2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
35 #include "opt_compat.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_tcpdebug.h"
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/callout.h>
44 #include <sys/eventhandler.h>
45 #include <sys/hhook.h>
46 #include <sys/kernel.h>
47 #include <sys/khelp.h>
48 #include <sys/sysctl.h>
50 #include <sys/malloc.h>
51 #include <sys/refcount.h>
54 #include <sys/domain.h>
59 #include <sys/socket.h>
60 #include <sys/socketvar.h>
61 #include <sys/protosw.h>
62 #include <sys/random.h>
66 #include <net/route.h>
68 #include <net/if_var.h>
71 #include <netinet/in.h>
72 #include <netinet/in_fib.h>
73 #include <netinet/in_kdtrace.h>
74 #include <netinet/in_pcb.h>
75 #include <netinet/in_systm.h>
76 #include <netinet/in_var.h>
77 #include <netinet/ip.h>
78 #include <netinet/ip_icmp.h>
79 #include <netinet/ip_var.h>
81 #include <netinet/ip6.h>
82 #include <netinet6/in6_fib.h>
83 #include <netinet6/in6_pcb.h>
84 #include <netinet6/ip6_var.h>
85 #include <netinet6/scope6_var.h>
86 #include <netinet6/nd6.h>
90 #include <netinet/tcp_fastopen.h>
92 #include <netinet/tcp.h>
93 #include <netinet/tcp_fsm.h>
94 #include <netinet/tcp_seq.h>
95 #include <netinet/tcp_timer.h>
96 #include <netinet/tcp_var.h>
97 #include <netinet/tcp_syncache.h>
98 #include <netinet/cc/cc.h>
100 #include <netinet6/tcp6_var.h>
102 #include <netinet/tcpip.h>
104 #include <netinet/tcp_pcap.h>
107 #include <netinet/tcp_debug.h>
110 #include <netinet6/ip6protosw.h>
113 #include <netinet/tcp_offload.h>
117 #include <netipsec/ipsec.h>
118 #include <netipsec/xform.h>
120 #include <netipsec/ipsec6.h>
122 #include <netipsec/key.h>
123 #include <sys/syslog.h>
126 #include <machine/in_cksum.h>
129 #include <security/mac/mac_framework.h>
131 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
133 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
136 struct rwlock tcp_function_lock;
139 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
144 error = sysctl_handle_int(oidp, &new, 0, req);
145 if (error == 0 && req->newptr) {
146 if (new < TCP_MINMSS)
154 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
155 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
156 &sysctl_net_inet_tcp_mss_check, "I",
157 "Default TCP Maximum Segment Size");
161 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
165 new = V_tcp_v6mssdflt;
166 error = sysctl_handle_int(oidp, &new, 0, req);
167 if (error == 0 && req->newptr) {
168 if (new < TCP_MINMSS)
171 V_tcp_v6mssdflt = new;
176 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
177 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
178 &sysctl_net_inet_tcp_mss_v6_check, "I",
179 "Default TCP Maximum Segment Size for IPv6");
183 * Minimum MSS we accept and use. This prevents DoS attacks where
184 * we are forced to a ridiculous low MSS like 20 and send hundreds
185 * of packets instead of one. The effect scales with the available
186 * bandwidth and quickly saturates the CPU and network interface
187 * with packet generation and sending. Set to zero to disable MINMSS
188 * checking. This setting prevents us from sending too small packets.
190 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
192 &VNET_NAME(tcp_minmss), 0,
193 "Minimum TCP Maximum Segment Size");
195 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
196 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
197 &VNET_NAME(tcp_do_rfc1323), 0,
198 "Enable rfc1323 (high performance TCP) extensions");
200 static int tcp_log_debug = 0;
201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
202 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
204 static int tcp_tcbhashsize;
205 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
206 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
208 static int do_tcpdrain = 1;
209 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
210 "Enable tcp_drain routine for extra help when low on mbufs");
212 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
213 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
215 static VNET_DEFINE(int, icmp_may_rst) = 1;
216 #define V_icmp_may_rst VNET(icmp_may_rst)
217 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
218 &VNET_NAME(icmp_may_rst), 0,
219 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
221 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
222 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
223 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
224 &VNET_NAME(tcp_isn_reseed_interval), 0,
225 "Seconds between reseeding of ISN secret");
227 static int tcp_soreceive_stream;
228 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
229 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
232 static int tcp_sig_checksigs = 1;
233 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
234 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
237 VNET_DEFINE(uma_zone_t, sack_hole_zone);
238 #define V_sack_hole_zone VNET(sack_hole_zone)
240 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
242 static struct inpcb *tcp_notify(struct inpcb *, int);
243 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
244 static void tcp_mtudisc(struct inpcb *, int);
245 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
246 void *ip4hdr, const void *ip6hdr);
247 static void tcp_timer_discard(struct tcpcb *, uint32_t);
250 static struct tcp_function_block tcp_def_funcblk = {
254 tcp_default_ctloutput,
266 int t_functions_inited = 0;
267 struct tcp_funchead t_functions;
268 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
271 init_tcp_functions(void)
273 if (t_functions_inited == 0) {
274 TAILQ_INIT(&t_functions);
275 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
276 t_functions_inited = 1;
280 static struct tcp_function_block *
281 find_tcp_functions_locked(struct tcp_function_set *fs)
283 struct tcp_function *f;
284 struct tcp_function_block *blk=NULL;
286 TAILQ_FOREACH(f, &t_functions, tf_next) {
287 if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) {
295 static struct tcp_function_block *
296 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
298 struct tcp_function_block *rblk=NULL;
299 struct tcp_function *f;
301 TAILQ_FOREACH(f, &t_functions, tf_next) {
302 if (f->tf_fb == blk) {
313 struct tcp_function_block *
314 find_and_ref_tcp_functions(struct tcp_function_set *fs)
316 struct tcp_function_block *blk;
318 rw_rlock(&tcp_function_lock);
319 blk = find_tcp_functions_locked(fs);
321 refcount_acquire(&blk->tfb_refcnt);
322 rw_runlock(&tcp_function_lock);
326 struct tcp_function_block *
327 find_and_ref_tcp_fb(struct tcp_function_block *blk)
329 struct tcp_function_block *rblk;
331 rw_rlock(&tcp_function_lock);
332 rblk = find_tcp_fb_locked(blk, NULL);
334 refcount_acquire(&rblk->tfb_refcnt);
335 rw_runlock(&tcp_function_lock);
341 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
344 struct tcp_function_set fs;
345 struct tcp_function_block *blk;
347 memset(&fs, 0, sizeof(fs));
348 rw_rlock(&tcp_function_lock);
349 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
352 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
353 fs.pcbcnt = blk->tfb_refcnt;
355 rw_runlock(&tcp_function_lock);
356 error = sysctl_handle_string(oidp, fs.function_set_name,
357 sizeof(fs.function_set_name), req);
359 /* Check for error or no change */
360 if (error != 0 || req->newptr == NULL)
363 rw_wlock(&tcp_function_lock);
364 blk = find_tcp_functions_locked(&fs);
366 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
370 tcp_func_set_ptr = blk;
372 rw_wunlock(&tcp_function_lock);
376 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
377 CTLTYPE_STRING | CTLFLAG_RW,
378 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
379 "Set/get the default TCP functions");
382 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
384 int error, cnt, linesz;
385 struct tcp_function *f;
390 rw_rlock(&tcp_function_lock);
391 TAILQ_FOREACH(f, &t_functions, tf_next) {
394 rw_runlock(&tcp_function_lock);
396 bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1;
397 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
402 linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count");
407 rw_rlock(&tcp_function_lock);
408 TAILQ_FOREACH(f, &t_functions, tf_next) {
409 linesz = snprintf(cp, bufsz, "%-32s%c %u\n",
410 f->tf_fb->tfb_tcp_block_name,
411 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
412 f->tf_fb->tfb_refcnt);
413 if (linesz >= bufsz) {
421 rw_runlock(&tcp_function_lock);
423 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
424 free(buffer, M_TEMP);
428 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
429 CTLTYPE_STRING|CTLFLAG_RD,
430 NULL, 0, sysctl_net_inet_list_available, "A",
431 "list available TCP Function sets");
434 * Target size of TCP PCB hash tables. Must be a power of two.
