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 int t_functions_inited = 0;
261 static int tcp_fb_cnt = 0;
262 struct tcp_funchead t_functions;
263 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
266 init_tcp_functions(void)
268 if (t_functions_inited == 0) {
269 TAILQ_INIT(&t_functions);
270 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
271 t_functions_inited = 1;
275 static struct tcp_function_block *
276 find_tcp_functions_locked(struct tcp_function_set *fs)
278 struct tcp_function *f;
279 struct tcp_function_block *blk=NULL;
281 TAILQ_FOREACH(f, &t_functions, tf_next) {
282 if (strcmp(f->tf_name, fs->function_set_name) == 0) {
290 static struct tcp_function_block *
291 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
293 struct tcp_function_block *rblk=NULL;
294 struct tcp_function *f;
296 TAILQ_FOREACH(f, &t_functions, tf_next) {
297 if (f->tf_fb == blk) {
308 struct tcp_function_block *
309 find_and_ref_tcp_functions(struct tcp_function_set *fs)
311 struct tcp_function_block *blk;
313 rw_rlock(&tcp_function_lock);
314 blk = find_tcp_functions_locked(fs);
316 refcount_acquire(&blk->tfb_refcnt);
317 rw_runlock(&tcp_function_lock);
321 struct tcp_function_block *
322 find_and_ref_tcp_fb(struct tcp_function_block *blk)
324 struct tcp_function_block *rblk;
326 rw_rlock(&tcp_function_lock);
327 rblk = find_tcp_fb_locked(blk, NULL);
329 refcount_acquire(&rblk->tfb_refcnt);
330 rw_runlock(&tcp_function_lock);
334 static struct tcp_function_block *
335 find_and_ref_tcp_default_fb(void)
337 struct tcp_function_block *rblk;
339 rw_rlock(&tcp_function_lock);
340 rblk = tcp_func_set_ptr;
341 refcount_acquire(&rblk->tfb_refcnt);
342 rw_runlock(&tcp_function_lock);
347 tcp_switch_back_to_default(struct tcpcb *tp)
349 struct tcp_function_block *tfb;
351 KASSERT(tp->t_fb != &tcp_def_funcblk,
352 ("%s: called by the built-in default stack", __func__));
355 * Release the old stack. This function will either find a new one
358 if (tp->t_fb->tfb_tcp_fb_fini != NULL)
359 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
360 refcount_release(&tp->t_fb->tfb_refcnt);
363 * Now, we'll find a new function block to use.
364 * Start by trying the current user-selected
365 * default, unless this stack is the user-selected
368 tfb = find_and_ref_tcp_default_fb();
369 if (tfb == tp->t_fb) {
370 refcount_release(&tfb->tfb_refcnt);
373 /* Does the stack accept this connection? */
374 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
375 (*tfb->tfb_tcp_handoff_ok)(tp)) {
376 refcount_release(&tfb->tfb_refcnt);
379 /* Try to use that stack. */
381 /* Initialize the new stack. If it succeeds, we are done. */
383 if (tp->t_fb->tfb_tcp_fb_init == NULL ||
384 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
388 * Initialization failed. Release the reference count on
391 refcount_release(&tfb->tfb_refcnt);
395 * If that wasn't feasible, use the built-in default
396 * stack which is not allowed to reject anyone.
398 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
400 /* there always should be a default */
401 panic("Can't refer to tcp_def_funcblk");
403 if (tfb->tfb_tcp_handoff_ok != NULL) {
404 if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
405 /* The default stack cannot say no */
406 panic("Default stack rejects a new session?");
410 if (tp->t_fb->tfb_tcp_fb_init != NULL &&
411 (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
412 /* The default stack cannot fail */
413 panic("Default stack initialization failed");
418 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
421 struct tcp_function_set fs;
422 struct tcp_function_block *blk;
424 memset(&fs, 0, sizeof(fs));
425 rw_rlock(&tcp_function_lock);
426 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
429 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
430 fs.pcbcnt = blk->tfb_refcnt;
432 rw_runlock(&tcp_function_lock);
433 error = sysctl_handle_string(oidp, fs.function_set_name,
434 sizeof(fs.function_set_name), req);
436 /* Check for error or no change */
437 if (error != 0 || req->newptr == NULL)
440 rw_wlock(&tcp_function_lock);
441 blk = find_tcp_functions_locked(&fs);
443 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
447 tcp_func_set_ptr = blk;
449 rw_wunlock(&tcp_function_lock);
453 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
454 CTLTYPE_STRING | CTLFLAG_RW,
455 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
456 "Set/get the default TCP functions");
459 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
461 int error, cnt, linesz;
462 struct tcp_function *f;
468 rw_rlock(&tcp_function_lock);
469 TAILQ_FOREACH(f, &t_functions, tf_next) {
472 rw_runlock(&tcp_function_lock);
474 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
475 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
480 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
481 "Alias", "PCB count");
486 rw_rlock(&tcp_function_lock);
487 TAILQ_FOREACH(f, &t_functions, tf_next) {
488 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
489 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
490 f->tf_fb->tfb_tcp_block_name,
491 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
492 alias ? f->tf_name : "-",
493 f->tf_fb->tfb_refcnt);
494 if (linesz >= bufsz) {
502 rw_runlock(&tcp_function_lock);
504 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
505 free(buffer, M_TEMP);
509 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
510 CTLTYPE_STRING|CTLFLAG_RD,
511 NULL, 0, sysctl_net_inet_list_available, "A",
512 "list available TCP Function sets");
515 * Exports one (struct tcp_function_info) for each alias/name.
518 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
521 struct tcp_function *f;
522 struct tcp_function_info tfi;
525 * We don't allow writes.
527 if (req->newptr != NULL)
531 * Wire the old buffer so we can directly copy the functions to
532 * user space without dropping the lock.
534 if (req->oldptr != NULL) {
535 error = sysctl_wire_old_buffer(req, 0);
541 * Walk the list and copy out matching entries. If INVARIANTS
542 * is compiled in, also walk the list to verify the length of
543 * the list matches what we have recorded.
545 rw_rlock(&tcp_function_lock);
549 if (req->oldptr == NULL) {
554 TAILQ_FOREACH(f, &t_functions, tf_next) {
558 if (req->oldptr != NULL) {
559 bzero(&tfi, sizeof(tfi));
560 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
561 tfi.tfi_id = f->tf_fb->tfb_id;
562 (void)strncpy(tfi.tfi_alias, f->tf_name,
563 TCP_FUNCTION_NAME_LEN_MAX);
564 tfi.tfi_alias[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
565 (void)strncpy(tfi.tfi_name,
566 f->tf_fb->tfb_tcp_block_name,
567 TCP_FUNCTION_NAME_LEN_MAX);
568 tfi.tfi_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
569 error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
571 * Don't stop on error, as that is the
572 * mechanism we use to accumulate length
573 * information if the buffer was too short.
577 KASSERT(cnt == tcp_fb_cnt,
578 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
582 rw_runlock(&tcp_function_lock);
583 if (req->oldptr == NULL)
584 error = SYSCTL_OUT(req, NULL,
585 (cnt + 1) * sizeof(struct tcp_function_info));
590 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
591 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
592 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
593 "List TCP function block name-to-ID mappings");
596 * tfb_tcp_handoff_ok() function for the default stack.
597 * Note that we'll basically try to take all comers.
600 tcp_default_handoff_ok(struct tcpcb *tp)
607 * tfb_tcp_fb_init() function for the default stack.
609 * This handles making sure we have appropriate timers set if you are
610 * transitioning a socket that has some amount of setup done.
612 * The init() fuction from the default can *never* return non-zero i.e.
613 * it is required to always succeed since it is the stack of last resort!
616 tcp_default_fb_init(struct tcpcb *tp)
621 INP_WLOCK_ASSERT(tp->t_inpcb);
623 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
624 ("%s: connection %p in unexpected state %d", __func__, tp,
628 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
629 * know what to do for unexpected states (which includes TIME_WAIT).
631 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
635 * Make sure some kind of transmission timer is set if there is
638 so = tp->t_inpcb->inp_socket;
639 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
640 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
641 tcp_timer_active(tp, TT_PERSIST))) {
643 * If the session has established and it looks like it should
644 * be in the persist state, set the persist timer. Otherwise,
645 * set the retransmit timer.
