2 * net/sched/ife.c Inter-FE action based on ForCES WG InterFE LFB
5 * draft-ietf-forces-interfelfb-03
8 * "Distributing Linux Traffic Control Classifier-Action
10 * Authors: Jamal Hadi Salim and Damascene M. Joachimpillai
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version
15 * 2 of the License, or (at your option) any later version.
17 * copyright Jamal Hadi Salim (2015)
21 #include <linux/types.h>
22 #include <linux/kernel.h>
23 #include <linux/string.h>
24 #include <linux/errno.h>
25 #include <linux/skbuff.h>
26 #include <linux/rtnetlink.h>
27 #include <linux/module.h>
28 #include <linux/init.h>
29 #include <net/net_namespace.h>
30 #include <net/netlink.h>
31 #include <net/pkt_sched.h>
32 #include <uapi/linux/tc_act/tc_ife.h>
33 #include <net/tc_act/tc_ife.h>
34 #include <linux/etherdevice.h>
37 static unsigned int ife_net_id
;
38 static int max_metacnt
= IFE_META_MAX
+ 1;
39 static struct tc_action_ops act_ife_ops
;
41 static const struct nla_policy ife_policy
[TCA_IFE_MAX
+ 1] = {
42 [TCA_IFE_PARMS
] = { .len
= sizeof(struct tc_ife
)},
43 [TCA_IFE_DMAC
] = { .len
= ETH_ALEN
},
44 [TCA_IFE_SMAC
] = { .len
= ETH_ALEN
},
45 [TCA_IFE_TYPE
] = { .type
= NLA_U16
},
48 int ife_encode_meta_u16(u16 metaval
, void *skbdata
, struct tcf_meta_info
*mi
)
53 edata
= *(u16
*)mi
->metaval
;
57 if (!edata
) /* will not encode */
61 return ife_tlv_meta_encode(skbdata
, mi
->metaid
, 2, &edata
);
63 EXPORT_SYMBOL_GPL(ife_encode_meta_u16
);
65 int ife_get_meta_u32(struct sk_buff
*skb
, struct tcf_meta_info
*mi
)
68 return nla_put_u32(skb
, mi
->metaid
, *(u32
*)mi
->metaval
);
70 return nla_put(skb
, mi
->metaid
, 0, NULL
);
72 EXPORT_SYMBOL_GPL(ife_get_meta_u32
);
74 int ife_check_meta_u32(u32 metaval
, struct tcf_meta_info
*mi
)
76 if (metaval
|| mi
->metaval
)
77 return 8; /* T+L+V == 2+2+4 */
81 EXPORT_SYMBOL_GPL(ife_check_meta_u32
);
83 int ife_check_meta_u16(u16 metaval
, struct tcf_meta_info
*mi
)
85 if (metaval
|| mi
->metaval
)
86 return 8; /* T+L+(V) == 2+2+(2+2bytepad) */
90 EXPORT_SYMBOL_GPL(ife_check_meta_u16
);
92 int ife_encode_meta_u32(u32 metaval
, void *skbdata
, struct tcf_meta_info
*mi
)
97 edata
= *(u32
*)mi
->metaval
;
101 if (!edata
) /* will not encode */
104 edata
= htonl(edata
);
105 return ife_tlv_meta_encode(skbdata
, mi
->metaid
, 4, &edata
);
107 EXPORT_SYMBOL_GPL(ife_encode_meta_u32
);
109 int ife_get_meta_u16(struct sk_buff
*skb
, struct tcf_meta_info
*mi
)
112 return nla_put_u16(skb
, mi
->metaid
, *(u16
*)mi
->metaval
);
