xref: /openbmc/linux/net/sched/act_ife.c (revision da2ef666)
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 #include <net/ife.h>
36 
37 static unsigned int ife_net_id;
38 static int max_metacnt = IFE_META_MAX + 1;
39 static struct tc_action_ops act_ife_ops;
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 int ife_encode_meta_u16(u16 metaval, void *skbdata, struct tcf_meta_info *mi)
49 {
50 	u16 edata = 0;
51 
52 	if (mi->metaval)
53 		edata = *(u16 *)mi->metaval;
54 	else if (metaval)
55 		edata = metaval;
56 
57 	if (!edata) /* will not encode */
58 		return 0;
59 
60 	edata = htons(edata);
61 	return ife_tlv_meta_encode(skbdata, mi->metaid, 2, &edata);
62 }
63 EXPORT_SYMBOL_GPL(ife_encode_meta_u16);
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_check_meta_u16(u16 metaval, struct tcf_meta_info *mi)
84 {
85 	if (metaval || mi->metaval)
86 		return 8; /* T+L+(V) == 2+2+(2+2bytepad) */
87 
88 	return 0;
89 }
90 EXPORT_SYMBOL_GPL(ife_check_meta_u16);
91 
92 int ife_encode_meta_u32(u32 metaval, void *skbdata, struct tcf_meta_info *mi)
93 {
94 	u32 edata = metaval;
95 
96 	if (mi->metaval)
97 		edata = *(u32 *)mi->metaval;
98 	else if (metaval)
99 		edata = metaval;
100 
101 	if (!edata) /* will not encode */
102 		return 0;
103 
104 	edata = htonl(edata);
105 	return ife_tlv_meta_encode(skbdata, mi->metaid, 4, &edata);
106 }
107 EXPORT_SYMBOL_GPL(ife_encode_meta_u32);
108 
109 int ife_get_meta_u16(struct sk_buff *skb, struct tcf_meta_info *mi)
110 {
111 	if (mi->metaval)
112 		return nla_put_u16(skb, mi->metaid, *(u16 *)mi->metaval);
113 	else
114 		return nla_put(skb, mi->metaid, 0, NULL);
115 }
116 EXPORT_SYMBOL_GPL(ife_get_meta_u16);
117 
118 int ife_alloc_meta_u32(struct tcf_meta_info *mi, void *metaval, gfp_t gfp)
119 {
120 	mi->metaval = kmemdup(metaval, sizeof(u32), gfp);
121 	if (!mi->metaval)
122 		return -ENOMEM;
123 
124 	return 0;
125 }
126 EXPORT_SYMBOL_GPL(ife_alloc_meta_u32);
127 
128 int ife_alloc_meta_u16(struct tcf_meta_info *mi, void *metaval, gfp_t gfp)
129 {
130 	mi->metaval = kmemdup(metaval, sizeof(u16), gfp);
131 	if (!mi->metaval)
132 		return -ENOMEM;
133 
134 	return 0;
135 }
136 EXPORT_SYMBOL_GPL(ife_alloc_meta_u16);
137 
138 void ife_release_meta_gen(struct tcf_meta_info *mi)
139 {
140 	kfree(mi->metaval);
141 }
142 EXPORT_SYMBOL_GPL(ife_release_meta_gen);
143 
144 int ife_validate_meta_u32(void *val, int len)
145 {
146 	if (len == sizeof(u32))
147 		return 0;
148 
149 	return -EINVAL;
150 }
151 EXPORT_SYMBOL_GPL(ife_validate_meta_u32);
152 
153 int ife_validate_meta_u16(void *val, int len)
154 {
155 	/* length will not include padding */
156 	if (len == sizeof(u16))
157 		return 0;
158 
159 	return -EINVAL;
160 }
161 EXPORT_SYMBOL_GPL(ife_validate_meta_u16);
162 
163 static LIST_HEAD(ifeoplist);
164 static DEFINE_RWLOCK(ife_mod_lock);
165 
166 static struct tcf_meta_ops *find_ife_oplist(u16 metaid)
167 {
168 	struct tcf_meta_ops *o;
169 
170 	read_lock(&ife_mod_lock);
171 	list_for_each_entry(o, &ifeoplist, list) {
172 		if (o->metaid == metaid) {
173 			if (!try_module_get(o->owner))
174 				o = NULL;
175 			read_unlock(&ife_mod_lock);
176 			return o;
177 		}
178 	}
179 	read_unlock(&ife_mod_lock);
