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>
18 * Passing parameters to the root seems to be done more awkwardly than really
19 * necessary. At least, u32 doesn't seem to use such dirty hacks. To be
23 #define PERFECT_HASH_THRESHOLD 64 /* use perfect hash if not bigger */
24 #define DEFAULT_HASH_SIZE 64 /* optimized for diffserv */
27 struct tcindex_filter_result {
29 struct tcf_result res;
32 struct tcindex_filter {
34 struct tcindex_filter_result result;
35 struct tcindex_filter __rcu *next;
41 struct tcindex_filter_result *perfect; /* perfect hash; NULL if none */
42 struct tcindex_filter __rcu **h; /* imperfect hash; */
44 u16 mask; /* AND key with mask */
45 u32 shift; /* shift ANDed key to the right */
46 u32 hash; /* hash table size; 0 if undefined */
47 u32 alloc_hash; /* allocated size */
48 u32 fall_through; /* 0: only classify if explicit match */
53 tcindex_filter_is_set(struct tcindex_filter_result *r)
55 return tcf_exts_is_predicative(&r->exts) || r->res.classid;
58 static struct tcindex_filter_result *
59 tcindex_lookup(struct tcindex_data *p, u16 key)
62 struct tcindex_filter_result *f = p->perfect + key;
64 return tcindex_filter_is_set(f) ? f : NULL;
66 struct tcindex_filter __rcu **fp;
67 struct tcindex_filter *f;
69 fp = &p->h[key % p->hash];
70 for (f = rcu_dereference_bh_rtnl(*fp);
72 fp = &f->next, f = rcu_dereference_bh_rtnl(*fp))
81 static int tcindex_classify(struct sk_buff *skb, const struct tcf_proto *tp,
82 struct tcf_result *res)
84 struct tcindex_data *p = rcu_dereference(tp->root);
85 struct tcindex_filter_result *f;
86 int key = (skb->tc_index & p->mask) >> p->shift;
88 pr_debug("tcindex_classify(skb %p,tp %p,res %p),p %p\n",
91 f = tcindex_lookup(p, key);
95 res->classid = TC_H_MAKE(TC_H_MAJ(tp->q->handle), key);
97 pr_debug("alg 0x%x\n", res->classid);
101 pr_debug("map 0x%x\n", res->classid);
103 return tcf_exts_exec(skb, &f->exts, res);
107 static unsigned long tcindex_get(struct tcf_proto *tp, u32 handle)
109 struct tcindex_data *p = rtnl_dereference(tp->root);
110 struct tcindex_filter_result *r;
112 pr_debug("tcindex_get(tp %p,handle 0x%08x)\n", tp, handle);
113 if (p->perfect && handle >= p->alloc_hash)
115 r = tcindex_lookup(p, handle);
116 return r && tcindex_filter_is_set(r) ? (unsigned long) r : 0UL;
120 static void tcindex_put(struct tcf_proto *tp, unsigned long f)
122 pr_debug("tcindex_put(tp %p,f 0x%lx)\n", tp, f);
126 static int tcindex_init(struct tcf_proto *tp)
128 struct tcindex_data *p;
130 pr_debug("tcindex_init(tp %p)\n", tp);
131 p = kzalloc(sizeof(struct tcindex_data), GFP_KERNEL);
136 p->hash = DEFAULT_HASH_SIZE;
139 rcu_assign_pointer(tp->root, p);
144 tcindex_delete(struct tcf_proto *tp, unsigned long arg)
146 struct tcindex_data *p = rtnl_dereference(tp->root);
147 struct tcindex_filter_result *r = (struct tcindex_filter_result *) arg;
148 struct tcindex_filter __rcu **walk;
149 struct tcindex_filter *f = NULL;
151 pr_debug("tcindex_delete(tp %p,arg 0x%lx),p %p\n", tp, arg, p);
158 for (i = 0; i < p->hash; i++) {
160 for (f = rtnl_dereference(*walk); f;
161 walk = &f->next, f = rtnl_dereference(*walk)) {
169 rcu_assign_pointer(*walk, rtnl_dereference(f->next));
171 tcf_unbind_filter(tp, &r->res);
172 tcf_exts_destroy(tp, &r->exts);
178 static int tcindex_destroy_element(struct tcf_proto *tp,
180 struct tcf_walker *walker)
182 return tcindex_delete(tp, arg);
185 static void __tcindex_destroy(struct rcu_head *head)
187 struct tcindex_data *p = container_of(head, struct tcindex_data, rcu);
195 valid_perfect_hash(struct tcindex_data *p)
197 return p->hash > (p->mask >> p->shift);
200 static const struct nla_policy tcindex_policy[TCA_TCINDEX_MAX + 1] = {
201 [TCA_TCINDEX_HASH] = { .type = NLA_U32 },
202 [TCA_TCINDEX_MASK] = { .type = NLA_U16 },
203 [TCA_TCINDEX_SHIFT] = { .type = NLA_U32 },
204 [TCA_TCINDEX_FALL_THROUGH] = { .type = NLA_U32 },
205 [TCA_TCINDEX_CLASSID] = { .type = NLA_U32 },
208 static void tcindex_filter_result_init(struct tcindex_filter_result *r)
210 memset(r, 0, sizeof(*r));
211 tcf_exts_init(&r->exts, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
214 static void __tcindex_partial_destroy(struct rcu_head *head)
216 struct tcindex_data *p = container_of(head, struct tcindex_data, rcu);
223 tcindex_set_parms(struct net *net, struct tcf_proto *tp, unsigned long base,
224 u32 handle, struct tcindex_data *p,
225 struct tcindex_filter_result *r, struct nlattr **tb,
226 struct nlattr *est, bool ovr)
229 struct tcindex_filter_result new_filter_result, *old_r = r;
230 struct tcindex_filter_result cr;
231 struct tcindex_data *cp, *oldp;
232 struct tcindex_filter *f = NULL; /* make gcc behave */
235 tcf_exts_init(&e, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
236 err = tcf_exts_validate(net, tp, tb, est, &e, ovr);
240 /* tcindex_data attributes must look atomic to classifier/lookup so
241 * allocate new tcindex data and RCU assign it onto root. Keeping
242 * perfect hash and hash pointers from old data.
