sys/netinet/tcp_stacks/rack_bbr_common.c

   1 /*-
   2  * Copyright (c) 2016-2018
   3  *      Netflix Inc.
   4  *      All rights reserved.
   5  *
   6  * Redistribution and use in source and binary forms, with or without
   7  * modification, are permitted provided that the following conditions
   8  * are met:
   9  * 1. Redistributions of source code must retain the above copyright
  10  *    notice, this list of conditions and the following disclaimer.
  11  * 2. Redistributions in binary form must reproduce the above copyright
  12  *    notice, this list of conditions and the following disclaimer in the
  13  *    documentation and/or other materials provided with the distribution.
  14  *
  15  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  25  * SUCH DAMAGE.
  26  *
  27  */
  28 /*
  29  * Author: Randall Stewart <rrs@netflix.com>
  30  * This work is based on the ACM Queue paper
  31  * BBR - Congestion Based Congestion Control
  32  * and also numerous discussions with Neal, Yuchung and Van.
  33  */
  34
  35 #include <sys/cdefs.h>
  36 __FBSDID("$FreeBSD$");
  37
  38 #include "opt_inet.h"
  39 #include "opt_inet6.h"
  40 #include "opt_ipsec.h"
  41 #include "opt_tcpdebug.h"
  42 #include "opt_ratelimit.h"
  43 /*#include "opt_kern_tls.h"*/
  44 #include <sys/param.h>
  45 #include <sys/module.h>
  46 #include <sys/kernel.h>
  47 #ifdef TCP_HHOOK
  48 #include <sys/hhook.h>
  49 #endif
  50 #include <sys/malloc.h>
  51 #include <sys/mbuf.h>
  52 #include <sys/proc.h>
  53 #include <sys/socket.h>
  54 #include <sys/socketvar.h>
  55 #ifdef KERN_TLS
  56 #include <sys/sockbuf_tls.h>
  57 #endif
  58 #include <sys/sysctl.h>
  59 #include <sys/systm.h>
  60 #include <sys/tree.h>
  61 #include <sys/refcount.h>
  62 #include <sys/queue.h>
  63 #include <sys/smp.h>
  64 #include <sys/kthread.h>
  65 #include <sys/lock.h>
  66 #include <sys/mutex.h>
  67 #include <sys/time.h>
  68 #include <vm/uma.h>
  69 #include <sys/kern_prefetch.h>
  70
  71 #include <net/route.h>
  72 #include <net/vnet.h>
  73 #include <net/ethernet.h>
  74 #include <net/bpf.h>
  75
  76 #define TCPSTATES               /* for logging */
  77
  78 #include <netinet/in.h>
  79 #include <netinet/in_kdtrace.h>
  80 #include <netinet/in_pcb.h>
  81 #include <netinet/ip.h>
  82 #include <netinet/ip_icmp.h>    /* required for icmp_var.h */
  83 #include <netinet/icmp_var.h>   /* for ICMP_BANDLIM */
  84 #include <netinet/ip_var.h>
  85 #include <netinet/ip6.h>
  86 #include <netinet6/in6_pcb.h>
  87 #include <netinet6/ip6_var.h>
  88 #define TCPOUTFLAGS
  89 #include <netinet/tcp.h>
  90 #include <netinet/tcp_fsm.h>
  91 #include <netinet/tcp_seq.h>
  92 #include <netinet/tcp_timer.h>
  93 #include <netinet/tcp_var.h>
  94 #include <netinet/tcpip.h>
  95 #include <netinet/tcp_hpts.h>
  96 #include <netinet/cc/cc.h>
  97 #include <netinet/tcp_log_buf.h>
  98 #ifdef TCPDEBUG
  99 #include <netinet/tcp_debug.h>
 100 #endif                          /* TCPDEBUG */
 101 #ifdef TCP_OFFLOAD
 102 #include <netinet/tcp_offload.h>
 103 #endif
 104 #ifdef INET6
 105 #include <netinet6/tcp6_var.h>
 106 #endif
 107 #include <netinet/tcp_fastopen.h>
 108
 109 #include <netipsec/ipsec_support.h>
 110 #include <net/if.h>
 111 #include <net/if_var.h>
 112
 113 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
 114 #include <netipsec/ipsec.h>
 115 #include <netipsec/ipsec6.h>
 116 #endif                          /* IPSEC */
 117
 118 #include <netinet/udp.h>
 119 #include <netinet/udp_var.h>
 120 #include <machine/in_cksum.h>
 121
 122 #ifdef MAC
 123 #include <security/mac/mac_framework.h>
 124 #endif
 125 #include "rack_bbr_common.h"
 126
 127 /*
 128  * Common TCP Functions - These are shared by borth
 129  * rack and BBR.
