?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.19.168
[linux/fpc-iii.git] / net / sched / cls_tcindex.c
blob0d7a0aac8dbb123694d7ef2249c3fa60e182f1b2
1 /*
2 * net/sched/cls_tcindex.c Packet classifier for skb->tc_index
3 *
4 * Written 1998,1999 by Werner Almesberger, EPFL ICA
5 */
6
7 #include <linux/module.h>
8 #include <linux/types.h>
9 #include <linux/kernel.h>
10 #include <linux/skbuff.h>
11 #include <linux/errno.h>
12 #include <linux/slab.h>
13 #include <net/act_api.h>
14 #include <net/netlink.h>
15 #include <net/pkt_cls.h>
16 #include <net/sch_generic.h>
17
18 /*
19 * Passing parameters to the root seems to be done more awkwardly than really
20 * necessary. At least, u32 doesn't seem to use such dirty hacks. To be
21 * verified. FIXME.
22 */
23
24 #define PERFECT_HASH_THRESHOLD 64 /* use perfect hash if not bigger */
25 #define DEFAULT_HASH_SIZE 64 /* optimized for diffserv */
26
27
28 struct tcindex_filter_result {
29 struct tcf_exts exts;
30 struct tcf_result res;
31 struct rcu_work rwork;
32 };
33
34 struct tcindex_filter {
35 u16 key;
36 struct tcindex_filter_result result;
37 struct tcindex_filter __rcu *next;
38 struct rcu_work rwork;
39 };
40
41
42 struct tcindex_data {
43 struct tcindex_filter_result *perfect; /* perfect hash; NULL if none */
44 struct tcindex_filter __rcu **h; /* imperfect hash; */
45 struct tcf_proto *tp;
46 u16 mask; /* AND key with mask */
47 u32 shift; /* shift ANDed key to the right */
48 u32 hash; /* hash table size; 0 if undefined */
49 u32 alloc_hash; /* allocated size */
50 u32 fall_through; /* 0: only classify if explicit match */
51 struct rcu_work rwork;
52 };
53
54 static inline int tcindex_filter_is_set(struct tcindex_filter_result *r)
55 {
56 return tcf_exts_has_actions(&r->exts) || r->res.classid;
57 }
58
59 static struct tcindex_filter_result *tcindex_lookup(struct tcindex_data *p,
60 u16 key)
61 {
62 if (p->perfect) {
63 struct tcindex_filter_result *f = p->perfect + key;
64
65 return tcindex_filter_is_set(f) ? f : NULL;
66 } else if (p->h) {
67 struct tcindex_filter __rcu **fp;
68 struct tcindex_filter *f;
69
70 fp = &p->h[key % p->hash];
71 for (f = rcu_dereference_bh_rtnl(*fp);
72 f;
73 fp = &f->next, f = rcu_dereference_bh_rtnl(*fp))
74 if (f->key == key)
75 return &f->result;
76 }
77
78 return NULL;
79 }
80
81
82 static int tcindex_classify(struct sk_buff *skb, const struct tcf_proto *tp,
83 struct tcf_result *res)
84 {
85 struct tcindex_data *p = rcu_dereference_bh(tp->root);
86 struct tcindex_filter_result *f;
87 int key = (skb->tc_index & p->mask) >> p->shift;
88
89 pr_debug("tcindex_classify(skb %p,tp %p,res %p),p %p\n",
90 skb, tp, res, p);
91
92 f = tcindex_lookup(p, key);
93 if (!f) {
94 struct Qdisc *q = tcf_block_q(tp->chain->block);
95
96 if (!p->fall_through)
97 return -1;
98 res->classid = TC_H_MAKE(TC_H_MAJ(q->handle), key);
99 res->class = 0;
100 pr_debug("alg 0x%x\n", res->classid);
101 return 0;
102 }
103 *res = f->res;
104 pr_debug("map 0x%x\n", res->classid);
105
106 return tcf_exts_exec(skb, &f->exts, res);
107 }
108
109
110 static void *tcindex_get(struct tcf_proto *tp, u32 handle)
111 {
112 struct tcindex_data *p = rtnl_dereference(tp->root);
