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 #ifdef CONFIG_CLS_U32_MARK
138 if ((skb
->mark
& n
->mask
) != n
->val
) {
139 n
= rcu_dereference_bh(n
->next
);
142 __this_cpu_inc(*n
->pcpu_success
);
146 for (i
= n
->sel
.nkeys
; i
> 0; i
--, key
++) {
147 int toff
= off
+ key
->off
+ (off2
& key
->offmask
);
150 if (skb_headroom(skb
) + toff
> INT_MAX
)
153 data
= skb_header_pointer(skb
, toff
, 4, &hdata
);
156 if ((*data
^ key
->val
) & key
->mask
) {
157 n
= rcu_dereference_bh(n
->next
);
160 #ifdef CONFIG_CLS_U32_PERF
161 __this_cpu_inc(n
->pf
->kcnts
[j
]);
166 ht
= rcu_dereference_bh(n
->ht_down
);
169 if (n
->sel
.flags
& TC_U32_TERMINAL
) {
172 #ifdef CONFIG_NET_CLS_IND
173 if (!tcf_match_indev(skb
, n
->ifindex
)) {
174 n
= rcu_dereference_bh(n
->next
);
178 #ifdef CONFIG_CLS_U32_PERF
179 __this_cpu_inc(n
->pf
->rhit
);
181 r
= tcf_exts_exec(skb
, &n
->exts
, res
);
183 n
= rcu_dereference_bh(n
->next
);
189 n
= rcu_dereference_bh(n
->next
);
194 if (sdepth
>= TC_U32_MAXDEPTH
)
196 stack
[sdepth
].knode
= n
;
197 stack
[sdepth
].off
= off
;
200 ht
= rcu_dereference_bh(n
->ht_down
);
205 data
= skb_header_pointer(skb
, off
+ n
->sel
.hoff
, 4,
209 sel
= ht
->divisor
& u32_hash_fold(*data
, &n
->sel
,
212 if (!(n
->sel
.flags
& (TC_U32_VAROFFSET
| TC_U32_OFFSET
| TC_U32_EAT
)))
215 if (n
->sel
.flags
& (TC_U32_OFFSET
| TC_U32_VAROFFSET
)) {
216 off2
= n
->sel
.off
+ 3;
217 if (n
->sel
.flags
& TC_U32_VAROFFSET
) {
220 data
= skb_header_pointer(skb
,
225 off2
+= ntohs(n
->sel
.offmask
& *data
) >>
230 if (n
->sel
.flags
& TC_U32_EAT
) {
241 n
= stack
[sdepth
].knode
;
242 ht
= rcu_dereference_bh(n
->ht_up
);
243 off
= stack
[sdepth
].off
;
250 net_warn_ratelimited("cls_u32: dead loop\n");
254 static struct tc_u_hnode
*
255 u32_lookup_ht(struct tc_u_common
*tp_c
, u32 handle
)
257 struct tc_u_hnode
*ht
;
259 for (ht
= rtnl_dereference(tp_c
->hlist
);
261 ht
= rtnl_dereference(ht
->next
))
262 if (ht
->handle
== handle
)
268 static struct tc_u_knode
*
269 u32_lookup_key(struct tc_u_hnode
*ht
, u32 handle
)
272 struct tc_u_knode
*n
= NULL
;
274 sel
= TC_U32_HASH(handle
);
275 if (sel
> ht
->divisor
)
278 for (n
= rtnl_dereference(ht
->ht
[sel
]);
280 n
= rtnl_dereference(n
->next
))
281 if (n
->handle
== handle
)
288 static unsigned long u32_get(struct tcf_proto
*tp
, u32 handle
)
290 struct tc_u_hnode
*ht
;
291 struct tc_u_common
*tp_c
= tp
->data
;
293 if (TC_U32_HTID(handle
) == TC_U32_ROOT
)
294 ht
= rtnl_dereference(tp
->root
);
296 ht
= u32_lookup_ht(tp_c
, TC_U32_HTID(handle
));
301 if (TC_U32_KEY(handle
) == 0)
302 return (unsigned long)ht
;
304 return (unsigned long)u32_lookup_key(ht
, handle
);
307 static u32
gen_new_htid(struct tc_u_common
*tp_c
)
311 /* hgenerator only used inside rtnl lock it is safe to increment
