1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * net/sched/cls_api.c Packet classifier API.
5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
9 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
12 #include <linux/module.h>
13 #include <linux/types.h>
14 #include <linux/kernel.h>
15 #include <linux/string.h>
16 #include <linux/errno.h>
17 #include <linux/err.h>
18 #include <linux/skbuff.h>
19 #include <linux/init.h>
20 #include <linux/kmod.h>
21 #include <linux/slab.h>
22 #include <linux/idr.h>
23 #include <linux/rhashtable.h>
24 #include <net/net_namespace.h>
26 #include <net/netlink.h>
27 #include <net/pkt_sched.h>
28 #include <net/pkt_cls.h>
29 #include <net/tc_act/tc_pedit.h>
30 #include <net/tc_act/tc_mirred.h>
31 #include <net/tc_act/tc_vlan.h>
32 #include <net/tc_act/tc_tunnel_key.h>
33 #include <net/tc_act/tc_csum.h>
34 #include <net/tc_act/tc_gact.h>
35 #include <net/tc_act/tc_police.h>
36 #include <net/tc_act/tc_sample.h>
37 #include <net/tc_act/tc_skbedit.h>
38 #include <net/tc_act/tc_ct.h>
39 #include <net/tc_act/tc_mpls.h>
40 #include <net/flow_offload.h>
42 extern const struct nla_policy rtm_tca_policy
[TCA_MAX
+ 1];
44 /* The list of all installed classifier types */
45 static LIST_HEAD(tcf_proto_base
);
47 /* Protects list of registered TC modules. It is pure SMP lock. */
48 static DEFINE_RWLOCK(cls_mod_lock
);
50 /* Find classifier type by string name */
52 static const struct tcf_proto_ops
*__tcf_proto_lookup_ops(const char *kind
)
54 const struct tcf_proto_ops
*t
, *res
= NULL
;
57 read_lock(&cls_mod_lock
);
58 list_for_each_entry(t
, &tcf_proto_base
, head
) {
59 if (strcmp(kind
, t
->kind
) == 0) {
60 if (try_module_get(t
->owner
))
65 read_unlock(&cls_mod_lock
);
70 static const struct tcf_proto_ops
*
71 tcf_proto_lookup_ops(const char *kind
, bool rtnl_held
,
72 struct netlink_ext_ack
*extack
)
74 const struct tcf_proto_ops
*ops
;
76 ops
= __tcf_proto_lookup_ops(kind
);
82 request_module("cls_%s", kind
);
85 ops
= __tcf_proto_lookup_ops(kind
);
86 /* We dropped the RTNL semaphore in order to perform
87 * the module load. So, even if we succeeded in loading
88 * the module we have to replay the request. We indicate
92 module_put(ops
->owner
);
93 return ERR_PTR(-EAGAIN
);
96 NL_SET_ERR_MSG(extack
, "TC classifier not found");
97 return ERR_PTR(-ENOENT
);
100 /* Register(unregister) new classifier type */
102 int register_tcf_proto_ops(struct tcf_proto_ops
*ops
)
104 struct tcf_proto_ops
*t
;
107 write_lock(&cls_mod_lock
);
108 list_for_each_entry(t
, &tcf_proto_base
, head
)
109 if (!strcmp(ops
->kind
, t
->kind
))
112 list_add_tail(&ops
->head
, &tcf_proto_base
);
115 write_unlock(&cls_mod_lock
);
118 EXPORT_SYMBOL(register_tcf_proto_ops
);
120 static struct workqueue_struct
*tc_filter_wq
;
122 int unregister_tcf_proto_ops(struct tcf_proto_ops
*ops
)
124 struct tcf_proto_ops
*t
;
127 /* Wait for outstanding call_rcu()s, if any, from a
128 * tcf_proto_ops's destroy() handler.
131 flush_workqueue(tc_filter_wq
);
133 write_lock(&cls_mod_lock
);
134 list_for_each_entry(t
, &tcf_proto_base
, head
) {
141 write_unlock(&cls_mod_lock
);
144 EXPORT_SYMBOL(unregister_tcf_proto_ops
);
146 bool tcf_queue_work(struct rcu_work
*rwork
, work_func_t func
)
148 INIT_RCU_WORK(rwork
, func
);
149 return queue_rcu_work(tc_filter_wq
, rwork
);
151 EXPORT_SYMBOL(tcf_queue_work
);
153 /* Select new prio value from the range, managed by kernel. */
155 static inline u32
tcf_auto_prio(struct tcf_proto
*tp
)
157 u32 first
= TC_H_MAKE(0xC0000000U
, 0U);
160 first
= tp
->prio
- 1;
162 return TC_H_MAJ(first
);
165 static bool tcf_proto_is_unlocked(const char *kind
)
167 const struct tcf_proto_ops
*ops
;
170 ops
= tcf_proto_lookup_ops(kind
, false, NULL
);
171 /* On error return false to take rtnl lock. Proto lookup/create
172 * functions will perform lookup again and properly handle errors.
177 ret
= !!(ops
->flags
& TCF_PROTO_OPS_DOIT_UNLOCKED
);
178 module_put(ops
->owner
);
182 static struct tcf_proto
*tcf_proto_create(const char *kind
, u32 protocol
,
183 u32 prio
, struct tcf_chain
*chain
,
185 struct netlink_ext_ack
*extack
)
187 struct tcf_proto
*tp
;
190 tp
= kzalloc(sizeof(*tp
), GFP_KERNEL
);
192 return ERR_PTR(-ENOBUFS
);
194 tp
->ops
= tcf_proto_lookup_ops(kind
, rtnl_held
, extack
);
195 if (IS_ERR(tp
->ops
)) {
196 err
= PTR_ERR(tp
->ops
);
199 tp
->classify
= tp
->ops
->classify
;
200 tp
->protocol
= protocol
;
203 spin_lock_init(&tp
->lock
);
204 refcount_set(&tp
->refcnt
, 1);
206 err
= tp
->ops
->init(tp
);
208 module_put(tp
->ops
->owner
);
218 static void tcf_proto_get(struct tcf_proto
*tp
)
220 refcount_inc(&tp
->refcnt
);
223 static void tcf_chain_put(struct tcf_chain
*chain
);
225 static void tcf_proto_destroy(struct tcf_proto
*tp
, bool rtnl_held
,
226 struct netlink_ext_ack
*extack
)
228 tp
->ops
->destroy(tp
, rtnl_held
, extack
);
229 tcf_chain_put(tp
->chain
);
230 module_put(tp
->ops
->owner
);
234 static void tcf_proto_put(struct tcf_proto
*tp
, bool rtnl_held
,
235 struct netlink_ext_ack
*extack
)
237 if (refcount_dec_and_test(&tp
->refcnt
))
238 tcf_proto_destroy(tp
, rtnl_held
, extack
);
241 static int walker_check_empty(struct tcf_proto
*tp
, void *fh
,
242 struct tcf_walker
*arg
)
245 arg
->nonempty
= true;
251 static bool tcf_proto_is_empty(struct tcf_proto
*tp
, bool rtnl_held
)
253 struct tcf_walker walker
= { .fn
= walker_check_empty
, };
256 tp
->ops
->walk(tp
, &walker
, rtnl_held
);
257 return !walker
.nonempty
;
262 static bool tcf_proto_check_delete(struct tcf_proto
*tp
, bool rtnl_held
)
264 spin_lock(&tp
->lock
);
265 if (tcf_proto_is_empty(tp
, rtnl_held
))
267 spin_unlock(&tp
->lock
);
271 static void tcf_proto_mark_delete(struct tcf_proto
*tp
)
273 spin_lock(&tp
->lock
);
275 spin_unlock(&tp
->lock
);
278 static bool tcf_proto_is_deleting(struct tcf_proto
*tp
)
282 spin_lock(&tp
->lock
);
283 deleting
= tp
->deleting
;
284 spin_unlock(&tp
->lock
);
289 #define ASSERT_BLOCK_LOCKED(block) \
290 lockdep_assert_held(&(block)->lock)
292 struct tcf_filter_chain_list_item
{
293 struct list_head list
;
294 tcf_chain_head_change_t
*chain_head_change
;
295 void *chain_head_change_priv
;
298 static struct tcf_chain
*tcf_chain_create(struct tcf_block
*block
,
301 struct tcf_chain
*chain
;
303 ASSERT_BLOCK_LOCKED(block
);
305 chain
= kzalloc(sizeof(*chain
), GFP_KERNEL
);
308 list_add_tail(&chain
->list
, &block
->chain_list
);
309 mutex_init(&chain
->filter_chain_lock
);
310 chain
->block
= block
;
311 chain
->index
= chain_index
;
314 block
->chain0
.chain
= chain
;
318 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item
*item
,
319 struct tcf_proto
*tp_head
)
321 if (item
->chain_head_change
)
322 item
->chain_head_change(tp_head
, item
->chain_head_change_priv
);
325 static void tcf_chain0_head_change(struct tcf_chain
*chain
,
326 struct tcf_proto
*tp_head
)
328 struct tcf_filter_chain_list_item
*item
;
329 struct tcf_block
*block
= chain
->block
;
334 mutex_lock(&block
->lock
);
335 list_for_each_entry(item
, &block
->chain0
.filter_chain_list
, list
)
336 tcf_chain_head_change_item(item
, tp_head
);
337 mutex_unlock(&block
->lock
);
340 /* Returns true if block can be safely freed. */
342 static bool tcf_chain_detach(struct tcf_chain
*chain
)
344 struct tcf_block
*block
= chain
->block
;
346 ASSERT_BLOCK_LOCKED(block
);
348 list_del(&chain
->list
);
350 block
->chain0
.chain
= NULL
;
352 if (list_empty(&block
->chain_list
) &&
353 refcount_read(&block
->refcnt
) == 0)
359 static void tcf_block_destroy(struct tcf_block
*block
)
361 mutex_destroy(&block
->lock
);
362 kfree_rcu(block
, rcu
);
365 static void tcf_chain_destroy(struct tcf_chain
*chain
, bool free_block
)
367 struct tcf_block
*block
= chain
->block
;
369 mutex_destroy(&chain
->filter_chain_lock
);
370 kfree_rcu(chain
, rcu
);
372 tcf_block_destroy(block
);
375 static void tcf_chain_hold(struct tcf_chain
*chain
)
377 ASSERT_BLOCK_LOCKED(chain
->block
);
382 static bool tcf_chain_held_by_acts_only(struct tcf_chain
*chain
)
384 ASSERT_BLOCK_LOCKED(chain
->block
);
386 /* In case all the references are action references, this
387 * chain should not be shown to the user.
389 return chain
->refcnt
== chain
->action_refcnt
;
392 static struct tcf_chain
*tcf_chain_lookup(struct tcf_block
*block
,
395 struct tcf_chain
*chain
;
397 ASSERT_BLOCK_LOCKED(block
);
399 list_for_each_entry(chain
, &block
->chain_list
, list
) {
400 if (chain
->index
== chain_index
)
406 static int tc_chain_notify(struct tcf_chain
*chain
, struct sk_buff
*oskb
,
407 u32 seq
, u16 flags
, int event
, bool unicast
);
409 static struct tcf_chain
*__tcf_chain_get(struct tcf_block
*block
,
410 u32 chain_index
, bool create
,
413 struct tcf_chain
*chain
= NULL
;
414 bool is_first_reference
;
416 mutex_lock(&block
->lock
);
417 chain
= tcf_chain_lookup(block
, chain_index
);
419 tcf_chain_hold(chain
);
423 chain
= tcf_chain_create(block
, chain_index
);
429 ++chain
->action_refcnt
;
430 is_first_reference
= chain
->refcnt
- chain
->action_refcnt
== 1;
431 mutex_unlock(&block
->lock
);
433 /* Send notification only in case we got the first
434 * non-action reference. Until then, the chain acts only as
435 * a placeholder for actions pointing to it and user ought
436 * not know about them.
438 if (is_first_reference
&& !by_act
)
439 tc_chain_notify(chain
, NULL
, 0, NLM_F_CREATE
| NLM_F_EXCL
,
440 RTM_NEWCHAIN
, false);
445 mutex_unlock(&block
->lock
);
449 static struct tcf_chain
*tcf_chain_get(struct tcf_block
*block
, u32 chain_index
,
452 return __tcf_chain_get(block
, chain_index
, create
, false);
455 struct tcf_chain
*tcf_chain_get_by_act(struct tcf_block
*block
, u32 chain_index
)
457 return __tcf_chain_get(block
, chain_index
, true, true);
459 EXPORT_SYMBOL(tcf_chain_get_by_act
);
461 static void tc_chain_tmplt_del(const struct tcf_proto_ops
*tmplt_ops
,
463 static int tc_chain_notify_delete(const struct tcf_proto_ops
*tmplt_ops
,
464 void *tmplt_priv
, u32 chain_index
,
465 struct tcf_block
*block
, struct sk_buff
*oskb
,
466 u32 seq
, u16 flags
, bool unicast
);
468 static void __tcf_chain_put(struct tcf_chain
*chain
, bool by_act
,
469 bool explicitly_created
)
471 struct tcf_block
*block
= chain
->block
;
472 const struct tcf_proto_ops
*tmplt_ops
;
473 bool free_block
= false;
477 mutex_lock(&block
->lock
);
478 if (explicitly_created
) {
479 if (!chain
->explicitly_created
) {
480 mutex_unlock(&block
->lock
);
483 chain
->explicitly_created
= false;
487 chain
->action_refcnt
--;
489 /* tc_chain_notify_delete can't be called while holding block lock.
490 * However, when block is unlocked chain can be changed concurrently, so
491 * save these to temporary variables.