180 
181 	return NULL;
182 }
183 
184 int register_ife_op(struct tcf_meta_ops *mops)
185 {
186 	struct tcf_meta_ops *m;
187 
188 	if (!mops->metaid || !mops->metatype || !mops->name ||
189 	    !mops->check_presence || !mops->encode || !mops->decode ||
190 	    !mops->get || !mops->alloc)
191 		return -EINVAL;
192 
193 	write_lock(&ife_mod_lock);
194 
195 	list_for_each_entry(m, &ifeoplist, list) {
196 		if (m->metaid == mops->metaid ||
197 		    (strcmp(mops->name, m->name) == 0)) {
198 			write_unlock(&ife_mod_lock);
199 			return -EEXIST;
200 		}
201 	}
202 
203 	if (!mops->release)
204 		mops->release = ife_release_meta_gen;
205 
206 	list_add_tail(&mops->list, &ifeoplist);
207 	write_unlock(&ife_mod_lock);
208 	return 0;
209 }
210 EXPORT_SYMBOL_GPL(unregister_ife_op);
211 
212 int unregister_ife_op(struct tcf_meta_ops *mops)
213 {
214 	struct tcf_meta_ops *m;
215 	int err = -ENOENT;
216 
217 	write_lock(&ife_mod_lock);
218 	list_for_each_entry(m, &ifeoplist, list) {
219 		if (m->metaid == mops->metaid) {
220 			list_del(&mops->list);
221 			err = 0;
222 			break;
223 		}
224 	}
225 	write_unlock(&ife_mod_lock);
226 
227 	return err;
228 }
229 EXPORT_SYMBOL_GPL(register_ife_op);
230 
231 static int ife_validate_metatype(struct tcf_meta_ops *ops, void *val, int len)
232 {
233 	int ret = 0;
234 	/* XXX: unfortunately cant use nla_policy at this point
235 	* because a length of 0 is valid in the case of
236 	* "allow". "use" semantics do enforce for proper
237 	* length and i couldve use nla_policy but it makes it hard
238 	* to use it just for that..
239 	*/
240 	if (ops->validate)
241 		return ops->validate(val, len);
242 
243 	if (ops->metatype == NLA_U32)
244 		ret = ife_validate_meta_u32(val, len);
245 	else if (ops->metatype == NLA_U16)
246 		ret = ife_validate_meta_u16(val, len);
247 
248 	return ret;
249 }
250 
251 #ifdef CONFIG_MODULES
252 static const char *ife_meta_id2name(u32 metaid)
253 {
254 	switch (metaid) {
255 	case IFE_META_SKBMARK:
256 		return "skbmark";
257 	case IFE_META_PRIO:
258 		return "skbprio";
259 	case IFE_META_TCINDEX:
260 		return "tcindex";
261 	default:
262 		return "unknown";
263 	}
264 }
265 #endif
266 
267 /* called when adding new meta information
268 */
269 static int load_metaops_and_vet(u32 metaid, void *val, int len, bool rtnl_held)
270 {
271 	struct tcf_meta_ops *ops = find_ife_oplist(metaid);
272 	int ret = 0;
273 
274 	if (!ops) {
275 		ret = -ENOENT;
276 #ifdef CONFIG_MODULES
277 		if (rtnl_held)
278 			rtnl_unlock();
279 		request_module("ife-meta-%s", ife_meta_id2name(metaid));
280 		if (rtnl_held)
281 			rtnl_lock();
282 		ops = find_ife_oplist(metaid);
283 #endif
284 	}
285 
286 	if (ops) {
287 		ret = 0;
288 		if (len)
289 			ret = ife_validate_metatype(ops, val, len);
290 
291 		module_put(ops->owner);
292 	}
293 
294 	return ret;
295 }
296 
297 /* called when adding new meta information
298 */
299 static int __add_metainfo(const struct tcf_meta_ops *ops,
300 			  struct tcf_ife_info *ife, u32 metaid, void *metaval,
301 			  int len, bool atomic, bool exists)
302 {
303 	struct tcf_meta_info *mi = NULL;
304 	int ret = 0;
305 
306 	mi = kzalloc(sizeof(*mi), atomic ? GFP_ATOMIC : GFP_KERNEL);
