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 static const char rcs_id[] = 32 "@(#) $FreeBSD$"; 33 34 #include "opt_inet6.h" 35 #include "opt_route.h" 36 #include <sys/param.h> 37 #include <sys/kernel.h> 38 #include <sys/limits.h> 39 #include <sys/mbuf.h> 40 #include <sys/syslog.h> 41 #include <sys/systm.h> 42 #include <sys/socket.h> 43 #include <sys/endian.h> 44 45 #include <machine/atomic.h> 46 #include <machine/stdarg.h> 47 48 #include <net/if.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->info.nfinfo_inact_t) 83 #define AGED(fle) (time_uptime - fle->f.first > priv->info.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 95 MALLOC_DECLARE(M_NETFLOW_HASH); 96 MALLOC_DEFINE(M_NETFLOW_HASH, "netflow_hash", "NetFlow hash"); 97 98 static int export_add(item_p, struct flow_entry *); 99 static int export_send(priv_p, fib_export_p, item_p, int); 100 101 static int hash_insert(priv_p, struct flow_hash_entry *, struct flow_rec *, int, uint8_t); 102 #ifdef INET6 103 static int hash6_insert(priv_p, struct flow6_hash_entry *, struct flow6_rec *, int, uint8_t); 104 #endif 105 106 static __inline 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 switch (r->r_ip_p) { 122 case IPPROTO_TCP: 123 case IPPROTO_UDP: 124 return FULL_HASH(r->r_src.s_addr, r->r_dst.s_addr, 125 r->r_sport, r->r_dport); 126 default: 127 return ADDR_HASH(r->r_src.s_addr, r->r_dst.s_addr); 128 } 129 } 130 131 #ifdef INET6 132 /* Generate hash for a given flow6 record. Use lower 4 octets from v6 addresses */ 133 static __inline uint32_t 134 ip6_hash(struct flow6_rec *r) 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 /* This is callback from uma(9), called on alloc. */ 150 static int 151 uma_ctor_flow(void *mem, int size, void *arg, int how) 152 { 153 priv_p priv = (priv_p )arg; 154 155 if (atomic_load_acq_32(&priv->info.nfinfo_used) >= CACHESIZE) 156 return (ENOMEM); 157 158 atomic_add_32(&priv->info.nfinfo_used, 1); 159 160 return (0); 161 } 162 163 /* This is callback from uma(9), called on free. */ 164 static void 165 uma_dtor_flow(void *mem, int size, void *arg) 166 { 167 priv_p priv = (priv_p )arg; 168 169 atomic_subtract_32(&priv->info.nfinfo_used, 1); 170 } 171 172 #ifdef INET6 173 /* This is callback from uma(9), called on alloc. */ 174 static int 175 uma_ctor_flow6(void *mem, int size, void *arg, int how) 176 { 177 priv_p priv = (priv_p )arg; 178 179 if (atomic_load_acq_32(&priv->info.nfinfo_used6) >= CACHESIZE) 180 return (ENOMEM); 181 182 atomic_add_32(&priv->info.nfinfo_used6, 1); 183 184 return (0); 185 } 186 187 /* This is callback from uma(9), called on free. */ 188 static void 189 uma_dtor_flow6(void *mem, int size, void *arg) 190 { 191 priv_p priv = (priv_p )arg; 192 193 atomic_subtract_32(&priv->info.nfinfo_used6, 1); 194 } 195 #endif 196 197 /* 198 * Detach export datagram from priv, if there is any. 199 * If there is no, allocate a new one. 200 */ 201 static item_p 202 get_export_dgram(priv_p priv, fib_export_p fe) 203 { 204 item_p item = NULL; 205 206 mtx_lock(&fe->export_mtx); 207 if (fe->exp.item != NULL) { 208 item = fe->exp.item; 209 fe->exp.item = NULL; 210 } 211 mtx_unlock(&fe->export_mtx); 212 213 if (item == NULL) { 214 struct netflow_v5_export_dgram *dgram; 215 struct mbuf *m; 216 217 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); 218 if (m == NULL) 219 return (NULL); 220 item = ng_package_data(m, NG_NOFLAGS); 221 if (item == NULL) 222 return (NULL); 223 dgram = mtod(m, struct netflow_v5_export_dgram *); 224 dgram->header.count = 0; 225 dgram->header.version = htons(NETFLOW_V5); 226 dgram->header.pad = 0; 227 228 } 229 230 return (item); 231 } 232 233 /* 234 * Re-attach incomplete datagram back to priv. 235 * If there is already another one, then send incomplete. */ 236 static void 237 return_export_dgram(priv_p priv, fib_export_p fe, item_p item, int flags) 238 { 239 /* 240 * It may happen on SMP, that some thread has already 241 * put its item there, in this case we bail out and 242 * send what we have to collector. 