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(NET_CLS_ALIAS_PREFIX
"%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 tp
->usesw
= !tp
->ops
->reoffload
;
394 spin_lock_init(&tp
->lock
);
395 refcount_set(&tp
->refcnt
, 1);
397 err
= tp
->ops
->init(tp
);
399 module_put(tp
->ops
->owner
);
409 static void tcf_proto_get(struct tcf_proto
*tp
)
411 refcount_inc(&tp
->refcnt
);
414 static void tcf_proto_count_usesw(struct tcf_proto
*tp
, bool add
)
416 #ifdef CONFIG_NET_CLS_ACT
417 struct tcf_block
*block
= tp
->chain
->block
;
418 bool counted
= false;
421 if (tp
->usesw
&& tp
->counted
) {
422 if (!atomic_dec_return(&block
->useswcnt
))
423 static_branch_dec(&tcf_sw_enabled_key
);
429 spin_lock(&tp
->lock
);
430 if (tp
->usesw
&& !tp
->counted
) {
434 spin_unlock(&tp
->lock
);
436 if (counted
&& atomic_inc_return(&block
->useswcnt
) == 1)
437 static_branch_inc(&tcf_sw_enabled_key
);
441 static void tcf_chain_put(struct tcf_chain
*chain
);
443 static void tcf_proto_destroy(struct tcf_proto
*tp
, bool rtnl_held
,
444 bool sig_destroy
, struct netlink_ext_ack
*extack
)
446 tp
->ops
->destroy(tp
, rtnl_held
, extack
);
447 tcf_proto_count_usesw(tp
, false);
449 tcf_proto_signal_destroyed(tp
->chain
, tp
);
450 tcf_chain_put(tp
->chain
);
451 module_put(tp
->ops
->owner
);
455 static void tcf_proto_put(struct tcf_proto
*tp
, bool rtnl_held
,
456 struct netlink_ext_ack
*extack
)
458 if (refcount_dec_and_test(&tp
->refcnt
))
459 tcf_proto_destroy(tp
, rtnl_held
, true, extack
);
462 static bool tcf_proto_check_delete(struct tcf_proto
*tp
)
464 if (tp
->ops
->delete_empty
)
465 return tp
->ops
->delete_empty(tp
);
471 static void tcf_proto_mark_delete(struct tcf_proto
*tp
)
473 spin_lock(&tp
->lock
);
475 spin_unlock(&tp
->lock
);
478 static bool tcf_proto_is_deleting(struct tcf_proto
*tp
)
482 spin_lock(&tp
->lock
);
483 deleting
= tp
->deleting
;
484 spin_unlock(&tp
->lock
);
489 #define ASSERT_BLOCK_LOCKED(block) \
490 lockdep_assert_held(&(block)->lock)
492 struct tcf_filter_chain_list_item
{
493 struct list_head list
;
494 tcf_chain_head_change_t
*chain_head_change
;
495 void *chain_head_change_priv
;
498 static struct tcf_chain
*tcf_chain_create(struct tcf_block
*block
,
501 struct tcf_chain
*chain
;
503 ASSERT_BLOCK_LOCKED(block
);
505 chain
= kzalloc(sizeof(*chain
), GFP_KERNEL
);
508 list_add_tail_rcu(&chain
->list
, &block
->chain_list
);
509 mutex_init(&chain
->filter_chain_lock
);
510 chain
->block
= block
;
511 chain
->index
= chain_index
;
514 block
->chain0
.chain
= chain
;
518 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item
*item
,
519 struct tcf_proto
*tp_head
)
521 if (item
->chain_head_change
)
522 item
->chain_head_change(tp_head
, item
->chain_head_change_priv
);
525 static void tcf_chain0_head_change(struct tcf_chain
*chain
,
526 struct tcf_proto
*tp_head
)
528 struct tcf_filter_chain_list_item
*item
;
529 struct tcf_block
*block
= chain
->block
;
534 mutex_lock(&block
->lock
);
535 list_for_each_entry(item
, &block
->chain0
.filter_chain_list
, list
)
536 tcf_chain_head_change_item(item
, tp_head
);
537 mutex_unlock(&block
->lock
);
540 /* Returns true if block can be safely freed. */
542 static bool tcf_chain_detach(struct tcf_chain
*chain
)
544 struct tcf_block
*block
= chain
->block
;
546 ASSERT_BLOCK_LOCKED(block
);
548 list_del_rcu(&chain
->list
);
550 block
->chain0
.chain
= NULL
;
552 if (list_empty(&block
->chain_list
) &&
553 refcount_read(&block
->refcnt
) == 0)
559 static void tcf_block_destroy(struct tcf_block
*block
)
561 mutex_destroy(&block
->lock
);
562 mutex_destroy(&block
->proto_destroy_lock
);
563 xa_destroy(&block
->ports
);
564 kfree_rcu(block
, rcu
);
567 static void tcf_chain_destroy(struct tcf_chain
*chain
, bool free_block
)
569 struct tcf_block
*block
= chain
->block
;
571 mutex_destroy(&chain
->filter_chain_lock
);
572 kfree_rcu(chain
, rcu
);
574 tcf_block_destroy(block
);
577 static void tcf_chain_hold(struct tcf_chain
*chain
)
579 ASSERT_BLOCK_LOCKED(chain
->block
);
584 static bool tcf_chain_held_by_acts_only(struct tcf_chain
*chain
)
586 ASSERT_BLOCK_LOCKED(chain
->block
);
588 /* In case all the references are action references, this
589 * chain should not be shown to the user.
591 return chain
->refcnt
== chain
->action_refcnt
;
594 static struct tcf_chain
*tcf_chain_lookup(struct tcf_block
*block
,
597 struct tcf_chain
*chain
;
599 ASSERT_BLOCK_LOCKED(block
);
601 list_for_each_entry(chain
, &block
->chain_list
, list
) {
602 if (chain
->index
== chain_index
)
608 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
609 static struct tcf_chain
*tcf_chain_lookup_rcu(const struct tcf_block
*block
,
612 struct tcf_chain
*chain
;
614 list_for_each_entry_rcu(chain
, &block
->chain_list
, list
) {
615 if (chain
->index
== chain_index
)
622 static int tc_chain_notify(struct tcf_chain
*chain
, struct sk_buff
*oskb
,
623 u32 seq
, u16 flags
, int event
, bool unicast
,
624 struct netlink_ext_ack
*extack
);
626 static struct tcf_chain
*__tcf_chain_get(struct tcf_block
*block
,
627 u32 chain_index
, bool create
,
630 struct tcf_chain
*chain
= NULL
;
631 bool is_first_reference
;
633 mutex_lock(&block
->lock
);
634 chain
= tcf_chain_lookup(block
, chain_index
);
636 tcf_chain_hold(chain
);
640 chain
= tcf_chain_create(block
, chain_index
);
646 ++chain
->action_refcnt
;
647 is_first_reference
= chain
->refcnt
- chain
->action_refcnt
== 1;
648 mutex_unlock(&block
->lock
);
650 /* Send notification only in case we got the first
651 * non-action reference. Until then, the chain acts only as
652 * a placeholder for actions pointing to it and user ought
653 * not know about them.
655 if (is_first_reference
&& !by_act
)
656 tc_chain_notify(chain
, NULL
, 0, NLM_F_CREATE
| NLM_F_EXCL
,
657 RTM_NEWCHAIN
, false, NULL
);
662 mutex_unlock(&block
->lock
);
666 static struct tcf_chain
*tcf_chain_get(struct tcf_block
*block
, u32 chain_index
,
669 return __tcf_chain_get(block
, chain_index
, create
, false);
672 struct tcf_chain
*tcf_chain_get_by_act(struct tcf_block
*block
, u32 chain_index
)
674 return __tcf_chain_get(block
, chain_index
, true, true);
676 EXPORT_SYMBOL(tcf_chain_get_by_act
);
678 static void tc_chain_tmplt_del(const struct tcf_proto_ops
*tmplt_ops
,
680 static int tc_chain_notify_delete(const struct tcf_proto_ops
*tmplt_ops
,
681 void *tmplt_priv
, u32 chain_index
,
682 struct tcf_block
*block
, struct sk_buff
*oskb
,
685 static void __tcf_chain_put(struct tcf_chain
*chain
, bool by_act
,
686 bool explicitly_created
)
688 struct tcf_block
*block
= chain
->block
;
689 const struct tcf_proto_ops
*tmplt_ops
;
690 unsigned int refcnt
, non_act_refcnt
;
691 bool free_block
= false;
694 mutex_lock(&block
->lock
);
695 if (explicitly_created
) {
696 if (!chain
->explicitly_created
) {
697 mutex_unlock(&block
->lock
);
700 chain
->explicitly_created
= false;
704 chain
->action_refcnt
--;
706 /* tc_chain_notify_delete can't be called while holding block lock.
707 * However, when block is unlocked chain can be changed concurrently, so
708 * save these to temporary variables.
710 refcnt
= --chain
->refcnt
;
711 non_act_refcnt
= refcnt
- chain
->action_refcnt
;
712 tmplt_ops
= chain
->tmplt_ops
;
713 tmplt_priv
= chain
->tmplt_priv
;
715 if (non_act_refcnt
== chain
->explicitly_created
&& !by_act
) {
716 if (non_act_refcnt
== 0)
717 tc_chain_notify_delete(tmplt_ops
, tmplt_priv
,
718 chain
->index
, block
, NULL
, 0, 0);
719 /* Last reference to chain, no need to lock. */
720 chain
->flushing
= false;
724 free_block
= tcf_chain_detach(chain
);
725 mutex_unlock(&block
->lock
);
728 tc_chain_tmplt_del(tmplt_ops
, tmplt_priv
);
729 tcf_chain_destroy(chain
, free_block
);
733 static void tcf_chain_put(struct tcf_chain
*chain
)
735 __tcf_chain_put(chain
, false, false);
738 void tcf_chain_put_by_act(struct tcf_chain
*chain
)
740 __tcf_chain_put(chain
, true, false);
742 EXPORT_SYMBOL(tcf_chain_put_by_act
);
744 static void tcf_chain_put_explicitly_created(struct tcf_chain
*chain
)
746 __tcf_chain_put(chain
, false, true);
749 static void tcf_chain_flush(struct tcf_chain
*chain
, bool rtnl_held
)
751 struct tcf_proto
*tp
, *tp_next
;
753 mutex_lock(&chain
->filter_chain_lock
);
754 tp
= tcf_chain_dereference(chain
->filter_chain
, chain
);
756 tp_next
= rcu_dereference_protected(tp
->next
, 1);
757 tcf_proto_signal_destroying(chain
, tp
);
760 tp
= tcf_chain_dereference(chain
->filter_chain
, chain
);
761 RCU_INIT_POINTER(chain
->filter_chain
, NULL
);
762 tcf_chain0_head_change(chain
, NULL
);
763 chain
->flushing
= true;
764 mutex_unlock(&chain
->filter_chain_lock
);
767 tp_next
= rcu_dereference_protected(tp
->next
, 1);
768 tcf_proto_put(tp
, rtnl_held
, NULL
);
773 static int tcf_block_setup(struct tcf_block
*block
,
774 struct flow_block_offload
*bo
);
776 static void tcf_block_offload_init(struct flow_block_offload
*bo
,
777 struct net_device
*dev
, struct Qdisc
*sch
,
778 enum flow_block_command command
,
779 enum flow_block_binder_type binder_type
,
780 struct flow_block
*flow_block
,
781 bool shared
, struct netlink_ext_ack
*extack
)
783 bo
->net
= dev_net(dev
);
784 bo
->command
= command
;
785 bo
->binder_type
= binder_type
;
786 bo
->block
= flow_block
;
787 bo
->block_shared
= shared
;
790 bo
->cb_list_head
= &flow_block
->cb_list
;
791 INIT_LIST_HEAD(&bo
->cb_list
);
794 static void tcf_block_unbind(struct tcf_block
*block
,
795 struct flow_block_offload
*bo
);
797 static void tc_block_indr_cleanup(struct flow_block_cb
*block_cb
)
799 struct tcf_block
*block
= block_cb
->indr
.data
;
800 struct net_device
*dev
= block_cb
->indr
.dev
;
801 struct Qdisc
*sch
= block_cb
->indr
.sch
;
802 struct netlink_ext_ack extack
= {};
803 struct flow_block_offload bo
= {};
805 tcf_block_offload_init(&bo
, dev
, sch
, FLOW_BLOCK_UNBIND
,
806 block_cb
->indr
.binder_type
,
807 &block
->flow_block
, tcf_block_shared(block
),
810 down_write(&block
->cb_lock
);
811 list_del(&block_cb
->driver_list
);
812 list_move(&block_cb
->list
, &bo
.cb_list
);
813 tcf_block_unbind(block
, &bo
);
814 up_write(&block
->cb_lock
);
818 static bool tcf_block_offload_in_use(struct tcf_block
*block
)
820 return atomic_read(&block
->offloadcnt
);
823 static int tcf_block_offload_cmd(struct tcf_block
*block
,
824 struct net_device
*dev
, struct Qdisc
*sch
,
825 struct tcf_block_ext_info
*ei
,
826 enum flow_block_command command
,
827 struct netlink_ext_ack
*extack
)
829 struct flow_block_offload bo
= {};
831 tcf_block_offload_init(&bo
, dev
, sch
, command
, ei
->binder_type
,
832 &block
->flow_block
, tcf_block_shared(block
),
835 if (dev
->netdev_ops
->ndo_setup_tc
) {
838 err
= dev
->netdev_ops
->ndo_setup_tc(dev
, TC_SETUP_BLOCK
, &bo
);
840 if (err
!= -EOPNOTSUPP
)
841 NL_SET_ERR_MSG(extack
, "Driver ndo_setup_tc failed");
845 return tcf_block_setup(block
, &bo
);
848 flow_indr_dev_setup_offload(dev
, sch
, TC_SETUP_BLOCK
, block
, &bo
,
849 tc_block_indr_cleanup
);
850 tcf_block_setup(block
, &bo
);
855 static int tcf_block_offload_bind(struct tcf_block
*block
, struct Qdisc
*q
,
856 struct tcf_block_ext_info
*ei
,
857 struct netlink_ext_ack
*extack
)
859 struct net_device
*dev
= q
->dev_queue
->dev
;
862 down_write(&block
->cb_lock
);
864 /* If tc offload feature is disabled and the block we try to bind
865 * to already has some offloaded filters, forbid to bind.
