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/netdevice.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
;
89 /* The 'ht' field MUST be the last field in structure to allow for
90 * more entries allocated at end of structure.
92 struct tc_u_knode __rcu
*ht
[1];
96 struct tc_u_hnode __rcu
*hlist
;
100 struct hlist_node hnode
;
104 static inline unsigned int u32_hash_fold(__be32 key
,
105 const struct tc_u32_sel
*sel
,
108 unsigned int h
= ntohl(key
& sel
->hmask
) >> fshift
;
113 static int u32_classify(struct sk_buff
*skb
, const struct tcf_proto
*tp
,
114 struct tcf_result
*res
)
117 struct tc_u_knode
*knode
;
119 } stack
[TC_U32_MAXDEPTH
];
121 struct tc_u_hnode
*ht
= rcu_dereference_bh(tp
->root
);
122 unsigned int off
= skb_network_offset(skb
);
123 struct tc_u_knode
*n
;
127 #ifdef CONFIG_CLS_U32_PERF
133 n
= rcu_dereference_bh(ht
->ht
[sel
]);
137 struct tc_u32_key
*key
= n
->sel
.keys
;
139 #ifdef CONFIG_CLS_U32_PERF
140 __this_cpu_inc(n
->pf
->rcnt
);
144 if (tc_skip_sw(n
->flags
)) {
145 n
= rcu_dereference_bh(n
->next
);
149 #ifdef CONFIG_CLS_U32_MARK
150 if ((skb
->mark
& n
->mask
) != n
->val
) {
151 n
= rcu_dereference_bh(n
->next
);
154 __this_cpu_inc(*n
->pcpu_success
);
158 for (i
= n
->sel
.nkeys
; i
> 0; i
--, key
++) {
159 int toff
= off
+ key
->off
+ (off2
& key
->offmask
);
162 if (skb_headroom(skb
) + toff
> INT_MAX
)
165 data
= skb_header_pointer(skb
, toff
, 4, &hdata
);
168 if ((*data
^ key
->val
) & key
->mask
) {
169 n
= rcu_dereference_bh(n
->next
);
172 #ifdef CONFIG_CLS_U32_PERF
173 __this_cpu_inc(n
->pf
->kcnts
[j
]);
178 ht
= rcu_dereference_bh(n
->ht_down
);
181 if (n
->sel
.flags
& TC_U32_TERMINAL
) {
184 #ifdef CONFIG_NET_CLS_IND
185 if (!tcf_match_indev(skb
, n
->ifindex
)) {
186 n
= rcu_dereference_bh(n
->next
);
190 #ifdef CONFIG_CLS_U32_PERF
191 __this_cpu_inc(n
->pf
->rhit
);
193 r
= tcf_exts_exec(skb
, &n
->exts
, res
);
195 n
= rcu_dereference_bh(n
->next
);
201 n
= rcu_dereference_bh(n
->next
);
206 if (sdepth
>= TC_U32_MAXDEPTH
)
208 stack
[sdepth
].knode
= n
;
209 stack
[sdepth
].off
= off
;
212 ht
= rcu_dereference_bh(n
->ht_down
);
217 data
= skb_header_pointer(skb
, off
+ n
->sel
.hoff
, 4,
221 sel
= ht
->divisor
& u32_hash_fold(*data
, &n
->sel
,
224 if (!(n
->sel
.flags
& (TC_U32_VAROFFSET
| TC_U32_OFFSET
| TC_U32_EAT
)))
227 if (n
->sel
.flags
& (TC_U32_OFFSET
| TC_U32_VAROFFSET
)) {
228 off2
= n
->sel
.off
+ 3;
229 if (n
->sel
.flags
& TC_U32_VAROFFSET
) {
232 data
= skb_header_pointer(skb
,
237 off2
+= ntohs(n
->sel
.offmask
& *data
) >>
242 if (n
->sel
.flags
& TC_U32_EAT
) {
253 n
= stack
[sdepth
].knode
;
254 ht
= rcu_dereference_bh(n
->ht_up
);
255 off
= stack
[sdepth
].off
;
