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 <linux/netdevice.h>
44 #include <linux/hash.h>
45 #include <net/netlink.h>
46 #include <net/act_api.h>
47 #include <net/pkt_cls.h>
48 #include <linux/idr.h>
51 struct tc_u_knode __rcu
*next
;
53 struct tc_u_hnode __rcu
*ht_up
;
55 #ifdef CONFIG_NET_CLS_IND
59 struct tcf_result res
;
60 struct tc_u_hnode __rcu
*ht_down
;
61 #ifdef CONFIG_CLS_U32_PERF
62 struct tc_u32_pcnt __percpu
*pf
;
65 #ifdef CONFIG_CLS_U32_MARK
68 u32 __percpu
*pcpu_success
;
72 struct work_struct work
;
75 /* The 'sel' field MUST be the last field in structure to allow for
76 * tc_u32_keys allocated at end of structure.
78 struct tc_u32_sel sel
;
82 struct tc_u_hnode __rcu
*next
;
85 struct tc_u_common
*tp_c
;
88 struct idr handle_idr
;
90 /* The 'ht' field MUST be the last field in structure to allow for
91 * more entries allocated at end of structure.
93 struct tc_u_knode __rcu
*ht
[1];
97 struct tc_u_hnode __rcu
*hlist
;
98 struct tcf_block
*block
;
100 struct idr handle_idr
;
101 struct hlist_node hnode
;
105 static inline unsigned int u32_hash_fold(__be32 key
,
106 const struct tc_u32_sel
*sel
,
109 unsigned int h
= ntohl(key
& sel
->hmask
) >> fshift
;
114 static int u32_classify(struct sk_buff
*skb
, const struct tcf_proto
*tp
,
115 struct tcf_result
*res
)
118 struct tc_u_knode
*knode
;
120 } stack
[TC_U32_MAXDEPTH
];
122 struct tc_u_hnode
*ht
= rcu_dereference_bh(tp
->root
);
123 unsigned int off
= skb_network_offset(skb
);
124 struct tc_u_knode
*n
;
128 #ifdef CONFIG_CLS_U32_PERF
134 n
= rcu_dereference_bh(ht
->ht
[sel
]);
138 struct tc_u32_key
*key
= n
->sel
.keys
;
140 #ifdef CONFIG_CLS_U32_PERF
141 __this_cpu_inc(n
->pf
->rcnt
);
145 if (tc_skip_sw(n
->flags
)) {
146 n
= rcu_dereference_bh(n
->next
);
150 #ifdef CONFIG_CLS_U32_MARK
151 if ((skb
->mark
& n
->mask
) != n
->val
) {
152 n
= rcu_dereference_bh(n
->next
);
155 __this_cpu_inc(*n
->pcpu_success
);
159 for (i
= n
->sel
.nkeys
; i
> 0; i
--, key
++) {
160 int toff
= off
+ key
->off
+ (off2
& key
->offmask
);
163 if (skb_headroom(skb
) + toff
> INT_MAX
)
166 data
= skb_header_pointer(skb
, toff
, 4, &hdata
);
169 if ((*data
^ key
->val
) & key
->mask
) {
170 n
= rcu_dereference_bh(n
->next
);
173 #ifdef CONFIG_CLS_U32_PERF
174 __this_cpu_inc(n
->pf
->kcnts
[j
]);
179 ht
= rcu_dereference_bh(n
->ht_down
);
182 if (n
->sel
.flags
& TC_U32_TERMINAL
) {
185 #ifdef CONFIG_NET_CLS_IND
186 if (!tcf_match_indev(skb
, n
->ifindex
)) {
187 n
= rcu_dereference_bh(n
->next
);
191 #ifdef CONFIG_CLS_U32_PERF
192 __this_cpu_inc(n
->pf
->rhit
);
194 r
= tcf_exts_exec(skb
, &n
->exts
, res
);
196 n
= rcu_dereference_bh(n
->next
);
202 n
= rcu_dereference_bh(n
->next
);
207 if (sdepth
>= TC_U32_MAXDEPTH
)
209 stack
[sdepth
].knode
= n
;
210 stack
[sdepth
].off
= off
;
213 ht
= rcu_dereference_bh(n
->ht_down
);
218 data
= skb_header_pointer(skb
, off
+ n
->sel
.hoff
, 4,
222 sel
= ht
->divisor
& u32_hash_fold(*data
, &n
->sel
,
225 if (!(n
->sel
.flags
& (TC_U32_VAROFFSET
| TC_U32_OFFSET
| TC_U32_EAT
)))
228 if (n
->sel
.flags
& (TC_U32_OFFSET
| TC_U32_VAROFFSET
)) {
229 off2
= n
->sel
.off
+ 3;
230 if (n
->sel
.flags
& TC_U32_VAROFFSET
) {
233 data
= skb_header_pointer(skb
,
238 off2
+= ntohs(n
->sel
.offmask
& *data
) >>
243 if (n
->sel
.flags
& TC_U32_EAT
) {
