Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | ||
1da177e4 LT |
13 | #include <linux/module.h> |
14 | #include <linux/types.h> | |
15 | #include <linux/kernel.h> | |
1da177e4 | 16 | #include <linux/string.h> |
1da177e4 | 17 | #include <linux/errno.h> |
1da177e4 | 18 | #include <linux/skbuff.h> |
1da177e4 LT |
19 | #include <linux/rtnetlink.h> |
20 | #include <linux/init.h> | |
5a0e3ad6 | 21 | #include <linux/slab.h> |
1da177e4 | 22 | #include <net/act_api.h> |
dc5fc579 | 23 | #include <net/netlink.h> |
1da177e4 | 24 | |
0e243218 JP |
25 | struct tcf_police { |
26 | struct tcf_common common; | |
27 | int tcfp_result; | |
28 | u32 tcfp_ewma_rate; | |
c6d14ff1 | 29 | s64 tcfp_burst; |
0e243218 | 30 | u32 tcfp_mtu; |
c6d14ff1 JP |
31 | s64 tcfp_toks; |
32 | s64 tcfp_ptoks; | |
33 | s64 tcfp_mtu_ptoks; | |
34 | s64 tcfp_t_c; | |
35 | struct psched_ratecfg rate; | |
36 | bool rate_present; | |
37 | struct psched_ratecfg peak; | |
38 | bool peak_present; | |
0e243218 JP |
39 | }; |
40 | #define to_police(pc) \ | |
41 | container_of(pc, struct tcf_police, common) | |
42 | ||
e9ce1cd3 DM |
43 | #define POL_TAB_MASK 15 |
44 | static struct tcf_common *tcf_police_ht[POL_TAB_MASK + 1]; | |
45 | static u32 police_idx_gen; | |
46 | static DEFINE_RWLOCK(police_lock); | |
1da177e4 | 47 | |
e9ce1cd3 DM |
48 | static struct tcf_hashinfo police_hash_info = { |
49 | .htab = tcf_police_ht, | |
50 | .hmask = POL_TAB_MASK, | |
51 | .lock = &police_lock, | |
52 | }; | |
1da177e4 | 53 | |
1e9b3d53 | 54 | /* old policer structure from before tc actions */ |
cc7ec456 | 55 | struct tc_police_compat { |
1e9b3d53 PM |
56 | u32 index; |
57 | int action; | |
58 | u32 limit; | |
59 | u32 burst; | |
60 | u32 mtu; | |
61 | struct tc_ratespec rate; | |
62 | struct tc_ratespec peakrate; | |
63 | }; | |
64 | ||
e9ce1cd3 | 65 | /* Each policer is serialized by its individual spinlock */ |
1da177e4 | 66 | |
83b950c8 | 67 | static int tcf_act_police_walker(struct sk_buff *skb, struct netlink_callback *cb, |
10297b99 | 68 | int type, struct tc_action *a) |
1da177e4 | 69 | { |
e9ce1cd3 | 70 | struct tcf_common *p; |
1da177e4 | 71 | int err = 0, index = -1, i = 0, s_i = 0, n_i = 0; |
4b3550ef | 72 | struct nlattr *nest; |
1da177e4 | 73 | |
e1e992e5 | 74 | read_lock_bh(&police_lock); |
1da177e4 LT |
75 | |
76 | s_i = cb->args[0]; | |
77 | ||
e9ce1cd3 DM |
78 | for (i = 0; i < (POL_TAB_MASK + 1); i++) { |
79 | p = tcf_police_ht[tcf_hash(i, POL_TAB_MASK)]; | |
1da177e4 | 80 | |
e9ce1cd3 | 81 | for (; p; p = p->tcfc_next) { |
1da177e4 LT |
82 | index++; |
83 | if (index < s_i) | |
84 | continue; | |
85 | a->priv = p; | |
86 | a->order = index; | |
4b3550ef PM |
87 | nest = nla_nest_start(skb, a->order); |
88 | if (nest == NULL) | |
89 | goto nla_put_failure; | |
1da177e4 LT |
90 | if (type == RTM_DELACTION) |
91 | err = tcf_action_dump_1(skb, a, 0, 1); | |
92 | else | |
93 | err = tcf_action_dump_1(skb, a, 0, 0); | |
94 | if (err < 0) { | |
95 | index--; | |
4b3550ef | 96 | nla_nest_cancel(skb, nest); |
1da177e4 LT |
97 | goto done; |
98 | } | |
4b3550ef | 99 | nla_nest_end(skb, nest); |
