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 unsigned int in_hw_count
;
66 #ifdef CONFIG_CLS_U32_MARK
69 u32 __percpu
*pcpu_success
;
71 struct rcu_work rwork
;
72 /* The 'sel' field MUST be the last field in structure to allow for
73 * tc_u32_keys allocated at end of structure.
75 struct tc_u32_sel sel
;
79 struct tc_u_hnode __rcu
*next
;
84 struct idr handle_idr
;
88 /* The 'ht' field MUST be the last field in structure to allow for
89 * more entries allocated at end of structure.
91 struct tc_u_knode __rcu
*ht
[1];
95 struct tc_u_hnode __rcu
*hlist
;
98 struct idr handle_idr
;
99 struct hlist_node hnode
;
103 static inline unsigned int u32_hash_fold(__be32 key
,
104 const struct tc_u32_sel
*sel
,
107 unsigned int h
= ntohl(key
& sel
->hmask
) >> fshift
;
112 static int u32_classify(struct sk_buff
*skb
, const struct tcf_proto
*tp
,
113 struct tcf_result
*res
)
116 struct tc_u_knode
*knode
;
118 } stack
[TC_U32_MAXDEPTH
];
120 struct tc_u_hnode
*ht
= rcu_dereference_bh(tp
->root
);
121 unsigned int off
= skb_network_offset(skb
);
122 struct tc_u_knode
*n
;
126 #ifdef CONFIG_CLS_U32_PERF
132 n
= rcu_dereference_bh(ht
->ht
[sel
]);
136 struct tc_u32_key
*key
= n
->sel
.keys
;
138 #ifdef CONFIG_CLS_U32_PERF
139 __this_cpu_inc(n
->pf
->rcnt
);
143 if (tc_skip_sw(n
->flags
)) {
144 n
= rcu_dereference_bh(n
->next
);
148 #ifdef CONFIG_CLS_U32_MARK
149 if ((skb
->mark
& n
->mask
) != n
->val
) {
150 n
= rcu_dereference_bh(n
->next
);
153 __this_cpu_inc(*n
->pcpu_success
);
157 for (i
= n
->sel
.nkeys
; i
> 0; i
--, key
++) {
158 int toff
= off
+ key
->off
+ (off2
& key
->offmask
);
161 if (skb_headroom(skb
) + toff
> INT_MAX
)
164 data
= skb_header_pointer(skb
, toff
, 4, &hdata
);
167 if ((*data
^ key
->val
) & key
->mask
) {
168 n
= rcu_dereference_bh(n
->next
);
171 #ifdef CONFIG_CLS_U32_PERF
172 __this_cpu_inc(n
->pf
->kcnts
[j
]);
177 ht
= rcu_dereference_bh(n
->ht_down
);
180 if (n
->sel
.flags
& TC_U32_TERMINAL
) {
183 #ifdef CONFIG_NET_CLS_IND
184 if (!tcf_match_indev(skb
, n
->ifindex
)) {
185 n
= rcu_dereference_bh(n
->next
);
189 #ifdef CONFIG_CLS_U32_PERF
190 __this_cpu_inc(n
->pf
->rhit
);
192 r
= tcf_exts_exec(skb
, &n
->exts
, res
);
194 n
= rcu_dereference_bh(n
->next
);
200 n
= rcu_dereference_bh(n
->next
);
205 if (sdepth
>= TC_U32_MAXDEPTH
)
207 stack
[sdepth
].knode
= n
;
208 stack
[sdepth
].off
= off
;
211 ht
= rcu_dereference_bh(n
->ht_down
);
216 data
= skb_header_pointer(skb
, off
+ n
->sel
.hoff
, 4,
220 sel
= ht
->divisor
& u32_hash_fold(*data
, &n
->sel
,
223 if (!(n
->sel
.flags
& (TC_U32_VAROFFSET
| TC_U32_OFFSET
| TC_U32_EAT
)))
226 if (n
->sel
.flags
& (TC_U32_OFFSET
| TC_U32_VAROFFSET
)) {
227 off2
= n
->sel
.off
+ 3;
228 if (n
->sel
.flags
& TC_U32_VAROFFSET
) {
231 data
= skb_header_pointer(skb
,
236 off2
+= ntohs(n
->sel
.offmask
& *data
) >>
241 if (n
->sel
.flags
& TC_U32_EAT
) {
252 n
= stack
[sdepth
].knode
;
253 ht
= rcu_dereference_bh(n
->ht_up
);
254 off
= stack
[sdepth
].off
;
261 net_warn_ratelimited("cls_u32: dead loop\n");
265 static struct tc_u_hnode
*u32_lookup_ht(struct tc_u_common
*tp_c
, u32 handle
)
267 struct tc_u_hnode
*ht
;
269 for (ht
= rtnl_dereference(tp_c
->hlist
);
271 ht
= rtnl_dereference(ht
->next
))
272 if (ht
->handle
== handle
)
278 static struct tc_u_knode
*u32_lookup_key(struct tc_u_hnode
*ht
, u32 handle
)
281 struct tc_u_knode
*n
= NULL
;
283 sel
= TC_U32_HASH(handle
);
284 if (sel
> ht
->divisor
)
287 for (n
= rtnl_dereference(ht
->ht
[sel
]);
289 n
= rtnl_dereference(n
->next
))
290 if (n
->handle
== handle
)
297 static void *u32_get(struct tcf_proto
*tp
, u32 handle
)
299 struct tc_u_hnode
*ht
;
300 struct tc_u_common
*tp_c
= tp
->data
;
302 if (TC_U32_HTID(handle
) == TC_U32_ROOT
)
303 ht
= rtnl_dereference(tp
->root
);
305 ht
= u32_lookup_ht(tp_c
, TC_U32_HTID(handle
));
310 if (TC_U32_KEY(handle
) == 0)
313 return u32_lookup_key(ht
, handle
);
316 /* Protected by rtnl lock */
317 static u32
gen_new_htid(struct tc_u_common
*tp_c
, struct tc_u_hnode
*ptr
)
319 int id
= idr_alloc_cyclic(&tp_c
->handle_idr
, ptr
, 1, 0x7FF, GFP_KERNEL
);
322 return (id
| 0x800U
) << 20;
325 static struct hlist_head
*tc_u_common_hash
;
327 #define U32_HASH_SHIFT 10
328 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
330 static void *tc_u_common_ptr(const struct tcf_proto
*tp
)
332 struct tcf_block
*block
= tp
->chain
->block
;
334 /* The block sharing is currently supported only
335 * for classless qdiscs. In that case we use block
336 * for tc_u_common identification. In case the
337 * block is not shared, block->q is a valid pointer
338 * and we can use that. That works for classful qdiscs.
