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 #include <net/tc_wrapper.h> 23 24 /* Each policer is serialized by its individual spinlock */ 25 26 static struct tc_action_ops act_police_ops; 27 28 static const struct nla_policy police_policy[TCA_POLICE_MAX + 1] = { 29 [TCA_POLICE_RATE] = { .len = TC_RTAB_SIZE }, 30 [TCA_POLICE_PEAKRATE] = { .len = TC_RTAB_SIZE }, 31 [TCA_POLICE_AVRATE] = { .type = NLA_U32 }, 32 [TCA_POLICE_RESULT] = { .type = NLA_U32 }, 33 [TCA_POLICE_RATE64] = { .type = NLA_U64 }, 34 [TCA_POLICE_PEAKRATE64] = { .type = NLA_U64 }, 35 [TCA_POLICE_PKTRATE64] = { .type = NLA_U64, .min = 1 }, 36 [TCA_POLICE_PKTBURST64] = { .type = NLA_U64, .min = 1 }, 37 }; 38 39 static int tcf_police_init(struct net *net, struct nlattr *nla, 40 struct nlattr *est, struct tc_action **a, 41 struct tcf_proto *tp, u32 flags, 42 struct netlink_ext_ack *extack) 43 { 44 int ret = 0, tcfp_result = TC_ACT_OK, err, size; 45 bool bind = flags & TCA_ACT_FLAGS_BIND; 46 struct nlattr *tb[TCA_POLICE_MAX + 1]; 47 struct tcf_chain *goto_ch = NULL; 48 struct tc_police *parm; 49 struct tcf_police *police; 50 struct qdisc_rate_table *R_tab = NULL, *P_tab = NULL; 51 struct tc_action_net *tn = net_generic(net, act_police_ops.net_id); 52 struct tcf_police_params *new; 53 bool exists = false; 54 u32 index; 55 u64 rate64, prate64; 56 u64 pps, ppsburst; 57 58 if (nla == NULL) 59 return -EINVAL; 60 61 err = nla_parse_nested_deprecated(tb, TCA_POLICE_MAX, nla, 62 police_policy, NULL); 63 if (err < 0) 64 return err; 65 66 if (tb[TCA_POLICE_TBF] == NULL) 67 return -EINVAL; 68 size = nla_len(tb[TCA_POLICE_TBF]); 69 if (size != sizeof(*parm) && size != sizeof(struct tc_police_compat)) 70 return -EINVAL; 71 72 parm = nla_data(tb[TCA_POLICE_TBF]); 73 index = parm->index; 74 err = tcf_idr_check_alloc(tn, &index, a, bind); 75 if (err < 0) 76 return err; 77 exists = err; 78 if (exists && bind) 79 return 0; 80 81 if (!exists) { 82 ret = tcf_idr_create(tn, index, NULL, a, 83 &act_police_ops, bind, true, flags); 84 if (ret) { 85 tcf_idr_cleanup(tn, index); 86 return ret; 87 } 88 ret = ACT_P_CREATED; 89 spin_lock_init(&(to_police(*a)->tcfp_lock)); 90 } else if (!(flags & TCA_ACT_FLAGS_REPLACE)) { 91 tcf_idr_release(*a, bind); 92 return -EEXIST; 93 } 94 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack); 95 if (err < 0) 96 goto release_idr; 97 98 police = to_police(*a); 99 if (parm->rate.rate) { 100 err = -ENOMEM; 101 R_tab = qdisc_get_rtab(&parm->rate, tb[TCA_POLICE_RATE], NULL); 102 if (R_tab == NULL) 103 goto failure; 104 105 if (parm->peakrate.rate) { 106 P_tab = qdisc_get_rtab(&parm->peakrate, 107 tb[TCA_POLICE_PEAKRATE], NULL); 108 if (P_tab == NULL) 109 goto failure; 110 } 111 } 112 113 if (est) { 114 err = gen_replace_estimator(&police->tcf_bstats, 115 police->common.cpu_bstats, 116 &police->tcf_rate_est, 117 &police->tcf_lock, 118 false, est); 119 if (err) 120 goto failure; 121 } else if (tb[TCA_POLICE_AVRATE] && 