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 *next;
40 struct tcindex_filter_result *perfect; /* perfect hash; NULL if none */
41 struct tcindex_filter **h; /* imperfect hash; only used if !perfect;
43 u16 mask; /* AND key with mask */
44 int shift; /* shift ANDed key to the right */
45 int hash; /* hash table size; 0 if undefined */
46 int alloc_hash; /* allocated size */
47 int fall_through; /* 0: only classify if explicit match */
51 tcindex_filter_is_set(struct tcindex_filter_result *r)
53 return tcf_exts_is_predicative(&r->exts) || r->res.classid;
56 static struct tcindex_filter_result *
57 tcindex_lookup(struct tcindex_data *p, u16 key)
59 struct tcindex_filter *f;
62 return tcindex_filter_is_set(p->perfect + key) ?
63 p->perfect + key : NULL;
65 for (f = p->h[key % p->hash]; f; f = f->next)
74 static int tcindex_classify(struct sk_buff *skb, const struct tcf_proto *tp,
75 struct tcf_result *res)
77 struct tcindex_data *p = tp->root;
78 struct tcindex_filter_result *f;
79 int key = (skb->tc_index & p->mask) >> p->shift;
81 pr_debug("tcindex_classify(skb %p,tp %p,res %p),p %p\n",
84 f = tcindex_lookup(p, key);
88 res->classid = TC_H_MAKE(TC_H_MAJ(tp->q->handle), key);
90 pr_debug("alg 0x%x\n", res->classid);
94 pr_debug("map 0x%x\n", res->classid);
96 return tcf_exts_exec(skb, &f->exts, res);
100 static unsigned long tcindex_get(struct tcf_proto *tp, u32 handle)
102 struct tcindex_data *p = tp->root;
103 struct tcindex_filter_result *r;
105 pr_debug("tcindex_get(tp %p,handle 0x%08x)\n", tp, handle);
106 if (p->perfect && handle >= p->alloc_hash)
108 r = tcindex_lookup(p, handle);
109 return r && tcindex_filter_is_set(r) ? (unsigned long) r : 0UL;
113 static void tcindex_put(struct tcf_proto *tp, unsigned long f)
115 pr_debug("tcindex_put(tp %p,f 0x%lx)\n", tp, f);
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;
138 __tcindex_delete(struct tcf_proto *tp, unsigned long arg, int lock)
140 struct tcindex_data *p = tp->root;
141 struct tcindex_filter_result *r = (struct tcindex_filter_result *) arg;
142 struct tcindex_filter *f = NULL;
144 pr_debug("tcindex_delete(tp %p,arg 0x%lx),p %p,f %p\n", tp, arg, p, f);
150 struct tcindex_filter **walk = NULL;
152 for (i = 0; i < p->hash; i++)
153 for (walk = p->h+i; *walk; walk = &(*walk)->next)
154 if (&(*walk)->result == r)
166 tcf_unbind_filter(tp, &r->res);
167 tcf_exts_destroy(tp, &r->exts);
172 static int tcindex_delete(struct tcf_proto *tp, unsigned long arg)
174 return __tcindex_delete(tp, arg, 1);
178 valid_perfect_hash(struct tcindex_data *p)
180 return p->hash > (p->mask >> p->shift);
183 static const struct nla_policy tcindex_policy[TCA_TCINDEX_MAX + 1] = {
184 [TCA_TCINDEX_HASH] = { .type = NLA_U32 },
185 [TCA_TCINDEX_MASK] = { .type = NLA_U16 },
186 [TCA_TCINDEX_SHIFT] = { .type = NLA_U32 },
187 [TCA_TCINDEX_FALL_THROUGH] = { .type = NLA_U32 },
188 [TCA_TCINDEX_CLASSID] = { .type = NLA_U32 },
191 static void tcindex_filter_result_init(struct tcindex_filter_result *r)
193 memset(r, 0, sizeof(*r));
194 tcf_exts_init(&r->exts, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
198 tcindex_set_parms(struct net *net, struct tcf_proto *tp, unsigned long base,
199 u32 handle, struct tcindex_data *p,
200 struct tcindex_filter_result *r, struct nlattr **tb,
201 struct nlattr *est, bool ovr)
204 struct tcindex_filter_result new_filter_result, *old_r = r;
205 struct tcindex_filter_result cr;
206 struct tcindex_data cp;
207 struct tcindex_filter *f = NULL; /* make gcc behave */
210 tcf_exts_init(&e, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
211 err = tcf_exts_validate(net, tp, tb, est, &e, ovr);
215 memcpy(&cp, p, sizeof(cp));
216 tcindex_filter_result_init(&new_filter_result);
218 tcindex_filter_result_init(&cr);
222 if (tb[TCA_TCINDEX_HASH])
223 cp.hash = nla_get_u32(tb[TCA_TCINDEX_HASH]);
225 if (tb[TCA_TCINDEX_MASK])
226 cp.mask = nla_get_u16(tb[TCA_TCINDEX_MASK]);
228 if (tb[TCA_TCINDEX_SHIFT])
229 cp.shift = nla_get_u32(tb[TCA_TCINDEX_SHIFT]);
232 /* Hash already allocated, make sure that we still meet the
233 * requirements for the allocated hash.