436 * Note that this can be overridden by the kernel environment
437 * variable net.inet.tcp.tcbhashsize
440 #define TCBHASHSIZE 0
445 * Callouts should be moved into struct tcp directly. They are currently
446 * separate because the tcpcb structure is exported to userland for sysctl
447 * parsing purposes, which do not know about callouts.
456 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
457 #define V_tcpcb_zone VNET(tcpcb_zone)
459 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
460 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
462 static struct mtx isn_mtx;
464 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
465 #define ISN_LOCK() mtx_lock(&isn_mtx)
466 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
469 * TCP initialization.
472 tcp_zone_change(void *tag)
475 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
476 uma_zone_set_max(V_tcpcb_zone, maxsockets);
477 tcp_tw_zone_change();
481 tcp_inpcb_init(void *mem, int size, int flags)
483 struct inpcb *inp = mem;
485 INP_LOCK_INIT(inp, "inp", "tcpinp");
490 * Take a value and get the next power of 2 that doesn't overflow.
491 * Used to size the tcp_inpcb hash buckets.
494 maketcp_hashsize(int size)
500 * get the next power of 2 higher than maxsockets.
502 hashsize = 1 << fls(size);
503 /* catch overflow, and just go one power of 2 smaller */
504 if (hashsize < size) {
505 hashsize = 1 << (fls(size) - 1);
511 register_tcp_functions(struct tcp_function_block *blk, int wait)
513 struct tcp_function_block *lblk;
514 struct tcp_function *n;
515 struct tcp_function_set fs;
517 if (t_functions_inited == 0) {
518 init_tcp_functions();
520 if ((blk->tfb_tcp_output == NULL) ||
521 (blk->tfb_tcp_do_segment == NULL) ||
522 (blk->tfb_tcp_ctloutput == NULL) ||
523 (strlen(blk->tfb_tcp_block_name) == 0)) {
525 * These functions are required and you
530 if (blk->tfb_tcp_timer_stop_all ||
531 blk->tfb_tcp_timers_left ||
532 blk->tfb_tcp_timer_activate ||
533 blk->tfb_tcp_timer_active ||
534 blk->tfb_tcp_timer_stop) {
536 * If you define one timer function you
537 * must have them all.
539 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
540 (blk->tfb_tcp_timers_left == NULL) ||
541 (blk->tfb_tcp_timer_activate == NULL) ||
542 (blk->tfb_tcp_timer_active == NULL) ||
543 (blk->tfb_tcp_timer_stop == NULL)) {
547 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
552 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
553 rw_wlock(&tcp_function_lock);
554 lblk = find_tcp_functions_locked(&fs);
556 /* Duplicate name space not allowed */
557 rw_wunlock(&tcp_function_lock);
558 free(n, M_TCPFUNCTIONS);
561 refcount_init(&blk->tfb_refcnt, 0);
563 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
564 rw_wunlock(&tcp_function_lock);
569 deregister_tcp_functions(struct tcp_function_block *blk)
571 struct tcp_function_block *lblk;
572 struct tcp_function *f;
575 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
576 /* You can't un-register the default */
579 rw_wlock(&tcp_function_lock);
580 if (blk == tcp_func_set_ptr) {
581 /* You can't free the current default */
582 rw_wunlock(&tcp_function_lock);
585 if (blk->tfb_refcnt) {
586 /* Still tcb attached, mark it. */
587 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
588 rw_wunlock(&tcp_function_lock);
591 lblk = find_tcp_fb_locked(blk, &f);
594 TAILQ_REMOVE(&t_functions, f, tf_next);
596 free(f, M_TCPFUNCTIONS);
599 rw_wunlock(&tcp_function_lock);
606 const char *tcbhash_tuneable;
609 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
611 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
612 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
613 printf("%s: WARNING: unable to register helper hook\n", __func__);
614 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
615 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
616 printf("%s: WARNING: unable to register helper hook\n", __func__);
617 hashsize = TCBHASHSIZE;
618 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
621 * Auto tune the hash size based on maxsockets.
622 * A perfect hash would have a 1:1 mapping
623 * (hashsize = maxsockets) however it's been
624 * suggested that O(2) average is better.
626 hashsize = maketcp_hashsize(maxsockets / 4);
628 * Our historical default is 512,
629 * do not autotune lower than this.
633 if (bootverbose && IS_DEFAULT_VNET(curvnet))
634 printf("%s: %s auto tuned to %d\n", __func__,
635 tcbhash_tuneable, hashsize);
638 * We require a hashsize to be a power of two.
639 * Previously if it was not a power of two we would just reset it
640 * back to 512, which could be a nasty surprise if you did not notice
642 * Instead what we do is clip it to the closest power of two lower
643 * than the specified hash value.
645 if (!powerof2(hashsize)) {
646 int oldhashsize = hashsize;
648 hashsize = maketcp_hashsize(hashsize);
649 /* prevent absurdly low value */
652 printf("%s: WARNING: TCB hash size not a power of 2, "
653 "clipped from %d to %d.\n", __func__, oldhashsize,
656 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
657 "tcp_inpcb", tcp_inpcb_init, NULL, 0, IPI_HASHFIELDS_4TUPLE);
660 * These have to be type stable for the benefit of the timers.
662 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
663 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
664 uma_zone_set_max(V_tcpcb_zone, maxsockets);
665 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
671 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
672 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
673 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
675 /* Skip initialization of globals for non-default instances. */
676 if (!IS_DEFAULT_VNET(curvnet))
679 tcp_reass_global_init();
681 /* XXX virtualize those bellow? */
682 tcp_delacktime = TCPTV_DELACK;
683 tcp_keepinit = TCPTV_KEEP_INIT;
684 tcp_keepidle = TCPTV_KEEP_IDLE;
685 tcp_keepintvl = TCPTV_KEEPINTVL;
686 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
688 tcp_rexmit_min = TCPTV_MIN;
689 if (tcp_rexmit_min < 1)
691 tcp_persmin = TCPTV_PERSMIN;
692 tcp_persmax = TCPTV_PERSMAX;
693 tcp_rexmit_slop = TCPTV_CPU_VAR;
694 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
695 tcp_tcbhashsize = hashsize;
696 /* Setup the tcp function block list */
697 init_tcp_functions();
698 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
700 if (tcp_soreceive_stream) {
702 tcp_usrreqs.pru_soreceive = soreceive_stream;
705 tcp6_usrreqs.pru_soreceive = soreceive_stream;
710 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
712 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
714 if (max_protohdr < TCP_MINPROTOHDR)
715 max_protohdr = TCP_MINPROTOHDR;
716 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
718 #undef TCP_MINPROTOHDR
721 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
722 SHUTDOWN_PRI_DEFAULT);
723 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
724 EVENTHANDLER_PRI_ANY);
741 * All our processes are gone, all our sockets should be cleaned
742 * up, which means, we should be past the tcp_discardcb() calls.
743 * Sleep to let all tcpcb timers really disappear and then cleanup.
744 * Timewait will cleanup its queue and will be ready to go.
745 * XXX-BZ In theory a few ticks should be good enough to make sure
746 * the timers are all really gone. We should see if we could use a
747 * better metric here and, e.g., check a tcbcb count as an optimization?
753 in_pcbinfo_destroy(&V_tcbinfo);
754 /* tcp_discardcb() clears the sack_holes up. */
755 uma_zdestroy(V_sack_hole_zone);
756 uma_zdestroy(V_tcpcb_zone);
760 * Cannot free the zone until all tcpcbs are released as we attach
761 * the allocations to them.
763 tcp_fastopen_destroy();
766 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
768 printf("%s: WARNING: unable to deregister helper hook "
769 "type=%d, id=%d: error %d returned\n", __func__,
770 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
772 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
774 printf("%s: WARNING: unable to deregister helper hook "
775 "type=%d, id=%d: error %d returned\n", __func__,
776 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
788 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
789 * tcp_template used to store this data in mbufs, but we now recopy it out
790 * of the tcpcb each time to conserve mbufs.
