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/netlink.h>
20 #include <net/pkt_cls.h>
21 #include <net/tc_act/tc_police.h>
22
23 /* Each policer is serialized by its individual spinlock */
24
25 static unsigned int police_net_id;
26 static struct tc_action_ops act_police_ops;
27
tcf_police_walker(struct net * net,struct sk_buff * skb,struct netlink_callback * cb,int type,const struct tc_action_ops * ops,struct netlink_ext_ack * extack)28 static int tcf_police_walker(struct net *net, struct sk_buff *skb,
29 struct netlink_callback *cb, int type,
30 const struct tc_action_ops *ops,
31 struct netlink_ext_ack *extack)
32 {
33 struct tc_action_net *tn = net_generic(net, police_net_id);
34
35 return tcf_generic_walker(tn, skb, cb, type, ops, extack);
36 }
37
38 static const struct nla_policy police_policy[TCA_POLICE_MAX + 1] = {
39 [TCA_POLICE_RATE] = { .len = TC_RTAB_SIZE },
40 [TCA_POLICE_PEAKRATE] = { .len = TC_RTAB_SIZE },
41 [TCA_POLICE_AVRATE] = { .type = NLA_U32 },
42 [TCA_POLICE_RESULT] = { .type = NLA_U32 },
43 [TCA_POLICE_RATE64] = { .type = NLA_U64 },
44 [TCA_POLICE_PEAKRATE64] = { .type = NLA_U64 },
45 };
46
tcf_police_init(struct net * net,struct nlattr * nla,struct nlattr * est,struct tc_action ** a,int ovr,int bind,bool rtnl_held,struct tcf_proto * tp,u32 flags,struct netlink_ext_ack * extack)47 static int tcf_police_init(struct net *net, struct nlattr *nla,
48 struct nlattr *est, struct tc_action **a,
49 int ovr, int bind, bool rtnl_held,
50 struct tcf_proto *tp, u32 flags,
51 struct netlink_ext_ack *extack)
52 {
53 int ret = 0, tcfp_result = TC_ACT_OK, err, size;
54 struct nlattr *tb[TCA_POLICE_MAX + 1];
55 struct tcf_chain *goto_ch = NULL;
56 struct tc_police *parm;
57 struct tcf_police *police;
58 struct qdisc_rate_table *R_tab = NULL, *P_tab = NULL;
59 struct tc_action_net *tn = net_generic(net, police_net_id);
60 struct tcf_police_params *new;
61 bool exists = false;
62 u32 index;
63 u64 rate64, prate64;
64
65 if (nla == NULL)
66 return -EINVAL;
67
68 err = nla_parse_nested_deprecated(tb, TCA_POLICE_MAX, nla,
69 police_policy, NULL);
70 if (err < 0)
71 return err;
72
73 if (tb[TCA_POLICE_TBF] == NULL)
74 return -EINVAL;
75 size = nla_len(tb[TCA_POLICE_TBF]);
76 if (size != sizeof(*parm) && size != sizeof(struct tc_police_compat))
77 return -EINVAL;
78
79 parm = nla_data(tb[TCA_POLICE_TBF]);
80 index = parm->index;
81 err = tcf_idr_check_alloc(tn, &index, a, bind);
82 if (err < 0)
83 return err;
84 exists = err;
85 if (exists && bind)
86 return 0;
87
88 if (!exists) {
89 ret = tcf_idr_create(tn, index, NULL, a,
90 &act_police_ops, bind, true, 0);
91 if (ret) {
92 tcf_idr_cleanup(tn, index);
93 return ret;
94 }
95 ret = ACT_P_CREATED;
96 spin_lock_init(&(to_police(*a)->tcfp_lock));
97 } else if (!ovr) {
98 tcf_idr_release(*a, bind);
99 return -EEXIST;
100 }
101 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
102 if (err < 0)
103 goto release_idr;
104
105 police = to_police(*a);
106 if (parm->rate.rate) {
107 err = -ENOMEM;
108 R_tab = qdisc_get_rtab(&parm->rate, tb[TCA_POLICE_RATE], NULL);
109 if (R_tab == NULL)
110 goto failure;
111
