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 * nfmark match added by Catalin(ux aka Dino) BOIE <catab at umbrella.ro>
26 #include <linux/module.h>
27 #include <linux/slab.h>
28 #include <linux/types.h>
29 #include <linux/kernel.h>
30 #include <linux/string.h>
31 #include <linux/errno.h>
32 #include <linux/percpu.h>
33 #include <linux/rtnetlink.h>
34 #include <linux/skbuff.h>
35 #include <linux/bitmap.h>
36 #include <linux/netdevice.h>
37 #include <linux/hash.h>
38 #include <net/netlink.h>
39 #include <net/act_api.h>
40 #include <net/pkt_cls.h>
41 #include <linux/idr.h>
44 struct tc_u_knode __rcu
*next
;
46 struct tc_u_hnode __rcu
*ht_up
;
50 struct tcf_result res
;
51 struct tc_u_hnode __rcu
*ht_down
;
52 #ifdef CONFIG_CLS_U32_PERF
53 struct tc_u32_pcnt __percpu
*pf
;
56 unsigned int in_hw_count
;
57 #ifdef CONFIG_CLS_U32_MARK
60 u32 __percpu
*pcpu_success
;
62 struct rcu_work rwork
;
63 /* The 'sel' field MUST be the last field in structure to allow for
64 * tc_u32_keys allocated at end of structure.
66 struct tc_u32_sel sel
;
70 struct tc_u_hnode __rcu
*next
;
75 struct idr handle_idr
;
79 /* The 'ht' field MUST be the last field in structure to allow for
80 * more entries allocated at end of structure.
82 struct tc_u_knode __rcu
*ht
[];
86 struct tc_u_hnode __rcu
*hlist
;
89 struct idr handle_idr
;
90 struct hlist_node hnode
;
94 static inline unsigned int u32_hash_fold(__be32 key
,
95 const struct tc_u32_sel
*sel
,
98 unsigned int h
= ntohl(key
& sel
->hmask
) >> fshift
;
103 static int u32_classify(struct sk_buff
*skb
, const struct tcf_proto
*tp
,
104 struct tcf_result
*res
)
107 struct tc_u_knode
*knode
;
109 } stack
[TC_U32_MAXDEPTH
];
111 struct tc_u_hnode
*ht
= rcu_dereference_bh(tp
->root
);
112 unsigned int off
= skb_network_offset(skb
);
113 struct tc_u_knode
*n
;
117 #ifdef CONFIG_CLS_U32_PERF
123 n
= rcu_dereference_bh(ht
->ht
[sel
]);
127 struct tc_u32_key
*key
= n
->sel
.keys
;
129 #ifdef CONFIG_CLS_U32_PERF
130 __this_cpu_inc(n
->pf
->rcnt
);
134 if (tc_skip_sw(n
->flags
)) {
135 n
= rcu_dereference_bh(n
->next
);
139 #ifdef CONFIG_CLS_U32_MARK
140 if ((skb
->mark
& n
->mask
) != n
->val
) {
141 n
= rcu_dereference_bh(n
->next
);
144 __this_cpu_inc(*n
->pcpu_success
);
148 for (i
= n
->sel
.nkeys
; i
> 0; i
--, key
++) {
149 int toff
= off
+ key
->off
+ (off2
& key
->offmask
);
152 if (skb_headroom(skb
) + toff
> INT_MAX
)
155 data
= skb_header_pointer(skb
, toff
, 4, &hdata
);
158 if ((*data
^ key
->val
) & key
->mask
) {
159 n
= rcu_dereference_bh(n
->next
);
162 #ifdef CONFIG_CLS_U32_PERF
163 __this_cpu_inc(n
->pf
->kcnts
[j
]);
168 ht
= rcu_dereference_bh(n
->ht_down
);
171 if (n
->sel
.flags
& TC_U32_TERMINAL
) {
174 if (!tcf_match_indev(skb
, n
->ifindex
)) {
175 n
= rcu_dereference_bh(n
->next
);
178 #ifdef CONFIG_CLS_U32_PERF
179 __this_cpu_inc(n
->pf
->rhit
);
181 r
= tcf_exts_exec(skb
, &n
->exts
, res
);
183 n
= rcu_dereference_bh(n
->next
);
189 n
= rcu_dereference_bh(n
->next
);
194 if (sdepth
>= TC_U32_MAXDEPTH
)
196 stack
[sdepth
].knode
= n
;
197 stack
[sdepth
].off
= off
;
200 ht
= rcu_dereference_bh(n
->ht_down
);
205 data
= skb_header_pointer(skb
, off
+ n
->sel
.hoff
, 4,
209 sel
= ht
->divisor
& u32_hash_fold(*data
, &n
->sel
,
212 if (!(n
->sel
.flags
& (TC_U32_VAROFFSET
| TC_U32_OFFSET
| TC_U32_EAT
)))
215 if (n
->sel
.flags
& (TC_U32_OFFSET
| TC_U32_VAROFFSET
)) {
216 off2
= n
->sel
.off
+ 3;
217 if (n
->sel
.flags
& TC_U32_VAROFFSET
) {
220 data
= skb_header_pointer(skb
,
225 off2
+= ntohs(n
->sel
.offmask
& *data
) >>
230 if (n
->sel
.flags
& TC_U32_EAT
) {
241 n
= stack
[sdepth
].knode
;
242 ht
= rcu_dereference_bh(n
->ht_up
);
243 off
= stack
[sdepth
].off
;
250 net_warn_ratelimited("cls_u32: dead loop\n");
254 static struct tc_u_hnode
*u32_lookup_ht(struct tc_u_common
*tp_c
, u32 handle
)
256 struct tc_u_hnode
*ht
;
258 for (ht
= rtnl_dereference(tp_c
->hlist
);
260 ht
= rtnl_dereference(ht
->next
))
261 if (ht
->handle
== handle
)
267 static struct tc_u_knode
*u32_lookup_key(struct tc_u_hnode
*ht
, u32 handle
