sys/netinet/siftr.c

   1 /*-
   2  * Copyright (c) 2007-2009
   3  *      Swinburne University of Technology, Melbourne, Australia.
   4  * Copyright (c) 2009-2010, The FreeBSD Foundation
   5  * All rights reserved.
   6  *
   7  * Portions of this software were developed at the Centre for Advanced
   8  * Internet Architectures, Swinburne University of Technology, Melbourne,
   9  * Australia by Lawrence Stewart under sponsorship from the FreeBSD Foundation.
  10  *
  11  * Redistribution and use in source and binary forms, with or without
  12  * modification, are permitted provided that the following conditions
  13  * are met:
  14  * 1. Redistributions of source code must retain the above copyright
  15  *    notice, this list of conditions and the following disclaimer.
  16  * 2. Redistributions in binary form must reproduce the above copyright
  17  *    notice, this list of conditions and the following disclaimer in the
  18  *    documentation and/or other materials provided with the distribution.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
  21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
  24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  30  * SUCH DAMAGE.
  31  */
  32
  33 /******************************************************
  34  * Statistical Information For TCP Research (SIFTR)
  35  *
  36  * A FreeBSD kernel module that adds very basic intrumentation to the
  37  * TCP stack, allowing internal stats to be recorded to a log file
  38  * for experimental, debugging and performance analysis purposes.
  39  *
  40  * SIFTR was first released in 2007 by James Healy and Lawrence Stewart whilst
  41  * working on the NewTCP research project at Swinburne University of
  42  * Technology's Centre for Advanced Internet Architectures, Melbourne,
  43  * Australia, which was made possible in part by a grant from the Cisco
  44  * University Research Program Fund at Community Foundation Silicon Valley.
  45  * 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: September 2010
  59  ******************************************************/
  60
  61 #include <sys/cdefs.h>
  62 __FBSDID("$FreeBSD$");
  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    4
 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
 159 static MALLOC_DEFINE(M_SIFTR, "siftr", "dynamic memory used by SIFTR");
 160 static MALLOC_DEFINE(M_SIFTR_PKTNODE, "siftr_pktnode",
 161     "SIFTR pkt_node struct");
 162 static MALLOC_DEFINE(M_SIFTR_HASHNODE, "siftr_hashnode",
 163     "SIFTR flow_hash_node struct");
 164
 165 /* Used as links in the pkt manager queue. */
 166 struct pkt_node {
 167         /* Timestamp of pkt as noted in the pfil hook. */
 168         struct timeval          tval;
 169         /* Direction pkt is travelling; either PFIL_IN or PFIL_OUT. */
 170         uint8_t                 direction;
 171         /* IP version pkt_node relates to; either INP_IPV4 or INP_IPV6. */
 172         uint8_t                 ipver;
 173         /* Hash of the pkt which triggered the log message. */
 174         uint32_t                hash;
 175         /* Local/foreign IP address. */
 176 #ifdef SIFTR_IPV6
 177         uint32_t                ip_laddr[4];
 178         uint32_t                ip_faddr[4];
 179 #else
 180         uint8_t                 ip_laddr[4];
 181         uint8_t                 ip_faddr[4];
 182 #endif
 183         /* Local TCP port. */
 184         uint16_t                tcp_localport;
 185         /* Foreign TCP port. */
 186         uint16_t                tcp_foreignport;
 187         /* Congestion Window (bytes). */
 188         u_long                  snd_cwnd;
 189         /* Sending Window (bytes). */
 190         u_long                  snd_wnd;
 191         /* Receive Window (bytes). */
 192         u_long                  rcv_wnd;
 193         /* Unused (was: Bandwidth Controlled Window (bytes)). */
 194         u_long                  snd_bwnd;
 195         /* Slow Start Threshold (bytes). */
 196         u_long                  snd_ssthresh;
 197         /* Current state of the TCP FSM. */
 198         int                     conn_state;
 199         /* Max Segment Size (bytes). */
 200         u_int                   max_seg_size;
 201         /*
 202          * Smoothed RTT stored as found in the TCP control block
 203          * in units of (TCP_RTT_SCALE*hz).
 204          */
 205         int                     smoothed_rtt;
 206         /* Is SACK enabled? */
 207         u_char                  sack_enabled;
 208         /* Window scaling for snd window. */
 209         u_char                  snd_scale;
 210         /* Window scaling for recv window. */
 211         u_char                  rcv_scale;
 212         /* TCP control block flags. */
 213         u_int                   flags;
 214         /* Retransmit timeout length. */
 215         int                     rxt_length;
 216         /* Size of the TCP send buffer in bytes. */
 217         u_int                   snd_buf_hiwater;
 218         /* Current num bytes in the send socket buffer. */
 219         u_int                   snd_buf_cc;
 220         /* Size of the TCP receive buffer in bytes. */
 221         u_int                   rcv_buf_hiwater;
 222         /* Current num bytes in the receive socket buffer. */
 223         u_int                   rcv_buf_cc;
 224         /* Number of bytes inflight that we are waiting on ACKs for. */
 225         u_int                   sent_inflight_bytes;
 226         /* Number of segments currently in the reassembly queue. */
 227         int                     t_segqlen;
 228         /* Flowid for the connection. */
 229         u_int                   flowid;
 230         /* Flow type for the connection. */
 231         u_int                   flowtype;
 232         /* Link to next pkt_node in the list. */
 233         STAILQ_ENTRY(pkt_node)  nodes;
 234 };
 235
 236 struct flow_hash_node
 237 {
 238         uint16_t counter;
 239         uint8_t key[FLOW_KEY_LEN];
 240         LIST_ENTRY(flow_hash_node) nodes;
 241 };
 242
 243 struct siftr_stats
 244 {
 245         /* # TCP pkts seen by the SIFTR PFIL hooks, including any skipped. */
 246         uint64_t n_in;
 247         uint64_t n_out;
 248         /* # pkts skipped due to failed malloc calls. */
 249         uint32_t nskip_in_malloc;
 250         uint32_t nskip_out_malloc;
 251         /* # pkts skipped due to failed mtx acquisition. */
 252         uint32_t nskip_in_mtx;
 253         uint32_t nskip_out_mtx;
 254         /* # pkts skipped due to failed inpcb lookups. */
 255         uint32_t nskip_in_inpcb;
 256         uint32_t nskip_out_inpcb;
 257         /* # pkts skipped due to failed tcpcb lookups. */
 258         uint32_t nskip_in_tcpcb;
 259         uint32_t nskip_out_tcpcb;
 260         /* # pkts skipped due to stack reinjection. */
 261         uint32_t nskip_in_dejavu;
 262         uint32_t nskip_out_dejavu;
 263 };
 264
 265 static DPCPU_DEFINE(struct siftr_stats, ss);
 266
 267 static volatile unsigned int siftr_exit_pkt_manager_thread = 0;
 268 static unsigned int siftr_enabled = 0;
 269 static unsigned int siftr_pkts_per_log = 1;
 270 static unsigned int siftr_generate_hashes = 0;
 271 /* static unsigned int siftr_binary_log = 0; */
 272 static char siftr_logfile[PATH_MAX] = "/var/log/siftr.log";
 273 static char siftr_logfile_shadow[PATH_MAX] = "/var/log/siftr.log";
 274 static u_long siftr_hashmask;
 275 STAILQ_HEAD(pkthead, pkt_node) pkt_queue = STAILQ_HEAD_INITIALIZER(pkt_queue);
 276 LIST_HEAD(listhead, flow_hash_node) *counter_hash;
 277 static int wait_for_pkt;
 278 static struct alq *siftr_alq = NULL;
 279 static struct mtx siftr_pkt_queue_mtx;
 280 static struct mtx siftr_pkt_mgr_mtx;
 281 static struct thread *siftr_pkt_manager_thr = NULL;
 282 /*
 283  * pfil.h defines PFIL_IN as 1 and PFIL_OUT as 2,
 284  * which we use as an index into this array.