114 return nla_put(skb
, mi
->metaid
, 0, NULL
);
116 EXPORT_SYMBOL_GPL(ife_get_meta_u16
);
118 int ife_alloc_meta_u32(struct tcf_meta_info
*mi
, void *metaval
, gfp_t gfp
)
120 mi
->metaval
= kmemdup(metaval
, sizeof(u32
), gfp
);
126 EXPORT_SYMBOL_GPL(ife_alloc_meta_u32
);
128 int ife_alloc_meta_u16(struct tcf_meta_info
*mi
, void *metaval
, gfp_t gfp
)
130 mi
->metaval
= kmemdup(metaval
, sizeof(u16
), gfp
);
136 EXPORT_SYMBOL_GPL(ife_alloc_meta_u16
);
138 void ife_release_meta_gen(struct tcf_meta_info
*mi
)
142 EXPORT_SYMBOL_GPL(ife_release_meta_gen
);
144 int ife_validate_meta_u32(void *val
, int len
)
146 if (len
== sizeof(u32
))
151 EXPORT_SYMBOL_GPL(ife_validate_meta_u32
);
153 int ife_validate_meta_u16(void *val
, int len
)
155 /* length will not include padding */
156 if (len
== sizeof(u16
))
161 EXPORT_SYMBOL_GPL(ife_validate_meta_u16
);
163 static LIST_HEAD(ifeoplist
);
164 static DEFINE_RWLOCK(ife_mod_lock
);
166 static struct tcf_meta_ops
*find_ife_oplist(u16 metaid
)
168 struct tcf_meta_ops
*o
;
170 read_lock(&ife_mod_lock
);
171 list_for_each_entry(o
, &ifeoplist
, list
) {
172 if (o
->metaid
== metaid
) {
173 if (!try_module_get(o
->owner
))
175 read_unlock(&ife_mod_lock
);
179 read_unlock(&ife_mod_lock
);
184 int register_ife_op(struct tcf_meta_ops
*mops
)
186 struct tcf_meta_ops
*m
;
188 if (!mops
->metaid
|| !mops
->metatype
|| !mops
->name
||
189 !mops
->check_presence
|| !mops
->encode
|| !mops
->decode
||
190 !mops
->get
|| !mops
->alloc
)
193 write_lock(&ife_mod_lock
);
195 list_for_each_entry(m
, &ifeoplist
, list
) {
196 if (m
->metaid
== mops
->metaid
||
197 (strcmp(mops
->name
, m
->name
) == 0)) {
198 write_unlock(&ife_mod_lock
);
204 mops
->release
= ife_release_meta_gen
;
206 list_add_tail(&mops
->list
, &ifeoplist
);
207 write_unlock(&ife_mod_lock
);
210 EXPORT_SYMBOL_GPL(unregister_ife_op
);
212 int unregister_ife_op(struct tcf_meta_ops
*mops
)
214 struct tcf_meta_ops
*m
;
217 write_lock(&ife_mod_lock
);
218 list_for_each_entry(m
, &ifeoplist
, list
) {
219 if (m
->metaid
== mops
->metaid
) {
220 list_del(&mops
->list
);
225 write_unlock(&ife_mod_lock
);
229 EXPORT_SYMBOL_GPL(register_ife_op
);
231 static int ife_validate_metatype(struct tcf_meta_ops
*ops
, void *val
, int len
)
234 /* XXX: unfortunately cant use nla_policy at this point
235 * because a length of 0 is valid in the case of
236 * "allow". "use" semantics do enforce for proper
237 * length and i couldve use nla_policy but it makes it hard
238 * to use it just for that..
