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_kdtrace.h>
73 #include <netinet/in_pcb.h>
74 #include <netinet/in_systm.h>
75 #include <netinet/in_var.h>
76 #include <netinet/ip.h>
77 #include <netinet/ip_icmp.h>
78 #include <netinet/ip_var.h>
80 #include <netinet/ip6.h>
81 #include <netinet6/in6_pcb.h>
82 #include <netinet6/ip6_var.h>
83 #include <netinet6/scope6_var.h>
84 #include <netinet6/nd6.h>
88 #include <netinet/tcp_fastopen.h>
90 #include <netinet/tcp.h>
91 #include <netinet/tcp_fsm.h>
92 #include <netinet/tcp_seq.h>
93 #include <netinet/tcp_timer.h>
94 #include <netinet/tcp_var.h>
95 #include <netinet/tcp_syncache.h>
96 #include <netinet/tcp_cc.h>
98 #include <netinet6/tcp6_var.h>
100 #include <netinet/tcpip.h>
102 #include <netinet/tcp_pcap.h>
105 #include <netinet/tcp_debug.h>
108 #include <netinet6/ip6protosw.h>
111 #include <netinet/tcp_offload.h>
115 #include <netipsec/ipsec.h>
116 #include <netipsec/xform.h>
118 #include <netipsec/ipsec6.h>
120 #include <netipsec/key.h>
121 #include <sys/syslog.h>
124 #include <machine/in_cksum.h>
127 #include <security/mac/mac_framework.h>
129 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
131 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
134 struct rwlock tcp_function_lock;
137 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
142 error = sysctl_handle_int(oidp, &new, 0, req);
143 if (error == 0 && req->newptr) {
144 if (new < TCP_MINMSS)
152 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
153 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
154 &sysctl_net_inet_tcp_mss_check, "I",
155 "Default TCP Maximum Segment Size");
159 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
163 new = V_tcp_v6mssdflt;
164 error = sysctl_handle_int(oidp, &new, 0, req);
165 if (error == 0 && req->newptr) {
166 if (new < TCP_MINMSS)
169 V_tcp_v6mssdflt = new;
174 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
175 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
176 &sysctl_net_inet_tcp_mss_v6_check, "I",
177 "Default TCP Maximum Segment Size for IPv6");
181 * Minimum MSS we accept and use. This prevents DoS attacks where
182 * we are forced to a ridiculous low MSS like 20 and send hundreds
183 * of packets instead of one. The effect scales with the available
184 * bandwidth and quickly saturates the CPU and network interface
185 * with packet generation and sending. Set to zero to disable MINMSS
186 * checking. This setting prevents us from sending too small packets.
188 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
190 &VNET_NAME(tcp_minmss), 0,
191 "Minimum TCP Maximum Segment Size");
193 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
194 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
195 &VNET_NAME(tcp_do_rfc1323), 0,
196 "Enable rfc1323 (high performance TCP) extensions");
198 static int tcp_log_debug = 0;
199 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
200 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
202 static int tcp_tcbhashsize;
203 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
204 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
206 static int do_tcpdrain = 1;
207 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
208 "Enable tcp_drain routine for extra help when low on mbufs");
210 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
211 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
213 static VNET_DEFINE(int, icmp_may_rst) = 1;
214 #define V_icmp_may_rst VNET(icmp_may_rst)
215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
216 &VNET_NAME(icmp_may_rst), 0,
217 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
219 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
220 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
222 &VNET_NAME(tcp_isn_reseed_interval), 0,
223 "Seconds between reseeding of ISN secret");
225 static int tcp_soreceive_stream;
226 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
227 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
230 static int tcp_sig_checksigs = 1;
231 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
232 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
235 VNET_DEFINE(uma_zone_t, sack_hole_zone);
236 #define V_sack_hole_zone VNET(sack_hole_zone)
238 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
240 static struct inpcb *tcp_notify(struct inpcb *, int);
241 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
242 static void tcp_mtudisc(struct inpcb *, int);
243 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
244 void *ip4hdr, const void *ip6hdr);
245 static void tcp_timer_discard(struct tcpcb *, uint32_t);
248 static struct tcp_function_block tcp_def_funcblk = {
252 tcp_default_ctloutput,
264 struct tcp_funchead t_functions;
265 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
267 static struct tcp_function_block *
268 find_tcp_functions_locked(struct tcp_function_set *fs)
270 struct tcp_function *f;
271 struct tcp_function_block *blk=NULL;
273 TAILQ_FOREACH(f, &t_functions, tf_next) {
274 if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) {
282 static struct tcp_function_block *
283 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
285 struct tcp_function_block *rblk=NULL;
286 struct tcp_function *f;
288 TAILQ_FOREACH(f, &t_functions, tf_next) {
289 if (f->tf_fb == blk) {
300 struct tcp_function_block *
301 find_and_ref_tcp_functions(struct tcp_function_set *fs)
303 struct tcp_function_block *blk;
305 rw_rlock(&tcp_function_lock);
306 blk = find_tcp_functions_locked(fs);
308 refcount_acquire(&blk->tfb_refcnt);
309 rw_runlock(&tcp_function_lock);
313 struct tcp_function_block *
314 find_and_ref_tcp_fb(struct tcp_function_block *blk)
316 struct tcp_function_block *rblk;
318 rw_rlock(&tcp_function_lock);
319 rblk = find_tcp_fb_locked(blk, NULL);
321 refcount_acquire(&rblk->tfb_refcnt);
322 rw_runlock(&tcp_function_lock);
328 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
331 struct tcp_function_set fs;
332 struct tcp_function_block *blk;
334 memset(&fs, 0, sizeof(fs));
335 rw_rlock(&tcp_function_lock);
336 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
339 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
340 fs.pcbcnt = blk->tfb_refcnt;
342 rw_runlock(&tcp_function_lock);
343 error = sysctl_handle_string(oidp, fs.function_set_name,
344 sizeof(fs.function_set_name), req);
346 /* Check for error or no change */
347 if (error != 0 || req->newptr == NULL)
350 rw_wlock(&tcp_function_lock);
351 blk = find_tcp_functions_locked(&fs);
353 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
357 tcp_func_set_ptr = blk;
359 rw_wunlock(&tcp_function_lock);
363 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
364 CTLTYPE_STRING | CTLFLAG_RW,
365 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
366 "Set/get the default TCP functions");
369 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
371 int error, cnt, linesz;
372 struct tcp_function *f;
377 rw_rlock(&tcp_function_lock);
378 TAILQ_FOREACH(f, &t_functions, tf_next) {
381 rw_runlock(&tcp_function_lock);
383 bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1;
384 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
389 linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count");
394 rw_rlock(&tcp_function_lock);
395 TAILQ_FOREACH(f, &t_functions, tf_next) {
396 linesz = snprintf(cp, bufsz, "%-32s%c %u\n",
397 f->tf_fb->tfb_tcp_block_name,
398 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
399 f->tf_fb->tfb_refcnt);
400 if (linesz >= bufsz) {
408 rw_runlock(&tcp_function_lock);
410 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
411 free(buffer, M_TEMP);
415 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
416 CTLTYPE_STRING|CTLFLAG_RD,
417 NULL, 0, sysctl_net_inet_list_available, "A",
418 "list available TCP Function sets");
421 * Target size of TCP PCB hash tables. Must be a power of two.
423 * Note that this can be overridden by the kernel environment
424 * variable net.inet.tcp.tcbhashsize
427 #define TCBHASHSIZE 0
432 * Callouts should be moved into struct tcp directly. They are currently
433 * separate because the tcpcb structure is exported to userland for sysctl
434 * parsing purposes, which do not know about callouts.
443 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
444 #define V_tcpcb_zone VNET(tcpcb_zone)
446 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
447 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
449 static struct mtx isn_mtx;
451 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
452 #define ISN_LOCK() mtx_lock(&isn_mtx)
453 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
456 * TCP initialization.
