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/qmath.h> 54 #include <sys/socket.h> 55 #include <sys/socketvar.h> 56 #ifdef KERN_TLS 57 #include <sys/ktls.h> 58 #endif 59 #include <sys/sysctl.h> 60 #include <sys/systm.h> 61 #include <sys/tree.h> 62 #ifdef NETFLIX_STATS 63 #include <sys/stats.h> /* Must come after qmath.h and tree.h */ 64 #endif 65 #include <sys/refcount.h> 66 #include <sys/queue.h> 67 #include <sys/smp.h> 68 #include <sys/kthread.h> 69 #include <sys/lock.h> 70 #include <sys/mutex.h> 71 #include <sys/tim_filter.h> 72 #include <sys/time.h> 73 #include <vm/uma.h> 74 #include <sys/kern_prefetch.h> 75 76 #include <net/route.h> 77 #include <net/vnet.h> 78 #include <net/ethernet.h> 79 #include <net/bpf.h> 80 81 #define TCPSTATES /* for logging */ 82 83 #include <netinet/in.h> 84 #include <netinet/in_kdtrace.h> 85 #include <netinet/in_pcb.h> 86 #include <netinet/ip.h> 87 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 88 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 89 #include <netinet/ip_var.h> 90 #include <netinet/ip6.h> 91 #include <netinet6/in6_pcb.h> 92 #include <netinet6/ip6_var.h> 93 #define TCPOUTFLAGS 94 #include <netinet/tcp.h> 95 #include <netinet/tcp_fsm.h> 96 #include <netinet/tcp_seq.h> 97 #include <netinet/tcp_timer.h> 98 #include <netinet/tcp_var.h> 99 #include <netinet/tcpip.h> 100 #include <netinet/tcp_hpts.h> 101 #include <netinet/cc/cc.h> 102 #include <netinet/tcp_log_buf.h> 103 #ifdef TCPDEBUG 104 #include <netinet/tcp_debug.h> 105 #endif /* TCPDEBUG */ 106 #ifdef TCP_OFFLOAD 107 #include <netinet/tcp_offload.h> 108 #endif 109 #ifdef INET6 110 #include <netinet6/tcp6_var.h> 111 #endif 112 #include <netinet/tcp_fastopen.h> 113 114 #include <netipsec/ipsec_support.h> 115 #include <net/if.h> 116 #include <net/if_var.h> 117 118 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 119 #include <netipsec/ipsec.h> 120 #include <netipsec/ipsec6.h> 121 #endif /* IPSEC */ 122 123 #include <netinet/udp.h> 124 #include <netinet/udp_var.h> 125 #include <machine/in_cksum.h> 126 127 #ifdef MAC 128 #include <security/mac/mac_framework.h> 129 #endif 130 #include "rack_bbr_common.h" 131 132 /* 133 * Common TCP Functions - These are shared by borth 134 * rack and BBR. 135 */ 136 137 138 #ifdef KERN_TLS 139 uint32_t 140 ctf_get_opt_tls_size(struct socket *so, uint32_t rwnd) 141 { 142 struct ktls_session *tls; 143 uint32_t len; 144 145 again: 146 tls = so->so_snd.sb_tls_info; 147 len = tls->params.max_frame_len; /* max tls payload */ 148 len += tls->params.tls_hlen; /* tls header len */ 149 len += tls->params.tls_tlen; /* tls trailer len */ 150 if ((len * 4) > rwnd) { 151 /* 152 * Stroke this will suck counter and what 153 * else should we do Drew? From the 154 * TCP perspective I am not sure 155 * what should be done... 156 */ 157 if (tls->params.max_frame_len > 4096) { 158 tls->params.max_frame_len -= 4096; 159 if (tls->params.max_frame_len < 4096) 160 tls->params.max_frame_len = 4096; 161 goto again; 162 } 163 } 164 return (len); 165 } 166 #endif 167 168 169 /* 170 * The function ctf_process_inbound_raw() is used by 171 * transport developers to do the steps needed to 172 * support MBUF Queuing i.e. the flags in 173 * inp->inp_flags2: 174 * 175 * - INP_SUPPORTS_MBUFQ 176 * - INP_MBUF_QUEUE_READY 177 * - INP_DONT_SACK_QUEUE 178 * 179 * These flags help control how LRO will deliver 180 * packets to the transport. You first set in inp_flags2 181 * the INP_SUPPORTS_MBUFQ to tell the LRO code that you 182 * will gladly take a queue of packets instead of a compressed 183 * single packet. You also set in your t_fb pointer the 184 * tfb_do_queued_segments to point to ctf_process_inbound_raw. 185 * 186 * This then gets you lists of inbound ACK's/Data instead 187 * of a condensed compressed ACK/DATA packet. Why would you 188 * want that? This will get you access to all the arrival 189 * times of at least LRO and possibly at the Hardware (if 190 * the interface card supports that) of the actual ACK/DATA. 191 * In some transport designs this is important since knowing 192 * the actual time we got the packet is useful information. 