?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Revert "ip6_vti: adjust vti mtu according to mtu of lower device"
[linux/fpc-iii.git] / net / sched / cls_tcindex.c
blob0751245a6aced60163601b94ab1386e0c5e30565
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
17 /*
18 * Passing parameters to the root seems to be done more awkwardly than really
19 * necessary. At least, u32 doesn't seem to use such dirty hacks. To be
20 * verified. FIXME.
21 */
22
23 #define PERFECT_HASH_THRESHOLD 64 /* use perfect hash if not bigger */
24 #define DEFAULT_HASH_SIZE 64 /* optimized for diffserv */
25
26
27 struct tcindex_filter_result {
28 struct tcf_exts exts;
29 struct tcf_result res;
30 struct rcu_head rcu;
31 };
32
33 struct tcindex_filter {
34 u16 key;
35 struct tcindex_filter_result result;
36 struct tcindex_filter __rcu *next;
37 struct rcu_head rcu;
38 };
39
40
41 struct tcindex_data {
42 struct tcindex_filter_result *perfect; /* perfect hash; NULL if none */
43 struct tcindex_filter __rcu **h; /* imperfect hash; */
44 struct tcf_proto *tp;
45 u16 mask; /* AND key with mask */
46 u32 shift; /* shift ANDed key to the right */
47 u32 hash; /* hash table size; 0 if undefined */
48 u32 alloc_hash; /* allocated size */
49 u32 fall_through; /* 0: only classify if explicit match */
50 struct rcu_head rcu;
51 };
52
53 static inline int tcindex_filter_is_set(struct tcindex_filter_result *r)
54 {
55 return tcf_exts_is_predicative(&r->exts) || r->res.classid;
56 }
57
58 static struct tcindex_filter_result *tcindex_lookup(struct tcindex_data *p,
59 u16 key)
60 {
61 if (p->perfect) {
62 struct tcindex_filter_result *f = p->perfect + key;
63
64 return tcindex_filter_is_set(f) ? f : NULL;
65 } else if (p->h) {
66 struct tcindex_filter __rcu **fp;
67 struct tcindex_filter *f;
68
69 fp = &p->h[key % p->hash];
70 for (f = rcu_dereference_bh_rtnl(*fp);
71 f;
72 fp = &f->next, f = rcu_dereference_bh_rtnl(*fp))
73 if (f->key == key)
74 return &f->result;
75 }
76
77 return NULL;
78 }
79
80
81 static int tcindex_classify(struct sk_buff *skb, const struct tcf_proto *tp,
82 struct tcf_result *res)
83 {
84 struct tcindex_data *p = rcu_dereference_bh(tp->root);
85 struct tcindex_filter_result *f;
86 int key = (skb->tc_index & p->mask) >> p->shift;
87
88 pr_debug("tcindex_classify(skb %p,tp %p,res %p),p %p\n",
89 skb, tp, res, p);
90
91 f = tcindex_lookup(p, key);
92 if (!f) {
93 if (!p->fall_through)
94 return -1;
95 res->classid = TC_H_MAKE(TC_H_MAJ(tp->q->handle), key);
96 res->class = 0;
97 pr_debug("alg 0x%x\n", res->classid);
98 return 0;
99 }
100 *res = f->res;
101 pr_debug("map 0x%x\n", res->classid);
102
103 return tcf_exts_exec(skb, &f->exts, res);
104 }
105
106
107 static unsigned long tcindex_get(struct tcf_proto *tp, u32 handle)
108 {
109 struct tcindex_data *p = rtnl_dereference(tp->root);
110 struct tcindex_filter_result *r;
111
112 pr_debug("tcindex_get(tp %p,handle 0x%08x)\n", tp, handle);
113 if (p->perfect && handle >= p->alloc_hash)
114 return 0;
115 r = tcindex_lookup(p, handle);
116 return r && tcindex_filter_is_set(r) ? (unsigned long) r : 0UL;
117 }
118
119 static int tcindex_init(struct tcf_proto *tp)
120 {
121 struct tcindex_data *p;
122
123 pr_debug("tcindex_init(tp %p)\n", tp);
124 p = kzalloc(sizeof(struct tcindex_data), GFP_KERNEL);
125 if (!p)
126 return -ENOMEM;
127
128 p->mask = 0xffff;
129 p->hash = DEFAULT_HASH_SIZE;
130 p->fall_through = 1;
131
132 rcu_assign_pointer(tp->root, p);
133 return 0;
134 }
135
136 static void tcindex_destroy_rexts(struct rcu_head *head)
137 {
138 struct tcindex_filter_result *r;
139
140 r = container_of(head, struct tcindex_filter_result, rcu);
141 tcf_exts_destroy(&r->exts);
142 }
143
144 static void tcindex_destroy_fexts(struct rcu_head *head)
145 {
146 struct tcindex_filter *f = container_of(head, struct tcindex_filter,
147 rcu);
148
149 tcf_exts_destroy(&f->result.exts);
150 kfree(f);
151 }
152
