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 <net/act_api.h>
13 #include <net/netlink.h>
14 #include <net/pkt_cls.h>
18 * Not quite sure if we need all the xchgs Alexey uses when accessing things.
19 * Can always add them later ... :)
23 * Passing parameters to the root seems to be done more awkwardly than really
24 * necessary. At least, u32 doesn't seem to use such dirty hacks. To be
28 #define PERFECT_HASH_THRESHOLD 64 /* use perfect hash if not bigger */
29 #define DEFAULT_HASH_SIZE 64 /* optimized for diffserv */
32 #define PRIV(tp) ((struct tcindex_data *) (tp)->root)
35 struct tcindex_filter_result {
37 struct tcf_result res;
40 struct tcindex_filter {
42 struct tcindex_filter_result result;
43 struct tcindex_filter *next;
48 struct tcindex_filter_result *perfect; /* perfect hash; NULL if none */
49 struct tcindex_filter **h; /* imperfect hash; only used if !perfect;
51 u16 mask; /* AND key with mask */
52 int shift; /* shift ANDed key to the right */
53 int hash; /* hash table size; 0 if undefined */
54 int alloc_hash; /* allocated size */
55 int fall_through; /* 0: only classify if explicit match */
58 static struct tcf_ext_map tcindex_ext_map = {
59 .police = TCA_TCINDEX_POLICE,
60 .action = TCA_TCINDEX_ACT
64 tcindex_filter_is_set(struct tcindex_filter_result *r)
66 return tcf_exts_is_predicative(&r->exts) || r->res.classid;
69 static struct tcindex_filter_result *
70 tcindex_lookup(struct tcindex_data *p, u16 key)
72 struct tcindex_filter *f;
75 return tcindex_filter_is_set(p->perfect + key) ?
76 p->perfect + key : NULL;
78 for (f = p->h[key % p->hash]; f; f = f->next)
87 static int tcindex_classify(struct sk_buff *skb, struct tcf_proto *tp,
88 struct tcf_result *res)
90 struct tcindex_data *p = PRIV(tp);
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 unsigned long tcindex_get(struct tcf_proto *tp, u32 handle)
115 struct tcindex_data *p = PRIV(tp);
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) ? (unsigned long) r : 0UL;
126 static void tcindex_put(struct tcf_proto *tp, unsigned long f)
128 pr_debug("tcindex_put(tp %p,f 0x%lx)\n", tp, f);
132 static int tcindex_init(struct tcf_proto *tp)
134 struct tcindex_data *p;
136 pr_debug("tcindex_init(tp %p)\n", tp);
137 p = kzalloc(sizeof(struct tcindex_data), GFP_KERNEL);
142 p->hash = DEFAULT_HASH_SIZE;
151 __tcindex_delete(struct tcf_proto *tp, unsigned long arg, int lock)
153 struct tcindex_data *p = PRIV(tp);
154 struct tcindex_filter_result *r = (struct tcindex_filter_result *) arg;
155 struct tcindex_filter *f = NULL;
157 pr_debug("tcindex_delete(tp %p,arg 0x%lx),p %p,f %p\n", tp, arg, p, f);
163 struct tcindex_filter **walk = NULL;
165 for (i = 0; i < p->hash; i++)
166 for (walk = p->h+i; *walk; walk = &(*walk)->next)
167 if (&(*walk)->result == r)
179 tcf_unbind_filter(tp, &r->res);
180 tcf_exts_destroy(tp, &r->exts);
185 static int tcindex_delete(struct tcf_proto *tp, unsigned long arg)
187 return __tcindex_delete(tp, arg, 1);
191 valid_perfect_hash(struct tcindex_data *p)
193 return p->hash > (p->mask >> p->shift);
