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 <asm/uaccess.h>
34 #include <asm/system.h>
35 #include <linux/bitops.h>
36 #include <linux/module.h>
37 #include <linux/types.h>
38 #include <linux/kernel.h>
39 #include <linux/string.h>
41 #include <linux/socket.h>
42 #include <linux/sockios.h>
44 #include <linux/errno.h>
45 #include <linux/interrupt.h>
46 #include <linux/if_ether.h>
47 #include <linux/inet.h>
48 #include <linux/netdevice.h>
49 #include <linux/etherdevice.h>
50 #include <linux/notifier.h>
51 #include <linux/rtnetlink.h>
53 #include <net/route.h>
54 #include <linux/skbuff.h>
56 #include <net/act_api.h>
57 #include <net/pkt_cls.h>
61 struct tc_u_knode *next;
63 struct tc_u_hnode *ht_up;
65 #ifdef CONFIG_NET_CLS_IND
69 struct tcf_result res;
70 struct tc_u_hnode *ht_down;
71 #ifdef CONFIG_CLS_U32_PERF
72 struct tc_u32_pcnt *pf;
74 #ifdef CONFIG_CLS_U32_MARK
75 struct tc_u32_mark mark;
77 struct tc_u32_sel sel;
82 struct tc_u_hnode *next;
85 struct tc_u_common *tp_c;
88 struct tc_u_knode *ht[1];
93 struct tc_u_common *next;
94 struct tc_u_hnode *hlist;
100 static struct tcf_ext_map u32_ext_map = {
101 .action = TCA_U32_ACT,
102 .police = TCA_U32_POLICE
105 static struct tc_u_common *u32_list;
107 static __inline__ unsigned u32_hash_fold(u32 key, struct tc_u32_sel *sel, u8 fshift)
109 unsigned h = (key & sel->hmask)>>fshift;
114 static int u32_classify(struct sk_buff *skb, struct tcf_proto *tp, struct tcf_result *res)
117 struct tc_u_knode *knode;
119 } stack[TC_U32_MAXDEPTH];
121 struct tc_u_hnode *ht = (struct tc_u_hnode*)tp->root;
122 u8 *ptr = skb->nh.raw;
123 struct tc_u_knode *n;
127 #ifdef CONFIG_CLS_U32_PERF
137 struct tc_u32_key *key = n->sel.keys;
139 #ifdef CONFIG_CLS_U32_PERF
144 #ifdef CONFIG_CLS_U32_MARK
145 if ((skb->mark & n->mark.mask) != n->mark.val) {
153 for (i = n->sel.nkeys; i>0; i--, key++) {
155 if ((*(u32*)(ptr+key->off+(off2&key->offmask))^key->val)&key->mask) {
159 #ifdef CONFIG_CLS_U32_PERF
164 if (n->ht_down == NULL) {
166 if (n->sel.flags&TC_U32_TERMINAL) {
169 #ifdef CONFIG_NET_CLS_IND
170 if (!tcf_match_indev(skb, n->indev)) {
175 #ifdef CONFIG_CLS_U32_PERF
178 r = tcf_exts_exec(skb, &n->exts, res);
191 if (sdepth >= TC_U32_MAXDEPTH)
193 stack[sdepth].knode = n;
194 stack[sdepth].ptr = ptr;
200 sel = ht->divisor&u32_hash_fold(*(u32*)(ptr+n->sel.hoff), &n->sel,n->fshift);
202 if (!(n->sel.flags&(TC_U32_VAROFFSET|TC_U32_OFFSET|TC_U32_EAT)))
205 if (n->sel.flags&(TC_U32_OFFSET|TC_U32_VAROFFSET)) {
206 off2 = n->sel.off + 3;
207 if (n->sel.flags&TC_U32_VAROFFSET)
208 off2 += ntohs(n->sel.offmask & *(u16*)(ptr+n->sel.offoff)) >>n->sel.offshift;
211 if (n->sel.flags&TC_U32_EAT) {
222 n = stack[sdepth].knode;
224 ptr = stack[sdepth].ptr;
231 printk("cls_u32: dead loop\n");
235 static __inline__ struct tc_u_hnode *
236 u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
238 struct tc_u_hnode *ht;
240 for (ht = tp_c->hlist; ht; ht = ht->next)
241 if (ht->handle == handle)
