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