493 refcnt
= --chain
->refcnt
;
494 tmplt_ops
= chain
->tmplt_ops
;
495 tmplt_priv
= chain
->tmplt_priv
;
497 /* The last dropped non-action reference will trigger notification. */
498 if (refcnt
- chain
->action_refcnt
== 0 && !by_act
) {
499 tc_chain_notify_delete(tmplt_ops
, tmplt_priv
, chain
->index
,
500 block
, NULL
, 0, 0, false);
501 /* Last reference to chain, no need to lock. */
502 chain
->flushing
= false;
506 free_block
= tcf_chain_detach(chain
);
507 mutex_unlock(&block
->lock
);
510 tc_chain_tmplt_del(tmplt_ops
, tmplt_priv
);
511 tcf_chain_destroy(chain
, free_block
);
515 static void tcf_chain_put(struct tcf_chain
*chain
)
517 __tcf_chain_put(chain
, false, false);
520 void tcf_chain_put_by_act(struct tcf_chain
*chain
)
522 __tcf_chain_put(chain
, true, false);
524 EXPORT_SYMBOL(tcf_chain_put_by_act
);
526 static void tcf_chain_put_explicitly_created(struct tcf_chain
*chain
)
528 __tcf_chain_put(chain
, false, true);
531 static void tcf_chain_flush(struct tcf_chain
*chain
, bool rtnl_held
)
533 struct tcf_proto
*tp
, *tp_next
;
535 mutex_lock(&chain
->filter_chain_lock
);
536 tp
= tcf_chain_dereference(chain
->filter_chain
, chain
);
537 RCU_INIT_POINTER(chain
->filter_chain
, NULL
);
538 tcf_chain0_head_change(chain
, NULL
);
539 chain
->flushing
= true;
540 mutex_unlock(&chain
->filter_chain_lock
);
543 tp_next
= rcu_dereference_protected(tp
->next
, 1);
544 tcf_proto_put(tp
, rtnl_held
, NULL
);
549 static int tcf_block_setup(struct tcf_block
*block
,
550 struct flow_block_offload
*bo
);
552 static void tc_indr_block_ing_cmd(struct net_device
*dev
,
553 struct tcf_block
*block
,
554 flow_indr_block_bind_cb_t
*cb
,
556 enum flow_block_command command
)
558 struct flow_block_offload bo
= {
560 .binder_type
= FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS
,
562 .block_shared
= tcf_block_non_null_shared(block
),
564 INIT_LIST_HEAD(&bo
.cb_list
);
569 bo
.block
= &block
->flow_block
;
571 down_write(&block
->cb_lock
);
572 cb(dev
, cb_priv
, TC_SETUP_BLOCK
, &bo
);
574 tcf_block_setup(block
, &bo
);
575 up_write(&block
->cb_lock
);
578 static struct tcf_block
*tc_dev_ingress_block(struct net_device
*dev
)
580 const struct Qdisc_class_ops
*cops
;
583 if (!dev_ingress_queue(dev
))
586 qdisc
= dev_ingress_queue(dev
)->qdisc_sleeping
;
590 cops
= qdisc
->ops
->cl_ops
;
594 if (!cops
->tcf_block
)
597 return cops
->tcf_block(qdisc
, TC_H_MIN_INGRESS
, NULL
);
600 static void tc_indr_block_get_and_ing_cmd(struct net_device
*dev
,
601 flow_indr_block_bind_cb_t
*cb
,
603 enum flow_block_command command
)
605 struct tcf_block
*block
= tc_dev_ingress_block(dev
);
607 tc_indr_block_ing_cmd(dev
, block
, cb
, cb_priv
, command
);
610 static void tc_indr_block_call(struct tcf_block
*block
,
611 struct net_device
*dev
,
612 struct tcf_block_ext_info
*ei
,
613 enum flow_block_command command
,
614 struct netlink_ext_ack
*extack
)
616 struct flow_block_offload bo
= {
618 .binder_type
= ei
->binder_type
,
620 .block
= &block
->flow_block
,
621 .block_shared
= tcf_block_shared(block
),
624 INIT_LIST_HEAD(&bo
.cb_list
);
626 flow_indr_block_call(dev
, &bo
, command
);
627 tcf_block_setup(block
, &bo
);
630 static bool tcf_block_offload_in_use(struct tcf_block
*block
)
632 return atomic_read(&block
->offloadcnt
);
635 static int tcf_block_offload_cmd(struct tcf_block
*block
,
636 struct net_device
*dev
,
637 struct tcf_block_ext_info
*ei
,
638 enum flow_block_command command
,
639 struct netlink_ext_ack
*extack
)
641 struct flow_block_offload bo
= {};
644 bo
.net
= dev_net(dev
);
645 bo
.command
= command
;
646 bo
.binder_type
= ei
->binder_type
;
647 bo
.block
= &block
->flow_block
;
648 bo
.block_shared
= tcf_block_shared(block
);
650 INIT_LIST_HEAD(&bo
.cb_list
);
652 err
= dev
->netdev_ops
->ndo_setup_tc(dev
, TC_SETUP_BLOCK
, &bo
);
656 return tcf_block_setup(block
, &bo
);
659 static int tcf_block_offload_bind(struct tcf_block
*block
, struct Qdisc
*q
,
660 struct tcf_block_ext_info
*ei
,
661 struct netlink_ext_ack
*extack
)
663 struct net_device
*dev
= q
->dev_queue
->dev
;
666 down_write(&block
->cb_lock
);
667 if (!dev
->netdev_ops
->ndo_setup_tc
)
668 goto no_offload_dev_inc
;
670 /* If tc offload feature is disabled and the block we try to bind
671 * to already has some offloaded filters, forbid to bind.
673 if (!tc_can_offload(dev
) && tcf_block_offload_in_use(block
)) {
674 NL_SET_ERR_MSG(extack
, "Bind to offloaded block failed as dev has offload disabled");
679 err
= tcf_block_offload_cmd(block
, dev
, ei
, FLOW_BLOCK_BIND
, extack
);
680 if (err
== -EOPNOTSUPP
)
681 goto no_offload_dev_inc
;
685 tc_indr_block_call(block
, dev
, ei
, FLOW_BLOCK_BIND
, extack
);
686 up_write(&block
->cb_lock
);
690 if (tcf_block_offload_in_use(block
)) {
695 block
->nooffloaddevcnt
++;
696 tc_indr_block_call(block
, dev
, ei
, FLOW_BLOCK_BIND
, extack
);
698 up_write(&block
->cb_lock
);
702 static void tcf_block_offload_unbind(struct tcf_block
*block
, struct Qdisc
*q
,
703 struct tcf_block_ext_info
*ei
)
705 struct net_device
*dev
= q
->dev_queue
->dev
;
708 down_write(&block
->cb_lock
);
709 tc_indr_block_call(block
, dev
, ei
, FLOW_BLOCK_UNBIND
, NULL
);
711 if (!dev
->netdev_ops
->ndo_setup_tc
)
712 goto no_offload_dev_dec
;
713 err
= tcf_block_offload_cmd(block
, dev
, ei
, FLOW_BLOCK_UNBIND
, NULL
);
714 if (err
== -EOPNOTSUPP
)
715 goto no_offload_dev_dec
;
716 up_write(&block
->cb_lock
);
720 WARN_ON(block
->nooffloaddevcnt
-- == 0);
721 up_write(&block
->cb_lock
);
725 tcf_chain0_head_change_cb_add(struct tcf_block
*block
,
726 struct tcf_block_ext_info
*ei
,
727 struct netlink_ext_ack
*extack
)
729 struct tcf_filter_chain_list_item
*item
;
730 struct tcf_chain
*chain0
;
732 item
= kmalloc(sizeof(*item
), GFP_KERNEL
);
734 NL_SET_ERR_MSG(extack
, "Memory allocation for head change callback item failed");
737 item
->chain_head_change
= ei
->chain_head_change
;
738 item
->chain_head_change_priv
= ei
->chain_head_change_priv
;
740 mutex_lock(&block
->lock
);
741 chain0
= block
->chain0
.chain
;
743 tcf_chain_hold(chain0
);
745 list_add(&item
->list
, &block
->chain0
.filter_chain_list
);
746 mutex_unlock(&block
->lock
);
749 struct tcf_proto
*tp_head
;
751 mutex_lock(&chain0
->filter_chain_lock
);
753 tp_head
= tcf_chain_dereference(chain0
->filter_chain
, chain0
);
755 tcf_chain_head_change_item(item
, tp_head
);
757 mutex_lock(&block
->lock
);
758 list_add(&item
->list
, &block
->chain0
.filter_chain_list
);
759 mutex_unlock(&block
->lock
);
761 mutex_unlock(&chain0
->filter_chain_lock
);
762 tcf_chain_put(chain0
);
769 tcf_chain0_head_change_cb_del(struct tcf_block
*block
,
770 struct tcf_block_ext_info
*ei
)
772 struct tcf_filter_chain_list_item
*item
;
774 mutex_lock(&block
->lock
);
775 list_for_each_entry(item
, &block
->chain0
.filter_chain_list
, list
) {
776 if ((!ei
->chain_head_change
&& !ei
->chain_head_change_priv
) ||
777 (item
->chain_head_change
== ei
->chain_head_change
&&
778 item
->chain_head_change_priv
== ei
->chain_head_change_priv
)) {
779 if (block
->chain0
.chain
)
780 tcf_chain_head_change_item(item
, NULL
);
781 list_del(&item
->list
);
782 mutex_unlock(&block
->lock
);
788 mutex_unlock(&block
->lock
);
793 spinlock_t idr_lock
; /* Protects idr */
797 static unsigned int tcf_net_id
;
799 static int tcf_block_insert(struct tcf_block
*block
, struct net
*net
,
800 struct netlink_ext_ack
*extack
)
802 struct tcf_net
*tn
= net_generic(net
, tcf_net_id
);
805 idr_preload(GFP_KERNEL
);
806 spin_lock(&tn
->idr_lock
);
807 err
= idr_alloc_u32(&tn
->idr
, block
, &block
->index
, block
->index
,
809 spin_unlock(&tn
->idr_lock
);
815 static void tcf_block_remove(struct tcf_block
*block
, struct net
*net
)
817 struct tcf_net
*tn
= net_generic(net
, tcf_net_id
);
819 spin_lock(&tn
->idr_lock
);
820 idr_remove(&tn
->idr
, block
->index
);
821 spin_unlock(&tn
->idr_lock
);
824 static struct tcf_block
*tcf_block_create(struct net
*net
, struct Qdisc
*q
,
826 struct netlink_ext_ack
*extack
)
828 struct tcf_block
*block
;
830 block
= kzalloc(sizeof(*block
), GFP_KERNEL
);
832 NL_SET_ERR_MSG(extack
, "Memory allocation for block failed");
833 return ERR_PTR(-ENOMEM
);
835 mutex_init(&block
->lock
);
836 init_rwsem(&block
->cb_lock
);
837 flow_block_init(&block
->flow_block
);
838 INIT_LIST_HEAD(&block
->chain_list
);
839 INIT_LIST_HEAD(&block
->owner_list
);
840 INIT_LIST_HEAD(&block
->chain0
.filter_chain_list
);
842 refcount_set(&block
->refcnt
, 1);
844 block
->index
= block_index
;
846 /* Don't store q pointer for blocks which are shared */
847 if (!tcf_block_shared(block
))
852 static struct tcf_block
*tcf_block_lookup(struct net
*net
, u32 block_index
)
854 struct tcf_net
*tn
= net_generic(net
, tcf_net_id
);
856 return idr_find(&tn
->idr
, block_index
);
859 static struct tcf_block
*tcf_block_refcnt_get(struct net
*net
, u32 block_index
)
861 struct tcf_block
*block
;
864 block
= tcf_block_lookup(net
, block_index
);
865 if (block
&& !refcount_inc_not_zero(&block
->refcnt
))
872 static struct tcf_chain
*
873 __tcf_get_next_chain(struct tcf_block
*block
, struct tcf_chain
*chain
)
875 mutex_lock(&block
->lock
);
877 chain
= list_is_last(&chain
->list
, &block
->chain_list
) ?
878 NULL
: list_next_entry(chain
, list
);
880 chain
= list_first_entry_or_null(&block
->chain_list
,
881 struct tcf_chain
, list
);
883 /* skip all action-only chains */
884 while (chain
&& tcf_chain_held_by_acts_only(chain
))
885 chain
= list_is_last(&chain
->list
, &block
->chain_list
) ?
886 NULL
: list_next_entry(chain
, list
);
889 tcf_chain_hold(chain
);
890 mutex_unlock(&block
->lock
);
895 /* Function to be used by all clients that want to iterate over all chains on
896 * block. It properly obtains block->lock and takes reference to chain before
897 * returning it. Users of this function must be tolerant to concurrent chain
898 * insertion/deletion or ensure that no concurrent chain modification is
899 * possible. Note that all netlink dump callbacks cannot guarantee to provide
900 * consistent dump because rtnl lock is released each time skb is filled with
901 * data and sent to user-space.
905 tcf_get_next_chain(struct tcf_block
*block
, struct tcf_chain
*chain
)
907 struct tcf_chain
*chain_next
= __tcf_get_next_chain(block
, chain
);
910 tcf_chain_put(chain
);
914 EXPORT_SYMBOL(tcf_get_next_chain
);
916 static struct tcf_proto
*
917 __tcf_get_next_proto(struct tcf_chain
*chain
, struct tcf_proto
*tp
)
922 mutex_lock(&chain
->filter_chain_lock
);
925 tp
= tcf_chain_dereference(chain
->filter_chain
, chain
);
926 } else if (tcf_proto_is_deleting(tp
)) {
927 /* 'deleting' flag is set and chain->filter_chain_lock was
928 * unlocked, which means next pointer could be invalid. Restart
932 tp
= tcf_chain_dereference(chain
->filter_chain
, chain
);
934 for (; tp
; tp
= tcf_chain_dereference(tp
->next
, chain
))
935 if (!tp
->deleting
&& tp
->prio
>= prio
)
938 tp
= tcf_chain_dereference(tp
->next
, chain
);
944 mutex_unlock(&chain
->filter_chain_lock
);
949 /* Function to be used by all clients that want to iterate over all tp's on
950 * chain. Users of this function must be tolerant to concurrent tp
951 * insertion/deletion or ensure that no concurrent chain modification is
952 * possible. Note that all netlink dump callbacks cannot guarantee to provide
953 * consistent dump because rtnl lock is released each time skb is filled with
954 * data and sent to user-space.
