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);
249 static struct tcp_function_block tcp_def_funcblk = {
253 tcp_default_ctloutput,
264 int t_functions_inited = 0;
265 struct tcp_funchead t_functions;
266 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
269 init_tcp_functions(void)
271 if (t_functions_inited == 0) {
272 TAILQ_INIT(&t_functions);
273 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
274 t_functions_inited = 1;
278 static struct tcp_function_block *
279 find_tcp_functions_locked(struct tcp_function_set *fs)
281 struct tcp_function *f;
282 struct tcp_function_block *blk=NULL;
284 TAILQ_FOREACH(f, &t_functions, tf_next) {
285 if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) {
293 static struct tcp_function_block *
294 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
296 struct tcp_function_block *rblk=NULL;
297 struct tcp_function *f;
299 TAILQ_FOREACH(f, &t_functions, tf_next) {
300 if (f->tf_fb == blk) {
311 struct tcp_function_block *
312 find_and_ref_tcp_functions(struct tcp_function_set *fs)
314 struct tcp_function_block *blk;
316 rw_rlock(&tcp_function_lock);
317 blk = find_tcp_functions_locked(fs);
319 refcount_acquire(&blk->tfb_refcnt);
320 rw_runlock(&tcp_function_lock);
324 struct tcp_function_block *
325 find_and_ref_tcp_fb(struct tcp_function_block *blk)
327 struct tcp_function_block *rblk;
329 rw_rlock(&tcp_function_lock);
330 rblk = find_tcp_fb_locked(blk, NULL);
332 refcount_acquire(&rblk->tfb_refcnt);
333 rw_runlock(&tcp_function_lock);
339 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
342 struct tcp_function_set fs;
343 struct tcp_function_block *blk;
345 memset(&fs, 0, sizeof(fs));
346 rw_rlock(&tcp_function_lock);
347 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
350 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
351 fs.pcbcnt = blk->tfb_refcnt;
353 rw_runlock(&tcp_function_lock);
354 error = sysctl_handle_string(oidp, fs.function_set_name,
355 sizeof(fs.function_set_name), req);
357 /* Check for error or no change */
358 if (error != 0 || req->newptr == NULL)
361 rw_wlock(&tcp_function_lock);
362 blk = find_tcp_functions_locked(&fs);
364 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
368 tcp_func_set_ptr = blk;
370 rw_wunlock(&tcp_function_lock);
374 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
375 CTLTYPE_STRING | CTLFLAG_RW,
376 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
377 "Set/get the default TCP functions");
380 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
382 int error, cnt, linesz;
383 struct tcp_function *f;
388 rw_rlock(&tcp_function_lock);
389 TAILQ_FOREACH(f, &t_functions, tf_next) {
392 rw_runlock(&tcp_function_lock);
394 bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1;
395 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
400 linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count");
405 rw_rlock(&tcp_function_lock);
406 TAILQ_FOREACH(f, &t_functions, tf_next) {
407 linesz = snprintf(cp, bufsz, "%-32s%c %u\n",
408 f->tf_fb->tfb_tcp_block_name,
409 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
410 f->tf_fb->tfb_refcnt);
411 if (linesz >= bufsz) {
419 rw_runlock(&tcp_function_lock);
421 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
422 free(buffer, M_TEMP);
426 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
427 CTLTYPE_STRING|CTLFLAG_RD,
428 NULL, 0, sysctl_net_inet_list_available, "A",
429 "list available TCP Function sets");
432 * Target size of TCP PCB hash tables. Must be a power of two.
434 * Note that this can be overridden by the kernel environment
435 * variable net.inet.tcp.tcbhashsize
438 #define TCBHASHSIZE 0
443 * Callouts should be moved into struct tcp directly. They are currently
444 * separate because the tcpcb structure is exported to userland for sysctl
445 * parsing purposes, which do not know about callouts.
454 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
455 #define V_tcpcb_zone VNET(tcpcb_zone)
457 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
458 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
460 static struct mtx isn_mtx;
462 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
463 #define ISN_LOCK() mtx_lock(&isn_mtx)
464 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
467 * TCP initialization.
470 tcp_zone_change(void *tag)
473 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
474 uma_zone_set_max(V_tcpcb_zone, maxsockets);
475 tcp_tw_zone_change();
479 tcp_inpcb_init(void *mem, int size, int flags)
481 struct inpcb *inp = mem;
483 INP_LOCK_INIT(inp, "inp", "tcpinp");
488 * Take a value and get the next power of 2 that doesn't overflow.
489 * Used to size the tcp_inpcb hash buckets.
492 maketcp_hashsize(int size)
498 * get the next power of 2 higher than maxsockets.
500 hashsize = 1 << fls(size);
501 /* catch overflow, and just go one power of 2 smaller */
502 if (hashsize < size) {
503 hashsize = 1 << (fls(size) - 1);
509 register_tcp_functions(struct tcp_function_block *blk, int wait)
511 struct tcp_function_block *lblk;
512 struct tcp_function *n;
513 struct tcp_function_set fs;
515 if (t_functions_inited == 0) {
516 init_tcp_functions();
518 if ((blk->tfb_tcp_output == NULL) ||
519 (blk->tfb_tcp_do_segment == NULL) ||
520 (blk->tfb_tcp_ctloutput == NULL) ||
521 (strlen(blk->tfb_tcp_block_name) == 0)) {
523 * These functions are required and you
528 if (blk->tfb_tcp_timer_stop_all ||
529 blk->tfb_tcp_timer_activate ||
530 blk->tfb_tcp_timer_active ||
531 blk->tfb_tcp_timer_stop) {
533 * If you define one timer function you
534 * must have them all.
536 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
537 (blk->tfb_tcp_timer_activate == NULL) ||
538 (blk->tfb_tcp_timer_active == NULL) ||
539 (blk->tfb_tcp_timer_stop == NULL)) {
543 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
548 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
549 rw_wlock(&tcp_function_lock);
550 lblk = find_tcp_functions_locked(&fs);
552 /* Duplicate name space not allowed */
553 rw_wunlock(&tcp_function_lock);
554 free(n, M_TCPFUNCTIONS);
557 refcount_init(&blk->tfb_refcnt, 0);
559 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
560 rw_wunlock(&tcp_function_lock);
565 deregister_tcp_functions(struct tcp_function_block *blk)
567 struct tcp_function_block *lblk;
568 struct tcp_function *f;
571 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
572 /* You can't un-register the default */
575 rw_wlock(&tcp_function_lock);
576 if (blk == tcp_func_set_ptr) {
577 /* You can't free the current default */
578 rw_wunlock(&tcp_function_lock);
581 if (blk->tfb_refcnt) {
582 /* Still tcb attached, mark it. */
583 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
584 rw_wunlock(&tcp_function_lock);
587 lblk = find_tcp_fb_locked(blk, &f);
590 TAILQ_REMOVE(&t_functions, f, tf_next);
592 free(f, M_TCPFUNCTIONS);
595 rw_wunlock(&tcp_function_lock);
602 const char *tcbhash_tuneable;
605 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
607 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
608 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
609 printf("%s: WARNING: unable to register helper hook\n", __func__);
610 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
611 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
612 printf("%s: WARNING: unable to register helper hook\n", __func__);
613 hashsize = TCBHASHSIZE;
614 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
617 * Auto tune the hash size based on maxsockets.
618 * A perfect hash would have a 1:1 mapping
619 * (hashsize = maxsockets) however it's been
620 * suggested that O(2) average is better.
622 hashsize = maketcp_hashsize(maxsockets / 4);
624 * Our historical default is 512,
625 * do not autotune lower than this.
