[NET_SCHED]: turn PSCHED_TDIFF_SAFE into inline function
[linux-3.10.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 #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 #ifdef CONFIG_NET_ESTIMATOR
122                         gen_kill_estimator(&p->tcf_bstats,
123                                            &p->tcf_rate_est);
124 #endif
125                         if (p->tcfp_R_tab)
126                                 qdisc_put_rtab(p->tcfp_R_tab);
127                         if (p->tcfp_P_tab)
128                                 qdisc_put_rtab(p->tcfp_P_tab);
129                         kfree(p);
130                         return;
131                 }
132         }
133         BUG_TRAP(0);
134 }
135
136 #ifdef CONFIG_NET_CLS_ACT
137 static int tcf_act_police_locate(struct rtattr *rta, struct rtattr *est,
138                                  struct tc_action *a, int ovr, int bind)
139 {
140         unsigned h;
141         int ret = 0, err;
142         struct rtattr *tb[TCA_POLICE_MAX];
143         struct tc_police *parm;
144         struct tcf_police *police;
145         struct qdisc_rate_table *R_tab = NULL, *P_tab = NULL;
146         int size;
147
148         if (rta == NULL || rtattr_parse_nested(tb, TCA_POLICE_MAX, rta) < 0)
149                 return -EINVAL;
150
151         if (tb[TCA_POLICE_TBF-1] == NULL)
152                 return -EINVAL;
153         size = RTA_PAYLOAD(tb[TCA_POLICE_TBF-1]);
154         if (size != sizeof(*parm) && size != sizeof(struct tc_police_compat))
155                 return -EINVAL;
156         parm = RTA_DATA(tb[TCA_POLICE_TBF-1]);
157
158         if (tb[TCA_POLICE_RESULT-1] != NULL &&
159             RTA_PAYLOAD(tb[TCA_POLICE_RESULT-1]) != sizeof(u32))
160                 return -EINVAL;
161         if (tb[TCA_POLICE_RESULT-1] != NULL &&
162             RTA_PAYLOAD(tb[TCA_POLICE_RESULT-1]) != sizeof(u32))
163                 return -EINVAL;
164
165         if (parm->index) {
166                 struct tcf_common *pc;
167
168                 pc = tcf_hash_lookup(parm->index, &police_hash_info);
169                 if (pc != NULL) {
170                         a->priv = pc;
171                         police = to_police(pc);
172                         if (bind) {
173                                 police->tcf_bindcnt += 1;
174                                 police->tcf_refcnt += 1;
175                         }
176                         if (ovr)
177                                 goto override;
178                         return ret;
179                 }
180         }
181
182         police = kzalloc(sizeof(*police), GFP_KERNEL);
183         if (police == NULL)
184                 return -ENOMEM;
185         ret = ACT_P_CREATED;
186         police->tcf_refcnt = 1;
187         spin_lock_init(&police->tcf_lock);
188         police->tcf_stats_lock = &police->tcf_lock;
189         if (bind)
190                 police->tcf_bindcnt = 1;
191 override:
192         if (parm->rate.rate) {
193                 err = -ENOMEM;
194                 R_tab = qdisc_get_rtab(&parm->rate, tb[TCA_POLICE_RATE-1]);
195                 if (R_tab == NULL)
196                         goto failure;
197                 if (parm->peakrate.rate) {
198                         P_tab = qdisc_get_rtab(&parm->peakrate,
199                                                tb[TCA_POLICE_PEAKRATE-1]);
200                         if (P_tab == NULL) {
201                                 qdisc_put_rtab(R_tab);
202                                 goto failure;
203                         }
204                 }
205         }
206         /* No failure allowed after this point */
207         spin_lock_bh(&police->tcf_lock);
208         if (R_tab != NULL) {
209                 qdisc_put_rtab(police->tcfp_R_tab);
210                 police->tcfp_R_tab = R_tab;
211         }
212         if (P_tab != NULL) {
