92649d2667ed1d8091d35ea755c0579ba5dcd587
[muen/linux.git] / net / sched / act_police.c
1 /*
2  * net/sched/act_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 struct tcf_police_params {
26         int                     tcfp_result;
27         u32                     tcfp_ewma_rate;
28         s64                     tcfp_burst;
29         u32                     tcfp_mtu;
30         s64                     tcfp_toks;
31         s64                     tcfp_ptoks;
32         s64                     tcfp_mtu_ptoks;
33         s64                     tcfp_t_c;
34         struct psched_ratecfg   rate;
35         bool                    rate_present;
36         struct psched_ratecfg   peak;
37         bool                    peak_present;
38         struct rcu_head rcu;
39 };
40
41 struct tcf_police {
42         struct tc_action        common;
43         struct tcf_police_params __rcu *params;
44 };
45
46 #define to_police(pc) ((struct tcf_police *)pc)
47
48 /* old policer structure from before tc actions */
49 struct tc_police_compat {
50         u32                     index;
51         int                     action;
52         u32                     limit;
53         u32                     burst;
54         u32                     mtu;
55         struct tc_ratespec      rate;
56         struct tc_ratespec      peakrate;
57 };
58
59 /* Each policer is serialized by its individual spinlock */
60
61 static unsigned int police_net_id;
62 static struct tc_action_ops act_police_ops;
63
64 static int tcf_police_walker(struct net *net, struct sk_buff *skb,
65                                  struct netlink_callback *cb, int type,
66                                  const struct tc_action_ops *ops,
67                                  struct netlink_ext_ack *extack)
68 {
69         struct tc_action_net *tn = net_generic(net, police_net_id);
70
71         return tcf_generic_walker(tn, skb, cb, type, ops, extack);
72 }
73
74 static const struct nla_policy police_policy[TCA_POLICE_MAX + 1] = {
75         [TCA_POLICE_RATE]       = { .len = TC_RTAB_SIZE },
76         [TCA_POLICE_PEAKRATE]   = { .len = TC_RTAB_SIZE },
77         [TCA_POLICE_AVRATE]     = { .type = NLA_U32 },
78         [TCA_POLICE_RESULT]     = { .type = NLA_U32 },
79 };
80
81 static int tcf_police_init(struct net *net, struct nlattr *nla,
82                                struct nlattr *est, struct tc_action **a,
83                                int ovr, int bind, bool rtnl_held,
84                                struct netlink_ext_ack *extack)
85 {
86         int ret = 0, err;
87         struct nlattr *tb[TCA_POLICE_MAX + 1];
88         struct tc_police *parm;
89         struct tcf_police *police;
90         struct qdisc_rate_table *R_tab = NULL, *P_tab = NULL;
91         struct tc_action_net *tn = net_generic(net, police_net_id);
92         struct tcf_police_params *new;
93         bool exists = false;
94         int size;
95
96         if (nla == NULL)
97                 return -EINVAL;
98
99         err = nla_parse_nested(tb, TCA_POLICE_MAX, nla, police_policy, NULL);
100         if (err < 0)
101                 return err;
102
103         if (tb[TCA_POLICE_TBF] == NULL)
104                 return -EINVAL;
105         size = nla_len(tb[TCA_POLICE_TBF]);
106         if (size != sizeof(*parm) && size != sizeof(struct tc_police_compat))
107                 return -EINVAL;
108
109         parm = nla_data(tb[TCA_POLICE_TBF]);
110         err = tcf_idr_check_alloc(tn, &parm->index, a, bind);
111         if (err < 0)
112                 return err;
113         exists = err;
114         if (exists && bind)
115                 return 0;
116
117         if (!exists) {
118                 ret = tcf_idr_create(tn, parm->index, NULL, a,
119                                      &act_police_ops, bind, true);
120                 if (ret) {
121                         tcf_idr_cleanup(tn, parm->index);
122                         return ret;
123                 }
124                 ret = ACT_P_CREATED;
125         } else if (!ovr) {
126                 tcf_idr_release(*a, bind);
127                 return -EEXIST;
128         }
129
130         police = to_police(*a);
131         if (parm->rate.rate) {
132                 err = -ENOMEM;
133                 R_tab = qdisc_get_rtab(&parm->rate, tb[TCA_POLICE_RATE], NULL);
134                 if (R_tab == NULL)
135                         goto failure;
136
137                 if (parm->peakrate.rate) {
138                         P_tab = qdisc_get_rtab(&parm->peakrate,
139                                                tb[TCA_POLICE_PEAKRATE], NULL);
140                         if (P_tab == NULL)
141                                 goto failure;
142                 }
143         }
144
145         if (est) {