 285  */
 286 static char direction[3] = {'\0', 'i','o'};
 287
 288 /* Required function prototypes. */
 289 static int siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS);
 290 static int siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS);
 291
 292
 293 /* Declare the net.inet.siftr sysctl tree and populate it. */
 294
 295 SYSCTL_DECL(_net_inet_siftr);
 296
 297 SYSCTL_NODE(_net_inet, OID_AUTO, siftr, CTLFLAG_RW, NULL,
 298     "siftr related settings");
 299
 300 SYSCTL_PROC(_net_inet_siftr, OID_AUTO, enabled, CTLTYPE_UINT|CTLFLAG_RW,
 301     &siftr_enabled, 0, &siftr_sysctl_enabled_handler, "IU",
 302     "switch siftr module operations on/off");
 303
 304 SYSCTL_PROC(_net_inet_siftr, OID_AUTO, logfile, CTLTYPE_STRING|CTLFLAG_RW,
 305     &siftr_logfile_shadow, sizeof(siftr_logfile_shadow), &siftr_sysctl_logfile_name_handler,
 306     "A", "file to save siftr log messages to");
 307
 308 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, ppl, CTLFLAG_RW,
 309     &siftr_pkts_per_log, 1,
 310     "number of packets between generating a log message");
 311
 312 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, genhashes, CTLFLAG_RW,
 313     &siftr_generate_hashes, 0,
 314     "enable packet hash generation");
 315
 316 /* XXX: TODO
 317 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, binary, CTLFLAG_RW,
 318     &siftr_binary_log, 0,
 319     "write log files in binary instead of ascii");
 320 */
 321
 322
 323 /* Begin functions. */
 324
 325 static void
 326 siftr_process_pkt(struct pkt_node * pkt_node)
 327 {
 328         struct flow_hash_node *hash_node;
 329         struct listhead *counter_list;
 330         struct siftr_stats *ss;
 331         struct ale *log_buf;
 332         uint8_t key[FLOW_KEY_LEN];
 333         uint8_t found_match, key_offset;
 334
 335         hash_node = NULL;
 336         ss = DPCPU_PTR(ss);
 337         found_match = 0;
 338         key_offset = 1;
 339
 340         /*
 341          * Create the key that will be used to create a hash index
 342          * into our hash table. Our key consists of:
 343          * ipversion, localip, localport, foreignip, foreignport
 344          */
 345         key[0] = pkt_node->ipver;
 346         memcpy(key + key_offset, &pkt_node->ip_laddr,
 347             sizeof(pkt_node->ip_laddr));
 348         key_offset += sizeof(pkt_node->ip_laddr);
 349         memcpy(key + key_offset, &pkt_node->tcp_localport,
 350             sizeof(pkt_node->tcp_localport));
 351         key_offset += sizeof(pkt_node->tcp_localport);
 352         memcpy(key + key_offset, &pkt_node->ip_faddr,
 353             sizeof(pkt_node->ip_faddr));
 354         key_offset += sizeof(pkt_node->ip_faddr);
 355         memcpy(key + key_offset, &pkt_node->tcp_foreignport,
 356             sizeof(pkt_node->tcp_foreignport));
 357
 358         counter_list = counter_hash +
 359             (hash32_buf(key, sizeof(key), 0) & siftr_hashmask);
 360
 361         /*
 362          * If the list is not empty i.e. the hash index has
 363          * been used by another flow previously.
 364          */
 365         if (LIST_FIRST(counter_list) != NULL) {
 366                 /*
 367                  * Loop through the hash nodes in the list.
 368                  * There should normally only be 1 hash node in the list,
 369                  * except if there have been collisions at the hash index
 370                  * computed by hash32_buf().
 371                  */
 372                 LIST_FOREACH(hash_node, counter_list, nodes) {
 373                         /*
 374                          * Check if the key for the pkt we are currently
 375                          * processing is the same as the key stored in the
 376                          * hash node we are currently processing.
 377                          * If they are the same, then we've found the
 378                          * hash node that stores the counter for the flow
 379                          * the pkt belongs to.
 380                          */
 381                         if (memcmp(hash_node->key, key, sizeof(key)) == 0) {
 382                                 found_match = 1;
 383                                 break;
 384                         }
 385                 }
 386         }
 387
 388         /* If this flow hash hasn't been seen before or we have a collision. */
 389         if (hash_node == NULL || !found_match) {
 390                 /* Create a new hash node to store the flow's counter. */
 391                 hash_node = malloc(sizeof(struct flow_hash_node),
 392                     M_SIFTR_HASHNODE, M_WAITOK);
 393
 394                 if (hash_node != NULL) {
 395                         /* Initialise our new hash node list entry. */
 396                         hash_node->counter = 0;
 397                         memcpy(hash_node->key, key, sizeof(key));
 398                         LIST_INSERT_HEAD(counter_list, hash_node, nodes);
 399                 } else {
 400                         /* Malloc failed. */
 401                         if (pkt_node->direction == PFIL_IN)
 402                                 ss->nskip_in_malloc++;
 403                         else
 404                                 ss->nskip_out_malloc++;
 405
 406                         return;
 407                 }
 408         } else if (siftr_pkts_per_log > 1) {
 409                 /*
 410                  * Taking the remainder of the counter divided
 411                  * by the current value of siftr_pkts_per_log
 412                  * and storing that in counter provides a neat
 413                  * way to modulate the frequency of log
 414                  * messages being written to the log file.
 415                  */
 416                 hash_node->counter = (hash_node->counter + 1) %
 417                     siftr_pkts_per_log;
 418
 419                 /*
 420                  * If we have not seen enough packets since the last time
 421                  * we wrote a log message for this connection, return.
