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 <net/pkt_cls.h>
33 #include <uapi/linux/tc_act/tc_ife.h>
34 #include <net/tc_act/tc_ife.h>
35 #include <linux/etherdevice.h>
38 static unsigned int ife_net_id;
39 static int max_metacnt = IFE_META_MAX + 1;
40 static struct tc_action_ops act_ife_ops;
42 static const struct nla_policy ife_policy[TCA_IFE_MAX + 1] = {
43 [TCA_IFE_PARMS] = { .len = sizeof(struct tc_ife)},
44 [TCA_IFE_DMAC] = { .len = ETH_ALEN},
45 [TCA_IFE_SMAC] = { .len = ETH_ALEN},
46 [TCA_IFE_TYPE] = { .type = NLA_U16},
49 int ife_encode_meta_u16(u16 metaval, void *skbdata, struct tcf_meta_info *mi)
54 edata = *(u16 *)mi->metaval;
58 if (!edata) /* will not encode */
62 return ife_tlv_meta_encode(skbdata, mi->metaid, 2, &edata);
64 EXPORT_SYMBOL_GPL(ife_encode_meta_u16);
66 int ife_get_meta_u32(struct sk_buff *skb, struct tcf_meta_info *mi)
69 return nla_put_u32(skb, mi->metaid, *(u32 *)mi->metaval);
71 return nla_put(skb, mi->metaid, 0, NULL);
73 EXPORT_SYMBOL_GPL(ife_get_meta_u32);
75 int ife_check_meta_u32(u32 metaval, struct tcf_meta_info *mi)
77 if (metaval || mi->metaval)
78 return 8; /* T+L+V == 2+2+4 */
82 EXPORT_SYMBOL_GPL(ife_check_meta_u32);
84 int ife_check_meta_u16(u16 metaval, struct tcf_meta_info *mi)
86 if (metaval || mi->metaval)
87 return 8; /* T+L+(V) == 2+2+(2+2bytepad) */
91 EXPORT_SYMBOL_GPL(ife_check_meta_u16);
93 int ife_encode_meta_u32(u32 metaval, void *skbdata, struct tcf_meta_info *mi)
98 edata = *(u32 *)mi->metaval;
102 if (!edata) /* will not encode */
105 edata = htonl(edata);
106 return ife_tlv_meta_encode(skbdata, mi->metaid, 4, &edata);
108 EXPORT_SYMBOL_GPL(ife_encode_meta_u32);
110 int ife_get_meta_u16(struct sk_buff *skb, struct tcf_meta_info *mi)
113 return nla_put_u16(skb, mi->metaid, *(u16 *)mi->metaval);
115 return nla_put(skb, mi->metaid, 0, NULL);
117 EXPORT_SYMBOL_GPL(ife_get_meta_u16);
119 int ife_alloc_meta_u32(struct tcf_meta_info *mi, void *metaval, gfp_t gfp)
121 mi->metaval = kmemdup(metaval, sizeof(u32), gfp);
127 EXPORT_SYMBOL_GPL(ife_alloc_meta_u32);
129 int ife_alloc_meta_u16(struct tcf_meta_info *mi, void *metaval, gfp_t gfp)
131 mi->metaval = kmemdup(metaval, sizeof(u16), gfp);
137 EXPORT_SYMBOL_GPL(ife_alloc_meta_u16);
139 void ife_release_meta_gen(struct tcf_meta_info *mi)
143 EXPORT_SYMBOL_GPL(ife_release_meta_gen);
145 int ife_validate_meta_u32(void *val, int len)
147 if (len == sizeof(u32))
152 EXPORT_SYMBOL_GPL(ife_validate_meta_u32);
154 int ife_validate_meta_u16(void *val, int len)
156 /* length will not include padding */
157 if (len == sizeof(u16))
162 EXPORT_SYMBOL_GPL(ife_validate_meta_u16);
164 static LIST_HEAD(ifeoplist);
165 static DEFINE_RWLOCK(ife_mod_lock);
167 static struct tcf_meta_ops *find_ife_oplist(u16 metaid)
169 struct tcf_meta_ops *o;
171 read_lock(&ife_mod_lock);
172 list_for_each_entry(o, &ifeoplist, list) {
173 if (o->metaid == metaid) {
174 if (!try_module_get(o->owner))
176 read_unlock(&ife_mod_lock);
180 read_unlock(&ife_mod_lock);
185 int register_ife_op(struct tcf_meta_ops *mops)
187 struct tcf_meta_ops *m;
189 if (!mops->metaid || !mops->metatype || !mops->name ||
190 !mops->check_presence || !mops->encode || !mops->decode ||
191 !mops->get || !mops->alloc)
194 write_lock(&ife_mod_lock);
196 list_for_each_entry(m, &ifeoplist, list) {
