Merge master.kernel.org:/pub/scm/linux/kernel/git/lethal/sh-2.6
[sfrench/cifs-2.6.git] / 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
34 #define L2T(p,L)   ((p)->tcfp_R_tab->data[(L)>>(p)->tcfp_R_tab->rate.cell_log])
35 #define L2T_P(p,L) ((p)->tcfp_P_tab->data[(L)>>(p)->tcfp_P_tab->rate.cell_log])
36
37 #define POL_TAB_MASK     15
38 static struct tcf_common *tcf_police_ht[POL_TAB_MASK + 1];
39 static u32 police_idx_gen;
40 static DEFINE_RWLOCK(police_lock);
41
42 static struct tcf_hashinfo police_hash_info = {
43         .htab   =       tcf_police_ht,
44         .hmask  =       POL_TAB_MASK,
45         .lock   =       &police_lock,
46 };
47
48 /* old policer structure from before tc actions */
49 struct tc_police_compat
50 {
51         u32                     index;
52         int                     action;
53         u32                     limit;
54         u32                     burst;
55         u32                     mtu;
56         struct tc_ratespec      rate;
57         struct tc_ratespec      peakrate;
58 };
59
60 /* Each policer is serialized by its individual spinlock */
61
62 #ifdef CONFIG_NET_CLS_ACT
63 static int tcf_act_police_walker(struct sk_buff *skb, struct netlink_callback *cb,
64                               int type, struct tc_action *a)
65 {
66         struct tcf_common *p;
67         int err = 0, index = -1, i = 0, s_i = 0, n_i = 0;
68         struct rtattr *r;
69
70         read_lock(&police_lock);
71
72         s_i = cb->args[0];
73
74         for (i = 0; i < (POL_TAB_MASK + 1); i++) {
75                 p = tcf_police_ht[tcf_hash(i, POL_TAB_MASK)];
76
77                 for (; p; p = p->tcfc_next) {
78                         index++;
79                         if (index < s_i)
80                                 continue;
81                         a->priv = p;
82                         a->order = index;
83                         r = (struct rtattr*) skb->tail;
84                         RTA_PUT(skb, a->order, 0, NULL);
85                         if (type == RTM_DELACTION)
86                                 err = tcf_action_dump_1(skb, a, 0, 1);
87                         else
88                                 err = tcf_action_dump_1(skb, a, 0, 0);
89                         if (err < 0) {
90                                 index--;
91                                 skb_trim(skb, (u8*)r - skb->data);
92                                 goto done;
93                         }
94                         r->rta_len = skb->tail - (u8*)r;
95                         n_i++;
96                 }
97         }
98 done:
99         read_unlock(&police_lock);
100         if (n_i)
101                 cb->args[0] += n_i;
102         return n_i;
103
104 rtattr_failure:
105         skb_trim(skb, (u8*)r - skb->data);
106         goto done;
107 }
108 #endif
109
110 void tcf_police_destroy(struct tcf_police *p)
111 {
112         unsigned int h = tcf_hash(p->tcf_index, POL_TAB_MASK);
113         struct tcf_common **p1p;
114
115         for (p1p = &tcf_police_ht[h]; *p1p; p1p = &(*p1p)->tcfc_next) {
116                 if (*p1p == &p->common) {
117                         write_lock_bh(&police_lock);
118                         *p1p = p->tcf_next;
119                         write_unlock_bh(&police_lock);
120 #ifdef CONFIG_NET_ESTIMATOR
121                         gen_kill_estimator(&p->tcf_bstats,
122                                            &p->tcf_rate_est);
123 #endif
124                         if (p->tcfp_R_tab)
125                                 qdisc_put_rtab(p->tcfp_R_tab);
126                         if (p->tcfp_P_tab)
127                                 qdisc_put_rtab(p->tcfp_P_tab);
128                         kfree(p);
129                         return;
