1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2007-2009 5 * Swinburne University of Technology, Melbourne, Australia. 6 * Copyright (c) 2009-2010, The FreeBSD Foundation 7 * All rights reserved. 8 * 9 * Portions of this software were developed at the Centre for Advanced 10 * Internet Architectures, Swinburne University of Technology, Melbourne, 11 * Australia by Lawrence Stewart under sponsorship from the FreeBSD Foundation. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 */ 34 35 /****************************************************** 36 * Statistical Information For TCP Research (SIFTR) 37 * 38 * A FreeBSD kernel module that adds very basic intrumentation to the 39 * TCP stack, allowing internal stats to be recorded to a log file 40 * for experimental, debugging and performance analysis purposes. 41 * 42 * SIFTR was first released in 2007 by James Healy and Lawrence Stewart whilst 43 * working on the NewTCP research project at Swinburne University of 44 * Technology's Centre for Advanced Internet Architectures, Melbourne, 45 * Australia, which was made possible in part by a grant from the Cisco 46 * University Research Program Fund at Community Foundation Silicon Valley. 47 * More details are available at: 48 * http://caia.swin.edu.au/urp/newtcp/ 49 * 50 * Work on SIFTR v1.2.x was sponsored by the FreeBSD Foundation as part of 51 * the "Enhancing the FreeBSD TCP Implementation" project 2008-2009. 52 * More details are available at: 53 * http://www.freebsdfoundation.org/ 54 * http://caia.swin.edu.au/freebsd/etcp09/ 55 * 56 * Lawrence Stewart is the current maintainer, and all contact regarding 57 * SIFTR should be directed to him via email: lastewart@swin.edu.au 58 * 59 * Initial release date: June 2007 60 * Most recent update: September 2010 61 ******************************************************/ 62 63 #include <sys/cdefs.h> 64 __FBSDID("$FreeBSD$"); 65 66 #include <sys/param.h> 67 #include <sys/alq.h> 68 #include <sys/errno.h> 69 #include <sys/eventhandler.h> 70 #include <sys/hash.h> 71 #include <sys/kernel.h> 72 #include <sys/kthread.h> 73 #include <sys/lock.h> 74 #include <sys/mbuf.h> 75 #include <sys/module.h> 76 #include <sys/mutex.h> 77 #include <sys/pcpu.h> 78 #include <sys/proc.h> 79 #include <sys/sbuf.h> 80 #include <sys/sdt.h> 81 #include <sys/smp.h> 82 #include <sys/socket.h> 83 #include <sys/socketvar.h> 84 #include <sys/sysctl.h> 85 #include <sys/unistd.h> 86 87 #include <net/if.h> 88 #include <net/if_var.h> 89 #include <net/pfil.h> 90 91 #include <netinet/in.h> 92 #include <netinet/in_kdtrace.h> 93 #include <netinet/in_pcb.h> 94 #include <netinet/in_systm.h> 95 #include <netinet/in_var.h> 96 #include <netinet/ip.h> 97 #include <netinet/ip_var.h> 98 #include <netinet/tcp_var.h> 99 100 #ifdef SIFTR_IPV6 101 #include <netinet/ip6.h> 102 #include <netinet6/ip6_var.h> 103 #include <netinet6/in6_pcb.h> 104 #endif /* SIFTR_IPV6 */ 105 106 #include <machine/in_cksum.h> 107 108 /* 109 * Three digit version number refers to X.Y.Z where: 110 * X is the major version number 111 * Y is bumped to mark backwards incompatible changes 112 * Z is bumped to mark backwards compatible changes 113 */ 114 #define V_MAJOR 1 115 #define V_BACKBREAK 3 116 #define V_BACKCOMPAT 0 117 #define MODVERSION __CONCAT(V_MAJOR, __CONCAT(V_BACKBREAK, V_BACKCOMPAT)) 118 #define MODVERSION_STR __XSTRING(V_MAJOR) "." __XSTRING(V_BACKBREAK) "." \ 119 __XSTRING(V_BACKCOMPAT) 120 121 #define HOOK 0 122 #define UNHOOK 1 123 #define SIFTR_EXPECTED_MAX_TCP_FLOWS 65536 124 #define SYS_NAME "FreeBSD" 125 #define PACKET_TAG_SIFTR 100 126 #define PACKET_COOKIE_SIFTR 21749576 127 #define SIFTR_LOG_FILE_MODE 0644 128 #define SIFTR_DISABLE 0 129 #define SIFTR_ENABLE 1 130 131 /* 132 * Hard upper limit on the length of log messages. Bump this up if you add new 133 * data fields such that the line length could exceed the below value. 134 */ 135 #define MAX_LOG_MSG_LEN 300 136 /* XXX: Make this a sysctl tunable. */ 137 #define SIFTR_ALQ_BUFLEN (1000*MAX_LOG_MSG_LEN) 138 139 /* 140 * 1 byte for IP version 141 * IPv4: src/dst IP (4+4) + src/dst port (2+2) = 12 bytes 142 * IPv6: src/dst IP (16+16) + src/dst port (2+2) = 36 bytes 143 */ 144 #ifdef SIFTR_IPV6 145 #define FLOW_KEY_LEN 37 146 #else 147 #define FLOW_KEY_LEN 13 148 #endif 149 150 #ifdef SIFTR_IPV6 151 #define SIFTR_IPMODE 6 152 #else 153 #define SIFTR_IPMODE 4 154 #endif 155 156 /* useful macros */ 157 #define UPPER_SHORT(X) (((X) & 0xFFFF0000) >> 16) 158 #define LOWER_SHORT(X) ((X) & 0x0000FFFF) 159 160 #define FIRST_OCTET(X) (((X) & 0xFF000000) >> 24) 161 #define SECOND_OCTET(X) (((X) & 0x00FF0000) >> 16) 162 #define THIRD_OCTET(X) (((X) & 0x0000FF00) >> 8) 163 #define FOURTH_OCTET(X) ((X) & 0x000000FF) 164 165 static MALLOC_DEFINE(M_SIFTR, "siftr", "dynamic memory used by SIFTR"); 166 static MALLOC_DEFINE(M_SIFTR_PKTNODE, "siftr_pktnode", 167 "SIFTR pkt_node struct"); 168 static MALLOC_DEFINE(M_SIFTR_HASHNODE, "siftr_hashnode", 169 "SIFTR flow_hash_node struct"); 170 171 /* Used as links in the pkt manager queue. */ 172 struct pkt_node { 173 /* Timestamp of pkt as noted in the pfil hook. */ 174 struct timeval tval; 175 /* Direction pkt is travelling. */ 176 enum { 177 DIR_IN = 0, 178 DIR_OUT = 1, 179 } direction; 180 /* IP version pkt_node relates to; either INP_IPV4 or INP_IPV6. */ 181 uint8_t ipver; 182 /* Local/foreign IP address. */ 183 #ifdef SIFTR_IPV6 184 uint32_t ip_laddr[4]; 185 uint32_t ip_faddr[4]; 186 #else 187 uint8_t ip_laddr[4]; 188 uint8_t ip_faddr[4]; 