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_bh(&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_bh(&ife_mod_lock);
179 read_unlock_bh(&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_bh(&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_bh(&ife_mod_lock);
204 mops->release = ife_release_meta_gen;
206 list_add_tail(&mops->list, &ifeoplist);
207 write_unlock_bh(&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_bh(&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_bh(&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
268 * under ife->tcf_lock for existing action
270 static int load_metaops_and_vet(struct tcf_ife_info *ife, u32 metaid,
271 void *val, int len, bool exists,
274 struct tcf_meta_ops *ops = find_ife_oplist(metaid);
279 #ifdef CONFIG_MODULES
281 spin_unlock_bh(&ife->tcf_lock);
284 request_module("ife-meta-%s", ife_meta_id2name(metaid));
288 spin_lock_bh(&ife->tcf_lock);
289 ops = find_ife_oplist(metaid);
296 ret = ife_validate_metatype(ops, val, len);
298 module_put(ops->owner);
304 /* called when adding new meta information
305 * under ife->tcf_lock for existing action
307 static int add_metainfo(struct tcf_ife_info *ife, u32 metaid, void *metaval,
308 int len, bool atomic)
310 struct tcf_meta_info *mi = NULL;
311 struct tcf_meta_ops *ops = find_ife_oplist(metaid);
317 mi = kzalloc(sizeof(*mi), atomic ? GFP_ATOMIC : GFP_KERNEL);
319 /*put back what find_ife_oplist took */
320 module_put(ops->owner);
327 ret = ops->alloc(mi, metaval, atomic ? GFP_ATOMIC : GFP_KERNEL);
330 module_put(ops->owner);
335 list_add_tail(&mi->metalist, &ife->metalist);
340 static int use_all_metadata(struct tcf_ife_info *ife)
342 struct tcf_meta_ops *o;
346 read_lock_bh(&ife_mod_lock);
347 list_for_each_entry(o, &ifeoplist, list) {
348 rc = add_metainfo(ife, o->metaid, NULL, 0, true);
352 read_unlock_bh(&ife_mod_lock);
360 static int dump_metalist(struct sk_buff *skb, struct tcf_ife_info *ife)
362 struct tcf_meta_info *e;
364 unsigned char *b = skb_tail_pointer(skb);
365 int total_encoded = 0;
367 /*can only happen on decode */
368 if (list_empty(&ife->metalist))
371 nest = nla_nest_start(skb, TCA_IFE_METALST);
375 list_for_each_entry(e, &ife->metalist, metalist) {
376 if (!e->ops->get(skb, e))
383 nla_nest_end(skb, nest);
392 /* under ife->tcf_lock */
393 static void _tcf_ife_cleanup(struct tc_action *a)
395 struct tcf_ife_info *ife = to_ife(a);
396 struct tcf_meta_info *e, *n;
398 list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
399 module_put(e->ops->owner);
400 list_del(&e->metalist);
411 static void tcf_ife_cleanup(struct tc_action *a)
413 struct tcf_ife_info *ife = to_ife(a);
414 struct tcf_ife_params *p;
416 spin_lock_bh(&ife->tcf_lock);
418 spin_unlock_bh(&ife->tcf_lock);
420 p = rcu_dereference_protected(ife->params, 1);
425 /* under ife->tcf_lock for existing action */
426 static int populate_metalist(struct tcf_ife_info *ife, struct nlattr **tb,
427 bool exists, bool rtnl_held)
434 for (i = 1; i < max_metacnt; i++) {
436 val = nla_data(tb[i]);
437 len = nla_len(tb[i]);
439 rc = load_metaops_and_vet(ife, i, val, len, exists,
444 rc = add_metainfo(ife, i, val, len, exists);
453 static int tcf_ife_init(struct net *net, struct nlattr *nla,
454 struct nlattr *est, struct tc_action **a,
455 int ovr, int bind, bool rtnl_held,
456 struct netlink_ext_ack *extack)
458 struct tc_action_net *tn = net_generic(net, ife_net_id);
459 struct nlattr *tb[TCA_IFE_MAX + 1];
460 struct nlattr *tb2[IFE_META_MAX + 1];
461 struct tcf_ife_params *p;
462 struct tcf_ife_info *ife;
463 u16 ife_type = ETH_P_IFE;
471 err = nla_parse_nested(tb, TCA_IFE_MAX, nla, ife_policy, NULL);
475 if (!tb[TCA_IFE_PARMS])
478 parm = nla_data(tb[TCA_IFE_PARMS]);
480 /* IFE_DECODE is 0 and indicates the opposite of IFE_ENCODE because
481 * they cannot run as the same time. Check on all other values which
482 * are not supported right now.
