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
,
108 struct tcf_result
*res
)
111 struct tc_u_knode
*knode
;
113 } stack
[TC_U32_MAXDEPTH
];
115 struct tc_u_hnode
*ht
= rcu_dereference_bh(tp
->root
);
116 unsigned int off
= skb_network_offset(skb
);
117 struct tc_u_knode
*n
;
121 #ifdef CONFIG_CLS_U32_PERF
127 n
= rcu_dereference_bh(ht
->ht
[sel
]);
131 struct tc_u32_key
*key
= n
->sel
.keys
;
133 #ifdef CONFIG_CLS_U32_PERF
134 __this_cpu_inc(n
->pf
->rcnt
);
138 if (tc_skip_sw(n
->flags
)) {
139 n
= rcu_dereference_bh(n
->next
);
143 #ifdef CONFIG_CLS_U32_MARK
144 if ((skb
->mark
& n
->mask
) != n
->val
) {
145 n
= rcu_dereference_bh(n
->next
);
148 __this_cpu_inc(*n
->pcpu_success
);
152 for (i
= n
->sel
.nkeys
; i
> 0; i
--, key
++) {
153 int toff
= off
+ key
->off
+ (off2
& key
->offmask
);
156 if (skb_headroom(skb
) + toff
> INT_MAX
)
159 data
= skb_header_pointer(skb
, toff
, 4, &hdata
);
162 if ((*data
^ key
->val
) & key
->mask
) {
163 n
= rcu_dereference_bh(n
->next
);
166 #ifdef CONFIG_CLS_U32_PERF
167 __this_cpu_inc(n
->pf
->kcnts
[j
]);
172 ht
= rcu_dereference_bh(n
->ht_down
);
175 if (n
->sel
.flags
& TC_U32_TERMINAL
) {
178 #ifdef CONFIG_NET_CLS_IND
179 if (!tcf_match_indev(skb
, n
->ifindex
)) {
180 n
= rcu_dereference_bh(n
->next
);
184 #ifdef CONFIG_CLS_U32_PERF
185 __this_cpu_inc(n
->pf
->rhit
);
187 r
= tcf_exts_exec(skb
, &n
->exts
, res
);
189 n
= rcu_dereference_bh(n
->next
);
195 n
= rcu_dereference_bh(n
->next
);
200 if (sdepth
>= TC_U32_MAXDEPTH
)
202 stack
[sdepth
].knode
= n
;
203 stack
[sdepth
].off
= off
;
206 ht
= rcu_dereference_bh(n
->ht_down
);
211 data
= skb_header_pointer(skb
, off
+ n
->sel
.hoff
, 4,
215 sel
= ht
->divisor
& u32_hash_fold(*data
, &n
->sel
,
218 if (!(n
->sel
.flags
& (TC_U32_VAROFFSET
| TC_U32_OFFSET
| TC_U32_EAT
)))
221 if (n
->sel
.flags
& (TC_U32_OFFSET
| TC_U32_VAROFFSET
)) {
222 off2
= n
->sel
.off
+ 3;
223 if (n
->sel
.flags
& TC_U32_VAROFFSET
) {
226 data
= skb_header_pointer(skb
,
231 off2
+= ntohs(n
->sel
.offmask
& *data
) >>
236 if (n
->sel
.flags
& TC_U32_EAT
) {
247 n
= stack
[sdepth
].knode
;
248 ht
= rcu_dereference_bh(n
->ht_up
);
249 off
= stack
[sdepth
].off
;
256 net_warn_ratelimited("cls_u32: dead loop\n");
260 static struct tc_u_hnode
*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
*u32_lookup_key(struct tc_u_hnode
*ht
, u32 handle
)
276 struct tc_u_knode
*n
= NULL
;
278 sel
= TC_U32_HASH(handle
);
279 if (sel
> ht
->divisor
)
282 for (n
= rtnl_dereference(ht
->ht
[sel
]);
284 n
= rtnl_dereference(n
->next
))
285 if (n
->handle
== handle
)
292 static unsigned long u32_get(struct tcf_proto
*tp
, u32 handle
)
294 struct tc_u_hnode
*ht
;
295 struct tc_u_common
*tp_c
= tp
->data
;
297 if (TC_U32_HTID(handle
) == TC_U32_ROOT
)
298 ht
= rtnl_dereference(tp
->root
);
300 ht
= u32_lookup_ht(tp_c
, TC_U32_HTID(handle
));
305 if (TC_U32_KEY(handle
) == 0)
