1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * net/sched/cls_u32.c Ugly (or Universal) 32bit key Packet Classifier.
5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
7 * The filters are packed to hash tables of key nodes
8 * with a set of 32bit key/mask pairs at every node.
9 * Nodes reference next level hash tables etc.
11 * This scheme is the best universal classifier I managed to
12 * invent; it is not super-fast, but it is not slow (provided you
13 * program it correctly), and general enough. And its relative
14 * speed grows as the number of rules becomes larger.
16 * It seems that it represents the best middle point between
17 * speed and manageability both by human and by machine.
19 * It is especially useful for link sharing combined with QoS;
20 * pure RSVP doesn't need such a general approach and can use
21 * much simpler (and faster) schemes, sort of cls_rsvp.c.
23 * JHS: We should remove the CONFIG_NET_CLS_IND from here
24 * eventually when the meta match extension is made available
26 * nfmark match added by Catalin(ux aka Dino) BOIE <catab at umbrella.ro>
29 #include <linux/module.h>
30 #include <linux/slab.h>
31 #include <linux/types.h>
32 #include <linux/kernel.h>
33 #include <linux/string.h>
34 #include <linux/errno.h>
35 #include <linux/percpu.h>
36 #include <linux/rtnetlink.h>
37 #include <linux/skbuff.h>
38 #include <linux/bitmap.h>
39 #include <linux/netdevice.h>
40 #include <linux/hash.h>
41 #include <net/netlink.h>
42 #include <net/act_api.h>
43 #include <net/pkt_cls.h>
44 #include <linux/idr.h>
47 struct tc_u_knode __rcu
*next
;
49 struct tc_u_hnode __rcu
*ht_up
;
51 #ifdef CONFIG_NET_CLS_IND
55 struct tcf_result res
;
56 struct tc_u_hnode __rcu
*ht_down
;
57 #ifdef CONFIG_CLS_U32_PERF
58 struct tc_u32_pcnt __percpu
*pf
;
61 unsigned int in_hw_count
;
62 #ifdef CONFIG_CLS_U32_MARK
65 u32 __percpu
*pcpu_success
;
67 struct rcu_work rwork
;
68 /* The 'sel' field MUST be the last field in structure to allow for
69 * tc_u32_keys allocated at end of structure.
71 struct tc_u32_sel sel
;
75 struct tc_u_hnode __rcu
*next
;
80 struct idr handle_idr
;
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
;
94 struct idr handle_idr
;
95 struct hlist_node hnode
;
99 static inline unsigned int u32_hash_fold(__be32 key
,
100 const struct tc_u32_sel
*sel
,
103 unsigned int h
= ntohl(key
& sel
->hmask
) >> fshift
;
108 static int u32_classify(struct sk_buff
*skb
, const struct tcf_proto
*tp
,
109 struct tcf_result
*res
)
112 struct tc_u_knode
*knode
;
114 } stack
[TC_U32_MAXDEPTH
];
116 struct tc_u_hnode
*ht
= rcu_dereference_bh(tp
->root
);
117 unsigned int off
= skb_network_offset(skb
);
118 struct tc_u_knode
*n
;
122 #ifdef CONFIG_CLS_U32_PERF
128 n
= rcu_dereference_bh(ht
->ht
[sel
]);
132 struct tc_u32_key
*key
= n
->sel
.keys
;
134 #ifdef CONFIG_CLS_U32_PERF
135 __this_cpu_inc(n
->pf
->rcnt
);
139 if (tc_skip_sw(n
->flags
)) {
140 n
= rcu_dereference_bh(n
->next
);
144 #ifdef CONFIG_CLS_U32_MARK
145 if ((skb
->mark
& n
->mask
) != n
->val
) {
146 n
= rcu_dereference_bh(n
->next
);
149 __this_cpu_inc(*n
->pcpu_success
);
153 for (i
= n
->sel
.nkeys
; i
> 0; i
--, key
++) {
154 int toff
= off
+ key
->off
+ (off2
& key
->offmask
);
157 if (skb_headroom(skb
) + toff
> INT_MAX
)
160 data
= skb_header_pointer(skb
, toff
, 4, &hdata
);
163 if ((*data
^ key
->val
) & key
->mask
) {
164 n
= rcu_dereference_bh(n
->next
);
167 #ifdef CONFIG_CLS_U32_PERF
168 __this_cpu_inc(n
->pf
->kcnts
[j
]);
173 ht
= rcu_dereference_bh(n
->ht_down
);
176 if (n