793 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
795 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
797 INP_WLOCK_ASSERT(inp);
800 if ((inp->inp_vflag & INP_IPV6) != 0) {
803 ip6 = (struct ip6_hdr *)ip_ptr;
804 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
805 (inp->inp_flow & IPV6_FLOWINFO_MASK);
806 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
807 (IPV6_VERSION & IPV6_VERSION_MASK);
808 ip6->ip6_nxt = IPPROTO_TCP;
809 ip6->ip6_plen = htons(sizeof(struct tcphdr));
810 ip6->ip6_src = inp->in6p_laddr;
811 ip6->ip6_dst = inp->in6p_faddr;
814 #if defined(INET6) && defined(INET)
821 ip = (struct ip *)ip_ptr;
822 ip->ip_v = IPVERSION;
824 ip->ip_tos = inp->inp_ip_tos;
828 ip->ip_ttl = inp->inp_ip_ttl;
830 ip->ip_p = IPPROTO_TCP;
831 ip->ip_src = inp->inp_laddr;
832 ip->ip_dst = inp->inp_faddr;
835 th->th_sport = inp->inp_lport;
836 th->th_dport = inp->inp_fport;
844 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
848 * Create template to be used to send tcp packets on a connection.
849 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
850 * use for this function is in keepalives, which use tcp_respond.
853 tcpip_maketemplate(struct inpcb *inp)
857 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
860 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
865 * Send a single message to the TCP at address specified by
866 * the given TCP/IP header. If m == NULL, then we make a copy
867 * of the tcpiphdr at th and send directly to the addressed host.
868 * This is used to force keep alive messages out using the TCP
869 * template for a connection. If flags are given then we send
870 * a message back to the TCP which originated the segment th,
871 * and discard the mbuf containing it and any other attached mbufs.
873 * In any case the ack and sequence number of the transmitted
874 * segment are as specified by the parameters.
876 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
879 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
880 tcp_seq ack, tcp_seq seq, int flags)
892 int optlen, tlen, win;
895 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
898 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
905 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
906 INP_WLOCK_ASSERT(inp);
913 if (!(flags & TH_RST)) {
914 win = sbspace(&inp->inp_socket->so_rcv);
915 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
916 win = (long)TCP_MAXWIN << tp->rcv_scale;
918 if ((tp->t_flags & TF_NOOPT) == 0)
922 m = m_gethdr(M_NOWAIT, MT_DATA);
925 m->m_data += max_linkhdr;
928 bcopy((caddr_t)ip6, mtod(m, caddr_t),
929 sizeof(struct ip6_hdr));
930 ip6 = mtod(m, struct ip6_hdr *);
931 nth = (struct tcphdr *)(ip6 + 1);
935 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
936 ip = mtod(m, struct ip *);
937 nth = (struct tcphdr *)(ip + 1);
939 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
944 * XXX MRT We inherrit the FIB, which is lucky.
948 m->m_data = (caddr_t)ipgen;
949 /* m_len is set later */
950 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
953 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
954 nth = (struct tcphdr *)(ip6 + 1);
958 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
959 nth = (struct tcphdr *)(ip + 1);
963 * this is usually a case when an extension header
964 * exists between the IPv6 header and the
967 nth->th_sport = th->th_sport;
968 nth->th_dport = th->th_dport;
970 xchg(nth->th_dport, nth->th_sport, uint16_t);
976 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
978 #if defined(INET) && defined(INET6)
982 tlen = sizeof (struct tcpiphdr);
986 KASSERT(M_TRAILINGSPACE(m) >= tlen,
987 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
988 m, tlen, (long)M_TRAILINGSPACE(m)));
993 /* Make sure we have room. */
994 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
995 m->m_next = m_get(M_NOWAIT, MT_DATA);
997 optp = mtod(m->m_next, u_char *);
1002 optp = (u_char *) (nth + 1);
1008 if (tp->t_flags & TF_RCVD_TSTMP) {
1009 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1010 to.to_tsecr = tp->ts_recent;
1011 to.to_flags |= TOF_TS;
1013 #ifdef TCP_SIGNATURE
1014 /* TCP-MD5 (RFC2385). */
1015 if (tp->t_flags & TF_SIGNATURE)
1016 to.to_flags |= TOF_SIGNATURE;
1019 /* Add the options. */
1020 tlen += optlen = tcp_addoptions(&to, optp);
1022 /* Update m_len in the correct mbuf. */
1023 optm->m_len += optlen;
1029 ip6->ip6_vfc = IPV6_VERSION;
1030 ip6->ip6_nxt = IPPROTO_TCP;
1031 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1034 #if defined(INET) && defined(INET6)
1039 ip->ip_len = htons(tlen);
1040 ip->ip_ttl = V_ip_defttl;
1041 if (V_path_mtu_discovery)
1042 ip->ip_off |= htons(IP_DF);
1045 m->m_pkthdr.len = tlen;
1046 m->m_pkthdr.rcvif = NULL;
1050 * Packet is associated with a socket, so allow the
1051 * label of the response to reflect the socket label.
1053 INP_WLOCK_ASSERT(inp);
1054 mac_inpcb_create_mbuf(inp, m);
1057 * Packet is not associated with a socket, so possibly
1058 * update the label in place.
1060 mac_netinet_tcp_reply(m);
1063 nth->th_seq = htonl(seq);
1064 nth->th_ack = htonl(ack);
1066 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1067 nth->th_flags = flags;
1069 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1071 nth->th_win = htons((u_short)win);
1074 #ifdef TCP_SIGNATURE
1075 if (to.to_flags & TOF_SIGNATURE) {
1076 tcp_signature_compute(m, 0, 0, optlen, to.to_signature,
1077 IPSEC_DIR_OUTBOUND);
1081 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1084 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1085 nth->th_sum = in6_cksum_pseudo(ip6,
1086 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1087 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1091 #if defined(INET6) && defined(INET)
1096 m->m_pkthdr.csum_flags = CSUM_TCP;
1097 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1098 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1102 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1103 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1105 TCP_PROBE3(debug__output, tp, th, mtod(m, const char *));
1107 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
1110 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
1113 (void) ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1115 #if defined(INET) && defined(INET6)
1119 (void) ip_output(m, NULL, NULL, 0, NULL, inp);
1124 * Create a new TCP control block, making an
1125 * empty reassembly queue and hooking it to the argument
1126 * protocol control block. The `inp' parameter must have
1127 * come from the zone allocator set up in tcp_init().
1130 tcp_newtcpcb(struct inpcb *inp)
1132 struct tcpcb_mem *tm;
1135 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1138 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1143 /* Initialise cc_var struct for this tcpcb. */
1145 tp->ccv->type = IPPROTO_TCP;
1146 tp->ccv->ccvc.tcp = tp;
1147 rw_rlock(&tcp_function_lock);
1148 tp->t_fb = tcp_func_set_ptr;
1149 refcount_acquire(&tp->t_fb->tfb_refcnt);
1150 rw_runlock(&tcp_function_lock);
1151 if (tp->t_fb->tfb_tcp_fb_init) {
1152 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1155 * Use the current system default CC algorithm.
1158 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1159 CC_ALGO(tp) = CC_DEFAULT();
1162 if (CC_ALGO(tp)->cb_init != NULL)
1163 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1164 if (tp->t_fb->tfb_tcp_fb_fini)
1165 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1166 refcount_release(&tp->t_fb->tfb_refcnt);
1167 uma_zfree(V_tcpcb_zone, tm);
1172 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1173 if (tp->t_fb->tfb_tcp_fb_fini)
1174 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1175 refcount_release(&tp->t_fb->tfb_refcnt);
1176 uma_zfree(V_tcpcb_zone, tm);
1181 tp->t_vnet = inp->inp_vnet;
1183 tp->t_timers = &tm->tt;
1184 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
1187 isipv6 ? V_tcp_v6mssdflt :
1191 /* Set up our timeouts. */
1192 callout_init(&tp->t_timers->tt_rexmt, 1);
1193 callout_init(&tp->t_timers->tt_persist, 1);
1194 callout_init(&tp->t_timers->tt_keep, 1);
1195 callout_init(&tp->t_timers->tt_2msl, 1);
1196 callout_init(&tp->t_timers->tt_delack, 1);
1198 if (V_tcp_do_rfc1323)
1199 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1201 tp->t_flags |= TF_SACK_PERMIT;
1202 TAILQ_INIT(&tp->snd_holes);
1204 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1207 in_pcbref(inp); /* Reference for tcpcb */
1211 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1212 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1213 * reasonable initial retransmit time.
