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>
46 struct tc_u_knode
*next
;
48 struct tc_u_hnode
*ht_up
;
50 #ifdef CONFIG_NET_CLS_IND
54 struct tcf_result res
;
55 struct tc_u_hnode
*ht_down
;
56 #ifdef CONFIG_CLS_U32_PERF
57 struct tc_u32_pcnt
*pf
;
59 #ifdef CONFIG_CLS_U32_MARK
60 struct tc_u32_mark mark
;
62 struct tc_u32_sel sel
;
66 struct tc_u_hnode
*next
;
69 struct tc_u_common
*tp_c
;
72 struct tc_u_knode
*ht
[1];
76 struct tc_u_hnode
*hlist
;
82 static inline unsigned int u32_hash_fold(__be32 key
,
83 const struct tc_u32_sel
*sel
,
86 unsigned int h
= ntohl(key
& sel
->hmask
) >> fshift
;
91 static int u32_classify(struct sk_buff
*skb
, const struct tcf_proto
*tp
, struct tcf_result
*res
)
94 struct tc_u_knode
*knode
;
96 } stack
[TC_U32_MAXDEPTH
];
98 struct tc_u_hnode
*ht
= tp
->root
;
99 unsigned int off
= skb_network_offset(skb
);
100 struct tc_u_knode
*n
;
104 #ifdef CONFIG_CLS_U32_PERF
114 struct tc_u32_key
*key
= n
->sel
.keys
;
116 #ifdef CONFIG_CLS_U32_PERF
121 #ifdef CONFIG_CLS_U32_MARK
122 if ((skb
->mark
& n
->mark
.mask
) != n
->mark
.val
) {
130 for (i
= n
->sel
.nkeys
; i
> 0; i
--, key
++) {
131 int toff
= off
+ key
->off
+ (off2
& key
->offmask
);
134 if (skb_headroom(skb
) + toff
> INT_MAX
)
137 data
= skb_header_pointer(skb
, toff
, 4, &hdata
);
140 if ((*data
^ key
->val
) & key
->mask
) {
144 #ifdef CONFIG_CLS_U32_PERF
145 n
->pf
->kcnts
[j
] += 1;
149 if (n
->ht_down
== NULL
) {
151 if (n
->sel
.flags
& TC_U32_TERMINAL
) {
154 #ifdef CONFIG_NET_CLS_IND
155 if (!tcf_match_indev(skb
, n
->ifindex
)) {
160 #ifdef CONFIG_CLS_U32_PERF
163 r
= tcf_exts_exec(skb
, &n
->exts
, res
);
176 if (sdepth
>= TC_U32_MAXDEPTH
)
178 stack
[sdepth
].knode
= n
;
179 stack
[sdepth
].off
= off
;
187 data
= skb_header_pointer(skb
, off
+ n
->sel
.hoff
, 4,
191 sel
= ht
->divisor
& u32_hash_fold(*data
, &n
->sel
,
194 if (!(n
->sel
.flags
& (TC_U32_VAROFFSET
| TC_U32_OFFSET
| TC_U32_EAT
)))
197 if (n
->sel
.flags
& (TC_U32_OFFSET
| TC_U32_VAROFFSET
)) {
198 off2
= n
->sel
.off
+ 3;
199 if (n
->sel
.flags
& TC_U32_VAROFFSET
) {
202 data
= skb_header_pointer(skb
,
207 off2
+= ntohs(n
->sel
.offmask
& *data
) >>
212 if (n
->sel
.flags
& TC_U32_EAT
) {
223 n
= stack
[sdepth
].knode
;
225 off
= stack
[sdepth
].off
;
232 net_warn_ratelimited("cls_u32: dead loop\n");
236 static struct tc_u_hnode
*
237 u32_lookup_ht(struct tc_u_common
*tp_c
, u32 handle
)
239 struct tc_u_hnode
*ht
;
241 for (ht
= tp_c
->hlist
; ht
; ht
= ht
->next
)
242 if (ht
->handle
== handle
)
248 static struct tc_u_knode
*
249 u32_lookup_key(struct tc_u_hnode
*ht
, u32 handle
)
252 struct tc_u_knode
*n
= NULL
;
254 sel
= TC_U32_HASH(handle
);
255 if (sel
> ht
->divisor
)
258 for (n
= ht
->ht
[sel
]; n
; n
= n
->next
)
259 if (n
->handle
== handle
)
266 static unsigned long u32_get(struct tcf_proto
*tp
, u32 handle
