xref: /openbmc/linux/net/sched/act_ife.c (revision 0984d159)
1 /*
2  * net/sched/ife.c	Inter-FE action based on ForCES WG InterFE LFB
3  *
4  *		Refer to:
5  *		draft-ietf-forces-interfelfb-03
6  *		and
7  *		netdev01 paper:
8  *		"Distributing Linux Traffic Control Classifier-Action
9  *		Subsystem"
10  *		Authors: Jamal Hadi Salim and Damascene M. Joachimpillai
11  *
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.
16  *
17  * copyright Jamal Hadi Salim (2015)
18  *
19 */
20 
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>
35 
36 #define IFE_TAB_MASK 15
37 
38 static int ife_net_id;
39 static int max_metacnt = IFE_META_MAX + 1;
40 
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},
46 };
47 
48 /* Caller takes care of presenting data in network order
49 */
50 int ife_tlv_meta_encode(void *skbdata, u16 attrtype, u16 dlen, const void *dval)
51 {
52 	u32 *tlv = (u32 *)(skbdata);
53 	u16 totlen = nla_total_size(dlen);	/*alignment + hdr */
54 	char *dptr = (char *)tlv + NLA_HDRLEN;
55 	u32 htlv = attrtype << 16 | totlen;
56 
57 	*tlv = htonl(htlv);
58 	memset(dptr, 0, totlen - NLA_HDRLEN);
59 	memcpy(dptr, dval, dlen);
60 
61 	return totlen;
62 }
63 EXPORT_SYMBOL_GPL(ife_tlv_meta_encode);
64 
65 int ife_get_meta_u32(struct sk_buff *skb, struct tcf_meta_info *mi)
66 {
67 	if (mi->metaval)
68 		return nla_put_u32(skb, mi->metaid, *(u32 *)mi->metaval);
69 	else
70 		return nla_put(skb, mi->metaid, 0, NULL);
71 }
72 EXPORT_SYMBOL_GPL(ife_get_meta_u32);
73 
74 int ife_check_meta_u32(u32 metaval, struct tcf_meta_info *mi)
75 {
76 	if (metaval || mi->metaval)
77 		return 8; /* T+L+V == 2+2+4 */
78 
79 	return 0;
80 }
81 EXPORT_SYMBOL_GPL(ife_check_meta_u32);
82 
83 int ife_encode_meta_u32(u32 metaval, void *skbdata, struct tcf_meta_info *mi)
84 {
85 	u32 edata = metaval;
86 
87 	if (mi->metaval)
88 		edata = *(u32 *)mi->metaval;
89 	else if (metaval)
90 		edata = metaval;
91 
92 	if (!edata) /* will not encode */
93 		return 0;
94 
95 	edata = htonl(edata);
96 	return ife_tlv_meta_encode(skbdata, mi->metaid, 4, &edata);
97 }
98 EXPORT_SYMBOL_GPL(ife_encode_meta_u32);
99 
100 int ife_get_meta_u16(struct sk_buff *skb, struct tcf_meta_info *mi)
101 {
102 	if (mi->metaval)
103 		return nla_put_u16(skb, mi->metaid, *(u16 *)mi->metaval);
104 	else
105 		return nla_put(skb, mi->metaid, 0, NULL);
106 }
107 EXPORT_SYMBOL_GPL(ife_get_meta_u16);
108 
109 int ife_alloc_meta_u32(struct tcf_meta_info *mi, void *metaval, gfp_t gfp)
110 {
111 	mi->metaval = kmemdup(metaval, sizeof(u32), gfp);
112 	if (!mi->metaval)
113 		return -ENOMEM;
114 
115 	return 0;
116 }
117 EXPORT_SYMBOL_GPL(ife_alloc_meta_u32);
118 
119 int ife_alloc_meta_u16(struct tcf_meta_info *mi, void *metaval, gfp_t gfp)
120 {
