1 /*- 2 * Copyright (c) 2010-2011 Alexander V. Chernikov <melifaro@ipfw.ru> 3 * Copyright (c) 2004-2005 Gleb Smirnoff <glebius@FreeBSD.org> 4 * Copyright (c) 2001-2003 Roman V. Palagin <romanp@unshadow.net> 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 * 28 * $SourceForge: netflow.c,v 1.41 2004/09/05 11:41:10 glebius Exp $ 29 */ 30 31 #include <sys/cdefs.h> 32 __FBSDID("$FreeBSD$"); 33 34 #include "opt_inet6.h" 35 #include "opt_route.h" 36 #include <sys/param.h> 37 #include <sys/systm.h> 38 #include <sys/counter.h> 39 #include <sys/kernel.h> 40 #include <sys/ktr.h> 41 #include <sys/limits.h> 42 #include <sys/mbuf.h> 43 #include <sys/syslog.h> 44 #include <sys/socket.h> 45 46 #include <net/if.h> 47 #include <net/if_dl.h> 48 #include <net/if_var.h> 49 #include <net/route.h> 50 #include <net/ethernet.h> 51 #include <netinet/in.h> 52 #include <netinet/in_systm.h> 53 #include <netinet/ip.h> 54 #include <netinet/ip6.h> 55 #include <netinet/tcp.h> 56 #include <netinet/udp.h> 57 58 #include <netgraph/ng_message.h> 59 #include <netgraph/netgraph.h> 60 61 #include <netgraph/netflow/netflow.h> 62 #include <netgraph/netflow/netflow_v9.h> 63 #include <netgraph/netflow/ng_netflow.h> 64 65 #define NBUCKETS (65536) /* must be power of 2 */ 66 67 /* This hash is for TCP or UDP packets. */ 68 #define FULL_HASH(addr1, addr2, port1, port2) \ 69 (((addr1 ^ (addr1 >> 16) ^ \ 70 htons(addr2 ^ (addr2 >> 16))) ^ \ 71 port1 ^ htons(port2)) & \ 72 (NBUCKETS - 1)) 73 74 /* This hash is for all other IP packets. */ 75 #define ADDR_HASH(addr1, addr2) \ 76 ((addr1 ^ (addr1 >> 16) ^ \ 77 htons(addr2 ^ (addr2 >> 16))) & \ 78 (NBUCKETS - 1)) 79 80 /* Macros to shorten logical constructions */ 81 /* XXX: priv must exist in namespace */ 82 #define INACTIVE(fle) (time_uptime - fle->f.last > priv->nfinfo_inact_t) 83 #define AGED(fle) (time_uptime - fle->f.first > priv->nfinfo_act_t) 84 #define ISFREE(fle) (fle->f.packets == 0) 85 86 /* 87 * 4 is a magical number: statistically number of 4-packet flows is 88 * bigger than 5,6,7...-packet flows by an order of magnitude. Most UDP/ICMP 89 * scans are 1 packet (~ 90% of flow cache). TCP scans are 2-packet in case 90 * of reachable host and 4-packet otherwise. 91 */ 92 #define SMALL(fle) (fle->f.packets <= 4) 93 94 MALLOC_DEFINE(M_NETFLOW_HASH, "netflow_hash", "NetFlow hash"); 95 96 static int export_add(item_p, struct flow_entry *); 97 static int export_send(priv_p, fib_export_p, item_p, int); 98 99 static int hash_insert(priv_p, struct flow_hash_entry *, struct flow_rec *, 100 int, uint8_t, uint8_t); 101 #ifdef INET6 102 static int hash6_insert(priv_p, struct flow_hash_entry *, struct flow6_rec *, 103 int, uint8_t, uint8_t); 104 #endif 105 106 static void expire_flow(priv_p, fib_export_p, struct flow_entry *, int); 107 108 /* 109 * Generate hash for a given flow record. 110 * 111 * FIB is not used here, because: 112 * most VRFS will carry public IPv4 addresses which are unique even 113 * without FIB private addresses can overlap, but this is worked out 114 * via flow_rec bcmp() containing fib id. In IPv6 world addresses are 115 * all globally unique (it's not fully true, there is FC00::/7 for example, 116 * but chances of address overlap are MUCH smaller) 117 */ 118 static inline uint32_t 119 ip_hash(struct flow_rec *r) 120 { 121 122 switch (r->r_ip_p) { 123 case IPPROTO_TCP: 124 case IPPROTO_UDP: 125 return FULL_HASH(r->r_src.s_addr, r->r_dst.s_addr, 126 r->r_sport, r->r_dport); 127 default: 128 return ADDR_HASH(r->r_src.s_addr, r->r_dst.s_addr); 129 } 130 } 131 132 #ifdef INET6 133 /* Generate hash for a given flow6 record. Use lower 4 octets from v6 addresses */ 134 static inline uint32_t 135 ip6_hash(struct flow6_rec *r) 136 { 137 138 switch (r->r_ip_p) { 139 case IPPROTO_TCP: 140 case IPPROTO_UDP: 141 return FULL_HASH(r->src.r_src6.__u6_addr.__u6_addr32[3], 142 r->dst.r_dst6.__u6_addr.__u6_addr32[3], r->r_sport, 143 r->r_dport); 144 default: 145 return ADDR_HASH(r->src.r_src6.__u6_addr.__u6_addr32[3], 146 r->dst.r_dst6.__u6_addr.