112 if (parm->peakrate.rate) {
113 P_tab = qdisc_get_rtab(&parm->peakrate,
114 tb[TCA_POLICE_PEAKRATE], NULL);
115 if (P_tab == NULL)
116 goto failure;
117 }
118 }
119
120 if (est) {
121 err = gen_replace_estimator(&police->tcf_bstats,
122 police->common.cpu_bstats,
123 &police->tcf_rate_est,
124 &police->tcf_lock,
125 NULL, est);
126 if (err)
127 goto failure;
128 } else if (tb[TCA_POLICE_AVRATE] &&
129 (ret == ACT_P_CREATED ||
130 !gen_estimator_active(&police->tcf_rate_est))) {
131 err = -EINVAL;
132 goto failure;
133 }
134
135 if (tb[TCA_POLICE_RESULT]) {
136 tcfp_result = nla_get_u32(tb[TCA_POLICE_RESULT]);
137 if (TC_ACT_EXT_CMP(tcfp_result, TC_ACT_GOTO_CHAIN)) {
138 NL_SET_ERR_MSG(extack,
139 "goto chain not allowed on fallback");
140 err = -EINVAL;
141 goto failure;
142 }
143 }
144
145 new = kzalloc(sizeof(*new), GFP_KERNEL);
146 if (unlikely(!new)) {
147 err = -ENOMEM;
148 goto failure;
149 }
150
151 /* No failure allowed after this point */
152 new->tcfp_result = tcfp_result;
153 new->tcfp_mtu = parm->mtu;
154 if (!new->tcfp_mtu) {
155 new->tcfp_mtu = ~0;
156 if (R_tab)
157 new->tcfp_mtu = 255 << R_tab->rate.cell_log;
158 }
159 if (R_tab) {
160 new->rate_present = true;
161 rate64 = tb[TCA_POLICE_RATE64] ?
162 nla_get_u64(tb[TCA_POLICE_RATE64]) : 0;
163 psched_ratecfg_precompute(&new->rate, &R_tab->rate, rate64);
164 qdisc_put_rtab(R_tab);
165 } else {
166 new->rate_present = false;
167 }
168 if (P_tab) {
169 new->peak_present = true;
170 prate64 = tb[TCA_POLICE_PEAKRATE64] ?
171 nla_get_u64(tb[TCA_POLICE_PEAKRATE64]) : 0;
172 psched_ratecfg_precompute(&new->peak, &P_tab->rate, prate64);
173 qdisc_put_rtab(P_tab);
174 } else {
175 new->peak_present = false;
176 }
177
178 new->tcfp_burst = PSCHED_TICKS2NS(parm->burst);
179 if (new->peak_present)
180 new->tcfp_mtu_ptoks = (s64)psched_l2t_ns(&new->peak,
181 new->tcfp_mtu);
182
183 if (tb[TCA_POLICE_AVRATE])
184 new->tcfp_ewma_rate = nla_get_u32(tb[TCA_POLICE_AVRATE]);
185
186 spin_lock_bh(&police->tcf_lock);
187 spin_lock_bh(&police->tcfp_lock);
188 police->tcfp_t_c = ktime_get_ns();
189 police->tcfp_toks = new->tcfp_burst;
190 if (new->peak_present)
191 police->tcfp_ptoks = new->tcfp_mtu_ptoks;
192 spin_unlock_bh(&police->tcfp_lock);
193 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
194 new = rcu_replace_pointer(police->params,
195 new,
196 lockdep_is_held(&police->tcf_lock));
197 spin_unlock_bh(&police->tcf_lock);
198
199 if (goto_ch)
200 tcf_chain_put_by_act(goto_ch);
201 if (new)
202 kfree_rcu(new, rcu);
203
204 return ret;
205
206 failure:
207 qdisc_put_rtab(P_tab);
208 qdisc_put_rtab(R_tab);
209 if (goto_ch)
210 tcf_chain_put_by_act(goto_ch);
211 release_idr:
212 tcf_idr_release(*a, bind);
213 return err;
214 }
215
tcf_police_mtu_check(struct sk_buff * skb,u32 limit)216 static bool tcf_police_mtu_check(struct sk_buff *skb, u32 limit)
217 {
218 u32 len;
219
220 if (skb_is_gso(skb))
221 return skb_gso_validate_mac_len(skb, limit);
222
223 len = qdisc_pkt_len(skb);
224 if (skb_at_tc_ingress(skb))
225 len += skb->mac_len;
226
227 return len <= limit;
228 }
229
