1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * net/sched/cls_api.c Packet classifier API.
5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
9 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
12 #include <linux/module.h>
13 #include <linux/types.h>
14 #include <linux/kernel.h>
15 #include <linux/string.h>
16 #include <linux/errno.h>
17 #include <linux/err.h>
18 #include <linux/skbuff.h>
19 #include <linux/init.h>
20 #include <linux/kmod.h>
21 #include <linux/slab.h>
22 #include <linux/idr.h>
23 #include <linux/jhash.h>
24 #include <linux/rculist.h>
25 #include <linux/rhashtable.h>
26 #include <net/net_namespace.h>
28 #include <net/netlink.h>
29 #include <net/pkt_sched.h>
30 #include <net/pkt_cls.h>
31 #include <net/tc_act/tc_pedit.h>
32 #include <net/tc_act/tc_mirred.h>
33 #include <net/tc_act/tc_vlan.h>
34 #include <net/tc_act/tc_tunnel_key.h>
35 #include <net/tc_act/tc_csum.h>
36 #include <net/tc_act/tc_gact.h>
37 #include <net/tc_act/tc_police.h>
38 #include <net/tc_act/tc_sample.h>
39 #include <net/tc_act/tc_skbedit.h>
40 #include <net/tc_act/tc_ct.h>
41 #include <net/tc_act/tc_mpls.h>
42 #include <net/tc_act/tc_gate.h>
43 #include <net/flow_offload.h>
44 #include <net/tc_wrapper.h>
46 /* The list of all installed classifier types */
47 static LIST_HEAD(tcf_proto_base);
49 /* Protects list of registered TC modules. It is pure SMP lock. */
50 static DEFINE_RWLOCK(cls_mod_lock);
52 static struct xarray tcf_exts_miss_cookies_xa;
53 struct tcf_exts_miss_cookie_node {
54 const struct tcf_chain *chain;
55 const struct tcf_proto *tp;
56 const struct tcf_exts *exts;
64 /* Each tc action entry cookie will be comprised of 32bit miss_cookie_base +
65 * action index in the exts tc actions array.
67 union tcf_exts_miss_cookie {
75 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
77 tcf_exts_miss_cookie_base_alloc(struct tcf_exts *exts, struct tcf_proto *tp,
80 struct tcf_exts_miss_cookie_node *n;
84 if (WARN_ON(!handle || !tp->ops->get_exts))
87 n = kzalloc(sizeof(*n), GFP_KERNEL);
91 n->chain_index = tp->chain->index;
93 n->tp_prio = tp->prio;
98 err = xa_alloc_cyclic(&tcf_exts_miss_cookies_xa, &n->miss_cookie_base,
99 n, xa_limit_32b, &next, GFP_KERNEL);
103 exts->miss_cookie_node = n;
111 static void tcf_exts_miss_cookie_base_destroy(struct tcf_exts *exts)
113 struct tcf_exts_miss_cookie_node *n;
115 if (!exts->miss_cookie_node)
118 n = exts->miss_cookie_node;
119 xa_erase(&tcf_exts_miss_cookies_xa, n->miss_cookie_base);
123 static struct tcf_exts_miss_cookie_node *
124 tcf_exts_miss_cookie_lookup(u64 miss_cookie, int *act_index)
126 union tcf_exts_miss_cookie mc = { .miss_cookie = miss_cookie, };
128 *act_index = mc.act_index;
129 return xa_load(&tcf_exts_miss_cookies_xa, mc.miss_cookie_base);
131 #else /* IS_ENABLED(CONFIG_NET_TC_SKB_EXT) */
133 tcf_exts_miss_cookie_base_alloc(struct tcf_exts *exts, struct tcf_proto *tp,
139 static void tcf_exts_miss_cookie_base_destroy(struct tcf_exts *exts)
142 #endif /* IS_ENABLED(CONFIG_NET_TC_SKB_EXT) */
144 static u64 tcf_exts_miss_cookie_get(u32 miss_cookie_base, int act_index)
146 union tcf_exts_miss_cookie mc = { .act_index = act_index, };
148 if (!miss_cookie_base)
151 mc.miss_cookie_base = miss_cookie_base;
152 return mc.miss_cookie;
155 #ifdef CONFIG_NET_CLS_ACT
156 DEFINE_STATIC_KEY_FALSE(tc_skb_ext_tc);
157 EXPORT_SYMBOL(tc_skb_ext_tc);
159 void tc_skb_ext_tc_enable(void)
161 static_branch_inc(&tc_skb_ext_tc);
163 EXPORT_SYMBOL(tc_skb_ext_tc_enable);
165 void tc_skb_ext_tc_disable(void)
167 static_branch_dec(&tc_skb_ext_tc);
169 EXPORT_SYMBOL(tc_skb_ext_tc_disable);
172 static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
174 return jhash_3words(tp->chain->index, tp->prio,
175 (__force __u32)tp->protocol, 0);
178 static void tcf_proto_signal_destroying(struct tcf_chain *chain,
179 struct tcf_proto *tp)
181 struct tcf_block *block = chain->block;
183 mutex_lock(&block->proto_destroy_lock);
184 hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
185 destroy_obj_hashfn(tp));
186 mutex_unlock(&block->proto_destroy_lock);
189 static bool tcf_proto_cmp(const struct tcf_proto *tp1,
190 const struct tcf_proto *tp2)
192 return tp1->chain->index == tp2->chain->index &&
193 tp1->prio == tp2->prio &&
194 tp1->protocol == tp2->protocol;
197 static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
198 struct tcf_proto *tp)
200 u32 hash = destroy_obj_hashfn(tp);
201 struct tcf_proto *iter;
205 hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
206 destroy_ht_node, hash) {
207 if (tcf_proto_cmp(tp, iter)) {
218 tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
220 struct tcf_block *block = chain->block;
222 mutex_lock(&block->proto_destroy_lock);
223 if (hash_hashed(&tp->destroy_ht_node))
224 hash_del_rcu(&tp->destroy_ht_node);
225 mutex_unlock(&block->proto_destroy_lock);
228 /* Find classifier type by string name */
230 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
232 const struct tcf_proto_ops *t, *res = NULL;
235 read_lock(&cls_mod_lock);
236 list_for_each_entry(t, &tcf_proto_base, head) {
237 if (strcmp(kind, t->kind) == 0) {
238 if (try_module_get(t->owner))
243 read_unlock(&cls_mod_lock);
248 static const struct tcf_proto_ops *
249 tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
250 struct netlink_ext_ack *extack)
252 const struct tcf_proto_ops *ops;
254 ops = __tcf_proto_lookup_ops(kind);
257 #ifdef CONFIG_MODULES
260 request_module("cls_%s", kind);
263 ops = __tcf_proto_lookup_ops(kind);
264 /* We dropped the RTNL semaphore in order to perform
265 * the module load. So, even if we succeeded in loading
266 * the module we have to replay the request. We indicate
267 * this using -EAGAIN.
270 module_put(ops->owner);
271 return ERR_PTR(-EAGAIN);
274 NL_SET_ERR_MSG(extack, "TC classifier not found");
275 return ERR_PTR(-ENOENT);
278 /* Register(unregister) new classifier type */
280 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
282 struct tcf_proto_ops *t;
285 write_lock(&cls_mod_lock);
286 list_for_each_entry(t, &tcf_proto_base, head)
287 if (!strcmp(ops->kind, t->kind))
290 list_add_tail(&ops->head, &tcf_proto_base);
293 write_unlock(&cls_mod_lock);
296 EXPORT_SYMBOL(register_tcf_proto_ops);
298 static struct workqueue_struct *tc_filter_wq;
300 void unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
302 struct tcf_proto_ops *t;
305 /* Wait for outstanding call_rcu()s, if any, from a
306 * tcf_proto_ops's destroy() handler.
309 flush_workqueue(tc_filter_wq);
311 write_lock(&cls_mod_lock);
312 list_for_each_entry(t, &tcf_proto_base, head) {
319 write_unlock(&cls_mod_lock);
321 WARN(rc, "unregister tc filter kind(%s) failed %d\n", ops->kind, rc);
323 EXPORT_SYMBOL(unregister_tcf_proto_ops);
325 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
327 INIT_RCU_WORK(rwork, func);
328 return queue_rcu_work(tc_filter_wq, rwork);
330 EXPORT_SYMBOL(tcf_queue_work);
332 /* Select new prio value from the range, managed by kernel. */
334 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
336 u32 first = TC_H_MAKE(0xC0000000U, 0U);
339 first = tp->prio - 1;
341 return TC_H_MAJ(first);
344 static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
347 return nla_strscpy(name, kind, IFNAMSIZ) < 0;
348 memset(name, 0, IFNAMSIZ);
352 static bool tcf_proto_is_unlocked(const char *kind)
354 const struct tcf_proto_ops *ops;
357 if (strlen(kind) == 0)
360 ops = tcf_proto_lookup_ops(kind, false, NULL);
361 /* On error return false to take rtnl lock. Proto lookup/create
362 * functions will perform lookup again and properly handle errors.