 422                  */
 423                 if (hash_node->counter > 0)
 424                         return;
 425         }
 426
 427         log_buf = alq_getn(siftr_alq, MAX_LOG_MSG_LEN, ALQ_WAITOK);
 428
 429         if (log_buf == NULL)
 430                 return; /* Should only happen if the ALQ is shutting down. */
 431
 432 #ifdef SIFTR_IPV6
 433         pkt_node->ip_laddr[3] = ntohl(pkt_node->ip_laddr[3]);
 434         pkt_node->ip_faddr[3] = ntohl(pkt_node->ip_faddr[3]);
 435
 436         if (pkt_node->ipver == INP_IPV6) { /* IPv6 packet */
 437                 pkt_node->ip_laddr[0] = ntohl(pkt_node->ip_laddr[0]);
 438                 pkt_node->ip_laddr[1] = ntohl(pkt_node->ip_laddr[1]);
 439                 pkt_node->ip_laddr[2] = ntohl(pkt_node->ip_laddr[2]);
 440                 pkt_node->ip_faddr[0] = ntohl(pkt_node->ip_faddr[0]);
 441                 pkt_node->ip_faddr[1] = ntohl(pkt_node->ip_faddr[1]);
 442                 pkt_node->ip_faddr[2] = ntohl(pkt_node->ip_faddr[2]);
 443
 444                 /* Construct an IPv6 log message. */
 445                 log_buf->ae_bytesused = snprintf(log_buf->ae_data,
 446                     MAX_LOG_MSG_LEN,
 447                     "%c,0x%08x,%zd.%06ld,%x:%x:%x:%x:%x:%x:%x:%x,%u,%x:%x:%x:"
 448                     "%x:%x:%x:%x:%x,%u,%ld,%ld,%ld,%ld,%ld,%u,%u,%u,%u,%u,%u,"
 449                     "%u,%d,%u,%u,%u,%u,%u,%u,%u,%u\n",
 450                     direction[pkt_node->direction],
 451                     pkt_node->hash,
 452                     pkt_node->tval.tv_sec,
 453                     pkt_node->tval.tv_usec,
 454                     UPPER_SHORT(pkt_node->ip_laddr[0]),
 455                     LOWER_SHORT(pkt_node->ip_laddr[0]),
 456                     UPPER_SHORT(pkt_node->ip_laddr[1]),
 457                     LOWER_SHORT(pkt_node->ip_laddr[1]),
 458                     UPPER_SHORT(pkt_node->ip_laddr[2]),
 459                     LOWER_SHORT(pkt_node->ip_laddr[2]),
 460                     UPPER_SHORT(pkt_node->ip_laddr[3]),
 461                     LOWER_SHORT(pkt_node->ip_laddr[3]),
 462                     ntohs(pkt_node->tcp_localport),
 463                     UPPER_SHORT(pkt_node->ip_faddr[0]),
 464                     LOWER_SHORT(pkt_node->ip_faddr[0]),
 465                     UPPER_SHORT(pkt_node->ip_faddr[1]),
 466                     LOWER_SHORT(pkt_node->ip_faddr[1]),
 467                     UPPER_SHORT(pkt_node->ip_faddr[2]),
 468                     LOWER_SHORT(pkt_node->ip_faddr[2]),
 469                     UPPER_SHORT(pkt_node->ip_faddr[3]),
 470                     LOWER_SHORT(pkt_node->ip_faddr[3]),
 471                     ntohs(pkt_node->tcp_foreignport),
 472                     pkt_node->snd_ssthresh,
 473                     pkt_node->snd_cwnd,
 474                     pkt_node->snd_bwnd,
 475                     pkt_node->snd_wnd,
 476                     pkt_node->rcv_wnd,
 477                     pkt_node->snd_scale,
 478                     pkt_node->rcv_scale,
 479                     pkt_node->conn_state,
 480                     pkt_node->max_seg_size,
 481                     pkt_node->smoothed_rtt,
 482                     pkt_node->sack_enabled,
 483                     pkt_node->flags,
 484                     pkt_node->rxt_length,
 485                     pkt_node->snd_buf_hiwater,
 486                     pkt_node->snd_buf_cc,
 487                     pkt_node->rcv_buf_hiwater,
 488                     pkt_node->rcv_buf_cc,
 489                     pkt_node->sent_inflight_bytes,
 490                     pkt_node->t_segqlen,
 491                     pkt_node->flowid,
 492                     pkt_node->flowtype);
 493         } else { /* IPv4 packet */
 494                 pkt_node->ip_laddr[0] = FIRST_OCTET(pkt_node->ip_laddr[3]);
 495                 pkt_node->ip_laddr[1] = SECOND_OCTET(pkt_node->ip_laddr[3]);
 496                 pkt_node->ip_laddr[2] = THIRD_OCTET(pkt_node->ip_laddr[3]);
 497                 pkt_node->ip_laddr[3] = FOURTH_OCTET(pkt_node->ip_laddr[3]);
 498                 pkt_node->ip_faddr[0] = FIRST_OCTET(pkt_node->ip_faddr[3]);
 499                 pkt_node->ip_faddr[1] = SECOND_OCTET(pkt_node->ip_faddr[3]);
 500                 pkt_node->ip_faddr[2] = THIRD_OCTET(pkt_node->ip_faddr[3]);
 501                 pkt_node->ip_faddr[3] = FOURTH_OCTET(pkt_node->ip_faddr[3]);
 502 #endif /* SIFTR_IPV6 */
 503
 504                 /* Construct an IPv4 log message. */
 505                 log_buf->ae_bytesused = snprintf(log_buf->ae_data,
 506                     MAX_LOG_MSG_LEN,
 507                     "%c,0x%08x,%jd.%06ld,%u.%u.%u.%u,%u,%u.%u.%u.%u,%u,%ld,%ld,"
 508                     "%ld,%ld,%ld,%u,%u,%u,%u,%u,%u,%u,%d,%u,%u,%u,%u,%u,%u,%u,%u\n",
 509                     direction[pkt_node->direction],
 510                     pkt_node->hash,
 511                     (intmax_t)pkt_node->tval.tv_sec,
 512                     pkt_node->tval.tv_usec,
 513                     pkt_node->ip_laddr[0],
 514                     pkt_node->ip_laddr[1],
 515                     pkt_node->ip_laddr[2],
 516                     pkt_node->ip_laddr[3],
 517                     ntohs(pkt_node->tcp_localport),
 518                     pkt_node->ip_faddr[0],
 519                     pkt_node->ip_faddr[1],
 520                     pkt_node->ip_faddr[2],
 521                     pkt_node->ip_faddr[3],
 522                     ntohs(pkt_node->tcp_foreignport),
 523                     pkt_node->snd_ssthresh,
 524                     pkt_node->snd_cwnd,
 525                     pkt_node->snd_bwnd,
 526                     pkt_node->snd_wnd,
 527                     pkt_node->rcv_wnd,
 528                     pkt_node->snd_scale,
 529                     pkt_node->rcv_scale,
 530                     pkt_node->conn_state,
 531                     pkt_node->max_seg_size,
 532                     pkt_node->smoothed_rtt,
 533                     pkt_node->sack_enabled,
 534                     pkt_node->flags,
 535                     pkt_node->rxt_length,
 536                     pkt_node->snd_buf_hiwater,
 537                     pkt_node->snd_buf_cc,
 538                     pkt_node->rcv_buf_hiwater,
 539                     pkt_node->rcv_buf_cc,
 540                     pkt_node->sent_inflight_bytes,
 541                     pkt_node->t_segqlen,
 542                     pkt_node->flowid,
 543                     pkt_node->flowtype);
 544 #ifdef SIFTR_IPV6
 545         }
 546 #endif
 547
 548         alq_post_flags(siftr_alq, log_buf, 0);
 549 }
 550
 551
 552 static void
 553 siftr_pkt_manager_thread(void *arg)
 554 {
 555         STAILQ_HEAD(pkthead, pkt_node) tmp_pkt_queue =
 556             STAILQ_HEAD_INITIALIZER(tmp_pkt_queue);
 557         struct pkt_node *pkt_node, *pkt_node_temp;
 558         uint8_t draining;
 559
 560         draining = 2;
 561
 562         mtx_lock(&siftr_pkt_mgr_mtx);
 563
 564         /* draining == 0 when queue has been flushed and it's safe to exit. */
 565         while (draining) {
 566                 /*
 567                  * Sleep until we are signalled to wake because thread has
 568                  * been told to exit or until 1 tick has passed.
 569                  */
 570                 mtx_sleep(&wait_for_pkt, &siftr_pkt_mgr_mtx, PWAIT, "pktwait",
 571                     1);
 572
 573                 /* Gain exclusive access to the pkt_node queue. */
 574                 mtx_lock(&siftr_pkt_queue_mtx);
 575
 576                 /*
 577                  * Move pkt_queue to tmp_pkt_queue, which leaves
 578                  * pkt_queue empty and ready to receive more pkt_nodes.
 579                  */
 580                 STAILQ_CONCAT(&tmp_pkt_queue, &pkt_queue);
 581
 582                 /*
 583                  * We've finished making changes to the list. Unlock it
 584                  * so the pfil hooks can continue queuing pkt_nodes.
 585                  */
 586                 mtx_unlock(&siftr_pkt_queue_mtx);
 587
 588                 /*
 589                  * We can't hold a mutex whilst calling siftr_process_pkt
 590                  * because ALQ might sleep waiting for buffer space.
 591                  */
 592                 mtx_unlock(&siftr_pkt_mgr_mtx);
 593
 594                 /* Flush all pkt_nodes to the log file. */
 595                 STAILQ_FOREACH_SAFE(pkt_node, &tmp_pkt_queue, nodes,
 596                     pkt_node_temp) {
 597                         siftr_process_pkt(pkt_node);
 598                         STAILQ_REMOVE_HEAD(&tmp_pkt_queue, nodes);
 599                         free(pkt_node, M_SIFTR_PKTNODE);
 600                 }
 601
 602                 KASSERT(STAILQ_EMPTY(&tmp_pkt_queue),
 603                     ("SIFTR tmp_pkt_queue not empty after flush"));
 604
 605                 mtx_lock(&siftr_pkt_mgr_mtx);
 606
 607                 /*
 608                  * If siftr_exit_pkt_manager_thread gets set during the window
 609                  * where we are draining the tmp_pkt_queue above, there might
 610                  * still be pkts in pkt_queue that need to be drained.
