]> err.no Git - linux-2.6/blob - net/sched/cls_tcindex.c
Merge branch 'upstream' of git://lost.foo-projects.org/~ahkok/git/netdev-2.6 into...
[linux-2.6] / net / sched / cls_tcindex.c
1 /*
2  * net/sched/cls_tcindex.c      Packet classifier for skb->tc_index
3  *
4  * Written 1998,1999 by Werner Almesberger, EPFL ICA
5  */
6
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/netdevice.h>
13 #include <net/ip.h>
14 #include <net/act_api.h>
15 #include <net/pkt_cls.h>
16 #include <net/route.h>
17
18
19 /*
20  * Not quite sure if we need all the xchgs Alexey uses when accessing things.
21  * Can always add them later ... :)
22  */
23
24 /*
25  * Passing parameters to the root seems to be done more awkwardly than really
26  * necessary. At least, u32 doesn't seem to use such dirty hacks. To be
27  * verified. FIXME.
28  */
29
30 #define PERFECT_HASH_THRESHOLD  64      /* use perfect hash if not bigger */
31 #define DEFAULT_HASH_SIZE       64      /* optimized for diffserv */
32
33
34 #if 1 /* control */
35 #define DPRINTK(format,args...) printk(KERN_DEBUG format,##args)
36 #else
37 #define DPRINTK(format,args...)
38 #endif
39
40 #if 0 /* data */
41 #define D2PRINTK(format,args...) printk(KERN_DEBUG format,##args)
42 #else
43 #define D2PRINTK(format,args...)
44 #endif
45
46
47 #define PRIV(tp)        ((struct tcindex_data *) (tp)->root)
48
49
50 struct tcindex_filter_result {
51         struct tcf_exts         exts;
52         struct tcf_result       res;
53 };
54
55 struct tcindex_filter {
56         u16 key;
57         struct tcindex_filter_result result;
58         struct tcindex_filter *next;
59 };
60
61
62 struct tcindex_data {
63         struct tcindex_filter_result *perfect; /* perfect hash; NULL if none */
64         struct tcindex_filter **h; /* imperfect hash; only used if !perfect;
65                                       NULL if unused */
66         u16 mask;               /* AND key with mask */
67         int shift;              /* shift ANDed key to the right */
68         int hash;               /* hash table size; 0 if undefined */
69         int alloc_hash;         /* allocated size */
70         int fall_through;       /* 0: only classify if explicit match */
71 };
72
73 static struct tcf_ext_map tcindex_ext_map = {
74         .police = TCA_TCINDEX_POLICE,
75         .action = TCA_TCINDEX_ACT
76 };
77
78 static inline int
79 tcindex_filter_is_set(struct tcindex_filter_result *r)
80 {
81         return tcf_exts_is_predicative(&r->exts) || r->res.classid;
82 }
83
84 static struct tcindex_filter_result *
85 tcindex_lookup(struct tcindex_data *p, u16 key)
86 {
87         struct tcindex_filter *f;
88
89         if (p->perfect)
90                 return tcindex_filter_is_set(p->perfect + key) ?
