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
33static unsigned int ife_net_id;
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 populate_metalist(struct tcf_ife_info *ife, struct nlattr **tb,
440 bool exists, bool rtnl_held)
441{
442 int len = 0;
443 int rc = 0;
444 int i = 0;
445 void *val;
446
447 for (i = 1; i < max_metacnt; i++) {
448 if (tb[i]) {
449 val = nla_data(tb[i]);
450 len = nla_len(tb[i]);
451
452 rc = load_metaops_and_vet(i, val, len, rtnl_held);
453 if (rc != 0)
454 return rc;
455
456 rc = add_metainfo(ife, i, val, len, exists);
457 if (rc)
458 return rc;
459 }
460 }
461
462 return rc;
463}
464
465static int tcf_ife_init(struct net *net, struct nlattr *nla,
466 struct nlattr *est, struct tc_action **a,
467 int ovr, int bind, bool rtnl_held,
468 struct tcf_proto *tp, struct netlink_ext_ack *extack)
469{
470 struct tc_action_net *tn = net_generic(net, ife_net_id);
471 struct nlattr *tb[TCA_IFE_MAX + 1];
472 struct nlattr *tb2[IFE_META_MAX + 1];
473 struct tcf_chain *goto_ch = NULL;
474 struct tcf_ife_params *p;
475 struct tcf_ife_info *ife;
476 u16 ife_type = ETH_P_IFE;
477 struct tc_ife *parm;
478 u8 *daddr = NULL;
479 u8 *saddr = NULL;
480 bool exists = false;
481 int ret = 0;
482 u32 index;
483 int err;
484
485 if (!nla) {
486 NL_SET_ERR_MSG_MOD(extack, "IFE requires attributes to be passed");
487 return -EINVAL;
488 }
489
490 err = nla_parse_nested_deprecated(tb, TCA_IFE_MAX, nla, ife_policy,
491 NULL);
492 if (err < 0)
493 return err;
494
495 if (!tb[TCA_IFE_PARMS])
496 return -EINVAL;
497
498 parm = nla_data(tb[TCA_IFE_PARMS]);
499
500 /* IFE_DECODE is 0 and indicates the opposite of IFE_ENCODE because
501 * they cannot run as the same time. Check on all other values which
502 * are not supported right now.
503 */
504 if (parm->flags & ~IFE_ENCODE)
505 return -EINVAL;
506
507 p = kzalloc(sizeof(*p), GFP_KERNEL);
508 if (!p)
509 return -ENOMEM;
510
511 index = parm->index;
512 err = tcf_idr_check_alloc(tn, &index, a, bind);
513 if (err < 0) {
514 kfree(p);
515 return err;
516 }
517 exists = err;
518 if (exists && bind) {
519 kfree(p);
520 return 0;
521 }
522
523 if (!exists) {
524 ret = tcf_idr_create(tn, index, est, a, &act_ife_ops,
525 bind, true);
526 if (ret) {
527 tcf_idr_cleanup(tn, index);
528 kfree(p);
529 return ret;
530 }
531 ret = ACT_P_CREATED;
532 } else if (!ovr) {
533 tcf_idr_release(*a, bind);
534 kfree(p);
535 return -EEXIST;
536 }
537
538 ife = to_ife(*a);
539 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
540 if (err < 0)
541 goto release_idr;
542
543 p->flags = parm->flags;
544
545 if (parm->flags & IFE_ENCODE) {
546 if (tb[TCA_IFE_TYPE])
547 ife_type = nla_get_u16(tb[TCA_IFE_TYPE]);
548 if (tb[TCA_IFE_DMAC])
549 daddr = nla_data(tb[TCA_IFE_DMAC]);
550 if (tb[TCA_IFE_SMAC])
551 saddr = nla_data(tb[TCA_IFE_SMAC]);
552 }
553
554 if (parm->flags & IFE_ENCODE) {
555 if (daddr)
556 ether_addr_copy(p->eth_dst, daddr);
557 else
558 eth_zero_addr(p->eth_dst);
559
560 if (saddr)
561 ether_addr_copy(p->eth_src, saddr);
562 else
563 eth_zero_addr(p->eth_src);
564
565 p->eth_type = ife_type;
566 }
567
568
569 if (ret == ACT_P_CREATED)
570 INIT_LIST_HEAD(&ife->metalist);
571
572 if (tb[TCA_IFE_METALST]) {
573 err = nla_parse_nested_deprecated(tb2, IFE_META_MAX,
574 tb[TCA_IFE_METALST], NULL,
575 NULL);
576 if (err)
577 goto metadata_parse_err;
578 err = populate_metalist(ife, tb2, exists, rtnl_held);
579 if (err)
580 goto metadata_parse_err;
581
582 } else {
583 /* if no passed metadata allow list or passed allow-all
584 * then here we process by adding as many supported metadatum
585 * as we can. You better have at least one else we are
586 * going to bail out
587 */
588 err = use_all_metadata(ife, exists);
589 if (err)
590 goto metadata_parse_err;
591 }
592
593 if (exists)
594 spin_lock_bh(&ife->tcf_lock);
595 /* protected by tcf_lock when modifying existing action */
596 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
597 rcu_swap_protected(ife->params, p, 1);
598
599 if (exists)
600 spin_unlock_bh(&ife->tcf_lock);
601 if (goto_ch)
602 tcf_chain_put_by_act(goto_ch);
603 if (p)
604 kfree_rcu(p, rcu);
605
606 if (ret == ACT_P_CREATED)
607 tcf_idr_insert(tn, *a);
608
609 return ret;
610metadata_parse_err:
611 if (goto_ch)
612 tcf_chain_put_by_act(goto_ch);
613release_idr:
614 kfree(p);
615 tcf_idr_release(*a, bind);