958 tcf_get_next_proto(struct tcf_chain
*chain
, struct tcf_proto
*tp
,
961 struct tcf_proto
*tp_next
= __tcf_get_next_proto(chain
, tp
);
964 tcf_proto_put(tp
, rtnl_held
, NULL
);
968 EXPORT_SYMBOL(tcf_get_next_proto
);
970 static void tcf_block_flush_all_chains(struct tcf_block
*block
, bool rtnl_held
)
972 struct tcf_chain
*chain
;
974 /* Last reference to block. At this point chains cannot be added or
975 * removed concurrently.
977 for (chain
= tcf_get_next_chain(block
, NULL
);
979 chain
= tcf_get_next_chain(block
, chain
)) {
980 tcf_chain_put_explicitly_created(chain
);
981 tcf_chain_flush(chain
, rtnl_held
);
985 /* Lookup Qdisc and increments its reference counter.
986 * Set parent, if necessary.
989 static int __tcf_qdisc_find(struct net
*net
, struct Qdisc
**q
,
990 u32
*parent
, int ifindex
, bool rtnl_held
,
991 struct netlink_ext_ack
*extack
)
993 const struct Qdisc_class_ops
*cops
;
994 struct net_device
*dev
;
997 if (ifindex
== TCM_IFINDEX_MAGIC_BLOCK
)
1003 dev
= dev_get_by_index_rcu(net
, ifindex
);
1012 *parent
= (*q
)->handle
;
1014 *q
= qdisc_lookup_rcu(dev
, TC_H_MAJ(*parent
));
1016 NL_SET_ERR_MSG(extack
, "Parent Qdisc doesn't exists");
1022 *q
= qdisc_refcount_inc_nz(*q
);
1024 NL_SET_ERR_MSG(extack
, "Parent Qdisc doesn't exists");
1029 /* Is it classful? */
1030 cops
= (*q
)->ops
->cl_ops
;
1032 NL_SET_ERR_MSG(extack
, "Qdisc not classful");
1037 if (!cops
->tcf_block
) {
1038 NL_SET_ERR_MSG(extack
, "Class doesn't support blocks");
1044 /* At this point we know that qdisc is not noop_qdisc,
1045 * which means that qdisc holds a reference to net_device
1046 * and we hold a reference to qdisc, so it is safe to release
1058 qdisc_put_unlocked(*q
);
1064 static int __tcf_qdisc_cl_find(struct Qdisc
*q
, u32 parent
, unsigned long *cl
,
1065 int ifindex
, struct netlink_ext_ack
*extack
)
1067 if (ifindex
== TCM_IFINDEX_MAGIC_BLOCK
)
1070 /* Do we search for filter, attached to class? */
1071 if (TC_H_MIN(parent
)) {
1072 const struct Qdisc_class_ops
*cops
= q
->ops
->cl_ops
;
1074 *cl
= cops
->find(q
, parent
);
1076 NL_SET_ERR_MSG(extack
, "Specified class doesn't exist");
1084 static struct tcf_block
*__tcf_block_find(struct net
*net
, struct Qdisc
*q
,
1085 unsigned long cl
, int ifindex
,
1087 struct netlink_ext_ack
*extack
)
1089 struct tcf_block
*block
;
1091 if (ifindex
== TCM_IFINDEX_MAGIC_BLOCK
) {
1092 block
= tcf_block_refcnt_get(net
, block_index
);
1094 NL_SET_ERR_MSG(extack
, "Block of given index was not found");
1095 return ERR_PTR(-EINVAL
);
1098 const struct Qdisc_class_ops
*cops
= q
->ops
->cl_ops
;
1100 block
= cops
->tcf_block(q
, cl
, extack
);
1102 return ERR_PTR(-EINVAL
);
1104 if (tcf_block_shared(block
)) {
1105 NL_SET_ERR_MSG(extack
, "This filter block is shared. Please use the block index to manipulate the filters");
1106 return ERR_PTR(-EOPNOTSUPP
);
1109 /* Always take reference to block in order to support execution
1110 * of rules update path of cls API without rtnl lock. Caller
1111 * must release block when it is finished using it. 'if' block
1112 * of this conditional obtain reference to block by calling
1113 * tcf_block_refcnt_get().
1115 refcount_inc(&block
->refcnt
);
1121 static void __tcf_block_put(struct tcf_block
*block
, struct Qdisc
*q
,
1122 struct tcf_block_ext_info
*ei
, bool rtnl_held
)
1124 if (refcount_dec_and_mutex_lock(&block
->refcnt
, &block
->lock
)) {
1125 /* Flushing/putting all chains will cause the block to be
1126 * deallocated when last chain is freed. However, if chain_list
1127 * is empty, block has to be manually deallocated. After block
1128 * reference counter reached 0, it is no longer possible to
1129 * increment it or add new chains to block.
1131 bool free_block
= list_empty(&block
->chain_list
);
1133 mutex_unlock(&block
->lock
);
1134 if (tcf_block_shared(block
))
1135 tcf_block_remove(block
, block
->net
);
1138 tcf_block_offload_unbind(block
, q
, ei
);
1141 tcf_block_destroy(block
);
1143 tcf_block_flush_all_chains(block
, rtnl_held
);
1145 tcf_block_offload_unbind(block
, q
, ei
);
1149 static void tcf_block_refcnt_put(struct tcf_block
*block
, bool rtnl_held
)
1151 __tcf_block_put(block
, NULL
, NULL
, rtnl_held
);
1155 * Set q, parent, cl when appropriate.
1158 static struct tcf_block
*tcf_block_find(struct net
*net
, struct Qdisc
**q
,
1159 u32
*parent
, unsigned long *cl
,
1160 int ifindex
, u32 block_index
,
1161 struct netlink_ext_ack
*extack
)
1163 struct tcf_block
*block
;
1168 err
= __tcf_qdisc_find(net
, q
, parent
, ifindex
, true, extack
);
1172 err
= __tcf_qdisc_cl_find(*q
, *parent
, cl
, ifindex
, extack
);
1176 block
= __tcf_block_find(net
, *q
, *cl
, ifindex
, block_index
, extack
);
1177 if (IS_ERR(block
)) {
1178 err
= PTR_ERR(block
);
1189 return ERR_PTR(err
);
1192 static void tcf_block_release(struct Qdisc
*q
, struct tcf_block
*block
,
1195 if (!IS_ERR_OR_NULL(block
))
1196 tcf_block_refcnt_put(block
, rtnl_held
);
1202 qdisc_put_unlocked(q
);
1206 struct tcf_block_owner_item
{
1207 struct list_head list
;
1209 enum flow_block_binder_type binder_type
;
1213 tcf_block_owner_netif_keep_dst(struct tcf_block
*block
,
1215 enum flow_block_binder_type binder_type
)
1217 if (block
->keep_dst
&&
1218 binder_type
!= FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS
&&
1219 binder_type
!= FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS
)
1220 netif_keep_dst(qdisc_dev(q
));
1223 void tcf_block_netif_keep_dst(struct tcf_block
*block
)
1225 struct tcf_block_owner_item
*item
;
1227 block
->keep_dst
= true;
1228 list_for_each_entry(item
, &block
->owner_list
, list
)
1229 tcf_block_owner_netif_keep_dst(block
, item
->q
,
1232 EXPORT_SYMBOL(tcf_block_netif_keep_dst
);
1234 static int tcf_block_owner_add(struct tcf_block
*block
,
1236 enum flow_block_binder_type binder_type
)
1238 struct tcf_block_owner_item
*item
;
1240 item
= kmalloc(sizeof(*item
), GFP_KERNEL
);
1244 item
->binder_type
= binder_type
;
1245 list_add(&item
->list
, &block
->owner_list
);
1249 static void tcf_block_owner_del(struct tcf_block
*block
,
1251 enum flow_block_binder_type binder_type
)
1253 struct tcf_block_owner_item
*item
;
1255 list_for_each_entry(item
, &block
->owner_list
, list
) {
1256 if (item
->q
== q
&& item
->binder_type
== binder_type
) {
1257 list_del(&item
->list
);
1265 int tcf_block_get_ext(struct tcf_block
**p_block
, struct Qdisc
*q
,
1266 struct tcf_block_ext_info
*ei
,
1267 struct netlink_ext_ack
*extack
)
1269 struct net
*net
= qdisc_net(q
);
1270 struct tcf_block
*block
= NULL
;
1273 if (ei
->block_index
)
1274 /* block_index not 0 means the shared block is requested */
1275 block
= tcf_block_refcnt_get(net
, ei
->block_index
);
1278 block
= tcf_block_create(net
, q
, ei
->block_index
, extack
);
1280 return PTR_ERR(block
);
1281 if (tcf_block_shared(block
)) {
1282 err
= tcf_block_insert(block
, net
, extack
);
1284 goto err_block_insert
;
1288 err
= tcf_block_owner_add(block
, q
, ei
->binder_type
);
1290 goto err_block_owner_add
;
1292 tcf_block_owner_netif_keep_dst(block
, q
, ei
->binder_type
);
1294 err
= tcf_chain0_head_change_cb_add(block
, ei
, extack
);
1296 goto err_chain0_head_change_cb_add
;
1298 err
= tcf_block_offload_bind(block
, q
, ei
, extack
);
1300 goto err_block_offload_bind
;
1305 err_block_offload_bind
:
1306 tcf_chain0_head_change_cb_del(block
, ei
);
1307 err_chain0_head_change_cb_add
:
1308 tcf_block_owner_del(block
, q
, ei
->binder_type
);
1309 err_block_owner_add
:
1311 tcf_block_refcnt_put(block
, true);
1314 EXPORT_SYMBOL(tcf_block_get_ext
);
1316 static void tcf_chain_head_change_dflt(struct tcf_proto
*tp_head
, void *priv
)
1318 struct tcf_proto __rcu
**p_filter_chain
= priv
;
1320 rcu_assign_pointer(*p_filter_chain
, tp_head
);
1323 int tcf_block_get(struct tcf_block
**p_block
,
1324 struct tcf_proto __rcu
**p_filter_chain
, struct Qdisc
*q
,
1325 struct netlink_ext_ack
*extack
)
1327 struct tcf_block_ext_info ei
= {
1328 .chain_head_change
= tcf_chain_head_change_dflt
,
1329 .chain_head_change_priv
= p_filter_chain
,
1332 WARN_ON(!p_filter_chain
);
1333 return tcf_block_get_ext(p_block
, q
, &ei
, extack
);
1335 EXPORT_SYMBOL(tcf_block_get
);
1337 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1338 * actions should be all removed after flushing.
1340 void tcf_block_put_ext(struct tcf_block
*block
, struct Qdisc
*q
,
1341 struct tcf_block_ext_info
*ei
)
1345 tcf_chain0_head_change_cb_del(block
, ei
);
1346 tcf_block_owner_del(block
, q
, ei
->binder_type
);
1348 __tcf_block_put(block
, q
, ei
, true);
1350 EXPORT_SYMBOL(tcf_block_put_ext
);
1352 void tcf_block_put(struct tcf_block
*block
)
1354 struct tcf_block_ext_info ei
= {0, };
1358 tcf_block_put_ext(block
, block
->q
, &ei
);
1361 EXPORT_SYMBOL(tcf_block_put
);
1364 tcf_block_playback_offloads(struct tcf_block
*block
, flow_setup_cb_t
*cb
,
1365 void *cb_priv
, bool add
, bool offload_in_use
,
1366 struct netlink_ext_ack
*extack
)
1368 struct tcf_chain
*chain
, *chain_prev
;
1369 struct tcf_proto
*tp
, *tp_prev
;
1372 lockdep_assert_held(&block
->cb_lock
);
1374 for (chain
= __tcf_get_next_chain(block
, NULL
);
1377 chain
= __tcf_get_next_chain(block
, chain
),
1378 tcf_chain_put(chain_prev
)) {
1379 for (tp
= __tcf_get_next_proto(chain
, NULL
); tp
;
1381 tp
= __tcf_get_next_proto(chain
, tp
),
1382 tcf_proto_put(tp_prev
, true, NULL
)) {
1383 if (tp
->ops
->reoffload
) {
1384 err
= tp
->ops
->reoffload(tp
, add
, cb
, cb_priv
,
1387 goto err_playback_remove
;
1388 } else if (add
&& offload_in_use
) {
1390 NL_SET_ERR_MSG(extack
, "Filter HW offload failed - classifier without re-offloading support");
1391 goto err_playback_remove
;
1398 err_playback_remove
:
1399 tcf_proto_put(tp
, true, NULL
);
1400 tcf_chain_put(chain
);
1401 tcf_block_playback_offloads(block
, cb
, cb_priv
, false, offload_in_use
,
1406 static int tcf_block_bind(struct tcf_block
*block
,
1407 struct flow_block_offload
*bo
)
1409 struct flow_block_cb
*block_cb
, *next
;
1412 lockdep_assert_held(&block
->cb_lock
);
1414 list_for_each_entry(block_cb
, &bo
->cb_list
, list
) {
1415 err
= tcf_block_playback_offloads(block
, block_cb
->cb
,
1416 block_cb
->cb_priv
, true,
1417 tcf_block_offload_in_use(block
),
1421 if (!bo
->unlocked_driver_cb
)
1422 block
->lockeddevcnt
++;
1426 list_splice(&bo
->cb_list
, &block
->flow_block
.cb_list
);
1431 list_for_each_entry_safe(block_cb
, next
, &bo
->cb_list
, list
) {
1433 list_del(&block_cb
->list
);
1434 tcf_block_playback_offloads(block
, block_cb
->cb
,
1435 block_cb
->cb_priv
, false,
1436 tcf_block_offload_in_use(block
),
1438 if (!bo
->unlocked_driver_cb
)
1439 block
->lockeddevcnt
--;
1441 flow_block_cb_free(block_cb
);
1447 static void tcf_block_unbind(struct tcf_block
*block
,
1448 struct flow_block_offload
*bo
)
1450 struct flow_block_cb
*block_cb
, *next
;
1452 lockdep_assert_held(&block
->cb_lock
);
1454 list_for_each_entry_safe(block_cb
, next
, &bo
->cb_list
, list
) {
1455 tcf_block_playback_offloads(block
, block_cb
->cb
,
1456 block_cb
->cb_priv
, false,
1457 tcf_block_offload_in_use(block
),
1459 list_del(&block_cb
->list
);
1460 flow_block_cb_free(block_cb
);
1461 if (!bo
->unlocked_driver_cb
)
1462 block
->lockeddevcnt
--;
1466 static int tcf_block_setup(struct tcf_block
*block
,
1467 struct flow_block_offload
*bo
)
1471 switch (bo
->command
) {
1472 case FLOW_BLOCK_BIND
:
1473 err
= tcf_block_bind(block
, bo
);
1475 case FLOW_BLOCK_UNBIND
:
1477 tcf_block_unbind(block
, bo
);
1487 /* Main classifier routine: scans classifier chain attached
1488 * to this qdisc, (optionally) tests for protocol and asks
1489 * specific classifiers.