312 * without read _copy_ update semantics
315 if (++tp_c
->hgenerator
== 0x7FF)
316 tp_c
->hgenerator
= 1;
317 } while (--i
> 0 && u32_lookup_ht(tp_c
, (tp_c
->hgenerator
|0x800)<<20));
319 return i
> 0 ? (tp_c
->hgenerator
|0x800)<<20 : 0;
322 static int u32_init(struct tcf_proto
*tp
)
324 struct tc_u_hnode
*root_ht
;
325 struct tc_u_common
*tp_c
;
327 tp_c
= tp
->q
->u32_node
;
329 root_ht
= kzalloc(sizeof(*root_ht
), GFP_KERNEL
);
333 root_ht
->divisor
= 0;
335 root_ht
->handle
= tp_c
? gen_new_htid(tp_c
) : 0x80000000;
336 root_ht
->prio
= tp
->prio
;
339 tp_c
= kzalloc(sizeof(*tp_c
), GFP_KERNEL
);
345 tp
->q
->u32_node
= tp_c
;
349 RCU_INIT_POINTER(root_ht
->next
, tp_c
->hlist
);
350 rcu_assign_pointer(tp_c
->hlist
, root_ht
);
351 root_ht
->tp_c
= tp_c
;
353 rcu_assign_pointer(tp
->root
, root_ht
);
358 static int u32_destroy_key(struct tcf_proto
*tp
,
359 struct tc_u_knode
*n
,
362 tcf_exts_destroy(&n
->exts
);
364 n
->ht_down
->refcnt
--;
365 #ifdef CONFIG_CLS_U32_PERF
369 #ifdef CONFIG_CLS_U32_MARK
371 free_percpu(n
->pcpu_success
);
377 /* u32_delete_key_rcu should be called when free'ing a copied
378 * version of a tc_u_knode obtained from u32_init_knode(). When
379 * copies are obtained from u32_init_knode() the statistics are
380 * shared between the old and new copies to allow readers to
381 * continue to update the statistics during the copy. To support
382 * this the u32_delete_key_rcu variant does not free the percpu
385 static void u32_delete_key_rcu(struct rcu_head
*rcu
)
387 struct tc_u_knode
*key
= container_of(rcu
, struct tc_u_knode
, rcu
);
389 u32_destroy_key(key
->tp
, key
, false);
392 /* u32_delete_key_freepf_rcu is the rcu callback variant
393 * that free's the entire structure including the statistics
394 * percpu variables. Only use this if the key is not a copy
395 * returned by u32_init_knode(). See u32_delete_key_rcu()
396 * for the variant that should be used with keys return from
399 static void u32_delete_key_freepf_rcu(struct rcu_head
*rcu
)
401 struct tc_u_knode
*key
= container_of(rcu
, struct tc_u_knode
, rcu
);
403 u32_destroy_key(key
->tp
, key
, true);
406 static int u32_delete_key(struct tcf_proto
*tp
, struct tc_u_knode
*key
)
408 struct tc_u_knode __rcu
**kp
;
409 struct tc_u_knode
*pkp
;
410 struct tc_u_hnode
*ht
= rtnl_dereference(key
->ht_up
);
413 kp
= &ht
->ht
[TC_U32_HASH(key
->handle
)];
414 for (pkp
= rtnl_dereference(*kp
); pkp
;
415 kp
= &pkp
->next
, pkp
= rtnl_dereference(*kp
)) {
417 RCU_INIT_POINTER(*kp
, key
->next
);
419 tcf_unbind_filter(tp
, &key
->res
);
420 call_rcu(&key
->rcu
, u32_delete_key_freepf_rcu
);
429 static void u32_remove_hw_knode(struct tcf_proto
*tp
, u32 handle
)
431 struct net_device
*dev
= tp
->q
->dev_queue
->dev
;
432 struct tc_cls_u32_offload u32_offload
= {0};