 130  */
 131
 132
 133 #ifdef KERN_TLS
 134 uint32_t
 135 ctf_get_opt_tls_size(struct socket *so, uint32_t rwnd)
 136 {
 137         struct sbtls_info *tls;
 138         uint32_t len;
 139
 140 again:
 141         tls = so->so_snd.sb_tls_info;
 142         len = tls->sb_params.sb_maxlen;         /* max tls payload */
 143         len += tls->sb_params.sb_tls_hlen;      /* tls header len  */
 144         len += tls->sb_params.sb_tls_tlen;      /* tls trailer len */
 145         if ((len * 4) > rwnd) {
 146                 /*
 147                  * Stroke this will suck counter and what
 148                  * else should we do Drew? From the
 149                  * TCP perspective I am not sure
 150                  * what should be done...
 151                  */
 152                 if (tls->sb_params.sb_maxlen > 4096) {
 153                         tls->sb_params.sb_maxlen -= 4096;
 154                         if (tls->sb_params.sb_maxlen < 4096)
 155                                 tls->sb_params.sb_maxlen = 4096;
 156                         goto again;
 157                 }
 158         }
 159         return (len);
 160 }
 161 #endif
 162
 163 int
 164 ctf_process_inbound_raw(struct tcpcb *tp, struct socket *so, struct mbuf *m, int has_pkt)
 165 {
 166         /*
 167          * We are passed a raw change of mbuf packets
 168          * that arrived in LRO. They are linked via
 169          * the m_nextpkt link in the pkt-headers.
 170          *
 171          * We process each one by:
 172          * a) saving off the next
 173          * b) stripping off the ether-header
 174          * c) formulating the arguments for
 175          *    the tfb_tcp_hpts_do_segment
 176          * d) calling each mbuf to tfb_tcp_hpts_do_segment
 177          *    after adjusting the time to match the arrival time.
 178          * Note that the LRO code assures no IP options are present.
 179          *
 180          * The symantics for calling tfb_tcp_hpts_do_segment are the
 181          * following:
 182          * 1) It returns 0 if all went well and you (the caller) need
 183          *    to release the lock.
 184          * 2) If nxt_pkt is set, then the function will surpress calls
 185          *    to tfb_tcp_output() since you are promising to call again
 186          *    with another packet.
 187          * 3) If it returns 1, then you must free all the packets being
 188          *    shipped in, the tcb has been destroyed (or about to be destroyed).
 189          */
 190         struct mbuf *m_save;
 191         struct ether_header *eh;
 192         struct epoch_tracker et;
 193         struct tcphdr *th;
 194 #ifdef INET6
 195         struct ip6_hdr *ip6 = NULL;     /* Keep compiler happy. */
 196 #endif
 197 #ifdef INET
 198         struct ip *ip = NULL;           /* Keep compiler happy. */
 199 #endif
 200         struct ifnet *ifp;
 201         struct timeval tv;
 202         int32_t retval, nxt_pkt, tlen, off;
 203         uint16_t etype;
 204         uint16_t drop_hdrlen;
 205         uint8_t iptos, no_vn=0, bpf_req=0;
 206
 207         /*
 208          * This is a bit deceptive, we get the
 209          * "info epoch" which is really the network
 210          * epoch. This covers us on both any INP
 211          * type change but also if the ifp goes
 212          * away it covers us as well.
 213          */
 214         INP_INFO_RLOCK_ET(&V_tcbinfo, et);
 215         if (m && m->m_pkthdr.rcvif)
 216                 ifp = m->m_pkthdr.rcvif;
 217         else
 218                 ifp = NULL;
 219         if (ifp) {
 220                 bpf_req = bpf_peers_present(ifp->if_bpf);
 221         } else  {
 222                 /*
 223                  * We probably should not work around
 224                  * but kassert, since lro alwasy sets rcvif.
