580698db578a44333d9c8f7a074300464e8a7039
[safe/jmp/linux-2.6] / net / sched / act_police.c
1 /*
2  * net/sched/police.c   Input police filter.
3  *
4  *              This program is free software; you can redistribute it and/or
5  *              modify it under the terms of the GNU General Public License
6  *              as published by the Free Software Foundation; either version
7  *              2 of the License, or (at your option) any later version.
8  *
9  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10  *              J Hadi Salim (action changes)
11  */
12
13 #include <asm/uaccess.h>
14 #include <asm/system.h>
15 #include <linux/bitops.h>
16 #include <linux/module.h>
17 #include <linux/types.h>
18 #include <linux/kernel.h>
19 #include <linux/string.h>
20 #include <linux/mm.h>
21 #include <linux/socket.h>
22 #include <linux/sockios.h>
23 #include <linux/in.h>
24 #include <linux/errno.h>
25 #include <linux/interrupt.h>
26 #include <linux/netdevice.h>
27 #include <linux/skbuff.h>
28 #include <linux/module.h>
29 #include <linux/rtnetlink.h>
30 #include <linux/init.h>
31 #include <net/sock.h>
32 #include <net/act_api.h>
33 #include <net/netlink.h>
34
35 #define L2T(p,L)   ((p)->tcfp_R_tab->data[(L)>>(p)->tcfp_R_tab->rate.cell_log])
36 #define L2T_P(p,L) ((p)->tcfp_P_tab->data[(L)>>(p)->tcfp_P_tab->rate.cell_log])
37
38 #define POL_TAB_MASK     15
39 static struct tcf_common *tcf_police_ht[POL_TAB_MASK + 1];
40 static u32 police_idx_gen;
41 static DEFINE_RWLOCK(police_lock);
42
43 static struct tcf_hashinfo police_hash_info = {
44         .htab   =       tcf_police_ht,
45         .hmask  =       POL_TAB_MASK,
46         .lock   =       &police_lock,
47 };
48
49 /* old policer structure from before tc actions */
50 struct tc_police_compat
51 {
52         u32                     index;
53         int                     action;
54         u32                     limit;
55         u32                     burst;
56         u32                     mtu;
57         struct tc_ratespec      rate;
58         struct tc_ratespec      peakrate;
59 };
60
61 /* Each policer is serialized by its individual spinlock */
62
63 #ifdef CONFIG_NET_CLS_ACT
64 static int tcf_act_police_walker(struct sk_buff *skb, struct netlink_callback *cb,
65                               int type, struct tc_action *a)
66 {
67         struct tcf_common *p;
68         int err = 0, index = -1, i = 0, s_i = 0, n_i = 0;
69         struct rtattr *r;
70
71         read_lock(&police_lock);
72
73         s_i = cb->args[0];
74
75         for (i = 0; i < (POL_TAB_MASK + 1); i++) {
76                 p = tcf_police_ht[tcf_hash(i, POL_TAB_MASK)];
77
78                 for (; p; p = p->tcfc_next) {
79                         index++;
80                         if (index < s_i)
81                                 continue;
82                         a->priv = p;
83                         a->order = index;
84                         r = (struct rtattr *)skb_tail_pointer(skb);
85                         RTA_PUT(skb, a->order, 0, NULL);
86                         if (type == RTM_DELACTION)
87                                 err = tcf_action_dump_1(skb, a, 0, 1);
88                         else
89                                 err = tcf_action_dump_1(skb, a, 0, 0);
90                         if (err < 0) {
91                                 index--;
92                                 nlmsg_trim(skb, r);
93                                 goto done;
94                         }
95                         r->rta_len = skb_tail_pointer(skb) - (u8 *)r;
96                         n_i++;
97                 }
98         }
99 done:
100         read_unlock(&police_lock);
101         if (n_i)
102                 cb->args[0] += n_i;
103         return n_i;
104
105 rtattr_failure:
106         nlmsg_trim(skb, r);
107         goto done;
108 }
109 #endif
110
