ipv4: tcp: set unicast_sock uc_ttl to -1
[linux-3.10.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 <linux/slab.h>
13 #include <net/act_api.h>
14 #include <net/netlink.h>
15 #include <net/pkt_cls.h>
16
17 /*
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
20  * verified. FIXME.
21  */
22
23 #define PERFECT_HASH_THRESHOLD  64      /* use perfect hash if not bigger */
24 #define DEFAULT_HASH_SIZE       64      /* optimized for diffserv */
25
26
27 #define PRIV(tp)        ((struct tcindex_data *) (tp)->root)
28
29
30 struct tcindex_filter_result {
31         struct tcf_exts         exts;
32         struct tcf_result       res;
33 };
34
35 struct tcindex_filter {
36         u16 key;
37         struct tcindex_filter_result result;
38         struct tcindex_filter *next;
39 };
40
41
42 struct tcindex_data {
43         struct tcindex_filter_result *perfect; /* perfect hash; NULL if none */
44         struct tcindex_filter **h; /* imperfect hash; only used if !perfect;
45                                       NULL if unused */
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 */
51 };
52
53 static const struct tcf_ext_map tcindex_ext_map = {
54         .police = TCA_TCINDEX_POLICE,
55         .action = TCA_TCINDEX_ACT
56 };
57
58 static inline int
59 tcindex_filter_is_set(struct tcindex_filter_result *r)
60 {
61         return tcf_exts_is_predicative(&r->exts) || r->res.classid;
62 }
63
64 static struct tcindex_filter_result *
65 tcindex_lookup(struct tcindex_data *p, u16 key)
66 {
67         struct tcindex_filter *f;
68
69         if (p->perfect)
70                 return tcindex_filter_is_set(p->perfect + key) ?
71                         p->perfect + key : NULL;
72         else if (p->h) {
73                 for (f = p->h[key % p->hash]; f; f = f->next)
74                         if (f->key == key)
75                                 return &f->result;
76         }
77
78         return NULL;
79 }
80
81
82 static int tcindex_classify(struct sk_buff *skb, const struct tcf_proto *tp,
83                             struct tcf_result *res)
84 {
85         struct tcindex_data *p = PRIV(tp);
86         struct tcindex_filter_result *f;
87         int key = (skb->tc_index & p->mask) >> p->shift;
88
89         pr_debug("tcindex_classify(skb %p,tp %p,res %p),p %p\n",
90                  skb, tp, res, p);
91
92         f = tcindex_lookup(p, key);
93         if (!f) {
94                 if (!p->fall_through)
95                         return -1;
96                 res->classid = TC_H_MAKE(TC_H_MAJ(tp->q->handle), key);
97                 res->class = 0;
98                 pr_debug("alg 0x%x\n", res->classid);
99                 return 0;
100         }
101         *res = f->res;
102         pr_debug("map 0x%x\n", res->classid);
103
104         return tcf_exts_exec(skb, &f->exts, res);
105 }
106
107
108 static unsigned long tcindex_get(struct tcf_proto *tp, u32 handle)
109 {
110         struct tcindex_data *p = PRIV(tp);
111         struct tcindex_filter_result *r;
112
113         pr_debug("tcindex_get(tp %p,handle 0x%08x)\n", tp, handle);
114         if (p->perfect && handle >= p->alloc_hash)
115                 return 0;
116         r = tcindex_lookup(p, handle);
117         return r && tcindex_filter_is_set(r) ? (unsigned long) r : 0UL;
118 }
119
120
121 static void tcindex_put(struct tcf_proto *tp, unsigned long f)
122 {
123         pr_debug("tcindex_put(tp %p,f 0x%lx)\n", tp, f);
124 }
125
126
127 static int tcindex_init(struct tcf_proto *tp)
128 {
129         struct tcindex_data *p;
130
131         pr_debug("tcindex_init(tp %p)\n", tp);
132         p = kzalloc(sizeof(struct tcindex_data), GFP_KERNEL);
133         if (!p)
134                 return -ENOMEM;
135
136         p->mask = 0xffff;
137         p->hash = DEFAULT_HASH_SIZE;
138         p->fall_through = 1;
139
140         tp->root = p;
141         return 0;
142 }
143
144
145 static int
146 __tcindex_delete(struct tcf_proto *tp, unsigned long arg, int lock)
147 {
148         struct tcindex_data *p = PRIV(tp);
149         struct tcindex_filter_result *r = (struct tcindex_filter_result *) arg;
