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/hhook.h>
45 #include <sys/kernel.h>
46 #include <sys/khelp.h>
47 #include <sys/sysctl.h>
49 #include <sys/malloc.h>
50 #include <sys/refcount.h>
53 #include <sys/domain.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/protosw.h>
61 #include <sys/random.h>
65 #include <net/route.h>
67 #include <net/if_var.h>
70 #include <netinet/cc.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_fsm.h>
91 #include <netinet/tcp_seq.h>
92 #include <netinet/tcp_timer.h>
93 #include <netinet/tcp_var.h>
94 #include <netinet/tcp_syncache.h>
96 #include <netinet6/tcp6_var.h>
98 #include <netinet/tcpip.h>
100 #include <netinet/tcp_pcap.h>
103 #include <netinet/tcp_debug.h>
106 #include <netinet6/ip6protosw.h>
109 #include <netinet/tcp_offload.h>
113 #include <netipsec/ipsec.h>
114 #include <netipsec/xform.h>
116 #include <netipsec/ipsec6.h>
118 #include <netipsec/key.h>
119 #include <sys/syslog.h>
122 #include <machine/in_cksum.h>
125 #include <security/mac/mac_framework.h>
127 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
129 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
132 struct rwlock tcp_function_lock;
135 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
140 error = sysctl_handle_int(oidp, &new, 0, req);
141 if (error == 0 && req->newptr) {
142 if (new < TCP_MINMSS)
150 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
151 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
152 &sysctl_net_inet_tcp_mss_check, "I",
153 "Default TCP Maximum Segment Size");
157 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
161 new = V_tcp_v6mssdflt;
162 error = sysctl_handle_int(oidp, &new, 0, req);
163 if (error == 0 && req->newptr) {
164 if (new < TCP_MINMSS)
167 V_tcp_v6mssdflt = new;
172 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
173 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
174 &sysctl_net_inet_tcp_mss_v6_check, "I",
175 "Default TCP Maximum Segment Size for IPv6");
179 * Minimum MSS we accept and use. This prevents DoS attacks where
180 * we are forced to a ridiculous low MSS like 20 and send hundreds
181 * of packets instead of one. The effect scales with the available
182 * bandwidth and quickly saturates the CPU and network interface
183 * with packet generation and sending. Set to zero to disable MINMSS
184 * checking. This setting prevents us from sending too small packets.
186 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
188 &VNET_NAME(tcp_minmss), 0,
189 "Minimum TCP Maximum Segment Size");
191 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
192 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
193 &VNET_NAME(tcp_do_rfc1323), 0,
194 "Enable rfc1323 (high performance TCP) extensions");
196 static int tcp_log_debug = 0;
197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
198 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
200 static int tcp_tcbhashsize;
201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
202 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
204 static int do_tcpdrain = 1;
205 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
206 "Enable tcp_drain routine for extra help when low on mbufs");
208 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
209 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
211 static VNET_DEFINE(int, icmp_may_rst) = 1;
212 #define V_icmp_may_rst VNET(icmp_may_rst)
213 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
214 &VNET_NAME(icmp_may_rst), 0,
215 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
217 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
218 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
219 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
220 &VNET_NAME(tcp_isn_reseed_interval), 0,
221 "Seconds between reseeding of ISN secret");
223 static int tcp_soreceive_stream;
224 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
225 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
228 static int tcp_sig_checksigs = 1;
229 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
230 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
233 VNET_DEFINE(uma_zone_t, sack_hole_zone);
234 #define V_sack_hole_zone VNET(sack_hole_zone)
236 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
238 static struct inpcb *tcp_notify(struct inpcb *, int);
239 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
240 static void tcp_mtudisc(struct inpcb *, int);
241 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
242 void *ip4hdr, const void *ip6hdr);
243 static void tcp_timer_discard(struct tcpcb *, uint32_t);
246 static struct tcp_function_block tcp_def_funcblk = {
250 tcp_default_ctloutput,
262 struct tcp_funchead t_functions;
263 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
265 static struct tcp_function_block *
266 find_tcp_functions_locked(struct tcp_function_set *fs)
268 struct tcp_function *f;
269 struct tcp_function_block *blk=NULL;
271 TAILQ_FOREACH(f, &t_functions, tf_next) {
272 if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) {
280 static struct tcp_function_block *
281 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
283 struct tcp_function_block *rblk=NULL;
284 struct tcp_function *f;
286 TAILQ_FOREACH(f, &t_functions, tf_next) {
287 if (f->tf_fb == blk) {
298 struct tcp_function_block *
299 find_and_ref_tcp_functions(struct tcp_function_set *fs)
301 struct tcp_function_block *blk;
303 rw_rlock(&tcp_function_lock);
304 blk = find_tcp_functions_locked(fs);
306 refcount_acquire(&blk->tfb_refcnt);
307 rw_runlock(&tcp_function_lock);
311 struct tcp_function_block *
312 find_and_ref_tcp_fb(struct tcp_function_block *blk)
314 struct tcp_function_block *rblk;
316 rw_rlock(&tcp_function_lock);
317 rblk = find_tcp_fb_locked(blk, NULL);
319 refcount_acquire(&rblk->tfb_refcnt);
320 rw_runlock(&tcp_function_lock);
326 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
329 struct tcp_function_set fs;
330 struct tcp_function_block *blk;
332 memset(&fs, 0, sizeof(fs));
333 rw_rlock(&tcp_function_lock);
334 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
337 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
338 fs.pcbcnt = blk->tfb_refcnt;
340 rw_runlock(&tcp_function_lock);
341 error = sysctl_handle_string(oidp, fs.function_set_name,
342 sizeof(fs.function_set_name), req);
344 /* Check for error or no change */
345 if (error != 0 || req->newptr == NULL)
348 rw_wlock(&tcp_function_lock);
349 blk = find_tcp_functions_locked(&fs);
351 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
355 tcp_func_set_ptr = blk;
357 rw_wunlock(&tcp_function_lock);
361 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
362 CTLTYPE_STRING | CTLFLAG_RW,
363 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
364 "Set/get the default TCP functions");
367 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
369 int error, cnt, linesz;
370 struct tcp_function *f;
375 rw_rlock(&tcp_function_lock);
376 TAILQ_FOREACH(f, &t_functions, tf_next) {
379 rw_runlock(&tcp_function_lock);
381 bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1;
382 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
387 linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count");
392 rw_rlock(&tcp_function_lock);
393 TAILQ_FOREACH(f, &t_functions, tf_next) {
394 linesz = snprintf(cp, bufsz, "%-32s%c %u\n",
395 f->tf_fb->tfb_tcp_block_name,
396 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
397 f->tf_fb->tfb_refcnt);
398 if (linesz >= bufsz) {
406 rw_runlock(&tcp_function_lock);
408 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
409 free(buffer, M_TEMP);
413 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
414 CTLTYPE_STRING|CTLFLAG_RD,
415 NULL, 0, sysctl_net_inet_list_available, "A",
416 "list available TCP Function sets");
419 * Target size of TCP PCB hash tables. Must be a power of two.
421 * Note that this can be overridden by the kernel environment
422 * variable net.inet.tcp.tcbhashsize
425 #define TCBHASHSIZE 0
430 * Callouts should be moved into struct tcp directly. They are currently
431 * separate because the tcpcb structure is exported to userland for sysctl
432 * parsing purposes, which do not know about callouts.
441 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
442 #define V_tcpcb_zone VNET(tcpcb_zone)
444 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
445 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
447 static struct mtx isn_mtx;
449 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
450 #define ISN_LOCK() mtx_lock(&isn_mtx)
451 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
454 * TCP initialization.
457 tcp_zone_change(void *tag)
460 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
461 uma_zone_set_max(V_tcpcb_zone, maxsockets);
462 tcp_tw_zone_change();
466 tcp_inpcb_init(void *mem, int size, int flags)
468 struct inpcb *inp = mem;
470 INP_LOCK_INIT(inp, "inp", "tcpinp");
475 * Take a value and get the next power of 2 that doesn't overflow.