243 */ 244 mtx_lock(&fe->export_mtx); 245 if (fe->exp.item == NULL) { 246 fe->exp.item = item; 247 mtx_unlock(&fe->export_mtx); 248 } else { 249 mtx_unlock(&fe->export_mtx); 250 export_send(priv, fe, item, flags); 251 } 252 } 253 254 /* 255 * The flow is over. Call export_add() and free it. If datagram is 256 * full, then call export_send(). 257 */ 258 static __inline void 259 expire_flow(priv_p priv, fib_export_p fe, struct flow_entry *fle, int flags) 260 { 261 struct netflow_export_item exp; 262 uint16_t version = fle->f.version; 263 264 if ((priv->export != NULL) && (version == IPVERSION)) { 265 exp.item = get_export_dgram(priv, fe); 266 if (exp.item == NULL) { 267 atomic_add_32(&priv->info.nfinfo_export_failed, 1); 268 if (priv->export9 != NULL) 269 atomic_add_32(&priv->info.nfinfo_export9_failed, 1); 270 /* fle definitely contains IPv4 flow */ 271 uma_zfree_arg(priv->zone, fle, priv); 272 return; 273 } 274 275 if (export_add(exp.item, fle) > 0) 276 export_send(priv, fe, exp.item, flags); 277 else 278 return_export_dgram(priv, fe, exp.item, NG_QUEUE); 279 } 280 281 if (priv->export9 != NULL) { 282 exp.item9 = get_export9_dgram(priv, fe, &exp.item9_opt); 283 if (exp.item9 == NULL) { 284 atomic_add_32(&priv->info.nfinfo_export9_failed, 1); 285 if (version == IPVERSION) 286 uma_zfree_arg(priv->zone, fle, priv); 287 #ifdef INET6 288 else if (version == IP6VERSION) 289 uma_zfree_arg(priv->zone6, fle, priv); 290 #endif 291 else 292 panic("ng_netflow: Unknown IP proto: %d", version); 293 return; 294 } 295 296 if (export9_add(exp.item9, exp.item9_opt, fle) > 0) 297 export9_send(priv, fe, exp.item9, exp.item9_opt, flags); 298 else 299 return_export9_dgram(priv, fe, exp.item9, exp.item9_opt, NG_QUEUE); 300 } 301 302 if (version == IPVERSION) 303 uma_zfree_arg(priv->zone, fle, priv); 304 #ifdef INET6 305 else if (version == IP6VERSION) 306 uma_zfree_arg(priv->zone6, fle, priv); 307 #endif 308 } 309 310 /* Get a snapshot of node statistics */ 311 void 312 ng_netflow_copyinfo(priv_p priv, struct ng_netflow_info *i) 313 { 314 /* XXX: atomic */ 315 memcpy((void *)i, (void *)&priv->info, sizeof(priv->info)); 316 } 317 318 /* 319 * Insert a record into defined slot. 320 * 321 * First we get for us a free flow entry, then fill in all 322 * possible fields in it. 323 * 324 * TODO: consider dropping hash mutex while filling in datagram, 325 * as this was done in previous version. Need to test & profile 326 * to be sure. 327 */ 328 static __inline int 329 hash_insert(priv_p priv, struct flow_hash_entry *hsh, struct flow_rec *r, 330 int plen, uint8_t tcp_flags) 331 { 332 struct flow_entry *fle; 333 struct sockaddr_in sin; 334 struct rtentry *rt; 335 336 mtx_assert(&hsh->mtx, MA_OWNED); 337 338 fle = uma_zalloc_arg(priv->zone, priv, M_NOWAIT); 339 if (fle == NULL) { 340 atomic_add_32(&priv->info.nfinfo_alloc_failed, 1); 341 return (ENOMEM); 342 } 343 344 /* 345 * Now fle is totally ours. It is detached from all lists, 346 * we can safely edit it. 347 */ 348 349 fle->f.version = IPVERSION; 350 bcopy(r, &fle->f.r, sizeof(struct flow_rec)); 351 fle->f.bytes = plen; 352 fle->f.packets = 1; 353 fle->f.tcp_flags = tcp_flags; 354 355 fle->f.first = fle->f.last = time_uptime; 356 357 /* 358 * First we do route table lookup on destination address. So we can 359 * fill in out_ifx, dst_mask, nexthop, and dst_as in future releases. 