Loading...
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#include <net/tc_wrapper.h>
33
34static int max_metacnt = IFE_META_MAX + 1;
35static struct tc_action_ops act_ife_ops;
36
37static 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
44int 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}
59EXPORT_SYMBOL_GPL(ife_encode_meta_u16);
60
61int 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}
68EXPORT_SYMBOL_GPL(ife_get_meta_u32);
69
70int 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}
77EXPORT_SYMBOL_GPL(ife_check_meta_u32);
78
79int 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}
86EXPORT_SYMBOL_GPL(ife_check_meta_u16);
87
88int 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}
103EXPORT_SYMBOL_GPL(ife_encode_meta_u32);
104
105int 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}
112EXPORT_SYMBOL_GPL(ife_get_meta_u16);
113
114int 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}
122EXPORT_SYMBOL_GPL(ife_alloc_meta_u32);
123
124int 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}
132EXPORT_SYMBOL_GPL(ife_alloc_meta_u16);
133
134void ife_release_meta_gen(struct tcf_meta_info *mi)
135{
136 kfree(mi->metaval);
137}
138EXPORT_SYMBOL_GPL(ife_release_meta_gen);
139
140int ife_validate_meta_u32(void *val, int len)
141{
142 if (len == sizeof(u32))
143 return 0;
144
145 return -EINVAL;
146}
147EXPORT_SYMBOL_GPL(ife_validate_meta_u32);
148
149int 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}
157EXPORT_SYMBOL_GPL(ife_validate_meta_u16);
158
159static LIST_HEAD(ifeoplist);
160static DEFINE_RWLOCK(ife_mod_lock);
161
162static 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
180int 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}
206EXPORT_SYMBOL_GPL(unregister_ife_op);
207
208int 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}
225EXPORT_SYMBOL_GPL(register_ife_op);
226
227static 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
248static 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*/
265static 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*/
295static 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
325static 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
339static 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
354static 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
374static 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
401out_nlmsg_trim:
402 nlmsg_trim(skb, b);
403 return -1;
404}
405
406/* under ife->tcf_lock */
407static 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
425static 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
439static 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
458static 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
480static int tcf_ife_init(struct net *net, struct nlattr *nla,
481 struct nlattr *est, struct tc_action **a,
482 struct tcf_proto *tp, u32 flags,
483 struct netlink_ext_ack *extack)
484{
485 struct tc_action_net *tn = net_generic(net, act_ife_ops.net_id);
486 bool bind = flags & TCA_ACT_FLAGS_BIND;
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, !(flags & TCA_ACT_FLAGS_NO_RTNL));
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 ACT_P_BOUND;
552 }
553
554 if (!exists) {
555 ret = tcf_idr_create(tn, index, est, a, &act_ife_ops,
556 bind, true, flags);
557 if (ret) {
558 tcf_idr_cleanup(tn, index);
559 kfree(p);
560 return ret;
561 }
562 ret = ACT_P_CREATED;
563 } else if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
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,
604 !(flags & TCA_ACT_FLAGS_NO_RTNL));
605 if (err)
606 goto metadata_parse_err;
607 } else {
608 /* if no passed metadata allow list or passed allow-all
609 * then here we process by adding as many supported metadatum
610 * as we can. You better have at least one else we are
611 * going to bail out
612 */
613 err = use_all_metadata(ife, exists);
614 if (err)
615 goto metadata_parse_err;
616 }
617
618 if (exists)
619 spin_lock_bh(&ife->tcf_lock);
620 /* protected by tcf_lock when modifying existing action */
621 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
622 p = rcu_replace_pointer(ife->params, p, 1);
623
624 if (exists)
625 spin_unlock_bh(&ife->tcf_lock);
626 if (goto_ch)
627 tcf_chain_put_by_act(goto_ch);
628 if (p)
629 kfree_rcu(p, rcu);
630
631 return ret;
632metadata_parse_err:
633 if (goto_ch)
634 tcf_chain_put_by_act(goto_ch);