459 tcp_zone_change(void *tag)
462 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
463 uma_zone_set_max(V_tcpcb_zone, maxsockets);
464 tcp_tw_zone_change();
468 tcp_inpcb_init(void *mem, int size, int flags)
470 struct inpcb *inp = mem;
472 INP_LOCK_INIT(inp, "inp", "tcpinp");
477 * Take a value and get the next power of 2 that doesn't overflow.
478 * Used to size the tcp_inpcb hash buckets.
481 maketcp_hashsize(int size)
487 * get the next power of 2 higher than maxsockets.
489 hashsize = 1 << fls(size);
490 /* catch overflow, and just go one power of 2 smaller */
491 if (hashsize < size) {
492 hashsize = 1 << (fls(size) - 1);
498 register_tcp_functions(struct tcp_function_block *blk, int wait)
500 struct tcp_function_block *lblk;
501 struct tcp_function *n;
502 struct tcp_function_set fs;
504 if ((blk->tfb_tcp_output == NULL) ||
505 (blk->tfb_tcp_do_segment == NULL) ||
506 (blk->tfb_tcp_ctloutput == NULL) ||
507 (strlen(blk->tfb_tcp_block_name) == 0)) {
509 * These functions are required and you
514 if (blk->tfb_tcp_timer_stop_all ||
515 blk->tfb_tcp_timers_left ||
516 blk->tfb_tcp_timer_activate ||
517 blk->tfb_tcp_timer_active ||
518 blk->tfb_tcp_timer_stop) {
520 * If you define one timer function you
521 * must have them all.
523 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
524 (blk->tfb_tcp_timers_left == NULL) ||
525 (blk->tfb_tcp_timer_activate == NULL) ||
526 (blk->tfb_tcp_timer_active == NULL) ||
527 (blk->tfb_tcp_timer_stop == NULL)) {
531 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
536 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
537 rw_wlock(&tcp_function_lock);
538 lblk = find_tcp_functions_locked(&fs);
540 /* Duplicate name space not allowed */
541 rw_wunlock(&tcp_function_lock);
542 free(n, M_TCPFUNCTIONS);
545 refcount_init(&blk->tfb_refcnt, 0);
547 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
548 rw_wunlock(&tcp_function_lock);
553 deregister_tcp_functions(struct tcp_function_block *blk)
555 struct tcp_function_block *lblk;
556 struct tcp_function *f;
559 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
560 /* You can't un-register the default */
563 rw_wlock(&tcp_function_lock);
564 if (blk == tcp_func_set_ptr) {
565 /* You can't free the current default */
566 rw_wunlock(&tcp_function_lock);
569 if (blk->tfb_refcnt) {
570 /* Still tcb attached, mark it. */
571 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
572 rw_wunlock(&tcp_function_lock);
575 lblk = find_tcp_fb_locked(blk, &f);
578 TAILQ_REMOVE(&t_functions, f, tf_next);
580 free(f, M_TCPFUNCTIONS);
583 rw_wunlock(&tcp_function_lock);
590 const char *tcbhash_tuneable;
593 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
595 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
596 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
597 printf("%s: WARNING: unable to register helper hook\n", __func__);
598 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
599 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
600 printf("%s: WARNING: unable to register helper hook\n", __func__);
601 hashsize = TCBHASHSIZE;
602 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
605 * Auto tune the hash size based on maxsockets.
606 * A perfect hash would have a 1:1 mapping
607 * (hashsize = maxsockets) however it's been
608 * suggested that O(2) average is better.
610 hashsize = maketcp_hashsize(maxsockets / 4);
612 * Our historical default is 512,
613 * do not autotune lower than this.
617 if (bootverbose && IS_DEFAULT_VNET(curvnet))
618 printf("%s: %s auto tuned to %d\n", __func__,
619 tcbhash_tuneable, hashsize);
622 * We require a hashsize to be a power of two.
623 * Previously if it was not a power of two we would just reset it
624 * back to 512, which could be a nasty surprise if you did not notice
626 * Instead what we do is clip it to the closest power of two lower
627 * than the specified hash value.
629 if (!powerof2(hashsize)) {
630 int oldhashsize = hashsize;
632 hashsize = maketcp_hashsize(hashsize);
633 /* prevent absurdly low value */
636 printf("%s: WARNING: TCB hash size not a power of 2, "
637 "clipped from %d to %d.\n", __func__, oldhashsize,
640 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
641 "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
642 IPI_HASHFIELDS_4TUPLE);
645 * These have to be type stable for the benefit of the timers.
647 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
648 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
649 uma_zone_set_max(V_tcpcb_zone, maxsockets);
650 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
656 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
657 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
658 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
660 /* Skip initialization of globals for non-default instances. */
661 if (!IS_DEFAULT_VNET(curvnet))
664 tcp_reass_global_init();
666 /* XXX virtualize those bellow? */
667 tcp_delacktime = TCPTV_DELACK;
668 tcp_keepinit = TCPTV_KEEP_INIT;
669 tcp_keepidle = TCPTV_KEEP_IDLE;
670 tcp_keepintvl = TCPTV_KEEPINTVL;
671 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
673 tcp_rexmit_min = TCPTV_MIN;
674 if (tcp_rexmit_min < 1)
676 tcp_rexmit_slop = TCPTV_CPU_VAR;
677 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
678 tcp_tcbhashsize = hashsize;
679 /* Setup the tcp function block list */
680 TAILQ_INIT(&t_functions);
681 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
682 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
684 if (tcp_soreceive_stream) {
686 tcp_usrreqs.pru_soreceive = soreceive_stream;
689 tcp6_usrreqs.pru_soreceive = soreceive_stream;
694 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
696 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
698 if (max_protohdr < TCP_MINPROTOHDR)
699 max_protohdr = TCP_MINPROTOHDR;
700 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
702 #undef TCP_MINPROTOHDR
705 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
706 SHUTDOWN_PRI_DEFAULT);
707 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
708 EVENTHANDLER_PRI_ANY);
725 tcp_fastopen_destroy();
730 in_pcbinfo_destroy(&V_tcbinfo);
731 uma_zdestroy(V_sack_hole_zone);
732 uma_zdestroy(V_tcpcb_zone);
734 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
736 printf("%s: WARNING: unable to deregister helper hook "
737 "type=%d, id=%d: error %d returned\n", __func__,
738 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
740 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
742 printf("%s: WARNING: unable to deregister helper hook "
743 "type=%d, id=%d: error %d returned\n", __func__,
744 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
756 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
757 * tcp_template used to store this data in mbufs, but we now recopy it out
758 * of the tcpcb each time to conserve mbufs.
761 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
763 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
765 INP_WLOCK_ASSERT(inp);
768 if ((inp->inp_vflag & INP_IPV6) != 0) {
771 ip6 = (struct ip6_hdr *)ip_ptr;
772 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
773 (inp->inp_flow & IPV6_FLOWINFO_MASK);
774 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
775 (IPV6_VERSION & IPV6_VERSION_MASK);
776 ip6->ip6_nxt = IPPROTO_TCP;
777 ip6->ip6_plen = htons(sizeof(struct tcphdr));
778 ip6->ip6_src = inp->in6p_laddr;
779 ip6->ip6_dst = inp->in6p_faddr;
782 #if defined(INET6) && defined(INET)
789 ip = (struct ip *)ip_ptr;
790 ip->ip_v = IPVERSION;
792 ip->ip_tos = inp->inp_ip_tos;
796 ip->ip_ttl = inp->inp_ip_ttl;
798 ip->ip_p = IPPROTO_TCP;
799 ip->ip_src = inp->inp_laddr;
800 ip->ip_dst = inp->inp_faddr;
803 th->th_sport = inp->inp_lport;
804 th->th_dport = inp->inp_fport;
812 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
816 * Create template to be used to send tcp packets on a connection.
817 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
818 * use for this function is in keepalives, which use tcp_respond.