360 */ 361 bzero(&sin, sizeof(sin)); 362 sin.sin_len = sizeof(struct sockaddr_in); 363 sin.sin_family = AF_INET; 364 sin.sin_addr = fle->f.r.r_dst; 365 rt = rtalloc1_fib((struct sockaddr *)&sin, 0, 0, r->fib); 366 if (rt != NULL) { 367 fle->f.fle_o_ifx = rt->rt_ifp->if_index; 368 369 if (rt->rt_flags & RTF_GATEWAY && 370 rt->rt_gateway->sa_family == AF_INET) 371 fle->f.next_hop = 372 ((struct sockaddr_in *)(rt->rt_gateway))->sin_addr; 373 374 if (rt_mask(rt)) 375 fle->f.dst_mask = bitcount32(((struct sockaddr_in *) 376 rt_mask(rt))->sin_addr.s_addr); 377 else if (rt->rt_flags & RTF_HOST) 378 /* Give up. We can't determine mask :( */ 379 fle->f.dst_mask = 32; 380 381 RTFREE_LOCKED(rt); 382 } 383 384 /* Do route lookup on source address, to fill in src_mask. */ 385 bzero(&sin, sizeof(sin)); 386 sin.sin_len = sizeof(struct sockaddr_in); 387 sin.sin_family = AF_INET; 388 sin.sin_addr = fle->f.r.r_src; 389 rt = rtalloc1_fib((struct sockaddr *)&sin, 0, 0, r->fib); 390 if (rt != NULL) { 391 if (rt_mask(rt)) 392 fle->f.src_mask = bitcount32(((struct sockaddr_in *) 393 rt_mask(rt))->sin_addr.s_addr); 394 else if (rt->rt_flags & RTF_HOST) 395 /* Give up. We can't determine mask :( */ 396 fle->f.src_mask = 32; 397 398 RTFREE_LOCKED(rt); 399 } 400 401 /* Push new flow at the and of hash. */ 402 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash); 403 404 return (0); 405 } 406 407 #ifdef INET6 408 /* XXX: make normal function, instead of.. */ 409 #define ipv6_masklen(x) bitcount32((x).__u6_addr.__u6_addr32[0]) + \ 410 bitcount32((x).__u6_addr.__u6_addr32[1]) + \ 411 bitcount32((x).__u6_addr.__u6_addr32[2]) + \ 412 bitcount32((x).__u6_addr.__u6_addr32[3]) 413 /* XXX: Do we need inline here ? */ 414 static __inline int 415 hash6_insert(priv_p priv, struct flow6_hash_entry *hsh6, struct flow6_rec *r, 416 int plen, uint8_t tcp_flags) 417 { 418 struct flow6_entry *fle6; 419 struct sockaddr_in6 *src, *dst; 420 struct rtentry *rt; 421 struct route_in6 rin6; 422 423 mtx_assert(&hsh6->mtx, MA_OWNED); 424 425 fle6 = uma_zalloc_arg(priv->zone6, priv, M_NOWAIT); 426 if (fle6 == NULL) { 427 atomic_add_32(&priv->info.nfinfo_alloc_failed, 1); 428 return (ENOMEM); 429 } 430 431 /* 432 * Now fle is totally ours. It is detached from all lists, 433 * we can safely edit it. 434 */ 435 436 fle6->f.version = IP6VERSION; 437 bcopy(r, &fle6->f.r, sizeof(struct flow6_rec)); 438 fle6->f.bytes = plen; 439 fle6->f.packets = 1; 440 fle6->f.tcp_flags = tcp_flags; 441 442 fle6->f.first = fle6->f.last = time_uptime; 443 444 /* 445 * First we do route table lookup on destination address. So we can 446 * fill in out_ifx, dst_mask, nexthop, and dst_as in future releases. 447 */ 448 bzero(&rin6, sizeof(struct route_in6)); 449 dst = (struct sockaddr_in6 *)&rin6.ro_dst; 450 dst->sin6_len = sizeof(struct sockaddr_in6); 451 dst->sin6_family = AF_INET6; 452 dst->sin6_addr = r->dst.r_dst6; 453 454 rin6.ro_rt = rtalloc1_fib((struct sockaddr *)dst, 0, 0, r->fib); 455 456 if (rin6.ro_rt != NULL) { 457 rt = rin6.ro_rt; 458 fle6->f.fle_o_ifx = rt->rt_ifp->if_index; 459 460 if (rt->rt_flags & RTF_GATEWAY && 461 rt->rt_gateway->sa_family == AF_INET6) 462 fle6->f.n.next_hop6 = 463 ((struct sockaddr_in6 *)(rt->rt_gateway))->sin6_addr; 464 465 if (rt_mask(rt)) 466 fle6->f.dst_mask = ipv6_masklen(((struct sockaddr_in6 *)rt_mask(rt))->sin6_addr); 467 else 468 fle6->f.dst_mask = 128; 469 470 RTFREE_LOCKED(rt); 471 } 472 473 /* Do route lookup on source address, to fill in src_mask. */ 474 bzero(&rin6, sizeof(struct route_in6)); 475 src = (struct sockaddr_in6 *)&rin6.ro_dst; 476 src->sin6_len = sizeof(struct sockaddr_in6); 477 src->sin6_family = AF_INET6; 478 src->sin6_addr = r->src.r_src6; 479 480 rin6.ro_rt = rtalloc1_fib((struct sockaddr *)src, 0, 0, r->fib); 481 482 if (rin6.ro_rt != NULL) { 483 rt = rin6.ro_rt; 