nlattr *tca[TCA_MAX + 1];
1985 struct Qdisc *q = NULL;
1986 struct tcf_chain_info chain_info;
1987 struct tcf_chain *chain = NULL;
1988 struct tcf_block *block;
1989 struct tcf_proto *tp;
1994 bool rtnl_held = false;
1996 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2002 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2003 rtm_tca_policy, extack);
2008 protocol = TC_H_MIN(t->tcm_info);
2009 prio = TC_H_MAJ(t->tcm_info);
2010 prio_allocate = false;
2011 parent = t->tcm_parent;
2017 /* If no priority is provided by the user,
2020 if (n->nlmsg_flags & NLM_F_CREATE) {
2021 prio = TC_H_MAKE(0x80000000U, 0U);
2022 prio_allocate = true;
2024 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2029 /* Find head of filter chain. */
2031 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2035 /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2036 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2037 * type is not specified, classifier is not unlocked.
2040 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2041 !tca[TCA_KIND] || !tcf_proto_is_unlocked(nla_data(tca[TCA_KIND]))) {
2046 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2050 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2052 if (IS_ERR(block)) {
2053 err = PTR_ERR(block);
2057 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2058 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2059 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2063 chain = tcf_chain_get(block, chain_index, true);
2065 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2070 mutex_lock(&chain->filter_chain_lock);
2071 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2072 prio, prio_allocate);
2074 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2080 struct tcf_proto *tp_new = NULL;
2082 if (chain->flushing) {
2087 /* Proto-tcf does not exist, create new one */
2089 if (tca[TCA_KIND] == NULL || !protocol) {
2090 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2095 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2096 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2102 prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2105 mutex_unlock(&chain->filter_chain_lock);
2106 tp_new = tcf_proto_create(nla_data(tca[TCA_KIND]),
2107 protocol, prio, chain, rtnl_held,
2109 if (IS_ERR(tp_new)) {
2110 err = PTR_ERR(tp_new);
2115 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2122 mutex_unlock(&chain->filter_chain_lock);
2125 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2126 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2131 fh = tp->ops->get(tp, t->tcm_handle);
2134 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2135 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2139 } else if (n->nlmsg_flags & NLM_F_EXCL) {
2140 tfilter_put(tp, fh);
2141 NL_SET_ERR_MSG(extack, "Filter already exists");
2146 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2147 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2152 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2153 n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
2156 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2157 RTM_NEWTFILTER, false, rtnl_held);
2158 tfilter_put(tp, fh);
2159 /* q pointer is NULL for shared blocks */
2161 q->flags &= ~TCQ_F_CAN_BYPASS;
2165 if (err && tp_created)
2166 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2169 if (tp && !IS_ERR(tp))
2170 tcf_proto_put(tp, rtnl_held, NULL);
2172 tcf_chain_put(chain);
2174 tcf_block_release(q, block, rtnl_held);
2179 if (err == -EAGAIN) {
2180 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2184 /* Replay the request. */
2190 mutex_unlock(&chain->filter_chain_lock);
2194 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2195 struct netlink_ext_ack *extack)
2197 struct net *net = sock_net(skb->sk);
2198 struct nlattr *tca[TCA_MAX + 1];
2204 struct Qdisc *q = NULL;
2205 struct tcf_chain_info chain_info;
2206 struct tcf_chain *chain = NULL;
2207 struct tcf_block *block = NULL;
2208 struct tcf_proto *tp = NULL;
2209 unsigned long cl = 0;
2212 bool rtnl_held = false;
2214 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2217 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2218 rtm_tca_policy, extack);
2223 protocol = TC_H_MIN(t->tcm_info);
2224 prio = TC_H_MAJ(t->tcm_info);
2225 parent = t->tcm_parent;
2227 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2228 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2232 /* Find head of filter chain. */
2234 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2238 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2239 * found), qdisc is not unlocked, classifier type is not specified,
2240 * classifier is not unlocked.
2243 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2244 !tca[TCA_KIND] || !tcf_proto_is_unlocked(nla_data(tca[TCA_KIND]))) {
2249 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2253 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2255 if (IS_ERR(block)) {
2256 err = PTR_ERR(block);
2260 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2261 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2262 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2266 chain = tcf_chain_get(block, chain_index, false);
2268 /* User requested flush on non-existent chain. Nothing to do,
2269 * so just return success.
