1 /*- 2 * 3 * SPDX-License-Identifier: BSD-3-Clause 4 * 5 * Copyright (c) 2018-2019 6 * Netflix Inc. 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28 * SUCH DAMAGE. 29 * 30 */ 31 /** 32 * Author: Randall Stewart <rrs@netflix.com> 33 */ 34 35 #include <sys/cdefs.h> 36 __FBSDID("$FreeBSD$"); 37 #include "opt_inet.h" 38 #include "opt_inet6.h" 39 #include "opt_ipsec.h" 40 #include "opt_tcpdebug.h" 41 #include "opt_ratelimit.h" 42 #include <sys/param.h> 43 #include <sys/kernel.h> 44 #include <sys/malloc.h> 45 #include <sys/mbuf.h> 46 #include <sys/socket.h> 47 #include <sys/socketvar.h> 48 #ifdef KERN_TLS 49 #include <sys/sockbuf_tls.h> 50 #endif 51 #include <sys/sysctl.h> 52 #include <sys/eventhandler.h> 53 #include <sys/mutex.h> 54 #include <sys/ck.h> 55 #define TCPSTATES /* for logging */ 56 #include <netinet/in.h> 57 #include <netinet/in_pcb.h> 58 #include <netinet/tcp_var.h> 59 #ifdef INET6 60 #include <netinet6/tcp6_var.h> 61 #endif 62 #include <netinet/tcp_ratelimit.h> 63 #ifndef USECS_IN_SECOND 64 #define USECS_IN_SECOND 1000000 65 #endif 66 /* 67 * For the purposes of each send, what is the size 68 * of an ethernet frame. 69 */ 70 #ifndef ETHERNET_SEGMENT_SIZE 71 #define ETHERNET_SEGMENT_SIZE 1500 72 #endif 73 MALLOC_DEFINE(M_TCPPACE, "tcp_hwpace", "TCP Hardware pacing memory"); 74 #ifdef RATELIMIT 75 76 #define COMMON_RATE 180500 77 uint64_t desired_rates[] = { 78 62500, /* 500Kbps */ 79 180500, /* 1.44Mpbs */ 80 375000, /* 3Mbps */ 81 500000, /* 4Mbps */ 82 625000, /* 5Mbps */ 83 750000, /* 6Mbps */ 84 1000000, /* 8Mbps */ 85 1250000, /* 10Mbps */ 86 2500000, /* 20Mbps */ 87 3750000, /* 30Mbps */ 88 5000000, /* 40Meg */ 89 6250000, /* 50Mbps */ 90 12500000, /* 100Mbps */ 91 25000000, /* 200Mbps */ 92 50000000, /* 400Mbps */ 93 100000000, /* 800Mbps */ 94 12500, /* 100kbps */ 95 25000, /* 200kbps */ 96 875000, /* 7Mbps */ 97 1125000, /* 9Mbps */ 98 1875000, /* 15Mbps */ 99 3125000, /* 25Mbps */ 100 8125000, /* 65Mbps */ 101 10000000, /* 80Mbps */ 102 18750000, /* 150Mbps */ 103 20000000, /* 250Mbps */ 104 37500000, /* 350Mbps */ 105 62500000, /* 500Mbps */ 106 78125000, /* 625Mbps */ 107 125000000, /* 1Gbps */ 108 }; 109 #define MAX_HDWR_RATES (sizeof(desired_rates)/sizeof(uint64_t)) 110 #define RS_ORDERED_COUNT 16 /* 111 * Number that are in order 112 * at the beginning of the table, 113 * over this a sort is required. 114 */ 115 #define RS_NEXT_ORDER_GROUP 16 /* 116 * The point in our table where 117 * we come fill in a second ordered 118 * group (index wise means -1). 119 */ 120 #define ALL_HARDWARE_RATES 1004 /* 121 * 1Meg - 1Gig in 1 Meg steps 122 * plus 100, 200k and 500k and 123 * 10Gig 124 */ 125 126 #define RS_ONE_MEGABIT_PERSEC 1000000 127 #define RS_ONE_GIGABIT_PERSEC 1000000000 128 #define RS_TEN_GIGABIT_PERSEC 10000000000 129 130 static struct head_tcp_rate_set int_rs; 131 static struct mtx rs_mtx; 132 uint32_t rs_number_alive; 133 uint32_t rs_number_dead; 134 135 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, rl, CTLFLAG_RW, 0, 136 "TCP Ratelimit stats"); 137 SYSCTL_UINT(_net_inet_tcp_rl, OID_AUTO, alive, CTLFLAG_RW, 138 &rs_number_alive, 0, 139 "Number of interfaces initialized for ratelimiting"); 140 SYSCTL_UINT(_net_inet_tcp_rl, OID_AUTO, dead, CTLFLAG_RW, 141 &rs_number_dead, 0, 142 "Number of interfaces departing from ratelimiting"); 143 144 static void 145 rl_add_syctl_entries(struct sysctl_oid *rl_sysctl_root, struct tcp_rate_set *rs) 146 { 147 /* 148 * Add sysctl entries for thus interface. 149 */ 150 if (rs->rs_flags & RS_INTF_NO_SUP) { 151 SYSCTL_ADD_S32(&rs->sysctl_ctx, 152 SYSCTL_CHILDREN(rl_sysctl_root), 153 OID_AUTO, "disable", CTLFLAG_RD, 154 &rs->rs_disable, 0, 155 "Disable this interface from new hdwr limiting?"); 156 } else { 157 SYSCTL_ADD_S32(&rs->sysctl_ctx, 158 SYSCTL_CHILDREN(rl_sysctl_root), 159 OID_AUTO, "disable", CTLFLAG_RW, 160 &rs->rs_disable, 0, 161 "Disable this interface from new hdwr limiting?"); 162 } 163 SYSCTL_ADD_S32(&rs->sysctl_ctx, 164 SYSCTL_CHILDREN(rl_sysctl_root), 165 OID_AUTO, "minseg", CTLFLAG_RW, 166 &rs->rs_min_seg, 0, 167 "What is the minimum we need to send on this interface?"); 168 SYSCTL_ADD_U64(&rs->sysctl_ctx, 169 SYSCTL_CHILDREN(rl_sysctl_root), 170 OID_AUTO, "flow_limit", CTLFLAG_RW, 171 &rs->rs_flow_limit, 0, 172 "What is the limit for number of flows (0=unlimited)?"); 173 