484 485 if (rt_mask(rt)) 486 fle6->f.src_mask = ipv6_masklen(((struct sockaddr_in6 *)rt_mask(rt))->sin6_addr); 487 else 488 fle6->f.src_mask = 128; 489 490 RTFREE_LOCKED(rt); 491 } 492 493 /* Push new flow at the and of hash. */ 494 TAILQ_INSERT_TAIL(&hsh6->head, fle6, fle6_hash); 495 496 return (0); 497 } 498 #endif 499 500 501 /* 502 * Non-static functions called from ng_netflow.c 503 */ 504 505 /* Allocate memory and set up flow cache */ 506 int 507 ng_netflow_cache_init(priv_p priv) 508 { 509 struct flow_hash_entry *hsh; 510 #ifdef INET6 511 struct flow6_hash_entry *hsh6; 512 #endif 513 int i; 514 515 /* Initialize cache UMA zone. */ 516 priv->zone = uma_zcreate("NetFlow IPv4 cache", sizeof(struct flow_entry), 517 uma_ctor_flow, uma_dtor_flow, NULL, NULL, UMA_ALIGN_CACHE, 0); 518 uma_zone_set_max(priv->zone, CACHESIZE); 519 #ifdef INET6 520 priv->zone6 = uma_zcreate("NetFlow IPv6 cache", sizeof(struct flow6_entry), 521 uma_ctor_flow6, uma_dtor_flow6, NULL, NULL, UMA_ALIGN_CACHE, 0); 522 uma_zone_set_max(priv->zone6, CACHESIZE); 523 #endif 524 525 /* Allocate hash. */ 526 priv->hash = malloc(NBUCKETS * sizeof(struct flow_hash_entry), 527 M_NETFLOW_HASH, M_WAITOK | M_ZERO); 528 529 /* Initialize hash. */ 530 for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++) { 531 mtx_init(&hsh->mtx, "hash mutex", NULL, MTX_DEF); 532 TAILQ_INIT(&hsh->head); 533 } 534 535 #ifdef INET6 536 /* Allocate hash. */ 537 priv->hash6 = malloc(NBUCKETS * sizeof(struct flow6_hash_entry), 538 M_NETFLOW_HASH, M_WAITOK | M_ZERO); 539 540 /* Initialize hash. */ 541 for (i = 0, hsh6 = priv->hash6; i < NBUCKETS; i++, hsh6++) { 542 mtx_init(&hsh6->mtx, "hash mutex", NULL, MTX_DEF); 543 TAILQ_INIT(&hsh6->head); 544 } 545 #endif 546 547 ng_netflow_v9_cache_init(priv); 548 CTR0(KTR_NET, "ng_netflow startup()"); 549 550 return (0); 551 } 552 553 /* Initialize new FIB table for v5 and v9 */ 554 int 555 ng_netflow_fib_init(priv_p priv, int fib) 556 { 557 fib_export_p fe = priv_to_fib(priv, fib); 558 559 CTR1(KTR_NET, "ng_netflow(): fib init: %d", fib); 560 561 if (fe != NULL) 562 return (0); 563 564 if ((fe = malloc(sizeof(struct fib_export), M_NETGRAPH, M_NOWAIT | M_ZERO)) == NULL) 565 return (1); 566 567 mtx_init(&fe->export_mtx, "export dgram lock", NULL, MTX_DEF); 568 mtx_init(&fe->export9_mtx, "export9 dgram lock", NULL, MTX_DEF); 569 fe->fib = fib; 570 fe->domain_id = fib; 571 572 if (atomic_cmpset_ptr((volatile uintptr_t *)&priv->fib_data[fib], (uintptr_t)NULL, (uintptr_t)fe) == 0) { 573 /* FIB already set up by other ISR */ 574 CTR3(KTR_NET, "ng_netflow(): fib init: %d setup %p but got %p", fib, fe, priv_to_fib(priv, fib)); 575 mtx_destroy(&fe->export_mtx); 576 mtx_destroy(&fe->export9_mtx); 577 free(fe, M_NETGRAPH); 578 } else { 579 /* Increase counter for statistics */ 580 CTR3(KTR_NET, "ng_netflow(): fib %d setup to %p (%p)", fib, fe, priv_to_fib(priv, fib)); 581 atomic_fetchadd_32(&priv->info.nfinfo_alloc_fibs, 1); 582 } 583 584 return (0); 585 } 586 587 /* Free all flow cache memory. Called from node close method. */ 588 void 589 ng_netflow_cache_flush(priv_p priv) 590 { 591 struct flow_entry *fle, *fle1; 592 struct flow_hash_entry *hsh; 593 #ifdef INET6 594 struct flow6_entry *fle6, *fle61; 595 struct flow6_hash_entry *hsh6; 596 #endif 597 struct netflow_export_item exp; 598 fib_export_p fe; 599 int i; 600 601 bzero(&exp, sizeof(exp)); 602 603 /* 604 * We are going to free probably billable data. 605 * Expire everything before freeing it. 606 * No locking is required since callout is already drained. 