476 * Used to size the tcp_inpcb hash buckets.
479 maketcp_hashsize(int size)
485 * get the next power of 2 higher than maxsockets.
487 hashsize = 1 << fls(size);
488 /* catch overflow, and just go one power of 2 smaller */
489 if (hashsize < size) {
490 hashsize = 1 << (fls(size) - 1);
496 register_tcp_functions(struct tcp_function_block *blk, int wait)
498 struct tcp_function_block *lblk;
499 struct tcp_function *n;
500 struct tcp_function_set fs;
502 if ((blk->tfb_tcp_output == NULL) ||
503 (blk->tfb_tcp_do_segment == NULL) ||
504 (blk->tfb_tcp_ctloutput == NULL) ||
505 (strlen(blk->tfb_tcp_block_name) == 0)) {
507 * These functions are required and you
512 if (blk->tfb_tcp_timer_stop_all ||
513 blk->tfb_tcp_timers_left ||
514 blk->tfb_tcp_timer_activate ||
515 blk->tfb_tcp_timer_active ||
516 blk->tfb_tcp_timer_stop) {
518 * If you define one timer function you
519 * must have them all.
521 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
522 (blk->tfb_tcp_timers_left == NULL) ||
523 (blk->tfb_tcp_timer_activate == NULL) ||
524 (blk->tfb_tcp_timer_active == NULL) ||
525 (blk->tfb_tcp_timer_stop == NULL)) {
529 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
534 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
535 rw_wlock(&tcp_function_lock);
536 lblk = find_tcp_functions_locked(&fs);
538 /* Duplicate name space not allowed */
539 rw_wunlock(&tcp_function_lock);
540 free(n, M_TCPFUNCTIONS);
543 refcount_init(&blk->tfb_refcnt, 0);
545 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
546 rw_wunlock(&tcp_function_lock);
551 deregister_tcp_functions(struct tcp_function_block *blk)
553 struct tcp_function_block *lblk;
554 struct tcp_function *f;
557 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
558 /* You can't un-register the default */
561 rw_wlock(&tcp_function_lock);
562 if (blk == tcp_func_set_ptr) {
563 /* You can't free the current default */
564 rw_wunlock(&tcp_function_lock);
567 if (blk->tfb_refcnt) {
568 /* Still tcb attached, mark it. */
569 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
570 rw_wunlock(&tcp_function_lock);
573 lblk = find_tcp_fb_locked(blk, &f);
576 TAILQ_REMOVE(&t_functions, f, tf_next);
578 free(f, M_TCPFUNCTIONS);
581 rw_wunlock(&tcp_function_lock);
588 const char *tcbhash_tuneable;
591 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
593 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
594 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
595 printf("%s: WARNING: unable to register helper hook\n", __func__);
596 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
597 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
598 printf("%s: WARNING: unable to register helper hook\n", __func__);
599 hashsize = TCBHASHSIZE;
600 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
603 * Auto tune the hash size based on maxsockets.
604 * A perfect hash would have a 1:1 mapping
605 * (hashsize = maxsockets) however it's been
606 * suggested that O(2) average is better.
608 hashsize = maketcp_hashsize(maxsockets / 4);
610 * Our historical default is 512,
611 * do not autotune lower than this.
615 if (bootverbose && IS_DEFAULT_VNET(curvnet))
616 printf("%s: %s auto tuned to %d\n", __func__,
617 tcbhash_tuneable, hashsize);
620 * We require a hashsize to be a power of two.
621 * Previously if it was not a power of two we would just reset it
622 * back to 512, which could be a nasty surprise if you did not notice
624 * Instead what we do is clip it to the closest power of two lower
625 * than the specified hash value.
627 if (!powerof2(hashsize)) {
628 int oldhashsize = hashsize;
630 hashsize = maketcp_hashsize(hashsize);
631 /* prevent absurdly low value */
634 printf("%s: WARNING: TCB hash size not a power of 2, "
635 "clipped from %d to %d.\n", __func__, oldhashsize,
638 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
639 "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
640 IPI_HASHFIELDS_4TUPLE);
643 * These have to be type stable for the benefit of the timers.
645 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
646 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
647 uma_zone_set_max(V_tcpcb_zone, maxsockets);
648 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
654 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
655 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
656 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
658 /* Skip initialization of globals for non-default instances. */
659 if (!IS_DEFAULT_VNET(curvnet))
662 tcp_reass_global_init();
664 /* XXX virtualize those bellow? */
665 tcp_delacktime = TCPTV_DELACK;
666 tcp_keepinit = TCPTV_KEEP_INIT;
667 tcp_keepidle = TCPTV_KEEP_IDLE;
668 tcp_keepintvl = TCPTV_KEEPINTVL;
669 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
671 tcp_rexmit_min = TCPTV_MIN;
672 if (tcp_rexmit_min < 1)
674 tcp_rexmit_slop = TCPTV_CPU_VAR;
675 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
676 tcp_tcbhashsize = hashsize;
677 /* Setup the tcp function block list */
678 TAILQ_INIT(&t_functions);
679 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
680 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
682 if (tcp_soreceive_stream) {
684 tcp_usrreqs.pru_soreceive = soreceive_stream;
687 tcp6_usrreqs.pru_soreceive = soreceive_stream;
692 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
694 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
696 if (max_protohdr < TCP_MINPROTOHDR)
697 max_protohdr = TCP_MINPROTOHDR;
698 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
700 #undef TCP_MINPROTOHDR
703 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
704 SHUTDOWN_PRI_DEFAULT);
705 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
706 EVENTHANDLER_PRI_ANY);
723 tcp_fastopen_destroy();
728 in_pcbinfo_destroy(&V_tcbinfo);
729 uma_zdestroy(V_sack_hole_zone);
730 uma_zdestroy(V_tcpcb_zone);
732 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
734 printf("%s: WARNING: unable to deregister helper hook "
735 "type=%d, id=%d: error %d returned\n", __func__,
736 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
738 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
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_OUT, error);
754 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
755 * tcp_template used to store this data in mbufs, but we now recopy it out
756 * of the tcpcb each time to conserve mbufs.
759 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
761 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
763 INP_WLOCK_ASSERT(inp);
766 if ((inp->inp_vflag & INP_IPV6) != 0) {
769 ip6 = (struct ip6_hdr *)ip_ptr;
770 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
771 (inp->inp_flow & IPV6_FLOWINFO_MASK);
772 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
773 (IPV6_VERSION & IPV6_VERSION_MASK);
774 ip6->ip6_nxt = IPPROTO_TCP;
775 ip6->ip6_plen = htons(sizeof(struct tcphdr));
776 ip6->ip6_src = inp->in6p_laddr;
777 ip6->ip6_dst = inp->in6p_faddr;
780 #if defined(INET6) && defined(INET)
787 ip = (struct ip *)ip_ptr;
788 ip->ip_v = IPVERSION;
790 ip->ip_tos = inp->inp_ip_tos;
794 ip->ip_ttl = inp->inp_ip_ttl;
796 ip->ip_p = IPPROTO_TCP;
797 ip->ip_src = inp->inp_laddr;
798 ip->ip_dst = inp->inp_faddr;
801 th->th_sport = inp->inp_lport;
802 th->th_dport = inp->inp_fport;
810 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
814 * Create template to be used to send tcp packets on a connection.
815 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
816 * use for this function is in keepalives, which use tcp_respond.
819 tcpip_maketemplate(struct inpcb *inp)
823 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
826 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
831 * Send a single message to the TCP at address specified by
832 * the given TCP/IP header. If m == NULL, then we make a copy
833 * of the tcpiphdr at th and send directly to the addressed host.
