act_ife.c

  1// SPDX-License-Identifier: GPL-2.0-or-later
  2/*
  3 * net/sched/ife.c	Inter-FE action based on ForCES WG InterFE LFB
  4 *
  5 *		Refer to:
  6 *		draft-ietf-forces-interfelfb-03
  7 *		and
  8 *		netdev01 paper:
  9 *		"Distributing Linux Traffic Control Classifier-Action
 10 *		Subsystem"
 11 *		Authors: Jamal Hadi Salim and Damascene M. Joachimpillai
 12 *
 13 * copyright Jamal Hadi Salim (2015)
 14*/
 15
 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>
 31#include <net/ife.h>
 32#include <net/tc_wrapper.h>
 33
 34static int max_metacnt = IFE_META_MAX + 1;
 35static struct tc_action_ops act_ife_ops;
 36
 37static 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},
 42};
 43
 44int ife_encode_meta_u16(u16 metaval, void *skbdata, struct tcf_meta_info *mi)
 45{
 46	u16 edata = 0;
 47
 48	if (mi->metaval)
 49		edata = *(u16 *)mi->metaval;
 50	else if (metaval)
 51		edata = metaval;
 52
 53	if (!edata) /* will not encode */
 54		return 0;
 55
 56	edata = htons(edata);
 57	return ife_tlv_meta_encode(skbdata, mi->metaid, 2, &edata);
 58}
 59EXPORT_SYMBOL_GPL(ife_encode_meta_u16);
 60
 61int ife_get_meta_u32(struct sk_buff *skb, struct tcf_meta_info *mi)
 62{
 63	if (mi->metaval)
 64		return nla_put_u32(skb, mi->metaid, *(u32 *)mi->metaval);
 65	else
 66		return nla_put(skb, mi->metaid, 0, NULL);
 67}
 68EXPORT_SYMBOL_GPL(ife_get_meta_u32);
 69
 70int ife_check_meta_u32(u32 metaval, struct tcf_meta_info *mi)
 71{
 72	if (metaval || mi->metaval)
 73		return 8; /* T+L+V == 2+2+4 */
 74
 75	return 0;
 76}
 77EXPORT_SYMBOL_GPL(ife_check_meta_u32);
 78
 79int ife_check_meta_u16(u16 metaval, struct tcf_meta_info *mi)
 80{
 81	if (metaval || mi->metaval)
 82		return 8; /* T+L+(V) == 2+2+(2+2bytepad) */
 83
 84	return 0;
 85}
 86EXPORT_SYMBOL_GPL(ife_check_meta_u16);
 87
 88int ife_encode_meta_u32(u32 metaval, void *skbdata, struct tcf_meta_info *mi)
 89{
 90	u32 edata = metaval;
 91
 92	if (mi->metaval)
 93		edata = *(u32 *)mi->metaval;
 94	else if (metaval)
 95		edata = metaval;
 96
 97	if (!edata) /* will not encode */
 98		return 0;
 99
100	edata = htonl(edata);
101	return ife_tlv_meta_encode(skbdata, mi->metaid, 4, &edata);
102}
103EXPORT_SYMBOL_GPL(ife_encode_meta_u32);
104
105int ife_get_meta_u16(struct sk_buff *skb, struct tcf_meta_info *mi)
106{
107	if (mi->metaval)
108		return nla_put_u16(skb, mi->metaid, *(u16 *)mi->metaval);
109	else
110		return nla_put(skb, mi->metaid, 0, NULL);
111}
112EXPORT_SYMBOL_GPL(ife_get_meta_u16);
113
114int ife_alloc_meta_u32(struct tcf_meta_info *mi, void *metaval, gfp_t gfp)
115{
116	mi->metaval = kmemdup(metaval, sizeof(u32), gfp);
117	if (!mi->metaval)
118		return -ENOMEM;
119
120	return 0;
121}
122EXPORT_SYMBOL_GPL(ife_alloc_meta_u32);
123
124int ife_alloc_meta_u16(struct tcf_meta_info *mi, void *metaval, gfp_t gfp)
125{
