Merge branch 'for-2.6.28' of git://linux-nfs.org/~bfields/linux
[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 <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 const 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 const struct nla_policy tcindex_policy[TCA_TCINDEX_MAX + 1] = {
197         [TCA_TCINDEX_HASH]              = { .type = NLA_U32 },
198         [TCA_TCINDEX_MASK]              = { .type = NLA_U16 },
199         [TCA_TCINDEX_SHIFT]             = { .type = NLA_U32 },
200         [TCA_TCINDEX_FALL_THROUGH]      = { .type = NLA_U32 },
201         [TCA_TCINDEX_CLASSID]           = { .type = NLA_U32 },
202 };
203
204 static int
205 tcindex_set_parms(struct tcf_proto *tp, unsigned long base, u32 handle,
206                   struct tcindex_data *p, struct tcindex_filter_result *r,
207                   struct nlattr **tb, struct nlattr *est)
208 {
209         int err, balloc = 0;
210         struct tcindex_filter_result new_filter_result, *old_r = r;
211         struct tcindex_filter_result cr;
212         struct tcindex_data cp;
213         struct tcindex_filter *f = NULL; /* make gcc behave */
214         struct tcf_exts e;
215
216         err = tcf_exts_validate(tp, tb, est, &e, &tcindex_ext_map);
217         if (err < 0)
218                 return err;
219
220         memcpy(&cp, p, sizeof(cp));
221         memset(&new_filter_result, 0, sizeof(new_filter_result));
222
223         if (old_r)
224                 memcpy(&cr, r, sizeof(cr));
225         else
226                 memset(&cr, 0, sizeof(cr));
227
228         if (tb[TCA_TCINDEX_HASH])
229                 cp.hash = nla_get_u32(tb[TCA_TCINDEX_HASH]);
230
231         if (tb[TCA_TCINDEX_MASK])
232                 cp.mask = nla_get_u16(tb[TCA_TCINDEX_MASK]);
233
234         if (tb[TCA_TCINDEX_SHIFT])
235                 cp.shift = nla_get_u32(tb[TCA_TCINDEX_SHIFT]);
236
237         err = -EBUSY;
238         /* Hash already allocated, make sure that we still meet the
239          * requirements for the allocated hash.
240          */
241         if (cp.perfect) {
242                 if (!valid_perfect_hash(&cp) ||
243                     cp.hash > cp.alloc_hash)
244                         goto errout;
245         } else if (cp.h && cp.hash != cp.alloc_hash)
246                 goto errout;
247
248         err = -EINVAL;
249         if (tb[TCA_TCINDEX_FALL_THROUGH])
250                 cp.fall_through = nla_get_u32(tb[TCA_TCINDEX_FALL_THROUGH]);
251
252         if (!cp.hash) {
253                 /* Hash not specified, use perfect hash if the upper limit
254                  * of the hashing index is below the threshold.
255                  */
256                 if ((cp.mask >> cp.shift) < PERFECT_HASH_THRESHOLD)
257                         cp.hash = (cp.mask >> cp.shift)+1;
258                 else
259                         cp.hash = DEFAULT_HASH_SIZE;
260         }
261
262         if (!cp.perfect && !cp.h)
263                 cp.alloc_hash = cp.hash;
264
265         /* Note: this could be as restrictive as if (handle & ~(mask >> shift))
266          * but then, we'd fail handles that may become valid after some future
267          * mask change. While this is extremely unlikely to ever matter,
268          * the check below is safer (and also more backwards-compatible).