635release_idr:
636 kfree(p);
637 tcf_idr_release(*a, bind);
638 return err;
639}
640
641static int tcf_ife_dump(struct sk_buff *skb, struct tc_action *a, int bind,
642 int ref)
643{
644 unsigned char *b = skb_tail_pointer(skb);
645 struct tcf_ife_info *ife = to_ife(a);
646 struct tcf_ife_params *p;
647 struct tc_ife opt = {
648 .index = ife->tcf_index,
649 .refcnt = refcount_read(&ife->tcf_refcnt) - ref,
650 .bindcnt = atomic_read(&ife->tcf_bindcnt) - bind,
651 };
652 struct tcf_t t;
653
654 spin_lock_bh(&ife->tcf_lock);
655 opt.action = ife->tcf_action;
656 p = rcu_dereference_protected(ife->params,
657 lockdep_is_held(&ife->tcf_lock));
658 opt.flags = p->flags;
659
660 if (nla_put(skb, TCA_IFE_PARMS, sizeof(opt), &opt))
661 goto nla_put_failure;
662
663 tcf_tm_dump(&t, &ife->tcf_tm);
664 if (nla_put_64bit(skb, TCA_IFE_TM, sizeof(t), &t, TCA_IFE_PAD))
665 goto nla_put_failure;
666
667 if (!is_zero_ether_addr(p->eth_dst)) {
668 if (nla_put(skb, TCA_IFE_DMAC, ETH_ALEN, p->eth_dst))
669 goto nla_put_failure;
670 }
671
672 if (!is_zero_ether_addr(p->eth_src)) {
673 if (nla_put(skb, TCA_IFE_SMAC, ETH_ALEN, p->eth_src))
674 goto nla_put_failure;
675 }
676
677 if (nla_put(skb, TCA_IFE_TYPE, 2, &p->eth_type))
678 goto nla_put_failure;
679
680 if (dump_metalist(skb, ife)) {
681 /*ignore failure to dump metalist */
682 pr_info("Failed to dump metalist\n");
683 }
684
685 spin_unlock_bh(&ife->tcf_lock);
686 return skb->len;
687
688nla_put_failure:
689 spin_unlock_bh(&ife->tcf_lock);
690 nlmsg_trim(skb, b);
691 return -1;
692}
693
694static int find_decode_metaid(struct sk_buff *skb, struct tcf_ife_info *ife,
695 u16 metaid, u16 mlen, void *mdata)
696{
697 struct tcf_meta_info *e;
698
699 /* XXX: use hash to speed up */
700 list_for_each_entry(e, &ife->metalist, metalist) {
701 if (metaid == e->metaid) {
702 if (e->ops) {
703 /* We check for decode presence already */
704 return e->ops->decode(skb, mdata, mlen);
705 }
706 }
707 }
708
709 return -ENOENT;
710}
711
712static int tcf_ife_decode(struct sk_buff *skb, const struct tc_action *a,
713 struct tcf_result *res)
714{
715 struct tcf_ife_info *ife = to_ife(a);
716 int action = ife->tcf_action;
717 u8 *ifehdr_end;
718 u8 *tlv_data;
719 u16 metalen;
720
721 bstats_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
722 tcf_lastuse_update(&ife->tcf_tm);
723
724 if (skb_at_tc_ingress(skb))
725 skb_push(skb, skb->dev->hard_header_len);
726
727 tlv_data = ife_decode(skb, &metalen);
728 if (unlikely(!tlv_data)) {
729 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
730 return TC_ACT_SHOT;
731 }
732
733 ifehdr_end = tlv_data + metalen;
734 for (; tlv_data < ifehdr_end; tlv_data = ife_tlv_meta_next(tlv_data)) {
735 u8 *curr_data;
736 u16 mtype;
737 u16 dlen;
738
739 curr_data = ife_tlv_meta_decode(tlv_data, ifehdr_end, &mtype,
740 &dlen, NULL);
741 if (!curr_data) {
742 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
743 return TC_ACT_SHOT;
744 }
745
746 if (find_decode_metaid(skb, ife, mtype, dlen, curr_data)) {
747 /* abuse overlimits to count when we receive metadata
748 * but dont have an ops for it
749 */
750 pr_info_ratelimited("Unknown metaid %d dlen %d\n",
751 mtype, dlen);
752 qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
753 }
754 }
755
756 if (WARN_ON(tlv_data != ifehdr_end)) {
757 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
758 return TC_ACT_SHOT;
759 }
760
761 skb->protocol = eth_type_trans(skb, skb->dev);
762 skb_reset_network_header(skb);
763
764 return action;
765}
766
767/*XXX: check if we can do this at install time instead of current
768 * send data path
769**/
770static int ife_get_sz(struct sk_buff *skb, struct tcf_ife_info *ife)
771{
772 struct tcf_meta_info *e, *n;
773 int tot_run_sz = 0, run_sz = 0;
774
775 list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
776 if (e->ops->check_presence) {
777 run_sz = e->ops->check_presence(skb, e);
778 tot_run_sz += run_sz;
779 }
780 }
781
782 return tot_run_sz;
783}
784
785static int tcf_ife_encode(struct sk_buff *skb, const struct tc_action *a,
786 struct tcf_result *res, struct tcf_ife_params *p)
787{
788 struct tcf_ife_info *ife = to_ife(a);
789 int action = ife->tcf_action;
790 struct ethhdr *oethh; /* outer ether header */
791 struct tcf_meta_info *e;
792 /*
793 OUTERHDR:TOTMETALEN:{TLVHDR:Metadatum:TLVHDR..}:ORIGDATA
794 where ORIGDATA = original ethernet header ...
