xref: /openbmc/linux/net/sched/act_ife.c (revision 911b8eac)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * net/sched/ife.c	Inter-FE action based on ForCES WG InterFE LFB
4  *
5  *		Refer to:
6  *		draft-ietf-forces-interfelfb-03
7  *		and
8  *		netdev01 paper:
9  *		"Distributing Linux Traffic Control Classifier-Action
10  *		Subsystem"
11  *		Authors: Jamal Hadi Salim and Damascene M. Joachimpillai
12  *
13  * copyright Jamal Hadi Salim (2015)
14 */
15 
16 #include <linux/types.h>
17 #include <linux/kernel.h>
18 #include <linux/string.h>
19 #include <linux/errno.h>
20 #include <linux/skbuff.h>
21 #include <linux/rtnetlink.h>
22 #include <linux/module.h>
23 #include <linux/init.h>
24 #include <net/net_namespace.h>
25 #include <net/netlink.h>
26 #include <net/pkt_sched.h>
27 #include <net/pkt_cls.h>
28 #include <uapi/linux/tc_act/tc_ife.h>
29 #include <net/tc_act/tc_ife.h>
30 #include <linux/etherdevice.h>
31 #include <net/ife.h>
32 
33 static unsigned int ife_net_id;
34 static int max_metacnt = IFE_META_MAX + 1;
35 static struct tc_action_ops act_ife_ops;
36 
37 static const struct nla_policy ife_policy[TCA_IFE_MAX + 1] = {
38 	[TCA_IFE_PARMS] = { .len = sizeof(struct tc_ife)},
39 	[TCA_IFE_DMAC] = { .len = ETH_ALEN},
40 	[TCA_IFE_SMAC] = { .len = ETH_ALEN},
41 	[TCA_IFE_TYPE] = { .type = NLA_U16},
42 };
43 
44 int ife_encode_meta_u16(u16 metaval, void *skbdata, struct tcf_meta_info *mi)
45 {
46 	u16 edata = 0;
47 
48 	if (mi->metaval)
49 		edata = *(u16 *)mi->metaval;
50 	else if (metaval)
51 		edata = metaval;
52 
53 	if (!edata) /* will not encode */
54 		return 0;
55 
56 	edata = htons(edata);
57 	return ife_tlv_meta_encode(skbdata, mi->metaid, 2, &edata);
58 }
59 EXPORT_SYMBOL_GPL(ife_encode_meta_u16);
60 
61 int ife_get_meta_u32(struct sk_buff *skb, struct tcf_meta_info *mi)
62 {
63 	if (mi->metaval)
64 		return nla_put_u32(skb, mi->metaid, *(u32 *)mi->metaval);
65 	else
66 		return nla_put(skb, mi->metaid, 0, NULL);
67 }
68 EXPORT_SYMBOL_GPL(ife_get_meta_u32);
69 
70 int ife_check_meta_u32(u32 metaval, struct tcf_meta_info *mi)
71 {
72 	if (metaval || mi->metaval)
73 		return 8; /* T+L+V == 2+2+4 */
74 
75 	return 0;
76 }
77 EXPORT_SYMBOL_GPL(ife_check_meta_u32);
78 
79 int ife_check_meta_u16(u16 metaval, struct tcf_meta_info *mi)
80 {
81 	if (metaval || mi->metaval)
82 		return 8; /* T+L+(V) == 2+2+(2+2bytepad) */
83 
84 	return 0;
85 }
86 EXPORT_SYMBOL_GPL(ife_check_meta_u16);
87 
88 int ife_encode_meta_u32(u32 metaval, void *skbdata, struct tcf_meta_info *mi)
89 {
90 	u32 edata = metaval;
91 
92 	if (mi->metaval)
93 		edata = *(u32 *)mi->metaval;
94 	else if (metaval)
95 		edata = metaval;
96 
97 	if (!edata) /* will not encode */
98 		return 0;
99 
100 	edata = htonl(edata);
101 	return ife_tlv_meta_encode(skbdata, mi->metaid, 4, &edata);
102 }
103 EXPORT_SYMBOL_GPL(ife_encode_meta_u32);
104 
105 int ife_get_meta_u16(struct sk_buff *skb, struct tcf_meta_info *mi)
106 {
107 	if (mi->metaval)
108 		return nla_put_u16(skb, mi->metaid, *(u16 *)mi->metaval);
109 	else
110 		return nla_put(skb, mi->metaid, 0, NULL);
111 }
112 EXPORT_SYMBOL_GPL(ife_get_meta_u16);