269          */
270         if (cp.perfect || valid_perfect_hash(&cp))
271                 if (handle >= cp.alloc_hash)
272                         goto errout;
273
274
275         err = -ENOMEM;
276         if (!cp.perfect && !cp.h) {
277                 if (valid_perfect_hash(&cp)) {
278                         cp.perfect = kcalloc(cp.hash, sizeof(*r), GFP_KERNEL);
279                         if (!cp.perfect)
280                                 goto errout;
281                         balloc = 1;
282                 } else {
283                         cp.h = kcalloc(cp.hash, sizeof(f), GFP_KERNEL);
284                         if (!cp.h)
285                                 goto errout;
286                         balloc = 2;
287                 }
288         }
289
290         if (cp.perfect)
291                 r = cp.perfect + handle;
292         else
293                 r = tcindex_lookup(&cp, handle) ? : &new_filter_result;
294
295         if (r == &new_filter_result) {
296                 f = kzalloc(sizeof(*f), GFP_KERNEL);
297                 if (!f)
298                         goto errout_alloc;
299         }
300
301         if (tb[TCA_TCINDEX_CLASSID]) {
302                 cr.res.classid = nla_get_u32(tb[TCA_TCINDEX_CLASSID]);
303                 tcf_bind_filter(tp, &cr.res, base);
304         }
305
306         tcf_exts_change(tp, &cr.exts, &e);
307
308         tcf_tree_lock(tp);
309         if (old_r && old_r != r)
310                 memset(old_r, 0, sizeof(*old_r));
311
312         memcpy(p, &cp, sizeof(cp));
313         memcpy(r, &cr, sizeof(cr));
314
315         if (r == &new_filter_result) {
316                 struct tcindex_filter **fp;
317
318                 f->key = handle;
319                 f->result = new_filter_result;
320                 f->next = NULL;
321                 for (fp = p->h+(handle % p->hash); *fp; fp = &(*fp)->next)
322                         /* nothing */;
323                 *fp = f;
324         }
325         tcf_tree_unlock(tp);
326
327         return 0;
328
329 errout_alloc:
330         if (balloc == 1)
331                 kfree(cp.perfect);
332         else if (balloc == 2)
333                 kfree(cp.h);
334 errout:
335         tcf_exts_destroy(tp, &e);
336         return err;
337 }
338
339 static int
340 tcindex_change(struct tcf_proto *tp, unsigned long base, u32 handle,
341                struct nlattr **tca, unsigned long *arg)
342 {
343         struct nlattr *opt = tca[TCA_OPTIONS];
344         struct nlattr *tb[TCA_TCINDEX_MAX + 1];
345         struct tcindex_data *p = PRIV(tp);
346         struct tcindex_filter_result *r = (struct tcindex_filter_result *) *arg;
347         int err;
348
349         pr_debug("tcindex_change(tp %p,handle 0x%08x,tca %p,arg %p),opt %p,"
350             "p %p,r %p,*arg 0x%lx\n",
351             tp, handle, tca, arg, opt, p, r, arg ? *arg : 0L);
352
353         if (!opt)
354                 return 0;
355
356         err = nla_parse_nested(tb, TCA_TCINDEX_MAX, opt, tcindex_policy);
357         if (err < 0)
358                 return err;
359
360         return tcindex_set_parms(tp, base, handle, p, r, tb, tca[TCA_RATE]);
361 }
362
363
364 static void tcindex_walk(struct tcf_proto *tp, struct tcf_walker *walker)
365 {
366         struct tcindex_data *p = PRIV(tp);
367         struct tcindex_filter *f, *next;
368         int i;
369
370         pr_debug("tcindex_walk(tp %p,walker %p),p %p\n", tp, walker, p);
371         if (p->perfect) {
372                 for (i = 0; i < p->hash; i++) {
373                         if (!p->perfect[i].res.class)
374                                 continue;
375                         if (walker->count >= walker->skip) {
376                                 if (walker->fn(tp,
377                                     (unsigned long) (p->perfect+i), walker)
378                                      < 0) {
379                                         walker->stop = 1;
380                                         return;
381                                 }
382                         }
383                         walker->count++;
384                 }
385         }
386         if (!p->h)
387                 return;
388         for (i = 0; i < p->hash; i++) {
389                 for (f = p->h[i]; f; f = next) {
390                         next = f->next;
391                         if (walker->count >= walker->skip) {