91                         p->perfect + key : NULL;
92         else if (p->h) {
93                 for (f = p->h[key % p->hash]; f; f = f->next)
94                         if (f->key == key)
95                                 return &f->result;
96         }
97
98         return NULL;
99 }
100
101
102 static int tcindex_classify(struct sk_buff *skb, struct tcf_proto *tp,
103                             struct tcf_result *res)
104 {
105         struct tcindex_data *p = PRIV(tp);
106         struct tcindex_filter_result *f;
107         int key = (skb->tc_index & p->mask) >> p->shift;
108
109         D2PRINTK("tcindex_classify(skb %p,tp %p,res %p),p %p\n",skb,tp,res,p);
110
111         f = tcindex_lookup(p, key);
112         if (!f) {
113                 if (!p->fall_through)
114                         return -1;
115                 res->classid = TC_H_MAKE(TC_H_MAJ(tp->q->handle), key);
116                 res->class = 0;
117                 D2PRINTK("alg 0x%x\n",res->classid);
118                 return 0;
119         }
120         *res = f->res;
121         D2PRINTK("map 0x%x\n",res->classid);
122
123         return tcf_exts_exec(skb, &f->exts, res);
124 }
125
126
127 static unsigned long tcindex_get(struct tcf_proto *tp, u32 handle)
128 {
129         struct tcindex_data *p = PRIV(tp);
130         struct tcindex_filter_result *r;
131
132         DPRINTK("tcindex_get(tp %p,handle 0x%08x)\n",tp,handle);
133         if (p->perfect && handle >= p->alloc_hash)
134                 return 0;
135         r = tcindex_lookup(p, handle);
136         return r && tcindex_filter_is_set(r) ? (unsigned long) r : 0UL;
137 }
138
139
140 static void tcindex_put(struct tcf_proto *tp, unsigned long f)
141 {
142         DPRINTK("tcindex_put(tp %p,f 0x%lx)\n",tp,f);
143 }
144
145
146 static int tcindex_init(struct tcf_proto *tp)
147 {
148         struct tcindex_data *p;
149
150         DPRINTK("tcindex_init(tp %p)\n",tp);
151         p = kmalloc(sizeof(struct tcindex_data),GFP_KERNEL);
152         if (!p)
153                 return -ENOMEM;
154
155         memset(p, 0, sizeof(*p));
156         p->mask = 0xffff;
157         p->hash = DEFAULT_HASH_SIZE;
158         p->fall_through = 1;
159
160         tp->root = p;
161         return 0;
162 }
163
164
165 static int
166 __tcindex_delete(struct tcf_proto *tp, unsigned long arg, int lock)
167 {
168         struct tcindex_data *p = PRIV(tp);
169         struct tcindex_filter_result *r = (struct tcindex_filter_result *) arg;
170         struct tcindex_filter *f = NULL;
171
172         DPRINTK("tcindex_delete(tp %p,arg 0x%lx),p %p,f %p\n",tp,arg,p,f);
173         if (p->perfect) {
174                 if (!r->res.class)
175                         return -ENOENT;
176         } else {
177                 int i;
178                 struct tcindex_filter **walk = NULL;
179
180                 for (i = 0; i < p->hash; i++)
181                         for (walk = p->h+i; *walk; walk = &(*walk)->next)
182                                 if (&(*walk)->result == r)
183                                         goto found;
184                 return -ENOENT;
185
186 found:
187                 f = *walk;
188                 if (lock)
189                         tcf_tree_lock(tp);
190                 *walk = f->next;
191                 if (lock)
192                         tcf_tree_unlock(tp);
193         }
194         tcf_unbind_filter(tp, &r->res);
195         tcf_exts_destroy(tp, &r->exts);
196         kfree(f);
197         return 0;
198 }
199
200 static int tcindex_delete(struct tcf_proto *tp, unsigned long arg)
201 {
202         return __tcindex_delete(tp, arg, 1);
203 }
204
205 static inline int
206 valid_perfect_hash(struct tcindex_data *p)
207 {
208         return  p->hash > (p->mask >> p->shift);
209 }
210
211 static int
212 tcindex_set_parms(struct tcf_proto *tp, unsigned long base, u32 handle,
213                   struct tcindex_data *p, struct tcindex_filter_result *r,
214                   struct rtattr **tb, struct rtattr *est)
215 {