241 return ops
->validate(val
, len
);
243 if (ops
->metatype
== NLA_U32
)
244 ret
= ife_validate_meta_u32(val
, len
);
245 else if (ops
->metatype
== NLA_U16
)
246 ret
= ife_validate_meta_u16(val
, len
);
251 /* called when adding new meta information
252 * under ife->tcf_lock for existing action
254 static int load_metaops_and_vet(struct tcf_ife_info
*ife
, u32 metaid
,
255 void *val
, int len
, bool exists
)
257 struct tcf_meta_ops
*ops
= find_ife_oplist(metaid
);
262 #ifdef CONFIG_MODULES
264 spin_unlock_bh(&ife
->tcf_lock
);
266 request_module("ifemeta%u", metaid
);
269 spin_lock_bh(&ife
->tcf_lock
);
270 ops
= find_ife_oplist(metaid
);
277 ret
= ife_validate_metatype(ops
, val
, len
);
279 module_put(ops
->owner
);
285 /* called when adding new meta information
286 * under ife->tcf_lock for existing action
288 static int add_metainfo(struct tcf_ife_info
*ife
, u32 metaid
, void *metaval
,
289 int len
, bool atomic
)
291 struct tcf_meta_info
*mi
= NULL
;
292 struct tcf_meta_ops
*ops
= find_ife_oplist(metaid
);
298 mi
= kzalloc(sizeof(*mi
), atomic
? GFP_ATOMIC
: GFP_KERNEL
);
300 /*put back what find_ife_oplist took */
301 module_put(ops
->owner
);
308 ret
= ops
->alloc(mi
, metaval
, atomic
? GFP_ATOMIC
: GFP_KERNEL
);
311 module_put(ops
->owner
);
316 list_add_tail(&mi
->metalist
, &ife
->metalist
);
321 static int use_all_metadata(struct tcf_ife_info
*ife
)
323 struct tcf_meta_ops
*o
;
327 read_lock(&ife_mod_lock
);
328 list_for_each_entry(o
, &ifeoplist
, list
) {
329 rc
= add_metainfo(ife
, o
->metaid
, NULL
, 0, true);
333 read_unlock(&ife_mod_lock
);
341 static int dump_metalist(struct sk_buff
*skb
, struct tcf_ife_info
*ife
)
343 struct tcf_meta_info
*e
;
345 unsigned char *b
= skb_tail_pointer(skb
);
346 int total_encoded
= 0;
348 /*can only happen on decode */
349 if (list_empty(&ife
->metalist
))
352 nest
= nla_nest_start(skb
, TCA_IFE_METALST
);
356 list_for_each_entry(e
, &ife
->metalist
, metalist
) {
357 if (!e
->ops
->get(skb
, e
))
364 nla_nest_end(skb
, nest
);
373 /* under ife->tcf_lock */
374 static void _tcf_ife_cleanup(struct tc_action
*a
, int bind
)
376 struct tcf_ife_info
*ife
= to_ife(a
);
377 struct tcf_meta_info
*e
, *n
;
379 list_for_each_entry_safe(e
, n
, &ife
->metalist
, metalist
) {
380 module_put(e
->ops
->owner
);
381 list_del(&e
->metalist
);
392 static void tcf_ife_cleanup(struct tc_action
*a
, int bind
)
394 struct tcf_ife_info
*ife
= to_ife(a
);
396 spin_lock_bh(&ife
->tcf_lock
);
397 _tcf_ife_cleanup(a
, bind
);
398 spin_unlock_bh(&ife
->tcf_lock
);
401 /* under ife->tcf_lock for existing action */
402 static int populate_metalist(struct tcf_ife_info
*ife
, struct nlattr
**tb
,
410 for (i
= 1; i
< max_metacnt
; i
++) {
412 val
= nla_data(tb
[i
]);
413 len
= nla_len(tb
[i
]);
415 rc
= load_metaops_and_vet(ife
, i
, val
, len
, exists
);
419 rc
= add_metainfo(ife
, i
, val
, len
, exists
);
428 static int tcf_ife_init(struct net
*net
, struct nlattr
*nla
,
429 struct nlattr
*est
, struct tc_action
**a
,
432 struct tc_action_net
*tn
= net_generic(net
, ife_net_id
);
433 struct nlattr
*tb
[TCA_IFE_MAX
+ 1];
434 struct nlattr
*tb2
[IFE_META_MAX
+ 1];