->sel
.flags
& TC_U32_TERMINAL
) {
179 #ifdef CONFIG_NET_CLS_IND
180 if (!tcf_match_indev(skb
, n
->ifindex
)) {
181 n
= rcu_dereference_bh(n
->next
);
185 #ifdef CONFIG_CLS_U32_PERF
186 __this_cpu_inc(n
->pf
->rhit
);
188 r
= tcf_exts_exec(skb
, &n
->exts
, res
);
190 n
= rcu_dereference_bh(n
->next
);
196 n
= rcu_dereference_bh(n
->next
);
201 if (sdepth
>= TC_U32_MAXDEPTH
)
203 stack
[sdepth
].knode
= n
;
204 stack
[sdepth
].off
= off
;
207 ht
= rcu_dereference_bh(n
->ht_down
);
212 data
= skb_header_pointer(skb
, off
+ n
->sel
.hoff
, 4,
216 sel
= ht
->divisor
& u32_hash_fold(*data
, &n
->sel
,
219 if (!(n
->sel
.flags
& (TC_U32_VAROFFSET
| TC_U32_OFFSET
| TC_U32_EAT
)))
222 if (n
->sel
.flags
& (TC_U32_OFFSET
| TC_U32_VAROFFSET
)) {
223 off2
= n
->sel
.off
+ 3;
224 if (n
->sel
.flags
& TC_U32_VAROFFSET
) {
227 data
= skb_header_pointer(skb
,
232 off2
+= ntohs(n
->sel
.offmask
& *data
) >>
237 if (n
->sel
.flags
& TC_U32_EAT
) {
248 n
= stack
[sdepth
].knode
;
249 ht
= rcu_dereference_bh(n
->ht_up
);
250 off
= stack
[sdepth
].off
;
257 net_warn_ratelimited("cls_u32: dead loop\n");
261 static struct tc_u_hnode
*u32_lookup_ht(struct tc_u_common
*tp_c
, u32 handle
)
263 struct tc_u_hnode
*ht
;
265 for (ht
= rtnl_dereference(tp_c
->hlist
);
267 ht
= rtnl_dereference(ht
->next
))
268 if (ht
->handle
== handle
)
274 static struct tc_u_knode
*u32_lookup_key(struct tc_u_hnode
*ht
, u32 handle
)
277 struct tc_u_knode
*n
= NULL
;
279 sel
= TC_U32_HASH(handle
);
280 if (sel
> ht
->divisor
)
283 for (n
= rtnl_dereference(ht
->ht
[sel
]);
285 n
= rtnl_dereference(n
->next
))
286 if (n
->handle
== handle
)
293 static void *u32_get(struct tcf_proto
*tp
, u32 handle
)
295 struct tc_u_hnode
*ht
;
296 struct tc_u_common
*tp_c
= tp
->data
;
298 if (TC_U32_HTID(handle
) == TC_U32_ROOT
)
299 ht
= rtnl_dereference(tp
->root
);
301 ht
= u32_lookup_ht(tp_c
, TC_U32_HTID(handle
));
306 if (TC_U32_KEY(handle
) == 0)
309 return u32_lookup_key(ht
, handle
);
312 /* Protected by rtnl lock */
313 static u32
gen_new_htid(struct tc_u_common
*tp_c
, struct tc_u_hnode
*ptr
)
315 int id
= idr_alloc_cyclic(&tp_c
->handle_idr
, ptr
, 1, 0x7FF, GFP_KERNEL
);
318 return (id
| 0x800U
) << 20;
321 static struct hlist_head
*tc_u_common_hash
;
323 #define U32_HASH_SHIFT 10
324 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
326 static void *tc_u_common_ptr(const struct tcf_proto
*tp
)
328 struct tcf_block
*block
= tp
->chain
->block
;
330 /* The block sharing is currently supported only
331 * for classless qdiscs. In that case we use block
332 * for tc_u_common identification. In case the
333 * block is not shared, block->q is a valid pointer
334 * and we can use that. That works for classful qdiscs.
336 if (tcf_block_shared(block
))
342 static struct hlist_head
*tc_u_hash(void *key
)
344 return tc_u_common_hash
+ hash_ptr(key
, U32_HASH_SHIFT
);
347 static struct tc_u_common
*tc_u_common_find(void *key
)
349 struct tc_u_common
*tc
;
350 hlist_for_each_entry(tc
, tc_u_hash(key
), hnode
) {
357 static int u32_init(struct tcf_proto
*tp
)
359 struct tc_u_hnode
*root_ht
;
360 void *key
= tc_u_common_ptr(tp
);
361 struct tc_u_common
*tp_c
= tc_u_common_find(key
);
363 root_ht
= kzalloc(sizeof(*root_ht
), GFP_KERNEL
);
368 root_ht
->handle
= tp_c
? gen_new_htid(tp_c
, root_ht
) : 0x80000000;
369 root_ht
->prio
= tp
->prio
;
370 root_ht
->is_root
= true;
371 idr_init(&root_ht
->handle_idr
);
374 tp_c
= kzalloc(sizeof(*tp_c
), GFP_KERNEL
);
380 INIT_HLIST_NODE(&tp_c
->hnode
);
381 idr_init(&tp_c
