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/percpu.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/skbuff.h>
42 #include <linux/bitmap.h>
43 #include <net/netlink.h>
44 #include <net/act_api.h>
45 #include <net/pkt_cls.h>
46 #include <linux/netdevice.h>
49 struct tc_u_knode __rcu
*next
;
51 struct tc_u_hnode __rcu
*ht_up
;
53 #ifdef CONFIG_NET_CLS_IND
57 struct tcf_result res
;
58 struct tc_u_hnode __rcu
*ht_down
;
59 #ifdef CONFIG_CLS_U32_PERF
60 struct tc_u32_pcnt __percpu
*pf
;
63 #ifdef CONFIG_CLS_U32_MARK
66 u32 __percpu
*pcpu_success
;
70 /* The 'sel' field MUST be the last field in structure to allow for
71 * tc_u32_keys allocated at end of structure.
73 struct tc_u32_sel sel
;
77 struct tc_u_hnode __rcu
*next
;
80 struct tc_u_common
*tp_c
;
84 /* The 'ht' field MUST be the last field in structure to allow for
85 * more entries allocated at end of structure.
87 struct tc_u_knode __rcu
*ht
[1];
91 struct tc_u_hnode __rcu
*hlist
;
98 static inline unsigned int u32_hash_fold(__be32 key
,
99 const struct tc_u32_sel
*sel
,
102 unsigned int h
= ntohl(key
& sel
->hmask
) >> fshift
;
107 static int u32_classify(struct sk_buff
*skb
, const struct tcf_proto
*tp
, struct tcf_result
*res
)
110 struct tc_u_knode
*knode
;
112 } stack
[TC_U32_MAXDEPTH
];
114 struct tc_u_hnode
*ht
= rcu_dereference_bh(tp
->root
);
115 unsigned int off
= skb_network_offset(skb
);
116 struct tc_u_knode
*n
;
120 #ifdef CONFIG_CLS_U32_PERF
126 n
= rcu_dereference_bh(ht
->ht
[sel
]);
130 struct tc_u32_key
*key
= n
->sel
.keys
;
132 #ifdef CONFIG_CLS_U32_PERF
133 __this_cpu_inc(n
->pf
->rcnt
);
137 if (tc_skip_sw(n
->flags
)) {
138 n
= rcu_dereference_bh(n
->next
);
142 #ifdef CONFIG_CLS_U32_MARK
143 if ((skb
->mark
& n
->mask
) != n
->val
) {
144 n
= rcu_dereference_bh(n
->next
);
147 __this_cpu_inc(*n
->pcpu_success
);
151 for (i
= n
->sel
.nkeys
; i
> 0; i
--, key
++) {
152 int toff
= off
+ key
->off
+ (off2
& key
->offmask
);
155 if (skb_headroom(skb
) + toff
> INT_MAX
)
158 data
= skb_header_pointer(skb
, toff
, 4, &hdata
);
161 if ((*data
^ key
->val
) & key
->mask
) {
162 n
= rcu_dereference_bh(n
->next
);
165 #ifdef CONFIG_CLS_U32_PERF
166 __this_cpu_inc(n
->pf
->kcnts
[j
]);
171 ht
= rcu_dereference_bh(n
->ht_down
);
174 if (n
->sel
.flags
& TC_U32_TERMINAL
) {
177 #ifdef CONFIG_NET_CLS_IND
178 if (!tcf_match_indev(skb
, n
->ifindex
)) {
179 n
= rcu_dereference_bh(n
->next
);
183 #ifdef CONFIG_CLS_U32_PERF
184 __this_cpu_inc(n
->pf
->rhit
);
186 r
= tcf_exts_exec(skb
, &n
->exts
, res
);
188 n
= rcu_dereference_bh(n
->next
);
194 n
= rcu_dereference_bh(n
->next
);
199 if (sdepth
>= TC_U32_MAXDEPTH
)
201 stack
[sdepth
].knode
= n
;
202 stack
[sdepth
].off
= off
;
205 ht
= rcu_dereference_bh(n
->ht_down
);
210 data
= skb_header_pointer(skb
, off
+ n
->sel
.hoff
, 4,
214 sel
= ht
->divisor
& u32_hash_fold(*data
, &n
->sel
,
217 if (!(n
->sel
.flags
& (TC_U32_VAROFFSET
| TC_U32_OFFSET
