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 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1428 struct tcf_block_ext_info *ei,
1429 struct netlink_ext_ack *extack)
1431 struct net *net = qdisc_net(q);
1432 struct tcf_block *block = NULL;
1435 if (ei->block_index)
1436 /* block_index not 0 means the shared block is requested */
1437 block = tcf_block_refcnt_get(net, ei->block_index);
1440 block = tcf_block_create(net, q, ei->block_index, extack);
1442 return PTR_ERR(block);
1443 if (tcf_block_shared(block)) {
1444 err = tcf_block_insert(block, net, extack);
1446 goto err_block_insert;
1450 err = tcf_block_owner_add(block, q, ei->binder_type);
1452 goto err_block_owner_add;
1454 tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1456 err = tcf_chain0_head_change_cb_add(block, ei, extack);
1458 goto err_chain0_head_change_cb_add;
1460 err = tcf_block_offload_bind(block, q, ei, extack);
1462 goto err_block_offload_bind;
1467 err_block_offload_bind:
1468 tcf_chain0_head_change_cb_del(block, ei);
1469 err_chain0_head_change_cb_add:
1470 tcf_block_owner_del(block, q, ei->binder_type);
1471 err_block_owner_add:
1473 tcf_block_refcnt_put(block, true);
1476 EXPORT_SYMBOL(tcf_block_get_ext);
1478 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1480 struct tcf_proto __rcu **p_filter_chain = priv;
1482 rcu_assign_pointer(*p_filter_chain, tp_head);
1485 int tcf_block_get(struct tcf_block **p_block,
1486 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1487 struct netlink_ext_ack *extack)
1489 struct tcf_block_ext_info ei = {
1490 .chain_head_change = tcf_chain_head_change_dflt,
1491 .chain_head_change_priv = p_filter_chain,
1494 WARN_ON(!p_filter_chain);
1495 return tcf_block_get_ext(p_block, q, &ei, extack);
1497 EXPORT_SYMBOL(tcf_block_get);
1499 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1500 * actions should be all removed after flushing.
1502 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1503 struct tcf_block_ext_info *ei)
1507 tcf_chain0_head_change_cb_del(block, ei);
1508 tcf_block_owner_del(block, q, ei->binder_type);
1510 __tcf_block_put(block, q, ei, true);
1512 EXPORT_SYMBOL(tcf_block_put_ext);
1514 void tcf_block_put(struct tcf_block *block)
1516 struct tcf_block_ext_info ei = {0, };
1520 tcf_block_put_ext(block, block->q, &ei);
1523 EXPORT_SYMBOL(tcf_block_put);
1526 tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1527 void *cb_priv, bool add, bool offload_in_use,
1528 struct netlink_ext_ack *extack)
1530 struct tcf_chain *chain, *chain_prev;
1531 struct tcf_proto *tp, *tp_prev;
1534 lockdep_assert_held(&block->cb_lock);
1536 for (chain = __tcf_get_next_chain(block, NULL);
1539 chain = __tcf_get_next_chain(block, chain),
1540 tcf_chain_put(chain_prev)) {
1541 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1543 tp = __tcf_get_next_proto(chain, tp),
1544 tcf_proto_put(tp_prev, true, NULL)) {
1545 if (tp->ops->reoffload) {
1546 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1549 goto err_playback_remove;
1550 } else if (add && offload_in_use) {
1552 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1553 goto err_playback_remove;
1560 err_playback_remove:
1561 tcf_proto_put(tp, true, NULL);
1562 tcf_chain_put(chain);
1563 tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1568 static int tcf_block_bind(struct tcf_block *block,
1569 struct flow_block_offload *bo)
1571 struct flow_block_cb *block_cb, *next;
1574 lockdep_assert_held(&block->cb_lock);
1576 list_for_each_entry(block_cb, &bo->cb_list, list) {
1577 err = tcf_block_playback_offloads(block, block_cb->cb,
1578 block_cb->cb_priv, true,
1579 tcf_block_offload_in_use(block),
1583 if (!bo->unlocked_driver_cb)
1584 block->lockeddevcnt++;
1588 list_splice(&bo->cb_list, &block->flow_block.cb_list);
1593 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1594 list_del(&block_cb->driver_list);
1596 list_del(&block_cb->list);
1597 tcf_block_playback_offloads(block, block_cb->cb,
1598 block_cb->cb_priv, false,
1599 tcf_block_offload_in_use(block),
1601 if (!bo->unlocked_driver_cb)
1602 block->lockeddevcnt--;
1604 flow_block_cb_free(block_cb);
1610 static void tcf_block_unbind(struct tcf_block *block,
1611 struct flow_block_offload *bo)
1613 struct flow_block_cb *block_cb, *next;
1615 lockdep_assert_held(&block->cb_lock);
1617 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1618 tcf_block_playback_offloads(block, block_cb->cb,
1619 block_cb->cb_priv, false,
1620 tcf_block_offload_in_use(block),
1622 list_del(&block_cb->list);
1623 flow_block_cb_free(block_cb);
1624 if (!bo->unlocked_driver_cb)
1625 block->lockeddevcnt--;
1629 static int tcf_block_setup(struct tcf_block *block,
1630 struct flow_block_offload *bo)
1634 switch (bo->command) {
1635 case FLOW_BLOCK_BIND:
1636 err = tcf_block_bind(block, bo);
1638 case FLOW_BLOCK_UNBIND:
1640 tcf_block_unbind(block, bo);
1650 /* Main classifier routine: scans classifier chain attached
1651 * to this qdisc, (optionally) tests for protocol and asks
1652 * specific classifiers.
1654 static inline int __tcf_classify(struct sk_buff *skb,
1655 const struct tcf_proto *tp,
1656 const struct tcf_proto *orig_tp,
1657 struct tcf_result *res,
1659 struct tcf_exts_miss_cookie_node *n,
1661 u32 *last_executed_chain)
1663 #ifdef CONFIG_NET_CLS_ACT
1664 const int max_reclassify_loop = 16;
1665 const struct tcf_proto *first_tp;
1670 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1671 __be16 protocol = skb_protocol(skb, false);
1675 struct tcf_exts *exts;
1677 if (n->tp_prio != tp->prio)
1680 /* We re-lookup the tp and chain based on index instead
1681 * of having hard refs and locks to them, so do a sanity
1682 * check if any of tp,chain,exts was replaced by the
1683 * time we got here with a cookie from hardware.
