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/tcp_hpts.h>
103 #include <netinet/cc/cc.h>
105 #include <netinet6/tcp6_var.h>
107 #include <netinet/tcpip.h>
108 #include <netinet/tcp_fastopen.h>
110 #include <netinet/tcp_pcap.h>
113 #include <netinet/tcp_debug.h>
116 #include <netinet6/ip6protosw.h>
119 #include <netinet/tcp_offload.h>
122 #include <netipsec/ipsec_support.h>
124 #include <machine/in_cksum.h>
127 #include <security/mac/mac_framework.h>
129 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
131 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
134 struct rwlock tcp_function_lock;
137 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
142 error = sysctl_handle_int(oidp, &new, 0, req);
143 if (error == 0 && req->newptr) {
144 if (new < TCP_MINMSS)
152 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
153 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
154 &sysctl_net_inet_tcp_mss_check, "I",
155 "Default TCP Maximum Segment Size");
159 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
163 new = V_tcp_v6mssdflt;
164 error = sysctl_handle_int(oidp, &new, 0, req);
165 if (error == 0 && req->newptr) {
166 if (new < TCP_MINMSS)
169 V_tcp_v6mssdflt = new;
174 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
175 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
176 &sysctl_net_inet_tcp_mss_v6_check, "I",
177 "Default TCP Maximum Segment Size for IPv6");
181 * Minimum MSS we accept and use. This prevents DoS attacks where
182 * we are forced to a ridiculous low MSS like 20 and send hundreds
183 * of packets instead of one. The effect scales with the available
184 * bandwidth and quickly saturates the CPU and network interface
185 * with packet generation and sending. Set to zero to disable MINMSS
186 * checking. This setting prevents us from sending too small packets.
188 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
190 &VNET_NAME(tcp_minmss), 0,
191 "Minimum TCP Maximum Segment Size");
193 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
194 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
195 &VNET_NAME(tcp_do_rfc1323), 0,
196 "Enable rfc1323 (high performance TCP) extensions");
198 static int tcp_log_debug = 0;
199 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
200 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
202 static int tcp_tcbhashsize;
203 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
204 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
206 static int do_tcpdrain = 1;
207 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
208 "Enable tcp_drain routine for extra help when low on mbufs");
210 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
211 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
213 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
214 #define V_icmp_may_rst VNET(icmp_may_rst)
215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
216 &VNET_NAME(icmp_may_rst), 0,
217 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
219 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
220 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
222 &VNET_NAME(tcp_isn_reseed_interval), 0,
223 "Seconds between reseeding of ISN secret");
225 static int tcp_soreceive_stream;
226 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
227 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
229 VNET_DEFINE(uma_zone_t, sack_hole_zone);
230 #define V_sack_hole_zone VNET(sack_hole_zone)
233 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
236 #define TS_OFFSET_SECRET_LENGTH 32
237 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
238 #define V_ts_offset_secret VNET(ts_offset_secret)
240 static int tcp_default_fb_init(struct tcpcb *tp);
241 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
242 static int tcp_default_handoff_ok(struct tcpcb *tp);
243 static struct inpcb *tcp_notify(struct inpcb *, int);
244 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
245 static void tcp_mtudisc(struct inpcb *, int);
246 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
247 void *ip4hdr, const void *ip6hdr);
250 static struct tcp_function_block tcp_def_funcblk = {
251 .tfb_tcp_block_name = "freebsd",
252 .tfb_tcp_output = tcp_output,
253 .tfb_tcp_do_segment = tcp_do_segment,
254 .tfb_tcp_ctloutput = tcp_default_ctloutput,
255 .tfb_tcp_handoff_ok = tcp_default_handoff_ok,
256 .tfb_tcp_fb_init = tcp_default_fb_init,
257 .tfb_tcp_fb_fini = tcp_default_fb_fini,
260 static int tcp_fb_cnt = 0;
261 struct tcp_funchead t_functions;
262 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
264 static struct tcp_function_block *
265 find_tcp_functions_locked(struct tcp_function_set *fs)
267 struct tcp_function *f;
268 struct tcp_function_block *blk=NULL;
270 TAILQ_FOREACH(f, &t_functions, tf_next) {
271 if (strcmp(f->tf_name, fs->function_set_name) == 0) {
279 static struct tcp_function_block *
280 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
282 struct tcp_function_block *rblk=NULL;
283 struct tcp_function *f;
285 TAILQ_FOREACH(f, &t_functions, tf_next) {
286 if (f->tf_fb == blk) {
297 struct tcp_function_block *
298 find_and_ref_tcp_functions(struct tcp_function_set *fs)
300 struct tcp_function_block *blk;
302 rw_rlock(&tcp_function_lock);
303 blk = find_tcp_functions_locked(fs);
305 refcount_acquire(&blk->tfb_refcnt);
306 rw_runlock(&tcp_function_lock);
310 struct tcp_function_block *
311 find_and_ref_tcp_fb(struct tcp_function_block *blk)
313 struct tcp_function_block *rblk;
315 rw_rlock(&tcp_function_lock);
316 rblk = find_tcp_fb_locked(blk, NULL);
318 refcount_acquire(&rblk->tfb_refcnt);
319 rw_runlock(&tcp_function_lock);
323 static struct tcp_function_block *
324 find_and_ref_tcp_default_fb(void)
326 struct tcp_function_block *rblk;
328 rw_rlock(&tcp_function_lock);
329 rblk = tcp_func_set_ptr;
330 refcount_acquire(&rblk->tfb_refcnt);
331 rw_runlock(&tcp_function_lock);
336 tcp_switch_back_to_default(struct tcpcb *tp)
338 struct tcp_function_block *tfb;
340 KASSERT(tp->t_fb != &tcp_def_funcblk,
341 ("%s: called by the built-in default stack", __func__));
344 * Release the old stack. This function will either find a new one
347 if (tp->t_fb->tfb_tcp_fb_fini != NULL)
348 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
349 refcount_release(&tp->t_fb->tfb_refcnt);
352 * Now, we'll find a new function block to use.
353 * Start by trying the current user-selected
354 * default, unless this stack is the user-selected
357 tfb = find_and_ref_tcp_default_fb();
358 if (tfb == tp->t_fb) {
359 refcount_release(&tfb->tfb_refcnt);
362 /* Does the stack accept this connection? */
363 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
364 (*tfb->tfb_tcp_handoff_ok)(tp)) {
365 refcount_release(&tfb->tfb_refcnt);
368 /* Try to use that stack. */
370 /* Initialize the new stack. If it succeeds, we are done. */
372 if (tp->t_fb->tfb_tcp_fb_init == NULL ||
373 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
377 * Initialization failed. Release the reference count on
380 refcount_release(&tfb->tfb_refcnt);
384 * If that wasn't feasible, use the built-in default
385 * stack which is not allowed to reject anyone.
387 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
389 /* there always should be a default */
390 panic("Can't refer to tcp_def_funcblk");
392 if (tfb->tfb_tcp_handoff_ok != NULL) {
393 if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
394 /* The default stack cannot say no */
395 panic("Default stack rejects a new session?");
399 if (tp->t_fb->tfb_tcp_fb_init != NULL &&
400 (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
401 /* The default stack cannot fail */
402 panic("Default stack initialization failed");
407 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
410 struct tcp_function_set fs;
411 struct tcp_function_block *blk;
413 memset(&fs, 0, sizeof(fs));
414 rw_rlock(&tcp_function_lock);
415 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
418 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
419 fs.pcbcnt = blk->tfb_refcnt;
421 rw_runlock(&tcp_function_lock);
422 error = sysctl_handle_string(oidp, fs.function_set_name,
423 sizeof(fs.function_set_name), req);
425 /* Check for error or no change */
426 if (error != 0 || req->newptr == NULL)
429 rw_wlock(&tcp_function_lock);
430 blk = find_tcp_functions_locked(&fs);
432 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
436 tcp_func_set_ptr = blk;
438 rw_wunlock(&tcp_function_lock);
442 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
443 CTLTYPE_STRING | CTLFLAG_RW,
444 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
445 "Set/get the default TCP functions");
448 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
450 int error, cnt, linesz;
451 struct tcp_function *f;
457 rw_rlock(&tcp_function_lock);
458 TAILQ_FOREACH(f, &t_functions, tf_next) {
461 rw_runlock(&tcp_function_lock);
463 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
464 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
469 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
470 "Alias", "PCB count");
475 rw_rlock(&tcp_function_lock);
476 TAILQ_FOREACH(f, &t_functions, tf_next) {
477 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
478 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
479 f->tf_fb->tfb_tcp_block_name,
480 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
481 alias ? f->tf_name : "-",
482 f->tf_fb->tfb_refcnt);
483 if (linesz >= bufsz) {
491 rw_runlock(&tcp_function_lock);
493 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
494 free(buffer, M_TEMP);
498 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
499 CTLTYPE_STRING|CTLFLAG_RD,
500 NULL, 0, sysctl_net_inet_list_available, "A",
501 "list available TCP Function sets");
504 * Exports one (struct tcp_function_info) for each alias/name.