795 */
796 u16 metalen = ife_get_sz(skb, ife);
797 int hdrm = metalen + skb->dev->hard_header_len + IFE_METAHDRLEN;
798 unsigned int skboff = 0;
799 int new_len = skb->len + hdrm;
800 bool exceed_mtu = false;
801 void *ife_meta;
802 int err = 0;
803
804 if (!skb_at_tc_ingress(skb)) {
805 if (new_len > skb->dev->mtu)
806 exceed_mtu = true;
807 }
808
809 bstats_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
810 tcf_lastuse_update(&ife->tcf_tm);
811
812 if (!metalen) { /* no metadata to send */
813 /* abuse overlimits to count when we allow packet
814 * with no metadata
815 */
816 qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
817 return action;
818 }
819 /* could be stupid policy setup or mtu config
820 * so lets be conservative.. */
821 if ((action == TC_ACT_SHOT) || exceed_mtu) {
822 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
823 return TC_ACT_SHOT;
824 }
825
826 if (skb_at_tc_ingress(skb))
827 skb_push(skb, skb->dev->hard_header_len);
828
829 ife_meta = ife_encode(skb, metalen);
830
831 spin_lock(&ife->tcf_lock);
832
833 /* XXX: we dont have a clever way of telling encode to
834 * not repeat some of the computations that are done by
835 * ops->presence_check...
836 */
837 list_for_each_entry(e, &ife->metalist, metalist) {
838 if (e->ops->encode) {
839 err = e->ops->encode(skb, (void *)(ife_meta + skboff),
840 e);
841 }
842 if (err < 0) {
843 /* too corrupt to keep around if overwritten */
844 spin_unlock(&ife->tcf_lock);
845 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
846 return TC_ACT_SHOT;
847 }
848 skboff += err;
849 }
850 spin_unlock(&ife->tcf_lock);
851 oethh = (struct ethhdr *)skb->data;
852
853 if (!is_zero_ether_addr(p->eth_src))
854 ether_addr_copy(oethh->h_source, p->eth_src);
855 if (!is_zero_ether_addr(p->eth_dst))
856 ether_addr_copy(oethh->h_dest, p->eth_dst);
857 oethh->h_proto = htons(p->eth_type);
858
859 if (skb_at_tc_ingress(skb))
860 skb_pull(skb, skb->dev->hard_header_len);
861
862 return action;
863}
864
865TC_INDIRECT_SCOPE int tcf_ife_act(struct sk_buff *skb,
866 const struct tc_action *a,
867 struct tcf_result *res)
868{
869 struct tcf_ife_info *ife = to_ife(a);
870 struct tcf_ife_params *p;
871 int ret;
872
873 p = rcu_dereference_bh(ife->params);
874 if (p->flags & IFE_ENCODE) {
875 ret = tcf_ife_encode(skb, a, res, p);
876 return ret;
877 }
878
879 return tcf_ife_decode(skb, a, res);
880}
881
882static struct tc_action_ops act_ife_ops = {
883 .kind = "ife",
884 .id = TCA_ID_IFE,
885 .owner = THIS_MODULE,
886 .act = tcf_ife_act,
887 .dump = tcf_ife_dump,
888 .cleanup = tcf_ife_cleanup,
889 .init = tcf_ife_init,
890 .size = sizeof(struct tcf_ife_info),
891};
892MODULE_ALIAS_NET_ACT("ife");
893
894static __net_init int ife_init_net(struct net *net)
895{
896 struct tc_action_net *tn = net_generic(net, act_ife_ops.net_id);
897
898 return tc_action_net_init(net, tn, &act_ife_ops);
899}
900
901static void __net_exit ife_exit_net(struct list_head *net_list)
902{
903 tc_action_net_exit(net_list, act_ife_ops.net_id);
904}
905
906static struct pernet_operations ife_net_ops = {
907 .init = ife_init_net,
908 .exit_batch = ife_exit_net,
909 .id = &act_ife_ops.net_id,
910 .size = sizeof(struct tc_action_net),
911};
912
913static int __init ife_init_module(void)
914{
915 return tcf_register_action(&act_ife_ops, &ife_net_ops);
916}
917
918static void __exit ife_cleanup_module(void)
919{
920 tcf_unregister_action(&act_ife_ops, &ife_net_ops);
921}
922
923module_init(ife_init_module);
