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 u8 *ptr = skb_network_header(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++) {
138 if ((*(__be32*)(ptr+key->off+(off2&key->offmask))^key->val)&key->mask) {
142 #ifdef CONFIG_CLS_U32_PERF
147 if (n->ht_down == NULL) {
149 if (n->sel.flags&TC_U32_TERMINAL) {
152 #ifdef CONFIG_NET_CLS_IND
153 if (!tcf_match_indev(skb, n->indev)) {
158 #ifdef CONFIG_CLS_U32_PERF
161 r = tcf_exts_exec(skb, &n->exts, res);
174 if (sdepth >= TC_U32_MAXDEPTH)
176 stack[sdepth].knode = n;
177 stack[sdepth].ptr = ptr;
183 sel = ht->divisor&u32_hash_fold(*(__be32*)(ptr+n->sel.hoff), &n->sel,n->fshift);
185 if (!(n->sel.flags&(TC_U32_VAROFFSET|TC_U32_OFFSET|TC_U32_EAT)))
188 if (n->sel.flags&(TC_U32_OFFSET|TC_U32_VAROFFSET)) {
189 off2 = n->sel.off + 3;
190 if (n->sel.flags&TC_U32_VAROFFSET)
191 off2 += ntohs(n->sel.offmask & *(__be16*)(ptr+n->sel.offoff)) >>n->sel.offshift;
194 if (n->sel.flags&TC_U32_EAT) {
199 if (ptr < skb_tail_pointer(skb))
205 n = stack[sdepth].knode;
207 ptr = stack[sdepth].ptr;
214 printk("cls_u32: dead loop\n");
218 static __inline__ struct tc_u_hnode *
219 u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
221 struct tc_u_hnode *ht;
223 for (ht = tp_c->hlist; ht; ht = ht->next)
224 if (ht->handle == handle)
230 static __inline__ struct tc_u_knode *
231 u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
234 struct tc_u_knode *n = NULL;
236 sel = TC_U32_HASH(handle);
237 if (sel > ht->divisor)
240 for (n = ht->ht[sel]; n; n = n->next)
241 if (n->handle == handle)
248 static unsigned long u32_get(struct tcf_proto *tp, u32 handle)
250 struct tc_u_hnode *ht;
251 struct tc_u_common *tp_c = tp->data;
253 if (TC_U32_HTID(handle) == TC_U32_ROOT)
256 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
261 if (TC_U32_KEY(handle) == 0)
262 return (unsigned long)ht;
264 return (unsigned long)u32_lookup_key(ht, handle);
267 static void u32_put(struct tcf_proto *tp, unsigned long f)
271 static u32 gen_new_htid(struct tc_u_common *tp_c)
276 if (++tp_c->hgenerator == 0x7FF)
277 tp_c->hgenerator = 1;
278 } while (--i>0 && u32_lookup_ht(tp_c, (tp_c->hgenerator|0x800)<<20));
280 return i > 0 ? (tp_c->hgenerator|0x800)<<20 : 0;
283 static int u32_init(struct tcf_proto *tp)
285 struct tc_u_hnode *root_ht;
286 struct tc_u_common *tp_c;
288 tp_c = tp->q->u32_node;
290 root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL);
294 root_ht->divisor = 0;
296 root_ht->handle = tp_c ? gen_new_htid(tp_c) : 0x80000000;
297 root_ht->prio = tp->prio;
300 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
306 tp->q->u32_node = tp_c;
310 root_ht->next = tp_c->hlist;
311 tp_c->hlist = root_ht;
312 root_ht->tp_c = tp_c;
319 static int u32_destroy_key(struct tcf_proto *tp, struct tc_u_knode *n)
321 tcf_unbind_filter(tp, &n->res);
322 tcf_exts_destroy(tp, &n->exts);
324 n->ht_down->refcnt--;
325 #ifdef CONFIG_CLS_U32_PERF
332 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode* key)
334 struct tc_u_knode **kp;
335 struct tc_u_hnode *ht = key->ht_up;
338 for (kp = &ht->ht[TC_U32_HASH(key->handle)]; *kp; kp = &(*kp)->next) {
344 u32_destroy_key(tp, key);
353 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
355 struct tc_u_knode *n;
358 for (h=0; h<=ht->divisor; h++) {
359 while ((n = ht->ht[h]) != NULL) {
362 u32_destroy_key(tp, n);
367 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
369 struct tc_u_common *tp_c = tp->data;