340 if (tcf_block_shared(block
))
346 static struct hlist_head
*tc_u_hash(void *key
)
348 return tc_u_common_hash
+ hash_ptr(key
, U32_HASH_SHIFT
);
351 static struct tc_u_common
*tc_u_common_find(void *key
)
353 struct tc_u_common
*tc
;
354 hlist_for_each_entry(tc
, tc_u_hash(key
), hnode
) {
361 static int u32_init(struct tcf_proto
*tp
)
363 struct tc_u_hnode
*root_ht
;
364 void *key
= tc_u_common_ptr(tp
);
365 struct tc_u_common
*tp_c
= tc_u_common_find(key
);
367 root_ht
= kzalloc(sizeof(*root_ht
), GFP_KERNEL
);
372 root_ht
->handle
= tp_c
? gen_new_htid(tp_c
, root_ht
) : 0x80000000;
373 root_ht
->prio
= tp
->prio
;
374 root_ht
->is_root
= true;
375 idr_init(&root_ht
->handle_idr
);
378 tp_c
= kzalloc(sizeof(*tp_c
), GFP_KERNEL
);
384 INIT_HLIST_NODE(&tp_c
->hnode
);
385 idr_init(&tp_c
->handle_idr
);
387 hlist_add_head(&tp_c
->hnode
, tc_u_hash(key
));
391 RCU_INIT_POINTER(root_ht
->next
, tp_c
->hlist
);
392 rcu_assign_pointer(tp_c
->hlist
, root_ht
);
395 rcu_assign_pointer(tp
->root
, root_ht
);
400 static int u32_destroy_key(struct tc_u_knode
*n
, bool free_pf
)
402 struct tc_u_hnode
*ht
= rtnl_dereference(n
->ht_down
);
404 tcf_exts_destroy(&n
->exts
);
405 tcf_exts_put_net(&n
->exts
);
406 if (ht
&& --ht
->refcnt
== 0)
408 #ifdef CONFIG_CLS_U32_PERF
412 #ifdef CONFIG_CLS_U32_MARK
414 free_percpu(n
->pcpu_success
);
420 /* u32_delete_key_rcu should be called when free'ing a copied
421 * version of a tc_u_knode obtained from u32_init_knode(). When
422 * copies are obtained from u32_init_knode() the statistics are
423 * shared between the old and new copies to allow readers to
424 * continue to update the statistics during the copy. To support
425 * this the u32_delete_key_rcu variant does not free the percpu
428 static void u32_delete_key_work(struct work_struct
*work
)
430 struct tc_u_knode
*key
= container_of(to_rcu_work(work
),
434 u32_destroy_key(key
, false);
438 /* u32_delete_key_freepf_rcu is the rcu callback variant
439 * that free's the entire structure including the statistics
440 * percpu variables. Only use this if the key is not a copy
441 * returned by u32_init_knode(). See u32_delete_key_rcu()
442 * for the variant that should be used with keys return from
445 static void u32_delete_key_freepf_work(struct work_struct
*work
)
447 struct tc_u_knode
*key
= container_of(to_rcu_work(work
),
451 u32_destroy_key(key
, true);
455 static int u32_delete_key(struct tcf_proto
*tp
, struct tc_u_knode
*key
)
457 struct tc_u_common
*tp_c
= tp
->data
;
458 struct tc_u_knode __rcu
**kp
;
459 struct tc_u_knode
*pkp
;
460 struct tc_u_hnode
*ht
= rtnl_dereference(key
->ht_up
);
463 kp
= &ht
->ht
[TC_U32_HASH(key
->handle
)];
464 for (pkp
= rtnl_dereference(*kp
); pkp
;
465 kp
= &pkp
->next
, pkp
= rtnl_dereference(*kp
)) {
467 RCU_INIT_POINTER(*kp
, key
->next
);
470 tcf_unbind_filter(tp
, &key
->res
);
471 idr_remove(&ht
->handle_idr
, key
->handle
);
472 tcf_exts_get_net(&key
->exts
);
473 tcf_queue_work(&key
->rwork
, u32_delete_key_freepf_work
);
482 static void u32_clear_hw_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*h
,
483 struct netlink_ext_ack
*extack
)
485 struct tcf_block
*block
= tp
->chain
->block
;
486 struct tc_cls_u32_offload cls_u32
= {};
488 tc_cls_common_offload_init(&cls_u32
.common
, tp
, h
->flags
, extack
);
489 cls_u32
.command
= TC_CLSU32_DELETE_HNODE
;
490 cls_u32
.hnode
.divisor
= h
->divisor
;
491 cls_u32
.hnode
.handle
= h
->handle
;
492 cls_u32
.hnode
.prio
= h
->prio
;
494 tc_setup_cb_call(block
, NULL