647 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
648 (int32_t)(tp->snd_nxt - tp->snd_una) <
649 (int32_t)sbavail(&so->so_snd))
652 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
655 /* All non-embryonic sessions get a keepalive timer. */
656 if (!tcp_timer_active(tp, TT_KEEP))
657 tcp_timer_activate(tp, TT_KEEP,
658 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
665 * tfb_tcp_fb_fini() function for the default stack.
667 * This changes state as necessary (or prudent) to prepare for another stack
668 * to assume responsibility for the connection.
671 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
674 INP_WLOCK_ASSERT(tp->t_inpcb);
679 * Target size of TCP PCB hash tables. Must be a power of two.
681 * Note that this can be overridden by the kernel environment
682 * variable net.inet.tcp.tcbhashsize
685 #define TCBHASHSIZE 0
690 * Callouts should be moved into struct tcp directly. They are currently
691 * separate because the tcpcb structure is exported to userland for sysctl
692 * parsing purposes, which do not know about callouts.
703 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
704 #define V_tcpcb_zone VNET(tcpcb_zone)
706 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
707 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
709 static struct mtx isn_mtx;
711 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
712 #define ISN_LOCK() mtx_lock(&isn_mtx)
713 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
716 * TCP initialization.
719 tcp_zone_change(void *tag)
722 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
723 uma_zone_set_max(V_tcpcb_zone, maxsockets);
724 tcp_tw_zone_change();
728 tcp_inpcb_init(void *mem, int size, int flags)
730 struct inpcb *inp = mem;
732 INP_LOCK_INIT(inp, "inp", "tcpinp");
737 * Take a value and get the next power of 2 that doesn't overflow.
738 * Used to size the tcp_inpcb hash buckets.
741 maketcp_hashsize(int size)
747 * get the next power of 2 higher than maxsockets.
749 hashsize = 1 << fls(size);
750 /* catch overflow, and just go one power of 2 smaller */
751 if (hashsize < size) {
752 hashsize = 1 << (fls(size) - 1);
757 static volatile int next_tcp_stack_id = 1;
760 * Register a TCP function block with the name provided in the names
761 * array. (Note that this function does NOT automatically register
762 * blk->tfb_tcp_block_name as a stack name. Therefore, you should
763 * explicitly include blk->tfb_tcp_block_name in the list of names if
764 * you wish to register the stack with that name.)
766 * Either all name registrations will succeed or all will fail. If
767 * a name registration fails, the function will update the num_names
768 * argument to point to the array index of the name that encountered
771 * Returns 0 on success, or an error code on failure.
774 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
775 const char *names[], int *num_names)
777 struct tcp_function *n;
778 struct tcp_function_set fs;
781 KASSERT(names != NULL && *num_names > 0,
782 ("%s: Called with 0-length name list", __func__));
783 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
785 if (t_functions_inited == 0) {
786 init_tcp_functions();
788 if ((blk->tfb_tcp_output == NULL) ||
789 (blk->tfb_tcp_do_segment == NULL) ||
790 (blk->tfb_tcp_ctloutput == NULL) ||
791 (strlen(blk->tfb_tcp_block_name) == 0)) {
793 * These functions are required and you
799 if (blk->tfb_tcp_timer_stop_all ||
800 blk->tfb_tcp_timer_activate ||
801 blk->tfb_tcp_timer_active ||
802 blk->tfb_tcp_timer_stop) {
804 * If you define one timer function you
805 * must have them all.
807 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
808 (blk->tfb_tcp_timer_activate == NULL) ||
809 (blk->tfb_tcp_timer_active == NULL) ||
810 (blk->tfb_tcp_timer_stop == NULL)) {
816 refcount_init(&blk->tfb_refcnt, 0);
818 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
819 for (i = 0; i < *num_names; i++) {
820 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
827 (void)strncpy(fs.function_set_name, names[i],
828 TCP_FUNCTION_NAME_LEN_MAX);
829 fs.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
830 rw_wlock(&tcp_function_lock);
831 if (find_tcp_functions_locked(&fs) != NULL) {
832 /* Duplicate name space not allowed */
833 rw_wunlock(&tcp_function_lock);
834 free(n, M_TCPFUNCTIONS);
838 (void)strncpy(n->tf_name, names[i], TCP_FUNCTION_NAME_LEN_MAX);
839 n->tf_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
840 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
842 rw_wunlock(&tcp_function_lock);
848 * Deregister the names we just added. Because registration failed
849 * for names[i], we don't need to deregister that name.
852 rw_wlock(&tcp_function_lock);
854 TAILQ_FOREACH(n, &t_functions, tf_next) {
855 if (!strncmp(n->tf_name, names[i],
856 TCP_FUNCTION_NAME_LEN_MAX)) {
857 TAILQ_REMOVE(&t_functions, n, tf_next);
860 free(n, M_TCPFUNCTIONS);
865 rw_wunlock(&tcp_function_lock);
870 * Register a TCP function block using the name provided in the name
873 * Returns 0 on success, or an error code on failure.
876 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
879 const char *name_list[1];
886 name_list[0] = blk->tfb_tcp_block_name;
887 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
892 * Register a TCP function block using the name defined in
893 * blk->tfb_tcp_block_name.
895 * Returns 0 on success, or an error code on failure.
898 register_tcp_functions(struct tcp_function_block *blk, int wait)
901 return (register_tcp_functions_as_name(blk, NULL, wait));
905 * Deregister all names associated with a function block. This
906 * functionally removes the function block from use within the system.
908 * When called with a true quiesce argument, mark the function block
909 * as being removed so no more stacks will use it and determine
910 * whether the removal would succeed.
912 * When called with a false quiesce argument, actually attempt the
915 * When called with a force argument, attempt to switch all TCBs to
916 * use the default stack instead of returning EBUSY.
918 * Returns 0 on success (or if the removal would succeed, or an error
922 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
925 struct tcp_function *f;
927 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
928 /* You can't un-register the default */
931 rw_wlock(&tcp_function_lock);
932 if (blk == tcp_func_set_ptr) {
933 /* You can't free the current default */
934 rw_wunlock(&tcp_function_lock);
937 /* Mark the block so no more stacks can use it. */
938 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
940 * If TCBs are still attached to the stack, attempt to switch them
941 * to the default stack.
943 if (force && blk->tfb_refcnt) {
946 VNET_ITERATOR_DECL(vnet_iter);
948 rw_wunlock(&tcp_function_lock);
951 VNET_FOREACH(vnet_iter) {
952 CURVNET_SET(vnet_iter);
953 INP_INFO_WLOCK(&V_tcbinfo);
954 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
956 if (inp->inp_flags & INP_TIMEWAIT) {
961 if (tp == NULL || tp->t_fb != blk) {
965 tcp_switch_back_to_default(tp);
968 INP_INFO_WUNLOCK(&V_tcbinfo);
973 rw_wlock(&tcp_function_lock);
975 if (blk->tfb_refcnt) {
976 /* TCBs still attached. */
977 rw_wunlock(&tcp_function_lock);
982 rw_wunlock(&tcp_function_lock);
985 /* Remove any function names that map to this function block. */
986 while (find_tcp_fb_locked(blk, &f) != NULL) {
987 TAILQ_REMOVE(&t_functions, f, tf_next);
990 free(f, M_TCPFUNCTIONS);
992 rw_wunlock(&tcp_function_lock);
999 const char *tcbhash_tuneable;
1002 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1005 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1006 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1007 printf("%s: WARNING: unable to register helper hook\n", __func__);
1008 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1009 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1010 printf("%s: WARNING: unable to register helper hook\n", __func__);
1012 hashsize = TCBHASHSIZE;
1013 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1014 if (hashsize == 0) {
1016 * Auto tune the hash size based on maxsockets.
1017 * A perfect hash would have a 1:1 mapping
1018 * (hashsize = maxsockets) however it's been
1019 * suggested that O(2) average is better.
1021 hashsize = maketcp_hashsize(maxsockets / 4);
1023 * Our historical default is 512,
1024 * do not autotune lower than this.
1028 if (bootverbose && IS_DEFAULT_VNET(curvnet))
1029 printf("%s: %s auto tuned to %d\n", __func__,
1030 tcbhash_tuneable, hashsize);
1033 * We require a hashsize to be a power of two.
1034 * Previously if it was not a power of two we would just reset it
1035 * back to 512, which could be a nasty surprise if you did not notice
1036 * the error message.
1037 * Instead what we do is clip it to the closest power of two lower
1038 * than the specified hash value.