193 * 194 * Now there are some interesting Caveats that the transport 195 * designer needs to take into account when using this feature. 196 * 197 * 1) It is used with HPTS and pacing, when the pacing timer 198 * for output calls it will first call the input. 199 * 2) When you set INP_MBUF_QUEUE_READY this tells LRO 200 * queue normal packets, I am busy pacing out data and 201 * will process the queued packets before my tfb_tcp_output 202 * call from pacing. If a non-normal packet arrives, (e.g. sack) 203 * you will be awoken immediately. 204 * 3) Finally you can add the INP_DONT_SACK_QUEUE to not even 205 * be awoken if a SACK has arrived. You would do this when 206 * you were not only running a pacing for output timer 207 * but a Rack timer as well i.e. you know you are in recovery 208 * and are in the process (via the timers) of dealing with 209 * the loss. 210 * 211 * Now a critical thing you must be aware of here is that the 212 * use of the flags has a far greater scope then just your 213 * typical LRO. Why? Well thats because in the normal compressed 214 * LRO case at the end of a driver interupt all packets are going 215 * to get presented to the transport no matter if there is one 216 * or 100. With the MBUF_QUEUE model, this is not true. You will 217 * only be awoken to process the queue of packets when: 218 * a) The flags discussed above allow it. 219 * <or> 220 * b) You exceed a ack or data limit (by default the 221 * ack limit is infinity (64k acks) and the data 222 * limit is 64k of new TCP data) 223 * <or> 224 * c) The push bit has been set by the peer 225 */ 226 227 int 228 ctf_process_inbound_raw(struct tcpcb *tp, struct socket *so, struct mbuf *m, int has_pkt) 229 { 230 /* 231 * We are passed a raw change of mbuf packets 232 * that arrived in LRO. They are linked via 233 * the m_nextpkt link in the pkt-headers. 234 * 235 * We process each one by: 236 * a) saving off the next 237 * b) stripping off the ether-header 238 * c) formulating the arguments for 239 * the tfb_tcp_hpts_do_segment 240 * d) calling each mbuf to tfb_tcp_hpts_do_segment 241 * after adjusting the time to match the arrival time. 242 * Note that the LRO code assures no IP options are present. 243 * 244 * The symantics for calling tfb_tcp_hpts_do_segment are the 245 * following: 246 * 1) It returns 0 if all went well and you (the caller) need 247 * to release the lock. 248 * 2) If nxt_pkt is set, then the function will surpress calls 249 * to tfb_tcp_output() since you are promising to call again 250 * with another packet. 251 * 3) If it returns 1, then you must free all the packets being 252 * shipped in, the tcb has been destroyed (or about to be destroyed). 