130                 }
131         }
132         BUG_TRAP(0);
133 }
134
135 #ifdef CONFIG_NET_CLS_ACT
136 static int tcf_act_police_locate(struct rtattr *rta, struct rtattr *est,
137                                  struct tc_action *a, int ovr, int bind)
138 {
139         unsigned h;
140         int ret = 0, err;
141         struct rtattr *tb[TCA_POLICE_MAX];
142         struct tc_police *parm;
143         struct tcf_police *police;
144         struct qdisc_rate_table *R_tab = NULL, *P_tab = NULL;
145         int size;
146
147         if (rta == NULL || rtattr_parse_nested(tb, TCA_POLICE_MAX, rta) < 0)
148                 return -EINVAL;
149
150         if (tb[TCA_POLICE_TBF-1] == NULL)
151                 return -EINVAL;
152         size = RTA_PAYLOAD(tb[TCA_POLICE_TBF-1]);
153         if (size != sizeof(*parm) && size != sizeof(struct tc_police_compat))
154                 return -EINVAL;
155         parm = RTA_DATA(tb[TCA_POLICE_TBF-1]);
156
157         if (tb[TCA_POLICE_RESULT-1] != NULL &&
158             RTA_PAYLOAD(tb[TCA_POLICE_RESULT-1]) != sizeof(u32))
159                 return -EINVAL;
160         if (tb[TCA_POLICE_RESULT-1] != NULL &&
161             RTA_PAYLOAD(tb[TCA_POLICE_RESULT-1]) != sizeof(u32))
162                 return -EINVAL;
163
164         if (parm->index) {
165                 struct tcf_common *pc;
166
167                 pc = tcf_hash_lookup(parm->index, &police_hash_info);
168                 if (pc != NULL) {
169                         a->priv = pc;
170                         police = to_police(pc);
171                         if (bind) {
172                                 police->tcf_bindcnt += 1;
173                                 police->tcf_refcnt += 1;
174                         }
175                         if (ovr)
176                                 goto override;
177                         return ret;
178                 }
179         }
180
181         police = kzalloc(sizeof(*police), GFP_KERNEL);
182         if (police == NULL)
183                 return -ENOMEM;
184         ret = ACT_P_CREATED;
185         police->tcf_refcnt = 1;
186         spin_lock_init(&police->tcf_lock);
187         police->tcf_stats_lock = &police->tcf_lock;
188         if (bind)
189                 police->tcf_bindcnt = 1;
190 override:
191         if (parm->rate.rate) {
192                 err = -ENOMEM;
193                 R_tab = qdisc_get_rtab(&parm->rate, tb[TCA_POLICE_RATE-1]);
194                 if (R_tab == NULL)
195                         goto failure;
196                 if (parm->peakrate.rate) {
197                         P_tab = qdisc_get_rtab(&parm->peakrate,
198                                                tb[TCA_POLICE_PEAKRATE-1]);
199                         if (P_tab == NULL) {
200                                 qdisc_put_rtab(R_tab);
201                                 goto failure;
202                         }
203                 }
204         }
205         /* No failure allowed after this point */
206         spin_lock_bh(&police->tcf_lock);
207         if (R_tab != NULL) {
208                 qdisc_put_rtab(police->tcfp_R_tab);
209                 police->tcfp_R_tab = R_tab;
210         }
211         if (P_tab != NULL) {
212                 qdisc_put_rtab(police->tcfp_P_tab);
213                 police->tcfp_P_tab = P_tab;
214         }
215
216         if (tb[TCA_POLICE_RESULT-1])
217                 police->tcfp_result = *(u32*)RTA_DATA(tb[TCA_POLICE_RESULT-1]);
218         police->tcfp_toks = police->tcfp_burst = parm->burst;
219         police->tcfp_mtu = parm->mtu;
220         if (police->tcfp_mtu == 0) {
221                 police->tcfp_mtu = ~0;
222                 if (police->tcfp_R_tab)
223                         police->tcfp_mtu = 255<<police->tcfp_R_tab->rate.cell_log;