1040 if (!powerof2(hashsize)) {
1041 int oldhashsize = hashsize;
1043 hashsize = maketcp_hashsize(hashsize);
1044 /* prevent absurdly low value */
1047 printf("%s: WARNING: TCB hash size not a power of 2, "
1048 "clipped from %d to %d.\n", __func__, oldhashsize,
1051 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
1052 "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
1055 * These have to be type stable for the benefit of the timers.
1057 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1058 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1059 uma_zone_set_max(V_tcpcb_zone, maxsockets);
1060 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1066 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1067 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1068 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1070 tcp_fastopen_init();
1072 /* Skip initialization of globals for non-default instances. */
1073 if (!IS_DEFAULT_VNET(curvnet))
1076 tcp_reass_global_init();
1078 /* XXX virtualize those bellow? */
1079 tcp_delacktime = TCPTV_DELACK;
1080 tcp_keepinit = TCPTV_KEEP_INIT;
1081 tcp_keepidle = TCPTV_KEEP_IDLE;
1082 tcp_keepintvl = TCPTV_KEEPINTVL;
1083 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1084 tcp_msl = TCPTV_MSL;
1085 tcp_rexmit_min = TCPTV_MIN;
1086 if (tcp_rexmit_min < 1)
1088 tcp_persmin = TCPTV_PERSMIN;
1089 tcp_persmax = TCPTV_PERSMAX;
1090 tcp_rexmit_slop = TCPTV_CPU_VAR;
1091 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1092 tcp_tcbhashsize = hashsize;
1093 /* Setup the tcp function block list */
1094 init_tcp_functions();
1095 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1097 /* Initialize the TCP logging data. */
1100 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1102 if (tcp_soreceive_stream) {
1104 tcp_usrreqs.pru_soreceive = soreceive_stream;
1107 tcp6_usrreqs.pru_soreceive = soreceive_stream;
1112 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1114 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1116 if (max_protohdr < TCP_MINPROTOHDR)
1117 max_protohdr = TCP_MINPROTOHDR;
1118 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1120 #undef TCP_MINPROTOHDR
1123 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1124 SHUTDOWN_PRI_DEFAULT);
1125 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
1126 EVENTHANDLER_PRI_ANY);
1134 tcp_destroy(void *unused __unused)
1142 * All our processes are gone, all our sockets should be cleaned
1143 * up, which means, we should be past the tcp_discardcb() calls.
1144 * Sleep to let all tcpcb timers really disappear and cleanup.
1147 INP_LIST_RLOCK(&V_tcbinfo);
1148 n = V_tcbinfo.ipi_count;
1149 INP_LIST_RUNLOCK(&V_tcbinfo);
1152 pause("tcpdes", hz / 10);
1157 in_pcbinfo_destroy(&V_tcbinfo);
1158 /* tcp_discardcb() clears the sack_holes up. */
1159 uma_zdestroy(V_sack_hole_zone);
1160 uma_zdestroy(V_tcpcb_zone);
1163 * Cannot free the zone until all tcpcbs are released as we attach
1164 * the allocations to them.
1166 tcp_fastopen_destroy();
1169 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1171 printf("%s: WARNING: unable to deregister helper hook "
1172 "type=%d, id=%d: error %d returned\n", __func__,
1173 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1175 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1177 printf("%s: WARNING: unable to deregister helper hook "
1178 "type=%d, id=%d: error %d returned\n", __func__,
1179 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1183 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1193 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1194 * tcp_template used to store this data in mbufs, but we now recopy it out
1195 * of the tcpcb each time to conserve mbufs.
1198 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
1200 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1202 INP_WLOCK_ASSERT(inp);
1205 if ((inp->inp_vflag & INP_IPV6) != 0) {
1206 struct ip6_hdr *ip6;
1208 ip6 = (struct ip6_hdr *)ip_ptr;
1209 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1210 (inp->inp_flow & IPV6_FLOWINFO_MASK);
1211 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1212 (IPV6_VERSION & IPV6_VERSION_MASK);
1213 ip6->ip6_nxt = IPPROTO_TCP;
1214 ip6->ip6_plen = htons(sizeof(struct tcphdr));
1215 ip6->ip6_src = inp->in6p_laddr;
1216 ip6->ip6_dst = inp->in6p_faddr;
1219 #if defined(INET6) && defined(INET)
1226 ip = (struct ip *)ip_ptr;
1227 ip->ip_v = IPVERSION;
1229 ip->ip_tos = inp->inp_ip_tos;
1233 ip->ip_ttl = inp->inp_ip_ttl;
1235 ip->ip_p = IPPROTO_TCP;
1236 ip->ip_src = inp->inp_laddr;
1237 ip->ip_dst = inp->inp_faddr;
1240 th->th_sport = inp->inp_lport;
1241 th->th_dport = inp->inp_fport;
1249 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
1253 * Create template to be used to send tcp packets on a connection.
1254 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
1255 * use for this function is in keepalives, which use tcp_respond.
1258 tcpip_maketemplate(struct inpcb *inp)
1262 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1265 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1270 * Send a single message to the TCP at address specified by
1271 * the given TCP/IP header. If m == NULL, then we make a copy
1272 * of the tcpiphdr at th and send directly to the addressed host.
1273 * This is used to force keep alive messages out using the TCP
1274 * template for a connection. If flags are given then we send
1275 * a message back to the TCP which originated the segment th,
1276 * and discard the mbuf containing it and any other attached mbufs.
1278 * In any case the ack and sequence number of the transmitted
1279 * segment are as specified by the parameters.
1281 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1284 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1285 tcp_seq ack, tcp_seq seq, int flags)
1294 struct ip6_hdr *ip6;
1297 int optlen, tlen, win;
1300 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1303 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1310 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1311 INP_WLOCK_ASSERT(inp);
1318 if (!(flags & TH_RST)) {
1319 win = sbspace(&inp->inp_socket->so_rcv);
1320 if (win > TCP_MAXWIN << tp->rcv_scale)
1321 win = TCP_MAXWIN << tp->rcv_scale;
1323 if ((tp->t_flags & TF_NOOPT) == 0)
1327 m = m_gethdr(M_NOWAIT, MT_DATA);
1330 m->m_data += max_linkhdr;
1333 bcopy((caddr_t)ip6, mtod(m, caddr_t),
1334 sizeof(struct ip6_hdr));
1335 ip6 = mtod(m, struct ip6_hdr *);
1336 nth = (struct tcphdr *)(ip6 + 1);
1340 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1341 ip = mtod(m, struct ip *);
1342 nth = (struct tcphdr *)(ip + 1);
1344 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1346 } else if (!M_WRITABLE(m)) {
1349 /* Can't reuse 'm', allocate a new mbuf. */
1350 n = m_gethdr(M_NOWAIT, MT_DATA);
1356 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1362 n->m_data += max_linkhdr;
1363 /* m_len is set later */
1364 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1367 bcopy((caddr_t)ip6, mtod(n, caddr_t),
1368 sizeof(struct ip6_hdr));
1369 ip6 = mtod(n, struct ip6_hdr *);
1370 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1371 nth = (struct tcphdr *)(ip6 + 1);
1375 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1376 ip = mtod(n, struct ip *);
1377 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1378 nth = (struct tcphdr *)(ip + 1);
1380 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1381 xchg(nth->th_dport, nth->th_sport, uint16_t);
1388 * XXX MRT We inherit the FIB, which is lucky.
1392 m->m_data = (caddr_t)ipgen;
1393 /* m_len is set later */
1396 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1397 nth = (struct tcphdr *)(ip6 + 1);
1401 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1402 nth = (struct tcphdr *)(ip + 1);
1406 * this is usually a case when an extension header
1407 * exists between the IPv6 header and the
1410 nth->th_sport = th->th_sport;
1411 nth->th_dport = th->th_dport;
1413 xchg(nth->th_dport, nth->th_sport, uint16_t);
1419 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1421 #if defined(INET) && defined(INET6)
1425 tlen = sizeof (struct tcpiphdr);
1429 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1430 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1431 m, tlen, (long)M_TRAILINGSPACE(m)));
1436 /* Make sure we have room. */
1437 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1438 m->m_next = m_get(M_NOWAIT, MT_DATA);
1440 optp = mtod(m->m_next, u_char *);
1445 optp = (u_char *) (nth + 1);
1451 if (tp->t_flags & TF_RCVD_TSTMP) {
1452 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1453 to.to_tsecr = tp->ts_recent;
1454 to.to_flags |= TOF_TS;
1456 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1457 /* TCP-MD5 (RFC2385). */
1458 if (tp->t_flags & TF_SIGNATURE)
1459 to.to_flags |= TOF_SIGNATURE;
1461 /* Add the options. */
1462 tlen += optlen = tcp_addoptions(&to, optp);
1464 /* Update m_len in the correct mbuf. */
1465 optm->m_len += optlen;
1471 ip6->ip6_vfc = IPV6_VERSION;
1472 ip6->ip6_nxt = IPPROTO_TCP;
1473 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1476 #if defined(INET) && defined(INET6)
1481 ip->ip_len = htons(tlen);
1482 ip->ip_ttl = V_ip_defttl;
1483 if (V_path_mtu_discovery)
1484 ip->ip_off |= htons(IP_DF);
1487 m->m_pkthdr.len = tlen;
1488 m->m_pkthdr.rcvif = NULL;
1492 * Packet is associated with a socket, so allow the
1493 * label of the response to reflect the socket label.