150         struct tcindex_filter *f = NULL;
151
152         pr_debug("tcindex_delete(tp %p,arg 0x%lx),p %p,f %p\n", tp, arg, p, f);
153         if (p->perfect) {
154                 if (!r->res.class)
155                         return -ENOENT;
156         } else {
157                 int i;
158                 struct tcindex_filter **walk = NULL;
159
160                 for (i = 0; i < p->hash; i++)
161                         for (walk = p->h+i; *walk; walk = &(*walk)->next)
162                                 if (&(*walk)->result == r)
163                                         goto found;
164                 return -ENOENT;
165
166 found:
167                 f = *walk;
168                 if (lock)
169                         tcf_tree_lock(tp);
170                 *walk = f->next;
171                 if (lock)
172                         tcf_tree_unlock(tp);
173         }
174         tcf_unbind_filter(tp, &r->res);
175         tcf_exts_destroy(tp, &r->exts);
176         kfree(f);
177         return 0;
178 }
179
180 static int tcindex_delete(struct tcf_proto *tp, unsigned long arg)
181 {
182         return __tcindex_delete(tp, arg, 1);
183 }
184
185 static inline int
186 valid_perfect_hash(struct tcindex_data *p)
187 {
188         return  p->hash > (p->mask >> p->shift);
189 }
190
191 static const struct nla_policy tcindex_policy[TCA_TCINDEX_MAX + 1] = {
192         [TCA_TCINDEX_HASH]              = { .type = NLA_U32 },
193         [TCA_TCINDEX_MASK]              = { .type = NLA_U16 },
194         [TCA_TCINDEX_SHIFT]             = { .type = NLA_U32 },
195         [TCA_TCINDEX_FALL_THROUGH]      = { .type = NLA_U32 },
196         [TCA_TCINDEX_CLASSID]           = { .type = NLA_U32 },
197 };
198
199 static int
200 tcindex_set_parms(struct tcf_proto *tp, unsigned long base, u32 handle,
201                   struct tcindex_data *p, struct tcindex_filter_result *r,
202                   struct nlattr **tb, struct nlattr *est)
203 {
204         int err, balloc = 0;
205         struct tcindex_filter_result new_filter_result, *old_r = r;
206         struct tcindex_filter_result cr;
207         struct tcindex_data cp;
208         struct tcindex_filter *f = NULL; /* make gcc behave */
209         struct tcf_exts e;
210
211         err = tcf_exts_validate(tp, tb, est, &e, &tcindex_ext_map);
212         if (err < 0)
213                 return err;
214
215         memcpy(&cp, p, sizeof(cp));
216         memset(&new_filter_result, 0, sizeof(new_filter_result));
217
218         if (old_r)
219                 memcpy(&cr, r, sizeof(cr));
220         else
221                 memset(&cr, 0, sizeof(cr));
222
223         if (tb[TCA_TCINDEX_HASH])
224                 cp.hash = nla_get_u32(tb[TCA_TCINDEX_HASH]);
225
226         if (tb[TCA_TCINDEX_MASK])
227                 cp.mask = nla_get_u16(tb[TCA_TCINDEX_MASK]);
228
229         if (tb[TCA_TCINDEX_SHIFT])
230                 cp.shift = nla_get_u32(tb[TCA_TCINDEX_SHIFT]);
231
232         err = -EBUSY;
233         /* Hash already allocated, make sure that we still meet the
234          * requirements for the allocated hash.
235          */
236         if (cp.perfect) {
237                 if (!valid_perfect_hash(&cp) ||
238                     cp.hash > cp.alloc_hash)
239                         goto errout;
240         } else if (cp.h && cp.hash != cp.alloc_hash)
241                 goto errout;
242
243         err = -EINVAL;
244         if (tb[TCA_TCINDEX_FALL_THROUGH])
245                 cp.fall_through = nla_get_u32(tb[TCA_TCINDEX_FALL_THROUGH]);
246
247         if (!cp.hash) {
248                 /* Hash not specified, use perfect hash if the upper limit
249                  * of the hashing index is below the threshold.
250                  */
251                 if ((cp.mask >> cp.shift) < PERFECT_HASH_THRESHOLD)
252                         cp.hash = (cp.mask >> cp.shift) + 1;
253                 else
254                         cp.hash = DEFAULT_HASH_SIZE;
255         }
256
257         if (!cp.perfect && !cp.h)
258                 cp.alloc_hash = cp.hash;
259
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).
