2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5 * The Regents of the University of California. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
38 #include "opt_inet6.h"
39 #include "opt_ipsec.h"
40 #include "opt_tcpdebug.h"
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/callout.h>
45 #include <sys/eventhandler.h>
47 #include <sys/hhook.h>
49 #include <sys/kernel.h>
51 #include <sys/khelp.h>
53 #include <sys/sysctl.h>
55 #include <sys/malloc.h>
56 #include <sys/refcount.h>
59 #include <sys/domain.h>
64 #include <sys/socket.h>
65 #include <sys/socketvar.h>
66 #include <sys/protosw.h>
67 #include <sys/random.h>
71 #include <net/route.h>
73 #include <net/if_var.h>
76 #include <netinet/in.h>
77 #include <netinet/in_fib.h>
78 #include <netinet/in_kdtrace.h>
79 #include <netinet/in_pcb.h>
80 #include <netinet/in_systm.h>
81 #include <netinet/in_var.h>
82 #include <netinet/ip.h>
83 #include <netinet/ip_icmp.h>
84 #include <netinet/ip_var.h>
86 #include <netinet/icmp6.h>
87 #include <netinet/ip6.h>
88 #include <netinet6/in6_fib.h>
89 #include <netinet6/in6_pcb.h>
90 #include <netinet6/ip6_var.h>
91 #include <netinet6/scope6_var.h>
92 #include <netinet6/nd6.h>
95 #include <netinet/tcp.h>
96 #include <netinet/tcp_fsm.h>
97 #include <netinet/tcp_seq.h>
98 #include <netinet/tcp_timer.h>
99 #include <netinet/tcp_var.h>
100 #include <netinet/tcp_log_buf.h>
101 #include <netinet/tcp_syncache.h>
102 #include <netinet/cc/cc.h>
104 #include <netinet6/tcp6_var.h>
106 #include <netinet/tcpip.h>
107 #include <netinet/tcp_fastopen.h>
109 #include <netinet/tcp_pcap.h>
112 #include <netinet/tcp_debug.h>
115 #include <netinet6/ip6protosw.h>
118 #include <netinet/tcp_offload.h>
121 #include <netipsec/ipsec_support.h>
123 #include <machine/in_cksum.h>
126 #include <security/mac/mac_framework.h>
128 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
130 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
133 struct rwlock tcp_function_lock;
136 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
141 error = sysctl_handle_int(oidp, &new, 0, req);
142 if (error == 0 && req->newptr) {
143 if (new < TCP_MINMSS)
151 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
152 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
153 &sysctl_net_inet_tcp_mss_check, "I",
154 "Default TCP Maximum Segment Size");
158 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
162 new = V_tcp_v6mssdflt;
163 error = sysctl_handle_int(oidp, &new, 0, req);
164 if (error == 0 && req->newptr) {
165 if (new < TCP_MINMSS)
168 V_tcp_v6mssdflt = new;
173 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
174 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
175 &sysctl_net_inet_tcp_mss_v6_check, "I",
176 "Default TCP Maximum Segment Size for IPv6");
180 * Minimum MSS we accept and use. This prevents DoS attacks where
181 * we are forced to a ridiculous low MSS like 20 and send hundreds
182 * of packets instead of one. The effect scales with the available
183 * bandwidth and quickly saturates the CPU and network interface
184 * with packet generation and sending. Set to zero to disable MINMSS
185 * checking. This setting prevents us from sending too small packets.
187 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
189 &VNET_NAME(tcp_minmss), 0,
190 "Minimum TCP Maximum Segment Size");
192 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
193 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
194 &VNET_NAME(tcp_do_rfc1323), 0,
195 "Enable rfc1323 (high performance TCP) extensions");
197 static int tcp_log_debug = 0;
198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
199 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
201 static int tcp_tcbhashsize;
202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
203 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
205 static int do_tcpdrain = 1;
206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
207 "Enable tcp_drain routine for extra help when low on mbufs");
209 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
210 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
212 static VNET_DEFINE(int, icmp_may_rst) = 1;
213 #define V_icmp_may_rst VNET(icmp_may_rst)
214 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
215 &VNET_NAME(icmp_may_rst), 0,
216 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
218 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
219 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
221 &VNET_NAME(tcp_isn_reseed_interval), 0,
222 "Seconds between reseeding of ISN secret");
224 static int tcp_soreceive_stream;
225 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
226 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
228 VNET_DEFINE(uma_zone_t, sack_hole_zone);
229 #define V_sack_hole_zone VNET(sack_hole_zone)
232 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
235 static int tcp_default_fb_init(struct tcpcb *tp);
236 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
237 static int tcp_default_handoff_ok(struct tcpcb *tp);
238 static struct inpcb *tcp_notify(struct inpcb *, int);
239 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
240 static void tcp_mtudisc(struct inpcb *, int);
241 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
242 void *ip4hdr, const void *ip6hdr);
245 static struct tcp_function_block tcp_def_funcblk = {
246 .tfb_tcp_block_name = "freebsd",
247 .tfb_tcp_output = tcp_output,
248 .tfb_tcp_do_segment = tcp_do_segment,
249 .tfb_tcp_ctloutput = tcp_default_ctloutput,
250 .tfb_tcp_handoff_ok = tcp_default_handoff_ok,
251 .tfb_tcp_fb_init = tcp_default_fb_init,
252 .tfb_tcp_fb_fini = tcp_default_fb_fini,
255 int t_functions_inited = 0;
256 static int tcp_fb_cnt = 0;
257 struct tcp_funchead t_functions;
258 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
261 init_tcp_functions(void)
263 if (t_functions_inited == 0) {
264 TAILQ_INIT(&t_functions);
265 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
266 t_functions_inited = 1;
270 static struct tcp_function_block *
271 find_tcp_functions_locked(struct tcp_function_set *fs)
273 struct tcp_function *f;
274 struct tcp_function_block *blk=NULL;
276 TAILQ_FOREACH(f, &t_functions, tf_next) {
277 if (strcmp(f->tf_name, fs->function_set_name) == 0) {
285 static struct tcp_function_block *
286 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
288 struct tcp_function_block *rblk=NULL;
289 struct tcp_function *f;
291 TAILQ_FOREACH(f, &t_functions, tf_next) {
292 if (f->tf_fb == blk) {
303 struct tcp_function_block *
304 find_and_ref_tcp_functions(struct tcp_function_set *fs)
306 struct tcp_function_block *blk;
308 rw_rlock(&tcp_function_lock);
309 blk = find_tcp_functions_locked(fs);
311 refcount_acquire(&blk->tfb_refcnt);
312 rw_runlock(&tcp_function_lock);
316 struct tcp_function_block *
317 find_and_ref_tcp_fb(struct tcp_function_block *blk)
319 struct tcp_function_block *rblk;
321 rw_rlock(&tcp_function_lock);
322 rblk = find_tcp_fb_locked(blk, NULL);
324 refcount_acquire(&rblk->tfb_refcnt);
325 rw_runlock(&tcp_function_lock);
329 static struct tcp_function_block *
330 find_and_ref_tcp_default_fb(void)
332 struct tcp_function_block *rblk;
334 rw_rlock(&tcp_function_lock);
335 rblk = tcp_func_set_ptr;
336 refcount_acquire(&rblk->tfb_refcnt);
337 rw_runlock(&tcp_function_lock);
342 tcp_switch_back_to_default(struct tcpcb *tp)
344 struct tcp_function_block *tfb;
346 KASSERT(tp->t_fb != &tcp_def_funcblk,
347 ("%s: called by the built-in default stack", __func__));
350 * Release the old stack. This function will either find a new one
353 if (tp->t_fb->tfb_tcp_fb_fini != NULL)
354 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
355 refcount_release(&tp->t_fb->tfb_refcnt);
358 * Now, we'll find a new function block to use.
359 * Start by trying the current user-selected
360 * default, unless this stack is the user-selected
363 tfb = find_and_ref_tcp_default_fb();
364 if (tfb == tp->t_fb) {
365 refcount_release(&tfb->tfb_refcnt);
368 /* Does the stack accept this connection? */
369 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
370 (*tfb->tfb_tcp_handoff_ok)(tp)) {
371 refcount_release(&tfb->tfb_refcnt);
374 /* Try to use that stack. */
376 /* Initialize the new stack. If it succeeds, we are done. */
378 if (tp->t_fb->tfb_tcp_fb_init == NULL ||
379 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
383 * Initialization failed. Release the reference count on
386 refcount_release(&tfb->tfb_refcnt);
390 * If that wasn't feasible, use the built-in default
391 * stack which is not allowed to reject anyone.