2275 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2281 tfilter_notify_chain(net, skb, block, q, parent, n,
2282 chain, RTM_DELTFILTER, rtnl_held);
2283 tcf_chain_flush(chain, rtnl_held);
2288 mutex_lock(&chain->filter_chain_lock);
2289 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2291 if (!tp || IS_ERR(tp)) {
2292 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2293 err = tp ? PTR_ERR(tp) : -ENOENT;
2295 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2296 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2299 } else if (t->tcm_handle == 0) {
2300 tcf_chain_tp_remove(chain, &chain_info, tp);
2301 mutex_unlock(&chain->filter_chain_lock);
2303 tcf_proto_put(tp, rtnl_held, NULL);
2304 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2305 RTM_DELTFILTER, false, rtnl_held);
2309 mutex_unlock(&chain->filter_chain_lock);
2311 fh = tp->ops->get(tp, t->tcm_handle);
2314 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2319 err = tfilter_del_notify(net, skb, n, tp, block,
2320 q, parent, fh, false, &last,
2326 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2331 if (tp && !IS_ERR(tp))
2332 tcf_proto_put(tp, rtnl_held, NULL);
2333 tcf_chain_put(chain);
2335 tcf_block_release(q, block, rtnl_held);
2343 mutex_unlock(&chain->filter_chain_lock);
2347 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2348 struct netlink_ext_ack *extack)
2350 struct net *net = sock_net(skb->sk);
2351 struct nlattr *tca[TCA_MAX + 1];
2357 struct Qdisc *q = NULL;
2358 struct tcf_chain_info chain_info;
2359 struct tcf_chain *chain = NULL;
2360 struct tcf_block *block = NULL;
2361 struct tcf_proto *tp = NULL;
2362 unsigned long cl = 0;
2365 bool rtnl_held = false;
2367 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2368 rtm_tca_policy, extack);
2373 protocol = TC_H_MIN(t->tcm_info);
2374 prio = TC_H_MAJ(t->tcm_info);
2375 parent = t->tcm_parent;
2378 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2382 /* Find head of filter chain. */
2384 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2388 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2389 * unlocked, classifier type is not specified, classifier is not
2392 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2393 !tca[TCA_KIND] || !tcf_proto_is_unlocked(nla_data(tca[TCA_KIND]))) {
2398 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2402 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2404 if (IS_ERR(block)) {
2405 err = PTR_ERR(block);
2409 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2410 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2411 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2415 chain = tcf_chain_get(block, chain_index, false);
2417 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2422 mutex_lock(&chain->filter_chain_lock);
2423 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2425 mutex_unlock(&chain->filter_chain_lock);
2426 if (!tp || IS_ERR(tp)) {
2427 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2428 err = tp ? PTR_ERR(tp) : -ENOENT;
2430 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2431 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2436 fh = tp->ops->get(tp, t->tcm_handle);
2439 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2442 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2443 fh, RTM_NEWTFILTER, true, rtnl_held);
2445 NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2448 tfilter_put(tp, fh);
2451 if (tp && !IS_ERR(tp))
2452 tcf_proto_put(tp, rtnl_held, NULL);
2453 tcf_chain_put(chain);
2455 tcf_block_release(q, block, rtnl_held);
2463 struct tcf_dump_args {
2464 struct tcf_walker w;
2465 struct sk_buff *skb;
2466 struct netlink_callback *cb;
2467 struct tcf_block *block;
2472 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2474 struct tcf_dump_args *a = (void *)arg;
2475 struct net *net = sock_net(a->skb->sk);
2477 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2478 n, NETLINK_CB(a->cb->skb).portid,
2479 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2480 RTM_NEWTFILTER, true);
2483 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2484 struct sk_buff *skb, struct netlink_callback *cb,
2485 long index_start, long *p_index)
2487 struct net *net = sock_net(skb->sk);
2488 struct tcf_block *block = chain->block;
2489 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2490 struct tcf_proto *tp, *tp_prev;
2491 struct tcf_dump_args arg;
2493 for (tp = __tcf_get_next_proto(chain, NULL);
2496 tp = __tcf_get_next_proto(chain, tp),
2497 tcf_proto_put(tp_prev, true, NULL),
2499 if (*p_index < index_start)
2501 if (TC_H_MAJ(tcm->tcm_info) &&
2502 TC_H_MAJ(tcm->tcm_info) != tp->prio)
2504 if (TC_H_MIN(tcm->tcm_info) &&
