7ebf7439b478d24b4246527c7313fb8cb774496b
[linux-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 <linux/module.h>
14 #include <linux/types.h>
15 #include <linux/kernel.h>
16 #include <linux/string.h>
17 #include <linux/errno.h>
18 #include <linux/skbuff.h>
19 #include <linux/rtnetlink.h>
20 #include <linux/init.h>
21 #include <linux/slab.h>
22 #include <net/act_api.h>
23 #include <net/netlink.h>
24
25 #define L2T(p,L)   qdisc_l2t((p)->tcfp_R_tab, L)
26 #define L2T_P(p,L) qdisc_l2t((p)->tcfp_P_tab, L)
27
28 #define POL_TAB_MASK     15
29 static struct tcf_common *tcf_police_ht[POL_TAB_MASK + 1];
30 static u32 police_idx_gen;
31 static DEFINE_RWLOCK(police_lock);
32
33 static struct tcf_hashinfo police_hash_info = {
34         .htab   =       tcf_police_ht,
35         .hmask  =       POL_TAB_MASK,
36         .lock   =       &police_lock,
37 };
38
39 /* old policer structure from before tc actions */
40 struct tc_police_compat
41 {
42         u32                     index;
43         int                     action;
44         u32                     limit;
45         u32                     burst;
46         u32                     mtu;
47         struct tc_ratespec      rate;
48         struct tc_ratespec      peakrate;
49 };
50
51 /* Each policer is serialized by its individual spinlock */
52
53 static int tcf_act_police_walker(struct sk_buff *skb, struct netlink_callback *cb,
54                               int type, struct tc_action *a)
55 {
56         struct tcf_common *p;
57         int err = 0, index = -1, i = 0, s_i = 0, n_i = 0;
58         struct nlattr *nest;
59
60         read_lock_bh(&police_lock);
61
62         s_i = cb->args[0];
63
64         for (i = 0; i < (POL_TAB_MASK + 1); i++) {
65                 p = tcf_police_ht[tcf_hash(i, POL_TAB_MASK)];
66
67                 for (; p; p = p->tcfc_next) {
68                         index++;
69                         if (index < s_i)
70                                 continue;
71                         a->priv = p;
72                         a->order = index;
73                         nest = nla_nest_start(skb, a->order);
74                         if (nest == NULL)
75                                 goto nla_put_failure;
76                         if (type == RTM_DELACTION)
77                                 err = tcf_action_dump_1(skb, a, 0, 1);
78                         else
79                                 err = tcf_action_dump_1(skb, a, 0, 0);
80                         if (err < 0) {
81                                 index--;
82                                 nla_nest_cancel(skb, nest);
83                                 goto done;
84                         }
85                         nla_nest_end(skb, nest);
86                         n_i++;
87                 }
88         }
89 done:
90         read_unlock_bh(&police_lock);
91         if (n_i)
92                 cb->args[0] += n_i;
93         return n_i;
94
95 nla_put_failure:
96         nla_nest_cancel(skb, nest);
97         goto done;
98 }
99
100 static void tcf_police_free_rcu(struct rcu_head *head)
101 {
102         kfree(container_of(head, struct tcf_police, tcf_rcu));
103 }
104
105 static void tcf_police_destroy(struct tcf_police *p)
106 {
107         unsigned int h = tcf_hash(p->tcf_index, POL_TAB_MASK);
108         struct tcf_common **p1p;
109
110         for (p1p = &tcf_police_ht[h]; *p1p; p1p = &(*p1p)->tcfc_next) {
111                 if (*p1p == &p->common) {
112                         write_lock_bh(&police_lock);
113                         *p1p = p->tcf_next;
114                         write_unlock_bh(&police_lock);
115                         gen_kill_estimator(&p->tcf_bstats,
116                                            &p->tcf_rate_est);
117                         if (p->tcfp_R_tab)
118                                 qdisc_put_rtab(p->tcfp_R_tab);
119                         if (p->tcfp_P_tab)
120                                 qdisc_put_rtab(p->tcfp_P_tab);
121                         /*
122                          * gen_estimator est_timer() might access p->tcf_lock
123                          * or bstats, wait a RCU grace period before freeing p
124                          */
125                         call_rcu(&p->tcf_rcu, tcf_police_free_rcu);
126                         return;
127                 }
128         }
129         WARN_ON(1);
130 }
131
132 static const struct nla_policy police_policy[TCA_POLICE_MAX + 1] = {
133         [TCA_POLICE_RATE]       = { .len = TC_RTAB_SIZE },
134         [TCA_POLICE_PEAKRATE]   = { .len = TC_RTAB_SIZE },
135         [TCA_POLICE_AVRATE]     = { .type = NLA_U32 },
136         [TCA_POLICE_RESULT]     = { .type = NLA_U32 },
137 };
138