)
270 struct tc_u_knode
*n
= NULL
;
272 sel
= TC_U32_HASH(handle
);
273 if (sel
> ht
->divisor
)
276 for (n
= rtnl_dereference(ht
->ht
[sel
]);
278 n
= rtnl_dereference(n
->next
))
279 if (n
->handle
== handle
)
286 static void *u32_get(struct tcf_proto
*tp
, u32 handle
)
288 struct tc_u_hnode
*ht
;
289 struct tc_u_common
*tp_c
= tp
->data
;
291 if (TC_U32_HTID(handle
) == TC_U32_ROOT
)
292 ht
= rtnl_dereference(tp
->root
);
294 ht
= u32_lookup_ht(tp_c
, TC_U32_HTID(handle
));
299 if (TC_U32_KEY(handle
) == 0)
302 return u32_lookup_key(ht
, handle
);
305 /* Protected by rtnl lock */
306 static u32
gen_new_htid(struct tc_u_common
*tp_c
, struct tc_u_hnode
*ptr
)
308 int id
= idr_alloc_cyclic(&tp_c
->handle_idr
, ptr
, 1, 0x7FF, GFP_KERNEL
);
311 return (id
| 0x800U
) << 20;
314 static struct hlist_head
*tc_u_common_hash
;
316 #define U32_HASH_SHIFT 10
317 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
319 static void *tc_u_common_ptr(const struct tcf_proto
*tp
)
321 struct tcf_block
*block
= tp
->chain
->block
;
323 /* The block sharing is currently supported only
324 * for classless qdiscs. In that case we use block
325 * for tc_u_common identification. In case the
326 * block is not shared, block->q is a valid pointer
327 * and we can use that. That works for classful qdiscs.
329 if (tcf_block_shared(block
))
335 static struct hlist_head
*tc_u_hash(void *key
)
337 return tc_u_common_hash
+ hash_ptr(key
, U32_HASH_SHIFT
);
340 static struct tc_u_common
*tc_u_common_find(void *key
)
342 struct tc_u_common
*tc
;
343 hlist_for_each_entry(tc
, tc_u_hash(key
), hnode
) {
350 static int u32_init(struct tcf_proto
*tp
)
352 struct tc_u_hnode
*root_ht
;
353 void *key
= tc_u_common_ptr(tp
);
354 struct tc_u_common
*tp_c
= tc_u_common_find(key
);
356 root_ht
= kzalloc(struct_size(root_ht
, ht
, 1), GFP_KERNEL
);
361 root_ht
->handle
= tp_c
? gen_new_htid(tp_c
, root_ht
) : 0x80000000;
362 root_ht
->prio
= tp
->prio
;
363 root_ht
->is_root
= true;
364 idr_init(&root_ht
->handle_idr
);
367 tp_c
= kzalloc(struct_size(tp_c
, hlist
->ht
, 1), GFP_KERNEL
);
373 INIT_HLIST_NODE(&tp_c
->hnode
);
374 idr_init(&tp_c
->handle_idr
);
376 hlist_add_head(&tp_c
->hnode
, tc_u_hash(key
));
380 RCU_INIT_POINTER(root_ht
->next
, tp_c
->hlist
);
381 rcu_assign_pointer(tp_c
->hlist
, root_ht
);
384 rcu_assign_pointer(tp
->root
, root_ht
);
389 static void __u32_destroy_key(struct tc_u_knode
*n
)
391 struct tc_u_hnode
*ht
= rtnl_dereference(n
->ht_down
);
393 tcf_exts_destroy(&n
->exts
);
394 if (ht
&& --ht
->refcnt
== 0)
399 static void u32_destroy_key(struct tc_u_knode
*n
, bool free_pf
)
401 tcf_exts_put_net(&n
->exts
);
402 #ifdef CONFIG_CLS_U32_PERF
406 #ifdef CONFIG_CLS_U32_MARK
408 free_percpu(n
->pcpu_success
);
410 __u32_destroy_key(n
);
413 /* u32_delete_key_rcu should be called when free'ing a copied
414 * version of a tc_u_knode obtained from u32_init_knode(). When
415 * copies are obtained from u32_init_knode() the statistics are
416 * shared between the old and new copies to allow readers to
417 * continue to update the statistics during the copy. To support
418 * this the u32_delete_key_rcu variant does not free the percpu
421 static void u32_delete_key_work(struct work_struct
*work
)
423 struct tc_u_knode
*key
= container_of(to_rcu_work(work
),
427 u32_destroy_key(key
, false);
431 /* u32_delete_key_freepf_rcu is the rcu callback variant
432 * that free's the entire structure including the statistics
433 * percpu variables. Only use this if the key is not a copy
434 * returned by u32_init_knode(). See u32_delete_key_rcu()
435 * for the variant that should be used with keys return from
438 static void u32_delete_key_freepf_work(struct work_struct
*work
)
440 struct tc_u_knode
*key
= container_of(to_rcu_work(work
),
444 u32_destroy_key(key
, true);
448 static int u32_delete_key(struct tcf_proto
*tp
, struct tc_u_knode
*key
)
450 struct tc_u_common
*tp_c
= tp
->data
;
451 struct tc_u_knode __rcu
**kp
;
452 struct tc_u_knode
*pkp
;
453 struct tc_u_hnode
*ht
= rtnl_dereference(key