253 */ 254 struct mbuf *m_save; 255 struct ether_header *eh; 256 struct epoch_tracker et; 257 struct tcphdr *th; 258 #ifdef INET6 259 struct ip6_hdr *ip6 = NULL; /* Keep compiler happy. */ 260 #endif 261 #ifdef INET 262 struct ip *ip = NULL; /* Keep compiler happy. */ 263 #endif 264 struct ifnet *ifp; 265 struct timeval tv; 266 int32_t retval, nxt_pkt, tlen, off; 267 uint16_t etype; 268 uint16_t drop_hdrlen; 269 uint8_t iptos, no_vn=0, bpf_req=0; 270 271 /* 272 * This is a bit deceptive, we get the 273 * "info epoch" which is really the network 274 * epoch. This covers us on both any INP 275 * type change but also if the ifp goes 276 * away it covers us as well. 277 */ 278 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 279 if (m && m->m_pkthdr.rcvif) 280 ifp = m->m_pkthdr.rcvif; 281 else 282 ifp = NULL; 283 if (ifp) { 284 bpf_req = bpf_peers_present(ifp->if_bpf); 285 } else { 286 /* 287 * We probably should not work around 288 * but kassert, since lro alwasy sets rcvif. 289 */ 290 no_vn = 1; 291 goto skip_vnet; 292 } 293 CURVNET_SET(ifp->if_vnet); 294 skip_vnet: 295 while (m) { 296 m_save = m->m_nextpkt; 297 m->m_nextpkt = NULL; 298 /* Now lets get the ether header */ 299 eh = mtod(m, struct ether_header *); 300 etype = ntohs(eh->ether_type); 301 /* Let the BPF see the packet */ 302 if (bpf_req && ifp) 303 ETHER_BPF_MTAP(ifp, m); 304 m_adj(m, sizeof(*eh)); 305 /* Trim off the ethernet header */ 306 switch (etype) { 307 #ifdef INET6 308 case ETHERTYPE_IPV6: 309 { 310 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 311 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 312 if (m == NULL) { 313 TCPSTAT_INC(tcps_rcvshort); 314 m_freem(m); 315 goto skipped_pkt; 316 } 317 } 318 ip6 = (struct ip6_hdr *)(eh + 1); 319 th = (struct tcphdr *)(ip6 + 1); 320 tlen = ntohs(ip6->ip6_plen); 321 drop_hdrlen = sizeof(*ip6); 322 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 323 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 324 th->th_sum = m->m_pkthdr.csum_data; 325 else 326 th->th_sum = in6_cksum_pseudo(ip6, tlen, 327 IPPROTO_TCP, m->m_pkthdr.csum_data); 328 th->th_sum ^= 0xffff; 329 } else 330 th->th_sum = in6_cksum(m, IPPROTO_TCP, drop_hdrlen, tlen); 331 if (th->th_sum) { 332 TCPSTAT_INC(tcps_rcvbadsum); 333 m_freem(m); 334 goto skipped_pkt; 335 } 336 /* 337 * Be proactive about unspecified IPv6 address in source. 338 * As we use all-zero to indicate unbounded/unconnected pcb, 339 * unspecified IPv6 address can be used to confuse us. 340 * 341 * Note that packets with unspecified IPv6 destination is 342 * already dropped in ip6_input. 343 */ 344 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 345 /* XXX stat */ 346 m_freem(m); 347 goto skipped_pkt; 348 } 349 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 350 break; 351 } 352 #endif 353 #ifdef INET 354 case ETHERTYPE_IP: 355 { 356 if (m->m_len < sizeof (struct tcpiphdr)) { 357 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 358 == NULL) { 359 TCPSTAT_INC(tcps_rcvshort); 360 m_freem(m); 361 goto skipped_pkt; 362 } 363 } 364 ip = (struct ip *)(eh + 1); 365 th = (struct tcphdr *)(ip + 1); 366 drop_hdrlen = sizeof(*ip); 367 iptos = ip->ip_tos; 368 tlen = ntohs(ip->ip_len) - sizeof(struct ip); 369 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 370 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 371 th->th_sum = m->m_pkthdr.csum_data; 372 else 373 th->th_sum = in_pseudo(ip->ip_src.s_addr, 374 ip->ip_dst.s_addr, 375 htonl(m->m_pkthdr.csum_data + tlen + 376 IPPROTO_TCP)); 377 th->th_sum ^= 0xffff; 378 } else { 379 int len; 380 struct ipovly *ipov = (struct ipovly *)ip; 381 /* 382 * Checksum extended TCP header and data. 383 */ 384 len = drop_hdrlen + tlen; 385 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 386 ipov->ih_len = htons(tlen); 387 th->th_sum = in_cksum(m, len); 388 /* Reset length for SDT probes. */ 389 ip->ip_len = htons(len); 390 /* Reset TOS bits */ 391 ip->ip_tos = iptos; 392 /* Re-initialization for later version check */ 393 ip->ip_v = IPVERSION; 394 ip->ip_hl = sizeof(*ip) >> 2; 395 } 396 if (th->th_sum) { 397 TCPSTAT_INC(tcps_rcvbadsum); 398 m_freem(m); 399 goto skipped_pkt; 400 } 401 break; 402 } 403 #endif 404 } 405 /* 406 * Convert TCP protocol specific fields to host format. 