->handle_idr
);
383 hlist_add_head(&tp_c
->hnode
, tc_u_hash(key
));
387 RCU_INIT_POINTER(root_ht
->next
, tp_c
->hlist
);
388 rcu_assign_pointer(tp_c
->hlist
, root_ht
);
391 rcu_assign_pointer(tp
->root
, root_ht
);
396 static int u32_destroy_key(struct tc_u_knode
*n
, bool free_pf
)
398 struct tc_u_hnode
*ht
= rtnl_dereference(n
->ht_down
);
400 tcf_exts_destroy(&n
->exts
);
401 tcf_exts_put_net(&n
->exts
);
402 if (ht
&& --ht
->refcnt
== 0)
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(to_rcu_work(work
),
430 u32_destroy_key(key
, false);
434 /* u32_delete_key_freepf_rcu is the rcu callback variant
435 * that free's the entire structure including the statistics
436 * percpu variables. Only use this if the key is not a copy
437 * returned by u32_init_knode(). See u32_delete_key_rcu()
438 * for the variant that should be used with keys return from
441 static void u32_delete_key_freepf_work(struct work_struct
*work
)
443 struct tc_u_knode
*key
= container_of(to_rcu_work(work
),
447 u32_destroy_key(key
, true);
451 static int u32_delete_key(struct tcf_proto
*tp
, struct tc_u_knode
*key
)
453 struct tc_u_common
*tp_c
= tp
->data
;
454 struct tc_u_knode __rcu
**kp
;
455 struct tc_u_knode
*pkp
;
456 struct tc_u_hnode
*ht
= rtnl_dereference(key
->ht_up
);
459 kp
= &ht
->ht
[TC_U32_HASH(key
->handle
)];
460 for (pkp
= rtnl_dereference(*kp
); pkp
;
461 kp
= &pkp
->next
, pkp
= rtnl_dereference(*kp
)) {
463 RCU_INIT_POINTER(*kp
, key
->next
);
466 tcf_unbind_filter(tp
, &key
->res
);
467 idr_remove(&ht
->handle_idr
, key
->handle
);
468 tcf_exts_get_net(&key
->exts
);
469 tcf_queue_work(&key
->rwork
, u32_delete_key_freepf_work
);
478 static void u32_clear_hw_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*h
,
479 struct netlink_ext_ack
*extack
)
481 struct tcf_block
*block
= tp
->chain
->block
;
482 struct tc_cls_u32_offload cls_u32
= {};
484 tc_cls_common_offload_init(&cls_u32
.common
, tp
, h
->flags
, extack
);
485 cls_u32
.command
= TC_CLSU32_DELETE_HNODE
;
486 cls_u32
.hnode
.divisor
= h
->divisor
;
487 cls_u32
.hnode
.handle
= h
->handle
;
488 cls_u32
.hnode
.prio
= h
->prio
;
490 tc_setup_cb_call(block
, TC_SETUP_CLSU32
, &cls_u32
, false);
493 static int u32_replace_hw_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*h
,
494 u32 flags
, struct netlink_ext_ack
*extack
)
496 struct tcf_block
*block
= tp
->chain
->block
;
497 struct tc_cls_u32_offload cls_u32
= {};
498 bool skip_sw
= tc_skip_sw(flags
);
499 bool offloaded
= false;
502 tc_cls_common_offload_init(&cls_u32
.common
, tp
, flags
, extack
);
503 cls_u32
.command
= TC_CLSU32_NEW_HNODE
;
504 cls_u32
.hnode
.divisor
= h
->divisor
;
505 cls_u32
.hnode
.handle
= h
->handle
;
506 cls_u32
.hnode
.prio
= h
->prio
;
508 err
= tc_setup_cb_call(block
, TC_SETUP_CLSU32
, &cls_u32
, skip_sw
);
510 u32_clear_hw_hnode(tp
, h
, NULL
);
512 } else if (err
> 0) {
516 if (skip_sw
&& !offloaded
)
522 static void u32_remove_hw_knode(struct tcf_proto
*tp
, struct tc_u_knode
*n
,
523 struct netlink_ext_ack
*extack
)
525 struct tcf_block
*block
= tp
->chain
->block
;
526 struct tc_cls_u32_offload cls_u32
= {};
528 tc_cls_common_offload_init(&cls_u32
.common
, tp
, n
->flags
, extack
);
529 cls_u32
.command
= TC_CLSU32_DELETE_KNODE
;
530 cls_u32
.knode
.handle
= n
->handle
;
532 tc_setup_cb_call(block
, TC_SETUP_CLSU32
, &cls_u32
, false);
533 tcf_block_offload_dec(block
, &n
->flags
);
536 static int u32_replace_hw_knode(struct tcf_proto
*tp
, struct tc_u_knode
*n
,
537 u32 flags
, struct netlink_ext_ack