| TC_U32_EAT
)))
220 if (n
->sel
.flags
& (TC_U32_OFFSET
| TC_U32_VAROFFSET
)) {
221 off2
= n
->sel
.off
+ 3;
222 if (n
->sel
.flags
& TC_U32_VAROFFSET
) {
225 data
= skb_header_pointer(skb
,
230 off2
+= ntohs(n
->sel
.offmask
& *data
) >>
235 if (n
->sel
.flags
& TC_U32_EAT
) {
246 n
= stack
[sdepth
].knode
;
247 ht
= rcu_dereference_bh(n
->ht_up
);
248 off
= stack
[sdepth
].off
;
255 net_warn_ratelimited("cls_u32: dead loop\n");
259 static struct tc_u_hnode
*
260 u32_lookup_ht(struct tc_u_common
*tp_c
, u32 handle
)
262 struct tc_u_hnode
*ht
;
264 for (ht
= rtnl_dereference(tp_c
->hlist
);
266 ht
= rtnl_dereference(ht
->next
))
267 if (ht
->handle
== handle
)
273 static struct tc_u_knode
*
274 u32_lookup_key(struct tc_u_hnode
*ht
, u32 handle
)
277 struct tc_u_knode
*n
= NULL
;
279 sel
= TC_U32_HASH(handle
);
280 if (sel
> ht
->divisor
)
283 for (n
= rtnl_dereference(ht
->ht
[sel
]);
285 n
= rtnl_dereference(n
->next
))
286 if (n
->handle
== handle
)
293 static unsigned long u32_get(struct tcf_proto
*tp
, u32 handle
)
295 struct tc_u_hnode
*ht
;
296 struct tc_u_common
*tp_c
= tp
->data
;
298 if (TC_U32_HTID(handle
) == TC_U32_ROOT
)
299 ht
= rtnl_dereference(tp
->root
);
301 ht
= u32_lookup_ht(tp_c
, TC_U32_HTID(handle
));
306 if (TC_U32_KEY(handle
) == 0)
307 return (unsigned long)ht
;
309 return (unsigned long)u32_lookup_key(ht
, handle
);
312 static u32
gen_new_htid(struct tc_u_common
*tp_c
)
316 /* hgenerator only used inside rtnl lock it is safe to increment
317 * without read _copy_ update semantics
320 if (++tp_c
->hgenerator
== 0x7FF)
321 tp_c
->hgenerator
= 1;
322 } while (--i
> 0 && u32_lookup_ht(tp_c
, (tp_c
->hgenerator
|0x800)<<20));
324 return i
> 0 ? (tp_c
->hgenerator
|0x800)<<20 : 0;
327 static int u32_init(struct tcf_proto
*tp
)
329 struct tc_u_hnode
*root_ht
;
330 struct tc_u_common
*tp_c
;
332 tp_c
= tp
->q
->u32_node
;
334 root_ht
= kzalloc(sizeof(*root_ht
), GFP_KERNEL
);
338 root_ht
->divisor
= 0;
340 root_ht
->handle
= tp_c
? gen_new_htid(tp_c
) : 0x80000000;
341 root_ht
->prio
= tp
->prio
;
344 tp_c
= kzalloc(sizeof(*tp_c
), GFP_KERNEL
);
350 tp
->q
->u32_node
= tp_c
;
354 RCU_INIT_POINTER(root_ht
->next
, tp_c
->hlist
);
355 rcu_assign_pointer(tp_c
->hlist
, root_ht
);
356 root_ht
->tp_c
= tp_c
;
358 rcu_assign_pointer(tp
->root
, root_ht
);
363 static int u32_destroy_key(struct tcf_proto
*tp
,
364 struct tc_u_knode
*n
,
367 tcf_exts_destroy(&n
->exts
);
369 n
->ht_down
->refcnt
--;
370 #ifdef CONFIG_CLS_U32_PERF
374 #ifdef CONFIG_CLS_U32_MARK
376 free_percpu(n
->pcpu_success
);
382 /* u32_delete_key_rcu should be called when free'ing a copied
383 * version of a tc_u_knode obtained from u32_init_knode(). When
384 * copies are obtained from u32_init_knode() the statistics are
385 * shared between the old and new copies to allow readers to
386 * continue to update the statistics during the copy. To support