306 return (unsigned long)ht
;
308 return (unsigned long)u32_lookup_key(ht
, handle
);
311 static u32
gen_new_htid(struct tc_u_common
*tp_c
)
315 /* hgenerator only used inside rtnl lock it is safe to increment
316 * without read _copy_ update semantics
319 if (++tp_c
->hgenerator
== 0x7FF)
320 tp_c
->hgenerator
= 1;
321 } while (--i
> 0 && u32_lookup_ht(tp_c
, (tp_c
->hgenerator
|0x800)<<20));
323 return i
> 0 ? (tp_c
->hgenerator
|0x800)<<20 : 0;
326 static int u32_init(struct tcf_proto
*tp
)
328 struct tc_u_hnode
*root_ht
;
329 struct tc_u_common
*tp_c
;
331 tp_c
= tp
->q
->u32_node
;
333 root_ht
= kzalloc(sizeof(*root_ht
), GFP_KERNEL
);
338 root_ht
->handle
= tp_c
? gen_new_htid(tp_c
) : 0x80000000;
339 root_ht
->prio
= tp
->prio
;
342 tp_c
= kzalloc(sizeof(*tp_c
), GFP_KERNEL
);
348 tp
->q
->u32_node
= tp_c
;
352 RCU_INIT_POINTER(root_ht
->next
, tp_c
->hlist
);
353 rcu_assign_pointer(tp_c
->hlist
, root_ht
);
354 root_ht
->tp_c
= tp_c
;
356 rcu_assign_pointer(tp
->root
, root_ht
);
361 static int u32_destroy_key(struct tcf_proto
*tp
, struct tc_u_knode
*n
,
364 tcf_exts_destroy(&n
->exts
);
366 n
->ht_down
->refcnt
--;
367 #ifdef CONFIG_CLS_U32_PERF
371 #ifdef CONFIG_CLS_U32_MARK
373 free_percpu(n
->pcpu_success
);
379 /* u32_delete_key_rcu should be called when free'ing a copied
380 * version of a tc_u_knode obtained from u32_init_knode(). When
381 * copies are obtained from u32_init_knode() the statistics are
382 * shared between the old and new copies to allow readers to
383 * continue to update the statistics during the copy. To support
384 * this the u32_delete_key_rcu variant does not free the percpu
387 static void u32_delete_key_rcu(struct rcu_head
*rcu
)
389 struct tc_u_knode
*key
= container_of(rcu
, struct tc_u_knode
, rcu
);
391 u32_destroy_key(key
->tp
, key
, false);
394 /* u32_delete_key_freepf_rcu is the rcu callback variant
395 * that free's the entire structure including the statistics
396 * percpu variables. Only use this if the key is not a copy
397 * returned by u32_init_knode(). See u32_delete_key_rcu()
398 * for the variant that should be used with keys return from
401 static void u32_delete_key_freepf_rcu(struct rcu_head
*rcu
)
403 struct tc_u_knode
*key
= container_of(rcu
, struct tc_u_knode
, rcu
);
405 u32_destroy_key(key
->tp
, key
, true);
408 static int u32_delete_key(struct tcf_proto
*tp
, struct tc_u_knode
*key
)
410 struct tc_u_knode __rcu
**kp
;
411 struct tc_u_knode
*pkp
;
412 struct tc_u_hnode
*ht
= rtnl_dereference(key
->ht_up
);
415 kp
= &ht
->ht
[TC_U32_HASH(key
->handle
)];
416 for (pkp
= rtnl_dereference(*kp
); pkp
;
417 kp
= &pkp
->next
, pkp
= rtnl_dereference(*kp
)) {
419 RCU_INIT_POINTER(*kp
, key
->next
);
421 tcf_unbind_filter(tp
, &key
->res
);
422 call_rcu(&key
->rcu
, u32_delete_key_freepf_rcu
);
431 static void u32_remove_hw_knode(struct tcf_proto
*tp
, u32 handle
)
433 struct net_device
*dev
= tp