393 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
395 /* there always should be a default */
396 panic("Can't refer to tcp_def_funcblk");
398 if (tfb->tfb_tcp_handoff_ok != NULL) {
399 if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
400 /* The default stack cannot say no */
401 panic("Default stack rejects a new session?");
405 if (tp->t_fb->tfb_tcp_fb_init != NULL &&
406 (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
407 /* The default stack cannot fail */
408 panic("Default stack initialization failed");
413 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
416 struct tcp_function_set fs;
417 struct tcp_function_block *blk;
419 memset(&fs, 0, sizeof(fs));
420 rw_rlock(&tcp_function_lock);
421 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
424 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
425 fs.pcbcnt = blk->tfb_refcnt;
427 rw_runlock(&tcp_function_lock);
428 error = sysctl_handle_string(oidp, fs.function_set_name,
429 sizeof(fs.function_set_name), req);
431 /* Check for error or no change */
432 if (error != 0 || req->newptr == NULL)
435 rw_wlock(&tcp_function_lock);
436 blk = find_tcp_functions_locked(&fs);
438 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
442 tcp_func_set_ptr = blk;
444 rw_wunlock(&tcp_function_lock);
448 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
449 CTLTYPE_STRING | CTLFLAG_RW,
450 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
451 "Set/get the default TCP functions");
454 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
456 int error, cnt, linesz;
457 struct tcp_function *f;
463 rw_rlock(&tcp_function_lock);
464 TAILQ_FOREACH(f, &t_functions, tf_next) {
467 rw_runlock(&tcp_function_lock);
469 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
470 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
475 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
476 "Alias", "PCB count");
481 rw_rlock(&tcp_function_lock);
482 TAILQ_FOREACH(f, &t_functions, tf_next) {
483 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
484 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
485 f->tf_fb->tfb_tcp_block_name,
486 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
487 alias ? f->tf_name : "-",
488 f->tf_fb->tfb_refcnt);
489 if (linesz >= bufsz) {
497 rw_runlock(&tcp_function_lock);
499 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
500 free(buffer, M_TEMP);
504 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
505 CTLTYPE_STRING|CTLFLAG_RD,
506 NULL, 0, sysctl_net_inet_list_available, "A",
507 "list available TCP Function sets");
510 * Exports one (struct tcp_function_info) for each alias/name.
513 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
516 struct tcp_function *f;
517 struct tcp_function_info tfi;
520 * We don't allow writes.
522 if (req->newptr != NULL)
526 * Wire the old buffer so we can directly copy the functions to
527 * user space without dropping the lock.
529 if (req->oldptr != NULL) {
530 error = sysctl_wire_old_buffer(req, 0);
536 * Walk the list and copy out matching entries. If INVARIANTS
537 * is compiled in, also walk the list to verify the length of
538 * the list matches what we have recorded.
540 rw_rlock(&tcp_function_lock);
544 if (req->oldptr == NULL) {
549 TAILQ_FOREACH(f, &t_functions, tf_next) {
553 if (req->oldptr != NULL) {
554 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
555 tfi.tfi_id = f->tf_fb->tfb_id;
556 (void)strncpy(tfi.tfi_alias, f->tf_name,
557 TCP_FUNCTION_NAME_LEN_MAX);
558 tfi.tfi_alias[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
559 (void)strncpy(tfi.tfi_name,
560 f->tf_fb->tfb_tcp_block_name,
561 TCP_FUNCTION_NAME_LEN_MAX);
562 tfi.tfi_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
563 error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
565 * Don't stop on error, as that is the
566 * mechanism we use to accumulate length
567 * information if the buffer was too short.
571 KASSERT(cnt == tcp_fb_cnt,
572 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
576 rw_runlock(&tcp_function_lock);
577 if (req->oldptr == NULL)
578 error = SYSCTL_OUT(req, NULL,
579 (cnt + 1) * sizeof(struct tcp_function_info));
584 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
585 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
586 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
587 "List TCP function block name-to-ID mappings");
590 * tfb_tcp_handoff_ok() function for the default stack.
591 * Note that we'll basically try to take all comers.
594 tcp_default_handoff_ok(struct tcpcb *tp)
601 * tfb_tcp_fb_init() function for the default stack.
603 * This handles making sure we have appropriate timers set if you are
604 * transitioning a socket that has some amount of setup done.
606 * The init() fuction from the default can *never* return non-zero i.e.
607 * it is required to always succeed since it is the stack of last resort!
610 tcp_default_fb_init(struct tcpcb *tp)
615 INP_WLOCK_ASSERT(tp->t_inpcb);
617 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
618 ("%s: connection %p in unexpected state %d", __func__, tp,
622 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
623 * know what to do for unexpected states (which includes TIME_WAIT).
625 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
629 * Make sure some kind of transmission timer is set if there is
632 so = tp->t_inpcb->inp_socket;
633 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
634 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
635 tcp_timer_active(tp, TT_PERSIST))) {
637 * If the session has established and it looks like it should
638 * be in the persist state, set the persist timer. Otherwise,
639 * set the retransmit timer.
641 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
642 (int32_t)(tp->snd_nxt - tp->snd_una) <
643 (int32_t)sbavail(&so->so_snd))
646 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
649 /* All non-embryonic sessions get a keepalive timer. */
650 if (!tcp_timer_active(tp, TT_KEEP))
651 tcp_timer_activate(tp, TT_KEEP,
652 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
659 * tfb_tcp_fb_fini() function for the default stack.
661 * This changes state as necessary (or prudent) to prepare for another stack
662 * to assume responsibility for the connection.
665 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
668 INP_WLOCK_ASSERT(tp->t_inpcb);
673 * Target size of TCP PCB hash tables. Must be a power of two.
675 * Note that this can be overridden by the kernel environment
676 * variable net.inet.tcp.tcbhashsize
679 #define TCBHASHSIZE 0
684 * Callouts should be moved into struct tcp directly. They are currently
685 * separate because the tcpcb structure is exported to userland for sysctl
686 * parsing purposes, which do not know about callouts.
697 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
698 #define V_tcpcb_zone VNET(tcpcb_zone)
700 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
701 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
703 static struct mtx isn_mtx;
705 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
706 #define ISN_LOCK() mtx_lock(&isn_mtx)
707 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
710 * TCP initialization.
713 tcp_zone_change(void *tag)
716 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
717 uma_zone_set_max(V_tcpcb_zone, maxsockets);
718 tcp_tw_zone_change();
722 tcp_inpcb_init(void *mem, int size, int flags)
724 struct inpcb *inp = mem;
726 INP_LOCK_INIT(inp, "inp", "tcpinp");
731 * Take a value and get the next power of 2 that doesn't overflow.
732 * Used to size the tcp_inpcb hash buckets.
735 maketcp_hashsize(int size)
741 * get the next power of 2 higher than maxsockets.
743 hashsize = 1 << fls(size);
744 /* catch overflow, and just go one power of 2 smaller */
745 if (hashsize < size) {
746 hashsize = 1 << (fls(size) - 1);
751 static volatile int next_tcp_stack_id = 1;
754 * Register a TCP function block with the name provided in the names
755 * array. (Note that this function does NOT automatically register
756 * blk->tfb_tcp_block_name as a stack name. Therefore, you should
757 * explicitly include blk->tfb_tcp_block_name in the list of names if
758 * you wish to register the stack with that name.)
760 * Either all name registrations will succeed or all will fail. If
761 * a name registration fails, the function will update the num_names
762 * argument to point to the array index of the name that encountered
765 * Returns 0 on success, or an error code on failure.
768 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
769 const char *names[], int *num_names)
771 struct tcp_function *n;
772 struct tcp_function_set fs;
775 KASSERT(names != NULL && *num_names > 0,
776 ("%s: Called with 0-length name list", __func__));
777 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
779 if (t_functions_inited == 0) {
780 init_tcp_functions();
782 if ((blk->tfb_tcp_output == NULL) ||
783 (blk->tfb_tcp_do_segment == NULL) ||
784 (blk->tfb_tcp_ctloutput == NULL) ||
785 (strlen(blk->tfb_tcp_block_name) == 0)) {
787 * These functions are required and you
793 if (blk->tfb_tcp_timer_stop_all ||
794 blk->tfb_tcp_timer_activate ||
795 blk->tfb_tcp_timer_active ||
796 blk->tfb_tcp_timer_stop) {
798 * If you define one timer function you
799 * must have them all.
801 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
802 (blk->tfb_tcp_timer_activate == NULL) ||
803 (blk->tfb_tcp_timer_active == NULL) ||
804 (blk->tfb_tcp_timer_stop == NULL)) {
810 refcount_init(&blk->tfb_refcnt, 0);
812 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
813 for (i = 0; i < *num_names; i++) {
814 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
821 (void)strncpy(fs.function_set_name, names[i],
822 TCP_FUNCTION_NAME_LEN_MAX);
823 fs.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
824 rw_wlock(&tcp_function_lock);
825 if (find_tcp_functions_locked(&fs) != NULL) {
826 /* Duplicate name space not allowed */
827 rw_wunlock(&tcp_function_lock);
828 free(n, M_TCPFUNCTIONS);
832 (void)strncpy(n->tf_name, names[i], TCP_FUNCTION_NAME_LEN_MAX);
833 n->tf_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
834 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
836 rw_wunlock(&tcp_function_lock);
842 * Deregister the names we just added. Because registration failed
843 * for names[i], we don't need to deregister that name.