507 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
510 struct tcp_function *f;
511 struct tcp_function_info tfi;
514 * We don't allow writes.
516 if (req->newptr != NULL)
520 * Wire the old buffer so we can directly copy the functions to
521 * user space without dropping the lock.
523 if (req->oldptr != NULL) {
524 error = sysctl_wire_old_buffer(req, 0);
530 * Walk the list and copy out matching entries. If INVARIANTS
531 * is compiled in, also walk the list to verify the length of
532 * the list matches what we have recorded.
534 rw_rlock(&tcp_function_lock);
538 if (req->oldptr == NULL) {
543 TAILQ_FOREACH(f, &t_functions, tf_next) {
547 if (req->oldptr != NULL) {
548 bzero(&tfi, sizeof(tfi));
549 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
550 tfi.tfi_id = f->tf_fb->tfb_id;
551 (void)strlcpy(tfi.tfi_alias, f->tf_name,
552 sizeof(tfi.tfi_alias));
553 (void)strlcpy(tfi.tfi_name,
554 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
555 error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
557 * Don't stop on error, as that is the
558 * mechanism we use to accumulate length
559 * information if the buffer was too short.
563 KASSERT(cnt == tcp_fb_cnt,
564 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
568 rw_runlock(&tcp_function_lock);
569 if (req->oldptr == NULL)
570 error = SYSCTL_OUT(req, NULL,
571 (cnt + 1) * sizeof(struct tcp_function_info));
576 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
577 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
578 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
579 "List TCP function block name-to-ID mappings");
582 * tfb_tcp_handoff_ok() function for the default stack.
583 * Note that we'll basically try to take all comers.
586 tcp_default_handoff_ok(struct tcpcb *tp)
593 * tfb_tcp_fb_init() function for the default stack.
595 * This handles making sure we have appropriate timers set if you are
596 * transitioning a socket that has some amount of setup done.
598 * The init() fuction from the default can *never* return non-zero i.e.
599 * it is required to always succeed since it is the stack of last resort!
602 tcp_default_fb_init(struct tcpcb *tp)
607 INP_WLOCK_ASSERT(tp->t_inpcb);
609 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
610 ("%s: connection %p in unexpected state %d", __func__, tp,
614 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
615 * know what to do for unexpected states (which includes TIME_WAIT).
617 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
621 * Make sure some kind of transmission timer is set if there is
624 so = tp->t_inpcb->inp_socket;
625 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
626 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
627 tcp_timer_active(tp, TT_PERSIST))) {
629 * If the session has established and it looks like it should
630 * be in the persist state, set the persist timer. Otherwise,
631 * set the retransmit timer.
633 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
634 (int32_t)(tp->snd_nxt - tp->snd_una) <
635 (int32_t)sbavail(&so->so_snd))
638 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
641 /* All non-embryonic sessions get a keepalive timer. */
642 if (!tcp_timer_active(tp, TT_KEEP))
643 tcp_timer_activate(tp, TT_KEEP,
644 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
651 * tfb_tcp_fb_fini() function for the default stack.
653 * This changes state as necessary (or prudent) to prepare for another stack
654 * to assume responsibility for the connection.
657 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
660 INP_WLOCK_ASSERT(tp->t_inpcb);
665 * Target size of TCP PCB hash tables. Must be a power of two.
667 * Note that this can be overridden by the kernel environment
668 * variable net.inet.tcp.tcbhashsize
671 #define TCBHASHSIZE 0
676 * Callouts should be moved into struct tcp directly. They are currently
677 * separate because the tcpcb structure is exported to userland for sysctl
678 * parsing purposes, which do not know about callouts.
689 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
690 #define V_tcpcb_zone VNET(tcpcb_zone)
692 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
693 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
695 static struct mtx isn_mtx;
697 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
698 #define ISN_LOCK() mtx_lock(&isn_mtx)
699 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
702 * TCP initialization.
705 tcp_zone_change(void *tag)
708 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
709 uma_zone_set_max(V_tcpcb_zone, maxsockets);
710 tcp_tw_zone_change();
714 tcp_inpcb_init(void *mem, int size, int flags)
716 struct inpcb *inp = mem;
718 INP_LOCK_INIT(inp, "inp", "tcpinp");
723 * Take a value and get the next power of 2 that doesn't overflow.
724 * Used to size the tcp_inpcb hash buckets.
727 maketcp_hashsize(int size)
733 * get the next power of 2 higher than maxsockets.
735 hashsize = 1 << fls(size);
736 /* catch overflow, and just go one power of 2 smaller */
737 if (hashsize < size) {
738 hashsize = 1 << (fls(size) - 1);
743 static volatile int next_tcp_stack_id = 1;
746 * Register a TCP function block with the name provided in the names
747 * array. (Note that this function does NOT automatically register
748 * blk->tfb_tcp_block_name as a stack name. Therefore, you should
749 * explicitly include blk->tfb_tcp_block_name in the list of names if
750 * you wish to register the stack with that name.)
752 * Either all name registrations will succeed or all will fail. If
753 * a name registration fails, the function will update the num_names
754 * argument to point to the array index of the name that encountered
757 * Returns 0 on success, or an error code on failure.
760 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
761 const char *names[], int *num_names)
763 struct tcp_function *n;
764 struct tcp_function_set fs;
767 KASSERT(names != NULL && *num_names > 0,
768 ("%s: Called with 0-length name list", __func__));
769 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
770 KASSERT(rw_initialized(&tcp_function_lock),
771 ("%s: called too early", __func__));
773 if ((blk->tfb_tcp_output == NULL) ||
774 (blk->tfb_tcp_do_segment == NULL) ||
775 (blk->tfb_tcp_ctloutput == NULL) ||
776 (strlen(blk->tfb_tcp_block_name) == 0)) {
778 * These functions are required and you
784 if (blk->tfb_tcp_timer_stop_all ||
785 blk->tfb_tcp_timer_activate ||
786 blk->tfb_tcp_timer_active ||
787 blk->tfb_tcp_timer_stop) {
789 * If you define one timer function you
790 * must have them all.
792 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
793 (blk->tfb_tcp_timer_activate == NULL) ||
794 (blk->tfb_tcp_timer_active == NULL) ||
795 (blk->tfb_tcp_timer_stop == NULL)) {
801 refcount_init(&blk->tfb_refcnt, 0);
803 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
804 for (i = 0; i < *num_names; i++) {
805 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
812 (void)strlcpy(fs.function_set_name, names[i],
813 sizeof(fs.function_set_name));
814 rw_wlock(&tcp_function_lock);
815 if (find_tcp_functions_locked(&fs) != NULL) {
816 /* Duplicate name space not allowed */
817 rw_wunlock(&tcp_function_lock);
818 free(n, M_TCPFUNCTIONS);
822 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
823 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
825 rw_wunlock(&tcp_function_lock);
831 * Deregister the names we just added. Because registration failed
832 * for names[i], we don't need to deregister that name.
835 rw_wlock(&tcp_function_lock);
837 TAILQ_FOREACH(n, &t_functions, tf_next) {
838 if (!strncmp(n->tf_name, names[i],
839 TCP_FUNCTION_NAME_LEN_MAX)) {
840 TAILQ_REMOVE(&t_functions, n, tf_next);
843 free(n, M_TCPFUNCTIONS);
848 rw_wunlock(&tcp_function_lock);
853 * Register a TCP function block using the name provided in the name
856 * Returns 0 on success, or an error code on failure.
859 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
862 const char *name_list[1];
869 name_list[0] = blk->tfb_tcp_block_name;
870 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
875 * Register a TCP function block using the name defined in
876 * blk->tfb_tcp_block_name.