1da177e4 LT |
100 | n_i++; |
101 | } | |
102 | } | |
103 | done: | |
e1e992e5 | 104 | read_unlock_bh(&police_lock); |
1da177e4 LT |
105 | if (n_i) |
106 | cb->args[0] += n_i; | |
107 | return n_i; | |
108 | ||
7ba699c6 | 109 | nla_put_failure: |
4b3550ef | 110 | nla_nest_cancel(skb, nest); |
1da177e4 LT |
111 | goto done; |
112 | } | |
1da177e4 | 113 | |
c3bc7cff | 114 | static void tcf_police_destroy(struct tcf_police *p) |
1da177e4 | 115 | { |
e9ce1cd3 DM |
116 | unsigned int h = tcf_hash(p->tcf_index, POL_TAB_MASK); |
117 | struct tcf_common **p1p; | |
10297b99 | 118 | |
e9ce1cd3 DM |
119 | for (p1p = &tcf_police_ht[h]; *p1p; p1p = &(*p1p)->tcfc_next) { |
120 | if (*p1p == &p->common) { | |
1da177e4 | 121 | write_lock_bh(&police_lock); |
e9ce1cd3 | 122 | *p1p = p->tcf_next; |
1da177e4 | 123 | write_unlock_bh(&police_lock); |
e9ce1cd3 DM |
124 | gen_kill_estimator(&p->tcf_bstats, |
125 | &p->tcf_rate_est); | |
c7de2cf0 ED |
126 | /* |
127 | * gen_estimator est_timer() might access p->tcf_lock | |
128 | * or bstats, wait a RCU grace period before freeing p | |
129 | */ | |
5957b1ac | 130 | kfree_rcu(p, tcf_rcu); |
1da177e4 LT |
131 | return; |
132 | } | |
133 | } | |
547b792c | 134 | WARN_ON(1); |
1da177e4 LT |
135 | } |
136 | ||
53b2bf3f PM |
137 | static const struct nla_policy police_policy[TCA_POLICE_MAX + 1] = { |
138 | [TCA_POLICE_RATE] = { .len = TC_RTAB_SIZE }, | |
139 | [TCA_POLICE_PEAKRATE] = { .len = TC_RTAB_SIZE }, | |
140 | [TCA_POLICE_AVRATE] = { .type = NLA_U32 }, | |
141 | [TCA_POLICE_RESULT] = { .type = NLA_U32 }, | |
142 | }; | |
143 | ||
c1b52739 BL |
144 | static int tcf_act_police_locate(struct net *net, struct nlattr *nla, |
145 | struct nlattr *est, struct tc_action *a, | |
146 | int ovr, int bind) | |
1da177e4 | 147 | { |
cc7ec456 | 148 | unsigned int h; |
1da177e4 | 149 | int ret = 0, err; |
7ba699c6 | 150 | struct nlattr *tb[TCA_POLICE_MAX + 1]; |
1da177e4 | 151 | struct tc_police *parm; |
e9ce1cd3 | 152 | struct tcf_police *police; |
1da177e4 | 153 | struct qdisc_rate_table *R_tab = NULL, *P_tab = NULL; |
1e9b3d53 | 154 | int size; |
1da177e4 | 155 | |
cee63723 | 156 | if (nla == NULL) |
1da177e4 LT |
157 | return -EINVAL; |
158 | ||
53b2bf3f | 159 | err = nla_parse_nested(tb, TCA_POLICE_MAX, nla, police_policy); |
cee63723 PM |
160 | if (err < 0) |
161 | return err; | |
162 | ||
7ba699c6 | 163 | if (tb[TCA_POLICE_TBF] == NULL) |
1e9b3d53 | 164 | return -EINVAL; |
7ba699c6 | 165 | size = nla_len(tb[TCA_POLICE_TBF]); |
1e9b3d53 | 166 | if (size != sizeof(*parm) && size != sizeof(struct tc_police_compat)) |
1da177e4 | 167 | return -EINVAL; |
7ba699c6 | 168 | parm = nla_data(tb[TCA_POLICE_TBF]); |
1da177e4 | 169 | |
e9ce1cd3 DM |
170 | if (parm->index) { |
171 | struct tcf_common *pc; | |
172 | ||
173 | pc = tcf_hash_lookup(parm->index, &police_hash_info); | |
174 | if (pc != NULL) { | |
175 | a->priv = pc; | |
176 | police = to_police(pc); | |
177 | if (bind) { | |
178 | police->tcf_bindcnt += 1; | |
179 | police->tcf_refcnt += 1; | |
180 | } | |
181 | if (ovr) | |
182 | goto override; | |
