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 #define PRIV(tp) ((struct tcindex_data *) (tp)->root)
30 struct tcindex_filter_result {
32 struct tcf_result res;
35 struct tcindex_filter {
37 struct tcindex_filter_result result;
38 struct tcindex_filter *next;
43 struct tcindex_filter_result *perfect; /* perfect hash; NULL if none */
44 struct tcindex_filter **h; /* imperfect hash; only used if !perfect;
46 u16 mask; /* AND key with mask */
47 int shift; /* shift ANDed key to the right */
48 int hash; /* hash table size; 0 if undefined */
49 int alloc_hash; /* allocated size */
50 int fall_through; /* 0: only classify if explicit match */
54 tcindex_filter_is_set(struct tcindex_filter_result *r)
56 return tcf_exts_is_predicative(&r->exts) || r->res.classid;
59 static struct tcindex_filter_result *
60 tcindex_lookup(struct tcindex_data *p, u16 key)
62 struct tcindex_filter *f;
65 return tcindex_filter_is_set(p->perfect + key) ?
66 p->perfect + key : NULL;
68 for (f = p->h[key % p->hash]; f; f = f->next)
77 static int tcindex_classify(struct sk_buff *skb, const struct tcf_proto *tp,
78 struct tcf_result *res)
80 struct tcindex_data *p = PRIV(tp);
81 struct tcindex_filter_result *f;
82 int key = (skb->tc_index & p->mask) >> p->shift;
84 pr_debug("tcindex_classify(skb %p,tp %p,res %p),p %p\n",
87 f = tcindex_lookup(p, key);
91 res->classid = TC_H_MAKE(TC_H_MAJ(tp->q->handle), key);
93 pr_debug("alg 0x%x\n", res->classid);
97 pr_debug("map 0x%x\n", res->classid);
99 return tcf_exts_exec(skb, &f->exts, res);
103 static unsigned long tcindex_get(struct tcf_proto *tp, u32 handle)
105 struct tcindex_data *p = PRIV(tp);
106 struct tcindex_filter_result *r;
108 pr_debug("tcindex_get(tp %p,handle 0x%08x)\n", tp, handle);
109 if (p->perfect && handle >= p->alloc_hash)
111 r = tcindex_lookup(p, handle);
112 return r && tcindex_filter_is_set(r) ? (unsigned long) r : 0UL;
116 static void tcindex_put(struct tcf_proto *tp, unsigned long f)
118 pr_debug("tcindex_put(tp %p,f 0x%lx)\n", tp, f);
122 static int tcindex_init(struct tcf_proto *tp)
124 struct tcindex_data *p;
126 pr_debug("tcindex_init(tp %p)\n", tp);
127 p = kzalloc(sizeof(struct tcindex_data), GFP_KERNEL);
132 p->hash = DEFAULT_HASH_SIZE;
141 __tcindex_delete(struct tcf_proto *tp, unsigned long arg, int lock)
143 struct tcindex_data *p = PRIV(tp);
144 struct tcindex_filter_result *r = (struct tcindex_filter_result *) arg;
145 struct tcindex_filter *f = NULL;
147 pr_debug("tcindex_delete(tp %p,arg 0x%lx),p %p,f %p\n", tp, arg, p, f);
153 struct tcindex_filter **walk = NULL;
155 for (i = 0; i < p->hash; i++)
156 for (walk = p->h+i; *walk; walk = &(*walk)->next)
157 if (&(*walk)->result == r)
169 tcf_unbind_filter(tp, &r->res);
170 tcf_exts_destroy(tp, &r->exts);
175 static int tcindex_delete(struct tcf_proto *tp, unsigned long arg)
177 return __tcindex_delete(tp, arg, 1);
181 valid_perfect_hash(struct tcindex_data *p)
183 return p->hash > (p->mask >> p->shift);
186 static const struct nla_policy tcindex_policy[TCA_TCINDEX_MAX + 1] = {
187 [TCA_TCINDEX_HASH] = { .type = NLA_U32 },
188 [TCA_TCINDEX_MASK] = { .type = NLA_U16 },
189 [TCA_TCINDEX_SHIFT] = { .type = NLA_U32 },
190 [TCA_TCINDEX_FALL_THROUGH] = { .type = NLA_U32 },