878 * Returns 0 on success, or an error code on failure.
881 register_tcp_functions(struct tcp_function_block *blk, int wait)
884 return (register_tcp_functions_as_name(blk, NULL, wait));
888 * Deregister all names associated with a function block. This
889 * functionally removes the function block from use within the system.
891 * When called with a true quiesce argument, mark the function block
892 * as being removed so no more stacks will use it and determine
893 * whether the removal would succeed.
895 * When called with a false quiesce argument, actually attempt the
898 * When called with a force argument, attempt to switch all TCBs to
899 * use the default stack instead of returning EBUSY.
901 * Returns 0 on success (or if the removal would succeed, or an error
905 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
908 struct tcp_function *f;
910 if (blk == &tcp_def_funcblk) {
911 /* You can't un-register the default */
914 rw_wlock(&tcp_function_lock);
915 if (blk == tcp_func_set_ptr) {
916 /* You can't free the current default */
917 rw_wunlock(&tcp_function_lock);
920 /* Mark the block so no more stacks can use it. */
921 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
923 * If TCBs are still attached to the stack, attempt to switch them
924 * to the default stack.
926 if (force && blk->tfb_refcnt) {
929 VNET_ITERATOR_DECL(vnet_iter);
931 rw_wunlock(&tcp_function_lock);
934 VNET_FOREACH(vnet_iter) {
935 CURVNET_SET(vnet_iter);
936 INP_INFO_WLOCK(&V_tcbinfo);
937 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
939 if (inp->inp_flags & INP_TIMEWAIT) {
944 if (tp == NULL || tp->t_fb != blk) {
948 tcp_switch_back_to_default(tp);
951 INP_INFO_WUNLOCK(&V_tcbinfo);
956 rw_wlock(&tcp_function_lock);
958 if (blk->tfb_refcnt) {
959 /* TCBs still attached. */
960 rw_wunlock(&tcp_function_lock);
965 rw_wunlock(&tcp_function_lock);
968 /* Remove any function names that map to this function block. */
969 while (find_tcp_fb_locked(blk, &f) != NULL) {
970 TAILQ_REMOVE(&t_functions, f, tf_next);
973 free(f, M_TCPFUNCTIONS);
975 rw_wunlock(&tcp_function_lock);
982 const char *tcbhash_tuneable;
985 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
988 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
989 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
990 printf("%s: WARNING: unable to register helper hook\n", __func__);
991 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
992 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
993 printf("%s: WARNING: unable to register helper hook\n", __func__);
995 hashsize = TCBHASHSIZE;
996 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
999 * Auto tune the hash size based on maxsockets.
1000 * A perfect hash would have a 1:1 mapping
1001 * (hashsize = maxsockets) however it's been
1002 * suggested that O(2) average is better.
1004 hashsize = maketcp_hashsize(maxsockets / 4);
1006 * Our historical default is 512,
1007 * do not autotune lower than this.
1011 if (bootverbose && IS_DEFAULT_VNET(curvnet))
1012 printf("%s: %s auto tuned to %d\n", __func__,
1013 tcbhash_tuneable, hashsize);
1016 * We require a hashsize to be a power of two.
1017 * Previously if it was not a power of two we would just reset it
1018 * back to 512, which could be a nasty surprise if you did not notice
1019 * the error message.
1020 * Instead what we do is clip it to the closest power of two lower
1021 * than the specified hash value.
1023 if (!powerof2(hashsize)) {
1024 int oldhashsize = hashsize;
1026 hashsize = maketcp_hashsize(hashsize);
1027 /* prevent absurdly low value */
1030 printf("%s: WARNING: TCB hash size not a power of 2, "
1031 "clipped from %d to %d.\n", __func__, oldhashsize,
1034 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
1035 "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
1038 * These have to be type stable for the benefit of the timers.
1040 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1041 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1042 uma_zone_set_max(V_tcpcb_zone, maxsockets);
1043 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1049 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1050 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1051 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1053 tcp_fastopen_init();
1055 /* Skip initialization of globals for non-default instances. */
1056 if (!IS_DEFAULT_VNET(curvnet))
1059 tcp_reass_global_init();
1061 /* XXX virtualize those bellow? */
1062 tcp_delacktime = TCPTV_DELACK;
1063 tcp_keepinit = TCPTV_KEEP_INIT;
1064 tcp_keepidle = TCPTV_KEEP_IDLE;
1065 tcp_keepintvl = TCPTV_KEEPINTVL;
1066 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1067 tcp_msl = TCPTV_MSL;
1068 tcp_rexmit_min = TCPTV_MIN;
1069 if (tcp_rexmit_min < 1)
1071 tcp_persmin = TCPTV_PERSMIN;
1072 tcp_persmax = TCPTV_PERSMAX;
1073 tcp_rexmit_slop = TCPTV_CPU_VAR;
1074 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1075 tcp_tcbhashsize = hashsize;
1077 /* Setup the tcp function block list */
1078 TAILQ_INIT(&t_functions);
1079 rw_init(&tcp_function_lock, "tcp_func_lock");
1080 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1082 /* Initialize the TCP logging data. */
1085 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1087 if (tcp_soreceive_stream) {
1089 tcp_usrreqs.pru_soreceive = soreceive_stream;
1092 tcp6_usrreqs.pru_soreceive = soreceive_stream;
1097 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1099 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1101 if (max_protohdr < TCP_MINPROTOHDR)
1102 max_protohdr = TCP_MINPROTOHDR;
1103 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1105 #undef TCP_MINPROTOHDR
1108 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1109 SHUTDOWN_PRI_DEFAULT);
1110 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
1111 EVENTHANDLER_PRI_ANY);
1119 tcp_destroy(void *unused __unused)
1127 * All our processes are gone, all our sockets should be cleaned
1128 * up, which means, we should be past the tcp_discardcb() calls.
1129 * Sleep to let all tcpcb timers really disappear and cleanup.
1132 INP_LIST_RLOCK(&V_tcbinfo);
1133 n = V_tcbinfo.ipi_count;
1134 INP_LIST_RUNLOCK(&V_tcbinfo);
1137 pause("tcpdes", hz / 10);
1142 in_pcbinfo_destroy(&V_tcbinfo);
1143 /* tcp_discardcb() clears the sack_holes up. */
1144 uma_zdestroy(V_sack_hole_zone);
1145 uma_zdestroy(V_tcpcb_zone);
1148 * Cannot free the zone until all tcpcbs are released as we attach
1149 * the allocations to them.
1151 tcp_fastopen_destroy();
1154 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1156 printf("%s: WARNING: unable to deregister helper hook "
1157 "type=%d, id=%d: error %d returned\n", __func__,
1158 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1160 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1162 printf("%s: WARNING: unable to deregister helper hook "
1163 "type=%d, id=%d: error %d returned\n", __func__,
1164 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1168 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1178 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1179 * tcp_template used to store this data in mbufs, but we now recopy it out
1180 * of the tcpcb each time to conserve mbufs.
1183 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
1185 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1187 INP_WLOCK_ASSERT(inp);
1190 if ((inp->inp_vflag & INP_IPV6) != 0) {
1191 struct ip6_hdr *ip6;
1193 ip6 = (struct ip6_hdr *)ip_ptr;
1194 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1195 (inp->inp_flow & IPV6_FLOWINFO_MASK);
1196 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1197 (IPV6_VERSION & IPV6_VERSION_MASK);
1198 ip6->ip6_nxt = IPPROTO_TCP;
1199 ip6->ip6_plen = htons(sizeof(struct tcphdr));
1200 ip6->ip6_src = inp->in6p_laddr;
1201 ip6->ip6_dst = inp->in6p_faddr;
1204 #if defined(INET6) && defined(INET)
1211 ip = (struct ip *)ip_ptr;
1212 ip->ip_v = IPVERSION;
1214 ip->ip_tos = inp->inp_ip_tos;
1218 ip->ip_ttl = inp->inp_ip_ttl;
1220 ip->ip_p = IPPROTO_TCP;
1221 ip->ip_src = inp->inp_laddr;
1222 ip->ip_dst = inp->inp_faddr;
1225 th->th_sport = inp->inp_lport;
1226 th->th_dport = inp->inp_fport;
1234 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
1238 * Create template to be used to send tcp packets on a connection.
