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;
31 struct work_struct work;
36 struct tcindex_filter {
38 struct tcindex_filter_result result;
39 struct tcindex_filter __rcu *next;
41 struct work_struct work;
48 struct tcindex_filter_result *perfect; /* perfect hash; NULL if none */
49 struct tcindex_filter __rcu **h; /* imperfect hash; */
51 u16 mask; /* AND key with mask */
52 u32 shift; /* shift ANDed key to the right */
53 u32 hash; /* hash table size; 0 if undefined */
54 u32 alloc_hash; /* allocated size */
55 u32 fall_through; /* 0: only classify if explicit match */
59 static inline int tcindex_filter_is_set(struct tcindex_filter_result *r)
61 return tcf_exts_has_actions(&r->exts) || r->res.classid;
64 static struct tcindex_filter_result *tcindex_lookup(struct tcindex_data *p,
68 struct tcindex_filter_result *f = p->perfect + key;
70 return tcindex_filter_is_set(f) ? f : NULL;
72 struct tcindex_filter __rcu **fp;
73 struct tcindex_filter *f;
75 fp = &p->h[key % p->hash];
76 for (f = rcu_dereference_bh_rtnl(*fp);
78 fp = &f->next, f = rcu_dereference_bh_rtnl(*fp))
87 static int tcindex_classify(struct sk_buff *skb, const struct tcf_proto *tp,
88 struct tcf_result *res)
90 struct tcindex_data *p = rcu_dereference_bh(tp->root);
91 struct tcindex_filter_result *f;
92 int key = (skb->tc_index & p->mask) >> p->shift;
94 pr_debug("tcindex_classify(skb %p,tp %p,res %p),p %p\n",
97 f = tcindex_lookup(p, key);
101 res->classid = TC_H_MAKE(TC_H_MAJ(tp->q->handle), key);
103 pr_debug("alg 0x%x\n", res->classid);
107 pr_debug("map 0x%x\n", res->classid);
109 return tcf_exts_exec(skb, &f->exts, res);
113 static void *tcindex_get(struct tcf_proto *tp, u32 handle)
115 struct tcindex_data *p = rtnl_dereference(tp->root);
116 struct tcindex_filter_result *r;
118 pr_debug("tcindex_get(tp %p,handle 0x%08x)\n", tp, handle);
119 if (p->perfect && handle >= p->alloc_hash)
121 r = tcindex_lookup(p, handle);
122 return r && tcindex_filter_is_set(r) ? r : NULL;
125 static int tcindex_init(struct tcf_proto *tp)
127 struct tcindex_data *p;
129 pr_debug("tcindex_init(tp %p)\n", tp);
130 p = kzalloc(sizeof(struct tcindex_data), GFP_KERNEL);
135 p->hash = DEFAULT_HASH_SIZE;
138 rcu_assign_pointer(tp->root, p);
142 static void tcindex_destroy_rexts_work(struct work_struct *work)
144 struct tcindex_filter_result *r;
146 r = container_of(work, struct tcindex_filter_result, work);
148 tcf_exts_destroy(&r->exts);
152 static void tcindex_destroy_rexts(struct rcu_head *head)
154 struct tcindex_filter_result *r;
156 r = container_of(head, struct tcindex_filter_result, rcu);
157 INIT_WORK(&r->work, tcindex_destroy_rexts_work);
158 tcf_queue_work(&r->work);
161 static void tcindex_destroy_fexts_work(struct work_struct *work)
163 struct tcindex_filter *f = container_of(work, struct tcindex_filter,
167 tcf_exts_destroy(&f->result.exts);
172 static void tcindex_destroy_fexts(struct rcu_head *head)
174 struct tcindex_filter *f = container_of(head, struct tcindex_filter,
177 INIT_WORK(&f->work, tcindex_destroy_fexts_work);
178 tcf_queue_work(&f->work);
181 static int tcindex_delete(struct tcf_proto *tp, void *arg, bool *last)
183 struct tcindex_data *p = rtnl_dereference(tp->root);
184 struct tcindex_filter_result *r = arg;
185 struct tcindex_filter __rcu **walk;
186 struct tcindex_filter *f = NULL;
188 pr_debug("tcindex_delete(tp %p,arg %p),p %p\n", tp, arg, p);
195 for (i = 0; i < p->hash; i++) {
