Linux 2.6.33-rc8
[linux-2.6/lguest.git] / net / sched / cls_u32.c
blob07372f60bee3364588cb94fcfcad1cc7bee514e5
1 /*
2 * net/sched/cls_u32.c Ugly (or Universal) 32bit key Packet Classifier.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
11 * The filters are packed to hash tables of key nodes
12 * with a set of 32bit key/mask pairs at every node.
13 * Nodes reference next level hash tables etc.
15 * This scheme is the best universal classifier I managed to
16 * invent; it is not super-fast, but it is not slow (provided you
17 * program it correctly), and general enough. And its relative
18 * speed grows as the number of rules becomes larger.
20 * It seems that it represents the best middle point between
21 * speed and manageability both by human and by machine.
23 * It is especially useful for link sharing combined with QoS;
24 * pure RSVP doesn't need such a general approach and can use
25 * much simpler (and faster) schemes, sort of cls_rsvp.c.
27 * JHS: We should remove the CONFIG_NET_CLS_IND from here
28 * eventually when the meta match extension is made available
30 * nfmark match added by Catalin(ux aka Dino) BOIE <catab at umbrella.ro>
33 #include <linux/module.h>
34 #include <linux/types.h>
35 #include <linux/kernel.h>
36 #include <linux/string.h>
37 #include <linux/errno.h>
38 #include <linux/rtnetlink.h>
39 #include <linux/skbuff.h>
40 #include <net/netlink.h>
41 #include <net/act_api.h>
42 #include <net/pkt_cls.h>
44 struct tc_u_knode
46 struct tc_u_knode *next;
47 u32 handle;
48 struct tc_u_hnode *ht_up;
49 struct tcf_exts exts;
50 #ifdef CONFIG_NET_CLS_IND
51 char indev[IFNAMSIZ];
52 #endif
53 u8 fshift;
54 struct tcf_result res;
55 struct tc_u_hnode *ht_down;
56 #ifdef CONFIG_CLS_U32_PERF
57 struct tc_u32_pcnt *pf;
58 #endif
59 #ifdef CONFIG_CLS_U32_MARK
60 struct tc_u32_mark mark;
61 #endif
62 struct tc_u32_sel sel;
65 struct tc_u_hnode
67 struct tc_u_hnode *next;
68 u32 handle;
69 u32 prio;
70 struct tc_u_common *tp_c;
71 int refcnt;
72 unsigned divisor;
73 struct tc_u_knode *ht[1];
76 struct tc_u_common
78 struct tc_u_hnode *hlist;
79 struct Qdisc *q;
80 int refcnt;
81 u32 hgenerator;
84 static const struct tcf_ext_map u32_ext_map = {
85 .action = TCA_U32_ACT,
86 .police = TCA_U32_POLICE
89 static __inline__ unsigned u32_hash_fold(__be32 key, struct tc_u32_sel *sel, u8 fshift)
91 unsigned h = ntohl(key & sel->hmask)>>fshift;
93 return h;
96 static int u32_classify(struct sk_buff *skb, struct tcf_proto *tp, struct tcf_result *res)
98 struct {
99 struct tc_u_knode *knode;
100 u8 *ptr;
101 } stack[TC_U32_MAXDEPTH];
103 struct tc_u_hnode *ht = (struct tc_u_hnode*)tp->root;
104 u8 *ptr = skb_network_header(skb);
105 struct tc_u_knode *n;
106 int sdepth = 0;
107 int off2 = 0;
108 int sel = 0;
109 #ifdef CONFIG_CLS_U32_PERF
110 int j;
111 #endif
112 int i, r;
114 next_ht:
115 n = ht->ht[sel];
117 next_knode:
118 if (n) {
119 struct tc_u32_key *key = n->sel.keys;
121 #ifdef CONFIG_CLS_U32_PERF
122 n->pf->rcnt +=1;
123 j = 0;
124 #endif
126 #ifdef CONFIG_CLS_U32_MARK
127 if ((skb->mark & n->mark.mask) != n->mark.val) {
128 n = n->next;
129 goto next_knode;
130 } else {
131 n->mark.success++;
133 #endif
135 for (i = n->sel.nkeys; i>0; i--, key++) {
137 if ((*(__be32*)(ptr+key->off+(off2&key->offmask))^key->val)&key->mask) {
138 n = n->next;
139 goto next_knode;
141 #ifdef CONFIG_CLS_U32_PERF
142 n->pf->kcnts[j] +=1;
143 j++;
144 #endif
146 if (n->ht_down == NULL) {