126	mi->metaval = kmemdup(metaval, sizeof(u16), gfp);
127	if (!mi->metaval)
128		return -ENOMEM;
129
130	return 0;
131}
132EXPORT_SYMBOL_GPL(ife_alloc_meta_u16);
133
134void ife_release_meta_gen(struct tcf_meta_info *mi)
135{
136	kfree(mi->metaval);
137}
138EXPORT_SYMBOL_GPL(ife_release_meta_gen);
139
140int ife_validate_meta_u32(void *val, int len)
141{
142	if (len == sizeof(u32))
143		return 0;
144
145	return -EINVAL;
146}
147EXPORT_SYMBOL_GPL(ife_validate_meta_u32);
148
149int ife_validate_meta_u16(void *val, int len)
150{
151	/* length will not include padding */
152	if (len == sizeof(u16))
153		return 0;
154
155	return -EINVAL;
156}
157EXPORT_SYMBOL_GPL(ife_validate_meta_u16);
158
159static LIST_HEAD(ifeoplist);
160static DEFINE_RWLOCK(ife_mod_lock);
161
162static struct tcf_meta_ops *find_ife_oplist(u16 metaid)
163{
164	struct tcf_meta_ops *o;
165
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))
170				o = NULL;
171			read_unlock(&ife_mod_lock);
172			return o;
173		}
174	}
175	read_unlock(&ife_mod_lock);
176
177	return NULL;
178}
179
180int register_ife_op(struct tcf_meta_ops *mops)
181{
182	struct tcf_meta_ops *m;
183
184	if (!mops->metaid || !mops->metatype || !mops->name ||
185	    !mops->check_presence || !mops->encode || !mops->decode ||
186	    !mops->get || !mops->alloc)
187		return -EINVAL;
188
189	write_lock(&ife_mod_lock);
190
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);
195			return -EEXIST;
196		}
197	}
198
199	if (!mops->release)
200		mops->release = ife_release_meta_gen;
201
202	list_add_tail(&mops->list, &ifeoplist);
203	write_unlock(&ife_mod_lock);
204	return 0;
205}
206EXPORT_SYMBOL_GPL(unregister_ife_op);
207
208int unregister_ife_op(struct tcf_meta_ops *mops)
209{
210	struct tcf_meta_ops *m;
211	int err = -ENOENT;
212
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);
217			err = 0;
218			break;
219		}
220	}
221	write_unlock(&ife_mod_lock);
222
223	return err;
224}
225EXPORT_SYMBOL_GPL(register_ife_op);
226
227static int ife_validate_metatype(struct tcf_meta_ops *ops, void *val, int len)
228{
229	int ret = 0;
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..
235	*/
236	if (ops->validate)
237		return ops->validate(val, len);
238
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);
243
244	return ret;
245}
246
247#ifdef CONFIG_MODULES
248static const char *ife_meta_id2name(u32 metaid)
249{
250	switch (metaid) {
251	case IFE_META_SKBMARK:
252		return "skbmark";
253	case IFE_META_PRIO:
254		return "skbprio";
255	case IFE_META_TCINDEX:
256		return "tcindex";
257	default:
258		return "unknown";
259	}
260}
261#endif
262
263/* called when adding new meta information
264*/
265static int load_metaops_and_vet(u32 metaid, void *val, int len, bool rtnl_held)
266{
267	struct tcf_meta_ops *ops = find_ife_oplist(metaid);
268	int ret = 0;
