1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2010-2011 Alexander V. Chernikov <melifaro@ipfw.ru> 5 * Copyright (c) 2004-2005 Gleb Smirnoff <glebius@FreeBSD.org> 6 * Copyright (c) 2001-2003 Roman V. Palagin <romanp@unshadow.net> 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 AUTHOR 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 AUTHOR 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 * $SourceForge: netflow.c,v 1.41 2004/09/05 11:41:10 glebius Exp $ 31 */ 32 33 #include <sys/cdefs.h> 34 __FBSDID("$FreeBSD$"); 35 36 #include "opt_inet6.h" 37 #include "opt_route.h" 38 #include <sys/param.h> 39 #include <sys/bitstring.h> 40 #include <sys/systm.h> 41 #include <sys/counter.h> 42 #include <sys/kernel.h> 43 #include <sys/ktr.h> 44 #include <sys/limits.h> 45 #include <sys/mbuf.h> 46 #include <sys/syslog.h> 47 #include <sys/socket.h> 48 #include <vm/uma.h> 49 50 #include <net/if.h> 51 #include <net/if_dl.h> 52 #include <net/if_var.h> 53 #include <net/route.h> 54 #include <net/ethernet.h> 55 #include <netinet/in.h> 56 #include <netinet/in_systm.h> 57 #include <netinet/ip.h> 58 #include <netinet/ip6.h> 59 #include <netinet/tcp.h> 60 #include <netinet/udp.h> 61 62 #include <netgraph/ng_message.h> 63 #include <netgraph/netgraph.h> 64 65 #include <netgraph/netflow/netflow.h> 66 #include <netgraph/netflow/netflow_v9.h> 67 #include <netgraph/netflow/ng_netflow.h> 68 69 #define NBUCKETS (65536) /* must be power of 2 */ 70 71 /* This hash is for TCP or UDP packets. */ 72 #define FULL_HASH(addr1, addr2, port1, port2) \ 73 (((addr1 ^ (addr1 >> 16) ^ \ 74 htons(addr2 ^ (addr2 >> 16))) ^ \ 75 port1 ^ htons(port2)) & \ 76 (NBUCKETS - 1)) 77 78 /* This hash is for all other IP packets. */ 79 #define ADDR_HASH(addr1, addr2) \ 80 ((addr1 ^ (addr1 >> 16) ^ \ 81 htons(addr2 ^ (addr2 >> 16))) & \ 82 (NBUCKETS - 1)) 83 84 /* Macros to shorten logical constructions */ 85 /* XXX: priv must exist in namespace */ 86 #define INACTIVE(fle) (time_uptime - fle->f.last > priv->nfinfo_inact_t) 87 #define AGED(fle) (time_uptime - fle->f.first > priv->nfinfo_act_t) 88 #define ISFREE(fle) (fle->f.packets == 0) 89 90 /* 91 * 4 is a magical number: statistically number of 4-packet flows is 92 * bigger than 5,6,7...-packet flows by an order of magnitude. Most UDP/ICMP 93 * scans are 1 packet (~ 90% of flow cache). TCP scans are 2-packet in case 94 * of reachable host and 4-packet otherwise. 95 */ 96 #define SMALL(fle) (fle->f.packets <= 4) 97 98 MALLOC_DEFINE(M_NETFLOW_HASH, "netflow_hash", "NetFlow hash"); 99 100 static int export_add(item_p, struct flow_entry *); 101 static int export_send(priv_p, fib_export_p, item_p, int); 102 103 static int hash_insert(priv_p, struct flow_hash_entry *, struct flow_rec *, 104 int, uint8_t, uint8_t); 105 #ifdef INET6 106 static int hash6_insert(priv_p, struct flow_hash_entry *, struct flow6_rec *, 107 int, uint8_t, uint8_t); 108 #endif 109 110 static void expire_flow(priv_p, fib_export_p, struct flow_entry *, int); 111 112 /* 113 * Generate hash for a given flow record. 114 * 115 * FIB is not used here, because: 116 * most VRFS will carry public IPv4 addresses which are unique even 117 * without FIB private addresses can overlap, but this is worked out 118 * via flow_rec bcmp() containing fib id. In IPv6 world addresses are 119 * all globally unique (it's not fully true, there is FC00::/7 for example, 120 * but chances of address overlap are MUCH smaller) 121 */ 122 static inline uint32_t 123 ip_hash(struct flow_rec *r) 124 { 125 126 switch (r->r_ip_p) { 127 case IPPROTO_TCP: 128 case IPPROTO_UDP: 129 return FULL_HASH(r->r_src.s_addr, r->r_dst.s_addr, 130 r->r_sport, r->r_dport); 131 default: 132 return ADDR_HASH(r->r_src.s_addr, r->r_dst.s_addr); 133 } 134 } 135 136 #ifdef INET6 137 /* Generate hash for a given flow6 record. Use lower 4 octets from v6 addresses */ 138 static inline uint32_t 139 ip6_hash(struct flow6_rec *r) 140 { 141 142 switch (r->r_ip_p) { 143 case IPPROTO_TCP: 144 case IPPROTO_UDP: 145 return FULL_HASH(r->src.r_src6.__u6_addr.__u6_addr32[3], 146 r->dst.r_dst6.__u6_addr.__u6_addr32[3], r->r_sport, 147 r->r_dport); 148 default: 149 return ADDR_HASH(r->src.r_src6.__u6_addr.__u6_addr32[3], 150 r->dst.r_dst6.__u6_addr.