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/slab.h>
35 #include <linux/types.h>
36 #include <linux/kernel.h>
37 #include <linux/string.h>
38 #include <linux/errno.h>
39 #include <linux/rtnetlink.h>
40 #include <linux/skbuff.h>
41 #include <net/netlink.h>
42 #include <net/act_api.h>
43 #include <net/pkt_cls.h>
47 struct tc_u_knode *next;
49 struct tc_u_hnode *ht_up;
51 #ifdef CONFIG_NET_CLS_IND
55 struct tcf_result res;
56 struct tc_u_hnode *ht_down;
57 #ifdef CONFIG_CLS_U32_PERF
58 struct tc_u32_pcnt *pf;
60 #ifdef CONFIG_CLS_U32_MARK
61 struct tc_u32_mark mark;
63 struct tc_u32_sel sel;
68 struct tc_u_hnode *next;
71 struct tc_u_common *tp_c;
74 struct tc_u_knode *ht[1];
79 struct tc_u_hnode *hlist;
85 static const struct tcf_ext_map u32_ext_map = {
86 .action = TCA_U32_ACT,
87 .police = TCA_U32_POLICE
90 static __inline__ unsigned u32_hash_fold(__be32 key, struct tc_u32_sel *sel, u8 fshift)
92 unsigned h = ntohl(key & sel->hmask)>>fshift;
97 static int u32_classify(struct sk_buff *skb, struct tcf_proto *tp, struct tcf_result *res)
100 struct tc_u_knode *knode;
102 } stack[TC_U32_MAXDEPTH];
104 struct tc_u_hnode *ht = (struct tc_u_hnode*)tp->root;
105 unsigned int off = skb_network_offset(skb);
106 struct tc_u_knode *n;
110 #ifdef CONFIG_CLS_U32_PERF
120 struct tc_u32_key *key = n->sel.keys;
122 #ifdef CONFIG_CLS_U32_PERF
127 #ifdef CONFIG_CLS_U32_MARK
128 if ((skb->mark & n->mark.mask) != n->mark.val) {
136 for (i = n->sel.nkeys; i>0; i--, key++) {
140 toff = off + key->off + (off2 & key->offmask);
141 data = skb_header_pointer(skb, toff, 4, &_data);
144 if ((*data ^ key->val) & key->mask) {
148 #ifdef CONFIG_CLS_U32_PERF
153 if (n->ht_down == NULL) {
155 if (n->sel.flags&TC_U32_TERMINAL) {
158 #ifdef CONFIG_NET_CLS_IND
159 if (!tcf_match_indev(skb, n->indev)) {
164 #ifdef CONFIG_CLS_U32_PERF
167 r = tcf_exts_exec(skb, &n->exts, res);
180 if (sdepth >= TC_U32_MAXDEPTH)
182 stack[sdepth].knode = n;
183 stack[sdepth].off = off;
191 data = skb_header_pointer(skb, off + n->sel.hoff, 4,
195 sel = ht->divisor & u32_hash_fold(*data, &n->sel,
198 if (!(n->sel.flags&(TC_U32_VAROFFSET|TC_U32_OFFSET|TC_U32_EAT)))
201 if (n->sel.flags&(TC_U32_OFFSET|TC_U32_VAROFFSET)) {
202 off2 = n->sel.off + 3;
203 if (n->sel.flags & TC_U32_VAROFFSET) {
206 data = skb_header_pointer(skb,
211 off2 += ntohs(n->sel.offmask & *data) >>
216 if (n->sel.flags&TC_U32_EAT) {
227 n = stack[sdepth].knode;
229 off = stack[sdepth].off;
237 printk(KERN_WARNING "cls_u32: dead loop\n");
241 static __inline__ struct tc_u_hnode *
242 u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
244 struct tc_u_hnode *ht;
246 for (ht = tp_c->hlist; ht; ht = ht->next)
247 if (ht->handle == handle)
253 static __inline__ struct tc_u_knode *
254 u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
257 struct tc_u_knode *n = NULL;
259 sel = TC_U32_HASH(handle);
260 if (sel > ht->divisor)
263 for (n = ht->ht[sel]; n; n = n->next)