122 (ret == ACT_P_CREATED || 123 !gen_estimator_active(&police->tcf_rate_est))) { 124 err = -EINVAL; 125 goto failure; 126 } 127 128 if (tb[TCA_POLICE_RESULT]) { 129 tcfp_result = nla_get_u32(tb[TCA_POLICE_RESULT]); 130 if (TC_ACT_EXT_CMP(tcfp_result, TC_ACT_GOTO_CHAIN)) { 131 NL_SET_ERR_MSG(extack, 132 "goto chain not allowed on fallback"); 133 err = -EINVAL; 134 goto failure; 135 } 136 } 137 138 if ((tb[TCA_POLICE_PKTRATE64] && !tb[TCA_POLICE_PKTBURST64]) || 139 (!tb[TCA_POLICE_PKTRATE64] && tb[TCA_POLICE_PKTBURST64])) { 140 NL_SET_ERR_MSG(extack, 141 "Both or neither packet-per-second burst and rate must be provided"); 142 err = -EINVAL; 143 goto failure; 144 } 145 146 if (tb[TCA_POLICE_PKTRATE64] && R_tab) { 147 NL_SET_ERR_MSG(extack, 148 "packet-per-second and byte-per-second rate limits not allowed in same action"); 149 err = -EINVAL; 150 goto failure; 151 } 152 153 new = kzalloc(sizeof(*new), GFP_KERNEL); 154 if (unlikely(!new)) { 155 err = -ENOMEM; 156 goto failure; 157 } 158 159 /* No failure allowed after this point */ 160 new->tcfp_result = tcfp_result; 161 new->tcfp_mtu = parm->mtu; 162 if (!new->tcfp_mtu) { 163 new->tcfp_mtu = ~0; 164 if (R_tab) 165 new->tcfp_mtu = 255 << R_tab->rate.cell_log; 166 } 167 if (R_tab) { 168 new->rate_present = true; 169 rate64 = tb[TCA_POLICE_RATE64] ? 170 nla_get_u64(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 = tb[TCA_POLICE_PEAKRATE64] ? 179 nla_get_u64(tb[TCA_POLICE_PEAKRATE64]) : 0; 180 psched_ratecfg_precompute(&new->peak, &P_tab->rate, prate64); 181 qdisc_put_rtab(P_tab); 182 } else { 183 new->peak_present = false; 184 } 185 186 new->tcfp_burst = PSCHED_TICKS2NS(parm->burst); 187 if (new->peak_present) 188 new->tcfp_mtu_ptoks = (s64)psched_l2t_ns(&new->peak, 189 new->tcfp_mtu); 190 191 if (tb[TCA_POLICE_AVRATE]) 192 new->tcfp_ewma_rate = nla_get_u32(tb[TCA_POLICE_AVRATE]); 193 194 if (tb[TCA_POLICE_PKTRATE64]) { 195 pps = nla_get_u64(tb[TCA_POLICE_PKTRATE64]); 196 ppsburst = nla_get_u64(tb[TCA_POLICE_PKTBURST64]); 197 new->pps_present = true; 198 new->tcfp_pkt_burst = PSCHED_TICKS2NS(ppsburst); 199 psched_ppscfg_precompute(&new->ppsrate, pps); 200 } 201 202 spin_lock_bh(&police->tcf_lock); 203 spin_lock_bh(&police->tcfp_lock); 204 police->tcfp_t_c = ktime_get_ns(); 205 police->tcfp_toks = new->tcfp_burst; 206 if (new->peak_present) 207 police->tcfp_ptoks = new->tcfp_mtu_ptoks; 208 spin_unlock_bh(&police->tcfp_lock); 209 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch); 210 new = rcu_replace_pointer(police->params, 211 new, 212 lockdep_is_held(&police->tcf_lock)); 213 spin_unlock_bh(&police->tcf_lock); 214 215 if (goto_ch) 216 tcf_chain_put_by_act(goto_ch); 217 if (new) 218 kfree_rcu(new, rcu); 219 220 return ret; 221 222 failure: 223 qdisc_put_rtab(P_tab); 224 qdisc_put_rtab(R_tab); 225 if (goto_ch) 226 tcf_chain_put_by_act(goto_ch); 227 release_idr: 228 tcf_idr_release(*a, bind); 229 return err; 230 } 231 232 static