xref: /openbmc/linux/net/sched/act_ife.c (revision 84cb8eb2)
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_and_get_ops(const struct tcf_meta_ops *ops,
330 				    struct tcf_ife_info *ife, u32 metaid,
331 				    bool exists)
332 {
333 	int ret;
334 
335 	if (!try_module_get(ops->owner))
336 		return -ENOENT;
337 	ret = __add_metainfo(ops, ife, metaid, NULL, 0, true, exists);
338 	if (ret)
339 		module_put(ops->owner);
340 	return ret;
341 }
342 
343 static int add_metainfo(struct tcf_ife_info *ife, u32 metaid, void *metaval,
344 			int len, bool exists)
345 {
346 	const struct tcf_meta_ops *ops = find_ife_oplist(metaid);
347 	int ret;
348 
349 	if (!ops)
350 		return -ENOENT;
351 	ret = __add_metainfo(ops, ife, metaid, metaval, len, false, exists);
352 	if (ret)
353 		/*put back what find_ife_oplist took */
354 		module_put(ops->owner);
355 	return ret;
356 }
357 
358 static int use_all_metadata(struct tcf_ife_info *ife, bool exists)
359 {
360 	struct tcf_meta_ops *o;
361 	int rc = 0;
362 	int installed = 0;
363 
364 	read_lock(&ife_mod_lock);
365 	list_for_each_entry(o, &ifeoplist, list) {
366 		rc = add_metainfo_and_get_ops(o, ife, o->metaid, exists);
367 		if (rc == 0)
368 			installed += 1;
369 	}
370 	read_unlock(&ife_mod_lock);
371 
372 	if (installed)
373 		return 0;
374 	else
375 		return -EINVAL;
376 }
377 
378 static int dump_metalist(struct sk_buff *skb, struct tcf_ife_info *ife)
379 {
380 	struct tcf_meta_info *e;
381 	struct nlattr *nest;
382 	unsigned char *b = skb_tail_pointer(skb);
383 	int total_encoded = 0;
384 
385 	/*can only happen on decode */
386 	if (list_empty(&ife->metalist))
387 		return 0;
388 
389 	nest = nla_nest_start(skb, TCA_IFE_METALST);
390 	if (!nest)
391 		goto out_nlmsg_trim;
392 
393 	list_for_each_entry(e, &ife->metalist, metalist) {
394 		if (!e->ops->get(skb, e))
395 			total_encoded += 1;
396 	}
397 
398 	if (!total_encoded)
399 		goto out_nlmsg_trim;
400 
401 	nla_nest_end(skb, nest);
402 
403 	return 0;
404 
405 out_nlmsg_trim:
406 	nlmsg_trim(skb, b);
407 	return -1;
408 }
409 
410 /* under ife->tcf_lock */
411 static void _tcf_ife_cleanup(struct tc_action *a)
412 {
413 	struct tcf_ife_info *ife = to_ife(a);
414 	struct tcf_meta_info *e, *n;
415 
416 	list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
417 		list_del(&e->metalist);
418 		if (e->metaval) {
419 			if (e->ops->release)
420 				e->ops->release(e);
421 			else
422 				kfree(e->metaval);
423 		}
424 		module_put(e->ops->owner);
425 		kfree(e);
426 	}
427 }
428 
429 static void tcf_ife_cleanup(struct tc_action *a)
430 {
431 	struct tcf_ife_info *ife = to_ife(a);
432 	struct tcf_ife_params *p;
433 
434 	spin_lock_bh(&ife->tcf_lock);
435 	_tcf_ife_cleanup(a);
436 	spin_unlock_bh(&ife->tcf_lock);
437 
438 	p = rcu_dereference_protected(ife->params, 1);
439 	if (p)
440 		kfree_rcu(p, rcu);
441 }
442 
443 static int populate_metalist(struct tcf_ife_info *ife, struct nlattr **tb,
444 			     bool exists, bool rtnl_held)
445 {
446 	int len = 0;
447 	int rc = 0;
448 	int i = 0;
449 	void *val;
450 
451 	for (i = 1; i < max_metacnt; i++) {
452 		if (tb[i]) {
453 			val = nla_data(tb[i]);
454 			len = nla_len(tb[i]);
455 
