2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
35 #include "opt_compat.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_tcpdebug.h"
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/callout.h>
44 #include <sys/eventhandler.h>
45 #include <sys/hhook.h>
46 #include <sys/kernel.h>
47 #include <sys/khelp.h>
48 #include <sys/sysctl.h>
50 #include <sys/malloc.h>
51 #include <sys/refcount.h>
54 #include <sys/domain.h>
59 #include <sys/socket.h>
60 #include <sys/socketvar.h>
61 #include <sys/protosw.h>
62 #include <sys/random.h>
66 #include <net/route.h>
68 #include <net/if_var.h>
71 #include <netinet/in.h>
72 #include <netinet/in_fib.h>
73 #include <netinet/in_kdtrace.h>
74 #include <netinet/in_pcb.h>
75 #include <netinet/in_systm.h>
76 #include <netinet/in_var.h>
77 #include <netinet/ip.h>
78 #include <netinet/ip_icmp.h>
79 #include <netinet/ip_var.h>
81 #include <netinet/ip6.h>
82 #include <netinet6/in6_fib.h>
83 #include <netinet6/in6_pcb.h>
84 #include <netinet6/ip6_var.h>
85 #include <netinet6/scope6_var.h>
86 #include <netinet6/nd6.h>
90 #include <netinet/tcp_fastopen.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/tcp_syncache.h>
98 #include <netinet/cc/cc.h>
100 #include <netinet6/tcp6_var.h>
102 #include <netinet/tcpip.h>
104 #include <netinet/tcp_pcap.h>
107 #include <netinet/tcp_debug.h>
110 #include <netinet6/ip6protosw.h>
113 #include <netinet/tcp_offload.h>
117 #include <netipsec/ipsec.h>
118 #include <netipsec/xform.h>
120 #include <netipsec/ipsec6.h>
122 #include <netipsec/key.h>
123 #include <sys/syslog.h>
126 #include <machine/in_cksum.h>
129 #include <security/mac/mac_framework.h>
131 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
133 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
136 struct rwlock tcp_function_lock;
139 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
144 error = sysctl_handle_int(oidp, &new, 0, req);
145 if (error == 0 && req->newptr) {
146 if (new < TCP_MINMSS)
154 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
155 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
156 &sysctl_net_inet_tcp_mss_check, "I",
157 "Default TCP Maximum Segment Size");
161 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
165 new = V_tcp_v6mssdflt;
166 error = sysctl_handle_int(oidp, &new, 0, req);
167 if (error == 0 && req->newptr) {
168 if (new < TCP_MINMSS)
171 V_tcp_v6mssdflt = new;
176 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
177 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
178 &sysctl_net_inet_tcp_mss_v6_check, "I",
179 "Default TCP Maximum Segment Size for IPv6");
183 * Minimum MSS we accept and use. This prevents DoS attacks where
184 * we are forced to a ridiculous low MSS like 20 and send hundreds
185 * of packets instead of one. The effect scales with the available
186 * bandwidth and quickly saturates the CPU and network interface
187 * with packet generation and sending. Set to zero to disable MINMSS
188 * checking. This setting prevents us from sending too small packets.
190 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
192 &VNET_NAME(tcp_minmss), 0,
193 "Minimum TCP Maximum Segment Size");
195 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
196 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
197 &VNET_NAME(tcp_do_rfc1323), 0,
198 "Enable rfc1323 (high performance TCP) extensions");
200 static int tcp_log_debug = 0;
201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
202 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
204 static int tcp_tcbhashsize;
205 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
206 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
208 static int do_tcpdrain = 1;
209 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
210 "Enable tcp_drain routine for extra help when low on mbufs");
212 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
213 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
215 static VNET_DEFINE(int, icmp_may_rst) = 1;
216 #define V_icmp_may_rst VNET(icmp_may_rst)
217 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
218 &VNET_NAME(icmp_may_rst), 0,
219 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
221 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
222 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
223 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
224 &VNET_NAME(tcp_isn_reseed_interval), 0,
225 "Seconds between reseeding of ISN secret");
227 static int tcp_soreceive_stream;
228 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
229 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
232 static int tcp_sig_checksigs = 1;
233 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
234 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
237 VNET_DEFINE(uma_zone_t, sack_hole_zone);
238 #define V_sack_hole_zone VNET(sack_hole_zone)
240 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
242 static struct inpcb *tcp_notify(struct inpcb *, int);
243 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
244 static void tcp_mtudisc(struct inpcb *, int);
245 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
246 void *ip4hdr, const void *ip6hdr);
247 static void tcp_timer_discard(struct tcpcb *, uint32_t);
250 static struct tcp_function_block tcp_def_funcblk = {
254 tcp_default_ctloutput,
266 struct tcp_funchead t_functions;
267 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
269 static struct tcp_function_block *
270 find_tcp_functions_locked(struct tcp_function_set *fs)
272 struct tcp_function *f;
273 struct tcp_function_block *blk=NULL;
275 TAILQ_FOREACH(f, &t_functions, tf_next) {
276 if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) {
284 static struct tcp_function_block *
285 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
287 struct tcp_function_block *rblk=NULL;
288 struct tcp_function *f;
290 TAILQ_FOREACH(f, &t_functions, tf_next) {
291 if (f->tf_fb == blk) {
302 struct tcp_function_block *
303 find_and_ref_tcp_functions(struct tcp_function_set *fs)
305 struct tcp_function_block *blk;
307 rw_rlock(&tcp_function_lock);
308 blk = find_tcp_functions_locked(fs);
310 refcount_acquire(&blk->tfb_refcnt);
311 rw_runlock(&tcp_function_lock);
315 struct tcp_function_block *
316 find_and_ref_tcp_fb(struct tcp_function_block *blk)
318 struct tcp_function_block *rblk;
320 rw_rlock(&tcp_function_lock);
321 rblk = find_tcp_fb_locked(blk, NULL);
323 refcount_acquire(&rblk->tfb_refcnt);
324 rw_runlock(&tcp_function_lock);
330 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
333 struct tcp_function_set fs;
334 struct tcp_function_block *blk;
336 memset(&fs, 0, sizeof(fs));
337 rw_rlock(&tcp_function_lock);
338 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
341 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
342 fs.pcbcnt = blk->tfb_refcnt;
344 rw_runlock(&tcp_function_lock);
345 error = sysctl_handle_string(oidp, fs.function_set_name,
346 sizeof(fs.function_set_name), req);
348 /* Check for error or no change */
349 if (error != 0 || req->newptr == NULL)
352 rw_wlock(&tcp_function_lock);
353 blk = find_tcp_functions_locked(&fs);
355 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
359 tcp_func_set_ptr = blk;
361 rw_wunlock(&tcp_function_lock);
365 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
366 CTLTYPE_STRING | CTLFLAG_RW,
367 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
368 "Set/get the default TCP functions");
371 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
373 int error, cnt, linesz;
374 struct tcp_function *f;
379 rw_rlock(&tcp_function_lock);
380 TAILQ_FOREACH(f, &t_functions, tf_next) {
383 rw_runlock(&tcp_function_lock);
385 bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1;
386 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
391 linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count");
396 rw_rlock(&tcp_function_lock);
397 TAILQ_FOREACH(f, &t_functions, tf_next) {
398 linesz = snprintf(cp, bufsz, "%-32s%c %u\n",
399 f->tf_fb->tfb_tcp_block_name,
400 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
401 f->tf_fb->tfb_refcnt);
402 if (linesz >= bufsz) {
410 rw_runlock(&tcp_function_lock);
412 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
413 free(buffer, M_TEMP);
417 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
418 CTLTYPE_STRING|CTLFLAG_RD,
419 NULL, 0, sysctl_net_inet_list_available, "A",
420 "list available TCP Function sets");
423 * Target size of TCP PCB hash tables. Must be a power of two.