1685 if (unlikely(n->tp != tp || n->tp->chain != n->chain ||
1686 !tp->ops->get_exts)) {
1687 tcf_set_drop_reason(skb,
1688 SKB_DROP_REASON_TC_COOKIE_ERROR);
1692 exts = tp->ops->get_exts(tp, n->handle);
1693 if (unlikely(!exts || n->exts != exts)) {
1694 tcf_set_drop_reason(skb,
1695 SKB_DROP_REASON_TC_COOKIE_ERROR);
1700 err = tcf_exts_exec_ex(skb, exts, act_index, res);
1702 if (tp->protocol != protocol &&
1703 tp->protocol != htons(ETH_P_ALL))
1706 err = tc_classify(skb, tp, res);
1708 #ifdef CONFIG_NET_CLS_ACT
1709 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1711 *last_executed_chain = first_tp->chain->index;
1713 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1714 first_tp = res->goto_tp;
1715 *last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1724 tcf_set_drop_reason(skb,
1725 SKB_DROP_REASON_TC_COOKIE_ERROR);
1729 return TC_ACT_UNSPEC; /* signal: continue lookup */
1730 #ifdef CONFIG_NET_CLS_ACT
1732 if (unlikely(limit++ >= max_reclassify_loop)) {
1733 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1734 tp->chain->block->index,
1736 ntohs(tp->protocol));
1737 tcf_set_drop_reason(skb,
1738 SKB_DROP_REASON_TC_RECLASSIFY_LOOP);
1747 int tcf_classify(struct sk_buff *skb,
1748 const struct tcf_block *block,
1749 const struct tcf_proto *tp,
1750 struct tcf_result *res, bool compat_mode)
1752 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1753 u32 last_executed_chain = 0;
1755 return __tcf_classify(skb, tp, tp, res, compat_mode, NULL, 0,
1756 &last_executed_chain);
1758 u32 last_executed_chain = tp ? tp->chain->index : 0;
1759 struct tcf_exts_miss_cookie_node *n = NULL;
1760 const struct tcf_proto *orig_tp = tp;
1761 struct tc_skb_ext *ext;
1766 ext = skb_ext_find(skb, TC_SKB_EXT);
1768 if (ext && (ext->chain || ext->act_miss)) {
1769 struct tcf_chain *fchain;
1772 if (ext->act_miss) {
1773 n = tcf_exts_miss_cookie_lookup(ext->act_miss_cookie,
1776 tcf_set_drop_reason(skb,
1777 SKB_DROP_REASON_TC_COOKIE_ERROR);
1781 chain = n->chain_index;
1786 fchain = tcf_chain_lookup_rcu(block, chain);
1788 tcf_set_drop_reason(skb,
1789 SKB_DROP_REASON_TC_CHAIN_NOTFOUND);
1794 /* Consume, so cloned/redirect skbs won't inherit ext */
1795 skb_ext_del(skb, TC_SKB_EXT);
1797 tp = rcu_dereference_bh(fchain->filter_chain);
1798 last_executed_chain = fchain->index;
1802 ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode, n, act_index,
1803 &last_executed_chain);
1805 if (tc_skb_ext_tc_enabled()) {
1806 /* If we missed on some chain */
1807 if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1808 struct tc_skb_cb *cb = tc_skb_cb(skb);
1810 ext = tc_skb_ext_alloc(skb);
1811 if (WARN_ON_ONCE(!ext)) {
1812 tcf_set_drop_reason(skb, SKB_DROP_REASON_NOMEM);
1815 ext->chain = last_executed_chain;
1817 ext->post_ct = cb->post_ct;
1818 ext->post_ct_snat = cb->post_ct_snat;
1819 ext->post_ct_dnat = cb->post_ct_dnat;
1820 ext->zone = cb->zone;
1827 EXPORT_SYMBOL(tcf_classify);
1829 struct tcf_chain_info {
1830 struct tcf_proto __rcu **pprev;
1831 struct tcf_proto __rcu *next;
1834 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1835 struct tcf_chain_info *chain_info)
1837 return tcf_chain_dereference(*chain_info->pprev, chain);
1840 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1841 struct tcf_chain_info *chain_info,
1842 struct tcf_proto *tp)
1844 if (chain->flushing)
1847 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1848 if (*chain_info->pprev == chain->filter_chain)
1849 tcf_chain0_head_change(chain, tp);
1851 rcu_assign_pointer(*chain_info->pprev, tp);
1856 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1857 struct tcf_chain_info *chain_info,
1858 struct tcf_proto *tp)
1860 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1862 tcf_proto_mark_delete(tp);
1863 if (tp == chain->filter_chain)
1864 tcf_chain0_head_change(chain, next);
1865 RCU_INIT_POINTER(*chain_info->pprev, next);
1868 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1869 struct tcf_chain_info *chain_info,
1870 u32 protocol, u32 prio,
1871 bool prio_allocate);
1873 /* Try to insert new proto.
1874 * If proto with specified priority already exists, free new proto
1875 * and return existing one.
1878 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1879 struct tcf_proto *tp_new,
1880 u32 protocol, u32 prio,
1883 struct tcf_chain_info chain_info;
1884 struct tcf_proto *tp;
1887 mutex_lock(&chain->filter_chain_lock);
1889 if (tcf_proto_exists_destroying(chain, tp_new)) {
1890 mutex_unlock(&chain->filter_chain_lock);
1891 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1892 return ERR_PTR(-EAGAIN);
1895 tp = tcf_chain_tp_find(chain, &chain_info,
1896 protocol, prio, false);
1898 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1899 mutex_unlock(&chain->filter_chain_lock);
1902 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1905 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1906 tp_new = ERR_PTR(err);
1912 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1913 struct tcf_proto *tp, bool rtnl_held,
1914 struct netlink_ext_ack *extack)
1916 struct tcf_chain_info chain_info;
1917 struct tcf_proto *tp_iter;
1918 struct tcf_proto **pprev;
1919 struct tcf_proto *next;
1921 mutex_lock(&chain->filter_chain_lock);
1923 /* Atomically find and remove tp from chain. */
1924 for (pprev = &chain->filter_chain;
1925 (tp_iter = tcf_chain_dereference(*pprev, chain));
1926 pprev = &tp_iter->next) {
1927 if (tp_iter == tp) {
1928 chain_info.pprev = pprev;
1929 chain_info.next = tp_iter->next;
1930 WARN_ON(tp_iter->deleting);
1934 /* Verify that tp still exists and no new filters were inserted
1936 * Mark tp for deletion if it is empty.