121 	mi->metaval = kmemdup(metaval, sizeof(u16), gfp);
122 	if (!mi->metaval)
123 		return -ENOMEM;
124 
125 	return 0;
126 }
127 EXPORT_SYMBOL_GPL(ife_alloc_meta_u16);
128 
129 void ife_release_meta_gen(struct tcf_meta_info *mi)
130 {
131 	kfree(mi->metaval);
132 }
133 EXPORT_SYMBOL_GPL(ife_release_meta_gen);
134 
135 int ife_validate_meta_u32(void *val, int len)
136 {
137 	if (len == 4)
138 		return 0;
139 
140 	return -EINVAL;
141 }
142 EXPORT_SYMBOL_GPL(ife_validate_meta_u32);
143 
144 int ife_validate_meta_u16(void *val, int len)
145 {
146 	/* length will include padding */
147 	if (len == NLA_ALIGN(2))
148 		return 0;
149 
150 	return -EINVAL;
151 }
152 EXPORT_SYMBOL_GPL(ife_validate_meta_u16);
153 
154 static LIST_HEAD(ifeoplist);
155 static DEFINE_RWLOCK(ife_mod_lock);
156 
157 static struct tcf_meta_ops *find_ife_oplist(u16 metaid)
158 {
159 	struct tcf_meta_ops *o;
160 
161 	read_lock(&ife_mod_lock);
162 	list_for_each_entry(o, &ifeoplist, list) {
163 		if (o->metaid == metaid) {
164 			if (!try_module_get(o->owner))
165 				o = NULL;
166 			read_unlock(&ife_mod_lock);
167 			return o;
168 		}
169 	}
170 	read_unlock(&ife_mod_lock);
171 
172 	return NULL;
173 }
174 
175 int register_ife_op(struct tcf_meta_ops *mops)
176 {
177 	struct tcf_meta_ops *m;
178 
179 	if (!mops->metaid || !mops->metatype || !mops->name ||
180 	    !mops->check_presence || !mops->encode || !mops->decode ||
181 	    !mops->get || !mops->alloc)
182 		return -EINVAL;
183 
184 	write_lock(&ife_mod_lock);
185 
186 	list_for_each_entry(m, &ifeoplist, list) {
187 		if (m->metaid == mops->metaid ||
188 		    (strcmp(mops->name, m->name) == 0)) {
189 			write_unlock(&ife_mod_lock);
190 			return -EEXIST;
191 		}
192 	}
193 
194 	if (!mops->release)
195 		mops->release = ife_release_meta_gen;
196 
197 	list_add_tail(&mops->list, &ifeoplist);
198 	write_unlock(&ife_mod_lock);
199 	return 0;
200 }
201 EXPORT_SYMBOL_GPL(unregister_ife_op);
202 
203 int unregister_ife_op(struct tcf_meta_ops *mops)
204 {
205 	struct tcf_meta_ops *m;
206 	int err = -ENOENT;
207 
208 	write_lock(&ife_mod_lock);
209 	list_for_each_entry(m, &ifeoplist, list) {
210 		if (m->metaid == mops->metaid) {
211 			list_del(&mops->list);
212 			err = 0;
213 			break;
214 		}
215 	}
216 	write_unlock(&ife_mod_lock);
217 
218 	return err;
219 }
220 EXPORT_SYMBOL_GPL(register_ife_op);
221 
222 static int ife_validate_metatype(struct tcf_meta_ops *ops, void *val, int len)
223 {
224 	int ret = 0;
225 	/* XXX: unfortunately cant use nla_policy at this point
226 	* because a length of 0 is valid in the case of
227 	* "allow". "use" semantics do enforce for proper
228 	* length and i couldve use nla_policy but it makes it hard
229 	* to use it just for that..