224         }
225         if (police->tcfp_P_tab)
226                 police->tcfp_ptoks = L2T_P(police, police->tcfp_mtu);
227         police->tcf_action = parm->action;
228
229 #ifdef CONFIG_NET_ESTIMATOR
230         if (tb[TCA_POLICE_AVRATE-1])
231                 police->tcfp_ewma_rate =
232                         *(u32*)RTA_DATA(tb[TCA_POLICE_AVRATE-1]);
233         if (est)
234                 gen_replace_estimator(&police->tcf_bstats,
235                                       &police->tcf_rate_est,
236                                       police->tcf_stats_lock, est);
237 #endif
238
239         spin_unlock_bh(&police->tcf_lock);
240         if (ret != ACT_P_CREATED)
241                 return ret;
242
243         PSCHED_GET_TIME(police->tcfp_t_c);
244         police->tcf_index = parm->index ? parm->index :
245                 tcf_hash_new_index(&police_idx_gen, &police_hash_info);
246         h = tcf_hash(police->tcf_index, POL_TAB_MASK);
247         write_lock_bh(&police_lock);
248         police->tcf_next = tcf_police_ht[h];
249         tcf_police_ht[h] = &police->common;
250         write_unlock_bh(&police_lock);
251
252         a->priv = police;
253         return ret;
254
255 failure:
256         if (ret == ACT_P_CREATED)
257                 kfree(police);
258         return err;
259 }
260
261 static int tcf_act_police_cleanup(struct tc_action *a, int bind)
262 {
263         struct tcf_police *p = a->priv;
264
265         if (p != NULL)
266                 return tcf_police_release(p, bind);
267         return 0;
268 }
269
270 static int tcf_act_police(struct sk_buff *skb, struct tc_action *a,
271                           struct tcf_result *res)
272 {
273         struct tcf_police *police = a->priv;
274         psched_time_t now;
275         long toks;
276         long ptoks = 0;
277
278         spin_lock(&police->tcf_lock);
279
280         police->tcf_bstats.bytes += skb->len;
281         police->tcf_bstats.packets++;
282
283 #ifdef CONFIG_NET_ESTIMATOR
284         if (police->tcfp_ewma_rate &&
285             police->tcf_rate_est.bps >= police->tcfp_ewma_rate) {
286                 police->tcf_qstats.overlimits++;
287                 spin_unlock(&police->tcf_lock);
288                 return police->tcf_action;
289         }
290 #endif
291
292         if (skb->len <= police->tcfp_mtu) {
293                 if (police->tcfp_R_tab == NULL) {
294                         spin_unlock(&police->tcf_lock);
295                         return police->tcfp_result;
296                 }
297
298                 PSCHED_GET_TIME(now);
299
300                 toks = PSCHED_TDIFF_SAFE(now, police->tcfp_t_c,
301                                          police->tcfp_burst);
302                 if (police->tcfp_P_tab) {
303                         ptoks = toks + police->tcfp_ptoks;
304                         if (ptoks > (long)L2T_P(police, police->tcfp_mtu))
305                                 ptoks = (long)L2T_P(police, police->tcfp_mtu);
306                         ptoks -= L2T_P(police, skb->len);
307                 }
308                 toks += police->tcfp_toks;
309                 if (toks > (long)police->tcfp_burst)
310                         toks = police->tcfp_burst;
311                 toks -= L2T(police, skb->len);
312                 if ((toks|ptoks) >= 0) {
313                         police->tcfp_t_c = now;
314                         police->tcfp_toks = toks;
315                         police->tcfp_ptoks = ptoks;
316                         spin_unlock(&police->tcf_lock);
317                         return police->tcfp_result;
318                 }
319         }
320
321         police->tcf_qstats.overlimits++;
322         spin_unlock(&police->tcf_lock);
