siftr.c revision 209982
1/*- 2 * Copyright (c) 2007-2009, Centre for Advanced Internet Architectures 3 * Swinburne University of Technology, Melbourne, Australia 4 * (CRICOS number 00111D). 5 * Copyright (c) 2009-2010, The FreeBSD Foundation 6 * All rights reserved. 7 * 8 * Portions of this software were developed at the Centre for Advanced 9 * Internet Architectures, Swinburne University of Technology, Melbourne, 10 * Australia by Lawrence Stewart under sponsorship from the FreeBSD Foundation. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 */ 33 34/****************************************************** 35 * Statistical Information For TCP Research (SIFTR) 36 * 37 * A FreeBSD kernel module that adds very basic intrumentation to the 38 * TCP stack, allowing internal stats to be recorded to a log file 39 * for experimental, debugging and performance analysis purposes. 40 * 41 * SIFTR was first released in 2007 by James Healy and Lawrence Stewart whilst 42 * working on the NewTCP research project at Swinburne University's Centre for 43 * Advanced Internet Architectures, Melbourne, Australia, which was made 44 * possible in part by a grant from the Cisco University Research Program Fund 45 * at Community Foundation Silicon Valley. More details are available at: 46 * http://caia.swin.edu.au/urp/newtcp/ 47 * 48 * Work on SIFTR v1.2.x was sponsored by the FreeBSD Foundation as part of 49 * the "Enhancing the FreeBSD TCP Implementation" project 2008-2009. 50 * More details are available at: 51 * http://www.freebsdfoundation.org/ 52 * http://caia.swin.edu.au/freebsd/etcp09/ 53 * 54 * Lawrence Stewart is the current maintainer, and all contact regarding 55 * SIFTR should be directed to him via email: lastewart@swin.edu.au 56 * 57 * Initial release date: June 2007 58 * Most recent update: June 2010 59 ******************************************************/ 60 61#include <sys/cdefs.h> 62__FBSDID("$FreeBSD: head/sys/netinet/siftr.c 209982 2010-07-13 08:23:46Z lstewart $"); 63 64#include <sys/param.h> 65#include <sys/alq.h> 66#include <sys/errno.h> 67#include <sys/hash.h> 68#include <sys/kernel.h> 69#include <sys/kthread.h> 70#include <sys/lock.h> 71#include <sys/mbuf.h> 72#include <sys/module.h> 73#include <sys/mutex.h> 74#include <sys/pcpu.h> 75#include <sys/proc.h> 76#include <sys/sbuf.h> 77#include <sys/smp.h> 78#include <sys/socket.h> 79#include <sys/socketvar.h> 80#include <sys/sysctl.h> 81#include <sys/unistd.h> 82 83#include <net/if.h> 84#include <net/pfil.h> 85 86#include <netinet/in.h> 87#include <netinet/in_pcb.h> 88#include <netinet/in_systm.h> 89#include <netinet/in_var.h> 90#include <netinet/ip.h> 91#include <netinet/tcp_var.h> 92 93#ifdef SIFTR_IPV6 94#include <netinet/ip6.h> 95#include <netinet6/in6_pcb.h> 96#endif /* SIFTR_IPV6 */ 97 98#include <machine/in_cksum.h> 99 100/* 101 * Three digit version number refers to X.Y.Z where: 102 * X is the major version number 103 * Y is bumped to mark backwards incompatible changes 104 * Z is bumped to mark backwards compatible changes 105 */ 106#define V_MAJOR 1 107#define V_BACKBREAK 2 108#define V_BACKCOMPAT 3 109#define MODVERSION __CONCAT(V_MAJOR, __CONCAT(V_BACKBREAK, V_BACKCOMPAT)) 110#define MODVERSION_STR __XSTRING(V_MAJOR) "." __XSTRING(V_BACKBREAK) "." \ 111 __XSTRING(V_BACKCOMPAT) 112 113#define HOOK 0 114#define UNHOOK 1 115#define SIFTR_EXPECTED_MAX_TCP_FLOWS 65536 116#define SYS_NAME "FreeBSD" 117#define PACKET_TAG_SIFTR 100 118#define PACKET_COOKIE_SIFTR 21749576 119#define SIFTR_LOG_FILE_MODE 0644 120#define SIFTR_DISABLE 0 121#define SIFTR_ENABLE 1 122 123/* 124 * Hard upper limit on the length of log messages. Bump this up if you add new 125 * data fields such that the line length could exceed the below value. 126 */ 127#define MAX_LOG_MSG_LEN 200 128/* XXX: Make this a sysctl tunable. */ 129#define SIFTR_ALQ_BUFLEN (1000*MAX_LOG_MSG_LEN) 130 131/* 132 * 1 byte for IP version 133 * IPv4: src/dst IP (4+4) + src/dst port (2+2) = 12 bytes 134 * IPv6: src/dst IP (16+16) + src/dst port (2+2) = 36 bytes 135 */ 136#ifdef SIFTR_IPV6 137#define FLOW_KEY_LEN 37 138#else 139#define FLOW_KEY_LEN 13 140#endif 141 142#ifdef SIFTR_IPV6 143#define SIFTR_IPMODE 6 144#else 145#define SIFTR_IPMODE 4 146#endif 147 148/* useful macros */ 149#define CAST_PTR_INT(X) (*((int*)(X))) 150 151#define UPPER_SHORT(X) (((X) & 0xFFFF0000) >> 16) 152#define LOWER_SHORT(X) ((X) & 0x0000FFFF) 153 154#define FIRST_OCTET(X) (((X) & 0xFF000000) >> 24) 155#define SECOND_OCTET(X) (((X) & 0x00FF0000) >> 16) 156#define THIRD_OCTET(X) (((X) & 0x0000FF00) >> 8) 157#define FOURTH_OCTET(X) ((X) & 0x000000FF) 158 159MALLOC_DECLARE(M_SIFTR); 160MALLOC_DEFINE(M_SIFTR, "siftr", "dynamic memory used by SIFTR"); 161 162MALLOC_DECLARE(M_SIFTR_PKTNODE); 163MALLOC_DEFINE(M_SIFTR_PKTNODE, "siftr_pktnode", "SIFTR pkt_node struct"); 164 165MALLOC_DECLARE(M_SIFTR_HASHNODE); 166MALLOC_DEFINE(M_SIFTR_HASHNODE, "siftr_hashnode", "SIFTR flow_hash_node struct"); 167 168/* Used as links in the pkt manager queue. */ 169struct pkt_node { 170 /* Timestamp of pkt as noted in the pfil hook. */ 171 struct timeval tval; 172 /* Direction pkt is travelling; either PFIL_IN or PFIL_OUT. */ 173 uint8_t direction; 174 /* IP version pkt_node relates to; either INP_IPV4 or INP_IPV6. */ 175 uint8_t ipver; 176 /* Hash of the pkt which triggered the log message. */ 177 uint32_t hash; 178 /* Local/foreign IP address. */ 179#ifdef SIFTR_IPV6 180 uint32_t ip_laddr[4]; 181 uint32_t ip_faddr[4]; 182#else 183 uint8_t ip_laddr[4]; 184 uint8_t ip_faddr[4]; 185#endif 186 /* Local TCP port. */ 187 uint16_t tcp_localport; 188 /* Foreign TCP port. */ 189 uint16_t tcp_foreignport; 190 /* Congestion Window (bytes). */ 191 u_long snd_cwnd; 192 /* Sending Window (bytes). */ 193 u_long snd_wnd; 194 /* Receive Window (bytes). */ 195 u_long rcv_wnd; 196 /* Bandwidth Controlled Window (bytes). */ 197 u_long snd_bwnd; 198 /* Slow Start Threshold (bytes). */ 199 u_long snd_ssthresh; 200 /* Current state of the TCP FSM. */ 201 int conn_state; 202 /* Max Segment Size (bytes). */ 203 u_int max_seg_size; 204 /* 205 * Smoothed RTT stored as found in the TCP control block 206 * in units of (TCP_RTT_SCALE*hz). 