1938 if (!tp_iter || !tcf_proto_check_delete(tp)) {
1939 mutex_unlock(&chain->filter_chain_lock);
1943 tcf_proto_signal_destroying(chain, tp);
1944 next = tcf_chain_dereference(chain_info.next, chain);
1945 if (tp == chain->filter_chain)
1946 tcf_chain0_head_change(chain, next);
1947 RCU_INIT_POINTER(*chain_info.pprev, next);
1948 mutex_unlock(&chain->filter_chain_lock);
1950 tcf_proto_put(tp, rtnl_held, extack);
1953 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1954 struct tcf_chain_info *chain_info,
1955 u32 protocol, u32 prio,
1958 struct tcf_proto **pprev;
1959 struct tcf_proto *tp;
1961 /* Check the chain for existence of proto-tcf with this priority */
1962 for (pprev = &chain->filter_chain;
1963 (tp = tcf_chain_dereference(*pprev, chain));
1964 pprev = &tp->next) {
1965 if (tp->prio >= prio) {
1966 if (tp->prio == prio) {
1967 if (prio_allocate ||
1968 (tp->protocol != protocol && protocol))
1969 return ERR_PTR(-EINVAL);
1976 chain_info->pprev = pprev;
1978 chain_info->next = tp->next;
1981 chain_info->next = NULL;
1986 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1987 struct tcf_proto *tp, struct tcf_block *block,
1988 struct Qdisc *q, u32 parent, void *fh,
1989 u32 portid, u32 seq, u16 flags, int event,
1990 bool terse_dump, bool rtnl_held,
1991 struct netlink_ext_ack *extack)
1994 struct nlmsghdr *nlh;
1995 unsigned char *b = skb_tail_pointer(skb);
1997 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1999 goto out_nlmsg_trim;
2000 tcm = nlmsg_data(nlh);
2001 tcm->tcm_family = AF_UNSPEC;
2005 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
2006 tcm->tcm_parent = parent;
2008 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2009 tcm->tcm_block_index = block->index;
2011 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
2012 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
2013 goto nla_put_failure;
2014 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
2015 goto nla_put_failure;
2017 tcm->tcm_handle = 0;
2018 } else if (terse_dump) {
2019 if (tp->ops->terse_dump) {
2020 if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
2022 goto nla_put_failure;
2024 goto cls_op_not_supp;
2027 if (tp->ops->dump &&
2028 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
2029 goto nla_put_failure;
2032 if (extack && extack->_msg &&
2033 nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2034 goto nla_put_failure;
2036 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2047 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
2048 struct nlmsghdr *n, struct tcf_proto *tp,
2049 struct tcf_block *block, struct Qdisc *q,
2050 u32 parent, void *fh, int event, bool unicast,
2051 bool rtnl_held, struct netlink_ext_ack *extack)
2053 struct sk_buff *skb;
2054 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2057 if (!unicast && !rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2060 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2064 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2065 n->nlmsg_seq, n->nlmsg_flags, event,
2066 false, rtnl_held, extack) <= 0) {
2072 err = rtnl_unicast(skb, net, portid);
2074 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2075 n->nlmsg_flags & NLM_F_ECHO);
2079 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
2080 struct nlmsghdr *n, struct tcf_proto *tp,
2081 struct tcf_block *block, struct Qdisc *q,
2082 u32 parent, void *fh, bool *last, bool rtnl_held,
2083 struct netlink_ext_ack *extack)
2085 struct sk_buff *skb;
2086 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2089 if (!rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2090 return tp->ops->delete(tp, fh, last, rtnl_held, extack);
2092 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2096 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2097 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
2098 false, rtnl_held, extack) <= 0) {
2099 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
2104 err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
2110 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2111 n->nlmsg_flags & NLM_F_ECHO);
2113 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
2118 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
2119 struct tcf_block *block, struct Qdisc *q,
2120 u32 parent, struct nlmsghdr *n,
2121 struct tcf_chain *chain, int event,
2122 struct netlink_ext_ack *extack)
2124 struct tcf_proto *tp;
2126 for (tp = tcf_get_next_proto(chain, NULL);
2127 tp; tp = tcf_get_next_proto(chain, tp))
2128 tfilter_notify(net, oskb, n, tp, block, q, parent, NULL,
2129 event, false, true, extack);
2132 static void tfilter_put(struct tcf_proto *tp, void *fh)
2134 if (tp->ops->put && fh)
2135 tp->ops->put(tp, fh);
2138 static bool is_qdisc_ingress(__u32 classid)
2140 return (TC_H_MIN(classid) == TC_H_MIN(TC_H_MIN_INGRESS));
2143 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2144 struct netlink_ext_ack *extack)
2146 struct net *net = sock_net(skb->sk);
2147 struct nlattr *tca[TCA_MAX + 1];
2148 char name[IFNAMSIZ];
2156 struct tcf_chain_info chain_info;
2157 struct tcf_chain *chain;
2158 struct tcf_block *block;
2159 struct tcf_proto *tp;
2164 bool rtnl_held = false;
2170 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2171 rtm_tca_policy, extack);
2176 protocol = TC_H_MIN(t->tcm_info);
2177 prio = TC_H_MAJ(t->tcm_info);
2178 prio_allocate = false;
2179 parent = t->tcm_parent;
2188 /* If no priority is provided by the user,
2191 if (n->nlmsg_flags & NLM_F_CREATE) {
2192 prio = TC_H_MAKE(0x80000000U, 0U);
2193 prio_allocate = true;
2195 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2200 /* Find head of filter chain. */
2202 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2206 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2207 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2212 /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2213 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2214 * type is not specified, classifier is not unlocked.
2217 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2218 !tcf_proto_is_unlocked(name)) {
2223 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2227 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2229 if (IS_ERR(block)) {
2230 err = PTR_ERR(block);
2233 block->classid = parent;
2235 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2236 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2237 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2241 chain = tcf_chain_get(block, chain_index, true);
2243 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2248 mutex_lock(&chain->filter_chain_lock);
2249 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2250 prio, prio_allocate);
2252 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2258 struct tcf_proto *tp_new = NULL;
2260 if (chain->flushing) {
2265 /* Proto-tcf does not exist, create new one */
2267 if (tca[TCA_KIND] == NULL || !protocol) {
2268 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2273 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2274 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2280 prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2283 mutex_unlock(&chain->filter_chain_lock);
2284 tp_new = tcf_proto_create(name, protocol, prio, chain,
2286 if (IS_ERR(tp_new)) {
2287 err = PTR_ERR(tp_new);
2292 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2299 mutex_unlock(&chain->filter_chain_lock);
2302 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2303 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2308 fh = tp->ops->get(tp, t->tcm_handle);
2311 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2312 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2316 } else if (n->nlmsg_flags & NLM_F_EXCL) {
2317 tfilter_put(tp, fh);
2318 NL_SET_ERR_MSG(extack, "Filter already exists");
2323 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2324 tfilter_put(tp, fh);
2325 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2330 if (!(n->nlmsg_flags & NLM_F_CREATE))
2331 flags |= TCA_ACT_FLAGS_REPLACE;
2333 flags |= TCA_ACT_FLAGS_NO_RTNL;
2334 if (is_qdisc_ingress(parent))
2335 flags |= TCA_ACT_FLAGS_AT_INGRESS;
2336 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2339 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2340 RTM_NEWTFILTER, false, rtnl_held, extack);
2341 tfilter_put(tp, fh);
2342 /* q pointer is NULL for shared blocks */
2344 q->flags &= ~TCQ_F_CAN_BYPASS;
2348 if (err && tp_created)
2349 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2352 if (tp && !IS_ERR(tp))
2353 tcf_proto_put(tp, rtnl_held, NULL);
2355 tcf_chain_put(chain);
2357 tcf_block_release(q, block, rtnl_held);
2362 if (err == -EAGAIN) {
2363 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2367 /* Replay the request. */
2373 mutex_unlock(&chain->filter_chain_lock);
2377 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2378 struct netlink_ext_ack *extack)
2380 struct net *net = sock_net(skb->sk);
2381 struct nlattr *tca[TCA_MAX + 1];
2382 char name[IFNAMSIZ];
2388 struct Qdisc *q = NULL;
2389 struct tcf_chain_info chain_info;
2390 struct tcf_chain *chain = NULL;
2391 struct tcf_block *block = NULL;
2392 struct tcf_proto *tp = NULL;
2393 unsigned long cl = 0;
2396 bool rtnl_held = false;
2398 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2399 rtm_tca_policy, extack);
2404 protocol = TC_H_MIN(t->tcm_info);
2405 prio = TC_H_MAJ(t->tcm_info);
2406 parent = t->tcm_parent;
2408 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2409 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2413 /* Find head of filter chain. */
2415 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2419 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2420 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2424 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2425 * found), qdisc is not unlocked, classifier type is not specified,
2426 * classifier is not unlocked.