__u6_addr32[3]); 147 } 148 } 149 #endif 150 151 /* 152 * Detach export datagram from priv, if there is any. 153 * If there is no, allocate a new one. 154 */ 155 static item_p 156 get_export_dgram(priv_p priv, fib_export_p fe) 157 { 158 item_p item = NULL; 159 160 mtx_lock(&fe->export_mtx); 161 if (fe->exp.item != NULL) { 162 item = fe->exp.item; 163 fe->exp.item = NULL; 164 } 165 mtx_unlock(&fe->export_mtx); 166 167 if (item == NULL) { 168 struct netflow_v5_export_dgram *dgram; 169 struct mbuf *m; 170 171 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 172 if (m == NULL) 173 return (NULL); 174 item = ng_package_data(m, NG_NOFLAGS); 175 if (item == NULL) 176 return (NULL); 177 dgram = mtod(m, struct netflow_v5_export_dgram *); 178 dgram->header.count = 0; 179 dgram->header.version = htons(NETFLOW_V5); 180 dgram->header.pad = 0; 181 } 182 183 return (item); 184 } 185 186 /* 187 * Re-attach incomplete datagram back to priv. 188 * If there is already another one, then send incomplete. */ 189 static void 190 return_export_dgram(priv_p priv, fib_export_p fe, item_p item, int flags) 191 { 192 193 /* 194 * It may happen on SMP, that some thread has already 195 * put its item there, in this case we bail out and 196 * send what we have to collector. 197 */ 198 mtx_lock(&fe->export_mtx); 199 if (fe->exp.item == NULL) { 200 fe->exp.item = item; 201 mtx_unlock(&fe->export_mtx); 202 } else { 203 mtx_unlock(&fe->export_mtx); 204 export_send(priv, fe, item, flags); 205 } 206 } 207 208 /* 209 * The flow is over. Call export_add() and free it. If datagram is 210 * full, then call export_send(). 211 */ 212 static void 213 expire_flow(priv_p priv, fib_export_p fe, struct flow_entry *fle, int flags) 214 { 215 struct netflow_export_item exp; 216 uint16_t version = fle->f.version; 217 218 if ((priv->export != NULL) && (version == IPVERSION)) { 219 exp.item = get_export_dgram(priv, fe); 220 if (exp.item == NULL) { 221 priv->nfinfo_export_failed++; 222 if (priv->export9 != NULL) 223 priv->nfinfo_export9_failed++; 224 /* fle definitely contains IPv4 flow. */ 225 uma_zfree_arg(priv->zone, fle, priv); 226 return; 227 } 228 229 if (export_add(exp.item, fle) > 0) 230 export_send(priv, fe, exp.item, flags); 231 else 232 return_export_dgram(priv, fe, exp.item, NG_QUEUE); 233 } 234 235 if (priv->export9 != NULL) { 236 exp.item9 = get_export9_dgram(priv, fe, &exp.item9_opt); 237 if (exp.item9 == NULL) { 238 priv->nfinfo_export9_failed++; 239 if (version == IPVERSION) 240 uma_zfree_arg(priv->zone, fle, priv); 241 #ifdef INET6 242 else if (version == IP6VERSION) 243 uma_zfree_arg(priv->zone6, fle, priv); 244 #endif 245 else 246 panic("ng_netflow: Unknown IP proto: %d", 247 version); 248 return; 249 } 250 251 if (export9_add(exp.item9, exp.item9_opt, fle) > 0) 252 export9_send(priv, fe, exp.item9, exp.item9_opt, flags); 253 else 254 return_export9_dgram(priv, fe, exp.item9, 255 exp.item9_opt, NG_QUEUE); 256 } 257 258 if (version == IPVERSION) 259 uma_zfree_arg(priv->zone, fle, priv); 260 #ifdef INET6 261 else if (version == IP6VERSION) 262 uma_zfree_arg(priv->zone6, fle, priv); 263 #endif 264 } 265 266 /* Get a snapshot of node statistics */ 267 void 268 ng_netflow_copyinfo(priv_p priv, struct ng_netflow_info *i) 269 { 270 271 i->nfinfo_bytes = counter_u64_fetch(priv->nfinfo_bytes); 272 i->nfinfo_packets = counter_u64_fetch(priv->nfinfo_packets); 273 i->nfinfo_bytes6 = counter_u64_fetch(priv->nfinfo_bytes6); 274 i->nfinfo_packets6 = counter_u64_fetch(priv->nfinfo_packets6); 275 i->nfinfo_sbytes = counter_u64_fetch(priv->nfinfo_sbytes); 276 i->nfinfo_spackets = counter_u64_fetch(priv->nfinfo_spackets); 277 i->nfinfo_sbytes6 = counter_u64_fetch(priv->nfinfo_sbytes6); 278 i->nfinfo_spackets6 = counter_u64_fetch(priv->nfinfo_spackets6); 279 i->nfinfo_act_exp = counter_u64_fetch(priv->nfinfo_act_exp); 280 i->nfinfo_inact_exp = counter_u64_fetch(priv->nfinfo_inact_exp); 281 282 i->nfinfo_used = uma_zone_get_cur(priv->zone); 283 #ifdef INET6 284 i->nfinfo_used6 = uma_zone_get_cur(priv->zone6); 285 #endif 286 287 i->nfinfo_alloc_failed = priv->nfinfo_alloc_failed; 288 i->nfinfo_export_failed = priv->nfinfo_export_failed; 289 i->nfinfo_export9_failed = priv->nfinfo_export9_failed; 290 i->nfinfo_realloc_mbuf = priv->nfinfo_realloc_mbuf; 291 i->nfinfo_alloc_fibs = priv->nfinfo_alloc_fibs; 292 i->nfinfo_inact_t = priv->nfinfo_inact_t; 293 i->nfinfo_act_t = priv->nfinfo_act_t; 294 } 295 296 /* 297 * Insert a record into defined slot. 298 * 299 * First we get for us a free flow entry, then fill in all 300 * possible fields in it. 301 * 302 * TODO: consider dropping hash mutex while filling in datagram, 303 * as this was done in previous version. Need to test & profile 304 * to be sure. 305 */ 306 static int 307 