1
2
3
4
5
6
7
8#include <linux/module.h>
9#include <linux/types.h>
10#include <linux/kernel.h>
11#include <linux/skbuff.h>
12#include <linux/errno.h>
13#include <linux/slab.h>
14#include <linux/refcount.h>
15#include <net/act_api.h>
16#include <net/netlink.h>
17#include <net/pkt_cls.h>
18#include <net/sch_generic.h>
19
20
21
22
23
24
25
26#define PERFECT_HASH_THRESHOLD 64
27#define DEFAULT_HASH_SIZE 64
28
29
30struct tcindex_data;
31
32struct tcindex_filter_result {
33 struct tcf_exts exts;
34 struct tcf_result res;
35 struct tcindex_data *p;
36 struct rcu_work rwork;
37};
38
39struct tcindex_filter {
40 u16 key;
41 struct tcindex_filter_result result;
42 struct tcindex_filter __rcu *next;
43 struct rcu_work rwork;
44};
45
46
47struct tcindex_data {
48 struct tcindex_filter_result *perfect;
49 struct tcindex_filter __rcu **h;
50 struct tcf_proto *tp;
51 u16 mask;
52 u32 shift;
53 u32 hash;
54 u32 alloc_hash;
55 u32 fall_through;
56 refcount_t refcnt;
57 struct rcu_work rwork;
58};
59
60static inline int tcindex_filter_is_set(struct tcindex_filter_result *r)
61{
62 return tcf_exts_has_actions(&r->exts) || r->res.classid;
63}
64
65static void tcindex_data_get(struct tcindex_data *p)
66{
67 refcount_inc(&p->refcnt);
68}
69
70static void tcindex_data_put(struct tcindex_data *p)
71{
72 if (refcount_dec_and_test(&p->refcnt)) {
73 kfree(p->perfect);
74 kfree(p->h);
75 kfree(p);
76 }
77}
78
79static struct tcindex_filter_result *tcindex_lookup(struct tcindex_data *p,
80 u16 key)
81{
82 if (p->perfect) {
83 struct tcindex_filter_result *f = p->perfect + key;
84
85 return tcindex_filter_is_set(f) ? f : NULL;
86 } else if (p->h) {
87 struct tcindex_filter __rcu **fp;
88 struct tcindex_filter *f;
89
90 fp = &p->h[key % p->hash];
91 for (f = rcu_dereference_bh_rtnl(*fp);
92 f;
93 fp = &f->next, f = rcu_dereference_bh_rtnl(*fp))
94 if (f->key == key)
95 return &f->result;
96 }
97
98 return NULL;
99}
100
101
102static int tcindex_classify(struct sk_buff *skb, const struct tcf_proto *tp,
103 struct tcf_result *res)
104{
105 struct tcindex_data *p = rcu_dereference_bh(tp->root);
106 struct tcindex_filter_result *f;
107 int key = (skb->tc_index & p->mask) >> p->shift;
108
109 pr_debug("tcindex_classify(skb %p,tp %p,res %p),p %p\n",
110 skb, tp, res, p);
111
112 f = tcindex_lookup(p, key);
113 if (!f) {
114 struct Qdisc *q = tcf_block_q(tp->chain->block);
115
116 if (!p->fall_through)
117 return -1;
118 res->classid = TC_H_MAKE(TC_H_MAJ(q->handle), key);
119 res->class = 0;
120 pr_debug("alg 0x%x\n", res->classid);
121 return 0;
122 }
123 *res = f->res;
124 pr_debug("map 0x%x\n", res->classid);
125
126 return tcf_exts_exec(skb, &f->exts, res);
127}
128
129
130static void *tcindex_get(struct tcf_proto *tp, u32 handle)
131{
132 struct tcindex_data *p = rtnl_dereference(tp->root);
133 struct tcindex_filter_result *r;
134
135 pr_debug("tcindex_get(tp %p,handle 0x%08x)\n", tp, handle);
