Revert "net: sched: act_ife: disable bh when taking ife_mod_lock"
[linux-block.git] / net / sched / act_ife.c
CommitLineData
ef6980b6
JHS
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>
295a6e06 35#include <net/ife.h>
ef6980b6 36
c7d03a00 37static unsigned int ife_net_id;
ef6980b6 38static int max_metacnt = IFE_META_MAX + 1;
a85a970a 39static struct tc_action_ops act_ife_ops;
ef6980b6
JHS
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
6a5d58b6
JHS
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
ef6980b6
JHS
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
6a5d58b6
JHS
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
ef6980b6
JHS
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
067a7cd0 118int ife_alloc_meta_u32(struct tcf_meta_info *mi, void *metaval, gfp_t gfp)
ef6980b6 119{
067a7cd0 120 mi->metaval = kmemdup(metaval, sizeof(u32), gfp);
ef6980b6
JHS
121 if (!mi->metaval)
122 return -ENOMEM;
123
124 return 0;
125}
126EXPORT_SYMBOL_GPL(ife_alloc_meta_u32);
127
067a7cd0 128int ife_alloc_meta_u16(struct tcf_meta_info *mi, void *metaval, gfp_t gfp)
ef6980b6 129{
067a7cd0 130 mi->metaval = kmemdup(metaval, sizeof(u16), gfp);
ef6980b6
JHS
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{
28a10c42 146 if (len == sizeof(u32))
ef6980b6
JHS
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{
28a10c42
JHS
155 /* length will not include padding */
156 if (len == sizeof(u16))
ef6980b6
JHS
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
8ce5be1c 170 read_lock(&ife_mod_lock);
ef6980b6
JHS
171 list_for_each_entry(o, &ifeoplist, list) {
172 if (o->metaid == metaid) {
173 if (!try_module_get(o->owner))
174 o = NULL;
8ce5be1c 175 read_unlock(&ife_mod_lock);
ef6980b6
JHS
176 return o;
177 }
178 }
8ce5be1c 179 read_unlock(&ife_mod_lock);
ef6980b6
JHS
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
8ce5be1c 193 write_lock(&ife_mod_lock);
ef6980b6
JHS
194
195 list_for_each_entry(m, &ifeoplist, list) {
196 if (m->metaid == mops->metaid ||
197 (strcmp(mops->name, m->name) == 0)) {
8ce5be1c 198 write_unlock(&ife_mod_lock);
ef6980b6
JHS
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);
8ce5be1c 207 write_unlock(&ife_mod_lock);
ef6980b6
JHS
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
8ce5be1c 217 write_lock(&ife_mod_lock);
ef6980b6
JHS
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 }
8ce5be1c 225 write_unlock(&ife_mod_lock);
ef6980b6
JHS
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
65787594 251#ifdef CONFIG_MODULES
d3f24ba8
RM
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}
65787594 265#endif
d3f24ba8 266
ef6980b6 267/* called when adding new meta information
067a7cd0 268 * under ife->tcf_lock for existing action
ef6980b6
JHS
269*/
270static int load_metaops_and_vet(struct tcf_ife_info *ife, u32 metaid,
54d0d423
VB
271 void *val, int len, bool exists,
272 bool rtnl_held)
ef6980b6
JHS
273{
274 struct tcf_meta_ops *ops = find_ife_oplist(metaid);