821 tcpip_maketemplate(struct inpcb *inp)
825 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
828 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
833 * Send a single message to the TCP at address specified by
834 * the given TCP/IP header. If m == NULL, then we make a copy
835 * of the tcpiphdr at th and send directly to the addressed host.
836 * This is used to force keep alive messages out using the TCP
837 * template for a connection. If flags are given then we send
838 * a message back to the TCP which originated the segment th,
839 * and discard the mbuf containing it and any other attached mbufs.
841 * In any case the ack and sequence number of the transmitted
842 * segment are as specified by the parameters.
844 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
847 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
848 tcp_seq ack, tcp_seq seq, int flags)
861 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
864 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
871 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
872 INP_WLOCK_ASSERT(inp);
877 if (!(flags & TH_RST)) {
878 win = sbspace(&inp->inp_socket->so_rcv);
879 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
880 win = (long)TCP_MAXWIN << tp->rcv_scale;
884 m = m_gethdr(M_NOWAIT, MT_DATA);
888 m->m_data += max_linkhdr;
891 bcopy((caddr_t)ip6, mtod(m, caddr_t),
892 sizeof(struct ip6_hdr));
893 ip6 = mtod(m, struct ip6_hdr *);
894 nth = (struct tcphdr *)(ip6 + 1);
898 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
899 ip = mtod(m, struct ip *);
900 nth = (struct tcphdr *)(ip + 1);
902 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
907 * XXX MRT We inherrit the FIB, which is lucky.
911 m->m_data = (caddr_t)ipgen;
912 /* m_len is set later */
914 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
917 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
918 nth = (struct tcphdr *)(ip6 + 1);
922 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
923 nth = (struct tcphdr *)(ip + 1);
927 * this is usually a case when an extension header
928 * exists between the IPv6 header and the
931 nth->th_sport = th->th_sport;
932 nth->th_dport = th->th_dport;
934 xchg(nth->th_dport, nth->th_sport, uint16_t);
940 ip6->ip6_vfc = IPV6_VERSION;
941 ip6->ip6_nxt = IPPROTO_TCP;
942 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
943 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
946 #if defined(INET) && defined(INET6)
951 tlen += sizeof (struct tcpiphdr);
952 ip->ip_len = htons(tlen);
953 ip->ip_ttl = V_ip_defttl;
954 if (V_path_mtu_discovery)
955 ip->ip_off |= htons(IP_DF);
959 m->m_pkthdr.len = tlen;
960 m->m_pkthdr.rcvif = NULL;
964 * Packet is associated with a socket, so allow the
965 * label of the response to reflect the socket label.
967 INP_WLOCK_ASSERT(inp);
968 mac_inpcb_create_mbuf(inp, m);
971 * Packet is not associated with a socket, so possibly
972 * update the label in place.
974 mac_netinet_tcp_reply(m);
977 nth->th_seq = htonl(seq);
978 nth->th_ack = htonl(ack);
980 nth->th_off = sizeof (struct tcphdr) >> 2;
981 nth->th_flags = flags;
983 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
985 nth->th_win = htons((u_short)win);
988 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
991 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
992 nth->th_sum = in6_cksum_pseudo(ip6,
993 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
994 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
998 #if defined(INET6) && defined(INET)
1003 m->m_pkthdr.csum_flags = CSUM_TCP;
1004 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1005 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1009 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1010 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1012 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
1014 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
1017 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
1020 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
1022 #if defined(INET) && defined(INET6)
1026 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
1031 * Create a new TCP control block, making an
1032 * empty reassembly queue and hooking it to the argument
1033 * protocol control block. The `inp' parameter must have
1034 * come from the zone allocator set up in tcp_init().
1037 tcp_newtcpcb(struct inpcb *inp)
1039 struct tcpcb_mem *tm;
1042 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1045 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1050 /* Initialise cc_var struct for this tcpcb. */
1052 tp->ccv->type = IPPROTO_TCP;
1053 tp->ccv->ccvc.tcp = tp;
1054 rw_rlock(&tcp_function_lock);
1055 tp->t_fb = tcp_func_set_ptr;
1056 refcount_acquire(&tp->t_fb->tfb_refcnt);
1057 rw_runlock(&tcp_function_lock);
1058 if (tp->t_fb->tfb_tcp_fb_init) {
1059 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1062 * Use the current system default CC algorithm.
1065 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1066 CC_ALGO(tp) = CC_DEFAULT();
1069 if (CC_ALGO(tp)->cb_init != NULL)
1070 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1071 if (tp->t_fb->tfb_tcp_fb_fini)
1072 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1073 refcount_release(&tp->t_fb->tfb_refcnt);
1074 uma_zfree(V_tcpcb_zone, tm);
1079 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1080 if (tp->t_fb->tfb_tcp_fb_fini)
1081 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1082 refcount_release(&tp->t_fb->tfb_refcnt);
1083 uma_zfree(V_tcpcb_zone, tm);
1088 tp->t_vnet = inp->inp_vnet;
1090 tp->t_timers = &tm->tt;
1091 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
1094 isipv6 ? V_tcp_v6mssdflt :
1098 /* Set up our timeouts. */
1099 callout_init(&tp->t_timers->tt_rexmt, 1);
1100 callout_init(&tp->t_timers->tt_persist, 1);
1101 callout_init(&tp->t_timers->tt_keep, 1);
1102 callout_init(&tp->t_timers->tt_2msl, 1);
1103 callout_init(&tp->t_timers->tt_delack, 1);
1105 if (V_tcp_do_rfc1323)
1106 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1108 tp->t_flags |= TF_SACK_PERMIT;
1109 TAILQ_INIT(&tp->snd_holes);
1111 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1114 in_pcbref(inp); /* Reference for tcpcb */
1118 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1119 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1120 * reasonable initial retransmit time.
1122 tp->t_srtt = TCPTV_SRTTBASE;
1123 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1124 tp->t_rttmin = tcp_rexmit_min;
1125 tp->t_rxtcur = TCPTV_RTOBASE;
1126 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1127 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1128 tp->t_rcvtime = ticks;
1130 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1131 * because the socket may be bound to an IPv6 wildcard address,
1132 * which may match an IPv4-mapped IPv6 address.
1134 inp->inp_ip_ttl = V_ip_defttl;
1138 * Init the TCP PCAP queues.
1140 tcp_pcap_tcpcb_init(tp);
1142 return (tp); /* XXX */
1146 * Switch the congestion control algorithm back to NewReno for any active
1147 * control blocks using an algorithm which is about to go away.
1148 * This ensures the CC framework can allow the unload to proceed without leaving
1149 * any dangling pointers which would trigger a panic.
1150 * Returning non-zero would inform the CC framework that something went wrong
1151 * and it would be unsafe to allow the unload to proceed. However, there is no
1152 * way for this to occur with this implementation so we always return zero.
1155 tcp_ccalgounload(struct cc_algo *unload_algo)
1157 struct cc_algo *tmpalgo;
1160 VNET_ITERATOR_DECL(vnet_iter);
1163 * Check all active control blocks across all network stacks and change
1164 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1165 * requires cleanup code to be run, call it.
1168 VNET_FOREACH(vnet_iter) {
1169 CURVNET_SET(vnet_iter);
1170 INP_INFO_WLOCK(&V_tcbinfo);
1172 * New connections already part way through being initialised
1173 * with the CC algo we're removing will not race with this code
1174 * because the INP_INFO_WLOCK is held during initialisation. We
1175 * therefore don't enter the loop below until the connection
1176 * list has stabilised.