433 struct tc_to_netdev offload
;
435 offload
.type
= TC_SETUP_CLSU32
;
436 offload
.cls_u32
= &u32_offload
;
438 if (tc_should_offload(dev
, 0)) {
439 offload
.cls_u32
->command
= TC_CLSU32_DELETE_KNODE
;
440 offload
.cls_u32
->knode
.handle
= handle
;
441 dev
->netdev_ops
->ndo_setup_tc(dev
, tp
->q
->handle
,
442 tp
->protocol
, &offload
);
446 static void u32_replace_hw_hnode(struct tcf_proto
*tp
,
447 struct tc_u_hnode
*h
,
450 struct net_device
*dev
= tp
->q
->dev_queue
->dev
;
451 struct tc_cls_u32_offload u32_offload
= {0};
452 struct tc_to_netdev offload
;
454 offload
.type
= TC_SETUP_CLSU32
;
455 offload
.cls_u32
= &u32_offload
;
457 if (tc_should_offload(dev
, flags
)) {
458 offload
.cls_u32
->command
= TC_CLSU32_NEW_HNODE
;
459 offload
.cls_u32
->hnode
.divisor
= h
->divisor
;
460 offload
.cls_u32
->hnode
.handle
= h
->handle
;
461 offload
.cls_u32
->hnode
.prio
= h
->prio
;
463 dev
->netdev_ops
->ndo_setup_tc(dev
, tp
->q
->handle
,
464 tp
->protocol
, &offload
);
468 static void u32_clear_hw_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*h
)
470 struct net_device
*dev
= tp
->q
->dev_queue
->dev
;
471 struct tc_cls_u32_offload u32_offload
= {0};
472 struct tc_to_netdev offload
;
474 offload
.type
= TC_SETUP_CLSU32
;
475 offload
.cls_u32
= &u32_offload
;
477 if (tc_should_offload(dev
, 0)) {
478 offload
.cls_u32
->command
= TC_CLSU32_DELETE_HNODE
;
479 offload
.cls_u32
->hnode
.divisor
= h
->divisor
;
480 offload
.cls_u32
->hnode
.handle
= h
->handle
;
481 offload
.cls_u32
->hnode
.prio
= h
->prio
;
483 dev
->netdev_ops
->ndo_setup_tc(dev
, tp
->q
->handle
,
484 tp
->protocol
, &offload
);
488 static void u32_replace_hw_knode(struct tcf_proto
*tp
,
489 struct tc_u_knode
*n
,
492 struct net_device
*dev
= tp
->q
->dev_queue
->dev
;
493 struct tc_cls_u32_offload u32_offload
= {0};
494 struct tc_to_netdev offload
;
496 offload
.type
= TC_SETUP_CLSU32
;
497 offload
.cls_u32
= &u32_offload
;
499 if (tc_should_offload(dev
, flags
)) {
500 offload
.cls_u32
->command
= TC_CLSU32_REPLACE_KNODE
;
501 offload
.cls_u32
->knode
.handle
= n
->handle
;
502 offload
.cls_u32
->knode
.fshift
= n
->fshift
;
503 #ifdef CONFIG_CLS_U32_MARK
504 offload
.cls_u32
->knode
.val
= n
->val
;
505 offload
.cls_u32
->knode
.mask
= n
->mask
;
507 offload
.cls_u32
->knode
.val
= 0;
508 offload
.cls_u32
->knode
.mask
= 0;
510 offload
.cls_u32
->knode
.sel
= &n
->sel
;
511 offload
.cls_u32
->knode
.exts
= &n
->exts
;
513 offload
.cls_u32
->knode
.link_handle
= n
->ht_down
->handle
;
515 dev
->netdev_ops
->ndo_setup_tc(dev
, tp
->q
->handle
,
516 tp
->protocol
, &offload
);
520 static void u32_clear_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
)
522 struct tc_u_knode
*n
;
525 for (h
= 0; h
<= ht
->divisor
; h
++) {
526 while ((n
= rtnl_dereference(ht
->ht
[h
])) != NULL