425 * Note that this can be overridden by the kernel environment
426 * variable net.inet.tcp.tcbhashsize
429 #define TCBHASHSIZE 0
434 * Callouts should be moved into struct tcp directly. They are currently
435 * separate because the tcpcb structure is exported to userland for sysctl
436 * parsing purposes, which do not know about callouts.
445 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
446 #define V_tcpcb_zone VNET(tcpcb_zone)
448 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
449 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
451 static struct mtx isn_mtx;
453 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
454 #define ISN_LOCK() mtx_lock(&isn_mtx)
455 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
458 * TCP initialization.
461 tcp_zone_change(void *tag)
464 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
465 uma_zone_set_max(V_tcpcb_zone, maxsockets);
466 tcp_tw_zone_change();
470 tcp_inpcb_init(void *mem, int size, int flags)
472 struct inpcb *inp = mem;
474 INP_LOCK_INIT(inp, "inp", "tcpinp");
479 * Take a value and get the next power of 2 that doesn't overflow.
480 * Used to size the tcp_inpcb hash buckets.
483 maketcp_hashsize(int size)
489 * get the next power of 2 higher than maxsockets.
491 hashsize = 1 << fls(size);
492 /* catch overflow, and just go one power of 2 smaller */
493 if (hashsize < size) {
494 hashsize = 1 << (fls(size) - 1);
500 register_tcp_functions(struct tcp_function_block *blk, int wait)
502 struct tcp_function_block *lblk;
503 struct tcp_function *n;
504 struct tcp_function_set fs;
506 if ((blk->tfb_tcp_output == NULL) ||
507 (blk->tfb_tcp_do_segment == NULL) ||
508 (blk->tfb_tcp_ctloutput == NULL) ||
509 (strlen(blk->tfb_tcp_block_name) == 0)) {
511 * These functions are required and you
516 if (blk->tfb_tcp_timer_stop_all ||
517 blk->tfb_tcp_timers_left ||
518 blk->tfb_tcp_timer_activate ||
519 blk->tfb_tcp_timer_active ||
520 blk->tfb_tcp_timer_stop) {
522 * If you define one timer function you
523 * must have them all.
525 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
526 (blk->tfb_tcp_timers_left == NULL) ||
527 (blk->tfb_tcp_timer_activate == NULL) ||
528 (blk->tfb_tcp_timer_active == NULL) ||
529 (blk->tfb_tcp_timer_stop == NULL)) {
533 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
538 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
539 rw_wlock(&tcp_function_lock);
540 lblk = find_tcp_functions_locked(&fs);
542 /* Duplicate name space not allowed */
543 rw_wunlock(&tcp_function_lock);
544 free(n, M_TCPFUNCTIONS);
547 refcount_init(&blk->tfb_refcnt, 0);
549 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
550 rw_wunlock(&tcp_function_lock);
555 deregister_tcp_functions(struct tcp_function_block *blk)
557 struct tcp_function_block *lblk;
558 struct tcp_function *f;
561 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
562 /* You can't un-register the default */
565 rw_wlock(&tcp_function_lock);
566 if (blk == tcp_func_set_ptr) {
567 /* You can't free the current default */
568 rw_wunlock(&tcp_function_lock);
571 if (blk->tfb_refcnt) {
572 /* Still tcb attached, mark it. */
573 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
574 rw_wunlock(&tcp_function_lock);
577 lblk = find_tcp_fb_locked(blk, &f);
580 TAILQ_REMOVE(&t_functions, f, tf_next);
582 free(f, M_TCPFUNCTIONS);
585 rw_wunlock(&tcp_function_lock);
592 const char *tcbhash_tuneable;
595 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
597 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
598 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
599 printf("%s: WARNING: unable to register helper hook\n", __func__);
600 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
601 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
602 printf("%s: WARNING: unable to register helper hook\n", __func__);
603 hashsize = TCBHASHSIZE;
604 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
607 * Auto tune the hash size based on maxsockets.
608 * A perfect hash would have a 1:1 mapping
609 * (hashsize = maxsockets) however it's been
610 * suggested that O(2) average is better.
612 hashsize = maketcp_hashsize(maxsockets / 4);
614 * Our historical default is 512,
615 * do not autotune lower than this.
619 if (bootverbose && IS_DEFAULT_VNET(curvnet))
620 printf("%s: %s auto tuned to %d\n", __func__,
621 tcbhash_tuneable, hashsize);
624 * We require a hashsize to be a power of two.
625 * Previously if it was not a power of two we would just reset it
626 * back to 512, which could be a nasty surprise if you did not notice
628 * Instead what we do is clip it to the closest power of two lower
629 * than the specified hash value.
631 if (!powerof2(hashsize)) {
632 int oldhashsize = hashsize;
634 hashsize = maketcp_hashsize(hashsize);
635 /* prevent absurdly low value */
638 printf("%s: WARNING: TCB hash size not a power of 2, "
639 "clipped from %d to %d.\n", __func__, oldhashsize,
642 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
643 "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
644 IPI_HASHFIELDS_4TUPLE);
647 * These have to be type stable for the benefit of the timers.
649 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
650 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
651 uma_zone_set_max(V_tcpcb_zone, maxsockets);
652 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
658 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
659 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
660 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
662 /* Skip initialization of globals for non-default instances. */
663 if (!IS_DEFAULT_VNET(curvnet))
666 tcp_reass_global_init();
668 /* XXX virtualize those bellow? */
669 tcp_delacktime = TCPTV_DELACK;
670 tcp_keepinit = TCPTV_KEEP_INIT;
671 tcp_keepidle = TCPTV_KEEP_IDLE;
672 tcp_keepintvl = TCPTV_KEEPINTVL;
673 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
675 tcp_rexmit_min = TCPTV_MIN;
676 if (tcp_rexmit_min < 1)
678 tcp_persmin = TCPTV_PERSMIN;
679 tcp_persmax = TCPTV_PERSMAX;
680 tcp_rexmit_slop = TCPTV_CPU_VAR;
681 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
682 tcp_tcbhashsize = hashsize;
683 /* Setup the tcp function block list */
684 TAILQ_INIT(&t_functions);
685 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
686 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
688 if (tcp_soreceive_stream) {
690 tcp_usrreqs.pru_soreceive = soreceive_stream;
693 tcp6_usrreqs.pru_soreceive = soreceive_stream;
698 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
700 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
702 if (max_protohdr < TCP_MINPROTOHDR)
703 max_protohdr = TCP_MINPROTOHDR;
704 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
706 #undef TCP_MINPROTOHDR
709 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
710 SHUTDOWN_PRI_DEFAULT);
711 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
712 EVENTHANDLER_PRI_ANY);
729 tcp_fastopen_destroy();
734 in_pcbinfo_destroy(&V_tcbinfo);
735 uma_zdestroy(V_sack_hole_zone);
736 uma_zdestroy(V_tcpcb_zone);
738 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
740 printf("%s: WARNING: unable to deregister helper hook "
741 "type=%d, id=%d: error %d returned\n", __func__,
742 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
744 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
746 printf("%s: WARNING: unable to deregister helper hook "
747 "type=%d, id=%d: error %d returned\n", __func__,
748 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
760 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
761 * tcp_template used to store this data in mbufs, but we now recopy it out
762 * of the tcpcb each time to conserve mbufs.
765 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
767 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
769 INP_WLOCK_ASSERT(inp);
772 if ((inp->inp_vflag & INP_IPV6) != 0) {
775 ip6 = (struct ip6_hdr *)ip_ptr;
776 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
777 (inp->inp_flow & IPV6_FLOWINFO_MASK);
778 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
779 (IPV6_VERSION & IPV6_VERSION_MASK);
780 ip6->ip6_nxt = IPPROTO_TCP;
781 ip6->ip6_plen = htons(sizeof(struct tcphdr));
782 ip6->ip6_src = inp->in6p_laddr;
783 ip6->ip6_dst = inp->in6p_faddr;
786 #if defined(INET6) && defined(INET)
793 ip = (struct ip *)ip_ptr;
794 ip->ip_v = IPVERSION;
796 ip->ip_tos = inp->inp_ip_tos;
800 ip->ip_ttl = inp->inp_ip_ttl;
802 ip->ip_p = IPPROTO_TCP;
803 ip->ip_src = inp->inp_laddr;
804 ip->ip_dst = inp->inp_faddr;
807 th->th_sport = inp->inp_lport;
808 th->th_dport = inp->inp_fport;
816 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
820 * Create template to be used to send tcp packets on a connection.