1178 LIST_FOREACH(inp, &V_tcb, inp_list) {
1180 /* Important to skip tcptw structs. */
1181 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1182 (tp = intotcpcb(inp)) != NULL) {
1184 * By holding INP_WLOCK here, we are assured
1185 * that the connection is not currently
1186 * executing inside the CC module's functions
1187 * i.e. it is safe to make the switch back to
1190 if (CC_ALGO(tp) == unload_algo) {
1191 tmpalgo = CC_ALGO(tp);
1192 /* NewReno does not require any init. */
1193 CC_ALGO(tp) = &newreno_cc_algo;
1194 if (tmpalgo->cb_destroy != NULL)
1195 tmpalgo->cb_destroy(tp->ccv);
1200 INP_INFO_WUNLOCK(&V_tcbinfo);
1203 VNET_LIST_RUNLOCK();
1209 * Drop a TCP connection, reporting
1210 * the specified error. If connection is synchronized,
1211 * then send a RST to peer.
1214 tcp_drop(struct tcpcb *tp, int errno)
1216 struct socket *so = tp->t_inpcb->inp_socket;
1218 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1219 INP_WLOCK_ASSERT(tp->t_inpcb);
1221 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1222 tcp_state_change(tp, TCPS_CLOSED);
1223 (void) tp->t_fb->tfb_tcp_output(tp);
1224 TCPSTAT_INC(tcps_drops);
1226 TCPSTAT_INC(tcps_conndrops);
1227 if (errno == ETIMEDOUT && tp->t_softerror)
1228 errno = tp->t_softerror;
1229 so->so_error = errno;
1230 return (tcp_close(tp));
1234 tcp_discardcb(struct tcpcb *tp)
1236 struct inpcb *inp = tp->t_inpcb;
1237 struct socket *so = inp->inp_socket;
1239 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1243 INP_WLOCK_ASSERT(inp);
1246 * Make sure that all of our timers are stopped before we delete the
1249 * If stopping a timer fails, we schedule a discard function in same
1250 * callout, and the last discard function called will take care of
1251 * deleting the tcpcb.
1253 tcp_timer_stop(tp, TT_REXMT);
1254 tcp_timer_stop(tp, TT_PERSIST);
1255 tcp_timer_stop(tp, TT_KEEP);
1256 tcp_timer_stop(tp, TT_2MSL);
1257 tcp_timer_stop(tp, TT_DELACK);
1258 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1259 /* Call the stop-all function of the methods */
1260 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1264 * If we got enough samples through the srtt filter,
1265 * save the rtt and rttvar in the routing entry.
1266 * 'Enough' is arbitrarily defined as 4 rtt samples.
1267 * 4 samples is enough for the srtt filter to converge
1268 * to within enough % of the correct value; fewer samples
1269 * and we could save a bogus rtt. The danger is not high
1270 * as tcp quickly recovers from everything.
1271 * XXX: Works very well but needs some more statistics!
1273 if (tp->t_rttupdated >= 4) {
1274 struct hc_metrics_lite metrics;
1277 bzero(&metrics, sizeof(metrics));
1279 * Update the ssthresh always when the conditions below
1280 * are satisfied. This gives us better new start value
1281 * for the congestion avoidance for new connections.
1282 * ssthresh is only set if packet loss occured on a session.
1284 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1285 * being torn down. Ideally this code would not use 'so'.
1287 ssthresh = tp->snd_ssthresh;
1288 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1290 * convert the limit from user data bytes to
1291 * packets then to packet data bytes.
1293 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1296 ssthresh *= (u_long)(tp->t_maxseg +
1298 (isipv6 ? sizeof (struct ip6_hdr) +
1299 sizeof (struct tcphdr) :
1301 sizeof (struct tcpiphdr)
1308 metrics.rmx_ssthresh = ssthresh;
1310 metrics.rmx_rtt = tp->t_srtt;
1311 metrics.rmx_rttvar = tp->t_rttvar;
1312 metrics.rmx_cwnd = tp->snd_cwnd;
1313 metrics.rmx_sendpipe = 0;
1314 metrics.rmx_recvpipe = 0;
1316 tcp_hc_update(&inp->inp_inc, &metrics);
1319 /* free the reassembly queue, if any */
1320 tcp_reass_flush(tp);
1323 /* Disconnect offload device, if any. */
1324 if (tp->t_flags & TF_TOE)
1325 tcp_offload_detach(tp);
1328 tcp_free_sackholes(tp);
1331 /* Free the TCP PCAP queues. */
1332 tcp_pcap_drain(&(tp->t_inpkts));
1333 tcp_pcap_drain(&(tp->t_outpkts));
1336 /* Allow the CC algorithm to clean up after itself. */
1337 if (CC_ALGO(tp)->cb_destroy != NULL)
1338 CC_ALGO(tp)->cb_destroy(tp->ccv);
1340 khelp_destroy_osd(tp->osd);
1343 inp->inp_ppcb = NULL;
1344 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1345 /* We own the last reference on tcpcb, let's free it. */
1346 if ((tp->t_fb->tfb_tcp_timers_left) &&
1347 (tp->t_fb->tfb_tcp_timers_left(tp))) {
1348 /* Some fb timers left running! */
1351 if (tp->t_fb->tfb_tcp_fb_fini)
1352 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1353 refcount_release(&tp->t_fb->tfb_refcnt);
1355 uma_zfree(V_tcpcb_zone, tp);
1356 released = in_pcbrele_wlocked(inp);
1357 KASSERT(!released, ("%s: inp %p should not have been released "
1358 "here", __func__, inp));
1363 tcp_timer_2msl_discard(void *xtp)
1366 tcp_timer_discard((struct tcpcb *)xtp, TT_2MSL);
1370 tcp_timer_keep_discard(void *xtp)
1373 tcp_timer_discard((struct tcpcb *)xtp, TT_KEEP);
1377 tcp_timer_persist_discard(void *xtp)
1380 tcp_timer_discard((struct tcpcb *)xtp, TT_PERSIST);
1384 tcp_timer_rexmt_discard(void *xtp)
1387 tcp_timer_discard((struct tcpcb *)xtp, TT_REXMT);
1391 tcp_timer_delack_discard(void *xtp)
1394 tcp_timer_discard((struct tcpcb *)xtp, TT_DELACK);
1398 tcp_timer_discard(struct tcpcb *tp, uint32_t timer_type)
1402 CURVNET_SET(tp->t_vnet);
1403 INP_INFO_RLOCK(&V_tcbinfo);
1405 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1408 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1409 ("%s: tcpcb has to be stopped here", __func__));
1410 KASSERT((tp->t_timers->tt_flags & timer_type) != 0,
1411 ("%s: discard callout should be running", __func__));
1412 tp->t_timers->tt_flags &= ~timer_type;
1413 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1414 /* We own the last reference on this tcpcb, let's free it. */
1415 if ((tp->t_fb->tfb_tcp_timers_left) &&
1416 (tp->t_fb->tfb_tcp_timers_left(tp))) {
1417 /* Some fb timers left running! */
1420 if (tp->t_fb->tfb_tcp_fb_fini)
1421 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1422 refcount_release(&tp->t_fb->tfb_refcnt);
1424 uma_zfree(V_tcpcb_zone, tp);
1425 if (in_pcbrele_wlocked(inp)) {
1426 INP_INFO_RUNLOCK(&V_tcbinfo);
1433 INP_INFO_RUNLOCK(&V_tcbinfo);
1438 * Attempt to close a TCP control block, marking it as dropped, and freeing
1439 * the socket if we hold the only reference.
1442 tcp_close(struct tcpcb *tp)
1444 struct inpcb *inp = tp->t_inpcb;
1447 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1448 INP_WLOCK_ASSERT(inp);
1451 if (tp->t_state == TCPS_LISTEN)
1452 tcp_offload_listen_stop(tp);
1456 * This releases the TFO pending counter resource for TFO listen
1457 * sockets as well as passively-created TFO sockets that transition
1458 * from SYN_RECEIVED to CLOSED.