867 if (dev
->netdev_ops
->ndo_setup_tc
&&
868 !tc_can_offload(dev
) &&
869 tcf_block_offload_in_use(block
)) {
870 NL_SET_ERR_MSG(extack
, "Bind to offloaded block failed as dev has offload disabled");
875 err
= tcf_block_offload_cmd(block
, dev
, q
, ei
, FLOW_BLOCK_BIND
, extack
);
876 if (err
== -EOPNOTSUPP
)
877 goto no_offload_dev_inc
;
881 up_write(&block
->cb_lock
);
885 if (tcf_block_offload_in_use(block
))
889 block
->nooffloaddevcnt
++;
891 up_write(&block
->cb_lock
);
895 static void tcf_block_offload_unbind(struct tcf_block
*block
, struct Qdisc
*q
,
896 struct tcf_block_ext_info
*ei
)
898 struct net_device
*dev
= q
->dev_queue
->dev
;
901 down_write(&block
->cb_lock
);
902 err
= tcf_block_offload_cmd(block
, dev
, q
, ei
, FLOW_BLOCK_UNBIND
, NULL
);
903 if (err
== -EOPNOTSUPP
)
904 goto no_offload_dev_dec
;
905 up_write(&block
->cb_lock
);
909 WARN_ON(block
->nooffloaddevcnt
-- == 0);
910 up_write(&block
->cb_lock
);
914 tcf_chain0_head_change_cb_add(struct tcf_block
*block
,
915 struct tcf_block_ext_info
*ei
,
916 struct netlink_ext_ack
*extack
)
918 struct tcf_filter_chain_list_item
*item
;
919 struct tcf_chain
*chain0
;
921 item
= kmalloc(sizeof(*item
), GFP_KERNEL
);
923 NL_SET_ERR_MSG(extack
, "Memory allocation for head change callback item failed");
926 item
->chain_head_change
= ei
->chain_head_change
;
927 item
->chain_head_change_priv
= ei
->chain_head_change_priv
;
929 mutex_lock(&block
->lock
);
930 chain0
= block
->chain0
.chain
;
932 tcf_chain_hold(chain0
);
934 list_add(&item
->list
, &block
->chain0
.filter_chain_list
);
935 mutex_unlock(&block
->lock
);
938 struct tcf_proto
*tp_head
;
940 mutex_lock(&chain0
->filter_chain_lock
);
942 tp_head
= tcf_chain_dereference(chain0
->filter_chain
, chain0
);
944 tcf_chain_head_change_item(item
, tp_head
);
946 mutex_lock(&block
->lock
);
947 list_add(&item
->list
, &block
->chain0
.filter_chain_list
);
948 mutex_unlock(&block
->lock
);
950 mutex_unlock(&chain0
->filter_chain_lock
);
951 tcf_chain_put(chain0
);
958 tcf_chain0_head_change_cb_del(struct tcf_block
*block
,
959 struct tcf_block_ext_info
*ei
)
961 struct tcf_filter_chain_list_item
*item
;
963 mutex_lock(&block
->lock
);
964 list_for_each_entry(item
, &block
->chain0
.filter_chain_list
, list
) {
965 if ((!ei
->chain_head_change
&& !ei
->chain_head_change_priv
) ||
966 (item
->chain_head_change
== ei
->chain_head_change
&&
967 item
->chain_head_change_priv
== ei
->chain_head_change_priv
)) {
968 if (block
->chain0
.chain
)
969 tcf_chain_head_change_item(item
, NULL
);
970 list_del(&item
->list
);
971 mutex_unlock(&block
->lock
);
977 mutex_unlock(&block
->lock
);
982 spinlock_t idr_lock
; /* Protects idr */
986 static unsigned int tcf_net_id
;
988 static int tcf_block_insert(struct tcf_block
*block
, struct net
*net
,
989 struct netlink_ext_ack
*extack
)
991 struct tcf_net
*tn
= net_generic(net
, tcf_net_id
);
994 idr_preload(GFP_KERNEL
);
995 spin_lock(&tn
->idr_lock
);
996 err
= idr_alloc_u32(&tn
->idr
, block
, &block
->index
, block
->index
,
998 spin_unlock(&tn
->idr_lock
);
1004 static void tcf_block_remove(struct tcf_block
*block
, struct net
*net
)
1006 struct tcf_net
*tn
= net_generic(net
, tcf_net_id
);
1008 spin_lock(&tn
->idr_lock
);
1009 idr_remove(&tn
->idr
, block
->index
);
1010 spin_unlock(&tn
->idr_lock
);
1013 static struct tcf_block
*tcf_block_create(struct net
*net
, struct Qdisc
*q
,
1015 struct netlink_ext_ack
*extack
)
1017 struct tcf_block
*block
;
1019 block
= kzalloc(sizeof(*block
), GFP_KERNEL
);
1021 NL_SET_ERR_MSG(extack
, "Memory allocation for block failed");
1022 return ERR_PTR(-ENOMEM
);
1024 mutex_init(&block
->lock
);
1025 mutex_init(&block
->proto_destroy_lock
);
1026 init_rwsem(&block
->cb_lock
);
1027 flow_block_init(&block
->flow_block
);
1028 INIT_LIST_HEAD(&block
->chain_list
);
1029 INIT_LIST_HEAD(&block
->owner_list
);
1030 INIT_LIST_HEAD(&block
->chain0
.filter_chain_list
);
1032 refcount_set(&block
->refcnt
, 1);
1034 block
->index
= block_index
;
1035 xa_init(&block
->ports
);
1037 /* Don't store q pointer for blocks which are shared */
1038 if (!tcf_block_shared(block
))
1043 struct tcf_block
*tcf_block_lookup(struct net
*net
, u32 block_index
)
1045 struct tcf_net
*tn
= net_generic(net
, tcf_net_id
);
1047 return idr_find(&tn
->idr
, block_index
);
1049 EXPORT_SYMBOL(tcf_block_lookup
);
1051 static struct tcf_block
*tcf_block_refcnt_get(struct net
*net
, u32 block_index
)
1053 struct tcf_block
*block
;
1056 block
= tcf_block_lookup(net
, block_index
);
1057 if (block
&& !refcount_inc_not_zero(&block
->refcnt
))
1064 static struct tcf_chain
*
1065 __tcf_get_next_chain(struct tcf_block
*block
, struct tcf_chain
*chain
)
1067 mutex_lock(&block
->lock
);
1069 chain
= list_is_last(&chain
->list
, &block
->chain_list
) ?
1070 NULL
: list_next_entry(chain
, list
);
1072 chain
= list_first_entry_or_null(&block
->chain_list
,
1073 struct tcf_chain
, list
);
1075 /* skip all action-only chains */
1076 while (chain
&& tcf_chain_held_by_acts_only(chain
))
1077 chain
= list_is_last(&chain
->list
, &block
->chain_list
) ?
1078 NULL
: list_next_entry(chain
, list
);
1081 tcf_chain_hold(chain
);
1082 mutex_unlock(&block
->lock
);
1087 /* Function to be used by all clients that want to iterate over all chains on
1088 * block. It properly obtains block->lock and takes reference to chain before
1089 * returning it. Users of this function must be tolerant to concurrent chain
1090 * insertion/deletion or ensure that no concurrent chain modification is
1091 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1092 * consistent dump because rtnl lock is released each time skb is filled with
1093 * data and sent to user-space.
1097 tcf_get_next_chain(struct tcf_block
*block
, struct tcf_chain
*chain
)
1099 struct tcf_chain
*chain_next
= __tcf_get_next_chain(block
, chain
);
1102 tcf_chain_put(chain
);
1106 EXPORT_SYMBOL(tcf_get_next_chain
);
1108 static struct tcf_proto
*
1109 __tcf_get_next_proto(struct tcf_chain
*chain
, struct tcf_proto
*tp
)
1114 mutex_lock(&chain
->filter_chain_lock
);
1117 tp
= tcf_chain_dereference(chain
->filter_chain
, chain
);
1118 } else if (tcf_proto_is_deleting(tp
)) {
1119 /* 'deleting' flag is set and chain->filter_chain_lock was
1120 * unlocked, which means next pointer could be invalid. Restart
1123 prio
= tp
->prio
+ 1;
1124 tp
= tcf_chain_dereference(chain
->filter_chain
, chain
);
1126 for (; tp
; tp
= tcf_chain_dereference(tp
->next
, chain
))
1127 if (!tp
->deleting
&& tp
->prio
>= prio
)
1130 tp
= tcf_chain_dereference(tp
->next
, chain
);
1136 mutex_unlock(&chain
->filter_chain_lock
);
1141 /* Function to be used by all clients that want to iterate over all tp's on
1142 * chain. Users of this function must be tolerant to concurrent tp
1143 * insertion/deletion or ensure that no concurrent chain modification is
1144 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1145 * consistent dump because rtnl lock is released each time skb is filled with
1146 * data and sent to user-space.
1150 tcf_get_next_proto(struct tcf_chain
*chain
, struct tcf_proto
*tp
)
1152 struct tcf_proto
*tp_next
= __tcf_get_next_proto(chain
, tp
);
1155 tcf_proto_put(tp
, true, NULL
);
1159 EXPORT_SYMBOL(tcf_get_next_proto
);
1161 static void tcf_block_flush_all_chains(struct tcf_block
*block
, bool rtnl_held
)
1163 struct tcf_chain
*chain
;
1165 /* Last reference to block. At this point chains cannot be added or
1166 * removed concurrently.
1168 for (chain
= tcf_get_next_chain(block
, NULL
);
1170 chain
= tcf_get_next_chain(block
, chain
)) {
1171 tcf_chain_put_explicitly_created(chain
);
1172 tcf_chain_flush(chain
, rtnl_held
);
1176 /* Lookup Qdisc and increments its reference counter.
1177 * Set parent, if necessary.
1180 static int __tcf_qdisc_find(struct net
*net
, struct Qdisc
**q
,
1181 u32
*parent
, int ifindex
, bool rtnl_held
,
1182 struct netlink_ext_ack
*extack
)
1184 const struct Qdisc_class_ops
*cops
;
1185 struct net_device
*dev
;
1188 if (ifindex
== TCM_IFINDEX_MAGIC_BLOCK
)
1194 dev
= dev_get_by_index_rcu(net
, ifindex
);
1202 *q
= rcu_dereference(dev
->qdisc
);
1203 *parent
= (*q
)->handle
;
1205 *q
= qdisc_lookup_rcu(dev
, TC_H_MAJ(*parent
));
1207 NL_SET_ERR_MSG(extack
, "Parent Qdisc doesn't exists");
1213 *q
= qdisc_refcount_inc_nz(*q
);
1215 NL_SET_ERR_MSG(extack
, "Parent Qdisc doesn't exists");
1220 /* Is it classful? */
1221 cops
= (*q
)->ops
->cl_ops
;
1223 NL_SET_ERR_MSG(extack
, "Qdisc not classful");
1228 if (!cops
->tcf_block
) {
1229 NL_SET_ERR_MSG(extack
, "Class doesn't support blocks");
1235 /* At this point we know that qdisc is not noop_qdisc,
1236 * which means that qdisc holds a reference to net_device
1237 * and we hold a reference to qdisc, so it is safe to release
1249 qdisc_put_unlocked(*q
);
1255 static int __tcf_qdisc_cl_find(struct Qdisc
*q
, u32 parent
, unsigned long *cl
,
1256 int ifindex
, struct netlink_ext_ack
*extack
)
1258 if (ifindex
== TCM_IFINDEX_MAGIC_BLOCK
)
1261 /* Do we search for filter, attached to class? */
1262 if (TC_H_MIN(parent
)) {
1263 const struct Qdisc_class_ops
*cops
= q
->ops
->cl_ops
;
1265 *cl
= cops
->find(q
, parent
);
1267 NL_SET_ERR_MSG(extack
, "Specified class doesn't exist");
1275 static struct tcf_block
*__tcf_block_find(struct net
*net
, struct Qdisc
*q
,
1276 unsigned long cl
, int ifindex
,
1278 struct netlink_ext_ack
*extack
)
1280 struct tcf_block
*block
;
1282 if (ifindex
== TCM_IFINDEX_MAGIC_BLOCK
) {
1283 block
= tcf_block_refcnt_get(net
, block_index
);
1285 NL_SET_ERR_MSG(extack
, "Block of given index was not found");
1286 return ERR_PTR(-EINVAL
);
1289 const struct Qdisc_class_ops
*cops
= q
->ops
->cl_ops
;
1291 block
= cops
->tcf_block(q
, cl
, extack
);
1293 return ERR_PTR(-EINVAL
);
1295 if (tcf_block_shared(block
)) {
1296 NL_SET_ERR_MSG(extack
, "This filter block is shared. Please use the block index to manipulate the filters");
1297 return ERR_PTR(-EOPNOTSUPP
);
1300 /* Always take reference to block in order to support execution
1301 * of rules update path of cls API without rtnl lock. Caller
1302 * must release block when it is finished using it. 'if' block
1303 * of this conditional obtain reference to block by calling
1304 * tcf_block_refcnt_get().
1306 refcount_inc(&block
->refcnt
);
1312 static void __tcf_block_put(struct tcf_block
*block
, struct Qdisc
*q
,
1313 struct tcf_block_ext_info
*ei
, bool rtnl_held
)
1315 if (refcount_dec_and_mutex_lock(&block
->refcnt
, &block
->lock
)) {
1316 /* Flushing/putting all chains will cause the block to be
1317 * deallocated when last chain is freed. However, if chain_list
1318 * is empty, block has to be manually deallocated. After block
1319 * reference counter reached 0, it is no longer possible to
1320 * increment it or add new chains to block.
1322 bool free_block
= list_empty(&block
->chain_list
);
1324 mutex_unlock(&block
->lock
);
1325 if (tcf_block_shared(block
))
1326 tcf_block_remove(block
, block
->net
);
1329 tcf_block_offload_unbind(block
, q
, ei
);
1332 tcf_block_destroy(block
);
1334 tcf_block_flush_all_chains(block
, rtnl_held
);
1336 tcf_block_offload_unbind(block
, q
, ei
);
1340 static void tcf_block_refcnt_put(struct tcf_block
*block
, bool rtnl_held
)
1342 __tcf_block_put(block
, NULL
, NULL
, rtnl_held
);
1346 * Set q, parent, cl when appropriate.
