1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * net/sched/ife.c Inter-FE action based on ForCES WG InterFE LFB
6 * draft-ietf-forces-interfelfb-03
9 * "Distributing Linux Traffic Control Classifier-Action
11 * Authors: Jamal Hadi Salim and Damascene M. Joachimpillai
13 * copyright Jamal Hadi Salim (2015)
16 #include <linux/types.h>
17 #include <linux/kernel.h>
18 #include <linux/string.h>
19 #include <linux/errno.h>
20 #include <linux/skbuff.h>
21 #include <linux/rtnetlink.h>
22 #include <linux/module.h>
23 #include <linux/init.h>
24 #include <net/net_namespace.h>
25 #include <net/netlink.h>
26 #include <net/pkt_sched.h>
27 #include <net/pkt_cls.h>
28 #include <uapi/linux/tc_act/tc_ife.h>
29 #include <net/tc_act/tc_ife.h>
30 #include <linux/etherdevice.h>
33 static unsigned int ife_net_id;
34 static int max_metacnt = IFE_META_MAX + 1;
35 static struct tc_action_ops act_ife_ops;
37 static const struct nla_policy ife_policy[TCA_IFE_MAX + 1] = {
38 [TCA_IFE_PARMS] = { .len = sizeof(struct tc_ife)},
39 [TCA_IFE_DMAC] = { .len = ETH_ALEN},
40 [TCA_IFE_SMAC] = { .len = ETH_ALEN},
41 [TCA_IFE_TYPE] = { .type = NLA_U16},
44 int ife_encode_meta_u16(u16 metaval, void *skbdata, struct tcf_meta_info *mi)
49 edata = *(u16 *)mi->metaval;
53 if (!edata) /* will not encode */
57 return ife_tlv_meta_encode(skbdata, mi->metaid, 2, &edata);
59 EXPORT_SYMBOL_GPL(ife_encode_meta_u16);
61 int ife_get_meta_u32(struct sk_buff *skb, struct tcf_meta_info *mi)
64 return nla_put_u32(skb, mi->metaid, *(u32 *)mi->metaval);
66 return nla_put(skb, mi->metaid, 0, NULL);
68 EXPORT_SYMBOL_GPL(ife_get_meta_u32);
70 int ife_check_meta_u32(u32 metaval, struct tcf_meta_info *mi)
72 if (metaval || mi->metaval)
73 return 8; /* T+L+V == 2+2+4 */
77 EXPORT_SYMBOL_GPL(ife_check_meta_u32);
79 int ife_check_meta_u16(u16 metaval, struct tcf_meta_info *mi)
81 if (metaval || mi->metaval)
82 return 8; /* T+L+(V) == 2+2+(2+2bytepad) */
86 EXPORT_SYMBOL_GPL(ife_check_meta_u16);
88 int ife_encode_meta_u32(u32 metaval, void *skbdata, struct tcf_meta_info *mi)
93 edata = *(u32 *)mi->metaval;
97 if (!edata) /* will not encode */
100 edata = htonl(edata);
101 return ife_tlv_meta_encode(skbdata, mi->metaid, 4, &edata);
103 EXPORT_SYMBOL_GPL(ife_encode_meta_u32);
105 int ife_get_meta_u16(struct sk_buff *skb, struct tcf_meta_info *mi)
108 return nla_put_u16(skb, mi->metaid, *(u16 *)mi->metaval);
110 return nla_put(skb, mi->metaid, 0, NULL);
112 EXPORT_SYMBOL_GPL(ife_get_meta_u16);
114 int ife_alloc_meta_u32(struct tcf_meta_info *mi, void *metaval, gfp_t gfp)
116 mi->metaval = kmemdup(metaval, sizeof(u32), gfp);
122 EXPORT_SYMBOL_GPL(ife_alloc_meta_u32);
124 int ife_alloc_meta_u16(struct tcf_meta_info *mi, void *metaval, gfp_t gfp)
126 mi->metaval = kmemdup(metaval, sizeof(u16), gfp);
132 EXPORT_SYMBOL_GPL(ife_alloc_meta_u16);
134 void ife_release_meta_gen(struct tcf_meta_info *mi)
138 EXPORT_SYMBOL_GPL(ife_release_meta_gen);
140 int ife_validate_meta_u32(void *val, int len)
142 if (len == sizeof(u32))
147 EXPORT_SYMBOL_GPL(ife_validate_meta_u32);