191 [TCA_TCINDEX_CLASSID] = { .type = NLA_U32 },
195 tcindex_set_parms(struct net *net, struct tcf_proto *tp, unsigned long base,
196 u32 handle, struct tcindex_data *p,
197 struct tcindex_filter_result *r, struct nlattr **tb,
201 struct tcindex_filter_result new_filter_result, *old_r = r;
202 struct tcindex_filter_result cr;
203 struct tcindex_data cp;
204 struct tcindex_filter *f = NULL; /* make gcc behave */
207 tcf_exts_init(&e, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
208 err = tcf_exts_validate(net, tp, tb, est, &e);
212 memcpy(&cp, p, sizeof(cp));
213 memset(&new_filter_result, 0, sizeof(new_filter_result));
214 tcf_exts_init(&new_filter_result.exts, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
217 memcpy(&cr, r, sizeof(cr));
219 memset(&cr, 0, sizeof(cr));
220 tcf_exts_init(&cr.exts, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
223 if (tb[TCA_TCINDEX_HASH])
224 cp.hash = nla_get_u32(tb[TCA_TCINDEX_HASH]);
226 if (tb[TCA_TCINDEX_MASK])
227 cp.mask = nla_get_u16(tb[TCA_TCINDEX_MASK]);
229 if (tb[TCA_TCINDEX_SHIFT])
230 cp.shift = nla_get_u32(tb[TCA_TCINDEX_SHIFT]);
233 /* Hash already allocated, make sure that we still meet the
234 * requirements for the allocated hash.
237 if (!valid_perfect_hash(&cp) ||
238 cp.hash > cp.alloc_hash)
240 } else if (cp.h && cp.hash != cp.alloc_hash)
244 if (tb[TCA_TCINDEX_FALL_THROUGH])
245 cp.fall_through = nla_get_u32(tb[TCA_TCINDEX_FALL_THROUGH]);
248 /* Hash not specified, use perfect hash if the upper limit
249 * of the hashing index is below the threshold.
251 if ((cp.mask >> cp.shift) < PERFECT_HASH_THRESHOLD)
252 cp.hash = (cp.mask >> cp.shift) + 1;
254 cp.hash = DEFAULT_HASH_SIZE;
257 if (!cp.perfect && !cp.h)
258 cp.alloc_hash = cp.hash;
260 /* Note: this could be as restrictive as if (handle & ~(mask >> shift))
261 * but then, we'd fail handles that may become valid after some future
262 * mask change. While this is extremely unlikely to ever matter,
263 * the check below is safer (and also more backwards-compatible).
265 if (cp.perfect || valid_perfect_hash(&cp))
266 if (handle >= cp.alloc_hash)
271 if (!cp.perfect && !cp.h) {
272 if (valid_perfect_hash(&cp)) {
273 cp.perfect = kcalloc(cp.hash, sizeof(*r), GFP_KERNEL);
278 cp.h = kcalloc(cp.hash, sizeof(f), GFP_KERNEL);
286 r = cp.perfect + handle;
288 r = tcindex_lookup(&cp, handle) ? : &new_filter_result;
290 if (r == &new_filter_result) {
291 f = kzalloc(sizeof(*f), GFP_KERNEL);
296 if (tb[TCA_TCINDEX_CLASSID]) {
297 cr.res.classid = nla_get_u32(tb[TCA_TCINDEX_CLASSID]);
298 tcf_bind_filter(tp, &cr.res, base);
301 tcf_exts_change(tp, &cr.exts, &e);
304 if (old_r && old_r != r)
305 memset(old_r, 0, sizeof(*old_r));
307 memcpy(p, &cp, sizeof(cp));
308 memcpy(r, &cr, sizeof(cr));
310 if (r == &new_filter_result) {
311 struct tcindex_filter **fp;
314 f->result = new_filter_result;
316 for (fp = p->h+(handle % p->hash); *fp; fp = &(*fp)->next)
327 else if (balloc == 2)
330 tcf_exts_destroy(tp, &e);
335 tcindex_change(struct net *net, struct sk_buff *in_skb,
336 struct tcf_proto *tp, unsigned long base, u32 handle,
337 struct nlattr **tca, unsigned long *arg)
339 struct nlattr *opt = tca[TCA_OPTIONS];
340 struct nlattr *tb[TCA_TCINDEX_MAX + 1];