821 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
822 * use for this function is in keepalives, which use tcp_respond.
825 tcpip_maketemplate(struct inpcb *inp)
829 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
832 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
837 * Send a single message to the TCP at address specified by
838 * the given TCP/IP header. If m == NULL, then we make a copy
839 * of the tcpiphdr at th and send directly to the addressed host.
840 * This is used to force keep alive messages out using the TCP
841 * template for a connection. If flags are given then we send
842 * a message back to the TCP which originated the segment th,
843 * and discard the mbuf containing it and any other attached mbufs.
845 * In any case the ack and sequence number of the transmitted
846 * segment are as specified by the parameters.
848 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
851 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
852 tcp_seq ack, tcp_seq seq, int flags)
865 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
868 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
875 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
876 INP_WLOCK_ASSERT(inp);
881 if (!(flags & TH_RST)) {
882 win = sbspace(&inp->inp_socket->so_rcv);
883 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
884 win = (long)TCP_MAXWIN << tp->rcv_scale;
888 m = m_gethdr(M_NOWAIT, MT_DATA);
892 m->m_data += max_linkhdr;
895 bcopy((caddr_t)ip6, mtod(m, caddr_t),
896 sizeof(struct ip6_hdr));
897 ip6 = mtod(m, struct ip6_hdr *);
898 nth = (struct tcphdr *)(ip6 + 1);
902 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
903 ip = mtod(m, struct ip *);
904 nth = (struct tcphdr *)(ip + 1);
906 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
911 * XXX MRT We inherrit the FIB, which is lucky.
915 m->m_data = (caddr_t)ipgen;
916 /* m_len is set later */
918 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
921 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
922 nth = (struct tcphdr *)(ip6 + 1);
926 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
927 nth = (struct tcphdr *)(ip + 1);
931 * this is usually a case when an extension header
932 * exists between the IPv6 header and the
935 nth->th_sport = th->th_sport;
936 nth->th_dport = th->th_dport;
938 xchg(nth->th_dport, nth->th_sport, uint16_t);
944 ip6->ip6_vfc = IPV6_VERSION;
945 ip6->ip6_nxt = IPPROTO_TCP;
946 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
947 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
950 #if defined(INET) && defined(INET6)
955 tlen += sizeof (struct tcpiphdr);
956 ip->ip_len = htons(tlen);
957 ip->ip_ttl = V_ip_defttl;
958 if (V_path_mtu_discovery)
959 ip->ip_off |= htons(IP_DF);
963 m->m_pkthdr.len = tlen;
964 m->m_pkthdr.rcvif = NULL;
968 * Packet is associated with a socket, so allow the
969 * label of the response to reflect the socket label.
971 INP_WLOCK_ASSERT(inp);
972 mac_inpcb_create_mbuf(inp, m);
975 * Packet is not associated with a socket, so possibly
976 * update the label in place.
978 mac_netinet_tcp_reply(m);
981 nth->th_seq = htonl(seq);
982 nth->th_ack = htonl(ack);
984 nth->th_off = sizeof (struct tcphdr) >> 2;
985 nth->th_flags = flags;
987 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
989 nth->th_win = htons((u_short)win);
992 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
995 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
996 nth->th_sum = in6_cksum_pseudo(ip6,
997 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
998 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1002 #if defined(INET6) && defined(INET)
1007 m->m_pkthdr.csum_flags = CSUM_TCP;
1008 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1009 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1013 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1014 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1016 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
1018 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
1021 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
1024 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
1026 #if defined(INET) && defined(INET6)
1030 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
1035 * Create a new TCP control block, making an
1036 * empty reassembly queue and hooking it to the argument
1037 * protocol control block. The `inp' parameter must have
1038 * come from the zone allocator set up in tcp_init().
1041 tcp_newtcpcb(struct inpcb *inp)
1043 struct tcpcb_mem *tm;
1046 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1049 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1054 /* Initialise cc_var struct for this tcpcb. */
1056 tp->ccv->type = IPPROTO_TCP;
1057 tp->ccv->ccvc.tcp = tp;
1058 rw_rlock(&tcp_function_lock);
1059 tp->t_fb = tcp_func_set_ptr;
1060 refcount_acquire(&tp->t_fb->tfb_refcnt);
1061 rw_runlock(&tcp_function_lock);
1062 if (tp->t_fb->tfb_tcp_fb_init) {
1063 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1066 * Use the current system default CC algorithm.
1069 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1070 CC_ALGO(tp) = CC_DEFAULT();
1073 if (CC_ALGO(tp)->cb_init != NULL)
1074 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1075 if (tp->t_fb->tfb_tcp_fb_fini)
1076 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1077 refcount_release(&tp->t_fb->tfb_refcnt);
1078 uma_zfree(V_tcpcb_zone, tm);
1083 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1084 if (tp->t_fb->tfb_tcp_fb_fini)
1085 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1086 refcount_release(&tp->t_fb->tfb_refcnt);
1087 uma_zfree(V_tcpcb_zone, tm);
1092 tp->t_vnet = inp->inp_vnet;
1094 tp->t_timers = &tm->tt;
1095 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
1098 isipv6 ? V_tcp_v6mssdflt :
1102 /* Set up our timeouts. */
1103 callout_init(&tp->t_timers->tt_rexmt, 1);
1104 callout_init(&tp->t_timers->tt_persist, 1);
1105 callout_init(&tp->t_timers->tt_keep, 1);
1106 callout_init(&tp->t_timers->tt_2msl, 1);
1107 callout_init(&tp->t_timers->tt_delack, 1);
1109 if (V_tcp_do_rfc1323)
1110 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1112 tp->t_flags |= TF_SACK_PERMIT;
1113 TAILQ_INIT(&tp->snd_holes);
1115 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1118 in_pcbref(inp); /* Reference for tcpcb */
1122 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1123 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1124 * reasonable initial retransmit time.
1126 tp->t_srtt = TCPTV_SRTTBASE;
1127 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1128 tp->t_rttmin = tcp_rexmit_min;
1129 tp->t_rxtcur = TCPTV_RTOBASE;
1130 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1131 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1132 tp->t_rcvtime = ticks;
1134 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1135 * because the socket may be bound to an IPv6 wildcard address,
1136 * which may match an IPv4-mapped IPv6 address.
1138 inp->inp_ip_ttl = V_ip_defttl;
1142 * Init the TCP PCAP queues.
1144 tcp_pcap_tcpcb_init(tp);
1146 return (tp); /* XXX */
1150 * Switch the congestion control algorithm back to NewReno for any active
1151 * control blocks using an algorithm which is about to go away.
1152 * This ensures the CC framework can allow the unload to proceed without leaving
1153 * any dangling pointers which would trigger a panic.
1154 * Returning non-zero would inform the CC framework that something went wrong
1155 * and it would be unsafe to allow the unload to proceed. However, there is no
1156 * way for this to occur with this implementation so we always return zero.
1159 tcp_ccalgounload(struct cc_algo *unload_algo)
1161 struct cc_algo *tmpalgo;
1164 VNET_ITERATOR_DECL(vnet_iter);
1167 * Check all active control blocks across all network stacks and change
1168 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1169 * requires cleanup code to be run, call it.
1172 VNET_FOREACH(vnet_iter) {
1173 CURVNET_SET(vnet_iter);
1174 INP_INFO_WLOCK(&V_tcbinfo);
1176 * New connections already part way through being initialised
1177 * with the CC algo we're removing will not race with this code
1178 * because the INP_INFO_WLOCK is held during initialisation. We
1179 * therefore don't enter the loop below until the connection
1180 * list has stabilised.