197 for (f = rtnl_dereference(*walk); f;
198 walk = &f->next, f = rtnl_dereference(*walk)) {
206 rcu_assign_pointer(*walk, rtnl_dereference(f->next));
208 tcf_unbind_filter(tp, &r->res);
209 /* all classifiers are required to call tcf_exts_destroy() after rcu
210 * grace period, since converted-to-rcu actions are relying on that
211 * in cleanup() callback
214 call_rcu(&f->rcu, tcindex_destroy_fexts);
216 call_rcu(&r->rcu, tcindex_destroy_rexts);
222 static int tcindex_destroy_element(struct tcf_proto *tp,
223 void *arg, struct tcf_walker *walker)
227 return tcindex_delete(tp, arg, &last);
230 static void __tcindex_destroy(struct rcu_head *head)
232 struct tcindex_data *p = container_of(head, struct tcindex_data, rcu);
240 valid_perfect_hash(struct tcindex_data *p)
242 return p->hash > (p->mask >> p->shift);
245 static const struct nla_policy tcindex_policy[TCA_TCINDEX_MAX + 1] = {
246 [TCA_TCINDEX_HASH] = { .type = NLA_U32 },
247 [TCA_TCINDEX_MASK] = { .type = NLA_U16 },
248 [TCA_TCINDEX_SHIFT] = { .type = NLA_U32 },
249 [TCA_TCINDEX_FALL_THROUGH] = { .type = NLA_U32 },
250 [TCA_TCINDEX_CLASSID] = { .type = NLA_U32 },
253 static int tcindex_filter_result_init(struct tcindex_filter_result *r)
255 memset(r, 0, sizeof(*r));
256 return tcf_exts_init(&r->exts, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
259 static void __tcindex_partial_destroy(struct rcu_head *head)
261 struct tcindex_data *p = container_of(head, struct tcindex_data, rcu);
267 static void tcindex_free_perfect_hash(struct tcindex_data *cp)
271 for (i = 0; i < cp->hash; i++)
272 tcf_exts_destroy(&cp->perfect[i].exts);
276 static int tcindex_alloc_perfect_hash(struct tcindex_data *cp)
280 cp->perfect = kcalloc(cp->hash, sizeof(struct tcindex_filter_result),
285 for (i = 0; i < cp->hash; i++) {
286 err = tcf_exts_init(&cp->perfect[i].exts,
287 TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
295 tcindex_free_perfect_hash(cp);
300 tcindex_set_parms(struct net *net, struct tcf_proto *tp, unsigned long base,
301 u32 handle, struct tcindex_data *p,
302 struct tcindex_filter_result *r, struct nlattr **tb,
303 struct nlattr *est, bool ovr)
305 struct tcindex_filter_result new_filter_result, *old_r = r;
306 struct tcindex_filter_result cr;
307 struct tcindex_data *cp = NULL, *oldp;
308 struct tcindex_filter *f = NULL; /* make gcc behave */
312 err = tcf_exts_init(&e, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
315 err = tcf_exts_validate(net, tp, tb, est, &e, ovr);
320 /* tcindex_data attributes must look atomic to classifier/lookup so
321 * allocate new tcindex data and RCU assign it onto root. Keeping
322 * perfect hash and hash pointers from old data.
324 cp = kzalloc(sizeof(*cp), GFP_KERNEL);
329 cp->shift = p->shift;
331 cp->alloc_hash = p->alloc_hash;
332 cp->fall_through = p->fall_through;
338 if (tcindex_alloc_perfect_hash(cp) < 0)
340 for (i = 0; i < cp->hash; i++)
341 cp->perfect[i].res = p->perfect[i].res;
346 err = tcindex_filter_result_init(&new_filter_result);
349 err = tcindex_filter_result_init(&cr);
355 if (tb[TCA_TCINDEX_HASH])
356 cp->hash = nla_get_u32(tb[TCA_TCINDEX_HASH]);
358 if (tb[TCA_TCINDEX_MASK])
359 cp->mask = nla_get_u16(tb[TCA_TCINDEX_MASK]);
361 if (tb[TCA_TCINDEX_SHIFT])
362 cp->shift = nla_get_u32(tb[TCA_TCINDEX_SHIFT]);
366 /* Hash already allocated, make sure that we still meet the
367 * requirements for the allocated hash.