924module_exit(ife_cleanup_module);
925
926MODULE_AUTHOR("Jamal Hadi Salim(2015)");
927MODULE_DESCRIPTION("Inter-FE LFB action");
928MODULE_LICENSE("GPL");
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
37static unsigned int ife_net_id;
38static int max_metacnt = IFE_META_MAX + 1;
39static struct tc_action_ops act_ife_ops;
40
41static 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
48int 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}
63EXPORT_SYMBOL_GPL(ife_encode_meta_u16);
64
65int 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}
72EXPORT_SYMBOL_GPL(ife_get_meta_u32);
73
74int 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}
81EXPORT_SYMBOL_GPL(ife_check_meta_u32);
82
83int 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}
90EXPORT_SYMBOL_GPL(ife_check_meta_u16);
91
92int 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}
107EXPORT_SYMBOL_GPL(ife_encode_meta_u32);
108
109int 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}
116EXPORT_SYMBOL_GPL(ife_get_meta_u16);
117
118int 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}
126EXPORT_SYMBOL_GPL(ife_alloc_meta_u32);
127
128int 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}
136EXPORT_SYMBOL_GPL(ife_alloc_meta_u16);
137
138void ife_release_meta_gen(struct tcf_meta_info *mi)
139{
140 kfree(mi->metaval);
141}
142EXPORT_SYMBOL_GPL(ife_release_meta_gen);
143
144int ife_validate_meta_u32(void *val, int len)
145{
146 if (len == sizeof(u32))
147 return 0;
148
149 return -EINVAL;
150}
151EXPORT_SYMBOL_GPL(ife_validate_meta_u32);
152
153int 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}
161EXPORT_SYMBOL_GPL(ife_validate_meta_u16);
162
163static LIST_HEAD(ifeoplist);
164static DEFINE_RWLOCK(ife_mod_lock);
165
166static 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
184int 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}
210EXPORT_SYMBOL_GPL(unregister_ife_op);
211
212int 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}
229EXPORT_SYMBOL_GPL(register_ife_op);
230
231static 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
252static 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 * under ife->tcf_lock for existing action
269*/
270static int load_metaops_and_vet(struct tcf_ife_info *ife, u32 metaid,
271 void *val, int len, bool exists)
272{
273 struct tcf_meta_ops *ops = find_ife_oplist(metaid);
274 int ret = 0;
275
276 if (!ops) {
277 ret = -ENOENT;
278#ifdef CONFIG_MODULES
279 if (exists)
280 spin_unlock_bh(&ife->tcf_lock);
281 rtnl_unlock();
282 request_module("ife-meta-%s", ife_meta_id2name(metaid));
283 rtnl_lock();
284 if (exists)
285 spin_lock_bh(&ife->tcf_lock);
286 ops = find_ife_oplist(metaid);
287#endif
288 }
289
290 if (ops) {
291 ret = 0;
292 if (len)
293 ret = ife_validate_metatype(ops, val, len);
294
295 module_put(ops->owner);
296 }
297
298 return ret;
299}
300
301/* called when adding new meta information
302 * under ife->tcf_lock for existing action
303*/
304static int add_metainfo(struct tcf_ife_info *ife, u32 metaid, void *metaval,
305 int len, bool atomic)
306{
307 struct tcf_meta_info *mi = NULL;
308 struct tcf_meta_ops *ops = find_ife_oplist(metaid);
309 int ret = 0;
310
311 if (!ops)
312 return -ENOENT;
313
314 mi = kzalloc(sizeof(*mi), atomic ? GFP_ATOMIC : GFP_KERNEL);
315 if (!mi) {
316 /*put back what find_ife_oplist took */
317 module_put(ops->owner);
318 return -ENOMEM;
319 }
320
321 mi->metaid = metaid;
322 mi->ops = ops;
323 if (len > 0) {
324 ret = ops->alloc(mi, metaval, atomic ? GFP_ATOMIC : GFP_KERNEL);