1182 LIST_FOREACH(inp, &V_tcb, inp_list) {
1184 /* Important to skip tcptw structs. */
1185 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1186 (tp = intotcpcb(inp)) != NULL) {
1188 * By holding INP_WLOCK here, we are assured
1189 * that the connection is not currently
1190 * executing inside the CC module's functions
1191 * i.e. it is safe to make the switch back to
1194 if (CC_ALGO(tp) == unload_algo) {
1195 tmpalgo = CC_ALGO(tp);
1196 /* NewReno does not require any init. */
1197 CC_ALGO(tp) = &newreno_cc_algo;
1198 if (tmpalgo->cb_destroy != NULL)
1199 tmpalgo->cb_destroy(tp->ccv);
1204 INP_INFO_WUNLOCK(&V_tcbinfo);
1207 VNET_LIST_RUNLOCK();
1213 * Drop a TCP connection, reporting
1214 * the specified error. If connection is synchronized,
1215 * then send a RST to peer.
1218 tcp_drop(struct tcpcb *tp, int errno)
1220 struct socket *so = tp->t_inpcb->inp_socket;
1222 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1223 INP_WLOCK_ASSERT(tp->t_inpcb);
1225 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1226 tcp_state_change(tp, TCPS_CLOSED);
1227 (void) tp->t_fb->tfb_tcp_output(tp);
1228 TCPSTAT_INC(tcps_drops);
1230 TCPSTAT_INC(tcps_conndrops);
1231 if (errno == ETIMEDOUT && tp->t_softerror)
1232 errno = tp->t_softerror;
1233 so->so_error = errno;
1234 return (tcp_close(tp));
1238 tcp_discardcb(struct tcpcb *tp)
1240 struct inpcb *inp = tp->t_inpcb;
1241 struct socket *so = inp->inp_socket;
1243 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1247 INP_WLOCK_ASSERT(inp);
1250 * Make sure that all of our timers are stopped before we delete the
1253 * If stopping a timer fails, we schedule a discard function in same
1254 * callout, and the last discard function called will take care of
1255 * deleting the tcpcb.
1257 tcp_timer_stop(tp, TT_REXMT);
1258 tcp_timer_stop(tp, TT_PERSIST);
1259 tcp_timer_stop(tp, TT_KEEP);
1260 tcp_timer_stop(tp, TT_2MSL);
1261 tcp_timer_stop(tp, TT_DELACK);
1262 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1263 /* Call the stop-all function of the methods */
1264 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1268 * If we got enough samples through the srtt filter,
1269 * save the rtt and rttvar in the routing entry.
1270 * 'Enough' is arbitrarily defined as 4 rtt samples.
1271 * 4 samples is enough for the srtt filter to converge
1272 * to within enough % of the correct value; fewer samples
1273 * and we could save a bogus rtt. The danger is not high
1274 * as tcp quickly recovers from everything.
1275 * XXX: Works very well but needs some more statistics!
1277 if (tp->t_rttupdated >= 4) {
1278 struct hc_metrics_lite metrics;
1281 bzero(&metrics, sizeof(metrics));
1283 * Update the ssthresh always when the conditions below
1284 * are satisfied. This gives us better new start value
1285 * for the congestion avoidance for new connections.
1286 * ssthresh is only set if packet loss occured on a session.
1288 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1289 * being torn down. Ideally this code would not use 'so'.
1291 ssthresh = tp->snd_ssthresh;
1292 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1294 * convert the limit from user data bytes to
1295 * packets then to packet data bytes.
1297 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1300 ssthresh *= (u_long)(tp->t_maxseg +
1302 (isipv6 ? sizeof (struct ip6_hdr) +
1303 sizeof (struct tcphdr) :
1305 sizeof (struct tcpiphdr)
1312 metrics.rmx_ssthresh = ssthresh;
1314 metrics.rmx_rtt = tp->t_srtt;
1315 metrics.rmx_rttvar = tp->t_rttvar;
1316 metrics.rmx_cwnd = tp->snd_cwnd;
1317 metrics.rmx_sendpipe = 0;
1318 metrics.rmx_recvpipe = 0;
1320 tcp_hc_update(&inp->inp_inc, &metrics);
1323 /* free the reassembly queue, if any */
1324 tcp_reass_flush(tp);
1327 /* Disconnect offload device, if any. */
1328 if (tp->t_flags & TF_TOE)
1329 tcp_offload_detach(tp);
1332 tcp_free_sackholes(tp);
1335 /* Free the TCP PCAP queues. */
1336 tcp_pcap_drain(&(tp->t_inpkts));
1337 tcp_pcap_drain(&(tp->t_outpkts));
1340 /* Allow the CC algorithm to clean up after itself. */
1341 if (CC_ALGO(tp)->cb_destroy != NULL)
1342 CC_ALGO(tp)->cb_destroy(tp->ccv);
1344 khelp_destroy_osd(tp->osd);
1347 inp->inp_ppcb = NULL;
1348 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1349 /* We own the last reference on tcpcb, let's free it. */
1350 if ((tp->t_fb->tfb_tcp_timers_left) &&
1351 (tp->t_fb->tfb_tcp_timers_left(tp))) {
1352 /* Some fb timers left running! */
1355 if (tp->t_fb->tfb_tcp_fb_fini)
1356 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1357 refcount_release(&tp->t_fb->tfb_refcnt);
1359 uma_zfree(V_tcpcb_zone, tp);
1360 released = in_pcbrele_wlocked(inp);
1361 KASSERT(!released, ("%s: inp %p should not have been released "
1362 "here", __func__, inp));
1367 tcp_timer_2msl_discard(void *xtp)
1370 tcp_timer_discard((struct tcpcb *)xtp, TT_2MSL);
1374 tcp_timer_keep_discard(void *xtp)
1377 tcp_timer_discard((struct tcpcb *)xtp, TT_KEEP);
1381 tcp_timer_persist_discard(void *xtp)
1384 tcp_timer_discard((struct tcpcb *)xtp, TT_PERSIST);
1388 tcp_timer_rexmt_discard(void *xtp)
1391 tcp_timer_discard((struct tcpcb *)xtp, TT_REXMT);
1395 tcp_timer_delack_discard(void *xtp)
1398 tcp_timer_discard((struct tcpcb *)xtp, TT_DELACK);
1402 tcp_timer_discard(struct tcpcb *tp, uint32_t timer_type)
1406 CURVNET_SET(tp->t_vnet);
1407 INP_INFO_RLOCK(&V_tcbinfo);
1409 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1412 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1413 ("%s: tcpcb has to be stopped here", __func__));
1414 KASSERT((tp->t_timers->tt_flags & timer_type) != 0,
1415 ("%s: discard callout should be running", __func__));
1416 tp->t_timers->tt_flags &= ~timer_type;
1417 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1418 /* We own the last reference on this tcpcb, let's free it. */
1419 if ((tp->t_fb->tfb_tcp_timers_left) &&
1420 (tp->t_fb->tfb_tcp_timers_left(tp))) {
1421 /* Some fb timers left running! */
1424 if (tp->t_fb->tfb_tcp_fb_fini)
1425 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1426 refcount_release(&tp->t_fb->tfb_refcnt);
1428 uma_zfree(V_tcpcb_zone, tp);
1429 if (in_pcbrele_wlocked(inp)) {
1430 INP_INFO_RUNLOCK(&V_tcbinfo);
1437 INP_INFO_RUNLOCK(&V_tcbinfo);
1442 * Attempt to close a TCP control block, marking it as dropped, and freeing
1443 * the socket if we hold the only reference.
1446 tcp_close(struct tcpcb *tp)
1448 struct inpcb *inp = tp->t_inpcb;
1451 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1452 INP_WLOCK_ASSERT(inp);
1455 if (tp->t_state == TCPS_LISTEN)
1456 tcp_offload_listen_stop(tp);
1460 * This releases the TFO pending counter resource for TFO listen
1461 * sockets as well as passively-created TFO sockets that transition
1462 * from SYN_RECEIVED to CLOSED.