407 */ 408 tcp_fields_to_host(th); 409 410 off = th->th_off << 2; 411 if (off < sizeof (struct tcphdr) || off > tlen) { 412 TCPSTAT_INC(tcps_rcvbadoff); 413 m_freem(m); 414 goto skipped_pkt; 415 } 416 tlen -= off; 417 drop_hdrlen += off; 418 /* 419 * Now lets setup the timeval to be when we should 420 * have been called (if we can). 421 */ 422 m->m_pkthdr.lro_nsegs = 1; 423 if (m->m_flags & M_TSTMP_LRO) { 424 tv.tv_sec = m->m_pkthdr.rcv_tstmp /1000000000; 425 tv.tv_usec = (m->m_pkthdr.rcv_tstmp % 1000000000)/1000; 426 } else { 427 /* Should not be should we kassert instead? */ 428 tcp_get_usecs(&tv); 429 } 430 /* Now what about next packet? */ 431 if (m_save || has_pkt) 432 nxt_pkt = 1; 433 else 434 nxt_pkt = 0; 435 retval = (*tp->t_fb->tfb_do_segment_nounlock)(m, th, so, tp, drop_hdrlen, tlen, 436 iptos, nxt_pkt, &tv); 437 if (retval) { 438 /* We lost the lock and tcb probably */ 439 m = m_save; 440 while(m) { 441 m_save = m->m_nextpkt; 442 m->m_nextpkt = NULL; 443 m_freem(m); 444 m = m_save; 445 } 446 if (no_vn == 0) 447 CURVNET_RESTORE(); 448 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 449 return(retval); 450 } 451 skipped_pkt: 452 m = m_save; 453 } 454 if (no_vn == 0) 455 CURVNET_RESTORE(); 456 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 457 return(retval); 458 } 459 460 int 461 ctf_do_queued_segments(struct socket *so, struct tcpcb *tp, int have_pkt) 462 { 463 struct mbuf *m; 464 465 /* First lets see if we have old packets */ 466 if (tp->t_in_pkt) { 467 m = tp->t_in_pkt; 468 tp->t_in_pkt = NULL; 469 tp->t_tail_pkt = NULL; 470 if (ctf_process_inbound_raw(tp, so, m, have_pkt)) { 471 /* We lost the tcpcb (maybe a RST came in)? */ 472 return(1); 473 } 474 } 475 return (0); 476 } 477 478 uint32_t 479 ctf_outstanding(struct tcpcb *tp) 480 { 481 return(tp->snd_max - tp->snd_una); 482 } 483 484 uint32_t 485 ctf_flight_size(struct tcpcb *tp, uint32_t rc_sacked) 486 { 487 if (rc_sacked <= ctf_outstanding(tp)) 488 return(ctf_outstanding(tp) - rc_sacked); 489 else { 490 /* TSNH */ 491 #ifdef INVARIANTS 492 panic("tp:%p rc_sacked:%d > out:%d", 493 tp, rc_sacked, ctf_outstanding(tp)); 494 #endif 495 return (0); 496 } 497 } 498 499 void 500 ctf_do_dropwithreset(struct mbuf *m, struct tcpcb *tp, struct tcphdr *th, 501 int32_t rstreason, int32_t tlen) 502 { 503 if (tp != NULL) { 504 tcp_dropwithreset(m, th, tp, tlen, rstreason); 505 INP_WUNLOCK(tp->t_inpcb); 506 } else 507 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 508 } 509 510 /* 511 * ctf_drop_checks returns 1 for you should not proceed. It places 512 * in ret_val what should be returned 1/0 by the caller. The 1 indicates 513 * that the TCB is unlocked and probably dropped. The 0 indicates the 514 * TCB is still valid and locked. 515 */ 516 int 517 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) 518 { 519 int32_t todrop; 520 int32_t thflags; 521 int32_t tlen; 522 523 thflags = *thf; 524 tlen = *tlenp; 525 todrop = tp->rcv_nxt - th->th_seq; 526 if (todrop > 0) { 527 if (thflags & TH_SYN) { 528 thflags &= ~TH_SYN; 529 th->th_seq++; 530 if (th->th_urp > 1) 531 th->th_urp--; 532 else 533 thflags &= ~TH_URG; 534 todrop--; 535 } 536 /* 537 * Following if statement from Stevens, vol. 2, p. 960. 538 */ 539 if (todrop > tlen 540 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 541 /* 542 * Any valid FIN must be to the left of the window. 