262 net_warn_ratelimited("cls_u32: dead loop\n");
266 static struct tc_u_hnode
*u32_lookup_ht(struct tc_u_common
*tp_c
, u32 handle
)
268 struct tc_u_hnode
*ht
;
270 for (ht
= rtnl_dereference(tp_c
->hlist
);
272 ht
= rtnl_dereference(ht
->next
))
273 if (ht
->handle
== handle
)
279 static struct tc_u_knode
*u32_lookup_key(struct tc_u_hnode
*ht
, u32 handle
)
282 struct tc_u_knode
*n
= NULL
;
284 sel
= TC_U32_HASH(handle
);
285 if (sel
> ht
->divisor
)
288 for (n
= rtnl_dereference(ht
->ht
[sel
]);
290 n
= rtnl_dereference(n
->next
))
291 if (n
->handle
== handle
)
298 static void *u32_get(struct tcf_proto
*tp
, u32 handle
)
300 struct tc_u_hnode
*ht
;
301 struct tc_u_common
*tp_c
= tp
->data
;
303 if (TC_U32_HTID(handle
) == TC_U32_ROOT
)
304 ht
= rtnl_dereference(tp
->root
);
306 ht
= u32_lookup_ht(tp_c
, TC_U32_HTID(handle
));
311 if (TC_U32_KEY(handle
) == 0)
314 return u32_lookup_key(ht
, handle
);
317 static u32
gen_new_htid(struct tc_u_common
*tp_c
)
321 /* hgenerator only used inside rtnl lock it is safe to increment
322 * without read _copy_ update semantics
325 if (++tp_c
->hgenerator
== 0x7FF)
326 tp_c
->hgenerator
= 1;
327 } while (--i
> 0 && u32_lookup_ht(tp_c
, (tp_c
->hgenerator
|0x800)<<20));
329 return i
> 0 ? (tp_c
->hgenerator
|0x800)<<20 : 0;
332 static struct hlist_head
*tc_u_common_hash
;
334 #define U32_HASH_SHIFT 10
335 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
337 static unsigned int tc_u_hash(const struct tcf_proto
*tp
)
339 struct net_device
*dev
= tp
->q
->dev_queue
->dev
;
340 u32 qhandle
= tp
->q
->handle
;
341 int ifindex
= dev
->ifindex
;
343 return hash_64((u64
)ifindex
<< 32 | qhandle
, U32_HASH_SHIFT
);
346 static struct tc_u_common
*tc_u_common_find(const struct tcf_proto
*tp
)
348 struct tc_u_common
*tc
;
352 hlist_for_each_entry(tc
, &tc_u_common_hash
[h
], hnode
) {
359 static int u32_init(struct tcf_proto
*tp
)
361 struct tc_u_hnode
*root_ht
;
362 struct tc_u_common
*tp_c
;
365 tp_c
= tc_u_common_find(tp
);
367 root_ht
= kzalloc(sizeof(*root_ht
), GFP_KERNEL
);
372 root_ht
->handle
= tp_c
? gen_new_htid(tp_c
) : 0x80000000;
373 root_ht
->prio
= tp
->prio
;
376 tp_c
= kzalloc(sizeof(*tp_c
), GFP_KERNEL
);
382 INIT_HLIST_NODE(&tp_c
->hnode
);
385 hlist_add_head(&tp_c
->hnode
, &tc_u_common_hash
[h
]);
389 RCU_INIT_POINTER(root_ht
->next
, tp_c
->hlist
);
390 rcu_assign_pointer(tp_c
->hlist
, root_ht
);
391 root_ht
->tp_c
= tp_c
;
393 rcu_assign_pointer(tp
->root
, root_ht
);
398 static int u32_destroy_key(struct tcf_proto
*tp
, struct tc_u_knode
*n
,
401 tcf_exts_destroy(&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 call_rcu(&key
->rcu
, u32_delete_key_freepf_rcu
);
488 static void u32_remove_hw_knode(struct tcf_proto
