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>
36 #define IFE_TAB_MASK 15
38 static int ife_net_id
;
39 static int max_metacnt
= IFE_META_MAX
+ 1;
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 /* Caller takes care of presenting data in network order
50 int ife_tlv_meta_encode(void *skbdata
, u16 attrtype
, u16 dlen
, const void *dval
)
52 u32
*tlv
= (u32
*)(skbdata
);
53 u16 totlen
= nla_total_size(dlen
); /*alignment + hdr */
54 char *dptr
= (char *)tlv
+ NLA_HDRLEN
;
55 u32 htlv
= attrtype
<< 16 | totlen
;
58 memset(dptr
, 0, totlen
- NLA_HDRLEN
);
59 memcpy(dptr
, dval
, dlen
);
63 EXPORT_SYMBOL_GPL(ife_tlv_meta_encode
);
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_encode_meta_u32(u32 metaval
, void *skbdata
, struct tcf_meta_info
*mi
)
88 edata
= *(u32
*)mi
->metaval
;
92 if (!edata
) /* will not encode */
96 return ife_tlv_meta_encode(skbdata
, mi
->metaid
, 4, &edata
);
98 EXPORT_SYMBOL_GPL(ife_encode_meta_u32
);
100 int ife_get_meta_u16(struct sk_buff
*skb
, struct tcf_meta_info
*mi
)
103 return nla_put_u16(skb
, mi
->metaid
, *(u16
*)mi
->metaval
);
105 return nla_put(skb
, mi
->metaid
, 0, NULL
);
107 EXPORT_SYMBOL_GPL(ife_get_meta_u16
);
109 int ife_alloc_meta_u32(struct tcf_meta_info
*mi
, void *metaval
)
111 mi
->metaval
= kmemdup(metaval
, sizeof(u32
), GFP_KERNEL
);
117 EXPORT_SYMBOL_GPL(ife_alloc_meta_u32
);
119 int ife_alloc_meta_u16(struct tcf_meta_info
*mi
, void *metaval
)
121 mi
->metaval
= kmemdup(metaval
, sizeof(u16
), GFP_KERNEL
);
127 EXPORT_SYMBOL_GPL(ife_alloc_meta_u16
);
129 void ife_release_meta_gen(struct tcf_meta_info
*mi
)
133 EXPORT_SYMBOL_GPL(ife_release_meta_gen
);
135 int ife_validate_meta_u32(void *val
, int len
)
142 EXPORT_SYMBOL_GPL(ife_validate_meta_u32
);
144 int ife_validate_meta_u16(void *val
, int len
)
146 /* length will include padding */
147 if (len
== NLA_ALIGN(2))
152 EXPORT_SYMBOL_GPL(ife_validate_meta_u16
);
154 static LIST_HEAD(ifeoplist
);
155 static DEFINE_RWLOCK(ife_mod_lock
);
157 static struct tcf_meta_ops
*find_ife_oplist(u16 metaid
)
159 struct tcf_meta_ops
*o
;
161 read_lock(&ife_mod_lock
);
162 list_for_each_entry(o
, &ifeoplist
, list
) {
163 if (o
->metaid
== metaid
) {
164 if (!try_module_get(o
->owner
))
166 read_unlock(&ife_mod_lock
);
170 read_unlock(&ife_mod_lock
);
175 int register_ife_op(struct tcf_meta_ops
*mops
)
177 struct tcf_meta_ops
*m
;
179 if (!mops
->metaid
|| !mops
->metatype
|| !mops
->name
||
180 !mops
->check_presence
|| !mops
->encode
|| !mops
->decode
||
181 !mops
->get
|| !mops
->alloc
)
184 write_lock(&ife_mod_lock
);
186 list_for_each_entry(m
, &ifeoplist
, list
) {
187 if (m
->metaid
== mops
->metaid
||
188 (strcmp(mops
->name
, m
->name
) == 0)) {
189 write_unlock(&ife_mod_lock
);
195 mops
->release
= ife_release_meta_gen
;
197 list_add_tail(&mops
->list
, &ifeoplist
);
198 write_unlock(&ife_mod_lock
);
201 EXPORT_SYMBOL_GPL(unregister_ife_op
);
203 int unregister_ife_op(struct tcf_meta_ops
*mops
)
205 struct tcf_meta_ops
*m
;
208 write_lock(&ife_mod_lock
);
209 list_for_each_entry(m
, &ifeoplist
, list
) {
210 if (m
->metaid
== mops
->metaid
) {
211 list_del(&mops
->list
);
216 write_unlock(&ife_mod_lock
);
220 EXPORT_SYMBOL_GPL(register_ife_op
);
222 static int ife_validate_metatype(struct tcf_meta_ops
*ops
, void *val
, int len
)
225 /* XXX: unfortunately cant use nla_policy at this point
226 * because a length of 0 is valid in the case of
227 * "allow". "use" semantics do enforce for proper
228 * length and i couldve use nla_policy but it makes it hard
229 * to use it just for that..