607 */ 608 for (hsh = priv->hash, 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 #ifdef INET6 615 for (hsh6 = priv->hash6, i = 0; i < NBUCKETS; hsh6++, i++) 616 TAILQ_FOREACH_SAFE(fle6, &hsh6->head, fle6_hash, fle61) { 617 TAILQ_REMOVE(&hsh6->head, fle6, fle6_hash); 618 fe = priv_to_fib(priv, fle6->f.r.fib); 619 expire_flow(priv, fe, (struct flow_entry *)fle6, NG_QUEUE); 620 } 621 #endif 622 623 uma_zdestroy(priv->zone); 624 /* Destroy hash mutexes. */ 625 for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++) 626 mtx_destroy(&hsh->mtx); 627 628 /* Free hash memory. */ 629 if (priv->hash != NULL) 630 free(priv->hash, M_NETFLOW_HASH); 631 #ifdef INET6 632 uma_zdestroy(priv->zone6); 633 /* Destroy hash mutexes. */ 634 for (i = 0, hsh6 = priv->hash6; i < NBUCKETS; i++, hsh6++) 635 mtx_destroy(&hsh6->mtx); 636 637 /* Free hash memory. */ 638 if (priv->hash6 != NULL) 639 free(priv->hash6, M_NETFLOW_HASH); 640 #endif 641 642 for (i = 0; i < RT_NUMFIBS; i++) { 643 if ((fe = priv_to_fib(priv, i)) == NULL) 644 continue; 645 646 if (fe->exp.item != NULL) 647 export_send(priv, fe, fe->exp.item, NG_QUEUE); 648 649 if (fe->exp.item9 != NULL) 650 export9_send(priv, fe, fe->exp.item9, fe->exp.item9_opt, NG_QUEUE); 651 652 mtx_destroy(&fe->export_mtx); 653 mtx_destroy(&fe->export9_mtx); 654 free(fe, M_NETGRAPH); 655 } 656 657 ng_netflow_v9_cache_flush(priv); 658 } 659 660 /* Insert packet from into flow cache. */ 661 int 662 ng_netflow_flow_add(priv_p priv, fib_export_p fe, struct ip *ip, caddr_t upper_ptr, uint8_t upper_proto, 663 uint8_t is_frag, unsigned int src_if_index) 664 { 665 register struct flow_entry *fle, *fle1; 666 struct flow_hash_entry *hsh; 667 struct flow_rec r; 668 int hlen, plen; 669 int error = 0; 670 uint8_t tcp_flags = 0; 671 uint16_t eproto; 672 673 /* Try to fill flow_rec r */ 674 bzero(&r, sizeof(r)); 675 /* check version */ 676 if (ip->ip_v != IPVERSION) 677 return (EINVAL); 678 679 /* verify min header length */ 680 hlen = ip->ip_hl << 2; 681 682 if (hlen < sizeof(struct ip)) 683 return (EINVAL); 684 685 eproto = ETHERTYPE_IP; 686 /* Assume L4 template by default */ 687 r.flow_type = NETFLOW_V9_FLOW_V4_L4; 688 689 r.r_src = ip->ip_src; 690 r.r_dst = ip->ip_dst; 691 r.fib = fe->fib; 692 693 /* save packet length */ 694 plen = ntohs(ip->ip_len); 695 696 r.r_ip_p = ip->ip_p; 697 r.r_tos = ip->ip_tos; 698 699 r.r_i_ifx = src_if_index; 700 701 /* 702 * XXX NOTE: only first fragment of fragmented TCP, UDP and 703 * ICMP packet will be recorded with proper s_port and d_port. 704 * Following fragments will be recorded simply as IP packet with 705 * ip_proto = ip->ip_p and s_port, d_port set to zero. 706 * I know, it looks like bug. But I don't want to re-implement 707 * ip packet assebmling here. Anyway, (in)famous trafd works this way - 708 * and nobody complains yet :) 709 */ 710 if ((ip->ip_off & htons(IP_OFFMASK)) == 0) 711 switch(r.r_ip_p) { 712 case IPPROTO_TCP: 713 { 714 register struct tcphdr *tcp; 715 716 tcp = (struct tcphdr *)((caddr_t )ip + hlen); 717 r.r_sport = tcp->th_sport; 718 r.r_dport = tcp->th_dport; 719 tcp_flags = tcp->th_flags; 720 break; 721 } 722 case IPPROTO_UDP: 723 r.r_ports = *(uint32_t *)((caddr_t )ip + hlen); 724 break; 725 } 726 727 atomic_fetchadd_32(&priv->info.nfinfo_packets, 1); 728 /* XXX: atomic */ 729 priv->info.nfinfo_bytes += plen; 730 731 /* Find hash slot. */ 732 hsh = &priv->hash[ip_hash(&r)]; 733 734 mtx_lock(&hsh->mtx); 735 736 /* 737 * Go through hash and find our entry. If we encounter an 738 * entry, that should be expired, purge it. We do a reverse 739 * search since most active entries are first, and most 740 * searches are done on most active entries. 