1215 tp->t_srtt = TCPTV_SRTTBASE;
1216 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1217 tp->t_rttmin = tcp_rexmit_min;
1218 tp->t_rxtcur = TCPTV_RTOBASE;
1219 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1220 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1221 tp->t_rcvtime = ticks;
1223 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1224 * because the socket may be bound to an IPv6 wildcard address,
1225 * which may match an IPv4-mapped IPv6 address.
1227 inp->inp_ip_ttl = V_ip_defttl;
1231 * Init the TCP PCAP queues.
1233 tcp_pcap_tcpcb_init(tp);
1235 return (tp); /* XXX */
1239 * Switch the congestion control algorithm back to NewReno for any active
1240 * control blocks using an algorithm which is about to go away.
1241 * This ensures the CC framework can allow the unload to proceed without leaving
1242 * any dangling pointers which would trigger a panic.
1243 * Returning non-zero would inform the CC framework that something went wrong
1244 * and it would be unsafe to allow the unload to proceed. However, there is no
1245 * way for this to occur with this implementation so we always return zero.
1248 tcp_ccalgounload(struct cc_algo *unload_algo)
1250 struct cc_algo *tmpalgo;
1253 VNET_ITERATOR_DECL(vnet_iter);
1256 * Check all active control blocks across all network stacks and change
1257 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1258 * requires cleanup code to be run, call it.
1261 VNET_FOREACH(vnet_iter) {
1262 CURVNET_SET(vnet_iter);
1263 INP_INFO_WLOCK(&V_tcbinfo);
1265 * New connections already part way through being initialised
1266 * with the CC algo we're removing will not race with this code
1267 * because the INP_INFO_WLOCK is held during initialisation. We
1268 * therefore don't enter the loop below until the connection
1269 * list has stabilised.
1271 LIST_FOREACH(inp, &V_tcb, inp_list) {
1273 /* Important to skip tcptw structs. */
1274 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1275 (tp = intotcpcb(inp)) != NULL) {
1277 * By holding INP_WLOCK here, we are assured
1278 * that the connection is not currently
1279 * executing inside the CC module's functions
1280 * i.e. it is safe to make the switch back to
1283 if (CC_ALGO(tp) == unload_algo) {
1284 tmpalgo = CC_ALGO(tp);
1285 /* NewReno does not require any init. */
1286 CC_ALGO(tp) = &newreno_cc_algo;
1287 if (tmpalgo->cb_destroy != NULL)
1288 tmpalgo->cb_destroy(tp->ccv);
1293 INP_INFO_WUNLOCK(&V_tcbinfo);
1296 VNET_LIST_RUNLOCK();
1302 * Drop a TCP connection, reporting
1303 * the specified error. If connection is synchronized,
1304 * then send a RST to peer.
1307 tcp_drop(struct tcpcb *tp, int errno)
1309 struct socket *so = tp->t_inpcb->inp_socket;
1311 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1312 INP_WLOCK_ASSERT(tp->t_inpcb);
1314 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1315 tcp_state_change(tp, TCPS_CLOSED);
1316 (void) tp->t_fb->tfb_tcp_output(tp);
1317 TCPSTAT_INC(tcps_drops);
1319 TCPSTAT_INC(tcps_conndrops);
1320 if (errno == ETIMEDOUT && tp->t_softerror)
1321 errno = tp->t_softerror;
1322 so->so_error = errno;
1323 return (tcp_close(tp));
1327 tcp_discardcb(struct tcpcb *tp)
1329 struct inpcb *inp = tp->t_inpcb;
1330 struct socket *so = inp->inp_socket;
1332 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1336 INP_WLOCK_ASSERT(inp);
1339 * Make sure that all of our timers are stopped before we delete the
1342 * If stopping a timer fails, we schedule a discard function in same
1343 * callout, and the last discard function called will take care of
1344 * deleting the tcpcb.
1346 tcp_timer_stop(tp, TT_REXMT);
1347 tcp_timer_stop(tp, TT_PERSIST);
1348 tcp_timer_stop(tp, TT_KEEP);
1349 tcp_timer_stop(tp, TT_2MSL);
1350 tcp_timer_stop(tp, TT_DELACK);
1351 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1352 /* Call the stop-all function of the methods */
1353 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1357 * If we got enough samples through the srtt filter,
1358 * save the rtt and rttvar in the routing entry.
1359 * 'Enough' is arbitrarily defined as 4 rtt samples.
1360 * 4 samples is enough for the srtt filter to converge
1361 * to within enough % of the correct value; fewer samples
1362 * and we could save a bogus rtt. The danger is not high
1363 * as tcp quickly recovers from everything.
1364 * XXX: Works very well but needs some more statistics!
1366 if (tp->t_rttupdated >= 4) {
1367 struct hc_metrics_lite metrics;
1370 bzero(&metrics, sizeof(metrics));
1372 * Update the ssthresh always when the conditions below
1373 * are satisfied. This gives us better new start value
1374 * for the congestion avoidance for new connections.
1375 * ssthresh is only set if packet loss occured on a session.
1377 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1378 * being torn down. Ideally this code would not use 'so'.
1380 ssthresh = tp->snd_ssthresh;
1381 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1383 * convert the limit from user data bytes to
1384 * packets then to packet data bytes.
1386 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1389 ssthresh *= (u_long)(tp->t_maxseg +
1391 (isipv6 ? sizeof (struct ip6_hdr) +
1392 sizeof (struct tcphdr) :
1394 sizeof (struct tcpiphdr)
1401 metrics.rmx_ssthresh = ssthresh;
1403 metrics.rmx_rtt = tp->t_srtt;
1404 metrics.rmx_rttvar = tp->t_rttvar;
1405 metrics.rmx_cwnd = tp->snd_cwnd;
1406 metrics.rmx_sendpipe = 0;
1407 metrics.rmx_recvpipe = 0;
1409 tcp_hc_update(&inp->inp_inc, &metrics);
1412 /* free the reassembly queue, if any */
1413 tcp_reass_flush(tp);
1416 /* Disconnect offload device, if any. */
1417 if (tp->t_flags & TF_TOE)
1418 tcp_offload_detach(tp);
1421 tcp_free_sackholes(tp);
1424 /* Free the TCP PCAP queues. */
1425 tcp_pcap_drain(&(tp->t_inpkts));
1426 tcp_pcap_drain(&(tp->t_outpkts));
1429 /* Allow the CC algorithm to clean up after itself. */
1430 if (CC_ALGO(tp)->cb_destroy != NULL)
1431 CC_ALGO(tp)->cb_destroy(tp->ccv);
1433 khelp_destroy_osd(tp->osd);
1436 inp->inp_ppcb = NULL;
1437 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1438 /* We own the last reference on tcpcb, let's free it. */
1439 if ((tp->t_fb->tfb_tcp_timers_left) &&
1440 (tp->t_fb->tfb_tcp_timers_left(tp))) {
1441 /* Some fb timers left running! */
1444 if (tp->t_fb->tfb_tcp_fb_fini)
1445 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1446 refcount_release(&tp->t_fb->tfb_refcnt);
1448 uma_zfree(V_tcpcb_zone, tp);
1449 released = in_pcbrele_wlocked(inp);
1450 KASSERT(!released, ("%s: inp %p should not have been released "
1451 "here", __func__, inp));
1456 tcp_timer_2msl_discard(void *xtp)
1459 tcp_timer_discard((struct tcpcb *)xtp, TT_2MSL);
1463 tcp_timer_keep_discard(void *xtp)
1466 tcp_timer_discard((struct tcpcb *)xtp, TT_KEEP);
1470 tcp_timer_persist_discard(void *xtp)
1473 tcp_timer_discard((struct tcpcb *)xtp, TT_PERSIST);
1477 tcp_timer_rexmt_discard(void *xtp)
1480 tcp_timer_discard((struct tcpcb *)xtp, TT_REXMT);
1484 tcp_timer_delack_discard(void *xtp)
1487 tcp_timer_discard((struct tcpcb *)xtp, TT_DELACK);
1491 tcp_timer_discard(struct tcpcb *tp, uint32_t timer_type)
1495 CURVNET_SET(tp->t_vnet);
1496 INP_INFO_RLOCK(&V_tcbinfo);
1498 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1501 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1502 ("%s: tcpcb has to be stopped here", __func__));
1503 KASSERT((tp->t_timers->tt_flags & timer_type) != 0,
1504 ("%s: discard callout should be running", __func__));
1505 tp->t_timers->tt_flags &= ~timer_type;
1506 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1507 /* We own the last reference on this tcpcb, let's free it. */
1508 if ((tp->t_fb->tfb_tcp_timers_left) &&
1509 (tp->t_fb->tfb_tcp_timers_left(tp))) {
1510 /* Some fb timers left running! */
1513 if (tp->t_fb->tfb_tcp_fb_fini)
1514 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1515 refcount_release(&tp->t_fb->tfb_refcnt);
1517 uma_zfree(V_tcpcb_zone, tp);
1518 if (in_pcbrele_wlocked(inp)) {
1519 INP_INFO_RUNLOCK(&V_tcbinfo);
1526 INP_INFO_RUNLOCK(&V_tcbinfo);
1531 * Attempt to close a TCP control block, marking it as dropped, and freeing
1532 * the socket if we hold the only reference.