387 * this the u32_delete_key_rcu variant does not free the percpu
390 static void u32_delete_key_rcu(struct rcu_head
*rcu
)
392 struct tc_u_knode
*key
= container_of(rcu
, struct tc_u_knode
, rcu
);
394 u32_destroy_key(key
->tp
, key
, false);
397 /* u32_delete_key_freepf_rcu is the rcu callback variant
398 * that free's the entire structure including the statistics
399 * percpu variables. Only use this if the key is not a copy
400 * returned by u32_init_knode(). See u32_delete_key_rcu()
401 * for the variant that should be used with keys return from
404 static void u32_delete_key_freepf_rcu(struct rcu_head
*rcu
)
406 struct tc_u_knode
*key
= container_of(rcu
, struct tc_u_knode
, rcu
);
408 u32_destroy_key(key
->tp
, key
, true);
411 static int u32_delete_key(struct tcf_proto
*tp
, struct tc_u_knode
*key
)
413 struct tc_u_knode __rcu
**kp
;
414 struct tc_u_knode
*pkp
;
415 struct tc_u_hnode
*ht
= rtnl_dereference(key
->ht_up
);
418 kp
= &ht
->ht
[TC_U32_HASH(key
->handle
)];
419 for (pkp
= rtnl_dereference(*kp
); pkp
;
420 kp
= &pkp
->next
, pkp
= rtnl_dereference(*kp
)) {
422 RCU_INIT_POINTER(*kp
, key
->next
);
424 tcf_unbind_filter(tp
, &key
->res
);
425 call_rcu(&key
->rcu
, u32_delete_key_freepf_rcu
);
434 static void u32_remove_hw_knode(struct tcf_proto
*tp
, u32 handle
)
436 struct net_device
*dev
= tp
->q
->dev_queue
->dev
;
437 struct tc_cls_u32_offload u32_offload
= {0};
438 struct tc_to_netdev offload
;
440 offload
.type
= TC_SETUP_CLSU32
;
441 offload
.cls_u32
= &u32_offload
;
443 if (tc_should_offload(dev
, 0)) {
444 offload
.cls_u32
->command
= TC_CLSU32_DELETE_KNODE
;
445 offload
.cls_u32
->knode
.handle
= handle
;
446 dev
->netdev_ops
->ndo_setup_tc(dev
, tp
->q
->handle
,
447 tp
->protocol
, &offload
);
451 static int u32_replace_hw_hnode(struct tcf_proto
*tp
,
452 struct tc_u_hnode
*h
,
455 struct net_device
*dev
= tp
->q
->dev_queue
->dev
;
456 struct tc_cls_u32_offload u32_offload
= {0};
457 struct tc_to_netdev offload
;
460 offload
.type
= TC_SETUP_CLSU32
;
461 offload
.cls_u32
= &u32_offload
;
463 if (tc_should_offload(dev
, flags
)) {
464 offload
.cls_u32
->command
= TC_CLSU32_NEW_HNODE
;
465 offload
.cls_u32
->hnode
.divisor
= h
->divisor
;
466 offload
.cls_u32
->hnode
.handle
= h
->handle
;
467 offload
.cls_u32
->hnode
.prio
= h
->prio
;
469 err
= dev
->netdev_ops
->ndo_setup_tc(dev
, tp
->q
->handle
,
470 tp
->protocol
, &offload
);
471 if (tc_skip_sw(flags
))
478 static void u32_clear_hw_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*h
)
480 struct net_device
*dev
= tp
->q
->dev_queue
->dev
;
481 struct tc_cls_u32_offload u32_offload
= {0};
482 struct tc_to_netdev offload
;
484 offload
.type
= TC_SETUP_CLSU32
;
485 offload
.cls_u32
= &u32_offload
;
487 if (tc_should_offload(dev
, 0)) {
488 offload
.cls_u32
->command
= TC_CLSU32_DELETE_HNODE
;
489 offload
.cls_u32
->hnode
.divisor
= h
->divisor
;
490 offload
.cls_u32
->hnode
.handle
= h
->handle
;
491 offload
.cls_u32