435 struct tcf_ife_info
*ife
;
436 u16 ife_type
= ETH_P_IFE
;
444 err
= nla_parse_nested(tb
, TCA_IFE_MAX
, nla
, ife_policy
, NULL
);
448 if (!tb
[TCA_IFE_PARMS
])
451 parm
= nla_data(tb
[TCA_IFE_PARMS
]);
453 exists
= tcf_idr_check(tn
, parm
->index
, a
, bind
);
458 ret
= tcf_idr_create(tn
, parm
->index
, est
, a
, &act_ife_ops
,
464 tcf_idr_release(*a
, bind
);
470 ife
->flags
= parm
->flags
;
472 if (parm
->flags
& IFE_ENCODE
) {
473 if (tb
[TCA_IFE_TYPE
])
474 ife_type
= nla_get_u16(tb
[TCA_IFE_TYPE
]);
475 if (tb
[TCA_IFE_DMAC
])
476 daddr
= nla_data(tb
[TCA_IFE_DMAC
]);
477 if (tb
[TCA_IFE_SMAC
])
478 saddr
= nla_data(tb
[TCA_IFE_SMAC
]);
482 spin_lock_bh(&ife
->tcf_lock
);
483 ife
->tcf_action
= parm
->action
;
485 if (parm
->flags
& IFE_ENCODE
) {
487 ether_addr_copy(ife
->eth_dst
, daddr
);
489 eth_zero_addr(ife
->eth_dst
);
492 ether_addr_copy(ife
->eth_src
, saddr
);
494 eth_zero_addr(ife
->eth_src
);
496 ife
->eth_type
= ife_type
;
499 if (ret
== ACT_P_CREATED
)
500 INIT_LIST_HEAD(&ife
->metalist
);
502 if (tb
[TCA_IFE_METALST
]) {
503 err
= nla_parse_nested(tb2
, IFE_META_MAX
, tb
[TCA_IFE_METALST
],
508 tcf_idr_release(*a
, bind
);
509 if (ret
== ACT_P_CREATED
)
510 _tcf_ife_cleanup(*a
, bind
);
513 spin_unlock_bh(&ife
->tcf_lock
);
517 err
= populate_metalist(ife
, tb2
, exists
);
519 goto metadata_parse_err
;
522 /* if no passed metadata allow list or passed allow-all
523 * then here we process by adding as many supported metadatum
524 * as we can. You better have at least one else we are
527 err
= use_all_metadata(ife
);
529 if (ret
== ACT_P_CREATED
)
530 _tcf_ife_cleanup(*a
, bind
);
533 spin_unlock_bh(&ife
->tcf_lock
);
539 spin_unlock_bh(&ife
->tcf_lock
);
541 if (ret
== ACT_P_CREATED
)
542 tcf_idr_insert(tn
, *a
);
547 static int tcf_ife_dump(struct sk_buff
*skb
, struct tc_action
*a
, int bind
,
550 unsigned char *b
= skb_tail_pointer(skb
);
551 struct tcf_ife_info
*ife
= to_ife(a
);
552 struct tc_ife opt
= {
553 .index
= ife
->tcf_index
,
554 .refcnt
= ife
->tcf_refcnt
- ref
,
555 .bindcnt
= ife
->tcf_bindcnt
- bind
,
556 .action
= ife
->tcf_action
,
561 if (nla_put(skb
, TCA_IFE_PARMS
, sizeof(opt
), &opt
))
562 goto nla_put_failure
;
564 tcf_tm_dump(&t
, &ife
->tcf_tm
);
565 if (nla_put_64bit(skb
, TCA_IFE_TM
, sizeof(t
), &t
, TCA_IFE_PAD
))
566 goto nla_put_failure
;
568 if (!is_zero_ether_addr(ife
->eth_dst
)) {
569 if (nla_put(skb
, TCA_IFE_DMAC
, ETH_ALEN
, ife
->eth_dst
))
570 goto nla_put_failure
;
573 if (!is_zero_ether_addr(ife
->eth_src
)) {
574 if (nla_put(skb
, TCA_IFE_SMAC
, ETH_ALEN
, ife
->eth_src
))
575 goto nla_put_failure
;
578 if (nla_put(skb
, TCA_IFE_TYPE
, 2, &ife
->eth_type
))
579 goto nla_put_failure
;
581 if (dump_metalist(skb
, ife
)) {
582 /*ignore failure to dump metalist */
583 pr_info("Failed to dump metalist\n");
593 static int find_decode_metaid(struct sk_buff
*skb
, struct tcf_ife_info
*ife
,
594 u16 metaid
, u16 mlen
, void *mdata
)
596 struct tcf_meta_info
*e
;
598 /* XXX: use hash to speed up */
599 list_for_each_entry(e
, &ife
->metalist
, metalist
) {
600 if (metaid
== e
->metaid
) {
602 /* We check for decode presence already */
