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/limits.h> 41 #include <sys/mbuf.h> 42 #include <sys/syslog.h> 43 #include <sys/socket.h> 44 45 #include <net/if.h> 46 #include <net/if_var.h> 47 #include <net/route.h> 48 #include <net/ethernet.h> 49 #include <netinet/in.h> 50 #include <netinet/in_systm.h> 51 #include <netinet/ip.h> 52 #include <netinet/ip6.h> 53 #include <netinet/tcp.h> 54 #include <netinet/udp.h> 55 56 #include <netgraph/ng_message.h> 57 #include <netgraph/netgraph.h> 58 59 #include <netgraph/netflow/netflow.h> 60 #include <netgraph/netflow/netflow_v9.h> 61 #include <netgraph/netflow/ng_netflow.h> 62 63 #define NBUCKETS (65536) /* must be power of 2 */ 64 65 /* This hash is for TCP or UDP packets. */ 66 #define FULL_HASH(addr1, addr2, port1, port2) \ 67 (((addr1 ^ (addr1 >> 16) ^ \ 68 htons(addr2 ^ (addr2 >> 16))) ^ \ 69 port1 ^ htons(port2)) & \ 70 (NBUCKETS - 1)) 71 72 /* This hash is for all other IP packets. */ 73 #define ADDR_HASH(addr1, addr2) \ 74 ((addr1 ^ (addr1 >> 16) ^ \ 75 htons(addr2 ^ (addr2 >> 16))) & \ 76 (NBUCKETS - 1)) 77 78 /* Macros to shorten logical constructions */ 79 /* XXX: priv must exist in namespace */ 80 #define INACTIVE(fle) (time_uptime - fle->f.last > priv->nfinfo_inact_t) 81 #define AGED(fle) (time_uptime - fle->f.first > priv->nfinfo_act_t) 82 #define ISFREE(fle) (fle->f.packets == 0) 83 84 /* 85 * 4 is a magical number: statistically number of 4-packet flows is 86 * bigger than 5,6,7...-packet flows by an order of magnitude. Most UDP/ICMP 87 * scans are 1 packet (~ 90% of flow cache). TCP scans are 2-packet in case 88 * of reachable host and 4-packet otherwise. 89 */ 90 #define SMALL(fle) (fle->f.packets <= 4) 91 92 MALLOC_DEFINE(M_NETFLOW_HASH, "netflow_hash", "NetFlow hash"); 93 94 static int export_add(item_p, struct flow_entry *); 95 static int export_send(priv_p, fib_export_p, item_p, int); 96 97 static int hash_insert(priv_p, struct flow_hash_entry *, struct flow_rec *, 98 int, uint8_t, uint8_t); 99 #ifdef INET6 100 static int hash6_insert(priv_p, struct flow_hash_entry *, struct flow6_rec *, 101 int, uint8_t, uint8_t); 102 #endif 103 104 static void expire_flow(priv_p, fib_export_p, struct flow_entry *, int); 105 106 /* 107 * Generate hash for a given flow record. 108 * 109 * FIB is not used here, because: 110 * most VRFS will carry public IPv4 addresses which are unique even 111 * without FIB private addresses can overlap, but this is worked out 112 * via flow_rec bcmp() containing fib id. In IPv6 world addresses are 113 * all globally unique (it's not fully true, there is FC00::/7 for example, 114 * but chances of address overlap are MUCH smaller) 115 */ 116 static inline uint32_t 117 ip_hash(struct flow_rec *r) 118 { 119 120 switch (r->r_ip_p) { 121 case IPPROTO_TCP: 122 case IPPROTO_UDP: 123 return FULL_HASH(r->r_src.s_addr, r->r_dst.s_addr, 124 r->r_sport, r->r_dport); 125 default: 126 return ADDR_HASH(r->r_src.s_addr, r->r_dst.s_addr); 127 } 128 } 129 130 #ifdef INET6 131 /* Generate hash for a given flow6 record. Use lower 4 octets from v6 addresses */ 132 static inline uint32_t 133 ip6_hash(struct flow6_rec *r) 134 { 135 136 switch (r->r_ip_p) { 137 case IPPROTO_TCP: 138 case IPPROTO_UDP: 139 return FULL_HASH(r->src.r_src6.__u6_addr.__u6_addr32[3], 140 r->dst.r_dst6.__u6_addr.__u6_addr32[3], r->r_sport, 141 r->r_dport); 142 default: 143 return ADDR_HASH(r->src.r_src6.__u6_addr.__u6_addr32[3], 144 r->dst.r_dst6.__u6_addr.__u6_addr32[3]); 145 } 146 } 147 #endif 148 149 /* 150 * Detach export datagram from priv, if there is any. 151 * If there is no, allocate a new one. 152 */ 153 static item_p 154 get_export_dgram(priv_p priv, fib_export_p fe) 155 { 156 item_p item = NULL; 157 158 mtx_lock(&fe->export_mtx); 159 if (fe->exp.item != NULL) { 160 item = fe->exp.item; 161 fe->exp.item = NULL; 162 } 163 mtx_unlock(&fe->export_mtx); 164 165 if (item == NULL) { 166 struct netflow_v5_export_dgram *dgram; 167 struct mbuf *m; 168 169 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 170 if (m == NULL) 171 return (NULL); 172 item = ng_package_data(m, NG_NOFLAGS); 173 if (item == NULL) 174 return (NULL); 175 dgram = mtod(m, struct netflow_v5_export_dgram *); 176 dgram->header.count = 0; 177 dgram->header.version = htons(NETFLOW_V5); 178 dgram->header.pad = 0; 179 } 180 181 return (item); 182 } 183 184 /* 185 * Re-attach incomplete datagram back to priv. 186 * If there is already another one, then send incomplete. */ 187 static void 188 return_export_dgram(priv_p priv, fib_export_p fe, item_p item, int flags) 189 { 190 191 /* 192 * It may happen on SMP, that some thread has already 193 * put its item there, in this case we bail out and 194 * send what we have to collector. 