323         return police->tcf_action;
324 }
325
326 static int
327 tcf_act_police_dump(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
328 {
329         unsigned char    *b = skb->tail;
330         struct tcf_police *police = a->priv;
331         struct tc_police opt;
332
333         opt.index = police->tcf_index;
334         opt.action = police->tcf_action;
335         opt.mtu = police->tcfp_mtu;
336         opt.burst = police->tcfp_burst;
337         opt.refcnt = police->tcf_refcnt - ref;
338         opt.bindcnt = police->tcf_bindcnt - bind;
339         if (police->tcfp_R_tab)
340                 opt.rate = police->tcfp_R_tab->rate;
341         else
342                 memset(&opt.rate, 0, sizeof(opt.rate));
343         if (police->tcfp_P_tab)
344                 opt.peakrate = police->tcfp_P_tab->rate;
345         else
346                 memset(&opt.peakrate, 0, sizeof(opt.peakrate));
347         RTA_PUT(skb, TCA_POLICE_TBF, sizeof(opt), &opt);
348         if (police->tcfp_result)
349                 RTA_PUT(skb, TCA_POLICE_RESULT, sizeof(int),
350                         &police->tcfp_result);
351 #ifdef CONFIG_NET_ESTIMATOR
352         if (police->tcfp_ewma_rate)
353                 RTA_PUT(skb, TCA_POLICE_AVRATE, 4, &police->tcfp_ewma_rate);
354 #endif
355         return skb->len;
356
357 rtattr_failure:
358         skb_trim(skb, b - skb->data);
359         return -1;
360 }
361
362 MODULE_AUTHOR("Alexey Kuznetsov");
363 MODULE_DESCRIPTION("Policing actions");
364 MODULE_LICENSE("GPL");
365
366 static struct tc_action_ops act_police_ops = {
367         .kind           =       "police",
368         .hinfo          =       &police_hash_info,
369         .type           =       TCA_ID_POLICE,
370         .capab          =       TCA_CAP_NONE,
371         .owner          =       THIS_MODULE,
372         .act            =       tcf_act_police,
373         .dump           =       tcf_act_police_dump,
374         .cleanup        =       tcf_act_police_cleanup,
375         .lookup         =       tcf_hash_search,
376         .init           =       tcf_act_police_locate,
377         .walk           =       tcf_act_police_walker
378 };
379
380 static int __init
381 police_init_module(void)
382 {
383         return tcf_register_action(&act_police_ops);
384 }
385
386 static void __exit
387 police_cleanup_module(void)
388 {
389         tcf_unregister_action(&act_police_ops);
390 }
391
392 module_init(police_init_module);
393 module_exit(police_cleanup_module);
394
395 #else /* CONFIG_NET_CLS_ACT */
396
397 static struct tcf_common *tcf_police_lookup(u32 index)
398 {
399         struct tcf_hashinfo *hinfo = &police_hash_info;
400         struct tcf_common *p;
401
402         read_lock(hinfo->lock);
403         for (p = hinfo->htab[tcf_hash(index, hinfo->hmask)]; p;
404              p = p->tcfc_next) {
405                 if (p->tcfc_index == index)
406                         break;
407         }
408         read_unlock(hinfo->lock);
409
410         return p;
411 }
412
413 static u32 tcf_police_new_index(void)
414 {
415         u32 *idx_gen = &police_idx_gen;
416         u32 val = *idx_gen;
417
418         do {
419                 if (++val == 0)
420                         val = 1;
421         } while (tcf_police_lookup(val));
422
423         return (*idx_gen = val);
424 }
425
426 struct tcf_police *tcf_police_locate(struct rtattr *rta, struct rtattr *est)
427 {
428         unsigned int h;
429         struct tcf_police *police;
430         struct rtattr *tb[TCA_POLICE_MAX];
431         struct tc_police *parm;
432         int size;
433
434         if (rtattr_parse_nested(tb, TCA_POLICE_MAX, rta) < 0)
435                 return NULL;
436
437         if (tb[TCA_POLICE_TBF-1] == NULL)
438                 return NULL;