254 n
= stack
[sdepth
].knode
;
255 ht
= rcu_dereference_bh(n
->ht_up
);
256 off
= stack
[sdepth
].off
;
263 net_warn_ratelimited("cls_u32: dead loop\n");
267 static struct tc_u_hnode
*u32_lookup_ht(struct tc_u_common
*tp_c
, u32 handle
)
269 struct tc_u_hnode
*ht
;
271 for (ht
= rtnl_dereference(tp_c
->hlist
);
273 ht
= rtnl_dereference(ht
->next
))
274 if (ht
->handle
== handle
)
280 static struct tc_u_knode
*u32_lookup_key(struct tc_u_hnode
*ht
, u32 handle
)
283 struct tc_u_knode
*n
= NULL
;
285 sel
= TC_U32_HASH(handle
);
286 if (sel
> ht
->divisor
)
289 for (n
= rtnl_dereference(ht
->ht
[sel
]);
291 n
= rtnl_dereference(n
->next
))
292 if (n
->handle
== handle
)
299 static void *u32_get(struct tcf_proto
*tp
, u32 handle
)
301 struct tc_u_hnode
*ht
;
302 struct tc_u_common
*tp_c
= tp
->data
;
304 if (TC_U32_HTID(handle
) == TC_U32_ROOT
)
305 ht
= rtnl_dereference(tp
->root
);
307 ht
= u32_lookup_ht(tp_c
, TC_U32_HTID(handle
));
312 if (TC_U32_KEY(handle
) == 0)
315 return u32_lookup_key(ht
, handle
);
318 static u32
gen_new_htid(struct tc_u_common
*tp_c
, struct tc_u_hnode
*ptr
)
320 unsigned long idr_index
;
323 /* This is only used inside rtnl lock it is safe to increment
324 * without read _copy_ update semantics
326 err
= idr_alloc_ext(&tp_c
->handle_idr
, ptr
, &idr_index
,
327 1, 0x7FF, GFP_KERNEL
);
330 return (u32
)(idr_index
| 0x800) << 20;
333 static struct hlist_head
*tc_u_common_hash
;
335 #define U32_HASH_SHIFT 10
336 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
338 static unsigned int tc_u_hash(const struct tcf_proto
*tp
)
340 return hash_ptr(tp
->chain
->block
, U32_HASH_SHIFT
);
343 static struct tc_u_common
*tc_u_common_find(const struct tcf_proto
*tp
)
345 struct tc_u_common
*tc
;
349 hlist_for_each_entry(tc
, &tc_u_common_hash
[h
], hnode
) {
350 if (tc
->block
== tp
->chain
->block
)
356 static int u32_init(struct tcf_proto
*tp
)
358 struct tc_u_hnode
*root_ht
;
359 struct tc_u_common
*tp_c
;
362 tp_c
= tc_u_common_find(tp
);
364 root_ht
= kzalloc(sizeof(*root_ht
), GFP_KERNEL
);
369 root_ht
->handle
= tp_c
? gen_new_htid(tp_c
, root_ht
) : 0x80000000;
370 root_ht
->prio
= tp
->prio
;
371 idr_init(&root_ht
->handle_idr
);
374 tp_c
= kzalloc(sizeof(*tp_c
), GFP_KERNEL
);
379 tp_c
->block
= tp
->chain
->block
;
380 INIT_HLIST_NODE(&tp_c
->hnode
);
381 idr_init(&tp_c
->handle_idr
);
384 hlist_add_head(&tp_c
->hnode
, &tc_u_common_hash
[h
]);
388 RCU_INIT_POINTER(root_ht
->next
, tp_c
->hlist
);
389 rcu_assign_pointer(tp_c
->hlist
, root_ht
);
390 root_ht
->tp_c
= tp_c
;
392 rcu_assign_pointer(tp
->root
, root_ht
);
397 static int u32_destroy_key(struct tcf_proto
*tp
, struct tc_u_knode
*n
,
400 tcf_exts_destroy(&n
->exts
);
401 tcf_exts_put_net(&n
->exts
);
403 n
->ht_down
->refcnt
--;
404 #ifdef CONFIG_CLS_U32_PERF
408 #ifdef CONFIG_CLS_U32_MARK
410 free_percpu(n
->pcpu_success
);
416 /* u32_delete_key_rcu should be called when free'ing a copied
417 * version of a tc_u_knode obtained from u32_init_knode(). When
418 * copies are obtained from u32_init_knode() the statistics are
419 * shared between the old and new copies to allow readers to
420 * continue to update the statistics during the copy. To support
421 * this the u32_delete_key_rcu variant does not free the percpu
424 static void u32_delete_key_work(struct work_struct
