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 <linux/bitmap.h>
42 #include <net/netlink.h>
43 #include <net/act_api.h>
44 #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
;
67 struct tc_u_hnode
*next
;
70 struct tc_u_common
*tp_c
;
73 struct tc_u_knode
*ht
[1];
77 struct tc_u_hnode
*hlist
;
83 static inline unsigned int u32_hash_fold(__be32 key
,
84 const struct tc_u32_sel
*sel
,
87 unsigned int h
= ntohl(key
& sel
->hmask
) >> fshift
;
92 static int u32_classify(struct sk_buff
*skb
, const struct tcf_proto
*tp
, struct tcf_result
*res
)
95 struct tc_u_knode
*knode
;
97 } stack
[TC_U32_MAXDEPTH
];
99 struct tc_u_hnode
*ht
= tp
->root
;
100 unsigned int off
= skb_network_offset(skb
);
101 struct tc_u_knode
*n
;
105 #ifdef CONFIG_CLS_U32_PERF
115 struct tc_u32_key
*key
= n
->sel
.keys
;
117 #ifdef CONFIG_CLS_U32_PERF
122 #ifdef CONFIG_CLS_U32_MARK
123 if ((skb
->mark
& n
->mark
.mask
) != n
->mark
.val
) {
131 for (i
= n
->sel
.nkeys
; i
> 0; i
--, key
++) {
132 int toff
= off
+ key
->off
+ (off2
& key
->offmask
);
135 if (skb_headroom(skb
) + toff
> INT_MAX
)
138 data
= skb_header_pointer(skb
, toff
, 4, &hdata
);
141 if ((*data
^ key
->val
) & key
->mask
) {
145 #ifdef CONFIG_CLS_U32_PERF
146 n
->pf
->kcnts
[j
] += 1;
150 if (n
->ht_down
== NULL
) {
152 if (n
->sel
.flags
& TC_U32_TERMINAL
) {
155 #ifdef CONFIG_NET_CLS_IND
156 if (!tcf_match_indev(skb
, n
->ifindex
)) {
161 #ifdef CONFIG_CLS_U32_PERF
164 r
= tcf_exts_exec(skb
, &n
->exts
, res
);
177 if (sdepth
>= TC_U32_MAXDEPTH
)
179 stack
[sdepth
].knode
= n
;
180 stack
[sdepth
].off
= off
;
188 data
= skb_header_pointer(skb
, off
+ n
->sel
.hoff
, 4,
192 sel
= ht
->divisor
& u32_hash_fold(*data
, &n
->sel
,
195 if (!(n
->sel
.flags
& (TC_U32_VAROFFSET
| TC_U32_OFFSET
| TC_U32_EAT
)))
198 if (n
->sel
.flags
& (TC_U32_OFFSET
| TC_U32_VAROFFSET
)) {
199 off2
= n
->sel
.off
+ 3;
200 if (n
->sel
.flags
& TC_U32_VAROFFSET
) {
203 data
= skb_header_pointer(skb
,
208 off2
+= ntohs(n
->sel
.offmask
& *data
) >>
213 if (n
->sel
.flags
& TC_U32_EAT
) {
224 n
= stack
[sdepth
].knode
;
226 off
= stack
[sdepth
].off
;
233 net_warn_ratelimited("cls_u32: dead loop\n");
237 static struct tc_u_hnode
*
238 u32_lookup_ht(struct tc_u_common
*tp_c
, u32 handle
)
240 struct tc_u_hnode
*ht
;
242 for (ht
= tp_c
->hlist
; ht
; ht
= ht
->next
)
243 if (ht
->handle
== handle
)
249 static struct tc_u_knode
*
250 u32_lookup_key(struct tc_u_hnode
*ht
, u32 handle
)
253 struct tc_u_knode
*n
= NULL
;
255 sel
= TC_U32_HASH(handle