1460 if (tp->t_tfo_pending) {
1461 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1462 tp->t_tfo_pending = NULL;
1466 TCPSTAT_INC(tcps_closed);
1467 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1468 so = inp->inp_socket;
1469 soisdisconnected(so);
1470 if (inp->inp_flags & INP_SOCKREF) {
1471 KASSERT(so->so_state & SS_PROTOREF,
1472 ("tcp_close: !SS_PROTOREF"));
1473 inp->inp_flags &= ~INP_SOCKREF;
1477 so->so_state &= ~SS_PROTOREF;
1487 VNET_ITERATOR_DECL(vnet_iter);
1492 VNET_LIST_RLOCK_NOSLEEP();
1493 VNET_FOREACH(vnet_iter) {
1494 CURVNET_SET(vnet_iter);
1499 * Walk the tcpbs, if existing, and flush the reassembly queue,
1500 * if there is one...
1501 * XXX: The "Net/3" implementation doesn't imply that the TCP
1502 * reassembly queue should be flushed, but in a situation
1503 * where we're really low on mbufs, this is potentially
1506 INP_INFO_WLOCK(&V_tcbinfo);
1507 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1508 if (inpb->inp_flags & INP_TIMEWAIT)
1511 if ((tcpb = intotcpcb(inpb)) != NULL) {
1512 tcp_reass_flush(tcpb);
1513 tcp_clean_sackreport(tcpb);
1517 INP_INFO_WUNLOCK(&V_tcbinfo);
1520 VNET_LIST_RUNLOCK_NOSLEEP();
1524 * Notify a tcp user of an asynchronous error;
1525 * store error as soft error, but wake up user
1526 * (for now, won't do anything until can select for soft error).
1528 * Do not wake up user since there currently is no mechanism for
1529 * reporting soft errors (yet - a kqueue filter may be added).
1531 static struct inpcb *
1532 tcp_notify(struct inpcb *inp, int error)
1536 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1537 INP_WLOCK_ASSERT(inp);
1539 if ((inp->inp_flags & INP_TIMEWAIT) ||
1540 (inp->inp_flags & INP_DROPPED))
1543 tp = intotcpcb(inp);
1544 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1547 * Ignore some errors if we are hooked up.
1548 * If connection hasn't completed, has retransmitted several times,
1549 * and receives a second error, give up now. This is better
1550 * than waiting a long time to establish a connection that
1551 * can never complete.
1553 if (tp->t_state == TCPS_ESTABLISHED &&
1554 (error == EHOSTUNREACH || error == ENETUNREACH ||
1555 error == EHOSTDOWN)) {
1557 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1559 tp = tcp_drop(tp, error);
1565 tp->t_softerror = error;
1569 wakeup( &so->so_timeo);
1576 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1578 int error, i, m, n, pcb_count;
1579 struct inpcb *inp, **inp_list;
1584 * The process of preparing the TCB list is too time-consuming and
1585 * resource-intensive to repeat twice on every request.
1587 if (req->oldptr == NULL) {
1588 n = V_tcbinfo.ipi_count + syncache_pcbcount();
1589 n += imax(n / 8, 10);
1590 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1594 if (req->newptr != NULL)
1598 * OK, now we're committed to doing something.
1600 INP_LIST_RLOCK(&V_tcbinfo);
1601 gencnt = V_tcbinfo.ipi_gencnt;
1602 n = V_tcbinfo.ipi_count;
1603 INP_LIST_RUNLOCK(&V_tcbinfo);
1605 m = syncache_pcbcount();
1607 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1608 + (n + m) * sizeof(struct xtcpcb));
1612 xig.xig_len = sizeof xig;
1613 xig.xig_count = n + m;
1614 xig.xig_gen = gencnt;
1615 xig.xig_sogen = so_gencnt;
1616 error = SYSCTL_OUT(req, &xig, sizeof xig);
1620 error = syncache_pcblist(req, m, &pcb_count);
1624 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1625 if (inp_list == NULL)
1628 INP_INFO_WLOCK(&V_tcbinfo);
1629 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1630 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1632 if (inp->inp_gencnt <= gencnt) {
1634 * XXX: This use of cr_cansee(), introduced with
1635 * TCP state changes, is not quite right, but for
1636 * now, better than nothing.
1638 if (inp->inp_flags & INP_TIMEWAIT) {
1639 if (intotw(inp) != NULL)
1640 error = cr_cansee(req->td->td_ucred,
1641 intotw(inp)->tw_cred);
1643 error = EINVAL; /* Skip this inp. */
1645 error = cr_canseeinpcb(req->td->td_ucred, inp);
1648 inp_list[i++] = inp;
1653 INP_INFO_WUNLOCK(&V_tcbinfo);
1657 for (i = 0; i < n; i++) {
1660 if (inp->inp_gencnt <= gencnt) {
1664 bzero(&xt, sizeof(xt));
1665 xt.xt_len = sizeof xt;
1666 /* XXX should avoid extra copy */
1667 bcopy(inp, &xt.xt_inp, sizeof *inp);
1668 inp_ppcb = inp->inp_ppcb;
1669 if (inp_ppcb == NULL)
1670 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1671 else if (inp->inp_flags & INP_TIMEWAIT) {
1672 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1673 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1675 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1676 if (xt.xt_tp.t_timers)
1677 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1679 if (inp->inp_socket != NULL)
1680 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1682 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1683 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1685 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1687 error = SYSCTL_OUT(req, &xt, sizeof xt);
1691 INP_INFO_RLOCK(&V_tcbinfo);
1692 for (i = 0; i < n; i++) {
1695 if (!in_pcbrele_rlocked(inp))
1698 INP_INFO_RUNLOCK(&V_tcbinfo);
1702 * Give the user an updated idea of our state.
1703 * If the generation differs from what we told
1704 * her before, she knows that something happened
1705 * while we were processing this request, and it
1706 * might be necessary to retry.
1708 INP_LIST_RLOCK(&V_tcbinfo);
1709 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1710 xig.xig_sogen = so_gencnt;
1711 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1712 INP_LIST_RUNLOCK(&V_tcbinfo);
1713 error = SYSCTL_OUT(req, &xig, sizeof xig);
1715 free(inp_list, M_TEMP);
1719 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1720 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1721 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1725 tcp_getcred(SYSCTL_HANDLER_ARGS)
1728 struct sockaddr_in addrs[2];
1732 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1735 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1738 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1739 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1741 if (inp->inp_socket == NULL)
1744 error = cr_canseeinpcb(req->td->td_ucred, inp);
1746 cru2x(inp->inp_cred, &xuc);
1751 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1755 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1756 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1757 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1762 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1765 struct sockaddr_in6 addrs[2];
1772 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1775 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1778 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1779 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1782 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1784 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1793 inp = in_pcblookup(&V_tcbinfo,
1794 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1796 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1797 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1800 inp = in6_pcblookup(&V_tcbinfo,
1801 &addrs[1].sin6_addr, addrs[1].sin6_port,
1802 &addrs[0].sin6_addr, addrs[0].sin6_port,
1803 INPLOOKUP_RLOCKPCB, NULL);
1805 if (inp->inp_socket == NULL)
1808 error = cr_canseeinpcb(req->td->td_ucred, inp);
1810 cru2x(inp->inp_cred, &xuc);
1815 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1819 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1820 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1821 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1827 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1829 struct ip *ip = vip;
1831 struct in_addr faddr;
1834 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1836 struct in_conninfo inc;
1837 tcp_seq icmp_tcp_seq;
1840 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1841 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1844 if (cmd == PRC_MSGSIZE)
1845 notify = tcp_mtudisc_notify;
1846 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1847 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1848 notify = tcp_drop_syn_sent;
1850 * Redirects don't need to be handled up here.
1852 else if (PRC_IS_REDIRECT(cmd))
1855 * Hostdead is ugly because it goes linearly through all PCBs.
1856 * XXX: We never get this from ICMP, otherwise it makes an
1857 * excellent DoS attack on machines with many connections.