hash_insert(priv_p priv, struct flow_hash_entry *hsh, struct flow_rec *r, 308 int plen, uint8_t flags, uint8_t tcp_flags) 309 { 310 struct flow_entry *fle; 311 struct sockaddr_in sin, sin_mask; 312 struct sockaddr_dl rt_gateway; 313 struct rt_addrinfo info; 314 315 mtx_assert(&hsh->mtx, MA_OWNED); 316 317 fle = uma_zalloc_arg(priv->zone, priv, M_NOWAIT); 318 if (fle == NULL) { 319 priv->nfinfo_alloc_failed++; 320 return (ENOMEM); 321 } 322 323 /* 324 * Now fle is totally ours. It is detached from all lists, 325 * we can safely edit it. 326 */ 327 fle->f.version = IPVERSION; 328 bcopy(r, &fle->f.r, sizeof(struct flow_rec)); 329 fle->f.bytes = plen; 330 fle->f.packets = 1; 331 fle->f.tcp_flags = tcp_flags; 332 333 fle->f.first = fle->f.last = time_uptime; 334 335 /* 336 * First we do route table lookup on destination address. So we can 337 * fill in out_ifx, dst_mask, nexthop, and dst_as in future releases. 338 */ 339 if ((flags & NG_NETFLOW_CONF_NODSTLOOKUP) == 0) { 340 bzero(&sin, sizeof(sin)); 341 sin.sin_len = sizeof(struct sockaddr_in); 342 sin.sin_family = AF_INET; 343 sin.sin_addr = fle->f.r.r_dst; 344 345 rt_gateway.sdl_len = sizeof(rt_gateway); 346 sin_mask.sin_len = sizeof(struct sockaddr_in); 347 bzero(&info, sizeof(info)); 348 349 info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&rt_gateway; 350 info.rti_info[RTAX_NETMASK] = (struct sockaddr *)&sin_mask; 351 352 if (rib_lookup_info(r->fib, (struct sockaddr *)&sin, NHR_REF, 0, 353 &info) == 0) { 354 fle->f.fle_o_ifx = info.rti_ifp->if_index; 355 356 if (info.rti_flags & RTF_GATEWAY && 357 rt_gateway.sdl_family == AF_INET) 358 fle->f.next_hop = 359 ((struct sockaddr_in *)&rt_gateway)->sin_addr; 360 361 if (info.rti_addrs & RTA_NETMASK) 362 fle->f.dst_mask = bitcount32(sin_mask.sin_addr.s_addr); 363 else if (info.rti_flags & RTF_HOST) 364 /* Give up. We can't determine mask :( */ 365 fle->f.dst_mask = 32; 366 367 rib_free_info(&info); 368 } 369 } 370 371 /* Do route lookup on source address, to fill in src_mask. */ 372 if ((flags & NG_NETFLOW_CONF_NOSRCLOOKUP) == 0) { 373 bzero(&sin, sizeof(sin)); 374 sin.sin_len = sizeof(struct sockaddr_in); 375 sin.sin_family = AF_INET; 376 sin.sin_addr = fle->f.r.r_src; 377 378 sin_mask.sin_len = sizeof(struct sockaddr_in); 379 bzero(&info, sizeof(info)); 380 381 info.rti_info[RTAX_NETMASK] = (struct sockaddr *)&sin_mask; 382 383 if (rib_lookup_info(r->fib, (struct sockaddr *)&sin, 0, 0, 384 &info) == 0) { 385 if (info.rti_addrs & RTA_NETMASK) 386 fle->f.src_mask = 387 bitcount32(sin_mask.sin_addr.s_addr); 388 else if (info.rti_flags & RTF_HOST) 389 /* Give up. We can't determine mask :( */ 390 fle->f.src_mask = 32; 391 } 392 } 393 394 /* Push new flow at the and of hash. */ 395 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash); 396 397 return (0); 398 } 399 400 #ifdef INET6 401 /* XXX: make normal function, instead of.. */ 402 #define ipv6_masklen(x) bitcount32((x).__u6_addr.__u6_addr32[0]) + \ 403 bitcount32((x).__u6_addr.__u6_addr32[1]) + \ 404 bitcount32((x).__u6_addr.__u6_addr32[2]) + \ 405 bitcount32((x).__u6_addr.__u6_addr32[3]) 406 static int 407 hash6_insert(priv_p priv, struct flow_hash_entry *hsh6, struct flow6_rec *r, 408 int plen, uint8_t flags, uint8_t tcp_flags) 409 { 410 struct flow6_entry *fle6; 411 struct sockaddr_in6 sin6, sin6_mask; 412 struct sockaddr_dl rt_gateway; 413 struct rt_addrinfo info; 414 415 mtx_assert(&hsh6->mtx, MA_OWNED); 416 417 fle6 = uma_zalloc_arg(priv->zone6, priv, M_NOWAIT); 418 if (fle6 == NULL) { 419 priv->nfinfo_alloc_failed++; 420 return (ENOMEM); 421 } 422 423 /* 424 * Now fle is totally ours. It is detached from all lists, 425 * we can safely edit it. 426 */ 427 428 fle6->f.version = IP6VERSION; 429 bcopy(r, &fle6->f.r, sizeof(struct flow6_rec)); 430 fle6->f.bytes = plen; 431 fle6->f.packets = 1; 432 fle6->f.tcp_flags = tcp_flags; 433 434 fle6->f.first = fle6->f.last = time_uptime; 435 436 /* 437 * First we do route table lookup on destination address. So we can 438 * fill in out_ifx, dst_mask, nexthop, and dst_as in future releases. 