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 27 static struct tc_action_ops act_police_ops; 28 29 static 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 40 static 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; 220 221 failure: 222 qdisc_put_rtab(P_tab); 223 qdisc_put_rtab(R_tab); 224 if (goto_ch) 225 tcf_chain_put_by_act(goto_ch); 226 release_idr: 227 tcf_idr_release(*a, bind); 228 return err; 229 } 230 231 static 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 } 244 245 TC_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 } 307 308 inc_overlimits: 309 qstats_overlimit_inc(this_cpu_ptr(police->common.cpu_qstats)); 310 inc_drops: 311 if (ret == TC_ACT_SHOT) 312 qstats_drop_inc(this_cpu_ptr(police->common.cpu_qstats)); 313 end: 314 return ret; 315 } 316 317 static 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 } 326 327 static 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 } 337 338 static 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; 398 399 nla_put_failure: 400 spin_unlock_bh(&police->tcf_lock); 401 nlmsg_trim(skb, b); 402 return -1; 403 } 404 405 static 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 } 435 436 static 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 } 486 487 MODULE_AUTHOR("Alexey Kuznetsov"); 488 MODULE_DESCRIPTION("Policing actions"); 489 MODULE_LICENSE("GPL"); 490 491 static 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 }; 503 MODULE_ALIAS_NET_ACT("police"); 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