1495 INP_WLOCK_ASSERT(inp);
1496 mac_inpcb_create_mbuf(inp, m);
1499 * Packet is not associated with a socket, so possibly
1500 * update the label in place.
1502 mac_netinet_tcp_reply(m);
1505 nth->th_seq = htonl(seq);
1506 nth->th_ack = htonl(ack);
1508 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1509 nth->th_flags = flags;
1511 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1513 nth->th_win = htons((u_short)win);
1516 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1517 if (to.to_flags & TOF_SIGNATURE) {
1518 if (!TCPMD5_ENABLED() ||
1519 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1526 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1529 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1530 nth->th_sum = in6_cksum_pseudo(ip6,
1531 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1532 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1536 #if defined(INET6) && defined(INET)
1541 m->m_pkthdr.csum_flags = CSUM_TCP;
1542 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1543 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1547 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1548 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1550 TCP_PROBE3(debug__output, tp, th, m);
1552 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1556 TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1557 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1560 #if defined(INET) && defined(INET6)
1565 TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1566 (void)ip_output(m, NULL, NULL, 0, NULL, inp);
1572 * Create a new TCP control block, making an
1573 * empty reassembly queue and hooking it to the argument
1574 * protocol control block. The `inp' parameter must have
1575 * come from the zone allocator set up in tcp_init().
1578 tcp_newtcpcb(struct inpcb *inp)
1580 struct tcpcb_mem *tm;
1583 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1586 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1591 /* Initialise cc_var struct for this tcpcb. */
1593 tp->ccv->type = IPPROTO_TCP;
1594 tp->ccv->ccvc.tcp = tp;
1595 rw_rlock(&tcp_function_lock);
1596 tp->t_fb = tcp_func_set_ptr;
1597 refcount_acquire(&tp->t_fb->tfb_refcnt);
1598 rw_runlock(&tcp_function_lock);
1600 * Use the current system default CC algorithm.
1603 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1604 CC_ALGO(tp) = CC_DEFAULT();
1607 if (CC_ALGO(tp)->cb_init != NULL)
1608 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1609 if (tp->t_fb->tfb_tcp_fb_fini)
1610 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1611 refcount_release(&tp->t_fb->tfb_refcnt);
1612 uma_zfree(V_tcpcb_zone, tm);
1618 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1619 if (tp->t_fb->tfb_tcp_fb_fini)
1620 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1621 refcount_release(&tp->t_fb->tfb_refcnt);
1622 uma_zfree(V_tcpcb_zone, tm);
1628 tp->t_vnet = inp->inp_vnet;
1630 tp->t_timers = &tm->tt;
1631 TAILQ_INIT(&tp->t_segq);
1634 isipv6 ? V_tcp_v6mssdflt :
1638 /* Set up our timeouts. */
1639 callout_init(&tp->t_timers->tt_rexmt, 1);
1640 callout_init(&tp->t_timers->tt_persist, 1);
1641 callout_init(&tp->t_timers->tt_keep, 1);
1642 callout_init(&tp->t_timers->tt_2msl, 1);
1643 callout_init(&tp->t_timers->tt_delack, 1);
1645 if (V_tcp_do_rfc1323)
1646 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1648 tp->t_flags |= TF_SACK_PERMIT;
1649 TAILQ_INIT(&tp->snd_holes);
1651 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1654 in_pcbref(inp); /* Reference for tcpcb */
1658 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1659 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1660 * reasonable initial retransmit time.
1662 tp->t_srtt = TCPTV_SRTTBASE;
1663 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1664 tp->t_rttmin = tcp_rexmit_min;
1665 tp->t_rxtcur = TCPTV_RTOBASE;
1666 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1667 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1668 tp->t_rcvtime = ticks;
1670 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1671 * because the socket may be bound to an IPv6 wildcard address,
1672 * which may match an IPv4-mapped IPv6 address.
1674 inp->inp_ip_ttl = V_ip_defttl;
1678 * Init the TCP PCAP queues.
1680 tcp_pcap_tcpcb_init(tp);
1683 /* Initialize the per-TCPCB log data. */
1684 tcp_log_tcpcbinit(tp);
1686 if (tp->t_fb->tfb_tcp_fb_init) {
1687 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1689 return (tp); /* XXX */
1693 * Switch the congestion control algorithm back to NewReno for any active
1694 * control blocks using an algorithm which is about to go away.
1695 * This ensures the CC framework can allow the unload to proceed without leaving
1696 * any dangling pointers which would trigger a panic.
1697 * Returning non-zero would inform the CC framework that something went wrong
1698 * and it would be unsafe to allow the unload to proceed. However, there is no
1699 * way for this to occur with this implementation so we always return zero.
1702 tcp_ccalgounload(struct cc_algo *unload_algo)
1704 struct cc_algo *tmpalgo;
1707 VNET_ITERATOR_DECL(vnet_iter);
1710 * Check all active control blocks across all network stacks and change
1711 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1712 * requires cleanup code to be run, call it.
1715 VNET_FOREACH(vnet_iter) {
1716 CURVNET_SET(vnet_iter);
1717 INP_INFO_WLOCK(&V_tcbinfo);
1719 * New connections already part way through being initialised
1720 * with the CC algo we're removing will not race with this code
1721 * because the INP_INFO_WLOCK is held during initialisation. We
1722 * therefore don't enter the loop below until the connection
1723 * list has stabilised.
1725 CK_LIST_FOREACH(inp, &V_tcb, inp_list) {
1727 /* Important to skip tcptw structs. */
1728 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1729 (tp = intotcpcb(inp)) != NULL) {
1731 * By holding INP_WLOCK here, we are assured
1732 * that the connection is not currently
1733 * executing inside the CC module's functions
1734 * i.e. it is safe to make the switch back to
1737 if (CC_ALGO(tp) == unload_algo) {
1738 tmpalgo = CC_ALGO(tp);
1739 if (tmpalgo->cb_destroy != NULL)
1740 tmpalgo->cb_destroy(tp->ccv);
1743 * NewReno may allocate memory on
1744 * demand for certain stateful
1745 * configuration as needed, but is
1746 * coded to never fail on memory
1747 * allocation failure so it is a safe
1750 CC_ALGO(tp) = &newreno_cc_algo;
1755 INP_INFO_WUNLOCK(&V_tcbinfo);
1758 VNET_LIST_RUNLOCK();
1764 * Drop a TCP connection, reporting
1765 * the specified error. If connection is synchronized,
1766 * then send a RST to peer.
1769 tcp_drop(struct tcpcb *tp, int errno)
1771 struct socket *so = tp->t_inpcb->inp_socket;
1773 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1774 INP_WLOCK_ASSERT(tp->t_inpcb);
1776 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1777 tcp_state_change(tp, TCPS_CLOSED);
1778 (void) tp->t_fb->tfb_tcp_output(tp);
1779 TCPSTAT_INC(tcps_drops);
1781 TCPSTAT_INC(tcps_conndrops);
1782 if (errno == ETIMEDOUT && tp->t_softerror)
1783 errno = tp->t_softerror;
1784 so->so_error = errno;
1785 return (tcp_close(tp));
1789 tcp_discardcb(struct tcpcb *tp)
1791 struct inpcb *inp = tp->t_inpcb;
1792 struct socket *so = inp->inp_socket;
1794 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1796 int released __unused;
1798 INP_WLOCK_ASSERT(inp);
1801 * Make sure that all of our timers are stopped before we delete the
1804 * If stopping a timer fails, we schedule a discard function in same
1805 * callout, and the last discard function called will take care of
1806 * deleting the tcpcb.