1859 else if (cmd == PRC_HOSTDEAD)
1861 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1865 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1869 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1870 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1871 INP_INFO_RLOCK(&V_tcbinfo);
1872 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
1873 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1875 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1876 !(inp->inp_flags & INP_DROPPED) &&
1877 !(inp->inp_socket == NULL)) {
1878 icmp_tcp_seq = ntohl(th->th_seq);
1879 tp = intotcpcb(inp);
1880 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1881 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1882 if (cmd == PRC_MSGSIZE) {
1885 * If we got a needfrag set the MTU
1886 * in the route to the suggested new
1887 * value (if given) and then notify.
1889 mtu = ntohs(icp->icmp_nextmtu);
1891 * If no alternative MTU was
1892 * proposed, try the next smaller
1897 ntohs(ip->ip_len), 1);
1898 if (mtu < V_tcp_minmss +
1899 sizeof(struct tcpiphdr))
1900 mtu = V_tcp_minmss +
1901 sizeof(struct tcpiphdr);
1903 * Only process the offered MTU if it
1904 * is smaller than the current one.
1906 if (mtu < tp->t_maxseg +
1907 sizeof(struct tcpiphdr)) {
1908 bzero(&inc, sizeof(inc));
1909 inc.inc_faddr = faddr;
1911 inp->inp_inc.inc_fibnum;
1912 tcp_hc_updatemtu(&inc, mtu);
1913 tcp_mtudisc(inp, mtu);
1916 inp = (*notify)(inp,
1917 inetctlerrmap[cmd]);
1923 bzero(&inc, sizeof(inc));
1924 inc.inc_fport = th->th_dport;
1925 inc.inc_lport = th->th_sport;
1926 inc.inc_faddr = faddr;
1927 inc.inc_laddr = ip->ip_src;
1928 syncache_unreach(&inc, th);
1930 INP_INFO_RUNLOCK(&V_tcbinfo);
1936 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1939 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1940 struct ip6_hdr *ip6;
1942 struct ip6ctlparam *ip6cp = NULL;
1943 const struct sockaddr_in6 *sa6_src = NULL;
1945 struct tcp_portonly {
1950 if (sa->sa_family != AF_INET6 ||
1951 sa->sa_len != sizeof(struct sockaddr_in6))
1954 if (cmd == PRC_MSGSIZE)
1955 notify = tcp_mtudisc_notify;
1956 else if (!PRC_IS_REDIRECT(cmd) &&
1957 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1960 /* if the parameter is from icmp6, decode it. */
1962 ip6cp = (struct ip6ctlparam *)d;
1964 ip6 = ip6cp->ip6c_ip6;
1965 off = ip6cp->ip6c_off;
1966 sa6_src = ip6cp->ip6c_src;
1970 off = 0; /* fool gcc */
1975 struct in_conninfo inc;
1977 * XXX: We assume that when IPV6 is non NULL,
1978 * M and OFF are valid.
1981 /* check if we can safely examine src and dst ports */
1982 if (m->m_pkthdr.len < off + sizeof(*thp))
1985 bzero(&th, sizeof(th));
1986 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1988 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1989 (struct sockaddr *)ip6cp->ip6c_src,
1990 th.th_sport, cmd, NULL, notify);
1992 bzero(&inc, sizeof(inc));
1993 inc.inc_fport = th.th_dport;
1994 inc.inc_lport = th.th_sport;
1995 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1996 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1997 inc.inc_flags |= INC_ISIPV6;
1998 INP_INFO_RLOCK(&V_tcbinfo);
1999 syncache_unreach(&inc, &th);
2000 INP_INFO_RUNLOCK(&V_tcbinfo);
2002 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
2003 0, cmd, NULL, notify);
2009 * Following is where TCP initial sequence number generation occurs.
2011 * There are two places where we must use initial sequence numbers:
2012 * 1. In SYN-ACK packets.
2013 * 2. In SYN packets.
2015 * All ISNs for SYN-ACK packets are generated by the syncache. See
2016 * tcp_syncache.c for details.
2018 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2019 * depends on this property. In addition, these ISNs should be
2020 * unguessable so as to prevent connection hijacking. To satisfy
2021 * the requirements of this situation, the algorithm outlined in
2022 * RFC 1948 is used, with only small modifications.
2024 * Implementation details:
2026 * Time is based off the system timer, and is corrected so that it
2027 * increases by one megabyte per second. This allows for proper
2028 * recycling on high speed LANs while still leaving over an hour
2031 * As reading the *exact* system time is too expensive to be done
2032 * whenever setting up a TCP connection, we increment the time
2033 * offset in two ways. First, a small random positive increment
2034 * is added to isn_offset for each connection that is set up.
2035 * Second, the function tcp_isn_tick fires once per clock tick
2036 * and increments isn_offset as necessary so that sequence numbers
2037 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2038 * random positive increments serve only to ensure that the same
2039 * exact sequence number is never sent out twice (as could otherwise
2040 * happen when a port is recycled in less than the system tick
2043 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2044 * between seeding of isn_secret. This is normally set to zero,
2045 * as reseeding should not be necessary.
2047 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2048 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
2049 * general, this means holding an exclusive (write) lock.
2052 #define ISN_BYTES_PER_SECOND 1048576
2053 #define ISN_STATIC_INCREMENT 4096
2054 #define ISN_RANDOM_INCREMENT (4096 - 1)
2056 static VNET_DEFINE(u_char, isn_secret[32]);
2057 static VNET_DEFINE(int, isn_last);
2058 static VNET_DEFINE(int, isn_last_reseed);
2059 static VNET_DEFINE(u_int32_t, isn_offset);
2060 static VNET_DEFINE(u_int32_t, isn_offset_old);
2062 #define V_isn_secret VNET(isn_secret)
2063 #define V_isn_last VNET(isn_last)
2064 #define V_isn_last_reseed VNET(isn_last_reseed)
2065 #define V_isn_offset VNET(isn_offset)
2066 #define V_isn_offset_old VNET(isn_offset_old)
2069 tcp_new_isn(struct tcpcb *tp)
2072 u_int32_t md5_buffer[4];
2074 u_int32_t projected_offset;
2076 INP_WLOCK_ASSERT(tp->t_inpcb);
2079 /* Seed if this is the first use, reseed if requested. */
2080 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2081 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2083 read_random(&V_isn_secret, sizeof(V_isn_secret));
2084 V_isn_last_reseed = ticks;
2087 /* Compute the md5 hash and return the ISN. */
2089 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2090 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2092 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2093 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2094 sizeof(struct in6_addr));
2095 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2096 sizeof(struct in6_addr));
2100 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2101 sizeof(struct in_addr));
2102 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2103 sizeof(struct in_addr));
2105 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2106 MD5Final((u_char *) &md5_buffer, &isn_ctx);
2107 new_isn = (tcp_seq) md5_buffer[0];
2108 V_isn_offset += ISN_STATIC_INCREMENT +
2109 (arc4random() & ISN_RANDOM_INCREMENT);
2110 if (ticks != V_isn_last) {
2111 projected_offset = V_isn_offset_old +
2112 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2113 if (SEQ_GT(projected_offset, V_isn_offset))
2114 V_isn_offset = projected_offset;
2115 V_isn_offset_old = V_isn_offset;
2118 new_isn += V_isn_offset;
2124 * When a specific ICMP unreachable message is received and the
2125 * connection state is SYN-SENT, drop the connection. This behavior
2126 * is controlled by the icmp_may_rst sysctl.
2129 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2133 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2134 INP_WLOCK_ASSERT(inp);
2136 if ((inp->inp_flags & INP_TIMEWAIT) ||
2137 (inp->inp_flags & INP_DROPPED))
2140 tp = intotcpcb(inp);
2141 if (tp->t_state != TCPS_SYN_SENT)
2144 tp = tcp_drop(tp, errno);
2152 * When `need fragmentation' ICMP is received, update our idea of the MSS
2153 * based on the new value. Also nudge TCP to send something, since we
2154 * know the packet we just sent was dropped.