, TC_SETUP_CLSU32
, &cls_u32
, false);
497 static int u32_replace_hw_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*h
,
498 u32 flags
, struct netlink_ext_ack
*extack
)
500 struct tcf_block
*block
= tp
->chain
->block
;
501 struct tc_cls_u32_offload cls_u32
= {};
502 bool skip_sw
= tc_skip_sw(flags
);
503 bool offloaded
= false;
506 tc_cls_common_offload_init(&cls_u32
.common
, tp
, flags
, extack
);
507 cls_u32
.command
= TC_CLSU32_NEW_HNODE
;
508 cls_u32
.hnode
.divisor
= h
->divisor
;
509 cls_u32
.hnode
.handle
= h
->handle
;
510 cls_u32
.hnode
.prio
= h
->prio
;
512 err
= tc_setup_cb_call(block
, NULL
, TC_SETUP_CLSU32
, &cls_u32
, skip_sw
);
514 u32_clear_hw_hnode(tp
, h
, NULL
);
516 } else if (err
> 0) {
520 if (skip_sw
&& !offloaded
)
526 static void u32_remove_hw_knode(struct tcf_proto
*tp
, struct tc_u_knode
*n
,
527 struct netlink_ext_ack
*extack
)
529 struct tcf_block
*block
= tp
->chain
->block
;
530 struct tc_cls_u32_offload cls_u32
= {};
532 tc_cls_common_offload_init(&cls_u32
.common
, tp
, n
->flags
, extack
);
533 cls_u32
.command
= TC_CLSU32_DELETE_KNODE
;
534 cls_u32
.knode
.handle
= n
->handle
;
536 tc_setup_cb_call(block
, NULL
, TC_SETUP_CLSU32
, &cls_u32
, false);
537 tcf_block_offload_dec(block
, &n
->flags
);
540 static int u32_replace_hw_knode(struct tcf_proto
*tp
, struct tc_u_knode
*n
,
541 u32 flags
, struct netlink_ext_ack
*extack
)
543 struct tc_u_hnode
*ht
= rtnl_dereference(n
->ht_down
);
544 struct tcf_block
*block
= tp
->chain
->block
;
545 struct tc_cls_u32_offload cls_u32
= {};
546 bool skip_sw
= tc_skip_sw(flags
);
549 tc_cls_common_offload_init(&cls_u32
.common
, tp
, flags
, extack
);
550 cls_u32
.command
= TC_CLSU32_REPLACE_KNODE
;
551 cls_u32
.knode
.handle
= n
->handle
;
552 cls_u32
.knode
.fshift
= n
->fshift
;
553 #ifdef CONFIG_CLS_U32_MARK
554 cls_u32
.knode
.val
= n
->val
;
555 cls_u32
.knode
.mask
= n
->mask
;
557 cls_u32
.knode
.val
= 0;
558 cls_u32
.knode
.mask
= 0;
560 cls_u32
.knode
.sel
= &n
->sel
;
561 cls_u32
.knode
.exts
= &n
->exts
;
563 cls_u32
.knode
.link_handle
= ht
->handle
;
565 err
= tc_setup_cb_call(block
, NULL
, TC_SETUP_CLSU32
, &cls_u32
, skip_sw
);
567 u32_remove_hw_knode(tp
, n
, NULL
);
569 } else if (err
> 0) {
570 n
->in_hw_count
= err
;
571 tcf_block_offload_inc(block
, &n
->flags
);
574 if (skip_sw
&& !(n
->flags
& TCA_CLS_FLAGS_IN_HW
))
580 static void u32_clear_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
,
581 struct netlink_ext_ack
*extack
)
583 struct tc_u_common
*tp_c
= tp
->data
;
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
));
592 tcf_unbind_filter(tp
, &n
->res
);
593 u32_remove_hw_knode(tp
, n
, extack
);
594 idr_remove(&ht
->handle_idr
, n
->handle
);
595 if (tcf_exts_get_net(&n
->exts
))
596 tcf_queue_work(&n
->rwork
, u32_delete_key_freepf_work
);
598 u32_destroy_key(n
, true);
603 static int u32_destroy_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
,
604 struct netlink_ext_ack
*extack
)
606 struct tc_u_common
*tp_c
= tp
->data
;
607 struct tc_u_hnode __rcu
**hn
;
608 struct tc_u_hnode
*phn
;
610 WARN_ON(--ht
->refcnt
);
612 u32_clear_hnode(tp
, ht
, extack
);
615 for (phn
= rtnl_dereference(*hn
);
617 hn
= &phn
->next
, phn
= rtnl_dereference(*hn
)) {
619 u32_clear_hw_hnode(tp
, ht
, extack
);
620 idr_destroy(&ht
->handle_idr
);
621 idr_remove(&tp_c
->handle_idr
, ht
->handle
);
622 RCU_INIT_POINTER(*hn
, ht
->next
);
631 static void u32_destroy(struct tcf_proto
*tp
, struct netlink_ext_ack
*extack
)
633 struct tc_u_common
*tp_c
= tp
->data