244 cp = kzalloc(sizeof(*cp), GFP_KERNEL);
249 cp->shift = p->shift;
251 cp->alloc_hash = p->alloc_hash;
252 cp->fall_through = p->fall_through;
256 cp->perfect = kmemdup(p->perfect,
257 sizeof(*r) * cp->hash, GFP_KERNEL);
263 memset(&new_filter_result, 0, sizeof(new_filter_result));
264 tcf_exts_init(&new_filter_result.exts,
265 TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
267 tcindex_filter_result_init(&cr);
271 if (tb[TCA_TCINDEX_HASH])
272 cp->hash = nla_get_u32(tb[TCA_TCINDEX_HASH]);
274 if (tb[TCA_TCINDEX_MASK])
275 cp->mask = nla_get_u16(tb[TCA_TCINDEX_MASK]);
277 if (tb[TCA_TCINDEX_SHIFT])
278 cp->shift = nla_get_u32(tb[TCA_TCINDEX_SHIFT]);
282 /* Hash already allocated, make sure that we still meet the
283 * requirements for the allocated hash.
286 if (!valid_perfect_hash(cp) ||
287 cp->hash > cp->alloc_hash)
289 } else if (cp->h && cp->hash != cp->alloc_hash) {
294 if (tb[TCA_TCINDEX_FALL_THROUGH])
295 cp->fall_through = nla_get_u32(tb[TCA_TCINDEX_FALL_THROUGH]);
298 /* Hash not specified, use perfect hash if the upper limit
299 * of the hashing index is below the threshold.
301 if ((cp->mask >> cp->shift) < PERFECT_HASH_THRESHOLD)
302 cp->hash = (cp->mask >> cp->shift) + 1;
304 cp->hash = DEFAULT_HASH_SIZE;
307 if (!cp->perfect && cp->h)
308 cp->alloc_hash = cp->hash;
310 /* Note: this could be as restrictive as if (handle & ~(mask >> shift))
311 * but then, we'd fail handles that may become valid after some future
312 * mask change. While this is extremely unlikely to ever matter,
313 * the check below is safer (and also more backwards-compatible).
315 if (cp->perfect || valid_perfect_hash(cp))
316 if (handle >= cp->alloc_hash)
321 if (!cp->perfect && !cp->h) {
322 if (valid_perfect_hash(cp)) {
325 cp->perfect = kcalloc(cp->hash, sizeof(*r), GFP_KERNEL);
328 for (i = 0; i < cp->hash; i++)
329 tcf_exts_init(&cp->perfect[i].exts,
334 struct tcindex_filter __rcu **hash;
336 hash = kcalloc(cp->hash,
337 sizeof(struct tcindex_filter *),
349 r = cp->perfect + handle;
351 r = tcindex_lookup(cp, handle) ? : &new_filter_result;
353 if (r == &new_filter_result) {
354 f = kzalloc(sizeof(*f), GFP_KERNEL);
359 if (tb[TCA_TCINDEX_CLASSID]) {
360 cr.res.classid = nla_get_u32(tb[TCA_TCINDEX_CLASSID]);
361 tcf_bind_filter(tp, &cr.res, base);
365 tcf_exts_change(tp, &r->exts, &e);
367 tcf_exts_change(tp, &cr.exts, &e);
369 if (old_r && old_r != r)
370 tcindex_filter_result_init(old_r);
374 rcu_assign_pointer(tp->root, cp);
376 if (r == &new_filter_result) {
377 struct tcindex_filter *nfp;
378 struct tcindex_filter __rcu **fp;
381 f->result = new_filter_result;
384 fp = p->h + (handle % p->hash);
385 for (nfp = rtnl_dereference(*fp);
387 fp = &nfp->next, nfp = rtnl_dereference(*fp))
390 rcu_assign_pointer(*fp, f);
394 call_rcu(&oldp->rcu, __tcindex_partial_destroy);
400 else if (balloc == 2)
404 tcf_exts_destroy(tp, &e);
409 tcindex_change(struct net *net, struct sk_buff *in_skb,
410 struct tcf_proto *tp, unsigned long base, u32 handle,
411 struct nlattr **tca, unsigned long *arg, bool ovr)
413 struct nlattr *opt = tca[TCA_OPTIONS];
414 struct nlattr *tb[TCA_TCINDEX_MAX + 1];
415 struct tcindex_data *p = rtnl_dereference(tp->root);
416 struct tcindex_filter_result *r = (struct tcindex_filter_result *) *arg;