1535 tcp_close(struct tcpcb *tp)
1537 struct inpcb *inp = tp->t_inpcb;
1540 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1541 INP_WLOCK_ASSERT(inp);
1544 if (tp->t_state == TCPS_LISTEN)
1545 tcp_offload_listen_stop(tp);
1549 * This releases the TFO pending counter resource for TFO listen
1550 * sockets as well as passively-created TFO sockets that transition
1551 * from SYN_RECEIVED to CLOSED.
1553 if (tp->t_tfo_pending) {
1554 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1555 tp->t_tfo_pending = NULL;
1559 TCPSTAT_INC(tcps_closed);
1560 TCPSTATES_DEC(tp->t_state);
1561 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1562 so = inp->inp_socket;
1563 soisdisconnected(so);
1564 if (inp->inp_flags & INP_SOCKREF) {
1565 KASSERT(so->so_state & SS_PROTOREF,
1566 ("tcp_close: !SS_PROTOREF"));
1567 inp->inp_flags &= ~INP_SOCKREF;
1571 so->so_state &= ~SS_PROTOREF;
1581 VNET_ITERATOR_DECL(vnet_iter);
1586 VNET_LIST_RLOCK_NOSLEEP();
1587 VNET_FOREACH(vnet_iter) {
1588 CURVNET_SET(vnet_iter);
1593 * Walk the tcpbs, if existing, and flush the reassembly queue,
1594 * if there is one...
1595 * XXX: The "Net/3" implementation doesn't imply that the TCP
1596 * reassembly queue should be flushed, but in a situation
1597 * where we're really low on mbufs, this is potentially
1600 INP_INFO_WLOCK(&V_tcbinfo);
1601 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1602 if (inpb->inp_flags & INP_TIMEWAIT)
1605 if ((tcpb = intotcpcb(inpb)) != NULL) {
1606 tcp_reass_flush(tcpb);
1607 tcp_clean_sackreport(tcpb);
1611 INP_INFO_WUNLOCK(&V_tcbinfo);
1614 VNET_LIST_RUNLOCK_NOSLEEP();
1618 * Notify a tcp user of an asynchronous error;
1619 * store error as soft error, but wake up user
1620 * (for now, won't do anything until can select for soft error).
1622 * Do not wake up user since there currently is no mechanism for
1623 * reporting soft errors (yet - a kqueue filter may be added).
1625 static struct inpcb *
1626 tcp_notify(struct inpcb *inp, int error)
1630 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1631 INP_WLOCK_ASSERT(inp);
1633 if ((inp->inp_flags & INP_TIMEWAIT) ||
1634 (inp->inp_flags & INP_DROPPED))
1637 tp = intotcpcb(inp);
1638 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1641 * Ignore some errors if we are hooked up.
1642 * If connection hasn't completed, has retransmitted several times,
1643 * and receives a second error, give up now. This is better
1644 * than waiting a long time to establish a connection that
1645 * can never complete.
1647 if (tp->t_state == TCPS_ESTABLISHED &&
1648 (error == EHOSTUNREACH || error == ENETUNREACH ||
1649 error == EHOSTDOWN)) {
1650 if (inp->inp_route.ro_rt) {
1651 RTFREE(inp->inp_route.ro_rt);
1652 inp->inp_route.ro_rt = (struct rtentry *)NULL;
1655 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1657 tp = tcp_drop(tp, error);
1663 tp->t_softerror = error;
1667 wakeup( &so->so_timeo);
1674 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1676 int error, i, m, n, pcb_count;
1677 struct inpcb *inp, **inp_list;
1682 * The process of preparing the TCB list is too time-consuming and
1683 * resource-intensive to repeat twice on every request.
1685 if (req->oldptr == NULL) {
1686 n = V_tcbinfo.ipi_count +
1687 counter_u64_fetch(VNET(tcps_states)[TCPS_SYN_RECEIVED]);
1688 n += imax(n / 8, 10);
1689 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1693 if (req->newptr != NULL)
1697 * OK, now we're committed to doing something.
1699 INP_LIST_RLOCK(&V_tcbinfo);
1700 gencnt = V_tcbinfo.ipi_gencnt;
1701 n = V_tcbinfo.ipi_count;
1702 INP_LIST_RUNLOCK(&V_tcbinfo);
1704 m = counter_u64_fetch(VNET(tcps_states)[TCPS_SYN_RECEIVED]);
1706 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1707 + (n + m) * sizeof(struct xtcpcb));
1711 xig.xig_len = sizeof xig;
1712 xig.xig_count = n + m;
1713 xig.xig_gen = gencnt;
1714 xig.xig_sogen = so_gencnt;
1715 error = SYSCTL_OUT(req, &xig, sizeof xig);
1719 error = syncache_pcblist(req, m, &pcb_count);
1723 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1725 INP_INFO_WLOCK(&V_tcbinfo);
1726 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1727 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1729 if (inp->inp_gencnt <= gencnt) {
1731 * XXX: This use of cr_cansee(), introduced with
1732 * TCP state changes, is not quite right, but for
1733 * now, better than nothing.
1735 if (inp->inp_flags & INP_TIMEWAIT) {
1736 if (intotw(inp) != NULL)
1737 error = cr_cansee(req->td->td_ucred,
1738 intotw(inp)->tw_cred);
1740 error = EINVAL; /* Skip this inp. */
1742 error = cr_canseeinpcb(req->td->td_ucred, inp);
1745 inp_list[i++] = inp;
1750 INP_INFO_WUNLOCK(&V_tcbinfo);
1754 for (i = 0; i < n; i++) {
1757 if (inp->inp_gencnt <= gencnt) {
1761 bzero(&xt, sizeof(xt));
1762 xt.xt_len = sizeof xt;
1763 /* XXX should avoid extra copy */
1764 bcopy(inp, &xt.xt_inp, sizeof *inp);
1765 inp_ppcb = inp->inp_ppcb;
1766 if (inp_ppcb == NULL)
1767 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1768 else if (inp->inp_flags & INP_TIMEWAIT) {
1769 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1770 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1772 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1773 if (xt.xt_tp.t_timers)
1774 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1776 if (inp->inp_socket != NULL)
1777 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1779 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1780 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1782 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1784 error = SYSCTL_OUT(req, &xt, sizeof xt);
1788 INP_INFO_RLOCK(&V_tcbinfo);
1789 for (i = 0; i < n; i++) {
1792 if (!in_pcbrele_rlocked(inp))
1795 INP_INFO_RUNLOCK(&V_tcbinfo);
1799 * Give the user an updated idea of our state.
1800 * If the generation differs from what we told
1801 * her before, she knows that something happened
1802 * while we were processing this request, and it
1803 * might be necessary to retry.
