siftr.c revision 241913
1193323Sed/*- 2193323Sed * Copyright (c) 2007-2009 3193323Sed * Swinburne University of Technology, Melbourne, Australia. 4193323Sed * Copyright (c) 2009-2010, The FreeBSD Foundation 5193323Sed * All rights reserved. 6193323Sed * 7193323Sed * Portions of this software were developed at the Centre for Advanced 8193323Sed * Internet Architectures, Swinburne University of Technology, Melbourne, 9193323Sed * Australia by Lawrence Stewart under sponsorship from the FreeBSD Foundation. 10193323Sed * 11193323Sed * Redistribution and use in source and binary forms, with or without 12193323Sed * modification, are permitted provided that the following conditions 13193323Sed * are met: 14193323Sed * 1. Redistributions of source code must retain the above copyright 15193323Sed * notice, this list of conditions and the following disclaimer. 16193323Sed * 2. Redistributions in binary form must reproduce the above copyright 17193323Sed * notice, this list of conditions and the following disclaimer in the 18224145Sdim * documentation and/or other materials provided with the distribution. 19226633Sdim * 20193323Sed * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 21224145Sdim * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22226633Sdim * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23208599Srdivacky * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 24224145Sdim * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25234353Sdim * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26223017Sdim * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27223017Sdim * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28193323Sed * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29223017Sdim * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30193323Sed * SUCH DAMAGE. 31193323Sed */ 32193323Sed 33223017Sdim/****************************************************** 34193323Sed * Statistical Information For TCP Research (SIFTR) 35234353Sdim * 36234353Sdim * A FreeBSD kernel module that adds very basic intrumentation to the 37234353Sdim * TCP stack, allowing internal stats to be recorded to a log file 38239462Sdim * for experimental, debugging and performance analysis purposes. 39239462Sdim * 40239462Sdim * SIFTR was first released in 2007 by James Healy and Lawrence Stewart whilst 41239462Sdim * working on the NewTCP research project at Swinburne University of 42234353Sdim * Technology's Centre for Advanced Internet Architectures, Melbourne, 43243830Sdim * Australia, which was made possible in part by a grant from the Cisco 44234353Sdim * University Research Program Fund at Community Foundation Silicon Valley. 45234353Sdim * More details are available at: 46239462Sdim * http://caia.swin.edu.au/urp/newtcp/ 47234353Sdim * 48239462Sdim * Work on SIFTR v1.2.x was sponsored by the FreeBSD Foundation as part of 49239462Sdim * the "Enhancing the FreeBSD TCP Implementation" project 2008-2009. 50234353Sdim * More details are available at: 51234353Sdim * http://www.freebsdfoundation.org/ 52234353Sdim * http://caia.swin.edu.au/freebsd/etcp09/ 53234353Sdim * 54234353Sdim * Lawrence Stewart is the current maintainer, and all contact regarding 55234353Sdim * SIFTR should be directed to him via email: lastewart@swin.edu.au 56234353Sdim * 57234353Sdim * Initial release date: June 2007 58234353Sdim * Most recent update: September 2010 59234353Sdim ******************************************************/ 60234353Sdim 61234353Sdim#include <sys/cdefs.h> 62234353Sdim__FBSDID("$FreeBSD: head/sys/netinet/siftr.c 241913 2012-10-22 21:09:03Z glebius $"); 63234353Sdim 64234353Sdim#include <sys/param.h> 65234353Sdim#include <sys/alq.h> 66234353Sdim#include <sys/errno.h> 67234353Sdim#include <sys/hash.h> 68234353Sdim#include <sys/kernel.h> 69234353Sdim#include <sys/kthread.h> 70234353Sdim#include <sys/lock.h> 71234353Sdim#include <sys/mbuf.h> 72234353Sdim#include <sys/module.h> 73234353Sdim#include <sys/mutex.h> 74234353Sdim#include <sys/pcpu.h> 75239462Sdim#include <sys/proc.h> 76234353Sdim#include <sys/sbuf.h> 77234353Sdim#include <sys/smp.h> 78234353Sdim#include <sys/socket.h> 79234353Sdim#include <sys/socketvar.h> 80234353Sdim#include <sys/sysctl.h> 81234353Sdim#include <sys/unistd.h> 82234353Sdim 83234353Sdim#include <net/if.h> 84234353Sdim#include <net/pfil.h> 85234353Sdim 86234353Sdim#include <netinet/in.h> 87243830Sdim#include <netinet/in_pcb.h> 88243830Sdim#include <netinet/in_systm.h> 89243830Sdim#include <netinet/in_var.h> 90234353Sdim#include <netinet/ip.h> 91234353Sdim#include <netinet/tcp_var.h> 92234353Sdim 93234353Sdim#ifdef SIFTR_IPV6 94193323Sed#include <netinet/ip6.h> 95193323Sed#include <netinet6/in6_pcb.h> 96193323Sed#endif /* SIFTR_IPV6 */ 97221345Sdim 98221345Sdim#include <machine/in_cksum.h> 99234353Sdim 100223017Sdim/* 101223017Sdim * Three digit version number refers to X.Y.Z where: 102234353Sdim * X is the major version number 103234353Sdim * Y is bumped to mark backwards incompatible changes 104223017Sdim * Z is bumped to mark backwards compatible changes 105223017Sdim */ 106223017Sdim#define V_MAJOR 1 107223017Sdim#define V_BACKBREAK 2 108223017Sdim#define V_BACKCOMPAT 4 109239462Sdim#define MODVERSION __CONCAT(V_MAJOR, __CONCAT(V_BACKBREAK, V_BACKCOMPAT)) 110239462Sdim#define MODVERSION_STR __XSTRING(V_MAJOR) "." __XSTRING(V_BACKBREAK) "." \ 111239462Sdim __XSTRING(V_BACKCOMPAT) 112239462Sdim 113239462Sdim#define HOOK 0 114223017Sdim#define UNHOOK 1 115239462Sdim#define SIFTR_EXPECTED_MAX_TCP_FLOWS 65536 116223017Sdim#define SYS_NAME "FreeBSD" 117239462Sdim#define PACKET_TAG_SIFTR 100 118239462Sdim#define PACKET_COOKIE_SIFTR 21749576 119239462Sdim#define SIFTR_LOG_FILE_MODE 0644 120239462Sdim#define SIFTR_DISABLE 0 121239462Sdim#define SIFTR_ENABLE 1 122239462Sdim 123239462Sdim/* 124223017Sdim * Hard upper limit on the length of log messages. Bump this up if you add new 125223017Sdim * data fields such that the line length could exceed the below value. 