113 
114 int ife_alloc_meta_u32(struct tcf_meta_info *mi, void *metaval, gfp_t gfp)
115 {
116 	mi->metaval = kmemdup(metaval, sizeof(u32), gfp);
117 	if (!mi->metaval)
118 		return -ENOMEM;
119 
120 	return 0;
121 }
122 EXPORT_SYMBOL_GPL(ife_alloc_meta_u32);
123 
124 int ife_alloc_meta_u16(struct tcf_meta_info *mi, void *metaval, gfp_t gfp)
125 {
126 	mi->metaval = kmemdup(metaval, sizeof(u16), gfp);
127 	if (!mi->metaval)
128 		return -ENOMEM;
129 
130 	return 0;
131 }
132 EXPORT_SYMBOL_GPL(ife_alloc_meta_u16);
133 
134 void ife_release_meta_gen(struct tcf_meta_info *mi)
135 {
136 	kfree(mi->metaval);
137 }
138 EXPORT_SYMBOL_GPL(ife_release_meta_gen);
139 
140 int ife_validate_meta_u32(void *val, int len)
141 {
142 	if (len == sizeof(u32))
143 		return 0;
144 
145 	return -EINVAL;
146 }
147 EXPORT_SYMBOL_GPL(ife_validate_meta_u32);
148 
149 int ife_validate_meta_u16(void *val, int len)
150 {
151 	/* length will not include padding */
152 	if (len == sizeof(u16))
153 		return 0;
154 
155 	return -EINVAL;
156 }
157 EXPORT_SYMBOL_GPL(ife_validate_meta_u16);
158 
159 static LIST_HEAD(ifeoplist);
160 static DEFINE_RWLOCK(ife_mod_lock);
161 
162 static struct tcf_meta_ops *find_ife_oplist(u16 metaid)
163 {
164 	struct tcf_meta_ops *o;
165 
166 	read_lock(&ife_mod_lock);
167 	list_for_each_entry(o, &ifeoplist, list) {
168 		if (o->metaid == metaid) {
169 			if (!try_module_get(o->owner))
170 				o = NULL;
171 			read_unlock(&ife_mod_lock);
172 			return o;
173 		}
174 	}
175 	read_unlock(&ife_mod_lock);
176 
177 	return NULL;
178 }
179 
180 int register_ife_op(struct tcf_meta_ops *mops)
181 {
182 	struct tcf_meta_ops *m;
183 
184 	if (!mops->metaid || !mops->metatype || !mops->name ||
185 	    !mops->check_presence || !mops->encode || !mops->decode ||
186 	    !mops->get || !mops->alloc)
187 		return -EINVAL;
188 
189 	write_lock(&ife_mod_lock);
190 
191 	list_for_each_entry(m, &ifeoplist, list) {
192 		if (m->metaid == mops->metaid ||
193 		    (strcmp(mops->name, m->name) == 0)) {
194 			write_unlock(&ife_mod_lock);
195 			return -EEXIST;
196 		}
197 	}
198 
199 	if (!mops->release)
200 		mops->release = ife_release_meta_gen;
201 
202 	list_add_tail(&mops->list, &ifeoplist);
203 	write_unlock(&ife_mod_lock);
204 	return 0;
205 }
206 EXPORT_SYMBOL_GPL(unregister_ife_op);
207 
208 int unregister_ife_op(struct tcf_meta_ops *mops)
209 {
210 	struct tcf_meta_ops *m;
211 	int err = -ENOENT;
212 
213 	write_lock(&ife_mod_lock);
214 	list_for_each_entry(m, &ifeoplist, list) {
215 		if (m->metaid == mops->metaid) {
216 			list_del(&mops->list);
217 			err = 0;
218 			break;
219 		}
220 	}
221 	write_unlock(&ife_mod_lock);
222 
223 	return err;
224 }
225 EXPORT_SYMBOL_GPL(register_ife_op);
226 
227 static int ife_validate_metatype(struct tcf_meta_ops *ops, void *val, int len)
228 {
229 	int ret = 0;
230 	/* XXX: unfortunately cant use nla_policy at this point
231 	* because a length of 0 is valid in the case of
232 	* "allow". "use" semantics do enforce for proper
233 	* length and i couldve use nla_policy but it makes it hard
234 	* to use it just for that..
