lib/conntrack-tcp.c

   1 /*-
   2  * Copyright (c) 2001 Daniel Hartmeier
   3  * Copyright (c) 2002 - 2008 Henning Brauer
   4  * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
   5  * Copyright (c) 2015, 2016 Nicira, Inc.
   6  * All rights reserved.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  *
  12  *    - Redistributions of source code must retain the above copyright
  13  *      notice, this list of conditions and the following disclaimer.
  14  *    - Redistributions in binary form must reproduce the above
  15  *      copyright notice, this list of conditions and the following
  16  *      disclaimer in the documentation and/or other materials provided
  17  *      with the distribution.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  20  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  21  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  22  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
  23  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  24  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
  25  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  26  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
  27  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
  29  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  30  * POSSIBILITY OF SUCH DAMAGE.
  31  *
  32  * Effort sponsored in part by the Defense Advanced Research Projects
  33  * Agency (DARPA) and Air Force Research Laboratory, Air Force
  34  * Materiel Command, USAF, under agreement number F30602-01-2-0537.
  35  *
  36  *      $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $
  37  */
  38
  39 #include <config.h>
  40
  41 #include "conntrack-private.h"
  42 #include "coverage.h"
  43 #include "ct-dpif.h"
  44 #include "dp-packet.h"
  45 #include "util.h"
  46
  47 COVERAGE_DEFINE(conntrack_tcp_seq_chk_bypass);
  48 COVERAGE_DEFINE(conntrack_tcp_seq_chk_failed);
  49 COVERAGE_DEFINE(conntrack_invalid_tcp_flags);
  50
  51 struct tcp_peer {
  52     uint32_t               seqlo;          /* Max sequence number sent     */
  53     uint32_t               seqhi;          /* Max the other end ACKd + win */
  54     uint16_t               max_win;        /* largest window (pre scaling) */
  55     uint8_t                wscale;         /* window scaling factor        */
  56     enum ct_dpif_tcp_state state;
  57 };
  58
  59 struct conn_tcp {
  60     struct conn up;
  61     struct tcp_peer peer[2]; /* 'conn' lock protected. */
  62 };
  63
  64 enum {
  65     TCPOPT_EOL,
  66     TCPOPT_NOP,
  67     TCPOPT_WINDOW = 3,
  68 };
  69
  70 /* TCP sequence numbers are 32 bit integers operated
  71  * on with modular arithmetic.  These macros can be
  72  * used to compare such integers. */
  73 #define SEQ_LT(a,b)     INT_MOD_LT(a, b)
  74 #define SEQ_LEQ(a,b)    INT_MOD_LEQ(a, b)
  75 #define SEQ_GT(a,b)     INT_MOD_GT(a, b)
  76 #define SEQ_GEQ(a,b)    INT_MOD_GEQ(a, b)
  77
  78 #define SEQ_MIN(a, b)   INT_MOD_MIN(a, b)
  79 #define SEQ_MAX(a, b)   INT_MOD_MAX(a, b)
  80
  81 static struct conn_tcp*
  82 conn_tcp_cast(const struct conn* conn)
  83 {
  84     return CONTAINER_OF(conn, struct conn_tcp, up);
  85 }
  86
  87 /* pf does this in in pf_normalize_tcp(), and it is called only if scrub
  88  * is enabled.  We're not scrubbing, but this check seems reasonable.  */
  89 static bool
  90 tcp_invalid_flags(uint16_t flags)
  91 {
  92
  93     if (flags & TCP_SYN) {
  94         if (flags & TCP_RST || flags & TCP_FIN) {
  95             return true;
  96         }
  97     } else {
  98         /* Illegal packet */
  99         if (!(flags & (TCP_ACK|TCP_RST))) {
 100             return true;
 101         }
 102     }
 103
 104     if (!(flags & TCP_ACK)) {
 105         /* These flags are only valid if ACK is set */
 106         if ((flags & TCP_FIN) || (flags & TCP_PSH) || (flags & TCP_URG)) {