 225                  */
 226                 no_vn = 1;
 227                 goto skip_vnet;
 228         }
 229         CURVNET_SET(ifp->if_vnet);
 230 skip_vnet:
 231         while (m) {
 232                 m_save = m->m_nextpkt;
 233                 m->m_nextpkt = NULL;
 234                 /* Now lets get the ether header */
 235                 eh = mtod(m, struct ether_header *);
 236                 etype = ntohs(eh->ether_type);
 237                 /* Let the BPF see the packet */
 238                 if (bpf_req && ifp)
 239                         ETHER_BPF_MTAP(ifp, m);
 240                 m_adj(m,  sizeof(*eh));
 241                 /* Trim off the ethernet header */
 242                 switch (etype) {
 243 #ifdef INET6
 244                 case ETHERTYPE_IPV6:
 245                 {
 246                         if (m->m_len < (sizeof(*ip6) + sizeof(*th))) {
 247                                 m = m_pullup(m, sizeof(*ip6) + sizeof(*th));
 248                                 if (m == NULL) {
 249                                         TCPSTAT_INC(tcps_rcvshort);
 250                                         m_freem(m);
 251                                         goto skipped_pkt;
 252                                 }
 253                         }
 254                         ip6 = (struct ip6_hdr *)(eh + 1);
 255                         th = (struct tcphdr *)(ip6 + 1);
 256                         tlen = ntohs(ip6->ip6_plen);
 257                         drop_hdrlen = sizeof(*ip6);
 258                         if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
 259                                 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
 260                                         th->th_sum = m->m_pkthdr.csum_data;
 261                                 else
 262                                         th->th_sum = in6_cksum_pseudo(ip6, tlen,
 263                                                                       IPPROTO_TCP, m->m_pkthdr.csum_data);
 264                                 th->th_sum ^= 0xffff;
 265                         } else
 266                                 th->th_sum = in6_cksum(m, IPPROTO_TCP, drop_hdrlen, tlen);
 267                         if (th->th_sum) {
 268                                 TCPSTAT_INC(tcps_rcvbadsum);
 269                                 m_freem(m);
 270                                 goto skipped_pkt;
 271                         }
 272                         /*
 273                          * Be proactive about unspecified IPv6 address in source.
 274                          * As we use all-zero to indicate unbounded/unconnected pcb,
 275                          * unspecified IPv6 address can be used to confuse us.
 276                          *
 277                          * Note that packets with unspecified IPv6 destination is
 278                          * already dropped in ip6_input.
 279                          */
 280                         if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
 281                                 /* XXX stat */
 282                                 m_freem(m);
 283                                 goto skipped_pkt;
 284                         }
 285                         iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
 286                         break;
 287                 }
 288 #endif
 289 #ifdef INET
 290                 case ETHERTYPE_IP:
 291                 {
 292                         if (m->m_len < sizeof (struct tcpiphdr)) {
 293                                 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
 294                                     == NULL) {
 295                                         TCPSTAT_INC(tcps_rcvshort);
 296                                         m_freem(m);
 297                                         goto skipped_pkt;
 298                                 }
 299                         }
 300                         ip = (struct ip *)(eh + 1);
 301                         th = (struct tcphdr *)(ip + 1);
 302                         drop_hdrlen = sizeof(*ip);
 303                         iptos = ip->ip_tos;
 304                         tlen = ntohs(ip->ip_len) - sizeof(struct ip);
 305                         if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
 306                                 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
 307                                         th->th_sum = m->m_pkthdr.csum_data;
 308                                 else
 309                                         th->th_sum = in_pseudo(ip->ip_src.s_addr,
 310                                                                ip->ip_dst.s_addr,
 311                                                                htonl(m->m_pkthdr.csum_data + tlen +
 312                                                                      IPPROTO_TCP));
 313                                 th->th_sum ^= 0xffff;
 314                         } else {
 315                                 int len;
 316                                 struct ipovly *ipov = (struct ipovly *)ip;
 317                                 /*
 318                                  * Checksum extended TCP header and data.