189 #endif 190 /* Local TCP port. */ 191 uint16_t tcp_localport; 192 /* Foreign TCP port. */ 193 uint16_t tcp_foreignport; 194 /* Congestion Window (bytes). */ 195 uint32_t snd_cwnd; 196 /* Sending Window (bytes). */ 197 uint32_t snd_wnd; 198 /* Receive Window (bytes). */ 199 uint32_t rcv_wnd; 200 /* More tcpcb flags storage */ 201 uint32_t t_flags2; 202 /* Slow Start Threshold (bytes). */ 203 uint32_t snd_ssthresh; 204 /* Current state of the TCP FSM. */ 205 int conn_state; 206 /* Max Segment Size (bytes). */ 207 u_int max_seg_size; 208 /* Smoothed RTT (usecs). */ 209 uint32_t srtt; 210 /* Is SACK enabled? */ 211 u_char sack_enabled; 212 /* Window scaling for snd window. */ 213 u_char snd_scale; 214 /* Window scaling for recv window. */ 215 u_char rcv_scale; 216 /* TCP control block flags. */ 217 u_int flags; 218 /* Retransmission timeout (usec). */ 219 uint32_t rto; 220 /* Size of the TCP send buffer in bytes. */ 221 u_int snd_buf_hiwater; 222 /* Current num bytes in the send socket buffer. */ 223 u_int snd_buf_cc; 224 /* Size of the TCP receive buffer in bytes. */ 225 u_int rcv_buf_hiwater; 226 /* Current num bytes in the receive socket buffer. */ 227 u_int rcv_buf_cc; 228 /* Number of bytes inflight that we are waiting on ACKs for. */ 229 u_int sent_inflight_bytes; 230 /* Number of segments currently in the reassembly queue. */ 231 int t_segqlen; 232 /* Flowid for the connection. */ 233 u_int flowid; 234 /* Flow type for the connection. */ 235 u_int flowtype; 236 /* Link to next pkt_node in the list. */ 237 STAILQ_ENTRY(pkt_node) nodes; 238 }; 239 240 struct flow_hash_node 241 { 242 uint16_t counter; 243 uint8_t key[FLOW_KEY_LEN]; 244 LIST_ENTRY(flow_hash_node) nodes; 245 }; 246 247 struct siftr_stats 248 { 249 /* # TCP pkts seen by the SIFTR PFIL hooks, including any skipped. */ 250 uint64_t n_in; 251 uint64_t n_out; 252 /* # pkts skipped due to failed malloc calls. */ 253 uint32_t nskip_in_malloc; 254 uint32_t nskip_out_malloc; 255 /* # pkts skipped due to failed inpcb lookups. */ 256 uint32_t nskip_in_inpcb; 257 uint32_t nskip_out_inpcb; 258 /* # pkts skipped due to failed tcpcb lookups. */ 259 uint32_t nskip_in_tcpcb; 260 uint32_t nskip_out_tcpcb; 261 /* # pkts skipped due to stack reinjection. */ 262 uint32_t nskip_in_dejavu; 263 uint32_t nskip_out_dejavu; 264 }; 265 266 DPCPU_DEFINE_STATIC(struct siftr_stats, ss); 267 268 static volatile unsigned int siftr_exit_pkt_manager_thread = 0; 269 static unsigned int siftr_enabled = 0; 270 static unsigned int siftr_pkts_per_log = 1; 271 static uint16_t siftr_port_filter = 0; 272 /* static unsigned int siftr_binary_log = 0; */ 273 static char siftr_logfile[PATH_MAX] = "/var/log/siftr.log"; 274 static char siftr_logfile_shadow[PATH_MAX] = "/var/log/siftr.log"; 275 static u_long siftr_hashmask; 276 STAILQ_HEAD(pkthead, pkt_node) pkt_queue = STAILQ_HEAD_INITIALIZER(pkt_queue); 277 LIST_HEAD(listhead, flow_hash_node) *counter_hash; 278 static int wait_for_pkt; 279 static struct alq *siftr_alq = NULL; 280 static struct mtx siftr_pkt_queue_mtx; 281 static struct mtx siftr_pkt_mgr_mtx; 282 static struct thread *siftr_pkt_manager_thr = NULL; 283 static char direction[2] = {'i','o'}; 284 285 /* Required function prototypes. */ 286 static int siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS); 287 static int siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS); 288 289 /* Declare the net.inet.siftr sysctl tree and populate it. */ 290 291 SYSCTL_DECL(_net_inet_siftr); 292 293 SYSCTL_NODE(_net_inet, OID_AUTO, siftr, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 294 "siftr related settings"); 295 296 SYSCTL_PROC(_net_inet_siftr, OID_AUTO, enabled, 297 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 298 &siftr_enabled, 0, &siftr_sysctl_enabled_handler, "IU", 299 "switch siftr module operations on/off"); 300 301 SYSCTL_PROC(_net_inet_siftr, OID_AUTO, logfile, 302 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &siftr_logfile_shadow, 303 sizeof(siftr_logfile_shadow), &siftr_sysctl_logfile_name_handler, "A", 304 "file to save siftr log messages to"); 305 306 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, ppl, CTLFLAG_RW, 307 &siftr_pkts_per_log, 1, 308 "number of packets between generating a log message"); 309 310 SYSCTL_U16(_net_inet_siftr, OID_AUTO, port_filter, CTLFLAG_RW, 311 &siftr_port_filter, 0, 312 "enable packet filter on a TCP port"); 313 314 /* XXX: TODO 315 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, binary, CTLFLAG_RW, 316 &siftr_binary_log, 0, 317 "write log files in binary instead of ascii"); 318 */ 319 320 /* Begin functions. */ 321 322 static void 323 siftr_process_pkt(struct pkt_node * pkt_node) 324 { 325 struct flow_hash_node *hash_node; 326 struct listhead *counter_list; 327 struct siftr_stats *ss; 328 struct ale *log_buf; 329 uint8_t key[FLOW_KEY_LEN]; 330 uint8_t found_match, key_offset; 331 332 hash_node = NULL; 333 ss = DPCPU_PTR(ss); 334 found_match = 0; 335 key_offset = 1; 336 337 /* 338 * Create the key that will be used to create a hash index 339 * into our hash table. Our key consists of: 340 * ipversion, localip, localport, foreignip, foreignport 341 */ 342 key[0] = pkt_node->ipver; 343 memcpy(key + key_offset, &pkt_node->ip_laddr, 344 sizeof(pkt_node->ip_laddr)); 345 key_offset += sizeof(pkt_node->ip_laddr); 346 memcpy(key + key_offset, &pkt_node->tcp_localport, 347 sizeof(pkt_node->tcp_localport)); 348 key_offset += sizeof(pkt_node->tcp_localport); 349 memcpy(key + key_offset, &pkt_node->ip_faddr, 350 sizeof(pkt_node->ip_faddr)); 351 key_offset += sizeof(pkt_node->ip_faddr); 352 memcpy(key + key_offset, &pkt_node->tcp_foreignport, 353 sizeof(pkt_node->tcp_foreignport)); 354 355 counter_list = counter_hash + 356 (hash32_buf(key, sizeof(key), 0) & siftr_hashmask); 357 358 /* 359 * If the list is not empty i.e. the hash index has 360 * been used by another flow previously. 