blob: 5af8a59e1503c7d5a90cf332af72ee5921d94950 [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
Linus Torvalds1da177e2005-04-16 15:20:36 -07007#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
50struct tcindex_filter_result {
51 struct tcf_exts exts;
52 struct tcf_result res;
53};
54
55struct tcindex_filter {
56 u16 key;
57 struct tcindex_filter_result result;
58 struct tcindex_filter *next;
59};
60
61
62struct 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
73static struct tcf_ext_map tcindex_ext_map = {
74 .police = TCA_TCINDEX_POLICE,
75 .action = TCA_TCINDEX_ACT
76};
77
78static inline int
79tcindex_filter_is_set(struct tcindex_filter_result *r)
80{
81 return tcf_exts_is_predicative(&r->exts) || r->res.classid;
82}
83
84static struct tcindex_filter_result *
85tcindex_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
102static 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
127static 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
140static 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
146static int tcindex_init(struct tcf_proto *tp)
147{
148 struct tcindex_data *p;
149
150 DPRINTK("tcindex_init(tp %p)\n",tp);
Panagiotis Issaris0da974f2006-07-21 14:51:30 -0700151 p = kzalloc(sizeof(struct tcindex_data),GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 if (!p)
153 return -ENOMEM;
154
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 p->mask = 0xffff;
156 p->hash = DEFAULT_HASH_SIZE;
157 p->fall_through = 1;
158
159 tp->root = p;
160 return 0;
161}
162
163
164static int
165__tcindex_delete(struct tcf_proto *tp, unsigned long arg, int lock)
166{
167 struct tcindex_data *p = PRIV(tp);
168 struct tcindex_filter_result *r = (struct tcindex_filter_result *) arg;
169 struct tcindex_filter *f = NULL;
170
171 DPRINTK("tcindex_delete(tp %p,arg 0x%lx),p %p,f %p\n",tp,arg,p,f);
172 if (p->perfect) {
173 if (!r->res.class)
174 return -ENOENT;
175 } else {
176 int i;
177 struct tcindex_filter **walk = NULL;
178
179 for (i = 0; i < p->hash; i++)
180 for (walk = p->h+i; *walk; walk = &(*walk)->next)
181 if (&(*walk)->result == r)
182 goto found;
183 return -ENOENT;
184
185found:
186 f = *walk;
187 if (lock)
188 tcf_tree_lock(tp);
189 *walk = f->next;
190 if (lock)
191 tcf_tree_unlock(tp);
192 }
193 tcf_unbind_filter(tp, &r->res);
194 tcf_exts_destroy(tp, &r->exts);
Jesper Juhla51482b2005-11-08 09:41:34 -0800195 kfree(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 return 0;
197}
198
199static int tcindex_delete(struct tcf_proto *tp, unsigned long arg)
200{
201 return __tcindex_delete(tp, arg, 1);
202}
203
204static inline int
205valid_perfect_hash(struct tcindex_data *p)
206{
207 return p->hash > (p->mask >> p->shift);
208}
209
210static int
211tcindex_set_parms(struct tcf_proto *tp, unsigned long base, u32 handle,
212 struct tcindex_data *p, struct tcindex_filter_result *r,
213 struct rtattr **tb, struct rtattr *est)
214{
215 int err, balloc = 0;
216 struct tcindex_filter_result new_filter_result, *old_r = r;
217 struct tcindex_filter_result cr;
218 struct tcindex_data cp;
219 struct tcindex_filter *f = NULL; /* make gcc behave */
220 struct tcf_exts e;
221
222 err = tcf_exts_validate(tp, tb, est, &e, &tcindex_ext_map);
223 if (err < 0)
224 return err;
225
226 memcpy(&cp, p, sizeof(cp));
227 memset(&new_filter_result, 0, sizeof(new_filter_result));
228
229 if (old_r)
230 memcpy(&cr, r, sizeof(cr));
231 else
232 memset(&cr, 0, sizeof(cr));
233
234 err = -EINVAL;
235 if (tb[TCA_TCINDEX_HASH-1]) {
236 if (RTA_PAYLOAD(tb[TCA_TCINDEX_HASH-1]) < sizeof(u32))
237 goto errout;
238 cp.hash = *(u32 *) RTA_DATA(tb[TCA_TCINDEX_HASH-1]);
239 }
240
241 if (tb[TCA_TCINDEX_MASK-1]) {
242 if (RTA_PAYLOAD(tb[TCA_TCINDEX_MASK-1]) < sizeof(u16))
243 goto errout;
244 cp.mask = *(u16 *) RTA_DATA(tb[TCA_TCINDEX_MASK-1]);
245 }
246
247 if (tb[TCA_TCINDEX_SHIFT-1]) {
248 if (RTA_PAYLOAD(tb[TCA_TCINDEX_SHIFT-1]) < sizeof(u16))
249 goto errout;
250 cp.shift = *(u16 *) RTA_DATA(tb[TCA_TCINDEX_SHIFT-1]);
251 }
252
253 err = -EBUSY;
254 /* Hash already allocated, make sure that we still meet the
255 * requirements for the allocated hash.