147 check_terminal:
148 if (n->sel.flags&TC_U32_TERMINAL) {
150 *res = n->res;
151 #ifdef CONFIG_NET_CLS_IND
152 if (!tcf_match_indev(skb, n->indev)) {
153 n = n->next;
154 goto next_knode;
156 #endif
157 #ifdef CONFIG_CLS_U32_PERF
158 n->pf->rhit +=1;
159 #endif
160 r = tcf_exts_exec(skb, &n->exts, res);
161 if (r < 0) {
162 n = n->next;
163 goto next_knode;
166 return r;
168 n = n->next;
169 goto next_knode;
172 /* PUSH */
173 if (sdepth >= TC_U32_MAXDEPTH)
174 goto deadloop;
175 stack[sdepth].knode = n;
176 stack[sdepth].ptr = ptr;
177 sdepth++;
179 ht = n->ht_down;
180 sel = 0;
181 if (ht->divisor)
182 sel = ht->divisor&u32_hash_fold(*(__be32*)(ptr+n->sel.hoff), &n->sel,n->fshift);
184 if (!(n->sel.flags&(TC_U32_VAROFFSET|TC_U32_OFFSET|TC_U32_EAT)))
185 goto next_ht;
187 if (n->sel.flags&(TC_U32_OFFSET|TC_U32_VAROFFSET)) {
188 off2 = n->sel.off + 3;
189 if (n->sel.flags&TC_U32_VAROFFSET)
190 off2 += ntohs(n->sel.offmask & *(__be16*)(ptr+n->sel.offoff)) >>n->sel.offshift;
191 off2 &= ~3;
193 if (n->sel.flags&TC_U32_EAT) {
194 ptr += off2;
195 off2 = 0;
198 if (ptr < skb_tail_pointer(skb))
199 goto next_ht;
202 /* POP */
203 if (sdepth--) {
204 n = stack[sdepth].knode;
205 ht = n->ht_up;
206 ptr = stack[sdepth].ptr;
207 goto check_terminal;
209 return -1;
211 deadloop:
212 if (net_ratelimit())
213 printk("cls_u32: dead loop\n");
214 return -1;
217 static __inline__ struct tc_u_hnode *
218 u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
220 struct tc_u_hnode *ht;
222 for (ht = tp_c->hlist; ht; ht = ht->next)
223 if (ht->handle == handle)
224 break;
226 return ht;
229 static __inline__ struct tc_u_knode *
230 u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
232 unsigned sel;
233 struct tc_u_knode *n = NULL;
235 sel = TC_U32_HASH(handle);
236 if (sel > ht->divisor)
237 goto out;
239 for (n = ht->ht[sel]; n; n = n->next)
240 if (n->handle == handle)
241 break;
242 out:
243 return n;
247 static unsigned long u32_get(struct tcf_proto *tp, u32 handle)
249 struct tc_u_hnode *ht;
250 struct tc_u_common *tp_c = tp->data;
252 if (TC_U32_HTID(handle) == TC_U32_ROOT)
253 ht = tp->root;
254 else
255 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
257 if (!ht)
258 return 0;
260 if (TC_U32_KEY(handle) == 0)
261 return (unsigned long)ht;
263 return (unsigned long)u32_lookup_key(ht, handle);
266 static void u32_put(struct tcf_proto *tp, unsigned long f)
270 static u32 gen_new_htid(struct tc_u_common *tp_c)
272 int i = 0x800;
274 do {
275 if (++tp_c->hgenerator == 0x7FF)
276 tp_c->hgenerator = 1;
277 } while (--i>0 && u32_lookup_ht(tp_c, (tp_c->hgenerator|0x800)<<20));
279 return i > 0 ? (tp_c->hgenerator|0x800)<<20 : 0;
282 static int u32_init(struct tcf_proto *tp)
284 struct tc_u_hnode *root_ht;
285 struct tc_u_common *tp_c;
287 tp_c = tp->q->u32_node;
289 root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL);
290 if (root_ht == NULL)
291 return -ENOBUFS;
293 root_ht->divisor = 0;
294 root_ht->refcnt++;
295 root_ht->handle = tp_c ? gen_new_htid(tp_c) : 0x80000000;
296 root_ht->prio = tp->prio;
298 if (tp_c == NULL) {
299 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
300 if (tp_c == NULL) {
301 kfree(root_ht);
302 return -ENOBUFS;
304 tp_c->q = tp->q;
305 tp->q->u32_node = tp_c;
308 tp_c->refcnt++;
309 root_ht->next = tp_c->hlist;
310 tp_c->hlist = root_ht;
311 root_ht->tp_c = tp_c;
313 tp->root = root_ht;
314 tp->data = tp_c;
315 return 0;