*tp
, u32 handle
)
490 struct net_device
*dev
= tp
->q
->dev_queue
->dev
;
491 struct tc_cls_u32_offload cls_u32
= {};
493 if (!tc_should_offload(dev
, 0))
496 tc_cls_common_offload_init(&cls_u32
.common
, tp
);
497 cls_u32
.command
= TC_CLSU32_DELETE_KNODE
;
498 cls_u32
.knode
.handle
= handle
;
500 dev
->netdev_ops
->ndo_setup_tc(dev
, TC_SETUP_CLSU32
, &cls_u32
);
503 static int u32_replace_hw_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*h
,
506 struct net_device
*dev
= tp
->q
->dev_queue
->dev
;
507 struct tc_cls_u32_offload cls_u32
= {};
510 if (!tc_should_offload(dev
, flags
))
511 return tc_skip_sw(flags
) ? -EINVAL
: 0;
513 tc_cls_common_offload_init(&cls_u32
.common
, tp
);
514 cls_u32
.command
= TC_CLSU32_NEW_HNODE
;
515 cls_u32
.hnode
.divisor
= h
->divisor
;
516 cls_u32
.hnode
.handle
= h
->handle
;
517 cls_u32
.hnode
.prio
= h
->prio
;
519 err
= dev
->netdev_ops
->ndo_setup_tc(dev
, TC_SETUP_CLSU32
, &cls_u32
);
520 if (tc_skip_sw(flags
))
526 static void u32_clear_hw_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*h
)
528 struct net_device
*dev
= tp
->q
->dev_queue
->dev
;
529 struct tc_cls_u32_offload cls_u32
= {};
531 if (!tc_should_offload(dev
, 0))
534 tc_cls_common_offload_init(&cls_u32
.common
, tp
);
535 cls_u32
.command
= TC_CLSU32_DELETE_HNODE
;
536 cls_u32
.hnode
.divisor
= h
->divisor
;
537 cls_u32
.hnode
.handle
= h
->handle
;
538 cls_u32
.hnode
.prio
= h
->prio
;
540 dev
->netdev_ops
->ndo_setup_tc(dev
, TC_SETUP_CLSU32
, &cls_u32
);
543 static int u32_replace_hw_knode(struct tcf_proto
*tp
, struct tc_u_knode
*n
,
546 struct net_device
*dev
= tp
->q
->dev_queue
->dev
;
547 struct tc_cls_u32_offload cls_u32
= {};
550 if (!tc_should_offload(dev
, flags
))
551 return tc_skip_sw(flags
) ? -EINVAL
: 0;
553 tc_cls_common_offload_init(&cls_u32
.common
, tp
);
554 cls_u32
.command
= TC_CLSU32_REPLACE_KNODE
;
555 cls_u32
.knode
.handle
= n
->handle
;
556 cls_u32
.knode
.fshift
= n
->fshift
;
557 #ifdef CONFIG_CLS_U32_MARK
558 cls_u32
.knode
.val
= n
->val
;
559 cls_u32
.knode
.mask
= n
->mask
;
561 cls_u32
.knode
.val
= 0;
562 cls_u32
.knode
.mask
= 0;
564 cls_u32
.knode
.sel
= &n
->sel
;
565 cls_u32
.knode
.exts
= &n
->exts
;
567 cls_u32
.knode
.link_handle
= n
->ht_down
->handle
;
569 err
= dev
->netdev_ops
->ndo_setup_tc(dev
, TC_SETUP_CLSU32
, &cls_u32
);
572 n
->flags
|= TCA_CLS_FLAGS_IN_HW
;
574 if (tc_skip_sw(flags
))
580 static void u32_clear_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
)
582 struct tc_u_knode
*n
;
585 for (h
= 0; h
<= ht
->divisor
; h
++) {
586 while ((n
= rtnl_dereference(ht
->ht
[h
])) != NULL
) {
587 RCU_INIT_POINTER(ht
->ht
[h
],
588 rtnl_dereference(n
->next
));