*extack
)
539 struct tc_u_hnode
*ht
= rtnl_dereference(n
->ht_down
);
540 struct tcf_block
*block
= tp
->chain
->block
;
541 struct tc_cls_u32_offload cls_u32
= {};
542 bool skip_sw
= tc_skip_sw(flags
);
545 tc_cls_common_offload_init(&cls_u32
.common
, tp
, flags
, extack
);
546 cls_u32
.command
= TC_CLSU32_REPLACE_KNODE
;
547 cls_u32
.knode
.handle
= n
->handle
;
548 cls_u32
.knode
.fshift
= n
->fshift
;
549 #ifdef CONFIG_CLS_U32_MARK
550 cls_u32
.knode
.val
= n
->val
;
551 cls_u32
.knode
.mask
= n
->mask
;
553 cls_u32
.knode
.val
= 0;
554 cls_u32
.knode
.mask
= 0;
556 cls_u32
.knode
.sel
= &n
->sel
;
557 cls_u32
.knode
.res
= &n
->res
;
558 cls_u32
.knode
.exts
= &n
->exts
;
560 cls_u32
.knode
.link_handle
= ht
->handle
;
562 err
= tc_setup_cb_call(block
, TC_SETUP_CLSU32
, &cls_u32
, skip_sw
);
564 u32_remove_hw_knode(tp
, n
, NULL
);
566 } else if (err
> 0) {
567 n
->in_hw_count
= err
;
568 tcf_block_offload_inc(block
, &n
->flags
);
571 if (skip_sw
&& !(n
->flags
& TCA_CLS_FLAGS_IN_HW
))
577 static void u32_clear_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
,
578 struct netlink_ext_ack
*extack
)
580 struct tc_u_common
*tp_c
= tp
->data
;
581 struct tc_u_knode
*n
;
584 for (h
= 0; h
<= ht
->divisor
; h
++) {
585 while ((n
= rtnl_dereference(ht
->ht
[h
])) != NULL
) {
586 RCU_INIT_POINTER(ht
->ht
[h
],
587 rtnl_dereference(n
->next
));
589 tcf_unbind_filter(tp
, &n
->res
);
590 u32_remove_hw_knode(tp
, n
, extack
);
591 idr_remove(&ht
->handle_idr
, n
->handle
);
592 if (tcf_exts_get_net(&n
->exts
))
593 tcf_queue_work(&n
->rwork
, u32_delete_key_freepf_work
);
595 u32_destroy_key(n
, true);
600 static int u32_destroy_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
,
601 struct netlink_ext_ack
*extack
)
603 struct tc_u_common
*tp_c
= tp
->data
;
604 struct tc_u_hnode __rcu
**hn
;
605 struct tc_u_hnode
*phn
;
607 WARN_ON(--ht
->refcnt
);
609 u32_clear_hnode(tp
, ht
, extack
);
612 for (phn
= rtnl_dereference(*hn
);
614 hn
= &phn
->next
, phn
= rtnl_dereference(*hn
)) {
616 u32_clear_hw_hnode(tp
, ht
, extack
);
617 idr_destroy(&ht
->handle_idr
);
618 idr_remove(&tp_c
->handle_idr
, ht
->handle
);
619 RCU_INIT_POINTER(*hn
, ht
->next
);
628 static void u32_destroy(struct tcf_proto
*tp
, bool rtnl_held
,
629 struct netlink_ext_ack
*extack
)
631 struct tc_u_common
*tp_c
= tp
->data
;
632 struct tc_u_hnode
*root_ht
= rtnl_dereference(tp
->root
);
634 WARN_ON(root_ht
== NULL
);
636 if (root_ht
&& --root_ht
->refcnt
== 1)
637 u32_destroy_hnode(tp
, root_ht
, extack
);
639 if (--tp_c
->refcnt
== 0) {
640 struct tc_u_hnode
*ht
;
642 hlist_del(&tp_c
->hnode
);
644 while ((ht
= rtnl_dereference(tp_c
->hlist
)) != NULL
) {
645 u32_clear_hnode(tp
, ht
, extack
);
646 RCU_INIT_POINTER(tp_c
->hlist
, ht
->next
);
648 /* u32_destroy_key() will later free ht for us, if it's
649 * still referenced by some knode
651 if (--ht
->refcnt
== 0)
655 idr_destroy(&tp_c
->handle_idr
);
662 static int u32_delete(struct tcf_proto
*tp
, void *arg
, bool *last
,
663 bool rtnl_held
, struct netlink_ext_ack
*extack
)
665 struct tc_u_hnode
*ht
= arg
;
666 struct tc_u_common
*tp_c
= tp
->data
;
669 if (TC_U32_KEY(ht
->handle
)) {
670 u32_remove_hw_knode(tp
, (struct tc_u_knode
*)ht
, extack
);
671 ret
= u32_delete_key(tp
, (struct tc_u_knode
*)ht
);
676 NL_SET_ERR_MSG_MOD(extack
, "Not allowed to delete root node");
680 if (ht
->refcnt
== 1) {
681 u32_destroy_hnode(tp
, ht
, extack
);
683 NL_SET_ERR_MSG_MOD(extack
, "Can not delete in-use filter");
688 *last
= tp_c
->refcnt
== 1 && tp_c