325 if (ret != 0) {
326 kfree(mi);
327 module_put(ops->owner);
328 return ret;
329 }
330 }
331
332 list_add_tail(&mi->metalist, &ife->metalist);
333
334 return ret;
335}
336
337static int use_all_metadata(struct tcf_ife_info *ife)
338{
339 struct tcf_meta_ops *o;
340 int rc = 0;
341 int installed = 0;
342
343 read_lock(&ife_mod_lock);
344 list_for_each_entry(o, &ifeoplist, list) {
345 rc = add_metainfo(ife, o->metaid, NULL, 0, true);
346 if (rc == 0)
347 installed += 1;
348 }
349 read_unlock(&ife_mod_lock);
350
351 if (installed)
352 return 0;
353 else
354 return -EINVAL;
355}
356
357static int dump_metalist(struct sk_buff *skb, struct tcf_ife_info *ife)
358{
359 struct tcf_meta_info *e;
360 struct nlattr *nest;
361 unsigned char *b = skb_tail_pointer(skb);
362 int total_encoded = 0;
363
364 /*can only happen on decode */
365 if (list_empty(&ife->metalist))
366 return 0;
367
368 nest = nla_nest_start(skb, TCA_IFE_METALST);
369 if (!nest)
370 goto out_nlmsg_trim;
371
372 list_for_each_entry(e, &ife->metalist, metalist) {
373 if (!e->ops->get(skb, e))
374 total_encoded += 1;
375 }
376
377 if (!total_encoded)
378 goto out_nlmsg_trim;
379
380 nla_nest_end(skb, nest);
381
382 return 0;
383
384out_nlmsg_trim:
385 nlmsg_trim(skb, b);
386 return -1;
387}
388
389/* under ife->tcf_lock */
390static void _tcf_ife_cleanup(struct tc_action *a)
391{
392 struct tcf_ife_info *ife = to_ife(a);
393 struct tcf_meta_info *e, *n;
394
395 list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
396 module_put(e->ops->owner);
397 list_del(&e->metalist);
398 if (e->metaval) {
399 if (e->ops->release)
400 e->ops->release(e);
401 else
402 kfree(e->metaval);
403 }
404 kfree(e);
405 }
406}
407
408static void tcf_ife_cleanup(struct tc_action *a)
409{
410 struct tcf_ife_info *ife = to_ife(a);
411 struct tcf_ife_params *p;
412
413 spin_lock_bh(&ife->tcf_lock);
414 _tcf_ife_cleanup(a);
415 spin_unlock_bh(&ife->tcf_lock);
416
417 p = rcu_dereference_protected(ife->params, 1);
418 kfree_rcu(p, rcu);
419}
420
421/* under ife->tcf_lock for existing action */
422static int populate_metalist(struct tcf_ife_info *ife, struct nlattr **tb,
423 bool exists)
424{
425 int len = 0;
426 int rc = 0;
427 int i = 0;
428 void *val;
429
430 for (i = 1; i < max_metacnt; i++) {
431 if (tb[i]) {
432 val = nla_data(tb[i]);
433 len = nla_len(tb[i]);
434
435 rc = load_metaops_and_vet(ife, i, val, len, exists);
436 if (rc != 0)
437 return rc;
438
439 rc = add_metainfo(ife, i, val, len, exists);
440 if (rc)
441 return rc;
442 }
443 }
444
445 return rc;
446}
447
448static int tcf_ife_init(struct net *net, struct nlattr *nla,
449 struct nlattr *est, struct tc_action **a,
450 int ovr, int bind, struct netlink_ext_ack *extack)
451{
452 struct tc_action_net *tn = net_generic(net, ife_net_id);
453 struct nlattr *tb[TCA_IFE_MAX + 1];
454 struct nlattr *tb2[IFE_META_MAX + 1];
455 struct tcf_ife_params *p, *p_old;
456 struct tcf_ife_info *ife;
457 u16 ife_type = ETH_P_IFE;
458 struct tc_ife *parm;
459 u8 *daddr = NULL;
460 u8 *saddr = NULL;
461 bool exists = false;
462 int ret = 0;
463 int err;
464
465 err = nla_parse_nested(tb, TCA_IFE_MAX, nla, ife_policy, NULL);
466 if (err < 0)
467 return err;
468
469 if (!tb[TCA_IFE_PARMS])
470 return -EINVAL;
471
472 parm = nla_data(tb[TCA_IFE_PARMS]);
473
474 /* IFE_DECODE is 0 and indicates the opposite of IFE_ENCODE because
475 * they cannot run as the same time. Check on all other values which
476 * are not supported right now.