439 */ 440 if ((flags & NG_NETFLOW_CONF_NODSTLOOKUP) == 0) { 441 bzero(&sin6, sizeof(struct sockaddr_in6)); 442 sin6.sin6_len = sizeof(struct sockaddr_in6); 443 sin6.sin6_family = AF_INET6; 444 sin6.sin6_addr = r->dst.r_dst6; 445 446 rt_gateway.sdl_len = sizeof(rt_gateway); 447 sin6_mask.sin6_len = sizeof(struct sockaddr_in6); 448 bzero(&info, sizeof(info)); 449 450 info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&rt_gateway; 451 info.rti_info[RTAX_NETMASK] = (struct sockaddr *)&sin6_mask; 452 453 if (rib_lookup_info(r->fib, (struct sockaddr *)&sin6, NHR_REF, 454 0, &info) == 0) { 455 fle6->f.fle_o_ifx = info.rti_ifp->if_index; 456 457 if (info.rti_flags & RTF_GATEWAY && 458 rt_gateway.sdl_family == AF_INET6) 459 fle6->f.n.next_hop6 = 460 ((struct sockaddr_in6 *)&rt_gateway)->sin6_addr; 461 462 if (info.rti_addrs & RTA_NETMASK) 463 fle6->f.dst_mask = 464 ipv6_masklen(sin6_mask.sin6_addr); 465 else 466 fle6->f.dst_mask = 128; 467 468 rib_free_info(&info); 469 } 470 } 471 472 if ((flags & NG_NETFLOW_CONF_NOSRCLOOKUP) == 0) { 473 /* Do route lookup on source address, to fill in src_mask. */ 474 bzero(&sin6, sizeof(struct sockaddr_in6)); 475 sin6.sin6_len = sizeof(struct sockaddr_in6); 476 sin6.sin6_family = AF_INET6; 477 sin6.sin6_addr = r->src.r_src6; 478 479 sin6_mask.sin6_len = sizeof(struct sockaddr_in6); 480 bzero(&info, sizeof(info)); 481 482 info.rti_info[RTAX_NETMASK] = (struct sockaddr *)&sin6_mask; 483 484 if (rib_lookup_info(r->fib, (struct sockaddr *)&sin6, 0, 0, 485 &info) == 0) { 486 if (info.rti_addrs & RTA_NETMASK) 487 fle6->f.src_mask = 488 ipv6_masklen(sin6_mask.sin6_addr); 489 else 490 fle6->f.src_mask = 128; 491 } 492 } 493 494 /* Push new flow at the and of hash. */ 495 TAILQ_INSERT_TAIL(&hsh6->head, (struct flow_entry *)fle6, fle_hash); 496 497 return (0); 498 } 499 #undef ipv6_masklen 500 #endif 501 502 503 /* 504 * Non-static functions called from ng_netflow.c 505 */ 506 507 /* Allocate memory and set up flow cache */ 508 void 509 ng_netflow_cache_init(priv_p priv) 510 { 511 struct flow_hash_entry *hsh; 512 int i; 513 514 /* Initialize cache UMA zone. */ 515 priv->zone = uma_zcreate("NetFlow IPv4 cache", 516 sizeof(struct flow_entry), NULL, NULL, NULL, NULL, 517 UMA_ALIGN_CACHE, 0); 518 uma_zone_set_max(priv->zone, CACHESIZE); 519 #ifdef INET6 520 priv->zone6 = uma_zcreate("NetFlow IPv6 cache", 521 sizeof(struct flow6_entry), NULL, NULL, NULL, NULL, 522 UMA_ALIGN_CACHE, 0); 523 uma_zone_set_max(priv->zone6, CACHESIZE); 524 #endif 525 526 /* Allocate hash. */ 527 priv->hash = malloc(NBUCKETS * sizeof(struct flow_hash_entry), 528 M_NETFLOW_HASH, M_WAITOK | M_ZERO); 529 530 /* Initialize hash. */ 531 for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++) { 532 mtx_init(&hsh->mtx, "hash mutex", NULL, MTX_DEF); 533 TAILQ_INIT(&hsh->head); 534 } 535 536 #ifdef INET6 537 /* Allocate hash. */ 538 priv->hash6 = malloc(NBUCKETS * sizeof(struct flow_hash_entry), 539 M_NETFLOW_HASH, M_WAITOK | M_ZERO); 540 541 /* Initialize hash. */ 542 for (i = 0, hsh = priv->hash6; i < NBUCKETS; i++, hsh++) { 543 mtx_init(&hsh->mtx, "hash mutex", NULL, MTX_DEF); 544 TAILQ_INIT(&hsh->head); 545 } 546 #endif 547 548 priv->nfinfo_bytes = counter_u64_alloc(M_WAITOK); 549 priv->nfinfo_packets = counter_u64_alloc(M_WAITOK); 550 priv->nfinfo_bytes6 = counter_u64_alloc(M_WAITOK); 551 priv->nfinfo_packets6 = counter_u64_alloc(M_WAITOK); 552 priv->nfinfo_sbytes = counter_u64_alloc(M_WAITOK); 553 priv->nfinfo_spackets = counter_u64_alloc(M_WAITOK); 554 priv->nfinfo_sbytes6 = counter_u64_alloc(M_WAITOK); 555 priv->nfinfo_spackets6 = counter_u64_alloc(M_WAITOK); 556 priv->nfinfo_act_exp = counter_u64_alloc(M_WAITOK); 557 priv->nfinfo_inact_exp = counter_u64_alloc(M_WAITOK); 558 559 ng_netflow_v9_cache_init(priv); 560 CTR0(KTR_NET, "ng_netflow startup()"); 561 } 562 563 /* Initialize new FIB table for v5 and v9 */ 564 int 565 ng_netflow_fib_init(priv_p priv, int fib) 566 { 567 fib_export_p fe = priv_to_fib(priv, fib); 568 569 CTR1(KTR_NET, "ng_netflow(): fib init: %d", fib); 570 571 if (fe != NULL) 572 return (0); 573 574 if ((fe = malloc(sizeof(struct fib_export), M_NETGRAPH, 575 M_NOWAIT | M_ZERO)) == NULL) 576 return (ENOMEM); 577 578 mtx_init(&fe->export_mtx, "export dgram lock", NULL, MTX_DEF); 579 mtx_init(&fe->export9_mtx, "export9 dgram lock", NULL, MTX_DEF); 580 fe->fib = fib; 581 fe->domain_id = fib; 582 583 if (atomic_cmpset_ptr((volatile uintptr_t *)&priv->fib_data[fib], 584 (uintptr_t)NULL, (uintptr_t)fe) == 0) { 585 /* FIB already set up by other ISR */ 586 CTR3(KTR_NET, "ng_netflow(): fib init: %d setup %p but got %p", 587 fib, fe, priv_to_fib(priv, fib)); 588 mtx_destroy(&fe->export_mtx); 589 mtx_destroy(&fe->export9_mtx); 590 