113 struct tcindex_filter_result *r;
114
115 pr_debug("tcindex_get(tp %p,handle 0x%08x)\n", tp, handle);
116 if (p->perfect && handle >= p->alloc_hash)
117 return NULL;
118 r = tcindex_lookup(p, handle);
119 return r && tcindex_filter_is_set(r) ? r : NULL;
120 }
121
122 static int tcindex_init(struct tcf_proto *tp)
123 {
124 struct tcindex_data *p;
125
126 pr_debug("tcindex_init(tp %p)\n", tp);
127 p = kzalloc(sizeof(struct tcindex_data), GFP_KERNEL);
128 if (!p)
129 return -ENOMEM;
130
131 p->mask = 0xffff;
132 p->hash = DEFAULT_HASH_SIZE;
133 p->fall_through = 1;
134
135 rcu_assign_pointer(tp->root, p);
136 return 0;
137 }
138
139 static void __tcindex_destroy_rexts(struct tcindex_filter_result *r)
140 {
141 tcf_exts_destroy(&r->exts);
142 tcf_exts_put_net(&r->exts);
143 }
144
145 static void tcindex_destroy_rexts_work(struct work_struct *work)
146 {
147 struct tcindex_filter_result *r;
148
149 r = container_of(to_rcu_work(work),
150 struct tcindex_filter_result,
151 rwork);
152 rtnl_lock();
153 __tcindex_destroy_rexts(r);
154 rtnl_unlock();
155 }
156
157 static void __tcindex_destroy_fexts(struct tcindex_filter *f)
158 {
159 tcf_exts_destroy(&f->result.exts);
160 tcf_exts_put_net(&f->result.exts);
161 kfree(f);
162 }
163
164 static void tcindex_destroy_fexts_work(struct work_struct *work)
165 {
166 struct tcindex_filter *f = container_of(to_rcu_work(work),
167 struct tcindex_filter,
168 rwork);
169
170 rtnl_lock();
171 __tcindex_destroy_fexts(f);
172 rtnl_unlock();
173 }
174
175 static int tcindex_delete(struct tcf_proto *tp, void *arg, bool *last,
176 struct netlink_ext_ack *extack)
177 {
178 struct tcindex_data *p = rtnl_dereference(tp->root);
179 struct tcindex_filter_result *r = arg;
180 struct tcindex_filter __rcu **walk;
181 struct tcindex_filter *f = NULL;
182
183 pr_debug("tcindex_delete(tp %p,arg %p),p %p\n", tp, arg, p);
184 if (p->perfect) {
185 if (!r->res.class)
186 return -ENOENT;
187 } else {
188 int i;
189
190 for (i = 0; i < p->hash; i++) {
191 walk = p->h + i;
192 for (f = rtnl_dereference(*walk); f;
193 walk = &f->next, f = rtnl_dereference(*walk)) {
194 if (&f->result == r)
195 goto found;
196 }
197 }
198 return -ENOENT;
199
200 found:
201 rcu_assign_pointer(*walk, rtnl_dereference(f->next));
202 }
203 tcf_unbind_filter(tp, &r->res);
204 /* all classifiers are required to call tcf_exts_destroy() after rcu
205 * grace period, since converted-to-rcu actions are relying on that
206 * in cleanup() callback
207 */
208 if (f) {
209 if (tcf_exts_get_net(&f->result.exts))
210 tcf_queue_work(&f->rwork, tcindex_destroy_fexts_work);
211 else
212 __tcindex_destroy_fexts(f);
213 } else {
214 if (tcf_exts_get_net(&r->exts))
215 tcf_queue_work(&r->rwork, tcindex_destroy_rexts_work);
216 else
217 __tcindex_destroy_rexts(r);
218 }
219
220 *last = false;
221 return 0;
222 }
223
224 static void tcindex_destroy_work(struct work_struct *work)
225 {
226 struct tcindex_data *p = container_of(to_rcu_work(work),
227 struct tcindex_data,
228 rwork);
229
230 kfree(p->perfect);
231 kfree(p->h);
232 kfree(p);
233 }
234
235 static inline int
236 valid_perfect_hash(struct tcindex_data *p)
237 {
238 return p->hash > (p->mask >> p->shift);
