2 * net/sched/cls_tcindex.c Packet classifier for skb->tc_index
4 * Written 1998,1999 by Werner Almesberger, EPFL ICA
7 #include <linux/module.h>
8 #include <linux/types.h>
9 #include <linux/kernel.h>
10 #include <linux/skbuff.h>
11 #include <linux/errno.h>
12 #include <linux/slab.h>
13 #include <net/act_api.h>
14 #include <net/netlink.h>
15 #include <net/pkt_cls.h>
16 #include <net/sch_generic.h>
19 * Passing parameters to the root seems to be done more awkwardly than really
20 * necessary. At least, u32 doesn't seem to use such dirty hacks. To be
24 #define PERFECT_HASH_THRESHOLD 64 /* use perfect hash if not bigger */
25 #define DEFAULT_HASH_SIZE 64 /* optimized for diffserv */
28 struct tcindex_filter_result
{
30 struct tcf_result res
;
32 struct work_struct work
;
37 struct tcindex_filter
{
39 struct tcindex_filter_result result
;
40 struct tcindex_filter __rcu
*next
;
42 struct work_struct work
;
49 struct tcindex_filter_result
*perfect
; /* perfect hash; NULL if none */
50 struct tcindex_filter __rcu
**h
; /* imperfect hash; */
52 u16 mask
; /* AND key with mask */
53 u32 shift
; /* shift ANDed key to the right */
54 u32 hash
; /* hash table size; 0 if undefined */
55 u32 alloc_hash
; /* allocated size */
56 u32 fall_through
; /* 0: only classify if explicit match */
60 static inline int tcindex_filter_is_set(struct tcindex_filter_result
*r
)
62 return tcf_exts_has_actions(&r
->exts
) || r
->res
.classid
;
65 static struct tcindex_filter_result
*tcindex_lookup(struct tcindex_data
*p
,
69 struct tcindex_filter_result
*f
= p
->perfect
+ key
;
71 return tcindex_filter_is_set(f
) ? f
: NULL
;
73 struct tcindex_filter __rcu
**fp
;
74 struct tcindex_filter
*f
;
76 fp
= &p
->h
[key
% p
->hash
];
77 for (f
= rcu_dereference_bh_rtnl(*fp
);
79 fp
= &f
->next
, f
= rcu_dereference_bh_rtnl(*fp
))
88 static int tcindex_classify(struct sk_buff
*skb
, const struct tcf_proto
*tp
,
89 struct tcf_result
*res
)
91 struct tcindex_data
*p
= rcu_dereference_bh(tp
->root
);
92 struct tcindex_filter_result
*f
;
93 int key
= (skb
->tc_index
& p
->mask
) >> p
->shift
;
95 pr_debug("tcindex_classify(skb %p,tp %p,res %p),p %p\n",
98 f
= tcindex_lookup(p
, key
);
100 struct Qdisc
*q
= tcf_block_q(tp
->chain
->block
);
102 if (!p
->fall_through
)
104 res
->classid
= TC_H_MAKE(TC_H_MAJ(q
->handle
), key
);
106 pr_debug("alg 0x%x\n", res
->classid
);
110 pr_debug("map 0x%x\n", res
->classid
);
112 return tcf_exts_exec(skb
, &f
->exts
, res
);
116 static void *tcindex_get(struct tcf_proto
*tp
, u32 handle
)
118 struct tcindex_data
*p
= rtnl_dereference(tp
->root
);
119 struct tcindex_filter_result
*r
;
121 pr_debug("tcindex_get(tp %p,handle 0x%08x)\n", tp
, handle
);
122 if (p
->perfect
&& handle
>= p
->alloc_hash
)
124 r
= tcindex_lookup(p
, handle
);
125 return r
&& tcindex_filter_is_set(r
) ? r
: NULL
;
128 static int tcindex_init(struct tcf_proto
*tp
)
130 struct tcindex_data
*p
;
132 pr_debug("tcindex_init(tp %p)\n", tp
);
133 p
= kzalloc(sizeof(struct tcindex_data
), GFP_KERNEL
);
138 p
->hash