195 */ 196 mtx_lock(&fe->export_mtx); 197 if (fe->exp.item == NULL) { 198 fe->exp.item = item; 199 mtx_unlock(&fe->export_mtx); 200 } else { 201 mtx_unlock(&fe->export_mtx); 202 export_send(priv, fe, item, flags); 203 } 204 } 205 206 /* 207 * The flow is over. Call export_add() and free it. If datagram is 208 * full, then call export_send(). 209 */ 210 static void 211 expire_flow(priv_p priv, fib_export_p fe, struct flow_entry *fle, int flags) 212 { 213 struct netflow_export_item exp; 214 uint16_t version = fle->f.version; 215 216 if ((priv->export != NULL) && (version == IPVERSION)) { 217 exp.item = get_export_dgram(priv, fe); 218 if (exp.item == NULL) { 219 priv->nfinfo_export_failed++; 220 if (priv->export9 != NULL) 221 priv->nfinfo_export9_failed++; 222 /* fle definitely contains IPv4 flow. */ 223 uma_zfree_arg(priv->zone, fle, priv); 224 return; 225 } 226 227 if (export_add(exp.item, fle) > 0) 228 export_send(priv, fe, exp.item, flags); 229 else 230 return_export_dgram(priv, fe, exp.item, NG_QUEUE); 231 } 232 233 if (priv->export9 != NULL) { 234 exp.item9 = get_export9_dgram(priv, fe, &exp.item9_opt); 235 if (exp.item9 == NULL) { 236 priv->nfinfo_export9_failed++; 237 if (version == IPVERSION) 238 uma_zfree_arg(priv->zone, fle, priv); 239 #ifdef INET6 240 else if (version == IP6VERSION) 241 uma_zfree_arg(priv->zone6, fle, priv); 242 #endif 243 else 244 panic("ng_netflow: Unknown IP proto: %d", 245 version); 246 return; 247 } 248 249 if (export9_add(exp.item9, exp.item9_opt, fle) > 0) 250 export9_send(priv, fe, exp.item9, exp.item9_opt, flags); 251 else 252 return_export9_dgram(priv, fe, exp.item9, 253 exp.item9_opt, NG_QUEUE); 254 } 255 256 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 } 263 264 /* Get a snapshot of node statistics */ 265 void 266 ng_netflow_copyinfo(priv_p priv, struct ng_netflow_info *i) 267 { 268 269 i->nfinfo_bytes = counter_u64_fetch(priv->nfinfo_bytes); 270 i->nfinfo_packets = counter_u64_fetch(priv->nfinfo_packets); 271 i->nfinfo_bytes6 = counter_u64_fetch(priv->nfinfo_bytes6); 272 i->nfinfo_packets6 = counter_u64_fetch(priv->nfinfo_packets6); 273 i->nfinfo_sbytes = counter_u64_fetch(priv->nfinfo_sbytes); 274 i->nfinfo_spackets = counter_u64_fetch(priv->nfinfo_spackets); 275 i->nfinfo_sbytes6 = counter_u64_fetch(priv->nfinfo_sbytes6); 276 i->nfinfo_spackets6 = counter_u64_fetch(priv->nfinfo_spackets6); 277 i->nfinfo_act_exp = counter_u64_fetch(priv->nfinfo_act_exp); 278 i->nfinfo_inact_exp = counter_u64_fetch(priv->nfinfo_inact_exp); 279 280 i->nfinfo_used = uma_zone_get_cur(priv->zone); 281 #ifdef INET6 282 i->nfinfo_used6 = uma_zone_get_cur(priv->zone6); 283 #endif 284 285 i->nfinfo_alloc_failed = priv->nfinfo_alloc_failed; 286 i->nfinfo_export_failed = priv->nfinfo_export_failed; 287 i->nfinfo_export9_failed = priv->nfinfo_export9_failed; 288 i->nfinfo_realloc_mbuf = priv->nfinfo_realloc_mbuf; 289 i->nfinfo_alloc_fibs = priv->nfinfo_alloc_fibs; 290 i->nfinfo_inact_t = priv->nfinfo_inact_t; 291 i->nfinfo_act_t = priv->nfinfo_act_t; 292 } 293 294 /* 295 * Insert a record into defined slot. 296 * 297 * First we get for us a free flow entry, then fill in all 298 * possible fields in it. 299 * 300 * TODO: consider dropping hash mutex while filling in datagram, 301 * as this was done in previous version. Need to test & profile 302 * to be sure. 303 */ 304 static int 305 hash_insert(priv_p priv, struct flow_hash_entry *hsh, struct flow_rec *r, 306 int plen, uint8_t flags, uint8_t tcp_flags) 307 { 308 struct flow_entry *fle; 309 struct sockaddr_in sin; 310 struct rtentry *rt; 311 312 mtx_assert(&hsh->mtx, MA_OWNED); 313 314 fle = uma_zalloc_arg(priv->zone, priv, M_NOWAIT); 315 if (fle == NULL) { 316 priv->nfinfo_alloc_failed++; 317 return (ENOMEM); 318 } 319 320 /* 321 * Now fle is totally ours. It is detached from all lists, 322 * we can safely edit it. 323 */ 324 fle->f.version = IPVERSION; 325 bcopy(r, &fle->f.r, sizeof(struct flow_rec)); 326 fle->f.bytes = plen; 327 fle->f.packets = 1; 328 fle->f.tcp_flags = tcp_flags; 329 330 fle->f.first = fle->f.last = time_uptime; 331 332 /* 333 * First we do route table lookup on destination address. So we can 334 * fill in out_ifx, dst_mask, nexthop, and dst_as in future releases. 