 319                                  */
 320                                 len = drop_hdrlen + tlen;
 321                                 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
 322                                 ipov->ih_len = htons(tlen);
 323                                 th->th_sum = in_cksum(m, len);
 324                                 /* Reset length for SDT probes. */
 325                                 ip->ip_len = htons(len);
 326                                 /* Reset TOS bits */
 327                                 ip->ip_tos = iptos;
 328                                 /* Re-initialization for later version check */
 329                                 ip->ip_v = IPVERSION;
 330                                 ip->ip_hl = sizeof(*ip) >> 2;
 331                         }
 332                         if (th->th_sum) {
 333                                 TCPSTAT_INC(tcps_rcvbadsum);
 334                                 m_freem(m);
 335                                 goto skipped_pkt;
 336                         }
 337                         break;
 338                 }
 339 #endif
 340                 }
 341                 /*
 342                  * Convert TCP protocol specific fields to host format.
 343                  */
 344                 tcp_fields_to_host(th);
 345
 346                 off = th->th_off << 2;
 347                 if (off < sizeof (struct tcphdr) || off > tlen) {
 348                         TCPSTAT_INC(tcps_rcvbadoff);
 349                                 m_freem(m);
 350                                 goto skipped_pkt;
 351                 }
 352                 tlen -= off;
 353                 drop_hdrlen += off;
 354                 /*
 355                  * Now lets setup the timeval to be when we should
 356                  * have been called (if we can).
 357                  */
 358                 m->m_pkthdr.lro_nsegs = 1;
 359                 if (m->m_flags & M_TSTMP_LRO) {
 360                         tv.tv_sec = m->m_pkthdr.rcv_tstmp / 1000000000;
 361                         tv.tv_usec = (m->m_pkthdr.rcv_tstmp % 1000000000) / 1000;
 362                 } else {
 363                         /* Should not be should we kassert instead? */
 364                         tcp_get_usecs(&tv);
 365                 }
 366                 /* Now what about next packet? */
 367                 if (m_save || has_pkt)
 368                         nxt_pkt = 1;
 369                 else
 370                         nxt_pkt = 0;
 371                 retval = (*tp->t_fb->tfb_do_segment_nounlock)(m, th, so, tp, drop_hdrlen, tlen,
 372                                                               iptos, nxt_pkt, &tv);
 373                 if (retval) {
 374                         /* We lost the lock and tcb probably */
 375                         m = m_save;
 376                         while (m) {
 377                                 m_save = m->m_nextpkt;
 378                                 m->m_nextpkt = NULL;
 379                                 m_freem(m);
 380                                 m = m_save;
 381                         }
 382                         if (no_vn == 0)
 383                                 CURVNET_RESTORE();
 384                         INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
 385                         return (retval);
 386                 }
 387 skipped_pkt:
 388                 m = m_save;
 389         }
 390         if (no_vn == 0)
 391                 CURVNET_RESTORE();
 392         INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
 393         return (retval);
 394 }
 395
 396 int
 397 ctf_do_queued_segments(struct socket *so, struct tcpcb *tp, int have_pkt)
 398 {
 399         struct mbuf *m;
 400
 401         /* First lets see if we have old packets */
 402         if (tp->t_in_pkt) {
 403                 m = tp->t_in_pkt;
 404                 tp->t_in_pkt = NULL;
 405                 tp->t_tail_pkt = NULL;
 406                 if (ctf_process_inbound_raw(tp, so, m, have_pkt)) {
 407                         /* We lost the tcpcb (maybe a RST came in)? */
 408                         return (1);
 409                 }
 410         }
 411         return (0);
 412 }
 413
 414 uint32_t
 415 ctf_outstanding(struct tcpcb *tp)
 416 {
 417         return (tp->snd_max - tp->snd_una);
 418 }
 419
 420 uint32_t
 421 ctf_flight_size(struct tcpcb *tp, uint32_t rc_sacked)
 422 {
 423         if (rc_sacked <= ctf_outstanding(tp))
 424                 return (ctf_outstanding(tp) - rc_sacked);
 425         else {
 426                 /* TSNH */
 427 #ifdef INVARIANTS
 428                 panic("tp:%p rc_sacked:%d > out:%d",
 429                       tp, rc_sacked, ctf_outstanding(tp));
 430 #endif
 431                 return (0);
 432         }
 433 }
 434
 435 void
 436 ctf_do_dropwithreset(struct mbuf *m, struct tcpcb *tp, struct tcphdr *th,
 437     int32_t rstreason, int32_t tlen)
 438 {
 439         if (tp != NULL) {
 440                 tcp_dropwithreset(m, th, tp, tlen, rstreason);
 441                 INP_WUNLOCK(tp->t_inpcb);
 442         } else
 443                 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
 444 }
 445
 446 /*
 447  * ctf_drop_checks returns 1 for you should not proceed. It places
 448  * in ret_val what should be returned 1/0 by the caller. The 1 indicates
 449  * that the TCB is unlocked and probably dropped. The 0 indicates the
 450  * TCB is still valid and locked.