free(fe, M_NETGRAPH); 591 } else { 592 /* Increase counter for statistics */ 593 CTR3(KTR_NET, "ng_netflow(): fib %d setup to %p (%p)", 594 fib, fe, priv_to_fib(priv, fib)); 595 priv->nfinfo_alloc_fibs++; 596 } 597 598 return (0); 599 } 600 601 /* Free all flow cache memory. Called from node close method. */ 602 void 603 ng_netflow_cache_flush(priv_p priv) 604 { 605 struct flow_entry *fle, *fle1; 606 struct flow_hash_entry *hsh; 607 struct netflow_export_item exp; 608 fib_export_p fe; 609 int i; 610 611 bzero(&exp, sizeof(exp)); 612 613 /* 614 * We are going to free probably billable data. 615 * Expire everything before freeing it. 616 * No locking is required since callout is already drained. 617 */ 618 for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++) 619 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) { 620 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 621 fe = priv_to_fib(priv, fle->f.r.fib); 622 expire_flow(priv, fe, fle, NG_QUEUE); 623 } 624 #ifdef INET6 625 for (hsh = priv->hash6, i = 0; i < NBUCKETS; hsh++, i++) 626 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) { 627 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 628 fe = priv_to_fib(priv, fle->f.r.fib); 629 expire_flow(priv, fe, fle, NG_QUEUE); 630 } 631 #endif 632 633 uma_zdestroy(priv->zone); 634 /* Destroy hash mutexes. */ 635 for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++) 636 mtx_destroy(&hsh->mtx); 637 638 /* Free hash memory. */ 639 if (priv->hash != NULL) 640 free(priv->hash, M_NETFLOW_HASH); 641 #ifdef INET6 642 uma_zdestroy(priv->zone6); 643 /* Destroy hash mutexes. */ 644 for (i = 0, hsh = priv->hash6; i < NBUCKETS; i++, hsh++) 645 mtx_destroy(&hsh->mtx); 646 647 /* Free hash memory. */ 648 if (priv->hash6 != NULL) 649 free(priv->hash6, M_NETFLOW_HASH); 650 #endif 651 652 for (i = 0; i < priv->maxfibs; i++) { 653 if ((fe = priv_to_fib(priv, i)) == NULL) 654 continue; 655 656 if (fe->exp.item != NULL) 657 export_send(priv, fe, fe->exp.item, NG_QUEUE); 658 659 if (fe->exp.item9 != NULL) 660 export9_send(priv, fe, fe->exp.item9, 661 fe->exp.item9_opt, NG_QUEUE); 662 663 mtx_destroy(&fe->export_mtx); 664 mtx_destroy(&fe->export9_mtx); 665 free(fe, M_NETGRAPH); 666 } 667 668 counter_u64_free(priv->nfinfo_bytes); 669 counter_u64_free(priv->nfinfo_packets); 670 counter_u64_free(priv->nfinfo_bytes6); 671 counter_u64_free(priv->nfinfo_packets6); 672 counter_u64_free(priv->nfinfo_sbytes); 673 counter_u64_free(priv->nfinfo_spackets); 674 counter_u64_free(priv->nfinfo_sbytes6); 675 counter_u64_free(priv->nfinfo_spackets6); 676 counter_u64_free(priv->nfinfo_act_exp); 677 counter_u64_free(priv->nfinfo_inact_exp); 678 679 ng_netflow_v9_cache_flush(priv); 680 } 681 682 /* Insert packet from into flow cache. */ 683 int 684 ng_netflow_flow_add(priv_p priv, fib_export_p fe, struct ip *ip, 685 caddr_t upper_ptr, uint8_t upper_proto, uint8_t flags, 686 unsigned int src_if_index) 687 { 688 struct flow_entry *fle, *fle1; 689 struct flow_hash_entry *hsh; 690 struct flow_rec r; 691 int hlen, plen; 692 int error = 0; 693 uint16_t eproto; 694 uint8_t tcp_flags = 0; 695 696 bzero(&r, sizeof(r)); 697 698 if (ip->ip_v != IPVERSION) 699 return (EINVAL); 700 701 hlen = ip->ip_hl << 2; 702 if (hlen < sizeof(struct ip)) 703 return (EINVAL); 704 705 eproto = ETHERTYPE_IP; 706 /* Assume L4 template by default */ 707 r.flow_type = NETFLOW_V9_FLOW_V4_L4; 708 709 r.r_src = ip->ip_src; 710 r.r_dst = ip->ip_dst; 711 r.fib = fe->fib; 712 713 plen = ntohs(ip->ip_len); 714 715 r.r_ip_p = ip->ip_p; 716 r.r_tos = ip->ip_tos; 717 718 r.r_i_ifx = src_if_index; 719 720 /* 721 * XXX NOTE: only first fragment of fragmented TCP, UDP and 722 * ICMP packet will be recorded with proper s_port and d_port. 723 * Following fragments will be recorded simply as IP packet with 724 * ip_proto = ip->ip_p and s_port, d_port set to zero. 725 * I know, it looks like bug. But I don't want to re-implement 726 * ip packet assebmling here. Anyway, (in)famous trafd works this way - 727 * and nobody complains yet :) 728 */ 729 if ((ip->ip_off & htons(IP_OFFMASK)) == 0) 730 switch(r.r_ip_p) { 731 case IPPROTO_TCP: 732 { 733 struct tcphdr *tcp; 734 735 tcp = (struct tcphdr *)((caddr_t )ip + hlen); 736 r.r_sport = tcp->th_sport; 737 r.r_dport = tcp->th_dport; 738 tcp_flags = tcp->th_flags; 739 break; 740 } 741 case IPPROTO_UDP: 742 r.r_ports = *(uint32_t *)((caddr_t )ip + hlen); 743 break; 744 } 745 746 counter_u64_add(priv->nfinfo_packets, 1); 747 counter_u64_add(priv->nfinfo_bytes, plen); 748 749 /* Find hash slot. */ 750 hsh = &priv->hash[ip_hash(&r)]; 751 752 mtx_lock(&hsh->mtx); 753 754 /* 755 * Go through hash and find our entry. If we encounter an 756 * entry, that should be expired, purge it. We do a reverse 757 * search since most active entries are first, and most 758 * searches are done on most active entries. 