blob: 404d9d83a7fab040efab32341db2732b31626cf9 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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/config.h>
8#include <linux/module.h>
9#include <linux/types.h>
10#include <linux/kernel.h>
11#include <linux/skbuff.h>
12#include <linux/errno.h>
13#include <linux/netdevice.h>
14#include <net/ip.h>
15#include <net/act_api.h>
16#include <net/pkt_cls.h>
17#include <net/route.h>
18
19
20/*
21 * Not quite sure if we need all the xchgs Alexey uses when accessing things.
22 * Can always add them later ... :)
23 */
24
25/*
26 * Passing parameters to the root seems to be done more awkwardly than really
27 * necessary. At least, u32 doesn't seem to use such dirty hacks. To be
28 * verified. FIXME.
29 */
30
31#define PERFECT_HASH_THRESHOLD 64 /* use perfect hash if not bigger */
32#define DEFAULT_HASH_SIZE 64 /* optimized for diffserv */
33
34
35#if 1 /* control */
36#define DPRINTK(format,args...) printk(KERN_DEBUG format,##args)
37#else
38#define DPRINTK(format,args...)
39#endif
40
41#if 0 /* data */
42#define D2PRINTK(format,args...) printk(KERN_DEBUG format,##args)
43#else
44#define D2PRINTK(format,args...)
45#endif
46
47
48#define PRIV(tp) ((struct tcindex_data *) (tp)->root)
49
50
51struct tcindex_filter_result {
52 struct tcf_exts exts;
53 struct tcf_result res;
54};
55
56struct tcindex_filter {
57 u16 key;
58 struct tcindex_filter_result result;
59 struct tcindex_filter *next;
60};
61
62
63struct tcindex_data {
64 struct tcindex_filter_result *perfect; /* perfect hash; NULL if none */
65 struct tcindex_filter **h; /* imperfect hash; only used if !perfect;
66 NULL if unused */
67 u16 mask; /* AND key with mask */
68 int shift; /* shift ANDed key to the right */
69 int hash; /* hash table size; 0 if undefined */
70 int alloc_hash; /* allocated size */
71 int fall_through; /* 0: only classify if explicit match */
72};
73
74static struct tcf_ext_map tcindex_ext_map = {
75 .police = TCA_TCINDEX_POLICE,
76 .action = TCA_TCINDEX_ACT
77};
78
79static inline int
80tcindex_filter_is_set(struct tcindex_filter_result *r)
81{
82 return tcf_exts_is_predicative(&r->exts) || r->res.classid;
83}
84
85static struct tcindex_filter_result *
86tcindex_lookup(struct tcindex_data *p, u16 key)
87{
88 struct tcindex_filter *f;
89
90 if (p->perfect)
91 return tcindex_filter_is_set(p->perfect + key) ?
92 p->perfect + key : NULL;
93 else if (p->h) {
94 for (f = p->h[key % p->hash]; f; f = f->next)
95 if (f->key == key)
96 return &f->result;
97 }
98
99 return NULL;
100}
101
102
103static int tcindex_classify(struct sk_buff *skb, struct tcf_proto *tp,
104 struct tcf_result *res)
105{
106 struct tcindex_data *p = PRIV(tp);
107 struct tcindex_filter_result *f;
108 int key = (skb->tc_index & p->mask) >> p->shift;
109
110 D2PRINTK("tcindex_classify(skb %p,tp %p,res %p),p %p\n",skb,tp,res,p);
111
112 f = tcindex_lookup(p, key);
113 if (!f) {
114 if (!p->fall_through)
115 return -1;
116 res->classid = TC_H_MAKE(TC_H_MAJ(tp->q->handle), key);
117 res->class = 0;
118 D2PRINTK("alg 0x%x\n",res->classid);
119 return 0;
120 }
121 *res = f->res;
122 D2PRINTK("map 0x%x\n",res->classid);
123
124 return tcf_exts_exec(skb, &f->exts, res);
125}
126
127
128static unsigned long tcindex_get(struct tcf_proto *tp, u32 handle)
129{
130 struct tcindex_data *p = PRIV(tp);
131 struct tcindex_filter_result *r;
132
133 DPRINTK("tcindex_get(tp %p,handle 0x%08x)\n",tp,handle);
134 if (p->perfect && handle >= p->alloc_hash)
135 return 0;
136 r = tcindex_lookup(p, handle);
137 return r && tcindex_filter_is_set(r) ? (unsigned long) r : 0UL;