235 	*/
236 	if (ops->validate)
237 		return ops->validate(val, len);
238 
239 	if (ops->metatype == NLA_U32)
240 		ret = ife_validate_meta_u32(val, len);
241 	else if (ops->metatype == NLA_U16)
242 		ret = ife_validate_meta_u16(val, len);
243 
244 	return ret;
245 }
246 
247 #ifdef CONFIG_MODULES
248 static const char *ife_meta_id2name(u32 metaid)
249 {
250 	switch (metaid) {
251 	case IFE_META_SKBMARK:
252 		return "skbmark";
253 	case IFE_META_PRIO:
254 		return "skbprio";
255 	case IFE_META_TCINDEX:
256 		return "tcindex";
257 	default:
258 		return "unknown";
259 	}
260 }
261 #endif
262 
263 /* called when adding new meta information
264 */
265 static int load_metaops_and_vet(u32 metaid, void *val, int len, bool rtnl_held)
266 {
267 	struct tcf_meta_ops *ops = find_ife_oplist(metaid);
268 	int ret = 0;
269 
270 	if (!ops) {
271 		ret = -ENOENT;
272 #ifdef CONFIG_MODULES
273 		if (rtnl_held)
274 			rtnl_unlock();
275 		request_module("ife-meta-%s", ife_meta_id2name(metaid));
276 		if (rtnl_held)
277 			rtnl_lock();
278 		ops = find_ife_oplist(metaid);
279 #endif
280 	}
281 
282 	if (ops) {
283 		ret = 0;
284 		if (len)
285 			ret = ife_validate_metatype(ops, val, len);
286 
287 		module_put(ops->owner);
288 	}
289 
290 	return ret;
291 }
292 
293 /* called when adding new meta information
294 */
295 static int __add_metainfo(const struct tcf_meta_ops *ops,
296 			  struct tcf_ife_info *ife, u32 metaid, void *metaval,
297 			  int len, bool atomic, bool exists)
298 {
299 	struct tcf_meta_info *mi = NULL;
300 	int ret = 0;
301 
302 	mi = kzalloc(sizeof(*mi), atomic ? GFP_ATOMIC : GFP_KERNEL);
303 	if (!mi)
304 		return -ENOMEM;
305 
306 	mi->metaid = metaid;
307 	mi->ops = ops;
308 	if (len > 0) {
309 		ret = ops->alloc(mi, metaval, atomic ? GFP_ATOMIC : GFP_KERNEL);
310 		if (ret != 0) {
311 			kfree(mi);
312 			return ret;
313 		}
314 	}
315 
316 	if (exists)
317 		spin_lock_bh(&ife->tcf_lock);
318 	list_add_tail(&mi->metalist, &ife->metalist);
319 	if (exists)
320 		spin_unlock_bh(&ife->tcf_lock);
321 
322 	return ret;
323 }
324 
325 static int add_metainfo_and_get_ops(const struct tcf_meta_ops *ops,
326 				    struct tcf_ife_info *ife, u32 metaid,
327 				    bool exists)
328 {
329 	int ret;
330 
331 	if (!try_module_get(ops->owner))
332 		return -ENOENT;
333 	ret = __add_metainfo(ops, ife, metaid, NULL, 0, true, exists);
334 	if (ret)
335 		module_put(ops->owner);
336 	return ret;
337 }
338 
339 static int add_metainfo(struct tcf_ife_info *ife, u32 metaid, void *metaval,
340 			int len, bool exists)
341 {
342 	const struct tcf_meta_ops *ops = find_ife_oplist(metaid);
343 	int ret;
344 
345 	if (!ops)
346 		return -ENOENT;
347 	ret = __add_metainfo(ops, ife, metaid, metaval, len, false, exists);
348 	if (ret)
349 		/*put back what find_ife_oplist took */
350 		module_put(ops->owner);
351 	return ret;
352 }
353 
354 static int use_all_metadata(struct tcf_ife_info *ife, bool exists)
355 {
356 	struct tcf_meta_ops *o;
357 	int rc = 0;
358 	int installed = 0;
359 
360 	read_lock(&ife_mod_lock);
361 	list_for_each_entry(o, &ifeoplist, list) {
362 		rc = add_metainfo_and_get_ops(o, ife, o->metaid, exists);
363 		if (rc == 0)
364 			installed += 1;
365 	}
366 	read_unlock(&ife_mod_lock);
367 
368 	if (installed)
369 		return 0;
370 	else
371 		return -EINVAL;
372 }
373 
374 static int dump_metalist(struct sk_buff *skb, struct tcf_ife_info *ife)
375 {
376 	struct tcf_meta_info *e;