543 * At this point the FIN must be a duplicate or out 544 * of sequence; drop it. 545 */ 546 thflags &= ~TH_FIN; 547 /* 548 * Send an ACK to resynchronize and drop any data. 549 * But keep on processing for RST or ACK. 550 */ 551 tp->t_flags |= TF_ACKNOW; 552 todrop = tlen; 553 TCPSTAT_INC(tcps_rcvduppack); 554 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 555 } else { 556 TCPSTAT_INC(tcps_rcvpartduppack); 557 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 558 } 559 /* 560 * DSACK - add SACK block for dropped range 561 */ 562 if (tp->t_flags & TF_SACK_PERMIT) { 563 tcp_update_sack_list(tp, th->th_seq, 564 th->th_seq + todrop); 565 /* 566 * ACK now, as the next in-sequence segment 567 * will clear the DSACK block again 568 */ 569 tp->t_flags |= TF_ACKNOW; 570 } 571 *drop_hdrlen += todrop; /* drop from the top afterwards */ 572 th->th_seq += todrop; 573 tlen -= todrop; 574 if (th->th_urp > todrop) 575 th->th_urp -= todrop; 576 else { 577 thflags &= ~TH_URG; 578 th->th_urp = 0; 579 } 580 } 581 /* 582 * If segment ends after window, drop trailing data (and PUSH and 583 * FIN); if nothing left, just ACK. 584 */ 585 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 586 if (todrop > 0) { 587 TCPSTAT_INC(tcps_rcvpackafterwin); 588 if (todrop >= tlen) { 589 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 590 /* 591 * If window is closed can only take segments at 592 * window edge, and have to drop data and PUSH from 593 * incoming segments. Continue processing, but 594 * remember to ack. Otherwise, drop segment and 595 * ack. 596 */ 597 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 598 tp->t_flags |= TF_ACKNOW; 599 TCPSTAT_INC(tcps_rcvwinprobe); 600 } else { 601 ctf_do_dropafterack(m, tp, th, thflags, tlen, ret_val); 602 return (1); 603 } 604 } else 605 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 606 m_adj(m, -todrop); 607 tlen -= todrop; 608 thflags &= ~(TH_PUSH | TH_FIN); 609 } 610 *thf = thflags; 611 *tlenp = tlen; 612 return (0); 613 } 614 615 /* 616 * The value in ret_val informs the caller 617 * if we dropped the tcb (and lock) or not. 618 * 1 = we dropped it, 0 = the TCB is still locked 619 * and valid. 620 */ 621 void 622 ctf_do_dropafterack(struct mbuf *m, struct tcpcb *tp, struct tcphdr *th, int32_t thflags, int32_t tlen, int32_t * ret_val) 623 { 624 /* 625 * Generate an ACK dropping incoming segment if it occupies sequence 626 * space, where the ACK reflects our state. 627 * 628 * We can now skip the test for the RST flag since all paths to this 629 * code happen after packets containing RST have been dropped. 630 * 631 * In the SYN-RECEIVED state, don't send an ACK unless the segment 632 * we received passes the SYN-RECEIVED ACK test. If it fails send a 633 * RST. This breaks the loop in the "LAND" DoS attack, and also 634 * prevents an ACK storm between two listening ports that have been 635 * sent forged SYN segments, each with the source address of the 636 * other. 637 */ 638 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 639 (SEQ_GT(tp->snd_una, th->th_ack) || 640 SEQ_GT(th->th_ack, tp->snd_max))) { 641 *ret_val = 1; 642 ctf_do_dropwithreset(m, tp, th, BANDLIM_RST_OPENPORT, tlen); 643 return; 644 } else 645 *ret_val = 0; 646 tp->t_flags |= TF_ACKNOW; 647 if (m) 648 m_freem(m); 649 } 650 651 void 652 ctf_do_drop(struct mbuf *m, struct tcpcb *tp) 653 { 654 655 /* 656 * Drop space held by incoming segment and return. 