239 }
240
241 static const struct nla_policy tcindex_policy[TCA_TCINDEX_MAX + 1] = {
242 [TCA_TCINDEX_HASH] = { .type = NLA_U32 },
243 [TCA_TCINDEX_MASK] = { .type = NLA_U16 },
244 [TCA_TCINDEX_SHIFT] = { .type = NLA_U32 },
245 [TCA_TCINDEX_FALL_THROUGH] = { .type = NLA_U32 },
246 [TCA_TCINDEX_CLASSID] = { .type = NLA_U32 },
247 };
248
249 static int tcindex_filter_result_init(struct tcindex_filter_result *r)
250 {
251 memset(r, 0, sizeof(*r));
252 return tcf_exts_init(&r->exts, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
253 }
254
255 static void tcindex_partial_destroy_work(struct work_struct *work)
256 {
257 struct tcindex_data *p = container_of(to_rcu_work(work),
258 struct tcindex_data,
259 rwork);
260
261 kfree(p->perfect);
262 kfree(p);
263 }
264
265 static void tcindex_free_perfect_hash(struct tcindex_data *cp)
266 {
267 int i;
268
269 for (i = 0; i < cp->hash; i++)
270 tcf_exts_destroy(&cp->perfect[i].exts);
271 kfree(cp->perfect);
272 }
273
274 static int tcindex_alloc_perfect_hash(struct net *net, struct tcindex_data *cp)
275 {
276 int i, err = 0;
277
278 cp->perfect = kcalloc(cp->hash, sizeof(struct tcindex_filter_result),
279 GFP_KERNEL);
280 if (!cp->perfect)
281 return -ENOMEM;
282
283 for (i = 0; i < cp->hash; i++) {
284 err = tcf_exts_init(&cp->perfect[i].exts,
285 TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
286 if (err < 0)
287 goto errout;
288 #ifdef CONFIG_NET_CLS_ACT
289 cp->perfect[i].exts.net = net;
290 #endif
291 }
292
293 return 0;
294
295 errout:
296 tcindex_free_perfect_hash(cp);
297 return err;
298 }
299
300 static int
301 tcindex_set_parms(struct net *net, struct tcf_proto *tp, unsigned long base,
302 u32 handle, struct tcindex_data *p,
303 struct tcindex_filter_result *r, struct nlattr **tb,
304 struct nlattr *est, bool ovr, struct netlink_ext_ack *extack)
305 {
306 struct tcindex_filter_result new_filter_result, *old_r = r;
307 struct tcindex_data *cp = NULL, *oldp;
308 struct tcindex_filter *f = NULL; /* make gcc behave */
309 struct tcf_result cr = {};
310 int err, balloc = 0;
311 struct tcf_exts e;
312
313 err = tcf_exts_init(&e, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
314 if (err < 0)
315 return err;
316 err = tcf_exts_validate(net, tp, tb, est, &e, ovr, extack);
317 if (err < 0)
318 goto errout;
319
320 err = -ENOMEM;
321 /* tcindex_data attributes must look atomic to classifier/lookup so
322 * allocate new tcindex data and RCU assign it onto root. Keeping
323 * perfect hash and hash pointers from old data.
324 */
325 cp = kzalloc(sizeof(*cp), GFP_KERNEL);
326 if (!cp)
327 goto errout;
328
329 cp->mask = p->mask;
330 cp->shift = p->shift;
331 cp->hash = p->hash;
332 cp->alloc_hash = p->alloc_hash;
333 cp->fall_through = p->fall_through;
334 cp->tp = tp;
335
336 if (tb[TCA_TCINDEX_HASH])
337 cp->hash = nla_get_u32(tb[TCA_TCINDEX_HASH]);
338
339 if (tb[TCA_TCINDEX_MASK])
340 cp->mask = nla_get_u16(tb[TCA_TCINDEX_MASK]);
341
342 if (tb[TCA_TCINDEX_SHIFT])
343 cp->shift = nla_get_u32(tb[TCA_TCINDEX_SHIFT]);
344
345 if (!cp->hash) {
346 /* Hash not specified, use perfect hash if the upper limit
347 * of the hashing index is below the threshold.