 611                  * Allow one further iteration to occur after
 612                  * siftr_exit_pkt_manager_thread has been set to ensure
 613                  * pkt_queue is completely empty before we kill the thread.
 614                  *
 615                  * siftr_exit_pkt_manager_thread is set only after the pfil
 616                  * hooks have been removed, so only 1 extra iteration
 617                  * is needed to drain the queue.
 618                  */
 619                 if (siftr_exit_pkt_manager_thread)
 620                         draining--;
 621         }
 622
 623         mtx_unlock(&siftr_pkt_mgr_mtx);
 624
 625         /* Calls wakeup on this thread's struct thread ptr. */
 626         kthread_exit();
 627 }
 628
 629
 630 static uint32_t
 631 hash_pkt(struct mbuf *m, uint32_t offset)
 632 {
 633         uint32_t hash;
 634
 635         hash = 0;
 636
 637         while (m != NULL && offset > m->m_len) {
 638                 /*
 639                  * The IP packet payload does not start in this mbuf, so
 640                  * need to figure out which mbuf it starts in and what offset
 641                  * into the mbuf's data region the payload starts at.
 642                  */
 643                 offset -= m->m_len;
 644                 m = m->m_next;
 645         }
 646
 647         while (m != NULL) {
 648                 /* Ensure there is data in the mbuf */
 649                 if ((m->m_len - offset) > 0)
 650                         hash = hash32_buf(m->m_data + offset,
 651                             m->m_len - offset, hash);
 652
 653                 m = m->m_next;
 654                 offset = 0;
 655         }
 656
 657         return (hash);
 658 }
 659
 660
 661 /*
 662  * Check if a given mbuf has the SIFTR mbuf tag. If it does, log the fact that
 663  * it's a reinjected packet and return. If it doesn't, tag the mbuf and return.
 664  * Return value >0 means the caller should skip processing this mbuf.
 665  */
 666 static inline int
 667 siftr_chkreinject(struct mbuf *m, int dir, struct siftr_stats *ss)
 668 {
 669         if (m_tag_locate(m, PACKET_COOKIE_SIFTR, PACKET_TAG_SIFTR, NULL)
 670             != NULL) {
 671                 if (dir == PFIL_IN)
 672                         ss->nskip_in_dejavu++;
 673                 else
 674                         ss->nskip_out_dejavu++;
 675
 676                 return (1);
 677         } else {
 678                 struct m_tag *tag = m_tag_alloc(PACKET_COOKIE_SIFTR,
 679                     PACKET_TAG_SIFTR, 0, M_NOWAIT);
 680                 if (tag == NULL) {
 681                         if (dir == PFIL_IN)
 682                                 ss->nskip_in_malloc++;
 683                         else
 684                                 ss->nskip_out_malloc++;
 685
 686                         return (1);
 687                 }
 688
 689                 m_tag_prepend(m, tag);
 690         }
 691
 692         return (0);
 693 }
 694
 695
 696 /*
 697  * Look up an inpcb for a packet. Return the inpcb pointer if found, or NULL
 698  * otherwise.
 699  */
 700 static inline struct inpcb *
 701 siftr_findinpcb(int ipver, struct ip *ip, struct mbuf *m, uint16_t sport,
 702     uint16_t dport, int dir, struct siftr_stats *ss)
 703 {
 704         struct inpcb *inp;
 705
 706         /* We need the tcbinfo lock. */
 707         INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
 708
 709         if (dir == PFIL_IN)
 710                 inp = (ipver == INP_IPV4 ?
 711                     in_pcblookup(&V_tcbinfo, ip->ip_src, sport, ip->ip_dst,
 712                     dport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif)
 713                     :
 714 #ifdef SIFTR_IPV6
 715                     in6_pcblookup(&V_tcbinfo,
 716                     &((struct ip6_hdr *)ip)->ip6_src, sport,
 717                     &((struct ip6_hdr *)ip)->ip6_dst, dport, INPLOOKUP_RLOCKPCB,
 718                     m->m_pkthdr.rcvif)
 719 #else
 720                     NULL
 721 #endif
 722                     );
 723
 724         else
 725                 inp = (ipver == INP_IPV4 ?
 726                     in_pcblookup(&V_tcbinfo, ip->ip_dst, dport, ip->ip_src,
 727                     sport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif)
 728                     :
 729 #ifdef SIFTR_IPV6
 730                     in6_pcblookup(&V_tcbinfo,
 731                     &((struct ip6_hdr *)ip)->ip6_dst, dport,
 732                     &((struct ip6_hdr *)ip)->ip6_src, sport, INPLOOKUP_RLOCKPCB,
 733                     m->m_pkthdr.rcvif)
 734 #else
 735                     NULL
 736 #endif
 737                     );
 738
 739         /* If we can't find the inpcb, bail. */
 740         if (inp == NULL) {
 741                 if (dir == PFIL_IN)
 742                         ss->nskip_in_inpcb++;
 743                 else
 744                         ss->nskip_out_inpcb++;
 745         }
 746
 747         return (inp);
 748 }
 749
 750
 751 static inline void
 752 siftr_siftdata(struct pkt_node *pn, struct inpcb *inp, struct tcpcb *tp,
 753     int ipver, int dir, int inp_locally_locked)
 754 {
 755 #ifdef SIFTR_IPV6
 756         if (ipver == INP_IPV4) {
 757                 pn->ip_laddr[3] = inp->inp_laddr.s_addr;
 758                 pn->ip_faddr[3] = inp->inp_faddr.s_addr;
 759 #else
 760                 *((uint32_t *)pn->ip_laddr) = inp->inp_laddr.s_addr;
 761                 *((uint32_t *)pn->ip_faddr) = inp->inp_faddr.s_addr;
 762 #endif
 763 #ifdef SIFTR_IPV6
 764         } else {
 765                 pn->ip_laddr[0] = inp->in6p_laddr.s6_addr32[0];
 766                 pn->ip_laddr[1] = inp->in6p_laddr.s6_addr32[1];
 767                 pn->ip_laddr[2] = inp->in6p_laddr.s6_addr32[2];
 768                 pn->ip_laddr[3] = inp->in6p_laddr.s6_addr32[3];
 769                 pn->ip_faddr[0] = inp->in6p_faddr.s6_addr32[0];
 770                 pn->ip_faddr[1] = inp->in6p_faddr.s6_addr32[1];
 771                 pn->ip_faddr[2] = inp->in6p_faddr.s6_addr32[2];
 772                 pn->ip_faddr[3] = inp->in6p_faddr.s6_addr32[3];
 773         }
 774 #endif
 775         pn->tcp_localport = inp->inp_lport;
 776         pn->tcp_foreignport = inp->inp_fport;
 777         pn->snd_cwnd = tp->snd_cwnd;
 778         pn->snd_wnd = tp->snd_wnd;
 779         pn->rcv_wnd = tp->rcv_wnd;
 780         pn->snd_bwnd = 0;               /* Unused, kept for compat. */
 781         pn->snd_ssthresh = tp->snd_ssthresh;
 782         pn->snd_scale = tp->snd_scale;
 783         pn->rcv_scale = tp->rcv_scale;
 784         pn->conn_state = tp->t_state;
 785         pn->max_seg_size = tp->t_maxseg;
 786         pn->smoothed_rtt = tp->t_srtt;
 787         pn->sack_enabled = (tp->t_flags & TF_SACK_PERMIT) != 0;
 788         pn->flags = tp->t_flags;
 789         pn->rxt_length = tp->t_rxtcur;
 790         pn->snd_buf_hiwater = inp->inp_socket->so_snd.sb_hiwat;
 791         pn->snd_buf_cc = inp->inp_socket->so_snd.sb_cc;
 792         pn->rcv_buf_hiwater = inp->inp_socket->so_rcv.sb_hiwat;
 793         pn->rcv_buf_cc = inp->inp_socket->so_rcv.sb_cc;
 794         pn->sent_inflight_bytes = tp->snd_max - tp->snd_una;
 795         pn->t_segqlen = tp->t_segqlen;
 796         pn->flowid = inp->inp_flowid;
 797         pn->flowtype = inp->inp_flowtype;
 798
 799         /* We've finished accessing the tcb so release the lock. */
 800         if (inp_locally_locked)
 801                 INP_RUNLOCK(inp);
 802
 803         pn->ipver = ipver;
 804         pn->direction = dir;
 805
 806         /*
 807          * Significantly more accurate than using getmicrotime(), but slower!