589 tcf_unbind_filter(tp
, &n
->res
);
590 u32_remove_hw_knode(tp
, n
->handle
);
591 call_rcu(&n
->rcu
, u32_delete_key_freepf_rcu
);
596 static int u32_destroy_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
)
598 struct tc_u_common
*tp_c
= tp
->data
;
599 struct tc_u_hnode __rcu
**hn
;
600 struct tc_u_hnode
*phn
;
604 u32_clear_hnode(tp
, ht
);
607 for (phn
= rtnl_dereference(*hn
);
609 hn
= &phn
->next
, phn
= rtnl_dereference(*hn
)) {
611 u32_clear_hw_hnode(tp
, ht
);
612 RCU_INIT_POINTER(*hn
, ht
->next
);
621 static bool ht_empty(struct tc_u_hnode
*ht
)
625 for (h
= 0; h
<= ht
->divisor
; h
++)
626 if (rcu_access_pointer(ht
->ht
[h
]))
632 static void u32_destroy(struct tcf_proto
*tp
)
634 struct tc_u_common
*tp_c
= tp
->data
;
635 struct tc_u_hnode
*root_ht
= rtnl_dereference(tp
->root
);
637 WARN_ON(root_ht
== NULL
);
639 if (root_ht
&& --root_ht
->refcnt
== 0)
640 u32_destroy_hnode(tp
, root_ht
);
642 if (--tp_c
->refcnt
== 0) {
643 struct tc_u_hnode
*ht
;
645 hlist_del(&tp_c
->hnode
);
647 for (ht
= rtnl_dereference(tp_c
->hlist
);
649 ht
= rtnl_dereference(ht
->next
)) {
651 u32_clear_hnode(tp
, ht
);
654 while ((ht
= rtnl_dereference(tp_c
->hlist
)) != NULL
) {
655 RCU_INIT_POINTER(tp_c
->hlist
, ht
->next
);
665 static int u32_delete(struct tcf_proto
*tp
, void *arg
, bool *last
)
667 struct tc_u_hnode
*ht
= arg
;
668 struct tc_u_hnode
*root_ht
= rtnl_dereference(tp
->root
);
669 struct tc_u_common
*tp_c
= tp
->data
;
675 if (TC_U32_KEY(ht
->handle
)) {
676 u32_remove_hw_knode(tp
, ht
->handle
);
677 ret
= u32_delete_key(tp
, (struct tc_u_knode
*)ht
);
684 if (ht
->refcnt
== 1) {
686 u32_destroy_hnode(tp
, ht
);
694 if (root_ht
->refcnt
> 1) {
698 if (root_ht
->refcnt
== 1) {
699 if (!ht_empty(root_ht
)) {
706 if (tp_c
->refcnt
> 1) {
711 if (tp_c
->refcnt
== 1) {
712 struct tc_u_hnode
*ht
;
714 for (ht
= rtnl_dereference(tp_c
->hlist
);
716 ht
= rtnl_dereference(ht
->next
))
727 #define NR_U32_NODE (1<<12)
728 static u32
gen_new_kid(struct tc_u_hnode
*ht
, u32 handle
)
730 struct tc_u_knode
*n
;
732 unsigned long *bitmap
= kzalloc(BITS_TO_LONGS(NR_U32_NODE
) * sizeof(unsigned long),
735 return handle
| 0xFFF;
737 for (n
= rtnl_dereference(ht
->ht
[TC_U32_HASH(handle
)]);
739 n
= rtnl_dereference(n
->next
))
740 set_bit(TC_U32_NODE(n
->handle
), bitmap
);
742 i
= find_next_zero_bit(bitmap
, NR_U32_NODE
, 0x800);
743 if (i
>= NR_U32_NODE
)
744 i
= find_next_zero_bit(bitmap
, NR_U32_NODE
, 1);
747 return handle
| (i
>= NR_U32_NODE
? 0xFFF : i
);
750 static const struct nla_policy u32_policy
[TCA_U32_MAX
+ 1] = {
751 [TCA_U32_CLASSID
] = { .type
= NLA_U32
},
752 [TCA_U32_HASH
] = { .type
= NLA_U32
},
753 [TCA_U32_LINK
] = { .type
= NLA_U32
},