->ht_up
);
456 kp
= &ht
->ht
[TC_U32_HASH(key
->handle
)];
457 for (pkp
= rtnl_dereference(*kp
); pkp
;
458 kp
= &pkp
->next
, pkp
= rtnl_dereference(*kp
)) {
460 RCU_INIT_POINTER(*kp
, key
->next
);
463 tcf_unbind_filter(tp
, &key
->res
);
464 idr_remove(&ht
->handle_idr
, key
->handle
);
465 tcf_exts_get_net(&key
->exts
);
466 tcf_queue_work(&key
->rwork
, u32_delete_key_freepf_work
);
475 static void u32_clear_hw_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*h
,
476 struct netlink_ext_ack
*extack
)
478 struct tcf_block
*block
= tp
->chain
->block
;
479 struct tc_cls_u32_offload cls_u32
= {};
481 tc_cls_common_offload_init(&cls_u32
.common
, tp
, h
->flags
, extack
);
482 cls_u32
.command
= TC_CLSU32_DELETE_HNODE
;
483 cls_u32
.hnode
.divisor
= h
->divisor
;
484 cls_u32
.hnode
.handle
= h
->handle
;
485 cls_u32
.hnode
.prio
= h
->prio
;
487 tc_setup_cb_call(block
, TC_SETUP_CLSU32
, &cls_u32
, false, true);
490 static int u32_replace_hw_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*h
,
491 u32 flags
, struct netlink_ext_ack
*extack
)
493 struct tcf_block
*block
= tp
->chain
->block
;
494 struct tc_cls_u32_offload cls_u32
= {};
495 bool skip_sw
= tc_skip_sw(flags
);
496 bool offloaded
= false;
499 tc_cls_common_offload_init(&cls_u32
.common
, tp
, flags
, extack
);
500 cls_u32
.command
= TC_CLSU32_NEW_HNODE
;
501 cls_u32
.hnode
.divisor
= h
->divisor
;
502 cls_u32
.hnode
.handle
= h
->handle
;
503 cls_u32
.hnode
.prio
= h
->prio
;
505 err
= tc_setup_cb_call(block
, TC_SETUP_CLSU32
, &cls_u32
, skip_sw
, true);
507 u32_clear_hw_hnode(tp
, h
, NULL
);
509 } else if (err
> 0) {
513 if (skip_sw
&& !offloaded
)
519 static void u32_remove_hw_knode(struct tcf_proto
*tp
, struct tc_u_knode
*n
,
520 struct netlink_ext_ack
*extack
)
522 struct tcf_block
*block
= tp
->chain
->block
;
523 struct tc_cls_u32_offload cls_u32
= {};
525 tc_cls_common_offload_init(&cls_u32
.common
, tp
, n
->flags
, extack
);
526 cls_u32
.command
= TC_CLSU32_DELETE_KNODE
;
527 cls_u32
.knode
.handle
= n
->handle
;
529 tc_setup_cb_destroy(block
, tp
, TC_SETUP_CLSU32
, &cls_u32
, false,
530 &n
->flags
, &n
->in_hw_count
, true);
533 static int u32_replace_hw_knode(struct tcf_proto
*tp
, struct tc_u_knode
*n
,
534 u32 flags
, struct netlink_ext_ack
*extack
)
536 struct tc_u_hnode
*ht
= rtnl_dereference(n
->ht_down
);
537 struct tcf_block
*block
= tp
->chain
->block
;
538 struct tc_cls_u32_offload cls_u32
= {};
539 bool skip_sw
= tc_skip_sw(flags
);
542 tc_cls_common_offload_init(&cls_u32
.common
, tp
, flags
, extack
);
543 cls_u32
.command
= TC_CLSU32_REPLACE_KNODE
;
544 cls_u32
.knode
.handle
= n
->handle
;
545 cls_u32
.knode
.fshift
= n
->fshift
;
546 #ifdef CONFIG_CLS_U32_MARK
547 cls_u32
.knode
.val
= n
->val
;
548 cls_u32
.knode
.mask
= n
->mask
;
550 cls_u32
.knode
.val
= 0;
551 cls_u32
.knode
.mask
= 0;
553 cls_u32
.knode
.sel
= &n
->sel
;
554 cls_u32
.knode
.res
= &n
->res
;
555 cls_u32
.knode
.exts
= &n
->exts
;
557 cls_u32
.knode
.link_handle
= ht
->handle
;
559 err
= tc_setup_cb_add(block
, tp
, TC_SETUP_CLSU32
, &cls_u32
, skip_sw
,
560 &n
->flags
, &n
->in_hw_count
, true);
562 u32_remove_hw_knode(tp
, n
, NULL
);
566 if (skip_sw
&& !(n
->flags
& TCA_CLS_FLAGS_IN_HW
))
572 static void u32_clear_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
,
573 struct netlink_ext_ack
*extack
)
575 struct tc_u_common
*tp_c
= tp
->data
;
576 struct tc_u_knode
*n
;
579 for (h
= 0; h
<= ht
->divisor
; h
++) {
580 while ((n
= rtnl_dereference(ht
->ht
[h
])) != NULL
) {
581 RCU_INIT_POINTER(ht
->ht
[h
],
582 rtnl_dereference(n
->next
));
584 tcf_unbind_filter(tp
, &n
->res
);
585 u32_remove_hw_knode(tp
, n
, extack
);
586 idr_remove(&ht
->handle_idr
, n
->handle
);
587 if (tcf_exts_get_net(&n
->exts
))
588 tcf_queue_work(&n
->rwork
, u32_delete_key_freepf_work
);
590 u32_destroy_key(n
, true);
595 static int u32_destroy_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
,
596 struct netlink_ext_ack
*extack
)
598 struct tc_u_common
*tp_c
= tp
->data
;
599 struct tc_u_hnode __rcu