370 if (!valid_perfect_hash(cp) ||
371 cp->hash > cp->alloc_hash)
373 } else if (cp->h && cp->hash != cp->alloc_hash) {
378 if (tb[TCA_TCINDEX_FALL_THROUGH])
379 cp->fall_through = nla_get_u32(tb[TCA_TCINDEX_FALL_THROUGH]);
382 /* Hash not specified, use perfect hash if the upper limit
383 * of the hashing index is below the threshold.
385 if ((cp->mask >> cp->shift) < PERFECT_HASH_THRESHOLD)
386 cp->hash = (cp->mask >> cp->shift) + 1;
388 cp->hash = DEFAULT_HASH_SIZE;
391 if (!cp->perfect && !cp->h)
392 cp->alloc_hash = cp->hash;
394 /* Note: this could be as restrictive as if (handle & ~(mask >> shift))
395 * but then, we'd fail handles that may become valid after some future
396 * mask change. While this is extremely unlikely to ever matter,
397 * the check below is safer (and also more backwards-compatible).
399 if (cp->perfect || valid_perfect_hash(cp))
400 if (handle >= cp->alloc_hash)
405 if (!cp->perfect && !cp->h) {
406 if (valid_perfect_hash(cp)) {
407 if (tcindex_alloc_perfect_hash(cp) < 0)
411 struct tcindex_filter __rcu **hash;
413 hash = kcalloc(cp->hash,
414 sizeof(struct tcindex_filter *),
426 r = cp->perfect + handle;
428 r = tcindex_lookup(cp, handle) ? : &new_filter_result;
430 if (r == &new_filter_result) {
431 f = kzalloc(sizeof(*f), GFP_KERNEL);
436 err = tcindex_filter_result_init(&f->result);
443 if (tb[TCA_TCINDEX_CLASSID]) {
444 cr.res.classid = nla_get_u32(tb[TCA_TCINDEX_CLASSID]);
445 tcf_bind_filter(tp, &cr.res, base);
449 tcf_exts_change(&r->exts, &e);
451 tcf_exts_change(&cr.exts, &e);
453 if (old_r && old_r != r) {
454 err = tcindex_filter_result_init(old_r);
463 rcu_assign_pointer(tp->root, cp);
465 if (r == &new_filter_result) {
466 struct tcindex_filter *nfp;
467 struct tcindex_filter __rcu **fp;
469 tcf_exts_change(&f->result.exts, &r->exts);
471 fp = cp->h + (handle % cp->hash);
472 for (nfp = rtnl_dereference(*fp);
474 fp = &nfp->next, nfp = rtnl_dereference(*fp))
477 rcu_assign_pointer(*fp, f);
481 call_rcu(&oldp->rcu, __tcindex_partial_destroy);
486 tcindex_free_perfect_hash(cp);
487 else if (balloc == 2)
490 tcf_exts_destroy(&cr.exts);
491 tcf_exts_destroy(&new_filter_result.exts);
494 tcf_exts_destroy(&e);
499 tcindex_change(struct net *net, struct sk_buff *in_skb,
500 struct tcf_proto *tp, unsigned long base, u32 handle,
501 struct nlattr **tca, void **arg, bool ovr)
503 struct nlattr *opt = tca[TCA_OPTIONS];
504 struct nlattr *tb[TCA_TCINDEX_MAX + 1];
505 struct tcindex_data *p = rtnl_dereference(tp->root);
506 struct tcindex_filter_result *r = *arg;
509 pr_debug("tcindex_change(tp %p,handle 0x%08x,tca %p,arg %p),opt %p,"
510 "p %p,r %p,*arg %p\n",
511 tp, handle, tca, arg, opt, p, r, arg ? *arg : NULL);
516 err = nla_parse_nested(tb, TCA_TCINDEX_MAX, opt, tcindex_policy, NULL);
520 return tcindex_set_parms(net, tp, base, handle, p, r, tb,
524 static void tcindex_walk(struct tcf_proto *tp, struct tcf_walker *walker)
526 struct tcindex_data *p = rtnl_dereference(tp->root);