232 return ops
->validate(val
, len
);
234 if (ops
->metatype
== NLA_U32
)
235 ret
= ife_validate_meta_u32(val
, len
);
236 else if (ops
->metatype
== NLA_U16
)
237 ret
= ife_validate_meta_u16(val
, len
);
242 /* called when adding new meta information
243 * under ife->tcf_lock
245 static int load_metaops_and_vet(struct tcf_ife_info
*ife
, u32 metaid
,
248 struct tcf_meta_ops
*ops
= find_ife_oplist(metaid
);
253 #ifdef CONFIG_MODULES
254 spin_unlock_bh(&ife
->tcf_lock
);
256 request_module("ifemeta%u", metaid
);
258 spin_lock_bh(&ife
->tcf_lock
);
259 ops
= find_ife_oplist(metaid
);
266 ret
= ife_validate_metatype(ops
, val
, len
);
268 module_put(ops
->owner
);
274 /* called when adding new meta information
275 * under ife->tcf_lock
277 static int add_metainfo(struct tcf_ife_info
*ife
, u32 metaid
, void *metaval
,
280 struct tcf_meta_info
*mi
= NULL
;
281 struct tcf_meta_ops
*ops
= find_ife_oplist(metaid
);
287 mi
= kzalloc(sizeof(*mi
), GFP_KERNEL
);
289 /*put back what find_ife_oplist took */
290 module_put(ops
->owner
);
297 ret
= ops
->alloc(mi
, metaval
);
300 module_put(ops
->owner
);
305 list_add_tail(&mi
->metalist
, &ife
->metalist
);
310 static int use_all_metadata(struct tcf_ife_info
*ife
)
312 struct tcf_meta_ops
*o
;
316 list_for_each_entry(o
, &ifeoplist
, list
) {
317 rc
= add_metainfo(ife
, o
->metaid
, NULL
, 0);
328 static int dump_metalist(struct sk_buff
*skb
, struct tcf_ife_info
*ife
)
330 struct tcf_meta_info
*e
;
332 unsigned char *b
= skb_tail_pointer(skb
);
333 int total_encoded
= 0;
335 /*can only happen on decode */
336 if (list_empty(&ife
->metalist
))
339 nest
= nla_nest_start(skb
, TCA_IFE_METALST
);
343 list_for_each_entry(e
, &ife
->metalist
, metalist
) {
344 if (!e
->ops
->get(skb
, e
))
351 nla_nest_end(skb
, nest
);
360 /* under ife->tcf_lock */
361 static void _tcf_ife_cleanup(struct tc_action
*a
, int bind
)
363 struct tcf_ife_info
*ife
= a
->priv
;
364 struct tcf_meta_info
*e
, *n
;
366 list_for_each_entry_safe(e
, n
, &ife
->metalist
, metalist
) {
367 module_put(e
->ops
->owner
);
368 list_del(&e
->metalist
);
379 static void tcf_ife_cleanup(struct tc_action
*a
, int bind
)
381 struct tcf_ife_info
*ife
= a
->priv
;
383 spin_lock_bh(&ife
->tcf_lock
);
384 _tcf_ife_cleanup(a
, bind
);
385 spin_unlock_bh(&ife
->tcf_lock
);
388 /* under ife->tcf_lock */
389 static int populate_metalist(struct tcf_ife_info
*ife
, struct nlattr
**tb
)
396 for (i
= 1; i
< max_metacnt
; i
++) {
398 val