2429 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2430 !tcf_proto_is_unlocked(name)) {
2435 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2439 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2441 if (IS_ERR(block)) {
2442 err = PTR_ERR(block);
2446 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2447 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2448 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2452 chain = tcf_chain_get(block, chain_index, false);
2454 /* User requested flush on non-existent chain. Nothing to do,
2455 * so just return success.
2461 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2467 tfilter_notify_chain(net, skb, block, q, parent, n,
2468 chain, RTM_DELTFILTER, extack);
2469 tcf_chain_flush(chain, rtnl_held);
2474 mutex_lock(&chain->filter_chain_lock);
2475 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2477 if (!tp || IS_ERR(tp)) {
2478 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2479 err = tp ? PTR_ERR(tp) : -ENOENT;
2481 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2482 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2485 } else if (t->tcm_handle == 0) {
2486 tcf_proto_signal_destroying(chain, tp);
2487 tcf_chain_tp_remove(chain, &chain_info, tp);
2488 mutex_unlock(&chain->filter_chain_lock);
2490 tcf_proto_put(tp, rtnl_held, NULL);
2491 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2492 RTM_DELTFILTER, false, rtnl_held, extack);
2496 mutex_unlock(&chain->filter_chain_lock);
2498 fh = tp->ops->get(tp, t->tcm_handle);
2501 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2506 err = tfilter_del_notify(net, skb, n, tp, block, q, parent, fh,
2507 &last, rtnl_held, extack);
2512 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2517 if (tp && !IS_ERR(tp))
2518 tcf_proto_put(tp, rtnl_held, NULL);
2519 tcf_chain_put(chain);
2521 tcf_block_release(q, block, rtnl_held);
2529 mutex_unlock(&chain->filter_chain_lock);
2533 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2534 struct netlink_ext_ack *extack)
2536 struct net *net = sock_net(skb->sk);
2537 struct nlattr *tca[TCA_MAX + 1];
2538 char name[IFNAMSIZ];
2544 struct Qdisc *q = NULL;
2545 struct tcf_chain_info chain_info;
2546 struct tcf_chain *chain = NULL;
2547 struct tcf_block *block = NULL;
2548 struct tcf_proto *tp = NULL;
2549 unsigned long cl = 0;
2552 bool rtnl_held = false;
2554 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2555 rtm_tca_policy, extack);
2560 protocol = TC_H_MIN(t->tcm_info);
2561 prio = TC_H_MAJ(t->tcm_info);
2562 parent = t->tcm_parent;
2565 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2569 /* Find head of filter chain. */
2571 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2575 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2576 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2580 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2581 * unlocked, classifier type is not specified, classifier is not
2584 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2585 !tcf_proto_is_unlocked(name)) {
2590 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2594 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2596 if (IS_ERR(block)) {
2597 err = PTR_ERR(block);
2601 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2602 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2603 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2607 chain = tcf_chain_get(block, chain_index, false);
2609 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2614 mutex_lock(&chain->filter_chain_lock);
2615 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2617 mutex_unlock(&chain->filter_chain_lock);
2618 if (!tp || IS_ERR(tp)) {
2619 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2620 err = tp ? PTR_ERR(tp) : -ENOENT;
2622 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2623 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2628 fh = tp->ops->get(tp, t->tcm_handle);
2631 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2634 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2635 fh, RTM_NEWTFILTER, true, rtnl_held, NULL);
2637 NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2640 tfilter_put(tp, fh);
2643 if (tp && !IS_ERR(tp))
2644 tcf_proto_put(tp, rtnl_held, NULL);
2645 tcf_chain_put(chain);
2647 tcf_block_release(q, block, rtnl_held);
2655 struct tcf_dump_args {
2656 struct tcf_walker w;
2657 struct sk_buff *skb;
2658 struct netlink_callback *cb;
2659 struct tcf_block *block;
2665 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2667 struct tcf_dump_args *a = (void *)arg;
2668 struct net *net = sock_net(a->skb->sk);
2670 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2671 n, NETLINK_CB(a->cb->skb).portid,
2672 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2673 RTM_NEWTFILTER, a->terse_dump, true, NULL);
2676 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2677 struct sk_buff *skb, struct netlink_callback *cb,
2678 long index_start, long *p_index, bool terse)
2680 struct net *net = sock_net(skb->sk);
2681 struct tcf_block *block = chain->block;
2682 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2683 struct tcf_proto *tp, *tp_prev;
2684 struct tcf_dump_args arg;
2686 for (tp = __tcf_get_next_proto(chain, NULL);
2689 tp = __tcf_get_next_proto(chain, tp),
2690 tcf_proto_put(tp_prev, true, NULL),
2692 if (*p_index < index_start)
2694 if (TC_H_MAJ(tcm->tcm_info) &&
2695 TC_H_MAJ(tcm->tcm_info) != tp->prio)
2697 if (TC_H_MIN(tcm->tcm_info) &&
2698 TC_H_MIN(tcm->tcm_info) != tp->protocol)
2700 if (*p_index > index_start)
2701 memset(&cb->args[1], 0,
2702 sizeof(cb->args) - sizeof(cb->args[0]));
2703 if (cb->args[1] == 0) {
2704 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2705 NETLINK_CB(cb->skb).portid,
2706 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2707 RTM_NEWTFILTER, false, true, NULL) <= 0)
2713 arg.w.fn = tcf_node_dump;
2718 arg.parent = parent;
2720 arg.w.skip = cb->args[1] - 1;
2722 arg.w.cookie = cb->args[2];
2723 arg.terse_dump = terse;
2724 tp->ops->walk(tp, &arg.w, true);
2725 cb->args[2] = arg.w.cookie;
2726 cb->args[1] = arg.w.count + 1;
2733 tcf_proto_put(tp, true, NULL);
2737 static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2738 [TCA_CHAIN] = { .type = NLA_U32 },
2739 [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2742 /* called with RTNL */
2743 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2745 struct tcf_chain *chain, *chain_prev;
2746 struct net *net = sock_net(skb->sk);
2747 struct nlattr *tca[TCA_MAX + 1];
2748 struct Qdisc *q = NULL;
2749 struct tcf_block *block;
2750 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2751 bool terse_dump = false;
2757 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2760 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2761 tcf_tfilter_dump_policy, cb->extack);
2765 if (tca[TCA_DUMP_FLAGS]) {
2766 struct nla_bitfield32 flags =
2767 nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2769 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2772 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2773 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2776 /* If we work with block index, q is NULL and parent value
2777 * will never be used in the following code. The check
2778 * in tcf_fill_node prevents it. However, compiler does not
2779 * see that far, so set parent to zero to silence the warning
2780 * about parent being uninitialized.