318 static int u32_destroy_key(struct tcf_proto *tp, struct tc_u_knode *n)
320 tcf_unbind_filter(tp, &n->res);
321 tcf_exts_destroy(tp, &n->exts);
322 if (n->ht_down)
323 n->ht_down->refcnt--;
324 #ifdef CONFIG_CLS_U32_PERF
325 kfree(n->pf);
326 #endif
327 kfree(n);
328 return 0;
331 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode* key)
333 struct tc_u_knode **kp;
334 struct tc_u_hnode *ht = key->ht_up;
336 if (ht) {
337 for (kp = &ht->ht[TC_U32_HASH(key->handle)]; *kp; kp = &(*kp)->next) {
338 if (*kp == key) {
339 tcf_tree_lock(tp);
340 *kp = key->next;
341 tcf_tree_unlock(tp);
343 u32_destroy_key(tp, key);
344 return 0;
348 WARN_ON(1);
349 return 0;
352 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
354 struct tc_u_knode *n;
355 unsigned h;
357 for (h=0; h<=ht->divisor; h++) {
358 while ((n = ht->ht[h]) != NULL) {
359 ht->ht[h] = n->next;
361 u32_destroy_key(tp, n);
366 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
368 struct tc_u_common *tp_c = tp->data;
369 struct tc_u_hnode **hn;
371 WARN_ON(ht->refcnt);
373 u32_clear_hnode(tp, ht);
375 for (hn = &tp_c->hlist; *hn; hn = &(*hn)->next) {
376 if (*hn == ht) {
377 *hn = ht->next;
378 kfree(ht);
379 return 0;
383 WARN_ON(1);
384 return -ENOENT;
387 static void u32_destroy(struct tcf_proto *tp)
389 struct tc_u_common *tp_c = tp->data;
390 struct tc_u_hnode *root_ht = tp->root;
392 WARN_ON(root_ht == NULL);
394 if (root_ht && --root_ht->refcnt == 0)
395 u32_destroy_hnode(tp, root_ht);
397 if (--tp_c->refcnt == 0) {
398 struct tc_u_hnode *ht;
400 tp->q->u32_node = NULL;
402 for (ht = tp_c->hlist; ht; ht = ht->next) {
403 ht->refcnt--;
404 u32_clear_hnode(tp, ht);
407 while ((ht = tp_c->hlist) != NULL) {
408 tp_c->hlist = ht->next;
410 WARN_ON(ht->refcnt != 0);
412 kfree(ht);
415 kfree(tp_c);
418 tp->data = NULL;
421 static int u32_delete(struct tcf_proto *tp, unsigned long arg)
423 struct tc_u_hnode *ht = (struct tc_u_hnode*)arg;
425 if (ht == NULL)
426 return 0;
428 if (TC_U32_KEY(ht->handle))
429 return u32_delete_key(tp, (struct tc_u_knode*)ht);
431 if (tp->root == ht)
432 return -EINVAL;
434 if (ht->refcnt == 1) {
435 ht->refcnt--;
436 u32_destroy_hnode(tp, ht);
437 } else {
438 return -EBUSY;
441 return 0;
444 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 handle)
446 struct tc_u_knode *n;
447 unsigned i = 0x7FF;
449 for (n=ht->ht[TC_U32_HASH(handle)]; n; n = n->next)
450 if (i < TC_U32_NODE(n->handle))
451 i = TC_U32_NODE(n->handle);
452 i++;
454 return handle|(i>0xFFF ? 0xFFF : i);
457 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
458 [TCA_U32_CLASSID] = { .type = NLA_U32 },
459 [TCA_U32_HASH] = { .type = NLA_U32 },
460 [TCA_U32_LINK] = { .type = NLA_U32 },
461 [TCA_U32_DIVISOR] = { .type = NLA_U32 },
462 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) },
463 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ },
464 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) },
467 static int u32_set_parms(struct tcf_proto *tp, unsigned long base,
468 struct tc_u_hnode *ht,
469 struct tc_u_knode *n, struct nlattr **tb,
470 struct nlattr *est)
472 int err;
473 struct tcf_exts e;
475 err = tcf_exts_validate(tp, tb, est, &e, &u32_ext_map);
476 if (err < 0)
477 return err;
479 err = -EINVAL;
480 if (tb[TCA_U32_LINK]) {
481 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
482 struct tc_u_hnode *ht_down = NULL, *ht_old;
484 if (TC_U32_KEY(handle))
485 goto errout;
487 if (handle) {