269
270	if (!ops) {
271		ret = -ENOENT;
272#ifdef CONFIG_MODULES
273		if (rtnl_held)
274			rtnl_unlock();
275		request_module("ife-meta-%s", ife_meta_id2name(metaid));
276		if (rtnl_held)
277			rtnl_lock();
278		ops = find_ife_oplist(metaid);
279#endif
280	}
281
282	if (ops) {
283		ret = 0;
284		if (len)
285			ret = ife_validate_metatype(ops, val, len);
286
287		module_put(ops->owner);
288	}
289
290	return ret;
291}
292
293/* called when adding new meta information
294*/
295static int __add_metainfo(const struct tcf_meta_ops *ops,
296			  struct tcf_ife_params *p, u32 metaid, void *metaval,
297			  int len, bool atomic)
298{
299	struct tcf_meta_info *mi = NULL;
300	int ret = 0;
301
302	mi = kzalloc(sizeof(*mi), atomic ? GFP_ATOMIC : GFP_KERNEL);
303	if (!mi)
304		return -ENOMEM;
305
306	mi->metaid = metaid;
307	mi->ops = ops;
308	if (len > 0) {
309		ret = ops->alloc(mi, metaval, atomic ? GFP_ATOMIC : GFP_KERNEL);
310		if (ret != 0) {
311			kfree(mi);
312			return ret;
313		}
314	}
315
316	list_add_tail(&mi->metalist, &p->metalist);
317
318	return ret;
319}
320
321static int add_metainfo_and_get_ops(const struct tcf_meta_ops *ops,
322				    struct tcf_ife_params *p, u32 metaid)
323{
324	int ret;
325
326	if (!try_module_get(ops->owner))
327		return -ENOENT;
328	ret = __add_metainfo(ops, p, metaid, NULL, 0, true);
329	if (ret)
330		module_put(ops->owner);
331	return ret;
332}
333
334static int add_metainfo(struct tcf_ife_params *p, u32 metaid, void *metaval,
335			int len)
336{
337	const struct tcf_meta_ops *ops = find_ife_oplist(metaid);
338	int ret;
339
340	if (!ops)
341		return -ENOENT;
342	ret = __add_metainfo(ops, p, metaid, metaval, len, false);
343	if (ret)
344		/*put back what find_ife_oplist took */
345		module_put(ops->owner);
346	return ret;
347}
348
349static int use_all_metadata(struct tcf_ife_params *p)
350{
351	struct tcf_meta_ops *o;
352	int rc = 0;
353	int installed = 0;
354
355	read_lock(&ife_mod_lock);
356	list_for_each_entry(o, &ifeoplist, list) {
357		rc = add_metainfo_and_get_ops(o, p, o->metaid);
358		if (rc == 0)
359			installed += 1;
360	}
361	read_unlock(&ife_mod_lock);
362
363	if (installed)
364		return 0;
365	else
366		return -EINVAL;
367}
368
369static int dump_metalist(struct sk_buff *skb, struct tcf_ife_params *p)
370{
371	struct tcf_meta_info *e;
372	struct nlattr *nest;
373	unsigned char *b = skb_tail_pointer(skb);
374	int total_encoded = 0;
375
376	/*can only happen on decode */
377	if (list_empty(&p->metalist))
378		return 0;
379
380	nest = nla_nest_start_noflag(skb, TCA_IFE_METALST);
381	if (!nest)
382		goto out_nlmsg_trim;
383
384	list_for_each_entry(e, &p->metalist, metalist) {
385		if (!e->ops->get(skb, e))
386			total_encoded += 1;
387	}
388
389	if (!total_encoded)
390		goto out_nlmsg_trim;
391
392	nla_nest_end(skb, nest);
393
394	return 0;
395