->q
->dev_queue
->dev
;
434 struct tc_cls_u32_offload u32_offload
= {0};
435 struct tc_to_netdev offload
;
437 offload
.type
= TC_SETUP_CLSU32
;
438 offload
.cls_u32
= &u32_offload
;
440 if (tc_should_offload(dev
, tp
, 0)) {
441 offload
.cls_u32
->command
= TC_CLSU32_DELETE_KNODE
;
442 offload
.cls_u32
->knode
.handle
= handle
;
443 dev
->netdev_ops
->ndo_setup_tc(dev
, tp
->q
->handle
,
444 tp
->protocol
, &offload
);
448 static int u32_replace_hw_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*h
,
451 struct net_device
*dev
= tp
->q
->dev_queue
->dev
;
452 struct tc_cls_u32_offload u32_offload
= {0};
453 struct tc_to_netdev offload
;
456 if (!tc_should_offload(dev
, tp
, flags
))
457 return tc_skip_sw(flags
) ? -EINVAL
: 0;
459 offload
.type
= TC_SETUP_CLSU32
;
460 offload
.cls_u32
= &u32_offload
;
462 offload
.cls_u32
->command
= TC_CLSU32_NEW_HNODE
;
463 offload
.cls_u32
->hnode
.divisor
= h
->divisor
;
464 offload
.cls_u32
->hnode
.handle
= h
->handle
;
465 offload
.cls_u32
->hnode
.prio
= h
->prio
;
467 err
= dev
->netdev_ops
->ndo_setup_tc(dev
, tp
->q
->handle
,
468 tp
->protocol
, &offload
);
469 if (tc_skip_sw(flags
))
475 static void u32_clear_hw_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*h
)
477 struct net_device
*dev
= tp
->q
->dev_queue
->dev
;
478 struct tc_cls_u32_offload u32_offload
= {0};
479 struct tc_to_netdev offload
;
481 offload
.type
= TC_SETUP_CLSU32
;
482 offload
.cls_u32
= &u32_offload
;
484 if (tc_should_offload(dev
, tp
, 0)) {
485 offload
.cls_u32
->command
= TC_CLSU32_DELETE_HNODE
;
486 offload
.cls_u32
->hnode
.divisor
= h
->divisor
;
487 offload
.cls_u32
->hnode
.handle
= h
->handle
;
488 offload
.cls_u32
->hnode
.prio
= h
->prio
;
490 dev
->netdev_ops
->ndo_setup_tc(dev
, tp
->q
->handle
,
491 tp
->protocol
, &offload
);
495 static int u32_replace_hw_knode(struct tcf_proto
*tp
, struct tc_u_knode
*n
,
498 struct net_device
*dev
= tp
->q
->dev_queue
->dev
;
499 struct tc_cls_u32_offload u32_offload
= {0};
500 struct tc_to_netdev offload
;
503 offload
.type
= TC_SETUP_CLSU32
;
504 offload
.cls_u32
= &u32_offload
;
506 if (!tc_should_offload(dev
, tp
, flags
))
507 return tc_skip_sw(flags
) ? -EINVAL
: 0;
509 offload
.cls_u32
->command
= TC_CLSU32_REPLACE_KNODE
;
510 offload
.cls_u32
->knode
.handle
= n
->handle
;
511 offload
.cls_u32
->knode
.fshift
= n
->fshift
;
512 #ifdef CONFIG_CLS_U32_MARK
513 offload
.cls_u32
->knode
.val
= n
->val
;
514 offload
.cls_u32
->knode
.mask
= n
->mask
;
516 offload
.cls_u32
->knode
.val
= 0;
517 offload
.cls_u32
->knode
.mask
= 0;
519 offload
.cls_u32
->knode
.sel
= &n
->sel
;
520 offload
.cls_u32
->knode
.exts
= &n
->exts
;
522 offload
.cls_u32
->knode
.link_handle
= n
->ht_down
->handle
;
524 err
= dev
->netdev_ops
->ndo_setup_tc(dev
, tp
->q
->handle
,
525 tp
->protocol
, &offload
);
528 n
->flags
|= TCA_CLS_FLAGS_IN_HW
;
530 if (tc_skip_sw(flags
))