153 static int tcindex_delete(struct tcf_proto *tp, unsigned long arg)
154 {
155 struct tcindex_data *p = rtnl_dereference(tp->root);
156 struct tcindex_filter_result *r = (struct tcindex_filter_result *) arg;
157 struct tcindex_filter __rcu **walk;
158 struct tcindex_filter *f = NULL;
159
160 pr_debug("tcindex_delete(tp %p,arg 0x%lx),p %p\n", tp, arg, p);
161 if (p->perfect) {
162 if (!r->res.class)
163 return -ENOENT;
164 } else {
165 int i;
166
167 for (i = 0; i < p->hash; i++) {
168 walk = p->h + i;
169 for (f = rtnl_dereference(*walk); f;
170 walk = &f->next, f = rtnl_dereference(*walk)) {
171 if (&f->result == r)
172 goto found;
173 }
174 }
175 return -ENOENT;
176
177 found:
178 rcu_assign_pointer(*walk, rtnl_dereference(f->next));
179 }
180 tcf_unbind_filter(tp, &r->res);
181 /* all classifiers are required to call tcf_exts_destroy() after rcu
182 * grace period, since converted-to-rcu actions are relying on that
183 * in cleanup() callback
184 */
185 if (f)
186 call_rcu(&f->rcu, tcindex_destroy_fexts);
187 else
188 call_rcu(&r->rcu, tcindex_destroy_rexts);
189 return 0;
190 }
191
192 static int tcindex_destroy_element(struct tcf_proto *tp,
193 unsigned long arg,
194 struct tcf_walker *walker)
195 {
196 return tcindex_delete(tp, arg);
197 }
198
199 static void __tcindex_destroy(struct rcu_head *head)
200 {
201 struct tcindex_data *p = container_of(head, struct tcindex_data, rcu);
202
203 kfree(p->perfect);
204 kfree(p->h);
205 kfree(p);
206 }
207
208 static inline int
209 valid_perfect_hash(struct tcindex_data *p)
210 {
211 return p->hash > (p->mask >> p->shift);
212 }
213
214 static const struct nla_policy tcindex_policy[TCA_TCINDEX_MAX + 1] = {
215 [TCA_TCINDEX_HASH] = { .type = NLA_U32 },
216 [TCA_TCINDEX_MASK] = { .type = NLA_U16 },
217 [TCA_TCINDEX_SHIFT] = { .type = NLA_U32 },
218 [TCA_TCINDEX_FALL_THROUGH] = { .type = NLA_U32 },
219 [TCA_TCINDEX_CLASSID] = { .type = NLA_U32 },
220 };
221
222 static int tcindex_filter_result_init(struct tcindex_filter_result *r)
223 {
224 memset(r, 0, sizeof(*r));
225 return tcf_exts_init(&r->exts, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
226 }
227
228 static void __tcindex_partial_destroy(struct rcu_head *head)
229 {
230 struct tcindex_data *p = container_of(head, struct tcindex_data, rcu);
231
232 kfree(p->perfect);
233 kfree(p);
234 }
235
236 static void tcindex_free_perfect_hash(struct tcindex_data *cp)
237 {
238 int i;
239
240 for (i = 0; i < cp->hash; i++)
241 tcf_exts_destroy(&cp->perfect[i].exts);
242 kfree(cp->perfect);
243 }
244
245 static int tcindex_alloc_perfect_hash(struct tcindex_data *cp)
246 {
247 int i, err = 0;
248
249 cp->perfect = kcalloc(cp->hash, sizeof(struct tcindex_filter_result),
250 GFP_KERNEL);
251 if (!cp->perfect)
252 return -ENOMEM;
253
254 for (i = 0; i < cp->hash; i++) {
255 err = tcf_exts_init(&cp->perfect[i].exts,
256 TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
257 if (err < 0)
258 goto errout;
259 }
260
261 return 0;
262
263 errout:
264 tcindex_free_perfect_hash(cp);
265 return err;
266 }
267
268 static int
269 tcindex_set_parms(struct net *net, struct tcf_proto *tp, unsigned long base,
270 u32 handle, struct tcindex_data *p,
271 struct tcindex_filter_result *r, struct nlattr **tb,
272 struct nlattr *est, bool ovr)
273 {
274 struct tcindex_filter_result new_filter_result, *old_r = r;
275 struct tcindex_filter_result cr;
276 struct tcindex_data *cp = NULL, *oldp;
277 struct tcindex_filter *f = NULL; /* make gcc behave */
278 int err, balloc = 0;
279 struct tcf_exts e;
280
281 err = tcf_exts_init(&e, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
282 if (err < 0)
283 return err;
284 err = tcf_exts_validate(net, tp, tb, est, &e, ovr);
285 if (err < 0)
286 goto errout;
287
288 err = -ENOMEM;
289 /* tcindex_data attributes must look atomic to classifier/lookup so
290 * allocate new tcindex data and RCU assign it onto root. Keeping
291 * perfect hash and hash pointers from old data.
