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/cc.h>
72 #include <netinet/in.h>
73 #include <netinet/in_kdtrace.h>
74 #include <netinet/in_pcb.h>
75 #include <netinet/in_systm.h>
76 #include <netinet/in_var.h>
77 #include <netinet/ip.h>
78 #include <netinet/ip_icmp.h>
79 #include <netinet/ip_var.h>
81 #include <netinet/ip6.h>
82 #include <netinet6/in6_pcb.h>
83 #include <netinet6/ip6_var.h>
84 #include <netinet6/scope6_var.h>
85 #include <netinet6/nd6.h>
89 #include <netinet/tcp_fastopen.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>
97 #include <netinet6/tcp6_var.h>
99 #include <netinet/tcpip.h>
101 #include <netinet/tcp_pcap.h>
104 #include <netinet/tcp_debug.h>
107 #include <netinet6/ip6protosw.h>
110 #include <netinet/tcp_offload.h>
114 #include <netipsec/ipsec.h>
115 #include <netipsec/xform.h>
117 #include <netipsec/ipsec6.h>
119 #include <netipsec/key.h>
120 #include <sys/syslog.h>
123 #include <machine/in_cksum.h>
126 #include <security/mac/mac_framework.h>
128 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
130 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
133 struct rwlock tcp_function_lock;
136 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
141 error = sysctl_handle_int(oidp, &new, 0, req);
142 if (error == 0 && req->newptr) {
143 if (new < TCP_MINMSS)
151 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
152 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
153 &sysctl_net_inet_tcp_mss_check, "I",
154 "Default TCP Maximum Segment Size");
158 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
162 new = V_tcp_v6mssdflt;
163 error = sysctl_handle_int(oidp, &new, 0, req);
164 if (error == 0 && req->newptr) {
165 if (new < TCP_MINMSS)
168 V_tcp_v6mssdflt = new;
173 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
174 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
175 &sysctl_net_inet_tcp_mss_v6_check, "I",
176 "Default TCP Maximum Segment Size for IPv6");
180 * Minimum MSS we accept and use. This prevents DoS attacks where
181 * we are forced to a ridiculous low MSS like 20 and send hundreds
182 * of packets instead of one. The effect scales with the available
183 * bandwidth and quickly saturates the CPU and network interface
184 * with packet generation and sending. Set to zero to disable MINMSS
185 * checking. This setting prevents us from sending too small packets.
187 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
189 &VNET_NAME(tcp_minmss), 0,
190 "Minimum TCP Maximum Segment Size");
192 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
193 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
194 &VNET_NAME(tcp_do_rfc1323), 0,
195 "Enable rfc1323 (high performance TCP) extensions");
197 static int tcp_log_debug = 0;
198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
199 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
201 static int tcp_tcbhashsize;
202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
203 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
205 static int do_tcpdrain = 1;
206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
207 "Enable tcp_drain routine for extra help when low on mbufs");
209 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
210 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
212 static VNET_DEFINE(int, icmp_may_rst) = 1;
213 #define V_icmp_may_rst VNET(icmp_may_rst)
214 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
215 &VNET_NAME(icmp_may_rst), 0,
216 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
218 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
219 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
221 &VNET_NAME(tcp_isn_reseed_interval), 0,
222 "Seconds between reseeding of ISN secret");
224 static int tcp_soreceive_stream;
225 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
226 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
229 static int tcp_sig_checksigs = 1;
230 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
231 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
234 VNET_DEFINE(uma_zone_t, sack_hole_zone);
235 #define V_sack_hole_zone VNET(sack_hole_zone)
237 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
239 static struct inpcb *tcp_notify(struct inpcb *, int);
240 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
241 static void tcp_mtudisc(struct inpcb *, int);
242 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
243 void *ip4hdr, const void *ip6hdr);
244 static void tcp_timer_discard(struct tcpcb *, uint32_t);
247 static struct tcp_function_block tcp_def_funcblk = {
251 tcp_default_ctloutput,
263 struct tcp_funchead t_functions;
264 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
266 static struct tcp_function_block *
267 find_tcp_functions_locked(struct tcp_function_set *fs)
269 struct tcp_function *f;
270 struct tcp_function_block *blk=NULL;
272 TAILQ_FOREACH(f, &t_functions, tf_next) {
273 if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) {
281 static struct tcp_function_block *
282 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
284 struct tcp_function_block *rblk=NULL;
285 struct tcp_function *f;
287 TAILQ_FOREACH(f, &t_functions, tf_next) {
288 if (f->tf_fb == blk) {
299 struct tcp_function_block *
300 find_and_ref_tcp_functions(struct tcp_function_set *fs)
302 struct tcp_function_block *blk;
304 rw_rlock(&tcp_function_lock);
305 blk = find_tcp_functions_locked(fs);
307 refcount_acquire(&blk->tfb_refcnt);
308 rw_runlock(&tcp_function_lock);
312 struct tcp_function_block *
313 find_and_ref_tcp_fb(struct tcp_function_block *blk)
315 struct tcp_function_block *rblk;
317 rw_rlock(&tcp_function_lock);
318 rblk = find_tcp_fb_locked(blk, NULL);
320 refcount_acquire(&rblk->tfb_refcnt);
321 rw_runlock(&tcp_function_lock);
327 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
330 struct tcp_function_set fs;
331 struct tcp_function_block *blk;
333 memset(&fs, 0, sizeof(fs));
334 rw_rlock(&tcp_function_lock);
335 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
338 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
339 fs.pcbcnt = blk->tfb_refcnt;
341 rw_runlock(&tcp_function_lock);
342 error = sysctl_handle_string(oidp, fs.function_set_name,
343 sizeof(fs.function_set_name), req);
345 /* Check for error or no change */
346 if (error != 0 || req->newptr == NULL)
349 rw_wlock(&tcp_function_lock);
350 blk = find_tcp_functions_locked(&fs);
352 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
356 tcp_func_set_ptr = blk;
358 rw_wunlock(&tcp_function_lock);
362 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
363 CTLTYPE_STRING | CTLFLAG_RW,
364 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
365 "Set/get the default TCP functions");
368 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
370 int error, cnt, linesz;
371 struct tcp_function *f;
376 rw_rlock(&tcp_function_lock);
377 TAILQ_FOREACH(f, &t_functions, tf_next) {
380 rw_runlock(&tcp_function_lock);
382 bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1;
383 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
388 linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count");
393 rw_rlock(&tcp_function_lock);
394 TAILQ_FOREACH(f, &t_functions, tf_next) {
395 linesz = snprintf(cp, bufsz, "%-32s%c %u\n",
396 f->tf_fb->tfb_tcp_block_name,
397 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
398 f->tf_fb->tfb_refcnt);
399 if (linesz >= bufsz) {
407 rw_runlock(&tcp_function_lock);
409 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
410 free(buffer, M_TEMP);
414 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
415 CTLTYPE_STRING|CTLFLAG_RD,
416 NULL, 0, sysctl_net_inet_list_available, "A",
417 "list available TCP Function sets");
420 * Target size of TCP PCB hash tables. Must be a power of two.
422 * Note that this can be overridden by the kernel environment
423 * variable net.inet.tcp.tcbhashsize
426 #define TCBHASHSIZE 0
431 * Callouts should be moved into struct tcp directly. They are currently
432 * separate because the tcpcb structure is exported to userland for sysctl
433 * parsing purposes, which do not know about callouts.
442 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
443 #define V_tcpcb_zone VNET(tcpcb_zone)
445 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
446 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
448 static struct mtx isn_mtx;
450 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
451 #define ISN_LOCK() mtx_lock(&isn_mtx)
452 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
455 * TCP initialization.
458 tcp_zone_change(void *tag)
461 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
462 uma_zone_set_max(V_tcpcb_zone, maxsockets);
463 tcp_tw_zone_change();
467 tcp_inpcb_init(void *mem, int size, int flags)
469 struct inpcb *inp = mem;
471 INP_LOCK_INIT(inp, "inp", "tcpinp");
476 * Take a value and get the next power of 2 that doesn't overflow.