*work
)
426 struct tc_u_knode
*key
= container_of(work
, struct tc_u_knode
, work
);
429 u32_destroy_key(key
->tp
, key
, false);
433 static void u32_delete_key_rcu(struct rcu_head
*rcu
)
435 struct tc_u_knode
*key
= container_of(rcu
, struct tc_u_knode
, rcu
);
437 INIT_WORK(&key
->work
, u32_delete_key_work
);
438 tcf_queue_work(&key
->work
);
441 /* u32_delete_key_freepf_rcu is the rcu callback variant
442 * that free's the entire structure including the statistics
443 * percpu variables. Only use this if the key is not a copy
444 * returned by u32_init_knode(). See u32_delete_key_rcu()
445 * for the variant that should be used with keys return from
448 static void u32_delete_key_freepf_work(struct work_struct
*work
)
450 struct tc_u_knode
*key
= container_of(work
, struct tc_u_knode
, work
);
453 u32_destroy_key(key
->tp
, key
, true);
457 static void u32_delete_key_freepf_rcu(struct rcu_head
*rcu
)
459 struct tc_u_knode
*key
= container_of(rcu
, struct tc_u_knode
, rcu
);
461 INIT_WORK(&key
->work
, u32_delete_key_freepf_work
);
462 tcf_queue_work(&key
->work
);
465 static int u32_delete_key(struct tcf_proto
*tp
, struct tc_u_knode
*key
)
467 struct tc_u_knode __rcu
**kp
;
468 struct tc_u_knode
*pkp
;
469 struct tc_u_hnode
*ht
= rtnl_dereference(key
->ht_up
);
472 kp
= &ht
->ht
[TC_U32_HASH(key
->handle
)];
473 for (pkp
= rtnl_dereference(*kp
); pkp
;
474 kp
= &pkp
->next
, pkp
= rtnl_dereference(*kp
)) {
476 RCU_INIT_POINTER(*kp
, key
->next
);
478 tcf_unbind_filter(tp
, &key
->res
);
479 tcf_exts_get_net(&key
->exts
);
480 call_rcu(&key
->rcu
, u32_delete_key_freepf_rcu
);
489 static void u32_clear_hw_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*h
)
491 struct tcf_block
*block
= tp
->chain
->block
;
492 struct tc_cls_u32_offload cls_u32
= {};
494 tc_cls_common_offload_init(&cls_u32
.common
, tp
);
495 cls_u32
.command
= TC_CLSU32_DELETE_HNODE
;
496 cls_u32
.hnode
.divisor
= h
->divisor
;
497 cls_u32
.hnode
.handle
= h
->handle
;
498 cls_u32
.hnode
.prio
= h
->prio
;
500 tc_setup_cb_call(block
, NULL
, TC_SETUP_CLSU32
, &cls_u32
, false);
503 static int u32_replace_hw_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*h
,
506 struct tcf_block
*block
= tp
->chain
->block
;
507 struct tc_cls_u32_offload cls_u32
= {};
508 bool skip_sw
= tc_skip_sw(flags
);
509 bool offloaded
= false;
512 tc_cls_common_offload_init(&cls_u32
.common
, tp
);
513 cls_u32
.command
= TC_CLSU32_NEW_HNODE
;
514 cls_u32
.hnode
.divisor
= h
->divisor
;
515 cls_u32
.hnode
.handle
= h
->handle
;
516 cls_u32
.hnode
.prio
= h
->prio
;
518 err
= tc_setup_cb_call(block
, NULL
, TC_SETUP_CLSU32
, &cls_u32
, skip_sw
);
520 u32_clear_hw_hnode(tp
, h
);
522 } else if (err
> 0) {
526 if (skip_sw
&& !offloaded
)
532 static void u32_remove_hw_knode(struct tcf_proto
*tp
, u32 handle
)
534 struct tcf_block
*block
= tp
->chain
->block
;
535 struct tc_cls_u32_offload cls_u32
= {};
537 tc_cls_common_offload_init(&cls_u32
.common
, tp
);
538 cls_u32
.command
= TC_CLSU32_DELETE_KNODE
;
539 cls_u32
.knode
.handle
= handle
;
541 tc_setup_cb_call(block
, NULL
, TC_SETUP_CLSU32
, &cls_u32
, false);
544 static int u32_replace_hw_knode(struct tcf_proto
*tp
, struct tc_u_knode
*n
,
547 struct tcf_block
*block
= tp
->chain
->block
;
548 struct tc_cls_u32_offload cls_u32