111 void tcf_police_destroy(struct tcf_police *p)
112 {
113         unsigned int h = tcf_hash(p->tcf_index, POL_TAB_MASK);
114         struct tcf_common **p1p;
115
116         for (p1p = &tcf_police_ht[h]; *p1p; p1p = &(*p1p)->tcfc_next) {
117                 if (*p1p == &p->common) {
118                         write_lock_bh(&police_lock);
119                         *p1p = p->tcf_next;
120                         write_unlock_bh(&police_lock);
121                         gen_kill_estimator(&p->tcf_bstats,
122                                            &p->tcf_rate_est);
123                         if (p->tcfp_R_tab)
124                                 qdisc_put_rtab(p->tcfp_R_tab);
125                         if (p->tcfp_P_tab)
126                                 qdisc_put_rtab(p->tcfp_P_tab);
127                         kfree(p);
128                         return;
129                 }
130         }
131         BUG_TRAP(0);
132 }
133
134 #ifdef CONFIG_NET_CLS_ACT
135 static int tcf_act_police_locate(struct rtattr *rta, struct rtattr *est,
136                                  struct tc_action *a, int ovr, int bind)
137 {
138         unsigned h;
139         int ret = 0, err;
140         struct rtattr *tb[TCA_POLICE_MAX];
141         struct tc_police *parm;
142         struct tcf_police *police;
143         struct qdisc_rate_table *R_tab = NULL, *P_tab = NULL;
144         int size;
145
146         if (rta == NULL || rtattr_parse_nested(tb, TCA_POLICE_MAX, rta) < 0)
147                 return -EINVAL;
148
149         if (tb[TCA_POLICE_TBF-1] == NULL)
150                 return -EINVAL;
151         size = RTA_PAYLOAD(tb[TCA_POLICE_TBF-1]);
152         if (size != sizeof(*parm) && size != sizeof(struct tc_police_compat))
153                 return -EINVAL;
154         parm = RTA_DATA(tb[TCA_POLICE_TBF-1]);
155
156         if (tb[TCA_POLICE_RESULT-1] != NULL &&
157             RTA_PAYLOAD(tb[TCA_POLICE_RESULT-1]) != sizeof(u32))
158                 return -EINVAL;
159         if (tb[TCA_POLICE_RESULT-1] != NULL &&
160             RTA_PAYLOAD(tb[TCA_POLICE_RESULT-1]) != sizeof(u32))
161                 return -EINVAL;
162
163         if (parm->index) {
164                 struct tcf_common *pc;
165
166                 pc = tcf_hash_lookup(parm->index, &police_hash_info);
167                 if (pc != NULL) {
168                         a->priv = pc;
169                         police = to_police(pc);
170                         if (bind) {
171                                 police->tcf_bindcnt += 1;
172                                 police->tcf_refcnt += 1;
173                         }
174                         if (ovr)
175                                 goto override;
176                         return ret;
177                 }
178         }
179
180         police = kzalloc(sizeof(*police), GFP_KERNEL);
181         if (police == NULL)
182                 return -ENOMEM;
183         ret = ACT_P_CREATED;
184         police->tcf_refcnt = 1;
185         spin_lock_init(&police->tcf_lock);
186         police->tcf_stats_lock = &police->tcf_lock;
187         if (bind)
188                 police->tcf_bindcnt = 1;
189 override:
190         if (parm->rate.rate) {
191                 err = -ENOMEM;
192                 R_tab = qdisc_get_rtab(&parm->rate, tb[TCA_POLICE_RATE-1]);
193                 if (R_tab == NULL)
194                         goto failure;
195                 if (parm->peakrate.rate) {
196                         P_tab = qdisc_get_rtab(&parm->peakrate,
197                                                tb[TCA_POLICE_PEAKRATE-1]);
198                         if (P_tab == NULL) {
199                                 qdisc_put_rtab(R_tab);
200                                 goto failure;
201                         }
202                 }
203         }
204         /* No failure allowed after this point */
