xref: /freebsd/sys/netinet/tcp_stacks/rack_bbr_common.c (revision d485c77f203fb0f4cdc08dea5ff81631b51d8809)
1 /*-
2  * Copyright (c) 2016-2020 Netflix, Inc.
3  *
4  * Redistribution and use in source and binary forms, with or without
5  * modification, are permitted provided that the following conditions
6  * are met:
7  * 1. Redistributions of source code must retain the above copyright
8  *    notice, this list of conditions and the following disclaimer.
9  * 2. Redistributions in binary form must reproduce the above copyright
10  *    notice, this list of conditions and the following disclaimer in the
11  *    documentation and/or other materials provided with the distribution.
12  *
13  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
14  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
17  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
23  * SUCH DAMAGE.
24  *
25  */
26 /*
27  * Author: Randall Stewart <rrs@netflix.com>
28  * This work is based on the ACM Queue paper
29  * BBR - Congestion Based Congestion Control
30  * and also numerous discussions with Neal, Yuchung and Van.
31  */
32 
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35 
36 #include "opt_inet.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_tcpdebug.h"
40 #include "opt_ratelimit.h"
41 #include "opt_kern_tls.h"
42 #include <sys/param.h>
43 #include <sys/arb.h>
44 #include <sys/module.h>
45 #include <sys/kernel.h>
46 #ifdef TCP_HHOOK
47 #include <sys/hhook.h>
48 #endif
49 #include <sys/malloc.h>
50 #include <sys/mbuf.h>
51 #include <sys/proc.h>
52 #include <sys/qmath.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #ifdef KERN_TLS
56 #include <sys/ktls.h>
57 #endif
58 #include <sys/sysctl.h>
59 #include <sys/systm.h>
60 #include <sys/tree.h>
61 #ifdef NETFLIX_STATS
62 #include <sys/stats.h> /* Must come after qmath.h and tree.h */
63 #endif
64 #include <sys/refcount.h>
65 #include <sys/queue.h>
66 #include <sys/smp.h>
67 #include <sys/kthread.h>
68 #include <sys/lock.h>
69 #include <sys/mutex.h>
70 #include <sys/tim_filter.h>
71 #include <sys/time.h>
72 #include <vm/uma.h>
73 #include <sys/kern_prefetch.h>
74 
75 #include <net/route.h>
76 #include <net/vnet.h>
77 #include <net/ethernet.h>
78 #include <net/bpf.h>
79 
80 #define TCPSTATES		/* for logging */
81 
82 #include <netinet/in.h>
83 #include <netinet/in_kdtrace.h>
84 #include <netinet/in_pcb.h>
85 #include <netinet/ip.h>
86 #include <netinet/ip_icmp.h>	/* required for icmp_var.h */
87 #include <netinet/icmp_var.h>	/* for ICMP_BANDLIM */
88 #include <netinet/ip_var.h>
89 #include <netinet/ip6.h>
90 #include <netinet6/in6_pcb.h>
91 #include <netinet6/ip6_var.h>
92 #include <netinet/tcp.h>
93 #include <netinet/tcp_fsm.h>
94 #include <netinet/tcp_seq.h>
95 #include <netinet/tcp_timer.h>
96 #include <netinet/tcp_var.h>
97 #include <netinet/tcpip.h>
98 #include <netinet/tcp_hpts.h>
99 #include <netinet/cc/cc.h>
100 #include <netinet/tcp_log_buf.h>
101 #ifdef TCPDEBUG
102 #include <netinet/tcp_debug.h>
103 #endif				/* TCPDEBUG */
104 #ifdef TCP_OFFLOAD
105 #include <netinet/tcp_offload.h>
106 #endif
107 #ifdef INET6
108 #include <netinet6/tcp6_var.h>
109 #endif
110 #include <netinet/tcp_fastopen.h>
111 
112 #include <netipsec/ipsec_support.h>
113 #include <net/if.h>
114 #include <net/if_var.h>
115 
116 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
117 #include <netipsec/ipsec.h>
118 #include <netipsec/ipsec6.h>
119 #endif				/* IPSEC */
120 
121 #include <netinet/udp.h>
122 #include <netinet/udp_var.h>
123 #include <machine/in_cksum.h>
124 
125 #ifdef MAC
126 #include <security/mac/mac_framework.h>
127 #endif
128 #include "rack_bbr_common.h"
129 
130 /*
131  * Common TCP Functions - These are shared by borth
132  * rack and BBR.