834 * This is used to force keep alive messages out using the TCP
835 * template for a connection. If flags are given then we send
836 * a message back to the TCP which originated the segment th,
837 * and discard the mbuf containing it and any other attached mbufs.
839 * In any case the ack and sequence number of the transmitted
840 * segment are as specified by the parameters.
842 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
845 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
846 tcp_seq ack, tcp_seq seq, int flags)
859 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
862 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
869 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
870 INP_WLOCK_ASSERT(inp);
875 if (!(flags & TH_RST)) {
876 win = sbspace(&inp->inp_socket->so_rcv);
877 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
878 win = (long)TCP_MAXWIN << tp->rcv_scale;
882 m = m_gethdr(M_NOWAIT, MT_DATA);
886 m->m_data += max_linkhdr;
889 bcopy((caddr_t)ip6, mtod(m, caddr_t),
890 sizeof(struct ip6_hdr));
891 ip6 = mtod(m, struct ip6_hdr *);
892 nth = (struct tcphdr *)(ip6 + 1);
896 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
897 ip = mtod(m, struct ip *);
898 nth = (struct tcphdr *)(ip + 1);
900 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
905 * XXX MRT We inherrit the FIB, which is lucky.
909 m->m_data = (caddr_t)ipgen;
910 /* m_len is set later */
912 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
915 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
916 nth = (struct tcphdr *)(ip6 + 1);
920 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
921 nth = (struct tcphdr *)(ip + 1);
925 * this is usually a case when an extension header
926 * exists between the IPv6 header and the
929 nth->th_sport = th->th_sport;
930 nth->th_dport = th->th_dport;
932 xchg(nth->th_dport, nth->th_sport, uint16_t);
938 ip6->ip6_vfc = IPV6_VERSION;
939 ip6->ip6_nxt = IPPROTO_TCP;
940 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
941 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
944 #if defined(INET) && defined(INET6)
949 tlen += sizeof (struct tcpiphdr);
950 ip->ip_len = htons(tlen);
951 ip->ip_ttl = V_ip_defttl;
952 if (V_path_mtu_discovery)
953 ip->ip_off |= htons(IP_DF);
957 m->m_pkthdr.len = tlen;
958 m->m_pkthdr.rcvif = NULL;
962 * Packet is associated with a socket, so allow the
963 * label of the response to reflect the socket label.
965 INP_WLOCK_ASSERT(inp);
966 mac_inpcb_create_mbuf(inp, m);
969 * Packet is not associated with a socket, so possibly
970 * update the label in place.
972 mac_netinet_tcp_reply(m);
975 nth->th_seq = htonl(seq);
976 nth->th_ack = htonl(ack);
978 nth->th_off = sizeof (struct tcphdr) >> 2;
979 nth->th_flags = flags;
981 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
983 nth->th_win = htons((u_short)win);
986 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
989 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
990 nth->th_sum = in6_cksum_pseudo(ip6,
991 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
992 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
996 #if defined(INET6) && defined(INET)
1001 m->m_pkthdr.csum_flags = CSUM_TCP;
1002 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1003 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1007 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1008 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1010 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
1012 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
1015 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
1018 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
1020 #if defined(INET) && defined(INET6)
1024 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
1029 * Create a new TCP control block, making an
1030 * empty reassembly queue and hooking it to the argument
1031 * protocol control block. The `inp' parameter must have
1032 * come from the zone allocator set up in tcp_init().
1035 tcp_newtcpcb(struct inpcb *inp)
1037 struct tcpcb_mem *tm;
1040 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1043 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1048 /* Initialise cc_var struct for this tcpcb. */
1050 tp->ccv->type = IPPROTO_TCP;
1051 tp->ccv->ccvc.tcp = tp;
1052 rw_rlock(&tcp_function_lock);
1053 tp->t_fb = tcp_func_set_ptr;
1054 refcount_acquire(&tp->t_fb->tfb_refcnt);
1055 rw_runlock(&tcp_function_lock);
1056 if (tp->t_fb->tfb_tcp_fb_init) {
1057 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1060 * Use the current system default CC algorithm.
1063 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1064 CC_ALGO(tp) = CC_DEFAULT();
1067 if (CC_ALGO(tp)->cb_init != NULL)
1068 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1069 if (tp->t_fb->tfb_tcp_fb_fini)
1070 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1071 refcount_release(&tp->t_fb->tfb_refcnt);
1072 uma_zfree(V_tcpcb_zone, tm);
1077 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1078 if (tp->t_fb->tfb_tcp_fb_fini)
1079 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1080 refcount_release(&tp->t_fb->tfb_refcnt);
1081 uma_zfree(V_tcpcb_zone, tm);
1086 tp->t_vnet = inp->inp_vnet;
1088 tp->t_timers = &tm->tt;
1089 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
1092 isipv6 ? V_tcp_v6mssdflt :
1096 /* Set up our timeouts. */
1097 callout_init(&tp->t_timers->tt_rexmt, 1);
1098 callout_init(&tp->t_timers->tt_persist, 1);
1099 callout_init(&tp->t_timers->tt_keep, 1);
1100 callout_init(&tp->t_timers->tt_2msl, 1);
1101 callout_init(&tp->t_timers->tt_delack, 1);
1103 if (V_tcp_do_rfc1323)
1104 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1106 tp->t_flags |= TF_SACK_PERMIT;
1107 TAILQ_INIT(&tp->snd_holes);
1109 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1112 in_pcbref(inp); /* Reference for tcpcb */
1116 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1117 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1118 * reasonable initial retransmit time.
1120 tp->t_srtt = TCPTV_SRTTBASE;
1121 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1122 tp->t_rttmin = tcp_rexmit_min;
1123 tp->t_rxtcur = TCPTV_RTOBASE;
1124 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1125 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1126 tp->t_rcvtime = ticks;
1128 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1129 * because the socket may be bound to an IPv6 wildcard address,
1130 * which may match an IPv4-mapped IPv6 address.
1132 inp->inp_ip_ttl = V_ip_defttl;
1136 * Init the TCP PCAP queues.
1138 tcp_pcap_tcpcb_init(tp);
1140 return (tp); /* XXX */
1144 * Switch the congestion control algorithm back to NewReno for any active
1145 * control blocks using an algorithm which is about to go away.
1146 * This ensures the CC framework can allow the unload to proceed without leaving
1147 * any dangling pointers which would trigger a panic.
1148 * Returning non-zero would inform the CC framework that something went wrong
1149 * and it would be unsafe to allow the unload to proceed. However, there is no
1150 * way for this to occur with this implementation so we always return zero.
1153 tcp_ccalgounload(struct cc_algo *unload_algo)
1155 struct cc_algo *tmpalgo;
1158 VNET_ITERATOR_DECL(vnet_iter);
1161 * Check all active control blocks across all network stacks and change
1162 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1163 * requires cleanup code to be run, call it.
1166 VNET_FOREACH(vnet_iter) {
1167 CURVNET_SET(vnet_iter);
1168 INP_INFO_WLOCK(&V_tcbinfo);
1170 * New connections already part way through being initialised
1171 * with the CC algo we're removing will not race with this code
1172 * because the INP_INFO_WLOCK is held during initialisation. We
1173 * therefore don't enter the loop below until the connection
1174 * list has stabilised.
1176 LIST_FOREACH(inp, &V_tcb, inp_list) {
1178 /* Important to skip tcptw structs. */
1179 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1180 (tp = intotcpcb(inp)) != NULL) {
1182 * By holding INP_WLOCK here, we are assured
1183 * that the connection is not currently
1184 * executing inside the CC module's functions
1185 * i.e. it is safe to make the switch back to
1188 if (CC_ALGO(tp) == unload_algo) {
1189 tmpalgo = CC_ALGO(tp);
1190 /* NewReno does not require any init. */
1191 CC_ALGO(tp) = &newreno_cc_algo;
1192 if (tmpalgo->cb_destroy != NULL)
1193 tmpalgo->cb_destroy(tp->ccv);
1198 INP_INFO_WUNLOCK(&V_tcbinfo);
1201 VNET_LIST_RUNLOCK();
1207 * Drop a TCP connection, reporting
1208 * the specified error. If connection is synchronized,
1209 * then send a RST to peer.