396out_nlmsg_trim:
397	nlmsg_trim(skb, b);
398	return -1;
399}
400
401static void __tcf_ife_cleanup(struct tcf_ife_params *p)
402{
403	struct tcf_meta_info *e, *n;
404
405	list_for_each_entry_safe(e, n, &p->metalist, metalist) {
406		list_del(&e->metalist);
407		if (e->metaval) {
408			if (e->ops->release)
409				e->ops->release(e);
410			else
411				kfree(e->metaval);
412		}
413		module_put(e->ops->owner);
414		kfree(e);
415	}
416}
417
418static void tcf_ife_cleanup_params(struct rcu_head *head)
419{
420	struct tcf_ife_params *p = container_of(head, struct tcf_ife_params,
421						rcu);
422
423	__tcf_ife_cleanup(p);
424	kfree(p);
425}
426
427static void tcf_ife_cleanup(struct tc_action *a)
428{
429	struct tcf_ife_info *ife = to_ife(a);
430	struct tcf_ife_params *p;
431
432	p = rcu_dereference_protected(ife->params, 1);
433	if (p)
434		call_rcu(&p->rcu, tcf_ife_cleanup_params);
435}
436
437static int load_metalist(struct nlattr **tb, bool rtnl_held)
438{
439	int i;
440
441	for (i = 1; i < max_metacnt; i++) {
442		if (tb[i]) {
443			void *val = nla_data(tb[i]);
444			int len = nla_len(tb[i]);
445			int rc;
446
447			rc = load_metaops_and_vet(i, val, len, rtnl_held);
448			if (rc != 0)
449				return rc;
450		}
451	}
452
453	return 0;
454}
455
456static int populate_metalist(struct tcf_ife_params *p, struct nlattr **tb)
457{
458	int len = 0;
459	int rc = 0;
460	int i = 0;
461	void *val;
462
463	for (i = 1; i < max_metacnt; i++) {
464		if (tb[i]) {
465			val = nla_data(tb[i]);
466			len = nla_len(tb[i]);
467
468			rc = add_metainfo(p, i, val, len);
469			if (rc)
470				return rc;
471		}
472	}
473
474	return rc;
475}
476
477static int tcf_ife_init(struct net *net, struct nlattr *nla,
478			struct nlattr *est, struct tc_action **a,
479			struct tcf_proto *tp, u32 flags,
480			struct netlink_ext_ack *extack)
481{
482	struct tc_action_net *tn = net_generic(net, act_ife_ops.net_id);
483	bool bind = flags & TCA_ACT_FLAGS_BIND;
484	struct nlattr *tb[TCA_IFE_MAX + 1];
485	struct nlattr *tb2[IFE_META_MAX + 1];
486	struct tcf_chain *goto_ch = NULL;
487	struct tcf_ife_params *p;
488	struct tcf_ife_info *ife;
489	u16 ife_type = ETH_P_IFE;
490	struct tc_ife *parm;
491	u8 *daddr = NULL;
492	u8 *saddr = NULL;
493	bool exists = false;
494	int ret = 0;
495	u32 index;
496	int err;
497
498	if (!nla) {
499		NL_SET_ERR_MSG_MOD(extack, "IFE requires attributes to be passed");
500		return -EINVAL;
501	}
502
503	err = nla_parse_nested_deprecated(tb, TCA_IFE_MAX, nla, ife_policy,
504					  NULL);
505	if (err < 0)
506		return err;
507
508	if (!tb[TCA_IFE_PARMS])
509		return -EINVAL;
510
511	parm = nla_data(tb[TCA_IFE_PARMS]);
512
513	/* IFE_DECODE is 0 and indicates the opposite of IFE_ENCODE because
514	 * they cannot run as the same time. Check on all other values which
515	 * are not supported right now.