bool tcf_police_mtu_check(struct sk_buff *skb, u32 limit) 233 { 234 u32 len; 235 236 if (skb_is_gso(skb)) 237 return skb_gso_validate_mac_len(skb, limit); 238 239 len = qdisc_pkt_len(skb); 240 if (skb_at_tc_ingress(skb)) 241 len += skb->mac_len; 242 243 return len <= limit; 244 } 245 246 TC_INDIRECT_SCOPE int tcf_police_act(struct sk_buff *skb, 247 const struct tc_action *a, 248 struct tcf_result *res) 249 { 250 struct tcf_police *police = to_police(a); 251 s64 now, toks, ppstoks = 0, ptoks = 0; 252 struct tcf_police_params *p; 253 int ret; 254 255 tcf_lastuse_update(&police->tcf_tm); 256 bstats_update(this_cpu_ptr(police->common.cpu_bstats), skb); 257 258 ret = READ_ONCE(police->tcf_action); 259 p = rcu_dereference_bh(police->params); 260 261 if (p->tcfp_ewma_rate) { 262 struct gnet_stats_rate_est64 sample; 263 264 if (!gen_estimator_read(&police->tcf_rate_est, &sample) || 265 sample.bps >= p->tcfp_ewma_rate) 266 goto inc_overlimits; 267 } 268 269 if (tcf_police_mtu_check(skb, p->tcfp_mtu)) { 270 if (!p->rate_present && !p->pps_present) { 271 ret = p->tcfp_result; 272 goto end; 273 } 274 275 now = ktime_get_ns(); 276 spin_lock_bh(&police->tcfp_lock); 277 toks = min_t(s64, now - police->tcfp_t_c, p->tcfp_burst); 278 if (p->peak_present) { 279 ptoks = toks + police->tcfp_ptoks; 280 if (ptoks > p->tcfp_mtu_ptoks) 281 ptoks = p->tcfp_mtu_ptoks; 282 ptoks -= (s64)psched_l2t_ns(&p->peak, 283 qdisc_pkt_len(skb)); 284 } 285 if (p->rate_present) { 286 toks += police->tcfp_toks; 287 if (toks > p->tcfp_burst) 288 toks = p->tcfp_burst; 289 toks -= (s64)psched_l2t_ns(&p->rate, qdisc_pkt_len(skb)); 290 } else if (p->pps_present) { 291 ppstoks = min_t(s64, now - police->tcfp_t_c, p->tcfp_pkt_burst); 292 ppstoks += police->tcfp_pkttoks; 293 if (ppstoks > p->tcfp_pkt_burst) 294 ppstoks = p->tcfp_pkt_burst; 295 ppstoks -= (s64)psched_pkt2t_ns(&p->ppsrate, 1); 296 } 297 if ((toks | ptoks | ppstoks) >= 0) { 298 police->tcfp_t_c = now; 299 police->tcfp_toks = toks; 300 police->tcfp_ptoks = ptoks; 301 police->tcfp_pkttoks = ppstoks; 302 spin_unlock_bh(&police->tcfp_lock); 303 ret = p->tcfp_result; 304 goto inc_drops; 305 } 306 spin_unlock_bh(&police->tcfp_lock); 307 } 308 309 inc_overlimits: 310 qstats_overlimit_inc(this_cpu_ptr(police->common.cpu_qstats)); 311 inc_drops: 312 if (ret == TC_ACT_SHOT) 313 qstats_drop_inc(this_cpu_ptr(police->common.cpu_qstats)); 314 end: 315 return ret; 316 } 317 318 static void tcf_police_cleanup(struct tc_action *a) 319 { 320 struct tcf_police *police = to_police(a); 321 struct tcf_police_params *p; 322 323 p = rcu_dereference_protected(police->params, 1); 324 if (p) 325 kfree_rcu(p, rcu); 326 } 327 328 static void tcf_police_stats_update(struct tc_action *a, 329 u64 bytes, u64 packets, u64 drops, 330 u64 lastuse, bool hw) 331 { 332 struct tcf_police *police = to_police(a); 333 struct tcf_t *tm = &police->tcf_tm; 334 335 tcf_action_update_stats(a, bytes, packets, drops, hw); 