484 if (parm->flags & ~IFE_ENCODE)
487 p = kzalloc(sizeof(*p), GFP_KERNEL);
491 err = tcf_idr_check_alloc(tn, &parm->index, a, bind);
497 if (exists && bind) {
503 ret = tcf_idr_create(tn, parm->index, est, a, &act_ife_ops,
506 tcf_idr_cleanup(tn, parm->index);
512 tcf_idr_release(*a, bind);
518 p->flags = parm->flags;
520 if (parm->flags & IFE_ENCODE) {
521 if (tb[TCA_IFE_TYPE])
522 ife_type = nla_get_u16(tb[TCA_IFE_TYPE]);
523 if (tb[TCA_IFE_DMAC])
524 daddr = nla_data(tb[TCA_IFE_DMAC]);
525 if (tb[TCA_IFE_SMAC])
526 saddr = nla_data(tb[TCA_IFE_SMAC]);
529 if (parm->flags & IFE_ENCODE) {
531 ether_addr_copy(p->eth_dst, daddr);
533 eth_zero_addr(p->eth_dst);
536 ether_addr_copy(p->eth_src, saddr);
538 eth_zero_addr(p->eth_src);
540 p->eth_type = ife_type;
544 spin_lock_bh(&ife->tcf_lock);
546 if (ret == ACT_P_CREATED)
547 INIT_LIST_HEAD(&ife->metalist);
549 if (tb[TCA_IFE_METALST]) {
550 err = nla_parse_nested(tb2, IFE_META_MAX, tb[TCA_IFE_METALST],
555 spin_unlock_bh(&ife->tcf_lock);
556 tcf_idr_release(*a, bind);
562 err = populate_metalist(ife, tb2, exists, rtnl_held);
564 goto metadata_parse_err;
567 /* if no passed metadata allow list or passed allow-all
568 * then here we process by adding as many supported metadatum
569 * as we can. You better have at least one else we are
572 err = use_all_metadata(ife);
575 spin_unlock_bh(&ife->tcf_lock);
576 tcf_idr_release(*a, bind);
583 ife->tcf_action = parm->action;
584 /* protected by tcf_lock when modifying existing action */
585 rcu_swap_protected(ife->params, p, 1);
588 spin_unlock_bh(&ife->tcf_lock);
592 if (ret == ACT_P_CREATED)
593 tcf_idr_insert(tn, *a);
598 static int tcf_ife_dump(struct sk_buff *skb, struct tc_action *a, int bind,
601 unsigned char *b = skb_tail_pointer(skb);
602 struct tcf_ife_info *ife = to_ife(a);
603 struct tcf_ife_params *p;
604 struct tc_ife opt = {
605 .index = ife->tcf_index,
606 .refcnt = refcount_read(&ife->tcf_refcnt) - ref,
607 .bindcnt = atomic_read(&ife->tcf_bindcnt) - bind,
611 spin_lock_bh(&ife->tcf_lock);
612 opt.action = ife->tcf_action;
613 p = rcu_dereference_protected(ife->params,
614 lockdep_is_held(&ife->tcf_lock));
615 opt.flags = p->flags;
617 if (nla_put(skb, TCA_IFE_PARMS, sizeof(opt), &opt))
618 goto nla_put_failure;
620 tcf_tm_dump(&t, &ife->tcf_tm);
621 if (nla_put_64bit(skb, TCA_IFE_TM, sizeof(t), &t, TCA_IFE_PAD))
622 goto nla_put_failure;
624 if (!is_zero_ether_addr(p->eth_dst)) {
625 if (nla_put(skb, TCA_IFE_DMAC, ETH_ALEN, p->eth_dst))
626 goto nla_put_failure;
629 if (!is_zero_ether_addr(p->eth_src)) {
630 if (nla_put(skb, TCA_IFE_SMAC, ETH_ALEN, p->eth_src))
631 goto nla_put_failure;
634 if (nla_put(skb, TCA_IFE_TYPE, 2, &p->eth_type))
635 goto nla_put_failure;
637 if (dump_metalist(skb, ife)) {
638 /*ignore failure to dump metalist */
639 pr_info("Failed to dump metalist\n");
642 spin_unlock_bh(&ife->tcf_lock);
646 spin_unlock_bh(&ife->tcf_lock);
651 static int find_decode_metaid(struct sk_buff *skb, struct tcf_ife_info *ife,
652 u16 metaid, u16 mlen, void *mdata)
654 struct tcf_meta_info *e;
656 /* XXX: use hash to speed up */
657 list_for_each_entry(e, &ife->metalist, metalist) {