1239 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
1240 * use for this function is in keepalives, which use tcp_respond.
1243 tcpip_maketemplate(struct inpcb *inp)
1247 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1250 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1255 * Send a single message to the TCP at address specified by
1256 * the given TCP/IP header. If m == NULL, then we make a copy
1257 * of the tcpiphdr at th and send directly to the addressed host.
1258 * This is used to force keep alive messages out using the TCP
1259 * template for a connection. If flags are given then we send
1260 * a message back to the TCP which originated the segment th,
1261 * and discard the mbuf containing it and any other attached mbufs.
1263 * In any case the ack and sequence number of the transmitted
1264 * segment are as specified by the parameters.
1266 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1269 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1270 tcp_seq ack, tcp_seq seq, int flags)
1279 struct ip6_hdr *ip6;
1282 int optlen, tlen, win;
1285 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1288 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1295 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1296 INP_WLOCK_ASSERT(inp);
1303 if (!(flags & TH_RST)) {
1304 win = sbspace(&inp->inp_socket->so_rcv);
1305 if (win > TCP_MAXWIN << tp->rcv_scale)
1306 win = TCP_MAXWIN << tp->rcv_scale;
1308 if ((tp->t_flags & TF_NOOPT) == 0)
1312 m = m_gethdr(M_NOWAIT, MT_DATA);
1315 m->m_data += max_linkhdr;
1318 bcopy((caddr_t)ip6, mtod(m, caddr_t),
1319 sizeof(struct ip6_hdr));
1320 ip6 = mtod(m, struct ip6_hdr *);
1321 nth = (struct tcphdr *)(ip6 + 1);
1325 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1326 ip = mtod(m, struct ip *);
1327 nth = (struct tcphdr *)(ip + 1);
1329 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1331 } else if (!M_WRITABLE(m)) {
1334 /* Can't reuse 'm', allocate a new mbuf. */
1335 n = m_gethdr(M_NOWAIT, MT_DATA);
1341 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1347 n->m_data += max_linkhdr;
1348 /* m_len is set later */
1349 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1352 bcopy((caddr_t)ip6, mtod(n, caddr_t),
1353 sizeof(struct ip6_hdr));
1354 ip6 = mtod(n, struct ip6_hdr *);
1355 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1356 nth = (struct tcphdr *)(ip6 + 1);
1360 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1361 ip = mtod(n, struct ip *);
1362 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1363 nth = (struct tcphdr *)(ip + 1);
1365 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1366 xchg(nth->th_dport, nth->th_sport, uint16_t);
1373 * XXX MRT We inherit the FIB, which is lucky.
1377 m->m_data = (caddr_t)ipgen;
1378 /* m_len is set later */
1381 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1382 nth = (struct tcphdr *)(ip6 + 1);
1386 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1387 nth = (struct tcphdr *)(ip + 1);
1391 * this is usually a case when an extension header
1392 * exists between the IPv6 header and the
1395 nth->th_sport = th->th_sport;
1396 nth->th_dport = th->th_dport;
1398 xchg(nth->th_dport, nth->th_sport, uint16_t);
1404 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1406 #if defined(INET) && defined(INET6)
1410 tlen = sizeof (struct tcpiphdr);
1414 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1415 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1416 m, tlen, (long)M_TRAILINGSPACE(m)));
1421 /* Make sure we have room. */
1422 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1423 m->m_next = m_get(M_NOWAIT, MT_DATA);
1425 optp = mtod(m->m_next, u_char *);
1430 optp = (u_char *) (nth + 1);
1436 if (tp->t_flags & TF_RCVD_TSTMP) {
1437 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1438 to.to_tsecr = tp->ts_recent;
1439 to.to_flags |= TOF_TS;
1441 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1442 /* TCP-MD5 (RFC2385). */
1443 if (tp->t_flags & TF_SIGNATURE)
1444 to.to_flags |= TOF_SIGNATURE;
1446 /* Add the options. */
1447 tlen += optlen = tcp_addoptions(&to, optp);
1449 /* Update m_len in the correct mbuf. */
1450 optm->m_len += optlen;
1456 ip6->ip6_vfc = IPV6_VERSION;
1457 ip6->ip6_nxt = IPPROTO_TCP;
1458 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1461 #if defined(INET) && defined(INET6)
1466 ip->ip_len = htons(tlen);
1467 ip->ip_ttl = V_ip_defttl;
1468 if (V_path_mtu_discovery)
1469 ip->ip_off |= htons(IP_DF);
1472 m->m_pkthdr.len = tlen;
1473 m->m_pkthdr.rcvif = NULL;
1477 * Packet is associated with a socket, so allow the
1478 * label of the response to reflect the socket label.
1480 INP_WLOCK_ASSERT(inp);
1481 mac_inpcb_create_mbuf(inp, m);
1484 * Packet is not associated with a socket, so possibly
1485 * update the label in place.
1487 mac_netinet_tcp_reply(m);
1490 nth->th_seq = htonl(seq);
1491 nth->th_ack = htonl(ack);
1493 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1494 nth->th_flags = flags;
1496 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1498 nth->th_win = htons((u_short)win);
1501 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1502 if (to.to_flags & TOF_SIGNATURE) {
1503 if (!TCPMD5_ENABLED() ||
1504 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1511 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1514 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1515 nth->th_sum = in6_cksum_pseudo(ip6,
1516 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1517 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1521 #if defined(INET6) && defined(INET)
1526 m->m_pkthdr.csum_flags = CSUM_TCP;
1527 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1528 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1532 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1533 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1535 TCP_PROBE3(debug__output, tp, th, m);
1537 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1541 TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1542 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1545 #if defined(INET) && defined(INET6)
1550 TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1551 (void)ip_output(m, NULL, NULL, 0, NULL, inp);
1557 * Create a new TCP control block, making an
1558 * empty reassembly queue and hooking it to the argument
1559 * protocol control block. The `inp' parameter must have
1560 * come from the zone allocator set up in tcp_init().
1563 tcp_newtcpcb(struct inpcb *inp)
1565 struct tcpcb_mem *tm;
1568 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1571 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1576 /* Initialise cc_var struct for this tcpcb. */
1578 tp->ccv->type = IPPROTO_TCP;
1579 tp->ccv->ccvc.tcp = tp;
1580 rw_rlock(&tcp_function_lock);
1581 tp->t_fb = tcp_func_set_ptr;
1582 refcount_acquire(&tp->t_fb->tfb_refcnt);
1583 rw_runlock(&tcp_function_lock);
1585 * Use the current system default CC algorithm.
1588 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1589 CC_ALGO(tp) = CC_DEFAULT();
1592 if (CC_ALGO(tp)->cb_init != NULL)
1593 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1594 if (tp->t_fb->tfb_tcp_fb_fini)
1595 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1596 refcount_release(&tp->t_fb->tfb_refcnt);
1597 uma_zfree(V_tcpcb_zone, tm);
1603 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1604 if (tp->t_fb->tfb_tcp_fb_fini)
1605 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1606 refcount_release(&tp->t_fb->tfb_refcnt);
1607 uma_zfree(V_tcpcb_zone, tm);
1613 tp->t_vnet = inp->inp_vnet;
1615 tp->t_timers = &tm->tt;
1616 TAILQ_INIT(&tp->t_segq);
1619 isipv6 ? V_tcp_v6mssdflt :
1623 /* Set up our timeouts. */
1624 callout_init(&tp->t_timers->tt_rexmt, 1);
1625 callout_init(&tp->t_timers->tt_persist, 1);
1626 callout_init(&tp->t_timers->tt_keep, 1);
1627 callout_init(&tp->t_timers->tt_2msl, 1);
1628 callout_init(&tp->t_timers->tt_delack, 1);
1630 if (V_tcp_do_rfc1323)
1631 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1633 tp->t_flags |= TF_SACK_PERMIT;
1634 TAILQ_INIT(&tp->snd_holes);
1636 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1639 in_pcbref(inp); /* Reference for tcpcb */
1643 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1644 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1645 * reasonable initial retransmit time.
1647 tp->t_srtt = TCPTV_SRTTBASE;
1648 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1649 tp->t_rttmin = tcp_rexmit_min;
1650 tp->t_rxtcur = TCPTV_RTOBASE;
1651 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1652 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1653 tp->t_rcvtime = ticks;
1655 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1656 * because the socket may be bound to an IPv6 wildcard address,
1657 * which may match an IPv4-mapped IPv6 address.
