[NET_SCHED]: Use nla_nest_start/nla_nest_end
[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 <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         BUG_TRAP(0);
120 }
121
122 static int tcf_act_police_locate(struct nlattr *nla, struct nlattr *est,
123                                  struct tc_action *a, int ovr, int bind)
124 {
125         unsigned h;
126         int ret = 0, err;
127         struct nlattr *tb[TCA_POLICE_MAX + 1];
128         struct tc_police *parm;
129         struct tcf_police *police;
130         struct qdisc_rate_table *R_tab = NULL, *P_tab = NULL;
131         int size;
132
133         if (nla == NULL)
134                 return -EINVAL;
135
136         err = nla_parse_nested(tb, TCA_POLICE_MAX, nla, NULL);
137         if (err < 0)
138                 return err;
139
140         if (tb[TCA_POLICE_TBF] == NULL)
141                 return -EINVAL;
142         size = nla_len(tb[TCA_POLICE_TBF]);
143         if (size != sizeof(*parm) && size != sizeof(struct tc_police_compat))
144                 return -EINVAL;
145         parm = nla_data(tb[TCA_POLICE_TBF]);
146
147         if (tb[TCA_POLICE_RESULT] != NULL &&
148             nla_len(tb[TCA_POLICE_RESULT]) != sizeof(u32))
149                 return -EINVAL;
150         if (tb[TCA_POLICE_RESULT] != NULL &&
151             nla_len(tb[TCA_POLICE_RESULT]) != sizeof(u32))
152                 return -EINVAL;
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 = *(u32*)nla_data(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 =
220                         *(u32*)nla_data(tb[TCA_POLICE_AVRATE]);
221         if (est)
222                 gen_replace_estimator(&police->tcf_bstats,
223                                       &police->tcf_rate_est,
224                                       &police->tcf_lock, est);
225
226         spin_unlock_bh(&police->tcf_lock);
227         if (ret != ACT_P_CREATED)
228                 return ret;
229
230         police->tcfp_t_c = psched_get_time();
231         police->tcf_index = parm->index ? parm->index :
232                 tcf_hash_new_index(&police_idx_gen, &police_hash_info);
233         h = tcf_hash(police->tcf_index, POL_TAB_MASK);
234         write_lock_bh(&police_lock);
235         police->tcf_next = tcf_police_ht[h];
236         tcf_police_ht[h] = &police->common;
237         write_unlock_bh(&police_lock);
238
239         a->priv = police;
240         return ret;
241
242 failure:
243         if (ret == ACT_P_CREATED)
244                 kfree(police);
245         return err;
246 }
247
248 static int tcf_act_police_cleanup(struct tc_action *a, int bind)
249 {
250         struct tcf_police *p = a->priv;
251         int ret = 0;
252
253         if (p != NULL) {
254                 if (bind)
255                         p->tcf_bindcnt--;
256
257                 p->tcf_refcnt--;
258                 if (p->tcf_refcnt <= 0 && !p->tcf_bindcnt) {
259                         tcf_police_destroy(p);
260                         ret = 1;
261                 }
262         }
263         return ret;
264 }
265
266 static int tcf_act_police(struct sk_buff *skb, struct tc_action *a,
267                           struct tcf_result *res)
268 {
269         struct tcf_police *police = a->priv;
270         psched_time_t now;
271         long toks;
272         long ptoks = 0;
273
274         spin_lock(&police->tcf_lock);
275
276         police->tcf_bstats.bytes += skb->len;
277         police->tcf_bstats.packets++;
278
279         if (police->tcfp_ewma_rate &&
280             police->tcf_rate_est.bps >= police->tcfp_ewma_rate) {
281                 police->tcf_qstats.overlimits++;
282                 spin_unlock(&police->tcf_lock);
283                 return police->tcf_action;
284         }
285
286         if (skb->len <= police->tcfp_mtu) {
287                 if (police->tcfp_R_tab == NULL) {
288                         spin_unlock(&police->tcf_lock);
289                         return police->tcfp_result;
290                 }
291
292                 now = psched_get_time();
293                 toks = psched_tdiff_bounded(now, police->tcfp_t_c,
294                                             police->tcfp_burst);
295                 if (police->tcfp_P_tab) {
296                         ptoks = toks + police->tcfp_ptoks;
297                         if (ptoks > (long)L2T_P(police, police->tcfp_mtu))
298                                 ptoks = (long)L2T_P(police, police->tcfp_mtu);
299                         ptoks -= L2T_P(police, skb->len);
300                 }
301                 toks += police->tcfp_toks;
302                 if (toks > (long)police->tcfp_burst)
303                         toks = police->tcfp_burst;
304                 toks -= L2T(police, skb->len);
305                 if ((toks|ptoks) >= 0) {
306                         police->tcfp_t_c = now;
307                         police->tcfp_toks = toks;
308                         police->tcfp_ptoks = ptoks;
309                         spin_unlock(&police->tcf_lock);
310                         return police->tcfp_result;
311                 }
312         }
313
314         police->tcf_qstats.overlimits++;
315         spin_unlock(&police->tcf_lock);
316         return police->tcf_action;
317 }
318
319 static int
320 tcf_act_police_dump(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
321 {
322         unsigned char *b = skb_tail_pointer(skb);
323         struct tcf_police *police = a->priv;
324         struct tc_police opt;
325
326         opt.index = police->tcf_index;
327         opt.action = police->tcf_action;
328         opt.mtu = police->tcfp_mtu;
329         opt.burst = police->tcfp_burst;
330         opt.refcnt = police->tcf_refcnt - ref;
331         opt.bindcnt = police->tcf_bindcnt - bind;
332         if (police->tcfp_R_tab)
333                 opt.rate = police->tcfp_R_tab->rate;
334         else
335                 memset(&opt.rate, 0, sizeof(opt.rate));
336         if (police->tcfp_P_tab)
337                 opt.peakrate = police->tcfp_P_tab->rate;
338         else
339                 memset(&opt.peakrate, 0, sizeof(opt.peakrate));
340         NLA_PUT(skb, TCA_POLICE_TBF, sizeof(opt), &opt);
341         if (police->tcfp_result)
342                 NLA_PUT(skb, TCA_POLICE_RESULT, sizeof(int),
343                         &police->tcfp_result);
344         if (police->tcfp_ewma_rate)
345                 NLA_PUT(skb, TCA_POLICE_AVRATE, 4, &police->tcfp_ewma_rate);
346         return skb->len;
347
348 nla_put_failure:
349         nlmsg_trim(skb, b);
350         return -1;
351 }
352
353 MODULE_AUTHOR("Alexey Kuznetsov");
354 MODULE_DESCRIPTION("Policing actions");
355 MODULE_LICENSE("GPL");
356
357 static struct tc_action_ops act_police_ops = {
358         .kind           =       "police",
359         .hinfo          =       &police_hash_info,
360         .type           =       TCA_ID_POLICE,
361         .capab          =       TCA_CAP_NONE,
362         .owner          =       THIS_MODULE,
363         .act            =       tcf_act_police,
364         .dump           =       tcf_act_police_dump,
365         .cleanup        =       tcf_act_police_cleanup,
366         .lookup         =       tcf_hash_search,
367         .init           =       tcf_act_police_locate,
368         .walk           =       tcf_act_police_walker
369 };
370
371 static int __init
372 police_init_module(void)
373 {
374         return tcf_register_action(&act_police_ops);
375 }
376
377 static void __exit
378 police_cleanup_module(void)
379 {
380         tcf_unregister_action(&act_police_ops);
381 }
382
383 module_init(police_init_module);
384 module_exit(police_cleanup_module);