2784 const struct Qdisc_class_ops *cops;
2785 struct net_device *dev;
2786 unsigned long cl = 0;
2788 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2792 parent = tcm->tcm_parent;
2794 q = rtnl_dereference(dev->qdisc);
2796 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2799 cops = q->ops->cl_ops;
2802 if (!cops->tcf_block)
2804 if (TC_H_MIN(tcm->tcm_parent)) {
2805 cl = cops->find(q, tcm->tcm_parent);
2809 block = cops->tcf_block(q, cl, NULL);
2812 parent = block->classid;
2813 if (tcf_block_shared(block))
2817 index_start = cb->args[0];
2820 for (chain = __tcf_get_next_chain(block, NULL);
2823 chain = __tcf_get_next_chain(block, chain),
2824 tcf_chain_put(chain_prev)) {
2825 if (tca[TCA_CHAIN] &&
2826 nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2828 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2829 index_start, &index, terse_dump)) {
2830 tcf_chain_put(chain);
2836 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2837 tcf_block_refcnt_put(block, true);
2838 cb->args[0] = index;
2841 /* If we did no progress, the error (EMSGSIZE) is real */
2842 if (skb->len == 0 && err)
2847 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2848 void *tmplt_priv, u32 chain_index,
2849 struct net *net, struct sk_buff *skb,
2850 struct tcf_block *block,
2851 u32 portid, u32 seq, u16 flags, int event,
2852 struct netlink_ext_ack *extack)
2854 unsigned char *b = skb_tail_pointer(skb);
2855 const struct tcf_proto_ops *ops;
2856 struct nlmsghdr *nlh;
2863 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2865 goto out_nlmsg_trim;
2866 tcm = nlmsg_data(nlh);
2867 tcm->tcm_family = AF_UNSPEC;
2870 tcm->tcm_handle = 0;
2872 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2873 tcm->tcm_parent = block->q->handle;
2875 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2876 tcm->tcm_block_index = block->index;
2879 if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2880 goto nla_put_failure;
2883 if (nla_put_string(skb, TCA_KIND, ops->kind))
2884 goto nla_put_failure;
2885 if (ops->tmplt_dump(skb, net, priv) < 0)
2886 goto nla_put_failure;
2889 if (extack && extack->_msg &&
2890 nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2891 goto out_nlmsg_trim;
2893 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2903 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2904 u32 seq, u16 flags, int event, bool unicast,
2905 struct netlink_ext_ack *extack)
2907 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2908 struct tcf_block *block = chain->block;
2909 struct net *net = block->net;
2910 struct sk_buff *skb;
2913 if (!unicast && !rtnl_notify_needed(net, flags, RTNLGRP_TC))
2916 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2920 if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2921 chain->index, net, skb, block, portid,
2922 seq, flags, event, extack) <= 0) {
2928 err = rtnl_unicast(skb, net, portid);
2930 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2931 flags & NLM_F_ECHO);
2936 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2937 void *tmplt_priv, u32 chain_index,
2938 struct tcf_block *block, struct sk_buff *oskb,
2941 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2942 struct net *net = block->net;
2943 struct sk_buff *skb;
2945 if (!rtnl_notify_needed(net, flags, RTNLGRP_TC))
2948 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2952 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2953 block, portid, seq, flags, RTM_DELCHAIN, NULL) <= 0) {
2958 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2961 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2962 struct nlattr **tca,
2963 struct netlink_ext_ack *extack)
2965 const struct tcf_proto_ops *ops;
2966 char name[IFNAMSIZ];
2969 /* If kind is not set, user did not specify template. */
2973 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2974 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
2978 ops = tcf_proto_lookup_ops(name, true, extack);
2980 return PTR_ERR(ops);
2981 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2982 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2983 module_put(ops->owner);
2987 tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2988 if (IS_ERR(tmplt_priv)) {
2989 module_put(ops->owner);
2990 return PTR_ERR(tmplt_priv);
2992 chain->tmplt_ops = ops;
2993 chain->tmplt_priv = tmplt_priv;
2997 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
3000 /* If template ops are set, no work to do for us. */
3004 tmplt_ops->tmplt_destroy(tmplt_priv);
3005 module_put(tmplt_ops->owner);
3008 /* Add/delete/get a chain */
3010 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
3011 struct netlink_ext_ack *extack)
3013 struct net *net = sock_net(skb->sk);
3014 struct nlattr *tca[TCA_MAX + 1];
3019 struct tcf_chain *chain;
3020 struct tcf_block *block;
3026 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
3027 rtm_tca_policy, extack);
3032 parent = t->tcm_parent;
3035 block = tcf_block_find(net, &q, &parent, &cl,
3036 t->tcm_ifindex, t->tcm_block_index, extack);
3038 return PTR_ERR(block);
3040 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
3041 if (chain_index > TC_ACT_EXT_VAL_MASK) {
3042 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
3047 mutex_lock(&block->lock);
3048 chain = tcf_chain_lookup(block, chain_index);
3049 if (n->nlmsg_type == RTM_NEWCHAIN) {
3051 if (tcf_chain_held_by_acts_only(chain)) {
3052 /* The chain exists only because there is
3053 * some action referencing it.
3055 tcf_chain_hold(chain);
3057 NL_SET_ERR_MSG(extack, "Filter chain already exists");
3059 goto errout_block_locked;
3062 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
3063 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
3065 goto errout_block_locked;
3067 chain = tcf_chain_create(block, chain_index);
3069 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
3071 goto errout_block_locked;
3075 if (!chain || tcf_chain_held_by_acts_only(chain)) {
3076 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
3078 goto errout_block_locked;
3080 tcf_chain_hold(chain);
3083 if (n->nlmsg_type == RTM_NEWCHAIN) {
3084 /* Modifying chain requires holding parent block lock. In case
3085 * the chain was successfully added, take a reference to the
3086 * chain. This ensures that an empty chain does not disappear at
3087 * the end of this function.