275 int ret = 0;
276
277 if (!ops) {
278 ret = -ENOENT;
279#ifdef CONFIG_MODULES
067a7cd0
WC
280 if (exists)
281 spin_unlock_bh(&ife->tcf_lock);
54d0d423
VB
282 if (rtnl_held)
283 rtnl_unlock();
d3f24ba8 284 request_module("ife-meta-%s", ife_meta_id2name(metaid));
54d0d423
VB
285 if (rtnl_held)
286 rtnl_lock();
067a7cd0
WC
287 if (exists)
288 spin_lock_bh(&ife->tcf_lock);
ef6980b6
JHS
289 ops = find_ife_oplist(metaid);
290#endif
291 }
292
293 if (ops) {
294 ret = 0;
295 if (len)
296 ret = ife_validate_metatype(ops, val, len);
297
298 module_put(ops->owner);
299 }
300
301 return ret;
302}
303
304/* called when adding new meta information
067a7cd0 305 * under ife->tcf_lock for existing action
ef6980b6
JHS
306*/
307static int add_metainfo(struct tcf_ife_info *ife, u32 metaid, void *metaval,
817e9f2c 308 int len, bool atomic)
ef6980b6
JHS
309{
310 struct tcf_meta_info *mi = NULL;
311 struct tcf_meta_ops *ops = find_ife_oplist(metaid);
312 int ret = 0;
313
314 if (!ops)
315 return -ENOENT;
316
817e9f2c 317 mi = kzalloc(sizeof(*mi), atomic ? GFP_ATOMIC : GFP_KERNEL);
ef6980b6
JHS
318 if (!mi) {
319 /*put back what find_ife_oplist took */
320 module_put(ops->owner);
321 return -ENOMEM;
322 }
323
324 mi->metaid = metaid;
325 mi->ops = ops;
326 if (len > 0) {
817e9f2c 327 ret = ops->alloc(mi, metaval, atomic ? GFP_ATOMIC : GFP_KERNEL);
ef6980b6
JHS
328 if (ret != 0) {
329 kfree(mi);
330 module_put(ops->owner);
331 return ret;
332 }
333 }
334
335 list_add_tail(&mi->metalist, &ife->metalist);
336
337 return ret;
338}
339
340static int use_all_metadata(struct tcf_ife_info *ife)
341{
342 struct tcf_meta_ops *o;
343 int rc = 0;
344 int installed = 0;
345
8ce5be1c 346 read_lock(&ife_mod_lock);
ef6980b6 347 list_for_each_entry(o, &ifeoplist, list) {
817e9f2c 348 rc = add_metainfo(ife, o->metaid, NULL, 0, true);
ef6980b6
JHS
349 if (rc == 0)
350 installed += 1;
351 }
8ce5be1c 352 read_unlock(&ife_mod_lock);
ef6980b6
JHS
353
354 if (installed)
355 return 0;
356 else
357 return -EINVAL;
358}
359
360static int dump_metalist(struct sk_buff *skb, struct tcf_ife_info *ife)
361{
362 struct tcf_meta_info *e;
363 struct nlattr *nest;
364 unsigned char *b = skb_tail_pointer(skb);
365 int total_encoded = 0;
366
367 /*can only happen on decode */
368 if (list_empty(&ife->metalist))
369 return 0;
370
371 nest = nla_nest_start(skb, TCA_IFE_METALST);
372 if (!nest)
373 goto out_nlmsg_trim;
374
375 list_for_each_entry(e, &ife->metalist, metalist) {
376 if (!e->ops->get(skb, e))
377 total_encoded += 1;
378 }
379
380 if (!total_encoded)
381 goto out_nlmsg_trim;
382
383 nla_nest_end(skb, nest);
384
385 return 0;
386
387out_nlmsg_trim:
388 nlmsg_trim(skb, b);
389 return -1;
390}
391
392/* under ife->tcf_lock */
9a63b255 393static void _tcf_ife_cleanup(struct tc_action *a)
ef6980b6 394{
a85a970a 395 struct tcf_ife_info *ife = to_ife(a);
ef6980b6
JHS
396 struct tcf_meta_info *e, *n;
397
398 list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