 808          * Gives true microsecond resolution at the expense of a hit to
 809          * maximum pps throughput processing when SIFTR is loaded and enabled.
 810          */
 811         microtime(&pn->tval);
 812 }
 813
 814
 815 /*
 816  * pfil hook that is called for each IPv4 packet making its way through the
 817  * stack in either direction.
 818  * The pfil subsystem holds a non-sleepable mutex somewhere when
 819  * calling our hook function, so we can't sleep at all.
 820  * It's very important to use the M_NOWAIT flag with all function calls
 821  * that support it so that they won't sleep, otherwise you get a panic.
 822  */
 823 static int
 824 siftr_chkpkt(void *arg, struct mbuf **m, struct ifnet *ifp, int dir,
 825     struct inpcb *inp)
 826 {
 827         struct pkt_node *pn;
 828         struct ip *ip;
 829         struct tcphdr *th;
 830         struct tcpcb *tp;
 831         struct siftr_stats *ss;
 832         unsigned int ip_hl;
 833         int inp_locally_locked;
 834
 835         inp_locally_locked = 0;
 836         ss = DPCPU_PTR(ss);
 837
 838         /*
 839          * m_pullup is not required here because ip_{input|output}
 840          * already do the heavy lifting for us.
 841          */
 842
 843         ip = mtod(*m, struct ip *);
 844
 845         /* Only continue processing if the packet is TCP. */
 846         if (ip->ip_p != IPPROTO_TCP)
 847                 goto ret;
 848
 849         /*
 850          * If a kernel subsystem reinjects packets into the stack, our pfil
 851          * hook will be called multiple times for the same packet.
 852          * Make sure we only process unique packets.
 853          */
 854         if (siftr_chkreinject(*m, dir, ss))
 855                 goto ret;
 856
 857         if (dir == PFIL_IN)
 858                 ss->n_in++;
 859         else
 860                 ss->n_out++;
 861
 862         /*
 863          * Create a tcphdr struct starting at the correct offset
 864          * in the IP packet. ip->ip_hl gives the ip header length
 865          * in 4-byte words, so multiply it to get the size in bytes.
 866          */
 867         ip_hl = (ip->ip_hl << 2);
 868         th = (struct tcphdr *)((caddr_t)ip + ip_hl);
 869
 870         /*
 871          * If the pfil hooks don't provide a pointer to the
 872          * inpcb, we need to find it ourselves and lock it.
 873          */
 874         if (!inp) {
 875                 /* Find the corresponding inpcb for this pkt. */
 876                 inp = siftr_findinpcb(INP_IPV4, ip, *m, th->th_sport,
 877                     th->th_dport, dir, ss);
 878
 879                 if (inp == NULL)
 880                         goto ret;
 881                 else
 882                         inp_locally_locked = 1;
 883         }
 884
 885         INP_LOCK_ASSERT(inp);
 886
 887         /* Find the TCP control block that corresponds with this packet */
 888         tp = intotcpcb(inp);
 889
 890         /*
 891          * If we can't find the TCP control block (happens occasionaly for a
 892          * packet sent during the shutdown phase of a TCP connection),
 893          * or we're in the timewait state, bail
 894          */
 895         if (tp == NULL || inp->inp_flags & INP_TIMEWAIT) {
 896                 if (dir == PFIL_IN)
 897                         ss->nskip_in_tcpcb++;
 898                 else
 899                         ss->nskip_out_tcpcb++;
 900
 901                 goto inp_unlock;
 902         }
 903
 904         pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO);
 905
 906         if (pn == NULL) {
 907                 if (dir == PFIL_IN)
 908                         ss->nskip_in_malloc++;
 909                 else
 910                         ss->nskip_out_malloc++;
 911
 912                 goto inp_unlock;
 913         }
 914
 915         siftr_siftdata(pn, inp, tp, INP_IPV4, dir, inp_locally_locked);
 916
 917         if (siftr_generate_hashes) {
 918                 if ((*m)->m_pkthdr.csum_flags & CSUM_TCP) {
 919                         /*
 920                          * For outbound packets, the TCP checksum isn't
 921                          * calculated yet. This is a problem for our packet
 922                          * hashing as the receiver will calc a different hash
 923                          * to ours if we don't include the correct TCP checksum
 924                          * in the bytes being hashed. To work around this
 925                          * problem, we manually calc the TCP checksum here in
 926                          * software. We unset the CSUM_TCP flag so the lower
 927                          * layers don't recalc it.
 928                          */
 929                         (*m)->m_pkthdr.csum_flags &= ~CSUM_TCP;
 930
 931                         /*
 932                          * Calculate the TCP checksum in software and assign
 933                          * to correct TCP header field, which will follow the
 934                          * packet mbuf down the stack. The trick here is that
 935                          * tcp_output() sets th->th_sum to the checksum of the
 936                          * pseudo header for us already. Because of the nature
 937                          * of the checksumming algorithm, we can sum over the
 938                          * entire IP payload (i.e. TCP header and data), which
 939                          * will include the already calculated pseduo header
 940                          * checksum, thus giving us the complete TCP checksum.
 941                          *
 942                          * To put it in simple terms, if checksum(1,2,3,4)=10,
 943                          * then checksum(1,2,3,4,5) == checksum(10,5).
 944                          * This property is what allows us to "cheat" and
 945                          * checksum only the IP payload which has the TCP
 946                          * th_sum field populated with the pseudo header's
 947                          * checksum, and not need to futz around checksumming
 948                          * pseudo header bytes and TCP header/data in one hit.
 949                          * Refer to RFC 1071 for more info.
 950                          *
 951                          * NB: in_cksum_skip(struct mbuf *m, int len, int skip)
 952                          * in_cksum_skip 2nd argument is NOT the number of
 953                          * bytes to read from the mbuf at "skip" bytes offset
 954                          * from the start of the mbuf (very counter intuitive!).
 955                          * The number of bytes to read is calculated internally
 956                          * by the function as len-skip i.e. to sum over the IP
 957                          * payload (TCP header + data) bytes, it is INCORRECT
 958                          * to call the function like this:
 959                          * in_cksum_skip(at, ip->ip_len - offset, offset)
 960                          * Rather, it should be called like this:
 961                          * in_cksum_skip(at, ip->ip_len, offset)
 962                          * which means read "ip->ip_len - offset" bytes from
 963                          * the mbuf cluster "at" at offset "offset" bytes from
 964                          * the beginning of the "at" mbuf's data pointer.
 965                          */
 966                         th->th_sum = in_cksum_skip(*m, ntohs(ip->ip_len),
 967                             ip_hl);
 968                 }
 969
 970                 /*
 971                  * XXX: Having to calculate the checksum in software and then
 972                  * hash over all bytes is really inefficient. Would be nice to
 973                  * find a way to create the hash and checksum in the same pass
 974                  * over the bytes.