1659 inp->inp_ip_ttl = V_ip_defttl;
1663 * Init the TCP PCAP queues.
1665 tcp_pcap_tcpcb_init(tp);
1668 /* Initialize the per-TCPCB log data. */
1669 tcp_log_tcpcbinit(tp);
1671 if (tp->t_fb->tfb_tcp_fb_init) {
1672 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1674 return (tp); /* XXX */
1678 * Switch the congestion control algorithm back to NewReno for any active
1679 * control blocks using an algorithm which is about to go away.
1680 * This ensures the CC framework can allow the unload to proceed without leaving
1681 * any dangling pointers which would trigger a panic.
1682 * Returning non-zero would inform the CC framework that something went wrong
1683 * and it would be unsafe to allow the unload to proceed. However, there is no
1684 * way for this to occur with this implementation so we always return zero.
1687 tcp_ccalgounload(struct cc_algo *unload_algo)
1689 struct cc_algo *tmpalgo;
1692 VNET_ITERATOR_DECL(vnet_iter);
1695 * Check all active control blocks across all network stacks and change
1696 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1697 * requires cleanup code to be run, call it.
1700 VNET_FOREACH(vnet_iter) {
1701 CURVNET_SET(vnet_iter);
1702 INP_INFO_WLOCK(&V_tcbinfo);
1704 * New connections already part way through being initialised
1705 * with the CC algo we're removing will not race with this code
1706 * because the INP_INFO_WLOCK is held during initialisation. We
1707 * therefore don't enter the loop below until the connection
1708 * list has stabilised.
1710 CK_LIST_FOREACH(inp, &V_tcb, inp_list) {
1712 /* Important to skip tcptw structs. */
1713 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1714 (tp = intotcpcb(inp)) != NULL) {
1716 * By holding INP_WLOCK here, we are assured
1717 * that the connection is not currently
1718 * executing inside the CC module's functions
1719 * i.e. it is safe to make the switch back to
1722 if (CC_ALGO(tp) == unload_algo) {
1723 tmpalgo = CC_ALGO(tp);
1724 if (tmpalgo->cb_destroy != NULL)
1725 tmpalgo->cb_destroy(tp->ccv);
1728 * NewReno may allocate memory on
1729 * demand for certain stateful
1730 * configuration as needed, but is
1731 * coded to never fail on memory
1732 * allocation failure so it is a safe
1735 CC_ALGO(tp) = &newreno_cc_algo;
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;
1781 int released __unused;
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);
1890 khelp_destroy_osd(tp->osd);
1894 inp->inp_ppcb = NULL;
1895 if (tp->t_timers->tt_draincnt == 0) {
1896 /* We own the last reference on tcpcb, let's free it. */
1898 tcp_log_tcpcbfini(tp);
1900 TCPSTATES_DEC(tp->t_state);
1901 if (tp->t_fb->tfb_tcp_fb_fini)
1902 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1903 refcount_release(&tp->t_fb->tfb_refcnt);
1905 uma_zfree(V_tcpcb_zone, tp);
1906 released = in_pcbrele_wlocked(inp);
1907 KASSERT(!released, ("%s: inp %p should not have been released "
1908 "here", __func__, inp));
1913 tcp_timer_discard(void *ptp)
1917 struct epoch_tracker et;
1919 tp = (struct tcpcb *)ptp;
1920 CURVNET_SET(tp->t_vnet);
1921 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
1923 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1926 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1927 ("%s: tcpcb has to be stopped here", __func__));
1928 tp->t_timers->tt_draincnt--;
1929 if (tp->t_timers->tt_draincnt == 0) {
1930 /* We own the last reference on this tcpcb, let's free it. */
1932 tcp_log_tcpcbfini(tp);
1934 TCPSTATES_DEC(tp->t_state);
1935 if (tp->t_fb->tfb_tcp_fb_fini)
1936 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1937 refcount_release(&tp->t_fb->tfb_refcnt);
1939 uma_zfree(V_tcpcb_zone, tp);
1940 if (in_pcbrele_wlocked(inp)) {
1941 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1947 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1952 * Attempt to close a TCP control block, marking it as dropped, and freeing
1953 * the socket if we hold the only reference.
1956 tcp_close(struct tcpcb *tp)
1958 struct inpcb *inp = tp->t_inpcb;
1961 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1962 INP_WLOCK_ASSERT(inp);
1965 if (tp->t_state == TCPS_LISTEN)
1966 tcp_offload_listen_stop(tp);
1969 * This releases the TFO pending counter resource for TFO listen
1970 * sockets as well as passively-created TFO sockets that transition
1971 * from SYN_RECEIVED to CLOSED.
1973 if (tp->t_tfo_pending) {
1974 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1975 tp->t_tfo_pending = NULL;
1978 TCPSTAT_INC(tcps_closed);
1979 if (tp->t_state != TCPS_CLOSED)
1980 tcp_state_change(tp, TCPS_CLOSED);
1981 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1982 so = inp->inp_socket;
1983 soisdisconnected(so);
1984 if (inp->inp_flags & INP_SOCKREF) {
1985 KASSERT(so->so_state & SS_PROTOREF,
1986 ("tcp_close: !SS_PROTOREF"));
1987 inp->inp_flags &= ~INP_SOCKREF;
1990 so->so_state &= ~SS_PROTOREF;
2000 VNET_ITERATOR_DECL(vnet_iter);
2005 VNET_LIST_RLOCK_NOSLEEP();
2006 VNET_FOREACH(vnet_iter) {
2007 CURVNET_SET(vnet_iter);
2012 * Walk the tcpbs, if existing, and flush the reassembly queue,
2013 * if there is one...
2014 * XXX: The "Net/3" implementation doesn't imply that the TCP
2015 * reassembly queue should be flushed, but in a situation
2016 * where we're really low on mbufs, this is potentially
2019 INP_INFO_WLOCK(&V_tcbinfo);
2020 CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
2022 if (inpb->inp_flags & INP_TIMEWAIT) {
2026 if ((tcpb = intotcpcb(inpb)) != NULL) {
2027 tcp_reass_flush(tcpb);
2028 tcp_clean_sackreport(tcpb);
2030 tcp_log_drain(tcpb);
2033 if (tcp_pcap_aggressive_free) {
2034 /* Free the TCP PCAP queues. */
2035 tcp_pcap_drain(&(tcpb->t_inpkts));
2036 tcp_pcap_drain(&(tcpb->t_outpkts));
2042 INP_INFO_WUNLOCK(&V_tcbinfo);
2045 VNET_LIST_RUNLOCK_NOSLEEP();
2049 * Notify a tcp user of an asynchronous error;
2050 * store error as soft error, but wake up user
2051 * (for now, won't do anything until can select for soft error).
2053 * Do not wake up user since there currently is no mechanism for
2054 * reporting soft errors (yet - a kqueue filter may be added).
2056 static struct inpcb *
2057 tcp_notify(struct inpcb *inp, int error)
2061 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2062 INP_WLOCK_ASSERT(inp);
2064 if ((inp->inp_flags & INP_TIMEWAIT) ||
2065 (inp->inp_flags & INP_DROPPED))
2068 tp = intotcpcb(inp);
2069 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2072 * Ignore some errors if we are hooked up.
2073 * If connection hasn't completed, has retransmitted several times,
2074 * and receives a second error, give up now. This is better
2075 * than waiting a long time to establish a connection that
2076 * can never complete.
2078 if (tp->t_state == TCPS_ESTABLISHED &&
2079 (error == EHOSTUNREACH || error == ENETUNREACH ||
2080 error == EHOSTDOWN)) {
2081 if (inp->inp_route.ro_rt) {
2082 RTFREE(inp->inp_route.ro_rt);
2083 inp->inp_route.ro_rt = (struct rtentry *)NULL;
2086 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2088 tp = tcp_drop(tp, error);
2094 tp->t_softerror = error;
2098 wakeup( &so->so_timeo);
2105 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2107 int error, i, m, n, pcb_count;
2108 struct inpcb *inp, **inp_list;
2111 struct epoch_tracker et;
2114 * The process of preparing the TCB list is too time-consuming and
2115 * resource-intensive to repeat twice on every request.