205         spin_lock_bh(&police->tcf_lock);
206         if (R_tab != NULL) {
207                 qdisc_put_rtab(police->tcfp_R_tab);
208                 police->tcfp_R_tab = R_tab;
209         }
210         if (P_tab != NULL) {
211                 qdisc_put_rtab(police->tcfp_P_tab);
212                 police->tcfp_P_tab = P_tab;
213         }
214
215         if (tb[TCA_POLICE_RESULT-1])
216                 police->tcfp_result = *(u32*)RTA_DATA(tb[TCA_POLICE_RESULT-1]);
217         police->tcfp_toks = police->tcfp_burst = parm->burst;
218         police->tcfp_mtu = parm->mtu;
219         if (police->tcfp_mtu == 0) {
220                 police->tcfp_mtu = ~0;
221                 if (police->tcfp_R_tab)
222                         police->tcfp_mtu = 255<<police->tcfp_R_tab->rate.cell_log;
223         }
224         if (police->tcfp_P_tab)
225                 police->tcfp_ptoks = L2T_P(police, police->tcfp_mtu);
226         police->tcf_action = parm->action;
227
228         if (tb[TCA_POLICE_AVRATE-1])
229                 police->tcfp_ewma_rate =
230                         *(u32*)RTA_DATA(tb[TCA_POLICE_AVRATE-1]);
231         if (est)
232                 gen_replace_estimator(&police->tcf_bstats,
233                                       &police->tcf_rate_est,
234                                       police->tcf_stats_lock, est);
235
236         spin_unlock_bh(&police->tcf_lock);
237         if (ret != ACT_P_CREATED)
238                 return ret;
239
240         police->tcfp_t_c = psched_get_time();
241         police->tcf_index = parm->index ? parm->index :
242                 tcf_hash_new_index(&police_idx_gen, &police_hash_info);
243         h = tcf_hash(police->tcf_index, POL_TAB_MASK);
244         write_lock_bh(&police_lock);
245         police->tcf_next = tcf_police_ht[h];
246         tcf_police_ht[h] = &police->common;
247         write_unlock_bh(&police_lock);
248
249         a->priv = police;
250         return ret;
251
252 failure:
253         if (ret == ACT_P_CREATED)
254                 kfree(police);
255         return err;
256 }
257
258 static int tcf_act_police_cleanup(struct tc_action *a, int bind)
259 {
260         struct tcf_police *p = a->priv;
261
262         if (p != NULL)
263                 return tcf_police_release(p, bind);
264         return 0;
265 }
266
267 static int tcf_act_police(struct sk_buff *skb, struct tc_action *a,
268                           struct tcf_result *res)
269 {
270         struct tcf_police *police = a->priv;
271         psched_time_t now;
272         long toks;
273         long ptoks = 0;
274
275         spin_lock(&police->tcf_lock);
276
277         police->tcf_bstats.bytes += skb->len;
278         police->tcf_bstats.packets++;
279
280         if (police->tcfp_ewma_rate &&
281             police->tcf_rate_est.bps >= police->tcfp_ewma_rate) {
282                 police->tcf_qstats.overlimits++;
283                 spin_unlock(&police->tcf_lock);
284                 return police->tcf_action;
285         }
286
287         if (skb->len <= police->tcfp_mtu) {
288                 if (police->tcfp_R_tab == NULL) {
289                         spin_unlock(&police->tcf_lock);
290                         return police->tcfp_result;
291                 }
292
293                 now = psched_get_time();
294                 toks = psched_tdiff_bounded(now, police->tcfp_t_c,
295                                             police->tcfp_burst);
296                 if (police->tcfp_P_tab) {
297                         ptoks = toks + police->tcfp_ptoks;
298                         if (ptoks > (long)L2T_P(police, police->tcfp_mtu))
299                                 ptoks = (long)L2T_P(police, police->tcfp_mtu);
300                         ptoks -= L2T_P(police, skb->len);
301                 }
302                 toks += police->tcfp_toks;
303                 if (toks > (long)police->tcfp_burst)