335 */ 336 if ((flags & NG_NETFLOW_CONF_NODSTLOOKUP) == 0) { 337 bzero(&sin, sizeof(sin)); 338 sin.sin_len = sizeof(struct sockaddr_in); 339 sin.sin_family = AF_INET; 340 sin.sin_addr = fle->f.r.r_dst; 341 rt = rtalloc1_fib((struct sockaddr *)&sin, 0, 0, r->fib); 342 if (rt != NULL) { 343 fle->f.fle_o_ifx = rt->rt_ifp->if_index; 344 345 if (rt->rt_flags & RTF_GATEWAY && 346 rt->rt_gateway->sa_family == AF_INET) 347 fle->f.next_hop = 348 ((struct sockaddr_in *)(rt->rt_gateway))->sin_addr; 349 350 if (rt_mask(rt)) 351 fle->f.dst_mask = 352 bitcount32(((struct sockaddr_in *)rt_mask(rt))->sin_addr.s_addr); 353 else if (rt->rt_flags & RTF_HOST) 354 /* Give up. We can't determine mask :( */ 355 fle->f.dst_mask = 32; 356 357 RTFREE_LOCKED(rt); 358 } 359 } 360 361 /* Do route lookup on source address, to fill in src_mask. */ 362 if ((flags & NG_NETFLOW_CONF_NOSRCLOOKUP) == 0) { 363 bzero(&sin, sizeof(sin)); 364 sin.sin_len = sizeof(struct sockaddr_in); 365 sin.sin_family = AF_INET; 366 sin.sin_addr = fle->f.r.r_src; 367 rt = rtalloc1_fib((struct sockaddr *)&sin, 0, 0, r->fib); 368 if (rt != NULL) { 369 if (rt_mask(rt)) 370 fle->f.src_mask = 371 bitcount32(((struct sockaddr_in *)rt_mask(rt))->sin_addr.s_addr); 372 else if (rt->rt_flags & RTF_HOST) 373 /* Give up. We can't determine mask :( */ 374 fle->f.src_mask = 32; 375 376 RTFREE_LOCKED(rt); 377 } 378 } 379 380 /* Push new flow at the and of hash. */ 381 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash); 382 383 return (0); 384 } 385 386 #ifdef INET6 387 /* XXX: make normal function, instead of.. */ 388 #define ipv6_masklen(x) bitcount32((x).__u6_addr.__u6_addr32[0]) + \ 389 bitcount32((x).__u6_addr.__u6_addr32[1]) + \ 390 bitcount32((x).__u6_addr.__u6_addr32[2]) + \ 391 bitcount32((x).__u6_addr.__u6_addr32[3]) 392 #define RT_MASK6(x) (ipv6_masklen(((struct sockaddr_in6 *)rt_mask(x))->sin6_addr)) 393 static int 394 hash6_insert(priv_p priv, struct flow_hash_entry *hsh6, struct flow6_rec *r, 395 int plen, uint8_t flags, uint8_t tcp_flags) 396 { 397 struct flow6_entry *fle6; 398 struct sockaddr_in6 *src, *dst; 399 struct rtentry *rt; 400 struct route_in6 rin6; 401 402 mtx_assert(&hsh6->mtx, MA_OWNED); 403 404 fle6 = uma_zalloc_arg(priv->zone6, priv, M_NOWAIT); 405 if (fle6 == NULL) { 406 priv->nfinfo_alloc_failed++; 407 return (ENOMEM); 408 } 409 410 /* 411 * Now fle is totally ours. It is detached from all lists, 412 * we can safely edit it. 413 */ 414 415 fle6->f.version = IP6VERSION; 416 bcopy(r, &fle6->f.r, sizeof(struct flow6_rec)); 417 fle6->f.bytes = plen; 418 fle6->f.packets = 1; 419 fle6->f.tcp_flags = tcp_flags; 420 421 fle6->f.first = fle6->f.last = time_uptime; 422 423 /* 424 * First we do route table lookup on destination address. So we can 425 * fill in out_ifx, dst_mask, nexthop, and dst_as in future releases. 426 */ 427 if ((flags & NG_NETFLOW_CONF_NODSTLOOKUP) == 0) { 428 bzero(&rin6, sizeof(struct route_in6)); 429 dst = (struct sockaddr_in6 *)&rin6.ro_dst; 430 dst->sin6_len = sizeof(struct sockaddr_in6); 431 dst->sin6_family = AF_INET6; 432 dst->sin6_addr = r->dst.r_dst6; 433 434 rin6.ro_rt = rtalloc1_fib((struct sockaddr *)dst, 0, 0, r->fib); 435 436 if (rin6.ro_rt != NULL) { 437 rt = rin6.ro_rt; 438 fle6->f.fle_o_ifx = rt->rt_ifp->if_index; 439 440 if (rt->rt_flags & RTF_GATEWAY && 441 rt->rt_gateway->sa_family == AF_INET6) 442 fle6->f.n.next_hop6 = 443 ((struct sockaddr_in6 *)(rt->rt_gateway))->sin6_addr; 444 445 if (rt_mask(rt)) 446 fle6->f.dst_mask = RT_MASK6(rt); 447 else 448 fle6->f.dst_mask = 128; 449 450 RTFREE_LOCKED(rt); 451 } 452 } 453 454 if ((flags & NG_NETFLOW_CONF_NODSTLOOKUP) == 0) { 455 /* Do route lookup on source address, to fill in src_mask. */ 456 