1491 int tcf_classify(struct sk_buff
*skb
, const struct tcf_proto
*tp
,
1492 struct tcf_result
*res
, bool compat_mode
)
1494 #ifdef CONFIG_NET_CLS_ACT
1495 const int max_reclassify_loop
= 4;
1496 const struct tcf_proto
*orig_tp
= tp
;
1497 const struct tcf_proto
*first_tp
;
1502 for (; tp
; tp
= rcu_dereference_bh(tp
->next
)) {
1503 __be16 protocol
= tc_skb_protocol(skb
);
1506 if (tp
->protocol
!= protocol
&&
1507 tp
->protocol
!= htons(ETH_P_ALL
))
1510 err
= tp
->classify(skb
, tp
, res
);
1511 #ifdef CONFIG_NET_CLS_ACT
1512 if (unlikely(err
== TC_ACT_RECLASSIFY
&& !compat_mode
)) {
1515 } else if (unlikely(TC_ACT_EXT_CMP(err
, TC_ACT_GOTO_CHAIN
))) {
1516 first_tp
= res
->goto_tp
;
1518 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1520 struct tc_skb_ext
*ext
;
1522 ext
= skb_ext_add(skb
, TC_SKB_EXT
);
1523 if (WARN_ON_ONCE(!ext
))
1526 ext
->chain
= err
& TC_ACT_EXT_VAL_MASK
;
1536 return TC_ACT_UNSPEC
; /* signal: continue lookup */
1537 #ifdef CONFIG_NET_CLS_ACT
1539 if (unlikely(limit
++ >= max_reclassify_loop
)) {
1540 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1541 tp
->chain
->block
->index
,
1543 ntohs(tp
->protocol
));
1551 EXPORT_SYMBOL(tcf_classify
);
1553 struct tcf_chain_info
{
1554 struct tcf_proto __rcu
**pprev
;
1555 struct tcf_proto __rcu
*next
;
1558 static struct tcf_proto
*tcf_chain_tp_prev(struct tcf_chain
*chain
,
1559 struct tcf_chain_info
*chain_info
)
1561 return tcf_chain_dereference(*chain_info
->pprev
, chain
);
1564 static int tcf_chain_tp_insert(struct tcf_chain
*chain
,
1565 struct tcf_chain_info
*chain_info
,
1566 struct tcf_proto
*tp
)
1568 if (chain
->flushing
)
1571 if (*chain_info
->pprev
== chain
->filter_chain
)
1572 tcf_chain0_head_change(chain
, tp
);
1574 RCU_INIT_POINTER(tp
->next
, tcf_chain_tp_prev(chain
, chain_info
));
1575 rcu_assign_pointer(*chain_info
->pprev
, tp
);
1580 static void tcf_chain_tp_remove(struct tcf_chain
*chain
,
1581 struct tcf_chain_info
*chain_info
,
1582 struct tcf_proto
*tp
)
1584 struct tcf_proto
*next
= tcf_chain_dereference(chain_info
->next
, chain
);
1586 tcf_proto_mark_delete(tp
);
1587 if (tp
== chain
->filter_chain
)
1588 tcf_chain0_head_change(chain
, next
);
1589 RCU_INIT_POINTER(*chain_info
->pprev
, next
);
1592 static struct tcf_proto
*tcf_chain_tp_find(struct tcf_chain
*chain
,
1593 struct tcf_chain_info
*chain_info
,
1594 u32 protocol
, u32 prio
,
1595 bool prio_allocate
);
1597 /* Try to insert new proto.
1598 * If proto with specified priority already exists, free new proto
1599 * and return existing one.
1602 static struct tcf_proto
*tcf_chain_tp_insert_unique(struct tcf_chain
*chain
,
1603 struct tcf_proto
*tp_new
,
1604 u32 protocol
, u32 prio
,
1607 struct tcf_chain_info chain_info
;
1608 struct tcf_proto
*tp
;
1611 mutex_lock(&chain
->filter_chain_lock
);
1613 tp
= tcf_chain_tp_find(chain
, &chain_info
,
1614 protocol
, prio
, false);
1616 err
= tcf_chain_tp_insert(chain
, &chain_info
, tp_new
);
1617 mutex_unlock(&chain
->filter_chain_lock
);
1620 tcf_proto_destroy(tp_new
, rtnl_held
, NULL
);
1623 tcf_proto_destroy(tp_new
, rtnl_held
, NULL
);
1624 tp_new
= ERR_PTR(err
);
1630 static void tcf_chain_tp_delete_empty(struct tcf_chain
*chain
,
1631 struct tcf_proto
*tp
, bool rtnl_held
,
1632 struct netlink_ext_ack
*extack
)
1634 struct tcf_chain_info chain_info
;
1635 struct tcf_proto
*tp_iter
;
1636 struct tcf_proto
**pprev
;
1637 struct tcf_proto
*next
;
1639 mutex_lock(&chain
->filter_chain_lock
);
1641 /* Atomically find and remove tp from chain. */
1642 for (pprev
= &chain
->filter_chain
;
1643 (tp_iter
= tcf_chain_dereference(*pprev
, chain
));
1644 pprev
= &tp_iter
->next
) {
1645 if (tp_iter
== tp
) {
1646 chain_info
.pprev
= pprev
;
1647 chain_info
.next
= tp_iter
->next
;
1648 WARN_ON(tp_iter
->deleting
);
1652 /* Verify that tp still exists and no new filters were inserted
1654 * Mark tp for deletion if it is empty.
1656 if (!tp_iter
|| !tcf_proto_check_delete(tp
, rtnl_held
)) {
1657 mutex_unlock(&chain
->filter_chain_lock
);
1661 next
= tcf_chain_dereference(chain_info
.next
, chain
);
1662 if (tp
== chain
->filter_chain
)
1663 tcf_chain0_head_change(chain
, next
);
1664 RCU_INIT_POINTER(*chain_info
.pprev
, next
);
1665 mutex_unlock(&chain
->filter_chain_lock
);
1667 tcf_proto_put(tp
, rtnl_held
, extack
);
1670 static struct tcf_proto
*tcf_chain_tp_find(struct tcf_chain
*chain
,
1671 struct tcf_chain_info
*chain_info
,
1672 u32 protocol
, u32 prio
,
1675 struct tcf_proto
**pprev
;
1676 struct tcf_proto
*tp
;
1678 /* Check the chain for existence of proto-tcf with this priority */
1679 for (pprev
= &chain
->filter_chain
;
1680 (tp
= tcf_chain_dereference(*pprev
, chain
));
1681 pprev
= &tp
->next
) {
1682 if (tp
->prio
>= prio
) {
1683 if (tp
->prio
== prio
) {
1684 if (prio_allocate
||
1685 (tp
->protocol
!= protocol
&& protocol
))
1686 return ERR_PTR(-EINVAL
);
1693 chain_info
->pprev
= pprev
;
1695 chain_info
->next
= tp
->next
;
1698 chain_info
->next
= NULL
;
1703 static int tcf_fill_node(struct net
*net
, struct sk_buff
*skb
,
1704 struct tcf_proto
*tp
, struct tcf_block
*block
,
1705 struct Qdisc
*q
, u32 parent
, void *fh
,
1706 u32 portid
, u32 seq
, u16 flags
, int event
,
1710 struct nlmsghdr
*nlh
;
1711 unsigned char *b
= skb_tail_pointer(skb
);
1713 nlh
= nlmsg_put(skb
, portid
, seq
, event
, sizeof(*tcm
), flags
);
1715 goto out_nlmsg_trim
;
1716 tcm
= nlmsg_data(nlh
);
1717 tcm
->tcm_family
= AF_UNSPEC
;
1721 tcm
->tcm_ifindex
= qdisc_dev(q
)->ifindex
;
1722 tcm
->tcm_parent
= parent
;
1724 tcm
->tcm_ifindex
= TCM_IFINDEX_MAGIC_BLOCK
;
1725 tcm
->tcm_block_index
= block
->index
;
1727 tcm
->tcm_info
= TC_H_MAKE(tp
->prio
, tp
->protocol
);
1728 if (nla_put_string(skb
, TCA_KIND
, tp
->ops
->kind
))
1729 goto nla_put_failure
;
1730 if (nla_put_u32(skb
, TCA_CHAIN
, tp
->chain
->index
))
1731 goto nla_put_failure
;
1733 tcm
->tcm_handle
= 0;
1735 if (tp
->ops
->dump
&&
1736 tp
->ops
->dump(net
, tp
, fh
, skb
, tcm
, rtnl_held
) < 0)
1737 goto nla_put_failure
;
1739 nlh
->nlmsg_len
= skb_tail_pointer(skb
) - b
;
1748 static int tfilter_notify(struct net
*net
, struct sk_buff
*oskb
,
1749 struct nlmsghdr
*n
, struct tcf_proto
*tp
,
1750 struct tcf_block
*block
, struct Qdisc
*q
,
1751 u32 parent
, void *fh
, int event
, bool unicast
,
1754 struct sk_buff
*skb
;
1755 u32 portid
= oskb
? NETLINK_CB(oskb
).portid
: 0;
1758 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
1762 if (tcf_fill_node(net
, skb
, tp
, block
, q
, parent
, fh
, portid
,
1763 n
->nlmsg_seq
, n
->nlmsg_flags
, event
,
1770 err
= netlink_unicast(net
->rtnl
, skb
, portid
, MSG_DONTWAIT
);
1772 err
= rtnetlink_send(skb
, net
, portid
, RTNLGRP_TC
,
1773 n
->nlmsg_flags
& NLM_F_ECHO
);
1780 static int tfilter_del_notify(struct net
*net
, struct sk_buff
*oskb
,
1781 struct nlmsghdr
*n
, struct tcf_proto
*tp
,
1782 struct tcf_block
*block
, struct Qdisc
*q
,
1783 u32 parent
, void *fh
, bool unicast
, bool *last
,
1784 bool rtnl_held
, struct netlink_ext_ack
*extack
)
1786 struct sk_buff
*skb
;
1787 u32 portid
= oskb
? NETLINK_CB(oskb
).portid
: 0;
1790 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
1794 if (tcf_fill_node(net
, skb
, tp
, block
, q
, parent
, fh
, portid
,
1795 n
->nlmsg_seq
, n
->nlmsg_flags
, RTM_DELTFILTER
,
1797 NL_SET_ERR_MSG(extack
, "Failed to build del event notification");
1802 err
= tp
->ops
->delete(tp
, fh
, last
, rtnl_held
, extack
);
1809 err
= netlink_unicast(net
->rtnl
, skb
, portid
, MSG_DONTWAIT
);
1811 err
= rtnetlink_send(skb
, net
, portid
, RTNLGRP_TC
,
1812 n
->nlmsg_flags
& NLM_F_ECHO
);
1814 NL_SET_ERR_MSG(extack
, "Failed to send filter delete notification");
1821 static void tfilter_notify_chain(struct net
*net
, struct sk_buff
*oskb
,
1822 struct tcf_block
*block
, struct Qdisc
*q
,
1823 u32 parent
, struct nlmsghdr
*n
,
1824 struct tcf_chain
*chain
, int event
,
1827 struct tcf_proto
*tp
;
1829 for (tp
= tcf_get_next_proto(chain
, NULL
, rtnl_held
);
1830 tp
; tp
= tcf_get_next_proto(chain
, tp
, rtnl_held
))
1831 tfilter_notify(net
, oskb
, n
, tp
, block
,
1832 q
, parent
, NULL
, event
, false, rtnl_held
);
1835 static void tfilter_put(struct tcf_proto
*tp
, void *fh
)
1837 if (tp
->ops
->put
&& fh
)
1838 tp
->ops
->put(tp
, fh
);
1841 static int tc_new_tfilter(struct sk_buff
*skb
, struct nlmsghdr
*n
,
1842 struct netlink_ext_ack
*extack
)
1844 struct net
*net
= sock_net(skb
->sk
);
1845 struct nlattr
*tca
[TCA_MAX
+ 1];
1852 struct Qdisc
*q
= NULL
;
1853 struct tcf_chain_info chain_info
;
1854 struct tcf_chain
*chain
= NULL
;
1855 struct tcf_block
*block
;
1856 struct tcf_proto
*tp
;
1861 bool rtnl_held
= false;
1863 if (!netlink_ns_capable(skb
, net
->user_ns
, CAP_NET_ADMIN
))
1869 err
= nlmsg_parse_deprecated(n
, sizeof(*t
), tca
, TCA_MAX
,
1870 rtm_tca_policy
, extack
);
1875 protocol
= TC_H_MIN(t
->tcm_info
);
1876 prio
= TC_H_MAJ(t
->tcm_info
);
1877 prio_allocate
= false;
1878 parent
= t
->tcm_parent
;
1884 /* If no priority is provided by the user,
1887 if (n
->nlmsg_flags
& NLM_F_CREATE
) {
1888 prio
= TC_H_MAKE(0x80000000U
, 0U);
1889 prio_allocate
= true;
1891 NL_SET_ERR_MSG(extack
, "Invalid filter command with priority of zero");
1896 /* Find head of filter chain. */
1898 err
= __tcf_qdisc_find(net
, &q
, &parent
, t
->tcm_ifindex
, false, extack
);
1902 /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
1903 * block is shared (no qdisc found), qdisc is not unlocked, classifier
1904 * type is not specified, classifier is not unlocked.