629 if (bootverbose && IS_DEFAULT_VNET(curvnet))
630 printf("%s: %s auto tuned to %d\n", __func__,
631 tcbhash_tuneable, hashsize);
634 * We require a hashsize to be a power of two.
635 * Previously if it was not a power of two we would just reset it
636 * back to 512, which could be a nasty surprise if you did not notice
638 * Instead what we do is clip it to the closest power of two lower
639 * than the specified hash value.
641 if (!powerof2(hashsize)) {
642 int oldhashsize = hashsize;
644 hashsize = maketcp_hashsize(hashsize);
645 /* prevent absurdly low value */
648 printf("%s: WARNING: TCB hash size not a power of 2, "
649 "clipped from %d to %d.\n", __func__, oldhashsize,
652 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
653 "tcp_inpcb", tcp_inpcb_init, NULL, 0, IPI_HASHFIELDS_4TUPLE);
656 * These have to be type stable for the benefit of the timers.
658 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
659 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
660 uma_zone_set_max(V_tcpcb_zone, maxsockets);
661 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
667 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
668 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
669 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
671 /* Skip initialization of globals for non-default instances. */
672 if (!IS_DEFAULT_VNET(curvnet))
675 tcp_reass_global_init();
677 /* XXX virtualize those bellow? */
678 tcp_delacktime = TCPTV_DELACK;
679 tcp_keepinit = TCPTV_KEEP_INIT;
680 tcp_keepidle = TCPTV_KEEP_IDLE;
681 tcp_keepintvl = TCPTV_KEEPINTVL;
682 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
684 tcp_rexmit_min = TCPTV_MIN;
685 if (tcp_rexmit_min < 1)
687 tcp_persmin = TCPTV_PERSMIN;
688 tcp_persmax = TCPTV_PERSMAX;
689 tcp_rexmit_slop = TCPTV_CPU_VAR;
690 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
691 tcp_tcbhashsize = hashsize;
692 /* Setup the tcp function block list */
693 init_tcp_functions();
694 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
696 if (tcp_soreceive_stream) {
698 tcp_usrreqs.pru_soreceive = soreceive_stream;
701 tcp6_usrreqs.pru_soreceive = soreceive_stream;
706 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
708 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
710 if (max_protohdr < TCP_MINPROTOHDR)
711 max_protohdr = TCP_MINPROTOHDR;
712 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
714 #undef TCP_MINPROTOHDR
717 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
718 SHUTDOWN_PRI_DEFAULT);
719 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
720 EVENTHANDLER_PRI_ANY);
732 tcp_destroy(void *unused __unused)
737 * All our processes are gone, all our sockets should be cleaned
738 * up, which means, we should be past the tcp_discardcb() calls.
739 * Sleep to let all tcpcb timers really disappear and cleanup.
742 INP_LIST_RLOCK(&V_tcbinfo);
743 n = V_tcbinfo.ipi_count;
744 INP_LIST_RUNLOCK(&V_tcbinfo);
746 pause("tcpdes", hz / 10);
751 in_pcbinfo_destroy(&V_tcbinfo);
752 /* tcp_discardcb() clears the sack_holes up. */
753 uma_zdestroy(V_sack_hole_zone);
754 uma_zdestroy(V_tcpcb_zone);
758 * Cannot free the zone until all tcpcbs are released as we attach
759 * the allocations to them.
761 tcp_fastopen_destroy();
764 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
766 printf("%s: WARNING: unable to deregister helper hook "
767 "type=%d, id=%d: error %d returned\n", __func__,
768 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
770 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
772 printf("%s: WARNING: unable to deregister helper hook "
773 "type=%d, id=%d: error %d returned\n", __func__,
774 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
777 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
787 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
788 * tcp_template used to store this data in mbufs, but we now recopy it out
789 * of the tcpcb each time to conserve mbufs.
792 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
794 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
796 INP_WLOCK_ASSERT(inp);
799 if ((inp->inp_vflag & INP_IPV6) != 0) {
802 ip6 = (struct ip6_hdr *)ip_ptr;
803 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
804 (inp->inp_flow & IPV6_FLOWINFO_MASK);
805 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
806 (IPV6_VERSION & IPV6_VERSION_MASK);
807 ip6->ip6_nxt = IPPROTO_TCP;
808 ip6->ip6_plen = htons(sizeof(struct tcphdr));
809 ip6->ip6_src = inp->in6p_laddr;
810 ip6->ip6_dst = inp->in6p_faddr;
813 #if defined(INET6) && defined(INET)
820 ip = (struct ip *)ip_ptr;
821 ip->ip_v = IPVERSION;
823 ip->ip_tos = inp->inp_ip_tos;
827 ip->ip_ttl = inp->inp_ip_ttl;
829 ip->ip_p = IPPROTO_TCP;
830 ip->ip_src = inp->inp_laddr;
831 ip->ip_dst = inp->inp_faddr;
834 th->th_sport = inp->inp_lport;
835 th->th_dport = inp->inp_fport;
843 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
847 * Create template to be used to send tcp packets on a connection.
848 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
849 * use for this function is in keepalives, which use tcp_respond.
852 tcpip_maketemplate(struct inpcb *inp)
856 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
859 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
864 * Send a single message to the TCP at address specified by
865 * the given TCP/IP header. If m == NULL, then we make a copy
866 * of the tcpiphdr at th and send directly to the addressed host.
867 * This is used to force keep alive messages out using the TCP
868 * template for a connection. If flags are given then we send
869 * a message back to the TCP which originated the segment th,
870 * and discard the mbuf containing it and any other attached mbufs.
872 * In any case the ack and sequence number of the transmitted
873 * segment are as specified by the parameters.
875 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
878 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
879 tcp_seq ack, tcp_seq seq, int flags)
891 int optlen, tlen, win;
894 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
897 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
904 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
905 INP_WLOCK_ASSERT(inp);
912 if (!(flags & TH_RST)) {
913 win = sbspace(&inp->inp_socket->so_rcv);
914 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
915 win = (long)TCP_MAXWIN << tp->rcv_scale;
917 if ((tp->t_flags & TF_NOOPT) == 0)
921 m = m_gethdr(M_NOWAIT, MT_DATA);
924 m->m_data += max_linkhdr;
927 bcopy((caddr_t)ip6, mtod(m, caddr_t),
928 sizeof(struct ip6_hdr));
929 ip6 = mtod(m, struct ip6_hdr *);
930 nth = (struct tcphdr *)(ip6 + 1);
934 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
935 ip = mtod(m, struct ip *);
936 nth = (struct tcphdr *)(ip + 1);
938 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
940 } else if (!M_WRITABLE(m)) {
943 /* Can't reuse 'm', allocate a new mbuf. */
944 n = m_gethdr(M_NOWAIT, MT_DATA);
950 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
956 n->m_data += max_linkhdr;
957 /* m_len is set later */
958 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
961 bcopy((caddr_t)ip6, mtod(n, caddr_t),
962 sizeof(struct ip6_hdr));
963 ip6 = mtod(n, struct ip6_hdr *);
964 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
965 nth = (struct tcphdr *)(ip6 + 1);
969 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
970 ip = mtod(n, struct ip *);
971 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
972 nth = (struct tcphdr *)(ip + 1);
974 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
975 xchg(nth->th_dport, nth->th_sport, uint16_t);
982 * XXX MRT We inherit the FIB, which is lucky.