616 return err;
617}
618
619static int tcf_ife_dump(struct sk_buff *skb, struct tc_action *a, int bind,
620 int ref)
621{
622 unsigned char *b = skb_tail_pointer(skb);
623 struct tcf_ife_info *ife = to_ife(a);
624 struct tcf_ife_params *p;
625 struct tc_ife opt = {
626 .index = ife->tcf_index,
627 .refcnt = refcount_read(&ife->tcf_refcnt) - ref,
628 .bindcnt = atomic_read(&ife->tcf_bindcnt) - bind,
629 };
630 struct tcf_t t;
631
632 spin_lock_bh(&ife->tcf_lock);
633 opt.action = ife->tcf_action;
634 p = rcu_dereference_protected(ife->params,
635 lockdep_is_held(&ife->tcf_lock));
636 opt.flags = p->flags;
637
638 if (nla_put(skb, TCA_IFE_PARMS, sizeof(opt), &opt))
639 goto nla_put_failure;
640
641 tcf_tm_dump(&t, &ife->tcf_tm);
642 if (nla_put_64bit(skb, TCA_IFE_TM, sizeof(t), &t, TCA_IFE_PAD))
643 goto nla_put_failure;
644
645 if (!is_zero_ether_addr(p->eth_dst)) {
646 if (nla_put(skb, TCA_IFE_DMAC, ETH_ALEN, p->eth_dst))
647 goto nla_put_failure;
648 }
649
650 if (!is_zero_ether_addr(p->eth_src)) {
651 if (nla_put(skb, TCA_IFE_SMAC, ETH_ALEN, p->eth_src))
652 goto nla_put_failure;
653 }
654
655 if (nla_put(skb, TCA_IFE_TYPE, 2, &p->eth_type))
656 goto nla_put_failure;
657
658 if (dump_metalist(skb, ife)) {
659 /*ignore failure to dump metalist */
660 pr_info("Failed to dump metalist\n");
661 }
662
663 spin_unlock_bh(&ife->tcf_lock);
664 return skb->len;
665
666nla_put_failure:
667 spin_unlock_bh(&ife->tcf_lock);
668 nlmsg_trim(skb, b);
669 return -1;
670}
671
672static int find_decode_metaid(struct sk_buff *skb, struct tcf_ife_info *ife,
673 u16 metaid, u16 mlen, void *mdata)
674{
675 struct tcf_meta_info *e;
676
677 /* XXX: use hash to speed up */
678 list_for_each_entry(e, &ife->metalist, metalist) {
679 if (metaid == e->metaid) {
680 if (e->ops) {
681 /* We check for decode presence already */
682 return e->ops->decode(skb, mdata, mlen);
683 }
684 }
685 }
686
687 return -ENOENT;
688}
689
690static int tcf_ife_decode(struct sk_buff *skb, const struct tc_action *a,
691 struct tcf_result *res)
692{
693 struct tcf_ife_info *ife = to_ife(a);
694 int action = ife->tcf_action;
695 u8 *ifehdr_end;
696 u8 *tlv_data;
697 u16 metalen;
698
699 bstats_cpu_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
700 tcf_lastuse_update(&ife->tcf_tm);
701
702 if (skb_at_tc_ingress(skb))
703 skb_push(skb, skb->dev->hard_header_len);
704
705 tlv_data = ife_decode(skb, &metalen);
706 if (unlikely(!tlv_data)) {
707 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
708 return TC_ACT_SHOT;
709 }
710
711 ifehdr_end = tlv_data + metalen;
712 for (; tlv_data < ifehdr_end; tlv_data = ife_tlv_meta_next(tlv_data)) {
713 u8 *curr_data;
714 u16 mtype;
715 u16 dlen;
716
717 curr_data = ife_tlv_meta_decode(tlv_data, ifehdr_end, &mtype,
718 &dlen, NULL);
719 if (!curr_data) {
720 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
721 return TC_ACT_SHOT;
722 }
723
724 if (find_decode_metaid(skb, ife, mtype, dlen, curr_data)) {
725 /* abuse overlimits to count when we receive metadata
726 * but dont have an ops for it
727 */
728 pr_info_ratelimited("Unknown metaid %d dlen %d\n",
729 mtype, dlen);
730 qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
731 }
732 }
733
734 if (WARN_ON(tlv_data != ifehdr_end)) {
735 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
736 return TC_ACT_SHOT;
737 }
738
739 skb->protocol = eth_type_trans(skb, skb->dev);
740 skb_reset_network_header(skb);
741
742 return action;
743}
744
745/*XXX: check if we can do this at install time instead of current
746 * send data path
747**/
748static int ife_get_sz(struct sk_buff *skb, struct tcf_ife_info *ife)
749{
750 struct tcf_meta_info *e, *n;
751 int tot_run_sz = 0, run_sz = 0;
752
753 list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
754 if (e->ops->check_presence) {
755 run_sz = e->ops->check_presence(skb, e);
756 tot_run_sz += run_sz;
757 }
758 }
759
760 return tot_run_sz;
761}
762
763static int tcf_ife_encode(struct sk_buff *skb, const struct tc_action *a,
764 struct tcf_result *res, struct tcf_ife_params *p)
765{
766 struct tcf_ife_info *ife = to_ife(a);
767 int action = ife->tcf_action;
768 struct ethhdr *oethh; /* outer ether header */
769 struct tcf_meta_info *e;
770 /*
771 OUTERHDR:TOTMETALEN:{TLVHDR:Metadatum:TLVHDR..}:ORIGDATA
772 where ORIGDATA = original ethernet header ...