1805 INP_LIST_RLOCK(&V_tcbinfo);
1806 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1807 xig.xig_sogen = so_gencnt;
1808 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1809 INP_LIST_RUNLOCK(&V_tcbinfo);
1810 error = SYSCTL_OUT(req, &xig, sizeof xig);
1812 free(inp_list, M_TEMP);
1816 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1817 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1818 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1822 tcp_getcred(SYSCTL_HANDLER_ARGS)
1825 struct sockaddr_in addrs[2];
1829 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1832 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1835 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1836 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1838 if (inp->inp_socket == NULL)
1841 error = cr_canseeinpcb(req->td->td_ucred, inp);
1843 cru2x(inp->inp_cred, &xuc);
1848 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1852 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1853 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1854 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1859 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1862 struct sockaddr_in6 addrs[2];
1869 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1872 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1875 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1876 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1879 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1881 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1890 inp = in_pcblookup(&V_tcbinfo,
1891 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1893 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1894 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1897 inp = in6_pcblookup(&V_tcbinfo,
1898 &addrs[1].sin6_addr, addrs[1].sin6_port,
1899 &addrs[0].sin6_addr, addrs[0].sin6_port,
1900 INPLOOKUP_RLOCKPCB, NULL);
1902 if (inp->inp_socket == NULL)
1905 error = cr_canseeinpcb(req->td->td_ucred, inp);
1907 cru2x(inp->inp_cred, &xuc);
1912 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1916 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1917 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1918 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1924 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1926 struct ip *ip = vip;
1928 struct in_addr faddr;
1931 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1933 struct in_conninfo inc;
1934 tcp_seq icmp_tcp_seq;
1937 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1938 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1941 if (cmd == PRC_MSGSIZE)
1942 notify = tcp_mtudisc_notify;
1943 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1944 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1945 notify = tcp_drop_syn_sent;
1946 else if (PRC_IS_REDIRECT(cmd)) {
1947 /* signal EHOSTDOWN, as it flushes the cached route */
1948 in_pcbnotifyall(&V_tcbinfo, faddr, EHOSTDOWN, notify);
1952 * Hostdead is ugly because it goes linearly through all PCBs.
1953 * XXX: We never get this from ICMP, otherwise it makes an
1954 * excellent DoS attack on machines with many connections.
1956 else if (cmd == PRC_HOSTDEAD)
1958 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1962 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1966 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1967 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1968 INP_INFO_RLOCK(&V_tcbinfo);
1969 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
1970 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1972 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1973 !(inp->inp_flags & INP_DROPPED) &&
1974 !(inp->inp_socket == NULL)) {
1975 icmp_tcp_seq = ntohl(th->th_seq);
1976 tp = intotcpcb(inp);
1977 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1978 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1979 if (cmd == PRC_MSGSIZE) {
1982 * If we got a needfrag set the MTU
1983 * in the route to the suggested new
1984 * value (if given) and then notify.
1986 mtu = ntohs(icp->icmp_nextmtu);
1988 * If no alternative MTU was
1989 * proposed, try the next smaller
1994 ntohs(ip->ip_len), 1);
1995 if (mtu < V_tcp_minmss +
1996 sizeof(struct tcpiphdr))
1997 mtu = V_tcp_minmss +
1998 sizeof(struct tcpiphdr);
2000 * Only process the offered MTU if it
2001 * is smaller than the current one.
2003 if (mtu < tp->t_maxseg +
2004 sizeof(struct tcpiphdr)) {
2005 bzero(&inc, sizeof(inc));
2006 inc.inc_faddr = faddr;
2008 inp->inp_inc.inc_fibnum;
2009 tcp_hc_updatemtu(&inc, mtu);
2010 tcp_mtudisc(inp, mtu);
2013 inp = (*notify)(inp,
2014 inetctlerrmap[cmd]);
2020 bzero(&inc, sizeof(inc));
2021 inc.inc_fport = th->th_dport;
2022 inc.inc_lport = th->th_sport;
2023 inc.inc_faddr = faddr;
2024 inc.inc_laddr = ip->ip_src;
2025 syncache_unreach(&inc, th);
2027 INP_INFO_RUNLOCK(&V_tcbinfo);
2033 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2036 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2037 struct ip6_hdr *ip6;
2039 struct ip6ctlparam *ip6cp = NULL;
2040 const struct sockaddr_in6 *sa6_src = NULL;
2042 struct tcp_portonly {
2047 if (sa->sa_family != AF_INET6 ||
2048 sa->sa_len != sizeof(struct sockaddr_in6))
2051 if (cmd == PRC_MSGSIZE)
2052 notify = tcp_mtudisc_notify;
2053 else if (!PRC_IS_REDIRECT(cmd) &&
2054 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
2057 /* if the parameter is from icmp6, decode it. */
2059 ip6cp = (struct ip6ctlparam *)d;
2061 ip6 = ip6cp->ip6c_ip6;
2062 off = ip6cp->ip6c_off;
2063 sa6_src = ip6cp->ip6c_src;
2067 off = 0; /* fool gcc */
2072 struct in_conninfo inc;
2074 * XXX: We assume that when IPV6 is non NULL,
2075 * M and OFF are valid.
2078 /* check if we can safely examine src and dst ports */
2079 if (m->m_pkthdr.len < off + sizeof(*thp))
2082 bzero(&th, sizeof(th));
2083 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
2085 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
2086 (struct sockaddr *)ip6cp->ip6c_src,
2087 th.th_sport, cmd, NULL, notify);
2089 bzero(&inc, sizeof(inc));
2090 inc.inc_fport = th.th_dport;
2091 inc.inc_lport = th.th_sport;
2092 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
2093 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
2094 inc.inc_flags |= INC_ISIPV6;
2095 INP_INFO_RLOCK(&V_tcbinfo);
2096 syncache_unreach(&inc, &th);
2097 INP_INFO_RUNLOCK(&V_tcbinfo);
2099 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
2100 0, cmd, NULL, notify);
2106 * Following is where TCP initial sequence number generation occurs.
2108 * There are two places where we must use initial sequence numbers:
2109 * 1. In SYN-ACK packets.
2110 * 2. In SYN packets.
2112 * All ISNs for SYN-ACK packets are generated by the syncache. See
2113 * tcp_syncache.c for details.
2115 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2116 * depends on this property. In addition, these ISNs should be
2117 * unguessable so as to prevent connection hijacking. To satisfy
2118 * the requirements of this situation, the algorithm outlined in
2119 * RFC 1948 is used, with only small modifications.
2121 * Implementation details:
2123 * Time is based off the system timer, and is corrected so that it
2124 * increases by one megabyte per second. This allows for proper
2125 * recycling on high speed LANs while still leaving over an hour
2128 * As reading the *exact* system time is too expensive to be done
2129 * whenever setting up a TCP connection, we increment the time
2130 * offset in two ways. First, a small random positive increment
2131 * is added to isn_offset for each connection that is set up.
2132 * Second, the function tcp_isn_tick fires once per clock tick
2133 * and increments isn_offset as necessary so that sequence numbers
2134 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2135 * random positive increments serve only to ensure that the same
2136 * exact sequence number is never sent out twice (as could otherwise
2137 * happen when a port is recycled in less than the system tick
2140 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2141 * between seeding of isn_secret. This is normally set to zero,
2142 * as reseeding should not be necessary.
2144 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2145 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
2146 * general, this means holding an exclusive (write) lock.
2149 #define ISN_BYTES_PER_SECOND 1048576
2150 #define ISN_STATIC_INCREMENT 4096
2151 #define ISN_RANDOM_INCREMENT (4096 - 1)
2153 static VNET_DEFINE(u_char, isn_secret[32]);
2154 static VNET_DEFINE(int, isn_last);
2155 static VNET_DEFINE(int, isn_last_reseed);
2156 static VNET_DEFINE(u_int32_t, isn_offset);
2157 static VNET_DEFINE(u_int32_t, isn_offset_old);
2159 #define V_isn_secret VNET(isn_secret)
2160 #define V_isn_last VNET(isn_last)
2161 #define V_isn_last_reseed VNET(isn_last_reseed)
2162 #define V_isn_offset VNET(isn_offset)
2163 #define V_isn_offset_old VNET(isn_offset_old)
2166 tcp_new_isn(struct tcpcb *tp)
2169 u_int32_t md5_buffer[4];
2171 u_int32_t projected_offset;
2173 INP_WLOCK_ASSERT(tp->t_inpcb);
2176 /* Seed if this is the first use, reseed if requested. */
2177 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2178 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2180 read_random(&V_isn_secret, sizeof(V_isn_secret));
2181 V_isn_last_reseed = ticks;
2184 /* Compute the md5 hash and return the ISN. */
2186 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2187 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2189 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2190 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2191 sizeof(struct in6_addr));
2192 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2193 sizeof(struct in6_addr));
2197 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2198 sizeof(struct in_addr));
2199 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2200 sizeof(struct in_addr));
2202 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2203 MD5Final((u_char *) &md5_buffer, &isn_ctx);
2204 new_isn = (tcp_seq) md5_buffer[0];
2205 V_isn_offset += ISN_STATIC_INCREMENT +
2206 (arc4random() & ISN_RANDOM_INCREMENT);
2207 if (ticks != V_isn_last) {
2208 projected_offset = V_isn_offset_old +
2209 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2210 if (SEQ_GT(projected_offset, V_isn_offset))
2211 V_isn_offset = projected_offset;
2212 V_isn_offset_old = V_isn_offset;
2215 new_isn += V_isn_offset;
2221 * When a specific ICMP unreachable message is received and the
2222 * connection state is SYN-SENT, drop the connection. This behavior
2223 * is controlled by the icmp_may_rst sysctl.