846 rw_wlock(&tcp_function_lock);
848 TAILQ_FOREACH(n, &t_functions, tf_next) {
849 if (!strncmp(n->tf_name, names[i],
850 TCP_FUNCTION_NAME_LEN_MAX)) {
851 TAILQ_REMOVE(&t_functions, n, tf_next);
854 free(n, M_TCPFUNCTIONS);
859 rw_wunlock(&tcp_function_lock);
864 * Register a TCP function block using the name provided in the name
867 * Returns 0 on success, or an error code on failure.
870 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
873 const char *name_list[1];
880 name_list[0] = blk->tfb_tcp_block_name;
881 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
886 * Register a TCP function block using the name defined in
887 * blk->tfb_tcp_block_name.
889 * Returns 0 on success, or an error code on failure.
892 register_tcp_functions(struct tcp_function_block *blk, int wait)
895 return (register_tcp_functions_as_name(blk, NULL, wait));
899 * Deregister all names associated with a function block. This
900 * functionally removes the function block from use within the system.
902 * When called with a true quiesce argument, mark the function block
903 * as being removed so no more stacks will use it and determine
904 * whether the removal would succeed.
906 * When called with a false quiesce argument, actually attempt the
909 * When called with a force argument, attempt to switch all TCBs to
910 * use the default stack instead of returning EBUSY.
912 * Returns 0 on success (or if the removal would succeed, or an error
916 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
919 struct tcp_function *f;
921 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
922 /* You can't un-register the default */
925 rw_wlock(&tcp_function_lock);
926 if (blk == tcp_func_set_ptr) {
927 /* You can't free the current default */
928 rw_wunlock(&tcp_function_lock);
931 /* Mark the block so no more stacks can use it. */
932 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
934 * If TCBs are still attached to the stack, attempt to switch them
935 * to the default stack.
937 if (force && blk->tfb_refcnt) {
940 VNET_ITERATOR_DECL(vnet_iter);
942 rw_wunlock(&tcp_function_lock);
945 VNET_FOREACH(vnet_iter) {
946 CURVNET_SET(vnet_iter);
947 INP_INFO_WLOCK(&V_tcbinfo);
948 LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
950 if (inp->inp_flags & INP_TIMEWAIT) {
955 if (tp == NULL || tp->t_fb != blk) {
959 tcp_switch_back_to_default(tp);
962 INP_INFO_WUNLOCK(&V_tcbinfo);
967 rw_wlock(&tcp_function_lock);
969 if (blk->tfb_refcnt) {
970 /* TCBs still attached. */
971 rw_wunlock(&tcp_function_lock);
976 rw_wunlock(&tcp_function_lock);
979 /* Remove any function names that map to this function block. */
980 while (find_tcp_fb_locked(blk, &f) != NULL) {
981 TAILQ_REMOVE(&t_functions, f, tf_next);
984 free(f, M_TCPFUNCTIONS);
986 rw_wunlock(&tcp_function_lock);
993 const char *tcbhash_tuneable;
996 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
999 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1000 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1001 printf("%s: WARNING: unable to register helper hook\n", __func__);
1002 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1003 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1004 printf("%s: WARNING: unable to register helper hook\n", __func__);
1006 hashsize = TCBHASHSIZE;
1007 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1008 if (hashsize == 0) {
1010 * Auto tune the hash size based on maxsockets.
1011 * A perfect hash would have a 1:1 mapping
1012 * (hashsize = maxsockets) however it's been
1013 * suggested that O(2) average is better.
1015 hashsize = maketcp_hashsize(maxsockets / 4);
1017 * Our historical default is 512,
1018 * do not autotune lower than this.
1022 if (bootverbose && IS_DEFAULT_VNET(curvnet))
1023 printf("%s: %s auto tuned to %d\n", __func__,
1024 tcbhash_tuneable, hashsize);
1027 * We require a hashsize to be a power of two.
1028 * Previously if it was not a power of two we would just reset it
1029 * back to 512, which could be a nasty surprise if you did not notice
1030 * the error message.
1031 * Instead what we do is clip it to the closest power of two lower
1032 * than the specified hash value.
1034 if (!powerof2(hashsize)) {
1035 int oldhashsize = hashsize;
1037 hashsize = maketcp_hashsize(hashsize);
1038 /* prevent absurdly low value */
1041 printf("%s: WARNING: TCB hash size not a power of 2, "
1042 "clipped from %d to %d.\n", __func__, oldhashsize,
1045 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
1046 "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
1049 * These have to be type stable for the benefit of the timers.
1051 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1052 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1053 uma_zone_set_max(V_tcpcb_zone, maxsockets);
1054 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1060 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1061 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1062 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1064 tcp_fastopen_init();
1066 /* Skip initialization of globals for non-default instances. */
1067 if (!IS_DEFAULT_VNET(curvnet))
1070 tcp_reass_global_init();
1072 /* XXX virtualize those bellow? */
1073 tcp_delacktime = TCPTV_DELACK;
1074 tcp_keepinit = TCPTV_KEEP_INIT;
1075 tcp_keepidle = TCPTV_KEEP_IDLE;
1076 tcp_keepintvl = TCPTV_KEEPINTVL;
1077 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1078 tcp_msl = TCPTV_MSL;
1079 tcp_rexmit_min = TCPTV_MIN;
1080 if (tcp_rexmit_min < 1)
1082 tcp_persmin = TCPTV_PERSMIN;
1083 tcp_persmax = TCPTV_PERSMAX;
1084 tcp_rexmit_slop = TCPTV_CPU_VAR;
1085 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1086 tcp_tcbhashsize = hashsize;
1087 /* Setup the tcp function block list */
1088 init_tcp_functions();
1089 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1091 /* Initialize the TCP logging data. */
1095 if (tcp_soreceive_stream) {
1097 tcp_usrreqs.pru_soreceive = soreceive_stream;
1100 tcp6_usrreqs.pru_soreceive = soreceive_stream;
1105 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1107 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1109 if (max_protohdr < TCP_MINPROTOHDR)
1110 max_protohdr = TCP_MINPROTOHDR;
1111 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1113 #undef TCP_MINPROTOHDR
1116 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1117 SHUTDOWN_PRI_DEFAULT);
1118 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
1119 EVENTHANDLER_PRI_ANY);
1127 tcp_destroy(void *unused __unused)
1135 * All our processes are gone, all our sockets should be cleaned
1136 * up, which means, we should be past the tcp_discardcb() calls.
1137 * Sleep to let all tcpcb timers really disappear and cleanup.
1140 INP_LIST_RLOCK(&V_tcbinfo);
1141 n = V_tcbinfo.ipi_count;
1142 INP_LIST_RUNLOCK(&V_tcbinfo);
1145 pause("tcpdes", hz / 10);
1150 in_pcbinfo_destroy(&V_tcbinfo);
1151 /* tcp_discardcb() clears the sack_holes up. */
1152 uma_zdestroy(V_sack_hole_zone);
1153 uma_zdestroy(V_tcpcb_zone);
1156 * Cannot free the zone until all tcpcbs are released as we attach
1157 * the allocations to them.
1159 tcp_fastopen_destroy();
1162 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1164 printf("%s: WARNING: unable to deregister helper hook "
1165 "type=%d, id=%d: error %d returned\n", __func__,
1166 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1168 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1170 printf("%s: WARNING: unable to deregister helper hook "
1171 "type=%d, id=%d: error %d returned\n", __func__,
1172 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1176 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1186 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1187 * tcp_template used to store this data in mbufs, but we now recopy it out
1188 * of the tcpcb each time to conserve mbufs.
1191 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
1193 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1195 INP_WLOCK_ASSERT(inp);
1198 if ((inp->inp_vflag & INP_IPV6) != 0) {
1199 struct ip6_hdr *ip6;
1201 ip6 = (struct ip6_hdr *)ip_ptr;
1202 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1203 (inp->inp_flow & IPV6_FLOWINFO_MASK);
1204 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1205 (IPV6_VERSION & IPV6_VERSION_MASK);
1206 ip6->ip6_nxt = IPPROTO_TCP;
1207 ip6->ip6_plen = htons(sizeof(struct tcphdr));
1208 ip6->ip6_src = inp->in6p_laddr;
1209 ip6->ip6_dst = inp->in6p_faddr;
1212 #if defined(INET6) && defined(INET)
1219 ip = (struct ip *)ip_ptr;
1220 ip->ip_v = IPVERSION;
1222 ip->ip_tos = inp->inp_ip_tos;
1226 ip->ip_ttl = inp->inp_ip_ttl;
1228 ip->ip_p = IPPROTO_TCP;
1229 ip->ip_src = inp->inp_laddr;
1230 ip->ip_dst = inp->inp_faddr;
1233 th->th_sport = inp->inp_lport;
1234 th->th_dport = inp->inp_fport;
1242 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
1246 * Create template to be used to send tcp packets on a connection.
1247 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
1248 * use for this function is in keepalives, which use tcp_respond.
1251 tcpip_maketemplate(struct inpcb *inp)
1255 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1258 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1263 * Send a single message to the TCP at address specified by
1264 * the given TCP/IP header. If m == NULL, then we make a copy
1265 * of the tcpiphdr at th and send directly to the addressed host.
1266 * This is used to force keep alive messages out using the TCP
1267 * template for a connection. If flags are given then we send
1268 * a message back to the TCP which originated the segment th,
1269 * and discard the mbuf containing it and any other attached mbufs.