367 ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
368 module_put(ops->owner);
372 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
373 u32 prio, struct tcf_chain *chain,
375 struct netlink_ext_ack *extack)
377 struct tcf_proto *tp;
380 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
382 return ERR_PTR(-ENOBUFS);
384 tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
385 if (IS_ERR(tp->ops)) {
386 err = PTR_ERR(tp->ops);
389 tp->classify = tp->ops->classify;
390 tp->protocol = protocol;
393 spin_lock_init(&tp->lock);
394 refcount_set(&tp->refcnt, 1);
396 err = tp->ops->init(tp);
398 module_put(tp->ops->owner);
408 static void tcf_proto_get(struct tcf_proto *tp)
410 refcount_inc(&tp->refcnt);
413 static void tcf_chain_put(struct tcf_chain *chain);
415 static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
416 bool sig_destroy, struct netlink_ext_ack *extack)
418 tp->ops->destroy(tp, rtnl_held, extack);
420 tcf_proto_signal_destroyed(tp->chain, tp);
421 tcf_chain_put(tp->chain);
422 module_put(tp->ops->owner);
426 static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
427 struct netlink_ext_ack *extack)
429 if (refcount_dec_and_test(&tp->refcnt))
430 tcf_proto_destroy(tp, rtnl_held, true, extack);
433 static bool tcf_proto_check_delete(struct tcf_proto *tp)
435 if (tp->ops->delete_empty)
436 return tp->ops->delete_empty(tp);
442 static void tcf_proto_mark_delete(struct tcf_proto *tp)
444 spin_lock(&tp->lock);
446 spin_unlock(&tp->lock);
449 static bool tcf_proto_is_deleting(struct tcf_proto *tp)
453 spin_lock(&tp->lock);
454 deleting = tp->deleting;
455 spin_unlock(&tp->lock);
460 #define ASSERT_BLOCK_LOCKED(block) \
461 lockdep_assert_held(&(block)->lock)
463 struct tcf_filter_chain_list_item {
464 struct list_head list;
465 tcf_chain_head_change_t *chain_head_change;
466 void *chain_head_change_priv;
469 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
472 struct tcf_chain *chain;
474 ASSERT_BLOCK_LOCKED(block);
476 chain = kzalloc(sizeof(*chain), GFP_KERNEL);
479 list_add_tail_rcu(&chain->list, &block->chain_list);
480 mutex_init(&chain->filter_chain_lock);
481 chain->block = block;
482 chain->index = chain_index;
485 block->chain0.chain = chain;
489 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
490 struct tcf_proto *tp_head)
492 if (item->chain_head_change)
493 item->chain_head_change(tp_head, item->chain_head_change_priv);
496 static void tcf_chain0_head_change(struct tcf_chain *chain,
497 struct tcf_proto *tp_head)
499 struct tcf_filter_chain_list_item *item;
500 struct tcf_block *block = chain->block;
505 mutex_lock(&block->lock);
506 list_for_each_entry(item, &block->chain0.filter_chain_list, list)
507 tcf_chain_head_change_item(item, tp_head);
508 mutex_unlock(&block->lock);
511 /* Returns true if block can be safely freed. */
513 static bool tcf_chain_detach(struct tcf_chain *chain)
515 struct tcf_block *block = chain->block;
517 ASSERT_BLOCK_LOCKED(block);
519 list_del_rcu(&chain->list);
521 block->chain0.chain = NULL;
523 if (list_empty(&block->chain_list) &&
524 refcount_read(&block->refcnt) == 0)
530 static void tcf_block_destroy(struct tcf_block *block)
532 mutex_destroy(&block->lock);
533 mutex_destroy(&block->proto_destroy_lock);
534 xa_destroy(&block->ports);
535 kfree_rcu(block, rcu);
538 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
540 struct tcf_block *block = chain->block;
542 mutex_destroy(&chain->filter_chain_lock);
543 kfree_rcu(chain, rcu);
545 tcf_block_destroy(block);
548 static void tcf_chain_hold(struct tcf_chain *chain)
550 ASSERT_BLOCK_LOCKED(chain->block);
555 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
557 ASSERT_BLOCK_LOCKED(chain->block);
559 /* In case all the references are action references, this
560 * chain should not be shown to the user.
562 return chain->refcnt == chain->action_refcnt;
565 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
568 struct tcf_chain *chain;
570 ASSERT_BLOCK_LOCKED(block);
572 list_for_each_entry(chain, &block->chain_list, list) {
573 if (chain->index == chain_index)
579 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
580 static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
583 struct tcf_chain *chain;
585 list_for_each_entry_rcu(chain, &block->chain_list, list) {
586 if (chain->index == chain_index)
593 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
594 u32 seq, u16 flags, int event, bool unicast,
595 struct netlink_ext_ack *extack);
597 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
598 u32 chain_index, bool create,
601 struct tcf_chain *chain = NULL;
602 bool is_first_reference;
604 mutex_lock(&block->lock);
605 chain = tcf_chain_lookup(block, chain_index);
607 tcf_chain_hold(chain);
611 chain = tcf_chain_create(block, chain_index);
617 ++chain->action_refcnt;
618 is_first_reference = chain->refcnt - chain->action_refcnt == 1;
619 mutex_unlock(&block->lock);
621 /* Send notification only in case we got the first
622 * non-action reference. Until then, the chain acts only as
623 * a placeholder for actions pointing to it and user ought
624 * not know about them.
626 if (is_first_reference && !by_act)
627 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
628 RTM_NEWCHAIN, false, NULL);
633 mutex_unlock(&block->lock);
637 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
640 return __tcf_chain_get(block, chain_index, create, false);
643 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
645 return __tcf_chain_get(block, chain_index, true, true);
647 EXPORT_SYMBOL(tcf_chain_get_by_act);
649 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
651 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
652 void *tmplt_priv, u32 chain_index,
653 struct tcf_block *block, struct sk_buff *oskb,
656 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
657 bool explicitly_created)
659 struct tcf_block *block = chain->block;
660 const struct tcf_proto_ops *tmplt_ops;
661 unsigned int refcnt, non_act_refcnt;
662 bool free_block = false;
665 mutex_lock(&block->lock);
666 if (explicitly_created) {
667 if (!chain->explicitly_created) {
668 mutex_unlock(&block->lock);
671 chain->explicitly_created = false;
675 chain->action_refcnt--;
677 /* tc_chain_notify_delete can't be called while holding block lock.
678 * However, when block is unlocked chain can be changed concurrently, so
679 * save these to temporary variables.
681 refcnt = --chain->refcnt;
682 non_act_refcnt = refcnt - chain->action_refcnt;
683 tmplt_ops = chain->tmplt_ops;
684 tmplt_priv = chain->tmplt_priv;
686 if (non_act_refcnt == chain->explicitly_created && !by_act) {
687 if (non_act_refcnt == 0)
688 tc_chain_notify_delete(tmplt_ops, tmplt_priv,
689 chain->index, block, NULL, 0, 0);
690 /* Last reference to chain, no need to lock. */
691 chain->flushing = false;
695 free_block = tcf_chain_detach(chain);
696 mutex_unlock(&block->lock);
699 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
700 tcf_chain_destroy(chain, free_block);
704 static void tcf_chain_put(struct tcf_chain *chain)
706 __tcf_chain_put(chain, false, false);
709 void tcf_chain_put_by_act(struct tcf_chain *chain)
711 __tcf_chain_put(chain, true, false);
713 EXPORT_SYMBOL(tcf_chain_put_by_act);
715 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
717 __tcf_chain_put(chain, false, true);
720 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
722 struct tcf_proto *tp, *tp_next;
724 mutex_lock(&chain->filter_chain_lock);
725 tp = tcf_chain_dereference(chain->filter_chain, chain);
727 tp_next = rcu_dereference_protected(tp->next, 1);
728 tcf_proto_signal_destroying(chain, tp);
731 tp = tcf_chain_dereference(chain->filter_chain, chain);
732 RCU_INIT_POINTER(chain->filter_chain, NULL);
733 tcf_chain0_head_change(chain, NULL);
734 chain->flushing = true;
735 mutex_unlock(&chain->filter_chain_lock);
738 tp_next = rcu_dereference_protected(tp->next, 1);
739 tcf_proto_put(tp, rtnl_held, NULL);
744 static int tcf_block_setup(struct tcf_block *block,
745 struct flow_block_offload *bo);
747 static void tcf_block_offload_init(struct flow_block_offload *bo,
748 struct net_device *dev, struct Qdisc *sch,
749 enum flow_block_command command,
750 enum flow_block_binder_type binder_type,
751 struct flow_block *flow_block,
752 bool shared, struct netlink_ext_ack *extack)
754 bo->net = dev_net(dev);
755 bo->command = command;
756 bo->binder_type = binder_type;
757 bo->block = flow_block;
758 bo->block_shared = shared;
761 bo->cb_list_head = &flow_block->cb_list;
762 INIT_LIST_HEAD(&bo->cb_list);
765 static void tcf_block_unbind(struct tcf_block *block,
766 struct flow_block_offload *bo);
768 static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
770 struct tcf_block *block = block_cb->indr.data;
771 struct net_device *dev = block_cb->indr.dev;
772 struct Qdisc *sch = block_cb->indr.sch;
773 struct netlink_ext_ack extack = {};
774 struct flow_block_offload bo = {};
776 tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
777 block_cb->indr.binder_type,
778 &block->flow_block, tcf_block_shared(block),
781 down_write(&block->cb_lock);
782 list_del(&block_cb->driver_list);
783 list_move(&block_cb->list, &bo.cb_list);
784 tcf_block_unbind(block, &bo);
785 up_write(&block->cb_lock);
789 static bool tcf_block_offload_in_use(struct tcf_block *block)
791 return atomic_read(&block->offloadcnt);
794 static int tcf_block_offload_cmd(struct tcf_block *block,
795 struct net_device *dev, struct Qdisc *sch,
796 struct tcf_block_ext_info *ei,
797 enum flow_block_command command,
798 struct netlink_ext_ack *extack)
800 struct flow_block_offload bo = {};
802 tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
803 &block->flow_block, tcf_block_shared(block),
806 if (dev->netdev_ops->ndo_setup_tc) {
809 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
811 if (err != -EOPNOTSUPP)
812 NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
816 return tcf_block_setup(block, &bo);
819 flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
820 tc_block_indr_cleanup);
821 tcf_block_setup(block, &bo);
826 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
827 struct tcf_block_ext_info *ei,
828 struct netlink_ext_ack *extack)
830 struct net_device *dev = q->dev_queue->dev;
833 down_write(&block->cb_lock);
835 /* If tc offload feature is disabled and the block we try to bind
836 * to already has some offloaded filters, forbid to bind.