);
256 if (sel
> ht
->divisor
)
259 for (n
= ht
->ht
[sel
]; n
; n
= n
->next
)
260 if (n
->handle
== handle
)
267 static unsigned long u32_get(struct tcf_proto
*tp
, u32 handle
)
269 struct tc_u_hnode
*ht
;
270 struct tc_u_common
*tp_c
= tp
->data
;
272 if (TC_U32_HTID(handle
) == TC_U32_ROOT
)
275 ht
= u32_lookup_ht(tp_c
, TC_U32_HTID(handle
));
280 if (TC_U32_KEY(handle
) == 0)
281 return (unsigned long)ht
;
283 return (unsigned long)u32_lookup_key(ht
, handle
);
286 static void u32_put(struct tcf_proto
*tp
, unsigned long f
)
290 static u32
gen_new_htid(struct tc_u_common
*tp_c
)
295 if (++tp_c
->hgenerator
== 0x7FF)
296 tp_c
->hgenerator
= 1;
297 } while (--i
> 0 && u32_lookup_ht(tp_c
, (tp_c
->hgenerator
|0x800)<<20));
299 return i
> 0 ? (tp_c
->hgenerator
|0x800)<<20 : 0;
302 static int u32_init(struct tcf_proto
*tp
)
304 struct tc_u_hnode
*root_ht
;
305 struct tc_u_common
*tp_c
;
307 tp_c
= tp
->q
->u32_node
;
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
->q
->u32_node
= tp_c
;
329 root_ht
->next
= tp_c
->hlist
;
330 tp_c
->hlist
= root_ht
;
331 root_ht
->tp_c
= tp_c
;
338 static int u32_destroy_key(struct tcf_proto
*tp
, struct tc_u_knode
*n
)
340 tcf_unbind_filter(tp
, &n
->res
);
341 tcf_exts_destroy(tp
, &n
->exts
);
343 n
->ht_down
->refcnt
--;
344 #ifdef CONFIG_CLS_U32_PERF
351 static int u32_delete_key(struct tcf_proto
*tp
, struct tc_u_knode
*key
)
353 struct tc_u_knode
**kp
;
354 struct tc_u_hnode
*ht
= key
->ht_up
;
357 for (kp
= &ht
->ht
[TC_U32_HASH(key
->handle
)]; *kp
; kp
= &(*kp
)->next
) {
363 u32_destroy_key(tp
, key
);
372 static void u32_clear_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
)
374 struct tc_u_knode
*n
;
377 for (h
= 0; h
<= ht
->divisor
; h
++) {
378 while ((n
= ht
->ht
[h
]) != NULL
) {
381 u32_destroy_key(tp
, n
);
386 static int u32_destroy_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
)
388 struct tc_u_common
*tp_c
= tp
->data
;
389 struct tc_u_hnode
**hn
;
393 u32_clear_hnode(tp
, ht
);
395 for (hn
= &tp_c
->hlist
; *hn
; hn
= &(*hn
)->next
) {
407 static void u32_destroy(struct tcf_proto
*tp
)
409 struct tc_u_common
*tp_c
= tp
->data
;
410 struct tc_u_hnode
*root_ht
= tp
->root
;
412 WARN_ON(root_ht
== NULL
);
414 if (root_ht
&& --root_ht
->refcnt
== 0)
415 u32_destroy_hnode(tp
, root_ht
);
417 if (--tp_c
->refcnt
== 0) {
418 struct tc_u_hnode
*ht
;
420 tp
->q
->u32_node
= NULL
;
422 for (ht
= tp_c
->hlist
; ht
; ht
= ht
->next
) {
424 u32_clear_hnode(tp
, ht
);
427 while ((ht
