xref: /openbmc/linux/net/sched/act_ife.c (revision 9854518e)
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)
110 {
111 	mi->metaval = kmemdup(metaval, sizeof(u32), GFP_KERNEL);
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)
120 {
121 	mi->metaval = kmemdup(metaval, sizeof(u16), GFP_KERNEL);
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
244 */
245 static int load_metaops_and_vet(struct tcf_ife_info *ife, u32 metaid,
246 				void *val, int len)
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 		spin_unlock_bh(&ife->tcf_lock);
255 		rtnl_unlock();
256 		request_module("ifemeta%u", metaid);
257 		rtnl_lock();
258 		spin_lock_bh(&ife->tcf_lock);
259 		ops = find_ife_oplist(metaid);
260 #endif
261 	}
262 
263 	if (ops) {
264 		ret = 0;
265 		if (len)
266 			ret = ife_validate_metatype(ops, val, len);
267 
268 		module_put(ops->owner);
269 	}
270 
271 	return ret;
272 }
273 
274 /* called when adding new meta information
275  * under ife->tcf_lock
276 */
277 static int add_metainfo(struct tcf_ife_info *ife, u32 metaid, void *metaval,
278 			int len)
279 {
280 	struct tcf_meta_info *mi = NULL;
281 	struct tcf_meta_ops *ops = find_ife_oplist(metaid);
282 	int ret = 0;
283 
284 	if (!ops)
285 		return -ENOENT;
286 
287 	mi = kzalloc(sizeof(*mi), GFP_KERNEL);
288 	if (!mi) {
289 		/*put back what find_ife_oplist took */
290 		module_put(ops->owner);
291 		return -ENOMEM;
292 	}
293 
294 	mi->metaid = metaid;
295 	mi->ops = ops;
296 	if (len > 0) {
297 		ret = ops->alloc(mi, metaval);
298 		if (ret != 0) {
299 			kfree(mi);
300 			module_put(ops->owner);
301 			return ret;
302 		}
303 	}
304 
305 	list_add_tail(&mi->metalist, &ife->metalist);
306 
307 	return ret;
308 }
309 
310 static int use_all_metadata(struct tcf_ife_info *ife)
311 {
312 	struct tcf_meta_ops *o;
313 	int rc = 0;
314 	int installed = 0;
315 
316 	list_for_each_entry(o, &ifeoplist, list) {
317 		rc = add_metainfo(ife, o->metaid, NULL, 0);
318 		if (rc == 0)
319 			installed += 1;
320 	}
321 
322 	if (installed)
323 		return 0;
324 	else
325 		return -EINVAL;
326 }
327 
328 static int dump_metalist(struct sk_buff *skb, struct tcf_ife_info *ife)
329 {
330 	struct tcf_meta_info *e;
331 	struct nlattr *nest;
332 	unsigned char *b = skb_tail_pointer(skb);
333 	int total_encoded = 0;
334 
335 	/*can only happen on decode */
336 	if (list_empty(&ife->metalist))
337 		return 0;
338 
339 	nest = nla_nest_start(skb, TCA_IFE_METALST);
340 	if (!nest)
341 		goto out_nlmsg_trim;
342 
343 	list_for_each_entry(e, &ife->metalist, metalist) {
344 		if (!e->ops->get(skb, e))
345 			total_encoded += 1;
346 	}
347 
348 	if (!total_encoded)
349 		goto out_nlmsg_trim;
350 
351 	nla_nest_end(skb, nest);
352 
353 	return 0;
354 
355 out_nlmsg_trim:
356 	nlmsg_trim(skb, b);
357 	return -1;
358 }
359 
360 /* under ife->tcf_lock */
361 static void _tcf_ife_cleanup(struct tc_action *a, int bind)
362 {
363 	struct tcf_ife_info *ife = a->priv;
364 	struct tcf_meta_info *e, *n;
365 
366 	list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
367 		module_put(e->ops->owner);
368 		list_del(&e->metalist);
369 		if (e->metaval) {
370 			if (e->ops->release)
371 				e->ops->release(e);
372 			else
373 				kfree(e->metaval);
374 		}
375 		kfree(e);
376 	}
377 }
378 
379 static void tcf_ife_cleanup(struct tc_action *a, int bind)
380 {