377 	struct nlattr *nest;
378 	unsigned char *b = skb_tail_pointer(skb);
379 	int total_encoded = 0;
380 
381 	/*can only happen on decode */
382 	if (list_empty(&ife->metalist))
383 		return 0;
384 
385 	nest = nla_nest_start_noflag(skb, TCA_IFE_METALST);
386 	if (!nest)
387 		goto out_nlmsg_trim;
388 
389 	list_for_each_entry(e, &ife->metalist, metalist) {
390 		if (!e->ops->get(skb, e))
391 			total_encoded += 1;
392 	}
393 
394 	if (!total_encoded)
395 		goto out_nlmsg_trim;
396 
397 	nla_nest_end(skb, nest);
398 
399 	return 0;
400 
401 out_nlmsg_trim:
402 	nlmsg_trim(skb, b);
403 	return -1;
404 }
405 
406 /* under ife->tcf_lock */
407 static void _tcf_ife_cleanup(struct tc_action *a)
408 {
409 	struct tcf_ife_info *ife = to_ife(a);
410 	struct tcf_meta_info *e, *n;
411 
412 	list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
413 		list_del(&e->metalist);
414 		if (e->metaval) {
415 			if (e->ops->release)
416 				e->ops->release(e);
417 			else
418 				kfree(e->metaval);
419 		}
420 		module_put(e->ops->owner);
421 		kfree(e);
422 	}
423 }
424 
425 static void tcf_ife_cleanup(struct tc_action *a)
426 {
427 	struct tcf_ife_info *ife = to_ife(a);
428 	struct tcf_ife_params *p;
429 
430 	spin_lock_bh(&ife->tcf_lock);
431 	_tcf_ife_cleanup(a);
432 	spin_unlock_bh(&ife->tcf_lock);
433 
434 	p = rcu_dereference_protected(ife->params, 1);
435 	if (p)
436 		kfree_rcu(p, rcu);
437 }
438 
439 static int load_metalist(struct nlattr **tb, bool rtnl_held)
440 {
441 	int i;
442 
443 	for (i = 1; i < max_metacnt; i++) {
444 		if (tb[i]) {
445 			void *val = nla_data(tb[i]);
446 			int len = nla_len(tb[i]);
447 			int rc;
448 
449 			rc = load_metaops_and_vet(i, val, len, rtnl_held);
450 			if (rc != 0)
451 				return rc;
452 		}
453 	}
454 
455 	return 0;
456 }
457 
458 static int populate_metalist(struct tcf_ife_info *ife, struct nlattr **tb,
459 			     bool exists, bool rtnl_held)
460 {
461 	int len = 0;
462 	int rc = 0;
463 	int i = 0;
464 	void *val;
465 
466 	for (i = 1; i < max_metacnt; i++) {
467 		if (tb[i]) {
468 			val = nla_data(tb[i]);
469 			len = nla_len(tb[i]);
470 
471 			rc = add_metainfo(ife, i, val, len, exists);
472 			if (rc)
473 				return rc;
474 		}
475 	}
476 
477 	return rc;
478 }
479 
480 static int tcf_ife_init(struct net *net, struct nlattr *nla,
481 			struct nlattr *est, struct tc_action **a,
482 			int ovr, int bind, bool rtnl_held,
483 			struct tcf_proto *tp, u32 flags,
484 			struct netlink_ext_ack *extack)
485 {
486 	struct tc_action_net *tn = net_generic(net, ife_net_id);
487 	struct nlattr *tb[TCA_IFE_MAX + 1];
488 	struct nlattr *tb2[IFE_META_MAX + 1];
489 	struct tcf_chain *goto_ch = NULL;
490 	struct tcf_ife_params *p;
491 	struct tcf_ife_info *ife;
492 	u16 ife_type = ETH_P_IFE;
493 	struct tc_ife *parm;
494 	u8 *daddr = NULL;
495 	u8 *saddr = NULL;
496 	bool exists = false;
497 	int ret = 0;
498 	u32 index;
499 	int err;
500 
501 	if (!nla) {
502 		NL_SET_ERR_MSG_MOD(extack, "IFE requires attributes to be passed");
503 		return -EINVAL;
504 	}
505 
506 	err = nla_parse_nested_deprecated(tb, TCA_IFE_MAX, nla, ife_policy,
507 					  NULL);
508 	if (err < 0)
509 		return err;
510 
511 	if (!tb[TCA_IFE_PARMS])
512 		return -EINVAL;
513 
514 	parm = nla_data(tb[TCA_IFE_PARMS]);
515 
516 	/* IFE_DECODE is 0 and indicates the opposite of IFE_ENCODE because
517 	 * they cannot run as the same time. Check on all other values which
518 	 * are not supported right now.