2505 TC_H_MIN(tcm->tcm_info) != tp->protocol)
2507 if (*p_index > index_start)
2508 memset(&cb->args[1], 0,
2509 sizeof(cb->args) - sizeof(cb->args[0]));
2510 if (cb->args[1] == 0) {
2511 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2512 NETLINK_CB(cb->skb).portid,
2513 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2514 RTM_NEWTFILTER, true) <= 0)
2520 arg.w.fn = tcf_node_dump;
2525 arg.parent = parent;
2527 arg.w.skip = cb->args[1] - 1;
2529 arg.w.cookie = cb->args[2];
2530 tp->ops->walk(tp, &arg.w, true);
2531 cb->args[2] = arg.w.cookie;
2532 cb->args[1] = arg.w.count + 1;
2539 tcf_proto_put(tp, true, NULL);
2543 /* called with RTNL */
2544 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2546 struct tcf_chain *chain, *chain_prev;
2547 struct net *net = sock_net(skb->sk);
2548 struct nlattr *tca[TCA_MAX + 1];
2549 struct Qdisc *q = NULL;
2550 struct tcf_block *block;
2551 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2557 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2560 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2565 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2566 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2569 /* If we work with block index, q is NULL and parent value
2570 * will never be used in the following code. The check
2571 * in tcf_fill_node prevents it. However, compiler does not
2572 * see that far, so set parent to zero to silence the warning
2573 * about parent being uninitialized.
2577 const struct Qdisc_class_ops *cops;
2578 struct net_device *dev;
2579 unsigned long cl = 0;
2581 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2585 parent = tcm->tcm_parent;
2590 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2594 cops = q->ops->cl_ops;
2597 if (!cops->tcf_block)
2599 if (TC_H_MIN(tcm->tcm_parent)) {
2600 cl = cops->find(q, tcm->tcm_parent);
2604 block = cops->tcf_block(q, cl, NULL);
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)) {
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 = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2714 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2715 flags & NLM_F_ECHO);
2722 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2723 void *tmplt_priv, u32 chain_index,
2724 struct tcf_block *block, struct sk_buff *oskb,
2725 u32 seq, u16 flags, bool unicast)
2727 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2728 struct net *net = block->net;
2729 struct sk_buff *skb;
2731 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2735 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2736 block, portid, seq, flags, RTM_DELCHAIN) <= 0) {
2742 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2744 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2747 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2748 struct nlattr **tca,
2749 struct netlink_ext_ack *extack)
2751 const struct tcf_proto_ops *ops;
2754 /* If kind is not set, user did not specify template. */
2758 ops = tcf_proto_lookup_ops(nla_data(tca[TCA_KIND]), true, extack);
2760 return PTR_ERR(ops);
2761 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2762 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2766 tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2767 if (IS_ERR(tmplt_priv)) {
2768 module_put(ops->owner);
2769 return PTR_ERR(tmplt_priv);
2771 chain->tmplt_ops = ops;
2772 chain->tmplt_priv = tmplt_priv;
2776 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2779 /* If template ops are set, no work to do for us. */
2783 tmplt_ops->tmplt_destroy(tmplt_priv);
2784 module_put(tmplt_ops->owner);
2787 /* Add/delete/get a chain */
2789 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2790 struct netlink_ext_ack *extack)
2792 struct net *net = sock_net(skb->sk);
2793 struct nlattr *tca[TCA_MAX + 1];
2797 struct Qdisc *q = NULL;
2798 struct tcf_chain *chain = NULL;
2799 struct tcf_block *block;
2803 if (n->nlmsg_type != RTM_GETCHAIN &&
2804 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2808 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2809 rtm_tca_policy, extack);
2814 parent = t->tcm_parent;
2817 block = tcf_block_find(net, &q, &parent, &cl,
2818 t->tcm_ifindex, t->tcm_block_index, extack);
2820 return PTR_ERR(block);
2822 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2823 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2824 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2829 mutex_lock(&block->lock);
2830 chain = tcf_chain_lookup(block, chain_index);
2831 if (n->nlmsg_type == RTM_NEWCHAIN) {
2833 if (tcf_chain_held_by_acts_only(chain)) {
2834 /* The chain exists only because there is
2835 * some action referencing it.