213                 qdisc_put_rtab(police->tcfp_P_tab);
214                 police->tcfp_P_tab = P_tab;
215         }
216
217         if (tb[TCA_POLICE_RESULT-1])
218                 police->tcfp_result = *(u32*)RTA_DATA(tb[TCA_POLICE_RESULT-1]);
219         police->tcfp_toks = police->tcfp_burst = parm->burst;
220         police->tcfp_mtu = parm->mtu;
221         if (police->tcfp_mtu == 0) {
222                 police->tcfp_mtu = ~0;
223                 if (police->tcfp_R_tab)
224                         police->tcfp_mtu = 255<<police->tcfp_R_tab->rate.cell_log;
225         }
226         if (police->tcfp_P_tab)
227                 police->tcfp_ptoks = L2T_P(police, police->tcfp_mtu);
228         police->tcf_action = parm->action;
229
230 #ifdef CONFIG_NET_ESTIMATOR
231         if (tb[TCA_POLICE_AVRATE-1])
232                 police->tcfp_ewma_rate =
233                         *(u32*)RTA_DATA(tb[TCA_POLICE_AVRATE-1]);
234         if (est)
235                 gen_replace_estimator(&police->tcf_bstats,
236                                       &police->tcf_rate_est,
237                                       police->tcf_stats_lock, est);
238 #endif
239
240         spin_unlock_bh(&police->tcf_lock);
241         if (ret != ACT_P_CREATED)
242                 return ret;
243
244         PSCHED_GET_TIME(police->tcfp_t_c);
245         police->tcf_index = parm->index ? parm->index :
246                 tcf_hash_new_index(&police_idx_gen, &police_hash_info);
247         h = tcf_hash(police->tcf_index, POL_TAB_MASK);
248         write_lock_bh(&police_lock);
249         police->tcf_next = tcf_police_ht[h];
250         tcf_police_ht[h] = &police->common;
251         write_unlock_bh(&police_lock);
252
253         a->priv = police;
254         return ret;
255
256 failure:
257         if (ret == ACT_P_CREATED)
258                 kfree(police);
259         return err;
260 }
261
262 static int tcf_act_police_cleanup(struct tc_action *a, int bind)
263 {
264         struct tcf_police *p = a->priv;
265
266         if (p != NULL)
267                 return tcf_police_release(p, bind);
268         return 0;
269 }
270
271 static int tcf_act_police(struct sk_buff *skb, struct tc_action *a,
272                           struct tcf_result *res)
273 {
274         struct tcf_police *police = a->priv;
275         psched_time_t now;
276         long toks;
277         long ptoks = 0;
278
279         spin_lock(&police->tcf_lock);
280
281         police->tcf_bstats.bytes += skb->len;
282         police->tcf_bstats.packets++;
283
284 #ifdef CONFIG_NET_ESTIMATOR
285         if (police->tcfp_ewma_rate &&
286             police->tcf_rate_est.bps >= police->tcfp_ewma_rate) {
287                 police->tcf_qstats.overlimits++;
288                 spin_unlock(&police->tcf_lock);
289                 return police->tcf_action;
290         }
291 #endif
292
293         if (skb->len <= police->tcfp_mtu) {
294                 if (police->tcfp_R_tab == NULL) {
295                         spin_unlock(&police->tcf_lock);
296                         return police->tcfp_result;
297                 }
298
299                 PSCHED_GET_TIME(now);
300
301                 toks = psched_tdiff_bounded(now, police->tcfp_t_c,
302                                             police->tcfp_burst);
303                 if (police->tcfp_P_tab) {
304                         ptoks = toks + police->tcfp_ptoks;
305                         if (ptoks > (long)L2T_P(police, police->tcfp_mtu))
306                                 ptoks = (long)L2T_P(police, police->tcfp_mtu);
307                         ptoks -= L2T_P(police, skb->len);
308                 }
309                 toks += police->tcfp_toks;
310                 if (toks > (long)police->tcfp_burst)
311                         toks = police->tcfp_burst;
312                 toks -= L2T(police, skb->len);
313                 if ((toks|ptoks) >= 0) {