292 */
293 cp = kzalloc(sizeof(*cp), GFP_KERNEL);
294 if (!cp)
295 goto errout;
296
297 cp->mask = p->mask;
298 cp->shift = p->shift;
299 cp->hash = p->hash;
300 cp->alloc_hash = p->alloc_hash;
301 cp->fall_through = p->fall_through;
302 cp->tp = tp;
303
304 if (p->perfect) {
305 int i;
306
307 if (tcindex_alloc_perfect_hash(cp) < 0)
308 goto errout;
309 for (i = 0; i < cp->hash; i++)
310 cp->perfect[i].res = p->perfect[i].res;
311 balloc = 1;
312 }
313 cp->h = p->h;
314
315 err = tcindex_filter_result_init(&new_filter_result);
316 if (err < 0)
317 goto errout1;
318 err = tcindex_filter_result_init(&cr);
319 if (err < 0)
320 goto errout1;
321 if (old_r)
322 cr.res = r->res;
323
324 if (tb[TCA_TCINDEX_HASH])
325 cp->hash = nla_get_u32(tb[TCA_TCINDEX_HASH]);
326
327 if (tb[TCA_TCINDEX_MASK])
328 cp->mask = nla_get_u16(tb[TCA_TCINDEX_MASK]);
329
330 if (tb[TCA_TCINDEX_SHIFT])
331 cp->shift = nla_get_u32(tb[TCA_TCINDEX_SHIFT]);
332
333 err = -EBUSY;
334
335 /* Hash already allocated, make sure that we still meet the
336 * requirements for the allocated hash.
337 */
338 if (cp->perfect) {
339 if (!valid_perfect_hash(cp) ||
340 cp->hash > cp->alloc_hash)
341 goto errout_alloc;
342 } else if (cp->h && cp->hash != cp->alloc_hash) {
343 goto errout_alloc;
344 }
345
346 err = -EINVAL;
347 if (tb[TCA_TCINDEX_FALL_THROUGH])
348 cp->fall_through = nla_get_u32(tb[TCA_TCINDEX_FALL_THROUGH]);
349
350 if (!cp->hash) {
351 /* Hash not specified, use perfect hash if the upper limit
352 * of the hashing index is below the threshold.
353 */
354 if ((cp->mask >> cp->shift) < PERFECT_HASH_THRESHOLD)
355 cp->hash = (cp->mask >> cp->shift) + 1;
356 else
357 cp->hash = DEFAULT_HASH_SIZE;
358 }
359
360 if (!cp->perfect && !cp->h)
361 cp->alloc_hash = cp->hash;
362
363 /* Note: this could be as restrictive as if (handle & ~(mask >> shift))
364 * but then, we'd fail handles that may become valid after some future
365 * mask change. While this is extremely unlikely to ever matter,
366 * the check below is safer (and also more backwards-compatible).