348 */
349 if ((cp->mask >> cp->shift) < PERFECT_HASH_THRESHOLD)
350 cp->hash = (cp->mask >> cp->shift) + 1;
351 else
352 cp->hash = DEFAULT_HASH_SIZE;
353 }
354
355 if (p->perfect) {
356 int i;
357
358 if (tcindex_alloc_perfect_hash(net, cp) < 0)
359 goto errout;
360 cp->alloc_hash = cp->hash;
361 for (i = 0; i < min(cp->hash, p->hash); i++)
362 cp->perfect[i].res = p->perfect[i].res;
363 balloc = 1;
364 }
365 cp->h = p->h;
366
367 err = tcindex_filter_result_init(&new_filter_result);
368 if (err < 0)
369 goto errout_alloc;
370 if (old_r)
371 cr = r->res;
372
373 err = -EBUSY;
374
375 /* Hash already allocated, make sure that we still meet the
376 * requirements for the allocated hash.
377 */
378 if (cp->perfect) {
379 if (!valid_perfect_hash(cp) ||
380 cp->hash > cp->alloc_hash)
381 goto errout_alloc;
382 } else if (cp->h && cp->hash != cp->alloc_hash) {
383 goto errout_alloc;
384 }
385
386 err = -EINVAL;
387 if (tb[TCA_TCINDEX_FALL_THROUGH])
388 cp->fall_through = nla_get_u32(tb[TCA_TCINDEX_FALL_THROUGH]);
389
390 if (!cp->perfect && !cp->h)
391 cp->alloc_hash = cp->hash;
392
393 /* Note: this could be as restrictive as if (handle & ~(mask >> shift))
394 * but then, we'd fail handles that may become valid after some future
395 * mask change. While this is extremely unlikely to ever matter,
396 * the check below is safer (and also more backwards-compatible).
397 */
398 if (cp->perfect || valid_perfect_hash(cp))
399 if (handle >= cp->alloc_hash)
400 goto errout_alloc;
401
402
403 err = -ENOMEM;
404 if (!cp->perfect && !cp->h) {
405 if (valid_perfect_hash(cp)) {
406 if (tcindex_alloc_perfect_hash(net, cp) < 0)
407 goto errout_alloc;
408 balloc = 1;
409 } else {
410 struct tcindex_filter __rcu **hash;
411
412 hash = kcalloc(cp->hash,
413 sizeof(struct tcindex_filter *),
414 GFP_KERNEL);
415
416 if (!hash)
417 goto errout_alloc;
418
419 cp->h = hash;
420 balloc = 2;
421 }
422 }
423
424 if (cp->perfect)
425 r = cp->perfect + handle;
426 else
427 r = tcindex_lookup(cp, handle) ? : &new_filter_result;
428
429 if (r == &new_filter_result) {
430 f = kzalloc(sizeof(*f), GFP_KERNEL);
431 if (!f)
432 goto errout_alloc;
433 f->key = handle;
434 f->next = NULL;
435 err = tcindex_filter_result_init(&f->result);
436 if (err < 0) {
437 kfree(f);
438 goto errout_alloc;
439 }
440 }
441
442 if (tb[TCA_TCINDEX_CLASSID]) {
443 cr.classid = nla_get_u32(tb[TCA_TCINDEX_CLASSID]);
444 tcf_bind_filter(tp, &cr, base);
445 }
446
447 if (old_r && old_r != r) {
448 err = tcindex_filter_result_init(old_r);
449 if (err < 0) {
450 kfree(f);
451 goto errout_alloc;
452 }
453 }
454
455 oldp = p;
456 r->res = cr;
457 tcf_exts_change(&r->exts, &e);
458
459 rcu_assign_pointer(tp->root, cp);
460
461 if (r == &new_filter_result) {
462 struct tcindex_filter *nfp;
463 struct tcindex_filter __rcu **fp;
464
465 f->result.res = r->res;
466 tcf_exts_change(&f->result.exts, &r->exts);
467
468 fp = cp->h + (handle % cp->hash);