986 m->m_data = (caddr_t)ipgen;
987 /* m_len is set later */
990 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
991 nth = (struct tcphdr *)(ip6 + 1);
995 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
996 nth = (struct tcphdr *)(ip + 1);
1000 * this is usually a case when an extension header
1001 * exists between the IPv6 header and the
1004 nth->th_sport = th->th_sport;
1005 nth->th_dport = th->th_dport;
1007 xchg(nth->th_dport, nth->th_sport, uint16_t);
1013 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1015 #if defined(INET) && defined(INET6)
1019 tlen = sizeof (struct tcpiphdr);
1023 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1024 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1025 m, tlen, (long)M_TRAILINGSPACE(m)));
1030 /* Make sure we have room. */
1031 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1032 m->m_next = m_get(M_NOWAIT, MT_DATA);
1034 optp = mtod(m->m_next, u_char *);
1039 optp = (u_char *) (nth + 1);
1045 if (tp->t_flags & TF_RCVD_TSTMP) {
1046 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1047 to.to_tsecr = tp->ts_recent;
1048 to.to_flags |= TOF_TS;
1050 #ifdef TCP_SIGNATURE
1051 /* TCP-MD5 (RFC2385). */
1052 if (tp->t_flags & TF_SIGNATURE)
1053 to.to_flags |= TOF_SIGNATURE;
1056 /* Add the options. */
1057 tlen += optlen = tcp_addoptions(&to, optp);
1059 /* Update m_len in the correct mbuf. */
1060 optm->m_len += optlen;
1066 ip6->ip6_vfc = IPV6_VERSION;
1067 ip6->ip6_nxt = IPPROTO_TCP;
1068 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1071 #if defined(INET) && defined(INET6)
1076 ip->ip_len = htons(tlen);
1077 ip->ip_ttl = V_ip_defttl;
1078 if (V_path_mtu_discovery)
1079 ip->ip_off |= htons(IP_DF);
1082 m->m_pkthdr.len = tlen;
1083 m->m_pkthdr.rcvif = NULL;
1087 * Packet is associated with a socket, so allow the
1088 * label of the response to reflect the socket label.
1090 INP_WLOCK_ASSERT(inp);
1091 mac_inpcb_create_mbuf(inp, m);
1094 * Packet is not associated with a socket, so possibly
1095 * update the label in place.
1097 mac_netinet_tcp_reply(m);
1100 nth->th_seq = htonl(seq);
1101 nth->th_ack = htonl(ack);
1103 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1104 nth->th_flags = flags;
1106 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1108 nth->th_win = htons((u_short)win);
1111 #ifdef TCP_SIGNATURE
1112 if (to.to_flags & TOF_SIGNATURE) {
1113 tcp_signature_compute(m, 0, 0, optlen, to.to_signature,
1114 IPSEC_DIR_OUTBOUND);
1118 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1121 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1122 nth->th_sum = in6_cksum_pseudo(ip6,
1123 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1124 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1128 #if defined(INET6) && defined(INET)
1133 m->m_pkthdr.csum_flags = CSUM_TCP;
1134 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1135 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1139 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1140 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1142 TCP_PROBE3(debug__output, tp, th, mtod(m, const char *));
1144 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
1147 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
1150 (void) ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1152 #if defined(INET) && defined(INET6)
1156 (void) ip_output(m, NULL, NULL, 0, NULL, inp);
1161 * Create a new TCP control block, making an
1162 * empty reassembly queue and hooking it to the argument
1163 * protocol control block. The `inp' parameter must have
1164 * come from the zone allocator set up in tcp_init().
1167 tcp_newtcpcb(struct inpcb *inp)
1169 struct tcpcb_mem *tm;
1172 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1175 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1180 /* Initialise cc_var struct for this tcpcb. */
1182 tp->ccv->type = IPPROTO_TCP;
1183 tp->ccv->ccvc.tcp = tp;
1184 rw_rlock(&tcp_function_lock);
1185 tp->t_fb = tcp_func_set_ptr;
1186 refcount_acquire(&tp->t_fb->tfb_refcnt);
1187 rw_runlock(&tcp_function_lock);
1188 if (tp->t_fb->tfb_tcp_fb_init) {
1189 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1192 * Use the current system default CC algorithm.
1195 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1196 CC_ALGO(tp) = CC_DEFAULT();
1199 if (CC_ALGO(tp)->cb_init != NULL)
1200 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1201 if (tp->t_fb->tfb_tcp_fb_fini)
1202 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1203 refcount_release(&tp->t_fb->tfb_refcnt);
1204 uma_zfree(V_tcpcb_zone, tm);
1209 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1210 if (tp->t_fb->tfb_tcp_fb_fini)
1211 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1212 refcount_release(&tp->t_fb->tfb_refcnt);
1213 uma_zfree(V_tcpcb_zone, tm);
1218 tp->t_vnet = inp->inp_vnet;
1220 tp->t_timers = &tm->tt;
1221 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
1224 isipv6 ? V_tcp_v6mssdflt :
1228 /* Set up our timeouts. */
1229 callout_init(&tp->t_timers->tt_rexmt, 1);
1230 callout_init(&tp->t_timers->tt_persist, 1);
1231 callout_init(&tp->t_timers->tt_keep, 1);
1232 callout_init(&tp->t_timers->tt_2msl, 1);
1233 callout_init(&tp->t_timers->tt_delack, 1);
1235 if (V_tcp_do_rfc1323)
1236 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1238 tp->t_flags |= TF_SACK_PERMIT;
1239 TAILQ_INIT(&tp->snd_holes);
1241 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1244 in_pcbref(inp); /* Reference for tcpcb */
1248 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1249 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1250 * reasonable initial retransmit time.
1252 tp->t_srtt = TCPTV_SRTTBASE;
1253 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1254 tp->t_rttmin = tcp_rexmit_min;
1255 tp->t_rxtcur = TCPTV_RTOBASE;
1256 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1257 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1258 tp->t_rcvtime = ticks;
1260 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1261 * because the socket may be bound to an IPv6 wildcard address,
1262 * which may match an IPv4-mapped IPv6 address.
1264 inp->inp_ip_ttl = V_ip_defttl;
1268 * Init the TCP PCAP queues.
1270 tcp_pcap_tcpcb_init(tp);
1272 return (tp); /* XXX */
1276 * Switch the congestion control algorithm back to NewReno for any active
1277 * control blocks using an algorithm which is about to go away.
1278 * This ensures the CC framework can allow the unload to proceed without leaving
1279 * any dangling pointers which would trigger a panic.
1280 * Returning non-zero would inform the CC framework that something went wrong
1281 * and it would be unsafe to allow the unload to proceed. However, there is no
1282 * way for this to occur with this implementation so we always return zero.
1285 tcp_ccalgounload(struct cc_algo *unload_algo)
1287 struct cc_algo *tmpalgo;
1290 VNET_ITERATOR_DECL(vnet_iter);
1293 * Check all active control blocks across all network stacks and change
1294 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1295 * requires cleanup code to be run, call it.
1298 VNET_FOREACH(vnet_iter) {
1299 CURVNET_SET(vnet_iter);
1300 INP_INFO_WLOCK(&V_tcbinfo);
1302 * New connections already part way through being initialised
1303 * with the CC algo we're removing will not race with this code
1304 * because the INP_INFO_WLOCK is held during initialisation. We
1305 * therefore don't enter the loop below until the connection
1306 * list has stabilised.