)
268 struct tc_u_hnode
*ht
;
269 struct tc_u_common
*tp_c
= tp
->data
;
271 if (TC_U32_HTID(handle
) == TC_U32_ROOT
)
274 ht
= u32_lookup_ht(tp_c
, TC_U32_HTID(handle
));
279 if (TC_U32_KEY(handle
) == 0)
280 return (unsigned long)ht
;
282 return (unsigned long)u32_lookup_key(ht
, handle
);
285 static void u32_put(struct tcf_proto
*tp
, unsigned long f
)
289 static u32
gen_new_htid(struct tc_u_common
*tp_c
)
294 if (++tp_c
->hgenerator
== 0x7FF)
295 tp_c
->hgenerator
= 1;
296 } while (--i
> 0 && u32_lookup_ht(tp_c
, (tp_c
->hgenerator
|0x800)<<20));
298 return i
> 0 ? (tp_c
->hgenerator
|0x800)<<20 : 0;
301 static int u32_init(struct tcf_proto
*tp
)
303 struct tc_u_hnode
*root_ht
;
304 struct tc_u_common
*tp_c
;
306 tp_c
= tp
->q
->u32_node
;
308 root_ht
= kzalloc(sizeof(*root_ht
), GFP_KERNEL
);
312 root_ht
->divisor
= 0;
314 root_ht
->handle
= tp_c
? gen_new_htid(tp_c
) : 0x80000000;
315 root_ht
->prio
= tp
->prio
;
318 tp_c
= kzalloc(sizeof(*tp_c
), GFP_KERNEL
);
324 tp
->q
->u32_node
= tp_c
;
328 root_ht
->next
= tp_c
->hlist
;
329 tp_c
->hlist
= root_ht
;
330 root_ht
->tp_c
= tp_c
;
337 static int u32_destroy_key(struct tcf_proto
*tp
, struct tc_u_knode
*n
)
339 tcf_unbind_filter(tp
, &n
->res
);
340 tcf_exts_destroy(tp
, &n
->exts
);
342 n
->ht_down
->refcnt
--;
343 #ifdef CONFIG_CLS_U32_PERF
350 static int u32_delete_key(struct tcf_proto
*tp
, struct tc_u_knode
*key
)
352 struct tc_u_knode
**kp
;
353 struct tc_u_hnode
*ht
= key
->ht_up
;
356 for (kp
= &ht
->ht
[TC_U32_HASH(key
->handle
)]; *kp
; kp
= &(*kp
)->next
) {
362 u32_destroy_key(tp
, key
);
371 static void u32_clear_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
)
373 struct tc_u_knode
*n
;
376 for (h
= 0; h
<= ht
->divisor
; h
++) {
377 while ((n
= ht
->ht
[h
]) != NULL
) {
380 u32_destroy_key(tp
, n
);
385 static int u32_destroy_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
)
387 struct tc_u_common
*tp_c
= tp
->data
;
388 struct tc_u_hnode
**hn
;
392 u32_clear_hnode(tp
, ht
);
394 for (hn
= &tp_c
->hlist
; *hn
; hn
= &(*hn
)->next
) {
406 static void u32_destroy(struct tcf_proto
*tp
)
408 struct tc_u_common
*tp_c
= tp
->data
;
409 struct tc_u_hnode
*root_ht
= tp
->root
;
411 WARN_ON(root_ht
== NULL
);
413 if (root_ht
&& --root_ht
->refcnt
== 0)
414 u32_destroy_hnode(tp
, root_ht
);
416 if (--tp_c
->refcnt
== 0) {
417 struct tc_u_hnode
*ht
;
419 tp
->q
->u32_node
= NULL
;
421 for (ht
= tp_c
->hlist
; ht
; ht
= ht
->next
) {
423 u32_clear_hnode(tp
, ht
);
426 while ((ht
= tp_c
->hlist
) != NULL
) {
427 tp_c
->hlist
= ht
->next
;
429 WARN_ON(ht
->refcnt
!= 0);
440 static int u32_delete(struct tcf_proto
*tp
, unsigned long arg
)
442 struct tc_u_hnode
*ht
= (struct tc_u_hnode
*)arg
;
447 if (TC_U32_KEY(ht
->handle
))
448 return u32_delete_key(tp
, (struct tc_u_knode
*)ht
);
453 if (ht
->refcnt
== 1) {