469 for (nfp = rtnl_dereference(*fp);
470 nfp;
471 fp = &nfp->next, nfp = rtnl_dereference(*fp))
472 ; /* nothing */
473
474 rcu_assign_pointer(*fp, f);
475 } else {
476 tcf_exts_destroy(&new_filter_result.exts);
477 }
478
479 if (oldp)
480 tcf_queue_work(&oldp->rwork, tcindex_partial_destroy_work);
481 return 0;
482
483 errout_alloc:
484 if (balloc == 1)
485 tcindex_free_perfect_hash(cp);
486 else if (balloc == 2)
487 kfree(cp->h);
488 tcf_exts_destroy(&new_filter_result.exts);
489 errout:
490 kfree(cp);
491 tcf_exts_destroy(&e);
492 return err;
493 }
494
495 static int
496 tcindex_change(struct net *net, struct sk_buff *in_skb,
497 struct tcf_proto *tp, unsigned long base, u32 handle,
498 struct nlattr **tca, void **arg, bool ovr,
499 struct netlink_ext_ack *extack)
500 {
501 struct nlattr *opt = tca[TCA_OPTIONS];
502 struct nlattr *tb[TCA_TCINDEX_MAX + 1];
503 struct tcindex_data *p = rtnl_dereference(tp->root);
504 struct tcindex_filter_result *r = *arg;
505 int err;
506
507 pr_debug("tcindex_change(tp %p,handle 0x%08x,tca %p,arg %p),opt %p,"
508 "p %p,r %p,*arg %p\n",
509 tp, handle, tca, arg, opt, p, r, arg ? *arg : NULL);
510
511 if (!opt)
512 return 0;
513
514 err = nla_parse_nested(tb, TCA_TCINDEX_MAX, opt, tcindex_policy, NULL);
515 if (err < 0)
516 return err;
517
518 return tcindex_set_parms(net, tp, base, handle, p, r, tb,
519 tca[TCA_RATE], ovr, extack);
520 }
521
522 static void tcindex_walk(struct tcf_proto *tp, struct tcf_walker *walker)
523 {
524 struct tcindex_data *p = rtnl_dereference(tp->root);
525 struct tcindex_filter *f, *next;
526 int i;
527
528 pr_debug("tcindex_walk(tp %p,walker %p),p %p\n", tp, walker, p);
529 if (p->perfect) {
530 for (i = 0; i < p->hash; i++) {
531 if (!p->perfect[i].res.class)
532 continue;
533 if (walker->count >= walker->skip) {
534 if (walker->fn(tp, p->perfect + i, walker) < 0) {
535 walker->stop = 1;
536 return;
537 }
538 }
539 walker->count++;
540 }
541 }
542 if (!p->h)
543 return;
544 for (i = 0; i < p->hash; i++) {
545 for (f = rtnl_dereference(p->h[i]); f; f = next) {
546 next = rtnl_dereference(f->next);
547 if (walker->count >= walker->skip) {
548 if (walker->fn(tp, &f->result, walker) < 0) {
549 walker->stop = 1;
550 return;
551 }
552 }
553 walker->count++;
554 }
555 }
556 }
557
558 static void tcindex_destroy(struct tcf_proto *tp,
559 struct netlink_ext_ack *extack)
560 {
561 struct tcindex_data *p = rtnl_dereference(tp->root);
562 int i;
563
564 pr_debug("tcindex_destroy(tp %p),p %p\n", tp, p);
565
566 if (p->perfect) {
567 for (i = 0; i < p->hash; i++) {
568 struct tcindex_filter_result *r = p->perfect + i;
569
570 tcf_unbind_filter(tp, &r->res);
571 if (tcf_exts_get_net(&r->exts))
572 tcf_queue_work(&r->rwork,
573 tcindex_destroy_rexts_work);
574 else
575 __tcindex_destroy_rexts(r);
576 }
577 }
578
579 for (i = 0; p->h && i < p->hash; i++) {
580 struct tcindex_filter *f, *next;
581 bool last;
582
583 for (f = rtnl_dereference(p->h[i]); f; f = next) {