773 */
774 u16 metalen = ife_get_sz(skb, ife);
775 int hdrm = metalen + skb->dev->hard_header_len + IFE_METAHDRLEN;
776 unsigned int skboff = 0;
777 int new_len = skb->len + hdrm;
778 bool exceed_mtu = false;
779 void *ife_meta;
780 int err = 0;
781
782 if (!skb_at_tc_ingress(skb)) {
783 if (new_len > skb->dev->mtu)
784 exceed_mtu = true;
785 }
786
787 bstats_cpu_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
788 tcf_lastuse_update(&ife->tcf_tm);
789
790 if (!metalen) { /* no metadata to send */
791 /* abuse overlimits to count when we allow packet
792 * with no metadata
793 */
794 qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
795 return action;
796 }
797 /* could be stupid policy setup or mtu config
798 * so lets be conservative.. */
799 if ((action == TC_ACT_SHOT) || exceed_mtu) {
800 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
801 return TC_ACT_SHOT;
802 }
803
804 if (skb_at_tc_ingress(skb))
805 skb_push(skb, skb->dev->hard_header_len);
806
807 ife_meta = ife_encode(skb, metalen);
808
809 spin_lock(&ife->tcf_lock);
810
811 /* XXX: we dont have a clever way of telling encode to
812 * not repeat some of the computations that are done by
813 * ops->presence_check...
814 */
815 list_for_each_entry(e, &ife->metalist, metalist) {
816 if (e->ops->encode) {
817 err = e->ops->encode(skb, (void *)(ife_meta + skboff),
818 e);
819 }
820 if (err < 0) {
821 /* too corrupt to keep around if overwritten */
822 spin_unlock(&ife->tcf_lock);
823 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
824 return TC_ACT_SHOT;
825 }
826 skboff += err;
827 }
828 spin_unlock(&ife->tcf_lock);
829 oethh = (struct ethhdr *)skb->data;
830
831 if (!is_zero_ether_addr(p->eth_src))
832 ether_addr_copy(oethh->h_source, p->eth_src);
833 if (!is_zero_ether_addr(p->eth_dst))
834 ether_addr_copy(oethh->h_dest, p->eth_dst);
835 oethh->h_proto = htons(p->eth_type);
836
837 if (skb_at_tc_ingress(skb))
838 skb_pull(skb, skb->dev->hard_header_len);
839
840 return action;
841}
842
843static int tcf_ife_act(struct sk_buff *skb, const struct tc_action *a,
844 struct tcf_result *res)
845{
846 struct tcf_ife_info *ife = to_ife(a);
847 struct tcf_ife_params *p;
848 int ret;
849
850 p = rcu_dereference_bh(ife->params);
851 if (p->flags & IFE_ENCODE) {
852 ret = tcf_ife_encode(skb, a, res, p);
853 return ret;
854 }
855
856 return tcf_ife_decode(skb, a, res);
857}
858
859static int tcf_ife_walker(struct net *net, struct sk_buff *skb,
860 struct netlink_callback *cb, int type,
861 const struct tc_action_ops *ops,
862 struct netlink_ext_ack *extack)
863{
864 struct tc_action_net *tn = net_generic(net, ife_net_id);
865
866 return tcf_generic_walker(tn, skb, cb, type, ops, extack);
867}
868
869static int tcf_ife_search(struct net *net, struct tc_action **a, u32 index)
870{
871 struct tc_action_net *tn = net_generic(net, ife_net_id);
872
873 return tcf_idr_search(tn, a, index);
874}
875
876static struct tc_action_ops act_ife_ops = {
877 .kind = "ife",
878 .id = TCA_ID_IFE,
879 .owner = THIS_MODULE,
880 .act = tcf_ife_act,
881 .dump = tcf_ife_dump,
882 .cleanup = tcf_ife_cleanup,
883 .init = tcf_ife_init,
884 .walk = tcf_ife_walker,
885 .lookup = tcf_ife_search,
886 .size = sizeof(struct tcf_ife_info),
887};
888
889static __net_init int ife_init_net(struct net *net)
890{
891 struct tc_action_net *tn = net_generic(net, ife_net_id);
892
893 return tc_action_net_init(net, tn, &act_ife_ops);
894}
895
896static void __net_exit ife_exit_net(struct list_head *net_list)
897{
898 tc_action_net_exit(net_list, ife_net_id);
899}
900
901static struct pernet_operations ife_net_ops = {
902 .init = ife_init_net,
903 .exit_batch = ife_exit_net,
904 .id = &ife_net_id,
905 .size = sizeof(struct tc_action_net),
906};
907
908static int __init ife_init_module(void)
909{
910 return tcf_register_action(&act_ife_ops, &ife_net_ops);
911}
912
913static void __exit ife_cleanup_module(void)
914{
915 tcf_unregister_action(&act_ife_ops, &ife_net_ops);
916}
917
918module_init(ife_init_module);
919module_exit(ife_cleanup_module);
920
921MODULE_AUTHOR("Jamal Hadi Salim(2015)");
922MODULE_DESCRIPTION("Inter-FE LFB action");
923MODULE_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
36#define IFE_TAB_MASK 15