519 	 */
520 	if (parm->flags & ~IFE_ENCODE)
521 		return -EINVAL;
522 
523 	p = kzalloc(sizeof(*p), GFP_KERNEL);
524 	if (!p)
525 		return -ENOMEM;
526 
527 	if (tb[TCA_IFE_METALST]) {
528 		err = nla_parse_nested_deprecated(tb2, IFE_META_MAX,
529 						  tb[TCA_IFE_METALST], NULL,
530 						  NULL);
531 		if (err) {
532 			kfree(p);
533 			return err;
534 		}
535 		err = load_metalist(tb2, rtnl_held);
536 		if (err) {
537 			kfree(p);
538 			return err;
539 		}
540 	}
541 
542 	index = parm->index;
543 	err = tcf_idr_check_alloc(tn, &index, a, bind);
544 	if (err < 0) {
545 		kfree(p);
546 		return err;
547 	}
548 	exists = err;
549 	if (exists && bind) {
550 		kfree(p);
551 		return 0;
552 	}
553 
554 	if (!exists) {
555 		ret = tcf_idr_create(tn, index, est, a, &act_ife_ops,
556 				     bind, true, 0);
557 		if (ret) {
558 			tcf_idr_cleanup(tn, index);
559 			kfree(p);
560 			return ret;
561 		}
562 		ret = ACT_P_CREATED;
563 	} else if (!ovr) {
564 		tcf_idr_release(*a, bind);
565 		kfree(p);
566 		return -EEXIST;
567 	}
568 
569 	ife = to_ife(*a);
570 	if (ret == ACT_P_CREATED)
571 		INIT_LIST_HEAD(&ife->metalist);
572 
573 	err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
574 	if (err < 0)
575 		goto release_idr;
576 
577 	p->flags = parm->flags;
578 
579 	if (parm->flags & IFE_ENCODE) {
580 		if (tb[TCA_IFE_TYPE])
581 			ife_type = nla_get_u16(tb[TCA_IFE_TYPE]);
582 		if (tb[TCA_IFE_DMAC])
583 			daddr = nla_data(tb[TCA_IFE_DMAC]);
584 		if (tb[TCA_IFE_SMAC])
585 			saddr = nla_data(tb[TCA_IFE_SMAC]);
586 	}
587 
588 	if (parm->flags & IFE_ENCODE) {
589 		if (daddr)
590 			ether_addr_copy(p->eth_dst, daddr);
591 		else
592 			eth_zero_addr(p->eth_dst);
593 
594 		if (saddr)
595 			ether_addr_copy(p->eth_src, saddr);
596 		else
597 			eth_zero_addr(p->eth_src);
598 
599 		p->eth_type = ife_type;
600 	}
601 
602 	if (tb[TCA_IFE_METALST]) {
603 		err = populate_metalist(ife, tb2, exists, rtnl_held);
604 		if (err)
605 			goto metadata_parse_err;
606 	} else {
607 		/* if no passed metadata allow list or passed allow-all
608 		 * then here we process by adding as many supported metadatum
609 		 * as we can. You better have at least one else we are
610 		 * going to bail out
611 		 */
612 		err = use_all_metadata(ife, exists);
613 		if (err)
614 			goto metadata_parse_err;
615 	}
616 
617 	if (exists)
618 		spin_lock_bh(&ife->tcf_lock);
619 	/* protected by tcf_lock when modifying existing action */
620 	goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
621 	p = rcu_replace_pointer(ife->params, p, 1);
622 
623 	if (exists)
624 		spin_unlock_bh(&ife->tcf_lock);
625 	if (goto_ch)
626 		tcf_chain_put_by_act(goto_ch);
627 	if (p)
628 		kfree_rcu(p, rcu);
629 
630 	if (ret == ACT_P_CREATED)
631 		tcf_idr_insert(tn, *a);
632 
633 	return ret;
634 metadata_parse_err:
635 	if (goto_ch)
636 		tcf_chain_put_by_act(goto_ch);
637 release_idr:
638 	kfree(p);
639 	tcf_idr_release(*a, bind);