->knodes
== 0;
692 static u32
gen_new_kid(struct tc_u_hnode
*ht
, u32 htid
)
694 u32 index
= htid
| 0x800;
695 u32 max
= htid
| 0xFFF;
697 if (idr_alloc_u32(&ht
->handle_idr
, NULL
, &index
, max
, GFP_KERNEL
)) {
699 if (idr_alloc_u32(&ht
->handle_idr
, NULL
, &index
, max
,
707 static const struct nla_policy u32_policy
[TCA_U32_MAX
+ 1] = {
708 [TCA_U32_CLASSID
] = { .type
= NLA_U32
},
709 [TCA_U32_HASH
] = { .type
= NLA_U32
},
710 [TCA_U32_LINK
] = { .type
= NLA_U32
},
711 [TCA_U32_DIVISOR
] = { .type
= NLA_U32
},
712 [TCA_U32_SEL
] = { .len
= sizeof(struct tc_u32_sel
) },
713 [TCA_U32_INDEV
] = { .type
= NLA_STRING
, .len
= IFNAMSIZ
},
714 [TCA_U32_MARK
] = { .len
= sizeof(struct tc_u32_mark
) },
715 [TCA_U32_FLAGS
] = { .type
= NLA_U32
},
718 static int u32_set_parms(struct net
*net
, struct tcf_proto
*tp
,
720 struct tc_u_knode
*n
, struct nlattr
**tb
,
721 struct nlattr
*est
, bool ovr
,
722 struct netlink_ext_ack
*extack
)
726 err
= tcf_exts_validate(net
, tp
, tb
, est
, &n
->exts
, ovr
, true, extack
);
730 if (tb
[TCA_U32_LINK
]) {
731 u32 handle
= nla_get_u32(tb
[TCA_U32_LINK
]);
732 struct tc_u_hnode
*ht_down
= NULL
, *ht_old
;
734 if (TC_U32_KEY(handle
)) {
735 NL_SET_ERR_MSG_MOD(extack
, "u32 Link handle must be a hash table");
740 ht_down
= u32_lookup_ht(tp
->data
, handle
);
743 NL_SET_ERR_MSG_MOD(extack
, "Link hash table not found");
746 if (ht_down
->is_root
) {
747 NL_SET_ERR_MSG_MOD(extack
, "Not linking to root node");
753 ht_old
= rtnl_dereference(n
->ht_down
);
754 rcu_assign_pointer(n
->ht_down
, ht_down
);
759 if (tb
[TCA_U32_CLASSID
]) {
760 n
->res
.classid
= nla_get_u32(tb
[TCA_U32_CLASSID
]);
761 tcf_bind_filter(tp
, &n
->res
, base
);
764 #ifdef CONFIG_NET_CLS_IND
765 if (tb
[TCA_U32_INDEV
]) {
767 ret
= tcf_change_indev(net
, tb
[TCA_U32_INDEV
], extack
);
776 static void u32_replace_knode(struct tcf_proto
*tp
, struct tc_u_common
*tp_c
,
777 struct tc_u_knode
*n
)
779 struct tc_u_knode __rcu
**ins
;
780 struct tc_u_knode
*pins
;
781 struct tc_u_hnode
*ht
;
783 if (TC_U32_HTID(n
->handle
) == TC_U32_ROOT
)
784 ht
= rtnl_dereference(tp
->root
);
786 ht
= u32_lookup_ht(tp_c
, TC_U32_HTID(n
->handle
));
788 ins
= &ht
->ht
[TC_U32_HASH(n
->handle
)];
790 /* The node must always exist for it to be replaced if this is not the
791 * case then something went very wrong elsewhere.
793 for (pins
= rtnl_dereference(*ins
); ;
794 ins
= &pins
->next
, pins
= rtnl_dereference(*ins
))
795 if (pins
->handle
== n
->handle
)
798 idr_replace(&ht
->handle_idr
, n
, n
->handle
);
799 RCU_INIT_POINTER(n
->next
, pins
->next
);
800 rcu_assign_pointer(*ins
, n
);
803 static struct tc_u_knode
*u32_init_knode(struct net
*net
, struct tcf_proto
*tp
,
804 struct tc_u_knode
*n
)
806 struct tc_u_hnode
*ht
= rtnl_dereference(n
->ht_down
);
807 struct tc_u32_sel
*s
= &n
->sel
;
808 struct tc_u_knode
*new;
810 new = kzalloc(sizeof(*n
) + s
->nkeys
*sizeof(struct tc_u32_key
),
816 RCU_INIT_POINTER(new->next
, n
->next
);
817 new->handle
= n
->handle
;
818 RCU_INIT_POINTER(new->ht_up
, n
->ht_up
);
820 #ifdef CONFIG_NET_CLS_IND
821 new->ifindex
= n
->ifindex
;
823 new->fshift
= n
->fshift
;
825 new->flags
= n
->flags
;
826 RCU_INIT_POINTER(new->ht_down
, ht
);
828 /* bump reference count as long as we hold pointer to structure */
832 #ifdef CONFIG_CLS_U32_PERF
833 /* Statistics may be incremented by readers during update
834 * so we must keep them in tact. When the node is later destroyed
835 * a special destroy call must be made to not free the pf memory.