) {
527 RCU_INIT_POINTER(ht
->ht
[h
],
528 rtnl_dereference(n
->next
));
529 tcf_unbind_filter(tp
, &n
->res
);
530 u32_remove_hw_knode(tp
, n
->handle
);
531 call_rcu(&n
->rcu
, u32_delete_key_freepf_rcu
);
536 static int u32_destroy_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
)
538 struct tc_u_common
*tp_c
= tp
->data
;
539 struct tc_u_hnode __rcu
**hn
;
540 struct tc_u_hnode
*phn
;
544 u32_clear_hnode(tp
, ht
);
547 for (phn
= rtnl_dereference(*hn
);
549 hn
= &phn
->next
, phn
= rtnl_dereference(*hn
)) {
551 u32_clear_hw_hnode(tp
, ht
);
552 RCU_INIT_POINTER(*hn
, ht
->next
);
561 static bool ht_empty(struct tc_u_hnode
*ht
)
565 for (h
= 0; h
<= ht
->divisor
; h
++)
566 if (rcu_access_pointer(ht
->ht
[h
]))
572 static bool u32_destroy(struct tcf_proto
*tp
, bool force
)
574 struct tc_u_common
*tp_c
= tp
->data
;
575 struct tc_u_hnode
*root_ht
= rtnl_dereference(tp
->root
);
577 WARN_ON(root_ht
== NULL
);
581 if (root_ht
->refcnt
> 1)
583 if (root_ht
->refcnt
== 1) {
584 if (!ht_empty(root_ht
))
589 if (tp_c
->refcnt
> 1)
592 if (tp_c
->refcnt
== 1) {
593 struct tc_u_hnode
*ht
;
595 for (ht
= rtnl_dereference(tp_c
->hlist
);
597 ht
= rtnl_dereference(ht
->next
))
603 if (root_ht
&& --root_ht
->refcnt
== 0)
604 u32_destroy_hnode(tp
, root_ht
);
606 if (--tp_c
->refcnt
== 0) {
607 struct tc_u_hnode
*ht
;
609 tp
->q
->u32_node
= NULL
;
611 for (ht
= rtnl_dereference(tp_c
->hlist
);
613 ht
= rtnl_dereference(ht
->next
)) {
615 u32_clear_hnode(tp
, ht
);
618 while ((ht
= rtnl_dereference(tp_c
->hlist
)) != NULL
) {
619 RCU_INIT_POINTER(tp_c
->hlist
, ht
->next
);
630 static int u32_delete(struct tcf_proto
*tp
, unsigned long arg
)
632 struct tc_u_hnode
*ht
= (struct tc_u_hnode
*)arg
;
633 struct tc_u_hnode
*root_ht
= rtnl_dereference(tp
->root
);
638 if (TC_U32_KEY(ht
->handle
)) {
639 u32_remove_hw_knode(tp
, ht
->handle
);
640 return u32_delete_key(tp
, (struct tc_u_knode
*)ht
);
646 if (ht
->refcnt
== 1) {
648 u32_destroy_hnode(tp
, ht
);
656 #define NR_U32_NODE (1<<12)
657 static u32
gen_new_kid(struct tc_u_hnode
*ht
, u32 handle
)
659 struct tc_u_knode
*n
;
661 unsigned long *bitmap
= kzalloc(BITS_TO_LONGS(NR_U32_NODE
) * sizeof(unsigned long),
664 return handle
| 0xFFF;
666 for (n
= rtnl_dereference(ht
->ht
[TC_U32_HASH(handle
)]);
668 n
= rtnl_dereference(n
->next
))
669 set_bit(TC_U32_NODE(n
->handle
), bitmap
);
671 i
= find_next_zero_bit(bitmap
, NR_U32_NODE
, 0x800);
672 if (i
>= NR_U32_NODE
)
673 i
= find_next_zero_bit(bitmap
, NR_U32_NODE
, 1);
676 return handle
| (i
>= NR_U32_NODE
? 0xFFF : i
);
679 static const struct nla_policy u32_policy
[TCA_U32_MAX
+ 1] = {
680 [TCA_U32_CLASSID
] = { .type
= NLA_U32
},
681 [TCA_U32_HASH
] = { .type
= NLA_U32
},
682 [TCA_U32_LINK
] = { .type
= NLA_U32
},
683 [TCA_U32_DIVISOR
] = { .type