= DEFAULT_HASH_SIZE
;
141 rcu_assign_pointer(tp
->root
, p
);
145 static void __tcindex_destroy_rexts(struct tcindex_filter_result
*r
)
147 tcf_exts_destroy(&r
->exts
);
148 tcf_exts_put_net(&r
->exts
);
151 static void tcindex_destroy_rexts_work(struct work_struct
*work
)
153 struct tcindex_filter_result
*r
;
155 r
= container_of(work
, struct tcindex_filter_result
, work
);
157 __tcindex_destroy_rexts(r
);
161 static void tcindex_destroy_rexts(struct rcu_head
*head
)
163 struct tcindex_filter_result
*r
;
165 r
= container_of(head
, struct tcindex_filter_result
, rcu
);
166 INIT_WORK(&r
->work
, tcindex_destroy_rexts_work
);
167 tcf_queue_work(&r
->work
);
170 static void __tcindex_destroy_fexts(struct tcindex_filter
*f
)
172 tcf_exts_destroy(&f
->result
.exts
);
173 tcf_exts_put_net(&f
->result
.exts
);
177 static void tcindex_destroy_fexts_work(struct work_struct
*work
)
179 struct tcindex_filter
*f
= container_of(work
, struct tcindex_filter
,
183 __tcindex_destroy_fexts(f
);
187 static void tcindex_destroy_fexts(struct rcu_head
*head
)
189 struct tcindex_filter
*f
= container_of(head
, struct tcindex_filter
,
192 INIT_WORK(&f
->work
, tcindex_destroy_fexts_work
);
193 tcf_queue_work(&f
->work
);
196 static int tcindex_delete(struct tcf_proto
*tp
, void *arg
, bool *last
)
198 struct tcindex_data
*p
= rtnl_dereference(tp
->root
);
199 struct tcindex_filter_result
*r
= arg
;
200 struct tcindex_filter __rcu
**walk
;
201 struct tcindex_filter
*f
= NULL
;
203 pr_debug("tcindex_delete(tp %p,arg %p),p %p\n", tp
, arg
, p
);
210 for (i
= 0; i
< p
->hash
; i
++) {
212 for (f
= rtnl_dereference(*walk
); f
;
213 walk
= &f
->next
, f
= rtnl_dereference(*walk
)) {
221 rcu_assign_pointer(*walk
, rtnl_dereference(f
->next
));
223 tcf_unbind_filter(tp
, &r
->res
);
224 /* all classifiers are required to call tcf_exts_destroy() after rcu
225 * grace period, since converted-to-rcu actions are relying on that
226 * in cleanup() callback
229 if (tcf_exts_get_net(&f
->result
.exts
))
230 call_rcu(&f
->rcu
, tcindex_destroy_fexts
);
232 __tcindex_destroy_fexts(f
);
234 if (tcf_exts_get_net(&r
->exts
))
235 call_rcu(&r
->rcu
, tcindex_destroy_rexts
);
237 __tcindex_destroy_rexts(r
);
244 static int tcindex_destroy_element(struct tcf_proto
*tp
,
245 void *arg
, struct tcf_walker
*walker
)
249 return tcindex_delete(tp
, arg
, &last
);
252 static void __tcindex_destroy(struct rcu_head
*head
)
254 struct tcindex_data
*p
= container_of(head
, struct tcindex_data
, rcu
);
262 valid_perfect_hash(struct tcindex_data
*p
)
264 return p
->hash
> (p
->mask
>> p
->shift
);
267 static const struct nla_policy tcindex_policy
[TCA_TCINDEX_MAX
+ 1] = {
268 [TCA_TCINDEX_HASH
] = { .type
= NLA_U32
},
269 [TCA_TCINDEX_MASK
] = { .type
= NLA_U16
},
270 [TCA_TCINDEX_SHIFT
] = { .type
= NLA_U32
},
271 [TCA_TCINDEX_FALL_THROUGH
] = { .type
= NLA_U32
},
272 [TCA_TCINDEX_CLASSID
] = { .type
= NLA_U32
},
275 static int tcindex_filter_result_init(struct tcindex_filter_result
*r
)
277 memset(r
, 0, sizeof(*r
));
278 return tcf_exts_init(&r