->hnode
.prio
= h
->prio
;
493 dev
->netdev_ops
->ndo_setup_tc(dev
, tp
->q
->handle
,
494 tp
->protocol
, &offload
);
498 static int u32_replace_hw_knode(struct tcf_proto
*tp
,
499 struct tc_u_knode
*n
,
502 struct net_device
*dev
= tp
->q
->dev_queue
->dev
;
503 struct tc_cls_u32_offload u32_offload
= {0};
504 struct tc_to_netdev offload
;
507 offload
.type
= TC_SETUP_CLSU32
;
508 offload
.cls_u32
= &u32_offload
;
510 if (tc_should_offload(dev
, flags
)) {
511 offload
.cls_u32
->command
= TC_CLSU32_REPLACE_KNODE
;
512 offload
.cls_u32
->knode
.handle
= n
->handle
;
513 offload
.cls_u32
->knode
.fshift
= n
->fshift
;
514 #ifdef CONFIG_CLS_U32_MARK
515 offload
.cls_u32
->knode
.val
= n
->val
;
516 offload
.cls_u32
->knode
.mask
= n
->mask
;
518 offload
.cls_u32
->knode
.val
= 0;
519 offload
.cls_u32
->knode
.mask
= 0;
521 offload
.cls_u32
->knode
.sel
= &n
->sel
;
522 offload
.cls_u32
->knode
.exts
= &n
->exts
;
524 offload
.cls_u32
->knode
.link_handle
= n
->ht_down
->handle
;
526 err
= dev
->netdev_ops
->ndo_setup_tc(dev
, tp
->q
->handle
,
527 tp
->protocol
, &offload
);
528 if (tc_skip_sw(flags
))
535 static void u32_clear_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
)
537 struct tc_u_knode
*n
;
540 for (h
= 0; h
<= ht
->divisor
; h
++) {
541 while ((n
= rtnl_dereference(ht
->ht
[h
])) != NULL
) {
542 RCU_INIT_POINTER(ht
->ht
[h
],
543 rtnl_dereference(n
->next
));
544 tcf_unbind_filter(tp
, &n
->res
);
545 u32_remove_hw_knode(tp
, n
->handle
);
546 call_rcu(&n
->rcu
, u32_delete_key_freepf_rcu
);
551 static int u32_destroy_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
)
553 struct tc_u_common
*tp_c
= tp
->data
;
554 struct tc_u_hnode __rcu
**hn
;
555 struct tc_u_hnode
*phn
;
559 u32_clear_hnode(tp
, ht
);
562 for (phn
= rtnl_dereference(*hn
);
564 hn
= &phn
->next
, phn
= rtnl_dereference(*hn
)) {
566 u32_clear_hw_hnode(tp
, ht
);
567 RCU_INIT_POINTER(*hn
, ht
->next
);
576 static bool ht_empty(struct tc_u_hnode
*ht
)
580 for (h
= 0; h
<= ht
->divisor
; h
++)
581 if (rcu_access_pointer(ht
->ht
[h
]))
587 static bool u32_destroy(struct tcf_proto
*tp
, bool force
)
589 struct tc_u_common
*tp_c
= tp
->data
;
590 struct tc_u_hnode
*root_ht
= rtnl_dereference(tp
->root
);
592 WARN_ON(root_ht
== NULL
);
596 if (root_ht
->refcnt
> 1)
598 if (root_ht
->refcnt
== 1) {
599 if (!ht_empty(root_ht
))
604 if (tp_c
->refcnt
> 1)
607 if (tp_c
->refcnt
== 1) {
608 struct tc_u_hnode
*ht
;
610 for (ht
= rtnl_dereference(tp_c
->hlist
);
612 ht
= rtnl_dereference(ht
->next
))
618 if (root_ht
&& --root_ht
->refcnt
== 0)
619 u32_destroy_hnode(tp
, root_ht
);
621 if (--tp_c
->refcnt
== 0) {
622 struct tc_u_hnode
*ht
;
624 tp
->q
->u32_node
= NULL
;
626 for (ht
= rtnl_dereference(tp_c
->hlist
);
628 ht
= rtnl_dereference(ht
->next
)) {
630 u32_clear_hnode(tp
, ht
);
633 while ((ht
= rtnl_dereference(tp_c
->hlist
)) != NULL
) {
634 RCU_INIT_POINTER(tp_c
->hlist
, ht
->next