488 ht_down = u32_lookup_ht(ht->tp_c, handle);
490 if (ht_down == NULL)
491 goto errout;
492 ht_down->refcnt++;
495 tcf_tree_lock(tp);
496 ht_old = n->ht_down;
497 n->ht_down = ht_down;
498 tcf_tree_unlock(tp);
500 if (ht_old)
501 ht_old->refcnt--;
503 if (tb[TCA_U32_CLASSID]) {
504 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
505 tcf_bind_filter(tp, &n->res, base);
508 #ifdef CONFIG_NET_CLS_IND
509 if (tb[TCA_U32_INDEV]) {
510 err = tcf_change_indev(tp, n->indev, tb[TCA_U32_INDEV]);
511 if (err < 0)
512 goto errout;
514 #endif
515 tcf_exts_change(tp, &n->exts, &e);
517 return 0;
518 errout:
519 tcf_exts_destroy(tp, &e);
520 return err;
523 static int u32_change(struct tcf_proto *tp, unsigned long base, u32 handle,
524 struct nlattr **tca,
525 unsigned long *arg)
527 struct tc_u_common *tp_c = tp->data;
528 struct tc_u_hnode *ht;
529 struct tc_u_knode *n;
530 struct tc_u32_sel *s;
531 struct nlattr *opt = tca[TCA_OPTIONS];
532 struct nlattr *tb[TCA_U32_MAX + 1];
533 u32 htid;
534 int err;
536 if (opt == NULL)
537 return handle ? -EINVAL : 0;
539 err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy);
540 if (err < 0)
541 return err;
543 if ((n = (struct tc_u_knode*)*arg) != NULL) {
544 if (TC_U32_KEY(n->handle) == 0)
545 return -EINVAL;
547 return u32_set_parms(tp, base, n->ht_up, n, tb, tca[TCA_RATE]);
550 if (tb[TCA_U32_DIVISOR]) {
551 unsigned divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
553 if (--divisor > 0x100)
554 return -EINVAL;
555 if (TC_U32_KEY(handle))
556 return -EINVAL;
557 if (handle == 0) {
558 handle = gen_new_htid(tp->data);
559 if (handle == 0)
560 return -ENOMEM;
562 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void*), GFP_KERNEL);
563 if (ht == NULL)
564 return -ENOBUFS;
565 ht->tp_c = tp_c;
566 ht->refcnt = 1;
567 ht->divisor = divisor;
568 ht->handle = handle;
569 ht->prio = tp->prio;
570 ht->next = tp_c->hlist;
571 tp_c->hlist = ht;
572 *arg = (unsigned long)ht;
573 return 0;
576 if (tb[TCA_U32_HASH]) {
577 htid = nla_get_u32(tb[TCA_U32_HASH]);
578 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
579 ht = tp->root;
580 htid = ht->handle;
581 } else {
582 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
583 if (ht == NULL)
584 return -EINVAL;
586 } else {
587 ht = tp->root;
588 htid = ht->handle;
591 if (ht->divisor < TC_U32_HASH(htid))
592 return -EINVAL;
594 if (handle) {
595 if (TC_U32_HTID(handle) && TC_U32_HTID(handle^htid))
596 return -EINVAL;
597 handle = htid | TC_U32_NODE(handle);
598 } else
599 handle = gen_new_kid(ht, htid);
601 if (tb[TCA_U32_SEL] == NULL)
602 return -EINVAL;
604 s = nla_data(tb[TCA_U32_SEL]);
606 n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL);
607 if (n == NULL)
608 return -ENOBUFS;
610 #ifdef CONFIG_CLS_U32_PERF
611 n->pf = kzalloc(sizeof(struct tc_u32_pcnt) + s->nkeys*sizeof(u64), GFP_KERNEL);
612 if (n->pf == NULL) {
613 kfree(n);
614 return -ENOBUFS;
616 #endif
618 memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
619 n->ht_up = ht;
620 n->handle = handle;
621 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
623 #ifdef CONFIG_CLS_U32_MARK
624 if (tb[TCA_U32_MARK]) {
625 struct tc_u32_mark *mark;
627 mark = nla_data(tb[TCA_U32_MARK]);
628 memcpy(&n->mark, mark, sizeof(struct tc_u32_mark));
629 n->mark.success = 0;
631 #endif
633 err = u32_set_parms(tp, base, ht, n, tb, tca[TCA_RATE]);
634 if (err == 0) {
635 struct tc_u_knode **ins;
636 for (ins = &ht->ht[TC_U32_HASH(handle)]; *ins; ins = &(*ins)->next)