640 	return err;
641 }
642 
643 static int tcf_ife_dump(struct sk_buff *skb, struct tc_action *a, int bind,
644 			int ref)
645 {
646 	unsigned char *b = skb_tail_pointer(skb);
647 	struct tcf_ife_info *ife = to_ife(a);
648 	struct tcf_ife_params *p;
649 	struct tc_ife opt = {
650 		.index = ife->tcf_index,
651 		.refcnt = refcount_read(&ife->tcf_refcnt) - ref,
652 		.bindcnt = atomic_read(&ife->tcf_bindcnt) - bind,
653 	};
654 	struct tcf_t t;
655 
656 	spin_lock_bh(&ife->tcf_lock);
657 	opt.action = ife->tcf_action;
658 	p = rcu_dereference_protected(ife->params,
659 				      lockdep_is_held(&ife->tcf_lock));
660 	opt.flags = p->flags;
661 
662 	if (nla_put(skb, TCA_IFE_PARMS, sizeof(opt), &opt))
663 		goto nla_put_failure;
664 
665 	tcf_tm_dump(&t, &ife->tcf_tm);
666 	if (nla_put_64bit(skb, TCA_IFE_TM, sizeof(t), &t, TCA_IFE_PAD))
667 		goto nla_put_failure;
668 
669 	if (!is_zero_ether_addr(p->eth_dst)) {
670 		if (nla_put(skb, TCA_IFE_DMAC, ETH_ALEN, p->eth_dst))
671 			goto nla_put_failure;
672 	}
673 
674 	if (!is_zero_ether_addr(p->eth_src)) {
675 		if (nla_put(skb, TCA_IFE_SMAC, ETH_ALEN, p->eth_src))
676 			goto nla_put_failure;
677 	}
678 
679 	if (nla_put(skb, TCA_IFE_TYPE, 2, &p->eth_type))
680 		goto nla_put_failure;
681 
682 	if (dump_metalist(skb, ife)) {
683 		/*ignore failure to dump metalist */
684 		pr_info("Failed to dump metalist\n");
685 	}
686 
687 	spin_unlock_bh(&ife->tcf_lock);
688 	return skb->len;
689 
690 nla_put_failure:
691 	spin_unlock_bh(&ife->tcf_lock);
692 	nlmsg_trim(skb, b);
693 	return -1;
694 }
695 
696 static int find_decode_metaid(struct sk_buff *skb, struct tcf_ife_info *ife,
697 			      u16 metaid, u16 mlen, void *mdata)
698 {
699 	struct tcf_meta_info *e;
700 
701 	/* XXX: use hash to speed up */
702 	list_for_each_entry(e, &ife->metalist, metalist) {
703 		if (metaid == e->metaid) {
704 			if (e->ops) {
705 				/* We check for decode presence already */
706 				return e->ops->decode(skb, mdata, mlen);
707 			}
708 		}
709 	}
710 
711 	return -ENOENT;
712 }
713 
714 static int tcf_ife_decode(struct sk_buff *skb, const struct tc_action *a,
715 			  struct tcf_result *res)
716 {
717 	struct tcf_ife_info *ife = to_ife(a);
718 	int action = ife->tcf_action;
719 	u8 *ifehdr_end;
720 	u8 *tlv_data;
721 	u16 metalen;
722 
723 	bstats_cpu_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
724 	tcf_lastuse_update(&ife->tcf_tm);
725 
726 	if (skb_at_tc_ingress(skb))
727 		skb_push(skb, skb->dev->hard_header_len);
728 
729 	tlv_data = ife_decode(skb, &metalen);
730 	if (unlikely(!tlv_data)) {
731 		qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
732 		return TC_ACT_SHOT;
733 	}
734 
735 	ifehdr_end = tlv_data + metalen;
736 	for (; tlv_data < ifehdr_end; tlv_data = ife_tlv_meta_next(tlv_data)) {
737 		u8 *curr_data;
738 		u16 mtype;
739 		u16 dlen;
740 
741 		curr_data = ife_tlv_meta_decode(tlv_data, ifehdr_end, &mtype,