bzero(&rin6, sizeof(struct route_in6)); 457 src = (struct sockaddr_in6 *)&rin6.ro_dst; 458 src->sin6_len = sizeof(struct sockaddr_in6); 459 src->sin6_family = AF_INET6; 460 src->sin6_addr = r->src.r_src6; 461 462 rin6.ro_rt = rtalloc1_fib((struct sockaddr *)src, 0, 0, r->fib); 463 464 if (rin6.ro_rt != NULL) { 465 rt = rin6.ro_rt; 466 467 if (rt_mask(rt)) 468 fle6->f.src_mask = RT_MASK6(rt); 469 else 470 fle6->f.src_mask = 128; 471 472 RTFREE_LOCKED(rt); 473 } 474 } 475 476 /* Push new flow at the and of hash. */ 477 TAILQ_INSERT_TAIL(&hsh6->head, (struct flow_entry *)fle6, fle_hash); 478 479 return (0); 480 } 481 #undef ipv6_masklen 482 #undef RT_MASK6 483 #endif 484 485 486 /* 487 * Non-static functions called from ng_netflow.c 488 */ 489 490 /* Allocate memory and set up flow cache */ 491 void 492 ng_netflow_cache_init(priv_p priv) 493 { 494 struct flow_hash_entry *hsh; 495 int i; 496 497 /* Initialize cache UMA zone. */ 498 priv->zone = uma_zcreate("NetFlow IPv4 cache", 499 sizeof(struct flow_entry), NULL, NULL, NULL, NULL, 500 UMA_ALIGN_CACHE, 0); 501 uma_zone_set_max(priv->zone, CACHESIZE); 502 #ifdef INET6 503 priv->zone6 = uma_zcreate("NetFlow IPv6 cache", 504 sizeof(struct flow6_entry), NULL, NULL, NULL, NULL, 505 UMA_ALIGN_CACHE, 0); 506 uma_zone_set_max(priv->zone6, CACHESIZE); 507 #endif 508 509 /* Allocate hash. */ 510 priv->hash = malloc(NBUCKETS * sizeof(struct flow_hash_entry), 511 M_NETFLOW_HASH, M_WAITOK | M_ZERO); 512 513 /* Initialize hash. */ 514 for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++) { 515 mtx_init(&hsh->mtx, "hash mutex", NULL, MTX_DEF); 516 TAILQ_INIT(&hsh->head); 517 } 518 519 #ifdef INET6 520 /* Allocate hash. */ 521 priv->hash6 = malloc(NBUCKETS * sizeof(struct flow_hash_entry), 522 M_NETFLOW_HASH, M_WAITOK | M_ZERO); 523 524 /* Initialize hash. */ 525 for (i = 0, hsh = priv->hash6; i < NBUCKETS; i++, hsh++) { 526 mtx_init(&hsh->mtx, "hash mutex", NULL, MTX_DEF); 527 TAILQ_INIT(&hsh->head); 528 } 529 #endif 530 531 priv->nfinfo_bytes = counter_u64_alloc(M_WAITOK); 532 priv->nfinfo_packets = counter_u64_alloc(M_WAITOK); 533 priv->nfinfo_bytes6 = counter_u64_alloc(M_WAITOK); 534 priv->nfinfo_packets6 = counter_u64_alloc(M_WAITOK); 535 priv->nfinfo_sbytes = counter_u64_alloc(M_WAITOK); 536 priv->nfinfo_spackets = counter_u64_alloc(M_WAITOK); 537 priv->nfinfo_sbytes6 = counter_u64_alloc(M_WAITOK); 538 priv->nfinfo_spackets6 = counter_u64_alloc(M_WAITOK); 539 priv->nfinfo_act_exp = counter_u64_alloc(M_WAITOK); 540 priv->nfinfo_inact_exp = counter_u64_alloc(M_WAITOK); 541 542 ng_netflow_v9_cache_init(priv); 543 CTR0(KTR_NET, "ng_netflow startup()"); 544 } 545 546 /* Initialize new FIB table for v5 and v9 */ 547 int 548 ng_netflow_fib_init(priv_p priv, int fib) 549 { 550 fib_export_p fe = priv_to_fib(priv, fib); 551 552 CTR1(KTR_NET, "ng_netflow(): fib init: %d", fib); 553 554 if (fe != NULL) 555 return (0); 556 557 if ((fe = malloc(sizeof(struct fib_export), M_NETGRAPH, 558 M_NOWAIT | M_ZERO)) == NULL) 559 return (ENOMEM); 560 561 mtx_init(&fe->export_mtx, "export dgram lock", NULL, MTX_DEF); 562 mtx_init(&fe->export9_mtx, "export9 dgram lock", NULL, MTX_DEF); 563 fe->fib = fib; 564 fe->domain_id = fib; 565 566 if (atomic_cmpset_ptr((volatile uintptr_t *)&priv->fib_data[fib], 567 (uintptr_t)NULL, (uintptr_t)fe) == 0) { 568 /* FIB already set up by other ISR */ 569 CTR3(KTR_NET, "ng_netflow(): fib init: %d setup %p but got %p", 570 fib, fe, priv_to_fib(priv, fib)); 571 mtx_destroy(&fe->export_mtx); 572 mtx_destroy(&fe->export9_mtx); 573 free(fe, M_NETGRAPH); 574 } else { 575 /* Increase counter for statistics */ 576 CTR3(KTR_NET, "ng_netflow(): fib %d setup to %p (%p)", 577 fib, fe, priv_to_fib(priv, fib)); 578 priv->nfinfo_alloc_fibs++; 579 } 580 581 return (0); 582 } 583 584 /* Free all flow cache memory. Called from node close method. */ 585 void 586 ng_netflow_cache_flush(priv_p priv) 587 { 588 struct flow_entry *fle, *fle1; 589 struct flow_hash_entry *hsh; 590 struct netflow_export_item exp; 591 fib_export_p fe; 592 int i; 593 594 bzero(&exp, sizeof(exp)); 595 596 /* 597 * We are going to free probably billable data. 598 * Expire everything before freeing it. 599 * No locking is required since callout is already drained. 