603 return e
->ops
->decode(skb
, mdata
, mlen
);
611 static int tcf_ife_decode(struct sk_buff
*skb
, const struct tc_action
*a
,
612 struct tcf_result
*res
)
614 struct tcf_ife_info
*ife
= to_ife(a
);
615 int action
= ife
->tcf_action
;
620 spin_lock(&ife
->tcf_lock
);
621 bstats_update(&ife
->tcf_bstats
, skb
);
622 tcf_lastuse_update(&ife
->tcf_tm
);
623 spin_unlock(&ife
->tcf_lock
);
625 if (skb_at_tc_ingress(skb
))
626 skb_push(skb
, skb
->dev
->hard_header_len
);
628 tlv_data
= ife_decode(skb
, &metalen
);
629 if (unlikely(!tlv_data
)) {
630 spin_lock(&ife
->tcf_lock
);
631 ife
->tcf_qstats
.drops
++;
632 spin_unlock(&ife
->tcf_lock
);
636 ifehdr_end
= tlv_data
+ metalen
;
637 for (; tlv_data
< ifehdr_end
; tlv_data
= ife_tlv_meta_next(tlv_data
)) {
642 curr_data
= ife_tlv_meta_decode(tlv_data
, &mtype
, &dlen
, NULL
);
644 if (find_decode_metaid(skb
, ife
, mtype
, dlen
, curr_data
)) {
645 /* abuse overlimits to count when we receive metadata
646 * but dont have an ops for it
648 pr_info_ratelimited("Unknown metaid %d dlen %d\n",
650 ife
->tcf_qstats
.overlimits
++;
654 if (WARN_ON(tlv_data
!= ifehdr_end
)) {
655 spin_lock(&ife
->tcf_lock
);
656 ife
->tcf_qstats
.drops
++;
657 spin_unlock(&ife
->tcf_lock
);
661 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
662 skb_reset_network_header(skb
);
667 /*XXX: check if we can do this at install time instead of current
670 static int ife_get_sz(struct sk_buff
*skb
, struct tcf_ife_info
*ife
)
672 struct tcf_meta_info
*e
, *n
;
673 int tot_run_sz
= 0, run_sz
= 0;
675 list_for_each_entry_safe(e
, n
, &ife
->metalist
, metalist
) {
676 if (e
->ops
->check_presence
) {
677 run_sz
= e
->ops
->check_presence(skb
, e
);
678 tot_run_sz
+= run_sz
;
685 static int tcf_ife_encode(struct sk_buff
*skb
, const struct tc_action
*a
,
686 struct tcf_result
*res
)
688 struct tcf_ife_info
*ife
= to_ife(a
);
689 int action
= ife
->tcf_action
;
690 struct ethhdr
*oethh
; /* outer ether header */
691 struct tcf_meta_info
*e
;
693 OUTERHDR:TOTMETALEN:{TLVHDR:Metadatum:TLVHDR..}:ORIGDATA
694 where ORIGDATA = original ethernet header ...
696 u16 metalen
= ife_get_sz(skb
, ife
);
697 int hdrm
= metalen
+ skb
->dev
->hard_header_len
+ IFE_METAHDRLEN
;
698 unsigned int skboff
= 0;
699 int new_len
= skb
->len
+ hdrm
;
700 bool exceed_mtu
= false;
704 if (!skb_at_tc_ingress(skb
)) {
705 if (new_len
> skb
->dev
->mtu
)
709 spin_lock(&ife
->tcf_lock
);
710 bstats_update(&ife
->tcf_bstats
, skb
);
711 tcf_lastuse_update(&ife
->tcf_tm
);
713 if (!metalen
) { /* no metadata to send */
714 /* abuse overlimits to count when we allow packet
717 ife
->tcf_qstats
.overlimits
++;
718 spin_unlock(&ife
->tcf_lock
);
721 /* could be stupid policy setup or mtu config
722 * so lets be conservative.. */
723 if ((action
== TC_ACT_SHOT
) || exceed_mtu
) {
724 ife
->tcf_qstats
.drops
++;
725 spin_unlock(&ife
->tcf_lock
);
729 if (skb_at_tc_ingress(skb
))
730 skb_push(skb
, skb
->dev
->hard_header_len
);
732 ife_meta
= ife_encode(skb
, metalen
);
734 /* XXX: we dont have a clever way of telling encode to
735 * not repeat some of the computations that are done by
736 * ops->presence_check...