**hn
;
600 struct tc_u_hnode
*phn
;
602 WARN_ON(--ht
->refcnt
);
604 u32_clear_hnode(tp
, ht
, extack
);
607 for (phn
= rtnl_dereference(*hn
);
609 hn
= &phn
->next
, phn
= rtnl_dereference(*hn
)) {
611 u32_clear_hw_hnode(tp
, ht
, extack
);
612 idr_destroy(&ht
->handle_idr
);
613 idr_remove(&tp_c
->handle_idr
, ht
->handle
);
614 RCU_INIT_POINTER(*hn
, ht
->next
);
623 static void u32_destroy(struct tcf_proto
*tp
, bool rtnl_held
,
624 struct netlink_ext_ack
*extack
)
626 struct tc_u_common
*tp_c
= tp
->data
;
627 struct tc_u_hnode
*root_ht
= rtnl_dereference(tp
->root
);
629 WARN_ON(root_ht
== NULL
);
631 if (root_ht
&& --root_ht
->refcnt
== 1)
632 u32_destroy_hnode(tp
, root_ht
, extack
);
634 if (--tp_c
->refcnt
== 0) {
635 struct tc_u_hnode
*ht
;
637 hlist_del(&tp_c
->hnode
);
639 while ((ht
= rtnl_dereference(tp_c
->hlist
)) != NULL
) {
640 u32_clear_hnode(tp
, ht
, extack
);
641 RCU_INIT_POINTER(tp_c
->hlist
, ht
->next
);
643 /* u32_destroy_key() will later free ht for us, if it's
644 * still referenced by some knode
646 if (--ht
->refcnt
== 0)
650 idr_destroy(&tp_c
->handle_idr
);
657 static int u32_delete(struct tcf_proto
*tp
, void *arg
, bool *last
,
658 bool rtnl_held
, struct netlink_ext_ack
*extack
)
660 struct tc_u_hnode
*ht
= arg
;
661 struct tc_u_common
*tp_c
= tp
->data
;
664 if (TC_U32_KEY(ht
->handle
)) {
665 u32_remove_hw_knode(tp
, (struct tc_u_knode
*)ht
, extack
);
666 ret
= u32_delete_key(tp
, (struct tc_u_knode
*)ht
);
671 NL_SET_ERR_MSG_MOD(extack
, "Not allowed to delete root node");
675 if (ht
->refcnt
== 1) {
676 u32_destroy_hnode(tp
, ht
, extack
);
678 NL_SET_ERR_MSG_MOD(extack
, "Can not delete in-use filter");
683 *last
= tp_c
->refcnt
== 1 && tp_c
->knodes
== 0;
687 static u32
gen_new_kid(struct tc_u_hnode
*ht
, u32 htid
)
689 u32 index
= htid
| 0x800;
690 u32 max
= htid
| 0xFFF;
692 if (idr_alloc_u32(&ht
->handle_idr
, NULL
, &index
, max
, GFP_KERNEL
)) {
694 if (idr_alloc_u32(&ht
->handle_idr
, NULL
, &index
, max
,
702 static const struct nla_policy u32_policy
[TCA_U32_MAX
+ 1] = {
703 [TCA_U32_CLASSID
] = { .type
= NLA_U32
},
704 [TCA_U32_HASH
] = { .type
= NLA_U32
},
705 [TCA_U32_LINK
] = { .type
= NLA_U32
},
706 [TCA_U32_DIVISOR
] = { .type
= NLA_U32
},
707 [TCA_U32_SEL
] = { .len
= sizeof(struct tc_u32_sel
) },
708 [TCA_U32_INDEV
] = { .type
= NLA_STRING
, .len
= IFNAMSIZ
},
709 [TCA_U32_MARK
] = { .len
= sizeof(struct tc_u32_mark
) },
710 [TCA_U32_FLAGS
] = { .type
= NLA_U32
},
713 static int u32_set_parms(struct net
*net
, struct tcf_proto
*tp
,
715 struct tc_u_knode
*n
, struct nlattr
**tb
,
716 struct nlattr
*est
, u32 flags
,
717 struct netlink_ext_ack
*extack
)
721 err
= tcf_exts_validate(net
, tp
, tb
, est
, &n
->exts
, flags
, extack
);
725 if (tb
[TCA_U32_LINK
]) {
726 u32 handle
= nla_get_u32(tb
[TCA_U32_LINK
]);
727 struct tc_u_hnode
*ht_down
= NULL
, *ht_old
;
729 if (TC_U32_KEY(handle
)) {
730 NL_SET_ERR_MSG_MOD(extack
, "u32 Link handle must be a hash table");
735 ht_down
= u32_lookup_ht(tp
->data
, handle
);
738 NL_SET_ERR_MSG_MOD(extack
, "Link hash table not found");
741 if (ht_down
->is_root
) {
742 NL_SET_ERR_MSG_MOD(extack
, "Not linking to root node");
748 ht_old
= rtnl_dereference(n
->ht_down
);
749 rcu_assign_pointer(n
->ht_down
, ht_down
);
754 if (tb
[TCA_U32_CLASSID
]) {
755 n
->res
.classid
= nla_get_u32(tb
[TCA_U32_CLASSID
]);
756 tcf_bind_filter(tp
, &n
->res
, base
);
759 if (tb
[TCA_U32_INDEV
]) {
761 ret
= tcf_change_indev(net
, tb
[TCA_U32_INDEV
], extack
);
769 static void u32_replace_knode(struct tcf_proto
*tp
, struct tc_u_common
*tp_c
,
770 struct tc_u_knode
*n
)
772 struct tc_u_knode __rcu
**ins
;
773 struct tc_u_knode
*pins
;
774 struct tc_u_hnode
*ht
;
776 if (TC_U32_HTID(n
->handle
) == TC_U32_ROOT
)
777 ht
= rtnl_dereference(tp
->root
);
779 ht
= u32_lookup_ht(tp_c
, TC_U32_HTID(n
->handle
));
781 ins
= &ht
->ht
[TC_U32_HASH(n
->handle
)];
783 /* The node must always exist for it to be replaced if this is not the
784 * case then something went very wrong elsewhere.