230 	*/
231 	if (ops->validate)
232 		return ops->validate(val, len);
233 
234 	if (ops->metatype == NLA_U32)
235 		ret = ife_validate_meta_u32(val, len);
236 	else if (ops->metatype == NLA_U16)
237 		ret = ife_validate_meta_u16(val, len);
238 
239 	return ret;
240 }
241 
242 /* called when adding new meta information
243  * under ife->tcf_lock for existing action
244 */
245 static int load_metaops_and_vet(struct tcf_ife_info *ife, u32 metaid,
246 				void *val, int len, bool exists)
247 {
248 	struct tcf_meta_ops *ops = find_ife_oplist(metaid);
249 	int ret = 0;
250 
251 	if (!ops) {
252 		ret = -ENOENT;
253 #ifdef CONFIG_MODULES
254 		if (exists)
255 			spin_unlock_bh(&ife->tcf_lock);
256 		rtnl_unlock();
257 		request_module("ifemeta%u", metaid);
258 		rtnl_lock();
259 		if (exists)
260 			spin_lock_bh(&ife->tcf_lock);
261 		ops = find_ife_oplist(metaid);
262 #endif
263 	}
264 
265 	if (ops) {
266 		ret = 0;
267 		if (len)
268 			ret = ife_validate_metatype(ops, val, len);
269 
270 		module_put(ops->owner);
271 	}
272 
273 	return ret;
274 }
275 
276 /* called when adding new meta information
277  * under ife->tcf_lock for existing action
278 */
279 static int add_metainfo(struct tcf_ife_info *ife, u32 metaid, void *metaval,
280 			int len, bool atomic)
281 {
282 	struct tcf_meta_info *mi = NULL;
283 	struct tcf_meta_ops *ops = find_ife_oplist(metaid);
284 	int ret = 0;
285 
286 	if (!ops)
287 		return -ENOENT;
288 
289 	mi = kzalloc(sizeof(*mi), atomic ? GFP_ATOMIC : GFP_KERNEL);
290 	if (!mi) {
291 		/*put back what find_ife_oplist took */
292 		module_put(ops->owner);
293 		return -ENOMEM;
294 	}
295 
296 	mi->metaid = metaid;
297 	mi->ops = ops;
298 	if (len > 0) {
299 		ret = ops->alloc(mi, metaval, atomic ? GFP_ATOMIC : GFP_KERNEL);
300 		if (ret != 0) {
301 			kfree(mi);
302 			module_put(ops->owner);
303 			return ret;
304 		}
305 	}
306 
307 	list_add_tail(&mi->metalist, &ife->metalist);
308 
309 	return ret;
310 }
311 
312 static int use_all_metadata(struct tcf_ife_info *ife)
313 {
314 	struct tcf_meta_ops *o;
315 	int rc = 0;
316 	int installed = 0;
317 
318 	read_lock(&ife_mod_lock);
319 	list_for_each_entry(o, &ifeoplist, list) {
320 		rc = add_metainfo(ife, o->metaid, NULL, 0, true);
321 		if (rc == 0)
322 			installed += 1;
323 	}
324 	read_unlock(&ife_mod_lock);
325 
326 	if (installed)
327 		return 0;
328 	else
329 		return -EINVAL;
330 }
331 
332 static int dump_metalist(struct sk_buff *skb, struct tcf_ife_info *ife)
333 {
334 	struct tcf_meta_info *e;
335 	struct nlattr *nest;
336 	unsigned char *b = skb_tail_pointer(skb);
337 	int total_encoded = 0;
338 
339 	/*can only happen on decode */