477 * Used to size the tcp_inpcb hash buckets.
480 maketcp_hashsize(int size)
486 * get the next power of 2 higher than maxsockets.
488 hashsize = 1 << fls(size);
489 /* catch overflow, and just go one power of 2 smaller */
490 if (hashsize < size) {
491 hashsize = 1 << (fls(size) - 1);
497 register_tcp_functions(struct tcp_function_block *blk, int wait)
499 struct tcp_function_block *lblk;
500 struct tcp_function *n;
501 struct tcp_function_set fs;
503 if ((blk->tfb_tcp_output == NULL) ||
504 (blk->tfb_tcp_do_segment == NULL) ||
505 (blk->tfb_tcp_ctloutput == NULL) ||
506 (strlen(blk->tfb_tcp_block_name) == 0)) {
508 * These functions are required and you
513 if (blk->tfb_tcp_timer_stop_all ||
514 blk->tfb_tcp_timers_left ||
515 blk->tfb_tcp_timer_activate ||
516 blk->tfb_tcp_timer_active ||
517 blk->tfb_tcp_timer_stop) {
519 * If you define one timer function you
520 * must have them all.
522 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
523 (blk->tfb_tcp_timers_left == NULL) ||
524 (blk->tfb_tcp_timer_activate == NULL) ||
525 (blk->tfb_tcp_timer_active == NULL) ||
526 (blk->tfb_tcp_timer_stop == NULL)) {
530 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
535 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
536 rw_wlock(&tcp_function_lock);
537 lblk = find_tcp_functions_locked(&fs);
539 /* Duplicate name space not allowed */
540 rw_wunlock(&tcp_function_lock);
541 free(n, M_TCPFUNCTIONS);
544 refcount_init(&blk->tfb_refcnt, 0);
546 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
547 rw_wunlock(&tcp_function_lock);
552 deregister_tcp_functions(struct tcp_function_block *blk)
554 struct tcp_function_block *lblk;
555 struct tcp_function *f;
558 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
559 /* You can't un-register the default */
562 rw_wlock(&tcp_function_lock);
563 if (blk == tcp_func_set_ptr) {
564 /* You can't free the current default */
565 rw_wunlock(&tcp_function_lock);
568 if (blk->tfb_refcnt) {
569 /* Still tcb attached, mark it. */
570 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
571 rw_wunlock(&tcp_function_lock);
574 lblk = find_tcp_fb_locked(blk, &f);
577 TAILQ_REMOVE(&t_functions, f, tf_next);
579 free(f, M_TCPFUNCTIONS);
582 rw_wunlock(&tcp_function_lock);
589 const char *tcbhash_tuneable;
592 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
594 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
595 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
596 printf("%s: WARNING: unable to register helper hook\n", __func__);
597 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
598 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
599 printf("%s: WARNING: unable to register helper hook\n", __func__);
600 hashsize = TCBHASHSIZE;
601 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
604 * Auto tune the hash size based on maxsockets.
605 * A perfect hash would have a 1:1 mapping
606 * (hashsize = maxsockets) however it's been
607 * suggested that O(2) average is better.
609 hashsize = maketcp_hashsize(maxsockets / 4);
611 * Our historical default is 512,
612 * do not autotune lower than this.
616 if (bootverbose && IS_DEFAULT_VNET(curvnet))
617 printf("%s: %s auto tuned to %d\n", __func__,
618 tcbhash_tuneable, hashsize);
621 * We require a hashsize to be a power of two.
622 * Previously if it was not a power of two we would just reset it
623 * back to 512, which could be a nasty surprise if you did not notice
625 * Instead what we do is clip it to the closest power of two lower
626 * than the specified hash value.
628 if (!powerof2(hashsize)) {
629 int oldhashsize = hashsize;
631 hashsize = maketcp_hashsize(hashsize);
632 /* prevent absurdly low value */
635 printf("%s: WARNING: TCB hash size not a power of 2, "
636 "clipped from %d to %d.\n", __func__, oldhashsize,
639 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
640 "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
641 IPI_HASHFIELDS_4TUPLE);
644 * These have to be type stable for the benefit of the timers.
646 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
647 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
648 uma_zone_set_max(V_tcpcb_zone, maxsockets);
649 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
655 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
656 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
657 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
659 /* Skip initialization of globals for non-default instances. */
660 if (!IS_DEFAULT_VNET(curvnet))
663 tcp_reass_global_init();
665 /* XXX virtualize those bellow? */
666 tcp_delacktime = TCPTV_DELACK;
667 tcp_keepinit = TCPTV_KEEP_INIT;
668 tcp_keepidle = TCPTV_KEEP_IDLE;
669 tcp_keepintvl = TCPTV_KEEPINTVL;
670 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
672 tcp_rexmit_min = TCPTV_MIN;
673 if (tcp_rexmit_min < 1)
675 tcp_rexmit_slop = TCPTV_CPU_VAR;
676 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
677 tcp_tcbhashsize = hashsize;
678 /* Setup the tcp function block list */
679 TAILQ_INIT(&t_functions);
680 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
681 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
683 if (tcp_soreceive_stream) {
685 tcp_usrreqs.pru_soreceive = soreceive_stream;
688 tcp6_usrreqs.pru_soreceive = soreceive_stream;
693 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
695 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
697 if (max_protohdr < TCP_MINPROTOHDR)
698 max_protohdr = TCP_MINPROTOHDR;
699 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
701 #undef TCP_MINPROTOHDR
704 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
705 SHUTDOWN_PRI_DEFAULT);
706 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
707 EVENTHANDLER_PRI_ANY);
724 tcp_fastopen_destroy();
729 in_pcbinfo_destroy(&V_tcbinfo);
730 uma_zdestroy(V_sack_hole_zone);
731 uma_zdestroy(V_tcpcb_zone);
733 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
735 printf("%s: WARNING: unable to deregister helper hook "
736 "type=%d, id=%d: error %d returned\n", __func__,
737 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
739 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
741 printf("%s: WARNING: unable to deregister helper hook "
742 "type=%d, id=%d: error %d returned\n", __func__,
743 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
755 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
756 * tcp_template used to store this data in mbufs, but we now recopy it out
757 * of the tcpcb each time to conserve mbufs.
760 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
762 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
764 INP_WLOCK_ASSERT(inp);
767 if ((inp->inp_vflag & INP_IPV6) != 0) {
770 ip6 = (struct ip6_hdr *)ip_ptr;
771 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
772 (inp->inp_flow & IPV6_FLOWINFO_MASK);
773 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
774 (IPV6_VERSION & IPV6_VERSION_MASK);
775 ip6->ip6_nxt = IPPROTO_TCP;
776 ip6->ip6_plen = htons(sizeof(struct tcphdr));
777 ip6->ip6_src = inp->in6p_laddr;
778 ip6->ip6_dst = inp->in6p_faddr;
781 #if defined(INET6) && defined(INET)
788 ip = (struct ip *)ip_ptr;
789 ip->ip_v = IPVERSION;
791 ip->ip_tos = inp->inp_ip_tos;
795 ip->ip_ttl = inp->inp_ip_ttl;
797 ip->ip_p = IPPROTO_TCP;
798 ip->ip_src = inp->inp_laddr;
799 ip->ip_dst = inp->inp_faddr;
802 th->th_sport = inp->inp_lport;
803 th->th_dport = inp->inp_fport;
811 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
815 * Create template to be used to send tcp packets on a connection.
816 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
817 * use for this function is in keepalives, which use tcp_respond.
820 tcpip_maketemplate(struct inpcb *inp)
824 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
827 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
832 * Send a single message to the TCP at address specified by
833 * the given TCP/IP header. If m == NULL, then we make a copy
834 * of the tcpiphdr at th and send directly to the addressed host.