516	 */
517	if (parm->flags & ~IFE_ENCODE)
518		return -EINVAL;
519
520	p = kzalloc(sizeof(*p), GFP_KERNEL);
521	if (!p)
522		return -ENOMEM;
523	INIT_LIST_HEAD(&p->metalist);
524
525	if (tb[TCA_IFE_METALST]) {
526		err = nla_parse_nested_deprecated(tb2, IFE_META_MAX,
527						  tb[TCA_IFE_METALST], NULL,
528						  NULL);
529		if (err) {
530			kfree(p);
531			return err;
532		}
533		err = load_metalist(tb2, !(flags & TCA_ACT_FLAGS_NO_RTNL));
534		if (err) {
535			kfree(p);
536			return err;
537		}
538	}
539
540	index = parm->index;
541	err = tcf_idr_check_alloc(tn, &index, a, bind);
542	if (err < 0) {
543		kfree(p);
544		return err;
545	}
546	exists = err;
547	if (exists && bind) {
548		kfree(p);
549		return ACT_P_BOUND;
550	}
551
552	if (!exists) {
553		ret = tcf_idr_create(tn, index, est, a, &act_ife_ops,
554				     bind, true, flags);
555		if (ret) {
556			tcf_idr_cleanup(tn, index);
557			kfree(p);
558			return ret;
559		}
560		ret = ACT_P_CREATED;
561	} else if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
562		tcf_idr_release(*a, bind);
563		kfree(p);
564		return -EEXIST;
565	}
566
567	ife = to_ife(*a);
568
569	err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
570	if (err < 0)
571		goto release_idr;
572
573	p->flags = parm->flags;
574
575	if (parm->flags & IFE_ENCODE) {
576		if (tb[TCA_IFE_TYPE])
577			ife_type = nla_get_u16(tb[TCA_IFE_TYPE]);
578		if (tb[TCA_IFE_DMAC])
579			daddr = nla_data(tb[TCA_IFE_DMAC]);
580		if (tb[TCA_IFE_SMAC])
581			saddr = nla_data(tb[TCA_IFE_SMAC]);
582	}
583
584	if (parm->flags & IFE_ENCODE) {
585		if (daddr)
586			ether_addr_copy(p->eth_dst, daddr);
587		else
588			eth_zero_addr(p->eth_dst);
589
590		if (saddr)
591			ether_addr_copy(p->eth_src, saddr);
592		else
593			eth_zero_addr(p->eth_src);
594
595		p->eth_type = ife_type;
596	}
597
598	if (tb[TCA_IFE_METALST]) {
599		err = populate_metalist(p, tb2);
600		if (err)
601			goto metadata_parse_err;
602	} else {
603		/* if no passed metadata allow list or passed allow-all
604		 * then here we process by adding as many supported metadatum
605		 * as we can. You better have at least one else we are
606		 * going to bail out
607		 */
608		err = use_all_metadata(p);
609		if (err)
610			goto metadata_parse_err;
611	}
612
613	if (exists)
614		spin_lock_bh(&ife->tcf_lock);
615	/* protected by tcf_lock when modifying existing action */
616	goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
617	p = rcu_replace_pointer(ife->params, p, 1);
618
619	if (exists)
620		spin_unlock_bh(&ife->tcf_lock);
621	if (goto_ch)
622		tcf_chain_put_by_act(goto_ch);
623	if (p)
624		call_rcu(&p->rcu, tcf_ife_cleanup_params);
625
626	return ret;
627metadata_parse_err:
628	if (goto_ch)
629		tcf_chain_put_by_act(goto_ch);
630release_idr:
631	__tcf_ife_cleanup(p);
632	kfree(p);
633	tcf_idr_release(*a, bind);
634	return err;
635}
636
637static int tcf_ife_dump(struct sk_buff *skb, struct tc_action *a, int bind,
638			int ref)
639{
640	unsigned char *b = skb_tail_pointer(skb);
641	struct tcf_ife_info *ife = to_ife(a);
642	struct tcf_ife_params *p;
643	struct tc_ife opt;
644	struct tcf_t t;
645
646	memset(&opt, 0, sizeof(opt));
647
648	opt.index = ife->tcf_index;
649	opt.refcnt = refcount_read(&ife->tcf_refcnt) - ref;
650	opt.bindcnt = atomic_read(&ife->tcf_bindcnt) - bind;
651
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;
657
658	if (nla_put(skb, TCA_IFE_PARMS, sizeof(opt), &opt))
659		goto nla_put_failure;
660
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;