340 	if (list_empty(&ife->metalist))
341 		return 0;
342 
343 	nest = nla_nest_start(skb, TCA_IFE_METALST);
344 	if (!nest)
345 		goto out_nlmsg_trim;
346 
347 	list_for_each_entry(e, &ife->metalist, metalist) {
348 		if (!e->ops->get(skb, e))
349 			total_encoded += 1;
350 	}
351 
352 	if (!total_encoded)
353 		goto out_nlmsg_trim;
354 
355 	nla_nest_end(skb, nest);
356 
357 	return 0;
358 
359 out_nlmsg_trim:
360 	nlmsg_trim(skb, b);
361 	return -1;
362 }
363 
364 /* under ife->tcf_lock */
365 static void _tcf_ife_cleanup(struct tc_action *a, int bind)
366 {
367 	struct tcf_ife_info *ife = a->priv;
368 	struct tcf_meta_info *e, *n;
369 
370 	list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
371 		module_put(e->ops->owner);
372 		list_del(&e->metalist);
373 		if (e->metaval) {
374 			if (e->ops->release)
375 				e->ops->release(e);
376 			else
377 				kfree(e->metaval);
378 		}
379 		kfree(e);
380 	}
381 }
382 
383 static void tcf_ife_cleanup(struct tc_action *a, int bind)
384 {
385 	struct tcf_ife_info *ife = a->priv;
386 
387 	spin_lock_bh(&ife->tcf_lock);
388 	_tcf_ife_cleanup(a, bind);
389 	spin_unlock_bh(&ife->tcf_lock);
390 }
391 
392 /* under ife->tcf_lock for existing action */
393 static int populate_metalist(struct tcf_ife_info *ife, struct nlattr **tb,
394 			     bool exists)
395 {
396 	int len = 0;
397 	int rc = 0;
398 	int i = 0;
399 	void *val;
400 
401 	for (i = 1; i < max_metacnt; i++) {
402 		if (tb[i]) {
403 			val = nla_data(tb[i]);
404 			len = nla_len(tb[i]);
405 
406 			rc = load_metaops_and_vet(ife, i, val, len, exists);
407 			if (rc != 0)
408 				return rc;
409 
410 			rc = add_metainfo(ife, i, val, len, exists);
411 			if (rc)
412 				return rc;
413 		}
414 	}
415 
416 	return rc;
417 }
418 
419 static int tcf_ife_init(struct net *net, struct nlattr *nla,
420 			struct nlattr *est, struct tc_action *a,
421 			int ovr, int bind)
422 {
423 	struct tc_action_net *tn = net_generic(net, ife_net_id);
424 	struct nlattr *tb[TCA_IFE_MAX + 1];
425 	struct nlattr *tb2[IFE_META_MAX + 1];
426 	struct tcf_ife_info *ife;
427 	struct tc_ife *parm;
428 	u16 ife_type = 0;
429 	u8 *daddr = NULL;
430 	u8 *saddr = NULL;
431 	int ret = 0, exists = 0;
432 	int err;
433 
434 	err = nla_parse_nested(tb, TCA_IFE_MAX, nla, ife_policy);
435 	if (err < 0)
436 		return err;
437 
438 	if (!tb[TCA_IFE_PARMS])
439 		return -EINVAL;
440 
441 	parm = nla_data(tb[TCA_IFE_PARMS]);
442 
443 	exists = tcf_hash_check(tn, parm->index, a, bind);
444 	if (exists && bind)
445 		return 0;
446 
447 	if (parm->flags & IFE_ENCODE) {
448 		/* Until we get issued the ethertype, we cant have
449 		 * a default..