;
634 struct tc_u_hnode
*root_ht
= rtnl_dereference(tp
->root
);
636 WARN_ON(root_ht
== NULL
);
638 if (root_ht
&& --root_ht
->refcnt
== 1)
639 u32_destroy_hnode(tp
, root_ht
, extack
);
641 if (--tp_c
->refcnt
== 0) {
642 struct tc_u_hnode
*ht
;
644 hlist_del(&tp_c
->hnode
);
646 while ((ht
= rtnl_dereference(tp_c
->hlist
)) != NULL
) {
647 u32_clear_hnode(tp
, ht
, extack
);
648 RCU_INIT_POINTER(tp_c
->hlist
, ht
->next
);
650 /* u32_destroy_key() will later free ht for us, if it's
651 * still referenced by some knode
653 if (--ht
->refcnt
== 0)
657 idr_destroy(&tp_c
->handle_idr
);
664 static int u32_delete(struct tcf_proto
*tp
, void *arg
, bool *last
,
665 struct netlink_ext_ack
*extack
)
667 struct tc_u_hnode
*ht
= arg
;
668 struct tc_u_common
*tp_c
= tp
->data
;
671 if (TC_U32_KEY(ht
->handle
)) {
672 u32_remove_hw_knode(tp
, (struct tc_u_knode
*)ht
, extack
);
673 ret
= u32_delete_key(tp
, (struct tc_u_knode
*)ht
);
678 NL_SET_ERR_MSG_MOD(extack
, "Not allowed to delete root node");
682 if (ht
->refcnt
== 1) {
683 u32_destroy_hnode(tp
, ht
, extack
);
685 NL_SET_ERR_MSG_MOD(extack
, "Can not delete in-use filter");
690 *last
= tp_c
->refcnt
== 1 && tp_c
->knodes
== 0;
694 static u32
gen_new_kid(struct tc_u_hnode
*ht
, u32 htid
)
696 u32 index
= htid
| 0x800;
697 u32 max
= htid
| 0xFFF;
699 if (idr_alloc_u32(&ht
->handle_idr
, NULL
, &index
, max
, GFP_KERNEL
)) {
701 if (idr_alloc_u32(&ht
->handle_idr
, NULL
, &index
, max
,
709 static const struct nla_policy u32_policy
[TCA_U32_MAX
+ 1] = {
710 [TCA_U32_CLASSID
] = { .type
= NLA_U32
},
711 [TCA_U32_HASH
] = { .type
= NLA_U32
},
712 [TCA_U32_LINK
] = { .type
= NLA_U32
},
713 [TCA_U32_DIVISOR
] = { .type
= NLA_U32
},
714 [TCA_U32_SEL
] = { .len
= sizeof(struct tc_u32_sel
) },
715 [TCA_U32_INDEV
] = { .type
= NLA_STRING
, .len
= IFNAMSIZ
},
716 [TCA_U32_MARK
] = { .len
= sizeof(struct tc_u32_mark
) },
717 [TCA_U32_FLAGS
] = { .type
= NLA_U32
},
720 static int u32_set_parms(struct net
*net
, struct tcf_proto
*tp
,
722 struct tc_u_knode
*n
, struct nlattr
**tb
,
723 struct nlattr
*est
, bool ovr
,
724 struct netlink_ext_ack
*extack
)
728 err
= tcf_exts_validate(net
, tp
, tb
, est
, &n
->exts
, ovr
, extack
);
732 if (tb
[TCA_U32_LINK
]) {
733 u32 handle
= nla_get_u32(tb
[TCA_U32_LINK
]);
734 struct tc_u_hnode
*ht_down
= NULL
, *ht_old
;
736 if (TC_U32_KEY(handle
)) {
737 NL_SET_ERR_MSG_MOD(extack
, "u32 Link handle must be a hash table");
742 ht_down
= u32_lookup_ht(tp
->data
, handle
);
745 NL_SET_ERR_MSG_MOD(extack
, "Link hash table not found");
748 if (ht_down
->is_root
) {
749 NL_SET_ERR_MSG_MOD(extack
, "Not linking to root node");
755 ht_old
= rtnl_dereference(n
->ht_down
);
756 rcu_assign_pointer(n
->ht_down
, ht_down
);
761 if (tb
[TCA_U32_CLASSID
]) {
762 n
->res
.classid
= nla_get_u32(tb
[TCA_U32_CLASSID
]);
763 tcf_bind_filter(tp
, &n
->res
, base
);
766 #ifdef CONFIG_NET_CLS_IND
767 if (tb
[TCA_U32_INDEV
]) {
769 ret
= tcf_change_indev(net
, tb
[TCA_U32_INDEV
], extack
);
778 static void u32_replace_knode(struct tcf_proto
*tp
, struct tc_u_common
*tp_c
,
779 struct tc_u_knode
*n
)
781 struct tc_u_knode __rcu
**ins
;
782 struct tc_u_knode
*pins
;
783 struct tc_u_hnode
*ht
;
785 if (TC_U32_HTID(n
->handle
) == TC_U32_ROOT
)
786 ht
= rtnl_dereference(tp
->root
);
788 ht
= u32_lookup_ht(tp_c
, TC_U32_HTID(n
->handle
));
790 ins
= &ht
->ht
[TC_U32_HASH(n
->handle
)];
792 /* The node must always exist for it to be replaced if this is not the
793 * case then something went very wrong elsewhere.