1212 tcp_drop(struct tcpcb *tp, int errno)
1214 struct socket *so = tp->t_inpcb->inp_socket;
1216 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1217 INP_WLOCK_ASSERT(tp->t_inpcb);
1219 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1220 tcp_state_change(tp, TCPS_CLOSED);
1221 (void) tp->t_fb->tfb_tcp_output(tp);
1222 TCPSTAT_INC(tcps_drops);
1224 TCPSTAT_INC(tcps_conndrops);
1225 if (errno == ETIMEDOUT && tp->t_softerror)
1226 errno = tp->t_softerror;
1227 so->so_error = errno;
1228 return (tcp_close(tp));
1232 tcp_discardcb(struct tcpcb *tp)
1234 struct inpcb *inp = tp->t_inpcb;
1235 struct socket *so = inp->inp_socket;
1237 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1241 INP_WLOCK_ASSERT(inp);
1244 * Make sure that all of our timers are stopped before we delete the
1247 * If stopping a timer fails, we schedule a discard function in same
1248 * callout, and the last discard function called will take care of
1249 * deleting the tcpcb.
1251 tcp_timer_stop(tp, TT_REXMT);
1252 tcp_timer_stop(tp, TT_PERSIST);
1253 tcp_timer_stop(tp, TT_KEEP);
1254 tcp_timer_stop(tp, TT_2MSL);
1255 tcp_timer_stop(tp, TT_DELACK);
1256 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1257 /* Call the stop-all function of the methods */
1258 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1262 * If we got enough samples through the srtt filter,
1263 * save the rtt and rttvar in the routing entry.
1264 * 'Enough' is arbitrarily defined as 4 rtt samples.
1265 * 4 samples is enough for the srtt filter to converge
1266 * to within enough % of the correct value; fewer samples
1267 * and we could save a bogus rtt. The danger is not high
1268 * as tcp quickly recovers from everything.
1269 * XXX: Works very well but needs some more statistics!
1271 if (tp->t_rttupdated >= 4) {
1272 struct hc_metrics_lite metrics;
1275 bzero(&metrics, sizeof(metrics));
1277 * Update the ssthresh always when the conditions below
1278 * are satisfied. This gives us better new start value
1279 * for the congestion avoidance for new connections.
1280 * ssthresh is only set if packet loss occured on a session.
1282 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1283 * being torn down. Ideally this code would not use 'so'.
1285 ssthresh = tp->snd_ssthresh;
1286 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1288 * convert the limit from user data bytes to
1289 * packets then to packet data bytes.
1291 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1294 ssthresh *= (u_long)(tp->t_maxseg +
1296 (isipv6 ? sizeof (struct ip6_hdr) +
1297 sizeof (struct tcphdr) :
1299 sizeof (struct tcpiphdr)
1306 metrics.rmx_ssthresh = ssthresh;
1308 metrics.rmx_rtt = tp->t_srtt;
1309 metrics.rmx_rttvar = tp->t_rttvar;
1310 metrics.rmx_cwnd = tp->snd_cwnd;
1311 metrics.rmx_sendpipe = 0;
1312 metrics.rmx_recvpipe = 0;
1314 tcp_hc_update(&inp->inp_inc, &metrics);
1317 /* free the reassembly queue, if any */
1318 tcp_reass_flush(tp);
1321 /* Disconnect offload device, if any. */
1322 if (tp->t_flags & TF_TOE)
1323 tcp_offload_detach(tp);
1326 tcp_free_sackholes(tp);
1329 /* Free the TCP PCAP queues. */
1330 tcp_pcap_drain(&(tp->t_inpkts));
1331 tcp_pcap_drain(&(tp->t_outpkts));
1334 /* Allow the CC algorithm to clean up after itself. */
1335 if (CC_ALGO(tp)->cb_destroy != NULL)
1336 CC_ALGO(tp)->cb_destroy(tp->ccv);
1338 khelp_destroy_osd(tp->osd);
1341 inp->inp_ppcb = NULL;
1342 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1343 /* We own the last reference on tcpcb, let's free it. */
1344 if ((tp->t_fb->tfb_tcp_timers_left) &&
1345 (tp->t_fb->tfb_tcp_timers_left(tp))) {
1346 /* Some fb timers left running! */
1349 if (tp->t_fb->tfb_tcp_fb_fini)
1350 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1351 refcount_release(&tp->t_fb->tfb_refcnt);
1353 uma_zfree(V_tcpcb_zone, tp);
1354 released = in_pcbrele_wlocked(inp);
1355 KASSERT(!released, ("%s: inp %p should not have been released "
1356 "here", __func__, inp));
1361 tcp_timer_2msl_discard(void *xtp)
1364 tcp_timer_discard((struct tcpcb *)xtp, TT_2MSL);
1368 tcp_timer_keep_discard(void *xtp)
1371 tcp_timer_discard((struct tcpcb *)xtp, TT_KEEP);
1375 tcp_timer_persist_discard(void *xtp)
1378 tcp_timer_discard((struct tcpcb *)xtp, TT_PERSIST);
1382 tcp_timer_rexmt_discard(void *xtp)
1385 tcp_timer_discard((struct tcpcb *)xtp, TT_REXMT);
1389 tcp_timer_delack_discard(void *xtp)
1392 tcp_timer_discard((struct tcpcb *)xtp, TT_DELACK);
1396 tcp_timer_discard(struct tcpcb *tp, uint32_t timer_type)
1400 CURVNET_SET(tp->t_vnet);
1401 INP_INFO_RLOCK(&V_tcbinfo);
1403 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1406 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1407 ("%s: tcpcb has to be stopped here", __func__));
1408 KASSERT((tp->t_timers->tt_flags & timer_type) != 0,
1409 ("%s: discard callout should be running", __func__));
1410 tp->t_timers->tt_flags &= ~timer_type;
1411 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1412 /* We own the last reference on this tcpcb, let's free it. */
1413 if ((tp->t_fb->tfb_tcp_timers_left) &&
1414 (tp->t_fb->tfb_tcp_timers_left(tp))) {
1415 /* Some fb timers left running! */
1418 if (tp->t_fb->tfb_tcp_fb_fini)
1419 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1420 refcount_release(&tp->t_fb->tfb_refcnt);
1422 uma_zfree(V_tcpcb_zone, tp);
1423 if (in_pcbrele_wlocked(inp)) {
1424 INP_INFO_RUNLOCK(&V_tcbinfo);
1431 INP_INFO_RUNLOCK(&V_tcbinfo);
1436 * Attempt to close a TCP control block, marking it as dropped, and freeing
1437 * the socket if we hold the only reference.
1440 tcp_close(struct tcpcb *tp)
1442 struct inpcb *inp = tp->t_inpcb;
1445 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1446 INP_WLOCK_ASSERT(inp);
1449 if (tp->t_state == TCPS_LISTEN)
1450 tcp_offload_listen_stop(tp);
1454 * This releases the TFO pending counter resource for TFO listen
1455 * sockets as well as passively-created TFO sockets that transition
1456 * from SYN_RECEIVED to CLOSED.
1458 if (tp->t_tfo_pending) {
1459 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1460 tp->t_tfo_pending = NULL;
1464 TCPSTAT_INC(tcps_closed);
1465 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1466 so = inp->inp_socket;
1467 soisdisconnected(so);
1468 if (inp->inp_flags & INP_SOCKREF) {
1469 KASSERT(so->so_state & SS_PROTOREF,
1470 ("tcp_close: !SS_PROTOREF"));
1471 inp->inp_flags &= ~INP_SOCKREF;
1475 so->so_state &= ~SS_PROTOREF;
1485 VNET_ITERATOR_DECL(vnet_iter);
1490 VNET_LIST_RLOCK_NOSLEEP();
1491 VNET_FOREACH(vnet_iter) {
1492 CURVNET_SET(vnet_iter);
1497 * Walk the tcpbs, if existing, and flush the reassembly queue,
1498 * if there is one...