456 			rc = load_metaops_and_vet(i, val, len, rtnl_held);
457 			if (rc != 0)
458 				return rc;
459 
460 			rc = add_metainfo(ife, i, val, len, exists);
461 			if (rc)
462 				return rc;
463 		}
464 	}
465 
466 	return rc;
467 }
468 
469 static int tcf_ife_init(struct net *net, struct nlattr *nla,
470 			struct nlattr *est, struct tc_action **a,
471 			int ovr, int bind, bool rtnl_held,
472 			struct netlink_ext_ack *extack)
473 {
474 	struct tc_action_net *tn = net_generic(net, ife_net_id);
475 	struct nlattr *tb[TCA_IFE_MAX + 1];
476 	struct nlattr *tb2[IFE_META_MAX + 1];
477 	struct tcf_ife_params *p;
478 	struct tcf_ife_info *ife;
479 	u16 ife_type = ETH_P_IFE;
480 	struct tc_ife *parm;
481 	u8 *daddr = NULL;
482 	u8 *saddr = NULL;
483 	bool exists = false;
484 	int ret = 0;
485 	int err;
486 
487 	err = nla_parse_nested(tb, TCA_IFE_MAX, nla, ife_policy, NULL);
488 	if (err < 0)
489 		return err;
490 
491 	if (!tb[TCA_IFE_PARMS])
492 		return -EINVAL;
493 
494 	parm = nla_data(tb[TCA_IFE_PARMS]);
495 
496 	/* IFE_DECODE is 0 and indicates the opposite of IFE_ENCODE because
497 	 * they cannot run as the same time. Check on all other values which
498 	 * are not supported right now.
499 	 */
500 	if (parm->flags & ~IFE_ENCODE)
501 		return -EINVAL;
502 
503 	p = kzalloc(sizeof(*p), GFP_KERNEL);
504 	if (!p)
505 		return -ENOMEM;
506 
507 	err = tcf_idr_check_alloc(tn, &parm->index, a, bind);
508 	if (err < 0) {
509 		kfree(p);
510 		return err;
511 	}
512 	exists = err;
513 	if (exists && bind) {
514 		kfree(p);
515 		return 0;
516 	}
517 
518 	if (!exists) {
519 		ret = tcf_idr_create(tn, parm->index, est, a, &act_ife_ops,
520 				     bind, true);
521 		if (ret) {
522 			tcf_idr_cleanup(tn, parm->index);
523 			kfree(p);
524 			return ret;
525 		}
526 		ret = ACT_P_CREATED;
527 	} else if (!ovr) {
528 		tcf_idr_release(*a, bind);
529 		kfree(p);
530 		return -EEXIST;
531 	}
532 
533 	ife = to_ife(*a);
534 	p->flags = parm->flags;
535 
536 	if (parm->flags & IFE_ENCODE) {
537 		if (tb[TCA_IFE_TYPE])
538 			ife_type = nla_get_u16(tb[TCA_IFE_TYPE]);
539 		if (tb[TCA_IFE_DMAC])
540 			daddr = nla_data(tb[TCA_IFE_DMAC]);
541 		if (tb[TCA_IFE_SMAC])
542 			saddr = nla_data(tb[TCA_IFE_SMAC]);
543 	}
544 
545 	if (parm->flags & IFE_ENCODE) {
546 		if (daddr)
547 			ether_addr_copy(p->eth_dst, daddr);
548 		else
549 			eth_zero_addr(p->eth_dst);
550 
551 		if (saddr)
552 			ether_addr_copy(p->eth_src, saddr);
553 		else
554 			eth_zero_addr(p->eth_src);
555 
556 		p->eth_type = ife_type;
557 	}
558 
559 
560 	if (ret == ACT_P_CREATED)
561 		INIT_LIST_HEAD(&ife->metalist);
562 
563 	if (tb[TCA_IFE_METALST]) {
564 		err = nla_parse_nested(tb2, IFE_META_MAX, tb[TCA_IFE_METALST],
565 				       NULL, NULL);
566 		if (err) {
567 metadata_parse_err:
568 			tcf_idr_release(*a, bind);
569 			kfree(p);
570 			return err;
571 		}
572 
573 		err = populate_metalist(ife, tb2, exists, rtnl_held);
574 		if (err)
575 			goto metadata_parse_err;
576 
577 	} else {
578 		/* if no passed metadata allow list or passed allow-all
579 		 * then here we process by adding as many supported metadatum
580 		 * as we can. You better have at least one else we are
581 		 * going to bail out
