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 #ifdef CONFIG_MODULES
252 static const char *ife_meta_id2name(u32 metaid)
255 case IFE_META_SKBMARK:
259 case IFE_META_TCINDEX:
267 /* called when adding new meta information
269 static int load_metaops_and_vet(u32 metaid, void *val, int len, bool rtnl_held)
271 struct tcf_meta_ops *ops = find_ife_oplist(metaid);
276 #ifdef CONFIG_MODULES
279 request_module("ife-meta-%s", ife_meta_id2name(metaid));
282 ops = find_ife_oplist(metaid);
289 ret = ife_validate_metatype(ops, val, len);
291 module_put(ops->owner);
297 /* called when adding new meta information
299 static int __add_metainfo(const struct tcf_meta_ops *ops,
300 struct tcf_ife_info *ife, u32 metaid, void *metaval,
301 int len, bool atomic, bool exists)
303 struct tcf_meta_info *mi = NULL;
306 mi = kzalloc(sizeof(*mi), atomic ? GFP_ATOMIC : GFP_KERNEL);
313 ret = ops->alloc(mi, metaval, atomic ? GFP_ATOMIC : GFP_KERNEL);
321 spin_lock_bh(&ife->tcf_lock);
322 list_add_tail(&mi->metalist, &ife->metalist);
324 spin_unlock_bh(&ife->tcf_lock);
329 static int add_metainfo_and_get_ops(const struct tcf_meta_ops *ops,
330 struct tcf_ife_info *ife, u32 metaid,
335 if (!try_module_get(ops->owner))
337 ret = __add_metainfo(ops, ife, metaid, NULL, 0, true, exists);
339 module_put(ops->owner);
343 static int add_metainfo(struct tcf_ife_info *ife, u32 metaid, void *metaval,
344 int len, bool exists)
346 const struct tcf_meta_ops *ops = find_ife_oplist(metaid);
351 ret = __add_metainfo(ops, ife, metaid, metaval, len, false, exists);
353 /*put back what find_ife_oplist took */
354 module_put(ops->owner);
358 static int use_all_metadata(struct tcf_ife_info *ife, bool exists)
360 struct tcf_meta_ops *o;
364 read_lock(&ife_mod_lock);
365 list_for_each_entry(o, &ifeoplist, list) {
366 rc = add_metainfo_and_get_ops(o, ife, o->metaid, exists);
370 read_unlock(&ife_mod_lock);
378 static int dump_metalist(struct sk_buff *skb, struct tcf_ife_info *ife)
380 struct tcf_meta_info *e;
382 unsigned char *b = skb_tail_pointer(skb);
383 int total_encoded = 0;
385 /*can only happen on decode */
386 if (list_empty(&ife->metalist))
389 nest = nla_nest_start(skb, TCA_IFE_METALST);
393 list_for_each_entry(e, &ife->metalist, metalist) {
394 if (!e->ops->get(skb, e))
401 nla_nest_end(skb, nest);
410 /* under ife->tcf_lock */
411 static void _tcf_ife_cleanup(struct tc_action *a)
413 struct tcf_ife_info *ife = to_ife(a);
414 struct tcf_meta_info *e, *n;
416 list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
417 list_del(&e->metalist);
424 module_put(e->ops->owner);
429 static void tcf_ife_cleanup(struct tc_action *a)
431 struct tcf_ife_info *ife = to_ife(a);
432 struct tcf_ife_params *p;
434 spin_lock_bh(&ife->tcf_lock);
436 spin_unlock_bh(&ife->tcf_lock);
438 p = rcu_dereference_protected(ife->params, 1);
443 static int populate_metalist(struct tcf_ife_info *ife, struct nlattr **tb,
444 bool exists, bool rtnl_held)
451 for (i = 1; i < max_metacnt; i++) {
453 val = nla_data(tb[i]);
454 len = nla_len(tb[i]);
456 rc = load_metaops_and_vet(i, val, len, rtnl_held);
460 rc = add_metainfo(ife, i, val, len, exists);
469 static int tcf_ife_init(struct net *net, struct nlattr *nla,
470 struct nlattr *est, struct tc_action **a,
471 int ovr, int bind, bool rtnl_held,
472 struct netlink_ext_ack *extack)
474 struct tc_action_net *tn = net_generic(net, ife_net_id);
475 struct nlattr *tb[TCA_IFE_MAX + 1];
476 struct nlattr *tb2[IFE_META_MAX + 1];
477 struct tcf_ife_params *p;
478 struct tcf_ife_info *ife;
479 u16 ife_type = ETH_P_IFE;
489 NL_SET_ERR_MSG_MOD(extack, "IFE requires attributes to be passed");
493 err = nla_parse_nested(tb, TCA_IFE_MAX, nla, ife_policy, NULL);
497 if (!tb[TCA_IFE_PARMS])
500 parm = nla_data(tb[TCA_IFE_PARMS]);
502 /* IFE_DECODE is 0 and indicates the opposite of IFE_ENCODE because
503 * they cannot run as the same time. Check on all other values which
504 * are not supported right now.