183 | return ret; | |
1da177e4 | 184 | } |
1da177e4 LT |
185 | } |
186 | ||
e9ce1cd3 DM |
187 | police = kzalloc(sizeof(*police), GFP_KERNEL); |
188 | if (police == NULL) | |
1da177e4 | 189 | return -ENOMEM; |
1da177e4 | 190 | ret = ACT_P_CREATED; |
e9ce1cd3 DM |
191 | police->tcf_refcnt = 1; |
192 | spin_lock_init(&police->tcf_lock); | |
1da177e4 | 193 | if (bind) |
e9ce1cd3 | 194 | police->tcf_bindcnt = 1; |
1da177e4 LT |
195 | override: |
196 | if (parm->rate.rate) { | |
197 | err = -ENOMEM; | |
7ba699c6 | 198 | R_tab = qdisc_get_rtab(&parm->rate, tb[TCA_POLICE_RATE]); |
1da177e4 LT |
199 | if (R_tab == NULL) |
200 | goto failure; | |
c1b56878 | 201 | |
1da177e4 LT |
202 | if (parm->peakrate.rate) { |
203 | P_tab = qdisc_get_rtab(&parm->peakrate, | |
7ba699c6 | 204 | tb[TCA_POLICE_PEAKRATE]); |
71bcb09a | 205 | if (P_tab == NULL) |
1da177e4 | 206 | goto failure; |
1da177e4 LT |
207 | } |
208 | } | |
71bcb09a | 209 | |
e9ce1cd3 | 210 | spin_lock_bh(&police->tcf_lock); |
71bcb09a SH |
211 | if (est) { |
212 | err = gen_replace_estimator(&police->tcf_bstats, | |
213 | &police->tcf_rate_est, | |
214 | &police->tcf_lock, est); | |
215 | if (err) | |
216 | goto failure_unlock; | |
a883bf56 JP |
217 | } else if (tb[TCA_POLICE_AVRATE] && |
218 | (ret == ACT_P_CREATED || | |
219 | !gen_estimator_active(&police->tcf_bstats, | |
220 | &police->tcf_rate_est))) { | |
221 | err = -EINVAL; | |
222 | goto failure_unlock; | |
71bcb09a SH |
223 | } |
224 | ||
225 | /* No failure allowed after this point */ | |
c6d14ff1 JP |
226 | police->tcfp_mtu = parm->mtu; |
227 | if (police->tcfp_mtu == 0) { | |
228 | police->tcfp_mtu = ~0; | |
229 | if (R_tab) | |
230 | police->tcfp_mtu = 255 << R_tab->rate.cell_log; | |
231 | } | |
232 | if (R_tab) { | |
233 | police->rate_present = true; | |
01cb71d2 | 234 | psched_ratecfg_precompute(&police->rate, &R_tab->rate); |
c6d14ff1 JP |
235 | qdisc_put_rtab(R_tab); |
236 | } else { | |
237 | police->rate_present = false; | |
1da177e4 | 238 | } |
c6d14ff1 JP |
239 | if (P_tab) { |
240 | police->peak_present = true; | |
01cb71d2 | 241 | psched_ratecfg_precompute(&police->peak, &P_tab->rate); |
c6d14ff1 JP |
242 | qdisc_put_rtab(P_tab); |
243 | } else { | |
244 | police->peak_present = false; | |
1da177e4 LT |
245 | } |
246 | ||
7ba699c6 | 247 | if (tb[TCA_POLICE_RESULT]) |
1587bac4 | 248 | police->tcfp_result = nla_get_u32(tb[TCA_POLICE_RESULT]); |
c6d14ff1 JP |
249 | police->tcfp_burst = PSCHED_TICKS2NS(parm->burst); |
250 | police->tcfp_toks = police->tcfp_burst; | |
251 | if (police->peak_present) { | |
252 | police->tcfp_mtu_ptoks = (s64) psched_l2t_ns(&police->peak, | |
253 | police->tcfp_mtu); | |
254 | police->tcfp_ptoks = police->tcfp_mtu_ptoks; | |
1da177e4 | 255 | } |
e9ce1cd3 | 256 | police->tcf_action = parm->action; |
1da177e4 | 257 | |
7ba699c6 | 258 | if (tb[TCA_POLICE_AVRATE]) |
1587bac4 | 259 | police->tcfp_ewma_rate = nla_get_u32(tb[TCA_POLICE_AVRATE]); |
1da177e4 | 260 | |
e9ce1cd3 | 261 | spin_unlock_bh(&police->tcf_lock); |
1da177e4 LT |
262 | if (ret != ACT_P_CREATED) |
263 | return ret; | |
264 | ||