207 */ 208 int smoothed_rtt; 209 /* Is SACK enabled? */ 210 u_char sack_enabled; 211 /* Window scaling for snd window. */ 212 u_char snd_scale; 213 /* Window scaling for recv window. */ 214 u_char rcv_scale; 215 /* TCP control block flags. */ 216 u_int flags; 217 /* Retransmit timeout length. */ 218 int rxt_length; 219 /* Size of the TCP send buffer in bytes. */ 220 u_int snd_buf_hiwater; 221 /* Current num bytes in the send socket buffer. */ 222 u_int snd_buf_cc; 223 /* Size of the TCP receive buffer in bytes. */ 224 u_int rcv_buf_hiwater; 225 /* Current num bytes in the receive socket buffer. */ 226 u_int rcv_buf_cc; 227 /* Number of bytes inflight that we are waiting on ACKs for. */ 228 u_int sent_inflight_bytes; 229 /* Link to next pkt_node in the list. */ 230 STAILQ_ENTRY(pkt_node) nodes; 231}; 232 233struct flow_hash_node 234{ 235 uint16_t counter; 236 uint8_t key[FLOW_KEY_LEN]; 237 LIST_ENTRY(flow_hash_node) nodes; 238}; 239 240struct siftr_stats 241{ 242 /* # TCP pkts seen by the SIFTR PFIL hooks, including any skipped. */ 243 uint64_t n_in; 244 uint64_t n_out; 245 /* # pkts skipped due to failed malloc calls. */ 246 uint32_t nskip_in_malloc; 247 uint32_t nskip_out_malloc; 248 /* # pkts skipped due to failed mtx acquisition. */ 249 uint32_t nskip_in_mtx; 250 uint32_t nskip_out_mtx; 251 /* # pkts skipped due to failed inpcb lookups. */ 252 uint32_t nskip_in_inpcb; 253 uint32_t nskip_out_inpcb; 254 /* # pkts skipped due to failed tcpcb lookups. */ 255 uint32_t nskip_in_tcpcb; 256 uint32_t nskip_out_tcpcb; 257 /* # pkts skipped due to stack reinjection. */ 258 uint32_t nskip_in_dejavu; 259 uint32_t nskip_out_dejavu; 260}; 261 262static DPCPU_DEFINE(struct siftr_stats, ss); 263 264static volatile unsigned int siftr_exit_pkt_manager_thread = 0; 265static unsigned int siftr_enabled = 0; 266static unsigned int siftr_pkts_per_log = 1; 267static unsigned int siftr_generate_hashes = 0; 268/* static unsigned int siftr_binary_log = 0; */ 269static char siftr_logfile[PATH_MAX] = "/var/log/siftr.log"; 270static u_long siftr_hashmask; 271STAILQ_HEAD(pkthead, pkt_node) pkt_queue = STAILQ_HEAD_INITIALIZER(pkt_queue); 272LIST_HEAD(listhead, flow_hash_node) *counter_hash; 273static int wait_for_pkt; 274static struct alq *siftr_alq = NULL; 275static struct mtx siftr_pkt_queue_mtx; 276static struct mtx siftr_pkt_mgr_mtx; 277static struct thread *siftr_pkt_manager_thr = NULL; 278/* 279 * pfil.h defines PFIL_IN as 1 and PFIL_OUT as 2, 280 * which we use as an index into this array. 281 */ 282static char direction[3] = {'\0', 'i','o'}; 283 284/* Required function prototypes. */ 285static int siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS); 286static int siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS); 287 288 289/* Declare the net.inet.siftr sysctl tree and populate it. */ 290 291SYSCTL_DECL(_net_inet_siftr); 292 293SYSCTL_NODE(_net_inet, OID_AUTO, siftr, CTLFLAG_RW, NULL, 294 "siftr related settings"); 295 296SYSCTL_PROC(_net_inet_siftr, OID_AUTO, enabled, CTLTYPE_UINT|CTLFLAG_RW, 297 &siftr_enabled, 0, &siftr_sysctl_enabled_handler, "IU", 298 "switch siftr module operations on/off"); 299 300SYSCTL_PROC(_net_inet_siftr, OID_AUTO, logfile, CTLTYPE_STRING|CTLFLAG_RW, 301 &siftr_logfile, sizeof(siftr_logfile), &siftr_sysctl_logfile_name_handler, 302 "A", "file to save siftr log messages to"); 303 304SYSCTL_UINT(_net_inet_siftr, OID_AUTO, ppl, CTLFLAG_RW, 305 &siftr_pkts_per_log, 1, 306 "number of packets between generating a log message"); 307 308SYSCTL_UINT(_net_inet_siftr, OID_AUTO, genhashes, CTLFLAG_RW, 309 &siftr_generate_hashes, 0, 310 "enable packet hash generation"); 311 312/* XXX: TODO 313SYSCTL_UINT(_net_inet_siftr, OID_AUTO, binary, CTLFLAG_RW, 314 &siftr_binary_log, 0, 315 "write log files in binary instead of ascii"); 316*/ 317 318 319/* Begin functions. */ 320 321static void 322siftr_process_pkt(struct pkt_node * pkt_node) 323{ 324 struct flow_hash_node *hash_node; 325 struct listhead *counter_list; 326 struct siftr_stats *ss; 327 struct ale *log_buf; 328 uint8_t key[FLOW_KEY_LEN]; 329 uint8_t found_match, key_offset; 330 331 hash_node = NULL; 332 ss = DPCPU_PTR(ss); 333 found_match = 0; 334 key_offset = 1; 335 336 /* 337 * Create the key that will be used to create a hash index 338 * into our hash table. Our key consists of: 339 * ipversion, localip, localport, foreignip, foreignport 340 */ 341 key[0] = pkt_node->ipver; 342 memcpy(key + key_offset, &pkt_node->ip_laddr, 343 sizeof(pkt_node->ip_laddr)); 344 key_offset += sizeof(pkt_node->ip_laddr); 345 memcpy(key + key_offset, &pkt_node->tcp_localport, 346 sizeof(pkt_node->tcp_localport)); 347 key_offset += sizeof(pkt_node->tcp_localport); 348 memcpy(key + key_offset, &pkt_node->ip_faddr, 349 sizeof(pkt_node->ip_faddr)); 350 key_offset += sizeof(pkt_node->ip_faddr); 351 memcpy(key + key_offset, &pkt_node->tcp_foreignport, 352 sizeof(pkt_node->tcp_foreignport)); 353 354 counter_list = counter_hash + 355 (hash32_buf(key, sizeof(key), 0) & siftr_hashmask); 356 357 /* 358 * If the list is not empty i.e. the hash index has 359 * been used by another flow previously. 360 */ 361 if (LIST_FIRST(counter_list) != NULL) { 362 /* 363 * Loop through the hash nodes in the list. 364 * There should normally only be 1 hash node in the list, 365 * except if there have been collisions at the hash index 366 * computed by hash32_buf(). 367 */ 368 LIST_FOREACH(hash_node, counter_list, nodes) { 369 /* 370 * Check if the key for the pkt we are currently 371 * processing is the same as the key stored in the 372 * hash node we are currently processing. 