835 * This is used to force keep alive messages out using the TCP
836 * template for a connection. If flags are given then we send
837 * a message back to the TCP which originated the segment th,
838 * and discard the mbuf containing it and any other attached mbufs.
840 * In any case the ack and sequence number of the transmitted
841 * segment are as specified by the parameters.
843 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
846 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
847 tcp_seq ack, tcp_seq seq, int flags)
860 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
863 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
870 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
871 INP_WLOCK_ASSERT(inp);
876 if (!(flags & TH_RST)) {
877 win = sbspace(&inp->inp_socket->so_rcv);
878 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
879 win = (long)TCP_MAXWIN << tp->rcv_scale;
883 m = m_gethdr(M_NOWAIT, MT_DATA);
887 m->m_data += max_linkhdr;
890 bcopy((caddr_t)ip6, mtod(m, caddr_t),
891 sizeof(struct ip6_hdr));
892 ip6 = mtod(m, struct ip6_hdr *);
893 nth = (struct tcphdr *)(ip6 + 1);
897 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
898 ip = mtod(m, struct ip *);
899 nth = (struct tcphdr *)(ip + 1);
901 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
906 * XXX MRT We inherrit the FIB, which is lucky.
910 m->m_data = (caddr_t)ipgen;
911 /* m_len is set later */
913 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
916 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
917 nth = (struct tcphdr *)(ip6 + 1);
921 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
922 nth = (struct tcphdr *)(ip + 1);
926 * this is usually a case when an extension header
927 * exists between the IPv6 header and the
930 nth->th_sport = th->th_sport;
931 nth->th_dport = th->th_dport;
933 xchg(nth->th_dport, nth->th_sport, uint16_t);
939 ip6->ip6_vfc = IPV6_VERSION;
940 ip6->ip6_nxt = IPPROTO_TCP;
941 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
942 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
945 #if defined(INET) && defined(INET6)
950 tlen += sizeof (struct tcpiphdr);
951 ip->ip_len = htons(tlen);
952 ip->ip_ttl = V_ip_defttl;
953 if (V_path_mtu_discovery)
954 ip->ip_off |= htons(IP_DF);
958 m->m_pkthdr.len = tlen;
959 m->m_pkthdr.rcvif = NULL;
963 * Packet is associated with a socket, so allow the
964 * label of the response to reflect the socket label.
966 INP_WLOCK_ASSERT(inp);
967 mac_inpcb_create_mbuf(inp, m);
970 * Packet is not associated with a socket, so possibly
971 * update the label in place.
973 mac_netinet_tcp_reply(m);
976 nth->th_seq = htonl(seq);
977 nth->th_ack = htonl(ack);
979 nth->th_off = sizeof (struct tcphdr) >> 2;
980 nth->th_flags = flags;
982 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
984 nth->th_win = htons((u_short)win);
987 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
990 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
991 nth->th_sum = in6_cksum_pseudo(ip6,
992 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
993 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
997 #if defined(INET6) && defined(INET)
1002 m->m_pkthdr.csum_flags = CSUM_TCP;
1003 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1004 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1008 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1009 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1011 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
1013 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
1016 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
1019 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
1021 #if defined(INET) && defined(INET6)
1025 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
1030 * Create a new TCP control block, making an
1031 * empty reassembly queue and hooking it to the argument
1032 * protocol control block. The `inp' parameter must have
1033 * come from the zone allocator set up in tcp_init().
1036 tcp_newtcpcb(struct inpcb *inp)
1038 struct tcpcb_mem *tm;
1041 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1044 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1049 /* Initialise cc_var struct for this tcpcb. */
1051 tp->ccv->type = IPPROTO_TCP;
1052 tp->ccv->ccvc.tcp = tp;
1053 rw_rlock(&tcp_function_lock);
1054 tp->t_fb = tcp_func_set_ptr;
1055 refcount_acquire(&tp->t_fb->tfb_refcnt);
1056 rw_runlock(&tcp_function_lock);
1057 if (tp->t_fb->tfb_tcp_fb_init) {
1058 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1061 * Use the current system default CC algorithm.
1064 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1065 CC_ALGO(tp) = CC_DEFAULT();
1068 if (CC_ALGO(tp)->cb_init != NULL)
1069 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1070 if (tp->t_fb->tfb_tcp_fb_fini)
1071 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1072 refcount_release(&tp->t_fb->tfb_refcnt);
1073 uma_zfree(V_tcpcb_zone, tm);
1078 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1079 if (tp->t_fb->tfb_tcp_fb_fini)
1080 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1081 refcount_release(&tp->t_fb->tfb_refcnt);
1082 uma_zfree(V_tcpcb_zone, tm);
1087 tp->t_vnet = inp->inp_vnet;
1089 tp->t_timers = &tm->tt;
1090 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
1093 isipv6 ? V_tcp_v6mssdflt :
1097 /* Set up our timeouts. */
1098 callout_init(&tp->t_timers->tt_rexmt, 1);
1099 callout_init(&tp->t_timers->tt_persist, 1);
1100 callout_init(&tp->t_timers->tt_keep, 1);
1101 callout_init(&tp->t_timers->tt_2msl, 1);
1102 callout_init(&tp->t_timers->tt_delack, 1);
1104 if (V_tcp_do_rfc1323)
1105 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1107 tp->t_flags |= TF_SACK_PERMIT;
1108 TAILQ_INIT(&tp->snd_holes);
1110 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1113 in_pcbref(inp); /* Reference for tcpcb */
1117 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1118 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1119 * reasonable initial retransmit time.
1121 tp->t_srtt = TCPTV_SRTTBASE;
1122 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1123 tp->t_rttmin = tcp_rexmit_min;
1124 tp->t_rxtcur = TCPTV_RTOBASE;
1125 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1126 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1127 tp->t_rcvtime = ticks;
1129 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1130 * because the socket may be bound to an IPv6 wildcard address,
1131 * which may match an IPv4-mapped IPv6 address.
1133 inp->inp_ip_ttl = V_ip_defttl;
1137 * Init the TCP PCAP queues.
1139 tcp_pcap_tcpcb_init(tp);
1141 return (tp); /* XXX */
1145 * Switch the congestion control algorithm back to NewReno for any active
1146 * control blocks using an algorithm which is about to go away.
1147 * This ensures the CC framework can allow the unload to proceed without leaving
1148 * any dangling pointers which would trigger a panic.
1149 * Returning non-zero would inform the CC framework that something went wrong
1150 * and it would be unsafe to allow the unload to proceed. However, there is no
1151 * way for this to occur with this implementation so we always return zero.
1154 tcp_ccalgounload(struct cc_algo *unload_algo)
1156 struct cc_algo *tmpalgo;
1159 VNET_ITERATOR_DECL(vnet_iter);
1162 * Check all active control blocks across all network stacks and change
1163 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1164 * requires cleanup code to be run, call it.
1167 VNET_FOREACH(vnet_iter) {
1168 CURVNET_SET(vnet_iter);
1169 INP_INFO_WLOCK(&V_tcbinfo);
1171 * New connections already part way through being initialised
1172 * with the CC algo we're removing will not race with this code
1173 * because the INP_INFO_WLOCK is held during initialisation. We
1174 * therefore don't enter the loop below until the connection
1175 * list has stabilised.
1177 LIST_FOREACH(inp, &V_tcb, inp_list) {
1179 /* Important to skip tcptw structs. */
1180 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1181 (tp = intotcpcb(inp)) != NULL) {
1183 * By holding INP_WLOCK here, we are assured
1184 * that the connection is not currently
1185 * executing inside the CC module's functions
1186 * i.e. it is safe to make the switch back to
1189 if (CC_ALGO(tp) == unload_algo) {
1190 tmpalgo = CC_ALGO(tp);
1191 /* NewReno does not require any init. */
1192 CC_ALGO(tp) = &newreno_cc_algo;
1193 if (tmpalgo->cb_destroy != NULL)
1194 tmpalgo->cb_destroy(tp->ccv);
1199 INP_INFO_WUNLOCK(&V_tcbinfo);
1202 VNET_LIST_RUNLOCK();
1208 * Drop a TCP connection, reporting
1209 * the specified error. If connection is synchronized,
1210 * then send a RST to peer.