c6d14ff1 | 265 | police->tcfp_t_c = ktime_to_ns(ktime_get()); |
e9ce1cd3 DM |
266 | police->tcf_index = parm->index ? parm->index : |
267 | tcf_hash_new_index(&police_idx_gen, &police_hash_info); | |
268 | h = tcf_hash(police->tcf_index, POL_TAB_MASK); | |
1da177e4 | 269 | write_lock_bh(&police_lock); |
e9ce1cd3 DM |
270 | police->tcf_next = tcf_police_ht[h]; |
271 | tcf_police_ht[h] = &police->common; | |
1da177e4 LT |
272 | write_unlock_bh(&police_lock); |
273 | ||
e9ce1cd3 | 274 | a->priv = police; |
1da177e4 LT |
275 | return ret; |
276 | ||
71bcb09a SH |
277 | failure_unlock: |
278 | spin_unlock_bh(&police->tcf_lock); | |
1da177e4 | 279 | failure: |
71bcb09a SH |
280 | if (P_tab) |
281 | qdisc_put_rtab(P_tab); | |
282 | if (R_tab) | |
283 | qdisc_put_rtab(R_tab); | |
1da177e4 | 284 | if (ret == ACT_P_CREATED) |
e9ce1cd3 | 285 | kfree(police); |
1da177e4 LT |
286 | return err; |
287 | } | |
288 | ||
289 | static int tcf_act_police_cleanup(struct tc_action *a, int bind) | |
290 | { | |
e9ce1cd3 | 291 | struct tcf_police *p = a->priv; |
c3bc7cff | 292 | int ret = 0; |
1da177e4 | 293 | |
c3bc7cff PM |
294 | if (p != NULL) { |
295 | if (bind) | |
296 | p->tcf_bindcnt--; | |
297 | ||
298 | p->tcf_refcnt--; | |
299 | if (p->tcf_refcnt <= 0 && !p->tcf_bindcnt) { | |
300 | tcf_police_destroy(p); | |
301 | ret = 1; | |
302 | } | |
303 | } | |
304 | return ret; | |
1da177e4 LT |
305 | } |
306 | ||
dc7f9f6e | 307 | static int tcf_act_police(struct sk_buff *skb, const struct tc_action *a, |
10297b99 | 308 | struct tcf_result *res) |
1da177e4 | 309 | { |
e9ce1cd3 | 310 | struct tcf_police *police = a->priv; |
c6d14ff1 JP |
311 | s64 now; |
312 | s64 toks; | |
313 | s64 ptoks = 0; | |
1da177e4 | 314 | |
e9ce1cd3 | 315 | spin_lock(&police->tcf_lock); |
1da177e4 | 316 | |
bfe0d029 | 317 | bstats_update(&police->tcf_bstats, skb); |
1da177e4 | 318 | |
e9ce1cd3 DM |
319 | if (police->tcfp_ewma_rate && |
320 | police->tcf_rate_est.bps >= police->tcfp_ewma_rate) { | |
321 | police->tcf_qstats.overlimits++; | |
b9647580 JP |
322 | if (police->tcf_action == TC_ACT_SHOT) |
323 | police->tcf_qstats.drops++; | |
e9ce1cd3 DM |
324 | spin_unlock(&police->tcf_lock); |
325 | return police->tcf_action; | |
1da177e4 | 326 | } |
1da177e4 | 327 | |
0abf77e5 | 328 | if (qdisc_pkt_len(skb) <= police->tcfp_mtu) { |
c6d14ff1 | 329 | if (!police->rate_present) { |
e9ce1cd3 DM |
330 | spin_unlock(&police->tcf_lock); |
331 | return police->tcfp_result; | |
1da177e4 LT |
332 | } |
333 | ||
c6d14ff1 JP |
334 | now = ktime_to_ns(ktime_get()); |
335 | toks = min_t(s64, now - police->tcfp_t_c, | |
336 | police->tcfp_burst); | |
337 | if (police->peak_present) { | |
e9ce1cd3 | 338 | ptoks = toks + police->tcfp_ptoks; |
c6d14ff1 JP |
339 | if (ptoks > police->tcfp_mtu_ptoks) |
340 | ptoks = police->tcfp_mtu_ptoks; | |
341 | ptoks -= (s64) psched_l2t_ns(&police->peak, | |
342 | qdisc_pkt_len(skb)); | |
1da177e4 | 343 | } |
e9ce1cd3 | 344 | toks += police->tcfp_toks; |
c6d14ff1 | 345 | if (toks > police->tcfp_burst) |
e9ce1cd3 | 346 | toks = police->tcfp_burst; |