= nla_data(tb
[i
]);
399 len
= nla_len(tb
[i
]);
401 rc
= load_metaops_and_vet(ife
, i
, val
, len
);
405 rc
= add_metainfo(ife
, i
, val
, len
);
414 static int tcf_ife_init(struct net
*net
, struct nlattr
*nla
,
415 struct nlattr
*est
, struct tc_action
*a
,
418 struct tc_action_net
*tn
= net_generic(net
, ife_net_id
);
419 struct nlattr
*tb
[TCA_IFE_MAX
+ 1];
420 struct nlattr
*tb2
[IFE_META_MAX
+ 1];
421 struct tcf_ife_info
*ife
;
430 err
= nla_parse_nested(tb
, TCA_IFE_MAX
, nla
, ife_policy
);
434 if (!tb
[TCA_IFE_PARMS
])
437 parm
= nla_data(tb
[TCA_IFE_PARMS
]);
439 exists
= tcf_hash_check(tn
, parm
->index
, a
, bind
);
443 if (parm
->flags
& IFE_ENCODE
) {
444 /* Until we get issued the ethertype, we cant have
447 if (!tb
[TCA_IFE_TYPE
]) {
449 tcf_hash_release(a
, bind
);
450 pr_info("You MUST pass etherype for encoding\n");
456 ret
= tcf_hash_create(tn
, parm
->index
, est
, a
, sizeof(*ife
),
462 tcf_hash_release(a
, bind
);
468 ife
->flags
= parm
->flags
;
470 if (parm
->flags
& IFE_ENCODE
) {
471 ife_type
= nla_get_u16(tb
[TCA_IFE_TYPE
]);
472 if (tb
[TCA_IFE_DMAC
])
473 daddr
= nla_data(tb
[TCA_IFE_DMAC
]);
474 if (tb
[TCA_IFE_SMAC
])
475 saddr
= nla_data(tb
[TCA_IFE_SMAC
]);
478 spin_lock_bh(&ife
->tcf_lock
);
479 ife
->tcf_action
= parm
->action
;
481 if (parm
->flags
& IFE_ENCODE
) {
483 ether_addr_copy(ife
->eth_dst
, daddr
);
485 eth_zero_addr(ife
->eth_dst
);
488 ether_addr_copy(ife
->eth_src
, saddr
);
490 eth_zero_addr(ife
->eth_src
);
492 ife
->eth_type
= ife_type
;
495 if (ret
== ACT_P_CREATED
)
496 INIT_LIST_HEAD(&ife
->metalist
);
498 if (tb
[TCA_IFE_METALST
]) {
499 err
= nla_parse_nested(tb2
, IFE_META_MAX
, tb
[TCA_IFE_METALST
],
504 tcf_hash_release(a
, bind
);
505 if (ret
== ACT_P_CREATED
)
506 _tcf_ife_cleanup(a
, bind
);
508 spin_unlock_bh(&ife
->tcf_lock
);
512 err
= populate_metalist(ife
, tb2
);
514 goto metadata_parse_err
;
517 /* if no passed metadata allow list or passed allow-all
518 * then here we process by adding as many supported metadatum
519 * as we can. You better have at least one else we are
522 err
= use_all_metadata(ife
);
524 if (ret
== ACT_P_CREATED
)
525 _tcf_ife_cleanup(a
, bind
);
527 spin_unlock_bh(&ife
->tcf_lock
);
532 spin_unlock_bh(&ife
->tcf_lock
);
534 if (ret
== ACT_P_CREATED
)
535 tcf_hash_insert(tn
, a
);
540 static int tcf_ife_dump(struct sk_buff
*skb
, struct tc_action
*a
, int bind
,
543 unsigned char *b
= skb_tail_pointer(skb
);
544 struct tcf_ife_info
*ife