 107             return true;
 108         }
 109     }
 110
 111     return false;
 112 }
 113
 114 #define TCP_MAX_WSCALE 14
 115 #define CT_WSCALE_FLAG 0x80
 116 #define CT_WSCALE_UNKNOWN 0x40
 117 #define CT_WSCALE_MASK 0xf
 118
 119 static uint8_t
 120 tcp_get_wscale(const struct tcp_header *tcp)
 121 {
 122     int len = TCP_OFFSET(tcp->tcp_ctl) * 4 - sizeof *tcp;
 123     const uint8_t *opt = (const uint8_t *)(tcp + 1);
 124     uint8_t wscale = 0;
 125     uint8_t optlen;
 126
 127     while (len >= 3) {
 128         switch (*opt) {
 129         case TCPOPT_EOL:
 130             return wscale;
 131         case TCPOPT_NOP:
 132             opt++;
 133             len--;
 134             break;
 135         case TCPOPT_WINDOW:
 136             wscale = MIN(opt[2], TCP_MAX_WSCALE);
 137             wscale |= CT_WSCALE_FLAG;
 138             /* fall through */
 139         default:
 140             optlen = opt[1];
 141             if (optlen < 2) {
 142                 optlen = 2;
 143             }
 144             len -= optlen;
 145             opt += optlen;
 146         }
 147     }
 148
 149     return wscale;
 150 }
 151
 152 static bool
 153 tcp_bypass_seq_chk(struct conntrack *ct)
 154 {
 155     if (!conntrack_get_tcp_seq_chk(ct)) {
 156         COVERAGE_INC(conntrack_tcp_seq_chk_bypass);
 157         return true;
 158     }
 159     return false;
 160 }
 161
 162 static enum ct_update_res
 163 tcp_conn_update(struct conntrack *ct, struct conn *conn_,
 164                 struct dp_packet *pkt, bool reply, long long now)
 165 {
 166     struct conn_tcp *conn = conn_tcp_cast(conn_);
 167     struct tcp_header *tcp = dp_packet_l4(pkt);
 168     /* The peer that sent 'pkt' */
 169     struct tcp_peer *src = &conn->peer[reply ? 1 : 0];
 170     /* The peer that should receive 'pkt' */
 171     struct tcp_peer *dst = &conn->peer[reply ? 0 : 1];
 172     uint8_t sws = 0, dws = 0;
 173     uint16_t tcp_flags = TCP_FLAGS(tcp->tcp_ctl);
 174
 175     uint16_t win = ntohs(tcp->tcp_winsz);
 176     uint32_t ack, end, seq, orig_seq;
 177     uint32_t p_len = tcp_payload_length(pkt);
 178
 179     if (tcp_invalid_flags(tcp_flags)) {
 180         COVERAGE_INC(conntrack_invalid_tcp_flags);
 181         return CT_UPDATE_INVALID;
 182     }
 183
 184     if (((tcp_flags & (TCP_SYN | TCP_ACK)) == TCP_SYN)
 185         && dst->state >= CT_DPIF_TCPS_FIN_WAIT_2
 186         && src->state >= CT_DPIF_TCPS_FIN_WAIT_2) {
 187         src->state = dst->state = CT_DPIF_TCPS_CLOSED;
 188         return CT_UPDATE_NEW;
 189     }
 190
 191     if (src->wscale & CT_WSCALE_FLAG
 192         && dst->wscale & CT_WSCALE_FLAG
 193         && !(tcp_flags & TCP_SYN)) {
 194
 195         sws = src->wscale & CT_WSCALE_MASK;
 196         dws = dst->wscale & CT_WSCALE_MASK;
 197
 198     } else if (src->wscale & CT_WSCALE_UNKNOWN
 199                && dst->wscale & CT_WSCALE_UNKNOWN
 200                && !(tcp_flags & TCP_SYN)) {
 201
 202         sws = TCP_MAX_WSCALE;
 203         dws = TCP_MAX_WSCALE;
 204     }
 205
 206     /*
 207      * Sequence tracking algorithm from Guido van Rooij's paper:
 208      *   http://www.madison-gurkha.com/publications/tcp_filtering/
 209      *      tcp_filtering.ps
 210      */
 211
 212     orig_seq = seq = ntohl(get_16aligned_be32(&tcp->tcp_seq));
 213     bool check_ackskew = true;
 214     if (src->state < CT_DPIF_TCPS_SYN_SENT) {
 215         /* First packet from this end. Set its state */
 216
 217         ack = ntohl(get_16aligned_be32(&tcp->tcp_ack));
 218
 219         end = seq + p_len;
 220         if (tcp_flags & TCP_SYN) {
 221             end++;
 222             if (dst->wscale & CT_WSCALE_FLAG) {
 223                 src->wscale = tcp_get_wscale(tcp);
 224                 if (src->wscale & CT_WSCALE_FLAG) {