264 if (n->handle == handle)
271 static unsigned long u32_get(struct tcf_proto *tp, u32 handle)
273 struct tc_u_hnode *ht;
274 struct tc_u_common *tp_c = tp->data;
276 if (TC_U32_HTID(handle) == TC_U32_ROOT)
279 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
284 if (TC_U32_KEY(handle) == 0)
285 return (unsigned long)ht;
287 return (unsigned long)u32_lookup_key(ht, handle);
290 static void u32_put(struct tcf_proto *tp, unsigned long f)
294 static u32 gen_new_htid(struct tc_u_common *tp_c)
299 if (++tp_c->hgenerator == 0x7FF)
300 tp_c->hgenerator = 1;
301 } while (--i>0 && u32_lookup_ht(tp_c, (tp_c->hgenerator|0x800)<<20));
303 return i > 0 ? (tp_c->hgenerator|0x800)<<20 : 0;
306 static int u32_init(struct tcf_proto *tp)
308 struct tc_u_hnode *root_ht;
309 struct tc_u_common *tp_c;
311 tp_c = tp->q->u32_node;
313 root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL);
317 root_ht->divisor = 0;
319 root_ht->handle = tp_c ? gen_new_htid(tp_c) : 0x80000000;
320 root_ht->prio = tp->prio;
323 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
329 tp->q->u32_node = tp_c;
333 root_ht->next = tp_c->hlist;
334 tp_c->hlist = root_ht;
335 root_ht->tp_c = tp_c;
342 static int u32_destroy_key(struct tcf_proto *tp, struct tc_u_knode *n)
344 tcf_unbind_filter(tp, &n->res);
345 tcf_exts_destroy(tp, &n->exts);
347 n->ht_down->refcnt--;
348 #ifdef CONFIG_CLS_U32_PERF
355 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode* key)
357 struct tc_u_knode **kp;
358 struct tc_u_hnode *ht = key->ht_up;
361 for (kp = &ht->ht[TC_U32_HASH(key->handle)]; *kp; kp = &(*kp)->next) {
367 u32_destroy_key(tp, key);
376 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
378 struct tc_u_knode *n;
381 for (h=0; h<=ht->divisor; h++) {
382 while ((n = ht->ht[h]) != NULL) {
385 u32_destroy_key(tp, n);
390 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
392 struct tc_u_common *tp_c = tp->data;
393 struct tc_u_hnode **hn;
397 u32_clear_hnode(tp, ht);
399 for (hn = &tp_c->hlist; *hn; hn = &(*hn)->next) {
411 static void u32_destroy(struct tcf_proto *tp)
413 struct tc_u_common *tp_c = tp->data;
414 struct tc_u_hnode *root_ht = tp->root;
416 WARN_ON(root_ht == NULL);
418 if (root_ht && --root_ht->refcnt == 0)
419 u32_destroy_hnode(tp, root_ht);
421 if (--tp_c->refcnt == 0) {
422 struct tc_u_hnode *ht;
424 tp->q->u32_node = NULL;
426 for (ht = tp_c->hlist; ht; ht = ht->next) {
428 u32_clear_hnode(tp, ht);
431 while ((ht = tp_c->hlist) != NULL) {
432 tp_c->hlist = ht->next;
434 WARN_ON(ht->refcnt != 0);
445 static int u32_delete(struct tcf_proto *tp, unsigned long arg)
447 struct tc_u_hnode *ht = (struct tc_u_hnode*)arg;
452 if (TC_U32_KEY(ht->handle))
453 return u32_delete_key(tp, (struct tc_u_knode*)ht);
458 if (ht->refcnt == 1) {
460 u32_destroy_hnode(tp, ht);
468 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 handle)
470 struct tc_u_knode *n;
473 for (n=ht->ht[TC_U32_HASH(handle)]; n; n = n->next)