477 */
478 if (parm->flags & ~IFE_ENCODE)
479 return -EINVAL;
480
481 p = kzalloc(sizeof(*p), GFP_KERNEL);
482 if (!p)
483 return -ENOMEM;
484
485 exists = tcf_idr_check(tn, parm->index, a, bind);
486 if (exists && bind) {
487 kfree(p);
488 return 0;
489 }
490
491 if (!exists) {
492 ret = tcf_idr_create(tn, parm->index, est, a, &act_ife_ops,
493 bind, true);
494 if (ret) {
495 kfree(p);
496 return ret;
497 }
498 ret = ACT_P_CREATED;
499 } else {
500 tcf_idr_release(*a, bind);
501 if (!ovr) {
502 kfree(p);
503 return -EEXIST;
504 }
505 }
506
507 ife = to_ife(*a);
508 p->flags = parm->flags;
509
510 if (parm->flags & IFE_ENCODE) {
511 if (tb[TCA_IFE_TYPE])
512 ife_type = nla_get_u16(tb[TCA_IFE_TYPE]);
513 if (tb[TCA_IFE_DMAC])
514 daddr = nla_data(tb[TCA_IFE_DMAC]);
515 if (tb[TCA_IFE_SMAC])
516 saddr = nla_data(tb[TCA_IFE_SMAC]);
517 }
518
519 ife->tcf_action = parm->action;
520
521 if (parm->flags & IFE_ENCODE) {
522 if (daddr)
523 ether_addr_copy(p->eth_dst, daddr);
524 else
525 eth_zero_addr(p->eth_dst);
526
527 if (saddr)
528 ether_addr_copy(p->eth_src, saddr);
529 else
530 eth_zero_addr(p->eth_src);
531
532 p->eth_type = ife_type;
533 }
534
535 if (exists)
536 spin_lock_bh(&ife->tcf_lock);
537
538 if (ret == ACT_P_CREATED)
539 INIT_LIST_HEAD(&ife->metalist);
540
541 if (tb[TCA_IFE_METALST]) {
542 err = nla_parse_nested(tb2, IFE_META_MAX, tb[TCA_IFE_METALST],
543 NULL, NULL);
544 if (err) {
545metadata_parse_err:
546 if (exists)
547 tcf_idr_release(*a, bind);
548 if (ret == ACT_P_CREATED)
549 _tcf_ife_cleanup(*a);
550
551 if (exists)
552 spin_unlock_bh(&ife->tcf_lock);
553 kfree(p);
554 return err;
555 }
556
557 err = populate_metalist(ife, tb2, exists);
558 if (err)
559 goto metadata_parse_err;
560
561 } else {
562 /* if no passed metadata allow list or passed allow-all
563 * then here we process by adding as many supported metadatum
564 * as we can. You better have at least one else we are
565 * going to bail out
566 */
567 err = use_all_metadata(ife);
568 if (err) {
569 if (ret == ACT_P_CREATED)
570 _tcf_ife_cleanup(*a);
571
572 if (exists)
573 spin_unlock_bh(&ife->tcf_lock);
574 kfree(p);
575 return err;
576 }
577 }
578
579 if (exists)
580 spin_unlock_bh(&ife->tcf_lock);
581
582 p_old = rtnl_dereference(ife->params);
583 rcu_assign_pointer(ife->params, p);
584 if (p_old)
585 kfree_rcu(p_old, rcu);
586
587 if (ret == ACT_P_CREATED)
588 tcf_idr_insert(tn, *a);
589
590 return ret;
591}
592
593static int tcf_ife_dump(struct sk_buff *skb, struct tc_action *a, int bind,
594 int ref)
595{
596 unsigned char *b = skb_tail_pointer(skb);
597 struct tcf_ife_info *ife = to_ife(a);
598 struct tcf_ife_params *p = rtnl_dereference(ife->params);
599 struct tc_ife opt = {
600 .index = ife->tcf_index,
601 .refcnt = ife->tcf_refcnt - ref,
602 .bindcnt = ife->tcf_bindcnt - bind,
603 .action = ife->tcf_action,
604 .flags = p->flags,
605 };
606 struct tcf_t t;
607
608 if (nla_put(skb, TCA_IFE_PARMS, sizeof(opt), &opt))
609 goto nla_put_failure;
610
611 tcf_tm_dump(&t, &ife->tcf_tm);
612 if (nla_put_64bit(skb, TCA_IFE_TM, sizeof(t), &t, TCA_IFE_PAD))
613 goto nla_put_failure;
614
615 if (!is_zero_ether_addr(p->eth_dst)) {
616 if (nla_put(skb, TCA_IFE_DMAC, ETH_ALEN, p->eth_dst))
617 goto nla_put_failure;
618 }
619
620 if (!is_zero_ether_addr(p->eth_src)) {
621 if (nla_put(skb, TCA_IFE_SMAC, ETH_ALEN, p->eth_src))
622 goto nla_put_failure;
623 }
624
625 if (nla_put(skb, TCA_IFE_TYPE, 2, &p->eth_type))
626 goto nla_put_failure;
627
628 if (dump_metalist(skb, ife)) {
629 /*ignore failure to dump metalist */
630 pr_info("Failed to dump metalist\n");
631 }
632
633 return skb->len;
634
635nla_put_failure:
636 nlmsg_trim(skb, b);
637 return -1;
638}
639
640static int find_decode_metaid(struct sk_buff *skb, struct tcf_ife_info *ife,
641 u16 metaid, u16 mlen, void *mdata)
642{
643 struct tcf_meta_info *e;
644
645 /* XXX: use hash to speed up */