);
645 static int u32_delete(struct tcf_proto
*tp
, unsigned long arg
)
647 struct tc_u_hnode
*ht
= (struct tc_u_hnode
*)arg
;
648 struct tc_u_hnode
*root_ht
= rtnl_dereference(tp
->root
);
653 if (TC_U32_KEY(ht
->handle
)) {
654 u32_remove_hw_knode(tp
, ht
->handle
);
655 return u32_delete_key(tp
, (struct tc_u_knode
*)ht
);
661 if (ht
->refcnt
== 1) {
663 u32_destroy_hnode(tp
, ht
);
671 #define NR_U32_NODE (1<<12)
672 static u32
gen_new_kid(struct tc_u_hnode
*ht
, u32 handle
)
674 struct tc_u_knode
*n
;
676 unsigned long *bitmap
= kzalloc(BITS_TO_LONGS(NR_U32_NODE
) * sizeof(unsigned long),
679 return handle
| 0xFFF;
681 for (n
= rtnl_dereference(ht
->ht
[TC_U32_HASH(handle
)]);
683 n
= rtnl_dereference(n
->next
))
684 set_bit(TC_U32_NODE(n
->handle
), bitmap
);
686 i
= find_next_zero_bit(bitmap
, NR_U32_NODE
, 0x800);
687 if (i
>= NR_U32_NODE
)
688 i
= find_next_zero_bit(bitmap
, NR_U32_NODE
, 1);
691 return handle
| (i
>= NR_U32_NODE
? 0xFFF : i
);
694 static const struct nla_policy u32_policy
[TCA_U32_MAX
+ 1] = {
695 [TCA_U32_CLASSID
] = { .type
= NLA_U32
},
696 [TCA_U32_HASH
] = { .type
= NLA_U32
},
697 [TCA_U32_LINK
] = { .type
= NLA_U32
},
698 [TCA_U32_DIVISOR
] = { .type
= NLA_U32
},
699 [TCA_U32_SEL
] = { .len
= sizeof(struct tc_u32_sel
) },
700 [TCA_U32_INDEV
] = { .type
= NLA_STRING
, .len
= IFNAMSIZ
},
701 [TCA_U32_MARK
] = { .len
= sizeof(struct tc_u32_mark
) },
702 [TCA_U32_FLAGS
] = { .type
= NLA_U32
},
705 static int u32_set_parms(struct net
*net
, struct tcf_proto
*tp
,
706 unsigned long base
, struct tc_u_hnode
*ht
,
707 struct tc_u_knode
*n
, struct nlattr
**tb
,
708 struct nlattr
*est
, bool ovr
)
713 tcf_exts_init(&e
, TCA_U32_ACT
, TCA_U32_POLICE
);
714 err
= tcf_exts_validate(net
, tp
, tb
, est
, &e
, ovr
);
719 if (tb
[TCA_U32_LINK
]) {
720 u32 handle
= nla_get_u32(tb
[TCA_U32_LINK
]);
721 struct tc_u_hnode
*ht_down
= NULL
, *ht_old
;
723 if (TC_U32_KEY(handle
))
727 ht_down
= u32_lookup_ht(ht
->tp_c
, handle
);
734 ht_old
= rtnl_dereference(n
->ht_down
);
735 rcu_assign_pointer(n
->ht_down
, ht_down
);
740 if (tb
[TCA_U32_CLASSID
]) {
741 n
->res
.classid
= nla_get_u32(tb
[TCA_U32_CLASSID
]);
742 tcf_bind_filter(tp
, &n
->res
, base
);
745 #ifdef CONFIG_NET_CLS_IND
746 if (tb
[TCA_U32_INDEV
]) {
748 ret
= tcf_change_indev(net
, tb
[TCA_U32_INDEV
]);
754 tcf_exts_change(tp
, &n
->exts
, &e
);
758 tcf_exts_destroy(&e
);
762 static void u32_replace_knode(struct tcf_proto
*tp
,
763 struct tc_u_common
*tp_c
,
764 struct tc_u_knode
*n
)
766 struct tc_u_knode __rcu
**ins
;
767 struct tc_u_knode
*pins
;
768 struct tc_u_hnode
*ht
;
770 if (TC_U32_HTID(n
->handle
) == TC_U32_ROOT
)
771 ht
= rtnl_dereference(tp
->root
);
773 ht
= u32_lookup_ht(tp_c
, TC_U32_HTID(n
->handle
));
775 ins
= &ht
->ht
[TC_U32_HASH(n
->handle
)];
777 /* The node must always exist for it to be replaced if this is not the
778 * case then something went very wrong elsewhere.