838 if (dev->netdev_ops->ndo_setup_tc &&
839 !tc_can_offload(dev) &&
840 tcf_block_offload_in_use(block)) {
841 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
846 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
847 if (err == -EOPNOTSUPP)
848 goto no_offload_dev_inc;
852 up_write(&block->cb_lock);
856 if (tcf_block_offload_in_use(block))
860 block->nooffloaddevcnt++;
862 up_write(&block->cb_lock);
866 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
867 struct tcf_block_ext_info *ei)
869 struct net_device *dev = q->dev_queue->dev;
872 down_write(&block->cb_lock);
873 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
874 if (err == -EOPNOTSUPP)
875 goto no_offload_dev_dec;
876 up_write(&block->cb_lock);
880 WARN_ON(block->nooffloaddevcnt-- == 0);
881 up_write(&block->cb_lock);
885 tcf_chain0_head_change_cb_add(struct tcf_block *block,
886 struct tcf_block_ext_info *ei,
887 struct netlink_ext_ack *extack)
889 struct tcf_filter_chain_list_item *item;
890 struct tcf_chain *chain0;
892 item = kmalloc(sizeof(*item), GFP_KERNEL);
894 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
897 item->chain_head_change = ei->chain_head_change;
898 item->chain_head_change_priv = ei->chain_head_change_priv;
900 mutex_lock(&block->lock);
901 chain0 = block->chain0.chain;
903 tcf_chain_hold(chain0);
905 list_add(&item->list, &block->chain0.filter_chain_list);
906 mutex_unlock(&block->lock);
909 struct tcf_proto *tp_head;
911 mutex_lock(&chain0->filter_chain_lock);
913 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
915 tcf_chain_head_change_item(item, tp_head);
917 mutex_lock(&block->lock);
918 list_add(&item->list, &block->chain0.filter_chain_list);
919 mutex_unlock(&block->lock);
921 mutex_unlock(&chain0->filter_chain_lock);
922 tcf_chain_put(chain0);
929 tcf_chain0_head_change_cb_del(struct tcf_block *block,
930 struct tcf_block_ext_info *ei)
932 struct tcf_filter_chain_list_item *item;
934 mutex_lock(&block->lock);
935 list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
936 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
937 (item->chain_head_change == ei->chain_head_change &&
938 item->chain_head_change_priv == ei->chain_head_change_priv)) {
939 if (block->chain0.chain)
940 tcf_chain_head_change_item(item, NULL);
941 list_del(&item->list);
942 mutex_unlock(&block->lock);
948 mutex_unlock(&block->lock);
953 spinlock_t idr_lock; /* Protects idr */
957 static unsigned int tcf_net_id;
959 static int tcf_block_insert(struct tcf_block *block, struct net *net,
960 struct netlink_ext_ack *extack)
962 struct tcf_net *tn = net_generic(net, tcf_net_id);
965 idr_preload(GFP_KERNEL);
966 spin_lock(&tn->idr_lock);
967 err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
969 spin_unlock(&tn->idr_lock);
975 static void tcf_block_remove(struct tcf_block *block, struct net *net)
977 struct tcf_net *tn = net_generic(net, tcf_net_id);
979 spin_lock(&tn->idr_lock);
980 idr_remove(&tn->idr, block->index);
981 spin_unlock(&tn->idr_lock);
984 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
986 struct netlink_ext_ack *extack)
988 struct tcf_block *block;
990 block = kzalloc(sizeof(*block), GFP_KERNEL);
992 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
993 return ERR_PTR(-ENOMEM);
995 mutex_init(&block->lock);
996 mutex_init(&block->proto_destroy_lock);
997 init_rwsem(&block->cb_lock);
998 flow_block_init(&block->flow_block);
999 INIT_LIST_HEAD(&block->chain_list);
1000 INIT_LIST_HEAD(&block->owner_list);
1001 INIT_LIST_HEAD(&block->chain0.filter_chain_list);
1003 refcount_set(&block->refcnt, 1);
1005 block->index = block_index;
1006 xa_init(&block->ports);
1008 /* Don't store q pointer for blocks which are shared */
1009 if (!tcf_block_shared(block))
1014 struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
1016 struct tcf_net *tn = net_generic(net, tcf_net_id);
1018 return idr_find(&tn->idr, block_index);
1020 EXPORT_SYMBOL(tcf_block_lookup);
1022 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
1024 struct tcf_block *block;
1027 block = tcf_block_lookup(net, block_index);
1028 if (block && !refcount_inc_not_zero(&block->refcnt))
1035 static struct tcf_chain *
1036 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1038 mutex_lock(&block->lock);
1040 chain = list_is_last(&chain->list, &block->chain_list) ?
1041 NULL : list_next_entry(chain, list);
1043 chain = list_first_entry_or_null(&block->chain_list,
1044 struct tcf_chain, list);
1046 /* skip all action-only chains */
1047 while (chain && tcf_chain_held_by_acts_only(chain))
1048 chain = list_is_last(&chain->list, &block->chain_list) ?
1049 NULL : list_next_entry(chain, list);
1052 tcf_chain_hold(chain);
1053 mutex_unlock(&block->lock);
1058 /* Function to be used by all clients that want to iterate over all chains on
1059 * block. It properly obtains block->lock and takes reference to chain before
1060 * returning it. Users of this function must be tolerant to concurrent chain
1061 * insertion/deletion or ensure that no concurrent chain modification is
1062 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1063 * consistent dump because rtnl lock is released each time skb is filled with
1064 * data and sent to user-space.
1068 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1070 struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
1073 tcf_chain_put(chain);
1077 EXPORT_SYMBOL(tcf_get_next_chain);
1079 static struct tcf_proto *
1080 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1085 mutex_lock(&chain->filter_chain_lock);
1088 tp = tcf_chain_dereference(chain->filter_chain, chain);
1089 } else if (tcf_proto_is_deleting(tp)) {
1090 /* 'deleting' flag is set and chain->filter_chain_lock was
1091 * unlocked, which means next pointer could be invalid. Restart
1094 prio = tp->prio + 1;
1095 tp = tcf_chain_dereference(chain->filter_chain, chain);
1097 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1098 if (!tp->deleting && tp->prio >= prio)
1101 tp = tcf_chain_dereference(tp->next, chain);
1107 mutex_unlock(&chain->filter_chain_lock);
1112 /* Function to be used by all clients that want to iterate over all tp's on
1113 * chain. Users of this function must be tolerant to concurrent tp
1114 * insertion/deletion or ensure that no concurrent chain modification is
1115 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1116 * consistent dump because rtnl lock is released each time skb is filled with
1117 * data and sent to user-space.
1121 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1123 struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1126 tcf_proto_put(tp, true, NULL);
1130 EXPORT_SYMBOL(tcf_get_next_proto);
1132 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1134 struct tcf_chain *chain;
1136 /* Last reference to block. At this point chains cannot be added or
1137 * removed concurrently.
1139 for (chain = tcf_get_next_chain(block, NULL);
1141 chain = tcf_get_next_chain(block, chain)) {
1142 tcf_chain_put_explicitly_created(chain);
1143 tcf_chain_flush(chain, rtnl_held);
1147 /* Lookup Qdisc and increments its reference counter.
1148 * Set parent, if necessary.
1151 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1152 u32 *parent, int ifindex, bool rtnl_held,
1153 struct netlink_ext_ack *extack)
1155 const struct Qdisc_class_ops *cops;
1156 struct net_device *dev;
1159 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1165 dev = dev_get_by_index_rcu(net, ifindex);
1173 *q = rcu_dereference(dev->qdisc);
1174 *parent = (*q)->handle;
1176 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1178 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1184 *q = qdisc_refcount_inc_nz(*q);
1186 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1191 /* Is it classful? */
1192 cops = (*q)->ops->cl_ops;
1194 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1199 if (!cops->tcf_block) {
1200 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1206 /* At this point we know that qdisc is not noop_qdisc,
1207 * which means that qdisc holds a reference to net_device
1208 * and we hold a reference to qdisc, so it is safe to release
1220 qdisc_put_unlocked(*q);
1226 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1227 int ifindex, struct netlink_ext_ack *extack)
1229 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1232 /* Do we search for filter, attached to class? */
1233 if (TC_H_MIN(parent)) {
1234 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1236 *cl = cops->find(q, parent);
1238 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1246 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1247 unsigned long cl, int ifindex,
1249 struct netlink_ext_ack *extack)
1251 struct tcf_block *block;
1253 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1254 block = tcf_block_refcnt_get(net, block_index);
1256 NL_SET_ERR_MSG(extack, "Block of given index was not found");
1257 return ERR_PTR(-EINVAL);
1260 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1262 block = cops->tcf_block(q, cl, extack);
1264 return ERR_PTR(-EINVAL);
1266 if (tcf_block_shared(block)) {
1267 NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1268 return ERR_PTR(-EOPNOTSUPP);
1271 /* Always take reference to block in order to support execution
1272 * of rules update path of cls API without rtnl lock. Caller
1273 * must release block when it is finished using it. 'if' block
1274 * of this conditional obtain reference to block by calling
1275 * tcf_block_refcnt_get().
1277 refcount_inc(&block->refcnt);
1283 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1284 struct tcf_block_ext_info *ei, bool rtnl_held)
1286 if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1287 /* Flushing/putting all chains will cause the block to be
1288 * deallocated when last chain is freed. However, if chain_list
1289 * is empty, block has to be manually deallocated. After block
1290 * reference counter reached 0, it is no longer possible to
1291 * increment it or add new chains to block.
1293 bool free_block = list_empty(&block->chain_list);
1295 mutex_unlock(&block->lock);
1296 if (tcf_block_shared(block))
1297 tcf_block_remove(block, block->net);
1300 tcf_block_offload_unbind(block, q, ei);
1303 tcf_block_destroy(block);
1305 tcf_block_flush_all_chains(block, rtnl_held);
1307 tcf_block_offload_unbind(block, q, ei);
1311 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1313 __tcf_block_put(block, NULL, NULL, rtnl_held);
1317 * Set q, parent, cl when appropriate.
1320 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1321 u32 *parent, unsigned long *cl,
1322 int ifindex, u32 block_index,
1323 struct netlink_ext_ack *extack)
1325 struct tcf_block *block;
1330 err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1334 err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1338 block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1339 if (IS_ERR(block)) {
1340 err = PTR_ERR(block);
1351 return ERR_PTR(err);
1354 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1357 if (!IS_ERR_OR_NULL(block))
1358 tcf_block_refcnt_put(block, rtnl_held);
1364 qdisc_put_unlocked(q);
1368 struct tcf_block_owner_item {
1369 struct list_head list;
1371 enum flow_block_binder_type binder_type;
1375 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1377 enum flow_block_binder_type binder_type)
1379 if (block->keep_dst &&
1380 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1381 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1382 netif_keep_dst(qdisc_dev(q));
1385 void tcf_block_netif_keep_dst(struct tcf_block *block)
1387 struct tcf_block_owner_item *item;
1389 block->keep_dst = true;
1390 list_for_each_entry(item, &block->owner_list, list)
1391 tcf_block_owner_netif_keep_dst(block, item->q,
1394 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1396 static int tcf_block_owner_add(struct tcf_block *block,
1398 enum flow_block_binder_type binder_type)
1400 struct tcf_block_owner_item *item;
1402 item = kmalloc(sizeof(*item), GFP_KERNEL);
1406 item->binder_type = binder_type;
1407 list_add(&item->list, &block->owner_list);
1411 static void tcf_block_owner_del(struct tcf_block *block,
1413 enum flow_block_binder_type binder_type)
1415 struct tcf_block_owner_item *item;
1417 list_for_each_entry(item, &block->owner_list, list) {
1418 if (item->q == q && item->binder_type == binder_type) {
1419 list_del(&item->list);
1427 static bool tcf_block_tracks_dev(struct tcf_block *block,
1428 struct tcf_block_ext_info *ei)
1430 return tcf_block_shared(block) &&
1431 (ei->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS ||
1432 ei->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS);
1435 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1436 struct tcf_block_ext_info *ei,
1437 struct netlink_ext_ack *extack)
1439 struct net_device *dev = qdisc_dev(q);
1440 struct net *net = qdisc_net(q);
1441 struct tcf_block *block = NULL;
1444 if (ei->block_index)
1445 /* block_index not 0 means the shared block is requested */
1446 block = tcf_block_refcnt_get(net, ei->block_index);
1449 block = tcf_block_create(net, q, ei->block_index, extack);
1451 return PTR_ERR(block);
1452 if (tcf_block_shared(block)) {
1453 err = tcf_block_insert(block, net, extack);
1455 goto err_block_insert;
1459 err = tcf_block_owner_add(block, q, ei->binder_type);
1461 goto err_block_owner_add;
1463 tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1465 err = tcf_chain0_head_change_cb_add(block, ei, extack);
1467 goto err_chain0_head_change_cb_add;
1469 err = tcf_block_offload_bind(block, q, ei, extack);
1471 goto err_block_offload_bind;
1473 if (tcf_block_tracks_dev(block, ei)) {
1474 err = xa_insert(&block->ports, dev->ifindex, dev, GFP_KERNEL);
1476 NL_SET_ERR_MSG(extack, "block dev insert failed");
1477 goto err_dev_insert;
1485 err_block_offload_bind:
1486 tcf_chain0_head_change_cb_del(block, ei);
1487 err_chain0_head_change_cb_add:
1488 tcf_block_owner_del(block, q, ei->binder_type);
1489 err_block_owner_add:
1491 tcf_block_refcnt_put(block, true);
1494 EXPORT_SYMBOL(tcf_block_get_ext);
1496 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1498 struct tcf_proto __rcu **p_filter_chain = priv;
1500 rcu_assign_pointer(*p_filter_chain, tp_head);
1503 int tcf_block_get(struct tcf_block **p_block,
1504 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1505 struct netlink_ext_ack *extack)
1507 struct tcf_block_ext_info ei = {
1508 .chain_head_change = tcf_chain_head_change_dflt,
1509 .chain_head_change_priv = p_filter_chain,
1512 WARN_ON(!p_filter_chain);
1513 return tcf_block_get_ext(p_block, q, &ei, extack);
1515 EXPORT_SYMBOL(tcf_block_get);
1517 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1518 * actions should be all removed after flushing.