759 */ 760 TAILQ_FOREACH_REVERSE_SAFE(fle, &hsh->head, fhead, fle_hash, fle1) { 761 if (bcmp(&r, &fle->f.r, sizeof(struct flow_rec)) == 0) 762 break; 763 if ((INACTIVE(fle) && SMALL(fle)) || AGED(fle)) { 764 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 765 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), 766 fle, NG_QUEUE); 767 counter_u64_add(priv->nfinfo_act_exp, 1); 768 } 769 } 770 771 if (fle) { /* An existent entry. */ 772 773 fle->f.bytes += plen; 774 fle->f.packets ++; 775 fle->f.tcp_flags |= tcp_flags; 776 fle->f.last = time_uptime; 777 778 /* 779 * We have the following reasons to expire flow in active way: 780 * - it hit active timeout 781 * - a TCP connection closed 782 * - it is going to overflow counter 783 */ 784 if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle) || 785 (fle->f.bytes >= (CNTR_MAX - IF_MAXMTU)) ) { 786 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 787 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), 788 fle, NG_QUEUE); 789 counter_u64_add(priv->nfinfo_act_exp, 1); 790 } else { 791 /* 792 * It is the newest, move it to the tail, 793 * if it isn't there already. Next search will 794 * locate it quicker. 795 */ 796 if (fle != TAILQ_LAST(&hsh->head, fhead)) { 797 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 798 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash); 799 } 800 } 801 } else /* A new flow entry. */ 802 error = hash_insert(priv, hsh, &r, plen, flags, tcp_flags); 803 804 mtx_unlock(&hsh->mtx); 805 806 return (error); 807 } 808 809 #ifdef INET6 810 /* Insert IPv6 packet from into flow cache. */ 811 int 812 ng_netflow_flow6_add(priv_p priv, fib_export_p fe, struct ip6_hdr *ip6, 813 caddr_t upper_ptr, uint8_t upper_proto, uint8_t flags, 814 unsigned int src_if_index) 815 { 816 struct flow_entry *fle = NULL, *fle1; 817 struct flow6_entry *fle6; 818 struct flow_hash_entry *hsh; 819 struct flow6_rec r; 820 int plen; 821 int error = 0; 822 uint8_t tcp_flags = 0; 823 824 /* check version */ 825 if ((ip6->ip6_vfc & IPV6_VERSION_MASK) != IPV6_VERSION) 826 return (EINVAL); 827 828 bzero(&r, sizeof(r)); 829 830 r.src.r_src6 = ip6->ip6_src; 831 r.dst.r_dst6 = ip6->ip6_dst; 832 r.fib = fe->fib; 833 834 /* Assume L4 template by default */ 835 r.flow_type = NETFLOW_V9_FLOW_V6_L4; 836 837 plen = ntohs(ip6->ip6_plen) + sizeof(struct ip6_hdr); 838 839 #if 0 840 /* XXX: set DSCP/CoS value */ 841 r.r_tos = ip->ip_tos; 842 #endif 843 if ((flags & NG_NETFLOW_IS_FRAG) == 0) { 844 switch(upper_proto) { 845 case IPPROTO_TCP: 846 { 847 struct tcphdr *tcp; 848 849 tcp = (struct tcphdr *)upper_ptr; 850 r.r_ports = *(uint32_t *)upper_ptr; 851 tcp_flags = tcp->th_flags; 852 break; 853 } 854 case IPPROTO_UDP: 855 case IPPROTO_SCTP: 856 r.r_ports = *(uint32_t *)upper_ptr; 857 break; 858 } 859 } 860 861 r.r_ip_p = upper_proto; 862 r.r_i_ifx = src_if_index; 863 864 counter_u64_add(priv->nfinfo_packets6, 1); 865 counter_u64_add(priv->nfinfo_bytes6, plen); 866 867 /* Find hash slot. */ 868 hsh = &priv->hash6[ip6_hash(&r)]; 869 870 mtx_lock(&hsh->mtx); 871 872 /* 873 * Go through hash and find our entry. If we encounter an 874 * entry, that should be expired, purge it. We do a reverse 875 * search since most active entries are first, and most 876 * searches are done on most active entries. 