367 */
368 if (cp->perfect || valid_perfect_hash(cp))
369 if (handle >= cp->alloc_hash)
370 goto errout_alloc;
371
372
373 err = -ENOMEM;
374 if (!cp->perfect && !cp->h) {
375 if (valid_perfect_hash(cp)) {
376 if (tcindex_alloc_perfect_hash(cp) < 0)
377 goto errout_alloc;
378 balloc = 1;
379 } else {
380 struct tcindex_filter __rcu **hash;
381
382 hash = kcalloc(cp->hash,
383 sizeof(struct tcindex_filter *),
384 GFP_KERNEL);
385
386 if (!hash)
387 goto errout_alloc;
388
389 cp->h = hash;
390 balloc = 2;
391 }
392 }
393
394 if (cp->perfect)
395 r = cp->perfect + handle;
396 else
397 r = tcindex_lookup(cp, handle) ? : &new_filter_result;
398
399 if (r == &new_filter_result) {
400 f = kzalloc(sizeof(*f), GFP_KERNEL);
401 if (!f)
402 goto errout_alloc;
403 f->key = handle;
404 f->next = NULL;
405 err = tcindex_filter_result_init(&f->result);
406 if (err < 0) {
407 kfree(f);
408 goto errout_alloc;
409 }
410 }
411
412 if (tb[TCA_TCINDEX_CLASSID]) {
413 cr.res.classid = nla_get_u32(tb[TCA_TCINDEX_CLASSID]);
414 tcf_bind_filter(tp, &cr.res, base);
415 }
416
417 if (old_r)
418 tcf_exts_change(tp, &r->exts, &e);
419 else
420 tcf_exts_change(tp, &cr.exts, &e);
421
422 if (old_r && old_r != r) {
423 err = tcindex_filter_result_init(old_r);
424 if (err < 0) {
425 kfree(f);
426 goto errout_alloc;
427 }
428 }
429
430 oldp = p;
431 r->res = cr.res;
432 rcu_assign_pointer(tp->root, cp);
433
434 if (r == &new_filter_result) {
435 struct tcindex_filter *nfp;
436 struct tcindex_filter __rcu **fp;
437
438 tcf_exts_change(tp, &f->result.exts, &r->exts);
439
440 fp = cp->h + (handle % cp->hash);
441 for (nfp = rtnl_dereference(*fp);
442 nfp;
443 fp = &nfp->next, nfp = rtnl_dereference(*fp))
444 ; /* nothing */
445
446 rcu_assign_pointer(*fp, f);
447 }
448
449 if (oldp)
450 call_rcu(&oldp->rcu, __tcindex_partial_destroy);
451 return 0;
452
453 errout_alloc:
454 if (balloc == 1)
455 tcindex_free_perfect_hash(cp);
456 else if (balloc == 2)
457 kfree(cp->h);
458 errout1:
459 tcf_exts_destroy(&cr.exts);
460 tcf_exts_destroy(&new_filter_result.exts);
461 errout:
462 kfree(cp);
463 tcf_exts_destroy(&e);
464 return err;
465 }
466
467 static int
468 tcindex_change(struct net *net, struct sk_buff *in_skb,
469 struct tcf_proto *tp, unsigned long base, u32 handle,
470 struct nlattr **tca, unsigned long *arg, bool ovr)
471 {
472 struct nlattr *opt = tca[TCA_OPTIONS];
473 struct nlattr *tb[TCA_TCINDEX_MAX + 1];
474 struct tcindex_data *p = rtnl_dereference(tp->root);
475 struct tcindex_filter_result *r = (struct tcindex_filter_result *) *arg;
476 int err;
477
478 pr_debug("tcindex_change(tp %p,handle 0x%08x,tca %p,arg %p),opt %p,"
479 "p %p,r %p,*arg 0x%lx\n",
480 tp, handle, tca, arg, opt, p, r, arg ? *arg : 0L);
481
482 if (!opt)
483 return 0;
484
485 err = nla_parse_nested(tb, TCA_TCINDEX_MAX, opt, tcindex_policy);
486 if (err < 0)
487 return err;
488
489 return tcindex_set_parms(net, tp, base, handle, p, r, tb,
490 tca[TCA_RATE], ovr);
491 }
492
493 static void tcindex_walk(struct tcf_proto *tp, struct tcf_walker *walker)
494 {
495 struct tcindex_data *p = rtnl_dereference(tp->root);
496 struct tcindex_filter *f, *next;
497 int i;
498
499 pr_debug("tcindex_walk(tp %p,walker %p),p %p\n", tp, walker, p);
500 if (p->perfect) {
501 for (i = 0; i < p->hash; i++) {
502 if (!p->perfect[i].res.class)
503 continue;
504 if (walker->count >= walker->skip) {
505 if (walker->fn(tp,