1308 LIST_FOREACH(inp, &V_tcb, inp_list) {
1310 /* Important to skip tcptw structs. */
1311 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1312 (tp = intotcpcb(inp)) != NULL) {
1314 * By holding INP_WLOCK here, we are assured
1315 * that the connection is not currently
1316 * executing inside the CC module's functions
1317 * i.e. it is safe to make the switch back to
1320 if (CC_ALGO(tp) == unload_algo) {
1321 tmpalgo = CC_ALGO(tp);
1322 /* NewReno does not require any init. */
1323 CC_ALGO(tp) = &newreno_cc_algo;
1324 if (tmpalgo->cb_destroy != NULL)
1325 tmpalgo->cb_destroy(tp->ccv);
1330 INP_INFO_WUNLOCK(&V_tcbinfo);
1333 VNET_LIST_RUNLOCK();
1339 * Drop a TCP connection, reporting
1340 * the specified error. If connection is synchronized,
1341 * then send a RST to peer.
1344 tcp_drop(struct tcpcb *tp, int errno)
1346 struct socket *so = tp->t_inpcb->inp_socket;
1348 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1349 INP_WLOCK_ASSERT(tp->t_inpcb);
1351 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1352 tcp_state_change(tp, TCPS_CLOSED);
1353 (void) tp->t_fb->tfb_tcp_output(tp);
1354 TCPSTAT_INC(tcps_drops);
1356 TCPSTAT_INC(tcps_conndrops);
1357 if (errno == ETIMEDOUT && tp->t_softerror)
1358 errno = tp->t_softerror;
1359 so->so_error = errno;
1360 return (tcp_close(tp));
1364 tcp_discardcb(struct tcpcb *tp)
1366 struct inpcb *inp = tp->t_inpcb;
1367 struct socket *so = inp->inp_socket;
1369 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1373 INP_WLOCK_ASSERT(inp);
1376 * Make sure that all of our timers are stopped before we delete the
1379 * If stopping a timer fails, we schedule a discard function in same
1380 * callout, and the last discard function called will take care of
1381 * deleting the tcpcb.
1383 tp->t_timers->tt_draincnt = 0;
1384 tcp_timer_stop(tp, TT_REXMT);
1385 tcp_timer_stop(tp, TT_PERSIST);
1386 tcp_timer_stop(tp, TT_KEEP);
1387 tcp_timer_stop(tp, TT_2MSL);
1388 tcp_timer_stop(tp, TT_DELACK);
1389 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1391 * Call the stop-all function of the methods,
1392 * this function should call the tcp_timer_stop()
1393 * method with each of the function specific timeouts.
1394 * That stop will be called via the tfb_tcp_timer_stop()
1395 * which should use the async drain function of the
1396 * callout system (see tcp_var.h).
1398 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1402 * If we got enough samples through the srtt filter,
1403 * save the rtt and rttvar in the routing entry.
1404 * 'Enough' is arbitrarily defined as 4 rtt samples.
1405 * 4 samples is enough for the srtt filter to converge
1406 * to within enough % of the correct value; fewer samples
1407 * and we could save a bogus rtt. The danger is not high
1408 * as tcp quickly recovers from everything.
1409 * XXX: Works very well but needs some more statistics!
1411 if (tp->t_rttupdated >= 4) {
1412 struct hc_metrics_lite metrics;
1415 bzero(&metrics, sizeof(metrics));
1417 * Update the ssthresh always when the conditions below
1418 * are satisfied. This gives us better new start value
1419 * for the congestion avoidance for new connections.
1420 * ssthresh is only set if packet loss occurred on a session.
1422 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1423 * being torn down. Ideally this code would not use 'so'.
1425 ssthresh = tp->snd_ssthresh;
1426 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1428 * convert the limit from user data bytes to
1429 * packets then to packet data bytes.
1431 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1434 ssthresh *= (u_long)(tp->t_maxseg +
1436 (isipv6 ? sizeof (struct ip6_hdr) +
1437 sizeof (struct tcphdr) :
1439 sizeof (struct tcpiphdr)
1446 metrics.rmx_ssthresh = ssthresh;
1448 metrics.rmx_rtt = tp->t_srtt;
1449 metrics.rmx_rttvar = tp->t_rttvar;
1450 metrics.rmx_cwnd = tp->snd_cwnd;
1451 metrics.rmx_sendpipe = 0;
1452 metrics.rmx_recvpipe = 0;
1454 tcp_hc_update(&inp->inp_inc, &metrics);
1457 /* free the reassembly queue, if any */
1458 tcp_reass_flush(tp);
1461 /* Disconnect offload device, if any. */
1462 if (tp->t_flags & TF_TOE)
1463 tcp_offload_detach(tp);
1466 tcp_free_sackholes(tp);
1469 /* Free the TCP PCAP queues. */
1470 tcp_pcap_drain(&(tp->t_inpkts));
1471 tcp_pcap_drain(&(tp->t_outpkts));
1474 /* Allow the CC algorithm to clean up after itself. */
1475 if (CC_ALGO(tp)->cb_destroy != NULL)
1476 CC_ALGO(tp)->cb_destroy(tp->ccv);
1478 khelp_destroy_osd(tp->osd);
1481 inp->inp_ppcb = NULL;
1482 if (tp->t_timers->tt_draincnt == 0) {
1483 /* We own the last reference on tcpcb, let's free it. */
1484 if (tp->t_fb->tfb_tcp_fb_fini)
1485 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1486 refcount_release(&tp->t_fb->tfb_refcnt);
1488 uma_zfree(V_tcpcb_zone, tp);
1489 released = in_pcbrele_wlocked(inp);
1490 KASSERT(!released, ("%s: inp %p should not have been released "
1491 "here", __func__, inp));
1496 tcp_timer_discard(void *ptp)
1501 tp = (struct tcpcb *)ptp;
1502 CURVNET_SET(tp->t_vnet);
1503 INP_INFO_RLOCK(&V_tcbinfo);
1505 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1508 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1509 ("%s: tcpcb has to be stopped here", __func__));
1510 tp->t_timers->tt_draincnt--;
1511 if (tp->t_timers->tt_draincnt == 0) {
1512 /* We own the last reference on this tcpcb, let's free it. */
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);
1525 INP_INFO_RUNLOCK(&V_tcbinfo);
1530 * Attempt to close a TCP control block, marking it as dropped, and freeing
1531 * the socket if we hold the only reference.
1534 tcp_close(struct tcpcb *tp)
1536 struct inpcb *inp = tp->t_inpcb;
1539 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1540 INP_WLOCK_ASSERT(inp);
1543 if (tp->t_state == TCPS_LISTEN)
1544 tcp_offload_listen_stop(tp);
1548 * This releases the TFO pending counter resource for TFO listen
1549 * sockets as well as passively-created TFO sockets that transition
1550 * from SYN_RECEIVED to CLOSED.
1552 if (tp->t_tfo_pending) {
1553 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1554 tp->t_tfo_pending = NULL;
1558 TCPSTAT_INC(tcps_closed);
1559 TCPSTATES_DEC(tp->t_state);
1560 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1561 so = inp->inp_socket;
1562 soisdisconnected(so);
1563 if (inp->inp_flags & INP_SOCKREF) {
1564 KASSERT(so->so_state & SS_PROTOREF,
1565 ("tcp_close: !SS_PROTOREF"));
1566 inp->inp_flags &= ~INP_SOCKREF;
1570 so->so_state &= ~SS_PROTOREF;
1580 VNET_ITERATOR_DECL(vnet_iter);
1585 VNET_LIST_RLOCK_NOSLEEP();
1586 VNET_FOREACH(vnet_iter) {
1587 CURVNET_SET(vnet_iter);
1592 * Walk the tcpbs, if existing, and flush the reassembly queue,
1593 * if there is one...