1499 * XXX: The "Net/3" implementation doesn't imply that the TCP
1500 * reassembly queue should be flushed, but in a situation
1501 * where we're really low on mbufs, this is potentially
1504 INP_INFO_WLOCK(&V_tcbinfo);
1505 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1506 if (inpb->inp_flags & INP_TIMEWAIT)
1509 if ((tcpb = intotcpcb(inpb)) != NULL) {
1510 tcp_reass_flush(tcpb);
1511 tcp_clean_sackreport(tcpb);
1515 INP_INFO_WUNLOCK(&V_tcbinfo);
1518 VNET_LIST_RUNLOCK_NOSLEEP();
1522 * Notify a tcp user of an asynchronous error;
1523 * store error as soft error, but wake up user
1524 * (for now, won't do anything until can select for soft error).
1526 * Do not wake up user since there currently is no mechanism for
1527 * reporting soft errors (yet - a kqueue filter may be added).
1529 static struct inpcb *
1530 tcp_notify(struct inpcb *inp, int error)
1534 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1535 INP_WLOCK_ASSERT(inp);
1537 if ((inp->inp_flags & INP_TIMEWAIT) ||
1538 (inp->inp_flags & INP_DROPPED))
1541 tp = intotcpcb(inp);
1542 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1545 * Ignore some errors if we are hooked up.
1546 * If connection hasn't completed, has retransmitted several times,
1547 * and receives a second error, give up now. This is better
1548 * than waiting a long time to establish a connection that
1549 * can never complete.
1551 if (tp->t_state == TCPS_ESTABLISHED &&
1552 (error == EHOSTUNREACH || error == ENETUNREACH ||
1553 error == EHOSTDOWN)) {
1555 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1557 tp = tcp_drop(tp, error);
1563 tp->t_softerror = error;
1567 wakeup( &so->so_timeo);
1574 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1576 int error, i, m, n, pcb_count;
1577 struct inpcb *inp, **inp_list;
1582 * The process of preparing the TCB list is too time-consuming and
1583 * resource-intensive to repeat twice on every request.
1585 if (req->oldptr == NULL) {
1586 n = V_tcbinfo.ipi_count + syncache_pcbcount();
1587 n += imax(n / 8, 10);
1588 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1592 if (req->newptr != NULL)
1596 * OK, now we're committed to doing something.
1598 INP_LIST_RLOCK(&V_tcbinfo);
1599 gencnt = V_tcbinfo.ipi_gencnt;
1600 n = V_tcbinfo.ipi_count;
1601 INP_LIST_RUNLOCK(&V_tcbinfo);
1603 m = syncache_pcbcount();
1605 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1606 + (n + m) * sizeof(struct xtcpcb));
1610 xig.xig_len = sizeof xig;
1611 xig.xig_count = n + m;
1612 xig.xig_gen = gencnt;
1613 xig.xig_sogen = so_gencnt;
1614 error = SYSCTL_OUT(req, &xig, sizeof xig);
1618 error = syncache_pcblist(req, m, &pcb_count);
1622 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1623 if (inp_list == NULL)
1626 INP_INFO_WLOCK(&V_tcbinfo);
1627 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1628 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1630 if (inp->inp_gencnt <= gencnt) {
1632 * XXX: This use of cr_cansee(), introduced with
1633 * TCP state changes, is not quite right, but for
1634 * now, better than nothing.
1636 if (inp->inp_flags & INP_TIMEWAIT) {
1637 if (intotw(inp) != NULL)
1638 error = cr_cansee(req->td->td_ucred,
1639 intotw(inp)->tw_cred);
1641 error = EINVAL; /* Skip this inp. */
1643 error = cr_canseeinpcb(req->td->td_ucred, inp);
1646 inp_list[i++] = inp;
1651 INP_INFO_WUNLOCK(&V_tcbinfo);
1655 for (i = 0; i < n; i++) {
1658 if (inp->inp_gencnt <= gencnt) {
1662 bzero(&xt, sizeof(xt));
1663 xt.xt_len = sizeof xt;
1664 /* XXX should avoid extra copy */
1665 bcopy(inp, &xt.xt_inp, sizeof *inp);
1666 inp_ppcb = inp->inp_ppcb;
1667 if (inp_ppcb == NULL)
1668 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1669 else if (inp->inp_flags & INP_TIMEWAIT) {
1670 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1671 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1673 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1674 if (xt.xt_tp.t_timers)
1675 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1677 if (inp->inp_socket != NULL)
1678 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1680 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1681 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1683 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1685 error = SYSCTL_OUT(req, &xt, sizeof xt);
1689 INP_INFO_RLOCK(&V_tcbinfo);
1690 for (i = 0; i < n; i++) {
1693 if (!in_pcbrele_rlocked(inp))
1696 INP_INFO_RUNLOCK(&V_tcbinfo);
1700 * Give the user an updated idea of our state.
1701 * If the generation differs from what we told
1702 * her before, she knows that something happened
1703 * while we were processing this request, and it
1704 * might be necessary to retry.
1706 INP_LIST_RLOCK(&V_tcbinfo);
1707 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1708 xig.xig_sogen = so_gencnt;
1709 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1710 INP_LIST_RUNLOCK(&V_tcbinfo);
1711 error = SYSCTL_OUT(req, &xig, sizeof xig);
1713 free(inp_list, M_TEMP);
1717 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1718 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1719 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1723 tcp_getcred(SYSCTL_HANDLER_ARGS)
1726 struct sockaddr_in addrs[2];
1730 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1733 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1736 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1737 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1739 if (inp->inp_socket == NULL)
1742 error = cr_canseeinpcb(req->td->td_ucred, inp);
1744 cru2x(inp->inp_cred, &xuc);
1749 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1753 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1754 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1755 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1760 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1763 struct sockaddr_in6 addrs[2];
1770 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1773 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1776 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1777 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1780 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1782 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1791 inp = in_pcblookup(&V_tcbinfo,
1792 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1794 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1795 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1798 inp = in6_pcblookup(&V_tcbinfo,
1799 &addrs[1].sin6_addr, addrs[1].sin6_port,
1800 &addrs[0].sin6_addr, addrs[0].sin6_port,
1801 INPLOOKUP_RLOCKPCB, NULL);
1803 if (inp->inp_socket == NULL)
1806 error = cr_canseeinpcb(req->td->td_ucred, inp);
1808 cru2x(inp->inp_cred, &xuc);
1813 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1817 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1818 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1819 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1825 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1827 struct ip *ip = vip;
1829 struct in_addr faddr;
1832 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1834 struct in_conninfo inc;
1835 tcp_seq icmp_tcp_seq;
1838 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1839 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1842 if (cmd == PRC_MSGSIZE)
1843 notify = tcp_mtudisc_notify;
1844 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1845 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1846 notify = tcp_drop_syn_sent;
1848 * Redirects don't need to be handled up here.
1850 else if (PRC_IS_REDIRECT(cmd))
1853 * Hostdead is ugly because it goes linearly through all PCBs.
1854 * XXX: We never get this from ICMP, otherwise it makes an
1855 * excellent DoS attack on machines with many connections.
1857 else if (cmd == PRC_HOSTDEAD)
1859 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1863 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1867 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1868 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1869 INP_INFO_RLOCK(&V_tcbinfo);
1870 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
1871 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1873 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1874 !(inp->inp_flags & INP_DROPPED) &&
1875 !(inp->inp_socket == NULL)) {
1876 icmp_tcp_seq = ntohl(th->th_seq);
1877 tp = intotcpcb(inp);
1878 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1879 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1880 if (cmd == PRC_MSGSIZE) {
1883 * If we got a needfrag set the MTU
1884 * in the route to the suggested new
1885 * value (if given) and then notify.
