Merge branch 'for-2.6.28' of git://linux-nfs.org/~bfields/linux
[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 <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 <net/act_api.h>
22 #include <net/netlink.h>
23
24 #define L2T(p,L)   qdisc_l2t((p)->tcfp_R_tab, L)
25 #define L2T_P(p,L) qdisc_l2t((p)->tcfp_P_tab, L)
26
27 #define POL_TAB_MASK     15
28 static struct tcf_common *tcf_police_ht[POL_TAB_MASK + 1];
29 static u32 police_idx_gen;
30 static DEFINE_RWLOCK(police_lock);
31
32 static struct tcf_hashinfo police_hash_info = {
33         .htab   =       tcf_police_ht,
34         .hmask  =       POL_TAB_MASK,
35         .lock   =       &police_lock,
36 };
37
38 /* old policer structure from before tc actions */
39 struct tc_police_compat
40 {
41         u32                     index;
42         int                     action;
43         u32                     limit;
44         u32                     burst;
45         u32                     mtu;
46         struct tc_ratespec      rate;
47         struct tc_ratespec      peakrate;
48 };
49
50 /* Each policer is serialized by its individual spinlock */
51
52 static int tcf_act_police_walker(struct sk_buff *skb, struct netlink_callback *cb,
53                               int type, struct tc_action *a)
54 {
55         struct tcf_common *p;
56         int err = 0, index = -1, i = 0, s_i = 0, n_i = 0;
57         struct nlattr *nest;
58
59         read_lock_bh(&police_lock);
60
61         s_i = cb->args[0];
62
63         for (i = 0; i < (POL_TAB_MASK + 1); i++) {
64                 p = tcf_police_ht[tcf_hash(i, POL_TAB_MASK)];
65
66                 for (; p; p = p->tcfc_next) {
67                         index++;
68                         if (index < s_i)
69                                 continue;
70                         a->priv = p;
71                         a->order = index;
72                         nest = nla_nest_start(skb, a->order);
73                         if (nest == NULL)
74                                 goto nla_put_failure;
75                         if (type == RTM_DELACTION)
76                                 err = tcf_action_dump_1(skb, a, 0, 1);
77                         else
78                                 err = tcf_action_dump_1(skb, a, 0, 0);
79                         if (err < 0) {
80                                 index--;
81                                 nla_nest_cancel(skb, nest);
82                                 goto done;
83                         }
84                         nla_nest_end(skb, nest);
85                         n_i++;
86                 }
87         }
88 done:
89         read_unlock_bh(&police_lock);
90         if (n_i)
91                 cb->args[0] += n_i;
92         return n_i;
93
94 nla_put_failure:
95         nla_nest_cancel(skb, nest);
96         goto done;
97 }
98
99 static void tcf_police_destroy(struct tcf_police *p)
100 {
101         unsigned int h = tcf_hash(p->tcf_index, POL_TAB_MASK);
102         struct tcf_common **p1p;
103
104         for (p1p = &tcf_police_ht[h]; *p1p; p1p = &(*p1p)->tcfc_next) {
105                 if (*p1p == &p->common) {
106                         write_lock_bh(&police_lock);
107                         *p1p = p->tcf_next;
108                         write_unlock_bh(&police_lock);
109                         gen_kill_estimator(&p->tcf_bstats,
110                                            &p->tcf_rate_est);
111                         if (p->tcfp_R_tab)
112                                 qdisc_put_rtab(p->tcfp_R_tab);
113                         if (p->tcfp_P_tab)
114                                 qdisc_put_rtab(p->tcfp_P_tab);
115                         kfree(p);
116                         return;
117                 }
118         }
119         WARN_ON(1);
120 }
121
122 static const struct nla_policy police_policy[TCA_POLICE_MAX + 1] = {
123         [TCA_POLICE_RATE]       = { .len = TC_RTAB_SIZE },
124         [TCA_POLICE_PEAKRATE]   = { .len = TC_RTAB_SIZE },
125         [TCA_POLICE_AVRATE]     = { .type = NLA_U32 },
126         [TCA_POLICE_RESULT]     = { .type = NLA_U32 },
127 };
128
129 static int tcf_act_police_locate(struct nlattr *nla, struct nlattr *est,
130                                  struct tc_action *a, int ovr, int bind)
131 {
132         unsigned h;
133         int ret = 0, err;
134         struct nlattr *tb[TCA_POLICE_MAX + 1];
135         struct tc_police *parm;
136         struct tcf_police *police;
137         struct qdisc_rate_table *R_tab = NULL, *P_tab = NULL;
138         int size;
139
140         if (nla == NULL)
141                 return -EINVAL;
142