1907 (q
&& !(q
->ops
->cl_ops
->flags
& QDISC_CLASS_OPS_DOIT_UNLOCKED
)) ||
1908 !tca
[TCA_KIND
] || !tcf_proto_is_unlocked(nla_data(tca
[TCA_KIND
]))) {
1913 err
= __tcf_qdisc_cl_find(q
, parent
, &cl
, t
->tcm_ifindex
, extack
);
1917 block
= __tcf_block_find(net
, q
, cl
, t
->tcm_ifindex
, t
->tcm_block_index
,
1919 if (IS_ERR(block
)) {
1920 err
= PTR_ERR(block
);
1924 chain_index
= tca
[TCA_CHAIN
] ? nla_get_u32(tca
[TCA_CHAIN
]) : 0;
1925 if (chain_index
> TC_ACT_EXT_VAL_MASK
) {
1926 NL_SET_ERR_MSG(extack
, "Specified chain index exceeds upper limit");
1930 chain
= tcf_chain_get(block
, chain_index
, true);
1932 NL_SET_ERR_MSG(extack
, "Cannot create specified filter chain");
1937 mutex_lock(&chain
->filter_chain_lock
);
1938 tp
= tcf_chain_tp_find(chain
, &chain_info
, protocol
,
1939 prio
, prio_allocate
);
1941 NL_SET_ERR_MSG(extack
, "Filter with specified priority/protocol not found");
1947 struct tcf_proto
*tp_new
= NULL
;
1949 if (chain
->flushing
) {
1954 /* Proto-tcf does not exist, create new one */
1956 if (tca
[TCA_KIND
] == NULL
|| !protocol
) {
1957 NL_SET_ERR_MSG(extack
, "Filter kind and protocol must be specified");
1962 if (!(n
->nlmsg_flags
& NLM_F_CREATE
)) {
1963 NL_SET_ERR_MSG(extack
, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
1969 prio
= tcf_auto_prio(tcf_chain_tp_prev(chain
,
1972 mutex_unlock(&chain
->filter_chain_lock
);
1973 tp_new
= tcf_proto_create(nla_data(tca
[TCA_KIND
]),
1974 protocol
, prio
, chain
, rtnl_held
,
1976 if (IS_ERR(tp_new
)) {
1977 err
= PTR_ERR(tp_new
);
1982 tp
= tcf_chain_tp_insert_unique(chain
, tp_new
, protocol
, prio
,
1989 mutex_unlock(&chain
->filter_chain_lock
);
1992 if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], tp
->ops
->kind
)) {
1993 NL_SET_ERR_MSG(extack
, "Specified filter kind does not match existing one");
1998 fh
= tp
->ops
->get(tp
, t
->tcm_handle
);
2001 if (!(n
->nlmsg_flags
& NLM_F_CREATE
)) {
2002 NL_SET_ERR_MSG(extack
, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2006 } else if (n
->nlmsg_flags
& NLM_F_EXCL
) {
2007 tfilter_put(tp
, fh
);
2008 NL_SET_ERR_MSG(extack
, "Filter already exists");
2013 if (chain
->tmplt_ops
&& chain
->tmplt_ops
!= tp
->ops
) {
2014 NL_SET_ERR_MSG(extack
, "Chain template is set to a different filter kind");
2019 err
= tp
->ops
->change(net
, skb
, tp
, cl
, t
->tcm_handle
, tca
, &fh
,
2020 n
->nlmsg_flags
& NLM_F_CREATE
? TCA_ACT_NOREPLACE
: TCA_ACT_REPLACE
,
2023 tfilter_notify(net
, skb
, n
, tp
, block
, q
, parent
, fh
,
2024 RTM_NEWTFILTER
, false, rtnl_held
);
2025 tfilter_put(tp
, fh
);
2026 /* q pointer is NULL for shared blocks */
2028 q
->flags
&= ~TCQ_F_CAN_BYPASS
;
2032 if (err
&& tp_created
)
2033 tcf_chain_tp_delete_empty(chain
, tp
, rtnl_held
, NULL
);
2036 if (tp
&& !IS_ERR(tp
))
2037 tcf_proto_put(tp
, rtnl_held
, NULL
);
2039 tcf_chain_put(chain
);
2041 tcf_block_release(q
, block
, rtnl_held
);
2046 if (err
== -EAGAIN
) {
2047 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2051 /* Replay the request. */
2057 mutex_unlock(&chain
->filter_chain_lock
);
2061 static int tc_del_tfilter(struct sk_buff
*skb
, struct nlmsghdr
*n
,
2062 struct netlink_ext_ack
*extack
)
2064 struct net
*net
= sock_net(skb
->sk
);
2065 struct nlattr
*tca
[TCA_MAX
+ 1];
2071 struct Qdisc
*q
= NULL
;
2072 struct tcf_chain_info chain_info
;
2073 struct tcf_chain
*chain
= NULL
;
2074 struct tcf_block
*block
= NULL
;
2075 struct tcf_proto
*tp
= NULL
;
2076 unsigned long cl
= 0;
2079 bool rtnl_held
= false;
2081 if (!netlink_ns_capable(skb
, net
->user_ns
, CAP_NET_ADMIN
))
2084 err
= nlmsg_parse_deprecated(n
, sizeof(*t
), tca
, TCA_MAX
,
2085 rtm_tca_policy
, extack
);
2090 protocol
= TC_H_MIN(t
->tcm_info
);
2091 prio
= TC_H_MAJ(t
->tcm_info
);
2092 parent
= t
->tcm_parent
;
2094 if (prio
== 0 && (protocol
|| t
->tcm_handle
|| tca
[TCA_KIND
])) {
2095 NL_SET_ERR_MSG(extack
, "Cannot flush filters with protocol, handle or kind set");
2099 /* Find head of filter chain. */
2101 err
= __tcf_qdisc_find(net
, &q
, &parent
, t
->tcm_ifindex
, false, extack
);
2105 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2106 * found), qdisc is not unlocked, classifier type is not specified,
2107 * classifier is not unlocked.
2110 (q
&& !(q
->ops
->cl_ops
->flags
& QDISC_CLASS_OPS_DOIT_UNLOCKED
)) ||
2111 !tca
[TCA_KIND
] || !tcf_proto_is_unlocked(nla_data(tca
[TCA_KIND
]))) {
2116 err
= __tcf_qdisc_cl_find(q
, parent
, &cl
, t
->tcm_ifindex
, extack
);
2120 block
= __tcf_block_find(net
, q
, cl
, t
->tcm_ifindex
, t
->tcm_block_index
,
2122 if (IS_ERR(block
)) {
2123 err
= PTR_ERR(block
);
2127 chain_index
= tca
[TCA_CHAIN
] ? nla_get_u32(tca
[TCA_CHAIN
]) : 0;
2128 if (chain_index
> TC_ACT_EXT_VAL_MASK
) {
2129 NL_SET_ERR_MSG(extack
, "Specified chain index exceeds upper limit");
2133 chain
= tcf_chain_get(block
, chain_index
, false);
2135 /* User requested flush on non-existent chain. Nothing to do,
2136 * so just return success.
2142 NL_SET_ERR_MSG(extack
, "Cannot find specified filter chain");
2148 tfilter_notify_chain(net
, skb
, block
, q
, parent
, n
,
2149 chain
, RTM_DELTFILTER
, rtnl_held
);
2150 tcf_chain_flush(chain
, rtnl_held
);
2155 mutex_lock(&chain
->filter_chain_lock
);
2156 tp
= tcf_chain_tp_find(chain
, &chain_info
, protocol
,
2158 if (!tp
|| IS_ERR(tp
)) {
2159 NL_SET_ERR_MSG(extack
, "Filter with specified priority/protocol not found");
2160 err
= tp
? PTR_ERR(tp
) : -ENOENT
;
2162 } else if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], tp
->ops
->kind
)) {
2163 NL_SET_ERR_MSG(extack
, "Specified filter kind does not match existing one");
2166 } else if (t
->tcm_handle
== 0) {
2167 tcf_chain_tp_remove(chain
, &chain_info
, tp
);
2168 mutex_unlock(&chain
->filter_chain_lock
);
2170 tcf_proto_put(tp
, rtnl_held
, NULL
);
2171 tfilter_notify(net
, skb
, n
, tp
, block
, q
, parent
, fh
,
2172 RTM_DELTFILTER
, false, rtnl_held
);
2176 mutex_unlock(&chain
->filter_chain_lock
);
2178 fh
= tp
->ops
->get(tp
, t
->tcm_handle
);
2181 NL_SET_ERR_MSG(extack
, "Specified filter handle not found");
2186 err
= tfilter_del_notify(net
, skb
, n
, tp
, block
,
2187 q
, parent
, fh
, false, &last
,
2193 tcf_chain_tp_delete_empty(chain
, tp
, rtnl_held
, extack
);
2198 if (tp
&& !IS_ERR(tp
))
2199 tcf_proto_put(tp
, rtnl_held
, NULL
);
2200 tcf_chain_put(chain
);
2202 tcf_block_release(q
, block
, rtnl_held
);
2210 mutex_unlock(&chain
->filter_chain_lock
);
2214 static int tc_get_tfilter(struct sk_buff
*skb
, struct nlmsghdr
*n
,
2215 struct netlink_ext_ack
*extack
)
2217 struct net
*net
= sock_net(skb
->sk
);
2218 struct nlattr
*tca
[TCA_MAX
+ 1];
2224 struct Qdisc
*q
= NULL
;
2225 struct tcf_chain_info chain_info
;
2226 struct tcf_chain
*chain
= NULL
;
2227 struct tcf_block
*block
= NULL
;
2228 struct tcf_proto
*tp
= NULL
;
2229 unsigned long cl
= 0;
2232 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 parent
= t
->tcm_parent
;
2245 NL_SET_ERR_MSG(extack
, "Invalid filter command with priority of zero");
2249 /* Find head of filter chain. */
2251 err
= __tcf_qdisc_find(net
, &q
, &parent
, t
->tcm_ifindex
, false, extack
);
2255 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2256 * unlocked, classifier type is not specified, classifier is not
2259 if ((q
&& !(q
->ops
->cl_ops
->flags
& QDISC_CLASS_OPS_DOIT_UNLOCKED
)) ||
2260 !tca
[TCA_KIND
] || !tcf_proto_is_unlocked(nla_data(tca
[TCA_KIND
]))) {
2265 err
= __tcf_qdisc_cl_find(q
, parent
, &cl
, t
->tcm_ifindex
, extack
);
2269 block
= __tcf_block_find(net
, q
, cl
, t
->tcm_ifindex
, t
->tcm_block_index
,
2271 if (IS_ERR(block
)) {
2272 err
= PTR_ERR(block
);
2276 chain_index
= tca
[TCA_CHAIN
] ? nla_get_u32(tca
[TCA_CHAIN
]) : 0;
2277 if (chain_index
> TC_ACT_EXT_VAL_MASK
) {
2278 NL_SET_ERR_MSG(extack
, "Specified chain index exceeds upper limit");
2282 chain
= tcf_chain_get(block
, chain_index
, false);
2284 NL_SET_ERR_MSG(extack
, "Cannot find specified filter chain");
2289 mutex_lock(&chain
->filter_chain_lock
);
2290 tp
= tcf_chain_tp_find(chain
, &chain_info
, protocol
,
2292 mutex_unlock(&chain
->filter_chain_lock
);
2293 if (!tp
|| IS_ERR(tp
)) {
2294 NL_SET_ERR_MSG(extack
, "Filter with specified priority/protocol not found");
2295 err
= tp
? PTR_ERR(tp
) : -ENOENT
;
2297 } else if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], tp
->ops
->kind
)) {
2298 NL_SET_ERR_MSG(extack
, "Specified filter kind does not match existing one");
2303 fh
= tp
->ops
->get(tp
, t
->tcm_handle
);
2306 NL_SET_ERR_MSG(extack
, "Specified filter handle not found");
2309 err
= tfilter_notify(net
, skb
, n
, tp
, block
, q
, parent
,
2310 fh
, RTM_NEWTFILTER
, true, rtnl_held
);
2312 NL_SET_ERR_MSG(extack
, "Failed to send filter notify message");
2315 tfilter_put(tp
, fh
);
2318 if (tp
&& !IS_ERR(tp
))
2319 tcf_proto_put(tp
, rtnl_held
, NULL
);
2320 tcf_chain_put(chain
);
2322 tcf_block_release(q
, block
, rtnl_held
);
2330 struct tcf_dump_args
{
2331 struct tcf_walker w
;
2332 struct sk_buff
*skb
;
2333 struct netlink_callback
*cb
;
2334 struct tcf_block
*block
;
2339 static int tcf_node_dump(struct tcf_proto
*tp
, void *n
, struct tcf_walker
*arg
)
2341 struct tcf_dump_args
*a
= (void *)arg
;
2342 struct net
*net
= sock_net(a
->skb
->sk
);
2344 return tcf_fill_node(net
, a
->skb
, tp
, a
->block
, a
->q
, a
->parent
,
2345 n
, NETLINK_CB(a
->cb
->skb
).portid
,
2346 a
->cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
2347 RTM_NEWTFILTER
, true);
2350 static bool tcf_chain_dump(struct tcf_chain
*chain
, struct Qdisc
*q
, u32 parent
,
2351 struct sk_buff
*skb
, struct netlink_callback
*cb
,
2352 long index_start
, long *p_index
)
2354 struct net
*net
= sock_net(skb
->sk
);
2355 struct tcf_block
*block
= chain
->block
;
2356 struct tcmsg
*tcm
= nlmsg_data(cb
->nlh
);
2357 struct tcf_proto
*tp
, *tp_prev
;
2358 struct tcf_dump_args arg
;
2360 for (tp
= __tcf_get_next_proto(chain
, NULL
);
2363 tp
= __tcf_get_next_proto(chain
, tp
),
2364 tcf_proto_put(tp_prev
, true, NULL
),
2366 if (*p_index
< index_start
)
2368 if (TC_H_MAJ(tcm
->tcm_info
) &&
2369 TC_H_MAJ(tcm
->tcm_info
) != tp
->prio
)
2371 if (TC_H_MIN(tcm
->tcm_info
) &&
2372 TC_H_MIN(tcm
->tcm_info
) != tp
->protocol
)
2374 if (*p_index
> index_start
)
2375 memset(&cb
->args
[1], 0,
2376 sizeof(cb
->args
) - sizeof(cb
->args
[0]));
2377 if (cb
->args
[1] == 0) {
2378 if (tcf_fill_node(net
, skb
, tp
, block
, q
, parent
, NULL
,
2379 NETLINK_CB(cb
->skb
).portid
,
2380 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
2381 RTM_NEWTFILTER
, true) <= 0)
2387 arg
.w
.fn
= tcf_node_dump
;
2392 arg
.parent
= parent
;
2394 arg
.w
.skip
= cb
->args
[1] - 1;
2396 arg
.w
.cookie
= cb
->args
[2];
2397 tp
->ops
->walk(tp
, &arg
.w
, true);
2398 cb
->args
[2] = arg
.w
.cookie
;
2399 cb
->args
[1] = arg
.w
.count
+ 1;
2406 tcf_proto_put(tp
, true, NULL
);
2410 /* called with RTNL */
2411 static int tc_dump_tfilter(struct sk_buff
*skb
, struct netlink_callback
*cb
)
2413 struct tcf_chain
*chain
, *chain_prev
;
2414 struct net
*net
= sock_net(skb
->sk
);
2415 struct nlattr
*tca
[TCA_MAX
+ 1];
2416 struct Qdisc
*q
= NULL
;
2417 struct tcf_block
*block
;
2418 struct tcmsg
*tcm
= nlmsg_data(cb
->nlh
);
2424 if (nlmsg_len(cb
->nlh
) < sizeof(*tcm
))
2427 err
= nlmsg_parse_deprecated(cb
->nlh
, sizeof(*tcm
), tca
, TCA_MAX
,
2432 if (tcm
->tcm_ifindex
== TCM_IFINDEX_MAGIC_BLOCK
) {
2433 block
= tcf_block_refcnt_get(net
, tcm
->tcm_block_index
);
2436 /* If we work with block index, q is NULL and parent value
2437 * will never be used in the following code. The check
2438 * in tcf_fill_node prevents it. However, compiler does not
2439 * see that far, so set parent to zero to silence the warning
2440 * about parent being uninitialized.