__u6_addr32[3]); 151 } 152 } 153 154 static inline int 155 ip6_masklen(struct in6_addr *saddr, struct rt_addrinfo *info) 156 { 157 const int nbits = sizeof(*saddr) * NBBY; 158 int mlen; 159 160 if (info->rti_addrs & RTA_NETMASK) 161 bit_count((bitstr_t *)saddr, 0, nbits, &mlen); 162 else 163 mlen = nbits; 164 return (mlen); 165 } 166 #endif 167 168 /* 169 * Detach export datagram from priv, if there is any. 170 * If there is no, allocate a new one. 171 */ 172 static item_p 173 get_export_dgram(priv_p priv, fib_export_p fe) 174 { 175 item_p item = NULL; 176 177 mtx_lock(&fe->export_mtx); 178 if (fe->exp.item != NULL) { 179 item = fe->exp.item; 180 fe->exp.item = NULL; 181 } 182 mtx_unlock(&fe->export_mtx); 183 184 if (item == NULL) { 185 struct netflow_v5_export_dgram *dgram; 186 struct mbuf *m; 187 188 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 189 if (m == NULL) 190 return (NULL); 191 item = ng_package_data(m, NG_NOFLAGS); 192 if (item == NULL) 193 return (NULL); 194 dgram = mtod(m, struct netflow_v5_export_dgram *); 195 dgram->header.count = 0; 196 dgram->header.version = htons(NETFLOW_V5); 197 dgram->header.pad = 0; 198 } 199 200 return (item); 201 } 202 203 /* 204 * Re-attach incomplete datagram back to priv. 205 * If there is already another one, then send incomplete. */ 206 static void 207 return_export_dgram(priv_p priv, fib_export_p fe, item_p item, int flags) 208 { 209 210 /* 211 * It may happen on SMP, that some thread has already 212 * put its item there, in this case we bail out and 213 * send what we have to collector. 214 */ 215 mtx_lock(&fe->export_mtx); 216 if (fe->exp.item == NULL) { 217 fe->exp.item = item; 218 mtx_unlock(&fe->export_mtx); 219 } else { 220 mtx_unlock(&fe->export_mtx); 221 export_send(priv, fe, item, flags); 222 } 223 } 224 225 /* 226 * The flow is over. Call export_add() and free it. If datagram is 227 * full, then call export_send(). 228 */ 229 static void 230 expire_flow(priv_p priv, fib_export_p fe, struct flow_entry *fle, int flags) 231 { 232 struct netflow_export_item exp; 233 uint16_t version = fle->f.version; 234 235 if ((priv->export != NULL) && (version == IPVERSION)) { 236 exp.item = get_export_dgram(priv, fe); 237 if (exp.item == NULL) { 238 priv->nfinfo_export_failed++; 239 if (priv->export9 != NULL) 240 priv->nfinfo_export9_failed++; 241 /* fle definitely contains IPv4 flow. */ 242 uma_zfree_arg(priv->zone, fle, priv); 243 return; 244 } 245 246 if (export_add(exp.item, fle) > 0) 247 export_send(priv, fe, exp.item, flags); 248 else 249 return_export_dgram(priv, fe, exp.item, NG_QUEUE); 250 } 251 252 if (priv->export9 != NULL) { 253 exp.item9 = get_export9_dgram(priv, fe, &exp.item9_opt); 254 if (exp.item9 == NULL) { 255 priv->nfinfo_export9_failed++; 256 if (version == IPVERSION) 257 uma_zfree_arg(priv->zone, fle, priv); 258 #ifdef INET6 259 else if (version == IP6VERSION) 260 uma_zfree_arg(priv->zone6, fle, priv); 261 #endif 262 else 263 panic("ng_netflow: Unknown IP proto: %d", 264 version); 265 return; 266 } 267 268 if (export9_add(exp.item9, exp.item9_opt, fle) > 0) 269 export9_send(priv, fe, exp.item9, exp.item9_opt, flags); 270 else 271 return_export9_dgram(priv, fe, exp.item9, 272 exp.item9_opt, NG_QUEUE); 273 } 274 275 if (version == IPVERSION) 276 uma_zfree_arg(priv->zone, fle, priv); 277 #ifdef INET6 278 else if (version == IP6VERSION) 279 uma_zfree_arg(priv->zone6, fle, priv); 280 #endif 281 } 282 283 /* Get a snapshot of node statistics */ 284 void 285 ng_netflow_copyinfo(priv_p priv, struct ng_netflow_info *i) 286 { 287 288 i->nfinfo_bytes = counter_u64_fetch(priv->nfinfo_bytes); 289 i->nfinfo_packets = counter_u64_fetch(priv->nfinfo_packets); 290 i->nfinfo_bytes6 = counter_u64_fetch(priv->nfinfo_bytes6); 291 i->nfinfo_packets6 = counter_u64_fetch(priv->nfinfo_packets6); 292 i->nfinfo_sbytes = counter_u64_fetch(priv->nfinfo_sbytes); 293 i->nfinfo_spackets = counter_u64_fetch(priv->nfinfo_spackets); 294 i->nfinfo_sbytes6 = counter_u64_fetch(priv->nfinfo_sbytes6); 295 i->nfinfo_spackets6 = counter_u64_fetch(priv->nfinfo_spackets6); 296 i->nfinfo_act_exp = counter_u64_fetch(priv->nfinfo_act_exp); 297 i->nfinfo_inact_exp = counter_u64_fetch(priv->nfinfo_inact_exp); 298 299 i->nfinfo_used = uma_zone_get_cur(priv->zone); 300 #ifdef INET6 301 i->nfinfo_used6 = uma_zone_get_cur(priv->zone6); 302 #endif 303 304 i->nfinfo_alloc_failed = priv->nfinfo_alloc_failed; 305 i->nfinfo_export_failed = priv->nfinfo_export_failed; 306 i->nfinfo_export9_failed = priv->nfinfo_export9_failed; 307 i->nfinfo_realloc_mbuf = priv->nfinfo_realloc_mbuf; 