582 		 */
583 		err = use_all_metadata(ife, exists);
584 		if (err) {
585 			tcf_idr_release(*a, bind);
586 			kfree(p);
587 			return err;
588 		}
589 	}
590 
591 	if (exists)
592 		spin_lock_bh(&ife->tcf_lock);
593 	ife->tcf_action = parm->action;
594 	/* protected by tcf_lock when modifying existing action */
595 	rcu_swap_protected(ife->params, p, 1);
596 
597 	if (exists)
598 		spin_unlock_bh(&ife->tcf_lock);
599 	if (p)
600 		kfree_rcu(p, rcu);
601 
602 	if (ret == ACT_P_CREATED)
603 		tcf_idr_insert(tn, *a);
604 
605 	return ret;
606 }
607 
608 static int tcf_ife_dump(struct sk_buff *skb, struct tc_action *a, int bind,
609 			int ref)
610 {
611 	unsigned char *b = skb_tail_pointer(skb);
612 	struct tcf_ife_info *ife = to_ife(a);
613 	struct tcf_ife_params *p;
614 	struct tc_ife opt = {
615 		.index = ife->tcf_index,
616 		.refcnt = refcount_read(&ife->tcf_refcnt) - ref,
617 		.bindcnt = atomic_read(&ife->tcf_bindcnt) - bind,
618 	};
619 	struct tcf_t t;
620 
621 	spin_lock_bh(&ife->tcf_lock);
622 	opt.action = ife->tcf_action;
623 	p = rcu_dereference_protected(ife->params,
624 				      lockdep_is_held(&ife->tcf_lock));
625 	opt.flags = p->flags;
626 
627 	if (nla_put(skb, TCA_IFE_PARMS, sizeof(opt), &opt))
628 		goto nla_put_failure;
629 
630 	tcf_tm_dump(&t, &ife->tcf_tm);
631 	if (nla_put_64bit(skb, TCA_IFE_TM, sizeof(t), &t, TCA_IFE_PAD))
632 		goto nla_put_failure;
633 
634 	if (!is_zero_ether_addr(p->eth_dst)) {
635 		if (nla_put(skb, TCA_IFE_DMAC, ETH_ALEN, p->eth_dst))
636 			goto nla_put_failure;
637 	}
638 
639 	if (!is_zero_ether_addr(p->eth_src)) {
640 		if (nla_put(skb, TCA_IFE_SMAC, ETH_ALEN, p->eth_src))
641 			goto nla_put_failure;
642 	}
643 
644 	if (nla_put(skb, TCA_IFE_TYPE, 2, &p->eth_type))
645 		goto nla_put_failure;
646 
647 	if (dump_metalist(skb, ife)) {
648 		/*ignore failure to dump metalist */
649 		pr_info("Failed to dump metalist\n");
650 	}
651 
652 	spin_unlock_bh(&ife->tcf_lock);
653 	return skb->len;
654 
655 nla_put_failure:
656 	spin_unlock_bh(&ife->tcf_lock);
657 	nlmsg_trim(skb, b);
658 	return -1;
659 }
660 
661 static int find_decode_metaid(struct sk_buff *skb, struct tcf_ife_info *ife,
662 			      u16 metaid, u16 mlen, void *mdata)
663 {
664 	struct tcf_meta_info *e;
665 
666 	/* XXX: use hash to speed up */
667 	list_for_each_entry(e, &ife->metalist, metalist) {
668 		if (metaid == e->metaid) {
669 			if (e->ops) {
670 				/* We check for decode presence already */
671 				return e->ops->decode(skb, mdata, mlen);
672 			}
673 		}
674 	}
675 
676 	return -ENOENT;
677 }
678 
679 static int tcf_ife_decode(struct sk_buff *skb, const struct tc_action *a,
680 			  struct tcf_result *res)
681 {
682 	struct tcf_ife_info *ife = to_ife(a);
683 	int action = ife->tcf_action;
684 	u8 *ifehdr_end;
685 	u8 *tlv_data;
686 	u16 metalen;
687 
688 	bstats_cpu_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
689 	tcf_lastuse_update(&ife->tcf_tm);
690 
691 	if (skb_at_tc_ingress(skb))
692 		skb_push(skb, skb->dev->hard_header_len);
693 
694 	tlv_data = ife_decode(skb, &metalen);
695 	if (unlikely(!tlv_data)) {
696 		qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
697 		return TC_ACT_SHOT;