877 */ 878 TAILQ_FOREACH_REVERSE_SAFE(fle, &hsh->head, fhead, fle_hash, fle1) { 879 if (fle->f.version != IP6VERSION) 880 continue; 881 fle6 = (struct flow6_entry *)fle; 882 if (bcmp(&r, &fle6->f.r, sizeof(struct flow6_rec)) == 0) 883 break; 884 if ((INACTIVE(fle6) && SMALL(fle6)) || AGED(fle6)) { 885 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 886 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), fle, 887 NG_QUEUE); 888 counter_u64_add(priv->nfinfo_act_exp, 1); 889 } 890 } 891 892 if (fle != NULL) { /* An existent entry. */ 893 fle6 = (struct flow6_entry *)fle; 894 895 fle6->f.bytes += plen; 896 fle6->f.packets ++; 897 fle6->f.tcp_flags |= tcp_flags; 898 fle6->f.last = time_uptime; 899 900 /* 901 * We have the following reasons to expire flow in active way: 902 * - it hit active timeout 903 * - a TCP connection closed 904 * - it is going to overflow counter 905 */ 906 if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle6) || 907 (fle6->f.bytes >= (CNTR_MAX - IF_MAXMTU)) ) { 908 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 909 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), fle, 910 NG_QUEUE); 911 counter_u64_add(priv->nfinfo_act_exp, 1); 912 } else { 913 /* 914 * It is the newest, move it to the tail, 915 * if it isn't there already. Next search will 916 * locate it quicker. 917 */ 918 if (fle != TAILQ_LAST(&hsh->head, fhead)) { 919 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 920 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash); 921 } 922 } 923 } else /* A new flow entry. */ 924 error = hash6_insert(priv, hsh, &r, plen, flags, tcp_flags); 925 926 mtx_unlock(&hsh->mtx); 927 928 return (error); 929 } 930 #endif 931 932 /* 933 * Return records from cache to userland. 934 * 935 * TODO: matching particular IP should be done in kernel, here. 936 */ 937 int 938 ng_netflow_flow_show(priv_p priv, struct ngnf_show_header *req, 939 struct ngnf_show_header *resp) 940 { 941 struct flow_hash_entry *hsh; 942 struct flow_entry *fle; 943 struct flow_entry_data *data = (struct flow_entry_data *)(resp + 1); 944 #ifdef INET6 945 struct flow6_entry_data *data6 = (struct flow6_entry_data *)(resp + 1); 946 #endif 947 int i, max; 948 949 i = req->hash_id; 950 if (i > NBUCKETS-1) 951 return (EINVAL); 952 953 #ifdef INET6 954 if (req->version == 6) { 955 resp->version = 6; 956 hsh = priv->hash6 + i; 957 max = NREC6_AT_ONCE; 958 } else 959 #endif 960 if (req->version == 4) { 961 resp->version = 4; 962 hsh = priv->hash + i; 963 max = NREC_AT_ONCE; 964 } else 965 return (EINVAL); 966 967 /* 968 * We will transfer not more than NREC_AT_ONCE. More data 969 * will come in next message. 970 * We send current hash index and current record number in list 971 * to userland, and userland should return it back to us. 972 * Then, we will restart with new entry. 973 * 974 * The resulting cache snapshot can be inaccurate if flow expiration 975 * is taking place on hash item between userland data requests for 976 * this hash item id. 977 */ 978 resp->nentries = 0; 979 for (; i < NBUCKETS; hsh++, i++) { 980 int list_id; 981 982 if (mtx_trylock(&hsh->mtx) == 0) { 983 /* 984 * Requested hash index is not available, 985 * relay decision to skip or re-request data 986 * to userland. 987 */ 988 resp->hash_id = i; 989 resp->list_id = 0; 990 return (0); 991 } 992 993 list_id = 0; 994 TAILQ_FOREACH(fle, &hsh->head, fle_hash) { 995 if (hsh->mtx.mtx_lock & MTX_CONTESTED) { 996 resp->hash_id = i; 997 resp->list_id = list_id; 998 mtx_unlock(&hsh->mtx); 999 return (0); 1000 } 1001 1002 list_id++; 1003 /* Search for particular record in list. */ 1004 if (req->list_id > 0) { 1005 if (list_id < req->list_id) 1006 continue; 1007 1008 /* Requested list position found. */ 1009 req->list_id = 0; 1010 } 1011 #ifdef INET6 1012 if (req->version == 6) { 1013 struct flow6_entry *fle6; 1014 1015 fle6 = (struct flow6_entry *)fle; 1016 bcopy(&fle6->f, data6 + resp->nentries, 1017 sizeof(fle6->f)); 1018 } else 1019 #endif 1020 bcopy(&fle->f, data + resp->nentries, 1021 sizeof(fle->f)); 1022 resp->nentries++; 1023 if (resp->nentries == max) { 1024 resp->hash_id = i; 1025 /* 1026 * If it was the last item in list 1027 * we simply skip to next hash_id. 1028 */ 1029 resp->list_id = list_id + 1; 1030 mtx_unlock(&hsh->mtx); 1031 return (0); 1032 } 1033 } 1034 mtx_unlock(&hsh->mtx); 1035 } 1036 1037 resp->hash_id = resp->list_id = 0; 1038 1039 return (0); 1040 } 1041 1042 /* We have full datagram in privdata. Send it to export hook. */ 1043 static int 1044 export_send(priv_p priv, fib_export_p fe, item_p item, int flags) 1045 { 1046 struct mbuf *m = NGI_M(item); 1047 struct netflow_v5_export_dgram *dgram = mtod(m, 1048 struct netflow_v5_export_dgram *); 1049 struct netflow_v5_header *header = &dgram->header; 1050 struct timespec ts; 1051 int error = 0; 1052 1053 /* Fill mbuf header. */ 1054 m->m_len = m->m_pkthdr.len = sizeof(struct netflow_v5_record) * 1055 header->count + sizeof(struct netflow_v5_header); 1056 1057 /* Fill export header. */ 1058 