= {};
549 bool skip_sw
= tc_skip_sw(flags
);
552 tc_cls_common_offload_init(&cls_u32
.common
, tp
);
553 cls_u32
.command
= TC_CLSU32_REPLACE_KNODE
;
554 cls_u32
.knode
.handle
= n
->handle
;
555 cls_u32
.knode
.fshift
= n
->fshift
;
556 #ifdef CONFIG_CLS_U32_MARK
557 cls_u32
.knode
.val
= n
->val
;
558 cls_u32
.knode
.mask
= n
->mask
;
560 cls_u32
.knode
.val
= 0;
561 cls_u32
.knode
.mask
= 0;
563 cls_u32
.knode
.sel
= &n
->sel
;
564 cls_u32
.knode
.exts
= &n
->exts
;
566 cls_u32
.knode
.link_handle
= n
->ht_down
->handle
;
568 err
= tc_setup_cb_call(block
, NULL
, TC_SETUP_CLSU32
, &cls_u32
, skip_sw
);
570 u32_remove_hw_knode(tp
, n
->handle
);
572 } else if (err
> 0) {
573 n
->flags
|= TCA_CLS_FLAGS_IN_HW
;
576 if (skip_sw
&& !(n
->flags
& TCA_CLS_FLAGS_IN_HW
))
582 static void u32_clear_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
)
584 struct tc_u_knode
*n
;
587 for (h
= 0; h
<= ht
->divisor
; h
++) {
588 while ((n
= rtnl_dereference(ht
->ht
[h
])) != NULL
) {
589 RCU_INIT_POINTER(ht
->ht
[h
],
590 rtnl_dereference(n
->next
));
591 tcf_unbind_filter(tp
, &n
->res
);
592 u32_remove_hw_knode(tp
, n
->handle
);
593 idr_remove_ext(&ht
->handle_idr
, n
->handle
);
594 if (tcf_exts_get_net(&n
->exts
))
595 call_rcu(&n
->rcu
, u32_delete_key_freepf_rcu
);
597 u32_destroy_key(n
->tp
, n
, true);
602 static int u32_destroy_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
)
604 struct tc_u_common
*tp_c
= tp
->data
;
605 struct tc_u_hnode __rcu
**hn
;
606 struct tc_u_hnode
*phn
;
610 u32_clear_hnode(tp
, ht
);
613 for (phn
= rtnl_dereference(*hn
);
615 hn
= &phn
->next
, phn
= rtnl_dereference(*hn
)) {
617 u32_clear_hw_hnode(tp
, ht
);
618 idr_destroy(&ht
->handle_idr
);
619 idr_remove_ext(&tp_c
->handle_idr
, ht
->handle
);
620 RCU_INIT_POINTER(*hn
, ht
->next
);
629 static bool ht_empty(struct tc_u_hnode
*ht
)
633 for (h
= 0; h
<= ht
->divisor
; h
++)
634 if (rcu_access_pointer(ht
->ht
[h
]))
640 static void u32_destroy(struct tcf_proto
*tp
)
642 struct tc_u_common
*tp_c
= tp
->data
;
643 struct tc_u_hnode
*root_ht
= rtnl_dereference(tp
->root
);
645 WARN_ON(root_ht
== NULL
);
647 if (root_ht
&& --root_ht
->refcnt
== 0)
648 u32_destroy_hnode(tp
, root_ht
);
650 if (--tp_c
->refcnt
== 0) {
651 struct tc_u_hnode
*ht
;
653 hlist_del(&tp_c
->hnode
);
655 for (ht
= rtnl_dereference(tp_c
->hlist
);
657 ht
= rtnl_dereference(ht
->next
)) {
659 u32_clear_hnode(tp
, ht
);
662 while ((ht
= rtnl_dereference(tp_c
->hlist
)) != NULL
) {
663 RCU_INIT_POINTER(tp_c
->hlist
, ht
->next
);
667 idr_destroy(&tp_c
->handle_idr
);
674 static int u32_delete(struct tcf_proto
*tp
, void *arg
, bool *last
)
676 struct tc_u_hnode
*ht
= arg
;
677 struct tc_u_hnode
*root_ht
= rtnl_dereference(tp
->root
);
678 struct tc_u_common
*tp_c
= tp
->data
;
684 if (TC_U32_KEY(ht
->handle
)) {
685 u32_remove_hw_knode(tp
, ht
->handle
);
686 ret
= u32_delete_key(tp
, (struct tc_u_knode
*)ht
);
693 if (ht
->refcnt
== 1) {
695 u32_destroy_hnode(tp
, ht
);
703 if (root_ht
->refcnt
> 1) {
707 if (root_ht
->refcnt
== 1) {
708 if (!ht_empty(root_ht
)) {
715 if (tp_c
->refcnt
> 1) {
720 if (tp_c
->refcnt
== 1) {
721 struct tc_u_hnode
*ht
;
723 for (ht
= rtnl_dereference(tp_c
->hlist
);
725 ht
= rtnl_dereference(ht
->next
))
736 static u32
gen_new_kid(struct tc_u_hnode