419 pr_debug("tcindex_change(tp %p,handle 0x%08x,tca %p,arg %p),opt %p,"
420 "p %p,r %p,*arg 0x%lx\n",
421 tp, handle, tca, arg, opt, p, r, arg ? *arg : 0L);
426 err = nla_parse_nested(tb, TCA_TCINDEX_MAX, opt, tcindex_policy);
430 return tcindex_set_parms(net, tp, base, handle, p, r, tb,
434 static void tcindex_walk(struct tcf_proto *tp, struct tcf_walker *walker)
436 struct tcindex_data *p = rtnl_dereference(tp->root);
437 struct tcindex_filter *f, *next;
440 pr_debug("tcindex_walk(tp %p,walker %p),p %p\n", tp, walker, p);
442 for (i = 0; i < p->hash; i++) {
443 if (!p->perfect[i].res.class)
445 if (walker->count >= walker->skip) {
447 (unsigned long) (p->perfect+i), walker)
458 for (i = 0; i < p->hash; i++) {
459 for (f = rtnl_dereference(p->h[i]); f; f = next) {
460 next = rtnl_dereference(f->next);
461 if (walker->count >= walker->skip) {
462 if (walker->fn(tp, (unsigned long) &f->result,
473 static void tcindex_destroy(struct tcf_proto *tp)
475 struct tcindex_data *p = rtnl_dereference(tp->root);
476 struct tcf_walker walker;
478 pr_debug("tcindex_destroy(tp %p),p %p\n", tp, p);
481 walker.fn = tcindex_destroy_element;
482 tcindex_walk(tp, &walker);
484 RCU_INIT_POINTER(tp->root, NULL);
485 call_rcu(&p->rcu, __tcindex_destroy);
489 static int tcindex_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
490 struct sk_buff *skb, struct tcmsg *t)
492 struct tcindex_data *p = rtnl_dereference(tp->root);
493 struct tcindex_filter_result *r = (struct tcindex_filter_result *) fh;
494 unsigned char *b = skb_tail_pointer(skb);
497 pr_debug("tcindex_dump(tp %p,fh 0x%lx,skb %p,t %p),p %p,r %p,b %p\n",
498 tp, fh, skb, t, p, r, b);
499 pr_debug("p->perfect %p p->h %p\n", p->perfect, p->h);
501 nest = nla_nest_start(skb, TCA_OPTIONS);
503 goto nla_put_failure;
506 t->tcm_handle = ~0; /* whatever ... */
507 if (nla_put_u32(skb, TCA_TCINDEX_HASH, p->hash) ||
508 nla_put_u16(skb, TCA_TCINDEX_MASK, p->mask) ||
509 nla_put_u32(skb, TCA_TCINDEX_SHIFT, p->shift) ||
510 nla_put_u32(skb, TCA_TCINDEX_FALL_THROUGH, p->fall_through))
511 goto nla_put_failure;
512 nla_nest_end(skb, nest);
515 t->tcm_handle = r - p->perfect;
517 struct tcindex_filter *f;
518 struct tcindex_filter __rcu **fp;
522 for (i = 0; !t->tcm_handle && i < p->hash; i++) {
524 for (f = rtnl_dereference(*fp);
526 fp = &f->next, f = rtnl_dereference(*fp)) {
528 t->tcm_handle = f->key;
532 pr_debug("handle = %d\n", t->tcm_handle);
534 nla_put_u32(skb, TCA_TCINDEX_CLASSID, r->res.classid))
535 goto nla_put_failure;
537 if (tcf_exts_dump(skb, &r->exts) < 0)
538 goto nla_put_failure;
539 nla_nest_end(skb, nest);
541 if (tcf_exts_dump_stats(skb, &r->exts) < 0)
542 goto nla_put_failure;
552 static struct tcf_proto_ops cls_tcindex_ops __read_mostly = {
554 .classify = tcindex_classify,
555 .init = tcindex_init,
556 .destroy = tcindex_destroy,
559 .change = tcindex_change,
560 .delete = tcindex_delete,
561 .walk = tcindex_walk,
562 .dump = tcindex_dump,
563 .owner = THIS_MODULE,
566 static int __init init_tcindex(void)
568 return register_tcf_proto_ops(&cls_tcindex_ops);
571 static void __exit exit_tcindex(void)
573 unregister_tcf_proto_ops(&cls_tcindex_ops);
576 module_init(init_tcindex)
577 module_exit(exit_tcindex)
578 MODULE_LICENSE("GPL");