216         int err, balloc = 0;
217         struct tcindex_filter_result new_filter_result, *old_r = r;
218         struct tcindex_filter_result cr;
219         struct tcindex_data cp;
220         struct tcindex_filter *f = NULL; /* make gcc behave */
221         struct tcf_exts e;
222
223         err = tcf_exts_validate(tp, tb, est, &e, &tcindex_ext_map);
224         if (err < 0)
225                 return err;
226         
227         memcpy(&cp, p, sizeof(cp));
228         memset(&new_filter_result, 0, sizeof(new_filter_result));
229
230         if (old_r)
231                 memcpy(&cr, r, sizeof(cr));
232         else
233                 memset(&cr, 0, sizeof(cr));
234
235         err = -EINVAL;
236         if (tb[TCA_TCINDEX_HASH-1]) {
237                 if (RTA_PAYLOAD(tb[TCA_TCINDEX_HASH-1]) < sizeof(u32))
238                         goto errout;
239                 cp.hash = *(u32 *) RTA_DATA(tb[TCA_TCINDEX_HASH-1]);
240         }
241
242         if (tb[TCA_TCINDEX_MASK-1]) {
243                 if (RTA_PAYLOAD(tb[TCA_TCINDEX_MASK-1]) < sizeof(u16))
244                         goto errout;
245                 cp.mask = *(u16 *) RTA_DATA(tb[TCA_TCINDEX_MASK-1]);
246         }
247
248         if (tb[TCA_TCINDEX_SHIFT-1]) {
249                 if (RTA_PAYLOAD(tb[TCA_TCINDEX_SHIFT-1]) < sizeof(u16))
250                         goto errout;
251                 cp.shift = *(u16 *) RTA_DATA(tb[TCA_TCINDEX_SHIFT-1]);
252         }
253
254         err = -EBUSY;
255         /* Hash already allocated, make sure that we still meet the
256          * requirements for the allocated hash.
257          */
258         if (cp.perfect) {
259                 if (!valid_perfect_hash(&cp) ||
260                     cp.hash > cp.alloc_hash)
261                         goto errout;
262         } else if (cp.h && cp.hash != cp.alloc_hash)
263                 goto errout;
264
265         err = -EINVAL;
266         if (tb[TCA_TCINDEX_FALL_THROUGH-1]) {
267                 if (RTA_PAYLOAD(tb[TCA_TCINDEX_FALL_THROUGH-1]) < sizeof(u32))
268                         goto errout;
269                 cp.fall_through =
270                         *(u32 *) RTA_DATA(tb[TCA_TCINDEX_FALL_THROUGH-1]);
271         }
272
273         if (!cp.hash) {
274                 /* Hash not specified, use perfect hash if the upper limit
275                  * of the hashing index is below the threshold.
276                  */
277                 if ((cp.mask >> cp.shift) < PERFECT_HASH_THRESHOLD)
278                         cp.hash = (cp.mask >> cp.shift)+1;
279                 else
280                         cp.hash = DEFAULT_HASH_SIZE;
281         }
282
283         if (!cp.perfect && !cp.h)
284                 cp.alloc_hash = cp.hash;
285
286         /* Note: this could be as restrictive as if (handle & ~(mask >> shift))
287          * but then, we'd fail handles that may become valid after some future
288          * mask change. While this is extremely unlikely to ever matter,
289          * the check below is safer (and also more backwards-compatible).
290          */
291         if (cp.perfect || valid_perfect_hash(&cp))
292                 if (handle >= cp.alloc_hash)
293                         goto errout;
294
295
296         err = -ENOMEM;
297         if (!cp.perfect && !cp.h) {
298                 if (valid_perfect_hash(&cp)) {
299                         cp.perfect = kmalloc(cp.hash * sizeof(*r), GFP_KERNEL);
300                         if (!cp.perfect)
301                                 goto errout;
302                         memset(cp.perfect, 0, cp.hash * sizeof(*r));
303                         balloc = 1;
304                 } else {
305                         cp.h = kmalloc(cp.hash * sizeof(f), GFP_KERNEL);
306                         if (!cp.h)
307                                 goto errout;
308                         memset(cp.h, 0, cp.hash * sizeof(f));
309                         balloc = 2;
310                 }
311         }
312
313         if (cp.perfect)
314                 r = cp.perfect + handle;