146                 err = gen_replace_estimator(&police->tcf_bstats,
147                                             police->common.cpu_bstats,
148                                             &police->tcf_rate_est,
149                                             &police->tcf_lock,
150                                             NULL, est);
151                 if (err)
152                         goto failure;
153         } else if (tb[TCA_POLICE_AVRATE] &&
154                    (ret == ACT_P_CREATED ||
155                     !gen_estimator_active(&police->tcf_rate_est))) {
156                 err = -EINVAL;
157                 goto failure;
158         }
159
160         new = kzalloc(sizeof(*new), GFP_KERNEL);
161         if (unlikely(!new)) {
162                 err = -ENOMEM;
163                 goto failure;
164         }
165
166         /* No failure allowed after this point */
167         new->tcfp_mtu = parm->mtu;
168         if (!new->tcfp_mtu) {
169                 new->tcfp_mtu = ~0;
170                 if (R_tab)
171                         new->tcfp_mtu = 255 << R_tab->rate.cell_log;
172         }
173         if (R_tab) {
174                 new->rate_present = true;
175                 psched_ratecfg_precompute(&new->rate, &R_tab->rate, 0);
176                 qdisc_put_rtab(R_tab);
177         } else {
178                 new->rate_present = false;
179         }
180         if (P_tab) {
181                 new->peak_present = true;
182                 psched_ratecfg_precompute(&new->peak, &P_tab->rate, 0);
183                 qdisc_put_rtab(P_tab);
184         } else {
185                 new->peak_present = false;
186         }
187
188         if (tb[TCA_POLICE_RESULT])
189                 new->tcfp_result = nla_get_u32(tb[TCA_POLICE_RESULT]);
190         new->tcfp_burst = PSCHED_TICKS2NS(parm->burst);
191         new->tcfp_toks = new->tcfp_burst;
192         if (new->peak_present) {
193                 new->tcfp_mtu_ptoks = (s64)psched_l2t_ns(&new->peak,
194                                                          new->tcfp_mtu);
195                 new->tcfp_ptoks = new->tcfp_mtu_ptoks;
196         }
197
198         if (tb[TCA_POLICE_AVRATE])
199                 new->tcfp_ewma_rate = nla_get_u32(tb[TCA_POLICE_AVRATE]);
200
201         spin_lock_bh(&police->tcf_lock);
202         new->tcfp_t_c = ktime_get_ns();
203         police->tcf_action = parm->action;
204         rcu_swap_protected(police->params,
205                            new,
206                            lockdep_is_held(&police->tcf_lock));
207         spin_unlock_bh(&police->tcf_lock);
208
209         if (new)
210                 kfree_rcu(new, rcu);
211
212         if (ret == ACT_P_CREATED)
213                 tcf_idr_insert(tn, *a);
214         return ret;
215
216 failure:
217         qdisc_put_rtab(P_tab);
218         qdisc_put_rtab(R_tab);
219         tcf_idr_release(*a, bind);
220         return err;
221 }
222
223 static int tcf_police_act(struct sk_buff *skb, const struct tc_action *a,
224                           struct tcf_result *res)
225 {
226         struct tcf_police *police = to_police(a);
227         struct tcf_police_params *p;
228         s64 now, toks, ptoks = 0;
229         int ret;
230
231         tcf_lastuse_update(&police->tcf_tm);
232         bstats_cpu_update(this_cpu_ptr(police->common.cpu_bstats), skb);
233
234         ret = READ_ONCE(police->tcf_action);
235         p = rcu_dereference_bh(police->params);
236
237         if (p->tcfp_ewma_rate) {
238                 struct gnet_stats_rate_est64 sample;
239
240                 if (!gen_estimator_read(&police->tcf_rate_est, &sample) ||
241                     sample.bps >= p->tcfp_ewma_rate)
242                         goto inc_overlimits;
243         }
244
245         if (qdisc_pkt_len(skb) <= p->tcfp_mtu) {
246                 if (!p->rate_present) {
247                         ret = p->tcfp_result;
248                         goto end;
249                 }
250
251                 now = ktime_get_ns();
252                 toks = min_t(s64, now - p->tcfp_t_c, p->tcfp_burst);
253                 if (p->peak_present) {
254                         ptoks = toks + p->tcfp_ptoks;
255                         if (ptoks > p->tcfp_mtu_ptoks)
256                                 ptoks = p->tcfp_mtu_ptoks;
257                         ptoks -= (s64)psched_l2t_ns(&p->peak,
258                                                     qdisc_pkt_len(skb));
259                 }
260                 toks += p->tcfp_toks;
261                 if (toks > p->tcfp_burst)
262                         toks = p->tcfp_burst;
263                 toks -= (s64)psched_l2t_ns(&p->rate, qdisc_pkt_len(skb));
264                 if ((toks|ptoks) >= 0) {
265                         p->tcfp_t_c = now;