584 next = rtnl_dereference(f->next);
585 tcindex_delete(tp, &f->result, &last, NULL);
586 }
587 }
588
589 tcf_queue_work(&p->rwork, tcindex_destroy_work);
590 }
591
592
593 static int tcindex_dump(struct net *net, struct tcf_proto *tp, void *fh,
594 struct sk_buff *skb, struct tcmsg *t)
595 {
596 struct tcindex_data *p = rtnl_dereference(tp->root);
597 struct tcindex_filter_result *r = fh;
598 struct nlattr *nest;
599
600 pr_debug("tcindex_dump(tp %p,fh %p,skb %p,t %p),p %p,r %p\n",
601 tp, fh, skb, t, p, r);
602 pr_debug("p->perfect %p p->h %p\n", p->perfect, p->h);
603
604 nest = nla_nest_start(skb, TCA_OPTIONS);
605 if (nest == NULL)
606 goto nla_put_failure;
607
608 if (!fh) {
609 t->tcm_handle = ~0; /* whatever ... */
610 if (nla_put_u32(skb, TCA_TCINDEX_HASH, p->hash) ||
611 nla_put_u16(skb, TCA_TCINDEX_MASK, p->mask) ||
612 nla_put_u32(skb, TCA_TCINDEX_SHIFT, p->shift) ||
613 nla_put_u32(skb, TCA_TCINDEX_FALL_THROUGH, p->fall_through))
614 goto nla_put_failure;
615 nla_nest_end(skb, nest);
616 } else {
617 if (p->perfect) {
618 t->tcm_handle = r - p->perfect;
619 } else {
620 struct tcindex_filter *f;
621 struct tcindex_filter __rcu **fp;
622 int i;
623
624 t->tcm_handle = 0;
625 for (i = 0; !t->tcm_handle && i < p->hash; i++) {
626 fp = &p->h[i];
627 for (f = rtnl_dereference(*fp);
628 !t->tcm_handle && f;
629 fp = &f->next, f = rtnl_dereference(*fp)) {
630 if (&f->result == r)
631 t->tcm_handle = f->key;
632 }
633 }
634 }
635 pr_debug("handle = %d\n", t->tcm_handle);
636 if (r->res.class &&
637 nla_put_u32(skb, TCA_TCINDEX_CLASSID, r->res.classid))
638 goto nla_put_failure;
639
640 if (tcf_exts_dump(skb, &r->exts) < 0)
641 goto nla_put_failure;
642 nla_nest_end(skb, nest);
643
644 if (tcf_exts_dump_stats(skb, &r->exts) < 0)
645 goto nla_put_failure;
646 }
647
648 return skb->len;
649
650 nla_put_failure:
651 nla_nest_cancel(skb, nest);
652 return -1;
653 }
654
655 static void tcindex_bind_class(void *fh, u32 classid, unsigned long cl,
656 void *q, unsigned long base)
657 {
658 struct tcindex_filter_result *r = fh;
659
660 if (r && r->res.classid == classid) {
661 if (cl)
662 __tcf_bind_filter(q, &r->res, base);
663 else
664 __tcf_unbind_filter(q, &r->res);
665 }
666 }
667
668 static struct tcf_proto_ops cls_tcindex_ops __read_mostly = {
669 .kind = "tcindex",
670 .classify = tcindex_classify,
671 .init = tcindex_init,
672 .destroy = tcindex_destroy,
673 .get = tcindex_get,
674 .change = tcindex_change,
675 .delete = tcindex_delete,
676 .walk = tcindex_walk,
677 .dump = tcindex_dump,
678 .bind_class = tcindex_bind_class,
679 .owner = THIS_MODULE,
680 };
681
682 static int __init init_tcindex(void)
683 {
684 return register_tcf_proto_ops(&cls_tcindex_ops);
685 }
686
687 static void __exit exit_tcindex(void)
688 {
689 unregister_tcf_proto_ops(&cls_tcindex_ops);
690 }
691
692 module_init(init_tcindex)
693 module_exit(exit_tcindex)
694 MODULE_LICENSE("GPL");
Linux with added FPC-III support