1887 mtu = ntohs(icp->icmp_nextmtu);
1889 * If no alternative MTU was
1890 * proposed, try the next smaller
1895 ntohs(ip->ip_len), 1);
1896 if (mtu < V_tcp_minmss +
1897 sizeof(struct tcpiphdr))
1898 mtu = V_tcp_minmss +
1899 sizeof(struct tcpiphdr);
1901 * Only process the offered MTU if it
1902 * is smaller than the current one.
1904 if (mtu < tp->t_maxseg +
1905 sizeof(struct tcpiphdr)) {
1906 bzero(&inc, sizeof(inc));
1907 inc.inc_faddr = faddr;
1909 inp->inp_inc.inc_fibnum;
1910 tcp_hc_updatemtu(&inc, mtu);
1911 tcp_mtudisc(inp, mtu);
1914 inp = (*notify)(inp,
1915 inetctlerrmap[cmd]);
1921 bzero(&inc, sizeof(inc));
1922 inc.inc_fport = th->th_dport;
1923 inc.inc_lport = th->th_sport;
1924 inc.inc_faddr = faddr;
1925 inc.inc_laddr = ip->ip_src;
1926 syncache_unreach(&inc, th);
1928 INP_INFO_RUNLOCK(&V_tcbinfo);
1934 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1937 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1938 struct ip6_hdr *ip6;
1940 struct ip6ctlparam *ip6cp = NULL;
1941 const struct sockaddr_in6 *sa6_src = NULL;
1943 struct tcp_portonly {
1948 if (sa->sa_family != AF_INET6 ||
1949 sa->sa_len != sizeof(struct sockaddr_in6))
1952 if (cmd == PRC_MSGSIZE)
1953 notify = tcp_mtudisc_notify;
1954 else if (!PRC_IS_REDIRECT(cmd) &&
1955 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1958 /* if the parameter is from icmp6, decode it. */
1960 ip6cp = (struct ip6ctlparam *)d;
1962 ip6 = ip6cp->ip6c_ip6;
1963 off = ip6cp->ip6c_off;
1964 sa6_src = ip6cp->ip6c_src;
1968 off = 0; /* fool gcc */
1973 struct in_conninfo inc;
1975 * XXX: We assume that when IPV6 is non NULL,
1976 * M and OFF are valid.
1979 /* check if we can safely examine src and dst ports */
1980 if (m->m_pkthdr.len < off + sizeof(*thp))
1983 bzero(&th, sizeof(th));
1984 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1986 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1987 (struct sockaddr *)ip6cp->ip6c_src,
1988 th.th_sport, cmd, NULL, notify);
1990 bzero(&inc, sizeof(inc));
1991 inc.inc_fport = th.th_dport;
1992 inc.inc_lport = th.th_sport;
1993 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1994 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1995 inc.inc_flags |= INC_ISIPV6;
1996 INP_INFO_RLOCK(&V_tcbinfo);
1997 syncache_unreach(&inc, &th);
1998 INP_INFO_RUNLOCK(&V_tcbinfo);
2000 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
2001 0, cmd, NULL, notify);
2007 * Following is where TCP initial sequence number generation occurs.
2009 * There are two places where we must use initial sequence numbers:
2010 * 1. In SYN-ACK packets.
2011 * 2. In SYN packets.
2013 * All ISNs for SYN-ACK packets are generated by the syncache. See
2014 * tcp_syncache.c for details.
2016 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2017 * depends on this property. In addition, these ISNs should be
2018 * unguessable so as to prevent connection hijacking. To satisfy
2019 * the requirements of this situation, the algorithm outlined in
2020 * RFC 1948 is used, with only small modifications.
2022 * Implementation details:
2024 * Time is based off the system timer, and is corrected so that it
2025 * increases by one megabyte per second. This allows for proper
2026 * recycling on high speed LANs while still leaving over an hour
2029 * As reading the *exact* system time is too expensive to be done
2030 * whenever setting up a TCP connection, we increment the time
2031 * offset in two ways. First, a small random positive increment
2032 * is added to isn_offset for each connection that is set up.
2033 * Second, the function tcp_isn_tick fires once per clock tick
2034 * and increments isn_offset as necessary so that sequence numbers
2035 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2036 * random positive increments serve only to ensure that the same
2037 * exact sequence number is never sent out twice (as could otherwise
2038 * happen when a port is recycled in less than the system tick
2041 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2042 * between seeding of isn_secret. This is normally set to zero,
2043 * as reseeding should not be necessary.
2045 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2046 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
2047 * general, this means holding an exclusive (write) lock.
2050 #define ISN_BYTES_PER_SECOND 1048576
2051 #define ISN_STATIC_INCREMENT 4096
2052 #define ISN_RANDOM_INCREMENT (4096 - 1)
2054 static VNET_DEFINE(u_char, isn_secret[32]);
2055 static VNET_DEFINE(int, isn_last);
2056 static VNET_DEFINE(int, isn_last_reseed);
2057 static VNET_DEFINE(u_int32_t, isn_offset);
2058 static VNET_DEFINE(u_int32_t, isn_offset_old);
2060 #define V_isn_secret VNET(isn_secret)
2061 #define V_isn_last VNET(isn_last)
2062 #define V_isn_last_reseed VNET(isn_last_reseed)
2063 #define V_isn_offset VNET(isn_offset)
2064 #define V_isn_offset_old VNET(isn_offset_old)
2067 tcp_new_isn(struct tcpcb *tp)
2070 u_int32_t md5_buffer[4];
2072 u_int32_t projected_offset;
2074 INP_WLOCK_ASSERT(tp->t_inpcb);
2077 /* Seed if this is the first use, reseed if requested. */
2078 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2079 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2081 read_random(&V_isn_secret, sizeof(V_isn_secret));
2082 V_isn_last_reseed = ticks;
2085 /* Compute the md5 hash and return the ISN. */
2087 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2088 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2090 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2091 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2092 sizeof(struct in6_addr));
2093 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2094 sizeof(struct in6_addr));
2098 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2099 sizeof(struct in_addr));
2100 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2101 sizeof(struct in_addr));
2103 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2104 MD5Final((u_char *) &md5_buffer, &isn_ctx);
2105 new_isn = (tcp_seq) md5_buffer[0];
2106 V_isn_offset += ISN_STATIC_INCREMENT +
2107 (arc4random() & ISN_RANDOM_INCREMENT);
2108 if (ticks != V_isn_last) {
2109 projected_offset = V_isn_offset_old +
2110 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2111 if (SEQ_GT(projected_offset, V_isn_offset))
2112 V_isn_offset = projected_offset;
2113 V_isn_offset_old = V_isn_offset;
2116 new_isn += V_isn_offset;
2122 * When a specific ICMP unreachable message is received and the
2123 * connection state is SYN-SENT, drop the connection. This behavior
2124 * is controlled by the icmp_may_rst sysctl.
2127 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2131 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2132 INP_WLOCK_ASSERT(inp);
2134 if ((inp->inp_flags & INP_TIMEWAIT) ||
2135 (inp->inp_flags & INP_DROPPED))
2138 tp = intotcpcb(inp);
2139 if (tp->t_state != TCPS_SYN_SENT)
2142 tp = tcp_drop(tp, errno);
2150 * When `need fragmentation' ICMP is received, update our idea of the MSS
2151 * based on the new value. Also nudge TCP to send something, since we
2152 * know the packet we just sent was dropped.