506 if (parm->flags & ~IFE_ENCODE)
509 p = kzalloc(sizeof(*p), GFP_KERNEL);
514 err = tcf_idr_check_alloc(tn, &index, a, bind);
520 if (exists && bind) {
526 ret = tcf_idr_create(tn, index, est, a, &act_ife_ops,
529 tcf_idr_cleanup(tn, index);
535 tcf_idr_release(*a, bind);
541 p->flags = parm->flags;
543 if (parm->flags & IFE_ENCODE) {
544 if (tb[TCA_IFE_TYPE])
545 ife_type = nla_get_u16(tb[TCA_IFE_TYPE]);
546 if (tb[TCA_IFE_DMAC])
547 daddr = nla_data(tb[TCA_IFE_DMAC]);
548 if (tb[TCA_IFE_SMAC])
549 saddr = nla_data(tb[TCA_IFE_SMAC]);
552 if (parm->flags & IFE_ENCODE) {
554 ether_addr_copy(p->eth_dst, daddr);
556 eth_zero_addr(p->eth_dst);
559 ether_addr_copy(p->eth_src, saddr);
561 eth_zero_addr(p->eth_src);
563 p->eth_type = ife_type;
567 if (ret == ACT_P_CREATED)
568 INIT_LIST_HEAD(&ife->metalist);
570 if (tb[TCA_IFE_METALST]) {
571 err = nla_parse_nested(tb2, IFE_META_MAX, tb[TCA_IFE_METALST],
575 tcf_idr_release(*a, bind);
580 err = populate_metalist(ife, tb2, exists, rtnl_held);
582 goto metadata_parse_err;
585 /* if no passed metadata allow list or passed allow-all
586 * then here we process by adding as many supported metadatum
587 * as we can. You better have at least one else we are
590 err = use_all_metadata(ife, exists);
592 tcf_idr_release(*a, bind);
599 spin_lock_bh(&ife->tcf_lock);
600 ife->tcf_action = parm->action;
601 /* protected by tcf_lock when modifying existing action */
602 rcu_swap_protected(ife->params, p, 1);
605 spin_unlock_bh(&ife->tcf_lock);
609 if (ret == ACT_P_CREATED)
610 tcf_idr_insert(tn, *a);
615 static int tcf_ife_dump(struct sk_buff *skb, struct tc_action *a, int bind,
618 unsigned char *b = skb_tail_pointer(skb);
619 struct tcf_ife_info *ife = to_ife(a);
620 struct tcf_ife_params *p;
621 struct tc_ife opt = {
622 .index = ife->tcf_index,
623 .refcnt = refcount_read(&ife->tcf_refcnt) - ref,
624 .bindcnt = atomic_read(&ife->tcf_bindcnt) - bind,
628 spin_lock_bh(&ife->tcf_lock);
629 opt.action = ife->tcf_action;
630 p = rcu_dereference_protected(ife->params,
631 lockdep_is_held(&ife->tcf_lock));
632 opt.flags = p->flags;
634 if (nla_put(skb, TCA_IFE_PARMS, sizeof(opt), &opt))
635 goto nla_put_failure;
637 tcf_tm_dump(&t, &ife->tcf_tm);
638 if (nla_put_64bit(skb, TCA_IFE_TM, sizeof(t), &t, TCA_IFE_PAD))
639 goto nla_put_failure;
641 if (!is_zero_ether_addr(p->eth_dst)) {
642 if (nla_put(skb, TCA_IFE_DMAC, ETH_ALEN, p->eth_dst))
643 goto nla_put_failure;
646 if (!is_zero_ether_addr(p->eth_src)) {
647 if (nla_put(skb, TCA_IFE_SMAC, ETH_ALEN, p->eth_src))
648 goto nla_put_failure;