= NLA_U32
},
684 [TCA_U32_SEL
] = { .len
= sizeof(struct tc_u32_sel
) },
685 [TCA_U32_INDEV
] = { .type
= NLA_STRING
, .len
= IFNAMSIZ
},
686 [TCA_U32_MARK
] = { .len
= sizeof(struct tc_u32_mark
) },
687 [TCA_U32_FLAGS
] = { .type
= NLA_U32
},
690 static int u32_set_parms(struct net
*net
, struct tcf_proto
*tp
,
691 unsigned long base
, struct tc_u_hnode
*ht
,
692 struct tc_u_knode
*n
, struct nlattr
**tb
,
693 struct nlattr
*est
, bool ovr
)
698 tcf_exts_init(&e
, TCA_U32_ACT
, TCA_U32_POLICE
);
699 err
= tcf_exts_validate(net
, tp
, tb
, est
, &e
, ovr
);
704 if (tb
[TCA_U32_LINK
]) {
705 u32 handle
= nla_get_u32(tb
[TCA_U32_LINK
]);
706 struct tc_u_hnode
*ht_down
= NULL
, *ht_old
;
708 if (TC_U32_KEY(handle
))
712 ht_down
= u32_lookup_ht(ht
->tp_c
, handle
);
719 ht_old
= rtnl_dereference(n
->ht_down
);
720 rcu_assign_pointer(n
->ht_down
, ht_down
);
725 if (tb
[TCA_U32_CLASSID
]) {
726 n
->res
.classid
= nla_get_u32(tb
[TCA_U32_CLASSID
]);
727 tcf_bind_filter(tp
, &n
->res
, base
);
730 #ifdef CONFIG_NET_CLS_IND
731 if (tb
[TCA_U32_INDEV
]) {
733 ret
= tcf_change_indev(net
, tb
[TCA_U32_INDEV
]);
739 tcf_exts_change(tp
, &n
->exts
, &e
);
743 tcf_exts_destroy(&e
);
747 static void u32_replace_knode(struct tcf_proto
*tp
,
748 struct tc_u_common
*tp_c
,
749 struct tc_u_knode
*n
)
751 struct tc_u_knode __rcu
**ins
;
752 struct tc_u_knode
*pins
;
753 struct tc_u_hnode
*ht
;
755 if (TC_U32_HTID(n
->handle
) == TC_U32_ROOT
)
756 ht
= rtnl_dereference(tp
->root
);
758 ht
= u32_lookup_ht(tp_c
, TC_U32_HTID(n
->handle
));
760 ins
= &ht
->ht
[TC_U32_HASH(n
->handle
)];
762 /* The node must always exist for it to be replaced if this is not the
763 * case then something went very wrong elsewhere.
765 for (pins
= rtnl_dereference(*ins
); ;
766 ins
= &pins
->next
, pins
= rtnl_dereference(*ins
))
767 if (pins
->handle
== n
->handle
)
770 RCU_INIT_POINTER(n
->next
, pins
->next
);
771 rcu_assign_pointer(*ins
, n
);
774 static struct tc_u_knode
*u32_init_knode(struct tcf_proto
*tp
,
775 struct tc_u_knode
*n
)
777 struct tc_u_knode
*new;
778 struct tc_u32_sel
*s
= &n
->sel
;
780 new = kzalloc(sizeof(*n
) + s
->nkeys
*sizeof(struct tc_u32_key
),
786 RCU_INIT_POINTER(new->next
, n
->next
);
787 new->handle
= n
->handle
;
788 RCU_INIT_POINTER(new->ht_up
, n
->ht_up
);
790 #ifdef CONFIG_NET_CLS_IND
791 new->ifindex
= n
->ifindex
;
793 new->fshift
= n
->fshift
;
795 new->flags
= n
->flags
;
796 RCU_INIT_POINTER(new->ht_down
, n
->ht_down
);
798 /* bump reference count as long as we hold pointer to structure */
800 new->ht_down
->refcnt
++;
802 #ifdef CONFIG_CLS_U32_PERF
803 /* Statistics may be incremented by readers during update
804 * so we must keep them in tact. When the node is later destroyed
805 * a special destroy call must be made to not free the pf memory.