399 module_put(e->ops->owner);
400 list_del(&e->metalist);
401 if (e->metaval) {
402 if (e->ops->release)
403 e->ops->release(e);
404 else
405 kfree(e->metaval);
406 }
407 kfree(e);
408 }
409}
410
9a63b255 411static void tcf_ife_cleanup(struct tc_action *a)
ef6980b6 412{
a85a970a 413 struct tcf_ife_info *ife = to_ife(a);
aa9fd9a3 414 struct tcf_ife_params *p;
ef6980b6
JHS
415
416 spin_lock_bh(&ife->tcf_lock);
9a63b255 417 _tcf_ife_cleanup(a);
ef6980b6 418 spin_unlock_bh(&ife->tcf_lock);
aa9fd9a3
AA
419
420 p = rcu_dereference_protected(ife->params, 1);
0a889b94
DC
421 if (p)
422 kfree_rcu(p, rcu);
ef6980b6
JHS
423}
424
067a7cd0
WC
425/* under ife->tcf_lock for existing action */
426static int populate_metalist(struct tcf_ife_info *ife, struct nlattr **tb,
54d0d423 427 bool exists, bool rtnl_held)
ef6980b6
JHS
428{
429 int len = 0;
430 int rc = 0;
431 int i = 0;
432 void *val;
433
434 for (i = 1; i < max_metacnt; i++) {
435 if (tb[i]) {
436 val = nla_data(tb[i]);
437 len = nla_len(tb[i]);
438
54d0d423
VB
439 rc = load_metaops_and_vet(ife, i, val, len, exists,
440 rtnl_held);
ef6980b6
JHS
441 if (rc != 0)
442 return rc;
443
067a7cd0 444 rc = add_metainfo(ife, i, val, len, exists);
ef6980b6
JHS
445 if (rc)
446 return rc;
447 }
448 }
449
450 return rc;
451}
452
453static int tcf_ife_init(struct net *net, struct nlattr *nla,
a85a970a 454 struct nlattr *est, struct tc_action **a,
789871bb
VB
455 int ovr, int bind, bool rtnl_held,
456 struct netlink_ext_ack *extack)
ef6980b6
JHS
457{
458 struct tc_action_net *tn = net_generic(net, ife_net_id);
459 struct nlattr *tb[TCA_IFE_MAX + 1];
460 struct nlattr *tb2[IFE_META_MAX + 1];
54d0d423 461 struct tcf_ife_params *p;
ef6980b6 462 struct tcf_ife_info *ife;
b522ed6e 463 u16 ife_type = ETH_P_IFE;
ef6980b6 464 struct tc_ife *parm;
ef6980b6
JHS
465 u8 *daddr = NULL;
466 u8 *saddr = NULL;
b2313077
WC
467 bool exists = false;
468 int ret = 0;
ef6980b6
JHS
469 int err;
470
fceb6435 471 err = nla_parse_nested(tb, TCA_IFE_MAX, nla, ife_policy, NULL);
ef6980b6
JHS
472 if (err < 0)
473 return err;
474
475 if (!tb[TCA_IFE_PARMS])
476 return -EINVAL;
477
478 parm = nla_data(tb[TCA_IFE_PARMS]);
479
734534e9
AA
480 /* IFE_DECODE is 0 and indicates the opposite of IFE_ENCODE because
481 * they cannot run as the same time. Check on all other values which
482 * are not supported right now.
483 */
484 if (parm->flags & ~IFE_ENCODE)
485 return -EINVAL;
486
aa9fd9a3
AA
487 p = kzalloc(sizeof(*p), GFP_KERNEL);
488 if (!p)
489 return -ENOMEM;
490
0190c1d4 491 err = tcf_idr_check_alloc(tn, &parm->index, a, bind);
01e866bf
VB
492 if (err < 0) {
493 kfree(p);
0190c1d4 494 return err;
01e866bf 495 }
0190c1d4 496 exists = err;
aa9fd9a3
AA
497 if (exists && bind) {
498 kfree(p);
4e8c8615 499 return 0;
aa9fd9a3 500 }
4e8c8615 501
4e8c8615 502 if (!exists) {
65a206c0 503 ret = tcf_idr_create(tn, parm->index, est, a, &act_ife_ops,
ced273ea 504 bind, true);
aa9fd9a3 505 if (ret) {