314                         police->tcfp_t_c = now;
315                         police->tcfp_toks = toks;
316                         police->tcfp_ptoks = ptoks;
317                         spin_unlock(&police->tcf_lock);
318                         return police->tcfp_result;
319                 }
320         }
321
322         police->tcf_qstats.overlimits++;
323         spin_unlock(&police->tcf_lock);
324         return police->tcf_action;
325 }
326
327 static int
328 tcf_act_police_dump(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
329 {
330         unsigned char *b = skb_tail_pointer(skb);
331         struct tcf_police *police = a->priv;
332         struct tc_police opt;
333
334         opt.index = police->tcf_index;
335         opt.action = police->tcf_action;
336         opt.mtu = police->tcfp_mtu;
337         opt.burst = police->tcfp_burst;
338         opt.refcnt = police->tcf_refcnt - ref;
339         opt.bindcnt = police->tcf_bindcnt - bind;
340         if (police->tcfp_R_tab)
341                 opt.rate = police->tcfp_R_tab->rate;
342         else
343                 memset(&opt.rate, 0, sizeof(opt.rate));
344         if (police->tcfp_P_tab)
345                 opt.peakrate = police->tcfp_P_tab->rate;
346         else
347                 memset(&opt.peakrate, 0, sizeof(opt.peakrate));
348         RTA_PUT(skb, TCA_POLICE_TBF, sizeof(opt), &opt);
349         if (police->tcfp_result)
350                 RTA_PUT(skb, TCA_POLICE_RESULT, sizeof(int),
351                         &police->tcfp_result);
352 #ifdef CONFIG_NET_ESTIMATOR
353         if (police->tcfp_ewma_rate)
354                 RTA_PUT(skb, TCA_POLICE_AVRATE, 4, &police->tcfp_ewma_rate);
355 #endif
356         return skb->len;
357
358 rtattr_failure:
359         nlmsg_trim(skb, b);
360         return -1;
361 }
362
363 MODULE_AUTHOR("Alexey Kuznetsov");
364 MODULE_DESCRIPTION("Policing actions");
365 MODULE_LICENSE("GPL");
366
367 static struct tc_action_ops act_police_ops = {
368         .kind           =       "police",
369         .hinfo          =       &police_hash_info,
370         .type           =       TCA_ID_POLICE,
371         .capab          =       TCA_CAP_NONE,
372         .owner          =       THIS_MODULE,
373         .act            =       tcf_act_police,
374         .dump           =       tcf_act_police_dump,
375         .cleanup        =       tcf_act_police_cleanup,
376         .lookup         =       tcf_hash_search,
377         .init           =       tcf_act_police_locate,
378         .walk           =       tcf_act_police_walker
379 };
380
381 static int __init
382 police_init_module(void)
383 {
384         return tcf_register_action(&act_police_ops);
385 }
386
387 static void __exit
388 police_cleanup_module(void)
389 {
390         tcf_unregister_action(&act_police_ops);
391 }
392
393 module_init(police_init_module);
394 module_exit(police_cleanup_module);
395
396 #else /* CONFIG_NET_CLS_ACT */
397
398 static struct tcf_common *tcf_police_lookup(u32 index)
399 {
400         struct tcf_hashinfo *hinfo = &police_hash_info;
401         struct tcf_common *p;
402
403         read_lock(hinfo->lock);
404         for (p = hinfo->htab[tcf_hash(index, hinfo->hmask)]; p;
405              p = p->tcfc_next) {
406                 if (p->tcfc_index == index)
407                         break;
408         }
409         read_unlock(hinfo->lock);
410
411         return p;
412 }
413
414 static u32 tcf_police_new_index(void)
415 {
416         u32 *idx_gen = &police_idx_gen;
417         u32 val = *idx_gen;
418
419         do {
420                 if (++val == 0)
421                         val = 1;
422         } while (tcf_police_lookup(val));
423
424         return (*idx_gen = val);
425 }
426
427 struct tcf_police *tcf_police_locate(struct rtattr *rta, struct rtattr *est)
428 {
429         unsigned int h;
430         struct tcf_police *police;
431         struct rtattr *tb[TCA_POLICE_MAX];