2117 if (req->oldptr == NULL) {
2118 n = V_tcbinfo.ipi_count +
2119 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2120 n += imax(n / 8, 10);
2121 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2125 if (req->newptr != NULL)
2129 * OK, now we're committed to doing something.
2131 INP_LIST_RLOCK(&V_tcbinfo);
2132 gencnt = V_tcbinfo.ipi_gencnt;
2133 n = V_tcbinfo.ipi_count;
2134 INP_LIST_RUNLOCK(&V_tcbinfo);
2136 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2138 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
2139 + (n + m) * sizeof(struct xtcpcb));
2143 bzero(&xig, sizeof(xig));
2144 xig.xig_len = sizeof xig;
2145 xig.xig_count = n + m;
2146 xig.xig_gen = gencnt;
2147 xig.xig_sogen = so_gencnt;
2148 error = SYSCTL_OUT(req, &xig, sizeof xig);
2152 error = syncache_pcblist(req, m, &pcb_count);
2156 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
2158 INP_INFO_WLOCK(&V_tcbinfo);
2159 for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
2160 inp != NULL && i < n; inp = CK_LIST_NEXT(inp, inp_list)) {
2162 if (inp->inp_gencnt <= gencnt) {
2164 * XXX: This use of cr_cansee(), introduced with
2165 * TCP state changes, is not quite right, but for
2166 * now, better than nothing.
2168 if (inp->inp_flags & INP_TIMEWAIT) {
2169 if (intotw(inp) != NULL)
2170 error = cr_cansee(req->td->td_ucred,
2171 intotw(inp)->tw_cred);
2173 error = EINVAL; /* Skip this inp. */
2175 error = cr_canseeinpcb(req->td->td_ucred, inp);
2178 inp_list[i++] = inp;
2183 INP_INFO_WUNLOCK(&V_tcbinfo);
2187 for (i = 0; i < n; i++) {
2190 if (inp->inp_gencnt <= gencnt) {
2193 tcp_inptoxtp(inp, &xt);
2195 error = SYSCTL_OUT(req, &xt, sizeof xt);
2199 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2200 for (i = 0; i < n; i++) {
2203 if (!in_pcbrele_rlocked(inp))
2206 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2210 * Give the user an updated idea of our state.
2211 * If the generation differs from what we told
2212 * her before, she knows that something happened
2213 * while we were processing this request, and it
2214 * might be necessary to retry.
2216 INP_LIST_RLOCK(&V_tcbinfo);
2217 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2218 xig.xig_sogen = so_gencnt;
2219 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
2220 INP_LIST_RUNLOCK(&V_tcbinfo);
2221 error = SYSCTL_OUT(req, &xig, sizeof xig);
2223 free(inp_list, M_TEMP);
2227 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2228 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
2229 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
2233 tcp_getcred(SYSCTL_HANDLER_ARGS)
2236 struct sockaddr_in addrs[2];
2240 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2243 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2246 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2247 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2249 if (inp->inp_socket == NULL)
2252 error = cr_canseeinpcb(req->td->td_ucred, inp);
2254 cru2x(inp->inp_cred, &xuc);
2259 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2263 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2264 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2265 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
2270 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2273 struct sockaddr_in6 addrs[2];
2280 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2283 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2286 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2287 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2290 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2292 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2301 inp = in_pcblookup(&V_tcbinfo,
2302 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2304 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2305 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2308 inp = in6_pcblookup(&V_tcbinfo,
2309 &addrs[1].sin6_addr, addrs[1].sin6_port,
2310 &addrs[0].sin6_addr, addrs[0].sin6_port,
2311 INPLOOKUP_RLOCKPCB, NULL);
2313 if (inp->inp_socket == NULL)
2316 error = cr_canseeinpcb(req->td->td_ucred, inp);
2318 cru2x(inp->inp_cred, &xuc);
2323 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2327 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2328 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2329 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
2335 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2337 struct ip *ip = vip;
2339 struct in_addr faddr;
2342 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2344 struct in_conninfo inc;
2345 struct epoch_tracker et;
2346 tcp_seq icmp_tcp_seq;
2349 faddr = ((struct sockaddr_in *)sa)->sin_addr;
2350 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2353 if (cmd == PRC_MSGSIZE)
2354 notify = tcp_mtudisc_notify;
2355 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2356 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2357 cmd == PRC_TIMXCEED_INTRANS) && ip)
2358 notify = tcp_drop_syn_sent;
2361 * Hostdead is ugly because it goes linearly through all PCBs.
2362 * XXX: We never get this from ICMP, otherwise it makes an
2363 * excellent DoS attack on machines with many connections.
2365 else if (cmd == PRC_HOSTDEAD)
2367 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2371 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2375 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2376 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2377 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2378 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2379 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2380 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2381 /* signal EHOSTDOWN, as it flushes the cached route */
2382 inp = (*notify)(inp, EHOSTDOWN);
2385 icmp_tcp_seq = th->th_seq;
2387 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2388 !(inp->inp_flags & INP_DROPPED) &&
2389 !(inp->inp_socket == NULL)) {
2390 tp = intotcpcb(inp);
2391 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2392 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2393 if (cmd == PRC_MSGSIZE) {
2396 * If we got a needfrag set the MTU
2397 * in the route to the suggested new
2398 * value (if given) and then notify.
2400 mtu = ntohs(icp->icmp_nextmtu);
2402 * If no alternative MTU was
2403 * proposed, try the next smaller
2408 ntohs(ip->ip_len), 1);
2409 if (mtu < V_tcp_minmss +
2410 sizeof(struct tcpiphdr))
2411 mtu = V_tcp_minmss +
2412 sizeof(struct tcpiphdr);
2414 * Only process the offered MTU if it
2415 * is smaller than the current one.
2417 if (mtu < tp->t_maxseg +
2418 sizeof(struct tcpiphdr)) {
2419 bzero(&inc, sizeof(inc));
2420 inc.inc_faddr = faddr;
2422 inp->inp_inc.inc_fibnum;
2423 tcp_hc_updatemtu(&inc, mtu);
2424 tcp_mtudisc(inp, mtu);
2427 inp = (*notify)(inp,
2428 inetctlerrmap[cmd]);
2432 bzero(&inc, sizeof(inc));
2433 inc.inc_fport = th->th_dport;
2434 inc.inc_lport = th->th_sport;
2435 inc.inc_faddr = faddr;
2436 inc.inc_laddr = ip->ip_src;
2437 syncache_unreach(&inc, icmp_tcp_seq);
2442 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2448 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2450 struct in6_addr *dst;
2451 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2452 struct ip6_hdr *ip6;
2456 struct icmp6_hdr *icmp6;
2457 struct ip6ctlparam *ip6cp = NULL;
2458 const struct sockaddr_in6 *sa6_src = NULL;
2459 struct in_conninfo inc;
2460 struct epoch_tracker et;
2465 tcp_seq icmp_tcp_seq;
2469 if (sa->sa_family != AF_INET6 ||
2470 sa->sa_len != sizeof(struct sockaddr_in6))
2473 /* if the parameter is from icmp6, decode it. */
2475 ip6cp = (struct ip6ctlparam *)d;
2476 icmp6 = ip6cp->ip6c_icmp6;
2478 ip6 = ip6cp->ip6c_ip6;
2479 off = ip6cp->ip6c_off;
2480 sa6_src = ip6cp->ip6c_src;
2481 dst = ip6cp->ip6c_finaldst;
2485 off = 0; /* fool gcc */
2490 if (cmd == PRC_MSGSIZE)
2491 notify = tcp_mtudisc_notify;
2492 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2493 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2494 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2495 notify = tcp_drop_syn_sent;
2498 * Hostdead is ugly because it goes linearly through all PCBs.
2499 * XXX: We never get this from ICMP, otherwise it makes an
2500 * excellent DoS attack on machines with many connections.
2502 else if (cmd == PRC_HOSTDEAD)
2504 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2508 in6_pcbnotify(&V_tcbinfo, sa, 0,
2509 (const struct sockaddr *)sa6_src,
2510 0, cmd, NULL, notify);
2514 /* Check if we can safely get the ports from the tcp hdr */
2517 (int32_t) (off + sizeof(struct tcp_ports)))) {
2520 bzero(&t_ports, sizeof(struct tcp_ports));
2521 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2522 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2523 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2524 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2525 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2526 /* signal EHOSTDOWN, as it flushes the cached route */
2527 inp = (*notify)(inp, EHOSTDOWN);
2530 off += sizeof(struct tcp_ports);
2531 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2534 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2536 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2537 !(inp->inp_flags & INP_DROPPED) &&
2538 !(inp->inp_socket == NULL)) {
2539 tp = intotcpcb(inp);
2540 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2541 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2542 if (cmd == PRC_MSGSIZE) {
2545 * If we got a needfrag set the MTU
2546 * in the route to the suggested new
2547 * value (if given) and then notify.