307 	if (!mi)
308 		return -ENOMEM;
309 
310 	mi->metaid = metaid;
311 	mi->ops = ops;
312 	if (len > 0) {
313 		ret = ops->alloc(mi, metaval, atomic ? GFP_ATOMIC : GFP_KERNEL);
314 		if (ret != 0) {
315 			kfree(mi);
316 			return ret;
317 		}
318 	}
319 
320 	if (exists)
321 		spin_lock_bh(&ife->tcf_lock);
322 	list_add_tail(&mi->metalist, &ife->metalist);
323 	if (exists)
324 		spin_unlock_bh(&ife->tcf_lock);
325 
326 	return ret;
327 }
328 
329 static int add_metainfo(struct tcf_ife_info *ife, u32 metaid, void *metaval,
330 			int len, bool exists)
331 {
332 	const struct tcf_meta_ops *ops = find_ife_oplist(metaid);
333 	int ret;
334 
335 	if (!ops)
336 		return -ENOENT;
337 	ret = __add_metainfo(ops, ife, metaid, metaval, len, false, exists);
338 	if (ret)
339 		/*put back what find_ife_oplist took */
340 		module_put(ops->owner);
341 	return ret;
342 }
343 
344 static int use_all_metadata(struct tcf_ife_info *ife, bool exists)
345 {
346 	struct tcf_meta_ops *o;
347 	int rc = 0;
348 	int installed = 0;
349 
350 	read_lock(&ife_mod_lock);
351 	list_for_each_entry(o, &ifeoplist, list) {
352 		rc = __add_metainfo(o, ife, o->metaid, NULL, 0, true, exists);
353 		if (rc == 0)
354 			installed += 1;
355 	}
356 	read_unlock(&ife_mod_lock);
357 
358 	if (installed)
359 		return 0;
360 	else
361 		return -EINVAL;
362 }
363 
364 static int dump_metalist(struct sk_buff *skb, struct tcf_ife_info *ife)
365 {
366 	struct tcf_meta_info *e;
367 	struct nlattr *nest;
368 	unsigned char *b = skb_tail_pointer(skb);
369 	int total_encoded = 0;
370 
371 	/*can only happen on decode */
372 	if (list_empty(&ife->metalist))
373 		return 0;
374 
375 	nest = nla_nest_start(skb, TCA_IFE_METALST);
376 	if (!nest)
377 		goto out_nlmsg_trim;
378 
379 	list_for_each_entry(e, &ife->metalist, metalist) {
380 		if (!e->ops->get(skb, e))
381 			total_encoded += 1;
382 	}
383 
384 	if (!total_encoded)
385 		goto out_nlmsg_trim;
386 
387 	nla_nest_end(skb, nest);
388 
389 	return 0;
390 
391 out_nlmsg_trim:
392 	nlmsg_trim(skb, b);
393 	return -1;
394 }
395 
396 /* under ife->tcf_lock */
397 static void _tcf_ife_cleanup(struct tc_action *a)
398 {
399 	struct tcf_ife_info *ife = to_ife(a);
400 	struct tcf_meta_info *e, *n;
401 
402 	list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
403 		module_put(e->ops->owner);
404 		list_del(&e->metalist);
405 		if (e->metaval) {
406 			if (e->ops->release)
407 				e->ops->release(e);
408 			else
409 				kfree(e->metaval);
410 		}
411 		kfree(e);
412 	}
413 }
414 
415 static void tcf_ife_cleanup(struct tc_action *a)
416 {
417 	struct tcf_ife_info *ife = to_ife(a);
418 	struct tcf_ife_params *p;
419 
420 	spin_lock_bh(&ife->tcf_lock);
421 	_tcf_ife_cleanup(a);
422 	spin_unlock_bh(&ife->tcf_lock);
423 
424 	p = rcu_dereference_protected(ife->params, 1);
425 	if (p)
426 		kfree_rcu(p, rcu);
427 }
428 
429 static int populate_metalist(struct tcf_ife_info *ife, struct nlattr **tb,
430 			     bool exists, bool rtnl_held)
431 {
432 	int len = 0;
433 	int rc = 0;
434 	int i = 0;
435 	void *val;
436 
437 	for (i = 1; i < max_metacnt; i++) {
438 		if (tb[i]) {
439 			val = nla_data(tb[i]);
440 			len = nla_len(tb[i]);
441 