133  */
134 #ifdef KERN_TLS
135 uint32_t
136 ctf_get_opt_tls_size(struct socket *so, uint32_t rwnd)
137 {
138 	struct ktls_session *tls;
139 	uint32_t len;
140 
141 again:
142 	tls = so->so_snd.sb_tls_info;
143 	len = tls->params.max_frame_len;         /* max tls payload */
144 	len += tls->params.tls_hlen;      /* tls header len  */
145 	len += tls->params.tls_tlen;      /* tls trailer len */
146 	if ((len * 4) > rwnd) {
147 		/*
148 		 * Stroke this will suck counter and what
149 		 * else should we do Drew? From the
150 		 * TCP perspective I am not sure
151 		 * what should be done...
152 		 */
153 		if (tls->params.max_frame_len > 4096) {
154 			tls->params.max_frame_len -= 4096;
155 			if (tls->params.max_frame_len < 4096)
156 				tls->params.max_frame_len = 4096;
157 			goto again;
158 		}
159 	}
160 	return (len);
161 }
162 #endif
163 
164 /*
165  * The function ctf_process_inbound_raw() is used by
166  * transport developers to do the steps needed to
167  * support MBUF Queuing i.e. the flags in
168  * inp->inp_flags2:
169  *
170  * - INP_SUPPORTS_MBUFQ
171  * - INP_MBUF_QUEUE_READY
172  * - INP_DONT_SACK_QUEUE
173  *
174  * These flags help control how LRO will deliver
175  * packets to the transport. You first set in inp_flags2
176  * the INP_SUPPORTS_MBUFQ to tell the LRO code that you
177  * will gladly take a queue of packets instead of a compressed
178  * single packet. You also set in your t_fb pointer the
179  * tfb_do_queued_segments to point to ctf_process_inbound_raw.
180  *
181  * This then gets you lists of inbound ACK's/Data instead
182  * of a condensed compressed ACK/DATA packet. Why would you
183  * want that? This will get you access to all the arrival
184  * times of at least LRO and possibly at the Hardware (if
185  * the interface card supports that) of the actual ACK/DATA.
186  * In some transport designs this is important since knowing
187  * the actual time we got the packet is useful information.
188  *
189  * Now there are some interesting Caveats that the transport
190  * designer needs to take into account when using this feature.
191  *
192  * 1) It is used with HPTS and pacing, when the pacing timer
193  *    for output calls it will first call the input.
194  * 2) When you set INP_MBUF_QUEUE_READY this tells LRO
195  *    queue normal packets, I am busy pacing out data and
196  *    will process the queued packets before my tfb_tcp_output
197  *    call from pacing. If a non-normal packet arrives, (e.g. sack)
198  *    you will be awoken immediately.
199  * 3) Finally you can add the INP_DONT_SACK_QUEUE to not even
200  *    be awoken if a SACK has arrived. You would do this when
201  *    you were not only running a pacing for output timer
202  *    but a Rack timer as well i.e. you know you are in recovery
203  *    and are in the process (via the timers) of dealing with
204  *    the loss.
205  *
206  * Now a critical thing you must be aware of here is that the
207  * use of the flags has a far greater scope then just your
208  * typical LRO. Why? Well thats because in the normal compressed
209  * LRO case at the end of a driver interupt all packets are going
210  * to get presented to the transport no matter if there is one
211  * or 100. With the MBUF_QUEUE model, this is not true. You will
212  * only be awoken to process the queue of packets when:
213  *     a) The flags discussed above allow it.
214  *          <or>
215  *     b) You exceed a ack or data limit (by default the
216  *        ack limit is infinity (64k acks) and the data
217  *        limit is 64k of new TCP data)
218  *         <or>
219  *     c) The push bit has been set by the peer
220  */
221 
222 int
223 ctf_process_inbound_raw(struct tcpcb *tp, struct socket *so, struct mbuf *m, int has_pkt)
224 {
225 	/*
226 	 * We are passed a raw change of mbuf packets
227 	 * that arrived in LRO. They are linked via
228 	 * the m_nextpkt link in the pkt-headers.
229 	 *
230 	 * We process each one by:
231 	 * a) saving off the next
232 	 * b) stripping off the ether-header
233 	 * c) formulating the arguments for
234 	 *    the tfb_tcp_hpts_do_segment
235 	 * d) calling each mbuf to tfb_tcp_hpts_do_segment
236 	 *    after adjusting the time to match the arrival time.