373 * If they are the same, then we've found the 374 * hash node that stores the counter for the flow 375 * the pkt belongs to. 376 */ 377 if (memcmp(hash_node->key, key, sizeof(key)) == 0) { 378 found_match = 1; 379 break; 380 } 381 } 382 } 383 384 /* If this flow hash hasn't been seen before or we have a collision. */ 385 if (hash_node == NULL || !found_match) { 386 /* Create a new hash node to store the flow's counter. */ 387 hash_node = malloc(sizeof(struct flow_hash_node), 388 M_SIFTR_HASHNODE, M_WAITOK); 389 390 if (hash_node != NULL) { 391 /* Initialise our new hash node list entry. */ 392 hash_node->counter = 0; 393 memcpy(hash_node->key, key, sizeof(key)); 394 LIST_INSERT_HEAD(counter_list, hash_node, nodes); 395 } else { 396 /* Malloc failed. */ 397 if (pkt_node->direction == PFIL_IN) 398 ss->nskip_in_malloc++; 399 else 400 ss->nskip_out_malloc++; 401 402 return; 403 } 404 } else if (siftr_pkts_per_log > 1) { 405 /* 406 * Taking the remainder of the counter divided 407 * by the current value of siftr_pkts_per_log 408 * and storing that in counter provides a neat 409 * way to modulate the frequency of log 410 * messages being written to the log file. 411 */ 412 hash_node->counter = (hash_node->counter + 1) % 413 siftr_pkts_per_log; 414 415 /* 416 * If we have not seen enough packets since the last time 417 * we wrote a log message for this connection, return. 418 */ 419 if (hash_node->counter > 0) 420 return; 421 } 422 423 log_buf = alq_getn(siftr_alq, MAX_LOG_MSG_LEN, ALQ_WAITOK); 424 425 if (log_buf == NULL) 426 return; /* Should only happen if the ALQ is shutting down. */ 427 428#ifdef SIFTR_IPV6 429 pkt_node->ip_laddr[3] = ntohl(pkt_node->ip_laddr[3]); 430 pkt_node->ip_faddr[3] = ntohl(pkt_node->ip_faddr[3]); 431 432 if (pkt_node->ipver == INP_IPV6) { /* IPv6 packet */ 433 pkt_node->ip_laddr[0] = ntohl(pkt_node->ip_laddr[0]); 434 pkt_node->ip_laddr[1] = ntohl(pkt_node->ip_laddr[1]); 435 pkt_node->ip_laddr[2] = ntohl(pkt_node->ip_laddr[2]); 436 pkt_node->ip_faddr[0] = ntohl(pkt_node->ip_faddr[0]); 437 pkt_node->ip_faddr[1] = ntohl(pkt_node->ip_faddr[1]); 438 pkt_node->ip_faddr[2] = ntohl(pkt_node->ip_faddr[2]); 439 440 /* Construct an IPv6 log message. */ 441 log_buf->ae_bytesused = snprintf(log_buf->ae_data, 442 MAX_LOG_MSG_LEN, 443 "%c,0x%08x,%zd.%06ld,%x:%x:%x:%x:%x:%x:%x:%x,%u,%x:%x:%x:" 444 "%x:%x:%x:%x:%x,%u,%ld,%ld,%ld,%ld,%ld,%u,%u,%u,%u,%u,%u," 445 "%u,%d,%u,%u,%u,%u,%u\n", 446 direction[pkt_node->direction], 447 pkt_node->hash, 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->snd_bwnd, 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->smoothed_rtt, 478 pkt_node->sack_enabled, 479 pkt_node->flags, 480 pkt_node->rxt_length, 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 } else { /* IPv4 packet */ 487 pkt_node->ip_laddr[0] = FIRST_OCTET(pkt_node->ip_laddr[3]); 488 pkt_node->ip_laddr[1] = SECOND_OCTET(pkt_node->ip_laddr[3]); 489 pkt_node->ip_laddr[2] = THIRD_OCTET(pkt_node->ip_laddr[3]); 490 pkt_node->ip_laddr[3] = FOURTH_OCTET(pkt_node->ip_laddr[3]); 491 pkt_node->ip_faddr[0] = FIRST_OCTET(pkt_node->ip_faddr[3]); 492 pkt_node->ip_faddr[1] = SECOND_OCTET(pkt_node->ip_faddr[3]); 493 pkt_node->ip_faddr[2] = THIRD_OCTET(pkt_node->ip_faddr[3]); 494 pkt_node->ip_faddr[3] = FOURTH_OCTET(pkt_node->ip_faddr[3]); 495#endif /* SIFTR_IPV6 */ 496 497 /* Construct an IPv4 log message. */ 498 log_buf->ae_bytesused = snprintf(log_buf->ae_data, 499 MAX_LOG_MSG_LEN, 500 "%c,0x%08x,%jd.%06ld,%u.%u.%u.%u,%u,%u.%u.%u.%u,%u,%ld,%ld," 501 "%ld,%ld,%ld,%u,%u,%u,%u,%u,%u,%u,%d,%u,%u,%u,%u,%u\n", 502 direction[pkt_node->direction], 503 pkt_node->hash, 504 (intmax_t)pkt_node->tval.tv_sec, 505 pkt_node->tval.tv_usec, 506 pkt_node->ip_laddr[0], 507 pkt_node->ip_laddr[1], 508 pkt_node->ip_laddr[2], 509 pkt_node->ip_laddr[3], 510 ntohs(pkt_node->tcp_localport), 511 pkt_node->ip_faddr[0], 512 pkt_node->ip_faddr[1], 513 pkt_node->ip_faddr[2], 514 pkt_node->ip_faddr[3], 515 ntohs(pkt_node->tcp_foreignport), 516 pkt_node->snd_ssthresh, 517 pkt_node->snd_cwnd, 518 pkt_node->snd_bwnd, 519 pkt_node->snd_wnd, 520 pkt_node->rcv_wnd, 521 pkt_node->snd_scale, 522 pkt_node->rcv_scale, 523 pkt_node->conn_state, 524 pkt_node->max_seg_size, 525 pkt_node->smoothed_rtt, 526 pkt_node->sack_enabled, 527 pkt_node->flags, 528 pkt_node->rxt_length, 529 pkt_node->snd_buf_hiwater, 530 pkt_node->snd_buf_cc, 531 pkt_node->rcv_buf_hiwater, 532 pkt_node->rcv_buf_cc, 533 pkt_node->sent_inflight_bytes); 534#ifdef SIFTR_IPV6 535 } 536#endif 537 538 alq_post_flags(siftr_alq, log_buf, 0); 539} 540 541 542static void 543siftr_pkt_manager_thread(void *arg) 544{ 545 STAILQ_HEAD(pkthead, pkt_node) tmp_pkt_queue = 546 STAILQ_HEAD_INITIALIZER(tmp_pkt_queue); 547 struct pkt_node *pkt_node, *pkt_node_temp; 548 uint8_t draining; 549 550 draining = 2; 551 552 mtx_lock(&siftr_pkt_mgr_mtx); 553 554 /* draining == 0 when queue has been flushed and it's safe to exit. */ 555 while (draining) { 556 /* 557 * Sleep until we are signalled to wake because thread has 558 * been told to exit or until 1 tick has passed. 559 */ 560 mtx_sleep(&wait_for_pkt, &siftr_pkt_mgr_mtx, PWAIT, "pktwait", 561 1); 562 563 /* Gain exclusive access to the pkt_node queue. */ 564 mtx_lock(&siftr_pkt_queue_mtx); 565 566 /* 567 * Move pkt_queue to tmp_pkt_queue, which leaves 568 * pkt_queue empty and ready to receive more pkt_nodes. 569 */ 570 STAILQ_CONCAT(&tmp_pkt_queue, &pkt_queue); 571 572 /* 573 * We've finished making changes to the list. Unlock it 574 * so the pfil hooks can continue queuing pkt_nodes. 575 */ 576 mtx_unlock(&siftr_pkt_queue_mtx); 577 578 /* 579 * We can't hold a mutex whilst calling siftr_process_pkt 580 * because ALQ might sleep waiting for buffer space. 