450 		**/
451 		if (!tb[TCA_IFE_TYPE]) {
452 			if (exists)
453 				tcf_hash_release(a, bind);
454 			pr_info("You MUST pass etherype for encoding\n");
455 			return -EINVAL;
456 		}
457 	}
458 
459 	if (!exists) {
460 		ret = tcf_hash_create(tn, parm->index, est, a, sizeof(*ife),
461 				      bind, false);
462 		if (ret)
463 			return ret;
464 		ret = ACT_P_CREATED;
465 	} else {
466 		tcf_hash_release(a, bind);
467 		if (!ovr)
468 			return -EEXIST;
469 	}
470 
471 	ife = to_ife(a);
472 	ife->flags = parm->flags;
473 
474 	if (parm->flags & IFE_ENCODE) {
475 		ife_type = nla_get_u16(tb[TCA_IFE_TYPE]);
476 		if (tb[TCA_IFE_DMAC])
477 			daddr = nla_data(tb[TCA_IFE_DMAC]);
478 		if (tb[TCA_IFE_SMAC])
479 			saddr = nla_data(tb[TCA_IFE_SMAC]);
480 	}
481 
482 	if (exists)
483 		spin_lock_bh(&ife->tcf_lock);
484 	ife->tcf_action = parm->action;
485 
486 	if (parm->flags & IFE_ENCODE) {
487 		if (daddr)
488 			ether_addr_copy(ife->eth_dst, daddr);
489 		else
490 			eth_zero_addr(ife->eth_dst);
491 
492 		if (saddr)
493 			ether_addr_copy(ife->eth_src, saddr);
494 		else
495 			eth_zero_addr(ife->eth_src);
496 
497 		ife->eth_type = ife_type;
498 	}
499 
500 	if (ret == ACT_P_CREATED)
501 		INIT_LIST_HEAD(&ife->metalist);
502 
503 	if (tb[TCA_IFE_METALST]) {
504 		err = nla_parse_nested(tb2, IFE_META_MAX, tb[TCA_IFE_METALST],
505 				       NULL);
506 		if (err) {
507 metadata_parse_err:
508 			if (exists)
509 				tcf_hash_release(a, bind);
510 			if (ret == ACT_P_CREATED)
511 				_tcf_ife_cleanup(a, bind);
512 
513 			if (exists)
514 				spin_unlock_bh(&ife->tcf_lock);
515 			return err;
516 		}
517 
518 		err = populate_metalist(ife, tb2, exists);
519 		if (err)
520 			goto metadata_parse_err;
521 
522 	} else {
523 		/* if no passed metadata allow list or passed allow-all
524 		 * then here we process by adding as many supported metadatum
525 		 * as we can. You better have at least one else we are
526 		 * going to bail out
527 		 */
528 		err = use_all_metadata(ife);
529 		if (err) {
530 			if (ret == ACT_P_CREATED)
531 				_tcf_ife_cleanup(a, bind);
532 
533 			if (exists)
534 				spin_unlock_bh(&ife->tcf_lock);
535 			return err;
536 		}
537 	}
538 
539 	if (exists)
540 		spin_unlock_bh(&ife->tcf_lock);
541 
542 	if (ret == ACT_P_CREATED)
543 		tcf_hash_insert(tn, a);
544 
545 	return ret;
546 }
547 
548 static int tcf_ife_dump(struct sk_buff *skb, struct tc_action *a, int bind,
549 			int ref)
550 {
551 	unsigned char *b = skb_tail_pointer(skb);
552 	struct tcf_ife_info *ife = a->priv;
553 	struct tc_ife opt = {
554 		.index = ife->tcf_index,
555 		.refcnt = ife->tcf_refcnt - ref,
556 		.bindcnt = ife->tcf_bindcnt - bind,
557 		.action = ife->tcf_action,
558 		.flags = ife->flags,
559 	};
560 	struct tcf_t t;
561 
562 	if (nla_put(skb, TCA_IFE_PARMS, sizeof(opt), &opt))
563 		goto nla_put_failure;
564 
565 	t.install = jiffies_to_clock_t(jiffies - ife->tcf_tm.install);
566 	t.lastuse = jiffies_to_clock_t(jiffies - ife->tcf_tm.lastuse);
567 	t.expires = jiffies_to_clock_t(ife->tcf_tm.expires);
568 	if (nla_put_64bit(skb, TCA_IFE_TM, sizeof(t), &t, TCA_IFE_PAD))
569 		goto nla_put_failure;
570 
571 	if (!is_zero_ether_addr(ife->eth_dst)) {
572 		if (nla_put(skb, TCA_IFE_DMAC, ETH_ALEN, ife->eth_dst))
573 			goto nla_put_failure;
574 	}
575 
576 	if (!is_zero_ether_addr(ife->eth_src)) {
577 		if (nla_put(skb, TCA_IFE_SMAC, ETH_ALEN, ife->eth_src))
578 			goto nla_put_failure;
579 	}
580 