657 */ 658 if (tp != NULL) 659 INP_WUNLOCK(tp->t_inpcb); 660 if (m) 661 m_freem(m); 662 } 663 664 int 665 ctf_process_rst(struct mbuf *m, struct tcphdr *th, struct socket *so, struct tcpcb *tp) 666 { 667 /* 668 * RFC5961 Section 3.2 669 * 670 * - RST drops connection only if SEG.SEQ == RCV.NXT. - If RST is in 671 * window, we send challenge ACK. 672 * 673 * Note: to take into account delayed ACKs, we should test against 674 * last_ack_sent instead of rcv_nxt. Note 2: we handle special case 675 * of closed window, not covered by the RFC. 676 */ 677 int dropped = 0; 678 679 if ((SEQ_GEQ(th->th_seq, (tp->last_ack_sent - 1)) && 680 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 681 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 682 683 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 684 KASSERT(tp->t_state != TCPS_SYN_SENT, 685 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 686 __func__, th, tp)); 687 688 if (V_tcp_insecure_rst || 689 (tp->last_ack_sent == th->th_seq) || 690 (tp->rcv_nxt == th->th_seq) || 691 ((tp->last_ack_sent - 1) == th->th_seq)) { 692 TCPSTAT_INC(tcps_drops); 693 /* Drop the connection. */ 694 switch (tp->t_state) { 695 case TCPS_SYN_RECEIVED: 696 so->so_error = ECONNREFUSED; 697 goto close; 698 case TCPS_ESTABLISHED: 699 case TCPS_FIN_WAIT_1: 700 case TCPS_FIN_WAIT_2: 701 case TCPS_CLOSE_WAIT: 702 case TCPS_CLOSING: 703 case TCPS_LAST_ACK: 704 so->so_error = ECONNRESET; 705 close: 706 tcp_state_change(tp, TCPS_CLOSED); 707 /* FALLTHROUGH */ 708 default: 709 tp = tcp_close(tp); 710 } 711 dropped = 1; 712 ctf_do_drop(m, tp); 713 } else { 714 TCPSTAT_INC(tcps_badrst); 715 /* Send challenge ACK. */ 716 tcp_respond(tp, mtod(m, void *), th, m, 717 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 718 tp->last_ack_sent = tp->rcv_nxt; 719 } 720 } else { 721 m_freem(m); 722 } 723 return (dropped); 724 } 725 726 /* 727 * The value in ret_val informs the caller 728 * if we dropped the tcb (and lock) or not. 729 * 1 = we dropped it, 0 = the TCB is still locked 730 * and valid. 731 */ 732 void 733 ctf_challenge_ack(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, int32_t * ret_val) 734 { 735 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 736 737 TCPSTAT_INC(tcps_badsyn); 738 if (V_tcp_insecure_syn && 739 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 740 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 741 tp = tcp_drop(tp, ECONNRESET); 742 *ret_val = 1; 743 ctf_do_drop(m, tp); 744 } else { 745 /* Send challenge ACK. */ 746 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 747 tp->snd_nxt, TH_ACK); 748 tp->last_ack_sent = tp->rcv_nxt; 749 m = NULL; 750 *ret_val = 0; 751 ctf_do_drop(m, NULL); 752 } 753 } 754 755 /* 756 * bbr_ts_check returns 1 for you should not proceed, the state 757 * machine should return. It places in ret_val what should 758 * be returned 1/0 by the caller (hpts_do_segment). The 1 indicates 759 * that the TCB is unlocked and probably dropped. The 0 indicates the 760 * TCB is still valid and locked. 761 */ 762 int 763 ctf_ts_check(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 764 int32_t tlen, int32_t thflags, int32_t * ret_val) 765 { 766 767 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 768 /* 769 * Invalidate ts_recent. If this segment updates ts_recent, 770 * the age will be reset later and ts_recent will get a 771 * valid value. If it does not, setting ts_recent to zero 772 * will at least satisfy the requirement that zero be placed 773 * in the timestamp echo reply when ts_recent isn't valid. 774 * The age isn't reset until we get a valid ts_recent 775 * because we don't want out-of-order segments to be dropped 776 * when ts_recent is old. 777 */ 778 tp->ts_recent = 0; 779 } else { 780 TCPSTAT_INC(tcps_rcvduppack); 781 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 782 TCPSTAT_INC(tcps_pawsdrop); 783 *ret_val = 0; 784 if (tlen) { 785 ctf_do_dropafterack(m, tp, th, thflags, tlen, ret_val); 786 } else { 787 ctf_do_drop(m, NULL); 788 } 789 return (1); 790 } 791 return (0); 792 } 793 794 void 795 ctf_calc_rwin(struct socket *so, struct tcpcb *tp) 796 { 797 int32_t win; 798 799 /* 800 * Calculate amount of space in receive window, and then do TCP 801 * input processing. Receive window is amount of space in rcv queue, 802 * but not less than advertised window. 