2226 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2230 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2231 INP_WLOCK_ASSERT(inp);
2233 if ((inp->inp_flags & INP_TIMEWAIT) ||
2234 (inp->inp_flags & INP_DROPPED))
2237 tp = intotcpcb(inp);
2238 if (tp->t_state != TCPS_SYN_SENT)
2241 tp = tcp_drop(tp, errno);
2249 * When `need fragmentation' ICMP is received, update our idea of the MSS
2250 * based on the new value. Also nudge TCP to send something, since we
2251 * know the packet we just sent was dropped.
2252 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2254 static struct inpcb *
2255 tcp_mtudisc_notify(struct inpcb *inp, int error)
2258 tcp_mtudisc(inp, -1);
2263 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2268 INP_WLOCK_ASSERT(inp);
2269 if ((inp->inp_flags & INP_TIMEWAIT) ||
2270 (inp->inp_flags & INP_DROPPED))
2273 tp = intotcpcb(inp);
2274 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2276 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2278 so = inp->inp_socket;
2279 SOCKBUF_LOCK(&so->so_snd);
2280 /* If the mss is larger than the socket buffer, decrease the mss. */
2281 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2282 tp->t_maxseg = so->so_snd.sb_hiwat;
2283 SOCKBUF_UNLOCK(&so->so_snd);
2285 TCPSTAT_INC(tcps_mturesent);
2287 tp->snd_nxt = tp->snd_una;
2288 tcp_free_sackholes(tp);
2289 tp->snd_recover = tp->snd_max;
2290 if (tp->t_flags & TF_SACK_PERMIT)
2291 EXIT_FASTRECOVERY(tp->t_flags);
2292 tp->t_fb->tfb_tcp_output(tp);
2297 * Look-up the routing entry to the peer of this inpcb. If no route
2298 * is found and it cannot be allocated, then return 0. This routine
2299 * is called by TCP routines that access the rmx structure and by
2300 * tcp_mss_update to get the peer/interface MTU.
2303 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2305 struct nhop4_extended nh4;
2309 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2311 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2313 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2314 NHR_REF, 0, &nh4) != 0)
2318 maxmtu = nh4.nh_mtu;
2320 /* Report additional interface capabilities. */
2322 if (ifp->if_capenable & IFCAP_TSO4 &&
2323 ifp->if_hwassist & CSUM_TSO) {
2324 cap->ifcap |= CSUM_TSO;
2325 cap->tsomax = ifp->if_hw_tsomax;
2326 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2327 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2330 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2338 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2340 struct nhop6_extended nh6;
2341 struct in6_addr dst6;
2346 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2348 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2349 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2350 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2355 maxmtu = nh6.nh_mtu;
2357 /* Report additional interface capabilities. */
2359 if (ifp->if_capenable & IFCAP_TSO6 &&
2360 ifp->if_hwassist & CSUM_TSO) {
2361 cap->ifcap |= CSUM_TSO;
2362 cap->tsomax = ifp->if_hw_tsomax;
2363 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2364 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2367 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2375 * Calculate effective SMSS per RFC5681 definition for a given TCP
2376 * connection at its current state, taking into account SACK and etc.
2379 tcp_maxseg(const struct tcpcb *tp)
2383 if (tp->t_flags & TF_NOOPT)
2384 return (tp->t_maxseg);
2387 * Here we have a simplified code from tcp_addoptions(),
2388 * without a proper loop, and having most of paddings hardcoded.
2389 * We might make mistakes with padding here in some edge cases,
2390 * but this is harmless, since result of tcp_maxseg() is used
2391 * only in cwnd and ssthresh estimations.
2393 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2394 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2395 if (tp->t_flags & TF_RCVD_TSTMP)
2396 optlen = TCPOLEN_TSTAMP_APPA;
2399 #ifdef TCP_SIGNATURE
2400 if (tp->t_flags & TF_SIGNATURE)
2401 optlen += PAD(TCPOLEN_SIGNATURE);
2403 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2404 optlen += TCPOLEN_SACKHDR;
2405 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2406 optlen = PAD(optlen);
2409 if (tp->t_flags & TF_REQ_TSTMP)
2410 optlen = TCPOLEN_TSTAMP_APPA;
2412 optlen = PAD(TCPOLEN_MAXSEG);
2413 if (tp->t_flags & TF_REQ_SCALE)
2414 optlen += PAD(TCPOLEN_WINDOW);
2415 #ifdef TCP_SIGNATURE
2416 if (tp->t_flags & TF_SIGNATURE)
2417 optlen += PAD(TCPOLEN_SIGNATURE);
2419 if (tp->t_flags & TF_SACK_PERMIT)
2420 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2423 optlen = min(optlen, TCP_MAXOLEN);
2424 return (tp->t_maxseg - optlen);
2428 /* compute ESP/AH header size for TCP, including outer IP header. */
2430 ipsec_hdrsiz_tcp(struct tcpcb *tp)
2437 struct ip6_hdr *ip6;
2441 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) ||
2442 (!key_havesp(IPSEC_DIR_OUTBOUND)))
2444 m = m_gethdr(M_NOWAIT, MT_DATA);
2449 if ((inp->inp_vflag & INP_IPV6) != 0) {
2450 ip6 = mtod(m, struct ip6_hdr *);
2451 th = (struct tcphdr *)(ip6 + 1);
2452 m->m_pkthdr.len = m->m_len =
2453 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
2454 tcpip_fillheaders(inp, ip6, th);
2455 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2459 ip = mtod(m, struct ip *);
2460 th = (struct tcphdr *)(ip + 1);
2461 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
2462 tcpip_fillheaders(inp, ip, th);
2463 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2471 #ifdef TCP_SIGNATURE
2473 * Callback function invoked by m_apply() to digest TCP segment data
2474 * contained within an mbuf chain.
2477 tcp_signature_apply(void *fstate, void *data, u_int len)
2480 MD5Update(fstate, (u_char *)data, len);
2485 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2486 * search with the destination IP address, and a 'magic SPI' to be
2487 * determined by the application. This is hardcoded elsewhere to 1179
2490 tcp_get_sav(struct mbuf *m, u_int direction)
2492 union sockaddr_union dst;
2493 struct secasvar *sav;
2496 struct ip6_hdr *ip6;
2497 char ip6buf[INET6_ADDRSTRLEN];
2500 /* Extract the destination from the IP header in the mbuf. */
2501 bzero(&dst, sizeof(union sockaddr_union));
2502 ip = mtod(m, struct ip *);
2504 ip6 = NULL; /* Make the compiler happy. */
2509 dst.sa.sa_len = sizeof(struct sockaddr_in);
2510 dst.sa.sa_family = AF_INET;
2511 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2512 ip->ip_src : ip->ip_dst;
2516 case (IPV6_VERSION >> 4):
2517 ip6 = mtod(m, struct ip6_hdr *);
2518 dst.sa.sa_len = sizeof(struct sockaddr_in6);
2519 dst.sa.sa_family = AF_INET6;
2520 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2521 ip6->ip6_src : ip6->ip6_dst;
2530 /* Look up an SADB entry which matches the address of the peer. */
2531 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2533 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2534 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2536 (ip->ip_v == (IPV6_VERSION >> 4)) ?
2537 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2546 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2549 * m pointer to head of mbuf chain
2550 * len length of TCP segment data, excluding options
2551 * optlen length of TCP segment options
2552 * buf pointer to storage for computed MD5 digest
2553 * sav pointer to security assosiation
2555 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2556 * When called from tcp_input(), we can be sure that th_sum has been
2557 * zeroed out and verified already.