1271 * In any case the ack and sequence number of the transmitted
1272 * segment are as specified by the parameters.
1274 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1277 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1278 tcp_seq ack, tcp_seq seq, int flags)
1287 struct ip6_hdr *ip6;
1290 int optlen, tlen, win;
1293 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1296 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1303 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1304 INP_WLOCK_ASSERT(inp);
1311 if (!(flags & TH_RST)) {
1312 win = sbspace(&inp->inp_socket->so_rcv);
1313 if (win > TCP_MAXWIN << tp->rcv_scale)
1314 win = TCP_MAXWIN << tp->rcv_scale;
1316 if ((tp->t_flags & TF_NOOPT) == 0)
1320 m = m_gethdr(M_NOWAIT, MT_DATA);
1323 m->m_data += max_linkhdr;
1326 bcopy((caddr_t)ip6, mtod(m, caddr_t),
1327 sizeof(struct ip6_hdr));
1328 ip6 = mtod(m, struct ip6_hdr *);
1329 nth = (struct tcphdr *)(ip6 + 1);
1333 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1334 ip = mtod(m, struct ip *);
1335 nth = (struct tcphdr *)(ip + 1);
1337 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1339 } else if (!M_WRITABLE(m)) {
1342 /* Can't reuse 'm', allocate a new mbuf. */
1343 n = m_gethdr(M_NOWAIT, MT_DATA);
1349 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1355 n->m_data += max_linkhdr;
1356 /* m_len is set later */
1357 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1360 bcopy((caddr_t)ip6, mtod(n, caddr_t),
1361 sizeof(struct ip6_hdr));
1362 ip6 = mtod(n, struct ip6_hdr *);
1363 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1364 nth = (struct tcphdr *)(ip6 + 1);
1368 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1369 ip = mtod(n, struct ip *);
1370 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1371 nth = (struct tcphdr *)(ip + 1);
1373 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1374 xchg(nth->th_dport, nth->th_sport, uint16_t);
1381 * XXX MRT We inherit the FIB, which is lucky.
1385 m->m_data = (caddr_t)ipgen;
1386 /* m_len is set later */
1389 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1390 nth = (struct tcphdr *)(ip6 + 1);
1394 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1395 nth = (struct tcphdr *)(ip + 1);
1399 * this is usually a case when an extension header
1400 * exists between the IPv6 header and the
1403 nth->th_sport = th->th_sport;
1404 nth->th_dport = th->th_dport;
1406 xchg(nth->th_dport, nth->th_sport, uint16_t);
1412 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1414 #if defined(INET) && defined(INET6)
1418 tlen = sizeof (struct tcpiphdr);
1422 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1423 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1424 m, tlen, (long)M_TRAILINGSPACE(m)));
1429 /* Make sure we have room. */
1430 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1431 m->m_next = m_get(M_NOWAIT, MT_DATA);
1433 optp = mtod(m->m_next, u_char *);
1438 optp = (u_char *) (nth + 1);
1444 if (tp->t_flags & TF_RCVD_TSTMP) {
1445 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1446 to.to_tsecr = tp->ts_recent;
1447 to.to_flags |= TOF_TS;
1449 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1450 /* TCP-MD5 (RFC2385). */
1451 if (tp->t_flags & TF_SIGNATURE)
1452 to.to_flags |= TOF_SIGNATURE;
1454 /* Add the options. */
1455 tlen += optlen = tcp_addoptions(&to, optp);
1457 /* Update m_len in the correct mbuf. */
1458 optm->m_len += optlen;
1464 ip6->ip6_vfc = IPV6_VERSION;
1465 ip6->ip6_nxt = IPPROTO_TCP;
1466 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1469 #if defined(INET) && defined(INET6)
1474 ip->ip_len = htons(tlen);
1475 ip->ip_ttl = V_ip_defttl;
1476 if (V_path_mtu_discovery)
1477 ip->ip_off |= htons(IP_DF);
1480 m->m_pkthdr.len = tlen;
1481 m->m_pkthdr.rcvif = NULL;
1485 * Packet is associated with a socket, so allow the
1486 * label of the response to reflect the socket label.
1488 INP_WLOCK_ASSERT(inp);
1489 mac_inpcb_create_mbuf(inp, m);
1492 * Packet is not associated with a socket, so possibly
1493 * update the label in place.
1495 mac_netinet_tcp_reply(m);
1498 nth->th_seq = htonl(seq);
1499 nth->th_ack = htonl(ack);
1501 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1502 nth->th_flags = flags;
1504 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1506 nth->th_win = htons((u_short)win);
1509 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1510 if (to.to_flags & TOF_SIGNATURE) {
1511 if (!TCPMD5_ENABLED() ||
1512 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1519 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1522 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1523 nth->th_sum = in6_cksum_pseudo(ip6,
1524 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1525 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1529 #if defined(INET6) && defined(INET)
1534 m->m_pkthdr.csum_flags = CSUM_TCP;
1535 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1536 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1540 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1541 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1543 TCP_PROBE3(debug__output, tp, th, m);
1545 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1549 TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1550 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1553 #if defined(INET) && defined(INET6)
1558 TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1559 (void)ip_output(m, NULL, NULL, 0, NULL, inp);
1565 * Create a new TCP control block, making an
1566 * empty reassembly queue and hooking it to the argument
1567 * protocol control block. The `inp' parameter must have
1568 * come from the zone allocator set up in tcp_init().
1571 tcp_newtcpcb(struct inpcb *inp)
1573 struct tcpcb_mem *tm;
1576 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1579 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1584 /* Initialise cc_var struct for this tcpcb. */
1586 tp->ccv->type = IPPROTO_TCP;
1587 tp->ccv->ccvc.tcp = tp;
1588 rw_rlock(&tcp_function_lock);
1589 tp->t_fb = tcp_func_set_ptr;
1590 refcount_acquire(&tp->t_fb->tfb_refcnt);
1591 rw_runlock(&tcp_function_lock);
1593 * Use the current system default CC algorithm.
1596 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1597 CC_ALGO(tp) = CC_DEFAULT();
1600 if (CC_ALGO(tp)->cb_init != NULL)
1601 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1602 if (tp->t_fb->tfb_tcp_fb_fini)
1603 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1604 refcount_release(&tp->t_fb->tfb_refcnt);
1605 uma_zfree(V_tcpcb_zone, tm);
1611 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1612 if (tp->t_fb->tfb_tcp_fb_fini)
1613 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1614 refcount_release(&tp->t_fb->tfb_refcnt);
1615 uma_zfree(V_tcpcb_zone, tm);
1621 tp->t_vnet = inp->inp_vnet;
1623 tp->t_timers = &tm->tt;
1624 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
1627 isipv6 ? V_tcp_v6mssdflt :
1631 /* Set up our timeouts. */
1632 callout_init(&tp->t_timers->tt_rexmt, 1);
1633 callout_init(&tp->t_timers->tt_persist, 1);
1634 callout_init(&tp->t_timers->tt_keep, 1);
1635 callout_init(&tp->t_timers->tt_2msl, 1);
1636 callout_init(&tp->t_timers->tt_delack, 1);
1638 if (V_tcp_do_rfc1323)
1639 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1641 tp->t_flags |= TF_SACK_PERMIT;
1642 TAILQ_INIT(&tp->snd_holes);
1644 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1647 in_pcbref(inp); /* Reference for tcpcb */
1651 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1652 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1653 * reasonable initial retransmit time.
1655 tp->t_srtt = TCPTV_SRTTBASE;
1656 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1657 tp->t_rttmin = tcp_rexmit_min;
1658 tp->t_rxtcur = TCPTV_RTOBASE;
1659 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1660 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1661 tp->t_rcvtime = ticks;
1663 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1664 * because the socket may be bound to an IPv6 wildcard address,
1665 * which may match an IPv4-mapped IPv6 address.
1667 inp->inp_ip_ttl = V_ip_defttl;
1671 * Init the TCP PCAP queues.
1673 tcp_pcap_tcpcb_init(tp);
1676 /* Initialize the per-TCPCB log data. */
1677 tcp_log_tcpcbinit(tp);
1679 if (tp->t_fb->tfb_tcp_fb_init) {
1680 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1682 return (tp); /* XXX */
1686 * Switch the congestion control algorithm back to NewReno for any active
1687 * control blocks using an algorithm which is about to go away.
1688 * This ensures the CC framework can allow the unload to proceed without leaving
1689 * any dangling pointers which would trigger a panic.
1690 * Returning non-zero would inform the CC framework that something went wrong
1691 * and it would be unsafe to allow the unload to proceed. However, there is no
1692 * way for this to occur with this implementation so we always return zero.
1695 tcp_ccalgounload(struct cc_algo *unload_algo)
1697 struct cc_algo *tmpalgo;
1700 VNET_ITERATOR_DECL(vnet_iter);
1703 * Check all active control blocks across all network stacks and change
1704 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1705 * requires cleanup code to be run, call it.