37
38static int ife_net_id;
39static int max_metacnt = IFE_META_MAX + 1;
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
48/* Caller takes care of presenting data in network order
49*/
50int ife_tlv_meta_encode(void *skbdata, u16 attrtype, u16 dlen, const void *dval)
51{
52 u32 *tlv = (u32 *)(skbdata);
53 u16 totlen = nla_total_size(dlen); /*alignment + hdr */
54 char *dptr = (char *)tlv + NLA_HDRLEN;
55 u32 htlv = attrtype << 16 | totlen;
56
57 *tlv = htonl(htlv);
58 memset(dptr, 0, totlen - NLA_HDRLEN);
59 memcpy(dptr, dval, dlen);
60
61 return totlen;
62}
63EXPORT_SYMBOL_GPL(ife_tlv_meta_encode);
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_encode_meta_u32(u32 metaval, void *skbdata, struct tcf_meta_info *mi)
84{
85 u32 edata = metaval;
86
87 if (mi->metaval)
88 edata = *(u32 *)mi->metaval;
89 else if (metaval)
90 edata = metaval;
91
92 if (!edata) /* will not encode */
93 return 0;
94
95 edata = htonl(edata);
96 return ife_tlv_meta_encode(skbdata, mi->metaid, 4, &edata);
97}
98EXPORT_SYMBOL_GPL(ife_encode_meta_u32);
99
100int ife_get_meta_u16(struct sk_buff *skb, struct tcf_meta_info *mi)
101{
102 if (mi->metaval)
103 return nla_put_u16(skb, mi->metaid, *(u16 *)mi->metaval);
104 else
105 return nla_put(skb, mi->metaid, 0, NULL);
106}
107EXPORT_SYMBOL_GPL(ife_get_meta_u16);
108
109int ife_alloc_meta_u32(struct tcf_meta_info *mi, void *metaval)
110{
111 mi->metaval = kmemdup(metaval, sizeof(u32), GFP_KERNEL);
112 if (!mi->metaval)
113 return -ENOMEM;
114
115 return 0;
116}
117EXPORT_SYMBOL_GPL(ife_alloc_meta_u32);
118
119int ife_alloc_meta_u16(struct tcf_meta_info *mi, void *metaval)
120{
121 mi->metaval = kmemdup(metaval, sizeof(u16), GFP_KERNEL);
122 if (!mi->metaval)
123 return -ENOMEM;
124
125 return 0;
126}
127EXPORT_SYMBOL_GPL(ife_alloc_meta_u16);
128
129void ife_release_meta_gen(struct tcf_meta_info *mi)
130{
131 kfree(mi->metaval);
132}
133EXPORT_SYMBOL_GPL(ife_release_meta_gen);
134
135int ife_validate_meta_u32(void *val, int len)
136{
137 if (len == 4)
138 return 0;
139
140 return -EINVAL;
141}
142EXPORT_SYMBOL_GPL(ife_validate_meta_u32);
143
144int ife_validate_meta_u16(void *val, int len)
145{
146 /* length will include padding */
147 if (len == NLA_ALIGN(2))
148 return 0;
149
150 return -EINVAL;
151}
152EXPORT_SYMBOL_GPL(ife_validate_meta_u16);
153
154static LIST_HEAD(ifeoplist);
155static DEFINE_RWLOCK(ife_mod_lock);
156
157static struct tcf_meta_ops *find_ife_oplist(u16 metaid)
158{
159 struct tcf_meta_ops *o;
160
161 read_lock(&ife_mod_lock);
162 list_for_each_entry(o, &ifeoplist, list) {
163 if (o->metaid == metaid) {
164 if (!try_module_get(o->owner))
165 o = NULL;
166 read_unlock(&ife_mod_lock);
167 return o;
168 }
169 }
170 read_unlock(&ife_mod_lock);
171
172 return NULL;
173}
174
175int register_ife_op(struct tcf_meta_ops *mops)
176{
177 struct tcf_meta_ops *m;
178
179 if (!mops->metaid || !mops->metatype || !mops->name ||
180 !mops->check_presence || !mops->encode || !mops->decode ||
181 !mops->get || !mops->alloc)
182 return -EINVAL;
183
184 write_lock(&ife_mod_lock);
185
186 list_for_each_entry(m, &ifeoplist, list) {
187 if (m->metaid == mops->metaid ||
188 (strcmp(mops->name, m->name) == 0)) {
189 write_unlock(&ife_mod_lock);
190 return -EEXIST;
191 }
192 }
193
194 if (!mops->release)
195 mops->release = ife_release_meta_gen;
196
197 list_add_tail(&mops->list, &ifeoplist);
198 write_unlock(&ife_mod_lock);
199 return 0;
200}
201EXPORT_SYMBOL_GPL(unregister_ife_op);
202
203int unregister_ife_op(struct tcf_meta_ops *mops)
204{
205 struct tcf_meta_ops *m;
206 int err = -ENOENT;
207
208 write_lock(&ife_mod_lock);
209 list_for_each_entry(m, &ifeoplist, list) {
210 if (m->metaid == mops->metaid) {
211 list_del(&mops->list);
212 err = 0;
213 break;
214 }
215 }
216 write_unlock(&ife_mod_lock);
217
218 return err;
219}
220EXPORT_SYMBOL_GPL(register_ife_op);
221
222static int ife_validate_metatype(struct tcf_meta_ops *ops, void *val, int len)
223{
224 int ret = 0;
225 /* XXX: unfortunately cant use nla_policy at this point
226 * because a length of 0 is valid in the case of
227 * "allow". "use" semantics do enforce for proper
228 * length and i couldve use nla_policy but it makes it hard
229 * to use it just for that..