361 */ 362 if (LIST_FIRST(counter_list) != NULL) { 363 /* 364 * Loop through the hash nodes in the list. 365 * There should normally only be 1 hash node in the list, 366 * except if there have been collisions at the hash index 367 * computed by hash32_buf(). 368 */ 369 LIST_FOREACH(hash_node, counter_list, nodes) { 370 /* 371 * Check if the key for the pkt we are currently 372 * processing is the same as the key stored in the 373 * hash node we are currently processing. 374 * If they are the same, then we've found the 375 * hash node that stores the counter for the flow 376 * the pkt belongs to. 377 */ 378 if (memcmp(hash_node->key, key, sizeof(key)) == 0) { 379 found_match = 1; 380 break; 381 } 382 } 383 } 384 385 /* If this flow hash hasn't been seen before or we have a collision. */ 386 if (hash_node == NULL || !found_match) { 387 /* Create a new hash node to store the flow's counter. */ 388 hash_node = malloc(sizeof(struct flow_hash_node), 389 M_SIFTR_HASHNODE, M_WAITOK); 390 391 if (hash_node != NULL) { 392 /* Initialise our new hash node list entry. */ 393 hash_node->counter = 0; 394 memcpy(hash_node->key, key, sizeof(key)); 395 LIST_INSERT_HEAD(counter_list, hash_node, nodes); 396 } else { 397 /* Malloc failed. */ 398 if (pkt_node->direction == DIR_IN) 399 ss->nskip_in_malloc++; 400 else 401 ss->nskip_out_malloc++; 402 403 return; 404 } 405 } else if (siftr_pkts_per_log > 1) { 406 /* 407 * Taking the remainder of the counter divided 408 * by the current value of siftr_pkts_per_log 409 * and storing that in counter provides a neat 410 * way to modulate the frequency of log 411 * messages being written to the log file. 412 */ 413 hash_node->counter = (hash_node->counter + 1) % 414 siftr_pkts_per_log; 415 416 /* 417 * If we have not seen enough packets since the last time 418 * we wrote a log message for this connection, return. 419 */ 420 if (hash_node->counter > 0) 421 return; 422 } 423 424 log_buf = alq_getn(siftr_alq, MAX_LOG_MSG_LEN, ALQ_WAITOK); 425 426 if (log_buf == NULL) 427 return; /* Should only happen if the ALQ is shutting down. */ 428 429 #ifdef SIFTR_IPV6 430 pkt_node->ip_laddr[3] = ntohl(pkt_node->ip_laddr[3]); 431 pkt_node->ip_faddr[3] = ntohl(pkt_node->ip_faddr[3]); 432 433 if (pkt_node->ipver == INP_IPV6) { /* IPv6 packet */ 434 pkt_node->ip_laddr[0] = ntohl(pkt_node->ip_laddr[0]); 435 pkt_node->ip_laddr[1] = ntohl(pkt_node->ip_laddr[1]); 436 pkt_node->ip_laddr[2] = ntohl(pkt_node->ip_laddr[2]); 437 pkt_node->ip_faddr[0] = ntohl(pkt_node->ip_faddr[0]); 438 pkt_node->ip_faddr[1] = ntohl(pkt_node->ip_faddr[1]); 439 pkt_node->ip_faddr[2] = ntohl(pkt_node->ip_faddr[2]); 440 441 /* Construct an IPv6 log message. */ 442 log_buf->ae_bytesused = snprintf(log_buf->ae_data, 443 MAX_LOG_MSG_LEN, 444 "%c,%zd.%06ld,%x:%x:%x:%x:%x:%x:%x:%x,%u,%x:%x:%x:" 445 "%x:%x:%x:%x:%x,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u," 446 "%u,%u,%u,%u,%u,%u,%u,%u,%u,%u\n", 447 direction[pkt_node->direction], 448 pkt_node->tval.tv_sec, 449 pkt_node->tval.tv_usec, 450 UPPER_SHORT(pkt_node->ip_laddr[0]), 451 LOWER_SHORT(pkt_node->ip_laddr[0]), 452 UPPER_SHORT(pkt_node->ip_laddr[1]), 453 LOWER_SHORT(pkt_node->ip_laddr[1]), 454 UPPER_SHORT(pkt_node->ip_laddr[2]), 455 LOWER_SHORT(pkt_node->ip_laddr[2]), 456 UPPER_SHORT(pkt_node->ip_laddr[3]), 457 LOWER_SHORT(pkt_node->ip_laddr[3]), 458 ntohs(pkt_node->tcp_localport), 459 UPPER_SHORT(pkt_node->ip_faddr[0]), 460 LOWER_SHORT(pkt_node->ip_faddr[0]), 461 UPPER_SHORT(pkt_node->ip_faddr[1]), 462 LOWER_SHORT(pkt_node->ip_faddr[1]), 463 UPPER_SHORT(pkt_node->ip_faddr[2]), 464 LOWER_SHORT(pkt_node->ip_faddr[2]), 465 UPPER_SHORT(pkt_node->ip_faddr[3]), 466 LOWER_SHORT(pkt_node->ip_faddr[3]), 467 ntohs(pkt_node->tcp_foreignport), 468 pkt_node->snd_ssthresh, 469 pkt_node->snd_cwnd, 470 pkt_node->t_flags2, 471 pkt_node->snd_wnd, 472 pkt_node->rcv_wnd, 473 pkt_node->snd_scale, 474 pkt_node->rcv_scale, 475 pkt_node->conn_state, 476 pkt_node->max_seg_size, 477 pkt_node->srtt, 478 pkt_node->sack_enabled, 479 pkt_node->flags, 480 pkt_node->rto, 481 pkt_node->snd_buf_hiwater, 482 pkt_node->snd_buf_cc, 483 pkt_node->rcv_buf_hiwater, 484 pkt_node->rcv_buf_cc, 485 pkt_node->sent_inflight_bytes, 486 pkt_node->t_segqlen, 487 pkt_node->flowid, 488 pkt_node->flowtype); 489 } else { /* IPv4 packet */ 490 pkt_node->ip_laddr[0] = FIRST_OCTET(pkt_node->ip_laddr[3]); 491 pkt_node->ip_laddr[1] = SECOND_OCTET(pkt_node->ip_laddr[3]); 492 pkt_node->ip_laddr[2] = THIRD_OCTET(pkt_node->ip_laddr[3]); 493 pkt_node->ip_laddr[3] = FOURTH_OCTET(pkt_node->ip_laddr[3]); 494 pkt_node->ip_faddr[0] = FIRST_OCTET(pkt_node->ip_faddr[3]); 495 pkt_node->ip_faddr[1] = SECOND_OCTET(pkt_node->ip_faddr[3]); 496 pkt_node->ip_faddr[2] = THIRD_OCTET(pkt_node->ip_faddr[3]); 497 pkt_node->ip_faddr[3] = FOURTH_OCTET(pkt_node->ip_faddr[3]); 498 #endif /* SIFTR_IPV6 */ 499 500 /* Construct an IPv4 log message. */ 501 log_buf->ae_bytesused = snprintf(log_buf->ae_data, 502 MAX_LOG_MSG_LEN, 503 "%c,%jd.%06ld,%u.%u.%u.%u,%u,%u.%u.%u.