698 	}
699 
700 	ifehdr_end = tlv_data + metalen;
701 	for (; tlv_data < ifehdr_end; tlv_data = ife_tlv_meta_next(tlv_data)) {
702 		u8 *curr_data;
703 		u16 mtype;
704 		u16 dlen;
705 
706 		curr_data = ife_tlv_meta_decode(tlv_data, ifehdr_end, &mtype,
707 						&dlen, NULL);
708 		if (!curr_data) {
709 			qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
710 			return TC_ACT_SHOT;
711 		}
712 
713 		if (find_decode_metaid(skb, ife, mtype, dlen, curr_data)) {
714 			/* abuse overlimits to count when we receive metadata
715 			 * but dont have an ops for it
716 			 */
717 			pr_info_ratelimited("Unknown metaid %d dlen %d\n",
718 					    mtype, dlen);
719 			qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
720 		}
721 	}
722 
723 	if (WARN_ON(tlv_data != ifehdr_end)) {
724 		qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
725 		return TC_ACT_SHOT;
726 	}
727 
728 	skb->protocol = eth_type_trans(skb, skb->dev);
729 	skb_reset_network_header(skb);
730 
731 	return action;
732 }
733 
734 /*XXX: check if we can do this at install time instead of current
735  * send data path
736 **/
737 static int ife_get_sz(struct sk_buff *skb, struct tcf_ife_info *ife)
738 {
739 	struct tcf_meta_info *e, *n;
740 	int tot_run_sz = 0, run_sz = 0;
741 
742 	list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
743 		if (e->ops->check_presence) {
744 			run_sz = e->ops->check_presence(skb, e);
745 			tot_run_sz += run_sz;
746 		}
747 	}
748 
749 	return tot_run_sz;
750 }
751 
752 static int tcf_ife_encode(struct sk_buff *skb, const struct tc_action *a,
753 			  struct tcf_result *res, struct tcf_ife_params *p)
754 {
755 	struct tcf_ife_info *ife = to_ife(a);
756 	int action = ife->tcf_action;
757 	struct ethhdr *oethh;	/* outer ether header */
758 	struct tcf_meta_info *e;
759 	/*
760 	   OUTERHDR:TOTMETALEN:{TLVHDR:Metadatum:TLVHDR..}:ORIGDATA
761 	   where ORIGDATA = original ethernet header ...
762 	 */
763 	u16 metalen = ife_get_sz(skb, ife);
764 	int hdrm = metalen + skb->dev->hard_header_len + IFE_METAHDRLEN;
765 	unsigned int skboff = 0;
766 	int new_len = skb->len + hdrm;
767 	bool exceed_mtu = false;
768 	void *ife_meta;
769 	int err = 0;
770 
771 	if (!skb_at_tc_ingress(skb)) {
772 		if (new_len > skb->dev->mtu)
773 			exceed_mtu = true;
774 	}
775 
776 	bstats_cpu_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
777 	tcf_lastuse_update(&ife->tcf_tm);
778 
779 	if (!metalen) {		/* no metadata to send */
780 		/* abuse overlimits to count when we allow packet
781 		 * with no metadata
782 		 */
783 		qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
784 		return action;
785 	}
786 	/* could be stupid policy setup or mtu config
787 	 * so lets be conservative.. */
788 	if ((action == TC_ACT_SHOT) || exceed_mtu) {
789 		qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
790 		return TC_ACT_SHOT;
791 	}
792 
793 	if (skb_at_tc_ingress(skb))
794 		skb_push(skb, skb->dev->hard_header_len);
795 
796 	ife_meta = ife_encode(skb, metalen);
797 
798 	spin_lock(&ife->tcf_lock);
799 
800 	/* XXX: we dont have a clever way of telling encode to
801 	 * not repeat some of the computations that are done by
802 	 * ops->presence_check...