1464 if (tp->t_tfo_pending) {
1465 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1466 tp->t_tfo_pending = NULL;
1470 TCPSTAT_INC(tcps_closed);
1471 TCPSTAT_DEC(tcps_states[tp->t_state]);
1472 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1473 so = inp->inp_socket;
1474 soisdisconnected(so);
1475 if (inp->inp_flags & INP_SOCKREF) {
1476 KASSERT(so->so_state & SS_PROTOREF,
1477 ("tcp_close: !SS_PROTOREF"));
1478 inp->inp_flags &= ~INP_SOCKREF;
1482 so->so_state &= ~SS_PROTOREF;
1492 VNET_ITERATOR_DECL(vnet_iter);
1497 VNET_LIST_RLOCK_NOSLEEP();
1498 VNET_FOREACH(vnet_iter) {
1499 CURVNET_SET(vnet_iter);
1504 * Walk the tcpbs, if existing, and flush the reassembly queue,
1505 * if there is one...
1506 * XXX: The "Net/3" implementation doesn't imply that the TCP
1507 * reassembly queue should be flushed, but in a situation
1508 * where we're really low on mbufs, this is potentially
1511 INP_INFO_WLOCK(&V_tcbinfo);
1512 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1513 if (inpb->inp_flags & INP_TIMEWAIT)
1516 if ((tcpb = intotcpcb(inpb)) != NULL) {
1517 tcp_reass_flush(tcpb);
1518 tcp_clean_sackreport(tcpb);
1522 INP_INFO_WUNLOCK(&V_tcbinfo);
1525 VNET_LIST_RUNLOCK_NOSLEEP();
1529 * Notify a tcp user of an asynchronous error;
1530 * store error as soft error, but wake up user
1531 * (for now, won't do anything until can select for soft error).
1533 * Do not wake up user since there currently is no mechanism for
1534 * reporting soft errors (yet - a kqueue filter may be added).
1536 static struct inpcb *
1537 tcp_notify(struct inpcb *inp, int error)
1541 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1542 INP_WLOCK_ASSERT(inp);
1544 if ((inp->inp_flags & INP_TIMEWAIT) ||
1545 (inp->inp_flags & INP_DROPPED))
1548 tp = intotcpcb(inp);
1549 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1552 * Ignore some errors if we are hooked up.
1553 * If connection hasn't completed, has retransmitted several times,
1554 * and receives a second error, give up now. This is better
1555 * than waiting a long time to establish a connection that
1556 * can never complete.
1558 if (tp->t_state == TCPS_ESTABLISHED &&
1559 (error == EHOSTUNREACH || error == ENETUNREACH ||
1560 error == EHOSTDOWN)) {
1562 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1564 tp = tcp_drop(tp, error);
1570 tp->t_softerror = error;
1574 wakeup( &so->so_timeo);
1581 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1583 int error, i, m, n, pcb_count;
1584 struct inpcb *inp, **inp_list;
1589 * The process of preparing the TCB list is too time-consuming and
1590 * resource-intensive to repeat twice on every request.
1592 if (req->oldptr == NULL) {
1593 n = V_tcbinfo.ipi_count +
1594 TCPSTAT_FETCH(tcps_states[TCPS_SYN_RECEIVED]);
1595 n += imax(n / 8, 10);
1596 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1600 if (req->newptr != NULL)
1604 * OK, now we're committed to doing something.
1606 INP_LIST_RLOCK(&V_tcbinfo);
1607 gencnt = V_tcbinfo.ipi_gencnt;
1608 n = V_tcbinfo.ipi_count;
1609 INP_LIST_RUNLOCK(&V_tcbinfo);
1611 m = TCPSTAT_FETCH(tcps_states[TCPS_SYN_RECEIVED]);
1613 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1614 + (n + m) * sizeof(struct xtcpcb));
1618 xig.xig_len = sizeof xig;
1619 xig.xig_count = n + m;
1620 xig.xig_gen = gencnt;
1621 xig.xig_sogen = so_gencnt;
1622 error = SYSCTL_OUT(req, &xig, sizeof xig);
1626 error = syncache_pcblist(req, m, &pcb_count);
1630 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1631 if (inp_list == NULL)
1634 INP_INFO_WLOCK(&V_tcbinfo);
1635 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1636 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1638 if (inp->inp_gencnt <= gencnt) {
1640 * XXX: This use of cr_cansee(), introduced with
1641 * TCP state changes, is not quite right, but for
1642 * now, better than nothing.
1644 if (inp->inp_flags & INP_TIMEWAIT) {
1645 if (intotw(inp) != NULL)
1646 error = cr_cansee(req->td->td_ucred,
1647 intotw(inp)->tw_cred);
1649 error = EINVAL; /* Skip this inp. */
1651 error = cr_canseeinpcb(req->td->td_ucred, inp);
1654 inp_list[i++] = inp;
1659 INP_INFO_WUNLOCK(&V_tcbinfo);
1663 for (i = 0; i < n; i++) {
1666 if (inp->inp_gencnt <= gencnt) {
1670 bzero(&xt, sizeof(xt));
1671 xt.xt_len = sizeof xt;
1672 /* XXX should avoid extra copy */
1673 bcopy(inp, &xt.xt_inp, sizeof *inp);
1674 inp_ppcb = inp->inp_ppcb;
1675 if (inp_ppcb == NULL)
1676 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1677 else if (inp->inp_flags & INP_TIMEWAIT) {
1678 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1679 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1681 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1682 if (xt.xt_tp.t_timers)
1683 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1685 if (inp->inp_socket != NULL)
1686 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1688 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1689 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1691 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1693 error = SYSCTL_OUT(req, &xt, sizeof xt);
1697 INP_INFO_RLOCK(&V_tcbinfo);
1698 for (i = 0; i < n; i++) {
1701 if (!in_pcbrele_rlocked(inp))
1704 INP_INFO_RUNLOCK(&V_tcbinfo);
1708 * Give the user an updated idea of our state.
1709 * If the generation differs from what we told
1710 * her before, she knows that something happened
1711 * while we were processing this request, and it
1712 * might be necessary to retry.
1714 INP_LIST_RLOCK(&V_tcbinfo);
1715 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1716 xig.xig_sogen = so_gencnt;
1717 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1718 INP_LIST_RUNLOCK(&V_tcbinfo);
1719 error = SYSCTL_OUT(req, &xig, sizeof xig);
1721 free(inp_list, M_TEMP);
1725 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1726 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1727 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1731 tcp_getcred(SYSCTL_HANDLER_ARGS)
1734 struct sockaddr_in addrs[2];
1738 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1741 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1744 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1745 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1747 if (inp->inp_socket == NULL)
1750 error = cr_canseeinpcb(req->td->td_ucred, inp);
1752 cru2x(inp->inp_cred, &xuc);
1757 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1761 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1762 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1763 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1768 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1771 struct sockaddr_in6 addrs[2];
1778 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1781 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1784 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1785 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1788 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1790 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1799 inp = in_pcblookup(&V_tcbinfo,
1800 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1802 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1803 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1806 inp = in6_pcblookup(&V_tcbinfo,
1807 &addrs[1].sin6_addr, addrs[1].sin6_port,
1808 &addrs[0].sin6_addr, addrs[0].sin6_port,
1809 INPLOOKUP_RLOCKPCB, NULL);
1811 if (inp->inp_socket == NULL)
1814 error = cr_canseeinpcb(req->td->td_ucred, inp);
1816 cru2x(inp->inp_cred, &xuc);
1821 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1825 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1826 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1827 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1833 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1835 struct ip *ip = vip;
1837 struct in_addr faddr;
1840 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1842 struct in_conninfo inc;
1843 tcp_seq icmp_tcp_seq;
1846 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1847 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1850 if (cmd == PRC_MSGSIZE)
1851 notify = tcp_mtudisc_notify;
1852 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1853 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1854 notify = tcp_drop_syn_sent;
1856 * Redirects don't need to be handled up here.