581 */ 582 mtx_unlock(&siftr_pkt_mgr_mtx); 583 584 /* Flush all pkt_nodes to the log file. */ 585 STAILQ_FOREACH_SAFE(pkt_node, &tmp_pkt_queue, nodes, 586 pkt_node_temp) { 587 siftr_process_pkt(pkt_node); 588 STAILQ_REMOVE_HEAD(&tmp_pkt_queue, nodes); 589 free(pkt_node, M_SIFTR_PKTNODE); 590 } 591 592 KASSERT(STAILQ_EMPTY(&tmp_pkt_queue), 593 ("SIFTR tmp_pkt_queue not empty after flush")); 594 595 mtx_lock(&siftr_pkt_mgr_mtx); 596 597 /* 598 * If siftr_exit_pkt_manager_thread gets set during the window 599 * where we are draining the tmp_pkt_queue above, there might 600 * still be pkts in pkt_queue that need to be drained. 601 * Allow one further iteration to occur after 602 * siftr_exit_pkt_manager_thread has been set to ensure 603 * pkt_queue is completely empty before we kill the thread. 604 * 605 * siftr_exit_pkt_manager_thread is set only after the pfil 606 * hooks have been removed, so only 1 extra iteration 607 * is needed to drain the queue. 608 */ 609 if (siftr_exit_pkt_manager_thread) 610 draining--; 611 } 612 613 mtx_unlock(&siftr_pkt_mgr_mtx); 614 615 /* Calls wakeup on this thread's struct thread ptr. */ 616 kthread_exit(); 617} 618 619 620static uint32_t 621hash_pkt(struct mbuf *m, uint32_t offset) 622{ 623 uint32_t hash; 624 625 hash = 0; 626 627 while (m != NULL && offset > m->m_len) { 628 /* 629 * The IP packet payload does not start in this mbuf, so 630 * need to figure out which mbuf it starts in and what offset 631 * into the mbuf's data region the payload starts at. 632 */ 633 offset -= m->m_len; 634 m = m->m_next; 635 } 636 637 while (m != NULL) { 638 /* Ensure there is data in the mbuf */ 639 if ((m->m_len - offset) > 0) 640 hash = hash32_buf(m->m_data + offset, 641 m->m_len - offset, hash); 642 643 m = m->m_next; 644 offset = 0; 645 } 646 647 return (hash); 648} 649 650 651/* 652 * Check if a given mbuf has the SIFTR mbuf tag. If it does, log the fact that 653 * it's a reinjected packet and return. If it doesn't, tag the mbuf and return. 654 * Return value >0 means the caller should skip processing this mbuf. 655 */ 656static inline int 657siftr_chkreinject(struct mbuf *m, int dir, struct siftr_stats *ss) 658{ 659 if (m_tag_locate(m, PACKET_COOKIE_SIFTR, PACKET_TAG_SIFTR, NULL) 660 != NULL) { 661 if (dir == PFIL_IN) 662 ss->nskip_in_dejavu++; 663 else 664 ss->nskip_out_dejavu++; 665 666 return (1); 667 } else { 668 struct m_tag *tag = m_tag_alloc(PACKET_COOKIE_SIFTR, 669 PACKET_TAG_SIFTR, 0, M_NOWAIT); 670 if (tag == NULL) { 671 if (dir == PFIL_IN) 672 ss->nskip_in_malloc++; 673 else 674 ss->nskip_out_malloc++; 675 676 return (1); 677 } 678 679 m_tag_prepend(m, tag); 680 } 681 682 return (0); 683} 684 685 686/* 687 * Look up an inpcb for a packet. Return the inpcb pointer if found, or NULL 688 * otherwise. 689 */ 690static inline struct inpcb * 691siftr_findinpcb(int ipver, struct ip *ip, struct mbuf *m, uint16_t sport, 692 uint16_t dport, int dir, struct siftr_stats *ss) 693{ 694 struct inpcb *inp; 695 696 /* We need the tcbinfo lock. */ 697 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 698 INP_INFO_RLOCK(&V_tcbinfo); 699 700 if (dir == PFIL_IN) 701 inp = (ipver == INP_IPV4 ? 702 in_pcblookup_hash(&V_tcbinfo, ip->ip_src, sport, ip->ip_dst, 703 dport, 0, m->m_pkthdr.rcvif) 704 : 705#ifdef SIFTR_IPV6 706 in6_pcblookup_hash(&V_tcbinfo, 707 &((struct ip6_hdr *)ip)->ip6_src, sport, 708 &((struct ip6_hdr *)ip)->ip6_dst, dport, 0, 709 m->m_pkthdr.rcvif) 710#else 711 NULL 712#endif 713 ); 714 715 else 716 inp = (ipver == INP_IPV4 ? 717 in_pcblookup_hash(&V_tcbinfo, ip->ip_dst, dport, ip->ip_src, 718 sport, 0, m->m_pkthdr.rcvif) 719 : 720#ifdef SIFTR_IPV6 721 in6_pcblookup_hash(&V_tcbinfo, 722 &((struct ip6_hdr *)ip)->ip6_dst, dport, 723 &((struct ip6_hdr *)ip)->ip6_src, sport, 0, 724 m->m_pkthdr.rcvif) 725#else 726 NULL 727#endif 728 ); 729 730 /* If we can't find the inpcb, bail. */ 731 if (inp == NULL) { 732 if (dir == PFIL_IN) 733 ss->nskip_in_inpcb++; 734 else 735 ss->nskip_out_inpcb++; 736 737 INP_INFO_RUNLOCK(&V_tcbinfo); 738 } else { 739 /* Acquire the inpcb lock. */ 740 INP_UNLOCK_ASSERT(inp); 741 INP_RLOCK(inp); 742 INP_INFO_RUNLOCK(&V_tcbinfo); 743 } 744 745 return (inp); 746} 747 748 749/* 750 * pfil hook that is called for each IPv4 packet making its way through the 751 * stack in either direction. 752 * The pfil subsystem holds a non-sleepable mutex somewhere when 753 * calling our hook function, so we can't sleep at all. 754 * It's very important to use the M_NOWAIT flag with all function calls 755 * that support it so that they won't sleep, otherwise you get a panic. 756 */ 757static int 758siftr_chkpkt(void *arg, struct mbuf **m, struct ifnet *ifp, int dir, 759 struct inpcb *inp) 760{ 761 struct pkt_node *pkt_node; 762 struct ip *ip; 763 struct tcphdr *th; 764 struct tcpcb *tp; 765 struct siftr_stats *ss; 766 unsigned int ip_hl; 767 uint8_t inp_locally_locked; 768 769 inp_locally_locked = 0; 770 ss = DPCPU_PTR(ss); 771 772 /* 773 * m_pullup is not required here because ip_{input|output} 774 * already do the heavy lifting for us. 775 */ 776 777 ip = mtod(*m, struct ip *); 778 779 /* Only continue processing if the packet is TCP. */ 780 if (ip->ip_p != IPPROTO_TCP) 781 goto ret; 782 783 /* 784 * If a kernel subsystem reinjects packets into the stack, our pfil 785 * hook will be called multiple times for the same packet. 786 * Make sure we only process unique packets. 787 */ 788 if (siftr_chkreinject(*m, dir, ss)) 789 goto ret; 790 791 if (dir == PFIL_IN) 792 ss->n_in++; 793 else 794 ss->n_out++; 795 796 /* 797 * Create a tcphdr struct starting at the correct offset 798 * in the IP packet. ip->ip_hl gives the ip header length 799 * in 4-byte words, so multiply it to get the size in bytes. 