2155 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2157 static struct inpcb *
2158 tcp_mtudisc_notify(struct inpcb *inp, int error)
2161 tcp_mtudisc(inp, -1);
2166 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2171 INP_WLOCK_ASSERT(inp);
2172 if ((inp->inp_flags & INP_TIMEWAIT) ||
2173 (inp->inp_flags & INP_DROPPED))
2176 tp = intotcpcb(inp);
2177 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2179 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2181 so = inp->inp_socket;
2182 SOCKBUF_LOCK(&so->so_snd);
2183 /* If the mss is larger than the socket buffer, decrease the mss. */
2184 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2185 tp->t_maxseg = so->so_snd.sb_hiwat;
2186 SOCKBUF_UNLOCK(&so->so_snd);
2188 TCPSTAT_INC(tcps_mturesent);
2190 tp->snd_nxt = tp->snd_una;
2191 tcp_free_sackholes(tp);
2192 tp->snd_recover = tp->snd_max;
2193 if (tp->t_flags & TF_SACK_PERMIT)
2194 EXIT_FASTRECOVERY(tp->t_flags);
2195 tp->t_fb->tfb_tcp_output(tp);
2200 * Look-up the routing entry to the peer of this inpcb. If no route
2201 * is found and it cannot be allocated, then return 0. This routine
2202 * is called by TCP routines that access the rmx structure and by
2203 * tcp_mss_update to get the peer/interface MTU.
2206 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2209 struct sockaddr_in *dst;
2213 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2215 bzero(&sro, sizeof(sro));
2216 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2217 dst = (struct sockaddr_in *)&sro.ro_dst;
2218 dst->sin_family = AF_INET;
2219 dst->sin_len = sizeof(*dst);
2220 dst->sin_addr = inc->inc_faddr;
2221 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
2223 if (sro.ro_rt != NULL) {
2224 ifp = sro.ro_rt->rt_ifp;
2225 if (sro.ro_rt->rt_mtu == 0)
2226 maxmtu = ifp->if_mtu;
2228 maxmtu = min(sro.ro_rt->rt_mtu, ifp->if_mtu);
2230 /* Report additional interface capabilities. */
2232 if (ifp->if_capenable & IFCAP_TSO4 &&
2233 ifp->if_hwassist & CSUM_TSO) {
2234 cap->ifcap |= CSUM_TSO;
2235 cap->tsomax = ifp->if_hw_tsomax;
2236 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2237 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2248 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2250 struct route_in6 sro6;
2254 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2256 bzero(&sro6, sizeof(sro6));
2257 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2258 sro6.ro_dst.sin6_family = AF_INET6;
2259 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
2260 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
2261 in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
2263 if (sro6.ro_rt != NULL) {
2264 ifp = sro6.ro_rt->rt_ifp;
2265 if (sro6.ro_rt->rt_mtu == 0)
2266 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
2268 maxmtu = min(sro6.ro_rt->rt_mtu,
2269 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
2271 /* Report additional interface capabilities. */
2273 if (ifp->if_capenable & IFCAP_TSO6 &&
2274 ifp->if_hwassist & CSUM_TSO) {
2275 cap->ifcap |= CSUM_TSO;
2276 cap->tsomax = ifp->if_hw_tsomax;
2277 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2278 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2289 * Calculate effective SMSS per RFC5681 definition for a given TCP
2290 * connection at its current state, taking into account SACK and etc.
2293 tcp_maxseg(const struct tcpcb *tp)
2297 if (tp->t_flags & TF_NOOPT)
2298 return (tp->t_maxseg);
2301 * Here we have a simplified code from tcp_addoptions(),
2302 * without a proper loop, and having most of paddings hardcoded.
2303 * We might make mistakes with padding here in some edge cases,
2304 * but this is harmless, since result of tcp_maxseg() is used
2305 * only in cwnd and ssthresh estimations.
2307 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2308 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2309 if (tp->t_flags & TF_RCVD_TSTMP)
2310 optlen = TCPOLEN_TSTAMP_APPA;
2313 #ifdef TCP_SIGNATURE
2314 if (tp->t_flags & TF_SIGNATURE)
2315 optlen += PAD(TCPOLEN_SIGNATURE);
2317 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2318 optlen += TCPOLEN_SACKHDR;
2319 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2320 optlen = PAD(optlen);
2323 if (tp->t_flags & TF_REQ_TSTMP)
2324 optlen = TCPOLEN_TSTAMP_APPA;
2326 optlen = PAD(TCPOLEN_MAXSEG);
2327 if (tp->t_flags & TF_REQ_SCALE)
2328 optlen += PAD(TCPOLEN_WINDOW);
2329 #ifdef TCP_SIGNATURE
2330 if (tp->t_flags & TF_SIGNATURE)
2331 optlen += PAD(TCPOLEN_SIGNATURE);
2333 if (tp->t_flags & TF_SACK_PERMIT)
2334 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2337 optlen = min(optlen, TCP_MAXOLEN);
2338 return (tp->t_maxseg - optlen);
2342 /* compute ESP/AH header size for TCP, including outer IP header. */
2344 ipsec_hdrsiz_tcp(struct tcpcb *tp)
2351 struct ip6_hdr *ip6;
2355 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) ||
2356 (!key_havesp(IPSEC_DIR_OUTBOUND)))
2358 m = m_gethdr(M_NOWAIT, MT_DATA);
2363 if ((inp->inp_vflag & INP_IPV6) != 0) {
2364 ip6 = mtod(m, struct ip6_hdr *);
2365 th = (struct tcphdr *)(ip6 + 1);
2366 m->m_pkthdr.len = m->m_len =
2367 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
2368 tcpip_fillheaders(inp, ip6, th);
2369 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2373 ip = mtod(m, struct ip *);
2374 th = (struct tcphdr *)(ip + 1);
2375 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
2376 tcpip_fillheaders(inp, ip, th);
2377 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2385 #ifdef TCP_SIGNATURE
2387 * Callback function invoked by m_apply() to digest TCP segment data
2388 * contained within an mbuf chain.
2391 tcp_signature_apply(void *fstate, void *data, u_int len)
2394 MD5Update(fstate, (u_char *)data, len);
2399 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2400 * search with the destination IP address, and a 'magic SPI' to be
2401 * determined by the application. This is hardcoded elsewhere to 1179
2404 tcp_get_sav(struct mbuf *m, u_int direction)
2406 union sockaddr_union dst;
2407 struct secasvar *sav;
2410 struct ip6_hdr *ip6;
2411 char ip6buf[INET6_ADDRSTRLEN];
2414 /* Extract the destination from the IP header in the mbuf. */
2415 bzero(&dst, sizeof(union sockaddr_union));
2416 ip = mtod(m, struct ip *);
2418 ip6 = NULL; /* Make the compiler happy. */
2423 dst.sa.sa_len = sizeof(struct sockaddr_in);
2424 dst.sa.sa_family = AF_INET;
2425 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2426 ip->ip_src : ip->ip_dst;
2430 case (IPV6_VERSION >> 4):
2431 ip6 = mtod(m, struct ip6_hdr *);
2432 dst.sa.sa_len = sizeof(struct sockaddr_in6);
2433 dst.sa.sa_family = AF_INET6;
2434 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2435 ip6->ip6_src : ip6->ip6_dst;
2444 /* Look up an SADB entry which matches the address of the peer. */
2445 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2447 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2448 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2450 (ip->ip_v == (IPV6_VERSION >> 4)) ?
2451 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2460 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2463 * m pointer to head of mbuf chain
2464 * len length of TCP segment data, excluding options
2465 * optlen length of TCP segment options
2466 * buf pointer to storage for computed MD5 digest
2467 * sav pointer to security assosiation
2469 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2470 * When called from tcp_input(), we can be sure that th_sum has been
2471 * zeroed out and verified already.
2473 * Releases reference to SADB key before return.
2475 * Return 0 if successful, otherwise return -1.