432         struct tc_police *parm;
433         int size;
434
435         if (rtattr_parse_nested(tb, TCA_POLICE_MAX, rta) < 0)
436                 return NULL;
437
438         if (tb[TCA_POLICE_TBF-1] == NULL)
439                 return NULL;
440         size = RTA_PAYLOAD(tb[TCA_POLICE_TBF-1]);
441         if (size != sizeof(*parm) && size != sizeof(struct tc_police_compat))
442                 return NULL;
443
444         parm = RTA_DATA(tb[TCA_POLICE_TBF-1]);
445
446         if (parm->index) {
447                 struct tcf_common *pc;
448
449                 pc = tcf_police_lookup(parm->index);
450                 if (pc) {
451                         police = to_police(pc);
452                         police->tcf_refcnt++;
453                         return police;
454                 }
455         }
456         police = kzalloc(sizeof(*police), GFP_KERNEL);
457         if (unlikely(!police))
458                 return NULL;
459
460         police->tcf_refcnt = 1;
461         spin_lock_init(&police->tcf_lock);
462         police->tcf_stats_lock = &police->tcf_lock;
463         if (parm->rate.rate) {
464                 police->tcfp_R_tab =
465                         qdisc_get_rtab(&parm->rate, tb[TCA_POLICE_RATE-1]);
466                 if (police->tcfp_R_tab == NULL)
467                         goto failure;
468                 if (parm->peakrate.rate) {
469                         police->tcfp_P_tab =
470                                 qdisc_get_rtab(&parm->peakrate,
471                                                tb[TCA_POLICE_PEAKRATE-1]);
472                         if (police->tcfp_P_tab == NULL)
473                                 goto failure;
474                 }
475         }
476         if (tb[TCA_POLICE_RESULT-1]) {
477                 if (RTA_PAYLOAD(tb[TCA_POLICE_RESULT-1]) != sizeof(u32))
478                         goto failure;
479                 police->tcfp_result = *(u32*)RTA_DATA(tb[TCA_POLICE_RESULT-1]);
480         }
481 #ifdef CONFIG_NET_ESTIMATOR
482         if (tb[TCA_POLICE_AVRATE-1]) {
483                 if (RTA_PAYLOAD(tb[TCA_POLICE_AVRATE-1]) != sizeof(u32))
484                         goto failure;
485                 police->tcfp_ewma_rate =
486                         *(u32*)RTA_DATA(tb[TCA_POLICE_AVRATE-1]);
487         }
488 #endif
489         police->tcfp_toks = police->tcfp_burst = parm->burst;
490         police->tcfp_mtu = parm->mtu;
491         if (police->tcfp_mtu == 0) {
492                 police->tcfp_mtu = ~0;
493                 if (police->tcfp_R_tab)
494                         police->tcfp_mtu = 255<<police->tcfp_R_tab->rate.cell_log;
495         }
496         if (police->tcfp_P_tab)
497                 police->tcfp_ptoks = L2T_P(police, police->tcfp_mtu);
498         PSCHED_GET_TIME(police->tcfp_t_c);
499         police->tcf_index = parm->index ? parm->index :
500                 tcf_police_new_index();
501         police->tcf_action = parm->action;
502 #ifdef CONFIG_NET_ESTIMATOR
503         if (est)
504                 gen_new_estimator(&police->tcf_bstats, &police->tcf_rate_est,
505                                   police->tcf_stats_lock, est);
506 #endif
507         h = tcf_hash(police->tcf_index, POL_TAB_MASK);
508         write_lock_bh(&police_lock);
509         police->tcf_next = tcf_police_ht[h];
510         tcf_police_ht[h] = &police->common;
511         write_unlock_bh(&police_lock);
512         return police;
513
514 failure:
515         if (police->tcfp_R_tab)
516                 qdisc_put_rtab(police->tcfp_R_tab);
517         kfree(police);
518         return NULL;
519 }
520
521 int tcf_police(struct sk_buff *skb, struct tcf_police *police)
522 {
523         psched_time_t now;
524         long toks;
525         long ptoks = 0;
526
527         spin_lock(&police->tcf_lock);
528
529         police->tcf_bstats.bytes += skb->len;