800 */ 801 ip_hl = (ip->ip_hl << 2); 802 th = (struct tcphdr *)((caddr_t)ip + ip_hl); 803 804 /* 805 * If the pfil hooks don't provide a pointer to the 806 * inpcb, we need to find it ourselves and lock it. 807 */ 808 if (!inp) { 809 /* Find the corresponding inpcb for this pkt. */ 810 inp = siftr_findinpcb(INP_IPV4, ip, *m, th->th_sport, 811 th->th_dport, dir, ss); 812 813 if (inp == NULL) 814 goto ret; 815 else 816 inp_locally_locked = 1; 817 } 818 819 INP_LOCK_ASSERT(inp); 820 821 pkt_node = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, 822 M_NOWAIT | M_ZERO); 823 824 if (pkt_node == NULL) { 825 if (dir == PFIL_IN) 826 ss->nskip_in_malloc++; 827 else 828 ss->nskip_out_malloc++; 829 830 goto inp_unlock; 831 } 832 833 /* Find the TCP control block that corresponds with this packet */ 834 tp = intotcpcb(inp); 835 836 /* 837 * If we can't find the TCP control block (happens occasionaly for a 838 * packet sent during the shutdown phase of a TCP connection), 839 * or we're in the timewait state, bail 840 */ 841 if (tp == NULL || inp->inp_flags & INP_TIMEWAIT) { 842 if (dir == PFIL_IN) 843 ss->nskip_in_tcpcb++; 844 else 845 ss->nskip_out_tcpcb++; 846 847 free(pkt_node, M_SIFTR_PKTNODE); 848 goto inp_unlock; 849 } 850 851 /* Fill in pkt_node data */ 852#ifdef SIFTR_IPV6 853 pkt_node->ip_laddr[3] = inp->inp_laddr.s_addr; 854 pkt_node->ip_faddr[3] = inp->inp_faddr.s_addr; 855#else 856 *((uint32_t *)pkt_node->ip_laddr) = inp->inp_laddr.s_addr; 857 *((uint32_t *)pkt_node->ip_faddr) = inp->inp_faddr.s_addr; 858#endif 859 pkt_node->tcp_localport = inp->inp_lport; 860 pkt_node->tcp_foreignport = inp->inp_fport; 861 pkt_node->snd_cwnd = tp->snd_cwnd; 862 pkt_node->snd_wnd = tp->snd_wnd; 863 pkt_node->rcv_wnd = tp->rcv_wnd; 864 pkt_node->snd_bwnd = tp->snd_bwnd; 865 pkt_node->snd_ssthresh = tp->snd_ssthresh; 866 pkt_node->snd_scale = tp->snd_scale; 867 pkt_node->rcv_scale = tp->rcv_scale; 868 pkt_node->conn_state = tp->t_state; 869 pkt_node->max_seg_size = tp->t_maxseg; 870 pkt_node->smoothed_rtt = tp->t_srtt; 871 pkt_node->sack_enabled = (tp->t_flags & TF_SACK_PERMIT) != 0; 872 pkt_node->flags = tp->t_flags; 873 pkt_node->rxt_length = tp->t_rxtcur; 874 pkt_node->snd_buf_hiwater = inp->inp_socket->so_snd.sb_hiwat; 875 pkt_node->snd_buf_cc = inp->inp_socket->so_snd.sb_cc; 876 pkt_node->rcv_buf_hiwater = inp->inp_socket->so_rcv.sb_hiwat; 877 pkt_node->rcv_buf_cc = inp->inp_socket->so_rcv.sb_cc; 878 pkt_node->sent_inflight_bytes = tp->snd_max - tp->snd_una; 879 880 /* We've finished accessing the tcb so release the lock. */ 881 if (inp_locally_locked) 882 INP_RUNLOCK(inp); 883 884 /* These are safe to access without the inp lock. */ 885 pkt_node->ipver = INP_IPV4; 886 pkt_node->direction = dir; 887 888 /* 889 * Significantly more accurate than using getmicrotime(), but slower! 890 * Gives true microsecond resolution at the expense of a hit to 891 * maximum pps throughput processing when SIFTR is loaded and enabled. 892 */ 893 microtime(&(pkt_node->tval)); 894 895 if (siftr_generate_hashes) { 896 if ((*m)->m_pkthdr.csum_flags & CSUM_TCP) { 897 /* 898 * For outbound packets, the TCP checksum isn't 899 * calculated yet. This is a problem for our packet 900 * hashing as the receiver will calc a different hash 901 * to ours if we don't include the correct TCP checksum 902 * in the bytes being hashed. To work around this 903 * problem, we manually calc the TCP checksum here in 904 * software. We unset the CSUM_TCP flag so the lower 905 * layers don't recalc it. 906 */ 907 (*m)->m_pkthdr.csum_flags &= ~CSUM_TCP; 908 909 /* 910 * Calculate the TCP checksum in software and assign 911 * to correct TCP header field, which will follow the 912 * packet mbuf down the stack. The trick here is that 913 * tcp_output() sets th->th_sum to the checksum of the 914 * pseudo header for us already. Because of the nature 915 * of the checksumming algorithm, we can sum over the 916 * entire IP payload (i.e. TCP header and data), which 917 * will include the already calculated pseduo header 918 * checksum, thus giving us the complete TCP checksum. 919 * 920 * To put it in simple terms, if checksum(1,2,3,4)=10, 921 * then checksum(1,2,3,4,5) == checksum(10,5). 922 * This property is what allows us to "cheat" and 923 * checksum only the IP payload which has the TCP 924 * th_sum field populated with the pseudo header's 925 * checksum, and not need to futz around checksumming 926 * pseudo header bytes and TCP header/data in one hit. 927 * Refer to RFC 1071 for more info. 928 * 929 * NB: in_cksum_skip(struct mbuf *m, int len, int skip) 930 * in_cksum_skip 2nd argument is NOT the number of 931 * bytes to read from the mbuf at "skip" bytes offset 932 * from the start of the mbuf (very counter intuitive!). 933 * The number of bytes to read is calculated internally 934 * by the function as len-skip i.e. to sum over the IP 935 * payload (TCP header + data) bytes, it is INCORRECT 936 * to call the function like this: 937 * in_cksum_skip(at, ip->ip_len - offset, offset) 938 * Rather, it should be called like this: 939 * in_cksum_skip(at, ip->ip_len, offset) 940 * which means read "ip->ip_len - offset" bytes from 941 * the mbuf cluster "at" at offset "offset" bytes from 942 * the beginning of the "at" mbuf's data pointer. 943 */ 944 th->th_sum = in_cksum_skip(*m, ip->ip_len, ip_hl); 945 } 946 947 /* 948 * XXX: Having to calculate the checksum in software and then 949 * hash over all bytes is really inefficient. Would be nice to 950 * find a way to create the hash and checksum in the same pass 951 * over the bytes. 952 */ 953 pkt_node->hash = hash_pkt(*m, ip_hl); 954 } 955 956 mtx_lock(&siftr_pkt_queue_mtx); 957 STAILQ_INSERT_TAIL(&pkt_queue, pkt_node, nodes); 958 mtx_unlock(&siftr_pkt_queue_mtx); 959 goto ret; 960 961inp_unlock: 962 if (inp_locally_locked) 963 INP_RUNLOCK(inp); 964 965ret: 966 /* Returning 0 ensures pfil will not discard the pkt */ 967 return (0); 968} 969 970 971#ifdef SIFTR_IPV6 972static int 973siftr_chkpkt6(void *arg, struct mbuf **m, struct ifnet *ifp, int dir, 974 struct inpcb *inp) 975{ 976 struct pkt_node *pkt_node; 977 struct ip6_hdr *ip6; 978 struct tcphdr *th; 979 struct tcpcb *tp; 980 struct siftr_stats *ss; 981 unsigned int ip6_hl; 982 uint8_t inp_locally_locked; 983 984 inp_locally_locked = 0; 985 ss = DPCPU_PTR(ss); 986 987 /* 988 * m_pullup is not required here because ip6_{input|output} 989 * already do the heavy lifting for us. 990 */ 991 992 ip6 = mtod(*m, struct ip6_hdr *); 993 994 /* 995 * Only continue processing if the packet is TCP 996 * XXX: We should follow the next header fields 997 * as shown on Pg 6 RFC 2460, but right now we'll 998 * only check pkts that have no extension headers. 