126223017Sdim */ 127223017Sdim#define MAX_LOG_MSG_LEN 200 128223017Sdim/* XXX: Make this a sysctl tunable. */ 129224145Sdim#define SIFTR_ALQ_BUFLEN (1000*MAX_LOG_MSG_LEN) 130234353Sdim 131234353Sdim/* 132224145Sdim * 1 byte for IP version 133239462Sdim * IPv4: src/dst IP (4+4) + src/dst port (2+2) = 12 bytes 134239462Sdim * IPv6: src/dst IP (16+16) + src/dst port (2+2) = 36 bytes 135239462Sdim */ 136239462Sdim#ifdef SIFTR_IPV6 137239462Sdim#define FLOW_KEY_LEN 37 138239462Sdim#else 139234353Sdim#define FLOW_KEY_LEN 13 140224145Sdim#endif 141239462Sdim 142239462Sdim#ifdef SIFTR_IPV6 143239462Sdim#define SIFTR_IPMODE 6 144224145Sdim#else 145224145Sdim#define SIFTR_IPMODE 4 146224145Sdim#endif 147224145Sdim 148224145Sdim/* useful macros */ 149239462Sdim#define CAST_PTR_INT(X) (*((int*)(X))) 150239462Sdim 151239462Sdim#define UPPER_SHORT(X) (((X) & 0xFFFF0000) >> 16) 152239462Sdim#define LOWER_SHORT(X) ((X) & 0x0000FFFF) 153239462Sdim 154239462Sdim#define FIRST_OCTET(X) (((X) & 0xFF000000) >> 24) 155239462Sdim#define SECOND_OCTET(X) (((X) & 0x00FF0000) >> 16) 156239462Sdim#define THIRD_OCTET(X) (((X) & 0x0000FF00) >> 8) 157239462Sdim#define FOURTH_OCTET(X) ((X) & 0x000000FF) 158239462Sdim 159239462Sdimstatic MALLOC_DEFINE(M_SIFTR, "siftr", "dynamic memory used by SIFTR"); 160239462Sdimstatic MALLOC_DEFINE(M_SIFTR_PKTNODE, "siftr_pktnode", 161239462Sdim "SIFTR pkt_node struct"); 162239462Sdimstatic MALLOC_DEFINE(M_SIFTR_HASHNODE, "siftr_hashnode", 163239462Sdim "SIFTR flow_hash_node struct"); 164239462Sdim 165239462Sdim/* Used as links in the pkt manager queue. */ 166234353Sdimstruct pkt_node { 167234353Sdim /* Timestamp of pkt as noted in the pfil hook. */ 168234353Sdim struct timeval tval; 169239462Sdim /* Direction pkt is travelling; either PFIL_IN or PFIL_OUT. */ 170239462Sdim uint8_t direction; 171239462Sdim /* IP version pkt_node relates to; either INP_IPV4 or INP_IPV6. */ 172234353Sdim uint8_t ipver; 173239462Sdim /* Hash of the pkt which triggered the log message. */ 174234353Sdim uint32_t hash; 175234353Sdim /* Local/foreign IP address. */ 176239462Sdim#ifdef SIFTR_IPV6 177239462Sdim uint32_t ip_laddr[4]; 178239462Sdim uint32_t ip_faddr[4]; 179239462Sdim#else 180239462Sdim uint8_t ip_laddr[4]; 181234353Sdim uint8_t ip_faddr[4]; 182234353Sdim#endif 183234353Sdim /* Local TCP port. */ 184234353Sdim uint16_t tcp_localport; 185234353Sdim /* Foreign TCP port. */ 186234353Sdim uint16_t tcp_foreignport; 187234353Sdim /* Congestion Window (bytes). */ 188234353Sdim u_long snd_cwnd; 189234353Sdim /* Sending Window (bytes). */ 190234353Sdim u_long snd_wnd; 191234353Sdim /* Receive Window (bytes). */ 192224145Sdim u_long rcv_wnd; 193224145Sdim /* Unused (was: Bandwidth Controlled Window (bytes)). */ 194224145Sdim u_long snd_bwnd; 195224145Sdim /* Slow Start Threshold (bytes). */ 196226633Sdim u_long snd_ssthresh; 197224145Sdim /* Current state of the TCP FSM. */ 198224145Sdim int conn_state; 199224145Sdim /* Max Segment Size (bytes). */ 200224145Sdim u_int max_seg_size; 201224145Sdim /* 202224145Sdim * Smoothed RTT stored as found in the TCP control block 203224145Sdim * in units of (TCP_RTT_SCALE*hz). 204239462Sdim */ 205239462Sdim int smoothed_rtt; 206239462Sdim /* Is SACK enabled? */ 207223017Sdim u_char sack_enabled; 208223017Sdim /* Window scaling for snd window. */ 209239462Sdim u_char snd_scale; 210239462Sdim /* Window scaling for recv window. */ 211239462Sdim u_char rcv_scale; 212239462Sdim /* TCP control block flags. */ 213239462Sdim u_int flags; 214239462Sdim /* Retransmit timeout length. */ 215239462Sdim int rxt_length; 216239462Sdim /* Size of the TCP send buffer in bytes. */ 217239462Sdim u_int snd_buf_hiwater; 218239462Sdim /* Current num bytes in the send socket buffer. */ 219239462Sdim u_int snd_buf_cc; 220239462Sdim /* Size of the TCP receive buffer in bytes. */ 221239462Sdim u_int rcv_buf_hiwater; 222223017Sdim /* Current num bytes in the receive socket buffer. */ 223224145Sdim u_int rcv_buf_cc; 224239462Sdim /* Number of bytes inflight that we are waiting on ACKs for. */ 225234353Sdim u_int sent_inflight_bytes; 226193323Sed /* Number of segments currently in the reassembly queue. */ 227193323Sed int t_segqlen; 228193323Sed /* Link to next pkt_node in the list. */ 229224145Sdim STAILQ_ENTRY(pkt_node) nodes; 230224145Sdim}; 231226633Sdim 232226633Sdimstruct flow_hash_node 233226633Sdim{ 234226633Sdim uint16_t counter; 235226633Sdim uint8_t key[FLOW_KEY_LEN]; 236226633Sdim LIST_ENTRY(flow_hash_node) nodes; 237226633Sdim}; 238226633Sdim 239226633Sdimstruct siftr_stats 240226633Sdim{ 241226633Sdim /* # TCP pkts seen by the SIFTR PFIL hooks, including any skipped. */ 242226633Sdim uint64_t n_in; 243226633Sdim uint64_t n_out; 244226633Sdim /* # pkts skipped due to failed malloc calls. */ 245234353Sdim uint32_t nskip_in_malloc; 246234353Sdim uint32_t nskip_out_malloc; 247234353Sdim /* # pkts skipped due to failed mtx acquisition. */ 248226633Sdim uint32_t nskip_in_mtx; 249234353Sdim uint32_t nskip_out_mtx; 250234353Sdim /* # pkts skipped due to failed inpcb lookups. */ 251234353Sdim uint32_t nskip_in_inpcb; 252234353Sdim uint32_t nskip_out_inpcb; 253234353Sdim /* # pkts skipped due to failed tcpcb lookups. */ 254234353Sdim uint32_t nskip_in_tcpcb; 255234353Sdim uint32_t nskip_out_tcpcb; 256234353Sdim /* # pkts skipped due to stack reinjection. */ 257239462Sdim uint32_t nskip_in_dejavu; 258239462Sdim uint32_t nskip_out_dejavu; 259239462Sdim}; 260239462Sdim 261224145Sdimstatic DPCPU_DEFINE(struct siftr_stats, ss); 262226633Sdim 263193323Sedstatic volatile unsigned int siftr_exit_pkt_manager_thread = 0; 264193323Sedstatic unsigned int siftr_enabled = 0; 265193323Sedstatic unsigned int siftr_pkts_per_log = 1; 266193323Sedstatic unsigned int siftr_generate_hashes = 0; 267193323Sed/* static unsigned int siftr_binary_log = 0; */ 268223017Sdimstatic char siftr_logfile[PATH_MAX] = "/var/log/siftr.log"; 269224145Sdimstatic u_long siftr_hashmask; 270193323SedSTAILQ_HEAD(pkthead, pkt_node) pkt_queue = STAILQ_HEAD_INITIALIZER(pkt_queue); 271226633SdimLIST_HEAD(listhead, flow_hash_node) *counter_hash; 272226633Sdimstatic int wait_for_pkt; 273226633Sdimstatic struct alq *siftr_alq = NULL; 274226633Sdimstatic struct mtx siftr_pkt_queue_mtx; 275226633Sdimstatic struct mtx siftr_pkt_mgr_mtx; 276226633Sdimstatic struct thread *siftr_pkt_manager_thr = NULL; 277193323Sed/* 278193323Sed * pfil.h defines PFIL_IN as 1 and PFIL_OUT as 2, 279221345Sdim * which we use as an index into this array. 