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, gfp_t gfp)
111 mi->metaval = kmemdup(metaval, sizeof(u32), gfp);
117 EXPORT_SYMBOL_GPL(ife_alloc_meta_u32);
119 int ife_alloc_meta_u16(struct tcf_meta_info *mi, void *metaval, gfp_t gfp)
121 mi->metaval = kmemdup(metaval, sizeof(u16), gfp);
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 for existing action
245 static int load_metaops_and_vet(struct tcf_ife_info *ife, u32 metaid,
246 void *val, int len, bool exists)
248 struct tcf_meta_ops *ops = find_ife_oplist(metaid);
253 #ifdef CONFIG_MODULES
255 spin_unlock_bh(&ife->tcf_lock);
257 request_module("ifemeta%u", metaid);
260 spin_lock_bh(&ife->tcf_lock);
261 ops = find_ife_oplist(metaid);
268 ret = ife_validate_metatype(ops, val, len);
270 module_put(ops->owner);
276 /* called when adding new meta information
277 * under ife->tcf_lock for existing action
279 static int add_metainfo(struct tcf_ife_info *ife, u32 metaid, void *metaval,
280 int len, bool atomic)
282 struct tcf_meta_info *mi = NULL;
283 struct tcf_meta_ops *ops = find_ife_oplist(metaid);
289 mi = kzalloc(sizeof(*mi), atomic ? GFP_ATOMIC : GFP_KERNEL);
291 /*put back what find_ife_oplist took */
292 module_put(ops->owner);
299 ret = ops->alloc(mi, metaval, atomic ? GFP_ATOMIC : GFP_KERNEL);
302 module_put(ops->owner);
307 list_add_tail(&mi->metalist, &ife->metalist);
312 static int use_all_metadata(struct tcf_ife_info *ife)
314 struct tcf_meta_ops *o;
318 read_lock(&ife_mod_lock);
319 list_for_each_entry(o, &ifeoplist, list) {
320 rc = add_metainfo(ife, o->metaid, NULL, 0, true);
324 read_unlock(&ife_mod_lock);
332 static int dump_metalist(struct sk_buff *skb, struct tcf_ife_info *ife)
334 struct tcf_meta_info *e;
336 unsigned char *b = skb_tail_pointer(skb);
337 int total_encoded = 0;
339 /*can only happen on decode */
340 if (list_empty(&ife->metalist))
343 nest = nla_nest_start(skb, TCA_IFE_METALST);
347 list_for_each_entry(e, &ife->metalist, metalist) {
348 if (!e->ops->get(skb, e))
355 nla_nest_end(skb, nest);
364 /* under ife->tcf_lock */
365 static void _tcf_ife_cleanup(struct tc_action *a, int bind)
367 struct tcf_ife_info *ife = a->priv;
368 struct tcf_meta_info *e, *n;
370 list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
371 module_put(e->ops->owner);
372 list_del(&e->metalist);
383 static void tcf_ife_cleanup(struct tc_action *a, int bind)
385 struct tcf_ife_info *ife = a->priv;
387 spin_lock_bh(&ife->tcf_lock);
388 _tcf_ife_cleanup(a, bind);
389 spin_unlock_bh(&ife->tcf_lock);
392 /* under ife->tcf_lock for existing action */
393 static int populate_metalist(struct tcf_ife_info *ife, struct nlattr **tb,
401 for (i = 1; i < max_metacnt; i++) {
403 val = nla_data(tb[i]);
404 len = nla_len(tb[i]);
406 rc = load_metaops_and_vet(ife, i, val, len, exists);
410 rc = add_metainfo(ife, i, val, len, exists);
419 static int tcf_ife_init(struct net *net, struct nlattr *nla,
420 struct nlattr *est, struct tc_action *a,
423 struct tc_action_net *tn = net_generic(net, ife_net_id);
424 struct nlattr *tb[TCA_IFE_MAX + 1];
425 struct nlattr *tb2[IFE_META_MAX + 1];
426 struct tcf_ife_info *ife;
431 int ret = 0, exists = 0;
434 err = nla_parse_nested(tb, TCA_IFE_MAX, nla, ife_policy);
438 if (!tb[TCA_IFE_PARMS])
441 parm = nla_data(tb[TCA_IFE_PARMS]);
443 exists = tcf_hash_check(tn, parm->index, a, bind);
