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