280193323Sed */ 281193323Sedstatic char direction[3] = {'\0', 'i','o'}; 282193323Sed 283234353Sdim/* Required function prototypes. */ 284193323Sedstatic int siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS); 285221345Sdimstatic int siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS); 286224145Sdim 287224145Sdim 288221345Sdim/* Declare the net.inet.siftr sysctl tree and populate it. */ 289224145Sdim 290212904SdimSYSCTL_DECL(_net_inet_siftr); 291221345Sdim 292212904SdimSYSCTL_NODE(_net_inet, OID_AUTO, siftr, CTLFLAG_RW, NULL, 293212904Sdim "siftr related settings"); 294212904Sdim 295212904SdimSYSCTL_PROC(_net_inet_siftr, OID_AUTO, enabled, CTLTYPE_UINT|CTLFLAG_RW, 296212904Sdim &siftr_enabled, 0, &siftr_sysctl_enabled_handler, "IU", 297212904Sdim "switch siftr module operations on/off"); 298226633Sdim 299226633SdimSYSCTL_PROC(_net_inet_siftr, OID_AUTO, logfile, CTLTYPE_STRING|CTLFLAG_RW, 300226633Sdim &siftr_logfile, sizeof(siftr_logfile), &siftr_sysctl_logfile_name_handler, 301193323Sed "A", "file to save siftr log messages to"); 302226633Sdim 303226633SdimSYSCTL_UINT(_net_inet_siftr, OID_AUTO, ppl, CTLFLAG_RW, 304234353Sdim &siftr_pkts_per_log, 1, 305234353Sdim "number of packets between generating a log message"); 306226633Sdim 307226633SdimSYSCTL_UINT(_net_inet_siftr, OID_AUTO, genhashes, CTLFLAG_RW, 308226633Sdim &siftr_generate_hashes, 0, 309234353Sdim "enable packet hash generation"); 310234353Sdim 311226633Sdim/* XXX: TODO 312226633SdimSYSCTL_UINT(_net_inet_siftr, OID_AUTO, binary, CTLFLAG_RW, 313226633Sdim &siftr_binary_log, 0, 314234353Sdim "write log files in binary instead of ascii"); 315234353Sdim*/ 316234353Sdim 317234353Sdim 318234353Sdim/* Begin functions. */ 319234353Sdim 320234353Sdimstatic void 321234353Sdimsiftr_process_pkt(struct pkt_node * pkt_node) 322234353Sdim{ 323234353Sdim struct flow_hash_node *hash_node; 324226633Sdim struct listhead *counter_list; 325226633Sdim struct siftr_stats *ss; 326226633Sdim struct ale *log_buf; 327226633Sdim uint8_t key[FLOW_KEY_LEN]; 328226633Sdim uint8_t found_match, key_offset; 329226633Sdim 330226633Sdim hash_node = NULL; 331226633Sdim ss = DPCPU_PTR(ss); 332226633Sdim found_match = 0; 333226633Sdim key_offset = 1; 334226633Sdim 335224145Sdim /* 336224145Sdim * Create the key that will be used to create a hash index 337224145Sdim * into our hash table. Our key consists of: 338224145Sdim * ipversion, localip, localport, foreignip, foreignport 339226633Sdim */ 340224145Sdim key[0] = pkt_node->ipver; 341212904Sdim memcpy(key + key_offset, &pkt_node->ip_laddr, 342224145Sdim sizeof(pkt_node->ip_laddr)); 343226633Sdim key_offset += sizeof(pkt_node->ip_laddr); 344212904Sdim memcpy(key + key_offset, &pkt_node->tcp_localport, 345226633Sdim sizeof(pkt_node->tcp_localport)); 346226633Sdim key_offset += sizeof(pkt_node->tcp_localport); 347226633Sdim memcpy(key + key_offset, &pkt_node->ip_faddr, 348226633Sdim sizeof(pkt_node->ip_faddr)); 349239462Sdim key_offset += sizeof(pkt_node->ip_faddr); 350239462Sdim memcpy(key + key_offset, &pkt_node->tcp_foreignport, 351239462Sdim sizeof(pkt_node->tcp_foreignport)); 352234353Sdim 353234353Sdim counter_list = counter_hash + 354234353Sdim (hash32_buf(key, sizeof(key), 0) & siftr_hashmask); 355224145Sdim 356226633Sdim /* 357226633Sdim * If the list is not empty i.e. the hash index has 358226633Sdim * been used by another flow previously. 359226633Sdim */ 360226633Sdim if (LIST_FIRST(counter_list) != NULL) { 361226633Sdim /* 362226633Sdim * Loop through the hash nodes in the list. 363226633Sdim * There should normally only be 1 hash node in the list, 364226633Sdim * except if there have been collisions at the hash index 365226633Sdim * computed by hash32_buf(). 366226633Sdim */ 367226633Sdim LIST_FOREACH(hash_node, counter_list, nodes) { 368226633Sdim /* 369226633Sdim * Check if the key for the pkt we are currently 370226633Sdim * processing is the same as the key stored in the 371226633Sdim * hash node we are currently processing. 372226633Sdim * If they are the same, then we've found the 373226633Sdim * hash node that stores the counter for the flow 374226633Sdim * the pkt belongs to. 375226633Sdim */ 376226633Sdim if (memcmp(hash_node->key, key, sizeof(key)) == 0) { 377226633Sdim found_match = 1; 378226633Sdim break; 379226633Sdim } 380226633Sdim } 381226633Sdim } 382226633Sdim 383239462Sdim /* If this flow hash hasn't been seen before or we have a collision. */ 384226633Sdim if (hash_node == NULL || !found_match) { 385226633Sdim /* Create a new hash node to store the flow's counter. */ 386226633Sdim hash_node = malloc(sizeof(struct flow_hash_node), 387193323Sed M_SIFTR_HASHNODE, M_WAITOK); 388223017Sdim 389239462Sdim if (hash_node != NULL) { 390239462Sdim /* Initialise our new hash node list entry. */ 391239462Sdim hash_node->counter = 0; 392239462Sdim memcpy(hash_node->key, key, sizeof(key)); 393239462Sdim LIST_INSERT_HEAD(counter_list, hash_node, nodes); 394239462Sdim } else { 395239462Sdim /* Malloc failed. */ 396239462Sdim if (pkt_node->direction == PFIL_IN) 397239462Sdim ss->nskip_in_malloc++; 398239462Sdim else 399239462Sdim ss->nskip_out_malloc++; 400239462Sdim 401239462Sdim return; 402239462Sdim } 403239462Sdim } else if (siftr_pkts_per_log > 1) { 404239462Sdim /* 405239462Sdim * Taking the remainder of the counter divided 406239462Sdim * by the current value of siftr_pkts_per_log 407239462Sdim * and storing that in counter provides a neat 408239462Sdim * way to modulate the frequency of log 409239462Sdim * messages being written to the log file. 410239462Sdim */ 411239462Sdim hash_node->counter = (hash_node->counter + 1) % 412239462Sdim siftr_pkts_per_log; 413239462Sdim 414234353Sdim /* 415234353Sdim * If we have not seen enough packets since the last time 416234353Sdim * we wrote a log message for this connection, return. 417234353Sdim */ 418234353Sdim if (hash_node->counter > 0) 419234353Sdim return; 420234353Sdim } 421234353Sdim 422239462Sdim log_buf = alq_getn(siftr_alq, MAX_LOG_MSG_LEN, ALQ_WAITOK); 423239462Sdim 424239462Sdim if (log_buf == NULL) 425239462Sdim return; /* Should only happen if the ALQ is shutting down. */ 426223017Sdim 427223017Sdim#ifdef SIFTR_IPV6 428223017Sdim pkt_node->ip_laddr[3] = ntohl(pkt_node->ip_laddr[3]); 429224145Sdim pkt_node->ip_faddr[3] = ntohl(pkt_node->ip_faddr[3]); 430224145Sdim 431234353Sdim if (pkt_node->ipver == INP_IPV6) { /* IPv6 packet */ 432234353Sdim pkt_node->ip_laddr[0] = ntohl(pkt_node->ip_laddr[0]); 433234353Sdim pkt_node->ip_laddr[1] = ntohl(pkt_node->ip_laddr[1]); 434234353Sdim pkt_node->ip_laddr[2] = ntohl(pkt_node->ip_laddr[2]); 435239462Sdim pkt_node->ip_faddr[0] = ntohl(pkt_node->ip_faddr[0]); 436239462Sdim pkt_node->ip_faddr[1] = ntohl(pkt_node->ip_faddr[1]); 437239462Sdim pkt_node->ip_faddr[2] = ntohl(pkt_node->ip_faddr[2]); 438239462Sdim 439239462Sdim /* Construct an IPv6 log message. */ 440239462Sdim log_buf->ae_bytesused = snprintf(log_buf->ae_data, 441234353Sdim MAX_LOG_MSG_LEN, 442224145Sdim "%c,0x%08x,%zd.