266                         p->tcfp_toks = toks;
267                         p->tcfp_ptoks = ptoks;
268                         ret = p->tcfp_result;
269                         goto inc_drops;
270                 }
271         }
272
273 inc_overlimits:
274         qstats_overlimit_inc(this_cpu_ptr(police->common.cpu_qstats));
275 inc_drops:
276         if (ret == TC_ACT_SHOT)
277                 qstats_drop_inc(this_cpu_ptr(police->common.cpu_qstats));
278 end:
279         return ret;
280 }
281
282 static void tcf_police_cleanup(struct tc_action *a)
283 {
284         struct tcf_police *police = to_police(a);
285         struct tcf_police_params *p;
286
287         p = rcu_dereference_protected(police->params, 1);
288         if (p)
289                 kfree_rcu(p, rcu);
290 }
291
292 static int tcf_police_dump(struct sk_buff *skb, struct tc_action *a,
293                                int bind, int ref)
294 {
295         unsigned char *b = skb_tail_pointer(skb);
296         struct tcf_police *police = to_police(a);
297         struct tcf_police_params *p;
298         struct tc_police opt = {
299                 .index = police->tcf_index,
300                 .refcnt = refcount_read(&police->tcf_refcnt) - ref,
301                 .bindcnt = atomic_read(&police->tcf_bindcnt) - bind,
302         };
303         struct tcf_t t;
304
305         spin_lock_bh(&police->tcf_lock);
306         opt.action = police->tcf_action;
307         p = rcu_dereference_protected(police->params,
308                                       lockdep_is_held(&police->tcf_lock));
309         opt.mtu = p->tcfp_mtu;
310         opt.burst = PSCHED_NS2TICKS(p->tcfp_burst);
311         if (p->rate_present)
312                 psched_ratecfg_getrate(&opt.rate, &p->rate);
313         if (p->peak_present)
314                 psched_ratecfg_getrate(&opt.peakrate, &p->peak);
315         if (nla_put(skb, TCA_POLICE_TBF, sizeof(opt), &opt))
316                 goto nla_put_failure;
317         if (p->tcfp_result &&
318             nla_put_u32(skb, TCA_POLICE_RESULT, p->tcfp_result))
319                 goto nla_put_failure;
320         if (p->tcfp_ewma_rate &&
321             nla_put_u32(skb, TCA_POLICE_AVRATE, p->tcfp_ewma_rate))
322                 goto nla_put_failure;
323
324         t.install = jiffies_to_clock_t(jiffies - police->tcf_tm.install);
325         t.lastuse = jiffies_to_clock_t(jiffies - police->tcf_tm.lastuse);
326         t.firstuse = jiffies_to_clock_t(jiffies - police->tcf_tm.firstuse);
327         t.expires = jiffies_to_clock_t(police->tcf_tm.expires);
328         if (nla_put_64bit(skb, TCA_POLICE_TM, sizeof(t), &t, TCA_POLICE_PAD))
329                 goto nla_put_failure;
330         spin_unlock_bh(&police->tcf_lock);
331
332         return skb->len;
333
334 nla_put_failure:
335         spin_unlock_bh(&police->tcf_lock);
336         nlmsg_trim(skb, b);
337         return -1;
338 }
339
340 static int tcf_police_search(struct net *net, struct tc_action **a, u32 index)
341 {
342         struct tc_action_net *tn = net_generic(net, police_net_id);
343
344         return tcf_idr_search(tn, a, index);
345 }
346
347 MODULE_AUTHOR("Alexey Kuznetsov");
348 MODULE_DESCRIPTION("Policing actions");
349 MODULE_LICENSE("GPL");
350
351 static struct tc_action_ops act_police_ops = {
352         .kind           =       "police",
353         .type           =       TCA_ID_POLICE,
354         .owner          =       THIS_MODULE,
355         .act            =       tcf_police_act,
356         .dump           =       tcf_police_dump,
357         .init           =       tcf_police_init,
358         .walk           =       tcf_police_walker,
359         .lookup         =       tcf_police_search,
360         .cleanup        =       tcf_police_cleanup,
361         .size           =       sizeof(struct tcf_police),
362 };
363
364 static __net_init int police_init_net(struct net *net)
365 {
366         struct tc_action_net *tn = net_generic(net, police_net_id);
367
368         return tc_action_net_init(tn, &act_police_ops);
369 }
370
371 static void __net_exit police_exit_net(struct list_head *net_list)
372 {
373         tc_action_net_exit(net_list, police_net_id);
374 }
375
376 static struct pernet_operations police_net_ops = {
377         .init = police_init_net,
378         .exit_batch = police_exit_net,
379         .id   = &police_net_id,
380         .size = sizeof(struct tc_action_net),
381 };
382
383 static int __init police_init_module(void)
384 {
385         return tcf_register_action(&act_police_ops, &police_net_ops);
386 }
387
388 static void __exit police_cleanup_module(void)
389 {
390         tcf_unregister_action(&act_police_ops, &police_net_ops);
391 }
392
393 module_init(police_init_module);
394 module_exit(police_cleanup_module);