786 for (pins
= rtnl_dereference(*ins
); ;
787 ins
= &pins
->next
, pins
= rtnl_dereference(*ins
))
788 if (pins
->handle
== n
->handle
)
791 idr_replace(&ht
->handle_idr
, n
, n
->handle
);
792 RCU_INIT_POINTER(n
->next
, pins
->next
);
793 rcu_assign_pointer(*ins
, n
);
796 static struct tc_u_knode
*u32_init_knode(struct net
*net
, struct tcf_proto
*tp
,
797 struct tc_u_knode
*n
)
799 struct tc_u_hnode
*ht
= rtnl_dereference(n
->ht_down
);
800 struct tc_u32_sel
*s
= &n
->sel
;
801 struct tc_u_knode
*new;
803 new = kzalloc(struct_size(new, sel
.keys
, s
->nkeys
), GFP_KERNEL
);
807 RCU_INIT_POINTER(new->next
, n
->next
);
808 new->handle
= n
->handle
;
809 RCU_INIT_POINTER(new->ht_up
, n
->ht_up
);
811 new->ifindex
= n
->ifindex
;
812 new->fshift
= n
->fshift
;
814 new->flags
= n
->flags
;
815 RCU_INIT_POINTER(new->ht_down
, ht
);
817 /* bump reference count as long as we hold pointer to structure */
821 #ifdef CONFIG_CLS_U32_PERF
822 /* Statistics may be incremented by readers during update
823 * so we must keep them in tact. When the node is later destroyed
824 * a special destroy call must be made to not free the pf memory.
829 #ifdef CONFIG_CLS_U32_MARK
832 /* Similarly success statistics must be moved as pointers */
833 new->pcpu_success
= n
->pcpu_success
;
835 memcpy(&new->sel
, s
, struct_size(s
, keys
, s
->nkeys
));
837 if (tcf_exts_init(&new->exts
, net
, TCA_U32_ACT
, TCA_U32_POLICE
)) {
845 static int u32_change(struct net
*net
, struct sk_buff
*in_skb
,
846 struct tcf_proto
*tp
, unsigned long base
, u32 handle
,
847 struct nlattr
**tca
, void **arg
, u32 flags
,
848 struct netlink_ext_ack
*extack
)
850 struct tc_u_common
*tp_c
= tp
->data
;
851 struct tc_u_hnode
*ht
;
852 struct tc_u_knode
*n
;
853 struct tc_u32_sel
*s
;
854 struct nlattr
*opt
= tca
[TCA_OPTIONS
];
855 struct nlattr
*tb
[TCA_U32_MAX
+ 1];
856 u32 htid
, userflags
= 0;
862 NL_SET_ERR_MSG_MOD(extack
, "Filter handle requires options");
869 err
= nla_parse_nested_deprecated(tb
, TCA_U32_MAX
, opt
, u32_policy
,
874 if (tb
[TCA_U32_FLAGS
]) {
875 userflags
= nla_get_u32(tb
[TCA_U32_FLAGS
]);
876 if (!tc_flags_valid(userflags
)) {
877 NL_SET_ERR_MSG_MOD(extack
, "Invalid filter flags");
884 struct tc_u_knode
*new;
886 if (TC_U32_KEY(n
->handle
) == 0) {
887 NL_SET_ERR_MSG_MOD(extack
, "Key node id cannot be zero");
891 if ((n
->flags
^ userflags
) &
892 ~(TCA_CLS_FLAGS_IN_HW
| TCA_CLS_FLAGS_NOT_IN_HW
)) {
893 NL_SET_ERR_MSG_MOD(extack
, "Key node flags do not match passed flags");
897 new = u32_init_knode(net
, tp
, n
);
901 err
= u32_set_parms(net
, tp
, base
, new, tb
,
902 tca
[TCA_RATE
], flags
, extack
);
905 __u32_destroy_key(new);
909 err
= u32_replace_hw_knode(tp
, new, flags
, extack
);
911 __u32_destroy_key(new);
915 if (!tc_in_hw(new->flags
))
916 new->flags
|= TCA_CLS_FLAGS_NOT_IN_HW
;
918 u32_replace_knode(tp
, tp_c
, new);
919 tcf_unbind_filter(tp
, &n
->res
);
920 tcf_exts_get_net(&n
->exts
);
921 tcf_queue_work(&n
->rwork
, u32_delete_key_work
);
925 if (tb
[TCA_U32_DIVISOR
]) {
926 unsigned int divisor
= nla_get_u32(tb
[TCA_U32_DIVISOR
]);
928 if (!is_power_of_2(divisor
)) {
929 NL_SET_ERR_MSG_MOD(extack
, "Divisor is not a power of 2");
932 if (divisor
-- > 0x100) {
933 NL_SET_ERR_MSG_MOD(extack
, "Exceeded maximum 256 hash buckets");
936 if (TC_U32_KEY(handle