138}
139
140
141static void tcindex_put(struct tcf_proto *tp, unsigned long f)
142{
143 DPRINTK("tcindex_put(tp %p,f 0x%lx)\n",tp,f);
144}
145
146
147static int tcindex_init(struct tcf_proto *tp)
148{
149 struct tcindex_data *p;
150
151 DPRINTK("tcindex_init(tp %p)\n",tp);
152 p = kmalloc(sizeof(struct tcindex_data),GFP_KERNEL);
153 if (!p)
154 return -ENOMEM;
155
156 memset(p, 0, sizeof(*p));
157 p->mask = 0xffff;
158 p->hash = DEFAULT_HASH_SIZE;
159 p->fall_through = 1;
160
161 tp->root = p;
162 return 0;
163}
164
165
166static int
167__tcindex_delete(struct tcf_proto *tp, unsigned long arg, int lock)
168{
169 struct tcindex_data *p = PRIV(tp);
170 struct tcindex_filter_result *r = (struct tcindex_filter_result *) arg;
171 struct tcindex_filter *f = NULL;
172
173 DPRINTK("tcindex_delete(tp %p,arg 0x%lx),p %p,f %p\n",tp,arg,p,f);
174 if (p->perfect) {
175 if (!r->res.class)
176 return -ENOENT;
177 } else {
178 int i;
179 struct tcindex_filter **walk = NULL;
180
181 for (i = 0; i < p->hash; i++)
182 for (walk = p->h+i; *walk; walk = &(*walk)->next)
183 if (&(*walk)->result == r)
184 goto found;
185 return -ENOENT;
186
187found:
188 f = *walk;
189 if (lock)
190 tcf_tree_lock(tp);
191 *walk = f->next;
192 if (lock)
193 tcf_tree_unlock(tp);
194 }
195 tcf_unbind_filter(tp, &r->res);
196 tcf_exts_destroy(tp, &r->exts);
197 if (f)
198 kfree(f);
199 return 0;
200}
201
202static int tcindex_delete(struct tcf_proto *tp, unsigned long arg)
203{
204 return __tcindex_delete(tp, arg, 1);
205}
206
207static inline int
208valid_perfect_hash(struct tcindex_data *p)
209{
210 return p->hash > (p->mask >> p->shift);
211}
212
213static int
214tcindex_set_parms(struct tcf_proto *tp, unsigned long base, u32 handle,
215 struct tcindex_data *p, struct tcindex_filter_result *r,
216 struct rtattr **tb, struct rtattr *est)
217{
218 int err, balloc = 0;
219 struct tcindex_filter_result new_filter_result, *old_r = r;
220 struct tcindex_filter_result cr;
221 struct tcindex_data cp;
222 struct tcindex_filter *f = NULL; /* make gcc behave */
223 struct tcf_exts e;
224
225 err = tcf_exts_validate(tp, tb, est, &e, &tcindex_ext_map);
226 if (err < 0)
227 return err;
228
229 memcpy(&cp, p, sizeof(cp));
230 memset(&new_filter_result, 0, sizeof(new_filter_result));
231
232 if (old_r)
233 memcpy(&cr, r, sizeof(cr));
234 else
235 memset(&cr, 0, sizeof(cr));
236
237 err = -EINVAL;
238 if (tb[TCA_TCINDEX_HASH-1]) {
239 if (RTA_PAYLOAD(tb[TCA_TCINDEX_HASH-1]) < sizeof(u32))
240 goto errout;
241 cp.hash = *(u32 *) RTA_DATA(tb[TCA_TCINDEX_HASH-1]);
242 }
243
244 if (tb[TCA_TCINDEX_MASK-1]) {
245 if (RTA_PAYLOAD(tb[TCA_TCINDEX_MASK-1]) < sizeof(u16))
246 goto errout;
247 cp.mask = *(u16 *) RTA_DATA(tb[TCA_TCINDEX_MASK-1]);
248 }
249
250 if (tb[TCA_TCINDEX_SHIFT-1]) {
251 if (RTA_PAYLOAD(tb[TCA_TCINDEX_SHIFT-1]) < sizeof(u16))
252 goto errout;
253 cp.shift = *(u16 *) RTA_DATA(tb[TCA_TCINDEX_SHIFT-1]);
254 }
255
256 err = -EBUSY;
257 /* Hash already allocated, make sure that we still meet the
258 * requirements for the allocated hash.
259 */
260 if (cp.perfect) {
261 if (!valid_perfect_hash(&cp) ||
262 cp.hash > cp.alloc_hash)
263 goto errout;
264 } else if (cp.h && cp.hash != cp.alloc_hash)
265 goto errout;
266
267 err = -EINVAL;
268 if (tb[TCA_TCINDEX_FALL_THROUGH-1]) {
269 if (RTA_PAYLOAD(tb[TCA_TCINDEX_FALL_THROUGH-1]) < sizeof(u32))
270 goto errout;
271 cp.fall_through =
272 *(u32 *) RTA_DATA(tb[TCA_TCINDEX_FALL_THROUGH-1]);
273 }
274
275 if (!cp.hash) {
276 /* Hash not specified, use perfect hash if the upper limit
277 * of the hashing index is below the threshold.