%06ld,%x:%x:%x:%x:%x:%x:%x:%x,%u,%x:%x:%x:" 443243830Sdim "%x:%x:%x:%x:%x,%u,%ld,%ld,%ld,%ld,%ld,%u,%u,%u,%u,%u,%u," 444223017Sdim "%u,%d,%u,%u,%u,%u,%u,%u\n", 445234353Sdim direction[pkt_node->direction], 446223017Sdim pkt_node->hash, 447239462Sdim pkt_node->tval.tv_sec, 448234353Sdim pkt_node->tval.tv_usec, 449239462Sdim UPPER_SHORT(pkt_node->ip_laddr[0]), 450239462Sdim LOWER_SHORT(pkt_node->ip_laddr[0]), 451239462Sdim UPPER_SHORT(pkt_node->ip_laddr[1]), 452226633Sdim LOWER_SHORT(pkt_node->ip_laddr[1]), 453226633Sdim UPPER_SHORT(pkt_node->ip_laddr[2]), 454224145Sdim LOWER_SHORT(pkt_node->ip_laddr[2]), 455226633Sdim UPPER_SHORT(pkt_node->ip_laddr[3]), 456226633Sdim LOWER_SHORT(pkt_node->ip_laddr[3]), 457224145Sdim ntohs(pkt_node->tcp_localport), 458234353Sdim UPPER_SHORT(pkt_node->ip_faddr[0]), 459234353Sdim LOWER_SHORT(pkt_node->ip_faddr[0]), 460234353Sdim UPPER_SHORT(pkt_node->ip_faddr[1]), 461234353Sdim LOWER_SHORT(pkt_node->ip_faddr[1]), 462234353Sdim UPPER_SHORT(pkt_node->ip_faddr[2]), 463234353Sdim LOWER_SHORT(pkt_node->ip_faddr[2]), 464234353Sdim UPPER_SHORT(pkt_node->ip_faddr[3]), 465234353Sdim LOWER_SHORT(pkt_node->ip_faddr[3]), 466234353Sdim ntohs(pkt_node->tcp_foreignport), 467226633Sdim pkt_node->snd_ssthresh, 468226633Sdim pkt_node->snd_cwnd, 469226633Sdim pkt_node->snd_bwnd, 470234353Sdim pkt_node->snd_wnd, 471234353Sdim pkt_node->rcv_wnd, 472234353Sdim pkt_node->snd_scale, 473234353Sdim pkt_node->rcv_scale, 474234353Sdim pkt_node->conn_state, 475226633Sdim pkt_node->max_seg_size, 476226633Sdim pkt_node->smoothed_rtt, 477234353Sdim pkt_node->sack_enabled, 478234353Sdim pkt_node->flags, 479234353Sdim pkt_node->rxt_length, 480234353Sdim pkt_node->snd_buf_hiwater, 481226633Sdim pkt_node->snd_buf_cc, 482234353Sdim pkt_node->rcv_buf_hiwater, 483234353Sdim pkt_node->rcv_buf_cc, 484223017Sdim pkt_node->sent_inflight_bytes, 485234353Sdim pkt_node->t_segqlen); 486234353Sdim } else { /* IPv4 packet */ 487234353Sdim pkt_node->ip_laddr[0] = FIRST_OCTET(pkt_node->ip_laddr[3]); 488234353Sdim pkt_node->ip_laddr[1] = SECOND_OCTET(pkt_node->ip_laddr[3]); 489234353Sdim pkt_node->ip_laddr[2] = THIRD_OCTET(pkt_node->ip_laddr[3]); 490234353Sdim pkt_node->ip_laddr[3] = FOURTH_OCTET(pkt_node->ip_laddr[3]); 491223017Sdim pkt_node->ip_faddr[0] = FIRST_OCTET(pkt_node->ip_faddr[3]); 492223017Sdim pkt_node->ip_faddr[1] = SECOND_OCTET(pkt_node->ip_faddr[3]); 493223017Sdim pkt_node->ip_faddr[2] = THIRD_OCTET(pkt_node->ip_faddr[3]); 494243830Sdim pkt_node->ip_faddr[3] = FOURTH_OCTET(pkt_node->ip_faddr[3]); 495243830Sdim#endif /* SIFTR_IPV6 */ 496243830Sdim 497223017Sdim /* Construct an IPv4 log message. */ 498223017Sdim log_buf->ae_bytesused = snprintf(log_buf->ae_data, 499223017Sdim MAX_LOG_MSG_LEN, 500224145Sdim "%c,0x%08x,%jd.%06ld,%u.%u.%u.%u,%u,%u.%u.%u.%u,%u,%ld,%ld," 501224145Sdim "%ld,%ld,%ld,%u,%u,%u,%u,%u,%u,%u,%d,%u,%u,%u,%u,%u,%u\n", 502223017Sdim direction[pkt_node->direction], 503223017Sdim pkt_node->hash, 504223017Sdim (intmax_t)pkt_node->tval.tv_sec, 505234353Sdim pkt_node->tval.tv_usec, 506223017Sdim pkt_node->ip_laddr[0], 507234353Sdim pkt_node->ip_laddr[1], 508234353Sdim pkt_node->ip_laddr[2], 509223017Sdim pkt_node->ip_laddr[3], 510223017Sdim ntohs(pkt_node->tcp_localport), 511234353Sdim pkt_node->ip_faddr[0], 512234353Sdim pkt_node->ip_faddr[1], 513223017Sdim pkt_node->ip_faddr[2], 514239462Sdim pkt_node->ip_faddr[3], 515239462Sdim ntohs(pkt_node->tcp_foreignport), 516239462Sdim pkt_node->snd_ssthresh, 517239462Sdim pkt_node->snd_cwnd, 518239462Sdim pkt_node->snd_bwnd, 519239462Sdim pkt_node->snd_wnd, 520239462Sdim pkt_node->rcv_wnd, 521239462Sdim pkt_node->snd_scale, 522239462Sdim pkt_node->rcv_scale, 523239462Sdim pkt_node->conn_state, 524239462Sdim pkt_node->max_seg_size, 525224145Sdim pkt_node->smoothed_rtt, 526243830Sdim pkt_node->sack_enabled, 527243830Sdim pkt_node->flags, 528243830Sdim pkt_node->rxt_length, 529223017Sdim pkt_node->snd_buf_hiwater, 530223017Sdim pkt_node->snd_buf_cc, 531223017Sdim pkt_node->rcv_buf_hiwater, 532223017Sdim pkt_node->rcv_buf_cc, 533234353Sdim pkt_node->sent_inflight_bytes, 534234353Sdim pkt_node->t_segqlen); 535234353Sdim#ifdef SIFTR_IPV6 536234353Sdim } 537234353Sdim#endif 538239462Sdim 539239462Sdim alq_post_flags(siftr_alq, log_buf, 0); 540239462Sdim} 541239462Sdim 542239462Sdim 543239462Sdimstatic void 544239462Sdimsiftr_pkt_manager_thread(void *arg) 545239462Sdim{ 546239462Sdim STAILQ_HEAD(pkthead, pkt_node) tmp_pkt_queue = 547239462Sdim STAILQ_HEAD_INITIALIZER(tmp_pkt_queue); 548239462Sdim struct pkt_node *pkt_node, *pkt_node_temp; 549239462Sdim uint8_t draining; 550239462Sdim 551239462Sdim draining = 2; 552239462Sdim 553239462Sdim mtx_lock(&siftr_pkt_mgr_mtx); 554239462Sdim 555239462Sdim /* draining == 0 when queue has been flushed and it's safe to exit. */ 556239462Sdim while (draining) { 557239462Sdim /* 558239462Sdim * Sleep until we are signalled to wake because thread has 559239462Sdim * been told to exit or until 1 tick has passed. 560234353Sdim */ 561234353Sdim mtx_sleep(&wait_for_pkt, &siftr_pkt_mgr_mtx, PWAIT, "pktwait", 562234353Sdim 1); 563239462Sdim 564239462Sdim /* Gain exclusive access to the pkt_node queue. */ 565239462Sdim mtx_lock(&siftr_pkt_queue_mtx); 566234353Sdim 567234353Sdim /* 568234353Sdim * Move pkt_queue to tmp_pkt_queue, which leaves 569234353Sdim * pkt_queue empty and ready to receive more pkt_nodes. 570234353Sdim */ 571234353Sdim STAILQ_CONCAT(&tmp_pkt_queue, &pkt_queue); 572234353Sdim 573234353Sdim /* 574234353Sdim * We've finished making changes to the list. Unlock it 575239462Sdim * so the pfil hooks can continue queuing pkt_nodes. 576239462Sdim */ 577239462Sdim mtx_unlock(&siftr_pkt_queue_mtx); 578239462Sdim 579239462Sdim /* 580239462Sdim * We can't hold a mutex whilst calling siftr_process_pkt 581239462Sdim * because ALQ might sleep waiting for buffer space. 582226633Sdim */ 583224145Sdim mtx_unlock(&siftr_pkt_mgr_mtx); 584224145Sdim 585224145Sdim /* Flush all pkt_nodes to the log file. */ 586224145Sdim STAILQ_FOREACH_SAFE(pkt_node, &tmp_pkt_queue, nodes, 587224145Sdim pkt_node_temp) { 588224145Sdim siftr_process_pkt(pkt_node); 589224145Sdim STAILQ_REMOVE_HEAD(&tmp_pkt_queue, nodes); 590224145Sdim free(pkt_node, M_SIFTR_PKTNODE); 591224145Sdim } 592224145Sdim 593224145Sdim KASSERT(STAILQ_EMPTY(&tmp_pkt_queue), 594224145Sdim ("SIFTR tmp_pkt_queue not empty after flush")); 595224145Sdim 596234353Sdim mtx_lock(&siftr_pkt_mgr_mtx); 597234353Sdim 598234353Sdim /* 599234353Sdim * If siftr_exit_pkt_manager_thread gets set during the window 600234353Sdim * where we are draining the tmp_pkt_queue above, there might 601239462Sdim * still be pkts in pkt_queue that need to be drained. 602234353Sdim * Allow one further iteration to occur after 603234353Sdim * siftr_exit_pkt_manager_thread has been set to ensure 604234353Sdim * pkt_queue is completely empty before we kill the thread. 605234353Sdim * 606234353Sdim * siftr_exit_pkt_manager_thread is set only after the pfil 607239462Sdim * hooks have been removed, so only 1 extra iteration 608234353Sdim * is needed to drain the queue. 