197 if (m->metaid == mops->metaid ||
198 (strcmp(mops->name, m->name) == 0)) {
199 write_unlock(&ife_mod_lock);
205 mops->release = ife_release_meta_gen;
207 list_add_tail(&mops->list, &ifeoplist);
208 write_unlock(&ife_mod_lock);
211 EXPORT_SYMBOL_GPL(unregister_ife_op);
213 int unregister_ife_op(struct tcf_meta_ops *mops)
215 struct tcf_meta_ops *m;
218 write_lock(&ife_mod_lock);
219 list_for_each_entry(m, &ifeoplist, list) {
220 if (m->metaid == mops->metaid) {
221 list_del(&mops->list);
226 write_unlock(&ife_mod_lock);
230 EXPORT_SYMBOL_GPL(register_ife_op);
232 static int ife_validate_metatype(struct tcf_meta_ops *ops, void *val, int len)
235 /* XXX: unfortunately cant use nla_policy at this point
236 * because a length of 0 is valid in the case of
237 * "allow". "use" semantics do enforce for proper
238 * length and i couldve use nla_policy but it makes it hard
239 * to use it just for that..
242 return ops->validate(val, len);
244 if (ops->metatype == NLA_U32)
245 ret = ife_validate_meta_u32(val, len);
246 else if (ops->metatype == NLA_U16)
247 ret = ife_validate_meta_u16(val, len);
252 #ifdef CONFIG_MODULES
253 static const char *ife_meta_id2name(u32 metaid)
256 case IFE_META_SKBMARK:
260 case IFE_META_TCINDEX:
268 /* called when adding new meta information
270 static int load_metaops_and_vet(u32 metaid, void *val, int len, bool rtnl_held)
272 struct tcf_meta_ops *ops = find_ife_oplist(metaid);
277 #ifdef CONFIG_MODULES
280 request_module("ife-meta-%s", ife_meta_id2name(metaid));
283 ops = find_ife_oplist(metaid);
290 ret = ife_validate_metatype(ops, val, len);
292 module_put(ops->owner);
298 /* called when adding new meta information
300 static int __add_metainfo(const struct tcf_meta_ops *ops,
301 struct tcf_ife_info *ife, u32 metaid, void *metaval,
302 int len, bool atomic, bool exists)
304 struct tcf_meta_info *mi = NULL;
307 mi = kzalloc(sizeof(*mi), atomic ? GFP_ATOMIC : GFP_KERNEL);
314 ret = ops->alloc(mi, metaval, atomic ? GFP_ATOMIC : GFP_KERNEL);
322 spin_lock_bh(&ife->tcf_lock);
323 list_add_tail(&mi->metalist, &ife->metalist);
325 spin_unlock_bh(&ife->tcf_lock);
330 static int add_metainfo_and_get_ops(const struct tcf_meta_ops *ops,
331 struct tcf_ife_info *ife, u32 metaid,
336 if (!try_module_get(ops->owner))
338 ret = __add_metainfo(ops, ife, metaid, NULL, 0, true, exists);
340 module_put(ops->owner);
344 static int add_metainfo(struct tcf_ife_info *ife, u32 metaid, void *metaval,
345 int len, bool exists)
347 const struct tcf_meta_ops *ops = find_ife_oplist(metaid);
352 ret = __add_metainfo(ops, ife, metaid, metaval, len, false, exists);
354 /*put back what find_ife_oplist took */
355 module_put(ops->owner);
359 static int use_all_metadata(struct tcf_ife_info *ife, bool exists)
361 struct tcf_meta_ops *o;
365 read_lock(&ife_mod_lock);
366 list_for_each_entry(o, &ifeoplist, list) {
367 rc = add_metainfo_and_get_ops(o, ife, o->metaid, exists);
371 read_unlock(&ife_mod_lock);
379 static int dump_metalist(struct sk_buff *skb, struct tcf_ife_info *ife)
381 struct tcf_meta_info *e;
383 unsigned char *b = skb_tail_pointer(skb);
384 int total_encoded = 0;
386 /*can only happen on decode */
387 if (list_empty(&ife->metalist))
390 nest = nla_nest_start_noflag(skb, TCA_IFE_METALST);