278 */
279 if ((cp.mask >> cp.shift) < PERFECT_HASH_THRESHOLD)
280 cp.hash = (cp.mask >> cp.shift)+1;
281 else
282 cp.hash = DEFAULT_HASH_SIZE;
283 }
284
285 if (!cp.perfect && !cp.h)
286 cp.alloc_hash = cp.hash;
287
288 /* Note: this could be as restrictive as if (handle & ~(mask >> shift))
289 * but then, we'd fail handles that may become valid after some future
290 * mask change. While this is extremely unlikely to ever matter,
291 * the check below is safer (and also more backwards-compatible).
292 */
293 if (cp.perfect || valid_perfect_hash(&cp))
294 if (handle >= cp.alloc_hash)
295 goto errout;
296
297
298 err = -ENOMEM;
299 if (!cp.perfect && !cp.h) {
300 if (valid_perfect_hash(&cp)) {
301 cp.perfect = kmalloc(cp.hash * sizeof(*r), GFP_KERNEL);
302 if (!cp.perfect)
303 goto errout;
304 memset(cp.perfect, 0, cp.hash * sizeof(*r));
305 balloc = 1;
306 } else {
307 cp.h = kmalloc(cp.hash * sizeof(f), GFP_KERNEL);
308 if (!cp.h)
309 goto errout;
310 memset(cp.h, 0, cp.hash * sizeof(f));
311 balloc = 2;
312 }
313 }
314
315 if (cp.perfect)
316 r = cp.perfect + handle;
317 else
318 r = tcindex_lookup(&cp, handle) ? : &new_filter_result;
319
320 if (r == &new_filter_result) {
321 f = kmalloc(sizeof(*f), GFP_KERNEL);
322 if (!f)
323 goto errout_alloc;
324 memset(f, 0, sizeof(*f));
325 }
326
327 if (tb[TCA_TCINDEX_CLASSID-1]) {
328 cr.res.classid = *(u32 *) RTA_DATA(tb[TCA_TCINDEX_CLASSID-1]);
329 tcf_bind_filter(tp, &cr.res, base);
330 }
331
332 tcf_exts_change(tp, &cr.exts, &e);
333
334 tcf_tree_lock(tp);
335 if (old_r && old_r != r)
336 memset(old_r, 0, sizeof(*old_r));
337
338 memcpy(p, &cp, sizeof(cp));
339 memcpy(r, &cr, sizeof(cr));
340
341 if (r == &new_filter_result) {
342 struct tcindex_filter **fp;
343
344 f->key = handle;
345 f->result = new_filter_result;
346 f->next = NULL;
347 for (fp = p->h+(handle % p->hash); *fp; fp = &(*fp)->next)
348 /* nothing */;
349 *fp = f;
350 }
351 tcf_tree_unlock(tp);
352
353 return 0;
354
355errout_alloc:
356 if (balloc == 1)
357 kfree(cp.perfect);
358 else if (balloc == 2)
359 kfree(cp.h);
360errout:
361 tcf_exts_destroy(tp, &e);
362 return err;
363}
364
365static int
366tcindex_change(struct tcf_proto *tp, unsigned long base, u32 handle,
367 struct rtattr **tca, unsigned long *arg)
368{
369 struct rtattr *opt = tca[TCA_OPTIONS-1];
370 struct rtattr *tb[TCA_TCINDEX_MAX];
371 struct tcindex_data *p = PRIV(tp);
372 struct tcindex_filter_result *r = (struct tcindex_filter_result *) *arg;
373
374 DPRINTK("tcindex_change(tp %p,handle 0x%08x,tca %p,arg %p),opt %p,"
375 "p %p,r %p,*arg 0x%lx\n",
376 tp, handle, tca, arg, opt, p, r, arg ? *arg : 0L);
377
378 if (!opt)
379 return 0;
380
381 if (rtattr_parse_nested(tb, TCA_TCINDEX_MAX, opt) < 0)
382 return -EINVAL;
383
384 return tcindex_set_parms(tp, base, handle, p, r, tb, tca[TCA_RATE-1]);
385}
386
387
388static void tcindex_walk(struct tcf_proto *tp, struct tcf_walker *walker)
389{
390 struct tcindex_data *p = PRIV(tp);
391 struct tcindex_filter *f,*next;
392 int i;
393
394 DPRINTK("tcindex_walk(tp %p,walker %p),p %p\n",tp,walker,p);
395 if (p->perfect) {
396 for (i = 0; i < p->hash; i++) {
397 if (!p->perfect[i].res.class)
398 continue;
399 if (walker->count >= walker->skip) {
400 if (walker->fn(tp,
401 (unsigned long) (p->perfect+i), walker)
402 < 0) {
403 walker->stop = 1;