1708 VNET_FOREACH(vnet_iter) {
1709 CURVNET_SET(vnet_iter);
1710 INP_INFO_WLOCK(&V_tcbinfo);
1712 * New connections already part way through being initialised
1713 * with the CC algo we're removing will not race with this code
1714 * because the INP_INFO_WLOCK is held during initialisation. We
1715 * therefore don't enter the loop below until the connection
1716 * list has stabilised.
1718 LIST_FOREACH(inp, &V_tcb, inp_list) {
1720 /* Important to skip tcptw structs. */
1721 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1722 (tp = intotcpcb(inp)) != NULL) {
1724 * By holding INP_WLOCK here, we are assured
1725 * that the connection is not currently
1726 * executing inside the CC module's functions
1727 * i.e. it is safe to make the switch back to
1730 if (CC_ALGO(tp) == unload_algo) {
1731 tmpalgo = CC_ALGO(tp);
1732 /* NewReno does not require any init. */
1733 CC_ALGO(tp) = &newreno_cc_algo;
1734 if (tmpalgo->cb_destroy != NULL)
1735 tmpalgo->cb_destroy(tp->ccv);
1740 INP_INFO_WUNLOCK(&V_tcbinfo);
1743 VNET_LIST_RUNLOCK();
1749 * Drop a TCP connection, reporting
1750 * the specified error. If connection is synchronized,
1751 * then send a RST to peer.
1754 tcp_drop(struct tcpcb *tp, int errno)
1756 struct socket *so = tp->t_inpcb->inp_socket;
1758 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1759 INP_WLOCK_ASSERT(tp->t_inpcb);
1761 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1762 tcp_state_change(tp, TCPS_CLOSED);
1763 (void) tp->t_fb->tfb_tcp_output(tp);
1764 TCPSTAT_INC(tcps_drops);
1766 TCPSTAT_INC(tcps_conndrops);
1767 if (errno == ETIMEDOUT && tp->t_softerror)
1768 errno = tp->t_softerror;
1769 so->so_error = errno;
1770 return (tcp_close(tp));
1774 tcp_discardcb(struct tcpcb *tp)
1776 struct inpcb *inp = tp->t_inpcb;
1777 struct socket *so = inp->inp_socket;
1779 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1783 INP_WLOCK_ASSERT(inp);
1786 * Make sure that all of our timers are stopped before we delete the
1789 * If stopping a timer fails, we schedule a discard function in same
1790 * callout, and the last discard function called will take care of
1791 * deleting the tcpcb.
1793 tp->t_timers->tt_draincnt = 0;
1794 tcp_timer_stop(tp, TT_REXMT);
1795 tcp_timer_stop(tp, TT_PERSIST);
1796 tcp_timer_stop(tp, TT_KEEP);
1797 tcp_timer_stop(tp, TT_2MSL);
1798 tcp_timer_stop(tp, TT_DELACK);
1799 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1801 * Call the stop-all function of the methods,
1802 * this function should call the tcp_timer_stop()
1803 * method with each of the function specific timeouts.
1804 * That stop will be called via the tfb_tcp_timer_stop()
1805 * which should use the async drain function of the
1806 * callout system (see tcp_var.h).
1808 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1812 * If we got enough samples through the srtt filter,
1813 * save the rtt and rttvar in the routing entry.
1814 * 'Enough' is arbitrarily defined as 4 rtt samples.
1815 * 4 samples is enough for the srtt filter to converge
1816 * to within enough % of the correct value; fewer samples
1817 * and we could save a bogus rtt. The danger is not high
1818 * as tcp quickly recovers from everything.
1819 * XXX: Works very well but needs some more statistics!
1821 if (tp->t_rttupdated >= 4) {
1822 struct hc_metrics_lite metrics;
1825 bzero(&metrics, sizeof(metrics));
1827 * Update the ssthresh always when the conditions below
1828 * are satisfied. This gives us better new start value
1829 * for the congestion avoidance for new connections.
1830 * ssthresh is only set if packet loss occurred on a session.
1832 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1833 * being torn down. Ideally this code would not use 'so'.
1835 ssthresh = tp->snd_ssthresh;
1836 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1838 * convert the limit from user data bytes to
1839 * packets then to packet data bytes.
1841 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1844 ssthresh *= (tp->t_maxseg +
1846 (isipv6 ? sizeof (struct ip6_hdr) +
1847 sizeof (struct tcphdr) :
1849 sizeof (struct tcpiphdr)
1856 metrics.rmx_ssthresh = ssthresh;
1858 metrics.rmx_rtt = tp->t_srtt;
1859 metrics.rmx_rttvar = tp->t_rttvar;
1860 metrics.rmx_cwnd = tp->snd_cwnd;
1861 metrics.rmx_sendpipe = 0;
1862 metrics.rmx_recvpipe = 0;
1864 tcp_hc_update(&inp->inp_inc, &metrics);
1867 /* free the reassembly queue, if any */
1868 tcp_reass_flush(tp);
1871 /* Disconnect offload device, if any. */
1872 if (tp->t_flags & TF_TOE)
1873 tcp_offload_detach(tp);
1876 tcp_free_sackholes(tp);
1879 /* Free the TCP PCAP queues. */
1880 tcp_pcap_drain(&(tp->t_inpkts));
1881 tcp_pcap_drain(&(tp->t_outpkts));
1884 /* Allow the CC algorithm to clean up after itself. */
1885 if (CC_ALGO(tp)->cb_destroy != NULL)
1886 CC_ALGO(tp)->cb_destroy(tp->ccv);
1889 khelp_destroy_osd(tp->osd);
1893 inp->inp_ppcb = NULL;
1894 if (tp->t_timers->tt_draincnt == 0) {
1895 /* We own the last reference on tcpcb, let's free it. */
1897 tcp_log_tcpcbfini(tp);
1899 TCPSTATES_DEC(tp->t_state);
1900 if (tp->t_fb->tfb_tcp_fb_fini)
1901 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1902 refcount_release(&tp->t_fb->tfb_refcnt);
1904 uma_zfree(V_tcpcb_zone, tp);
1905 released = in_pcbrele_wlocked(inp);
1906 KASSERT(!released, ("%s: inp %p should not have been released "
1907 "here", __func__, inp));
1912 tcp_timer_discard(void *ptp)
1917 tp = (struct tcpcb *)ptp;
1918 CURVNET_SET(tp->t_vnet);
1919 INP_INFO_RLOCK(&V_tcbinfo);
1921 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1924 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1925 ("%s: tcpcb has to be stopped here", __func__));
1926 tp->t_timers->tt_draincnt--;
1927 if (tp->t_timers->tt_draincnt == 0) {
1928 /* We own the last reference on this tcpcb, let's free it. */
1930 tcp_log_tcpcbfini(tp);
1932 TCPSTATES_DEC(tp->t_state);
1933 if (tp->t_fb->tfb_tcp_fb_fini)
1934 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1935 refcount_release(&tp->t_fb->tfb_refcnt);
1937 uma_zfree(V_tcpcb_zone, tp);
1938 if (in_pcbrele_wlocked(inp)) {
1939 INP_INFO_RUNLOCK(&V_tcbinfo);
1945 INP_INFO_RUNLOCK(&V_tcbinfo);
1950 * Attempt to close a TCP control block, marking it as dropped, and freeing
1951 * the socket if we hold the only reference.
1954 tcp_close(struct tcpcb *tp)
1956 struct inpcb *inp = tp->t_inpcb;
1959 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1960 INP_WLOCK_ASSERT(inp);
1963 if (tp->t_state == TCPS_LISTEN)
1964 tcp_offload_listen_stop(tp);
1967 * This releases the TFO pending counter resource for TFO listen
1968 * sockets as well as passively-created TFO sockets that transition
1969 * from SYN_RECEIVED to CLOSED.
1971 if (tp->t_tfo_pending) {
1972 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1973 tp->t_tfo_pending = NULL;
1976 TCPSTAT_INC(tcps_closed);
1977 if (tp->t_state != TCPS_CLOSED)
1978 tcp_state_change(tp, TCPS_CLOSED);
1979 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1980 so = inp->inp_socket;
1981 soisdisconnected(so);
1982 if (inp->inp_flags & INP_SOCKREF) {
1983 KASSERT(so->so_state & SS_PROTOREF,
1984 ("tcp_close: !SS_PROTOREF"));
1985 inp->inp_flags &= ~INP_SOCKREF;
1988 so->so_state &= ~SS_PROTOREF;
1998 VNET_ITERATOR_DECL(vnet_iter);
2003 VNET_LIST_RLOCK_NOSLEEP();
2004 VNET_FOREACH(vnet_iter) {
2005 CURVNET_SET(vnet_iter);
2010 * Walk the tcpbs, if existing, and flush the reassembly queue,
2011 * if there is one...