1349 static struct tcf_block
*tcf_block_find(struct net
*net
, struct Qdisc
**q
,
1350 u32
*parent
, unsigned long *cl
,
1351 int ifindex
, u32 block_index
,
1352 struct netlink_ext_ack
*extack
)
1354 struct tcf_block
*block
;
1359 err
= __tcf_qdisc_find(net
, q
, parent
, ifindex
, true, extack
);
1363 err
= __tcf_qdisc_cl_find(*q
, *parent
, cl
, ifindex
, extack
);
1367 block
= __tcf_block_find(net
, *q
, *cl
, ifindex
, block_index
, extack
);
1368 if (IS_ERR(block
)) {
1369 err
= PTR_ERR(block
);
1380 return ERR_PTR(err
);
1383 static void tcf_block_release(struct Qdisc
*q
, struct tcf_block
*block
,
1386 if (!IS_ERR_OR_NULL(block
))
1387 tcf_block_refcnt_put(block
, rtnl_held
);
1393 qdisc_put_unlocked(q
);
1397 struct tcf_block_owner_item
{
1398 struct list_head list
;
1400 enum flow_block_binder_type binder_type
;
1404 tcf_block_owner_netif_keep_dst(struct tcf_block
*block
,
1406 enum flow_block_binder_type binder_type
)
1408 if (block
->keep_dst
&&
1409 binder_type
!= FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS
&&
1410 binder_type
!= FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS
)
1411 netif_keep_dst(qdisc_dev(q
));
1414 void tcf_block_netif_keep_dst(struct tcf_block
*block
)
1416 struct tcf_block_owner_item
*item
;
1418 block
->keep_dst
= true;
1419 list_for_each_entry(item
, &block
->owner_list
, list
)
1420 tcf_block_owner_netif_keep_dst(block
, item
->q
,
1423 EXPORT_SYMBOL(tcf_block_netif_keep_dst
);
1425 static int tcf_block_owner_add(struct tcf_block
*block
,
1427 enum flow_block_binder_type binder_type
)
1429 struct tcf_block_owner_item
*item
;
1431 item
= kmalloc(sizeof(*item
), GFP_KERNEL
);
1435 item
->binder_type
= binder_type
;
1436 list_add(&item
->list
, &block
->owner_list
);
1440 static void tcf_block_owner_del(struct tcf_block
*block
,
1442 enum flow_block_binder_type binder_type
)
1444 struct tcf_block_owner_item
*item
;
1446 list_for_each_entry(item
, &block
->owner_list
, list
) {
1447 if (item
->q
== q
&& item
->binder_type
== binder_type
) {
1448 list_del(&item
->list
);
1456 static bool tcf_block_tracks_dev(struct tcf_block
*block
,
1457 struct tcf_block_ext_info
*ei
)
1459 return tcf_block_shared(block
) &&
1460 (ei
->binder_type
== FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS
||
1461 ei
->binder_type
== FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS
);
1464 int tcf_block_get_ext(struct tcf_block
**p_block
, struct Qdisc
*q
,
1465 struct tcf_block_ext_info
*ei
,
1466 struct netlink_ext_ack
*extack
)
1468 struct net_device
*dev
= qdisc_dev(q
);
1469 struct net
*net
= qdisc_net(q
);
1470 struct tcf_block
*block
= NULL
;
1473 if (ei
->block_index
)
1474 /* block_index not 0 means the shared block is requested */
1475 block
= tcf_block_refcnt_get(net
, ei
->block_index
);
1478 block
= tcf_block_create(net
, q
, ei
->block_index
, extack
);
1480 return PTR_ERR(block
);
1481 if (tcf_block_shared(block
)) {
1482 err
= tcf_block_insert(block
, net
, extack
);
1484 goto err_block_insert
;
1488 err
= tcf_block_owner_add(block
, q
, ei
->binder_type
);
1490 goto err_block_owner_add
;
1492 tcf_block_owner_netif_keep_dst(block
, q
, ei
->binder_type
);
1494 err
= tcf_chain0_head_change_cb_add(block
, ei
, extack
);
1496 goto err_chain0_head_change_cb_add
;
1498 err
= tcf_block_offload_bind(block
, q
, ei
, extack
);
1500 goto err_block_offload_bind
;
1502 if (tcf_block_tracks_dev(block
, ei
)) {
1503 err
= xa_insert(&block
->ports
, dev
->ifindex
, dev
, GFP_KERNEL
);
1505 NL_SET_ERR_MSG(extack
, "block dev insert failed");
1506 goto err_dev_insert
;
1514 tcf_block_offload_unbind(block
, q
, ei
);
1515 err_block_offload_bind
:
1516 tcf_chain0_head_change_cb_del(block
, ei
);
1517 err_chain0_head_change_cb_add
:
1518 tcf_block_owner_del(block
, q
, ei
->binder_type
);
1519 err_block_owner_add
:
1521 tcf_block_refcnt_put(block
, true);
1524 EXPORT_SYMBOL(tcf_block_get_ext
);
1526 static void tcf_chain_head_change_dflt(struct tcf_proto
*tp_head
, void *priv
)
1528 struct tcf_proto __rcu
**p_filter_chain
= priv
;
1530 rcu_assign_pointer(*p_filter_chain
, tp_head
);
1533 int tcf_block_get(struct tcf_block
**p_block
,
1534 struct tcf_proto __rcu
**p_filter_chain
, struct Qdisc
*q
,
1535 struct netlink_ext_ack
*extack
)
1537 struct tcf_block_ext_info ei
= {
1538 .chain_head_change
= tcf_chain_head_change_dflt
,
1539 .chain_head_change_priv
= p_filter_chain
,
1542 WARN_ON(!p_filter_chain
);
1543 return tcf_block_get_ext(p_block
, q
, &ei
, extack
);
1545 EXPORT_SYMBOL(tcf_block_get
);
1547 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1548 * actions should be all removed after flushing.
1550 void tcf_block_put_ext(struct tcf_block
*block
, struct Qdisc
*q
,
1551 struct tcf_block_ext_info
*ei
)
1553 struct net_device
*dev
= qdisc_dev(q
);
1557 if (tcf_block_tracks_dev(block
, ei
))
1558 xa_erase(&block
->ports
, dev
->ifindex
);
1559 tcf_chain0_head_change_cb_del(block
, ei
);
1560 tcf_block_owner_del(block
, q
, ei
->binder_type
);
1562 __tcf_block_put(block
, q
, ei
, true);
1564 EXPORT_SYMBOL(tcf_block_put_ext
);
1566 void tcf_block_put(struct tcf_block
*block
)
1568 struct tcf_block_ext_info ei
= {0, };
1572 tcf_block_put_ext(block
, block
->q
, &ei
);
1575 EXPORT_SYMBOL(tcf_block_put
);
1578 tcf_block_playback_offloads(struct tcf_block
*block
, flow_setup_cb_t
*cb
,
1579 void *cb_priv
, bool add
, bool offload_in_use
,
1580 struct netlink_ext_ack
*extack
)
1582 struct tcf_chain
*chain
, *chain_prev
;
1583 struct tcf_proto
*tp
, *tp_prev
;
1586 lockdep_assert_held(&block
->cb_lock
);
1588 for (chain
= __tcf_get_next_chain(block
, NULL
);
1591 chain
= __tcf_get_next_chain(block
, chain
),
1592 tcf_chain_put(chain_prev
)) {
1593 if (chain
->tmplt_ops
&& add
)
1594 chain
->tmplt_ops
->tmplt_reoffload(chain
, true, cb
,
1596 for (tp
= __tcf_get_next_proto(chain
, NULL
); tp
;
1598 tp
= __tcf_get_next_proto(chain
, tp
),
1599 tcf_proto_put(tp_prev
, true, NULL
)) {
1600 if (tp
->ops
->reoffload
) {
1601 err
= tp
->ops
->reoffload(tp
, add
, cb
, cb_priv
,
1604 goto err_playback_remove
;
1605 } else if (add
&& offload_in_use
) {
1607 NL_SET_ERR_MSG(extack
, "Filter HW offload failed - classifier without re-offloading support");
1608 goto err_playback_remove
;
1611 if (chain
->tmplt_ops
&& !add
)
1612 chain
->tmplt_ops
->tmplt_reoffload(chain
, false, cb
,
1618 err_playback_remove
:
1619 tcf_proto_put(tp
, true, NULL
);
1620 tcf_chain_put(chain
);
1621 tcf_block_playback_offloads(block
, cb
, cb_priv
, false, offload_in_use
,
1626 static int tcf_block_bind(struct tcf_block
*block
,
1627 struct flow_block_offload
*bo
)
1629 struct flow_block_cb
*block_cb
, *next
;
1632 lockdep_assert_held(&block
->cb_lock
);
1634 list_for_each_entry(block_cb
, &bo
->cb_list
, list
) {
1635 err
= tcf_block_playback_offloads(block
, block_cb
->cb
,
1636 block_cb
->cb_priv
, true,
1637 tcf_block_offload_in_use(block
),
1641 if (!bo
->unlocked_driver_cb
)
1642 block
->lockeddevcnt
++;
1646 list_splice(&bo
->cb_list
, &block
->flow_block
.cb_list
);
1651 list_for_each_entry_safe(block_cb
, next
, &bo
->cb_list
, list
) {
1652 list_del(&block_cb
->driver_list
);
1654 list_del(&block_cb
->list
);
1655 tcf_block_playback_offloads(block
, block_cb
->cb
,
1656 block_cb
->cb_priv
, false,
1657 tcf_block_offload_in_use(block
),
1659 if (!bo
->unlocked_driver_cb
)
1660 block
->lockeddevcnt
--;
1662 flow_block_cb_free(block_cb
);
1668 static void tcf_block_unbind(struct tcf_block
*block
,
1669 struct flow_block_offload
*bo
)
1671 struct flow_block_cb
*block_cb
, *next
;
1673 lockdep_assert_held(&block
->cb_lock
);
1675 list_for_each_entry_safe(block_cb
, next
, &bo
->cb_list
, list
) {
1676 tcf_block_playback_offloads(block
, block_cb
->cb
,
1677 block_cb
->cb_priv
, false,
1678 tcf_block_offload_in_use(block
),
1680 list_del(&block_cb
->list
);
1681 flow_block_cb_free(block_cb
);
1682 if (!bo
->unlocked_driver_cb
)
1683 block
->lockeddevcnt
--;
1687 static int tcf_block_setup(struct tcf_block
*block
,
1688 struct flow_block_offload
*bo
)
1692 switch (bo
->command
) {
1693 case FLOW_BLOCK_BIND
:
1694 err
= tcf_block_bind(block
, bo
);
1696 case FLOW_BLOCK_UNBIND
:
1698 tcf_block_unbind(block
, bo
);
1708 /* Main classifier routine: scans classifier chain attached
1709 * to this qdisc, (optionally) tests for protocol and asks
1710 * specific classifiers.
1712 static inline int __tcf_classify(struct sk_buff
*skb
,
1713 const struct tcf_proto
*tp
,
1714 const struct tcf_proto
*orig_tp
,
1715 struct tcf_result
*res
,
1717 struct tcf_exts_miss_cookie_node
*n
,
1719 u32
*last_executed_chain
)
1721 #ifdef CONFIG_NET_CLS_ACT
1722 const int max_reclassify_loop
= 16;
1723 const struct tcf_proto
*first_tp
;
1728 for (; tp
; tp
= rcu_dereference_bh(tp
->next
)) {
1729 __be16 protocol
= skb_protocol(skb
, false);
1733 struct tcf_exts
*exts
;
1735 if (n
->tp_prio
!= tp
->prio
)
1738 /* We re-lookup the tp and chain based on index instead
1739 * of having hard refs and locks to them, so do a sanity
1740 * check if any of tp,chain,exts was replaced by the
1741 * time we got here with a cookie from hardware.