304                         toks = police->tcfp_burst;
305                 toks -= L2T(police, skb->len);
306                 if ((toks|ptoks) >= 0) {
307                         police->tcfp_t_c = now;
308                         police->tcfp_toks = toks;
309                         police->tcfp_ptoks = ptoks;
310                         spin_unlock(&police->tcf_lock);
311                         return police->tcfp_result;
312                 }
313         }
314
315         police->tcf_qstats.overlimits++;
316         spin_unlock(&police->tcf_lock);
317         return police->tcf_action;
318 }
319
320 static int
321 tcf_act_police_dump(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
322 {
323         unsigned char *b = skb_tail_pointer(skb);
324         struct tcf_police *police = a->priv;
325         struct tc_police opt;
326
327         opt.index = police->tcf_index;
328         opt.action = police->tcf_action;
329         opt.mtu = police->tcfp_mtu;
330         opt.burst = police->tcfp_burst;
331         opt.refcnt = police->tcf_refcnt - ref;
332         opt.bindcnt = police->tcf_bindcnt - bind;
333         if (police->tcfp_R_tab)
334                 opt.rate = police->tcfp_R_tab->rate;
335         else
336                 memset(&opt.rate, 0, sizeof(opt.rate));
337         if (police->tcfp_P_tab)
338                 opt.peakrate = police->tcfp_P_tab->rate;
339         else
340                 memset(&opt.peakrate, 0, sizeof(opt.peakrate));
341         RTA_PUT(skb, TCA_POLICE_TBF, sizeof(opt), &opt);
342         if (police->tcfp_result)
343                 RTA_PUT(skb, TCA_POLICE_RESULT, sizeof(int),
344                         &police->tcfp_result);
345         if (police->tcfp_ewma_rate)
346                 RTA_PUT(skb, TCA_POLICE_AVRATE, 4, &police->tcfp_ewma_rate);
347         return skb->len;
348
349 rtattr_failure:
350         nlmsg_trim(skb, b);
351         return -1;
352 }
353
354 MODULE_AUTHOR("Alexey Kuznetsov");
355 MODULE_DESCRIPTION("Policing actions");
356 MODULE_LICENSE("GPL");
357
358 static struct tc_action_ops act_police_ops = {
359         .kind           =       "police",
360         .hinfo          =       &police_hash_info,
361         .type           =       TCA_ID_POLICE,
362         .capab          =       TCA_CAP_NONE,
363         .owner          =       THIS_MODULE,
364         .act            =       tcf_act_police,
365         .dump           =       tcf_act_police_dump,
366         .cleanup        =       tcf_act_police_cleanup,
367         .lookup         =       tcf_hash_search,
368         .init           =       tcf_act_police_locate,
369         .walk           =       tcf_act_police_walker
370 };
371
372 static int __init
373 police_init_module(void)
374 {
375         return tcf_register_action(&act_police_ops);
376 }
377
378 static void __exit
379 police_cleanup_module(void)
380 {
381         tcf_unregister_action(&act_police_ops);
382 }
383
384 module_init(police_init_module);
385 module_exit(police_cleanup_module);
386
387 #else /* CONFIG_NET_CLS_ACT */
388
389 static struct tcf_common *tcf_police_lookup(u32 index)
390 {
391         struct tcf_hashinfo *hinfo = &police_hash_info;
392         struct tcf_common *p;
393
394         read_lock(hinfo->lock);
395         for (p = hinfo->htab[tcf_hash(index, hinfo->hmask)]; p;
396              p = p->tcfc_next) {
397                 if (p->tcfc_index == index)
398                         break;
399         }
400         read_unlock(hinfo->lock);
401
402         return p;
403 }
404
405 static u32 tcf_police_new_index(void)
406 {
407         u32 *idx_gen = &police_idx_gen;
408         u32 val = *idx_gen;
409
410         do {
411                 if (++val == 0)
412                         val = 1;
413         } while (tcf_police_lookup(val));
414
415         return (*idx_gen = val);
416 }
417