236 if (!valid_perfect_hash(&cp) ||
237 cp.hash > cp.alloc_hash)
239 } else if (cp.h && cp.hash != cp.alloc_hash)
243 if (tb[TCA_TCINDEX_FALL_THROUGH])
244 cp.fall_through = nla_get_u32(tb[TCA_TCINDEX_FALL_THROUGH]);
247 /* Hash not specified, use perfect hash if the upper limit
248 * of the hashing index is below the threshold.
250 if ((cp.mask >> cp.shift) < PERFECT_HASH_THRESHOLD)
251 cp.hash = (cp.mask >> cp.shift) + 1;
253 cp.hash = DEFAULT_HASH_SIZE;
256 if (!cp.perfect && !cp.h)
257 cp.alloc_hash = cp.hash;
259 /* Note: this could be as restrictive as if (handle & ~(mask >> shift))
260 * but then, we'd fail handles that may become valid after some future
261 * mask change. While this is extremely unlikely to ever matter,
262 * the check below is safer (and also more backwards-compatible).
264 if (cp.perfect || valid_perfect_hash(&cp))
265 if (handle >= cp.alloc_hash)
270 if (!cp.perfect && !cp.h) {
271 if (valid_perfect_hash(&cp)) {
274 cp.perfect = kcalloc(cp.hash, sizeof(*r), GFP_KERNEL);
277 for (i = 0; i < cp.hash; i++)
278 tcf_exts_init(&cp.perfect[i].exts, TCA_TCINDEX_ACT,
282 cp.h = kcalloc(cp.hash, sizeof(f), GFP_KERNEL);
290 r = cp.perfect + handle;
292 r = tcindex_lookup(&cp, handle) ? : &new_filter_result;
294 if (r == &new_filter_result) {
295 f = kzalloc(sizeof(*f), GFP_KERNEL);
300 if (tb[TCA_TCINDEX_CLASSID]) {
301 cr.res.classid = nla_get_u32(tb[TCA_TCINDEX_CLASSID]);
302 tcf_bind_filter(tp, &cr.res, base);
306 tcf_exts_change(tp, &r->exts, &e);
308 tcf_exts_change(tp, &cr.exts, &e);
311 if (old_r && old_r != r)
312 tcindex_filter_result_init(old_r);
314 memcpy(p, &cp, sizeof(cp));
317 if (r == &new_filter_result) {
318 struct tcindex_filter **fp;
321 f->result = new_filter_result;
323 for (fp = p->h+(handle % p->hash); *fp; fp = &(*fp)->next)
334 else if (balloc == 2)
337 tcf_exts_destroy(tp, &e);
342 tcindex_change(struct net *net, struct sk_buff *in_skb,
343 struct tcf_proto *tp, unsigned long base, u32 handle,
344 struct nlattr **tca, unsigned long *arg, bool ovr)
346 struct nlattr *opt = tca[TCA_OPTIONS];
347 struct nlattr *tb[TCA_TCINDEX_MAX + 1];
348 struct tcindex_data *p = tp->root;
349 struct tcindex_filter_result *r = (struct tcindex_filter_result *) *arg;
352 pr_debug("tcindex_change(tp %p,handle 0x%08x,tca %p,arg %p),opt %p,"
353 "p %p,r %p,*arg 0x%lx\n",
354 tp, handle, tca, arg, opt, p, r, arg ? *arg : 0L);
359 err = nla_parse_nested(tb, TCA_TCINDEX_MAX, opt, tcindex_policy);
363 return tcindex_set_parms(net, tp, base, handle, p, r, tb,
368 static void tcindex_walk(struct tcf_proto *tp, struct tcf_walker *walker)
370 struct tcindex_data *p = tp->root;
371 struct tcindex_filter *f, *next;