 975                  */
 976                 pn->hash = hash_pkt(*m, ip_hl);
 977         }
 978
 979         mtx_lock(&siftr_pkt_queue_mtx);
 980         STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes);
 981         mtx_unlock(&siftr_pkt_queue_mtx);
 982         goto ret;
 983
 984 inp_unlock:
 985         if (inp_locally_locked)
 986                 INP_RUNLOCK(inp);
 987
 988 ret:
 989         /* Returning 0 ensures pfil will not discard the pkt */
 990         return (0);
 991 }
 992
 993
 994 #ifdef SIFTR_IPV6
 995 static int
 996 siftr_chkpkt6(void *arg, struct mbuf **m, struct ifnet *ifp, int dir,
 997     struct inpcb *inp)
 998 {
 999         struct pkt_node *pn;
1000         struct ip6_hdr *ip6;
1001         struct tcphdr *th;
1002         struct tcpcb *tp;
1003         struct siftr_stats *ss;
1004         unsigned int ip6_hl;
1005         int inp_locally_locked;
1006
1007         inp_locally_locked = 0;
1008         ss = DPCPU_PTR(ss);
1009
1010         /*
1011          * m_pullup is not required here because ip6_{input|output}
1012          * already do the heavy lifting for us.
1013          */
1014
1015         ip6 = mtod(*m, struct ip6_hdr *);
1016
1017         /*
1018          * Only continue processing if the packet is TCP
1019          * XXX: We should follow the next header fields
1020          * as shown on Pg 6 RFC 2460, but right now we'll
1021          * only check pkts that have no extension headers.
1022          */
1023         if (ip6->ip6_nxt != IPPROTO_TCP)
1024                 goto ret6;
1025
1026         /*
1027          * If a kernel subsystem reinjects packets into the stack, our pfil
1028          * hook will be called multiple times for the same packet.
1029          * Make sure we only process unique packets.
1030          */
1031         if (siftr_chkreinject(*m, dir, ss))
1032                 goto ret6;
1033
1034         if (dir == PFIL_IN)
1035                 ss->n_in++;
1036         else
1037                 ss->n_out++;
1038
1039         ip6_hl = sizeof(struct ip6_hdr);
1040
1041         /*
1042          * Create a tcphdr struct starting at the correct offset
1043          * in the ipv6 packet. ip->ip_hl gives the ip header length
1044          * in 4-byte words, so multiply it to get the size in bytes.
1045          */
1046         th = (struct tcphdr *)((caddr_t)ip6 + ip6_hl);
1047
1048         /*
1049          * For inbound packets, the pfil hooks don't provide a pointer to the
1050          * inpcb, so we need to find it ourselves and lock it.
1051          */
1052         if (!inp) {
1053                 /* Find the corresponding inpcb for this pkt. */
1054                 inp = siftr_findinpcb(INP_IPV6, (struct ip *)ip6, *m,
1055                     th->th_sport, th->th_dport, dir, ss);
1056
1057                 if (inp == NULL)
1058                         goto ret6;
1059                 else
1060                         inp_locally_locked = 1;
1061         }
1062
1063         /* Find the TCP control block that corresponds with this packet. */
1064         tp = intotcpcb(inp);
1065
1066         /*
1067          * If we can't find the TCP control block (happens occasionaly for a
1068          * packet sent during the shutdown phase of a TCP connection),
1069          * or we're in the timewait state, bail.
1070          */
1071         if (tp == NULL || inp->inp_flags & INP_TIMEWAIT) {
1072                 if (dir == PFIL_IN)
1073                         ss->nskip_in_tcpcb++;
1074                 else
1075                         ss->nskip_out_tcpcb++;
1076
1077                 goto inp_unlock6;
1078         }
1079
1080         pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO);
1081
1082         if (pn == NULL) {
1083                 if (dir == PFIL_IN)
1084                         ss->nskip_in_malloc++;
1085                 else
1086                         ss->nskip_out_malloc++;
1087
1088                 goto inp_unlock6;
1089         }
1090
1091         siftr_siftdata(pn, inp, tp, INP_IPV6, dir, inp_locally_locked);
1092
1093         /* XXX: Figure out how to generate hashes for IPv6 packets. */
1094
1095         mtx_lock(&siftr_pkt_queue_mtx);
1096         STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes);
1097         mtx_unlock(&siftr_pkt_queue_mtx);
1098         goto ret6;
1099
1100 inp_unlock6:
1101         if (inp_locally_locked)
1102                 INP_RUNLOCK(inp);
1103
1104 ret6:
1105         /* Returning 0 ensures pfil will not discard the pkt. */
1106         return (0);
1107 }
1108 #endif /* #ifdef SIFTR_IPV6 */
1109
1110
1111 static int
1112 siftr_pfil(int action)
1113 {
1114         struct pfil_head *pfh_inet;
1115 #ifdef SIFTR_IPV6
1116         struct pfil_head *pfh_inet6;
1117 #endif
1118         VNET_ITERATOR_DECL(vnet_iter);
1119
1120         VNET_LIST_RLOCK();
1121         VNET_FOREACH(vnet_iter) {
1122                 CURVNET_SET(vnet_iter);
1123                 pfh_inet = pfil_head_get(PFIL_TYPE_AF, AF_INET);
1124 #ifdef SIFTR_IPV6
1125                 pfh_inet6 = pfil_head_get(PFIL_TYPE_AF, AF_INET6);
1126 #endif
1127
1128                 if (action == HOOK) {
1129                         pfil_add_hook(siftr_chkpkt, NULL,
1130                             PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet);
1131 #ifdef SIFTR_IPV6
1132                         pfil_add_hook(siftr_chkpkt6, NULL,
1133                             PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet6);
1134 #endif
1135                 } else if (action == UNHOOK) {
1136                         pfil_remove_hook(siftr_chkpkt, NULL,
1137                             PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet);
1138 #ifdef SIFTR_IPV6
1139                         pfil_remove_hook(siftr_chkpkt6, NULL,
1140                             PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet6);
1141 #endif
1142                 }
1143                 CURVNET_RESTORE();
1144         }
1145         VNET_LIST_RUNLOCK();
1146
1147         return (0);
1148 }
1149
1150
1151 static int
1152 siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS)
1153 {
1154         struct alq *new_alq;
1155         int error;
1156
1157         error = sysctl_handle_string(oidp, arg1, arg2, req);
1158
1159         /* Check for error or same filename */
1160         if (error != 0 || req->newptr == NULL ||
1161             strncmp(siftr_logfile, arg1, arg2) == 0)
1162                 goto done;
1163
1164         /* Filname changed */
1165         error = alq_open(&new_alq, arg1, curthread->td_ucred,
1166             SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0);
1167         if (error != 0)
1168                 goto done;
1169
1170         /*
1171          * If disabled, siftr_alq == NULL so we simply close
1172          * the alq as we've proved it can be opened.
1173          * If enabled, close the existing alq and switch the old
1174          * for the new.
1175          */
1176         if (siftr_alq == NULL) {
1177                 alq_close(new_alq);
1178         } else {
1179                 alq_close(siftr_alq);
1180                 siftr_alq = new_alq;
1181         }
1182
1183         /* Update filename upon success */
1184         strlcpy(siftr_logfile, arg1, arg2);
1185 done:
1186         return (error);
1187 }
1188
1189 static int
1190 siftr_manage_ops(uint8_t action)
1191 {
1192         struct siftr_stats totalss;
1193         struct timeval tval;
1194         struct flow_hash_node *counter, *tmp_counter;
1195         struct sbuf *s;
1196         int i, key_index, ret, error;
1197         uint32_t bytes_to_write, total_skipped_pkts;
1198         uint16_t lport, fport;
1199         uint8_t *key, ipver;
1200
1201 #ifdef SIFTR_IPV6
1202         uint32_t laddr[4];
1203         uint32_t faddr[4];
1204 #else
1205         uint8_t laddr[4];
1206         uint8_t faddr[4];
1207 #endif
1208
1209         error = 0;
1210         total_skipped_pkts = 0;
1211
1212         /* Init an autosizing sbuf that initially holds 200 chars. */
1213         if ((s = sbuf_new(NULL, NULL, 200, SBUF_AUTOEXTEND)) == NULL)
1214                 return (-1);
1215
1216         if (action == SIFTR_ENABLE) {
1217                 /*
1218                  * Create our alq
1219                  * XXX: We should abort if alq_open fails!