308 i->nfinfo_alloc_fibs = priv->nfinfo_alloc_fibs; 309 i->nfinfo_inact_t = priv->nfinfo_inact_t; 310 i->nfinfo_act_t = priv->nfinfo_act_t; 311 } 312 313 /* 314 * Insert a record into defined slot. 315 * 316 * First we get for us a free flow entry, then fill in all 317 * possible fields in it. 318 * 319 * TODO: consider dropping hash mutex while filling in datagram, 320 * as this was done in previous version. Need to test & profile 321 * to be sure. 322 */ 323 static int 324 hash_insert(priv_p priv, struct flow_hash_entry *hsh, struct flow_rec *r, 325 int plen, uint8_t flags, uint8_t tcp_flags) 326 { 327 struct flow_entry *fle; 328 struct sockaddr_in sin, sin_mask; 329 struct sockaddr_dl rt_gateway; 330 struct rt_addrinfo info; 331 332 mtx_assert(&hsh->mtx, MA_OWNED); 333 334 fle = uma_zalloc_arg(priv->zone, priv, M_NOWAIT); 335 if (fle == NULL) { 336 priv->nfinfo_alloc_failed++; 337 return (ENOMEM); 338 } 339 340 /* 341 * Now fle is totally ours. It is detached from all lists, 342 * we can safely edit it. 343 */ 344 fle->f.version = IPVERSION; 345 bcopy(r, &fle->f.r, sizeof(struct flow_rec)); 346 fle->f.bytes = plen; 347 fle->f.packets = 1; 348 fle->f.tcp_flags = tcp_flags; 349 350 fle->f.first = fle->f.last = time_uptime; 351 352 /* 353 * First we do route table lookup on destination address. So we can 354 * fill in out_ifx, dst_mask, nexthop, and dst_as in future releases. 355 */ 356 if ((flags & NG_NETFLOW_CONF_NODSTLOOKUP) == 0) { 357 bzero(&sin, sizeof(sin)); 358 sin.sin_len = sizeof(struct sockaddr_in); 359 sin.sin_family = AF_INET; 360 sin.sin_addr = fle->f.r.r_dst; 361 362 rt_gateway.sdl_len = sizeof(rt_gateway); 363 sin_mask.sin_len = sizeof(struct sockaddr_in); 364 bzero(&info, sizeof(info)); 365 366 info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&rt_gateway; 367 info.rti_info[RTAX_NETMASK] = (struct sockaddr *)&sin_mask; 368 369 if (rib_lookup_info(r->fib, (struct sockaddr *)&sin, NHR_REF, 0, 370 &info) == 0) { 371 fle->f.fle_o_ifx = info.rti_ifp->if_index; 372 373 if (info.rti_flags & RTF_GATEWAY && 374 rt_gateway.sdl_family == AF_INET) 375 fle->f.next_hop = 376 ((struct sockaddr_in *)&rt_gateway)->sin_addr; 377 378 if (info.rti_addrs & RTA_NETMASK) 379 fle->f.dst_mask = bitcount32(sin_mask.sin_addr.s_addr); 380 else if (info.rti_flags & RTF_HOST) 381 /* Give up. We can't determine mask :( */ 382 fle->f.dst_mask = 32; 383 384 rib_free_info(&info); 385 } 386 } 387 388 /* Do route lookup on source address, to fill in src_mask. */ 389 if ((flags & NG_NETFLOW_CONF_NOSRCLOOKUP) == 0) { 390 bzero(&sin, sizeof(sin)); 391 sin.sin_len = sizeof(struct sockaddr_in); 392 sin.sin_family = AF_INET; 393 sin.sin_addr = fle->f.r.r_src; 394 395 sin_mask.sin_len = sizeof(struct sockaddr_in); 396 bzero(&info, sizeof(info)); 397 398 info.rti_info[RTAX_NETMASK] = (struct sockaddr *)&sin_mask; 399 400 if (rib_lookup_info(r->fib, (struct sockaddr *)&sin, 0, 0, 401 &info) == 0) { 402 if (info.rti_addrs & RTA_NETMASK) 403 fle->f.src_mask = 404 bitcount32(sin_mask.sin_addr.s_addr); 405 else if (info.rti_flags & RTF_HOST) 406 /* Give up. We can't determine mask :( */ 407 fle->f.src_mask = 32; 408 } 409 } 410 411 /* Push new flow at the and of hash. */ 412 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash); 413 414 return (0); 415 } 416 417 #ifdef INET6 418 static int 419 hash6_insert(priv_p priv, struct flow_hash_entry *hsh6, struct flow6_rec *r, 420 int plen, uint8_t flags, uint8_t tcp_flags) 421 { 422 struct flow6_entry *fle6; 423 struct sockaddr_in6 sin6, sin6_mask; 424 struct sockaddr_dl rt_gateway; 425 struct rt_addrinfo info; 426 427 mtx_assert(&hsh6->mtx, MA_OWNED); 428 429 fle6 = uma_zalloc_arg(priv->zone6, priv, M_NOWAIT); 430 if (fle6 == NULL) { 431 priv->nfinfo_alloc_failed++; 432 return (ENOMEM); 433 } 434 435 /* 436 * Now fle is totally ours. It is detached from all lists, 437 * we can safely edit it. 438 */ 439 440 fle6->f.version = IP6VERSION; 441 bcopy(r, &fle6->f.r, sizeof(struct flow6_rec)); 442 fle6->f.bytes = plen; 443 fle6->f.packets = 1; 444 fle6->f.tcp_flags = tcp_flags; 445 446 fle6->f.first = fle6->f.last = time_uptime; 447 448 /* 449 * First we do route table lookup on destination address. So we can 450 * fill in out_ifx, dst_mask, nexthop, and dst_as in future releases. 