 451  */
 452 int
 453 ctf_drop_checks(struct tcpopt *to, struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, int32_t * tlenp,  int32_t * thf, int32_t * drop_hdrlen, int32_t * ret_val)
 454 {
 455         int32_t todrop;
 456         int32_t thflags;
 457         int32_t tlen;
 458
 459         thflags = *thf;
 460         tlen = *tlenp;
 461         todrop = tp->rcv_nxt - th->th_seq;
 462         if (todrop > 0) {
 463                 if (thflags & TH_SYN) {
 464                         thflags &= ~TH_SYN;
 465                         th->th_seq++;
 466                         if (th->th_urp > 1)
 467                                 th->th_urp--;
 468                         else
 469                                 thflags &= ~TH_URG;
 470                         todrop--;
 471                 }
 472                 /*
 473                  * Following if statement from Stevens, vol. 2, p. 960.
 474                  */
 475                 if (todrop > tlen
 476                     || (todrop == tlen && (thflags & TH_FIN) == 0)) {
 477                         /*
 478                          * Any valid FIN must be to the left of the window.
 479                          * At this point the FIN must be a duplicate or out
 480                          * of sequence; drop it.
 481                          */
 482                         thflags &= ~TH_FIN;
 483                         /*
 484                          * Send an ACK to resynchronize and drop any data.
 485                          * But keep on processing for RST or ACK.
 486                          */
 487                         tp->t_flags |= TF_ACKNOW;
 488                         todrop = tlen;
 489                         TCPSTAT_INC(tcps_rcvduppack);
 490                         TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
 491                 } else {
 492                         TCPSTAT_INC(tcps_rcvpartduppack);
 493                         TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
 494                 }
 495                 /*
 496                  * DSACK - add SACK block for dropped range
 497                  */
 498                 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
 499                         tcp_update_sack_list(tp, th->th_seq, th->th_seq + tlen);
 500                         /*
 501                          * ACK now, as the next in-sequence segment
 502                          * will clear the DSACK block again
 503                          */
 504                         tp->t_flags |= TF_ACKNOW;
 505                 }
 506                 *drop_hdrlen += todrop; /* drop from the top afterwards */
 507                 th->th_seq += todrop;
 508                 tlen -= todrop;
 509                 if (th->th_urp > todrop)
 510                         th->th_urp -= todrop;
 511                 else {
 512                         thflags &= ~TH_URG;
 513                         th->th_urp = 0;
 514                 }
 515         }
 516         /*
 517          * If segment ends after window, drop trailing data (and PUSH and
 518          * FIN); if nothing left, just ACK.
 519          */
 520         todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
 521         if (todrop > 0) {
 522                 TCPSTAT_INC(tcps_rcvpackafterwin);
 523                 if (todrop >= tlen) {
 524                         TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
 525                         /*
 526                          * If window is closed can only take segments at
 527                          * window edge, and have to drop data and PUSH from
 528                          * incoming segments.  Continue processing, but
 529                          * remember to ack.  Otherwise, drop segment and
 530                          * ack.
 531                          */
 532                         if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
 533                                 tp->t_flags |= TF_ACKNOW;
 534                                 TCPSTAT_INC(tcps_rcvwinprobe);
 535                         } else {
 536                                 ctf_do_dropafterack(m, tp, th, thflags, tlen, ret_val);
 537                                 return (1);
 538                         }
 539                 } else
 540                         TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
 541                 m_adj(m, -todrop);
 542                 tlen -= todrop;
 543                 thflags &= ~(TH_PUSH | TH_FIN);
 544         }
 545         *thf = thflags;
 546         *tlenp = tlen;
 547         return (0);
 548 }
 549
 550 /*
 551  * The value in ret_val informs the caller
 552  * if we dropped the tcb (and lock) or not.