795 for (pins
= rtnl_dereference(*ins
); ;
796 ins
= &pins
->next
, pins
= rtnl_dereference(*ins
))
797 if (pins
->handle
== n
->handle
)
800 idr_replace(&ht
->handle_idr
, n
, n
->handle
);
801 RCU_INIT_POINTER(n
->next
, pins
->next
);
802 rcu_assign_pointer(*ins
, n
);
805 static struct tc_u_knode
*u32_init_knode(struct tcf_proto
*tp
,
806 struct tc_u_knode
*n
)
808 struct tc_u_hnode
*ht
= rtnl_dereference(n
->ht_down
);
809 struct tc_u32_sel
*s
= &n
->sel
;
810 struct tc_u_knode
*new;
812 new = kzalloc(sizeof(*n
) + s
->nkeys
*sizeof(struct tc_u32_key
),
818 RCU_INIT_POINTER(new->next
, n
->next
);
819 new->handle
= n
->handle
;
820 RCU_INIT_POINTER(new->ht_up
, n
->ht_up
);
822 #ifdef CONFIG_NET_CLS_IND
823 new->ifindex
= n
->ifindex
;
825 new->fshift
= n
->fshift
;
827 new->flags
= n
->flags
;
828 RCU_INIT_POINTER(new->ht_down
, ht
);
830 /* bump reference count as long as we hold pointer to structure */
834 #ifdef CONFIG_CLS_U32_PERF
835 /* Statistics may be incremented by readers during update
836 * so we must keep them in tact. When the node is later destroyed
837 * a special destroy call must be made to not free the pf memory.
842 #ifdef CONFIG_CLS_U32_MARK
845 /* Similarly success statistics must be moved as pointers */
846 new->pcpu_success
= n
->pcpu_success
;
848 memcpy(&new->sel
, s
, sizeof(*s
) + s
->nkeys
*sizeof(struct tc_u32_key
));
850 if (tcf_exts_init(&new->exts
, TCA_U32_ACT
, TCA_U32_POLICE
)) {
858 static int u32_change(struct net
*net
, struct sk_buff
*in_skb
,
859 struct tcf_proto
*tp
, unsigned long base
, u32 handle
,
860 struct nlattr
**tca
, void **arg
, bool ovr
,
861 struct netlink_ext_ack
*extack
)
863 struct tc_u_common
*tp_c
= tp
->data
;
864 struct tc_u_hnode
*ht
;
865 struct tc_u_knode
*n
;
866 struct tc_u32_sel
*s
;
867 struct nlattr
*opt
= tca
[TCA_OPTIONS
];
868 struct nlattr
*tb
[TCA_U32_MAX
+ 1];
872 #ifdef CONFIG_CLS_U32_PERF
878 NL_SET_ERR_MSG_MOD(extack
, "Filter handle requires options");
885 err
= nla_parse_nested(tb
, TCA_U32_MAX
, opt
, u32_policy
, extack
);
889 if (tb
[TCA_U32_FLAGS
]) {
890 flags
= nla_get_u32(tb
[TCA_U32_FLAGS
]);
891 if (!tc_flags_valid(flags
)) {
892 NL_SET_ERR_MSG_MOD(extack
, "Invalid filter flags");
899 struct tc_u_knode
*new;
901 if (TC_U32_KEY(n
->handle
) == 0) {
902 NL_SET_ERR_MSG_MOD(extack
, "Key node id cannot be zero");
906 if ((n
->flags
^ flags
) &
907 ~(TCA_CLS_FLAGS_IN_HW
| TCA_CLS_FLAGS_NOT_IN_HW
)) {
908 NL_SET_ERR_MSG_MOD(extack
, "Key node flags do not match passed flags");
912 new = u32_init_knode(tp
, n
);
916 err
= u32_set_parms(net
, tp
, base
, new, tb
,
917 tca
[TCA_RATE
], ovr
, extack
);
920 u32_destroy_key(new, false);
924 err
= u32_replace_hw_knode(tp
, new, flags
, extack
);
926 u32_destroy_key(new, false);
930 if (!tc_in_hw(new->flags
))
931 new->flags
|= TCA_CLS_FLAGS_NOT_IN_HW
;
933 u32_replace_knode(tp
, tp_c
, new);
934 tcf_unbind_filter(tp
, &n
->res
);
935 tcf_exts_get_net(&n
->exts
);
936 tcf_queue_work(&n
->rwork
, u32_delete_key_work
);
940 if (tb
[TCA_U32_DIVISOR
]) {
941 unsigned int divisor
= nla_get_u32(tb
[TCA_U32_DIVISOR
]);
943 if (!is_power_of_2(divisor
)) {
944 NL_SET_ERR_MSG_MOD(extack
, "Divisor is not a power of 2");
947 if (divisor
-- > 0x100) {
948 NL_SET_ERR_MSG_MOD(extack
, "Exceeded maximum 256 hash buckets");