tcf_police_act(struct sk_buff * skb,const struct tc_action * a,struct tcf_result * res)230 static int tcf_police_act(struct sk_buff *skb, const struct tc_action *a,
231 struct tcf_result *res)
232 {
233 struct tcf_police *police = to_police(a);
234 struct tcf_police_params *p;
235 s64 now, toks, ptoks = 0;
236 int ret;
237
238 tcf_lastuse_update(&police->tcf_tm);
239 bstats_cpu_update(this_cpu_ptr(police->common.cpu_bstats), skb);
240
241 ret = READ_ONCE(police->tcf_action);
242 p = rcu_dereference_bh(police->params);
243
244 if (p->tcfp_ewma_rate) {
245 struct gnet_stats_rate_est64 sample;
246
247 if (!gen_estimator_read(&police->tcf_rate_est, &sample) ||
248 sample.bps >= p->tcfp_ewma_rate)
249 goto inc_overlimits;
250 }
251
252 if (tcf_police_mtu_check(skb, p->tcfp_mtu)) {
253 if (!p->rate_present) {
254 ret = p->tcfp_result;
255 goto end;
256 }
257
258 now = ktime_get_ns();
259 spin_lock_bh(&police->tcfp_lock);
260 toks = min_t(s64, now - police->tcfp_t_c, p->tcfp_burst);
261 if (p->peak_present) {
262 ptoks = toks + police->tcfp_ptoks;
263 if (ptoks > p->tcfp_mtu_ptoks)
264 ptoks = p->tcfp_mtu_ptoks;
265 ptoks -= (s64)psched_l2t_ns(&p->peak,
266 qdisc_pkt_len(skb));
267 }
268 toks += police->tcfp_toks;
269 if (toks > p->tcfp_burst)
270 toks = p->tcfp_burst;
271 toks -= (s64)psched_l2t_ns(&p->rate, qdisc_pkt_len(skb));
272 if ((toks|ptoks) >= 0) {
273 police->tcfp_t_c = now;
274 police->tcfp_toks = toks;
275 police->tcfp_ptoks = ptoks;
276 spin_unlock_bh(&police->tcfp_lock);
277 ret = p->tcfp_result;
278 goto inc_drops;
279 }
280 spin_unlock_bh(&police->tcfp_lock);
281 }
282
283 inc_overlimits:
284 qstats_overlimit_inc(this_cpu_ptr(police->common.cpu_qstats));
285 inc_drops:
286 if (ret == TC_ACT_SHOT)
287 qstats_drop_inc(this_cpu_ptr(police->common.cpu_qstats));
288 end:
289 return ret;
290 }
291
tcf_police_cleanup(struct tc_action * a)292 static void tcf_police_cleanup(struct tc_action *a)
293 {
294 struct tcf_police *police = to_police(a);
295 struct tcf_police_params *p;
296
297 p = rcu_dereference_protected(police->params, 1);
298 if (p)
299 kfree_rcu(p, rcu);
300 }
301
tcf_police_stats_update(struct tc_action * a,u64 bytes,u64 packets,u64 drops,u64 lastuse,bool hw)302 static void tcf_police_stats_update(struct tc_action *a,
303 u64 bytes, u64 packets, u64 drops,
304 u64 lastuse, bool hw)
305 {
306 struct tcf_police *police = to_police(a);
307 struct tcf_t *tm = &police->tcf_tm;
308
309 tcf_action_update_stats(a, bytes, packets, drops, hw);
310 tm->lastuse = max_t(u64, tm->lastuse, lastuse);
311 }
312
tcf_police_dump(struct sk_buff * skb,struct tc_action * a,int bind,int ref)313 static int tcf_police_dump(struct sk_buff *skb, struct tc_action *a,
314 int bind, int ref)
315 {
316 unsigned char *b = skb_tail_pointer(skb);
317 struct tcf_police *police = to_police(a);
318 struct tcf_police_params *p;
319 struct tc_police opt = {
320 .index = police->tcf_index,
321 .refcnt = refcount_read(&police->tcf_refcnt) - ref,
322 .bindcnt = atomic_read(&police->tcf_bindcnt) - bind,
323 };
324 struct tcf_t t;
325
326 spin_lock_bh(&police->tcf_lock);
327 opt.action = police->tcf_action;
328 p = rcu_dereference_protected(police->params,