 553  * 1 = we dropped it, 0 = the TCB is still locked
 554  * and valid.
 555  */
 556 void
 557 ctf_do_dropafterack(struct mbuf *m, struct tcpcb *tp, struct tcphdr *th, int32_t thflags, int32_t tlen, int32_t * ret_val)
 558 {
 559         /*
 560          * Generate an ACK dropping incoming segment if it occupies sequence
 561          * space, where the ACK reflects our state.
 562          *
 563          * We can now skip the test for the RST flag since all paths to this
 564          * code happen after packets containing RST have been dropped.
 565          *
 566          * In the SYN-RECEIVED state, don't send an ACK unless the segment
 567          * we received passes the SYN-RECEIVED ACK test. If it fails send a
 568          * RST.  This breaks the loop in the "LAND" DoS attack, and also
 569          * prevents an ACK storm between two listening ports that have been
 570          * sent forged SYN segments, each with the source address of the
 571          * other.
 572          */
 573         if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
 574             (SEQ_GT(tp->snd_una, th->th_ack) ||
 575             SEQ_GT(th->th_ack, tp->snd_max))) {
 576                 *ret_val = 1;
 577                 ctf_do_dropwithreset(m, tp, th, BANDLIM_RST_OPENPORT, tlen);
 578                 return;
 579         } else
 580                 *ret_val = 0;
 581         tp->t_flags |= TF_ACKNOW;
 582         if (m)
 583                 m_freem(m);
 584 }
 585
 586 void
 587 ctf_do_drop(struct mbuf *m, struct tcpcb *tp)
 588 {
 589
 590         /*
 591          * Drop space held by incoming segment and return.
 592          */
 593         if (tp != NULL)
 594                 INP_WUNLOCK(tp->t_inpcb);
 595         if (m)
 596                 m_freem(m);
 597 }
 598
 599 int
 600 ctf_process_rst(struct mbuf *m, struct tcphdr *th, struct socket *so, struct tcpcb *tp)
 601 {
 602         /*
 603          * RFC5961 Section 3.2
 604          *
 605          * - RST drops connection only if SEG.SEQ == RCV.NXT. - If RST is in
 606          * window, we send challenge ACK.
 607          *
 608          * Note: to take into account delayed ACKs, we should test against
 609          * last_ack_sent instead of rcv_nxt. Note 2: we handle special case
 610          * of closed window, not covered by the RFC.
 611          */
 612         int dropped = 0;
 613
 614         if ((SEQ_GEQ(th->th_seq, (tp->last_ack_sent - 1)) &&
 615             SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
 616             (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
 617
 618                 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
 619                 KASSERT(tp->t_state != TCPS_SYN_SENT,
 620                     ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
 621                     __func__, th, tp));
 622
 623                 if (V_tcp_insecure_rst ||
 624                     (tp->last_ack_sent == th->th_seq) ||
 625                     (tp->rcv_nxt == th->th_seq) ||
 626                     ((tp->last_ack_sent - 1) == th->th_seq)) {
 627                         TCPSTAT_INC(tcps_drops);
 628                         /* Drop the connection. */
 629                         switch (tp->t_state) {
 630                         case TCPS_SYN_RECEIVED:
 631                                 so->so_error = ECONNREFUSED;
 632                                 goto close;
 633                         case TCPS_ESTABLISHED:
 634                         case TCPS_FIN_WAIT_1:
 635                         case TCPS_FIN_WAIT_2:
 636                         case TCPS_CLOSE_WAIT:
 637                         case TCPS_CLOSING:
 638                         case TCPS_LAST_ACK:
 639                                 so->so_error = ECONNRESET;
 640                 close:
 641                                 tcp_state_change(tp, TCPS_CLOSED);
 642                                 /* FALLTHROUGH */
 643                         default:
 644                                 tp = tcp_close(tp);
 645                         }
 646                         dropped = 1;
 647                         ctf_do_drop(m, tp);
 648                 } else {
 649                         TCPSTAT_INC(tcps_badrst);
 650                         /* Send challenge ACK. */
 651                         tcp_respond(tp, mtod(m, void *), th, m,
 652                             tp->rcv_nxt, tp->snd_nxt, TH_ACK);
 653                         tp->last_ack_sent = tp->rcv_nxt;
 654                 }
 655         } else {
 656                 m_freem(m);
 657         }
 658         return (dropped);
 659 }
 660
 661 /*
 662  * The value in ret_val informs the caller
 663  * if we dropped the tcb (and lock) or not.