536 static void u32_clear_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
)
538 struct tc_u_knode
*n
;
541 for (h
= 0; h
<= ht
->divisor
; h
++) {
542 while ((n
= rtnl_dereference(ht
->ht
[h
])) != NULL
) {
543 RCU_INIT_POINTER(ht
->ht
[h
],
544 rtnl_dereference(n
->next
));
545 tcf_unbind_filter(tp
, &n
->res
);
546 u32_remove_hw_knode(tp
, n
->handle
);
547 call_rcu(&n
->rcu
, u32_delete_key_freepf_rcu
);
552 static int u32_destroy_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
)
554 struct tc_u_common
*tp_c
= tp
->data
;
555 struct tc_u_hnode __rcu
**hn
;
556 struct tc_u_hnode
*phn
;
560 u32_clear_hnode(tp
, ht
);
563 for (phn
= rtnl_dereference(*hn
);
565 hn
= &phn
->next
, phn
= rtnl_dereference(*hn
)) {
567 u32_clear_hw_hnode(tp
, ht
);
568 RCU_INIT_POINTER(*hn
, ht
->next
);
577 static bool ht_empty(struct tc_u_hnode
*ht
)
581 for (h
= 0; h
<= ht
->divisor
; h
++)
582 if (rcu_access_pointer(ht
->ht
[h
]))
588 static bool u32_destroy(struct tcf_proto
*tp
, bool force
)
590 struct tc_u_common
*tp_c
= tp
->data
;
591 struct tc_u_hnode
*root_ht
= rtnl_dereference(tp
->root
);
593 WARN_ON(root_ht
== NULL
);
597 if (root_ht
->refcnt
> 1)
599 if (root_ht
->refcnt
== 1) {
600 if (!ht_empty(root_ht
))
605 if (tp_c
->refcnt
> 1)
608 if (tp_c
->refcnt
== 1) {
609 struct tc_u_hnode
*ht
;
611 for (ht
= rtnl_dereference(tp_c
->hlist
);
613 ht
= rtnl_dereference(ht
->next
))
619 if (root_ht
&& --root_ht
->refcnt
== 0)
620 u32_destroy_hnode(tp
, root_ht
);
622 if (--tp_c
->refcnt
== 0) {
623 struct tc_u_hnode
*ht
;
625 tp
->q
->u32_node
= NULL
;
627 for (ht
= rtnl_dereference(tp_c
->hlist
);
629 ht
= rtnl_dereference(ht
->next
)) {
631 u32_clear_hnode(tp
, ht
);
634 while ((ht
= rtnl_dereference(tp_c
->hlist
)) != NULL
) {
635 RCU_INIT_POINTER(tp_c
->hlist
, ht
->next
);
646 static int u32_delete(struct tcf_proto
*tp
, unsigned long arg
)
648 struct tc_u_hnode
*ht
= (struct tc_u_hnode
*)arg
;
649 struct tc_u_hnode
*root_ht
= rtnl_dereference(tp
->root
);
654 if (TC_U32_KEY(ht
->handle
)) {
655 u32_remove_hw_knode(tp
, ht
->handle
);
656 return u32_delete_key(tp
, (struct tc_u_knode
*)ht
);
662 if (ht
->refcnt
== 1) {
664 u32_destroy_hnode(tp
, ht
);
672 #define NR_U32_NODE (1<<12)
673 static u32
gen_new_kid(struct tc_u_hnode
*ht
, u32 handle
)
675 struct tc_u_knode
*n
;
677 unsigned long *bitmap
= kzalloc(BITS_TO_LONGS(NR_U32_NODE
) * sizeof(unsigned long),
680 return handle
| 0xFFF;
682 for (n
= rtnl_dereference(ht
->ht
[TC_U32_HASH(handle
)]);
684 n
= rtnl_dereference(n
->next
))
685 set_bit(TC_U32_NODE(n
->handle
), bitmap
);
687 i
= find_next_zero_bit(bitmap
, NR_U32_NODE
, 0x800);
688 if (i
>= NR_U32_NODE
)
689 i
= find_next_zero_bit(bitmap
, NR_U32_NODE
, 1);
692 return handle
| (i
>= NR_U32_NODE
? 0xFFF : i
);
695 static const struct nla_policy u32_policy
[TCA_U32_MAX
+ 1] = {
696 [TCA_U32_CLASSID
] = { .type
= NLA_U32
},
697 [TCA_U32_HASH