230 */
231 if (ops->validate)
232 return ops->validate(val, len);
233
234 if (ops->metatype == NLA_U32)
235 ret = ife_validate_meta_u32(val, len);
236 else if (ops->metatype == NLA_U16)
237 ret = ife_validate_meta_u16(val, len);
238
239 return ret;
240}
241
242/* called when adding new meta information
243 * under ife->tcf_lock
244*/
245static int load_metaops_and_vet(struct tcf_ife_info *ife, u32 metaid,
246 void *val, int len)
247{
248 struct tcf_meta_ops *ops = find_ife_oplist(metaid);
249 int ret = 0;
250
251 if (!ops) {
252 ret = -ENOENT;
253#ifdef CONFIG_MODULES
254 spin_unlock_bh(&ife->tcf_lock);
255 rtnl_unlock();
256 request_module("ifemeta%u", metaid);
257 rtnl_lock();
258 spin_lock_bh(&ife->tcf_lock);
259 ops = find_ife_oplist(metaid);
260#endif
261 }
262
263 if (ops) {
264 ret = 0;
265 if (len)
266 ret = ife_validate_metatype(ops, val, len);
267
268 module_put(ops->owner);
269 }
270
271 return ret;
272}
273
274/* called when adding new meta information
275 * under ife->tcf_lock
276*/
277static int add_metainfo(struct tcf_ife_info *ife, u32 metaid, void *metaval,
278 int len)
279{
280 struct tcf_meta_info *mi = NULL;
281 struct tcf_meta_ops *ops = find_ife_oplist(metaid);
282 int ret = 0;
283
284 if (!ops)
285 return -ENOENT;
286
287 mi = kzalloc(sizeof(*mi), GFP_KERNEL);
288 if (!mi) {
289 /*put back what find_ife_oplist took */
290 module_put(ops->owner);
291 return -ENOMEM;
292 }
293
294 mi->metaid = metaid;
295 mi->ops = ops;
296 if (len > 0) {
297 ret = ops->alloc(mi, metaval);
298 if (ret != 0) {
299 kfree(mi);
300 module_put(ops->owner);
301 return ret;
302 }
303 }
304
305 list_add_tail(&mi->metalist, &ife->metalist);
306
307 return ret;
308}
309
310static int use_all_metadata(struct tcf_ife_info *ife)
311{
312 struct tcf_meta_ops *o;
313 int rc = 0;
314 int installed = 0;
315
316 list_for_each_entry(o, &ifeoplist, list) {
317 rc = add_metainfo(ife, o->metaid, NULL, 0);
318 if (rc == 0)
319 installed += 1;
320 }
321
322 if (installed)
323 return 0;
324 else
325 return -EINVAL;
326}
327
328static int dump_metalist(struct sk_buff *skb, struct tcf_ife_info *ife)
329{
330 struct tcf_meta_info *e;
331 struct nlattr *nest;
332 unsigned char *b = skb_tail_pointer(skb);
333 int total_encoded = 0;
334
335 /*can only happen on decode */
336 if (list_empty(&ife->metalist))
337 return 0;
338
339 nest = nla_nest_start(skb, TCA_IFE_METALST);
340 if (!nest)
341 goto out_nlmsg_trim;
342
343 list_for_each_entry(e, &ife->metalist, metalist) {
344 if (!e->ops->get(skb, e))
345 total_encoded += 1;
346 }
347
348 if (!total_encoded)
349 goto out_nlmsg_trim;
350
351 nla_nest_end(skb, nest);
352
353 return 0;
354
355out_nlmsg_trim:
356 nlmsg_trim(skb, b);
357 return -1;
358}
359
360/* under ife->tcf_lock */
361static void _tcf_ife_cleanup(struct tc_action *a, int bind)
362{
363 struct tcf_ife_info *ife = a->priv;
364 struct tcf_meta_info *e, *n;
365
366 list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
367 module_put(e->ops->owner);
368 list_del(&e->metalist);
369 if (e->metaval) {
370 if (e->ops->release)
371 e->ops->release(e);
372 else
373 kfree(e->metaval);
374 }
375 kfree(e);
376 }
377}
378
379static void tcf_ife_cleanup(struct tc_action *a, int bind)
380{
381 struct tcf_ife_info *ife = a->priv;
382
383 spin_lock_bh(&ife->tcf_lock);
384 _tcf_ife_cleanup(a, bind);
385 spin_unlock_bh(&ife->tcf_lock);
386}
387
388/* under ife->tcf_lock */
389static int populate_metalist(struct tcf_ife_info *ife, struct nlattr **tb)
390{
391 int len = 0;
392 int rc = 0;
393 int i = 0;
394 void *val;
395
396 for (i = 1; i < max_metacnt; i++) {
397 if (tb[i]) {
398 val = nla_data(tb[i]);
399 len = nla_len(tb[i]);
400
401 rc = load_metaops_and_vet(ife, i, val, len);
402 if (rc != 0)
403 return rc;
404
405 rc = add_metainfo(ife, i, val, len);
406 if (rc)
407 return rc;
408 }
409 }
410
411 return rc;
412}
413
414static int tcf_ife_init(struct net *net, struct nlattr *nla,
415 struct nlattr *est, struct tc_action *a,
416 int ovr, int bind)
417{
418 struct tc_action_net *tn = net_generic(net, ife_net_id);
419 struct nlattr *tb[TCA_IFE_MAX + 1];
420 struct nlattr *tb2[IFE_META_MAX + 1];
421 struct tcf_ife_info *ife;
422 struct tc_ife *parm;
423 u16 ife_type = 0;
424 u8 *daddr = NULL;
425 u8 *saddr = NULL;
426 int ret = 0, exists = 0;
427 int err;
428
429 err = nla_parse_nested(tb, TCA_IFE_MAX, nla, ife_policy);
430 if (err < 0)
431 return err;
432
433 if (!tb[TCA_IFE_PARMS])
434 return -EINVAL;
435
436 parm = nla_data(tb[TCA_IFE_PARMS]);
437
438 exists = tcf_hash_check(tn, parm->index, a, bind);
439 if (exists && bind)
440 return 0;
441
442 if (parm->flags & IFE_ENCODE) {
443 /* Until we get issued the ethertype, we cant have
444 * a default..