780 for (pins
= rtnl_dereference(*ins
); ;
781 ins
= &pins
->next
, pins
= rtnl_dereference(*ins
))
782 if (pins
->handle
== n
->handle
)
785 RCU_INIT_POINTER(n
->next
, pins
->next
);
786 rcu_assign_pointer(*ins
, n
);
789 static struct tc_u_knode
*u32_init_knode(struct tcf_proto
*tp
,
790 struct tc_u_knode
*n
)
792 struct tc_u_knode
*new;
793 struct tc_u32_sel
*s
= &n
->sel
;
795 new = kzalloc(sizeof(*n
) + s
->nkeys
*sizeof(struct tc_u32_key
),
801 RCU_INIT_POINTER(new->next
, n
->next
);
802 new->handle
= n
->handle
;
803 RCU_INIT_POINTER(new->ht_up
, n
->ht_up
);
805 #ifdef CONFIG_NET_CLS_IND
806 new->ifindex
= n
->ifindex
;
808 new->fshift
= n
->fshift
;
810 new->flags
= n
->flags
;
811 RCU_INIT_POINTER(new->ht_down
, n
->ht_down
);
813 /* bump reference count as long as we hold pointer to structure */
815 new->ht_down
->refcnt
++;
817 #ifdef CONFIG_CLS_U32_PERF
818 /* Statistics may be incremented by readers during update
819 * so we must keep them in tact. When the node is later destroyed
820 * a special destroy call must be made to not free the pf memory.
825 #ifdef CONFIG_CLS_U32_MARK
828 /* Similarly success statistics must be moved as pointers */
829 new->pcpu_success
= n
->pcpu_success
;
832 memcpy(&new->sel
, s
, sizeof(*s
) + s
->nkeys
*sizeof(struct tc_u32_key
));
834 tcf_exts_init(&new->exts
, TCA_U32_ACT
, TCA_U32_POLICE
);
839 static int u32_change(struct net
*net
, struct sk_buff
*in_skb
,
840 struct tcf_proto
*tp
, unsigned long base
, u32 handle
,
842 unsigned long *arg
, bool ovr
)
844 struct tc_u_common
*tp_c
= tp
->data
;
845 struct tc_u_hnode
*ht
;
846 struct tc_u_knode
*n
;
847 struct tc_u32_sel
*s
;
848 struct nlattr
*opt
= tca
[TCA_OPTIONS
];
849 struct nlattr
*tb
[TCA_U32_MAX
+ 1];
852 #ifdef CONFIG_CLS_U32_PERF
857 return handle
? -EINVAL
: 0;
859 err
= nla_parse_nested(tb
, TCA_U32_MAX
, opt
, u32_policy
);
863 if (tb
[TCA_U32_FLAGS
]) {
864 flags
= nla_get_u32(tb
[TCA_U32_FLAGS
]);
865 if (!tc_flags_valid(flags
))
869 n
= (struct tc_u_knode
*)*arg
;
871 struct tc_u_knode
*new;
873 if (TC_U32_KEY(n
->handle
) == 0)
876 if (n
->flags
!= flags
)
879 new = u32_init_knode(tp
, n
);
883 err
= u32_set_parms(net
, tp
, base
,
884 rtnl_dereference(n
->ht_up
), new, tb
,
888 u32_destroy_key(tp
, new, false);
892 err
= u32_replace_hw_knode(tp
, new, flags
);
894 u32_destroy_key(tp
, new, false);
898 u32_replace_knode(tp
, tp_c
, new);
899 tcf_unbind_filter(tp
, &n
->res
);
900 call_rcu(&n
->rcu
, u32_delete_key_rcu
);
904 if (tb
[TCA_U32_DIVISOR
]) {
905 unsigned int divisor
= nla_get_u32(tb
[TCA_U32_DIVISOR
]);
907 if (--divisor
> 0x100)
909 if (TC_U32_KEY(handle
))
912 handle
= gen_new_htid(tp
->data
);
916 ht
= kzalloc(sizeof(*ht
) + divisor
*sizeof(void *), GFP_KERNEL
);
921 ht
->divisor
= divisor
;
924 RCU_INIT_POINTER(ht
->next
, tp_c
->hlist
);
925 rcu_assign_pointer(tp_c
->hlist
, ht
);
926 *arg
= (unsigned long)ht
;
928 u32_replace_hw_hnode(tp