247 static __inline__ struct tc_u_knode *
248 u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
251 struct tc_u_knode *n = NULL;
253 sel = TC_U32_HASH(handle);
254 if (sel > ht->divisor)
257 for (n = ht->ht[sel]; n; n = n->next)
258 if (n->handle == handle)
265 static unsigned long u32_get(struct tcf_proto *tp, u32 handle)
267 struct tc_u_hnode *ht;
268 struct tc_u_common *tp_c = tp->data;
270 if (TC_U32_HTID(handle) == TC_U32_ROOT)
273 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
278 if (TC_U32_KEY(handle) == 0)
279 return (unsigned long)ht;
281 return (unsigned long)u32_lookup_key(ht, handle);
284 static void u32_put(struct tcf_proto *tp, unsigned long f)
288 static u32 gen_new_htid(struct tc_u_common *tp_c)
293 if (++tp_c->hgenerator == 0x7FF)
294 tp_c->hgenerator = 1;
295 } while (--i>0 && u32_lookup_ht(tp_c, (tp_c->hgenerator|0x800)<<20));
297 return i > 0 ? (tp_c->hgenerator|0x800)<<20 : 0;
300 static int u32_init(struct tcf_proto *tp)
302 struct tc_u_hnode *root_ht;
303 struct tc_u_common *tp_c;
305 for (tp_c = u32_list; tp_c; tp_c = tp_c->next)
306 if (tp_c->q == tp->q)
309 root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL);
313 root_ht->divisor = 0;
315 root_ht->handle = tp_c ? gen_new_htid(tp_c) : 0x80000000;
316 root_ht->prio = tp->prio;
319 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
325 tp_c->next = u32_list;
330 root_ht->next = tp_c->hlist;
331 tp_c->hlist = root_ht;
332 root_ht->tp_c = tp_c;
339 static int u32_destroy_key(struct tcf_proto *tp, struct tc_u_knode *n)
341 tcf_unbind_filter(tp, &n->res);
342 tcf_exts_destroy(tp, &n->exts);
344 n->ht_down->refcnt--;
345 #ifdef CONFIG_CLS_U32_PERF
352 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode* key)
354 struct tc_u_knode **kp;
355 struct tc_u_hnode *ht = key->ht_up;
358 for (kp = &ht->ht[TC_U32_HASH(key->handle)]; *kp; kp = &(*kp)->next) {
364 u32_destroy_key(tp, key);
373 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
375 struct tc_u_knode *n;
378 for (h=0; h<=ht->divisor; h++) {
379 while ((n = ht->ht[h]) != NULL) {
382 u32_destroy_key(tp, n);
387 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
389 struct tc_u_common *tp_c = tp->data;
390 struct tc_u_hnode **hn;
392 BUG_TRAP(!ht->refcnt);
394 u32_clear_hnode(tp, ht);
396 for (hn = &tp_c->hlist; *hn; hn = &(*hn)->next) {
408 static void u32_destroy(struct tcf_proto *tp)
410 struct tc_u_common *tp_c = tp->data;
411 struct tc_u_hnode *root_ht = xchg(&tp->root, NULL);
413 BUG_TRAP(root_ht != NULL);
415 if (root_ht && --root_ht->refcnt == 0)
416 u32_destroy_hnode(tp, root_ht);
418 if (--tp_c->refcnt == 0) {
419 struct tc_u_hnode *ht;
420 struct tc_u_common **tp_cp;
422 for (tp_cp = &u32_list; *tp_cp; tp_cp = &(*tp_cp)->next) {
423 if (*tp_cp == tp_c) {
429 for (ht=tp_c->hlist; ht; ht = ht->next)
430 u32_clear_hnode(tp, ht);
432 while ((ht = tp_c->hlist) != NULL) {
433 tp_c->hlist = ht->next;
435 BUG_TRAP(ht->refcnt == 0);
446 static int u32_delete(struct tcf_proto *tp, unsigned long arg)