3089 tcf_chain_hold(chain);
3090 chain->explicitly_created = true;
3092 mutex_unlock(&block->lock);
3094 switch (n->nlmsg_type) {
3096 err = tc_chain_tmplt_add(chain, net, tca, extack);
3098 tcf_chain_put_explicitly_created(chain);
3102 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
3103 RTM_NEWCHAIN, false, extack);
3106 tfilter_notify_chain(net, skb, block, q, parent, n,
3107 chain, RTM_DELTFILTER, extack);
3108 /* Flush the chain first as the user requested chain removal. */
3109 tcf_chain_flush(chain, true);
3110 /* In case the chain was successfully deleted, put a reference
3111 * to the chain previously taken during addition.
3113 tcf_chain_put_explicitly_created(chain);
3116 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
3117 n->nlmsg_flags, n->nlmsg_type, true, extack);
3119 NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
3123 NL_SET_ERR_MSG(extack, "Unsupported message type");
3128 tcf_chain_put(chain);
3130 tcf_block_release(q, block, true);
3132 /* Replay the request. */
3136 errout_block_locked:
3137 mutex_unlock(&block->lock);
3141 /* called with RTNL */
3142 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
3144 struct net *net = sock_net(skb->sk);
3145 struct nlattr *tca[TCA_MAX + 1];
3146 struct Qdisc *q = NULL;
3147 struct tcf_block *block;
3148 struct tcmsg *tcm = nlmsg_data(cb->nlh);
3149 struct tcf_chain *chain;
3154 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
3157 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
3158 rtm_tca_policy, cb->extack);
3162 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
3163 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
3167 const struct Qdisc_class_ops *cops;
3168 struct net_device *dev;
3169 unsigned long cl = 0;
3171 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
3175 if (!tcm->tcm_parent)
3176 q = rtnl_dereference(dev->qdisc);
3178 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
3182 cops = q->ops->cl_ops;
3185 if (!cops->tcf_block)
3187 if (TC_H_MIN(tcm->tcm_parent)) {
3188 cl = cops->find(q, tcm->tcm_parent);
3192 block = cops->tcf_block(q, cl, NULL);
3195 if (tcf_block_shared(block))
3199 index_start = cb->args[0];
3202 mutex_lock(&block->lock);
3203 list_for_each_entry(chain, &block->chain_list, list) {
3204 if ((tca[TCA_CHAIN] &&
3205 nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3207 if (index < index_start) {
3211 if (tcf_chain_held_by_acts_only(chain))
3213 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3214 chain->index, net, skb, block,
3215 NETLINK_CB(cb->skb).portid,
3216 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3217 RTM_NEWCHAIN, NULL);
3222 mutex_unlock(&block->lock);
3224 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3225 tcf_block_refcnt_put(block, true);
3226 cb->args[0] = index;
3229 /* If we did no progress, the error (EMSGSIZE) is real */
3230 if (skb->len == 0 && err)
3235 int tcf_exts_init_ex(struct tcf_exts *exts, struct net *net, int action,
3236 int police, struct tcf_proto *tp, u32 handle,
3237 bool use_action_miss)
3241 #ifdef CONFIG_NET_CLS_ACT
3243 exts->nr_actions = 0;
3244 exts->miss_cookie_node = NULL;
3245 /* Note: we do not own yet a reference on net.
3246 * This reference might be taken later from tcf_exts_get_net().
3249 exts->actions = kcalloc(TCA_ACT_MAX_PRIO, sizeof(struct tc_action *),
3255 exts->action = action;
3256 exts->police = police;
3258 if (!use_action_miss)
3261 err = tcf_exts_miss_cookie_base_alloc(exts, tp, handle);
3263 goto err_miss_alloc;
3268 tcf_exts_destroy(exts);
3269 #ifdef CONFIG_NET_CLS_ACT
3270 exts->actions = NULL;
3274 EXPORT_SYMBOL(tcf_exts_init_ex);
3276 void tcf_exts_destroy(struct tcf_exts *exts)
3278 tcf_exts_miss_cookie_base_destroy(exts);
3280 #ifdef CONFIG_NET_CLS_ACT
3281 if (exts->actions) {
3282 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3283 kfree(exts->actions);
3285 exts->nr_actions = 0;
3288 EXPORT_SYMBOL(tcf_exts_destroy);
3290 int tcf_exts_validate_ex(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3291 struct nlattr *rate_tlv, struct tcf_exts *exts,
3292 u32 flags, u32 fl_flags, struct netlink_ext_ack *extack)
3294 #ifdef CONFIG_NET_CLS_ACT
3296 int init_res[TCA_ACT_MAX_PRIO] = {};
3297 struct tc_action *act;
3298 size_t attr_size = 0;
3300 if (exts->police && tb[exts->police]) {
3301 struct tc_action_ops *a_o;
3303 a_o = tc_action_load_ops(tb[exts->police], true,
3304 !(flags & TCA_ACT_FLAGS_NO_RTNL),
3307 return PTR_ERR(a_o);
3308 flags |= TCA_ACT_FLAGS_POLICE | TCA_ACT_FLAGS_BIND;
3309 act = tcf_action_init_1(net, tp, tb[exts->police],
3310 rate_tlv, a_o, init_res, flags,
3312 module_put(a_o->owner);
3314 return PTR_ERR(act);
3316 act->type = exts->type = TCA_OLD_COMPAT;
3317 exts->actions[0] = act;
3318 exts->nr_actions = 1;
3319 tcf_idr_insert_many(exts->actions, init_res);
3320 } else if (exts->action && tb[exts->action]) {
3323 flags |= TCA_ACT_FLAGS_BIND;
3324 err = tcf_action_init(net, tp, tb[exts->action],
3325 rate_tlv, exts->actions, init_res,
3326 &attr_size, flags, fl_flags,
3330 exts->nr_actions = err;
3334 if ((exts->action && tb[exts->action]) ||
3335 (exts->police && tb[exts->police])) {
3336 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3343 EXPORT_SYMBOL(tcf_exts_validate_ex);
3345 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3346 struct nlattr *rate_tlv, struct tcf_exts *exts,
3347 u32 flags, struct netlink_ext_ack *extack)
3349 return tcf_exts_validate_ex(net, tp, tb, rate_tlv, exts,
3352 EXPORT_SYMBOL(tcf_exts_validate);
3354 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3356 #ifdef CONFIG_NET_CLS_ACT
3357 struct tcf_exts old = *dst;
3360 tcf_exts_destroy(&old);
3363 EXPORT_SYMBOL(tcf_exts_change);
3365 #ifdef CONFIG_NET_CLS_ACT
3366 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3368 if (exts->nr_actions == 0)
3371 return exts->actions[0];
3375 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3377 #ifdef CONFIG_NET_CLS_ACT
3378 struct nlattr *nest;
3380 if (exts->action && tcf_exts_has_actions(exts)) {
3382 * again for backward compatible mode - we want
3383 * to work with both old and new modes of entering
3384 * tc data even if iproute2 was newer - jhs
3386 if (exts->type != TCA_OLD_COMPAT) {
3387 nest = nla_nest_start_noflag(skb, exts->action);