1743 if (unlikely(n
->tp
!= tp
|| n
->tp
->chain
!= n
->chain
||
1744 !tp
->ops
->get_exts
)) {
1745 tcf_set_drop_reason(skb
,
1746 SKB_DROP_REASON_TC_COOKIE_ERROR
);
1750 exts
= tp
->ops
->get_exts(tp
, n
->handle
);
1751 if (unlikely(!exts
|| n
->exts
!= exts
)) {
1752 tcf_set_drop_reason(skb
,
1753 SKB_DROP_REASON_TC_COOKIE_ERROR
);
1758 err
= tcf_exts_exec_ex(skb
, exts
, act_index
, res
);
1760 if (tp
->protocol
!= protocol
&&
1761 tp
->protocol
!= htons(ETH_P_ALL
))
1764 err
= tc_classify(skb
, tp
, res
);
1766 #ifdef CONFIG_NET_CLS_ACT
1767 if (unlikely(err
== TC_ACT_RECLASSIFY
&& !compat_mode
)) {
1769 *last_executed_chain
= first_tp
->chain
->index
;
1771 } else if (unlikely(TC_ACT_EXT_CMP(err
, TC_ACT_GOTO_CHAIN
))) {
1772 first_tp
= res
->goto_tp
;
1773 *last_executed_chain
= err
& TC_ACT_EXT_VAL_MASK
;
1782 tcf_set_drop_reason(skb
,
1783 SKB_DROP_REASON_TC_COOKIE_ERROR
);
1787 return TC_ACT_UNSPEC
; /* signal: continue lookup */
1788 #ifdef CONFIG_NET_CLS_ACT
1790 if (unlikely(limit
++ >= max_reclassify_loop
)) {
1791 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1792 tp
->chain
->block
->index
,
1794 ntohs(tp
->protocol
));
1795 tcf_set_drop_reason(skb
,
1796 SKB_DROP_REASON_TC_RECLASSIFY_LOOP
);
1805 int tcf_classify(struct sk_buff
*skb
,
1806 const struct tcf_block
*block
,
1807 const struct tcf_proto
*tp
,
1808 struct tcf_result
*res
, bool compat_mode
)
1810 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1811 u32 last_executed_chain
= 0;
1813 return __tcf_classify(skb
, tp
, tp
, res
, compat_mode
, NULL
, 0,
1814 &last_executed_chain
);
1816 u32 last_executed_chain
= tp
? tp
->chain
->index
: 0;
1817 struct tcf_exts_miss_cookie_node
*n
= NULL
;
1818 const struct tcf_proto
*orig_tp
= tp
;
1819 struct tc_skb_ext
*ext
;
1824 ext
= skb_ext_find(skb
, TC_SKB_EXT
);
1826 if (ext
&& (ext
->chain
|| ext
->act_miss
)) {
1827 struct tcf_chain
*fchain
;
1830 if (ext
->act_miss
) {
1831 n
= tcf_exts_miss_cookie_lookup(ext
->act_miss_cookie
,
1834 tcf_set_drop_reason(skb
,
1835 SKB_DROP_REASON_TC_COOKIE_ERROR
);
1839 chain
= n
->chain_index
;
1844 fchain
= tcf_chain_lookup_rcu(block
, chain
);
1846 tcf_set_drop_reason(skb
,
1847 SKB_DROP_REASON_TC_CHAIN_NOTFOUND
);
1852 /* Consume, so cloned/redirect skbs won't inherit ext */
1853 skb_ext_del(skb
, TC_SKB_EXT
);
1855 tp
= rcu_dereference_bh(fchain
->filter_chain
);
1856 last_executed_chain
= fchain
->index
;
1860 ret
= __tcf_classify(skb
, tp
, orig_tp
, res
, compat_mode
, n
, act_index
,
1861 &last_executed_chain
);
1863 if (tc_skb_ext_tc_enabled()) {
1864 /* If we missed on some chain */
1865 if (ret
== TC_ACT_UNSPEC
&& last_executed_chain
) {
1866 struct tc_skb_cb
*cb
= tc_skb_cb(skb
);
1868 ext
= tc_skb_ext_alloc(skb
);
1869 if (WARN_ON_ONCE(!ext
)) {
1870 tcf_set_drop_reason(skb
, SKB_DROP_REASON_NOMEM
);
1873 ext
->chain
= last_executed_chain
;
1875 ext
->post_ct
= cb
->post_ct
;
1876 ext
->post_ct_snat
= cb
->post_ct_snat
;
1877 ext
->post_ct_dnat
= cb
->post_ct_dnat
;
1878 ext
->zone
= cb
->zone
;
1885 EXPORT_SYMBOL(tcf_classify
);
1887 struct tcf_chain_info
{
1888 struct tcf_proto __rcu
**pprev
;
1889 struct tcf_proto __rcu
*next
;
1892 static struct tcf_proto
*tcf_chain_tp_prev(struct tcf_chain
*chain
,
1893 struct tcf_chain_info
*chain_info
)
1895 return tcf_chain_dereference(*chain_info
->pprev
, chain
);
1898 static int tcf_chain_tp_insert(struct tcf_chain
*chain
,
1899 struct tcf_chain_info
*chain_info
,
1900 struct tcf_proto
*tp
)
1902 if (chain
->flushing
)
1905 RCU_INIT_POINTER(tp
->next
, tcf_chain_tp_prev(chain
, chain_info
));
1906 if (*chain_info
->pprev
== chain
->filter_chain
)
1907 tcf_chain0_head_change(chain
, tp
);
1909 rcu_assign_pointer(*chain_info
->pprev
, tp
);
1914 static void tcf_chain_tp_remove(struct tcf_chain
*chain
,
1915 struct tcf_chain_info
*chain_info
,
1916 struct tcf_proto
*tp
)
1918 struct tcf_proto
*next
= tcf_chain_dereference(chain_info
->next
, chain
);
1920 tcf_proto_mark_delete(tp
);
1921 if (tp
== chain
->filter_chain
)
1922 tcf_chain0_head_change(chain
, next
);
1923 RCU_INIT_POINTER(*chain_info
->pprev
, next
);
1926 static struct tcf_proto
*tcf_chain_tp_find(struct tcf_chain
*chain
,
1927 struct tcf_chain_info
*chain_info
,
1928 u32 protocol
, u32 prio
,
1930 struct netlink_ext_ack
*extack
);
1932 /* Try to insert new proto.
1933 * If proto with specified priority already exists, free new proto
1934 * and return existing one.
1937 static struct tcf_proto
*tcf_chain_tp_insert_unique(struct tcf_chain
*chain
,
1938 struct tcf_proto
*tp_new
,
1939 u32 protocol
, u32 prio
,
1942 struct tcf_chain_info chain_info
;
1943 struct tcf_proto
*tp
;
1946 mutex_lock(&chain
->filter_chain_lock
);
1948 if (tcf_proto_exists_destroying(chain
, tp_new
)) {
1949 mutex_unlock(&chain
->filter_chain_lock
);
1950 tcf_proto_destroy(tp_new
, rtnl_held
, false, NULL
);
1951 return ERR_PTR(-EAGAIN
);
1954 tp
= tcf_chain_tp_find(chain
, &chain_info
, protocol
, prio
, false, NULL
);
1956 err
= tcf_chain_tp_insert(chain
, &chain_info
, tp_new
);
1957 mutex_unlock(&chain
->filter_chain_lock
);
1960 tcf_proto_destroy(tp_new
, rtnl_held
, false, NULL
);
1963 tcf_proto_destroy(tp_new
, rtnl_held
, false, NULL
);
1964 tp_new
= ERR_PTR(err
);
1970 static void tcf_chain_tp_delete_empty(struct tcf_chain
*chain
,
1971 struct tcf_proto
*tp
, bool rtnl_held
,
1972 struct netlink_ext_ack
*extack
)
1974 struct tcf_chain_info chain_info
;
1975 struct tcf_proto
*tp_iter
;
1976 struct tcf_proto
**pprev
;
1977 struct tcf_proto
*next
;
1979 mutex_lock(&chain
->filter_chain_lock
);
1981 /* Atomically find and remove tp from chain. */
1982 for (pprev
= &chain
->filter_chain
;
1983 (tp_iter
= tcf_chain_dereference(*pprev
, chain
));
1984 pprev
= &tp_iter
->next
) {
1985 if (tp_iter
== tp
) {
1986 chain_info
.pprev
= pprev
;
1987 chain_info
.next
= tp_iter
->next
;
1988 WARN_ON(tp_iter
->deleting
);
1992 /* Verify that tp still exists and no new filters were inserted
1994 * Mark tp for deletion if it is empty.
1996 if (!tp_iter
|| !tcf_proto_check_delete(tp
)) {
1997 mutex_unlock(&chain
->filter_chain_lock
);
2001 tcf_proto_signal_destroying(chain
, tp
);
2002 next
= tcf_chain_dereference(chain_info
.next
, chain
);
2003 if (tp
== chain
->filter_chain
)
2004 tcf_chain0_head_change(chain
, next
);
2005 RCU_INIT_POINTER(*chain_info
.pprev
, next
);
2006 mutex_unlock(&chain
->filter_chain_lock
);
2008 tcf_proto_put(tp
, rtnl_held
, extack
);
2011 static struct tcf_proto
*tcf_chain_tp_find(struct tcf_chain
*chain
,
2012 struct tcf_chain_info
*chain_info
,
2013 u32 protocol
, u32 prio
,
2015 struct netlink_ext_ack
*extack
)
2017 struct tcf_proto
**pprev
;
2018 struct tcf_proto
*tp
;
2020 /* Check the chain for existence of proto-tcf with this priority */
2021 for (pprev
= &chain
->filter_chain
;
2022 (tp
= tcf_chain_dereference(*pprev
, chain
));
2023 pprev
= &tp
->next
) {
2024 if (tp
->prio
>= prio
) {
2025 if (tp
->prio
== prio
) {
2026 if (prio_allocate
) {
2027 NL_SET_ERR_MSG(extack
, "Lowest ID from auto-alloc range already in use");
2028 return ERR_PTR(-ENOSPC
);
2030 if (tp
->protocol
!= protocol
&& protocol
) {
2031 NL_SET_ERR_MSG(extack
, "Protocol mismatch for filter with specified priority");
2032 return ERR_PTR(-EINVAL
);
2040 chain_info
->pprev
= pprev
;
2042 chain_info
->next
= tp
->next
;
2045 chain_info
->next
= NULL
;
2050 static int tcf_fill_node(struct net
*net
, struct sk_buff
*skb
,
2051 struct tcf_proto
*tp
, struct tcf_block
*block
,
2052 struct Qdisc
*q
, u32 parent
, void *fh
,
2053 u32 portid
, u32 seq
, u16 flags
, int event
,
2054 bool terse_dump
, bool rtnl_held
,
2055 struct netlink_ext_ack
*extack
)
2058 struct nlmsghdr
*nlh
;
2059 unsigned char *b
= skb_tail_pointer(skb
);
2061 nlh
= nlmsg_put(skb
, portid
, seq
, event
, sizeof(*tcm
), flags
);
2063 goto out_nlmsg_trim
;
2064 tcm
= nlmsg_data(nlh
);
2065 tcm
->tcm_family
= AF_UNSPEC
;
2069 tcm
->tcm_ifindex
= qdisc_dev(q
)->ifindex
;
2070 tcm
->tcm_parent
= parent
;
2072 tcm
->tcm_ifindex
= TCM_IFINDEX_MAGIC_BLOCK
;
2073 tcm
->tcm_block_index
= block
->index
;
2075 tcm
->tcm_info
= TC_H_MAKE(tp
->prio
, tp
->protocol
);
2076 if (nla_put_string(skb
, TCA_KIND
, tp
->ops
->kind
))
2077 goto nla_put_failure
;
2078 if (nla_put_u32(skb
, TCA_CHAIN
, tp
->chain
->index
))
2079 goto nla_put_failure
;
2081 tcm
->tcm_handle
= 0;
2082 } else if (terse_dump
) {
2083 if (tp
->ops
->terse_dump
) {
2084 if (tp
->ops
->terse_dump(net
, tp
, fh
, skb
, tcm
,
2086 goto nla_put_failure
;
2088 goto cls_op_not_supp
;
2091 if (tp
->ops
->dump
&&
2092 tp
->ops
->dump(net
, tp
, fh
, skb
, tcm
, rtnl_held
) < 0)
2093 goto nla_put_failure
;
2096 if (extack
&& extack
->_msg
&&
2097 nla_put_string(skb
, TCA_EXT_WARN_MSG
, extack
->_msg
))
2098 goto nla_put_failure
;
2100 nlh
->nlmsg_len
= skb_tail_pointer(skb
) - b
;
2111 static int tfilter_notify(struct net
*net
, struct sk_buff
*oskb
,
2112 struct nlmsghdr
*n
, struct tcf_proto
*tp
,
2113 struct tcf_block
*block
, struct Qdisc
*q
,
2114 u32 parent
, void *fh
, int event
, bool unicast
,
2115 bool rtnl_held
, struct netlink_ext_ack
*extack
)
2117 struct sk_buff
*skb
;
2118 u32 portid
= oskb
? NETLINK_CB(oskb
).portid
: 0;
2121 if (!unicast
&& !rtnl_notify_needed(net
, n
->nlmsg_flags
, RTNLGRP_TC
))
2124 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
2128 if (tcf_fill_node(net
, skb
, tp
, block
, q
, parent
, fh
, portid
,
2129 n
->nlmsg_seq
, n
->nlmsg_flags
, event
,
2130 false, rtnl_held
, extack
) <= 0) {
2136 err
= rtnl_unicast(skb
, net
, portid
);
2138 err
= rtnetlink_send(skb
, net
, portid
, RTNLGRP_TC
,
2139 n
->nlmsg_flags
& NLM_F_ECHO
);
2143 static int tfilter_del_notify(struct net
*net
, struct sk_buff
*oskb
,
2144 struct nlmsghdr
*n
, struct tcf_proto
*tp
,
2145 struct tcf_block
*block
, struct Qdisc
*q
,
2146 u32 parent
, void *fh
, bool *last
, bool rtnl_held
,
2147 struct netlink_ext_ack
*extack
)
2149 struct sk_buff
*skb
;
2150 u32 portid
= oskb
? NETLINK_CB(oskb
).portid
: 0;
2153 if (!rtnl_notify_needed(net
, n
->nlmsg_flags
, RTNLGRP_TC
))
2154 return tp
->ops
->delete(tp
, fh
, last
, rtnl_held
, extack
);
2156 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
2160 if (tcf_fill_node(net
, skb
, tp
, block
, q
, parent
, fh
, portid
,
2161 n
->nlmsg_seq
, n
->nlmsg_flags
, RTM_DELTFILTER
,
2162 false, rtnl_held
, extack
) <= 0) {
2163 NL_SET_ERR_MSG(extack
, "Failed to build del event notification");
2168 err
= tp
->ops
->delete(tp
, fh
, last
, rtnl_held
, extack
);
2174 err
= rtnetlink_send(skb
, net
, portid
, RTNLGRP_TC
,
2175 n
->nlmsg_flags
& NLM_F_ECHO
);
2177 NL_SET_ERR_MSG(extack
, "Failed to send filter delete notification");
2182 static void tfilter_notify_chain(struct net
*net
, struct sk_buff
*oskb
,
2183 struct tcf_block
*block
, struct Qdisc
*q
,
2184 u32 parent
, struct nlmsghdr
*n
,
2185 struct tcf_chain
*chain
, int event
,
2186 struct netlink_ext_ack
*extack
)
2188 struct tcf_proto
*tp
;
2190 for (tp
= tcf_get_next_proto(chain
, NULL
);
2191 tp
; tp
= tcf_get_next_proto(chain
, tp
))
2192 tfilter_notify(net
, oskb
, n
, tp
, block
, q
, parent
, NULL
,
2193 event
, false, true, extack
);
2196 static void tfilter_put(struct tcf_proto
*tp
, void *fh
)
2198 if (tp
->ops
->put
&& fh
)
2199 tp
->ops
->put(tp
, fh
);
2202 static bool is_qdisc_ingress(__u32 classid
)
2204 return (TC_H_MIN(classid
) == TC_H_MIN(TC_H_MIN_INGRESS
));
2207 static int tc_new_tfilter(struct sk_buff
*skb
, struct nlmsghdr
*n
,
2208 struct netlink_ext_ack
*extack
)
2210 struct net
*net
= sock_net(skb
->sk
);
2211 struct nlattr
*tca
[TCA_MAX
+ 1];
2212 char name
[IFNAMSIZ
];
2220 struct tcf_chain_info chain_info
;
2221 struct tcf_chain
*chain
;
2222 struct tcf_block
*block
;
2223 struct tcf_proto
*tp
;
2228 bool rtnl_held
= false;
2234 err
= nlmsg_parse_deprecated(n
, sizeof(*t
), tca
, TCA_MAX
,
2235 rtm_tca_policy
, extack
);
2240 protocol
= TC_H_MIN(t
->tcm_info
);
2241 prio
= TC_H_MAJ(t
->tcm_info
);
2242 prio_allocate
= false;
2243 parent
= t
->tcm_parent
;
2252 /* If no priority is provided by the user,
2255 if (n
->nlmsg_flags
& NLM_F_CREATE
) {
2256 prio
= TC_H_MAKE(0x80000000U
, 0U);
2257 prio_allocate
= true;
2259 NL_SET_ERR_MSG(extack
, "Invalid filter command with priority of zero");
2264 /* Find head of filter chain. */
2266 err
= __tcf_qdisc_find(net
, &q
, &parent
, t
->tcm_ifindex
, false, extack
);
2270 if (tcf_proto_check_kind(tca
[TCA_KIND
], name
)) {
2271 NL_SET_ERR_MSG(extack
, "Specified TC filter name too long");
2276 /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2277 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2278 * type is not specified, classifier is not unlocked.