)) {
937 NL_SET_ERR_MSG_MOD(extack
, "Divisor can only be used on a hash table");
940 ht
= kzalloc(struct_size(ht
, ht
, divisor
+ 1), GFP_KERNEL
);
944 handle
= gen_new_htid(tp
->data
, ht
);
950 err
= idr_alloc_u32(&tp_c
->handle_idr
, ht
, &handle
,
958 ht
->divisor
= divisor
;
961 idr_init(&ht
->handle_idr
);
962 ht
->flags
= userflags
;
964 err
= u32_replace_hw_hnode(tp
, ht
, userflags
, extack
);
966 idr_remove(&tp_c
->handle_idr
, handle
);
971 RCU_INIT_POINTER(ht
->next
, tp_c
->hlist
);
972 rcu_assign_pointer(tp_c
->hlist
, ht
);
978 if (tb
[TCA_U32_HASH
]) {
979 htid
= nla_get_u32(tb
[TCA_U32_HASH
]);
980 if (TC_U32_HTID(htid
) == TC_U32_ROOT
) {
981 ht
= rtnl_dereference(tp
->root
);
984 ht
= u32_lookup_ht(tp
->data
, TC_U32_HTID(htid
));
986 NL_SET_ERR_MSG_MOD(extack
, "Specified hash table not found");
991 ht
= rtnl_dereference(tp
->root
);
995 if (ht
->divisor
< TC_U32_HASH(htid
)) {
996 NL_SET_ERR_MSG_MOD(extack
, "Specified hash table buckets exceed configured value");
1001 if (TC_U32_HTID(handle
) && TC_U32_HTID(handle
^ htid
)) {
1002 NL_SET_ERR_MSG_MOD(extack
, "Handle specified hash table address mismatch");
1005 handle
= htid
| TC_U32_NODE(handle
);
1006 err
= idr_alloc_u32(&ht
->handle_idr
, NULL
, &handle
, handle
,
1011 handle
= gen_new_kid(ht
, htid
);
1013 if (tb
[TCA_U32_SEL
] == NULL
) {
1014 NL_SET_ERR_MSG_MOD(extack
, "Selector not specified");
1019 s
= nla_data(tb
[TCA_U32_SEL
]);
1020 sel_size
= struct_size(s
, keys
, s
->nkeys
);
1021 if (nla_len(tb
[TCA_U32_SEL
]) < sel_size
) {
1026 n
= kzalloc(struct_size(n
, sel
.keys
, s
->nkeys
), GFP_KERNEL
);
1032 #ifdef CONFIG_CLS_U32_PERF
1033 n
->pf
= __alloc_percpu(struct_size(n
->pf
, kcnts
, s
->nkeys
),
1034 __alignof__(struct tc_u32_pcnt
));
1041 memcpy(&n
->sel
, s
, sel_size
);
1042 RCU_INIT_POINTER(n
->ht_up
, ht
);
1044 n
->fshift
= s
->hmask
? ffs(ntohl(s
->hmask
)) - 1 : 0;
1045 n
->flags
= userflags
;
1047 err
= tcf_exts_init(&n
->exts
, net
, TCA_U32_ACT
, TCA_U32_POLICE
);
1051 #ifdef CONFIG_CLS_U32_MARK
1052 n
->pcpu_success
= alloc_percpu(u32
);
1053 if (!n
->pcpu_success
) {
1058 if (tb
[TCA_U32_MARK
]) {
1059 struct tc_u32_mark
*mark
;
1061 mark
= nla_data(tb
[TCA_U32_MARK
]);
1063 n
->mask
= mark
->mask
;
1067 err
= u32_set_parms(net
, tp
, base
, n
, tb
, tca
[TCA_RATE
], flags
,
1070 struct tc_u_knode __rcu
**ins
;
1071 struct tc_u_knode
*pins
;
1073 err
= u32_replace_hw_knode(tp
, n
, flags
, extack
);
1077 if (!tc_in_hw(n
->flags
))
1078 n
->flags
|= TCA_CLS_FLAGS_NOT_IN_HW
;
1080 ins
= &ht
->ht
[TC_U32_HASH(handle
)];
1081 for (pins
= rtnl_dereference(*ins
); pins
;
1082 ins
= &pins
->next
, pins
= rtnl_dereference(*ins
))
1083 if (TC_U32_NODE(handle
) < TC_U32_NODE(pins
->handle
))
1086 RCU_INIT_POINTER(n
->next
, pins
);
1087 rcu_assign_pointer(*ins
, n
);
1094 #ifdef CONFIG_CLS_U32_MARK
1095 free_percpu(n
->pcpu_success
);
1099 tcf_exts_destroy(&n
->exts
);
1100 #ifdef CONFIG_CLS_U32_PERF
1106 idr_remove(&ht
->handle_idr
, handle
);
1110 static void u32_walk(struct tcf_proto
*tp
, struct tcf_walker
*arg
,
1113 struct tc_u_common
*tp_c
= tp
->data
;
1114 struct tc_u_hnode
*ht
;
1115 struct tc_u_knode
*n
;
1121 for (ht
= rtnl_dereference(tp_c
->hlist
);
1123 ht
= rtnl_dereference(ht
->next
)) {
1124 if (ht
->prio
!= tp
->prio
)
1126 if (arg
->count