451 */ 452 if ((flags & NG_NETFLOW_CONF_NODSTLOOKUP) == 0) { 453 bzero(&sin6, sizeof(struct sockaddr_in6)); 454 sin6.sin6_len = sizeof(struct sockaddr_in6); 455 sin6.sin6_family = AF_INET6; 456 sin6.sin6_addr = r->dst.r_dst6; 457 458 rt_gateway.sdl_len = sizeof(rt_gateway); 459 sin6_mask.sin6_len = sizeof(struct sockaddr_in6); 460 bzero(&info, sizeof(info)); 461 462 info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&rt_gateway; 463 info.rti_info[RTAX_NETMASK] = (struct sockaddr *)&sin6_mask; 464 465 if (rib_lookup_info(r->fib, (struct sockaddr *)&sin6, NHR_REF, 466 0, &info) == 0) { 467 fle6->f.fle_o_ifx = info.rti_ifp->if_index; 468 469 if (info.rti_flags & RTF_GATEWAY && 470 rt_gateway.sdl_family == AF_INET6) 471 fle6->f.n.next_hop6 = 472 ((struct sockaddr_in6 *)&rt_gateway)->sin6_addr; 473 474 fle6->f.dst_mask = 475 ip6_masklen(&sin6_mask.sin6_addr, &info); 476 477 rib_free_info(&info); 478 } 479 } 480 481 if ((flags & NG_NETFLOW_CONF_NOSRCLOOKUP) == 0) { 482 /* Do route lookup on source address, to fill in src_mask. */ 483 bzero(&sin6, sizeof(struct sockaddr_in6)); 484 sin6.sin6_len = sizeof(struct sockaddr_in6); 485 sin6.sin6_family = AF_INET6; 486 sin6.sin6_addr = r->src.r_src6; 487 488 sin6_mask.sin6_len = sizeof(struct sockaddr_in6); 489 bzero(&info, sizeof(info)); 490 491 info.rti_info[RTAX_NETMASK] = (struct sockaddr *)&sin6_mask; 492 493 if (rib_lookup_info(r->fib, (struct sockaddr *)&sin6, 0, 0, 494 &info) == 0) 495 fle6->f.src_mask = 496 ip6_masklen(&sin6_mask.sin6_addr, &info); 497 } 498 499 /* Push new flow at the and of hash. */ 500 TAILQ_INSERT_TAIL(&hsh6->head, (struct flow_entry *)fle6, fle_hash); 501 502 return (0); 503 } 504 #endif 505 506 /* 507 * Non-static functions called from ng_netflow.c 508 */ 509 510 /* Allocate memory and set up flow cache */ 511 void 512 ng_netflow_cache_init(priv_p priv) 513 { 514 struct flow_hash_entry *hsh; 515 int i; 516 517 /* Initialize cache UMA zone. */ 518 priv->zone = uma_zcreate("NetFlow IPv4 cache", 519 sizeof(struct flow_entry), NULL, NULL, NULL, NULL, 520 UMA_ALIGN_CACHE, 0); 521 uma_zone_set_max(priv->zone, CACHESIZE); 522 #ifdef INET6 523 priv->zone6 = uma_zcreate("NetFlow IPv6 cache", 524 sizeof(struct flow6_entry), NULL, NULL, NULL, NULL, 525 UMA_ALIGN_CACHE, 0); 526 uma_zone_set_max(priv->zone6, CACHESIZE); 527 #endif 528 529 /* Allocate hash. */ 530 priv->hash = malloc(NBUCKETS * sizeof(struct flow_hash_entry), 531 M_NETFLOW_HASH, M_WAITOK | M_ZERO); 532 533 /* Initialize hash. */ 534 for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++) { 535 mtx_init(&hsh->mtx, "hash mutex", NULL, MTX_DEF); 536 TAILQ_INIT(&hsh->head); 537 } 538 539 #ifdef INET6 540 /* Allocate hash. */ 541 priv->hash6 = malloc(NBUCKETS * sizeof(struct flow_hash_entry), 542 M_NETFLOW_HASH, M_WAITOK | M_ZERO); 543 544 /* Initialize hash. */ 545 for (i = 0, hsh = priv->hash6; i < NBUCKETS; i++, hsh++) { 546 mtx_init(&hsh->mtx, "hash mutex", NULL, MTX_DEF); 547 TAILQ_INIT(&hsh->head); 548 } 549 #endif 550 551 priv->nfinfo_bytes = counter_u64_alloc(M_WAITOK); 552 priv->nfinfo_packets = counter_u64_alloc(M_WAITOK); 553 priv->nfinfo_bytes6 = counter_u64_alloc(M_WAITOK); 554 priv->nfinfo_packets6 = counter_u64_alloc(M_WAITOK); 555 priv->nfinfo_sbytes = counter_u64_alloc(M_WAITOK); 556 priv->nfinfo_spackets = counter_u64_alloc(M_WAITOK); 557 priv->nfinfo_sbytes6 = counter_u64_alloc(M_WAITOK); 558 priv->nfinfo_spackets6 = counter_u64_alloc(M_WAITOK); 559 priv->nfinfo_act_exp = counter_u64_alloc(M_WAITOK); 560 priv->nfinfo_inact_exp = counter_u64_alloc(M_WAITOK); 561 562 ng_netflow_v9_cache_init(priv); 563 CTR0(KTR_NET, "ng_netflow startup()"); 564 } 565 566 /* Initialize new FIB table for v5 and v9 */ 567 int 568 ng_netflow_fib_init(priv_p priv, int fib) 569 { 570 fib_export_p fe = priv_to_fib(priv, fib); 571 572 CTR1(KTR_NET, "ng_netflow(): fib init: %d", fib); 573 574 if (fe != NULL) 575 return (0); 576 577 if ((fe = malloc(sizeof(struct fib_export), M_NETGRAPH, 578 M_NOWAIT | M_ZERO)) == NULL) 579 return (ENOMEM); 580 581 mtx_init(&fe->export_mtx, "export dgram lock", NULL, MTX_DEF); 582 mtx_init(&fe->export9_mtx, "export9 dgram lock", NULL, MTX_DEF); 583 fe->fib = fib; 584 fe->domain_id = fib; 585 586 if (atomic_cmpset_ptr((volatile uintptr_t *)&priv->fib_data[fib], 587 (uintptr_t)NULL, (uintptr_t)fe) == 0) { 588 /* FIB already set up by other ISR */ 589 CTR3(KTR_NET, "ng_netflow(): fib init: %d setup %p but got %p", 590 fib, fe, priv_to_fib(priv, fib)); 591 mtx_destroy(&fe->export_mtx); 592 