2281 (q
&& !(q
->ops
->cl_ops
->flags
& QDISC_CLASS_OPS_DOIT_UNLOCKED
)) ||
2282 !tcf_proto_is_unlocked(name
)) {
2287 err
= __tcf_qdisc_cl_find(q
, parent
, &cl
, t
->tcm_ifindex
, extack
);
2291 block
= __tcf_block_find(net
, q
, cl
, t
->tcm_ifindex
, t
->tcm_block_index
,
2293 if (IS_ERR(block
)) {
2294 err
= PTR_ERR(block
);
2297 block
->classid
= parent
;
2299 chain_index
= nla_get_u32_default(tca
[TCA_CHAIN
], 0);
2300 if (chain_index
> TC_ACT_EXT_VAL_MASK
) {
2301 NL_SET_ERR_MSG(extack
, "Specified chain index exceeds upper limit");
2305 chain
= tcf_chain_get(block
, chain_index
, true);
2307 NL_SET_ERR_MSG(extack
, "Cannot create specified filter chain");
2312 mutex_lock(&chain
->filter_chain_lock
);
2313 tp
= tcf_chain_tp_find(chain
, &chain_info
, protocol
,
2314 prio
, prio_allocate
, extack
);
2321 struct tcf_proto
*tp_new
= NULL
;
2323 if (chain
->flushing
) {
2328 /* Proto-tcf does not exist, create new one */
2330 if (tca
[TCA_KIND
] == NULL
|| !protocol
) {
2331 NL_SET_ERR_MSG(extack
, "Filter kind and protocol must be specified");
2336 if (!(n
->nlmsg_flags
& NLM_F_CREATE
)) {
2337 NL_SET_ERR_MSG(extack
, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2343 prio
= tcf_auto_prio(tcf_chain_tp_prev(chain
,
2346 mutex_unlock(&chain
->filter_chain_lock
);
2347 tp_new
= tcf_proto_create(name
, protocol
, prio
, chain
,
2349 if (IS_ERR(tp_new
)) {
2350 err
= PTR_ERR(tp_new
);
2355 tp
= tcf_chain_tp_insert_unique(chain
, tp_new
, protocol
, prio
,
2362 mutex_unlock(&chain
->filter_chain_lock
);
2365 if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], tp
->ops
->kind
)) {
2366 NL_SET_ERR_MSG(extack
, "Specified filter kind does not match existing one");
2371 fh
= tp
->ops
->get(tp
, t
->tcm_handle
);
2374 if (!(n
->nlmsg_flags
& NLM_F_CREATE
)) {
2375 NL_SET_ERR_MSG(extack
, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2379 } else if (n
->nlmsg_flags
& NLM_F_EXCL
) {
2380 tfilter_put(tp
, fh
);
2381 NL_SET_ERR_MSG(extack
, "Filter already exists");
2386 if (chain
->tmplt_ops
&& chain
->tmplt_ops
!= tp
->ops
) {
2387 tfilter_put(tp
, fh
);
2388 NL_SET_ERR_MSG(extack
, "Chain template is set to a different filter kind");
2393 if (!(n
->nlmsg_flags
& NLM_F_CREATE
))
2394 flags
|= TCA_ACT_FLAGS_REPLACE
;
2396 flags
|= TCA_ACT_FLAGS_NO_RTNL
;
2397 if (is_qdisc_ingress(parent
))
2398 flags
|= TCA_ACT_FLAGS_AT_INGRESS
;
2399 err
= tp
->ops
->change(net
, skb
, tp
, cl
, t
->tcm_handle
, tca
, &fh
,
2402 tfilter_notify(net
, skb
, n
, tp
, block
, q
, parent
, fh
,
2403 RTM_NEWTFILTER
, false, rtnl_held
, extack
);
2404 tfilter_put(tp
, fh
);
2405 tcf_proto_count_usesw(tp
, true);
2406 /* q pointer is NULL for shared blocks */
2408 q
->flags
&= ~TCQ_F_CAN_BYPASS
;
2412 if (err
&& tp_created
)
2413 tcf_chain_tp_delete_empty(chain
, tp
, rtnl_held
, NULL
);
2416 if (tp
&& !IS_ERR(tp
))
2417 tcf_proto_put(tp
, rtnl_held
, NULL
);
2419 tcf_chain_put(chain
);
2421 tcf_block_release(q
, block
, rtnl_held
);
2426 if (err
== -EAGAIN
) {
2427 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2431 /* Replay the request. */
2437 mutex_unlock(&chain
->filter_chain_lock
);
2441 static int tc_del_tfilter(struct sk_buff
*skb
, struct nlmsghdr
*n
,
2442 struct netlink_ext_ack
*extack
)
2444 struct net
*net
= sock_net(skb
->sk
);
2445 struct nlattr
*tca
[TCA_MAX
+ 1];
2446 char name
[IFNAMSIZ
];
2452 struct Qdisc
*q
= NULL
;
2453 struct tcf_chain_info chain_info
;
2454 struct tcf_chain
*chain
= NULL
;
2455 struct tcf_block
*block
= NULL
;
2456 struct tcf_proto
*tp
= NULL
;
2457 unsigned long cl
= 0;
2460 bool rtnl_held
= false;
2462 err
= nlmsg_parse_deprecated(n
, sizeof(*t
), tca
, TCA_MAX
,
2463 rtm_tca_policy
, extack
);
2468 protocol
= TC_H_MIN(t
->tcm_info
);
2469 prio
= TC_H_MAJ(t
->tcm_info
);
2470 parent
= t
->tcm_parent
;
2472 if (prio
== 0 && (protocol
|| t
->tcm_handle
|| tca
[TCA_KIND
])) {
2473 NL_SET_ERR_MSG(extack
, "Cannot flush filters with protocol, handle or kind set");
2477 /* Find head of filter chain. */
2479 err
= __tcf_qdisc_find(net
, &q
, &parent
, t
->tcm_ifindex
, false, extack
);
2483 if (tcf_proto_check_kind(tca
[TCA_KIND
], name
)) {
2484 NL_SET_ERR_MSG(extack
, "Specified TC filter name too long");
2488 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2489 * found), qdisc is not unlocked, classifier type is not specified,
2490 * classifier is not unlocked.
2493 (q
&& !(q
->ops
->cl_ops
->flags
& QDISC_CLASS_OPS_DOIT_UNLOCKED
)) ||
2494 !tcf_proto_is_unlocked(name
)) {
2499 err
= __tcf_qdisc_cl_find(q
, parent
, &cl
, t
->tcm_ifindex
, extack
);
2503 block
= __tcf_block_find(net
, q
, cl
, t
->tcm_ifindex
, t
->tcm_block_index
,
2505 if (IS_ERR(block
)) {
2506 err
= PTR_ERR(block
);
2510 chain_index
= nla_get_u32_default(tca
[TCA_CHAIN
], 0);
2511 if (chain_index
> TC_ACT_EXT_VAL_MASK
) {
2512 NL_SET_ERR_MSG(extack
, "Specified chain index exceeds upper limit");
2516 chain
= tcf_chain_get(block
, chain_index
, false);
2518 /* User requested flush on non-existent chain. Nothing to do,
2519 * so just return success.
2525 NL_SET_ERR_MSG(extack
, "Cannot find specified filter chain");
2531 tfilter_notify_chain(net
, skb
, block
, q
, parent
, n
,
2532 chain
, RTM_DELTFILTER
, extack
);
2533 tcf_chain_flush(chain
, rtnl_held
);
2538 mutex_lock(&chain
->filter_chain_lock
);
2539 tp
= tcf_chain_tp_find(chain
, &chain_info
, protocol
,
2540 prio
, false, extack
);
2543 NL_SET_ERR_MSG(extack
, "Filter with specified priority/protocol not found");
2545 } else if (IS_ERR(tp
)) {
2548 } else if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], tp
->ops
->kind
)) {
2549 NL_SET_ERR_MSG(extack
, "Specified filter kind does not match existing one");
2552 } else if (t
->tcm_handle
== 0) {
2553 tcf_proto_signal_destroying(chain
, tp
);
2554 tcf_chain_tp_remove(chain
, &chain_info
, tp
);
2555 mutex_unlock(&chain
->filter_chain_lock
);
2557 tcf_proto_put(tp
, rtnl_held
, NULL
);
2558 tfilter_notify(net
, skb
, n
, tp
, block
, q
, parent
, fh
,
2559 RTM_DELTFILTER
, false, rtnl_held
, extack
);
2563 mutex_unlock(&chain
->filter_chain_lock
);
2565 fh
= tp
->ops
->get(tp
, t
->tcm_handle
);
2568 NL_SET_ERR_MSG(extack
, "Specified filter handle not found");
2573 err
= tfilter_del_notify(net
, skb
, n
, tp
, block
, q
, parent
, fh
,
2574 &last
, rtnl_held
, extack
);
2579 tcf_chain_tp_delete_empty(chain
, tp
, rtnl_held
, extack
);
2584 if (tp
&& !IS_ERR(tp
))
2585 tcf_proto_put(tp
, rtnl_held
, NULL
);
2586 tcf_chain_put(chain
);
2588 tcf_block_release(q
, block
, rtnl_held
);
2596 mutex_unlock(&chain
->filter_chain_lock
);
2600 static int tc_get_tfilter(struct sk_buff
*skb
, struct nlmsghdr
*n
,
2601 struct netlink_ext_ack
*extack
)
2603 struct net
*net
= sock_net(skb
->sk
);
2604 struct nlattr
*tca
[TCA_MAX
+ 1];
2605 char name
[IFNAMSIZ
];
2611 struct Qdisc
*q
= NULL
;
2612 struct tcf_chain_info chain_info
;
2613 struct tcf_chain
*chain
= NULL
;
2614 struct tcf_block
*block
= NULL
;
2615 struct tcf_proto
*tp
= NULL
;
2616 unsigned long cl
= 0;
2619 bool rtnl_held
= false;
2621 err
= nlmsg_parse_deprecated(n
, sizeof(*t
), tca
, TCA_MAX
,
2622 rtm_tca_policy
, extack
);
2627 protocol
= TC_H_MIN(t
->tcm_info
);
2628 prio
= TC_H_MAJ(t
->tcm_info
);
2629 parent
= t
->tcm_parent
;
2632 NL_SET_ERR_MSG(extack
, "Invalid filter command with priority of zero");
2636 /* Find head of filter chain. */
2638 err
= __tcf_qdisc_find(net
, &q
, &parent
, t
->tcm_ifindex
, false, extack
);
2642 if (tcf_proto_check_kind(tca
[TCA_KIND
], name
)) {
2643 NL_SET_ERR_MSG(extack
, "Specified TC filter name too long");
2647 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2648 * unlocked, classifier type is not specified, classifier is not
2651 if ((q
&& !(q
->ops
->cl_ops
->flags
& QDISC_CLASS_OPS_DOIT_UNLOCKED
)) ||
2652 !tcf_proto_is_unlocked(name
)) {
2657 err
= __tcf_qdisc_cl_find(q
, parent
, &cl
, t
->tcm_ifindex
, extack
);
2661 block
= __tcf_block_find(net
, q
, cl
, t
->tcm_ifindex
, t
->tcm_block_index
,
2663 if (IS_ERR(block
)) {
2664 err
= PTR_ERR(block
);
2668 chain_index
= nla_get_u32_default(tca
[TCA_CHAIN
], 0);
2669 if (chain_index
> TC_ACT_EXT_VAL_MASK
) {
2670 NL_SET_ERR_MSG(extack
, "Specified chain index exceeds upper limit");
2674 chain
= tcf_chain_get(block
, chain_index
, false);
2676 NL_SET_ERR_MSG(extack
, "Cannot find specified filter chain");
2681 mutex_lock(&chain
->filter_chain_lock
);
2682 tp
= tcf_chain_tp_find(chain
, &chain_info
, protocol
,
2683 prio
, false, extack
);
2684 mutex_unlock(&chain
->filter_chain_lock
);
2687 NL_SET_ERR_MSG(extack
, "Filter with specified priority/protocol not found");
2689 } else if (IS_ERR(tp
)) {
2692 } else if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], tp
->ops
->kind
)) {
2693 NL_SET_ERR_MSG(extack
, "Specified filter kind does not match existing one");
2698 fh
= tp
->ops
->get(tp
, t
->tcm_handle
);
2701 NL_SET_ERR_MSG(extack
, "Specified filter handle not found");
2704 err
= tfilter_notify(net
, skb
, n
, tp
, block
, q
, parent
,
2705 fh
, RTM_NEWTFILTER
, true, rtnl_held
, NULL
);
2707 NL_SET_ERR_MSG(extack
, "Failed to send filter notify message");
2710 tfilter_put(tp
, fh
);
2713 if (tp
&& !IS_ERR(tp
))
2714 tcf_proto_put(tp
, rtnl_held
, NULL
);
2715 tcf_chain_put(chain
);
2717 tcf_block_release(q
, block
, rtnl_held
);
2725 struct tcf_dump_args
{
2726 struct tcf_walker w
;
2727 struct sk_buff
*skb
;
2728 struct netlink_callback
*cb
;
2729 struct tcf_block
*block
;
2735 static int tcf_node_dump(struct tcf_proto
*tp
, void *n
, struct tcf_walker
*arg
)
2737 struct tcf_dump_args
*a
= (void *)arg
;
2738 struct net
*net
= sock_net(a
->skb
->sk
);
2740 return tcf_fill_node(net
, a
->skb
, tp
, a
->block
, a
->q
, a
->parent
,
2741 n
, NETLINK_CB(a
->cb
->skb
).portid
,
2742 a
->cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
2743 RTM_NEWTFILTER
, a
->terse_dump
, true, NULL
);
2746 static bool tcf_chain_dump(struct tcf_chain
*chain
, struct Qdisc
*q
, u32 parent
,
2747 struct sk_buff
*skb
, struct netlink_callback
*cb
,
2748 long index_start
, long *p_index
, bool terse
)
2750 struct net
*net
= sock_net(skb
->sk
);
2751 struct tcf_block
*block
= chain
->block
;
2752 struct tcmsg
*tcm
= nlmsg_data(cb
->nlh
);
2753 struct tcf_proto
*tp
, *tp_prev
;
2754 struct tcf_dump_args arg
;
2756 for (tp
= __tcf_get_next_proto(chain
, NULL
);
2759 tp
= __tcf_get_next_proto(chain
, tp
),
2760 tcf_proto_put(tp_prev
, true, NULL
),
2762 if (*p_index
< index_start
)
2764 if (TC_H_MAJ(tcm
->tcm_info
) &&
2765 TC_H_MAJ(tcm
->tcm_info
) != tp
->prio
)
2767 if (TC_H_MIN(tcm
->tcm_info
) &&
2768 TC_H_MIN(tcm
->tcm_info
) != tp
->protocol
)
2770 if (*p_index
> index_start
)
2771 memset(&cb
->args
[1], 0,
2772 sizeof(cb
->args
) - sizeof(cb
->args
[0]));
2773 if (cb
->args
[1] == 0) {
2774 if (tcf_fill_node(net
, skb
, tp
, block
, q
, parent
, NULL
,
2775 NETLINK_CB(cb
->skb
).portid
,
2776 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
2777 RTM_NEWTFILTER
, false, true, NULL
) <= 0)
2783 arg
.w
.fn
= tcf_node_dump
;
2788 arg
.parent
= parent
;
2790 arg
.w
.skip
= cb
->args
[1] - 1;
2792 arg
.w
.cookie
= cb
->args
[2];
2793 arg
.terse_dump
= terse
;
2794 tp
->ops
->walk(tp
, &arg
.w
, true);
2795 cb
->args
[2] = arg
.w
.cookie
;
2796 cb
->args
[1] = arg
.w
.count
+ 1;
2803 tcf_proto_put(tp
, true, NULL
);
2807 static const struct nla_policy tcf_tfilter_dump_policy
[TCA_MAX
+ 1] = {
2808 [TCA_CHAIN
] = { .type
= NLA_U32
},
2809 [TCA_DUMP_FLAGS
] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE
),
2812 /* called with RTNL */
2813 static int tc_dump_tfilter(struct sk_buff
*skb
, struct netlink_callback
*cb
)
2815 struct tcf_chain
*chain
, *chain_prev
;
2816 struct net
*net
= sock_net(skb
->sk
);
2817 struct nlattr
*tca
[TCA_MAX
+ 1];
2818 struct Qdisc
*q
= NULL
;
2819 struct tcf_block
*block
;
2820 struct tcmsg
*tcm
= nlmsg_data(cb
->nlh
);
2821 bool terse_dump
= false;
2827 if (nlmsg_len(cb
->nlh
) < sizeof(*tcm
))
2830 err
= nlmsg_parse_deprecated(cb
->nlh
, sizeof(*tcm
), tca
, TCA_MAX
,
2831 tcf_tfilter_dump_policy
, cb
->extack
);
2835 if (tca
[TCA_DUMP_FLAGS
]) {
2836 struct nla_bitfield32 flags
=
2837 nla_get_bitfield32(tca
[TCA_DUMP_FLAGS
]);
2839 terse_dump
= flags
.value
& TCA_DUMP_FLAGS_TERSE
;
2842 if (tcm
->tcm_ifindex
== TCM_IFINDEX_MAGIC_BLOCK
) {
2843 block
= tcf_block_refcnt_get(net
, tcm
->tcm_block_index
);
2846 /* If we work with block index, q is NULL and parent value
2847 * will never be used in the following code. The check
2848 * in tcf_fill_node prevents it. However, compiler does not
2849 * see that far, so set parent to zero to silence the warning
2850 * about parent being uninitialized.