741 */ 742 TAILQ_FOREACH_REVERSE_SAFE(fle, &hsh->head, fhead, fle_hash, fle1) { 743 if (bcmp(&r, &fle->f.r, sizeof(struct flow_rec)) == 0) 744 break; 745 if ((INACTIVE(fle) && SMALL(fle)) || AGED(fle)) { 746 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 747 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), fle, NG_QUEUE); 748 atomic_add_32(&priv->info.nfinfo_act_exp, 1); 749 } 750 } 751 752 if (fle) { /* An existent entry. */ 753 754 fle->f.bytes += plen; 755 fle->f.packets ++; 756 fle->f.tcp_flags |= tcp_flags; 757 fle->f.last = time_uptime; 758 759 /* 760 * We have the following reasons to expire flow in active way: 761 * - it hit active timeout 762 * - a TCP connection closed 763 * - it is going to overflow counter 764 */ 765 if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle) || 766 (fle->f.bytes >= (CNTR_MAX - IF_MAXMTU)) ) { 767 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 768 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), fle, NG_QUEUE); 769 atomic_add_32(&priv->info.nfinfo_act_exp, 1); 770 } else { 771 /* 772 * It is the newest, move it to the tail, 773 * if it isn't there already. Next search will 774 * locate it quicker. 775 */ 776 if (fle != TAILQ_LAST(&hsh->head, fhead)) { 777 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 778 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash); 779 } 780 } 781 } else /* A new flow entry. */ 782 error = hash_insert(priv, hsh, &r, plen, tcp_flags); 783 784 mtx_unlock(&hsh->mtx); 785 786 return (error); 787 } 788 789 #ifdef INET6 790 /* Insert IPv6 packet from into flow cache. */ 791 int 792 ng_netflow_flow6_add(priv_p priv, fib_export_p fe, struct ip6_hdr *ip6, caddr_t upper_ptr, uint8_t upper_proto, 793 uint8_t is_frag, unsigned int src_if_index) 794 { 795 register struct flow6_entry *fle6 = NULL, *fle61; 796 struct flow6_hash_entry *hsh6; 797 struct flow6_rec r; 798 int plen; 799 int error = 0; 800 uint8_t tcp_flags = 0; 801 802 /* check version */ 803 if ((ip6->ip6_vfc & IPV6_VERSION_MASK) != IPV6_VERSION) 804 return (EINVAL); 805 806 bzero(&r, sizeof(r)); 807 808 r.src.r_src6 = ip6->ip6_src; 809 r.dst.r_dst6 = ip6->ip6_dst; 810 r.fib = fe->fib; 811 812 /* Assume L4 template by default */ 813 r.flow_type = NETFLOW_V9_FLOW_V6_L4; 814 815 /* save packet length */ 816 plen = ntohs(ip6->ip6_plen) + sizeof(struct ip6_hdr); 817 818 /* XXX: set DSCP/CoS value */ 819 #if 0 820 r.r_tos = ip->ip_tos; 821 #endif 822 if (is_frag == 0) { 823 switch(upper_proto) { 824 case IPPROTO_TCP: 825 { 826 register struct tcphdr *tcp; 827 828 tcp = (struct tcphdr *)upper_ptr; 829 r.r_ports = *(uint32_t *)upper_ptr; 830 tcp_flags = tcp->th_flags; 831 break; 832 } 833 case IPPROTO_UDP: 834 case IPPROTO_SCTP: 835 { 836 r.r_ports = *(uint32_t *)upper_ptr; 837 break; 838 } 839 840 } 841 } 842 843 r.r_ip_p = upper_proto; 844 r.r_i_ifx = src_if_index; 845 846 atomic_fetchadd_32(&priv->info.nfinfo_packets6, 1); 847 /* XXX: atomic */ 848 priv->info.nfinfo_bytes6 += plen; 849 850 /* Find hash slot. */ 851 hsh6 = &priv->hash6[ip6_hash(&r)]; 852 853 mtx_lock(&hsh6->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(fle6, &hsh6->head, f6head, fle6_hash, fle61) { 862 if (fle6->f.version != IP6VERSION) 863 continue; 864 if (bcmp(&r, &fle6->f.r, sizeof(struct flow6_rec)) == 0) 865 break; 866 if ((INACTIVE(fle6) && SMALL(fle6)) || AGED(fle6)) { 867 TAILQ_REMOVE(&hsh6->head, fle6, fle6_hash); 868 expire_flow(priv, priv_to_fib(priv, fle6->f.r.fib), (struct flow_entry *)fle6, NG_QUEUE); 869 atomic_add_32(&priv->info.nfinfo_act_exp, 1); 870 } 871 } 872 873 if (fle6 != NULL) { /* An existent entry. */ 874 875 fle6->f.bytes += plen; 876 fle6->f.packets ++; 877 fle6->f.tcp_flags |= tcp_flags; 878 fle6->f.last = time_uptime; 879 880 /* 881 * We have the following reasons to expire flow in active way: 882 * - it hit active timeout 883 * - a TCP connection closed 884 * - it is going to overflow counter 885 */ 886 if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle6) || 887 (fle6->f.bytes >= (CNTR_MAX - IF_MAXMTU)) ) { 888 TAILQ_REMOVE(&hsh6->head, fle6, fle6_hash); 889 expire_flow(priv, priv_to_fib(priv, fle6->f.r.fib), (struct flow_entry *)fle6, NG_QUEUE); 890 atomic_add_32(&priv->info.nfinfo_act_exp, 1); 891 } else { 892 /* 893 * It is the newest, move it to the tail, 894 * if it isn't there already. Next search will 895 * locate it quicker. 