] = { .type
= NLA_U32
},
698 [TCA_U32_LINK
] = { .type
= NLA_U32
},
699 [TCA_U32_DIVISOR
] = { .type
= NLA_U32
},
700 [TCA_U32_SEL
] = { .len
= sizeof(struct tc_u32_sel
) },
701 [TCA_U32_INDEV
] = { .type
= NLA_STRING
, .len
= IFNAMSIZ
},
702 [TCA_U32_MARK
] = { .len
= sizeof(struct tc_u32_mark
) },
703 [TCA_U32_FLAGS
] = { .type
= NLA_U32
},
706 static int u32_set_parms(struct net
*net
, struct tcf_proto
*tp
,
707 unsigned long base
, struct tc_u_hnode
*ht
,
708 struct tc_u_knode
*n
, struct nlattr
**tb
,
709 struct nlattr
*est
, bool ovr
)
714 err
= tcf_exts_init(&e
, TCA_U32_ACT
, TCA_U32_POLICE
);
717 err
= tcf_exts_validate(net
, tp
, tb
, est
, &e
, ovr
);
722 if (tb
[TCA_U32_LINK
]) {
723 u32 handle
= nla_get_u32(tb
[TCA_U32_LINK
]);
724 struct tc_u_hnode
*ht_down
= NULL
, *ht_old
;
726 if (TC_U32_KEY(handle
))
730 ht_down
= u32_lookup_ht(ht
->tp_c
, handle
);
737 ht_old
= rtnl_dereference(n
->ht_down
);
738 rcu_assign_pointer(n
->ht_down
, ht_down
);
743 if (tb
[TCA_U32_CLASSID
]) {
744 n
->res
.classid
= nla_get_u32(tb
[TCA_U32_CLASSID
]);
745 tcf_bind_filter(tp
, &n
->res
, base
);
748 #ifdef CONFIG_NET_CLS_IND
749 if (tb
[TCA_U32_INDEV
]) {
751 ret
= tcf_change_indev(net
, tb
[TCA_U32_INDEV
]);
757 tcf_exts_change(tp
, &n
->exts
, &e
);
761 tcf_exts_destroy(&e
);
765 static void u32_replace_knode(struct tcf_proto
*tp
, struct tc_u_common
*tp_c
,
766 struct tc_u_knode
*n
)
768 struct tc_u_knode __rcu
**ins
;
769 struct tc_u_knode
*pins
;
770 struct tc_u_hnode
*ht
;
772 if (TC_U32_HTID(n
->handle
) == TC_U32_ROOT
)
773 ht
= rtnl_dereference(tp
->root
);
775 ht
= u32_lookup_ht(tp_c
, TC_U32_HTID(n
->handle
));
777 ins
= &ht
->ht
[TC_U32_HASH(n
->handle
)];
779 /* The node must always exist for it to be replaced if this is not the
780 * case then something went very wrong elsewhere.
782 for (pins
= rtnl_dereference(*ins
); ;
783 ins
= &pins
->next
, pins
= rtnl_dereference(*ins
))
784 if (pins
->handle
== n
->handle
)
787 RCU_INIT_POINTER(n
->next
, pins
->next
);
788 rcu_assign_pointer(*ins
, n
);
791 static struct tc_u_knode
*u32_init_knode(struct tcf_proto
*tp
,
792 struct tc_u_knode
*n
)
794 struct tc_u_knode
*new;
795 struct tc_u32_sel
*s
= &n
->sel
;
797 new = kzalloc(sizeof(*n
) + s
->nkeys
*sizeof(struct tc_u32_key
),
803 RCU_INIT_POINTER(new->next
, n
->next
);
804 new->handle
= n
->handle
;
805 RCU_INIT_POINTER(new->ht_up
, n
->ht_up
);
807 #ifdef CONFIG_NET_CLS_IND
808 new->ifindex
= n
->ifindex
;
810 new->fshift
= n
->fshift
;
812 new->flags
= n
->flags
;
813 RCU_INIT_POINTER(new->ht_down
, n
->ht_down
);
815 /* bump reference count as long as we hold pointer to structure */
817 new->ht_down
->refcnt
++;
819 #ifdef CONFIG_CLS_U32_PERF
820 /* Statistics may be incremented by readers during update
821 * so we must keep them in tact. When the node is later destroyed
822 * a special destroy call must be made to not free the pf memory.