664
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;
668	}
669
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;
673	}
674
675	if (nla_put(skb, TCA_IFE_TYPE, 2, &p->eth_type))
676		goto nla_put_failure;
677
678	if (dump_metalist(skb, p)) {
679		/*ignore failure to dump metalist */
680		pr_info("Failed to dump metalist\n");
681	}
682
683	spin_unlock_bh(&ife->tcf_lock);
684	return skb->len;
685
686nla_put_failure:
687	spin_unlock_bh(&ife->tcf_lock);
688	nlmsg_trim(skb, b);
689	return -1;
690}
691
692static int find_decode_metaid(struct sk_buff *skb, struct tcf_ife_params *p,
693			      u16 metaid, u16 mlen, void *mdata)
694{
695	struct tcf_meta_info *e;
696
697	/* XXX: use hash to speed up */
698	list_for_each_entry_rcu(e, &p->metalist, metalist) {
699		if (metaid == e->metaid) {
700			if (e->ops) {
701				/* We check for decode presence already */
702				return e->ops->decode(skb, mdata, mlen);
703			}
704		}
705	}
706
707	return -ENOENT;
708}
709
710static int tcf_ife_decode(struct sk_buff *skb, const struct tc_action *a,
711			  struct tcf_result *res)
712{
713	struct tcf_ife_info *ife = to_ife(a);
714	int action = ife->tcf_action;
715	struct tcf_ife_params *p;
716	u8 *ifehdr_end;
717	u8 *tlv_data;
718	u16 metalen;
719
720	p = rcu_dereference_bh(ife->params);
721
722	bstats_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
723	tcf_lastuse_update(&ife->tcf_tm);
724
725	if (skb_at_tc_ingress(skb))
726		skb_push(skb, skb->dev->hard_header_len);
727
728	tlv_data = ife_decode(skb, &metalen);
729	if (unlikely(!tlv_data)) {
730		qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
731		return TC_ACT_SHOT;
732	}
733
734	ifehdr_end = tlv_data + metalen;
735	for (; tlv_data < ifehdr_end; tlv_data = ife_tlv_meta_next(tlv_data)) {
736		u8 *curr_data;
737		u16 mtype;
738		u16 dlen;
739
740		curr_data = ife_tlv_meta_decode(tlv_data, ifehdr_end, &mtype,
741						&dlen, NULL);
742		if (!curr_data) {
743			qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
744			return TC_ACT_SHOT;
745		}
746
747		if (find_decode_metaid(skb, p, mtype, dlen, curr_data)) {
748			/* abuse overlimits to count when we receive metadata
749			 * but dont have an ops for it
750			 */
751			pr_info_ratelimited("Unknown metaid %d dlen %d\n",
752					    mtype, dlen);
753			qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
754		}
755	}
756
757	if (WARN_ON(tlv_data != ifehdr_end)) {
758		qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
759		return TC_ACT_SHOT;
760	}
761
762	skb->protocol = eth_type_trans(skb, skb->dev);
763	skb_reset_network_header(skb);
764
765	return action;
766}
767
768/*XXX: check if we can do this at install time instead of current
769 * send data path
770**/
771static int ife_get_sz(struct sk_buff *skb, struct tcf_ife_params *p)
772{
773	struct tcf_meta_info *e;
774	int tot_run_sz = 0, run_sz = 0;
775
776	list_for_each_entry_rcu(e, &p->metalist, metalist) {
777		if (e->ops->check_presence) {
778			run_sz = e->ops->check_presence(skb, e);
779			tot_run_sz += run_sz;
780		}
781	}
782
783	return tot_run_sz;
784}
785
786static int tcf_ife_encode(struct sk_buff *skb, const struct tc_action *a,
787			  struct tcf_result *res, struct tcf_ife_params *p)
788{
789	struct tcf_ife_info *ife = to_ife(a);
790	int action = ife->tcf_action;
791	struct ethhdr *oethh;	/* outer ether header */
792	struct tcf_meta_info *e;
793	/*
794	   OUTERHDR:TOTMETALEN:{TLVHDR:Metadatum:TLVHDR..}:ORIGDATA
795	   where ORIGDATA = original ethernet header ...