SYSCTL_ADD_S32(&rs->sysctl_ctx, 174 SYSCTL_CHILDREN(rl_sysctl_root), 175 OID_AUTO, "highest", CTLFLAG_RD, 176 &rs->rs_highest_valid, 0, 177 "Highest valid rate"); 178 SYSCTL_ADD_S32(&rs->sysctl_ctx, 179 SYSCTL_CHILDREN(rl_sysctl_root), 180 OID_AUTO, "lowest", CTLFLAG_RD, 181 &rs->rs_lowest_valid, 0, 182 "Lowest valid rate"); 183 SYSCTL_ADD_S32(&rs->sysctl_ctx, 184 SYSCTL_CHILDREN(rl_sysctl_root), 185 OID_AUTO, "flags", CTLFLAG_RD, 186 &rs->rs_flags, 0, 187 "What lags are on the entry?"); 188 SYSCTL_ADD_S32(&rs->sysctl_ctx, 189 SYSCTL_CHILDREN(rl_sysctl_root), 190 OID_AUTO, "numrates", CTLFLAG_RD, 191 &rs->rs_rate_cnt, 0, 192 "How many rates re there?"); 193 SYSCTL_ADD_U64(&rs->sysctl_ctx, 194 SYSCTL_CHILDREN(rl_sysctl_root), 195 OID_AUTO, "flows_using", CTLFLAG_RD, 196 &rs->rs_flows_using, 0, 197 "How many flows are using this interface now?"); 198 #ifdef DETAILED_RATELIMIT_SYSCTL 199 if (rs->rs_rlt && rs->rs_rate_cnt > 0) { 200 /* Lets display the rates */ 201 int i; 202 struct sysctl_oid *rl_rates; 203 struct sysctl_oid *rl_rate_num; 204 char rate_num[16]; 205 rl_rates = SYSCTL_ADD_NODE(&rs->sysctl_ctx, 206 SYSCTL_CHILDREN(rl_sysctl_root), 207 OID_AUTO, 208 "rate", 209 CTLFLAG_RW, 0, 210 "Ratelist"); 211 for( i = 0; i < rs->rs_rate_cnt; i++) { 212 sprintf(rate_num, "%d", i); 213 rl_rate_num = SYSCTL_ADD_NODE(&rs->sysctl_ctx, 214 SYSCTL_CHILDREN(rl_rates), 215 OID_AUTO, 216 rate_num, 217 CTLFLAG_RW, 0, 218 "Individual Rate"); 219 SYSCTL_ADD_U32(&rs->sysctl_ctx, 220 SYSCTL_CHILDREN(rl_rate_num), 221 OID_AUTO, "flags", CTLFLAG_RD, 222 &rs->rs_rlt[i].flags, 0, 223 "Flags on this rate"); 224 SYSCTL_ADD_U32(&rs->sysctl_ctx, 225 SYSCTL_CHILDREN(rl_rate_num), 226 OID_AUTO, "pacetime", CTLFLAG_RD, 227 &rs->rs_rlt[i].time_between, 0, 228 "Time hardware inserts between 1500 byte sends"); 229 SYSCTL_ADD_U64(&rs->sysctl_ctx, 230 SYSCTL_CHILDREN(rl_rate_num), 231 OID_AUTO, "rate", CTLFLAG_RD, 232 &rs->rs_rlt[i].rate, 0, 233 "Rate in bytes per second"); 234 } 235 } 236 #endif 237 } 238 239 static void 240 rs_destroy(epoch_context_t ctx) 241 { 242 struct tcp_rate_set *rs; 243 244 rs = __containerof(ctx, struct tcp_rate_set, rs_epoch_ctx); 245 mtx_lock(&rs_mtx); 246 rs->rs_flags &= ~RS_FUNERAL_SCHD; 247 if (rs->rs_flows_using == 0) { 248 /* 249 * In theory its possible (but unlikely) 250 * that while the delete was occuring 251 * and we were applying the DEAD flag 252 * someone slipped in and found the 253 * interface in a lookup. While we 254 * decided rs_flows_using were 0 and 255 * scheduling the epoch_call, the other 256 * thread incremented rs_flow_using. This 257 * is because users have a pointer and 258 * we only use the rs_flows_using in an 259 * atomic fashion, i.e. the other entities 260 * are not protected. To assure this did 261 * not occur, we check rs_flows_using here 262 * before deleteing. 263 */ 264 sysctl_ctx_free(&rs->sysctl_ctx); 265 free(rs->rs_rlt, M_TCPPACE); 266 free(rs, M_TCPPACE); 267 rs_number_dead--; 268 } 269 mtx_unlock(&rs_mtx); 270 271 } 272 273 #ifdef INET 274 extern counter_u64_t rate_limit_set_ok; 275 extern counter_u64_t rate_limit_active; 276 extern counter_u64_t rate_limit_alloc_fail; 277 #endif 278 279 static int 280 rl_attach_txrtlmt(struct ifnet *ifp, 281 uint32_t flowtype, 282 int flowid, 283 uint64_t cfg_rate, 284 struct m_snd_tag **tag) 285 { 286 int error; 287 union if_snd_tag_alloc_params params = { 288 .rate_limit.hdr.type = IF_SND_TAG_TYPE_RATE_LIMIT, 289 .rate_limit.hdr.flowid = flowid, 290 .rate_limit.hdr.flowtype = flowtype, 291 .rate_limit.max_rate = cfg_rate, 292 .rate_limit.flags = M_NOWAIT, 293 }; 294 295 if (ifp->if_snd_tag_alloc == NULL) { 296 error = EOPNOTSUPP; 297 } else { 298 error = ifp->if_snd_tag_alloc(ifp, ¶ms, tag); 299 #ifdef INET 300 if (error == 0) { 301 if_ref((*tag)->ifp); 302 counter_u64_add(rate_limit_set_ok, 1); 303 counter_u64_add(rate_limit_active, 1); 304 } else 305 counter_u64_add(rate_limit_alloc_fail, 1); 306 #endif 307 } 308 return (error); 309 } 310 311 static void 312 populate_canned_table(struct tcp_rate_set *rs, const uint64_t *rate_table_act) 313 { 314 /* 315 * The internal table is "special", it 316 * is two seperate ordered tables that 317 * must be merged. We get here when the 318 * adapter specifies a number of rates that 319 * covers both ranges in the table in some 320 * form. 321 */ 322 int i, at_low, at_high; 