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;
33 struct tcindex_filter {
35 struct tcindex_filter_result result;
36 struct tcindex_filter __rcu *next;
42 struct tcindex_filter_result *perfect; /* perfect hash; NULL if none */
43 struct tcindex_filter __rcu **h; /* imperfect hash; */
45 u16 mask; /* AND key with mask */
46 u32 shift; /* shift ANDed key to the right */
47 u32 hash; /* hash table size; 0 if undefined */
48 u32 alloc_hash; /* allocated size */
49 u32 fall_through; /* 0: only classify if explicit match */
53 static inline int tcindex_filter_is_set(struct tcindex_filter_result *r)
55 return tcf_exts_has_actions(&r->exts) || r->res.classid;
58 static struct tcindex_filter_result *tcindex_lookup(struct tcindex_data *p,
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_bh(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 void *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) ? r : NULL;
119 static int tcindex_init(struct tcf_proto *tp)
121 struct tcindex_data *p;
123 pr_debug("tcindex_init(tp %p)\n", tp);
124 p = kzalloc(sizeof(struct tcindex_data), GFP_KERNEL);
129 p->hash = DEFAULT_HASH_SIZE;
132 rcu_assign_pointer(tp->root, p);
136 static void tcindex_destroy_rexts(struct rcu_head *head)
138 struct tcindex_filter_result *r;
140 r = container_of(head, struct tcindex_filter_result, rcu);
141 tcf_exts_destroy(&r->exts);
144 static void tcindex_destroy_fexts(struct rcu_head *head)
146 struct tcindex_filter *f = container_of(head, struct tcindex_filter,
149 tcf_exts_destroy(&f->result.exts);
153 static int tcindex_delete(struct tcf_proto *tp, void *arg, bool *last)
155 struct tcindex_data *p = rtnl_dereference(tp->root);
156 struct tcindex_filter_result *r = arg;
157 struct tcindex_filter __rcu **walk;
158 struct tcindex_filter *f = NULL;
160 pr_debug("tcindex_delete(tp %p,arg %p),p %p\n", tp, arg, p);
167 for (i = 0; i < p->hash; i++) {
169 for (f = rtnl_dereference(*walk); f;
170 walk = &f->next, f = rtnl_dereference(*walk)) {
178 rcu_assign_pointer(*walk, rtnl_dereference(f->next));
180 tcf_unbind_filter(tp, &r->res);
181 /* all classifiers are required to call tcf_exts_destroy() after rcu
182 * grace period, since converted-to-rcu actions are relying on that
183 * in cleanup() callback
186 call_rcu(&f->rcu, tcindex_destroy_fexts);
188 call_rcu(&r->rcu, tcindex_destroy_rexts);
194 static int tcindex_destroy_element(struct tcf_proto *tp,
195 void *arg, struct tcf_walker *walker)
199 return tcindex_delete(tp, arg, &last);
202 static void __tcindex_destroy(struct rcu_head *head)
204 struct tcindex_data *p = container_of(head, struct tcindex_data, rcu);
212 valid_perfect_hash(struct tcindex_data *p)
214 return p->hash > (p->mask >> p->shift);
217 static const struct nla_policy tcindex_policy[TCA_TCINDEX_MAX + 1] = {
218 [TCA_TCINDEX_HASH] = { .type = NLA_U32 },
219 [TCA_TCINDEX_MASK] = { .type = NLA_U16 },
220 [TCA_TCINDEX_SHIFT] = { .type = NLA_U32 },
221 [TCA_TCINDEX_FALL_THROUGH] = { .type = NLA_U32 },
222 [TCA_TCINDEX_CLASSID] = { .type = NLA_U32 },
225 static int tcindex_filter_result_init(struct tcindex_filter_result *r)
227 memset(r, 0, sizeof(*r));
228 return tcf_exts_init(&r->exts, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
231 static void __tcindex_partial_destroy(struct rcu_head *head)
233 struct tcindex_data *p = container_of(head, struct tcindex_data, rcu);
239 static void tcindex_free_perfect_hash(struct tcindex_data *cp)
243 for (i = 0; i < cp->hash; i++)
244 tcf_exts_destroy(&cp->perfect[i].exts);
248 static int tcindex_alloc_perfect_hash(struct tcindex_data *cp)
252 cp->perfect = kcalloc(cp->hash, sizeof(struct tcindex_filter_result),
257 for (i = 0; i < cp->hash; i++) {
258 err = tcf_exts_init(&cp->perfect[i].exts,
259 TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
267 tcindex_free_perfect_hash(cp);
272 tcindex_set_parms(struct net *net, struct tcf_proto *tp, unsigned long base,
273 u32 handle, struct tcindex_data *p,
274 struct tcindex_filter_result *r, struct nlattr **tb,
275 struct nlattr *est, bool ovr)
277 struct tcindex_filter_result new_filter_result, *old_r = r;
278 struct tcindex_filter_result cr;
279 struct tcindex_data *cp = NULL, *oldp;
280 struct tcindex_filter *f = NULL; /* make gcc behave */
284 err = tcf_exts_init(&e, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
287 err = tcf_exts_validate(net, tp, tb, est, &e, ovr);
292 /* tcindex_data attributes must look atomic to classifier/lookup so
293 * allocate new tcindex data and RCU assign it onto root. Keeping
294 * perfect hash and hash pointers from old data.