442 			rc = load_metaops_and_vet(i, val, len, rtnl_held);
443 			if (rc != 0)
444 				return rc;
445 
446 			rc = add_metainfo(ife, i, val, len, exists);
447 			if (rc)
448 				return rc;
449 		}
450 	}
451 
452 	return rc;
453 }
454 
455 static int tcf_ife_init(struct net *net, struct nlattr *nla,
456 			struct nlattr *est, struct tc_action **a,
457 			int ovr, int bind, bool rtnl_held,
458 			struct netlink_ext_ack *extack)
459 {
460 	struct tc_action_net *tn = net_generic(net, ife_net_id);
461 	struct nlattr *tb[TCA_IFE_MAX + 1];
462 	struct nlattr *tb2[IFE_META_MAX + 1];
463 	struct tcf_ife_params *p;
464 	struct tcf_ife_info *ife;
465 	u16 ife_type = ETH_P_IFE;
466 	struct tc_ife *parm;
467 	u8 *daddr = NULL;
468 	u8 *saddr = NULL;
469 	bool exists = false;
470 	int ret = 0;
471 	int err;
472 
473 	err = nla_parse_nested(tb, TCA_IFE_MAX, nla, ife_policy, NULL);
474 	if (err < 0)
475 		return err;
476 
477 	if (!tb[TCA_IFE_PARMS])
478 		return -EINVAL;
479 
480 	parm = nla_data(tb[TCA_IFE_PARMS]);
481 
482 	/* IFE_DECODE is 0 and indicates the opposite of IFE_ENCODE because
483 	 * they cannot run as the same time. Check on all other values which
484 	 * are not supported right now.
485 	 */
486 	if (parm->flags & ~IFE_ENCODE)
487 		return -EINVAL;
488 
489 	p = kzalloc(sizeof(*p), GFP_KERNEL);
490 	if (!p)
491 		return -ENOMEM;
492 
493 	err = tcf_idr_check_alloc(tn, &parm->index, a, bind);
494 	if (err < 0) {
495 		kfree(p);
496 		return err;
497 	}
498 	exists = err;
499 	if (exists && bind) {
500 		kfree(p);
501 		return 0;
502 	}
503 
504 	if (!exists) {
505 		ret = tcf_idr_create(tn, parm->index, est, a, &act_ife_ops,
506 				     bind, true);
507 		if (ret) {
508 			tcf_idr_cleanup(tn, parm->index);
509 			kfree(p);
510 			return ret;
511 		}
512 		ret = ACT_P_CREATED;
513 	} else if (!ovr) {
514 		tcf_idr_release(*a, bind);
515 		kfree(p);
516 		return -EEXIST;
517 	}
518 
519 	ife = to_ife(*a);
520 	p->flags = parm->flags;
521 
522 	if (parm->flags & IFE_ENCODE) {
523 		if (tb[TCA_IFE_TYPE])
524 			ife_type = nla_get_u16(tb[TCA_IFE_TYPE]);
525 		if (tb[TCA_IFE_DMAC])
526 			daddr = nla_data(tb[TCA_IFE_DMAC]);
527 		if (tb[TCA_IFE_SMAC])
528 			saddr = nla_data(tb[TCA_IFE_SMAC]);
529 	}
530 
531 	if (parm->flags & IFE_ENCODE) {
532 		if (daddr)
533 			ether_addr_copy(p->eth_dst, daddr);
534 		else
535 			eth_zero_addr(p->eth_dst);
536 
537 		if (saddr)
538 			ether_addr_copy(p->eth_src, saddr);
539 		else
540 			eth_zero_addr(p->eth_src);
541 
542 		p->eth_type = ife_type;
543 	}
544 
545 
546 	if (ret == ACT_P_CREATED)
547 		INIT_LIST_HEAD(&ife->metalist);
548 
549 	if (tb[TCA_IFE_METALST]) {
550 		err = nla_parse_nested(tb2, IFE_META_MAX, tb[TCA_IFE_METALST],
551 				       NULL, NULL);
552 		if (err) {
553 metadata_parse_err:
554 			tcf_idr_release(*a, bind);
555 			kfree(p);
556 			return err;
557 		}
558 
559 		err = populate_metalist(ife, tb2, exists, rtnl_held);
560 		if (err)
561 			goto metadata_parse_err;
562 
563 	} else {
564 		/* if no passed metadata allow list or passed allow-all
565 		 * then here we process by adding as many supported metadatum
566 		 * as we can. You better have at least one else we are
567 		 * going to bail out