264          */
265         if (cp.perfect || valid_perfect_hash(&cp))
266                 if (handle >= cp.alloc_hash)
267                         goto errout;
268
269
270         err = -ENOMEM;
271         if (!cp.perfect && !cp.h) {
272                 if (valid_perfect_hash(&cp)) {
273                         cp.perfect = kcalloc(cp.hash, sizeof(*r), GFP_KERNEL);
274                         if (!cp.perfect)
275                                 goto errout;
276                         balloc = 1;
277                 } else {
278                         cp.h = kcalloc(cp.hash, sizeof(f), GFP_KERNEL);
279                         if (!cp.h)
280                                 goto errout;
281                         balloc = 2;
282                 }
283         }
284
285         if (cp.perfect)
286                 r = cp.perfect + handle;
287         else
288                 r = tcindex_lookup(&cp, handle) ? : &new_filter_result;
289
290         if (r == &new_filter_result) {
291                 f = kzalloc(sizeof(*f), GFP_KERNEL);
292                 if (!f)
293                         goto errout_alloc;
294         }
295
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);
299         }
300
301         tcf_exts_change(tp, &cr.exts, &e);
302
303         tcf_tree_lock(tp);
304         if (old_r && old_r != r)
305                 memset(old_r, 0, sizeof(*old_r));
306
307         memcpy(p, &cp, sizeof(cp));
308         memcpy(r, &cr, sizeof(cr));
309
310         if (r == &new_filter_result) {
311                 struct tcindex_filter **fp;
312
313                 f->key = handle;
314                 f->result = new_filter_result;
315                 f->next = NULL;
316                 for (fp = p->h+(handle % p->hash); *fp; fp = &(*fp)->next)
317                         /* nothing */;
318                 *fp = f;
319         }
320         tcf_tree_unlock(tp);
321
322         return 0;
323
324 errout_alloc:
325         if (balloc == 1)
326                 kfree(cp.perfect);
327         else if (balloc == 2)
328                 kfree(cp.h);
329 errout:
330         tcf_exts_destroy(tp, &e);
331         return err;
332 }
333
334 static int
335 tcindex_change(struct tcf_proto *tp, unsigned long base, u32 handle,
336                struct nlattr **tca, unsigned long *arg)
337 {
338         struct nlattr *opt = tca[TCA_OPTIONS];
339         struct nlattr *tb[TCA_TCINDEX_MAX + 1];
340         struct tcindex_data *p = PRIV(tp);
341         struct tcindex_filter_result *r = (struct tcindex_filter_result *) *arg;
342         int err;
343
344         pr_debug("tcindex_change(tp %p,handle 0x%08x,tca %p,arg %p),opt %p,"
345             "p %p,r %p,*arg 0x%lx\n",
346             tp, handle, tca, arg, opt, p, r, arg ? *arg : 0L);
347
348         if (!opt)
349                 return 0;
350
351         err = nla_parse_nested(tb, TCA_TCINDEX_MAX, opt, tcindex_policy);
352         if (err < 0)
353                 return err;
354
355         return tcindex_set_parms(tp, base, handle, p, r, tb, tca[TCA_RATE]);
356 }
357
358
359 static void tcindex_walk(struct tcf_proto *tp, struct tcf_walker *walker)
360 {
361         struct tcindex_data *p = PRIV(tp);
362         struct tcindex_filter *f, *next;
363         int i;
364
365         pr_debug("tcindex_walk(tp %p,walker %p),p %p\n", tp, walker, p);
366         if (p->perfect) {
367                 for (i = 0; i < p->hash; i++) {
368                         if (!p->perfect[i].res.class)
369                                 continue;
370                         if (walker->count >= walker->skip) {
371                                 if (walker->fn(tp,
372                                     (unsigned long) (p->perfect+i), walker)
373                                      < 0) {
374                                         walker->stop = 1;
375                                         return;
376                                 }
377                         }
378                         walker->count++;
379                 }
380         }
381         if (!p->h)
382                 return;
383         for (i = 0; i < p->hash; i++) {
384                 for (f = p->h[i]; f; f = next) {
385                         next = f->next;
386                         if (walker->count >= walker->skip) {
387                                 if (walker->fn(tp, (unsigned long) &f->result,
388                                     walker) < 0) {