609234353Sdim */ 610234353Sdim if (siftr_exit_pkt_manager_thread) 611234353Sdim draining--; 612234353Sdim } 613234353Sdim 614234353Sdim mtx_unlock(&siftr_pkt_mgr_mtx); 615234353Sdim 616234353Sdim /* Calls wakeup on this thread's struct thread ptr. */ 617234353Sdim kthread_exit(); 618234353Sdim} 619234353Sdim 620234353Sdim 621234353Sdimstatic uint32_t 622234353Sdimhash_pkt(struct mbuf *m, uint32_t offset) 623234353Sdim{ 624234353Sdim uint32_t hash; 625234353Sdim 626223017Sdim hash = 0; 627223017Sdim 628224145Sdim while (m != NULL && offset > m->m_len) { 629234353Sdim /* 630234353Sdim * The IP packet payload does not start in this mbuf, so 631234353Sdim * need to figure out which mbuf it starts in and what offset 632234353Sdim * into the mbuf's data region the payload starts at. 633234353Sdim */ 634234353Sdim offset -= m->m_len; 635234353Sdim m = m->m_next; 636234353Sdim } 637223017Sdim 638193323Sed while (m != NULL) { 639193323Sed /* Ensure there is data in the mbuf */ 640193323Sed if ((m->m_len - offset) > 0) 641 hash = hash32_buf(m->m_data + offset, 642 m->m_len - offset, hash); 643 644 m = m->m_next; 645 offset = 0; 646 } 647 648 return (hash); 649} 650 651 652/* 653 * Check if a given mbuf has the SIFTR mbuf tag. If it does, log the fact that 654 * it's a reinjected packet and return. If it doesn't, tag the mbuf and return. 655 * Return value >0 means the caller should skip processing this mbuf. 656 */ 657static inline int 658siftr_chkreinject(struct mbuf *m, int dir, struct siftr_stats *ss) 659{ 660 if (m_tag_locate(m, PACKET_COOKIE_SIFTR, PACKET_TAG_SIFTR, NULL) 661 != NULL) { 662 if (dir == PFIL_IN) 663 ss->nskip_in_dejavu++; 664 else 665 ss->nskip_out_dejavu++; 666 667 return (1); 668 } else { 669 struct m_tag *tag = m_tag_alloc(PACKET_COOKIE_SIFTR, 670 PACKET_TAG_SIFTR, 0, M_NOWAIT); 671 if (tag == NULL) { 672 if (dir == PFIL_IN) 673 ss->nskip_in_malloc++; 674 else 675 ss->nskip_out_malloc++; 676 677 return (1); 678 } 679 680 m_tag_prepend(m, tag); 681 } 682 683 return (0); 684} 685 686 687/* 688 * Look up an inpcb for a packet. Return the inpcb pointer if found, or NULL 689 * otherwise. 690 */ 691static inline struct inpcb * 692siftr_findinpcb(int ipver, struct ip *ip, struct mbuf *m, uint16_t sport, 693 uint16_t dport, int dir, struct siftr_stats *ss) 694{ 695 struct inpcb *inp; 696 697 /* We need the tcbinfo lock. */ 698 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 699 700 if (dir == PFIL_IN) 701 inp = (ipver == INP_IPV4 ? 702 in_pcblookup(&V_tcbinfo, ip->ip_src, sport, ip->ip_dst, 703 dport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif) 704 : 705#ifdef SIFTR_IPV6 706 in6_pcblookup(&V_tcbinfo, 707 &((struct ip6_hdr *)ip)->ip6_src, sport, 708 &((struct ip6_hdr *)ip)->ip6_dst, dport, INPLOOKUP_RLOCKPCB, 709 m->m_pkthdr.rcvif) 710#else 711 NULL 712#endif 713 ); 714 715 else 716 inp = (ipver == INP_IPV4 ? 717 in_pcblookup(&V_tcbinfo, ip->ip_dst, dport, ip->ip_src, 718 sport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif) 719 : 720#ifdef SIFTR_IPV6 721 in6_pcblookup(&V_tcbinfo, 722 &((struct ip6_hdr *)ip)->ip6_dst, dport, 723 &((struct ip6_hdr *)ip)->ip6_src, sport, INPLOOKUP_RLOCKPCB, 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 } 737 738 return (inp); 739} 740 741 742static inline void 743siftr_siftdata(struct pkt_node *pn, struct inpcb *inp, struct tcpcb *tp, 744 int ipver, int dir, int inp_locally_locked) 745{ 746#ifdef SIFTR_IPV6 747 if (ipver == INP_IPV4) { 748 pn->ip_laddr[3] = inp->inp_laddr.s_addr; 749 pn->ip_faddr[3] = inp->inp_faddr.s_addr; 750#else 751 *((uint32_t *)pn->ip_laddr) = inp->inp_laddr.s_addr; 752 *((uint32_t *)pn->ip_faddr) = inp->inp_faddr.s_addr; 753#endif 754#ifdef SIFTR_IPV6 755 } else { 756 pn->ip_laddr[0] = inp->in6p_laddr.s6_addr32[0]; 757 pn->ip_laddr[1] = inp->in6p_laddr.s6_addr32[1]; 758 pn->ip_laddr[2] = inp->in6p_laddr.s6_addr32[2]; 759 pn->ip_laddr[3] = inp->in6p_laddr.s6_addr32[3]; 760 pn->ip_faddr[0] = inp->in6p_faddr.s6_addr32[0]; 761 pn->ip_faddr[1] = inp->in6p_faddr.s6_addr32[1]; 762 pn->ip_faddr[2] = inp->in6p_faddr.s6_addr32[2]; 763 pn->ip_faddr[3] = inp->in6p_faddr.s6_addr32[3]; 764 } 765#endif 766 pn->tcp_localport = inp->inp_lport; 767 pn->tcp_foreignport = inp->inp_fport; 768 pn->snd_cwnd = tp->snd_cwnd; 769 pn->snd_wnd = tp->snd_wnd; 770 pn->rcv_wnd = tp->rcv_wnd; 771 pn->snd_bwnd = 0; /* Unused, kept for compat. */ 772 pn->snd_ssthresh = tp->snd_ssthresh; 773 pn->snd_scale = tp->snd_scale; 774 pn->rcv_scale = tp->rcv_scale; 775 pn->conn_state = tp->t_state; 776 pn->max_seg_size = tp->t_maxseg; 777 pn->smoothed_rtt = tp->t_srtt; 778 pn->sack_enabled = (tp->t_flags & TF_SACK_PERMIT) != 0; 779 pn->flags = tp->t_flags; 780 pn->rxt_length = tp->t_rxtcur; 781 pn->snd_buf_hiwater = inp->inp_socket->so_snd.sb_hiwat; 782 pn->snd_buf_cc = inp->inp_socket->so_snd.sb_cc; 783 pn->rcv_buf_hiwater = inp->inp_socket->so_rcv.sb_hiwat; 784 pn->rcv_buf_cc = inp->inp_socket->so_rcv.sb_cc; 785 pn->sent_inflight_bytes = tp->snd_max - tp->snd_una; 786 pn->t_segqlen = tp->t_segqlen; 787 788 /* We've finished accessing the tcb so release the lock. */ 789 if (inp_locally_locked) 790 INP_RUNLOCK(inp); 791 792 pn->ipver = ipver; 793 pn->direction = dir; 794 795 /* 796 * Significantly more accurate than using getmicrotime(), but slower! 797 * Gives true microsecond resolution at the expense of a hit to 798 * maximum pps throughput processing when SIFTR is loaded and enabled. 799 */ 800 microtime(&pn->tval); 801} 802 803 804/* 805 * pfil hook that is called for each IPv4 packet making its way through the 806 * stack in either direction. 807 * The pfil subsystem holds a non-sleepable mutex somewhere when 808 * calling our hook function, so we can't sleep at all. 809 * It's very important to use the M_NOWAIT flag with all function calls 810 * that support it so that they won't sleep, otherwise you get a panic. 811 */ 812static int 813siftr_chkpkt(void *arg, struct mbuf **m, struct ifnet *ifp, int dir, 814 struct inpcb *inp) 815{ 816 struct pkt_node *pn; 817 struct ip *ip; 818 struct tcphdr *th; 819 struct tcpcb *tp; 820 struct siftr_stats *ss; 821 unsigned int ip_hl; 822 int inp_locally_locked; 823 824 inp_locally_locked = 0; 825 ss = DPCPU_PTR(ss); 826 827 /* 828 * m_pullup is not required here because ip_{input|output} 829 * already do the heavy lifting for us. 830 */ 831 832 ip = mtod(*m, struct ip *); 833 834 /* Only continue processing if the packet is TCP. */ 835 if (ip->ip_p != IPPROTO_TCP) 836 goto ret; 837 838 /* 839 * If a kernel subsystem reinjects packets into the stack, our pfil 840 * hook will be called multiple times for the same packet. 841 * Make sure we only process unique packets. 