2479 tcp_signature_do_compute(struct mbuf *m, int len, int optlen,
2480 u_char *buf, struct secasvar *sav)
2483 struct ippseudo ippseudo;
2489 struct ipovly *ipovly;
2493 struct ip6_hdr *ip6;
2494 struct in6_addr in6;
2500 KASSERT(m != NULL, ("NULL mbuf chain"));
2501 KASSERT(buf != NULL, ("NULL signature pointer"));
2503 /* Extract the destination from the IP header in the mbuf. */
2504 ip = mtod(m, struct ip *);
2506 ip6 = NULL; /* Make the compiler happy. */
2511 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2513 * XXX The ippseudo header MUST be digested in network byte order,
2514 * or else we'll fail the regression test. Assume all fields we've
2515 * been doing arithmetic on have been in host byte order.
2516 * XXX One cannot depend on ipovly->ih_len here. When called from
2517 * tcp_output(), the underlying ip_len member has not yet been set.
2522 ipovly = (struct ipovly *)ip;
2523 ippseudo.ippseudo_src = ipovly->ih_src;
2524 ippseudo.ippseudo_dst = ipovly->ih_dst;
2525 ippseudo.ippseudo_pad = 0;
2526 ippseudo.ippseudo_p = IPPROTO_TCP;
2527 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2529 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2531 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2532 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2537 * RFC 2385, 2.0 Proposal
2538 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2539 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2540 * extended next header value (to form 32 bits), and 32-bit segment
2542 * Note: Upper-Layer Packet Length comes before Next Header.
2544 case (IPV6_VERSION >> 4):
2546 in6_clearscope(&in6);
2547 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2549 in6_clearscope(&in6);
2550 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2551 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2552 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2554 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2555 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2556 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2558 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2560 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2561 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2573 * Step 2: Update MD5 hash with TCP header, excluding options.
2574 * The TCP checksum must be set to zero.
2576 savecsum = th->th_sum;
2578 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2579 th->th_sum = savecsum;
2582 * Step 3: Update MD5 hash with TCP segment data.
2583 * Use m_apply() to avoid an early m_pullup().
2586 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2589 * Step 4: Update MD5 hash with shared secret.
2591 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2592 MD5Final(buf, &ctx);
2594 key_sa_recordxfer(sav, m);
2600 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2602 * Return 0 if successful, otherwise return -1.
2605 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2606 u_char *buf, u_int direction)
2608 struct secasvar *sav;
2610 if ((sav = tcp_get_sav(m, direction)) == NULL)
2613 return (tcp_signature_do_compute(m, len, optlen, buf, sav));
2617 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2620 * m pointer to head of mbuf chain
2621 * len length of TCP segment data, excluding options
2622 * optlen length of TCP segment options
2623 * buf pointer to storage for computed MD5 digest
2624 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2626 * Return 1 if successful, otherwise return 0.
2629 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2630 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2632 char tmpdigest[TCP_SIGLEN];
2634 if (tcp_sig_checksigs == 0)
2636 if ((tcpbflag & TF_SIGNATURE) == 0) {
2637 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2640 * If this socket is not expecting signature but
2641 * the segment contains signature just fail.
2643 TCPSTAT_INC(tcps_sig_err_sigopt);
2644 TCPSTAT_INC(tcps_sig_rcvbadsig);
2648 /* Signature is not expected, and not present in segment. */
2653 * If this socket is expecting signature but the segment does not
2654 * contain any just fail.
2656 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2657 TCPSTAT_INC(tcps_sig_err_nosigopt);
2658 TCPSTAT_INC(tcps_sig_rcvbadsig);
2661 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2662 IPSEC_DIR_INBOUND) == -1) {
2663 TCPSTAT_INC(tcps_sig_err_buildsig);
2664 TCPSTAT_INC(tcps_sig_rcvbadsig);
2668 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2669 TCPSTAT_INC(tcps_sig_rcvbadsig);
2672 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2675 #endif /* TCP_SIGNATURE */
2678 sysctl_drop(SYSCTL_HANDLER_ARGS)
2680 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2681 struct sockaddr_storage addrs[2];
2685 struct sockaddr_in *fin, *lin;
2687 struct sockaddr_in6 *fin6, *lin6;
2698 if (req->oldptr != NULL || req->oldlen != 0)
2700 if (req->newptr == NULL)
2702 if (req->newlen < sizeof(addrs))
2704 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2708 switch (addrs[0].ss_family) {
2711 fin6 = (struct sockaddr_in6 *)&addrs[0];
2712 lin6 = (struct sockaddr_in6 *)&addrs[1];
2713 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2714 lin6->sin6_len != sizeof(struct sockaddr_in6))
2716 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2717 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2719 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2720 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2721 fin = (struct sockaddr_in *)&addrs[0];
2722 lin = (struct sockaddr_in *)&addrs[1];
2725 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2728 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2735 fin = (struct sockaddr_in *)&addrs[0];
2736 lin = (struct sockaddr_in *)&addrs[1];
2737 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2738 lin->sin_len != sizeof(struct sockaddr_in))
2745 INP_INFO_RLOCK(&V_tcbinfo);
2746 switch (addrs[0].ss_family) {
2749 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2750 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2751 INPLOOKUP_WLOCKPCB, NULL);
2756 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2757 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2762 if (inp->inp_flags & INP_TIMEWAIT) {
2764 * XXXRW: There currently exists a state where an
2765 * inpcb is present, but its timewait state has been
2766 * discarded. For now, don't allow dropping of this
2774 } else if (!(inp->inp_flags & INP_DROPPED) &&
2775 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2776 tp = intotcpcb(inp);
2777 tp = tcp_drop(tp, ECONNABORTED);
2784 INP_INFO_RUNLOCK(&V_tcbinfo);
2788 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2789 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2790 0, sysctl_drop, "", "Drop TCP connection");
2793 * Generate a standardized TCP log line for use throughout the
2794 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2795 * allow use in the interrupt context.
2797 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2798 * NB: The function may return NULL if memory allocation failed.
2800 * Due to header inclusion and ordering limitations the struct ip
2801 * and ip6_hdr pointers have to be passed as void pointers.
2804 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2808 /* Is logging enabled? */
2809 if (tcp_log_in_vain == 0)
2812 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2816 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2820 /* Is logging enabled? */
2821 if (tcp_log_debug == 0)
2824 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2828 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2835 const struct ip6_hdr *ip6;
2837 ip6 = (const struct ip6_hdr *)ip6hdr;
2839 ip = (struct ip *)ip4hdr;
2842 * The log line looks like this:
2843 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2845 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2846 sizeof(PRINT_TH_FLAGS) + 1 +
2848 2 * INET6_ADDRSTRLEN;
2850 2 * INET_ADDRSTRLEN;
2853 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2857 strcat(s, "TCP: [");
2860 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2861 inet_ntoa_r(inc->inc_faddr, sp);
2863 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2865 inet_ntoa_r(inc->inc_laddr, sp);
2867 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2870 ip6_sprintf(sp, &inc->inc6_faddr);
2872 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2874 ip6_sprintf(sp, &inc->inc6_laddr);
2876 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2877 } else if (ip6 && th) {
2878 ip6_sprintf(sp, &ip6->ip6_src);
2880 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2882 ip6_sprintf(sp, &ip6->ip6_dst);
2884 sprintf(sp, "]:%i", ntohs(th->th_dport));
2887 } else if (ip && th) {
2888 inet_ntoa_r(ip->ip_src, sp);
2890 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2892 inet_ntoa_r(ip->ip_dst, sp);
2894 sprintf(sp, "]:%i", ntohs(th->th_dport));
2902 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2903 if (*(s + size - 1) != '\0')
2904 panic("%s: string too long", __func__);
2909 * A subroutine which makes it easy to track TCP state changes with DTrace.
2910 * This function shouldn't be called for t_state initializations that don't
2911 * correspond to actual TCP state transitions.
2914 tcp_state_change(struct tcpcb *tp, int newstate)
2916 #if defined(KDTRACE_HOOKS)
2917 int pstate = tp->t_state;
2920 tp->t_state = newstate;
2921 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);