1520 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1521 struct tcf_block_ext_info *ei)
1523 struct net_device *dev = qdisc_dev(q);
1527 if (tcf_block_tracks_dev(block, ei))
1528 xa_erase(&block->ports, dev->ifindex);
1529 tcf_chain0_head_change_cb_del(block, ei);
1530 tcf_block_owner_del(block, q, ei->binder_type);
1532 __tcf_block_put(block, q, ei, true);
1534 EXPORT_SYMBOL(tcf_block_put_ext);
1536 void tcf_block_put(struct tcf_block *block)
1538 struct tcf_block_ext_info ei = {0, };
1542 tcf_block_put_ext(block, block->q, &ei);
1545 EXPORT_SYMBOL(tcf_block_put);
1548 tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1549 void *cb_priv, bool add, bool offload_in_use,
1550 struct netlink_ext_ack *extack)
1552 struct tcf_chain *chain, *chain_prev;
1553 struct tcf_proto *tp, *tp_prev;
1556 lockdep_assert_held(&block->cb_lock);
1558 for (chain = __tcf_get_next_chain(block, NULL);
1561 chain = __tcf_get_next_chain(block, chain),
1562 tcf_chain_put(chain_prev)) {
1563 if (chain->tmplt_ops && add)
1564 chain->tmplt_ops->tmplt_reoffload(chain, true, cb,
1566 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1568 tp = __tcf_get_next_proto(chain, tp),
1569 tcf_proto_put(tp_prev, true, NULL)) {
1570 if (tp->ops->reoffload) {
1571 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1574 goto err_playback_remove;
1575 } else if (add && offload_in_use) {
1577 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1578 goto err_playback_remove;
1581 if (chain->tmplt_ops && !add)
1582 chain->tmplt_ops->tmplt_reoffload(chain, false, cb,
1588 err_playback_remove:
1589 tcf_proto_put(tp, true, NULL);
1590 tcf_chain_put(chain);
1591 tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1596 static int tcf_block_bind(struct tcf_block *block,
1597 struct flow_block_offload *bo)
1599 struct flow_block_cb *block_cb, *next;
1602 lockdep_assert_held(&block->cb_lock);
1604 list_for_each_entry(block_cb, &bo->cb_list, list) {
1605 err = tcf_block_playback_offloads(block, block_cb->cb,
1606 block_cb->cb_priv, true,
1607 tcf_block_offload_in_use(block),
1611 if (!bo->unlocked_driver_cb)
1612 block->lockeddevcnt++;
1616 list_splice(&bo->cb_list, &block->flow_block.cb_list);
1621 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1622 list_del(&block_cb->driver_list);
1624 list_del(&block_cb->list);
1625 tcf_block_playback_offloads(block, block_cb->cb,
1626 block_cb->cb_priv, false,
1627 tcf_block_offload_in_use(block),
1629 if (!bo->unlocked_driver_cb)
1630 block->lockeddevcnt--;
1632 flow_block_cb_free(block_cb);
1638 static void tcf_block_unbind(struct tcf_block *block,
1639 struct flow_block_offload *bo)
1641 struct flow_block_cb *block_cb, *next;
1643 lockdep_assert_held(&block->cb_lock);
1645 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1646 tcf_block_playback_offloads(block, block_cb->cb,
1647 block_cb->cb_priv, false,
1648 tcf_block_offload_in_use(block),
1650 list_del(&block_cb->list);
1651 flow_block_cb_free(block_cb);
1652 if (!bo->unlocked_driver_cb)
1653 block->lockeddevcnt--;
1657 static int tcf_block_setup(struct tcf_block *block,
1658 struct flow_block_offload *bo)
1662 switch (bo->command) {
1663 case FLOW_BLOCK_BIND:
1664 err = tcf_block_bind(block, bo);
1666 case FLOW_BLOCK_UNBIND:
1668 tcf_block_unbind(block, bo);
1678 /* Main classifier routine: scans classifier chain attached
1679 * to this qdisc, (optionally) tests for protocol and asks
1680 * specific classifiers.
1682 static inline int __tcf_classify(struct sk_buff *skb,
1683 const struct tcf_proto *tp,
1684 const struct tcf_proto *orig_tp,
1685 struct tcf_result *res,
1687 struct tcf_exts_miss_cookie_node *n,
1689 u32 *last_executed_chain)
1691 #ifdef CONFIG_NET_CLS_ACT
1692 const int max_reclassify_loop = 16;
1693 const struct tcf_proto *first_tp;
1698 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1699 __be16 protocol = skb_protocol(skb, false);
1703 struct tcf_exts *exts;
1705 if (n->tp_prio != tp->prio)
1708 /* We re-lookup the tp and chain based on index instead
1709 * of having hard refs and locks to them, so do a sanity
1710 * check if any of tp,chain,exts was replaced by the
1711 * time we got here with a cookie from hardware.
1713 if (unlikely(n->tp != tp || n->tp->chain != n->chain ||
1714 !tp->ops->get_exts)) {
1715 tcf_set_drop_reason(skb,
1716 SKB_DROP_REASON_TC_COOKIE_ERROR);
1720 exts = tp->ops->get_exts(tp, n->handle);
1721 if (unlikely(!exts || n->exts != exts)) {
1722 tcf_set_drop_reason(skb,
1723 SKB_DROP_REASON_TC_COOKIE_ERROR);
1728 err = tcf_exts_exec_ex(skb, exts, act_index, res);
1730 if (tp->protocol != protocol &&
1731 tp->protocol != htons(ETH_P_ALL))
1734 err = tc_classify(skb, tp, res);
1736 #ifdef CONFIG_NET_CLS_ACT
1737 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1739 *last_executed_chain = first_tp->chain->index;
1741 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1742 first_tp = res->goto_tp;
1743 *last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1752 tcf_set_drop_reason(skb,
1753 SKB_DROP_REASON_TC_COOKIE_ERROR);
1757 return TC_ACT_UNSPEC; /* signal: continue lookup */
1758 #ifdef CONFIG_NET_CLS_ACT
1760 if (unlikely(limit++ >= max_reclassify_loop)) {
1761 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1762 tp->chain->block->index,
1764 ntohs(tp->protocol));
1765 tcf_set_drop_reason(skb,
1766 SKB_DROP_REASON_TC_RECLASSIFY_LOOP);
1775 int tcf_classify(struct sk_buff *skb,
1776 const struct tcf_block *block,
1777 const struct tcf_proto *tp,
1778 struct tcf_result *res, bool compat_mode)
1780 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1781 u32 last_executed_chain = 0;
1783 return __tcf_classify(skb, tp, tp, res, compat_mode, NULL, 0,
1784 &last_executed_chain);
1786 u32 last_executed_chain = tp ? tp->chain->index : 0;
1787 struct tcf_exts_miss_cookie_node *n = NULL;
1788 const struct tcf_proto *orig_tp = tp;
1789 struct tc_skb_ext *ext;
1794 ext = skb_ext_find(skb, TC_SKB_EXT);
1796 if (ext && (ext->chain || ext->act_miss)) {
1797 struct tcf_chain *fchain;
1800 if (ext->act_miss) {
1801 n = tcf_exts_miss_cookie_lookup(ext->act_miss_cookie,
1804 tcf_set_drop_reason(skb,
1805 SKB_DROP_REASON_TC_COOKIE_ERROR);
1809 chain = n->chain_index;
1814 fchain = tcf_chain_lookup_rcu(block, chain);
1816 tcf_set_drop_reason(skb,
1817 SKB_DROP_REASON_TC_CHAIN_NOTFOUND);
1822 /* Consume, so cloned/redirect skbs won't inherit ext */
1823 skb_ext_del(skb, TC_SKB_EXT);
1825 tp = rcu_dereference_bh(fchain->filter_chain);
1826 last_executed_chain = fchain->index;
1830 ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode, n, act_index,
1831 &last_executed_chain);
1833 if (tc_skb_ext_tc_enabled()) {
1834 /* If we missed on some chain */
1835 if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1836 struct tc_skb_cb *cb = tc_skb_cb(skb);
1838 ext = tc_skb_ext_alloc(skb);
1839 if (WARN_ON_ONCE(!ext)) {
1840 tcf_set_drop_reason(skb, SKB_DROP_REASON_NOMEM);
1843 ext->chain = last_executed_chain;
1845 ext->post_ct = cb->post_ct;
1846 ext->post_ct_snat = cb->post_ct_snat;
1847 ext->post_ct_dnat = cb->post_ct_dnat;
1848 ext->zone = cb->zone;
1855 EXPORT_SYMBOL(tcf_classify);
1857 struct tcf_chain_info {
1858 struct tcf_proto __rcu **pprev;
1859 struct tcf_proto __rcu *next;
1862 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1863 struct tcf_chain_info *chain_info)
1865 return tcf_chain_dereference(*chain_info->pprev, chain);
1868 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1869 struct tcf_chain_info *chain_info,
1870 struct tcf_proto *tp)
1872 if (chain->flushing)
1875 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1876 if (*chain_info->pprev == chain->filter_chain)
1877 tcf_chain0_head_change(chain, tp);
1879 rcu_assign_pointer(*chain_info->pprev, tp);
1884 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1885 struct tcf_chain_info *chain_info,
1886 struct tcf_proto *tp)
1888 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1890 tcf_proto_mark_delete(tp);
1891 if (tp == chain->filter_chain)
1892 tcf_chain0_head_change(chain, next);
1893 RCU_INIT_POINTER(*chain_info->pprev, next);
1896 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1897 struct tcf_chain_info *chain_info,
1898 u32 protocol, u32 prio,
1899 bool prio_allocate);
1901 /* Try to insert new proto.