568 		 */
569 		err = use_all_metadata(ife, exists);
570 		if (err) {
571 			tcf_idr_release(*a, bind);
572 			kfree(p);
573 			return err;
574 		}
575 	}
576 
577 	if (exists)
578 		spin_lock_bh(&ife->tcf_lock);
579 	ife->tcf_action = parm->action;
580 	/* protected by tcf_lock when modifying existing action */
581 	rcu_swap_protected(ife->params, p, 1);
582 
583 	if (exists)
584 		spin_unlock_bh(&ife->tcf_lock);
585 	if (p)
586 		kfree_rcu(p, rcu);
587 
588 	if (ret == ACT_P_CREATED)
589 		tcf_idr_insert(tn, *a);
590 
591 	return ret;
592 }
593 
594 static int tcf_ife_dump(struct sk_buff *skb, struct tc_action *a, int bind,
595 			int ref)
596 {
597 	unsigned char *b = skb_tail_pointer(skb);
598 	struct tcf_ife_info *ife = to_ife(a);
599 	struct tcf_ife_params *p;
600 	struct tc_ife opt = {
601 		.index = ife->tcf_index,
602 		.refcnt = refcount_read(&ife->tcf_refcnt) - ref,
603 		.bindcnt = atomic_read(&ife->tcf_bindcnt) - bind,
604 	};
605 	struct tcf_t t;
606 
607 	spin_lock_bh(&ife->tcf_lock);
608 	opt.action = ife->tcf_action;
609 	p = rcu_dereference_protected(ife->params,
610 				      lockdep_is_held(&ife->tcf_lock));
611 	opt.flags = p->flags;
612 
613 	if (nla_put(skb, TCA_IFE_PARMS, sizeof(opt), &opt))
614 		goto nla_put_failure;
615 
616 	tcf_tm_dump(&t, &ife->tcf_tm);
617 	if (nla_put_64bit(skb, TCA_IFE_TM, sizeof(t), &t, TCA_IFE_PAD))
618 		goto nla_put_failure;
619 
620 	if (!is_zero_ether_addr(p->eth_dst)) {
621 		if (nla_put(skb, TCA_IFE_DMAC, ETH_ALEN, p->eth_dst))
622 			goto nla_put_failure;
623 	}
624 
625 	if (!is_zero_ether_addr(p->eth_src)) {
626 		if (nla_put(skb, TCA_IFE_SMAC, ETH_ALEN, p->eth_src))
627 			goto nla_put_failure;
628 	}
629 
630 	if (nla_put(skb, TCA_IFE_TYPE, 2, &p->eth_type))
631 		goto nla_put_failure;
632 
633 	if (dump_metalist(skb, ife)) {
634 		/*ignore failure to dump metalist */
635 		pr_info("Failed to dump metalist\n");
636 	}
637 
638 	spin_unlock_bh(&ife->tcf_lock);
639 	return skb->len;
640 
641 nla_put_failure:
642 	spin_unlock_bh(&ife->tcf_lock);
643 	nlmsg_trim(skb, b);
644 	return -1;
645 }
646 
647 static int find_decode_metaid(struct sk_buff *skb, struct tcf_ife_info *ife,
648 			      u16 metaid, u16 mlen, void *mdata)
649 {
650 	struct tcf_meta_info *e;
651 
652 	/* XXX: use hash to speed up */
653 	list_for_each_entry(e, &ife->metalist, metalist) {
654 		if (metaid == e->metaid) {
655 			if (e->ops) {
656 				/* We check for decode presence already */
657 				return e->ops->decode(skb, mdata, mlen);
658 			}
659 		}
660 	}
661 
662 	return -ENOENT;
663 }
664 
665 static int tcf_ife_decode(struct sk_buff *skb, const struct tc_action *a,
666 			  struct tcf_result *res)
667 {
668 	struct tcf_ife_info *ife = to_ife(a);
669 	int action = ife->tcf_action;
670 	u8 *ifehdr_end;
671 	u8 *tlv_data;
672 	u16 metalen;
673 
674 	bstats_cpu_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
675 	tcf_lastuse_update(&ife->tcf_tm);
676 
677 	if (skb_at_tc_ingress(skb))
678 		skb_push(skb, skb->dev->hard_header_len);
679 
680 	tlv_data = ife_decode(skb, &metalen);
681 	if (unlikely(!tlv_data)) {
682 		qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
683 		return TC_ACT_SHOT;