600 */ 601 for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++) 602 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) { 603 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 604 fe = priv_to_fib(priv, fle->f.r.fib); 605 expire_flow(priv, fe, fle, NG_QUEUE); 606 } 607 #ifdef INET6 608 for (hsh = priv->hash6, i = 0; i < NBUCKETS; hsh++, i++) 609 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) { 610 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 611 fe = priv_to_fib(priv, fle->f.r.fib); 612 expire_flow(priv, fe, fle, NG_QUEUE); 613 } 614 #endif 615 616 uma_zdestroy(priv->zone); 617 /* Destroy hash mutexes. */ 618 for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++) 619 mtx_destroy(&hsh->mtx); 620 621 /* Free hash memory. */ 622 if (priv->hash != NULL) 623 free(priv->hash, M_NETFLOW_HASH); 624 #ifdef INET6 625 uma_zdestroy(priv->zone6); 626 /* Destroy hash mutexes. */ 627 for (i = 0, hsh = priv->hash6; i < NBUCKETS; i++, hsh++) 628 mtx_destroy(&hsh->mtx); 629 630 /* Free hash memory. */ 631 if (priv->hash6 != NULL) 632 free(priv->hash6, M_NETFLOW_HASH); 633 #endif 634 635 for (i = 0; i < priv->maxfibs; i++) { 636 if ((fe = priv_to_fib(priv, i)) == NULL) 637 continue; 638 639 if (fe->exp.item != NULL) 640 export_send(priv, fe, fe->exp.item, NG_QUEUE); 641 642 if (fe->exp.item9 != NULL) 643 export9_send(priv, fe, fe->exp.item9, 644 fe->exp.item9_opt, NG_QUEUE); 645 646 mtx_destroy(&fe->export_mtx); 647 mtx_destroy(&fe->export9_mtx); 648 free(fe, M_NETGRAPH); 649 } 650 651 counter_u64_free(priv->nfinfo_bytes); 652 counter_u64_free(priv->nfinfo_packets); 653 counter_u64_free(priv->nfinfo_bytes6); 654 counter_u64_free(priv->nfinfo_packets6); 655 counter_u64_free(priv->nfinfo_sbytes); 656 counter_u64_free(priv->nfinfo_spackets); 657 counter_u64_free(priv->nfinfo_sbytes6); 658 counter_u64_free(priv->nfinfo_spackets6); 659 counter_u64_free(priv->nfinfo_act_exp); 660 counter_u64_free(priv->nfinfo_inact_exp); 661 662 ng_netflow_v9_cache_flush(priv); 663 } 664 665 /* Insert packet from into flow cache. */ 666 int 667 ng_netflow_flow_add(priv_p priv, fib_export_p fe, struct ip *ip, 668 caddr_t upper_ptr, uint8_t upper_proto, uint8_t flags, 669 unsigned int src_if_index) 670 { 671 struct flow_entry *fle, *fle1; 672 struct flow_hash_entry *hsh; 673 struct flow_rec r; 674 int hlen, plen; 675 int error = 0; 676 uint16_t eproto; 677 uint8_t tcp_flags = 0; 678 679 bzero(&r, sizeof(r)); 680 681 if (ip->ip_v != IPVERSION) 682 return (EINVAL); 683 684 hlen = ip->ip_hl << 2; 685 if (hlen < sizeof(struct ip)) 686 return (EINVAL); 687 688 eproto = ETHERTYPE_IP; 689 /* Assume L4 template by default */ 690 r.flow_type = NETFLOW_V9_FLOW_V4_L4; 691 692 r.r_src = ip->ip_src; 693 r.r_dst = ip->ip_dst; 694 r.fib = fe->fib; 695 696 plen = ntohs(ip->ip_len); 697 698 r.r_ip_p = ip->ip_p; 699 r.r_tos = ip->ip_tos; 700 701 r.r_i_ifx = src_if_index; 702 703 /* 704 * XXX NOTE: only first fragment of fragmented TCP, UDP and 705 * ICMP packet will be recorded with proper s_port and d_port. 706 * Following fragments will be recorded simply as IP packet with 707 * ip_proto = ip->ip_p and s_port, d_port set to zero. 708 * I know, it looks like bug. But I don't want to re-implement 709 * ip packet assebmling here. Anyway, (in)famous trafd works this way - 710 * and nobody complains yet :) 711 */ 712 if ((ip->ip_off & htons(IP_OFFMASK)) == 0) 713 switch(r.r_ip_p) { 714 case IPPROTO_TCP: 715 { 716 struct tcphdr *tcp; 717 718 tcp = (struct tcphdr *)((caddr_t )ip + hlen); 719 r.r_sport = tcp->th_sport; 720 r.r_dport = tcp->th_dport; 721 tcp_flags = tcp->th_flags; 722 break; 723 } 724 case IPPROTO_UDP: 725 r.r_ports = *(uint32_t *)((caddr_t )ip + hlen); 726 break; 727 } 728 729 counter_u64_add(priv->nfinfo_packets, 1); 730 counter_u64_add(priv->nfinfo_bytes, plen); 731 732 /* Find hash slot. */ 733 hsh = &priv->hash[ip_hash(&r)]; 734 735 mtx_lock(&hsh->mtx); 736 737 /* 738 * Go through hash and find our entry. If we encounter an 739 * entry, that should be expired, purge it. We do a reverse 740 * search since most active entries are first, and most 741 * searches are done on most active entries. 