754 [TCA_U32_DIVISOR
] = { .type
= NLA_U32
},
755 [TCA_U32_SEL
] = { .len
= sizeof(struct tc_u32_sel
) },
756 [TCA_U32_INDEV
] = { .type
= NLA_STRING
, .len
= IFNAMSIZ
},
757 [TCA_U32_MARK
] = { .len
= sizeof(struct tc_u32_mark
) },
758 [TCA_U32_FLAGS
] = { .type
= NLA_U32
},
761 static int u32_set_parms(struct net
*net
, struct tcf_proto
*tp
,
762 unsigned long base
, struct tc_u_hnode
*ht
,
763 struct tc_u_knode
*n
, struct nlattr
**tb
,
764 struct nlattr
*est
, bool ovr
)
768 err
= tcf_exts_validate(net
, tp
, tb
, est
, &n
->exts
, ovr
);
772 if (tb
[TCA_U32_LINK
]) {
773 u32 handle
= nla_get_u32(tb
[TCA_U32_LINK
]);
774 struct tc_u_hnode
*ht_down
= NULL
, *ht_old
;
776 if (TC_U32_KEY(handle
))
780 ht_down
= u32_lookup_ht(ht
->tp_c
, handle
);
787 ht_old
= rtnl_dereference(n
->ht_down
);
788 rcu_assign_pointer(n
->ht_down
, ht_down
);
793 if (tb
[TCA_U32_CLASSID
]) {
794 n
->res
.classid
= nla_get_u32(tb
[TCA_U32_CLASSID
]);
795 tcf_bind_filter(tp
, &n
->res
, base
);
798 #ifdef CONFIG_NET_CLS_IND
799 if (tb
[TCA_U32_INDEV
]) {
801 ret
= tcf_change_indev(net
, tb
[TCA_U32_INDEV
]);
810 static void u32_replace_knode(struct tcf_proto
*tp
, struct tc_u_common
*tp_c
,
811 struct tc_u_knode
*n
)
813 struct tc_u_knode __rcu
**ins
;
814 struct tc_u_knode
*pins
;
815 struct tc_u_hnode
*ht
;
817 if (TC_U32_HTID(n
->handle
) == TC_U32_ROOT
)
818 ht
= rtnl_dereference(tp
->root
);
820 ht
= u32_lookup_ht(tp_c
, TC_U32_HTID(n
->handle
));
822 ins
= &ht
->ht
[TC_U32_HASH(n
->handle
)];
824 /* The node must always exist for it to be replaced if this is not the
825 * case then something went very wrong elsewhere.
827 for (pins
= rtnl_dereference(*ins
); ;
828 ins
= &pins
->next
, pins
= rtnl_dereference(*ins
))
829 if (pins
->handle
== n
->handle
)
832 RCU_INIT_POINTER(n
->next
, pins
->next
);
833 rcu_assign_pointer(*ins
, n
);
836 static struct tc_u_knode
*u32_init_knode(struct tcf_proto
*tp
,
837 struct tc_u_knode
*n
)
839 struct tc_u_knode
*new;
840 struct tc_u32_sel
*s
= &n
->sel
;
842 new = kzalloc(sizeof(*n
) + s
->nkeys
*sizeof(struct tc_u32_key
),
848 RCU_INIT_POINTER(new->next
, n
->next
);
849 new->handle
= n
->handle
;
850 RCU_INIT_POINTER(new->ht_up
, n
->ht_up
);
852 #ifdef CONFIG_NET_CLS_IND
853 new->ifindex
= n
->ifindex
;
855 new->fshift
= n
->fshift
;
857 new->flags
= n
->flags
;
858 RCU_INIT_POINTER(new->ht_down
, n
->ht_down
);
860 /* bump reference count as long as we hold pointer to structure */
862 new->ht_down
->refcnt
++;
864 #ifdef CONFIG_CLS_U32_PERF
865 /* Statistics may be incremented by readers during update
866 * so we must keep them in tact. When the node is later destroyed
867 * a special destroy call must be made to not free the pf memory.