237 	 * Note that the LRO code assures no IP options are present.
238 	 *
239 	 * The symantics for calling tfb_tcp_hpts_do_segment are the
240 	 * following:
241 	 * 1) It returns 0 if all went well and you (the caller) need
242 	 *    to release the lock.
243 	 * 2) If nxt_pkt is set, then the function will surpress calls
244 	 *    to tfb_tcp_output() since you are promising to call again
245 	 *    with another packet.
246 	 * 3) If it returns 1, then you must free all the packets being
247 	 *    shipped in, the tcb has been destroyed (or about to be destroyed).
248 	 */
249 	struct mbuf *m_save;
250 	struct ether_header *eh;
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 	NET_EPOCH_ASSERT();
266 
267 	if (m && m->m_pkthdr.rcvif)
268 		ifp = m->m_pkthdr.rcvif;
269 	else
270 		ifp = NULL;
271 	if (ifp) {
272 		bpf_req = bpf_peers_present(ifp->if_bpf);
273 	} else  {
274 		/*
275 		 * We probably should not work around
276 		 * but kassert, since lro alwasy sets rcvif.
277 		 */
278 		no_vn = 1;
279 		goto skip_vnet;
280 	}
281 	CURVNET_SET(ifp->if_vnet);
282 skip_vnet:
283 	while (m) {
284 		m_save = m->m_nextpkt;
285 		m->m_nextpkt = NULL;
286 		/* Now lets get the ether header */
287 		eh = mtod(m, struct ether_header *);
288 		etype = ntohs(eh->ether_type);
289 		/* Let the BPF see the packet */
290 		if (bpf_req && ifp)
291 			ETHER_BPF_MTAP(ifp, m);
292 		m_adj(m,  sizeof(*eh));
293 		/* Trim off the ethernet header */
294 		switch (etype) {
295 #ifdef INET6
296 		case ETHERTYPE_IPV6:
297 		{
298 			if (m->m_len < (sizeof(*ip6) + sizeof(*th))) {
299 				m = m_pullup(m, sizeof(*ip6) + sizeof(*th));
300 				if (m == NULL) {
301 					KMOD_TCPSTAT_INC(tcps_rcvshort);
302 					m_freem(m);
303 					goto skipped_pkt;
304 				}
305 			}
306 			ip6 = (struct ip6_hdr *)(eh + 1);
307 			th = (struct tcphdr *)(ip6 + 1);
308 			tlen = ntohs(ip6->ip6_plen);
309 			drop_hdrlen = sizeof(*ip6);
310 			if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
311 				if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
312 					th->th_sum = m->m_pkthdr.csum_data;
313 				else
314 					th->th_sum = in6_cksum_pseudo(ip6, tlen,
315 								      IPPROTO_TCP, m->m_pkthdr.csum_data);
316 				th->th_sum ^= 0xffff;
317 			} else
318 				th->th_sum = in6_cksum(m, IPPROTO_TCP, drop_hdrlen, tlen);
319 			if (th->th_sum) {
320 				KMOD_TCPSTAT_INC(tcps_rcvbadsum);
321 				m_freem(m);
322 				goto skipped_pkt;
323 			}
324 			/*
325 			 * Be proactive about unspecified IPv6 address in source.
326 			 * As we use all-zero to indicate unbounded/unconnected pcb,
327 			 * unspecified IPv6 address can be used to confuse us.
328 			 *
329 			 * Note that packets with unspecified IPv6 destination is
330 			 * already dropped in ip6_input.
331 			 */
332 			if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
333 				/* XXX stat */
334 				m_freem(m);
335 				goto skipped_pkt;
336 			}
337 			iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
338 			break;
339 		}
340 #endif
341 #ifdef INET
342 		case ETHERTYPE_IP:
343 		{
344 			if (m->m_len < sizeof (struct tcpiphdr)) {
345 				if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
346 				    == NULL) {
347 					KMOD_TCPSTAT_INC(tcps_rcvshort);
348 					m_freem(m);
349 					goto skipped_pkt;
350 				}
351 			}
352 			ip = (struct ip *)(eh + 1);
353 			th = (struct tcphdr *)(ip + 1);
354 			drop_hdrlen = sizeof(*ip);
355 			iptos = ip->ip_tos;
356 			tlen = ntohs(ip->ip_len) - sizeof(struct ip);
357 			if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
358 				if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
359 					th->th_sum = m->m_pkthdr.csum_data;
360 				else
361 					th->th_sum = in_pseudo(ip->ip_src.s_addr,
362 							       ip->ip_dst.s_addr,
363 							       htonl(m->m_pkthdr.csum_data + tlen +
364 								     IPPROTO_TCP));
365 				th->th_sum ^= 0xffff;
366 			} else {
367 				int len;
368 				struct ipovly *ipov = (struct ipovly *)ip;
369 				/*
370 				 * Checksum extended TCP header and data.