999 */ 1000 if (ip6->ip6_nxt != IPPROTO_TCP) 1001 goto ret6; 1002 1003 /* 1004 * If a kernel subsystem reinjects packets into the stack, our pfil 1005 * hook will be called multiple times for the same packet. 1006 * Make sure we only process unique packets. 1007 */ 1008 if (siftr_chkreinject(*m, dir, ss)) 1009 goto ret6; 1010 1011 if (dir == PFIL_IN) 1012 ss->n_in++; 1013 else 1014 ss->n_out++; 1015 1016 ip6_hl = sizeof(struct ip6_hdr); 1017 1018 /* 1019 * Create a tcphdr struct starting at the correct offset 1020 * in the ipv6 packet. ip->ip_hl gives the ip header length 1021 * in 4-byte words, so multiply it to get the size in bytes. 1022 */ 1023 th = (struct tcphdr *)((caddr_t)ip6 + ip6_hl); 1024 1025 /* 1026 * For inbound packets, the pfil hooks don't provide a pointer to the 1027 * inpcb, so we need to find it ourselves and lock it. 1028 */ 1029 if (!inp) { 1030 /* Find the corresponding inpcb for this pkt. */ 1031 inp = siftr_findinpcb(INP_IPV6, (struct ip *)ip6, *m, 1032 th->th_sport, th->th_dport, dir, ss); 1033 1034 if (inp == NULL) 1035 goto ret6; 1036 else 1037 inp_locally_locked = 1; 1038 } 1039 1040 pkt_node = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, 1041 M_NOWAIT | M_ZERO); 1042 1043 if (pkt_node == NULL) { 1044 if (dir == PFIL_IN) 1045 ss->nskip_in_malloc++; 1046 else 1047 ss->nskip_out_malloc++; 1048 1049 goto inp_unlock6; 1050 } 1051 1052 /* Find the TCP control block that corresponds with this packet. */ 1053 tp = intotcpcb(inp); 1054 1055 /* 1056 * If we can't find the TCP control block (happens occasionaly for a 1057 * packet sent during the shutdown phase of a TCP connection), 1058 * or we're in the timewait state, bail. 1059 */ 1060 if (tp == NULL || inp->inp_flags & INP_TIMEWAIT) { 1061 if (dir == PFIL_IN) 1062 ss->nskip_in_tcpcb++; 1063 else 1064 ss->nskip_out_tcpcb++; 1065 1066 free(pkt_node, M_SIFTR_PKTNODE); 1067 goto inp_unlock6; 1068 } 1069 1070 /* Fill in pkt_node data. */ 1071 pkt_node->ip_laddr[0] = inp->in6p_laddr.s6_addr32[0]; 1072 pkt_node->ip_laddr[1] = inp->in6p_laddr.s6_addr32[1]; 1073 pkt_node->ip_laddr[2] = inp->in6p_laddr.s6_addr32[2]; 1074 pkt_node->ip_laddr[3] = inp->in6p_laddr.s6_addr32[3]; 1075 pkt_node->ip_faddr[0] = inp->in6p_faddr.s6_addr32[0]; 1076 pkt_node->ip_faddr[1] = inp->in6p_faddr.s6_addr32[1]; 1077 pkt_node->ip_faddr[2] = inp->in6p_faddr.s6_addr32[2]; 1078 pkt_node->ip_faddr[3] = inp->in6p_faddr.s6_addr32[3]; 1079 pkt_node->tcp_localport = inp->inp_lport; 1080 pkt_node->tcp_foreignport = inp->inp_fport; 1081 pkt_node->snd_cwnd = tp->snd_cwnd; 1082 pkt_node->snd_wnd = tp->snd_wnd; 1083 pkt_node->rcv_wnd = tp->rcv_wnd; 1084 pkt_node->snd_bwnd = tp->snd_bwnd; 1085 pkt_node->snd_ssthresh = tp->snd_ssthresh; 1086 pkt_node->snd_scale = tp->snd_scale; 1087 pkt_node->rcv_scale = tp->rcv_scale; 1088 pkt_node->conn_state = tp->t_state; 1089 pkt_node->max_seg_size = tp->t_maxseg; 1090 pkt_node->smoothed_rtt = tp->t_srtt; 1091 pkt_node->sack_enabled = (tp->t_flags & TF_SACK_PERMIT) != 0; 1092 pkt_node->flags = tp->t_flags; 1093 pkt_node->rxt_length = tp->t_rxtcur; 1094 pkt_node->snd_buf_hiwater = inp->inp_socket->so_snd.sb_hiwat; 1095 pkt_node->snd_buf_cc = inp->inp_socket->so_snd.sb_cc; 1096 pkt_node->rcv_buf_hiwater = inp->inp_socket->so_rcv.sb_hiwat; 1097 pkt_node->rcv_buf_cc = inp->inp_socket->so_rcv.sb_cc; 1098 pkt_node->sent_inflight_bytes = tp->snd_max - tp->snd_una; 1099 1100 /* We've finished accessing the tcb so release the lock. */ 1101 if (inp_locally_locked) 1102 INP_RUNLOCK(inp); 1103 1104 /* These are safe to access without the inp lock. */ 1105 pkt_node->ipver = INP_IPV6; 1106 pkt_node->direction = dir; 1107 1108 /* 1109 * Significantly more accurate than using getmicrotime(), but slower! 1110 * Gives true microsecond resolution at the expense of a hit to 1111 * maximum pps throughput processing when SIFTR is loaded and enabled. 1112 */ 1113 microtime(&(pkt_node->tval)); 1114 1115 /* XXX: Figure out how to do hash calcs for IPv6 */ 1116 1117 mtx_lock(&siftr_pkt_queue_mtx); 1118 STAILQ_INSERT_TAIL(&pkt_queue, pkt_node, nodes); 1119 mtx_unlock(&siftr_pkt_queue_mtx); 1120 goto ret6; 1121 1122inp_unlock6: 1123 if (inp_locally_locked) 1124 INP_RUNLOCK(inp); 1125 1126ret6: 1127 /* Returning 0 ensures pfil will not discard the pkt. */ 1128 return (0); 1129} 1130#endif /* #ifdef SIFTR_IPV6 */ 1131 1132 1133static int 1134siftr_pfil(int action) 1135{ 1136 struct pfil_head *pfh_inet = pfil_head_get(PFIL_TYPE_AF, AF_INET); 1137#ifdef SIFTR_IPV6 1138 struct pfil_head *pfh_inet6 = pfil_head_get(PFIL_TYPE_AF, AF_INET6); 1139#endif 1140 1141 if (action == HOOK) { 1142 pfil_add_hook(siftr_chkpkt, NULL, 1143 PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet); 1144#ifdef SIFTR_IPV6 1145 pfil_add_hook(siftr_chkpkt6, NULL, 1146 PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet6); 1147#endif 1148 } else if (action == UNHOOK) { 1149 pfil_remove_hook(siftr_chkpkt, NULL, 1150 PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet); 1151#ifdef SIFTR_IPV6 1152 pfil_remove_hook(siftr_chkpkt6, NULL, 1153 PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet6); 1154#endif 1155 } 1156 1157 return (0); 1158} 1159 1160 1161static int 1162siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS) 1163{ 1164 struct alq *new_alq; 1165 int error; 1166 1167 if (req->newptr == NULL) 1168 goto skip; 1169 1170 /* If old filename and new filename are different. */ 1171 if (strncmp(siftr_logfile, (char *)req->newptr, PATH_MAX)) { 1172 1173 error = alq_open(&new_alq, req->newptr, curthread->td_ucred, 1174 SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0); 1175 1176 /* Bail if unable to create new alq. */ 1177 if (error) 1178 return (1); 1179 1180 /* 1181 * If disabled, siftr_alq == NULL so we simply close 1182 * the alq as we've proved it can be opened. 