323 uint8_t low_disabled = 0, high_disabled = 0; 324 325 for(i = 0, at_low = 0, at_high = RS_NEXT_ORDER_GROUP; i < rs->rs_rate_cnt; i++) { 326 rs->rs_rlt[i].flags = 0; 327 rs->rs_rlt[i].time_between = 0; 328 if ((low_disabled == 0) && 329 (high_disabled || 330 (rate_table_act[at_low] < rate_table_act[at_high]))) { 331 rs->rs_rlt[i].rate = rate_table_act[at_low]; 332 at_low++; 333 if (at_low == RS_NEXT_ORDER_GROUP) 334 low_disabled = 1; 335 } else if (high_disabled == 0) { 336 rs->rs_rlt[i].rate = rate_table_act[at_high]; 337 at_high++; 338 if (at_high == MAX_HDWR_RATES) 339 high_disabled = 1; 340 } 341 } 342 } 343 344 static struct tcp_rate_set * 345 rt_setup_new_rs(struct ifnet *ifp, int *error) 346 { 347 struct tcp_rate_set *rs; 348 const uint64_t *rate_table_act; 349 uint64_t lentim, res; 350 size_t sz; 351 uint32_t hash_type; 352 int i; 353 struct if_ratelimit_query_results rl; 354 struct sysctl_oid *rl_sysctl_root; 355 /* 356 * We expect to enter with the 357 * mutex locked. 358 */ 359 360 if (ifp->if_ratelimit_query == NULL) { 361 /* 362 * We can do nothing if we cannot 363 * get a query back from the driver. 364 */ 365 return (NULL); 366 } 367 rs = malloc(sizeof(struct tcp_rate_set), M_TCPPACE, M_NOWAIT | M_ZERO); 368 if (rs == NULL) { 369 if (error) 370 *error = ENOMEM; 371 return (NULL); 372 } 373 rl.flags = RT_NOSUPPORT; 374 ifp->if_ratelimit_query(ifp, &rl); 375 if (rl.flags & RT_IS_UNUSABLE) { 376 /* 377 * The interface does not really support 378 * the rate-limiting. 379 */ 380 memset(rs, 0, sizeof(struct tcp_rate_set)); 381 rs->rs_ifp = ifp; 382 rs->rs_if_dunit = ifp->if_dunit; 383 rs->rs_flags = RS_INTF_NO_SUP; 384 rs->rs_disable = 1; 385 rs_number_alive++; 386 sysctl_ctx_init(&rs->sysctl_ctx); 387 rl_sysctl_root = SYSCTL_ADD_NODE(&rs->sysctl_ctx, 388 SYSCTL_STATIC_CHILDREN(_net_inet_tcp_rl), 389 OID_AUTO, 390 rs->rs_ifp->if_xname, 391 CTLFLAG_RW, 0, 392 ""); 393 rl_add_syctl_entries(rl_sysctl_root, rs); 394 mtx_lock(&rs_mtx); 395 CK_LIST_INSERT_HEAD(&int_rs, rs, next); 396 mtx_unlock(&rs_mtx); 397 return (rs); 398 } else if ((rl.flags & RT_IS_INDIRECT) == RT_IS_INDIRECT) { 399 memset(rs, 0, sizeof(struct tcp_rate_set)); 400 rs->rs_ifp = ifp; 401 rs->rs_if_dunit = ifp->if_dunit; 402 rs->rs_flags = RS_IS_DEFF; 403 rs_number_alive++; 404 sysctl_ctx_init(&rs->sysctl_ctx); 405 rl_sysctl_root = SYSCTL_ADD_NODE(&rs->sysctl_ctx, 406 SYSCTL_STATIC_CHILDREN(_net_inet_tcp_rl), 407 OID_AUTO, 408 rs->rs_ifp->if_xname, 409 CTLFLAG_RW, 0, 410 ""); 411 rl_add_syctl_entries(rl_sysctl_root, rs); 412 mtx_lock(&rs_mtx); 413 CK_LIST_INSERT_HEAD(&int_rs, rs, next); 414 mtx_unlock(&rs_mtx); 415 return (rs); 416 } else if ((rl.flags & RT_IS_FIXED_TABLE) == RT_IS_FIXED_TABLE) { 417 /* Mellanox most likely */ 418 rs->rs_ifp = ifp; 419 rs->rs_if_dunit = ifp->if_dunit; 420 rs->rs_rate_cnt = rl.number_of_rates; 421 rs->rs_min_seg = rl.min_segment_burst; 422 rs->rs_highest_valid = 0; 423 rs->rs_flow_limit = rl.max_flows; 424 rs->rs_flags = RS_IS_INTF | RS_NO_PRE; 425 rs->rs_disable = 0; 426 rate_table_act = rl.rate_table; 427 } else if ((rl.flags & RT_IS_SELECTABLE) == RT_IS_SELECTABLE) { 428 /* Chelsio */ 429 rs->rs_ifp = ifp; 430 rs->rs_if_dunit = ifp->if_dunit; 431 rs->rs_rate_cnt = rl.number_of_rates; 432 rs->rs_min_seg = rl.min_segment_burst; 433 rs->rs_disable = 0; 434 rs->rs_flow_limit = rl.max_flows; 435 rate_table_act = desired_rates; 436 if ((rs->rs_rate_cnt > MAX_HDWR_RATES) && 437 (rs->rs_rate_cnt < ALL_HARDWARE_RATES)) { 438 /* 439 * Our desired table is not big 440 * enough, do what we can. 441 */ 442 rs->rs_rate_cnt = MAX_HDWR_RATES; 443 } 444 if (rs->rs_rate_cnt <= RS_ORDERED_COUNT) 445 rs->rs_flags = RS_IS_INTF; 446 else 447 rs->rs_flags = RS_IS_INTF | RS_INT_TBL; 448 if (rs->rs_rate_cnt >= ALL_HARDWARE_RATES) 449 rs->rs_rate_cnt = ALL_HARDWARE_RATES; 450 } else { 451 printf("Interface:%s unit:%d not one known to have rate-limits\n", 452 ifp->if_dname, 453 ifp->if_dunit); 454 free(rs, M_TCPPACE); 455 return (NULL); 456 } 457 sz = sizeof(struct tcp_hwrate_limit_table) * rs->rs_rate_cnt; 458 rs->rs_rlt = malloc(sz, M_TCPPACE, M_NOWAIT); 459 if (rs->rs_rlt == NULL) { 460 if (error) 461 *error = ENOMEM; 462 bail: 463 free(rs, M_TCPPACE); 464 return (NULL); 465 } 466 if (rs->rs_rate_cnt >= ALL_HARDWARE_RATES) { 467 /* 468 * The interface supports all 469 * the rates we could possibly want. 470 */ 471 uint64_t rat; 472 473 rs->rs_rlt[0].rate = 12500; /* 100k */ 474 rs->rs_rlt[1].rate = 25000; /* 200k */ 475 rs->rs_rlt[2].rate = 62500; /* 500k */ 476 /* Note 125000 == 1Megabit 477 * populate 1Meg - 1000meg. 478 */ 479 for(i = 3, rat = 125000; i< (ALL_HARDWARE_RATES-1); i++) { 480 rs->rs_rlt[i].rate = rat; 481 rat += 125000; 482 } 483 rs->rs_rlt[(ALL_HARDWARE_RATES-1)].rate = 1250000000; 484 } else if (rs->rs_flags & RS_INT_TBL) { 485 /* We populate this in a special way */ 486 populate_canned_table(rs, rate_table_act); 487 } else { 488 /* 489 * Just copy in the rates from 490 * the table, it is in order. 