951 if (TC_U32_KEY(handle
)) {
952 NL_SET_ERR_MSG_MOD(extack
, "Divisor can only be used on a hash table");
955 ht
= kzalloc(sizeof(*ht
) + divisor
*sizeof(void *), GFP_KERNEL
);
959 handle
= gen_new_htid(tp
->data
, ht
);
965 err
= idr_alloc_u32(&tp_c
->handle_idr
, ht
, &handle
,
973 ht
->divisor
= divisor
;
976 idr_init(&ht
->handle_idr
);
979 err
= u32_replace_hw_hnode(tp
, ht
, flags
, extack
);
981 idr_remove(&tp_c
->handle_idr
, handle
);
986 RCU_INIT_POINTER(ht
->next
, tp_c
->hlist
);
987 rcu_assign_pointer(tp_c
->hlist
, ht
);
993 if (tb
[TCA_U32_HASH
]) {
994 htid
= nla_get_u32(tb
[TCA_U32_HASH
]);
995 if (TC_U32_HTID(htid
) == TC_U32_ROOT
) {
996 ht
= rtnl_dereference(tp
->root
);
999 ht
= u32_lookup_ht(tp
->data
, TC_U32_HTID(htid
));
1001 NL_SET_ERR_MSG_MOD(extack
, "Specified hash table not found");
1006 ht
= rtnl_dereference(tp
->root
);
1010 if (ht
->divisor
< TC_U32_HASH(htid
)) {
1011 NL_SET_ERR_MSG_MOD(extack
, "Specified hash table buckets exceed configured value");
1016 if (TC_U32_HTID(handle
) && TC_U32_HTID(handle
^ htid
)) {
1017 NL_SET_ERR_MSG_MOD(extack
, "Handle specified hash table address mismatch");
1020 handle
= htid
| TC_U32_NODE(handle
);
1021 err
= idr_alloc_u32(&ht
->handle_idr
, NULL
, &handle
, handle
,
1026 handle
= gen_new_kid(ht
, htid
);
1028 if (tb
[TCA_U32_SEL
] == NULL
) {
1029 NL_SET_ERR_MSG_MOD(extack
, "Selector not specified");
1034 s
= nla_data(tb
[TCA_U32_SEL
]);
1035 sel_size
= struct_size(s
, keys
, s
->nkeys
);
1036 if (nla_len(tb
[TCA_U32_SEL
]) < sel_size
) {
1041 n
= kzalloc(offsetof(typeof(*n
), sel
) + sel_size
, GFP_KERNEL
);
1047 #ifdef CONFIG_CLS_U32_PERF
1048 size
= sizeof(struct tc_u32_pcnt
) + s
->nkeys
* sizeof(u64
);
1049 n
->pf
= __alloc_percpu(size
, __alignof__(struct tc_u32_pcnt
));
1056 memcpy(&n
->sel
, s
, sel_size
);
1057 RCU_INIT_POINTER(n
->ht_up
, ht
);
1059 n
->fshift
= s
->hmask
? ffs(ntohl(s
->hmask
)) - 1 : 0;
1062 err
= tcf_exts_init(&n
->exts
, TCA_U32_ACT
, TCA_U32_POLICE
);
1066 #ifdef CONFIG_CLS_U32_MARK
1067 n
->pcpu_success
= alloc_percpu(u32
);
1068 if (!n
->pcpu_success
) {
1073 if (tb
[TCA_U32_MARK
]) {
1074 struct tc_u32_mark
*mark
;
1076 mark
= nla_data(tb
[TCA_U32_MARK
]);
1078 n
->mask
= mark
->mask
;
1082 err
= u32_set_parms(net
, tp
, base
, n
, tb
, tca
[TCA_RATE
], ovr
,
1085 struct tc_u_knode __rcu
**ins
;
1086 struct tc_u_knode
*pins
;
1088 err
= u32_replace_hw_knode(tp
, n
, flags
, extack
);
1092 if (!tc_in_hw(n
->flags
))
1093 n
->flags
|= TCA_CLS_FLAGS_NOT_IN_HW
;
1095 ins
= &ht
->ht
[TC_U32_HASH(handle
)];
1096 for (pins
= rtnl_dereference(*ins
); pins
;
1097 ins
= &pins
->next
, pins
= rtnl_dereference(*ins
))
1098 if (TC_U32_NODE(handle
) < TC_U32_NODE(pins
->handle
))
1101 RCU_INIT_POINTER(n
->next
, pins
);
1102 rcu_assign_pointer(*ins
, n
);
1109 #ifdef CONFIG_CLS_U32_MARK
1110 free_percpu(n
->pcpu_success
);
1114 tcf_exts_destroy(&n
->exts
);
1115 #ifdef CONFIG_CLS_U32_PERF
1121 idr_remove(&ht
->handle_idr
, handle
);
1125 static void u32_walk(struct tcf_proto
*tp
, struct tcf_walker
*arg
)
1127 struct tc_u_common
*tp_c
= tp
->data
;
1128 struct tc_u_hnode
*ht
;
1129 struct tc_u_knode
*n
;
1135 for (ht
= rtnl_dereference(tp_c
->hlist
);
1137 ht
= rtnl_dereference(ht
->next
)) {
1138 if (ht
->prio
!= tp
->prio
)
1140 if (arg
->count
>= arg
->skip
) {