149 int ife_validate_meta_u16(void *val, int len)
151 /* length will not include padding */
152 if (len == sizeof(u16))
157 EXPORT_SYMBOL_GPL(ife_validate_meta_u16);
159 static LIST_HEAD(ifeoplist);
160 static DEFINE_RWLOCK(ife_mod_lock);
162 static struct tcf_meta_ops *find_ife_oplist(u16 metaid)
164 struct tcf_meta_ops *o;
166 read_lock(&ife_mod_lock);
167 list_for_each_entry(o, &ifeoplist, list) {
168 if (o->metaid == metaid) {
169 if (!try_module_get(o->owner))
171 read_unlock(&ife_mod_lock);
175 read_unlock(&ife_mod_lock);
180 int register_ife_op(struct tcf_meta_ops *mops)
182 struct tcf_meta_ops *m;
184 if (!mops->metaid || !mops->metatype || !mops->name ||
185 !mops->check_presence || !mops->encode || !mops->decode ||
186 !mops->get || !mops->alloc)
189 write_lock(&ife_mod_lock);
191 list_for_each_entry(m, &ifeoplist, list) {
192 if (m->metaid == mops->metaid ||
193 (strcmp(mops->name, m->name) == 0)) {
194 write_unlock(&ife_mod_lock);
200 mops->release = ife_release_meta_gen;
202 list_add_tail(&mops->list, &ifeoplist);
203 write_unlock(&ife_mod_lock);
206 EXPORT_SYMBOL_GPL(unregister_ife_op);
208 int unregister_ife_op(struct tcf_meta_ops *mops)
210 struct tcf_meta_ops *m;
213 write_lock(&ife_mod_lock);
214 list_for_each_entry(m, &ifeoplist, list) {
215 if (m->metaid == mops->metaid) {
216 list_del(&mops->list);
221 write_unlock(&ife_mod_lock);
225 EXPORT_SYMBOL_GPL(register_ife_op);
227 static int ife_validate_metatype(struct tcf_meta_ops *ops, void *val, int len)
230 /* XXX: unfortunately cant use nla_policy at this point
231 * because a length of 0 is valid in the case of
232 * "allow". "use" semantics do enforce for proper
233 * length and i couldve use nla_policy but it makes it hard
234 * to use it just for that..
237 return ops->validate(val, len);
239 if (ops->metatype == NLA_U32)
240 ret = ife_validate_meta_u32(val, len);
241 else if (ops->metatype == NLA_U16)
242 ret = ife_validate_meta_u16(val, len);
247 #ifdef CONFIG_MODULES
248 static const char *ife_meta_id2name(u32 metaid)
251 case IFE_META_SKBMARK:
255 case IFE_META_TCINDEX:
263 /* called when adding new meta information
265 static int load_metaops_and_vet(u32 metaid, void *val, int len, bool rtnl_held)
267 struct tcf_meta_ops *ops = find_ife_oplist(metaid);
272 #ifdef CONFIG_MODULES
275 request_module("ife-meta-%s", ife_meta_id2name(metaid));
278 ops = find_ife_oplist(metaid);
285 ret = ife_validate_metatype(ops, val, len);
287 module_put(ops->owner);
293 /* called when adding new meta information
295 static int __add_metainfo(const struct tcf_meta_ops *ops,
296 struct tcf_ife_info *ife, u32 metaid, void *metaval,
297 int len, bool atomic, bool exists)
299 struct tcf_meta_info *mi = NULL;
302 mi = kzalloc(sizeof(*mi), atomic ? GFP_ATOMIC : GFP_KERNEL);
309 ret = ops->alloc(mi, metaval, atomic ? GFP_ATOMIC : GFP_KERNEL);
317 spin_lock_bh(&ife->tcf_lock);
318 list_add_tail(&mi->metalist, &ife->metalist);
320 spin_unlock_bh(&ife->tcf_lock);
325 static int add_metainfo_and_get_ops(const struct tcf_meta_ops *ops,