474 if (i < TC_U32_NODE(n->handle))
475 i = TC_U32_NODE(n->handle);
478 return handle|(i>0xFFF ? 0xFFF : i);
481 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
482 [TCA_U32_CLASSID] = { .type = NLA_U32 },
483 [TCA_U32_HASH] = { .type = NLA_U32 },
484 [TCA_U32_LINK] = { .type = NLA_U32 },
485 [TCA_U32_DIVISOR] = { .type = NLA_U32 },
486 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) },
487 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ },
488 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) },
491 static int u32_set_parms(struct tcf_proto *tp, unsigned long base,
492 struct tc_u_hnode *ht,
493 struct tc_u_knode *n, struct nlattr **tb,
499 err = tcf_exts_validate(tp, tb, est, &e, &u32_ext_map);
504 if (tb[TCA_U32_LINK]) {
505 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
506 struct tc_u_hnode *ht_down = NULL, *ht_old;
508 if (TC_U32_KEY(handle))
512 ht_down = u32_lookup_ht(ht->tp_c, handle);
521 n->ht_down = ht_down;
527 if (tb[TCA_U32_CLASSID]) {
528 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
529 tcf_bind_filter(tp, &n->res, base);
532 #ifdef CONFIG_NET_CLS_IND
533 if (tb[TCA_U32_INDEV]) {
534 err = tcf_change_indev(tp, n->indev, tb[TCA_U32_INDEV]);
539 tcf_exts_change(tp, &n->exts, &e);
543 tcf_exts_destroy(tp, &e);
547 static int u32_change(struct tcf_proto *tp, unsigned long base, u32 handle,
551 struct tc_u_common *tp_c = tp->data;
552 struct tc_u_hnode *ht;
553 struct tc_u_knode *n;
554 struct tc_u32_sel *s;
555 struct nlattr *opt = tca[TCA_OPTIONS];
556 struct nlattr *tb[TCA_U32_MAX + 1];
561 return handle ? -EINVAL : 0;
563 err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy);
567 if ((n = (struct tc_u_knode*)*arg) != NULL) {
568 if (TC_U32_KEY(n->handle) == 0)
571 return u32_set_parms(tp, base, n->ht_up, n, tb, tca[TCA_RATE]);
574 if (tb[TCA_U32_DIVISOR]) {
575 unsigned divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
577 if (--divisor > 0x100)
579 if (TC_U32_KEY(handle))
582 handle = gen_new_htid(tp->data);
586 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void*), GFP_KERNEL);
591 ht->divisor = divisor;
594 ht->next = tp_c->hlist;
596 *arg = (unsigned long)ht;
600 if (tb[TCA_U32_HASH]) {
601 htid = nla_get_u32(tb[TCA_U32_HASH]);
602 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
606 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
615 if (ht->divisor < TC_U32_HASH(htid))
619 if (TC_U32_HTID(handle) && TC_U32_HTID(handle^htid))
621 handle = htid | TC_U32_NODE(handle);
623 handle = gen_new_kid(ht, htid);
625 if (tb[TCA_U32_SEL] == NULL)
628 s = nla_data(tb[TCA_U32_SEL]);
630 n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL);
634 #ifdef CONFIG_CLS_U32_PERF
635 n->pf = kzalloc(sizeof(struct tc_u32_pcnt) + s->nkeys*sizeof(u64), GFP_KERNEL);
642 memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
645 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
