xref: /openbmc/linux/net/sched/act_ife.c (revision 77a87824)
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 static struct tc_action_ops act_ife_ops;
41 
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},
47 };
48 
49 /* Caller takes care of presenting data in network order
50 */
51 int ife_tlv_meta_encode(void *skbdata, u16 attrtype, u16 dlen, const void *dval)
52 {
53 	u32 *tlv = (u32 *)(skbdata);
54 	u16 totlen = nla_total_size(dlen);	/*alignment + hdr */
55 	char *dptr = (char *)tlv + NLA_HDRLEN;
56 	u32 htlv = attrtype << 16 | totlen;
57 
58 	*tlv = htonl(htlv);
59 	memset(dptr, 0, totlen - NLA_HDRLEN);
60 	memcpy(dptr, dval, dlen);
61 
62 	return totlen;
63 }
64 EXPORT_SYMBOL_GPL(ife_tlv_meta_encode);
65 
66 int ife_get_meta_u32(struct sk_buff *skb, struct tcf_meta_info *mi)
67 {
68 	if (mi->metaval)
69 		return nla_put_u32(skb, mi->metaid, *(u32 *)mi->metaval);
70 	else
71 		return nla_put(skb, mi->metaid, 0, NULL);
72 }
73 EXPORT_SYMBOL_GPL(ife_get_meta_u32);
74 
75 int ife_check_meta_u32(u32 metaval, struct tcf_meta_info *mi)
76 {
77 	if (metaval || mi->metaval)
78 		return 8; /* T+L+V == 2+2+4 */
79 
80 	return 0;
81 }
82 EXPORT_SYMBOL_GPL(ife_check_meta_u32);
83 
84 int ife_encode_meta_u32(u32 metaval, void *skbdata, struct tcf_meta_info *mi)
85 {
86 	u32 edata = metaval;
87 
88 	if (mi->metaval)
89 		edata = *(u32 *)mi->metaval;
90 	else if (metaval)
91 		edata = metaval;
92 
93 	if (!edata) /* will not encode */
94 		return 0;
95 
96 	edata = htonl(edata);
97 	return ife_tlv_meta_encode(skbdata, mi->metaid, 4, &edata);
98 }
99 EXPORT_SYMBOL_GPL(ife_encode_meta_u32);
100 
101 int ife_get_meta_u16(struct sk_buff *skb, struct tcf_meta_info *mi)
102 {
103 	if (mi->metaval)
104 		return nla_put_u16(skb, mi->metaid, *(u16 *)mi->metaval);
105 	else
106 		return nla_put(skb, mi->metaid, 0, NULL);
107 }
108 EXPORT_SYMBOL_GPL(ife_get_meta_u16);
109 
110 int ife_alloc_meta_u32(struct tcf_meta_info *mi, void *metaval, gfp_t gfp)
111 {
112 	mi->metaval = kmemdup(metaval, sizeof(u32), gfp);
113 	if (!mi->metaval)
114 		return -ENOMEM;
115 
116 	return 0;
117 }
118 EXPORT_SYMBOL_GPL(ife_alloc_meta_u32);
119 
120 int ife_alloc_meta_u16(struct tcf_meta_info *mi, void *metaval, gfp_t gfp)
121 {
122 	mi->metaval = kmemdup(metaval, sizeof(u16), gfp);
123 	if (!mi->metaval)
124 		return -ENOMEM;
125 
126 	return 0;
127 }
128 EXPORT_SYMBOL_GPL(ife_alloc_meta_u16);
129 
130 void ife_release_meta_gen(struct tcf_meta_info *mi)
131 {
132 	kfree(mi->metaval);