2153 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2155 static struct inpcb *
2156 tcp_mtudisc_notify(struct inpcb *inp, int error)
2159 tcp_mtudisc(inp, -1);
2164 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2169 INP_WLOCK_ASSERT(inp);
2170 if ((inp->inp_flags & INP_TIMEWAIT) ||
2171 (inp->inp_flags & INP_DROPPED))
2174 tp = intotcpcb(inp);
2175 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2177 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2179 so = inp->inp_socket;
2180 SOCKBUF_LOCK(&so->so_snd);
2181 /* If the mss is larger than the socket buffer, decrease the mss. */
2182 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2183 tp->t_maxseg = so->so_snd.sb_hiwat;
2184 SOCKBUF_UNLOCK(&so->so_snd);
2186 TCPSTAT_INC(tcps_mturesent);
2188 tp->snd_nxt = tp->snd_una;
2189 tcp_free_sackholes(tp);
2190 tp->snd_recover = tp->snd_max;
2191 if (tp->t_flags & TF_SACK_PERMIT)
2192 EXIT_FASTRECOVERY(tp->t_flags);
2193 tp->t_fb->tfb_tcp_output(tp);
2198 * Look-up the routing entry to the peer of this inpcb. If no route
2199 * is found and it cannot be allocated, then return 0. This routine
2200 * is called by TCP routines that access the rmx structure and by
2201 * tcp_mss_update to get the peer/interface MTU.
2204 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2207 struct sockaddr_in *dst;
2211 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2213 bzero(&sro, sizeof(sro));
2214 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2215 dst = (struct sockaddr_in *)&sro.ro_dst;
2216 dst->sin_family = AF_INET;
2217 dst->sin_len = sizeof(*dst);
2218 dst->sin_addr = inc->inc_faddr;
2219 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
2221 if (sro.ro_rt != NULL) {
2222 ifp = sro.ro_rt->rt_ifp;
2223 if (sro.ro_rt->rt_mtu == 0)
2224 maxmtu = ifp->if_mtu;
2226 maxmtu = min(sro.ro_rt->rt_mtu, ifp->if_mtu);
2228 /* Report additional interface capabilities. */
2230 if (ifp->if_capenable & IFCAP_TSO4 &&
2231 ifp->if_hwassist & CSUM_TSO) {
2232 cap->ifcap |= CSUM_TSO;
2233 cap->tsomax = ifp->if_hw_tsomax;
2234 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2235 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2246 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2248 struct route_in6 sro6;
2252 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2254 bzero(&sro6, sizeof(sro6));
2255 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2256 sro6.ro_dst.sin6_family = AF_INET6;
2257 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
2258 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
2259 in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
2261 if (sro6.ro_rt != NULL) {
2262 ifp = sro6.ro_rt->rt_ifp;
2263 if (sro6.ro_rt->rt_mtu == 0)
2264 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
2266 maxmtu = min(sro6.ro_rt->rt_mtu,
2267 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
2269 /* Report additional interface capabilities. */
2271 if (ifp->if_capenable & IFCAP_TSO6 &&
2272 ifp->if_hwassist & CSUM_TSO) {
2273 cap->ifcap |= CSUM_TSO;
2274 cap->tsomax = ifp->if_hw_tsomax;
2275 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2276 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2287 * Calculate effective SMSS per RFC5681 definition for a given TCP
2288 * connection at its current state, taking into account SACK and etc.
2291 tcp_maxseg(const struct tcpcb *tp)
2295 if (tp->t_flags & TF_NOOPT)
2296 return (tp->t_maxseg);
2299 * Here we have a simplified code from tcp_addoptions(),
2300 * without a proper loop, and having most of paddings hardcoded.
2301 * We might make mistakes with padding here in some edge cases,
2302 * but this is harmless, since result of tcp_maxseg() is used
2303 * only in cwnd and ssthresh estimations.
2305 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2306 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2307 if (tp->t_flags & TF_RCVD_TSTMP)
2308 optlen = TCPOLEN_TSTAMP_APPA;
2311 #ifdef TCP_SIGNATURE
2312 if (tp->t_flags & TF_SIGNATURE)
2313 optlen += PAD(TCPOLEN_SIGNATURE);
2315 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2316 optlen += TCPOLEN_SACKHDR;
2317 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2318 optlen = PAD(optlen);
2321 if (tp->t_flags & TF_REQ_TSTMP)
2322 optlen = TCPOLEN_TSTAMP_APPA;
2324 optlen = PAD(TCPOLEN_MAXSEG);
2325 if (tp->t_flags & TF_REQ_SCALE)
2326 optlen += PAD(TCPOLEN_WINDOW);
2327 #ifdef TCP_SIGNATURE
2328 if (tp->t_flags & TF_SIGNATURE)
2329 optlen += PAD(TCPOLEN_SIGNATURE);
2331 if (tp->t_flags & TF_SACK_PERMIT)
2332 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2335 optlen = min(optlen, TCP_MAXOLEN);
2336 return (tp->t_maxseg - optlen);
2340 /* compute ESP/AH header size for TCP, including outer IP header. */
2342 ipsec_hdrsiz_tcp(struct tcpcb *tp)
2349 struct ip6_hdr *ip6;
2353 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) ||
2354 (!key_havesp(IPSEC_DIR_OUTBOUND)))
2356 m = m_gethdr(M_NOWAIT, MT_DATA);
2361 if ((inp->inp_vflag & INP_IPV6) != 0) {
2362 ip6 = mtod(m, struct ip6_hdr *);
2363 th = (struct tcphdr *)(ip6 + 1);
2364 m->m_pkthdr.len = m->m_len =
2365 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
2366 tcpip_fillheaders(inp, ip6, th);
2367 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2371 ip = mtod(m, struct ip *);
2372 th = (struct tcphdr *)(ip + 1);
2373 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
2374 tcpip_fillheaders(inp, ip, th);
2375 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2383 #ifdef TCP_SIGNATURE
2385 * Callback function invoked by m_apply() to digest TCP segment data
2386 * contained within an mbuf chain.
2389 tcp_signature_apply(void *fstate, void *data, u_int len)
2392 MD5Update(fstate, (u_char *)data, len);
2397 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2398 * search with the destination IP address, and a 'magic SPI' to be
2399 * determined by the application. This is hardcoded elsewhere to 1179
2402 tcp_get_sav(struct mbuf *m, u_int direction)
2404 union sockaddr_union dst;
2405 struct secasvar *sav;
2408 struct ip6_hdr *ip6;
2409 char ip6buf[INET6_ADDRSTRLEN];
2412 /* Extract the destination from the IP header in the mbuf. */
2413 bzero(&dst, sizeof(union sockaddr_union));
2414 ip = mtod(m, struct ip *);
2416 ip6 = NULL; /* Make the compiler happy. */
2421 dst.sa.sa_len = sizeof(struct sockaddr_in);
2422 dst.sa.sa_family = AF_INET;
2423 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2424 ip->ip_src : ip->ip_dst;
2428 case (IPV6_VERSION >> 4):
2429 ip6 = mtod(m, struct ip6_hdr *);
2430 dst.sa.sa_len = sizeof(struct sockaddr_in6);
2431 dst.sa.sa_family = AF_INET6;
2432 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2433 ip6->ip6_src : ip6->ip6_dst;
2442 /* Look up an SADB entry which matches the address of the peer. */
2443 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2445 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2446 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2448 (ip->ip_v == (IPV6_VERSION >> 4)) ?
2449 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2458 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2461 * m pointer to head of mbuf chain
2462 * len length of TCP segment data, excluding options
2463 * optlen length of TCP segment options
2464 * buf pointer to storage for computed MD5 digest
2465 * sav pointer to security assosiation
2467 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2468 * When called from tcp_input(), we can be sure that th_sum has been
2469 * zeroed out and verified already.
2471 * Releases reference to SADB key before return.
2473 * Return 0 if successful, otherwise return -1.