1858 else if (PRC_IS_REDIRECT(cmd))
1861 * Hostdead is ugly because it goes linearly through all PCBs.
1862 * XXX: We never get this from ICMP, otherwise it makes an
1863 * excellent DoS attack on machines with many connections.
1865 else if (cmd == PRC_HOSTDEAD)
1867 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1871 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1875 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1876 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1877 INP_INFO_RLOCK(&V_tcbinfo);
1878 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
1879 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1881 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1882 !(inp->inp_flags & INP_DROPPED) &&
1883 !(inp->inp_socket == NULL)) {
1884 icmp_tcp_seq = ntohl(th->th_seq);
1885 tp = intotcpcb(inp);
1886 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1887 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1888 if (cmd == PRC_MSGSIZE) {
1891 * If we got a needfrag set the MTU
1892 * in the route to the suggested new
1893 * value (if given) and then notify.
1895 mtu = ntohs(icp->icmp_nextmtu);
1897 * If no alternative MTU was
1898 * proposed, try the next smaller
1903 ntohs(ip->ip_len), 1);
1904 if (mtu < V_tcp_minmss +
1905 sizeof(struct tcpiphdr))
1906 mtu = V_tcp_minmss +
1907 sizeof(struct tcpiphdr);
1909 * Only process the offered MTU if it
1910 * is smaller than the current one.
1912 if (mtu < tp->t_maxseg +
1913 sizeof(struct tcpiphdr)) {
1914 bzero(&inc, sizeof(inc));
1915 inc.inc_faddr = faddr;
1917 inp->inp_inc.inc_fibnum;
1918 tcp_hc_updatemtu(&inc, mtu);
1919 tcp_mtudisc(inp, mtu);
1922 inp = (*notify)(inp,
1923 inetctlerrmap[cmd]);
1929 bzero(&inc, sizeof(inc));
1930 inc.inc_fport = th->th_dport;
1931 inc.inc_lport = th->th_sport;
1932 inc.inc_faddr = faddr;
1933 inc.inc_laddr = ip->ip_src;
1934 syncache_unreach(&inc, th);
1936 INP_INFO_RUNLOCK(&V_tcbinfo);
1942 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1945 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1946 struct ip6_hdr *ip6;
1948 struct ip6ctlparam *ip6cp = NULL;
1949 const struct sockaddr_in6 *sa6_src = NULL;
1951 struct tcp_portonly {
1956 if (sa->sa_family != AF_INET6 ||
1957 sa->sa_len != sizeof(struct sockaddr_in6))
1960 if (cmd == PRC_MSGSIZE)
1961 notify = tcp_mtudisc_notify;
1962 else if (!PRC_IS_REDIRECT(cmd) &&
1963 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1966 /* if the parameter is from icmp6, decode it. */
1968 ip6cp = (struct ip6ctlparam *)d;
1970 ip6 = ip6cp->ip6c_ip6;
1971 off = ip6cp->ip6c_off;
1972 sa6_src = ip6cp->ip6c_src;
1976 off = 0; /* fool gcc */
1981 struct in_conninfo inc;
1983 * XXX: We assume that when IPV6 is non NULL,
1984 * M and OFF are valid.
1987 /* check if we can safely examine src and dst ports */
1988 if (m->m_pkthdr.len < off + sizeof(*thp))
1991 bzero(&th, sizeof(th));
1992 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1994 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1995 (struct sockaddr *)ip6cp->ip6c_src,
1996 th.th_sport, cmd, NULL, notify);
1998 bzero(&inc, sizeof(inc));
1999 inc.inc_fport = th.th_dport;
2000 inc.inc_lport = th.th_sport;
2001 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
2002 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
2003 inc.inc_flags |= INC_ISIPV6;
2004 INP_INFO_RLOCK(&V_tcbinfo);
2005 syncache_unreach(&inc, &th);
2006 INP_INFO_RUNLOCK(&V_tcbinfo);
2008 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
2009 0, cmd, NULL, notify);
2015 * Following is where TCP initial sequence number generation occurs.
2017 * There are two places where we must use initial sequence numbers:
2018 * 1. In SYN-ACK packets.
2019 * 2. In SYN packets.
2021 * All ISNs for SYN-ACK packets are generated by the syncache. See
2022 * tcp_syncache.c for details.
2024 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2025 * depends on this property. In addition, these ISNs should be
2026 * unguessable so as to prevent connection hijacking. To satisfy
2027 * the requirements of this situation, the algorithm outlined in
2028 * RFC 1948 is used, with only small modifications.
2030 * Implementation details:
2032 * Time is based off the system timer, and is corrected so that it
2033 * increases by one megabyte per second. This allows for proper
2034 * recycling on high speed LANs while still leaving over an hour
2037 * As reading the *exact* system time is too expensive to be done
2038 * whenever setting up a TCP connection, we increment the time
2039 * offset in two ways. First, a small random positive increment
2040 * is added to isn_offset for each connection that is set up.
2041 * Second, the function tcp_isn_tick fires once per clock tick
2042 * and increments isn_offset as necessary so that sequence numbers
2043 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2044 * random positive increments serve only to ensure that the same
2045 * exact sequence number is never sent out twice (as could otherwise
2046 * happen when a port is recycled in less than the system tick
2049 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2050 * between seeding of isn_secret. This is normally set to zero,
2051 * as reseeding should not be necessary.
2053 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2054 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
2055 * general, this means holding an exclusive (write) lock.
2058 #define ISN_BYTES_PER_SECOND 1048576
2059 #define ISN_STATIC_INCREMENT 4096
2060 #define ISN_RANDOM_INCREMENT (4096 - 1)
2062 static VNET_DEFINE(u_char, isn_secret[32]);
2063 static VNET_DEFINE(int, isn_last);
2064 static VNET_DEFINE(int, isn_last_reseed);
2065 static VNET_DEFINE(u_int32_t, isn_offset);
2066 static VNET_DEFINE(u_int32_t, isn_offset_old);
2068 #define V_isn_secret VNET(isn_secret)
2069 #define V_isn_last VNET(isn_last)
2070 #define V_isn_last_reseed VNET(isn_last_reseed)
2071 #define V_isn_offset VNET(isn_offset)
2072 #define V_isn_offset_old VNET(isn_offset_old)
2075 tcp_new_isn(struct tcpcb *tp)
2078 u_int32_t md5_buffer[4];
2080 u_int32_t projected_offset;
2082 INP_WLOCK_ASSERT(tp->t_inpcb);
2085 /* Seed if this is the first use, reseed if requested. */
2086 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2087 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2089 read_random(&V_isn_secret, sizeof(V_isn_secret));
2090 V_isn_last_reseed = ticks;
2093 /* Compute the md5 hash and return the ISN. */
2095 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2096 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2098 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2099 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2100 sizeof(struct in6_addr));
2101 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2102 sizeof(struct in6_addr));
2106 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2107 sizeof(struct in_addr));
2108 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2109 sizeof(struct in_addr));
2111 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2112 MD5Final((u_char *) &md5_buffer, &isn_ctx);
2113 new_isn = (tcp_seq) md5_buffer[0];
2114 V_isn_offset += ISN_STATIC_INCREMENT +
2115 (arc4random() & ISN_RANDOM_INCREMENT);
2116 if (ticks != V_isn_last) {
2117 projected_offset = V_isn_offset_old +
2118 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2119 if (SEQ_GT(projected_offset, V_isn_offset))
2120 V_isn_offset = projected_offset;
2121 V_isn_offset_old = V_isn_offset;
2124 new_isn += V_isn_offset;
2130 * When a specific ICMP unreachable message is received and the
2131 * connection state is SYN-SENT, drop the connection. This behavior
2132 * is controlled by the icmp_may_rst sysctl.