448 struct tc_u_hnode *ht = (struct tc_u_hnode*)arg;
453 if (TC_U32_KEY(ht->handle))
454 return u32_delete_key(tp, (struct tc_u_knode*)ht);
459 if (--ht->refcnt == 0)
460 u32_destroy_hnode(tp, ht);
465 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 handle)
467 struct tc_u_knode *n;
470 for (n=ht->ht[TC_U32_HASH(handle)]; n; n = n->next)
471 if (i < TC_U32_NODE(n->handle))
472 i = TC_U32_NODE(n->handle);
475 return handle|(i>0xFFF ? 0xFFF : i);
478 static int u32_set_parms(struct tcf_proto *tp, unsigned long base,
479 struct tc_u_hnode *ht,
480 struct tc_u_knode *n, struct rtattr **tb,
486 err = tcf_exts_validate(tp, tb, est, &e, &u32_ext_map);
491 if (tb[TCA_U32_LINK-1]) {
492 u32 handle = *(u32*)RTA_DATA(tb[TCA_U32_LINK-1]);
493 struct tc_u_hnode *ht_down = NULL;
495 if (TC_U32_KEY(handle))
499 ht_down = u32_lookup_ht(ht->tp_c, handle);
507 ht_down = xchg(&n->ht_down, ht_down);
513 if (tb[TCA_U32_CLASSID-1]) {
514 n->res.classid = *(u32*)RTA_DATA(tb[TCA_U32_CLASSID-1]);
515 tcf_bind_filter(tp, &n->res, base);
518 #ifdef CONFIG_NET_CLS_IND
519 if (tb[TCA_U32_INDEV-1]) {
520 int err = tcf_change_indev(tp, n->indev, tb[TCA_U32_INDEV-1]);
525 tcf_exts_change(tp, &n->exts, &e);
529 tcf_exts_destroy(tp, &e);
533 static int u32_change(struct tcf_proto *tp, unsigned long base, u32 handle,
537 struct tc_u_common *tp_c = tp->data;
538 struct tc_u_hnode *ht;
539 struct tc_u_knode *n;
540 struct tc_u32_sel *s;
541 struct rtattr *opt = tca[TCA_OPTIONS-1];
542 struct rtattr *tb[TCA_U32_MAX];
547 return handle ? -EINVAL : 0;
549 if (rtattr_parse_nested(tb, TCA_U32_MAX, opt) < 0)
552 if ((n = (struct tc_u_knode*)*arg) != NULL) {
553 if (TC_U32_KEY(n->handle) == 0)
556 return u32_set_parms(tp, base, n->ht_up, n, tb, tca[TCA_RATE-1]);
559 if (tb[TCA_U32_DIVISOR-1]) {
560 unsigned divisor = *(unsigned*)RTA_DATA(tb[TCA_U32_DIVISOR-1]);
562 if (--divisor > 0x100)
564 if (TC_U32_KEY(handle))
567 handle = gen_new_htid(tp->data);
571 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void*), GFP_KERNEL);
576 ht->divisor = divisor;
579 ht->next = tp_c->hlist;
581 *arg = (unsigned long)ht;
585 if (tb[TCA_U32_HASH-1]) {
586 htid = *(unsigned*)RTA_DATA(tb[TCA_U32_HASH-1]);
587 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
591 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
600 if (ht->divisor < TC_U32_HASH(htid))
604 if (TC_U32_HTID(handle) && TC_U32_HTID(handle^htid))
606 handle = htid | TC_U32_NODE(handle);
608 handle = gen_new_kid(ht, htid);
610 if (tb[TCA_U32_SEL-1] == 0 ||
611 RTA_PAYLOAD(tb[TCA_U32_SEL-1]) < sizeof(struct tc_u32_sel))
614 s = RTA_DATA(tb[TCA_U32_SEL-1]);
616 n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL);
620 #ifdef CONFIG_CLS_U32_PERF
621 n->pf = kzalloc(sizeof(struct tc_u32_pcnt) + s->nkeys*sizeof(u64), GFP_KERNEL);
628 memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
635 while (!(mask & 1)) {
643 #ifdef CONFIG_CLS_U32_MARK
644 if (tb[TCA_U32_MARK-1]) {
645 struct tc_u32_mark *mark;