742 						&dlen, NULL);
743 		if (!curr_data) {
744 			qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
745 			return TC_ACT_SHOT;
746 		}
747 
748 		if (find_decode_metaid(skb, ife, mtype, dlen, curr_data)) {
749 			/* abuse overlimits to count when we receive metadata
750 			 * but dont have an ops for it
751 			 */
752 			pr_info_ratelimited("Unknown metaid %d dlen %d\n",
753 					    mtype, dlen);
754 			qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
755 		}
756 	}
757 
758 	if (WARN_ON(tlv_data != ifehdr_end)) {
759 		qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
760 		return TC_ACT_SHOT;
761 	}
762 
763 	skb->protocol = eth_type_trans(skb, skb->dev);
764 	skb_reset_network_header(skb);
765 
766 	return action;
767 }
768 
769 /*XXX: check if we can do this at install time instead of current
770  * send data path
771 **/
772 static int ife_get_sz(struct sk_buff *skb, struct tcf_ife_info *ife)
773 {
774 	struct tcf_meta_info *e, *n;
775 	int tot_run_sz = 0, run_sz = 0;
776 
777 	list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
778 		if (e->ops->check_presence) {
779 			run_sz = e->ops->check_presence(skb, e);
780 			tot_run_sz += run_sz;
781 		}
782 	}
783 
784 	return tot_run_sz;
785 }
786 
787 static int tcf_ife_encode(struct sk_buff *skb, const struct tc_action *a,
788 			  struct tcf_result *res, struct tcf_ife_params *p)
789 {
790 	struct tcf_ife_info *ife = to_ife(a);
791 	int action = ife->tcf_action;
792 	struct ethhdr *oethh;	/* outer ether header */
793 	struct tcf_meta_info *e;
794 	/*
795 	   OUTERHDR:TOTMETALEN:{TLVHDR:Metadatum:TLVHDR..}:ORIGDATA
796 	   where ORIGDATA = original ethernet header ...
797 	 */
798 	u16 metalen = ife_get_sz(skb, ife);
799 	int hdrm = metalen + skb->dev->hard_header_len + IFE_METAHDRLEN;
800 	unsigned int skboff = 0;
801 	int new_len = skb->len + hdrm;
802 	bool exceed_mtu = false;
803 	void *ife_meta;
804 	int err = 0;
805 
806 	if (!skb_at_tc_ingress(skb)) {
807 		if (new_len > skb->dev->mtu)
808 			exceed_mtu = true;
809 	}
810 
811 	bstats_cpu_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
812 	tcf_lastuse_update(&ife->tcf_tm);
813 
814 	if (!metalen) {		/* no metadata to send */
815 		/* abuse overlimits to count when we allow packet
816 		 * with no metadata
817 		 */
818 		qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
819 		return action;
820 	}
821 	/* could be stupid policy setup or mtu config
822 	 * so lets be conservative.. */
823 	if ((action == TC_ACT_SHOT) || exceed_mtu) {
824 		qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
825 		return TC_ACT_SHOT;
826 	}
827 
828 	if (skb_at_tc_ingress(skb))
829 		skb_push(skb, skb->dev->hard_header_len);
830 
831 	ife_meta = ife_encode(skb, metalen);
832 
833 	spin_lock(&ife->tcf_lock);
834 
835 	/* XXX: we dont have a clever way of telling encode to
836 	 * not repeat some of the computations that are done by
837 	 * ops->presence_check...