1141 if (arg
->fn(tp
, ht
, arg
) < 0) {
1147 for (h
= 0; h
<= ht
->divisor
; h
++) {
1148 for (n
= rtnl_dereference(ht
->ht
[h
]);
1150 n
= rtnl_dereference(n
->next
)) {
1151 if (arg
->count
< arg
->skip
) {
1155 if (arg
->fn(tp
, n
, arg
) < 0) {
1165 static int u32_reoffload_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
,
1166 bool add
, tc_setup_cb_t
*cb
, void *cb_priv
,
1167 struct netlink_ext_ack
*extack
)
1169 struct tc_cls_u32_offload cls_u32
= {};
1172 tc_cls_common_offload_init(&cls_u32
.common
, tp
, ht
->flags
, extack
);
1173 cls_u32
.command
= add
? TC_CLSU32_NEW_HNODE
: TC_CLSU32_DELETE_HNODE
;
1174 cls_u32
.hnode
.divisor
= ht
->divisor
;
1175 cls_u32
.hnode
.handle
= ht
->handle
;
1176 cls_u32
.hnode
.prio
= ht
->prio
;
1178 err
= cb(TC_SETUP_CLSU32
, &cls_u32
, cb_priv
);
1179 if (err
&& add
&& tc_skip_sw(ht
->flags
))
1185 static int u32_reoffload_knode(struct tcf_proto
*tp
, struct tc_u_knode
*n
,
1186 bool add
, tc_setup_cb_t
*cb
, void *cb_priv
,
1187 struct netlink_ext_ack
*extack
)
1189 struct tc_u_hnode
*ht
= rtnl_dereference(n
->ht_down
);
1190 struct tcf_block
*block
= tp
->chain
->block
;
1191 struct tc_cls_u32_offload cls_u32
= {};
1194 tc_cls_common_offload_init(&cls_u32
.common
, tp
, n
->flags
, extack
);
1195 cls_u32
.command
= add
?
1196 TC_CLSU32_REPLACE_KNODE
: TC_CLSU32_DELETE_KNODE
;
1197 cls_u32
.knode
.handle
= n
->handle
;
1200 cls_u32
.knode
.fshift
= n
->fshift
;
1201 #ifdef CONFIG_CLS_U32_MARK
1202 cls_u32
.knode
.val
= n
->val
;
1203 cls_u32
.knode
.mask
= n
->mask
;
1205 cls_u32
.knode
.val
= 0;
1206 cls_u32
.knode
.mask
= 0;
1208 cls_u32
.knode
.sel
= &n
->sel
;
1209 cls_u32
.knode
.exts
= &n
->exts
;
1211 cls_u32
.knode
.link_handle
= ht
->handle
;
1214 err
= cb(TC_SETUP_CLSU32
, &cls_u32
, cb_priv
);
1216 if (add
&& tc_skip_sw(n
->flags
))
1221 tc_cls_offload_cnt_update(block
, &n
->in_hw_count
, &n
->flags
, add
);
1226 static int u32_reoffload(struct tcf_proto
*tp
, bool add
, tc_setup_cb_t
*cb
,
1227 void *cb_priv
, struct netlink_ext_ack
*extack
)
1229 struct tc_u_common
*tp_c
= tp
->data
;
1230 struct tc_u_hnode
*ht
;
1231 struct tc_u_knode
*n
;
1235 for (ht
= rtnl_dereference(tp_c
->hlist
);
1237 ht
= rtnl_dereference(ht
->next
)) {
1238 if (ht
->prio
!= tp
->prio
)
1241 /* When adding filters to a new dev, try to offload the
1242 * hashtable first. When removing, do the filters before the
1245 if (add
&& !tc_skip_hw(ht
->flags
)) {
1246 err
= u32_reoffload_hnode(tp
, ht
, add
, cb
, cb_priv
,
1252 for (h
= 0; h
<= ht
->divisor
; h
++) {
1253 for (n
= rtnl_dereference(ht
->ht
[h
]);
1255 n
= rtnl_dereference(n
->next
)) {
1256 if (tc_skip_hw(n
->flags
))
1259 err
= u32_reoffload_knode(tp
, n
, add
, cb
,
1266 if (!add
&& !tc_skip_hw(ht
->flags
))
1267 u32_reoffload_hnode(tp
, ht
, add
, cb
, cb_priv
, extack
);
1273 static void u32_bind_class(void *fh
, u32 classid
, unsigned long cl
)
1275 struct tc_u_knode
*n
= fh
;
1277 if (n
&& n
->res
.classid
== classid
)
1281 static int u32_dump(struct net
*net
, struct tcf_proto
*tp
, void *fh
,
1282 struct sk_buff
*skb
, struct tcmsg
*t
)
1284 struct tc_u_knode
*n
= fh
;
1285 struct tc_u_hnode
*ht_up
, *ht_down
;
1286 struct nlattr
*nest
;
1291 t
->tcm_handle
= n
->handle
;
1293 nest
= nla_nest_start(skb
, TCA_OPTIONS
);
1295 goto nla_put_failure