1808 tp->t_timers->tt_draincnt = 0;
1809 tcp_timer_stop(tp, TT_REXMT);
1810 tcp_timer_stop(tp, TT_PERSIST);
1811 tcp_timer_stop(tp, TT_KEEP);
1812 tcp_timer_stop(tp, TT_2MSL);
1813 tcp_timer_stop(tp, TT_DELACK);
1814 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1816 * Call the stop-all function of the methods,
1817 * this function should call the tcp_timer_stop()
1818 * method with each of the function specific timeouts.
1819 * That stop will be called via the tfb_tcp_timer_stop()
1820 * which should use the async drain function of the
1821 * callout system (see tcp_var.h).
1823 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1827 * If we got enough samples through the srtt filter,
1828 * save the rtt and rttvar in the routing entry.
1829 * 'Enough' is arbitrarily defined as 4 rtt samples.
1830 * 4 samples is enough for the srtt filter to converge
1831 * to within enough % of the correct value; fewer samples
1832 * and we could save a bogus rtt. The danger is not high
1833 * as tcp quickly recovers from everything.
1834 * XXX: Works very well but needs some more statistics!
1836 if (tp->t_rttupdated >= 4) {
1837 struct hc_metrics_lite metrics;
1840 bzero(&metrics, sizeof(metrics));
1842 * Update the ssthresh always when the conditions below
1843 * are satisfied. This gives us better new start value
1844 * for the congestion avoidance for new connections.
1845 * ssthresh is only set if packet loss occurred on a session.
1847 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1848 * being torn down. Ideally this code would not use 'so'.
1850 ssthresh = tp->snd_ssthresh;
1851 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1853 * convert the limit from user data bytes to
1854 * packets then to packet data bytes.
1856 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1859 ssthresh *= (tp->t_maxseg +
1861 (isipv6 ? sizeof (struct ip6_hdr) +
1862 sizeof (struct tcphdr) :
1864 sizeof (struct tcpiphdr)
1871 metrics.rmx_ssthresh = ssthresh;
1873 metrics.rmx_rtt = tp->t_srtt;
1874 metrics.rmx_rttvar = tp->t_rttvar;
1875 metrics.rmx_cwnd = tp->snd_cwnd;
1876 metrics.rmx_sendpipe = 0;
1877 metrics.rmx_recvpipe = 0;
1879 tcp_hc_update(&inp->inp_inc, &metrics);
1882 /* free the reassembly queue, if any */
1883 tcp_reass_flush(tp);
1886 /* Disconnect offload device, if any. */
1887 if (tp->t_flags & TF_TOE)
1888 tcp_offload_detach(tp);
1891 tcp_free_sackholes(tp);
1894 /* Free the TCP PCAP queues. */
1895 tcp_pcap_drain(&(tp->t_inpkts));
1896 tcp_pcap_drain(&(tp->t_outpkts));
1899 /* Allow the CC algorithm to clean up after itself. */
1900 if (CC_ALGO(tp)->cb_destroy != NULL)
1901 CC_ALGO(tp)->cb_destroy(tp->ccv);
1905 khelp_destroy_osd(tp->osd);
1909 inp->inp_ppcb = NULL;
1910 if (tp->t_timers->tt_draincnt == 0) {
1911 /* We own the last reference on tcpcb, let's free it. */
1913 tcp_log_tcpcbfini(tp);
1915 TCPSTATES_DEC(tp->t_state);
1916 if (tp->t_fb->tfb_tcp_fb_fini)
1917 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1918 refcount_release(&tp->t_fb->tfb_refcnt);
1920 uma_zfree(V_tcpcb_zone, tp);
1921 released = in_pcbrele_wlocked(inp);
1922 KASSERT(!released, ("%s: inp %p should not have been released "
1923 "here", __func__, inp));
1928 tcp_timer_discard(void *ptp)
1932 struct epoch_tracker et;
1934 tp = (struct tcpcb *)ptp;
1935 CURVNET_SET(tp->t_vnet);
1936 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
1938 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1941 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1942 ("%s: tcpcb has to be stopped here", __func__));
1943 tp->t_timers->tt_draincnt--;
1944 if (tp->t_timers->tt_draincnt == 0) {
1945 /* We own the last reference on this tcpcb, let's free it. */
1947 tcp_log_tcpcbfini(tp);
1949 TCPSTATES_DEC(tp->t_state);
1950 if (tp->t_fb->tfb_tcp_fb_fini)
1951 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1952 refcount_release(&tp->t_fb->tfb_refcnt);
1954 uma_zfree(V_tcpcb_zone, tp);
1955 if (in_pcbrele_wlocked(inp)) {
1956 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1962 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1967 * Attempt to close a TCP control block, marking it as dropped, and freeing
1968 * the socket if we hold the only reference.
1971 tcp_close(struct tcpcb *tp)
1973 struct inpcb *inp = tp->t_inpcb;
1976 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1977 INP_WLOCK_ASSERT(inp);
1980 if (tp->t_state == TCPS_LISTEN)
1981 tcp_offload_listen_stop(tp);
1984 * This releases the TFO pending counter resource for TFO listen
1985 * sockets as well as passively-created TFO sockets that transition
1986 * from SYN_RECEIVED to CLOSED.
1988 if (tp->t_tfo_pending) {
1989 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1990 tp->t_tfo_pending = NULL;
1993 TCPSTAT_INC(tcps_closed);
1994 if (tp->t_state != TCPS_CLOSED)
1995 tcp_state_change(tp, TCPS_CLOSED);
1996 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1997 so = inp->inp_socket;
1998 soisdisconnected(so);
1999 if (inp->inp_flags & INP_SOCKREF) {
2000 KASSERT(so->so_state & SS_PROTOREF,
2001 ("tcp_close: !SS_PROTOREF"));
2002 inp->inp_flags &= ~INP_SOCKREF;
2005 so->so_state &= ~SS_PROTOREF;
2015 VNET_ITERATOR_DECL(vnet_iter);
2020 VNET_LIST_RLOCK_NOSLEEP();
2021 VNET_FOREACH(vnet_iter) {
2022 CURVNET_SET(vnet_iter);
2027 * Walk the tcpbs, if existing, and flush the reassembly queue,
2028 * if there is one...
2029 * XXX: The "Net/3" implementation doesn't imply that the TCP
2030 * reassembly queue should be flushed, but in a situation
2031 * where we're really low on mbufs, this is potentially
2034 INP_INFO_WLOCK(&V_tcbinfo);
2035 CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
2037 if (inpb->inp_flags & INP_TIMEWAIT) {
2041 if ((tcpb = intotcpcb(inpb)) != NULL) {
2042 tcp_reass_flush(tcpb);
2043 tcp_clean_sackreport(tcpb);
2045 tcp_log_drain(tcpb);
2048 if (tcp_pcap_aggressive_free) {
2049 /* Free the TCP PCAP queues. */
2050 tcp_pcap_drain(&(tcpb->t_inpkts));
2051 tcp_pcap_drain(&(tcpb->t_outpkts));
2057 INP_INFO_WUNLOCK(&V_tcbinfo);
2060 VNET_LIST_RUNLOCK_NOSLEEP();
2064 * Notify a tcp user of an asynchronous error;
2065 * store error as soft error, but wake up user
2066 * (for now, won't do anything until can select for soft error).
2068 * Do not wake up user since there currently is no mechanism for
2069 * reporting soft errors (yet - a kqueue filter may be added).
2071 static struct inpcb *
2072 tcp_notify(struct inpcb *inp, int error)
2076 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2077 INP_WLOCK_ASSERT(inp);
2079 if ((inp->inp_flags & INP_TIMEWAIT) ||
2080 (inp->inp_flags & INP_DROPPED))
2083 tp = intotcpcb(inp);
2084 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2087 * Ignore some errors if we are hooked up.
2088 * If connection hasn't completed, has retransmitted several times,
2089 * and receives a second error, give up now. This is better
2090 * than waiting a long time to establish a connection that
2091 * can never complete.