1213 tcp_drop(struct tcpcb *tp, int errno)
1215 struct socket *so = tp->t_inpcb->inp_socket;
1217 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1218 INP_WLOCK_ASSERT(tp->t_inpcb);
1220 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1221 tcp_state_change(tp, TCPS_CLOSED);
1222 (void) tp->t_fb->tfb_tcp_output(tp);
1223 TCPSTAT_INC(tcps_drops);
1225 TCPSTAT_INC(tcps_conndrops);
1226 if (errno == ETIMEDOUT && tp->t_softerror)
1227 errno = tp->t_softerror;
1228 so->so_error = errno;
1229 return (tcp_close(tp));
1233 tcp_discardcb(struct tcpcb *tp)
1235 struct inpcb *inp = tp->t_inpcb;
1236 struct socket *so = inp->inp_socket;
1238 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1242 INP_WLOCK_ASSERT(inp);
1245 * Make sure that all of our timers are stopped before we delete the
1248 * If stopping a timer fails, we schedule a discard function in same
1249 * callout, and the last discard function called will take care of
1250 * deleting the tcpcb.
1252 tcp_timer_stop(tp, TT_REXMT);
1253 tcp_timer_stop(tp, TT_PERSIST);
1254 tcp_timer_stop(tp, TT_KEEP);
1255 tcp_timer_stop(tp, TT_2MSL);
1256 tcp_timer_stop(tp, TT_DELACK);
1257 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1258 /* Call the stop-all function of the methods */
1259 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1263 * If we got enough samples through the srtt filter,
1264 * save the rtt and rttvar in the routing entry.
1265 * 'Enough' is arbitrarily defined as 4 rtt samples.
1266 * 4 samples is enough for the srtt filter to converge
1267 * to within enough % of the correct value; fewer samples
1268 * and we could save a bogus rtt. The danger is not high
1269 * as tcp quickly recovers from everything.
1270 * XXX: Works very well but needs some more statistics!
1272 if (tp->t_rttupdated >= 4) {
1273 struct hc_metrics_lite metrics;
1276 bzero(&metrics, sizeof(metrics));
1278 * Update the ssthresh always when the conditions below
1279 * are satisfied. This gives us better new start value
1280 * for the congestion avoidance for new connections.
1281 * ssthresh is only set if packet loss occured on a session.
1283 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1284 * being torn down. Ideally this code would not use 'so'.
1286 ssthresh = tp->snd_ssthresh;
1287 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1289 * convert the limit from user data bytes to
1290 * packets then to packet data bytes.
1292 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1295 ssthresh *= (u_long)(tp->t_maxseg +
1297 (isipv6 ? sizeof (struct ip6_hdr) +
1298 sizeof (struct tcphdr) :
1300 sizeof (struct tcpiphdr)
1307 metrics.rmx_ssthresh = ssthresh;
1309 metrics.rmx_rtt = tp->t_srtt;
1310 metrics.rmx_rttvar = tp->t_rttvar;
1311 metrics.rmx_cwnd = tp->snd_cwnd;
1312 metrics.rmx_sendpipe = 0;
1313 metrics.rmx_recvpipe = 0;
1315 tcp_hc_update(&inp->inp_inc, &metrics);
1318 /* free the reassembly queue, if any */
1319 tcp_reass_flush(tp);
1322 /* Disconnect offload device, if any. */
1323 if (tp->t_flags & TF_TOE)
1324 tcp_offload_detach(tp);
1327 tcp_free_sackholes(tp);
1330 /* Free the TCP PCAP queues. */
1331 tcp_pcap_drain(&(tp->t_inpkts));
1332 tcp_pcap_drain(&(tp->t_outpkts));
1335 /* Allow the CC algorithm to clean up after itself. */
1336 if (CC_ALGO(tp)->cb_destroy != NULL)
1337 CC_ALGO(tp)->cb_destroy(tp->ccv);
1339 khelp_destroy_osd(tp->osd);
1342 inp->inp_ppcb = NULL;
1343 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1344 /* We own the last reference on tcpcb, let's free it. */
1345 if ((tp->t_fb->tfb_tcp_timers_left) &&
1346 (tp->t_fb->tfb_tcp_timers_left(tp))) {
1347 /* Some fb timers left running! */
1350 if (tp->t_fb->tfb_tcp_fb_fini)
1351 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1352 refcount_release(&tp->t_fb->tfb_refcnt);
1354 uma_zfree(V_tcpcb_zone, tp);
1355 released = in_pcbrele_wlocked(inp);
1356 KASSERT(!released, ("%s: inp %p should not have been released "
1357 "here", __func__, inp));
1362 tcp_timer_2msl_discard(void *xtp)
1365 tcp_timer_discard((struct tcpcb *)xtp, TT_2MSL);
1369 tcp_timer_keep_discard(void *xtp)
1372 tcp_timer_discard((struct tcpcb *)xtp, TT_KEEP);
1376 tcp_timer_persist_discard(void *xtp)
1379 tcp_timer_discard((struct tcpcb *)xtp, TT_PERSIST);
1383 tcp_timer_rexmt_discard(void *xtp)
1386 tcp_timer_discard((struct tcpcb *)xtp, TT_REXMT);
1390 tcp_timer_delack_discard(void *xtp)
1393 tcp_timer_discard((struct tcpcb *)xtp, TT_DELACK);
1397 tcp_timer_discard(struct tcpcb *tp, uint32_t timer_type)
1401 CURVNET_SET(tp->t_vnet);
1402 INP_INFO_RLOCK(&V_tcbinfo);
1404 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1407 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1408 ("%s: tcpcb has to be stopped here", __func__));
1409 KASSERT((tp->t_timers->tt_flags & timer_type) != 0,
1410 ("%s: discard callout should be running", __func__));
1411 tp->t_timers->tt_flags &= ~timer_type;
1412 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1413 /* We own the last reference on this tcpcb, let's free it. */
1414 if ((tp->t_fb->tfb_tcp_timers_left) &&
1415 (tp->t_fb->tfb_tcp_timers_left(tp))) {
1416 /* Some fb timers left running! */
1419 if (tp->t_fb->tfb_tcp_fb_fini)
1420 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1421 refcount_release(&tp->t_fb->tfb_refcnt);
1423 uma_zfree(V_tcpcb_zone, tp);
1424 if (in_pcbrele_wlocked(inp)) {
1425 INP_INFO_RUNLOCK(&V_tcbinfo);
1432 INP_INFO_RUNLOCK(&V_tcbinfo);
1437 * Attempt to close a TCP control block, marking it as dropped, and freeing
1438 * the socket if we hold the only reference.
1441 tcp_close(struct tcpcb *tp)
1443 struct inpcb *inp = tp->t_inpcb;
1446 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1447 INP_WLOCK_ASSERT(inp);
1450 if (tp->t_state == TCPS_LISTEN)
1451 tcp_offload_listen_stop(tp);
1455 * This releases the TFO pending counter resource for TFO listen
1456 * sockets as well as passively-created TFO sockets that transition
1457 * from SYN_RECEIVED to CLOSED.
1459 if (tp->t_tfo_pending) {
1460 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1461 tp->t_tfo_pending = NULL;
1465 TCPSTAT_INC(tcps_closed);
1466 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1467 so = inp->inp_socket;
1468 soisdisconnected(so);
1469 if (inp->inp_flags & INP_SOCKREF) {
1470 KASSERT(so->so_state & SS_PROTOREF,
1471 ("tcp_close: !SS_PROTOREF"));
1472 inp->inp_flags &= ~INP_SOCKREF;
1476 so->so_state &= ~SS_PROTOREF;
1486 VNET_ITERATOR_DECL(vnet_iter);
1491 VNET_LIST_RLOCK_NOSLEEP();
1492 VNET_FOREACH(vnet_iter) {
1493 CURVNET_SET(vnet_iter);
1498 * Walk the tcpbs, if existing, and flush the reassembly queue,
1499 * if there is one...