= tp_c
->hlist
) != NULL
) {
428 tp_c
->hlist
= ht
->next
;
430 WARN_ON(ht
->refcnt
!= 0);
441 static int u32_delete(struct tcf_proto
*tp
, unsigned long arg
)
443 struct tc_u_hnode
*ht
= (struct tc_u_hnode
*)arg
;
448 if (TC_U32_KEY(ht
->handle
))
449 return u32_delete_key(tp
, (struct tc_u_knode
*)ht
);
454 if (ht
->refcnt
== 1) {
456 u32_destroy_hnode(tp
, ht
);
464 #define NR_U32_NODE (1<<12)
465 static u32
gen_new_kid(struct tc_u_hnode
*ht
, u32 handle
)
467 struct tc_u_knode
*n
;
469 unsigned long *bitmap
= kzalloc(BITS_TO_LONGS(NR_U32_NODE
) * sizeof(unsigned long),
472 return handle
| 0xFFF;
474 for (n
= ht
->ht
[TC_U32_HASH(handle
)]; n
; n
= n
->next
)
475 set_bit(TC_U32_NODE(n
->handle
), bitmap
);
477 i
= find_next_zero_bit(bitmap
, NR_U32_NODE
, 0x800);
478 if (i
>= NR_U32_NODE
)
479 i
= find_next_zero_bit(bitmap
, NR_U32_NODE
, 1);
482 return handle
| (i
>= NR_U32_NODE
? 0xFFF : i
);
485 static const struct nla_policy u32_policy
[TCA_U32_MAX
+ 1] = {
486 [TCA_U32_CLASSID
] = { .type
= NLA_U32
},
487 [TCA_U32_HASH
] = { .type
= NLA_U32
},
488 [TCA_U32_LINK
] = { .type
= NLA_U32
},
489 [TCA_U32_DIVISOR
] = { .type
= NLA_U32
},
490 [TCA_U32_SEL
] = { .len
= sizeof(struct tc_u32_sel
) },
491 [TCA_U32_INDEV
] = { .type
= NLA_STRING
, .len
= IFNAMSIZ
},
492 [TCA_U32_MARK
] = { .len
= sizeof(struct tc_u32_mark
) },
495 static int u32_set_parms(struct net
*net
, struct tcf_proto
*tp
,
496 unsigned long base
, struct tc_u_hnode
*ht
,
497 struct tc_u_knode
*n
, struct nlattr
**tb
,
498 struct nlattr
*est
, bool ovr
)
503 tcf_exts_init(&e
, TCA_U32_ACT
, TCA_U32_POLICE
);
504 err
= tcf_exts_validate(net
, tp
, tb
, est
, &e
, ovr
);
509 if (tb
[TCA_U32_LINK
]) {
510 u32 handle
= nla_get_u32(tb
[TCA_U32_LINK
]);
511 struct tc_u_hnode
*ht_down
= NULL
, *ht_old
;
513 if (TC_U32_KEY(handle
))
517 ht_down
= u32_lookup_ht(ht
->tp_c
, handle
);
526 n
->ht_down
= ht_down
;
532 if (tb
[TCA_U32_CLASSID
]) {
533 n
->res
.classid
= nla_get_u32(tb
[TCA_U32_CLASSID
]);
534 tcf_bind_filter(tp
, &n
->res
, base
);
537 #ifdef CONFIG_NET_CLS_IND
538 if (tb
[TCA_U32_INDEV
]) {
540 ret
= tcf_change_indev(net
, tb
[TCA_U32_INDEV
]);
546 tcf_exts_change(tp
, &n
->exts
, &e
);
550 tcf_exts_destroy(tp
, &e
);
554 static int u32_change(struct net
*net
, struct sk_buff
*in_skb
,
555 struct tcf_proto
*tp
, unsigned long base
, u32 handle
,
557 unsigned long *arg
, bool ovr
)
559 struct tc_u_common
*tp_c
= tp
->data
;
560 struct tc_u_hnode
*ht
;
561 struct tc_u_knode
*n
;