530         police->tcf_bstats.packets++;
531
532 #ifdef CONFIG_NET_ESTIMATOR
533         if (police->tcfp_ewma_rate &&
534             police->tcf_rate_est.bps >= police->tcfp_ewma_rate) {
535                 police->tcf_qstats.overlimits++;
536                 spin_unlock(&police->tcf_lock);
537                 return police->tcf_action;
538         }
539 #endif
540         if (skb->len <= police->tcfp_mtu) {
541                 if (police->tcfp_R_tab == NULL) {
542                         spin_unlock(&police->tcf_lock);
543                         return police->tcfp_result;
544                 }
545
546                 PSCHED_GET_TIME(now);
547                 toks = psched_tdiff_bounded(now, police->tcfp_t_c,
548                                             police->tcfp_burst);
549                 if (police->tcfp_P_tab) {
550                         ptoks = toks + police->tcfp_ptoks;
551                         if (ptoks > (long)L2T_P(police, police->tcfp_mtu))
552                                 ptoks = (long)L2T_P(police, police->tcfp_mtu);
553                         ptoks -= L2T_P(police, skb->len);
554                 }
555                 toks += police->tcfp_toks;
556                 if (toks > (long)police->tcfp_burst)
557                         toks = police->tcfp_burst;
558                 toks -= L2T(police, skb->len);
559                 if ((toks|ptoks) >= 0) {
560                         police->tcfp_t_c = now;
561                         police->tcfp_toks = toks;
562                         police->tcfp_ptoks = ptoks;
563                         spin_unlock(&police->tcf_lock);
564                         return police->tcfp_result;
565                 }
566         }
567
568         police->tcf_qstats.overlimits++;
569         spin_unlock(&police->tcf_lock);
570         return police->tcf_action;
571 }
572 EXPORT_SYMBOL(tcf_police);
573
574 int tcf_police_dump(struct sk_buff *skb, struct tcf_police *police)
575 {
576         unsigned char *b = skb_tail_pointer(skb);
577         struct tc_police opt;
578
579         opt.index = police->tcf_index;
580         opt.action = police->tcf_action;
581         opt.mtu = police->tcfp_mtu;
582         opt.burst = police->tcfp_burst;
583         if (police->tcfp_R_tab)
584                 opt.rate = police->tcfp_R_tab->rate;
585         else
586                 memset(&opt.rate, 0, sizeof(opt.rate));
587         if (police->tcfp_P_tab)
588                 opt.peakrate = police->tcfp_P_tab->rate;
589         else
590                 memset(&opt.peakrate, 0, sizeof(opt.peakrate));
591         RTA_PUT(skb, TCA_POLICE_TBF, sizeof(opt), &opt);
592         if (police->tcfp_result)
593                 RTA_PUT(skb, TCA_POLICE_RESULT, sizeof(int),
594                         &police->tcfp_result);
595 #ifdef CONFIG_NET_ESTIMATOR
596         if (police->tcfp_ewma_rate)
597                 RTA_PUT(skb, TCA_POLICE_AVRATE, 4, &police->tcfp_ewma_rate);
598 #endif
599         return skb->len;
600
601 rtattr_failure:
602         nlmsg_trim(skb, b);
603         return -1;
604 }
605
606 int tcf_police_dump_stats(struct sk_buff *skb, struct tcf_police *police)
607 {
608         struct gnet_dump d;
609
610         if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
611                                          TCA_XSTATS, police->tcf_stats_lock,
612                                          &d) < 0)
613                 goto errout;
614
615         if (gnet_stats_copy_basic(&d, &police->tcf_bstats) < 0 ||
616 #ifdef CONFIG_NET_ESTIMATOR
617             gnet_stats_copy_rate_est(&d, &police->tcf_rate_est) < 0 ||
618 #endif
619             gnet_stats_copy_queue(&d, &police->tcf_qstats) < 0)
620                 goto errout;
621
622         if (gnet_stats_finish_copy(&d) < 0)
623                 goto errout;
624
625         return 0;
626
627 errout:
628         return -1;
629 }
630
631 #endif /* CONFIG_NET_CLS_ACT */