1220                  */
1221                 alq_open(&siftr_alq, siftr_logfile, curthread->td_ucred,
1222                     SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0);
1223
1224                 STAILQ_INIT(&pkt_queue);
1225
1226                 DPCPU_ZERO(ss);
1227
1228                 siftr_exit_pkt_manager_thread = 0;
1229
1230                 ret = kthread_add(&siftr_pkt_manager_thread, NULL, NULL,
1231                     &siftr_pkt_manager_thr, RFNOWAIT, 0,
1232                     "siftr_pkt_manager_thr");
1233
1234                 siftr_pfil(HOOK);
1235
1236                 microtime(&tval);
1237
1238                 sbuf_printf(s,
1239                     "enable_time_secs=%jd\tenable_time_usecs=%06ld\t"
1240                     "siftrver=%s\thz=%u\ttcp_rtt_scale=%u\tsysname=%s\t"
1241                     "sysver=%u\tipmode=%u\n",
1242                     (intmax_t)tval.tv_sec, tval.tv_usec, MODVERSION_STR, hz,
1243                     TCP_RTT_SCALE, SYS_NAME, __FreeBSD_version, SIFTR_IPMODE);
1244
1245                 sbuf_finish(s);
1246                 alq_writen(siftr_alq, sbuf_data(s), sbuf_len(s), ALQ_WAITOK);
1247
1248         } else if (action == SIFTR_DISABLE && siftr_pkt_manager_thr != NULL) {
1249                 /*
1250                  * Remove the pfil hook functions. All threads currently in
1251                  * the hook functions are allowed to exit before siftr_pfil()
1252                  * returns.
1253                  */
1254                 siftr_pfil(UNHOOK);
1255
1256                 /* This will block until the pkt manager thread unlocks it. */
1257                 mtx_lock(&siftr_pkt_mgr_mtx);
1258
1259                 /* Tell the pkt manager thread that it should exit now. */
1260                 siftr_exit_pkt_manager_thread = 1;
1261
1262                 /*
1263                  * Wake the pkt_manager thread so it realises that
1264                  * siftr_exit_pkt_manager_thread == 1 and exits gracefully.
1265                  * The wakeup won't be delivered until we unlock
1266                  * siftr_pkt_mgr_mtx so this isn't racy.
1267                  */
1268                 wakeup(&wait_for_pkt);
1269
1270                 /* Wait for the pkt_manager thread to exit. */
1271                 mtx_sleep(siftr_pkt_manager_thr, &siftr_pkt_mgr_mtx, PWAIT,
1272                     "thrwait", 0);
1273
1274                 siftr_pkt_manager_thr = NULL;
1275                 mtx_unlock(&siftr_pkt_mgr_mtx);
1276
1277                 totalss.n_in = DPCPU_VARSUM(ss, n_in);
1278                 totalss.n_out = DPCPU_VARSUM(ss, n_out);
1279                 totalss.nskip_in_malloc = DPCPU_VARSUM(ss, nskip_in_malloc);
1280                 totalss.nskip_out_malloc = DPCPU_VARSUM(ss, nskip_out_malloc);
1281                 totalss.nskip_in_mtx = DPCPU_VARSUM(ss, nskip_in_mtx);
1282                 totalss.nskip_out_mtx = DPCPU_VARSUM(ss, nskip_out_mtx);
1283                 totalss.nskip_in_tcpcb = DPCPU_VARSUM(ss, nskip_in_tcpcb);
1284                 totalss.nskip_out_tcpcb = DPCPU_VARSUM(ss, nskip_out_tcpcb);
1285                 totalss.nskip_in_inpcb = DPCPU_VARSUM(ss, nskip_in_inpcb);
1286                 totalss.nskip_out_inpcb = DPCPU_VARSUM(ss, nskip_out_inpcb);
1287
1288                 total_skipped_pkts = totalss.nskip_in_malloc +
1289                     totalss.nskip_out_malloc + totalss.nskip_in_mtx +
1290                     totalss.nskip_out_mtx + totalss.nskip_in_tcpcb +
1291                     totalss.nskip_out_tcpcb + totalss.nskip_in_inpcb +
1292                     totalss.nskip_out_inpcb;
1293
1294                 microtime(&tval);
1295
1296                 sbuf_printf(s,
1297                     "disable_time_secs=%jd\tdisable_time_usecs=%06ld\t"
1298                     "num_inbound_tcp_pkts=%ju\tnum_outbound_tcp_pkts=%ju\t"
1299                     "total_tcp_pkts=%ju\tnum_inbound_skipped_pkts_malloc=%u\t"
1300                     "num_outbound_skipped_pkts_malloc=%u\t"
1301                     "num_inbound_skipped_pkts_mtx=%u\t"
1302                     "num_outbound_skipped_pkts_mtx=%u\t"
1303                     "num_inbound_skipped_pkts_tcpcb=%u\t"
1304                     "num_outbound_skipped_pkts_tcpcb=%u\t"
1305                     "num_inbound_skipped_pkts_inpcb=%u\t"
1306                     "num_outbound_skipped_pkts_inpcb=%u\t"
1307                     "total_skipped_tcp_pkts=%u\tflow_list=",
1308                     (intmax_t)tval.tv_sec,
1309                     tval.tv_usec,
1310                     (uintmax_t)totalss.n_in,
1311                     (uintmax_t)totalss.n_out,
1312                     (uintmax_t)(totalss.n_in + totalss.n_out),
1313                     totalss.nskip_in_malloc,
1314                     totalss.nskip_out_malloc,
1315                     totalss.nskip_in_mtx,
1316                     totalss.nskip_out_mtx,
1317                     totalss.nskip_in_tcpcb,
1318                     totalss.nskip_out_tcpcb,
1319                     totalss.nskip_in_inpcb,
1320                     totalss.nskip_out_inpcb,
1321                     total_skipped_pkts);
1322
1323                 /*
1324                  * Iterate over the flow hash, printing a summary of each
1325                  * flow seen and freeing any malloc'd memory.
1326                  * The hash consists of an array of LISTs (man 3 queue).