2549 mtu = ntohl(icmp6->icmp6_mtu);
2551 * If no alternative MTU was
2552 * proposed, or the proposed
2553 * MTU was too small, set to
2556 if (mtu < IPV6_MMTU)
2557 mtu = IPV6_MMTU - 8;
2558 bzero(&inc, sizeof(inc));
2559 inc.inc_fibnum = M_GETFIB(m);
2560 inc.inc_flags |= INC_ISIPV6;
2561 inc.inc6_faddr = *dst;
2562 if (in6_setscope(&inc.inc6_faddr,
2563 m->m_pkthdr.rcvif, NULL))
2566 * Only process the offered MTU if it
2567 * is smaller than the current one.
2569 if (mtu < tp->t_maxseg +
2570 sizeof (struct tcphdr) +
2571 sizeof (struct ip6_hdr)) {
2572 tcp_hc_updatemtu(&inc, mtu);
2573 tcp_mtudisc(inp, mtu);
2574 ICMP6STAT_INC(icp6s_pmtuchg);
2577 inp = (*notify)(inp,
2578 inet6ctlerrmap[cmd]);
2582 bzero(&inc, sizeof(inc));
2583 inc.inc_fibnum = M_GETFIB(m);
2584 inc.inc_flags |= INC_ISIPV6;
2585 inc.inc_fport = t_ports.th_dport;
2586 inc.inc_lport = t_ports.th_sport;
2587 inc.inc6_faddr = *dst;
2588 inc.inc6_laddr = ip6->ip6_src;
2589 syncache_unreach(&inc, icmp_tcp_seq);
2594 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2599 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
2605 MD5Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
2606 MD5Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
2607 switch (inc->inc_flags & INC_ISIPV6) {
2610 MD5Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
2611 MD5Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
2616 MD5Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
2617 MD5Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
2621 MD5Update(&ctx, key, len);
2622 MD5Final((unsigned char *)hash, &ctx);
2628 tcp_new_ts_offset(struct in_conninfo *inc)
2630 return (tcp_keyed_hash(inc, V_ts_offset_secret,
2631 sizeof(V_ts_offset_secret)));
2635 * Following is where TCP initial sequence number generation occurs.
2637 * There are two places where we must use initial sequence numbers:
2638 * 1. In SYN-ACK packets.
2639 * 2. In SYN packets.
2641 * All ISNs for SYN-ACK packets are generated by the syncache. See
2642 * tcp_syncache.c for details.
2644 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2645 * depends on this property. In addition, these ISNs should be
2646 * unguessable so as to prevent connection hijacking. To satisfy
2647 * the requirements of this situation, the algorithm outlined in
2648 * RFC 1948 is used, with only small modifications.
2650 * Implementation details:
2652 * Time is based off the system timer, and is corrected so that it
2653 * increases by one megabyte per second. This allows for proper
2654 * recycling on high speed LANs while still leaving over an hour
2657 * As reading the *exact* system time is too expensive to be done
2658 * whenever setting up a TCP connection, we increment the time
2659 * offset in two ways. First, a small random positive increment
2660 * is added to isn_offset for each connection that is set up.
2661 * Second, the function tcp_isn_tick fires once per clock tick
2662 * and increments isn_offset as necessary so that sequence numbers
2663 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2664 * random positive increments serve only to ensure that the same
2665 * exact sequence number is never sent out twice (as could otherwise
2666 * happen when a port is recycled in less than the system tick
2669 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2670 * between seeding of isn_secret. This is normally set to zero,
2671 * as reseeding should not be necessary.
2673 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2674 * isn_offset_old, and isn_ctx is performed using the ISN lock. In
2675 * general, this means holding an exclusive (write) lock.
2678 #define ISN_BYTES_PER_SECOND 1048576
2679 #define ISN_STATIC_INCREMENT 4096
2680 #define ISN_RANDOM_INCREMENT (4096 - 1)
2681 #define ISN_SECRET_LENGTH 32
2683 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
2684 VNET_DEFINE_STATIC(int, isn_last);
2685 VNET_DEFINE_STATIC(int, isn_last_reseed);
2686 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
2687 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
2689 #define V_isn_secret VNET(isn_secret)
2690 #define V_isn_last VNET(isn_last)
2691 #define V_isn_last_reseed VNET(isn_last_reseed)
2692 #define V_isn_offset VNET(isn_offset)
2693 #define V_isn_offset_old VNET(isn_offset_old)
2696 tcp_new_isn(struct in_conninfo *inc)
2699 u_int32_t projected_offset;
2702 /* Seed if this is the first use, reseed if requested. */
2703 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2704 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2706 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
2707 V_isn_last_reseed = ticks;
2710 /* Compute the md5 hash and return the ISN. */
2711 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
2712 sizeof(V_isn_secret));
2713 V_isn_offset += ISN_STATIC_INCREMENT +
2714 (arc4random() & ISN_RANDOM_INCREMENT);
2715 if (ticks != V_isn_last) {
2716 projected_offset = V_isn_offset_old +
2717 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2718 if (SEQ_GT(projected_offset, V_isn_offset))
2719 V_isn_offset = projected_offset;
2720 V_isn_offset_old = V_isn_offset;
2723 new_isn += V_isn_offset;
2729 * When a specific ICMP unreachable message is received and the
2730 * connection state is SYN-SENT, drop the connection. This behavior
2731 * is controlled by the icmp_may_rst sysctl.
2734 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2738 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2739 INP_WLOCK_ASSERT(inp);
2741 if ((inp->inp_flags & INP_TIMEWAIT) ||
2742 (inp->inp_flags & INP_DROPPED))
2745 tp = intotcpcb(inp);
2746 if (tp->t_state != TCPS_SYN_SENT)
2749 if (IS_FASTOPEN(tp->t_flags))
2750 tcp_fastopen_disable_path(tp);
2752 tp = tcp_drop(tp, errno);
2760 * When `need fragmentation' ICMP is received, update our idea of the MSS
2761 * based on the new value. Also nudge TCP to send something, since we
2762 * know the packet we just sent was dropped.
2763 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2765 static struct inpcb *
2766 tcp_mtudisc_notify(struct inpcb *inp, int error)
2769 tcp_mtudisc(inp, -1);
2774 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2779 INP_WLOCK_ASSERT(inp);
2780 if ((inp->inp_flags & INP_TIMEWAIT) ||
2781 (inp->inp_flags & INP_DROPPED))
2784 tp = intotcpcb(inp);
2785 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2787 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2789 so = inp->inp_socket;
2790 SOCKBUF_LOCK(&so->so_snd);
2791 /* If the mss is larger than the socket buffer, decrease the mss. */
2792 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2793 tp->t_maxseg = so->so_snd.sb_hiwat;
2794 SOCKBUF_UNLOCK(&so->so_snd);
2796 TCPSTAT_INC(tcps_mturesent);
2798 tp->snd_nxt = tp->snd_una;
2799 tcp_free_sackholes(tp);
2800 tp->snd_recover = tp->snd_max;
2801 if (tp->t_flags & TF_SACK_PERMIT)
2802 EXIT_FASTRECOVERY(tp->t_flags);
2803 tp->t_fb->tfb_tcp_output(tp);
2808 * Look-up the routing entry to the peer of this inpcb. If no route
2809 * is found and it cannot be allocated, then return 0. This routine
2810 * is called by TCP routines that access the rmx structure and by
2811 * tcp_mss_update to get the peer/interface MTU.