506 (unsigned long) (p->perfect+i), walker)
507 < 0) {
508 walker->stop = 1;
509 return;
510 }
511 }
512 walker->count++;
513 }
514 }
515 if (!p->h)
516 return;
517 for (i = 0; i < p->hash; i++) {
518 for (f = rtnl_dereference(p->h[i]); f; f = next) {
519 next = rtnl_dereference(f->next);
520 if (walker->count >= walker->skip) {
521 if (walker->fn(tp, (unsigned long) &f->result,
522 walker) < 0) {
523 walker->stop = 1;
524 return;
525 }
526 }
527 walker->count++;
528 }
529 }
530 }
531
532 static bool tcindex_destroy(struct tcf_proto *tp, bool force)
533 {
534 struct tcindex_data *p = rtnl_dereference(tp->root);
535 struct tcf_walker walker;
536
537 if (!force)
538 return false;
539
540 pr_debug("tcindex_destroy(tp %p),p %p\n", tp, p);
541 walker.count = 0;
542 walker.skip = 0;
543 walker.fn = tcindex_destroy_element;
544 tcindex_walk(tp, &walker);
545
546 call_rcu(&p->rcu, __tcindex_destroy);
547 return true;
548 }
549
550
551 static int tcindex_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
552 struct sk_buff *skb, struct tcmsg *t)
553 {
554 struct tcindex_data *p = rtnl_dereference(tp->root);
555 struct tcindex_filter_result *r = (struct tcindex_filter_result *) fh;
556 struct nlattr *nest;
557
558 pr_debug("tcindex_dump(tp %p,fh 0x%lx,skb %p,t %p),p %p,r %p\n",
559 tp, fh, skb, t, p, r);
560 pr_debug("p->perfect %p p->h %p\n", p->perfect, p->h);
561
562 nest = nla_nest_start(skb, TCA_OPTIONS);
563 if (nest == NULL)
564 goto nla_put_failure;
565
566 if (!fh) {
567 t->tcm_handle = ~0; /* whatever ... */
568 if (nla_put_u32(skb, TCA_TCINDEX_HASH, p->hash) ||
569 nla_put_u16(skb, TCA_TCINDEX_MASK, p->mask) ||
570 nla_put_u32(skb, TCA_TCINDEX_SHIFT, p->shift) ||
571 nla_put_u32(skb, TCA_TCINDEX_FALL_THROUGH, p->fall_through))
572 goto nla_put_failure;
573 nla_nest_end(skb, nest);
574 } else {
575 if (p->perfect) {
576 t->tcm_handle = r - p->perfect;
577 } else {
578 struct tcindex_filter *f;
579 struct tcindex_filter __rcu **fp;
580 int i;
581
582 t->tcm_handle = 0;
583 for (i = 0; !t->tcm_handle && i < p->hash; i++) {
584 fp = &p->h[i];
585 for (f = rtnl_dereference(*fp);
586 !t->tcm_handle && f;
587 fp = &f->next, f = rtnl_dereference(*fp)) {
588 if (&f->result == r)
589 t->tcm_handle = f->key;
590 }
591 }
592 }
593 pr_debug("handle = %d\n", t->tcm_handle);
594 if (r->res.class &&
595 nla_put_u32(skb, TCA_TCINDEX_CLASSID, r->res.classid))
596 goto nla_put_failure;
597
598 if (tcf_exts_dump(skb, &r->exts) < 0)
599 goto nla_put_failure;
600 nla_nest_end(skb, nest);
601
602 if (tcf_exts_dump_stats(skb, &r->exts) < 0)
603 goto nla_put_failure;
604 }
605
606 return skb->len;
607
608 nla_put_failure:
609 nla_nest_cancel(skb, nest);
610 return -1;
611 }
612
613 static struct tcf_proto_ops cls_tcindex_ops __read_mostly = {
614 .kind = "tcindex",
615 .classify = tcindex_classify,
616 .init = tcindex_init,
617 .destroy = tcindex_destroy,
618 .get = tcindex_get,
619 .change = tcindex_change,
620 .delete = tcindex_delete,
621 .walk = tcindex_walk,
622 .dump = tcindex_dump,
623 .owner = THIS_MODULE,
624 };
625
626 static int __init init_tcindex(void)
627 {
628 return register_tcf_proto_ops(&cls_tcindex_ops);
629 }
630
631 static void __exit exit_tcindex(void)
632 {
633 unregister_tcf_proto_ops(&cls_tcindex_ops);
634 }
635
636 module_init(init_tcindex)
637 module_exit(exit_tcindex)
638 MODULE_LICENSE("GPL");
Linux with added FPC-III support