>= arg
->skip
) {
1127 if (arg
->fn(tp
, ht
, arg
) < 0) {
1133 for (h
= 0; h
<= ht
->divisor
; h
++) {
1134 for (n
= rtnl_dereference(ht
->ht
[h
]);
1136 n
= rtnl_dereference(n
->next
)) {
1137 if (arg
->count
< arg
->skip
) {
1141 if (arg
->fn(tp
, n
, arg
) < 0) {
1151 static int u32_reoffload_hnode(struct tcf_proto
*tp
, struct tc_u_hnode
*ht
,
1152 bool add
, flow_setup_cb_t
*cb
, void *cb_priv
,
1153 struct netlink_ext_ack
*extack
)
1155 struct tc_cls_u32_offload cls_u32
= {};
1158 tc_cls_common_offload_init(&cls_u32
.common
, tp
, ht
->flags
, extack
);
1159 cls_u32
.command
= add
? TC_CLSU32_NEW_HNODE
: TC_CLSU32_DELETE_HNODE
;
1160 cls_u32
.hnode
.divisor
= ht
->divisor
;
1161 cls_u32
.hnode
.handle
= ht
->handle
;
1162 cls_u32
.hnode
.prio
= ht
->prio
;
1164 err
= cb(TC_SETUP_CLSU32
, &cls_u32
, cb_priv
);
1165 if (err
&& add
&& tc_skip_sw(ht
->flags
))
1171 static int u32_reoffload_knode(struct tcf_proto
*tp
, struct tc_u_knode
*n
,
1172 bool add
, flow_setup_cb_t
*cb
, void *cb_priv
,
1173 struct netlink_ext_ack
*extack
)
1175 struct tc_u_hnode
*ht
= rtnl_dereference(n
->ht_down
);
1176 struct tcf_block
*block
= tp
->chain
->block
;
1177 struct tc_cls_u32_offload cls_u32
= {};
1179 tc_cls_common_offload_init(&cls_u32
.common
, tp
, n
->flags
, extack
);
1180 cls_u32
.command
= add
?
1181 TC_CLSU32_REPLACE_KNODE
: TC_CLSU32_DELETE_KNODE
;
1182 cls_u32
.knode
.handle
= n
->handle
;
1185 cls_u32
.knode
.fshift
= n
->fshift
;
1186 #ifdef CONFIG_CLS_U32_MARK
1187 cls_u32
.knode
.val
= n
->val
;
1188 cls_u32
.knode
.mask
= n
->mask
;
1190 cls_u32
.knode
.val
= 0;
1191 cls_u32
.knode
.mask
= 0;
1193 cls_u32
.knode
.sel
= &n
->sel
;
1194 cls_u32
.knode
.res
= &n
->res
;
1195 cls_u32
.knode
.exts
= &n
->exts
;
1197 cls_u32
.knode
.link_handle
= ht
->handle
;
1200 return tc_setup_cb_reoffload(block
, tp
, add
, cb
, TC_SETUP_CLSU32
,
1201 &cls_u32
, cb_priv
, &n
->flags
,
1205 static int u32_reoffload(struct tcf_proto
*tp
, bool add
, flow_setup_cb_t
*cb
,
1206 void *cb_priv
, struct netlink_ext_ack
*extack
)
1208 struct tc_u_common
*tp_c
= tp
->data
;
1209 struct tc_u_hnode
*ht
;
1210 struct tc_u_knode
*n
;
1214 for (ht
= rtnl_dereference(tp_c
->hlist
);
1216 ht
= rtnl_dereference(ht
->next
)) {
1217 if (ht
->prio
!= tp
->prio
)
1220 /* When adding filters to a new dev, try to offload the
1221 * hashtable first. When removing, do the filters before the
1224 if (add
&& !tc_skip_hw(ht
->flags
)) {
1225 err
= u32_reoffload_hnode(tp
, ht
, add
, cb
, cb_priv
,
1231 for (h
= 0; h
<= ht
->divisor
; h
++) {
1232 for (n
= rtnl_dereference(ht
->ht
[h
]);
1234 n
= rtnl_dereference(n
->next
)) {
1235 if (tc_skip_hw(n
->flags
))
1238 err
= u32_reoffload_knode(tp
, n
, add
, cb
,
1245 if (!add
&& !tc_skip_hw(ht
->flags
))
1246 u32_reoffload_hnode(tp
, ht
, add
, cb
, cb_priv
, extack
);
1252 static void u32_bind_class(void *fh
, u32 classid
, unsigned long cl
, void *q
,
1255 struct tc_u_knode
*n
= fh
;
1257 if (n
&& n
->res
.classid
== classid
) {
1259 __tcf_bind_filter(q
, &n
->res
, base
);
1261 __tcf_unbind_filter(q
, &n
->res
);
1265 static int u32_dump(struct net
*net
, struct tcf_proto
*tp
, void *fh
,
1266 struct sk_buff
*skb
, struct tcmsg
*t
, bool rtnl_held
)
1268 struct tc_u_knode
*n
= fh
;
1269 struct tc_u_hnode
*ht_up
, *ht_down
;
1270 struct nlattr