326 struct tcf_ife_info *ife, u32 metaid,
331 if (!try_module_get(ops->owner))
333 ret = __add_metainfo(ops, ife, metaid, NULL, 0, true, exists);
335 module_put(ops->owner);
339 static int add_metainfo(struct tcf_ife_info *ife, u32 metaid, void *metaval,
340 int len, bool exists)
342 const struct tcf_meta_ops *ops = find_ife_oplist(metaid);
347 ret = __add_metainfo(ops, ife, metaid, metaval, len, false, exists);
349 /*put back what find_ife_oplist took */
350 module_put(ops->owner);
354 static int use_all_metadata(struct tcf_ife_info *ife, bool exists)
356 struct tcf_meta_ops *o;
360 read_lock(&ife_mod_lock);
361 list_for_each_entry(o, &ifeoplist, list) {
362 rc = add_metainfo_and_get_ops(o, ife, o->metaid, exists);
366 read_unlock(&ife_mod_lock);
374 static int dump_metalist(struct sk_buff *skb, struct tcf_ife_info *ife)
376 struct tcf_meta_info *e;
378 unsigned char *b = skb_tail_pointer(skb);
379 int total_encoded = 0;
381 /*can only happen on decode */
382 if (list_empty(&ife->metalist))
385 nest = nla_nest_start_noflag(skb, TCA_IFE_METALST);
389 list_for_each_entry(e, &ife->metalist, metalist) {
390 if (!e->ops->get(skb, e))
397 nla_nest_end(skb, nest);
406 /* under ife->tcf_lock */
407 static void _tcf_ife_cleanup(struct tc_action *a)
409 struct tcf_ife_info *ife = to_ife(a);
410 struct tcf_meta_info *e, *n;
412 list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
413 list_del(&e->metalist);
420 module_put(e->ops->owner);
425 static void tcf_ife_cleanup(struct tc_action *a)
427 struct tcf_ife_info *ife = to_ife(a);
428 struct tcf_ife_params *p;
430 spin_lock_bh(&ife->tcf_lock);
432 spin_unlock_bh(&ife->tcf_lock);
434 p = rcu_dereference_protected(ife->params, 1);
439 static int load_metalist(struct nlattr **tb, bool rtnl_held)
443 for (i = 1; i < max_metacnt; i++) {
445 void *val = nla_data(tb[i]);
446 int len = nla_len(tb[i]);
449 rc = load_metaops_and_vet(i, val, len, rtnl_held);
458 static int populate_metalist(struct tcf_ife_info *ife, struct nlattr **tb,
459 bool exists, bool rtnl_held)
466 for (i = 1; i < max_metacnt; i++) {
468 val = nla_data(tb[i]);
469 len = nla_len(tb[i]);
471 rc = add_metainfo(ife, i, val, len, exists);
480 static int tcf_ife_init(struct net *net, struct nlattr *nla,
481 struct nlattr *est, struct tc_action **a,
482 int ovr, int bind, bool rtnl_held,
483 struct tcf_proto *tp, struct netlink_ext_ack *extack)
485 struct tc_action_net *tn = net_generic(net, ife_net_id);
486 struct nlattr *tb[TCA_IFE_MAX + 1];
487 struct nlattr *tb2[IFE_META_MAX + 1];
488 struct tcf_chain *goto_ch = NULL;
489 struct tcf_ife_params *p;
490 struct tcf_ife_info *ife;
491 u16 ife_type = ETH_P_IFE;
501 NL_SET_ERR_MSG_MOD(extack, "IFE requires attributes to be passed");
505 err = nla_parse_nested_deprecated(tb, TCA_IFE_MAX, nla, ife_policy,
510 if (!tb[TCA_IFE_PARMS])
513 parm = nla_data(tb[TCA_IFE_PARMS]);
515 /* IFE_DECODE is 0 and indicates the opposite of IFE_ENCODE because
516 * they cannot run as the same time. Check on all other values which
517 * are not supported right now.