447 if (parm->flags & IFE_ENCODE) {
448 /* Until we get issued the ethertype, we cant have
451 if (!tb[TCA_IFE_TYPE]) {
453 tcf_hash_release(a, bind);
454 pr_info("You MUST pass etherype for encoding\n");
460 ret = tcf_hash_create(tn, parm->index, est, a, sizeof(*ife),
466 tcf_hash_release(a, bind);
472 ife->flags = parm->flags;
474 if (parm->flags & IFE_ENCODE) {
475 ife_type = nla_get_u16(tb[TCA_IFE_TYPE]);
476 if (tb[TCA_IFE_DMAC])
477 daddr = nla_data(tb[TCA_IFE_DMAC]);
478 if (tb[TCA_IFE_SMAC])
479 saddr = nla_data(tb[TCA_IFE_SMAC]);
483 spin_lock_bh(&ife->tcf_lock);
484 ife->tcf_action = parm->action;
486 if (parm->flags & IFE_ENCODE) {
488 ether_addr_copy(ife->eth_dst, daddr);
490 eth_zero_addr(ife->eth_dst);
493 ether_addr_copy(ife->eth_src, saddr);
495 eth_zero_addr(ife->eth_src);
497 ife->eth_type = ife_type;
500 if (ret == ACT_P_CREATED)
501 INIT_LIST_HEAD(&ife->metalist);
503 if (tb[TCA_IFE_METALST]) {
504 err = nla_parse_nested(tb2, IFE_META_MAX, tb[TCA_IFE_METALST],
509 tcf_hash_release(a, bind);
510 if (ret == ACT_P_CREATED)
511 _tcf_ife_cleanup(a, bind);
514 spin_unlock_bh(&ife->tcf_lock);
518 err = populate_metalist(ife, tb2, exists);
520 goto metadata_parse_err;
523 /* if no passed metadata allow list or passed allow-all
524 * then here we process by adding as many supported metadatum
525 * as we can. You better have at least one else we are
528 err = use_all_metadata(ife);
530 if (ret == ACT_P_CREATED)
531 _tcf_ife_cleanup(a, bind);
534 spin_unlock_bh(&ife->tcf_lock);
540 spin_unlock_bh(&ife->tcf_lock);
542 if (ret == ACT_P_CREATED)
543 tcf_hash_insert(tn, a);
548 static int tcf_ife_dump(struct sk_buff *skb, struct tc_action *a, int bind,
551 unsigned char *b = skb_tail_pointer(skb);
552 struct tcf_ife_info *ife = a->priv;
553 struct tc_ife opt = {
554 .index = ife->tcf_index,
555 .refcnt = ife->tcf_refcnt - ref,
556 .bindcnt = ife->tcf_bindcnt - bind,
557 .action = ife->tcf_action,
562 if (nla_put(skb, TCA_IFE_PARMS, sizeof(opt), &opt))
563 goto nla_put_failure;
565 t.install = jiffies_to_clock_t(jiffies - ife->tcf_tm.install);
566 t.lastuse = jiffies_to_clock_t(jiffies - ife->tcf_tm.lastuse);
567 t.expires = jiffies_to_clock_t(ife->tcf_tm.expires);
568 if (nla_put_64bit(skb, TCA_IFE_TM, sizeof(t), &t, TCA_IFE_PAD))
569 goto nla_put_failure;
571 if (!is_zero_ether_addr(ife->eth_dst)) {
572 if (nla_put(skb, TCA_IFE_DMAC, ETH_ALEN, ife->eth_dst))
573 goto nla_put_failure;
576 if (!is_zero_ether_addr(ife->eth_src)) {
577 if (nla_put(skb, TCA_IFE_SMAC, ETH_ALEN, ife->eth_src))
578 goto nla_put_failure;
581 if (nla_put(skb, TCA_IFE_TYPE, 2, &ife->eth_type))
582 goto nla_put_failure;
584 if (dump_metalist(skb, ife)) {
585 /*ignore failure to dump metalist */
586 pr_info("Failed to dump metalist\n");
596 int find_decode_metaid(struct sk_buff *skb, struct tcf_ife_info *ife,
597 u16 metaid, u16 mlen, void *mdata)
599 struct tcf_meta_info *e;
601 /* XXX: use hash to speed up */
602 list_for_each_entry(e, &ife->metalist, metalist) {
603 if (metaid == e->metaid) {
605 /* We check for decode presence already */
606 return e->ops->decode(skb, mdata, mlen);
624 static int tcf_ife_decode(struct sk_buff *skb, const struct tc_action *a,
625 struct tcf_result *res)
627 struct tcf_ife_info *ife = a->priv;
628 int action = ife->tcf_action;