1902 * If proto with specified priority already exists, free new proto
1903 * and return existing one.
1906 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1907 struct tcf_proto *tp_new,
1908 u32 protocol, u32 prio,
1911 struct tcf_chain_info chain_info;
1912 struct tcf_proto *tp;
1915 mutex_lock(&chain->filter_chain_lock);
1917 if (tcf_proto_exists_destroying(chain, tp_new)) {
1918 mutex_unlock(&chain->filter_chain_lock);
1919 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1920 return ERR_PTR(-EAGAIN);
1923 tp = tcf_chain_tp_find(chain, &chain_info,
1924 protocol, prio, false);
1926 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1927 mutex_unlock(&chain->filter_chain_lock);
1930 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1933 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1934 tp_new = ERR_PTR(err);
1940 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1941 struct tcf_proto *tp, bool rtnl_held,
1942 struct netlink_ext_ack *extack)
1944 struct tcf_chain_info chain_info;
1945 struct tcf_proto *tp_iter;
1946 struct tcf_proto **pprev;
1947 struct tcf_proto *next;
1949 mutex_lock(&chain->filter_chain_lock);
1951 /* Atomically find and remove tp from chain. */
1952 for (pprev = &chain->filter_chain;
1953 (tp_iter = tcf_chain_dereference(*pprev, chain));
1954 pprev = &tp_iter->next) {
1955 if (tp_iter == tp) {
1956 chain_info.pprev = pprev;
1957 chain_info.next = tp_iter->next;
1958 WARN_ON(tp_iter->deleting);
1962 /* Verify that tp still exists and no new filters were inserted
1964 * Mark tp for deletion if it is empty.
1966 if (!tp_iter || !tcf_proto_check_delete(tp)) {
1967 mutex_unlock(&chain->filter_chain_lock);
1971 tcf_proto_signal_destroying(chain, tp);
1972 next = tcf_chain_dereference(chain_info.next, chain);
1973 if (tp == chain->filter_chain)
1974 tcf_chain0_head_change(chain, next);
1975 RCU_INIT_POINTER(*chain_info.pprev, next);
1976 mutex_unlock(&chain->filter_chain_lock);
1978 tcf_proto_put(tp, rtnl_held, extack);
1981 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1982 struct tcf_chain_info *chain_info,
1983 u32 protocol, u32 prio,
1986 struct tcf_proto **pprev;
1987 struct tcf_proto *tp;
1989 /* Check the chain for existence of proto-tcf with this priority */
1990 for (pprev = &chain->filter_chain;
1991 (tp = tcf_chain_dereference(*pprev, chain));
1992 pprev = &tp->next) {
1993 if (tp->prio >= prio) {
1994 if (tp->prio == prio) {
1995 if (prio_allocate ||
1996 (tp->protocol != protocol && protocol))
1997 return ERR_PTR(-EINVAL);
2004 chain_info->pprev = pprev;
2006 chain_info->next = tp->next;
2009 chain_info->next = NULL;
2014 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
2015 struct tcf_proto *tp, struct tcf_block *block,
2016 struct Qdisc *q, u32 parent, void *fh,
2017 u32 portid, u32 seq, u16 flags, int event,
2018 bool terse_dump, bool rtnl_held,
2019 struct netlink_ext_ack *extack)
2022 struct nlmsghdr *nlh;
2023 unsigned char *b = skb_tail_pointer(skb);
2025 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2027 goto out_nlmsg_trim;
2028 tcm = nlmsg_data(nlh);
2029 tcm->tcm_family = AF_UNSPEC;
2033 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
2034 tcm->tcm_parent = parent;
2036 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2037 tcm->tcm_block_index = block->index;
2039 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
2040 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
2041 goto nla_put_failure;
2042 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
2043 goto nla_put_failure;
2045 tcm->tcm_handle = 0;
2046 } else if (terse_dump) {
2047 if (tp->ops->terse_dump) {
2048 if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
2050 goto nla_put_failure;
2052 goto cls_op_not_supp;
2055 if (tp->ops->dump &&
2056 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
2057 goto nla_put_failure;
2060 if (extack && extack->_msg &&
2061 nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2062 goto nla_put_failure;
2064 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2075 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
2076 struct nlmsghdr *n, struct tcf_proto *tp,
2077 struct tcf_block *block, struct Qdisc *q,
2078 u32 parent, void *fh, int event, bool unicast,
2079 bool rtnl_held, struct netlink_ext_ack *extack)
2081 struct sk_buff *skb;
2082 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2085 if (!unicast && !rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2088 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2092 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2093 n->nlmsg_seq, n->nlmsg_flags, event,
2094 false, rtnl_held, extack) <= 0) {
2100 err = rtnl_unicast(skb, net, portid);
2102 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2103 n->nlmsg_flags & NLM_F_ECHO);
2107 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
2108 struct nlmsghdr *n, struct tcf_proto *tp,
2109 struct tcf_block *block, struct Qdisc *q,
2110 u32 parent, void *fh, bool *last, bool rtnl_held,
2111 struct netlink_ext_ack *extack)
2113 struct sk_buff *skb;
2114 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2117 if (!rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2118 return tp->ops->delete(tp, fh, last, rtnl_held, extack);
2120 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2124 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2125 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
2126 false, rtnl_held, extack) <= 0) {
2127 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
2132 err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
2138 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2139 n->nlmsg_flags & NLM_F_ECHO);
2141 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
2146 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
2147 struct tcf_block *block, struct Qdisc *q,
2148 u32 parent, struct nlmsghdr *n,
2149 struct tcf_chain *chain, int event,
2150 struct netlink_ext_ack *extack)
2152 struct tcf_proto *tp;
2154 for (tp = tcf_get_next_proto(chain, NULL);
2155 tp; tp = tcf_get_next_proto(chain, tp))
2156 tfilter_notify(net, oskb, n, tp, block, q, parent, NULL,
2157 event, false, true, extack);
2160 static void tfilter_put(struct tcf_proto *tp, void *fh)
2162 if (tp->ops->put && fh)
2163 tp->ops->put(tp, fh);
2166 static bool is_qdisc_ingress(__u32 classid)
2168 return (TC_H_MIN(classid) == TC_H_MIN(TC_H_MIN_INGRESS));
2171 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2172 struct netlink_ext_ack *extack)
2174 struct net *net = sock_net(skb->sk);
2175 struct nlattr *tca[TCA_MAX + 1];
2176 char name[IFNAMSIZ];
2184 struct tcf_chain_info chain_info;
2185 struct tcf_chain *chain;
2186 struct tcf_block *block;
2187 struct tcf_proto *tp;
2192 bool rtnl_held = false;
2198 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2199 rtm_tca_policy, extack);
2204 protocol = TC_H_MIN(t->tcm_info);
2205 prio = TC_H_MAJ(t->tcm_info);
2206 prio_allocate = false;
2207 parent = t->tcm_parent;
2216 /* If no priority is provided by the user,
2219 if (n->nlmsg_flags & NLM_F_CREATE) {
2220 prio = TC_H_MAKE(0x80000000U, 0U);
2221 prio_allocate = true;
2223 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2228 /* Find head of filter chain. */
2230 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2234 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2235 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2240 /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2241 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2242 * type is not specified, classifier is not unlocked.
2245 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2246 !tcf_proto_is_unlocked(name)) {
2251 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2255 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2257 if (IS_ERR(block)) {
2258 err = PTR_ERR(block);
2261 block->classid = parent;
2263 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2264 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2265 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2269 chain = tcf_chain_get(block, chain_index, true);
2271 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2276 mutex_lock(&chain->filter_chain_lock);
2277 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2278 prio, prio_allocate);
2280 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2286 struct tcf_proto *tp_new = NULL;
2288 if (chain->flushing) {
2293 /* Proto-tcf does not exist, create new one */
2295 if (tca[TCA_KIND] == NULL || !protocol) {
2296 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2301 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2302 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2308 prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2311 mutex_unlock(&chain->filter_chain_lock);
2312 tp_new = tcf_proto_create(name, protocol, prio, chain,
2314 if (IS_ERR(tp_new)) {
2315 err = PTR_ERR(tp_new);
2320 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2327 mutex_unlock(&chain->filter_chain_lock);
2330 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2331 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2336 fh = tp->ops->get(tp, t->tcm_handle);
2339 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2340 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2344 } else if (n->nlmsg_flags & NLM_F_EXCL) {
2345 tfilter_put(tp, fh);
2346 NL_SET_ERR_MSG(extack, "Filter already exists");
2351 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2352 tfilter_put(tp, fh);
2353 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2358 if (!(n->nlmsg_flags & NLM_F_CREATE))
2359 flags |= TCA_ACT_FLAGS_REPLACE;
2361 flags |= TCA_ACT_FLAGS_NO_RTNL;
2362 if (is_qdisc_ingress(parent))
2363 flags |= TCA_ACT_FLAGS_AT_INGRESS;
2364 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2367 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2368 RTM_NEWTFILTER, false, rtnl_held, extack);
2369 tfilter_put(tp, fh);
2370 /* q pointer is NULL for shared blocks */
2372 q->flags &= ~TCQ_F_CAN_BYPASS;
2376 if (err && tp_created)
2377 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2380 if (tp && !IS_ERR(tp))
2381 tcf_proto_put(tp, rtnl_held, NULL);
2383 tcf_chain_put(chain);
2385 tcf_block_release(q, block, rtnl_held);
2390 if (err == -EAGAIN) {
2391 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2395 /* Replay the request. */
2401 mutex_unlock(&chain->filter_chain_lock);
2405 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2406 struct netlink_ext_ack *extack)
2408 struct net *net = sock_net(skb->sk);
2409 struct nlattr *tca[TCA_MAX + 1];
2410 char name[IFNAMSIZ];
2416 struct Qdisc *q = NULL;
2417 struct tcf_chain_info chain_info;
2418 struct tcf_chain *chain = NULL;
2419 struct tcf_block *block = NULL;
2420 struct tcf_proto *tp = NULL;
2421 unsigned long cl = 0;
2424 bool rtnl_held = false;
2426 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2427 rtm_tca_policy, extack);
2432 protocol = TC_H_MIN(t->tcm_info);
2433 prio = TC_H_MAJ(t->tcm_info);
2434 parent = t->tcm_parent;
2436 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2437 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2441 /* Find head of filter chain. */
2443 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2447 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2448 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2452 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2453 * found), qdisc is not unlocked, classifier type is not specified,
2454 * classifier is not unlocked.