519 if (parm->flags & ~IFE_ENCODE)
522 p = kzalloc(sizeof(*p), GFP_KERNEL);
526 if (tb[TCA_IFE_METALST]) {
527 err = nla_parse_nested_deprecated(tb2, IFE_META_MAX,
528 tb[TCA_IFE_METALST], NULL,
534 err = load_metalist(tb2, rtnl_held);
542 err = tcf_idr_check_alloc(tn, &index, a, bind);
548 if (exists && bind) {
554 ret = tcf_idr_create(tn, index, est, a, &act_ife_ops,
557 tcf_idr_cleanup(tn, index);
563 tcf_idr_release(*a, bind);
569 if (ret == ACT_P_CREATED)
570 INIT_LIST_HEAD(&ife->metalist);
572 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
576 p->flags = parm->flags;
578 if (parm->flags & IFE_ENCODE) {
579 if (tb[TCA_IFE_TYPE])
580 ife_type = nla_get_u16(tb[TCA_IFE_TYPE]);
581 if (tb[TCA_IFE_DMAC])
582 daddr = nla_data(tb[TCA_IFE_DMAC]);
583 if (tb[TCA_IFE_SMAC])
584 saddr = nla_data(tb[TCA_IFE_SMAC]);
587 if (parm->flags & IFE_ENCODE) {
589 ether_addr_copy(p->eth_dst, daddr);
591 eth_zero_addr(p->eth_dst);
594 ether_addr_copy(p->eth_src, saddr);
596 eth_zero_addr(p->eth_src);
598 p->eth_type = ife_type;
601 if (tb[TCA_IFE_METALST]) {
602 err = populate_metalist(ife, tb2, exists, rtnl_held);
604 goto metadata_parse_err;
606 /* if no passed metadata allow list or passed allow-all
607 * then here we process by adding as many supported metadatum
608 * as we can. You better have at least one else we are
611 err = use_all_metadata(ife, exists);
613 goto metadata_parse_err;
617 spin_lock_bh(&ife->tcf_lock);
618 /* protected by tcf_lock when modifying existing action */
619 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
620 rcu_swap_protected(ife->params, p, 1);
623 spin_unlock_bh(&ife->tcf_lock);
625 tcf_chain_put_by_act(goto_ch);
632 tcf_chain_put_by_act(goto_ch);
635 tcf_idr_release(*a, bind);
639 static int tcf_ife_dump(struct sk_buff *skb, struct tc_action *a, int bind,
642 unsigned char *b = skb_tail_pointer(skb);
643 struct tcf_ife_info *ife = to_ife(a);
644 struct tcf_ife_params *p;
645 struct tc_ife opt = {
646 .index = ife->tcf_index,
647 .refcnt = refcount_read(&ife->tcf_refcnt) - ref,
648 .bindcnt = atomic_read(&ife->tcf_bindcnt) - bind,
652 spin_lock_bh(&ife->tcf_lock);
653 opt.action = ife->tcf_action;
654 p = rcu_dereference_protected(ife->params,
655 lockdep_is_held(&ife->tcf_lock));
656 opt.flags = p->flags;
658 if (nla_put(skb, TCA_IFE_PARMS, sizeof(opt), &opt))
659 goto nla_put_failure;
661 tcf_tm_dump(&t, &ife->tcf_tm);
662 if (nla_put_64bit(skb, TCA_IFE_TM, sizeof(t), &t, TCA_IFE_PAD))
663 goto nla_put_failure;
665 if (!is_zero_ether_addr(p->eth_dst)) {
666 if (nla_put(skb, TCA_IFE_DMAC, ETH_ALEN, p->eth_dst))
667 goto nla_put_failure;
670 if (!is_zero_ether_addr(p->eth_src)) {