136 if (p->perfect && handle >= p->alloc_hash)
137 return NULL;
138 r = tcindex_lookup(p, handle);
139 return r && tcindex_filter_is_set(r) ? r : NULL;
140}
141
142static int tcindex_init(struct tcf_proto *tp)
143{
144 struct tcindex_data *p;
145
146 pr_debug("tcindex_init(tp %p)\n", tp);
147 p = kzalloc(sizeof(struct tcindex_data), GFP_KERNEL);
148 if (!p)
149 return -ENOMEM;
150
151 p->mask = 0xffff;
152 p->hash = DEFAULT_HASH_SIZE;
153 p->fall_through = 1;
154 refcount_set(&p->refcnt, 1);
155
156 rcu_assign_pointer(tp->root, p);
157 return 0;
158}
159
160static void __tcindex_destroy_rexts(struct tcindex_filter_result *r)
161{
162 tcf_exts_destroy(&r->exts);
163 tcf_exts_put_net(&r->exts);
164 tcindex_data_put(r->p);
165}
166
167static void tcindex_destroy_rexts_work(struct work_struct *work)
168{
169 struct tcindex_filter_result *r;
170
171 r = container_of(to_rcu_work(work),
172 struct tcindex_filter_result,
173 rwork);
174 rtnl_lock();
175 __tcindex_destroy_rexts(r);
176 rtnl_unlock();
177}
178
179static void __tcindex_destroy_fexts(struct tcindex_filter *f)
180{
181 tcf_exts_destroy(&f->result.exts);
182 tcf_exts_put_net(&f->result.exts);
183 kfree(f);
184}
185
186static void tcindex_destroy_fexts_work(struct work_struct *work)
187{
188 struct tcindex_filter *f = container_of(to_rcu_work(work),
189 struct tcindex_filter,
190 rwork);
191
192 rtnl_lock();
193 __tcindex_destroy_fexts(f);
194 rtnl_unlock();
195}
196
197static int tcindex_delete(struct tcf_proto *tp, void *arg, bool *last,
198 bool rtnl_held, struct netlink_ext_ack *extack)
199{
200 struct tcindex_data *p = rtnl_dereference(tp->root);
201 struct tcindex_filter_result *r = arg;
202 struct tcindex_filter __rcu **walk;
203 struct tcindex_filter *f = NULL;
204
205 pr_debug("tcindex_delete(tp %p,arg %p),p %p\n", tp, arg, p);
206 if (p->perfect) {
207 if (!r->res.class)
208 return -ENOENT;
209 } else {
210 int i;
211
212 for (i = 0; i < p->hash; i++) {
213 walk = p->h + i;
214 for (f = rtnl_dereference(*walk); f;
215 walk = &f->next, f = rtnl_dereference(*walk)) {
216 if (&f->result == r)
217 goto found;
218 }
219 }
220 return -ENOENT;
221
222found:
223 rcu_assign_pointer(*walk, rtnl_dereference(f->next));
224 }
225 tcf_unbind_filter(tp, &r->res);
226
227
228
229
230 if (f) {
231 if (tcf_exts_get_net(&f->result.exts))
232 tcf_queue_work(&f->rwork, tcindex_destroy_fexts_work);
233 else
234 __tcindex_destroy_fexts(f);
235 } else {
236 tcindex_data_get(p);
237
238 if (tcf_exts_get_net(&r->exts))
239 tcf_queue_work(&r->rwork, tcindex_destroy_rexts_work);
240 else
241 __tcindex_destroy_rexts(r);
242 }
243
244 *last = false;
245 return 0;
246}
247
248static void tcindex_destroy_work(struct work_struct *work)
249{
250 struct tcindex_data *p = container_of(to_rcu_work(work),
251 struct tcindex_data,
252 rwork);
253
254 tcindex_data_put(p);
255}
256
257static inline int
258valid_perfect_hash(struct tcindex_data *p)
259{