896 */ 897 if (fle6 != TAILQ_LAST(&hsh6->head, f6head)) { 898 TAILQ_REMOVE(&hsh6->head, fle6, fle6_hash); 899 TAILQ_INSERT_TAIL(&hsh6->head, fle6, fle6_hash); 900 } 901 } 902 } else /* A new flow entry. */ 903 error = hash6_insert(priv, hsh6, &r, plen, tcp_flags); 904 905 mtx_unlock(&hsh6->mtx); 906 907 return (error); 908 } 909 #endif 910 911 /* 912 * Return records from cache to userland. 913 * 914 * TODO: matching particular IP should be done in kernel, here. 915 * XXX: IPv6 flows will return random data 916 */ 917 int 918 ng_netflow_flow_show(priv_p priv, uint32_t last, struct ng_mesg *resp) 919 { 920 struct flow_hash_entry *hsh; 921 struct flow_entry *fle; 922 struct ngnf_flows *data; 923 int i; 924 925 data = (struct ngnf_flows *)resp->data; 926 data->last = 0; 927 data->nentries = 0; 928 929 /* Check if this is a first run */ 930 if (last == 0) { 931 hsh = priv->hash; 932 i = 0; 933 } else { 934 if (last > NBUCKETS-1) 935 return (EINVAL); 936 hsh = priv->hash + last; 937 i = last; 938 } 939 940 /* 941 * We will transfer not more than NREC_AT_ONCE. More data 942 * will come in next message. 943 * We send current hash index to userland, and userland should 944 * return it back to us. Then, we will restart with new entry. 945 * 946 * The resulting cache snapshot is inaccurate for the 947 * following reasons: 948 * - we skip locked hash entries 949 * - we bail out, if someone wants our entry 950 * - we skip rest of entry, when hit NREC_AT_ONCE 951 */ 952 for (; i < NBUCKETS; hsh++, i++) { 953 if (mtx_trylock(&hsh->mtx) == 0) 954 continue; 955 956 TAILQ_FOREACH(fle, &hsh->head, fle_hash) { 957 if (hsh->mtx.mtx_lock & MTX_CONTESTED) 958 break; 959 960 bcopy(&fle->f, &(data->entries[data->nentries]), 961 sizeof(fle->f)); 962 data->nentries++; 963 if (data->nentries == NREC_AT_ONCE) { 964 mtx_unlock(&hsh->mtx); 965 if (++i < NBUCKETS) 966 data->last = i; 967 return (0); 968 } 969 } 970 mtx_unlock(&hsh->mtx); 971 } 972 973 return (0); 974 } 975 976 /* We have full datagram in privdata. Send it to export hook. */ 977 static int 978 export_send(priv_p priv, fib_export_p fe, item_p item, int flags) 979 { 980 struct mbuf *m = NGI_M(item); 981 struct netflow_v5_export_dgram *dgram = mtod(m, 982 struct netflow_v5_export_dgram *); 983 struct netflow_v5_header *header = &dgram->header; 984 struct timespec ts; 985 int error = 0; 986 987 /* Fill mbuf header. */ 988 m->m_len = m->m_pkthdr.len = sizeof(struct netflow_v5_record) * 989 header->count + sizeof(struct netflow_v5_header); 990 991 /* Fill export header. */ 992 header->sys_uptime = htonl(MILLIUPTIME(time_uptime)); 993 getnanotime(&ts); 994 header->unix_secs = htonl(ts.tv_sec); 995 header->unix_nsecs = htonl(ts.tv_nsec); 996 header->engine_type = 0; 997 header->engine_id = fe->domain_id; 998 header->pad = 0; 999 header->flow_seq = htonl(atomic_fetchadd_32(&fe->flow_seq, 1000 header->count)); 1001 header->count = htons(header->count); 1002 1003 if (priv->export != NULL) 1004 NG_FWD_ITEM_HOOK_FLAGS(error, item, priv->export, flags); 1005 else 1006 NG_FREE_ITEM(item); 1007 1008 return (error); 1009 } 1010 1011 1012 /* Add export record to dgram. */ 1013 static int 1014 export_add(item_p item, struct flow_entry *fle) 1015 { 1016 struct netflow_v5_export_dgram *dgram = mtod(NGI_M(item), 1017 struct