2012 * XXX: The "Net/3" implementation doesn't imply that the TCP
2013 * reassembly queue should be flushed, but in a situation
2014 * where we're really low on mbufs, this is potentially
2017 INP_INFO_WLOCK(&V_tcbinfo);
2018 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
2019 if (inpb->inp_flags & INP_TIMEWAIT)
2022 if ((tcpb = intotcpcb(inpb)) != NULL) {
2023 tcp_reass_flush(tcpb);
2024 tcp_clean_sackreport(tcpb);
2026 tcp_log_drain(tcpb);
2029 if (tcp_pcap_aggressive_free) {
2030 /* Free the TCP PCAP queues. */
2031 tcp_pcap_drain(&(tcpb->t_inpkts));
2032 tcp_pcap_drain(&(tcpb->t_outpkts));
2038 INP_INFO_WUNLOCK(&V_tcbinfo);
2041 VNET_LIST_RUNLOCK_NOSLEEP();
2045 * Notify a tcp user of an asynchronous error;
2046 * store error as soft error, but wake up user
2047 * (for now, won't do anything until can select for soft error).
2049 * Do not wake up user since there currently is no mechanism for
2050 * reporting soft errors (yet - a kqueue filter may be added).
2052 static struct inpcb *
2053 tcp_notify(struct inpcb *inp, int error)
2057 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2058 INP_WLOCK_ASSERT(inp);
2060 if ((inp->inp_flags & INP_TIMEWAIT) ||
2061 (inp->inp_flags & INP_DROPPED))
2064 tp = intotcpcb(inp);
2065 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2068 * Ignore some errors if we are hooked up.
2069 * If connection hasn't completed, has retransmitted several times,
2070 * and receives a second error, give up now. This is better
2071 * than waiting a long time to establish a connection that
2072 * can never complete.
2074 if (tp->t_state == TCPS_ESTABLISHED &&
2075 (error == EHOSTUNREACH || error == ENETUNREACH ||
2076 error == EHOSTDOWN)) {
2077 if (inp->inp_route.ro_rt) {
2078 RTFREE(inp->inp_route.ro_rt);
2079 inp->inp_route.ro_rt = (struct rtentry *)NULL;
2082 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2084 tp = tcp_drop(tp, error);
2090 tp->t_softerror = error;
2094 wakeup( &so->so_timeo);
2101 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2103 int error, i, m, n, pcb_count;
2104 struct inpcb *inp, **inp_list;
2109 * The process of preparing the TCB list is too time-consuming and
2110 * resource-intensive to repeat twice on every request.
2112 if (req->oldptr == NULL) {
2113 n = V_tcbinfo.ipi_count +
2114 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2115 n += imax(n / 8, 10);
2116 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2120 if (req->newptr != NULL)
2124 * OK, now we're committed to doing something.
2126 INP_LIST_RLOCK(&V_tcbinfo);
2127 gencnt = V_tcbinfo.ipi_gencnt;
2128 n = V_tcbinfo.ipi_count;
2129 INP_LIST_RUNLOCK(&V_tcbinfo);
2131 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2133 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
2134 + (n + m) * sizeof(struct xtcpcb));
2138 xig.xig_len = sizeof xig;
2139 xig.xig_count = n + m;
2140 xig.xig_gen = gencnt;
2141 xig.xig_sogen = so_gencnt;
2142 error = SYSCTL_OUT(req, &xig, sizeof xig);
2146 error = syncache_pcblist(req, m, &pcb_count);
2150 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
2152 INP_INFO_WLOCK(&V_tcbinfo);
2153 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
2154 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
2156 if (inp->inp_gencnt <= gencnt) {
2158 * XXX: This use of cr_cansee(), introduced with
2159 * TCP state changes, is not quite right, but for
2160 * now, better than nothing.
2162 if (inp->inp_flags & INP_TIMEWAIT) {
2163 if (intotw(inp) != NULL)
2164 error = cr_cansee(req->td->td_ucred,
2165 intotw(inp)->tw_cred);
2167 error = EINVAL; /* Skip this inp. */
2169 error = cr_canseeinpcb(req->td->td_ucred, inp);
2172 inp_list[i++] = inp;
2177 INP_INFO_WUNLOCK(&V_tcbinfo);
2181 for (i = 0; i < n; i++) {
2184 if (inp->inp_gencnt <= gencnt) {
2187 tcp_inptoxtp(inp, &xt);
2189 error = SYSCTL_OUT(req, &xt, sizeof xt);
2193 INP_INFO_RLOCK(&V_tcbinfo);
2194 for (i = 0; i < n; i++) {
2197 if (!in_pcbrele_rlocked(inp))
2200 INP_INFO_RUNLOCK(&V_tcbinfo);
2204 * Give the user an updated idea of our state.
2205 * If the generation differs from what we told
2206 * her before, she knows that something happened
2207 * while we were processing this request, and it
2208 * might be necessary to retry.
2210 INP_LIST_RLOCK(&V_tcbinfo);
2211 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2212 xig.xig_sogen = so_gencnt;
2213 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
2214 INP_LIST_RUNLOCK(&V_tcbinfo);
2215 error = SYSCTL_OUT(req, &xig, sizeof xig);
2217 free(inp_list, M_TEMP);
2221 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2222 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
2223 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
2227 tcp_getcred(SYSCTL_HANDLER_ARGS)
2230 struct sockaddr_in addrs[2];
2234 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2237 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2240 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2241 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2243 if (inp->inp_socket == NULL)
2246 error = cr_canseeinpcb(req->td->td_ucred, inp);
2248 cru2x(inp->inp_cred, &xuc);
2253 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2257 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2258 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2259 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
2264 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2267 struct sockaddr_in6 addrs[2];
2274 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2277 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2280 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2281 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2284 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2286 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2295 inp = in_pcblookup(&V_tcbinfo,
2296 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2298 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2299 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2302 inp = in6_pcblookup(&V_tcbinfo,
2303 &addrs[1].sin6_addr, addrs[1].sin6_port,
2304 &addrs[0].sin6_addr, addrs[0].sin6_port,
2305 INPLOOKUP_RLOCKPCB, NULL);
2307 if (inp->inp_socket == NULL)
2310 error = cr_canseeinpcb(req->td->td_ucred, inp);
2312 cru2x(inp->inp_cred, &xuc);
2317 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2321 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2322 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2323 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
2329 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2331 struct ip *ip = vip;
2333 struct in_addr faddr;
2336 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2338 struct in_conninfo inc;
2339 tcp_seq icmp_tcp_seq;
2342 faddr = ((struct sockaddr_in *)sa)->sin_addr;
2343 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2346 if (cmd == PRC_MSGSIZE)
2347 notify = tcp_mtudisc_notify;
2348 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2349 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2350 cmd == PRC_TIMXCEED_INTRANS) && ip)
2351 notify = tcp_drop_syn_sent;
2354 * Hostdead is ugly because it goes linearly through all PCBs.
2355 * XXX: We never get this from ICMP, otherwise it makes an
2356 * excellent DoS attack on machines with many connections.
2358 else if (cmd == PRC_HOSTDEAD)
2360 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2364 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2368 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2369 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2370 INP_INFO_RLOCK(&V_tcbinfo);
2371 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2372 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2373 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2374 /* signal EHOSTDOWN, as it flushes the cached route */
2375 inp = (*notify)(inp, EHOSTDOWN);
2378 icmp_tcp_seq = th->th_seq;
2380 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2381 !(inp->inp_flags & INP_DROPPED) &&
2382 !(inp->inp_socket == NULL)) {
2383 tp = intotcpcb(inp);
2384 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2385 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2386 if (cmd == PRC_MSGSIZE) {
2389 * If we got a needfrag set the MTU
2390 * in the route to the suggested new
2391 * value (if given) and then notify.
2393 mtu = ntohs(icp->icmp_nextmtu);
2395 * If no alternative MTU was
2396 * proposed, try the next smaller
2401 ntohs(ip->ip_len), 1);
2402 if (mtu < V_tcp_minmss +
2403 sizeof(struct tcpiphdr))
2404 mtu = V_tcp_minmss +
2405 sizeof(struct tcpiphdr);
2407 * Only process the offered MTU if it
2408 * is smaller than the current one.
2410 if (mtu < tp->t_maxseg +
2411 sizeof(struct tcpiphdr)) {
2412 bzero(&inc, sizeof(inc));
2413 inc.inc_faddr = faddr;
2415 inp->inp_inc.inc_fibnum;
2416 tcp_hc_updatemtu(&inc, mtu);
2417 tcp_mtudisc(inp, mtu);
2420 inp = (*notify)(inp,
2421 inetctlerrmap[cmd]);
2425 bzero(&inc, sizeof(inc));
2426 inc.inc_fport = th->th_dport;
2427 inc.inc_lport = th->th_sport;
2428 inc.inc_faddr = faddr;
2429 inc.inc_laddr = ip->ip_src;
2430 syncache_unreach(&inc, icmp_tcp_seq);
2435 INP_INFO_RUNLOCK(&V_tcbinfo);
2441 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2443 struct in6_addr *dst;
2444 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2445 struct ip6_hdr *ip6;
2449 struct icmp6_hdr *icmp6;
2450 struct ip6ctlparam *ip6cp = NULL;
2451 const struct sockaddr_in6 *sa6_src = NULL;
2452 struct in_conninfo inc;
2457 tcp_seq icmp_tcp_seq;
2461 if (sa->sa_family != AF_INET6 ||
2462 sa->sa_len != sizeof(struct sockaddr_in6))
2465 /* if the parameter is from icmp6, decode it. */
2467 ip6cp = (struct ip6ctlparam *)d;
2468 icmp6 = ip6cp->ip6c_icmp6;
2470 ip6 = ip6cp->ip6c_ip6;
2471 off = ip6cp->ip6c_off;
2472 sa6_src = ip6cp->ip6c_src;
2473 dst = ip6cp->ip6c_finaldst;
2477 off = 0; /* fool gcc */
2482 if (cmd == PRC_MSGSIZE)
2483 notify = tcp_mtudisc_notify;
2484 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2485 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2486 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2487 notify = tcp_drop_syn_sent;
2490 * Hostdead is ugly because it goes linearly through all PCBs.