1500 * XXX: The "Net/3" implementation doesn't imply that the TCP
1501 * reassembly queue should be flushed, but in a situation
1502 * where we're really low on mbufs, this is potentially
1505 INP_INFO_WLOCK(&V_tcbinfo);
1506 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1507 if (inpb->inp_flags & INP_TIMEWAIT)
1510 if ((tcpb = intotcpcb(inpb)) != NULL) {
1511 tcp_reass_flush(tcpb);
1512 tcp_clean_sackreport(tcpb);
1516 INP_INFO_WUNLOCK(&V_tcbinfo);
1519 VNET_LIST_RUNLOCK_NOSLEEP();
1523 * Notify a tcp user of an asynchronous error;
1524 * store error as soft error, but wake up user
1525 * (for now, won't do anything until can select for soft error).
1527 * Do not wake up user since there currently is no mechanism for
1528 * reporting soft errors (yet - a kqueue filter may be added).
1530 static struct inpcb *
1531 tcp_notify(struct inpcb *inp, int error)
1535 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1536 INP_WLOCK_ASSERT(inp);
1538 if ((inp->inp_flags & INP_TIMEWAIT) ||
1539 (inp->inp_flags & INP_DROPPED))
1542 tp = intotcpcb(inp);
1543 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1546 * Ignore some errors if we are hooked up.
1547 * If connection hasn't completed, has retransmitted several times,
1548 * and receives a second error, give up now. This is better
1549 * than waiting a long time to establish a connection that
1550 * can never complete.
1552 if (tp->t_state == TCPS_ESTABLISHED &&
1553 (error == EHOSTUNREACH || error == ENETUNREACH ||
1554 error == EHOSTDOWN)) {
1556 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1558 tp = tcp_drop(tp, error);
1564 tp->t_softerror = error;
1568 wakeup( &so->so_timeo);
1575 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1577 int error, i, m, n, pcb_count;
1578 struct inpcb *inp, **inp_list;
1583 * The process of preparing the TCB list is too time-consuming and
1584 * resource-intensive to repeat twice on every request.
1586 if (req->oldptr == NULL) {
1587 n = V_tcbinfo.ipi_count + syncache_pcbcount();
1588 n += imax(n / 8, 10);
1589 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1593 if (req->newptr != NULL)
1597 * OK, now we're committed to doing something.
1599 INP_LIST_RLOCK(&V_tcbinfo);
1600 gencnt = V_tcbinfo.ipi_gencnt;
1601 n = V_tcbinfo.ipi_count;
1602 INP_LIST_RUNLOCK(&V_tcbinfo);
1604 m = syncache_pcbcount();
1606 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1607 + (n + m) * sizeof(struct xtcpcb));
1611 xig.xig_len = sizeof xig;
1612 xig.xig_count = n + m;
1613 xig.xig_gen = gencnt;
1614 xig.xig_sogen = so_gencnt;
1615 error = SYSCTL_OUT(req, &xig, sizeof xig);
1619 error = syncache_pcblist(req, m, &pcb_count);
1623 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1624 if (inp_list == NULL)
1627 INP_INFO_WLOCK(&V_tcbinfo);
1628 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1629 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1631 if (inp->inp_gencnt <= gencnt) {
1633 * XXX: This use of cr_cansee(), introduced with
1634 * TCP state changes, is not quite right, but for
1635 * now, better than nothing.
1637 if (inp->inp_flags & INP_TIMEWAIT) {
1638 if (intotw(inp) != NULL)
1639 error = cr_cansee(req->td->td_ucred,
1640 intotw(inp)->tw_cred);
1642 error = EINVAL; /* Skip this inp. */
1644 error = cr_canseeinpcb(req->td->td_ucred, inp);
1647 inp_list[i++] = inp;
1652 INP_INFO_WUNLOCK(&V_tcbinfo);
1656 for (i = 0; i < n; i++) {
1659 if (inp->inp_gencnt <= gencnt) {
1663 bzero(&xt, sizeof(xt));
1664 xt.xt_len = sizeof xt;
1665 /* XXX should avoid extra copy */
1666 bcopy(inp, &xt.xt_inp, sizeof *inp);
1667 inp_ppcb = inp->inp_ppcb;
1668 if (inp_ppcb == NULL)
1669 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1670 else if (inp->inp_flags & INP_TIMEWAIT) {
1671 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1672 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1674 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1675 if (xt.xt_tp.t_timers)
1676 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1678 if (inp->inp_socket != NULL)
1679 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1681 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1682 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1684 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1686 error = SYSCTL_OUT(req, &xt, sizeof xt);
1690 INP_INFO_RLOCK(&V_tcbinfo);
1691 for (i = 0; i < n; i++) {
1694 if (!in_pcbrele_rlocked(inp))
1697 INP_INFO_RUNLOCK(&V_tcbinfo);
1701 * Give the user an updated idea of our state.
1702 * If the generation differs from what we told
1703 * her before, she knows that something happened
1704 * while we were processing this request, and it
1705 * might be necessary to retry.
1707 INP_LIST_RLOCK(&V_tcbinfo);
1708 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1709 xig.xig_sogen = so_gencnt;
1710 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1711 INP_LIST_RUNLOCK(&V_tcbinfo);
1712 error = SYSCTL_OUT(req, &xig, sizeof xig);
1714 free(inp_list, M_TEMP);
1718 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1719 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1720 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1724 tcp_getcred(SYSCTL_HANDLER_ARGS)
1727 struct sockaddr_in addrs[2];
1731 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1734 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1737 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1738 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1740 if (inp->inp_socket == NULL)
1743 error = cr_canseeinpcb(req->td->td_ucred, inp);
1745 cru2x(inp->inp_cred, &xuc);
1750 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1754 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1755 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1756 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1761 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1764 struct sockaddr_in6 addrs[2];
1771 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1774 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1777 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1778 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1781 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1783 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1792 inp = in_pcblookup(&V_tcbinfo,
1793 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1795 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1796 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1799 inp = in6_pcblookup(&V_tcbinfo,
1800 &addrs[1].sin6_addr, addrs[1].sin6_port,
1801 &addrs[0].sin6_addr, addrs[0].sin6_port,
1802 INPLOOKUP_RLOCKPCB, NULL);
1804 if (inp->inp_socket == NULL)
1807 error = cr_canseeinpcb(req->td->td_ucred, inp);
1809 cru2x(inp->inp_cred, &xuc);
1814 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1818 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1819 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1820 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1826 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1828 struct ip *ip = vip;
1830 struct in_addr faddr;
1833 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1835 struct in_conninfo inc;
1836 tcp_seq icmp_tcp_seq;
1839 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1840 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1843 if (cmd == PRC_MSGSIZE)
1844 notify = tcp_mtudisc_notify;
1845 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1846 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1847 notify = tcp_drop_syn_sent;
1849 * Redirects don't need to be handled up here.
1851 else if (PRC_IS_REDIRECT(cmd))
1854 * Hostdead is ugly because it goes linearly through all PCBs.
1855 * XXX: We never get this from ICMP, otherwise it makes an
1856 * excellent DoS attack on machines with many connections.
1858 else if (cmd == PRC_HOSTDEAD)
1860 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1864 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1868 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1869 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1870 INP_INFO_RLOCK(&V_tcbinfo);
1871 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
1872 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1874 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1875 !(inp->inp_flags & INP_DROPPED) &&
1876 !(inp->inp_socket == NULL)) {
1877 icmp_tcp_seq = ntohl(th->th_seq);
1878 tp = intotcpcb(inp);
1879 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1880 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1881 if (cmd == PRC_MSGSIZE) {
1884 * If we got a needfrag set the MTU
1885 * in the route to the suggested new
1886 * value (if given) and then notify.