2135 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2139 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2140 INP_WLOCK_ASSERT(inp);
2142 if ((inp->inp_flags & INP_TIMEWAIT) ||
2143 (inp->inp_flags & INP_DROPPED))
2146 tp = intotcpcb(inp);
2147 if (tp->t_state != TCPS_SYN_SENT)
2150 tp = tcp_drop(tp, errno);
2158 * When `need fragmentation' ICMP is received, update our idea of the MSS
2159 * based on the new value. Also nudge TCP to send something, since we
2160 * know the packet we just sent was dropped.
2161 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2163 static struct inpcb *
2164 tcp_mtudisc_notify(struct inpcb *inp, int error)
2167 tcp_mtudisc(inp, -1);
2172 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2177 INP_WLOCK_ASSERT(inp);
2178 if ((inp->inp_flags & INP_TIMEWAIT) ||
2179 (inp->inp_flags & INP_DROPPED))
2182 tp = intotcpcb(inp);
2183 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2185 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2187 so = inp->inp_socket;
2188 SOCKBUF_LOCK(&so->so_snd);
2189 /* If the mss is larger than the socket buffer, decrease the mss. */
2190 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2191 tp->t_maxseg = so->so_snd.sb_hiwat;
2192 SOCKBUF_UNLOCK(&so->so_snd);
2194 TCPSTAT_INC(tcps_mturesent);
2196 tp->snd_nxt = tp->snd_una;
2197 tcp_free_sackholes(tp);
2198 tp->snd_recover = tp->snd_max;
2199 if (tp->t_flags & TF_SACK_PERMIT)
2200 EXIT_FASTRECOVERY(tp->t_flags);
2201 tp->t_fb->tfb_tcp_output(tp);
2206 * Look-up the routing entry to the peer of this inpcb. If no route
2207 * is found and it cannot be allocated, then return 0. This routine
2208 * is called by TCP routines that access the rmx structure and by
2209 * tcp_mss_update to get the peer/interface MTU.
2212 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2214 struct nhop4_extended nh4;
2218 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2220 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2222 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2223 NHR_REF, 0, &nh4) != 0)
2227 maxmtu = nh4.nh_mtu;
2229 /* Report additional interface capabilities. */
2231 if (ifp->if_capenable & IFCAP_TSO4 &&
2232 ifp->if_hwassist & CSUM_TSO) {
2233 cap->ifcap |= CSUM_TSO;
2234 cap->tsomax = ifp->if_hw_tsomax;
2235 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2236 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2239 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2247 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2249 struct nhop6_extended nh6;
2250 struct in6_addr dst6;
2255 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2257 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2258 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2259 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2264 maxmtu = nh6.nh_mtu;
2266 /* Report additional interface capabilities. */
2268 if (ifp->if_capenable & IFCAP_TSO6 &&
2269 ifp->if_hwassist & CSUM_TSO) {
2270 cap->ifcap |= CSUM_TSO;
2271 cap->tsomax = ifp->if_hw_tsomax;
2272 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2273 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2276 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2284 * Calculate effective SMSS per RFC5681 definition for a given TCP
2285 * connection at its current state, taking into account SACK and etc.
2288 tcp_maxseg(const struct tcpcb *tp)
2292 if (tp->t_flags & TF_NOOPT)
2293 return (tp->t_maxseg);
2296 * Here we have a simplified code from tcp_addoptions(),
2297 * without a proper loop, and having most of paddings hardcoded.
2298 * We might make mistakes with padding here in some edge cases,
2299 * but this is harmless, since result of tcp_maxseg() is used
2300 * only in cwnd and ssthresh estimations.
2302 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2303 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2304 if (tp->t_flags & TF_RCVD_TSTMP)
2305 optlen = TCPOLEN_TSTAMP_APPA;
2308 #ifdef TCP_SIGNATURE
2309 if (tp->t_flags & TF_SIGNATURE)
2310 optlen += PAD(TCPOLEN_SIGNATURE);
2312 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2313 optlen += TCPOLEN_SACKHDR;
2314 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2315 optlen = PAD(optlen);
2318 if (tp->t_flags & TF_REQ_TSTMP)
2319 optlen = TCPOLEN_TSTAMP_APPA;
2321 optlen = PAD(TCPOLEN_MAXSEG);
2322 if (tp->t_flags & TF_REQ_SCALE)
2323 optlen += PAD(TCPOLEN_WINDOW);
2324 #ifdef TCP_SIGNATURE
2325 if (tp->t_flags & TF_SIGNATURE)
2326 optlen += PAD(TCPOLEN_SIGNATURE);
2328 if (tp->t_flags & TF_SACK_PERMIT)
2329 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2332 optlen = min(optlen, TCP_MAXOLEN);
2333 return (tp->t_maxseg - optlen);
2337 /* compute ESP/AH header size for TCP, including outer IP header. */
2339 ipsec_hdrsiz_tcp(struct tcpcb *tp)
2346 struct ip6_hdr *ip6;
2350 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) ||
2351 (!key_havesp(IPSEC_DIR_OUTBOUND)))
2353 m = m_gethdr(M_NOWAIT, MT_DATA);
2358 if ((inp->inp_vflag & INP_IPV6) != 0) {
2359 ip6 = mtod(m, struct ip6_hdr *);
2360 th = (struct tcphdr *)(ip6 + 1);
2361 m->m_pkthdr.len = m->m_len =
2362 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
2363 tcpip_fillheaders(inp, ip6, th);
2364 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2368 ip = mtod(m, struct ip *);
2369 th = (struct tcphdr *)(ip + 1);
2370 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
2371 tcpip_fillheaders(inp, ip, th);
2372 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2380 #ifdef TCP_SIGNATURE
2382 * Callback function invoked by m_apply() to digest TCP segment data
2383 * contained within an mbuf chain.
2386 tcp_signature_apply(void *fstate, void *data, u_int len)
2389 MD5Update(fstate, (u_char *)data, len);
2394 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2395 * search with the destination IP address, and a 'magic SPI' to be
2396 * determined by the application. This is hardcoded elsewhere to 1179
2399 tcp_get_sav(struct mbuf *m, u_int direction)
2401 union sockaddr_union dst;
2402 struct secasvar *sav;
2405 struct ip6_hdr *ip6;
2406 char ip6buf[INET6_ADDRSTRLEN];
2409 /* Extract the destination from the IP header in the mbuf. */
2410 bzero(&dst, sizeof(union sockaddr_union));
2411 ip = mtod(m, struct ip *);
2413 ip6 = NULL; /* Make the compiler happy. */
2418 dst.sa.sa_len = sizeof(struct sockaddr_in);
2419 dst.sa.sa_family = AF_INET;
2420 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2421 ip->ip_src : ip->ip_dst;
2425 case (IPV6_VERSION >> 4):
2426 ip6 = mtod(m, struct ip6_hdr *);
2427 dst.sa.sa_len = sizeof(struct sockaddr_in6);
2428 dst.sa.sa_family = AF_INET6;
2429 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2430 ip6->ip6_src : ip6->ip6_dst;
2439 /* Look up an SADB entry which matches the address of the peer. */
2440 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2442 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2443 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2445 (ip->ip_v == (IPV6_VERSION >> 4)) ?
2446 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2455 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2458 * m pointer to head of mbuf chain
2459 * len length of TCP segment data, excluding options
2460 * optlen length of TCP segment options
2461 * buf pointer to storage for computed MD5 digest
2462 * sav pointer to security assosiation
2464 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2465 * When called from tcp_input(), we can be sure that th_sum has been
2466 * zeroed out and verified already.
2468 * Releases reference to SADB key before return.
2470 * Return 0 if successful, otherwise return -1.