c6d14ff1 | 347 | toks -= (s64) psched_l2t_ns(&police->rate, qdisc_pkt_len(skb)); |
1da177e4 | 348 | if ((toks|ptoks) >= 0) { |
e9ce1cd3 DM |
349 | police->tcfp_t_c = now; |
350 | police->tcfp_toks = toks; | |
351 | police->tcfp_ptoks = ptoks; | |
352 | spin_unlock(&police->tcf_lock); | |
353 | return police->tcfp_result; | |
1da177e4 LT |
354 | } |
355 | } | |
356 | ||
e9ce1cd3 | 357 | police->tcf_qstats.overlimits++; |
b9647580 JP |
358 | if (police->tcf_action == TC_ACT_SHOT) |
359 | police->tcf_qstats.drops++; | |
e9ce1cd3 DM |
360 | spin_unlock(&police->tcf_lock); |
361 | return police->tcf_action; | |
1da177e4 LT |
362 | } |
363 | ||
364 | static int | |
365 | tcf_act_police_dump(struct sk_buff *skb, struct tc_action *a, int bind, int ref) | |
366 | { | |
27a884dc | 367 | unsigned char *b = skb_tail_pointer(skb); |
e9ce1cd3 | 368 | struct tcf_police *police = a->priv; |
0f04cfd0 JM |
369 | struct tc_police opt = { |
370 | .index = police->tcf_index, | |
371 | .action = police->tcf_action, | |
372 | .mtu = police->tcfp_mtu, | |
c6d14ff1 | 373 | .burst = PSCHED_NS2TICKS(police->tcfp_burst), |
0f04cfd0 JM |
374 | .refcnt = police->tcf_refcnt - ref, |
375 | .bindcnt = police->tcf_bindcnt - bind, | |
376 | }; | |
377 | ||
c6d14ff1 | 378 | if (police->rate_present) |
01cb71d2 | 379 | psched_ratecfg_getrate(&opt.rate, &police->rate); |
c6d14ff1 | 380 | if (police->peak_present) |
01cb71d2 | 381 | psched_ratecfg_getrate(&opt.peakrate, &police->peak); |
1b34ec43 DM |
382 | if (nla_put(skb, TCA_POLICE_TBF, sizeof(opt), &opt)) |
383 | goto nla_put_failure; | |
384 | if (police->tcfp_result && | |
385 | nla_put_u32(skb, TCA_POLICE_RESULT, police->tcfp_result)) | |
386 | goto nla_put_failure; | |
387 | if (police->tcfp_ewma_rate && | |
388 | nla_put_u32(skb, TCA_POLICE_AVRATE, police->tcfp_ewma_rate)) | |
389 | goto nla_put_failure; | |
1da177e4 LT |
390 | return skb->len; |
391 | ||
7ba699c6 | 392 | nla_put_failure: |
dc5fc579 | 393 | nlmsg_trim(skb, b); |
1da177e4 LT |
394 | return -1; |
395 | } | |
396 | ||
397 | MODULE_AUTHOR("Alexey Kuznetsov"); | |
398 | MODULE_DESCRIPTION("Policing actions"); | |
399 | MODULE_LICENSE("GPL"); | |
400 | ||
401 | static struct tc_action_ops act_police_ops = { | |
402 | .kind = "police", | |
e9ce1cd3 | 403 | .hinfo = &police_hash_info, |
1da177e4 LT |
404 | .type = TCA_ID_POLICE, |
405 | .capab = TCA_CAP_NONE, | |
406 | .owner = THIS_MODULE, | |
407 | .act = tcf_act_police, | |
408 | .dump = tcf_act_police_dump, | |
409 | .cleanup = tcf_act_police_cleanup, | |
e9ce1cd3 | 410 | .lookup = tcf_hash_search, |
1da177e4 | 411 | .init = tcf_act_police_locate, |
83b950c8 | 412 | .walk = tcf_act_police_walker |
1da177e4 LT |
413 | }; |
414 | ||
415 | static int __init | |
416 | police_init_module(void) | |
417 | { | |
418 | return tcf_register_action(&act_police_ops); | |
419 | } | |
420 | ||
421 | static void __exit | |
422 | police_cleanup_module(void) | |
423 | { | |
424 | tcf_unregister_action(&act_police_ops); | |
425 | } | |
426 | ||
427 | module_init(police_init_module); | |
428 | module_exit(police_cleanup_module); |