1888 mtu = ntohs(icp->icmp_nextmtu);
1890 * If no alternative MTU was
1891 * proposed, try the next smaller
1896 ntohs(ip->ip_len), 1);
1897 if (mtu < V_tcp_minmss +
1898 sizeof(struct tcpiphdr))
1899 mtu = V_tcp_minmss +
1900 sizeof(struct tcpiphdr);
1902 * Only process the offered MTU if it
1903 * is smaller than the current one.
1905 if (mtu < tp->t_maxseg +
1906 sizeof(struct tcpiphdr)) {
1907 bzero(&inc, sizeof(inc));
1908 inc.inc_faddr = faddr;
1910 inp->inp_inc.inc_fibnum;
1911 tcp_hc_updatemtu(&inc, mtu);
1912 tcp_mtudisc(inp, mtu);
1915 inp = (*notify)(inp,
1916 inetctlerrmap[cmd]);
1922 bzero(&inc, sizeof(inc));
1923 inc.inc_fport = th->th_dport;
1924 inc.inc_lport = th->th_sport;
1925 inc.inc_faddr = faddr;
1926 inc.inc_laddr = ip->ip_src;
1927 syncache_unreach(&inc, th);
1929 INP_INFO_RUNLOCK(&V_tcbinfo);
1935 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1938 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1939 struct ip6_hdr *ip6;
1941 struct ip6ctlparam *ip6cp = NULL;
1942 const struct sockaddr_in6 *sa6_src = NULL;
1944 struct tcp_portonly {
1949 if (sa->sa_family != AF_INET6 ||
1950 sa->sa_len != sizeof(struct sockaddr_in6))
1953 if (cmd == PRC_MSGSIZE)
1954 notify = tcp_mtudisc_notify;
1955 else if (!PRC_IS_REDIRECT(cmd) &&
1956 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1959 /* if the parameter is from icmp6, decode it. */
1961 ip6cp = (struct ip6ctlparam *)d;
1963 ip6 = ip6cp->ip6c_ip6;
1964 off = ip6cp->ip6c_off;
1965 sa6_src = ip6cp->ip6c_src;
1969 off = 0; /* fool gcc */
1974 struct in_conninfo inc;
1976 * XXX: We assume that when IPV6 is non NULL,
1977 * M and OFF are valid.
1980 /* check if we can safely examine src and dst ports */
1981 if (m->m_pkthdr.len < off + sizeof(*thp))
1984 bzero(&th, sizeof(th));
1985 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1987 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1988 (struct sockaddr *)ip6cp->ip6c_src,
1989 th.th_sport, cmd, NULL, notify);
1991 bzero(&inc, sizeof(inc));
1992 inc.inc_fport = th.th_dport;
1993 inc.inc_lport = th.th_sport;
1994 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1995 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1996 inc.inc_flags |= INC_ISIPV6;
1997 INP_INFO_RLOCK(&V_tcbinfo);
1998 syncache_unreach(&inc, &th);
1999 INP_INFO_RUNLOCK(&V_tcbinfo);
2001 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
2002 0, cmd, NULL, notify);
2008 * Following is where TCP initial sequence number generation occurs.
2010 * There are two places where we must use initial sequence numbers:
2011 * 1. In SYN-ACK packets.
2012 * 2. In SYN packets.
2014 * All ISNs for SYN-ACK packets are generated by the syncache. See
2015 * tcp_syncache.c for details.
2017 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2018 * depends on this property. In addition, these ISNs should be
2019 * unguessable so as to prevent connection hijacking. To satisfy
2020 * the requirements of this situation, the algorithm outlined in
2021 * RFC 1948 is used, with only small modifications.
2023 * Implementation details:
2025 * Time is based off the system timer, and is corrected so that it
2026 * increases by one megabyte per second. This allows for proper
2027 * recycling on high speed LANs while still leaving over an hour
2030 * As reading the *exact* system time is too expensive to be done
2031 * whenever setting up a TCP connection, we increment the time
2032 * offset in two ways. First, a small random positive increment
2033 * is added to isn_offset for each connection that is set up.
2034 * Second, the function tcp_isn_tick fires once per clock tick
2035 * and increments isn_offset as necessary so that sequence numbers
2036 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2037 * random positive increments serve only to ensure that the same
2038 * exact sequence number is never sent out twice (as could otherwise
2039 * happen when a port is recycled in less than the system tick
2042 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2043 * between seeding of isn_secret. This is normally set to zero,
2044 * as reseeding should not be necessary.
2046 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2047 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
2048 * general, this means holding an exclusive (write) lock.
2051 #define ISN_BYTES_PER_SECOND 1048576
2052 #define ISN_STATIC_INCREMENT 4096
2053 #define ISN_RANDOM_INCREMENT (4096 - 1)
2055 static VNET_DEFINE(u_char, isn_secret[32]);
2056 static VNET_DEFINE(int, isn_last);
2057 static VNET_DEFINE(int, isn_last_reseed);
2058 static VNET_DEFINE(u_int32_t, isn_offset);
2059 static VNET_DEFINE(u_int32_t, isn_offset_old);
2061 #define V_isn_secret VNET(isn_secret)
2062 #define V_isn_last VNET(isn_last)
2063 #define V_isn_last_reseed VNET(isn_last_reseed)
2064 #define V_isn_offset VNET(isn_offset)
2065 #define V_isn_offset_old VNET(isn_offset_old)
2068 tcp_new_isn(struct tcpcb *tp)
2071 u_int32_t md5_buffer[4];
2073 u_int32_t projected_offset;
2075 INP_WLOCK_ASSERT(tp->t_inpcb);
2078 /* Seed if this is the first use, reseed if requested. */
2079 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2080 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2082 read_random(&V_isn_secret, sizeof(V_isn_secret));
2083 V_isn_last_reseed = ticks;
2086 /* Compute the md5 hash and return the ISN. */
2088 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2089 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2091 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2092 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2093 sizeof(struct in6_addr));
2094 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2095 sizeof(struct in6_addr));
2099 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2100 sizeof(struct in_addr));
2101 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2102 sizeof(struct in_addr));
2104 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2105 MD5Final((u_char *) &md5_buffer, &isn_ctx);
2106 new_isn = (tcp_seq) md5_buffer[0];
2107 V_isn_offset += ISN_STATIC_INCREMENT +
2108 (arc4random() & ISN_RANDOM_INCREMENT);
2109 if (ticks != V_isn_last) {
2110 projected_offset = V_isn_offset_old +
2111 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2112 if (SEQ_GT(projected_offset, V_isn_offset))
2113 V_isn_offset = projected_offset;
2114 V_isn_offset_old = V_isn_offset;
2117 new_isn += V_isn_offset;
2123 * When a specific ICMP unreachable message is received and the
2124 * connection state is SYN-SENT, drop the connection. This behavior
2125 * is controlled by the icmp_may_rst sysctl.
2128 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2132 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2133 INP_WLOCK_ASSERT(inp);
2135 if ((inp->inp_flags & INP_TIMEWAIT) ||
2136 (inp->inp_flags & INP_DROPPED))
2139 tp = intotcpcb(inp);
2140 if (tp->t_state != TCPS_SYN_SENT)
2143 tp = tcp_drop(tp, errno);
2151 * When `need fragmentation' ICMP is received, update our idea of the MSS
2152 * based on the new value. Also nudge TCP to send something, since we
2153 * know the packet we just sent was dropped.
