Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * net/sched/act_police.c Input police filter
4 *
5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6 * J Hadi Salim (action changes)
7 */
8
9#include <linux/module.h>
10#include <linux/types.h>
11#include <linux/kernel.h>
12#include <linux/string.h>
13#include <linux/errno.h>
14#include <linux/skbuff.h>
15#include <linux/rtnetlink.h>
16#include <linux/init.h>
17#include <linux/slab.h>
18#include <net/act_api.h>
19#include <net/gso.h>
20#include <net/netlink.h>
21#include <net/pkt_cls.h>
22#include <net/tc_act/tc_police.h>
23#include <net/tc_wrapper.h>
24
25/* Each policer is serialized by its individual spinlock */
26
27static struct tc_action_ops act_police_ops;
28
29static const struct nla_policy police_policy[TCA_POLICE_MAX + 1] = {
30 [TCA_POLICE_RATE] = { .len = TC_RTAB_SIZE },
31 [TCA_POLICE_PEAKRATE] = { .len = TC_RTAB_SIZE },
32 [TCA_POLICE_AVRATE] = { .type = NLA_U32 },
33 [TCA_POLICE_RESULT] = { .type = NLA_U32 },
34 [TCA_POLICE_RATE64] = { .type = NLA_U64 },
35 [TCA_POLICE_PEAKRATE64] = { .type = NLA_U64 },
36 [TCA_POLICE_PKTRATE64] = { .type = NLA_U64, .min = 1 },
37 [TCA_POLICE_PKTBURST64] = { .type = NLA_U64, .min = 1 },
38};
39
40static int tcf_police_init(struct net *net, struct nlattr *nla,
41 struct nlattr *est, struct tc_action **a,
42 struct tcf_proto *tp, u32 flags,
43 struct netlink_ext_ack *extack)
44{
45 int ret = 0, tcfp_result = TC_ACT_OK, err, size;
46 bool bind = flags & TCA_ACT_FLAGS_BIND;
47 struct nlattr *tb[TCA_POLICE_MAX + 1];
48 struct tcf_chain *goto_ch = NULL;
49 struct tc_police *parm;
50 struct tcf_police *police;
51 struct qdisc_rate_table *R_tab = NULL, *P_tab = NULL;
52 struct tc_action_net *tn = net_generic(net, act_police_ops.net_id);
53 struct tcf_police_params *new;
54 bool exists = false;
55 u32 index;
56 u64 rate64, prate64;
57 u64 pps, ppsburst;
58
59 if (nla == NULL)
60 return -EINVAL;
61
62 err = nla_parse_nested_deprecated(tb, TCA_POLICE_MAX, nla,
63 police_policy, NULL);
64 if (err < 0)
65 return err;
66
67 if (tb[TCA_POLICE_TBF] == NULL)
68 return -EINVAL;
69 size = nla_len(tb[TCA_POLICE_TBF]);
70 if (size != sizeof(*parm) && size != sizeof(struct tc_police_compat))
71 return -EINVAL;
72
73 parm = nla_data(tb[TCA_POLICE_TBF]);
74 index = parm->index;
75 err = tcf_idr_check_alloc(tn, &index, a, bind);
76 if (err < 0)
77 return err;
78 exists = err;
79 if (exists && bind)
80 return ACT_P_BOUND;
81
82 if (!exists) {
83 ret = tcf_idr_create(tn, index, NULL, a,
84 &act_police_ops, bind, true, flags);
85 if (ret) {
86 tcf_idr_cleanup(tn, index);
87 return ret;
88 }
89 ret = ACT_P_CREATED;
90 spin_lock_init(&(to_police(*a)->tcfp_lock));
91 } else if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
92 tcf_idr_release(*a, bind);
93 return -EEXIST;
94 }
95 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
96 if (err < 0)
97 goto release_idr;
98
99 police = to_police(*a);
100 if (parm->rate.rate) {
101 err = -ENOMEM;