296 cp = kzalloc(sizeof(*cp), GFP_KERNEL);
301 cp->shift = p->shift;
303 cp->alloc_hash = p->alloc_hash;
304 cp->fall_through = p->fall_through;
310 if (tcindex_alloc_perfect_hash(cp) < 0)
312 for (i = 0; i < cp->hash; i++)
313 cp->perfect[i].res = p->perfect[i].res;
318 err = tcindex_filter_result_init(&new_filter_result);
321 err = tcindex_filter_result_init(&cr);
327 if (tb[TCA_TCINDEX_HASH])
328 cp->hash = nla_get_u32(tb[TCA_TCINDEX_HASH]);
330 if (tb[TCA_TCINDEX_MASK])
331 cp->mask = nla_get_u16(tb[TCA_TCINDEX_MASK]);
333 if (tb[TCA_TCINDEX_SHIFT])
334 cp->shift = nla_get_u32(tb[TCA_TCINDEX_SHIFT]);
338 /* Hash already allocated, make sure that we still meet the
339 * requirements for the allocated hash.
342 if (!valid_perfect_hash(cp) ||
343 cp->hash > cp->alloc_hash)
345 } else if (cp->h && cp->hash != cp->alloc_hash) {
350 if (tb[TCA_TCINDEX_FALL_THROUGH])
351 cp->fall_through = nla_get_u32(tb[TCA_TCINDEX_FALL_THROUGH]);
354 /* Hash not specified, use perfect hash if the upper limit
355 * of the hashing index is below the threshold.
357 if ((cp->mask >> cp->shift) < PERFECT_HASH_THRESHOLD)
358 cp->hash = (cp->mask >> cp->shift) + 1;
360 cp->hash = DEFAULT_HASH_SIZE;
363 if (!cp->perfect && !cp->h)
364 cp->alloc_hash = cp->hash;
366 /* Note: this could be as restrictive as if (handle & ~(mask >> shift))
367 * but then, we'd fail handles that may become valid after some future
368 * mask change. While this is extremely unlikely to ever matter,
369 * the check below is safer (and also more backwards-compatible).
371 if (cp->perfect || valid_perfect_hash(cp))
372 if (handle >= cp->alloc_hash)
377 if (!cp->perfect && !cp->h) {
378 if (valid_perfect_hash(cp)) {
379 if (tcindex_alloc_perfect_hash(cp) < 0)
383 struct tcindex_filter __rcu **hash;
385 hash = kcalloc(cp->hash,
386 sizeof(struct tcindex_filter *),
398 r = cp->perfect + handle;
400 r = tcindex_lookup(cp, handle) ? : &new_filter_result;
402 if (r == &new_filter_result) {
403 f = kzalloc(sizeof(*f), GFP_KERNEL);
408 err = tcindex_filter_result_init(&f->result);
415 if (tb[TCA_TCINDEX_CLASSID]) {
416 cr.res.classid = nla_get_u32(tb[TCA_TCINDEX_CLASSID]);
417 tcf_bind_filter(tp, &cr.res, base);
421 tcf_exts_change(&r->exts, &e);
423 tcf_exts_change(&cr.exts, &e);
425 if (old_r && old_r != r) {
426 err = tcindex_filter_result_init(old_r);
435 rcu_assign_pointer(tp->root, cp);
437 if (r == &new_filter_result) {
438 struct tcindex_filter *nfp;
439 struct tcindex_filter __rcu **fp;
441 tcf_exts_change(&f->result.exts, &r->exts);
443 fp = cp->h + (handle % cp->hash);
444 for (nfp = rtnl_dereference(*fp);
446 fp = &nfp->next, nfp = rtnl_dereference(*fp))
449 rcu_assign_pointer(*fp, f);
453 call_rcu(&oldp->rcu, __tcindex_partial_destroy);
458 tcindex_free_perfect_hash(cp);
459 else if (balloc == 2)
462 tcf_exts_destroy(&cr.exts);
463 tcf_exts_destroy(&new_filter_result.exts);
466 tcf_exts_destroy(&e);
471 tcindex_change(struct net *net, struct sk_buff *in_skb,
472 struct tcf_proto *tp, unsigned long base, u32 handle,
473 struct nlattr **tca, void **arg, bool ovr)
475 struct nlattr *opt = tca[TCA_OPTIONS];
476 struct nlattr *tb[TCA_TCINDEX_MAX + 1];
477 struct tcindex_data *p = rtnl_dereference(tp->root);
478 struct tcindex_filter_result *r = *arg;
481 pr_debug("tcindex_change(tp %p,handle 0x%08x,tca %p,arg %p),opt %p,"