2837 tcf_chain_hold(chain);
2839 NL_SET_ERR_MSG(extack, "Filter chain already exists");
2841 goto errout_block_locked;
2844 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2845 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
2847 goto errout_block_locked;
2849 chain = tcf_chain_create(block, chain_index);
2851 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
2853 goto errout_block_locked;
2857 if (!chain || tcf_chain_held_by_acts_only(chain)) {
2858 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2860 goto errout_block_locked;
2862 tcf_chain_hold(chain);
2865 if (n->nlmsg_type == RTM_NEWCHAIN) {
2866 /* Modifying chain requires holding parent block lock. In case
2867 * the chain was successfully added, take a reference to the
2868 * chain. This ensures that an empty chain does not disappear at
2869 * the end of this function.
2871 tcf_chain_hold(chain);
2872 chain->explicitly_created = true;
2874 mutex_unlock(&block->lock);
2876 switch (n->nlmsg_type) {
2878 err = tc_chain_tmplt_add(chain, net, tca, extack);
2880 tcf_chain_put_explicitly_created(chain);
2884 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2885 RTM_NEWCHAIN, false);
2888 tfilter_notify_chain(net, skb, block, q, parent, n,
2889 chain, RTM_DELTFILTER, true);
2890 /* Flush the chain first as the user requested chain removal. */
2891 tcf_chain_flush(chain, true);
2892 /* In case the chain was successfully deleted, put a reference
2893 * to the chain previously taken during addition.
2895 tcf_chain_put_explicitly_created(chain);
2898 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
2899 n->nlmsg_seq, n->nlmsg_type, true);
2901 NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2905 NL_SET_ERR_MSG(extack, "Unsupported message type");
2910 tcf_chain_put(chain);
2912 tcf_block_release(q, block, true);
2914 /* Replay the request. */
2918 errout_block_locked:
2919 mutex_unlock(&block->lock);
2923 /* called with RTNL */
2924 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2926 struct net *net = sock_net(skb->sk);
2927 struct nlattr *tca[TCA_MAX + 1];
2928 struct Qdisc *q = NULL;
2929 struct tcf_block *block;
2930 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2931 struct tcf_chain *chain;
2937 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2940 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2941 rtm_tca_policy, cb->extack);
2945 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2946 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2949 /* If we work with block index, q is NULL and parent value
2950 * will never be used in the following code. The check
2951 * in tcf_fill_node prevents it. However, compiler does not
2952 * see that far, so set parent to zero to silence the warning
2953 * about parent being uninitialized.
2957 const struct Qdisc_class_ops *cops;
2958 struct net_device *dev;
2959 unsigned long cl = 0;
2961 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2965 parent = tcm->tcm_parent;
2970 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2974 cops = q->ops->cl_ops;
2977 if (!cops->tcf_block)
2979 if (TC_H_MIN(tcm->tcm_parent)) {
2980 cl = cops->find(q, tcm->tcm_parent);
2984 block = cops->tcf_block(q, cl, NULL);
2987 if (tcf_block_shared(block))
2991 index_start = cb->args[0];
2994 mutex_lock(&block->lock);
2995 list_for_each_entry(chain, &block->chain_list, list) {
2996 if ((tca[TCA_CHAIN] &&
2997 nla_get_u32(tca[TCA_CHAIN]) != chain->index))
2999 if (index < index_start) {
3003 if (tcf_chain_held_by_acts_only(chain))
3005 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3006 chain->index, net, skb, block,
3007 NETLINK_CB(cb->skb).portid,
3008 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3014 mutex_unlock(&block->lock);
3016 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3017 tcf_block_refcnt_put(block, true);
3018 cb->args[0] = index;
3021 /* If we did no progress, the error (EMSGSIZE) is real */
3022 if (skb->len == 0 && err)