827 #ifdef CONFIG_CLS_U32_MARK
830 /* Similarly success statistics must be moved as pointers */
831 new->pcpu_success
= n
->pcpu_success
;
834 memcpy(&new->sel
, s
, sizeof(*s
) + s
->nkeys
*sizeof(struct tc_u32_key
));
836 if (tcf_exts_init(&new->exts
, TCA_U32_ACT
, TCA_U32_POLICE
)) {
844 static int u32_change(struct net
*net
, struct sk_buff
*in_skb
,
845 struct tcf_proto
*tp
, unsigned long base
, u32 handle
,
846 struct nlattr
**tca
, unsigned long *arg
, bool ovr
)
848 struct tc_u_common
*tp_c
= tp
->data
;
849 struct tc_u_hnode
*ht
;
850 struct tc_u_knode
*n
;
851 struct tc_u32_sel
*s
;
852 struct nlattr
*opt
= tca
[TCA_OPTIONS
];
853 struct nlattr
*tb
[TCA_U32_MAX
+ 1];
856 #ifdef CONFIG_CLS_U32_PERF
861 return handle
? -EINVAL
: 0;
863 err
= nla_parse_nested(tb
, TCA_U32_MAX
, opt
, u32_policy
);
867 if (tb
[TCA_U32_FLAGS
]) {
868 flags
= nla_get_u32(tb
[TCA_U32_FLAGS
]);
869 if (!tc_flags_valid(flags
))
873 n
= (struct tc_u_knode
*)*arg
;
875 struct tc_u_knode
*new;
877 if (TC_U32_KEY(n
->handle
) == 0)
880 if (n
->flags
!= flags
)
883 new = u32_init_knode(tp
, n
);
887 err
= u32_set_parms(net
, tp
, base
,
888 rtnl_dereference(n
->ht_up
), new, tb
,
892 u32_destroy_key(tp
, new, false);
896 err
= u32_replace_hw_knode(tp
, new, flags
);
898 u32_destroy_key(tp
, new, false);
902 if (!tc_in_hw(new->flags
))
903 new->flags
|= TCA_CLS_FLAGS_NOT_IN_HW
;
905 u32_replace_knode(tp
, tp_c
, new);
906 tcf_unbind_filter(tp
, &n
->res
);
907 call_rcu(&n
->rcu
, u32_delete_key_rcu
);
911 if (tb
[TCA_U32_DIVISOR
]) {
912 unsigned int divisor
= nla_get_u32(tb
[TCA_U32_DIVISOR
]);
914 if (--divisor
> 0x100)
916 if (TC_U32_KEY(handle
))
919 handle
= gen_new_htid(tp
->data
);
923 ht
= kzalloc(sizeof(*ht
) + divisor
*sizeof(void *), GFP_KERNEL
);
928 ht
->divisor
= divisor
;
932 err
= u32_replace_hw_hnode(tp
, ht
, flags
);
938 RCU_INIT_POINTER(ht
->next
, tp_c
->hlist
);
939 rcu_assign_pointer(tp_c
->hlist
, ht
);
940 *arg
= (unsigned long)ht
;
945 if (tb
[TCA_U32_HASH
]) {
946 htid
= nla_get_u32(tb
[TCA_U32_HASH
]);
947 if (TC_U32_HTID(htid
) == TC_U32_ROOT
) {
948 ht
= rtnl_dereference(tp
->root
);
951 ht
= u32_lookup_ht(tp
->data
, TC_U32_HTID(htid
));
956 ht
= rtnl_dereference(tp
->root
);
960 if (ht
->divisor
< TC_U32_HASH(htid
))
964 if (TC_U32_HTID(handle
) && TC_U32_HTID(handle
^htid
))
966 handle
= htid
| TC_U32_NODE(handle
);
968 handle
= gen_new_kid(ht
, htid
);
970 if (tb
[TCA_U32_SEL
] == NULL
)
973 s
= nla_data(tb
[TCA_U32_SEL
]);
975 n
= kzalloc(sizeof(*n
) + s
->nkeys
*sizeof(struct tc_u32_key
), GFP_KERNEL
);
979 #ifdef CONFIG_CLS_U32_PERF
980 size
= sizeof(struct tc_u32_pcnt
) + s