netflow_v5_export_dgram *); 1018 struct netflow_v5_header *header = &dgram->header; 1019 struct netflow_v5_record *rec; 1020 1021 rec = &dgram->r[header->count]; 1022 header->count ++; 1023 1024 KASSERT(header->count <= NETFLOW_V5_MAX_RECORDS, 1025 ("ng_netflow: export too big")); 1026 1027 /* Fill in export record. */ 1028 rec->src_addr = fle->f.r.r_src.s_addr; 1029 rec->dst_addr = fle->f.r.r_dst.s_addr; 1030 rec->next_hop = fle->f.next_hop.s_addr; 1031 rec->i_ifx = htons(fle->f.fle_i_ifx); 1032 rec->o_ifx = htons(fle->f.fle_o_ifx); 1033 rec->packets = htonl(fle->f.packets); 1034 rec->octets = htonl(fle->f.bytes); 1035 rec->first = htonl(MILLIUPTIME(fle->f.first)); 1036 rec->last = htonl(MILLIUPTIME(fle->f.last)); 1037 rec->s_port = fle->f.r.r_sport; 1038 rec->d_port = fle->f.r.r_dport; 1039 rec->flags = fle->f.tcp_flags; 1040 rec->prot = fle->f.r.r_ip_p; 1041 rec->tos = fle->f.r.r_tos; 1042 rec->dst_mask = fle->f.dst_mask; 1043 rec->src_mask = fle->f.src_mask; 1044 rec->pad1 = 0; 1045 rec->pad2 = 0; 1046 1047 /* Not supported fields. */ 1048 rec->src_as = rec->dst_as = 0; 1049 1050 if (header->count == NETFLOW_V5_MAX_RECORDS) 1051 return (1); /* end of datagram */ 1052 else 1053 return (0); 1054 } 1055 1056 /* Periodic flow expiry run. */ 1057 void 1058 ng_netflow_expire(void *arg) 1059 { 1060 struct flow_entry *fle, *fle1; 1061 struct flow_hash_entry *hsh; 1062 #ifdef INET6 1063 struct flow6_entry *fle6, *fle61; 1064 struct flow6_hash_entry *hsh6; 1065 #endif 1066 priv_p priv = (priv_p )arg; 1067 uint32_t used; 1068 int i; 1069 1070 /* 1071 * Going through all the cache. 1072 */ 1073 for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++) { 1074 /* 1075 * Skip entries, that are already being worked on. 1076 */ 1077 if (mtx_trylock(&hsh->mtx) == 0) 1078 continue; 1079 1080 used = atomic_load_acq_32(&priv->info.nfinfo_used); 1081 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) { 1082 /* 1083 * Interrupt thread wants this entry! 1084 * Quick! Quick! Bail out! 1085 */ 1086 if (hsh->mtx.mtx_lock & MTX_CONTESTED) 1087 break; 1088 1089 /* 1090 * Don't expire aggressively while hash collision 1091 * ratio is predicted small. 1092 */ 1093 if (used <= (NBUCKETS*2) && !INACTIVE(fle)) 1094 break; 1095 1096 if ((INACTIVE(fle) && (SMALL(fle) || 1097 (used > (NBUCKETS*2)))) || AGED(fle)) { 1098 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 1099 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), fle, NG_NOFLAGS); 1100 used--; 1101 atomic_add_32(&priv->info.nfinfo_inact_exp, 1); 1102 } 1103 } 1104 mtx_unlock(&hsh->mtx); 1105 } 1106 1107 #ifdef INET6 1108 for (hsh6 = priv->hash6, i = 0; i < NBUCKETS; hsh6++, i++) { 1109 /* 1110 * Skip entries, that are already being worked on. 1111 */ 1112 if (mtx_trylock(&hsh6->mtx) == 0) 1113 continue; 1114 1115 used = atomic_load_acq_32(&priv->info.nfinfo_used6); 1116 TAILQ_FOREACH_SAFE(fle6, &hsh6->head, fle6_hash, fle61) { 1117 /* 1118 * Interrupt thread wants this entry! 1119 * Quick! Quick! Bail out! 1120 */ 1121 if (hsh6->mtx.mtx_lock & MTX_CONTESTED) 1122 break; 1123 1124 /* 1125 * Don't expire aggressively while hash collision 1126 * ratio is predicted small. 1127 */ 1128 if (used <= (NBUCKETS*2) && !INACTIVE(fle6)) 1129 break; 1130 1131 if ((INACTIVE(fle6) && (SMALL(fle6) || 1132 (used > (NBUCKETS*2)))) || AGED(fle6)) { 1133 TAILQ_REMOVE(&hsh6->head, fle6, fle6_hash); 1134 expire_flow(priv, priv_to_fib(priv, fle6->f.r.fib), (struct flow_entry *)fle6, NG_NOFLAGS); 1135 used--; 1136 atomic_add_32(&priv->info.nfinfo_inact_exp, 1); 1137 } 1138 } 1139 mtx_unlock(&hsh6->mtx); 1140 } 1141 #endif 1142 1143 /* Schedule next expire. */ 1144 callout_reset(&priv->exp_callout, (1*hz), &ng_netflow_expire, 1145 (void *)priv); 1146 } 1147