 664  * 1 = we dropped it, 0 = the TCB is still locked
 665  * and valid.
 666  */
 667 void
 668 ctf_challenge_ack(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, int32_t * ret_val)
 669 {
 670         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
 671
 672         TCPSTAT_INC(tcps_badsyn);
 673         if (V_tcp_insecure_syn &&
 674             SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
 675             SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
 676                 tp = tcp_drop(tp, ECONNRESET);
 677                 *ret_val = 1;
 678                 ctf_do_drop(m, tp);
 679         } else {
 680                 /* Send challenge ACK. */
 681                 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
 682                     tp->snd_nxt, TH_ACK);
 683                 tp->last_ack_sent = tp->rcv_nxt;
 684                 m = NULL;
 685                 *ret_val = 0;
 686                 ctf_do_drop(m, NULL);
 687         }
 688 }
 689
 690 /*
 691  * bbr_ts_check returns 1 for you should not proceed, the state
 692  * machine should return. It places in ret_val what should
 693  * be returned 1/0 by the caller (hpts_do_segment). The 1 indicates
 694  * that the TCB is unlocked and probably dropped. The 0 indicates the
 695  * TCB is still valid and locked.
 696  */
 697 int
 698 ctf_ts_check(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
 699     int32_t tlen, int32_t thflags, int32_t * ret_val)
 700 {
 701
 702         if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
 703                 /*
 704                  * Invalidate ts_recent.  If this segment updates ts_recent,
 705                  * the age will be reset later and ts_recent will get a
 706                  * valid value.  If it does not, setting ts_recent to zero
 707                  * will at least satisfy the requirement that zero be placed
 708                  * in the timestamp echo reply when ts_recent isn't valid.
 709                  * The age isn't reset until we get a valid ts_recent
 710                  * because we don't want out-of-order segments to be dropped
 711                  * when ts_recent is old.
 712                  */
 713                 tp->ts_recent = 0;
 714         } else {
 715                 TCPSTAT_INC(tcps_rcvduppack);
 716                 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
 717                 TCPSTAT_INC(tcps_pawsdrop);
 718                 *ret_val = 0;
 719                 if (tlen) {
 720                         ctf_do_dropafterack(m, tp, th, thflags, tlen, ret_val);
 721                 } else {
 722                         ctf_do_drop(m, NULL);
 723                 }
 724                 return (1);
 725         }
 726         return (0);
 727 }
 728
 729 void
 730 ctf_calc_rwin(struct socket *so, struct tcpcb *tp)
 731 {
 732         int32_t win;
 733
 734         /*
 735          * Calculate amount of space in receive window, and then do TCP
 736          * input processing. Receive window is amount of space in rcv queue,
 737          * but not less than advertised window.
 738          */
 739         win = sbspace(&so->so_rcv);
 740         if (win < 0)
 741                 win = 0;
 742         tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
 743 }
 744
 745 void
 746 ctf_do_dropwithreset_conn(struct mbuf *m, struct tcpcb *tp, struct tcphdr *th,
 747     int32_t rstreason, int32_t tlen)
 748 {
 749
 750         if (tp->t_inpcb) {
 751                 tcp_set_inp_to_drop(tp->t_inpcb, ETIMEDOUT);
 752         }
 753         tcp_dropwithreset(m, th, tp, tlen, rstreason);
 754         INP_WUNLOCK(tp->t_inpcb);
 755 }
 756
 757 uint32_t
 758 ctf_fixed_maxseg(struct tcpcb *tp)
 759 {
 760         int optlen;
 761
 762         if (tp->t_flags & TF_NOOPT)
 763                 return (tp->t_maxseg);
 764
 765         /*
 766          * Here we have a simplified code from tcp_addoptions(),
 767          * without a proper loop, and having most of paddings hardcoded.