394 list_for_each_entry(e, &ife->metalist, metalist) {
395 if (!e->ops->get(skb, e))
402 nla_nest_end(skb, nest);
411 /* under ife->tcf_lock */
412 static void _tcf_ife_cleanup(struct tc_action *a)
414 struct tcf_ife_info *ife = to_ife(a);
415 struct tcf_meta_info *e, *n;
417 list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
418 list_del(&e->metalist);
425 module_put(e->ops->owner);
430 static void tcf_ife_cleanup(struct tc_action *a)
432 struct tcf_ife_info *ife = to_ife(a);
433 struct tcf_ife_params *p;
435 spin_lock_bh(&ife->tcf_lock);
437 spin_unlock_bh(&ife->tcf_lock);
439 p = rcu_dereference_protected(ife->params, 1);
444 static int populate_metalist(struct tcf_ife_info *ife, struct nlattr **tb,
445 bool exists, bool rtnl_held)
452 for (i = 1; i < max_metacnt; i++) {
454 val = nla_data(tb[i]);
455 len = nla_len(tb[i]);
457 rc = load_metaops_and_vet(i, val, len, rtnl_held);
461 rc = add_metainfo(ife, i, val, len, exists);
470 static int tcf_ife_init(struct net *net, struct nlattr *nla,
471 struct nlattr *est, struct tc_action **a,
472 int ovr, int bind, bool rtnl_held,
473 struct tcf_proto *tp, struct netlink_ext_ack *extack)
475 struct tc_action_net *tn = net_generic(net, ife_net_id);
476 struct nlattr *tb[TCA_IFE_MAX + 1];
477 struct nlattr *tb2[IFE_META_MAX + 1];
478 struct tcf_chain *goto_ch = NULL;
479 struct tcf_ife_params *p;
480 struct tcf_ife_info *ife;
481 u16 ife_type = ETH_P_IFE;
489 err = nla_parse_nested_deprecated(tb, TCA_IFE_MAX, nla, ife_policy,
494 if (!tb[TCA_IFE_PARMS])
497 parm = nla_data(tb[TCA_IFE_PARMS]);
499 /* IFE_DECODE is 0 and indicates the opposite of IFE_ENCODE because
500 * they cannot run as the same time. Check on all other values which
501 * are not supported right now.
503 if (parm->flags & ~IFE_ENCODE)
506 p = kzalloc(sizeof(*p), GFP_KERNEL);
510 err = tcf_idr_check_alloc(tn, &parm->index, a, bind);
516 if (exists && bind) {
522 ret = tcf_idr_create(tn, parm->index, est, a, &act_ife_ops,
525 tcf_idr_cleanup(tn, parm->index);
531 tcf_idr_release(*a, bind);
537 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
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_deprecated(tb2, IFE_META_MAX,
572 tb[TCA_IFE_METALST], NULL,
575 goto metadata_parse_err;
576 err = populate_metalist(ife, tb2, exists, rtnl_held);
578 goto metadata_parse_err;
581 /* if no passed metadata allow list or passed allow-all
582 * then here we process by adding as many supported metadatum
583 * as we can. You better have at least one else we are
586 err = use_all_metadata(ife, exists);
588 goto metadata_parse_err;
592 spin_lock_bh(&ife->tcf_lock);
593 /* protected by tcf_lock when modifying existing action */
594 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
595 rcu_swap_protected(ife->params, p, 1);
598 spin_unlock_bh(&ife->tcf_lock);
600 tcf_chain_put_by_act(goto_ch);
604 if (ret == ACT_P_CREATED)
605 tcf_idr_insert(tn, *a);
610 tcf_chain_put_by_act(goto_ch);
613 tcf_idr_release(*a, bind);
617 static int tcf_ife_dump(struct sk_buff *skb, struct tc_action *a, int bind,
620 unsigned char *b = skb_tail_pointer(skb);
621 struct tcf_ife_info *ife = to_ife(a);
622 struct tcf_ife_params *p;
623 struct tc_ife opt = {
624 .index = ife->tcf_index,
625 .refcnt = refcount_read(&ife->tcf_refcnt) - ref,