2457 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2458 !tcf_proto_is_unlocked(name)) {
2463 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2467 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2469 if (IS_ERR(block)) {
2470 err = PTR_ERR(block);
2474 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2475 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2476 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2480 chain = tcf_chain_get(block, chain_index, false);
2482 /* User requested flush on non-existent chain. Nothing to do,
2483 * so just return success.
2489 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2495 tfilter_notify_chain(net, skb, block, q, parent, n,
2496 chain, RTM_DELTFILTER, extack);
2497 tcf_chain_flush(chain, rtnl_held);
2502 mutex_lock(&chain->filter_chain_lock);
2503 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2505 if (!tp || IS_ERR(tp)) {
2506 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2507 err = tp ? PTR_ERR(tp) : -ENOENT;
2509 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2510 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2513 } else if (t->tcm_handle == 0) {
2514 tcf_proto_signal_destroying(chain, tp);
2515 tcf_chain_tp_remove(chain, &chain_info, tp);
2516 mutex_unlock(&chain->filter_chain_lock);
2518 tcf_proto_put(tp, rtnl_held, NULL);
2519 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2520 RTM_DELTFILTER, false, rtnl_held, extack);
2524 mutex_unlock(&chain->filter_chain_lock);
2526 fh = tp->ops->get(tp, t->tcm_handle);
2529 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2534 err = tfilter_del_notify(net, skb, n, tp, block, q, parent, fh,
2535 &last, rtnl_held, extack);
2540 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2545 if (tp && !IS_ERR(tp))
2546 tcf_proto_put(tp, rtnl_held, NULL);
2547 tcf_chain_put(chain);
2549 tcf_block_release(q, block, rtnl_held);
2557 mutex_unlock(&chain->filter_chain_lock);
2561 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2562 struct netlink_ext_ack *extack)
2564 struct net *net = sock_net(skb->sk);
2565 struct nlattr *tca[TCA_MAX + 1];
2566 char name[IFNAMSIZ];
2572 struct Qdisc *q = NULL;
2573 struct tcf_chain_info chain_info;
2574 struct tcf_chain *chain = NULL;
2575 struct tcf_block *block = NULL;
2576 struct tcf_proto *tp = NULL;
2577 unsigned long cl = 0;
2580 bool rtnl_held = false;
2582 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2583 rtm_tca_policy, extack);
2588 protocol = TC_H_MIN(t->tcm_info);
2589 prio = TC_H_MAJ(t->tcm_info);
2590 parent = t->tcm_parent;
2593 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2597 /* Find head of filter chain. */
2599 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2603 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2604 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2608 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2609 * unlocked, classifier type is not specified, classifier is not
2612 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2613 !tcf_proto_is_unlocked(name)) {
2618 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2622 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2624 if (IS_ERR(block)) {
2625 err = PTR_ERR(block);
2629 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2630 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2631 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2635 chain = tcf_chain_get(block, chain_index, false);
2637 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2642 mutex_lock(&chain->filter_chain_lock);
2643 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2645 mutex_unlock(&chain->filter_chain_lock);
2646 if (!tp || IS_ERR(tp)) {
2647 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2648 err = tp ? PTR_ERR(tp) : -ENOENT;
2650 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2651 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2656 fh = tp->ops->get(tp, t->tcm_handle);
2659 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2662 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2663 fh, RTM_NEWTFILTER, true, rtnl_held, NULL);
2665 NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2668 tfilter_put(tp, fh);
2671 if (tp && !IS_ERR(tp))
2672 tcf_proto_put(tp, rtnl_held, NULL);
2673 tcf_chain_put(chain);
2675 tcf_block_release(q, block, rtnl_held);
2683 struct tcf_dump_args {
2684 struct tcf_walker w;
2685 struct sk_buff *skb;
2686 struct netlink_callback *cb;
2687 struct tcf_block *block;
2693 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2695 struct tcf_dump_args *a = (void *)arg;
2696 struct net *net = sock_net(a->skb->sk);
2698 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2699 n, NETLINK_CB(a->cb->skb).portid,
2700 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2701 RTM_NEWTFILTER, a->terse_dump, true, NULL);
2704 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2705 struct sk_buff *skb, struct netlink_callback *cb,
2706 long index_start, long *p_index, bool terse)
2708 struct net *net = sock_net(skb->sk);
2709 struct tcf_block *block = chain->block;
2710 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2711 struct tcf_proto *tp, *tp_prev;
2712 struct tcf_dump_args arg;
2714 for (tp = __tcf_get_next_proto(chain, NULL);
2717 tp = __tcf_get_next_proto(chain, tp),
2718 tcf_proto_put(tp_prev, true, NULL),
2720 if (*p_index < index_start)
2722 if (TC_H_MAJ(tcm->tcm_info) &&
2723 TC_H_MAJ(tcm->tcm_info) != tp->prio)
2725 if (TC_H_MIN(tcm->tcm_info) &&
2726 TC_H_MIN(tcm->tcm_info) != tp->protocol)
2728 if (*p_index > index_start)
2729 memset(&cb->args[1], 0,
2730 sizeof(cb->args) - sizeof(cb->args[0]));
2731 if (cb->args[1] == 0) {
2732 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2733 NETLINK_CB(cb->skb).portid,
2734 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2735 RTM_NEWTFILTER, false, true, NULL) <= 0)
2741 arg.w.fn = tcf_node_dump;
2746 arg.parent = parent;
2748 arg.w.skip = cb->args[1] - 1;
2750 arg.w.cookie = cb->args[2];
2751 arg.terse_dump = terse;
2752 tp->ops->walk(tp, &arg.w, true);
2753 cb->args[2] = arg.w.cookie;
2754 cb->args[1] = arg.w.count + 1;
2761 tcf_proto_put(tp, true, NULL);
2765 static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2766 [TCA_CHAIN] = { .type = NLA_U32 },
2767 [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2770 /* called with RTNL */
2771 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2773 struct tcf_chain *chain, *chain_prev;
2774 struct net *net = sock_net(skb->sk);
2775 struct nlattr *tca[TCA_MAX + 1];
2776 struct Qdisc *q = NULL;
2777 struct tcf_block *block;
2778 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2779 bool terse_dump = false;
2785 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2788 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2789 tcf_tfilter_dump_policy, cb->extack);
2793 if (tca[TCA_DUMP_FLAGS]) {
2794 struct nla_bitfield32 flags =
2795 nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2797 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2800 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2801 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2804 /* If we work with block index, q is NULL and parent value
2805 * will never be used in the following code. The check
2806 * in tcf_fill_node prevents it. However, compiler does not
2807 * see that far, so set parent to zero to silence the warning
2808 * about parent being uninitialized.
2812 const struct Qdisc_class_ops *cops;
2813 struct net_device *dev;
2814 unsigned long cl = 0;
2816 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2820 parent = tcm->tcm_parent;
2822 q = rtnl_dereference(dev->qdisc);
2824 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2827 cops = q->ops->cl_ops;
2830 if (!cops->tcf_block)
2832 if (TC_H_MIN(tcm->tcm_parent)) {
2833 cl = cops->find(q, tcm->tcm_parent);
2837 block = cops->tcf_block(q, cl, NULL);
2840 parent = block->classid;
2841 if (tcf_block_shared(block))
2845 index_start = cb->args[0];
2848 for (chain = __tcf_get_next_chain(block, NULL);
2851 chain = __tcf_get_next_chain(block, chain),
2852 tcf_chain_put(chain_prev)) {
2853 if (tca[TCA_CHAIN] &&
2854 nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2856 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2857 index_start, &index, terse_dump)) {
2858 tcf_chain_put(chain);
2864 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2865 tcf_block_refcnt_put(block, true);
2866 cb->args[0] = index;
2869 /* If we did no progress, the error (EMSGSIZE) is real */
2870 if (skb->len == 0 && err)
2875 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2876 void *tmplt_priv, u32 chain_index,
2877 struct net *net, struct sk_buff *skb,
2878 struct tcf_block *block,
2879 u32 portid, u32 seq, u16 flags, int event,
2880 struct netlink_ext_ack *extack)
2882 unsigned char *b = skb_tail_pointer(skb);
2883 const struct tcf_proto_ops *ops;
2884 struct nlmsghdr *nlh;
2891 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2893 goto out_nlmsg_trim;
2894 tcm = nlmsg_data(nlh);
2895 tcm->tcm_family = AF_UNSPEC;
2898 tcm->tcm_handle = 0;
2900 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2901 tcm->tcm_parent = block->q->handle;
2903 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2904 tcm->tcm_block_index = block->index;
2907 if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2908 goto nla_put_failure;
2911 if (nla_put_string(skb, TCA_KIND, ops->kind))
2912 goto nla_put_failure;
2913 if (ops->tmplt_dump(skb, net, priv) < 0)
2914 goto nla_put_failure;
2917 if (extack && extack->_msg &&
2918 nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2919 goto out_nlmsg_trim;
2921 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2931 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2932 u32 seq, u16 flags, int event, bool unicast,
2933 struct netlink_ext_ack *extack)
2935 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2936 struct tcf_block *block = chain->block;
2937 struct net *net = block->net;
2938 struct sk_buff *skb;
2941 if (!unicast && !rtnl_notify_needed(net, flags, RTNLGRP_TC))
2944 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2948 if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2949 chain->index, net, skb, block, portid,
2950 seq, flags, event, extack) <= 0) {
2956 err = rtnl_unicast(skb, net, portid);
2958 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2959 flags & NLM_F_ECHO);
2964 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2965 void *tmplt_priv, u32 chain_index,
2966 struct tcf_block *block, struct sk_buff *oskb,
2969 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2970 struct net *net = block->net;
2971 struct sk_buff *skb;
2973 if (!rtnl_notify_needed(net, flags, RTNLGRP_TC))
2976 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2980 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2981 block, portid, seq, flags, RTM_DELCHAIN, NULL) <= 0) {
2986 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2989 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2990 struct nlattr **tca,
2991 struct netlink_ext_ack *extack)
2993 const struct tcf_proto_ops *ops;
2994 char name[IFNAMSIZ];
2997 /* If kind is not set, user did not specify template. */
3001 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
3002 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
3006 ops = tcf_proto_lookup_ops(name, true, extack);
3008 return PTR_ERR(ops);
3009 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump ||
3010 !ops->tmplt_reoffload) {
3011 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
3012 module_put(ops->owner);
3016 tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
3017 if (IS_ERR(tmplt_priv)) {
3018 module_put(ops->owner);
3019 return PTR_ERR(tmplt_priv);
3021 chain->tmplt_ops = ops;
3022 chain->tmplt_priv = tmplt_priv;
3026 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
3029 /* If template ops are set, no work to do for us. */
3033 tmplt_ops->tmplt_destroy(tmplt_priv);
3034 module_put(tmplt_ops->owner);
3037 /* Add/delete/get a chain */
3039 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
3040 struct netlink_ext_ack *extack)
3042 struct net *net = sock_net(skb->sk);
3043 struct nlattr *tca[TCA_MAX + 1];
3048 struct tcf_chain *chain;
3049 struct tcf_block *block;
3055 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
3056 rtm_tca_policy, extack);
3061 parent = t->tcm_parent;
3064 block = tcf_block_find(net, &q, &parent, &cl,
3065 t->tcm_ifindex, t->tcm_block_index, extack);
3067 return PTR_ERR(block);
3069 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
3070 if (chain_index > TC_ACT_EXT_VAL_MASK) {
3071 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
3076 mutex_lock(&block->lock);
3077 chain = tcf_chain_lookup(block, chain_index);
3078 if (n->nlmsg_type == RTM_NEWCHAIN) {
3080 if (tcf_chain_held_by_acts_only(chain)) {
3081 /* The chain exists only because there is
3082 * some action referencing it.