374 pr_debug("tcindex_walk(tp %p,walker %p),p %p\n", tp, walker, p);
376 for (i = 0; i < p->hash; i++) {
377 if (!p->perfect[i].res.class)
379 if (walker->count >= walker->skip) {
381 (unsigned long) (p->perfect+i), walker)
392 for (i = 0; i < p->hash; i++) {
393 for (f = p->h[i]; f; f = next) {
395 if (walker->count >= walker->skip) {
396 if (walker->fn(tp, (unsigned long) &f->result,
408 static int tcindex_destroy_element(struct tcf_proto *tp,
409 unsigned long arg, struct tcf_walker *walker)
411 return __tcindex_delete(tp, arg, 0);
415 static void tcindex_destroy(struct tcf_proto *tp)
417 struct tcindex_data *p = tp->root;
418 struct tcf_walker walker;
420 pr_debug("tcindex_destroy(tp %p),p %p\n", tp, p);
423 walker.fn = tcindex_destroy_element;
424 tcindex_walk(tp, &walker);
432 static int tcindex_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
433 struct sk_buff *skb, struct tcmsg *t)
435 struct tcindex_data *p = tp->root;
436 struct tcindex_filter_result *r = (struct tcindex_filter_result *) fh;
437 unsigned char *b = skb_tail_pointer(skb);
440 pr_debug("tcindex_dump(tp %p,fh 0x%lx,skb %p,t %p),p %p,r %p,b %p\n",
441 tp, fh, skb, t, p, r, b);
442 pr_debug("p->perfect %p p->h %p\n", p->perfect, p->h);
444 nest = nla_nest_start(skb, TCA_OPTIONS);
446 goto nla_put_failure;
449 t->tcm_handle = ~0; /* whatever ... */
450 if (nla_put_u32(skb, TCA_TCINDEX_HASH, p->hash) ||
451 nla_put_u16(skb, TCA_TCINDEX_MASK, p->mask) ||
452 nla_put_u32(skb, TCA_TCINDEX_SHIFT, p->shift) ||
453 nla_put_u32(skb, TCA_TCINDEX_FALL_THROUGH, p->fall_through))
454 goto nla_put_failure;
455 nla_nest_end(skb, nest);
458 t->tcm_handle = r-p->perfect;
460 struct tcindex_filter *f;
464 for (i = 0; !t->tcm_handle && i < p->hash; i++) {
465 for (f = p->h[i]; !t->tcm_handle && f;
468 t->tcm_handle = f->key;
472 pr_debug("handle = %d\n", t->tcm_handle);
474 nla_put_u32(skb, TCA_TCINDEX_CLASSID, r->res.classid))
475 goto nla_put_failure;
477 if (tcf_exts_dump(skb, &r->exts) < 0)
478 goto nla_put_failure;
479 nla_nest_end(skb, nest);
481 if (tcf_exts_dump_stats(skb, &r->exts) < 0)
482 goto nla_put_failure;
492 static struct tcf_proto_ops cls_tcindex_ops __read_mostly = {
494 .classify = tcindex_classify,
495 .init = tcindex_init,
496 .destroy = tcindex_destroy,
499 .change = tcindex_change,
500 .delete = tcindex_delete,
501 .walk = tcindex_walk,
502 .dump = tcindex_dump,
503 .owner = THIS_MODULE,
506 static int __init init_tcindex(void)
508 return register_tcf_proto_ops(&cls_tcindex_ops);
511 static void __exit exit_tcindex(void)
513 unregister_tcf_proto_ops(&cls_tcindex_ops);
516 module_init(init_tcindex)
517 module_exit(exit_tcindex)
518 MODULE_LICENSE("GPL");