102 R_tab = qdisc_get_rtab(&parm->rate, tb[TCA_POLICE_RATE], NULL);
103 if (R_tab == NULL)
104 goto failure;
105
106 if (parm->peakrate.rate) {
107 P_tab = qdisc_get_rtab(&parm->peakrate,
108 tb[TCA_POLICE_PEAKRATE], NULL);
109 if (P_tab == NULL)
110 goto failure;
111 }
112 }
113
114 if (est) {
115 err = gen_replace_estimator(&police->tcf_bstats,
116 police->common.cpu_bstats,
117 &police->tcf_rate_est,
118 &police->tcf_lock,
119 false, est);
120 if (err)
121 goto failure;
122 } else if (tb[TCA_POLICE_AVRATE] &&
123 (ret == ACT_P_CREATED ||
124 !gen_estimator_active(&police->tcf_rate_est))) {
125 err = -EINVAL;
126 goto failure;
127 }
128
129 if (tb[TCA_POLICE_RESULT]) {
130 tcfp_result = nla_get_u32(tb[TCA_POLICE_RESULT]);
131 if (TC_ACT_EXT_CMP(tcfp_result, TC_ACT_GOTO_CHAIN)) {
132 NL_SET_ERR_MSG(extack,
133 "goto chain not allowed on fallback");
134 err = -EINVAL;
135 goto failure;
136 }
137 }
138
139 if ((tb[TCA_POLICE_PKTRATE64] && !tb[TCA_POLICE_PKTBURST64]) ||
140 (!tb[TCA_POLICE_PKTRATE64] && tb[TCA_POLICE_PKTBURST64])) {
141 NL_SET_ERR_MSG(extack,
142 "Both or neither packet-per-second burst and rate must be provided");
143 err = -EINVAL;
144 goto failure;
145 }
146
147 if (tb[TCA_POLICE_PKTRATE64] && R_tab) {
148 NL_SET_ERR_MSG(extack,
149 "packet-per-second and byte-per-second rate limits not allowed in same action");
150 err = -EINVAL;
151 goto failure;
152 }
153
154 new = kzalloc(sizeof(*new), GFP_KERNEL);
155 if (unlikely(!new)) {
156 err = -ENOMEM;
157 goto failure;
158 }
159
160 /* No failure allowed after this point */
161 new->tcfp_result = tcfp_result;
162 new->tcfp_mtu = parm->mtu;
163 if (!new->tcfp_mtu) {
164 new->tcfp_mtu = ~0;
165 if (R_tab)
166 new->tcfp_mtu = 255 << R_tab->rate.cell_log;
167 }
168 if (R_tab) {
169 new->rate_present = true;
170 rate64 = nla_get_u64_default(tb[TCA_POLICE_RATE64], 0);
171 psched_ratecfg_precompute(&new->rate, &R_tab->rate, rate64);
172 qdisc_put_rtab(R_tab);
173 } else {
174 new->rate_present = false;
175 }
176 if (P_tab) {
177 new->peak_present = true;
178 prate64 = nla_get_u64_default(tb[TCA_POLICE_PEAKRATE64], 0);
179 psched_ratecfg_precompute(&new->peak, &P_tab->rate, prate64);
180 qdisc_put_rtab(P_tab);
181 } else {
182 new->peak_present = false;
183 }
184
185 new->tcfp_burst = PSCHED_TICKS2NS(parm->burst);
186 if (new->peak_present)
187 new->tcfp_mtu_ptoks = (s64)psched_l2t_ns(&new->peak,
188 new->tcfp_mtu);
189
190 if (tb[TCA_POLICE_AVRATE])
191 new->tcfp_ewma_rate = nla_get_u32(tb[TCA_POLICE_AVRATE]);
192
193 if (tb[TCA_POLICE_PKTRATE64]) {
194 pps = nla_get_u64(tb[TCA_POLICE_PKTRATE64]);
195 ppsburst = nla_get_u64(tb[TCA_POLICE_PKTBURST64]);
196 new->pps_present = true;
197 new->tcfp_pkt_burst = PSCHED_TICKS2NS(ppsburst);
198 psched_ppscfg_precompute(&new->ppsrate, pps);
199 }
200
201 spin_lock_bh(&police->tcf_lock);
202 spin_lock_bh(&police->tcfp_lock);
203 police->tcfp_t_c = ktime_get_ns();
204 police->tcfp_toks = new->tcfp_burst;
205 if (new->peak_present)
206 police->tcfp_ptoks = new->tcfp_mtu_ptoks;