872 #ifdef CONFIG_CLS_U32_MARK
875 /* Similarly success statistics must be moved as pointers */
876 new->pcpu_success
= n
->pcpu_success
;
879 memcpy(&new->sel
, s
, sizeof(*s
) + s
->nkeys
*sizeof(struct tc_u32_key
));
881 if (tcf_exts_init(&new->exts
, TCA_U32_ACT
, TCA_U32_POLICE
)) {
889 static int u32_change(struct net
*net
, struct sk_buff
*in_skb
,
890 struct tcf_proto
*tp
, unsigned long base
, u32 handle
,
891 struct nlattr
**tca
, void **arg
, bool ovr
)
893 struct tc_u_common
*tp_c
= tp
->data
;
894 struct tc_u_hnode
*ht
;
895 struct tc_u_knode
*n
;
896 struct tc_u32_sel
*s
;
897 struct nlattr
*opt
= tca
[TCA_OPTIONS
];
898 struct nlattr
*tb
[TCA_U32_MAX
+ 1];
901 #ifdef CONFIG_CLS_U32_PERF
906 return handle
? -EINVAL
: 0;
908 err
= nla_parse_nested(tb
, TCA_U32_MAX
, opt
, u32_policy
, NULL
);
912 if (tb
[TCA_U32_FLAGS
]) {
913 flags
= nla_get_u32(tb
[TCA_U32_FLAGS
]);
914 if (!tc_flags_valid(flags
))
920 struct tc_u_knode
*new;
922 if (TC_U32_KEY(n
->handle
) == 0)
925 if (n
->flags
!= flags
)
928 new = u32_init_knode(tp
, n
);
932 err
= u32_set_parms(net
, tp
, base
,
933 rtnl_dereference(n
->ht_up
), new, tb
,
937 u32_destroy_key(tp
, new, false);
941 err
= u32_replace_hw_knode(tp
, new, flags
);
943 u32_destroy_key(tp
, new, false);
947 if (!tc_in_hw(new->flags
))
948 new->flags
|= TCA_CLS_FLAGS_NOT_IN_HW
;
950 u32_replace_knode(tp
, tp_c
, new);
951 tcf_unbind_filter(tp
, &n
->res
);
952 call_rcu(&n
->rcu
, u32_delete_key_rcu
);
956 if (tb
[TCA_U32_DIVISOR
]) {
957 unsigned int divisor
= nla_get_u32(tb
[TCA_U32_DIVISOR
]);
959 if (--divisor
> 0x100)
961 if (TC_U32_KEY(handle
))
964 handle
= gen_new_htid(tp
->data
);
968 ht
= kzalloc(sizeof(*ht
) + divisor
*sizeof(void *), GFP_KERNEL
);
973 ht
->divisor
= divisor
;
977 err
= u32_replace_hw_hnode(tp
, ht
, flags
);
983 RCU_INIT_POINTER(ht
->next
, tp_c
->hlist
);
984 rcu_assign_pointer(tp_c
->hlist
, ht
);
990 if (tb
[TCA_U32_HASH
]) {
991 htid
= nla_get_u32(tb
[TCA_U32_HASH
]);
992 if (TC_U32_HTID(htid
) == TC_U32_ROOT
) {
993 ht
= rtnl_dereference(tp
->root
);
996 ht
= u32_lookup_ht(tp
->data
, TC_U32_HTID(htid
));
1001 ht
= rtnl_dereference(tp
->root
);
1005 if (ht
->divisor
< TC_U32_HASH(htid
))
1009 if (TC_U32_HTID(handle
) && TC_U32_HTID(handle
^htid
))
1011 handle
= htid
| TC_U32_NODE(handle
);
1013 handle
= gen_new_kid(ht
, htid
);
1015 if (tb
[TCA_U32_SEL
] == NULL
)
1018 s
= nla_data(tb
[TCA_U32_SEL
]);
1020 n
= kzalloc(sizeof(*n
) + s
->nkeys
*sizeof(struct tc_u32_key
), GFP_KERNEL
);
1024 #ifdef CONFIG_CLS_U32_PERF
1025 size
= sizeof(struct tc_u32_pcnt
) + s
->nkeys
* sizeof(u64
);
1026 n
->pf
= __alloc_percpu(size
, __alignof__(struct tc_u32_pcnt
));
1033 memcpy(&n
->sel
, s
, sizeof(*s
) + s
->nkeys