;
1297 if (TC_U32_KEY(n
->handle
) == 0) {
1298 struct tc_u_hnode
*ht
= fh
;
1299 u32 divisor
= ht
->divisor
+ 1;
1301 if (nla_put_u32(skb
, TCA_U32_DIVISOR
, divisor
))
1302 goto nla_put_failure
;
1304 #ifdef CONFIG_CLS_U32_PERF
1305 struct tc_u32_pcnt
*gpf
;
1309 if (nla_put(skb
, TCA_U32_SEL
,
1310 sizeof(n
->sel
) + n
->sel
.nkeys
*sizeof(struct tc_u32_key
),
1312 goto nla_put_failure
;
1314 ht_up
= rtnl_dereference(n
->ht_up
);
1316 u32 htid
= n
->handle
& 0xFFFFF000;
1317 if (nla_put_u32(skb
, TCA_U32_HASH
, htid
))
1318 goto nla_put_failure
;
1320 if (n
->res
.classid
&&
1321 nla_put_u32(skb
, TCA_U32_CLASSID
, n
->res
.classid
))
1322 goto nla_put_failure
;
1324 ht_down
= rtnl_dereference(n
->ht_down
);
1326 nla_put_u32(skb
, TCA_U32_LINK
, ht_down
->handle
))
1327 goto nla_put_failure
;
1329 if (n
->flags
&& nla_put_u32(skb
, TCA_U32_FLAGS
, n
->flags
))
1330 goto nla_put_failure
;
1332 #ifdef CONFIG_CLS_U32_MARK
1333 if ((n
->val
|| n
->mask
)) {
1334 struct tc_u32_mark mark
= {.val
= n
->val
,
1339 for_each_possible_cpu(cpum
) {
1340 __u32 cnt
= *per_cpu_ptr(n
->pcpu_success
, cpum
);
1342 mark
.success
+= cnt
;
1345 if (nla_put(skb
, TCA_U32_MARK
, sizeof(mark
), &mark
))
1346 goto nla_put_failure
;
1350 if (tcf_exts_dump(skb
, &n
->exts
) < 0)
1351 goto nla_put_failure
;
1353 #ifdef CONFIG_NET_CLS_IND
1355 struct net_device
*dev
;
1356 dev
= __dev_get_by_index(net
, n
->ifindex
);
1357 if (dev
&& nla_put_string(skb
, TCA_U32_INDEV
, dev
->name
))
1358 goto nla_put_failure
;
1361 #ifdef CONFIG_CLS_U32_PERF
1362 gpf
= kzalloc(sizeof(struct tc_u32_pcnt
) +
1363 n
->sel
.nkeys
* sizeof(u64
),
1366 goto nla_put_failure
;
1368 for_each_possible_cpu(cpu
) {
1370 struct tc_u32_pcnt
*pf
= per_cpu_ptr(n
->pf
, cpu
);
1372 gpf
->rcnt
+= pf
->rcnt
;
1373 gpf
->rhit
+= pf
->rhit
;
1374 for (i
= 0; i
< n
->sel
.nkeys
; i
++)
1375 gpf
->kcnts
[i
] += pf
->kcnts
[i
];
1378 if (nla_put_64bit(skb
, TCA_U32_PCNT
,
1379 sizeof(struct tc_u32_pcnt
) +
1380 n
->sel
.nkeys
* sizeof(u64
),
1381 gpf
, TCA_U32_PAD
)) {
1383 goto nla_put_failure
;
1389 nla_nest_end(skb
, nest
);
1391 if (TC_U32_KEY(n
->handle
))
1392 if (tcf_exts_dump_stats(skb
, &n
->exts
) < 0)
1393 goto nla_put_failure
;
1397 nla_nest_cancel(skb
, nest
);
1401 static struct tcf_proto_ops cls_u32_ops __read_mostly
= {
1403 .classify
= u32_classify
,
1405 .destroy
= u32_destroy
,
1407 .change
= u32_change
,
1408 .delete = u32_delete
,
1410 .reoffload
= u32_reoffload
,
1412 .bind_class
= u32_bind_class
,
1413 .owner
= THIS_MODULE
,
1416 static int __init
init_u32(void)
1420 pr_info("u32 classifier\n");
1421 #ifdef CONFIG_CLS_U32_PERF
1422 pr_info(" Performance counters on\n");
1424 #ifdef CONFIG_NET_CLS_IND
1425 pr_info(" input device check on\n");
1427 #ifdef CONFIG_NET_CLS_ACT
1428 pr_info(" Actions configured\n");
1430 tc_u_common_hash
= kvmalloc_array(U32_HASH_SIZE
,
1431 sizeof(struct hlist_head
),
1433 if (!tc_u_common_hash
)
1436 for (i
= 0; i
< U32_HASH_SIZE
; i
++)
1437 INIT_HLIST_HEAD(&tc_u_common_hash
[i
]);
1439 ret
= register_tcf_proto_ops(&cls_u32_ops
);
1441 kvfree(tc_u_common_hash
);
1445 static void __exit
exit_u32(void)
1447 unregister_tcf_proto_ops(&cls_u32_ops
);
1448 kvfree(tc_u_common_hash
);
1451 module_init(init_u32
)
1452 module_exit(exit_u32
)
1453 MODULE_LICENSE("GPL");