2093 if (tp->t_state == TCPS_ESTABLISHED &&
2094 (error == EHOSTUNREACH || error == ENETUNREACH ||
2095 error == EHOSTDOWN)) {
2096 if (inp->inp_route.ro_rt) {
2097 RTFREE(inp->inp_route.ro_rt);
2098 inp->inp_route.ro_rt = (struct rtentry *)NULL;
2101 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2103 tp = tcp_drop(tp, error);
2109 tp->t_softerror = error;
2113 wakeup( &so->so_timeo);
2120 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2122 int error, i, m, n, pcb_count;
2123 struct inpcb *inp, **inp_list;
2126 struct epoch_tracker et;
2129 * The process of preparing the TCB list is too time-consuming and
2130 * resource-intensive to repeat twice on every request.
2132 if (req->oldptr == NULL) {
2133 n = V_tcbinfo.ipi_count +
2134 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2135 n += imax(n / 8, 10);
2136 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2140 if (req->newptr != NULL)
2144 * OK, now we're committed to doing something.
2146 INP_LIST_RLOCK(&V_tcbinfo);
2147 gencnt = V_tcbinfo.ipi_gencnt;
2148 n = V_tcbinfo.ipi_count;
2149 INP_LIST_RUNLOCK(&V_tcbinfo);
2151 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2153 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
2154 + (n + m) * sizeof(struct xtcpcb));
2158 bzero(&xig, sizeof(xig));
2159 xig.xig_len = sizeof xig;
2160 xig.xig_count = n + m;
2161 xig.xig_gen = gencnt;
2162 xig.xig_sogen = so_gencnt;
2163 error = SYSCTL_OUT(req, &xig, sizeof xig);
2167 error = syncache_pcblist(req, m, &pcb_count);
2171 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
2173 INP_INFO_WLOCK(&V_tcbinfo);
2174 for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
2175 inp != NULL && i < n; inp = CK_LIST_NEXT(inp, inp_list)) {
2177 if (inp->inp_gencnt <= gencnt) {
2179 * XXX: This use of cr_cansee(), introduced with
2180 * TCP state changes, is not quite right, but for
2181 * now, better than nothing.
2183 if (inp->inp_flags & INP_TIMEWAIT) {
2184 if (intotw(inp) != NULL)
2185 error = cr_cansee(req->td->td_ucred,
2186 intotw(inp)->tw_cred);
2188 error = EINVAL; /* Skip this inp. */
2190 error = cr_canseeinpcb(req->td->td_ucred, inp);
2193 inp_list[i++] = inp;
2198 INP_INFO_WUNLOCK(&V_tcbinfo);
2202 for (i = 0; i < n; i++) {
2205 if (inp->inp_gencnt <= gencnt) {
2208 tcp_inptoxtp(inp, &xt);
2210 error = SYSCTL_OUT(req, &xt, sizeof xt);
2214 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2215 for (i = 0; i < n; i++) {
2218 if (!in_pcbrele_rlocked(inp))
2221 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2225 * Give the user an updated idea of our state.
2226 * If the generation differs from what we told
2227 * her before, she knows that something happened
2228 * while we were processing this request, and it
2229 * might be necessary to retry.
2231 INP_LIST_RLOCK(&V_tcbinfo);
2232 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2233 xig.xig_sogen = so_gencnt;
2234 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
2235 INP_LIST_RUNLOCK(&V_tcbinfo);
2236 error = SYSCTL_OUT(req, &xig, sizeof xig);
2238 free(inp_list, M_TEMP);
2242 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2243 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
2244 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
2248 tcp_getcred(SYSCTL_HANDLER_ARGS)
2251 struct sockaddr_in addrs[2];
2255 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2258 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2261 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2262 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2264 if (inp->inp_socket == NULL)
2267 error = cr_canseeinpcb(req->td->td_ucred, inp);
2269 cru2x(inp->inp_cred, &xuc);
2274 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2278 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2279 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2280 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
2285 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2288 struct sockaddr_in6 addrs[2];
2295 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2298 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2301 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2302 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2305 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2307 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2316 inp = in_pcblookup(&V_tcbinfo,
2317 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2319 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2320 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2323 inp = in6_pcblookup(&V_tcbinfo,
2324 &addrs[1].sin6_addr, addrs[1].sin6_port,
2325 &addrs[0].sin6_addr, addrs[0].sin6_port,
2326 INPLOOKUP_RLOCKPCB, NULL);
2328 if (inp->inp_socket == NULL)
2331 error = cr_canseeinpcb(req->td->td_ucred, inp);
2333 cru2x(inp->inp_cred, &xuc);
2338 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2342 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2343 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2344 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
2350 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2352 struct ip *ip = vip;
2354 struct in_addr faddr;
2357 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2359 struct in_conninfo inc;
2360 struct epoch_tracker et;
2361 tcp_seq icmp_tcp_seq;
2364 faddr = ((struct sockaddr_in *)sa)->sin_addr;
2365 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2368 if (cmd == PRC_MSGSIZE)
2369 notify = tcp_mtudisc_notify;
2370 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2371 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2372 cmd == PRC_TIMXCEED_INTRANS) && ip)
2373 notify = tcp_drop_syn_sent;
2376 * Hostdead is ugly because it goes linearly through all PCBs.
2377 * XXX: We never get this from ICMP, otherwise it makes an
2378 * excellent DoS attack on machines with many connections.
2380 else if (cmd == PRC_HOSTDEAD)
2382 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2386 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2390 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2391 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2392 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2393 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2394 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2395 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2396 /* signal EHOSTDOWN, as it flushes the cached route */
2397 inp = (*notify)(inp, EHOSTDOWN);
2400 icmp_tcp_seq = th->th_seq;
2402 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2403 !(inp->inp_flags & INP_DROPPED) &&
2404 !(inp->inp_socket == NULL)) {
2405 tp = intotcpcb(inp);
2406 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2407 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2408 if (cmd == PRC_MSGSIZE) {
2411 * If we got a needfrag set the MTU
2412 * in the route to the suggested new
2413 * value (if given) and then notify.
2415 mtu = ntohs(icp->icmp_nextmtu);
2417 * If no alternative MTU was
2418 * proposed, try the next smaller
2423 ntohs(ip->ip_len), 1);
2424 if (mtu < V_tcp_minmss +
2425 sizeof(struct tcpiphdr))
2426 mtu = V_tcp_minmss +
2427 sizeof(struct tcpiphdr);
2429 * Only process the offered MTU if it
2430 * is smaller than the current one.
2432 if (mtu < tp->t_maxseg +
2433 sizeof(struct tcpiphdr)) {
2434 bzero(&inc, sizeof(inc));
2435 inc.inc_faddr = faddr;
2437 inp->inp_inc.inc_fibnum;
2438 tcp_hc_updatemtu(&inc, mtu);
2439 tcp_mtudisc(inp, mtu);
2442 inp = (*notify)(inp,
2443 inetctlerrmap[cmd]);
2447 bzero(&inc, sizeof(inc));
2448 inc.inc_fport = th->th_dport;
2449 inc.inc_lport = th->th_sport;
2450 inc.inc_faddr = faddr;
2451 inc.inc_laddr = ip->ip_src;
2452 syncache_unreach(&inc, icmp_tcp_seq);
2457 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2463 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2465 struct in6_addr *dst;
2466 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2467 struct ip6_hdr *ip6;
2471 struct icmp6_hdr *icmp6;
2472 struct ip6ctlparam *ip6cp = NULL;
2473 const struct sockaddr_in6 *sa6_src = NULL;
2474 struct in_conninfo inc;
2475 struct epoch_tracker et;
2480 tcp_seq icmp_tcp_seq;
2484 if (sa->sa_family != AF_INET6 ||
2485 sa->sa_len != sizeof(struct sockaddr_in6))
2488 /* if the parameter is from icmp6, decode it. */
2490 ip6cp = (struct ip6ctlparam *)d;
2491 icmp6 = ip6cp->ip6c_icmp6;
2493 ip6 = ip6cp->ip6c_ip6;
2494 off = ip6cp->ip6c_off;
2495 sa6_src = ip6cp->ip6c_src;
2496 dst = ip6cp->ip6c_finaldst;
2500 off = 0; /* fool gcc */
2505 if (cmd == PRC_MSGSIZE)
2506 notify = tcp_mtudisc_notify;
2507 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2508 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2509 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2510 notify = tcp_drop_syn_sent;
2513 * Hostdead is ugly because it goes linearly through all PCBs.
2514 * XXX: We never get this from ICMP, otherwise it makes an
2515 * excellent DoS attack on machines with many connections.