742 */ 743 TAILQ_FOREACH_REVERSE_SAFE(fle, &hsh->head, fhead, fle_hash, fle1) { 744 if (bcmp(&r, &fle->f.r, sizeof(struct flow_rec)) == 0) 745 break; 746 if ((INACTIVE(fle) && SMALL(fle)) || AGED(fle)) { 747 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 748 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), 749 fle, NG_QUEUE); 750 counter_u64_add(priv->nfinfo_act_exp, 1); 751 } 752 } 753 754 if (fle) { /* An existent entry. */ 755 756 fle->f.bytes += plen; 757 fle->f.packets ++; 758 fle->f.tcp_flags |= tcp_flags; 759 fle->f.last = time_uptime; 760 761 /* 762 * We have the following reasons to expire flow in active way: 763 * - it hit active timeout 764 * - a TCP connection closed 765 * - it is going to overflow counter 766 */ 767 if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle) || 768 (fle->f.bytes >= (CNTR_MAX - IF_MAXMTU)) ) { 769 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 770 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), 771 fle, NG_QUEUE); 772 counter_u64_add(priv->nfinfo_act_exp, 1); 773 } else { 774 /* 775 * It is the newest, move it to the tail, 776 * if it isn't there already. Next search will 777 * locate it quicker. 778 */ 779 if (fle != TAILQ_LAST(&hsh->head, fhead)) { 780 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 781 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash); 782 } 783 } 784 } else /* A new flow entry. */ 785 error = hash_insert(priv, hsh, &r, plen, flags, tcp_flags); 786 787 mtx_unlock(&hsh->mtx); 788 789 return (error); 790 } 791 792 #ifdef INET6 793 /* Insert IPv6 packet from into flow cache. */ 794 int 795 ng_netflow_flow6_add(priv_p priv, fib_export_p fe, struct ip6_hdr *ip6, 796 caddr_t upper_ptr, uint8_t upper_proto, uint8_t flags, 797 unsigned int src_if_index) 798 { 799 struct flow_entry *fle = NULL, *fle1; 800 struct flow6_entry *fle6; 801 struct flow_hash_entry *hsh; 802 struct flow6_rec r; 803 int plen; 804 int error = 0; 805 uint8_t tcp_flags = 0; 806 807 /* check version */ 808 if ((ip6->ip6_vfc & IPV6_VERSION_MASK) != IPV6_VERSION) 809 return (EINVAL); 810 811 bzero(&r, sizeof(r)); 812 813 r.src.r_src6 = ip6->ip6_src; 814 r.dst.r_dst6 = ip6->ip6_dst; 815 r.fib = fe->fib; 816 817 /* Assume L4 template by default */ 818 r.flow_type = NETFLOW_V9_FLOW_V6_L4; 819 820 plen = ntohs(ip6->ip6_plen) + sizeof(struct ip6_hdr); 821 822 #if 0 823 /* XXX: set DSCP/CoS value */ 824 r.r_tos = ip->ip_tos; 825 #endif 826 if ((flags & NG_NETFLOW_IS_FRAG) == 0) { 827 switch(upper_proto) { 828 case IPPROTO_TCP: 829 { 830 struct tcphdr *tcp; 831 832 tcp = (struct tcphdr *)upper_ptr; 833 r.r_ports = *(uint32_t *)upper_ptr; 834 tcp_flags = tcp->th_flags; 835 break; 836 } 837 case IPPROTO_UDP: 838 case IPPROTO_SCTP: 839 r.r_ports = *(uint32_t *)upper_ptr; 840 break; 841 } 842 } 843 844 r.r_ip_p = upper_proto; 845 r.r_i_ifx = src_if_index; 846 847 counter_u64_add(priv->nfinfo_packets6, 1); 848 counter_u64_add(priv->nfinfo_bytes6, plen); 849 850 /* Find hash slot. */ 851 hsh = &priv->hash6[ip6_hash(&r)]; 852 853 mtx_lock(&hsh->mtx); 854 855 /* 856 * Go through hash and find our entry. If we encounter an 857 * entry, that should be expired, purge it. We do a reverse 858 * search since most active entries are first, and most 859 * searches are done on most active entries. 860 */ 861 TAILQ_FOREACH_REVERSE_SAFE(fle, &hsh->head, fhead, fle_hash, fle1) { 862 if (fle->f.version != IP6VERSION) 863 continue; 864 fle6 = (struct flow6_entry *)fle; 865 if (bcmp(&r, &fle6->f.r, sizeof(struct flow6_rec)) == 0) 866 break; 867 if ((INACTIVE(fle6) && SMALL(fle6)) || AGED(fle6)) { 868 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 869 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), fle, 870 NG_QUEUE); 871 counter_u64_add(priv->nfinfo_act_exp, 1); 872 } 873 } 874 875 if (fle != NULL) { /* An existent entry. */ 876 fle6 = (struct flow6_entry *)fle; 877 878 fle6->f.bytes += plen; 879 fle6->f.packets ++; 880 fle6->f.tcp_flags |= tcp_flags; 881 fle6->f.last = time_uptime; 882 883 /* 884 * We have the following reasons to expire flow in active way: 885 * - it hit active timeout 886 * - a TCP connection closed 887 * - it is going to overflow counter 888 */ 889 if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle6) || 890 (fle6->f.bytes >= (CNTR_MAX - IF_MAXMTU)) ) { 891 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 892 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), fle, 893 NG_QUEUE); 894 counter_u64_add(priv->nfinfo_act_exp, 1); 895 } else { 896 /* 897 * It is the newest, move it to the tail, 898 * if it isn't there already. Next search will 899 * locate it quicker. 