371 				 */
372 				len = drop_hdrlen + tlen;
373 				bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
374 				ipov->ih_len = htons(tlen);
375 				th->th_sum = in_cksum(m, len);
376 				/* Reset length for SDT probes. */
377 				ip->ip_len = htons(len);
378 				/* Reset TOS bits */
379 				ip->ip_tos = iptos;
380 				/* Re-initialization for later version check */
381 				ip->ip_v = IPVERSION;
382 				ip->ip_hl = sizeof(*ip) >> 2;
383 			}
384 			if (th->th_sum) {
385 				KMOD_TCPSTAT_INC(tcps_rcvbadsum);
386 				m_freem(m);
387 				goto skipped_pkt;
388 			}
389 			break;
390 		}
391 #endif
392 		}
393 		/*
394 		 * Convert TCP protocol specific fields to host format.
395 		 */
396 		tcp_fields_to_host(th);
397 
398 		off = th->th_off << 2;
399 		if (off < sizeof (struct tcphdr) || off > tlen) {
400 			KMOD_TCPSTAT_INC(tcps_rcvbadoff);
401 				m_freem(m);
402 				goto skipped_pkt;
403 		}
404 		tlen -= off;
405 		drop_hdrlen += off;
406 		/*
407 		 * Now lets setup the timeval to be when we should
408 		 * have been called (if we can).
409 		 */
410 		m->m_pkthdr.lro_nsegs = 1;
411 		if (m->m_flags & M_TSTMP_LRO) {
412 			tv.tv_sec = m->m_pkthdr.rcv_tstmp /1000000000;
413 			tv.tv_usec = (m->m_pkthdr.rcv_tstmp % 1000000000)/1000;
414 		} else {
415 			/* Should not be should we kassert instead? */
416 			tcp_get_usecs(&tv);
417 		}
418 		/* Now what about next packet? */
419 		if (m_save || has_pkt)
420 			nxt_pkt = 1;
421 		else
422 			nxt_pkt = 0;
423 		KMOD_TCPSTAT_INC(tcps_rcvtotal);
424 		retval = (*tp->t_fb->tfb_do_segment_nounlock)(m, th, so, tp, drop_hdrlen, tlen,
425 							      iptos, nxt_pkt, &tv);
426 		if (retval) {
427 			/* We lost the lock and tcb probably */
428 			m = m_save;
429 			while(m) {
430 				m_save = m->m_nextpkt;
431 				m->m_nextpkt = NULL;
432 				m_freem(m);
433 				m = m_save;
434 			}
435 			if (no_vn == 0)
436 				CURVNET_RESTORE();
437 			return(retval);
438 		}
439 skipped_pkt:
440 		m = m_save;
441 	}
442 	if (no_vn == 0)
443 		CURVNET_RESTORE();
444 	return(retval);
445 }
446 
447 int
448 ctf_do_queued_segments(struct socket *so, struct tcpcb *tp, int have_pkt)
449 {
450 	struct mbuf *m;
451 
452 	/* First lets see if we have old packets */
453 	if (tp->t_in_pkt) {
454 		m = tp->t_in_pkt;
455 		tp->t_in_pkt = NULL;
456 		tp->t_tail_pkt = NULL;
457 		if (ctf_process_inbound_raw(tp, so, m, have_pkt)) {
458 			/* We lost the tcpcb (maybe a RST came in)? */
459 			return(1);
460 		}
461 		tcp_handle_wakeup(tp, so);
462 	}
463 	return (0);
464 }
465 
466 uint32_t
467 ctf_outstanding(struct tcpcb *tp)
468 {
469 	uint32_t bytes_out;
470 
471 	bytes_out = tp->snd_max - tp->snd_una;
472 	if (tp->t_state < TCPS_ESTABLISHED)
473 		bytes_out++;
474 	if (tp->t_flags & TF_SENTFIN)
475 		bytes_out++;
476 	return (bytes_out);
477 }
478 
479 uint32_t
480 ctf_flight_size(struct tcpcb *tp, uint32_t rc_sacked)
481 {
482 	if (rc_sacked <= ctf_outstanding(tp))
483 		return(ctf_outstanding(tp) - rc_sacked);
484 	else {
485 		/* TSNH */
486 #ifdef INVARIANTS
487 		panic("tp:%p rc_sacked:%d > out:%d",
488 		      tp, rc_sacked, ctf_outstanding(tp));
489 #endif
490 		return (0);
491 	}
492 }
493 
494 void
495 ctf_do_dropwithreset(struct mbuf *m, struct tcpcb *tp, struct tcphdr *th,
496     int32_t rstreason, int32_t tlen)
497 {
498 	if (tp != NULL) {
499 		tcp_dropwithreset(m, th, tp, tlen, rstreason);
500 		INP_WUNLOCK(tp->t_inpcb);
501 	} else
502 		tcp_dropwithreset(m, th, NULL, tlen, rstreason);
503 }
504 
505 /*
506  * ctf_drop_checks returns 1 for you should not proceed. It places
507  * in ret_val what should be returned 1/0 by the caller. The 1 indicates
508  * that the TCB is unlocked and probably dropped. The 0 indicates the
509  * TCB is still valid and locked.