647 #ifdef CONFIG_CLS_U32_MARK
648 if (tb[TCA_U32_MARK]) {
649 struct tc_u32_mark *mark;
651 mark = nla_data(tb[TCA_U32_MARK]);
652 memcpy(&n->mark, mark, sizeof(struct tc_u32_mark));
657 err = u32_set_parms(tp, base, ht, n, tb, tca[TCA_RATE]);
659 struct tc_u_knode **ins;
660 for (ins = &ht->ht[TC_U32_HASH(handle)]; *ins; ins = &(*ins)->next)
661 if (TC_U32_NODE(handle) < TC_U32_NODE((*ins)->handle))
669 *arg = (unsigned long)n;
672 #ifdef CONFIG_CLS_U32_PERF
679 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg)
681 struct tc_u_common *tp_c = tp->data;
682 struct tc_u_hnode *ht;
683 struct tc_u_knode *n;
689 for (ht = tp_c->hlist; ht; ht = ht->next) {
690 if (ht->prio != tp->prio)
692 if (arg->count >= arg->skip) {
693 if (arg->fn(tp, (unsigned long)ht, arg) < 0) {
699 for (h = 0; h <= ht->divisor; h++) {
700 for (n = ht->ht[h]; n; n = n->next) {
701 if (arg->count < arg->skip) {
705 if (arg->fn(tp, (unsigned long)n, arg) < 0) {
715 static int u32_dump(struct tcf_proto *tp, unsigned long fh,
716 struct sk_buff *skb, struct tcmsg *t)
718 struct tc_u_knode *n = (struct tc_u_knode*)fh;
724 t->tcm_handle = n->handle;
726 nest = nla_nest_start(skb, TCA_OPTIONS);
728 goto nla_put_failure;
730 if (TC_U32_KEY(n->handle) == 0) {
731 struct tc_u_hnode *ht = (struct tc_u_hnode*)fh;
732 u32 divisor = ht->divisor+1;
733 NLA_PUT_U32(skb, TCA_U32_DIVISOR, divisor);
735 NLA_PUT(skb, TCA_U32_SEL,
736 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
739 u32 htid = n->handle & 0xFFFFF000;
740 NLA_PUT_U32(skb, TCA_U32_HASH, htid);
743 NLA_PUT_U32(skb, TCA_U32_CLASSID, n->res.classid);
745 NLA_PUT_U32(skb, TCA_U32_LINK, n->ht_down->handle);
747 #ifdef CONFIG_CLS_U32_MARK
748 if (n->mark.val || n->mark.mask)
749 NLA_PUT(skb, TCA_U32_MARK, sizeof(n->mark), &n->mark);
752 if (tcf_exts_dump(skb, &n->exts, &u32_ext_map) < 0)
753 goto nla_put_failure;
755 #ifdef CONFIG_NET_CLS_IND
757 NLA_PUT_STRING(skb, TCA_U32_INDEV, n->indev);
759 #ifdef CONFIG_CLS_U32_PERF
760 NLA_PUT(skb, TCA_U32_PCNT,
761 sizeof(struct tc_u32_pcnt) + n->sel.nkeys*sizeof(u64),
766 nla_nest_end(skb, nest);
768 if (TC_U32_KEY(n->handle))
769 if (tcf_exts_dump_stats(skb, &n->exts, &u32_ext_map) < 0)
770 goto nla_put_failure;
774 nla_nest_cancel(skb, nest);
778 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
780 .classify = u32_classify,
782 .destroy = u32_destroy,
785 .change = u32_change,
786 .delete = u32_delete,
789 .owner = THIS_MODULE,
792 static int __init init_u32(void)
794 pr_info("u32 classifier\n");
795 #ifdef CONFIG_CLS_U32_PERF
796 pr_info(" Performance counters on\n");
798 #ifdef CONFIG_NET_CLS_IND
799 pr_info(" input device check on\n");
801 #ifdef CONFIG_NET_CLS_ACT
802 pr_info(" Actions configured\n");
804 return register_tcf_proto_ops(&cls_u32_ops);
807 static void __exit exit_u32(void)
809 unregister_tcf_proto_ops(&cls_u32_ops);
812 module_init(init_u32)
813 module_exit(exit_u32)
814 MODULE_LICENSE("GPL");