581 	if (nla_put(skb, TCA_IFE_TYPE, 2, &ife->eth_type))
582 		goto nla_put_failure;
583 
584 	if (dump_metalist(skb, ife)) {
585 		/*ignore failure to dump metalist */
586 		pr_info("Failed to dump metalist\n");
587 	}
588 
589 	return skb->len;
590 
591 nla_put_failure:
592 	nlmsg_trim(skb, b);
593 	return -1;
594 }
595 
596 int find_decode_metaid(struct sk_buff *skb, struct tcf_ife_info *ife,
597 		       u16 metaid, u16 mlen, void *mdata)
598 {
599 	struct tcf_meta_info *e;
600 
601 	/* XXX: use hash to speed up */
602 	list_for_each_entry(e, &ife->metalist, metalist) {
603 		if (metaid == e->metaid) {
604 			if (e->ops) {
605 				/* We check for decode presence already */
606 				return e->ops->decode(skb, mdata, mlen);
607 			}
608 		}
609 	}
610 
611 	return 0;
612 }
613 
614 struct ifeheadr {
615 	__be16 metalen;
616 	u8 tlv_data[];
617 };
618 
619 struct meta_tlvhdr {
620 	__be16 type;
621 	__be16 len;
622 };
623 
624 static int tcf_ife_decode(struct sk_buff *skb, const struct tc_action *a,
625 			  struct tcf_result *res)
626 {
627 	struct tcf_ife_info *ife = a->priv;
628 	int action = ife->tcf_action;
629 	struct ifeheadr *ifehdr = (struct ifeheadr *)skb->data;
630 	u16 ifehdrln = ifehdr->metalen;
631 	struct meta_tlvhdr *tlv = (struct meta_tlvhdr *)(ifehdr->tlv_data);
632 
633 	spin_lock(&ife->tcf_lock);
634 	bstats_update(&ife->tcf_bstats, skb);
635 	ife->tcf_tm.lastuse = jiffies;
636 	spin_unlock(&ife->tcf_lock);
637 
638 	ifehdrln = ntohs(ifehdrln);
639 	if (unlikely(!pskb_may_pull(skb, ifehdrln))) {
640 		spin_lock(&ife->tcf_lock);
641 		ife->tcf_qstats.drops++;
642 		spin_unlock(&ife->tcf_lock);
643 		return TC_ACT_SHOT;
644 	}
645 
646 	skb_set_mac_header(skb, ifehdrln);
647 	__skb_pull(skb, ifehdrln);
648 	skb->protocol = eth_type_trans(skb, skb->dev);
649 	ifehdrln -= IFE_METAHDRLEN;
650 
651 	while (ifehdrln > 0) {
652 		u8 *tlvdata = (u8 *)tlv;
653 		u16 mtype = tlv->type;
654 		u16 mlen = tlv->len;
655 
656 		mtype = ntohs(mtype);
657 		mlen = ntohs(mlen);
658 
659 		if (find_decode_metaid(skb, ife, mtype, (mlen - 4),
660 				       (void *)(tlvdata + 4))) {
661 			/* abuse overlimits to count when we receive metadata
662 			 * but dont have an ops for it
663 			 */
664 			pr_info_ratelimited("Unknown metaid %d alnlen %d\n",
665 					    mtype, mlen);
666 			ife->tcf_qstats.overlimits++;
667 		}
668 
669 		tlvdata += mlen;
670 		ifehdrln -= mlen;
671 		tlv = (struct meta_tlvhdr *)tlvdata;
672 	}
673 
674 	skb_reset_network_header(skb);
675 	return action;
676 }
677 
678 /*XXX: check if we can do this at install time instead of current
679  * send data path
680 **/
681 static int ife_get_sz(struct sk_buff *skb, struct tcf_ife_info *ife)
682 {
683 	struct tcf_meta_info *e, *n;
684 	int tot_run_sz = 0, run_sz = 0;
685 
686 	list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
687 		if (e->ops->check_presence) {
688 			run_sz = e->ops->check_presence(skb, e);
689 			tot_run_sz += run_sz;
690 		}
691 	}
692 
693 	return tot_run_sz;
694 }
695 
696 static int tcf_ife_encode(struct sk_buff *skb, const struct tc_action *a,
697 			  struct tcf_result *res)
698 {
699 	struct tcf_ife_info *ife = a->priv;
700 	int action = ife->tcf_action;
701 	struct ethhdr *oethh;	/* outer ether header */
702 	struct ethhdr *iethh;	/* inner eth header */
703 	struct tcf_meta_info *e;
704 	/*
705 	   OUTERHDR:TOTMETALEN:{TLVHDR:Metadatum:TLVHDR..}:ORIGDATA
706 	   where ORIGDATA = original ethernet header ...