482 "p %p,r %p,*arg %p\n",
483 tp, handle, tca, arg, opt, p, r, arg ? *arg : NULL);
488 err = nla_parse_nested(tb, TCA_TCINDEX_MAX, opt, tcindex_policy, NULL);
492 return tcindex_set_parms(net, tp, base, handle, p, r, tb,
496 static void tcindex_walk(struct tcf_proto *tp, struct tcf_walker *walker)
498 struct tcindex_data *p = rtnl_dereference(tp->root);
499 struct tcindex_filter *f, *next;
502 pr_debug("tcindex_walk(tp %p,walker %p),p %p\n", tp, walker, p);
504 for (i = 0; i < p->hash; i++) {
505 if (!p->perfect[i].res.class)
507 if (walker->count >= walker->skip) {
508 if (walker->fn(tp, p->perfect + i, walker) < 0) {
518 for (i = 0; i < p->hash; i++) {
519 for (f = rtnl_dereference(p->h[i]); f; f = next) {
520 next = rtnl_dereference(f->next);
521 if (walker->count >= walker->skip) {
522 if (walker->fn(tp, &f->result, walker) < 0) {
532 static void tcindex_destroy(struct tcf_proto *tp)
534 struct tcindex_data *p = rtnl_dereference(tp->root);
535 struct tcf_walker walker;
537 pr_debug("tcindex_destroy(tp %p),p %p\n", tp, p);
540 walker.fn = tcindex_destroy_element;
541 tcindex_walk(tp, &walker);
543 call_rcu(&p->rcu, __tcindex_destroy);
547 static int tcindex_dump(struct net *net, struct tcf_proto *tp, void *fh,
548 struct sk_buff *skb, struct tcmsg *t)
550 struct tcindex_data *p = rtnl_dereference(tp->root);
551 struct tcindex_filter_result *r = fh;
554 pr_debug("tcindex_dump(tp %p,fh %p,skb %p,t %p),p %p,r %p\n",
555 tp, fh, skb, t, p, r);
556 pr_debug("p->perfect %p p->h %p\n", p->perfect, p->h);
558 nest = nla_nest_start(skb, TCA_OPTIONS);
560 goto nla_put_failure;
563 t->tcm_handle = ~0; /* whatever ... */
564 if (nla_put_u32(skb, TCA_TCINDEX_HASH, p->hash) ||
565 nla_put_u16(skb, TCA_TCINDEX_MASK, p->mask) ||
566 nla_put_u32(skb, TCA_TCINDEX_SHIFT, p->shift) ||
567 nla_put_u32(skb, TCA_TCINDEX_FALL_THROUGH, p->fall_through))
568 goto nla_put_failure;
569 nla_nest_end(skb, nest);
572 t->tcm_handle = r - p->perfect;
574 struct tcindex_filter *f;
575 struct tcindex_filter __rcu **fp;
579 for (i = 0; !t->tcm_handle && i < p->hash; i++) {
581 for (f = rtnl_dereference(*fp);
583 fp = &f->next, f = rtnl_dereference(*fp)) {
585 t->tcm_handle = f->key;
589 pr_debug("handle = %d\n", t->tcm_handle);
591 nla_put_u32(skb, TCA_TCINDEX_CLASSID, r->res.classid))
592 goto nla_put_failure;
594 if (tcf_exts_dump(skb, &r->exts) < 0)
595 goto nla_put_failure;
596 nla_nest_end(skb, nest);
598 if (tcf_exts_dump_stats(skb, &r->exts) < 0)
599 goto nla_put_failure;
605 nla_nest_cancel(skb, nest);
609 static void tcindex_bind_class(void *fh, u32 classid, unsigned long cl)
611 struct tcindex_filter_result *r = fh;
613 if (r && r->res.classid == classid)
617 static struct tcf_proto_ops cls_tcindex_ops __read_mostly = {
619 .classify = tcindex_classify,
620 .init = tcindex_init,
621 .destroy = tcindex_destroy,
623 .change = tcindex_change,
624 .delete = tcindex_delete,
625 .walk = tcindex_walk,
626 .dump = tcindex_dump,
627 .bind_class = tcindex_bind_class,
628 .owner = THIS_MODULE,
631 static int __init init_tcindex(void)
633 return register_tcf_proto_ops(&cls_tcindex_ops);
636 static void __exit exit_tcindex(void)
638 unregister_tcf_proto_ops(&cls_tcindex_ops);
641 module_init(init_tcindex)
642 module_exit(exit_tcindex)
643 MODULE_LICENSE("GPL");