370 struct tc_u_hnode **hn;
374 u32_clear_hnode(tp, ht);
376 for (hn = &tp_c->hlist; *hn; hn = &(*hn)->next) {
388 static void u32_destroy(struct tcf_proto *tp)
390 struct tc_u_common *tp_c = tp->data;
391 struct tc_u_hnode *root_ht = tp->root;
393 WARN_ON(root_ht == NULL);
395 if (root_ht && --root_ht->refcnt == 0)
396 u32_destroy_hnode(tp, root_ht);
398 if (--tp_c->refcnt == 0) {
399 struct tc_u_hnode *ht;
401 tp->q->u32_node = NULL;
403 for (ht = tp_c->hlist; ht; ht = ht->next) {
405 u32_clear_hnode(tp, ht);
408 while ((ht = tp_c->hlist) != NULL) {
409 tp_c->hlist = ht->next;
411 WARN_ON(ht->refcnt != 0);
422 static int u32_delete(struct tcf_proto *tp, unsigned long arg)
424 struct tc_u_hnode *ht = (struct tc_u_hnode*)arg;
429 if (TC_U32_KEY(ht->handle))
430 return u32_delete_key(tp, (struct tc_u_knode*)ht);
435 if (ht->refcnt == 1) {
437 u32_destroy_hnode(tp, ht);
445 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 handle)
447 struct tc_u_knode *n;
450 for (n=ht->ht[TC_U32_HASH(handle)]; n; n = n->next)
451 if (i < TC_U32_NODE(n->handle))
452 i = TC_U32_NODE(n->handle);
455 return handle|(i>0xFFF ? 0xFFF : i);
458 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
459 [TCA_U32_CLASSID] = { .type = NLA_U32 },
460 [TCA_U32_HASH] = { .type = NLA_U32 },
461 [TCA_U32_LINK] = { .type = NLA_U32 },
462 [TCA_U32_DIVISOR] = { .type = NLA_U32 },
463 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) },
464 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ },
465 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) },
468 static int u32_set_parms(struct tcf_proto *tp, unsigned long base,
469 struct tc_u_hnode *ht,
470 struct tc_u_knode *n, struct nlattr **tb,
476 err = tcf_exts_validate(tp, tb, est, &e, &u32_ext_map);
481 if (tb[TCA_U32_LINK]) {
482 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
483 struct tc_u_hnode *ht_down = NULL, *ht_old;
485 if (TC_U32_KEY(handle))
489 ht_down = u32_lookup_ht(ht->tp_c, handle);
498 n->ht_down = ht_down;
504 if (tb[TCA_U32_CLASSID]) {
505 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
506 tcf_bind_filter(tp, &n->res, base);
509 #ifdef CONFIG_NET_CLS_IND
510 if (tb[TCA_U32_INDEV]) {
511 err = tcf_change_indev(tp, n->indev, tb[TCA_U32_INDEV]);
516 tcf_exts_change(tp, &n->exts, &e);
520 tcf_exts_destroy(tp, &e);
524 static int u32_change(struct tcf_proto *tp, unsigned long base, u32 handle,
528 struct tc_u_common *tp_c = tp->data;
529 struct tc_u_hnode *ht;
530 struct tc_u_knode *n;
531 struct tc_u32_sel *s;
532 struct nlattr *opt = tca[TCA_OPTIONS];
533 struct nlattr *tb[TCA_U32_MAX + 1];
538 return handle ? -EINVAL : 0;
540 err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy);
544 if ((n = (struct tc_u_knode*)*arg) != NULL) {
545 if (TC_U32_KEY(n->handle) == 0)
548 return u32_set_parms(tp, base, n->ht_up, n, tb, tca[TCA_RATE]);
551 if (tb[TCA_U32_DIVISOR]) {
552 unsigned divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
554 if (--divisor > 0x100)
556 if (TC_U32_KEY(handle))
559 handle = gen_new_htid(tp->data);
563 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void*), GFP_KERNEL);
568 ht->divisor = divisor;
571 ht->next = tp_c->hlist;
573 *arg = (unsigned long)ht;
577 if (tb[TCA_U32_HASH]) {
578 htid = nla_get_u32(tb[TCA_U32_HASH]);
579 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
583 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
592 if (ht->divisor < TC_U32_HASH(htid))