*sizeof(struct tc_u32_key
));
1034 RCU_INIT_POINTER(n
->ht_up
, ht
);
1036 n
->fshift
= s
->hmask
? ffs(ntohl(s
->hmask
)) - 1 : 0;
1040 err
= tcf_exts_init(&n
->exts
, TCA_U32_ACT
, TCA_U32_POLICE
);
1044 #ifdef CONFIG_CLS_U32_MARK
1045 n
->pcpu_success
= alloc_percpu(u32
);
1046 if (!n
->pcpu_success
) {
1051 if (tb
[TCA_U32_MARK
]) {
1052 struct tc_u32_mark
*mark
;
1054 mark
= nla_data(tb
[TCA_U32_MARK
]);
1056 n
->mask
= mark
->mask
;
1060 err
= u32_set_parms(net
, tp
, base
, ht
, n
, tb
, tca
[TCA_RATE
], ovr
);
1062 struct tc_u_knode __rcu
**ins
;
1063 struct tc_u_knode
*pins
;
1065 err
= u32_replace_hw_knode(tp
, n
, flags
);
1069 if (!tc_in_hw(n
->flags
))
1070 n
->flags
|= TCA_CLS_FLAGS_NOT_IN_HW
;
1072 ins
= &ht
->ht
[TC_U32_HASH(handle
)];
1073 for (pins
= rtnl_dereference(*ins
); pins
;
1074 ins
= &pins
->next
, pins
= rtnl_dereference(*ins
))
1075 if (TC_U32_NODE(handle
) < TC_U32_NODE(pins
->handle
))
1078 RCU_INIT_POINTER(n
->next
, pins
);
1079 rcu_assign_pointer(*ins
, n
);
1085 #ifdef CONFIG_CLS_U32_MARK
1086 free_percpu(n
->pcpu_success
);
1090 tcf_exts_destroy(&n
->exts
);
1091 #ifdef CONFIG_CLS_U32_PERF
1098 static void u32_walk(struct tcf_proto
*tp
, struct tcf_walker
*arg
)
1100 struct tc_u_common
*tp_c
= tp
->data
;
1101 struct tc_u_hnode
*ht
;
1102 struct tc_u_knode
*n
;
1108 for (ht
= rtnl_dereference(tp_c
->hlist
);
1110 ht
= rtnl_dereference(ht
->next
)) {
1111 if (ht
->prio
!= tp
->prio
)
1113 if (arg
->count
>= arg
->skip
) {
1114 if (arg
->fn(tp
, ht
, arg
) < 0) {
1120 for (h
= 0; h
<= ht
->divisor
; h
++) {
1121 for (n
= rtnl_dereference(ht
->ht
[h
]);
1123 n
= rtnl_dereference(n
->next
)) {
1124 if (arg
->count
< arg
->skip
) {
1128 if (arg
->fn(tp
, n
, arg
) < 0) {
1138 static void u32_bind_class(void *fh
, u32 classid
, unsigned long cl
)
1140 struct tc_u_knode
*n
= fh
;
1142 if (n
&& n
->res
.classid
== classid
)
1146 static int u32_dump(struct net
*net
, struct tcf_proto
*tp
, void *fh
,
1147 struct sk_buff
*skb
, struct tcmsg
*t
)
1149 struct tc_u_knode
*n
= fh
;
1150 struct tc_u_hnode
*ht_up
, *ht_down
;
1151 struct nlattr
*nest
;
1156 t
->tcm_handle
= n
->handle
;
1158 nest
= nla_nest_start(skb
, TCA_OPTIONS
);
1160 goto nla_put_failure
;
1162 if (TC_U32_KEY(n
->handle
) == 0) {
1163 struct tc_u_hnode
*ht
= fh
;
1164 u32 divisor
= ht
->divisor
+ 1;
1166 if (nla_put_u32(skb
, TCA_U32_DIVISOR
, divisor
))
1167 goto nla_put_failure
;
1169 #ifdef CONFIG_CLS_U32_PERF
1170 struct tc_u32_pcnt
*gpf
;
1174 if (nla_put(skb
, TCA_U32_SEL
,
1175 sizeof(n
->sel
) + n
->sel
.nkeys
*sizeof(struct tc_u32_key
),
1177 goto nla_put_failure
;
1179 ht_up
= rtnl_dereference(n
->ht_up
);
1181 u32 htid
= n
->handle
& 0xFFFFF000;
1182 if (nla_put_u32(skb
, TCA_U32_HASH
, htid
))