646 list_for_each_entry(e, &ife->metalist, metalist) {
647 if (metaid == e->metaid) {
648 if (e->ops) {
649 /* We check for decode presence already */
650 return e->ops->decode(skb, mdata, mlen);
651 }
652 }
653 }
654
655 return -ENOENT;
656}
657
658static int tcf_ife_decode(struct sk_buff *skb, const struct tc_action *a,
659 struct tcf_result *res)
660{
661 struct tcf_ife_info *ife = to_ife(a);
662 int action = ife->tcf_action;
663 u8 *ifehdr_end;
664 u8 *tlv_data;
665 u16 metalen;
666
667 bstats_cpu_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
668 tcf_lastuse_update(&ife->tcf_tm);
669
670 if (skb_at_tc_ingress(skb))
671 skb_push(skb, skb->dev->hard_header_len);
672
673 tlv_data = ife_decode(skb, &metalen);
674 if (unlikely(!tlv_data)) {
675 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
676 return TC_ACT_SHOT;
677 }
678
679 ifehdr_end = tlv_data + metalen;
680 for (; tlv_data < ifehdr_end; tlv_data = ife_tlv_meta_next(tlv_data)) {
681 u8 *curr_data;
682 u16 mtype;
683 u16 dlen;
684
685 curr_data = ife_tlv_meta_decode(tlv_data, ifehdr_end, &mtype,
686 &dlen, NULL);
687 if (!curr_data) {
688 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
689 return TC_ACT_SHOT;
690 }
691
692 if (find_decode_metaid(skb, ife, mtype, dlen, curr_data)) {
693 /* abuse overlimits to count when we receive metadata
694 * but dont have an ops for it
695 */
696 pr_info_ratelimited("Unknown metaid %d dlen %d\n",
697 mtype, dlen);
698 qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
699 }
700 }
701
702 if (WARN_ON(tlv_data != ifehdr_end)) {
703 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
704 return TC_ACT_SHOT;
705 }
706
707 skb->protocol = eth_type_trans(skb, skb->dev);
708 skb_reset_network_header(skb);
709
710 return action;
711}
712
713/*XXX: check if we can do this at install time instead of current
714 * send data path
715**/
716static int ife_get_sz(struct sk_buff *skb, struct tcf_ife_info *ife)
717{
718 struct tcf_meta_info *e, *n;
719 int tot_run_sz = 0, run_sz = 0;
720
721 list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
722 if (e->ops->check_presence) {
723 run_sz = e->ops->check_presence(skb, e);
724 tot_run_sz += run_sz;
725 }
726 }
727
728 return tot_run_sz;
729}
730
731static int tcf_ife_encode(struct sk_buff *skb, const struct tc_action *a,
732 struct tcf_result *res, struct tcf_ife_params *p)
733{
734 struct tcf_ife_info *ife = to_ife(a);
735 int action = ife->tcf_action;
736 struct ethhdr *oethh; /* outer ether header */
737 struct tcf_meta_info *e;
738 /*
739 OUTERHDR:TOTMETALEN:{TLVHDR:Metadatum:TLVHDR..}:ORIGDATA
740 where ORIGDATA = original ethernet header ...
741 */
742 u16 metalen = ife_get_sz(skb, ife);
743 int hdrm = metalen + skb->dev->hard_header_len + IFE_METAHDRLEN;
744 unsigned int skboff = 0;
745 int new_len = skb->len + hdrm;
746 bool exceed_mtu = false;
747 void *ife_meta;
748 int err = 0;
749
750 if (!skb_at_tc_ingress(skb)) {
751 if (new_len > skb->dev->mtu)
752 exceed_mtu = true;
753 }
754
755 bstats_cpu_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
756 tcf_lastuse_update(&ife->tcf_tm);
757
758 if (!metalen) { /* no metadata to send */
759 /* abuse overlimits to count when we allow packet
760 * with no metadata
761 */
762 qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
763 return action;
764 }
765 /* could be stupid policy setup or mtu config
766 * so lets be conservative.. */
767 if ((action == TC_ACT_SHOT) || exceed_mtu) {
768 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
769 return TC_ACT_SHOT;
770 }
771
772 if (skb_at_tc_ingress(skb))
773 skb_push(skb, skb->dev->hard_header_len);
774
775 ife_meta = ife_encode(skb, metalen);
776
777 spin_lock(&ife->tcf_lock);
778
779 /* XXX: we dont have a clever way of telling encode to
780 * not repeat some of the computations that are done by
781 * ops->presence_check...