738 list_for_each_entry(e
, &ife
->metalist
, metalist
) {
739 if (e
->ops
->encode
) {
740 err
= e
->ops
->encode(skb
, (void *)(ife_meta
+ skboff
),
744 /* too corrupt to keep around if overwritten */
745 ife
->tcf_qstats
.drops
++;
746 spin_unlock(&ife
->tcf_lock
);
751 oethh
= (struct ethhdr
*)skb
->data
;
753 if (!is_zero_ether_addr(ife
->eth_src
))
754 ether_addr_copy(oethh
->h_source
, ife
->eth_src
);
755 if (!is_zero_ether_addr(ife
->eth_dst
))
756 ether_addr_copy(oethh
->h_dest
, ife
->eth_dst
);
757 oethh
->h_proto
= htons(ife
->eth_type
);
759 if (skb_at_tc_ingress(skb
))
760 skb_pull(skb
, skb
->dev
->hard_header_len
);
762 spin_unlock(&ife
->tcf_lock
);
767 static int tcf_ife_act(struct sk_buff
*skb
, const struct tc_action
*a
,
768 struct tcf_result
*res
)
770 struct tcf_ife_info
*ife
= to_ife(a
);
772 if (ife
->flags
& IFE_ENCODE
)
773 return tcf_ife_encode(skb
, a
, res
);
775 if (!(ife
->flags
& IFE_ENCODE
))
776 return tcf_ife_decode(skb
, a
, res
);
778 pr_info_ratelimited("unknown failure(policy neither de/encode\n");
779 spin_lock(&ife
->tcf_lock
);
780 bstats_update(&ife
->tcf_bstats
, skb
);
781 tcf_lastuse_update(&ife
->tcf_tm
);
782 ife
->tcf_qstats
.drops
++;
783 spin_unlock(&ife
->tcf_lock
);
788 static int tcf_ife_walker(struct net
*net
, struct sk_buff
*skb
,
789 struct netlink_callback
*cb
, int type
,
790 const struct tc_action_ops
*ops
)
792 struct tc_action_net
*tn
= net_generic(net
, ife_net_id
);
794 return tcf_generic_walker(tn
, skb
, cb
, type
, ops
);
797 static int tcf_ife_search(struct net
*net
, struct tc_action
**a
, u32 index
)
799 struct tc_action_net
*tn
= net_generic(net
, ife_net_id
);
801 return tcf_idr_search(tn
, a
, index
);
804 static struct tc_action_ops act_ife_ops
= {
807 .owner
= THIS_MODULE
,
809 .dump
= tcf_ife_dump
,
810 .cleanup
= tcf_ife_cleanup
,
811 .init
= tcf_ife_init
,
812 .walk
= tcf_ife_walker
,
813 .lookup
= tcf_ife_search
,
814 .size
= sizeof(struct tcf_ife_info
),
817 static __net_init
int ife_init_net(struct net
*net
)
819 struct tc_action_net
*tn
= net_generic(net
, ife_net_id
);
821 return tc_action_net_init(tn
, &act_ife_ops
);
824 static void __net_exit
ife_exit_net(struct net
*net
)
826 struct tc_action_net
*tn
= net_generic(net
, ife_net_id
);
828 tc_action_net_exit(tn
);
831 static struct pernet_operations ife_net_ops
= {
832 .init
= ife_init_net
,
833 .exit
= ife_exit_net
,
835 .size
= sizeof(struct tc_action_net
),
838 static int __init
ife_init_module(void)
840 return tcf_register_action(&act_ife_ops
, &ife_net_ops
);
843 static void __exit
ife_cleanup_module(void)
845 tcf_unregister_action(&act_ife_ops
, &ife_net_ops
);
848 module_init(ife_init_module
);
849 module_exit(ife_cleanup_module
);
851 MODULE_AUTHOR("Jamal Hadi Salim(2015)");
852 MODULE_DESCRIPTION("Inter-FE LFB action");
853 MODULE_LICENSE("GPL");