2814 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2816 struct nhop4_extended nh4;
2818 uint32_t maxmtu = 0;
2820 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2822 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2824 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2825 NHR_REF, 0, &nh4) != 0)
2829 maxmtu = nh4.nh_mtu;
2831 /* Report additional interface capabilities. */
2833 if (ifp->if_capenable & IFCAP_TSO4 &&
2834 ifp->if_hwassist & CSUM_TSO) {
2835 cap->ifcap |= CSUM_TSO;
2836 cap->tsomax = ifp->if_hw_tsomax;
2837 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2838 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2841 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2849 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2851 struct nhop6_extended nh6;
2852 struct in6_addr dst6;
2855 uint32_t maxmtu = 0;
2857 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2859 if (inc->inc_flags & INC_IPV6MINMTU)
2862 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2863 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2864 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2869 maxmtu = nh6.nh_mtu;
2871 /* Report additional interface capabilities. */
2873 if (ifp->if_capenable & IFCAP_TSO6 &&
2874 ifp->if_hwassist & CSUM_TSO) {
2875 cap->ifcap |= CSUM_TSO;
2876 cap->tsomax = ifp->if_hw_tsomax;
2877 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2878 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2881 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2889 * Calculate effective SMSS per RFC5681 definition for a given TCP
2890 * connection at its current state, taking into account SACK and etc.
2893 tcp_maxseg(const struct tcpcb *tp)
2897 if (tp->t_flags & TF_NOOPT)
2898 return (tp->t_maxseg);
2901 * Here we have a simplified code from tcp_addoptions(),
2902 * without a proper loop, and having most of paddings hardcoded.
2903 * We might make mistakes with padding here in some edge cases,
2904 * but this is harmless, since result of tcp_maxseg() is used
2905 * only in cwnd and ssthresh estimations.
2907 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2908 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2909 if (tp->t_flags & TF_RCVD_TSTMP)
2910 optlen = TCPOLEN_TSTAMP_APPA;
2913 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2914 if (tp->t_flags & TF_SIGNATURE)
2915 optlen += PAD(TCPOLEN_SIGNATURE);
2917 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2918 optlen += TCPOLEN_SACKHDR;
2919 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2920 optlen = PAD(optlen);
2923 if (tp->t_flags & TF_REQ_TSTMP)
2924 optlen = TCPOLEN_TSTAMP_APPA;
2926 optlen = PAD(TCPOLEN_MAXSEG);
2927 if (tp->t_flags & TF_REQ_SCALE)
2928 optlen += PAD(TCPOLEN_WINDOW);
2929 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2930 if (tp->t_flags & TF_SIGNATURE)
2931 optlen += PAD(TCPOLEN_SIGNATURE);
2933 if (tp->t_flags & TF_SACK_PERMIT)
2934 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2937 optlen = min(optlen, TCP_MAXOLEN);
2938 return (tp->t_maxseg - optlen);
2942 sysctl_drop(SYSCTL_HANDLER_ARGS)
2944 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2945 struct sockaddr_storage addrs[2];
2949 struct sockaddr_in *fin, *lin;
2950 struct epoch_tracker et;
2952 struct sockaddr_in6 *fin6, *lin6;
2963 if (req->oldptr != NULL || req->oldlen != 0)
2965 if (req->newptr == NULL)
2967 if (req->newlen < sizeof(addrs))
2969 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2973 switch (addrs[0].ss_family) {
2976 fin6 = (struct sockaddr_in6 *)&addrs[0];
2977 lin6 = (struct sockaddr_in6 *)&addrs[1];
2978 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2979 lin6->sin6_len != sizeof(struct sockaddr_in6))
2981 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2982 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2984 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2985 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2986 fin = (struct sockaddr_in *)&addrs[0];
2987 lin = (struct sockaddr_in *)&addrs[1];
2990 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2993 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3000 fin = (struct sockaddr_in *)&addrs[0];
3001 lin = (struct sockaddr_in *)&addrs[1];
3002 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3003 lin->sin_len != sizeof(struct sockaddr_in))
3010 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
3011 switch (addrs[0].ss_family) {
3014 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3015 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3016 INPLOOKUP_WLOCKPCB, NULL);
3021 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3022 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3027 if (inp->inp_flags & INP_TIMEWAIT) {
3029 * XXXRW: There currently exists a state where an
3030 * inpcb is present, but its timewait state has been
3031 * discarded. For now, don't allow dropping of this
3039 } else if (!(inp->inp_flags & INP_DROPPED) &&
3040 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
3041 tp = intotcpcb(inp);
3042 tp = tcp_drop(tp, ECONNABORTED);
3049 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
3053 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3054 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
3055 0, sysctl_drop, "", "Drop TCP connection");
3058 * Generate a standardized TCP log line for use throughout the
3059 * tcp subsystem. Memory allocation is done with M_NOWAIT to
3060 * allow use in the interrupt context.
3062 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3063 * NB: The function may return NULL if memory allocation failed.
3065 * Due to header inclusion and ordering limitations the struct ip
3066 * and ip6_hdr pointers have to be passed as void pointers.
3069 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3073 /* Is logging enabled? */
3074 if (tcp_log_in_vain == 0)
3077 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3081 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3085 /* Is logging enabled? */
3086 if (tcp_log_debug == 0)
3089 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3093 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3100 const struct ip6_hdr *ip6;
3102 ip6 = (const struct ip6_hdr *)ip6hdr;
3104 ip = (struct ip *)ip4hdr;
3107 * The log line looks like this:
3108 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3110 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3111 sizeof(PRINT_TH_FLAGS) + 1 +
3113 2 * INET6_ADDRSTRLEN;
3115 2 * INET_ADDRSTRLEN;
3118 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3122 strcat(s, "TCP: [");
3125 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3126 inet_ntoa_r(inc->inc_faddr, sp);
3128 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3130 inet_ntoa_r(inc->inc_laddr, sp);
3132 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3135 ip6_sprintf(sp, &inc->inc6_faddr);
3137 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3139 ip6_sprintf(sp, &inc->inc6_laddr);
3141 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3142 } else if (ip6 && th) {
3143 ip6_sprintf(sp, &ip6->ip6_src);
3145 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3147 ip6_sprintf(sp, &ip6->ip6_dst);
3149 sprintf(sp, "]:%i", ntohs(th->th_dport));
3152 } else if (ip && th) {
3153 inet_ntoa_r(ip->ip_src, sp);
3155 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3157 inet_ntoa_r(ip->ip_dst, sp);
3159 sprintf(sp, "]:%i", ntohs(th->th_dport));
3167 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3168 if (*(s + size - 1) != '\0')
3169 panic("%s: string too long", __func__);
3174 * A subroutine which makes it easy to track TCP state changes with DTrace.
3175 * This function shouldn't be called for t_state initializations that don't
3176 * correspond to actual TCP state transitions.
3179 tcp_state_change(struct tcpcb *tp, int newstate)
3181 #if defined(KDTRACE_HOOKS)
3182 int pstate = tp->t_state;
3185 TCPSTATES_DEC(tp->t_state);
3186 TCPSTATES_INC(newstate);
3187 tp->t_state = newstate;
3188 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3192 * Create an external-format (``xtcpcb'') structure using the information in
3193 * the kernel-format tcpcb structure pointed to by tp. This is done to
3194 * reduce the spew of irrelevant information over this interface, to isolate
3195 * user code from changes in the kernel structure, and potentially to provide
3196 * information-hiding if we decide that some of this information should be
3197 * hidden from users.
3200 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3202 struct tcpcb *tp = intotcpcb(inp);
3205 bzero(xt, sizeof(*xt));
3206 if (inp->inp_flags & INP_TIMEWAIT) {
3207 xt->t_state = TCPS_TIME_WAIT;
3209 xt->t_state = tp->t_state;
3210 xt->t_logstate = tp->t_logstate;
3211 xt->t_flags = tp->t_flags;
3212 xt->t_sndzerowin = tp->t_sndzerowin;
3213 xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3214 xt->t_rcvoopack = tp->t_rcvoopack;
3216 now = getsbinuptime();
3217 #define COPYTIMER(ttt) do { \
3218 if (callout_active(&tp->t_timers->ttt)) \
3219 xt->ttt = (tp->t_timers->ttt.c_time - now) / \
3224 COPYTIMER(tt_delack);
3225 COPYTIMER(tt_rexmt);
3226 COPYTIMER(tt_persist);
3230 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3232 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3233 TCP_FUNCTION_NAME_LEN_MAX);
3235 (void)tcp_log_get_id(tp, xt->xt_logid);
3239 xt->xt_len = sizeof(struct xtcpcb);
3240 in_pcbtoxinpcb(inp, &xt->xt_inp);
3241 if (inp->inp_socket == NULL)
3242 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;