562 struct tc_u32_sel
*s
;
563 struct nlattr
*opt
= tca
[TCA_OPTIONS
];
564 struct nlattr
*tb
[TCA_U32_MAX
+ 1];
569 return handle
? -EINVAL
: 0;
571 err
= nla_parse_nested(tb
, TCA_U32_MAX
, opt
, u32_policy
);
575 n
= (struct tc_u_knode
*)*arg
;
577 if (TC_U32_KEY(n
->handle
) == 0)
580 return u32_set_parms(net
, tp
, base
, n
->ht_up
, n
, tb
,
584 if (tb
[TCA_U32_DIVISOR
]) {
585 unsigned int divisor
= nla_get_u32(tb
[TCA_U32_DIVISOR
]);
587 if (--divisor
> 0x100)
589 if (TC_U32_KEY(handle
))
592 handle
= gen_new_htid(tp
->data
);
596 ht
= kzalloc(sizeof(*ht
) + divisor
*sizeof(void *), GFP_KERNEL
);
601 ht
->divisor
= divisor
;
604 ht
->next
= tp_c
->hlist
;
606 *arg
= (unsigned long)ht
;
610 if (tb
[TCA_U32_HASH
]) {
611 htid
= nla_get_u32(tb
[TCA_U32_HASH
]);
612 if (TC_U32_HTID(htid
) == TC_U32_ROOT
) {
616 ht
= u32_lookup_ht(tp
->data
, TC_U32_HTID(htid
));
625 if (ht
->divisor
< TC_U32_HASH(htid
))
629 if (TC_U32_HTID(handle
) && TC_U32_HTID(handle
^htid
))
631 handle
= htid
| TC_U32_NODE(handle
);
633 handle
= gen_new_kid(ht
, htid
);
635 if (tb
[TCA_U32_SEL
] == NULL
)
638 s
= nla_data(tb
[TCA_U32_SEL
]);
640 n
= kzalloc(sizeof(*n
) + s
->nkeys
*sizeof(struct tc_u32_key
), GFP_KERNEL
);
644 #ifdef CONFIG_CLS_U32_PERF
645 n
->pf
= kzalloc(sizeof(struct tc_u32_pcnt
) + s
->nkeys
*sizeof(u64
), GFP_KERNEL
);
652 memcpy(&n
->sel
, s
, sizeof(*s
) + s
->nkeys
*sizeof(struct tc_u32_key
));
655 n
->fshift
= s
->hmask
? ffs(ntohl(s
->hmask
)) - 1 : 0;
656 tcf_exts_init(&n
->exts
, TCA_U32_ACT
, TCA_U32_POLICE
);
658 #ifdef CONFIG_CLS_U32_MARK
659 if (tb
[TCA_U32_MARK
]) {
660 struct tc_u32_mark
*mark
;
662 mark
= nla_data(tb
[TCA_U32_MARK
]);
663 memcpy(&n
->mark
, mark
, sizeof(struct tc_u32_mark
));
668 err
= u32_set_parms(net
, tp
, base
, ht
, n
, tb
, tca
[TCA_RATE
], ovr
);
670 struct tc_u_knode
**ins
;
671 for (ins
= &ht
->ht
[TC_U32_HASH(handle
)]; *ins
; ins
= &(*ins
)->next
)
672 if (TC_U32_NODE(handle
) < TC_U32_NODE((*ins
)->handle
))
680 *arg
= (unsigned long)n
;
683 #ifdef CONFIG_CLS_U32_PERF
690 static void u32_walk(struct tcf_proto
*tp
, struct tcf_walker
*arg
)
692 struct tc_u_common
*tp_c
= tp
->data
;
693 struct tc_u_hnode
*ht
;
694 struct tc_u_knode
*n
;
700 for (ht
= tp_c
->hlist
; ht
; ht
= ht
->next
) {
701 if (ht
->prio
!= tp
->prio
)
703 if (arg
->count
>= arg
->skip
) {
704 if (arg
->fn(tp
, (unsigned long)ht
, arg
) < 0) {
710 for (h
= 0; h
<= ht
->divisor
; h
++) {
711 for (n
= ht
->ht
[h
]; n
; n
= n
->next
) {
712 if (arg
->count