260 return p->hash > (p->mask >> p->shift);
261}
262
263static const struct nla_policy tcindex_policy[TCA_TCINDEX_MAX + 1] = {
264 [TCA_TCINDEX_HASH] = { .type = NLA_U32 },
265 [TCA_TCINDEX_MASK] = { .type = NLA_U16 },
266 [TCA_TCINDEX_SHIFT] = { .type = NLA_U32 },
267 [TCA_TCINDEX_FALL_THROUGH] = { .type = NLA_U32 },
268 [TCA_TCINDEX_CLASSID] = { .type = NLA_U32 },
269};
270
271static int tcindex_filter_result_init(struct tcindex_filter_result *r,
272 struct tcindex_data *p,
273 struct net *net)
274{
275 memset(r, 0, sizeof(*r));
276 r->p = p;
277 return tcf_exts_init(&r->exts, net, TCA_TCINDEX_ACT,
278 TCA_TCINDEX_POLICE);
279}
280
281static void tcindex_free_perfect_hash(struct tcindex_data *cp);
282
283static void tcindex_partial_destroy_work(struct work_struct *work)
284{
285 struct tcindex_data *p = container_of(to_rcu_work(work),
286 struct tcindex_data,
287 rwork);
288
289 rtnl_lock();
290 if (p->perfect)
291 tcindex_free_perfect_hash(p);
292 kfree(p);
293 rtnl_unlock();
294}
295
296static void tcindex_free_perfect_hash(struct tcindex_data *cp)
297{
298 int i;
299
300 for (i = 0; i < cp->hash; i++)
301 tcf_exts_destroy(&cp->perfect[i].exts);
302 kfree(cp->perfect);
303}
304
305static int tcindex_alloc_perfect_hash(struct net *net, struct tcindex_data *cp)
306{
307 int i, err = 0;
308
309 cp->perfect = kcalloc(cp->hash, sizeof(struct tcindex_filter_result),
310 GFP_KERNEL | __GFP_NOWARN);
311 if (!cp->perfect)
312 return -ENOMEM;
313
314 for (i = 0; i < cp->hash; i++) {
315 err = tcf_exts_init(&cp->perfect[i].exts, net,
316 TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
317 if (err < 0)
318 goto errout;
319 cp->perfect[i].p = cp;
320 }
321
322 return 0;
323
324errout:
325 tcindex_free_perfect_hash(cp);
326 return err;
327}
328
329static int
330tcindex_set_parms(struct net *net, struct tcf_proto *tp, unsigned long base,
331 u32 handle, struct tcindex_data *p,
332 struct tcindex_filter_result *r, struct nlattr **tb,
333 struct nlattr *est, bool ovr, struct netlink_ext_ack *extack)
334{
335 struct tcindex_filter_result new_filter_result, *old_r = r;
336 struct tcindex_data *cp = NULL, *oldp;
337 struct tcindex_filter *f = NULL;
338 struct tcf_result cr = {};
339 int err, balloc = 0;
340 struct tcf_exts e;
341
342 err = tcf_exts_init(&e, net, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
343 if (err < 0)
344 return err;
345 err = tcf_exts_validate(net, tp, tb, est, &e, ovr, true, extack);
346 if (err < 0)
347 goto errout;
348
349 err = -ENOMEM;
350
351
352
353
354 cp = kzalloc(sizeof(*cp), GFP_KERNEL);
355 if (!cp)
356 goto errout;
357
358 cp->mask = p->mask;
359 cp->shift = p->shift;
360 cp->hash = p->hash;
361 cp->alloc_hash = p->alloc_hash;
362 cp->fall_through = p->fall_through;
363 cp->tp = tp;
364 refcount_set(&cp->refcnt, 1);
365
366 if (tb[TCA_TCINDEX_HASH])
367 cp->hash = nla_get_u32(tb[TCA_TCINDEX_HASH]);
368
369 if (tb[TCA_TCINDEX_MASK])