143         err = nla_parse_nested(tb, TCA_POLICE_MAX, nla, police_policy);
144         if (err < 0)
145                 return err;
146
147         if (tb[TCA_POLICE_TBF] == NULL)
148                 return -EINVAL;
149         size = nla_len(tb[TCA_POLICE_TBF]);
150         if (size != sizeof(*parm) && size != sizeof(struct tc_police_compat))
151                 return -EINVAL;
152         parm = nla_data(tb[TCA_POLICE_TBF]);
153
154         if (parm->index) {
155                 struct tcf_common *pc;
156
157                 pc = tcf_hash_lookup(parm->index, &police_hash_info);
158                 if (pc != NULL) {
159                         a->priv = pc;
160                         police = to_police(pc);
161                         if (bind) {
162                                 police->tcf_bindcnt += 1;
163                                 police->tcf_refcnt += 1;
164                         }
165                         if (ovr)
166                                 goto override;
167                         return ret;
168                 }
169         }
170
171         police = kzalloc(sizeof(*police), GFP_KERNEL);
172         if (police == NULL)
173                 return -ENOMEM;
174         ret = ACT_P_CREATED;
175         police->tcf_refcnt = 1;
176         spin_lock_init(&police->tcf_lock);
177         if (bind)
178                 police->tcf_bindcnt = 1;
179 override:
180         if (parm->rate.rate) {
181                 err = -ENOMEM;
182                 R_tab = qdisc_get_rtab(&parm->rate, tb[TCA_POLICE_RATE]);
183                 if (R_tab == NULL)
184                         goto failure;
185                 if (parm->peakrate.rate) {
186                         P_tab = qdisc_get_rtab(&parm->peakrate,
187                                                tb[TCA_POLICE_PEAKRATE]);
188                         if (P_tab == NULL) {
189                                 qdisc_put_rtab(R_tab);
190                                 goto failure;
191                         }
192                 }
193         }
194         /* No failure allowed after this point */
195         spin_lock_bh(&police->tcf_lock);
196         if (R_tab != NULL) {
197                 qdisc_put_rtab(police->tcfp_R_tab);
198                 police->tcfp_R_tab = R_tab;
199         }
200         if (P_tab != NULL) {
201                 qdisc_put_rtab(police->tcfp_P_tab);
202                 police->tcfp_P_tab = P_tab;
203         }
204
205         if (tb[TCA_POLICE_RESULT])
206                 police->tcfp_result = nla_get_u32(tb[TCA_POLICE_RESULT]);
207         police->tcfp_toks = police->tcfp_burst = parm->burst;
208         police->tcfp_mtu = parm->mtu;
209         if (police->tcfp_mtu == 0) {
210                 police->tcfp_mtu = ~0;
211                 if (police->tcfp_R_tab)
212                         police->tcfp_mtu = 255<<police->tcfp_R_tab->rate.cell_log;
213         }
214         if (police->tcfp_P_tab)
215                 police->tcfp_ptoks = L2T_P(police, police->tcfp_mtu);
216         police->tcf_action = parm->action;
217
218         if (tb[TCA_POLICE_AVRATE])
219                 police->tcfp_ewma_rate = nla_get_u32(tb[TCA_POLICE_AVRATE]);
220         if (est)
221                 gen_replace_estimator(&police->tcf_bstats,
222                                       &police->tcf_rate_est,
223                                       &police->tcf_lock, est);
224
225         spin_unlock_bh(&police->tcf_lock);
226         if (ret != ACT_P_CREATED)
227                 return ret;
228
229         police->tcfp_t_c = psched_get_time();
230         police->tcf_index = parm->index ? parm->index :
231                 tcf_hash_new_index(&police_idx_gen, &police_hash_info);
232         h = tcf_hash(police->tcf_index, POL_TAB_MASK);
233         write_lock_bh(&police_lock);
234         police->tcf_next = tcf_police_ht[h];
235         tcf_police_ht[h] = &police->common;
236         write_unlock_bh(&police_lock);
237
238         a->priv = police;
239         return ret;
240
241 failure:
242         if (ret == ACT_P_CREATED)
243                 kfree(police);
244         return err;
245 }
246
247 static int tcf_act_police_cleanup(struct tc_action *a, int bind)
248 {
249         struct tcf_police *p = a->priv;
250         int ret = 0;
251
252         if (p != NULL) {
253                 if (bind)
254                         p->tcf_bindcnt--;
255
256                 p->tcf_refcnt--;
257                 if (p->tcf_refcnt <= 0 && !p->tcf_bindcnt) {
258                         tcf_police_destroy(p);
259                         ret = 1;
260                 }
261         }