< arg
->skip
) {
716 if (arg
->fn(tp
, (unsigned long)n
, arg
) < 0) {
726 static int u32_dump(struct net
*net
, struct tcf_proto
*tp
, unsigned long fh
,
727 struct sk_buff
*skb
, struct tcmsg
*t
)
729 struct tc_u_knode
*n
= (struct tc_u_knode
*)fh
;
735 t
->tcm_handle
= n
->handle
;
737 nest
= nla_nest_start(skb
, TCA_OPTIONS
);
739 goto nla_put_failure
;
741 if (TC_U32_KEY(n
->handle
) == 0) {
742 struct tc_u_hnode
*ht
= (struct tc_u_hnode
*)fh
;
743 u32 divisor
= ht
->divisor
+ 1;
745 if (nla_put_u32(skb
, TCA_U32_DIVISOR
, divisor
))
746 goto nla_put_failure
;
748 if (nla_put(skb
, TCA_U32_SEL
,
749 sizeof(n
->sel
) + n
->sel
.nkeys
*sizeof(struct tc_u32_key
),
751 goto nla_put_failure
;
753 u32 htid
= n
->handle
& 0xFFFFF000;
754 if (nla_put_u32(skb
, TCA_U32_HASH
, htid
))
755 goto nla_put_failure
;
757 if (n
->res
.classid
&&
758 nla_put_u32(skb
, TCA_U32_CLASSID
, n
->res
.classid
))
759 goto nla_put_failure
;
761 nla_put_u32(skb
, TCA_U32_LINK
, n
->ht_down
->handle
))
762 goto nla_put_failure
;
764 #ifdef CONFIG_CLS_U32_MARK
765 if ((n
->mark
.val
|| n
->mark
.mask
) &&
766 nla_put(skb
, TCA_U32_MARK
, sizeof(n
->mark
), &n
->mark
))
767 goto nla_put_failure
;
770 if (tcf_exts_dump(skb
, &n
->exts
) < 0)
771 goto nla_put_failure
;
773 #ifdef CONFIG_NET_CLS_IND
775 struct net_device
*dev
;
776 dev
= __dev_get_by_index(net
, n
->ifindex
);
777 if (dev
&& nla_put_string(skb
, TCA_U32_INDEV
, dev
->name
))
778 goto nla_put_failure
;
781 #ifdef CONFIG_CLS_U32_PERF
782 if (nla_put(skb
, TCA_U32_PCNT
,
783 sizeof(struct tc_u32_pcnt
) + n
->sel
.nkeys
*sizeof(u64
),
785 goto nla_put_failure
;
789 nla_nest_end(skb
, nest
);
791 if (TC_U32_KEY(n
->handle
))
792 if (tcf_exts_dump_stats(skb
, &n
->exts
) < 0)
793 goto nla_put_failure
;
797 nla_nest_cancel(skb
, nest
);
801 static struct tcf_proto_ops cls_u32_ops __read_mostly
= {
803 .classify
= u32_classify
,
805 .destroy
= u32_destroy
,
808 .change
= u32_change
,
809 .delete = u32_delete
,
812 .owner
= THIS_MODULE
,
815 static int __init
init_u32(void)
817 pr_info("u32 classifier\n");
818 #ifdef CONFIG_CLS_U32_PERF
819 pr_info(" Performance counters on\n");
821 #ifdef CONFIG_NET_CLS_IND
822 pr_info(" input device check on\n");
824 #ifdef CONFIG_NET_CLS_ACT
825 pr_info(" Actions configured\n");
827 return register_tcf_proto_ops(&cls_u32_ops
);
830 static void __exit
exit_u32(void)
832 unregister_tcf_proto_ops(&cls_u32_ops
);
835 module_init(init_u32
)
836 module_exit(exit_u32
)
837 MODULE_LICENSE("GPL");