510  */
511 int
512 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)
513 {
514 	int32_t todrop;
515 	int32_t thflags;
516 	int32_t tlen;
517 
518 	thflags = *thf;
519 	tlen = *tlenp;
520 	todrop = tp->rcv_nxt - th->th_seq;
521 	if (todrop > 0) {
522 		if (thflags & TH_SYN) {
523 			thflags &= ~TH_SYN;
524 			th->th_seq++;
525 			if (th->th_urp > 1)
526 				th->th_urp--;
527 			else
528 				thflags &= ~TH_URG;
529 			todrop--;
530 		}
531 		/*
532 		 * Following if statement from Stevens, vol. 2, p. 960.
533 		 */
534 		if (todrop > tlen
535 		    || (todrop == tlen && (thflags & TH_FIN) == 0)) {
536 			/*
537 			 * Any valid FIN must be to the left of the window.
538 			 * At this point the FIN must be a duplicate or out
539 			 * of sequence; drop it.
540 			 */
541 			thflags &= ~TH_FIN;
542 			/*
543 			 * Send an ACK to resynchronize and drop any data.
544 			 * But keep on processing for RST or ACK.
545 			 */
546 			tp->t_flags |= TF_ACKNOW;
547 			todrop = tlen;
548 			KMOD_TCPSTAT_INC(tcps_rcvduppack);
549 			KMOD_TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
550 		} else {
551 			KMOD_TCPSTAT_INC(tcps_rcvpartduppack);
552 			KMOD_TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
553 		}
554 		/*
555 		 * DSACK - add SACK block for dropped range
556 		 */
557 		if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
558 			tcp_update_sack_list(tp, th->th_seq,
559 			    th->th_seq + todrop);
560 			/*
561 			 * ACK now, as the next in-sequence segment
562 			 * will clear the DSACK block again
563 			 */
564 			tp->t_flags |= TF_ACKNOW;
565 		}
566 		*drop_hdrlen += todrop;	/* drop from the top afterwards */
567 		th->th_seq += todrop;
568 		tlen -= todrop;
569 		if (th->th_urp > todrop)
570 			th->th_urp -= todrop;
571 		else {
572 			thflags &= ~TH_URG;
573 			th->th_urp = 0;
574 		}
575 	}
576 	/*
577 	 * If segment ends after window, drop trailing data (and PUSH and
578 	 * FIN); if nothing left, just ACK.
579 	 */
580 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
581 	if (todrop > 0) {
582 		KMOD_TCPSTAT_INC(tcps_rcvpackafterwin);
583 		if (todrop >= tlen) {
584 			KMOD_TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
585 			/*
586 			 * If window is closed can only take segments at
587 			 * window edge, and have to drop data and PUSH from
588 			 * incoming segments.  Continue processing, but
589 			 * remember to ack.  Otherwise, drop segment and
590 			 * ack.