%u,%u,%u,%u," 504 "%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u\n", 505 direction[pkt_node->direction], 506 (intmax_t)pkt_node->tval.tv_sec, 507 pkt_node->tval.tv_usec, 508 pkt_node->ip_laddr[0], 509 pkt_node->ip_laddr[1], 510 pkt_node->ip_laddr[2], 511 pkt_node->ip_laddr[3], 512 ntohs(pkt_node->tcp_localport), 513 pkt_node->ip_faddr[0], 514 pkt_node->ip_faddr[1], 515 pkt_node->ip_faddr[2], 516 pkt_node->ip_faddr[3], 517 ntohs(pkt_node->tcp_foreignport), 518 pkt_node->snd_ssthresh, 519 pkt_node->snd_cwnd, 520 pkt_node->t_flags2, 521 pkt_node->snd_wnd, 522 pkt_node->rcv_wnd, 523 pkt_node->snd_scale, 524 pkt_node->rcv_scale, 525 pkt_node->conn_state, 526 pkt_node->max_seg_size, 527 pkt_node->srtt, 528 pkt_node->sack_enabled, 529 pkt_node->flags, 530 pkt_node->rto, 531 pkt_node->snd_buf_hiwater, 532 pkt_node->snd_buf_cc, 533 pkt_node->rcv_buf_hiwater, 534 pkt_node->rcv_buf_cc, 535 pkt_node->sent_inflight_bytes, 536 pkt_node->t_segqlen, 537 pkt_node->flowid, 538 pkt_node->flowtype); 539 #ifdef SIFTR_IPV6 540 } 541 #endif 542 543 alq_post_flags(siftr_alq, log_buf, 0); 544 } 545 546 static void 547 siftr_pkt_manager_thread(void *arg) 548 { 549 STAILQ_HEAD(pkthead, pkt_node) tmp_pkt_queue = 550 STAILQ_HEAD_INITIALIZER(tmp_pkt_queue); 551 struct pkt_node *pkt_node, *pkt_node_temp; 552 uint8_t draining; 553 554 draining = 2; 555 556 mtx_lock(&siftr_pkt_mgr_mtx); 557 558 /* draining == 0 when queue has been flushed and it's safe to exit. */ 559 while (draining) { 560 /* 561 * Sleep until we are signalled to wake because thread has 562 * been told to exit or until 1 tick has passed. 563 */ 564 mtx_sleep(&wait_for_pkt, &siftr_pkt_mgr_mtx, PWAIT, "pktwait", 565 1); 566 567 /* Gain exclusive access to the pkt_node queue. */ 568 mtx_lock(&siftr_pkt_queue_mtx); 569 570 /* 571 * Move pkt_queue to tmp_pkt_queue, which leaves 572 * pkt_queue empty and ready to receive more pkt_nodes. 573 */ 574 STAILQ_CONCAT(&tmp_pkt_queue, &pkt_queue); 575 576 /* 577 * We've finished making changes to the list. Unlock it 578 * so the pfil hooks can continue queuing pkt_nodes. 579 */ 580 mtx_unlock(&siftr_pkt_queue_mtx); 581 582 /* 583 * We can't hold a mutex whilst calling siftr_process_pkt 584 * because ALQ might sleep waiting for buffer space. 585 */ 586 mtx_unlock(&siftr_pkt_mgr_mtx); 587 588 /* Flush all pkt_nodes to the log file. */ 589 STAILQ_FOREACH_SAFE(pkt_node, &tmp_pkt_queue, nodes, 590 pkt_node_temp) { 591 siftr_process_pkt(pkt_node); 592 STAILQ_REMOVE_HEAD(&tmp_pkt_queue, nodes); 593 free(pkt_node, M_SIFTR_PKTNODE); 594 } 595 596 KASSERT(STAILQ_EMPTY(&tmp_pkt_queue), 597 ("SIFTR tmp_pkt_queue not empty after flush")); 598 599 mtx_lock(&siftr_pkt_mgr_mtx); 600 601 /* 602 * If siftr_exit_pkt_manager_thread gets set during the window 603 * where we are draining the tmp_pkt_queue above, there might 604 * still be pkts in pkt_queue that need to be drained. 605 * Allow one further iteration to occur after 606 * siftr_exit_pkt_manager_thread has been set to ensure 607 * pkt_queue is completely empty before we kill the thread. 608 * 609 * siftr_exit_pkt_manager_thread is set only after the pfil 610 * hooks have been removed, so only 1 extra iteration 611 * is needed to drain the queue. 612 */ 613 if (siftr_exit_pkt_manager_thread) 614 draining--; 615 } 616 617 mtx_unlock(&siftr_pkt_mgr_mtx); 618 619 /* Calls wakeup on this thread's struct thread ptr. */ 620 kthread_exit(); 621 } 622 623 /* 624 * Check if a given mbuf has the SIFTR mbuf tag. If it does, log the fact that 625 * it's a reinjected packet and return. If it doesn't, tag the mbuf and return. 626 * Return value >0 means the caller should skip processing this mbuf. 627 */ 628 static inline int 629 siftr_chkreinject(struct mbuf *m, int dir, struct siftr_stats *ss) 630 { 631 if (m_tag_locate(m, PACKET_COOKIE_SIFTR, PACKET_TAG_SIFTR, NULL) 632 != NULL) { 633 if (dir == PFIL_IN) 634 ss->nskip_in_dejavu++; 635 else 636 ss->nskip_out_dejavu++; 637 638 return (1); 639 } else { 640 struct m_tag *tag = m_tag_alloc(PACKET_COOKIE_SIFTR, 641 PACKET_TAG_SIFTR, 0, M_NOWAIT); 642 if (tag == NULL) { 643 if (dir == PFIL_IN) 644 ss->nskip_in_malloc++; 645 else 646 ss->nskip_out_malloc++; 647 648 return (1); 649 } 650 651 m_tag_prepend(m, tag); 652 } 653 654 return (0); 655 } 656 657 /* 658 * Look up an inpcb for a packet. Return the inpcb pointer if found, or NULL 659 * otherwise. 660 */ 661 static inline struct inpcb * 662 siftr_findinpcb(int ipver, struct ip *ip, struct mbuf *m, uint16_t sport, 663 uint16_t dport, int dir, struct siftr_stats *ss) 664 { 665 struct inpcb *inp; 666 667 /* We need the tcbinfo lock. */ 668 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 669 670 if (dir == PFIL_IN) 671 inp = (ipver == INP_IPV4 ? 672 in_pcblookup(&V_tcbinfo, ip->ip_src, sport, ip->ip_dst, 673 dport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif) 674 : 675 #ifdef SIFTR_IPV6 676 in6_pcblookup(&V_tcbinfo, 677 &((struct ip6_hdr *)ip)->ip6_src, sport, 678 &((struct ip6_hdr *)ip)->ip6_dst, dport, INPLOOKUP_RLOCKPCB, 679 m->m_pkthdr.rcvif) 680 #else 681 NULL 682 #endif 683 ); 684 685 else 686 inp = (ipver == INP_IPV4 ? 