439         size = RTA_PAYLOAD(tb[TCA_POLICE_TBF-1]);
440         if (size != sizeof(*parm) && size != sizeof(struct tc_police_compat))
441                 return NULL;
442
443         parm = RTA_DATA(tb[TCA_POLICE_TBF-1]);
444
445         if (parm->index) {
446                 struct tcf_common *pc;
447
448                 pc = tcf_police_lookup(parm->index);
449                 if (pc) {
450                         police = to_police(pc);
451                         police->tcf_refcnt++;
452                         return police;
453                 }
454         }
455         police = kzalloc(sizeof(*police), GFP_KERNEL);
456         if (unlikely(!police))
457                 return NULL;
458
459         police->tcf_refcnt = 1;
460         spin_lock_init(&police->tcf_lock);
461         police->tcf_stats_lock = &police->tcf_lock;
462         if (parm->rate.rate) {
463                 police->tcfp_R_tab =
464                         qdisc_get_rtab(&parm->rate, tb[TCA_POLICE_RATE-1]);
465                 if (police->tcfp_R_tab == NULL)
466                         goto failure;
467                 if (parm->peakrate.rate) {
468                         police->tcfp_P_tab =
469                                 qdisc_get_rtab(&parm->peakrate,
470                                                tb[TCA_POLICE_PEAKRATE-1]);
471                         if (police->tcfp_P_tab == NULL)
472                                 goto failure;
473                 }
474         }
475         if (tb[TCA_POLICE_RESULT-1]) {
476                 if (RTA_PAYLOAD(tb[TCA_POLICE_RESULT-1]) != sizeof(u32))
477                         goto failure;
478                 police->tcfp_result = *(u32*)RTA_DATA(tb[TCA_POLICE_RESULT-1]);
479         }
480 #ifdef CONFIG_NET_ESTIMATOR
481         if (tb[TCA_POLICE_AVRATE-1]) {
482                 if (RTA_PAYLOAD(tb[TCA_POLICE_AVRATE-1]) != sizeof(u32))
483                         goto failure;
484                 police->tcfp_ewma_rate =
485                         *(u32*)RTA_DATA(tb[TCA_POLICE_AVRATE-1]);
486         }
487 #endif
488         police->tcfp_toks = police->tcfp_burst = parm->burst;
489         police->tcfp_mtu = parm->mtu;
490         if (police->tcfp_mtu == 0) {
491                 police->tcfp_mtu = ~0;
492                 if (police->tcfp_R_tab)
493                         police->tcfp_mtu = 255<<police->tcfp_R_tab->rate.cell_log;
494         }
495         if (police->tcfp_P_tab)
496                 police->tcfp_ptoks = L2T_P(police, police->tcfp_mtu);
497         PSCHED_GET_TIME(police->tcfp_t_c);
498         police->tcf_index = parm->index ? parm->index :
499                 tcf_police_new_index();
500         police->tcf_action = parm->action;
501 #ifdef CONFIG_NET_ESTIMATOR
502         if (est)
503                 gen_new_estimator(&police->tcf_bstats, &police->tcf_rate_est,
504                                   police->tcf_stats_lock, est);
505 #endif
506         h = tcf_hash(police->tcf_index, POL_TAB_MASK);
507         write_lock_bh(&police_lock);
508         police->tcf_next = tcf_police_ht[h];
509         tcf_police_ht[h] = &police->common;
510         write_unlock_bh(&police_lock);
511         return police;
512
513 failure:
514         if (police->tcfp_R_tab)
515                 qdisc_put_rtab(police->tcfp_R_tab);
516         kfree(police);
517         return NULL;
518 }
519
520 int tcf_police(struct sk_buff *skb, struct tcf_police *police)
521 {
522         psched_time_t now;
523         long toks;
524         long ptoks = 0;
525
526         spin_lock(&police->tcf_lock);
527
528         police->tcf_bstats.bytes += skb->len;
529         police->tcf_bstats.packets++;
530
531 #ifdef CONFIG_NET_ESTIMATOR
532         if (police->tcfp_ewma_rate &&