591 			 */
592 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
593 				tp->t_flags |= TF_ACKNOW;
594 				KMOD_TCPSTAT_INC(tcps_rcvwinprobe);
595 			} else {
596 				ctf_do_dropafterack(m, tp, th, thflags, tlen, ret_val);
597 				return (1);
598 			}
599 		} else
600 			KMOD_TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
601 		m_adj(m, -todrop);
602 		tlen -= todrop;
603 		thflags &= ~(TH_PUSH | TH_FIN);
604 	}
605 	*thf = thflags;
606 	*tlenp = tlen;
607 	return (0);
608 }
609 
610 /*
611  * The value in ret_val informs the caller
612  * if we dropped the tcb (and lock) or not.
613  * 1 = we dropped it, 0 = the TCB is still locked
614  * and valid.
615  */
616 void
617 ctf_do_dropafterack(struct mbuf *m, struct tcpcb *tp, struct tcphdr *th, int32_t thflags, int32_t tlen, int32_t * ret_val)
618 {
619 	/*
620 	 * Generate an ACK dropping incoming segment if it occupies sequence
621 	 * space, where the ACK reflects our state.
622 	 *
623 	 * We can now skip the test for the RST flag since all paths to this
624 	 * code happen after packets containing RST have been dropped.
625 	 *
626 	 * In the SYN-RECEIVED state, don't send an ACK unless the segment
627 	 * we received passes the SYN-RECEIVED ACK test. If it fails send a
628 	 * RST.  This breaks the loop in the "LAND" DoS attack, and also
629 	 * prevents an ACK storm between two listening ports that have been
630 	 * sent forged SYN segments, each with the source address of the
631 	 * other.
632 	 */
633 	if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
634 	    (SEQ_GT(tp->snd_una, th->th_ack) ||
635 	    SEQ_GT(th->th_ack, tp->snd_max))) {
636 		*ret_val = 1;
637 		ctf_do_dropwithreset(m, tp, th, BANDLIM_RST_OPENPORT, tlen);
638 		return;
639 	} else
640 		*ret_val = 0;
641 	tp->t_flags |= TF_ACKNOW;
642 	if (m)
643 		m_freem(m);
644 }
645 
646 void
647 ctf_do_drop(struct mbuf *m, struct tcpcb *tp)
648 {
649 
650 	/*
651 	 * Drop space held by incoming segment and return.
652 	 */
653 	if (tp != NULL)
654 		INP_WUNLOCK(tp->t_inpcb);
655 	if (m)
656 		m_freem(m);
657 }
658 
659 int
660 ctf_process_rst(struct mbuf *m, struct tcphdr *th, struct socket *so, struct tcpcb *tp)
661 {
662 	/*
663 	 * RFC5961 Section 3.2
664 	 *
665 	 * - RST drops connection only if SEG.SEQ == RCV.NXT. - If RST is in
666 	 * window, we send challenge ACK.
667 	 *
668 	 * Note: to take into account delayed ACKs, we should test against
669 	 * last_ack_sent instead of rcv_nxt. Note 2: we handle special case
670 	 * of closed window, not covered by the RFC.
671 	 */
672 	int dropped = 0;
673 
674 	if ((SEQ_GEQ(th->th_seq, (tp->last_ack_sent - 1)) &&
675 	    SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
676 	    (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
677 		KASSERT(tp->t_state != TCPS_SYN_SENT,
678 		    ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
679 		    __func__, th, tp));
680 
681 		if (V_tcp_insecure_rst ||
682 		    (tp->last_ack_sent == th->th_seq) ||
683 		    (tp->rcv_nxt == th->th_seq) ||
684 		    ((tp->last_ack_sent - 1) == th->th_seq)) {
685 			KMOD_TCPSTAT_INC(tcps_drops);
686 			/* Drop the connection. */
687 			switch (tp->t_state) {
688 			case TCPS_SYN_RECEIVED:
689 				so->so_error = ECONNREFUSED;
690 				goto close;
691 			case TCPS_ESTABLISHED:
692 			case TCPS_FIN_WAIT_1:
693 			case TCPS_FIN_WAIT_2:
694 			case TCPS_CLOSE_WAIT:
695 			case TCPS_CLOSING:
696 			case TCPS_LAST_ACK:
697 				so->so_error = ECONNRESET;
698 		close:
699 				tcp_state_change(tp, TCPS_CLOSED);
700 				/* FALLTHROUGH */
701 			default:
702 				tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_RST);
703 				tp = tcp_close(tp);
704 			}
705 			dropped = 1;
706 			ctf_do_drop(m, tp);
707 		} else {
708 			KMOD_TCPSTAT_INC(tcps_badrst);
709 			/* Send challenge ACK. */
710 			tcp_respond(tp, mtod(m, void *), th, m,
711 			    tp->rcv_nxt, tp->snd_nxt, TH_ACK);
712 			tp->last_ack_sent = tp->rcv_nxt;
713 		}
714 	} else {
715 		m_freem(m);
716 	}
717 	return (dropped);
718 }
719 
720 /*
721  * The value in ret_val informs the caller
722  * if we dropped the tcb (and lock) or not.