315         else
316                 r = tcindex_lookup(&cp, handle) ? : &new_filter_result;
317
318         if (r == &new_filter_result) {
319                 f = kmalloc(sizeof(*f), GFP_KERNEL);
320                 if (!f)
321                         goto errout_alloc;
322                 memset(f, 0, sizeof(*f));
323         }
324
325         if (tb[TCA_TCINDEX_CLASSID-1]) {
326                 cr.res.classid = *(u32 *) RTA_DATA(tb[TCA_TCINDEX_CLASSID-1]);
327                 tcf_bind_filter(tp, &cr.res, base);
328         }
329
330         tcf_exts_change(tp, &cr.exts, &e);
331
332         tcf_tree_lock(tp);
333         if (old_r && old_r != r)
334                 memset(old_r, 0, sizeof(*old_r));
335
336         memcpy(p, &cp, sizeof(cp));
337         memcpy(r, &cr, sizeof(cr));
338
339         if (r == &new_filter_result) {
340                 struct tcindex_filter **fp;
341
342                 f->key = handle;
343                 f->result = new_filter_result;
344                 f->next = NULL;
345                 for (fp = p->h+(handle % p->hash); *fp; fp = &(*fp)->next)
346                         /* nothing */;
347                 *fp = f;
348         }
349         tcf_tree_unlock(tp);
350
351         return 0;
352
353 errout_alloc:
354         if (balloc == 1)
355                 kfree(cp.perfect);
356         else if (balloc == 2)
357                 kfree(cp.h);
358 errout:
359         tcf_exts_destroy(tp, &e);
360         return err;
361 }
362
363 static int
364 tcindex_change(struct tcf_proto *tp, unsigned long base, u32 handle,
365                struct rtattr **tca, unsigned long *arg)
366 {
367         struct rtattr *opt = tca[TCA_OPTIONS-1];
368         struct rtattr *tb[TCA_TCINDEX_MAX];
369         struct tcindex_data *p = PRIV(tp);
370         struct tcindex_filter_result *r = (struct tcindex_filter_result *) *arg;
371
372         DPRINTK("tcindex_change(tp %p,handle 0x%08x,tca %p,arg %p),opt %p,"
373             "p %p,r %p,*arg 0x%lx\n",
374             tp, handle, tca, arg, opt, p, r, arg ? *arg : 0L);
375
376         if (!opt)
377                 return 0;
378
379         if (rtattr_parse_nested(tb, TCA_TCINDEX_MAX, opt) < 0)
380                 return -EINVAL;
381
382         return tcindex_set_parms(tp, base, handle, p, r, tb, tca[TCA_RATE-1]);
383 }
384
385
386 static void tcindex_walk(struct tcf_proto *tp, struct tcf_walker *walker)
387 {
388         struct tcindex_data *p = PRIV(tp);
389         struct tcindex_filter *f,*next;
390         int i;
391
392         DPRINTK("tcindex_walk(tp %p,walker %p),p %p\n",tp,walker,p);
393         if (p->perfect) {
394                 for (i = 0; i < p->hash; i++) {
395                         if (!p->perfect[i].res.class)
396                                 continue;
397                         if (walker->count >= walker->skip) {
398                                 if (walker->fn(tp,
399                                     (unsigned long) (p->perfect+i), walker)
400                                      < 0) {
401                                         walker->stop = 1;
402                                         return;
403                                 }
404                         }
405                         walker->count++;
406                 }
407         }
408         if (!p->h)
409                 return;
410         for (i = 0; i < p->hash; i++) {
411                 for (f = p->h[i]; f; f = next) {
412                         next = f->next;
413                         if (walker->count >= walker->skip) {
414                                 if (walker->fn(tp,(unsigned long) &f->result,
415                                     walker) < 0) {
416                                         walker->stop = 1;
417                                         return;
418                                 }
419                         }
420                         walker->count++;
421                 }
422         }
423 }
424
425
426 static int tcindex_destroy_element(struct tcf_proto *tp,