671 if (nla_put(skb, TCA_IFE_SMAC, ETH_ALEN, p->eth_src))
672 goto nla_put_failure;
675 if (nla_put(skb, TCA_IFE_TYPE, 2, &p->eth_type))
676 goto nla_put_failure;
678 if (dump_metalist(skb, ife)) {
679 /*ignore failure to dump metalist */
680 pr_info("Failed to dump metalist\n");
683 spin_unlock_bh(&ife->tcf_lock);
687 spin_unlock_bh(&ife->tcf_lock);
692 static int find_decode_metaid(struct sk_buff *skb, struct tcf_ife_info *ife,
693 u16 metaid, u16 mlen, void *mdata)
695 struct tcf_meta_info *e;
697 /* XXX: use hash to speed up */
698 list_for_each_entry(e, &ife->metalist, metalist) {
699 if (metaid == e->metaid) {
701 /* We check for decode presence already */
702 return e->ops->decode(skb, mdata, mlen);
710 static int tcf_ife_decode(struct sk_buff *skb, const struct tc_action *a,
711 struct tcf_result *res)
713 struct tcf_ife_info *ife = to_ife(a);
714 int action = ife->tcf_action;
719 bstats_cpu_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
720 tcf_lastuse_update(&ife->tcf_tm);
722 if (skb_at_tc_ingress(skb))
723 skb_push(skb, skb->dev->hard_header_len);
725 tlv_data = ife_decode(skb, &metalen);
726 if (unlikely(!tlv_data)) {
727 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
731 ifehdr_end = tlv_data + metalen;
732 for (; tlv_data < ifehdr_end; tlv_data = ife_tlv_meta_next(tlv_data)) {
737 curr_data = ife_tlv_meta_decode(tlv_data, ifehdr_end, &mtype,
740 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
744 if (find_decode_metaid(skb, ife, mtype, dlen, curr_data)) {
745 /* abuse overlimits to count when we receive metadata
746 * but dont have an ops for it
748 pr_info_ratelimited("Unknown metaid %d dlen %d\n",
750 qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
754 if (WARN_ON(tlv_data != ifehdr_end)) {
755 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
759 skb->protocol = eth_type_trans(skb, skb->dev);
760 skb_reset_network_header(skb);
765 /*XXX: check if we can do this at install time instead of current
768 static int ife_get_sz(struct sk_buff *skb, struct tcf_ife_info *ife)
770 struct tcf_meta_info *e, *n;
771 int tot_run_sz = 0, run_sz = 0;
773 list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
774 if (e->ops->check_presence) {
775 run_sz = e->ops->check_presence(skb, e);
776 tot_run_sz += run_sz;
783 static int tcf_ife_encode(struct sk_buff *skb, const struct tc_action *a,
784 struct tcf_result *res, struct tcf_ife_params *p)
786 struct tcf_ife_info *ife = to_ife(a);
787 int action = ife->tcf_action;
788 struct ethhdr *oethh; /* outer ether header */
789 struct tcf_meta_info *e;
791 OUTERHDR:TOTMETALEN:{TLVHDR:Metadatum:TLVHDR..}:ORIGDATA
792 where ORIGDATA = original ethernet header ...