491 */ 492 for (i=0; i<rs->rs_rate_cnt; i++) { 493 rs->rs_rlt[i].rate = rate_table_act[i]; 494 rs->rs_rlt[i].time_between = 0; 495 rs->rs_rlt[i].flags = 0; 496 } 497 } 498 for (i = (rs->rs_rate_cnt - 1); i >= 0; i--) { 499 /* 500 * We go backwards through the list so that if we can't get 501 * a rate and fail to init one, we have at least a chance of 502 * getting the highest one. 503 */ 504 rs->rs_rlt[i].ptbl = rs; 505 rs->rs_rlt[i].tag = NULL; 506 /* 507 * Calculate the time between. 508 */ 509 lentim = ETHERNET_SEGMENT_SIZE * USECS_IN_SECOND; 510 res = lentim / rs->rs_rlt[i].rate; 511 if (res > 0) 512 rs->rs_rlt[i].time_between = res; 513 else 514 rs->rs_rlt[i].time_between = 1; 515 if (rs->rs_flags & RS_NO_PRE) { 516 rs->rs_rlt[i].flags = HDWRPACE_INITED; 517 rs->rs_lowest_valid = i; 518 } else { 519 int err; 520 #ifdef RSS 521 hash_type = M_HASHTYPE_RSS_TCP_IPV4; 522 #else 523 hash_type = M_HASHTYPE_OPAQUE_HASH; 524 #endif 525 err = rl_attach_txrtlmt(ifp, 526 hash_type, 527 (i + 1), 528 rs->rs_rlt[i].rate, 529 &rs->rs_rlt[i].tag); 530 if (err) { 531 if (i == (rs->rs_rate_cnt - 1)) { 532 /* 533 * Huh - first rate and we can't get 534 * it? 535 */ 536 free(rs->rs_rlt, M_TCPPACE); 537 if (error) 538 *error = err; 539 goto bail; 540 } else { 541 if (error) 542 *error = err; 543 } 544 break; 545 } else { 546 rs->rs_rlt[i].flags = HDWRPACE_INITED | HDWRPACE_TAGPRESENT; 547 rs->rs_lowest_valid = i; 548 } 549 } 550 } 551 /* Did we get at least 1 rate? */ 552 if (rs->rs_rlt[(rs->rs_rate_cnt - 1)].flags & HDWRPACE_INITED) 553 rs->rs_highest_valid = rs->rs_rate_cnt - 1; 554 else { 555 free(rs->rs_rlt, M_TCPPACE); 556 goto bail; 557 } 558 rs_number_alive++; 559 sysctl_ctx_init(&rs->sysctl_ctx); 560 rl_sysctl_root = SYSCTL_ADD_NODE(&rs->sysctl_ctx, 561 SYSCTL_STATIC_CHILDREN(_net_inet_tcp_rl), 562 OID_AUTO, 563 rs->rs_ifp->if_xname, 564 CTLFLAG_RW, 0, 565 ""); 566 rl_add_syctl_entries(rl_sysctl_root, rs); 567 mtx_lock(&rs_mtx); 568 CK_LIST_INSERT_HEAD(&int_rs, rs, next); 569 mtx_unlock(&rs_mtx); 570 return (rs); 571 } 572 573 static const struct tcp_hwrate_limit_table * 574 tcp_int_find_suitable_rate(const struct tcp_rate_set *rs, 575 uint64_t bytes_per_sec, uint32_t flags) 576 { 577 struct tcp_hwrate_limit_table *arte = NULL, *rte = NULL; 578 uint64_t mbits_per_sec, ind_calc; 579 int i; 580 581 mbits_per_sec = (bytes_per_sec * 8); 582 if (flags & RS_PACING_LT) { 583 if ((mbits_per_sec < RS_ONE_MEGABIT_PERSEC) && 584 (rs->rs_lowest_valid <= 2)){ 585 /* 586 * Smaller than 1Meg, only 587 * 3 entries can match it. 588 */ 589 for(i = rs->rs_lowest_valid; i < 3; i++) { 590 if (bytes_per_sec <= rs->rs_rlt[i].rate) { 591 rte = &rs->rs_rlt[i]; 592 break; 593 } else if (rs->rs_rlt[i].flags & HDWRPACE_INITED) { 594 arte = &rs->rs_rlt[i]; 595 } 596 } 597 goto done; 598 } else if ((mbits_per_sec > RS_ONE_GIGABIT_PERSEC) && 599 (rs->rs_rlt[(ALL_HARDWARE_RATES-1)].flags & HDWRPACE_INITED)){ 600 /* 601 * Larger than 1G (the majority of 602 * our table. 603 */ 604 if (mbits_per_sec < RS_TEN_GIGABIT_PERSEC) 605 rte = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)]; 606 else 607 arte = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)]; 608 goto done; 609 } 610 /* 611 * If we reach here its in our table (between 1Meg - 1000Meg), 612 * just take the rounded down mbits per second, and add 613 * 1Megabit to it, from this we can calculate 614 * the index in the table. 