392                                 if (walker->fn(tp, (unsigned long) &f->result,
393                                     walker) < 0) {
394                                         walker->stop = 1;
395                                         return;
396                                 }
397                         }
398                         walker->count++;
399                 }
400         }
401 }
402
403
404 static int tcindex_destroy_element(struct tcf_proto *tp,
405     unsigned long arg, struct tcf_walker *walker)
406 {
407         return __tcindex_delete(tp, arg, 0);
408 }
409
410
411 static void tcindex_destroy(struct tcf_proto *tp)
412 {
413         struct tcindex_data *p = PRIV(tp);
414         struct tcf_walker walker;
415
416         pr_debug("tcindex_destroy(tp %p),p %p\n", tp, p);
417         walker.count = 0;
418         walker.skip = 0;
419         walker.fn = &tcindex_destroy_element;
420         tcindex_walk(tp, &walker);
421         kfree(p->perfect);
422         kfree(p->h);
423         kfree(p);
424         tp->root = NULL;
425 }
426
427
428 static int tcindex_dump(struct tcf_proto *tp, unsigned long fh,
429     struct sk_buff *skb, struct tcmsg *t)
430 {
431         struct tcindex_data *p = PRIV(tp);
432         struct tcindex_filter_result *r = (struct tcindex_filter_result *) fh;
433         unsigned char *b = skb_tail_pointer(skb);
434         struct nlattr *nest;
435
436         pr_debug("tcindex_dump(tp %p,fh 0x%lx,skb %p,t %p),p %p,r %p,b %p\n",
437                  tp, fh, skb, t, p, r, b);
438         pr_debug("p->perfect %p p->h %p\n", p->perfect, p->h);
439
440         nest = nla_nest_start(skb, TCA_OPTIONS);
441         if (nest == NULL)
442                 goto nla_put_failure;
443
444         if (!fh) {
445                 t->tcm_handle = ~0; /* whatever ... */
446                 NLA_PUT_U32(skb, TCA_TCINDEX_HASH, p->hash);
447                 NLA_PUT_U16(skb, TCA_TCINDEX_MASK, p->mask);
448                 NLA_PUT_U32(skb, TCA_TCINDEX_SHIFT, p->shift);
449                 NLA_PUT_U32(skb, TCA_TCINDEX_FALL_THROUGH, p->fall_through);
450                 nla_nest_end(skb, nest);
451         } else {
452                 if (p->perfect) {
453                         t->tcm_handle = r-p->perfect;
454                 } else {
455                         struct tcindex_filter *f;
456                         int i;
457
458                         t->tcm_handle = 0;
459                         for (i = 0; !t->tcm_handle && i < p->hash; i++) {
460                                 for (f = p->h[i]; !t->tcm_handle && f;
461                                      f = f->next) {
462                                         if (&f->result == r)
463                                                 t->tcm_handle = f->key;
464                                 }
465                         }
466                 }
467                 pr_debug("handle = %d\n", t->tcm_handle);
468                 if (r->res.class)
469                         NLA_PUT_U32(skb, TCA_TCINDEX_CLASSID, r->res.classid);
470
471                 if (tcf_exts_dump(skb, &r->exts, &tcindex_ext_map) < 0)
472                         goto nla_put_failure;
473                 nla_nest_end(skb, nest);
474
475                 if (tcf_exts_dump_stats(skb, &r->exts, &tcindex_ext_map) < 0)
476                         goto nla_put_failure;
477         }
478
479         return skb->len;
480
481 nla_put_failure:
482         nlmsg_trim(skb, b);
483         return -1;
484 }
485
486 static struct tcf_proto_ops cls_tcindex_ops __read_mostly = {
487         .kind           =       "tcindex",
488         .classify       =       tcindex_classify,
489         .init           =       tcindex_init,
490         .destroy        =       tcindex_destroy,
491         .get            =       tcindex_get,
492         .put            =       tcindex_put,
493         .change         =       tcindex_change,
494         .delete         =       tcindex_delete,
495         .walk           =       tcindex_walk,
496         .dump           =       tcindex_dump,
497         .owner          =       THIS_MODULE,
498 };
499
500 static int __init init_tcindex(void)
501 {
502         return register_tcf_proto_ops(&cls_tcindex_ops);
503 }
504
505 static void __exit exit_tcindex(void)
506 {
507         unregister_tcf_proto_ops(&cls_tcindex_ops);
508 }
509
510 module_init(init_tcindex)
511 module_exit(exit_tcindex)
512 MODULE_LICENSE("GPL");