 225                     /* Remove scale factor from initial window */
 226                     sws = src->wscale & CT_WSCALE_MASK;
 227                     win = DIV_ROUND_UP((uint32_t) win, 1 << sws);
 228                     dws = dst->wscale & CT_WSCALE_MASK;
 229                 } else {
 230                     /* fixup other window */
 231                     dst->max_win <<= dst->wscale & CT_WSCALE_MASK;
 232                     /* in case of a retrans SYN|ACK */
 233                     dst->wscale = 0;
 234                 }
 235             }
 236         }
 237         if (tcp_flags & TCP_FIN) {
 238             end++;
 239         }
 240
 241         src->seqlo = seq;
 242         src->state = CT_DPIF_TCPS_SYN_SENT;
 243         /*
 244          * May need to slide the window (seqhi may have been set by
 245          * the crappy stack check or if we picked up the connection
 246          * after establishment)
 247          */
 248         if (src->seqhi == 1
 249                 || SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) {
 250             src->seqhi = end + MAX(1, dst->max_win << dws);
 251             /* We are either picking up a new connection or a connection which
 252              * was already in place.  We are more permissive in terms of
 253              * ackskew checking in these cases.
 254              */
 255             check_ackskew = false;
 256         }
 257         if (win > src->max_win) {
 258             src->max_win = win;
 259         }
 260
 261     } else {
 262         ack = ntohl(get_16aligned_be32(&tcp->tcp_ack));
 263         end = seq + p_len;
 264         if (tcp_flags & TCP_SYN) {
 265             end++;
 266         }
 267         if (tcp_flags & TCP_FIN) {
 268             end++;
 269         }
 270     }
 271
 272     if ((tcp_flags & TCP_ACK) == 0) {
 273         /* Let it pass through the ack skew check */
 274         ack = dst->seqlo;
 275     } else if ((ack == 0
 276                 && (tcp_flags & (TCP_ACK|TCP_RST)) == (TCP_ACK|TCP_RST))
 277                /* broken tcp stacks do not set ack */) {
 278         /* Many stacks (ours included) will set the ACK number in an
 279          * FIN|ACK if the SYN times out -- no sequence to ACK. */
 280         ack = dst->seqlo;
 281     }
 282
 283     if (seq == end) {
 284         /* Ease sequencing restrictions on no data packets */
 285         seq = src->seqlo;
 286         end = seq;
 287     }
 288
 289     int ackskew = check_ackskew ? dst->seqlo - ack : 0;
 290 #define MAXACKWINDOW (0xffff + 1500)    /* 1500 is an arbitrary fudge factor */
 291     if ((SEQ_GEQ(src->seqhi, end)
 292         /* Last octet inside other's window space */
 293         && SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws))
 294         /* Retrans: not more than one window back */
 295         && (ackskew >= -MAXACKWINDOW)
 296         /* Acking not more than one reassembled fragment backwards */
 297         && (ackskew <= (MAXACKWINDOW << sws))
 298         /* Acking not more than one window forward */
 299         && ((tcp_flags & TCP_RST) == 0 || orig_seq == src->seqlo
 300             || (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo)))
 301         || tcp_bypass_seq_chk(ct)) {
 302         /* Require an exact/+1 sequence match on resets when possible */
 303
 304         /* update max window */
 305         if (src->max_win < win) {
 306             src->max_win = win;
 307         }
 308         /* synchronize sequencing */
 309         if (SEQ_GT(end, src->seqlo)) {
 310             src->seqlo = end;
 311         }
 312         /* slide the window of what the other end can send */
 313         if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) {
 314             dst->seqhi = ack + MAX((win << sws), 1);
 315         }
 316
 317         /* update states */
 318         if (tcp_flags & TCP_SYN && src->state < CT_DPIF_TCPS_SYN_SENT) {