197 tcindex_set_parms(struct tcf_proto *tp, unsigned long base, u32 handle,
198 struct tcindex_data *p, struct tcindex_filter_result *r,
199 struct nlattr **tb, struct nlattr *est)
202 struct tcindex_filter_result new_filter_result, *old_r = r;
203 struct tcindex_filter_result cr;
204 struct tcindex_data cp;
205 struct tcindex_filter *f = NULL; /* make gcc behave */
208 err = tcf_exts_validate(tp, tb, est, &e, &tcindex_ext_map);
212 memcpy(&cp, p, sizeof(cp));
213 memset(&new_filter_result, 0, sizeof(new_filter_result));
216 memcpy(&cr, r, sizeof(cr));
218 memset(&cr, 0, sizeof(cr));
221 if (tb[TCA_TCINDEX_HASH]) {
222 if (nla_len(tb[TCA_TCINDEX_HASH]) < sizeof(u32))
224 cp.hash = nla_get_u32(tb[TCA_TCINDEX_HASH]);
227 if (tb[TCA_TCINDEX_MASK]) {
228 if (nla_len(tb[TCA_TCINDEX_MASK]) < sizeof(u16))
230 cp.mask = nla_get_u16(tb[TCA_TCINDEX_MASK]);
233 if (tb[TCA_TCINDEX_SHIFT]) {
234 if (nla_len(tb[TCA_TCINDEX_SHIFT]) < sizeof(int))
236 cp.shift = nla_get_u32(tb[TCA_TCINDEX_SHIFT]);
240 /* Hash already allocated, make sure that we still meet the
241 * requirements for the allocated hash.
244 if (!valid_perfect_hash(&cp) ||
245 cp.hash > cp.alloc_hash)
247 } else if (cp.h && cp.hash != cp.alloc_hash)
251 if (tb[TCA_TCINDEX_FALL_THROUGH]) {
252 if (nla_len(tb[TCA_TCINDEX_FALL_THROUGH]) < sizeof(u32))
254 cp.fall_through = nla_get_u32(tb[TCA_TCINDEX_FALL_THROUGH]);
258 /* Hash not specified, use perfect hash if the upper limit
259 * of the hashing index is below the threshold.
261 if ((cp.mask >> cp.shift) < PERFECT_HASH_THRESHOLD)
262 cp.hash = (cp.mask >> cp.shift)+1;
264 cp.hash = DEFAULT_HASH_SIZE;
267 if (!cp.perfect && !cp.h)
268 cp.alloc_hash = cp.hash;
270 /* Note: this could be as restrictive as if (handle & ~(mask >> shift))
271 * but then, we'd fail handles that may become valid after some future
272 * mask change. While this is extremely unlikely to ever matter,
273 * the check below is safer (and also more backwards-compatible).
275 if (cp.perfect || valid_perfect_hash(&cp))
276 if (handle >= cp.alloc_hash)
281 if (!cp.perfect && !cp.h) {
282 if (valid_perfect_hash(&cp)) {
283 cp.perfect = kcalloc(cp.hash, sizeof(*r), GFP_KERNEL);
288 cp.h = kcalloc(cp.hash, sizeof(f), GFP_KERNEL);
296 r = cp.perfect + handle;
298 r = tcindex_lookup(&cp, handle) ? : &new_filter_result;
300 if (r == &new_filter_result) {
301 f = kzalloc(sizeof(*f), GFP_KERNEL);
306 if (tb[TCA_TCINDEX_CLASSID]) {
307 cr.res.classid = nla_get_u32(tb[TCA_TCINDEX_CLASSID]);
308 tcf_bind_filter(tp, &cr.res, base);
311 tcf_exts_change(tp, &cr.exts, &e);
314 if (old_r && old_r != r)
315 memset(old_r, 0, sizeof(*old_r));
317 memcpy(p, &cp, sizeof(cp));
318 memcpy(r, &cr, sizeof(cr));
320 if (r == &new_filter_result) {
321 struct tcindex_filter **fp;
324 f->result = new_filter_result;
326 for (fp = p->h+(handle % p->hash); *fp; fp = &(*fp)->next)
337 else if (balloc == 2)
340 tcf_exts_destroy(tp, &e);
345 tcindex_change(struct tcf_proto *tp, unsigned long base, u32 handle,
346 struct nlattr **tca, unsigned long *arg)