341 struct tcindex_data *p = PRIV(tp);
342 struct tcindex_filter_result *r = (struct tcindex_filter_result *) *arg;
345 pr_debug("tcindex_change(tp %p,handle 0x%08x,tca %p,arg %p),opt %p,"
346 "p %p,r %p,*arg 0x%lx\n",
347 tp, handle, tca, arg, opt, p, r, arg ? *arg : 0L);
352 err = nla_parse_nested(tb, TCA_TCINDEX_MAX, opt, tcindex_policy);
356 return tcindex_set_parms(net, tp, base, handle, p, r, tb,
361 static void tcindex_walk(struct tcf_proto *tp, struct tcf_walker *walker)
363 struct tcindex_data *p = PRIV(tp);
364 struct tcindex_filter *f, *next;
367 pr_debug("tcindex_walk(tp %p,walker %p),p %p\n", tp, walker, p);
369 for (i = 0; i < p->hash; i++) {
370 if (!p->perfect[i].res.class)
372 if (walker->count >= walker->skip) {
374 (unsigned long) (p->perfect+i), walker)
385 for (i = 0; i < p->hash; i++) {
386 for (f = p->h[i]; f; f = next) {
388 if (walker->count >= walker->skip) {
389 if (walker->fn(tp, (unsigned long) &f->result,
401 static int tcindex_destroy_element(struct tcf_proto *tp,
402 unsigned long arg, struct tcf_walker *walker)
404 return __tcindex_delete(tp, arg, 0);
408 static void tcindex_destroy(struct tcf_proto *tp)
410 struct tcindex_data *p = PRIV(tp);
411 struct tcf_walker walker;
413 pr_debug("tcindex_destroy(tp %p),p %p\n", tp, p);
416 walker.fn = &tcindex_destroy_element;
417 tcindex_walk(tp, &walker);
425 static int tcindex_dump(struct tcf_proto *tp, unsigned long fh,
426 struct sk_buff *skb, struct tcmsg *t)
428 struct tcindex_data *p = PRIV(tp);
429 struct tcindex_filter_result *r = (struct tcindex_filter_result *) fh;
430 unsigned char *b = skb_tail_pointer(skb);
433 pr_debug("tcindex_dump(tp %p,fh 0x%lx,skb %p,t %p),p %p,r %p,b %p\n",
434 tp, fh, skb, t, p, r, b);
435 pr_debug("p->perfect %p p->h %p\n", p->perfect, p->h);
437 nest = nla_nest_start(skb, TCA_OPTIONS);
439 goto nla_put_failure;
442 t->tcm_handle = ~0; /* whatever ... */
443 if (nla_put_u32(skb, TCA_TCINDEX_HASH, p->hash) ||
444 nla_put_u16(skb, TCA_TCINDEX_MASK, p->mask) ||
445 nla_put_u32(skb, TCA_TCINDEX_SHIFT, p->shift) ||
446 nla_put_u32(skb, TCA_TCINDEX_FALL_THROUGH, p->fall_through))
447 goto nla_put_failure;
448 nla_nest_end(skb, nest);
451 t->tcm_handle = r-p->perfect;
453 struct tcindex_filter *f;
457 for (i = 0; !t->tcm_handle && i < p->hash; i++) {
458 for (f = p->h[i]; !t->tcm_handle && f;
461 t->tcm_handle = f->key;
465 pr_debug("handle = %d\n", t->tcm_handle);
467 nla_put_u32(skb, TCA_TCINDEX_CLASSID, r->res.classid))
468 goto nla_put_failure;
470 if (tcf_exts_dump(skb, &r->exts) < 0)
471 goto nla_put_failure;
472 nla_nest_end(skb, nest);
474 if (tcf_exts_dump_stats(skb, &r->exts) < 0)
475 goto nla_put_failure;
485 static struct tcf_proto_ops cls_tcindex_ops __read_mostly = {
487 .classify = tcindex_classify,
488 .init = tcindex_init,
489 .destroy = tcindex_destroy,
492 .change = tcindex_change,
493 .delete = tcindex_delete,
494 .walk = tcindex_walk,
495 .dump = tcindex_dump,
496 .owner = THIS_MODULE,
499 static int __init init_tcindex(void)
501 return register_tcf_proto_ops(&cls_tcindex_ops);
504 static void __exit exit_tcindex(void)
506 unregister_tcf_proto_ops(&cls_tcindex_ops);
509 module_init(init_tcindex)
510 module_exit(exit_tcindex)
511 MODULE_LICENSE("GPL");