445 **/
446 if (!tb[TCA_IFE_TYPE]) {
447 if (exists)
448 tcf_hash_release(a, bind);
449 pr_info("You MUST pass etherype for encoding\n");
450 return -EINVAL;
451 }
452 }
453
454 if (!exists) {
455 ret = tcf_hash_create(tn, parm->index, est, a, sizeof(*ife),
456 bind, false);
457 if (ret)
458 return ret;
459 ret = ACT_P_CREATED;
460 } else {
461 tcf_hash_release(a, bind);
462 if (!ovr)
463 return -EEXIST;
464 }
465
466 ife = to_ife(a);
467 ife->flags = parm->flags;
468
469 if (parm->flags & IFE_ENCODE) {
470 ife_type = nla_get_u16(tb[TCA_IFE_TYPE]);
471 if (tb[TCA_IFE_DMAC])
472 daddr = nla_data(tb[TCA_IFE_DMAC]);
473 if (tb[TCA_IFE_SMAC])
474 saddr = nla_data(tb[TCA_IFE_SMAC]);
475 }
476
477 spin_lock_bh(&ife->tcf_lock);
478 ife->tcf_action = parm->action;
479
480 if (parm->flags & IFE_ENCODE) {
481 if (daddr)
482 ether_addr_copy(ife->eth_dst, daddr);
483 else
484 eth_zero_addr(ife->eth_dst);
485
486 if (saddr)
487 ether_addr_copy(ife->eth_src, saddr);
488 else
489 eth_zero_addr(ife->eth_src);
490
491 ife->eth_type = ife_type;
492 }
493
494 if (ret == ACT_P_CREATED)
495 INIT_LIST_HEAD(&ife->metalist);
496
497 if (tb[TCA_IFE_METALST]) {
498 err = nla_parse_nested(tb2, IFE_META_MAX, tb[TCA_IFE_METALST],
499 NULL);
500 if (err) {
501metadata_parse_err:
502 if (exists)
503 tcf_hash_release(a, bind);
504 if (ret == ACT_P_CREATED)
505 _tcf_ife_cleanup(a, bind);
506
507 spin_unlock_bh(&ife->tcf_lock);
508 return err;
509 }
510
511 err = populate_metalist(ife, tb2);
512 if (err)
513 goto metadata_parse_err;
514
515 } else {
516 /* if no passed metadata allow list or passed allow-all
517 * then here we process by adding as many supported metadatum
518 * as we can. You better have at least one else we are
519 * going to bail out
520 */
521 err = use_all_metadata(ife);
522 if (err) {
523 if (ret == ACT_P_CREATED)
524 _tcf_ife_cleanup(a, bind);
525
526 spin_unlock_bh(&ife->tcf_lock);
527 return err;
528 }
529 }
530
531 spin_unlock_bh(&ife->tcf_lock);
532
533 if (ret == ACT_P_CREATED)
534 tcf_hash_insert(tn, a);
535
536 return ret;
537}
538
539static int tcf_ife_dump(struct sk_buff *skb, struct tc_action *a, int bind,
540 int ref)
541{
542 unsigned char *b = skb_tail_pointer(skb);
543 struct tcf_ife_info *ife = a->priv;
544 struct tc_ife opt = {
545 .index = ife->tcf_index,
546 .refcnt = ife->tcf_refcnt - ref,
547 .bindcnt = ife->tcf_bindcnt - bind,
548 .action = ife->tcf_action,
549 .flags = ife->flags,
550 };
551 struct tcf_t t;
552
553 if (nla_put(skb, TCA_IFE_PARMS, sizeof(opt), &opt))
554 goto nla_put_failure;
555
556 t.install = jiffies_to_clock_t(jiffies - ife->tcf_tm.install);
557 t.lastuse = jiffies_to_clock_t(jiffies - ife->tcf_tm.lastuse);
558 t.expires = jiffies_to_clock_t(ife->tcf_tm.expires);
559 if (nla_put(skb, TCA_IFE_TM, sizeof(t), &t))
560 goto nla_put_failure;
561
562 if (!is_zero_ether_addr(ife->eth_dst)) {
563 if (nla_put(skb, TCA_IFE_DMAC, ETH_ALEN, ife->eth_dst))
564 goto nla_put_failure;
565 }
566
567 if (!is_zero_ether_addr(ife->eth_src)) {
568 if (nla_put(skb, TCA_IFE_SMAC, ETH_ALEN, ife->eth_src))
569 goto nla_put_failure;
570 }
571
572 if (nla_put(skb, TCA_IFE_TYPE, 2, &ife->eth_type))
573 goto nla_put_failure;