796	 */
797	u16 metalen = ife_get_sz(skb, p);
798	int hdrm = metalen + skb->dev->hard_header_len + IFE_METAHDRLEN;
799	unsigned int skboff = 0;
800	int new_len = skb->len + hdrm;
801	bool exceed_mtu = false;
802	void *ife_meta;
803	int err = 0;
804
805	if (!skb_at_tc_ingress(skb)) {
806		if (new_len > skb->dev->mtu)
807			exceed_mtu = true;
808	}
809
810	bstats_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
811	tcf_lastuse_update(&ife->tcf_tm);
812
813	if (!metalen) {		/* no metadata to send */
814		/* abuse overlimits to count when we allow packet
815		 * with no metadata
816		 */
817		qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
818		return action;
819	}
820	/* could be stupid policy setup or mtu config
821	 * so lets be conservative.. */
822	if ((action == TC_ACT_SHOT) || exceed_mtu) {
823drop:
824		qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
825		return TC_ACT_SHOT;
826	}
827
828	if (skb_at_tc_ingress(skb))
829		skb_push(skb, skb->dev->hard_header_len);
830
831	ife_meta = ife_encode(skb, metalen);
832	if (!ife_meta)
833		goto drop;
834
835	/* XXX: we dont have a clever way of telling encode to
836	 * not repeat some of the computations that are done by
837	 * ops->presence_check...
838	 */
839	list_for_each_entry_rcu(e, &p->metalist, metalist) {
840		if (e->ops->encode) {
841			err = e->ops->encode(skb, (void *)(ife_meta + skboff),
842					     e);
843		}
844		if (err < 0) {
845			/* too corrupt to keep around if overwritten */
846			goto drop;
847		}
848		skboff += err;
849	}
850	oethh = (struct ethhdr *)skb->data;
851
852	if (!is_zero_ether_addr(p->eth_src))
853		ether_addr_copy(oethh->h_source, p->eth_src);
854	if (!is_zero_ether_addr(p->eth_dst))
855		ether_addr_copy(oethh->h_dest, p->eth_dst);
856	oethh->h_proto = htons(p->eth_type);
857
858	if (skb_at_tc_ingress(skb))
859		skb_pull(skb, skb->dev->hard_header_len);
860
861	return action;
862}
863
864TC_INDIRECT_SCOPE int tcf_ife_act(struct sk_buff *skb,
865				  const struct tc_action *a,
866				  struct tcf_result *res)
867{
868	struct tcf_ife_info *ife = to_ife(a);
869	struct tcf_ife_params *p;
870	int ret;
871
872	p = rcu_dereference_bh(ife->params);
873	if (p->flags & IFE_ENCODE) {
874		ret = tcf_ife_encode(skb, a, res, p);
875		return ret;
876	}
877
878	return tcf_ife_decode(skb, a, res);
879}
880
881static struct tc_action_ops act_ife_ops = {
882	.kind = "ife",
883	.id = TCA_ID_IFE,
884	.owner = THIS_MODULE,
885	.act = tcf_ife_act,
886	.dump = tcf_ife_dump,
887	.cleanup = tcf_ife_cleanup,
888	.init = tcf_ife_init,
889	.size =	sizeof(struct tcf_ife_info),
890};
891MODULE_ALIAS_NET_ACT("ife");
892
893static __net_init int ife_init_net(struct net *net)
894{
895	struct tc_action_net *tn = net_generic(net, act_ife_ops.net_id);
896
897	return tc_action_net_init(net, tn, &act_ife_ops);
898}
899
900static void __net_exit ife_exit_net(struct list_head *net_list)
901{
902	tc_action_net_exit(net_list, act_ife_ops.net_id);
903}
904
905static struct pernet_operations ife_net_ops = {
906	.init = ife_init_net,
907	.exit_batch = ife_exit_net,
908	.id   = &act_ife_ops.net_id,
909	.size = sizeof(struct tc_action_net),
910};
911
912static int __init ife_init_module(void)
913{
914	return tcf_register_action(&act_ife_ops, &ife_net_ops);
915}
916
917static void __exit ife_cleanup_module(void)
918{
919	tcf_unregister_action(&act_ife_ops, &ife_net_ops);
920}
921
922module_init(ife_init_module);
923module_exit(ife_cleanup_module);
924
925MODULE_AUTHOR("Jamal Hadi Salim(2015)");
926MODULE_DESCRIPTION("Inter-FE LFB action");
927MODULE_LICENSE("GPL");