2444 const struct Qdisc_class_ops
*cops
;
2445 struct net_device
*dev
;
2446 unsigned long cl
= 0;
2448 dev
= __dev_get_by_index(net
, tcm
->tcm_ifindex
);
2452 parent
= tcm
->tcm_parent
;
2457 q
= qdisc_lookup(dev
, TC_H_MAJ(tcm
->tcm_parent
));
2461 cops
= q
->ops
->cl_ops
;
2464 if (!cops
->tcf_block
)
2466 if (TC_H_MIN(tcm
->tcm_parent
)) {
2467 cl
= cops
->find(q
, tcm
->tcm_parent
);
2471 block
= cops
->tcf_block(q
, cl
, NULL
);
2474 if (tcf_block_shared(block
))
2478 index_start
= cb
->args
[0];
2481 for (chain
= __tcf_get_next_chain(block
, NULL
);
2484 chain
= __tcf_get_next_chain(block
, chain
),
2485 tcf_chain_put(chain_prev
)) {
2486 if (tca
[TCA_CHAIN
] &&
2487 nla_get_u32(tca
[TCA_CHAIN
]) != chain
->index
)
2489 if (!tcf_chain_dump(chain
, q
, parent
, skb
, cb
,
2490 index_start
, &index
)) {
2491 tcf_chain_put(chain
);
2497 if (tcm
->tcm_ifindex
== TCM_IFINDEX_MAGIC_BLOCK
)
2498 tcf_block_refcnt_put(block
, true);
2499 cb
->args
[0] = index
;
2502 /* If we did no progress, the error (EMSGSIZE) is real */
2503 if (skb
->len
== 0 && err
)
2508 static int tc_chain_fill_node(const struct tcf_proto_ops
*tmplt_ops
,
2509 void *tmplt_priv
, u32 chain_index
,
2510 struct net
*net
, struct sk_buff
*skb
,
2511 struct tcf_block
*block
,
2512 u32 portid
, u32 seq
, u16 flags
, int event
)
2514 unsigned char *b
= skb_tail_pointer(skb
);
2515 const struct tcf_proto_ops
*ops
;
2516 struct nlmsghdr
*nlh
;
2523 nlh
= nlmsg_put(skb
, portid
, seq
, event
, sizeof(*tcm
), flags
);
2525 goto out_nlmsg_trim
;
2526 tcm
= nlmsg_data(nlh
);
2527 tcm
->tcm_family
= AF_UNSPEC
;
2530 tcm
->tcm_handle
= 0;
2532 tcm
->tcm_ifindex
= qdisc_dev(block
->q
)->ifindex
;
2533 tcm
->tcm_parent
= block
->q
->handle
;
2535 tcm
->tcm_ifindex
= TCM_IFINDEX_MAGIC_BLOCK
;
2536 tcm
->tcm_block_index
= block
->index
;
2539 if (nla_put_u32(skb
, TCA_CHAIN
, chain_index
))
2540 goto nla_put_failure
;
2543 if (nla_put_string(skb
, TCA_KIND
, ops
->kind
))
2544 goto nla_put_failure
;
2545 if (ops
->tmplt_dump(skb
, net
, priv
) < 0)
2546 goto nla_put_failure
;
2549 nlh
->nlmsg_len
= skb_tail_pointer(skb
) - b
;
2558 static int tc_chain_notify(struct tcf_chain
*chain
, struct sk_buff
*oskb
,
2559 u32 seq
, u16 flags
, int event
, bool unicast
)
2561 u32 portid
= oskb
? NETLINK_CB(oskb
).portid
: 0;
2562 struct tcf_block
*block
= chain
->block
;
2563 struct net
*net
= block
->net
;
2564 struct sk_buff
*skb
;
2567 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
2571 if (tc_chain_fill_node(chain
->tmplt_ops
, chain
->tmplt_priv
,
2572 chain
->index
, net
, skb
, block
, portid
,
2573 seq
, flags
, event
) <= 0) {
2579 err
= netlink_unicast(net
->rtnl
, skb
, portid
, MSG_DONTWAIT
);
2581 err
= rtnetlink_send(skb
, net
, portid
, RTNLGRP_TC
,
2582 flags
& NLM_F_ECHO
);
2589 static int tc_chain_notify_delete(const struct tcf_proto_ops
*tmplt_ops
,
2590 void *tmplt_priv
, u32 chain_index
,
2591 struct tcf_block
*block
, struct sk_buff
*oskb
,
2592 u32 seq
, u16 flags
, bool unicast
)
2594 u32 portid
= oskb
? NETLINK_CB(oskb
).portid
: 0;
2595 struct net
*net
= block
->net
;
2596 struct sk_buff
*skb
;
2598 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
2602 if (tc_chain_fill_node(tmplt_ops
, tmplt_priv
, chain_index
, net
, skb
,
2603 block
, portid
, seq
, flags
, RTM_DELCHAIN
) <= 0) {
2609 return netlink_unicast(net
->rtnl
, skb
, portid
, MSG_DONTWAIT
);
2611 return rtnetlink_send(skb
, net
, portid
, RTNLGRP_TC
, flags
& NLM_F_ECHO
);
2614 static int tc_chain_tmplt_add(struct tcf_chain
*chain
, struct net
*net
,
2615 struct nlattr
**tca
,
2616 struct netlink_ext_ack
*extack
)
2618 const struct tcf_proto_ops
*ops
;
2621 /* If kind is not set, user did not specify template. */
2625 ops
= tcf_proto_lookup_ops(nla_data(tca
[TCA_KIND
]), true, extack
);
2627 return PTR_ERR(ops
);
2628 if (!ops
->tmplt_create
|| !ops
->tmplt_destroy
|| !ops
->tmplt_dump
) {
2629 NL_SET_ERR_MSG(extack
, "Chain templates are not supported with specified classifier");
2633 tmplt_priv
= ops
->tmplt_create(net
, chain
, tca
, extack
);
2634 if (IS_ERR(tmplt_priv
)) {
2635 module_put(ops
->owner
);
2636 return PTR_ERR(tmplt_priv
);
2638 chain
->tmplt_ops
= ops
;
2639 chain
->tmplt_priv
= tmplt_priv
;
2643 static void tc_chain_tmplt_del(const struct tcf_proto_ops
*tmplt_ops
,
2646 /* If template ops are set, no work to do for us. */
2650 tmplt_ops
->tmplt_destroy(tmplt_priv
);
2651 module_put(tmplt_ops
->owner
);
2654 /* Add/delete/get a chain */
2656 static int tc_ctl_chain(struct sk_buff
*skb
, struct nlmsghdr
*n
,
2657 struct netlink_ext_ack
*extack
)
2659 struct net
*net
= sock_net(skb
->sk
);
2660 struct nlattr
*tca
[TCA_MAX
+ 1];
2664 struct Qdisc
*q
= NULL
;
2665 struct tcf_chain
*chain
= NULL
;
2666 struct tcf_block
*block
;
2670 if (n
->nlmsg_type
!= RTM_GETCHAIN
&&
2671 !netlink_ns_capable(skb
, net
->user_ns
, CAP_NET_ADMIN
))
2675 err
= nlmsg_parse_deprecated(n
, sizeof(*t
), tca
, TCA_MAX
,
2676 rtm_tca_policy
, extack
);
2681 parent
= t
->tcm_parent
;
2684 block
= tcf_block_find(net
, &q
, &parent
, &cl
,
2685 t
->tcm_ifindex
, t
->tcm_block_index
, extack
);
2687 return PTR_ERR(block
);
2689 chain_index
= tca
[TCA_CHAIN
] ? nla_get_u32(tca
[TCA_CHAIN
]) : 0;
2690 if (chain_index
> TC_ACT_EXT_VAL_MASK
) {
2691 NL_SET_ERR_MSG(extack
, "Specified chain index exceeds upper limit");
2696 mutex_lock(&block
->lock
);
2697 chain
= tcf_chain_lookup(block
, chain_index
);
2698 if (n
->nlmsg_type
== RTM_NEWCHAIN
) {
2700 if (tcf_chain_held_by_acts_only(chain
)) {
2701 /* The chain exists only because there is
2702 * some action referencing it.
2704 tcf_chain_hold(chain
);
2706 NL_SET_ERR_MSG(extack
, "Filter chain already exists");
2708 goto errout_block_locked
;
2711 if (!(n
->nlmsg_flags
& NLM_F_CREATE
)) {
2712 NL_SET_ERR_MSG(extack
, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
2714 goto errout_block_locked
;
2716 chain
= tcf_chain_create(block
, chain_index
);
2718 NL_SET_ERR_MSG(extack
, "Failed to create filter chain");
2720 goto errout_block_locked
;
2724 if (!chain
|| tcf_chain_held_by_acts_only(chain
)) {
2725 NL_SET_ERR_MSG(extack
, "Cannot find specified filter chain");
2727 goto errout_block_locked
;
2729 tcf_chain_hold(chain
);
2732 if (n
->nlmsg_type
== RTM_NEWCHAIN
) {
2733 /* Modifying chain requires holding parent block lock. In case
2734 * the chain was successfully added, take a reference to the
2735 * chain. This ensures that an empty chain does not disappear at
2736 * the end of this function.
2738 tcf_chain_hold(chain
);
2739 chain
->explicitly_created
= true;
2741 mutex_unlock(&block
->lock
);
2743 switch (n
->nlmsg_type
) {
2745 err
= tc_chain_tmplt_add(chain
, net
, tca
, extack
);
2747 tcf_chain_put_explicitly_created(chain
);
2751 tc_chain_notify(chain
, NULL
, 0, NLM_F_CREATE
| NLM_F_EXCL
,
2752 RTM_NEWCHAIN
, false);
2755 tfilter_notify_chain(net
, skb
, block
, q
, parent
, n
,
2756 chain
, RTM_DELTFILTER
, true);
2757 /* Flush the chain first as the user requested chain removal. */
2758 tcf_chain_flush(chain
, true);
2759 /* In case the chain was successfully deleted, put a reference
2760 * to the chain previously taken during addition.