 768          * We only consider fixed options that we would send every
 769          * time I.e. SACK is not considered.
 770          *
 771          */
 772 #define PAD(len)        ((((len) / 4) + !!((len) % 4)) * 4)
 773         if (TCPS_HAVEESTABLISHED(tp->t_state)) {
 774                 if (tp->t_flags & TF_RCVD_TSTMP)
 775                         optlen = TCPOLEN_TSTAMP_APPA;
 776                 else
 777                         optlen = 0;
 778 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
 779                 if (tp->t_flags & TF_SIGNATURE)
 780                         optlen += PAD(TCPOLEN_SIGNATURE);
 781 #endif
 782         } else {
 783                 if (tp->t_flags & TF_REQ_TSTMP)
 784                         optlen = TCPOLEN_TSTAMP_APPA;
 785                 else
 786                         optlen = PAD(TCPOLEN_MAXSEG);
 787                 if (tp->t_flags & TF_REQ_SCALE)
 788                         optlen += PAD(TCPOLEN_WINDOW);
 789 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
 790                 if (tp->t_flags & TF_SIGNATURE)
 791                         optlen += PAD(TCPOLEN_SIGNATURE);
 792 #endif
 793                 if (tp->t_flags & TF_SACK_PERMIT)
 794                         optlen += PAD(TCPOLEN_SACK_PERMITTED);
 795         }
 796 #undef PAD
 797         optlen = min(optlen, TCP_MAXOLEN);
 798         return (tp->t_maxseg - optlen);
 799 }
 800
 801 void
 802 ctf_log_sack_filter(struct tcpcb *tp, int num_sack_blks, struct sackblk *sack_blocks)
 803 {
 804         if (tp->t_logstate != TCP_LOG_STATE_OFF) {
 805                 union tcp_log_stackspecific log;
 806                 struct timeval tv;
 807
 808                 memset(&log, 0, sizeof(log));
 809                 log.u_bbr.timeStamp = tcp_get_usecs(&tv);
 810                 log.u_bbr.flex8 = num_sack_blks;
 811                 if (num_sack_blks > 0) {
 812                         log.u_bbr.flex1 = sack_blocks[0].start;
 813                         log.u_bbr.flex2 = sack_blocks[0].end;
 814                 }
 815                 if (num_sack_blks > 1) {
 816                         log.u_bbr.flex3 = sack_blocks[1].start;
 817                         log.u_bbr.flex4 = sack_blocks[1].end;
 818                 }
 819                 if (num_sack_blks > 2) {
 820                         log.u_bbr.flex5 = sack_blocks[2].start;
 821                         log.u_bbr.flex6 = sack_blocks[2].end;
 822                 }
 823                 if (num_sack_blks > 3) {
 824                         log.u_bbr.applimited = sack_blocks[3].start;
 825                         log.u_bbr.pkts_out = sack_blocks[3].end;
 826                 }
 827                 TCP_LOG_EVENTP(tp, NULL,
 828                     &tp->t_inpcb->inp_socket->so_rcv,
 829                     &tp->t_inpcb->inp_socket->so_snd,
 830                     TCP_SACK_FILTER_RES, 0,
 831                     0, &log, false, &tv);
 832         }
 833 }
 834
 835 uint32_t
 836 ctf_decay_count(uint32_t count, uint32_t decay)
 837 {
 838         /*
 839          * Given a count, decay it by a set percentage. The
 840          * percentage is in thousands i.e. 100% = 1000,
 841          * 19.3% = 193.
 842          */
 843         uint64_t perc_count, decay_per;
 844         uint32_t decayed_count;
 845         if (decay > 1000) {
 846                 /* We don't raise it */
 847                 return (count);
 848         }
 849         perc_count = count;
 850         decay_per = decay;
 851         perc_count *= decay_per;
 852         perc_count /= 1000;
 853         /*
 854          * So now perc_count holds the
 855          * count decay value.
 856          */
 857         decayed_count = count - (uint32_t)perc_count;
 858         return (decayed_count);
 859 }