637 if (TC_U32_NODE(handle) < TC_U32_NODE((*ins)->handle))
638 break;
640 n->next = *ins;
641 tcf_tree_lock(tp);
642 *ins = n;
643 tcf_tree_unlock(tp);
645 *arg = (unsigned long)n;
646 return 0;
648 #ifdef CONFIG_CLS_U32_PERF
649 kfree(n->pf);
650 #endif
651 kfree(n);
652 return err;
655 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg)
657 struct tc_u_common *tp_c = tp->data;
658 struct tc_u_hnode *ht;
659 struct tc_u_knode *n;
660 unsigned h;
662 if (arg->stop)
663 return;
665 for (ht = tp_c->hlist; ht; ht = ht->next) {
666 if (ht->prio != tp->prio)
667 continue;
668 if (arg->count >= arg->skip) {
669 if (arg->fn(tp, (unsigned long)ht, arg) < 0) {
670 arg->stop = 1;
671 return;
674 arg->count++;
675 for (h = 0; h <= ht->divisor; h++) {
676 for (n = ht->ht[h]; n; n = n->next) {
677 if (arg->count < arg->skip) {
678 arg->count++;
679 continue;
681 if (arg->fn(tp, (unsigned long)n, arg) < 0) {
682 arg->stop = 1;
683 return;
685 arg->count++;
691 static int u32_dump(struct tcf_proto *tp, unsigned long fh,
692 struct sk_buff *skb, struct tcmsg *t)
694 struct tc_u_knode *n = (struct tc_u_knode*)fh;
695 struct nlattr *nest;
697 if (n == NULL)
698 return skb->len;
700 t->tcm_handle = n->handle;
702 nest = nla_nest_start(skb, TCA_OPTIONS);
703 if (nest == NULL)
704 goto nla_put_failure;
706 if (TC_U32_KEY(n->handle) == 0) {
707 struct tc_u_hnode *ht = (struct tc_u_hnode*)fh;
708 u32 divisor = ht->divisor+1;
709 NLA_PUT_U32(skb, TCA_U32_DIVISOR, divisor);
710 } else {
711 NLA_PUT(skb, TCA_U32_SEL,
712 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
713 &n->sel);
714 if (n->ht_up) {
715 u32 htid = n->handle & 0xFFFFF000;
716 NLA_PUT_U32(skb, TCA_U32_HASH, htid);
718 if (n->res.classid)
719 NLA_PUT_U32(skb, TCA_U32_CLASSID, n->res.classid);
720 if (n->ht_down)
721 NLA_PUT_U32(skb, TCA_U32_LINK, n->ht_down->handle);
723 #ifdef CONFIG_CLS_U32_MARK
724 if (n->mark.val || n->mark.mask)
725 NLA_PUT(skb, TCA_U32_MARK, sizeof(n->mark), &n->mark);
726 #endif
728 if (tcf_exts_dump(skb, &n->exts, &u32_ext_map) < 0)
729 goto nla_put_failure;
731 #ifdef CONFIG_NET_CLS_IND
732 if(strlen(n->indev))
733 NLA_PUT_STRING(skb, TCA_U32_INDEV, n->indev);
734 #endif
735 #ifdef CONFIG_CLS_U32_PERF
736 NLA_PUT(skb, TCA_U32_PCNT,
737 sizeof(struct tc_u32_pcnt) + n->sel.nkeys*sizeof(u64),
738 n->pf);
739 #endif
742 nla_nest_end(skb, nest);
744 if (TC_U32_KEY(n->handle))
745 if (tcf_exts_dump_stats(skb, &n->exts, &u32_ext_map) < 0)
746 goto nla_put_failure;
747 return skb->len;
749 nla_put_failure:
750 nla_nest_cancel(skb, nest);
751 return -1;
754 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
755 .kind = "u32",
756 .classify = u32_classify,
757 .init = u32_init,
758 .destroy = u32_destroy,
759 .get = u32_get,
760 .put = u32_put,
761 .change = u32_change,
762 .delete = u32_delete,
763 .walk = u32_walk,
764 .dump = u32_dump,
765 .owner = THIS_MODULE,
768 static int __init init_u32(void)
770 printk("u32 classifier\n");
771 #ifdef CONFIG_CLS_U32_PERF
772 printk(" Performance counters on\n");
773 #endif
774 #ifdef CONFIG_NET_CLS_IND
775 printk(" input device check on \n");
776 #endif
777 #ifdef CONFIG_NET_CLS_ACT
778 printk(" Actions configured \n");
779 #endif
780 return register_tcf_proto_ops(&cls_u32_ops);
783 static void __exit exit_u32(void)
785 unregister_tcf_proto_ops(&cls_u32_ops);
788 module_init(init_u32)
789 module_exit(exit_u32)
790 MODULE_LICENSE("GPL");