900 */ 901 if (fle != TAILQ_LAST(&hsh->head, fhead)) { 902 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 903 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash); 904 } 905 } 906 } else /* A new flow entry. */ 907 error = hash6_insert(priv, hsh, &r, plen, flags, tcp_flags); 908 909 mtx_unlock(&hsh->mtx); 910 911 return (error); 912 } 913 #endif 914 915 /* 916 * Return records from cache to userland. 917 * 918 * TODO: matching particular IP should be done in kernel, here. 919 */ 920 int 921 ng_netflow_flow_show(priv_p priv, struct ngnf_show_header *req, 922 struct ngnf_show_header *resp) 923 { 924 struct flow_hash_entry *hsh; 925 struct flow_entry *fle; 926 struct flow_entry_data *data = (struct flow_entry_data *)(resp + 1); 927 #ifdef INET6 928 struct flow6_entry_data *data6 = (struct flow6_entry_data *)(resp + 1); 929 #endif 930 int i, max; 931 932 i = req->hash_id; 933 if (i > NBUCKETS-1) 934 return (EINVAL); 935 936 #ifdef INET6 937 if (req->version == 6) { 938 resp->version = 6; 939 hsh = priv->hash6 + i; 940 max = NREC6_AT_ONCE; 941 } else 942 #endif 943 if (req->version == 4) { 944 resp->version = 4; 945 hsh = priv->hash + i; 946 max = NREC_AT_ONCE; 947 } else 948 return (EINVAL); 949 950 /* 951 * We will transfer not more than NREC_AT_ONCE. More data 952 * will come in next message. 953 * We send current hash index and current record number in list 954 * to userland, and userland should return it back to us. 955 * Then, we will restart with new entry. 956 * 957 * The resulting cache snapshot can be inaccurate if flow expiration 958 * is taking place on hash item between userland data requests for 959 * this hash item id. 960 */ 961 resp->nentries = 0; 962 for (; i < NBUCKETS; hsh++, i++) { 963 int list_id; 964 965 if (mtx_trylock(&hsh->mtx) == 0) { 966 /* 967 * Requested hash index is not available, 968 * relay decision to skip or re-request data 969 * to userland. 970 */ 971 resp->hash_id = i; 972 resp->list_id = 0; 973 return (0); 974 } 975 976 list_id = 0; 977 TAILQ_FOREACH(fle, &hsh->head, fle_hash) { 978 if (hsh->mtx.mtx_lock & MTX_CONTESTED) { 979 resp->hash_id = i; 980 resp->list_id = list_id; 981 mtx_unlock(&hsh->mtx); 982 return (0); 983 } 984 985 list_id++; 986 /* Search for particular record in list. */ 987 if (req->list_id > 0) { 988 if (list_id < req->list_id) 989 continue; 990 991 /* Requested list position found. */ 992 req->list_id = 0; 993 } 994 #ifdef INET6 995 if (req->version == 6) { 996 struct flow6_entry *fle6; 997 998 fle6 = (struct flow6_entry *)fle; 999 bcopy(&fle6->f, data6 + resp->nentries, 1000 sizeof(fle6->f)); 1001 } else 1002 #endif 1003 bcopy(&fle->f, data + resp->nentries, 1004 sizeof(fle->f)); 1005 resp->nentries++; 1006 if (resp->nentries == max) { 1007 resp->hash_id = i; 1008 /* 1009 * If it was the last item in list 1010 * we simply skip to next hash_id. 1011 */ 1012 resp->list_id = list_id + 1; 1013 mtx_unlock(&hsh->mtx); 1014 return (0); 1015 } 1016 } 1017 mtx_unlock(&hsh->mtx); 1018 } 1019 1020 resp->hash_id = resp->list_id = 0; 1021 1022 return (0); 1023 } 1024 1025 /* We have full datagram in privdata. Send it to export hook. */ 1026 static int 1027 export_send(priv_p priv, fib_export_p fe, item_p item, int flags) 1028 { 1029 struct mbuf *m = NGI_M(item); 1030 struct netflow_v5_export_dgram *dgram = mtod(m, 1031 struct netflow_v5_export_dgram *); 1032 struct netflow_v5_header *header = &dgram->header; 1033 struct timespec ts; 1034 int error = 0; 1035 1036 /* Fill mbuf header. */ 1037 m->m_len = m->m_pkthdr.len = sizeof(struct netflow_v5_record) * 1038 header->count + sizeof(struct netflow_v5_header); 1039 1040 /* Fill export header. */ 1041 header->sys_uptime = htonl(MILLIUPTIME(time_uptime)); 1042 getnanotime(&ts); 1043 header->unix_secs = htonl(ts.tv_sec); 1044 header->unix_nsecs = htonl(ts.tv_nsec); 1045 header->engine_type = 0; 1046 header->engine_id = fe->domain_id; 1047 header->pad = 0; 1048 header->flow_seq = htonl(atomic_fetchadd_32(&fe->flow_seq, 1049 header->count)); 1050 header->count = htons(header->count); 1051 1052 if (priv->export != NULL) 1053 NG_FWD_ITEM_HOOK_FLAGS(error, item, priv->export, flags); 1054 else 1055 NG_FREE_ITEM(item); 1056 1057 return (error); 1058 } 1059 1060 1061 /* Add export record to dgram. */ 1062 static int 1063 export_add(item_p item, struct flow_entry *fle) 1064 { 1065 struct netflow_v5_export_dgram *dgram = mtod(NGI_M(item), 1066 struct netflow_v5_export_dgram *); 1067 struct netflow_v5_header *header = &dgram->header; 1068 struct netflow_v5_record *rec; 1069 1070 rec = &dgram->r[header->count]; 1071 header->count ++; 1072 1073 KASSERT(header->count <= NETFLOW_V5_MAX_RECORDS, 1074 ("ng_netflow: export too big")); 1075 1076 /* Fill in export record. */ 1077 rec->src_addr = fle->f.r.r_src.s_addr; 1078 rec->dst_addr = fle->f.r.r_dst.s_addr; 1079 rec->next_hop = fle->f.next_hop.s_addr; 1080 rec->i_ifx = htons(fle->f.fle_i_ifx); 1081 rec->o_ifx = htons(fle->f.fle_o_ifx); 1082 rec->packets = htonl(fle->f.packets); 1083 rec->octets = htonl(fle->f.bytes); 1084 rec->first = htonl(MILLIUPTIME(fle->f.first)); 1085 rec->last = htonl(MILLIUPTIME(fle->f.last)); 1086 rec->s_port = fle->f.r.r_sport; 1087 rec->d_port = fle->f.r.r_dport; 1088 rec->flags = fle->f.tcp_flags; 1089 rec->prot = fle->f.r.r_ip_p; 1090 rec->tos = fle->f.r.r_tos; 1091 rec->dst_mask = fle->f.dst_mask; 1092 rec->src_mask = fle->f.src_mask; 1093 rec->pad1 = 0; 1094 rec->pad2 = 0; 1095 1096 /* Not supported fields. */ 1097 rec->src_as = rec->dst_as = 0; 1098 1099 if (header->count == NETFLOW_V5_MAX_RECORDS) 1100 return (1); /* end of datagram */ 1101 else 1102 return (0); 1103 } 1104 1105 /* Periodic flow expiry run. */ 1106 void 1107 ng_netflow_expire(void *arg) 1108 { 1109 struct flow_entry *fle, *fle1; 1110 struct flow_hash_entry *hsh; 1111 priv_p priv = (priv_p )arg; 1112 int used, i; 1113 1114 /* 1115 * Going through all the cache. 1116 */ 1117 used = uma_zone_get_cur(priv->zone); 1118 for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++) { 1119 /* 1120 * Skip entries, that are already being worked on. 1121 */ 1122 if (mtx_trylock(&hsh->mtx) == 0) 1123 continue; 1124 1125 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) { 1126 /* 1127 * Interrupt thread wants this entry! 1128 * Quick! Quick! Bail out! 1129 */ 1130 if (hsh->mtx.mtx_lock & MTX_CONTESTED) 1131 break; 1132 1133 /* 1134 * Don't expire aggressively while hash collision 1135 * ratio is predicted small. 1136 */ 1137 if (used <= (NBUCKETS*2) && !INACTIVE(fle)) 1138 break; 1139 1140 if ((INACTIVE(fle) && (SMALL(fle) || 1141 (used > (NBUCKETS*2)))) || AGED(fle)) { 1142 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 1143 expire_flow(priv, priv_to_fib(priv, 1144 fle->f.r.fib), fle, NG_NOFLAGS); 1145 used--; 1146 counter_u64_add(priv->nfinfo_inact_exp, 1); 1147 } 1148 } 1149 mtx_unlock(&hsh->mtx); 1150 } 1151 1152 #ifdef INET6 1153 used = uma_zone_get_cur(priv->zone6); 1154 for (hsh = priv->hash6, i = 0; i < NBUCKETS; hsh++, i++) { 1155 struct flow6_entry *fle6; 1156 1157 /* 1158 * Skip entries, that are already being worked on. 1159 */ 1160 if (mtx_trylock(&hsh->mtx) == 0) 1161 continue; 1162 1163 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) { 1164 fle6 = (struct flow6_entry *)fle; 1165 /* 1166 * Interrupt thread wants this entry! 1167 * Quick! Quick! Bail out! 1168 */ 1169 if (hsh->mtx.mtx_lock & MTX_CONTESTED) 1170 break; 1171 1172 /* 1173 * Don't expire aggressively while hash collision 1174 * ratio is predicted small. 1175 */ 1176 if (used <= (NBUCKETS*2) && !INACTIVE(fle6)) 1177 break; 1178 1179 if ((INACTIVE(fle6) && (SMALL(fle6) || 1180 (used > (NBUCKETS*2)))) || AGED(fle6)) { 1181 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 1182 expire_flow(priv, priv_to_fib(priv, 1183 fle->f.r.fib), fle, NG_NOFLAGS); 1184 used--; 1185 counter_u64_add(priv->nfinfo_inact_exp, 1); 1186 } 1187 } 1188 mtx_unlock(&hsh->mtx); 1189 } 1190 #endif 1191 1192 /* Schedule next expire. */ 1193 callout_reset(&priv->exp_callout, (1*hz), &ng_netflow_expire, 1194 (void *)priv); 1195 } 1196