207 spin_unlock_bh(&police->tcfp_lock);
208 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
209 new = rcu_replace_pointer(police->params,
210 new,
211 lockdep_is_held(&police->tcf_lock));
212 spin_unlock_bh(&police->tcf_lock);
213
214 if (goto_ch)
215 tcf_chain_put_by_act(goto_ch);
216 if (new)
217 kfree_rcu(new, rcu);
218
219 return ret;
220
221failure:
222 qdisc_put_rtab(P_tab);
223 qdisc_put_rtab(R_tab);
224 if (goto_ch)
225 tcf_chain_put_by_act(goto_ch);
226release_idr:
227 tcf_idr_release(*a, bind);
228 return err;
229}
230
231static bool tcf_police_mtu_check(struct sk_buff *skb, u32 limit)
232{
233 u32 len;
234
235 if (skb_is_gso(skb))
236 return skb_gso_validate_mac_len(skb, limit);
237
238 len = qdisc_pkt_len(skb);
239 if (skb_at_tc_ingress(skb))
240 len += skb->mac_len;
241
242 return len <= limit;
243}
244
245TC_INDIRECT_SCOPE int tcf_police_act(struct sk_buff *skb,
246 const struct tc_action *a,
247 struct tcf_result *res)
248{
249 struct tcf_police *police = to_police(a);
250 s64 now, toks, ppstoks = 0, ptoks = 0;
251 struct tcf_police_params *p;
252 int ret;
253
254 tcf_lastuse_update(&police->tcf_tm);
255 bstats_update(this_cpu_ptr(police->common.cpu_bstats), skb);
256
257 ret = READ_ONCE(police->tcf_action);
258 p = rcu_dereference_bh(police->params);
259
260 if (p->tcfp_ewma_rate) {
261 struct gnet_stats_rate_est64 sample;
262
263 if (!gen_estimator_read(&police->tcf_rate_est, &sample) ||
264 sample.bps >= p->tcfp_ewma_rate)
265 goto inc_overlimits;
266 }
267
268 if (tcf_police_mtu_check(skb, p->tcfp_mtu)) {
269 if (!p->rate_present && !p->pps_present) {
270 ret = p->tcfp_result;
271 goto end;
272 }
273
274 now = ktime_get_ns();
275 spin_lock_bh(&police->tcfp_lock);
276 toks = min_t(s64, now - police->tcfp_t_c, p->tcfp_burst);
277 if (p->peak_present) {
278 ptoks = toks + police->tcfp_ptoks;
279 if (ptoks > p->tcfp_mtu_ptoks)
280 ptoks = p->tcfp_mtu_ptoks;
281 ptoks -= (s64)psched_l2t_ns(&p->peak,
282 qdisc_pkt_len(skb));
283 }
284 if (p->rate_present) {
285 toks += police->tcfp_toks;
286 if (toks > p->tcfp_burst)
287 toks = p->tcfp_burst;
288 toks -= (s64)psched_l2t_ns(&p->rate, qdisc_pkt_len(skb));
289 } else if (p->pps_present) {
290 ppstoks = min_t(s64, now - police->tcfp_t_c, p->tcfp_pkt_burst);
291 ppstoks += police->tcfp_pkttoks;
292 if (ppstoks > p->tcfp_pkt_burst)
293 ppstoks = p->tcfp_pkt_burst;
294 ppstoks -= (s64)psched_pkt2t_ns(&p->ppsrate, 1);
295 }
296 if ((toks | ptoks | ppstoks) >= 0) {
297 police->tcfp_t_c = now;
298 police->tcfp_toks = toks;
299 police->tcfp_ptoks = ptoks;
300 police->tcfp_pkttoks = ppstoks;
301 spin_unlock_bh(&police->tcfp_lock);
302 ret = p->tcfp_result;
303 goto inc_drops;
304 }
305 spin_unlock_bh(&police->tcfp_lock);
306 }
307
308inc_overlimits:
309 qstats_overlimit_inc(this_cpu_ptr(police->common.cpu_qstats));
310inc_drops:
311 if (ret == TC_ACT_SHOT)
312 qstats_drop_inc(this_cpu_ptr(police->common.cpu_qstats));
313end:
314 return ret;
315}
316
317static void tcf_police_cleanup(struct tc_action *a)
318{