1594 * XXX: The "Net/3" implementation doesn't imply that the TCP
1595 * reassembly queue should be flushed, but in a situation
1596 * where we're really low on mbufs, this is potentially
1599 INP_INFO_WLOCK(&V_tcbinfo);
1600 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1601 if (inpb->inp_flags & INP_TIMEWAIT)
1604 if ((tcpb = intotcpcb(inpb)) != NULL) {
1605 tcp_reass_flush(tcpb);
1606 tcp_clean_sackreport(tcpb);
1610 INP_INFO_WUNLOCK(&V_tcbinfo);
1613 VNET_LIST_RUNLOCK_NOSLEEP();
1617 * Notify a tcp user of an asynchronous error;
1618 * store error as soft error, but wake up user
1619 * (for now, won't do anything until can select for soft error).
1621 * Do not wake up user since there currently is no mechanism for
1622 * reporting soft errors (yet - a kqueue filter may be added).
1624 static struct inpcb *
1625 tcp_notify(struct inpcb *inp, int error)
1629 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1630 INP_WLOCK_ASSERT(inp);
1632 if ((inp->inp_flags & INP_TIMEWAIT) ||
1633 (inp->inp_flags & INP_DROPPED))
1636 tp = intotcpcb(inp);
1637 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1640 * Ignore some errors if we are hooked up.
1641 * If connection hasn't completed, has retransmitted several times,
1642 * and receives a second error, give up now. This is better
1643 * than waiting a long time to establish a connection that
1644 * can never complete.
1646 if (tp->t_state == TCPS_ESTABLISHED &&
1647 (error == EHOSTUNREACH || error == ENETUNREACH ||
1648 error == EHOSTDOWN)) {
1649 if (inp->inp_route.ro_rt) {
1650 RTFREE(inp->inp_route.ro_rt);
1651 inp->inp_route.ro_rt = (struct rtentry *)NULL;
1654 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1656 tp = tcp_drop(tp, error);
1662 tp->t_softerror = error;
1666 wakeup( &so->so_timeo);
1673 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1675 int error, i, m, n, pcb_count;
1676 struct inpcb *inp, **inp_list;
1681 * The process of preparing the TCB list is too time-consuming and
1682 * resource-intensive to repeat twice on every request.
1684 if (req->oldptr == NULL) {
1685 n = V_tcbinfo.ipi_count +
1686 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1687 n += imax(n / 8, 10);
1688 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1692 if (req->newptr != NULL)
1696 * OK, now we're committed to doing something.
1698 INP_LIST_RLOCK(&V_tcbinfo);
1699 gencnt = V_tcbinfo.ipi_gencnt;
1700 n = V_tcbinfo.ipi_count;
1701 INP_LIST_RUNLOCK(&V_tcbinfo);
1703 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1705 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1706 + (n + m) * sizeof(struct xtcpcb));
1710 xig.xig_len = sizeof xig;
1711 xig.xig_count = n + m;
1712 xig.xig_gen = gencnt;
1713 xig.xig_sogen = so_gencnt;
1714 error = SYSCTL_OUT(req, &xig, sizeof xig);
1718 error = syncache_pcblist(req, m, &pcb_count);
1722 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1724 INP_INFO_WLOCK(&V_tcbinfo);
1725 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1726 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1728 if (inp->inp_gencnt <= gencnt) {
1730 * XXX: This use of cr_cansee(), introduced with
1731 * TCP state changes, is not quite right, but for
1732 * now, better than nothing.
1734 if (inp->inp_flags & INP_TIMEWAIT) {
1735 if (intotw(inp) != NULL)
1736 error = cr_cansee(req->td->td_ucred,
1737 intotw(inp)->tw_cred);
1739 error = EINVAL; /* Skip this inp. */
1741 error = cr_canseeinpcb(req->td->td_ucred, inp);
1744 inp_list[i++] = inp;
1749 INP_INFO_WUNLOCK(&V_tcbinfo);
1753 for (i = 0; i < n; i++) {
1756 if (inp->inp_gencnt <= gencnt) {
1760 bzero(&xt, sizeof(xt));
1761 xt.xt_len = sizeof xt;
1762 /* XXX should avoid extra copy */
1763 bcopy(inp, &xt.xt_inp, sizeof *inp);
1764 inp_ppcb = inp->inp_ppcb;
1765 if (inp_ppcb == NULL)
1766 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1767 else if (inp->inp_flags & INP_TIMEWAIT) {
1768 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1769 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1771 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1772 if (xt.xt_tp.t_timers)
1773 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1775 if (inp->inp_socket != NULL)
1776 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1778 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1779 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1781 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1783 error = SYSCTL_OUT(req, &xt, sizeof xt);
1787 INP_INFO_RLOCK(&V_tcbinfo);
1788 for (i = 0; i < n; i++) {
1791 if (!in_pcbrele_rlocked(inp))
1794 INP_INFO_RUNLOCK(&V_tcbinfo);
1798 * Give the user an updated idea of our state.
1799 * If the generation differs from what we told
1800 * her before, she knows that something happened
1801 * while we were processing this request, and it
1802 * might be necessary to retry.
1804 INP_LIST_RLOCK(&V_tcbinfo);
1805 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1806 xig.xig_sogen = so_gencnt;
1807 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1808 INP_LIST_RUNLOCK(&V_tcbinfo);
1809 error = SYSCTL_OUT(req, &xig, sizeof xig);
1811 free(inp_list, M_TEMP);
1815 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1816 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1817 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1821 tcp_getcred(SYSCTL_HANDLER_ARGS)
1824 struct sockaddr_in addrs[2];
1828 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1831 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1834 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1835 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1837 if (inp->inp_socket == NULL)
1840 error = cr_canseeinpcb(req->td->td_ucred, inp);
1842 cru2x(inp->inp_cred, &xuc);
1847 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1851 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1852 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1853 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1858 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1861 struct sockaddr_in6 addrs[2];
1868 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1871 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1874 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1875 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1878 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1880 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1889 inp = in_pcblookup(&V_tcbinfo,
1890 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1892 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1893 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1896 inp = in6_pcblookup(&V_tcbinfo,
1897 &addrs[1].sin6_addr, addrs[1].sin6_port,
1898 &addrs[0].sin6_addr, addrs[0].sin6_port,
1899 INPLOOKUP_RLOCKPCB, NULL);
1901 if (inp->inp_socket == NULL)
1904 error = cr_canseeinpcb(req->td->td_ucred, inp);
1906 cru2x(inp->inp_cred, &xuc);
1911 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1915 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1916 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1917 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1923 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1925 struct ip *ip = vip;
1927 struct in_addr faddr;
1930 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1932 struct in_conninfo inc;
1933 tcp_seq icmp_tcp_seq;
1936 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1937 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1940 if (cmd == PRC_MSGSIZE)
1941 notify = tcp_mtudisc_notify;
1942 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1943 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1944 notify = tcp_drop_syn_sent;
1945 else if (PRC_IS_REDIRECT(cmd)) {
1946 /* signal EHOSTDOWN, as it flushes the cached route */
1947 in_pcbnotifyall(&V_tcbinfo, faddr, EHOSTDOWN, notify);
1951 * Hostdead is ugly because it goes linearly through all PCBs.
1952 * XXX: We never get this from ICMP, otherwise it makes an
1953 * excellent DoS attack on machines with many connections.
1955 else if (cmd == PRC_HOSTDEAD)
1957 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1961 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1965 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1966 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1967 INP_INFO_RLOCK(&V_tcbinfo);
1968 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
1969 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1971 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1972 !(inp->inp_flags & INP_DROPPED) &&
1973 !(inp->inp_socket == NULL)) {
1974 icmp_tcp_seq = ntohl(th->th_seq);
1975 tp = intotcpcb(inp);
1976 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1977 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1978 if (cmd == PRC_MSGSIZE) {
1981 * If we got a needfrag set the MTU
1982 * in the route to the suggested new
1983 * value (if given) and then notify.