615 */ 616 ind_calc = mbits_per_sec/RS_ONE_MEGABIT_PERSEC; 617 if ((ind_calc * RS_ONE_MEGABIT_PERSEC) != mbits_per_sec) 618 ind_calc++; 619 /* our table is offset by 3, we add 2 */ 620 ind_calc += 2; 621 if (ind_calc > (ALL_HARDWARE_RATES-1)) { 622 /* This should not happen */ 623 ind_calc = ALL_HARDWARE_RATES-1; 624 } 625 if ((ind_calc >= rs->rs_lowest_valid) && 626 (ind_calc <= rs->rs_highest_valid)) 627 rte = &rs->rs_rlt[ind_calc]; 628 } else if (flags & RS_PACING_EXACT_MATCH) { 629 if ((mbits_per_sec < RS_ONE_MEGABIT_PERSEC) && 630 (rs->rs_lowest_valid <= 2)){ 631 for(i = rs->rs_lowest_valid; i < 3; i++) { 632 if (bytes_per_sec == rs->rs_rlt[i].rate) { 633 rte = &rs->rs_rlt[i]; 634 break; 635 } 636 } 637 } else if ((mbits_per_sec > RS_ONE_GIGABIT_PERSEC) && 638 (rs->rs_rlt[(ALL_HARDWARE_RATES-1)].flags & HDWRPACE_INITED)) { 639 /* > 1Gbps only one rate */ 640 if (bytes_per_sec == rs->rs_rlt[(ALL_HARDWARE_RATES-1)].rate) { 641 /* Its 10G wow */ 642 rte = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)]; 643 } 644 } else { 645 /* Ok it must be a exact meg (its between 1G and 1Meg) */ 646 ind_calc = mbits_per_sec/RS_ONE_MEGABIT_PERSEC; 647 if ((ind_calc * RS_ONE_MEGABIT_PERSEC) == mbits_per_sec) { 648 /* its an exact Mbps */ 649 ind_calc += 2; 650 if (ind_calc > (ALL_HARDWARE_RATES-1)) { 651 /* This should not happen */ 652 ind_calc = ALL_HARDWARE_RATES-1; 653 } 654 if (rs->rs_rlt[ind_calc].flags & HDWRPACE_INITED) 655 rte = &rs->rs_rlt[ind_calc]; 656 } 657 } 658 } else { 659 /* we want greater than the requested rate */ 660 if ((mbits_per_sec < RS_ONE_MEGABIT_PERSEC) && 661 (rs->rs_lowest_valid <= 2)){ 662 arte = &rs->rs_rlt[3]; /* set alternate to 1Meg */ 663 for (i=2; i>=rs->rs_lowest_valid; i--) { 664 if (bytes_per_sec < rs->rs_rlt[i].rate) { 665 rte = &rs->rs_rlt[i]; 666 break; 667 } else if ((flags & RS_PACING_GEQ) && 668 (bytes_per_sec == rs->rs_rlt[i].rate)) { 669 rte = &rs->rs_rlt[i]; 670 break; 671 } else { 672 arte = &rs->rs_rlt[i]; /* new alternate */ 673 } 674 } 675 } else if (mbits_per_sec > RS_ONE_GIGABIT_PERSEC) { 676 if ((bytes_per_sec < rs->rs_rlt[(ALL_HARDWARE_RATES-1)].rate) && 677 (rs->rs_rlt[(ALL_HARDWARE_RATES-1)].flags & HDWRPACE_INITED)){ 678 /* Our top rate is larger than the request */ 679 rte = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)]; 680 } else if ((flags & RS_PACING_GEQ) && 681 (bytes_per_sec == rs->rs_rlt[(ALL_HARDWARE_RATES-1)].rate) && 682 (rs->rs_rlt[(ALL_HARDWARE_RATES-1)].flags & HDWRPACE_INITED)) { 683 /* It matches our top rate */ 684 rte = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)]; 685 } else if (rs->rs_rlt[(ALL_HARDWARE_RATES-1)].flags & HDWRPACE_INITED) { 686 /* The top rate is an alternative */ 687 arte = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)]; 688 } 689 } else { 690 /* Its in our range 1Meg - 1Gig */ 691 if (flags & RS_PACING_GEQ) { 692 ind_calc = mbits_per_sec/RS_ONE_MEGABIT_PERSEC; 693 if ((ind_calc * RS_ONE_MEGABIT_PERSEC) == mbits_per_sec) { 694 if (ind_calc > (ALL_HARDWARE_RATES-1)) { 695 /* This should not happen */ 696 ind_calc = (ALL_HARDWARE_RATES-1); 697 } 698 rte = &rs->rs_rlt[ind_calc]; 699 } 700 goto done; 701 } 702 ind_calc = (mbits_per_sec + (RS_ONE_MEGABIT_PERSEC-1))/RS_ONE_MEGABIT_PERSEC; 703 ind_calc += 2; 704 if (ind_calc > (ALL_HARDWARE_RATES-1)) { 705 /* This should not happen */ 706 ind_calc = ALL_HARDWARE_RATES-1; 707 } 708 if (rs->rs_rlt[ind_calc].flags & HDWRPACE_INITED) 709 rte = &rs->rs_rlt[ind_calc]; 710 } 711 } 712 done: 713 if ((rte == NULL) && 714 (arte != NULL) && 715 (flags & RS_PACING_SUB_OK)) { 716 /* We can use the substitute */ 717 rte = arte; 718 } 719 return (rte); 720 } 721 722 static const struct tcp_hwrate_limit_table * 723 tcp_find_suitable_rate(const struct tcp_rate_set *rs, uint64_t bytes_per_sec, uint32_t flags) 724 { 725 /** 726 * Hunt the rate table with the restrictions in flags and find a 727 * suitable rate if possible. 728 * RS_PACING_EXACT_MATCH - look for an exact match to rate. 729 * RS_PACING_GT - must be greater than. 730 * RS_PACING_GEQ - must be greater than or equal. 731 * RS_PACING_LT - must be less than. 732 * RS_PACING_SUB_OK - If we don't meet criteria a 733 * substitute is ok. 734 */ 735 int i, matched; 736 struct tcp_hwrate_limit_table *rte = NULL; 737 738 739 if ((rs->rs_flags & RS_INT_TBL) && 740 (rs->rs_rate_cnt >= ALL_HARDWARE_RATES)) { 741 /* 742 * Here we don't want to paw thru 743 * a big table, we have everything 744 * from 1Meg - 1000Meg in 1Meg increments. 745 * Use an alternate method to "lookup". 