mtx_destroy(&fe->export9_mtx); 593 free(fe, M_NETGRAPH); 594 } else { 595 /* Increase counter for statistics */ 596 CTR3(KTR_NET, "ng_netflow(): fib %d setup to %p (%p)", 597 fib, fe, priv_to_fib(priv, fib)); 598 priv->nfinfo_alloc_fibs++; 599 } 600 601 return (0); 602 } 603 604 /* Free all flow cache memory. Called from node close method. */ 605 void 606 ng_netflow_cache_flush(priv_p priv) 607 { 608 struct flow_entry *fle, *fle1; 609 struct flow_hash_entry *hsh; 610 struct netflow_export_item exp; 611 fib_export_p fe; 612 int i; 613 614 bzero(&exp, sizeof(exp)); 615 616 /* 617 * We are going to free probably billable data. 618 * Expire everything before freeing it. 619 * No locking is required since callout is already drained. 620 */ 621 for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++) 622 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) { 623 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 624 fe = priv_to_fib(priv, fle->f.r.fib); 625 expire_flow(priv, fe, fle, NG_QUEUE); 626 } 627 #ifdef INET6 628 for (hsh = priv->hash6, i = 0; i < NBUCKETS; hsh++, i++) 629 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) { 630 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 631 fe = priv_to_fib(priv, fle->f.r.fib); 632 expire_flow(priv, fe, fle, NG_QUEUE); 633 } 634 #endif 635 636 uma_zdestroy(priv->zone); 637 /* Destroy hash mutexes. */ 638 for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++) 639 mtx_destroy(&hsh->mtx); 640 641 /* Free hash memory. */ 642 if (priv->hash != NULL) 643 free(priv->hash, M_NETFLOW_HASH); 644 #ifdef INET6 645 uma_zdestroy(priv->zone6); 646 /* Destroy hash mutexes. */ 647 for (i = 0, hsh = priv->hash6; i < NBUCKETS; i++, hsh++) 648 mtx_destroy(&hsh->mtx); 649 650 /* Free hash memory. */ 651 if (priv->hash6 != NULL) 652 free(priv->hash6, M_NETFLOW_HASH); 653 #endif 654 655 for (i = 0; i < priv->maxfibs; i++) { 656 if ((fe = priv_to_fib(priv, i)) == NULL) 657 continue; 658 659 if (fe->exp.item != NULL) 660 export_send(priv, fe, fe->exp.item, NG_QUEUE); 661 662 if (fe->exp.item9 != NULL) 663 export9_send(priv, fe, fe->exp.item9, 664 fe->exp.item9_opt, NG_QUEUE); 665 666 mtx_destroy(&fe->export_mtx); 667 mtx_destroy(&fe->export9_mtx); 668 free(fe, M_NETGRAPH); 669 } 670 671 counter_u64_free(priv->nfinfo_bytes); 672 counter_u64_free(priv->nfinfo_packets); 673 counter_u64_free(priv->nfinfo_bytes6); 674 counter_u64_free(priv->nfinfo_packets6); 675 counter_u64_free(priv->nfinfo_sbytes); 676 counter_u64_free(priv->nfinfo_spackets); 677 counter_u64_free(priv->nfinfo_sbytes6); 678 counter_u64_free(priv->nfinfo_spackets6); 679 counter_u64_free(priv->nfinfo_act_exp); 680 counter_u64_free(priv->nfinfo_inact_exp); 681 682 ng_netflow_v9_cache_flush(priv); 683 } 684 685 /* Insert packet from into flow cache. */ 686 int 687 ng_netflow_flow_add(priv_p priv, fib_export_p fe, struct ip *ip, 688 caddr_t upper_ptr, uint8_t upper_proto, uint8_t flags, 689 unsigned int src_if_index) 690 { 691 struct flow_entry *fle, *fle1; 692 struct flow_hash_entry *hsh; 693 struct flow_rec r; 694 int hlen, plen; 695 int error = 0; 696 uint16_t eproto; 697 uint8_t tcp_flags = 0; 698 699 bzero(&r, sizeof(r)); 700 701 if (ip->ip_v != IPVERSION) 702 return (EINVAL); 703 704 hlen = ip->ip_hl << 2; 705 if (hlen < sizeof(struct ip)) 706 return (EINVAL); 707 708 eproto = ETHERTYPE_IP; 709 /* Assume L4 template by default */ 710 r.flow_type = NETFLOW_V9_FLOW_V4_L4; 711 712 r.r_src = ip->ip_src; 713 r.r_dst = ip->ip_dst; 714 r.fib = fe->fib; 715 716 plen = ntohs(ip->ip_len); 717 718 r.r_ip_p = ip->ip_p; 719 r.r_tos = ip->ip_tos; 720 721 r.r_i_ifx = src_if_index; 722 723 /* 724 * XXX NOTE: only first fragment of fragmented TCP, UDP and 725 * ICMP packet will be recorded with proper s_port and d_port. 726 * Following fragments will be recorded simply as IP packet with 727 * ip_proto = ip->ip_p and s_port, d_port set to zero. 728 * I know, it looks like bug. But I don't want to re-implement 729 * ip packet assebmling here. Anyway, (in)famous trafd works this way - 730 * and nobody complains yet :) 731 */ 732 if ((ip->ip_off & htons(IP_OFFMASK)) == 0) 733 switch(r.r_ip_p) { 734 case IPPROTO_TCP: 735 { 736 struct tcphdr *tcp; 737 738 tcp = (struct tcphdr *)((caddr_t )ip + hlen); 739 r.r_sport = tcp->th_sport; 740 r.r_dport = tcp->th_dport; 741 tcp_flags = tcp->th_flags; 742 break; 743 } 744 case IPPROTO_UDP: 745 r.r_ports = *(uint32_t *)((caddr_t )ip + hlen); 746 break; 747 } 748 749 counter_u64_add(priv->nfinfo_packets, 1); 750 counter_u64_add(priv->nfinfo_bytes, plen); 751 752 /* Find hash slot. */ 753 hsh = &priv->hash[ip_hash(&r)]; 754 755 mtx_lock(&hsh->mtx); 756 757 /* 758 * Go through hash and find our entry. If we encounter an 759 * entry, that should be expired, purge it. We do a reverse 760 * search since most active entries are first, and most 761 * searches are done on most active entries. 