 319                 src->state = CT_DPIF_TCPS_SYN_SENT;
 320         }
 321         if (tcp_flags & TCP_FIN && src->state < CT_DPIF_TCPS_CLOSING) {
 322                 src->state = CT_DPIF_TCPS_CLOSING;
 323         }
 324         if (tcp_flags & TCP_ACK) {
 325             if (dst->state == CT_DPIF_TCPS_SYN_SENT) {
 326                 dst->state = CT_DPIF_TCPS_ESTABLISHED;
 327             } else if (dst->state == CT_DPIF_TCPS_CLOSING) {
 328                 dst->state = CT_DPIF_TCPS_FIN_WAIT_2;
 329             }
 330         }
 331         if (tcp_flags & TCP_RST) {
 332             src->state = dst->state = CT_DPIF_TCPS_TIME_WAIT;
 333         }
 334
 335         if (src->state >= CT_DPIF_TCPS_FIN_WAIT_2
 336             && dst->state >= CT_DPIF_TCPS_FIN_WAIT_2) {
 337             conn_update_expiration(ct, &conn->up, CT_TM_TCP_CLOSED, now);
 338         } else if (src->state >= CT_DPIF_TCPS_CLOSING
 339                    && dst->state >= CT_DPIF_TCPS_CLOSING) {
 340             conn_update_expiration(ct, &conn->up, CT_TM_TCP_FIN_WAIT, now);
 341         } else if (src->state < CT_DPIF_TCPS_ESTABLISHED
 342                    || dst->state < CT_DPIF_TCPS_ESTABLISHED) {
 343             conn_update_expiration(ct, &conn->up, CT_TM_TCP_OPENING, now);
 344         } else if (src->state >= CT_DPIF_TCPS_CLOSING
 345                    || dst->state >= CT_DPIF_TCPS_CLOSING) {
 346             conn_update_expiration(ct, &conn->up, CT_TM_TCP_CLOSING, now);
 347         } else {
 348             conn_update_expiration(ct, &conn->up, CT_TM_TCP_ESTABLISHED, now);
 349         }
 350     } else if ((dst->state < CT_DPIF_TCPS_SYN_SENT
 351                 || dst->state >= CT_DPIF_TCPS_FIN_WAIT_2
 352                 || src->state >= CT_DPIF_TCPS_FIN_WAIT_2)
 353                && SEQ_GEQ(src->seqhi + MAXACKWINDOW, end)
 354                /* Within a window forward of the originating packet */
 355                && SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
 356                /* Within a window backward of the originating packet */
 357
 358         /*
 359          * This currently handles three situations:
 360          *  1) Stupid stacks will shotgun SYNs before their peer
 361          *     replies.
 362          *  2) When PF catches an already established stream (the
 363          *     firewall rebooted, the state table was flushed, routes
 364          *     changed...)
 365          *  3) Packets get funky immediately after the connection
 366          *     closes (this should catch Solaris spurious ACK|FINs
 367          *     that web servers like to spew after a close)
 368          *
 369          * This must be a little more careful than the above code
 370          * since packet floods will also be caught here. We don't
 371          * update the TTL here to mitigate the damage of a packet
 372          * flood and so the same code can handle awkward establishment
 373          * and a loosened connection close.
 374          * In the establishment case, a correct peer response will
 375          * validate the connection, go through the normal state code
 376          * and keep updating the state TTL.
 377          */
 378
 379         /* update max window */
 380         if (src->max_win < win) {
 381             src->max_win = win;
 382         }
 383         /* synchronize sequencing */
 384         if (SEQ_GT(end, src->seqlo)) {
 385             src->seqlo = end;
 386         }
 387         /* slide the window of what the other end can send */
 388         if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) {
 389             dst->seqhi = ack + MAX((win << sws), 1);
 390         }
 391
 392         /*
 393          * Cannot set dst->seqhi here since this could be a shotgunned
 394          * SYN and not an already established connection.