629 struct ifeheadr *ifehdr = (struct ifeheadr *)skb->data;
630 u16 ifehdrln = ifehdr->metalen;
631 struct meta_tlvhdr *tlv = (struct meta_tlvhdr *)(ifehdr->tlv_data);
633 spin_lock(&ife->tcf_lock);
634 bstats_update(&ife->tcf_bstats, skb);
635 ife->tcf_tm.lastuse = jiffies;
636 spin_unlock(&ife->tcf_lock);
638 ifehdrln = ntohs(ifehdrln);
639 if (unlikely(!pskb_may_pull(skb, ifehdrln))) {
640 spin_lock(&ife->tcf_lock);
641 ife->tcf_qstats.drops++;
642 spin_unlock(&ife->tcf_lock);
646 skb_set_mac_header(skb, ifehdrln);
647 __skb_pull(skb, ifehdrln);
648 skb->protocol = eth_type_trans(skb, skb->dev);
649 ifehdrln -= IFE_METAHDRLEN;
651 while (ifehdrln > 0) {
652 u8 *tlvdata = (u8 *)tlv;
653 u16 mtype = tlv->type;
656 mtype = ntohs(mtype);
659 if (find_decode_metaid(skb, ife, mtype, (mlen - 4),
660 (void *)(tlvdata + 4))) {
661 /* abuse overlimits to count when we receive metadata
662 * but dont have an ops for it
664 pr_info_ratelimited("Unknown metaid %d alnlen %d\n",
666 ife->tcf_qstats.overlimits++;
671 tlv = (struct meta_tlvhdr *)tlvdata;
674 skb_reset_network_header(skb);
678 /*XXX: check if we can do this at install time instead of current
681 static int ife_get_sz(struct sk_buff *skb, struct tcf_ife_info *ife)
683 struct tcf_meta_info *e, *n;
684 int tot_run_sz = 0, run_sz = 0;
686 list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
687 if (e->ops->check_presence) {
688 run_sz = e->ops->check_presence(skb, e);
689 tot_run_sz += run_sz;
696 static int tcf_ife_encode(struct sk_buff *skb, const struct tc_action *a,
697 struct tcf_result *res)
699 struct tcf_ife_info *ife = a->priv;
700 int action = ife->tcf_action;
701 struct ethhdr *oethh; /* outer ether header */
702 struct ethhdr *iethh; /* inner eth header */
703 struct tcf_meta_info *e;
705 OUTERHDR:TOTMETALEN:{TLVHDR:Metadatum:TLVHDR..}:ORIGDATA
706 where ORIGDATA = original ethernet header ...
708 u16 metalen = ife_get_sz(skb, ife);
709 int hdrm = metalen + skb->dev->hard_header_len + IFE_METAHDRLEN;
710 unsigned int skboff = skb->dev->hard_header_len;
711 u32 at = G_TC_AT(skb->tc_verd);
712 int new_len = skb->len + hdrm;
713 bool exceed_mtu = false;
716 if (at & AT_EGRESS) {
717 if (new_len > skb->dev->mtu)
721 spin_lock(&ife->tcf_lock);
722 bstats_update(&ife->tcf_bstats, skb);
723 ife->tcf_tm.lastuse = jiffies;
725 if (!metalen) { /* no metadata to send */
726 /* abuse overlimits to count when we allow packet
729 ife->tcf_qstats.overlimits++;
730 spin_unlock(&ife->tcf_lock);
733 /* could be stupid policy setup or mtu config
734 * so lets be conservative.. */
735 if ((action == TC_ACT_SHOT) || exceed_mtu) {
736 ife->tcf_qstats.drops++;
737 spin_unlock(&ife->tcf_lock);
741 iethh = eth_hdr(skb);
743 err = skb_cow_head(skb, hdrm);
745 ife->tcf_qstats.drops++;
746 spin_unlock(&ife->tcf_lock);
750 if (!(at & AT_EGRESS))
751 skb_push(skb, skb->dev->hard_header_len);
753 __skb_push(skb, hdrm);
754 memcpy(skb->data, iethh, skb->mac_len);
755 skb_reset_mac_header(skb);
756 oethh = eth_hdr(skb);
758 /*total metadata length */
759 metalen += IFE_METAHDRLEN;
760 metalen = htons(metalen);
761 memcpy((skb->data + skboff), &metalen, IFE_METAHDRLEN);
762 skboff += IFE_METAHDRLEN;
764 /* XXX: we dont have a clever way of telling encode to
765 * not repeat some of the computations that are done by
766 * ops->presence_check...