840 #ifdef CONFIG_CLS_U32_MARK
843 /* Similarly success statistics must be moved as pointers */
844 new->pcpu_success
= n
->pcpu_success
;
846 memcpy(&new->sel
, s
, struct_size(s
, keys
, s
->nkeys
));
848 if (tcf_exts_init(&new->exts
, net
, TCA_U32_ACT
, TCA_U32_POLICE
)) {
856 static int u32_change(struct net
*net
, struct sk_buff
*in_skb
,
857 struct tcf_proto
*tp
, unsigned long base
, u32 handle
,
858 struct nlattr
**tca
, void **arg
, bool ovr
, bool rtnl_held
,
859 struct netlink_ext_ack
*extack
)
861 struct tc_u_common
*tp_c
= tp
->data
;
862 struct tc_u_hnode
*ht
;
863 struct tc_u_knode
*n
;
864 struct tc_u32_sel
*s
;
865 struct nlattr
*opt
= tca
[TCA_OPTIONS
];
866 struct nlattr
*tb
[TCA_U32_MAX
+ 1];
870 #ifdef CONFIG_CLS_U32_PERF
876 NL_SET_ERR_MSG_MOD(extack
, "Filter handle requires options");
883 err
= nla_parse_nested_deprecated(tb
, TCA_U32_MAX
, opt
, u32_policy
,
888 if (tb
[TCA_U32_FLAGS
]) {
889 flags
= nla_get_u32(tb
[TCA_U32_FLAGS
]);
890 if (!tc_flags_valid(flags
)) {
891 NL_SET_ERR_MSG_MOD(extack
, "Invalid filter flags");
898 struct tc_u_knode
*new;
900 if (TC_U32_KEY(n
->handle
) == 0) {
901 NL_SET_ERR_MSG_MOD(extack
, "Key node id cannot be zero");
905 if ((n
->flags
^ flags
) &
906 ~(TCA_CLS_FLAGS_IN_HW
| TCA_CLS_FLAGS_NOT_IN_HW
)) {
907 NL_SET_ERR_MSG_MOD(extack
, "Key node flags do not match passed flags");
911 new = u32_init_knode(net
, tp
, n
);
915 err
= u32_set_parms(net
, tp
, base
, new, tb
,
916 tca
[TCA_RATE
], ovr
, extack
);
919 u32_destroy_key(new, false);
923 err
= u32_replace_hw_knode(tp
, new, flags
, extack
);
925 u32_destroy_key(new, false);
929 if (!tc_in_hw(new->flags
))
930 new->flags
|= TCA_CLS_FLAGS_NOT_IN_HW
;
932 u32_replace_knode(tp
, tp_c
, new);
933 tcf_unbind_filter(tp
, &n
->res
);
934 tcf_exts_get_net(&n
->exts
);
935 tcf_queue_work(&n
->rwork
, u32_delete_key_work
);
939 if (tb
[TCA_U32_DIVISOR
]) {
940 unsigned int divisor
= nla_get_u32(tb
[TCA_U32_DIVISOR
]);
942 if (!is_power_of_2(divisor
)) {
943 NL_SET_ERR_MSG_MOD(extack
, "Divisor is not a power of 2");
946 if (divisor
-- > 0x100) {
947 NL_SET_ERR_MSG_MOD(extack
, "Exceeded maximum 256 hash buckets");
950 if (TC_U32_KEY(handle
)) {
951 NL_SET_ERR_MSG_MOD(extack
, "Divisor can only be used on a hash table");
954 ht
= kzalloc(sizeof(*ht
) + divisor
*sizeof(void *), GFP_KERNEL
);
958 handle
= gen_new_htid(tp
->data
, ht
);
964 err
= idr_alloc_u32(&tp_c
->handle_idr
, ht
, &handle
,
972 ht
->divisor
= divisor
;
975 idr_init(&ht
->handle_idr
);
978 err
= u32_replace_hw_hnode(tp
, ht
, flags
, extack
);
980 idr_remove(&tp_c
->handle_idr
, handle
);
985 RCU_INIT_POINTER(ht
->next
, tp_c
->hlist
);
986 rcu_assign_pointer(tp_c
->hlist
, ht
);
992 if (tb
[TCA_U32_HASH
]) {
993 htid
= nla_get_u32(tb
[TCA_U32_HASH
]);
994 if (TC_U32_HTID(htid
) == TC_U32_ROOT
) {
995 ht
= rtnl_dereference(tp
->root
);
998 ht
= u32_lookup_ht(tp
->data
, TC_U32_HTID(htid
));
1000 NL_SET_ERR_MSG_MOD(extack
, "Specified hash table not found");
1005 ht
= rtnl_dereference(tp
->root
);
1009 if (ht
->divisor
< TC_U32_HASH(htid
)) {
1010 NL_SET_ERR_MSG_MOD(extack
, "Specified hash table buckets exceed configured value");
1015 if (TC_U32_HTID(handle
) && TC_U32_HTID(handle
^ htid
)) {
1016 NL_SET_ERR_MSG_MOD(extack
, "Handle specified hash table address mismatch");
1019 handle
= htid
| TC_U32_NODE(handle
);
1020 err
= idr_alloc_u32(&ht
->handle_idr
, NULL
, &handle