2854 const struct Qdisc_class_ops
*cops
;
2855 struct net_device
*dev
;
2856 unsigned long cl
= 0;
2858 dev
= __dev_get_by_index(net
, tcm
->tcm_ifindex
);
2862 parent
= tcm
->tcm_parent
;
2864 q
= rtnl_dereference(dev
->qdisc
);
2866 q
= qdisc_lookup(dev
, TC_H_MAJ(tcm
->tcm_parent
));
2869 cops
= q
->ops
->cl_ops
;
2872 if (!cops
->tcf_block
)
2874 if (TC_H_MIN(tcm
->tcm_parent
)) {
2875 cl
= cops
->find(q
, tcm
->tcm_parent
);
2879 block
= cops
->tcf_block(q
, cl
, NULL
);
2882 parent
= block
->classid
;
2883 if (tcf_block_shared(block
))
2887 index_start
= cb
->args
[0];
2890 for (chain
= __tcf_get_next_chain(block
, NULL
);
2893 chain
= __tcf_get_next_chain(block
, chain
),
2894 tcf_chain_put(chain_prev
)) {
2895 if (tca
[TCA_CHAIN
] &&
2896 nla_get_u32(tca
[TCA_CHAIN
]) != chain
->index
)
2898 if (!tcf_chain_dump(chain
, q
, parent
, skb
, cb
,
2899 index_start
, &index
, terse_dump
)) {
2900 tcf_chain_put(chain
);
2906 if (tcm
->tcm_ifindex
== TCM_IFINDEX_MAGIC_BLOCK
)
2907 tcf_block_refcnt_put(block
, true);
2908 cb
->args
[0] = index
;
2911 /* If we did no progress, the error (EMSGSIZE) is real */
2912 if (skb
->len
== 0 && err
)
2917 static int tc_chain_fill_node(const struct tcf_proto_ops
*tmplt_ops
,
2918 void *tmplt_priv
, u32 chain_index
,
2919 struct net
*net
, struct sk_buff
*skb
,
2920 struct tcf_block
*block
,
2921 u32 portid
, u32 seq
, u16 flags
, int event
,
2922 struct netlink_ext_ack
*extack
)
2924 unsigned char *b
= skb_tail_pointer(skb
);
2925 const struct tcf_proto_ops
*ops
;
2926 struct nlmsghdr
*nlh
;
2933 nlh
= nlmsg_put(skb
, portid
, seq
, event
, sizeof(*tcm
), flags
);
2935 goto out_nlmsg_trim
;
2936 tcm
= nlmsg_data(nlh
);
2937 tcm
->tcm_family
= AF_UNSPEC
;
2940 tcm
->tcm_handle
= 0;
2942 tcm
->tcm_ifindex
= qdisc_dev(block
->q
)->ifindex
;
2943 tcm
->tcm_parent
= block
->q
->handle
;
2945 tcm
->tcm_ifindex
= TCM_IFINDEX_MAGIC_BLOCK
;
2946 tcm
->tcm_block_index
= block
->index
;
2949 if (nla_put_u32(skb
, TCA_CHAIN
, chain_index
))
2950 goto nla_put_failure
;
2953 if (nla_put_string(skb
, TCA_KIND
, ops
->kind
))
2954 goto nla_put_failure
;
2955 if (ops
->tmplt_dump(skb
, net
, priv
) < 0)
2956 goto nla_put_failure
;
2959 if (extack
&& extack
->_msg
&&
2960 nla_put_string(skb
, TCA_EXT_WARN_MSG
, extack
->_msg
))
2961 goto out_nlmsg_trim
;
2963 nlh
->nlmsg_len
= skb_tail_pointer(skb
) - b
;
2973 static int tc_chain_notify(struct tcf_chain
*chain
, struct sk_buff
*oskb
,
2974 u32 seq
, u16 flags
, int event
, bool unicast
,
2975 struct netlink_ext_ack
*extack
)
2977 u32 portid
= oskb
? NETLINK_CB(oskb
).portid
: 0;
2978 struct tcf_block
*block
= chain
->block
;
2979 struct net
*net
= block
->net
;
2980 struct sk_buff
*skb
;
2983 if (!unicast
&& !rtnl_notify_needed(net
, flags
, RTNLGRP_TC
))
2986 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
2990 if (tc_chain_fill_node(chain
->tmplt_ops
, chain
->tmplt_priv
,
2991 chain
->index
, net
, skb
, block
, portid
,
2992 seq
, flags
, event
, extack
) <= 0) {
2998 err
= rtnl_unicast(skb
, net
, portid
);
3000 err
= rtnetlink_send(skb
, net
, portid
, RTNLGRP_TC
,
3001 flags
& NLM_F_ECHO
);
3006 static int tc_chain_notify_delete(const struct tcf_proto_ops
*tmplt_ops
,
3007 void *tmplt_priv
, u32 chain_index
,
3008 struct tcf_block
*block
, struct sk_buff
*oskb
,
3011 u32 portid
= oskb
? NETLINK_CB(oskb
).portid
: 0;
3012 struct net
*net
= block
->net
;
3013 struct sk_buff
*skb
;
3015 if (!rtnl_notify_needed(net
, flags
, RTNLGRP_TC
))
3018 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
3022 if (tc_chain_fill_node(tmplt_ops
, tmplt_priv
, chain_index
, net
, skb
,
3023 block
, portid
, seq
, flags
, RTM_DELCHAIN
, NULL
) <= 0) {
3028 return rtnetlink_send(skb
, net
, portid
, RTNLGRP_TC
, flags
& NLM_F_ECHO
);
3031 static int tc_chain_tmplt_add(struct tcf_chain
*chain
, struct net
*net
,
3032 struct nlattr
**tca
,
3033 struct netlink_ext_ack
*extack
)
3035 const struct tcf_proto_ops
*ops
;
3036 char name
[IFNAMSIZ
];
3039 /* If kind is not set, user did not specify template. */
3043 if (tcf_proto_check_kind(tca
[TCA_KIND
], name
)) {
3044 NL_SET_ERR_MSG(extack
, "Specified TC chain template name too long");
3048 ops
= tcf_proto_lookup_ops(name
, true, extack
);
3050 return PTR_ERR(ops
);
3051 if (!ops
->tmplt_create
|| !ops
->tmplt_destroy
|| !ops
->tmplt_dump
||
3052 !ops
->tmplt_reoffload
) {
3053 NL_SET_ERR_MSG(extack
, "Chain templates are not supported with specified classifier");
3054 module_put(ops
->owner
);
3058 tmplt_priv
= ops
->tmplt_create(net
, chain
, tca
, extack
);
3059 if (IS_ERR(tmplt_priv
)) {
3060 module_put(ops
->owner
);
3061 return PTR_ERR(tmplt_priv
);
3063 chain
->tmplt_ops
= ops
;
3064 chain
->tmplt_priv
= tmplt_priv
;
3068 static void tc_chain_tmplt_del(const struct tcf_proto_ops
*tmplt_ops
,
3071 /* If template ops are set, no work to do for us. */
3075 tmplt_ops
->tmplt_destroy(tmplt_priv
);
3076 module_put(tmplt_ops
->owner
);
3079 /* Add/delete/get a chain */
3081 static int tc_ctl_chain(struct sk_buff
*skb
, struct nlmsghdr
*n
,
3082 struct netlink_ext_ack
*extack
)
3084 struct net
*net
= sock_net(skb
->sk
);
3085 struct nlattr
*tca
[TCA_MAX
+ 1];
3090 struct tcf_chain
*chain
;
3091 struct tcf_block
*block
;
3097 err
= nlmsg_parse_deprecated(n
, sizeof(*t
), tca
, TCA_MAX
,
3098 rtm_tca_policy
, extack
);
3103 parent
= t
->tcm_parent
;
3106 block
= tcf_block_find(net
, &q
, &parent
, &cl
,
3107 t
->tcm_ifindex
, t
->tcm_block_index
, extack
);
3109 return PTR_ERR(block
);
3111 chain_index
= nla_get_u32_default(tca
[TCA_CHAIN
], 0);
3112 if (chain_index
> TC_ACT_EXT_VAL_MASK
) {
3113 NL_SET_ERR_MSG(extack
, "Specified chain index exceeds upper limit");
3118 mutex_lock(&block
->lock
);
3119 chain
= tcf_chain_lookup(block
, chain_index
);
3120 if (n
->nlmsg_type
== RTM_NEWCHAIN
) {
3122 if (tcf_chain_held_by_acts_only(chain
)) {
3123 /* The chain exists only because there is
3124 * some action referencing it.
3126 tcf_chain_hold(chain
);
3128 NL_SET_ERR_MSG(extack
, "Filter chain already exists");
3130 goto errout_block_locked
;
3133 if (!(n
->nlmsg_flags
& NLM_F_CREATE
)) {
3134 NL_SET_ERR_MSG(extack
, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
3136 goto errout_block_locked
;
3138 chain
= tcf_chain_create(block
, chain_index
);
3140 NL_SET_ERR_MSG(extack
, "Failed to create filter chain");
3142 goto errout_block_locked
;
3146 if (!chain
|| tcf_chain_held_by_acts_only(chain
)) {
3147 NL_SET_ERR_MSG(extack
, "Cannot find specified filter chain");
3149 goto errout_block_locked
;
3151 tcf_chain_hold(chain
);
3154 if (n
->nlmsg_type
== RTM_NEWCHAIN
) {
3155 /* Modifying chain requires holding parent block lock. In case
3156 * the chain was successfully added, take a reference to the
3157 * chain. This ensures that an empty chain does not disappear at
3158 * the end of this function.
3160 tcf_chain_hold(chain
);
3161 chain
->explicitly_created
= true;
3163 mutex_unlock(&block
->lock
);
3165 switch (n
->nlmsg_type
) {
3167 err
= tc_chain_tmplt_add(chain
, net
, tca
, extack
);
3169 tcf_chain_put_explicitly_created(chain
);
3173 tc_chain_notify(chain
, NULL
, 0, NLM_F_CREATE
| NLM_F_EXCL
,
3174 RTM_NEWCHAIN
, false, extack
);
3177 tfilter_notify_chain(net
, skb
, block
, q
, parent
, n
,
3178 chain
, RTM_DELTFILTER
, extack
);
3179 /* Flush the chain first as the user requested chain removal. */
3180 tcf_chain_flush(chain
, true);
3181 /* In case the chain was successfully deleted, put a reference
3182 * to the chain previously taken during addition.