3084 tcf_chain_hold(chain);
3086 NL_SET_ERR_MSG(extack, "Filter chain already exists");
3088 goto errout_block_locked;
3091 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
3092 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
3094 goto errout_block_locked;
3096 chain = tcf_chain_create(block, chain_index);
3098 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
3100 goto errout_block_locked;
3104 if (!chain || tcf_chain_held_by_acts_only(chain)) {
3105 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
3107 goto errout_block_locked;
3109 tcf_chain_hold(chain);
3112 if (n->nlmsg_type == RTM_NEWCHAIN) {
3113 /* Modifying chain requires holding parent block lock. In case
3114 * the chain was successfully added, take a reference to the
3115 * chain. This ensures that an empty chain does not disappear at
3116 * the end of this function.
3118 tcf_chain_hold(chain);
3119 chain->explicitly_created = true;
3121 mutex_unlock(&block->lock);
3123 switch (n->nlmsg_type) {
3125 err = tc_chain_tmplt_add(chain, net, tca, extack);
3127 tcf_chain_put_explicitly_created(chain);
3131 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
3132 RTM_NEWCHAIN, false, extack);
3135 tfilter_notify_chain(net, skb, block, q, parent, n,
3136 chain, RTM_DELTFILTER, extack);
3137 /* Flush the chain first as the user requested chain removal. */
3138 tcf_chain_flush(chain, true);
3139 /* In case the chain was successfully deleted, put a reference
3140 * to the chain previously taken during addition.
3142 tcf_chain_put_explicitly_created(chain);
3145 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
3146 n->nlmsg_flags, n->nlmsg_type, true, extack);
3148 NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
3152 NL_SET_ERR_MSG(extack, "Unsupported message type");
3157 tcf_chain_put(chain);
3159 tcf_block_release(q, block, true);
3161 /* Replay the request. */
3165 errout_block_locked:
3166 mutex_unlock(&block->lock);
3170 /* called with RTNL */
3171 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
3173 struct net *net = sock_net(skb->sk);
3174 struct nlattr *tca[TCA_MAX + 1];
3175 struct Qdisc *q = NULL;
3176 struct tcf_block *block;
3177 struct tcmsg *tcm = nlmsg_data(cb->nlh);
3178 struct tcf_chain *chain;
3183 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
3186 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
3187 rtm_tca_policy, cb->extack);
3191 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
3192 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
3196 const struct Qdisc_class_ops *cops;
3197 struct net_device *dev;
3198 unsigned long cl = 0;
3200 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
3204 if (!tcm->tcm_parent)
3205 q = rtnl_dereference(dev->qdisc);
3207 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
3211 cops = q->ops->cl_ops;
3214 if (!cops->tcf_block)
3216 if (TC_H_MIN(tcm->tcm_parent)) {
3217 cl = cops->find(q, tcm->tcm_parent);
3221 block = cops->tcf_block(q, cl, NULL);
3224 if (tcf_block_shared(block))
3228 index_start = cb->args[0];
3231 mutex_lock(&block->lock);
3232 list_for_each_entry(chain, &block->chain_list, list) {
3233 if ((tca[TCA_CHAIN] &&
3234 nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3236 if (index < index_start) {
3240 if (tcf_chain_held_by_acts_only(chain))
3242 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3243 chain->index, net, skb, block,
3244 NETLINK_CB(cb->skb).portid,
3245 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3246 RTM_NEWCHAIN, NULL);
3251 mutex_unlock(&block->lock);
3253 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3254 tcf_block_refcnt_put(block, true);
3255 cb->args[0] = index;
3258 /* If we did no progress, the error (EMSGSIZE) is real */
3259 if (skb->len == 0 && err)
3264 int tcf_exts_init_ex(struct tcf_exts *exts, struct net *net, int action,
3265 int police, struct tcf_proto *tp, u32 handle,
3266 bool use_action_miss)
3270 #ifdef CONFIG_NET_CLS_ACT
3272 exts->nr_actions = 0;
3273 exts->miss_cookie_node = NULL;
3274 /* Note: we do not own yet a reference on net.
3275 * This reference might be taken later from tcf_exts_get_net().
3278 exts->actions = kcalloc(TCA_ACT_MAX_PRIO, sizeof(struct tc_action *),
3284 exts->action = action;
3285 exts->police = police;
3287 if (!use_action_miss)
3290 err = tcf_exts_miss_cookie_base_alloc(exts, tp, handle);
3292 goto err_miss_alloc;
3297 tcf_exts_destroy(exts);
3298 #ifdef CONFIG_NET_CLS_ACT
3299 exts->actions = NULL;
3303 EXPORT_SYMBOL(tcf_exts_init_ex);
3305 void tcf_exts_destroy(struct tcf_exts *exts)
3307 tcf_exts_miss_cookie_base_destroy(exts);
3309 #ifdef CONFIG_NET_CLS_ACT
3310 if (exts->actions) {
3311 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3312 kfree(exts->actions);
3314 exts->nr_actions = 0;
3317 EXPORT_SYMBOL(tcf_exts_destroy);
3319 int tcf_exts_validate_ex(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3320 struct nlattr *rate_tlv, struct tcf_exts *exts,
3321 u32 flags, u32 fl_flags, struct netlink_ext_ack *extack)
3323 #ifdef CONFIG_NET_CLS_ACT
3325 int init_res[TCA_ACT_MAX_PRIO] = {};
3326 struct tc_action *act;
3327 size_t attr_size = 0;
3329 if (exts->police && tb[exts->police]) {
3330 struct tc_action_ops *a_o;
3332 flags |= TCA_ACT_FLAGS_POLICE | TCA_ACT_FLAGS_BIND;
3333 a_o = tc_action_load_ops(tb[exts->police], flags,
3336 return PTR_ERR(a_o);
3337 act = tcf_action_init_1(net, tp, tb[exts->police],
3338 rate_tlv, a_o, init_res, flags,
3340 module_put(a_o->owner);
3342 return PTR_ERR(act);
3344 act->type = exts->type = TCA_OLD_COMPAT;
3345 exts->actions[0] = act;
3346 exts->nr_actions = 1;
3347 tcf_idr_insert_many(exts->actions, init_res);
3348 } else if (exts->action && tb[exts->action]) {
3351 flags |= TCA_ACT_FLAGS_BIND;
3352 err = tcf_action_init(net, tp, tb[exts->action],
3353 rate_tlv, exts->actions, init_res,
3354 &attr_size, flags, fl_flags,
3358 exts->nr_actions = err;
3362 if ((exts->action && tb[exts->action]) ||
3363 (exts->police && tb[exts->police])) {
3364 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3371 EXPORT_SYMBOL(tcf_exts_validate_ex);
3373 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3374 struct nlattr *rate_tlv, struct tcf_exts *exts,
3375 u32 flags, struct netlink_ext_ack *extack)
3377 return tcf_exts_validate_ex(net, tp, tb, rate_tlv, exts,
3380 EXPORT_SYMBOL(tcf_exts_validate);
3382 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3384 #ifdef CONFIG_NET_CLS_ACT
3385 struct tcf_exts old = *dst;
3388 tcf_exts_destroy(&old);
3391 EXPORT_SYMBOL(tcf_exts_change);
3393 #ifdef CONFIG_NET_CLS_ACT
3394 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3396 if (exts->nr_actions == 0)
3399 return exts->actions[0];
3403 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3405 #ifdef CONFIG_NET_CLS_ACT
3406 struct nlattr *nest;
3408 if (exts->action && tcf_exts_has_actions(exts)) {
3410 * again for backward compatible mode - we want
3411 * to work with both old and new modes of entering
3412 * tc data even if iproute2 was newer - jhs
3414 if (exts->type != TCA_OLD_COMPAT) {
3415 nest = nla_nest_start_noflag(skb, exts->action);
3417 goto nla_put_failure;
3419 if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3421 goto nla_put_failure;
3422 nla_nest_end(skb, nest);
3423 } else if (exts->police) {
3424 struct tc_action *act = tcf_exts_first_act(exts);
3425 nest = nla_nest_start_noflag(skb, exts->police);
3426 if (nest == NULL || !act)
3427 goto nla_put_failure;
3428 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3429 goto nla_put_failure;
3430 nla_nest_end(skb, nest);
3436 nla_nest_cancel(skb, nest);
3442 EXPORT_SYMBOL(tcf_exts_dump);
3444 int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3446 #ifdef CONFIG_NET_CLS_ACT
3447 struct nlattr *nest;
3449 if (!exts->action || !tcf_exts_has_actions(exts))
3452 nest = nla_nest_start_noflag(skb, exts->action);
3454 goto nla_put_failure;
3456 if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3457 goto nla_put_failure;
3458 nla_nest_end(skb, nest);
3462 nla_nest_cancel(skb, nest);
3468 EXPORT_SYMBOL(tcf_exts_terse_dump);
3470 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3472 #ifdef CONFIG_NET_CLS_ACT
3473 struct tc_action *a = tcf_exts_first_act(exts);
3474 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3479 EXPORT_SYMBOL(tcf_exts_dump_stats);
3481 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3483 if (*flags & TCA_CLS_FLAGS_IN_HW)
3485 *flags |= TCA_CLS_FLAGS_IN_HW;
3486 atomic_inc(&block->offloadcnt);
3489 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3491 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3493 *flags &= ~TCA_CLS_FLAGS_IN_HW;
3494 atomic_dec(&block->offloadcnt);
3497 static void tc_cls_offload_cnt_update(struct tcf_block *block,
3498 struct tcf_proto *tp, u32 *cnt,
3499 u32 *flags, u32 diff, bool add)
3501 lockdep_assert_held(&block->cb_lock);
3503 spin_lock(&tp->lock);
3506 tcf_block_offload_inc(block, flags);
3511 tcf_block_offload_dec(block, flags);
3513 spin_unlock(&tp->lock);
3517 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3518 u32 *cnt, u32 *flags)
3520 lockdep_assert_held(&block->cb_lock);
3522 spin_lock(&tp->lock);
3523 tcf_block_offload_dec(block, flags);
3525 spin_unlock(&tp->lock);
3529 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3530 void *type_data, bool err_stop)
3532 struct flow_block_cb *block_cb;
3536 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3537 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3548 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3549 void *type_data, bool err_stop, bool rtnl_held)
3551 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3557 down_read(&block->cb_lock);
3558 /* Need to obtain rtnl lock if block is bound to devs that require it.