*nest
;
1275 t
->tcm_handle
= n
->handle
;
1277 nest
= nla_nest_start_noflag(skb
, TCA_OPTIONS
);
1279 goto nla_put_failure
;
1281 if (TC_U32_KEY(n
->handle
) == 0) {
1282 struct tc_u_hnode
*ht
= fh
;
1283 u32 divisor
= ht
->divisor
+ 1;
1285 if (nla_put_u32(skb
, TCA_U32_DIVISOR
, divisor
))
1286 goto nla_put_failure
;
1288 #ifdef CONFIG_CLS_U32_PERF
1289 struct tc_u32_pcnt
*gpf
;
1293 if (nla_put(skb
, TCA_U32_SEL
, struct_size(&n
->sel
, keys
, n
->sel
.nkeys
),
1295 goto nla_put_failure
;
1297 ht_up
= rtnl_dereference(n
->ht_up
);
1299 u32 htid
= n
->handle
& 0xFFFFF000;
1300 if (nla_put_u32(skb
, TCA_U32_HASH
, htid
))
1301 goto nla_put_failure
;
1303 if (n
->res
.classid
&&
1304 nla_put_u32(skb
, TCA_U32_CLASSID
, n
->res
.classid
))
1305 goto nla_put_failure
;
1307 ht_down
= rtnl_dereference(n
->ht_down
);
1309 nla_put_u32(skb
, TCA_U32_LINK
, ht_down
->handle
))
1310 goto nla_put_failure
;
1312 if (n
->flags
&& nla_put_u32(skb
, TCA_U32_FLAGS
, n
->flags
))
1313 goto nla_put_failure
;
1315 #ifdef CONFIG_CLS_U32_MARK
1316 if ((n
->val
|| n
->mask
)) {
1317 struct tc_u32_mark mark
= {.val
= n
->val
,
1322 for_each_possible_cpu(cpum
) {
1323 __u32 cnt
= *per_cpu_ptr(n
->pcpu_success
, cpum
);
1325 mark
.success
+= cnt
;
1328 if (nla_put(skb
, TCA_U32_MARK
, sizeof(mark
), &mark
))
1329 goto nla_put_failure
;
1333 if (tcf_exts_dump(skb
, &n
->exts
) < 0)
1334 goto nla_put_failure
;
1337 struct net_device
*dev
;
1338 dev
= __dev_get_by_index(net
, n
->ifindex
);
1339 if (dev
&& nla_put_string(skb
, TCA_U32_INDEV
, dev
->name
))
1340 goto nla_put_failure
;
1342 #ifdef CONFIG_CLS_U32_PERF
1343 gpf
= kzalloc(struct_size(gpf
, kcnts
, n
->sel
.nkeys
), GFP_KERNEL
);
1345 goto nla_put_failure
;
1347 for_each_possible_cpu(cpu
) {
1349 struct tc_u32_pcnt
*pf
= per_cpu_ptr(n
->pf
, cpu
);
1351 gpf
->rcnt
+= pf
->rcnt
;
1352 gpf
->rhit
+= pf
->rhit
;
1353 for (i
= 0; i
< n
->sel
.nkeys
; i
++)
1354 gpf
->kcnts
[i
] += pf
->kcnts
[i
];
1357 if (nla_put_64bit(skb
, TCA_U32_PCNT
, struct_size(gpf
, kcnts
, n
->sel
.nkeys
),
1358 gpf
, TCA_U32_PAD
)) {
1360 goto nla_put_failure
;
1366 nla_nest_end(skb
, nest
);
1368 if (TC_U32_KEY(n
->handle
))
1369 if (tcf_exts_dump_stats(skb
, &n
->exts
) < 0)
1370 goto nla_put_failure
;
1374 nla_nest_cancel(skb
, nest
);
1378 static struct tcf_proto_ops cls_u32_ops __read_mostly
= {
1380 .classify
= u32_classify
,
1382 .destroy
= u32_destroy
,
1384 .change
= u32_change
,
1385 .delete = u32_delete
,
1387 .reoffload
= u32_reoffload
,
1389 .bind_class
= u32_bind_class
,
1390 .owner
= THIS_MODULE
,
1393 static int __init
init_u32(void)
1397 pr_info("u32 classifier\n");
1398 #ifdef CONFIG_CLS_U32_PERF
1399 pr_info(" Performance counters on\n");
1401 pr_info(" input device check on\n");
1402 #ifdef CONFIG_NET_CLS_ACT
1403 pr_info(" Actions configured\n");
1405 tc_u_common_hash
= kvmalloc_array(U32_HASH_SIZE
,
1406 sizeof(struct hlist_head
),
1408 if (!tc_u_common_hash
)
1411 for (i
= 0; i
< U32_HASH_SIZE
; i
++)
1412 INIT_HLIST_HEAD(&tc_u_common_hash
[i
]);
1414 ret
= register_tcf_proto_ops(&cls_u32_ops
);
1416 kvfree(tc_u_common_hash
);
1420 static void __exit
exit_u32(void)
1422 unregister_tcf_proto_ops(&cls_u32_ops
);
1423 kvfree(tc_u_common_hash
);
1426 module_init(init_u32
)
1427 module_exit(exit_u32
)
1428 MODULE_LICENSE("GPL");