256 */
257 if (cp.perfect) {
258 if (!valid_perfect_hash(&cp) ||
259 cp.hash > cp.alloc_hash)
260 goto errout;
261 } else if (cp.h && cp.hash != cp.alloc_hash)
262 goto errout;
263
264 err = -EINVAL;
265 if (tb[TCA_TCINDEX_FALL_THROUGH-1]) {
266 if (RTA_PAYLOAD(tb[TCA_TCINDEX_FALL_THROUGH-1]) < sizeof(u32))
267 goto errout;
268 cp.fall_through =
269 *(u32 *) RTA_DATA(tb[TCA_TCINDEX_FALL_THROUGH-1]);
270 }
271
272 if (!cp.hash) {
273 /* Hash not specified, use perfect hash if the upper limit
274 * of the hashing index is below the threshold.
275 */
276 if ((cp.mask >> cp.shift) < PERFECT_HASH_THRESHOLD)
277 cp.hash = (cp.mask >> cp.shift)+1;
278 else
279 cp.hash = DEFAULT_HASH_SIZE;
280 }
281
282 if (!cp.perfect && !cp.h)
283 cp.alloc_hash = cp.hash;
284
285 /* Note: this could be as restrictive as if (handle & ~(mask >> shift))
286 * but then, we'd fail handles that may become valid after some future
287 * mask change. While this is extremely unlikely to ever matter,
288 * the check below is safer (and also more backwards-compatible).
289 */
290 if (cp.perfect || valid_perfect_hash(&cp))
291 if (handle >= cp.alloc_hash)
292 goto errout;
293
294
295 err = -ENOMEM;
296 if (!cp.perfect && !cp.h) {
297 if (valid_perfect_hash(&cp)) {
Panagiotis Issaris0da974f2006-07-21 14:51:30 -0700298 cp.perfect = kcalloc(cp.hash, sizeof(*r), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 if (!cp.perfect)
300 goto errout;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 balloc = 1;
302 } else {
Panagiotis Issaris0da974f2006-07-21 14:51:30 -0700303 cp.h = kcalloc(cp.hash, sizeof(f), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 if (!cp.h)
305 goto errout;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 balloc = 2;
307 }
308 }
309
310 if (cp.perfect)
311 r = cp.perfect + handle;
312 else
313 r = tcindex_lookup(&cp, handle) ? : &new_filter_result;
314
315 if (r == &new_filter_result) {
Panagiotis Issaris0da974f2006-07-21 14:51:30 -0700316 f = kzalloc(sizeof(*f), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 if (!f)
318 goto errout_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 }
320
321 if (tb[TCA_TCINDEX_CLASSID-1]) {
322 cr.res.classid = *(u32 *) RTA_DATA(tb[TCA_TCINDEX_CLASSID-1]);
323 tcf_bind_filter(tp, &cr.res, base);
324 }
325
326 tcf_exts_change(tp, &cr.exts, &e);
327
328 tcf_tree_lock(tp);
329 if (old_r && old_r != r)
330 memset(old_r, 0, sizeof(*old_r));
331
332 memcpy(p, &cp, sizeof(cp));
333 memcpy(r, &cr, sizeof(cr));
334
335 if (r == &new_filter_result) {
336 struct tcindex_filter **fp;
337
338 f->key = handle;
339 f->result = new_filter_result;
340 f->next = NULL;
341 for (fp = p->h+(handle % p->hash); *fp; fp = &(*fp)->next)
342 /* nothing */;
343 *fp = f;
344 }
345 tcf_tree_unlock(tp);
346
347 return 0;
348
349errout_alloc:
350 if (balloc == 1)
351 kfree(cp.perfect);
352 else if (balloc == 2)
353 kfree(cp.h);
354errout:
355 tcf_exts_destroy(tp, &e);
356 return err;
357}
358
359static int
360tcindex_change(struct tcf_proto *tp, unsigned long base, u32 handle,
361 struct rtattr **tca, unsigned long *arg)
362{
363 struct rtattr *opt = tca[TCA_OPTIONS-1];
364 struct rtattr *tb[TCA_TCINDEX_MAX];
365 struct tcindex_data *p = PRIV(tp);
366 struct tcindex_filter_result *r = (struct tcindex_filter_result *) *arg;
367
368 DPRINTK("tcindex_change(tp %p,handle 0x%08x,tca %p,arg %p),opt %p,"
369 "p %p,r %p,*arg 0x%lx\n",
370 tp, handle, tca, arg, opt, p, r, arg ? *arg : 0L);
371
372 if (!opt)
373 return 0;
374
375 if (rtattr_parse_nested(tb, TCA_TCINDEX_MAX, opt) < 0)
376 return -EINVAL;
377
378 return tcindex_set_parms(tp, base, handle, p, r, tb, tca[TCA_RATE-1]);
379}
380
381
382static void tcindex_walk(struct tcf_proto *tp, struct tcf_walker *walker)
383{