746 */ 747 return (tcp_int_find_suitable_rate(rs, bytes_per_sec, flags)); 748 } 749 if ((flags & RS_PACING_LT) || 750 (flags & RS_PACING_EXACT_MATCH)) { 751 /* 752 * For exact and less than we go forward through the table. 753 * This way when we find one larger we stop (exact was a 754 * toss up). 755 */ 756 for (i = rs->rs_lowest_valid, matched = 0; i <= rs->rs_highest_valid; i++) { 757 if ((flags & RS_PACING_EXACT_MATCH) && 758 (bytes_per_sec == rs->rs_rlt[i].rate)) { 759 rte = &rs->rs_rlt[i]; 760 matched = 1; 761 break; 762 } else if ((flags & RS_PACING_LT) && 763 (bytes_per_sec <= rs->rs_rlt[i].rate)) { 764 rte = &rs->rs_rlt[i]; 765 matched = 1; 766 break; 767 } 768 if (bytes_per_sec > rs->rs_rlt[i].rate) 769 break; 770 } 771 if ((matched == 0) && 772 (flags & RS_PACING_LT) && 773 (flags & RS_PACING_SUB_OK)) { 774 /* Kick in a substitute (the lowest) */ 775 rte = &rs->rs_rlt[rs->rs_lowest_valid]; 776 } 777 } else { 778 /* 779 * Here we go backward through the table so that we can find 780 * the one greater in theory faster (but its probably a 781 * wash). 782 */ 783 for (i = rs->rs_highest_valid, matched = 0; i >= rs->rs_lowest_valid; i--) { 784 if (rs->rs_rlt[i].rate > bytes_per_sec) { 785 /* A possible candidate */ 786 rte = &rs->rs_rlt[i]; 787 } 788 if ((flags & RS_PACING_GEQ) && 789 (bytes_per_sec == rs->rs_rlt[i].rate)) { 790 /* An exact match and we want equal */ 791 matched = 1; 792 rte = &rs->rs_rlt[i]; 793 break; 794 } else if (rte) { 795 /* 796 * Found one that is larger than but don't 797 * stop, there may be a more closer match. 798 */ 799 matched = 1; 800 } 801 if (rs->rs_rlt[i].rate < bytes_per_sec) { 802 /* 803 * We found a table entry that is smaller, 804 * stop there will be none greater or equal. 805 */ 806 break; 807 } 808 } 809 if ((matched == 0) && 810 (flags & RS_PACING_SUB_OK)) { 811 /* Kick in a substitute (the highest) */ 812 rte = &rs->rs_rlt[rs->rs_highest_valid]; 813 } 814 } 815 return (rte); 816 } 817 818 static struct ifnet * 819 rt_find_real_interface(struct ifnet *ifp, struct inpcb *inp, int *error) 820 { 821 struct ifnet *tifp; 822 struct m_snd_tag *tag; 823 union if_snd_tag_alloc_params params = { 824 .rate_limit.hdr.type = IF_SND_TAG_TYPE_RATE_LIMIT, 825 .rate_limit.hdr.flowid = 1, 826 .rate_limit.max_rate = COMMON_RATE, 827 .rate_limit.flags = M_NOWAIT, 828 }; 829 int err; 830 #ifdef RSS 831 params.rate_limit.hdr.flowtype = ((inp->inp_vflag & INP_IPV6) ? 832 M_HASHTYPE_RSS_TCP_IPV6 : M_HASHTYPE_RSS_TCP_IPV4); 833 #else 834 params.rate_limit.hdr.flowtype = M_HASHTYPE_OPAQUE_HASH; 835 #endif 836 tag = NULL; 837 if (ifp->if_snd_tag_alloc) { 838 if (error) 839 *error = ENODEV; 840 return (NULL); 841 } 842 err = ifp->if_snd_tag_alloc(ifp, ¶ms, &tag); 843 if (err) { 844 /* Failed to setup a tag? */ 845 if (error) 846 *error = err; 847 return (NULL); 848 } 849 tifp = tag->ifp; 850 tifp->if_snd_tag_free(tag); 851 return (tifp); 852 } 853 854 static const struct tcp_hwrate_limit_table * 855 rt_setup_rate(struct inpcb *inp, struct ifnet *ifp, uint64_t bytes_per_sec, 856 uint32_t flags, int *error) 857 { 858 /* First lets find the interface if it exists */ 859 const struct tcp_hwrate_limit_table *rte; 860 struct tcp_rate_set *rs; 861 struct epoch_tracker et; 862 int err; 863 864 epoch_enter_preempt(net_epoch_preempt, &et); 865 use_real_interface: 866 CK_LIST_FOREACH(rs, &int_rs, next) { 867 /* 868 * Note we don't look with the lock since we either see a 869 * new entry or will get one when we try to add it. 870 */ 871 if (rs->rs_flags & RS_IS_DEAD) { 872 /* The dead are not looked at */ 873 continue; 874 } 875 if ((rs->rs_ifp == ifp) && 876 (rs->rs_if_dunit == ifp->if_dunit)) { 877 /* Ok we found it */ 878 break; 879 } 880 } 881 if ((rs == NULL) || 882 (rs->rs_flags & RS_INTF_NO_SUP) || 883 (rs->rs_flags & RS_IS_DEAD)) { 884 /* 885 * This means we got a packet *before* 886 * the IF-UP was processed below, <or> 887 * while or after we already received an interface 888 * departed event. In either case we really don't 889 * want to do anything with pacing, in 890 * the departing case the packet is not 891 * going to go very far. The new case 892 * might be arguable, but its impossible 893 * to tell from the departing case. 894 */ 895 if (rs->rs_disable && error) 896 *error = ENODEV; 897 epoch_exit_preempt(net_epoch_preempt, &et); 898 return (NULL); 899 } 900 901 if ((rs == NULL) || (rs->rs_disable != 0)) { 902 if (rs->rs_disable && error) 903 *error = ENOSPC; 904 epoch_exit_preempt(net_epoch_preempt, &et); 905 return (NULL); 906 } 907 if (rs->rs_flags & RS_IS_DEFF) { 908 /* We need to find the real interface */ 909 struct ifnet *tifp; 910 911 tifp = rt_find_real_interface(ifp, inp, error); 912 if (tifp == NULL) { 913 if (rs->rs_disable && error) 914 *error = ENOTSUP; 915 epoch_exit_preempt(net_epoch_preempt, &et); 916 return (NULL); 917 } 918 goto use_real_interface; 919 } 920 if (rs->rs_flow_limit && 921 ((rs->rs_flows_using + 1) > rs->rs_flow_limit)) { 922 if (error) 923 *error = ENOSPC; 924 epoch_exit_preempt(net_epoch_preempt, &et); 925 return (NULL); 926 } 927 rte = tcp_find_suitable_rate(rs, bytes_per_sec, flags); 928 if (rte) { 929 err = in_pcbattach_txrtlmt(inp, rs->rs_ifp, 930 inp->inp_flowtype, 931 inp->inp_flowid, 932 rte->rate, 933 &inp->inp_snd_tag); 934 if (err) { 935 /* Failed to attach */ 936 if (error) 937 *error = err; 938 rte = NULL; 939 } 940 } 941 if (rte) { 942 /* 943 * We use an atomic here for accounting so we don't have to 944 * use locks when freeing. 945 */ 946 atomic_add_64(&rs->rs_flows_using, 1); 947 } 948 epoch_exit_preempt(net_epoch_preempt, &et); 949 return (rte); 950 } 951 952 static void 953 tcp_rl_ifnet_link(void *arg __unused, struct ifnet *ifp, int link_state) 954 { 955 int error; 956 struct tcp_rate_set *rs; 957 958 if (((ifp->if_capabilities & IFCAP_TXRTLMT) == 0) || 959 (link_state != LINK_STATE_UP)) { 960 /* 961 * We only care on an interface going up that is rate-limit 962 * capable. 963 */ 964 return; 965 } 966 mtx_lock(&rs_mtx); 967 CK_LIST_FOREACH(rs, &int_rs, next) { 968 if ((rs->rs_ifp == ifp) && 969 (rs->rs_if_dunit == ifp->if_dunit)) { 970 /* We already have initialized this guy */ 971 mtx_unlock(&rs_mtx); 972 return; 973 } 974 } 975 mtx_unlock(&rs_mtx); 976 rt_setup_new_rs(ifp, &error); 977 } 978 979 static void 980 tcp_rl_ifnet_departure(void *arg __unused, struct ifnet *ifp) 981 { 982 struct tcp_rate_set *rs, *nrs; 983 struct ifnet *tifp; 984 int i; 985 986 mtx_lock(&rs_mtx); 987 CK_LIST_FOREACH_SAFE(rs, &int_rs, next, nrs) { 988 if ((rs->rs_ifp == ifp) && 989 (rs->rs_if_dunit == ifp->if_dunit)) { 990 CK_LIST_REMOVE(rs, next); 991 rs_number_alive--; 992 rs_number_dead++; 993 rs->rs_flags |= RS_IS_DEAD; 994 for (i = 0; i < rs->rs_rate_cnt; i++) { 995 if (rs->rs_rlt[i].flags & HDWRPACE_TAGPRESENT) { 996 tifp = rs->rs_rlt[i].tag->ifp; 997 in_pcbdetach_tag(tifp, rs->rs_rlt[i].tag); 998 rs->rs_rlt[i].tag = NULL; 999 } 1000 rs->rs_rlt[i].flags = HDWRPACE_IFPDEPARTED; 1001 } 1002 if (rs->rs_flows_using == 0) { 1003 /* 1004 * No references left, so we can schedule the 1005 * destruction after the epoch (with a caveat). 1006 */ 1007 rs->rs_flags |= RS_FUNERAL_SCHD; 1008 epoch_call(net_epoch, &rs->rs_epoch_ctx, rs_destroy); 1009 } 1010 break; 1011 } 1012 } 1013 mtx_unlock(&rs_mtx); 1014 } 1015 1016 static void 1017 tcp_rl_shutdown(void *arg __unused, int howto __unused) 1018 { 1019 struct tcp_rate_set *rs, *nrs; 1020 struct ifnet *tifp; 1021 int i; 1022 1023 mtx_lock(&rs_mtx); 1024 CK_LIST_FOREACH_SAFE(rs, &int_rs, next, nrs) { 1025 CK_LIST_REMOVE(rs, next); 1026 rs_number_alive--; 1027 rs_number_dead++; 1028 rs->rs_flags |= RS_IS_DEAD; 1029 for (i = 0; i < rs->rs_rate_cnt; i++) { 1030 if (rs->rs_rlt[i].flags & HDWRPACE_TAGPRESENT) { 1031 tifp = rs->rs_rlt[i].tag->ifp; 1032 in_pcbdetach_tag(tifp, rs->rs_rlt[i].tag); 1033 rs->rs_rlt[i].tag = NULL; 1034 } 1035 rs->rs_rlt[i].flags = HDWRPACE_IFPDEPARTED; 1036 } 1037 if (rs->rs_flows_using != 0) { 1038 /* 1039 * We dont hold a reference 1040 * so we have nothing left to 1041 * do. 1042 */ 1043 } else { 1044 /* 1045 * No references left, so we can destroy it 1046 * after the epoch. 1047 */ 1048 rs->rs_flags |= RS_FUNERAL_SCHD; 1049 epoch_call(net_epoch, &rs->rs_epoch_ctx, rs_destroy); 1050 } 1051 } 1052 mtx_unlock(&rs_mtx); 1053 } 1054 1055 const struct tcp_hwrate_limit_table * 1056 tcp_set_pacing_rate(struct tcpcb *tp, struct ifnet *ifp, 1057 uint64_t bytes_per_sec, int flags, int *error) 1058 { 1059 const struct tcp_hwrate_limit_table *rte; 1060 1061 if (tp->t_inpcb->inp_snd_tag == NULL) { 1062 /* 1063 * We are setting up a rate for the first time. 1064 */ 1065 if ((ifp->if_capabilities & IFCAP_TXRTLMT) == 0) { 1066 /* Not supported by the egress */ 1067 if (error) 1068 *error = ENODEV; 1069 return (NULL); 1070 } 1071 #ifdef KERN_TLS 1072 if (tp->t_inpcb->inp_socket->so_snd.sb_tls_flags & SB_TLS_IFNET) { 1073 /* 1074 * We currently can't do both TLS and hardware 1075 * pacing 1076 */ 1077 if (error) 1078 *error = EINVAL; 1079 return (NULL); 1080 } 1081 #endif 1082 rte = rt_setup_rate(tp->t_inpcb, ifp, bytes_per_sec, flags, error); 1083 } else { 1084 /* 1085 * We are modifying a rate, wrong interface? 