381 	struct tcf_ife_info *ife = a->priv;
382 
383 	spin_lock_bh(&ife->tcf_lock);
384 	_tcf_ife_cleanup(a, bind);
385 	spin_unlock_bh(&ife->tcf_lock);
386 }
387 
388 /* under ife->tcf_lock */
389 static int populate_metalist(struct tcf_ife_info *ife, struct nlattr **tb)
390 {
391 	int len = 0;
392 	int rc = 0;
393 	int i = 0;
394 	void *val;
395 
396 	for (i = 1; i < max_metacnt; i++) {
397 		if (tb[i]) {
398 			val = nla_data(tb[i]);
399 			len = nla_len(tb[i]);
400 
401 			rc = load_metaops_and_vet(ife, i, val, len);
402 			if (rc != 0)
403 				return rc;
404 
405 			rc = add_metainfo(ife, i, val, len);
406 			if (rc)
407 				return rc;
408 		}
409 	}
410 
411 	return rc;
412 }
413 
414 static int tcf_ife_init(struct net *net, struct nlattr *nla,
415 			struct nlattr *est, struct tc_action *a,
416 			int ovr, int bind)
417 {
418 	struct tc_action_net *tn = net_generic(net, ife_net_id);
419 	struct nlattr *tb[TCA_IFE_MAX + 1];
420 	struct nlattr *tb2[IFE_META_MAX + 1];
421 	struct tcf_ife_info *ife;
422 	struct tc_ife *parm;
423 	u16 ife_type = 0;
424 	u8 *daddr = NULL;
425 	u8 *saddr = NULL;
426 	int ret = 0;
427 	int err;
428 
429 	err = nla_parse_nested(tb, TCA_IFE_MAX, nla, ife_policy);
430 	if (err < 0)
431 		return err;
432 
433 	if (!tb[TCA_IFE_PARMS])
434 		return -EINVAL;
435 
436 	parm = nla_data(tb[TCA_IFE_PARMS]);
437 
438 	if (parm->flags & IFE_ENCODE) {
439 		/* Until we get issued the ethertype, we cant have
440 		 * a default..
441 		**/
442 		if (!tb[TCA_IFE_TYPE]) {
443 			pr_info("You MUST pass etherype for encoding\n");
444 			return -EINVAL;
445 		}
446 	}
447 
448 	if (!tcf_hash_check(tn, parm->index, a, bind)) {
449 		ret = tcf_hash_create(tn, parm->index, est, a, sizeof(*ife),
450 				      bind, false);
451 		if (ret)
452 			return ret;
453 		ret = ACT_P_CREATED;
454 	} else {
455 		if (bind)	/* dont override defaults */
456 			return 0;
457 		tcf_hash_release(a, bind);
458 		if (!ovr)
459 			return -EEXIST;
460 	}
461 
462 	ife = to_ife(a);
463 	ife->flags = parm->flags;
464 
465 	if (parm->flags & IFE_ENCODE) {
466 		ife_type = nla_get_u16(tb[TCA_IFE_TYPE]);
467 		if (tb[TCA_IFE_DMAC])
468 			daddr = nla_data(tb[TCA_IFE_DMAC]);
469 		if (tb[TCA_IFE_SMAC])
470 			saddr = nla_data(tb[TCA_IFE_SMAC]);
471 	}
472 
473 	spin_lock_bh(&ife->tcf_lock);
474 	ife->tcf_action = parm->action;
475 
476 	if (parm->flags & IFE_ENCODE) {
477 		if (daddr)
478 			ether_addr_copy(ife->eth_dst, daddr);
479 		else
480 			eth_zero_addr(ife->eth_dst);
481 
482 		if (saddr)
483 			ether_addr_copy(ife->eth_src, saddr);
484 		else
485 			eth_zero_addr(ife->eth_src);
486 
487 		ife->eth_type = ife_type;
488 	}
489 
490 	if (ret == ACT_P_CREATED)
491 		INIT_LIST_HEAD(&ife->metalist);
492 
493 	if (tb[TCA_IFE_METALST]) {
494 		err = nla_parse_nested(tb2, IFE_META_MAX, tb[TCA_IFE_METALST],
495 				       NULL);
496 		if (err) {
497 metadata_parse_err:
498 			if (ret == ACT_P_CREATED)
499 				_tcf_ife_cleanup(a, bind);
500 
501 			spin_unlock_bh(&ife->tcf_lock);
502 			return err;
503 		}
504 
505 		err = populate_metalist(ife, tb2);
506 		if (err)
507 			goto metadata_parse_err;
508 
509 	} else {
510 		/* if no passed metadata allow list or passed allow-all
511 		 * then here we process by adding as many supported metadatum
512 		 * as we can. You better have at least one else we are