*ht
, u32 htid
)
738 unsigned long idr_index
;
739 u32 start
= htid
| 0x800;
740 u32 max
= htid
| 0xFFF;
743 if (idr_alloc_ext(&ht
->handle_idr
, NULL
, &idr_index
,
744 start
, max
+ 1, GFP_KERNEL
)) {
745 if (idr_alloc_ext(&ht
->handle_idr
, NULL
, &idr_index
,
746 min
+ 1, max
+ 1, GFP_KERNEL
))
750 return (u32
)idr_index
;
753 static const struct nla_policy u32_policy
[TCA_U32_MAX
+ 1] = {
754 [TCA_U32_CLASSID
] = { .type
= NLA_U32
},
755 [TCA_U32_HASH
] = { .type
= NLA_U32
},
756 [TCA_U32_LINK
] = { .type
= NLA_U32
},
757 [TCA_U32_DIVISOR
] = { .type
= NLA_U32
},
758 [TCA_U32_SEL
] = { .len
= sizeof(struct tc_u32_sel
) },
759 [TCA_U32_INDEV
] = { .type
= NLA_STRING
, .len
= IFNAMSIZ
},
760 [TCA_U32_MARK
] = { .len
= sizeof(struct tc_u32_mark
) },
761 [TCA_U32_FLAGS
] = { .type
= NLA_U32
},
764 static int u32_set_parms(struct net
*net
, struct tcf_proto
*tp
,
765 unsigned long base
, struct tc_u_hnode
*ht
,
766 struct tc_u_knode
*n
, struct nlattr
**tb
,
767 struct nlattr
*est
, bool ovr
)
771 err
= tcf_exts_validate(net
, tp
, tb
, est
, &n
->exts
, ovr
);
775 if (tb
[TCA_U32_LINK
]) {
776 u32 handle
= nla_get_u32(tb
[TCA_U32_LINK
]);
777 struct tc_u_hnode
*ht_down
= NULL
, *ht_old
;
779 if (TC_U32_KEY(handle
))
783 ht_down
= u32_lookup_ht(ht
->tp_c
, handle
);
790 ht_old
= rtnl_dereference(n
->ht_down
);
791 rcu_assign_pointer(n
->ht_down
, ht_down
);
796 if (tb
[TCA_U32_CLASSID
]) {
797 n
->res
.classid
= nla_get_u32(tb
[TCA_U32_CLASSID
]);
798 tcf_bind_filter(tp
, &n
->res
, base
);
801 #ifdef CONFIG_NET_CLS_IND
802 if (tb
[TCA_U32_INDEV
]) {
804 ret
= tcf_change_indev(net
, tb
[TCA_U32_INDEV
]);
813 static void u32_replace_knode(struct tcf_proto
*tp
, struct tc_u_common
*tp_c
,
814 struct tc_u_knode
*n
)
816 struct tc_u_knode __rcu
**ins
;
817 struct tc_u_knode
*pins
;
818 struct tc_u_hnode
*ht
;
820 if (TC_U32_HTID(n
->handle
) == TC_U32_ROOT
)
821 ht
= rtnl_dereference(tp
->root
);
823 ht
= u32_lookup_ht(tp_c
, TC_U32_HTID(n
->handle
));
825 ins
= &ht
->ht
[TC_U32_HASH(n
->handle
)];
827 /* The node must always exist for it to be replaced if this is not the
828 * case then something went very wrong elsewhere.
830 for (pins
= rtnl_dereference(*ins
); ;
831 ins
= &pins
->next
, pins
= rtnl_dereference(*ins
))
832 if (pins
->handle
== n
->handle
)
835 idr_replace_ext(&ht
->handle_idr
, n
, n
->handle
);
836 RCU_INIT_POINTER(n
->next
, pins
->next
);
837 rcu_assign_pointer(*ins
, n
);
840 static struct tc_u_knode
*u32_init_knode(struct tcf_proto
*tp
,
841 struct tc_u_knode
*n
)
843 struct tc_u_knode
*new;
844 struct tc_u32_sel
*s
= &n
->sel
;
846 new = kzalloc(sizeof(*n
) + s
->nkeys
*sizeof(struct tc_u32_key
),
852 RCU_INIT_POINTER(new->next
, n
->next
);
853 new->handle
= n
->handle
;
854 RCU_INIT_POINTER(new->ht_up
, n
->ht_up
);
856 #ifdef CONFIG_NET_CLS_IND
857 new->ifindex
= n
->ifindex
;
859 new->fshift
= n
->fshift
;
861 new->flags
= n
->flags
;
862 RCU_INIT_POINTER(new->ht_down
, n
->ht_down
);
864 /* bump reference count as long as we hold pointer to structure */
866 new->ht_down
->refcnt
++;
868 #ifdef CONFIG_CLS_U32_PERF
869 /* Statistics may be incremented by readers during update
870 * so we must keep them in tact. When the node is later destroyed
871 * a special destroy call must be made to not free the pf memory.