687 in_pcblookup(&V_tcbinfo, ip->ip_dst, dport, ip->ip_src, 688 sport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif) 689 : 690 #ifdef SIFTR_IPV6 691 in6_pcblookup(&V_tcbinfo, 692 &((struct ip6_hdr *)ip)->ip6_dst, dport, 693 &((struct ip6_hdr *)ip)->ip6_src, sport, INPLOOKUP_RLOCKPCB, 694 m->m_pkthdr.rcvif) 695 #else 696 NULL 697 #endif 698 ); 699 700 /* If we can't find the inpcb, bail. */ 701 if (inp == NULL) { 702 if (dir == PFIL_IN) 703 ss->nskip_in_inpcb++; 704 else 705 ss->nskip_out_inpcb++; 706 } 707 708 return (inp); 709 } 710 711 static inline void 712 siftr_siftdata(struct pkt_node *pn, struct inpcb *inp, struct tcpcb *tp, 713 int ipver, int dir, int inp_locally_locked) 714 { 715 #ifdef SIFTR_IPV6 716 if (ipver == INP_IPV4) { 717 pn->ip_laddr[3] = inp->inp_laddr.s_addr; 718 pn->ip_faddr[3] = inp->inp_faddr.s_addr; 719 #else 720 *((uint32_t *)pn->ip_laddr) = inp->inp_laddr.s_addr; 721 *((uint32_t *)pn->ip_faddr) = inp->inp_faddr.s_addr; 722 #endif 723 #ifdef SIFTR_IPV6 724 } else { 725 pn->ip_laddr[0] = inp->in6p_laddr.s6_addr32[0]; 726 pn->ip_laddr[1] = inp->in6p_laddr.s6_addr32[1]; 727 pn->ip_laddr[2] = inp->in6p_laddr.s6_addr32[2]; 728 pn->ip_laddr[3] = inp->in6p_laddr.s6_addr32[3]; 729 pn->ip_faddr[0] = inp->in6p_faddr.s6_addr32[0]; 730 pn->ip_faddr[1] = inp->in6p_faddr.s6_addr32[1]; 731 pn->ip_faddr[2] = inp->in6p_faddr.s6_addr32[2]; 732 pn->ip_faddr[3] = inp->in6p_faddr.s6_addr32[3]; 733 } 734 #endif 735 pn->tcp_localport = inp->inp_lport; 736 pn->tcp_foreignport = inp->inp_fport; 737 pn->snd_cwnd = tp->snd_cwnd; 738 pn->snd_wnd = tp->snd_wnd; 739 pn->rcv_wnd = tp->rcv_wnd; 740 pn->t_flags2 = tp->t_flags2; 741 pn->snd_ssthresh = tp->snd_ssthresh; 742 pn->snd_scale = tp->snd_scale; 743 pn->rcv_scale = tp->rcv_scale; 744 pn->conn_state = tp->t_state; 745 pn->max_seg_size = tp->t_maxseg; 746 pn->srtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT; 747 pn->sack_enabled = (tp->t_flags & TF_SACK_PERMIT) != 0; 748 pn->flags = tp->t_flags; 749 pn->rto = tp->t_rxtcur * tick; 750 pn->snd_buf_hiwater = inp->inp_socket->so_snd.sb_hiwat; 751 pn->snd_buf_cc = sbused(&inp->inp_socket->so_snd); 752 pn->rcv_buf_hiwater = inp->inp_socket->so_rcv.sb_hiwat; 753 pn->rcv_buf_cc = sbused(&inp->inp_socket->so_rcv); 754 pn->sent_inflight_bytes = tp->snd_max - tp->snd_una; 755 pn->t_segqlen = tp->t_segqlen; 756 pn->flowid = inp->inp_flowid; 757 pn->flowtype = inp->inp_flowtype; 758 759 /* We've finished accessing the tcb so release the lock. */ 760 if (inp_locally_locked) 761 INP_RUNLOCK(inp); 762 763 pn->ipver = ipver; 764 pn->direction = (dir == PFIL_IN ? DIR_IN : DIR_OUT); 765 766 /* 767 * Significantly more accurate than using getmicrotime(), but slower! 768 * Gives true microsecond resolution at the expense of a hit to 769 * maximum pps throughput processing when SIFTR is loaded and enabled. 770 */ 771 microtime(&pn->tval); 772 TCP_PROBE1(siftr, &pn); 773 774 } 775 776 /* 777 * pfil hook that is called for each IPv4 packet making its way through the 778 * stack in either direction. 779 * The pfil subsystem holds a non-sleepable mutex somewhere when 780 * calling our hook function, so we can't sleep at all. 781 * It's very important to use the M_NOWAIT flag with all function calls 782 * that support it so that they won't sleep, otherwise you get a panic. 783 */ 784 static pfil_return_t 785 siftr_chkpkt(struct mbuf **m, struct ifnet *ifp, int flags, 786 void *ruleset __unused, struct inpcb *inp) 787 { 788 struct pkt_node *pn; 789 struct ip *ip; 790 struct tcphdr *th; 791 struct tcpcb *tp; 792 struct siftr_stats *ss; 793 unsigned int ip_hl; 794 int inp_locally_locked, dir; 795 796 inp_locally_locked = 0; 797 dir = PFIL_DIR(flags); 798 ss = DPCPU_PTR(ss); 799 800 /* 801 * m_pullup is not required here because ip_{input|output} 802 * already do the heavy lifting for us. 803 */ 804 805 ip = mtod(*m, struct ip *); 806 807 /* Only continue processing if the packet is TCP. */ 808 if (ip->ip_p != IPPROTO_TCP) 809 goto ret; 810 811 /* 812 * Create a tcphdr struct starting at the correct offset 813 * in the IP packet. ip->ip_hl gives the ip header length 814 * in 4-byte words, so multiply it to get the size in bytes. 815 */ 816 ip_hl = (ip->ip_hl << 2); 817 th = (struct tcphdr *)((caddr_t)ip + ip_hl); 818 819 /* 820 * Only pkts selected by the tcp port filter 821 * can be inserted into the pkt_queue 822 */ 823 if ((siftr_port_filter != 0) && 824 (siftr_port_filter != ntohs(th->th_sport)) && 825 (siftr_port_filter != ntohs(th->th_dport))) { 826 goto ret; 827 } 828 829 /* 830 * If a kernel subsystem reinjects packets into the stack, our pfil 831 * hook will be called multiple times for the same packet. 832 * Make sure we only process unique packets. 833 */ 834 if (siftr_chkreinject(*m, dir, ss)) 835 goto ret; 836 837 if (dir == PFIL_IN) 838 ss->n_in++; 839 else 840 ss->n_out++; 841 842 /* 843 * If the pfil hooks don't provide a pointer to the 844 * inpcb, we need to find it ourselves and lock it. 845 */ 846 if (!inp) { 847 /* Find the corresponding inpcb for this pkt. */ 848 inp = siftr_findinpcb(INP_IPV4, ip, *m, th->th_sport, 849 th->th_dport, dir, ss); 850 851 if (inp == NULL) 852 goto ret; 853 else 854 inp_locally_locked = 1; 855 } 856 857 INP_LOCK_ASSERT(inp); 858 859 /* Find the TCP control block that corresponds with this packet */ 860 tp = intotcpcb(inp); 861 862 /* 863 * If we can't find the TCP control block (happens occasionaly for a 864 * packet sent during the shutdown phase of a TCP connection), bail 865 */ 866 if (tp == NULL) { 867 if (dir == PFIL_IN) 868 ss->nskip_in_tcpcb++; 869 else 870 ss->nskip_out_tcpcb++; 871 872 goto inp_unlock; 873 } 874 875 876 pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO); 877 878 if (pn == NULL) { 879 if (dir == PFIL_IN) 880 ss->nskip_in_malloc++; 881 else 882 ss->nskip_out_malloc++; 883 884 goto inp_unlock; 885 } 886 887 siftr_siftdata(pn, inp, tp, INP_IPV4, dir, inp_locally_locked); 888 889 