723  * 1 = we dropped it, 0 = the TCB is still locked
724  * and valid.
725  */
726 void
727 ctf_challenge_ack(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, int32_t * ret_val)
728 {
729 
730 	NET_EPOCH_ASSERT();
731 
732 	KMOD_TCPSTAT_INC(tcps_badsyn);
733 	if (V_tcp_insecure_syn &&
734 	    SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
735 	    SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
736 		tp = tcp_drop(tp, ECONNRESET);
737 		*ret_val = 1;
738 		ctf_do_drop(m, tp);
739 	} else {
740 		/* Send challenge ACK. */
741 		tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
742 		    tp->snd_nxt, TH_ACK);
743 		tp->last_ack_sent = tp->rcv_nxt;
744 		m = NULL;
745 		*ret_val = 0;
746 		ctf_do_drop(m, NULL);
747 	}
748 }
749 
750 /*
751  * bbr_ts_check returns 1 for you should not proceed, the state
752  * machine should return. It places in ret_val what should
753  * be returned 1/0 by the caller (hpts_do_segment). The 1 indicates
754  * that the TCB is unlocked and probably dropped. The 0 indicates the
755  * TCB is still valid and locked.
756  */
757 int
758 ctf_ts_check(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
759     int32_t tlen, int32_t thflags, int32_t * ret_val)
760 {
761 
762 	if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
763 		/*
764 		 * Invalidate ts_recent.  If this segment updates ts_recent,
765 		 * the age will be reset later and ts_recent will get a
766 		 * valid value.  If it does not, setting ts_recent to zero
767 		 * will at least satisfy the requirement that zero be placed
768 		 * in the timestamp echo reply when ts_recent isn't valid.
769 		 * The age isn't reset until we get a valid ts_recent
770 		 * because we don't want out-of-order segments to be dropped
771 		 * when ts_recent is old.
772 		 */
773 		tp->ts_recent = 0;
774 	} else {
775 		KMOD_TCPSTAT_INC(tcps_rcvduppack);
776 		KMOD_TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
777 		KMOD_TCPSTAT_INC(tcps_pawsdrop);
778 		*ret_val = 0;
779 		if (tlen) {
780 			ctf_do_dropafterack(m, tp, th, thflags, tlen, ret_val);
781 		} else {
782 			ctf_do_drop(m, NULL);
783 		}
784 		return (1);
785 	}
786 	return (0);
787 }
788 
789 void
790 ctf_calc_rwin(struct socket *so, struct tcpcb *tp)
791 {
792 	int32_t win;
793 
794 	/*
795 	 * Calculate amount of space in receive window, and then do TCP
796 	 * input processing. Receive window is amount of space in rcv queue,
797 	 * but not less than advertised window.
798 	 */
799 	win = sbspace(&so->so_rcv);
800 	if (win < 0)
801 		win = 0;
802 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
803 }
804 
805 void
806 ctf_do_dropwithreset_conn(struct mbuf *m, struct tcpcb *tp, struct tcphdr *th,
807     int32_t rstreason, int32_t tlen)
808 {
809 
810 	if (tp->t_inpcb) {
811 		tcp_set_inp_to_drop(tp->t_inpcb, ETIMEDOUT);
812 	}
813 	tcp_dropwithreset(m, th, tp, tlen, rstreason);
814 	INP_WUNLOCK(tp->t_inpcb);
815 }
816 
817 uint32_t
818 ctf_fixed_maxseg(struct tcpcb *tp)
819 {
820 	int optlen;
821 
822 	if (tp->t_flags & TF_NOOPT)
823 		return (tp->t_maxseg);
824 
825 	/*
826 	 * Here we have a simplified code from tcp_addoptions(),
827 	 * without a proper loop, and having most of paddings hardcoded.
828 	 * We only consider fixed options that we would send every
829 	 * time I.e. SACK is not considered.