1183 * If enabled, close the existing alq and switch the old 1184 * for the new. 1185 */ 1186 if (siftr_alq == NULL) 1187 alq_close(new_alq); 1188 else { 1189 alq_close(siftr_alq); 1190 siftr_alq = new_alq; 1191 } 1192 } 1193 1194skip: 1195 return (sysctl_handle_string(oidp, arg1, arg2, req)); 1196} 1197 1198 1199static int 1200siftr_manage_ops(uint8_t action) 1201{ 1202 struct siftr_stats totalss; 1203 struct timeval tval; 1204 struct flow_hash_node *counter, *tmp_counter; 1205 struct sbuf *s; 1206 int i, key_index, ret, error; 1207 uint32_t bytes_to_write, total_skipped_pkts; 1208 uint16_t lport, fport; 1209 uint8_t *key, ipver; 1210 1211#ifdef SIFTR_IPV6 1212 uint32_t laddr[4]; 1213 uint32_t faddr[4]; 1214#else 1215 uint8_t laddr[4]; 1216 uint8_t faddr[4]; 1217#endif 1218 1219 error = 0; 1220 total_skipped_pkts = 0; 1221 1222 /* Init an autosizing sbuf that initially holds 200 chars. */ 1223 if ((s = sbuf_new(NULL, NULL, 200, SBUF_AUTOEXTEND)) == NULL) 1224 return (-1); 1225 1226 if (action == SIFTR_ENABLE) { 1227 /* 1228 * Create our alq 1229 * XXX: We should abort if alq_open fails! 1230 */ 1231 alq_open(&siftr_alq, siftr_logfile, curthread->td_ucred, 1232 SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0); 1233 1234 STAILQ_INIT(&pkt_queue); 1235 1236 DPCPU_ZERO(ss); 1237 1238 siftr_exit_pkt_manager_thread = 0; 1239 1240 ret = kthread_add(&siftr_pkt_manager_thread, NULL, NULL, 1241 &siftr_pkt_manager_thr, RFNOWAIT, 0, 1242 "siftr_pkt_manager_thr"); 1243 1244 siftr_pfil(HOOK); 1245 1246 microtime(&tval); 1247 1248 sbuf_printf(s, 1249 "enable_time_secs=%jd\tenable_time_usecs=%06ld\t" 1250 "siftrver=%s\thz=%u\ttcp_rtt_scale=%u\tsysname=%s\t" 1251 "sysver=%u\tipmode=%u\n", 1252 (intmax_t)tval.tv_sec, tval.tv_usec, MODVERSION_STR, hz, 1253 TCP_RTT_SCALE, SYS_NAME, __FreeBSD_version, SIFTR_IPMODE); 1254 1255 sbuf_finish(s); 1256 alq_writen(siftr_alq, sbuf_data(s), sbuf_len(s), ALQ_WAITOK); 1257 1258 } else if (action == SIFTR_DISABLE && siftr_pkt_manager_thr != NULL) { 1259 /* 1260 * Remove the pfil hook functions. All threads currently in 1261 * the hook functions are allowed to exit before siftr_pfil() 1262 * returns. 1263 */ 1264 siftr_pfil(UNHOOK); 1265 1266 /* This will block until the pkt manager thread unlocks it. */ 1267 mtx_lock(&siftr_pkt_mgr_mtx); 1268 1269 /* Tell the pkt manager thread that it should exit now. */ 1270 siftr_exit_pkt_manager_thread = 1; 1271 1272 /* 1273 * Wake the pkt_manager thread so it realises that 1274 * siftr_exit_pkt_manager_thread == 1 and exits gracefully. 1275 * The wakeup won't be delivered until we unlock 1276 * siftr_pkt_mgr_mtx so this isn't racy. 1277 */ 1278 wakeup(&wait_for_pkt); 1279 1280 /* Wait for the pkt_manager thread to exit. */ 1281 mtx_sleep(siftr_pkt_manager_thr, &siftr_pkt_mgr_mtx, PWAIT, 1282 "thrwait", 0); 1283 1284 siftr_pkt_manager_thr = NULL; 1285 mtx_unlock(&siftr_pkt_mgr_mtx); 1286 1287 totalss.n_in = DPCPU_VARSUM(ss, n_in); 1288 totalss.n_out = DPCPU_VARSUM(ss, n_out); 1289 totalss.nskip_in_malloc = DPCPU_VARSUM(ss, nskip_in_malloc); 1290 totalss.nskip_out_malloc = DPCPU_VARSUM(ss, nskip_out_malloc); 1291 totalss.nskip_in_mtx = DPCPU_VARSUM(ss, nskip_in_mtx); 1292 totalss.nskip_out_mtx = DPCPU_VARSUM(ss, nskip_out_mtx); 1293 totalss.nskip_in_tcpcb = DPCPU_VARSUM(ss, nskip_in_tcpcb); 1294 totalss.nskip_out_tcpcb = DPCPU_VARSUM(ss, nskip_out_tcpcb); 1295 totalss.nskip_in_inpcb = DPCPU_VARSUM(ss, nskip_in_inpcb); 1296 totalss.nskip_out_inpcb = DPCPU_VARSUM(ss, nskip_out_inpcb); 1297 1298 total_skipped_pkts = totalss.nskip_in_malloc + 1299 totalss.nskip_out_malloc + totalss.nskip_in_mtx + 1300 totalss.nskip_out_mtx + totalss.nskip_in_tcpcb + 1301 totalss.nskip_out_tcpcb + totalss.nskip_in_inpcb + 1302 totalss.nskip_out_inpcb; 1303 1304 microtime(&tval); 1305 1306 sbuf_printf(s, 1307 "disable_time_secs=%jd\tdisable_time_usecs=%06ld\t" 1308 "num_inbound_tcp_pkts=%ju\tnum_outbound_tcp_pkts=%ju\t" 1309 "total_tcp_pkts=%ju\tnum_inbound_skipped_pkts_malloc=%u\t" 1310 "num_outbound_skipped_pkts_malloc=%u\t" 1311 "num_inbound_skipped_pkts_mtx=%u\t" 1312 "num_outbound_skipped_pkts_mtx=%u\t" 1313 "num_inbound_skipped_pkts_tcpcb=%u\t" 1314 "num_outbound_skipped_pkts_tcpcb=%u\t" 1315 "num_inbound_skipped_pkts_inpcb=%u\t" 1316 "num_outbound_skipped_pkts_inpcb=%u\t" 1317 "total_skipped_tcp_pkts=%u\tflow_list=", 1318 (intmax_t)tval.tv_sec, 1319 tval.tv_usec, 1320 (uintmax_t)totalss.n_in, 1321 (uintmax_t)totalss.n_out, 1322 (uintmax_t)(totalss.n_in + totalss.n_out), 1323 totalss.nskip_in_malloc, 1324 totalss.nskip_out_malloc, 1325 totalss.nskip_in_mtx, 1326 totalss.nskip_out_mtx, 1327 totalss.nskip_in_tcpcb, 1328 totalss.nskip_out_tcpcb, 1329 totalss.nskip_in_inpcb, 1330 totalss.nskip_out_inpcb, 1331 total_skipped_pkts); 1332 1333 /* 1334 * Iterate over the flow hash, printing a summary of each 1335 * flow seen and freeing any malloc'd memory. 1336 * The hash consists of an array of LISTs (man 3 queue). 