, handle
,
1025 handle
= gen_new_kid(ht
, htid
);
1027 if (tb
[TCA_U32_SEL
] == NULL
) {
1028 NL_SET_ERR_MSG_MOD(extack
, "Selector not specified");
1033 s
= nla_data(tb
[TCA_U32_SEL
]);
1034 sel_size
= struct_size(s
, keys
, s
->nkeys
);
1035 if (nla_len(tb
[TCA_U32_SEL
]) < sel_size
) {
1040 n
= kzalloc(offsetof(typeof(*n
), sel
) + sel_size
, GFP_KERNEL
);
1046 #ifdef CONFIG_CLS_U32_PERF
1047 size
= sizeof(struct tc_u32_pcnt
) + s
->nkeys
* sizeof(u64
);
1048 n
->pf
= __alloc_percpu(size
, __alignof__(struct tc_u32_pcnt
));
1055 memcpy(&n
->sel
, s
, sel_size
);
1056 RCU_INIT_POINTER(n
->ht_up
, ht
);
1058 n
->fshift
= s
->hmask
? ffs(ntohl(s
->hmask
)) - 1 : 0;
1061 err
= tcf_exts_init(&n
->exts
, net
, TCA_U32_ACT
, TCA_U32_POLICE
);
1065 #ifdef CONFIG_CLS_U32_MARK
1066 n
->pcpu_success
= alloc_percpu(u32
);
1067 if (!n
->pcpu_success
) {
1072 if (tb
[TCA_U32_MARK
]) {
1073 struct tc_u32_mark
*mark
;
1075 mark
= nla_data(tb
[TCA_U32_MARK
]);
1077 n
->mask
= mark
->mask
;
1081 err
= u32_set_parms(net
, tp
, base
, n
, tb
, tca
[TCA_RATE
], ovr
,
1084 struct tc_u_knode __rcu
**ins
;
1085 struct tc_u_knode
*pins
;
1087 err
= u32_replace_hw_knode(tp
, n
, flags
, extack
);
1091 if (!tc_in_hw(n
->flags
))
1092 n
->flags
|= TCA_CLS_FLAGS_NOT_IN_HW
;
1094 ins
= &ht
->ht
[TC_U32_HASH(handle
)];
1095 for (pins
= rtnl_dereference(*ins
); pins
;
1096 ins
= &pins
->next
, pins
= rtnl_dereference(*ins
))
1097 if (TC_U32_NODE(handle
) < TC_U32_NODE(pins
->handle
))
1100 RCU_INIT_POINTER(n
->next
, pins
);
1101 rcu_assign_pointer(*ins
, n
);
1108 #ifdef CONFIG_CLS_U32_MARK
1109 free_percpu(n
->pcpu_success
);
1113 tcf_exts_destroy(&n
->exts
);
1114 #ifdef CONFIG_CLS_U32_PERF
1120 idr_remove(&ht
->handle_idr
, handle
);
1124 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
.res
= &n
->res
;
1210 cls_u32
.knode
.exts
= &n
->exts
;
1212 cls_u32
.knode
.link_handle
= ht
->handle
;
1215 err
= cb(TC_SETUP_CLSU32
, &cls_u32
, cb_priv
);
1217 if (add
&& tc_skip_sw(n
->flags
))
1222 tc_cls_offload_cnt_update(block
, &n
->in_hw_count
, &n
->flags
, add
);
1227 static int u32_reoffload(struct tcf_proto
*tp
, bool add
, tc_setup_cb_t
*cb
,
1228 void *cb_priv
, struct netlink_ext_ack
*extack
)
1230 struct tc_u_common
*tp_c
= tp
->data
;
1231 struct tc_u_hnode
*ht
;
1232 struct tc_u_knode
*n
;
1236 for (ht
= rtnl_dereference(tp_c
->hlist
);
1238 ht
= rtnl_dereference(ht
->next
)) {
1239 if (ht
->prio
!= tp
->prio
)
1242 /* When adding filters to a new dev, try to offload the
1243 * hashtable first. When removing, do the filters before the
1246 if (add
&& !tc_skip_hw(ht
->flags
)) {
1247 err
= u32_reoffload_hnode(tp
, ht
, add
, cb
, cb_priv
,
1253 for (h
= 0; h
<= ht
->divisor
; h
++) {
1254 for (n
= rtnl_dereference(ht
->ht
[h
]);
1256 n
= rtnl_dereference(n
->next
)) {
1257 if (tc_skip_hw(n
->flags
))
1260 err
= u32_reoffload_knode(tp
, n
, add
, cb
,
1267 if (!add
&& !tc_skip_hw(ht
->flags
))
1268 u32_reoffload_hnode(tp
, ht
, add
, cb
, cb_priv
, extack
);
1274 static void u32_bind_class(void *fh
, u32 classid
, unsigned long cl
)
1276 struct tc_u_knode
*n
= fh
;
1278 if (n
&& n
->res
.classid
== classid
)
1282 static int u32_dump(struct net
*net
, struct tcf_proto
*tp
, void *fh
,
1283 struct sk_buff
*skb
, struct tcmsg
*t
, bool rtnl_held
)
1285 struct tc_u_knode
*n
= fh
;
1286 struct tc_u_hnode
*ht_up
, *ht_down
;
1287 struct nlattr
*nest
;
1292 t
->tcm_handle