684 	}
685 
686 	ifehdr_end = tlv_data + metalen;
687 	for (; tlv_data < ifehdr_end; tlv_data = ife_tlv_meta_next(tlv_data)) {
688 		u8 *curr_data;
689 		u16 mtype;
690 		u16 dlen;
691 
692 		curr_data = ife_tlv_meta_decode(tlv_data, ifehdr_end, &mtype,
693 						&dlen, NULL);
694 		if (!curr_data) {
695 			qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
696 			return TC_ACT_SHOT;
697 		}
698 
699 		if (find_decode_metaid(skb, ife, mtype, dlen, curr_data)) {
700 			/* abuse overlimits to count when we receive metadata
701 			 * but dont have an ops for it
702 			 */
703 			pr_info_ratelimited("Unknown metaid %d dlen %d\n",
704 					    mtype, dlen);
705 			qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
706 		}
707 	}
708 
709 	if (WARN_ON(tlv_data != ifehdr_end)) {
710 		qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
711 		return TC_ACT_SHOT;
712 	}
713 
714 	skb->protocol = eth_type_trans(skb, skb->dev);
715 	skb_reset_network_header(skb);
716 
717 	return action;
718 }
719 
720 /*XXX: check if we can do this at install time instead of current
721  * send data path
722 **/
723 static int ife_get_sz(struct sk_buff *skb, struct tcf_ife_info *ife)
724 {
725 	struct tcf_meta_info *e, *n;
726 	int tot_run_sz = 0, run_sz = 0;
727 
728 	list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
729 		if (e->ops->check_presence) {
730 			run_sz = e->ops->check_presence(skb, e);
731 			tot_run_sz += run_sz;
732 		}
733 	}
734 
735 	return tot_run_sz;
736 }
737 
738 static int tcf_ife_encode(struct sk_buff *skb, const struct tc_action *a,
739 			  struct tcf_result *res, struct tcf_ife_params *p)
740 {
741 	struct tcf_ife_info *ife = to_ife(a);
742 	int action = ife->tcf_action;
743 	struct ethhdr *oethh;	/* outer ether header */
744 	struct tcf_meta_info *e;
745 	/*
746 	   OUTERHDR:TOTMETALEN:{TLVHDR:Metadatum:TLVHDR..}:ORIGDATA
747 	   where ORIGDATA = original ethernet header ...
748 	 */
749 	u16 metalen = ife_get_sz(skb, ife);
750 	int hdrm = metalen + skb->dev->hard_header_len + IFE_METAHDRLEN;
751 	unsigned int skboff = 0;
752 	int new_len = skb->len + hdrm;
753 	bool exceed_mtu = false;
754 	void *ife_meta;
755 	int err = 0;
756 
757 	if (!skb_at_tc_ingress(skb)) {
758 		if (new_len > skb->dev->mtu)
759 			exceed_mtu = true;
760 	}
761 
762 	bstats_cpu_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
763 	tcf_lastuse_update(&ife->tcf_tm);
764 
765 	if (!metalen) {		/* no metadata to send */
766 		/* abuse overlimits to count when we allow packet
767 		 * with no metadata
768 		 */
769 		qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
770 		return action;
771 	}
772 	/* could be stupid policy setup or mtu config
773 	 * so lets be conservative.. */
774 	if ((action == TC_ACT_SHOT) || exceed_mtu) {
775 		qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
776 		return TC_ACT_SHOT;
777 	}
778 
779 	if (skb_at_tc_ingress(skb))
780 		skb_push(skb, skb->dev->hard_header_len);
781 
782 	ife_meta = ife_encode(skb, metalen);
783 
784 	spin_lock(&ife->tcf_lock);
785 
786 	/* XXX: we dont have a clever way of telling encode to
787 	 * not repeat some of the computations that are done by
788 	 * ops->presence_check...