 395          */
 396
 397         if (tcp_flags & TCP_FIN && src->state < CT_DPIF_TCPS_CLOSING) {
 398             src->state = CT_DPIF_TCPS_CLOSING;
 399         }
 400
 401         if (tcp_flags & TCP_RST) {
 402             src->state = dst->state = CT_DPIF_TCPS_TIME_WAIT;
 403         }
 404     } else {
 405         COVERAGE_INC(conntrack_tcp_seq_chk_failed);
 406         return CT_UPDATE_INVALID;
 407     }
 408
 409     return CT_UPDATE_VALID;
 410 }
 411
 412 static bool
 413 tcp_valid_new(struct dp_packet *pkt)
 414 {
 415     struct tcp_header *tcp = dp_packet_l4(pkt);
 416     uint16_t tcp_flags = TCP_FLAGS(tcp->tcp_ctl);
 417
 418     if (tcp_invalid_flags(tcp_flags)) {
 419         return false;
 420     }
 421
 422     /* A syn+ack is not allowed to create a connection.  We want to allow
 423      * totally new connections (syn) or already established, not partially
 424      * open (syn+ack). */
 425     if ((tcp_flags & TCP_SYN) && (tcp_flags & TCP_ACK)) {
 426         return false;
 427     }
 428
 429     return true;
 430 }
 431
 432 static struct conn *
 433 tcp_new_conn(struct conntrack *ct, struct dp_packet *pkt, long long now)
 434 {
 435     struct conn_tcp* newconn = NULL;
 436     struct tcp_header *tcp = dp_packet_l4(pkt);
 437     struct tcp_peer *src, *dst;
 438     uint16_t tcp_flags = TCP_FLAGS(tcp->tcp_ctl);
 439
 440     newconn = xzalloc(sizeof *newconn);
 441
 442     src = &newconn->peer[0];
 443     dst = &newconn->peer[1];
 444
 445     src->seqlo = ntohl(get_16aligned_be32(&tcp->tcp_seq));
 446     src->seqhi = src->seqlo + tcp_payload_length(pkt) + 1;
 447
 448     if (tcp_flags & TCP_SYN) {
 449         src->seqhi++;
 450         src->wscale = tcp_get_wscale(tcp);
 451     } else {
 452         src->wscale = CT_WSCALE_UNKNOWN;
 453         dst->wscale = CT_WSCALE_UNKNOWN;
 454     }
 455     src->max_win = MAX(ntohs(tcp->tcp_winsz), 1);
 456     if (src->wscale & CT_WSCALE_MASK) {
 457         /* Remove scale factor from initial window */
 458         uint8_t sws = src->wscale & CT_WSCALE_MASK;
 459         src->max_win = DIV_ROUND_UP((uint32_t) src->max_win, 1 << sws);
 460     }
 461     if (tcp_flags & TCP_FIN) {
 462         src->seqhi++;
 463     }
 464     dst->seqhi = 1;
 465     dst->max_win = 1;
 466     src->state = CT_DPIF_TCPS_SYN_SENT;
 467     dst->state = CT_DPIF_TCPS_CLOSED;
 468
 469     conn_init_expiration(ct, &newconn->up, CT_TM_TCP_FIRST_PACKET, now);
 470
 471     return &newconn->up;
 472 }
 473
 474 static uint8_t
 475 tcp_peer_to_protoinfo_flags(const struct tcp_peer *peer)
 476 {
 477     uint8_t res = 0;
 478
 479     if (peer->wscale & CT_WSCALE_FLAG) {
 480         res |= CT_DPIF_TCPF_WINDOW_SCALE;
 481     }
 482
 483     if (peer->wscale & CT_WSCALE_UNKNOWN) {
 484         res |= CT_DPIF_TCPF_BE_LIBERAL;
 485     }
 486
 487     return res;
 488 }
 489
 490 static void
 491 tcp_conn_get_protoinfo(const struct conn *conn_,
 492                        struct ct_dpif_protoinfo *protoinfo)
 493 {
 494     const struct conn_tcp *conn = conn_tcp_cast(conn_);
 495
 496     protoinfo->proto = IPPROTO_TCP;
 497     protoinfo->tcp.state_orig = conn->peer[0].state;
 498     protoinfo->tcp.state_reply = conn->peer[1].state;
 499
 500     protoinfo->tcp.wscale_orig = conn->peer[0].wscale & CT_WSCALE_MASK;
 501     protoinfo->tcp.wscale_reply = conn->peer[1].wscale & CT_WSCALE_MASK;
 502
 503     protoinfo->tcp.flags_orig = tcp_peer_to_protoinfo_flags(&conn->peer[0]);
 504     protoinfo->tcp.flags_reply = tcp_peer_to_protoinfo_flags(&conn->peer[1]);
 505 }
 506
 507 struct ct_l4_proto ct_proto_tcp = {
 508     .new_conn = tcp_new_conn,
 509     .valid_new = tcp_valid_new,
 510     .conn_update = tcp_conn_update,
 511     .conn_get_protoinfo = tcp_conn_get_protoinfo,
 512 };