418 struct tcf_police *tcf_police_locate(struct rtattr *rta, struct rtattr *est)
419 {
420         unsigned int h;
421         struct tcf_police *police;
422         struct rtattr *tb[TCA_POLICE_MAX];
423         struct tc_police *parm;
424         int size;
425
426         if (rtattr_parse_nested(tb, TCA_POLICE_MAX, rta) < 0)
427                 return NULL;
428
429         if (tb[TCA_POLICE_TBF-1] == NULL)
430                 return NULL;
431         size = RTA_PAYLOAD(tb[TCA_POLICE_TBF-1]);
432         if (size != sizeof(*parm) && size != sizeof(struct tc_police_compat))
433                 return NULL;
434
435         parm = RTA_DATA(tb[TCA_POLICE_TBF-1]);
436
437         if (parm->index) {
438                 struct tcf_common *pc;
439
440                 pc = tcf_police_lookup(parm->index);
441                 if (pc) {
442                         police = to_police(pc);
443                         police->tcf_refcnt++;
444                         return police;
445                 }
446         }
447         police = kzalloc(sizeof(*police), GFP_KERNEL);
448         if (unlikely(!police))
449                 return NULL;
450
451         police->tcf_refcnt = 1;
452         spin_lock_init(&police->tcf_lock);
453         police->tcf_stats_lock = &police->tcf_lock;
454         if (parm->rate.rate) {
455                 police->tcfp_R_tab =
456                         qdisc_get_rtab(&parm->rate, tb[TCA_POLICE_RATE-1]);
457                 if (police->tcfp_R_tab == NULL)
458                         goto failure;
459                 if (parm->peakrate.rate) {
460                         police->tcfp_P_tab =
461                                 qdisc_get_rtab(&parm->peakrate,
462                                                tb[TCA_POLICE_PEAKRATE-1]);
463                         if (police->tcfp_P_tab == NULL)
464                                 goto failure;
465                 }
466         }
467         if (tb[TCA_POLICE_RESULT-1]) {
468                 if (RTA_PAYLOAD(tb[TCA_POLICE_RESULT-1]) != sizeof(u32))
469                         goto failure;
470                 police->tcfp_result = *(u32*)RTA_DATA(tb[TCA_POLICE_RESULT-1]);
471         }
472         if (tb[TCA_POLICE_AVRATE-1]) {
473                 if (RTA_PAYLOAD(tb[TCA_POLICE_AVRATE-1]) != sizeof(u32))
474                         goto failure;
475                 police->tcfp_ewma_rate =
476                         *(u32*)RTA_DATA(tb[TCA_POLICE_AVRATE-1]);
477         }
478         police->tcfp_toks = police->tcfp_burst = parm->burst;
479         police->tcfp_mtu = parm->mtu;
480         if (police->tcfp_mtu == 0) {
481                 police->tcfp_mtu = ~0;
482                 if (police->tcfp_R_tab)
483                         police->tcfp_mtu = 255<<police->tcfp_R_tab->rate.cell_log;
484         }
485         if (police->tcfp_P_tab)
486                 police->tcfp_ptoks = L2T_P(police, police->tcfp_mtu);
487         police->tcfp_t_c = psched_get_time();
488         police->tcf_index = parm->index ? parm->index :
489                 tcf_police_new_index();
490         police->tcf_action = parm->action;
491         if (est)
492                 gen_new_estimator(&police->tcf_bstats, &police->tcf_rate_est,
493                                   police->tcf_stats_lock, est);
494         h = tcf_hash(police->tcf_index, POL_TAB_MASK);
495         write_lock_bh(&police_lock);
496         police->tcf_next = tcf_police_ht[h];
497         tcf_police_ht[h] = &police->common;
498         write_unlock_bh(&police_lock);
499         return police;
500
501 failure:
502         if (police->tcfp_R_tab)
503                 qdisc_put_rtab(police->tcfp_R_tab);
504         kfree(police);
505         return NULL;
506 }
507
508 int tcf_police(struct sk_buff *skb, struct tcf_police *police)
509 {
510         psched_time_t now;
511         long toks;
512         long ptoks = 0;
513