651 if (nla_put(skb, TCA_IFE_TYPE, 2, &p->eth_type))
652 goto nla_put_failure;
654 if (dump_metalist(skb, ife)) {
655 /*ignore failure to dump metalist */
656 pr_info("Failed to dump metalist\n");
659 spin_unlock_bh(&ife->tcf_lock);
663 spin_unlock_bh(&ife->tcf_lock);
668 static int find_decode_metaid(struct sk_buff *skb, struct tcf_ife_info *ife,
669 u16 metaid, u16 mlen, void *mdata)
671 struct tcf_meta_info *e;
673 /* XXX: use hash to speed up */
674 list_for_each_entry(e, &ife->metalist, metalist) {
675 if (metaid == e->metaid) {
677 /* We check for decode presence already */
678 return e->ops->decode(skb, mdata, mlen);
686 static int tcf_ife_decode(struct sk_buff *skb, const struct tc_action *a,
687 struct tcf_result *res)
689 struct tcf_ife_info *ife = to_ife(a);
690 int action = ife->tcf_action;
695 bstats_cpu_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
696 tcf_lastuse_update(&ife->tcf_tm);
698 if (skb_at_tc_ingress(skb))
699 skb_push(skb, skb->dev->hard_header_len);
701 tlv_data = ife_decode(skb, &metalen);
702 if (unlikely(!tlv_data)) {
703 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
707 ifehdr_end = tlv_data + metalen;
708 for (; tlv_data < ifehdr_end; tlv_data = ife_tlv_meta_next(tlv_data)) {
713 curr_data = ife_tlv_meta_decode(tlv_data, ifehdr_end, &mtype,
716 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
720 if (find_decode_metaid(skb, ife, mtype, dlen, curr_data)) {
721 /* abuse overlimits to count when we receive metadata
722 * but dont have an ops for it
724 pr_info_ratelimited("Unknown metaid %d dlen %d\n",
726 qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
730 if (WARN_ON(tlv_data != ifehdr_end)) {
731 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
735 skb->protocol = eth_type_trans(skb, skb->dev);
736 skb_reset_network_header(skb);
741 /*XXX: check if we can do this at install time instead of current
744 static int ife_get_sz(struct sk_buff *skb, struct tcf_ife_info *ife)
746 struct tcf_meta_info *e, *n;
747 int tot_run_sz = 0, run_sz = 0;
749 list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
750 if (e->ops->check_presence) {
751 run_sz = e->ops->check_presence(skb, e);
752 tot_run_sz += run_sz;
759 static int tcf_ife_encode(struct sk_buff *skb, const struct tc_action *a,
760 struct tcf_result *res, struct tcf_ife_params *p)
762 struct tcf_ife_info *ife = to_ife(a);
763 int action = ife->tcf_action;
764 struct ethhdr *oethh; /* outer ether header */
765 struct tcf_meta_info *e;
767 OUTERHDR:TOTMETALEN:{TLVHDR:Metadatum:TLVHDR..}:ORIGDATA
768 where ORIGDATA = original ethernet header ...