810 #ifdef CONFIG_CLS_U32_MARK
813 /* Similarly success statistics must be moved as pointers */
814 new->pcpu_success
= n
->pcpu_success
;
817 memcpy(&new->sel
, s
, sizeof(*s
) + s
->nkeys
*sizeof(struct tc_u32_key
));
819 tcf_exts_init(&new->exts
, TCA_U32_ACT
, TCA_U32_POLICE
);
824 static int u32_change(struct net
*net
, struct sk_buff
*in_skb
,
825 struct tcf_proto
*tp
, unsigned long base
, u32 handle
,
827 unsigned long *arg
, bool ovr
)
829 struct tc_u_common
*tp_c
= tp
->data
;
830 struct tc_u_hnode
*ht
;
831 struct tc_u_knode
*n
;
832 struct tc_u32_sel
*s
;
833 struct nlattr
*opt
= tca
[TCA_OPTIONS
];
834 struct nlattr
*tb
[TCA_U32_MAX
+ 1];
837 #ifdef CONFIG_CLS_U32_PERF
842 return handle
? -EINVAL
: 0;
844 err
= nla_parse_nested(tb
, TCA_U32_MAX
, opt
, u32_policy
);
848 if (tb
[TCA_U32_FLAGS
])
849 flags
= nla_get_u32(tb
[TCA_U32_FLAGS
]);
851 n
= (struct tc_u_knode
*)*arg
;
853 struct tc_u_knode
*new;
855 if (TC_U32_KEY(n
->handle
) == 0)
858 if (n
->flags
!= flags
)
861 new = u32_init_knode(tp
, n
);
865 err
= u32_set_parms(net
, tp
, base
,
866 rtnl_dereference(n
->ht_up
), new, tb
,
870 u32_destroy_key(tp
, new, false);
874 u32_replace_knode(tp
, tp_c
, new);
875 tcf_unbind_filter(tp
, &n
->res
);
876 call_rcu(&n
->rcu
, u32_delete_key_rcu
);
877 u32_replace_hw_knode(tp
, new, flags
);
881 if (tb
[TCA_U32_DIVISOR
]) {
882 unsigned int divisor
= nla_get_u32(tb
[TCA_U32_DIVISOR
]);
884 if (--divisor
> 0x100)
886 if (TC_U32_KEY(handle
))
889 handle
= gen_new_htid(tp
->data
);
893 ht
= kzalloc(sizeof(*ht
) + divisor
*sizeof(void *), GFP_KERNEL
);
898 ht
->divisor
= divisor
;
901 RCU_INIT_POINTER(ht
->next
, tp_c
->hlist
);
902 rcu_assign_pointer(tp_c
->hlist
, ht
);
903 *arg
= (unsigned long)ht
;
905 u32_replace_hw_hnode(tp
, ht
, flags
);
909 if (tb
[TCA_U32_HASH
]) {
910 htid
= nla_get_u32(tb
[TCA_U32_HASH
]);
911 if (TC_U32_HTID(htid
) == TC_U32_ROOT
) {
912 ht
= rtnl_dereference(tp
->root
);
915 ht
= u32_lookup_ht(tp
->data
, TC_U32_HTID(htid
));
920 ht
= rtnl_dereference(tp
->root
);
924 if (ht
->divisor
< TC_U32_HASH(htid
))
928 if (TC_U32_HTID(handle
) && TC_U32_HTID(handle
^htid
))
930 handle
= htid
| TC_U32_NODE(handle
);
932 handle
= gen_new_kid(ht
, htid
);
934 if (tb
[TCA_U32_SEL
] == NULL
)
937 s
= nla_data(tb
[TCA_U32_SEL
]);
939 n
= kzalloc(sizeof(*n
) + s
->nkeys
*sizeof(struct tc_u32_key
), GFP_KERNEL
);
943 #ifdef CONFIG_CLS_U32_PERF
944 size
= sizeof(struct tc_u32_pcnt
) + s
->nkeys
* sizeof(u64
);
945 n
->pf
= __alloc_percpu(size