3389 goto nla_put_failure;
3391 if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3393 goto nla_put_failure;
3394 nla_nest_end(skb, nest);
3395 } else if (exts->police) {
3396 struct tc_action *act = tcf_exts_first_act(exts);
3397 nest = nla_nest_start_noflag(skb, exts->police);
3398 if (nest == NULL || !act)
3399 goto nla_put_failure;
3400 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3401 goto nla_put_failure;
3402 nla_nest_end(skb, nest);
3408 nla_nest_cancel(skb, nest);
3414 EXPORT_SYMBOL(tcf_exts_dump);
3416 int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3418 #ifdef CONFIG_NET_CLS_ACT
3419 struct nlattr *nest;
3421 if (!exts->action || !tcf_exts_has_actions(exts))
3424 nest = nla_nest_start_noflag(skb, exts->action);
3426 goto nla_put_failure;
3428 if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3429 goto nla_put_failure;
3430 nla_nest_end(skb, nest);
3434 nla_nest_cancel(skb, nest);
3440 EXPORT_SYMBOL(tcf_exts_terse_dump);
3442 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3444 #ifdef CONFIG_NET_CLS_ACT
3445 struct tc_action *a = tcf_exts_first_act(exts);
3446 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3451 EXPORT_SYMBOL(tcf_exts_dump_stats);
3453 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3455 if (*flags & TCA_CLS_FLAGS_IN_HW)
3457 *flags |= TCA_CLS_FLAGS_IN_HW;
3458 atomic_inc(&block->offloadcnt);
3461 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3463 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3465 *flags &= ~TCA_CLS_FLAGS_IN_HW;
3466 atomic_dec(&block->offloadcnt);
3469 static void tc_cls_offload_cnt_update(struct tcf_block *block,
3470 struct tcf_proto *tp, u32 *cnt,
3471 u32 *flags, u32 diff, bool add)
3473 lockdep_assert_held(&block->cb_lock);
3475 spin_lock(&tp->lock);
3478 tcf_block_offload_inc(block, flags);
3483 tcf_block_offload_dec(block, flags);
3485 spin_unlock(&tp->lock);
3489 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3490 u32 *cnt, u32 *flags)
3492 lockdep_assert_held(&block->cb_lock);
3494 spin_lock(&tp->lock);
3495 tcf_block_offload_dec(block, flags);
3497 spin_unlock(&tp->lock);
3501 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3502 void *type_data, bool err_stop)
3504 struct flow_block_cb *block_cb;
3508 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3509 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3520 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3521 void *type_data, bool err_stop, bool rtnl_held)
3523 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3529 down_read(&block->cb_lock);
3530 /* Need to obtain rtnl lock if block is bound to devs that require it.
3531 * In block bind code cb_lock is obtained while holding rtnl, so we must
3532 * obtain the locks in same order here.
3534 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3535 up_read(&block->cb_lock);
3540 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3542 up_read(&block->cb_lock);
3547 EXPORT_SYMBOL(tc_setup_cb_call);
3549 /* Non-destructive filter add. If filter that wasn't already in hardware is
3550 * successfully offloaded, increment block offloads counter. On failure,
3551 * previously offloaded filter is considered to be intact and offloads counter
3552 * is not decremented.
3555 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3556 enum tc_setup_type type, void *type_data, bool err_stop,
3557 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3559 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3565 down_read(&block->cb_lock);
3566 /* Need to obtain rtnl lock if block is bound to devs that require it.
3567 * In block bind code cb_lock is obtained while holding rtnl, so we must
3568 * obtain the locks in same order here.
3570 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3571 up_read(&block->cb_lock);
3576 /* Make sure all netdevs sharing this block are offload-capable. */
3577 if (block->nooffloaddevcnt && err_stop) {
3578 ok_count = -EOPNOTSUPP;
3582 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3586 if (tp->ops->hw_add)
3587 tp->ops->hw_add(tp, type_data);
3589 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3592 up_read(&block->cb_lock);
3595 return min(ok_count, 0);
3597 EXPORT_SYMBOL(tc_setup_cb_add);
3599 /* Destructive filter replace. If filter that wasn't already in hardware is
3600 * successfully offloaded, increment block offload counter. On failure,
3601 * previously offloaded filter is considered to be destroyed and offload counter
3605 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3606 enum tc_setup_type type, void *type_data, bool err_stop,
3607 u32 *old_flags, unsigned int *old_in_hw_count,
3608 u32 *new_flags, unsigned int *new_in_hw_count,
3611 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3617 down_read(&block->cb_lock);
3618 /* Need to obtain rtnl lock if block is bound to devs that require it.
3619 * In block bind code cb_lock is obtained while holding rtnl, so we must
3620 * obtain the locks in same order here.
3622 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3623 up_read(&block->cb_lock);
3628 /* Make sure all netdevs sharing this block are offload-capable. */
3629 if (block->nooffloaddevcnt && err_stop) {
3630 ok_count = -EOPNOTSUPP;
3634 tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3635 if (tp->ops->hw_del)
3636 tp->ops->hw_del(tp, type_data);
3638 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3642 if (tp->ops->hw_add)
3643 tp->ops->hw_add(tp, type_data);
3645 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3646 new_flags, ok_count, true);
3648 up_read(&block->cb_lock);
3651 return min(ok_count, 0);
3653 EXPORT_SYMBOL(tc_setup_cb_replace);
3655 /* Destroy filter and decrement block offload counter, if filter was previously
3659 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3660 enum tc_setup_type type, void *type_data, bool err_stop,
3661 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3663 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3669 down_read(&block->cb_lock);
3670 /* Need to obtain rtnl lock if block is bound to devs that require it.
3671 * In block bind code cb_lock is obtained while holding rtnl, so we must
3672 * obtain the locks in same order here.