336 tm->lastuse = max_t(u64, tm->lastuse, lastuse); 337 } 338 339 static int tcf_police_dump(struct sk_buff *skb, struct tc_action *a, 340 int bind, int ref) 341 { 342 unsigned char *b = skb_tail_pointer(skb); 343 struct tcf_police *police = to_police(a); 344 struct tcf_police_params *p; 345 struct tc_police opt = { 346 .index = police->tcf_index, 347 .refcnt = refcount_read(&police->tcf_refcnt) - ref, 348 .bindcnt = atomic_read(&police->tcf_bindcnt) - bind, 349 }; 350 struct tcf_t t; 351 352 spin_lock_bh(&police->tcf_lock); 353 opt.action = police->tcf_action; 354 p = rcu_dereference_protected(police->params, 355 lockdep_is_held(&police->tcf_lock)); 356 opt.mtu = p->tcfp_mtu; 357 opt.burst = PSCHED_NS2TICKS(p->tcfp_burst); 358 if (p->rate_present) { 359 psched_ratecfg_getrate(&opt.rate, &p->rate); 360 if ((police->params->rate.rate_bytes_ps >= (1ULL << 32)) && 361 nla_put_u64_64bit(skb, TCA_POLICE_RATE64, 362 police->params->rate.rate_bytes_ps, 363 TCA_POLICE_PAD)) 364 goto nla_put_failure; 365 } 366 if (p->peak_present) { 367 psched_ratecfg_getrate(&opt.peakrate, &p->peak); 368 if ((police->params->peak.rate_bytes_ps >= (1ULL << 32)) && 369 nla_put_u64_64bit(skb, TCA_POLICE_PEAKRATE64, 370 police->params->peak.rate_bytes_ps, 371 TCA_POLICE_PAD)) 372 goto nla_put_failure; 373 } 374 if (p->pps_present) { 375 if (nla_put_u64_64bit(skb, TCA_POLICE_PKTRATE64, 376 police->params->ppsrate.rate_pkts_ps, 377 TCA_POLICE_PAD)) 378 goto nla_put_failure; 379 if (nla_put_u64_64bit(skb, TCA_POLICE_PKTBURST64, 380 PSCHED_NS2TICKS(p->tcfp_pkt_burst), 381 TCA_POLICE_PAD)) 382 goto nla_put_failure; 383 } 384 if (nla_put(skb, TCA_POLICE_TBF, sizeof(opt), &opt)) 385 goto nla_put_failure; 386 if (p->tcfp_result && 387 nla_put_u32(skb, TCA_POLICE_RESULT, p->tcfp_result)) 388 goto nla_put_failure; 389 if (p->tcfp_ewma_rate && 390 nla_put_u32(skb, TCA_POLICE_AVRATE, p->tcfp_ewma_rate)) 391 goto nla_put_failure; 392 393 tcf_tm_dump(&t, &police->tcf_tm); 394 if (nla_put_64bit(skb, TCA_POLICE_TM, sizeof(t), &t, TCA_POLICE_PAD)) 395 goto nla_put_failure; 396 spin_unlock_bh(&police->tcf_lock); 397 398 return skb->len; 399 400 nla_put_failure: 401 spin_unlock_bh(&police->tcf_lock); 402 nlmsg_trim(skb, b); 403 return -1; 404 } 405 406 static int tcf_police_act_to_flow_act(int tc_act, u32 *extval, 407 struct netlink_ext_ack *extack) 408 { 409 int act_id = -EOPNOTSUPP; 410 411 if (!TC_ACT_EXT_OPCODE(tc_act)) { 412 if (tc_act == TC_ACT_OK) 413 act_id = FLOW_ACTION_ACCEPT; 414 else if (tc_act == TC_ACT_SHOT) 415 act_id = FLOW_ACTION_DROP; 416 else if (tc_act == TC_ACT_PIPE) 417 act_id = FLOW_ACTION_PIPE; 418 else if (tc_act == TC_ACT_RECLASSIFY) 419 NL_SET_ERR_MSG_MOD(extack, "Offload not supported when conform/exceed action is \"reclassify\""); 420 else 421 NL_SET_ERR_MSG_MOD(extack, "Unsupported conform/exceed action offload"); 422 } else if (TC_ACT_EXT_CMP(tc_act, TC_ACT_GOTO_CHAIN)) { 423 act_id = FLOW_ACTION_GOTO; 