384 struct tcindex_data *p = PRIV(tp);
385 struct tcindex_filter *f,*next;
386 int i;
387
388 DPRINTK("tcindex_walk(tp %p,walker %p),p %p\n",tp,walker,p);
389 if (p->perfect) {
390 for (i = 0; i < p->hash; i++) {
391 if (!p->perfect[i].res.class)
392 continue;
393 if (walker->count >= walker->skip) {
394 if (walker->fn(tp,
395 (unsigned long) (p->perfect+i), walker)
396 < 0) {
397 walker->stop = 1;
398 return;
399 }
400 }
401 walker->count++;
402 }
403 }
404 if (!p->h)
405 return;
406 for (i = 0; i < p->hash; i++) {
407 for (f = p->h[i]; f; f = next) {
408 next = f->next;
409 if (walker->count >= walker->skip) {
410 if (walker->fn(tp,(unsigned long) &f->result,
411 walker) < 0) {
412 walker->stop = 1;
413 return;
414 }
415 }
416 walker->count++;
417 }
418 }
419}
420
421
422static int tcindex_destroy_element(struct tcf_proto *tp,
423 unsigned long arg, struct tcf_walker *walker)
424{
425 return __tcindex_delete(tp, arg, 0);
426}
427
428
429static void tcindex_destroy(struct tcf_proto *tp)
430{
431 struct tcindex_data *p = PRIV(tp);
432 struct tcf_walker walker;
433
434 DPRINTK("tcindex_destroy(tp %p),p %p\n",tp,p);
435 walker.count = 0;
436 walker.skip = 0;
437 walker.fn = &tcindex_destroy_element;
438 tcindex_walk(tp,&walker);
Jesper Juhla51482b2005-11-08 09:41:34 -0800439 kfree(p->perfect);
440 kfree(p->h);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 kfree(p);
442 tp->root = NULL;
443}
444
445
446static int tcindex_dump(struct tcf_proto *tp, unsigned long fh,
447 struct sk_buff *skb, struct tcmsg *t)
448{
449 struct tcindex_data *p = PRIV(tp);
450 struct tcindex_filter_result *r = (struct tcindex_filter_result *) fh;
451 unsigned char *b = skb->tail;
452 struct rtattr *rta;
453
454 DPRINTK("tcindex_dump(tp %p,fh 0x%lx,skb %p,t %p),p %p,r %p,b %p\n",
455 tp,fh,skb,t,p,r,b);
456 DPRINTK("p->perfect %p p->h %p\n",p->perfect,p->h);
457 rta = (struct rtattr *) b;
458 RTA_PUT(skb,TCA_OPTIONS,0,NULL);
459 if (!fh) {
460 t->tcm_handle = ~0; /* whatever ... */
461 RTA_PUT(skb,TCA_TCINDEX_HASH,sizeof(p->hash),&p->hash);
462 RTA_PUT(skb,TCA_TCINDEX_MASK,sizeof(p->mask),&p->mask);
463 RTA_PUT(skb,TCA_TCINDEX_SHIFT,sizeof(p->shift),&p->shift);
464 RTA_PUT(skb,TCA_TCINDEX_FALL_THROUGH,sizeof(p->fall_through),
465 &p->fall_through);
466 rta->rta_len = skb->tail-b;
467 } else {
468 if (p->perfect) {
469 t->tcm_handle = r-p->perfect;
470 } else {
471 struct tcindex_filter *f;
472 int i;
473
474 t->tcm_handle = 0;
475 for (i = 0; !t->tcm_handle && i < p->hash; i++) {
476 for (f = p->h[i]; !t->tcm_handle && f;
477 f = f->next) {
478 if (&f->result == r)
479 t->tcm_handle = f->key;
480 }
481 }
482 }
483 DPRINTK("handle = %d\n",t->tcm_handle);
484 if (r->res.class)
485 RTA_PUT(skb, TCA_TCINDEX_CLASSID, 4, &r->res.classid);
486
487 if (tcf_exts_dump(skb, &r->exts, &tcindex_ext_map) < 0)
488 goto rtattr_failure;
489 rta->rta_len = skb->tail-b;
490
491 if (tcf_exts_dump_stats(skb, &r->exts, &tcindex_ext_map) < 0)
492 goto rtattr_failure;
493 }
494
495 return skb->len;
496
497rtattr_failure:
498 skb_trim(skb, b - skb->data);
499 return -1;
500}
501
502static struct tcf_proto_ops cls_tcindex_ops = {
503 .next = NULL,
504 .kind = "tcindex",
505 .classify = tcindex_classify,
506 .init = tcindex_init,
507 .destroy = tcindex_destroy,
508 .get = tcindex_get,
509 .put = tcindex_put,
510 .change = tcindex_change,
511 .delete = tcindex_delete,
512 .walk = tcindex_walk,
513 .dump = tcindex_dump,
514 .owner = THIS_MODULE,
515};
516
517static int __init init_tcindex(void)
518{
519 return register_tcf_proto_ops(&cls_tcindex_ops);
520}
521
522static void __exit exit_tcindex(void)
523{
524 unregister_tcf_proto_ops(&cls_tcindex_ops);
525}
526
527module_init(init_tcindex)
528module_exit(exit_tcindex)
529MODULE_LICENSE("GPL");