626 .bindcnt = atomic_read(&ife->tcf_bindcnt) - bind,
630 spin_lock_bh(&ife->tcf_lock);
631 opt.action = ife->tcf_action;
632 p = rcu_dereference_protected(ife->params,
633 lockdep_is_held(&ife->tcf_lock));
634 opt.flags = p->flags;
636 if (nla_put(skb, TCA_IFE_PARMS, sizeof(opt), &opt))
637 goto nla_put_failure;
639 tcf_tm_dump(&t, &ife->tcf_tm);
640 if (nla_put_64bit(skb, TCA_IFE_TM, sizeof(t), &t, TCA_IFE_PAD))
641 goto nla_put_failure;
643 if (!is_zero_ether_addr(p->eth_dst)) {
644 if (nla_put(skb, TCA_IFE_DMAC, ETH_ALEN, p->eth_dst))
645 goto nla_put_failure;
648 if (!is_zero_ether_addr(p->eth_src)) {
649 if (nla_put(skb, TCA_IFE_SMAC, ETH_ALEN, p->eth_src))
650 goto nla_put_failure;
653 if (nla_put(skb, TCA_IFE_TYPE, 2, &p->eth_type))
654 goto nla_put_failure;
656 if (dump_metalist(skb, ife)) {
657 /*ignore failure to dump metalist */
658 pr_info("Failed to dump metalist\n");
661 spin_unlock_bh(&ife->tcf_lock);
665 spin_unlock_bh(&ife->tcf_lock);
670 static int find_decode_metaid(struct sk_buff *skb, struct tcf_ife_info *ife,
671 u16 metaid, u16 mlen, void *mdata)
673 struct tcf_meta_info *e;
675 /* XXX: use hash to speed up */
676 list_for_each_entry(e, &ife->metalist, metalist) {
677 if (metaid == e->metaid) {
679 /* We check for decode presence already */
680 return e->ops->decode(skb, mdata, mlen);
688 static int tcf_ife_decode(struct sk_buff *skb, const struct tc_action *a,
689 struct tcf_result *res)
691 struct tcf_ife_info *ife = to_ife(a);
692 int action = ife->tcf_action;
697 bstats_cpu_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
698 tcf_lastuse_update(&ife->tcf_tm);
700 if (skb_at_tc_ingress(skb))
701 skb_push(skb, skb->dev->hard_header_len);
703 tlv_data = ife_decode(skb, &metalen);
704 if (unlikely(!tlv_data)) {
705 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
709 ifehdr_end = tlv_data + metalen;
710 for (; tlv_data < ifehdr_end; tlv_data = ife_tlv_meta_next(tlv_data)) {
715 curr_data = ife_tlv_meta_decode(tlv_data, ifehdr_end, &mtype,
718 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
722 if (find_decode_metaid(skb, ife, mtype, dlen, curr_data)) {
723 /* abuse overlimits to count when we receive metadata
724 * but dont have an ops for it
726 pr_info_ratelimited("Unknown metaid %d dlen %d\n",
728 qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
732 if (WARN_ON(tlv_data != ifehdr_end)) {
733 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
737 skb->protocol = eth_type_trans(skb, skb->dev);
738 skb_reset_network_header(skb);
743 /*XXX: check if we can do this at install time instead of current
746 static int ife_get_sz(struct sk_buff *skb, struct tcf_ife_info *ife)
748 struct tcf_meta_info *e, *n;
749 int tot_run_sz = 0, run_sz = 0;
751 list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
752 if (e->ops->check_presence) {
753 run_sz = e->ops->check_presence(skb, e);
754 tot_run_sz += run_sz;
761 static int tcf_ife_encode(struct sk_buff *skb, const struct tc_action *a,
762 struct tcf_result *res, struct tcf_ife_params *p)
764 struct tcf_ife_info *ife = to_ife(a);
765 int action = ife->tcf_action;
766 struct ethhdr *oethh; /* outer ether header */
767 struct tcf_meta_info *e;
769 OUTERHDR:TOTMETALEN:{TLVHDR:Metadatum:TLVHDR..}:ORIGDATA
770 where ORIGDATA = original ethernet header ...