3027 void tcf_exts_destroy(struct tcf_exts *exts)
3029 #ifdef CONFIG_NET_CLS_ACT
3030 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3031 kfree(exts->actions);
3032 exts->nr_actions = 0;
3035 EXPORT_SYMBOL(tcf_exts_destroy);
3037 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3038 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
3039 bool rtnl_held, struct netlink_ext_ack *extack)
3041 #ifdef CONFIG_NET_CLS_ACT
3043 struct tc_action *act;
3044 size_t attr_size = 0;
3046 if (exts->police && tb[exts->police]) {
3047 act = tcf_action_init_1(net, tp, tb[exts->police],
3048 rate_tlv, "police", ovr,
3049 TCA_ACT_BIND, rtnl_held,
3052 return PTR_ERR(act);
3054 act->type = exts->type = TCA_OLD_COMPAT;
3055 exts->actions[0] = act;
3056 exts->nr_actions = 1;
3057 } else if (exts->action && tb[exts->action]) {
3060 err = tcf_action_init(net, tp, tb[exts->action],
3061 rate_tlv, NULL, ovr, TCA_ACT_BIND,
3062 exts->actions, &attr_size,
3066 exts->nr_actions = err;
3070 if ((exts->action && tb[exts->action]) ||
3071 (exts->police && tb[exts->police])) {
3072 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3079 EXPORT_SYMBOL(tcf_exts_validate);
3081 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3083 #ifdef CONFIG_NET_CLS_ACT
3084 struct tcf_exts old = *dst;
3087 tcf_exts_destroy(&old);
3090 EXPORT_SYMBOL(tcf_exts_change);
3092 #ifdef CONFIG_NET_CLS_ACT
3093 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3095 if (exts->nr_actions == 0)
3098 return exts->actions[0];
3102 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3104 #ifdef CONFIG_NET_CLS_ACT
3105 struct nlattr *nest;
3107 if (exts->action && tcf_exts_has_actions(exts)) {
3109 * again for backward compatible mode - we want
3110 * to work with both old and new modes of entering
3111 * tc data even if iproute2 was newer - jhs
3113 if (exts->type != TCA_OLD_COMPAT) {
3114 nest = nla_nest_start_noflag(skb, exts->action);
3116 goto nla_put_failure;
3118 if (tcf_action_dump(skb, exts->actions, 0, 0) < 0)
3119 goto nla_put_failure;
3120 nla_nest_end(skb, nest);
3121 } else if (exts->police) {
3122 struct tc_action *act = tcf_exts_first_act(exts);
3123 nest = nla_nest_start_noflag(skb, exts->police);
3124 if (nest == NULL || !act)
3125 goto nla_put_failure;
3126 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3127 goto nla_put_failure;
3128 nla_nest_end(skb, nest);
3134 nla_nest_cancel(skb, nest);
3140 EXPORT_SYMBOL(tcf_exts_dump);
3143 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3145 #ifdef CONFIG_NET_CLS_ACT
3146 struct tc_action *a = tcf_exts_first_act(exts);
3147 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3152 EXPORT_SYMBOL(tcf_exts_dump_stats);
3154 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3155 void *type_data, bool err_stop)
3157 struct flow_block_cb *block_cb;
3161 /* Make sure all netdevs sharing this block are offload-capable. */
3162 if (block->nooffloaddevcnt && err_stop)
3165 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3166 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3176 EXPORT_SYMBOL(tc_setup_cb_call);
3178 int tc_setup_flow_action(struct flow_action *flow_action,
3179 const struct tcf_exts *exts)
3181 const struct tc_action *act;
3188 tcf_exts_for_each_action(i, act, exts) {
3189 struct flow_action_entry *entry;
3191 entry = &flow_action->entries[j];
3192 if (is_tcf_gact_ok(act)) {
3193 entry->id = FLOW_ACTION_ACCEPT;
3194 } else if (is_tcf_gact_shot(act)) {
3195 entry->id = FLOW_ACTION_DROP;
3196 } else if (is_tcf_gact_trap(act)) {
3197 entry->id = FLOW_ACTION_TRAP;
3198 } else if (is_tcf_gact_goto_chain(act)) {
3199 entry->id = FLOW_ACTION_GOTO;
3200 entry->chain_index = tcf_gact_goto_chain_index(act);
3201 } else if (is_tcf_mirred_egress_redirect(act)) {
3202 entry->id = FLOW_ACTION_REDIRECT;
3203 entry->dev = tcf_mirred_dev(act);
3204 } else if (is_tcf_mirred_egress_mirror(act)) {
3205 entry->id = FLOW_ACTION_MIRRED;
3206 entry->dev = tcf_mirred_dev(act);