2762 tcf_chain_put_explicitly_created(chain
);
2765 err
= tc_chain_notify(chain
, skb
, n
->nlmsg_seq
,
2766 n
->nlmsg_seq
, n
->nlmsg_type
, true);
2768 NL_SET_ERR_MSG(extack
, "Failed to send chain notify message");
2772 NL_SET_ERR_MSG(extack
, "Unsupported message type");
2777 tcf_chain_put(chain
);
2779 tcf_block_release(q
, block
, true);
2781 /* Replay the request. */
2785 errout_block_locked
:
2786 mutex_unlock(&block
->lock
);
2790 /* called with RTNL */
2791 static int tc_dump_chain(struct sk_buff
*skb
, struct netlink_callback
*cb
)
2793 struct net
*net
= sock_net(skb
->sk
);
2794 struct nlattr
*tca
[TCA_MAX
+ 1];
2795 struct Qdisc
*q
= NULL
;
2796 struct tcf_block
*block
;
2797 struct tcmsg
*tcm
= nlmsg_data(cb
->nlh
);
2798 struct tcf_chain
*chain
;
2804 if (nlmsg_len(cb
->nlh
) < sizeof(*tcm
))
2807 err
= nlmsg_parse_deprecated(cb
->nlh
, sizeof(*tcm
), tca
, TCA_MAX
,
2808 rtm_tca_policy
, cb
->extack
);
2812 if (tcm
->tcm_ifindex
== TCM_IFINDEX_MAGIC_BLOCK
) {
2813 block
= tcf_block_refcnt_get(net
, tcm
->tcm_block_index
);
2816 /* If we work with block index, q is NULL and parent value
2817 * will never be used in the following code. The check
2818 * in tcf_fill_node prevents it. However, compiler does not
2819 * see that far, so set parent to zero to silence the warning
2820 * about parent being uninitialized.
2824 const struct Qdisc_class_ops
*cops
;
2825 struct net_device
*dev
;
2826 unsigned long cl
= 0;
2828 dev
= __dev_get_by_index(net
, tcm
->tcm_ifindex
);
2832 parent
= tcm
->tcm_parent
;
2837 q
= qdisc_lookup(dev
, TC_H_MAJ(tcm
->tcm_parent
));
2841 cops
= q
->ops
->cl_ops
;
2844 if (!cops
->tcf_block
)
2846 if (TC_H_MIN(tcm
->tcm_parent
)) {
2847 cl
= cops
->find(q
, tcm
->tcm_parent
);
2851 block
= cops
->tcf_block(q
, cl
, NULL
);
2854 if (tcf_block_shared(block
))
2858 index_start
= cb
->args
[0];
2861 mutex_lock(&block
->lock
);
2862 list_for_each_entry(chain
, &block
->chain_list
, list
) {
2863 if ((tca
[TCA_CHAIN
] &&
2864 nla_get_u32(tca
[TCA_CHAIN
]) != chain
->index
))
2866 if (index
< index_start
) {
2870 if (tcf_chain_held_by_acts_only(chain
))
2872 err
= tc_chain_fill_node(chain
->tmplt_ops
, chain
->tmplt_priv
,
2873 chain
->index
, net
, skb
, block
,
2874 NETLINK_CB(cb
->skb
).portid
,
2875 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
2881 mutex_unlock(&block
->lock
);
2883 if (tcm
->tcm_ifindex
== TCM_IFINDEX_MAGIC_BLOCK
)
2884 tcf_block_refcnt_put(block
, true);
2885 cb
->args
[0] = index
;
2888 /* If we did no progress, the error (EMSGSIZE) is real */
2889 if (skb
->len
== 0 && err
)
2894 void tcf_exts_destroy(struct tcf_exts
*exts
)
2896 #ifdef CONFIG_NET_CLS_ACT
2897 tcf_action_destroy(exts
->actions
, TCA_ACT_UNBIND
);
2898 kfree(exts
->actions
);
2899 exts
->nr_actions
= 0;
2902 EXPORT_SYMBOL(tcf_exts_destroy
);
2904 int tcf_exts_validate(struct net
*net
, struct tcf_proto
*tp
, struct nlattr
**tb
,
2905 struct nlattr
*rate_tlv
, struct tcf_exts
*exts
, bool ovr
,
2906 bool rtnl_held
, struct netlink_ext_ack
*extack
)
2908 #ifdef CONFIG_NET_CLS_ACT
2910 struct tc_action
*act
;
2911 size_t attr_size
= 0;
2913 if (exts
->police
&& tb
[exts
->police
]) {
2914 act
= tcf_action_init_1(net
, tp
, tb
[exts
->police
],
2915 rate_tlv
, "police", ovr
,
2916 TCA_ACT_BIND
, rtnl_held
,
2919 return PTR_ERR(act
);
2921 act
->type
= exts
->type
= TCA_OLD_COMPAT
;
2922 exts
->actions
[0] = act
;
2923 exts
->nr_actions
= 1;
2924 } else if (exts
->action
&& tb
[exts
->action
]) {
2927 err
= tcf_action_init(net
, tp
, tb
[exts
->action
],
2928 rate_tlv
, NULL
, ovr
, TCA_ACT_BIND
,
2929 exts
->actions
, &attr_size
,
2933 exts
->nr_actions
= err
;
2937 if ((exts
->action
&& tb
[exts
->action
]) ||
2938 (exts
->police
&& tb
[exts
->police
])) {
2939 NL_SET_ERR_MSG(extack
, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
2946 EXPORT_SYMBOL(tcf_exts_validate
);
2948 void tcf_exts_change(struct tcf_exts
*dst
, struct tcf_exts
*src
)
2950 #ifdef CONFIG_NET_CLS_ACT
2951 struct tcf_exts old
= *dst
;
2954 tcf_exts_destroy(&old
);
2957 EXPORT_SYMBOL(tcf_exts_change
);
2959 #ifdef CONFIG_NET_CLS_ACT
2960 static struct tc_action
*tcf_exts_first_act(struct tcf_exts
*exts
)
2962 if (exts
->nr_actions
== 0)
2965 return exts
->actions
[0];
2969 int tcf_exts_dump(struct sk_buff
*skb
, struct tcf_exts
*exts
)
2971 #ifdef CONFIG_NET_CLS_ACT
2972 struct nlattr
*nest
;
2974 if (exts
->action
&& tcf_exts_has_actions(exts
)) {
2976 * again for backward compatible mode - we want
2977 * to work with both old and new modes of entering
2978 * tc data even if iproute2 was newer - jhs
2980 if (exts
->type
!= TCA_OLD_COMPAT
) {
2981 nest
= nla_nest_start_noflag(skb
, exts
->action
);
2983 goto nla_put_failure
;
2985 if (tcf_action_dump(skb
, exts
->actions
, 0, 0) < 0)
2986 goto nla_put_failure
;
2987 nla_nest_end(skb
, nest
);
2988 } else if (exts
->police
) {
2989 struct tc_action
*act
= tcf_exts_first_act(exts
);
2990 nest
= nla_nest_start_noflag(skb
, exts
->police
);
2991 if (nest
== NULL
|| !act
)
2992 goto nla_put_failure
;
2993 if (tcf_action_dump_old(skb
, act
, 0, 0) < 0)
2994 goto nla_put_failure
;
2995 nla_nest_end(skb
, nest
);
3001 nla_nest_cancel(skb
, nest
);
3007 EXPORT_SYMBOL(tcf_exts_dump
);
3010 int tcf_exts_dump_stats(struct sk_buff
*skb
, struct tcf_exts
*exts
)
3012 #ifdef CONFIG_NET_CLS_ACT
3013 struct tc_action
*a
= tcf_exts_first_act(exts
);
3014 if (a
!= NULL
&& tcf_action_copy_stats(skb
, a
, 1) < 0)
3019 EXPORT_SYMBOL(tcf_exts_dump_stats
);
3021 static void tcf_block_offload_inc(struct tcf_block
*block
, u32
*flags
)
3023 if (*flags
& TCA_CLS_FLAGS_IN_HW
)
3025 *flags
|= TCA_CLS_FLAGS_IN_HW
;
3026 atomic_inc(&block
->offloadcnt
);
3029 static void tcf_block_offload_dec(struct tcf_block
*block
, u32
*flags
)
3031 if (!(*flags
& TCA_CLS_FLAGS_IN_HW
))
3033 *flags
&= ~TCA_CLS_FLAGS_IN_HW
;
3034 atomic_dec(&block
->offloadcnt
);
3037 static void tc_cls_offload_cnt_update(struct tcf_block
*block
,
3038 struct tcf_proto
*tp
, u32
*cnt
,
3039 u32
*flags
, u32 diff
, bool add
)
3041 lockdep_assert_held(&block
->cb_lock
);
3043 spin_lock(&tp
->lock
);
3046 tcf_block_offload_inc(block
, flags
);
3051 tcf_block_offload_dec(block
, flags
);
3053 spin_unlock(&tp
->lock
);
3057 tc_cls_offload_cnt_reset(struct tcf_block
*block
, struct tcf_proto
*tp
,
3058 u32
*cnt
, u32
*flags
)
3060 lockdep_assert_held(&block
->cb_lock
);
3062 spin_lock(&tp
->lock
);
3063 tcf_block_offload_dec(block
, flags
);
3065 spin_unlock(&tp
->lock
);
3069 __tc_setup_cb_call(struct tcf_block
*block
, enum tc_setup_type type
,
3070 void *type_data
, bool err_stop
)
3072 struct flow_block_cb
*block_cb
;
3076 list_for_each_entry(block_cb
, &block
->flow_block
.cb_list
, list
) {
3077 err
= block_cb
->cb(type
, type_data
, block_cb
->cb_priv
);
3088 int tc_setup_cb_call(struct tcf_block
*block
, enum tc_setup_type type
,
3089 void *type_data
, bool err_stop
, bool rtnl_held
)
3091 bool take_rtnl
= READ_ONCE(block
->lockeddevcnt
) && !rtnl_held
;
3097 down_read(&block
->cb_lock
);
3098 /* Need to obtain rtnl lock if block is bound to devs that require it.
3099 * In block bind code cb_lock is obtained while holding rtnl, so we must
3100 * obtain the locks in same order here.
3102 if (!rtnl_held
&& !take_rtnl
&& block
->lockeddevcnt
) {
3103 up_read(&block
->cb_lock
);
3108 ok_count
= __tc_setup_cb_call(block
, type
, type_data
, err_stop
);
3110 up_read(&block
->cb_lock
);
3115 EXPORT_SYMBOL(tc_setup_cb_call
);
3117 /* Non-destructive filter add. If filter that wasn't already in hardware is
3118 * successfully offloaded, increment block offloads counter. On failure,
3119 * previously offloaded filter is considered to be intact and offloads counter
3120 * is not decremented.
3123 int tc_setup_cb_add(struct tcf_block
*block
, struct tcf_proto
*tp
,
3124 enum tc_setup_type type
, void *type_data
, bool err_stop
,
3125 u32
*flags
, unsigned int *in_hw_count
, bool rtnl_held
)
3127 bool take_rtnl
= READ_ONCE(block
->lockeddevcnt
) && !rtnl_held
;
3133 down_read(&block
->cb_lock
);
3134 /* Need to obtain rtnl lock if block is bound to devs that require it.
3135 * In block bind code cb_lock is obtained while holding rtnl, so we must
3136 * obtain the locks in same order here.
3138 if (!rtnl_held
&& !take_rtnl
&& block
->lockeddevcnt
) {
3139 up_read(&block
->cb_lock
);
3144 /* Make sure all netdevs sharing this block are offload-capable. */
3145 if (block
->nooffloaddevcnt
&& err_stop
) {
3146 ok_count
= -EOPNOTSUPP
;
3150 ok_count
= __tc_setup_cb_call(block
, type
, type_data
, err_stop
);
3154 if (tp
->ops
->hw_add
)
3155 tp
->ops
->hw_add(tp
, type_data
);
3157 tc_cls_offload_cnt_update(block
, tp
, in_hw_count
, flags
,
3160 up_read(&block
->cb_lock
);
3163 return ok_count
< 0 ? ok_count
: 0;
3165 EXPORT_SYMBOL(tc_setup_cb_add
);
3167 /* Destructive filter replace. If filter that wasn't already in hardware is
3168 * successfully offloaded, increment block offload counter. On failure,
3169 * previously offloaded filter is considered to be destroyed and offload counter
3173 int tc_setup_cb_replace(struct tcf_block
*block
, struct tcf_proto
*tp
,
3174 enum tc_setup_type type
, void *type_data
, bool err_stop
,
3175 u32
*old_flags
, unsigned int *old_in_hw_count
,
3176 u32
*new_flags
, unsigned int *new_in_hw_count
,
3179 bool take_rtnl
= READ_ONCE(block
->lockeddevcnt
) && !rtnl_held
;
3185 down_read(&block
->cb_lock
);
3186 /* Need to obtain rtnl lock if block is bound to devs that require it.
3187 * In block bind code cb_lock is obtained while holding rtnl, so we must
3188 * obtain the locks in same order here.