782 */
783 list_for_each_entry(e, &ife->metalist, metalist) {
784 if (e->ops->encode) {
785 err = e->ops->encode(skb, (void *)(ife_meta + skboff),
786 e);
787 }
788 if (err < 0) {
789 /* too corrupt to keep around if overwritten */
790 spin_unlock(&ife->tcf_lock);
791 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
792 return TC_ACT_SHOT;
793 }
794 skboff += err;
795 }
796 spin_unlock(&ife->tcf_lock);
797 oethh = (struct ethhdr *)skb->data;
798
799 if (!is_zero_ether_addr(p->eth_src))
800 ether_addr_copy(oethh->h_source, p->eth_src);
801 if (!is_zero_ether_addr(p->eth_dst))
802 ether_addr_copy(oethh->h_dest, p->eth_dst);
803 oethh->h_proto = htons(p->eth_type);
804
805 if (skb_at_tc_ingress(skb))
806 skb_pull(skb, skb->dev->hard_header_len);
807
808 return action;
809}
810
811static int tcf_ife_act(struct sk_buff *skb, const struct tc_action *a,
812 struct tcf_result *res)
813{
814 struct tcf_ife_info *ife = to_ife(a);
815 struct tcf_ife_params *p;
816 int ret;
817
818 rcu_read_lock();
819 p = rcu_dereference(ife->params);
820 if (p->flags & IFE_ENCODE) {
821 ret = tcf_ife_encode(skb, a, res, p);
822 rcu_read_unlock();
823 return ret;
824 }
825 rcu_read_unlock();
826
827 return tcf_ife_decode(skb, a, res);
828}
829
830static int tcf_ife_walker(struct net *net, struct sk_buff *skb,
831 struct netlink_callback *cb, int type,
832 const struct tc_action_ops *ops,
833 struct netlink_ext_ack *extack)
834{
835 struct tc_action_net *tn = net_generic(net, ife_net_id);
836
837 return tcf_generic_walker(tn, skb, cb, type, ops, extack);
838}
839
840static int tcf_ife_search(struct net *net, struct tc_action **a, u32 index,
841 struct netlink_ext_ack *extack)
842{
843 struct tc_action_net *tn = net_generic(net, ife_net_id);
844
845 return tcf_idr_search(tn, a, index);
846}
847
848static struct tc_action_ops act_ife_ops = {
849 .kind = "ife",
850 .type = TCA_ACT_IFE,
851 .owner = THIS_MODULE,
852 .act = tcf_ife_act,
853 .dump = tcf_ife_dump,
854 .cleanup = tcf_ife_cleanup,
855 .init = tcf_ife_init,
856 .walk = tcf_ife_walker,
857 .lookup = tcf_ife_search,
858 .size = sizeof(struct tcf_ife_info),
859};
860
861static __net_init int ife_init_net(struct net *net)
862{
863 struct tc_action_net *tn = net_generic(net, ife_net_id);
864
865 return tc_action_net_init(tn, &act_ife_ops);
866}
867
868static void __net_exit ife_exit_net(struct list_head *net_list)
869{
870 tc_action_net_exit(net_list, ife_net_id);
871}
872
873static struct pernet_operations ife_net_ops = {
874 .init = ife_init_net,
875 .exit_batch = ife_exit_net,
876 .id = &ife_net_id,
877 .size = sizeof(struct tc_action_net),
878};
879
880static int __init ife_init_module(void)
881{
882 return tcf_register_action(&act_ife_ops, &ife_net_ops);
883}
884
885static void __exit ife_cleanup_module(void)
886{
887 tcf_unregister_action(&act_ife_ops, &ife_net_ops);
888}
889
890module_init(ife_init_module);
891module_exit(ife_cleanup_module);
892
893MODULE_AUTHOR("Jamal Hadi Salim(2015)");
894MODULE_DESCRIPTION("Inter-FE LFB action");
895MODULE_LICENSE("GPL");