->exts
, TCA_TCINDEX_ACT
, TCA_TCINDEX_POLICE
);
281 static void __tcindex_partial_destroy(struct rcu_head
*head
)
283 struct tcindex_data
*p
= container_of(head
, struct tcindex_data
, rcu
);
289 static void tcindex_free_perfect_hash(struct tcindex_data
*cp
)
293 for (i
= 0; i
< cp
->hash
; i
++)
294 tcf_exts_destroy(&cp
->perfect
[i
].exts
);
298 static int tcindex_alloc_perfect_hash(struct tcindex_data
*cp
)
302 cp
->perfect
= kcalloc(cp
->hash
, sizeof(struct tcindex_filter_result
),
307 for (i
= 0; i
< cp
->hash
; i
++) {
308 err
= tcf_exts_init(&cp
->perfect
[i
].exts
,
309 TCA_TCINDEX_ACT
, TCA_TCINDEX_POLICE
);
317 tcindex_free_perfect_hash(cp
);
322 tcindex_set_parms(struct net
*net
, struct tcf_proto
*tp
, unsigned long base
,
323 u32 handle
, struct tcindex_data
*p
,
324 struct tcindex_filter_result
*r
, struct nlattr
**tb
,
325 struct nlattr
*est
, bool ovr
)
327 struct tcindex_filter_result new_filter_result
, *old_r
= r
;
328 struct tcindex_filter_result cr
;
329 struct tcindex_data
*cp
= NULL
, *oldp
;
330 struct tcindex_filter
*f
= NULL
; /* make gcc behave */
334 err
= tcf_exts_init(&e
, TCA_TCINDEX_ACT
, TCA_TCINDEX_POLICE
);
337 err
= tcf_exts_validate(net
, tp
, tb
, est
, &e
, ovr
);
342 /* tcindex_data attributes must look atomic to classifier/lookup so
343 * allocate new tcindex data and RCU assign it onto root. Keeping
344 * perfect hash and hash pointers from old data.
346 cp
= kzalloc(sizeof(*cp
), GFP_KERNEL
);
351 cp
->shift
= p
->shift
;
353 cp
->alloc_hash
= p
->alloc_hash
;
354 cp
->fall_through
= p
->fall_through
;
360 if (tcindex_alloc_perfect_hash(cp
) < 0)
362 for (i
= 0; i
< cp
->hash
; i
++)
363 cp
->perfect
[i
].res
= p
->perfect
[i
].res
;
368 err
= tcindex_filter_result_init(&new_filter_result
);
371 err
= tcindex_filter_result_init(&cr
);
377 if (tb
[TCA_TCINDEX_HASH
])
378 cp
->hash
= nla_get_u32(tb
[TCA_TCINDEX_HASH
]);
380 if (tb
[TCA_TCINDEX_MASK
])
381 cp
->mask
= nla_get_u16(tb
[TCA_TCINDEX_MASK
]);
383 if (tb
[TCA_TCINDEX_SHIFT
])
384 cp
->shift
= nla_get_u32(tb
[TCA_TCINDEX_SHIFT
]);
388 /* Hash already allocated, make sure that we still meet the
389 * requirements for the allocated hash.
392 if (!valid_perfect_hash(cp
) ||
393 cp
->hash
> cp
->alloc_hash
)
395 } else if (cp
->h
&& cp
->hash
!= cp
->alloc_hash
) {
400 if (tb
[TCA_TCINDEX_FALL_THROUGH
])
401 cp
->fall_through
= nla_get_u32(tb
[TCA_TCINDEX_FALL_THROUGH
]);
404 /* Hash not specified, use perfect hash if the upper limit
405 * of the hashing index is below the threshold.
407 if ((cp
->mask
>> cp
->shift
) < PERFECT_HASH_THRESHOLD
)
408 cp
->hash
= (cp
->mask
>> cp
->shift
) + 1;
410 cp
->hash
= DEFAULT_HASH_SIZE
;
413 if (!cp
->perfect
&& !cp
->h
)
414 cp
->alloc_hash
= cp
->hash
;
416 /* Note: this could be as restrictive as if (handle & ~(mask >> shift))
417 * but then, we'd fail handles that may become valid after some future
418 * mask change. While this is extremely unlikely to ever matter,
419 * the check below is safer (and also more backwards-compatible).