0190c1d4 506 tcf_idr_cleanup(tn, parm->index);
aa9fd9a3 507 kfree(p);
ef6980b6 508 return ret;
aa9fd9a3 509 }
ef6980b6 510 ret = ACT_P_CREATED;
4e8ddd7f 511 } else if (!ovr) {
65a206c0 512 tcf_idr_release(*a, bind);
4e8ddd7f
VB
513 kfree(p);
514 return -EEXIST;
ef6980b6
JHS
515 }
516
a85a970a 517 ife = to_ife(*a);
aa9fd9a3 518 p->flags = parm->flags;
ef6980b6
JHS
519
520 if (parm->flags & IFE_ENCODE) {
b522ed6e
AA
521 if (tb[TCA_IFE_TYPE])
522 ife_type = nla_get_u16(tb[TCA_IFE_TYPE]);
ef6980b6
JHS
523 if (tb[TCA_IFE_DMAC])
524 daddr = nla_data(tb[TCA_IFE_DMAC]);
525 if (tb[TCA_IFE_SMAC])
526 saddr = nla_data(tb[TCA_IFE_SMAC]);
527 }
528
ef6980b6
JHS
529 if (parm->flags & IFE_ENCODE) {
530 if (daddr)
aa9fd9a3 531 ether_addr_copy(p->eth_dst, daddr);
ef6980b6 532 else
aa9fd9a3 533 eth_zero_addr(p->eth_dst);
ef6980b6
JHS
534
535 if (saddr)
aa9fd9a3 536 ether_addr_copy(p->eth_src, saddr);
ef6980b6 537 else
aa9fd9a3 538 eth_zero_addr(p->eth_src);
ef6980b6 539
aa9fd9a3 540 p->eth_type = ife_type;
ef6980b6
JHS
541 }
542
aa9fd9a3
AA
543 if (exists)
544 spin_lock_bh(&ife->tcf_lock);
545
ef6980b6
JHS
546 if (ret == ACT_P_CREATED)
547 INIT_LIST_HEAD(&ife->metalist);
548
549 if (tb[TCA_IFE_METALST]) {
550 err = nla_parse_nested(tb2, IFE_META_MAX, tb[TCA_IFE_METALST],
fceb6435 551 NULL, NULL);
ef6980b6
JHS
552 if (err) {
553metadata_parse_err:
067a7cd0
WC
554 if (exists)
555 spin_unlock_bh(&ife->tcf_lock);
4e8ddd7f
VB
556 tcf_idr_release(*a, bind);
557
aa9fd9a3 558 kfree(p);
ef6980b6
JHS
559 return err;
560 }
561
54d0d423 562 err = populate_metalist(ife, tb2, exists, rtnl_held);
ef6980b6
JHS
563 if (err)
564 goto metadata_parse_err;
565
566 } else {
567 /* if no passed metadata allow list or passed allow-all
568 * then here we process by adding as many supported metadatum
569 * as we can. You better have at least one else we are
570 * going to bail out
571 */
572 err = use_all_metadata(ife);
573 if (err) {
067a7cd0
WC
574 if (exists)
575 spin_unlock_bh(&ife->tcf_lock);
32039eac
VB
576 tcf_idr_release(*a, bind);
577
aa9fd9a3 578 kfree(p);
ef6980b6
JHS
579 return err;
580 }
581 }
582
cbf56c29 583 ife->tcf_action = parm->action;
54d0d423
VB
584 /* protected by tcf_lock when modifying existing action */
585 rcu_swap_protected(ife->params, p, 1);
586
067a7cd0
WC
587 if (exists)
588 spin_unlock_bh(&ife->tcf_lock);
54d0d423
VB
589 if (p)
590 kfree_rcu(p, rcu);
aa9fd9a3 591
ef6980b6 592 if (ret == ACT_P_CREATED)
65a206c0 593 tcf_idr_insert(tn, *a);
ef6980b6
JHS
594
595 return ret;
596}
597
598static int tcf_ife_dump(struct sk_buff *skb, struct tc_action *a, int bind,
599 int ref)
600{
601 unsigned char *b = skb_tail_pointer(skb);
a85a970a 602 struct tcf_ife_info *ife = to_ife(a);
54d0d423 603 struct tcf_ife_params *p;
ef6980b6
JHS
604 struct tc_ife opt = {
605 .index = ife->tcf_index,
036bb443
VB
606 .refcnt = refcount_read(&ife->tcf_refcnt) - ref,
607 .bindcnt = atomic_read(&ife->tcf_bindcnt) - bind,
ef6980b6
JHS
608 };