3184 tcf_chain_put_explicitly_created(chain
);
3187 err
= tc_chain_notify(chain
, skb
, n
->nlmsg_seq
,
3188 n
->nlmsg_flags
, n
->nlmsg_type
, true, extack
);
3190 NL_SET_ERR_MSG(extack
, "Failed to send chain notify message");
3194 NL_SET_ERR_MSG(extack
, "Unsupported message type");
3199 tcf_chain_put(chain
);
3201 tcf_block_release(q
, block
, true);
3203 /* Replay the request. */
3207 errout_block_locked
:
3208 mutex_unlock(&block
->lock
);
3212 /* called with RTNL */
3213 static int tc_dump_chain(struct sk_buff
*skb
, struct netlink_callback
*cb
)
3215 struct net
*net
= sock_net(skb
->sk
);
3216 struct nlattr
*tca
[TCA_MAX
+ 1];
3217 struct Qdisc
*q
= NULL
;
3218 struct tcf_block
*block
;
3219 struct tcmsg
*tcm
= nlmsg_data(cb
->nlh
);
3220 struct tcf_chain
*chain
;
3225 if (nlmsg_len(cb
->nlh
) < sizeof(*tcm
))
3228 err
= nlmsg_parse_deprecated(cb
->nlh
, sizeof(*tcm
), tca
, TCA_MAX
,
3229 rtm_tca_policy
, cb
->extack
);
3233 if (tcm
->tcm_ifindex
== TCM_IFINDEX_MAGIC_BLOCK
) {
3234 block
= tcf_block_refcnt_get(net
, tcm
->tcm_block_index
);
3238 const struct Qdisc_class_ops
*cops
;
3239 struct net_device
*dev
;
3240 unsigned long cl
= 0;
3242 dev
= __dev_get_by_index(net
, tcm
->tcm_ifindex
);
3246 if (!tcm
->tcm_parent
)
3247 q
= rtnl_dereference(dev
->qdisc
);
3249 q
= qdisc_lookup(dev
, TC_H_MAJ(tcm
->tcm_parent
));
3253 cops
= q
->ops
->cl_ops
;
3256 if (!cops
->tcf_block
)
3258 if (TC_H_MIN(tcm
->tcm_parent
)) {
3259 cl
= cops
->find(q
, tcm
->tcm_parent
);
3263 block
= cops
->tcf_block(q
, cl
, NULL
);
3266 if (tcf_block_shared(block
))
3270 index_start
= cb
->args
[0];
3273 mutex_lock(&block
->lock
);
3274 list_for_each_entry(chain
, &block
->chain_list
, list
) {
3275 if ((tca
[TCA_CHAIN
] &&
3276 nla_get_u32(tca
[TCA_CHAIN
]) != chain
->index
))
3278 if (index
< index_start
) {
3282 if (tcf_chain_held_by_acts_only(chain
))
3284 err
= tc_chain_fill_node(chain
->tmplt_ops
, chain
->tmplt_priv
,
3285 chain
->index
, net
, skb
, block
,
3286 NETLINK_CB(cb
->skb
).portid
,
3287 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
3288 RTM_NEWCHAIN
, NULL
);
3293 mutex_unlock(&block
->lock
);
3295 if (tcm
->tcm_ifindex
== TCM_IFINDEX_MAGIC_BLOCK
)
3296 tcf_block_refcnt_put(block
, true);
3297 cb
->args
[0] = index
;
3300 /* If we did no progress, the error (EMSGSIZE) is real */
3301 if (skb
->len
== 0 && err
)
3306 int tcf_exts_init_ex(struct tcf_exts
*exts
, struct net
*net
, int action
,
3307 int police
, struct tcf_proto
*tp
, u32 handle
,
3308 bool use_action_miss
)
3312 #ifdef CONFIG_NET_CLS_ACT
3314 exts
->nr_actions
= 0;
3315 exts
->miss_cookie_node
= NULL
;
3316 /* Note: we do not own yet a reference on net.
3317 * This reference might be taken later from tcf_exts_get_net().
3320 exts
->actions
= kcalloc(TCA_ACT_MAX_PRIO
, sizeof(struct tc_action
*),
3326 exts
->action
= action
;
3327 exts
->police
= police
;
3329 if (!use_action_miss
)
3332 err
= tcf_exts_miss_cookie_base_alloc(exts
, tp
, handle
);
3334 goto err_miss_alloc
;
3339 tcf_exts_destroy(exts
);
3340 #ifdef CONFIG_NET_CLS_ACT
3341 exts
->actions
= NULL
;
3345 EXPORT_SYMBOL(tcf_exts_init_ex
);
3347 void tcf_exts_destroy(struct tcf_exts
*exts
)
3349 tcf_exts_miss_cookie_base_destroy(exts
);
3351 #ifdef CONFIG_NET_CLS_ACT
3352 if (exts
->actions
) {
3353 tcf_action_destroy(exts
->actions
, TCA_ACT_UNBIND
);
3354 kfree(exts
->actions
);
3356 exts
->nr_actions
= 0;
3359 EXPORT_SYMBOL(tcf_exts_destroy
);
3361 int tcf_exts_validate_ex(struct net
*net
, struct tcf_proto
*tp
, struct nlattr
**tb
,
3362 struct nlattr
*rate_tlv
, struct tcf_exts
*exts
,
3363 u32 flags
, u32 fl_flags
, struct netlink_ext_ack
*extack
)
3365 #ifdef CONFIG_NET_CLS_ACT
3367 int init_res
[TCA_ACT_MAX_PRIO
] = {};
3368 struct tc_action
*act
;
3369 size_t attr_size
= 0;
3371 if (exts
->police
&& tb
[exts
->police
]) {
3372 struct tc_action_ops
*a_o
;
3374 flags
|= TCA_ACT_FLAGS_POLICE
| TCA_ACT_FLAGS_BIND
;
3375 a_o
= tc_action_load_ops(tb
[exts
->police
], flags
,
3378 return PTR_ERR(a_o
);
3379 act
= tcf_action_init_1(net
, tp
, tb
[exts
->police
],
3380 rate_tlv
, a_o
, init_res
, flags
,
3382 module_put(a_o
->owner
);
3384 return PTR_ERR(act
);
3386 act
->type
= exts
->type
= TCA_OLD_COMPAT
;
3387 exts
->actions
[0] = act
;
3388 exts
->nr_actions
= 1;
3389 tcf_idr_insert_many(exts
->actions
, init_res
);
3390 } else if (exts
->action
&& tb
[exts
->action
]) {
3393 flags
|= TCA_ACT_FLAGS_BIND
;
3394 err
= tcf_action_init(net
, tp
, tb
[exts
->action
],
3395 rate_tlv
, exts
->actions
, init_res
,
3396 &attr_size
, flags
, fl_flags
,
3400 exts
->nr_actions
= err
;
3404 if ((exts
->action
&& tb
[exts
->action
]) ||
3405 (exts
->police
&& tb
[exts
->police
])) {
3406 NL_SET_ERR_MSG(extack
, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3413 EXPORT_SYMBOL(tcf_exts_validate_ex
);
3415 int tcf_exts_validate(struct net
*net
, struct tcf_proto
*tp
, struct nlattr
**tb
,
3416 struct nlattr
*rate_tlv
, struct tcf_exts
*exts
,
3417 u32 flags
, struct netlink_ext_ack
*extack
)
3419 return tcf_exts_validate_ex(net
, tp
, tb
, rate_tlv
, exts
,
3422 EXPORT_SYMBOL(tcf_exts_validate
);
3424 void tcf_exts_change(struct tcf_exts
*dst
, struct tcf_exts
*src
)
3426 #ifdef CONFIG_NET_CLS_ACT
3427 struct tcf_exts old
= *dst
;
3430 tcf_exts_destroy(&old
);
3433 EXPORT_SYMBOL(tcf_exts_change
);
3435 #ifdef CONFIG_NET_CLS_ACT
3436 static struct tc_action
*tcf_exts_first_act(struct tcf_exts
*exts
)
3438 if (exts
->nr_actions
== 0)
3441 return exts
->actions
[0];
3445 int tcf_exts_dump(struct sk_buff
*skb
, struct tcf_exts
*exts
)
3447 #ifdef CONFIG_NET_CLS_ACT
3448 struct nlattr
*nest
;
3450 if (exts
->action
&& tcf_exts_has_actions(exts
)) {
3452 * again for backward compatible mode - we want
3453 * to work with both old and new modes of entering
3454 * tc data even if iproute2 was newer - jhs
3456 if (exts
->type
!= TCA_OLD_COMPAT
) {
3457 nest
= nla_nest_start_noflag(skb
, exts
->action
);
3459 goto nla_put_failure
;
3461 if (tcf_action_dump(skb
, exts
->actions
, 0, 0, false)
3463 goto nla_put_failure
;
3464 nla_nest_end(skb
, nest
);
3465 } else if (exts
->police
) {
3466 struct tc_action
*act
= tcf_exts_first_act(exts
);
3467 nest
= nla_nest_start_noflag(skb
, exts
->police
);
3468 if (nest
== NULL
|| !act
)
3469 goto nla_put_failure
;
3470 if (tcf_action_dump_old(skb
, act
, 0, 0) < 0)
3471 goto nla_put_failure
;
3472 nla_nest_end(skb
, nest
);
3478 nla_nest_cancel(skb
, nest
);
3484 EXPORT_SYMBOL(tcf_exts_dump
);
3486 int tcf_exts_terse_dump(struct sk_buff
*skb
, struct tcf_exts
*exts
)
3488 #ifdef CONFIG_NET_CLS_ACT
3489 struct nlattr
*nest
;
3491 if (!exts
->action
|| !tcf_exts_has_actions(exts
))
3494 nest
= nla_nest_start_noflag(skb
, exts
->action
);
3496 goto nla_put_failure
;
3498 if (tcf_action_dump(skb
, exts
->actions
, 0, 0, true) < 0)
3499 goto nla_put_failure
;
3500 nla_nest_end(skb
, nest
);
3504 nla_nest_cancel(skb
, nest
);
3510 EXPORT_SYMBOL(tcf_exts_terse_dump
);
3512 int tcf_exts_dump_stats(struct sk_buff
*skb
, struct tcf_exts
*exts
)
3514 #ifdef CONFIG_NET_CLS_ACT
3515 struct tc_action
*a
= tcf_exts_first_act(exts
);
3516 if (a
!= NULL
&& tcf_action_copy_stats(skb
, a
, 1) < 0)
3521 EXPORT_SYMBOL(tcf_exts_dump_stats
);
3523 static void tcf_block_offload_inc(struct tcf_block
*block
, u32
*flags
)
3525 if (*flags
& TCA_CLS_FLAGS_IN_HW
)
3527 *flags
|= TCA_CLS_FLAGS_IN_HW
;
3528 atomic_inc(&block
->offloadcnt
);
3531 static void tcf_block_offload_dec(struct tcf_block
*block
, u32
*flags
)
3533 if (!(*flags
& TCA_CLS_FLAGS_IN_HW
))
3535 *flags
&= ~TCA_CLS_FLAGS_IN_HW
;
3536 atomic_dec(&block
->offloadcnt
);
3539 static void tc_cls_offload_cnt_update(struct tcf_block
*block
,
3540 struct tcf_proto
*tp
, u32
*cnt
,
3541 u32
*flags
, u32 diff
, bool add
)
3543 lockdep_assert_held(&block
->cb_lock
);
3545 spin_lock(&tp
->lock
);
3548 tcf_block_offload_inc(block
, flags
);
3553 tcf_block_offload_dec(block
, flags
);
3555 spin_unlock(&tp
->lock
);
3559 tc_cls_offload_cnt_reset(struct tcf_block
*block
, struct tcf_proto
*tp
,
3560 u32
*cnt
, u32
*flags
)
3562 lockdep_assert_held(&block
->cb_lock
);
3564 spin_lock(&tp
->lock
);
3565 tcf_block_offload_dec(block
, flags
);
3567 spin_unlock(&tp
->lock
);
3571 __tc_setup_cb_call(struct tcf_block
*block
, enum tc_setup_type type
,
3572 void *type_data
, bool err_stop
)
3574 struct flow_block_cb
*block_cb
;
3578 list_for_each_entry(block_cb
, &block
->flow_block
.cb_list
, list
) {
3579 err
= block_cb
->cb(type
, type_data
, block_cb
->cb_priv
);
3590 int tc_setup_cb_call(struct tcf_block
*block
, enum tc_setup_type type
,
3591 void *type_data
, bool err_stop
, bool rtnl_held
)
3593 bool take_rtnl
= READ_ONCE(block
->lockeddevcnt
) && !rtnl_held
;
3599 down_read(&block
->cb_lock
);
3600 /* Need to obtain rtnl lock if block is bound to devs that require it.
3601 * In block bind code cb_lock is obtained while holding rtnl, so we must
3602 * obtain the locks in same order here.
3604 if (!rtnl_held
&& !take_rtnl
&& block
->lockeddevcnt
) {
3605 up_read(&block
->cb_lock
);
3610 ok_count
= __tc_setup_cb_call(block
, type
, type_data
, err_stop
);
3612 up_read(&block
->cb_lock
);
3617 EXPORT_SYMBOL(tc_setup_cb_call
);
3619 /* Non-destructive filter add. If filter that wasn't already in hardware is
3620 * successfully offloaded, increment block offloads counter. On failure,
3621 * previously offloaded filter is considered to be intact and offloads counter
3622 * is not decremented.
3625 int tc_setup_cb_add(struct tcf_block
*block
, struct tcf_proto
*tp
,
3626 enum tc_setup_type type
, void *type_data
, bool err_stop
,
3627 u32
*flags
, unsigned int *in_hw_count
, bool rtnl_held
)
3629 bool take_rtnl
= READ_ONCE(block
->lockeddevcnt
) && !rtnl_held
;
3635 down_read(&block
->cb_lock
);
3636 /* Need to obtain rtnl lock if block is bound to devs that require it.
3637 * In block bind code cb_lock is obtained while holding rtnl, so we must
3638 * obtain the locks in same order here.