319 struct tcf_police *police = to_police(a);
320 struct tcf_police_params *p;
321
322 p = rcu_dereference_protected(police->params, 1);
323 if (p)
324 kfree_rcu(p, rcu);
325}
326
327static void tcf_police_stats_update(struct tc_action *a,
328 u64 bytes, u64 packets, u64 drops,
329 u64 lastuse, bool hw)
330{
331 struct tcf_police *police = to_police(a);
332 struct tcf_t *tm = &police->tcf_tm;
333
334 tcf_action_update_stats(a, bytes, packets, drops, hw);
335 tm->lastuse = max_t(u64, tm->lastuse, lastuse);
336}
337
338static int tcf_police_dump(struct sk_buff *skb, struct tc_action *a,
339 int bind, int ref)
340{
341 unsigned char *b = skb_tail_pointer(skb);
342 struct tcf_police *police = to_police(a);
343 struct tcf_police_params *p;
344 struct tc_police opt = {
345 .index = police->tcf_index,
346 .refcnt = refcount_read(&police->tcf_refcnt) - ref,
347 .bindcnt = atomic_read(&police->tcf_bindcnt) - bind,
348 };
349 struct tcf_t t;
350
351 spin_lock_bh(&police->tcf_lock);
352 opt.action = police->tcf_action;
353 p = rcu_dereference_protected(police->params,
354 lockdep_is_held(&police->tcf_lock));
355 opt.mtu = p->tcfp_mtu;
356 opt.burst = PSCHED_NS2TICKS(p->tcfp_burst);
357 if (p->rate_present) {
358 psched_ratecfg_getrate(&opt.rate, &p->rate);
359 if ((p->rate.rate_bytes_ps >= (1ULL << 32)) &&
360 nla_put_u64_64bit(skb, TCA_POLICE_RATE64,
361 p->rate.rate_bytes_ps,
362 TCA_POLICE_PAD))
363 goto nla_put_failure;
364 }
365 if (p->peak_present) {
366 psched_ratecfg_getrate(&opt.peakrate, &p->peak);
367 if ((p->peak.rate_bytes_ps >= (1ULL << 32)) &&
368 nla_put_u64_64bit(skb, TCA_POLICE_PEAKRATE64,
369 p->peak.rate_bytes_ps,
370 TCA_POLICE_PAD))
371 goto nla_put_failure;
372 }
373 if (p->pps_present) {
374 if (nla_put_u64_64bit(skb, TCA_POLICE_PKTRATE64,
375 p->ppsrate.rate_pkts_ps,
376 TCA_POLICE_PAD))
377 goto nla_put_failure;
378 if (nla_put_u64_64bit(skb, TCA_POLICE_PKTBURST64,
379 PSCHED_NS2TICKS(p->tcfp_pkt_burst),
380 TCA_POLICE_PAD))
381 goto nla_put_failure;
382 }
383 if (nla_put(skb, TCA_POLICE_TBF, sizeof(opt), &opt))
384 goto nla_put_failure;
385 if (p->tcfp_result &&
386 nla_put_u32(skb, TCA_POLICE_RESULT, p->tcfp_result))
387 goto nla_put_failure;
388 if (p->tcfp_ewma_rate &&
389 nla_put_u32(skb, TCA_POLICE_AVRATE, p->tcfp_ewma_rate))
390 goto nla_put_failure;
391
392 tcf_tm_dump(&t, &police->tcf_tm);
393 if (nla_put_64bit(skb, TCA_POLICE_TM, sizeof(t), &t, TCA_POLICE_PAD))
394 goto nla_put_failure;
395 spin_unlock_bh(&police->tcf_lock);
396
397 return skb->len;
398
399nla_put_failure:
400 spin_unlock_bh(&police->tcf_lock);
401 nlmsg_trim(skb, b);
402 return -1;
403}
404
405static int tcf_police_act_to_flow_act(int tc_act, u32 *extval,
406 struct netlink_ext_ack *extack)
407{
408 int act_id = -EOPNOTSUPP;
409
410 if (!TC_ACT_EXT_OPCODE(tc_act)) {
411 if (tc_act == TC_ACT_OK)
412 act_id = FLOW_ACTION_ACCEPT;
413 else if (tc_act == TC_ACT_SHOT)
414 act_id = FLOW_ACTION_DROP;
415 else if (tc_act == TC_ACT_PIPE)
416 act_id = FLOW_ACTION_PIPE;