609 struct tcf_t t;
610
54d0d423
VB
611 spin_lock_bh(&ife->tcf_lock);
612 opt.action = ife->tcf_action;
613 p = rcu_dereference_protected(ife->params,
614 lockdep_is_held(&ife->tcf_lock));
615 opt.flags = p->flags;
616
ef6980b6
JHS
617 if (nla_put(skb, TCA_IFE_PARMS, sizeof(opt), &opt))
618 goto nla_put_failure;
619
48d8ee16 620 tcf_tm_dump(&t, &ife->tcf_tm);
9854518e 621 if (nla_put_64bit(skb, TCA_IFE_TM, sizeof(t), &t, TCA_IFE_PAD))
ef6980b6
JHS
622 goto nla_put_failure;
623
aa9fd9a3
AA
624 if (!is_zero_ether_addr(p->eth_dst)) {
625 if (nla_put(skb, TCA_IFE_DMAC, ETH_ALEN, p->eth_dst))
ef6980b6
JHS
626 goto nla_put_failure;
627 }
628
aa9fd9a3
AA
629 if (!is_zero_ether_addr(p->eth_src)) {
630 if (nla_put(skb, TCA_IFE_SMAC, ETH_ALEN, p->eth_src))
ef6980b6
JHS
631 goto nla_put_failure;
632 }
633
aa9fd9a3 634 if (nla_put(skb, TCA_IFE_TYPE, 2, &p->eth_type))
ef6980b6
JHS
635 goto nla_put_failure;
636
637 if (dump_metalist(skb, ife)) {
638 /*ignore failure to dump metalist */
639 pr_info("Failed to dump metalist\n");
640 }
641
54d0d423 642 spin_unlock_bh(&ife->tcf_lock);
ef6980b6
JHS
643 return skb->len;
644
645nla_put_failure:
54d0d423 646 spin_unlock_bh(&ife->tcf_lock);
ef6980b6
JHS
647 nlmsg_trim(skb, b);
648 return -1;
649}
650
4dba87b0
OG
651static int find_decode_metaid(struct sk_buff *skb, struct tcf_ife_info *ife,
652 u16 metaid, u16 mlen, void *mdata)
ef6980b6
JHS
653{
654 struct tcf_meta_info *e;
655
656 /* XXX: use hash to speed up */
657 list_for_each_entry(e, &ife->metalist, metalist) {
658 if (metaid == e->metaid) {
659 if (e->ops) {
660 /* We check for decode presence already */
661 return e->ops->decode(skb, mdata, mlen);
662 }
663 }
664 }
665
f6cd1453 666 return -ENOENT;
ef6980b6
JHS
667}
668
ef6980b6
JHS
669static int tcf_ife_decode(struct sk_buff *skb, const struct tc_action *a,
670 struct tcf_result *res)
671{
a85a970a 672 struct tcf_ife_info *ife = to_ife(a);
ef6980b6 673 int action = ife->tcf_action;
295a6e06
YG
674 u8 *ifehdr_end;
675 u8 *tlv_data;
676 u16 metalen;
ef6980b6 677
ced273ea 678 bstats_cpu_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
9c4a4e48 679 tcf_lastuse_update(&ife->tcf_tm);
ef6980b6 680
295a6e06
YG
681 if (skb_at_tc_ingress(skb))
682 skb_push(skb, skb->dev->hard_header_len);
683
684 tlv_data = ife_decode(skb, &metalen);
685 if (unlikely(!tlv_data)) {
ced273ea 686 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
ef6980b6
JHS
687 return TC_ACT_SHOT;
688 }
689
295a6e06
YG
690 ifehdr_end = tlv_data + metalen;
691 for (; tlv_data < ifehdr_end; tlv_data = ife_tlv_meta_next(tlv_data)) {
692 u8 *curr_data;
693 u16 mtype;
694 u16 dlen;
ef6980b6 695
cc74eddd
AA
696 curr_data = ife_tlv_meta_decode(tlv_data, ifehdr_end, &mtype,
697 &dlen, NULL);
698 if (!curr_data) {
699 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
700 return TC_ACT_SHOT;
701 }
ef6980b6 702
295a6e06 703 if (find_decode_metaid(skb, ife, mtype, dlen, curr_data)) {
ef6980b6
JHS