1086 */ 1087 if (error) 1088 *error = EINVAL; 1089 rte = NULL; 1090 } 1091 return (rte); 1092 } 1093 1094 const struct tcp_hwrate_limit_table * 1095 tcp_chg_pacing_rate(const struct tcp_hwrate_limit_table *crte, 1096 struct tcpcb *tp, struct ifnet *ifp, 1097 uint64_t bytes_per_sec, int flags, int *error) 1098 { 1099 const struct tcp_hwrate_limit_table *nrte; 1100 const struct tcp_rate_set *rs; 1101 int is_indirect = 0; 1102 int err; 1103 1104 1105 if ((tp->t_inpcb->inp_snd_tag == NULL) || 1106 (crte == NULL)) { 1107 /* Wrong interface */ 1108 if (error) 1109 *error = EINVAL; 1110 return (NULL); 1111 } 1112 rs = crte->ptbl; 1113 if ((rs->rs_flags & RS_IS_DEAD) || 1114 (crte->flags & HDWRPACE_IFPDEPARTED)) { 1115 /* Release the rate, and try anew */ 1116 re_rate: 1117 tcp_rel_pacing_rate(crte, tp); 1118 nrte = tcp_set_pacing_rate(tp, ifp, 1119 bytes_per_sec, flags, error); 1120 return (nrte); 1121 } 1122 if ((rs->rs_flags & RT_IS_INDIRECT ) == RT_IS_INDIRECT) 1123 is_indirect = 1; 1124 else 1125 is_indirect = 0; 1126 if ((is_indirect == 0) && 1127 ((ifp != rs->rs_ifp) || 1128 (ifp->if_dunit != rs->rs_if_dunit))) { 1129 /* 1130 * Something changed, the user is not pointing to the same 1131 * ifp? Maybe a route updated on this guy? 1132 */ 1133 goto re_rate; 1134 } else if (is_indirect) { 1135 /* 1136 * For indirect we have to dig in and find the real interface. 1137 */ 1138 struct ifnet *rifp; 1139 1140 rifp = rt_find_real_interface(ifp, tp->t_inpcb, error); 1141 if (rifp == NULL) { 1142 /* Can't find it? */ 1143 goto re_rate; 1144 } 1145 if ((rifp != rs->rs_ifp) || 1146 (ifp->if_dunit != rs->rs_if_dunit)) { 1147 goto re_rate; 1148 } 1149 } 1150 nrte = tcp_find_suitable_rate(rs, bytes_per_sec, flags); 1151 if (nrte == crte) { 1152 /* No change */ 1153 if (error) 1154 *error = 0; 1155 return (crte); 1156 } 1157 if (nrte == NULL) { 1158 /* Release the old rate */ 1159 tcp_rel_pacing_rate(crte, tp); 1160 return (NULL); 1161 } 1162 /* Change rates to our new entry */ 1163 err = in_pcbmodify_txrtlmt(tp->t_inpcb, nrte->rate); 1164 if (err) { 1165 if (error) 1166 *error = err; 1167 return (NULL); 1168 } 1169 if (error) 1170 *error = 0; 1171 return (nrte); 1172 } 1173 1174 void 1175 tcp_rel_pacing_rate(const struct tcp_hwrate_limit_table *crte, struct tcpcb *tp) 1176 { 1177 const struct tcp_rate_set *crs; 1178 struct tcp_rate_set *rs; 1179 uint64_t pre; 1180 1181 crs = crte->ptbl; 1182 /* 1183 * Now we must break the const 1184 * in order to release our refcount. 1185 */ 1186 rs = __DECONST(struct tcp_rate_set *, crs); 1187 pre = atomic_fetchadd_64(&rs->rs_flows_using, -1); 1188 if (pre == 1) { 1189 mtx_lock(&rs_mtx); 1190 /* 1191 * Is it dead? 1192 */ 1193 if ((rs->rs_flags & RS_IS_DEAD) && 1194 ((rs->rs_flags & RS_FUNERAL_SCHD) == 0)){ 1195 /* 1196 * We were the last, 1197 * and a funeral is not pending, so 1198 * we must schedule it. 1199 */ 1200 rs->rs_flags |= RS_FUNERAL_SCHD; 1201 epoch_call(net_epoch, &rs->rs_epoch_ctx, rs_destroy); 1202 } 1203 mtx_unlock(&rs_mtx); 1204 } 1205 in_pcbdetach_txrtlmt(tp->t_inpcb); 1206 } 1207 1208 static eventhandler_tag rl_ifnet_departs; 1209 static eventhandler_tag rl_ifnet_arrives; 1210 static eventhandler_tag rl_shutdown_start; 1211 1212 static void 1213 tcp_rs_init(void *st __unused) 1214 { 1215 CK_LIST_INIT(&int_rs); 1216 rs_number_alive = 0; 1217 rs_number_dead = 0;; 1218 mtx_init(&rs_mtx, "tcp_rs_mtx", "rsmtx", MTX_DEF); 1219 rl_ifnet_departs = EVENTHANDLER_REGISTER(ifnet_departure_event, 1220 tcp_rl_ifnet_departure, 1221 NULL, EVENTHANDLER_PRI_ANY); 1222 rl_ifnet_arrives = EVENTHANDLER_REGISTER(ifnet_link_event, 1223 tcp_rl_ifnet_link, 1224 NULL, EVENTHANDLER_PRI_ANY); 1225 rl_shutdown_start = EVENTHANDLER_REGISTER(shutdown_pre_sync, 1226 tcp_rl_shutdown, NULL, 1227 SHUTDOWN_PRI_FIRST); 1228 printf("TCP_ratelimit: Is now initialized\n"); 1229 } 1230 1231 SYSINIT(tcp_rl_init, SI_SUB_SMP + 1, SI_ORDER_ANY, tcp_rs_init, NULL); 1232 #endif 1233