370 cp->mask = nla_get_u16(tb[TCA_TCINDEX_MASK]);
371
372 if (tb[TCA_TCINDEX_SHIFT]) {
373 cp->shift = nla_get_u32(tb[TCA_TCINDEX_SHIFT]);
374 if (cp->shift > 16) {
375 err = -EINVAL;
376 goto errout;
377 }
378 }
379 if (!cp->hash) {
380
381
382
383 if ((cp->mask >> cp->shift) < PERFECT_HASH_THRESHOLD)
384 cp->hash = (cp->mask >> cp->shift) + 1;
385 else
386 cp->hash = DEFAULT_HASH_SIZE;
387 }
388
389 if (p->perfect) {
390 int i;
391
392 if (tcindex_alloc_perfect_hash(net, cp) < 0)
393 goto errout;
394 cp->alloc_hash = cp->hash;
395 for (i = 0; i < min(cp->hash, p->hash); i++)
396 cp->perfect[i].res = p->perfect[i].res;
397 balloc = 1;
398 }
399 cp->h = p->h;
400
401 err = tcindex_filter_result_init(&new_filter_result, cp, net);
402 if (err < 0)
403 goto errout_alloc;
404 if (old_r)
405 cr = r->res;
406
407 err = -EBUSY;
408
409
410
411
412 if (cp->perfect) {
413 if (!valid_perfect_hash(cp) ||
414 cp->hash > cp->alloc_hash)
415 goto errout_alloc;
416 } else if (cp->h && cp->hash != cp->alloc_hash) {
417 goto errout_alloc;
418 }
419
420 err = -EINVAL;
421 if (tb[TCA_TCINDEX_FALL_THROUGH])
422 cp->fall_through = nla_get_u32(tb[TCA_TCINDEX_FALL_THROUGH]);
423
424 if (!cp->perfect && !cp->h)
425 cp->alloc_hash = cp->hash;
426
427
428
429
430
431
432 if (cp->perfect || valid_perfect_hash(cp))
433 if (handle >= cp->alloc_hash)
434 goto errout_alloc;
435
436
437 err = -ENOMEM;
438 if (!cp->perfect && !cp->h) {
439 if (valid_perfect_hash(cp)) {
440 if (tcindex_alloc_perfect_hash(net, cp) < 0)
441 goto errout_alloc;
442 balloc = 1;
443 } else {
444 struct tcindex_filter __rcu **hash;
445
446 hash = kcalloc(cp->hash,
447 sizeof(struct tcindex_filter *),
448 GFP_KERNEL);
449
450 if (!hash)
451 goto errout_alloc;
452
453 cp->h = hash;
454 balloc = 2;
455 }
456 }
457
458 if (cp->perfect)
459 r = cp->perfect + handle;
460 else
461 r = tcindex_lookup(cp, handle) ? : &new_filter_result;
462
463 if (r == &new_filter_result) {
464 f = kzalloc(sizeof(*f), GFP_KERNEL);
465 if (!f)
466 goto errout_alloc;
467 f->key = handle;
468 f->next = NULL;
469 err = tcindex_filter_result_init(&f->result, cp, net);
470 if (err < 0) {
471 kfree(f);
472 goto errout_alloc;
473 }
474 }
475
476 if (tb[TCA_TCINDEX_CLASSID]) {
477 cr.classid = nla_get_u32(tb[TCA_TCINDEX_CLASSID]);
478 tcf_bind_filter(tp, &cr, base);
479 }
480
481 if (old_r && old_r != r) {
482 err = tcindex_filter_result_init(old_r, cp, net);
483 if (err < 0) {
484 kfree(f);
485 goto errout_alloc;
486 }
487 }
488
489 oldp = p;
490 r->res = cr;
491 tcf_exts_change(&r->exts, &e);
492
493 rcu_assign_pointer(tp->root, cp);
494
495 if (r == &new_filter_result) {
496 struct tcindex_filter *nfp;
497 struct tcindex_filter __rcu **fp;
498
499 f->result.res = r->res;
500 tcf_exts_change(&f->result.exts, &r->exts);
501
502 fp = cp->h + (handle % cp->hash);
503 for (nfp = rtnl_dereference(*fp);
504 nfp;
505 fp = &nfp->next, nfp = rtnl_dereference(*fp))
506 ;
507