3190 if (!rtnl_held
&& !take_rtnl
&& block
->lockeddevcnt
) {
3191 up_read(&block
->cb_lock
);
3196 /* Make sure all netdevs sharing this block are offload-capable. */
3197 if (block
->nooffloaddevcnt
&& err_stop
) {
3198 ok_count
= -EOPNOTSUPP
;
3202 tc_cls_offload_cnt_reset(block
, tp
, old_in_hw_count
, old_flags
);
3203 if (tp
->ops
->hw_del
)
3204 tp
->ops
->hw_del(tp
, type_data
);
3206 ok_count
= __tc_setup_cb_call(block
, type
, type_data
, err_stop
);
3210 if (tp
->ops
->hw_add
)
3211 tp
->ops
->hw_add(tp
, type_data
);
3213 tc_cls_offload_cnt_update(block
, tp
, new_in_hw_count
,
3214 new_flags
, ok_count
, true);
3216 up_read(&block
->cb_lock
);
3219 return ok_count
< 0 ? ok_count
: 0;
3221 EXPORT_SYMBOL(tc_setup_cb_replace
);
3223 /* Destroy filter and decrement block offload counter, if filter was previously
3227 int tc_setup_cb_destroy(struct tcf_block
*block
, struct tcf_proto
*tp
,
3228 enum tc_setup_type type
, void *type_data
, bool err_stop
,
3229 u32
*flags
, unsigned int *in_hw_count
, bool rtnl_held
)
3231 bool take_rtnl
= READ_ONCE(block
->lockeddevcnt
) && !rtnl_held
;
3237 down_read(&block
->cb_lock
);
3238 /* Need to obtain rtnl lock if block is bound to devs that require it.
3239 * In block bind code cb_lock is obtained while holding rtnl, so we must
3240 * obtain the locks in same order here.
3242 if (!rtnl_held
&& !take_rtnl
&& block
->lockeddevcnt
) {
3243 up_read(&block
->cb_lock
);
3248 ok_count
= __tc_setup_cb_call(block
, type
, type_data
, err_stop
);
3250 tc_cls_offload_cnt_reset(block
, tp
, in_hw_count
, flags
);
3251 if (tp
->ops
->hw_del
)
3252 tp
->ops
->hw_del(tp
, type_data
);
3254 up_read(&block
->cb_lock
);
3257 return ok_count
< 0 ? ok_count
: 0;
3259 EXPORT_SYMBOL(tc_setup_cb_destroy
);
3261 int tc_setup_cb_reoffload(struct tcf_block
*block
, struct tcf_proto
*tp
,
3262 bool add
, flow_setup_cb_t
*cb
,
3263 enum tc_setup_type type
, void *type_data
,
3264 void *cb_priv
, u32
*flags
, unsigned int *in_hw_count
)
3266 int err
= cb(type
, type_data
, cb_priv
);
3269 if (add
&& tc_skip_sw(*flags
))
3272 tc_cls_offload_cnt_update(block
, tp
, in_hw_count
, flags
, 1,
3278 EXPORT_SYMBOL(tc_setup_cb_reoffload
);
3280 void tc_cleanup_flow_action(struct flow_action
*flow_action
)
3282 struct flow_action_entry
*entry
;
3285 flow_action_for_each(i
, entry
, flow_action
)
3286 if (entry
->destructor
)
3287 entry
->destructor(entry
->destructor_priv
);
3289 EXPORT_SYMBOL(tc_cleanup_flow_action
);
3291 static void tcf_mirred_get_dev(struct flow_action_entry
*entry
,
3292 const struct tc_action
*act
)
3294 #ifdef CONFIG_NET_CLS_ACT
3295 entry
->dev
= act
->ops
->get_dev(act
, &entry
->destructor
);
3298 entry
->destructor_priv
= entry
->dev
;
3302 static void tcf_tunnel_encap_put_tunnel(void *priv
)
3304 struct ip_tunnel_info
*tunnel
= priv
;
3309 static int tcf_tunnel_encap_get_tunnel(struct flow_action_entry
*entry
,
3310 const struct tc_action
*act
)
3312 entry
->tunnel
= tcf_tunnel_info_copy(act
);
3315 entry
->destructor
= tcf_tunnel_encap_put_tunnel
;
3316 entry
->destructor_priv
= entry
->tunnel
;
3320 static void tcf_sample_get_group(struct flow_action_entry
*entry
,
3321 const struct tc_action
*act
)
3323 #ifdef CONFIG_NET_CLS_ACT
3324 entry
->sample
.psample_group
=
3325 act
->ops
->get_psample_group(act
, &entry
->destructor
);
3326 entry
->destructor_priv
= entry
->sample
.psample_group
;
3330 int tc_setup_flow_action(struct flow_action
*flow_action
,
3331 const struct tcf_exts
*exts
, bool rtnl_held
)
3333 const struct tc_action
*act
;
3334 int i
, j
, k
, err
= 0;
3343 tcf_exts_for_each_action(i
, act
, exts
) {
3344 struct flow_action_entry
*entry
;
3346 entry
= &flow_action
->entries
[j
];
3347 if (is_tcf_gact_ok(act
)) {
3348 entry
->id
= FLOW_ACTION_ACCEPT
;
3349 } else if (is_tcf_gact_shot(act
)) {
3350 entry
->id
= FLOW_ACTION_DROP
;
3351 } else if (is_tcf_gact_trap(act
)) {
3352 entry
->id
= FLOW_ACTION_TRAP
;
3353 } else if (is_tcf_gact_goto_chain(act
)) {
3354 entry
->id
= FLOW_ACTION_GOTO
;
3355 entry
->chain_index
= tcf_gact_goto_chain_index(act
);
3356 } else if (is_tcf_mirred_egress_redirect(act
)) {
3357 entry
->id
= FLOW_ACTION_REDIRECT
;
3358 tcf_mirred_get_dev(entry
, act
);
3359 } else if (is_tcf_mirred_egress_mirror(act
)) {
3360 entry
->id
= FLOW_ACTION_MIRRED
;
3361 tcf_mirred_get_dev(entry
, act
);
3362 } else if (is_tcf_mirred_ingress_redirect(act
)) {
3363 entry
->id
= FLOW_ACTION_REDIRECT_INGRESS
;
3364 tcf_mirred_get_dev(entry
, act
);
3365 } else if (is_tcf_mirred_ingress_mirror(act
)) {
3366 entry
->id
= FLOW_ACTION_MIRRED_INGRESS
;
3367 tcf_mirred_get_dev(entry
, act
);
3368 } else if (is_tcf_vlan(act
)) {
3369 switch (tcf_vlan_action(act
)) {
3370 case TCA_VLAN_ACT_PUSH
:
3371 entry
->id
= FLOW_ACTION_VLAN_PUSH
;
3372 entry
->vlan
.vid
= tcf_vlan_push_vid(act
);
3373 entry
->vlan
.proto
= tcf_vlan_push_proto(act
);
3374 entry
->vlan
.prio
= tcf_vlan_push_prio(act
);
3376 case TCA_VLAN_ACT_POP
:
3377 entry
->id
= FLOW_ACTION_VLAN_POP
;
3379 case TCA_VLAN_ACT_MODIFY
:
3380 entry
->id
= FLOW_ACTION_VLAN_MANGLE
;
3381 entry
->vlan
.vid
= tcf_vlan_push_vid(act
);
3382 entry
->vlan
.proto
= tcf_vlan_push_proto(act
);
3383 entry
->vlan
.prio
= tcf_vlan_push_prio(act
);
3389 } else if (is_tcf_tunnel_set(act
)) {
3390 entry
->id
= FLOW_ACTION_TUNNEL_ENCAP
;
3391 err
= tcf_tunnel_encap_get_tunnel(entry
, act
);
3394 } else if (is_tcf_tunnel_release(act
)) {
3395 entry
->id
= FLOW_ACTION_TUNNEL_DECAP
;
3396 } else if (is_tcf_pedit(act
)) {
3397 for (k
= 0; k
< tcf_pedit_nkeys(act
); k
++) {
3398 switch (tcf_pedit_cmd(act
, k
)) {
3399 case TCA_PEDIT_KEY_EX_CMD_SET
:
3400 entry
->id
= FLOW_ACTION_MANGLE
;
3402 case TCA_PEDIT_KEY_EX_CMD_ADD
:
3403 entry
->id
= FLOW_ACTION_ADD
;
3409 entry
->mangle
.htype
= tcf_pedit_htype(act
, k
);
3410 entry
->mangle
.mask
= tcf_pedit_mask(act
, k
);
3411 entry
->mangle
.val
= tcf_pedit_val(act
, k
);
3412 entry
->mangle
.offset
= tcf_pedit_offset(act
, k
);
3413 entry
= &flow_action
->entries
[++j
];
3415 } else if (is_tcf_csum(act
)) {
3416 entry
->id
= FLOW_ACTION_CSUM
;
3417 entry
->csum_flags
= tcf_csum_update_flags(act
);
3418 } else if (is_tcf_skbedit_mark(act
)) {
3419 entry
->id
= FLOW_ACTION_MARK
;
3420 entry
->mark
= tcf_skbedit_mark(act
);
3421 } else if (is_tcf_sample(act
)) {
3422 entry
->id
= FLOW_ACTION_SAMPLE
;
3423 entry
->sample
.trunc_size
= tcf_sample_trunc_size(act
);
3424 entry
->sample
.truncate
= tcf_sample_truncate(act
);
3425 entry
->sample
.rate
= tcf_sample_rate(act
);
3426 tcf_sample_get_group(entry
, act
);
3427 } else if (is_tcf_police(act
)) {
3428 entry
->id
= FLOW_ACTION_POLICE
;
3429 entry
->police
.burst
= tcf_police_tcfp_burst(act
);
3430 entry
->police
.rate_bytes_ps
=
3431 tcf_police_rate_bytes_ps(act
);
3432 } else if (is_tcf_ct(act
)) {
3433 entry
->id
= FLOW_ACTION_CT
;
3434 entry
->ct
.action
= tcf_ct_action(act
);
3435 entry
->ct
.zone
= tcf_ct_zone(act
);
3436 } else if (is_tcf_mpls(act
)) {
3437 switch (tcf_mpls_action(act
)) {
3438 case TCA_MPLS_ACT_PUSH
:
3439 entry
->id
= FLOW_ACTION_MPLS_PUSH
;
3440 entry
->mpls_push
.proto
= tcf_mpls_proto(act
);
3441 entry
->mpls_push
.label
= tcf_mpls_label(act
);
3442 entry
->mpls_push
.tc
= tcf_mpls_tc(act
);
3443 entry
->mpls_push
.bos
= tcf_mpls_bos(act
);
3444 entry
->mpls_push
.ttl
= tcf_mpls_ttl(act
);
3446 case TCA_MPLS_ACT_POP
:
3447 entry
->id
= FLOW_ACTION_MPLS_POP
;
3448 entry
->mpls_pop
.proto
= tcf_mpls_proto(act
);
3450 case TCA_MPLS_ACT_MODIFY
:
3451 entry
->id
= FLOW_ACTION_MPLS_MANGLE
;
3452 entry
->mpls_mangle
.label
= tcf_mpls_label(act
);
3453 entry
->mpls_mangle
.tc
= tcf_mpls_tc(act
);
3454 entry
->mpls_mangle
.bos
= tcf_mpls_bos(act
);
3455 entry
->mpls_mangle
.ttl
= tcf_mpls_ttl(act
);
3460 } else if (is_tcf_skbedit_ptype(act
)) {
3461 entry
->id
= FLOW_ACTION_PTYPE
;
3462 entry
->ptype
= tcf_skbedit_ptype(act
);
3468 if (!is_tcf_pedit(act
))
3477 tc_cleanup_flow_action(flow_action
);
3481 EXPORT_SYMBOL(tc_setup_flow_action
);
3483 unsigned int tcf_exts_num_actions(struct tcf_exts
*exts
)
3485 unsigned int num_acts
= 0;
3486 struct tc_action
*act
;
3489 tcf_exts_for_each_action(i
, act
, exts
) {
3490 if (is_tcf_pedit(act
))
3491 num_acts
+= tcf_pedit_nkeys(act
);
3497 EXPORT_SYMBOL(tcf_exts_num_actions
);
3499 static __net_init
int tcf_net_init(struct net
*net
)
3501 struct tcf_net
*tn
= net_generic(net
, tcf_net_id
);
3503 spin_lock_init(&tn
->idr_lock
);
3508 static void __net_exit
tcf_net_exit(struct net
*net
)
3510 struct tcf_net
*tn
= net_generic(net
, tcf_net_id
);
3512 idr_destroy(&tn
->idr
);
3515 static struct pernet_operations tcf_net_ops
= {
3516 .init
= tcf_net_init
,
3517 .exit
= tcf_net_exit
,
3519 .size
= sizeof(struct tcf_net
),
3522 static struct flow_indr_block_ing_entry block_ing_entry
= {
3523 .cb
= tc_indr_block_get_and_ing_cmd
,
3524 .list
= LIST_HEAD_INIT(block_ing_entry
.list
),
3527 static int __init
tc_filter_init(void)
3531 tc_filter_wq
= alloc_ordered_workqueue("tc_filter_workqueue", 0);
3535 err
= register_pernet_subsys(&tcf_net_ops
);
3537 goto err_register_pernet_subsys
;
3539 flow_indr_add_block_ing_cb(&block_ing_entry
);
3541 rtnl_register(PF_UNSPEC
, RTM_NEWTFILTER
, tc_new_tfilter
, NULL
,
3542 RTNL_FLAG_DOIT_UNLOCKED
);
3543 rtnl_register(PF_UNSPEC
, RTM_DELTFILTER
, tc_del_tfilter
, NULL
,
3544 RTNL_FLAG_DOIT_UNLOCKED
);
3545 rtnl_register(PF_UNSPEC
, RTM_GETTFILTER
, tc_get_tfilter
,
3546 tc_dump_tfilter
, RTNL_FLAG_DOIT_UNLOCKED
);
3547 rtnl_register(PF_UNSPEC
, RTM_NEWCHAIN
, tc_ctl_chain
, NULL
, 0);
3548 rtnl_register(PF_UNSPEC
, RTM_DELCHAIN
, tc_ctl_chain
, NULL
, 0);
3549 rtnl_register(PF_UNSPEC
, RTM_GETCHAIN
, tc_ctl_chain
,
3554 err_register_pernet_subsys
:
3555 destroy_workqueue(tc_filter_wq
);
3559 subsys_initcall(tc_filter_init
);