658 if (metaid == e->metaid) {
660 /* We check for decode presence already */
661 return e->ops->decode(skb, mdata, mlen);
669 static int tcf_ife_decode(struct sk_buff *skb, const struct tc_action *a,
670 struct tcf_result *res)
672 struct tcf_ife_info *ife = to_ife(a);
673 int action = ife->tcf_action;
678 bstats_cpu_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
679 tcf_lastuse_update(&ife->tcf_tm);
681 if (skb_at_tc_ingress(skb))
682 skb_push(skb, skb->dev->hard_header_len);
684 tlv_data = ife_decode(skb, &metalen);
685 if (unlikely(!tlv_data)) {
686 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
690 ifehdr_end = tlv_data + metalen;
691 for (; tlv_data < ifehdr_end; tlv_data = ife_tlv_meta_next(tlv_data)) {
696 curr_data = ife_tlv_meta_decode(tlv_data, ifehdr_end, &mtype,
699 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
703 if (find_decode_metaid(skb, ife, mtype, dlen, curr_data)) {
704 /* abuse overlimits to count when we receive metadata
705 * but dont have an ops for it
707 pr_info_ratelimited("Unknown metaid %d dlen %d\n",
709 qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
713 if (WARN_ON(tlv_data != ifehdr_end)) {
714 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
718 skb->protocol = eth_type_trans(skb, skb->dev);
719 skb_reset_network_header(skb);
724 /*XXX: check if we can do this at install time instead of current
727 static int ife_get_sz(struct sk_buff *skb, struct tcf_ife_info *ife)
729 struct tcf_meta_info *e, *n;
730 int tot_run_sz = 0, run_sz = 0;
732 list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
733 if (e->ops->check_presence) {
734 run_sz = e->ops->check_presence(skb, e);
735 tot_run_sz += run_sz;
742 static int tcf_ife_encode(struct sk_buff *skb, const struct tc_action *a,
743 struct tcf_result *res, struct tcf_ife_params *p)
745 struct tcf_ife_info *ife = to_ife(a);
746 int action = ife->tcf_action;
747 struct ethhdr *oethh; /* outer ether header */
748 struct tcf_meta_info *e;
750 OUTERHDR:TOTMETALEN:{TLVHDR:Metadatum:TLVHDR..}:ORIGDATA
751 where ORIGDATA = original ethernet header ...
753 u16 metalen = ife_get_sz(skb, ife);
754 int hdrm = metalen + skb->dev->hard_header_len + IFE_METAHDRLEN;
755 unsigned int skboff = 0;
756 int new_len = skb->len + hdrm;
757 bool exceed_mtu = false;
761 if (!skb_at_tc_ingress(skb)) {
762 if (new_len > skb->dev->mtu)
766 bstats_cpu_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
767 tcf_lastuse_update(&ife->tcf_tm);
769 if (!metalen) { /* no metadata to send */
770 /* abuse overlimits to count when we allow packet
773 qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
776 /* could be stupid policy setup or mtu config
777 * so lets be conservative.. */
778 if ((action == TC_ACT_SHOT) || exceed_mtu) {
779 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
783 if (skb_at_tc_ingress(skb))
784 skb_push(skb, skb->dev->hard_header_len);
786 ife_meta = ife_encode(skb, metalen);
788 spin_lock(&ife->tcf_lock);
790 /* XXX: we dont have a clever way of telling encode to
791 * not repeat some of the computations that are done by
792 * ops->presence_check...