513 		 * going to bail out
514 		 */
515 		err = use_all_metadata(ife);
516 		if (err) {
517 			if (ret == ACT_P_CREATED)
518 				_tcf_ife_cleanup(a, bind);
519 
520 			spin_unlock_bh(&ife->tcf_lock);
521 			return err;
522 		}
523 	}
524 
525 	spin_unlock_bh(&ife->tcf_lock);
526 
527 	if (ret == ACT_P_CREATED)
528 		tcf_hash_insert(tn, a);
529 
530 	return ret;
531 }
532 
533 static int tcf_ife_dump(struct sk_buff *skb, struct tc_action *a, int bind,
534 			int ref)
535 {
536 	unsigned char *b = skb_tail_pointer(skb);
537 	struct tcf_ife_info *ife = a->priv;
538 	struct tc_ife opt = {
539 		.index = ife->tcf_index,
540 		.refcnt = ife->tcf_refcnt - ref,
541 		.bindcnt = ife->tcf_bindcnt - bind,
542 		.action = ife->tcf_action,
543 		.flags = ife->flags,
544 	};
545 	struct tcf_t t;
546 
547 	if (nla_put(skb, TCA_IFE_PARMS, sizeof(opt), &opt))
548 		goto nla_put_failure;
549 
550 	t.install = jiffies_to_clock_t(jiffies - ife->tcf_tm.install);
551 	t.lastuse = jiffies_to_clock_t(jiffies - ife->tcf_tm.lastuse);
552 	t.expires = jiffies_to_clock_t(ife->tcf_tm.expires);
553 	if (nla_put_64bit(skb, TCA_IFE_TM, sizeof(t), &t, TCA_IFE_PAD))
554 		goto nla_put_failure;
555 
556 	if (!is_zero_ether_addr(ife->eth_dst)) {
557 		if (nla_put(skb, TCA_IFE_DMAC, ETH_ALEN, ife->eth_dst))
558 			goto nla_put_failure;
559 	}
560 
561 	if (!is_zero_ether_addr(ife->eth_src)) {
562 		if (nla_put(skb, TCA_IFE_SMAC, ETH_ALEN, ife->eth_src))
563 			goto nla_put_failure;
564 	}
565 
566 	if (nla_put(skb, TCA_IFE_TYPE, 2, &ife->eth_type))
567 		goto nla_put_failure;
568 
569 	if (dump_metalist(skb, ife)) {
570 		/*ignore failure to dump metalist */
571 		pr_info("Failed to dump metalist\n");
572 	}
573 
574 	return skb->len;
575 
576 nla_put_failure:
577 	nlmsg_trim(skb, b);
578 	return -1;
579 }
580 
581 int find_decode_metaid(struct sk_buff *skb, struct tcf_ife_info *ife,
582 		       u16 metaid, u16 mlen, void *mdata)
583 {
584 	struct tcf_meta_info *e;
585 
586 	/* XXX: use hash to speed up */
587 	list_for_each_entry(e, &ife->metalist, metalist) {
588 		if (metaid == e->metaid) {
589 			if (e->ops) {
590 				/* We check for decode presence already */
591 				return e->ops->decode(skb, mdata, mlen);
592 			}
593 		}
594 	}
595 
596 	return 0;
597 }
598 
599 struct ifeheadr {
600 	__be16 metalen;
601 	u8 tlv_data[];
602 };
603 
604 struct meta_tlvhdr {
605 	__be16 type;
606 	__be16 len;
607 };
608 
609 static int tcf_ife_decode(struct sk_buff *skb, const struct tc_action *a,
610 			  struct tcf_result *res)
611 {
612 	struct tcf_ife_info *ife = a->priv;
613 	int action = ife->tcf_action;
614 	struct ifeheadr *ifehdr = (struct ifeheadr *)skb->data;
615 	u16 ifehdrln = ifehdr->metalen;
616 	struct meta_tlvhdr *tlv = (struct meta_tlvhdr *)(ifehdr->tlv_data);
617 
618 	spin_lock(&ife->tcf_lock);
619 	bstats_update(&ife->tcf_bstats, skb);
620 	ife->tcf_tm.lastuse = jiffies;
621 	spin_unlock(&ife->tcf_lock);
622 
623 	ifehdrln = ntohs(ifehdrln);
624 	if (unlikely(!pskb_may_pull(skb, ifehdrln))) {
625 		spin_lock(&ife->tcf_lock);
626 		ife->tcf_qstats.drops++;
627 		spin_unlock(&ife->tcf_lock);
628 		return TC_ACT_SHOT;
629 	}
630 
631 	skb_set_mac_header(skb, ifehdrln);
632 	__skb_pull(skb, ifehdrln);
633 	skb->protocol = eth_type_trans(skb, skb->dev);
634 	ifehdrln -= IFE_METAHDRLEN;
635 
636 	while (ifehdrln > 0) {