803 	 */
804 	list_for_each_entry(e, &ife->metalist, metalist) {
805 		if (e->ops->encode) {
806 			err = e->ops->encode(skb, (void *)(ife_meta + skboff),
807 					     e);
808 		}
809 		if (err < 0) {
810 			/* too corrupt to keep around if overwritten */
811 			spin_unlock(&ife->tcf_lock);
812 			qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
813 			return TC_ACT_SHOT;
814 		}
815 		skboff += err;
816 	}
817 	spin_unlock(&ife->tcf_lock);
818 	oethh = (struct ethhdr *)skb->data;
819 
820 	if (!is_zero_ether_addr(p->eth_src))
821 		ether_addr_copy(oethh->h_source, p->eth_src);
822 	if (!is_zero_ether_addr(p->eth_dst))
823 		ether_addr_copy(oethh->h_dest, p->eth_dst);
824 	oethh->h_proto = htons(p->eth_type);
825 
826 	if (skb_at_tc_ingress(skb))
827 		skb_pull(skb, skb->dev->hard_header_len);
828 
829 	return action;
830 }
831 
832 static int tcf_ife_act(struct sk_buff *skb, const struct tc_action *a,
833 		       struct tcf_result *res)
834 {
835 	struct tcf_ife_info *ife = to_ife(a);
836 	struct tcf_ife_params *p;
837 	int ret;
838 
839 	p = rcu_dereference_bh(ife->params);
840 	if (p->flags & IFE_ENCODE) {
841 		ret = tcf_ife_encode(skb, a, res, p);
842 		return ret;
843 	}
844 
845 	return tcf_ife_decode(skb, a, res);
846 }
847 
848 static int tcf_ife_walker(struct net *net, struct sk_buff *skb,
849 			  struct netlink_callback *cb, int type,
850 			  const struct tc_action_ops *ops,
851 			  struct netlink_ext_ack *extack)
852 {
853 	struct tc_action_net *tn = net_generic(net, ife_net_id);
854 
855 	return tcf_generic_walker(tn, skb, cb, type, ops, extack);
856 }
857 
858 static int tcf_ife_search(struct net *net, struct tc_action **a, u32 index,
859 			  struct netlink_ext_ack *extack)
860 {
861 	struct tc_action_net *tn = net_generic(net, ife_net_id);
862 
863 	return tcf_idr_search(tn, a, index);
864 }
865 
866 static struct tc_action_ops act_ife_ops = {
867 	.kind = "ife",
868 	.type = TCA_ACT_IFE,
869 	.owner = THIS_MODULE,
870 	.act = tcf_ife_act,
871 	.dump = tcf_ife_dump,
872 	.cleanup = tcf_ife_cleanup,
873 	.init = tcf_ife_init,
874 	.walk = tcf_ife_walker,
875 	.lookup = tcf_ife_search,
876 	.size =	sizeof(struct tcf_ife_info),
877 };
878 
879 static __net_init int ife_init_net(struct net *net)
880 {
881 	struct tc_action_net *tn = net_generic(net, ife_net_id);
882 
883 	return tc_action_net_init(tn, &act_ife_ops);
884 }
885 
886 static void __net_exit ife_exit_net(struct list_head *net_list)
887 {
888 	tc_action_net_exit(net_list, ife_net_id);
889 }
890 
891 static struct pernet_operations ife_net_ops = {
892 	.init = ife_init_net,
893 	.exit_batch = ife_exit_net,
894 	.id   = &ife_net_id,
895 	.size = sizeof(struct tc_action_net),
896 };
897 
898 static int __init ife_init_module(void)
899 {
900 	return tcf_register_action(&act_ife_ops, &ife_net_ops);
901 }
902 
903 static void __exit ife_cleanup_module(void)
904 {
905 	tcf_unregister_action(&act_ife_ops, &ife_net_ops);
906 }
907 
908 module_init(ife_init_module);
909 module_exit(ife_cleanup_module);
910 
911 MODULE_AUTHOR("Jamal Hadi Salim(2015)");
912 MODULE_DESCRIPTION("Inter-FE LFB action");
913 MODULE_LICENSE("GPL");
914