= a
->priv
;
545 struct tc_ife opt
= {
546 .index
= ife
->tcf_index
,
547 .refcnt
= ife
->tcf_refcnt
- ref
,
548 .bindcnt
= ife
->tcf_bindcnt
- bind
,
549 .action
= ife
->tcf_action
,
554 if (nla_put(skb
, TCA_IFE_PARMS
, sizeof(opt
), &opt
))
555 goto nla_put_failure
;
557 tcf_tm_dump(&t
, &ife
->tcf_tm
);
558 if (nla_put_64bit(skb
, TCA_IFE_TM
, sizeof(t
), &t
, TCA_IFE_PAD
))
559 goto nla_put_failure
;
561 if (!is_zero_ether_addr(ife
->eth_dst
)) {
562 if (nla_put(skb
, TCA_IFE_DMAC
, ETH_ALEN
, ife
->eth_dst
))
563 goto nla_put_failure
;
566 if (!is_zero_ether_addr(ife
->eth_src
)) {
567 if (nla_put(skb
, TCA_IFE_SMAC
, ETH_ALEN
, ife
->eth_src
))
568 goto nla_put_failure
;
571 if (nla_put(skb
, TCA_IFE_TYPE
, 2, &ife
->eth_type
))
572 goto nla_put_failure
;
574 if (dump_metalist(skb
, ife
)) {
575 /*ignore failure to dump metalist */
576 pr_info("Failed to dump metalist\n");
586 int find_decode_metaid(struct sk_buff
*skb
, struct tcf_ife_info
*ife
,
587 u16 metaid
, u16 mlen
, void *mdata
)
589 struct tcf_meta_info
*e
;
591 /* XXX: use hash to speed up */
592 list_for_each_entry(e
, &ife
->metalist
, metalist
) {
593 if (metaid
== e
->metaid
) {
595 /* We check for decode presence already */
596 return e
->ops
->decode(skb
, mdata
, mlen
);
614 static int tcf_ife_decode(struct sk_buff
*skb
, const struct tc_action
*a
,
615 struct tcf_result
*res
)
617 struct tcf_ife_info
*ife
= a
->priv
;
618 int action
= ife
->tcf_action
;
619 struct ifeheadr
*ifehdr
= (struct ifeheadr
*)skb
->data
;
620 u16 ifehdrln
= ifehdr
->metalen
;
621 struct meta_tlvhdr
*tlv
= (struct meta_tlvhdr
*)(ifehdr
->tlv_data
);
623 spin_lock(&ife
->tcf_lock
);
624 bstats_update(&ife
->tcf_bstats
, skb
);
625 tcf_lastuse_update(&ife
->tcf_tm
);
626 spin_unlock(&ife
->tcf_lock
);
628 ifehdrln
= ntohs(ifehdrln
);
629 if (unlikely(!pskb_may_pull(skb
, ifehdrln
))) {
630 spin_lock(&ife
->tcf_lock
);
631 ife
->tcf_qstats
.drops
++;
632 spin_unlock(&ife
->tcf_lock
);
636 skb_set_mac_header(skb
, ifehdrln
);
637 __skb_pull(skb
, ifehdrln
);
638 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
639 ifehdrln
-= IFE_METAHDRLEN
;
641 while (ifehdrln
> 0) {
642 u8
*tlvdata
= (u8
*)tlv
;
643 u16 mtype
= tlv
->type
;
646 mtype
= ntohs(mtype
);
649 if (find_decode_metaid(skb
, ife
, mtype
, (mlen
- 4),
650 (void *)(tlvdata
+ 4))) {
651 /* abuse overlimits to count when we receive metadata
652 * but dont have an ops for it
654 pr_info_ratelimited("Unknown metaid %d alnlen %d\n",