2474 tcp_signature_do_compute(struct mbuf *m, int len, int optlen,
2475 u_char *buf, struct secasvar *sav)
2478 struct ippseudo ippseudo;
2484 struct ipovly *ipovly;
2488 struct ip6_hdr *ip6;
2489 struct in6_addr in6;
2495 KASSERT(m != NULL, ("NULL mbuf chain"));
2496 KASSERT(buf != NULL, ("NULL signature pointer"));
2498 /* Extract the destination from the IP header in the mbuf. */
2499 ip = mtod(m, struct ip *);
2501 ip6 = NULL; /* Make the compiler happy. */
2506 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2508 * XXX The ippseudo header MUST be digested in network byte order,
2509 * or else we'll fail the regression test. Assume all fields we've
2510 * been doing arithmetic on have been in host byte order.
2511 * XXX One cannot depend on ipovly->ih_len here. When called from
2512 * tcp_output(), the underlying ip_len member has not yet been set.
2517 ipovly = (struct ipovly *)ip;
2518 ippseudo.ippseudo_src = ipovly->ih_src;
2519 ippseudo.ippseudo_dst = ipovly->ih_dst;
2520 ippseudo.ippseudo_pad = 0;
2521 ippseudo.ippseudo_p = IPPROTO_TCP;
2522 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2524 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2526 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2527 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2532 * RFC 2385, 2.0 Proposal
2533 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2534 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2535 * extended next header value (to form 32 bits), and 32-bit segment
2537 * Note: Upper-Layer Packet Length comes before Next Header.
2539 case (IPV6_VERSION >> 4):
2541 in6_clearscope(&in6);
2542 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2544 in6_clearscope(&in6);
2545 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2546 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2547 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2549 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2550 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2551 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2553 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2555 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2556 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2568 * Step 2: Update MD5 hash with TCP header, excluding options.
2569 * The TCP checksum must be set to zero.
2571 savecsum = th->th_sum;
2573 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2574 th->th_sum = savecsum;
2577 * Step 3: Update MD5 hash with TCP segment data.
2578 * Use m_apply() to avoid an early m_pullup().
2581 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2584 * Step 4: Update MD5 hash with shared secret.
2586 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2587 MD5Final(buf, &ctx);
2589 key_sa_recordxfer(sav, m);
2595 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2597 * Return 0 if successful, otherwise return -1.
2600 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2601 u_char *buf, u_int direction)
2603 struct secasvar *sav;
2605 if ((sav = tcp_get_sav(m, direction)) == NULL)
2608 return (tcp_signature_do_compute(m, len, optlen, buf, sav));
2612 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2615 * m pointer to head of mbuf chain
2616 * len length of TCP segment data, excluding options
2617 * optlen length of TCP segment options
2618 * buf pointer to storage for computed MD5 digest
2619 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2621 * Return 1 if successful, otherwise return 0.
2624 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2625 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2627 char tmpdigest[TCP_SIGLEN];
2629 if (tcp_sig_checksigs == 0)
2631 if ((tcpbflag & TF_SIGNATURE) == 0) {
2632 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2635 * If this socket is not expecting signature but
2636 * the segment contains signature just fail.
2638 TCPSTAT_INC(tcps_sig_err_sigopt);
2639 TCPSTAT_INC(tcps_sig_rcvbadsig);
2643 /* Signature is not expected, and not present in segment. */
2648 * If this socket is expecting signature but the segment does not
2649 * contain any just fail.
2651 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2652 TCPSTAT_INC(tcps_sig_err_nosigopt);
2653 TCPSTAT_INC(tcps_sig_rcvbadsig);
2656 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2657 IPSEC_DIR_INBOUND) == -1) {
2658 TCPSTAT_INC(tcps_sig_err_buildsig);
2659 TCPSTAT_INC(tcps_sig_rcvbadsig);
2663 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2664 TCPSTAT_INC(tcps_sig_rcvbadsig);
2667 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2670 #endif /* TCP_SIGNATURE */
2673 sysctl_drop(SYSCTL_HANDLER_ARGS)
2675 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2676 struct sockaddr_storage addrs[2];
2680 struct sockaddr_in *fin, *lin;
2682 struct sockaddr_in6 *fin6, *lin6;
2693 if (req->oldptr != NULL || req->oldlen != 0)
2695 if (req->newptr == NULL)
2697 if (req->newlen < sizeof(addrs))
2699 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2703 switch (addrs[0].ss_family) {
2706 fin6 = (struct sockaddr_in6 *)&addrs[0];
2707 lin6 = (struct sockaddr_in6 *)&addrs[1];
2708 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2709 lin6->sin6_len != sizeof(struct sockaddr_in6))
2711 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2712 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2714 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2715 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2716 fin = (struct sockaddr_in *)&addrs[0];
2717 lin = (struct sockaddr_in *)&addrs[1];
2720 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2723 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2730 fin = (struct sockaddr_in *)&addrs[0];
2731 lin = (struct sockaddr_in *)&addrs[1];
2732 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2733 lin->sin_len != sizeof(struct sockaddr_in))
2740 INP_INFO_RLOCK(&V_tcbinfo);
2741 switch (addrs[0].ss_family) {
2744 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2745 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2746 INPLOOKUP_WLOCKPCB, NULL);
2751 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2752 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2757 if (inp->inp_flags & INP_TIMEWAIT) {
2759 * XXXRW: There currently exists a state where an
2760 * inpcb is present, but its timewait state has been
2761 * discarded. For now, don't allow dropping of this
2769 } else if (!(inp->inp_flags & INP_DROPPED) &&
2770 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2771 tp = intotcpcb(inp);
2772 tp = tcp_drop(tp, ECONNABORTED);
2779 INP_INFO_RUNLOCK(&V_tcbinfo);
2783 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2784 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2785 0, sysctl_drop, "", "Drop TCP connection");
2788 * Generate a standardized TCP log line for use throughout the
2789 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2790 * allow use in the interrupt context.
2792 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2793 * NB: The function may return NULL if memory allocation failed.
2795 * Due to header inclusion and ordering limitations the struct ip
2796 * and ip6_hdr pointers have to be passed as void pointers.
2799 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2803 /* Is logging enabled? */
2804 if (tcp_log_in_vain == 0)
2807 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2811 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2815 /* Is logging enabled? */
2816 if (tcp_log_debug == 0)
2819 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2823 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2830 const struct ip6_hdr *ip6;
2832 ip6 = (const struct ip6_hdr *)ip6hdr;
2834 ip = (struct ip *)ip4hdr;
2837 * The log line looks like this:
2838 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2840 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2841 sizeof(PRINT_TH_FLAGS) + 1 +
2843 2 * INET6_ADDRSTRLEN;
2845 2 * INET_ADDRSTRLEN;
2848 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2852 strcat(s, "TCP: [");
2855 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2856 inet_ntoa_r(inc->inc_faddr, sp);
2858 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2860 inet_ntoa_r(inc->inc_laddr, sp);
2862 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2865 ip6_sprintf(sp, &inc->inc6_faddr);
2867 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2869 ip6_sprintf(sp, &inc->inc6_laddr);
2871 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2872 } else if (ip6 && th) {
2873 ip6_sprintf(sp, &ip6->ip6_src);
2875 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2877 ip6_sprintf(sp, &ip6->ip6_dst);
2879 sprintf(sp, "]:%i", ntohs(th->th_dport));
2882 } else if (ip && th) {
2883 inet_ntoa_r(ip->ip_src, sp);
2885 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2887 inet_ntoa_r(ip->ip_dst, sp);
2889 sprintf(sp, "]:%i", ntohs(th->th_dport));
2897 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2898 if (*(s + size - 1) != '\0')
2899 panic("%s: string too long", __func__);
2904 * A subroutine which makes it easy to track TCP state changes with DTrace.
2905 * This function shouldn't be called for t_state initializations that don't
2906 * correspond to actual TCP state transitions.
2909 tcp_state_change(struct tcpcb *tp, int newstate)
2911 #if defined(KDTRACE_HOOKS)
2912 int pstate = tp->t_state;
2915 TCPSTAT_DEC(tcps_states[tp->t_state]);
2916 TCPSTAT_INC(tcps_states[newstate]);
2917 tp->t_state = newstate;
2918 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);