2154 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2156 static struct inpcb *
2157 tcp_mtudisc_notify(struct inpcb *inp, int error)
2160 tcp_mtudisc(inp, -1);
2165 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2170 INP_WLOCK_ASSERT(inp);
2171 if ((inp->inp_flags & INP_TIMEWAIT) ||
2172 (inp->inp_flags & INP_DROPPED))
2175 tp = intotcpcb(inp);
2176 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2178 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2180 so = inp->inp_socket;
2181 SOCKBUF_LOCK(&so->so_snd);
2182 /* If the mss is larger than the socket buffer, decrease the mss. */
2183 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2184 tp->t_maxseg = so->so_snd.sb_hiwat;
2185 SOCKBUF_UNLOCK(&so->so_snd);
2187 TCPSTAT_INC(tcps_mturesent);
2189 tp->snd_nxt = tp->snd_una;
2190 tcp_free_sackholes(tp);
2191 tp->snd_recover = tp->snd_max;
2192 if (tp->t_flags & TF_SACK_PERMIT)
2193 EXIT_FASTRECOVERY(tp->t_flags);
2194 tp->t_fb->tfb_tcp_output(tp);
2199 * Look-up the routing entry to the peer of this inpcb. If no route
2200 * is found and it cannot be allocated, then return 0. This routine
2201 * is called by TCP routines that access the rmx structure and by
2202 * tcp_mss_update to get the peer/interface MTU.
2205 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2208 struct sockaddr_in *dst;
2212 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2214 bzero(&sro, sizeof(sro));
2215 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2216 dst = (struct sockaddr_in *)&sro.ro_dst;
2217 dst->sin_family = AF_INET;
2218 dst->sin_len = sizeof(*dst);
2219 dst->sin_addr = inc->inc_faddr;
2220 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
2222 if (sro.ro_rt != NULL) {
2223 ifp = sro.ro_rt->rt_ifp;
2224 if (sro.ro_rt->rt_mtu == 0)
2225 maxmtu = ifp->if_mtu;
2227 maxmtu = min(sro.ro_rt->rt_mtu, ifp->if_mtu);
2229 /* Report additional interface capabilities. */
2231 if (ifp->if_capenable & IFCAP_TSO4 &&
2232 ifp->if_hwassist & CSUM_TSO) {
2233 cap->ifcap |= CSUM_TSO;
2234 cap->tsomax = ifp->if_hw_tsomax;
2235 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2236 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2247 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2249 struct route_in6 sro6;
2253 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2255 bzero(&sro6, sizeof(sro6));
2256 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2257 sro6.ro_dst.sin6_family = AF_INET6;
2258 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
2259 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
2260 in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
2262 if (sro6.ro_rt != NULL) {
2263 ifp = sro6.ro_rt->rt_ifp;
2264 if (sro6.ro_rt->rt_mtu == 0)
2265 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
2267 maxmtu = min(sro6.ro_rt->rt_mtu,
2268 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
2270 /* Report additional interface capabilities. */
2272 if (ifp->if_capenable & IFCAP_TSO6 &&
2273 ifp->if_hwassist & CSUM_TSO) {
2274 cap->ifcap |= CSUM_TSO;
2275 cap->tsomax = ifp->if_hw_tsomax;
2276 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2277 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2288 * Calculate effective SMSS per RFC5681 definition for a given TCP
2289 * connection at its current state, taking into account SACK and etc.
2292 tcp_maxseg(const struct tcpcb *tp)
2296 if (tp->t_flags & TF_NOOPT)
2297 return (tp->t_maxseg);
2300 * Here we have a simplified code from tcp_addoptions(),
2301 * without a proper loop, and having most of paddings hardcoded.
2302 * We might make mistakes with padding here in some edge cases,
2303 * but this is harmless, since result of tcp_maxseg() is used
2304 * only in cwnd and ssthresh estimations.
2306 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2307 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2308 if (tp->t_flags & TF_RCVD_TSTMP)
2309 optlen = TCPOLEN_TSTAMP_APPA;
2312 #ifdef TCP_SIGNATURE
2313 if (tp->t_flags & TF_SIGNATURE)
2314 optlen += PAD(TCPOLEN_SIGNATURE);
2316 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2317 optlen += TCPOLEN_SACKHDR;
2318 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2319 optlen = PAD(optlen);
2322 if (tp->t_flags & TF_REQ_TSTMP)
2323 optlen = TCPOLEN_TSTAMP_APPA;
2325 optlen = PAD(TCPOLEN_MAXSEG);
2326 if (tp->t_flags & TF_REQ_SCALE)
2327 optlen += PAD(TCPOLEN_WINDOW);
2328 #ifdef TCP_SIGNATURE
2329 if (tp->t_flags & TF_SIGNATURE)
2330 optlen += PAD(TCPOLEN_SIGNATURE);
2332 if (tp->t_flags & TF_SACK_PERMIT)
2333 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2336 optlen = min(optlen, TCP_MAXOLEN);
2337 return (tp->t_maxseg - optlen);
2341 /* compute ESP/AH header size for TCP, including outer IP header. */
2343 ipsec_hdrsiz_tcp(struct tcpcb *tp)
2350 struct ip6_hdr *ip6;
2354 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) ||
2355 (!key_havesp(IPSEC_DIR_OUTBOUND)))
2357 m = m_gethdr(M_NOWAIT, MT_DATA);
2362 if ((inp->inp_vflag & INP_IPV6) != 0) {
2363 ip6 = mtod(m, struct ip6_hdr *);
2364 th = (struct tcphdr *)(ip6 + 1);
2365 m->m_pkthdr.len = m->m_len =
2366 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
2367 tcpip_fillheaders(inp, ip6, th);
2368 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2372 ip = mtod(m, struct ip *);
2373 th = (struct tcphdr *)(ip + 1);
2374 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
2375 tcpip_fillheaders(inp, ip, th);
2376 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2384 #ifdef TCP_SIGNATURE
2386 * Callback function invoked by m_apply() to digest TCP segment data
2387 * contained within an mbuf chain.
2390 tcp_signature_apply(void *fstate, void *data, u_int len)
2393 MD5Update(fstate, (u_char *)data, len);
2398 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2399 * search with the destination IP address, and a 'magic SPI' to be
2400 * determined by the application. This is hardcoded elsewhere to 1179
2403 tcp_get_sav(struct mbuf *m, u_int direction)
2405 union sockaddr_union dst;
2406 struct secasvar *sav;
2409 struct ip6_hdr *ip6;
2410 char ip6buf[INET6_ADDRSTRLEN];
2413 /* Extract the destination from the IP header in the mbuf. */
2414 bzero(&dst, sizeof(union sockaddr_union));
2415 ip = mtod(m, struct ip *);
2417 ip6 = NULL; /* Make the compiler happy. */
2422 dst.sa.sa_len = sizeof(struct sockaddr_in);
2423 dst.sa.sa_family = AF_INET;
2424 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2425 ip->ip_src : ip->ip_dst;
2429 case (IPV6_VERSION >> 4):
2430 ip6 = mtod(m, struct ip6_hdr *);
2431 dst.sa.sa_len = sizeof(struct sockaddr_in6);
2432 dst.sa.sa_family = AF_INET6;
2433 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2434 ip6->ip6_src : ip6->ip6_dst;
2443 /* Look up an SADB entry which matches the address of the peer. */
2444 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2446 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2447 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2449 (ip->ip_v == (IPV6_VERSION >> 4)) ?
2450 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2459 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2462 * m pointer to head of mbuf chain
2463 * len length of TCP segment data, excluding options
2464 * optlen length of TCP segment options
2465 * buf pointer to storage for computed MD5 digest
2466 * sav pointer to security assosiation
2468 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2469 * When called from tcp_input(), we can be sure that th_sum has been
2470 * zeroed out and verified already.
2472 * Releases reference to SADB key before return.
2474 * Return 0 if successful, otherwise return -1.