707 	 */
708 	u16 metalen = ife_get_sz(skb, ife);
709 	int hdrm = metalen + skb->dev->hard_header_len + IFE_METAHDRLEN;
710 	unsigned int skboff = skb->dev->hard_header_len;
711 	u32 at = G_TC_AT(skb->tc_verd);
712 	int new_len = skb->len + hdrm;
713 	bool exceed_mtu = false;
714 	int err;
715 
716 	if (at & AT_EGRESS) {
717 		if (new_len > skb->dev->mtu)
718 			exceed_mtu = true;
719 	}
720 
721 	spin_lock(&ife->tcf_lock);
722 	bstats_update(&ife->tcf_bstats, skb);
723 	ife->tcf_tm.lastuse = jiffies;
724 
725 	if (!metalen) {		/* no metadata to send */
726 		/* abuse overlimits to count when we allow packet
727 		 * with no metadata
728 		 */
729 		ife->tcf_qstats.overlimits++;
730 		spin_unlock(&ife->tcf_lock);
731 		return action;
732 	}
733 	/* could be stupid policy setup or mtu config
734 	 * so lets be conservative.. */
735 	if ((action == TC_ACT_SHOT) || exceed_mtu) {
736 		ife->tcf_qstats.drops++;
737 		spin_unlock(&ife->tcf_lock);
738 		return TC_ACT_SHOT;
739 	}
740 
741 	iethh = eth_hdr(skb);
742 
743 	err = skb_cow_head(skb, hdrm);
744 	if (unlikely(err)) {
745 		ife->tcf_qstats.drops++;
746 		spin_unlock(&ife->tcf_lock);
747 		return TC_ACT_SHOT;
748 	}
749 
750 	if (!(at & AT_EGRESS))
751 		skb_push(skb, skb->dev->hard_header_len);
752 
753 	__skb_push(skb, hdrm);
754 	memcpy(skb->data, iethh, skb->mac_len);
755 	skb_reset_mac_header(skb);
756 	oethh = eth_hdr(skb);
757 
758 	/*total metadata length */
759 	metalen += IFE_METAHDRLEN;
760 	metalen = htons(metalen);
761 	memcpy((skb->data + skboff), &metalen, IFE_METAHDRLEN);
762 	skboff += IFE_METAHDRLEN;
763 
764 	/* XXX: we dont have a clever way of telling encode to
765 	 * not repeat some of the computations that are done by
766 	 * ops->presence_check...