527 struct tcindex_filter *f, *next;
530 pr_debug("tcindex_walk(tp %p,walker %p),p %p\n", tp, walker, p);
532 for (i = 0; i < p->hash; i++) {
533 if (!p->perfect[i].res.class)
535 if (walker->count >= walker->skip) {
536 if (walker->fn(tp, p->perfect + i, walker) < 0) {
546 for (i = 0; i < p->hash; i++) {
547 for (f = rtnl_dereference(p->h[i]); f; f = next) {
548 next = rtnl_dereference(f->next);
549 if (walker->count >= walker->skip) {
550 if (walker->fn(tp, &f->result, walker) < 0) {
560 static void tcindex_destroy(struct tcf_proto *tp)
562 struct tcindex_data *p = rtnl_dereference(tp->root);
563 struct tcf_walker walker;
565 pr_debug("tcindex_destroy(tp %p),p %p\n", tp, p);
568 walker.fn = tcindex_destroy_element;
569 tcindex_walk(tp, &walker);
571 call_rcu(&p->rcu, __tcindex_destroy);
575 static int tcindex_dump(struct net *net, struct tcf_proto *tp, void *fh,
576 struct sk_buff *skb, struct tcmsg *t)
578 struct tcindex_data *p = rtnl_dereference(tp->root);
579 struct tcindex_filter_result *r = fh;
582 pr_debug("tcindex_dump(tp %p,fh %p,skb %p,t %p),p %p,r %p\n",
583 tp, fh, skb, t, p, r);
584 pr_debug("p->perfect %p p->h %p\n", p->perfect, p->h);
586 nest = nla_nest_start(skb, TCA_OPTIONS);
588 goto nla_put_failure;
591 t->tcm_handle = ~0; /* whatever ... */
592 if (nla_put_u32(skb, TCA_TCINDEX_HASH, p->hash) ||
593 nla_put_u16(skb, TCA_TCINDEX_MASK, p->mask) ||
594 nla_put_u32(skb, TCA_TCINDEX_SHIFT, p->shift) ||
595 nla_put_u32(skb, TCA_TCINDEX_FALL_THROUGH, p->fall_through))
596 goto nla_put_failure;
597 nla_nest_end(skb, nest);
600 t->tcm_handle = r - p->perfect;
602 struct tcindex_filter *f;
603 struct tcindex_filter __rcu **fp;
607 for (i = 0; !t->tcm_handle && i < p->hash; i++) {
609 for (f = rtnl_dereference(*fp);
611 fp = &f->next, f = rtnl_dereference(*fp)) {
613 t->tcm_handle = f->key;
617 pr_debug("handle = %d\n", t->tcm_handle);
619 nla_put_u32(skb, TCA_TCINDEX_CLASSID, r->res.classid))
620 goto nla_put_failure;
622 if (tcf_exts_dump(skb, &r->exts) < 0)
623 goto nla_put_failure;
624 nla_nest_end(skb, nest);
626 if (tcf_exts_dump_stats(skb, &r->exts) < 0)
627 goto nla_put_failure;
633 nla_nest_cancel(skb, nest);
637 static void tcindex_bind_class(void *fh, u32 classid, unsigned long cl)
639 struct tcindex_filter_result *r = fh;
641 if (r && r->res.classid == classid)
645 static struct tcf_proto_ops cls_tcindex_ops __read_mostly = {
647 .classify = tcindex_classify,
648 .init = tcindex_init,
649 .destroy = tcindex_destroy,
651 .change = tcindex_change,
652 .delete = tcindex_delete,
653 .walk = tcindex_walk,
654 .dump = tcindex_dump,
655 .bind_class = tcindex_bind_class,
656 .owner = THIS_MODULE,
659 static int __init init_tcindex(void)
661 return register_tcf_proto_ops(&cls_tcindex_ops);
664 static void __exit exit_tcindex(void)
666 unregister_tcf_proto_ops(&cls_tcindex_ops);
669 module_init(init_tcindex)
670 module_exit(exit_tcindex)
671 MODULE_LICENSE("GPL");