768 list_for_each_entry(e, &ife->metalist, metalist) {
769 if (e->ops->encode) {
770 err = e->ops->encode(skb, (void *)(skb->data + skboff),
774 /* too corrupt to keep around if overwritten */
775 ife->tcf_qstats.drops++;
776 spin_unlock(&ife->tcf_lock);
782 if (!is_zero_ether_addr(ife->eth_src))
783 ether_addr_copy(oethh->h_source, ife->eth_src);
785 ether_addr_copy(oethh->h_source, iethh->h_source);
786 if (!is_zero_ether_addr(ife->eth_dst))
787 ether_addr_copy(oethh->h_dest, ife->eth_dst);
789 ether_addr_copy(oethh->h_dest, iethh->h_dest);
790 oethh->h_proto = htons(ife->eth_type);
792 if (!(at & AT_EGRESS))
793 skb_pull(skb, skb->dev->hard_header_len);
795 spin_unlock(&ife->tcf_lock);
800 static int tcf_ife_act(struct sk_buff *skb, const struct tc_action *a,
801 struct tcf_result *res)
803 struct tcf_ife_info *ife = a->priv;
805 if (ife->flags & IFE_ENCODE)
806 return tcf_ife_encode(skb, a, res);
808 if (!(ife->flags & IFE_ENCODE))
809 return tcf_ife_decode(skb, a, res);
811 pr_info_ratelimited("unknown failure(policy neither de/encode\n");
812 spin_lock(&ife->tcf_lock);
813 bstats_update(&ife->tcf_bstats, skb);
814 ife->tcf_tm.lastuse = jiffies;
815 ife->tcf_qstats.drops++;
816 spin_unlock(&ife->tcf_lock);
821 static int tcf_ife_walker(struct net *net, struct sk_buff *skb,
822 struct netlink_callback *cb, int type,
825 struct tc_action_net *tn = net_generic(net, ife_net_id);
827 return tcf_generic_walker(tn, skb, cb, type, a);
830 static int tcf_ife_search(struct net *net, struct tc_action *a, u32 index)
832 struct tc_action_net *tn = net_generic(net, ife_net_id);
834 return tcf_hash_search(tn, a, index);
837 static struct tc_action_ops act_ife_ops = {
840 .owner = THIS_MODULE,
842 .dump = tcf_ife_dump,
843 .cleanup = tcf_ife_cleanup,
844 .init = tcf_ife_init,
845 .walk = tcf_ife_walker,
846 .lookup = tcf_ife_search,
849 static __net_init int ife_init_net(struct net *net)
851 struct tc_action_net *tn = net_generic(net, ife_net_id);
853 return tc_action_net_init(tn, &act_ife_ops, IFE_TAB_MASK);
856 static void __net_exit ife_exit_net(struct net *net)
858 struct tc_action_net *tn = net_generic(net, ife_net_id);
860 tc_action_net_exit(tn);
863 static struct pernet_operations ife_net_ops = {
864 .init = ife_init_net,
865 .exit = ife_exit_net,
867 .size = sizeof(struct tc_action_net),
870 static int __init ife_init_module(void)
872 return tcf_register_action(&act_ife_ops, &ife_net_ops);
875 static void __exit ife_cleanup_module(void)
877 tcf_unregister_action(&act_ife_ops, &ife_net_ops);
880 module_init(ife_init_module);
881 module_exit(ife_cleanup_module);
883 MODULE_AUTHOR("Jamal Hadi Salim(2015)");
884 MODULE_DESCRIPTION("Inter-FE LFB action");
885 MODULE_LICENSE("GPL");