, ht
, flags
);
932 if (tb
[TCA_U32_HASH
]) {
933 htid
= nla_get_u32(tb
[TCA_U32_HASH
]);
934 if (TC_U32_HTID(htid
) == TC_U32_ROOT
) {
935 ht
= rtnl_dereference(tp
->root
);
938 ht
= u32_lookup_ht(tp
->data
, TC_U32_HTID(htid
));
943 ht
= rtnl_dereference(tp
->root
);
947 if (ht
->divisor
< TC_U32_HASH(htid
))
951 if (TC_U32_HTID(handle
) && TC_U32_HTID(handle
^htid
))
953 handle
= htid
| TC_U32_NODE(handle
);
955 handle
= gen_new_kid(ht
, htid
);
957 if (tb
[TCA_U32_SEL
] == NULL
)
960 s
= nla_data(tb
[TCA_U32_SEL
]);
962 n
= kzalloc(sizeof(*n
) + s
->nkeys
*sizeof(struct tc_u32_key
), GFP_KERNEL
);
966 #ifdef CONFIG_CLS_U32_PERF
967 size
= sizeof(struct tc_u32_pcnt
) + s
->nkeys
* sizeof(u64
);
968 n
->pf
= __alloc_percpu(size
, __alignof__(struct tc_u32_pcnt
));
975 memcpy(&n
->sel
, s
, sizeof(*s
) + s
->nkeys
*sizeof(struct tc_u32_key
));
976 RCU_INIT_POINTER(n
->ht_up
, ht
);
978 n
->fshift
= s
->hmask
? ffs(ntohl(s
->hmask
)) - 1 : 0;
980 tcf_exts_init(&n
->exts
, TCA_U32_ACT
, TCA_U32_POLICE
);
983 #ifdef CONFIG_CLS_U32_MARK
984 n
->pcpu_success
= alloc_percpu(u32
);
985 if (!n
->pcpu_success
) {
990 if (tb
[TCA_U32_MARK
]) {
991 struct tc_u32_mark
*mark
;
993 mark
= nla_data(tb
[TCA_U32_MARK
]);
995 n
->mask
= mark
->mask
;
999 err
= u32_set_parms(net
, tp
, base
, ht
, n
, tb
, tca
[TCA_RATE
], ovr
);
1001 struct tc_u_knode __rcu
**ins
;
1002 struct tc_u_knode
*pins
;
1004 err
= u32_replace_hw_knode(tp
, n
, flags
);
1008 ins
= &ht
->ht
[TC_U32_HASH(handle
)];
1009 for (pins
= rtnl_dereference(*ins
); pins
;
1010 ins
= &pins
->next
, pins
= rtnl_dereference(*ins
))
1011 if (TC_U32_NODE(handle
) < TC_U32_NODE(pins
->handle
))
1014 RCU_INIT_POINTER(n
->next
, pins
);
1015 rcu_assign_pointer(*ins
, n
);
1016 *arg
= (unsigned long)n
;
1021 #ifdef CONFIG_CLS_U32_MARK
1022 free_percpu(n
->pcpu_success
);
1026 #ifdef CONFIG_CLS_U32_PERF
1033 static void u32_walk(struct tcf_proto
*tp
, struct tcf_walker
*arg
)
1035 struct tc_u_common
*tp_c
= tp
->data
;
1036 struct tc_u_hnode
*ht
;
1037 struct tc_u_knode
*n
;
1043 for (ht
= rtnl_dereference(tp_c
->hlist
);
1045 ht
= rtnl_dereference(ht
->next
)) {
1046 if (ht
->prio
!= tp
->prio
)
1048 if (arg
->count
>= arg
->skip
) {
1049 if (arg
->fn(tp
, (unsigned long)ht
, arg
) < 0) {
1055 for (h
= 0; h
<= ht
->divisor
; h
++) {
1056 for (n
= rtnl_dereference(ht
->ht
[h
]);
1058 n
= rtnl_dereference(n
->next
)) {
1059 if (arg
->count
< arg
->skip
) {
1063 if (arg
->fn(tp
, (unsigned long)n
, arg
) < 0) {
1073 static int u32_dump(struct net
*net
, struct tcf_proto
*tp
, unsigned long fh
,
1074 struct sk_buff
*skb
, struct tcmsg
*t
)
1076 struct tc_u_knode
*n
= (struct tc_u_knode
*)fh
;
1077 struct tc_u_hnode
*ht_up
, *ht_down
;
1078 struct nlattr
*nest
;
1083 t
->tcm_handle
= n
->handle
;
1085 nest
= nla_nest_start(skb
, TCA_OPTIONS
);
1087 goto nla_put_failure
;
1089 if (TC_U32_KEY(n
->handle
) == 0) {
1090 struct tc_u_hnode
