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
++) {
137 int toff
= off
+ key
->off
+ (off2
& key
->offmask
);
140 if (skb_headroom(skb
) + toff
> INT_MAX
)
143 data
= skb_header_pointer(skb
, toff
, 4, &_data
);
146 if ((*data
^ key
->val
) & key
->mask
) {
150 #ifdef CONFIG_CLS_U32_PERF
155 if (n
->ht_down
== NULL
) {
157 if (n
->sel
.flags
&TC_U32_TERMINAL
) {
160 #ifdef CONFIG_NET_CLS_IND
161 if (!tcf_match_indev(skb
, n
->indev
)) {
166 #ifdef CONFIG_CLS_U32_PERF
169 r
= tcf_exts_exec(skb
, &n
->exts
, res
);
182 if (sdepth
>= TC_U32_MAXDEPTH
)
184 stack
[sdepth
].knode
= n
;
185 stack
[sdepth
].off
= off
;
193 data
= skb_header_pointer(skb
, off
+ n
->sel
.hoff
, 4,
197 sel
= ht
->divisor
& u32_hash_fold(*data
, &n
->sel
,
200 if (!(n
->sel
.flags
&(TC_U32_VAROFFSET
|TC_U32_OFFSET
|TC_U32_EAT
)))
203 if (n
->sel
.flags
&(TC_U32_OFFSET
|TC_U32_VAROFFSET
)) {
204 off2
= n
->sel
.off
+ 3;
205 if (n
->sel
.flags
& TC_U32_VAROFFSET
) {
208 data
= skb_header_pointer(skb
,
213 off2
+= ntohs(n
->sel
.offmask
& *data
) >>
218 if (n
->sel
.flags
&TC_U32_EAT
) {
229 n
= stack
[sdepth
].knode
;
231 off
= stack
[sdepth
].off
;
239 printk(KERN_WARNING
"cls_u32: dead loop\n");
243 static __inline__
struct tc_u_hnode
*
244 u32_lookup_ht(struct tc_u_common
*tp_c
, u32 handle
)
246 struct tc_u_hnode
*ht
;
248 for (ht
= tp_c
->hlist
; ht
; ht
= ht
->next
)
249 if (ht
->handle
== handle
)
255 static __inline__
struct tc_u_knode
*
256 u32_lookup_key(struct tc_u_hnode
*ht
, u32 handle
)
259 struct tc_u_knode
*n
= NULL
;
261 sel
= TC_U32_HASH(handle
);
262 if (sel
> ht
->divisor
)
265 for (n
= ht
->ht
[sel
]; n
; n
= n
->next
)
266 if (n
->handle
== handle
)
273 static unsigned long u32_get(struct tcf_proto
*tp
, u32 handle
)
275 struct tc_u_hnode
*ht
;
276 struct tc_u_common
*tp_c
= tp
->data
;
278 if (TC_U32_HTID(handle
) == TC_U32_ROOT
)
281 ht
= u32_lookup_ht(tp_c
, TC_U32_HTID(handle
));
286 if (TC_U32_KEY(handle
) == 0)
287 return (unsigned long)ht
;
289 return (unsigned long)u32_lookup_key(ht
, handle
);
292 static void u32_put(struct tcf_proto
*tp
, unsigned long f
)
296 static u32
gen_new_htid(struct tc_u_common
*tp_c
)
301 if (++tp_c
->hgenerator
== 0x7FF)
302 tp_c
->hgenerator
= 1;
303 } while (--i
>0 && u32_lookup_ht(tp_c
, (tp_c
->hgenerator
|0x800)<<20));
305 return i
> 0 ? (tp_c
->hgenerator
|0x800)<<20 : 0;
308 static int u32_init(struct tcf_proto
*tp
)
310 struct tc_u_hnode
*root_ht
;
311 struct tc_u_common
*tp_c
;
313 tp_c
= tp