421 if (cp
->perfect
|| valid_perfect_hash(cp
))
422 if (handle
>= cp
->alloc_hash
)
427 if (!cp
->perfect
&& !cp
->h
) {
428 if (valid_perfect_hash(cp
)) {
429 if (tcindex_alloc_perfect_hash(cp
) < 0)
433 struct tcindex_filter __rcu
**hash
;
435 hash
= kcalloc(cp
->hash
,
436 sizeof(struct tcindex_filter
*),
448 r
= cp
->perfect
+ handle
;
450 r
= tcindex_lookup(cp
, handle
) ? : &new_filter_result
;
452 if (r
== &new_filter_result
) {
453 f
= kzalloc(sizeof(*f
), GFP_KERNEL
);
458 err
= tcindex_filter_result_init(&f
->result
);
465 if (tb
[TCA_TCINDEX_CLASSID
]) {
466 cr
.res
.classid
= nla_get_u32(tb
[TCA_TCINDEX_CLASSID
]);
467 tcf_bind_filter(tp
, &cr
.res
, base
);
471 tcf_exts_change(&r
->exts
, &e
);
473 tcf_exts_change(&cr
.exts
, &e
);
475 if (old_r
&& old_r
!= r
) {
476 err
= tcindex_filter_result_init(old_r
);
485 rcu_assign_pointer(tp
->root
, cp
);
487 if (r
== &new_filter_result
) {
488 struct tcindex_filter
*nfp
;
489 struct tcindex_filter __rcu
**fp
;
491 tcf_exts_change(&f
->result
.exts
, &r
->exts
);
493 fp
= cp
->h
+ (handle
% cp
->hash
);
494 for (nfp
= rtnl_dereference(*fp
);
496 fp
= &nfp
->next
, nfp
= rtnl_dereference(*fp
))
499 rcu_assign_pointer(*fp
, f
);
503 call_rcu(&oldp
->rcu
, __tcindex_partial_destroy
);
508 tcindex_free_perfect_hash(cp
);
509 else if (balloc
== 2)
512 tcf_exts_destroy(&cr
.exts
);
513 tcf_exts_destroy(&new_filter_result
.exts
);
516 tcf_exts_destroy(&e
);
521 tcindex_change(struct net
*net
, struct sk_buff
*in_skb
,
522 struct tcf_proto
*tp
, unsigned long base
, u32 handle
,
523 struct nlattr
**tca
, void **arg
, bool ovr
)
525 struct nlattr
*opt
= tca
[TCA_OPTIONS
];
526 struct nlattr
*tb
[TCA_TCINDEX_MAX
+ 1];
527 struct tcindex_data
*p
= rtnl_dereference(tp
->root
);
528 struct tcindex_filter_result
*r
= *arg
;
531 pr_debug("tcindex_change(tp %p,handle 0x%08x,tca %p,arg %p),opt %p,"
532 "p %p,r %p,*arg %p\n",
533 tp
, handle
, tca
, arg
, opt
, p
, r
, arg
? *arg
: NULL
);
538 err
= nla_parse_nested(tb
, TCA_TCINDEX_MAX
, opt
, tcindex_policy
, NULL
);
542 return tcindex_set_parms(net
, tp
, base
, handle
, p
, r
, tb
,
546 static void tcindex_walk(struct tcf_proto
*tp
, struct tcf_walker
*walker
)
548 struct tcindex_data
*p
= rtnl_dereference(tp
->root
);
549 struct tcindex_filter
*f
, *next
;
552 pr_debug("tcindex_walk(tp %p,walker %p),p %p\n", tp
, walker
, p
);
554 for (i
= 0; i
< p
->hash
; i
++) {
555 if (!p
->perfect
[i
].res
.class)
557 if (walker
->count
>= walker
->skip
) {
558 if (walker
->fn(tp
, p
->perfect
+ i
, walker
) < 0) {
568 for (i
= 0; i
< p
->hash
; i
++) {
569 for (f
= rtnl_dereference(p
->h
[i
]); f
; f
= next
) {
570 next
= rtnl_dereference(f