->nkeys
* sizeof(u64
);
981 n
->pf
= __alloc_percpu(size
, __alignof__(struct tc_u32_pcnt
));
988 memcpy(&n
->sel
, s
, sizeof(*s
) + s
->nkeys
*sizeof(struct tc_u32_key
));
989 RCU_INIT_POINTER(n
->ht_up
, ht
);
991 n
->fshift
= s
->hmask
? ffs(ntohl(s
->hmask
)) - 1 : 0;
995 err
= tcf_exts_init(&n
->exts
, TCA_U32_ACT
, TCA_U32_POLICE
);
999 #ifdef CONFIG_CLS_U32_MARK
1000 n
->pcpu_success
= alloc_percpu(u32
);
1001 if (!n
->pcpu_success
) {
1006 if (tb
[TCA_U32_MARK
]) {
1007 struct tc_u32_mark
*mark
;
1009 mark
= nla_data(tb
[TCA_U32_MARK
]);
1011 n
->mask
= mark
->mask
;
1015 err
= u32_set_parms(net
, tp
, base
, ht
, n
, tb
, tca
[TCA_RATE
], ovr
);
1017 struct tc_u_knode __rcu
**ins
;
1018 struct tc_u_knode
*pins
;
1020 err
= u32_replace_hw_knode(tp
, n
, flags
);
1024 if (!tc_in_hw(n
->flags
))
1025 n
->flags
|= TCA_CLS_FLAGS_NOT_IN_HW
;
1027 ins
= &ht
->ht
[TC_U32_HASH(handle
)];
1028 for (pins
= rtnl_dereference(*ins
); pins
;
1029 ins
= &pins
->next
, pins
= rtnl_dereference(*ins
))
1030 if (TC_U32_NODE(handle
) < TC_U32_NODE(pins
->handle
))
1033 RCU_INIT_POINTER(n
->next
, pins
);
1034 rcu_assign_pointer(*ins
, n
);
1035 *arg
= (unsigned long)n
;
1040 #ifdef CONFIG_CLS_U32_MARK
1041 free_percpu(n
->pcpu_success
);
1045 tcf_exts_destroy(&n
->exts
);
1046 #ifdef CONFIG_CLS_U32_PERF
1053 static void u32_walk(struct tcf_proto
*tp
, struct tcf_walker
*arg
)
1055 struct tc_u_common
*tp_c
= tp
->data
;
1056 struct tc_u_hnode
*ht
;
1057 struct tc_u_knode
*n
;
1063 for (ht
= rtnl_dereference(tp_c
->hlist
);
1065 ht
= rtnl_dereference(ht
->next
)) {
1066 if (ht
->prio
!= tp
->prio
)
1068 if (arg
->count
>= arg
->skip
) {
1069 if (arg
->fn(tp
, (unsigned long)ht
, arg
) < 0) {
1075 for (h
= 0; h
<= ht
->divisor
; h
++) {
1076 for (n
= rtnl_dereference(ht
->ht
[h
]);
1078 n
= rtnl_dereference(n
->next
)) {
1079 if (arg
->count
< arg
->skip
) {
1083 if (arg
->fn(tp
, (unsigned long)n
, arg
) < 0) {
1093 static int u32_dump(struct net
*net
, struct tcf_proto
*tp
, unsigned long fh
,
1094 struct sk_buff
*skb
, struct tcmsg
*t
)
1096 struct tc_u_knode
*n
= (struct tc_u_knode
*)fh
;
1097 struct tc_u_hnode
*ht_up
, *ht_down
;
1098 struct nlattr
*nest
;
1103 t
->tcm_handle
= n
->handle
;
1105 nest
= nla_nest_start(skb
, TCA_OPTIONS
);
1107 goto nla_put_failure
;
1109 if (TC_U32_KEY(n
->handle
) == 0) {
1110 struct tc_u_hnode
*ht
= (struct tc_u_hnode
*)fh
;
1111 u32 divisor
= ht
->divisor
+ 1;
1113 if (nla_put_u32(skb
, TCA_U32_DIVISOR
, divisor
))
1114 goto nla_put_failure
;
1116 #ifdef CONFIG_CLS_U32_PERF
1117 struct tc_u32_pcnt
*gpf
;
1121 if (nla_put(skb
, TCA_U32_SEL
,
1122 sizeof(n
->sel
) + n
->sel