->q
->u32_node
;
315 root_ht
= kzalloc(sizeof(*root_ht
), GFP_KERNEL
);
319 root_ht
->divisor
= 0;
321 root_ht
->handle
= tp_c
? gen_new_htid(tp_c
) : 0x80000000;
322 root_ht
->prio
= tp
->prio
;
325 tp_c
= kzalloc(sizeof(*tp_c
), GFP_KERNEL
);
331 tp
->q
->u32_node
= tp_c
;
335 root_ht
->next
= tp_c
->hlist
;
336 tp_c
->hlist
= root_ht
;
337 root_ht
->tp_c
= tp_c
;
344 static int u32_destroy_key(struct tcf_proto
*tp
, struct tc_u_knode
*n
)
346 tcf_unbind_filter(tp
, &n
->res
);
347 tcf_exts_destroy(tp
, &n
->exts
);
349 n
->ht_down
->refcnt
--;
350 #ifdef CONFIG_CLS_U32_PERF
357 static int u32_delete_key(struct tcf_proto
*tp
, struct tc_u_knode
* key
)
359 struct tc_u_knode
**kp
;
360 struct tc_u_hnode
*ht
= key
->ht_up
;
363 for (kp
= &ht
->ht
[TC_U32_HASH(key
->handle
)]; *kp
; kp
= &(*kp
)->next
) {
369 u32_destroy_key(tp
, key
);
378 static void u32_clear_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
)
380 struct tc_u_knode
*n
;
383 for (h
=0; h
<=ht
->divisor
; h
++) {
384 while ((n
= ht
->ht
[h
]) != NULL
) {
387 u32_destroy_key(tp
, n
);
392 static int u32_destroy_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
)
394 struct tc_u_common
*tp_c
= tp
->data
;
395 struct tc_u_hnode
**hn
;
399 u32_clear_hnode(tp
, ht
);
401 for (hn
= &tp_c
->hlist
; *hn
; hn
= &(*hn
)->next
) {
413 static void u32_destroy(struct tcf_proto
*tp
)
415 struct tc_u_common
*tp_c
= tp
->data
;
416 struct tc_u_hnode
*root_ht
= tp
->root
;
418 WARN_ON(root_ht
== NULL
);
420 if (root_ht
&& --root_ht
->refcnt
== 0)
421 u32_destroy_hnode(tp
, root_ht
);
423 if (--tp_c
->refcnt
== 0) {
424 struct tc_u_hnode
*ht
;
426 tp
->q
->u32_node
= NULL
;
428 for (ht
= tp_c
->hlist
; ht
; ht
= ht
->next
) {
430 u32_clear_hnode(tp
, ht
);
433 while ((ht
= tp_c
->hlist
) != NULL
) {
434 tp_c
->hlist
= ht
->next
;
436 WARN_ON(ht
->refcnt
!= 0);
447 static int u32_delete(struct tcf_proto
*tp
, unsigned long arg
)
449 struct tc_u_hnode
*ht
= (struct tc_u_hnode
*)arg
;
454 if (TC_U32_KEY(ht
->handle
))
455 return u32_delete_key(tp
, (struct tc_u_knode
*)ht
);
460 if (ht
->refcnt
== 1) {
462 u32_destroy_hnode(tp
, ht
);
470 static u32
gen_new_kid(struct tc_u_hnode
*ht
, u32 handle
)
472 struct tc_u_knode
*n
;
475 for (n
=ht
->ht
[TC_U32_HASH(handle
)]; n
; n
= n
->next
)
476 if (i
< TC_U32_NODE(n
->handle
))
477 i
= TC_U32_NODE(n
->handle
);
480 return handle
|(i
>0xFFF ? 0xFFF : i
);
483 static const struct nla_policy u32_policy
[TCA_U32_MAX
+ 1] = {
484 [TCA_U32_CLASSID