2478 tcp_signature_do_compute(struct mbuf *m, int len, int optlen,
2479 u_char *buf, struct secasvar *sav)
2482 struct ippseudo ippseudo;
2488 struct ipovly *ipovly;
2492 struct ip6_hdr *ip6;
2493 struct in6_addr in6;
2499 KASSERT(m != NULL, ("NULL mbuf chain"));
2500 KASSERT(buf != NULL, ("NULL signature pointer"));
2502 /* Extract the destination from the IP header in the mbuf. */
2503 ip = mtod(m, struct ip *);
2505 ip6 = NULL; /* Make the compiler happy. */
2510 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2512 * XXX The ippseudo header MUST be digested in network byte order,
2513 * or else we'll fail the regression test. Assume all fields we've
2514 * been doing arithmetic on have been in host byte order.
2515 * XXX One cannot depend on ipovly->ih_len here. When called from
2516 * tcp_output(), the underlying ip_len member has not yet been set.
2521 ipovly = (struct ipovly *)ip;
2522 ippseudo.ippseudo_src = ipovly->ih_src;
2523 ippseudo.ippseudo_dst = ipovly->ih_dst;
2524 ippseudo.ippseudo_pad = 0;
2525 ippseudo.ippseudo_p = IPPROTO_TCP;
2526 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2528 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2530 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2531 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2536 * RFC 2385, 2.0 Proposal
2537 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2538 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2539 * extended next header value (to form 32 bits), and 32-bit segment
2541 * Note: Upper-Layer Packet Length comes before Next Header.
2543 case (IPV6_VERSION >> 4):
2545 in6_clearscope(&in6);
2546 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2548 in6_clearscope(&in6);
2549 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2550 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2551 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2553 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2554 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2555 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2557 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2559 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2560 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2572 * Step 2: Update MD5 hash with TCP header, excluding options.
2573 * The TCP checksum must be set to zero.
2575 savecsum = th->th_sum;
2577 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2578 th->th_sum = savecsum;
2581 * Step 3: Update MD5 hash with TCP segment data.
2582 * Use m_apply() to avoid an early m_pullup().
2585 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2588 * Step 4: Update MD5 hash with shared secret.
2590 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2591 MD5Final(buf, &ctx);
2593 key_sa_recordxfer(sav, m);
2599 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2601 * Return 0 if successful, otherwise return -1.
2604 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2605 u_char *buf, u_int direction)
2607 struct secasvar *sav;
2609 if ((sav = tcp_get_sav(m, direction)) == NULL)
2612 return (tcp_signature_do_compute(m, len, optlen, buf, sav));
2616 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2619 * m pointer to head of mbuf chain
2620 * len length of TCP segment data, excluding options
2621 * optlen length of TCP segment options
2622 * buf pointer to storage for computed MD5 digest
2623 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2625 * Return 1 if successful, otherwise return 0.
2628 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2629 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2631 char tmpdigest[TCP_SIGLEN];
2633 if (tcp_sig_checksigs == 0)
2635 if ((tcpbflag & TF_SIGNATURE) == 0) {
2636 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2639 * If this socket is not expecting signature but
2640 * the segment contains signature just fail.
2642 TCPSTAT_INC(tcps_sig_err_sigopt);
2643 TCPSTAT_INC(tcps_sig_rcvbadsig);
2647 /* Signature is not expected, and not present in segment. */
2652 * If this socket is expecting signature but the segment does not
2653 * contain any just fail.
2655 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2656 TCPSTAT_INC(tcps_sig_err_nosigopt);
2657 TCPSTAT_INC(tcps_sig_rcvbadsig);
2660 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2661 IPSEC_DIR_INBOUND) == -1) {
2662 TCPSTAT_INC(tcps_sig_err_buildsig);
2663 TCPSTAT_INC(tcps_sig_rcvbadsig);
2667 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2668 TCPSTAT_INC(tcps_sig_rcvbadsig);
2671 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2674 #endif /* TCP_SIGNATURE */
2677 sysctl_drop(SYSCTL_HANDLER_ARGS)
2679 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2680 struct sockaddr_storage addrs[2];
2684 struct sockaddr_in *fin, *lin;
2686 struct sockaddr_in6 *fin6, *lin6;
2697 if (req->oldptr != NULL || req->oldlen != 0)
2699 if (req->newptr == NULL)
2701 if (req->newlen < sizeof(addrs))
2703 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2707 switch (addrs[0].ss_family) {
2710 fin6 = (struct sockaddr_in6 *)&addrs[0];
2711 lin6 = (struct sockaddr_in6 *)&addrs[1];
2712 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2713 lin6->sin6_len != sizeof(struct sockaddr_in6))
2715 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2716 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2718 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2719 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2720 fin = (struct sockaddr_in *)&addrs[0];
2721 lin = (struct sockaddr_in *)&addrs[1];
2724 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2727 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2734 fin = (struct sockaddr_in *)&addrs[0];
2735 lin = (struct sockaddr_in *)&addrs[1];
2736 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2737 lin->sin_len != sizeof(struct sockaddr_in))
2744 INP_INFO_RLOCK(&V_tcbinfo);
2745 switch (addrs[0].ss_family) {
2748 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2749 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2750 INPLOOKUP_WLOCKPCB, NULL);
2755 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2756 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2761 if (inp->inp_flags & INP_TIMEWAIT) {
2763 * XXXRW: There currently exists a state where an
2764 * inpcb is present, but its timewait state has been
2765 * discarded. For now, don't allow dropping of this
2773 } else if (!(inp->inp_flags & INP_DROPPED) &&
2774 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2775 tp = intotcpcb(inp);
2776 tp = tcp_drop(tp, ECONNABORTED);
2783 INP_INFO_RUNLOCK(&V_tcbinfo);
2787 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2788 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2789 0, sysctl_drop, "", "Drop TCP connection");
2792 * Generate a standardized TCP log line for use throughout the
2793 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2794 * allow use in the interrupt context.
2796 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2797 * NB: The function may return NULL if memory allocation failed.
2799 * Due to header inclusion and ordering limitations the struct ip
2800 * and ip6_hdr pointers have to be passed as void pointers.
2803 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2807 /* Is logging enabled? */
2808 if (tcp_log_in_vain == 0)
2811 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2815 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2819 /* Is logging enabled? */
2820 if (tcp_log_debug == 0)
2823 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2827 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2834 const struct ip6_hdr *ip6;
2836 ip6 = (const struct ip6_hdr *)ip6hdr;
2838 ip = (struct ip *)ip4hdr;
2841 * The log line looks like this:
2842 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2844 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2845 sizeof(PRINT_TH_FLAGS) + 1 +
2847 2 * INET6_ADDRSTRLEN;
2849 2 * INET_ADDRSTRLEN;
2852 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2856 strcat(s, "TCP: [");
2859 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2860 inet_ntoa_r(inc->inc_faddr, sp);
2862 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2864 inet_ntoa_r(inc->inc_laddr, sp);
2866 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2869 ip6_sprintf(sp, &inc->inc6_faddr);
2871 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2873 ip6_sprintf(sp, &inc->inc6_laddr);
2875 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2876 } else if (ip6 && th) {
2877 ip6_sprintf(sp, &ip6->ip6_src);
2879 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2881 ip6_sprintf(sp, &ip6->ip6_dst);
2883 sprintf(sp, "]:%i", ntohs(th->th_dport));
2886 } else if (ip && th) {
2887 inet_ntoa_r(ip->ip_src, sp);
2889 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2891 inet_ntoa_r(ip->ip_dst, sp);
2893 sprintf(sp, "]:%i", ntohs(th->th_dport));
2901 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2902 if (*(s + size - 1) != '\0')
2903 panic("%s: string too long", __func__);
2908 * A subroutine which makes it easy to track TCP state changes with DTrace.
2909 * This function shouldn't be called for t_state initializations that don't
2910 * correspond to actual TCP state transitions.
2913 tcp_state_change(struct tcpcb *tp, int newstate)
2915 #if defined(KDTRACE_HOOKS)
2916 int pstate = tp->t_state;
2919 tp->t_state = newstate;
2920 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);