704 /* abuse overlimits to count when we receive metadata
705 * but dont have an ops for it
706 */
295a6e06
YG
707 pr_info_ratelimited("Unknown metaid %d dlen %d\n",
708 mtype, dlen);
ced273ea 709 qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
ef6980b6 710 }
295a6e06 711 }
ef6980b6 712
295a6e06 713 if (WARN_ON(tlv_data != ifehdr_end)) {
ced273ea 714 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
295a6e06 715 return TC_ACT_SHOT;
ef6980b6
JHS
716 }
717
295a6e06 718 skb->protocol = eth_type_trans(skb, skb->dev);
ef6980b6 719 skb_reset_network_header(skb);
295a6e06 720
ef6980b6
JHS
721 return action;
722}
723
724/*XXX: check if we can do this at install time instead of current
725 * send data path
726**/
727static int ife_get_sz(struct sk_buff *skb, struct tcf_ife_info *ife)
728{
729 struct tcf_meta_info *e, *n;
730 int tot_run_sz = 0, run_sz = 0;
731
732 list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
733 if (e->ops->check_presence) {
734 run_sz = e->ops->check_presence(skb, e);
735 tot_run_sz += run_sz;
736 }
737 }
738
739 return tot_run_sz;
740}
741
742static int tcf_ife_encode(struct sk_buff *skb, const struct tc_action *a,
aa9fd9a3 743 struct tcf_result *res, struct tcf_ife_params *p)
ef6980b6 744{
a85a970a 745 struct tcf_ife_info *ife = to_ife(a);
ef6980b6
JHS
746 int action = ife->tcf_action;
747 struct ethhdr *oethh; /* outer ether header */
ef6980b6
JHS
748 struct tcf_meta_info *e;
749 /*
750 OUTERHDR:TOTMETALEN:{TLVHDR:Metadatum:TLVHDR..}:ORIGDATA
751 where ORIGDATA = original ethernet header ...
752 */
753 u16 metalen = ife_get_sz(skb, ife);
754 int hdrm = metalen + skb->dev->hard_header_len + IFE_METAHDRLEN;
295a6e06 755 unsigned int skboff = 0;
ef6980b6
JHS
756 int new_len = skb->len + hdrm;
757 bool exceed_mtu = false;
295a6e06
YG
758 void *ife_meta;
759 int err = 0;
ef6980b6 760
a5135bcf 761 if (!skb_at_tc_ingress(skb)) {
ef6980b6
JHS
762 if (new_len > skb->dev->mtu)
763 exceed_mtu = true;
764 }
765
ced273ea 766 bstats_cpu_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
9c4a4e48 767 tcf_lastuse_update(&ife->tcf_tm);
ef6980b6
JHS
768
769 if (!metalen) { /* no metadata to send */
770 /* abuse overlimits to count when we allow packet
771 * with no metadata
772 */
ced273ea 773 qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
ef6980b6
JHS
774 return action;
775 }
776 /* could be stupid policy setup or mtu config
777 * so lets be conservative.. */
778 if ((action == TC_ACT_SHOT) || exceed_mtu) {
ced273ea 779 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
ef6980b6
JHS
780 return TC_ACT_SHOT;
781 }
782
a5135bcf 783 if (skb_at_tc_ingress(skb))
ef6980b6
JHS
784 skb_push(skb, skb->dev->hard_header_len);
785
295a6e06 786 ife_meta = ife_encode(skb, metalen);
ef6980b6 787
ced273ea
AA
788 spin_lock(&ife->tcf_lock);
789
ef6980b6
JHS
790 /* XXX: we dont have a clever way of telling encode to
791 * not repeat some of the computations that are done by
792 * ops->presence_check...