508 rcu_assign_pointer(*fp, f);
509 } else {
510 tcf_exts_destroy(&new_filter_result.exts);
511 }
512
513 if (oldp)
514 tcf_queue_work(&oldp->rwork, tcindex_partial_destroy_work);
515 return 0;
516
517errout_alloc:
518 if (balloc == 1)
519 tcindex_free_perfect_hash(cp);
520 else if (balloc == 2)
521 kfree(cp->h);
522 tcf_exts_destroy(&new_filter_result.exts);
523errout:
524 kfree(cp);
525 tcf_exts_destroy(&e);
526 return err;
527}
528
529static int
530tcindex_change(struct net *net, struct sk_buff *in_skb,
531 struct tcf_proto *tp, unsigned long base, u32 handle,
532 struct nlattr **tca, void **arg, bool ovr,
533 bool rtnl_held, struct netlink_ext_ack *extack)
534{
535 struct nlattr *opt = tca[TCA_OPTIONS];
536 struct nlattr *tb[TCA_TCINDEX_MAX + 1];
537 struct tcindex_data *p = rtnl_dereference(tp->root);
538 struct tcindex_filter_result *r = *arg;
539 int err;
540
541 pr_debug("tcindex_change(tp %p,handle 0x%08x,tca %p,arg %p),opt %p,"
542 "p %p,r %p,*arg %p\n",
543 tp, handle, tca, arg, opt, p, r, *arg);
544
545 if (!opt)
546 return 0;
547
548 err = nla_parse_nested_deprecated(tb, TCA_TCINDEX_MAX, opt,
549 tcindex_policy, NULL);
550 if (err < 0)
551 return err;
552
553 return tcindex_set_parms(net, tp, base, handle, p, r, tb,
554 tca[TCA_RATE], ovr, extack);
555}
556
557static void tcindex_walk(struct tcf_proto *tp, struct tcf_walker *walker,
558 bool rtnl_held)
559{
560 struct tcindex_data *p = rtnl_dereference(tp->root);
561 struct tcindex_filter *f, *next;
562 int i;
563
564 pr_debug("tcindex_walk(tp %p,walker %p),p %p\n", tp, walker, p);
565 if (p->perfect) {
566 for (i = 0; i < p->hash; i++) {
567 if (!p->perfect[i].res.class)
568 continue;
569 if (walker->count >= walker->skip) {
570 if (walker->fn(tp, p->perfect + i, walker) < 0) {
571 walker->stop = 1;
572 return;
573 }
574 }
575 walker->count++;
576 }
577 }
578 if (!p->h)
579 return;
580 for (i = 0; i < p->hash; i++) {
581 for (f = rtnl_dereference(p->h[i]); f; f = next) {
582 next = rtnl_dereference(f->next);
583 if (walker->count >= walker->skip) {
584 if (walker->fn(tp, &f->result, walker) < 0) {
585 walker->stop = 1;
586 return;
587 }
588 }
589 walker->count++;
590 }
591 }
592}
593
594static void tcindex_destroy(struct tcf_proto *tp, bool rtnl_held,
595 struct netlink_ext_ack *extack)
596{
597 struct tcindex_data *p = rtnl_dereference(tp->root);
598 int i;
599
600 pr_debug("tcindex_destroy(tp %p),p %p\n", tp, p);
601
602 if (p->perfect) {
603 for (i = 0; i < p->hash; i++) {
604 struct tcindex_filter_result *r = p->perfect + i;
605
606
607
608
609
610
611
612 tcindex_data_get(p);
613
614 tcf_unbind_filter(tp, &r->res);
615 if (tcf_exts_get_net(&r->exts))
616 tcf_queue_work(&r->rwork,
617 tcindex_destroy_rexts_work);
618 else
619 __tcindex_destroy_rexts(r);
620 }
621 }
622
623 for (i = 0; p->h && i < p->hash; i++) {
624 struct tcindex_filter *f, *next;
625 bool last;
626
627 for (f = rtnl_dereference(p->h[i]); f; f = next) {