637 		u8 *tlvdata = (u8 *)tlv;
638 		u16 mtype = tlv->type;
639 		u16 mlen = tlv->len;
640 
641 		mtype = ntohs(mtype);
642 		mlen = ntohs(mlen);
643 
644 		if (find_decode_metaid(skb, ife, mtype, (mlen - 4),
645 				       (void *)(tlvdata + 4))) {
646 			/* abuse overlimits to count when we receive metadata
647 			 * but dont have an ops for it
648 			 */
649 			pr_info_ratelimited("Unknown metaid %d alnlen %d\n",
650 					    mtype, mlen);
651 			ife->tcf_qstats.overlimits++;
652 		}
653 
654 		tlvdata += mlen;
655 		ifehdrln -= mlen;
656 		tlv = (struct meta_tlvhdr *)tlvdata;
657 	}
658 
659 	skb_reset_network_header(skb);
660 	return action;
661 }
662 
663 /*XXX: check if we can do this at install time instead of current
664  * send data path
665 **/
666 static int ife_get_sz(struct sk_buff *skb, struct tcf_ife_info *ife)
667 {
668 	struct tcf_meta_info *e, *n;
669 	int tot_run_sz = 0, run_sz = 0;
670 
671 	list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
672 		if (e->ops->check_presence) {
673 			run_sz = e->ops->check_presence(skb, e);
674 			tot_run_sz += run_sz;
675 		}
676 	}
677 
678 	return tot_run_sz;
679 }
680 
681 static int tcf_ife_encode(struct sk_buff *skb, const struct tc_action *a,
682 			  struct tcf_result *res)
683 {
684 	struct tcf_ife_info *ife = a->priv;
685 	int action = ife->tcf_action;
686 	struct ethhdr *oethh;	/* outer ether header */
687 	struct ethhdr *iethh;	/* inner eth header */
688 	struct tcf_meta_info *e;
689 	/*
690 	   OUTERHDR:TOTMETALEN:{TLVHDR:Metadatum:TLVHDR..}:ORIGDATA
691 	   where ORIGDATA = original ethernet header ...
692 	 */
693 	u16 metalen = ife_get_sz(skb, ife);
694 	int hdrm = metalen + skb->dev->hard_header_len + IFE_METAHDRLEN;
695 	unsigned int skboff = skb->dev->hard_header_len;
696 	u32 at = G_TC_AT(skb->tc_verd);
697 	int new_len = skb->len + hdrm;
698 	bool exceed_mtu = false;
699 	int err;
700 
701 	if (at & AT_EGRESS) {
702 		if (new_len > skb->dev->mtu)
703 			exceed_mtu = true;
704 	}
705 
706 	spin_lock(&ife->tcf_lock);
707 	bstats_update(&ife->tcf_bstats, skb);
708 	ife->tcf_tm.lastuse = jiffies;
709 
710 	if (!metalen) {		/* no metadata to send */
711 		/* abuse overlimits to count when we allow packet
712 		 * with no metadata
713 		 */
714 		ife->tcf_qstats.overlimits++;
715 		spin_unlock(&ife->tcf_lock);
716 		return action;
717 	}
718 	/* could be stupid policy setup or mtu config
719 	 * so lets be conservative.. */
720 	if ((action == TC_ACT_SHOT) || exceed_mtu) {
721 		ife->tcf_qstats.drops++;
722 		spin_unlock(&ife->tcf_lock);
723 		return TC_ACT_SHOT;
724 	}
725 
726 	iethh = eth_hdr(skb);
727 
728 	err = skb_cow_head(skb, hdrm);
729 	if (unlikely(err)) {
730 		ife->tcf_qstats.drops++;
731 		spin_unlock(&ife->tcf_lock);
732 		return TC_ACT_SHOT;
733 	}
734 
735 	if (!(at & AT_EGRESS))
736 		skb_push(skb, skb->dev->hard_header_len);
737 
738 	__skb_push(skb, hdrm);
739 	memcpy(skb->data, iethh, skb->mac_len);
740 	skb_reset_mac_header(skb);
741 	oethh = eth_hdr(skb);
742 
743 	/*total metadata length */
744 	metalen += IFE_METAHDRLEN;
745 	metalen = htons(metalen);
746 	memcpy((skb->data + skboff), &metalen, IFE_METAHDRLEN);
747 	skboff += IFE_METAHDRLEN;
748 
749 	/* XXX: we dont have a clever way of telling encode to
750 	 * not repeat some of the computations that are done by
751 	 * ops->presence_check...