427     unsigned long arg, struct tcf_walker *walker)
428 {
429         return __tcindex_delete(tp, arg, 0);
430 }
431
432
433 static void tcindex_destroy(struct tcf_proto *tp)
434 {
435         struct tcindex_data *p = PRIV(tp);
436         struct tcf_walker walker;
437
438         DPRINTK("tcindex_destroy(tp %p),p %p\n",tp,p);
439         walker.count = 0;
440         walker.skip = 0;
441         walker.fn = &tcindex_destroy_element;
442         tcindex_walk(tp,&walker);
443         kfree(p->perfect);
444         kfree(p->h);
445         kfree(p);
446         tp->root = NULL;
447 }
448
449
450 static int tcindex_dump(struct tcf_proto *tp, unsigned long fh,
451     struct sk_buff *skb, struct tcmsg *t)
452 {
453         struct tcindex_data *p = PRIV(tp);
454         struct tcindex_filter_result *r = (struct tcindex_filter_result *) fh;
455         unsigned char *b = skb->tail;
456         struct rtattr *rta;
457
458         DPRINTK("tcindex_dump(tp %p,fh 0x%lx,skb %p,t %p),p %p,r %p,b %p\n",
459             tp,fh,skb,t,p,r,b);
460         DPRINTK("p->perfect %p p->h %p\n",p->perfect,p->h);
461         rta = (struct rtattr *) b;
462         RTA_PUT(skb,TCA_OPTIONS,0,NULL);
463         if (!fh) {
464                 t->tcm_handle = ~0; /* whatever ... */
465                 RTA_PUT(skb,TCA_TCINDEX_HASH,sizeof(p->hash),&p->hash);
466                 RTA_PUT(skb,TCA_TCINDEX_MASK,sizeof(p->mask),&p->mask);
467                 RTA_PUT(skb,TCA_TCINDEX_SHIFT,sizeof(p->shift),&p->shift);
468                 RTA_PUT(skb,TCA_TCINDEX_FALL_THROUGH,sizeof(p->fall_through),
469                     &p->fall_through);
470                 rta->rta_len = skb->tail-b;
471         } else {
472                 if (p->perfect) {
473                         t->tcm_handle = r-p->perfect;
474                 } else {
475                         struct tcindex_filter *f;
476                         int i;
477
478                         t->tcm_handle = 0;
479                         for (i = 0; !t->tcm_handle && i < p->hash; i++) {
480                                 for (f = p->h[i]; !t->tcm_handle && f;
481                                      f = f->next) {
482                                         if (&f->result == r)
483                                                 t->tcm_handle = f->key;
484                                 }
485                         }
486                 }
487                 DPRINTK("handle = %d\n",t->tcm_handle);
488                 if (r->res.class)
489                         RTA_PUT(skb, TCA_TCINDEX_CLASSID, 4, &r->res.classid);
490
491                 if (tcf_exts_dump(skb, &r->exts, &tcindex_ext_map) < 0)
492                         goto rtattr_failure;
493                 rta->rta_len = skb->tail-b;
494
495                 if (tcf_exts_dump_stats(skb, &r->exts, &tcindex_ext_map) < 0)
496                         goto rtattr_failure;
497         }
498         
499         return skb->len;
500
501 rtattr_failure:
502         skb_trim(skb, b - skb->data);
503         return -1;
504 }
505
506 static struct tcf_proto_ops cls_tcindex_ops = {
507         .next           =       NULL,
508         .kind           =       "tcindex",
509         .classify       =       tcindex_classify,
510         .init           =       tcindex_init,
511         .destroy        =       tcindex_destroy,
512         .get            =       tcindex_get,
513         .put            =       tcindex_put,
514         .change         =       tcindex_change,
515         .delete         =       tcindex_delete,
516         .walk           =       tcindex_walk,
517         .dump           =       tcindex_dump,
518         .owner          =       THIS_MODULE,
519 };
520
521 static int __init init_tcindex(void)
522 {
523         return register_tcf_proto_ops(&cls_tcindex_ops);
524 }
525
526 static void __exit exit_tcindex(void) 
527 {
528         unregister_tcf_proto_ops(&cls_tcindex_ops);
529 }
530
531 module_init(init_tcindex)
532 module_exit(exit_tcindex)
533 MODULE_LICENSE("GPL");