, __alignof__(struct tc_u32_pcnt
));
952 memcpy(&n
->sel
, s
, sizeof(*s
) + s
->nkeys
*sizeof(struct tc_u32_key
));
953 RCU_INIT_POINTER(n
->ht_up
, ht
);
955 n
->fshift
= s
->hmask
? ffs(ntohl(s
->hmask
)) - 1 : 0;
957 tcf_exts_init(&n
->exts
, TCA_U32_ACT
, TCA_U32_POLICE
);
960 #ifdef CONFIG_CLS_U32_MARK
961 n
->pcpu_success
= alloc_percpu(u32
);
962 if (!n
->pcpu_success
) {
967 if (tb
[TCA_U32_MARK
]) {
968 struct tc_u32_mark
*mark
;
970 mark
= nla_data(tb
[TCA_U32_MARK
]);
972 n
->mask
= mark
->mask
;
976 err
= u32_set_parms(net
, tp
, base
, ht
, n
, tb
, tca
[TCA_RATE
], ovr
);
978 struct tc_u_knode __rcu
**ins
;
979 struct tc_u_knode
*pins
;
981 ins
= &ht
->ht
[TC_U32_HASH(handle
)];
982 for (pins
= rtnl_dereference(*ins
); pins
;
983 ins
= &pins
->next
, pins
= rtnl_dereference(*ins
))
984 if (TC_U32_NODE(handle
) < TC_U32_NODE(pins
->handle
))
987 RCU_INIT_POINTER(n
->next
, pins
);
988 rcu_assign_pointer(*ins
, n
);
989 u32_replace_hw_knode(tp
, n
, flags
);
990 *arg
= (unsigned long)n
;
994 #ifdef CONFIG_CLS_U32_MARK
995 free_percpu(n
->pcpu_success
);
999 #ifdef CONFIG_CLS_U32_PERF
1006 static void u32_walk(struct tcf_proto
*tp
, struct tcf_walker
*arg
)
1008 struct tc_u_common
*tp_c
= tp
->data
;
1009 struct tc_u_hnode
*ht
;
1010 struct tc_u_knode
*n
;
1016 for (ht
= rtnl_dereference(tp_c
->hlist
);
1018 ht
= rtnl_dereference(ht
->next
)) {
1019 if (ht
->prio
!= tp
->prio
)
1021 if (arg
->count
>= arg
->skip
) {
1022 if (arg
->fn(tp
, (unsigned long)ht
, arg
) < 0) {
1028 for (h
= 0; h
<= ht
->divisor
; h
++) {
1029 for (n
= rtnl_dereference(ht
->ht
[h
]);
1031 n
= rtnl_dereference(n
->next
)) {
1032 if (arg
->count
< arg
->skip
) {
1036 if (arg
->fn(tp
, (unsigned long)n
, arg
) < 0) {
1046 static int u32_dump(struct net
*net
, struct tcf_proto
*tp
, unsigned long fh
,
1047 struct sk_buff
*skb
, struct tcmsg
*t
)
1049 struct tc_u_knode
*n
= (struct tc_u_knode
*)fh
;
1050 struct tc_u_hnode
*ht_up
, *ht_down
;
1051 struct nlattr
*nest
;
1056 t
->tcm_handle
= n
->handle
;
1058 nest
= nla_nest_start(skb
, TCA_OPTIONS
);
1060 goto nla_put_failure
;
1062 if (TC_U32_KEY(n
->handle
) == 0) {
1063 struct tc_u_hnode
*ht
= (struct tc_u_hnode
*)fh
;
1064 u32 divisor
= ht
->divisor
+ 1;
1066 if (nla_put_u32(skb
, TCA_U32_DIVISOR
, divisor
))
1067 goto nla_put_failure
;
1069 #ifdef CONFIG_CLS_U32_PERF
1070 struct tc_u32_pcnt
*gpf
;
1074 if (nla_put(skb
, TCA_U32_SEL
,
1075 sizeof(n
->sel
) + n
->sel
.nkeys
*sizeof(struct tc_u32_key
),
1077 goto nla_put_failure
;
1079 ht_up