533             police->tcf_rate_est.bps >= police->tcfp_ewma_rate) {
534                 police->tcf_qstats.overlimits++;
535                 spin_unlock(&police->tcf_lock);
536                 return police->tcf_action;
537         }
538 #endif
539         if (skb->len <= police->tcfp_mtu) {
540                 if (police->tcfp_R_tab == NULL) {
541                         spin_unlock(&police->tcf_lock);
542                         return police->tcfp_result;
543                 }
544
545                 PSCHED_GET_TIME(now);
546                 toks = PSCHED_TDIFF_SAFE(now, police->tcfp_t_c,
547                                          police->tcfp_burst);
548                 if (police->tcfp_P_tab) {
549                         ptoks = toks + police->tcfp_ptoks;
550                         if (ptoks > (long)L2T_P(police, police->tcfp_mtu))
551                                 ptoks = (long)L2T_P(police, police->tcfp_mtu);
552                         ptoks -= L2T_P(police, skb->len);
553                 }
554                 toks += police->tcfp_toks;
555                 if (toks > (long)police->tcfp_burst)
556                         toks = police->tcfp_burst;
557                 toks -= L2T(police, skb->len);
558                 if ((toks|ptoks) >= 0) {
559                         police->tcfp_t_c = now;
560                         police->tcfp_toks = toks;
561                         police->tcfp_ptoks = ptoks;
562                         spin_unlock(&police->tcf_lock);
563                         return police->tcfp_result;
564                 }
565         }
566
567         police->tcf_qstats.overlimits++;
568         spin_unlock(&police->tcf_lock);
569         return police->tcf_action;
570 }
571 EXPORT_SYMBOL(tcf_police);
572
573 int tcf_police_dump(struct sk_buff *skb, struct tcf_police *police)
574 {
575         unsigned char *b = skb->tail;
576         struct tc_police opt;
577
578         opt.index = police->tcf_index;
579         opt.action = police->tcf_action;
580         opt.mtu = police->tcfp_mtu;
581         opt.burst = police->tcfp_burst;
582         if (police->tcfp_R_tab)
583                 opt.rate = police->tcfp_R_tab->rate;
584         else
585                 memset(&opt.rate, 0, sizeof(opt.rate));
586         if (police->tcfp_P_tab)
587                 opt.peakrate = police->tcfp_P_tab->rate;
588         else
589                 memset(&opt.peakrate, 0, sizeof(opt.peakrate));
590         RTA_PUT(skb, TCA_POLICE_TBF, sizeof(opt), &opt);
591         if (police->tcfp_result)
592                 RTA_PUT(skb, TCA_POLICE_RESULT, sizeof(int),
593                         &police->tcfp_result);
594 #ifdef CONFIG_NET_ESTIMATOR
595         if (police->tcfp_ewma_rate)
596                 RTA_PUT(skb, TCA_POLICE_AVRATE, 4, &police->tcfp_ewma_rate);
597 #endif
598         return skb->len;
599
600 rtattr_failure:
601         skb_trim(skb, b - skb->data);
602         return -1;
603 }
604
605 int tcf_police_dump_stats(struct sk_buff *skb, struct tcf_police *police)
606 {
607         struct gnet_dump d;
608
609         if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
610                                          TCA_XSTATS, police->tcf_stats_lock,
611                                          &d) < 0)
612                 goto errout;
613
614         if (gnet_stats_copy_basic(&d, &police->tcf_bstats) < 0 ||
615 #ifdef CONFIG_NET_ESTIMATOR
616             gnet_stats_copy_rate_est(&d, &police->tcf_rate_est) < 0 ||
617 #endif
618             gnet_stats_copy_queue(&d, &police->tcf_qstats) < 0)
619                 goto errout;
620
621         if (gnet_stats_finish_copy(&d) < 0)
622                 goto errout;
623
624         return 0;
625
626 errout:
627         return -1;
628 }
629
630 #endif /* CONFIG_NET_CLS_ACT */