789 	 */
790 	list_for_each_entry(e, &ife->metalist, metalist) {
791 		if (e->ops->encode) {
792 			err = e->ops->encode(skb, (void *)(ife_meta + skboff),
793 					     e);
794 		}
795 		if (err < 0) {
796 			/* too corrupt to keep around if overwritten */
797 			spin_unlock(&ife->tcf_lock);
798 			qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
799 			return TC_ACT_SHOT;
800 		}
801 		skboff += err;
802 	}
803 	spin_unlock(&ife->tcf_lock);
804 	oethh = (struct ethhdr *)skb->data;
805 
806 	if (!is_zero_ether_addr(p->eth_src))
807 		ether_addr_copy(oethh->h_source, p->eth_src);
808 	if (!is_zero_ether_addr(p->eth_dst))
809 		ether_addr_copy(oethh->h_dest, p->eth_dst);
810 	oethh->h_proto = htons(p->eth_type);
811 
812 	if (skb_at_tc_ingress(skb))
813 		skb_pull(skb, skb->dev->hard_header_len);
814 
815 	return action;
816 }
817 
818 static int tcf_ife_act(struct sk_buff *skb, const struct tc_action *a,
819 		       struct tcf_result *res)
820 {
821 	struct tcf_ife_info *ife = to_ife(a);
822 	struct tcf_ife_params *p;
823 	int ret;
824 
825 	p = rcu_dereference_bh(ife->params);
826 	if (p->flags & IFE_ENCODE) {
827 		ret = tcf_ife_encode(skb, a, res, p);
828 		return ret;
829 	}
830 
831 	return tcf_ife_decode(skb, a, res);
832 }
833 
834 static int tcf_ife_walker(struct net *net, struct sk_buff *skb,
835 			  struct netlink_callback *cb, int type,
836 			  const struct tc_action_ops *ops,
837 			  struct netlink_ext_ack *extack)
838 {
839 	struct tc_action_net *tn = net_generic(net, ife_net_id);
840 
841 	return tcf_generic_walker(tn, skb, cb, type, ops, extack);
842 }
843 
844 static int tcf_ife_search(struct net *net, struct tc_action **a, u32 index,
845 			  struct netlink_ext_ack *extack)
846 {
847 	struct tc_action_net *tn = net_generic(net, ife_net_id);
848 
849 	return tcf_idr_search(tn, a, index);
850 }
851 
852 static struct tc_action_ops act_ife_ops = {
853 	.kind = "ife",
854 	.type = TCA_ACT_IFE,
855 	.owner = THIS_MODULE,
856 	.act = tcf_ife_act,
857 	.dump = tcf_ife_dump,
858 	.cleanup = tcf_ife_cleanup,
859 	.init = tcf_ife_init,
860 	.walk = tcf_ife_walker,
861 	.lookup = tcf_ife_search,
862 	.size =	sizeof(struct tcf_ife_info),
863 };
864 
865 static __net_init int ife_init_net(struct net *net)
866 {
867 	struct tc_action_net *tn = net_generic(net, ife_net_id);
868 
869 	return tc_action_net_init(tn, &act_ife_ops);
870 }
871 
872 static void __net_exit ife_exit_net(struct list_head *net_list)
873 {
874 	tc_action_net_exit(net_list, ife_net_id);
875 }
876 
877 static struct pernet_operations ife_net_ops = {
878 	.init = ife_init_net,
879 	.exit_batch = ife_exit_net,
880 	.id   = &ife_net_id,
881 	.size = sizeof(struct tc_action_net),
882 };
883 
884 static int __init ife_init_module(void)
885 {
886 	return tcf_register_action(&act_ife_ops, &ife_net_ops);
887 }
888 
889 static void __exit ife_cleanup_module(void)
890 {
891 	tcf_unregister_action(&act_ife_ops, &ife_net_ops);
892 }
893 
894 module_init(ife_init_module);
895 module_exit(ife_cleanup_module);
896 
897 MODULE_AUTHOR("Jamal Hadi Salim(2015)");
898 MODULE_DESCRIPTION("Inter-FE LFB action");
899 MODULE_LICENSE("GPL");
900