139 static int tcf_act_police_locate(struct nlattr *nla, struct nlattr *est,
140                                  struct tc_action *a, int ovr, int bind)
141 {
142         unsigned h;
143         int ret = 0, err;
144         struct nlattr *tb[TCA_POLICE_MAX + 1];
145         struct tc_police *parm;
146         struct tcf_police *police;
147         struct qdisc_rate_table *R_tab = NULL, *P_tab = NULL;
148         int size;
149
150         if (nla == NULL)
151                 return -EINVAL;
152
153         err = nla_parse_nested(tb, TCA_POLICE_MAX, nla, police_policy);
154         if (err < 0)
155                 return err;
156
157         if (tb[TCA_POLICE_TBF] == NULL)
158                 return -EINVAL;
159         size = nla_len(tb[TCA_POLICE_TBF]);
160         if (size != sizeof(*parm) && size != sizeof(struct tc_police_compat))
161                 return -EINVAL;
162         parm = nla_data(tb[TCA_POLICE_TBF]);
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         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]);
193                 if (R_tab == NULL)
194                         goto failure;
195
196                 if (parm->peakrate.rate) {
197                         P_tab = qdisc_get_rtab(&parm->peakrate,
198                                                tb[TCA_POLICE_PEAKRATE]);
199                         if (P_tab == NULL)
200                                 goto failure;
201                 }
202         }
203
204         spin_lock_bh(&police->tcf_lock);
205         if (est) {
206                 err = gen_replace_estimator(&police->tcf_bstats,
207                                             &police->tcf_rate_est,
208                                             &police->tcf_lock, est);
209                 if (err)
210                         goto failure_unlock;
211         } else if (tb[TCA_POLICE_AVRATE] &&
212                    (ret == ACT_P_CREATED ||
213                     !gen_estimator_active(&police->tcf_bstats,
214                                           &police->tcf_rate_est))) {
215                 err = -EINVAL;
216                 goto failure_unlock;
217         }
218
219         /* No failure allowed after this point */
220         if (R_tab != NULL) {
221                 qdisc_put_rtab(police->tcfp_R_tab);
222                 police->tcfp_R_tab = R_tab;
223         }
224         if (P_tab != NULL) {
225                 qdisc_put_rtab(police->tcfp_P_tab);
226                 police->tcfp_P_tab = P_tab;
227         }
228
229         if (tb[TCA_POLICE_RESULT])
230                 police->tcfp_result = nla_get_u32(tb[TCA_POLICE_RESULT]);
231         police->tcfp_toks = police->tcfp_burst = parm->burst;
232         police->tcfp_mtu = parm->mtu;
233         if (police->tcfp_mtu == 0) {
234                 police->tcfp_mtu = ~0;
235                 if (police->tcfp_R_tab)
236                         police->tcfp_mtu = 255<<police->tcfp_R_tab->rate.cell_log;
237         }
238         if (police->tcfp_P_tab)
239                 police->tcfp_ptoks = L2T_P(police, police->tcfp_mtu);
240         police->tcf_action = parm->action;
241
242         if (tb[TCA_POLICE_AVRATE])
243                 police->tcfp_ewma_rate = nla_get_u32(tb[TCA_POLICE_AVRATE]);
244
245         spin_unlock_bh(&police->tcf_lock);
246         if (ret != ACT_P_CREATED)
247                 return ret;
248
249         police->tcfp_t_c = psched_get_time();
250         police->tcf_index = parm->index ? parm->index :
251                 tcf_hash_new_index(&police_idx_gen, &police_hash_info);
252         h = tcf_hash(police->tcf_index, POL_TAB_MASK);
253         write_lock_bh(&police_lock);
254         police->tcf_next = tcf_police_ht[h];
255         tcf_police_ht[h] = &police->common;
256         write_unlock_bh(&police_lock);
257
258         a->priv = police;
259         return ret;
260
261 failure_unlock:
262         spin_unlock_bh(&police->tcf_lock);
263 failure:
264         if (P_tab)
265                 qdisc_put_rtab(P_tab);
266         if (R_tab)
267                 qdisc_put_rtab(R_tab);
268         if (ret == ACT_P_CREATED)
269                 kfree(police);
270         return err;
271 }
272
273 static int tcf_act_police_cleanup(struct tc_action *a, int bind)
274 {
275         struct tcf_police *p = a->priv;
276         int ret = 0;
277
278         if (p != NULL) {
279                 if (bind)
280                         p->tcf_bindcnt--;
281
282                 p->tcf_refcnt--;
283                 if (p->tcf_refcnt <= 0 && !p->tcf_bindcnt) {
284                         tcf_police_destroy(p);
285                         ret = 1;
286                 }
287         }
288         return ret;
289 }
290
291 static int tcf_act_police(struct sk_buff *skb, struct tc_action *a,