417 else if (tc_act == TC_ACT_RECLASSIFY)
418 NL_SET_ERR_MSG_MOD(extack, "Offload not supported when conform/exceed action is \"reclassify\"");
419 else
420 NL_SET_ERR_MSG_MOD(extack, "Unsupported conform/exceed action offload");
421 } else if (TC_ACT_EXT_CMP(tc_act, TC_ACT_GOTO_CHAIN)) {
422 act_id = FLOW_ACTION_GOTO;
423 *extval = tc_act & TC_ACT_EXT_VAL_MASK;
424 } else if (TC_ACT_EXT_CMP(tc_act, TC_ACT_JUMP)) {
425 act_id = FLOW_ACTION_JUMP;
426 *extval = tc_act & TC_ACT_EXT_VAL_MASK;
427 } else if (tc_act == TC_ACT_UNSPEC) {
428 act_id = FLOW_ACTION_CONTINUE;
429 } else {
430 NL_SET_ERR_MSG_MOD(extack, "Unsupported conform/exceed action offload");
431 }
432
433 return act_id;
434}
435
436static int tcf_police_offload_act_setup(struct tc_action *act, void *entry_data,
437 u32 *index_inc, bool bind,
438 struct netlink_ext_ack *extack)
439{
440 if (bind) {
441 struct flow_action_entry *entry = entry_data;
442 struct tcf_police *police = to_police(act);
443 struct tcf_police_params *p;
444 int act_id;
445
446 p = rcu_dereference_protected(police->params,
447 lockdep_is_held(&police->tcf_lock));
448
449 entry->id = FLOW_ACTION_POLICE;
450 entry->police.burst = tcf_police_burst(act);
451 entry->police.rate_bytes_ps =
452 tcf_police_rate_bytes_ps(act);
453 entry->police.peakrate_bytes_ps = tcf_police_peakrate_bytes_ps(act);
454 entry->police.avrate = tcf_police_tcfp_ewma_rate(act);
455 entry->police.overhead = tcf_police_rate_overhead(act);
456 entry->police.burst_pkt = tcf_police_burst_pkt(act);
457 entry->police.rate_pkt_ps =
458 tcf_police_rate_pkt_ps(act);
459 entry->police.mtu = tcf_police_tcfp_mtu(act);
460
461 act_id = tcf_police_act_to_flow_act(police->tcf_action,
462 &entry->police.exceed.extval,
463 extack);
464 if (act_id < 0)
465 return act_id;
466
467 entry->police.exceed.act_id = act_id;
468
469 act_id = tcf_police_act_to_flow_act(p->tcfp_result,
470 &entry->police.notexceed.extval,
471 extack);
472 if (act_id < 0)
473 return act_id;
474
475 entry->police.notexceed.act_id = act_id;
476
477 *index_inc = 1;
478 } else {
479 struct flow_offload_action *fl_action = entry_data;
480
481 fl_action->id = FLOW_ACTION_POLICE;
482 }
483
484 return 0;
485}
486
487MODULE_AUTHOR("Alexey Kuznetsov");
488MODULE_DESCRIPTION("Policing actions");
489MODULE_LICENSE("GPL");
490
491static struct tc_action_ops act_police_ops = {
492 .kind = "police",
493 .id = TCA_ID_POLICE,
494 .owner = THIS_MODULE,
495 .stats_update = tcf_police_stats_update,
496 .act = tcf_police_act,
497 .dump = tcf_police_dump,
498 .init = tcf_police_init,
499 .cleanup = tcf_police_cleanup,
500 .offload_act_setup = tcf_police_offload_act_setup,
501 .size = sizeof(struct tcf_police),
502};
503MODULE_ALIAS_NET_ACT("police");
504
505static __net_init int police_init_net(struct net *net)
506{
507 struct tc_action_net *tn = net_generic(net, act_police_ops.net_id);
508
509 return tc_action_net_init(net, tn, &act_police_ops);
510}
511
512static void __net_exit police_exit_net(struct list_head *net_list)