133 }
134 EXPORT_SYMBOL_GPL(ife_release_meta_gen);
135 
136 int ife_validate_meta_u32(void *val, int len)
137 {
138 	if (len == 4)
139 		return 0;
140 
141 	return -EINVAL;
142 }
143 EXPORT_SYMBOL_GPL(ife_validate_meta_u32);
144 
145 int ife_validate_meta_u16(void *val, int len)
146 {
147 	/* length will include padding */
148 	if (len == NLA_ALIGN(2))
149 		return 0;
150 
151 	return -EINVAL;
152 }
153 EXPORT_SYMBOL_GPL(ife_validate_meta_u16);
154 
155 static LIST_HEAD(ifeoplist);
156 static DEFINE_RWLOCK(ife_mod_lock);
157 
158 static struct tcf_meta_ops *find_ife_oplist(u16 metaid)
159 {
160 	struct tcf_meta_ops *o;
161 
162 	read_lock(&ife_mod_lock);
163 	list_for_each_entry(o, &ifeoplist, list) {
164 		if (o->metaid == metaid) {
165 			if (!try_module_get(o->owner))
166 				o = NULL;
167 			read_unlock(&ife_mod_lock);
168 			return o;
169 		}
170 	}
171 	read_unlock(&ife_mod_lock);
172 
173 	return NULL;
174 }
175 
176 int register_ife_op(struct tcf_meta_ops *mops)
177 {
178 	struct tcf_meta_ops *m;
179 
180 	if (!mops->metaid || !mops->metatype || !mops->name ||
181 	    !mops->check_presence || !mops->encode || !mops->decode ||
182 	    !mops->get || !mops->alloc)
183 		return -EINVAL;
184 
185 	write_lock(&ife_mod_lock);
186 
187 	list_for_each_entry(m, &ifeoplist, list) {
188 		if (m->metaid == mops->metaid ||
189 		    (strcmp(mops->name, m->name) == 0)) {
190 			write_unlock(&ife_mod_lock);
191 			return -EEXIST;
192 		}
193 	}
194 
195 	if (!mops->release)
196 		mops->release = ife_release_meta_gen;
197 
198 	list_add_tail(&mops->list, &ifeoplist);
199 	write_unlock(&ife_mod_lock);
200 	return 0;
201 }
202 EXPORT_SYMBOL_GPL(unregister_ife_op);
203 
204 int unregister_ife_op(struct tcf_meta_ops *mops)
205 {
206 	struct tcf_meta_ops *m;
207 	int err = -ENOENT;
208 
209 	write_lock(&ife_mod_lock);
210 	list_for_each_entry(m, &ifeoplist, list) {
211 		if (m->metaid == mops->metaid) {
212 			list_del(&mops->list);
213 			err = 0;
214 			break;
215 		}
216 	}
217 	write_unlock(&ife_mod_lock);
218 
219 	return err;
220 }
221 EXPORT_SYMBOL_GPL(register_ife_op);
222 
223 static int ife_validate_metatype(struct tcf_meta_ops *ops, void *val, int len)
224 {
225 	int ret = 0;
226 	/* XXX: unfortunately cant use nla_policy at this point
227 	* because a length of 0 is valid in the case of
228 	* "allow". "use" semantics do enforce for proper
229 	* length and i couldve use nla_policy but it makes it hard
230 	* to use it just for that..