3674 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3675 up_read(&block->cb_lock);
3680 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3682 tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3683 if (tp->ops->hw_del)
3684 tp->ops->hw_del(tp, type_data);
3686 up_read(&block->cb_lock);
3689 return min(ok_count, 0);
3691 EXPORT_SYMBOL(tc_setup_cb_destroy);
3693 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3694 bool add, flow_setup_cb_t *cb,
3695 enum tc_setup_type type, void *type_data,
3696 void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3698 int err = cb(type, type_data, cb_priv);
3701 if (add && tc_skip_sw(*flags))
3704 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3710 EXPORT_SYMBOL(tc_setup_cb_reoffload);
3712 static int tcf_act_get_user_cookie(struct flow_action_entry *entry,
3713 const struct tc_action *act)
3715 struct tc_cookie *user_cookie;
3719 user_cookie = rcu_dereference(act->user_cookie);
3721 entry->user_cookie = flow_action_cookie_create(user_cookie->data,
3724 if (!entry->user_cookie)
3731 static void tcf_act_put_user_cookie(struct flow_action_entry *entry)
3733 flow_action_cookie_destroy(entry->user_cookie);
3736 void tc_cleanup_offload_action(struct flow_action *flow_action)
3738 struct flow_action_entry *entry;
3741 flow_action_for_each(i, entry, flow_action) {
3742 tcf_act_put_user_cookie(entry);
3743 if (entry->destructor)
3744 entry->destructor(entry->destructor_priv);
3747 EXPORT_SYMBOL(tc_cleanup_offload_action);
3749 static int tc_setup_offload_act(struct tc_action *act,
3750 struct flow_action_entry *entry,
3752 struct netlink_ext_ack *extack)
3754 #ifdef CONFIG_NET_CLS_ACT
3755 if (act->ops->offload_act_setup) {
3756 return act->ops->offload_act_setup(act, entry, index_inc, true,
3759 NL_SET_ERR_MSG(extack, "Action does not support offload");
3767 int tc_setup_action(struct flow_action *flow_action,
3768 struct tc_action *actions[],
3769 u32 miss_cookie_base,
3770 struct netlink_ext_ack *extack)
3772 int i, j, k, index, err = 0;
3773 struct tc_action *act;
3775 BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3776 BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3777 BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3783 tcf_act_for_each_action(i, act, actions) {
3784 struct flow_action_entry *entry;
3786 entry = &flow_action->entries[j];
3787 spin_lock_bh(&act->tcfa_lock);
3788 err = tcf_act_get_user_cookie(entry, act);
3790 goto err_out_locked;
3793 err = tc_setup_offload_act(act, entry, &index, extack);
3795 goto err_out_locked;
3797 for (k = 0; k < index ; k++) {
3798 entry[k].hw_stats = tc_act_hw_stats(act->hw_stats);
3799 entry[k].hw_index = act->tcfa_index;
3800 entry[k].cookie = (unsigned long)act;
3801 entry[k].miss_cookie =
3802 tcf_exts_miss_cookie_get(miss_cookie_base, i);
3807 spin_unlock_bh(&act->tcfa_lock);
3812 tc_cleanup_offload_action(flow_action);
3816 spin_unlock_bh(&act->tcfa_lock);
3820 int tc_setup_offload_action(struct flow_action *flow_action,
3821 const struct tcf_exts *exts,
3822 struct netlink_ext_ack *extack)
3824 #ifdef CONFIG_NET_CLS_ACT
3825 u32 miss_cookie_base;
3830 miss_cookie_base = exts->miss_cookie_node ?
3831 exts->miss_cookie_node->miss_cookie_base : 0;
3832 return tc_setup_action(flow_action, exts->actions, miss_cookie_base,
3838 EXPORT_SYMBOL(tc_setup_offload_action);
3840 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3842 unsigned int num_acts = 0;
3843 struct tc_action *act;
3846 tcf_exts_for_each_action(i, act, exts) {
3847 if (is_tcf_pedit(act))
3848 num_acts += tcf_pedit_nkeys(act);
3854 EXPORT_SYMBOL(tcf_exts_num_actions);
3856 #ifdef CONFIG_NET_CLS_ACT
3857 static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3859 struct netlink_ext_ack *extack)
3861 *p_block_index = nla_get_u32(block_index_attr);
3862 if (!*p_block_index) {
3863 NL_SET_ERR_MSG(extack, "Block number may not be zero");
3870 int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3871 enum flow_block_binder_type binder_type,
3872 struct nlattr *block_index_attr,
3873 struct netlink_ext_ack *extack)
3878 if (!block_index_attr)
3881 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3885 qe->info.binder_type = binder_type;
3886 qe->info.chain_head_change = tcf_chain_head_change_dflt;
3887 qe->info.chain_head_change_priv = &qe->filter_chain;
3888 qe->info.block_index = block_index;
3890 return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3892 EXPORT_SYMBOL(tcf_qevent_init);
3894 void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3896 if (qe->info.block_index)
3897 tcf_block_put_ext(qe->block, sch, &qe->info);
3899 EXPORT_SYMBOL(tcf_qevent_destroy);
3901 int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3902 struct netlink_ext_ack *extack)
3907 if (!block_index_attr)
3910 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3914 /* Bounce newly-configured block or change in block. */
3915 if (block_index != qe->info.block_index) {
3916 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3922 EXPORT_SYMBOL(tcf_qevent_validate_change);
3924 struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
3925 struct sk_buff **to_free, int *ret)
3927 struct tcf_result cl_res;
3928 struct tcf_proto *fl;
3930 if (!qe->info.block_index)
3933 fl = rcu_dereference_bh(qe->filter_chain);
3935 switch (tcf_classify(skb, NULL, fl, &cl_res, false)) {
3937 qdisc_qstats_drop(sch);
3938 __qdisc_drop(skb, to_free);
3939 *ret = __NET_XMIT_BYPASS;
3944 __qdisc_drop(skb, to_free);
3945 *ret = __NET_XMIT_STOLEN;
3947 case TC_ACT_REDIRECT:
3948 skb_do_redirect(skb);
3949 *ret = __NET_XMIT_STOLEN;
3955 EXPORT_SYMBOL(tcf_qevent_handle);
3957 int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
3959 if (!qe->info.block_index)
3961 return nla_put_u32(skb, attr_name, qe->info.block_index);
3963 EXPORT_SYMBOL(tcf_qevent_dump);
3966 static __net_init int tcf_net_init(struct net *net)
3968 struct tcf_net *tn = net_generic(net, tcf_net_id);
3970 spin_lock_init(&tn->idr_lock);
3975 static void __net_exit tcf_net_exit(struct net *net)
3977 struct tcf_net *tn = net_generic(net, tcf_net_id);
3979 idr_destroy(&tn->idr);
3982 static struct pernet_operations tcf_net_ops = {
3983 .init = tcf_net_init,
3984 .exit = tcf_net_exit,
3986 .size = sizeof(struct tcf_net),
3989 static int __init tc_filter_init(void)
3993 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3997 err = register_pernet_subsys(&tcf_net_ops);
3999 goto err_register_pernet_subsys;
4001 xa_init_flags(&tcf_exts_miss_cookies_xa, XA_FLAGS_ALLOC1);
4003 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
4004 RTNL_FLAG_DOIT_UNLOCKED);
4005 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
4006 RTNL_FLAG_DOIT_UNLOCKED);
4007 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
4008 tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
4009 rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
4010 rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
4011 rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
4016 err_register_pernet_subsys:
4017 destroy_workqueue(tc_filter_wq);
4021 subsys_initcall(tc_filter_init);