*ht
= (struct tc_u_hnode
*)fh
;
1091 u32 divisor
= ht
->divisor
+ 1;
1093 if (nla_put_u32(skb
, TCA_U32_DIVISOR
, divisor
))
1094 goto nla_put_failure
;
1096 #ifdef CONFIG_CLS_U32_PERF
1097 struct tc_u32_pcnt
*gpf
;
1101 if (nla_put(skb
, TCA_U32_SEL
,
1102 sizeof(n
->sel
) + n
->sel
.nkeys
*sizeof(struct tc_u32_key
),
1104 goto nla_put_failure
;
1106 ht_up
= rtnl_dereference(n
->ht_up
);
1108 u32 htid
= n
->handle
& 0xFFFFF000;
1109 if (nla_put_u32(skb
, TCA_U32_HASH
, htid
))
1110 goto nla_put_failure
;
1112 if (n
->res
.classid
&&
1113 nla_put_u32(skb
, TCA_U32_CLASSID
, n
->res
.classid
))
1114 goto nla_put_failure
;
1116 ht_down
= rtnl_dereference(n
->ht_down
);
1118 nla_put_u32(skb
, TCA_U32_LINK
, ht_down
->handle
))
1119 goto nla_put_failure
;
1121 if (n
->flags
&& nla_put_u32(skb
, TCA_U32_FLAGS
, n
->flags
))
1122 goto nla_put_failure
;
1124 #ifdef CONFIG_CLS_U32_MARK
1125 if ((n
->val
|| n
->mask
)) {
1126 struct tc_u32_mark mark
= {.val
= n
->val
,
1131 for_each_possible_cpu(cpum
) {
1132 __u32 cnt
= *per_cpu_ptr(n
->pcpu_success
, cpum
);
1134 mark
.success
+= cnt
;
1137 if (nla_put(skb
, TCA_U32_MARK
, sizeof(mark
), &mark
))
1138 goto nla_put_failure
;
1142 if (tcf_exts_dump(skb
, &n
->exts
) < 0)
1143 goto nla_put_failure
;
1145 #ifdef CONFIG_NET_CLS_IND
1147 struct net_device
*dev
;
1148 dev
= __dev_get_by_index(net
, n
->ifindex
);
1149 if (dev
&& nla_put_string(skb
, TCA_U32_INDEV
, dev
->name
))
1150 goto nla_put_failure
;
1153 #ifdef CONFIG_CLS_U32_PERF
1154 gpf
= kzalloc(sizeof(struct tc_u32_pcnt
) +
1155 n
->sel
.nkeys
* sizeof(u64
),
1158 goto nla_put_failure
;
1160 for_each_possible_cpu(cpu
) {
1162 struct tc_u32_pcnt
*pf
= per_cpu_ptr(n
->pf
, cpu
);
1164 gpf
->rcnt
+= pf
->rcnt
;
1165 gpf
->rhit
+= pf
->rhit
;
1166 for (i
= 0; i
< n
->sel
.nkeys
; i
++)
1167 gpf
->kcnts
[i
] += pf
->kcnts
[i
];
1170 if (nla_put_64bit(skb
, TCA_U32_PCNT
,
1171 sizeof(struct tc_u32_pcnt
) +
1172 n
->sel
.nkeys
* sizeof(u64
),
1173 gpf
, TCA_U32_PAD
)) {
1175 goto nla_put_failure
;
1181 nla_nest_end(skb
, nest
);
1183 if (TC_U32_KEY(n
->handle
))
1184 if (tcf_exts_dump_stats(skb
, &n
->exts
) < 0)
1185 goto nla_put_failure
;
1189 nla_nest_cancel(skb
, nest
);
1193 static struct tcf_proto_ops cls_u32_ops __read_mostly
= {
1195 .classify
= u32_classify
,
1197 .destroy
= u32_destroy
,
1199 .change
= u32_change
,
1200 .delete = u32_delete
,
1203 .owner
= THIS_MODULE
,
1206 static int __init
init_u32(void)
1208 pr_info("u32 classifier\n");
1209 #ifdef CONFIG_CLS_U32_PERF
1210 pr_info(" Performance counters on\n");
1212 #ifdef CONFIG_NET_CLS_IND
1213 pr_info(" input device check on\n");
1215 #ifdef CONFIG_NET_CLS_ACT
1216 pr_info(" Actions configured\n");
1218 return register_tcf_proto_ops(&cls_u32_ops
);
1221 static void __exit
exit_u32(void)
1223 unregister_tcf_proto_ops(&cls_u32_ops
);
1226 module_init(init_u32
)
1227 module_exit(exit_u32
)
1228 MODULE_LICENSE("GPL");