3559 * In block bind code cb_lock is obtained while holding rtnl, so we must
3560 * obtain the locks in same order here.
3562 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3563 up_read(&block->cb_lock);
3568 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3570 up_read(&block->cb_lock);
3575 EXPORT_SYMBOL(tc_setup_cb_call);
3577 /* Non-destructive filter add. If filter that wasn't already in hardware is
3578 * successfully offloaded, increment block offloads counter. On failure,
3579 * previously offloaded filter is considered to be intact and offloads counter
3580 * is not decremented.
3583 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3584 enum tc_setup_type type, void *type_data, bool err_stop,
3585 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3587 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3593 down_read(&block->cb_lock);
3594 /* Need to obtain rtnl lock if block is bound to devs that require it.
3595 * In block bind code cb_lock is obtained while holding rtnl, so we must
3596 * obtain the locks in same order here.
3598 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3599 up_read(&block->cb_lock);
3604 /* Make sure all netdevs sharing this block are offload-capable. */
3605 if (block->nooffloaddevcnt && err_stop) {
3606 ok_count = -EOPNOTSUPP;
3610 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3614 if (tp->ops->hw_add)
3615 tp->ops->hw_add(tp, type_data);
3617 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3620 up_read(&block->cb_lock);
3623 return min(ok_count, 0);
3625 EXPORT_SYMBOL(tc_setup_cb_add);
3627 /* Destructive filter replace. If filter that wasn't already in hardware is
3628 * successfully offloaded, increment block offload counter. On failure,
3629 * previously offloaded filter is considered to be destroyed and offload counter
3633 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3634 enum tc_setup_type type, void *type_data, bool err_stop,
3635 u32 *old_flags, unsigned int *old_in_hw_count,
3636 u32 *new_flags, unsigned int *new_in_hw_count,
3639 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3645 down_read(&block->cb_lock);
3646 /* Need to obtain rtnl lock if block is bound to devs that require it.
3647 * In block bind code cb_lock is obtained while holding rtnl, so we must
3648 * obtain the locks in same order here.
3650 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3651 up_read(&block->cb_lock);
3656 /* Make sure all netdevs sharing this block are offload-capable. */
3657 if (block->nooffloaddevcnt && err_stop) {
3658 ok_count = -EOPNOTSUPP;
3662 tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3663 if (tp->ops->hw_del)
3664 tp->ops->hw_del(tp, type_data);
3666 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3670 if (tp->ops->hw_add)
3671 tp->ops->hw_add(tp, type_data);
3673 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3674 new_flags, ok_count, true);
3676 up_read(&block->cb_lock);
3679 return min(ok_count, 0);
3681 EXPORT_SYMBOL(tc_setup_cb_replace);
3683 /* Destroy filter and decrement block offload counter, if filter was previously
3687 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3688 enum tc_setup_type type, void *type_data, bool err_stop,
3689 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3691 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3697 down_read(&block->cb_lock);
3698 /* Need to obtain rtnl lock if block is bound to devs that require it.
3699 * In block bind code cb_lock is obtained while holding rtnl, so we must
3700 * obtain the locks in same order here.
3702 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3703 up_read(&block->cb_lock);
3708 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3710 tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3711 if (tp->ops->hw_del)
3712 tp->ops->hw_del(tp, type_data);
3714 up_read(&block->cb_lock);
3717 return min(ok_count, 0);
3719 EXPORT_SYMBOL(tc_setup_cb_destroy);
3721 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3722 bool add, flow_setup_cb_t *cb,
3723 enum tc_setup_type type, void *type_data,
3724 void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3726 int err = cb(type, type_data, cb_priv);
3729 if (add && tc_skip_sw(*flags))
3732 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3738 EXPORT_SYMBOL(tc_setup_cb_reoffload);
3740 static int tcf_act_get_user_cookie(struct flow_action_entry *entry,
3741 const struct tc_action *act)
3743 struct tc_cookie *user_cookie;
3747 user_cookie = rcu_dereference(act->user_cookie);
3749 entry->user_cookie = flow_action_cookie_create(user_cookie->data,
3752 if (!entry->user_cookie)
3759 static void tcf_act_put_user_cookie(struct flow_action_entry *entry)
3761 flow_action_cookie_destroy(entry->user_cookie);
3764 void tc_cleanup_offload_action(struct flow_action *flow_action)
3766 struct flow_action_entry *entry;
3769 flow_action_for_each(i, entry, flow_action) {
3770 tcf_act_put_user_cookie(entry);
3771 if (entry->destructor)
3772 entry->destructor(entry->destructor_priv);
3775 EXPORT_SYMBOL(tc_cleanup_offload_action);
3777 static int tc_setup_offload_act(struct tc_action *act,
3778 struct flow_action_entry *entry,
3780 struct netlink_ext_ack *extack)
3782 #ifdef CONFIG_NET_CLS_ACT
3783 if (act->ops->offload_act_setup) {
3784 return act->ops->offload_act_setup(act, entry, index_inc, true,
3787 NL_SET_ERR_MSG(extack, "Action does not support offload");
3795 int tc_setup_action(struct flow_action *flow_action,
3796 struct tc_action *actions[],
3797 u32 miss_cookie_base,
3798 struct netlink_ext_ack *extack)
3800 int i, j, k, index, err = 0;
3801 struct tc_action *act;
3803 BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3804 BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3805 BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3811 tcf_act_for_each_action(i, act, actions) {
3812 struct flow_action_entry *entry;
3814 entry = &flow_action->entries[j];
3815 spin_lock_bh(&act->tcfa_lock);
3816 err = tcf_act_get_user_cookie(entry, act);
3818 goto err_out_locked;
3821 err = tc_setup_offload_act(act, entry, &index, extack);
3823 goto err_out_locked;
3825 for (k = 0; k < index ; k++) {
3826 entry[k].hw_stats = tc_act_hw_stats(act->hw_stats);
3827 entry[k].hw_index = act->tcfa_index;
3828 entry[k].cookie = (unsigned long)act;
3829 entry[k].miss_cookie =
3830 tcf_exts_miss_cookie_get(miss_cookie_base, i);
3835 spin_unlock_bh(&act->tcfa_lock);
3840 tc_cleanup_offload_action(flow_action);
3844 spin_unlock_bh(&act->tcfa_lock);
3848 int tc_setup_offload_action(struct flow_action *flow_action,
3849 const struct tcf_exts *exts,
3850 struct netlink_ext_ack *extack)
3852 #ifdef CONFIG_NET_CLS_ACT
3853 u32 miss_cookie_base;
3858 miss_cookie_base = exts->miss_cookie_node ?
3859 exts->miss_cookie_node->miss_cookie_base : 0;
3860 return tc_setup_action(flow_action, exts->actions, miss_cookie_base,
3866 EXPORT_SYMBOL(tc_setup_offload_action);
3868 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3870 unsigned int num_acts = 0;
3871 struct tc_action *act;
3874 tcf_exts_for_each_action(i, act, exts) {
3875 if (is_tcf_pedit(act))
3876 num_acts += tcf_pedit_nkeys(act);
3882 EXPORT_SYMBOL(tcf_exts_num_actions);
3884 #ifdef CONFIG_NET_CLS_ACT
3885 static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3887 struct netlink_ext_ack *extack)
3889 *p_block_index = nla_get_u32(block_index_attr);
3890 if (!*p_block_index) {
3891 NL_SET_ERR_MSG(extack, "Block number may not be zero");
3898 int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3899 enum flow_block_binder_type binder_type,
3900 struct nlattr *block_index_attr,
3901 struct netlink_ext_ack *extack)
3906 if (!block_index_attr)
3909 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3913 qe->info.binder_type = binder_type;
3914 qe->info.chain_head_change = tcf_chain_head_change_dflt;
3915 qe->info.chain_head_change_priv = &qe->filter_chain;
3916 qe->info.block_index = block_index;
3918 return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3920 EXPORT_SYMBOL(tcf_qevent_init);
3922 void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3924 if (qe->info.block_index)
3925 tcf_block_put_ext(qe->block, sch, &qe->info);
3927 EXPORT_SYMBOL(tcf_qevent_destroy);
3929 int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3930 struct netlink_ext_ack *extack)
3935 if (!block_index_attr)
3938 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3942 /* Bounce newly-configured block or change in block. */
3943 if (block_index != qe->info.block_index) {
3944 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3950 EXPORT_SYMBOL(tcf_qevent_validate_change);
3952 struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
3953 struct sk_buff **to_free, int *ret)
3955 struct tcf_result cl_res;
3956 struct tcf_proto *fl;
3958 if (!qe->info.block_index)
3961 fl = rcu_dereference_bh(qe->filter_chain);
3963 switch (tcf_classify(skb, NULL, fl, &cl_res, false)) {
3965 qdisc_qstats_drop(sch);
3966 __qdisc_drop(skb, to_free);
3967 *ret = __NET_XMIT_BYPASS;
3972 __qdisc_drop(skb, to_free);
3973 *ret = __NET_XMIT_STOLEN;
3975 case TC_ACT_REDIRECT:
3976 skb_do_redirect(skb);
3977 *ret = __NET_XMIT_STOLEN;
3983 EXPORT_SYMBOL(tcf_qevent_handle);
3985 int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
3987 if (!qe->info.block_index)
3989 return nla_put_u32(skb, attr_name, qe->info.block_index);
3991 EXPORT_SYMBOL(tcf_qevent_dump);
3994 static __net_init int tcf_net_init(struct net *net)
3996 struct tcf_net *tn = net_generic(net, tcf_net_id);
3998 spin_lock_init(&tn->idr_lock);
4003 static void __net_exit tcf_net_exit(struct net *net)
4005 struct tcf_net *tn = net_generic(net, tcf_net_id);
4007 idr_destroy(&tn->idr);
4010 static struct pernet_operations tcf_net_ops = {
4011 .init = tcf_net_init,
4012 .exit = tcf_net_exit,
4014 .size = sizeof(struct tcf_net),
4017 static int __init tc_filter_init(void)
4021 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
4025 err = register_pernet_subsys(&tcf_net_ops);
4027 goto err_register_pernet_subsys;
4029 xa_init_flags(&tcf_exts_miss_cookies_xa, XA_FLAGS_ALLOC1);
4031 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
4032 RTNL_FLAG_DOIT_UNLOCKED);
4033 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
4034 RTNL_FLAG_DOIT_UNLOCKED);
4035 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
4036 tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
4037 rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
4038 rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
4039 rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
4044 err_register_pernet_subsys:
4045 destroy_workqueue(tc_filter_wq);
4049 subsys_initcall(tc_filter_init);