842 */ 843 if (siftr_chkreinject(*m, dir, ss)) 844 goto ret; 845 846 if (dir == PFIL_IN) 847 ss->n_in++; 848 else 849 ss->n_out++; 850 851 /* 852 * Create a tcphdr struct starting at the correct offset 853 * in the IP packet. ip->ip_hl gives the ip header length 854 * in 4-byte words, so multiply it to get the size in bytes. 855 */ 856 ip_hl = (ip->ip_hl << 2); 857 th = (struct tcphdr *)((caddr_t)ip + ip_hl); 858 859 /* 860 * If the pfil hooks don't provide a pointer to the 861 * inpcb, we need to find it ourselves and lock it. 862 */ 863 if (!inp) { 864 /* Find the corresponding inpcb for this pkt. */ 865 inp = siftr_findinpcb(INP_IPV4, ip, *m, th->th_sport, 866 th->th_dport, dir, ss); 867 868 if (inp == NULL) 869 goto ret; 870 else 871 inp_locally_locked = 1; 872 } 873 874 INP_LOCK_ASSERT(inp); 875 876 /* Find the TCP control block that corresponds with this packet */ 877 tp = intotcpcb(inp); 878 879 /* 880 * If we can't find the TCP control block (happens occasionaly for a 881 * packet sent during the shutdown phase of a TCP connection), 882 * or we're in the timewait state, bail 883 */ 884 if (tp == NULL || inp->inp_flags & INP_TIMEWAIT) { 885 if (dir == PFIL_IN) 886 ss->nskip_in_tcpcb++; 887 else 888 ss->nskip_out_tcpcb++; 889 890 goto inp_unlock; 891 } 892 893 pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO); 894 895 if (pn == NULL) { 896 if (dir == PFIL_IN) 897 ss->nskip_in_malloc++; 898 else 899 ss->nskip_out_malloc++; 900 901 goto inp_unlock; 902 } 903 904 siftr_siftdata(pn, inp, tp, INP_IPV4, dir, inp_locally_locked); 905 906 if (siftr_generate_hashes) { 907 if ((*m)->m_pkthdr.csum_flags & CSUM_TCP) { 908 /* 909 * For outbound packets, the TCP checksum isn't 910 * calculated yet. This is a problem for our packet 911 * hashing as the receiver will calc a different hash 912 * to ours if we don't include the correct TCP checksum 913 * in the bytes being hashed. To work around this 914 * problem, we manually calc the TCP checksum here in 915 * software. We unset the CSUM_TCP flag so the lower 916 * layers don't recalc it. 917 */ 918 (*m)->m_pkthdr.csum_flags &= ~CSUM_TCP; 919 920 /* 921 * Calculate the TCP checksum in software and assign 922 * to correct TCP header field, which will follow the 923 * packet mbuf down the stack. The trick here is that 924 * tcp_output() sets th->th_sum to the checksum of the 925 * pseudo header for us already. Because of the nature 926 * of the checksumming algorithm, we can sum over the 927 * entire IP payload (i.e. TCP header and data), which 928 * will include the already calculated pseduo header 929 * checksum, thus giving us the complete TCP checksum. 930 * 931 * To put it in simple terms, if checksum(1,2,3,4)=10, 932 * then checksum(1,2,3,4,5) == checksum(10,5). 933 * This property is what allows us to "cheat" and 934 * checksum only the IP payload which has the TCP 935 * th_sum field populated with the pseudo header's 936 * checksum, and not need to futz around checksumming 937 * pseudo header bytes and TCP header/data in one hit. 938 * Refer to RFC 1071 for more info. 939 * 940 * NB: in_cksum_skip(struct mbuf *m, int len, int skip) 941 * in_cksum_skip 2nd argument is NOT the number of 942 * bytes to read from the mbuf at "skip" bytes offset 943 * from the start of the mbuf (very counter intuitive!). 944 * The number of bytes to read is calculated internally 945 * by the function as len-skip i.e. to sum over the IP 946 * payload (TCP header + data) bytes, it is INCORRECT 947 * to call the function like this: 948 * in_cksum_skip(at, ip->ip_len - offset, offset) 949 * Rather, it should be called like this: 950 * in_cksum_skip(at, ip->ip_len, offset) 951 * which means read "ip->ip_len - offset" bytes from 952 * the mbuf cluster "at" at offset "offset" bytes from 953 * the beginning of the "at" mbuf's data pointer. 954 */ 955 th->th_sum = in_cksum_skip(*m, ntohs(ip->ip_len), 956 ip_hl); 957 } 958 959 /* 960 * XXX: Having to calculate the checksum in software and then 961 * hash over all bytes is really inefficient. Would be nice to 962 * find a way to create the hash and checksum in the same pass 963 * over the bytes. 964 */ 965 pn->hash = hash_pkt(*m, ip_hl); 966 } 967 968 mtx_lock(&siftr_pkt_queue_mtx); 969 STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes); 970 mtx_unlock(&siftr_pkt_queue_mtx); 971 goto ret; 972 973inp_unlock: 974 if (inp_locally_locked) 975 INP_RUNLOCK(inp); 976 977ret: 978 /* Returning 0 ensures pfil will not discard the pkt */ 979 return (0); 980} 981 982 983#ifdef SIFTR_IPV6 984static int 985siftr_chkpkt6(void *arg, struct mbuf **m, struct ifnet *ifp, int dir, 986 struct inpcb *inp) 987{ 988 struct pkt_node *pn; 989 struct ip6_hdr *ip6; 990 struct tcphdr *th; 991 struct tcpcb *tp; 992 struct siftr_stats *ss; 993 unsigned int ip6_hl; 994 int inp_locally_locked; 995 996 inp_locally_locked = 0; 997 ss = DPCPU_PTR(ss); 998 999 /* 1000 * m_pullup is not required here because ip6_{input|output} 1001 * already do the heavy lifting for us. 1002 */ 1003 1004 ip6 = mtod(*m, struct ip6_hdr *); 1005 1006 /* 1007 * Only continue processing if the packet is TCP 1008 * XXX: We should follow the next header fields 1009 * as shown on Pg 6 RFC 2460, but right now we'll 1010 * only check pkts that have no extension headers. 1011 */ 1012 if (ip6->ip6_nxt != IPPROTO_TCP) 1013 goto ret6; 1014 1015 /* 1016 * If a kernel subsystem reinjects packets into the stack, our pfil 1017 * hook will be called multiple times for the same packet. 1018 * Make sure we only process unique packets. 1019 */ 1020 if (siftr_chkreinject(*m, dir, ss)) 1021 goto ret6; 1022 1023 if (dir == PFIL_IN) 1024 ss->n_in++; 1025 else 1026 ss->n_out++; 1027 1028 ip6_hl = sizeof(struct ip6_hdr); 1029 1030 /* 1031 * Create a tcphdr struct starting at the correct offset 1032 * in the ipv6 packet. ip->ip_hl gives the ip header length 1033 * in 4-byte words, so multiply it to get the size in bytes. 