2477 tcp_signature_do_compute(struct mbuf *m, int len, int optlen,
2478 u_char *buf, struct secasvar *sav)
2481 struct ippseudo ippseudo;
2487 struct ipovly *ipovly;
2491 struct ip6_hdr *ip6;
2492 struct in6_addr in6;
2498 KASSERT(m != NULL, ("NULL mbuf chain"));
2499 KASSERT(buf != NULL, ("NULL signature pointer"));
2501 /* Extract the destination from the IP header in the mbuf. */
2502 ip = mtod(m, struct ip *);
2504 ip6 = NULL; /* Make the compiler happy. */
2509 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2511 * XXX The ippseudo header MUST be digested in network byte order,
2512 * or else we'll fail the regression test. Assume all fields we've
2513 * been doing arithmetic on have been in host byte order.
2514 * XXX One cannot depend on ipovly->ih_len here. When called from
2515 * tcp_output(), the underlying ip_len member has not yet been set.
2520 ipovly = (struct ipovly *)ip;
2521 ippseudo.ippseudo_src = ipovly->ih_src;
2522 ippseudo.ippseudo_dst = ipovly->ih_dst;
2523 ippseudo.ippseudo_pad = 0;
2524 ippseudo.ippseudo_p = IPPROTO_TCP;
2525 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2527 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2529 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2530 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2535 * RFC 2385, 2.0 Proposal
2536 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2537 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2538 * extended next header value (to form 32 bits), and 32-bit segment
2540 * Note: Upper-Layer Packet Length comes before Next Header.
2542 case (IPV6_VERSION >> 4):
2544 in6_clearscope(&in6);
2545 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2547 in6_clearscope(&in6);
2548 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2549 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2550 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2552 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2553 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2554 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2556 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2558 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2559 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2571 * Step 2: Update MD5 hash with TCP header, excluding options.
2572 * The TCP checksum must be set to zero.
2574 savecsum = th->th_sum;
2576 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2577 th->th_sum = savecsum;
2580 * Step 3: Update MD5 hash with TCP segment data.
2581 * Use m_apply() to avoid an early m_pullup().
2584 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2587 * Step 4: Update MD5 hash with shared secret.
2589 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2590 MD5Final(buf, &ctx);
2592 key_sa_recordxfer(sav, m);
2598 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2600 * Return 0 if successful, otherwise return -1.
2603 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2604 u_char *buf, u_int direction)
2606 struct secasvar *sav;
2608 if ((sav = tcp_get_sav(m, direction)) == NULL)
2611 return (tcp_signature_do_compute(m, len, optlen, buf, sav));
2615 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2618 * m pointer to head of mbuf chain
2619 * len length of TCP segment data, excluding options
2620 * optlen length of TCP segment options
2621 * buf pointer to storage for computed MD5 digest
2622 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2624 * Return 1 if successful, otherwise return 0.
2627 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2628 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2630 char tmpdigest[TCP_SIGLEN];
2632 if (tcp_sig_checksigs == 0)
2634 if ((tcpbflag & TF_SIGNATURE) == 0) {
2635 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2638 * If this socket is not expecting signature but
2639 * the segment contains signature just fail.
2641 TCPSTAT_INC(tcps_sig_err_sigopt);
2642 TCPSTAT_INC(tcps_sig_rcvbadsig);
2646 /* Signature is not expected, and not present in segment. */
2651 * If this socket is expecting signature but the segment does not
2652 * contain any just fail.
2654 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2655 TCPSTAT_INC(tcps_sig_err_nosigopt);
2656 TCPSTAT_INC(tcps_sig_rcvbadsig);
2659 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2660 IPSEC_DIR_INBOUND) == -1) {
2661 TCPSTAT_INC(tcps_sig_err_buildsig);
2662 TCPSTAT_INC(tcps_sig_rcvbadsig);
2666 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2667 TCPSTAT_INC(tcps_sig_rcvbadsig);
2670 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2673 #endif /* TCP_SIGNATURE */
2676 sysctl_drop(SYSCTL_HANDLER_ARGS)
2678 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2679 struct sockaddr_storage addrs[2];
2683 struct sockaddr_in *fin, *lin;
2685 struct sockaddr_in6 *fin6, *lin6;
2696 if (req->oldptr != NULL || req->oldlen != 0)
2698 if (req->newptr == NULL)
2700 if (req->newlen < sizeof(addrs))
2702 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2706 switch (addrs[0].ss_family) {
2709 fin6 = (struct sockaddr_in6 *)&addrs[0];
2710 lin6 = (struct sockaddr_in6 *)&addrs[1];
2711 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2712 lin6->sin6_len != sizeof(struct sockaddr_in6))
2714 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2715 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2717 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2718 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2719 fin = (struct sockaddr_in *)&addrs[0];
2720 lin = (struct sockaddr_in *)&addrs[1];
2723 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2726 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2733 fin = (struct sockaddr_in *)&addrs[0];
2734 lin = (struct sockaddr_in *)&addrs[1];
2735 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2736 lin->sin_len != sizeof(struct sockaddr_in))
2743 INP_INFO_RLOCK(&V_tcbinfo);
2744 switch (addrs[0].ss_family) {
2747 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2748 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2749 INPLOOKUP_WLOCKPCB, NULL);
2754 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2755 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2760 if (inp->inp_flags & INP_TIMEWAIT) {
2762 * XXXRW: There currently exists a state where an
2763 * inpcb is present, but its timewait state has been
2764 * discarded. For now, don't allow dropping of this
2772 } else if (!(inp->inp_flags & INP_DROPPED) &&
2773 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2774 tp = intotcpcb(inp);
2775 tp = tcp_drop(tp, ECONNABORTED);
2782 INP_INFO_RUNLOCK(&V_tcbinfo);
2786 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2787 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2788 0, sysctl_drop, "", "Drop TCP connection");
2791 * Generate a standardized TCP log line for use throughout the
2792 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2793 * allow use in the interrupt context.
2795 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2796 * NB: The function may return NULL if memory allocation failed.
2798 * Due to header inclusion and ordering limitations the struct ip
2799 * and ip6_hdr pointers have to be passed as void pointers.
2802 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2806 /* Is logging enabled? */
2807 if (tcp_log_in_vain == 0)
2810 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2814 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2818 /* Is logging enabled? */
2819 if (tcp_log_debug == 0)
2822 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2826 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2833 const struct ip6_hdr *ip6;
2835 ip6 = (const struct ip6_hdr *)ip6hdr;
2837 ip = (struct ip *)ip4hdr;
2840 * The log line looks like this:
2841 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2843 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2844 sizeof(PRINT_TH_FLAGS) + 1 +
2846 2 * INET6_ADDRSTRLEN;
2848 2 * INET_ADDRSTRLEN;
2851 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2855 strcat(s, "TCP: [");
2858 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2859 inet_ntoa_r(inc->inc_faddr, sp);
2861 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2863 inet_ntoa_r(inc->inc_laddr, sp);
2865 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2868 ip6_sprintf(sp, &inc->inc6_faddr);
2870 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2872 ip6_sprintf(sp, &inc->inc6_laddr);
2874 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2875 } else if (ip6 && th) {
2876 ip6_sprintf(sp, &ip6->ip6_src);
2878 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2880 ip6_sprintf(sp, &ip6->ip6_dst);
2882 sprintf(sp, "]:%i", ntohs(th->th_dport));
2885 } else if (ip && th) {
2886 inet_ntoa_r(ip->ip_src, sp);
2888 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2890 inet_ntoa_r(ip->ip_dst, sp);
2892 sprintf(sp, "]:%i", ntohs(th->th_dport));
2900 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2901 if (*(s + size - 1) != '\0')
2902 panic("%s: string too long", __func__);
2907 * A subroutine which makes it easy to track TCP state changes with DTrace.
2908 * This function shouldn't be called for t_state initializations that don't
2909 * correspond to actual TCP state transitions.
2912 tcp_state_change(struct tcpcb *tp, int newstate)
2914 #if defined(KDTRACE_HOOKS)
2915 int pstate = tp->t_state;
2918 tp->t_state = newstate;
2919 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);