762 */ 763 TAILQ_FOREACH_REVERSE_SAFE(fle, &hsh->head, fhead, fle_hash, fle1) { 764 if (bcmp(&r, &fle->f.r, sizeof(struct flow_rec)) == 0) 765 break; 766 if ((INACTIVE(fle) && SMALL(fle)) || AGED(fle)) { 767 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 768 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), 769 fle, NG_QUEUE); 770 counter_u64_add(priv->nfinfo_act_exp, 1); 771 } 772 } 773 774 if (fle) { /* An existent entry. */ 775 776 fle->f.bytes += plen; 777 fle->f.packets ++; 778 fle->f.tcp_flags |= tcp_flags; 779 fle->f.last = time_uptime; 780 781 /* 782 * We have the following reasons to expire flow in active way: 783 * - it hit active timeout 784 * - a TCP connection closed 785 * - it is going to overflow counter 786 */ 787 if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle) || 788 (fle->f.bytes >= (CNTR_MAX - IF_MAXMTU)) ) { 789 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 790 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), 791 fle, NG_QUEUE); 792 counter_u64_add(priv->nfinfo_act_exp, 1); 793 } else { 794 /* 795 * It is the newest, move it to the tail, 796 * if it isn't there already. Next search will 797 * locate it quicker. 798 */ 799 if (fle != TAILQ_LAST(&hsh->head, fhead)) { 800 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 801 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash); 802 } 803 } 804 } else /* A new flow entry. */ 805 error = hash_insert(priv, hsh, &r, plen, flags, tcp_flags); 806 807 mtx_unlock(&hsh->mtx); 808 809 return (error); 810 } 811 812 #ifdef INET6 813 /* Insert IPv6 packet from into flow cache. */ 814 int 815 ng_netflow_flow6_add(priv_p priv, fib_export_p fe, struct ip6_hdr *ip6, 816 caddr_t upper_ptr, uint8_t upper_proto, uint8_t flags, 817 unsigned int src_if_index) 818 { 819 struct flow_entry *fle = NULL, *fle1; 820 struct flow6_entry *fle6; 821 struct flow_hash_entry *hsh; 822 struct flow6_rec r; 823 int plen; 824 int error = 0; 825 uint8_t tcp_flags = 0; 826 827 /* check version */ 828 if ((ip6->ip6_vfc & IPV6_VERSION_MASK) != IPV6_VERSION) 829 return (EINVAL); 830 831 bzero(&r, sizeof(r)); 832 833 r.src.r_src6 = ip6->ip6_src; 834 r.dst.r_dst6 = ip6->ip6_dst; 835 r.fib = fe->fib; 836 837 /* Assume L4 template by default */ 838 r.flow_type = NETFLOW_V9_FLOW_V6_L4; 839 840 plen = ntohs(ip6->ip6_plen) + sizeof(struct ip6_hdr); 841 842 #if 0 843 /* XXX: set DSCP/CoS value */ 844 r.r_tos = ip->ip_tos; 845 #endif 846 if ((flags & NG_NETFLOW_IS_FRAG) == 0) { 847 switch(upper_proto) { 848 case IPPROTO_TCP: 849 { 850 struct tcphdr *tcp; 851 852 tcp = (struct tcphdr *)upper_ptr; 853 r.r_ports = *(uint32_t *)upper_ptr; 854 tcp_flags = tcp->th_flags; 855 break; 856 } 857 case IPPROTO_UDP: 858 case IPPROTO_SCTP: 859 r.r_ports = *(uint32_t *)upper_ptr; 860 break; 861 } 862 } 863 864 r.r_ip_p = upper_proto; 865 r.r_i_ifx = src_if_index; 866 867 counter_u64_add(priv->nfinfo_packets6, 1); 868 counter_u64_add(priv->nfinfo_bytes6, plen); 869 870 /* Find hash slot. */ 871 hsh = &priv->hash6[ip6_hash(&r)]; 872 873 mtx_lock(&hsh->mtx); 874 875 /* 876 * Go through hash and find our entry. If we encounter an 877 * entry, that should be expired, purge it. We do a reverse 878 * search since most active entries are first, and most 879 * searches are done on most active entries. 