838 	 */
839 	list_for_each_entry(e, &ife->metalist, metalist) {
840 		if (e->ops->encode) {
841 			err = e->ops->encode(skb, (void *)(ife_meta + skboff),
842 					     e);
843 		}
844 		if (err < 0) {
845 			/* too corrupt to keep around if overwritten */
846 			spin_unlock(&ife->tcf_lock);
847 			qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
848 			return TC_ACT_SHOT;
849 		}
850 		skboff += err;
851 	}
852 	spin_unlock(&ife->tcf_lock);
853 	oethh = (struct ethhdr *)skb->data;
854 
855 	if (!is_zero_ether_addr(p->eth_src))
856 		ether_addr_copy(oethh->h_source, p->eth_src);
857 	if (!is_zero_ether_addr(p->eth_dst))
858 		ether_addr_copy(oethh->h_dest, p->eth_dst);
859 	oethh->h_proto = htons(p->eth_type);
860 
861 	if (skb_at_tc_ingress(skb))
862 		skb_pull(skb, skb->dev->hard_header_len);
863 
864 	return action;
865 }
866 
867 static int tcf_ife_act(struct sk_buff *skb, const struct tc_action *a,
868 		       struct tcf_result *res)
869 {
870 	struct tcf_ife_info *ife = to_ife(a);
871 	struct tcf_ife_params *p;
872 	int ret;
873 
874 	p = rcu_dereference_bh(ife->params);
875 	if (p->flags & IFE_ENCODE) {
876 		ret = tcf_ife_encode(skb, a, res, p);
877 		return ret;
878 	}
879 
880 	return tcf_ife_decode(skb, a, res);
881 }
882 
883 static int tcf_ife_walker(struct net *net, struct sk_buff *skb,
884 			  struct netlink_callback *cb, int type,
885 			  const struct tc_action_ops *ops,
886 			  struct netlink_ext_ack *extack)
887 {
888 	struct tc_action_net *tn = net_generic(net, ife_net_id);
889 
890 	return tcf_generic_walker(tn, skb, cb, type, ops, extack);
891 }
892 
893 static int tcf_ife_search(struct net *net, struct tc_action **a, u32 index)
894 {
895 	struct tc_action_net *tn = net_generic(net, ife_net_id);
896 
897 	return tcf_idr_search(tn, a, index);
898 }
899 
900 static struct tc_action_ops act_ife_ops = {
901 	.kind = "ife",
902 	.id = TCA_ID_IFE,
903 	.owner = THIS_MODULE,
904 	.act = tcf_ife_act,
905 	.dump = tcf_ife_dump,
906 	.cleanup = tcf_ife_cleanup,
907 	.init = tcf_ife_init,
908 	.walk = tcf_ife_walker,
909 	.lookup = tcf_ife_search,
910 	.size =	sizeof(struct tcf_ife_info),
911 };
912 
913 static __net_init int ife_init_net(struct net *net)
914 {
915 	struct tc_action_net *tn = net_generic(net, ife_net_id);
916 
917 	return tc_action_net_init(net, tn, &act_ife_ops);
918 }
919 
920 static void __net_exit ife_exit_net(struct list_head *net_list)
921 {
922 	tc_action_net_exit(net_list, ife_net_id);
923 }
924 
925 static struct pernet_operations ife_net_ops = {
926 	.init = ife_init_net,
927 	.exit_batch = ife_exit_net,
928 	.id   = &ife_net_id,
929 	.size = sizeof(struct tc_action_net),
930 };
931 
932 static int __init ife_init_module(void)
933 {
934 	return tcf_register_action(&act_ife_ops, &ife_net_ops);
935 }
936 
937 static void __exit ife_cleanup_module(void)
938 {
939 	tcf_unregister_action(&act_ife_ops, &ife_net_ops);
940 }
941 
942 module_init(ife_init_module);
943 module_exit(ife_cleanup_module);
944 
945 MODULE_AUTHOR("Jamal Hadi Salim(2015)");
946 MODULE_DESCRIPTION("Inter-FE LFB action");
947 MODULE_LICENSE("GPL");
948