455 u32_destroy_hnode(tp
, ht
);
463 static u32
gen_new_kid(struct tc_u_hnode
*ht
, u32 handle
)
465 struct tc_u_knode
*n
;
466 unsigned int i
= 0x7FF;
468 for (n
= ht
->ht
[TC_U32_HASH(handle
)]; n
; n
= n
->next
)
469 if (i
< TC_U32_NODE(n
->handle
))
470 i
= TC_U32_NODE(n
->handle
);
473 return handle
| (i
> 0xFFF ? 0xFFF : i
);
476 static const struct nla_policy u32_policy
[TCA_U32_MAX
+ 1] = {
477 [TCA_U32_CLASSID
] = { .type
= NLA_U32
},
478 [TCA_U32_HASH
] = { .type
= NLA_U32
},
479 [TCA_U32_LINK
] = { .type
= NLA_U32
},
480 [TCA_U32_DIVISOR
] = { .type
= NLA_U32
},
481 [TCA_U32_SEL
] = { .len
= sizeof(struct tc_u32_sel
) },
482 [TCA_U32_INDEV
] = { .type
= NLA_STRING
, .len
= IFNAMSIZ
},
483 [TCA_U32_MARK
] = { .len
= sizeof(struct tc_u32_mark
) },
486 static int u32_set_parms(struct net
*net
, struct tcf_proto
*tp
,
487 unsigned long base
, struct tc_u_hnode
*ht
,
488 struct tc_u_knode
*n
, struct nlattr
**tb
,
489 struct nlattr
*est
, bool ovr
)
494 tcf_exts_init(&e
, TCA_U32_ACT
, TCA_U32_POLICE
);
495 err
= tcf_exts_validate(net
, tp
, tb
, est
, &e
, ovr
);
500 if (tb
[TCA_U32_LINK
]) {
501 u32 handle
= nla_get_u32(tb
[TCA_U32_LINK
]);
502 struct tc_u_hnode
*ht_down
= NULL
, *ht_old
;
504 if (TC_U32_KEY(handle
))
508 ht_down
= u32_lookup_ht(ht
->tp_c
, handle
);
517 n
->ht_down
= ht_down
;
523 if (tb
[TCA_U32_CLASSID
]) {
524 n
->res
.classid
= nla_get_u32(tb
[TCA_U32_CLASSID
]);
525 tcf_bind_filter(tp
, &n
->res
, base
);
528 #ifdef CONFIG_NET_CLS_IND
529 if (tb
[TCA_U32_INDEV
]) {
531 ret
= tcf_change_indev(net
, tb
[TCA_U32_INDEV
]);
537 tcf_exts_change(tp
, &n
->exts
, &e
);
541 tcf_exts_destroy(tp
, &e
);
545 static int u32_change(struct net
*net
, struct sk_buff
*in_skb
,
546 struct tcf_proto
*tp
, unsigned long base
, u32 handle
,
548 unsigned long *arg
, bool ovr
)
550 struct tc_u_common
*tp_c
= tp
->data
;
551 struct tc_u_hnode
*ht
;
552 struct tc_u_knode
*n
;
553 struct tc_u32_sel
*s
;
554 struct nlattr
*opt
= tca
[TCA_OPTIONS
];
555 struct nlattr
*tb
[TCA_U32_MAX
+ 1];
560 return handle
? -EINVAL
: 0;
562 err
= nla_parse_nested(tb
, TCA_U32_MAX
, opt
, u32_policy
);
566 n
= (struct tc_u_knode
*)*arg
;
568 if (TC_U32_KEY(n
->handle
) == 0)
571 return u32_set_parms(net
, tp
, base
, n
->ht_up
, n
, tb
,
575 if (tb
[TCA_U32_DIVISOR
]) {
576 unsigned int divisor
= nla_get_u32(tb
[TCA_U32_DIVISOR
]);
578 if (--divisor
> 0x100)
580 if (TC_U32_KEY(handle
))
583 handle
= gen_new_htid(tp
->data
);
587 ht
= kzalloc(sizeof(*ht
) + divisor
*sizeof(void *), GFP_KERNEL
);
592 ht
->divisor
= divisor
;
595 ht
->next
= tp_c
->hlist
;
597 *arg
= (unsigned long)ht
;
601 if (tb
[TCA_U32_HASH
]) {
602 htid
= nla_get_u32(tb
[TCA_U32_HASH
]);
603 if (TC_U32_HTID(htid
) == TC_U32_ROOT
) {
607 ht
= u32_lookup_ht(tp
->data
, TC_U32_HTID(htid
));
616 if (ht
->divisor
< TC_U32_HASH(htid
))
620 if (TC_U32_HTID(handle