830 	 *
831 	 */
832 #define	PAD(len)	((((len) / 4) + !!((len) % 4)) * 4)
833 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
834 		if (tp->t_flags & TF_RCVD_TSTMP)
835 			optlen = TCPOLEN_TSTAMP_APPA;
836 		else
837 			optlen = 0;
838 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
839 		if (tp->t_flags & TF_SIGNATURE)
840 			optlen += PAD(TCPOLEN_SIGNATURE);
841 #endif
842 	} else {
843 		if (tp->t_flags & TF_REQ_TSTMP)
844 			optlen = TCPOLEN_TSTAMP_APPA;
845 		else
846 			optlen = PAD(TCPOLEN_MAXSEG);
847 		if (tp->t_flags & TF_REQ_SCALE)
848 			optlen += PAD(TCPOLEN_WINDOW);
849 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
850 		if (tp->t_flags & TF_SIGNATURE)
851 			optlen += PAD(TCPOLEN_SIGNATURE);
852 #endif
853 		if (tp->t_flags & TF_SACK_PERMIT)
854 			optlen += PAD(TCPOLEN_SACK_PERMITTED);
855 	}
856 #undef PAD
857 	optlen = min(optlen, TCP_MAXOLEN);
858 	return (tp->t_maxseg - optlen);
859 }
860 
861 void
862 ctf_log_sack_filter(struct tcpcb *tp, int num_sack_blks, struct sackblk *sack_blocks)
863 {
864 	if (tp->t_logstate != TCP_LOG_STATE_OFF) {
865 		union tcp_log_stackspecific log;
866 		struct timeval tv;
867 
868 		memset(&log, 0, sizeof(log));
869 		log.u_bbr.timeStamp = tcp_get_usecs(&tv);
870 		log.u_bbr.flex8 = num_sack_blks;
871 		if (num_sack_blks > 0) {
872 			log.u_bbr.flex1 = sack_blocks[0].start;
873 			log.u_bbr.flex2 = sack_blocks[0].end;
874 		}
875 		if (num_sack_blks > 1) {
876 			log.u_bbr.flex3 = sack_blocks[1].start;
877 			log.u_bbr.flex4 = sack_blocks[1].end;
878 		}
879 		if (num_sack_blks > 2) {
880 			log.u_bbr.flex5 = sack_blocks[2].start;
881 			log.u_bbr.flex6 = sack_blocks[2].end;
882 		}
883 		if (num_sack_blks > 3) {
884 			log.u_bbr.applimited = sack_blocks[3].start;
885 			log.u_bbr.pkts_out = sack_blocks[3].end;
886 		}
887 		TCP_LOG_EVENTP(tp, NULL,
888 		    &tp->t_inpcb->inp_socket->so_rcv,
889 		    &tp->t_inpcb->inp_socket->so_snd,
890 		    TCP_SACK_FILTER_RES, 0,
891 		    0, &log, false, &tv);
892 	}
893 }
894 
895 uint32_t
896 ctf_decay_count(uint32_t count, uint32_t decay)
897 {
898 	/*
899 	 * Given a count, decay it by a set percentage. The
900 	 * percentage is in thousands i.e. 100% = 1000,
901 	 * 19.3% = 193.
902 	 */
903 	uint64_t perc_count, decay_per;
904 	uint32_t decayed_count;
905 	if (decay > 1000) {
906 		/* We don't raise it */
907 		return (count);
908 	}
909 	perc_count = count;
910 	decay_per = decay;
911 	perc_count *= decay_per;
912 	perc_count /= 1000;
913 	/*
914 	 * So now perc_count holds the
915 	 * count decay value.
916 	 */
917 	decayed_count = count - (uint32_t)perc_count;
918 	return(decayed_count);
919 }
920 
921 int32_t
922 ctf_progress_timeout_check(struct tcpcb *tp, bool log)
923 {
924 	if (tp->t_maxunacktime && tp->t_acktime && TSTMP_GT(ticks, tp->t_acktime)) {
925 		if ((ticks - tp->t_acktime) >= tp->t_maxunacktime) {
926 			/*
927 			 * There is an assumption that the caller
928 			 * will drop the connection so we will
929 			 * increment the counters here.
930 			 */
931 			if (log)
932 				tcp_log_end_status(tp, TCP_EI_STATUS_PROGRESS);
933 #ifdef NETFLIX_STATS
934 			KMOD_TCPSTAT_INC(tcps_progdrops);
935 #endif
936 			return (1);
937 		}
938 	}
939 	return (0);
940 }
941