3640 if (!rtnl_held
&& !take_rtnl
&& block
->lockeddevcnt
) {
3641 up_read(&block
->cb_lock
);
3646 /* Make sure all netdevs sharing this block are offload-capable. */
3647 if (block
->nooffloaddevcnt
&& err_stop
) {
3648 ok_count
= -EOPNOTSUPP
;
3652 ok_count
= __tc_setup_cb_call(block
, type
, type_data
, err_stop
);
3656 if (tp
->ops
->hw_add
)
3657 tp
->ops
->hw_add(tp
, type_data
);
3659 tc_cls_offload_cnt_update(block
, tp
, in_hw_count
, flags
,
3662 up_read(&block
->cb_lock
);
3665 return min(ok_count
, 0);
3667 EXPORT_SYMBOL(tc_setup_cb_add
);
3669 /* Destructive filter replace. If filter that wasn't already in hardware is
3670 * successfully offloaded, increment block offload counter. On failure,
3671 * previously offloaded filter is considered to be destroyed and offload counter
3675 int tc_setup_cb_replace(struct tcf_block
*block
, struct tcf_proto
*tp
,
3676 enum tc_setup_type type
, void *type_data
, bool err_stop
,
3677 u32
*old_flags
, unsigned int *old_in_hw_count
,
3678 u32
*new_flags
, unsigned int *new_in_hw_count
,
3681 bool take_rtnl
= READ_ONCE(block
->lockeddevcnt
) && !rtnl_held
;
3687 down_read(&block
->cb_lock
);
3688 /* Need to obtain rtnl lock if block is bound to devs that require it.
3689 * In block bind code cb_lock is obtained while holding rtnl, so we must
3690 * obtain the locks in same order here.
3692 if (!rtnl_held
&& !take_rtnl
&& block
->lockeddevcnt
) {
3693 up_read(&block
->cb_lock
);
3698 /* Make sure all netdevs sharing this block are offload-capable. */
3699 if (block
->nooffloaddevcnt
&& err_stop
) {
3700 ok_count
= -EOPNOTSUPP
;
3704 tc_cls_offload_cnt_reset(block
, tp
, old_in_hw_count
, old_flags
);
3705 if (tp
->ops
->hw_del
)
3706 tp
->ops
->hw_del(tp
, type_data
);
3708 ok_count
= __tc_setup_cb_call(block
, type
, type_data
, err_stop
);
3712 if (tp
->ops
->hw_add
)
3713 tp
->ops
->hw_add(tp
, type_data
);
3715 tc_cls_offload_cnt_update(block
, tp
, new_in_hw_count
,
3716 new_flags
, ok_count
, true);
3718 up_read(&block
->cb_lock
);
3721 return min(ok_count
, 0);
3723 EXPORT_SYMBOL(tc_setup_cb_replace
);
3725 /* Destroy filter and decrement block offload counter, if filter was previously
3729 int tc_setup_cb_destroy(struct tcf_block
*block
, struct tcf_proto
*tp
,
3730 enum tc_setup_type type
, void *type_data
, bool err_stop
,
3731 u32
*flags
, unsigned int *in_hw_count
, bool rtnl_held
)
3733 bool take_rtnl
= READ_ONCE(block
->lockeddevcnt
) && !rtnl_held
;
3739 down_read(&block
->cb_lock
);
3740 /* Need to obtain rtnl lock if block is bound to devs that require it.
3741 * In block bind code cb_lock is obtained while holding rtnl, so we must
3742 * obtain the locks in same order here.
3744 if (!rtnl_held
&& !take_rtnl
&& block
->lockeddevcnt
) {
3745 up_read(&block
->cb_lock
);
3750 ok_count
= __tc_setup_cb_call(block
, type
, type_data
, err_stop
);
3752 tc_cls_offload_cnt_reset(block
, tp
, in_hw_count
, flags
);
3753 if (tp
->ops
->hw_del
)
3754 tp
->ops
->hw_del(tp
, type_data
);
3756 up_read(&block
->cb_lock
);
3759 return min(ok_count
, 0);
3761 EXPORT_SYMBOL(tc_setup_cb_destroy
);
3763 int tc_setup_cb_reoffload(struct tcf_block
*block
, struct tcf_proto
*tp
,
3764 bool add
, flow_setup_cb_t
*cb
,
3765 enum tc_setup_type type
, void *type_data
,
3766 void *cb_priv
, u32
*flags
, unsigned int *in_hw_count
)
3768 int err
= cb(type
, type_data
, cb_priv
);
3771 if (add
&& tc_skip_sw(*flags
))
3774 tc_cls_offload_cnt_update(block
, tp
, in_hw_count
, flags
, 1,
3780 EXPORT_SYMBOL(tc_setup_cb_reoffload
);
3782 static int tcf_act_get_user_cookie(struct flow_action_entry
*entry
,
3783 const struct tc_action
*act
)
3785 struct tc_cookie
*user_cookie
;
3789 user_cookie
= rcu_dereference(act
->user_cookie
);
3791 entry
->user_cookie
= flow_action_cookie_create(user_cookie
->data
,
3794 if (!entry
->user_cookie
)
3801 static void tcf_act_put_user_cookie(struct flow_action_entry
*entry
)
3803 flow_action_cookie_destroy(entry
->user_cookie
);
3806 void tc_cleanup_offload_action(struct flow_action
*flow_action
)
3808 struct flow_action_entry
*entry
;
3811 flow_action_for_each(i
, entry
, flow_action
) {
3812 tcf_act_put_user_cookie(entry
);
3813 if (entry
->destructor
)
3814 entry
->destructor(entry
->destructor_priv
);
3817 EXPORT_SYMBOL(tc_cleanup_offload_action
);
3819 static int tc_setup_offload_act(struct tc_action
*act
,
3820 struct flow_action_entry
*entry
,
3822 struct netlink_ext_ack
*extack
)
3824 #ifdef CONFIG_NET_CLS_ACT
3825 if (act
->ops
->offload_act_setup
) {
3826 return act
->ops
->offload_act_setup(act
, entry
, index_inc
, true,
3829 NL_SET_ERR_MSG(extack
, "Action does not support offload");
3837 int tc_setup_action(struct flow_action
*flow_action
,
3838 struct tc_action
*actions
[],
3839 u32 miss_cookie_base
,
3840 struct netlink_ext_ack
*extack
)
3842 int i
, j
, k
, index
, err
= 0;
3843 struct tc_action
*act
;
3845 BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY
!= FLOW_ACTION_HW_STATS_ANY
);
3846 BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE
!= FLOW_ACTION_HW_STATS_IMMEDIATE
);
3847 BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED
!= FLOW_ACTION_HW_STATS_DELAYED
);
3853 tcf_act_for_each_action(i
, act
, actions
) {
3854 struct flow_action_entry
*entry
;
3856 entry
= &flow_action
->entries
[j
];
3857 spin_lock_bh(&act
->tcfa_lock
);
3858 err
= tcf_act_get_user_cookie(entry
, act
);
3860 goto err_out_locked
;
3863 err
= tc_setup_offload_act(act
, entry
, &index
, extack
);
3865 goto err_out_locked
;
3867 for (k
= 0; k
< index
; k
++) {
3868 entry
[k
].hw_stats
= tc_act_hw_stats(act
->hw_stats
);
3869 entry
[k
].hw_index
= act
->tcfa_index
;
3870 entry
[k
].cookie
= (unsigned long)act
;
3871 entry
[k
].miss_cookie
=
3872 tcf_exts_miss_cookie_get(miss_cookie_base
, i
);
3877 spin_unlock_bh(&act
->tcfa_lock
);
3882 tc_cleanup_offload_action(flow_action
);
3886 spin_unlock_bh(&act
->tcfa_lock
);
3890 int tc_setup_offload_action(struct flow_action
*flow_action
,
3891 const struct tcf_exts
*exts
,
3892 struct netlink_ext_ack
*extack
)
3894 #ifdef CONFIG_NET_CLS_ACT
3895 u32 miss_cookie_base
;
3900 miss_cookie_base
= exts
->miss_cookie_node
?
3901 exts
->miss_cookie_node
->miss_cookie_base
: 0;
3902 return tc_setup_action(flow_action
, exts
->actions
, miss_cookie_base
,
3908 EXPORT_SYMBOL(tc_setup_offload_action
);
3910 unsigned int tcf_exts_num_actions(struct tcf_exts
*exts
)
3912 unsigned int num_acts
= 0;
3913 struct tc_action
*act
;
3916 tcf_exts_for_each_action(i
, act
, exts
) {
3917 if (is_tcf_pedit(act
))
3918 num_acts
+= tcf_pedit_nkeys(act
);
3924 EXPORT_SYMBOL(tcf_exts_num_actions
);
3926 #ifdef CONFIG_NET_CLS_ACT
3927 static int tcf_qevent_parse_block_index(struct nlattr
*block_index_attr
,
3929 struct netlink_ext_ack
*extack
)
3931 *p_block_index
= nla_get_u32(block_index_attr
);
3932 if (!*p_block_index
) {
3933 NL_SET_ERR_MSG(extack
, "Block number may not be zero");
3940 int tcf_qevent_init(struct tcf_qevent
*qe
, struct Qdisc
*sch
,
3941 enum flow_block_binder_type binder_type
,
3942 struct nlattr
*block_index_attr
,
3943 struct netlink_ext_ack
*extack
)
3948 if (!block_index_attr
)
3951 err
= tcf_qevent_parse_block_index(block_index_attr
, &block_index
, extack
);
3955 qe
->info
.binder_type
= binder_type
;
3956 qe
->info
.chain_head_change
= tcf_chain_head_change_dflt
;
3957 qe
->info
.chain_head_change_priv
= &qe
->filter_chain
;
3958 qe
->info
.block_index
= block_index
;
3960 return tcf_block_get_ext(&qe
->block
, sch
, &qe
->info
, extack
);
3962 EXPORT_SYMBOL(tcf_qevent_init
);
3964 void tcf_qevent_destroy(struct tcf_qevent
*qe
, struct Qdisc
*sch
)
3966 if (qe
->info
.block_index
)
3967 tcf_block_put_ext(qe
->block
, sch
, &qe
->info
);
3969 EXPORT_SYMBOL(tcf_qevent_destroy
);
3971 int tcf_qevent_validate_change(struct tcf_qevent
*qe
, struct nlattr
*block_index_attr
,
3972 struct netlink_ext_ack
*extack
)
3977 if (!block_index_attr
)
3980 err
= tcf_qevent_parse_block_index(block_index_attr
, &block_index
, extack
);
3984 /* Bounce newly-configured block or change in block. */
3985 if (block_index
!= qe
->info
.block_index
) {
3986 NL_SET_ERR_MSG(extack
, "Change of blocks is not supported");
3992 EXPORT_SYMBOL(tcf_qevent_validate_change
);
3994 struct sk_buff
*tcf_qevent_handle(struct tcf_qevent
*qe
, struct Qdisc
*sch
, struct sk_buff
*skb
,
3995 struct sk_buff
**to_free
, int *ret
)
3997 struct tcf_result cl_res
;
3998 struct tcf_proto
*fl
;
4000 if (!qe
->info
.block_index
)
4003 fl
= rcu_dereference_bh(qe
->filter_chain
);
4005 switch (tcf_classify(skb
, NULL
, fl
, &cl_res
, false)) {
4007 qdisc_qstats_drop(sch
);
4008 __qdisc_drop(skb
, to_free
);
4009 *ret
= __NET_XMIT_BYPASS
;
4014 __qdisc_drop(skb
, to_free
);
4015 *ret
= __NET_XMIT_STOLEN
;
4017 case TC_ACT_REDIRECT
:
4018 skb_do_redirect(skb
);
4019 *ret
= __NET_XMIT_STOLEN
;
4025 EXPORT_SYMBOL(tcf_qevent_handle
);
4027 int tcf_qevent_dump(struct sk_buff
*skb
, int attr_name
, struct tcf_qevent
*qe
)
4029 if (!qe
->info
.block_index
)
4031 return nla_put_u32(skb
, attr_name
, qe
->info
.block_index
);
4033 EXPORT_SYMBOL(tcf_qevent_dump
);
4036 static __net_init
int tcf_net_init(struct net
*net
)
4038 struct tcf_net
*tn
= net_generic(net
, tcf_net_id
);
4040 spin_lock_init(&tn
->idr_lock
);
4045 static void __net_exit
tcf_net_exit(struct net
*net
)
4047 struct tcf_net
*tn
= net_generic(net
, tcf_net_id
);
4049 idr_destroy(&tn
->idr
);
4052 static struct pernet_operations tcf_net_ops
= {
4053 .init
= tcf_net_init
,
4054 .exit
= tcf_net_exit
,
4056 .size
= sizeof(struct tcf_net
),
4059 static const struct rtnl_msg_handler tc_filter_rtnl_msg_handlers
[] __initconst
= {
4060 {.msgtype
= RTM_NEWTFILTER
, .doit
= tc_new_tfilter
,
4061 .flags
= RTNL_FLAG_DOIT_UNLOCKED
},
4062 {.msgtype
= RTM_DELTFILTER
, .doit
= tc_del_tfilter
,
4063 .flags
= RTNL_FLAG_DOIT_UNLOCKED
},
4064 {.msgtype
= RTM_GETTFILTER
, .doit
= tc_get_tfilter
,
4065 .dumpit
= tc_dump_tfilter
, .flags
= RTNL_FLAG_DOIT_UNLOCKED
},
4066 {.msgtype
= RTM_NEWCHAIN
, .doit
= tc_ctl_chain
},
4067 {.msgtype
= RTM_DELCHAIN
, .doit
= tc_ctl_chain
},
4068 {.msgtype
= RTM_GETCHAIN
, .doit
= tc_ctl_chain
,
4069 .dumpit
= tc_dump_chain
},
4072 static int __init
tc_filter_init(void)
4076 tc_filter_wq
= alloc_ordered_workqueue("tc_filter_workqueue", 0);
4080 err
= register_pernet_subsys(&tcf_net_ops
);
4082 goto err_register_pernet_subsys
;
4084 xa_init_flags(&tcf_exts_miss_cookies_xa
, XA_FLAGS_ALLOC1
);
4085 rtnl_register_many(tc_filter_rtnl_msg_handlers
);
4089 err_register_pernet_subsys
:
4090 destroy_workqueue(tc_filter_wq
);
4094 subsys_initcall(tc_filter_init
);