767 	 */
768 	list_for_each_entry(e, &ife->metalist, metalist) {
769 		if (e->ops->encode) {
770 			err = e->ops->encode(skb, (void *)(skb->data + skboff),
771 					     e);
772 		}
773 		if (err < 0) {
774 			/* too corrupt to keep around if overwritten */
775 			ife->tcf_qstats.drops++;
776 			spin_unlock(&ife->tcf_lock);
777 			return TC_ACT_SHOT;
778 		}
779 		skboff += err;
780 	}
781 
782 	if (!is_zero_ether_addr(ife->eth_src))
783 		ether_addr_copy(oethh->h_source, ife->eth_src);
784 	else
785 		ether_addr_copy(oethh->h_source, iethh->h_source);
786 	if (!is_zero_ether_addr(ife->eth_dst))
787 		ether_addr_copy(oethh->h_dest, ife->eth_dst);
788 	else
789 		ether_addr_copy(oethh->h_dest, iethh->h_dest);
790 	oethh->h_proto = htons(ife->eth_type);
791 
792 	if (!(at & AT_EGRESS))
793 		skb_pull(skb, skb->dev->hard_header_len);
794 
795 	spin_unlock(&ife->tcf_lock);
796 
797 	return action;
798 }
799 
800 static int tcf_ife_act(struct sk_buff *skb, const struct tc_action *a,
801 		       struct tcf_result *res)
802 {
803 	struct tcf_ife_info *ife = a->priv;
804 
805 	if (ife->flags & IFE_ENCODE)
806 		return tcf_ife_encode(skb, a, res);
807 
808 	if (!(ife->flags & IFE_ENCODE))
809 		return tcf_ife_decode(skb, a, res);
810 
811 	pr_info_ratelimited("unknown failure(policy neither de/encode\n");
812 	spin_lock(&ife->tcf_lock);
813 	bstats_update(&ife->tcf_bstats, skb);
814 	ife->tcf_tm.lastuse = jiffies;
815 	ife->tcf_qstats.drops++;
816 	spin_unlock(&ife->tcf_lock);
817 
818 	return TC_ACT_SHOT;
819 }
820 
821 static int tcf_ife_walker(struct net *net, struct sk_buff *skb,
822 			  struct netlink_callback *cb, int type,
823 			  struct tc_action *a)
824 {
825 	struct tc_action_net *tn = net_generic(net, ife_net_id);
826 
827 	return tcf_generic_walker(tn, skb, cb, type, a);
828 }
829 
830 static int tcf_ife_search(struct net *net, struct tc_action *a, u32 index)
831 {
832 	struct tc_action_net *tn = net_generic(net, ife_net_id);
833 
834 	return tcf_hash_search(tn, a, index);
835 }
836 
837 static struct tc_action_ops act_ife_ops = {
838 	.kind = "ife",
839 	.type = TCA_ACT_IFE,
840 	.owner = THIS_MODULE,
841 	.act = tcf_ife_act,
842 	.dump = tcf_ife_dump,
843 	.cleanup = tcf_ife_cleanup,
844 	.init = tcf_ife_init,
845 	.walk = tcf_ife_walker,
846 	.lookup = tcf_ife_search,
847 };
848 
849 static __net_init int ife_init_net(struct net *net)
850 {
851 	struct tc_action_net *tn = net_generic(net, ife_net_id);
852 
853 	return tc_action_net_init(tn, &act_ife_ops, IFE_TAB_MASK);
854 }
855 
856 static void __net_exit ife_exit_net(struct net *net)
857 {
858 	struct tc_action_net *tn = net_generic(net, ife_net_id);
859 
860 	tc_action_net_exit(tn);
861 }
862 
863 static struct pernet_operations ife_net_ops = {
864 	.init = ife_init_net,
865 	.exit = ife_exit_net,
866 	.id   = &ife_net_id,
867 	.size = sizeof(struct tc_action_net),
868 };
869 
870 static int __init ife_init_module(void)
871 {
872 	return tcf_register_action(&act_ife_ops, &ife_net_ops);
873 }
874 
875 static void __exit ife_cleanup_module(void)
876 {
877 	tcf_unregister_action(&act_ife_ops, &ife_net_ops);
878 }
879 
880 module_init(ife_init_module);
881 module_exit(ife_cleanup_module);
882 
883 MODULE_AUTHOR("Jamal Hadi Salim(2015)");
884 MODULE_DESCRIPTION("Inter-FE LFB action");
885 MODULE_LICENSE("GPL");
886