876 #ifdef CONFIG_CLS_U32_MARK
879 /* Similarly success statistics must be moved as pointers */
880 new->pcpu_success
= n
->pcpu_success
;
883 memcpy(&new->sel
, s
, sizeof(*s
) + s
->nkeys
*sizeof(struct tc_u32_key
));
885 if (tcf_exts_init(&new->exts
, TCA_U32_ACT
, TCA_U32_POLICE
)) {
893 static int u32_change(struct net
*net
, struct sk_buff
*in_skb
,
894 struct tcf_proto
*tp
, unsigned long base
, u32 handle
,
895 struct nlattr
**tca
, void **arg
, bool ovr
)
897 struct tc_u_common
*tp_c
= tp
->data
;
898 struct tc_u_hnode
*ht
;
899 struct tc_u_knode
*n
;
900 struct tc_u32_sel
*s
;
901 struct nlattr
*opt
= tca
[TCA_OPTIONS
];
902 struct nlattr
*tb
[TCA_U32_MAX
+ 1];
905 #ifdef CONFIG_CLS_U32_PERF
910 return handle
? -EINVAL
: 0;
912 err
= nla_parse_nested(tb
, TCA_U32_MAX
, opt
, u32_policy
, NULL
);
916 if (tb
[TCA_U32_FLAGS
]) {
917 flags
= nla_get_u32(tb
[TCA_U32_FLAGS
]);
918 if (!tc_flags_valid(flags
))
924 struct tc_u_knode
*new;
926 if (TC_U32_KEY(n
->handle
) == 0)
929 if (n
->flags
!= flags
)
932 new = u32_init_knode(tp
, n
);
936 err
= u32_set_parms(net
, tp
, base
,
937 rtnl_dereference(n
->ht_up
), new, tb
,
941 u32_destroy_key(tp
, new, false);
945 err
= u32_replace_hw_knode(tp
, new, flags
);
947 u32_destroy_key(tp
, new, false);
951 if (!tc_in_hw(new->flags
))
952 new->flags
|= TCA_CLS_FLAGS_NOT_IN_HW
;
954 u32_replace_knode(tp
, tp_c
, new);
955 tcf_unbind_filter(tp
, &n
->res
);
956 tcf_exts_get_net(&n
->exts
);
957 call_rcu(&n
->rcu
, u32_delete_key_rcu
);
961 if (tb
[TCA_U32_DIVISOR
]) {
962 unsigned int divisor
= nla_get_u32(tb
[TCA_U32_DIVISOR
]);
964 if (--divisor
> 0x100)
966 if (TC_U32_KEY(handle
))
968 ht
= kzalloc(sizeof(*ht
) + divisor
*sizeof(void *), GFP_KERNEL
);
972 handle
= gen_new_htid(tp
->data
, ht
);
978 err
= idr_alloc_ext(&tp_c
->handle_idr
, ht
, NULL
,
979 handle
, handle
+ 1, GFP_KERNEL
);
987 ht
->divisor
= divisor
;
990 idr_init(&ht
->handle_idr
);
992 err
= u32_replace_hw_hnode(tp
, ht
, flags
);
994 idr_remove_ext(&tp_c
->handle_idr
, handle
);
999 RCU_INIT_POINTER(ht
->next
, tp_c
->hlist
);
1000 rcu_assign_pointer(tp_c
->hlist
, ht
);
1006 if (tb
[TCA_U32_HASH
]) {
1007 htid
= nla_get_u32(tb
[TCA_U32_HASH
]);
1008 if (TC_U32_HTID(htid
) == TC_U32_ROOT
) {
1009 ht
= rtnl_dereference(tp
->root
);
1012 ht
= u32_lookup_ht(tp
->data
, TC_U32_HTID(htid
));
1017 ht
= rtnl_dereference(tp
->root
);
1021 if (ht
->divisor
< TC_U32_HASH(htid
))
1025 if (TC_U32_HTID(handle
) && TC_U32_HTID(handle
^htid
))
1027 handle
= htid
| TC_U32_NODE(handle
);
1028 err
= idr_alloc_ext(&ht
->handle_idr
, NULL
, NULL
,
1034 handle
= gen_new_kid(ht
, htid
);
1036 if (tb
[TCA_U32_SEL
] == NULL
) {
1041 s
= nla_data(tb
[TCA_U32_SEL
]);
1043 n
= kzalloc(sizeof(*n
) + s
->nkeys
*sizeof(struct tc_u32_key
), GFP_KERNEL
);
1049 #ifdef CONFIG_CLS_U32_PERF
1050 size
= sizeof(struct tc_u32_pcnt