770 u16 metalen = ife_get_sz(skb, ife);
771 int hdrm = metalen + skb->dev->hard_header_len + IFE_METAHDRLEN;
772 unsigned int skboff = 0;
773 int new_len = skb->len + hdrm;
774 bool exceed_mtu = false;
778 if (!skb_at_tc_ingress(skb)) {
779 if (new_len > skb->dev->mtu)
783 bstats_cpu_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
784 tcf_lastuse_update(&ife->tcf_tm);
786 if (!metalen) { /* no metadata to send */
787 /* abuse overlimits to count when we allow packet
790 qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
793 /* could be stupid policy setup or mtu config
794 * so lets be conservative.. */
795 if ((action == TC_ACT_SHOT) || exceed_mtu) {
796 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
800 if (skb_at_tc_ingress(skb))
801 skb_push(skb, skb->dev->hard_header_len);
803 ife_meta = ife_encode(skb, metalen);
805 spin_lock(&ife->tcf_lock);
807 /* XXX: we dont have a clever way of telling encode to
808 * not repeat some of the computations that are done by
809 * ops->presence_check...
811 list_for_each_entry(e, &ife->metalist, metalist) {
812 if (e->ops->encode) {
813 err = e->ops->encode(skb, (void *)(ife_meta + skboff),
817 /* too corrupt to keep around if overwritten */
818 spin_unlock(&ife->tcf_lock);
819 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
824 spin_unlock(&ife->tcf_lock);
825 oethh = (struct ethhdr *)skb->data;
827 if (!is_zero_ether_addr(p->eth_src))
828 ether_addr_copy(oethh->h_source, p->eth_src);
829 if (!is_zero_ether_addr(p->eth_dst))
830 ether_addr_copy(oethh->h_dest, p->eth_dst);
831 oethh->h_proto = htons(p->eth_type);
833 if (skb_at_tc_ingress(skb))
834 skb_pull(skb, skb->dev->hard_header_len);
839 static int tcf_ife_act(struct sk_buff *skb, const struct tc_action *a,
840 struct tcf_result *res)
842 struct tcf_ife_info *ife = to_ife(a);
843 struct tcf_ife_params *p;
846 p = rcu_dereference_bh(ife->params);
847 if (p->flags & IFE_ENCODE) {
848 ret = tcf_ife_encode(skb, a, res, p);
852 return tcf_ife_decode(skb, a, res);
855 static int tcf_ife_walker(struct net *net, struct sk_buff *skb,
856 struct netlink_callback *cb, int type,
857 const struct tc_action_ops *ops,
858 struct netlink_ext_ack *extack)
860 struct tc_action_net *tn = net_generic(net, ife_net_id);
862 return tcf_generic_walker(tn, skb, cb, type, ops, extack);
865 static int tcf_ife_search(struct net *net, struct tc_action **a, u32 index,
866 struct netlink_ext_ack *extack)
868 struct tc_action_net *tn = net_generic(net, ife_net_id);
870 return tcf_idr_search(tn, a, index);
873 static struct tc_action_ops act_ife_ops = {
876 .owner = THIS_MODULE,
878 .dump = tcf_ife_dump,
879 .cleanup = tcf_ife_cleanup,
880 .init = tcf_ife_init,
881 .walk = tcf_ife_walker,
882 .lookup = tcf_ife_search,
883 .size = sizeof(struct tcf_ife_info),
886 static __net_init int ife_init_net(struct net *net)
888 struct tc_action_net *tn = net_generic(net, ife_net_id);
890 return tc_action_net_init(net, tn, &act_ife_ops);
893 static void __net_exit ife_exit_net(struct list_head *net_list)
895 tc_action_net_exit(net_list, ife_net_id);
898 static struct pernet_operations ife_net_ops = {
899 .init = ife_init_net,
900 .exit_batch = ife_exit_net,
902 .size = sizeof(struct tc_action_net),
905 static int __init ife_init_module(void)
907 return tcf_register_action(&act_ife_ops, &ife_net_ops);
910 static void __exit ife_cleanup_module(void)
912 tcf_unregister_action(&act_ife_ops, &ife_net_ops);
915 module_init(ife_init_module);
916 module_exit(ife_cleanup_module);
918 MODULE_AUTHOR("Jamal Hadi Salim(2015)");
919 MODULE_DESCRIPTION("Inter-FE LFB action");
920 MODULE_LICENSE("GPL");