880 */ 881 TAILQ_FOREACH_REVERSE_SAFE(fle, &hsh->head, fhead, fle_hash, fle1) { 882 if (fle->f.version != IP6VERSION) 883 continue; 884 fle6 = (struct flow6_entry *)fle; 885 if (bcmp(&r, &fle6->f.r, sizeof(struct flow6_rec)) == 0) 886 break; 887 if ((INACTIVE(fle6) && SMALL(fle6)) || AGED(fle6)) { 888 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 889 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), fle, 890 NG_QUEUE); 891 counter_u64_add(priv->nfinfo_act_exp, 1); 892 } 893 } 894 895 if (fle != NULL) { /* An existent entry. */ 896 fle6 = (struct flow6_entry *)fle; 897 898 fle6->f.bytes += plen; 899 fle6->f.packets ++; 900 fle6->f.tcp_flags |= tcp_flags; 901 fle6->f.last = time_uptime; 902 903 /* 904 * We have the following reasons to expire flow in active way: 905 * - it hit active timeout 906 * - a TCP connection closed 907 * - it is going to overflow counter 908 */ 909 if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle6) || 910 (fle6->f.bytes >= (CNTR_MAX - IF_MAXMTU)) ) { 911 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 912 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), fle, 913 NG_QUEUE); 914 counter_u64_add(priv->nfinfo_act_exp, 1); 915 } else { 916 /* 917 * It is the newest, move it to the tail, 918 * if it isn't there already. Next search will 919 * locate it quicker. 920 */ 921 if (fle != TAILQ_LAST(&hsh->head, fhead)) { 922 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 923 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash); 924 } 925 } 926 } else /* A new flow entry. */ 927 error = hash6_insert(priv, hsh, &r, plen, flags, tcp_flags); 928 929 mtx_unlock(&hsh->mtx); 930 931 return (error); 932 } 933 #endif 934 935 /* 936 * Return records from cache to userland. 937 * 938 * TODO: matching particular IP should be done in kernel, here. 939 */ 940 int 941 ng_netflow_flow_show(priv_p priv, struct ngnf_show_header *req, 942 struct ngnf_show_header *resp) 943 { 944 struct flow_hash_entry *hsh; 945 struct flow_entry *fle; 946 struct flow_entry_data *data = (struct flow_entry_data *)(resp + 1); 947 #ifdef INET6 948 struct flow6_entry_data *data6 = (struct flow6_entry_data *)(resp + 1); 949 #endif 950 int i, max; 951 952 i = req->hash_id; 953 if (i > NBUCKETS-1) 954 return (EINVAL); 955 956 #ifdef INET6 957 if (req->version == 6) { 958 resp->version = 6; 959 hsh = priv->hash6 + i; 960 max = NREC6_AT_ONCE; 961 } else 962 #endif 963 if (req->version == 4) { 964 resp->version = 4; 965 hsh = priv->hash + i; 966 max = NREC_AT_ONCE; 967 } else 968 return (EINVAL); 969 970 /* 971 * We will transfer not more than NREC_AT_ONCE. More data 972 * will come in next message. 973 * We send current hash index and current record number in list 974 * to userland, and userland should return it back to us. 975 * Then, we will restart with new entry. 976 * 977 * The resulting cache snapshot can be inaccurate if flow expiration 978 * is taking place on hash item between userland data requests for 979 * this hash item id. 980 */ 981 resp->nentries = 0; 982 for (; i < NBUCKETS; hsh++, i++) { 983 int list_id; 984 985 if (mtx_trylock(&hsh->mtx) == 0) { 986 /* 987 * Requested hash index is not available, 988 * relay decision to skip or re-request data 989 * to userland. 990 */ 991 resp->hash_id = i; 992 resp->list_id = 0; 993 return (0); 994 } 995 996 list_id = 0; 997 TAILQ_FOREACH(fle, &hsh->head, fle_hash) { 998 if (hsh->mtx.mtx_lock & MTX_CONTESTED) { 999 resp->hash_id = i; 1000 resp->list_id = list_id; 1001 mtx_unlock(&hsh->mtx); 1002 return (0); 1003 } 1004 1005 list_id++; 1006 /* Search for particular record in list. */ 1007 if (req->list_id > 0) { 1008 if (list_id < req->list_id) 1009 continue; 1010 1011 /* Requested list position found. */ 1012 req->list_id = 0; 1013 } 1014 #ifdef INET6 1015 if (req->version == 6) { 1016 struct flow6_entry *fle6; 1017 1018 fle6 = (struct flow6_entry *)fle; 1019 bcopy(&fle6->f, data6 + resp->nentries, 1020 sizeof(fle6->f)); 1021 } else 1022 #endif 1023 bcopy(&fle->f, data + resp->nentries, 1024 sizeof(fle->f)); 1025 resp->nentries++; 1026 if (resp->nentries == max) { 1027 resp->hash_id = i; 1028 /* 1029 * If it was the last item in list 1030 * we simply skip to next hash_id. 1031 */ 1032 resp->list_id = list_id + 1; 1033 mtx_unlock(&hsh->mtx); 1034 return (0); 1035 } 1036 } 1037 mtx_unlock(&hsh->mtx); 1038 } 1039 1040 resp->hash_id = resp->list_id = 0; 1041 1042 return (0); 1043 } 1044 1045 /* We have full datagram in privdata. Send it to export hook. */ 1046 static int 1047 export_send(priv_p priv, fib_export_p fe, item_p item, int flags) 1048 { 1049 struct mbuf *m = NGI_M(item); 1050 struct netflow_v5_export_dgram *dgram = mtod(m, 1051 struct netflow_v5_export_dgram *); 1052 struct netflow_v5_header *header = &dgram->header; 1053 struct timespec ts; 1054 int error = 0; 1055 1056 /* Fill mbuf header. */ 1057 m->m_len = m->m_pkthdr.len = sizeof(struct netflow_v5_record) * 1058 header->count + sizeof(struct netflow_v5_header); 1059 1060 /* Fill export header. */ 