2517 else if (cmd == PRC_HOSTDEAD)
2519 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2523 in6_pcbnotify(&V_tcbinfo, sa, 0,
2524 (const struct sockaddr *)sa6_src,
2525 0, cmd, NULL, notify);
2529 /* Check if we can safely get the ports from the tcp hdr */
2532 (int32_t) (off + sizeof(struct tcp_ports)))) {
2535 bzero(&t_ports, sizeof(struct tcp_ports));
2536 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2537 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2538 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2539 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2540 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2541 /* signal EHOSTDOWN, as it flushes the cached route */
2542 inp = (*notify)(inp, EHOSTDOWN);
2545 off += sizeof(struct tcp_ports);
2546 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2549 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2551 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2552 !(inp->inp_flags & INP_DROPPED) &&
2553 !(inp->inp_socket == NULL)) {
2554 tp = intotcpcb(inp);
2555 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2556 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2557 if (cmd == PRC_MSGSIZE) {
2560 * If we got a needfrag set the MTU
2561 * in the route to the suggested new
2562 * value (if given) and then notify.
2564 mtu = ntohl(icmp6->icmp6_mtu);
2566 * If no alternative MTU was
2567 * proposed, or the proposed
2568 * MTU was too small, set to
2571 if (mtu < IPV6_MMTU)
2572 mtu = IPV6_MMTU - 8;
2573 bzero(&inc, sizeof(inc));
2574 inc.inc_fibnum = M_GETFIB(m);
2575 inc.inc_flags |= INC_ISIPV6;
2576 inc.inc6_faddr = *dst;
2577 if (in6_setscope(&inc.inc6_faddr,
2578 m->m_pkthdr.rcvif, NULL))
2581 * Only process the offered MTU if it
2582 * is smaller than the current one.
2584 if (mtu < tp->t_maxseg +
2585 sizeof (struct tcphdr) +
2586 sizeof (struct ip6_hdr)) {
2587 tcp_hc_updatemtu(&inc, mtu);
2588 tcp_mtudisc(inp, mtu);
2589 ICMP6STAT_INC(icp6s_pmtuchg);
2592 inp = (*notify)(inp,
2593 inet6ctlerrmap[cmd]);
2597 bzero(&inc, sizeof(inc));
2598 inc.inc_fibnum = M_GETFIB(m);
2599 inc.inc_flags |= INC_ISIPV6;
2600 inc.inc_fport = t_ports.th_dport;
2601 inc.inc_lport = t_ports.th_sport;
2602 inc.inc6_faddr = *dst;
2603 inc.inc6_laddr = ip6->ip6_src;
2604 syncache_unreach(&inc, icmp_tcp_seq);
2609 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2614 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
2620 MD5Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
2621 MD5Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
2622 switch (inc->inc_flags & INC_ISIPV6) {
2625 MD5Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
2626 MD5Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
2631 MD5Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
2632 MD5Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
2636 MD5Update(&ctx, key, len);
2637 MD5Final((unsigned char *)hash, &ctx);
2643 tcp_new_ts_offset(struct in_conninfo *inc)
2645 return (tcp_keyed_hash(inc, V_ts_offset_secret,
2646 sizeof(V_ts_offset_secret)));
2650 * Following is where TCP initial sequence number generation occurs.
2652 * There are two places where we must use initial sequence numbers:
2653 * 1. In SYN-ACK packets.
2654 * 2. In SYN packets.
2656 * All ISNs for SYN-ACK packets are generated by the syncache. See
2657 * tcp_syncache.c for details.
2659 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2660 * depends on this property. In addition, these ISNs should be
2661 * unguessable so as to prevent connection hijacking. To satisfy
2662 * the requirements of this situation, the algorithm outlined in
2663 * RFC 1948 is used, with only small modifications.
2665 * Implementation details:
2667 * Time is based off the system timer, and is corrected so that it
2668 * increases by one megabyte per second. This allows for proper
2669 * recycling on high speed LANs while still leaving over an hour
2672 * As reading the *exact* system time is too expensive to be done
2673 * whenever setting up a TCP connection, we increment the time
2674 * offset in two ways. First, a small random positive increment
2675 * is added to isn_offset for each connection that is set up.
2676 * Second, the function tcp_isn_tick fires once per clock tick
2677 * and increments isn_offset as necessary so that sequence numbers
2678 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2679 * random positive increments serve only to ensure that the same
2680 * exact sequence number is never sent out twice (as could otherwise
2681 * happen when a port is recycled in less than the system tick
2684 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2685 * between seeding of isn_secret. This is normally set to zero,
2686 * as reseeding should not be necessary.
2688 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2689 * isn_offset_old, and isn_ctx is performed using the ISN lock. In
2690 * general, this means holding an exclusive (write) lock.
2693 #define ISN_BYTES_PER_SECOND 1048576
2694 #define ISN_STATIC_INCREMENT 4096
2695 #define ISN_RANDOM_INCREMENT (4096 - 1)
2696 #define ISN_SECRET_LENGTH 32
2698 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
2699 VNET_DEFINE_STATIC(int, isn_last);
2700 VNET_DEFINE_STATIC(int, isn_last_reseed);
2701 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
2702 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
2704 #define V_isn_secret VNET(isn_secret)
2705 #define V_isn_last VNET(isn_last)
2706 #define V_isn_last_reseed VNET(isn_last_reseed)
2707 #define V_isn_offset VNET(isn_offset)
2708 #define V_isn_offset_old VNET(isn_offset_old)
2711 tcp_new_isn(struct in_conninfo *inc)
2714 u_int32_t projected_offset;
2717 /* Seed if this is the first use, reseed if requested. */
2718 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2719 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2721 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
2722 V_isn_last_reseed = ticks;
2725 /* Compute the md5 hash and return the ISN. */
2726 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
2727 sizeof(V_isn_secret));
2728 V_isn_offset += ISN_STATIC_INCREMENT +
2729 (arc4random() & ISN_RANDOM_INCREMENT);
2730 if (ticks != V_isn_last) {
2731 projected_offset = V_isn_offset_old +
2732 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2733 if (SEQ_GT(projected_offset, V_isn_offset))
2734 V_isn_offset = projected_offset;
2735 V_isn_offset_old = V_isn_offset;
2738 new_isn += V_isn_offset;
2744 * When a specific ICMP unreachable message is received and the
2745 * connection state is SYN-SENT, drop the connection. This behavior
2746 * is controlled by the icmp_may_rst sysctl.
2749 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2753 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2754 INP_WLOCK_ASSERT(inp);
2756 if ((inp->inp_flags & INP_TIMEWAIT) ||
2757 (inp->inp_flags & INP_DROPPED))
2760 tp = intotcpcb(inp);
2761 if (tp->t_state != TCPS_SYN_SENT)
2764 if (IS_FASTOPEN(tp->t_flags))
2765 tcp_fastopen_disable_path(tp);
2767 tp = tcp_drop(tp, errno);
2775 * When `need fragmentation' ICMP is received, update our idea of the MSS
2776 * based on the new value. Also nudge TCP to send something, since we
2777 * know the packet we just sent was dropped.
2778 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2780 static struct inpcb *
2781 tcp_mtudisc_notify(struct inpcb *inp, int error)
2784 tcp_mtudisc(inp, -1);
2789 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2794 INP_WLOCK_ASSERT(inp);
2795 if ((inp->inp_flags & INP_TIMEWAIT) ||
2796 (inp->inp_flags & INP_DROPPED))
2799 tp = intotcpcb(inp);
2800 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2802 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2804 so = inp->inp_socket;
2805 SOCKBUF_LOCK(&so->so_snd);
2806 /* If the mss is larger than the socket buffer, decrease the mss. */
2807 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2808 tp->t_maxseg = so->so_snd.sb_hiwat;
2809 SOCKBUF_UNLOCK(&so->so_snd);
2811 TCPSTAT_INC(tcps_mturesent);
2813 tp->snd_nxt = tp->snd_una;
2814 tcp_free_sackholes(tp);
2815 tp->snd_recover = tp->snd_max;
2816 if (tp->t_flags & TF_SACK_PERMIT)
2817 EXIT_FASTRECOVERY(tp->t_flags);
2818 tp->t_fb->tfb_tcp_output(tp);
2823 * Look-up the routing entry to the peer of this inpcb. If no route
2824 * is found and it cannot be allocated, then return 0. This routine
2825 * is called by TCP routines that access the rmx structure and by
2826 * tcp_mss_update to get the peer/interface MTU.