513{
514 tc_action_net_exit(net_list, act_police_ops.net_id);
515}
516
517static struct pernet_operations police_net_ops = {
518 .init = police_init_net,
519 .exit_batch = police_exit_net,
520 .id = &act_police_ops.net_id,
521 .size = sizeof(struct tc_action_net),
522};
523
524static int __init police_init_module(void)
525{
526 return tcf_register_action(&act_police_ops, &police_net_ops);
527}
528
529static void __exit police_cleanup_module(void)
530{
531 tcf_unregister_action(&act_police_ops, &police_net_ops);
532}
533
534module_init(police_init_module);
535module_exit(police_cleanup_module);
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
29static struct tcf_common *tcf_police_ht[POL_TAB_MASK + 1];
30static u32 police_idx_gen;
31static DEFINE_RWLOCK(police_lock);
32
33static 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 */
40struct tc_police_compat {
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
52static 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 }
88done:
89 read_unlock_bh(&police_lock);
90 if (n_i)
91 cb->args[0] += n_i;
92 return n_i;
93
94nla_put_failure:
95 nla_nest_cancel(skb, nest);
96 goto done;
97}
98
99static 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 /*
116 * gen_estimator est_timer() might access p->tcf_lock
117 * or bstats, wait a RCU grace period before freeing p
118 */
119 kfree_rcu(p, tcf_rcu);
120 return;
121 }
122 }
123 WARN_ON(1);
124}
125
126static const struct nla_policy police_policy[TCA_POLICE_MAX + 1] = {
127 [TCA_POLICE_RATE] = { .len = TC_RTAB_SIZE },
128 [TCA_POLICE_PEAKRATE] = { .len = TC_RTAB_SIZE },
129 [TCA_POLICE_AVRATE] = { .type = NLA_U32 },
130 [TCA_POLICE_RESULT] = { .type = NLA_U32 },
131};
132
133static int tcf_act_police_locate(struct nlattr *nla, struct nlattr *est,
134 struct tc_action *a, int ovr, int bind)
135{
136 unsigned int h;
137 int ret = 0, err;
138 struct nlattr *tb[TCA_POLICE_MAX + 1];
139 struct tc_police *parm;
140 struct tcf_police *police;
141 struct qdisc_rate_table *R_tab = NULL, *P_tab = NULL;
142 int size;
143
144 if (nla == NULL)
145 return -EINVAL;
146
147 err = nla_parse_nested(tb, TCA_POLICE_MAX, nla, police_policy);
148 if (err < 0)
149 return err;
150
151 if (tb[TCA_POLICE_TBF] == NULL)
152 return -EINVAL;
153 size = nla_len(tb[TCA_POLICE_TBF]);
154 if (size != sizeof(*parm) && size != sizeof(struct tc_police_compat))
155 return -EINVAL;
156 parm = nla_data(tb[TCA_POLICE_TBF]);
157
158 if (parm->index) {
159 struct tcf_common *pc;
160
161 pc = tcf_hash_lookup(parm->index, &police_hash_info);
162 if (pc != NULL) {
163 a->priv = pc;
164 police = to_police(pc);
165 if (bind) {
166 police->tcf_bindcnt += 1;
167 police->tcf_refcnt += 1;
168 }
169 if (ovr)
170 goto override;
171 return ret;
172 }
173 }
174
175 police = kzalloc(sizeof(*police), GFP_KERNEL);
176 if (police == NULL)
177 return -ENOMEM;
178 ret = ACT_P_CREATED;
179 police->tcf_refcnt = 1;
180 spin_lock_init(&police->tcf_lock);
181 if (bind)
182 police->tcf_bindcnt = 1;
183override:
184 if (parm->rate.rate) {
185 err = -ENOMEM;
186 R_tab = qdisc_get_rtab(&parm->rate, tb[TCA_POLICE_RATE]);
187 if (R_tab == NULL)
188 goto failure;
189
190 if (parm->peakrate.rate) {
191 P_tab = qdisc_get_rtab(&parm->peakrate,
192 tb[TCA_POLICE_PEAKRATE]);
193 if (P_tab == NULL)
194 goto failure;
195 }
196 }
197
198 spin_lock_bh(&police->tcf_lock);
199 if (est) {
200 err = gen_replace_estimator(&police->tcf_bstats,
201 &police->tcf_rate_est,
202 &police->tcf_lock, est);
203 if (err)
204 goto failure_unlock;
205 } else if (tb[TCA_POLICE_AVRATE] &&
206 (ret == ACT_P_CREATED ||
207 !gen_estimator_active(&police->tcf_bstats,
208 &police->tcf_rate_est))) {
209 err = -EINVAL;
210 goto failure_unlock;
211 }
212
213 /* No failure allowed after this point */
214 if (R_tab != NULL) {
215 qdisc_put_rtab(police->tcfp_R_tab);
216 police->tcfp_R_tab = R_tab;
217 }
218 if (P_tab != NULL) {
219 qdisc_put_rtab(police->tcfp_P_tab);
220 police->tcfp_P_tab = P_tab;
221 }
222
223 if (tb[TCA_POLICE_RESULT])
224 police->tcfp_result = nla_get_u32(tb[TCA_POLICE_RESULT]);
225 police->tcfp_toks = police->tcfp_burst = parm->burst;
226 police->tcfp_mtu = parm->mtu;
227 if (police->tcfp_mtu == 0) {
228 police->tcfp_mtu = ~0;
229 if (police->tcfp_R_tab)
230 police->tcfp_mtu = 255<<police->tcfp_R_tab->rate.cell_log;
231 }
232 if (police->tcfp_P_tab)
233 police->tcfp_ptoks = L2T_P(police, police->tcfp_mtu);
234 police->tcf_action = parm->action;
235
236 if (tb[TCA_POLICE_AVRATE])
237 police->tcfp_ewma_rate = nla_get_u32(tb[TCA_POLICE_AVRATE]);
238
239 spin_unlock_bh(&police->tcf_lock);
240 if (ret != ACT_P_CREATED)
241 return ret;
242
243 police->tcfp_t_c = psched_get_time();
244 police->tcf_index = parm->index ? parm->index :
245 tcf_hash_new_index(&police_idx_gen, &police_hash_info);
246 h = tcf_hash(police->tcf_index, POL_TAB_MASK);
247 write_lock_bh(&police_lock);
248 police->tcf_next = tcf_police_ht[h];
249 tcf_police_ht[h] = &police->common;
250 write_unlock_bh(&police_lock);
251
252 a->priv = police;
253 return ret;
254
255failure_unlock:
256 spin_unlock_bh(&police->tcf_lock);
257failure:
258 if (P_tab)
259 qdisc_put_rtab(P_tab);
260 if (R_tab)
261 qdisc_put_rtab(R_tab);
262 if (ret == ACT_P_CREATED)
263 kfree(police);
264 return err;
265}
266
267static int tcf_act_police_cleanup(struct tc_action *a, int bind)
268{
269 struct tcf_police *p = a->priv;
270 int ret = 0;
271
272 if (p != NULL) {
273 if (bind)
274 p->tcf_bindcnt--;
275
276 p->tcf_refcnt--;
277 if (p->tcf_refcnt <= 0 && !p->tcf_bindcnt) {
278 tcf_police_destroy(p);
279 ret = 1;
280 }
281 }
282 return ret;
283}
284
285static int tcf_act_police(struct sk_buff *skb, const struct tc_action *a,
286 struct tcf_result *res)
287{
288 struct tcf_police *police = a->priv;
289 psched_time_t now;
290 long toks;
291 long ptoks = 0;
292
293 spin_lock(&police->tcf_lock);
294
295 bstats_update(&police->tcf_bstats, skb);
296
297 if (police->tcfp_ewma_rate &&
298 police->tcf_rate_est.bps >= police->tcfp_ewma_rate) {
299 police->tcf_qstats.overlimits++;