1327                  */
1328                 for (i = 0; i <= siftr_hashmask; i++) {
1329                         LIST_FOREACH_SAFE(counter, counter_hash + i, nodes,
1330                             tmp_counter) {
1331                                 key = counter->key;
1332                                 key_index = 1;
1333
1334                                 ipver = key[0];
1335
1336                                 memcpy(laddr, key + key_index, sizeof(laddr));
1337                                 key_index += sizeof(laddr);
1338                                 memcpy(&lport, key + key_index, sizeof(lport));
1339                                 key_index += sizeof(lport);
1340                                 memcpy(faddr, key + key_index, sizeof(faddr));
1341                                 key_index += sizeof(faddr);
1342                                 memcpy(&fport, key + key_index, sizeof(fport));
1343
1344 #ifdef SIFTR_IPV6
1345                                 laddr[3] = ntohl(laddr[3]);
1346                                 faddr[3] = ntohl(faddr[3]);
1347
1348                                 if (ipver == INP_IPV6) {
1349                                         laddr[0] = ntohl(laddr[0]);
1350                                         laddr[1] = ntohl(laddr[1]);
1351                                         laddr[2] = ntohl(laddr[2]);
1352                                         faddr[0] = ntohl(faddr[0]);
1353                                         faddr[1] = ntohl(faddr[1]);
1354                                         faddr[2] = ntohl(faddr[2]);
1355
1356                                         sbuf_printf(s,
1357                                             "%x:%x:%x:%x:%x:%x:%x:%x;%u-"
1358                                             "%x:%x:%x:%x:%x:%x:%x:%x;%u,",
1359                                             UPPER_SHORT(laddr[0]),
1360                                             LOWER_SHORT(laddr[0]),
1361                                             UPPER_SHORT(laddr[1]),
1362                                             LOWER_SHORT(laddr[1]),
1363                                             UPPER_SHORT(laddr[2]),
1364                                             LOWER_SHORT(laddr[2]),
1365                                             UPPER_SHORT(laddr[3]),
1366                                             LOWER_SHORT(laddr[3]),
1367                                             ntohs(lport),
1368                                             UPPER_SHORT(faddr[0]),
1369                                             LOWER_SHORT(faddr[0]),
1370                                             UPPER_SHORT(faddr[1]),
1371                                             LOWER_SHORT(faddr[1]),
1372                                             UPPER_SHORT(faddr[2]),
1373                                             LOWER_SHORT(faddr[2]),
1374                                             UPPER_SHORT(faddr[3]),
1375                                             LOWER_SHORT(faddr[3]),
1376                                             ntohs(fport));
1377                                 } else {
1378                                         laddr[0] = FIRST_OCTET(laddr[3]);
1379                                         laddr[1] = SECOND_OCTET(laddr[3]);
1380                                         laddr[2] = THIRD_OCTET(laddr[3]);
1381                                         laddr[3] = FOURTH_OCTET(laddr[3]);
1382                                         faddr[0] = FIRST_OCTET(faddr[3]);
1383                                         faddr[1] = SECOND_OCTET(faddr[3]);
1384                                         faddr[2] = THIRD_OCTET(faddr[3]);
1385                                         faddr[3] = FOURTH_OCTET(faddr[3]);
1386 #endif
1387                                         sbuf_printf(s,
1388                                             "%u.%u.%u.%u;%u-%u.%u.%u.%u;%u,",
1389                                             laddr[0],
1390                                             laddr[1],
1391                                             laddr[2],
1392                                             laddr[3],
1393                                             ntohs(lport),
1394                                             faddr[0],
1395                                             faddr[1],
1396                                             faddr[2],
1397                                             faddr[3],
1398                                             ntohs(fport));
1399 #ifdef SIFTR_IPV6
1400                                 }
1401 #endif
1402
1403                                 free(counter, M_SIFTR_HASHNODE);
1404                         }
1405
1406                         LIST_INIT(counter_hash + i);
1407                 }
1408
1409                 sbuf_printf(s, "\n");
1410                 sbuf_finish(s);
1411
1412                 i = 0;
1413                 do {
1414                         bytes_to_write = min(SIFTR_ALQ_BUFLEN, sbuf_len(s)-i);
1415                         alq_writen(siftr_alq, sbuf_data(s)+i, bytes_to_write, ALQ_WAITOK);
1416                         i += bytes_to_write;
1417                 } while (i < sbuf_len(s));
1418
1419                 alq_close(siftr_alq);
1420                 siftr_alq = NULL;
1421         }
1422
1423         sbuf_delete(s);
1424
1425         /*
1426          * XXX: Should be using ret to check if any functions fail
1427          * and set error appropriately
1428          */
1429
1430         return (error);
1431 }
1432
1433
1434 static int
1435 siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS)
1436 {
1437         if (req->newptr == NULL)
1438                 goto skip;
1439
1440         /* If the value passed in isn't 0 or 1, return an error. */
1441         if (CAST_PTR_INT(req->newptr) != 0 && CAST_PTR_INT(req->newptr) != 1)
1442                 return (1);
1443
1444         /* If we are changing state (0 to 1 or 1 to 0). */
1445         if (CAST_PTR_INT(req->newptr) != siftr_enabled )
1446                 if (siftr_manage_ops(CAST_PTR_INT(req->newptr))) {
1447                         siftr_manage_ops(SIFTR_DISABLE);
1448                         return (1);
1449                 }
1450
1451 skip:
1452         return (sysctl_handle_int(oidp, arg1, arg2, req));
1453 }
1454
1455
1456 static void
1457 siftr_shutdown_handler(void *arg)
1458 {
1459         siftr_manage_ops(SIFTR_DISABLE);
1460 }
1461
1462
1463 /*
1464  * Module is being unloaded or machine is shutting down. Take care of cleanup.
1465  */
1466 static int
1467 deinit_siftr(void)
1468 {
1469         /* Cleanup. */
1470         siftr_manage_ops(SIFTR_DISABLE);
1471         hashdestroy(counter_hash, M_SIFTR, siftr_hashmask);
1472         mtx_destroy(&siftr_pkt_queue_mtx);
1473         mtx_destroy(&siftr_pkt_mgr_mtx);
1474
1475         return (0);
1476 }
1477
1478
1479 /*
1480  * Module has just been loaded into the kernel.
1481  */
1482 static int
1483 init_siftr(void)
1484 {
1485         EVENTHANDLER_REGISTER(shutdown_pre_sync, siftr_shutdown_handler, NULL,
1486             SHUTDOWN_PRI_FIRST);
1487
1488         /* Initialise our flow counter hash table. */
1489         counter_hash = hashinit(SIFTR_EXPECTED_MAX_TCP_FLOWS, M_SIFTR,
1490             &siftr_hashmask);
1491
1492         mtx_init(&siftr_pkt_queue_mtx, "siftr_pkt_queue_mtx", NULL, MTX_DEF);
1493         mtx_init(&siftr_pkt_mgr_mtx, "siftr_pkt_mgr_mtx", NULL, MTX_DEF);
1494
1495         /* Print message to the user's current terminal. */
1496         uprintf("\nStatistical Information For TCP Research (SIFTR) %s\n"
1497             "          http://caia.swin.edu.au/urp/newtcp\n\n",
1498             MODVERSION_STR);
1499
1500         return (0);
1501 }
1502
1503
1504 /*
1505  * This is the function that is called to load and unload the module.
1506  * When the module is loaded, this function is called once with
1507  * "what" == MOD_LOAD
1508  * When the module is unloaded, this function is called twice with
1509  * "what" = MOD_QUIESCE first, followed by "what" = MOD_UNLOAD second
1510  * When the system is shut down e.g. CTRL-ALT-DEL or using the shutdown command,
1511  * this function is called once with "what" = MOD_SHUTDOWN
1512  * When the system is shut down, the handler isn't called until the very end
1513  * of the shutdown sequence i.e. after the disks have been synced.
1514  */
1515 static int
1516 siftr_load_handler(module_t mod, int what, void *arg)
1517 {
1518         int ret;
1519
1520         switch (what) {
1521         case MOD_LOAD:
1522                 ret = init_siftr();
1523                 break;
1524
1525         case MOD_QUIESCE:
1526         case MOD_SHUTDOWN:
1527                 ret = deinit_siftr();
1528                 break;
1529
1530         case MOD_UNLOAD:
1531                 ret = 0;
1532                 break;
1533
1534         default:
1535                 ret = EINVAL;
1536                 break;
1537         }
1538
1539         return (ret);
1540 }
1541
1542
1543 static moduledata_t siftr_mod = {
1544         .name = "siftr",
1545         .evhand = siftr_load_handler,
1546 };
1547
1548 /*
1549  * Param 1: name of the kernel module
1550  * Param 2: moduledata_t struct containing info about the kernel module
1551  *          and the execution entry point for the module
1552  * Param 3: From sysinit_sub_id enumeration in /usr/include/sys/kernel.h
1553  *          Defines the module initialisation order
1554  * Param 4: From sysinit_elem_order enumeration in /usr/include/sys/kernel.h
1555  *          Defines the initialisation order of this kld relative to others
1556  *          within the same subsystem as defined by param 3
1557  */
1558 DECLARE_MODULE(siftr, siftr_mod, SI_SUB_SMP, SI_ORDER_ANY);
1559 MODULE_DEPEND(siftr, alq, 1, 1, 1);
1560 MODULE_VERSION(siftr, MODVERSION);