596 if (TC_U32_HTID(handle) && TC_U32_HTID(handle^htid))
598 handle = htid | TC_U32_NODE(handle);
600 handle = gen_new_kid(ht, htid);
602 if (tb[TCA_U32_SEL] == NULL)
605 s = nla_data(tb[TCA_U32_SEL]);
607 n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL);
611 #ifdef CONFIG_CLS_U32_PERF
612 n->pf = kzalloc(sizeof(struct tc_u32_pcnt) + s->nkeys*sizeof(u64), GFP_KERNEL);
619 memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
622 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
624 #ifdef CONFIG_CLS_U32_MARK
625 if (tb[TCA_U32_MARK]) {
626 struct tc_u32_mark *mark;
628 mark = nla_data(tb[TCA_U32_MARK]);
629 memcpy(&n->mark, mark, sizeof(struct tc_u32_mark));
634 err = u32_set_parms(tp, base, ht, n, tb, tca[TCA_RATE]);
636 struct tc_u_knode **ins;
637 for (ins = &ht->ht[TC_U32_HASH(handle)]; *ins; ins = &(*ins)->next)
638 if (TC_U32_NODE(handle) < TC_U32_NODE((*ins)->handle))
646 *arg = (unsigned long)n;
649 #ifdef CONFIG_CLS_U32_PERF
656 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg)
658 struct tc_u_common *tp_c = tp->data;
659 struct tc_u_hnode *ht;
660 struct tc_u_knode *n;
666 for (ht = tp_c->hlist; ht; ht = ht->next) {
667 if (ht->prio != tp->prio)
669 if (arg->count >= arg->skip) {
670 if (arg->fn(tp, (unsigned long)ht, arg) < 0) {
676 for (h = 0; h <= ht->divisor; h++) {
677 for (n = ht->ht[h]; n; n = n->next) {
678 if (arg->count < arg->skip) {
682 if (arg->fn(tp, (unsigned long)n, arg) < 0) {
692 static int u32_dump(struct tcf_proto *tp, unsigned long fh,
693 struct sk_buff *skb, struct tcmsg *t)
695 struct tc_u_knode *n = (struct tc_u_knode*)fh;
701 t->tcm_handle = n->handle;
703 nest = nla_nest_start(skb, TCA_OPTIONS);
705 goto nla_put_failure;
707 if (TC_U32_KEY(n->handle) == 0) {
708 struct tc_u_hnode *ht = (struct tc_u_hnode*)fh;
709 u32 divisor = ht->divisor+1;
710 NLA_PUT_U32(skb, TCA_U32_DIVISOR, divisor);
712 NLA_PUT(skb, TCA_U32_SEL,
713 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
716 u32 htid = n->handle & 0xFFFFF000;
717 NLA_PUT_U32(skb, TCA_U32_HASH, htid);
720 NLA_PUT_U32(skb, TCA_U32_CLASSID, n->res.classid);
722 NLA_PUT_U32(skb, TCA_U32_LINK, n->ht_down->handle);
724 #ifdef CONFIG_CLS_U32_MARK
725 if (n->mark.val || n->mark.mask)
726 NLA_PUT(skb, TCA_U32_MARK, sizeof(n->mark), &n->mark);
729 if (tcf_exts_dump(skb, &n->exts, &u32_ext_map) < 0)
730 goto nla_put_failure;
732 #ifdef CONFIG_NET_CLS_IND
734 NLA_PUT_STRING(skb, TCA_U32_INDEV, n->indev);
736 #ifdef CONFIG_CLS_U32_PERF
737 NLA_PUT(skb, TCA_U32_PCNT,
738 sizeof(struct tc_u32_pcnt) + n->sel.nkeys*sizeof(u64),
743 nla_nest_end(skb, nest);
745 if (TC_U32_KEY(n->handle))
746 if (tcf_exts_dump_stats(skb, &n->exts, &u32_ext_map) < 0)
747 goto nla_put_failure;
751 nla_nest_cancel(skb, nest);
755 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
757 .classify = u32_classify,
759 .destroy = u32_destroy,
762 .change = u32_change,
763 .delete = u32_delete,
766 .owner = THIS_MODULE,
769 static int __init init_u32(void)
771 printk("u32 classifier\n");
772 #ifdef CONFIG_CLS_U32_PERF
773 printk(" Performance counters on\n");
775 #ifdef CONFIG_NET_CLS_IND
776 printk(" input device check on \n");
778 #ifdef CONFIG_NET_CLS_ACT
779 printk(" Actions configured \n");
781 return register_tcf_proto_ops(&cls_u32_ops);
784 static void __exit exit_u32(void)
786 unregister_tcf_proto_ops(&cls_u32_ops);
789 module_init(init_u32)
790 module_exit(exit_u32)
791 MODULE_LICENSE("GPL");