1183 goto nla_put_failure
;
1185 if (n
->res
.classid
&&
1186 nla_put_u32(skb
, TCA_U32_CLASSID
, n
->res
.classid
))
1187 goto nla_put_failure
;
1189 ht_down
= rtnl_dereference(n
->ht_down
);
1191 nla_put_u32(skb
, TCA_U32_LINK
, ht_down
->handle
))
1192 goto nla_put_failure
;
1194 if (n
->flags
&& nla_put_u32(skb
, TCA_U32_FLAGS
, n
->flags
))
1195 goto nla_put_failure
;
1197 #ifdef CONFIG_CLS_U32_MARK
1198 if ((n
->val
|| n
->mask
)) {
1199 struct tc_u32_mark mark
= {.val
= n
->val
,
1204 for_each_possible_cpu(cpum
) {
1205 __u32 cnt
= *per_cpu_ptr(n
->pcpu_success
, cpum
);
1207 mark
.success
+= cnt
;
1210 if (nla_put(skb
, TCA_U32_MARK
, sizeof(mark
), &mark
))
1211 goto nla_put_failure
;
1215 if (tcf_exts_dump(skb
, &n
->exts
) < 0)
1216 goto nla_put_failure
;
1218 #ifdef CONFIG_NET_CLS_IND
1220 struct net_device
*dev
;
1221 dev
= __dev_get_by_index(net
, n
->ifindex
);
1222 if (dev
&& nla_put_string(skb
, TCA_U32_INDEV
, dev
->name
))
1223 goto nla_put_failure
;
1226 #ifdef CONFIG_CLS_U32_PERF
1227 gpf
= kzalloc(sizeof(struct tc_u32_pcnt
) +
1228 n
->sel
.nkeys
* sizeof(u64
),
1231 goto nla_put_failure
;
1233 for_each_possible_cpu(cpu
) {
1235 struct tc_u32_pcnt
*pf
= per_cpu_ptr(n
->pf
, cpu
);
1237 gpf
->rcnt
+= pf
->rcnt
;
1238 gpf
->rhit
+= pf
->rhit
;
1239 for (i
= 0; i
< n
->sel
.nkeys
; i
++)
1240 gpf
->kcnts
[i
] += pf
->kcnts
[i
];
1243 if (nla_put_64bit(skb
, TCA_U32_PCNT
,
1244 sizeof(struct tc_u32_pcnt
) +
1245 n
->sel
.nkeys
* sizeof(u64
),
1246 gpf
, TCA_U32_PAD
)) {
1248 goto nla_put_failure
;
1254 nla_nest_end(skb
, nest
);
1256 if (TC_U32_KEY(n
->handle
))
1257 if (tcf_exts_dump_stats(skb
, &n
->exts
) < 0)
1258 goto nla_put_failure
;
1262 nla_nest_cancel(skb
, nest
);
1266 static struct tcf_proto_ops cls_u32_ops __read_mostly
= {
1268 .classify
= u32_classify
,
1270 .destroy
= u32_destroy
,
1272 .change
= u32_change
,
1273 .delete = u32_delete
,
1276 .bind_class
= u32_bind_class
,
1277 .owner
= THIS_MODULE
,
1280 static int __init
init_u32(void)
1284 pr_info("u32 classifier\n");
1285 #ifdef CONFIG_CLS_U32_PERF
1286 pr_info(" Performance counters on\n");
1288 #ifdef CONFIG_NET_CLS_IND
1289 pr_info(" input device check on\n");
1291 #ifdef CONFIG_NET_CLS_ACT
1292 pr_info(" Actions configured\n");
1294 tc_u_common_hash
= kvmalloc_array(U32_HASH_SIZE
,
1295 sizeof(struct hlist_head
),
1297 if (!tc_u_common_hash
)
1300 for (i
= 0; i
< U32_HASH_SIZE
; i
++)
1301 INIT_HLIST_HEAD(&tc_u_common_hash
[i
]);
1303 ret
= register_tcf_proto_ops(&cls_u32_ops
);
1305 kvfree(tc_u_common_hash
);
1309 static void __exit
exit_u32(void)
1311 unregister_tcf_proto_ops(&cls_u32_ops
);
1312 kvfree(tc_u_common_hash
);
1315 module_init(init_u32
)
1316 module_exit(exit_u32
)
1317 MODULE_LICENSE("GPL");