1034 */ 1035 th = (struct tcphdr *)((caddr_t)ip6 + ip6_hl); 1036 1037 /* 1038 * For inbound packets, the pfil hooks don't provide a pointer to the 1039 * inpcb, so we need to find it ourselves and lock it. 1040 */ 1041 if (!inp) { 1042 /* Find the corresponding inpcb for this pkt. */ 1043 inp = siftr_findinpcb(INP_IPV6, (struct ip *)ip6, *m, 1044 th->th_sport, th->th_dport, dir, ss); 1045 1046 if (inp == NULL) 1047 goto ret6; 1048 else 1049 inp_locally_locked = 1; 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 goto inp_unlock6; 1067 } 1068 1069 pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO); 1070 1071 if (pn == NULL) { 1072 if (dir == PFIL_IN) 1073 ss->nskip_in_malloc++; 1074 else 1075 ss->nskip_out_malloc++; 1076 1077 goto inp_unlock6; 1078 } 1079 1080 siftr_siftdata(pn, inp, tp, INP_IPV6, dir, inp_locally_locked); 1081 1082 /* XXX: Figure out how to generate hashes for IPv6 packets. */ 1083 1084 mtx_lock(&siftr_pkt_queue_mtx); 1085 STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes); 1086 mtx_unlock(&siftr_pkt_queue_mtx); 1087 goto ret6; 1088 1089inp_unlock6: 1090 if (inp_locally_locked) 1091 INP_RUNLOCK(inp); 1092 1093ret6: 1094 /* Returning 0 ensures pfil will not discard the pkt. */ 1095 return (0); 1096} 1097#endif /* #ifdef SIFTR_IPV6 */ 1098 1099 1100static int 1101siftr_pfil(int action) 1102{ 1103 struct pfil_head *pfh_inet; 1104#ifdef SIFTR_IPV6 1105 struct pfil_head *pfh_inet6; 1106#endif 1107 VNET_ITERATOR_DECL(vnet_iter); 1108 1109 VNET_LIST_RLOCK(); 1110 VNET_FOREACH(vnet_iter) { 1111 CURVNET_SET(vnet_iter); 1112 pfh_inet = pfil_head_get(PFIL_TYPE_AF, AF_INET); 1113#ifdef SIFTR_IPV6 1114 pfh_inet6 = pfil_head_get(PFIL_TYPE_AF, AF_INET6); 1115#endif 1116 1117 if (action == HOOK) { 1118 pfil_add_hook(siftr_chkpkt, NULL, 1119 PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet); 1120#ifdef SIFTR_IPV6 1121 pfil_add_hook(siftr_chkpkt6, NULL, 1122 PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet6); 1123#endif 1124 } else if (action == UNHOOK) { 1125 pfil_remove_hook(siftr_chkpkt, NULL, 1126 PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet); 1127#ifdef SIFTR_IPV6 1128 pfil_remove_hook(siftr_chkpkt6, NULL, 1129 PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet6); 1130#endif 1131 } 1132 CURVNET_RESTORE(); 1133 } 1134 VNET_LIST_RUNLOCK(); 1135 1136 return (0); 1137} 1138 1139 1140static int 1141siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS) 1142{ 1143 struct alq *new_alq; 1144 int error; 1145 1146 if (req->newptr == NULL) 1147 goto skip; 1148 1149 /* If old filename and new filename are different. */ 1150 if (strncmp(siftr_logfile, (char *)req->newptr, PATH_MAX)) { 1151 1152 error = alq_open(&new_alq, req->newptr, curthread->td_ucred, 1153 SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0); 1154 1155 /* Bail if unable to create new alq. */ 1156 if (error) 1157 return (1); 1158 1159 /* 1160 * If disabled, siftr_alq == NULL so we simply close 1161 * the alq as we've proved it can be opened. 1162 * If enabled, close the existing alq and switch the old 1163 * for the new. 1164 */ 1165 if (siftr_alq == NULL) 1166 alq_close(new_alq); 1167 else { 1168 alq_close(siftr_alq); 1169 siftr_alq = new_alq; 1170 } 1171 } 1172 1173skip: 1174 return (sysctl_handle_string(oidp, arg1, arg2, req)); 1175} 1176 1177 1178static int 1179siftr_manage_ops(uint8_t action) 1180{ 1181 struct siftr_stats totalss; 1182 struct timeval tval; 1183 struct flow_hash_node *counter, *tmp_counter; 1184 struct sbuf *s; 1185 int i, key_index, ret, error; 1186 uint32_t bytes_to_write, total_skipped_pkts; 1187 uint16_t lport, fport; 1188 uint8_t *key, ipver; 1189 1190#ifdef SIFTR_IPV6 1191 uint32_t laddr[4]; 1192 uint32_t faddr[4]; 1193#else 1194 uint8_t laddr[4]; 1195 uint8_t faddr[4]; 1196#endif 1197 1198 error = 0; 1199 total_skipped_pkts = 0; 1200 1201 /* Init an autosizing sbuf that initially holds 200 chars. */ 1202 if ((s = sbuf_new(NULL, NULL, 200, SBUF_AUTOEXTEND)) == NULL) 1203 return (-1); 1204 1205 if (action == SIFTR_ENABLE) { 1206 /* 1207 * Create our alq 1208 * XXX: We should abort if alq_open fails! 1209 */ 1210 alq_open(&siftr_alq, siftr_logfile, curthread->td_ucred, 1211 SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0); 1212 1213 STAILQ_INIT(&pkt_queue); 1214 1215 DPCPU_ZERO(ss); 1216 1217 siftr_exit_pkt_manager_thread = 0; 1218 1219 ret = kthread_add(&siftr_pkt_manager_thread, NULL, NULL, 1220 &siftr_pkt_manager_thr, RFNOWAIT, 0, 1221 "siftr_pkt_manager_thr"); 1222 1223 siftr_pfil(HOOK); 1224 1225 microtime(&tval); 1226 1227 sbuf_printf(s, 1228 "enable_time_secs=%jd\tenable_time_usecs=%06ld\t" 1229 "siftrver=%s\thz=%u\ttcp_rtt_scale=%u\tsysname=%s\t" 1230 "sysver=%u\tipmode=%u\n", 1231 (intmax_t)tval.tv_sec, tval.tv_usec, MODVERSION_STR, hz, 1232 TCP_RTT_SCALE, SYS_NAME, __FreeBSD_version, SIFTR_IPMODE); 1233 1234 sbuf_finish(s); 1235 alq_writen(siftr_alq, sbuf_data(s), sbuf_len(s), ALQ_WAITOK); 1236 1237 } else if (action == SIFTR_DISABLE && siftr_pkt_manager_thr != NULL) { 1238 /* 1239 * Remove the pfil hook functions. All threads currently in 1240 * the hook functions are allowed to exit before siftr_pfil() 1241 * returns. 1242 */ 1243 siftr_pfil(UNHOOK); 1244 1245 /* This will block until the pkt manager thread unlocks it. */ 1246 mtx_lock(&siftr_pkt_mgr_mtx); 1247 1248 /* Tell the pkt manager thread that it should exit now. */ 1249 siftr_exit_pkt_manager_thread = 1; 1250 1251 /* 1252 * Wake the pkt_manager thread so it realises that 1253 * siftr_exit_pkt_manager_thread == 1 and exits gracefully. 1254 * The wakeup won't be delivered until we unlock 1255 * siftr_pkt_mgr_mtx so this isn't racy. 1256 */ 1257 wakeup(&wait_for_pkt); 1258 1259 /* Wait for the pkt_manager thread to exit. */ 1260 mtx_sleep(siftr_pkt_manager_thr, &siftr_pkt_mgr_mtx, PWAIT, 1261 "thrwait", 0); 1262 1263 siftr_pkt_manager_thr = NULL; 1264 mtx_unlock(&siftr_pkt_mgr_mtx); 1265 1266 totalss.n_in = DPCPU_VARSUM(ss, n_in); 1267 totalss.n_out = DPCPU_VARSUM(ss, n_out); 1268 totalss.nskip_in_malloc = DPCPU_VARSUM(ss, nskip_in_malloc); 1269 totalss.nskip_out_malloc = DPCPU_VARSUM(ss, nskip_out_malloc); 1270 totalss.nskip_in_mtx = DPCPU_VARSUM(ss, nskip_in_mtx); 1271 totalss.nskip_out_mtx = DPCPU_VARSUM(ss, nskip_out_mtx); 1272 totalss.nskip_in_tcpcb = DPCPU_VARSUM(ss, nskip_in_tcpcb); 1273 totalss.nskip_out_tcpcb = DPCPU_VARSUM(ss, nskip_out_tcpcb); 1274 totalss.nskip_in_inpcb = DPCPU_VARSUM(ss, nskip_in_inpcb); 1275 totalss.nskip_out_inpcb = DPCPU_VARSUM(ss, nskip_out_inpcb); 1276 1277 total_skipped_pkts = totalss.nskip_in_malloc + 1278 totalss.nskip_out_malloc + totalss.nskip_in_mtx + 1279 totalss.nskip_out_mtx + totalss.nskip_in_tcpcb + 1280 totalss.nskip_out_tcpcb + totalss.nskip_in_inpcb + 1281 totalss.nskip_out_inpcb; 1282 1283 microtime(&tval); 1284 1285 sbuf_printf(s, 1286 "disable_time_secs=%jd\tdisable_time_usecs=%06ld\t" 1287 "num_inbound_tcp_pkts=%ju\tnum_outbound_tcp_pkts=%ju\t" 1288 "total_tcp_pkts=%ju\tnum_inbound_skipped_pkts_malloc=%u\t" 1289 "num_outbound_skipped_pkts_malloc=%u\t" 1290 "num_inbound_skipped_pkts_mtx=%u\t" 1291 "num_outbound_skipped_pkts_mtx=%u\t" 1292 "num_inbound_skipped_pkts_tcpcb=%u\t" 1293 "num_outbound_skipped_pkts_tcpcb=%u\t" 1294 "num_inbound_skipped_pkts_inpcb=%u\t" 1295 "num_outbound_skipped_pkts_inpcb=%u\t" 1296 "total_skipped_tcp_pkts=%u\tflow_list=", 1297 (intmax_t)tval.tv_sec, 1298 tval.tv_usec, 1299 (uintmax_t)totalss.n_in, 1300 (uintmax_t)totalss.n_out, 1301 (uintmax_t)(totalss.n_in + totalss.n_out), 1302 totalss.nskip_in_malloc, 1303 totalss.nskip_out_malloc, 1304 totalss.nskip_in_mtx, 1305 totalss.nskip_out_mtx, 1306 totalss.nskip_in_tcpcb, 1307 totalss.nskip_out_tcpcb, 1308 totalss.nskip_in_inpcb, 1309 totalss.nskip_out_inpcb, 1310 total_skipped_pkts); 1311 1312 /* 1313 * Iterate over the flow hash, printing a summary of each 1314 * flow seen and freeing any malloc'd memory. 