mtx_lock(&siftr_pkt_queue_mtx); 890 STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes); 891 mtx_unlock(&siftr_pkt_queue_mtx); 892 goto ret; 893 894 inp_unlock: 895 if (inp_locally_locked) 896 INP_RUNLOCK(inp); 897 898 ret: 899 return (PFIL_PASS); 900 } 901 902 #ifdef SIFTR_IPV6 903 static pfil_return_t 904 siftr_chkpkt6(struct mbuf **m, struct ifnet *ifp, int flags, 905 void *ruleset __unused, struct inpcb *inp) 906 { 907 struct pkt_node *pn; 908 struct ip6_hdr *ip6; 909 struct tcphdr *th; 910 struct tcpcb *tp; 911 struct siftr_stats *ss; 912 unsigned int ip6_hl; 913 int inp_locally_locked, dir; 914 915 inp_locally_locked = 0; 916 dir = PFIL_DIR(flags); 917 ss = DPCPU_PTR(ss); 918 919 /* 920 * m_pullup is not required here because ip6_{input|output} 921 * already do the heavy lifting for us. 922 */ 923 924 ip6 = mtod(*m, struct ip6_hdr *); 925 926 /* 927 * Only continue processing if the packet is TCP 928 * XXX: We should follow the next header fields 929 * as shown on Pg 6 RFC 2460, but right now we'll 930 * only check pkts that have no extension headers. 931 */ 932 if (ip6->ip6_nxt != IPPROTO_TCP) 933 goto ret6; 934 935 /* 936 * Create a tcphdr struct starting at the correct offset 937 * in the ipv6 packet. 938 */ 939 ip6_hl = sizeof(struct ip6_hdr); 940 th = (struct tcphdr *)((caddr_t)ip6 + ip6_hl); 941 942 /* 943 * Only pkts selected by the tcp port filter 944 * can be inserted into the pkt_queue 945 */ 946 if ((siftr_port_filter != 0) && 947 (siftr_port_filter != ntohs(th->th_sport)) && 948 (siftr_port_filter != ntohs(th->th_dport))) { 949 goto ret6; 950 } 951 952 /* 953 * If a kernel subsystem reinjects packets into the stack, our pfil 954 * hook will be called multiple times for the same packet. 955 * Make sure we only process unique packets. 956 */ 957 if (siftr_chkreinject(*m, dir, ss)) 958 goto ret6; 959 960 if (dir == PFIL_IN) 961 ss->n_in++; 962 else 963 ss->n_out++; 964 965 /* 966 * For inbound packets, the pfil hooks don't provide a pointer to the 967 * inpcb, so we need to find it ourselves and lock it. 968 */ 969 if (!inp) { 970 /* Find the corresponding inpcb for this pkt. */ 971 inp = siftr_findinpcb(INP_IPV6, (struct ip *)ip6, *m, 972 th->th_sport, th->th_dport, dir, ss); 973 974 if (inp == NULL) 975 goto ret6; 976 else 977 inp_locally_locked = 1; 978 } 979 980 /* Find the TCP control block that corresponds with this packet. */ 981 tp = intotcpcb(inp); 982 983 /* 984 * If we can't find the TCP control block (happens occasionaly for a 985 * packet sent during the shutdown phase of a TCP connection), bail 986 */ 987 if (tp == NULL) { 988 if (dir == PFIL_IN) 989 ss->nskip_in_tcpcb++; 990 else 991 ss->nskip_out_tcpcb++; 992 993 goto inp_unlock6; 994 } 995 996 997 pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO); 998 999 if (pn == NULL) { 1000 if (dir == PFIL_IN) 1001 ss->nskip_in_malloc++; 1002 else 1003 ss->nskip_out_malloc++; 1004 1005 goto inp_unlock6; 1006 } 1007 1008 siftr_siftdata(pn, inp, tp, INP_IPV6, dir, inp_locally_locked); 1009 1010 /* XXX: Figure out how to generate hashes for IPv6 packets. */ 1011 1012 mtx_lock(&siftr_pkt_queue_mtx); 1013 STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes); 1014 mtx_unlock(&siftr_pkt_queue_mtx); 1015 goto ret6; 1016 1017 inp_unlock6: 1018 if (inp_locally_locked) 1019 INP_RUNLOCK(inp); 1020 1021 ret6: 1022 return (PFIL_PASS); 1023 } 1024 #endif /* #ifdef SIFTR_IPV6 */ 1025 1026 VNET_DEFINE_STATIC(pfil_hook_t, siftr_inet_hook); 1027 #define V_siftr_inet_hook VNET(siftr_inet_hook) 1028 #ifdef SIFTR_IPV6 1029 VNET_DEFINE_STATIC(pfil_hook_t, siftr_inet6_hook); 1030 #define V_siftr_inet6_hook VNET(siftr_inet6_hook) 1031 #endif 1032 static int 1033 siftr_pfil(int action) 1034 { 1035 struct pfil_hook_args pha = { 1036 .pa_version = PFIL_VERSION, 1037 .pa_flags = PFIL_IN | PFIL_OUT, 1038 .pa_modname = "siftr", 1039 .pa_rulname = "default", 1040 }; 1041 struct pfil_link_args pla = { 1042 .pa_version = PFIL_VERSION, 1043 .pa_flags = PFIL_IN | PFIL_OUT | PFIL_HEADPTR | PFIL_HOOKPTR, 1044 }; 1045 1046 VNET_ITERATOR_DECL(vnet_iter); 1047 1048 VNET_LIST_RLOCK(); 1049 VNET_FOREACH(vnet_iter) { 1050 CURVNET_SET(vnet_iter); 1051 1052 if (action == HOOK) { 1053 pha.pa_mbuf_chk = siftr_chkpkt; 1054 pha.pa_type = PFIL_TYPE_IP4; 1055 V_siftr_inet_hook = pfil_add_hook(&pha); 1056 pla.pa_hook = V_siftr_inet_hook; 1057 pla.pa_head = V_inet_pfil_head; 1058 (void)pfil_link(&pla); 1059 #ifdef SIFTR_IPV6 1060 pha.pa_mbuf_chk = siftr_chkpkt6; 1061 pha.pa_type = PFIL_TYPE_IP6; 1062 V_siftr_inet6_hook = pfil_add_hook(&pha); 1063 pla.pa_hook = V_siftr_inet6_hook; 1064 pla.pa_head = V_inet6_pfil_head; 1065 (void)pfil_link(&pla); 1066 #endif 1067 } else if (action == UNHOOK) { 1068 pfil_remove_hook(V_siftr_inet_hook); 1069 #ifdef SIFTR_IPV6 1070 pfil_remove_hook(V_siftr_inet6_hook); 1071 #endif 1072 } 1073 CURVNET_RESTORE(); 1074 } 1075 VNET_LIST_RUNLOCK(); 1076 1077 return (0); 1078 } 1079 1080 static int 1081 siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS) 1082 { 1083 struct alq *new_alq; 1084 int error; 1085 1086 error = sysctl_handle_string(oidp, arg1, arg2, req); 1087 1088 /* Check for error or same filename */ 1089 if (error != 0 || req->newptr == NULL || 1090 strncmp(siftr_logfile, arg1, arg2) == 0) 1091 goto done; 1092 1093 /* Filname changed */ 1094 error = alq_open(&new_alq, arg1, curthread->td_ucred, 1095 SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0); 1096 if (error != 0) 1097 goto done; 1098 1099 /* 1100 * If disabled, siftr_alq == NULL so we simply close 1101 * the alq as we've proved it can be opened. 