803 */ 804 win = sbspace(&so->so_rcv); 805 if (win < 0) 806 win = 0; 807 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 808 } 809 810 void 811 ctf_do_dropwithreset_conn(struct mbuf *m, struct tcpcb *tp, struct tcphdr *th, 812 int32_t rstreason, int32_t tlen) 813 { 814 815 if (tp->t_inpcb) { 816 tcp_set_inp_to_drop(tp->t_inpcb, ETIMEDOUT); 817 } 818 tcp_dropwithreset(m, th, tp, tlen, rstreason); 819 INP_WUNLOCK(tp->t_inpcb); 820 } 821 822 uint32_t 823 ctf_fixed_maxseg(struct tcpcb *tp) 824 { 825 int optlen; 826 827 if (tp->t_flags & TF_NOOPT) 828 return (tp->t_maxseg); 829 830 /* 831 * Here we have a simplified code from tcp_addoptions(), 832 * without a proper loop, and having most of paddings hardcoded. 833 * We only consider fixed options that we would send every 834 * time I.e. SACK is not considered. 835 * 836 */ 837 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) 838 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 839 if (tp->t_flags & TF_RCVD_TSTMP) 840 optlen = TCPOLEN_TSTAMP_APPA; 841 else 842 optlen = 0; 843 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 844 if (tp->t_flags & TF_SIGNATURE) 845 optlen += PAD(TCPOLEN_SIGNATURE); 846 #endif 847 } else { 848 if (tp->t_flags & TF_REQ_TSTMP) 849 optlen = TCPOLEN_TSTAMP_APPA; 850 else 851 optlen = PAD(TCPOLEN_MAXSEG); 852 if (tp->t_flags & TF_REQ_SCALE) 853 optlen += PAD(TCPOLEN_WINDOW); 854 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 855 if (tp->t_flags & TF_SIGNATURE) 856 optlen += PAD(TCPOLEN_SIGNATURE); 857 #endif 858 if (tp->t_flags & TF_SACK_PERMIT) 859 optlen += PAD(TCPOLEN_SACK_PERMITTED); 860 } 861 #undef PAD 862 optlen = min(optlen, TCP_MAXOLEN); 863 return (tp->t_maxseg - optlen); 864 } 865 866 void 867 ctf_log_sack_filter(struct tcpcb *tp, int num_sack_blks, struct sackblk *sack_blocks) 868 { 869 if (tp->t_logstate != TCP_LOG_STATE_OFF) { 870 union tcp_log_stackspecific log; 871 struct timeval tv; 872 873 memset(&log, 0, sizeof(log)); 874 log.u_bbr.timeStamp = tcp_get_usecs(&tv); 875 log.u_bbr.flex8 = num_sack_blks; 876 if (num_sack_blks > 0) { 877 log.u_bbr.flex1 = sack_blocks[0].start; 878 log.u_bbr.flex2 = sack_blocks[0].end; 879 } 880 if (num_sack_blks > 1) { 881 log.u_bbr.flex3 = sack_blocks[1].start; 882 log.u_bbr.flex4 = sack_blocks[1].end; 883 } 884 if (num_sack_blks > 2) { 885 log.u_bbr.flex5 = sack_blocks[2].start; 886 log.u_bbr.flex6 = sack_blocks[2].end; 887 } 888 if (num_sack_blks > 3) { 889 log.u_bbr.applimited = sack_blocks[3].start; 890 log.u_bbr.pkts_out = sack_blocks[3].end; 891 } 892 TCP_LOG_EVENTP(tp, NULL, 893 &tp->t_inpcb->inp_socket->so_rcv, 894 &tp->t_inpcb->inp_socket->so_snd, 895 TCP_SACK_FILTER_RES, 0, 896 0, &log, false, &tv); 897 } 898 } 899 900 uint32_t 901 ctf_decay_count(uint32_t count, uint32_t decay) 902 { 903 /* 904 * Given a count, decay it by a set percentage. The 905 * percentage is in thousands i.e. 100% = 1000, 906 * 19.3% = 193. 907 */ 908 uint64_t perc_count, decay_per; 909 uint32_t decayed_count; 910 if (decay > 1000) { 911 /* We don't raise it */ 912 return (count); 913 } 914 perc_count = count; 915 decay_per = decay; 916 perc_count *= decay_per; 917 perc_count /= 1000; 918 /* 919 * So now perc_count holds the 920 * count decay value. 921 */ 922 decayed_count = count - (uint32_t)perc_count; 923 return(decayed_count); 924 } 925