= rtnl_dereference(n
->ht_up
);
1081 u32 htid
= n
->handle
& 0xFFFFF000;
1082 if (nla_put_u32(skb
, TCA_U32_HASH
, htid
))
1083 goto nla_put_failure
;
1085 if (n
->res
.classid
&&
1086 nla_put_u32(skb
, TCA_U32_CLASSID
, n
->res
.classid
))
1087 goto nla_put_failure
;
1089 ht_down
= rtnl_dereference(n
->ht_down
);
1091 nla_put_u32(skb
, TCA_U32_LINK
, ht_down
->handle
))
1092 goto nla_put_failure
;
1094 if (n
->flags
&& nla_put_u32(skb
, TCA_U32_FLAGS
, n
->flags
))
1095 goto nla_put_failure
;
1097 #ifdef CONFIG_CLS_U32_MARK
1098 if ((n
->val
|| n
->mask
)) {
1099 struct tc_u32_mark mark
= {.val
= n
->val
,
1104 for_each_possible_cpu(cpum
) {
1105 __u32 cnt
= *per_cpu_ptr(n
->pcpu_success
, cpum
);
1107 mark
.success
+= cnt
;
1110 if (nla_put(skb
, TCA_U32_MARK
, sizeof(mark
), &mark
))
1111 goto nla_put_failure
;
1115 if (tcf_exts_dump(skb
, &n
->exts
) < 0)
1116 goto nla_put_failure
;
1118 #ifdef CONFIG_NET_CLS_IND
1120 struct net_device
*dev
;
1121 dev
= __dev_get_by_index(net
, n
->ifindex
);
1122 if (dev
&& nla_put_string(skb
, TCA_U32_INDEV
, dev
->name
))
1123 goto nla_put_failure
;
1126 #ifdef CONFIG_CLS_U32_PERF
1127 gpf
= kzalloc(sizeof(struct tc_u32_pcnt
) +
1128 n
->sel
.nkeys
* sizeof(u64
),
1131 goto nla_put_failure
;
1133 for_each_possible_cpu(cpu
) {
1135 struct tc_u32_pcnt
*pf
= per_cpu_ptr(n
->pf
, cpu
);
1137 gpf
->rcnt
+= pf
->rcnt
;
1138 gpf
->rhit
+= pf
->rhit
;
1139 for (i
= 0; i
< n
->sel
.nkeys
; i
++)
1140 gpf
->kcnts
[i
] += pf
->kcnts
[i
];
1143 if (nla_put(skb
, TCA_U32_PCNT
,
1144 sizeof(struct tc_u32_pcnt
) + n
->sel
.nkeys
*sizeof(u64
),
1147 goto nla_put_failure
;
1153 nla_nest_end(skb
, nest
);
1155 if (TC_U32_KEY(n
->handle
))
1156 if (tcf_exts_dump_stats(skb
, &n
->exts
) < 0)
1157 goto nla_put_failure
;
1161 nla_nest_cancel(skb
, nest
);
1165 static struct tcf_proto_ops cls_u32_ops __read_mostly
= {
1167 .classify
= u32_classify
,
1169 .destroy
= u32_destroy
,
1171 .change
= u32_change
,
1172 .delete = u32_delete
,
1175 .owner
= THIS_MODULE
,
1178 static int __init
init_u32(void)
1180 pr_info("u32 classifier\n");
1181 #ifdef CONFIG_CLS_U32_PERF
1182 pr_info(" Performance counters on\n");
1184 #ifdef CONFIG_NET_CLS_IND
1185 pr_info(" input device check on\n");
1187 #ifdef CONFIG_NET_CLS_ACT
1188 pr_info(" Actions configured\n");
1190 return register_tcf_proto_ops(&cls_u32_ops
);
1193 static void __exit
exit_u32(void)
1195 unregister_tcf_proto_ops(&cls_u32_ops
);
1198 module_init(init_u32
)
1199 module_exit(exit_u32
)
1200 MODULE_LICENSE("GPL");