1102 * If enabled, close the existing alq and switch the old 1103 * for the new. 1104 */ 1105 if (siftr_alq == NULL) { 1106 alq_close(new_alq); 1107 } else { 1108 alq_close(siftr_alq); 1109 siftr_alq = new_alq; 1110 } 1111 1112 /* Update filename upon success */ 1113 strlcpy(siftr_logfile, arg1, arg2); 1114 done: 1115 return (error); 1116 } 1117 1118 static int 1119 siftr_manage_ops(uint8_t action) 1120 { 1121 struct siftr_stats totalss; 1122 struct timeval tval; 1123 struct flow_hash_node *counter, *tmp_counter; 1124 struct sbuf *s; 1125 int i, key_index, error; 1126 uint32_t bytes_to_write, total_skipped_pkts; 1127 uint16_t lport, fport; 1128 uint8_t *key, ipver __unused; 1129 1130 #ifdef SIFTR_IPV6 1131 uint32_t laddr[4]; 1132 uint32_t faddr[4]; 1133 #else 1134 uint8_t laddr[4]; 1135 uint8_t faddr[4]; 1136 #endif 1137 1138 error = 0; 1139 total_skipped_pkts = 0; 1140 1141 /* Init an autosizing sbuf that initially holds 200 chars. */ 1142 if ((s = sbuf_new(NULL, NULL, 200, SBUF_AUTOEXTEND)) == NULL) 1143 return (-1); 1144 1145 if (action == SIFTR_ENABLE && siftr_pkt_manager_thr == NULL) { 1146 /* 1147 * Create our alq 1148 * XXX: We should abort if alq_open fails! 1149 */ 1150 alq_open(&siftr_alq, siftr_logfile, curthread->td_ucred, 1151 SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0); 1152 1153 STAILQ_INIT(&pkt_queue); 1154 1155 DPCPU_ZERO(ss); 1156 1157 siftr_exit_pkt_manager_thread = 0; 1158 1159 kthread_add(&siftr_pkt_manager_thread, NULL, NULL, 1160 &siftr_pkt_manager_thr, RFNOWAIT, 0, 1161 "siftr_pkt_manager_thr"); 1162 1163 siftr_pfil(HOOK); 1164 1165 microtime(&tval); 1166 1167 sbuf_printf(s, 1168 "enable_time_secs=%jd\tenable_time_usecs=%06ld\t" 1169 "siftrver=%s\tsysname=%s\tsysver=%u\tipmode=%u\n", 1170 (intmax_t)tval.tv_sec, tval.tv_usec, MODVERSION_STR, 1171 SYS_NAME, __FreeBSD_version, SIFTR_IPMODE); 1172 1173 sbuf_finish(s); 1174 alq_writen(siftr_alq, sbuf_data(s), sbuf_len(s), ALQ_WAITOK); 1175 1176 } else if (action == SIFTR_DISABLE && siftr_pkt_manager_thr != NULL) { 1177 /* 1178 * Remove the pfil hook functions. All threads currently in 1179 * the hook functions are allowed to exit before siftr_pfil() 1180 * returns. 1181 */ 1182 siftr_pfil(UNHOOK); 1183 1184 /* This will block until the pkt manager thread unlocks it. */ 1185 mtx_lock(&siftr_pkt_mgr_mtx); 1186 1187 /* Tell the pkt manager thread that it should exit now. */ 1188 siftr_exit_pkt_manager_thread = 1; 1189 1190 /* 1191 * Wake the pkt_manager thread so it realises that 1192 * siftr_exit_pkt_manager_thread == 1 and exits gracefully. 1193 * The wakeup won't be delivered until we unlock 1194 * siftr_pkt_mgr_mtx so this isn't racy. 1195 */ 1196 wakeup(&wait_for_pkt); 1197 1198 /* Wait for the pkt_manager thread to exit. */ 1199 mtx_sleep(siftr_pkt_manager_thr, &siftr_pkt_mgr_mtx, PWAIT, 1200 "thrwait", 0); 1201 1202 siftr_pkt_manager_thr = NULL; 1203 mtx_unlock(&siftr_pkt_mgr_mtx); 1204 1205 totalss.n_in = DPCPU_VARSUM(ss, n_in); 1206 totalss.n_out = DPCPU_VARSUM(ss, n_out); 1207 totalss.nskip_in_malloc = DPCPU_VARSUM(ss, nskip_in_malloc); 1208 totalss.nskip_out_malloc = DPCPU_VARSUM(ss, nskip_out_malloc); 1209 totalss.nskip_in_tcpcb = DPCPU_VARSUM(ss, nskip_in_tcpcb); 1210 totalss.nskip_out_tcpcb = DPCPU_VARSUM(ss, nskip_out_tcpcb); 1211 totalss.nskip_in_inpcb = DPCPU_VARSUM(ss, nskip_in_inpcb); 1212 totalss.nskip_out_inpcb = DPCPU_VARSUM(ss, nskip_out_inpcb); 1213 1214 total_skipped_pkts = totalss.nskip_in_malloc + 1215 totalss.nskip_out_malloc + totalss.nskip_in_tcpcb + 1216 totalss.nskip_out_tcpcb + totalss.nskip_in_inpcb + 1217 totalss.nskip_out_inpcb; 1218 1219 microtime(&tval); 1220 1221 sbuf_printf(s, 1222 "disable_time_secs=%jd\tdisable_time_usecs=%06ld\t" 1223 "num_inbound_tcp_pkts=%ju\tnum_outbound_tcp_pkts=%ju\t" 1224 "total_tcp_pkts=%ju\tnum_inbound_skipped_pkts_malloc=%u\t" 1225 "num_outbound_skipped_pkts_malloc=%u\t" 1226 "num_inbound_skipped_pkts_tcpcb=%u\t" 1227 "num_outbound_skipped_pkts_tcpcb=%u\t" 1228 "num_inbound_skipped_pkts_inpcb=%u\t" 1229 "num_outbound_skipped_pkts_inpcb=%u\t" 1230 "total_skipped_tcp_pkts=%u\tflow_list=", 1231 (intmax_t)tval.tv_sec, 1232 tval.tv_usec, 1233 (uintmax_t)totalss.n_in, 1234 (uintmax_t)totalss.n_out, 1235 (uintmax_t)(totalss.n_in + totalss.n_out), 1236 totalss.nskip_in_malloc, 1237 totalss.nskip_out_malloc, 1238 totalss.nskip_in_tcpcb, 1239 totalss.nskip_out_tcpcb, 1240 totalss.nskip_in_inpcb, 1241 totalss.nskip_out_inpcb, 1242 total_skipped_pkts); 1243 1244 /* 1245 * Iterate over the flow hash, printing a summary of each 1246 * flow seen and freeing any malloc'd memory. 1247 * The hash consists of an array of LISTs (man 3 queue). 