794 list_for_each_entry(e, &ife->metalist, metalist) {
795 if (e->ops->encode) {
796 err = e->ops->encode(skb, (void *)(ife_meta + skboff),
800 /* too corrupt to keep around if overwritten */
801 spin_unlock(&ife->tcf_lock);
802 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
807 spin_unlock(&ife->tcf_lock);
808 oethh = (struct ethhdr *)skb->data;
810 if (!is_zero_ether_addr(p->eth_src))
811 ether_addr_copy(oethh->h_source, p->eth_src);
812 if (!is_zero_ether_addr(p->eth_dst))
813 ether_addr_copy(oethh->h_dest, p->eth_dst);
814 oethh->h_proto = htons(p->eth_type);
816 if (skb_at_tc_ingress(skb))
817 skb_pull(skb, skb->dev->hard_header_len);
822 static int tcf_ife_act(struct sk_buff *skb, const struct tc_action *a,
823 struct tcf_result *res)
825 struct tcf_ife_info *ife = to_ife(a);
826 struct tcf_ife_params *p;
829 p = rcu_dereference_bh(ife->params);
830 if (p->flags & IFE_ENCODE) {
831 ret = tcf_ife_encode(skb, a, res, p);
835 return tcf_ife_decode(skb, a, res);
838 static int tcf_ife_walker(struct net *net, struct sk_buff *skb,
839 struct netlink_callback *cb, int type,
840 const struct tc_action_ops *ops,
841 struct netlink_ext_ack *extack)
843 struct tc_action_net *tn = net_generic(net, ife_net_id);
845 return tcf_generic_walker(tn, skb, cb, type, ops, extack);
848 static int tcf_ife_search(struct net *net, struct tc_action **a, u32 index,
849 struct netlink_ext_ack *extack)
851 struct tc_action_net *tn = net_generic(net, ife_net_id);
853 return tcf_idr_search(tn, a, index);
856 static int tcf_ife_delete(struct net *net, u32 index)
858 struct tc_action_net *tn = net_generic(net, ife_net_id);
860 return tcf_idr_delete_index(tn, index);
863 static struct tc_action_ops act_ife_ops = {
866 .owner = THIS_MODULE,
868 .dump = tcf_ife_dump,
869 .cleanup = tcf_ife_cleanup,
870 .init = tcf_ife_init,
871 .walk = tcf_ife_walker,
872 .lookup = tcf_ife_search,
873 .delete = tcf_ife_delete,
874 .size = sizeof(struct tcf_ife_info),
877 static __net_init int ife_init_net(struct net *net)
879 struct tc_action_net *tn = net_generic(net, ife_net_id);
881 return tc_action_net_init(tn, &act_ife_ops);
884 static void __net_exit ife_exit_net(struct list_head *net_list)
886 tc_action_net_exit(net_list, ife_net_id);
889 static struct pernet_operations ife_net_ops = {
890 .init = ife_init_net,
891 .exit_batch = ife_exit_net,
893 .size = sizeof(struct tc_action_net),
896 static int __init ife_init_module(void)
898 return tcf_register_action(&act_ife_ops, &ife_net_ops);
901 static void __exit ife_cleanup_module(void)
903 tcf_unregister_action(&act_ife_ops, &ife_net_ops);
906 module_init(ife_init_module);
907 module_exit(ife_cleanup_module);
909 MODULE_AUTHOR("Jamal Hadi Salim(2015)");
910 MODULE_DESCRIPTION("Inter-FE LFB action");
911 MODULE_LICENSE("GPL");