794 u16 metalen = ife_get_sz(skb, ife);
795 int hdrm = metalen + skb->dev->hard_header_len + IFE_METAHDRLEN;
796 unsigned int skboff = 0;
797 int new_len = skb->len + hdrm;
798 bool exceed_mtu = false;
802 if (!skb_at_tc_ingress(skb)) {
803 if (new_len > skb->dev->mtu)
807 bstats_cpu_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
808 tcf_lastuse_update(&ife->tcf_tm);
810 if (!metalen) { /* no metadata to send */
811 /* abuse overlimits to count when we allow packet
814 qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
817 /* could be stupid policy setup or mtu config
818 * so lets be conservative.. */
819 if ((action == TC_ACT_SHOT) || exceed_mtu) {
820 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
824 if (skb_at_tc_ingress(skb))
825 skb_push(skb, skb->dev->hard_header_len);
827 ife_meta = ife_encode(skb, metalen);
829 spin_lock(&ife->tcf_lock);
831 /* XXX: we dont have a clever way of telling encode to
832 * not repeat some of the computations that are done by
833 * ops->presence_check...
835 list_for_each_entry(e, &ife->metalist, metalist) {
836 if (e->ops->encode) {
837 err = e->ops->encode(skb, (void *)(ife_meta + skboff),
841 /* too corrupt to keep around if overwritten */
842 spin_unlock(&ife->tcf_lock);
843 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
848 spin_unlock(&ife->tcf_lock);
849 oethh = (struct ethhdr *)skb->data;
851 if (!is_zero_ether_addr(p->eth_src))
852 ether_addr_copy(oethh->h_source, p->eth_src);
853 if (!is_zero_ether_addr(p->eth_dst))
854 ether_addr_copy(oethh->h_dest, p->eth_dst);
855 oethh->h_proto = htons(p->eth_type);
857 if (skb_at_tc_ingress(skb))
858 skb_pull(skb, skb->dev->hard_header_len);
863 static int tcf_ife_act(struct sk_buff *skb, const struct tc_action *a,
864 struct tcf_result *res)
866 struct tcf_ife_info *ife = to_ife(a);
867 struct tcf_ife_params *p;
870 p = rcu_dereference_bh(ife->params);
871 if (p->flags & IFE_ENCODE) {
872 ret = tcf_ife_encode(skb, a, res, p);
876 return tcf_ife_decode(skb, a, res);
879 static int tcf_ife_walker(struct net *net, struct sk_buff *skb,
880 struct netlink_callback *cb, int type,
881 const struct tc_action_ops *ops,
882 struct netlink_ext_ack *extack)
884 struct tc_action_net *tn = net_generic(net, ife_net_id);
886 return tcf_generic_walker(tn, skb, cb, type, ops, extack);
889 static int tcf_ife_search(struct net *net, struct tc_action **a, u32 index)
891 struct tc_action_net *tn = net_generic(net, ife_net_id);
893 return tcf_idr_search(tn, a, index);
896 static struct tc_action_ops act_ife_ops = {
899 .owner = THIS_MODULE,
901 .dump = tcf_ife_dump,
902 .cleanup = tcf_ife_cleanup,
903 .init = tcf_ife_init,
904 .walk = tcf_ife_walker,
905 .lookup = tcf_ife_search,
906 .size = sizeof(struct tcf_ife_info),
909 static __net_init int ife_init_net(struct net *net)
911 struct tc_action_net *tn = net_generic(net, ife_net_id);
913 return tc_action_net_init(net, tn, &act_ife_ops);
916 static void __net_exit ife_exit_net(struct list_head *net_list)
918 tc_action_net_exit(net_list, ife_net_id);
921 static struct pernet_operations ife_net_ops = {
922 .init = ife_init_net,
923 .exit_batch = ife_exit_net,
925 .size = sizeof(struct tc_action_net),
928 static int __init ife_init_module(void)
930 return tcf_register_action(&act_ife_ops, &ife_net_ops);
933 static void __exit ife_cleanup_module(void)
935 tcf_unregister_action(&act_ife_ops, &ife_net_ops);
938 module_init(ife_init_module);
939 module_exit(ife_cleanup_module);
941 MODULE_AUTHOR("Jamal Hadi Salim(2015)");
942 MODULE_DESCRIPTION("Inter-FE LFB action");
943 MODULE_LICENSE("GPL");