404 return;
405 }
406 }
407 walker->count++;
408 }
409 }
410 if (!p->h)
411 return;
412 for (i = 0; i < p->hash; i++) {
413 for (f = p->h[i]; f; f = next) {
414 next = f->next;
415 if (walker->count >= walker->skip) {
416 if (walker->fn(tp,(unsigned long) &f->result,
417 walker) < 0) {
418 walker->stop = 1;
419 return;
420 }
421 }
422 walker->count++;
423 }
424 }
425}
426
427
428static int tcindex_destroy_element(struct tcf_proto *tp,
429 unsigned long arg, struct tcf_walker *walker)
430{
431 return __tcindex_delete(tp, arg, 0);
432}
433
434
435static void tcindex_destroy(struct tcf_proto *tp)
436{
437 struct tcindex_data *p = PRIV(tp);
438 struct tcf_walker walker;
439
440 DPRINTK("tcindex_destroy(tp %p),p %p\n",tp,p);
441 walker.count = 0;
442 walker.skip = 0;
443 walker.fn = &tcindex_destroy_element;
444 tcindex_walk(tp,&walker);
445 if (p->perfect)
446 kfree(p->perfect);
447 if (p->h)
448 kfree(p->h);
449 kfree(p);
450 tp->root = NULL;
451}
452
453
454static int tcindex_dump(struct tcf_proto *tp, unsigned long fh,
455 struct sk_buff *skb, struct tcmsg *t)
456{
457 struct tcindex_data *p = PRIV(tp);
458 struct tcindex_filter_result *r = (struct tcindex_filter_result *) fh;
459 unsigned char *b = skb->tail;
460 struct rtattr *rta;
461
462 DPRINTK("tcindex_dump(tp %p,fh 0x%lx,skb %p,t %p),p %p,r %p,b %p\n",
463 tp,fh,skb,t,p,r,b);
464 DPRINTK("p->perfect %p p->h %p\n",p->perfect,p->h);
465 rta = (struct rtattr *) b;
466 RTA_PUT(skb,TCA_OPTIONS,0,NULL);
467 if (!fh) {
468 t->tcm_handle = ~0; /* whatever ... */
469 RTA_PUT(skb,TCA_TCINDEX_HASH,sizeof(p->hash),&p->hash);
470 RTA_PUT(skb,TCA_TCINDEX_MASK,sizeof(p->mask),&p->mask);
471 RTA_PUT(skb,TCA_TCINDEX_SHIFT,sizeof(p->shift),&p->shift);
472 RTA_PUT(skb,TCA_TCINDEX_FALL_THROUGH,sizeof(p->fall_through),
473 &p->fall_through);
474 rta->rta_len = skb->tail-b;
475 } else {
476 if (p->perfect) {
477 t->tcm_handle = r-p->perfect;
478 } else {
479 struct tcindex_filter *f;
480 int i;
481
482 t->tcm_handle = 0;
483 for (i = 0; !t->tcm_handle && i < p->hash; i++) {
484 for (f = p->h[i]; !t->tcm_handle && f;
485 f = f->next) {
486 if (&f->result == r)
487 t->tcm_handle = f->key;
488 }
489 }
490 }
491 DPRINTK("handle = %d\n",t->tcm_handle);
492 if (r->res.class)
493 RTA_PUT(skb, TCA_TCINDEX_CLASSID, 4, &r->res.classid);
494
495 if (tcf_exts_dump(skb, &r->exts, &tcindex_ext_map) < 0)
496 goto rtattr_failure;
497 rta->rta_len = skb->tail-b;
498
499 if (tcf_exts_dump_stats(skb, &r->exts, &tcindex_ext_map) < 0)
500 goto rtattr_failure;
501 }
502
503 return skb->len;
504
505rtattr_failure:
506 skb_trim(skb, b - skb->data);
507 return -1;
508}
509
510static struct tcf_proto_ops cls_tcindex_ops = {
511 .next = NULL,
512 .kind = "tcindex",
513 .classify = tcindex_classify,
514 .init = tcindex_init,
515 .destroy = tcindex_destroy,
516 .get = tcindex_get,
517 .put = tcindex_put,
518 .change = tcindex_change,
519 .delete = tcindex_delete,
520 .walk = tcindex_walk,
521 .dump = tcindex_dump,
522 .owner = THIS_MODULE,
523};
524
525static int __init init_tcindex(void)
526{
527 return register_tcf_proto_ops(&cls_tcindex_ops);
528}
529
530static void __exit exit_tcindex(void)
531{
532 unregister_tcf_proto_ops(&cls_tcindex_ops);
533}
534
535module_init(init_tcindex)
536module_exit(exit_tcindex)
537MODULE_LICENSE("GPL");