389                                         walker->stop = 1;
390                                         return;
391                                 }
392                         }
393                         walker->count++;
394                 }
395         }
396 }
397
398
399 static int tcindex_destroy_element(struct tcf_proto *tp,
400     unsigned long arg, struct tcf_walker *walker)
401 {
402         return __tcindex_delete(tp, arg, 0);
403 }
404
405
406 static void tcindex_destroy(struct tcf_proto *tp)
407 {
408         struct tcindex_data *p = PRIV(tp);
409         struct tcf_walker walker;
410
411         pr_debug("tcindex_destroy(tp %p),p %p\n", tp, p);
412         walker.count = 0;
413         walker.skip = 0;
414         walker.fn = &tcindex_destroy_element;
415         tcindex_walk(tp, &walker);
416         kfree(p->perfect);
417         kfree(p->h);
418         kfree(p);
419         tp->root = NULL;
420 }
421
422
423 static int tcindex_dump(struct tcf_proto *tp, unsigned long fh,
424     struct sk_buff *skb, struct tcmsg *t)
425 {
426         struct tcindex_data *p = PRIV(tp);
427         struct tcindex_filter_result *r = (struct tcindex_filter_result *) fh;
428         unsigned char *b = skb_tail_pointer(skb);
429         struct nlattr *nest;
430
431         pr_debug("tcindex_dump(tp %p,fh 0x%lx,skb %p,t %p),p %p,r %p,b %p\n",
432                  tp, fh, skb, t, p, r, b);
433         pr_debug("p->perfect %p p->h %p\n", p->perfect, p->h);
434
435         nest = nla_nest_start(skb, TCA_OPTIONS);
436         if (nest == NULL)
437                 goto nla_put_failure;
438
439         if (!fh) {
440                 t->tcm_handle = ~0; /* whatever ... */
441                 if (nla_put_u32(skb, TCA_TCINDEX_HASH, p->hash) ||
442                     nla_put_u16(skb, TCA_TCINDEX_MASK, p->mask) ||
443                     nla_put_u32(skb, TCA_TCINDEX_SHIFT, p->shift) ||
444                     nla_put_u32(skb, TCA_TCINDEX_FALL_THROUGH, p->fall_through))
445                         goto nla_put_failure;
446                 nla_nest_end(skb, nest);
447         } else {
448                 if (p->perfect) {
449                         t->tcm_handle = r-p->perfect;
450                 } else {
451                         struct tcindex_filter *f;
452                         int i;
453
454                         t->tcm_handle = 0;
455                         for (i = 0; !t->tcm_handle && i < p->hash; i++) {
456                                 for (f = p->h[i]; !t->tcm_handle && f;
457                                      f = f->next) {
458                                         if (&f->result == r)
459                                                 t->tcm_handle = f->key;
460                                 }
461                         }
462                 }
463                 pr_debug("handle = %d\n", t->tcm_handle);
464                 if (r->res.class &&
465                     nla_put_u32(skb, TCA_TCINDEX_CLASSID, r->res.classid))
466                         goto nla_put_failure;
467
468                 if (tcf_exts_dump(skb, &r->exts, &tcindex_ext_map) < 0)
469                         goto nla_put_failure;
470                 nla_nest_end(skb, nest);
471
472                 if (tcf_exts_dump_stats(skb, &r->exts, &tcindex_ext_map) < 0)
473                         goto nla_put_failure;
474         }
475
476         return skb->len;
477
478 nla_put_failure:
479         nlmsg_trim(skb, b);
480         return -1;
481 }
482
483 static struct tcf_proto_ops cls_tcindex_ops __read_mostly = {
484         .kind           =       "tcindex",
485         .classify       =       tcindex_classify,
486         .init           =       tcindex_init,
487         .destroy        =       tcindex_destroy,
488         .get            =       tcindex_get,
489         .put            =       tcindex_put,
490         .change         =       tcindex_change,
491         .delete         =       tcindex_delete,
492         .walk           =       tcindex_walk,
493         .dump           =       tcindex_dump,
494         .owner          =       THIS_MODULE,
495 };
496
497 static int __init init_tcindex(void)
498 {
499         return register_tcf_proto_ops(&cls_tcindex_ops);
500 }
501
502 static void __exit exit_tcindex(void)
503 {
504         unregister_tcf_proto_ops(&cls_tcindex_ops);
505 }
506
507 module_init(init_tcindex)
508 module_exit(exit_tcindex)
509 MODULE_LICENSE("GPL");