) + s
->nkeys
* sizeof(u64
);
1051 n
->pf
= __alloc_percpu(size
, __alignof__(struct tc_u32_pcnt
));
1058 memcpy(&n
->sel
, s
, sizeof(*s
) + s
->nkeys
*sizeof(struct tc_u32_key
));
1059 RCU_INIT_POINTER(n
->ht_up
, ht
);
1061 n
->fshift
= s
->hmask
? ffs(ntohl(s
->hmask
)) - 1 : 0;
1065 err
= tcf_exts_init(&n
->exts
, TCA_U32_ACT
, TCA_U32_POLICE
);
1069 #ifdef CONFIG_CLS_U32_MARK
1070 n
->pcpu_success
= alloc_percpu(u32
);
1071 if (!n
->pcpu_success
) {
1076 if (tb
[TCA_U32_MARK
]) {
1077 struct tc_u32_mark
*mark
;
1079 mark
= nla_data(tb
[TCA_U32_MARK
]);
1081 n
->mask
= mark
->mask
;
1085 err
= u32_set_parms(net
, tp
, base
, ht
, n
, tb
, tca
[TCA_RATE
], ovr
);
1087 struct tc_u_knode __rcu
**ins
;
1088 struct tc_u_knode
*pins
;
1090 err
= u32_replace_hw_knode(tp
, n
, flags
);
1094 if (!tc_in_hw(n
->flags
))
1095 n
->flags
|= TCA_CLS_FLAGS_NOT_IN_HW
;
1097 ins
= &ht
->ht
[TC_U32_HASH(handle
)];
1098 for (pins
= rtnl_dereference(*ins
); pins
;
1099 ins
= &pins
->next
, pins
= rtnl_dereference(*ins
))
1100 if (TC_U32_NODE(handle
) < TC_U32_NODE(pins
->handle
))
1103 RCU_INIT_POINTER(n
->next
, pins
);
1104 rcu_assign_pointer(*ins
, n
);
1110 #ifdef CONFIG_CLS_U32_MARK
1111 free_percpu(n
->pcpu_success
);
1115 tcf_exts_destroy(&n
->exts
);
1116 #ifdef CONFIG_CLS_U32_PERF
1122 idr_remove_ext(&ht
->handle_idr
, handle
);
1126 static void u32_walk(struct tcf_proto
*tp
, struct tcf_walker
*arg
)
1128 struct tc_u_common
*tp_c
= tp
->data
;
1129 struct tc_u_hnode
*ht
;
1130 struct tc_u_knode
*n
;
1136 for (ht
= rtnl_dereference(tp_c
->hlist
);
1138 ht
= rtnl_dereference(ht
->next
)) {
1139 if (ht
->prio
!= tp
->prio
)
1141 if (arg
->count
>= arg
->skip
) {
1142 if (arg
->fn(tp
, ht
, arg
) < 0) {
1148 for (h
= 0; h
<= ht
->divisor
; h
++) {
1149 for (n
= rtnl_dereference(ht
->ht
[h
]);
1151 n
= rtnl_dereference(n
->next
)) {
1152 if (arg
->count
< arg
->skip
) {
1156 if (arg
->fn(tp
, n
, arg
) < 0) {
1166 static void u32_bind_class(void *fh
, u32 classid
, unsigned long cl
)
1168 struct tc_u_knode
*n
= fh
;
1170 if (n
&& n
->res
.classid
== classid
)
1174 static int u32_dump(struct net
*net
, struct tcf_proto
*tp
, void *fh
,
1175 struct sk_buff
*skb
, struct tcmsg
*t
)
1177 struct tc_u_knode
*n
= fh
;
1178 struct tc_u_hnode
*ht_up
, *ht_down
;
1179 struct nlattr
*nest
;
1184 t
->tcm_handle
= n
->handle
;
1186 nest
= nla_nest_start(skb
, TCA_OPTIONS
);
1188 goto nla_put_failure
;
1190 if (TC_U32_KEY(n
->handle
) == 0) {
1191 struct tc_u_hnode
*ht
= fh
;
1192 u32 divisor
= ht
->divisor
+ 1;
1194 if (nla_put_u32(skb
, TCA_U32_DIVISOR
, divisor
))
1195 goto nla_put_failure
;
1197 #ifdef CONFIG_CLS_U32_PERF
1198 struct tc_u32_pcnt
*gpf
;
1202 if (nla_put(skb
, TCA_U32_SEL
,
1203 sizeof(n
->sel
) + n
->sel
.nkeys
*sizeof(struct tc_u32_key
),
1205 goto nla_put_failure
;
1207 ht_up
= rtnl_dereference(n