514         spin_lock(&police->tcf_lock);
515
516         police->tcf_bstats.bytes += skb->len;
517         police->tcf_bstats.packets++;
518
519         if (police->tcfp_ewma_rate &&
520             police->tcf_rate_est.bps >= police->tcfp_ewma_rate) {
521                 police->tcf_qstats.overlimits++;
522                 spin_unlock(&police->tcf_lock);
523                 return police->tcf_action;
524         }
525         if (skb->len <= police->tcfp_mtu) {
526                 if (police->tcfp_R_tab == NULL) {
527                         spin_unlock(&police->tcf_lock);
528                         return police->tcfp_result;
529                 }
530
531                 now = psched_get_time();
532                 toks = psched_tdiff_bounded(now, police->tcfp_t_c,
533                                             police->tcfp_burst);
534                 if (police->tcfp_P_tab) {
535                         ptoks = toks + police->tcfp_ptoks;
536                         if (ptoks > (long)L2T_P(police, police->tcfp_mtu))
537                                 ptoks = (long)L2T_P(police, police->tcfp_mtu);
538                         ptoks -= L2T_P(police, skb->len);
539                 }
540                 toks += police->tcfp_toks;
541                 if (toks > (long)police->tcfp_burst)
542                         toks = police->tcfp_burst;
543                 toks -= L2T(police, skb->len);
544                 if ((toks|ptoks) >= 0) {
545                         police->tcfp_t_c = now;
546                         police->tcfp_toks = toks;
547                         police->tcfp_ptoks = ptoks;
548                         spin_unlock(&police->tcf_lock);
549                         return police->tcfp_result;
550                 }
551         }
552
553         police->tcf_qstats.overlimits++;
554         spin_unlock(&police->tcf_lock);
555         return police->tcf_action;
556 }
557 EXPORT_SYMBOL(tcf_police);
558
559 int tcf_police_dump(struct sk_buff *skb, struct tcf_police *police)
560 {
561         unsigned char *b = skb_tail_pointer(skb);
562         struct tc_police opt;
563
564         opt.index = police->tcf_index;
565         opt.action = police->tcf_action;
566         opt.mtu = police->tcfp_mtu;
567         opt.burst = police->tcfp_burst;
568         if (police->tcfp_R_tab)
569                 opt.rate = police->tcfp_R_tab->rate;
570         else
571                 memset(&opt.rate, 0, sizeof(opt.rate));
572         if (police->tcfp_P_tab)
573                 opt.peakrate = police->tcfp_P_tab->rate;
574         else
575                 memset(&opt.peakrate, 0, sizeof(opt.peakrate));
576         RTA_PUT(skb, TCA_POLICE_TBF, sizeof(opt), &opt);
577         if (police->tcfp_result)
578                 RTA_PUT(skb, TCA_POLICE_RESULT, sizeof(int),
579                         &police->tcfp_result);
580         if (police->tcfp_ewma_rate)
581                 RTA_PUT(skb, TCA_POLICE_AVRATE, 4, &police->tcfp_ewma_rate);
582         return skb->len;
583
584 rtattr_failure:
585         nlmsg_trim(skb, b);
586         return -1;
587 }
588
589 int tcf_police_dump_stats(struct sk_buff *skb, struct tcf_police *police)
590 {
591         struct gnet_dump d;
592
593         if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
594                                          TCA_XSTATS, police->tcf_stats_lock,
595                                          &d) < 0)
596                 goto errout;
597
598         if (gnet_stats_copy_basic(&d, &police->tcf_bstats) < 0 ||
599             gnet_stats_copy_rate_est(&d, &police->tcf_rate_est) < 0 ||
600             gnet_stats_copy_queue(&d, &police->tcf_qstats) < 0)
601                 goto errout;
602
603         if (gnet_stats_finish_copy(&d) < 0)
604                 goto errout;
605
606         return 0;
607
608 errout:
609         return -1;
610 }
611
612 #endif /* CONFIG_NET_CLS_ACT */