752 	 */
753 	list_for_each_entry(e, &ife->metalist, metalist) {
754 		if (e->ops->encode) {
755 			err = e->ops->encode(skb, (void *)(skb->data + skboff),
756 					     e);
757 		}
758 		if (err < 0) {
759 			/* too corrupt to keep around if overwritten */
760 			ife->tcf_qstats.drops++;
761 			spin_unlock(&ife->tcf_lock);
762 			return TC_ACT_SHOT;
763 		}
764 		skboff += err;
765 	}
766 
767 	if (!is_zero_ether_addr(ife->eth_src))
768 		ether_addr_copy(oethh->h_source, ife->eth_src);
769 	else
770 		ether_addr_copy(oethh->h_source, iethh->h_source);
771 	if (!is_zero_ether_addr(ife->eth_dst))
772 		ether_addr_copy(oethh->h_dest, ife->eth_dst);
773 	else
774 		ether_addr_copy(oethh->h_dest, iethh->h_dest);
775 	oethh->h_proto = htons(ife->eth_type);
776 
777 	if (!(at & AT_EGRESS))
778 		skb_pull(skb, skb->dev->hard_header_len);
779 
780 	spin_unlock(&ife->tcf_lock);
781 
782 	return action;
783 }
784 
785 static int tcf_ife_act(struct sk_buff *skb, const struct tc_action *a,
786 		       struct tcf_result *res)
787 {
788 	struct tcf_ife_info *ife = a->priv;
789 
790 	if (ife->flags & IFE_ENCODE)
791 		return tcf_ife_encode(skb, a, res);
792 
793 	if (!(ife->flags & IFE_ENCODE))
794 		return tcf_ife_decode(skb, a, res);
795 
796 	pr_info_ratelimited("unknown failure(policy neither de/encode\n");
797 	spin_lock(&ife->tcf_lock);
798 	bstats_update(&ife->tcf_bstats, skb);
799 	ife->tcf_tm.lastuse = jiffies;
800 	ife->tcf_qstats.drops++;
801 	spin_unlock(&ife->tcf_lock);
802 
803 	return TC_ACT_SHOT;
804 }
805 
806 static int tcf_ife_walker(struct net *net, struct sk_buff *skb,
807 			  struct netlink_callback *cb, int type,
808 			  struct tc_action *a)
809 {
810 	struct tc_action_net *tn = net_generic(net, ife_net_id);
811 
812 	return tcf_generic_walker(tn, skb, cb, type, a);
813 }
814 
815 static int tcf_ife_search(struct net *net, struct tc_action *a, u32 index)
816 {
817 	struct tc_action_net *tn = net_generic(net, ife_net_id);
818 
819 	return tcf_hash_search(tn, a, index);
820 }
821 
822 static struct tc_action_ops act_ife_ops = {
823 	.kind = "ife",
824 	.type = TCA_ACT_IFE,
825 	.owner = THIS_MODULE,
826 	.act = tcf_ife_act,
827 	.dump = tcf_ife_dump,
828 	.cleanup = tcf_ife_cleanup,
829 	.init = tcf_ife_init,
830 	.walk = tcf_ife_walker,
831 	.lookup = tcf_ife_search,
832 };
833 
834 static __net_init int ife_init_net(struct net *net)
835 {
836 	struct tc_action_net *tn = net_generic(net, ife_net_id);
837 
838 	return tc_action_net_init(tn, &act_ife_ops, IFE_TAB_MASK);
839 }
840 
841 static void __net_exit ife_exit_net(struct net *net)
842 {
843 	struct tc_action_net *tn = net_generic(net, ife_net_id);
844 
845 	tc_action_net_exit(tn);
846 }
847 
848 static struct pernet_operations ife_net_ops = {
849 	.init = ife_init_net,
850 	.exit = ife_exit_net,
851 	.id   = &ife_net_id,
852 	.size = sizeof(struct tc_action_net),
853 };
854 
855 static int __init ife_init_module(void)
856 {
857 	return tcf_register_action(&act_ife_ops, &ife_net_ops);
858 }
859 
860 static void __exit ife_cleanup_module(void)
861 {
862 	tcf_unregister_action(&act_ife_ops, &ife_net_ops);
863 }
864 
865 module_init(ife_init_module);
866 module_exit(ife_cleanup_module);
867 
868 MODULE_AUTHOR("Jamal Hadi Salim(2015)");
869 MODULE_DESCRIPTION("Inter-FE LFB action");
870 MODULE_LICENSE("GPL");
871