2491 * XXX: We never get this from ICMP, otherwise it makes an
2492 * excellent DoS attack on machines with many connections.
2494 else if (cmd == PRC_HOSTDEAD)
2496 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2500 in6_pcbnotify(&V_tcbinfo, sa, 0,
2501 (const struct sockaddr *)sa6_src,
2502 0, cmd, NULL, notify);
2506 /* Check if we can safely get the ports from the tcp hdr */
2509 (int32_t) (off + sizeof(struct tcp_ports)))) {
2512 bzero(&t_ports, sizeof(struct tcp_ports));
2513 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2514 INP_INFO_RLOCK(&V_tcbinfo);
2515 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2516 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2517 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2518 /* signal EHOSTDOWN, as it flushes the cached route */
2519 inp = (*notify)(inp, EHOSTDOWN);
2522 off += sizeof(struct tcp_ports);
2523 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2526 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2528 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2529 !(inp->inp_flags & INP_DROPPED) &&
2530 !(inp->inp_socket == NULL)) {
2531 tp = intotcpcb(inp);
2532 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2533 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2534 if (cmd == PRC_MSGSIZE) {
2537 * If we got a needfrag set the MTU
2538 * in the route to the suggested new
2539 * value (if given) and then notify.
2541 mtu = ntohl(icmp6->icmp6_mtu);
2543 * If no alternative MTU was
2544 * proposed, or the proposed
2545 * MTU was too small, set to
2548 if (mtu < IPV6_MMTU)
2549 mtu = IPV6_MMTU - 8;
2550 bzero(&inc, sizeof(inc));
2551 inc.inc_fibnum = M_GETFIB(m);
2552 inc.inc_flags |= INC_ISIPV6;
2553 inc.inc6_faddr = *dst;
2554 if (in6_setscope(&inc.inc6_faddr,
2555 m->m_pkthdr.rcvif, NULL))
2558 * Only process the offered MTU if it
2559 * is smaller than the current one.
2561 if (mtu < tp->t_maxseg +
2562 sizeof (struct tcphdr) +
2563 sizeof (struct ip6_hdr)) {
2564 tcp_hc_updatemtu(&inc, mtu);
2565 tcp_mtudisc(inp, mtu);
2566 ICMP6STAT_INC(icp6s_pmtuchg);
2569 inp = (*notify)(inp,
2570 inet6ctlerrmap[cmd]);
2574 bzero(&inc, sizeof(inc));
2575 inc.inc_fibnum = M_GETFIB(m);
2576 inc.inc_flags |= INC_ISIPV6;
2577 inc.inc_fport = t_ports.th_dport;
2578 inc.inc_lport = t_ports.th_sport;
2579 inc.inc6_faddr = *dst;
2580 inc.inc6_laddr = ip6->ip6_src;
2581 syncache_unreach(&inc, icmp_tcp_seq);
2586 INP_INFO_RUNLOCK(&V_tcbinfo);
2592 * Following is where TCP initial sequence number generation occurs.
2594 * There are two places where we must use initial sequence numbers:
2595 * 1. In SYN-ACK packets.
2596 * 2. In SYN packets.
2598 * All ISNs for SYN-ACK packets are generated by the syncache. See
2599 * tcp_syncache.c for details.
2601 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2602 * depends on this property. In addition, these ISNs should be
2603 * unguessable so as to prevent connection hijacking. To satisfy
2604 * the requirements of this situation, the algorithm outlined in
2605 * RFC 1948 is used, with only small modifications.
2607 * Implementation details:
2609 * Time is based off the system timer, and is corrected so that it
2610 * increases by one megabyte per second. This allows for proper
2611 * recycling on high speed LANs while still leaving over an hour
2614 * As reading the *exact* system time is too expensive to be done
2615 * whenever setting up a TCP connection, we increment the time
2616 * offset in two ways. First, a small random positive increment
2617 * is added to isn_offset for each connection that is set up.
2618 * Second, the function tcp_isn_tick fires once per clock tick
2619 * and increments isn_offset as necessary so that sequence numbers
2620 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2621 * random positive increments serve only to ensure that the same
2622 * exact sequence number is never sent out twice (as could otherwise
2623 * happen when a port is recycled in less than the system tick
2626 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2627 * between seeding of isn_secret. This is normally set to zero,
2628 * as reseeding should not be necessary.
2630 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2631 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
2632 * general, this means holding an exclusive (write) lock.
2635 #define ISN_BYTES_PER_SECOND 1048576
2636 #define ISN_STATIC_INCREMENT 4096
2637 #define ISN_RANDOM_INCREMENT (4096 - 1)
2639 static VNET_DEFINE(u_char, isn_secret[32]);
2640 static VNET_DEFINE(int, isn_last);
2641 static VNET_DEFINE(int, isn_last_reseed);
2642 static VNET_DEFINE(u_int32_t, isn_offset);
2643 static VNET_DEFINE(u_int32_t, isn_offset_old);
2645 #define V_isn_secret VNET(isn_secret)
2646 #define V_isn_last VNET(isn_last)
2647 #define V_isn_last_reseed VNET(isn_last_reseed)
2648 #define V_isn_offset VNET(isn_offset)
2649 #define V_isn_offset_old VNET(isn_offset_old)
2652 tcp_new_isn(struct tcpcb *tp)
2655 u_int32_t md5_buffer[4];
2657 u_int32_t projected_offset;
2659 INP_WLOCK_ASSERT(tp->t_inpcb);
2662 /* Seed if this is the first use, reseed if requested. */
2663 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2664 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2666 read_random(&V_isn_secret, sizeof(V_isn_secret));
2667 V_isn_last_reseed = ticks;
2670 /* Compute the md5 hash and return the ISN. */
2672 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2673 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2675 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2676 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2677 sizeof(struct in6_addr));
2678 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2679 sizeof(struct in6_addr));
2683 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2684 sizeof(struct in_addr));
2685 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2686 sizeof(struct in_addr));
2688 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2689 MD5Final((u_char *) &md5_buffer, &isn_ctx);
2690 new_isn = (tcp_seq) md5_buffer[0];
2691 V_isn_offset += ISN_STATIC_INCREMENT +
2692 (arc4random() & ISN_RANDOM_INCREMENT);
2693 if (ticks != V_isn_last) {
2694 projected_offset = V_isn_offset_old +
2695 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2696 if (SEQ_GT(projected_offset, V_isn_offset))
2697 V_isn_offset = projected_offset;
2698 V_isn_offset_old = V_isn_offset;
2701 new_isn += V_isn_offset;
2707 * When a specific ICMP unreachable message is received and the
2708 * connection state is SYN-SENT, drop the connection. This behavior
2709 * is controlled by the icmp_may_rst sysctl.
2712 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2716 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2717 INP_WLOCK_ASSERT(inp);
2719 if ((inp->inp_flags & INP_TIMEWAIT) ||
2720 (inp->inp_flags & INP_DROPPED))
2723 tp = intotcpcb(inp);
2724 if (tp->t_state != TCPS_SYN_SENT)
2727 if (IS_FASTOPEN(tp->t_flags))
2728 tcp_fastopen_disable_path(tp);
2730 tp = tcp_drop(tp, errno);
2738 * When `need fragmentation' ICMP is received, update our idea of the MSS
2739 * based on the new value. Also nudge TCP to send something, since we
2740 * know the packet we just sent was dropped.
2741 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2743 static struct inpcb *
2744 tcp_mtudisc_notify(struct inpcb *inp, int error)
2747 tcp_mtudisc(inp, -1);
2752 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2757 INP_WLOCK_ASSERT(inp);
2758 if ((inp->inp_flags & INP_TIMEWAIT) ||
2759 (inp->inp_flags & INP_DROPPED))
2762 tp = intotcpcb(inp);
2763 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2765 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2767 so = inp->inp_socket;
2768 SOCKBUF_LOCK(&so->so_snd);
2769 /* If the mss is larger than the socket buffer, decrease the mss. */
2770 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2771 tp->t_maxseg = so->so_snd.sb_hiwat;
2772 SOCKBUF_UNLOCK(&so->so_snd);
2774 TCPSTAT_INC(tcps_mturesent);
2776 tp->snd_nxt = tp->snd_una;
2777 tcp_free_sackholes(tp);
2778 tp->snd_recover = tp->snd_max;
2779 if (tp->t_flags & TF_SACK_PERMIT)
2780 EXIT_FASTRECOVERY(tp->t_flags);
2781 tp->t_fb->tfb_tcp_output(tp);
2786 * Look-up the routing entry to the peer of this inpcb. If no route
2787 * is found and it cannot be allocated, then return 0. This routine
2788 * is called by TCP routines that access the rmx structure and by
2789 * tcp_mss_update to get the peer/interface MTU.