.nkeys
*sizeof(struct tc_u32_key
),
1124 goto nla_put_failure
;
1126 ht_up
= rtnl_dereference(n
->ht_up
);
1128 u32 htid
= n
->handle
& 0xFFFFF000;
1129 if (nla_put_u32(skb
, TCA_U32_HASH
, htid
))
1130 goto nla_put_failure
;
1132 if (n
->res
.classid
&&
1133 nla_put_u32(skb
, TCA_U32_CLASSID
, n
->res
.classid
))
1134 goto nla_put_failure
;
1136 ht_down
= rtnl_dereference(n
->ht_down
);
1138 nla_put_u32(skb
, TCA_U32_LINK
, ht_down
->handle
))
1139 goto nla_put_failure
;
1141 if (n
->flags
&& nla_put_u32(skb
, TCA_U32_FLAGS
, n
->flags
))
1142 goto nla_put_failure
;
1144 #ifdef CONFIG_CLS_U32_MARK
1145 if ((n
->val
|| n
->mask
)) {
1146 struct tc_u32_mark mark
= {.val
= n
->val
,
1151 for_each_possible_cpu(cpum
) {
1152 __u32 cnt
= *per_cpu_ptr(n
->pcpu_success
, cpum
);
1154 mark
.success
+= cnt
;
1157 if (nla_put(skb
, TCA_U32_MARK
, sizeof(mark
), &mark
))
1158 goto nla_put_failure
;
1162 if (tcf_exts_dump(skb
, &n
->exts
) < 0)
1163 goto nla_put_failure
;
1165 #ifdef CONFIG_NET_CLS_IND
1167 struct net_device
*dev
;
1168 dev
= __dev_get_by_index(net
, n
->ifindex
);
1169 if (dev
&& nla_put_string(skb
, TCA_U32_INDEV
, dev
->name
))
1170 goto nla_put_failure
;
1173 #ifdef CONFIG_CLS_U32_PERF
1174 gpf
= kzalloc(sizeof(struct tc_u32_pcnt
) +
1175 n
->sel
.nkeys
* sizeof(u64
),
1178 goto nla_put_failure
;
1180 for_each_possible_cpu(cpu
) {
1182 struct tc_u32_pcnt
*pf
= per_cpu_ptr(n
->pf
, cpu
);
1184 gpf
->rcnt
+= pf
->rcnt
;
1185 gpf
->rhit
+= pf
->rhit
;
1186 for (i
= 0; i
< n
->sel
.nkeys
; i
++)
1187 gpf
->kcnts
[i
] += pf
->kcnts
[i
];
1190 if (nla_put_64bit(skb
, TCA_U32_PCNT
,
1191 sizeof(struct tc_u32_pcnt
) +
1192 n
->sel
.nkeys
* sizeof(u64
),
1193 gpf
, TCA_U32_PAD
)) {
1195 goto nla_put_failure
;
1201 nla_nest_end(skb
, nest
);
1203 if (TC_U32_KEY(n
->handle
))
1204 if (tcf_exts_dump_stats(skb
, &n
->exts
) < 0)
1205 goto nla_put_failure
;
1209 nla_nest_cancel(skb
, nest
);
1213 static struct tcf_proto_ops cls_u32_ops __read_mostly
= {
1215 .classify
= u32_classify
,
1217 .destroy
= u32_destroy
,
1219 .change
= u32_change
,
1220 .delete = u32_delete
,
1223 .owner
= THIS_MODULE
,
1226 static int __init
init_u32(void)
1228 pr_info("u32 classifier\n");
1229 #ifdef CONFIG_CLS_U32_PERF
1230 pr_info(" Performance counters on\n");
1232 #ifdef CONFIG_NET_CLS_IND
1233 pr_info(" input device check on\n");
1235 #ifdef CONFIG_NET_CLS_ACT
1236 pr_info(" Actions configured\n");
1238 return register_tcf_proto_ops(&cls_u32_ops
);
1241 static void __exit
exit_u32(void)
1243 unregister_tcf_proto_ops(&cls_u32_ops
);
1246 module_init(init_u32
)
1247 module_exit(exit_u32
)
1248 MODULE_LICENSE("GPL");