647 if (RTA_PAYLOAD(tb[TCA_U32_MARK-1]) < sizeof(struct tc_u32_mark)) {
648 #ifdef CONFIG_CLS_U32_PERF
654 mark = RTA_DATA(tb[TCA_U32_MARK-1]);
655 memcpy(&n->mark, mark, sizeof(struct tc_u32_mark));
660 err = u32_set_parms(tp, base, ht, n, tb, tca[TCA_RATE-1]);
662 struct tc_u_knode **ins;
663 for (ins = &ht->ht[TC_U32_HASH(handle)]; *ins; ins = &(*ins)->next)
664 if (TC_U32_NODE(handle) < TC_U32_NODE((*ins)->handle))
671 *arg = (unsigned long)n;
674 #ifdef CONFIG_CLS_U32_PERF
681 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg)
683 struct tc_u_common *tp_c = tp->data;
684 struct tc_u_hnode *ht;
685 struct tc_u_knode *n;
691 for (ht = tp_c->hlist; ht; ht = ht->next) {
692 if (ht->prio != tp->prio)
694 if (arg->count >= arg->skip) {
695 if (arg->fn(tp, (unsigned long)ht, arg) < 0) {
701 for (h = 0; h <= ht->divisor; h++) {
702 for (n = ht->ht[h]; n; n = n->next) {
703 if (arg->count < arg->skip) {
707 if (arg->fn(tp, (unsigned long)n, arg) < 0) {
717 static int u32_dump(struct tcf_proto *tp, unsigned long fh,
718 struct sk_buff *skb, struct tcmsg *t)
720 struct tc_u_knode *n = (struct tc_u_knode*)fh;
721 unsigned char *b = skb->tail;
727 t->tcm_handle = n->handle;
729 rta = (struct rtattr*)b;
730 RTA_PUT(skb, TCA_OPTIONS, 0, NULL);
732 if (TC_U32_KEY(n->handle) == 0) {
733 struct tc_u_hnode *ht = (struct tc_u_hnode*)fh;
734 u32 divisor = ht->divisor+1;
735 RTA_PUT(skb, TCA_U32_DIVISOR, 4, &divisor);
737 RTA_PUT(skb, TCA_U32_SEL,
738 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
741 u32 htid = n->handle & 0xFFFFF000;
742 RTA_PUT(skb, TCA_U32_HASH, 4, &htid);
745 RTA_PUT(skb, TCA_U32_CLASSID, 4, &n->res.classid);
747 RTA_PUT(skb, TCA_U32_LINK, 4, &n->ht_down->handle);
749 #ifdef CONFIG_CLS_U32_MARK
750 if (n->mark.val || n->mark.mask)
751 RTA_PUT(skb, TCA_U32_MARK, sizeof(n->mark), &n->mark);
754 if (tcf_exts_dump(skb, &n->exts, &u32_ext_map) < 0)
757 #ifdef CONFIG_NET_CLS_IND
759 RTA_PUT(skb, TCA_U32_INDEV, IFNAMSIZ, n->indev);
761 #ifdef CONFIG_CLS_U32_PERF
762 RTA_PUT(skb, TCA_U32_PCNT,
763 sizeof(struct tc_u32_pcnt) + n->sel.nkeys*sizeof(u64),
768 rta->rta_len = skb->tail - b;
769 if (TC_U32_KEY(n->handle))
770 if (tcf_exts_dump_stats(skb, &n->exts, &u32_ext_map) < 0)
775 skb_trim(skb, b - skb->data);
779 static struct tcf_proto_ops cls_u32_ops = {
782 .classify = u32_classify,
784 .destroy = u32_destroy,
787 .change = u32_change,
788 .delete = u32_delete,
791 .owner = THIS_MODULE,
794 static int __init init_u32(void)
796 printk("u32 classifier\n");
797 #ifdef CONFIG_CLS_U32_PERF
798 printk(" Performance counters on\n");
800 #ifdef CONFIG_NET_CLS_POLICE
801 printk(" OLD policer on \n");
803 #ifdef CONFIG_NET_CLS_IND
804 printk(" input device check on \n");
806 #ifdef CONFIG_NET_CLS_ACT
807 printk(" Actions configured \n");
809 return register_tcf_proto_ops(&cls_u32_ops);
812 static void __exit exit_u32(void)
814 unregister_tcf_proto_ops(&cls_u32_ops);
817 module_init(init_u32)
818 module_exit(exit_u32)
819 MODULE_LICENSE("GPL");