header->sys_uptime = htonl(MILLIUPTIME(time_uptime)); 1059 getnanotime(&ts); 1060 header->unix_secs = htonl(ts.tv_sec); 1061 header->unix_nsecs = htonl(ts.tv_nsec); 1062 header->engine_type = 0; 1063 header->engine_id = fe->domain_id; 1064 header->pad = 0; 1065 header->flow_seq = htonl(atomic_fetchadd_32(&fe->flow_seq, 1066 header->count)); 1067 header->count = htons(header->count); 1068 1069 if (priv->export != NULL) 1070 NG_FWD_ITEM_HOOK_FLAGS(error, item, priv->export, flags); 1071 else 1072 NG_FREE_ITEM(item); 1073 1074 return (error); 1075 } 1076 1077 1078 /* Add export record to dgram. */ 1079 static int 1080 export_add(item_p item, struct flow_entry *fle) 1081 { 1082 struct netflow_v5_export_dgram *dgram = mtod(NGI_M(item), 1083 struct netflow_v5_export_dgram *); 1084 struct netflow_v5_header *header = &dgram->header; 1085 struct netflow_v5_record *rec; 1086 1087 rec = &dgram->r[header->count]; 1088 header->count ++; 1089 1090 KASSERT(header->count <= NETFLOW_V5_MAX_RECORDS, 1091 ("ng_netflow: export too big")); 1092 1093 /* Fill in export record. */ 1094 rec->src_addr = fle->f.r.r_src.s_addr; 1095 rec->dst_addr = fle->f.r.r_dst.s_addr; 1096 rec->next_hop = fle->f.next_hop.s_addr; 1097 rec->i_ifx = htons(fle->f.fle_i_ifx); 1098 rec->o_ifx = htons(fle->f.fle_o_ifx); 1099 rec->packets = htonl(fle->f.packets); 1100 rec->octets = htonl(fle->f.bytes); 1101 rec->first = htonl(MILLIUPTIME(fle->f.first)); 1102 rec->last = htonl(MILLIUPTIME(fle->f.last)); 1103 rec->s_port = fle->f.r.r_sport; 1104 rec->d_port = fle->f.r.r_dport; 1105 rec->flags = fle->f.tcp_flags; 1106 rec->prot = fle->f.r.r_ip_p; 1107 rec->tos = fle->f.r.r_tos; 1108 rec->dst_mask = fle->f.dst_mask; 1109 rec->src_mask = fle->f.src_mask; 1110 rec->pad1 = 0; 1111 rec->pad2 = 0; 1112 1113 /* Not supported fields. */ 1114 rec->src_as = rec->dst_as = 0; 1115 1116 if (header->count == NETFLOW_V5_MAX_RECORDS) 1117 return (1); /* end of datagram */ 1118 else 1119 return (0); 1120 } 1121 1122 /* Periodic flow expiry run. */ 1123 void 1124 ng_netflow_expire(void *arg) 1125 { 1126 struct flow_entry *fle, *fle1; 1127 struct flow_hash_entry *hsh; 1128 priv_p priv = (priv_p )arg; 1129 int used, i; 1130 1131 /* 1132 * Going through all the cache. 1133 */ 1134 used = uma_zone_get_cur(priv->zone); 1135 for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++) { 1136 /* 1137 * Skip entries, that are already being worked on. 1138 */ 1139 if (mtx_trylock(&hsh->mtx) == 0) 1140 continue; 1141 1142 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) { 1143 /* 1144 * Interrupt thread wants this entry! 1145 * Quick! Quick! Bail out! 1146 */ 1147 if (hsh->mtx.mtx_lock & MTX_CONTESTED) 1148 break; 1149 1150 /* 1151 * Don't expire aggressively while hash collision 1152 * ratio is predicted small. 1153 */ 1154 if (used <= (NBUCKETS*2) && !INACTIVE(fle)) 1155 break; 1156 1157 if ((INACTIVE(fle) && (SMALL(fle) || 1158 (used > (NBUCKETS*2)))) || AGED(fle)) { 1159 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 1160 expire_flow(priv, priv_to_fib(priv, 1161 fle->f.r.fib), fle, NG_NOFLAGS); 1162 used--; 1163 counter_u64_add(priv->nfinfo_inact_exp, 1); 1164 } 1165 } 1166 mtx_unlock(&hsh->mtx); 1167 } 1168 1169 #ifdef INET6 1170 used = uma_zone_get_cur(priv->zone6); 1171 for (hsh = priv->hash6, i = 0; i < NBUCKETS; hsh++, i++) { 1172 struct flow6_entry *fle6; 1173 1174 /* 1175 * Skip entries, that are already being worked on. 1176 */ 1177 if (mtx_trylock(&hsh->mtx) == 0) 1178 continue; 1179 1180 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) { 1181 fle6 = (struct flow6_entry *)fle; 1182 /* 1183 * Interrupt thread wants this entry! 1184 * Quick! Quick! Bail out! 1185 */ 1186 if (hsh->mtx.mtx_lock & MTX_CONTESTED) 1187 break; 1188 1189 /* 1190 * Don't expire aggressively while hash collision 1191 * ratio is predicted small. 1192 */ 1193 if (used <= (NBUCKETS*2) && !INACTIVE(fle6)) 1194 break; 1195 1196 if ((INACTIVE(fle6) && (SMALL(fle6) || 1197 (used > (NBUCKETS*2)))) || AGED(fle6)) { 1198 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 1199 expire_flow(priv, priv_to_fib(priv, 1200 fle->f.r.fib), fle, NG_NOFLAGS); 1201 used--; 1202 counter_u64_add(priv->nfinfo_inact_exp, 1); 1203 } 1204 } 1205 mtx_unlock(&hsh->mtx); 1206 } 1207 #endif 1208 1209 /* Schedule next expire. */ 1210 callout_reset(&priv->exp_callout, (1*hz), &ng_netflow_expire, 1211 (void *)priv); 1212 } 1213