628 next = rtnl_dereference(f->next);
629 tcindex_delete(tp, &f->result, &last, rtnl_held, NULL);
630 }
631 }
632
633 tcf_queue_work(&p->rwork, tcindex_destroy_work);
634}
635
636
637static int tcindex_dump(struct net *net, struct tcf_proto *tp, void *fh,
638 struct sk_buff *skb, struct tcmsg *t, bool rtnl_held)
639{
640 struct tcindex_data *p = rtnl_dereference(tp->root);
641 struct tcindex_filter_result *r = fh;
642 struct nlattr *nest;
643
644 pr_debug("tcindex_dump(tp %p,fh %p,skb %p,t %p),p %p,r %p\n",
645 tp, fh, skb, t, p, r);
646 pr_debug("p->perfect %p p->h %p\n", p->perfect, p->h);
647
648 nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
649 if (nest == NULL)
650 goto nla_put_failure;
651
652 if (!fh) {
653 t->tcm_handle = ~0;
654 if (nla_put_u32(skb, TCA_TCINDEX_HASH, p->hash) ||
655 nla_put_u16(skb, TCA_TCINDEX_MASK, p->mask) ||
656 nla_put_u32(skb, TCA_TCINDEX_SHIFT, p->shift) ||
657 nla_put_u32(skb, TCA_TCINDEX_FALL_THROUGH, p->fall_through))
658 goto nla_put_failure;
659 nla_nest_end(skb, nest);
660 } else {
661 if (p->perfect) {
662 t->tcm_handle = r - p->perfect;
663 } else {
664 struct tcindex_filter *f;
665 struct tcindex_filter __rcu **fp;
666 int i;
667
668 t->tcm_handle = 0;
669 for (i = 0; !t->tcm_handle && i < p->hash; i++) {
670 fp = &p->h[i];
671 for (f = rtnl_dereference(*fp);
672 !t->tcm_handle && f;
673 fp = &f->next, f = rtnl_dereference(*fp)) {
674 if (&f->result == r)
675 t->tcm_handle = f->key;
676 }
677 }
678 }
679 pr_debug("handle = %d\n", t->tcm_handle);
680 if (r->res.class &&
681 nla_put_u32(skb, TCA_TCINDEX_CLASSID, r->res.classid))
682 goto nla_put_failure;
683
684 if (tcf_exts_dump(skb, &r->exts) < 0)
685 goto nla_put_failure;
686 nla_nest_end(skb, nest);
687
688 if (tcf_exts_dump_stats(skb, &r->exts) < 0)
689 goto nla_put_failure;
690 }
691
692 return skb->len;
693
694nla_put_failure:
695 nla_nest_cancel(skb, nest);
696 return -1;
697}
698
699static void tcindex_bind_class(void *fh, u32 classid, unsigned long cl,
700 void *q, unsigned long base)
701{
702 struct tcindex_filter_result *r = fh;
703
704 if (r && r->res.classid == classid) {
705 if (cl)
706 __tcf_bind_filter(q, &r->res, base);
707 else
708 __tcf_unbind_filter(q, &r->res);
709 }
710}
711
712static struct tcf_proto_ops cls_tcindex_ops __read_mostly = {
713 .kind = "tcindex",
714 .classify = tcindex_classify,
715 .init = tcindex_init,
716 .destroy = tcindex_destroy,
717 .get = tcindex_get,
718 .change = tcindex_change,
719 .delete = tcindex_delete,
720 .walk = tcindex_walk,
721 .dump = tcindex_dump,
722 .bind_class = tcindex_bind_class,
723 .owner = THIS_MODULE,
724};
725
726static int __init init_tcindex(void)
727{
728 return register_tcf_proto_ops(&cls_tcindex_ops);
729}
730
731static void __exit exit_tcindex(void)
732{
733 unregister_tcf_proto_ops(&cls_tcindex_ops);
734}
735
736module_init(init_tcindex)
737module_exit(exit_tcindex)
738MODULE_LICENSE("GPL");
739