2559 * Releases reference to SADB key before return.
2561 * Return 0 if successful, otherwise return -1.
2565 tcp_signature_do_compute(struct mbuf *m, int len, int optlen,
2566 u_char *buf, struct secasvar *sav)
2569 struct ippseudo ippseudo;
2575 struct ipovly *ipovly;
2579 struct ip6_hdr *ip6;
2580 struct in6_addr in6;
2586 KASSERT(m != NULL, ("NULL mbuf chain"));
2587 KASSERT(buf != NULL, ("NULL signature pointer"));
2589 /* Extract the destination from the IP header in the mbuf. */
2590 ip = mtod(m, struct ip *);
2592 ip6 = NULL; /* Make the compiler happy. */
2597 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2599 * XXX The ippseudo header MUST be digested in network byte order,
2600 * or else we'll fail the regression test. Assume all fields we've
2601 * been doing arithmetic on have been in host byte order.
2602 * XXX One cannot depend on ipovly->ih_len here. When called from
2603 * tcp_output(), the underlying ip_len member has not yet been set.
2608 ipovly = (struct ipovly *)ip;
2609 ippseudo.ippseudo_src = ipovly->ih_src;
2610 ippseudo.ippseudo_dst = ipovly->ih_dst;
2611 ippseudo.ippseudo_pad = 0;
2612 ippseudo.ippseudo_p = IPPROTO_TCP;
2613 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2615 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2617 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2618 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2623 * RFC 2385, 2.0 Proposal
2624 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2625 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2626 * extended next header value (to form 32 bits), and 32-bit segment
2628 * Note: Upper-Layer Packet Length comes before Next Header.
2630 case (IPV6_VERSION >> 4):
2632 in6_clearscope(&in6);
2633 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2635 in6_clearscope(&in6);
2636 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2637 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2638 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2640 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2641 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2642 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2644 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2646 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2647 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2659 * Step 2: Update MD5 hash with TCP header, excluding options.
2660 * The TCP checksum must be set to zero.
2662 savecsum = th->th_sum;
2664 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2665 th->th_sum = savecsum;
2668 * Step 3: Update MD5 hash with TCP segment data.
2669 * Use m_apply() to avoid an early m_pullup().
2672 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2675 * Step 4: Update MD5 hash with shared secret.
2677 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2678 MD5Final(buf, &ctx);
2680 key_sa_recordxfer(sav, m);
2686 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2688 * Return 0 if successful, otherwise return -1.
2691 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2692 u_char *buf, u_int direction)
2694 struct secasvar *sav;
2696 if ((sav = tcp_get_sav(m, direction)) == NULL)
2699 return (tcp_signature_do_compute(m, len, optlen, buf, sav));
2703 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2706 * m pointer to head of mbuf chain
2707 * len length of TCP segment data, excluding options
2708 * optlen length of TCP segment options
2709 * buf pointer to storage for computed MD5 digest
2710 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2712 * Return 1 if successful, otherwise return 0.
2715 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2716 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2718 char tmpdigest[TCP_SIGLEN];
2720 if (tcp_sig_checksigs == 0)
2722 if ((tcpbflag & TF_SIGNATURE) == 0) {
2723 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2726 * If this socket is not expecting signature but
2727 * the segment contains signature just fail.
2729 TCPSTAT_INC(tcps_sig_err_sigopt);
2730 TCPSTAT_INC(tcps_sig_rcvbadsig);
2734 /* Signature is not expected, and not present in segment. */
2739 * If this socket is expecting signature but the segment does not
2740 * contain any just fail.
2742 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2743 TCPSTAT_INC(tcps_sig_err_nosigopt);
2744 TCPSTAT_INC(tcps_sig_rcvbadsig);
2747 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2748 IPSEC_DIR_INBOUND) == -1) {
2749 TCPSTAT_INC(tcps_sig_err_buildsig);
2750 TCPSTAT_INC(tcps_sig_rcvbadsig);
2754 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2755 TCPSTAT_INC(tcps_sig_rcvbadsig);
2758 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2761 #endif /* TCP_SIGNATURE */
2764 sysctl_drop(SYSCTL_HANDLER_ARGS)
2766 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2767 struct sockaddr_storage addrs[2];
2771 struct sockaddr_in *fin, *lin;
2773 struct sockaddr_in6 *fin6, *lin6;
2784 if (req->oldptr != NULL || req->oldlen != 0)
2786 if (req->newptr == NULL)
2788 if (req->newlen < sizeof(addrs))
2790 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2794 switch (addrs[0].ss_family) {
2797 fin6 = (struct sockaddr_in6 *)&addrs[0];
2798 lin6 = (struct sockaddr_in6 *)&addrs[1];
2799 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2800 lin6->sin6_len != sizeof(struct sockaddr_in6))
2802 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2803 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2805 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2806 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2807 fin = (struct sockaddr_in *)&addrs[0];
2808 lin = (struct sockaddr_in *)&addrs[1];
2811 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2814 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2821 fin = (struct sockaddr_in *)&addrs[0];
2822 lin = (struct sockaddr_in *)&addrs[1];
2823 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2824 lin->sin_len != sizeof(struct sockaddr_in))
2831 INP_INFO_RLOCK(&V_tcbinfo);
2832 switch (addrs[0].ss_family) {
2835 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2836 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2837 INPLOOKUP_WLOCKPCB, NULL);
2842 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2843 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2848 if (inp->inp_flags & INP_TIMEWAIT) {
2850 * XXXRW: There currently exists a state where an
2851 * inpcb is present, but its timewait state has been
2852 * discarded. For now, don't allow dropping of this
2860 } else if (!(inp->inp_flags & INP_DROPPED) &&
2861 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2862 tp = intotcpcb(inp);
2863 tp = tcp_drop(tp, ECONNABORTED);
2870 INP_INFO_RUNLOCK(&V_tcbinfo);
2874 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2875 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2876 0, sysctl_drop, "", "Drop TCP connection");
2879 * Generate a standardized TCP log line for use throughout the
2880 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2881 * allow use in the interrupt context.
2883 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2884 * NB: The function may return NULL if memory allocation failed.
2886 * Due to header inclusion and ordering limitations the struct ip
2887 * and ip6_hdr pointers have to be passed as void pointers.
2890 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2894 /* Is logging enabled? */
2895 if (tcp_log_in_vain == 0)
2898 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2902 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2906 /* Is logging enabled? */
2907 if (tcp_log_debug == 0)
2910 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2914 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2921 const struct ip6_hdr *ip6;
2923 ip6 = (const struct ip6_hdr *)ip6hdr;
2925 ip = (struct ip *)ip4hdr;
2928 * The log line looks like this:
2929 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2931 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2932 sizeof(PRINT_TH_FLAGS) + 1 +
2934 2 * INET6_ADDRSTRLEN;
2936 2 * INET_ADDRSTRLEN;
2939 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2943 strcat(s, "TCP: [");
2946 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2947 inet_ntoa_r(inc->inc_faddr, sp);
2949 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2951 inet_ntoa_r(inc->inc_laddr, sp);
2953 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2956 ip6_sprintf(sp, &inc->inc6_faddr);
2958 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2960 ip6_sprintf(sp, &inc->inc6_laddr);
2962 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2963 } else if (ip6 && th) {
2964 ip6_sprintf(sp, &ip6->ip6_src);
2966 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2968 ip6_sprintf(sp, &ip6->ip6_dst);
2970 sprintf(sp, "]:%i", ntohs(th->th_dport));
2973 } else if (ip && th) {
2974 inet_ntoa_r(ip->ip_src, sp);
2976 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2978 inet_ntoa_r(ip->ip_dst, sp);
2980 sprintf(sp, "]:%i", ntohs(th->th_dport));
2988 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2989 if (*(s + size - 1) != '\0')
2990 panic("%s: string too long", __func__);
2995 * A subroutine which makes it easy to track TCP state changes with DTrace.
2996 * This function shouldn't be called for t_state initializations that don't
2997 * correspond to actual TCP state transitions.
3000 tcp_state_change(struct tcpcb *tp, int newstate)
3002 #if defined(KDTRACE_HOOKS)
3003 int pstate = tp->t_state;
3006 TCPSTATES_DEC(tp->t_state);
3007 TCPSTATES_INC(newstate);
3008 tp->t_state = newstate;
3009 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);