656 ife
->tcf_qstats
.overlimits
++;
661 tlv
= (struct meta_tlvhdr
*)tlvdata
;
664 skb_reset_network_header(skb
);
668 /*XXX: check if we can do this at install time instead of current
671 static int ife_get_sz(struct sk_buff
*skb
, struct tcf_ife_info
*ife
)
673 struct tcf_meta_info
*e
, *n
;
674 int tot_run_sz
= 0, run_sz
= 0;
676 list_for_each_entry_safe(e
, n
, &ife
->metalist
, metalist
) {
677 if (e
->ops
->check_presence
) {
678 run_sz
= e
->ops
->check_presence(skb
, e
);
679 tot_run_sz
+= run_sz
;
686 static int tcf_ife_encode(struct sk_buff
*skb
, const struct tc_action
*a
,
687 struct tcf_result
*res
)
689 struct tcf_ife_info
*ife
= a
->priv
;
690 int action
= ife
->tcf_action
;
691 struct ethhdr
*oethh
; /* outer ether header */
692 struct ethhdr
*iethh
; /* inner eth header */
693 struct tcf_meta_info
*e
;
695 OUTERHDR:TOTMETALEN:{TLVHDR:Metadatum:TLVHDR..}:ORIGDATA
696 where ORIGDATA = original ethernet header ...
698 u16 metalen
= ife_get_sz(skb
, ife
);
699 int hdrm
= metalen
+ skb
->dev
->hard_header_len
+ IFE_METAHDRLEN
;
700 unsigned int skboff
= skb
->dev
->hard_header_len
;
701 u32 at
= G_TC_AT(skb
->tc_verd
);
702 int new_len
= skb
->len
+ hdrm
;
703 bool exceed_mtu
= false;
706 if (at
& AT_EGRESS
) {
707 if (new_len
> skb
->dev
->mtu
)
711 spin_lock(&ife
->tcf_lock
);
712 bstats_update(&ife
->tcf_bstats
, skb
);
713 tcf_lastuse_update(&ife
->tcf_tm
);
715 if (!metalen
) { /* no metadata to send */
716 /* abuse overlimits to count when we allow packet
719 ife
->tcf_qstats
.overlimits
++;
720 spin_unlock(&ife
->tcf_lock
);
723 /* could be stupid policy setup or mtu config
724 * so lets be conservative.. */
725 if ((action
== TC_ACT_SHOT
) || exceed_mtu
) {
726 ife
->tcf_qstats
.drops
++;
727 spin_unlock(&ife
->tcf_lock
);
731 iethh
= eth_hdr(skb
);
733 err
= skb_cow_head(skb
, hdrm
);
735 ife
->tcf_qstats
.drops
++;
736 spin_unlock(&ife
->tcf_lock
);
740 if (!(at
& AT_EGRESS
))
741 skb_push(skb
, skb
->dev
->hard_header_len
);
743 __skb_push(skb
, hdrm
);
744 memcpy(skb
->data
, iethh
, skb
->mac_len
);
745 skb_reset_mac_header(skb
);
746 oethh
= eth_hdr(skb
);
748 /*total metadata length */
749 metalen
+= IFE_METAHDRLEN
;
750 metalen
= htons(metalen
);
751 memcpy((skb
->data
+ skboff
), &metalen
, IFE_METAHDRLEN
);
752 skboff
+= IFE_METAHDRLEN
;
754 /* XXX: we dont have a clever way of telling encode to
755 * not repeat some of the computations that are done by
756 * ops->presence_check...