->ht_up
);
1209 u32 htid
= n
->handle
& 0xFFFFF000;
1210 if (nla_put_u32(skb
, TCA_U32_HASH
, htid
))
1211 goto nla_put_failure
;
1213 if (n
->res
.classid
&&
1214 nla_put_u32(skb
, TCA_U32_CLASSID
, n
->res
.classid
))
1215 goto nla_put_failure
;
1217 ht_down
= rtnl_dereference(n
->ht_down
);
1219 nla_put_u32(skb
, TCA_U32_LINK
, ht_down
->handle
))
1220 goto nla_put_failure
;
1222 if (n
->flags
&& nla_put_u32(skb
, TCA_U32_FLAGS
, n
->flags
))
1223 goto nla_put_failure
;
1225 #ifdef CONFIG_CLS_U32_MARK
1226 if ((n
->val
|| n
->mask
)) {
1227 struct tc_u32_mark mark
= {.val
= n
->val
,
1232 for_each_possible_cpu(cpum
) {
1233 __u32 cnt
= *per_cpu_ptr(n
->pcpu_success
, cpum
);
1235 mark
.success
+= cnt
;
1238 if (nla_put(skb
, TCA_U32_MARK
, sizeof(mark
), &mark
))
1239 goto nla_put_failure
;
1243 if (tcf_exts_dump(skb
, &n
->exts
) < 0)
1244 goto nla_put_failure
;
1246 #ifdef CONFIG_NET_CLS_IND
1248 struct net_device
*dev
;
1249 dev
= __dev_get_by_index(net
, n
->ifindex
);
1250 if (dev
&& nla_put_string(skb
, TCA_U32_INDEV
, dev
->name
))
1251 goto nla_put_failure
;
1254 #ifdef CONFIG_CLS_U32_PERF
1255 gpf
= kzalloc(sizeof(struct tc_u32_pcnt
) +
1256 n
->sel
.nkeys
* sizeof(u64
),
1259 goto nla_put_failure
;
1261 for_each_possible_cpu(cpu
) {
1263 struct tc_u32_pcnt
*pf
= per_cpu_ptr(n
->pf
, cpu
);
1265 gpf
->rcnt
+= pf
->rcnt
;
1266 gpf
->rhit
+= pf
->rhit
;
1267 for (i
= 0; i
< n
->sel
.nkeys
; i
++)
1268 gpf
->kcnts
[i
] += pf
->kcnts
[i
];
1271 if (nla_put_64bit(skb
, TCA_U32_PCNT
,
1272 sizeof(struct tc_u32_pcnt
) +
1273 n
->sel
.nkeys
* sizeof(u64
),
1274 gpf
, TCA_U32_PAD
)) {
1276 goto nla_put_failure
;
1282 nla_nest_end(skb
, nest
);
1284 if (TC_U32_KEY(n
->handle
))
1285 if (tcf_exts_dump_stats(skb
, &n
->exts
) < 0)
1286 goto nla_put_failure
;
1290 nla_nest_cancel(skb
, nest
);
1294 static struct tcf_proto_ops cls_u32_ops __read_mostly
= {
1296 .classify
= u32_classify
,
1298 .destroy
= u32_destroy
,
1300 .change
= u32_change
,
1301 .delete = u32_delete
,
1304 .bind_class
= u32_bind_class
,
1305 .owner
= THIS_MODULE
,
1308 static int __init
init_u32(void)
1312 pr_info("u32 classifier\n");
1313 #ifdef CONFIG_CLS_U32_PERF
1314 pr_info(" Performance counters on\n");
1316 #ifdef CONFIG_NET_CLS_IND
1317 pr_info(" input device check on\n");
1319 #ifdef CONFIG_NET_CLS_ACT
1320 pr_info(" Actions configured\n");
1322 tc_u_common_hash
= kvmalloc_array(U32_HASH_SIZE
,
1323 sizeof(struct hlist_head
),
1325 if (!tc_u_common_hash
)
1328 for (i
= 0; i
< U32_HASH_SIZE
; i
++)
1329 INIT_HLIST_HEAD(&tc_u_common_hash
[i
]);
1331 ret
= register_tcf_proto_ops(&cls_u32_ops
);
1333 kvfree(tc_u_common_hash
);
1337 static void __exit
exit_u32(void)
1339 unregister_tcf_proto_ops(&cls_u32_ops
);
1340 kvfree(tc_u_common_hash
);
1343 module_init(init_u32
)
1344 module_exit(exit_u32
)
1345 MODULE_LICENSE("GPL");