329 lockdep_is_held(&police->tcf_lock));
330 opt.mtu = p->tcfp_mtu;
331 opt.burst = PSCHED_NS2TICKS(p->tcfp_burst);
332 if (p->rate_present) {
333 psched_ratecfg_getrate(&opt.rate, &p->rate);
334 if ((police->params->rate.rate_bytes_ps >= (1ULL << 32)) &&
335 nla_put_u64_64bit(skb, TCA_POLICE_RATE64,
336 police->params->rate.rate_bytes_ps,
337 TCA_POLICE_PAD))
338 goto nla_put_failure;
339 }
340 if (p->peak_present) {
341 psched_ratecfg_getrate(&opt.peakrate, &p->peak);
342 if ((police->params->peak.rate_bytes_ps >= (1ULL << 32)) &&
343 nla_put_u64_64bit(skb, TCA_POLICE_PEAKRATE64,
344 police->params->peak.rate_bytes_ps,
345 TCA_POLICE_PAD))
346 goto nla_put_failure;
347 }
348 if (nla_put(skb, TCA_POLICE_TBF, sizeof(opt), &opt))
349 goto nla_put_failure;
350 if (p->tcfp_result &&
351 nla_put_u32(skb, TCA_POLICE_RESULT, p->tcfp_result))
352 goto nla_put_failure;
353 if (p->tcfp_ewma_rate &&
354 nla_put_u32(skb, TCA_POLICE_AVRATE, p->tcfp_ewma_rate))
355 goto nla_put_failure;
356
357 tcf_tm_dump(&t, &police->tcf_tm);
358 if (nla_put_64bit(skb, TCA_POLICE_TM, sizeof(t), &t, TCA_POLICE_PAD))
359 goto nla_put_failure;
360 spin_unlock_bh(&police->tcf_lock);
361
362 return skb->len;
363
364 nla_put_failure:
365 spin_unlock_bh(&police->tcf_lock);
366 nlmsg_trim(skb, b);
367 return -1;
368 }
369
tcf_police_search(struct net * net,struct tc_action ** a,u32 index)370 static int tcf_police_search(struct net *net, struct tc_action **a, u32 index)
371 {
372 struct tc_action_net *tn = net_generic(net, police_net_id);
373
374 return tcf_idr_search(tn, a, index);
375 }
376
377 MODULE_AUTHOR("Alexey Kuznetsov");
378 MODULE_DESCRIPTION("Policing actions");
379 MODULE_LICENSE("GPL");
380
381 static struct tc_action_ops act_police_ops = {
382 .kind = "police",
383 .id = TCA_ID_POLICE,
384 .owner = THIS_MODULE,
385 .stats_update = tcf_police_stats_update,
386 .act = tcf_police_act,
387 .dump = tcf_police_dump,
388 .init = tcf_police_init,
389 .walk = tcf_police_walker,
390 .lookup = tcf_police_search,
391 .cleanup = tcf_police_cleanup,
392 .size = sizeof(struct tcf_police),
393 };
394
police_init_net(struct net * net)395 static __net_init int police_init_net(struct net *net)
396 {
397 struct tc_action_net *tn = net_generic(net, police_net_id);
398
399 return tc_action_net_init(net, tn, &act_police_ops);
400 }
401
police_exit_net(struct list_head * net_list)402 static void __net_exit police_exit_net(struct list_head *net_list)
403 {
404 tc_action_net_exit(net_list, police_net_id);
405 }
406
407 static struct pernet_operations police_net_ops = {
408 .init = police_init_net,
409 .exit_batch = police_exit_net,
410 .id = &police_net_id,
411 .size = sizeof(struct tc_action_net),
412 };
413
police_init_module(void)414 static int __init police_init_module(void)
415 {
416 return tcf_register_action(&act_police_ops, &police_net_ops);
417 }
418
police_cleanup_module(void)419 static void __exit police_cleanup_module(void)
420 {
421 tcf_unregister_action(&act_police_ops, &police_net_ops);
422 }
423
424 module_init(police_init_module);
425 module_exit(police_cleanup_module);
426