2829 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2831 struct nhop4_extended nh4;
2833 uint32_t maxmtu = 0;
2835 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2837 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2839 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2840 NHR_REF, 0, &nh4) != 0)
2844 maxmtu = nh4.nh_mtu;
2846 /* Report additional interface capabilities. */
2848 if (ifp->if_capenable & IFCAP_TSO4 &&
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 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2864 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2866 struct nhop6_extended nh6;
2867 struct in6_addr dst6;
2870 uint32_t maxmtu = 0;
2872 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2874 if (inc->inc_flags & INC_IPV6MINMTU)
2877 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2878 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2879 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2884 maxmtu = nh6.nh_mtu;
2886 /* Report additional interface capabilities. */
2888 if (ifp->if_capenable & IFCAP_TSO6 &&
2889 ifp->if_hwassist & CSUM_TSO) {
2890 cap->ifcap |= CSUM_TSO;
2891 cap->tsomax = ifp->if_hw_tsomax;
2892 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2893 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2896 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2904 * Calculate effective SMSS per RFC5681 definition for a given TCP
2905 * connection at its current state, taking into account SACK and etc.
2908 tcp_maxseg(const struct tcpcb *tp)
2912 if (tp->t_flags & TF_NOOPT)
2913 return (tp->t_maxseg);
2916 * Here we have a simplified code from tcp_addoptions(),
2917 * without a proper loop, and having most of paddings hardcoded.
2918 * We might make mistakes with padding here in some edge cases,
2919 * but this is harmless, since result of tcp_maxseg() is used
2920 * only in cwnd and ssthresh estimations.
2922 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2923 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2924 if (tp->t_flags & TF_RCVD_TSTMP)
2925 optlen = TCPOLEN_TSTAMP_APPA;
2928 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2929 if (tp->t_flags & TF_SIGNATURE)
2930 optlen += PAD(TCPOLEN_SIGNATURE);
2932 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2933 optlen += TCPOLEN_SACKHDR;
2934 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2935 optlen = PAD(optlen);
2938 if (tp->t_flags & TF_REQ_TSTMP)
2939 optlen = TCPOLEN_TSTAMP_APPA;
2941 optlen = PAD(TCPOLEN_MAXSEG);
2942 if (tp->t_flags & TF_REQ_SCALE)
2943 optlen += PAD(TCPOLEN_WINDOW);
2944 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2945 if (tp->t_flags & TF_SIGNATURE)
2946 optlen += PAD(TCPOLEN_SIGNATURE);
2948 if (tp->t_flags & TF_SACK_PERMIT)
2949 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2952 optlen = min(optlen, TCP_MAXOLEN);
2953 return (tp->t_maxseg - optlen);
2957 sysctl_drop(SYSCTL_HANDLER_ARGS)
2959 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2960 struct sockaddr_storage addrs[2];
2964 struct sockaddr_in *fin, *lin;
2965 struct epoch_tracker et;
2967 struct sockaddr_in6 *fin6, *lin6;
2978 if (req->oldptr != NULL || req->oldlen != 0)
2980 if (req->newptr == NULL)
2982 if (req->newlen < sizeof(addrs))
2984 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2988 switch (addrs[0].ss_family) {
2991 fin6 = (struct sockaddr_in6 *)&addrs[0];
2992 lin6 = (struct sockaddr_in6 *)&addrs[1];
2993 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2994 lin6->sin6_len != sizeof(struct sockaddr_in6))
2996 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2997 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2999 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3000 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3001 fin = (struct sockaddr_in *)&addrs[0];
3002 lin = (struct sockaddr_in *)&addrs[1];
3005 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3008 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3015 fin = (struct sockaddr_in *)&addrs[0];
3016 lin = (struct sockaddr_in *)&addrs[1];
3017 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3018 lin->sin_len != sizeof(struct sockaddr_in))
3025 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
3026 switch (addrs[0].ss_family) {
3029 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3030 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3031 INPLOOKUP_WLOCKPCB, NULL);
3036 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3037 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3042 if (inp->inp_flags & INP_TIMEWAIT) {
3044 * XXXRW: There currently exists a state where an
3045 * inpcb is present, but its timewait state has been
3046 * discarded. For now, don't allow dropping of this
3054 } else if (!(inp->inp_flags & INP_DROPPED) &&
3055 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
3056 tp = intotcpcb(inp);
3057 tp = tcp_drop(tp, ECONNABORTED);
3064 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
3068 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3069 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
3070 0, sysctl_drop, "", "Drop TCP connection");
3073 * Generate a standardized TCP log line for use throughout the
3074 * tcp subsystem. Memory allocation is done with M_NOWAIT to
3075 * allow use in the interrupt context.
3077 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3078 * NB: The function may return NULL if memory allocation failed.
3080 * Due to header inclusion and ordering limitations the struct ip
3081 * and ip6_hdr pointers have to be passed as void pointers.
3084 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3088 /* Is logging enabled? */
3089 if (tcp_log_in_vain == 0)
3092 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3096 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3100 /* Is logging enabled? */
3101 if (tcp_log_debug == 0)
3104 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3108 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3115 const struct ip6_hdr *ip6;
3117 ip6 = (const struct ip6_hdr *)ip6hdr;
3119 ip = (struct ip *)ip4hdr;
3122 * The log line looks like this:
3123 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3125 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3126 sizeof(PRINT_TH_FLAGS) + 1 +
3128 2 * INET6_ADDRSTRLEN;
3130 2 * INET_ADDRSTRLEN;
3133 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3137 strcat(s, "TCP: [");
3140 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3141 inet_ntoa_r(inc->inc_faddr, sp);
3143 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3145 inet_ntoa_r(inc->inc_laddr, sp);
3147 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3150 ip6_sprintf(sp, &inc->inc6_faddr);
3152 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3154 ip6_sprintf(sp, &inc->inc6_laddr);
3156 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3157 } else if (ip6 && th) {
3158 ip6_sprintf(sp, &ip6->ip6_src);
3160 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3162 ip6_sprintf(sp, &ip6->ip6_dst);
3164 sprintf(sp, "]:%i", ntohs(th->th_dport));
3167 } else if (ip && th) {
3168 inet_ntoa_r(ip->ip_src, sp);
3170 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3172 inet_ntoa_r(ip->ip_dst, sp);
3174 sprintf(sp, "]:%i", ntohs(th->th_dport));
3182 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3183 if (*(s + size - 1) != '\0')
3184 panic("%s: string too long", __func__);
3189 * A subroutine which makes it easy to track TCP state changes with DTrace.
3190 * This function shouldn't be called for t_state initializations that don't
3191 * correspond to actual TCP state transitions.
3194 tcp_state_change(struct tcpcb *tp, int newstate)
3196 #if defined(KDTRACE_HOOKS)
3197 int pstate = tp->t_state;
3200 TCPSTATES_DEC(tp->t_state);
3201 TCPSTATES_INC(newstate);
3202 tp->t_state = newstate;
3203 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3207 * Create an external-format (``xtcpcb'') structure using the information in
3208 * the kernel-format tcpcb structure pointed to by tp. This is done to
3209 * reduce the spew of irrelevant information over this interface, to isolate
3210 * user code from changes in the kernel structure, and potentially to provide
3211 * information-hiding if we decide that some of this information should be
3212 * hidden from users.
3215 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3217 struct tcpcb *tp = intotcpcb(inp);
3220 bzero(xt, sizeof(*xt));
3221 if (inp->inp_flags & INP_TIMEWAIT) {
3222 xt->t_state = TCPS_TIME_WAIT;
3224 xt->t_state = tp->t_state;
3225 xt->t_logstate = tp->t_logstate;
3226 xt->t_flags = tp->t_flags;
3227 xt->t_sndzerowin = tp->t_sndzerowin;
3228 xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3229 xt->t_rcvoopack = tp->t_rcvoopack;
3231 now = getsbinuptime();
3232 #define COPYTIMER(ttt) do { \
3233 if (callout_active(&tp->t_timers->ttt)) \
3234 xt->ttt = (tp->t_timers->ttt.c_time - now) / \
3239 COPYTIMER(tt_delack);
3240 COPYTIMER(tt_rexmt);
3241 COPYTIMER(tt_persist);
3245 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3247 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3248 TCP_FUNCTION_NAME_LEN_MAX);
3250 (void)tcp_log_get_id(tp, xt->xt_logid);
3254 xt->xt_len = sizeof(struct xtcpcb);
3255 in_pcbtoxinpcb(inp, &xt->xt_inp);
3256 if (inp->inp_socket == NULL)
3257 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;