262         return ret;
263 }
264
265 static int tcf_act_police(struct sk_buff *skb, struct tc_action *a,
266                           struct tcf_result *res)
267 {
268         struct tcf_police *police = a->priv;
269         psched_time_t now;
270         long toks;
271         long ptoks = 0;
272
273         spin_lock(&police->tcf_lock);
274
275         police->tcf_bstats.bytes += qdisc_pkt_len(skb);
276         police->tcf_bstats.packets++;
277
278         if (police->tcfp_ewma_rate &&
279             police->tcf_rate_est.bps >= police->tcfp_ewma_rate) {
280                 police->tcf_qstats.overlimits++;
281                 spin_unlock(&police->tcf_lock);
282                 return police->tcf_action;
283         }
284
285         if (qdisc_pkt_len(skb) <= police->tcfp_mtu) {
286                 if (police->tcfp_R_tab == NULL) {
287                         spin_unlock(&police->tcf_lock);
288                         return police->tcfp_result;
289                 }
290
291                 now = psched_get_time();
292                 toks = psched_tdiff_bounded(now, police->tcfp_t_c,
293                                             police->tcfp_burst);
294                 if (police->tcfp_P_tab) {
295                         ptoks = toks + police->tcfp_ptoks;
296                         if (ptoks > (long)L2T_P(police, police->tcfp_mtu))
297                                 ptoks = (long)L2T_P(police, police->tcfp_mtu);
298                         ptoks -= L2T_P(police, qdisc_pkt_len(skb));
299                 }
300                 toks += police->tcfp_toks;
301                 if (toks > (long)police->tcfp_burst)
302                         toks = police->tcfp_burst;
303                 toks -= L2T(police, qdisc_pkt_len(skb));
304                 if ((toks|ptoks) >= 0) {
305                         police->tcfp_t_c = now;
306                         police->tcfp_toks = toks;
307                         police->tcfp_ptoks = ptoks;
308                         spin_unlock(&police->tcf_lock);
309                         return police->tcfp_result;
310                 }
311         }
312
313         police->tcf_qstats.overlimits++;
314         spin_unlock(&police->tcf_lock);
315         return police->tcf_action;
316 }
317
318 static int
319 tcf_act_police_dump(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
320 {
321         unsigned char *b = skb_tail_pointer(skb);
322         struct tcf_police *police = a->priv;
323         struct tc_police opt;
324
325         opt.index = police->tcf_index;
326         opt.action = police->tcf_action;
327         opt.mtu = police->tcfp_mtu;
328         opt.burst = police->tcfp_burst;
329         opt.refcnt = police->tcf_refcnt - ref;
330         opt.bindcnt = police->tcf_bindcnt - bind;
331         if (police->tcfp_R_tab)
332                 opt.rate = police->tcfp_R_tab->rate;
333         else
334                 memset(&opt.rate, 0, sizeof(opt.rate));
335         if (police->tcfp_P_tab)
336                 opt.peakrate = police->tcfp_P_tab->rate;
337         else
338                 memset(&opt.peakrate, 0, sizeof(opt.peakrate));
339         NLA_PUT(skb, TCA_POLICE_TBF, sizeof(opt), &opt);
340         if (police->tcfp_result)
341                 NLA_PUT_U32(skb, TCA_POLICE_RESULT, police->tcfp_result);
342         if (police->tcfp_ewma_rate)
343                 NLA_PUT_U32(skb, TCA_POLICE_AVRATE, police->tcfp_ewma_rate);
344         return skb->len;
345
346 nla_put_failure:
347         nlmsg_trim(skb, b);
348         return -1;
349 }
350
351 MODULE_AUTHOR("Alexey Kuznetsov");
352 MODULE_DESCRIPTION("Policing actions");
353 MODULE_LICENSE("GPL");
354
355 static struct tc_action_ops act_police_ops = {
356         .kind           =       "police",
357         .hinfo          =       &police_hash_info,
358         .type           =       TCA_ID_POLICE,
359         .capab          =       TCA_CAP_NONE,
360         .owner          =       THIS_MODULE,
361         .act            =       tcf_act_police,
362         .dump           =       tcf_act_police_dump,
363         .cleanup        =       tcf_act_police_cleanup,
364         .lookup         =       tcf_hash_search,
365         .init           =       tcf_act_police_locate,
366         .walk           =       tcf_act_police_walker
367 };
368
369 static int __init
370 police_init_module(void)
371 {
372         return tcf_register_action(&act_police_ops);
373 }
374
375 static void __exit
376 police_cleanup_module(void)
377 {
378         tcf_unregister_action(&act_police_ops);
379 }
380
381 module_init(police_init_module);
382 module_exit(police_cleanup_module);