292                           struct tcf_result *res)
293 {
294         struct tcf_police *police = a->priv;
295         psched_time_t now;
296         long toks;
297         long ptoks = 0;
298
299         spin_lock(&police->tcf_lock);
300
301         police->tcf_bstats.bytes += qdisc_pkt_len(skb);
302         police->tcf_bstats.packets++;
303
304         if (police->tcfp_ewma_rate &&
305             police->tcf_rate_est.bps >= police->tcfp_ewma_rate) {
306                 police->tcf_qstats.overlimits++;
307                 if (police->tcf_action == TC_ACT_SHOT)
308                         police->tcf_qstats.drops++;
309                 spin_unlock(&police->tcf_lock);
310                 return police->tcf_action;
311         }
312
313         if (qdisc_pkt_len(skb) <= police->tcfp_mtu) {
314                 if (police->tcfp_R_tab == NULL) {
315                         spin_unlock(&police->tcf_lock);
316                         return police->tcfp_result;
317                 }
318
319                 now = psched_get_time();
320                 toks = psched_tdiff_bounded(now, police->tcfp_t_c,
321                                             police->tcfp_burst);
322                 if (police->tcfp_P_tab) {
323                         ptoks = toks + police->tcfp_ptoks;
324                         if (ptoks > (long)L2T_P(police, police->tcfp_mtu))
325                                 ptoks = (long)L2T_P(police, police->tcfp_mtu);
326                         ptoks -= L2T_P(police, qdisc_pkt_len(skb));
327                 }
328                 toks += police->tcfp_toks;
329                 if (toks > (long)police->tcfp_burst)
330                         toks = police->tcfp_burst;
331                 toks -= L2T(police, qdisc_pkt_len(skb));
332                 if ((toks|ptoks) >= 0) {
333                         police->tcfp_t_c = now;
334                         police->tcfp_toks = toks;
335                         police->tcfp_ptoks = ptoks;
336                         spin_unlock(&police->tcf_lock);
337                         return police->tcfp_result;
338                 }
339         }
340
341         police->tcf_qstats.overlimits++;
342         if (police->tcf_action == TC_ACT_SHOT)
343                 police->tcf_qstats.drops++;
344         spin_unlock(&police->tcf_lock);
345         return police->tcf_action;
346 }
347
348 static int
349 tcf_act_police_dump(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
350 {
351         unsigned char *b = skb_tail_pointer(skb);
352         struct tcf_police *police = a->priv;
353         struct tc_police opt = {
354                 .index = police->tcf_index,
355                 .action = police->tcf_action,
356                 .mtu = police->tcfp_mtu,
357                 .burst = police->tcfp_burst,
358                 .refcnt = police->tcf_refcnt - ref,
359                 .bindcnt = police->tcf_bindcnt - bind,
360         };
361
362         if (police->tcfp_R_tab)
363                 opt.rate = police->tcfp_R_tab->rate;
364         if (police->tcfp_P_tab)
365                 opt.peakrate = police->tcfp_P_tab->rate;
366         NLA_PUT(skb, TCA_POLICE_TBF, sizeof(opt), &opt);
367         if (police->tcfp_result)
368                 NLA_PUT_U32(skb, TCA_POLICE_RESULT, police->tcfp_result);
369         if (police->tcfp_ewma_rate)
370                 NLA_PUT_U32(skb, TCA_POLICE_AVRATE, police->tcfp_ewma_rate);
371         return skb->len;
372
373 nla_put_failure:
374         nlmsg_trim(skb, b);
375         return -1;
376 }
377
378 MODULE_AUTHOR("Alexey Kuznetsov");
379 MODULE_DESCRIPTION("Policing actions");
380 MODULE_LICENSE("GPL");
381
382 static struct tc_action_ops act_police_ops = {
383         .kind           =       "police",
384         .hinfo          =       &police_hash_info,
385         .type           =       TCA_ID_POLICE,
386         .capab          =       TCA_CAP_NONE,
387         .owner          =       THIS_MODULE,
388         .act            =       tcf_act_police,
389         .dump           =       tcf_act_police_dump,
390         .cleanup        =       tcf_act_police_cleanup,
391         .lookup         =       tcf_hash_search,
392         .init           =       tcf_act_police_locate,
393         .walk           =       tcf_act_police_walker
394 };
395
396 static int __init
397 police_init_module(void)
398 {
399         return tcf_register_action(&act_police_ops);
400 }
401
402 static void __exit
403 police_cleanup_module(void)
404 {
405         tcf_unregister_action(&act_police_ops);
406         rcu_barrier(); /* Wait for completion of call_rcu()'s (tcf_police_free_rcu) */
407 }
408
409 module_init(police_init_module);
410 module_exit(police_cleanup_module);