574
575 if (dump_metalist(skb, ife)) {
576 /*ignore failure to dump metalist */
577 pr_info("Failed to dump metalist\n");
578 }
579
580 return skb->len;
581
582nla_put_failure:
583 nlmsg_trim(skb, b);
584 return -1;
585}
586
587int find_decode_metaid(struct sk_buff *skb, struct tcf_ife_info *ife,
588 u16 metaid, u16 mlen, void *mdata)
589{
590 struct tcf_meta_info *e;
591
592 /* XXX: use hash to speed up */
593 list_for_each_entry(e, &ife->metalist, metalist) {
594 if (metaid == e->metaid) {
595 if (e->ops) {
596 /* We check for decode presence already */
597 return e->ops->decode(skb, mdata, mlen);
598 }
599 }
600 }
601
602 return 0;
603}
604
605struct ifeheadr {
606 __be16 metalen;
607 u8 tlv_data[];
608};
609
610struct meta_tlvhdr {
611 __be16 type;
612 __be16 len;
613};
614
615static int tcf_ife_decode(struct sk_buff *skb, const struct tc_action *a,
616 struct tcf_result *res)
617{
618 struct tcf_ife_info *ife = a->priv;
619 int action = ife->tcf_action;
620 struct ifeheadr *ifehdr = (struct ifeheadr *)skb->data;
621 u16 ifehdrln = ifehdr->metalen;
622 struct meta_tlvhdr *tlv = (struct meta_tlvhdr *)(ifehdr->tlv_data);
623
624 spin_lock(&ife->tcf_lock);
625 bstats_update(&ife->tcf_bstats, skb);
626 ife->tcf_tm.lastuse = jiffies;
627 spin_unlock(&ife->tcf_lock);
628
629 ifehdrln = ntohs(ifehdrln);
630 if (unlikely(!pskb_may_pull(skb, ifehdrln))) {
631 spin_lock(&ife->tcf_lock);
632 ife->tcf_qstats.drops++;
633 spin_unlock(&ife->tcf_lock);
634 return TC_ACT_SHOT;
635 }
636
637 skb_set_mac_header(skb, ifehdrln);
638 __skb_pull(skb, ifehdrln);
639 skb->protocol = eth_type_trans(skb, skb->dev);
640 ifehdrln -= IFE_METAHDRLEN;
641
642 while (ifehdrln > 0) {
643 u8 *tlvdata = (u8 *)tlv;
644 u16 mtype = tlv->type;
645 u16 mlen = tlv->len;
646
647 mtype = ntohs(mtype);
648 mlen = ntohs(mlen);
649
650 if (find_decode_metaid(skb, ife, mtype, (mlen - 4),
651 (void *)(tlvdata + 4))) {
652 /* abuse overlimits to count when we receive metadata
653 * but dont have an ops for it
654 */
655 pr_info_ratelimited("Unknown metaid %d alnlen %d\n",
656 mtype, mlen);
657 ife->tcf_qstats.overlimits++;
658 }
659
660 tlvdata += mlen;
661 ifehdrln -= mlen;
662 tlv = (struct meta_tlvhdr *)tlvdata;
663 }
664
665 skb_reset_network_header(skb);
666 return action;
667}
668
669/*XXX: check if we can do this at install time instead of current
670 * send data path
671**/
672static int ife_get_sz(struct sk_buff *skb, struct tcf_ife_info *ife)
673{
674 struct tcf_meta_info *e, *n;
675 int tot_run_sz = 0, run_sz = 0;
676
677 list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
678 if (e->ops->check_presence) {
679 run_sz = e->ops->check_presence(skb, e);
680 tot_run_sz += run_sz;
681 }
682 }
683
684 return tot_run_sz;
685}
686
687static int tcf_ife_encode(struct sk_buff *skb, const struct tc_action *a,
688 struct tcf_result *res)
689{
690 struct tcf_ife_info *ife = a->priv;
691 int action = ife->tcf_action;
692 struct ethhdr *oethh; /* outer ether header */
693 struct ethhdr *iethh; /* inner eth header */
694 struct tcf_meta_info *e;
695 /*
696 OUTERHDR:TOTMETALEN:{TLVHDR:Metadatum:TLVHDR..}:ORIGDATA
697 where ORIGDATA = original ethernet header ...