1985 mtu = ntohs(icp->icmp_nextmtu);
1987 * If no alternative MTU was
1988 * proposed, try the next smaller
1993 ntohs(ip->ip_len), 1);
1994 if (mtu < V_tcp_minmss +
1995 sizeof(struct tcpiphdr))
1996 mtu = V_tcp_minmss +
1997 sizeof(struct tcpiphdr);
1999 * Only process the offered MTU if it
2000 * is smaller than the current one.
2002 if (mtu < tp->t_maxseg +
2003 sizeof(struct tcpiphdr)) {
2004 bzero(&inc, sizeof(inc));
2005 inc.inc_faddr = faddr;
2007 inp->inp_inc.inc_fibnum;
2008 tcp_hc_updatemtu(&inc, mtu);
2009 tcp_mtudisc(inp, mtu);
2012 inp = (*notify)(inp,
2013 inetctlerrmap[cmd]);
2019 bzero(&inc, sizeof(inc));
2020 inc.inc_fport = th->th_dport;
2021 inc.inc_lport = th->th_sport;
2022 inc.inc_faddr = faddr;
2023 inc.inc_laddr = ip->ip_src;
2024 syncache_unreach(&inc, th);
2026 INP_INFO_RUNLOCK(&V_tcbinfo);
2032 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2035 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2036 struct ip6_hdr *ip6;
2038 struct ip6ctlparam *ip6cp = NULL;
2039 const struct sockaddr_in6 *sa6_src = NULL;
2041 struct tcp_portonly {
2046 if (sa->sa_family != AF_INET6 ||
2047 sa->sa_len != sizeof(struct sockaddr_in6))
2050 if (cmd == PRC_MSGSIZE)
2051 notify = tcp_mtudisc_notify;
2052 else if (!PRC_IS_REDIRECT(cmd) &&
2053 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
2056 /* if the parameter is from icmp6, decode it. */
2058 ip6cp = (struct ip6ctlparam *)d;
2060 ip6 = ip6cp->ip6c_ip6;
2061 off = ip6cp->ip6c_off;
2062 sa6_src = ip6cp->ip6c_src;
2066 off = 0; /* fool gcc */
2071 struct in_conninfo inc;
2073 * XXX: We assume that when IPV6 is non NULL,
2074 * M and OFF are valid.
2077 /* check if we can safely examine src and dst ports */
2078 if (m->m_pkthdr.len < off + sizeof(*thp))
2081 bzero(&th, sizeof(th));
2082 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
2084 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
2085 (struct sockaddr *)ip6cp->ip6c_src,
2086 th.th_sport, cmd, NULL, notify);
2088 bzero(&inc, sizeof(inc));
2089 inc.inc_fport = th.th_dport;
2090 inc.inc_lport = th.th_sport;
2091 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
2092 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
2093 inc.inc_flags |= INC_ISIPV6;
2094 INP_INFO_RLOCK(&V_tcbinfo);
2095 syncache_unreach(&inc, &th);
2096 INP_INFO_RUNLOCK(&V_tcbinfo);
2098 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
2099 0, cmd, NULL, notify);
2105 * Following is where TCP initial sequence number generation occurs.
2107 * There are two places where we must use initial sequence numbers:
2108 * 1. In SYN-ACK packets.
2109 * 2. In SYN packets.
2111 * All ISNs for SYN-ACK packets are generated by the syncache. See
2112 * tcp_syncache.c for details.
2114 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2115 * depends on this property. In addition, these ISNs should be
2116 * unguessable so as to prevent connection hijacking. To satisfy
2117 * the requirements of this situation, the algorithm outlined in
2118 * RFC 1948 is used, with only small modifications.
2120 * Implementation details:
2122 * Time is based off the system timer, and is corrected so that it
2123 * increases by one megabyte per second. This allows for proper
2124 * recycling on high speed LANs while still leaving over an hour
2127 * As reading the *exact* system time is too expensive to be done
2128 * whenever setting up a TCP connection, we increment the time
2129 * offset in two ways. First, a small random positive increment
2130 * is added to isn_offset for each connection that is set up.
2131 * Second, the function tcp_isn_tick fires once per clock tick
2132 * and increments isn_offset as necessary so that sequence numbers
2133 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2134 * random positive increments serve only to ensure that the same
2135 * exact sequence number is never sent out twice (as could otherwise
2136 * happen when a port is recycled in less than the system tick
2139 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2140 * between seeding of isn_secret. This is normally set to zero,
2141 * as reseeding should not be necessary.
2143 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2144 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
2145 * general, this means holding an exclusive (write) lock.
2148 #define ISN_BYTES_PER_SECOND 1048576
2149 #define ISN_STATIC_INCREMENT 4096
2150 #define ISN_RANDOM_INCREMENT (4096 - 1)
2152 static VNET_DEFINE(u_char, isn_secret[32]);
2153 static VNET_DEFINE(int, isn_last);
2154 static VNET_DEFINE(int, isn_last_reseed);
2155 static VNET_DEFINE(u_int32_t, isn_offset);
2156 static VNET_DEFINE(u_int32_t, isn_offset_old);
2158 #define V_isn_secret VNET(isn_secret)
2159 #define V_isn_last VNET(isn_last)
2160 #define V_isn_last_reseed VNET(isn_last_reseed)
2161 #define V_isn_offset VNET(isn_offset)
2162 #define V_isn_offset_old VNET(isn_offset_old)
2165 tcp_new_isn(struct tcpcb *tp)
2168 u_int32_t md5_buffer[4];
2170 u_int32_t projected_offset;
2172 INP_WLOCK_ASSERT(tp->t_inpcb);
2175 /* Seed if this is the first use, reseed if requested. */
2176 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2177 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2179 read_random(&V_isn_secret, sizeof(V_isn_secret));
2180 V_isn_last_reseed = ticks;
2183 /* Compute the md5 hash and return the ISN. */
2185 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2186 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2188 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2189 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2190 sizeof(struct in6_addr));
2191 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2192 sizeof(struct in6_addr));
2196 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2197 sizeof(struct in_addr));
2198 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2199 sizeof(struct in_addr));
2201 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2202 MD5Final((u_char *) &md5_buffer, &isn_ctx);
2203 new_isn = (tcp_seq) md5_buffer[0];
2204 V_isn_offset += ISN_STATIC_INCREMENT +
2205 (arc4random() & ISN_RANDOM_INCREMENT);
2206 if (ticks != V_isn_last) {
2207 projected_offset = V_isn_offset_old +
2208 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2209 if (SEQ_GT(projected_offset, V_isn_offset))
2210 V_isn_offset = projected_offset;
2211 V_isn_offset_old = V_isn_offset;
2214 new_isn += V_isn_offset;
2220 * When a specific ICMP unreachable message is received and the
2221 * connection state is SYN-SENT, drop the connection. This behavior
2222 * is controlled by the icmp_may_rst sysctl.
2225 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2229 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2230 INP_WLOCK_ASSERT(inp);
2232 if ((inp->inp_flags & INP_TIMEWAIT) ||
2233 (inp->inp_flags & INP_DROPPED))
2236 tp = intotcpcb(inp);
2237 if (tp->t_state != TCPS_SYN_SENT)
2240 tp = tcp_drop(tp, errno);
2248 * When `need fragmentation' ICMP is received, update our idea of the MSS
2249 * based on the new value. Also nudge TCP to send something, since we
2250 * know the packet we just sent was dropped.