) && TC_U32_HTID(handle
^htid
))
622 handle
= htid
| TC_U32_NODE(handle
);
624 handle
= gen_new_kid(ht
, htid
);
626 if (tb
[TCA_U32_SEL
] == NULL
)
629 s
= nla_data(tb
[TCA_U32_SEL
]);
631 n
= kzalloc(sizeof(*n
) + s
->nkeys
*sizeof(struct tc_u32_key
), GFP_KERNEL
);
635 #ifdef CONFIG_CLS_U32_PERF
636 n
->pf
= kzalloc(sizeof(struct tc_u32_pcnt
) + s
->nkeys
*sizeof(u64
), GFP_KERNEL
);
643 memcpy(&n
->sel
, s
, sizeof(*s
) + s
->nkeys
*sizeof(struct tc_u32_key
));
646 n
->fshift
= s
->hmask
? ffs(ntohl(s
->hmask
)) - 1 : 0;
647 tcf_exts_init(&n
->exts
, TCA_U32_ACT
, TCA_U32_POLICE
);
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(net
, tp
, base
, ht
, n
, tb
, tca
[TCA_RATE
], ovr
);
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 net
*net
, 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;
736 if (nla_put_u32(skb
, TCA_U32_DIVISOR
, divisor
))
737 goto nla_put_failure
;
739 if (nla_put(skb
, TCA_U32_SEL
,
740 sizeof(n
->sel
) + n
->sel
.nkeys
*sizeof(struct tc_u32_key
),
742 goto nla_put_failure
;
744 u32 htid
= n
->handle
& 0xFFFFF000;
745 if (nla_put_u32(skb
, TCA_U32_HASH
, htid
))
746 goto nla_put_failure
;
748 if (n
->res
.classid
&&
749 nla_put_u32(skb
, TCA_U32_CLASSID
, n
->res
.classid
))
750 goto nla_put_failure
;
752 nla_put_u32(skb
, TCA_U32_LINK
, n
->ht_down
->handle
))
753 goto nla_put_failure
;
755 #ifdef CONFIG_CLS_U32_MARK
756 if ((n
->mark
.val
|| n
->mark
.mask
) &&
757 nla_put(skb
, TCA_U32_MARK
, sizeof(n
->mark
), &n
->mark
))
758 goto nla_put_failure
;
761 if (tcf_exts_dump(skb
, &n
->exts
) < 0)
762 goto nla_put_failure
;
764 #ifdef CONFIG_NET_CLS_IND
766 struct net_device
*dev
;
767 dev
= __dev_get_by_index(net
, n
->ifindex
);
768 if (dev
&& nla_put_string(skb
, TCA_U32_INDEV
, dev
->name
))
769 goto nla_put_failure
;
772 #ifdef CONFIG_CLS_U32_PERF
773 if (nla_put(skb
, TCA_U32_PCNT
,
774 sizeof(struct tc_u32_pcnt
) + n
->sel
.nkeys
*sizeof(u64
),
776 goto nla_put_failure
;
780 nla_nest_end(skb
, nest
);
782 if (TC_U32_KEY(n
->handle
))
783 if (tcf_exts_dump_stats(skb
, &n
->exts
) < 0)
784 goto nla_put_failure
;
788 nla_nest_cancel(skb
, nest
);
792 static struct tcf_proto_ops cls_u32_ops __read_mostly
= {
794 .classify
= u32_classify
,
796 .destroy
= u32_destroy
,
799 .change
= u32_change
,
800 .delete = u32_delete
,
803 .owner
= THIS_MODULE
,
806 static int __init
init_u32(void)
808 pr_info("u32 classifier\n");
809 #ifdef CONFIG_CLS_U32_PERF
810 pr_info(" Performance counters on\n");
812 #ifdef CONFIG_NET_CLS_IND
813 pr_info(" input device check on\n");
815 #ifdef CONFIG_NET_CLS_ACT
816 pr_info(" Actions configured\n");
818 return register_tcf_proto_ops(&cls_u32_ops
);
821 static void __exit
exit_u32(void)
823 unregister_tcf_proto_ops(&cls_u32_ops
);
826 module_init(init_u32
)
827 module_exit(exit_u32
)
828 MODULE_LICENSE("GPL");