772 u16 metalen = ife_get_sz(skb, ife);
773 int hdrm = metalen + skb->dev->hard_header_len + IFE_METAHDRLEN;
774 unsigned int skboff = 0;
775 int new_len = skb->len + hdrm;
776 bool exceed_mtu = false;
780 if (!skb_at_tc_ingress(skb)) {
781 if (new_len > skb->dev->mtu)
785 bstats_cpu_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
786 tcf_lastuse_update(&ife->tcf_tm);
788 if (!metalen) { /* no metadata to send */
789 /* abuse overlimits to count when we allow packet
792 qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
795 /* could be stupid policy setup or mtu config
796 * so lets be conservative.. */
797 if ((action == TC_ACT_SHOT) || exceed_mtu) {
798 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
802 if (skb_at_tc_ingress(skb))
803 skb_push(skb, skb->dev->hard_header_len);
805 ife_meta = ife_encode(skb, metalen);
807 spin_lock(&ife->tcf_lock);
809 /* XXX: we dont have a clever way of telling encode to
810 * not repeat some of the computations that are done by
811 * ops->presence_check...
813 list_for_each_entry(e, &ife->metalist, metalist) {
814 if (e->ops->encode) {
815 err = e->ops->encode(skb, (void *)(ife_meta + skboff),
819 /* too corrupt to keep around if overwritten */
820 spin_unlock(&ife->tcf_lock);
821 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
826 spin_unlock(&ife->tcf_lock);
827 oethh = (struct ethhdr *)skb->data;
829 if (!is_zero_ether_addr(p->eth_src))
830 ether_addr_copy(oethh->h_source, p->eth_src);
831 if (!is_zero_ether_addr(p->eth_dst))
832 ether_addr_copy(oethh->h_dest, p->eth_dst);
833 oethh->h_proto = htons(p->eth_type);
835 if (skb_at_tc_ingress(skb))
836 skb_pull(skb, skb->dev->hard_header_len);
841 static int tcf_ife_act(struct sk_buff *skb, const struct tc_action *a,
842 struct tcf_result *res)
844 struct tcf_ife_info *ife = to_ife(a);
845 struct tcf_ife_params *p;
848 p = rcu_dereference_bh(ife->params);
849 if (p->flags & IFE_ENCODE) {
850 ret = tcf_ife_encode(skb, a, res, p);
854 return tcf_ife_decode(skb, a, res);
857 static int tcf_ife_walker(struct net *net, struct sk_buff *skb,
858 struct netlink_callback *cb, int type,
859 const struct tc_action_ops *ops,
860 struct netlink_ext_ack *extack)
862 struct tc_action_net *tn = net_generic(net, ife_net_id);
864 return tcf_generic_walker(tn, skb, cb, type, ops, extack);
867 static int tcf_ife_search(struct net *net, struct tc_action **a, u32 index)
869 struct tc_action_net *tn = net_generic(net, ife_net_id);
871 return tcf_idr_search(tn, a, index);
874 static struct tc_action_ops act_ife_ops = {
877 .owner = THIS_MODULE,
879 .dump = tcf_ife_dump,
880 .cleanup = tcf_ife_cleanup,
881 .init = tcf_ife_init,
882 .walk = tcf_ife_walker,
883 .lookup = tcf_ife_search,
884 .size = sizeof(struct tcf_ife_info),
887 static __net_init int ife_init_net(struct net *net)
889 struct tc_action_net *tn = net_generic(net, ife_net_id);
891 return tc_action_net_init(tn, &act_ife_ops);
894 static void __net_exit ife_exit_net(struct list_head *net_list)
896 tc_action_net_exit(net_list, ife_net_id);
899 static struct pernet_operations ife_net_ops = {
900 .init = ife_init_net,
901 .exit_batch = ife_exit_net,
903 .size = sizeof(struct tc_action_net),
906 static int __init ife_init_module(void)
908 return tcf_register_action(&act_ife_ops, &ife_net_ops);
911 static void __exit ife_cleanup_module(void)
913 tcf_unregister_action(&act_ife_ops, &ife_net_ops);
916 module_init(ife_init_module);
917 module_exit(ife_cleanup_module);
919 MODULE_AUTHOR("Jamal Hadi Salim(2015)");
920 MODULE_DESCRIPTION("Inter-FE LFB action");
921 MODULE_LICENSE("GPL");