424 *extval = tc_act & TC_ACT_EXT_VAL_MASK; 425 } else if (TC_ACT_EXT_CMP(tc_act, TC_ACT_JUMP)) { 426 act_id = FLOW_ACTION_JUMP; 427 *extval = tc_act & TC_ACT_EXT_VAL_MASK; 428 } else if (tc_act == TC_ACT_UNSPEC) { 429 act_id = FLOW_ACTION_CONTINUE; 430 } else { 431 NL_SET_ERR_MSG_MOD(extack, "Unsupported conform/exceed action offload"); 432 } 433 434 return act_id; 435 } 436 437 static int tcf_police_offload_act_setup(struct tc_action *act, void *entry_data, 438 u32 *index_inc, bool bind, 439 struct netlink_ext_ack *extack) 440 { 441 if (bind) { 442 struct flow_action_entry *entry = entry_data; 443 struct tcf_police *police = to_police(act); 444 struct tcf_police_params *p; 445 int act_id; 446 447 p = rcu_dereference_protected(police->params, 448 lockdep_is_held(&police->tcf_lock)); 449 450 entry->id = FLOW_ACTION_POLICE; 451 entry->police.burst = tcf_police_burst(act); 452 entry->police.rate_bytes_ps = 453 tcf_police_rate_bytes_ps(act); 454 entry->police.peakrate_bytes_ps = tcf_police_peakrate_bytes_ps(act); 455 entry->police.avrate = tcf_police_tcfp_ewma_rate(act); 456 entry->police.overhead = tcf_police_rate_overhead(act); 457 entry->police.burst_pkt = tcf_police_burst_pkt(act); 458 entry->police.rate_pkt_ps = 459 tcf_police_rate_pkt_ps(act); 460 entry->police.mtu = tcf_police_tcfp_mtu(act); 461 462 act_id = tcf_police_act_to_flow_act(police->tcf_action, 463 &entry->police.exceed.extval, 464 extack); 465 if (act_id < 0) 466 return act_id; 467 468 entry->police.exceed.act_id = act_id; 469 470 act_id = tcf_police_act_to_flow_act(p->tcfp_result, 471 &entry->police.notexceed.extval, 472 extack); 473 if (act_id < 0) 474 return act_id; 475 476 entry->police.notexceed.act_id = act_id; 477 478 *index_inc = 1; 479 } else { 480 struct flow_offload_action *fl_action = entry_data; 481 482 fl_action->id = FLOW_ACTION_POLICE; 483 } 484 485 return 0; 486 } 487 488 MODULE_AUTHOR("Alexey Kuznetsov"); 489 MODULE_DESCRIPTION("Policing actions"); 490 MODULE_LICENSE("GPL"); 491 492 static struct tc_action_ops act_police_ops = { 493 .kind = "police", 494 .id = TCA_ID_POLICE, 495 .owner = THIS_MODULE, 496 .stats_update = tcf_police_stats_update, 497 .act = tcf_police_act, 498 .dump = tcf_police_dump, 499 .init = tcf_police_init, 500 .cleanup = tcf_police_cleanup, 501 .offload_act_setup = tcf_police_offload_act_setup, 502 .size = sizeof(struct tcf_police), 503 }; 504 505 static __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 } 511 512 static 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 } 516 517 static 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 }; 523 524 static int __init police_init_module(void) 525 { 526 return tcf_register_action(&act_police_ops, &police_net_ops); 527 } 528 529 static void __exit police_cleanup_module(void) 530 { 531 tcf_unregister_action(&act_police_ops, &police_net_ops); 532 } 533 534 module_init(police_init_module); 535 module_exit(police_cleanup_module); 536