= n
->handle
;
1294 nest
= nla_nest_start_noflag(skb
, TCA_OPTIONS
);
1296 goto nla_put_failure
;
1298 if (TC_U32_KEY(n
->handle
) == 0) {
1299 struct tc_u_hnode
*ht
= fh
;
1300 u32 divisor
= ht
->divisor
+ 1;
1302 if (nla_put_u32(skb
, TCA_U32_DIVISOR
, divisor
))
1303 goto nla_put_failure
;
1305 #ifdef CONFIG_CLS_U32_PERF
1306 struct tc_u32_pcnt
*gpf
;
1310 if (nla_put(skb
, TCA_U32_SEL
,
1311 sizeof(n
->sel
) + n
->sel
.nkeys
*sizeof(struct tc_u32_key
),
1313 goto nla_put_failure
;
1315 ht_up
= rtnl_dereference(n
->ht_up
);
1317 u32 htid
= n
->handle
& 0xFFFFF000;
1318 if (nla_put_u32(skb
, TCA_U32_HASH
, htid
))
1319 goto nla_put_failure
;
1321 if (n
->res
.classid
&&
1322 nla_put_u32(skb
, TCA_U32_CLASSID
, n
->res
.classid
))
1323 goto nla_put_failure
;
1325 ht_down
= rtnl_dereference(n
->ht_down
);
1327 nla_put_u32(skb
, TCA_U32_LINK
, ht_down
->handle
))
1328 goto nla_put_failure
;
1330 if (n
->flags
&& nla_put_u32(skb
, TCA_U32_FLAGS
, n
->flags
))
1331 goto nla_put_failure
;
1333 #ifdef CONFIG_CLS_U32_MARK
1334 if ((n
->val
|| n
->mask
)) {
1335 struct tc_u32_mark mark
= {.val
= n
->val
,
1340 for_each_possible_cpu(cpum
) {
1341 __u32 cnt
= *per_cpu_ptr(n
->pcpu_success
, cpum
);
1343 mark
.success
+= cnt
;
1346 if (nla_put(skb
, TCA_U32_MARK
, sizeof(mark
), &mark
))
1347 goto nla_put_failure
;
1351 if (tcf_exts_dump(skb
, &n
->exts
) < 0)
1352 goto nla_put_failure
;
1354 #ifdef CONFIG_NET_CLS_IND
1356 struct net_device
*dev
;
1357 dev
= __dev_get_by_index(net
, n
->ifindex
);
1358 if (dev
&& nla_put_string(skb
, TCA_U32_INDEV
, dev
->name
))
1359 goto nla_put_failure
;
1362 #ifdef CONFIG_CLS_U32_PERF
1363 gpf
= kzalloc(sizeof(struct tc_u32_pcnt
) +
1364 n
->sel
.nkeys
* sizeof(u64
),
1367 goto nla_put_failure
;
1369 for_each_possible_cpu(cpu
) {
1371 struct tc_u32_pcnt
*pf
= per_cpu_ptr(n
->pf
, cpu
);
1373 gpf
->rcnt
+= pf
->rcnt
;
1374 gpf
->rhit
+= pf
->rhit
;
1375 for (i
= 0; i
< n
->sel
.nkeys
; i
++)
1376 gpf
->kcnts
[i
] += pf
->kcnts
[i
];
1379 if (nla_put_64bit(skb
, TCA_U32_PCNT
,
1380 sizeof(struct tc_u32_pcnt
) +
1381 n
->sel
.nkeys
* sizeof(u64
),
1382 gpf
, TCA_U32_PAD
)) {
1384 goto nla_put_failure
;
1390 nla_nest_end(skb
, nest
);
1392 if (TC_U32_KEY(n
->handle
))
1393 if (tcf_exts_dump_stats(skb
, &n
->exts
) < 0)
1394 goto nla_put_failure
;
1398 nla_nest_cancel(skb
, nest
);
1402 static struct tcf_proto_ops cls_u32_ops __read_mostly
= {
1404 .classify
= u32_classify
,
1406 .destroy
= u32_destroy
,
1408 .change
= u32_change
,
1409 .delete = u32_delete
,
1411 .reoffload
= u32_reoffload
,
1413 .bind_class
= u32_bind_class
,
1414 .owner
= THIS_MODULE
,
1417 static int __init
init_u32(void)
1421 pr_info("u32 classifier\n");
1422 #ifdef CONFIG_CLS_U32_PERF
1423 pr_info(" Performance counters on\n");
1425 #ifdef CONFIG_NET_CLS_IND
1426 pr_info(" input device check on\n");
1428 #ifdef CONFIG_NET_CLS_ACT
1429 pr_info(" Actions configured\n");
1431 tc_u_common_hash
= kvmalloc_array(U32_HASH_SIZE
,
1432 sizeof(struct hlist_head
),
1434 if (!tc_u_common_hash
)
1437 for (i
= 0; i
< U32_HASH_SIZE
; i
++)
1438 INIT_HLIST_HEAD(&tc_u_common_hash
[i
]);
1440 ret
= register_tcf_proto_ops(&cls_u32_ops
);
1442 kvfree(tc_u_common_hash
);
1446 static void __exit
exit_u32(void)
1448 unregister_tcf_proto_ops(&cls_u32_ops
);
1449 kvfree(tc_u_common_hash
);
1452 module_init(init_u32
)
1453 module_exit(exit_u32
)
1454 MODULE_LICENSE("GPL");