231 	*/
232 	if (ops->validate)
233 		return ops->validate(val, len);
234 
235 	if (ops->metatype == NLA_U32)
236 		ret = ife_validate_meta_u32(val, len);
237 	else if (ops->metatype == NLA_U16)
238 		ret = ife_validate_meta_u16(val, len);
239 
240 	return ret;
241 }
242 
243 /* called when adding new meta information
244  * under ife->tcf_lock for existing action
245 */
246 static int load_metaops_and_vet(struct tcf_ife_info *ife, u32 metaid,
247 				void *val, int len, bool exists)
248 {
249 	struct tcf_meta_ops *ops = find_ife_oplist(metaid);
250 	int ret = 0;
251 
252 	if (!ops) {
253 		ret = -ENOENT;
254 #ifdef CONFIG_MODULES
255 		if (exists)
256 			spin_unlock_bh(&ife->tcf_lock);
257 		rtnl_unlock();
258 		request_module("ifemeta%u", metaid);
259 		rtnl_lock();
260 		if (exists)
261 			spin_lock_bh(&ife->tcf_lock);
262 		ops = find_ife_oplist(metaid);
263 #endif
264 	}
265 
266 	if (ops) {
267 		ret = 0;
268 		if (len)
269 			ret = ife_validate_metatype(ops, val, len);
270 
271 		module_put(ops->owner);
272 	}
273 
274 	return ret;
275 }
276 
277 /* called when adding new meta information
278  * under ife->tcf_lock for existing action
279 */
280 static int add_metainfo(struct tcf_ife_info *ife, u32 metaid, void *metaval,
281 			int len, bool atomic)
282 {
283 	struct tcf_meta_info *mi = NULL;
284 	struct tcf_meta_ops *ops = find_ife_oplist(metaid);
285 	int ret = 0;
286 
287 	if (!ops)
288 		return -ENOENT;
289 
290 	mi = kzalloc(sizeof(*mi), atomic ? GFP_ATOMIC : GFP_KERNEL);
291 	if (!mi) {
292 		/*put back what find_ife_oplist took */
293 		module_put(ops->owner);
294 		return -ENOMEM;
295 	}
296 
297 	mi->metaid = metaid;
298 	mi->ops = ops;
299 	if (len > 0) {
300 		ret = ops->alloc(mi, metaval, atomic ? GFP_ATOMIC : GFP_KERNEL);
301 		if (ret != 0) {
302 			kfree(mi);
303 			module_put(ops->owner);
304 			return ret;
305 		}
306 	}
307 
308 	list_add_tail(&mi->metalist, &ife->metalist);
309 
310 	return ret;
311 }
312 
313 static int use_all_metadata(struct tcf_ife_info *ife)
314 {
315 	struct tcf_meta_ops *o;
316 	int rc = 0;
317 	int installed = 0;
318 
319 	read_lock(&ife_mod_lock);
320 	list_for_each_entry(o, &ifeoplist, list) {
321 		rc = add_metainfo(ife, o->metaid, NULL, 0, true);
322 		if (rc == 0)
323 			installed += 1;
324 	}
325 	read_unlock(&ife_mod_lock);
326 
327 	if (installed)
328 		return 0;
329 	else
330 		return -EINVAL;
331 }
332 
333 static int dump_metalist(struct sk_buff *skb, struct tcf_ife_info *ife)
334 {
335 	struct tcf_meta_info *e;
336 	struct nlattr *nest;
337 	unsigned char *b = skb_tail_pointer(skb);
338 	int total_encoded = 0;
339 
340 	/*can only happen on decode */
341 	if (list_empty(&ife->metalist))
342 		return 0;
343 
344 	nest = nla_nest_start(skb, TCA_IFE_METALST);
345 	if (!nest)
346 		goto out_nlmsg_trim;
347 
348 	list_for_each_entry(e, &ife->metalist, metalist) {
349 		if (!e->ops->get(skb, e))
350 			total_encoded += 1;
351 	}
352 
353 	if (!total_encoded)
354 		goto out_nlmsg_trim;
355 
356 	nla_nest_end(skb, nest);
357 
358 	return 0;
359 
360 out_nlmsg_trim:
361 	nlmsg_trim(skb, b);
362 	return -1;
363 }
364 
365 /* under ife->tcf_lock */
366 static void _tcf_ife_cleanup(struct tc_action *a, int bind)
367 {
368 	struct tcf_ife_info *ife = to_ife(a);
369 	struct tcf_meta_info *e, *n;
370 
371 	list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