] = { .type
= NLA_U32
},
485 [TCA_U32_HASH
] = { .type
= NLA_U32
},
486 [TCA_U32_LINK
] = { .type
= NLA_U32
},
487 [TCA_U32_DIVISOR
] = { .type
= NLA_U32
},
488 [TCA_U32_SEL
] = { .len
= sizeof(struct tc_u32_sel
) },
489 [TCA_U32_INDEV
] = { .type
= NLA_STRING
, .len
= IFNAMSIZ
},
490 [TCA_U32_MARK
] = { .len
= sizeof(struct tc_u32_mark
) },
493 static int u32_set_parms(struct tcf_proto
*tp
, unsigned long base
,
494 struct tc_u_hnode
*ht
,
495 struct tc_u_knode
*n
, struct nlattr
**tb
,
501 err
= tcf_exts_validate(tp
, tb
, est
, &e
, &u32_ext_map
);
506 if (tb
[TCA_U32_LINK
]) {
507 u32 handle
= nla_get_u32(tb
[TCA_U32_LINK
]);
508 struct tc_u_hnode
*ht_down
= NULL
, *ht_old
;
510 if (TC_U32_KEY(handle
))
514 ht_down
= u32_lookup_ht(ht
->tp_c
, handle
);
523 n
->ht_down
= ht_down
;
529 if (tb
[TCA_U32_CLASSID
]) {
530 n
->res
.classid
= nla_get_u32(tb
[TCA_U32_CLASSID
]);
531 tcf_bind_filter(tp
, &n
->res
, base
);
534 #ifdef CONFIG_NET_CLS_IND
535 if (tb
[TCA_U32_INDEV
]) {
536 err
= tcf_change_indev(tp
, n
->indev
, tb
[TCA_U32_INDEV
]);
541 tcf_exts_change(tp
, &n
->exts
, &e
);
545 tcf_exts_destroy(tp
, &e
);
549 static int u32_change(struct tcf_proto
*tp
, unsigned long base
, u32 handle
,
553 struct tc_u_common
*tp_c
= tp
->data
;
554 struct tc_u_hnode
*ht
;
555 struct tc_u_knode
*n
;
556 struct tc_u32_sel
*s
;
557 struct nlattr
*opt
= tca
[TCA_OPTIONS
];
558 struct nlattr
*tb
[TCA_U32_MAX
+ 1];
563 return handle
? -EINVAL
: 0;
565 err
= nla_parse_nested(tb
, TCA_U32_MAX
, opt
, u32_policy
);
569 if ((n
= (struct tc_u_knode
*)*arg
) != NULL
) {
570 if (TC_U32_KEY(n
->handle
) == 0)
573 return u32_set_parms(tp
, base
, n
->ht_up
, n
, tb
, tca
[TCA_RATE
]);
576 if (tb
[TCA_U32_DIVISOR
]) {
577 unsigned divisor
= nla_get_u32(tb
[TCA_U32_DIVISOR
]);
579 if (--divisor
> 0x100)
581 if (TC_U32_KEY(handle
))
584 handle
= gen_new_htid(tp
->data
);
588 ht
= kzalloc(sizeof(*ht
) + divisor
*sizeof(void*), GFP_KERNEL
);
593 ht
->divisor
= divisor
;
596 ht
->next
= tp_c
->hlist
;
598 *arg
= (unsigned long)ht
;
602 if (tb
[TCA_U32_HASH
]) {
603 htid
= nla_get_u32(tb
[TCA_U32_HASH
]);
604 if (TC_U32_HTID(htid
) == TC_U32_ROOT
) {
608 ht
= u32_lookup_ht(tp
->data
, TC_U32_HTID(htid
));
617 if (ht
->divisor
< TC_U32_HASH(htid
))
621 if (TC_U32_HTID(handle
) && TC_U32_HTID(handle
^htid
))
623 handle
= htid
| TC_U32_NODE(handle
);
625 handle
= gen_new_kid(ht
, htid
);
627 if (tb
[TCA_U32_SEL
] == NULL
)
630 s
= nla_data(tb
[TCA_U32_SEL
]);
632 n