2251 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2253 static struct inpcb *
2254 tcp_mtudisc_notify(struct inpcb *inp, int error)
2257 tcp_mtudisc(inp, -1);
2262 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2267 INP_WLOCK_ASSERT(inp);
2268 if ((inp->inp_flags & INP_TIMEWAIT) ||
2269 (inp->inp_flags & INP_DROPPED))
2272 tp = intotcpcb(inp);
2273 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2275 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2277 so = inp->inp_socket;
2278 SOCKBUF_LOCK(&so->so_snd);
2279 /* If the mss is larger than the socket buffer, decrease the mss. */
2280 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2281 tp->t_maxseg = so->so_snd.sb_hiwat;
2282 SOCKBUF_UNLOCK(&so->so_snd);
2284 TCPSTAT_INC(tcps_mturesent);
2286 tp->snd_nxt = tp->snd_una;
2287 tcp_free_sackholes(tp);
2288 tp->snd_recover = tp->snd_max;
2289 if (tp->t_flags & TF_SACK_PERMIT)
2290 EXIT_FASTRECOVERY(tp->t_flags);
2291 tp->t_fb->tfb_tcp_output(tp);
2296 * Look-up the routing entry to the peer of this inpcb. If no route
2297 * is found and it cannot be allocated, then return 0. This routine
2298 * is called by TCP routines that access the rmx structure and by
2299 * tcp_mss_update to get the peer/interface MTU.
2302 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2304 struct nhop4_extended nh4;
2308 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2310 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2312 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2313 NHR_REF, 0, &nh4) != 0)
2317 maxmtu = nh4.nh_mtu;
2319 /* Report additional interface capabilities. */
2321 if (ifp->if_capenable & IFCAP_TSO4 &&
2322 ifp->if_hwassist & CSUM_TSO) {
2323 cap->ifcap |= CSUM_TSO;
2324 cap->tsomax = ifp->if_hw_tsomax;
2325 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2326 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2329 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2337 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2339 struct nhop6_extended nh6;
2340 struct in6_addr dst6;
2345 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2347 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2348 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2349 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2354 maxmtu = nh6.nh_mtu;
2356 /* Report additional interface capabilities. */
2358 if (ifp->if_capenable & IFCAP_TSO6 &&
2359 ifp->if_hwassist & CSUM_TSO) {
2360 cap->ifcap |= CSUM_TSO;
2361 cap->tsomax = ifp->if_hw_tsomax;
2362 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2363 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2366 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2374 * Calculate effective SMSS per RFC5681 definition for a given TCP
2375 * connection at its current state, taking into account SACK and etc.
2378 tcp_maxseg(const struct tcpcb *tp)
2382 if (tp->t_flags & TF_NOOPT)
2383 return (tp->t_maxseg);
2386 * Here we have a simplified code from tcp_addoptions(),
2387 * without a proper loop, and having most of paddings hardcoded.
2388 * We might make mistakes with padding here in some edge cases,
2389 * but this is harmless, since result of tcp_maxseg() is used
2390 * only in cwnd and ssthresh estimations.
2392 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2393 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2394 if (tp->t_flags & TF_RCVD_TSTMP)
2395 optlen = TCPOLEN_TSTAMP_APPA;
2398 #ifdef TCP_SIGNATURE
2399 if (tp->t_flags & TF_SIGNATURE)
2400 optlen += PAD(TCPOLEN_SIGNATURE);
2402 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2403 optlen += TCPOLEN_SACKHDR;
2404 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2405 optlen = PAD(optlen);
2408 if (tp->t_flags & TF_REQ_TSTMP)
2409 optlen = TCPOLEN_TSTAMP_APPA;
2411 optlen = PAD(TCPOLEN_MAXSEG);
2412 if (tp->t_flags & TF_REQ_SCALE)
2413 optlen += PAD(TCPOLEN_WINDOW);
2414 #ifdef TCP_SIGNATURE
2415 if (tp->t_flags & TF_SIGNATURE)
2416 optlen += PAD(TCPOLEN_SIGNATURE);
2418 if (tp->t_flags & TF_SACK_PERMIT)
2419 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2422 optlen = min(optlen, TCP_MAXOLEN);
2423 return (tp->t_maxseg - optlen);
2427 /* compute ESP/AH header size for TCP, including outer IP header. */
2429 ipsec_hdrsiz_tcp(struct tcpcb *tp)
2436 struct ip6_hdr *ip6;
2440 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) ||
2441 (!key_havesp(IPSEC_DIR_OUTBOUND)))
2443 m = m_gethdr(M_NOWAIT, MT_DATA);
2448 if ((inp->inp_vflag & INP_IPV6) != 0) {
2449 ip6 = mtod(m, struct ip6_hdr *);
2450 th = (struct tcphdr *)(ip6 + 1);
2451 m->m_pkthdr.len = m->m_len =
2452 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
2453 tcpip_fillheaders(inp, ip6, th);
2454 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2458 ip = mtod(m, struct ip *);
2459 th = (struct tcphdr *)(ip + 1);
2460 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
2461 tcpip_fillheaders(inp, ip, th);
2462 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2470 #ifdef TCP_SIGNATURE
2472 * Callback function invoked by m_apply() to digest TCP segment data
2473 * contained within an mbuf chain.
2476 tcp_signature_apply(void *fstate, void *data, u_int len)
2479 MD5Update(fstate, (u_char *)data, len);
2484 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2485 * search with the destination IP address, and a 'magic SPI' to be
2486 * determined by the application. This is hardcoded elsewhere to 1179
2489 tcp_get_sav(struct mbuf *m, u_int direction)
2491 union sockaddr_union dst;
2492 struct secasvar *sav;
2495 struct ip6_hdr *ip6;
2496 char ip6buf[INET6_ADDRSTRLEN];
2499 /* Extract the destination from the IP header in the mbuf. */
2500 bzero(&dst, sizeof(union sockaddr_union));
2501 ip = mtod(m, struct ip *);
2503 ip6 = NULL; /* Make the compiler happy. */
2508 dst.sa.sa_len = sizeof(struct sockaddr_in);
2509 dst.sa.sa_family = AF_INET;
2510 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2511 ip->ip_src : ip->ip_dst;
2515 case (IPV6_VERSION >> 4):
2516 ip6 = mtod(m, struct ip6_hdr *);
2517 dst.sa.sa_len = sizeof(struct sockaddr_in6);
2518 dst.sa.sa_family = AF_INET6;
2519 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2520 ip6->ip6_src : ip6->ip6_dst;
2529 /* Look up an SADB entry which matches the address of the peer. */
2530 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2532 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2533 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2535 (ip->ip_v == (IPV6_VERSION >> 4)) ?
2536 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2545 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2548 * m pointer to head of mbuf chain
2549 * len length of TCP segment data, excluding options
2550 * optlen length of TCP segment options
2551 * buf pointer to storage for computed MD5 digest
2552 * sav pointer to security assosiation
2554 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2555 * When called from tcp_input(), we can be sure that th_sum has been
2556 * zeroed out and verified already.
2558 * Releases reference to SADB key before return.
2560 * Return 0 if successful, otherwise return -1.
2564 tcp_signature_do_compute(struct mbuf *m, int len, int optlen,
2565 u_char *buf, struct secasvar *sav)
2568 struct ippseudo ippseudo;
2574 struct ipovly *ipovly;
2578 struct ip6_hdr *ip6;
2579 struct in6_addr in6;
2585 KASSERT(m != NULL, ("NULL mbuf chain"));
2586 KASSERT(buf != NULL, ("NULL signature pointer"));
2588 /* Extract the destination from the IP header in the mbuf. */
2589 ip = mtod(m, struct ip *);
2591 ip6 = NULL; /* Make the compiler happy. */
2596 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2598 * XXX The ippseudo header MUST be digested in network byte order,
2599 * or else we'll fail the regression test. Assume all fields we've
2600 * been doing arithmetic on have been in host byte order.