1248 */ 1249 for (i = 0; i <= siftr_hashmask; i++) { 1250 LIST_FOREACH_SAFE(counter, counter_hash + i, nodes, 1251 tmp_counter) { 1252 key = counter->key; 1253 key_index = 1; 1254 1255 ipver = key[0]; 1256 1257 memcpy(laddr, key + key_index, sizeof(laddr)); 1258 key_index += sizeof(laddr); 1259 memcpy(&lport, key + key_index, sizeof(lport)); 1260 key_index += sizeof(lport); 1261 memcpy(faddr, key + key_index, sizeof(faddr)); 1262 key_index += sizeof(faddr); 1263 memcpy(&fport, key + key_index, sizeof(fport)); 1264 1265 #ifdef SIFTR_IPV6 1266 laddr[3] = ntohl(laddr[3]); 1267 faddr[3] = ntohl(faddr[3]); 1268 1269 if (ipver == INP_IPV6) { 1270 laddr[0] = ntohl(laddr[0]); 1271 laddr[1] = ntohl(laddr[1]); 1272 laddr[2] = ntohl(laddr[2]); 1273 faddr[0] = ntohl(faddr[0]); 1274 faddr[1] = ntohl(faddr[1]); 1275 faddr[2] = ntohl(faddr[2]); 1276 1277 sbuf_printf(s, 1278 "%x:%x:%x:%x:%x:%x:%x:%x;%u-" 1279 "%x:%x:%x:%x:%x:%x:%x:%x;%u,", 1280 UPPER_SHORT(laddr[0]), 1281 LOWER_SHORT(laddr[0]), 1282 UPPER_SHORT(laddr[1]), 1283 LOWER_SHORT(laddr[1]), 1284 UPPER_SHORT(laddr[2]), 1285 LOWER_SHORT(laddr[2]), 1286 UPPER_SHORT(laddr[3]), 1287 LOWER_SHORT(laddr[3]), 1288 ntohs(lport), 1289 UPPER_SHORT(faddr[0]), 1290 LOWER_SHORT(faddr[0]), 1291 UPPER_SHORT(faddr[1]), 1292 LOWER_SHORT(faddr[1]), 1293 UPPER_SHORT(faddr[2]), 1294 LOWER_SHORT(faddr[2]), 1295 UPPER_SHORT(faddr[3]), 1296 LOWER_SHORT(faddr[3]), 1297 ntohs(fport)); 1298 } else { 1299 laddr[0] = FIRST_OCTET(laddr[3]); 1300 laddr[1] = SECOND_OCTET(laddr[3]); 1301 laddr[2] = THIRD_OCTET(laddr[3]); 1302 laddr[3] = FOURTH_OCTET(laddr[3]); 1303 faddr[0] = FIRST_OCTET(faddr[3]); 1304 faddr[1] = SECOND_OCTET(faddr[3]); 1305 faddr[2] = THIRD_OCTET(faddr[3]); 1306 faddr[3] = FOURTH_OCTET(faddr[3]); 1307 #endif 1308 sbuf_printf(s, 1309 "%u.%u.%u.%u;%u-%u.%u.%u.%u;%u,", 1310 laddr[0], 1311 laddr[1], 1312 laddr[2], 1313 laddr[3], 1314 ntohs(lport), 1315 faddr[0], 1316 faddr[1], 1317 faddr[2], 1318 faddr[3], 1319 ntohs(fport)); 1320 #ifdef SIFTR_IPV6 1321 } 1322 #endif 1323 1324 free(counter, M_SIFTR_HASHNODE); 1325 } 1326 1327 LIST_INIT(counter_hash + i); 1328 } 1329 1330 sbuf_printf(s, "\n"); 1331 sbuf_finish(s); 1332 1333 i = 0; 1334 do { 1335 bytes_to_write = min(SIFTR_ALQ_BUFLEN, sbuf_len(s)-i); 1336 alq_writen(siftr_alq, sbuf_data(s)+i, bytes_to_write, ALQ_WAITOK); 1337 i += bytes_to_write; 1338 } while (i < sbuf_len(s)); 1339 1340 alq_close(siftr_alq); 1341 siftr_alq = NULL; 1342 } else 1343 error = EINVAL; 1344 1345 sbuf_delete(s); 1346 1347 /* 1348 * XXX: Should be using ret to check if any functions fail 1349 * and set error appropriately 1350 */ 1351 1352 return (error); 1353 } 1354 1355 static int 1356 siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS) 1357 { 1358 int error; 1359 uint32_t new; 1360 1361 new = siftr_enabled; 1362 error = sysctl_handle_int(oidp, &new, 0, req); 1363 if (error == 0 && req->newptr != NULL) { 1364 if (new > 1) 1365 return (EINVAL); 1366 else if (new != siftr_enabled) { 1367 if ((error = siftr_manage_ops(new)) == 0) { 1368 siftr_enabled = new; 1369 } else { 1370 siftr_manage_ops(SIFTR_DISABLE); 1371 } 1372 } 1373 } 1374 1375 return (error); 1376 } 1377 1378 static void 1379 siftr_shutdown_handler(void *arg) 1380 { 1381 if (siftr_enabled == 1) { 1382 siftr_manage_ops(SIFTR_DISABLE); 1383 } 1384 } 1385 1386 /* 1387 * Module is being unloaded or machine is shutting down. Take care of cleanup. 1388 */ 1389 static int 1390 deinit_siftr(void) 1391 { 1392 /* Cleanup. */ 1393 siftr_manage_ops(SIFTR_DISABLE); 1394 hashdestroy(counter_hash, M_SIFTR, siftr_hashmask); 1395 mtx_destroy(&siftr_pkt_queue_mtx); 1396 mtx_destroy(&siftr_pkt_mgr_mtx); 1397 1398 return (0); 1399 } 1400 1401 /* 1402 * Module has just been loaded into the kernel. 1403 */ 1404 static int 1405 init_siftr(void) 1406 { 1407 EVENTHANDLER_REGISTER(shutdown_pre_sync, siftr_shutdown_handler, NULL, 1408 SHUTDOWN_PRI_FIRST); 1409 1410 /* Initialise our flow counter hash table. */ 1411 counter_hash = hashinit(SIFTR_EXPECTED_MAX_TCP_FLOWS, M_SIFTR, 1412 &siftr_hashmask); 1413 1414 mtx_init(&siftr_pkt_queue_mtx, "siftr_pkt_queue_mtx", NULL, MTX_DEF); 1415 mtx_init(&siftr_pkt_mgr_mtx, "siftr_pkt_mgr_mtx", NULL, MTX_DEF); 1416 1417 /* Print message to the user's current terminal. */ 1418 uprintf("\nStatistical Information For TCP Research (SIFTR) %s\n" 1419 " http://caia.swin.edu.au/urp/newtcp\n\n", 1420 MODVERSION_STR); 1421 1422 return (0); 1423 } 1424 1425 /* 1426 * This is the function that is called to load and unload the module. 1427 * When the module is loaded, this function is called once with 1428 * "what" == MOD_LOAD 1429 * When the module is unloaded, this function is called twice with 1430 * "what" = MOD_QUIESCE first, followed by "what" = MOD_UNLOAD second 1431 * When the system is shut down e.g. CTRL-ALT-DEL or using the shutdown command, 1432 * this function is called once with "what" = MOD_SHUTDOWN 1433 * When the system is shut down, the handler isn't called until the very end 1434 * of the shutdown sequence i.e. after the disks have been synced. 1435 */ 1436 static int 1437 siftr_load_handler(module_t mod, int what, void *arg) 1438 { 1439 int ret; 1440 1441 switch (what) { 1442 case MOD_LOAD: 1443 ret = init_siftr(); 1444 break; 1445 1446 case MOD_QUIESCE: 1447 case MOD_SHUTDOWN: 1448 ret = deinit_siftr(); 1449 break; 1450 1451 case MOD_UNLOAD: 1452 ret = 0; 1453 break; 1454 1455 default: 1456 ret = EINVAL; 1457 break; 1458 } 1459 1460 return (ret); 1461 } 1462 1463 static moduledata_t siftr_mod = { 1464 .name = "siftr", 1465 .evhand = siftr_load_handler, 1466 }; 1467 1468 /* 1469 * Param 1: name of the kernel module 1470 * Param 2: moduledata_t struct containing info about the kernel module 1471 * and the execution entry point for the module 1472 * Param 3: From sysinit_sub_id enumeration in /usr/include/sys/kernel.h 1473 * Defines the module initialisation order 1474 * Param 4: From sysinit_elem_order enumeration in /usr/include/sys/kernel.h 1475 * Defines the initialisation order of this kld relative to others 1476 * within the same subsystem as defined by param 3 1477 */ 1478 DECLARE_MODULE(siftr, siftr_mod, SI_SUB_LAST, SI_ORDER_ANY); 1479 MODULE_DEPEND(siftr, alq, 1, 1, 1); 1480 MODULE_VERSION(siftr, MODVERSION); 1481