758 list_for_each_entry(e
, &ife
->metalist
, metalist
) {
759 if (e
->ops
->encode
) {
760 err
= e
->ops
->encode(skb
, (void *)(skb
->data
+ skboff
),
764 /* too corrupt to keep around if overwritten */
765 ife
->tcf_qstats
.drops
++;
766 spin_unlock(&ife
->tcf_lock
);
772 if (!is_zero_ether_addr(ife
->eth_src
))
773 ether_addr_copy(oethh
->h_source
, ife
->eth_src
);
775 ether_addr_copy(oethh
->h_source
, iethh
->h_source
);
776 if (!is_zero_ether_addr(ife
->eth_dst
))
777 ether_addr_copy(oethh
->h_dest
, ife
->eth_dst
);
779 ether_addr_copy(oethh
->h_dest
, iethh
->h_dest
);
780 oethh
->h_proto
= htons(ife
->eth_type
);
782 if (!(at
& AT_EGRESS
))
783 skb_pull(skb
, skb
->dev
->hard_header_len
);
785 spin_unlock(&ife
->tcf_lock
);
790 static int tcf_ife_act(struct sk_buff
*skb
, const struct tc_action
*a
,
791 struct tcf_result
*res
)
793 struct tcf_ife_info
*ife
= a
->priv
;
795 if (ife
->flags
& IFE_ENCODE
)
796 return tcf_ife_encode(skb
, a
, res
);
798 if (!(ife
->flags
& IFE_ENCODE
))
799 return tcf_ife_decode(skb
, a
, res
);
801 pr_info_ratelimited("unknown failure(policy neither de/encode\n");
802 spin_lock(&ife
->tcf_lock
);
803 bstats_update(&ife
->tcf_bstats
, skb
);
804 tcf_lastuse_update(&ife
->tcf_tm
);
805 ife
->tcf_qstats
.drops
++;
806 spin_unlock(&ife
->tcf_lock
);
811 static int tcf_ife_walker(struct net
*net
, struct sk_buff
*skb
,
812 struct netlink_callback
*cb
, int type
,
815 struct tc_action_net
*tn
= net_generic(net
, ife_net_id
);
817 return tcf_generic_walker(tn
, skb
, cb
, type
, a
);
820 static int tcf_ife_search(struct net
*net
, struct tc_action
*a
, u32 index
)
822 struct tc_action_net
*tn
= net_generic(net
, ife_net_id
);
824 return tcf_hash_search(tn
, a
, index
);
827 static struct tc_action_ops act_ife_ops
= {
830 .owner
= THIS_MODULE
,
832 .dump
= tcf_ife_dump
,
833 .cleanup
= tcf_ife_cleanup
,
834 .init
= tcf_ife_init
,
835 .walk
= tcf_ife_walker
,
836 .lookup
= tcf_ife_search
,
839 static __net_init
int ife_init_net(struct net
*net
)
841 struct tc_action_net
*tn
= net_generic(net
, ife_net_id
);
843 return tc_action_net_init(tn
, &act_ife_ops
, IFE_TAB_MASK
);
846 static void __net_exit
ife_exit_net(struct net
*net
)
848 struct tc_action_net
*tn
= net_generic(net
, ife_net_id
);
850 tc_action_net_exit(tn
);
853 static struct pernet_operations ife_net_ops
= {
854 .init
= ife_init_net
,
855 .exit
= ife_exit_net
,
857 .size
= sizeof(struct tc_action_net
),
860 static int __init
ife_init_module(void)
862 return tcf_register_action(&act_ife_ops
, &ife_net_ops
);
865 static void __exit
ife_cleanup_module(void)
867 tcf_unregister_action(&act_ife_ops
, &ife_net_ops
);
870 module_init(ife_init_module
);
871 module_exit(ife_cleanup_module
);
873 MODULE_AUTHOR("Jamal Hadi Salim(2015)");
874 MODULE_DESCRIPTION("Inter-FE LFB action");
875 MODULE_LICENSE("GPL");