348 struct nlattr *opt = tca[TCA_OPTIONS];
349 struct nlattr *tb[TCA_TCINDEX_MAX + 1];
350 struct tcindex_data *p = PRIV(tp);
351 struct tcindex_filter_result *r = (struct tcindex_filter_result *) *arg;
354 pr_debug("tcindex_change(tp %p,handle 0x%08x,tca %p,arg %p),opt %p,"
355 "p %p,r %p,*arg 0x%lx\n",
356 tp, handle, tca, arg, opt, p, r, arg ? *arg : 0L);
361 err = nla_parse_nested(tb, TCA_TCINDEX_MAX, opt, NULL);
365 return tcindex_set_parms(tp, base, handle, p, r, tb, tca[TCA_RATE]);
369 static void tcindex_walk(struct tcf_proto *tp, struct tcf_walker *walker)
371 struct tcindex_data *p = PRIV(tp);
372 struct tcindex_filter *f, *next;
375 pr_debug("tcindex_walk(tp %p,walker %p),p %p\n", tp, walker, p);
377 for (i = 0; i < p->hash; i++) {
378 if (!p->perfect[i].res.class)
380 if (walker->count >= walker->skip) {
382 (unsigned long) (p->perfect+i), walker)
393 for (i = 0; i < p->hash; i++) {
394 for (f = p->h[i]; f; f = next) {
396 if (walker->count >= walker->skip) {
397 if (walker->fn(tp, (unsigned long) &f->result,
409 static int tcindex_destroy_element(struct tcf_proto *tp,
410 unsigned long arg, struct tcf_walker *walker)
412 return __tcindex_delete(tp, arg, 0);
416 static void tcindex_destroy(struct tcf_proto *tp)
418 struct tcindex_data *p = PRIV(tp);
419 struct tcf_walker walker;
421 pr_debug("tcindex_destroy(tp %p),p %p\n", tp, p);
424 walker.fn = &tcindex_destroy_element;
425 tcindex_walk(tp, &walker);
433 static int tcindex_dump(struct tcf_proto *tp, unsigned long fh,
434 struct sk_buff *skb, struct tcmsg *t)
436 struct tcindex_data *p = PRIV(tp);
437 struct tcindex_filter_result *r = (struct tcindex_filter_result *) fh;
438 unsigned char *b = skb_tail_pointer(skb);
441 pr_debug("tcindex_dump(tp %p,fh 0x%lx,skb %p,t %p),p %p,r %p,b %p\n",
442 tp, fh, skb, t, p, r, b);
443 pr_debug("p->perfect %p p->h %p\n", p->perfect, p->h);
445 nest = nla_nest_start(skb, TCA_OPTIONS);
447 goto nla_put_failure;
450 t->tcm_handle = ~0; /* whatever ... */
451 NLA_PUT_U32(skb, TCA_TCINDEX_HASH, p->hash);
452 NLA_PUT_U16(skb, TCA_TCINDEX_MASK, p->mask);
453 NLA_PUT_U32(skb, TCA_TCINDEX_SHIFT, p->shift);
454 NLA_PUT_U32(skb, TCA_TCINDEX_FALL_THROUGH, p->fall_through);
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);
476 if (tcf_exts_dump(skb, &r->exts, &tcindex_ext_map) < 0)
477 goto nla_put_failure;
478 nla_nest_end(skb, nest);
480 if (tcf_exts_dump_stats(skb, &r->exts, &tcindex_ext_map) < 0)
481 goto nla_put_failure;
491 static struct tcf_proto_ops cls_tcindex_ops __read_mostly = {
493 .classify = tcindex_classify,
494 .init = tcindex_init,
495 .destroy = tcindex_destroy,
498 .change = tcindex_change,
499 .delete = tcindex_delete,
500 .walk = tcindex_walk,
501 .dump = tcindex_dump,
502 .owner = THIS_MODULE,
505 static int __init init_tcindex(void)
507 return register_tcf_proto_ops(&cls_tcindex_ops);
510 static void __exit exit_tcindex(void)
512 unregister_tcf_proto_ops(&cls_tcindex_ops);
515 module_init(init_tcindex)
516 module_exit(exit_tcindex)
517 MODULE_LICENSE("GPL");