372 		module_put(e->ops->owner);
373 		list_del(&e->metalist);
374 		if (e->metaval) {
375 			if (e->ops->release)
376 				e->ops->release(e);
377 			else
378 				kfree(e->metaval);
379 		}
380 		kfree(e);
381 	}
382 }
383 
384 static void tcf_ife_cleanup(struct tc_action *a, int bind)
385 {
386 	struct tcf_ife_info *ife = to_ife(a);
387 
388 	spin_lock_bh(&ife->tcf_lock);
389 	_tcf_ife_cleanup(a, bind);
390 	spin_unlock_bh(&ife->tcf_lock);
391 }
392 
393 /* under ife->tcf_lock for existing action */
394 static int populate_metalist(struct tcf_ife_info *ife, struct nlattr **tb,
395 			     bool exists)
396 {
397 	int len = 0;
398 	int rc = 0;
399 	int i = 0;
400 	void *val;
401 
402 	for (i = 1; i < max_metacnt; i++) {
403 		if (tb[i]) {
404 			val = nla_data(tb[i]);
405 			len = nla_len(tb[i]);
406 
407 			rc = load_metaops_and_vet(ife, i, val, len, exists);
408 			if (rc != 0)
409 				return rc;
410 
411 			rc = add_metainfo(ife, i, val, len, exists);
412 			if (rc)
413 				return rc;
414 		}
415 	}
416 
417 	return rc;
418 }
419 
420 static int tcf_ife_init(struct net *net, struct nlattr *nla,
421 			struct nlattr *est, struct tc_action **a,
422 			int ovr, int bind)
423 {
424 	struct tc_action_net *tn = net_generic(net, ife_net_id);
425 	struct nlattr *tb[TCA_IFE_MAX + 1];
426 	struct nlattr *tb2[IFE_META_MAX + 1];
427 	struct tcf_ife_info *ife;
428 	struct tc_ife *parm;
429 	u16 ife_type = 0;
430 	u8 *daddr = NULL;
431 	u8 *saddr = NULL;
432 	bool exists = false;
433 	int ret = 0;
434 	int err;
435 
436 	err = nla_parse_nested(tb, TCA_IFE_MAX, nla, ife_policy);
437 	if (err < 0)
438 		return err;
439 
440 	if (!tb[TCA_IFE_PARMS])
441 		return -EINVAL;
442 
443 	parm = nla_data(tb[TCA_IFE_PARMS]);
444 
445 	exists = tcf_hash_check(tn, parm->index, a, bind);
446 	if (exists && bind)
447 		return 0;
448 
449 	if (parm->flags & IFE_ENCODE) {
450 		/* Until we get issued the ethertype, we cant have
451 		 * a default..
452 		**/
453 		if (!tb[TCA_IFE_TYPE]) {
454 			if (exists)
455 				tcf_hash_release(*a, bind);
456 			pr_info("You MUST pass etherype for encoding\n");
457 			return -EINVAL;
458 		}
459 	}
460 
461 	if (!exists) {
462 		ret = tcf_hash_create(tn, parm->index, est, a, &act_ife_ops,
463 				      bind, false);
464 		if (ret)
465 			return ret;
466 		ret = ACT_P_CREATED;
467 	} else {
468 		tcf_hash_release(*a, bind);
469 		if (!ovr)
470 			return -EEXIST;
471 	}
472 
473 	ife = to_ife(*a);
474 	ife->flags = parm->flags;
475 
476 	if (parm->flags & IFE_ENCODE) {
477 		ife_type = nla_get_u16(tb[TCA_IFE_TYPE]);
478 		if (tb[TCA_IFE_DMAC])
479 			daddr = nla_data(tb[TCA_IFE_DMAC]);
480 		if (tb[TCA_IFE_SMAC])
481 			saddr = nla_data(tb[TCA_IFE_SMAC]);
482 	}
483 
484 	if (exists)
485 		spin_lock_bh(&ife->tcf_lock);
486 	ife->tcf_action = parm->action;
487 
488 	if (parm->flags & IFE_ENCODE) {
489 		if (daddr)
490 			ether_addr_copy(ife->eth_dst, daddr);
491 		else
492 			eth_zero_addr(ife->eth_dst);
493 
494 		if (saddr)
495 			ether_addr_copy(ife->eth_src, saddr);
496 		else
497 			eth_zero_addr(ife->eth_src);
498 
499 		ife->eth_type = ife_type;
500 	}
501 
502 	if (ret == ACT_P_CREATED)
503 		INIT_LIST_HEAD(&ife->metalist);
504 
505 	if (tb[TCA_IFE_METALST]) {
506 		err = nla_parse_nested(tb2, IFE_META_MAX, tb[TCA_IFE_METALST],