300 if (police->tcf_action == TC_ACT_SHOT)
301 police->tcf_qstats.drops++;
302 spin_unlock(&police->tcf_lock);
303 return police->tcf_action;
304 }
305
306 if (qdisc_pkt_len(skb) <= police->tcfp_mtu) {
307 if (police->tcfp_R_tab == NULL) {
308 spin_unlock(&police->tcf_lock);
309 return police->tcfp_result;
310 }
311
312 now = psched_get_time();
313 toks = psched_tdiff_bounded(now, police->tcfp_t_c,
314 police->tcfp_burst);
315 if (police->tcfp_P_tab) {
316 ptoks = toks + police->tcfp_ptoks;
317 if (ptoks > (long)L2T_P(police, police->tcfp_mtu))
318 ptoks = (long)L2T_P(police, police->tcfp_mtu);
319 ptoks -= L2T_P(police, qdisc_pkt_len(skb));
320 }
321 toks += police->tcfp_toks;
322 if (toks > (long)police->tcfp_burst)
323 toks = police->tcfp_burst;
324 toks -= L2T(police, qdisc_pkt_len(skb));
325 if ((toks|ptoks) >= 0) {
326 police->tcfp_t_c = now;
327 police->tcfp_toks = toks;
328 police->tcfp_ptoks = ptoks;
329 spin_unlock(&police->tcf_lock);
330 return police->tcfp_result;
331 }
332 }
333
334 police->tcf_qstats.overlimits++;
335 if (police->tcf_action == TC_ACT_SHOT)
336 police->tcf_qstats.drops++;
337 spin_unlock(&police->tcf_lock);
338 return police->tcf_action;
339}
340
341static int
342tcf_act_police_dump(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
343{
344 unsigned char *b = skb_tail_pointer(skb);
345 struct tcf_police *police = a->priv;
346 struct tc_police opt = {
347 .index = police->tcf_index,
348 .action = police->tcf_action,
349 .mtu = police->tcfp_mtu,
350 .burst = police->tcfp_burst,
351 .refcnt = police->tcf_refcnt - ref,
352 .bindcnt = police->tcf_bindcnt - bind,
353 };
354
355 if (police->tcfp_R_tab)
356 opt.rate = police->tcfp_R_tab->rate;
357 if (police->tcfp_P_tab)
358 opt.peakrate = police->tcfp_P_tab->rate;
359 if (nla_put(skb, TCA_POLICE_TBF, sizeof(opt), &opt))
360 goto nla_put_failure;
361 if (police->tcfp_result &&
362 nla_put_u32(skb, TCA_POLICE_RESULT, police->tcfp_result))
363 goto nla_put_failure;
364 if (police->tcfp_ewma_rate &&
365 nla_put_u32(skb, TCA_POLICE_AVRATE, police->tcfp_ewma_rate))
366 goto nla_put_failure;
367 return skb->len;
368
369nla_put_failure:
370 nlmsg_trim(skb, b);
371 return -1;
372}
373
374MODULE_AUTHOR("Alexey Kuznetsov");
375MODULE_DESCRIPTION("Policing actions");
376MODULE_LICENSE("GPL");
377
378static struct tc_action_ops act_police_ops = {
379 .kind = "police",
380 .hinfo = &police_hash_info,
381 .type = TCA_ID_POLICE,
382 .capab = TCA_CAP_NONE,
383 .owner = THIS_MODULE,
384 .act = tcf_act_police,
385 .dump = tcf_act_police_dump,
386 .cleanup = tcf_act_police_cleanup,
387 .lookup = tcf_hash_search,
388 .init = tcf_act_police_locate,
389 .walk = tcf_act_police_walker
390};
391
392static int __init
393police_init_module(void)
394{
395 return tcf_register_action(&act_police_ops);
396}
397
398static void __exit
399police_cleanup_module(void)
400{
401 tcf_unregister_action(&act_police_ops);
402}
403
404module_init(police_init_module);
405module_exit(police_cleanup_module);