3207 } else if (is_tcf_vlan(act)) {
3208 switch (tcf_vlan_action(act)) {
3209 case TCA_VLAN_ACT_PUSH:
3210 entry->id = FLOW_ACTION_VLAN_PUSH;
3211 entry->vlan.vid = tcf_vlan_push_vid(act);
3212 entry->vlan.proto = tcf_vlan_push_proto(act);
3213 entry->vlan.prio = tcf_vlan_push_prio(act);
3215 case TCA_VLAN_ACT_POP:
3216 entry->id = FLOW_ACTION_VLAN_POP;
3218 case TCA_VLAN_ACT_MODIFY:
3219 entry->id = FLOW_ACTION_VLAN_MANGLE;
3220 entry->vlan.vid = tcf_vlan_push_vid(act);
3221 entry->vlan.proto = tcf_vlan_push_proto(act);
3222 entry->vlan.prio = tcf_vlan_push_prio(act);
3227 } else if (is_tcf_tunnel_set(act)) {
3228 entry->id = FLOW_ACTION_TUNNEL_ENCAP;
3229 entry->tunnel = tcf_tunnel_info(act);
3230 } else if (is_tcf_tunnel_release(act)) {
3231 entry->id = FLOW_ACTION_TUNNEL_DECAP;
3232 } else if (is_tcf_pedit(act)) {
3233 for (k = 0; k < tcf_pedit_nkeys(act); k++) {
3234 switch (tcf_pedit_cmd(act, k)) {
3235 case TCA_PEDIT_KEY_EX_CMD_SET:
3236 entry->id = FLOW_ACTION_MANGLE;
3238 case TCA_PEDIT_KEY_EX_CMD_ADD:
3239 entry->id = FLOW_ACTION_ADD;
3244 entry->mangle.htype = tcf_pedit_htype(act, k);
3245 entry->mangle.mask = tcf_pedit_mask(act, k);
3246 entry->mangle.val = tcf_pedit_val(act, k);
3247 entry->mangle.offset = tcf_pedit_offset(act, k);
3248 entry = &flow_action->entries[++j];
3250 } else if (is_tcf_csum(act)) {
3251 entry->id = FLOW_ACTION_CSUM;
3252 entry->csum_flags = tcf_csum_update_flags(act);
3253 } else if (is_tcf_skbedit_mark(act)) {
3254 entry->id = FLOW_ACTION_MARK;
3255 entry->mark = tcf_skbedit_mark(act);
3256 } else if (is_tcf_sample(act)) {
3257 entry->id = FLOW_ACTION_SAMPLE;
3258 entry->sample.psample_group =
3259 tcf_sample_psample_group(act);
3260 entry->sample.trunc_size = tcf_sample_trunc_size(act);
3261 entry->sample.truncate = tcf_sample_truncate(act);
3262 entry->sample.rate = tcf_sample_rate(act);
3263 } else if (is_tcf_police(act)) {
3264 entry->id = FLOW_ACTION_POLICE;
3265 entry->police.burst = tcf_police_tcfp_burst(act);
3266 entry->police.rate_bytes_ps =
3267 tcf_police_rate_bytes_ps(act);
3268 } else if (is_tcf_ct(act)) {
3269 entry->id = FLOW_ACTION_CT;
3270 entry->ct.action = tcf_ct_action(act);
3271 entry->ct.zone = tcf_ct_zone(act);
3276 if (!is_tcf_pedit(act))
3283 EXPORT_SYMBOL(tc_setup_flow_action);
3285 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3287 unsigned int num_acts = 0;
3288 struct tc_action *act;
3291 tcf_exts_for_each_action(i, act, exts) {
3292 if (is_tcf_pedit(act))
3293 num_acts += tcf_pedit_nkeys(act);
3299 EXPORT_SYMBOL(tcf_exts_num_actions);
3301 static __net_init int tcf_net_init(struct net *net)
3303 struct tcf_net *tn = net_generic(net, tcf_net_id);
3305 spin_lock_init(&tn->idr_lock);
3310 static void __net_exit tcf_net_exit(struct net *net)
3312 struct tcf_net *tn = net_generic(net, tcf_net_id);
3314 idr_destroy(&tn->idr);
3317 static struct pernet_operations tcf_net_ops = {
3318 .init = tcf_net_init,
3319 .exit = tcf_net_exit,
3321 .size = sizeof(struct tcf_net),
3324 static int __init tc_filter_init(void)
3328 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3332 err = register_pernet_subsys(&tcf_net_ops);
3334 goto err_register_pernet_subsys;
3336 err = rhashtable_init(&indr_setup_block_ht,
3337 &tc_indr_setup_block_ht_params);
3339 goto err_rhash_setup_block_ht;
3341 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3342 RTNL_FLAG_DOIT_UNLOCKED);
3343 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3344 RTNL_FLAG_DOIT_UNLOCKED);
3345 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3346 tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3347 rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3348 rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3349 rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3354 err_rhash_setup_block_ht:
3355 unregister_pernet_subsys(&tcf_net_ops);
3356 err_register_pernet_subsys:
3357 destroy_workqueue(tc_filter_wq);
3361 subsys_initcall(tc_filter_init);