507 				       NULL);
508 		if (err) {
509 metadata_parse_err:
510 			if (exists)
511 				tcf_hash_release(*a, bind);
512 			if (ret == ACT_P_CREATED)
513 				_tcf_ife_cleanup(*a, bind);
514 
515 			if (exists)
516 				spin_unlock_bh(&ife->tcf_lock);
517 			return err;
518 		}
519 
520 		err = populate_metalist(ife, tb2, exists);
521 		if (err)
522 			goto metadata_parse_err;
523 
524 	} else {
525 		/* if no passed metadata allow list or passed allow-all
526 		 * then here we process by adding as many supported metadatum
527 		 * as we can. You better have at least one else we are
528 		 * going to bail out
529 		 */
530 		err = use_all_metadata(ife);
531 		if (err) {
532 			if (ret == ACT_P_CREATED)
533 				_tcf_ife_cleanup(*a, bind);
534 
535 			if (exists)
536 				spin_unlock_bh(&ife->tcf_lock);
537 			return err;
538 		}
539 	}
540 
541 	if (exists)
542 		spin_unlock_bh(&ife->tcf_lock);
543 
544 	if (ret == ACT_P_CREATED)
545 		tcf_hash_insert(tn, *a);
546 
547 	return ret;
548 }
549 
550 static int tcf_ife_dump(struct sk_buff *skb, struct tc_action *a, int bind,
551 			int ref)
552 {
553 	unsigned char *b = skb_tail_pointer(skb);
554 	struct tcf_ife_info *ife = to_ife(a);
555 	struct tc_ife opt = {
556 		.index = ife->tcf_index,
557 		.refcnt = ife->tcf_refcnt - ref,
558 		.bindcnt = ife->tcf_bindcnt - bind,
559 		.action = ife->tcf_action,
560 		.flags = ife->flags,
561 	};
562 	struct tcf_t t;
563 
564 	if (nla_put(skb, TCA_IFE_PARMS, sizeof(opt), &opt))
565 		goto nla_put_failure;
566 
567 	tcf_tm_dump(&t, &ife->tcf_tm);
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 = to_ife(a);
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 	tcf_lastuse_update(&ife->tcf_tm);
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 = to_ife(a);
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 	tcf_lastuse_update(&ife->tcf_tm);
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 = to_ife(a);
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 	tcf_lastuse_update(&ife->tcf_tm);
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 			  const struct tc_action_ops *ops)
824 {
825 	struct tc_action_net *tn = net_generic(net, ife_net_id);
826 
827 	return tcf_generic_walker(tn, skb, cb, type, ops);
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 	.size =	sizeof(struct tcf_ife_info),
848 };
849 
850 static __net_init int ife_init_net(struct net *net)
851 {
852 	struct tc_action_net *tn = net_generic(net, ife_net_id);
853 
854 	return tc_action_net_init(tn, &act_ife_ops, IFE_TAB_MASK);
855 }
856 
857 static void __net_exit ife_exit_net(struct net *net)
858 {
859 	struct tc_action_net *tn = net_generic(net, ife_net_id);
860 
861 	tc_action_net_exit(tn);
862 }
863 
864 static struct pernet_operations ife_net_ops = {
865 	.init = ife_init_net,
866 	.exit = ife_exit_net,
867 	.id   = &ife_net_id,
868 	.size = sizeof(struct tc_action_net),
869 };
870 
871 static int __init ife_init_module(void)
872 {
873 	return tcf_register_action(&act_ife_ops, &ife_net_ops);
874 }
875 
876 static void __exit ife_cleanup_module(void)
877 {
878 	tcf_unregister_action(&act_ife_ops, &ife_net_ops);
879 }
880 
881 module_init(ife_init_module);
882 module_exit(ife_cleanup_module);
883 
884 MODULE_AUTHOR("Jamal Hadi Salim(2015)");
885 MODULE_DESCRIPTION("Inter-FE LFB action");
886 MODULE_LICENSE("GPL");
887