793 */
794 list_for_each_entry(e, &ife->metalist, metalist) {
795 if (e->ops->encode) {
295a6e06 796 err = e->ops->encode(skb, (void *)(ife_meta + skboff),
ef6980b6
JHS
797 e);
798 }
799 if (err < 0) {
800 /* too corrupt to keep around if overwritten */
ef6980b6 801 spin_unlock(&ife->tcf_lock);
ced273ea 802 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
ef6980b6
JHS
803 return TC_ACT_SHOT;
804 }
805 skboff += err;
806 }
aa9fd9a3 807 spin_unlock(&ife->tcf_lock);
295a6e06 808 oethh = (struct ethhdr *)skb->data;
ef6980b6 809
aa9fd9a3
AA
810 if (!is_zero_ether_addr(p->eth_src))
811 ether_addr_copy(oethh->h_source, p->eth_src);
812 if (!is_zero_ether_addr(p->eth_dst))
813 ether_addr_copy(oethh->h_dest, p->eth_dst);
814 oethh->h_proto = htons(p->eth_type);
ef6980b6 815
a5135bcf 816 if (skb_at_tc_ingress(skb))
ef6980b6
JHS
817 skb_pull(skb, skb->dev->hard_header_len);
818
ef6980b6
JHS
819 return action;
820}
821
822static int tcf_ife_act(struct sk_buff *skb, const struct tc_action *a,
823 struct tcf_result *res)
824{
a85a970a 825 struct tcf_ife_info *ife = to_ife(a);
aa9fd9a3
AA
826 struct tcf_ife_params *p;
827 int ret;
828
7fd4b288 829 p = rcu_dereference_bh(ife->params);
aa9fd9a3
AA
830 if (p->flags & IFE_ENCODE) {
831 ret = tcf_ife_encode(skb, a, res, p);
aa9fd9a3
AA
832 return ret;
833 }
ef6980b6 834
734534e9 835 return tcf_ife_decode(skb, a, res);
ef6980b6
JHS
836}
837
838static int tcf_ife_walker(struct net *net, struct sk_buff *skb,
839 struct netlink_callback *cb, int type,
41780105
AA
840 const struct tc_action_ops *ops,
841 struct netlink_ext_ack *extack)
ef6980b6
JHS
842{
843 struct tc_action_net *tn = net_generic(net, ife_net_id);
844
b3620145 845 return tcf_generic_walker(tn, skb, cb, type, ops, extack);
ef6980b6
JHS
846}
847
331a9295
AA
848static int tcf_ife_search(struct net *net, struct tc_action **a, u32 index,
849 struct netlink_ext_ack *extack)
ef6980b6
JHS
850{
851 struct tc_action_net *tn = net_generic(net, ife_net_id);
852
65a206c0 853 return tcf_idr_search(tn, a, index);
ef6980b6
JHS
854}
855
856static struct tc_action_ops act_ife_ops = {
857 .kind = "ife",
858 .type = TCA_ACT_IFE,
859 .owner = THIS_MODULE,
860 .act = tcf_ife_act,
861 .dump = tcf_ife_dump,
862 .cleanup = tcf_ife_cleanup,
863 .init = tcf_ife_init,
864 .walk = tcf_ife_walker,
865 .lookup = tcf_ife_search,
a85a970a 866 .size = sizeof(struct tcf_ife_info),
ef6980b6
JHS
867};
868
869static __net_init int ife_init_net(struct net *net)
870{
871 struct tc_action_net *tn = net_generic(net, ife_net_id);
872
c7e460ce 873 return tc_action_net_init(tn, &act_ife_ops);
ef6980b6
JHS
874}
875
039af9c6 876static void __net_exit ife_exit_net(struct list_head *net_list)
ef6980b6 877{
039af9c6 878 tc_action_net_exit(net_list, ife_net_id);
ef6980b6
JHS
879}
880
881static struct pernet_operations ife_net_ops = {
882 .init = ife_init_net,
039af9c6 883 .exit_batch = ife_exit_net,
ef6980b6
JHS
884 .id = &ife_net_id,
885 .size = sizeof(struct tc_action_net),
886};
887
888static int __init ife_init_module(void)
889{
890 return tcf_register_action(&act_ife_ops, &ife_net_ops);
891}
892
893static void __exit ife_cleanup_module(void)
894{
895 tcf_unregister_action(&act_ife_ops, &ife_net_ops);
896}
897
898module_init(ife_init_module);
899module_exit(ife_cleanup_module);
900
901MODULE_AUTHOR("Jamal Hadi Salim(2015)");
902MODULE_DESCRIPTION("Inter-FE LFB action");
903MODULE_LICENSE("GPL");