1315 * The hash consists of an array of LISTs (man 3 queue). 1316 */ 1317 for (i = 0; i < siftr_hashmask; i++) { 1318 LIST_FOREACH_SAFE(counter, counter_hash + i, nodes, 1319 tmp_counter) { 1320 key = counter->key; 1321 key_index = 1; 1322 1323 ipver = key[0]; 1324 1325 memcpy(laddr, key + key_index, sizeof(laddr)); 1326 key_index += sizeof(laddr); 1327 memcpy(&lport, key + key_index, sizeof(lport)); 1328 key_index += sizeof(lport); 1329 memcpy(faddr, key + key_index, sizeof(faddr)); 1330 key_index += sizeof(faddr); 1331 memcpy(&fport, key + key_index, sizeof(fport)); 1332 1333#ifdef SIFTR_IPV6 1334 laddr[3] = ntohl(laddr[3]); 1335 faddr[3] = ntohl(faddr[3]); 1336 1337 if (ipver == INP_IPV6) { 1338 laddr[0] = ntohl(laddr[0]); 1339 laddr[1] = ntohl(laddr[1]); 1340 laddr[2] = ntohl(laddr[2]); 1341 faddr[0] = ntohl(faddr[0]); 1342 faddr[1] = ntohl(faddr[1]); 1343 faddr[2] = ntohl(faddr[2]); 1344 1345 sbuf_printf(s, 1346 "%x:%x:%x:%x:%x:%x:%x:%x;%u-" 1347 "%x:%x:%x:%x:%x:%x:%x:%x;%u,", 1348 UPPER_SHORT(laddr[0]), 1349 LOWER_SHORT(laddr[0]), 1350 UPPER_SHORT(laddr[1]), 1351 LOWER_SHORT(laddr[1]), 1352 UPPER_SHORT(laddr[2]), 1353 LOWER_SHORT(laddr[2]), 1354 UPPER_SHORT(laddr[3]), 1355 LOWER_SHORT(laddr[3]), 1356 ntohs(lport), 1357 UPPER_SHORT(faddr[0]), 1358 LOWER_SHORT(faddr[0]), 1359 UPPER_SHORT(faddr[1]), 1360 LOWER_SHORT(faddr[1]), 1361 UPPER_SHORT(faddr[2]), 1362 LOWER_SHORT(faddr[2]), 1363 UPPER_SHORT(faddr[3]), 1364 LOWER_SHORT(faddr[3]), 1365 ntohs(fport)); 1366 } else { 1367 laddr[0] = FIRST_OCTET(laddr[3]); 1368 laddr[1] = SECOND_OCTET(laddr[3]); 1369 laddr[2] = THIRD_OCTET(laddr[3]); 1370 laddr[3] = FOURTH_OCTET(laddr[3]); 1371 faddr[0] = FIRST_OCTET(faddr[3]); 1372 faddr[1] = SECOND_OCTET(faddr[3]); 1373 faddr[2] = THIRD_OCTET(faddr[3]); 1374 faddr[3] = FOURTH_OCTET(faddr[3]); 1375#endif 1376 sbuf_printf(s, 1377 "%u.%u.%u.%u;%u-%u.%u.%u.%u;%u,", 1378 laddr[0], 1379 laddr[1], 1380 laddr[2], 1381 laddr[3], 1382 ntohs(lport), 1383 faddr[0], 1384 faddr[1], 1385 faddr[2], 1386 faddr[3], 1387 ntohs(fport)); 1388#ifdef SIFTR_IPV6 1389 } 1390#endif 1391 1392 free(counter, M_SIFTR_HASHNODE); 1393 } 1394 1395 LIST_INIT(counter_hash + i); 1396 } 1397 1398 sbuf_printf(s, "\n"); 1399 sbuf_finish(s); 1400 1401 i = 0; 1402 do { 1403 bytes_to_write = min(SIFTR_ALQ_BUFLEN, sbuf_len(s)-i); 1404 alq_writen(siftr_alq, sbuf_data(s)+i, bytes_to_write, ALQ_WAITOK); 1405 i += bytes_to_write; 1406 } while (i < sbuf_len(s)); 1407 1408 alq_close(siftr_alq); 1409 siftr_alq = NULL; 1410 } 1411 1412 sbuf_delete(s); 1413 1414 /* 1415 * XXX: Should be using ret to check if any functions fail 1416 * and set error appropriately 1417 */ 1418 1419 return (error); 1420} 1421 1422 1423static int 1424siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS) 1425{ 1426 if (req->newptr == NULL) 1427 goto skip; 1428 1429 /* If the value passed in isn't 0 or 1, return an error. */ 1430 if (CAST_PTR_INT(req->newptr) != 0 && CAST_PTR_INT(req->newptr) != 1) 1431 return (1); 1432 1433 /* If we are changing state (0 to 1 or 1 to 0). */ 1434 if (CAST_PTR_INT(req->newptr) != siftr_enabled ) 1435 if (siftr_manage_ops(CAST_PTR_INT(req->newptr))) { 1436 siftr_manage_ops(SIFTR_DISABLE); 1437 return (1); 1438 } 1439 1440skip: 1441 return (sysctl_handle_int(oidp, arg1, arg2, req)); 1442} 1443 1444 1445static void 1446siftr_shutdown_handler(void *arg) 1447{ 1448 siftr_manage_ops(SIFTR_DISABLE); 1449} 1450 1451 1452/* 1453 * Module is being unloaded or machine is shutting down. Take care of cleanup. 1454 */ 1455static int 1456deinit_siftr(void) 1457{ 1458 /* Cleanup. */ 1459 siftr_manage_ops(SIFTR_DISABLE); 1460 hashdestroy(counter_hash, M_SIFTR, siftr_hashmask); 1461 mtx_destroy(&siftr_pkt_queue_mtx); 1462 mtx_destroy(&siftr_pkt_mgr_mtx); 1463 1464 return (0); 1465} 1466 1467 1468/* 1469 * Module has just been loaded into the kernel. 1470 */ 1471static int 1472init_siftr(void) 1473{ 1474 EVENTHANDLER_REGISTER(shutdown_pre_sync, siftr_shutdown_handler, NULL, 1475 SHUTDOWN_PRI_FIRST); 1476 1477 /* Initialise our flow counter hash table. */ 1478 counter_hash = hashinit(SIFTR_EXPECTED_MAX_TCP_FLOWS, M_SIFTR, 1479 &siftr_hashmask); 1480 1481 mtx_init(&siftr_pkt_queue_mtx, "siftr_pkt_queue_mtx", NULL, MTX_DEF); 1482 mtx_init(&siftr_pkt_mgr_mtx, "siftr_pkt_mgr_mtx", NULL, MTX_DEF); 1483 1484 /* Print message to the user's current terminal. */ 1485 uprintf("\nStatistical Information For TCP Research (SIFTR) %s\n" 1486 " http://caia.swin.edu.au/urp/newtcp\n\n", 1487 MODVERSION_STR); 1488 1489 return (0); 1490} 1491 1492 1493/* 1494 * This is the function that is called to load and unload the module. 1495 * When the module is loaded, this function is called once with 1496 * "what" == MOD_LOAD 1497 * When the module is unloaded, this function is called twice with 1498 * "what" = MOD_QUIESCE first, followed by "what" = MOD_UNLOAD second 1499 * When the system is shut down e.g. CTRL-ALT-DEL or using the shutdown command, 1500 * this function is called once with "what" = MOD_SHUTDOWN 1501 * When the system is shut down, the handler isn't called until the very end 1502 * of the shutdown sequence i.e. after the disks have been synced. 1503 */ 1504static int 1505siftr_load_handler(module_t mod, int what, void *arg) 1506{ 1507 int ret; 1508 1509 switch (what) { 1510 case MOD_LOAD: 1511 ret = init_siftr(); 1512 break; 1513 1514 case MOD_QUIESCE: 1515 case MOD_SHUTDOWN: 1516 ret = deinit_siftr(); 1517 break; 1518 1519 case MOD_UNLOAD: 1520 ret = 0; 1521 break; 1522 1523 default: 1524 ret = EINVAL; 1525 break; 1526 } 1527 1528 return (ret); 1529} 1530 1531 1532static moduledata_t siftr_mod = { 1533 .name = "siftr", 1534 .evhand = siftr_load_handler, 1535}; 1536 1537/* 1538 * Param 1: name of the kernel module 1539 * Param 2: moduledata_t struct containing info about the kernel module 1540 * and the execution entry point for the module 1541 * Param 3: From sysinit_sub_id enumeration in /usr/include/sys/kernel.h 1542 * Defines the module initialisation order 1543 * Param 4: From sysinit_elem_order enumeration in /usr/include/sys/kernel.h 1544 * Defines the initialisation order of this kld relative to others 1545 * within the same subsystem as defined by param 3 1546 */ 1547DECLARE_MODULE(siftr, siftr_mod, SI_SUB_SMP, SI_ORDER_ANY); 1548MODULE_DEPEND(siftr, alq, 1, 1, 1); 1549MODULE_VERSION(siftr, MODVERSION); 1550