698 */
699 u16 metalen = ife_get_sz(skb, ife);
700 int hdrm = metalen + skb->dev->hard_header_len + IFE_METAHDRLEN;
701 unsigned int skboff = skb->dev->hard_header_len;
702 u32 at = G_TC_AT(skb->tc_verd);
703 int new_len = skb->len + hdrm;
704 bool exceed_mtu = false;
705 int err;
706
707 if (at & AT_EGRESS) {
708 if (new_len > skb->dev->mtu)
709 exceed_mtu = true;
710 }
711
712 spin_lock(&ife->tcf_lock);
713 bstats_update(&ife->tcf_bstats, skb);
714 ife->tcf_tm.lastuse = jiffies;
715
716 if (!metalen) { /* no metadata to send */
717 /* abuse overlimits to count when we allow packet
718 * with no metadata
719 */
720 ife->tcf_qstats.overlimits++;
721 spin_unlock(&ife->tcf_lock);
722 return action;
723 }
724 /* could be stupid policy setup or mtu config
725 * so lets be conservative.. */
726 if ((action == TC_ACT_SHOT) || exceed_mtu) {
727 ife->tcf_qstats.drops++;
728 spin_unlock(&ife->tcf_lock);
729 return TC_ACT_SHOT;
730 }
731
732 iethh = eth_hdr(skb);
733
734 err = skb_cow_head(skb, hdrm);
735 if (unlikely(err)) {
736 ife->tcf_qstats.drops++;
737 spin_unlock(&ife->tcf_lock);
738 return TC_ACT_SHOT;
739 }
740
741 if (!(at & AT_EGRESS))
742 skb_push(skb, skb->dev->hard_header_len);
743
744 __skb_push(skb, hdrm);
745 memcpy(skb->data, iethh, skb->mac_len);
746 skb_reset_mac_header(skb);
747 oethh = eth_hdr(skb);
748
749 /*total metadata length */
750 metalen += IFE_METAHDRLEN;
751 metalen = htons(metalen);
752 memcpy((skb->data + skboff), &metalen, IFE_METAHDRLEN);
753 skboff += IFE_METAHDRLEN;
754
755 /* XXX: we dont have a clever way of telling encode to
756 * not repeat some of the computations that are done by
757 * ops->presence_check...
758 */
759 list_for_each_entry(e, &ife->metalist, metalist) {
760 if (e->ops->encode) {
761 err = e->ops->encode(skb, (void *)(skb->data + skboff),
762 e);
763 }
764 if (err < 0) {
765 /* too corrupt to keep around if overwritten */
766 ife->tcf_qstats.drops++;
767 spin_unlock(&ife->tcf_lock);
768 return TC_ACT_SHOT;
769 }
770 skboff += err;
771 }
772
773 if (!is_zero_ether_addr(ife->eth_src))
774 ether_addr_copy(oethh->h_source, ife->eth_src);
775 else
776 ether_addr_copy(oethh->h_source, iethh->h_source);
777 if (!is_zero_ether_addr(ife->eth_dst))
778 ether_addr_copy(oethh->h_dest, ife->eth_dst);
779 else
780 ether_addr_copy(oethh->h_dest, iethh->h_dest);
781 oethh->h_proto = htons(ife->eth_type);
782
783 if (!(at & AT_EGRESS))
784 skb_pull(skb, skb->dev->hard_header_len);
785
786 spin_unlock(&ife->tcf_lock);
787
788 return action;
789}
790
791static int tcf_ife_act(struct sk_buff *skb, const struct tc_action *a,
792 struct tcf_result *res)
793{
794 struct tcf_ife_info *ife = a->priv;
795
796 if (ife->flags & IFE_ENCODE)
797 return tcf_ife_encode(skb, a, res);
798
799 if (!(ife->flags & IFE_ENCODE))
800 return tcf_ife_decode(skb, a, res);
801
802 pr_info_ratelimited("unknown failure(policy neither de/encode\n");
803 spin_lock(&ife->tcf_lock);
804 bstats_update(&ife->tcf_bstats, skb);
805 ife->tcf_tm.lastuse = jiffies;
806 ife->tcf_qstats.drops++;
807 spin_unlock(&ife->tcf_lock);
808
809 return TC_ACT_SHOT;
810}
811
812static int tcf_ife_walker(struct net *net, struct sk_buff *skb,
813 struct netlink_callback *cb, int type,
814 struct tc_action *a)
815{
816 struct tc_action_net *tn = net_generic(net, ife_net_id);
817
818 return tcf_generic_walker(tn, skb, cb, type, a);
819}
820
821static int tcf_ife_search(struct net *net, struct tc_action *a, u32 index)
822{
823 struct tc_action_net *tn = net_generic(net, ife_net_id);
824
825 return tcf_hash_search(tn, a, index);
826}
827
828static struct tc_action_ops act_ife_ops = {
829 .kind = "ife",
830 .type = TCA_ACT_IFE,
831 .owner = THIS_MODULE,
832 .act = tcf_ife_act,
833 .dump = tcf_ife_dump,
834 .cleanup = tcf_ife_cleanup,
835 .init = tcf_ife_init,
836 .walk = tcf_ife_walker,
837 .lookup = tcf_ife_search,
838};
839
840static __net_init int ife_init_net(struct net *net)
841{
842 struct tc_action_net *tn = net_generic(net, ife_net_id);
843
844 return tc_action_net_init(tn, &act_ife_ops, IFE_TAB_MASK);
845}
846
847static void __net_exit ife_exit_net(struct net *net)
848{
849 struct tc_action_net *tn = net_generic(net, ife_net_id);
850
851 tc_action_net_exit(tn);
852}
853
854static struct pernet_operations ife_net_ops = {
855 .init = ife_init_net,
856 .exit = ife_exit_net,
857 .id = &ife_net_id,
858 .size = sizeof(struct tc_action_net),
859};
860
861static int __init ife_init_module(void)
862{
863 return tcf_register_action(&act_ife_ops, &ife_net_ops);
864}
865
866static void __exit ife_cleanup_module(void)
867{
868 tcf_unregister_action(&act_ife_ops, &ife_net_ops);
869}
870
871module_init(ife_init_module);
872module_exit(ife_cleanup_module);
873
874MODULE_AUTHOR("Jamal Hadi Salim(2015)");
875MODULE_DESCRIPTION("Inter-FE LFB action");
876MODULE_LICENSE("GPL");