->next
);
571 if (walker
->count
>= walker
->skip
) {
572 if (walker
->fn(tp
, &f
->result
, walker
) < 0) {
582 static void tcindex_destroy(struct tcf_proto
*tp
)
584 struct tcindex_data
*p
= rtnl_dereference(tp
->root
);
585 struct tcf_walker walker
;
587 pr_debug("tcindex_destroy(tp %p),p %p\n", tp
, p
);
590 walker
.fn
= tcindex_destroy_element
;
591 tcindex_walk(tp
, &walker
);
593 call_rcu(&p
->rcu
, __tcindex_destroy
);
597 static int tcindex_dump(struct net
*net
, struct tcf_proto
*tp
, void *fh
,
598 struct sk_buff
*skb
, struct tcmsg
*t
)
600 struct tcindex_data
*p
= rtnl_dereference(tp
->root
);
601 struct tcindex_filter_result
*r
= fh
;
604 pr_debug("tcindex_dump(tp %p,fh %p,skb %p,t %p),p %p,r %p\n",
605 tp
, fh
, skb
, t
, p
, r
);
606 pr_debug("p->perfect %p p->h %p\n", p
->perfect
, p
->h
);
608 nest
= nla_nest_start(skb
, TCA_OPTIONS
);
610 goto nla_put_failure
;
613 t
->tcm_handle
= ~0; /* whatever ... */
614 if (nla_put_u32(skb
, TCA_TCINDEX_HASH
, p
->hash
) ||
615 nla_put_u16(skb
, TCA_TCINDEX_MASK
, p
->mask
) ||
616 nla_put_u32(skb
, TCA_TCINDEX_SHIFT
, p
->shift
) ||
617 nla_put_u32(skb
, TCA_TCINDEX_FALL_THROUGH
, p
->fall_through
))
618 goto nla_put_failure
;
619 nla_nest_end(skb
, nest
);
622 t
->tcm_handle
= r
- p
->perfect
;
624 struct tcindex_filter
*f
;
625 struct tcindex_filter __rcu
**fp
;
629 for (i
= 0; !t
->tcm_handle
&& i
< p
->hash
; i
++) {
631 for (f
= rtnl_dereference(*fp
);
633 fp
= &f
->next
, f
= rtnl_dereference(*fp
)) {
635 t
->tcm_handle
= f
->key
;
639 pr_debug("handle = %d\n", t
->tcm_handle
);
641 nla_put_u32(skb
, TCA_TCINDEX_CLASSID
, r
->res
.classid
))
642 goto nla_put_failure
;
644 if (tcf_exts_dump(skb
, &r
->exts
) < 0)
645 goto nla_put_failure
;
646 nla_nest_end(skb
, nest
);
648 if (tcf_exts_dump_stats(skb
, &r
->exts
) < 0)
649 goto nla_put_failure
;
655 nla_nest_cancel(skb
, nest
);
659 static void tcindex_bind_class(void *fh
, u32 classid
, unsigned long cl
)
661 struct tcindex_filter_result
*r
= fh
;
663 if (r
&& r
->res
.classid
== classid
)
667 static struct tcf_proto_ops cls_tcindex_ops __read_mostly
= {
669 .classify
= tcindex_classify
,
670 .init
= tcindex_init
,
671 .destroy
= tcindex_destroy
,
673 .change
= tcindex_change
,
674 .delete = tcindex_delete
,
675 .walk
= tcindex_walk
,
676 .dump
= tcindex_dump
,
677 .bind_class
= tcindex_bind_class
,
678 .owner
= THIS_MODULE
,
681 static int __init
init_tcindex(void)
683 return register_tcf_proto_ops(&cls_tcindex_ops
);
686 static void __exit
exit_tcindex(void)
688 unregister_tcf_proto_ops(&cls_tcindex_ops
);
691 module_init(init_tcindex
)
692 module_exit(exit_tcindex
)
693 MODULE_LICENSE("GPL");