1061 header->sys_uptime = htonl(MILLIUPTIME(time_uptime)); 1062 getnanotime(&ts); 1063 header->unix_secs = htonl(ts.tv_sec); 1064 header->unix_nsecs = htonl(ts.tv_nsec); 1065 header->engine_type = 0; 1066 header->engine_id = fe->domain_id; 1067 header->pad = 0; 1068 header->flow_seq = htonl(atomic_fetchadd_32(&fe->flow_seq, 1069 header->count)); 1070 header->count = htons(header->count); 1071 1072 if (priv->export != NULL) 1073 NG_FWD_ITEM_HOOK_FLAGS(error, item, priv->export, flags); 1074 else 1075 NG_FREE_ITEM(item); 1076 1077 return (error); 1078 } 1079 1080 /* Add export record to dgram. */ 1081 static int 1082 export_add(item_p item, struct flow_entry *fle) 1083 { 1084 struct netflow_v5_export_dgram *dgram = mtod(NGI_M(item), 1085 struct netflow_v5_export_dgram *); 1086 struct netflow_v5_header *header = &dgram->header; 1087 struct netflow_v5_record *rec; 1088 1089 rec = &dgram->r[header->count]; 1090 header->count ++; 1091 1092 KASSERT(header->count <= NETFLOW_V5_MAX_RECORDS, 1093 ("ng_netflow: export too big")); 1094 1095 /* Fill in export record. */ 1096 rec->src_addr = fle->f.r.r_src.s_addr; 1097 rec->dst_addr = fle->f.r.r_dst.s_addr; 1098 rec->next_hop = fle->f.next_hop.s_addr; 1099 rec->i_ifx = htons(fle->f.fle_i_ifx); 1100 rec->o_ifx = htons(fle->f.fle_o_ifx); 1101 rec->packets = htonl(fle->f.packets); 1102 rec->octets = htonl(fle->f.bytes); 1103 rec->first = htonl(MILLIUPTIME(fle->f.first)); 1104 rec->last = htonl(MILLIUPTIME(fle->f.last)); 1105 rec->s_port = fle->f.r.r_sport; 1106 rec->d_port = fle->f.r.r_dport; 1107 rec->flags = fle->f.tcp_flags; 1108 rec->prot = fle->f.r.r_ip_p; 1109 rec->tos = fle->f.r.r_tos; 1110 rec->dst_mask = fle->f.dst_mask; 1111 rec->src_mask = fle->f.src_mask; 1112 rec->pad1 = 0; 1113 rec->pad2 = 0; 1114 1115 /* Not supported fields. */ 1116 rec->src_as = rec->dst_as = 0; 1117 1118 if (header->count == NETFLOW_V5_MAX_RECORDS) 1119 return (1); /* end of datagram */ 1120 else 1121 return (0); 1122 } 1123 1124 /* Periodic flow expiry run. */ 1125 void 1126 ng_netflow_expire(void *arg) 1127 { 1128 struct flow_entry *fle, *fle1; 1129 struct flow_hash_entry *hsh; 1130 priv_p priv = (priv_p )arg; 1131 int used, i; 1132 1133 /* 1134 * Going through all the cache. 1135 */ 1136 used = uma_zone_get_cur(priv->zone); 1137 for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++) { 1138 /* 1139 * Skip entries, that are already being worked on. 1140 */ 1141 if (mtx_trylock(&hsh->mtx) == 0) 1142 continue; 1143 1144 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) { 1145 /* 1146 * Interrupt thread wants this entry! 1147 * Quick! Quick! Bail out! 1148 */ 1149 if (hsh->mtx.mtx_lock & MTX_CONTESTED) 1150 break; 1151 1152 /* 1153 * Don't expire aggressively while hash collision 1154 * ratio is predicted small. 1155 */ 1156 if (used <= (NBUCKETS*2) && !INACTIVE(fle)) 1157 break; 1158 1159 if ((INACTIVE(fle) && (SMALL(fle) || 1160 (used > (NBUCKETS*2)))) || AGED(fle)) { 1161 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 1162 expire_flow(priv, priv_to_fib(priv, 1163 fle->f.r.fib), fle, NG_NOFLAGS); 1164 used--; 1165 counter_u64_add(priv->nfinfo_inact_exp, 1); 1166 } 1167 } 1168 mtx_unlock(&hsh->mtx); 1169 } 1170 1171 #ifdef INET6 1172 used = uma_zone_get_cur(priv->zone6); 1173 for (hsh = priv->hash6, i = 0; i < NBUCKETS; hsh++, i++) { 1174 struct flow6_entry *fle6; 1175 1176 /* 1177 * Skip entries, that are already being worked on. 1178 */ 1179 if (mtx_trylock(&hsh->mtx) == 0) 1180 continue; 1181 1182 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) { 1183 fle6 = (struct flow6_entry *)fle; 1184 /* 1185 * Interrupt thread wants this entry! 1186 * Quick! Quick! Bail out! 1187 */ 1188 if (hsh->mtx.mtx_lock & MTX_CONTESTED) 1189 break; 1190 1191 /* 1192 * Don't expire aggressively while hash collision 1193 * ratio is predicted small. 1194 */ 1195 if (used <= (NBUCKETS*2) && !INACTIVE(fle6)) 1196 break; 1197 1198 if ((INACTIVE(fle6) && (SMALL(fle6) || 1199 (used > (NBUCKETS*2)))) || AGED(fle6)) { 1200 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 1201 expire_flow(priv, priv_to_fib(priv, 1202 fle->f.r.fib), fle, NG_NOFLAGS); 1203 used--; 1204 counter_u64_add(priv->nfinfo_inact_exp, 1); 1205 } 1206 } 1207 mtx_unlock(&hsh->mtx); 1208 } 1209 #endif 1210 1211 /* Schedule next expire. */ 1212 callout_reset(&priv->exp_callout, (1*hz), &ng_netflow_expire, 1213 (void *)priv); 1214 } 1215