2792 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2794 struct nhop4_extended nh4;
2796 uint32_t maxmtu = 0;
2798 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2800 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2802 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2803 NHR_REF, 0, &nh4) != 0)
2807 maxmtu = nh4.nh_mtu;
2809 /* Report additional interface capabilities. */
2811 if (ifp->if_capenable & IFCAP_TSO4 &&
2812 ifp->if_hwassist & CSUM_TSO) {
2813 cap->ifcap |= CSUM_TSO;
2814 cap->tsomax = ifp->if_hw_tsomax;
2815 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2816 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2819 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2827 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2829 struct nhop6_extended nh6;
2830 struct in6_addr dst6;
2833 uint32_t maxmtu = 0;
2835 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2837 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2838 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2839 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2844 maxmtu = nh6.nh_mtu;
2846 /* Report additional interface capabilities. */
2848 if (ifp->if_capenable & IFCAP_TSO6 &&
2849 ifp->if_hwassist & CSUM_TSO) {
2850 cap->ifcap |= CSUM_TSO;
2851 cap->tsomax = ifp->if_hw_tsomax;
2852 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2853 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2856 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2864 * Calculate effective SMSS per RFC5681 definition for a given TCP
2865 * connection at its current state, taking into account SACK and etc.
2868 tcp_maxseg(const struct tcpcb *tp)
2872 if (tp->t_flags & TF_NOOPT)
2873 return (tp->t_maxseg);
2876 * Here we have a simplified code from tcp_addoptions(),
2877 * without a proper loop, and having most of paddings hardcoded.
2878 * We might make mistakes with padding here in some edge cases,
2879 * but this is harmless, since result of tcp_maxseg() is used
2880 * only in cwnd and ssthresh estimations.
2882 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2883 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2884 if (tp->t_flags & TF_RCVD_TSTMP)
2885 optlen = TCPOLEN_TSTAMP_APPA;
2888 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2889 if (tp->t_flags & TF_SIGNATURE)
2890 optlen += PAD(TCPOLEN_SIGNATURE);
2892 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2893 optlen += TCPOLEN_SACKHDR;
2894 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2895 optlen = PAD(optlen);
2898 if (tp->t_flags & TF_REQ_TSTMP)
2899 optlen = TCPOLEN_TSTAMP_APPA;
2901 optlen = PAD(TCPOLEN_MAXSEG);
2902 if (tp->t_flags & TF_REQ_SCALE)
2903 optlen += PAD(TCPOLEN_WINDOW);
2904 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2905 if (tp->t_flags & TF_SIGNATURE)
2906 optlen += PAD(TCPOLEN_SIGNATURE);
2908 if (tp->t_flags & TF_SACK_PERMIT)
2909 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2912 optlen = min(optlen, TCP_MAXOLEN);
2913 return (tp->t_maxseg - optlen);
2917 sysctl_drop(SYSCTL_HANDLER_ARGS)
2919 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2920 struct sockaddr_storage addrs[2];
2924 struct sockaddr_in *fin, *lin;
2926 struct sockaddr_in6 *fin6, *lin6;
2937 if (req->oldptr != NULL || req->oldlen != 0)
2939 if (req->newptr == NULL)
2941 if (req->newlen < sizeof(addrs))
2943 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2947 switch (addrs[0].ss_family) {
2950 fin6 = (struct sockaddr_in6 *)&addrs[0];
2951 lin6 = (struct sockaddr_in6 *)&addrs[1];
2952 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2953 lin6->sin6_len != sizeof(struct sockaddr_in6))
2955 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2956 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2958 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2959 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2960 fin = (struct sockaddr_in *)&addrs[0];
2961 lin = (struct sockaddr_in *)&addrs[1];
2964 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2967 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2974 fin = (struct sockaddr_in *)&addrs[0];
2975 lin = (struct sockaddr_in *)&addrs[1];
2976 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2977 lin->sin_len != sizeof(struct sockaddr_in))
2984 INP_INFO_RLOCK(&V_tcbinfo);
2985 switch (addrs[0].ss_family) {
2988 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2989 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2990 INPLOOKUP_WLOCKPCB, NULL);
2995 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2996 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3001 if (inp->inp_flags & INP_TIMEWAIT) {
3003 * XXXRW: There currently exists a state where an
3004 * inpcb is present, but its timewait state has been
3005 * discarded. For now, don't allow dropping of this
3013 } else if (!(inp->inp_flags & INP_DROPPED) &&
3014 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
3015 tp = intotcpcb(inp);
3016 tp = tcp_drop(tp, ECONNABORTED);
3023 INP_INFO_RUNLOCK(&V_tcbinfo);
3027 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3028 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
3029 0, sysctl_drop, "", "Drop TCP connection");
3032 * Generate a standardized TCP log line for use throughout the
3033 * tcp subsystem. Memory allocation is done with M_NOWAIT to
3034 * allow use in the interrupt context.
3036 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3037 * NB: The function may return NULL if memory allocation failed.
3039 * Due to header inclusion and ordering limitations the struct ip
3040 * and ip6_hdr pointers have to be passed as void pointers.
3043 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3047 /* Is logging enabled? */
3048 if (tcp_log_in_vain == 0)
3051 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3055 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3059 /* Is logging enabled? */
3060 if (tcp_log_debug == 0)
3063 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3067 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3074 const struct ip6_hdr *ip6;
3076 ip6 = (const struct ip6_hdr *)ip6hdr;
3078 ip = (struct ip *)ip4hdr;
3081 * The log line looks like this:
3082 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3084 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3085 sizeof(PRINT_TH_FLAGS) + 1 +
3087 2 * INET6_ADDRSTRLEN;
3089 2 * INET_ADDRSTRLEN;
3092 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3096 strcat(s, "TCP: [");
3099 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3100 inet_ntoa_r(inc->inc_faddr, sp);
3102 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3104 inet_ntoa_r(inc->inc_laddr, sp);
3106 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3109 ip6_sprintf(sp, &inc->inc6_faddr);
3111 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3113 ip6_sprintf(sp, &inc->inc6_laddr);
3115 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3116 } else if (ip6 && th) {
3117 ip6_sprintf(sp, &ip6->ip6_src);
3119 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3121 ip6_sprintf(sp, &ip6->ip6_dst);
3123 sprintf(sp, "]:%i", ntohs(th->th_dport));
3126 } else if (ip && th) {
3127 inet_ntoa_r(ip->ip_src, sp);
3129 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3131 inet_ntoa_r(ip->ip_dst, sp);
3133 sprintf(sp, "]:%i", ntohs(th->th_dport));
3141 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3142 if (*(s + size - 1) != '\0')
3143 panic("%s: string too long", __func__);
3148 * A subroutine which makes it easy to track TCP state changes with DTrace.
3149 * This function shouldn't be called for t_state initializations that don't
3150 * correspond to actual TCP state transitions.
3153 tcp_state_change(struct tcpcb *tp, int newstate)
3155 #if defined(KDTRACE_HOOKS)
3156 int pstate = tp->t_state;
3159 TCPSTATES_DEC(tp->t_state);
3160 TCPSTATES_INC(newstate);
3161 tp->t_state = newstate;
3162 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3166 * Create an external-format (``xtcpcb'') structure using the information in
3167 * the kernel-format tcpcb structure pointed to by tp. This is done to
3168 * reduce the spew of irrelevant information over this interface, to isolate
3169 * user code from changes in the kernel structure, and potentially to provide
3170 * information-hiding if we decide that some of this information should be
3171 * hidden from users.
3174 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3176 struct tcpcb *tp = intotcpcb(inp);
3179 if (inp->inp_flags & INP_TIMEWAIT) {
3180 bzero(xt, sizeof(struct xtcpcb));
3181 xt->t_state = TCPS_TIME_WAIT;
3183 xt->t_state = tp->t_state;
3184 xt->t_logstate = tp->t_logstate;
3185 xt->t_flags = tp->t_flags;
3186 xt->t_sndzerowin = tp->t_sndzerowin;
3187 xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3188 xt->t_rcvoopack = tp->t_rcvoopack;
3190 now = getsbinuptime();
3191 #define COPYTIMER(ttt) do { \
3192 if (callout_active(&tp->t_timers->ttt)) \
3193 xt->ttt = (tp->t_timers->ttt.c_time - now) / \
3198 COPYTIMER(tt_delack);
3199 COPYTIMER(tt_rexmt);
3200 COPYTIMER(tt_persist);
3204 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3206 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3207 TCP_FUNCTION_NAME_LEN_MAX);
3208 bzero(xt->xt_logid, TCP_LOG_ID_LEN);
3210 (void)tcp_log_get_id(tp, xt->xt_logid);
3214 xt->xt_len = sizeof(struct xtcpcb);
3215 in_pcbtoxinpcb(inp, &xt->xt_inp);
3216 if (inp->inp_socket == NULL)
3217 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;