= kzalloc(sizeof(*n
) + s
->nkeys
*sizeof(struct tc_u32_key
), GFP_KERNEL
);
636 #ifdef CONFIG_CLS_U32_PERF
637 n
->pf
= kzalloc(sizeof(struct tc_u32_pcnt
) + s
->nkeys
*sizeof(u64
), GFP_KERNEL
);
644 memcpy(&n
->sel
, s
, sizeof(*s
) + s
->nkeys
*sizeof(struct tc_u32_key
));
647 n
->fshift
= s
->hmask
? ffs(ntohl(s
->hmask
)) - 1 : 0;
649 #ifdef CONFIG_CLS_U32_MARK
650 if (tb
[TCA_U32_MARK
]) {
651 struct tc_u32_mark
*mark
;
653 mark
= nla_data(tb
[TCA_U32_MARK
]);
654 memcpy(&n
->mark
, mark
, sizeof(struct tc_u32_mark
));
659 err
= u32_set_parms(tp
, base
, ht
, n
, tb
, tca
[TCA_RATE
]);
661 struct tc_u_knode
**ins
;
662 for (ins
= &ht
->ht
[TC_U32_HASH(handle
)]; *ins
; ins
= &(*ins
)->next
)
663 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
;
726 t
->tcm_handle
= n
->handle
;
728 nest
= nla_nest_start(skb
, TCA_OPTIONS
);
730 goto nla_put_failure
;
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 NLA_PUT_U32(skb
, TCA_U32_DIVISOR
, divisor
);
737 NLA_PUT(skb
, TCA_U32_SEL
,
738 sizeof(n
->sel
) + n
->sel
.nkeys
*sizeof(struct tc_u32_key
),
741 u32 htid
= n
->handle
& 0xFFFFF000;
742 NLA_PUT_U32(skb
, TCA_U32_HASH
, htid
);
745 NLA_PUT_U32(skb
, TCA_U32_CLASSID
, n
->res
.classid
);
747 NLA_PUT_U32(skb
, TCA_U32_LINK
, n
->ht_down
->handle
);
749 #ifdef CONFIG_CLS_U32_MARK
750 if (n
->mark
.val
|| n
->mark
.mask
)
751 NLA_PUT(skb
, TCA_U32_MARK
, sizeof(n
->mark
), &n
->mark
);
754 if (tcf_exts_dump(skb
, &n
->exts
, &u32_ext_map
) < 0)
755 goto nla_put_failure
;
757 #ifdef CONFIG_NET_CLS_IND
759 NLA_PUT_STRING(skb
, TCA_U32_INDEV
, n
->indev
);
761 #ifdef CONFIG_CLS_U32_PERF
762 NLA_PUT(skb
, TCA_U32_PCNT
,
763 sizeof(struct tc_u32_pcnt
) + n
->sel
.nkeys
*sizeof(u64
),
768 nla_nest_end(skb
, nest
);
770 if (TC_U32_KEY(n
->handle
))
771 if (tcf_exts_dump_stats(skb
, &n
->exts
, &u32_ext_map
) < 0)
772 goto nla_put_failure
;
776 nla_nest_cancel(skb
, nest
);
780 static struct tcf_proto_ops cls_u32_ops __read_mostly
= {
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 pr_info("u32 classifier\n");
797 #ifdef CONFIG_CLS_U32_PERF
798 pr_info(" Performance counters on\n");
800 #ifdef CONFIG_NET_CLS_IND
801 pr_info(" input device check on\n");
803 #ifdef CONFIG_NET_CLS_ACT
804 pr_info(" Actions configured\n");
806 return register_tcf_proto_ops(&cls_u32_ops
);
809 static void __exit
exit_u32(void)
811 unregister_tcf_proto_ops(&cls_u32_ops
);
814 module_init(init_u32
)
815 module_exit(exit_u32
)
816 MODULE_LICENSE("GPL");