2601 * XXX One cannot depend on ipovly->ih_len here. When called from
2602 * tcp_output(), the underlying ip_len member has not yet been set.
2607 ipovly = (struct ipovly *)ip;
2608 ippseudo.ippseudo_src = ipovly->ih_src;
2609 ippseudo.ippseudo_dst = ipovly->ih_dst;
2610 ippseudo.ippseudo_pad = 0;
2611 ippseudo.ippseudo_p = IPPROTO_TCP;
2612 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2614 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2616 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2617 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2622 * RFC 2385, 2.0 Proposal
2623 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2624 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2625 * extended next header value (to form 32 bits), and 32-bit segment
2627 * Note: Upper-Layer Packet Length comes before Next Header.
2629 case (IPV6_VERSION >> 4):
2631 in6_clearscope(&in6);
2632 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2634 in6_clearscope(&in6);
2635 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2636 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2637 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2639 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2640 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2641 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2643 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2645 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2646 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2658 * Step 2: Update MD5 hash with TCP header, excluding options.
2659 * The TCP checksum must be set to zero.
2661 savecsum = th->th_sum;
2663 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2664 th->th_sum = savecsum;
2667 * Step 3: Update MD5 hash with TCP segment data.
2668 * Use m_apply() to avoid an early m_pullup().
2671 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2674 * Step 4: Update MD5 hash with shared secret.
2676 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2677 MD5Final(buf, &ctx);
2679 key_sa_recordxfer(sav, m);
2685 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2687 * Return 0 if successful, otherwise return -1.
2690 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2691 u_char *buf, u_int direction)
2693 struct secasvar *sav;
2695 if ((sav = tcp_get_sav(m, direction)) == NULL)
2698 return (tcp_signature_do_compute(m, len, optlen, buf, sav));
2702 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2705 * m pointer to head of mbuf chain
2706 * len length of TCP segment data, excluding options
2707 * optlen length of TCP segment options
2708 * buf pointer to storage for computed MD5 digest
2709 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2711 * Return 1 if successful, otherwise return 0.
2714 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2715 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2717 char tmpdigest[TCP_SIGLEN];
2719 if (tcp_sig_checksigs == 0)
2721 if ((tcpbflag & TF_SIGNATURE) == 0) {
2722 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2725 * If this socket is not expecting signature but
2726 * the segment contains signature just fail.
2728 TCPSTAT_INC(tcps_sig_err_sigopt);
2729 TCPSTAT_INC(tcps_sig_rcvbadsig);
2733 /* Signature is not expected, and not present in segment. */
2738 * If this socket is expecting signature but the segment does not
2739 * contain any just fail.
2741 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2742 TCPSTAT_INC(tcps_sig_err_nosigopt);
2743 TCPSTAT_INC(tcps_sig_rcvbadsig);
2746 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2747 IPSEC_DIR_INBOUND) == -1) {
2748 TCPSTAT_INC(tcps_sig_err_buildsig);
2749 TCPSTAT_INC(tcps_sig_rcvbadsig);
2753 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2754 TCPSTAT_INC(tcps_sig_rcvbadsig);
2757 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2760 #endif /* TCP_SIGNATURE */
2763 sysctl_drop(SYSCTL_HANDLER_ARGS)
2765 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2766 struct sockaddr_storage addrs[2];
2770 struct sockaddr_in *fin, *lin;
2772 struct sockaddr_in6 *fin6, *lin6;
2783 if (req->oldptr != NULL || req->oldlen != 0)
2785 if (req->newptr == NULL)
2787 if (req->newlen < sizeof(addrs))
2789 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2793 switch (addrs[0].ss_family) {
2796 fin6 = (struct sockaddr_in6 *)&addrs[0];
2797 lin6 = (struct sockaddr_in6 *)&addrs[1];
2798 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2799 lin6->sin6_len != sizeof(struct sockaddr_in6))
2801 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2802 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2804 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2805 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2806 fin = (struct sockaddr_in *)&addrs[0];
2807 lin = (struct sockaddr_in *)&addrs[1];
2810 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2813 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2820 fin = (struct sockaddr_in *)&addrs[0];
2821 lin = (struct sockaddr_in *)&addrs[1];
2822 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2823 lin->sin_len != sizeof(struct sockaddr_in))
2830 INP_INFO_RLOCK(&V_tcbinfo);
2831 switch (addrs[0].ss_family) {
2834 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2835 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2836 INPLOOKUP_WLOCKPCB, NULL);
2841 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2842 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2847 if (inp->inp_flags & INP_TIMEWAIT) {
2849 * XXXRW: There currently exists a state where an
2850 * inpcb is present, but its timewait state has been
2851 * discarded. For now, don't allow dropping of this
2859 } else if (!(inp->inp_flags & INP_DROPPED) &&
2860 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2861 tp = intotcpcb(inp);
2862 tp = tcp_drop(tp, ECONNABORTED);
2869 INP_INFO_RUNLOCK(&V_tcbinfo);
2873 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2874 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2875 0, sysctl_drop, "", "Drop TCP connection");
2878 * Generate a standardized TCP log line for use throughout the
2879 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2880 * allow use in the interrupt context.
2882 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2883 * NB: The function may return NULL if memory allocation failed.
2885 * Due to header inclusion and ordering limitations the struct ip
2886 * and ip6_hdr pointers have to be passed as void pointers.
2889 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2893 /* Is logging enabled? */
2894 if (tcp_log_in_vain == 0)
2897 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2901 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2905 /* Is logging enabled? */
2906 if (tcp_log_debug == 0)
2909 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2913 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2920 const struct ip6_hdr *ip6;
2922 ip6 = (const struct ip6_hdr *)ip6hdr;
2924 ip = (struct ip *)ip4hdr;
2927 * The log line looks like this:
2928 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2930 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2931 sizeof(PRINT_TH_FLAGS) + 1 +
2933 2 * INET6_ADDRSTRLEN;
2935 2 * INET_ADDRSTRLEN;
2938 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2942 strcat(s, "TCP: [");
2945 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2946 inet_ntoa_r(inc->inc_faddr, sp);
2948 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2950 inet_ntoa_r(inc->inc_laddr, sp);
2952 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2955 ip6_sprintf(sp, &inc->inc6_faddr);
2957 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2959 ip6_sprintf(sp, &inc->inc6_laddr);
2961 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2962 } else if (ip6 && th) {
2963 ip6_sprintf(sp, &ip6->ip6_src);
2965 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2967 ip6_sprintf(sp, &ip6->ip6_dst);
2969 sprintf(sp, "]:%i", ntohs(th->th_dport));
2972 } else if (ip && th) {
2973 inet_ntoa_r(ip->ip_src, sp);
2975 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2977 inet_ntoa_r(ip->ip_dst, sp);
2979 sprintf(sp, "]:%i", ntohs(th->th_dport));
2987 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2988 if (*(s + size - 1) != '\0')
2989 panic("%s: string too long", __func__);
2994 * A subroutine which makes it easy to track TCP state changes with DTrace.
2995 * This function shouldn't be called for t_state initializations that don't
2996 * correspond to actual TCP state transitions.
2999 tcp_state_change(struct tcpcb *tp, int newstate)
3001 #if defined(KDTRACE_HOOKS)
3002 int pstate = tp->t_state;
3005 TCPSTATES_DEC(tp->t_state);
3006 TCPSTATES_INC(newstate);
3007 tp->t_state = newstate;
3008 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);