1337 */ 1338 for (i = 0; i < siftr_hashmask; i++) { 1339 LIST_FOREACH_SAFE(counter, counter_hash + i, nodes, 1340 tmp_counter) { 1341 key = counter->key; 1342 key_index = 1; 1343 1344 ipver = key[0]; 1345 1346 memcpy(laddr, key + key_index, sizeof(laddr)); 1347 key_index += sizeof(laddr); 1348 memcpy(&lport, key + key_index, sizeof(lport)); 1349 key_index += sizeof(lport); 1350 memcpy(faddr, key + key_index, sizeof(faddr)); 1351 key_index += sizeof(faddr); 1352 memcpy(&fport, key + key_index, sizeof(fport)); 1353 1354#ifdef SIFTR_IPV6 1355 laddr[3] = ntohl(laddr[3]); 1356 faddr[3] = ntohl(faddr[3]); 1357 1358 if (ipver == INP_IPV6) { 1359 laddr[0] = ntohl(laddr[0]); 1360 laddr[1] = ntohl(laddr[1]); 1361 laddr[2] = ntohl(laddr[2]); 1362 faddr[0] = ntohl(faddr[0]); 1363 faddr[1] = ntohl(faddr[1]); 1364 faddr[2] = ntohl(faddr[2]); 1365 1366 sbuf_printf(s, 1367 "%x:%x:%x:%x:%x:%x:%x:%x;%u-" 1368 "%x:%x:%x:%x:%x:%x:%x:%x;%u,", 1369 UPPER_SHORT(laddr[0]), 1370 LOWER_SHORT(laddr[0]), 1371 UPPER_SHORT(laddr[1]), 1372 LOWER_SHORT(laddr[1]), 1373 UPPER_SHORT(laddr[2]), 1374 LOWER_SHORT(laddr[2]), 1375 UPPER_SHORT(laddr[3]), 1376 LOWER_SHORT(laddr[3]), 1377 ntohs(lport), 1378 UPPER_SHORT(faddr[0]), 1379 LOWER_SHORT(faddr[0]), 1380 UPPER_SHORT(faddr[1]), 1381 LOWER_SHORT(faddr[1]), 1382 UPPER_SHORT(faddr[2]), 1383 LOWER_SHORT(faddr[2]), 1384 UPPER_SHORT(faddr[3]), 1385 LOWER_SHORT(faddr[3]), 1386 ntohs(fport)); 1387 } else { 1388 laddr[0] = FIRST_OCTET(laddr[3]); 1389 laddr[1] = SECOND_OCTET(laddr[3]); 1390 laddr[2] = THIRD_OCTET(laddr[3]); 1391 laddr[3] = FOURTH_OCTET(laddr[3]); 1392 faddr[0] = FIRST_OCTET(faddr[3]); 1393 faddr[1] = SECOND_OCTET(faddr[3]); 1394 faddr[2] = THIRD_OCTET(faddr[3]); 1395 faddr[3] = FOURTH_OCTET(faddr[3]); 1396#endif 1397 sbuf_printf(s, 1398 "%u.%u.%u.%u;%u-%u.%u.%u.%u;%u,", 1399 laddr[0], 1400 laddr[1], 1401 laddr[2], 1402 laddr[3], 1403 ntohs(lport), 1404 faddr[0], 1405 faddr[1], 1406 faddr[2], 1407 faddr[3], 1408 ntohs(fport)); 1409#ifdef SIFTR_IPV6 1410 } 1411#endif 1412 1413 free(counter, M_SIFTR_HASHNODE); 1414 } 1415 1416 LIST_INIT(counter_hash + i); 1417 } 1418 1419 sbuf_printf(s, "\n"); 1420 sbuf_finish(s); 1421 1422 i = 0; 1423 do { 1424 bytes_to_write = min(SIFTR_ALQ_BUFLEN, sbuf_len(s)-i); 1425 alq_writen(siftr_alq, sbuf_data(s)+i, bytes_to_write, ALQ_WAITOK); 1426 i += bytes_to_write; 1427 } while (i < sbuf_len(s)); 1428 1429 alq_close(siftr_alq); 1430 siftr_alq = NULL; 1431 } 1432 1433 sbuf_delete(s); 1434 1435 /* 1436 * XXX: Should be using ret to check if any functions fail 1437 * and set error appropriately 1438 */ 1439 1440 return (error); 1441} 1442 1443 1444static int 1445siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS) 1446{ 1447 if (req->newptr == NULL) 1448 goto skip; 1449 1450 /* If the value passed in isn't 0 or 1, return an error. */ 1451 if (CAST_PTR_INT(req->newptr) != 0 && CAST_PTR_INT(req->newptr) != 1) 1452 return (1); 1453 1454 /* If we are changing state (0 to 1 or 1 to 0). */ 1455 if (CAST_PTR_INT(req->newptr) != siftr_enabled ) 1456 if (siftr_manage_ops(CAST_PTR_INT(req->newptr))) { 1457 siftr_manage_ops(SIFTR_DISABLE); 1458 return (1); 1459 } 1460 1461skip: 1462 return (sysctl_handle_int(oidp, arg1, arg2, req)); 1463} 1464 1465 1466static void 1467siftr_shutdown_handler(void *arg) 1468{ 1469 siftr_manage_ops(SIFTR_DISABLE); 1470} 1471 1472 1473/* 1474 * Module is being unloaded or machine is shutting down. Take care of cleanup. 1475 */ 1476static int 1477deinit_siftr(void) 1478{ 1479 /* Cleanup. */ 1480 siftr_manage_ops(SIFTR_DISABLE); 1481 hashdestroy(counter_hash, M_SIFTR, siftr_hashmask); 1482 mtx_destroy(&siftr_pkt_queue_mtx); 1483 mtx_destroy(&siftr_pkt_mgr_mtx); 1484 1485 return (0); 1486} 1487 1488 1489/* 1490 * Module has just been loaded into the kernel. 1491 */ 1492static int 1493init_siftr(void) 1494{ 1495 EVENTHANDLER_REGISTER(shutdown_pre_sync, siftr_shutdown_handler, NULL, 1496 SHUTDOWN_PRI_FIRST); 1497 1498 /* Initialise our flow counter hash table. */ 1499 counter_hash = hashinit(SIFTR_EXPECTED_MAX_TCP_FLOWS, M_SIFTR, 1500 &siftr_hashmask); 1501 1502 mtx_init(&siftr_pkt_queue_mtx, "siftr_pkt_queue_mtx", NULL, MTX_DEF); 1503 mtx_init(&siftr_pkt_mgr_mtx, "siftr_pkt_mgr_mtx", NULL, MTX_DEF); 1504 1505 /* Print message to the user's current terminal. */ 1506 uprintf("\nStatistical Information For TCP Research (SIFTR) %s\n" 1507 " http://caia.swin.edu.au/urp/newtcp\n\n", 1508 MODVERSION_STR); 1509 1510 return (0); 1511} 1512 1513 1514/* 1515 * This is the function that is called to load and unload the module. 1516 * When the module is loaded, this function is called once with 1517 * "what" == MOD_LOAD 1518 * When the module is unloaded, this function is called twice with 1519 * "what" = MOD_QUIESCE first, followed by "what" = MOD_UNLOAD second 1520 * When the system is shut down e.g. CTRL-ALT-DEL or using the shutdown command, 1521 * this function is called once with "what" = MOD_SHUTDOWN 1522 * When the system is shut down, the handler isn't called until the very end 1523 * of the shutdown sequence i.e. after the disks have been synced. 1524 */ 1525static int 1526siftr_load_handler(module_t mod, int what, void *arg) 1527{ 1528 int ret; 1529 1530 switch (what) { 1531 case MOD_LOAD: 1532 ret = init_siftr(); 1533 break; 1534 1535 case MOD_QUIESCE: 1536 case MOD_SHUTDOWN: 1537 ret = deinit_siftr(); 1538 break; 1539 1540 case MOD_UNLOAD: 1541 ret = 0; 1542 break; 1543 1544 default: 1545 ret = EINVAL; 1546 break; 1547 } 1548 1549 return (ret); 1550} 1551 1552 1553static moduledata_t siftr_mod = { 1554 .name = "siftr", 1555 .evhand = siftr_load_handler, 1556}; 1557 1558/* 1559 * Param 1: name of the kernel module 1560 * Param 2: moduledata_t struct containing info about the kernel module 1561 * and the execution entry point for the module 1562 * Param 3: From sysinit_sub_id enumeration in /usr/include/sys/kernel.h 1563 * Defines the module initialisation order 1564 * Param 4: From sysinit_elem_order enumeration in /usr/include/sys/kernel.h 1565 * Defines the initialisation order of this kld relative to others 1566 * within the same subsystem as defined by param 3 1567 */ 1568DECLARE_MODULE(siftr, siftr_mod, SI_SUB_SMP, SI_ORDER_ANY); 1569MODULE_DEPEND(siftr, alq, 1, 1, 1); 1570MODULE_VERSION(siftr, MODVERSION); 1571